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 "dev.h" 155 #include "net-sysfs.h" 156 157 158 static DEFINE_SPINLOCK(ptype_lock); 159 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; 160 struct list_head ptype_all __read_mostly; /* Taps */ 161 162 static int netif_rx_internal(struct sk_buff *skb); 163 static int call_netdevice_notifiers_info(unsigned long val, 164 struct netdev_notifier_info *info); 165 static int call_netdevice_notifiers_extack(unsigned long val, 166 struct net_device *dev, 167 struct netlink_ext_ack *extack); 168 static struct napi_struct *napi_by_id(unsigned int napi_id); 169 170 /* 171 * The @dev_base_head list is protected by @dev_base_lock and the rtnl 172 * semaphore. 173 * 174 * Pure readers hold dev_base_lock for reading, or rcu_read_lock() 175 * 176 * Writers must hold the rtnl semaphore while they loop through the 177 * dev_base_head list, and hold dev_base_lock for writing when they do the 178 * actual updates. This allows pure readers to access the list even 179 * while a writer is preparing to update it. 180 * 181 * To put it another way, dev_base_lock is held for writing only to 182 * protect against pure readers; the rtnl semaphore provides the 183 * protection against other writers. 184 * 185 * See, for example usages, register_netdevice() and 186 * unregister_netdevice(), which must be called with the rtnl 187 * semaphore held. 188 */ 189 DEFINE_RWLOCK(dev_base_lock); 190 EXPORT_SYMBOL(dev_base_lock); 191 192 static DEFINE_MUTEX(ifalias_mutex); 193 194 /* protects napi_hash addition/deletion and napi_gen_id */ 195 static DEFINE_SPINLOCK(napi_hash_lock); 196 197 static unsigned int napi_gen_id = NR_CPUS; 198 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8); 199 200 static DECLARE_RWSEM(devnet_rename_sem); 201 202 static inline void dev_base_seq_inc(struct net *net) 203 { 204 while (++net->dev_base_seq == 0) 205 ; 206 } 207 208 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) 209 { 210 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ)); 211 212 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; 213 } 214 215 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) 216 { 217 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; 218 } 219 220 static inline void rps_lock_irqsave(struct softnet_data *sd, 221 unsigned long *flags) 222 { 223 if (IS_ENABLED(CONFIG_RPS)) 224 spin_lock_irqsave(&sd->input_pkt_queue.lock, *flags); 225 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 226 local_irq_save(*flags); 227 } 228 229 static inline void rps_lock_irq_disable(struct softnet_data *sd) 230 { 231 if (IS_ENABLED(CONFIG_RPS)) 232 spin_lock_irq(&sd->input_pkt_queue.lock); 233 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 234 local_irq_disable(); 235 } 236 237 static inline void rps_unlock_irq_restore(struct softnet_data *sd, 238 unsigned long *flags) 239 { 240 if (IS_ENABLED(CONFIG_RPS)) 241 spin_unlock_irqrestore(&sd->input_pkt_queue.lock, *flags); 242 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 243 local_irq_restore(*flags); 244 } 245 246 static inline void rps_unlock_irq_enable(struct softnet_data *sd) 247 { 248 if (IS_ENABLED(CONFIG_RPS)) 249 spin_unlock_irq(&sd->input_pkt_queue.lock); 250 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 251 local_irq_enable(); 252 } 253 254 static struct netdev_name_node *netdev_name_node_alloc(struct net_device *dev, 255 const char *name) 256 { 257 struct netdev_name_node *name_node; 258 259 name_node = kmalloc(sizeof(*name_node), GFP_KERNEL); 260 if (!name_node) 261 return NULL; 262 INIT_HLIST_NODE(&name_node->hlist); 263 name_node->dev = dev; 264 name_node->name = name; 265 return name_node; 266 } 267 268 static struct netdev_name_node * 269 netdev_name_node_head_alloc(struct net_device *dev) 270 { 271 struct netdev_name_node *name_node; 272 273 name_node = netdev_name_node_alloc(dev, dev->name); 274 if (!name_node) 275 return NULL; 276 INIT_LIST_HEAD(&name_node->list); 277 return name_node; 278 } 279 280 static void netdev_name_node_free(struct netdev_name_node *name_node) 281 { 282 kfree(name_node); 283 } 284 285 static void netdev_name_node_add(struct net *net, 286 struct netdev_name_node *name_node) 287 { 288 hlist_add_head_rcu(&name_node->hlist, 289 dev_name_hash(net, name_node->name)); 290 } 291 292 static void netdev_name_node_del(struct netdev_name_node *name_node) 293 { 294 hlist_del_rcu(&name_node->hlist); 295 } 296 297 static struct netdev_name_node *netdev_name_node_lookup(struct net *net, 298 const char *name) 299 { 300 struct hlist_head *head = dev_name_hash(net, name); 301 struct netdev_name_node *name_node; 302 303 hlist_for_each_entry(name_node, head, hlist) 304 if (!strcmp(name_node->name, name)) 305 return name_node; 306 return NULL; 307 } 308 309 static struct netdev_name_node *netdev_name_node_lookup_rcu(struct net *net, 310 const char *name) 311 { 312 struct hlist_head *head = dev_name_hash(net, name); 313 struct netdev_name_node *name_node; 314 315 hlist_for_each_entry_rcu(name_node, head, hlist) 316 if (!strcmp(name_node->name, name)) 317 return name_node; 318 return NULL; 319 } 320 321 bool netdev_name_in_use(struct net *net, const char *name) 322 { 323 return netdev_name_node_lookup(net, name); 324 } 325 EXPORT_SYMBOL(netdev_name_in_use); 326 327 int netdev_name_node_alt_create(struct net_device *dev, const char *name) 328 { 329 struct netdev_name_node *name_node; 330 struct net *net = dev_net(dev); 331 332 name_node = netdev_name_node_lookup(net, name); 333 if (name_node) 334 return -EEXIST; 335 name_node = netdev_name_node_alloc(dev, name); 336 if (!name_node) 337 return -ENOMEM; 338 netdev_name_node_add(net, name_node); 339 /* The node that holds dev->name acts as a head of per-device list. */ 340 list_add_tail(&name_node->list, &dev->name_node->list); 341 342 return 0; 343 } 344 345 static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node) 346 { 347 list_del(&name_node->list); 348 netdev_name_node_del(name_node); 349 kfree(name_node->name); 350 netdev_name_node_free(name_node); 351 } 352 353 int netdev_name_node_alt_destroy(struct net_device *dev, const char *name) 354 { 355 struct netdev_name_node *name_node; 356 struct net *net = dev_net(dev); 357 358 name_node = netdev_name_node_lookup(net, name); 359 if (!name_node) 360 return -ENOENT; 361 /* lookup might have found our primary name or a name belonging 362 * to another device. 363 */ 364 if (name_node == dev->name_node || name_node->dev != dev) 365 return -EINVAL; 366 367 __netdev_name_node_alt_destroy(name_node); 368 369 return 0; 370 } 371 372 static void netdev_name_node_alt_flush(struct net_device *dev) 373 { 374 struct netdev_name_node *name_node, *tmp; 375 376 list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list) 377 __netdev_name_node_alt_destroy(name_node); 378 } 379 380 /* Device list insertion */ 381 static void list_netdevice(struct net_device *dev) 382 { 383 struct net *net = dev_net(dev); 384 385 ASSERT_RTNL(); 386 387 write_lock(&dev_base_lock); 388 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); 389 netdev_name_node_add(net, dev->name_node); 390 hlist_add_head_rcu(&dev->index_hlist, 391 dev_index_hash(net, dev->ifindex)); 392 write_unlock(&dev_base_lock); 393 394 dev_base_seq_inc(net); 395 } 396 397 /* Device list removal 398 * caller must respect a RCU grace period before freeing/reusing dev 399 */ 400 static void unlist_netdevice(struct net_device *dev, bool lock) 401 { 402 ASSERT_RTNL(); 403 404 /* Unlink dev from the device chain */ 405 if (lock) 406 write_lock(&dev_base_lock); 407 list_del_rcu(&dev->dev_list); 408 netdev_name_node_del(dev->name_node); 409 hlist_del_rcu(&dev->index_hlist); 410 if (lock) 411 write_unlock(&dev_base_lock); 412 413 dev_base_seq_inc(dev_net(dev)); 414 } 415 416 /* 417 * Our notifier list 418 */ 419 420 static RAW_NOTIFIER_HEAD(netdev_chain); 421 422 /* 423 * Device drivers call our routines to queue packets here. We empty the 424 * queue in the local softnet handler. 425 */ 426 427 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); 428 EXPORT_PER_CPU_SYMBOL(softnet_data); 429 430 #ifdef CONFIG_LOCKDEP 431 /* 432 * register_netdevice() inits txq->_xmit_lock and sets lockdep class 433 * according to dev->type 434 */ 435 static const unsigned short netdev_lock_type[] = { 436 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, 437 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, 438 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, 439 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, 440 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, 441 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, 442 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, 443 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, 444 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, 445 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, 446 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, 447 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, 448 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, 449 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, 450 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; 451 452 static const char *const netdev_lock_name[] = { 453 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", 454 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", 455 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", 456 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", 457 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", 458 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", 459 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", 460 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", 461 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", 462 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", 463 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", 464 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", 465 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", 466 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", 467 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; 468 469 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; 470 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; 471 472 static inline unsigned short netdev_lock_pos(unsigned short dev_type) 473 { 474 int i; 475 476 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) 477 if (netdev_lock_type[i] == dev_type) 478 return i; 479 /* the last key is used by default */ 480 return ARRAY_SIZE(netdev_lock_type) - 1; 481 } 482 483 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 484 unsigned short dev_type) 485 { 486 int i; 487 488 i = netdev_lock_pos(dev_type); 489 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], 490 netdev_lock_name[i]); 491 } 492 493 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 494 { 495 int i; 496 497 i = netdev_lock_pos(dev->type); 498 lockdep_set_class_and_name(&dev->addr_list_lock, 499 &netdev_addr_lock_key[i], 500 netdev_lock_name[i]); 501 } 502 #else 503 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 504 unsigned short dev_type) 505 { 506 } 507 508 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 509 { 510 } 511 #endif 512 513 /******************************************************************************* 514 * 515 * Protocol management and registration routines 516 * 517 *******************************************************************************/ 518 519 520 /* 521 * Add a protocol ID to the list. Now that the input handler is 522 * smarter we can dispense with all the messy stuff that used to be 523 * here. 524 * 525 * BEWARE!!! Protocol handlers, mangling input packets, 526 * MUST BE last in hash buckets and checking protocol handlers 527 * MUST start from promiscuous ptype_all chain in net_bh. 528 * It is true now, do not change it. 529 * Explanation follows: if protocol handler, mangling packet, will 530 * be the first on list, it is not able to sense, that packet 531 * is cloned and should be copied-on-write, so that it will 532 * change it and subsequent readers will get broken packet. 533 * --ANK (980803) 534 */ 535 536 static inline struct list_head *ptype_head(const struct packet_type *pt) 537 { 538 if (pt->type == htons(ETH_P_ALL)) 539 return pt->dev ? &pt->dev->ptype_all : &ptype_all; 540 else 541 return pt->dev ? &pt->dev->ptype_specific : 542 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; 543 } 544 545 /** 546 * dev_add_pack - add packet handler 547 * @pt: packet type declaration 548 * 549 * Add a protocol handler to the networking stack. The passed &packet_type 550 * is linked into kernel lists and may not be freed until it has been 551 * removed from the kernel lists. 552 * 553 * This call does not sleep therefore it can not 554 * guarantee all CPU's that are in middle of receiving packets 555 * will see the new packet type (until the next received packet). 556 */ 557 558 void dev_add_pack(struct packet_type *pt) 559 { 560 struct list_head *head = ptype_head(pt); 561 562 spin_lock(&ptype_lock); 563 list_add_rcu(&pt->list, head); 564 spin_unlock(&ptype_lock); 565 } 566 EXPORT_SYMBOL(dev_add_pack); 567 568 /** 569 * __dev_remove_pack - remove packet handler 570 * @pt: packet type declaration 571 * 572 * Remove a protocol handler that was previously added to the kernel 573 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 574 * from the kernel lists and can be freed or reused once this function 575 * returns. 576 * 577 * The packet type might still be in use by receivers 578 * and must not be freed until after all the CPU's have gone 579 * through a quiescent state. 580 */ 581 void __dev_remove_pack(struct packet_type *pt) 582 { 583 struct list_head *head = ptype_head(pt); 584 struct packet_type *pt1; 585 586 spin_lock(&ptype_lock); 587 588 list_for_each_entry(pt1, head, list) { 589 if (pt == pt1) { 590 list_del_rcu(&pt->list); 591 goto out; 592 } 593 } 594 595 pr_warn("dev_remove_pack: %p not found\n", pt); 596 out: 597 spin_unlock(&ptype_lock); 598 } 599 EXPORT_SYMBOL(__dev_remove_pack); 600 601 /** 602 * dev_remove_pack - remove packet handler 603 * @pt: packet type declaration 604 * 605 * Remove a protocol handler that was previously added to the kernel 606 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 607 * from the kernel lists and can be freed or reused once this function 608 * returns. 609 * 610 * This call sleeps to guarantee that no CPU is looking at the packet 611 * type after return. 612 */ 613 void dev_remove_pack(struct packet_type *pt) 614 { 615 __dev_remove_pack(pt); 616 617 synchronize_net(); 618 } 619 EXPORT_SYMBOL(dev_remove_pack); 620 621 622 /******************************************************************************* 623 * 624 * Device Interface Subroutines 625 * 626 *******************************************************************************/ 627 628 /** 629 * dev_get_iflink - get 'iflink' value of a interface 630 * @dev: targeted interface 631 * 632 * Indicates the ifindex the interface is linked to. 633 * Physical interfaces have the same 'ifindex' and 'iflink' values. 634 */ 635 636 int dev_get_iflink(const struct net_device *dev) 637 { 638 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink) 639 return dev->netdev_ops->ndo_get_iflink(dev); 640 641 return dev->ifindex; 642 } 643 EXPORT_SYMBOL(dev_get_iflink); 644 645 /** 646 * dev_fill_metadata_dst - Retrieve tunnel egress information. 647 * @dev: targeted interface 648 * @skb: The packet. 649 * 650 * For better visibility of tunnel traffic OVS needs to retrieve 651 * egress tunnel information for a packet. Following API allows 652 * user to get this info. 653 */ 654 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) 655 { 656 struct ip_tunnel_info *info; 657 658 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst) 659 return -EINVAL; 660 661 info = skb_tunnel_info_unclone(skb); 662 if (!info) 663 return -ENOMEM; 664 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX))) 665 return -EINVAL; 666 667 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb); 668 } 669 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst); 670 671 static struct net_device_path *dev_fwd_path(struct net_device_path_stack *stack) 672 { 673 int k = stack->num_paths++; 674 675 if (WARN_ON_ONCE(k >= NET_DEVICE_PATH_STACK_MAX)) 676 return NULL; 677 678 return &stack->path[k]; 679 } 680 681 int dev_fill_forward_path(const struct net_device *dev, const u8 *daddr, 682 struct net_device_path_stack *stack) 683 { 684 const struct net_device *last_dev; 685 struct net_device_path_ctx ctx = { 686 .dev = dev, 687 }; 688 struct net_device_path *path; 689 int ret = 0; 690 691 memcpy(ctx.daddr, daddr, sizeof(ctx.daddr)); 692 stack->num_paths = 0; 693 while (ctx.dev && ctx.dev->netdev_ops->ndo_fill_forward_path) { 694 last_dev = ctx.dev; 695 path = dev_fwd_path(stack); 696 if (!path) 697 return -1; 698 699 memset(path, 0, sizeof(struct net_device_path)); 700 ret = ctx.dev->netdev_ops->ndo_fill_forward_path(&ctx, path); 701 if (ret < 0) 702 return -1; 703 704 if (WARN_ON_ONCE(last_dev == ctx.dev)) 705 return -1; 706 } 707 708 if (!ctx.dev) 709 return ret; 710 711 path = dev_fwd_path(stack); 712 if (!path) 713 return -1; 714 path->type = DEV_PATH_ETHERNET; 715 path->dev = ctx.dev; 716 717 return ret; 718 } 719 EXPORT_SYMBOL_GPL(dev_fill_forward_path); 720 721 /** 722 * __dev_get_by_name - find a device by its name 723 * @net: the applicable net namespace 724 * @name: name to find 725 * 726 * Find an interface by name. Must be called under RTNL semaphore 727 * or @dev_base_lock. If the name is found a pointer to the device 728 * is returned. If the name is not found then %NULL is returned. The 729 * reference counters are not incremented so the caller must be 730 * careful with locks. 731 */ 732 733 struct net_device *__dev_get_by_name(struct net *net, const char *name) 734 { 735 struct netdev_name_node *node_name; 736 737 node_name = netdev_name_node_lookup(net, name); 738 return node_name ? node_name->dev : NULL; 739 } 740 EXPORT_SYMBOL(__dev_get_by_name); 741 742 /** 743 * dev_get_by_name_rcu - find a device by its name 744 * @net: the applicable net namespace 745 * @name: name to find 746 * 747 * Find an interface by name. 748 * If the name is found a pointer to the device is returned. 749 * If the name is not found then %NULL is returned. 750 * The reference counters are not incremented so the caller must be 751 * careful with locks. The caller must hold RCU lock. 752 */ 753 754 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) 755 { 756 struct netdev_name_node *node_name; 757 758 node_name = netdev_name_node_lookup_rcu(net, name); 759 return node_name ? node_name->dev : NULL; 760 } 761 EXPORT_SYMBOL(dev_get_by_name_rcu); 762 763 /** 764 * dev_get_by_name - find a device by its name 765 * @net: the applicable net namespace 766 * @name: name to find 767 * 768 * Find an interface by name. This can be called from any 769 * context and does its own locking. The returned handle has 770 * the usage count incremented and the caller must use dev_put() to 771 * release it when it is no longer needed. %NULL is returned if no 772 * matching device is found. 773 */ 774 775 struct net_device *dev_get_by_name(struct net *net, const char *name) 776 { 777 struct net_device *dev; 778 779 rcu_read_lock(); 780 dev = dev_get_by_name_rcu(net, name); 781 dev_hold(dev); 782 rcu_read_unlock(); 783 return dev; 784 } 785 EXPORT_SYMBOL(dev_get_by_name); 786 787 /** 788 * __dev_get_by_index - find a device by its ifindex 789 * @net: the applicable net namespace 790 * @ifindex: index of device 791 * 792 * Search for an interface by index. Returns %NULL if the device 793 * is not found or a pointer to the device. The device has not 794 * had its reference counter increased so the caller must be careful 795 * about locking. The caller must hold either the RTNL semaphore 796 * or @dev_base_lock. 797 */ 798 799 struct net_device *__dev_get_by_index(struct net *net, int ifindex) 800 { 801 struct net_device *dev; 802 struct hlist_head *head = dev_index_hash(net, ifindex); 803 804 hlist_for_each_entry(dev, head, index_hlist) 805 if (dev->ifindex == ifindex) 806 return dev; 807 808 return NULL; 809 } 810 EXPORT_SYMBOL(__dev_get_by_index); 811 812 /** 813 * dev_get_by_index_rcu - find a device by its ifindex 814 * @net: the applicable net namespace 815 * @ifindex: index of device 816 * 817 * Search for an interface by index. Returns %NULL if the device 818 * is not found or a pointer to the device. The device has not 819 * had its reference counter increased so the caller must be careful 820 * about locking. The caller must hold RCU lock. 821 */ 822 823 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) 824 { 825 struct net_device *dev; 826 struct hlist_head *head = dev_index_hash(net, ifindex); 827 828 hlist_for_each_entry_rcu(dev, head, index_hlist) 829 if (dev->ifindex == ifindex) 830 return dev; 831 832 return NULL; 833 } 834 EXPORT_SYMBOL(dev_get_by_index_rcu); 835 836 837 /** 838 * dev_get_by_index - find a device by its ifindex 839 * @net: the applicable net namespace 840 * @ifindex: index of device 841 * 842 * Search for an interface by index. Returns NULL if the device 843 * is not found or a pointer to the device. The device returned has 844 * had a reference added and the pointer is safe until the user calls 845 * dev_put to indicate they have finished with it. 846 */ 847 848 struct net_device *dev_get_by_index(struct net *net, int ifindex) 849 { 850 struct net_device *dev; 851 852 rcu_read_lock(); 853 dev = dev_get_by_index_rcu(net, ifindex); 854 dev_hold(dev); 855 rcu_read_unlock(); 856 return dev; 857 } 858 EXPORT_SYMBOL(dev_get_by_index); 859 860 /** 861 * dev_get_by_napi_id - find a device by napi_id 862 * @napi_id: ID of the NAPI struct 863 * 864 * Search for an interface by NAPI ID. Returns %NULL if the device 865 * is not found or a pointer to the device. The device has not had 866 * its reference counter increased so the caller must be careful 867 * about locking. The caller must hold RCU lock. 868 */ 869 870 struct net_device *dev_get_by_napi_id(unsigned int napi_id) 871 { 872 struct napi_struct *napi; 873 874 WARN_ON_ONCE(!rcu_read_lock_held()); 875 876 if (napi_id < MIN_NAPI_ID) 877 return NULL; 878 879 napi = napi_by_id(napi_id); 880 881 return napi ? napi->dev : NULL; 882 } 883 EXPORT_SYMBOL(dev_get_by_napi_id); 884 885 /** 886 * netdev_get_name - get a netdevice name, knowing its ifindex. 887 * @net: network namespace 888 * @name: a pointer to the buffer where the name will be stored. 889 * @ifindex: the ifindex of the interface to get the name from. 890 */ 891 int netdev_get_name(struct net *net, char *name, int ifindex) 892 { 893 struct net_device *dev; 894 int ret; 895 896 down_read(&devnet_rename_sem); 897 rcu_read_lock(); 898 899 dev = dev_get_by_index_rcu(net, ifindex); 900 if (!dev) { 901 ret = -ENODEV; 902 goto out; 903 } 904 905 strcpy(name, dev->name); 906 907 ret = 0; 908 out: 909 rcu_read_unlock(); 910 up_read(&devnet_rename_sem); 911 return ret; 912 } 913 914 /** 915 * dev_getbyhwaddr_rcu - find a device by its hardware address 916 * @net: the applicable net namespace 917 * @type: media type of device 918 * @ha: hardware address 919 * 920 * Search for an interface by MAC address. Returns NULL if the device 921 * is not found or a pointer to the device. 922 * The caller must hold RCU or RTNL. 923 * The returned device has not had its ref count increased 924 * and the caller must therefore be careful about locking 925 * 926 */ 927 928 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, 929 const char *ha) 930 { 931 struct net_device *dev; 932 933 for_each_netdev_rcu(net, dev) 934 if (dev->type == type && 935 !memcmp(dev->dev_addr, ha, dev->addr_len)) 936 return dev; 937 938 return NULL; 939 } 940 EXPORT_SYMBOL(dev_getbyhwaddr_rcu); 941 942 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) 943 { 944 struct net_device *dev, *ret = NULL; 945 946 rcu_read_lock(); 947 for_each_netdev_rcu(net, dev) 948 if (dev->type == type) { 949 dev_hold(dev); 950 ret = dev; 951 break; 952 } 953 rcu_read_unlock(); 954 return ret; 955 } 956 EXPORT_SYMBOL(dev_getfirstbyhwtype); 957 958 /** 959 * __dev_get_by_flags - find any device with given flags 960 * @net: the applicable net namespace 961 * @if_flags: IFF_* values 962 * @mask: bitmask of bits in if_flags to check 963 * 964 * Search for any interface with the given flags. Returns NULL if a device 965 * is not found or a pointer to the device. Must be called inside 966 * rtnl_lock(), and result refcount is unchanged. 967 */ 968 969 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags, 970 unsigned short mask) 971 { 972 struct net_device *dev, *ret; 973 974 ASSERT_RTNL(); 975 976 ret = NULL; 977 for_each_netdev(net, dev) { 978 if (((dev->flags ^ if_flags) & mask) == 0) { 979 ret = dev; 980 break; 981 } 982 } 983 return ret; 984 } 985 EXPORT_SYMBOL(__dev_get_by_flags); 986 987 /** 988 * dev_valid_name - check if name is okay for network device 989 * @name: name string 990 * 991 * Network device names need to be valid file names to 992 * allow sysfs to work. We also disallow any kind of 993 * whitespace. 994 */ 995 bool dev_valid_name(const char *name) 996 { 997 if (*name == '\0') 998 return false; 999 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ) 1000 return false; 1001 if (!strcmp(name, ".") || !strcmp(name, "..")) 1002 return false; 1003 1004 while (*name) { 1005 if (*name == '/' || *name == ':' || isspace(*name)) 1006 return false; 1007 name++; 1008 } 1009 return true; 1010 } 1011 EXPORT_SYMBOL(dev_valid_name); 1012 1013 /** 1014 * __dev_alloc_name - allocate a name for a device 1015 * @net: network namespace to allocate the device name in 1016 * @name: name format string 1017 * @buf: scratch buffer and result name string 1018 * 1019 * Passed a format string - eg "lt%d" it will try and find a suitable 1020 * id. It scans list of devices to build up a free map, then chooses 1021 * the first empty slot. The caller must hold the dev_base or rtnl lock 1022 * while allocating the name and adding the device in order to avoid 1023 * duplicates. 1024 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1025 * Returns the number of the unit assigned or a negative errno code. 1026 */ 1027 1028 static int __dev_alloc_name(struct net *net, const char *name, char *buf) 1029 { 1030 int i = 0; 1031 const char *p; 1032 const int max_netdevices = 8*PAGE_SIZE; 1033 unsigned long *inuse; 1034 struct net_device *d; 1035 1036 if (!dev_valid_name(name)) 1037 return -EINVAL; 1038 1039 p = strchr(name, '%'); 1040 if (p) { 1041 /* 1042 * Verify the string as this thing may have come from 1043 * the user. There must be either one "%d" and no other "%" 1044 * characters. 1045 */ 1046 if (p[1] != 'd' || strchr(p + 2, '%')) 1047 return -EINVAL; 1048 1049 /* Use one page as a bit array of possible slots */ 1050 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); 1051 if (!inuse) 1052 return -ENOMEM; 1053 1054 for_each_netdev(net, d) { 1055 struct netdev_name_node *name_node; 1056 list_for_each_entry(name_node, &d->name_node->list, list) { 1057 if (!sscanf(name_node->name, name, &i)) 1058 continue; 1059 if (i < 0 || i >= max_netdevices) 1060 continue; 1061 1062 /* avoid cases where sscanf is not exact inverse of printf */ 1063 snprintf(buf, IFNAMSIZ, name, i); 1064 if (!strncmp(buf, name_node->name, IFNAMSIZ)) 1065 __set_bit(i, inuse); 1066 } 1067 if (!sscanf(d->name, name, &i)) 1068 continue; 1069 if (i < 0 || i >= max_netdevices) 1070 continue; 1071 1072 /* avoid cases where sscanf is not exact inverse of printf */ 1073 snprintf(buf, IFNAMSIZ, name, i); 1074 if (!strncmp(buf, d->name, IFNAMSIZ)) 1075 __set_bit(i, inuse); 1076 } 1077 1078 i = find_first_zero_bit(inuse, max_netdevices); 1079 free_page((unsigned long) inuse); 1080 } 1081 1082 snprintf(buf, IFNAMSIZ, name, i); 1083 if (!netdev_name_in_use(net, buf)) 1084 return i; 1085 1086 /* It is possible to run out of possible slots 1087 * when the name is long and there isn't enough space left 1088 * for the digits, or if all bits are used. 1089 */ 1090 return -ENFILE; 1091 } 1092 1093 static int dev_alloc_name_ns(struct net *net, 1094 struct net_device *dev, 1095 const char *name) 1096 { 1097 char buf[IFNAMSIZ]; 1098 int ret; 1099 1100 BUG_ON(!net); 1101 ret = __dev_alloc_name(net, name, buf); 1102 if (ret >= 0) 1103 strscpy(dev->name, buf, IFNAMSIZ); 1104 return ret; 1105 } 1106 1107 /** 1108 * dev_alloc_name - allocate a name for a device 1109 * @dev: device 1110 * @name: name format string 1111 * 1112 * Passed a format string - eg "lt%d" it will try and find a suitable 1113 * id. It scans list of devices to build up a free map, then chooses 1114 * the first empty slot. The caller must hold the dev_base or rtnl lock 1115 * while allocating the name and adding the device in order to avoid 1116 * duplicates. 1117 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1118 * Returns the number of the unit assigned or a negative errno code. 1119 */ 1120 1121 int dev_alloc_name(struct net_device *dev, const char *name) 1122 { 1123 return dev_alloc_name_ns(dev_net(dev), dev, name); 1124 } 1125 EXPORT_SYMBOL(dev_alloc_name); 1126 1127 static int dev_get_valid_name(struct net *net, struct net_device *dev, 1128 const char *name) 1129 { 1130 BUG_ON(!net); 1131 1132 if (!dev_valid_name(name)) 1133 return -EINVAL; 1134 1135 if (strchr(name, '%')) 1136 return dev_alloc_name_ns(net, dev, name); 1137 else if (netdev_name_in_use(net, name)) 1138 return -EEXIST; 1139 else if (dev->name != name) 1140 strscpy(dev->name, name, IFNAMSIZ); 1141 1142 return 0; 1143 } 1144 1145 /** 1146 * dev_change_name - change name of a device 1147 * @dev: device 1148 * @newname: name (or format string) must be at least IFNAMSIZ 1149 * 1150 * Change name of a device, can pass format strings "eth%d". 1151 * for wildcarding. 1152 */ 1153 int dev_change_name(struct net_device *dev, const char *newname) 1154 { 1155 unsigned char old_assign_type; 1156 char oldname[IFNAMSIZ]; 1157 int err = 0; 1158 int ret; 1159 struct net *net; 1160 1161 ASSERT_RTNL(); 1162 BUG_ON(!dev_net(dev)); 1163 1164 net = dev_net(dev); 1165 1166 /* Some auto-enslaved devices e.g. failover slaves are 1167 * special, as userspace might rename the device after 1168 * the interface had been brought up and running since 1169 * the point kernel initiated auto-enslavement. Allow 1170 * live name change even when these slave devices are 1171 * up and running. 1172 * 1173 * Typically, users of these auto-enslaving devices 1174 * don't actually care about slave name change, as 1175 * they are supposed to operate on master interface 1176 * directly. 1177 */ 1178 if (dev->flags & IFF_UP && 1179 likely(!(dev->priv_flags & IFF_LIVE_RENAME_OK))) 1180 return -EBUSY; 1181 1182 down_write(&devnet_rename_sem); 1183 1184 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { 1185 up_write(&devnet_rename_sem); 1186 return 0; 1187 } 1188 1189 memcpy(oldname, dev->name, IFNAMSIZ); 1190 1191 err = dev_get_valid_name(net, dev, newname); 1192 if (err < 0) { 1193 up_write(&devnet_rename_sem); 1194 return err; 1195 } 1196 1197 if (oldname[0] && !strchr(oldname, '%')) 1198 netdev_info(dev, "renamed from %s\n", oldname); 1199 1200 old_assign_type = dev->name_assign_type; 1201 dev->name_assign_type = NET_NAME_RENAMED; 1202 1203 rollback: 1204 ret = device_rename(&dev->dev, dev->name); 1205 if (ret) { 1206 memcpy(dev->name, oldname, IFNAMSIZ); 1207 dev->name_assign_type = old_assign_type; 1208 up_write(&devnet_rename_sem); 1209 return ret; 1210 } 1211 1212 up_write(&devnet_rename_sem); 1213 1214 netdev_adjacent_rename_links(dev, oldname); 1215 1216 write_lock(&dev_base_lock); 1217 netdev_name_node_del(dev->name_node); 1218 write_unlock(&dev_base_lock); 1219 1220 synchronize_rcu(); 1221 1222 write_lock(&dev_base_lock); 1223 netdev_name_node_add(net, dev->name_node); 1224 write_unlock(&dev_base_lock); 1225 1226 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); 1227 ret = notifier_to_errno(ret); 1228 1229 if (ret) { 1230 /* err >= 0 after dev_alloc_name() or stores the first errno */ 1231 if (err >= 0) { 1232 err = ret; 1233 down_write(&devnet_rename_sem); 1234 memcpy(dev->name, oldname, IFNAMSIZ); 1235 memcpy(oldname, newname, IFNAMSIZ); 1236 dev->name_assign_type = old_assign_type; 1237 old_assign_type = NET_NAME_RENAMED; 1238 goto rollback; 1239 } else { 1240 netdev_err(dev, "name change rollback failed: %d\n", 1241 ret); 1242 } 1243 } 1244 1245 return err; 1246 } 1247 1248 /** 1249 * dev_set_alias - change ifalias of a device 1250 * @dev: device 1251 * @alias: name up to IFALIASZ 1252 * @len: limit of bytes to copy from info 1253 * 1254 * Set ifalias for a device, 1255 */ 1256 int dev_set_alias(struct net_device *dev, const char *alias, size_t len) 1257 { 1258 struct dev_ifalias *new_alias = NULL; 1259 1260 if (len >= IFALIASZ) 1261 return -EINVAL; 1262 1263 if (len) { 1264 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL); 1265 if (!new_alias) 1266 return -ENOMEM; 1267 1268 memcpy(new_alias->ifalias, alias, len); 1269 new_alias->ifalias[len] = 0; 1270 } 1271 1272 mutex_lock(&ifalias_mutex); 1273 new_alias = rcu_replace_pointer(dev->ifalias, new_alias, 1274 mutex_is_locked(&ifalias_mutex)); 1275 mutex_unlock(&ifalias_mutex); 1276 1277 if (new_alias) 1278 kfree_rcu(new_alias, rcuhead); 1279 1280 return len; 1281 } 1282 EXPORT_SYMBOL(dev_set_alias); 1283 1284 /** 1285 * dev_get_alias - get ifalias of a device 1286 * @dev: device 1287 * @name: buffer to store name of ifalias 1288 * @len: size of buffer 1289 * 1290 * get ifalias for a device. Caller must make sure dev cannot go 1291 * away, e.g. rcu read lock or own a reference count to device. 1292 */ 1293 int dev_get_alias(const struct net_device *dev, char *name, size_t len) 1294 { 1295 const struct dev_ifalias *alias; 1296 int ret = 0; 1297 1298 rcu_read_lock(); 1299 alias = rcu_dereference(dev->ifalias); 1300 if (alias) 1301 ret = snprintf(name, len, "%s", alias->ifalias); 1302 rcu_read_unlock(); 1303 1304 return ret; 1305 } 1306 1307 /** 1308 * netdev_features_change - device changes features 1309 * @dev: device to cause notification 1310 * 1311 * Called to indicate a device has changed features. 1312 */ 1313 void netdev_features_change(struct net_device *dev) 1314 { 1315 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); 1316 } 1317 EXPORT_SYMBOL(netdev_features_change); 1318 1319 /** 1320 * netdev_state_change - device changes state 1321 * @dev: device to cause notification 1322 * 1323 * Called to indicate a device has changed state. This function calls 1324 * the notifier chains for netdev_chain and sends a NEWLINK message 1325 * to the routing socket. 1326 */ 1327 void netdev_state_change(struct net_device *dev) 1328 { 1329 if (dev->flags & IFF_UP) { 1330 struct netdev_notifier_change_info change_info = { 1331 .info.dev = dev, 1332 }; 1333 1334 call_netdevice_notifiers_info(NETDEV_CHANGE, 1335 &change_info.info); 1336 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL); 1337 } 1338 } 1339 EXPORT_SYMBOL(netdev_state_change); 1340 1341 /** 1342 * __netdev_notify_peers - notify network peers about existence of @dev, 1343 * to be called when rtnl lock is already held. 1344 * @dev: network device 1345 * 1346 * Generate traffic such that interested network peers are aware of 1347 * @dev, such as by generating a gratuitous ARP. This may be used when 1348 * a device wants to inform the rest of the network about some sort of 1349 * reconfiguration such as a failover event or virtual machine 1350 * migration. 1351 */ 1352 void __netdev_notify_peers(struct net_device *dev) 1353 { 1354 ASSERT_RTNL(); 1355 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); 1356 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev); 1357 } 1358 EXPORT_SYMBOL(__netdev_notify_peers); 1359 1360 /** 1361 * netdev_notify_peers - notify network peers about existence of @dev 1362 * @dev: network device 1363 * 1364 * Generate traffic such that interested network peers are aware of 1365 * @dev, such as by generating a gratuitous ARP. This may be used when 1366 * a device wants to inform the rest of the network about some sort of 1367 * reconfiguration such as a failover event or virtual machine 1368 * migration. 1369 */ 1370 void netdev_notify_peers(struct net_device *dev) 1371 { 1372 rtnl_lock(); 1373 __netdev_notify_peers(dev); 1374 rtnl_unlock(); 1375 } 1376 EXPORT_SYMBOL(netdev_notify_peers); 1377 1378 static int napi_threaded_poll(void *data); 1379 1380 static int napi_kthread_create(struct napi_struct *n) 1381 { 1382 int err = 0; 1383 1384 /* Create and wake up the kthread once to put it in 1385 * TASK_INTERRUPTIBLE mode to avoid the blocked task 1386 * warning and work with loadavg. 1387 */ 1388 n->thread = kthread_run(napi_threaded_poll, n, "napi/%s-%d", 1389 n->dev->name, n->napi_id); 1390 if (IS_ERR(n->thread)) { 1391 err = PTR_ERR(n->thread); 1392 pr_err("kthread_run failed with err %d\n", err); 1393 n->thread = NULL; 1394 } 1395 1396 return err; 1397 } 1398 1399 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1400 { 1401 const struct net_device_ops *ops = dev->netdev_ops; 1402 int ret; 1403 1404 ASSERT_RTNL(); 1405 dev_addr_check(dev); 1406 1407 if (!netif_device_present(dev)) { 1408 /* may be detached because parent is runtime-suspended */ 1409 if (dev->dev.parent) 1410 pm_runtime_resume(dev->dev.parent); 1411 if (!netif_device_present(dev)) 1412 return -ENODEV; 1413 } 1414 1415 /* Block netpoll from trying to do any rx path servicing. 1416 * If we don't do this there is a chance ndo_poll_controller 1417 * or ndo_poll may be running while we open the device 1418 */ 1419 netpoll_poll_disable(dev); 1420 1421 ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack); 1422 ret = notifier_to_errno(ret); 1423 if (ret) 1424 return ret; 1425 1426 set_bit(__LINK_STATE_START, &dev->state); 1427 1428 if (ops->ndo_validate_addr) 1429 ret = ops->ndo_validate_addr(dev); 1430 1431 if (!ret && ops->ndo_open) 1432 ret = ops->ndo_open(dev); 1433 1434 netpoll_poll_enable(dev); 1435 1436 if (ret) 1437 clear_bit(__LINK_STATE_START, &dev->state); 1438 else { 1439 dev->flags |= IFF_UP; 1440 dev_set_rx_mode(dev); 1441 dev_activate(dev); 1442 add_device_randomness(dev->dev_addr, dev->addr_len); 1443 } 1444 1445 return ret; 1446 } 1447 1448 /** 1449 * dev_open - prepare an interface for use. 1450 * @dev: device to open 1451 * @extack: netlink extended ack 1452 * 1453 * Takes a device from down to up state. The device's private open 1454 * function is invoked and then the multicast lists are loaded. Finally 1455 * the device is moved into the up state and a %NETDEV_UP message is 1456 * sent to the netdev notifier chain. 1457 * 1458 * Calling this function on an active interface is a nop. On a failure 1459 * a negative errno code is returned. 1460 */ 1461 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1462 { 1463 int ret; 1464 1465 if (dev->flags & IFF_UP) 1466 return 0; 1467 1468 ret = __dev_open(dev, extack); 1469 if (ret < 0) 1470 return ret; 1471 1472 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); 1473 call_netdevice_notifiers(NETDEV_UP, dev); 1474 1475 return ret; 1476 } 1477 EXPORT_SYMBOL(dev_open); 1478 1479 static void __dev_close_many(struct list_head *head) 1480 { 1481 struct net_device *dev; 1482 1483 ASSERT_RTNL(); 1484 might_sleep(); 1485 1486 list_for_each_entry(dev, head, close_list) { 1487 /* Temporarily disable netpoll until the interface is down */ 1488 netpoll_poll_disable(dev); 1489 1490 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); 1491 1492 clear_bit(__LINK_STATE_START, &dev->state); 1493 1494 /* Synchronize to scheduled poll. We cannot touch poll list, it 1495 * can be even on different cpu. So just clear netif_running(). 1496 * 1497 * dev->stop() will invoke napi_disable() on all of it's 1498 * napi_struct instances on this device. 1499 */ 1500 smp_mb__after_atomic(); /* Commit netif_running(). */ 1501 } 1502 1503 dev_deactivate_many(head); 1504 1505 list_for_each_entry(dev, head, close_list) { 1506 const struct net_device_ops *ops = dev->netdev_ops; 1507 1508 /* 1509 * Call the device specific close. This cannot fail. 1510 * Only if device is UP 1511 * 1512 * We allow it to be called even after a DETACH hot-plug 1513 * event. 1514 */ 1515 if (ops->ndo_stop) 1516 ops->ndo_stop(dev); 1517 1518 dev->flags &= ~IFF_UP; 1519 netpoll_poll_enable(dev); 1520 } 1521 } 1522 1523 static void __dev_close(struct net_device *dev) 1524 { 1525 LIST_HEAD(single); 1526 1527 list_add(&dev->close_list, &single); 1528 __dev_close_many(&single); 1529 list_del(&single); 1530 } 1531 1532 void dev_close_many(struct list_head *head, bool unlink) 1533 { 1534 struct net_device *dev, *tmp; 1535 1536 /* Remove the devices that don't need to be closed */ 1537 list_for_each_entry_safe(dev, tmp, head, close_list) 1538 if (!(dev->flags & IFF_UP)) 1539 list_del_init(&dev->close_list); 1540 1541 __dev_close_many(head); 1542 1543 list_for_each_entry_safe(dev, tmp, head, close_list) { 1544 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); 1545 call_netdevice_notifiers(NETDEV_DOWN, dev); 1546 if (unlink) 1547 list_del_init(&dev->close_list); 1548 } 1549 } 1550 EXPORT_SYMBOL(dev_close_many); 1551 1552 /** 1553 * dev_close - shutdown an interface. 1554 * @dev: device to shutdown 1555 * 1556 * This function moves an active device into down state. A 1557 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device 1558 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier 1559 * chain. 1560 */ 1561 void dev_close(struct net_device *dev) 1562 { 1563 if (dev->flags & IFF_UP) { 1564 LIST_HEAD(single); 1565 1566 list_add(&dev->close_list, &single); 1567 dev_close_many(&single, true); 1568 list_del(&single); 1569 } 1570 } 1571 EXPORT_SYMBOL(dev_close); 1572 1573 1574 /** 1575 * dev_disable_lro - disable Large Receive Offload on a device 1576 * @dev: device 1577 * 1578 * Disable Large Receive Offload (LRO) on a net device. Must be 1579 * called under RTNL. This is needed if received packets may be 1580 * forwarded to another interface. 1581 */ 1582 void dev_disable_lro(struct net_device *dev) 1583 { 1584 struct net_device *lower_dev; 1585 struct list_head *iter; 1586 1587 dev->wanted_features &= ~NETIF_F_LRO; 1588 netdev_update_features(dev); 1589 1590 if (unlikely(dev->features & NETIF_F_LRO)) 1591 netdev_WARN(dev, "failed to disable LRO!\n"); 1592 1593 netdev_for_each_lower_dev(dev, lower_dev, iter) 1594 dev_disable_lro(lower_dev); 1595 } 1596 EXPORT_SYMBOL(dev_disable_lro); 1597 1598 /** 1599 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device 1600 * @dev: device 1601 * 1602 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be 1603 * called under RTNL. This is needed if Generic XDP is installed on 1604 * the device. 1605 */ 1606 static void dev_disable_gro_hw(struct net_device *dev) 1607 { 1608 dev->wanted_features &= ~NETIF_F_GRO_HW; 1609 netdev_update_features(dev); 1610 1611 if (unlikely(dev->features & NETIF_F_GRO_HW)) 1612 netdev_WARN(dev, "failed to disable GRO_HW!\n"); 1613 } 1614 1615 const char *netdev_cmd_to_name(enum netdev_cmd cmd) 1616 { 1617 #define N(val) \ 1618 case NETDEV_##val: \ 1619 return "NETDEV_" __stringify(val); 1620 switch (cmd) { 1621 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER) 1622 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE) 1623 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE) 1624 N(POST_INIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) N(CHANGEUPPER) 1625 N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) N(BONDING_INFO) 1626 N(PRECHANGEUPPER) N(CHANGELOWERSTATE) N(UDP_TUNNEL_PUSH_INFO) 1627 N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN) 1628 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO) 1629 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO) 1630 N(PRE_CHANGEADDR) N(OFFLOAD_XSTATS_ENABLE) N(OFFLOAD_XSTATS_DISABLE) 1631 N(OFFLOAD_XSTATS_REPORT_USED) N(OFFLOAD_XSTATS_REPORT_DELTA) 1632 } 1633 #undef N 1634 return "UNKNOWN_NETDEV_EVENT"; 1635 } 1636 EXPORT_SYMBOL_GPL(netdev_cmd_to_name); 1637 1638 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, 1639 struct net_device *dev) 1640 { 1641 struct netdev_notifier_info info = { 1642 .dev = dev, 1643 }; 1644 1645 return nb->notifier_call(nb, val, &info); 1646 } 1647 1648 static int call_netdevice_register_notifiers(struct notifier_block *nb, 1649 struct net_device *dev) 1650 { 1651 int err; 1652 1653 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); 1654 err = notifier_to_errno(err); 1655 if (err) 1656 return err; 1657 1658 if (!(dev->flags & IFF_UP)) 1659 return 0; 1660 1661 call_netdevice_notifier(nb, NETDEV_UP, dev); 1662 return 0; 1663 } 1664 1665 static void call_netdevice_unregister_notifiers(struct notifier_block *nb, 1666 struct net_device *dev) 1667 { 1668 if (dev->flags & IFF_UP) { 1669 call_netdevice_notifier(nb, NETDEV_GOING_DOWN, 1670 dev); 1671 call_netdevice_notifier(nb, NETDEV_DOWN, dev); 1672 } 1673 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); 1674 } 1675 1676 static int call_netdevice_register_net_notifiers(struct notifier_block *nb, 1677 struct net *net) 1678 { 1679 struct net_device *dev; 1680 int err; 1681 1682 for_each_netdev(net, dev) { 1683 err = call_netdevice_register_notifiers(nb, dev); 1684 if (err) 1685 goto rollback; 1686 } 1687 return 0; 1688 1689 rollback: 1690 for_each_netdev_continue_reverse(net, dev) 1691 call_netdevice_unregister_notifiers(nb, dev); 1692 return err; 1693 } 1694 1695 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb, 1696 struct net *net) 1697 { 1698 struct net_device *dev; 1699 1700 for_each_netdev(net, dev) 1701 call_netdevice_unregister_notifiers(nb, dev); 1702 } 1703 1704 static int dev_boot_phase = 1; 1705 1706 /** 1707 * register_netdevice_notifier - register a network notifier block 1708 * @nb: notifier 1709 * 1710 * Register a notifier to be called when network device events occur. 1711 * The notifier passed is linked into the kernel structures and must 1712 * not be reused until it has been unregistered. A negative errno code 1713 * is returned on a failure. 1714 * 1715 * When registered all registration and up events are replayed 1716 * to the new notifier to allow device to have a race free 1717 * view of the network device list. 1718 */ 1719 1720 int register_netdevice_notifier(struct notifier_block *nb) 1721 { 1722 struct net *net; 1723 int err; 1724 1725 /* Close race with setup_net() and cleanup_net() */ 1726 down_write(&pernet_ops_rwsem); 1727 rtnl_lock(); 1728 err = raw_notifier_chain_register(&netdev_chain, nb); 1729 if (err) 1730 goto unlock; 1731 if (dev_boot_phase) 1732 goto unlock; 1733 for_each_net(net) { 1734 err = call_netdevice_register_net_notifiers(nb, net); 1735 if (err) 1736 goto rollback; 1737 } 1738 1739 unlock: 1740 rtnl_unlock(); 1741 up_write(&pernet_ops_rwsem); 1742 return err; 1743 1744 rollback: 1745 for_each_net_continue_reverse(net) 1746 call_netdevice_unregister_net_notifiers(nb, net); 1747 1748 raw_notifier_chain_unregister(&netdev_chain, nb); 1749 goto unlock; 1750 } 1751 EXPORT_SYMBOL(register_netdevice_notifier); 1752 1753 /** 1754 * unregister_netdevice_notifier - unregister a network notifier block 1755 * @nb: notifier 1756 * 1757 * Unregister a notifier previously registered by 1758 * register_netdevice_notifier(). The notifier is unlinked into the 1759 * kernel structures and may then be reused. A negative errno code 1760 * is returned on a failure. 1761 * 1762 * After unregistering unregister and down device events are synthesized 1763 * for all devices on the device list to the removed notifier to remove 1764 * the need for special case cleanup code. 1765 */ 1766 1767 int unregister_netdevice_notifier(struct notifier_block *nb) 1768 { 1769 struct net *net; 1770 int err; 1771 1772 /* Close race with setup_net() and cleanup_net() */ 1773 down_write(&pernet_ops_rwsem); 1774 rtnl_lock(); 1775 err = raw_notifier_chain_unregister(&netdev_chain, nb); 1776 if (err) 1777 goto unlock; 1778 1779 for_each_net(net) 1780 call_netdevice_unregister_net_notifiers(nb, net); 1781 1782 unlock: 1783 rtnl_unlock(); 1784 up_write(&pernet_ops_rwsem); 1785 return err; 1786 } 1787 EXPORT_SYMBOL(unregister_netdevice_notifier); 1788 1789 static int __register_netdevice_notifier_net(struct net *net, 1790 struct notifier_block *nb, 1791 bool ignore_call_fail) 1792 { 1793 int err; 1794 1795 err = raw_notifier_chain_register(&net->netdev_chain, nb); 1796 if (err) 1797 return err; 1798 if (dev_boot_phase) 1799 return 0; 1800 1801 err = call_netdevice_register_net_notifiers(nb, net); 1802 if (err && !ignore_call_fail) 1803 goto chain_unregister; 1804 1805 return 0; 1806 1807 chain_unregister: 1808 raw_notifier_chain_unregister(&net->netdev_chain, nb); 1809 return err; 1810 } 1811 1812 static int __unregister_netdevice_notifier_net(struct net *net, 1813 struct notifier_block *nb) 1814 { 1815 int err; 1816 1817 err = raw_notifier_chain_unregister(&net->netdev_chain, nb); 1818 if (err) 1819 return err; 1820 1821 call_netdevice_unregister_net_notifiers(nb, net); 1822 return 0; 1823 } 1824 1825 /** 1826 * register_netdevice_notifier_net - register a per-netns network notifier block 1827 * @net: network namespace 1828 * @nb: notifier 1829 * 1830 * Register a notifier to be called when network device events occur. 1831 * The notifier passed is linked into the kernel structures and must 1832 * not be reused until it has been unregistered. A negative errno code 1833 * is returned on a failure. 1834 * 1835 * When registered all registration and up events are replayed 1836 * to the new notifier to allow device to have a race free 1837 * view of the network device list. 1838 */ 1839 1840 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb) 1841 { 1842 int err; 1843 1844 rtnl_lock(); 1845 err = __register_netdevice_notifier_net(net, nb, false); 1846 rtnl_unlock(); 1847 return err; 1848 } 1849 EXPORT_SYMBOL(register_netdevice_notifier_net); 1850 1851 /** 1852 * unregister_netdevice_notifier_net - unregister a per-netns 1853 * network notifier block 1854 * @net: network namespace 1855 * @nb: notifier 1856 * 1857 * Unregister a notifier previously registered by 1858 * register_netdevice_notifier(). The notifier is unlinked into the 1859 * kernel structures and may then be reused. A negative errno code 1860 * is returned on a failure. 1861 * 1862 * After unregistering unregister and down device events are synthesized 1863 * for all devices on the device list to the removed notifier to remove 1864 * the need for special case cleanup code. 1865 */ 1866 1867 int unregister_netdevice_notifier_net(struct net *net, 1868 struct notifier_block *nb) 1869 { 1870 int err; 1871 1872 rtnl_lock(); 1873 err = __unregister_netdevice_notifier_net(net, nb); 1874 rtnl_unlock(); 1875 return err; 1876 } 1877 EXPORT_SYMBOL(unregister_netdevice_notifier_net); 1878 1879 int register_netdevice_notifier_dev_net(struct net_device *dev, 1880 struct notifier_block *nb, 1881 struct netdev_net_notifier *nn) 1882 { 1883 int err; 1884 1885 rtnl_lock(); 1886 err = __register_netdevice_notifier_net(dev_net(dev), nb, false); 1887 if (!err) { 1888 nn->nb = nb; 1889 list_add(&nn->list, &dev->net_notifier_list); 1890 } 1891 rtnl_unlock(); 1892 return err; 1893 } 1894 EXPORT_SYMBOL(register_netdevice_notifier_dev_net); 1895 1896 int unregister_netdevice_notifier_dev_net(struct net_device *dev, 1897 struct notifier_block *nb, 1898 struct netdev_net_notifier *nn) 1899 { 1900 int err; 1901 1902 rtnl_lock(); 1903 list_del(&nn->list); 1904 err = __unregister_netdevice_notifier_net(dev_net(dev), nb); 1905 rtnl_unlock(); 1906 return err; 1907 } 1908 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net); 1909 1910 static void move_netdevice_notifiers_dev_net(struct net_device *dev, 1911 struct net *net) 1912 { 1913 struct netdev_net_notifier *nn; 1914 1915 list_for_each_entry(nn, &dev->net_notifier_list, list) { 1916 __unregister_netdevice_notifier_net(dev_net(dev), nn->nb); 1917 __register_netdevice_notifier_net(net, nn->nb, true); 1918 } 1919 } 1920 1921 /** 1922 * call_netdevice_notifiers_info - call all network notifier blocks 1923 * @val: value passed unmodified to notifier function 1924 * @info: notifier information data 1925 * 1926 * Call all network notifier blocks. Parameters and return value 1927 * are as for raw_notifier_call_chain(). 1928 */ 1929 1930 static int call_netdevice_notifiers_info(unsigned long val, 1931 struct netdev_notifier_info *info) 1932 { 1933 struct net *net = dev_net(info->dev); 1934 int ret; 1935 1936 ASSERT_RTNL(); 1937 1938 /* Run per-netns notifier block chain first, then run the global one. 1939 * Hopefully, one day, the global one is going to be removed after 1940 * all notifier block registrators get converted to be per-netns. 1941 */ 1942 ret = raw_notifier_call_chain(&net->netdev_chain, val, info); 1943 if (ret & NOTIFY_STOP_MASK) 1944 return ret; 1945 return raw_notifier_call_chain(&netdev_chain, val, info); 1946 } 1947 1948 /** 1949 * call_netdevice_notifiers_info_robust - call per-netns notifier blocks 1950 * for and rollback on error 1951 * @val_up: value passed unmodified to notifier function 1952 * @val_down: value passed unmodified to the notifier function when 1953 * recovering from an error on @val_up 1954 * @info: notifier information data 1955 * 1956 * Call all per-netns network notifier blocks, but not notifier blocks on 1957 * the global notifier chain. Parameters and return value are as for 1958 * raw_notifier_call_chain_robust(). 1959 */ 1960 1961 static int 1962 call_netdevice_notifiers_info_robust(unsigned long val_up, 1963 unsigned long val_down, 1964 struct netdev_notifier_info *info) 1965 { 1966 struct net *net = dev_net(info->dev); 1967 1968 ASSERT_RTNL(); 1969 1970 return raw_notifier_call_chain_robust(&net->netdev_chain, 1971 val_up, val_down, info); 1972 } 1973 1974 static int call_netdevice_notifiers_extack(unsigned long val, 1975 struct net_device *dev, 1976 struct netlink_ext_ack *extack) 1977 { 1978 struct netdev_notifier_info info = { 1979 .dev = dev, 1980 .extack = extack, 1981 }; 1982 1983 return call_netdevice_notifiers_info(val, &info); 1984 } 1985 1986 /** 1987 * call_netdevice_notifiers - call all network notifier blocks 1988 * @val: value passed unmodified to notifier function 1989 * @dev: net_device pointer passed unmodified to notifier function 1990 * 1991 * Call all network notifier blocks. Parameters and return value 1992 * are as for raw_notifier_call_chain(). 1993 */ 1994 1995 int call_netdevice_notifiers(unsigned long val, struct net_device *dev) 1996 { 1997 return call_netdevice_notifiers_extack(val, dev, NULL); 1998 } 1999 EXPORT_SYMBOL(call_netdevice_notifiers); 2000 2001 /** 2002 * call_netdevice_notifiers_mtu - call all network notifier blocks 2003 * @val: value passed unmodified to notifier function 2004 * @dev: net_device pointer passed unmodified to notifier function 2005 * @arg: additional u32 argument passed to the notifier function 2006 * 2007 * Call all network notifier blocks. Parameters and return value 2008 * are as for raw_notifier_call_chain(). 2009 */ 2010 static int call_netdevice_notifiers_mtu(unsigned long val, 2011 struct net_device *dev, u32 arg) 2012 { 2013 struct netdev_notifier_info_ext info = { 2014 .info.dev = dev, 2015 .ext.mtu = arg, 2016 }; 2017 2018 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0); 2019 2020 return call_netdevice_notifiers_info(val, &info.info); 2021 } 2022 2023 #ifdef CONFIG_NET_INGRESS 2024 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key); 2025 2026 void net_inc_ingress_queue(void) 2027 { 2028 static_branch_inc(&ingress_needed_key); 2029 } 2030 EXPORT_SYMBOL_GPL(net_inc_ingress_queue); 2031 2032 void net_dec_ingress_queue(void) 2033 { 2034 static_branch_dec(&ingress_needed_key); 2035 } 2036 EXPORT_SYMBOL_GPL(net_dec_ingress_queue); 2037 #endif 2038 2039 #ifdef CONFIG_NET_EGRESS 2040 static DEFINE_STATIC_KEY_FALSE(egress_needed_key); 2041 2042 void net_inc_egress_queue(void) 2043 { 2044 static_branch_inc(&egress_needed_key); 2045 } 2046 EXPORT_SYMBOL_GPL(net_inc_egress_queue); 2047 2048 void net_dec_egress_queue(void) 2049 { 2050 static_branch_dec(&egress_needed_key); 2051 } 2052 EXPORT_SYMBOL_GPL(net_dec_egress_queue); 2053 #endif 2054 2055 DEFINE_STATIC_KEY_FALSE(netstamp_needed_key); 2056 EXPORT_SYMBOL(netstamp_needed_key); 2057 #ifdef CONFIG_JUMP_LABEL 2058 static atomic_t netstamp_needed_deferred; 2059 static atomic_t netstamp_wanted; 2060 static void netstamp_clear(struct work_struct *work) 2061 { 2062 int deferred = atomic_xchg(&netstamp_needed_deferred, 0); 2063 int wanted; 2064 2065 wanted = atomic_add_return(deferred, &netstamp_wanted); 2066 if (wanted > 0) 2067 static_branch_enable(&netstamp_needed_key); 2068 else 2069 static_branch_disable(&netstamp_needed_key); 2070 } 2071 static DECLARE_WORK(netstamp_work, netstamp_clear); 2072 #endif 2073 2074 void net_enable_timestamp(void) 2075 { 2076 #ifdef CONFIG_JUMP_LABEL 2077 int wanted; 2078 2079 while (1) { 2080 wanted = atomic_read(&netstamp_wanted); 2081 if (wanted <= 0) 2082 break; 2083 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted) 2084 return; 2085 } 2086 atomic_inc(&netstamp_needed_deferred); 2087 schedule_work(&netstamp_work); 2088 #else 2089 static_branch_inc(&netstamp_needed_key); 2090 #endif 2091 } 2092 EXPORT_SYMBOL(net_enable_timestamp); 2093 2094 void net_disable_timestamp(void) 2095 { 2096 #ifdef CONFIG_JUMP_LABEL 2097 int wanted; 2098 2099 while (1) { 2100 wanted = atomic_read(&netstamp_wanted); 2101 if (wanted <= 1) 2102 break; 2103 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted) 2104 return; 2105 } 2106 atomic_dec(&netstamp_needed_deferred); 2107 schedule_work(&netstamp_work); 2108 #else 2109 static_branch_dec(&netstamp_needed_key); 2110 #endif 2111 } 2112 EXPORT_SYMBOL(net_disable_timestamp); 2113 2114 static inline void net_timestamp_set(struct sk_buff *skb) 2115 { 2116 skb->tstamp = 0; 2117 skb->mono_delivery_time = 0; 2118 if (static_branch_unlikely(&netstamp_needed_key)) 2119 skb->tstamp = ktime_get_real(); 2120 } 2121 2122 #define net_timestamp_check(COND, SKB) \ 2123 if (static_branch_unlikely(&netstamp_needed_key)) { \ 2124 if ((COND) && !(SKB)->tstamp) \ 2125 (SKB)->tstamp = ktime_get_real(); \ 2126 } \ 2127 2128 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb) 2129 { 2130 return __is_skb_forwardable(dev, skb, true); 2131 } 2132 EXPORT_SYMBOL_GPL(is_skb_forwardable); 2133 2134 static int __dev_forward_skb2(struct net_device *dev, struct sk_buff *skb, 2135 bool check_mtu) 2136 { 2137 int ret = ____dev_forward_skb(dev, skb, check_mtu); 2138 2139 if (likely(!ret)) { 2140 skb->protocol = eth_type_trans(skb, dev); 2141 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); 2142 } 2143 2144 return ret; 2145 } 2146 2147 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2148 { 2149 return __dev_forward_skb2(dev, skb, true); 2150 } 2151 EXPORT_SYMBOL_GPL(__dev_forward_skb); 2152 2153 /** 2154 * dev_forward_skb - loopback an skb to another netif 2155 * 2156 * @dev: destination network device 2157 * @skb: buffer to forward 2158 * 2159 * return values: 2160 * NET_RX_SUCCESS (no congestion) 2161 * NET_RX_DROP (packet was dropped, but freed) 2162 * 2163 * dev_forward_skb can be used for injecting an skb from the 2164 * start_xmit function of one device into the receive queue 2165 * of another device. 2166 * 2167 * The receiving device may be in another namespace, so 2168 * we have to clear all information in the skb that could 2169 * impact namespace isolation. 2170 */ 2171 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2172 { 2173 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); 2174 } 2175 EXPORT_SYMBOL_GPL(dev_forward_skb); 2176 2177 int dev_forward_skb_nomtu(struct net_device *dev, struct sk_buff *skb) 2178 { 2179 return __dev_forward_skb2(dev, skb, false) ?: netif_rx_internal(skb); 2180 } 2181 2182 static inline int deliver_skb(struct sk_buff *skb, 2183 struct packet_type *pt_prev, 2184 struct net_device *orig_dev) 2185 { 2186 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 2187 return -ENOMEM; 2188 refcount_inc(&skb->users); 2189 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 2190 } 2191 2192 static inline void deliver_ptype_list_skb(struct sk_buff *skb, 2193 struct packet_type **pt, 2194 struct net_device *orig_dev, 2195 __be16 type, 2196 struct list_head *ptype_list) 2197 { 2198 struct packet_type *ptype, *pt_prev = *pt; 2199 2200 list_for_each_entry_rcu(ptype, ptype_list, list) { 2201 if (ptype->type != type) 2202 continue; 2203 if (pt_prev) 2204 deliver_skb(skb, pt_prev, orig_dev); 2205 pt_prev = ptype; 2206 } 2207 *pt = pt_prev; 2208 } 2209 2210 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) 2211 { 2212 if (!ptype->af_packet_priv || !skb->sk) 2213 return false; 2214 2215 if (ptype->id_match) 2216 return ptype->id_match(ptype, skb->sk); 2217 else if ((struct sock *)ptype->af_packet_priv == skb->sk) 2218 return true; 2219 2220 return false; 2221 } 2222 2223 /** 2224 * dev_nit_active - return true if any network interface taps are in use 2225 * 2226 * @dev: network device to check for the presence of taps 2227 */ 2228 bool dev_nit_active(struct net_device *dev) 2229 { 2230 return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all); 2231 } 2232 EXPORT_SYMBOL_GPL(dev_nit_active); 2233 2234 /* 2235 * Support routine. Sends outgoing frames to any network 2236 * taps currently in use. 2237 */ 2238 2239 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) 2240 { 2241 struct packet_type *ptype; 2242 struct sk_buff *skb2 = NULL; 2243 struct packet_type *pt_prev = NULL; 2244 struct list_head *ptype_list = &ptype_all; 2245 2246 rcu_read_lock(); 2247 again: 2248 list_for_each_entry_rcu(ptype, ptype_list, list) { 2249 if (ptype->ignore_outgoing) 2250 continue; 2251 2252 /* Never send packets back to the socket 2253 * they originated from - MvS (miquels@drinkel.ow.org) 2254 */ 2255 if (skb_loop_sk(ptype, skb)) 2256 continue; 2257 2258 if (pt_prev) { 2259 deliver_skb(skb2, pt_prev, skb->dev); 2260 pt_prev = ptype; 2261 continue; 2262 } 2263 2264 /* need to clone skb, done only once */ 2265 skb2 = skb_clone(skb, GFP_ATOMIC); 2266 if (!skb2) 2267 goto out_unlock; 2268 2269 net_timestamp_set(skb2); 2270 2271 /* skb->nh should be correctly 2272 * set by sender, so that the second statement is 2273 * just protection against buggy protocols. 2274 */ 2275 skb_reset_mac_header(skb2); 2276 2277 if (skb_network_header(skb2) < skb2->data || 2278 skb_network_header(skb2) > skb_tail_pointer(skb2)) { 2279 net_crit_ratelimited("protocol %04x is buggy, dev %s\n", 2280 ntohs(skb2->protocol), 2281 dev->name); 2282 skb_reset_network_header(skb2); 2283 } 2284 2285 skb2->transport_header = skb2->network_header; 2286 skb2->pkt_type = PACKET_OUTGOING; 2287 pt_prev = ptype; 2288 } 2289 2290 if (ptype_list == &ptype_all) { 2291 ptype_list = &dev->ptype_all; 2292 goto again; 2293 } 2294 out_unlock: 2295 if (pt_prev) { 2296 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC)) 2297 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); 2298 else 2299 kfree_skb(skb2); 2300 } 2301 rcu_read_unlock(); 2302 } 2303 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit); 2304 2305 /** 2306 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change 2307 * @dev: Network device 2308 * @txq: number of queues available 2309 * 2310 * If real_num_tx_queues is changed the tc mappings may no longer be 2311 * valid. To resolve this verify the tc mapping remains valid and if 2312 * not NULL the mapping. With no priorities mapping to this 2313 * offset/count pair it will no longer be used. In the worst case TC0 2314 * is invalid nothing can be done so disable priority mappings. If is 2315 * expected that drivers will fix this mapping if they can before 2316 * calling netif_set_real_num_tx_queues. 2317 */ 2318 static void netif_setup_tc(struct net_device *dev, unsigned int txq) 2319 { 2320 int i; 2321 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2322 2323 /* If TC0 is invalidated disable TC mapping */ 2324 if (tc->offset + tc->count > txq) { 2325 netdev_warn(dev, "Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); 2326 dev->num_tc = 0; 2327 return; 2328 } 2329 2330 /* Invalidated prio to tc mappings set to TC0 */ 2331 for (i = 1; i < TC_BITMASK + 1; i++) { 2332 int q = netdev_get_prio_tc_map(dev, i); 2333 2334 tc = &dev->tc_to_txq[q]; 2335 if (tc->offset + tc->count > txq) { 2336 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", 2337 i, q); 2338 netdev_set_prio_tc_map(dev, i, 0); 2339 } 2340 } 2341 } 2342 2343 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq) 2344 { 2345 if (dev->num_tc) { 2346 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2347 int i; 2348 2349 /* walk through the TCs and see if it falls into any of them */ 2350 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) { 2351 if ((txq - tc->offset) < tc->count) 2352 return i; 2353 } 2354 2355 /* didn't find it, just return -1 to indicate no match */ 2356 return -1; 2357 } 2358 2359 return 0; 2360 } 2361 EXPORT_SYMBOL(netdev_txq_to_tc); 2362 2363 #ifdef CONFIG_XPS 2364 static struct static_key xps_needed __read_mostly; 2365 static struct static_key xps_rxqs_needed __read_mostly; 2366 static DEFINE_MUTEX(xps_map_mutex); 2367 #define xmap_dereference(P) \ 2368 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) 2369 2370 static bool remove_xps_queue(struct xps_dev_maps *dev_maps, 2371 struct xps_dev_maps *old_maps, int tci, u16 index) 2372 { 2373 struct xps_map *map = NULL; 2374 int pos; 2375 2376 if (dev_maps) 2377 map = xmap_dereference(dev_maps->attr_map[tci]); 2378 if (!map) 2379 return false; 2380 2381 for (pos = map->len; pos--;) { 2382 if (map->queues[pos] != index) 2383 continue; 2384 2385 if (map->len > 1) { 2386 map->queues[pos] = map->queues[--map->len]; 2387 break; 2388 } 2389 2390 if (old_maps) 2391 RCU_INIT_POINTER(old_maps->attr_map[tci], NULL); 2392 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2393 kfree_rcu(map, rcu); 2394 return false; 2395 } 2396 2397 return true; 2398 } 2399 2400 static bool remove_xps_queue_cpu(struct net_device *dev, 2401 struct xps_dev_maps *dev_maps, 2402 int cpu, u16 offset, u16 count) 2403 { 2404 int num_tc = dev_maps->num_tc; 2405 bool active = false; 2406 int tci; 2407 2408 for (tci = cpu * num_tc; num_tc--; tci++) { 2409 int i, j; 2410 2411 for (i = count, j = offset; i--; j++) { 2412 if (!remove_xps_queue(dev_maps, NULL, tci, j)) 2413 break; 2414 } 2415 2416 active |= i < 0; 2417 } 2418 2419 return active; 2420 } 2421 2422 static void reset_xps_maps(struct net_device *dev, 2423 struct xps_dev_maps *dev_maps, 2424 enum xps_map_type type) 2425 { 2426 static_key_slow_dec_cpuslocked(&xps_needed); 2427 if (type == XPS_RXQS) 2428 static_key_slow_dec_cpuslocked(&xps_rxqs_needed); 2429 2430 RCU_INIT_POINTER(dev->xps_maps[type], NULL); 2431 2432 kfree_rcu(dev_maps, rcu); 2433 } 2434 2435 static void clean_xps_maps(struct net_device *dev, enum xps_map_type type, 2436 u16 offset, u16 count) 2437 { 2438 struct xps_dev_maps *dev_maps; 2439 bool active = false; 2440 int i, j; 2441 2442 dev_maps = xmap_dereference(dev->xps_maps[type]); 2443 if (!dev_maps) 2444 return; 2445 2446 for (j = 0; j < dev_maps->nr_ids; j++) 2447 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, count); 2448 if (!active) 2449 reset_xps_maps(dev, dev_maps, type); 2450 2451 if (type == XPS_CPUS) { 2452 for (i = offset + (count - 1); count--; i--) 2453 netdev_queue_numa_node_write( 2454 netdev_get_tx_queue(dev, i), NUMA_NO_NODE); 2455 } 2456 } 2457 2458 static void netif_reset_xps_queues(struct net_device *dev, u16 offset, 2459 u16 count) 2460 { 2461 if (!static_key_false(&xps_needed)) 2462 return; 2463 2464 cpus_read_lock(); 2465 mutex_lock(&xps_map_mutex); 2466 2467 if (static_key_false(&xps_rxqs_needed)) 2468 clean_xps_maps(dev, XPS_RXQS, offset, count); 2469 2470 clean_xps_maps(dev, XPS_CPUS, offset, count); 2471 2472 mutex_unlock(&xps_map_mutex); 2473 cpus_read_unlock(); 2474 } 2475 2476 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) 2477 { 2478 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index); 2479 } 2480 2481 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index, 2482 u16 index, bool is_rxqs_map) 2483 { 2484 struct xps_map *new_map; 2485 int alloc_len = XPS_MIN_MAP_ALLOC; 2486 int i, pos; 2487 2488 for (pos = 0; map && pos < map->len; pos++) { 2489 if (map->queues[pos] != index) 2490 continue; 2491 return map; 2492 } 2493 2494 /* Need to add tx-queue to this CPU's/rx-queue's existing map */ 2495 if (map) { 2496 if (pos < map->alloc_len) 2497 return map; 2498 2499 alloc_len = map->alloc_len * 2; 2500 } 2501 2502 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's 2503 * map 2504 */ 2505 if (is_rxqs_map) 2506 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL); 2507 else 2508 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, 2509 cpu_to_node(attr_index)); 2510 if (!new_map) 2511 return NULL; 2512 2513 for (i = 0; i < pos; i++) 2514 new_map->queues[i] = map->queues[i]; 2515 new_map->alloc_len = alloc_len; 2516 new_map->len = pos; 2517 2518 return new_map; 2519 } 2520 2521 /* Copy xps maps at a given index */ 2522 static void xps_copy_dev_maps(struct xps_dev_maps *dev_maps, 2523 struct xps_dev_maps *new_dev_maps, int index, 2524 int tc, bool skip_tc) 2525 { 2526 int i, tci = index * dev_maps->num_tc; 2527 struct xps_map *map; 2528 2529 /* copy maps belonging to foreign traffic classes */ 2530 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2531 if (i == tc && skip_tc) 2532 continue; 2533 2534 /* fill in the new device map from the old device map */ 2535 map = xmap_dereference(dev_maps->attr_map[tci]); 2536 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2537 } 2538 } 2539 2540 /* Must be called under cpus_read_lock */ 2541 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask, 2542 u16 index, enum xps_map_type type) 2543 { 2544 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL, *old_dev_maps = NULL; 2545 const unsigned long *online_mask = NULL; 2546 bool active = false, copy = false; 2547 int i, j, tci, numa_node_id = -2; 2548 int maps_sz, num_tc = 1, tc = 0; 2549 struct xps_map *map, *new_map; 2550 unsigned int nr_ids; 2551 2552 if (dev->num_tc) { 2553 /* Do not allow XPS on subordinate device directly */ 2554 num_tc = dev->num_tc; 2555 if (num_tc < 0) 2556 return -EINVAL; 2557 2558 /* If queue belongs to subordinate dev use its map */ 2559 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev; 2560 2561 tc = netdev_txq_to_tc(dev, index); 2562 if (tc < 0) 2563 return -EINVAL; 2564 } 2565 2566 mutex_lock(&xps_map_mutex); 2567 2568 dev_maps = xmap_dereference(dev->xps_maps[type]); 2569 if (type == XPS_RXQS) { 2570 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues); 2571 nr_ids = dev->num_rx_queues; 2572 } else { 2573 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc); 2574 if (num_possible_cpus() > 1) 2575 online_mask = cpumask_bits(cpu_online_mask); 2576 nr_ids = nr_cpu_ids; 2577 } 2578 2579 if (maps_sz < L1_CACHE_BYTES) 2580 maps_sz = L1_CACHE_BYTES; 2581 2582 /* The old dev_maps could be larger or smaller than the one we're 2583 * setting up now, as dev->num_tc or nr_ids could have been updated in 2584 * between. We could try to be smart, but let's be safe instead and only 2585 * copy foreign traffic classes if the two map sizes match. 2586 */ 2587 if (dev_maps && 2588 dev_maps->num_tc == num_tc && dev_maps->nr_ids == nr_ids) 2589 copy = true; 2590 2591 /* allocate memory for queue storage */ 2592 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids), 2593 j < nr_ids;) { 2594 if (!new_dev_maps) { 2595 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); 2596 if (!new_dev_maps) { 2597 mutex_unlock(&xps_map_mutex); 2598 return -ENOMEM; 2599 } 2600 2601 new_dev_maps->nr_ids = nr_ids; 2602 new_dev_maps->num_tc = num_tc; 2603 } 2604 2605 tci = j * num_tc + tc; 2606 map = copy ? xmap_dereference(dev_maps->attr_map[tci]) : NULL; 2607 2608 map = expand_xps_map(map, j, index, type == XPS_RXQS); 2609 if (!map) 2610 goto error; 2611 2612 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2613 } 2614 2615 if (!new_dev_maps) 2616 goto out_no_new_maps; 2617 2618 if (!dev_maps) { 2619 /* Increment static keys at most once per type */ 2620 static_key_slow_inc_cpuslocked(&xps_needed); 2621 if (type == XPS_RXQS) 2622 static_key_slow_inc_cpuslocked(&xps_rxqs_needed); 2623 } 2624 2625 for (j = 0; j < nr_ids; j++) { 2626 bool skip_tc = false; 2627 2628 tci = j * num_tc + tc; 2629 if (netif_attr_test_mask(j, mask, nr_ids) && 2630 netif_attr_test_online(j, online_mask, nr_ids)) { 2631 /* add tx-queue to CPU/rx-queue maps */ 2632 int pos = 0; 2633 2634 skip_tc = true; 2635 2636 map = xmap_dereference(new_dev_maps->attr_map[tci]); 2637 while ((pos < map->len) && (map->queues[pos] != index)) 2638 pos++; 2639 2640 if (pos == map->len) 2641 map->queues[map->len++] = index; 2642 #ifdef CONFIG_NUMA 2643 if (type == XPS_CPUS) { 2644 if (numa_node_id == -2) 2645 numa_node_id = cpu_to_node(j); 2646 else if (numa_node_id != cpu_to_node(j)) 2647 numa_node_id = -1; 2648 } 2649 #endif 2650 } 2651 2652 if (copy) 2653 xps_copy_dev_maps(dev_maps, new_dev_maps, j, tc, 2654 skip_tc); 2655 } 2656 2657 rcu_assign_pointer(dev->xps_maps[type], new_dev_maps); 2658 2659 /* Cleanup old maps */ 2660 if (!dev_maps) 2661 goto out_no_old_maps; 2662 2663 for (j = 0; j < dev_maps->nr_ids; j++) { 2664 for (i = num_tc, tci = j * dev_maps->num_tc; i--; tci++) { 2665 map = xmap_dereference(dev_maps->attr_map[tci]); 2666 if (!map) 2667 continue; 2668 2669 if (copy) { 2670 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2671 if (map == new_map) 2672 continue; 2673 } 2674 2675 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2676 kfree_rcu(map, rcu); 2677 } 2678 } 2679 2680 old_dev_maps = dev_maps; 2681 2682 out_no_old_maps: 2683 dev_maps = new_dev_maps; 2684 active = true; 2685 2686 out_no_new_maps: 2687 if (type == XPS_CPUS) 2688 /* update Tx queue numa node */ 2689 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), 2690 (numa_node_id >= 0) ? 2691 numa_node_id : NUMA_NO_NODE); 2692 2693 if (!dev_maps) 2694 goto out_no_maps; 2695 2696 /* removes tx-queue from unused CPUs/rx-queues */ 2697 for (j = 0; j < dev_maps->nr_ids; j++) { 2698 tci = j * dev_maps->num_tc; 2699 2700 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2701 if (i == tc && 2702 netif_attr_test_mask(j, mask, dev_maps->nr_ids) && 2703 netif_attr_test_online(j, online_mask, dev_maps->nr_ids)) 2704 continue; 2705 2706 active |= remove_xps_queue(dev_maps, 2707 copy ? old_dev_maps : NULL, 2708 tci, index); 2709 } 2710 } 2711 2712 if (old_dev_maps) 2713 kfree_rcu(old_dev_maps, rcu); 2714 2715 /* free map if not active */ 2716 if (!active) 2717 reset_xps_maps(dev, dev_maps, type); 2718 2719 out_no_maps: 2720 mutex_unlock(&xps_map_mutex); 2721 2722 return 0; 2723 error: 2724 /* remove any maps that we added */ 2725 for (j = 0; j < nr_ids; j++) { 2726 for (i = num_tc, tci = j * num_tc; i--; tci++) { 2727 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2728 map = copy ? 2729 xmap_dereference(dev_maps->attr_map[tci]) : 2730 NULL; 2731 if (new_map && new_map != map) 2732 kfree(new_map); 2733 } 2734 } 2735 2736 mutex_unlock(&xps_map_mutex); 2737 2738 kfree(new_dev_maps); 2739 return -ENOMEM; 2740 } 2741 EXPORT_SYMBOL_GPL(__netif_set_xps_queue); 2742 2743 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, 2744 u16 index) 2745 { 2746 int ret; 2747 2748 cpus_read_lock(); 2749 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, XPS_CPUS); 2750 cpus_read_unlock(); 2751 2752 return ret; 2753 } 2754 EXPORT_SYMBOL(netif_set_xps_queue); 2755 2756 #endif 2757 static void netdev_unbind_all_sb_channels(struct net_device *dev) 2758 { 2759 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2760 2761 /* Unbind any subordinate channels */ 2762 while (txq-- != &dev->_tx[0]) { 2763 if (txq->sb_dev) 2764 netdev_unbind_sb_channel(dev, txq->sb_dev); 2765 } 2766 } 2767 2768 void netdev_reset_tc(struct net_device *dev) 2769 { 2770 #ifdef CONFIG_XPS 2771 netif_reset_xps_queues_gt(dev, 0); 2772 #endif 2773 netdev_unbind_all_sb_channels(dev); 2774 2775 /* Reset TC configuration of device */ 2776 dev->num_tc = 0; 2777 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq)); 2778 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map)); 2779 } 2780 EXPORT_SYMBOL(netdev_reset_tc); 2781 2782 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset) 2783 { 2784 if (tc >= dev->num_tc) 2785 return -EINVAL; 2786 2787 #ifdef CONFIG_XPS 2788 netif_reset_xps_queues(dev, offset, count); 2789 #endif 2790 dev->tc_to_txq[tc].count = count; 2791 dev->tc_to_txq[tc].offset = offset; 2792 return 0; 2793 } 2794 EXPORT_SYMBOL(netdev_set_tc_queue); 2795 2796 int netdev_set_num_tc(struct net_device *dev, u8 num_tc) 2797 { 2798 if (num_tc > TC_MAX_QUEUE) 2799 return -EINVAL; 2800 2801 #ifdef CONFIG_XPS 2802 netif_reset_xps_queues_gt(dev, 0); 2803 #endif 2804 netdev_unbind_all_sb_channels(dev); 2805 2806 dev->num_tc = num_tc; 2807 return 0; 2808 } 2809 EXPORT_SYMBOL(netdev_set_num_tc); 2810 2811 void netdev_unbind_sb_channel(struct net_device *dev, 2812 struct net_device *sb_dev) 2813 { 2814 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2815 2816 #ifdef CONFIG_XPS 2817 netif_reset_xps_queues_gt(sb_dev, 0); 2818 #endif 2819 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq)); 2820 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map)); 2821 2822 while (txq-- != &dev->_tx[0]) { 2823 if (txq->sb_dev == sb_dev) 2824 txq->sb_dev = NULL; 2825 } 2826 } 2827 EXPORT_SYMBOL(netdev_unbind_sb_channel); 2828 2829 int netdev_bind_sb_channel_queue(struct net_device *dev, 2830 struct net_device *sb_dev, 2831 u8 tc, u16 count, u16 offset) 2832 { 2833 /* Make certain the sb_dev and dev are already configured */ 2834 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc) 2835 return -EINVAL; 2836 2837 /* We cannot hand out queues we don't have */ 2838 if ((offset + count) > dev->real_num_tx_queues) 2839 return -EINVAL; 2840 2841 /* Record the mapping */ 2842 sb_dev->tc_to_txq[tc].count = count; 2843 sb_dev->tc_to_txq[tc].offset = offset; 2844 2845 /* Provide a way for Tx queue to find the tc_to_txq map or 2846 * XPS map for itself. 2847 */ 2848 while (count--) 2849 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev; 2850 2851 return 0; 2852 } 2853 EXPORT_SYMBOL(netdev_bind_sb_channel_queue); 2854 2855 int netdev_set_sb_channel(struct net_device *dev, u16 channel) 2856 { 2857 /* Do not use a multiqueue device to represent a subordinate channel */ 2858 if (netif_is_multiqueue(dev)) 2859 return -ENODEV; 2860 2861 /* We allow channels 1 - 32767 to be used for subordinate channels. 2862 * Channel 0 is meant to be "native" mode and used only to represent 2863 * the main root device. We allow writing 0 to reset the device back 2864 * to normal mode after being used as a subordinate channel. 2865 */ 2866 if (channel > S16_MAX) 2867 return -EINVAL; 2868 2869 dev->num_tc = -channel; 2870 2871 return 0; 2872 } 2873 EXPORT_SYMBOL(netdev_set_sb_channel); 2874 2875 /* 2876 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues 2877 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed. 2878 */ 2879 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) 2880 { 2881 bool disabling; 2882 int rc; 2883 2884 disabling = txq < dev->real_num_tx_queues; 2885 2886 if (txq < 1 || txq > dev->num_tx_queues) 2887 return -EINVAL; 2888 2889 if (dev->reg_state == NETREG_REGISTERED || 2890 dev->reg_state == NETREG_UNREGISTERING) { 2891 ASSERT_RTNL(); 2892 2893 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, 2894 txq); 2895 if (rc) 2896 return rc; 2897 2898 if (dev->num_tc) 2899 netif_setup_tc(dev, txq); 2900 2901 dev_qdisc_change_real_num_tx(dev, txq); 2902 2903 dev->real_num_tx_queues = txq; 2904 2905 if (disabling) { 2906 synchronize_net(); 2907 qdisc_reset_all_tx_gt(dev, txq); 2908 #ifdef CONFIG_XPS 2909 netif_reset_xps_queues_gt(dev, txq); 2910 #endif 2911 } 2912 } else { 2913 dev->real_num_tx_queues = txq; 2914 } 2915 2916 return 0; 2917 } 2918 EXPORT_SYMBOL(netif_set_real_num_tx_queues); 2919 2920 #ifdef CONFIG_SYSFS 2921 /** 2922 * netif_set_real_num_rx_queues - set actual number of RX queues used 2923 * @dev: Network device 2924 * @rxq: Actual number of RX queues 2925 * 2926 * This must be called either with the rtnl_lock held or before 2927 * registration of the net device. Returns 0 on success, or a 2928 * negative error code. If called before registration, it always 2929 * succeeds. 2930 */ 2931 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) 2932 { 2933 int rc; 2934 2935 if (rxq < 1 || rxq > dev->num_rx_queues) 2936 return -EINVAL; 2937 2938 if (dev->reg_state == NETREG_REGISTERED) { 2939 ASSERT_RTNL(); 2940 2941 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, 2942 rxq); 2943 if (rc) 2944 return rc; 2945 } 2946 2947 dev->real_num_rx_queues = rxq; 2948 return 0; 2949 } 2950 EXPORT_SYMBOL(netif_set_real_num_rx_queues); 2951 #endif 2952 2953 /** 2954 * netif_set_real_num_queues - set actual number of RX and TX queues used 2955 * @dev: Network device 2956 * @txq: Actual number of TX queues 2957 * @rxq: Actual number of RX queues 2958 * 2959 * Set the real number of both TX and RX queues. 2960 * Does nothing if the number of queues is already correct. 2961 */ 2962 int netif_set_real_num_queues(struct net_device *dev, 2963 unsigned int txq, unsigned int rxq) 2964 { 2965 unsigned int old_rxq = dev->real_num_rx_queues; 2966 int err; 2967 2968 if (txq < 1 || txq > dev->num_tx_queues || 2969 rxq < 1 || rxq > dev->num_rx_queues) 2970 return -EINVAL; 2971 2972 /* Start from increases, so the error path only does decreases - 2973 * decreases can't fail. 2974 */ 2975 if (rxq > dev->real_num_rx_queues) { 2976 err = netif_set_real_num_rx_queues(dev, rxq); 2977 if (err) 2978 return err; 2979 } 2980 if (txq > dev->real_num_tx_queues) { 2981 err = netif_set_real_num_tx_queues(dev, txq); 2982 if (err) 2983 goto undo_rx; 2984 } 2985 if (rxq < dev->real_num_rx_queues) 2986 WARN_ON(netif_set_real_num_rx_queues(dev, rxq)); 2987 if (txq < dev->real_num_tx_queues) 2988 WARN_ON(netif_set_real_num_tx_queues(dev, txq)); 2989 2990 return 0; 2991 undo_rx: 2992 WARN_ON(netif_set_real_num_rx_queues(dev, old_rxq)); 2993 return err; 2994 } 2995 EXPORT_SYMBOL(netif_set_real_num_queues); 2996 2997 /** 2998 * netif_set_tso_max_size() - set the max size of TSO frames supported 2999 * @dev: netdev to update 3000 * @size: max skb->len of a TSO frame 3001 * 3002 * Set the limit on the size of TSO super-frames the device can handle. 3003 * Unless explicitly set the stack will assume the value of 3004 * %GSO_LEGACY_MAX_SIZE. 3005 */ 3006 void netif_set_tso_max_size(struct net_device *dev, unsigned int size) 3007 { 3008 dev->tso_max_size = min(GSO_MAX_SIZE, size); 3009 if (size < READ_ONCE(dev->gso_max_size)) 3010 netif_set_gso_max_size(dev, size); 3011 } 3012 EXPORT_SYMBOL(netif_set_tso_max_size); 3013 3014 /** 3015 * netif_set_tso_max_segs() - set the max number of segs supported for TSO 3016 * @dev: netdev to update 3017 * @segs: max number of TCP segments 3018 * 3019 * Set the limit on the number of TCP segments the device can generate from 3020 * a single TSO super-frame. 3021 * Unless explicitly set the stack will assume the value of %GSO_MAX_SEGS. 3022 */ 3023 void netif_set_tso_max_segs(struct net_device *dev, unsigned int segs) 3024 { 3025 dev->tso_max_segs = segs; 3026 if (segs < READ_ONCE(dev->gso_max_segs)) 3027 netif_set_gso_max_segs(dev, segs); 3028 } 3029 EXPORT_SYMBOL(netif_set_tso_max_segs); 3030 3031 /** 3032 * netif_inherit_tso_max() - copy all TSO limits from a lower device to an upper 3033 * @to: netdev to update 3034 * @from: netdev from which to copy the limits 3035 */ 3036 void netif_inherit_tso_max(struct net_device *to, const struct net_device *from) 3037 { 3038 netif_set_tso_max_size(to, from->tso_max_size); 3039 netif_set_tso_max_segs(to, from->tso_max_segs); 3040 } 3041 EXPORT_SYMBOL(netif_inherit_tso_max); 3042 3043 /** 3044 * netif_get_num_default_rss_queues - default number of RSS queues 3045 * 3046 * Default value is the number of physical cores if there are only 1 or 2, or 3047 * divided by 2 if there are more. 3048 */ 3049 int netif_get_num_default_rss_queues(void) 3050 { 3051 cpumask_var_t cpus; 3052 int cpu, count = 0; 3053 3054 if (unlikely(is_kdump_kernel() || !zalloc_cpumask_var(&cpus, GFP_KERNEL))) 3055 return 1; 3056 3057 cpumask_copy(cpus, cpu_online_mask); 3058 for_each_cpu(cpu, cpus) { 3059 ++count; 3060 cpumask_andnot(cpus, cpus, topology_sibling_cpumask(cpu)); 3061 } 3062 free_cpumask_var(cpus); 3063 3064 return count > 2 ? DIV_ROUND_UP(count, 2) : count; 3065 } 3066 EXPORT_SYMBOL(netif_get_num_default_rss_queues); 3067 3068 static void __netif_reschedule(struct Qdisc *q) 3069 { 3070 struct softnet_data *sd; 3071 unsigned long flags; 3072 3073 local_irq_save(flags); 3074 sd = this_cpu_ptr(&softnet_data); 3075 q->next_sched = NULL; 3076 *sd->output_queue_tailp = q; 3077 sd->output_queue_tailp = &q->next_sched; 3078 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3079 local_irq_restore(flags); 3080 } 3081 3082 void __netif_schedule(struct Qdisc *q) 3083 { 3084 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) 3085 __netif_reschedule(q); 3086 } 3087 EXPORT_SYMBOL(__netif_schedule); 3088 3089 struct dev_kfree_skb_cb { 3090 enum skb_free_reason reason; 3091 }; 3092 3093 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) 3094 { 3095 return (struct dev_kfree_skb_cb *)skb->cb; 3096 } 3097 3098 void netif_schedule_queue(struct netdev_queue *txq) 3099 { 3100 rcu_read_lock(); 3101 if (!netif_xmit_stopped(txq)) { 3102 struct Qdisc *q = rcu_dereference(txq->qdisc); 3103 3104 __netif_schedule(q); 3105 } 3106 rcu_read_unlock(); 3107 } 3108 EXPORT_SYMBOL(netif_schedule_queue); 3109 3110 void netif_tx_wake_queue(struct netdev_queue *dev_queue) 3111 { 3112 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { 3113 struct Qdisc *q; 3114 3115 rcu_read_lock(); 3116 q = rcu_dereference(dev_queue->qdisc); 3117 __netif_schedule(q); 3118 rcu_read_unlock(); 3119 } 3120 } 3121 EXPORT_SYMBOL(netif_tx_wake_queue); 3122 3123 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason) 3124 { 3125 unsigned long flags; 3126 3127 if (unlikely(!skb)) 3128 return; 3129 3130 if (likely(refcount_read(&skb->users) == 1)) { 3131 smp_rmb(); 3132 refcount_set(&skb->users, 0); 3133 } else if (likely(!refcount_dec_and_test(&skb->users))) { 3134 return; 3135 } 3136 get_kfree_skb_cb(skb)->reason = reason; 3137 local_irq_save(flags); 3138 skb->next = __this_cpu_read(softnet_data.completion_queue); 3139 __this_cpu_write(softnet_data.completion_queue, skb); 3140 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3141 local_irq_restore(flags); 3142 } 3143 EXPORT_SYMBOL(__dev_kfree_skb_irq); 3144 3145 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason) 3146 { 3147 if (in_hardirq() || irqs_disabled()) 3148 __dev_kfree_skb_irq(skb, reason); 3149 else 3150 dev_kfree_skb(skb); 3151 } 3152 EXPORT_SYMBOL(__dev_kfree_skb_any); 3153 3154 3155 /** 3156 * netif_device_detach - mark device as removed 3157 * @dev: network device 3158 * 3159 * Mark device as removed from system and therefore no longer available. 3160 */ 3161 void netif_device_detach(struct net_device *dev) 3162 { 3163 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && 3164 netif_running(dev)) { 3165 netif_tx_stop_all_queues(dev); 3166 } 3167 } 3168 EXPORT_SYMBOL(netif_device_detach); 3169 3170 /** 3171 * netif_device_attach - mark device as attached 3172 * @dev: network device 3173 * 3174 * Mark device as attached from system and restart if needed. 3175 */ 3176 void netif_device_attach(struct net_device *dev) 3177 { 3178 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && 3179 netif_running(dev)) { 3180 netif_tx_wake_all_queues(dev); 3181 __netdev_watchdog_up(dev); 3182 } 3183 } 3184 EXPORT_SYMBOL(netif_device_attach); 3185 3186 /* 3187 * Returns a Tx hash based on the given packet descriptor a Tx queues' number 3188 * to be used as a distribution range. 3189 */ 3190 static u16 skb_tx_hash(const struct net_device *dev, 3191 const struct net_device *sb_dev, 3192 struct sk_buff *skb) 3193 { 3194 u32 hash; 3195 u16 qoffset = 0; 3196 u16 qcount = dev->real_num_tx_queues; 3197 3198 if (dev->num_tc) { 3199 u8 tc = netdev_get_prio_tc_map(dev, skb->priority); 3200 3201 qoffset = sb_dev->tc_to_txq[tc].offset; 3202 qcount = sb_dev->tc_to_txq[tc].count; 3203 if (unlikely(!qcount)) { 3204 net_warn_ratelimited("%s: invalid qcount, qoffset %u for tc %u\n", 3205 sb_dev->name, qoffset, tc); 3206 qoffset = 0; 3207 qcount = dev->real_num_tx_queues; 3208 } 3209 } 3210 3211 if (skb_rx_queue_recorded(skb)) { 3212 hash = skb_get_rx_queue(skb); 3213 if (hash >= qoffset) 3214 hash -= qoffset; 3215 while (unlikely(hash >= qcount)) 3216 hash -= qcount; 3217 return hash + qoffset; 3218 } 3219 3220 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; 3221 } 3222 3223 static void skb_warn_bad_offload(const struct sk_buff *skb) 3224 { 3225 static const netdev_features_t null_features; 3226 struct net_device *dev = skb->dev; 3227 const char *name = ""; 3228 3229 if (!net_ratelimit()) 3230 return; 3231 3232 if (dev) { 3233 if (dev->dev.parent) 3234 name = dev_driver_string(dev->dev.parent); 3235 else 3236 name = netdev_name(dev); 3237 } 3238 skb_dump(KERN_WARNING, skb, false); 3239 WARN(1, "%s: caps=(%pNF, %pNF)\n", 3240 name, dev ? &dev->features : &null_features, 3241 skb->sk ? &skb->sk->sk_route_caps : &null_features); 3242 } 3243 3244 /* 3245 * Invalidate hardware checksum when packet is to be mangled, and 3246 * complete checksum manually on outgoing path. 3247 */ 3248 int skb_checksum_help(struct sk_buff *skb) 3249 { 3250 __wsum csum; 3251 int ret = 0, offset; 3252 3253 if (skb->ip_summed == CHECKSUM_COMPLETE) 3254 goto out_set_summed; 3255 3256 if (unlikely(skb_is_gso(skb))) { 3257 skb_warn_bad_offload(skb); 3258 return -EINVAL; 3259 } 3260 3261 /* Before computing a checksum, we should make sure no frag could 3262 * be modified by an external entity : checksum could be wrong. 3263 */ 3264 if (skb_has_shared_frag(skb)) { 3265 ret = __skb_linearize(skb); 3266 if (ret) 3267 goto out; 3268 } 3269 3270 offset = skb_checksum_start_offset(skb); 3271 ret = -EINVAL; 3272 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { 3273 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false); 3274 goto out; 3275 } 3276 csum = skb_checksum(skb, offset, skb->len - offset, 0); 3277 3278 offset += skb->csum_offset; 3279 if (WARN_ON_ONCE(offset + sizeof(__sum16) > skb_headlen(skb))) { 3280 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false); 3281 goto out; 3282 } 3283 ret = skb_ensure_writable(skb, offset + sizeof(__sum16)); 3284 if (ret) 3285 goto out; 3286 3287 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0; 3288 out_set_summed: 3289 skb->ip_summed = CHECKSUM_NONE; 3290 out: 3291 return ret; 3292 } 3293 EXPORT_SYMBOL(skb_checksum_help); 3294 3295 int skb_crc32c_csum_help(struct sk_buff *skb) 3296 { 3297 __le32 crc32c_csum; 3298 int ret = 0, offset, start; 3299 3300 if (skb->ip_summed != CHECKSUM_PARTIAL) 3301 goto out; 3302 3303 if (unlikely(skb_is_gso(skb))) 3304 goto out; 3305 3306 /* Before computing a checksum, we should make sure no frag could 3307 * be modified by an external entity : checksum could be wrong. 3308 */ 3309 if (unlikely(skb_has_shared_frag(skb))) { 3310 ret = __skb_linearize(skb); 3311 if (ret) 3312 goto out; 3313 } 3314 start = skb_checksum_start_offset(skb); 3315 offset = start + offsetof(struct sctphdr, checksum); 3316 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { 3317 ret = -EINVAL; 3318 goto out; 3319 } 3320 3321 ret = skb_ensure_writable(skb, offset + sizeof(__le32)); 3322 if (ret) 3323 goto out; 3324 3325 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start, 3326 skb->len - start, ~(__u32)0, 3327 crc32c_csum_stub)); 3328 *(__le32 *)(skb->data + offset) = crc32c_csum; 3329 skb->ip_summed = CHECKSUM_NONE; 3330 skb->csum_not_inet = 0; 3331 out: 3332 return ret; 3333 } 3334 3335 __be16 skb_network_protocol(struct sk_buff *skb, int *depth) 3336 { 3337 __be16 type = skb->protocol; 3338 3339 /* Tunnel gso handlers can set protocol to ethernet. */ 3340 if (type == htons(ETH_P_TEB)) { 3341 struct ethhdr *eth; 3342 3343 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) 3344 return 0; 3345 3346 eth = (struct ethhdr *)skb->data; 3347 type = eth->h_proto; 3348 } 3349 3350 return __vlan_get_protocol(skb, type, depth); 3351 } 3352 3353 /* openvswitch calls this on rx path, so we need a different check. 3354 */ 3355 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path) 3356 { 3357 if (tx_path) 3358 return skb->ip_summed != CHECKSUM_PARTIAL && 3359 skb->ip_summed != CHECKSUM_UNNECESSARY; 3360 3361 return skb->ip_summed == CHECKSUM_NONE; 3362 } 3363 3364 /** 3365 * __skb_gso_segment - Perform segmentation on skb. 3366 * @skb: buffer to segment 3367 * @features: features for the output path (see dev->features) 3368 * @tx_path: whether it is called in TX path 3369 * 3370 * This function segments the given skb and returns a list of segments. 3371 * 3372 * It may return NULL if the skb requires no segmentation. This is 3373 * only possible when GSO is used for verifying header integrity. 3374 * 3375 * Segmentation preserves SKB_GSO_CB_OFFSET bytes of previous skb cb. 3376 */ 3377 struct sk_buff *__skb_gso_segment(struct sk_buff *skb, 3378 netdev_features_t features, bool tx_path) 3379 { 3380 struct sk_buff *segs; 3381 3382 if (unlikely(skb_needs_check(skb, tx_path))) { 3383 int err; 3384 3385 /* We're going to init ->check field in TCP or UDP header */ 3386 err = skb_cow_head(skb, 0); 3387 if (err < 0) 3388 return ERR_PTR(err); 3389 } 3390 3391 /* Only report GSO partial support if it will enable us to 3392 * support segmentation on this frame without needing additional 3393 * work. 3394 */ 3395 if (features & NETIF_F_GSO_PARTIAL) { 3396 netdev_features_t partial_features = NETIF_F_GSO_ROBUST; 3397 struct net_device *dev = skb->dev; 3398 3399 partial_features |= dev->features & dev->gso_partial_features; 3400 if (!skb_gso_ok(skb, features | partial_features)) 3401 features &= ~NETIF_F_GSO_PARTIAL; 3402 } 3403 3404 BUILD_BUG_ON(SKB_GSO_CB_OFFSET + 3405 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb)); 3406 3407 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb); 3408 SKB_GSO_CB(skb)->encap_level = 0; 3409 3410 skb_reset_mac_header(skb); 3411 skb_reset_mac_len(skb); 3412 3413 segs = skb_mac_gso_segment(skb, features); 3414 3415 if (segs != skb && unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs))) 3416 skb_warn_bad_offload(skb); 3417 3418 return segs; 3419 } 3420 EXPORT_SYMBOL(__skb_gso_segment); 3421 3422 /* Take action when hardware reception checksum errors are detected. */ 3423 #ifdef CONFIG_BUG 3424 static void do_netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3425 { 3426 netdev_err(dev, "hw csum failure\n"); 3427 skb_dump(KERN_ERR, skb, true); 3428 dump_stack(); 3429 } 3430 3431 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3432 { 3433 DO_ONCE_LITE(do_netdev_rx_csum_fault, dev, skb); 3434 } 3435 EXPORT_SYMBOL(netdev_rx_csum_fault); 3436 #endif 3437 3438 /* XXX: check that highmem exists at all on the given machine. */ 3439 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) 3440 { 3441 #ifdef CONFIG_HIGHMEM 3442 int i; 3443 3444 if (!(dev->features & NETIF_F_HIGHDMA)) { 3445 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 3446 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 3447 3448 if (PageHighMem(skb_frag_page(frag))) 3449 return 1; 3450 } 3451 } 3452 #endif 3453 return 0; 3454 } 3455 3456 /* If MPLS offload request, verify we are testing hardware MPLS features 3457 * instead of standard features for the netdev. 3458 */ 3459 #if IS_ENABLED(CONFIG_NET_MPLS_GSO) 3460 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3461 netdev_features_t features, 3462 __be16 type) 3463 { 3464 if (eth_p_mpls(type)) 3465 features &= skb->dev->mpls_features; 3466 3467 return features; 3468 } 3469 #else 3470 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3471 netdev_features_t features, 3472 __be16 type) 3473 { 3474 return features; 3475 } 3476 #endif 3477 3478 static netdev_features_t harmonize_features(struct sk_buff *skb, 3479 netdev_features_t features) 3480 { 3481 __be16 type; 3482 3483 type = skb_network_protocol(skb, NULL); 3484 features = net_mpls_features(skb, features, type); 3485 3486 if (skb->ip_summed != CHECKSUM_NONE && 3487 !can_checksum_protocol(features, type)) { 3488 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); 3489 } 3490 if (illegal_highdma(skb->dev, skb)) 3491 features &= ~NETIF_F_SG; 3492 3493 return features; 3494 } 3495 3496 netdev_features_t passthru_features_check(struct sk_buff *skb, 3497 struct net_device *dev, 3498 netdev_features_t features) 3499 { 3500 return features; 3501 } 3502 EXPORT_SYMBOL(passthru_features_check); 3503 3504 static netdev_features_t dflt_features_check(struct sk_buff *skb, 3505 struct net_device *dev, 3506 netdev_features_t features) 3507 { 3508 return vlan_features_check(skb, features); 3509 } 3510 3511 static netdev_features_t gso_features_check(const struct sk_buff *skb, 3512 struct net_device *dev, 3513 netdev_features_t features) 3514 { 3515 u16 gso_segs = skb_shinfo(skb)->gso_segs; 3516 3517 if (gso_segs > READ_ONCE(dev->gso_max_segs)) 3518 return features & ~NETIF_F_GSO_MASK; 3519 3520 if (!skb_shinfo(skb)->gso_type) { 3521 skb_warn_bad_offload(skb); 3522 return features & ~NETIF_F_GSO_MASK; 3523 } 3524 3525 /* Support for GSO partial features requires software 3526 * intervention before we can actually process the packets 3527 * so we need to strip support for any partial features now 3528 * and we can pull them back in after we have partially 3529 * segmented the frame. 3530 */ 3531 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)) 3532 features &= ~dev->gso_partial_features; 3533 3534 /* Make sure to clear the IPv4 ID mangling feature if the 3535 * IPv4 header has the potential to be fragmented. 3536 */ 3537 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { 3538 struct iphdr *iph = skb->encapsulation ? 3539 inner_ip_hdr(skb) : ip_hdr(skb); 3540 3541 if (!(iph->frag_off & htons(IP_DF))) 3542 features &= ~NETIF_F_TSO_MANGLEID; 3543 } 3544 3545 return features; 3546 } 3547 3548 netdev_features_t netif_skb_features(struct sk_buff *skb) 3549 { 3550 struct net_device *dev = skb->dev; 3551 netdev_features_t features = dev->features; 3552 3553 if (skb_is_gso(skb)) 3554 features = gso_features_check(skb, dev, features); 3555 3556 /* If encapsulation offload request, verify we are testing 3557 * hardware encapsulation features instead of standard 3558 * features for the netdev 3559 */ 3560 if (skb->encapsulation) 3561 features &= dev->hw_enc_features; 3562 3563 if (skb_vlan_tagged(skb)) 3564 features = netdev_intersect_features(features, 3565 dev->vlan_features | 3566 NETIF_F_HW_VLAN_CTAG_TX | 3567 NETIF_F_HW_VLAN_STAG_TX); 3568 3569 if (dev->netdev_ops->ndo_features_check) 3570 features &= dev->netdev_ops->ndo_features_check(skb, dev, 3571 features); 3572 else 3573 features &= dflt_features_check(skb, dev, features); 3574 3575 return harmonize_features(skb, features); 3576 } 3577 EXPORT_SYMBOL(netif_skb_features); 3578 3579 static int xmit_one(struct sk_buff *skb, struct net_device *dev, 3580 struct netdev_queue *txq, bool more) 3581 { 3582 unsigned int len; 3583 int rc; 3584 3585 if (dev_nit_active(dev)) 3586 dev_queue_xmit_nit(skb, dev); 3587 3588 len = skb->len; 3589 trace_net_dev_start_xmit(skb, dev); 3590 rc = netdev_start_xmit(skb, dev, txq, more); 3591 trace_net_dev_xmit(skb, rc, dev, len); 3592 3593 return rc; 3594 } 3595 3596 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, 3597 struct netdev_queue *txq, int *ret) 3598 { 3599 struct sk_buff *skb = first; 3600 int rc = NETDEV_TX_OK; 3601 3602 while (skb) { 3603 struct sk_buff *next = skb->next; 3604 3605 skb_mark_not_on_list(skb); 3606 rc = xmit_one(skb, dev, txq, next != NULL); 3607 if (unlikely(!dev_xmit_complete(rc))) { 3608 skb->next = next; 3609 goto out; 3610 } 3611 3612 skb = next; 3613 if (netif_tx_queue_stopped(txq) && skb) { 3614 rc = NETDEV_TX_BUSY; 3615 break; 3616 } 3617 } 3618 3619 out: 3620 *ret = rc; 3621 return skb; 3622 } 3623 3624 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, 3625 netdev_features_t features) 3626 { 3627 if (skb_vlan_tag_present(skb) && 3628 !vlan_hw_offload_capable(features, skb->vlan_proto)) 3629 skb = __vlan_hwaccel_push_inside(skb); 3630 return skb; 3631 } 3632 3633 int skb_csum_hwoffload_help(struct sk_buff *skb, 3634 const netdev_features_t features) 3635 { 3636 if (unlikely(skb_csum_is_sctp(skb))) 3637 return !!(features & NETIF_F_SCTP_CRC) ? 0 : 3638 skb_crc32c_csum_help(skb); 3639 3640 if (features & NETIF_F_HW_CSUM) 3641 return 0; 3642 3643 if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) { 3644 switch (skb->csum_offset) { 3645 case offsetof(struct tcphdr, check): 3646 case offsetof(struct udphdr, check): 3647 return 0; 3648 } 3649 } 3650 3651 return skb_checksum_help(skb); 3652 } 3653 EXPORT_SYMBOL(skb_csum_hwoffload_help); 3654 3655 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again) 3656 { 3657 netdev_features_t features; 3658 3659 features = netif_skb_features(skb); 3660 skb = validate_xmit_vlan(skb, features); 3661 if (unlikely(!skb)) 3662 goto out_null; 3663 3664 skb = sk_validate_xmit_skb(skb, dev); 3665 if (unlikely(!skb)) 3666 goto out_null; 3667 3668 if (netif_needs_gso(skb, features)) { 3669 struct sk_buff *segs; 3670 3671 segs = skb_gso_segment(skb, features); 3672 if (IS_ERR(segs)) { 3673 goto out_kfree_skb; 3674 } else if (segs) { 3675 consume_skb(skb); 3676 skb = segs; 3677 } 3678 } else { 3679 if (skb_needs_linearize(skb, features) && 3680 __skb_linearize(skb)) 3681 goto out_kfree_skb; 3682 3683 /* If packet is not checksummed and device does not 3684 * support checksumming for this protocol, complete 3685 * checksumming here. 3686 */ 3687 if (skb->ip_summed == CHECKSUM_PARTIAL) { 3688 if (skb->encapsulation) 3689 skb_set_inner_transport_header(skb, 3690 skb_checksum_start_offset(skb)); 3691 else 3692 skb_set_transport_header(skb, 3693 skb_checksum_start_offset(skb)); 3694 if (skb_csum_hwoffload_help(skb, features)) 3695 goto out_kfree_skb; 3696 } 3697 } 3698 3699 skb = validate_xmit_xfrm(skb, features, again); 3700 3701 return skb; 3702 3703 out_kfree_skb: 3704 kfree_skb(skb); 3705 out_null: 3706 dev_core_stats_tx_dropped_inc(dev); 3707 return NULL; 3708 } 3709 3710 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again) 3711 { 3712 struct sk_buff *next, *head = NULL, *tail; 3713 3714 for (; skb != NULL; skb = next) { 3715 next = skb->next; 3716 skb_mark_not_on_list(skb); 3717 3718 /* in case skb wont be segmented, point to itself */ 3719 skb->prev = skb; 3720 3721 skb = validate_xmit_skb(skb, dev, again); 3722 if (!skb) 3723 continue; 3724 3725 if (!head) 3726 head = skb; 3727 else 3728 tail->next = skb; 3729 /* If skb was segmented, skb->prev points to 3730 * the last segment. If not, it still contains skb. 3731 */ 3732 tail = skb->prev; 3733 } 3734 return head; 3735 } 3736 EXPORT_SYMBOL_GPL(validate_xmit_skb_list); 3737 3738 static void qdisc_pkt_len_init(struct sk_buff *skb) 3739 { 3740 const struct skb_shared_info *shinfo = skb_shinfo(skb); 3741 3742 qdisc_skb_cb(skb)->pkt_len = skb->len; 3743 3744 /* To get more precise estimation of bytes sent on wire, 3745 * we add to pkt_len the headers size of all segments 3746 */ 3747 if (shinfo->gso_size && skb_transport_header_was_set(skb)) { 3748 unsigned int hdr_len; 3749 u16 gso_segs = shinfo->gso_segs; 3750 3751 /* mac layer + network layer */ 3752 hdr_len = skb_transport_header(skb) - skb_mac_header(skb); 3753 3754 /* + transport layer */ 3755 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { 3756 const struct tcphdr *th; 3757 struct tcphdr _tcphdr; 3758 3759 th = skb_header_pointer(skb, skb_transport_offset(skb), 3760 sizeof(_tcphdr), &_tcphdr); 3761 if (likely(th)) 3762 hdr_len += __tcp_hdrlen(th); 3763 } else { 3764 struct udphdr _udphdr; 3765 3766 if (skb_header_pointer(skb, skb_transport_offset(skb), 3767 sizeof(_udphdr), &_udphdr)) 3768 hdr_len += sizeof(struct udphdr); 3769 } 3770 3771 if (shinfo->gso_type & SKB_GSO_DODGY) 3772 gso_segs = DIV_ROUND_UP(skb->len - hdr_len, 3773 shinfo->gso_size); 3774 3775 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; 3776 } 3777 } 3778 3779 static int dev_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *q, 3780 struct sk_buff **to_free, 3781 struct netdev_queue *txq) 3782 { 3783 int rc; 3784 3785 rc = q->enqueue(skb, q, to_free) & NET_XMIT_MASK; 3786 if (rc == NET_XMIT_SUCCESS) 3787 trace_qdisc_enqueue(q, txq, skb); 3788 return rc; 3789 } 3790 3791 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, 3792 struct net_device *dev, 3793 struct netdev_queue *txq) 3794 { 3795 spinlock_t *root_lock = qdisc_lock(q); 3796 struct sk_buff *to_free = NULL; 3797 bool contended; 3798 int rc; 3799 3800 qdisc_calculate_pkt_len(skb, q); 3801 3802 if (q->flags & TCQ_F_NOLOCK) { 3803 if (q->flags & TCQ_F_CAN_BYPASS && nolock_qdisc_is_empty(q) && 3804 qdisc_run_begin(q)) { 3805 /* Retest nolock_qdisc_is_empty() within the protection 3806 * of q->seqlock to protect from racing with requeuing. 3807 */ 3808 if (unlikely(!nolock_qdisc_is_empty(q))) { 3809 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3810 __qdisc_run(q); 3811 qdisc_run_end(q); 3812 3813 goto no_lock_out; 3814 } 3815 3816 qdisc_bstats_cpu_update(q, skb); 3817 if (sch_direct_xmit(skb, q, dev, txq, NULL, true) && 3818 !nolock_qdisc_is_empty(q)) 3819 __qdisc_run(q); 3820 3821 qdisc_run_end(q); 3822 return NET_XMIT_SUCCESS; 3823 } 3824 3825 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3826 qdisc_run(q); 3827 3828 no_lock_out: 3829 if (unlikely(to_free)) 3830 kfree_skb_list_reason(to_free, 3831 SKB_DROP_REASON_QDISC_DROP); 3832 return rc; 3833 } 3834 3835 /* 3836 * Heuristic to force contended enqueues to serialize on a 3837 * separate lock before trying to get qdisc main lock. 3838 * This permits qdisc->running owner to get the lock more 3839 * often and dequeue packets faster. 3840 * On PREEMPT_RT it is possible to preempt the qdisc owner during xmit 3841 * and then other tasks will only enqueue packets. The packets will be 3842 * sent after the qdisc owner is scheduled again. To prevent this 3843 * scenario the task always serialize on the lock. 3844 */ 3845 contended = qdisc_is_running(q) || IS_ENABLED(CONFIG_PREEMPT_RT); 3846 if (unlikely(contended)) 3847 spin_lock(&q->busylock); 3848 3849 spin_lock(root_lock); 3850 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { 3851 __qdisc_drop(skb, &to_free); 3852 rc = NET_XMIT_DROP; 3853 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && 3854 qdisc_run_begin(q)) { 3855 /* 3856 * This is a work-conserving queue; there are no old skbs 3857 * waiting to be sent out; and the qdisc is not running - 3858 * xmit the skb directly. 3859 */ 3860 3861 qdisc_bstats_update(q, skb); 3862 3863 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { 3864 if (unlikely(contended)) { 3865 spin_unlock(&q->busylock); 3866 contended = false; 3867 } 3868 __qdisc_run(q); 3869 } 3870 3871 qdisc_run_end(q); 3872 rc = NET_XMIT_SUCCESS; 3873 } else { 3874 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3875 if (qdisc_run_begin(q)) { 3876 if (unlikely(contended)) { 3877 spin_unlock(&q->busylock); 3878 contended = false; 3879 } 3880 __qdisc_run(q); 3881 qdisc_run_end(q); 3882 } 3883 } 3884 spin_unlock(root_lock); 3885 if (unlikely(to_free)) 3886 kfree_skb_list_reason(to_free, SKB_DROP_REASON_QDISC_DROP); 3887 if (unlikely(contended)) 3888 spin_unlock(&q->busylock); 3889 return rc; 3890 } 3891 3892 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) 3893 static void skb_update_prio(struct sk_buff *skb) 3894 { 3895 const struct netprio_map *map; 3896 const struct sock *sk; 3897 unsigned int prioidx; 3898 3899 if (skb->priority) 3900 return; 3901 map = rcu_dereference_bh(skb->dev->priomap); 3902 if (!map) 3903 return; 3904 sk = skb_to_full_sk(skb); 3905 if (!sk) 3906 return; 3907 3908 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data); 3909 3910 if (prioidx < map->priomap_len) 3911 skb->priority = map->priomap[prioidx]; 3912 } 3913 #else 3914 #define skb_update_prio(skb) 3915 #endif 3916 3917 /** 3918 * dev_loopback_xmit - loop back @skb 3919 * @net: network namespace this loopback is happening in 3920 * @sk: sk needed to be a netfilter okfn 3921 * @skb: buffer to transmit 3922 */ 3923 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) 3924 { 3925 skb_reset_mac_header(skb); 3926 __skb_pull(skb, skb_network_offset(skb)); 3927 skb->pkt_type = PACKET_LOOPBACK; 3928 if (skb->ip_summed == CHECKSUM_NONE) 3929 skb->ip_summed = CHECKSUM_UNNECESSARY; 3930 DEBUG_NET_WARN_ON_ONCE(!skb_dst(skb)); 3931 skb_dst_force(skb); 3932 netif_rx(skb); 3933 return 0; 3934 } 3935 EXPORT_SYMBOL(dev_loopback_xmit); 3936 3937 #ifdef CONFIG_NET_EGRESS 3938 static struct sk_buff * 3939 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 3940 { 3941 #ifdef CONFIG_NET_CLS_ACT 3942 struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress); 3943 struct tcf_result cl_res; 3944 3945 if (!miniq) 3946 return skb; 3947 3948 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */ 3949 tc_skb_cb(skb)->mru = 0; 3950 tc_skb_cb(skb)->post_ct = false; 3951 mini_qdisc_bstats_cpu_update(miniq, skb); 3952 3953 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) { 3954 case TC_ACT_OK: 3955 case TC_ACT_RECLASSIFY: 3956 skb->tc_index = TC_H_MIN(cl_res.classid); 3957 break; 3958 case TC_ACT_SHOT: 3959 mini_qdisc_qstats_cpu_drop(miniq); 3960 *ret = NET_XMIT_DROP; 3961 kfree_skb_reason(skb, SKB_DROP_REASON_TC_EGRESS); 3962 return NULL; 3963 case TC_ACT_STOLEN: 3964 case TC_ACT_QUEUED: 3965 case TC_ACT_TRAP: 3966 *ret = NET_XMIT_SUCCESS; 3967 consume_skb(skb); 3968 return NULL; 3969 case TC_ACT_REDIRECT: 3970 /* No need to push/pop skb's mac_header here on egress! */ 3971 skb_do_redirect(skb); 3972 *ret = NET_XMIT_SUCCESS; 3973 return NULL; 3974 default: 3975 break; 3976 } 3977 #endif /* CONFIG_NET_CLS_ACT */ 3978 3979 return skb; 3980 } 3981 3982 static struct netdev_queue * 3983 netdev_tx_queue_mapping(struct net_device *dev, struct sk_buff *skb) 3984 { 3985 int qm = skb_get_queue_mapping(skb); 3986 3987 return netdev_get_tx_queue(dev, netdev_cap_txqueue(dev, qm)); 3988 } 3989 3990 static bool netdev_xmit_txqueue_skipped(void) 3991 { 3992 return __this_cpu_read(softnet_data.xmit.skip_txqueue); 3993 } 3994 3995 void netdev_xmit_skip_txqueue(bool skip) 3996 { 3997 __this_cpu_write(softnet_data.xmit.skip_txqueue, skip); 3998 } 3999 EXPORT_SYMBOL_GPL(netdev_xmit_skip_txqueue); 4000 #endif /* CONFIG_NET_EGRESS */ 4001 4002 #ifdef CONFIG_XPS 4003 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb, 4004 struct xps_dev_maps *dev_maps, unsigned int tci) 4005 { 4006 int tc = netdev_get_prio_tc_map(dev, skb->priority); 4007 struct xps_map *map; 4008 int queue_index = -1; 4009 4010 if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids) 4011 return queue_index; 4012 4013 tci *= dev_maps->num_tc; 4014 tci += tc; 4015 4016 map = rcu_dereference(dev_maps->attr_map[tci]); 4017 if (map) { 4018 if (map->len == 1) 4019 queue_index = map->queues[0]; 4020 else 4021 queue_index = map->queues[reciprocal_scale( 4022 skb_get_hash(skb), map->len)]; 4023 if (unlikely(queue_index >= dev->real_num_tx_queues)) 4024 queue_index = -1; 4025 } 4026 return queue_index; 4027 } 4028 #endif 4029 4030 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev, 4031 struct sk_buff *skb) 4032 { 4033 #ifdef CONFIG_XPS 4034 struct xps_dev_maps *dev_maps; 4035 struct sock *sk = skb->sk; 4036 int queue_index = -1; 4037 4038 if (!static_key_false(&xps_needed)) 4039 return -1; 4040 4041 rcu_read_lock(); 4042 if (!static_key_false(&xps_rxqs_needed)) 4043 goto get_cpus_map; 4044 4045 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]); 4046 if (dev_maps) { 4047 int tci = sk_rx_queue_get(sk); 4048 4049 if (tci >= 0) 4050 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4051 tci); 4052 } 4053 4054 get_cpus_map: 4055 if (queue_index < 0) { 4056 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]); 4057 if (dev_maps) { 4058 unsigned int tci = skb->sender_cpu - 1; 4059 4060 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4061 tci); 4062 } 4063 } 4064 rcu_read_unlock(); 4065 4066 return queue_index; 4067 #else 4068 return -1; 4069 #endif 4070 } 4071 4072 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb, 4073 struct net_device *sb_dev) 4074 { 4075 return 0; 4076 } 4077 EXPORT_SYMBOL(dev_pick_tx_zero); 4078 4079 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb, 4080 struct net_device *sb_dev) 4081 { 4082 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues; 4083 } 4084 EXPORT_SYMBOL(dev_pick_tx_cpu_id); 4085 4086 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb, 4087 struct net_device *sb_dev) 4088 { 4089 struct sock *sk = skb->sk; 4090 int queue_index = sk_tx_queue_get(sk); 4091 4092 sb_dev = sb_dev ? : dev; 4093 4094 if (queue_index < 0 || skb->ooo_okay || 4095 queue_index >= dev->real_num_tx_queues) { 4096 int new_index = get_xps_queue(dev, sb_dev, skb); 4097 4098 if (new_index < 0) 4099 new_index = skb_tx_hash(dev, sb_dev, skb); 4100 4101 if (queue_index != new_index && sk && 4102 sk_fullsock(sk) && 4103 rcu_access_pointer(sk->sk_dst_cache)) 4104 sk_tx_queue_set(sk, new_index); 4105 4106 queue_index = new_index; 4107 } 4108 4109 return queue_index; 4110 } 4111 EXPORT_SYMBOL(netdev_pick_tx); 4112 4113 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev, 4114 struct sk_buff *skb, 4115 struct net_device *sb_dev) 4116 { 4117 int queue_index = 0; 4118 4119 #ifdef CONFIG_XPS 4120 u32 sender_cpu = skb->sender_cpu - 1; 4121 4122 if (sender_cpu >= (u32)NR_CPUS) 4123 skb->sender_cpu = raw_smp_processor_id() + 1; 4124 #endif 4125 4126 if (dev->real_num_tx_queues != 1) { 4127 const struct net_device_ops *ops = dev->netdev_ops; 4128 4129 if (ops->ndo_select_queue) 4130 queue_index = ops->ndo_select_queue(dev, skb, sb_dev); 4131 else 4132 queue_index = netdev_pick_tx(dev, skb, sb_dev); 4133 4134 queue_index = netdev_cap_txqueue(dev, queue_index); 4135 } 4136 4137 skb_set_queue_mapping(skb, queue_index); 4138 return netdev_get_tx_queue(dev, queue_index); 4139 } 4140 4141 /** 4142 * __dev_queue_xmit() - transmit a buffer 4143 * @skb: buffer to transmit 4144 * @sb_dev: suboordinate device used for L2 forwarding offload 4145 * 4146 * Queue a buffer for transmission to a network device. The caller must 4147 * have set the device and priority and built the buffer before calling 4148 * this function. The function can be called from an interrupt. 4149 * 4150 * When calling this method, interrupts MUST be enabled. This is because 4151 * the BH enable code must have IRQs enabled so that it will not deadlock. 4152 * 4153 * Regardless of the return value, the skb is consumed, so it is currently 4154 * difficult to retry a send to this method. (You can bump the ref count 4155 * before sending to hold a reference for retry if you are careful.) 4156 * 4157 * Return: 4158 * * 0 - buffer successfully transmitted 4159 * * positive qdisc return code - NET_XMIT_DROP etc. 4160 * * negative errno - other errors 4161 */ 4162 int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev) 4163 { 4164 struct net_device *dev = skb->dev; 4165 struct netdev_queue *txq = NULL; 4166 struct Qdisc *q; 4167 int rc = -ENOMEM; 4168 bool again = false; 4169 4170 skb_reset_mac_header(skb); 4171 skb_assert_len(skb); 4172 4173 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) 4174 __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED); 4175 4176 /* Disable soft irqs for various locks below. Also 4177 * stops preemption for RCU. 4178 */ 4179 rcu_read_lock_bh(); 4180 4181 skb_update_prio(skb); 4182 4183 qdisc_pkt_len_init(skb); 4184 #ifdef CONFIG_NET_CLS_ACT 4185 skb->tc_at_ingress = 0; 4186 #endif 4187 #ifdef CONFIG_NET_EGRESS 4188 if (static_branch_unlikely(&egress_needed_key)) { 4189 if (nf_hook_egress_active()) { 4190 skb = nf_hook_egress(skb, &rc, dev); 4191 if (!skb) 4192 goto out; 4193 } 4194 4195 netdev_xmit_skip_txqueue(false); 4196 4197 nf_skip_egress(skb, true); 4198 skb = sch_handle_egress(skb, &rc, dev); 4199 if (!skb) 4200 goto out; 4201 nf_skip_egress(skb, false); 4202 4203 if (netdev_xmit_txqueue_skipped()) 4204 txq = netdev_tx_queue_mapping(dev, skb); 4205 } 4206 #endif 4207 /* If device/qdisc don't need skb->dst, release it right now while 4208 * its hot in this cpu cache. 4209 */ 4210 if (dev->priv_flags & IFF_XMIT_DST_RELEASE) 4211 skb_dst_drop(skb); 4212 else 4213 skb_dst_force(skb); 4214 4215 if (!txq) 4216 txq = netdev_core_pick_tx(dev, skb, sb_dev); 4217 4218 q = rcu_dereference_bh(txq->qdisc); 4219 4220 trace_net_dev_queue(skb); 4221 if (q->enqueue) { 4222 rc = __dev_xmit_skb(skb, q, dev, txq); 4223 goto out; 4224 } 4225 4226 /* The device has no queue. Common case for software devices: 4227 * loopback, all the sorts of tunnels... 4228 4229 * Really, it is unlikely that netif_tx_lock protection is necessary 4230 * here. (f.e. loopback and IP tunnels are clean ignoring statistics 4231 * counters.) 4232 * However, it is possible, that they rely on protection 4233 * made by us here. 4234 4235 * Check this and shot the lock. It is not prone from deadlocks. 4236 *Either shot noqueue qdisc, it is even simpler 8) 4237 */ 4238 if (dev->flags & IFF_UP) { 4239 int cpu = smp_processor_id(); /* ok because BHs are off */ 4240 4241 /* Other cpus might concurrently change txq->xmit_lock_owner 4242 * to -1 or to their cpu id, but not to our id. 4243 */ 4244 if (READ_ONCE(txq->xmit_lock_owner) != cpu) { 4245 if (dev_xmit_recursion()) 4246 goto recursion_alert; 4247 4248 skb = validate_xmit_skb(skb, dev, &again); 4249 if (!skb) 4250 goto out; 4251 4252 HARD_TX_LOCK(dev, txq, cpu); 4253 4254 if (!netif_xmit_stopped(txq)) { 4255 dev_xmit_recursion_inc(); 4256 skb = dev_hard_start_xmit(skb, dev, txq, &rc); 4257 dev_xmit_recursion_dec(); 4258 if (dev_xmit_complete(rc)) { 4259 HARD_TX_UNLOCK(dev, txq); 4260 goto out; 4261 } 4262 } 4263 HARD_TX_UNLOCK(dev, txq); 4264 net_crit_ratelimited("Virtual device %s asks to queue packet!\n", 4265 dev->name); 4266 } else { 4267 /* Recursion is detected! It is possible, 4268 * unfortunately 4269 */ 4270 recursion_alert: 4271 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", 4272 dev->name); 4273 } 4274 } 4275 4276 rc = -ENETDOWN; 4277 rcu_read_unlock_bh(); 4278 4279 dev_core_stats_tx_dropped_inc(dev); 4280 kfree_skb_list(skb); 4281 return rc; 4282 out: 4283 rcu_read_unlock_bh(); 4284 return rc; 4285 } 4286 EXPORT_SYMBOL(__dev_queue_xmit); 4287 4288 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id) 4289 { 4290 struct net_device *dev = skb->dev; 4291 struct sk_buff *orig_skb = skb; 4292 struct netdev_queue *txq; 4293 int ret = NETDEV_TX_BUSY; 4294 bool again = false; 4295 4296 if (unlikely(!netif_running(dev) || 4297 !netif_carrier_ok(dev))) 4298 goto drop; 4299 4300 skb = validate_xmit_skb_list(skb, dev, &again); 4301 if (skb != orig_skb) 4302 goto drop; 4303 4304 skb_set_queue_mapping(skb, queue_id); 4305 txq = skb_get_tx_queue(dev, skb); 4306 4307 local_bh_disable(); 4308 4309 dev_xmit_recursion_inc(); 4310 HARD_TX_LOCK(dev, txq, smp_processor_id()); 4311 if (!netif_xmit_frozen_or_drv_stopped(txq)) 4312 ret = netdev_start_xmit(skb, dev, txq, false); 4313 HARD_TX_UNLOCK(dev, txq); 4314 dev_xmit_recursion_dec(); 4315 4316 local_bh_enable(); 4317 return ret; 4318 drop: 4319 dev_core_stats_tx_dropped_inc(dev); 4320 kfree_skb_list(skb); 4321 return NET_XMIT_DROP; 4322 } 4323 EXPORT_SYMBOL(__dev_direct_xmit); 4324 4325 /************************************************************************* 4326 * Receiver routines 4327 *************************************************************************/ 4328 4329 int netdev_max_backlog __read_mostly = 1000; 4330 EXPORT_SYMBOL(netdev_max_backlog); 4331 4332 int netdev_tstamp_prequeue __read_mostly = 1; 4333 unsigned int sysctl_skb_defer_max __read_mostly = 64; 4334 int netdev_budget __read_mostly = 300; 4335 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */ 4336 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ; 4337 int weight_p __read_mostly = 64; /* old backlog weight */ 4338 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ 4339 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ 4340 int dev_rx_weight __read_mostly = 64; 4341 int dev_tx_weight __read_mostly = 64; 4342 4343 /* Called with irq disabled */ 4344 static inline void ____napi_schedule(struct softnet_data *sd, 4345 struct napi_struct *napi) 4346 { 4347 struct task_struct *thread; 4348 4349 lockdep_assert_irqs_disabled(); 4350 4351 if (test_bit(NAPI_STATE_THREADED, &napi->state)) { 4352 /* Paired with smp_mb__before_atomic() in 4353 * napi_enable()/dev_set_threaded(). 4354 * Use READ_ONCE() to guarantee a complete 4355 * read on napi->thread. Only call 4356 * wake_up_process() when it's not NULL. 4357 */ 4358 thread = READ_ONCE(napi->thread); 4359 if (thread) { 4360 /* Avoid doing set_bit() if the thread is in 4361 * INTERRUPTIBLE state, cause napi_thread_wait() 4362 * makes sure to proceed with napi polling 4363 * if the thread is explicitly woken from here. 4364 */ 4365 if (READ_ONCE(thread->__state) != TASK_INTERRUPTIBLE) 4366 set_bit(NAPI_STATE_SCHED_THREADED, &napi->state); 4367 wake_up_process(thread); 4368 return; 4369 } 4370 } 4371 4372 list_add_tail(&napi->poll_list, &sd->poll_list); 4373 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4374 } 4375 4376 #ifdef CONFIG_RPS 4377 4378 /* One global table that all flow-based protocols share. */ 4379 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; 4380 EXPORT_SYMBOL(rps_sock_flow_table); 4381 u32 rps_cpu_mask __read_mostly; 4382 EXPORT_SYMBOL(rps_cpu_mask); 4383 4384 struct static_key_false rps_needed __read_mostly; 4385 EXPORT_SYMBOL(rps_needed); 4386 struct static_key_false rfs_needed __read_mostly; 4387 EXPORT_SYMBOL(rfs_needed); 4388 4389 static struct rps_dev_flow * 4390 set_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4391 struct rps_dev_flow *rflow, u16 next_cpu) 4392 { 4393 if (next_cpu < nr_cpu_ids) { 4394 #ifdef CONFIG_RFS_ACCEL 4395 struct netdev_rx_queue *rxqueue; 4396 struct rps_dev_flow_table *flow_table; 4397 struct rps_dev_flow *old_rflow; 4398 u32 flow_id; 4399 u16 rxq_index; 4400 int rc; 4401 4402 /* Should we steer this flow to a different hardware queue? */ 4403 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || 4404 !(dev->features & NETIF_F_NTUPLE)) 4405 goto out; 4406 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); 4407 if (rxq_index == skb_get_rx_queue(skb)) 4408 goto out; 4409 4410 rxqueue = dev->_rx + rxq_index; 4411 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4412 if (!flow_table) 4413 goto out; 4414 flow_id = skb_get_hash(skb) & flow_table->mask; 4415 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, 4416 rxq_index, flow_id); 4417 if (rc < 0) 4418 goto out; 4419 old_rflow = rflow; 4420 rflow = &flow_table->flows[flow_id]; 4421 rflow->filter = rc; 4422 if (old_rflow->filter == rflow->filter) 4423 old_rflow->filter = RPS_NO_FILTER; 4424 out: 4425 #endif 4426 rflow->last_qtail = 4427 per_cpu(softnet_data, next_cpu).input_queue_head; 4428 } 4429 4430 rflow->cpu = next_cpu; 4431 return rflow; 4432 } 4433 4434 /* 4435 * get_rps_cpu is called from netif_receive_skb and returns the target 4436 * CPU from the RPS map of the receiving queue for a given skb. 4437 * rcu_read_lock must be held on entry. 4438 */ 4439 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4440 struct rps_dev_flow **rflowp) 4441 { 4442 const struct rps_sock_flow_table *sock_flow_table; 4443 struct netdev_rx_queue *rxqueue = dev->_rx; 4444 struct rps_dev_flow_table *flow_table; 4445 struct rps_map *map; 4446 int cpu = -1; 4447 u32 tcpu; 4448 u32 hash; 4449 4450 if (skb_rx_queue_recorded(skb)) { 4451 u16 index = skb_get_rx_queue(skb); 4452 4453 if (unlikely(index >= dev->real_num_rx_queues)) { 4454 WARN_ONCE(dev->real_num_rx_queues > 1, 4455 "%s received packet on queue %u, but number " 4456 "of RX queues is %u\n", 4457 dev->name, index, dev->real_num_rx_queues); 4458 goto done; 4459 } 4460 rxqueue += index; 4461 } 4462 4463 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ 4464 4465 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4466 map = rcu_dereference(rxqueue->rps_map); 4467 if (!flow_table && !map) 4468 goto done; 4469 4470 skb_reset_network_header(skb); 4471 hash = skb_get_hash(skb); 4472 if (!hash) 4473 goto done; 4474 4475 sock_flow_table = rcu_dereference(rps_sock_flow_table); 4476 if (flow_table && sock_flow_table) { 4477 struct rps_dev_flow *rflow; 4478 u32 next_cpu; 4479 u32 ident; 4480 4481 /* First check into global flow table if there is a match */ 4482 ident = sock_flow_table->ents[hash & sock_flow_table->mask]; 4483 if ((ident ^ hash) & ~rps_cpu_mask) 4484 goto try_rps; 4485 4486 next_cpu = ident & rps_cpu_mask; 4487 4488 /* OK, now we know there is a match, 4489 * we can look at the local (per receive queue) flow table 4490 */ 4491 rflow = &flow_table->flows[hash & flow_table->mask]; 4492 tcpu = rflow->cpu; 4493 4494 /* 4495 * If the desired CPU (where last recvmsg was done) is 4496 * different from current CPU (one in the rx-queue flow 4497 * table entry), switch if one of the following holds: 4498 * - Current CPU is unset (>= nr_cpu_ids). 4499 * - Current CPU is offline. 4500 * - The current CPU's queue tail has advanced beyond the 4501 * last packet that was enqueued using this table entry. 4502 * This guarantees that all previous packets for the flow 4503 * have been dequeued, thus preserving in order delivery. 4504 */ 4505 if (unlikely(tcpu != next_cpu) && 4506 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || 4507 ((int)(per_cpu(softnet_data, tcpu).input_queue_head - 4508 rflow->last_qtail)) >= 0)) { 4509 tcpu = next_cpu; 4510 rflow = set_rps_cpu(dev, skb, rflow, next_cpu); 4511 } 4512 4513 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { 4514 *rflowp = rflow; 4515 cpu = tcpu; 4516 goto done; 4517 } 4518 } 4519 4520 try_rps: 4521 4522 if (map) { 4523 tcpu = map->cpus[reciprocal_scale(hash, map->len)]; 4524 if (cpu_online(tcpu)) { 4525 cpu = tcpu; 4526 goto done; 4527 } 4528 } 4529 4530 done: 4531 return cpu; 4532 } 4533 4534 #ifdef CONFIG_RFS_ACCEL 4535 4536 /** 4537 * rps_may_expire_flow - check whether an RFS hardware filter may be removed 4538 * @dev: Device on which the filter was set 4539 * @rxq_index: RX queue index 4540 * @flow_id: Flow ID passed to ndo_rx_flow_steer() 4541 * @filter_id: Filter ID returned by ndo_rx_flow_steer() 4542 * 4543 * Drivers that implement ndo_rx_flow_steer() should periodically call 4544 * this function for each installed filter and remove the filters for 4545 * which it returns %true. 4546 */ 4547 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, 4548 u32 flow_id, u16 filter_id) 4549 { 4550 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; 4551 struct rps_dev_flow_table *flow_table; 4552 struct rps_dev_flow *rflow; 4553 bool expire = true; 4554 unsigned int cpu; 4555 4556 rcu_read_lock(); 4557 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4558 if (flow_table && flow_id <= flow_table->mask) { 4559 rflow = &flow_table->flows[flow_id]; 4560 cpu = READ_ONCE(rflow->cpu); 4561 if (rflow->filter == filter_id && cpu < nr_cpu_ids && 4562 ((int)(per_cpu(softnet_data, cpu).input_queue_head - 4563 rflow->last_qtail) < 4564 (int)(10 * flow_table->mask))) 4565 expire = false; 4566 } 4567 rcu_read_unlock(); 4568 return expire; 4569 } 4570 EXPORT_SYMBOL(rps_may_expire_flow); 4571 4572 #endif /* CONFIG_RFS_ACCEL */ 4573 4574 /* Called from hardirq (IPI) context */ 4575 static void rps_trigger_softirq(void *data) 4576 { 4577 struct softnet_data *sd = data; 4578 4579 ____napi_schedule(sd, &sd->backlog); 4580 sd->received_rps++; 4581 } 4582 4583 #endif /* CONFIG_RPS */ 4584 4585 /* Called from hardirq (IPI) context */ 4586 static void trigger_rx_softirq(void *data) 4587 { 4588 struct softnet_data *sd = data; 4589 4590 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4591 smp_store_release(&sd->defer_ipi_scheduled, 0); 4592 } 4593 4594 /* 4595 * Check if this softnet_data structure is another cpu one 4596 * If yes, queue it to our IPI list and return 1 4597 * If no, return 0 4598 */ 4599 static int napi_schedule_rps(struct softnet_data *sd) 4600 { 4601 struct softnet_data *mysd = this_cpu_ptr(&softnet_data); 4602 4603 #ifdef CONFIG_RPS 4604 if (sd != mysd) { 4605 sd->rps_ipi_next = mysd->rps_ipi_list; 4606 mysd->rps_ipi_list = sd; 4607 4608 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4609 return 1; 4610 } 4611 #endif /* CONFIG_RPS */ 4612 __napi_schedule_irqoff(&mysd->backlog); 4613 return 0; 4614 } 4615 4616 #ifdef CONFIG_NET_FLOW_LIMIT 4617 int netdev_flow_limit_table_len __read_mostly = (1 << 12); 4618 #endif 4619 4620 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) 4621 { 4622 #ifdef CONFIG_NET_FLOW_LIMIT 4623 struct sd_flow_limit *fl; 4624 struct softnet_data *sd; 4625 unsigned int old_flow, new_flow; 4626 4627 if (qlen < (READ_ONCE(netdev_max_backlog) >> 1)) 4628 return false; 4629 4630 sd = this_cpu_ptr(&softnet_data); 4631 4632 rcu_read_lock(); 4633 fl = rcu_dereference(sd->flow_limit); 4634 if (fl) { 4635 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); 4636 old_flow = fl->history[fl->history_head]; 4637 fl->history[fl->history_head] = new_flow; 4638 4639 fl->history_head++; 4640 fl->history_head &= FLOW_LIMIT_HISTORY - 1; 4641 4642 if (likely(fl->buckets[old_flow])) 4643 fl->buckets[old_flow]--; 4644 4645 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { 4646 fl->count++; 4647 rcu_read_unlock(); 4648 return true; 4649 } 4650 } 4651 rcu_read_unlock(); 4652 #endif 4653 return false; 4654 } 4655 4656 /* 4657 * enqueue_to_backlog is called to queue an skb to a per CPU backlog 4658 * queue (may be a remote CPU queue). 4659 */ 4660 static int enqueue_to_backlog(struct sk_buff *skb, int cpu, 4661 unsigned int *qtail) 4662 { 4663 enum skb_drop_reason reason; 4664 struct softnet_data *sd; 4665 unsigned long flags; 4666 unsigned int qlen; 4667 4668 reason = SKB_DROP_REASON_NOT_SPECIFIED; 4669 sd = &per_cpu(softnet_data, cpu); 4670 4671 rps_lock_irqsave(sd, &flags); 4672 if (!netif_running(skb->dev)) 4673 goto drop; 4674 qlen = skb_queue_len(&sd->input_pkt_queue); 4675 if (qlen <= READ_ONCE(netdev_max_backlog) && !skb_flow_limit(skb, qlen)) { 4676 if (qlen) { 4677 enqueue: 4678 __skb_queue_tail(&sd->input_pkt_queue, skb); 4679 input_queue_tail_incr_save(sd, qtail); 4680 rps_unlock_irq_restore(sd, &flags); 4681 return NET_RX_SUCCESS; 4682 } 4683 4684 /* Schedule NAPI for backlog device 4685 * We can use non atomic operation since we own the queue lock 4686 */ 4687 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) 4688 napi_schedule_rps(sd); 4689 goto enqueue; 4690 } 4691 reason = SKB_DROP_REASON_CPU_BACKLOG; 4692 4693 drop: 4694 sd->dropped++; 4695 rps_unlock_irq_restore(sd, &flags); 4696 4697 dev_core_stats_rx_dropped_inc(skb->dev); 4698 kfree_skb_reason(skb, reason); 4699 return NET_RX_DROP; 4700 } 4701 4702 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb) 4703 { 4704 struct net_device *dev = skb->dev; 4705 struct netdev_rx_queue *rxqueue; 4706 4707 rxqueue = dev->_rx; 4708 4709 if (skb_rx_queue_recorded(skb)) { 4710 u16 index = skb_get_rx_queue(skb); 4711 4712 if (unlikely(index >= dev->real_num_rx_queues)) { 4713 WARN_ONCE(dev->real_num_rx_queues > 1, 4714 "%s received packet on queue %u, but number " 4715 "of RX queues is %u\n", 4716 dev->name, index, dev->real_num_rx_queues); 4717 4718 return rxqueue; /* Return first rxqueue */ 4719 } 4720 rxqueue += index; 4721 } 4722 return rxqueue; 4723 } 4724 4725 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp, 4726 struct bpf_prog *xdp_prog) 4727 { 4728 void *orig_data, *orig_data_end, *hard_start; 4729 struct netdev_rx_queue *rxqueue; 4730 bool orig_bcast, orig_host; 4731 u32 mac_len, frame_sz; 4732 __be16 orig_eth_type; 4733 struct ethhdr *eth; 4734 u32 metalen, act; 4735 int off; 4736 4737 /* The XDP program wants to see the packet starting at the MAC 4738 * header. 4739 */ 4740 mac_len = skb->data - skb_mac_header(skb); 4741 hard_start = skb->data - skb_headroom(skb); 4742 4743 /* SKB "head" area always have tailroom for skb_shared_info */ 4744 frame_sz = (void *)skb_end_pointer(skb) - hard_start; 4745 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 4746 4747 rxqueue = netif_get_rxqueue(skb); 4748 xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq); 4749 xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len, 4750 skb_headlen(skb) + mac_len, true); 4751 4752 orig_data_end = xdp->data_end; 4753 orig_data = xdp->data; 4754 eth = (struct ethhdr *)xdp->data; 4755 orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr); 4756 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest); 4757 orig_eth_type = eth->h_proto; 4758 4759 act = bpf_prog_run_xdp(xdp_prog, xdp); 4760 4761 /* check if bpf_xdp_adjust_head was used */ 4762 off = xdp->data - orig_data; 4763 if (off) { 4764 if (off > 0) 4765 __skb_pull(skb, off); 4766 else if (off < 0) 4767 __skb_push(skb, -off); 4768 4769 skb->mac_header += off; 4770 skb_reset_network_header(skb); 4771 } 4772 4773 /* check if bpf_xdp_adjust_tail was used */ 4774 off = xdp->data_end - orig_data_end; 4775 if (off != 0) { 4776 skb_set_tail_pointer(skb, xdp->data_end - xdp->data); 4777 skb->len += off; /* positive on grow, negative on shrink */ 4778 } 4779 4780 /* check if XDP changed eth hdr such SKB needs update */ 4781 eth = (struct ethhdr *)xdp->data; 4782 if ((orig_eth_type != eth->h_proto) || 4783 (orig_host != ether_addr_equal_64bits(eth->h_dest, 4784 skb->dev->dev_addr)) || 4785 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) { 4786 __skb_push(skb, ETH_HLEN); 4787 skb->pkt_type = PACKET_HOST; 4788 skb->protocol = eth_type_trans(skb, skb->dev); 4789 } 4790 4791 /* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull 4792 * before calling us again on redirect path. We do not call do_redirect 4793 * as we leave that up to the caller. 4794 * 4795 * Caller is responsible for managing lifetime of skb (i.e. calling 4796 * kfree_skb in response to actions it cannot handle/XDP_DROP). 4797 */ 4798 switch (act) { 4799 case XDP_REDIRECT: 4800 case XDP_TX: 4801 __skb_push(skb, mac_len); 4802 break; 4803 case XDP_PASS: 4804 metalen = xdp->data - xdp->data_meta; 4805 if (metalen) 4806 skb_metadata_set(skb, metalen); 4807 break; 4808 } 4809 4810 return act; 4811 } 4812 4813 static u32 netif_receive_generic_xdp(struct sk_buff *skb, 4814 struct xdp_buff *xdp, 4815 struct bpf_prog *xdp_prog) 4816 { 4817 u32 act = XDP_DROP; 4818 4819 /* Reinjected packets coming from act_mirred or similar should 4820 * not get XDP generic processing. 4821 */ 4822 if (skb_is_redirected(skb)) 4823 return XDP_PASS; 4824 4825 /* XDP packets must be linear and must have sufficient headroom 4826 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also 4827 * native XDP provides, thus we need to do it here as well. 4828 */ 4829 if (skb_cloned(skb) || skb_is_nonlinear(skb) || 4830 skb_headroom(skb) < XDP_PACKET_HEADROOM) { 4831 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb); 4832 int troom = skb->tail + skb->data_len - skb->end; 4833 4834 /* In case we have to go down the path and also linearize, 4835 * then lets do the pskb_expand_head() work just once here. 4836 */ 4837 if (pskb_expand_head(skb, 4838 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0, 4839 troom > 0 ? troom + 128 : 0, GFP_ATOMIC)) 4840 goto do_drop; 4841 if (skb_linearize(skb)) 4842 goto do_drop; 4843 } 4844 4845 act = bpf_prog_run_generic_xdp(skb, xdp, xdp_prog); 4846 switch (act) { 4847 case XDP_REDIRECT: 4848 case XDP_TX: 4849 case XDP_PASS: 4850 break; 4851 default: 4852 bpf_warn_invalid_xdp_action(skb->dev, xdp_prog, act); 4853 fallthrough; 4854 case XDP_ABORTED: 4855 trace_xdp_exception(skb->dev, xdp_prog, act); 4856 fallthrough; 4857 case XDP_DROP: 4858 do_drop: 4859 kfree_skb(skb); 4860 break; 4861 } 4862 4863 return act; 4864 } 4865 4866 /* When doing generic XDP we have to bypass the qdisc layer and the 4867 * network taps in order to match in-driver-XDP behavior. This also means 4868 * that XDP packets are able to starve other packets going through a qdisc, 4869 * and DDOS attacks will be more effective. In-driver-XDP use dedicated TX 4870 * queues, so they do not have this starvation issue. 4871 */ 4872 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog) 4873 { 4874 struct net_device *dev = skb->dev; 4875 struct netdev_queue *txq; 4876 bool free_skb = true; 4877 int cpu, rc; 4878 4879 txq = netdev_core_pick_tx(dev, skb, NULL); 4880 cpu = smp_processor_id(); 4881 HARD_TX_LOCK(dev, txq, cpu); 4882 if (!netif_xmit_frozen_or_drv_stopped(txq)) { 4883 rc = netdev_start_xmit(skb, dev, txq, 0); 4884 if (dev_xmit_complete(rc)) 4885 free_skb = false; 4886 } 4887 HARD_TX_UNLOCK(dev, txq); 4888 if (free_skb) { 4889 trace_xdp_exception(dev, xdp_prog, XDP_TX); 4890 dev_core_stats_tx_dropped_inc(dev); 4891 kfree_skb(skb); 4892 } 4893 } 4894 4895 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key); 4896 4897 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb) 4898 { 4899 if (xdp_prog) { 4900 struct xdp_buff xdp; 4901 u32 act; 4902 int err; 4903 4904 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog); 4905 if (act != XDP_PASS) { 4906 switch (act) { 4907 case XDP_REDIRECT: 4908 err = xdp_do_generic_redirect(skb->dev, skb, 4909 &xdp, xdp_prog); 4910 if (err) 4911 goto out_redir; 4912 break; 4913 case XDP_TX: 4914 generic_xdp_tx(skb, xdp_prog); 4915 break; 4916 } 4917 return XDP_DROP; 4918 } 4919 } 4920 return XDP_PASS; 4921 out_redir: 4922 kfree_skb_reason(skb, SKB_DROP_REASON_XDP); 4923 return XDP_DROP; 4924 } 4925 EXPORT_SYMBOL_GPL(do_xdp_generic); 4926 4927 static int netif_rx_internal(struct sk_buff *skb) 4928 { 4929 int ret; 4930 4931 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb); 4932 4933 trace_netif_rx(skb); 4934 4935 #ifdef CONFIG_RPS 4936 if (static_branch_unlikely(&rps_needed)) { 4937 struct rps_dev_flow voidflow, *rflow = &voidflow; 4938 int cpu; 4939 4940 rcu_read_lock(); 4941 4942 cpu = get_rps_cpu(skb->dev, skb, &rflow); 4943 if (cpu < 0) 4944 cpu = smp_processor_id(); 4945 4946 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 4947 4948 rcu_read_unlock(); 4949 } else 4950 #endif 4951 { 4952 unsigned int qtail; 4953 4954 ret = enqueue_to_backlog(skb, smp_processor_id(), &qtail); 4955 } 4956 return ret; 4957 } 4958 4959 /** 4960 * __netif_rx - Slightly optimized version of netif_rx 4961 * @skb: buffer to post 4962 * 4963 * This behaves as netif_rx except that it does not disable bottom halves. 4964 * As a result this function may only be invoked from the interrupt context 4965 * (either hard or soft interrupt). 4966 */ 4967 int __netif_rx(struct sk_buff *skb) 4968 { 4969 int ret; 4970 4971 lockdep_assert_once(hardirq_count() | softirq_count()); 4972 4973 trace_netif_rx_entry(skb); 4974 ret = netif_rx_internal(skb); 4975 trace_netif_rx_exit(ret); 4976 return ret; 4977 } 4978 EXPORT_SYMBOL(__netif_rx); 4979 4980 /** 4981 * netif_rx - post buffer to the network code 4982 * @skb: buffer to post 4983 * 4984 * This function receives a packet from a device driver and queues it for 4985 * the upper (protocol) levels to process via the backlog NAPI device. It 4986 * always succeeds. The buffer may be dropped during processing for 4987 * congestion control or by the protocol layers. 4988 * The network buffer is passed via the backlog NAPI device. Modern NIC 4989 * driver should use NAPI and GRO. 4990 * This function can used from interrupt and from process context. The 4991 * caller from process context must not disable interrupts before invoking 4992 * this function. 4993 * 4994 * return values: 4995 * NET_RX_SUCCESS (no congestion) 4996 * NET_RX_DROP (packet was dropped) 4997 * 4998 */ 4999 int netif_rx(struct sk_buff *skb) 5000 { 5001 bool need_bh_off = !(hardirq_count() | softirq_count()); 5002 int ret; 5003 5004 if (need_bh_off) 5005 local_bh_disable(); 5006 trace_netif_rx_entry(skb); 5007 ret = netif_rx_internal(skb); 5008 trace_netif_rx_exit(ret); 5009 if (need_bh_off) 5010 local_bh_enable(); 5011 return ret; 5012 } 5013 EXPORT_SYMBOL(netif_rx); 5014 5015 static __latent_entropy void net_tx_action(struct softirq_action *h) 5016 { 5017 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 5018 5019 if (sd->completion_queue) { 5020 struct sk_buff *clist; 5021 5022 local_irq_disable(); 5023 clist = sd->completion_queue; 5024 sd->completion_queue = NULL; 5025 local_irq_enable(); 5026 5027 while (clist) { 5028 struct sk_buff *skb = clist; 5029 5030 clist = clist->next; 5031 5032 WARN_ON(refcount_read(&skb->users)); 5033 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED)) 5034 trace_consume_skb(skb); 5035 else 5036 trace_kfree_skb(skb, net_tx_action, 5037 SKB_DROP_REASON_NOT_SPECIFIED); 5038 5039 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) 5040 __kfree_skb(skb); 5041 else 5042 __kfree_skb_defer(skb); 5043 } 5044 } 5045 5046 if (sd->output_queue) { 5047 struct Qdisc *head; 5048 5049 local_irq_disable(); 5050 head = sd->output_queue; 5051 sd->output_queue = NULL; 5052 sd->output_queue_tailp = &sd->output_queue; 5053 local_irq_enable(); 5054 5055 rcu_read_lock(); 5056 5057 while (head) { 5058 struct Qdisc *q = head; 5059 spinlock_t *root_lock = NULL; 5060 5061 head = head->next_sched; 5062 5063 /* We need to make sure head->next_sched is read 5064 * before clearing __QDISC_STATE_SCHED 5065 */ 5066 smp_mb__before_atomic(); 5067 5068 if (!(q->flags & TCQ_F_NOLOCK)) { 5069 root_lock = qdisc_lock(q); 5070 spin_lock(root_lock); 5071 } else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, 5072 &q->state))) { 5073 /* There is a synchronize_net() between 5074 * STATE_DEACTIVATED flag being set and 5075 * qdisc_reset()/some_qdisc_is_busy() in 5076 * dev_deactivate(), so we can safely bail out 5077 * early here to avoid data race between 5078 * qdisc_deactivate() and some_qdisc_is_busy() 5079 * for lockless qdisc. 5080 */ 5081 clear_bit(__QDISC_STATE_SCHED, &q->state); 5082 continue; 5083 } 5084 5085 clear_bit(__QDISC_STATE_SCHED, &q->state); 5086 qdisc_run(q); 5087 if (root_lock) 5088 spin_unlock(root_lock); 5089 } 5090 5091 rcu_read_unlock(); 5092 } 5093 5094 xfrm_dev_backlog(sd); 5095 } 5096 5097 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) 5098 /* This hook is defined here for ATM LANE */ 5099 int (*br_fdb_test_addr_hook)(struct net_device *dev, 5100 unsigned char *addr) __read_mostly; 5101 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); 5102 #endif 5103 5104 static inline struct sk_buff * 5105 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 5106 struct net_device *orig_dev, bool *another) 5107 { 5108 #ifdef CONFIG_NET_CLS_ACT 5109 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress); 5110 struct tcf_result cl_res; 5111 5112 /* If there's at least one ingress present somewhere (so 5113 * we get here via enabled static key), remaining devices 5114 * that are not configured with an ingress qdisc will bail 5115 * out here. 5116 */ 5117 if (!miniq) 5118 return skb; 5119 5120 if (*pt_prev) { 5121 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5122 *pt_prev = NULL; 5123 } 5124 5125 qdisc_skb_cb(skb)->pkt_len = skb->len; 5126 tc_skb_cb(skb)->mru = 0; 5127 tc_skb_cb(skb)->post_ct = false; 5128 skb->tc_at_ingress = 1; 5129 mini_qdisc_bstats_cpu_update(miniq, skb); 5130 5131 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) { 5132 case TC_ACT_OK: 5133 case TC_ACT_RECLASSIFY: 5134 skb->tc_index = TC_H_MIN(cl_res.classid); 5135 break; 5136 case TC_ACT_SHOT: 5137 mini_qdisc_qstats_cpu_drop(miniq); 5138 kfree_skb_reason(skb, SKB_DROP_REASON_TC_INGRESS); 5139 return NULL; 5140 case TC_ACT_STOLEN: 5141 case TC_ACT_QUEUED: 5142 case TC_ACT_TRAP: 5143 consume_skb(skb); 5144 return NULL; 5145 case TC_ACT_REDIRECT: 5146 /* skb_mac_header check was done by cls/act_bpf, so 5147 * we can safely push the L2 header back before 5148 * redirecting to another netdev 5149 */ 5150 __skb_push(skb, skb->mac_len); 5151 if (skb_do_redirect(skb) == -EAGAIN) { 5152 __skb_pull(skb, skb->mac_len); 5153 *another = true; 5154 break; 5155 } 5156 return NULL; 5157 case TC_ACT_CONSUMED: 5158 return NULL; 5159 default: 5160 break; 5161 } 5162 #endif /* CONFIG_NET_CLS_ACT */ 5163 return skb; 5164 } 5165 5166 /** 5167 * netdev_is_rx_handler_busy - check if receive handler is registered 5168 * @dev: device to check 5169 * 5170 * Check if a receive handler is already registered for a given device. 5171 * Return true if there one. 5172 * 5173 * The caller must hold the rtnl_mutex. 5174 */ 5175 bool netdev_is_rx_handler_busy(struct net_device *dev) 5176 { 5177 ASSERT_RTNL(); 5178 return dev && rtnl_dereference(dev->rx_handler); 5179 } 5180 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); 5181 5182 /** 5183 * netdev_rx_handler_register - register receive handler 5184 * @dev: device to register a handler for 5185 * @rx_handler: receive handler to register 5186 * @rx_handler_data: data pointer that is used by rx handler 5187 * 5188 * Register a receive handler for a device. This handler will then be 5189 * called from __netif_receive_skb. A negative errno code is returned 5190 * on a failure. 5191 * 5192 * The caller must hold the rtnl_mutex. 5193 * 5194 * For a general description of rx_handler, see enum rx_handler_result. 5195 */ 5196 int netdev_rx_handler_register(struct net_device *dev, 5197 rx_handler_func_t *rx_handler, 5198 void *rx_handler_data) 5199 { 5200 if (netdev_is_rx_handler_busy(dev)) 5201 return -EBUSY; 5202 5203 if (dev->priv_flags & IFF_NO_RX_HANDLER) 5204 return -EINVAL; 5205 5206 /* Note: rx_handler_data must be set before rx_handler */ 5207 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); 5208 rcu_assign_pointer(dev->rx_handler, rx_handler); 5209 5210 return 0; 5211 } 5212 EXPORT_SYMBOL_GPL(netdev_rx_handler_register); 5213 5214 /** 5215 * netdev_rx_handler_unregister - unregister receive handler 5216 * @dev: device to unregister a handler from 5217 * 5218 * Unregister a receive handler from a device. 5219 * 5220 * The caller must hold the rtnl_mutex. 5221 */ 5222 void netdev_rx_handler_unregister(struct net_device *dev) 5223 { 5224 5225 ASSERT_RTNL(); 5226 RCU_INIT_POINTER(dev->rx_handler, NULL); 5227 /* a reader seeing a non NULL rx_handler in a rcu_read_lock() 5228 * section has a guarantee to see a non NULL rx_handler_data 5229 * as well. 5230 */ 5231 synchronize_net(); 5232 RCU_INIT_POINTER(dev->rx_handler_data, NULL); 5233 } 5234 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); 5235 5236 /* 5237 * Limit the use of PFMEMALLOC reserves to those protocols that implement 5238 * the special handling of PFMEMALLOC skbs. 5239 */ 5240 static bool skb_pfmemalloc_protocol(struct sk_buff *skb) 5241 { 5242 switch (skb->protocol) { 5243 case htons(ETH_P_ARP): 5244 case htons(ETH_P_IP): 5245 case htons(ETH_P_IPV6): 5246 case htons(ETH_P_8021Q): 5247 case htons(ETH_P_8021AD): 5248 return true; 5249 default: 5250 return false; 5251 } 5252 } 5253 5254 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, 5255 int *ret, struct net_device *orig_dev) 5256 { 5257 if (nf_hook_ingress_active(skb)) { 5258 int ingress_retval; 5259 5260 if (*pt_prev) { 5261 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5262 *pt_prev = NULL; 5263 } 5264 5265 rcu_read_lock(); 5266 ingress_retval = nf_hook_ingress(skb); 5267 rcu_read_unlock(); 5268 return ingress_retval; 5269 } 5270 return 0; 5271 } 5272 5273 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc, 5274 struct packet_type **ppt_prev) 5275 { 5276 struct packet_type *ptype, *pt_prev; 5277 rx_handler_func_t *rx_handler; 5278 struct sk_buff *skb = *pskb; 5279 struct net_device *orig_dev; 5280 bool deliver_exact = false; 5281 int ret = NET_RX_DROP; 5282 __be16 type; 5283 5284 net_timestamp_check(!READ_ONCE(netdev_tstamp_prequeue), skb); 5285 5286 trace_netif_receive_skb(skb); 5287 5288 orig_dev = skb->dev; 5289 5290 skb_reset_network_header(skb); 5291 if (!skb_transport_header_was_set(skb)) 5292 skb_reset_transport_header(skb); 5293 skb_reset_mac_len(skb); 5294 5295 pt_prev = NULL; 5296 5297 another_round: 5298 skb->skb_iif = skb->dev->ifindex; 5299 5300 __this_cpu_inc(softnet_data.processed); 5301 5302 if (static_branch_unlikely(&generic_xdp_needed_key)) { 5303 int ret2; 5304 5305 migrate_disable(); 5306 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb); 5307 migrate_enable(); 5308 5309 if (ret2 != XDP_PASS) { 5310 ret = NET_RX_DROP; 5311 goto out; 5312 } 5313 } 5314 5315 if (eth_type_vlan(skb->protocol)) { 5316 skb = skb_vlan_untag(skb); 5317 if (unlikely(!skb)) 5318 goto out; 5319 } 5320 5321 if (skb_skip_tc_classify(skb)) 5322 goto skip_classify; 5323 5324 if (pfmemalloc) 5325 goto skip_taps; 5326 5327 list_for_each_entry_rcu(ptype, &ptype_all, list) { 5328 if (pt_prev) 5329 ret = deliver_skb(skb, pt_prev, orig_dev); 5330 pt_prev = ptype; 5331 } 5332 5333 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { 5334 if (pt_prev) 5335 ret = deliver_skb(skb, pt_prev, orig_dev); 5336 pt_prev = ptype; 5337 } 5338 5339 skip_taps: 5340 #ifdef CONFIG_NET_INGRESS 5341 if (static_branch_unlikely(&ingress_needed_key)) { 5342 bool another = false; 5343 5344 nf_skip_egress(skb, true); 5345 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev, 5346 &another); 5347 if (another) 5348 goto another_round; 5349 if (!skb) 5350 goto out; 5351 5352 nf_skip_egress(skb, false); 5353 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) 5354 goto out; 5355 } 5356 #endif 5357 skb_reset_redirect(skb); 5358 skip_classify: 5359 if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) 5360 goto drop; 5361 5362 if (skb_vlan_tag_present(skb)) { 5363 if (pt_prev) { 5364 ret = deliver_skb(skb, pt_prev, orig_dev); 5365 pt_prev = NULL; 5366 } 5367 if (vlan_do_receive(&skb)) 5368 goto another_round; 5369 else if (unlikely(!skb)) 5370 goto out; 5371 } 5372 5373 rx_handler = rcu_dereference(skb->dev->rx_handler); 5374 if (rx_handler) { 5375 if (pt_prev) { 5376 ret = deliver_skb(skb, pt_prev, orig_dev); 5377 pt_prev = NULL; 5378 } 5379 switch (rx_handler(&skb)) { 5380 case RX_HANDLER_CONSUMED: 5381 ret = NET_RX_SUCCESS; 5382 goto out; 5383 case RX_HANDLER_ANOTHER: 5384 goto another_round; 5385 case RX_HANDLER_EXACT: 5386 deliver_exact = true; 5387 break; 5388 case RX_HANDLER_PASS: 5389 break; 5390 default: 5391 BUG(); 5392 } 5393 } 5394 5395 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) { 5396 check_vlan_id: 5397 if (skb_vlan_tag_get_id(skb)) { 5398 /* Vlan id is non 0 and vlan_do_receive() above couldn't 5399 * find vlan device. 5400 */ 5401 skb->pkt_type = PACKET_OTHERHOST; 5402 } else if (eth_type_vlan(skb->protocol)) { 5403 /* Outer header is 802.1P with vlan 0, inner header is 5404 * 802.1Q or 802.1AD and vlan_do_receive() above could 5405 * not find vlan dev for vlan id 0. 5406 */ 5407 __vlan_hwaccel_clear_tag(skb); 5408 skb = skb_vlan_untag(skb); 5409 if (unlikely(!skb)) 5410 goto out; 5411 if (vlan_do_receive(&skb)) 5412 /* After stripping off 802.1P header with vlan 0 5413 * vlan dev is found for inner header. 5414 */ 5415 goto another_round; 5416 else if (unlikely(!skb)) 5417 goto out; 5418 else 5419 /* We have stripped outer 802.1P vlan 0 header. 5420 * But could not find vlan dev. 5421 * check again for vlan id to set OTHERHOST. 5422 */ 5423 goto check_vlan_id; 5424 } 5425 /* Note: we might in the future use prio bits 5426 * and set skb->priority like in vlan_do_receive() 5427 * For the time being, just ignore Priority Code Point 5428 */ 5429 __vlan_hwaccel_clear_tag(skb); 5430 } 5431 5432 type = skb->protocol; 5433 5434 /* deliver only exact match when indicated */ 5435 if (likely(!deliver_exact)) { 5436 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5437 &ptype_base[ntohs(type) & 5438 PTYPE_HASH_MASK]); 5439 } 5440 5441 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5442 &orig_dev->ptype_specific); 5443 5444 if (unlikely(skb->dev != orig_dev)) { 5445 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5446 &skb->dev->ptype_specific); 5447 } 5448 5449 if (pt_prev) { 5450 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 5451 goto drop; 5452 *ppt_prev = pt_prev; 5453 } else { 5454 drop: 5455 if (!deliver_exact) 5456 dev_core_stats_rx_dropped_inc(skb->dev); 5457 else 5458 dev_core_stats_rx_nohandler_inc(skb->dev); 5459 kfree_skb_reason(skb, SKB_DROP_REASON_UNHANDLED_PROTO); 5460 /* Jamal, now you will not able to escape explaining 5461 * me how you were going to use this. :-) 5462 */ 5463 ret = NET_RX_DROP; 5464 } 5465 5466 out: 5467 /* The invariant here is that if *ppt_prev is not NULL 5468 * then skb should also be non-NULL. 5469 * 5470 * Apparently *ppt_prev assignment above holds this invariant due to 5471 * skb dereferencing near it. 5472 */ 5473 *pskb = skb; 5474 return ret; 5475 } 5476 5477 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc) 5478 { 5479 struct net_device *orig_dev = skb->dev; 5480 struct packet_type *pt_prev = NULL; 5481 int ret; 5482 5483 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5484 if (pt_prev) 5485 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb, 5486 skb->dev, pt_prev, orig_dev); 5487 return ret; 5488 } 5489 5490 /** 5491 * netif_receive_skb_core - special purpose version of netif_receive_skb 5492 * @skb: buffer to process 5493 * 5494 * More direct receive version of netif_receive_skb(). It should 5495 * only be used by callers that have a need to skip RPS and Generic XDP. 5496 * Caller must also take care of handling if ``(page_is_)pfmemalloc``. 5497 * 5498 * This function may only be called from softirq context and interrupts 5499 * should be enabled. 5500 * 5501 * Return values (usually ignored): 5502 * NET_RX_SUCCESS: no congestion 5503 * NET_RX_DROP: packet was dropped 5504 */ 5505 int netif_receive_skb_core(struct sk_buff *skb) 5506 { 5507 int ret; 5508 5509 rcu_read_lock(); 5510 ret = __netif_receive_skb_one_core(skb, false); 5511 rcu_read_unlock(); 5512 5513 return ret; 5514 } 5515 EXPORT_SYMBOL(netif_receive_skb_core); 5516 5517 static inline void __netif_receive_skb_list_ptype(struct list_head *head, 5518 struct packet_type *pt_prev, 5519 struct net_device *orig_dev) 5520 { 5521 struct sk_buff *skb, *next; 5522 5523 if (!pt_prev) 5524 return; 5525 if (list_empty(head)) 5526 return; 5527 if (pt_prev->list_func != NULL) 5528 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv, 5529 ip_list_rcv, head, pt_prev, orig_dev); 5530 else 5531 list_for_each_entry_safe(skb, next, head, list) { 5532 skb_list_del_init(skb); 5533 pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 5534 } 5535 } 5536 5537 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc) 5538 { 5539 /* Fast-path assumptions: 5540 * - There is no RX handler. 5541 * - Only one packet_type matches. 5542 * If either of these fails, we will end up doing some per-packet 5543 * processing in-line, then handling the 'last ptype' for the whole 5544 * sublist. This can't cause out-of-order delivery to any single ptype, 5545 * because the 'last ptype' must be constant across the sublist, and all 5546 * other ptypes are handled per-packet. 5547 */ 5548 /* Current (common) ptype of sublist */ 5549 struct packet_type *pt_curr = NULL; 5550 /* Current (common) orig_dev of sublist */ 5551 struct net_device *od_curr = NULL; 5552 struct list_head sublist; 5553 struct sk_buff *skb, *next; 5554 5555 INIT_LIST_HEAD(&sublist); 5556 list_for_each_entry_safe(skb, next, head, list) { 5557 struct net_device *orig_dev = skb->dev; 5558 struct packet_type *pt_prev = NULL; 5559 5560 skb_list_del_init(skb); 5561 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5562 if (!pt_prev) 5563 continue; 5564 if (pt_curr != pt_prev || od_curr != orig_dev) { 5565 /* dispatch old sublist */ 5566 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5567 /* start new sublist */ 5568 INIT_LIST_HEAD(&sublist); 5569 pt_curr = pt_prev; 5570 od_curr = orig_dev; 5571 } 5572 list_add_tail(&skb->list, &sublist); 5573 } 5574 5575 /* dispatch final sublist */ 5576 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5577 } 5578 5579 static int __netif_receive_skb(struct sk_buff *skb) 5580 { 5581 int ret; 5582 5583 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { 5584 unsigned int noreclaim_flag; 5585 5586 /* 5587 * PFMEMALLOC skbs are special, they should 5588 * - be delivered to SOCK_MEMALLOC sockets only 5589 * - stay away from userspace 5590 * - have bounded memory usage 5591 * 5592 * Use PF_MEMALLOC as this saves us from propagating the allocation 5593 * context down to all allocation sites. 5594 */ 5595 noreclaim_flag = memalloc_noreclaim_save(); 5596 ret = __netif_receive_skb_one_core(skb, true); 5597 memalloc_noreclaim_restore(noreclaim_flag); 5598 } else 5599 ret = __netif_receive_skb_one_core(skb, false); 5600 5601 return ret; 5602 } 5603 5604 static void __netif_receive_skb_list(struct list_head *head) 5605 { 5606 unsigned long noreclaim_flag = 0; 5607 struct sk_buff *skb, *next; 5608 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */ 5609 5610 list_for_each_entry_safe(skb, next, head, list) { 5611 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) { 5612 struct list_head sublist; 5613 5614 /* Handle the previous sublist */ 5615 list_cut_before(&sublist, head, &skb->list); 5616 if (!list_empty(&sublist)) 5617 __netif_receive_skb_list_core(&sublist, pfmemalloc); 5618 pfmemalloc = !pfmemalloc; 5619 /* See comments in __netif_receive_skb */ 5620 if (pfmemalloc) 5621 noreclaim_flag = memalloc_noreclaim_save(); 5622 else 5623 memalloc_noreclaim_restore(noreclaim_flag); 5624 } 5625 } 5626 /* Handle the remaining sublist */ 5627 if (!list_empty(head)) 5628 __netif_receive_skb_list_core(head, pfmemalloc); 5629 /* Restore pflags */ 5630 if (pfmemalloc) 5631 memalloc_noreclaim_restore(noreclaim_flag); 5632 } 5633 5634 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp) 5635 { 5636 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog); 5637 struct bpf_prog *new = xdp->prog; 5638 int ret = 0; 5639 5640 switch (xdp->command) { 5641 case XDP_SETUP_PROG: 5642 rcu_assign_pointer(dev->xdp_prog, new); 5643 if (old) 5644 bpf_prog_put(old); 5645 5646 if (old && !new) { 5647 static_branch_dec(&generic_xdp_needed_key); 5648 } else if (new && !old) { 5649 static_branch_inc(&generic_xdp_needed_key); 5650 dev_disable_lro(dev); 5651 dev_disable_gro_hw(dev); 5652 } 5653 break; 5654 5655 default: 5656 ret = -EINVAL; 5657 break; 5658 } 5659 5660 return ret; 5661 } 5662 5663 static int netif_receive_skb_internal(struct sk_buff *skb) 5664 { 5665 int ret; 5666 5667 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb); 5668 5669 if (skb_defer_rx_timestamp(skb)) 5670 return NET_RX_SUCCESS; 5671 5672 rcu_read_lock(); 5673 #ifdef CONFIG_RPS 5674 if (static_branch_unlikely(&rps_needed)) { 5675 struct rps_dev_flow voidflow, *rflow = &voidflow; 5676 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5677 5678 if (cpu >= 0) { 5679 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5680 rcu_read_unlock(); 5681 return ret; 5682 } 5683 } 5684 #endif 5685 ret = __netif_receive_skb(skb); 5686 rcu_read_unlock(); 5687 return ret; 5688 } 5689 5690 void netif_receive_skb_list_internal(struct list_head *head) 5691 { 5692 struct sk_buff *skb, *next; 5693 struct list_head sublist; 5694 5695 INIT_LIST_HEAD(&sublist); 5696 list_for_each_entry_safe(skb, next, head, list) { 5697 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb); 5698 skb_list_del_init(skb); 5699 if (!skb_defer_rx_timestamp(skb)) 5700 list_add_tail(&skb->list, &sublist); 5701 } 5702 list_splice_init(&sublist, head); 5703 5704 rcu_read_lock(); 5705 #ifdef CONFIG_RPS 5706 if (static_branch_unlikely(&rps_needed)) { 5707 list_for_each_entry_safe(skb, next, head, list) { 5708 struct rps_dev_flow voidflow, *rflow = &voidflow; 5709 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5710 5711 if (cpu >= 0) { 5712 /* Will be handled, remove from list */ 5713 skb_list_del_init(skb); 5714 enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5715 } 5716 } 5717 } 5718 #endif 5719 __netif_receive_skb_list(head); 5720 rcu_read_unlock(); 5721 } 5722 5723 /** 5724 * netif_receive_skb - process receive buffer from network 5725 * @skb: buffer to process 5726 * 5727 * netif_receive_skb() is the main receive data processing function. 5728 * It always succeeds. The buffer may be dropped during processing 5729 * for congestion control or by the protocol layers. 5730 * 5731 * This function may only be called from softirq context and interrupts 5732 * should be enabled. 5733 * 5734 * Return values (usually ignored): 5735 * NET_RX_SUCCESS: no congestion 5736 * NET_RX_DROP: packet was dropped 5737 */ 5738 int netif_receive_skb(struct sk_buff *skb) 5739 { 5740 int ret; 5741 5742 trace_netif_receive_skb_entry(skb); 5743 5744 ret = netif_receive_skb_internal(skb); 5745 trace_netif_receive_skb_exit(ret); 5746 5747 return ret; 5748 } 5749 EXPORT_SYMBOL(netif_receive_skb); 5750 5751 /** 5752 * netif_receive_skb_list - process many receive buffers from network 5753 * @head: list of skbs to process. 5754 * 5755 * Since return value of netif_receive_skb() is normally ignored, and 5756 * wouldn't be meaningful for a list, this function returns void. 5757 * 5758 * This function may only be called from softirq context and interrupts 5759 * should be enabled. 5760 */ 5761 void netif_receive_skb_list(struct list_head *head) 5762 { 5763 struct sk_buff *skb; 5764 5765 if (list_empty(head)) 5766 return; 5767 if (trace_netif_receive_skb_list_entry_enabled()) { 5768 list_for_each_entry(skb, head, list) 5769 trace_netif_receive_skb_list_entry(skb); 5770 } 5771 netif_receive_skb_list_internal(head); 5772 trace_netif_receive_skb_list_exit(0); 5773 } 5774 EXPORT_SYMBOL(netif_receive_skb_list); 5775 5776 static DEFINE_PER_CPU(struct work_struct, flush_works); 5777 5778 /* Network device is going away, flush any packets still pending */ 5779 static void flush_backlog(struct work_struct *work) 5780 { 5781 struct sk_buff *skb, *tmp; 5782 struct softnet_data *sd; 5783 5784 local_bh_disable(); 5785 sd = this_cpu_ptr(&softnet_data); 5786 5787 rps_lock_irq_disable(sd); 5788 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { 5789 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5790 __skb_unlink(skb, &sd->input_pkt_queue); 5791 dev_kfree_skb_irq(skb); 5792 input_queue_head_incr(sd); 5793 } 5794 } 5795 rps_unlock_irq_enable(sd); 5796 5797 skb_queue_walk_safe(&sd->process_queue, skb, tmp) { 5798 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5799 __skb_unlink(skb, &sd->process_queue); 5800 kfree_skb(skb); 5801 input_queue_head_incr(sd); 5802 } 5803 } 5804 local_bh_enable(); 5805 } 5806 5807 static bool flush_required(int cpu) 5808 { 5809 #if IS_ENABLED(CONFIG_RPS) 5810 struct softnet_data *sd = &per_cpu(softnet_data, cpu); 5811 bool do_flush; 5812 5813 rps_lock_irq_disable(sd); 5814 5815 /* as insertion into process_queue happens with the rps lock held, 5816 * process_queue access may race only with dequeue 5817 */ 5818 do_flush = !skb_queue_empty(&sd->input_pkt_queue) || 5819 !skb_queue_empty_lockless(&sd->process_queue); 5820 rps_unlock_irq_enable(sd); 5821 5822 return do_flush; 5823 #endif 5824 /* without RPS we can't safely check input_pkt_queue: during a 5825 * concurrent remote skb_queue_splice() we can detect as empty both 5826 * input_pkt_queue and process_queue even if the latter could end-up 5827 * containing a lot of packets. 5828 */ 5829 return true; 5830 } 5831 5832 static void flush_all_backlogs(void) 5833 { 5834 static cpumask_t flush_cpus; 5835 unsigned int cpu; 5836 5837 /* since we are under rtnl lock protection we can use static data 5838 * for the cpumask and avoid allocating on stack the possibly 5839 * large mask 5840 */ 5841 ASSERT_RTNL(); 5842 5843 cpus_read_lock(); 5844 5845 cpumask_clear(&flush_cpus); 5846 for_each_online_cpu(cpu) { 5847 if (flush_required(cpu)) { 5848 queue_work_on(cpu, system_highpri_wq, 5849 per_cpu_ptr(&flush_works, cpu)); 5850 cpumask_set_cpu(cpu, &flush_cpus); 5851 } 5852 } 5853 5854 /* we can have in flight packet[s] on the cpus we are not flushing, 5855 * synchronize_net() in unregister_netdevice_many() will take care of 5856 * them 5857 */ 5858 for_each_cpu(cpu, &flush_cpus) 5859 flush_work(per_cpu_ptr(&flush_works, cpu)); 5860 5861 cpus_read_unlock(); 5862 } 5863 5864 static void net_rps_send_ipi(struct softnet_data *remsd) 5865 { 5866 #ifdef CONFIG_RPS 5867 while (remsd) { 5868 struct softnet_data *next = remsd->rps_ipi_next; 5869 5870 if (cpu_online(remsd->cpu)) 5871 smp_call_function_single_async(remsd->cpu, &remsd->csd); 5872 remsd = next; 5873 } 5874 #endif 5875 } 5876 5877 /* 5878 * net_rps_action_and_irq_enable sends any pending IPI's for rps. 5879 * Note: called with local irq disabled, but exits with local irq enabled. 5880 */ 5881 static void net_rps_action_and_irq_enable(struct softnet_data *sd) 5882 { 5883 #ifdef CONFIG_RPS 5884 struct softnet_data *remsd = sd->rps_ipi_list; 5885 5886 if (remsd) { 5887 sd->rps_ipi_list = NULL; 5888 5889 local_irq_enable(); 5890 5891 /* Send pending IPI's to kick RPS processing on remote cpus. */ 5892 net_rps_send_ipi(remsd); 5893 } else 5894 #endif 5895 local_irq_enable(); 5896 } 5897 5898 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) 5899 { 5900 #ifdef CONFIG_RPS 5901 return sd->rps_ipi_list != NULL; 5902 #else 5903 return false; 5904 #endif 5905 } 5906 5907 static int process_backlog(struct napi_struct *napi, int quota) 5908 { 5909 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); 5910 bool again = true; 5911 int work = 0; 5912 5913 /* Check if we have pending ipi, its better to send them now, 5914 * not waiting net_rx_action() end. 5915 */ 5916 if (sd_has_rps_ipi_waiting(sd)) { 5917 local_irq_disable(); 5918 net_rps_action_and_irq_enable(sd); 5919 } 5920 5921 napi->weight = READ_ONCE(dev_rx_weight); 5922 while (again) { 5923 struct sk_buff *skb; 5924 5925 while ((skb = __skb_dequeue(&sd->process_queue))) { 5926 rcu_read_lock(); 5927 __netif_receive_skb(skb); 5928 rcu_read_unlock(); 5929 input_queue_head_incr(sd); 5930 if (++work >= quota) 5931 return work; 5932 5933 } 5934 5935 rps_lock_irq_disable(sd); 5936 if (skb_queue_empty(&sd->input_pkt_queue)) { 5937 /* 5938 * Inline a custom version of __napi_complete(). 5939 * only current cpu owns and manipulates this napi, 5940 * and NAPI_STATE_SCHED is the only possible flag set 5941 * on backlog. 5942 * We can use a plain write instead of clear_bit(), 5943 * and we dont need an smp_mb() memory barrier. 5944 */ 5945 napi->state = 0; 5946 again = false; 5947 } else { 5948 skb_queue_splice_tail_init(&sd->input_pkt_queue, 5949 &sd->process_queue); 5950 } 5951 rps_unlock_irq_enable(sd); 5952 } 5953 5954 return work; 5955 } 5956 5957 /** 5958 * __napi_schedule - schedule for receive 5959 * @n: entry to schedule 5960 * 5961 * The entry's receive function will be scheduled to run. 5962 * Consider using __napi_schedule_irqoff() if hard irqs are masked. 5963 */ 5964 void __napi_schedule(struct napi_struct *n) 5965 { 5966 unsigned long flags; 5967 5968 local_irq_save(flags); 5969 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 5970 local_irq_restore(flags); 5971 } 5972 EXPORT_SYMBOL(__napi_schedule); 5973 5974 /** 5975 * napi_schedule_prep - check if napi can be scheduled 5976 * @n: napi context 5977 * 5978 * Test if NAPI routine is already running, and if not mark 5979 * it as running. This is used as a condition variable to 5980 * insure only one NAPI poll instance runs. We also make 5981 * sure there is no pending NAPI disable. 5982 */ 5983 bool napi_schedule_prep(struct napi_struct *n) 5984 { 5985 unsigned long val, new; 5986 5987 do { 5988 val = READ_ONCE(n->state); 5989 if (unlikely(val & NAPIF_STATE_DISABLE)) 5990 return false; 5991 new = val | NAPIF_STATE_SCHED; 5992 5993 /* Sets STATE_MISSED bit if STATE_SCHED was already set 5994 * This was suggested by Alexander Duyck, as compiler 5995 * emits better code than : 5996 * if (val & NAPIF_STATE_SCHED) 5997 * new |= NAPIF_STATE_MISSED; 5998 */ 5999 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * 6000 NAPIF_STATE_MISSED; 6001 } while (cmpxchg(&n->state, val, new) != val); 6002 6003 return !(val & NAPIF_STATE_SCHED); 6004 } 6005 EXPORT_SYMBOL(napi_schedule_prep); 6006 6007 /** 6008 * __napi_schedule_irqoff - schedule for receive 6009 * @n: entry to schedule 6010 * 6011 * Variant of __napi_schedule() assuming hard irqs are masked. 6012 * 6013 * On PREEMPT_RT enabled kernels this maps to __napi_schedule() 6014 * because the interrupt disabled assumption might not be true 6015 * due to force-threaded interrupts and spinlock substitution. 6016 */ 6017 void __napi_schedule_irqoff(struct napi_struct *n) 6018 { 6019 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6020 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6021 else 6022 __napi_schedule(n); 6023 } 6024 EXPORT_SYMBOL(__napi_schedule_irqoff); 6025 6026 bool napi_complete_done(struct napi_struct *n, int work_done) 6027 { 6028 unsigned long flags, val, new, timeout = 0; 6029 bool ret = true; 6030 6031 /* 6032 * 1) Don't let napi dequeue from the cpu poll list 6033 * just in case its running on a different cpu. 6034 * 2) If we are busy polling, do nothing here, we have 6035 * the guarantee we will be called later. 6036 */ 6037 if (unlikely(n->state & (NAPIF_STATE_NPSVC | 6038 NAPIF_STATE_IN_BUSY_POLL))) 6039 return false; 6040 6041 if (work_done) { 6042 if (n->gro_bitmask) 6043 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6044 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs); 6045 } 6046 if (n->defer_hard_irqs_count > 0) { 6047 n->defer_hard_irqs_count--; 6048 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6049 if (timeout) 6050 ret = false; 6051 } 6052 if (n->gro_bitmask) { 6053 /* When the NAPI instance uses a timeout and keeps postponing 6054 * it, we need to bound somehow the time packets are kept in 6055 * the GRO layer 6056 */ 6057 napi_gro_flush(n, !!timeout); 6058 } 6059 6060 gro_normal_list(n); 6061 6062 if (unlikely(!list_empty(&n->poll_list))) { 6063 /* If n->poll_list is not empty, we need to mask irqs */ 6064 local_irq_save(flags); 6065 list_del_init(&n->poll_list); 6066 local_irq_restore(flags); 6067 } 6068 6069 do { 6070 val = READ_ONCE(n->state); 6071 6072 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); 6073 6074 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED | 6075 NAPIF_STATE_SCHED_THREADED | 6076 NAPIF_STATE_PREFER_BUSY_POLL); 6077 6078 /* If STATE_MISSED was set, leave STATE_SCHED set, 6079 * because we will call napi->poll() one more time. 6080 * This C code was suggested by Alexander Duyck to help gcc. 6081 */ 6082 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * 6083 NAPIF_STATE_SCHED; 6084 } while (cmpxchg(&n->state, val, new) != val); 6085 6086 if (unlikely(val & NAPIF_STATE_MISSED)) { 6087 __napi_schedule(n); 6088 return false; 6089 } 6090 6091 if (timeout) 6092 hrtimer_start(&n->timer, ns_to_ktime(timeout), 6093 HRTIMER_MODE_REL_PINNED); 6094 return ret; 6095 } 6096 EXPORT_SYMBOL(napi_complete_done); 6097 6098 /* must be called under rcu_read_lock(), as we dont take a reference */ 6099 static struct napi_struct *napi_by_id(unsigned int napi_id) 6100 { 6101 unsigned int hash = napi_id % HASH_SIZE(napi_hash); 6102 struct napi_struct *napi; 6103 6104 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) 6105 if (napi->napi_id == napi_id) 6106 return napi; 6107 6108 return NULL; 6109 } 6110 6111 #if defined(CONFIG_NET_RX_BUSY_POLL) 6112 6113 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule) 6114 { 6115 if (!skip_schedule) { 6116 gro_normal_list(napi); 6117 __napi_schedule(napi); 6118 return; 6119 } 6120 6121 if (napi->gro_bitmask) { 6122 /* flush too old packets 6123 * If HZ < 1000, flush all packets. 6124 */ 6125 napi_gro_flush(napi, HZ >= 1000); 6126 } 6127 6128 gro_normal_list(napi); 6129 clear_bit(NAPI_STATE_SCHED, &napi->state); 6130 } 6131 6132 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll, 6133 u16 budget) 6134 { 6135 bool skip_schedule = false; 6136 unsigned long timeout; 6137 int rc; 6138 6139 /* Busy polling means there is a high chance device driver hard irq 6140 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was 6141 * set in napi_schedule_prep(). 6142 * Since we are about to call napi->poll() once more, we can safely 6143 * clear NAPI_STATE_MISSED. 6144 * 6145 * Note: x86 could use a single "lock and ..." instruction 6146 * to perform these two clear_bit() 6147 */ 6148 clear_bit(NAPI_STATE_MISSED, &napi->state); 6149 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); 6150 6151 local_bh_disable(); 6152 6153 if (prefer_busy_poll) { 6154 napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs); 6155 timeout = READ_ONCE(napi->dev->gro_flush_timeout); 6156 if (napi->defer_hard_irqs_count && timeout) { 6157 hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED); 6158 skip_schedule = true; 6159 } 6160 } 6161 6162 /* All we really want here is to re-enable device interrupts. 6163 * Ideally, a new ndo_busy_poll_stop() could avoid another round. 6164 */ 6165 rc = napi->poll(napi, budget); 6166 /* We can't gro_normal_list() here, because napi->poll() might have 6167 * rearmed the napi (napi_complete_done()) in which case it could 6168 * already be running on another CPU. 6169 */ 6170 trace_napi_poll(napi, rc, budget); 6171 netpoll_poll_unlock(have_poll_lock); 6172 if (rc == budget) 6173 __busy_poll_stop(napi, skip_schedule); 6174 local_bh_enable(); 6175 } 6176 6177 void napi_busy_loop(unsigned int napi_id, 6178 bool (*loop_end)(void *, unsigned long), 6179 void *loop_end_arg, bool prefer_busy_poll, u16 budget) 6180 { 6181 unsigned long start_time = loop_end ? busy_loop_current_time() : 0; 6182 int (*napi_poll)(struct napi_struct *napi, int budget); 6183 void *have_poll_lock = NULL; 6184 struct napi_struct *napi; 6185 6186 restart: 6187 napi_poll = NULL; 6188 6189 rcu_read_lock(); 6190 6191 napi = napi_by_id(napi_id); 6192 if (!napi) 6193 goto out; 6194 6195 preempt_disable(); 6196 for (;;) { 6197 int work = 0; 6198 6199 local_bh_disable(); 6200 if (!napi_poll) { 6201 unsigned long val = READ_ONCE(napi->state); 6202 6203 /* If multiple threads are competing for this napi, 6204 * we avoid dirtying napi->state as much as we can. 6205 */ 6206 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | 6207 NAPIF_STATE_IN_BUSY_POLL)) { 6208 if (prefer_busy_poll) 6209 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6210 goto count; 6211 } 6212 if (cmpxchg(&napi->state, val, 6213 val | NAPIF_STATE_IN_BUSY_POLL | 6214 NAPIF_STATE_SCHED) != val) { 6215 if (prefer_busy_poll) 6216 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6217 goto count; 6218 } 6219 have_poll_lock = netpoll_poll_lock(napi); 6220 napi_poll = napi->poll; 6221 } 6222 work = napi_poll(napi, budget); 6223 trace_napi_poll(napi, work, budget); 6224 gro_normal_list(napi); 6225 count: 6226 if (work > 0) 6227 __NET_ADD_STATS(dev_net(napi->dev), 6228 LINUX_MIB_BUSYPOLLRXPACKETS, work); 6229 local_bh_enable(); 6230 6231 if (!loop_end || loop_end(loop_end_arg, start_time)) 6232 break; 6233 6234 if (unlikely(need_resched())) { 6235 if (napi_poll) 6236 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6237 preempt_enable(); 6238 rcu_read_unlock(); 6239 cond_resched(); 6240 if (loop_end(loop_end_arg, start_time)) 6241 return; 6242 goto restart; 6243 } 6244 cpu_relax(); 6245 } 6246 if (napi_poll) 6247 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6248 preempt_enable(); 6249 out: 6250 rcu_read_unlock(); 6251 } 6252 EXPORT_SYMBOL(napi_busy_loop); 6253 6254 #endif /* CONFIG_NET_RX_BUSY_POLL */ 6255 6256 static void napi_hash_add(struct napi_struct *napi) 6257 { 6258 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state)) 6259 return; 6260 6261 spin_lock(&napi_hash_lock); 6262 6263 /* 0..NR_CPUS range is reserved for sender_cpu use */ 6264 do { 6265 if (unlikely(++napi_gen_id < MIN_NAPI_ID)) 6266 napi_gen_id = MIN_NAPI_ID; 6267 } while (napi_by_id(napi_gen_id)); 6268 napi->napi_id = napi_gen_id; 6269 6270 hlist_add_head_rcu(&napi->napi_hash_node, 6271 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); 6272 6273 spin_unlock(&napi_hash_lock); 6274 } 6275 6276 /* Warning : caller is responsible to make sure rcu grace period 6277 * is respected before freeing memory containing @napi 6278 */ 6279 static void napi_hash_del(struct napi_struct *napi) 6280 { 6281 spin_lock(&napi_hash_lock); 6282 6283 hlist_del_init_rcu(&napi->napi_hash_node); 6284 6285 spin_unlock(&napi_hash_lock); 6286 } 6287 6288 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) 6289 { 6290 struct napi_struct *napi; 6291 6292 napi = container_of(timer, struct napi_struct, timer); 6293 6294 /* Note : we use a relaxed variant of napi_schedule_prep() not setting 6295 * NAPI_STATE_MISSED, since we do not react to a device IRQ. 6296 */ 6297 if (!napi_disable_pending(napi) && 6298 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) { 6299 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6300 __napi_schedule_irqoff(napi); 6301 } 6302 6303 return HRTIMER_NORESTART; 6304 } 6305 6306 static void init_gro_hash(struct napi_struct *napi) 6307 { 6308 int i; 6309 6310 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6311 INIT_LIST_HEAD(&napi->gro_hash[i].list); 6312 napi->gro_hash[i].count = 0; 6313 } 6314 napi->gro_bitmask = 0; 6315 } 6316 6317 int dev_set_threaded(struct net_device *dev, bool threaded) 6318 { 6319 struct napi_struct *napi; 6320 int err = 0; 6321 6322 if (dev->threaded == threaded) 6323 return 0; 6324 6325 if (threaded) { 6326 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6327 if (!napi->thread) { 6328 err = napi_kthread_create(napi); 6329 if (err) { 6330 threaded = false; 6331 break; 6332 } 6333 } 6334 } 6335 } 6336 6337 dev->threaded = threaded; 6338 6339 /* Make sure kthread is created before THREADED bit 6340 * is set. 6341 */ 6342 smp_mb__before_atomic(); 6343 6344 /* Setting/unsetting threaded mode on a napi might not immediately 6345 * take effect, if the current napi instance is actively being 6346 * polled. In this case, the switch between threaded mode and 6347 * softirq mode will happen in the next round of napi_schedule(). 6348 * This should not cause hiccups/stalls to the live traffic. 6349 */ 6350 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6351 if (threaded) 6352 set_bit(NAPI_STATE_THREADED, &napi->state); 6353 else 6354 clear_bit(NAPI_STATE_THREADED, &napi->state); 6355 } 6356 6357 return err; 6358 } 6359 EXPORT_SYMBOL(dev_set_threaded); 6360 6361 void netif_napi_add_weight(struct net_device *dev, struct napi_struct *napi, 6362 int (*poll)(struct napi_struct *, int), int weight) 6363 { 6364 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state))) 6365 return; 6366 6367 INIT_LIST_HEAD(&napi->poll_list); 6368 INIT_HLIST_NODE(&napi->napi_hash_node); 6369 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); 6370 napi->timer.function = napi_watchdog; 6371 init_gro_hash(napi); 6372 napi->skb = NULL; 6373 INIT_LIST_HEAD(&napi->rx_list); 6374 napi->rx_count = 0; 6375 napi->poll = poll; 6376 if (weight > NAPI_POLL_WEIGHT) 6377 netdev_err_once(dev, "%s() called with weight %d\n", __func__, 6378 weight); 6379 napi->weight = weight; 6380 napi->dev = dev; 6381 #ifdef CONFIG_NETPOLL 6382 napi->poll_owner = -1; 6383 #endif 6384 set_bit(NAPI_STATE_SCHED, &napi->state); 6385 set_bit(NAPI_STATE_NPSVC, &napi->state); 6386 list_add_rcu(&napi->dev_list, &dev->napi_list); 6387 napi_hash_add(napi); 6388 napi_get_frags_check(napi); 6389 /* Create kthread for this napi if dev->threaded is set. 6390 * Clear dev->threaded if kthread creation failed so that 6391 * threaded mode will not be enabled in napi_enable(). 6392 */ 6393 if (dev->threaded && napi_kthread_create(napi)) 6394 dev->threaded = 0; 6395 } 6396 EXPORT_SYMBOL(netif_napi_add_weight); 6397 6398 void napi_disable(struct napi_struct *n) 6399 { 6400 unsigned long val, new; 6401 6402 might_sleep(); 6403 set_bit(NAPI_STATE_DISABLE, &n->state); 6404 6405 for ( ; ; ) { 6406 val = READ_ONCE(n->state); 6407 if (val & (NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC)) { 6408 usleep_range(20, 200); 6409 continue; 6410 } 6411 6412 new = val | NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC; 6413 new &= ~(NAPIF_STATE_THREADED | NAPIF_STATE_PREFER_BUSY_POLL); 6414 6415 if (cmpxchg(&n->state, val, new) == val) 6416 break; 6417 } 6418 6419 hrtimer_cancel(&n->timer); 6420 6421 clear_bit(NAPI_STATE_DISABLE, &n->state); 6422 } 6423 EXPORT_SYMBOL(napi_disable); 6424 6425 /** 6426 * napi_enable - enable NAPI scheduling 6427 * @n: NAPI context 6428 * 6429 * Resume NAPI from being scheduled on this context. 6430 * Must be paired with napi_disable. 6431 */ 6432 void napi_enable(struct napi_struct *n) 6433 { 6434 unsigned long val, new; 6435 6436 do { 6437 val = READ_ONCE(n->state); 6438 BUG_ON(!test_bit(NAPI_STATE_SCHED, &val)); 6439 6440 new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC); 6441 if (n->dev->threaded && n->thread) 6442 new |= NAPIF_STATE_THREADED; 6443 } while (cmpxchg(&n->state, val, new) != val); 6444 } 6445 EXPORT_SYMBOL(napi_enable); 6446 6447 static void flush_gro_hash(struct napi_struct *napi) 6448 { 6449 int i; 6450 6451 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6452 struct sk_buff *skb, *n; 6453 6454 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list) 6455 kfree_skb(skb); 6456 napi->gro_hash[i].count = 0; 6457 } 6458 } 6459 6460 /* Must be called in process context */ 6461 void __netif_napi_del(struct napi_struct *napi) 6462 { 6463 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state)) 6464 return; 6465 6466 napi_hash_del(napi); 6467 list_del_rcu(&napi->dev_list); 6468 napi_free_frags(napi); 6469 6470 flush_gro_hash(napi); 6471 napi->gro_bitmask = 0; 6472 6473 if (napi->thread) { 6474 kthread_stop(napi->thread); 6475 napi->thread = NULL; 6476 } 6477 } 6478 EXPORT_SYMBOL(__netif_napi_del); 6479 6480 static int __napi_poll(struct napi_struct *n, bool *repoll) 6481 { 6482 int work, weight; 6483 6484 weight = n->weight; 6485 6486 /* This NAPI_STATE_SCHED test is for avoiding a race 6487 * with netpoll's poll_napi(). Only the entity which 6488 * obtains the lock and sees NAPI_STATE_SCHED set will 6489 * actually make the ->poll() call. Therefore we avoid 6490 * accidentally calling ->poll() when NAPI is not scheduled. 6491 */ 6492 work = 0; 6493 if (test_bit(NAPI_STATE_SCHED, &n->state)) { 6494 work = n->poll(n, weight); 6495 trace_napi_poll(n, work, weight); 6496 } 6497 6498 if (unlikely(work > weight)) 6499 netdev_err_once(n->dev, "NAPI poll function %pS returned %d, exceeding its budget of %d.\n", 6500 n->poll, work, weight); 6501 6502 if (likely(work < weight)) 6503 return work; 6504 6505 /* Drivers must not modify the NAPI state if they 6506 * consume the entire weight. In such cases this code 6507 * still "owns" the NAPI instance and therefore can 6508 * move the instance around on the list at-will. 6509 */ 6510 if (unlikely(napi_disable_pending(n))) { 6511 napi_complete(n); 6512 return work; 6513 } 6514 6515 /* The NAPI context has more processing work, but busy-polling 6516 * is preferred. Exit early. 6517 */ 6518 if (napi_prefer_busy_poll(n)) { 6519 if (napi_complete_done(n, work)) { 6520 /* If timeout is not set, we need to make sure 6521 * that the NAPI is re-scheduled. 6522 */ 6523 napi_schedule(n); 6524 } 6525 return work; 6526 } 6527 6528 if (n->gro_bitmask) { 6529 /* flush too old packets 6530 * If HZ < 1000, flush all packets. 6531 */ 6532 napi_gro_flush(n, HZ >= 1000); 6533 } 6534 6535 gro_normal_list(n); 6536 6537 /* Some drivers may have called napi_schedule 6538 * prior to exhausting their budget. 6539 */ 6540 if (unlikely(!list_empty(&n->poll_list))) { 6541 pr_warn_once("%s: Budget exhausted after napi rescheduled\n", 6542 n->dev ? n->dev->name : "backlog"); 6543 return work; 6544 } 6545 6546 *repoll = true; 6547 6548 return work; 6549 } 6550 6551 static int napi_poll(struct napi_struct *n, struct list_head *repoll) 6552 { 6553 bool do_repoll = false; 6554 void *have; 6555 int work; 6556 6557 list_del_init(&n->poll_list); 6558 6559 have = netpoll_poll_lock(n); 6560 6561 work = __napi_poll(n, &do_repoll); 6562 6563 if (do_repoll) 6564 list_add_tail(&n->poll_list, repoll); 6565 6566 netpoll_poll_unlock(have); 6567 6568 return work; 6569 } 6570 6571 static int napi_thread_wait(struct napi_struct *napi) 6572 { 6573 bool woken = false; 6574 6575 set_current_state(TASK_INTERRUPTIBLE); 6576 6577 while (!kthread_should_stop()) { 6578 /* Testing SCHED_THREADED bit here to make sure the current 6579 * kthread owns this napi and could poll on this napi. 6580 * Testing SCHED bit is not enough because SCHED bit might be 6581 * set by some other busy poll thread or by napi_disable(). 6582 */ 6583 if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state) || woken) { 6584 WARN_ON(!list_empty(&napi->poll_list)); 6585 __set_current_state(TASK_RUNNING); 6586 return 0; 6587 } 6588 6589 schedule(); 6590 /* woken being true indicates this thread owns this napi. */ 6591 woken = true; 6592 set_current_state(TASK_INTERRUPTIBLE); 6593 } 6594 __set_current_state(TASK_RUNNING); 6595 6596 return -1; 6597 } 6598 6599 static int napi_threaded_poll(void *data) 6600 { 6601 struct napi_struct *napi = data; 6602 void *have; 6603 6604 while (!napi_thread_wait(napi)) { 6605 for (;;) { 6606 bool repoll = false; 6607 6608 local_bh_disable(); 6609 6610 have = netpoll_poll_lock(napi); 6611 __napi_poll(napi, &repoll); 6612 netpoll_poll_unlock(have); 6613 6614 local_bh_enable(); 6615 6616 if (!repoll) 6617 break; 6618 6619 cond_resched(); 6620 } 6621 } 6622 return 0; 6623 } 6624 6625 static void skb_defer_free_flush(struct softnet_data *sd) 6626 { 6627 struct sk_buff *skb, *next; 6628 unsigned long flags; 6629 6630 /* Paired with WRITE_ONCE() in skb_attempt_defer_free() */ 6631 if (!READ_ONCE(sd->defer_list)) 6632 return; 6633 6634 spin_lock_irqsave(&sd->defer_lock, flags); 6635 skb = sd->defer_list; 6636 sd->defer_list = NULL; 6637 sd->defer_count = 0; 6638 spin_unlock_irqrestore(&sd->defer_lock, flags); 6639 6640 while (skb != NULL) { 6641 next = skb->next; 6642 napi_consume_skb(skb, 1); 6643 skb = next; 6644 } 6645 } 6646 6647 static __latent_entropy void net_rx_action(struct softirq_action *h) 6648 { 6649 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 6650 unsigned long time_limit = jiffies + 6651 usecs_to_jiffies(READ_ONCE(netdev_budget_usecs)); 6652 int budget = READ_ONCE(netdev_budget); 6653 LIST_HEAD(list); 6654 LIST_HEAD(repoll); 6655 6656 local_irq_disable(); 6657 list_splice_init(&sd->poll_list, &list); 6658 local_irq_enable(); 6659 6660 for (;;) { 6661 struct napi_struct *n; 6662 6663 skb_defer_free_flush(sd); 6664 6665 if (list_empty(&list)) { 6666 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll)) 6667 goto end; 6668 break; 6669 } 6670 6671 n = list_first_entry(&list, struct napi_struct, poll_list); 6672 budget -= napi_poll(n, &repoll); 6673 6674 /* If softirq window is exhausted then punt. 6675 * Allow this to run for 2 jiffies since which will allow 6676 * an average latency of 1.5/HZ. 6677 */ 6678 if (unlikely(budget <= 0 || 6679 time_after_eq(jiffies, time_limit))) { 6680 sd->time_squeeze++; 6681 break; 6682 } 6683 } 6684 6685 local_irq_disable(); 6686 6687 list_splice_tail_init(&sd->poll_list, &list); 6688 list_splice_tail(&repoll, &list); 6689 list_splice(&list, &sd->poll_list); 6690 if (!list_empty(&sd->poll_list)) 6691 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 6692 6693 net_rps_action_and_irq_enable(sd); 6694 end:; 6695 } 6696 6697 struct netdev_adjacent { 6698 struct net_device *dev; 6699 netdevice_tracker dev_tracker; 6700 6701 /* upper master flag, there can only be one master device per list */ 6702 bool master; 6703 6704 /* lookup ignore flag */ 6705 bool ignore; 6706 6707 /* counter for the number of times this device was added to us */ 6708 u16 ref_nr; 6709 6710 /* private field for the users */ 6711 void *private; 6712 6713 struct list_head list; 6714 struct rcu_head rcu; 6715 }; 6716 6717 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, 6718 struct list_head *adj_list) 6719 { 6720 struct netdev_adjacent *adj; 6721 6722 list_for_each_entry(adj, adj_list, list) { 6723 if (adj->dev == adj_dev) 6724 return adj; 6725 } 6726 return NULL; 6727 } 6728 6729 static int ____netdev_has_upper_dev(struct net_device *upper_dev, 6730 struct netdev_nested_priv *priv) 6731 { 6732 struct net_device *dev = (struct net_device *)priv->data; 6733 6734 return upper_dev == dev; 6735 } 6736 6737 /** 6738 * netdev_has_upper_dev - Check if device is linked to an upper device 6739 * @dev: device 6740 * @upper_dev: upper device to check 6741 * 6742 * Find out if a device is linked to specified upper device and return true 6743 * in case it is. Note that this checks only immediate upper device, 6744 * not through a complete stack of devices. The caller must hold the RTNL lock. 6745 */ 6746 bool netdev_has_upper_dev(struct net_device *dev, 6747 struct net_device *upper_dev) 6748 { 6749 struct netdev_nested_priv priv = { 6750 .data = (void *)upper_dev, 6751 }; 6752 6753 ASSERT_RTNL(); 6754 6755 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 6756 &priv); 6757 } 6758 EXPORT_SYMBOL(netdev_has_upper_dev); 6759 6760 /** 6761 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device 6762 * @dev: device 6763 * @upper_dev: upper device to check 6764 * 6765 * Find out if a device is linked to specified upper device and return true 6766 * in case it is. Note that this checks the entire upper device chain. 6767 * The caller must hold rcu lock. 6768 */ 6769 6770 bool netdev_has_upper_dev_all_rcu(struct net_device *dev, 6771 struct net_device *upper_dev) 6772 { 6773 struct netdev_nested_priv priv = { 6774 .data = (void *)upper_dev, 6775 }; 6776 6777 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 6778 &priv); 6779 } 6780 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); 6781 6782 /** 6783 * netdev_has_any_upper_dev - Check if device is linked to some device 6784 * @dev: device 6785 * 6786 * Find out if a device is linked to an upper device and return true in case 6787 * it is. The caller must hold the RTNL lock. 6788 */ 6789 bool netdev_has_any_upper_dev(struct net_device *dev) 6790 { 6791 ASSERT_RTNL(); 6792 6793 return !list_empty(&dev->adj_list.upper); 6794 } 6795 EXPORT_SYMBOL(netdev_has_any_upper_dev); 6796 6797 /** 6798 * netdev_master_upper_dev_get - Get master upper device 6799 * @dev: device 6800 * 6801 * Find a master upper device and return pointer to it or NULL in case 6802 * it's not there. The caller must hold the RTNL lock. 6803 */ 6804 struct net_device *netdev_master_upper_dev_get(struct net_device *dev) 6805 { 6806 struct netdev_adjacent *upper; 6807 6808 ASSERT_RTNL(); 6809 6810 if (list_empty(&dev->adj_list.upper)) 6811 return NULL; 6812 6813 upper = list_first_entry(&dev->adj_list.upper, 6814 struct netdev_adjacent, list); 6815 if (likely(upper->master)) 6816 return upper->dev; 6817 return NULL; 6818 } 6819 EXPORT_SYMBOL(netdev_master_upper_dev_get); 6820 6821 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev) 6822 { 6823 struct netdev_adjacent *upper; 6824 6825 ASSERT_RTNL(); 6826 6827 if (list_empty(&dev->adj_list.upper)) 6828 return NULL; 6829 6830 upper = list_first_entry(&dev->adj_list.upper, 6831 struct netdev_adjacent, list); 6832 if (likely(upper->master) && !upper->ignore) 6833 return upper->dev; 6834 return NULL; 6835 } 6836 6837 /** 6838 * netdev_has_any_lower_dev - Check if device is linked to some device 6839 * @dev: device 6840 * 6841 * Find out if a device is linked to a lower device and return true in case 6842 * it is. The caller must hold the RTNL lock. 6843 */ 6844 static bool netdev_has_any_lower_dev(struct net_device *dev) 6845 { 6846 ASSERT_RTNL(); 6847 6848 return !list_empty(&dev->adj_list.lower); 6849 } 6850 6851 void *netdev_adjacent_get_private(struct list_head *adj_list) 6852 { 6853 struct netdev_adjacent *adj; 6854 6855 adj = list_entry(adj_list, struct netdev_adjacent, list); 6856 6857 return adj->private; 6858 } 6859 EXPORT_SYMBOL(netdev_adjacent_get_private); 6860 6861 /** 6862 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list 6863 * @dev: device 6864 * @iter: list_head ** of the current position 6865 * 6866 * Gets the next device from the dev's upper list, starting from iter 6867 * position. The caller must hold RCU read lock. 6868 */ 6869 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, 6870 struct list_head **iter) 6871 { 6872 struct netdev_adjacent *upper; 6873 6874 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 6875 6876 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 6877 6878 if (&upper->list == &dev->adj_list.upper) 6879 return NULL; 6880 6881 *iter = &upper->list; 6882 6883 return upper->dev; 6884 } 6885 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); 6886 6887 static struct net_device *__netdev_next_upper_dev(struct net_device *dev, 6888 struct list_head **iter, 6889 bool *ignore) 6890 { 6891 struct netdev_adjacent *upper; 6892 6893 upper = list_entry((*iter)->next, struct netdev_adjacent, list); 6894 6895 if (&upper->list == &dev->adj_list.upper) 6896 return NULL; 6897 6898 *iter = &upper->list; 6899 *ignore = upper->ignore; 6900 6901 return upper->dev; 6902 } 6903 6904 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, 6905 struct list_head **iter) 6906 { 6907 struct netdev_adjacent *upper; 6908 6909 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 6910 6911 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 6912 6913 if (&upper->list == &dev->adj_list.upper) 6914 return NULL; 6915 6916 *iter = &upper->list; 6917 6918 return upper->dev; 6919 } 6920 6921 static int __netdev_walk_all_upper_dev(struct net_device *dev, 6922 int (*fn)(struct net_device *dev, 6923 struct netdev_nested_priv *priv), 6924 struct netdev_nested_priv *priv) 6925 { 6926 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 6927 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 6928 int ret, cur = 0; 6929 bool ignore; 6930 6931 now = dev; 6932 iter = &dev->adj_list.upper; 6933 6934 while (1) { 6935 if (now != dev) { 6936 ret = fn(now, priv); 6937 if (ret) 6938 return ret; 6939 } 6940 6941 next = NULL; 6942 while (1) { 6943 udev = __netdev_next_upper_dev(now, &iter, &ignore); 6944 if (!udev) 6945 break; 6946 if (ignore) 6947 continue; 6948 6949 next = udev; 6950 niter = &udev->adj_list.upper; 6951 dev_stack[cur] = now; 6952 iter_stack[cur++] = iter; 6953 break; 6954 } 6955 6956 if (!next) { 6957 if (!cur) 6958 return 0; 6959 next = dev_stack[--cur]; 6960 niter = iter_stack[cur]; 6961 } 6962 6963 now = next; 6964 iter = niter; 6965 } 6966 6967 return 0; 6968 } 6969 6970 int netdev_walk_all_upper_dev_rcu(struct net_device *dev, 6971 int (*fn)(struct net_device *dev, 6972 struct netdev_nested_priv *priv), 6973 struct netdev_nested_priv *priv) 6974 { 6975 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 6976 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 6977 int ret, cur = 0; 6978 6979 now = dev; 6980 iter = &dev->adj_list.upper; 6981 6982 while (1) { 6983 if (now != dev) { 6984 ret = fn(now, priv); 6985 if (ret) 6986 return ret; 6987 } 6988 6989 next = NULL; 6990 while (1) { 6991 udev = netdev_next_upper_dev_rcu(now, &iter); 6992 if (!udev) 6993 break; 6994 6995 next = udev; 6996 niter = &udev->adj_list.upper; 6997 dev_stack[cur] = now; 6998 iter_stack[cur++] = iter; 6999 break; 7000 } 7001 7002 if (!next) { 7003 if (!cur) 7004 return 0; 7005 next = dev_stack[--cur]; 7006 niter = iter_stack[cur]; 7007 } 7008 7009 now = next; 7010 iter = niter; 7011 } 7012 7013 return 0; 7014 } 7015 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); 7016 7017 static bool __netdev_has_upper_dev(struct net_device *dev, 7018 struct net_device *upper_dev) 7019 { 7020 struct netdev_nested_priv priv = { 7021 .flags = 0, 7022 .data = (void *)upper_dev, 7023 }; 7024 7025 ASSERT_RTNL(); 7026 7027 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev, 7028 &priv); 7029 } 7030 7031 /** 7032 * netdev_lower_get_next_private - Get the next ->private from the 7033 * lower neighbour list 7034 * @dev: device 7035 * @iter: list_head ** of the current position 7036 * 7037 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7038 * list, starting from iter position. The caller must hold either hold the 7039 * RTNL lock or its own locking that guarantees that the neighbour lower 7040 * list will remain unchanged. 7041 */ 7042 void *netdev_lower_get_next_private(struct net_device *dev, 7043 struct list_head **iter) 7044 { 7045 struct netdev_adjacent *lower; 7046 7047 lower = list_entry(*iter, struct netdev_adjacent, list); 7048 7049 if (&lower->list == &dev->adj_list.lower) 7050 return NULL; 7051 7052 *iter = lower->list.next; 7053 7054 return lower->private; 7055 } 7056 EXPORT_SYMBOL(netdev_lower_get_next_private); 7057 7058 /** 7059 * netdev_lower_get_next_private_rcu - Get the next ->private from the 7060 * lower neighbour list, RCU 7061 * variant 7062 * @dev: device 7063 * @iter: list_head ** of the current position 7064 * 7065 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7066 * list, starting from iter position. The caller must hold RCU read lock. 7067 */ 7068 void *netdev_lower_get_next_private_rcu(struct net_device *dev, 7069 struct list_head **iter) 7070 { 7071 struct netdev_adjacent *lower; 7072 7073 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 7074 7075 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7076 7077 if (&lower->list == &dev->adj_list.lower) 7078 return NULL; 7079 7080 *iter = &lower->list; 7081 7082 return lower->private; 7083 } 7084 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); 7085 7086 /** 7087 * netdev_lower_get_next - Get the next device from the lower neighbour 7088 * list 7089 * @dev: device 7090 * @iter: list_head ** of the current position 7091 * 7092 * Gets the next netdev_adjacent from the dev's lower neighbour 7093 * list, starting from iter position. The caller must hold RTNL lock or 7094 * its own locking that guarantees that the neighbour lower 7095 * list will remain unchanged. 7096 */ 7097 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) 7098 { 7099 struct netdev_adjacent *lower; 7100 7101 lower = list_entry(*iter, struct netdev_adjacent, list); 7102 7103 if (&lower->list == &dev->adj_list.lower) 7104 return NULL; 7105 7106 *iter = lower->list.next; 7107 7108 return lower->dev; 7109 } 7110 EXPORT_SYMBOL(netdev_lower_get_next); 7111 7112 static struct net_device *netdev_next_lower_dev(struct net_device *dev, 7113 struct list_head **iter) 7114 { 7115 struct netdev_adjacent *lower; 7116 7117 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7118 7119 if (&lower->list == &dev->adj_list.lower) 7120 return NULL; 7121 7122 *iter = &lower->list; 7123 7124 return lower->dev; 7125 } 7126 7127 static struct net_device *__netdev_next_lower_dev(struct net_device *dev, 7128 struct list_head **iter, 7129 bool *ignore) 7130 { 7131 struct netdev_adjacent *lower; 7132 7133 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7134 7135 if (&lower->list == &dev->adj_list.lower) 7136 return NULL; 7137 7138 *iter = &lower->list; 7139 *ignore = lower->ignore; 7140 7141 return lower->dev; 7142 } 7143 7144 int netdev_walk_all_lower_dev(struct net_device *dev, 7145 int (*fn)(struct net_device *dev, 7146 struct netdev_nested_priv *priv), 7147 struct netdev_nested_priv *priv) 7148 { 7149 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7150 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7151 int ret, cur = 0; 7152 7153 now = dev; 7154 iter = &dev->adj_list.lower; 7155 7156 while (1) { 7157 if (now != dev) { 7158 ret = fn(now, priv); 7159 if (ret) 7160 return ret; 7161 } 7162 7163 next = NULL; 7164 while (1) { 7165 ldev = netdev_next_lower_dev(now, &iter); 7166 if (!ldev) 7167 break; 7168 7169 next = ldev; 7170 niter = &ldev->adj_list.lower; 7171 dev_stack[cur] = now; 7172 iter_stack[cur++] = iter; 7173 break; 7174 } 7175 7176 if (!next) { 7177 if (!cur) 7178 return 0; 7179 next = dev_stack[--cur]; 7180 niter = iter_stack[cur]; 7181 } 7182 7183 now = next; 7184 iter = niter; 7185 } 7186 7187 return 0; 7188 } 7189 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); 7190 7191 static int __netdev_walk_all_lower_dev(struct net_device *dev, 7192 int (*fn)(struct net_device *dev, 7193 struct netdev_nested_priv *priv), 7194 struct netdev_nested_priv *priv) 7195 { 7196 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7197 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7198 int ret, cur = 0; 7199 bool ignore; 7200 7201 now = dev; 7202 iter = &dev->adj_list.lower; 7203 7204 while (1) { 7205 if (now != dev) { 7206 ret = fn(now, priv); 7207 if (ret) 7208 return ret; 7209 } 7210 7211 next = NULL; 7212 while (1) { 7213 ldev = __netdev_next_lower_dev(now, &iter, &ignore); 7214 if (!ldev) 7215 break; 7216 if (ignore) 7217 continue; 7218 7219 next = ldev; 7220 niter = &ldev->adj_list.lower; 7221 dev_stack[cur] = now; 7222 iter_stack[cur++] = iter; 7223 break; 7224 } 7225 7226 if (!next) { 7227 if (!cur) 7228 return 0; 7229 next = dev_stack[--cur]; 7230 niter = iter_stack[cur]; 7231 } 7232 7233 now = next; 7234 iter = niter; 7235 } 7236 7237 return 0; 7238 } 7239 7240 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, 7241 struct list_head **iter) 7242 { 7243 struct netdev_adjacent *lower; 7244 7245 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7246 if (&lower->list == &dev->adj_list.lower) 7247 return NULL; 7248 7249 *iter = &lower->list; 7250 7251 return lower->dev; 7252 } 7253 EXPORT_SYMBOL(netdev_next_lower_dev_rcu); 7254 7255 static u8 __netdev_upper_depth(struct net_device *dev) 7256 { 7257 struct net_device *udev; 7258 struct list_head *iter; 7259 u8 max_depth = 0; 7260 bool ignore; 7261 7262 for (iter = &dev->adj_list.upper, 7263 udev = __netdev_next_upper_dev(dev, &iter, &ignore); 7264 udev; 7265 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) { 7266 if (ignore) 7267 continue; 7268 if (max_depth < udev->upper_level) 7269 max_depth = udev->upper_level; 7270 } 7271 7272 return max_depth; 7273 } 7274 7275 static u8 __netdev_lower_depth(struct net_device *dev) 7276 { 7277 struct net_device *ldev; 7278 struct list_head *iter; 7279 u8 max_depth = 0; 7280 bool ignore; 7281 7282 for (iter = &dev->adj_list.lower, 7283 ldev = __netdev_next_lower_dev(dev, &iter, &ignore); 7284 ldev; 7285 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) { 7286 if (ignore) 7287 continue; 7288 if (max_depth < ldev->lower_level) 7289 max_depth = ldev->lower_level; 7290 } 7291 7292 return max_depth; 7293 } 7294 7295 static int __netdev_update_upper_level(struct net_device *dev, 7296 struct netdev_nested_priv *__unused) 7297 { 7298 dev->upper_level = __netdev_upper_depth(dev) + 1; 7299 return 0; 7300 } 7301 7302 #ifdef CONFIG_LOCKDEP 7303 static LIST_HEAD(net_unlink_list); 7304 7305 static void net_unlink_todo(struct net_device *dev) 7306 { 7307 if (list_empty(&dev->unlink_list)) 7308 list_add_tail(&dev->unlink_list, &net_unlink_list); 7309 } 7310 #endif 7311 7312 static int __netdev_update_lower_level(struct net_device *dev, 7313 struct netdev_nested_priv *priv) 7314 { 7315 dev->lower_level = __netdev_lower_depth(dev) + 1; 7316 7317 #ifdef CONFIG_LOCKDEP 7318 if (!priv) 7319 return 0; 7320 7321 if (priv->flags & NESTED_SYNC_IMM) 7322 dev->nested_level = dev->lower_level - 1; 7323 if (priv->flags & NESTED_SYNC_TODO) 7324 net_unlink_todo(dev); 7325 #endif 7326 return 0; 7327 } 7328 7329 int netdev_walk_all_lower_dev_rcu(struct net_device *dev, 7330 int (*fn)(struct net_device *dev, 7331 struct netdev_nested_priv *priv), 7332 struct netdev_nested_priv *priv) 7333 { 7334 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7335 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7336 int ret, cur = 0; 7337 7338 now = dev; 7339 iter = &dev->adj_list.lower; 7340 7341 while (1) { 7342 if (now != dev) { 7343 ret = fn(now, priv); 7344 if (ret) 7345 return ret; 7346 } 7347 7348 next = NULL; 7349 while (1) { 7350 ldev = netdev_next_lower_dev_rcu(now, &iter); 7351 if (!ldev) 7352 break; 7353 7354 next = ldev; 7355 niter = &ldev->adj_list.lower; 7356 dev_stack[cur] = now; 7357 iter_stack[cur++] = iter; 7358 break; 7359 } 7360 7361 if (!next) { 7362 if (!cur) 7363 return 0; 7364 next = dev_stack[--cur]; 7365 niter = iter_stack[cur]; 7366 } 7367 7368 now = next; 7369 iter = niter; 7370 } 7371 7372 return 0; 7373 } 7374 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); 7375 7376 /** 7377 * netdev_lower_get_first_private_rcu - Get the first ->private from the 7378 * lower neighbour list, RCU 7379 * variant 7380 * @dev: device 7381 * 7382 * Gets the first netdev_adjacent->private from the dev's lower neighbour 7383 * list. The caller must hold RCU read lock. 7384 */ 7385 void *netdev_lower_get_first_private_rcu(struct net_device *dev) 7386 { 7387 struct netdev_adjacent *lower; 7388 7389 lower = list_first_or_null_rcu(&dev->adj_list.lower, 7390 struct netdev_adjacent, list); 7391 if (lower) 7392 return lower->private; 7393 return NULL; 7394 } 7395 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); 7396 7397 /** 7398 * netdev_master_upper_dev_get_rcu - Get master upper device 7399 * @dev: device 7400 * 7401 * Find a master upper device and return pointer to it or NULL in case 7402 * it's not there. The caller must hold the RCU read lock. 7403 */ 7404 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) 7405 { 7406 struct netdev_adjacent *upper; 7407 7408 upper = list_first_or_null_rcu(&dev->adj_list.upper, 7409 struct netdev_adjacent, list); 7410 if (upper && likely(upper->master)) 7411 return upper->dev; 7412 return NULL; 7413 } 7414 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); 7415 7416 static int netdev_adjacent_sysfs_add(struct net_device *dev, 7417 struct net_device *adj_dev, 7418 struct list_head *dev_list) 7419 { 7420 char linkname[IFNAMSIZ+7]; 7421 7422 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7423 "upper_%s" : "lower_%s", adj_dev->name); 7424 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), 7425 linkname); 7426 } 7427 static void netdev_adjacent_sysfs_del(struct net_device *dev, 7428 char *name, 7429 struct list_head *dev_list) 7430 { 7431 char linkname[IFNAMSIZ+7]; 7432 7433 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7434 "upper_%s" : "lower_%s", name); 7435 sysfs_remove_link(&(dev->dev.kobj), linkname); 7436 } 7437 7438 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, 7439 struct net_device *adj_dev, 7440 struct list_head *dev_list) 7441 { 7442 return (dev_list == &dev->adj_list.upper || 7443 dev_list == &dev->adj_list.lower) && 7444 net_eq(dev_net(dev), dev_net(adj_dev)); 7445 } 7446 7447 static int __netdev_adjacent_dev_insert(struct net_device *dev, 7448 struct net_device *adj_dev, 7449 struct list_head *dev_list, 7450 void *private, bool master) 7451 { 7452 struct netdev_adjacent *adj; 7453 int ret; 7454 7455 adj = __netdev_find_adj(adj_dev, dev_list); 7456 7457 if (adj) { 7458 adj->ref_nr += 1; 7459 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", 7460 dev->name, adj_dev->name, adj->ref_nr); 7461 7462 return 0; 7463 } 7464 7465 adj = kmalloc(sizeof(*adj), GFP_KERNEL); 7466 if (!adj) 7467 return -ENOMEM; 7468 7469 adj->dev = adj_dev; 7470 adj->master = master; 7471 adj->ref_nr = 1; 7472 adj->private = private; 7473 adj->ignore = false; 7474 netdev_hold(adj_dev, &adj->dev_tracker, GFP_KERNEL); 7475 7476 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", 7477 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); 7478 7479 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { 7480 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); 7481 if (ret) 7482 goto free_adj; 7483 } 7484 7485 /* Ensure that master link is always the first item in list. */ 7486 if (master) { 7487 ret = sysfs_create_link(&(dev->dev.kobj), 7488 &(adj_dev->dev.kobj), "master"); 7489 if (ret) 7490 goto remove_symlinks; 7491 7492 list_add_rcu(&adj->list, dev_list); 7493 } else { 7494 list_add_tail_rcu(&adj->list, dev_list); 7495 } 7496 7497 return 0; 7498 7499 remove_symlinks: 7500 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7501 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7502 free_adj: 7503 netdev_put(adj_dev, &adj->dev_tracker); 7504 kfree(adj); 7505 7506 return ret; 7507 } 7508 7509 static void __netdev_adjacent_dev_remove(struct net_device *dev, 7510 struct net_device *adj_dev, 7511 u16 ref_nr, 7512 struct list_head *dev_list) 7513 { 7514 struct netdev_adjacent *adj; 7515 7516 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", 7517 dev->name, adj_dev->name, ref_nr); 7518 7519 adj = __netdev_find_adj(adj_dev, dev_list); 7520 7521 if (!adj) { 7522 pr_err("Adjacency does not exist for device %s from %s\n", 7523 dev->name, adj_dev->name); 7524 WARN_ON(1); 7525 return; 7526 } 7527 7528 if (adj->ref_nr > ref_nr) { 7529 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", 7530 dev->name, adj_dev->name, ref_nr, 7531 adj->ref_nr - ref_nr); 7532 adj->ref_nr -= ref_nr; 7533 return; 7534 } 7535 7536 if (adj->master) 7537 sysfs_remove_link(&(dev->dev.kobj), "master"); 7538 7539 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7540 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7541 7542 list_del_rcu(&adj->list); 7543 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", 7544 adj_dev->name, dev->name, adj_dev->name); 7545 netdev_put(adj_dev, &adj->dev_tracker); 7546 kfree_rcu(adj, rcu); 7547 } 7548 7549 static int __netdev_adjacent_dev_link_lists(struct net_device *dev, 7550 struct net_device *upper_dev, 7551 struct list_head *up_list, 7552 struct list_head *down_list, 7553 void *private, bool master) 7554 { 7555 int ret; 7556 7557 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, 7558 private, master); 7559 if (ret) 7560 return ret; 7561 7562 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, 7563 private, false); 7564 if (ret) { 7565 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); 7566 return ret; 7567 } 7568 7569 return 0; 7570 } 7571 7572 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, 7573 struct net_device *upper_dev, 7574 u16 ref_nr, 7575 struct list_head *up_list, 7576 struct list_head *down_list) 7577 { 7578 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); 7579 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); 7580 } 7581 7582 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, 7583 struct net_device *upper_dev, 7584 void *private, bool master) 7585 { 7586 return __netdev_adjacent_dev_link_lists(dev, upper_dev, 7587 &dev->adj_list.upper, 7588 &upper_dev->adj_list.lower, 7589 private, master); 7590 } 7591 7592 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, 7593 struct net_device *upper_dev) 7594 { 7595 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, 7596 &dev->adj_list.upper, 7597 &upper_dev->adj_list.lower); 7598 } 7599 7600 static int __netdev_upper_dev_link(struct net_device *dev, 7601 struct net_device *upper_dev, bool master, 7602 void *upper_priv, void *upper_info, 7603 struct netdev_nested_priv *priv, 7604 struct netlink_ext_ack *extack) 7605 { 7606 struct netdev_notifier_changeupper_info changeupper_info = { 7607 .info = { 7608 .dev = dev, 7609 .extack = extack, 7610 }, 7611 .upper_dev = upper_dev, 7612 .master = master, 7613 .linking = true, 7614 .upper_info = upper_info, 7615 }; 7616 struct net_device *master_dev; 7617 int ret = 0; 7618 7619 ASSERT_RTNL(); 7620 7621 if (dev == upper_dev) 7622 return -EBUSY; 7623 7624 /* To prevent loops, check if dev is not upper device to upper_dev. */ 7625 if (__netdev_has_upper_dev(upper_dev, dev)) 7626 return -EBUSY; 7627 7628 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV) 7629 return -EMLINK; 7630 7631 if (!master) { 7632 if (__netdev_has_upper_dev(dev, upper_dev)) 7633 return -EEXIST; 7634 } else { 7635 master_dev = __netdev_master_upper_dev_get(dev); 7636 if (master_dev) 7637 return master_dev == upper_dev ? -EEXIST : -EBUSY; 7638 } 7639 7640 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7641 &changeupper_info.info); 7642 ret = notifier_to_errno(ret); 7643 if (ret) 7644 return ret; 7645 7646 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, 7647 master); 7648 if (ret) 7649 return ret; 7650 7651 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 7652 &changeupper_info.info); 7653 ret = notifier_to_errno(ret); 7654 if (ret) 7655 goto rollback; 7656 7657 __netdev_update_upper_level(dev, NULL); 7658 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 7659 7660 __netdev_update_lower_level(upper_dev, priv); 7661 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 7662 priv); 7663 7664 return 0; 7665 7666 rollback: 7667 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 7668 7669 return ret; 7670 } 7671 7672 /** 7673 * netdev_upper_dev_link - Add a link to the upper device 7674 * @dev: device 7675 * @upper_dev: new upper device 7676 * @extack: netlink extended ack 7677 * 7678 * Adds a link to device which is upper to this one. The caller must hold 7679 * the RTNL lock. On a failure a negative errno code is returned. 7680 * On success the reference counts are adjusted and the function 7681 * returns zero. 7682 */ 7683 int netdev_upper_dev_link(struct net_device *dev, 7684 struct net_device *upper_dev, 7685 struct netlink_ext_ack *extack) 7686 { 7687 struct netdev_nested_priv priv = { 7688 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7689 .data = NULL, 7690 }; 7691 7692 return __netdev_upper_dev_link(dev, upper_dev, false, 7693 NULL, NULL, &priv, extack); 7694 } 7695 EXPORT_SYMBOL(netdev_upper_dev_link); 7696 7697 /** 7698 * netdev_master_upper_dev_link - Add a master link to the upper device 7699 * @dev: device 7700 * @upper_dev: new upper device 7701 * @upper_priv: upper device private 7702 * @upper_info: upper info to be passed down via notifier 7703 * @extack: netlink extended ack 7704 * 7705 * Adds a link to device which is upper to this one. In this case, only 7706 * one master upper device can be linked, although other non-master devices 7707 * might be linked as well. The caller must hold the RTNL lock. 7708 * On a failure a negative errno code is returned. On success the reference 7709 * counts are adjusted and the function returns zero. 7710 */ 7711 int netdev_master_upper_dev_link(struct net_device *dev, 7712 struct net_device *upper_dev, 7713 void *upper_priv, void *upper_info, 7714 struct netlink_ext_ack *extack) 7715 { 7716 struct netdev_nested_priv priv = { 7717 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7718 .data = NULL, 7719 }; 7720 7721 return __netdev_upper_dev_link(dev, upper_dev, true, 7722 upper_priv, upper_info, &priv, extack); 7723 } 7724 EXPORT_SYMBOL(netdev_master_upper_dev_link); 7725 7726 static void __netdev_upper_dev_unlink(struct net_device *dev, 7727 struct net_device *upper_dev, 7728 struct netdev_nested_priv *priv) 7729 { 7730 struct netdev_notifier_changeupper_info changeupper_info = { 7731 .info = { 7732 .dev = dev, 7733 }, 7734 .upper_dev = upper_dev, 7735 .linking = false, 7736 }; 7737 7738 ASSERT_RTNL(); 7739 7740 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; 7741 7742 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7743 &changeupper_info.info); 7744 7745 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 7746 7747 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 7748 &changeupper_info.info); 7749 7750 __netdev_update_upper_level(dev, NULL); 7751 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 7752 7753 __netdev_update_lower_level(upper_dev, priv); 7754 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 7755 priv); 7756 } 7757 7758 /** 7759 * netdev_upper_dev_unlink - Removes a link to upper device 7760 * @dev: device 7761 * @upper_dev: new upper device 7762 * 7763 * Removes a link to device which is upper to this one. The caller must hold 7764 * the RTNL lock. 7765 */ 7766 void netdev_upper_dev_unlink(struct net_device *dev, 7767 struct net_device *upper_dev) 7768 { 7769 struct netdev_nested_priv priv = { 7770 .flags = NESTED_SYNC_TODO, 7771 .data = NULL, 7772 }; 7773 7774 __netdev_upper_dev_unlink(dev, upper_dev, &priv); 7775 } 7776 EXPORT_SYMBOL(netdev_upper_dev_unlink); 7777 7778 static void __netdev_adjacent_dev_set(struct net_device *upper_dev, 7779 struct net_device *lower_dev, 7780 bool val) 7781 { 7782 struct netdev_adjacent *adj; 7783 7784 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower); 7785 if (adj) 7786 adj->ignore = val; 7787 7788 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper); 7789 if (adj) 7790 adj->ignore = val; 7791 } 7792 7793 static void netdev_adjacent_dev_disable(struct net_device *upper_dev, 7794 struct net_device *lower_dev) 7795 { 7796 __netdev_adjacent_dev_set(upper_dev, lower_dev, true); 7797 } 7798 7799 static void netdev_adjacent_dev_enable(struct net_device *upper_dev, 7800 struct net_device *lower_dev) 7801 { 7802 __netdev_adjacent_dev_set(upper_dev, lower_dev, false); 7803 } 7804 7805 int netdev_adjacent_change_prepare(struct net_device *old_dev, 7806 struct net_device *new_dev, 7807 struct net_device *dev, 7808 struct netlink_ext_ack *extack) 7809 { 7810 struct netdev_nested_priv priv = { 7811 .flags = 0, 7812 .data = NULL, 7813 }; 7814 int err; 7815 7816 if (!new_dev) 7817 return 0; 7818 7819 if (old_dev && new_dev != old_dev) 7820 netdev_adjacent_dev_disable(dev, old_dev); 7821 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv, 7822 extack); 7823 if (err) { 7824 if (old_dev && new_dev != old_dev) 7825 netdev_adjacent_dev_enable(dev, old_dev); 7826 return err; 7827 } 7828 7829 return 0; 7830 } 7831 EXPORT_SYMBOL(netdev_adjacent_change_prepare); 7832 7833 void netdev_adjacent_change_commit(struct net_device *old_dev, 7834 struct net_device *new_dev, 7835 struct net_device *dev) 7836 { 7837 struct netdev_nested_priv priv = { 7838 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7839 .data = NULL, 7840 }; 7841 7842 if (!new_dev || !old_dev) 7843 return; 7844 7845 if (new_dev == old_dev) 7846 return; 7847 7848 netdev_adjacent_dev_enable(dev, old_dev); 7849 __netdev_upper_dev_unlink(old_dev, dev, &priv); 7850 } 7851 EXPORT_SYMBOL(netdev_adjacent_change_commit); 7852 7853 void netdev_adjacent_change_abort(struct net_device *old_dev, 7854 struct net_device *new_dev, 7855 struct net_device *dev) 7856 { 7857 struct netdev_nested_priv priv = { 7858 .flags = 0, 7859 .data = NULL, 7860 }; 7861 7862 if (!new_dev) 7863 return; 7864 7865 if (old_dev && new_dev != old_dev) 7866 netdev_adjacent_dev_enable(dev, old_dev); 7867 7868 __netdev_upper_dev_unlink(new_dev, dev, &priv); 7869 } 7870 EXPORT_SYMBOL(netdev_adjacent_change_abort); 7871 7872 /** 7873 * netdev_bonding_info_change - Dispatch event about slave change 7874 * @dev: device 7875 * @bonding_info: info to dispatch 7876 * 7877 * Send NETDEV_BONDING_INFO to netdev notifiers with info. 7878 * The caller must hold the RTNL lock. 7879 */ 7880 void netdev_bonding_info_change(struct net_device *dev, 7881 struct netdev_bonding_info *bonding_info) 7882 { 7883 struct netdev_notifier_bonding_info info = { 7884 .info.dev = dev, 7885 }; 7886 7887 memcpy(&info.bonding_info, bonding_info, 7888 sizeof(struct netdev_bonding_info)); 7889 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, 7890 &info.info); 7891 } 7892 EXPORT_SYMBOL(netdev_bonding_info_change); 7893 7894 static int netdev_offload_xstats_enable_l3(struct net_device *dev, 7895 struct netlink_ext_ack *extack) 7896 { 7897 struct netdev_notifier_offload_xstats_info info = { 7898 .info.dev = dev, 7899 .info.extack = extack, 7900 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, 7901 }; 7902 int err; 7903 int rc; 7904 7905 dev->offload_xstats_l3 = kzalloc(sizeof(*dev->offload_xstats_l3), 7906 GFP_KERNEL); 7907 if (!dev->offload_xstats_l3) 7908 return -ENOMEM; 7909 7910 rc = call_netdevice_notifiers_info_robust(NETDEV_OFFLOAD_XSTATS_ENABLE, 7911 NETDEV_OFFLOAD_XSTATS_DISABLE, 7912 &info.info); 7913 err = notifier_to_errno(rc); 7914 if (err) 7915 goto free_stats; 7916 7917 return 0; 7918 7919 free_stats: 7920 kfree(dev->offload_xstats_l3); 7921 dev->offload_xstats_l3 = NULL; 7922 return err; 7923 } 7924 7925 int netdev_offload_xstats_enable(struct net_device *dev, 7926 enum netdev_offload_xstats_type type, 7927 struct netlink_ext_ack *extack) 7928 { 7929 ASSERT_RTNL(); 7930 7931 if (netdev_offload_xstats_enabled(dev, type)) 7932 return -EALREADY; 7933 7934 switch (type) { 7935 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 7936 return netdev_offload_xstats_enable_l3(dev, extack); 7937 } 7938 7939 WARN_ON(1); 7940 return -EINVAL; 7941 } 7942 EXPORT_SYMBOL(netdev_offload_xstats_enable); 7943 7944 static void netdev_offload_xstats_disable_l3(struct net_device *dev) 7945 { 7946 struct netdev_notifier_offload_xstats_info info = { 7947 .info.dev = dev, 7948 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, 7949 }; 7950 7951 call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_DISABLE, 7952 &info.info); 7953 kfree(dev->offload_xstats_l3); 7954 dev->offload_xstats_l3 = NULL; 7955 } 7956 7957 int netdev_offload_xstats_disable(struct net_device *dev, 7958 enum netdev_offload_xstats_type type) 7959 { 7960 ASSERT_RTNL(); 7961 7962 if (!netdev_offload_xstats_enabled(dev, type)) 7963 return -EALREADY; 7964 7965 switch (type) { 7966 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 7967 netdev_offload_xstats_disable_l3(dev); 7968 return 0; 7969 } 7970 7971 WARN_ON(1); 7972 return -EINVAL; 7973 } 7974 EXPORT_SYMBOL(netdev_offload_xstats_disable); 7975 7976 static void netdev_offload_xstats_disable_all(struct net_device *dev) 7977 { 7978 netdev_offload_xstats_disable(dev, NETDEV_OFFLOAD_XSTATS_TYPE_L3); 7979 } 7980 7981 static struct rtnl_hw_stats64 * 7982 netdev_offload_xstats_get_ptr(const struct net_device *dev, 7983 enum netdev_offload_xstats_type type) 7984 { 7985 switch (type) { 7986 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 7987 return dev->offload_xstats_l3; 7988 } 7989 7990 WARN_ON(1); 7991 return NULL; 7992 } 7993 7994 bool netdev_offload_xstats_enabled(const struct net_device *dev, 7995 enum netdev_offload_xstats_type type) 7996 { 7997 ASSERT_RTNL(); 7998 7999 return netdev_offload_xstats_get_ptr(dev, type); 8000 } 8001 EXPORT_SYMBOL(netdev_offload_xstats_enabled); 8002 8003 struct netdev_notifier_offload_xstats_ru { 8004 bool used; 8005 }; 8006 8007 struct netdev_notifier_offload_xstats_rd { 8008 struct rtnl_hw_stats64 stats; 8009 bool used; 8010 }; 8011 8012 static void netdev_hw_stats64_add(struct rtnl_hw_stats64 *dest, 8013 const struct rtnl_hw_stats64 *src) 8014 { 8015 dest->rx_packets += src->rx_packets; 8016 dest->tx_packets += src->tx_packets; 8017 dest->rx_bytes += src->rx_bytes; 8018 dest->tx_bytes += src->tx_bytes; 8019 dest->rx_errors += src->rx_errors; 8020 dest->tx_errors += src->tx_errors; 8021 dest->rx_dropped += src->rx_dropped; 8022 dest->tx_dropped += src->tx_dropped; 8023 dest->multicast += src->multicast; 8024 } 8025 8026 static int netdev_offload_xstats_get_used(struct net_device *dev, 8027 enum netdev_offload_xstats_type type, 8028 bool *p_used, 8029 struct netlink_ext_ack *extack) 8030 { 8031 struct netdev_notifier_offload_xstats_ru report_used = {}; 8032 struct netdev_notifier_offload_xstats_info info = { 8033 .info.dev = dev, 8034 .info.extack = extack, 8035 .type = type, 8036 .report_used = &report_used, 8037 }; 8038 int rc; 8039 8040 WARN_ON(!netdev_offload_xstats_enabled(dev, type)); 8041 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_USED, 8042 &info.info); 8043 *p_used = report_used.used; 8044 return notifier_to_errno(rc); 8045 } 8046 8047 static int netdev_offload_xstats_get_stats(struct net_device *dev, 8048 enum netdev_offload_xstats_type type, 8049 struct rtnl_hw_stats64 *p_stats, 8050 bool *p_used, 8051 struct netlink_ext_ack *extack) 8052 { 8053 struct netdev_notifier_offload_xstats_rd report_delta = {}; 8054 struct netdev_notifier_offload_xstats_info info = { 8055 .info.dev = dev, 8056 .info.extack = extack, 8057 .type = type, 8058 .report_delta = &report_delta, 8059 }; 8060 struct rtnl_hw_stats64 *stats; 8061 int rc; 8062 8063 stats = netdev_offload_xstats_get_ptr(dev, type); 8064 if (WARN_ON(!stats)) 8065 return -EINVAL; 8066 8067 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_DELTA, 8068 &info.info); 8069 8070 /* Cache whatever we got, even if there was an error, otherwise the 8071 * successful stats retrievals would get lost. 8072 */ 8073 netdev_hw_stats64_add(stats, &report_delta.stats); 8074 8075 if (p_stats) 8076 *p_stats = *stats; 8077 *p_used = report_delta.used; 8078 8079 return notifier_to_errno(rc); 8080 } 8081 8082 int netdev_offload_xstats_get(struct net_device *dev, 8083 enum netdev_offload_xstats_type type, 8084 struct rtnl_hw_stats64 *p_stats, bool *p_used, 8085 struct netlink_ext_ack *extack) 8086 { 8087 ASSERT_RTNL(); 8088 8089 if (p_stats) 8090 return netdev_offload_xstats_get_stats(dev, type, p_stats, 8091 p_used, extack); 8092 else 8093 return netdev_offload_xstats_get_used(dev, type, p_used, 8094 extack); 8095 } 8096 EXPORT_SYMBOL(netdev_offload_xstats_get); 8097 8098 void 8099 netdev_offload_xstats_report_delta(struct netdev_notifier_offload_xstats_rd *report_delta, 8100 const struct rtnl_hw_stats64 *stats) 8101 { 8102 report_delta->used = true; 8103 netdev_hw_stats64_add(&report_delta->stats, stats); 8104 } 8105 EXPORT_SYMBOL(netdev_offload_xstats_report_delta); 8106 8107 void 8108 netdev_offload_xstats_report_used(struct netdev_notifier_offload_xstats_ru *report_used) 8109 { 8110 report_used->used = true; 8111 } 8112 EXPORT_SYMBOL(netdev_offload_xstats_report_used); 8113 8114 void netdev_offload_xstats_push_delta(struct net_device *dev, 8115 enum netdev_offload_xstats_type type, 8116 const struct rtnl_hw_stats64 *p_stats) 8117 { 8118 struct rtnl_hw_stats64 *stats; 8119 8120 ASSERT_RTNL(); 8121 8122 stats = netdev_offload_xstats_get_ptr(dev, type); 8123 if (WARN_ON(!stats)) 8124 return; 8125 8126 netdev_hw_stats64_add(stats, p_stats); 8127 } 8128 EXPORT_SYMBOL(netdev_offload_xstats_push_delta); 8129 8130 /** 8131 * netdev_get_xmit_slave - Get the xmit slave of master device 8132 * @dev: device 8133 * @skb: The packet 8134 * @all_slaves: assume all the slaves are active 8135 * 8136 * The reference counters are not incremented so the caller must be 8137 * careful with locks. The caller must hold RCU lock. 8138 * %NULL is returned if no slave is found. 8139 */ 8140 8141 struct net_device *netdev_get_xmit_slave(struct net_device *dev, 8142 struct sk_buff *skb, 8143 bool all_slaves) 8144 { 8145 const struct net_device_ops *ops = dev->netdev_ops; 8146 8147 if (!ops->ndo_get_xmit_slave) 8148 return NULL; 8149 return ops->ndo_get_xmit_slave(dev, skb, all_slaves); 8150 } 8151 EXPORT_SYMBOL(netdev_get_xmit_slave); 8152 8153 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev, 8154 struct sock *sk) 8155 { 8156 const struct net_device_ops *ops = dev->netdev_ops; 8157 8158 if (!ops->ndo_sk_get_lower_dev) 8159 return NULL; 8160 return ops->ndo_sk_get_lower_dev(dev, sk); 8161 } 8162 8163 /** 8164 * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket 8165 * @dev: device 8166 * @sk: the socket 8167 * 8168 * %NULL is returned if no lower device is found. 8169 */ 8170 8171 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev, 8172 struct sock *sk) 8173 { 8174 struct net_device *lower; 8175 8176 lower = netdev_sk_get_lower_dev(dev, sk); 8177 while (lower) { 8178 dev = lower; 8179 lower = netdev_sk_get_lower_dev(dev, sk); 8180 } 8181 8182 return dev; 8183 } 8184 EXPORT_SYMBOL(netdev_sk_get_lowest_dev); 8185 8186 static void netdev_adjacent_add_links(struct net_device *dev) 8187 { 8188 struct netdev_adjacent *iter; 8189 8190 struct net *net = dev_net(dev); 8191 8192 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8193 if (!net_eq(net, dev_net(iter->dev))) 8194 continue; 8195 netdev_adjacent_sysfs_add(iter->dev, dev, 8196 &iter->dev->adj_list.lower); 8197 netdev_adjacent_sysfs_add(dev, iter->dev, 8198 &dev->adj_list.upper); 8199 } 8200 8201 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8202 if (!net_eq(net, dev_net(iter->dev))) 8203 continue; 8204 netdev_adjacent_sysfs_add(iter->dev, dev, 8205 &iter->dev->adj_list.upper); 8206 netdev_adjacent_sysfs_add(dev, iter->dev, 8207 &dev->adj_list.lower); 8208 } 8209 } 8210 8211 static void netdev_adjacent_del_links(struct net_device *dev) 8212 { 8213 struct netdev_adjacent *iter; 8214 8215 struct net *net = dev_net(dev); 8216 8217 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8218 if (!net_eq(net, dev_net(iter->dev))) 8219 continue; 8220 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8221 &iter->dev->adj_list.lower); 8222 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8223 &dev->adj_list.upper); 8224 } 8225 8226 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8227 if (!net_eq(net, dev_net(iter->dev))) 8228 continue; 8229 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8230 &iter->dev->adj_list.upper); 8231 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8232 &dev->adj_list.lower); 8233 } 8234 } 8235 8236 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) 8237 { 8238 struct netdev_adjacent *iter; 8239 8240 struct net *net = dev_net(dev); 8241 8242 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8243 if (!net_eq(net, dev_net(iter->dev))) 8244 continue; 8245 netdev_adjacent_sysfs_del(iter->dev, oldname, 8246 &iter->dev->adj_list.lower); 8247 netdev_adjacent_sysfs_add(iter->dev, dev, 8248 &iter->dev->adj_list.lower); 8249 } 8250 8251 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8252 if (!net_eq(net, dev_net(iter->dev))) 8253 continue; 8254 netdev_adjacent_sysfs_del(iter->dev, oldname, 8255 &iter->dev->adj_list.upper); 8256 netdev_adjacent_sysfs_add(iter->dev, dev, 8257 &iter->dev->adj_list.upper); 8258 } 8259 } 8260 8261 void *netdev_lower_dev_get_private(struct net_device *dev, 8262 struct net_device *lower_dev) 8263 { 8264 struct netdev_adjacent *lower; 8265 8266 if (!lower_dev) 8267 return NULL; 8268 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); 8269 if (!lower) 8270 return NULL; 8271 8272 return lower->private; 8273 } 8274 EXPORT_SYMBOL(netdev_lower_dev_get_private); 8275 8276 8277 /** 8278 * netdev_lower_state_changed - Dispatch event about lower device state change 8279 * @lower_dev: device 8280 * @lower_state_info: state to dispatch 8281 * 8282 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. 8283 * The caller must hold the RTNL lock. 8284 */ 8285 void netdev_lower_state_changed(struct net_device *lower_dev, 8286 void *lower_state_info) 8287 { 8288 struct netdev_notifier_changelowerstate_info changelowerstate_info = { 8289 .info.dev = lower_dev, 8290 }; 8291 8292 ASSERT_RTNL(); 8293 changelowerstate_info.lower_state_info = lower_state_info; 8294 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, 8295 &changelowerstate_info.info); 8296 } 8297 EXPORT_SYMBOL(netdev_lower_state_changed); 8298 8299 static void dev_change_rx_flags(struct net_device *dev, int flags) 8300 { 8301 const struct net_device_ops *ops = dev->netdev_ops; 8302 8303 if (ops->ndo_change_rx_flags) 8304 ops->ndo_change_rx_flags(dev, flags); 8305 } 8306 8307 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) 8308 { 8309 unsigned int old_flags = dev->flags; 8310 kuid_t uid; 8311 kgid_t gid; 8312 8313 ASSERT_RTNL(); 8314 8315 dev->flags |= IFF_PROMISC; 8316 dev->promiscuity += inc; 8317 if (dev->promiscuity == 0) { 8318 /* 8319 * Avoid overflow. 8320 * If inc causes overflow, untouch promisc and return error. 8321 */ 8322 if (inc < 0) 8323 dev->flags &= ~IFF_PROMISC; 8324 else { 8325 dev->promiscuity -= inc; 8326 netdev_warn(dev, "promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n"); 8327 return -EOVERFLOW; 8328 } 8329 } 8330 if (dev->flags != old_flags) { 8331 pr_info("device %s %s promiscuous mode\n", 8332 dev->name, 8333 dev->flags & IFF_PROMISC ? "entered" : "left"); 8334 if (audit_enabled) { 8335 current_uid_gid(&uid, &gid); 8336 audit_log(audit_context(), GFP_ATOMIC, 8337 AUDIT_ANOM_PROMISCUOUS, 8338 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", 8339 dev->name, (dev->flags & IFF_PROMISC), 8340 (old_flags & IFF_PROMISC), 8341 from_kuid(&init_user_ns, audit_get_loginuid(current)), 8342 from_kuid(&init_user_ns, uid), 8343 from_kgid(&init_user_ns, gid), 8344 audit_get_sessionid(current)); 8345 } 8346 8347 dev_change_rx_flags(dev, IFF_PROMISC); 8348 } 8349 if (notify) 8350 __dev_notify_flags(dev, old_flags, IFF_PROMISC); 8351 return 0; 8352 } 8353 8354 /** 8355 * dev_set_promiscuity - update promiscuity count on a device 8356 * @dev: device 8357 * @inc: modifier 8358 * 8359 * Add or remove promiscuity from a device. While the count in the device 8360 * remains above zero the interface remains promiscuous. Once it hits zero 8361 * the device reverts back to normal filtering operation. A negative inc 8362 * value is used to drop promiscuity on the device. 8363 * Return 0 if successful or a negative errno code on error. 8364 */ 8365 int dev_set_promiscuity(struct net_device *dev, int inc) 8366 { 8367 unsigned int old_flags = dev->flags; 8368 int err; 8369 8370 err = __dev_set_promiscuity(dev, inc, true); 8371 if (err < 0) 8372 return err; 8373 if (dev->flags != old_flags) 8374 dev_set_rx_mode(dev); 8375 return err; 8376 } 8377 EXPORT_SYMBOL(dev_set_promiscuity); 8378 8379 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) 8380 { 8381 unsigned int old_flags = dev->flags, old_gflags = dev->gflags; 8382 8383 ASSERT_RTNL(); 8384 8385 dev->flags |= IFF_ALLMULTI; 8386 dev->allmulti += inc; 8387 if (dev->allmulti == 0) { 8388 /* 8389 * Avoid overflow. 8390 * If inc causes overflow, untouch allmulti and return error. 8391 */ 8392 if (inc < 0) 8393 dev->flags &= ~IFF_ALLMULTI; 8394 else { 8395 dev->allmulti -= inc; 8396 netdev_warn(dev, "allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n"); 8397 return -EOVERFLOW; 8398 } 8399 } 8400 if (dev->flags ^ old_flags) { 8401 dev_change_rx_flags(dev, IFF_ALLMULTI); 8402 dev_set_rx_mode(dev); 8403 if (notify) 8404 __dev_notify_flags(dev, old_flags, 8405 dev->gflags ^ old_gflags); 8406 } 8407 return 0; 8408 } 8409 8410 /** 8411 * dev_set_allmulti - update allmulti count on a device 8412 * @dev: device 8413 * @inc: modifier 8414 * 8415 * Add or remove reception of all multicast frames to a device. While the 8416 * count in the device remains above zero the interface remains listening 8417 * to all interfaces. Once it hits zero the device reverts back to normal 8418 * filtering operation. A negative @inc value is used to drop the counter 8419 * when releasing a resource needing all multicasts. 8420 * Return 0 if successful or a negative errno code on error. 8421 */ 8422 8423 int dev_set_allmulti(struct net_device *dev, int inc) 8424 { 8425 return __dev_set_allmulti(dev, inc, true); 8426 } 8427 EXPORT_SYMBOL(dev_set_allmulti); 8428 8429 /* 8430 * Upload unicast and multicast address lists to device and 8431 * configure RX filtering. When the device doesn't support unicast 8432 * filtering it is put in promiscuous mode while unicast addresses 8433 * are present. 8434 */ 8435 void __dev_set_rx_mode(struct net_device *dev) 8436 { 8437 const struct net_device_ops *ops = dev->netdev_ops; 8438 8439 /* dev_open will call this function so the list will stay sane. */ 8440 if (!(dev->flags&IFF_UP)) 8441 return; 8442 8443 if (!netif_device_present(dev)) 8444 return; 8445 8446 if (!(dev->priv_flags & IFF_UNICAST_FLT)) { 8447 /* Unicast addresses changes may only happen under the rtnl, 8448 * therefore calling __dev_set_promiscuity here is safe. 8449 */ 8450 if (!netdev_uc_empty(dev) && !dev->uc_promisc) { 8451 __dev_set_promiscuity(dev, 1, false); 8452 dev->uc_promisc = true; 8453 } else if (netdev_uc_empty(dev) && dev->uc_promisc) { 8454 __dev_set_promiscuity(dev, -1, false); 8455 dev->uc_promisc = false; 8456 } 8457 } 8458 8459 if (ops->ndo_set_rx_mode) 8460 ops->ndo_set_rx_mode(dev); 8461 } 8462 8463 void dev_set_rx_mode(struct net_device *dev) 8464 { 8465 netif_addr_lock_bh(dev); 8466 __dev_set_rx_mode(dev); 8467 netif_addr_unlock_bh(dev); 8468 } 8469 8470 /** 8471 * dev_get_flags - get flags reported to userspace 8472 * @dev: device 8473 * 8474 * Get the combination of flag bits exported through APIs to userspace. 8475 */ 8476 unsigned int dev_get_flags(const struct net_device *dev) 8477 { 8478 unsigned int flags; 8479 8480 flags = (dev->flags & ~(IFF_PROMISC | 8481 IFF_ALLMULTI | 8482 IFF_RUNNING | 8483 IFF_LOWER_UP | 8484 IFF_DORMANT)) | 8485 (dev->gflags & (IFF_PROMISC | 8486 IFF_ALLMULTI)); 8487 8488 if (netif_running(dev)) { 8489 if (netif_oper_up(dev)) 8490 flags |= IFF_RUNNING; 8491 if (netif_carrier_ok(dev)) 8492 flags |= IFF_LOWER_UP; 8493 if (netif_dormant(dev)) 8494 flags |= IFF_DORMANT; 8495 } 8496 8497 return flags; 8498 } 8499 EXPORT_SYMBOL(dev_get_flags); 8500 8501 int __dev_change_flags(struct net_device *dev, unsigned int flags, 8502 struct netlink_ext_ack *extack) 8503 { 8504 unsigned int old_flags = dev->flags; 8505 int ret; 8506 8507 ASSERT_RTNL(); 8508 8509 /* 8510 * Set the flags on our device. 8511 */ 8512 8513 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | 8514 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | 8515 IFF_AUTOMEDIA)) | 8516 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | 8517 IFF_ALLMULTI)); 8518 8519 /* 8520 * Load in the correct multicast list now the flags have changed. 8521 */ 8522 8523 if ((old_flags ^ flags) & IFF_MULTICAST) 8524 dev_change_rx_flags(dev, IFF_MULTICAST); 8525 8526 dev_set_rx_mode(dev); 8527 8528 /* 8529 * Have we downed the interface. We handle IFF_UP ourselves 8530 * according to user attempts to set it, rather than blindly 8531 * setting it. 8532 */ 8533 8534 ret = 0; 8535 if ((old_flags ^ flags) & IFF_UP) { 8536 if (old_flags & IFF_UP) 8537 __dev_close(dev); 8538 else 8539 ret = __dev_open(dev, extack); 8540 } 8541 8542 if ((flags ^ dev->gflags) & IFF_PROMISC) { 8543 int inc = (flags & IFF_PROMISC) ? 1 : -1; 8544 unsigned int old_flags = dev->flags; 8545 8546 dev->gflags ^= IFF_PROMISC; 8547 8548 if (__dev_set_promiscuity(dev, inc, false) >= 0) 8549 if (dev->flags != old_flags) 8550 dev_set_rx_mode(dev); 8551 } 8552 8553 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI 8554 * is important. Some (broken) drivers set IFF_PROMISC, when 8555 * IFF_ALLMULTI is requested not asking us and not reporting. 8556 */ 8557 if ((flags ^ dev->gflags) & IFF_ALLMULTI) { 8558 int inc = (flags & IFF_ALLMULTI) ? 1 : -1; 8559 8560 dev->gflags ^= IFF_ALLMULTI; 8561 __dev_set_allmulti(dev, inc, false); 8562 } 8563 8564 return ret; 8565 } 8566 8567 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, 8568 unsigned int gchanges) 8569 { 8570 unsigned int changes = dev->flags ^ old_flags; 8571 8572 if (gchanges) 8573 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC); 8574 8575 if (changes & IFF_UP) { 8576 if (dev->flags & IFF_UP) 8577 call_netdevice_notifiers(NETDEV_UP, dev); 8578 else 8579 call_netdevice_notifiers(NETDEV_DOWN, dev); 8580 } 8581 8582 if (dev->flags & IFF_UP && 8583 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { 8584 struct netdev_notifier_change_info change_info = { 8585 .info = { 8586 .dev = dev, 8587 }, 8588 .flags_changed = changes, 8589 }; 8590 8591 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info); 8592 } 8593 } 8594 8595 /** 8596 * dev_change_flags - change device settings 8597 * @dev: device 8598 * @flags: device state flags 8599 * @extack: netlink extended ack 8600 * 8601 * Change settings on device based state flags. The flags are 8602 * in the userspace exported format. 8603 */ 8604 int dev_change_flags(struct net_device *dev, unsigned int flags, 8605 struct netlink_ext_ack *extack) 8606 { 8607 int ret; 8608 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; 8609 8610 ret = __dev_change_flags(dev, flags, extack); 8611 if (ret < 0) 8612 return ret; 8613 8614 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); 8615 __dev_notify_flags(dev, old_flags, changes); 8616 return ret; 8617 } 8618 EXPORT_SYMBOL(dev_change_flags); 8619 8620 int __dev_set_mtu(struct net_device *dev, int new_mtu) 8621 { 8622 const struct net_device_ops *ops = dev->netdev_ops; 8623 8624 if (ops->ndo_change_mtu) 8625 return ops->ndo_change_mtu(dev, new_mtu); 8626 8627 /* Pairs with all the lockless reads of dev->mtu in the stack */ 8628 WRITE_ONCE(dev->mtu, new_mtu); 8629 return 0; 8630 } 8631 EXPORT_SYMBOL(__dev_set_mtu); 8632 8633 int dev_validate_mtu(struct net_device *dev, int new_mtu, 8634 struct netlink_ext_ack *extack) 8635 { 8636 /* MTU must be positive, and in range */ 8637 if (new_mtu < 0 || new_mtu < dev->min_mtu) { 8638 NL_SET_ERR_MSG(extack, "mtu less than device minimum"); 8639 return -EINVAL; 8640 } 8641 8642 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { 8643 NL_SET_ERR_MSG(extack, "mtu greater than device maximum"); 8644 return -EINVAL; 8645 } 8646 return 0; 8647 } 8648 8649 /** 8650 * dev_set_mtu_ext - Change maximum transfer unit 8651 * @dev: device 8652 * @new_mtu: new transfer unit 8653 * @extack: netlink extended ack 8654 * 8655 * Change the maximum transfer size of the network device. 8656 */ 8657 int dev_set_mtu_ext(struct net_device *dev, int new_mtu, 8658 struct netlink_ext_ack *extack) 8659 { 8660 int err, orig_mtu; 8661 8662 if (new_mtu == dev->mtu) 8663 return 0; 8664 8665 err = dev_validate_mtu(dev, new_mtu, extack); 8666 if (err) 8667 return err; 8668 8669 if (!netif_device_present(dev)) 8670 return -ENODEV; 8671 8672 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); 8673 err = notifier_to_errno(err); 8674 if (err) 8675 return err; 8676 8677 orig_mtu = dev->mtu; 8678 err = __dev_set_mtu(dev, new_mtu); 8679 8680 if (!err) { 8681 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8682 orig_mtu); 8683 err = notifier_to_errno(err); 8684 if (err) { 8685 /* setting mtu back and notifying everyone again, 8686 * so that they have a chance to revert changes. 8687 */ 8688 __dev_set_mtu(dev, orig_mtu); 8689 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8690 new_mtu); 8691 } 8692 } 8693 return err; 8694 } 8695 8696 int dev_set_mtu(struct net_device *dev, int new_mtu) 8697 { 8698 struct netlink_ext_ack extack; 8699 int err; 8700 8701 memset(&extack, 0, sizeof(extack)); 8702 err = dev_set_mtu_ext(dev, new_mtu, &extack); 8703 if (err && extack._msg) 8704 net_err_ratelimited("%s: %s\n", dev->name, extack._msg); 8705 return err; 8706 } 8707 EXPORT_SYMBOL(dev_set_mtu); 8708 8709 /** 8710 * dev_change_tx_queue_len - Change TX queue length of a netdevice 8711 * @dev: device 8712 * @new_len: new tx queue length 8713 */ 8714 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len) 8715 { 8716 unsigned int orig_len = dev->tx_queue_len; 8717 int res; 8718 8719 if (new_len != (unsigned int)new_len) 8720 return -ERANGE; 8721 8722 if (new_len != orig_len) { 8723 dev->tx_queue_len = new_len; 8724 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev); 8725 res = notifier_to_errno(res); 8726 if (res) 8727 goto err_rollback; 8728 res = dev_qdisc_change_tx_queue_len(dev); 8729 if (res) 8730 goto err_rollback; 8731 } 8732 8733 return 0; 8734 8735 err_rollback: 8736 netdev_err(dev, "refused to change device tx_queue_len\n"); 8737 dev->tx_queue_len = orig_len; 8738 return res; 8739 } 8740 8741 /** 8742 * dev_set_group - Change group this device belongs to 8743 * @dev: device 8744 * @new_group: group this device should belong to 8745 */ 8746 void dev_set_group(struct net_device *dev, int new_group) 8747 { 8748 dev->group = new_group; 8749 } 8750 8751 /** 8752 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR. 8753 * @dev: device 8754 * @addr: new address 8755 * @extack: netlink extended ack 8756 */ 8757 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr, 8758 struct netlink_ext_ack *extack) 8759 { 8760 struct netdev_notifier_pre_changeaddr_info info = { 8761 .info.dev = dev, 8762 .info.extack = extack, 8763 .dev_addr = addr, 8764 }; 8765 int rc; 8766 8767 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info); 8768 return notifier_to_errno(rc); 8769 } 8770 EXPORT_SYMBOL(dev_pre_changeaddr_notify); 8771 8772 /** 8773 * dev_set_mac_address - Change Media Access Control Address 8774 * @dev: device 8775 * @sa: new address 8776 * @extack: netlink extended ack 8777 * 8778 * Change the hardware (MAC) address of the device 8779 */ 8780 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa, 8781 struct netlink_ext_ack *extack) 8782 { 8783 const struct net_device_ops *ops = dev->netdev_ops; 8784 int err; 8785 8786 if (!ops->ndo_set_mac_address) 8787 return -EOPNOTSUPP; 8788 if (sa->sa_family != dev->type) 8789 return -EINVAL; 8790 if (!netif_device_present(dev)) 8791 return -ENODEV; 8792 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack); 8793 if (err) 8794 return err; 8795 err = ops->ndo_set_mac_address(dev, sa); 8796 if (err) 8797 return err; 8798 dev->addr_assign_type = NET_ADDR_SET; 8799 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); 8800 add_device_randomness(dev->dev_addr, dev->addr_len); 8801 return 0; 8802 } 8803 EXPORT_SYMBOL(dev_set_mac_address); 8804 8805 static DECLARE_RWSEM(dev_addr_sem); 8806 8807 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa, 8808 struct netlink_ext_ack *extack) 8809 { 8810 int ret; 8811 8812 down_write(&dev_addr_sem); 8813 ret = dev_set_mac_address(dev, sa, extack); 8814 up_write(&dev_addr_sem); 8815 return ret; 8816 } 8817 EXPORT_SYMBOL(dev_set_mac_address_user); 8818 8819 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name) 8820 { 8821 size_t size = sizeof(sa->sa_data); 8822 struct net_device *dev; 8823 int ret = 0; 8824 8825 down_read(&dev_addr_sem); 8826 rcu_read_lock(); 8827 8828 dev = dev_get_by_name_rcu(net, dev_name); 8829 if (!dev) { 8830 ret = -ENODEV; 8831 goto unlock; 8832 } 8833 if (!dev->addr_len) 8834 memset(sa->sa_data, 0, size); 8835 else 8836 memcpy(sa->sa_data, dev->dev_addr, 8837 min_t(size_t, size, dev->addr_len)); 8838 sa->sa_family = dev->type; 8839 8840 unlock: 8841 rcu_read_unlock(); 8842 up_read(&dev_addr_sem); 8843 return ret; 8844 } 8845 EXPORT_SYMBOL(dev_get_mac_address); 8846 8847 /** 8848 * dev_change_carrier - Change device carrier 8849 * @dev: device 8850 * @new_carrier: new value 8851 * 8852 * Change device carrier 8853 */ 8854 int dev_change_carrier(struct net_device *dev, bool new_carrier) 8855 { 8856 const struct net_device_ops *ops = dev->netdev_ops; 8857 8858 if (!ops->ndo_change_carrier) 8859 return -EOPNOTSUPP; 8860 if (!netif_device_present(dev)) 8861 return -ENODEV; 8862 return ops->ndo_change_carrier(dev, new_carrier); 8863 } 8864 8865 /** 8866 * dev_get_phys_port_id - Get device physical port ID 8867 * @dev: device 8868 * @ppid: port ID 8869 * 8870 * Get device physical port ID 8871 */ 8872 int dev_get_phys_port_id(struct net_device *dev, 8873 struct netdev_phys_item_id *ppid) 8874 { 8875 const struct net_device_ops *ops = dev->netdev_ops; 8876 8877 if (!ops->ndo_get_phys_port_id) 8878 return -EOPNOTSUPP; 8879 return ops->ndo_get_phys_port_id(dev, ppid); 8880 } 8881 8882 /** 8883 * dev_get_phys_port_name - Get device physical port name 8884 * @dev: device 8885 * @name: port name 8886 * @len: limit of bytes to copy to name 8887 * 8888 * Get device physical port name 8889 */ 8890 int dev_get_phys_port_name(struct net_device *dev, 8891 char *name, size_t len) 8892 { 8893 const struct net_device_ops *ops = dev->netdev_ops; 8894 int err; 8895 8896 if (ops->ndo_get_phys_port_name) { 8897 err = ops->ndo_get_phys_port_name(dev, name, len); 8898 if (err != -EOPNOTSUPP) 8899 return err; 8900 } 8901 return devlink_compat_phys_port_name_get(dev, name, len); 8902 } 8903 8904 /** 8905 * dev_get_port_parent_id - Get the device's port parent identifier 8906 * @dev: network device 8907 * @ppid: pointer to a storage for the port's parent identifier 8908 * @recurse: allow/disallow recursion to lower devices 8909 * 8910 * Get the devices's port parent identifier 8911 */ 8912 int dev_get_port_parent_id(struct net_device *dev, 8913 struct netdev_phys_item_id *ppid, 8914 bool recurse) 8915 { 8916 const struct net_device_ops *ops = dev->netdev_ops; 8917 struct netdev_phys_item_id first = { }; 8918 struct net_device *lower_dev; 8919 struct list_head *iter; 8920 int err; 8921 8922 if (ops->ndo_get_port_parent_id) { 8923 err = ops->ndo_get_port_parent_id(dev, ppid); 8924 if (err != -EOPNOTSUPP) 8925 return err; 8926 } 8927 8928 err = devlink_compat_switch_id_get(dev, ppid); 8929 if (!recurse || err != -EOPNOTSUPP) 8930 return err; 8931 8932 netdev_for_each_lower_dev(dev, lower_dev, iter) { 8933 err = dev_get_port_parent_id(lower_dev, ppid, true); 8934 if (err) 8935 break; 8936 if (!first.id_len) 8937 first = *ppid; 8938 else if (memcmp(&first, ppid, sizeof(*ppid))) 8939 return -EOPNOTSUPP; 8940 } 8941 8942 return err; 8943 } 8944 EXPORT_SYMBOL(dev_get_port_parent_id); 8945 8946 /** 8947 * netdev_port_same_parent_id - Indicate if two network devices have 8948 * the same port parent identifier 8949 * @a: first network device 8950 * @b: second network device 8951 */ 8952 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b) 8953 { 8954 struct netdev_phys_item_id a_id = { }; 8955 struct netdev_phys_item_id b_id = { }; 8956 8957 if (dev_get_port_parent_id(a, &a_id, true) || 8958 dev_get_port_parent_id(b, &b_id, true)) 8959 return false; 8960 8961 return netdev_phys_item_id_same(&a_id, &b_id); 8962 } 8963 EXPORT_SYMBOL(netdev_port_same_parent_id); 8964 8965 /** 8966 * dev_change_proto_down - set carrier according to proto_down. 8967 * 8968 * @dev: device 8969 * @proto_down: new value 8970 */ 8971 int dev_change_proto_down(struct net_device *dev, bool proto_down) 8972 { 8973 if (!(dev->priv_flags & IFF_CHANGE_PROTO_DOWN)) 8974 return -EOPNOTSUPP; 8975 if (!netif_device_present(dev)) 8976 return -ENODEV; 8977 if (proto_down) 8978 netif_carrier_off(dev); 8979 else 8980 netif_carrier_on(dev); 8981 dev->proto_down = proto_down; 8982 return 0; 8983 } 8984 8985 /** 8986 * dev_change_proto_down_reason - proto down reason 8987 * 8988 * @dev: device 8989 * @mask: proto down mask 8990 * @value: proto down value 8991 */ 8992 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask, 8993 u32 value) 8994 { 8995 int b; 8996 8997 if (!mask) { 8998 dev->proto_down_reason = value; 8999 } else { 9000 for_each_set_bit(b, &mask, 32) { 9001 if (value & (1 << b)) 9002 dev->proto_down_reason |= BIT(b); 9003 else 9004 dev->proto_down_reason &= ~BIT(b); 9005 } 9006 } 9007 } 9008 9009 struct bpf_xdp_link { 9010 struct bpf_link link; 9011 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */ 9012 int flags; 9013 }; 9014 9015 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags) 9016 { 9017 if (flags & XDP_FLAGS_HW_MODE) 9018 return XDP_MODE_HW; 9019 if (flags & XDP_FLAGS_DRV_MODE) 9020 return XDP_MODE_DRV; 9021 if (flags & XDP_FLAGS_SKB_MODE) 9022 return XDP_MODE_SKB; 9023 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB; 9024 } 9025 9026 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode) 9027 { 9028 switch (mode) { 9029 case XDP_MODE_SKB: 9030 return generic_xdp_install; 9031 case XDP_MODE_DRV: 9032 case XDP_MODE_HW: 9033 return dev->netdev_ops->ndo_bpf; 9034 default: 9035 return NULL; 9036 } 9037 } 9038 9039 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev, 9040 enum bpf_xdp_mode mode) 9041 { 9042 return dev->xdp_state[mode].link; 9043 } 9044 9045 static struct bpf_prog *dev_xdp_prog(struct net_device *dev, 9046 enum bpf_xdp_mode mode) 9047 { 9048 struct bpf_xdp_link *link = dev_xdp_link(dev, mode); 9049 9050 if (link) 9051 return link->link.prog; 9052 return dev->xdp_state[mode].prog; 9053 } 9054 9055 u8 dev_xdp_prog_count(struct net_device *dev) 9056 { 9057 u8 count = 0; 9058 int i; 9059 9060 for (i = 0; i < __MAX_XDP_MODE; i++) 9061 if (dev->xdp_state[i].prog || dev->xdp_state[i].link) 9062 count++; 9063 return count; 9064 } 9065 EXPORT_SYMBOL_GPL(dev_xdp_prog_count); 9066 9067 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode) 9068 { 9069 struct bpf_prog *prog = dev_xdp_prog(dev, mode); 9070 9071 return prog ? prog->aux->id : 0; 9072 } 9073 9074 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode, 9075 struct bpf_xdp_link *link) 9076 { 9077 dev->xdp_state[mode].link = link; 9078 dev->xdp_state[mode].prog = NULL; 9079 } 9080 9081 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode, 9082 struct bpf_prog *prog) 9083 { 9084 dev->xdp_state[mode].link = NULL; 9085 dev->xdp_state[mode].prog = prog; 9086 } 9087 9088 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode, 9089 bpf_op_t bpf_op, struct netlink_ext_ack *extack, 9090 u32 flags, struct bpf_prog *prog) 9091 { 9092 struct netdev_bpf xdp; 9093 int err; 9094 9095 memset(&xdp, 0, sizeof(xdp)); 9096 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG; 9097 xdp.extack = extack; 9098 xdp.flags = flags; 9099 xdp.prog = prog; 9100 9101 /* Drivers assume refcnt is already incremented (i.e, prog pointer is 9102 * "moved" into driver), so they don't increment it on their own, but 9103 * they do decrement refcnt when program is detached or replaced. 9104 * Given net_device also owns link/prog, we need to bump refcnt here 9105 * to prevent drivers from underflowing it. 9106 */ 9107 if (prog) 9108 bpf_prog_inc(prog); 9109 err = bpf_op(dev, &xdp); 9110 if (err) { 9111 if (prog) 9112 bpf_prog_put(prog); 9113 return err; 9114 } 9115 9116 if (mode != XDP_MODE_HW) 9117 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog); 9118 9119 return 0; 9120 } 9121 9122 static void dev_xdp_uninstall(struct net_device *dev) 9123 { 9124 struct bpf_xdp_link *link; 9125 struct bpf_prog *prog; 9126 enum bpf_xdp_mode mode; 9127 bpf_op_t bpf_op; 9128 9129 ASSERT_RTNL(); 9130 9131 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) { 9132 prog = dev_xdp_prog(dev, mode); 9133 if (!prog) 9134 continue; 9135 9136 bpf_op = dev_xdp_bpf_op(dev, mode); 9137 if (!bpf_op) 9138 continue; 9139 9140 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9141 9142 /* auto-detach link from net device */ 9143 link = dev_xdp_link(dev, mode); 9144 if (link) 9145 link->dev = NULL; 9146 else 9147 bpf_prog_put(prog); 9148 9149 dev_xdp_set_link(dev, mode, NULL); 9150 } 9151 } 9152 9153 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack, 9154 struct bpf_xdp_link *link, struct bpf_prog *new_prog, 9155 struct bpf_prog *old_prog, u32 flags) 9156 { 9157 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES); 9158 struct bpf_prog *cur_prog; 9159 struct net_device *upper; 9160 struct list_head *iter; 9161 enum bpf_xdp_mode mode; 9162 bpf_op_t bpf_op; 9163 int err; 9164 9165 ASSERT_RTNL(); 9166 9167 /* either link or prog attachment, never both */ 9168 if (link && (new_prog || old_prog)) 9169 return -EINVAL; 9170 /* link supports only XDP mode flags */ 9171 if (link && (flags & ~XDP_FLAGS_MODES)) { 9172 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment"); 9173 return -EINVAL; 9174 } 9175 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */ 9176 if (num_modes > 1) { 9177 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set"); 9178 return -EINVAL; 9179 } 9180 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */ 9181 if (!num_modes && dev_xdp_prog_count(dev) > 1) { 9182 NL_SET_ERR_MSG(extack, 9183 "More than one program loaded, unset mode is ambiguous"); 9184 return -EINVAL; 9185 } 9186 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */ 9187 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) { 9188 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified"); 9189 return -EINVAL; 9190 } 9191 9192 mode = dev_xdp_mode(dev, flags); 9193 /* can't replace attached link */ 9194 if (dev_xdp_link(dev, mode)) { 9195 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link"); 9196 return -EBUSY; 9197 } 9198 9199 /* don't allow if an upper device already has a program */ 9200 netdev_for_each_upper_dev_rcu(dev, upper, iter) { 9201 if (dev_xdp_prog_count(upper) > 0) { 9202 NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program"); 9203 return -EEXIST; 9204 } 9205 } 9206 9207 cur_prog = dev_xdp_prog(dev, mode); 9208 /* can't replace attached prog with link */ 9209 if (link && cur_prog) { 9210 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link"); 9211 return -EBUSY; 9212 } 9213 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) { 9214 NL_SET_ERR_MSG(extack, "Active program does not match expected"); 9215 return -EEXIST; 9216 } 9217 9218 /* put effective new program into new_prog */ 9219 if (link) 9220 new_prog = link->link.prog; 9221 9222 if (new_prog) { 9223 bool offload = mode == XDP_MODE_HW; 9224 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB 9225 ? XDP_MODE_DRV : XDP_MODE_SKB; 9226 9227 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) { 9228 NL_SET_ERR_MSG(extack, "XDP program already attached"); 9229 return -EBUSY; 9230 } 9231 if (!offload && dev_xdp_prog(dev, other_mode)) { 9232 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time"); 9233 return -EEXIST; 9234 } 9235 if (!offload && bpf_prog_is_dev_bound(new_prog->aux)) { 9236 NL_SET_ERR_MSG(extack, "Using device-bound program without HW_MODE flag is not supported"); 9237 return -EINVAL; 9238 } 9239 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) { 9240 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device"); 9241 return -EINVAL; 9242 } 9243 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) { 9244 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device"); 9245 return -EINVAL; 9246 } 9247 } 9248 9249 /* don't call drivers if the effective program didn't change */ 9250 if (new_prog != cur_prog) { 9251 bpf_op = dev_xdp_bpf_op(dev, mode); 9252 if (!bpf_op) { 9253 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode"); 9254 return -EOPNOTSUPP; 9255 } 9256 9257 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog); 9258 if (err) 9259 return err; 9260 } 9261 9262 if (link) 9263 dev_xdp_set_link(dev, mode, link); 9264 else 9265 dev_xdp_set_prog(dev, mode, new_prog); 9266 if (cur_prog) 9267 bpf_prog_put(cur_prog); 9268 9269 return 0; 9270 } 9271 9272 static int dev_xdp_attach_link(struct net_device *dev, 9273 struct netlink_ext_ack *extack, 9274 struct bpf_xdp_link *link) 9275 { 9276 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags); 9277 } 9278 9279 static int dev_xdp_detach_link(struct net_device *dev, 9280 struct netlink_ext_ack *extack, 9281 struct bpf_xdp_link *link) 9282 { 9283 enum bpf_xdp_mode mode; 9284 bpf_op_t bpf_op; 9285 9286 ASSERT_RTNL(); 9287 9288 mode = dev_xdp_mode(dev, link->flags); 9289 if (dev_xdp_link(dev, mode) != link) 9290 return -EINVAL; 9291 9292 bpf_op = dev_xdp_bpf_op(dev, mode); 9293 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9294 dev_xdp_set_link(dev, mode, NULL); 9295 return 0; 9296 } 9297 9298 static void bpf_xdp_link_release(struct bpf_link *link) 9299 { 9300 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9301 9302 rtnl_lock(); 9303 9304 /* if racing with net_device's tear down, xdp_link->dev might be 9305 * already NULL, in which case link was already auto-detached 9306 */ 9307 if (xdp_link->dev) { 9308 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link)); 9309 xdp_link->dev = NULL; 9310 } 9311 9312 rtnl_unlock(); 9313 } 9314 9315 static int bpf_xdp_link_detach(struct bpf_link *link) 9316 { 9317 bpf_xdp_link_release(link); 9318 return 0; 9319 } 9320 9321 static void bpf_xdp_link_dealloc(struct bpf_link *link) 9322 { 9323 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9324 9325 kfree(xdp_link); 9326 } 9327 9328 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link, 9329 struct seq_file *seq) 9330 { 9331 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9332 u32 ifindex = 0; 9333 9334 rtnl_lock(); 9335 if (xdp_link->dev) 9336 ifindex = xdp_link->dev->ifindex; 9337 rtnl_unlock(); 9338 9339 seq_printf(seq, "ifindex:\t%u\n", ifindex); 9340 } 9341 9342 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link, 9343 struct bpf_link_info *info) 9344 { 9345 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9346 u32 ifindex = 0; 9347 9348 rtnl_lock(); 9349 if (xdp_link->dev) 9350 ifindex = xdp_link->dev->ifindex; 9351 rtnl_unlock(); 9352 9353 info->xdp.ifindex = ifindex; 9354 return 0; 9355 } 9356 9357 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog, 9358 struct bpf_prog *old_prog) 9359 { 9360 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9361 enum bpf_xdp_mode mode; 9362 bpf_op_t bpf_op; 9363 int err = 0; 9364 9365 rtnl_lock(); 9366 9367 /* link might have been auto-released already, so fail */ 9368 if (!xdp_link->dev) { 9369 err = -ENOLINK; 9370 goto out_unlock; 9371 } 9372 9373 if (old_prog && link->prog != old_prog) { 9374 err = -EPERM; 9375 goto out_unlock; 9376 } 9377 old_prog = link->prog; 9378 if (old_prog->type != new_prog->type || 9379 old_prog->expected_attach_type != new_prog->expected_attach_type) { 9380 err = -EINVAL; 9381 goto out_unlock; 9382 } 9383 9384 if (old_prog == new_prog) { 9385 /* no-op, don't disturb drivers */ 9386 bpf_prog_put(new_prog); 9387 goto out_unlock; 9388 } 9389 9390 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags); 9391 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode); 9392 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL, 9393 xdp_link->flags, new_prog); 9394 if (err) 9395 goto out_unlock; 9396 9397 old_prog = xchg(&link->prog, new_prog); 9398 bpf_prog_put(old_prog); 9399 9400 out_unlock: 9401 rtnl_unlock(); 9402 return err; 9403 } 9404 9405 static const struct bpf_link_ops bpf_xdp_link_lops = { 9406 .release = bpf_xdp_link_release, 9407 .dealloc = bpf_xdp_link_dealloc, 9408 .detach = bpf_xdp_link_detach, 9409 .show_fdinfo = bpf_xdp_link_show_fdinfo, 9410 .fill_link_info = bpf_xdp_link_fill_link_info, 9411 .update_prog = bpf_xdp_link_update, 9412 }; 9413 9414 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog) 9415 { 9416 struct net *net = current->nsproxy->net_ns; 9417 struct bpf_link_primer link_primer; 9418 struct bpf_xdp_link *link; 9419 struct net_device *dev; 9420 int err, fd; 9421 9422 rtnl_lock(); 9423 dev = dev_get_by_index(net, attr->link_create.target_ifindex); 9424 if (!dev) { 9425 rtnl_unlock(); 9426 return -EINVAL; 9427 } 9428 9429 link = kzalloc(sizeof(*link), GFP_USER); 9430 if (!link) { 9431 err = -ENOMEM; 9432 goto unlock; 9433 } 9434 9435 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog); 9436 link->dev = dev; 9437 link->flags = attr->link_create.flags; 9438 9439 err = bpf_link_prime(&link->link, &link_primer); 9440 if (err) { 9441 kfree(link); 9442 goto unlock; 9443 } 9444 9445 err = dev_xdp_attach_link(dev, NULL, link); 9446 rtnl_unlock(); 9447 9448 if (err) { 9449 link->dev = NULL; 9450 bpf_link_cleanup(&link_primer); 9451 goto out_put_dev; 9452 } 9453 9454 fd = bpf_link_settle(&link_primer); 9455 /* link itself doesn't hold dev's refcnt to not complicate shutdown */ 9456 dev_put(dev); 9457 return fd; 9458 9459 unlock: 9460 rtnl_unlock(); 9461 9462 out_put_dev: 9463 dev_put(dev); 9464 return err; 9465 } 9466 9467 /** 9468 * dev_change_xdp_fd - set or clear a bpf program for a device rx path 9469 * @dev: device 9470 * @extack: netlink extended ack 9471 * @fd: new program fd or negative value to clear 9472 * @expected_fd: old program fd that userspace expects to replace or clear 9473 * @flags: xdp-related flags 9474 * 9475 * Set or clear a bpf program for a device 9476 */ 9477 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, 9478 int fd, int expected_fd, u32 flags) 9479 { 9480 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags); 9481 struct bpf_prog *new_prog = NULL, *old_prog = NULL; 9482 int err; 9483 9484 ASSERT_RTNL(); 9485 9486 if (fd >= 0) { 9487 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP, 9488 mode != XDP_MODE_SKB); 9489 if (IS_ERR(new_prog)) 9490 return PTR_ERR(new_prog); 9491 } 9492 9493 if (expected_fd >= 0) { 9494 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP, 9495 mode != XDP_MODE_SKB); 9496 if (IS_ERR(old_prog)) { 9497 err = PTR_ERR(old_prog); 9498 old_prog = NULL; 9499 goto err_out; 9500 } 9501 } 9502 9503 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags); 9504 9505 err_out: 9506 if (err && new_prog) 9507 bpf_prog_put(new_prog); 9508 if (old_prog) 9509 bpf_prog_put(old_prog); 9510 return err; 9511 } 9512 9513 /** 9514 * dev_new_index - allocate an ifindex 9515 * @net: the applicable net namespace 9516 * 9517 * Returns a suitable unique value for a new device interface 9518 * number. The caller must hold the rtnl semaphore or the 9519 * dev_base_lock to be sure it remains unique. 9520 */ 9521 static int dev_new_index(struct net *net) 9522 { 9523 int ifindex = net->ifindex; 9524 9525 for (;;) { 9526 if (++ifindex <= 0) 9527 ifindex = 1; 9528 if (!__dev_get_by_index(net, ifindex)) 9529 return net->ifindex = ifindex; 9530 } 9531 } 9532 9533 /* Delayed registration/unregisteration */ 9534 LIST_HEAD(net_todo_list); 9535 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); 9536 9537 static void net_set_todo(struct net_device *dev) 9538 { 9539 list_add_tail(&dev->todo_list, &net_todo_list); 9540 atomic_inc(&dev_net(dev)->dev_unreg_count); 9541 } 9542 9543 static netdev_features_t netdev_sync_upper_features(struct net_device *lower, 9544 struct net_device *upper, netdev_features_t features) 9545 { 9546 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9547 netdev_features_t feature; 9548 int feature_bit; 9549 9550 for_each_netdev_feature(upper_disables, feature_bit) { 9551 feature = __NETIF_F_BIT(feature_bit); 9552 if (!(upper->wanted_features & feature) 9553 && (features & feature)) { 9554 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", 9555 &feature, upper->name); 9556 features &= ~feature; 9557 } 9558 } 9559 9560 return features; 9561 } 9562 9563 static void netdev_sync_lower_features(struct net_device *upper, 9564 struct net_device *lower, netdev_features_t features) 9565 { 9566 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9567 netdev_features_t feature; 9568 int feature_bit; 9569 9570 for_each_netdev_feature(upper_disables, feature_bit) { 9571 feature = __NETIF_F_BIT(feature_bit); 9572 if (!(features & feature) && (lower->features & feature)) { 9573 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", 9574 &feature, lower->name); 9575 lower->wanted_features &= ~feature; 9576 __netdev_update_features(lower); 9577 9578 if (unlikely(lower->features & feature)) 9579 netdev_WARN(upper, "failed to disable %pNF on %s!\n", 9580 &feature, lower->name); 9581 else 9582 netdev_features_change(lower); 9583 } 9584 } 9585 } 9586 9587 static netdev_features_t netdev_fix_features(struct net_device *dev, 9588 netdev_features_t features) 9589 { 9590 /* Fix illegal checksum combinations */ 9591 if ((features & NETIF_F_HW_CSUM) && 9592 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { 9593 netdev_warn(dev, "mixed HW and IP checksum settings.\n"); 9594 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); 9595 } 9596 9597 /* TSO requires that SG is present as well. */ 9598 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { 9599 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); 9600 features &= ~NETIF_F_ALL_TSO; 9601 } 9602 9603 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && 9604 !(features & NETIF_F_IP_CSUM)) { 9605 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); 9606 features &= ~NETIF_F_TSO; 9607 features &= ~NETIF_F_TSO_ECN; 9608 } 9609 9610 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && 9611 !(features & NETIF_F_IPV6_CSUM)) { 9612 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); 9613 features &= ~NETIF_F_TSO6; 9614 } 9615 9616 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ 9617 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) 9618 features &= ~NETIF_F_TSO_MANGLEID; 9619 9620 /* TSO ECN requires that TSO is present as well. */ 9621 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) 9622 features &= ~NETIF_F_TSO_ECN; 9623 9624 /* Software GSO depends on SG. */ 9625 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { 9626 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); 9627 features &= ~NETIF_F_GSO; 9628 } 9629 9630 /* GSO partial features require GSO partial be set */ 9631 if ((features & dev->gso_partial_features) && 9632 !(features & NETIF_F_GSO_PARTIAL)) { 9633 netdev_dbg(dev, 9634 "Dropping partially supported GSO features since no GSO partial.\n"); 9635 features &= ~dev->gso_partial_features; 9636 } 9637 9638 if (!(features & NETIF_F_RXCSUM)) { 9639 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet 9640 * successfully merged by hardware must also have the 9641 * checksum verified by hardware. If the user does not 9642 * want to enable RXCSUM, logically, we should disable GRO_HW. 9643 */ 9644 if (features & NETIF_F_GRO_HW) { 9645 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n"); 9646 features &= ~NETIF_F_GRO_HW; 9647 } 9648 } 9649 9650 /* LRO/HW-GRO features cannot be combined with RX-FCS */ 9651 if (features & NETIF_F_RXFCS) { 9652 if (features & NETIF_F_LRO) { 9653 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n"); 9654 features &= ~NETIF_F_LRO; 9655 } 9656 9657 if (features & NETIF_F_GRO_HW) { 9658 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n"); 9659 features &= ~NETIF_F_GRO_HW; 9660 } 9661 } 9662 9663 if ((features & NETIF_F_GRO_HW) && (features & NETIF_F_LRO)) { 9664 netdev_dbg(dev, "Dropping LRO feature since HW-GRO is requested.\n"); 9665 features &= ~NETIF_F_LRO; 9666 } 9667 9668 if (features & NETIF_F_HW_TLS_TX) { 9669 bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) == 9670 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM); 9671 bool hw_csum = features & NETIF_F_HW_CSUM; 9672 9673 if (!ip_csum && !hw_csum) { 9674 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n"); 9675 features &= ~NETIF_F_HW_TLS_TX; 9676 } 9677 } 9678 9679 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) { 9680 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n"); 9681 features &= ~NETIF_F_HW_TLS_RX; 9682 } 9683 9684 return features; 9685 } 9686 9687 int __netdev_update_features(struct net_device *dev) 9688 { 9689 struct net_device *upper, *lower; 9690 netdev_features_t features; 9691 struct list_head *iter; 9692 int err = -1; 9693 9694 ASSERT_RTNL(); 9695 9696 features = netdev_get_wanted_features(dev); 9697 9698 if (dev->netdev_ops->ndo_fix_features) 9699 features = dev->netdev_ops->ndo_fix_features(dev, features); 9700 9701 /* driver might be less strict about feature dependencies */ 9702 features = netdev_fix_features(dev, features); 9703 9704 /* some features can't be enabled if they're off on an upper device */ 9705 netdev_for_each_upper_dev_rcu(dev, upper, iter) 9706 features = netdev_sync_upper_features(dev, upper, features); 9707 9708 if (dev->features == features) 9709 goto sync_lower; 9710 9711 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", 9712 &dev->features, &features); 9713 9714 if (dev->netdev_ops->ndo_set_features) 9715 err = dev->netdev_ops->ndo_set_features(dev, features); 9716 else 9717 err = 0; 9718 9719 if (unlikely(err < 0)) { 9720 netdev_err(dev, 9721 "set_features() failed (%d); wanted %pNF, left %pNF\n", 9722 err, &features, &dev->features); 9723 /* return non-0 since some features might have changed and 9724 * it's better to fire a spurious notification than miss it 9725 */ 9726 return -1; 9727 } 9728 9729 sync_lower: 9730 /* some features must be disabled on lower devices when disabled 9731 * on an upper device (think: bonding master or bridge) 9732 */ 9733 netdev_for_each_lower_dev(dev, lower, iter) 9734 netdev_sync_lower_features(dev, lower, features); 9735 9736 if (!err) { 9737 netdev_features_t diff = features ^ dev->features; 9738 9739 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) { 9740 /* udp_tunnel_{get,drop}_rx_info both need 9741 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the 9742 * device, or they won't do anything. 9743 * Thus we need to update dev->features 9744 * *before* calling udp_tunnel_get_rx_info, 9745 * but *after* calling udp_tunnel_drop_rx_info. 9746 */ 9747 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) { 9748 dev->features = features; 9749 udp_tunnel_get_rx_info(dev); 9750 } else { 9751 udp_tunnel_drop_rx_info(dev); 9752 } 9753 } 9754 9755 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) { 9756 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) { 9757 dev->features = features; 9758 err |= vlan_get_rx_ctag_filter_info(dev); 9759 } else { 9760 vlan_drop_rx_ctag_filter_info(dev); 9761 } 9762 } 9763 9764 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) { 9765 if (features & NETIF_F_HW_VLAN_STAG_FILTER) { 9766 dev->features = features; 9767 err |= vlan_get_rx_stag_filter_info(dev); 9768 } else { 9769 vlan_drop_rx_stag_filter_info(dev); 9770 } 9771 } 9772 9773 dev->features = features; 9774 } 9775 9776 return err < 0 ? 0 : 1; 9777 } 9778 9779 /** 9780 * netdev_update_features - recalculate device features 9781 * @dev: the device to check 9782 * 9783 * Recalculate dev->features set and send notifications if it 9784 * has changed. Should be called after driver or hardware dependent 9785 * conditions might have changed that influence the features. 9786 */ 9787 void netdev_update_features(struct net_device *dev) 9788 { 9789 if (__netdev_update_features(dev)) 9790 netdev_features_change(dev); 9791 } 9792 EXPORT_SYMBOL(netdev_update_features); 9793 9794 /** 9795 * netdev_change_features - recalculate device features 9796 * @dev: the device to check 9797 * 9798 * Recalculate dev->features set and send notifications even 9799 * if they have not changed. Should be called instead of 9800 * netdev_update_features() if also dev->vlan_features might 9801 * have changed to allow the changes to be propagated to stacked 9802 * VLAN devices. 9803 */ 9804 void netdev_change_features(struct net_device *dev) 9805 { 9806 __netdev_update_features(dev); 9807 netdev_features_change(dev); 9808 } 9809 EXPORT_SYMBOL(netdev_change_features); 9810 9811 /** 9812 * netif_stacked_transfer_operstate - transfer operstate 9813 * @rootdev: the root or lower level device to transfer state from 9814 * @dev: the device to transfer operstate to 9815 * 9816 * Transfer operational state from root to device. This is normally 9817 * called when a stacking relationship exists between the root 9818 * device and the device(a leaf device). 9819 */ 9820 void netif_stacked_transfer_operstate(const struct net_device *rootdev, 9821 struct net_device *dev) 9822 { 9823 if (rootdev->operstate == IF_OPER_DORMANT) 9824 netif_dormant_on(dev); 9825 else 9826 netif_dormant_off(dev); 9827 9828 if (rootdev->operstate == IF_OPER_TESTING) 9829 netif_testing_on(dev); 9830 else 9831 netif_testing_off(dev); 9832 9833 if (netif_carrier_ok(rootdev)) 9834 netif_carrier_on(dev); 9835 else 9836 netif_carrier_off(dev); 9837 } 9838 EXPORT_SYMBOL(netif_stacked_transfer_operstate); 9839 9840 static int netif_alloc_rx_queues(struct net_device *dev) 9841 { 9842 unsigned int i, count = dev->num_rx_queues; 9843 struct netdev_rx_queue *rx; 9844 size_t sz = count * sizeof(*rx); 9845 int err = 0; 9846 9847 BUG_ON(count < 1); 9848 9849 rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 9850 if (!rx) 9851 return -ENOMEM; 9852 9853 dev->_rx = rx; 9854 9855 for (i = 0; i < count; i++) { 9856 rx[i].dev = dev; 9857 9858 /* XDP RX-queue setup */ 9859 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0); 9860 if (err < 0) 9861 goto err_rxq_info; 9862 } 9863 return 0; 9864 9865 err_rxq_info: 9866 /* Rollback successful reg's and free other resources */ 9867 while (i--) 9868 xdp_rxq_info_unreg(&rx[i].xdp_rxq); 9869 kvfree(dev->_rx); 9870 dev->_rx = NULL; 9871 return err; 9872 } 9873 9874 static void netif_free_rx_queues(struct net_device *dev) 9875 { 9876 unsigned int i, count = dev->num_rx_queues; 9877 9878 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */ 9879 if (!dev->_rx) 9880 return; 9881 9882 for (i = 0; i < count; i++) 9883 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq); 9884 9885 kvfree(dev->_rx); 9886 } 9887 9888 static void netdev_init_one_queue(struct net_device *dev, 9889 struct netdev_queue *queue, void *_unused) 9890 { 9891 /* Initialize queue lock */ 9892 spin_lock_init(&queue->_xmit_lock); 9893 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); 9894 queue->xmit_lock_owner = -1; 9895 netdev_queue_numa_node_write(queue, NUMA_NO_NODE); 9896 queue->dev = dev; 9897 #ifdef CONFIG_BQL 9898 dql_init(&queue->dql, HZ); 9899 #endif 9900 } 9901 9902 static void netif_free_tx_queues(struct net_device *dev) 9903 { 9904 kvfree(dev->_tx); 9905 } 9906 9907 static int netif_alloc_netdev_queues(struct net_device *dev) 9908 { 9909 unsigned int count = dev->num_tx_queues; 9910 struct netdev_queue *tx; 9911 size_t sz = count * sizeof(*tx); 9912 9913 if (count < 1 || count > 0xffff) 9914 return -EINVAL; 9915 9916 tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 9917 if (!tx) 9918 return -ENOMEM; 9919 9920 dev->_tx = tx; 9921 9922 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); 9923 spin_lock_init(&dev->tx_global_lock); 9924 9925 return 0; 9926 } 9927 9928 void netif_tx_stop_all_queues(struct net_device *dev) 9929 { 9930 unsigned int i; 9931 9932 for (i = 0; i < dev->num_tx_queues; i++) { 9933 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 9934 9935 netif_tx_stop_queue(txq); 9936 } 9937 } 9938 EXPORT_SYMBOL(netif_tx_stop_all_queues); 9939 9940 /** 9941 * register_netdevice() - register a network device 9942 * @dev: device to register 9943 * 9944 * Take a prepared network device structure and make it externally accessible. 9945 * A %NETDEV_REGISTER message is sent to the netdev notifier chain. 9946 * Callers must hold the rtnl lock - you may want register_netdev() 9947 * instead of this. 9948 */ 9949 int register_netdevice(struct net_device *dev) 9950 { 9951 int ret; 9952 struct net *net = dev_net(dev); 9953 9954 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE < 9955 NETDEV_FEATURE_COUNT); 9956 BUG_ON(dev_boot_phase); 9957 ASSERT_RTNL(); 9958 9959 might_sleep(); 9960 9961 /* When net_device's are persistent, this will be fatal. */ 9962 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); 9963 BUG_ON(!net); 9964 9965 ret = ethtool_check_ops(dev->ethtool_ops); 9966 if (ret) 9967 return ret; 9968 9969 spin_lock_init(&dev->addr_list_lock); 9970 netdev_set_addr_lockdep_class(dev); 9971 9972 ret = dev_get_valid_name(net, dev, dev->name); 9973 if (ret < 0) 9974 goto out; 9975 9976 ret = -ENOMEM; 9977 dev->name_node = netdev_name_node_head_alloc(dev); 9978 if (!dev->name_node) 9979 goto out; 9980 9981 /* Init, if this function is available */ 9982 if (dev->netdev_ops->ndo_init) { 9983 ret = dev->netdev_ops->ndo_init(dev); 9984 if (ret) { 9985 if (ret > 0) 9986 ret = -EIO; 9987 goto err_free_name; 9988 } 9989 } 9990 9991 if (((dev->hw_features | dev->features) & 9992 NETIF_F_HW_VLAN_CTAG_FILTER) && 9993 (!dev->netdev_ops->ndo_vlan_rx_add_vid || 9994 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { 9995 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); 9996 ret = -EINVAL; 9997 goto err_uninit; 9998 } 9999 10000 ret = -EBUSY; 10001 if (!dev->ifindex) 10002 dev->ifindex = dev_new_index(net); 10003 else if (__dev_get_by_index(net, dev->ifindex)) 10004 goto err_uninit; 10005 10006 /* Transfer changeable features to wanted_features and enable 10007 * software offloads (GSO and GRO). 10008 */ 10009 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF); 10010 dev->features |= NETIF_F_SOFT_FEATURES; 10011 10012 if (dev->udp_tunnel_nic_info) { 10013 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10014 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10015 } 10016 10017 dev->wanted_features = dev->features & dev->hw_features; 10018 10019 if (!(dev->flags & IFF_LOOPBACK)) 10020 dev->hw_features |= NETIF_F_NOCACHE_COPY; 10021 10022 /* If IPv4 TCP segmentation offload is supported we should also 10023 * allow the device to enable segmenting the frame with the option 10024 * of ignoring a static IP ID value. This doesn't enable the 10025 * feature itself but allows the user to enable it later. 10026 */ 10027 if (dev->hw_features & NETIF_F_TSO) 10028 dev->hw_features |= NETIF_F_TSO_MANGLEID; 10029 if (dev->vlan_features & NETIF_F_TSO) 10030 dev->vlan_features |= NETIF_F_TSO_MANGLEID; 10031 if (dev->mpls_features & NETIF_F_TSO) 10032 dev->mpls_features |= NETIF_F_TSO_MANGLEID; 10033 if (dev->hw_enc_features & NETIF_F_TSO) 10034 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; 10035 10036 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. 10037 */ 10038 dev->vlan_features |= NETIF_F_HIGHDMA; 10039 10040 /* Make NETIF_F_SG inheritable to tunnel devices. 10041 */ 10042 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; 10043 10044 /* Make NETIF_F_SG inheritable to MPLS. 10045 */ 10046 dev->mpls_features |= NETIF_F_SG; 10047 10048 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); 10049 ret = notifier_to_errno(ret); 10050 if (ret) 10051 goto err_uninit; 10052 10053 ret = netdev_register_kobject(dev); 10054 write_lock(&dev_base_lock); 10055 dev->reg_state = ret ? NETREG_UNREGISTERED : NETREG_REGISTERED; 10056 write_unlock(&dev_base_lock); 10057 if (ret) 10058 goto err_uninit; 10059 10060 __netdev_update_features(dev); 10061 10062 /* 10063 * Default initial state at registry is that the 10064 * device is present. 10065 */ 10066 10067 set_bit(__LINK_STATE_PRESENT, &dev->state); 10068 10069 linkwatch_init_dev(dev); 10070 10071 dev_init_scheduler(dev); 10072 10073 netdev_hold(dev, &dev->dev_registered_tracker, GFP_KERNEL); 10074 list_netdevice(dev); 10075 10076 add_device_randomness(dev->dev_addr, dev->addr_len); 10077 10078 /* If the device has permanent device address, driver should 10079 * set dev_addr and also addr_assign_type should be set to 10080 * NET_ADDR_PERM (default value). 10081 */ 10082 if (dev->addr_assign_type == NET_ADDR_PERM) 10083 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); 10084 10085 /* Notify protocols, that a new device appeared. */ 10086 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); 10087 ret = notifier_to_errno(ret); 10088 if (ret) { 10089 /* Expect explicit free_netdev() on failure */ 10090 dev->needs_free_netdev = false; 10091 unregister_netdevice_queue(dev, NULL); 10092 goto out; 10093 } 10094 /* 10095 * Prevent userspace races by waiting until the network 10096 * device is fully setup before sending notifications. 10097 */ 10098 if (!dev->rtnl_link_ops || 10099 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 10100 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); 10101 10102 out: 10103 return ret; 10104 10105 err_uninit: 10106 if (dev->netdev_ops->ndo_uninit) 10107 dev->netdev_ops->ndo_uninit(dev); 10108 if (dev->priv_destructor) 10109 dev->priv_destructor(dev); 10110 err_free_name: 10111 netdev_name_node_free(dev->name_node); 10112 goto out; 10113 } 10114 EXPORT_SYMBOL(register_netdevice); 10115 10116 /** 10117 * init_dummy_netdev - init a dummy network device for NAPI 10118 * @dev: device to init 10119 * 10120 * This takes a network device structure and initialize the minimum 10121 * amount of fields so it can be used to schedule NAPI polls without 10122 * registering a full blown interface. This is to be used by drivers 10123 * that need to tie several hardware interfaces to a single NAPI 10124 * poll scheduler due to HW limitations. 10125 */ 10126 int init_dummy_netdev(struct net_device *dev) 10127 { 10128 /* Clear everything. Note we don't initialize spinlocks 10129 * are they aren't supposed to be taken by any of the 10130 * NAPI code and this dummy netdev is supposed to be 10131 * only ever used for NAPI polls 10132 */ 10133 memset(dev, 0, sizeof(struct net_device)); 10134 10135 /* make sure we BUG if trying to hit standard 10136 * register/unregister code path 10137 */ 10138 dev->reg_state = NETREG_DUMMY; 10139 10140 /* NAPI wants this */ 10141 INIT_LIST_HEAD(&dev->napi_list); 10142 10143 /* a dummy interface is started by default */ 10144 set_bit(__LINK_STATE_PRESENT, &dev->state); 10145 set_bit(__LINK_STATE_START, &dev->state); 10146 10147 /* napi_busy_loop stats accounting wants this */ 10148 dev_net_set(dev, &init_net); 10149 10150 /* Note : We dont allocate pcpu_refcnt for dummy devices, 10151 * because users of this 'device' dont need to change 10152 * its refcount. 10153 */ 10154 10155 return 0; 10156 } 10157 EXPORT_SYMBOL_GPL(init_dummy_netdev); 10158 10159 10160 /** 10161 * register_netdev - register a network device 10162 * @dev: device to register 10163 * 10164 * Take a completed network device structure and add it to the kernel 10165 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 10166 * chain. 0 is returned on success. A negative errno code is returned 10167 * on a failure to set up the device, or if the name is a duplicate. 10168 * 10169 * This is a wrapper around register_netdevice that takes the rtnl semaphore 10170 * and expands the device name if you passed a format string to 10171 * alloc_netdev. 10172 */ 10173 int register_netdev(struct net_device *dev) 10174 { 10175 int err; 10176 10177 if (rtnl_lock_killable()) 10178 return -EINTR; 10179 err = register_netdevice(dev); 10180 rtnl_unlock(); 10181 return err; 10182 } 10183 EXPORT_SYMBOL(register_netdev); 10184 10185 int netdev_refcnt_read(const struct net_device *dev) 10186 { 10187 #ifdef CONFIG_PCPU_DEV_REFCNT 10188 int i, refcnt = 0; 10189 10190 for_each_possible_cpu(i) 10191 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); 10192 return refcnt; 10193 #else 10194 return refcount_read(&dev->dev_refcnt); 10195 #endif 10196 } 10197 EXPORT_SYMBOL(netdev_refcnt_read); 10198 10199 int netdev_unregister_timeout_secs __read_mostly = 10; 10200 10201 #define WAIT_REFS_MIN_MSECS 1 10202 #define WAIT_REFS_MAX_MSECS 250 10203 /** 10204 * netdev_wait_allrefs_any - wait until all references are gone. 10205 * @list: list of net_devices to wait on 10206 * 10207 * This is called when unregistering network devices. 10208 * 10209 * Any protocol or device that holds a reference should register 10210 * for netdevice notification, and cleanup and put back the 10211 * reference if they receive an UNREGISTER event. 10212 * We can get stuck here if buggy protocols don't correctly 10213 * call dev_put. 10214 */ 10215 static struct net_device *netdev_wait_allrefs_any(struct list_head *list) 10216 { 10217 unsigned long rebroadcast_time, warning_time; 10218 struct net_device *dev; 10219 int wait = 0; 10220 10221 rebroadcast_time = warning_time = jiffies; 10222 10223 list_for_each_entry(dev, list, todo_list) 10224 if (netdev_refcnt_read(dev) == 1) 10225 return dev; 10226 10227 while (true) { 10228 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { 10229 rtnl_lock(); 10230 10231 /* Rebroadcast unregister notification */ 10232 list_for_each_entry(dev, list, todo_list) 10233 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10234 10235 __rtnl_unlock(); 10236 rcu_barrier(); 10237 rtnl_lock(); 10238 10239 list_for_each_entry(dev, list, todo_list) 10240 if (test_bit(__LINK_STATE_LINKWATCH_PENDING, 10241 &dev->state)) { 10242 /* We must not have linkwatch events 10243 * pending on unregister. If this 10244 * happens, we simply run the queue 10245 * unscheduled, resulting in a noop 10246 * for this device. 10247 */ 10248 linkwatch_run_queue(); 10249 break; 10250 } 10251 10252 __rtnl_unlock(); 10253 10254 rebroadcast_time = jiffies; 10255 } 10256 10257 if (!wait) { 10258 rcu_barrier(); 10259 wait = WAIT_REFS_MIN_MSECS; 10260 } else { 10261 msleep(wait); 10262 wait = min(wait << 1, WAIT_REFS_MAX_MSECS); 10263 } 10264 10265 list_for_each_entry(dev, list, todo_list) 10266 if (netdev_refcnt_read(dev) == 1) 10267 return dev; 10268 10269 if (time_after(jiffies, warning_time + 10270 READ_ONCE(netdev_unregister_timeout_secs) * HZ)) { 10271 list_for_each_entry(dev, list, todo_list) { 10272 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", 10273 dev->name, netdev_refcnt_read(dev)); 10274 ref_tracker_dir_print(&dev->refcnt_tracker, 10); 10275 } 10276 10277 warning_time = jiffies; 10278 } 10279 } 10280 } 10281 10282 /* The sequence is: 10283 * 10284 * rtnl_lock(); 10285 * ... 10286 * register_netdevice(x1); 10287 * register_netdevice(x2); 10288 * ... 10289 * unregister_netdevice(y1); 10290 * unregister_netdevice(y2); 10291 * ... 10292 * rtnl_unlock(); 10293 * free_netdev(y1); 10294 * free_netdev(y2); 10295 * 10296 * We are invoked by rtnl_unlock(). 10297 * This allows us to deal with problems: 10298 * 1) We can delete sysfs objects which invoke hotplug 10299 * without deadlocking with linkwatch via keventd. 10300 * 2) Since we run with the RTNL semaphore not held, we can sleep 10301 * safely in order to wait for the netdev refcnt to drop to zero. 10302 * 10303 * We must not return until all unregister events added during 10304 * the interval the lock was held have been completed. 10305 */ 10306 void netdev_run_todo(void) 10307 { 10308 struct net_device *dev, *tmp; 10309 struct list_head list; 10310 #ifdef CONFIG_LOCKDEP 10311 struct list_head unlink_list; 10312 10313 list_replace_init(&net_unlink_list, &unlink_list); 10314 10315 while (!list_empty(&unlink_list)) { 10316 struct net_device *dev = list_first_entry(&unlink_list, 10317 struct net_device, 10318 unlink_list); 10319 list_del_init(&dev->unlink_list); 10320 dev->nested_level = dev->lower_level - 1; 10321 } 10322 #endif 10323 10324 /* Snapshot list, allow later requests */ 10325 list_replace_init(&net_todo_list, &list); 10326 10327 __rtnl_unlock(); 10328 10329 /* Wait for rcu callbacks to finish before next phase */ 10330 if (!list_empty(&list)) 10331 rcu_barrier(); 10332 10333 list_for_each_entry_safe(dev, tmp, &list, todo_list) { 10334 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { 10335 netdev_WARN(dev, "run_todo but not unregistering\n"); 10336 list_del(&dev->todo_list); 10337 continue; 10338 } 10339 10340 write_lock(&dev_base_lock); 10341 dev->reg_state = NETREG_UNREGISTERED; 10342 write_unlock(&dev_base_lock); 10343 linkwatch_forget_dev(dev); 10344 } 10345 10346 while (!list_empty(&list)) { 10347 dev = netdev_wait_allrefs_any(&list); 10348 list_del(&dev->todo_list); 10349 10350 /* paranoia */ 10351 BUG_ON(netdev_refcnt_read(dev) != 1); 10352 BUG_ON(!list_empty(&dev->ptype_all)); 10353 BUG_ON(!list_empty(&dev->ptype_specific)); 10354 WARN_ON(rcu_access_pointer(dev->ip_ptr)); 10355 WARN_ON(rcu_access_pointer(dev->ip6_ptr)); 10356 10357 if (dev->priv_destructor) 10358 dev->priv_destructor(dev); 10359 if (dev->needs_free_netdev) 10360 free_netdev(dev); 10361 10362 if (atomic_dec_and_test(&dev_net(dev)->dev_unreg_count)) 10363 wake_up(&netdev_unregistering_wq); 10364 10365 /* Free network device */ 10366 kobject_put(&dev->dev.kobj); 10367 } 10368 } 10369 10370 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has 10371 * all the same fields in the same order as net_device_stats, with only 10372 * the type differing, but rtnl_link_stats64 may have additional fields 10373 * at the end for newer counters. 10374 */ 10375 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, 10376 const struct net_device_stats *netdev_stats) 10377 { 10378 #if BITS_PER_LONG == 64 10379 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats)); 10380 memcpy(stats64, netdev_stats, sizeof(*netdev_stats)); 10381 /* zero out counters that only exist in rtnl_link_stats64 */ 10382 memset((char *)stats64 + sizeof(*netdev_stats), 0, 10383 sizeof(*stats64) - sizeof(*netdev_stats)); 10384 #else 10385 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long); 10386 const unsigned long *src = (const unsigned long *)netdev_stats; 10387 u64 *dst = (u64 *)stats64; 10388 10389 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); 10390 for (i = 0; i < n; i++) 10391 dst[i] = src[i]; 10392 /* zero out counters that only exist in rtnl_link_stats64 */ 10393 memset((char *)stats64 + n * sizeof(u64), 0, 10394 sizeof(*stats64) - n * sizeof(u64)); 10395 #endif 10396 } 10397 EXPORT_SYMBOL(netdev_stats_to_stats64); 10398 10399 struct net_device_core_stats __percpu *netdev_core_stats_alloc(struct net_device *dev) 10400 { 10401 struct net_device_core_stats __percpu *p; 10402 10403 p = alloc_percpu_gfp(struct net_device_core_stats, 10404 GFP_ATOMIC | __GFP_NOWARN); 10405 10406 if (p && cmpxchg(&dev->core_stats, NULL, p)) 10407 free_percpu(p); 10408 10409 /* This READ_ONCE() pairs with the cmpxchg() above */ 10410 return READ_ONCE(dev->core_stats); 10411 } 10412 EXPORT_SYMBOL(netdev_core_stats_alloc); 10413 10414 /** 10415 * dev_get_stats - get network device statistics 10416 * @dev: device to get statistics from 10417 * @storage: place to store stats 10418 * 10419 * Get network statistics from device. Return @storage. 10420 * The device driver may provide its own method by setting 10421 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; 10422 * otherwise the internal statistics structure is used. 10423 */ 10424 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, 10425 struct rtnl_link_stats64 *storage) 10426 { 10427 const struct net_device_ops *ops = dev->netdev_ops; 10428 const struct net_device_core_stats __percpu *p; 10429 10430 if (ops->ndo_get_stats64) { 10431 memset(storage, 0, sizeof(*storage)); 10432 ops->ndo_get_stats64(dev, storage); 10433 } else if (ops->ndo_get_stats) { 10434 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); 10435 } else { 10436 netdev_stats_to_stats64(storage, &dev->stats); 10437 } 10438 10439 /* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */ 10440 p = READ_ONCE(dev->core_stats); 10441 if (p) { 10442 const struct net_device_core_stats *core_stats; 10443 int i; 10444 10445 for_each_possible_cpu(i) { 10446 core_stats = per_cpu_ptr(p, i); 10447 storage->rx_dropped += READ_ONCE(core_stats->rx_dropped); 10448 storage->tx_dropped += READ_ONCE(core_stats->tx_dropped); 10449 storage->rx_nohandler += READ_ONCE(core_stats->rx_nohandler); 10450 storage->rx_otherhost_dropped += READ_ONCE(core_stats->rx_otherhost_dropped); 10451 } 10452 } 10453 return storage; 10454 } 10455 EXPORT_SYMBOL(dev_get_stats); 10456 10457 /** 10458 * dev_fetch_sw_netstats - get per-cpu network device statistics 10459 * @s: place to store stats 10460 * @netstats: per-cpu network stats to read from 10461 * 10462 * Read per-cpu network statistics and populate the related fields in @s. 10463 */ 10464 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s, 10465 const struct pcpu_sw_netstats __percpu *netstats) 10466 { 10467 int cpu; 10468 10469 for_each_possible_cpu(cpu) { 10470 u64 rx_packets, rx_bytes, tx_packets, tx_bytes; 10471 const struct pcpu_sw_netstats *stats; 10472 unsigned int start; 10473 10474 stats = per_cpu_ptr(netstats, cpu); 10475 do { 10476 start = u64_stats_fetch_begin_irq(&stats->syncp); 10477 rx_packets = u64_stats_read(&stats->rx_packets); 10478 rx_bytes = u64_stats_read(&stats->rx_bytes); 10479 tx_packets = u64_stats_read(&stats->tx_packets); 10480 tx_bytes = u64_stats_read(&stats->tx_bytes); 10481 } while (u64_stats_fetch_retry_irq(&stats->syncp, start)); 10482 10483 s->rx_packets += rx_packets; 10484 s->rx_bytes += rx_bytes; 10485 s->tx_packets += tx_packets; 10486 s->tx_bytes += tx_bytes; 10487 } 10488 } 10489 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats); 10490 10491 /** 10492 * dev_get_tstats64 - ndo_get_stats64 implementation 10493 * @dev: device to get statistics from 10494 * @s: place to store stats 10495 * 10496 * Populate @s from dev->stats and dev->tstats. Can be used as 10497 * ndo_get_stats64() callback. 10498 */ 10499 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s) 10500 { 10501 netdev_stats_to_stats64(s, &dev->stats); 10502 dev_fetch_sw_netstats(s, dev->tstats); 10503 } 10504 EXPORT_SYMBOL_GPL(dev_get_tstats64); 10505 10506 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) 10507 { 10508 struct netdev_queue *queue = dev_ingress_queue(dev); 10509 10510 #ifdef CONFIG_NET_CLS_ACT 10511 if (queue) 10512 return queue; 10513 queue = kzalloc(sizeof(*queue), GFP_KERNEL); 10514 if (!queue) 10515 return NULL; 10516 netdev_init_one_queue(dev, queue, NULL); 10517 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); 10518 queue->qdisc_sleeping = &noop_qdisc; 10519 rcu_assign_pointer(dev->ingress_queue, queue); 10520 #endif 10521 return queue; 10522 } 10523 10524 static const struct ethtool_ops default_ethtool_ops; 10525 10526 void netdev_set_default_ethtool_ops(struct net_device *dev, 10527 const struct ethtool_ops *ops) 10528 { 10529 if (dev->ethtool_ops == &default_ethtool_ops) 10530 dev->ethtool_ops = ops; 10531 } 10532 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); 10533 10534 void netdev_freemem(struct net_device *dev) 10535 { 10536 char *addr = (char *)dev - dev->padded; 10537 10538 kvfree(addr); 10539 } 10540 10541 /** 10542 * alloc_netdev_mqs - allocate network device 10543 * @sizeof_priv: size of private data to allocate space for 10544 * @name: device name format string 10545 * @name_assign_type: origin of device name 10546 * @setup: callback to initialize device 10547 * @txqs: the number of TX subqueues to allocate 10548 * @rxqs: the number of RX subqueues to allocate 10549 * 10550 * Allocates a struct net_device with private data area for driver use 10551 * and performs basic initialization. Also allocates subqueue structs 10552 * for each queue on the device. 10553 */ 10554 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, 10555 unsigned char name_assign_type, 10556 void (*setup)(struct net_device *), 10557 unsigned int txqs, unsigned int rxqs) 10558 { 10559 struct net_device *dev; 10560 unsigned int alloc_size; 10561 struct net_device *p; 10562 10563 BUG_ON(strlen(name) >= sizeof(dev->name)); 10564 10565 if (txqs < 1) { 10566 pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); 10567 return NULL; 10568 } 10569 10570 if (rxqs < 1) { 10571 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); 10572 return NULL; 10573 } 10574 10575 alloc_size = sizeof(struct net_device); 10576 if (sizeof_priv) { 10577 /* ensure 32-byte alignment of private area */ 10578 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); 10579 alloc_size += sizeof_priv; 10580 } 10581 /* ensure 32-byte alignment of whole construct */ 10582 alloc_size += NETDEV_ALIGN - 1; 10583 10584 p = kvzalloc(alloc_size, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10585 if (!p) 10586 return NULL; 10587 10588 dev = PTR_ALIGN(p, NETDEV_ALIGN); 10589 dev->padded = (char *)dev - (char *)p; 10590 10591 ref_tracker_dir_init(&dev->refcnt_tracker, 128); 10592 #ifdef CONFIG_PCPU_DEV_REFCNT 10593 dev->pcpu_refcnt = alloc_percpu(int); 10594 if (!dev->pcpu_refcnt) 10595 goto free_dev; 10596 __dev_hold(dev); 10597 #else 10598 refcount_set(&dev->dev_refcnt, 1); 10599 #endif 10600 10601 if (dev_addr_init(dev)) 10602 goto free_pcpu; 10603 10604 dev_mc_init(dev); 10605 dev_uc_init(dev); 10606 10607 dev_net_set(dev, &init_net); 10608 10609 dev->gso_max_size = GSO_LEGACY_MAX_SIZE; 10610 dev->gso_max_segs = GSO_MAX_SEGS; 10611 dev->gro_max_size = GRO_LEGACY_MAX_SIZE; 10612 dev->tso_max_size = TSO_LEGACY_MAX_SIZE; 10613 dev->tso_max_segs = TSO_MAX_SEGS; 10614 dev->upper_level = 1; 10615 dev->lower_level = 1; 10616 #ifdef CONFIG_LOCKDEP 10617 dev->nested_level = 0; 10618 INIT_LIST_HEAD(&dev->unlink_list); 10619 #endif 10620 10621 INIT_LIST_HEAD(&dev->napi_list); 10622 INIT_LIST_HEAD(&dev->unreg_list); 10623 INIT_LIST_HEAD(&dev->close_list); 10624 INIT_LIST_HEAD(&dev->link_watch_list); 10625 INIT_LIST_HEAD(&dev->adj_list.upper); 10626 INIT_LIST_HEAD(&dev->adj_list.lower); 10627 INIT_LIST_HEAD(&dev->ptype_all); 10628 INIT_LIST_HEAD(&dev->ptype_specific); 10629 INIT_LIST_HEAD(&dev->net_notifier_list); 10630 #ifdef CONFIG_NET_SCHED 10631 hash_init(dev->qdisc_hash); 10632 #endif 10633 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; 10634 setup(dev); 10635 10636 if (!dev->tx_queue_len) { 10637 dev->priv_flags |= IFF_NO_QUEUE; 10638 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; 10639 } 10640 10641 dev->num_tx_queues = txqs; 10642 dev->real_num_tx_queues = txqs; 10643 if (netif_alloc_netdev_queues(dev)) 10644 goto free_all; 10645 10646 dev->num_rx_queues = rxqs; 10647 dev->real_num_rx_queues = rxqs; 10648 if (netif_alloc_rx_queues(dev)) 10649 goto free_all; 10650 10651 strcpy(dev->name, name); 10652 dev->name_assign_type = name_assign_type; 10653 dev->group = INIT_NETDEV_GROUP; 10654 if (!dev->ethtool_ops) 10655 dev->ethtool_ops = &default_ethtool_ops; 10656 10657 nf_hook_netdev_init(dev); 10658 10659 return dev; 10660 10661 free_all: 10662 free_netdev(dev); 10663 return NULL; 10664 10665 free_pcpu: 10666 #ifdef CONFIG_PCPU_DEV_REFCNT 10667 free_percpu(dev->pcpu_refcnt); 10668 free_dev: 10669 #endif 10670 netdev_freemem(dev); 10671 return NULL; 10672 } 10673 EXPORT_SYMBOL(alloc_netdev_mqs); 10674 10675 /** 10676 * free_netdev - free network device 10677 * @dev: device 10678 * 10679 * This function does the last stage of destroying an allocated device 10680 * interface. The reference to the device object is released. If this 10681 * is the last reference then it will be freed.Must be called in process 10682 * context. 10683 */ 10684 void free_netdev(struct net_device *dev) 10685 { 10686 struct napi_struct *p, *n; 10687 10688 might_sleep(); 10689 10690 /* When called immediately after register_netdevice() failed the unwind 10691 * handling may still be dismantling the device. Handle that case by 10692 * deferring the free. 10693 */ 10694 if (dev->reg_state == NETREG_UNREGISTERING) { 10695 ASSERT_RTNL(); 10696 dev->needs_free_netdev = true; 10697 return; 10698 } 10699 10700 netif_free_tx_queues(dev); 10701 netif_free_rx_queues(dev); 10702 10703 kfree(rcu_dereference_protected(dev->ingress_queue, 1)); 10704 10705 /* Flush device addresses */ 10706 dev_addr_flush(dev); 10707 10708 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) 10709 netif_napi_del(p); 10710 10711 ref_tracker_dir_exit(&dev->refcnt_tracker); 10712 #ifdef CONFIG_PCPU_DEV_REFCNT 10713 free_percpu(dev->pcpu_refcnt); 10714 dev->pcpu_refcnt = NULL; 10715 #endif 10716 free_percpu(dev->core_stats); 10717 dev->core_stats = NULL; 10718 free_percpu(dev->xdp_bulkq); 10719 dev->xdp_bulkq = NULL; 10720 10721 /* Compatibility with error handling in drivers */ 10722 if (dev->reg_state == NETREG_UNINITIALIZED) { 10723 netdev_freemem(dev); 10724 return; 10725 } 10726 10727 BUG_ON(dev->reg_state != NETREG_UNREGISTERED); 10728 dev->reg_state = NETREG_RELEASED; 10729 10730 /* will free via device release */ 10731 put_device(&dev->dev); 10732 } 10733 EXPORT_SYMBOL(free_netdev); 10734 10735 /** 10736 * synchronize_net - Synchronize with packet receive processing 10737 * 10738 * Wait for packets currently being received to be done. 10739 * Does not block later packets from starting. 10740 */ 10741 void synchronize_net(void) 10742 { 10743 might_sleep(); 10744 if (rtnl_is_locked()) 10745 synchronize_rcu_expedited(); 10746 else 10747 synchronize_rcu(); 10748 } 10749 EXPORT_SYMBOL(synchronize_net); 10750 10751 /** 10752 * unregister_netdevice_queue - remove device from the kernel 10753 * @dev: device 10754 * @head: list 10755 * 10756 * This function shuts down a device interface and removes it 10757 * from the kernel tables. 10758 * If head not NULL, device is queued to be unregistered later. 10759 * 10760 * Callers must hold the rtnl semaphore. You may want 10761 * unregister_netdev() instead of this. 10762 */ 10763 10764 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) 10765 { 10766 ASSERT_RTNL(); 10767 10768 if (head) { 10769 list_move_tail(&dev->unreg_list, head); 10770 } else { 10771 LIST_HEAD(single); 10772 10773 list_add(&dev->unreg_list, &single); 10774 unregister_netdevice_many(&single); 10775 } 10776 } 10777 EXPORT_SYMBOL(unregister_netdevice_queue); 10778 10779 /** 10780 * unregister_netdevice_many - unregister many devices 10781 * @head: list of devices 10782 * 10783 * Note: As most callers use a stack allocated list_head, 10784 * we force a list_del() to make sure stack wont be corrupted later. 10785 */ 10786 void unregister_netdevice_many(struct list_head *head) 10787 { 10788 struct net_device *dev, *tmp; 10789 LIST_HEAD(close_head); 10790 10791 BUG_ON(dev_boot_phase); 10792 ASSERT_RTNL(); 10793 10794 if (list_empty(head)) 10795 return; 10796 10797 list_for_each_entry_safe(dev, tmp, head, unreg_list) { 10798 /* Some devices call without registering 10799 * for initialization unwind. Remove those 10800 * devices and proceed with the remaining. 10801 */ 10802 if (dev->reg_state == NETREG_UNINITIALIZED) { 10803 pr_debug("unregister_netdevice: device %s/%p never was registered\n", 10804 dev->name, dev); 10805 10806 WARN_ON(1); 10807 list_del(&dev->unreg_list); 10808 continue; 10809 } 10810 dev->dismantle = true; 10811 BUG_ON(dev->reg_state != NETREG_REGISTERED); 10812 } 10813 10814 /* If device is running, close it first. */ 10815 list_for_each_entry(dev, head, unreg_list) 10816 list_add_tail(&dev->close_list, &close_head); 10817 dev_close_many(&close_head, true); 10818 10819 list_for_each_entry(dev, head, unreg_list) { 10820 /* And unlink it from device chain. */ 10821 write_lock(&dev_base_lock); 10822 unlist_netdevice(dev, false); 10823 dev->reg_state = NETREG_UNREGISTERING; 10824 write_unlock(&dev_base_lock); 10825 } 10826 flush_all_backlogs(); 10827 10828 synchronize_net(); 10829 10830 list_for_each_entry(dev, head, unreg_list) { 10831 struct sk_buff *skb = NULL; 10832 10833 /* Shutdown queueing discipline. */ 10834 dev_shutdown(dev); 10835 10836 dev_xdp_uninstall(dev); 10837 10838 netdev_offload_xstats_disable_all(dev); 10839 10840 /* Notify protocols, that we are about to destroy 10841 * this device. They should clean all the things. 10842 */ 10843 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10844 10845 if (!dev->rtnl_link_ops || 10846 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 10847 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0, 10848 GFP_KERNEL, NULL, 0); 10849 10850 /* 10851 * Flush the unicast and multicast chains 10852 */ 10853 dev_uc_flush(dev); 10854 dev_mc_flush(dev); 10855 10856 netdev_name_node_alt_flush(dev); 10857 netdev_name_node_free(dev->name_node); 10858 10859 if (dev->netdev_ops->ndo_uninit) 10860 dev->netdev_ops->ndo_uninit(dev); 10861 10862 if (skb) 10863 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL); 10864 10865 /* Notifier chain MUST detach us all upper devices. */ 10866 WARN_ON(netdev_has_any_upper_dev(dev)); 10867 WARN_ON(netdev_has_any_lower_dev(dev)); 10868 10869 /* Remove entries from kobject tree */ 10870 netdev_unregister_kobject(dev); 10871 #ifdef CONFIG_XPS 10872 /* Remove XPS queueing entries */ 10873 netif_reset_xps_queues_gt(dev, 0); 10874 #endif 10875 } 10876 10877 synchronize_net(); 10878 10879 list_for_each_entry(dev, head, unreg_list) { 10880 netdev_put(dev, &dev->dev_registered_tracker); 10881 net_set_todo(dev); 10882 } 10883 10884 list_del(head); 10885 } 10886 EXPORT_SYMBOL(unregister_netdevice_many); 10887 10888 /** 10889 * unregister_netdev - remove device from the kernel 10890 * @dev: device 10891 * 10892 * This function shuts down a device interface and removes it 10893 * from the kernel tables. 10894 * 10895 * This is just a wrapper for unregister_netdevice that takes 10896 * the rtnl semaphore. In general you want to use this and not 10897 * unregister_netdevice. 10898 */ 10899 void unregister_netdev(struct net_device *dev) 10900 { 10901 rtnl_lock(); 10902 unregister_netdevice(dev); 10903 rtnl_unlock(); 10904 } 10905 EXPORT_SYMBOL(unregister_netdev); 10906 10907 /** 10908 * __dev_change_net_namespace - move device to different nethost namespace 10909 * @dev: device 10910 * @net: network namespace 10911 * @pat: If not NULL name pattern to try if the current device name 10912 * is already taken in the destination network namespace. 10913 * @new_ifindex: If not zero, specifies device index in the target 10914 * namespace. 10915 * 10916 * This function shuts down a device interface and moves it 10917 * to a new network namespace. On success 0 is returned, on 10918 * a failure a netagive errno code is returned. 10919 * 10920 * Callers must hold the rtnl semaphore. 10921 */ 10922 10923 int __dev_change_net_namespace(struct net_device *dev, struct net *net, 10924 const char *pat, int new_ifindex) 10925 { 10926 struct net *net_old = dev_net(dev); 10927 int err, new_nsid; 10928 10929 ASSERT_RTNL(); 10930 10931 /* Don't allow namespace local devices to be moved. */ 10932 err = -EINVAL; 10933 if (dev->features & NETIF_F_NETNS_LOCAL) 10934 goto out; 10935 10936 /* Ensure the device has been registrered */ 10937 if (dev->reg_state != NETREG_REGISTERED) 10938 goto out; 10939 10940 /* Get out if there is nothing todo */ 10941 err = 0; 10942 if (net_eq(net_old, net)) 10943 goto out; 10944 10945 /* Pick the destination device name, and ensure 10946 * we can use it in the destination network namespace. 10947 */ 10948 err = -EEXIST; 10949 if (netdev_name_in_use(net, dev->name)) { 10950 /* We get here if we can't use the current device name */ 10951 if (!pat) 10952 goto out; 10953 err = dev_get_valid_name(net, dev, pat); 10954 if (err < 0) 10955 goto out; 10956 } 10957 10958 /* Check that new_ifindex isn't used yet. */ 10959 err = -EBUSY; 10960 if (new_ifindex && __dev_get_by_index(net, new_ifindex)) 10961 goto out; 10962 10963 /* 10964 * And now a mini version of register_netdevice unregister_netdevice. 10965 */ 10966 10967 /* If device is running close it first. */ 10968 dev_close(dev); 10969 10970 /* And unlink it from device chain */ 10971 unlist_netdevice(dev, true); 10972 10973 synchronize_net(); 10974 10975 /* Shutdown queueing discipline. */ 10976 dev_shutdown(dev); 10977 10978 /* Notify protocols, that we are about to destroy 10979 * this device. They should clean all the things. 10980 * 10981 * Note that dev->reg_state stays at NETREG_REGISTERED. 10982 * This is wanted because this way 8021q and macvlan know 10983 * the device is just moving and can keep their slaves up. 10984 */ 10985 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10986 rcu_barrier(); 10987 10988 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL); 10989 /* If there is an ifindex conflict assign a new one */ 10990 if (!new_ifindex) { 10991 if (__dev_get_by_index(net, dev->ifindex)) 10992 new_ifindex = dev_new_index(net); 10993 else 10994 new_ifindex = dev->ifindex; 10995 } 10996 10997 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid, 10998 new_ifindex); 10999 11000 /* 11001 * Flush the unicast and multicast chains 11002 */ 11003 dev_uc_flush(dev); 11004 dev_mc_flush(dev); 11005 11006 /* Send a netdev-removed uevent to the old namespace */ 11007 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); 11008 netdev_adjacent_del_links(dev); 11009 11010 /* Move per-net netdevice notifiers that are following the netdevice */ 11011 move_netdevice_notifiers_dev_net(dev, net); 11012 11013 /* Actually switch the network namespace */ 11014 dev_net_set(dev, net); 11015 dev->ifindex = new_ifindex; 11016 11017 /* Send a netdev-add uevent to the new namespace */ 11018 kobject_uevent(&dev->dev.kobj, KOBJ_ADD); 11019 netdev_adjacent_add_links(dev); 11020 11021 /* Fixup kobjects */ 11022 err = device_rename(&dev->dev, dev->name); 11023 WARN_ON(err); 11024 11025 /* Adapt owner in case owning user namespace of target network 11026 * namespace is different from the original one. 11027 */ 11028 err = netdev_change_owner(dev, net_old, net); 11029 WARN_ON(err); 11030 11031 /* Add the device back in the hashes */ 11032 list_netdevice(dev); 11033 11034 /* Notify protocols, that a new device appeared. */ 11035 call_netdevice_notifiers(NETDEV_REGISTER, dev); 11036 11037 /* 11038 * Prevent userspace races by waiting until the network 11039 * device is fully setup before sending notifications. 11040 */ 11041 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); 11042 11043 synchronize_net(); 11044 err = 0; 11045 out: 11046 return err; 11047 } 11048 EXPORT_SYMBOL_GPL(__dev_change_net_namespace); 11049 11050 static int dev_cpu_dead(unsigned int oldcpu) 11051 { 11052 struct sk_buff **list_skb; 11053 struct sk_buff *skb; 11054 unsigned int cpu; 11055 struct softnet_data *sd, *oldsd, *remsd = NULL; 11056 11057 local_irq_disable(); 11058 cpu = smp_processor_id(); 11059 sd = &per_cpu(softnet_data, cpu); 11060 oldsd = &per_cpu(softnet_data, oldcpu); 11061 11062 /* Find end of our completion_queue. */ 11063 list_skb = &sd->completion_queue; 11064 while (*list_skb) 11065 list_skb = &(*list_skb)->next; 11066 /* Append completion queue from offline CPU. */ 11067 *list_skb = oldsd->completion_queue; 11068 oldsd->completion_queue = NULL; 11069 11070 /* Append output queue from offline CPU. */ 11071 if (oldsd->output_queue) { 11072 *sd->output_queue_tailp = oldsd->output_queue; 11073 sd->output_queue_tailp = oldsd->output_queue_tailp; 11074 oldsd->output_queue = NULL; 11075 oldsd->output_queue_tailp = &oldsd->output_queue; 11076 } 11077 /* Append NAPI poll list from offline CPU, with one exception : 11078 * process_backlog() must be called by cpu owning percpu backlog. 11079 * We properly handle process_queue & input_pkt_queue later. 11080 */ 11081 while (!list_empty(&oldsd->poll_list)) { 11082 struct napi_struct *napi = list_first_entry(&oldsd->poll_list, 11083 struct napi_struct, 11084 poll_list); 11085 11086 list_del_init(&napi->poll_list); 11087 if (napi->poll == process_backlog) 11088 napi->state = 0; 11089 else 11090 ____napi_schedule(sd, napi); 11091 } 11092 11093 raise_softirq_irqoff(NET_TX_SOFTIRQ); 11094 local_irq_enable(); 11095 11096 #ifdef CONFIG_RPS 11097 remsd = oldsd->rps_ipi_list; 11098 oldsd->rps_ipi_list = NULL; 11099 #endif 11100 /* send out pending IPI's on offline CPU */ 11101 net_rps_send_ipi(remsd); 11102 11103 /* Process offline CPU's input_pkt_queue */ 11104 while ((skb = __skb_dequeue(&oldsd->process_queue))) { 11105 netif_rx(skb); 11106 input_queue_head_incr(oldsd); 11107 } 11108 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { 11109 netif_rx(skb); 11110 input_queue_head_incr(oldsd); 11111 } 11112 11113 return 0; 11114 } 11115 11116 /** 11117 * netdev_increment_features - increment feature set by one 11118 * @all: current feature set 11119 * @one: new feature set 11120 * @mask: mask feature set 11121 * 11122 * Computes a new feature set after adding a device with feature set 11123 * @one to the master device with current feature set @all. Will not 11124 * enable anything that is off in @mask. Returns the new feature set. 11125 */ 11126 netdev_features_t netdev_increment_features(netdev_features_t all, 11127 netdev_features_t one, netdev_features_t mask) 11128 { 11129 if (mask & NETIF_F_HW_CSUM) 11130 mask |= NETIF_F_CSUM_MASK; 11131 mask |= NETIF_F_VLAN_CHALLENGED; 11132 11133 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; 11134 all &= one | ~NETIF_F_ALL_FOR_ALL; 11135 11136 /* If one device supports hw checksumming, set for all. */ 11137 if (all & NETIF_F_HW_CSUM) 11138 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); 11139 11140 return all; 11141 } 11142 EXPORT_SYMBOL(netdev_increment_features); 11143 11144 static struct hlist_head * __net_init netdev_create_hash(void) 11145 { 11146 int i; 11147 struct hlist_head *hash; 11148 11149 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL); 11150 if (hash != NULL) 11151 for (i = 0; i < NETDEV_HASHENTRIES; i++) 11152 INIT_HLIST_HEAD(&hash[i]); 11153 11154 return hash; 11155 } 11156 11157 /* Initialize per network namespace state */ 11158 static int __net_init netdev_init(struct net *net) 11159 { 11160 BUILD_BUG_ON(GRO_HASH_BUCKETS > 11161 8 * sizeof_field(struct napi_struct, gro_bitmask)); 11162 11163 INIT_LIST_HEAD(&net->dev_base_head); 11164 11165 net->dev_name_head = netdev_create_hash(); 11166 if (net->dev_name_head == NULL) 11167 goto err_name; 11168 11169 net->dev_index_head = netdev_create_hash(); 11170 if (net->dev_index_head == NULL) 11171 goto err_idx; 11172 11173 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain); 11174 11175 return 0; 11176 11177 err_idx: 11178 kfree(net->dev_name_head); 11179 err_name: 11180 return -ENOMEM; 11181 } 11182 11183 /** 11184 * netdev_drivername - network driver for the device 11185 * @dev: network device 11186 * 11187 * Determine network driver for device. 11188 */ 11189 const char *netdev_drivername(const struct net_device *dev) 11190 { 11191 const struct device_driver *driver; 11192 const struct device *parent; 11193 const char *empty = ""; 11194 11195 parent = dev->dev.parent; 11196 if (!parent) 11197 return empty; 11198 11199 driver = parent->driver; 11200 if (driver && driver->name) 11201 return driver->name; 11202 return empty; 11203 } 11204 11205 static void __netdev_printk(const char *level, const struct net_device *dev, 11206 struct va_format *vaf) 11207 { 11208 if (dev && dev->dev.parent) { 11209 dev_printk_emit(level[1] - '0', 11210 dev->dev.parent, 11211 "%s %s %s%s: %pV", 11212 dev_driver_string(dev->dev.parent), 11213 dev_name(dev->dev.parent), 11214 netdev_name(dev), netdev_reg_state(dev), 11215 vaf); 11216 } else if (dev) { 11217 printk("%s%s%s: %pV", 11218 level, netdev_name(dev), netdev_reg_state(dev), vaf); 11219 } else { 11220 printk("%s(NULL net_device): %pV", level, vaf); 11221 } 11222 } 11223 11224 void netdev_printk(const char *level, const struct net_device *dev, 11225 const char *format, ...) 11226 { 11227 struct va_format vaf; 11228 va_list args; 11229 11230 va_start(args, format); 11231 11232 vaf.fmt = format; 11233 vaf.va = &args; 11234 11235 __netdev_printk(level, dev, &vaf); 11236 11237 va_end(args); 11238 } 11239 EXPORT_SYMBOL(netdev_printk); 11240 11241 #define define_netdev_printk_level(func, level) \ 11242 void func(const struct net_device *dev, const char *fmt, ...) \ 11243 { \ 11244 struct va_format vaf; \ 11245 va_list args; \ 11246 \ 11247 va_start(args, fmt); \ 11248 \ 11249 vaf.fmt = fmt; \ 11250 vaf.va = &args; \ 11251 \ 11252 __netdev_printk(level, dev, &vaf); \ 11253 \ 11254 va_end(args); \ 11255 } \ 11256 EXPORT_SYMBOL(func); 11257 11258 define_netdev_printk_level(netdev_emerg, KERN_EMERG); 11259 define_netdev_printk_level(netdev_alert, KERN_ALERT); 11260 define_netdev_printk_level(netdev_crit, KERN_CRIT); 11261 define_netdev_printk_level(netdev_err, KERN_ERR); 11262 define_netdev_printk_level(netdev_warn, KERN_WARNING); 11263 define_netdev_printk_level(netdev_notice, KERN_NOTICE); 11264 define_netdev_printk_level(netdev_info, KERN_INFO); 11265 11266 static void __net_exit netdev_exit(struct net *net) 11267 { 11268 kfree(net->dev_name_head); 11269 kfree(net->dev_index_head); 11270 if (net != &init_net) 11271 WARN_ON_ONCE(!list_empty(&net->dev_base_head)); 11272 } 11273 11274 static struct pernet_operations __net_initdata netdev_net_ops = { 11275 .init = netdev_init, 11276 .exit = netdev_exit, 11277 }; 11278 11279 static void __net_exit default_device_exit_net(struct net *net) 11280 { 11281 struct net_device *dev, *aux; 11282 /* 11283 * Push all migratable network devices back to the 11284 * initial network namespace 11285 */ 11286 ASSERT_RTNL(); 11287 for_each_netdev_safe(net, dev, aux) { 11288 int err; 11289 char fb_name[IFNAMSIZ]; 11290 11291 /* Ignore unmoveable devices (i.e. loopback) */ 11292 if (dev->features & NETIF_F_NETNS_LOCAL) 11293 continue; 11294 11295 /* Leave virtual devices for the generic cleanup */ 11296 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund) 11297 continue; 11298 11299 /* Push remaining network devices to init_net */ 11300 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); 11301 if (netdev_name_in_use(&init_net, fb_name)) 11302 snprintf(fb_name, IFNAMSIZ, "dev%%d"); 11303 err = dev_change_net_namespace(dev, &init_net, fb_name); 11304 if (err) { 11305 pr_emerg("%s: failed to move %s to init_net: %d\n", 11306 __func__, dev->name, err); 11307 BUG(); 11308 } 11309 } 11310 } 11311 11312 static void __net_exit default_device_exit_batch(struct list_head *net_list) 11313 { 11314 /* At exit all network devices most be removed from a network 11315 * namespace. Do this in the reverse order of registration. 11316 * Do this across as many network namespaces as possible to 11317 * improve batching efficiency. 11318 */ 11319 struct net_device *dev; 11320 struct net *net; 11321 LIST_HEAD(dev_kill_list); 11322 11323 rtnl_lock(); 11324 list_for_each_entry(net, net_list, exit_list) { 11325 default_device_exit_net(net); 11326 cond_resched(); 11327 } 11328 11329 list_for_each_entry(net, net_list, exit_list) { 11330 for_each_netdev_reverse(net, dev) { 11331 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) 11332 dev->rtnl_link_ops->dellink(dev, &dev_kill_list); 11333 else 11334 unregister_netdevice_queue(dev, &dev_kill_list); 11335 } 11336 } 11337 unregister_netdevice_many(&dev_kill_list); 11338 rtnl_unlock(); 11339 } 11340 11341 static struct pernet_operations __net_initdata default_device_ops = { 11342 .exit_batch = default_device_exit_batch, 11343 }; 11344 11345 /* 11346 * Initialize the DEV module. At boot time this walks the device list and 11347 * unhooks any devices that fail to initialise (normally hardware not 11348 * present) and leaves us with a valid list of present and active devices. 11349 * 11350 */ 11351 11352 /* 11353 * This is called single threaded during boot, so no need 11354 * to take the rtnl semaphore. 11355 */ 11356 static int __init net_dev_init(void) 11357 { 11358 int i, rc = -ENOMEM; 11359 11360 BUG_ON(!dev_boot_phase); 11361 11362 if (dev_proc_init()) 11363 goto out; 11364 11365 if (netdev_kobject_init()) 11366 goto out; 11367 11368 INIT_LIST_HEAD(&ptype_all); 11369 for (i = 0; i < PTYPE_HASH_SIZE; i++) 11370 INIT_LIST_HEAD(&ptype_base[i]); 11371 11372 if (register_pernet_subsys(&netdev_net_ops)) 11373 goto out; 11374 11375 /* 11376 * Initialise the packet receive queues. 11377 */ 11378 11379 for_each_possible_cpu(i) { 11380 struct work_struct *flush = per_cpu_ptr(&flush_works, i); 11381 struct softnet_data *sd = &per_cpu(softnet_data, i); 11382 11383 INIT_WORK(flush, flush_backlog); 11384 11385 skb_queue_head_init(&sd->input_pkt_queue); 11386 skb_queue_head_init(&sd->process_queue); 11387 #ifdef CONFIG_XFRM_OFFLOAD 11388 skb_queue_head_init(&sd->xfrm_backlog); 11389 #endif 11390 INIT_LIST_HEAD(&sd->poll_list); 11391 sd->output_queue_tailp = &sd->output_queue; 11392 #ifdef CONFIG_RPS 11393 INIT_CSD(&sd->csd, rps_trigger_softirq, sd); 11394 sd->cpu = i; 11395 #endif 11396 INIT_CSD(&sd->defer_csd, trigger_rx_softirq, sd); 11397 spin_lock_init(&sd->defer_lock); 11398 11399 init_gro_hash(&sd->backlog); 11400 sd->backlog.poll = process_backlog; 11401 sd->backlog.weight = weight_p; 11402 } 11403 11404 dev_boot_phase = 0; 11405 11406 /* The loopback device is special if any other network devices 11407 * is present in a network namespace the loopback device must 11408 * be present. Since we now dynamically allocate and free the 11409 * loopback device ensure this invariant is maintained by 11410 * keeping the loopback device as the first device on the 11411 * list of network devices. Ensuring the loopback devices 11412 * is the first device that appears and the last network device 11413 * that disappears. 11414 */ 11415 if (register_pernet_device(&loopback_net_ops)) 11416 goto out; 11417 11418 if (register_pernet_device(&default_device_ops)) 11419 goto out; 11420 11421 open_softirq(NET_TX_SOFTIRQ, net_tx_action); 11422 open_softirq(NET_RX_SOFTIRQ, net_rx_action); 11423 11424 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", 11425 NULL, dev_cpu_dead); 11426 WARN_ON(rc < 0); 11427 rc = 0; 11428 out: 11429 return rc; 11430 } 11431 11432 subsys_initcall(net_dev_init); 11433