1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * NET3 Protocol independent device support routines. 4 * 5 * Derived from the non IP parts of dev.c 1.0.19 6 * Authors: Ross Biro 7 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 8 * Mark Evans, <evansmp@uhura.aston.ac.uk> 9 * 10 * Additional Authors: 11 * Florian la Roche <rzsfl@rz.uni-sb.de> 12 * Alan Cox <gw4pts@gw4pts.ampr.org> 13 * David Hinds <dahinds@users.sourceforge.net> 14 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru> 15 * Adam Sulmicki <adam@cfar.umd.edu> 16 * Pekka Riikonen <priikone@poesidon.pspt.fi> 17 * 18 * Changes: 19 * D.J. Barrow : Fixed bug where dev->refcnt gets set 20 * to 2 if register_netdev gets called 21 * before net_dev_init & also removed a 22 * few lines of code in the process. 23 * Alan Cox : device private ioctl copies fields back. 24 * Alan Cox : Transmit queue code does relevant 25 * stunts to keep the queue safe. 26 * Alan Cox : Fixed double lock. 27 * Alan Cox : Fixed promisc NULL pointer trap 28 * ???????? : Support the full private ioctl range 29 * Alan Cox : Moved ioctl permission check into 30 * drivers 31 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI 32 * Alan Cox : 100 backlog just doesn't cut it when 33 * you start doing multicast video 8) 34 * Alan Cox : Rewrote net_bh and list manager. 35 * Alan Cox : Fix ETH_P_ALL echoback lengths. 36 * Alan Cox : Took out transmit every packet pass 37 * Saved a few bytes in the ioctl handler 38 * Alan Cox : Network driver sets packet type before 39 * calling netif_rx. Saves a function 40 * call a packet. 41 * Alan Cox : Hashed net_bh() 42 * Richard Kooijman: Timestamp fixes. 43 * Alan Cox : Wrong field in SIOCGIFDSTADDR 44 * Alan Cox : Device lock protection. 45 * Alan Cox : Fixed nasty side effect of device close 46 * changes. 47 * Rudi Cilibrasi : Pass the right thing to 48 * set_mac_address() 49 * Dave Miller : 32bit quantity for the device lock to 50 * make it work out on a Sparc. 51 * Bjorn Ekwall : Added KERNELD hack. 52 * Alan Cox : Cleaned up the backlog initialise. 53 * Craig Metz : SIOCGIFCONF fix if space for under 54 * 1 device. 55 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there 56 * is no device open function. 57 * Andi Kleen : Fix error reporting for SIOCGIFCONF 58 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF 59 * Cyrus Durgin : Cleaned for KMOD 60 * Adam Sulmicki : Bug Fix : Network Device Unload 61 * A network device unload needs to purge 62 * the backlog queue. 63 * Paul Rusty Russell : SIOCSIFNAME 64 * Pekka Riikonen : Netdev boot-time settings code 65 * Andrew Morton : Make unregister_netdevice wait 66 * indefinitely on dev->refcnt 67 * J Hadi Salim : - Backlog queue sampling 68 * - netif_rx() feedback 69 */ 70 71 #include <linux/uaccess.h> 72 #include <linux/bitops.h> 73 #include <linux/capability.h> 74 #include <linux/cpu.h> 75 #include <linux/types.h> 76 #include <linux/kernel.h> 77 #include <linux/hash.h> 78 #include <linux/slab.h> 79 #include <linux/sched.h> 80 #include <linux/sched/mm.h> 81 #include <linux/mutex.h> 82 #include <linux/rwsem.h> 83 #include <linux/string.h> 84 #include <linux/mm.h> 85 #include <linux/socket.h> 86 #include <linux/sockios.h> 87 #include <linux/errno.h> 88 #include <linux/interrupt.h> 89 #include <linux/if_ether.h> 90 #include <linux/netdevice.h> 91 #include <linux/etherdevice.h> 92 #include <linux/ethtool.h> 93 #include <linux/skbuff.h> 94 #include <linux/kthread.h> 95 #include <linux/bpf.h> 96 #include <linux/bpf_trace.h> 97 #include <net/net_namespace.h> 98 #include <net/sock.h> 99 #include <net/busy_poll.h> 100 #include <linux/rtnetlink.h> 101 #include <linux/stat.h> 102 #include <net/dsa.h> 103 #include <net/dst.h> 104 #include <net/dst_metadata.h> 105 #include <net/gro.h> 106 #include <net/pkt_sched.h> 107 #include <net/pkt_cls.h> 108 #include <net/checksum.h> 109 #include <net/xfrm.h> 110 #include <linux/highmem.h> 111 #include <linux/init.h> 112 #include <linux/module.h> 113 #include <linux/netpoll.h> 114 #include <linux/rcupdate.h> 115 #include <linux/delay.h> 116 #include <net/iw_handler.h> 117 #include <asm/current.h> 118 #include <linux/audit.h> 119 #include <linux/dmaengine.h> 120 #include <linux/err.h> 121 #include <linux/ctype.h> 122 #include <linux/if_arp.h> 123 #include <linux/if_vlan.h> 124 #include <linux/ip.h> 125 #include <net/ip.h> 126 #include <net/mpls.h> 127 #include <linux/ipv6.h> 128 #include <linux/in.h> 129 #include <linux/jhash.h> 130 #include <linux/random.h> 131 #include <trace/events/napi.h> 132 #include <trace/events/net.h> 133 #include <trace/events/skb.h> 134 #include <trace/events/qdisc.h> 135 #include <linux/inetdevice.h> 136 #include <linux/cpu_rmap.h> 137 #include <linux/static_key.h> 138 #include <linux/hashtable.h> 139 #include <linux/vmalloc.h> 140 #include <linux/if_macvlan.h> 141 #include <linux/errqueue.h> 142 #include <linux/hrtimer.h> 143 #include <linux/netfilter_netdev.h> 144 #include <linux/crash_dump.h> 145 #include <linux/sctp.h> 146 #include <net/udp_tunnel.h> 147 #include <linux/net_namespace.h> 148 #include <linux/indirect_call_wrapper.h> 149 #include <net/devlink.h> 150 #include <linux/pm_runtime.h> 151 #include <linux/prandom.h> 152 #include <linux/once_lite.h> 153 154 #include "net-sysfs.h" 155 156 157 static DEFINE_SPINLOCK(ptype_lock); 158 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; 159 struct list_head ptype_all __read_mostly; /* Taps */ 160 161 static int netif_rx_internal(struct sk_buff *skb); 162 static int call_netdevice_notifiers_info(unsigned long val, 163 struct netdev_notifier_info *info); 164 static int call_netdevice_notifiers_extack(unsigned long val, 165 struct net_device *dev, 166 struct netlink_ext_ack *extack); 167 static struct napi_struct *napi_by_id(unsigned int napi_id); 168 169 /* 170 * The @dev_base_head list is protected by @dev_base_lock and the rtnl 171 * semaphore. 172 * 173 * Pure readers hold dev_base_lock for reading, or rcu_read_lock() 174 * 175 * Writers must hold the rtnl semaphore while they loop through the 176 * dev_base_head list, and hold dev_base_lock for writing when they do the 177 * actual updates. This allows pure readers to access the list even 178 * while a writer is preparing to update it. 179 * 180 * To put it another way, dev_base_lock is held for writing only to 181 * protect against pure readers; the rtnl semaphore provides the 182 * protection against other writers. 183 * 184 * See, for example usages, register_netdevice() and 185 * unregister_netdevice(), which must be called with the rtnl 186 * semaphore held. 187 */ 188 DEFINE_RWLOCK(dev_base_lock); 189 EXPORT_SYMBOL(dev_base_lock); 190 191 static DEFINE_MUTEX(ifalias_mutex); 192 193 /* protects napi_hash addition/deletion and napi_gen_id */ 194 static DEFINE_SPINLOCK(napi_hash_lock); 195 196 static unsigned int napi_gen_id = NR_CPUS; 197 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8); 198 199 static DECLARE_RWSEM(devnet_rename_sem); 200 201 static inline void dev_base_seq_inc(struct net *net) 202 { 203 while (++net->dev_base_seq == 0) 204 ; 205 } 206 207 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) 208 { 209 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ)); 210 211 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; 212 } 213 214 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) 215 { 216 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; 217 } 218 219 static inline void rps_lock_irqsave(struct softnet_data *sd, 220 unsigned long *flags) 221 { 222 if (IS_ENABLED(CONFIG_RPS)) 223 spin_lock_irqsave(&sd->input_pkt_queue.lock, *flags); 224 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 225 local_irq_save(*flags); 226 } 227 228 static inline void rps_lock_irq_disable(struct softnet_data *sd) 229 { 230 if (IS_ENABLED(CONFIG_RPS)) 231 spin_lock_irq(&sd->input_pkt_queue.lock); 232 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 233 local_irq_disable(); 234 } 235 236 static inline void rps_unlock_irq_restore(struct softnet_data *sd, 237 unsigned long *flags) 238 { 239 if (IS_ENABLED(CONFIG_RPS)) 240 spin_unlock_irqrestore(&sd->input_pkt_queue.lock, *flags); 241 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 242 local_irq_restore(*flags); 243 } 244 245 static inline void rps_unlock_irq_enable(struct softnet_data *sd) 246 { 247 if (IS_ENABLED(CONFIG_RPS)) 248 spin_unlock_irq(&sd->input_pkt_queue.lock); 249 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 250 local_irq_enable(); 251 } 252 253 static struct netdev_name_node *netdev_name_node_alloc(struct net_device *dev, 254 const char *name) 255 { 256 struct netdev_name_node *name_node; 257 258 name_node = kmalloc(sizeof(*name_node), GFP_KERNEL); 259 if (!name_node) 260 return NULL; 261 INIT_HLIST_NODE(&name_node->hlist); 262 name_node->dev = dev; 263 name_node->name = name; 264 return name_node; 265 } 266 267 static struct netdev_name_node * 268 netdev_name_node_head_alloc(struct net_device *dev) 269 { 270 struct netdev_name_node *name_node; 271 272 name_node = netdev_name_node_alloc(dev, dev->name); 273 if (!name_node) 274 return NULL; 275 INIT_LIST_HEAD(&name_node->list); 276 return name_node; 277 } 278 279 static void netdev_name_node_free(struct netdev_name_node *name_node) 280 { 281 kfree(name_node); 282 } 283 284 static void netdev_name_node_add(struct net *net, 285 struct netdev_name_node *name_node) 286 { 287 hlist_add_head_rcu(&name_node->hlist, 288 dev_name_hash(net, name_node->name)); 289 } 290 291 static void netdev_name_node_del(struct netdev_name_node *name_node) 292 { 293 hlist_del_rcu(&name_node->hlist); 294 } 295 296 static struct netdev_name_node *netdev_name_node_lookup(struct net *net, 297 const char *name) 298 { 299 struct hlist_head *head = dev_name_hash(net, name); 300 struct netdev_name_node *name_node; 301 302 hlist_for_each_entry(name_node, head, hlist) 303 if (!strcmp(name_node->name, name)) 304 return name_node; 305 return NULL; 306 } 307 308 static struct netdev_name_node *netdev_name_node_lookup_rcu(struct net *net, 309 const char *name) 310 { 311 struct hlist_head *head = dev_name_hash(net, name); 312 struct netdev_name_node *name_node; 313 314 hlist_for_each_entry_rcu(name_node, head, hlist) 315 if (!strcmp(name_node->name, name)) 316 return name_node; 317 return NULL; 318 } 319 320 bool netdev_name_in_use(struct net *net, const char *name) 321 { 322 return netdev_name_node_lookup(net, name); 323 } 324 EXPORT_SYMBOL(netdev_name_in_use); 325 326 int netdev_name_node_alt_create(struct net_device *dev, const char *name) 327 { 328 struct netdev_name_node *name_node; 329 struct net *net = dev_net(dev); 330 331 name_node = netdev_name_node_lookup(net, name); 332 if (name_node) 333 return -EEXIST; 334 name_node = netdev_name_node_alloc(dev, name); 335 if (!name_node) 336 return -ENOMEM; 337 netdev_name_node_add(net, name_node); 338 /* The node that holds dev->name acts as a head of per-device list. */ 339 list_add_tail(&name_node->list, &dev->name_node->list); 340 341 return 0; 342 } 343 344 static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node) 345 { 346 list_del(&name_node->list); 347 netdev_name_node_del(name_node); 348 kfree(name_node->name); 349 netdev_name_node_free(name_node); 350 } 351 352 int netdev_name_node_alt_destroy(struct net_device *dev, const char *name) 353 { 354 struct netdev_name_node *name_node; 355 struct net *net = dev_net(dev); 356 357 name_node = netdev_name_node_lookup(net, name); 358 if (!name_node) 359 return -ENOENT; 360 /* lookup might have found our primary name or a name belonging 361 * to another device. 362 */ 363 if (name_node == dev->name_node || name_node->dev != dev) 364 return -EINVAL; 365 366 __netdev_name_node_alt_destroy(name_node); 367 368 return 0; 369 } 370 371 static void netdev_name_node_alt_flush(struct net_device *dev) 372 { 373 struct netdev_name_node *name_node, *tmp; 374 375 list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list) 376 __netdev_name_node_alt_destroy(name_node); 377 } 378 379 /* Device list insertion */ 380 static void list_netdevice(struct net_device *dev) 381 { 382 struct net *net = dev_net(dev); 383 384 ASSERT_RTNL(); 385 386 write_lock(&dev_base_lock); 387 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); 388 netdev_name_node_add(net, dev->name_node); 389 hlist_add_head_rcu(&dev->index_hlist, 390 dev_index_hash(net, dev->ifindex)); 391 write_unlock(&dev_base_lock); 392 393 dev_base_seq_inc(net); 394 } 395 396 /* Device list removal 397 * caller must respect a RCU grace period before freeing/reusing dev 398 */ 399 static void unlist_netdevice(struct net_device *dev) 400 { 401 ASSERT_RTNL(); 402 403 /* Unlink dev from the device chain */ 404 write_lock(&dev_base_lock); 405 list_del_rcu(&dev->dev_list); 406 netdev_name_node_del(dev->name_node); 407 hlist_del_rcu(&dev->index_hlist); 408 write_unlock(&dev_base_lock); 409 410 dev_base_seq_inc(dev_net(dev)); 411 } 412 413 /* 414 * Our notifier list 415 */ 416 417 static RAW_NOTIFIER_HEAD(netdev_chain); 418 419 /* 420 * Device drivers call our routines to queue packets here. We empty the 421 * queue in the local softnet handler. 422 */ 423 424 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); 425 EXPORT_PER_CPU_SYMBOL(softnet_data); 426 427 #ifdef CONFIG_LOCKDEP 428 /* 429 * register_netdevice() inits txq->_xmit_lock and sets lockdep class 430 * according to dev->type 431 */ 432 static const unsigned short netdev_lock_type[] = { 433 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, 434 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, 435 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, 436 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, 437 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, 438 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, 439 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, 440 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, 441 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, 442 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, 443 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, 444 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, 445 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, 446 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, 447 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; 448 449 static const char *const netdev_lock_name[] = { 450 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", 451 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", 452 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", 453 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", 454 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", 455 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", 456 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", 457 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", 458 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", 459 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", 460 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", 461 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", 462 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", 463 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", 464 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; 465 466 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; 467 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; 468 469 static inline unsigned short netdev_lock_pos(unsigned short dev_type) 470 { 471 int i; 472 473 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) 474 if (netdev_lock_type[i] == dev_type) 475 return i; 476 /* the last key is used by default */ 477 return ARRAY_SIZE(netdev_lock_type) - 1; 478 } 479 480 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 481 unsigned short dev_type) 482 { 483 int i; 484 485 i = netdev_lock_pos(dev_type); 486 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], 487 netdev_lock_name[i]); 488 } 489 490 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 491 { 492 int i; 493 494 i = netdev_lock_pos(dev->type); 495 lockdep_set_class_and_name(&dev->addr_list_lock, 496 &netdev_addr_lock_key[i], 497 netdev_lock_name[i]); 498 } 499 #else 500 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 501 unsigned short dev_type) 502 { 503 } 504 505 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 506 { 507 } 508 #endif 509 510 /******************************************************************************* 511 * 512 * Protocol management and registration routines 513 * 514 *******************************************************************************/ 515 516 517 /* 518 * Add a protocol ID to the list. Now that the input handler is 519 * smarter we can dispense with all the messy stuff that used to be 520 * here. 521 * 522 * BEWARE!!! Protocol handlers, mangling input packets, 523 * MUST BE last in hash buckets and checking protocol handlers 524 * MUST start from promiscuous ptype_all chain in net_bh. 525 * It is true now, do not change it. 526 * Explanation follows: if protocol handler, mangling packet, will 527 * be the first on list, it is not able to sense, that packet 528 * is cloned and should be copied-on-write, so that it will 529 * change it and subsequent readers will get broken packet. 530 * --ANK (980803) 531 */ 532 533 static inline struct list_head *ptype_head(const struct packet_type *pt) 534 { 535 if (pt->type == htons(ETH_P_ALL)) 536 return pt->dev ? &pt->dev->ptype_all : &ptype_all; 537 else 538 return pt->dev ? &pt->dev->ptype_specific : 539 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; 540 } 541 542 /** 543 * dev_add_pack - add packet handler 544 * @pt: packet type declaration 545 * 546 * Add a protocol handler to the networking stack. The passed &packet_type 547 * is linked into kernel lists and may not be freed until it has been 548 * removed from the kernel lists. 549 * 550 * This call does not sleep therefore it can not 551 * guarantee all CPU's that are in middle of receiving packets 552 * will see the new packet type (until the next received packet). 553 */ 554 555 void dev_add_pack(struct packet_type *pt) 556 { 557 struct list_head *head = ptype_head(pt); 558 559 spin_lock(&ptype_lock); 560 list_add_rcu(&pt->list, head); 561 spin_unlock(&ptype_lock); 562 } 563 EXPORT_SYMBOL(dev_add_pack); 564 565 /** 566 * __dev_remove_pack - remove packet handler 567 * @pt: packet type declaration 568 * 569 * Remove a protocol handler that was previously added to the kernel 570 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 571 * from the kernel lists and can be freed or reused once this function 572 * returns. 573 * 574 * The packet type might still be in use by receivers 575 * and must not be freed until after all the CPU's have gone 576 * through a quiescent state. 577 */ 578 void __dev_remove_pack(struct packet_type *pt) 579 { 580 struct list_head *head = ptype_head(pt); 581 struct packet_type *pt1; 582 583 spin_lock(&ptype_lock); 584 585 list_for_each_entry(pt1, head, list) { 586 if (pt == pt1) { 587 list_del_rcu(&pt->list); 588 goto out; 589 } 590 } 591 592 pr_warn("dev_remove_pack: %p not found\n", pt); 593 out: 594 spin_unlock(&ptype_lock); 595 } 596 EXPORT_SYMBOL(__dev_remove_pack); 597 598 /** 599 * dev_remove_pack - remove packet handler 600 * @pt: packet type declaration 601 * 602 * Remove a protocol handler that was previously added to the kernel 603 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 604 * from the kernel lists and can be freed or reused once this function 605 * returns. 606 * 607 * This call sleeps to guarantee that no CPU is looking at the packet 608 * type after return. 609 */ 610 void dev_remove_pack(struct packet_type *pt) 611 { 612 __dev_remove_pack(pt); 613 614 synchronize_net(); 615 } 616 EXPORT_SYMBOL(dev_remove_pack); 617 618 619 /******************************************************************************* 620 * 621 * Device Interface Subroutines 622 * 623 *******************************************************************************/ 624 625 /** 626 * dev_get_iflink - get 'iflink' value of a interface 627 * @dev: targeted interface 628 * 629 * Indicates the ifindex the interface is linked to. 630 * Physical interfaces have the same 'ifindex' and 'iflink' values. 631 */ 632 633 int dev_get_iflink(const struct net_device *dev) 634 { 635 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink) 636 return dev->netdev_ops->ndo_get_iflink(dev); 637 638 return dev->ifindex; 639 } 640 EXPORT_SYMBOL(dev_get_iflink); 641 642 /** 643 * dev_fill_metadata_dst - Retrieve tunnel egress information. 644 * @dev: targeted interface 645 * @skb: The packet. 646 * 647 * For better visibility of tunnel traffic OVS needs to retrieve 648 * egress tunnel information for a packet. Following API allows 649 * user to get this info. 650 */ 651 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) 652 { 653 struct ip_tunnel_info *info; 654 655 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst) 656 return -EINVAL; 657 658 info = skb_tunnel_info_unclone(skb); 659 if (!info) 660 return -ENOMEM; 661 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX))) 662 return -EINVAL; 663 664 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb); 665 } 666 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst); 667 668 static struct net_device_path *dev_fwd_path(struct net_device_path_stack *stack) 669 { 670 int k = stack->num_paths++; 671 672 if (WARN_ON_ONCE(k >= NET_DEVICE_PATH_STACK_MAX)) 673 return NULL; 674 675 return &stack->path[k]; 676 } 677 678 int dev_fill_forward_path(const struct net_device *dev, const u8 *daddr, 679 struct net_device_path_stack *stack) 680 { 681 const struct net_device *last_dev; 682 struct net_device_path_ctx ctx = { 683 .dev = dev, 684 .daddr = daddr, 685 }; 686 struct net_device_path *path; 687 int ret = 0; 688 689 stack->num_paths = 0; 690 while (ctx.dev && ctx.dev->netdev_ops->ndo_fill_forward_path) { 691 last_dev = ctx.dev; 692 path = dev_fwd_path(stack); 693 if (!path) 694 return -1; 695 696 memset(path, 0, sizeof(struct net_device_path)); 697 ret = ctx.dev->netdev_ops->ndo_fill_forward_path(&ctx, path); 698 if (ret < 0) 699 return -1; 700 701 if (WARN_ON_ONCE(last_dev == ctx.dev)) 702 return -1; 703 } 704 path = dev_fwd_path(stack); 705 if (!path) 706 return -1; 707 path->type = DEV_PATH_ETHERNET; 708 path->dev = ctx.dev; 709 710 return ret; 711 } 712 EXPORT_SYMBOL_GPL(dev_fill_forward_path); 713 714 /** 715 * __dev_get_by_name - find a device by its name 716 * @net: the applicable net namespace 717 * @name: name to find 718 * 719 * Find an interface by name. Must be called under RTNL semaphore 720 * or @dev_base_lock. If the name is found a pointer to the device 721 * is returned. If the name is not found then %NULL is returned. The 722 * reference counters are not incremented so the caller must be 723 * careful with locks. 724 */ 725 726 struct net_device *__dev_get_by_name(struct net *net, const char *name) 727 { 728 struct netdev_name_node *node_name; 729 730 node_name = netdev_name_node_lookup(net, name); 731 return node_name ? node_name->dev : NULL; 732 } 733 EXPORT_SYMBOL(__dev_get_by_name); 734 735 /** 736 * dev_get_by_name_rcu - find a device by its name 737 * @net: the applicable net namespace 738 * @name: name to find 739 * 740 * Find an interface by name. 741 * If the name is found a pointer to the device is returned. 742 * If the name is not found then %NULL is returned. 743 * The reference counters are not incremented so the caller must be 744 * careful with locks. The caller must hold RCU lock. 745 */ 746 747 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) 748 { 749 struct netdev_name_node *node_name; 750 751 node_name = netdev_name_node_lookup_rcu(net, name); 752 return node_name ? node_name->dev : NULL; 753 } 754 EXPORT_SYMBOL(dev_get_by_name_rcu); 755 756 /** 757 * dev_get_by_name - find a device by its name 758 * @net: the applicable net namespace 759 * @name: name to find 760 * 761 * Find an interface by name. This can be called from any 762 * context and does its own locking. The returned handle has 763 * the usage count incremented and the caller must use dev_put() to 764 * release it when it is no longer needed. %NULL is returned if no 765 * matching device is found. 766 */ 767 768 struct net_device *dev_get_by_name(struct net *net, const char *name) 769 { 770 struct net_device *dev; 771 772 rcu_read_lock(); 773 dev = dev_get_by_name_rcu(net, name); 774 dev_hold(dev); 775 rcu_read_unlock(); 776 return dev; 777 } 778 EXPORT_SYMBOL(dev_get_by_name); 779 780 /** 781 * __dev_get_by_index - find a device by its ifindex 782 * @net: the applicable net namespace 783 * @ifindex: index of device 784 * 785 * Search for an interface by index. Returns %NULL if the device 786 * is not found or a pointer to the device. The device has not 787 * had its reference counter increased so the caller must be careful 788 * about locking. The caller must hold either the RTNL semaphore 789 * or @dev_base_lock. 790 */ 791 792 struct net_device *__dev_get_by_index(struct net *net, int ifindex) 793 { 794 struct net_device *dev; 795 struct hlist_head *head = dev_index_hash(net, ifindex); 796 797 hlist_for_each_entry(dev, head, index_hlist) 798 if (dev->ifindex == ifindex) 799 return dev; 800 801 return NULL; 802 } 803 EXPORT_SYMBOL(__dev_get_by_index); 804 805 /** 806 * dev_get_by_index_rcu - find a device by its ifindex 807 * @net: the applicable net namespace 808 * @ifindex: index of device 809 * 810 * Search for an interface by index. Returns %NULL if the device 811 * is not found or a pointer to the device. The device has not 812 * had its reference counter increased so the caller must be careful 813 * about locking. The caller must hold RCU lock. 814 */ 815 816 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) 817 { 818 struct net_device *dev; 819 struct hlist_head *head = dev_index_hash(net, ifindex); 820 821 hlist_for_each_entry_rcu(dev, head, index_hlist) 822 if (dev->ifindex == ifindex) 823 return dev; 824 825 return NULL; 826 } 827 EXPORT_SYMBOL(dev_get_by_index_rcu); 828 829 830 /** 831 * dev_get_by_index - find a device by its ifindex 832 * @net: the applicable net namespace 833 * @ifindex: index of device 834 * 835 * Search for an interface by index. Returns NULL if the device 836 * is not found or a pointer to the device. The device returned has 837 * had a reference added and the pointer is safe until the user calls 838 * dev_put to indicate they have finished with it. 839 */ 840 841 struct net_device *dev_get_by_index(struct net *net, int ifindex) 842 { 843 struct net_device *dev; 844 845 rcu_read_lock(); 846 dev = dev_get_by_index_rcu(net, ifindex); 847 dev_hold(dev); 848 rcu_read_unlock(); 849 return dev; 850 } 851 EXPORT_SYMBOL(dev_get_by_index); 852 853 /** 854 * dev_get_by_napi_id - find a device by napi_id 855 * @napi_id: ID of the NAPI struct 856 * 857 * Search for an interface by NAPI ID. Returns %NULL if the device 858 * is not found or a pointer to the device. The device has not had 859 * its reference counter increased so the caller must be careful 860 * about locking. The caller must hold RCU lock. 861 */ 862 863 struct net_device *dev_get_by_napi_id(unsigned int napi_id) 864 { 865 struct napi_struct *napi; 866 867 WARN_ON_ONCE(!rcu_read_lock_held()); 868 869 if (napi_id < MIN_NAPI_ID) 870 return NULL; 871 872 napi = napi_by_id(napi_id); 873 874 return napi ? napi->dev : NULL; 875 } 876 EXPORT_SYMBOL(dev_get_by_napi_id); 877 878 /** 879 * netdev_get_name - get a netdevice name, knowing its ifindex. 880 * @net: network namespace 881 * @name: a pointer to the buffer where the name will be stored. 882 * @ifindex: the ifindex of the interface to get the name from. 883 */ 884 int netdev_get_name(struct net *net, char *name, int ifindex) 885 { 886 struct net_device *dev; 887 int ret; 888 889 down_read(&devnet_rename_sem); 890 rcu_read_lock(); 891 892 dev = dev_get_by_index_rcu(net, ifindex); 893 if (!dev) { 894 ret = -ENODEV; 895 goto out; 896 } 897 898 strcpy(name, dev->name); 899 900 ret = 0; 901 out: 902 rcu_read_unlock(); 903 up_read(&devnet_rename_sem); 904 return ret; 905 } 906 907 /** 908 * dev_getbyhwaddr_rcu - find a device by its hardware address 909 * @net: the applicable net namespace 910 * @type: media type of device 911 * @ha: hardware address 912 * 913 * Search for an interface by MAC address. Returns NULL if the device 914 * is not found or a pointer to the device. 915 * The caller must hold RCU or RTNL. 916 * The returned device has not had its ref count increased 917 * and the caller must therefore be careful about locking 918 * 919 */ 920 921 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, 922 const char *ha) 923 { 924 struct net_device *dev; 925 926 for_each_netdev_rcu(net, dev) 927 if (dev->type == type && 928 !memcmp(dev->dev_addr, ha, dev->addr_len)) 929 return dev; 930 931 return NULL; 932 } 933 EXPORT_SYMBOL(dev_getbyhwaddr_rcu); 934 935 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) 936 { 937 struct net_device *dev, *ret = NULL; 938 939 rcu_read_lock(); 940 for_each_netdev_rcu(net, dev) 941 if (dev->type == type) { 942 dev_hold(dev); 943 ret = dev; 944 break; 945 } 946 rcu_read_unlock(); 947 return ret; 948 } 949 EXPORT_SYMBOL(dev_getfirstbyhwtype); 950 951 /** 952 * __dev_get_by_flags - find any device with given flags 953 * @net: the applicable net namespace 954 * @if_flags: IFF_* values 955 * @mask: bitmask of bits in if_flags to check 956 * 957 * Search for any interface with the given flags. Returns NULL if a device 958 * is not found or a pointer to the device. Must be called inside 959 * rtnl_lock(), and result refcount is unchanged. 960 */ 961 962 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags, 963 unsigned short mask) 964 { 965 struct net_device *dev, *ret; 966 967 ASSERT_RTNL(); 968 969 ret = NULL; 970 for_each_netdev(net, dev) { 971 if (((dev->flags ^ if_flags) & mask) == 0) { 972 ret = dev; 973 break; 974 } 975 } 976 return ret; 977 } 978 EXPORT_SYMBOL(__dev_get_by_flags); 979 980 /** 981 * dev_valid_name - check if name is okay for network device 982 * @name: name string 983 * 984 * Network device names need to be valid file names to 985 * allow sysfs to work. We also disallow any kind of 986 * whitespace. 987 */ 988 bool dev_valid_name(const char *name) 989 { 990 if (*name == '\0') 991 return false; 992 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ) 993 return false; 994 if (!strcmp(name, ".") || !strcmp(name, "..")) 995 return false; 996 997 while (*name) { 998 if (*name == '/' || *name == ':' || isspace(*name)) 999 return false; 1000 name++; 1001 } 1002 return true; 1003 } 1004 EXPORT_SYMBOL(dev_valid_name); 1005 1006 /** 1007 * __dev_alloc_name - allocate a name for a device 1008 * @net: network namespace to allocate the device name in 1009 * @name: name format string 1010 * @buf: scratch buffer and result name string 1011 * 1012 * Passed a format string - eg "lt%d" it will try and find a suitable 1013 * id. It scans list of devices to build up a free map, then chooses 1014 * the first empty slot. The caller must hold the dev_base or rtnl lock 1015 * while allocating the name and adding the device in order to avoid 1016 * duplicates. 1017 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1018 * Returns the number of the unit assigned or a negative errno code. 1019 */ 1020 1021 static int __dev_alloc_name(struct net *net, const char *name, char *buf) 1022 { 1023 int i = 0; 1024 const char *p; 1025 const int max_netdevices = 8*PAGE_SIZE; 1026 unsigned long *inuse; 1027 struct net_device *d; 1028 1029 if (!dev_valid_name(name)) 1030 return -EINVAL; 1031 1032 p = strchr(name, '%'); 1033 if (p) { 1034 /* 1035 * Verify the string as this thing may have come from 1036 * the user. There must be either one "%d" and no other "%" 1037 * characters. 1038 */ 1039 if (p[1] != 'd' || strchr(p + 2, '%')) 1040 return -EINVAL; 1041 1042 /* Use one page as a bit array of possible slots */ 1043 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); 1044 if (!inuse) 1045 return -ENOMEM; 1046 1047 for_each_netdev(net, d) { 1048 struct netdev_name_node *name_node; 1049 list_for_each_entry(name_node, &d->name_node->list, list) { 1050 if (!sscanf(name_node->name, name, &i)) 1051 continue; 1052 if (i < 0 || i >= max_netdevices) 1053 continue; 1054 1055 /* avoid cases where sscanf is not exact inverse of printf */ 1056 snprintf(buf, IFNAMSIZ, name, i); 1057 if (!strncmp(buf, name_node->name, IFNAMSIZ)) 1058 __set_bit(i, inuse); 1059 } 1060 if (!sscanf(d->name, name, &i)) 1061 continue; 1062 if (i < 0 || i >= max_netdevices) 1063 continue; 1064 1065 /* avoid cases where sscanf is not exact inverse of printf */ 1066 snprintf(buf, IFNAMSIZ, name, i); 1067 if (!strncmp(buf, d->name, IFNAMSIZ)) 1068 __set_bit(i, inuse); 1069 } 1070 1071 i = find_first_zero_bit(inuse, max_netdevices); 1072 free_page((unsigned long) inuse); 1073 } 1074 1075 snprintf(buf, IFNAMSIZ, name, i); 1076 if (!netdev_name_in_use(net, buf)) 1077 return i; 1078 1079 /* It is possible to run out of possible slots 1080 * when the name is long and there isn't enough space left 1081 * for the digits, or if all bits are used. 1082 */ 1083 return -ENFILE; 1084 } 1085 1086 static int dev_alloc_name_ns(struct net *net, 1087 struct net_device *dev, 1088 const char *name) 1089 { 1090 char buf[IFNAMSIZ]; 1091 int ret; 1092 1093 BUG_ON(!net); 1094 ret = __dev_alloc_name(net, name, buf); 1095 if (ret >= 0) 1096 strlcpy(dev->name, buf, IFNAMSIZ); 1097 return ret; 1098 } 1099 1100 /** 1101 * dev_alloc_name - allocate a name for a device 1102 * @dev: device 1103 * @name: name format string 1104 * 1105 * Passed a format string - eg "lt%d" it will try and find a suitable 1106 * id. It scans list of devices to build up a free map, then chooses 1107 * the first empty slot. The caller must hold the dev_base or rtnl lock 1108 * while allocating the name and adding the device in order to avoid 1109 * duplicates. 1110 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1111 * Returns the number of the unit assigned or a negative errno code. 1112 */ 1113 1114 int dev_alloc_name(struct net_device *dev, const char *name) 1115 { 1116 return dev_alloc_name_ns(dev_net(dev), dev, name); 1117 } 1118 EXPORT_SYMBOL(dev_alloc_name); 1119 1120 static int dev_get_valid_name(struct net *net, struct net_device *dev, 1121 const char *name) 1122 { 1123 BUG_ON(!net); 1124 1125 if (!dev_valid_name(name)) 1126 return -EINVAL; 1127 1128 if (strchr(name, '%')) 1129 return dev_alloc_name_ns(net, dev, name); 1130 else if (netdev_name_in_use(net, name)) 1131 return -EEXIST; 1132 else if (dev->name != name) 1133 strlcpy(dev->name, name, IFNAMSIZ); 1134 1135 return 0; 1136 } 1137 1138 /** 1139 * dev_change_name - change name of a device 1140 * @dev: device 1141 * @newname: name (or format string) must be at least IFNAMSIZ 1142 * 1143 * Change name of a device, can pass format strings "eth%d". 1144 * for wildcarding. 1145 */ 1146 int dev_change_name(struct net_device *dev, const char *newname) 1147 { 1148 unsigned char old_assign_type; 1149 char oldname[IFNAMSIZ]; 1150 int err = 0; 1151 int ret; 1152 struct net *net; 1153 1154 ASSERT_RTNL(); 1155 BUG_ON(!dev_net(dev)); 1156 1157 net = dev_net(dev); 1158 1159 /* Some auto-enslaved devices e.g. failover slaves are 1160 * special, as userspace might rename the device after 1161 * the interface had been brought up and running since 1162 * the point kernel initiated auto-enslavement. Allow 1163 * live name change even when these slave devices are 1164 * up and running. 1165 * 1166 * Typically, users of these auto-enslaving devices 1167 * don't actually care about slave name change, as 1168 * they are supposed to operate on master interface 1169 * directly. 1170 */ 1171 if (dev->flags & IFF_UP && 1172 likely(!(dev->priv_flags & IFF_LIVE_RENAME_OK))) 1173 return -EBUSY; 1174 1175 down_write(&devnet_rename_sem); 1176 1177 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { 1178 up_write(&devnet_rename_sem); 1179 return 0; 1180 } 1181 1182 memcpy(oldname, dev->name, IFNAMSIZ); 1183 1184 err = dev_get_valid_name(net, dev, newname); 1185 if (err < 0) { 1186 up_write(&devnet_rename_sem); 1187 return err; 1188 } 1189 1190 if (oldname[0] && !strchr(oldname, '%')) 1191 netdev_info(dev, "renamed from %s\n", oldname); 1192 1193 old_assign_type = dev->name_assign_type; 1194 dev->name_assign_type = NET_NAME_RENAMED; 1195 1196 rollback: 1197 ret = device_rename(&dev->dev, dev->name); 1198 if (ret) { 1199 memcpy(dev->name, oldname, IFNAMSIZ); 1200 dev->name_assign_type = old_assign_type; 1201 up_write(&devnet_rename_sem); 1202 return ret; 1203 } 1204 1205 up_write(&devnet_rename_sem); 1206 1207 netdev_adjacent_rename_links(dev, oldname); 1208 1209 write_lock(&dev_base_lock); 1210 netdev_name_node_del(dev->name_node); 1211 write_unlock(&dev_base_lock); 1212 1213 synchronize_rcu(); 1214 1215 write_lock(&dev_base_lock); 1216 netdev_name_node_add(net, dev->name_node); 1217 write_unlock(&dev_base_lock); 1218 1219 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); 1220 ret = notifier_to_errno(ret); 1221 1222 if (ret) { 1223 /* err >= 0 after dev_alloc_name() or stores the first errno */ 1224 if (err >= 0) { 1225 err = ret; 1226 down_write(&devnet_rename_sem); 1227 memcpy(dev->name, oldname, IFNAMSIZ); 1228 memcpy(oldname, newname, IFNAMSIZ); 1229 dev->name_assign_type = old_assign_type; 1230 old_assign_type = NET_NAME_RENAMED; 1231 goto rollback; 1232 } else { 1233 netdev_err(dev, "name change rollback failed: %d\n", 1234 ret); 1235 } 1236 } 1237 1238 return err; 1239 } 1240 1241 /** 1242 * dev_set_alias - change ifalias of a device 1243 * @dev: device 1244 * @alias: name up to IFALIASZ 1245 * @len: limit of bytes to copy from info 1246 * 1247 * Set ifalias for a device, 1248 */ 1249 int dev_set_alias(struct net_device *dev, const char *alias, size_t len) 1250 { 1251 struct dev_ifalias *new_alias = NULL; 1252 1253 if (len >= IFALIASZ) 1254 return -EINVAL; 1255 1256 if (len) { 1257 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL); 1258 if (!new_alias) 1259 return -ENOMEM; 1260 1261 memcpy(new_alias->ifalias, alias, len); 1262 new_alias->ifalias[len] = 0; 1263 } 1264 1265 mutex_lock(&ifalias_mutex); 1266 new_alias = rcu_replace_pointer(dev->ifalias, new_alias, 1267 mutex_is_locked(&ifalias_mutex)); 1268 mutex_unlock(&ifalias_mutex); 1269 1270 if (new_alias) 1271 kfree_rcu(new_alias, rcuhead); 1272 1273 return len; 1274 } 1275 EXPORT_SYMBOL(dev_set_alias); 1276 1277 /** 1278 * dev_get_alias - get ifalias of a device 1279 * @dev: device 1280 * @name: buffer to store name of ifalias 1281 * @len: size of buffer 1282 * 1283 * get ifalias for a device. Caller must make sure dev cannot go 1284 * away, e.g. rcu read lock or own a reference count to device. 1285 */ 1286 int dev_get_alias(const struct net_device *dev, char *name, size_t len) 1287 { 1288 const struct dev_ifalias *alias; 1289 int ret = 0; 1290 1291 rcu_read_lock(); 1292 alias = rcu_dereference(dev->ifalias); 1293 if (alias) 1294 ret = snprintf(name, len, "%s", alias->ifalias); 1295 rcu_read_unlock(); 1296 1297 return ret; 1298 } 1299 1300 /** 1301 * netdev_features_change - device changes features 1302 * @dev: device to cause notification 1303 * 1304 * Called to indicate a device has changed features. 1305 */ 1306 void netdev_features_change(struct net_device *dev) 1307 { 1308 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); 1309 } 1310 EXPORT_SYMBOL(netdev_features_change); 1311 1312 /** 1313 * netdev_state_change - device changes state 1314 * @dev: device to cause notification 1315 * 1316 * Called to indicate a device has changed state. This function calls 1317 * the notifier chains for netdev_chain and sends a NEWLINK message 1318 * to the routing socket. 1319 */ 1320 void netdev_state_change(struct net_device *dev) 1321 { 1322 if (dev->flags & IFF_UP) { 1323 struct netdev_notifier_change_info change_info = { 1324 .info.dev = dev, 1325 }; 1326 1327 call_netdevice_notifiers_info(NETDEV_CHANGE, 1328 &change_info.info); 1329 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL); 1330 } 1331 } 1332 EXPORT_SYMBOL(netdev_state_change); 1333 1334 /** 1335 * __netdev_notify_peers - notify network peers about existence of @dev, 1336 * to be called when rtnl lock is already held. 1337 * @dev: network device 1338 * 1339 * Generate traffic such that interested network peers are aware of 1340 * @dev, such as by generating a gratuitous ARP. This may be used when 1341 * a device wants to inform the rest of the network about some sort of 1342 * reconfiguration such as a failover event or virtual machine 1343 * migration. 1344 */ 1345 void __netdev_notify_peers(struct net_device *dev) 1346 { 1347 ASSERT_RTNL(); 1348 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); 1349 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev); 1350 } 1351 EXPORT_SYMBOL(__netdev_notify_peers); 1352 1353 /** 1354 * netdev_notify_peers - notify network peers about existence of @dev 1355 * @dev: network device 1356 * 1357 * Generate traffic such that interested network peers are aware of 1358 * @dev, such as by generating a gratuitous ARP. This may be used when 1359 * a device wants to inform the rest of the network about some sort of 1360 * reconfiguration such as a failover event or virtual machine 1361 * migration. 1362 */ 1363 void netdev_notify_peers(struct net_device *dev) 1364 { 1365 rtnl_lock(); 1366 __netdev_notify_peers(dev); 1367 rtnl_unlock(); 1368 } 1369 EXPORT_SYMBOL(netdev_notify_peers); 1370 1371 static int napi_threaded_poll(void *data); 1372 1373 static int napi_kthread_create(struct napi_struct *n) 1374 { 1375 int err = 0; 1376 1377 /* Create and wake up the kthread once to put it in 1378 * TASK_INTERRUPTIBLE mode to avoid the blocked task 1379 * warning and work with loadavg. 1380 */ 1381 n->thread = kthread_run(napi_threaded_poll, n, "napi/%s-%d", 1382 n->dev->name, n->napi_id); 1383 if (IS_ERR(n->thread)) { 1384 err = PTR_ERR(n->thread); 1385 pr_err("kthread_run failed with err %d\n", err); 1386 n->thread = NULL; 1387 } 1388 1389 return err; 1390 } 1391 1392 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1393 { 1394 const struct net_device_ops *ops = dev->netdev_ops; 1395 int ret; 1396 1397 ASSERT_RTNL(); 1398 dev_addr_check(dev); 1399 1400 if (!netif_device_present(dev)) { 1401 /* may be detached because parent is runtime-suspended */ 1402 if (dev->dev.parent) 1403 pm_runtime_resume(dev->dev.parent); 1404 if (!netif_device_present(dev)) 1405 return -ENODEV; 1406 } 1407 1408 /* Block netpoll from trying to do any rx path servicing. 1409 * If we don't do this there is a chance ndo_poll_controller 1410 * or ndo_poll may be running while we open the device 1411 */ 1412 netpoll_poll_disable(dev); 1413 1414 ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack); 1415 ret = notifier_to_errno(ret); 1416 if (ret) 1417 return ret; 1418 1419 set_bit(__LINK_STATE_START, &dev->state); 1420 1421 if (ops->ndo_validate_addr) 1422 ret = ops->ndo_validate_addr(dev); 1423 1424 if (!ret && ops->ndo_open) 1425 ret = ops->ndo_open(dev); 1426 1427 netpoll_poll_enable(dev); 1428 1429 if (ret) 1430 clear_bit(__LINK_STATE_START, &dev->state); 1431 else { 1432 dev->flags |= IFF_UP; 1433 dev_set_rx_mode(dev); 1434 dev_activate(dev); 1435 add_device_randomness(dev->dev_addr, dev->addr_len); 1436 } 1437 1438 return ret; 1439 } 1440 1441 /** 1442 * dev_open - prepare an interface for use. 1443 * @dev: device to open 1444 * @extack: netlink extended ack 1445 * 1446 * Takes a device from down to up state. The device's private open 1447 * function is invoked and then the multicast lists are loaded. Finally 1448 * the device is moved into the up state and a %NETDEV_UP message is 1449 * sent to the netdev notifier chain. 1450 * 1451 * Calling this function on an active interface is a nop. On a failure 1452 * a negative errno code is returned. 1453 */ 1454 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1455 { 1456 int ret; 1457 1458 if (dev->flags & IFF_UP) 1459 return 0; 1460 1461 ret = __dev_open(dev, extack); 1462 if (ret < 0) 1463 return ret; 1464 1465 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); 1466 call_netdevice_notifiers(NETDEV_UP, dev); 1467 1468 return ret; 1469 } 1470 EXPORT_SYMBOL(dev_open); 1471 1472 static void __dev_close_many(struct list_head *head) 1473 { 1474 struct net_device *dev; 1475 1476 ASSERT_RTNL(); 1477 might_sleep(); 1478 1479 list_for_each_entry(dev, head, close_list) { 1480 /* Temporarily disable netpoll until the interface is down */ 1481 netpoll_poll_disable(dev); 1482 1483 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); 1484 1485 clear_bit(__LINK_STATE_START, &dev->state); 1486 1487 /* Synchronize to scheduled poll. We cannot touch poll list, it 1488 * can be even on different cpu. So just clear netif_running(). 1489 * 1490 * dev->stop() will invoke napi_disable() on all of it's 1491 * napi_struct instances on this device. 1492 */ 1493 smp_mb__after_atomic(); /* Commit netif_running(). */ 1494 } 1495 1496 dev_deactivate_many(head); 1497 1498 list_for_each_entry(dev, head, close_list) { 1499 const struct net_device_ops *ops = dev->netdev_ops; 1500 1501 /* 1502 * Call the device specific close. This cannot fail. 1503 * Only if device is UP 1504 * 1505 * We allow it to be called even after a DETACH hot-plug 1506 * event. 1507 */ 1508 if (ops->ndo_stop) 1509 ops->ndo_stop(dev); 1510 1511 dev->flags &= ~IFF_UP; 1512 netpoll_poll_enable(dev); 1513 } 1514 } 1515 1516 static void __dev_close(struct net_device *dev) 1517 { 1518 LIST_HEAD(single); 1519 1520 list_add(&dev->close_list, &single); 1521 __dev_close_many(&single); 1522 list_del(&single); 1523 } 1524 1525 void dev_close_many(struct list_head *head, bool unlink) 1526 { 1527 struct net_device *dev, *tmp; 1528 1529 /* Remove the devices that don't need to be closed */ 1530 list_for_each_entry_safe(dev, tmp, head, close_list) 1531 if (!(dev->flags & IFF_UP)) 1532 list_del_init(&dev->close_list); 1533 1534 __dev_close_many(head); 1535 1536 list_for_each_entry_safe(dev, tmp, head, close_list) { 1537 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); 1538 call_netdevice_notifiers(NETDEV_DOWN, dev); 1539 if (unlink) 1540 list_del_init(&dev->close_list); 1541 } 1542 } 1543 EXPORT_SYMBOL(dev_close_many); 1544 1545 /** 1546 * dev_close - shutdown an interface. 1547 * @dev: device to shutdown 1548 * 1549 * This function moves an active device into down state. A 1550 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device 1551 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier 1552 * chain. 1553 */ 1554 void dev_close(struct net_device *dev) 1555 { 1556 if (dev->flags & IFF_UP) { 1557 LIST_HEAD(single); 1558 1559 list_add(&dev->close_list, &single); 1560 dev_close_many(&single, true); 1561 list_del(&single); 1562 } 1563 } 1564 EXPORT_SYMBOL(dev_close); 1565 1566 1567 /** 1568 * dev_disable_lro - disable Large Receive Offload on a device 1569 * @dev: device 1570 * 1571 * Disable Large Receive Offload (LRO) on a net device. Must be 1572 * called under RTNL. This is needed if received packets may be 1573 * forwarded to another interface. 1574 */ 1575 void dev_disable_lro(struct net_device *dev) 1576 { 1577 struct net_device *lower_dev; 1578 struct list_head *iter; 1579 1580 dev->wanted_features &= ~NETIF_F_LRO; 1581 netdev_update_features(dev); 1582 1583 if (unlikely(dev->features & NETIF_F_LRO)) 1584 netdev_WARN(dev, "failed to disable LRO!\n"); 1585 1586 netdev_for_each_lower_dev(dev, lower_dev, iter) 1587 dev_disable_lro(lower_dev); 1588 } 1589 EXPORT_SYMBOL(dev_disable_lro); 1590 1591 /** 1592 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device 1593 * @dev: device 1594 * 1595 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be 1596 * called under RTNL. This is needed if Generic XDP is installed on 1597 * the device. 1598 */ 1599 static void dev_disable_gro_hw(struct net_device *dev) 1600 { 1601 dev->wanted_features &= ~NETIF_F_GRO_HW; 1602 netdev_update_features(dev); 1603 1604 if (unlikely(dev->features & NETIF_F_GRO_HW)) 1605 netdev_WARN(dev, "failed to disable GRO_HW!\n"); 1606 } 1607 1608 const char *netdev_cmd_to_name(enum netdev_cmd cmd) 1609 { 1610 #define N(val) \ 1611 case NETDEV_##val: \ 1612 return "NETDEV_" __stringify(val); 1613 switch (cmd) { 1614 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER) 1615 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE) 1616 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE) 1617 N(POST_INIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) N(CHANGEUPPER) 1618 N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) N(BONDING_INFO) 1619 N(PRECHANGEUPPER) N(CHANGELOWERSTATE) N(UDP_TUNNEL_PUSH_INFO) 1620 N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN) 1621 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO) 1622 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO) 1623 N(PRE_CHANGEADDR) N(OFFLOAD_XSTATS_ENABLE) N(OFFLOAD_XSTATS_DISABLE) 1624 N(OFFLOAD_XSTATS_REPORT_USED) N(OFFLOAD_XSTATS_REPORT_DELTA) 1625 } 1626 #undef N 1627 return "UNKNOWN_NETDEV_EVENT"; 1628 } 1629 EXPORT_SYMBOL_GPL(netdev_cmd_to_name); 1630 1631 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, 1632 struct net_device *dev) 1633 { 1634 struct netdev_notifier_info info = { 1635 .dev = dev, 1636 }; 1637 1638 return nb->notifier_call(nb, val, &info); 1639 } 1640 1641 static int call_netdevice_register_notifiers(struct notifier_block *nb, 1642 struct net_device *dev) 1643 { 1644 int err; 1645 1646 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); 1647 err = notifier_to_errno(err); 1648 if (err) 1649 return err; 1650 1651 if (!(dev->flags & IFF_UP)) 1652 return 0; 1653 1654 call_netdevice_notifier(nb, NETDEV_UP, dev); 1655 return 0; 1656 } 1657 1658 static void call_netdevice_unregister_notifiers(struct notifier_block *nb, 1659 struct net_device *dev) 1660 { 1661 if (dev->flags & IFF_UP) { 1662 call_netdevice_notifier(nb, NETDEV_GOING_DOWN, 1663 dev); 1664 call_netdevice_notifier(nb, NETDEV_DOWN, dev); 1665 } 1666 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); 1667 } 1668 1669 static int call_netdevice_register_net_notifiers(struct notifier_block *nb, 1670 struct net *net) 1671 { 1672 struct net_device *dev; 1673 int err; 1674 1675 for_each_netdev(net, dev) { 1676 err = call_netdevice_register_notifiers(nb, dev); 1677 if (err) 1678 goto rollback; 1679 } 1680 return 0; 1681 1682 rollback: 1683 for_each_netdev_continue_reverse(net, dev) 1684 call_netdevice_unregister_notifiers(nb, dev); 1685 return err; 1686 } 1687 1688 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb, 1689 struct net *net) 1690 { 1691 struct net_device *dev; 1692 1693 for_each_netdev(net, dev) 1694 call_netdevice_unregister_notifiers(nb, dev); 1695 } 1696 1697 static int dev_boot_phase = 1; 1698 1699 /** 1700 * register_netdevice_notifier - register a network notifier block 1701 * @nb: notifier 1702 * 1703 * Register a notifier to be called when network device events occur. 1704 * The notifier passed is linked into the kernel structures and must 1705 * not be reused until it has been unregistered. A negative errno code 1706 * is returned on a failure. 1707 * 1708 * When registered all registration and up events are replayed 1709 * to the new notifier to allow device to have a race free 1710 * view of the network device list. 1711 */ 1712 1713 int register_netdevice_notifier(struct notifier_block *nb) 1714 { 1715 struct net *net; 1716 int err; 1717 1718 /* Close race with setup_net() and cleanup_net() */ 1719 down_write(&pernet_ops_rwsem); 1720 rtnl_lock(); 1721 err = raw_notifier_chain_register(&netdev_chain, nb); 1722 if (err) 1723 goto unlock; 1724 if (dev_boot_phase) 1725 goto unlock; 1726 for_each_net(net) { 1727 err = call_netdevice_register_net_notifiers(nb, net); 1728 if (err) 1729 goto rollback; 1730 } 1731 1732 unlock: 1733 rtnl_unlock(); 1734 up_write(&pernet_ops_rwsem); 1735 return err; 1736 1737 rollback: 1738 for_each_net_continue_reverse(net) 1739 call_netdevice_unregister_net_notifiers(nb, net); 1740 1741 raw_notifier_chain_unregister(&netdev_chain, nb); 1742 goto unlock; 1743 } 1744 EXPORT_SYMBOL(register_netdevice_notifier); 1745 1746 /** 1747 * unregister_netdevice_notifier - unregister a network notifier block 1748 * @nb: notifier 1749 * 1750 * Unregister a notifier previously registered by 1751 * register_netdevice_notifier(). The notifier is unlinked into the 1752 * kernel structures and may then be reused. A negative errno code 1753 * is returned on a failure. 1754 * 1755 * After unregistering unregister and down device events are synthesized 1756 * for all devices on the device list to the removed notifier to remove 1757 * the need for special case cleanup code. 1758 */ 1759 1760 int unregister_netdevice_notifier(struct notifier_block *nb) 1761 { 1762 struct net *net; 1763 int err; 1764 1765 /* Close race with setup_net() and cleanup_net() */ 1766 down_write(&pernet_ops_rwsem); 1767 rtnl_lock(); 1768 err = raw_notifier_chain_unregister(&netdev_chain, nb); 1769 if (err) 1770 goto unlock; 1771 1772 for_each_net(net) 1773 call_netdevice_unregister_net_notifiers(nb, net); 1774 1775 unlock: 1776 rtnl_unlock(); 1777 up_write(&pernet_ops_rwsem); 1778 return err; 1779 } 1780 EXPORT_SYMBOL(unregister_netdevice_notifier); 1781 1782 static int __register_netdevice_notifier_net(struct net *net, 1783 struct notifier_block *nb, 1784 bool ignore_call_fail) 1785 { 1786 int err; 1787 1788 err = raw_notifier_chain_register(&net->netdev_chain, nb); 1789 if (err) 1790 return err; 1791 if (dev_boot_phase) 1792 return 0; 1793 1794 err = call_netdevice_register_net_notifiers(nb, net); 1795 if (err && !ignore_call_fail) 1796 goto chain_unregister; 1797 1798 return 0; 1799 1800 chain_unregister: 1801 raw_notifier_chain_unregister(&net->netdev_chain, nb); 1802 return err; 1803 } 1804 1805 static int __unregister_netdevice_notifier_net(struct net *net, 1806 struct notifier_block *nb) 1807 { 1808 int err; 1809 1810 err = raw_notifier_chain_unregister(&net->netdev_chain, nb); 1811 if (err) 1812 return err; 1813 1814 call_netdevice_unregister_net_notifiers(nb, net); 1815 return 0; 1816 } 1817 1818 /** 1819 * register_netdevice_notifier_net - register a per-netns network notifier block 1820 * @net: network namespace 1821 * @nb: notifier 1822 * 1823 * Register a notifier to be called when network device events occur. 1824 * The notifier passed is linked into the kernel structures and must 1825 * not be reused until it has been unregistered. A negative errno code 1826 * is returned on a failure. 1827 * 1828 * When registered all registration and up events are replayed 1829 * to the new notifier to allow device to have a race free 1830 * view of the network device list. 1831 */ 1832 1833 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb) 1834 { 1835 int err; 1836 1837 rtnl_lock(); 1838 err = __register_netdevice_notifier_net(net, nb, false); 1839 rtnl_unlock(); 1840 return err; 1841 } 1842 EXPORT_SYMBOL(register_netdevice_notifier_net); 1843 1844 /** 1845 * unregister_netdevice_notifier_net - unregister a per-netns 1846 * network notifier block 1847 * @net: network namespace 1848 * @nb: notifier 1849 * 1850 * Unregister a notifier previously registered by 1851 * register_netdevice_notifier(). The notifier is unlinked into the 1852 * kernel structures and may then be reused. A negative errno code 1853 * is returned on a failure. 1854 * 1855 * After unregistering unregister and down device events are synthesized 1856 * for all devices on the device list to the removed notifier to remove 1857 * the need for special case cleanup code. 1858 */ 1859 1860 int unregister_netdevice_notifier_net(struct net *net, 1861 struct notifier_block *nb) 1862 { 1863 int err; 1864 1865 rtnl_lock(); 1866 err = __unregister_netdevice_notifier_net(net, nb); 1867 rtnl_unlock(); 1868 return err; 1869 } 1870 EXPORT_SYMBOL(unregister_netdevice_notifier_net); 1871 1872 int register_netdevice_notifier_dev_net(struct net_device *dev, 1873 struct notifier_block *nb, 1874 struct netdev_net_notifier *nn) 1875 { 1876 int err; 1877 1878 rtnl_lock(); 1879 err = __register_netdevice_notifier_net(dev_net(dev), nb, false); 1880 if (!err) { 1881 nn->nb = nb; 1882 list_add(&nn->list, &dev->net_notifier_list); 1883 } 1884 rtnl_unlock(); 1885 return err; 1886 } 1887 EXPORT_SYMBOL(register_netdevice_notifier_dev_net); 1888 1889 int unregister_netdevice_notifier_dev_net(struct net_device *dev, 1890 struct notifier_block *nb, 1891 struct netdev_net_notifier *nn) 1892 { 1893 int err; 1894 1895 rtnl_lock(); 1896 list_del(&nn->list); 1897 err = __unregister_netdevice_notifier_net(dev_net(dev), nb); 1898 rtnl_unlock(); 1899 return err; 1900 } 1901 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net); 1902 1903 static void move_netdevice_notifiers_dev_net(struct net_device *dev, 1904 struct net *net) 1905 { 1906 struct netdev_net_notifier *nn; 1907 1908 list_for_each_entry(nn, &dev->net_notifier_list, list) { 1909 __unregister_netdevice_notifier_net(dev_net(dev), nn->nb); 1910 __register_netdevice_notifier_net(net, nn->nb, true); 1911 } 1912 } 1913 1914 /** 1915 * call_netdevice_notifiers_info - call all network notifier blocks 1916 * @val: value passed unmodified to notifier function 1917 * @info: notifier information data 1918 * 1919 * Call all network notifier blocks. Parameters and return value 1920 * are as for raw_notifier_call_chain(). 1921 */ 1922 1923 static int call_netdevice_notifiers_info(unsigned long val, 1924 struct netdev_notifier_info *info) 1925 { 1926 struct net *net = dev_net(info->dev); 1927 int ret; 1928 1929 ASSERT_RTNL(); 1930 1931 /* Run per-netns notifier block chain first, then run the global one. 1932 * Hopefully, one day, the global one is going to be removed after 1933 * all notifier block registrators get converted to be per-netns. 1934 */ 1935 ret = raw_notifier_call_chain(&net->netdev_chain, val, info); 1936 if (ret & NOTIFY_STOP_MASK) 1937 return ret; 1938 return raw_notifier_call_chain(&netdev_chain, val, info); 1939 } 1940 1941 /** 1942 * call_netdevice_notifiers_info_robust - call per-netns notifier blocks 1943 * for and rollback on error 1944 * @val_up: value passed unmodified to notifier function 1945 * @val_down: value passed unmodified to the notifier function when 1946 * recovering from an error on @val_up 1947 * @info: notifier information data 1948 * 1949 * Call all per-netns network notifier blocks, but not notifier blocks on 1950 * the global notifier chain. Parameters and return value are as for 1951 * raw_notifier_call_chain_robust(). 1952 */ 1953 1954 static int 1955 call_netdevice_notifiers_info_robust(unsigned long val_up, 1956 unsigned long val_down, 1957 struct netdev_notifier_info *info) 1958 { 1959 struct net *net = dev_net(info->dev); 1960 1961 ASSERT_RTNL(); 1962 1963 return raw_notifier_call_chain_robust(&net->netdev_chain, 1964 val_up, val_down, info); 1965 } 1966 1967 static int call_netdevice_notifiers_extack(unsigned long val, 1968 struct net_device *dev, 1969 struct netlink_ext_ack *extack) 1970 { 1971 struct netdev_notifier_info info = { 1972 .dev = dev, 1973 .extack = extack, 1974 }; 1975 1976 return call_netdevice_notifiers_info(val, &info); 1977 } 1978 1979 /** 1980 * call_netdevice_notifiers - call all network notifier blocks 1981 * @val: value passed unmodified to notifier function 1982 * @dev: net_device pointer passed unmodified to notifier function 1983 * 1984 * Call all network notifier blocks. Parameters and return value 1985 * are as for raw_notifier_call_chain(). 1986 */ 1987 1988 int call_netdevice_notifiers(unsigned long val, struct net_device *dev) 1989 { 1990 return call_netdevice_notifiers_extack(val, dev, NULL); 1991 } 1992 EXPORT_SYMBOL(call_netdevice_notifiers); 1993 1994 /** 1995 * call_netdevice_notifiers_mtu - call all network notifier blocks 1996 * @val: value passed unmodified to notifier function 1997 * @dev: net_device pointer passed unmodified to notifier function 1998 * @arg: additional u32 argument passed to the notifier function 1999 * 2000 * Call all network notifier blocks. Parameters and return value 2001 * are as for raw_notifier_call_chain(). 2002 */ 2003 static int call_netdevice_notifiers_mtu(unsigned long val, 2004 struct net_device *dev, u32 arg) 2005 { 2006 struct netdev_notifier_info_ext info = { 2007 .info.dev = dev, 2008 .ext.mtu = arg, 2009 }; 2010 2011 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0); 2012 2013 return call_netdevice_notifiers_info(val, &info.info); 2014 } 2015 2016 #ifdef CONFIG_NET_INGRESS 2017 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key); 2018 2019 void net_inc_ingress_queue(void) 2020 { 2021 static_branch_inc(&ingress_needed_key); 2022 } 2023 EXPORT_SYMBOL_GPL(net_inc_ingress_queue); 2024 2025 void net_dec_ingress_queue(void) 2026 { 2027 static_branch_dec(&ingress_needed_key); 2028 } 2029 EXPORT_SYMBOL_GPL(net_dec_ingress_queue); 2030 #endif 2031 2032 #ifdef CONFIG_NET_EGRESS 2033 static DEFINE_STATIC_KEY_FALSE(egress_needed_key); 2034 2035 void net_inc_egress_queue(void) 2036 { 2037 static_branch_inc(&egress_needed_key); 2038 } 2039 EXPORT_SYMBOL_GPL(net_inc_egress_queue); 2040 2041 void net_dec_egress_queue(void) 2042 { 2043 static_branch_dec(&egress_needed_key); 2044 } 2045 EXPORT_SYMBOL_GPL(net_dec_egress_queue); 2046 #endif 2047 2048 DEFINE_STATIC_KEY_FALSE(netstamp_needed_key); 2049 EXPORT_SYMBOL(netstamp_needed_key); 2050 #ifdef CONFIG_JUMP_LABEL 2051 static atomic_t netstamp_needed_deferred; 2052 static atomic_t netstamp_wanted; 2053 static void netstamp_clear(struct work_struct *work) 2054 { 2055 int deferred = atomic_xchg(&netstamp_needed_deferred, 0); 2056 int wanted; 2057 2058 wanted = atomic_add_return(deferred, &netstamp_wanted); 2059 if (wanted > 0) 2060 static_branch_enable(&netstamp_needed_key); 2061 else 2062 static_branch_disable(&netstamp_needed_key); 2063 } 2064 static DECLARE_WORK(netstamp_work, netstamp_clear); 2065 #endif 2066 2067 void net_enable_timestamp(void) 2068 { 2069 #ifdef CONFIG_JUMP_LABEL 2070 int wanted; 2071 2072 while (1) { 2073 wanted = atomic_read(&netstamp_wanted); 2074 if (wanted <= 0) 2075 break; 2076 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted) 2077 return; 2078 } 2079 atomic_inc(&netstamp_needed_deferred); 2080 schedule_work(&netstamp_work); 2081 #else 2082 static_branch_inc(&netstamp_needed_key); 2083 #endif 2084 } 2085 EXPORT_SYMBOL(net_enable_timestamp); 2086 2087 void net_disable_timestamp(void) 2088 { 2089 #ifdef CONFIG_JUMP_LABEL 2090 int wanted; 2091 2092 while (1) { 2093 wanted = atomic_read(&netstamp_wanted); 2094 if (wanted <= 1) 2095 break; 2096 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted) 2097 return; 2098 } 2099 atomic_dec(&netstamp_needed_deferred); 2100 schedule_work(&netstamp_work); 2101 #else 2102 static_branch_dec(&netstamp_needed_key); 2103 #endif 2104 } 2105 EXPORT_SYMBOL(net_disable_timestamp); 2106 2107 static inline void net_timestamp_set(struct sk_buff *skb) 2108 { 2109 skb->tstamp = 0; 2110 skb->mono_delivery_time = 0; 2111 if (static_branch_unlikely(&netstamp_needed_key)) 2112 skb->tstamp = ktime_get_real(); 2113 } 2114 2115 #define net_timestamp_check(COND, SKB) \ 2116 if (static_branch_unlikely(&netstamp_needed_key)) { \ 2117 if ((COND) && !(SKB)->tstamp) \ 2118 (SKB)->tstamp = ktime_get_real(); \ 2119 } \ 2120 2121 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb) 2122 { 2123 return __is_skb_forwardable(dev, skb, true); 2124 } 2125 EXPORT_SYMBOL_GPL(is_skb_forwardable); 2126 2127 static int __dev_forward_skb2(struct net_device *dev, struct sk_buff *skb, 2128 bool check_mtu) 2129 { 2130 int ret = ____dev_forward_skb(dev, skb, check_mtu); 2131 2132 if (likely(!ret)) { 2133 skb->protocol = eth_type_trans(skb, dev); 2134 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); 2135 } 2136 2137 return ret; 2138 } 2139 2140 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2141 { 2142 return __dev_forward_skb2(dev, skb, true); 2143 } 2144 EXPORT_SYMBOL_GPL(__dev_forward_skb); 2145 2146 /** 2147 * dev_forward_skb - loopback an skb to another netif 2148 * 2149 * @dev: destination network device 2150 * @skb: buffer to forward 2151 * 2152 * return values: 2153 * NET_RX_SUCCESS (no congestion) 2154 * NET_RX_DROP (packet was dropped, but freed) 2155 * 2156 * dev_forward_skb can be used for injecting an skb from the 2157 * start_xmit function of one device into the receive queue 2158 * of another device. 2159 * 2160 * The receiving device may be in another namespace, so 2161 * we have to clear all information in the skb that could 2162 * impact namespace isolation. 2163 */ 2164 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2165 { 2166 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); 2167 } 2168 EXPORT_SYMBOL_GPL(dev_forward_skb); 2169 2170 int dev_forward_skb_nomtu(struct net_device *dev, struct sk_buff *skb) 2171 { 2172 return __dev_forward_skb2(dev, skb, false) ?: netif_rx_internal(skb); 2173 } 2174 2175 static inline int deliver_skb(struct sk_buff *skb, 2176 struct packet_type *pt_prev, 2177 struct net_device *orig_dev) 2178 { 2179 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 2180 return -ENOMEM; 2181 refcount_inc(&skb->users); 2182 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 2183 } 2184 2185 static inline void deliver_ptype_list_skb(struct sk_buff *skb, 2186 struct packet_type **pt, 2187 struct net_device *orig_dev, 2188 __be16 type, 2189 struct list_head *ptype_list) 2190 { 2191 struct packet_type *ptype, *pt_prev = *pt; 2192 2193 list_for_each_entry_rcu(ptype, ptype_list, list) { 2194 if (ptype->type != type) 2195 continue; 2196 if (pt_prev) 2197 deliver_skb(skb, pt_prev, orig_dev); 2198 pt_prev = ptype; 2199 } 2200 *pt = pt_prev; 2201 } 2202 2203 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) 2204 { 2205 if (!ptype->af_packet_priv || !skb->sk) 2206 return false; 2207 2208 if (ptype->id_match) 2209 return ptype->id_match(ptype, skb->sk); 2210 else if ((struct sock *)ptype->af_packet_priv == skb->sk) 2211 return true; 2212 2213 return false; 2214 } 2215 2216 /** 2217 * dev_nit_active - return true if any network interface taps are in use 2218 * 2219 * @dev: network device to check for the presence of taps 2220 */ 2221 bool dev_nit_active(struct net_device *dev) 2222 { 2223 return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all); 2224 } 2225 EXPORT_SYMBOL_GPL(dev_nit_active); 2226 2227 /* 2228 * Support routine. Sends outgoing frames to any network 2229 * taps currently in use. 2230 */ 2231 2232 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) 2233 { 2234 struct packet_type *ptype; 2235 struct sk_buff *skb2 = NULL; 2236 struct packet_type *pt_prev = NULL; 2237 struct list_head *ptype_list = &ptype_all; 2238 2239 rcu_read_lock(); 2240 again: 2241 list_for_each_entry_rcu(ptype, ptype_list, list) { 2242 if (ptype->ignore_outgoing) 2243 continue; 2244 2245 /* Never send packets back to the socket 2246 * they originated from - MvS (miquels@drinkel.ow.org) 2247 */ 2248 if (skb_loop_sk(ptype, skb)) 2249 continue; 2250 2251 if (pt_prev) { 2252 deliver_skb(skb2, pt_prev, skb->dev); 2253 pt_prev = ptype; 2254 continue; 2255 } 2256 2257 /* need to clone skb, done only once */ 2258 skb2 = skb_clone(skb, GFP_ATOMIC); 2259 if (!skb2) 2260 goto out_unlock; 2261 2262 net_timestamp_set(skb2); 2263 2264 /* skb->nh should be correctly 2265 * set by sender, so that the second statement is 2266 * just protection against buggy protocols. 2267 */ 2268 skb_reset_mac_header(skb2); 2269 2270 if (skb_network_header(skb2) < skb2->data || 2271 skb_network_header(skb2) > skb_tail_pointer(skb2)) { 2272 net_crit_ratelimited("protocol %04x is buggy, dev %s\n", 2273 ntohs(skb2->protocol), 2274 dev->name); 2275 skb_reset_network_header(skb2); 2276 } 2277 2278 skb2->transport_header = skb2->network_header; 2279 skb2->pkt_type = PACKET_OUTGOING; 2280 pt_prev = ptype; 2281 } 2282 2283 if (ptype_list == &ptype_all) { 2284 ptype_list = &dev->ptype_all; 2285 goto again; 2286 } 2287 out_unlock: 2288 if (pt_prev) { 2289 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC)) 2290 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); 2291 else 2292 kfree_skb(skb2); 2293 } 2294 rcu_read_unlock(); 2295 } 2296 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit); 2297 2298 /** 2299 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change 2300 * @dev: Network device 2301 * @txq: number of queues available 2302 * 2303 * If real_num_tx_queues is changed the tc mappings may no longer be 2304 * valid. To resolve this verify the tc mapping remains valid and if 2305 * not NULL the mapping. With no priorities mapping to this 2306 * offset/count pair it will no longer be used. In the worst case TC0 2307 * is invalid nothing can be done so disable priority mappings. If is 2308 * expected that drivers will fix this mapping if they can before 2309 * calling netif_set_real_num_tx_queues. 2310 */ 2311 static void netif_setup_tc(struct net_device *dev, unsigned int txq) 2312 { 2313 int i; 2314 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2315 2316 /* If TC0 is invalidated disable TC mapping */ 2317 if (tc->offset + tc->count > txq) { 2318 netdev_warn(dev, "Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); 2319 dev->num_tc = 0; 2320 return; 2321 } 2322 2323 /* Invalidated prio to tc mappings set to TC0 */ 2324 for (i = 1; i < TC_BITMASK + 1; i++) { 2325 int q = netdev_get_prio_tc_map(dev, i); 2326 2327 tc = &dev->tc_to_txq[q]; 2328 if (tc->offset + tc->count > txq) { 2329 netdev_warn(dev, "Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n", 2330 i, q); 2331 netdev_set_prio_tc_map(dev, i, 0); 2332 } 2333 } 2334 } 2335 2336 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq) 2337 { 2338 if (dev->num_tc) { 2339 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2340 int i; 2341 2342 /* walk through the TCs and see if it falls into any of them */ 2343 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) { 2344 if ((txq - tc->offset) < tc->count) 2345 return i; 2346 } 2347 2348 /* didn't find it, just return -1 to indicate no match */ 2349 return -1; 2350 } 2351 2352 return 0; 2353 } 2354 EXPORT_SYMBOL(netdev_txq_to_tc); 2355 2356 #ifdef CONFIG_XPS 2357 static struct static_key xps_needed __read_mostly; 2358 static struct static_key xps_rxqs_needed __read_mostly; 2359 static DEFINE_MUTEX(xps_map_mutex); 2360 #define xmap_dereference(P) \ 2361 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) 2362 2363 static bool remove_xps_queue(struct xps_dev_maps *dev_maps, 2364 struct xps_dev_maps *old_maps, int tci, u16 index) 2365 { 2366 struct xps_map *map = NULL; 2367 int pos; 2368 2369 if (dev_maps) 2370 map = xmap_dereference(dev_maps->attr_map[tci]); 2371 if (!map) 2372 return false; 2373 2374 for (pos = map->len; pos--;) { 2375 if (map->queues[pos] != index) 2376 continue; 2377 2378 if (map->len > 1) { 2379 map->queues[pos] = map->queues[--map->len]; 2380 break; 2381 } 2382 2383 if (old_maps) 2384 RCU_INIT_POINTER(old_maps->attr_map[tci], NULL); 2385 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2386 kfree_rcu(map, rcu); 2387 return false; 2388 } 2389 2390 return true; 2391 } 2392 2393 static bool remove_xps_queue_cpu(struct net_device *dev, 2394 struct xps_dev_maps *dev_maps, 2395 int cpu, u16 offset, u16 count) 2396 { 2397 int num_tc = dev_maps->num_tc; 2398 bool active = false; 2399 int tci; 2400 2401 for (tci = cpu * num_tc; num_tc--; tci++) { 2402 int i, j; 2403 2404 for (i = count, j = offset; i--; j++) { 2405 if (!remove_xps_queue(dev_maps, NULL, tci, j)) 2406 break; 2407 } 2408 2409 active |= i < 0; 2410 } 2411 2412 return active; 2413 } 2414 2415 static void reset_xps_maps(struct net_device *dev, 2416 struct xps_dev_maps *dev_maps, 2417 enum xps_map_type type) 2418 { 2419 static_key_slow_dec_cpuslocked(&xps_needed); 2420 if (type == XPS_RXQS) 2421 static_key_slow_dec_cpuslocked(&xps_rxqs_needed); 2422 2423 RCU_INIT_POINTER(dev->xps_maps[type], NULL); 2424 2425 kfree_rcu(dev_maps, rcu); 2426 } 2427 2428 static void clean_xps_maps(struct net_device *dev, enum xps_map_type type, 2429 u16 offset, u16 count) 2430 { 2431 struct xps_dev_maps *dev_maps; 2432 bool active = false; 2433 int i, j; 2434 2435 dev_maps = xmap_dereference(dev->xps_maps[type]); 2436 if (!dev_maps) 2437 return; 2438 2439 for (j = 0; j < dev_maps->nr_ids; j++) 2440 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, count); 2441 if (!active) 2442 reset_xps_maps(dev, dev_maps, type); 2443 2444 if (type == XPS_CPUS) { 2445 for (i = offset + (count - 1); count--; i--) 2446 netdev_queue_numa_node_write( 2447 netdev_get_tx_queue(dev, i), NUMA_NO_NODE); 2448 } 2449 } 2450 2451 static void netif_reset_xps_queues(struct net_device *dev, u16 offset, 2452 u16 count) 2453 { 2454 if (!static_key_false(&xps_needed)) 2455 return; 2456 2457 cpus_read_lock(); 2458 mutex_lock(&xps_map_mutex); 2459 2460 if (static_key_false(&xps_rxqs_needed)) 2461 clean_xps_maps(dev, XPS_RXQS, offset, count); 2462 2463 clean_xps_maps(dev, XPS_CPUS, offset, count); 2464 2465 mutex_unlock(&xps_map_mutex); 2466 cpus_read_unlock(); 2467 } 2468 2469 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) 2470 { 2471 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index); 2472 } 2473 2474 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index, 2475 u16 index, bool is_rxqs_map) 2476 { 2477 struct xps_map *new_map; 2478 int alloc_len = XPS_MIN_MAP_ALLOC; 2479 int i, pos; 2480 2481 for (pos = 0; map && pos < map->len; pos++) { 2482 if (map->queues[pos] != index) 2483 continue; 2484 return map; 2485 } 2486 2487 /* Need to add tx-queue to this CPU's/rx-queue's existing map */ 2488 if (map) { 2489 if (pos < map->alloc_len) 2490 return map; 2491 2492 alloc_len = map->alloc_len * 2; 2493 } 2494 2495 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's 2496 * map 2497 */ 2498 if (is_rxqs_map) 2499 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL); 2500 else 2501 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, 2502 cpu_to_node(attr_index)); 2503 if (!new_map) 2504 return NULL; 2505 2506 for (i = 0; i < pos; i++) 2507 new_map->queues[i] = map->queues[i]; 2508 new_map->alloc_len = alloc_len; 2509 new_map->len = pos; 2510 2511 return new_map; 2512 } 2513 2514 /* Copy xps maps at a given index */ 2515 static void xps_copy_dev_maps(struct xps_dev_maps *dev_maps, 2516 struct xps_dev_maps *new_dev_maps, int index, 2517 int tc, bool skip_tc) 2518 { 2519 int i, tci = index * dev_maps->num_tc; 2520 struct xps_map *map; 2521 2522 /* copy maps belonging to foreign traffic classes */ 2523 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2524 if (i == tc && skip_tc) 2525 continue; 2526 2527 /* fill in the new device map from the old device map */ 2528 map = xmap_dereference(dev_maps->attr_map[tci]); 2529 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2530 } 2531 } 2532 2533 /* Must be called under cpus_read_lock */ 2534 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask, 2535 u16 index, enum xps_map_type type) 2536 { 2537 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL, *old_dev_maps = NULL; 2538 const unsigned long *online_mask = NULL; 2539 bool active = false, copy = false; 2540 int i, j, tci, numa_node_id = -2; 2541 int maps_sz, num_tc = 1, tc = 0; 2542 struct xps_map *map, *new_map; 2543 unsigned int nr_ids; 2544 2545 if (dev->num_tc) { 2546 /* Do not allow XPS on subordinate device directly */ 2547 num_tc = dev->num_tc; 2548 if (num_tc < 0) 2549 return -EINVAL; 2550 2551 /* If queue belongs to subordinate dev use its map */ 2552 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev; 2553 2554 tc = netdev_txq_to_tc(dev, index); 2555 if (tc < 0) 2556 return -EINVAL; 2557 } 2558 2559 mutex_lock(&xps_map_mutex); 2560 2561 dev_maps = xmap_dereference(dev->xps_maps[type]); 2562 if (type == XPS_RXQS) { 2563 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues); 2564 nr_ids = dev->num_rx_queues; 2565 } else { 2566 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc); 2567 if (num_possible_cpus() > 1) 2568 online_mask = cpumask_bits(cpu_online_mask); 2569 nr_ids = nr_cpu_ids; 2570 } 2571 2572 if (maps_sz < L1_CACHE_BYTES) 2573 maps_sz = L1_CACHE_BYTES; 2574 2575 /* The old dev_maps could be larger or smaller than the one we're 2576 * setting up now, as dev->num_tc or nr_ids could have been updated in 2577 * between. We could try to be smart, but let's be safe instead and only 2578 * copy foreign traffic classes if the two map sizes match. 2579 */ 2580 if (dev_maps && 2581 dev_maps->num_tc == num_tc && dev_maps->nr_ids == nr_ids) 2582 copy = true; 2583 2584 /* allocate memory for queue storage */ 2585 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids), 2586 j < nr_ids;) { 2587 if (!new_dev_maps) { 2588 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); 2589 if (!new_dev_maps) { 2590 mutex_unlock(&xps_map_mutex); 2591 return -ENOMEM; 2592 } 2593 2594 new_dev_maps->nr_ids = nr_ids; 2595 new_dev_maps->num_tc = num_tc; 2596 } 2597 2598 tci = j * num_tc + tc; 2599 map = copy ? xmap_dereference(dev_maps->attr_map[tci]) : NULL; 2600 2601 map = expand_xps_map(map, j, index, type == XPS_RXQS); 2602 if (!map) 2603 goto error; 2604 2605 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2606 } 2607 2608 if (!new_dev_maps) 2609 goto out_no_new_maps; 2610 2611 if (!dev_maps) { 2612 /* Increment static keys at most once per type */ 2613 static_key_slow_inc_cpuslocked(&xps_needed); 2614 if (type == XPS_RXQS) 2615 static_key_slow_inc_cpuslocked(&xps_rxqs_needed); 2616 } 2617 2618 for (j = 0; j < nr_ids; j++) { 2619 bool skip_tc = false; 2620 2621 tci = j * num_tc + tc; 2622 if (netif_attr_test_mask(j, mask, nr_ids) && 2623 netif_attr_test_online(j, online_mask, nr_ids)) { 2624 /* add tx-queue to CPU/rx-queue maps */ 2625 int pos = 0; 2626 2627 skip_tc = true; 2628 2629 map = xmap_dereference(new_dev_maps->attr_map[tci]); 2630 while ((pos < map->len) && (map->queues[pos] != index)) 2631 pos++; 2632 2633 if (pos == map->len) 2634 map->queues[map->len++] = index; 2635 #ifdef CONFIG_NUMA 2636 if (type == XPS_CPUS) { 2637 if (numa_node_id == -2) 2638 numa_node_id = cpu_to_node(j); 2639 else if (numa_node_id != cpu_to_node(j)) 2640 numa_node_id = -1; 2641 } 2642 #endif 2643 } 2644 2645 if (copy) 2646 xps_copy_dev_maps(dev_maps, new_dev_maps, j, tc, 2647 skip_tc); 2648 } 2649 2650 rcu_assign_pointer(dev->xps_maps[type], new_dev_maps); 2651 2652 /* Cleanup old maps */ 2653 if (!dev_maps) 2654 goto out_no_old_maps; 2655 2656 for (j = 0; j < dev_maps->nr_ids; j++) { 2657 for (i = num_tc, tci = j * dev_maps->num_tc; i--; tci++) { 2658 map = xmap_dereference(dev_maps->attr_map[tci]); 2659 if (!map) 2660 continue; 2661 2662 if (copy) { 2663 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2664 if (map == new_map) 2665 continue; 2666 } 2667 2668 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2669 kfree_rcu(map, rcu); 2670 } 2671 } 2672 2673 old_dev_maps = dev_maps; 2674 2675 out_no_old_maps: 2676 dev_maps = new_dev_maps; 2677 active = true; 2678 2679 out_no_new_maps: 2680 if (type == XPS_CPUS) 2681 /* update Tx queue numa node */ 2682 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), 2683 (numa_node_id >= 0) ? 2684 numa_node_id : NUMA_NO_NODE); 2685 2686 if (!dev_maps) 2687 goto out_no_maps; 2688 2689 /* removes tx-queue from unused CPUs/rx-queues */ 2690 for (j = 0; j < dev_maps->nr_ids; j++) { 2691 tci = j * dev_maps->num_tc; 2692 2693 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2694 if (i == tc && 2695 netif_attr_test_mask(j, mask, dev_maps->nr_ids) && 2696 netif_attr_test_online(j, online_mask, dev_maps->nr_ids)) 2697 continue; 2698 2699 active |= remove_xps_queue(dev_maps, 2700 copy ? old_dev_maps : NULL, 2701 tci, index); 2702 } 2703 } 2704 2705 if (old_dev_maps) 2706 kfree_rcu(old_dev_maps, rcu); 2707 2708 /* free map if not active */ 2709 if (!active) 2710 reset_xps_maps(dev, dev_maps, type); 2711 2712 out_no_maps: 2713 mutex_unlock(&xps_map_mutex); 2714 2715 return 0; 2716 error: 2717 /* remove any maps that we added */ 2718 for (j = 0; j < nr_ids; j++) { 2719 for (i = num_tc, tci = j * num_tc; i--; tci++) { 2720 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2721 map = copy ? 2722 xmap_dereference(dev_maps->attr_map[tci]) : 2723 NULL; 2724 if (new_map && new_map != map) 2725 kfree(new_map); 2726 } 2727 } 2728 2729 mutex_unlock(&xps_map_mutex); 2730 2731 kfree(new_dev_maps); 2732 return -ENOMEM; 2733 } 2734 EXPORT_SYMBOL_GPL(__netif_set_xps_queue); 2735 2736 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, 2737 u16 index) 2738 { 2739 int ret; 2740 2741 cpus_read_lock(); 2742 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, XPS_CPUS); 2743 cpus_read_unlock(); 2744 2745 return ret; 2746 } 2747 EXPORT_SYMBOL(netif_set_xps_queue); 2748 2749 #endif 2750 static void netdev_unbind_all_sb_channels(struct net_device *dev) 2751 { 2752 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2753 2754 /* Unbind any subordinate channels */ 2755 while (txq-- != &dev->_tx[0]) { 2756 if (txq->sb_dev) 2757 netdev_unbind_sb_channel(dev, txq->sb_dev); 2758 } 2759 } 2760 2761 void netdev_reset_tc(struct net_device *dev) 2762 { 2763 #ifdef CONFIG_XPS 2764 netif_reset_xps_queues_gt(dev, 0); 2765 #endif 2766 netdev_unbind_all_sb_channels(dev); 2767 2768 /* Reset TC configuration of device */ 2769 dev->num_tc = 0; 2770 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq)); 2771 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map)); 2772 } 2773 EXPORT_SYMBOL(netdev_reset_tc); 2774 2775 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset) 2776 { 2777 if (tc >= dev->num_tc) 2778 return -EINVAL; 2779 2780 #ifdef CONFIG_XPS 2781 netif_reset_xps_queues(dev, offset, count); 2782 #endif 2783 dev->tc_to_txq[tc].count = count; 2784 dev->tc_to_txq[tc].offset = offset; 2785 return 0; 2786 } 2787 EXPORT_SYMBOL(netdev_set_tc_queue); 2788 2789 int netdev_set_num_tc(struct net_device *dev, u8 num_tc) 2790 { 2791 if (num_tc > TC_MAX_QUEUE) 2792 return -EINVAL; 2793 2794 #ifdef CONFIG_XPS 2795 netif_reset_xps_queues_gt(dev, 0); 2796 #endif 2797 netdev_unbind_all_sb_channels(dev); 2798 2799 dev->num_tc = num_tc; 2800 return 0; 2801 } 2802 EXPORT_SYMBOL(netdev_set_num_tc); 2803 2804 void netdev_unbind_sb_channel(struct net_device *dev, 2805 struct net_device *sb_dev) 2806 { 2807 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2808 2809 #ifdef CONFIG_XPS 2810 netif_reset_xps_queues_gt(sb_dev, 0); 2811 #endif 2812 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq)); 2813 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map)); 2814 2815 while (txq-- != &dev->_tx[0]) { 2816 if (txq->sb_dev == sb_dev) 2817 txq->sb_dev = NULL; 2818 } 2819 } 2820 EXPORT_SYMBOL(netdev_unbind_sb_channel); 2821 2822 int netdev_bind_sb_channel_queue(struct net_device *dev, 2823 struct net_device *sb_dev, 2824 u8 tc, u16 count, u16 offset) 2825 { 2826 /* Make certain the sb_dev and dev are already configured */ 2827 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc) 2828 return -EINVAL; 2829 2830 /* We cannot hand out queues we don't have */ 2831 if ((offset + count) > dev->real_num_tx_queues) 2832 return -EINVAL; 2833 2834 /* Record the mapping */ 2835 sb_dev->tc_to_txq[tc].count = count; 2836 sb_dev->tc_to_txq[tc].offset = offset; 2837 2838 /* Provide a way for Tx queue to find the tc_to_txq map or 2839 * XPS map for itself. 2840 */ 2841 while (count--) 2842 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev; 2843 2844 return 0; 2845 } 2846 EXPORT_SYMBOL(netdev_bind_sb_channel_queue); 2847 2848 int netdev_set_sb_channel(struct net_device *dev, u16 channel) 2849 { 2850 /* Do not use a multiqueue device to represent a subordinate channel */ 2851 if (netif_is_multiqueue(dev)) 2852 return -ENODEV; 2853 2854 /* We allow channels 1 - 32767 to be used for subordinate channels. 2855 * Channel 0 is meant to be "native" mode and used only to represent 2856 * the main root device. We allow writing 0 to reset the device back 2857 * to normal mode after being used as a subordinate channel. 2858 */ 2859 if (channel > S16_MAX) 2860 return -EINVAL; 2861 2862 dev->num_tc = -channel; 2863 2864 return 0; 2865 } 2866 EXPORT_SYMBOL(netdev_set_sb_channel); 2867 2868 /* 2869 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues 2870 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed. 2871 */ 2872 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) 2873 { 2874 bool disabling; 2875 int rc; 2876 2877 disabling = txq < dev->real_num_tx_queues; 2878 2879 if (txq < 1 || txq > dev->num_tx_queues) 2880 return -EINVAL; 2881 2882 if (dev->reg_state == NETREG_REGISTERED || 2883 dev->reg_state == NETREG_UNREGISTERING) { 2884 ASSERT_RTNL(); 2885 2886 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, 2887 txq); 2888 if (rc) 2889 return rc; 2890 2891 if (dev->num_tc) 2892 netif_setup_tc(dev, txq); 2893 2894 dev_qdisc_change_real_num_tx(dev, txq); 2895 2896 dev->real_num_tx_queues = txq; 2897 2898 if (disabling) { 2899 synchronize_net(); 2900 qdisc_reset_all_tx_gt(dev, txq); 2901 #ifdef CONFIG_XPS 2902 netif_reset_xps_queues_gt(dev, txq); 2903 #endif 2904 } 2905 } else { 2906 dev->real_num_tx_queues = txq; 2907 } 2908 2909 return 0; 2910 } 2911 EXPORT_SYMBOL(netif_set_real_num_tx_queues); 2912 2913 #ifdef CONFIG_SYSFS 2914 /** 2915 * netif_set_real_num_rx_queues - set actual number of RX queues used 2916 * @dev: Network device 2917 * @rxq: Actual number of RX queues 2918 * 2919 * This must be called either with the rtnl_lock held or before 2920 * registration of the net device. Returns 0 on success, or a 2921 * negative error code. If called before registration, it always 2922 * succeeds. 2923 */ 2924 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) 2925 { 2926 int rc; 2927 2928 if (rxq < 1 || rxq > dev->num_rx_queues) 2929 return -EINVAL; 2930 2931 if (dev->reg_state == NETREG_REGISTERED) { 2932 ASSERT_RTNL(); 2933 2934 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, 2935 rxq); 2936 if (rc) 2937 return rc; 2938 } 2939 2940 dev->real_num_rx_queues = rxq; 2941 return 0; 2942 } 2943 EXPORT_SYMBOL(netif_set_real_num_rx_queues); 2944 #endif 2945 2946 /** 2947 * netif_set_real_num_queues - set actual number of RX and TX queues used 2948 * @dev: Network device 2949 * @txq: Actual number of TX queues 2950 * @rxq: Actual number of RX queues 2951 * 2952 * Set the real number of both TX and RX queues. 2953 * Does nothing if the number of queues is already correct. 2954 */ 2955 int netif_set_real_num_queues(struct net_device *dev, 2956 unsigned int txq, unsigned int rxq) 2957 { 2958 unsigned int old_rxq = dev->real_num_rx_queues; 2959 int err; 2960 2961 if (txq < 1 || txq > dev->num_tx_queues || 2962 rxq < 1 || rxq > dev->num_rx_queues) 2963 return -EINVAL; 2964 2965 /* Start from increases, so the error path only does decreases - 2966 * decreases can't fail. 2967 */ 2968 if (rxq > dev->real_num_rx_queues) { 2969 err = netif_set_real_num_rx_queues(dev, rxq); 2970 if (err) 2971 return err; 2972 } 2973 if (txq > dev->real_num_tx_queues) { 2974 err = netif_set_real_num_tx_queues(dev, txq); 2975 if (err) 2976 goto undo_rx; 2977 } 2978 if (rxq < dev->real_num_rx_queues) 2979 WARN_ON(netif_set_real_num_rx_queues(dev, rxq)); 2980 if (txq < dev->real_num_tx_queues) 2981 WARN_ON(netif_set_real_num_tx_queues(dev, txq)); 2982 2983 return 0; 2984 undo_rx: 2985 WARN_ON(netif_set_real_num_rx_queues(dev, old_rxq)); 2986 return err; 2987 } 2988 EXPORT_SYMBOL(netif_set_real_num_queues); 2989 2990 /** 2991 * netif_get_num_default_rss_queues - default number of RSS queues 2992 * 2993 * Default value is the number of physical cores if there are only 1 or 2, or 2994 * divided by 2 if there are more. 2995 */ 2996 int netif_get_num_default_rss_queues(void) 2997 { 2998 cpumask_var_t cpus; 2999 int cpu, count = 0; 3000 3001 if (unlikely(is_kdump_kernel() || !zalloc_cpumask_var(&cpus, GFP_KERNEL))) 3002 return 1; 3003 3004 cpumask_copy(cpus, cpu_online_mask); 3005 for_each_cpu(cpu, cpus) { 3006 ++count; 3007 cpumask_andnot(cpus, cpus, topology_sibling_cpumask(cpu)); 3008 } 3009 free_cpumask_var(cpus); 3010 3011 return count > 2 ? DIV_ROUND_UP(count, 2) : count; 3012 } 3013 EXPORT_SYMBOL(netif_get_num_default_rss_queues); 3014 3015 static void __netif_reschedule(struct Qdisc *q) 3016 { 3017 struct softnet_data *sd; 3018 unsigned long flags; 3019 3020 local_irq_save(flags); 3021 sd = this_cpu_ptr(&softnet_data); 3022 q->next_sched = NULL; 3023 *sd->output_queue_tailp = q; 3024 sd->output_queue_tailp = &q->next_sched; 3025 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3026 local_irq_restore(flags); 3027 } 3028 3029 void __netif_schedule(struct Qdisc *q) 3030 { 3031 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) 3032 __netif_reschedule(q); 3033 } 3034 EXPORT_SYMBOL(__netif_schedule); 3035 3036 struct dev_kfree_skb_cb { 3037 enum skb_free_reason reason; 3038 }; 3039 3040 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) 3041 { 3042 return (struct dev_kfree_skb_cb *)skb->cb; 3043 } 3044 3045 void netif_schedule_queue(struct netdev_queue *txq) 3046 { 3047 rcu_read_lock(); 3048 if (!netif_xmit_stopped(txq)) { 3049 struct Qdisc *q = rcu_dereference(txq->qdisc); 3050 3051 __netif_schedule(q); 3052 } 3053 rcu_read_unlock(); 3054 } 3055 EXPORT_SYMBOL(netif_schedule_queue); 3056 3057 void netif_tx_wake_queue(struct netdev_queue *dev_queue) 3058 { 3059 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { 3060 struct Qdisc *q; 3061 3062 rcu_read_lock(); 3063 q = rcu_dereference(dev_queue->qdisc); 3064 __netif_schedule(q); 3065 rcu_read_unlock(); 3066 } 3067 } 3068 EXPORT_SYMBOL(netif_tx_wake_queue); 3069 3070 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason) 3071 { 3072 unsigned long flags; 3073 3074 if (unlikely(!skb)) 3075 return; 3076 3077 if (likely(refcount_read(&skb->users) == 1)) { 3078 smp_rmb(); 3079 refcount_set(&skb->users, 0); 3080 } else if (likely(!refcount_dec_and_test(&skb->users))) { 3081 return; 3082 } 3083 get_kfree_skb_cb(skb)->reason = reason; 3084 local_irq_save(flags); 3085 skb->next = __this_cpu_read(softnet_data.completion_queue); 3086 __this_cpu_write(softnet_data.completion_queue, skb); 3087 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3088 local_irq_restore(flags); 3089 } 3090 EXPORT_SYMBOL(__dev_kfree_skb_irq); 3091 3092 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason) 3093 { 3094 if (in_hardirq() || irqs_disabled()) 3095 __dev_kfree_skb_irq(skb, reason); 3096 else 3097 dev_kfree_skb(skb); 3098 } 3099 EXPORT_SYMBOL(__dev_kfree_skb_any); 3100 3101 3102 /** 3103 * netif_device_detach - mark device as removed 3104 * @dev: network device 3105 * 3106 * Mark device as removed from system and therefore no longer available. 3107 */ 3108 void netif_device_detach(struct net_device *dev) 3109 { 3110 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && 3111 netif_running(dev)) { 3112 netif_tx_stop_all_queues(dev); 3113 } 3114 } 3115 EXPORT_SYMBOL(netif_device_detach); 3116 3117 /** 3118 * netif_device_attach - mark device as attached 3119 * @dev: network device 3120 * 3121 * Mark device as attached from system and restart if needed. 3122 */ 3123 void netif_device_attach(struct net_device *dev) 3124 { 3125 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && 3126 netif_running(dev)) { 3127 netif_tx_wake_all_queues(dev); 3128 __netdev_watchdog_up(dev); 3129 } 3130 } 3131 EXPORT_SYMBOL(netif_device_attach); 3132 3133 /* 3134 * Returns a Tx hash based on the given packet descriptor a Tx queues' number 3135 * to be used as a distribution range. 3136 */ 3137 static u16 skb_tx_hash(const struct net_device *dev, 3138 const struct net_device *sb_dev, 3139 struct sk_buff *skb) 3140 { 3141 u32 hash; 3142 u16 qoffset = 0; 3143 u16 qcount = dev->real_num_tx_queues; 3144 3145 if (dev->num_tc) { 3146 u8 tc = netdev_get_prio_tc_map(dev, skb->priority); 3147 3148 qoffset = sb_dev->tc_to_txq[tc].offset; 3149 qcount = sb_dev->tc_to_txq[tc].count; 3150 if (unlikely(!qcount)) { 3151 net_warn_ratelimited("%s: invalid qcount, qoffset %u for tc %u\n", 3152 sb_dev->name, qoffset, tc); 3153 qoffset = 0; 3154 qcount = dev->real_num_tx_queues; 3155 } 3156 } 3157 3158 if (skb_rx_queue_recorded(skb)) { 3159 hash = skb_get_rx_queue(skb); 3160 if (hash >= qoffset) 3161 hash -= qoffset; 3162 while (unlikely(hash >= qcount)) 3163 hash -= qcount; 3164 return hash + qoffset; 3165 } 3166 3167 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; 3168 } 3169 3170 static void skb_warn_bad_offload(const struct sk_buff *skb) 3171 { 3172 static const netdev_features_t null_features; 3173 struct net_device *dev = skb->dev; 3174 const char *name = ""; 3175 3176 if (!net_ratelimit()) 3177 return; 3178 3179 if (dev) { 3180 if (dev->dev.parent) 3181 name = dev_driver_string(dev->dev.parent); 3182 else 3183 name = netdev_name(dev); 3184 } 3185 skb_dump(KERN_WARNING, skb, false); 3186 WARN(1, "%s: caps=(%pNF, %pNF)\n", 3187 name, dev ? &dev->features : &null_features, 3188 skb->sk ? &skb->sk->sk_route_caps : &null_features); 3189 } 3190 3191 /* 3192 * Invalidate hardware checksum when packet is to be mangled, and 3193 * complete checksum manually on outgoing path. 3194 */ 3195 int skb_checksum_help(struct sk_buff *skb) 3196 { 3197 __wsum csum; 3198 int ret = 0, offset; 3199 3200 if (skb->ip_summed == CHECKSUM_COMPLETE) 3201 goto out_set_summed; 3202 3203 if (unlikely(skb_is_gso(skb))) { 3204 skb_warn_bad_offload(skb); 3205 return -EINVAL; 3206 } 3207 3208 /* Before computing a checksum, we should make sure no frag could 3209 * be modified by an external entity : checksum could be wrong. 3210 */ 3211 if (skb_has_shared_frag(skb)) { 3212 ret = __skb_linearize(skb); 3213 if (ret) 3214 goto out; 3215 } 3216 3217 offset = skb_checksum_start_offset(skb); 3218 BUG_ON(offset >= skb_headlen(skb)); 3219 csum = skb_checksum(skb, offset, skb->len - offset, 0); 3220 3221 offset += skb->csum_offset; 3222 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); 3223 3224 ret = skb_ensure_writable(skb, offset + sizeof(__sum16)); 3225 if (ret) 3226 goto out; 3227 3228 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0; 3229 out_set_summed: 3230 skb->ip_summed = CHECKSUM_NONE; 3231 out: 3232 return ret; 3233 } 3234 EXPORT_SYMBOL(skb_checksum_help); 3235 3236 int skb_crc32c_csum_help(struct sk_buff *skb) 3237 { 3238 __le32 crc32c_csum; 3239 int ret = 0, offset, start; 3240 3241 if (skb->ip_summed != CHECKSUM_PARTIAL) 3242 goto out; 3243 3244 if (unlikely(skb_is_gso(skb))) 3245 goto out; 3246 3247 /* Before computing a checksum, we should make sure no frag could 3248 * be modified by an external entity : checksum could be wrong. 3249 */ 3250 if (unlikely(skb_has_shared_frag(skb))) { 3251 ret = __skb_linearize(skb); 3252 if (ret) 3253 goto out; 3254 } 3255 start = skb_checksum_start_offset(skb); 3256 offset = start + offsetof(struct sctphdr, checksum); 3257 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { 3258 ret = -EINVAL; 3259 goto out; 3260 } 3261 3262 ret = skb_ensure_writable(skb, offset + sizeof(__le32)); 3263 if (ret) 3264 goto out; 3265 3266 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start, 3267 skb->len - start, ~(__u32)0, 3268 crc32c_csum_stub)); 3269 *(__le32 *)(skb->data + offset) = crc32c_csum; 3270 skb->ip_summed = CHECKSUM_NONE; 3271 skb->csum_not_inet = 0; 3272 out: 3273 return ret; 3274 } 3275 3276 __be16 skb_network_protocol(struct sk_buff *skb, int *depth) 3277 { 3278 __be16 type = skb->protocol; 3279 3280 /* Tunnel gso handlers can set protocol to ethernet. */ 3281 if (type == htons(ETH_P_TEB)) { 3282 struct ethhdr *eth; 3283 3284 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) 3285 return 0; 3286 3287 eth = (struct ethhdr *)skb->data; 3288 type = eth->h_proto; 3289 } 3290 3291 return __vlan_get_protocol(skb, type, depth); 3292 } 3293 3294 /* openvswitch calls this on rx path, so we need a different check. 3295 */ 3296 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path) 3297 { 3298 if (tx_path) 3299 return skb->ip_summed != CHECKSUM_PARTIAL && 3300 skb->ip_summed != CHECKSUM_UNNECESSARY; 3301 3302 return skb->ip_summed == CHECKSUM_NONE; 3303 } 3304 3305 /** 3306 * __skb_gso_segment - Perform segmentation on skb. 3307 * @skb: buffer to segment 3308 * @features: features for the output path (see dev->features) 3309 * @tx_path: whether it is called in TX path 3310 * 3311 * This function segments the given skb and returns a list of segments. 3312 * 3313 * It may return NULL if the skb requires no segmentation. This is 3314 * only possible when GSO is used for verifying header integrity. 3315 * 3316 * Segmentation preserves SKB_GSO_CB_OFFSET bytes of previous skb cb. 3317 */ 3318 struct sk_buff *__skb_gso_segment(struct sk_buff *skb, 3319 netdev_features_t features, bool tx_path) 3320 { 3321 struct sk_buff *segs; 3322 3323 if (unlikely(skb_needs_check(skb, tx_path))) { 3324 int err; 3325 3326 /* We're going to init ->check field in TCP or UDP header */ 3327 err = skb_cow_head(skb, 0); 3328 if (err < 0) 3329 return ERR_PTR(err); 3330 } 3331 3332 /* Only report GSO partial support if it will enable us to 3333 * support segmentation on this frame without needing additional 3334 * work. 3335 */ 3336 if (features & NETIF_F_GSO_PARTIAL) { 3337 netdev_features_t partial_features = NETIF_F_GSO_ROBUST; 3338 struct net_device *dev = skb->dev; 3339 3340 partial_features |= dev->features & dev->gso_partial_features; 3341 if (!skb_gso_ok(skb, features | partial_features)) 3342 features &= ~NETIF_F_GSO_PARTIAL; 3343 } 3344 3345 BUILD_BUG_ON(SKB_GSO_CB_OFFSET + 3346 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb)); 3347 3348 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb); 3349 SKB_GSO_CB(skb)->encap_level = 0; 3350 3351 skb_reset_mac_header(skb); 3352 skb_reset_mac_len(skb); 3353 3354 segs = skb_mac_gso_segment(skb, features); 3355 3356 if (segs != skb && unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs))) 3357 skb_warn_bad_offload(skb); 3358 3359 return segs; 3360 } 3361 EXPORT_SYMBOL(__skb_gso_segment); 3362 3363 /* Take action when hardware reception checksum errors are detected. */ 3364 #ifdef CONFIG_BUG 3365 static void do_netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3366 { 3367 netdev_err(dev, "hw csum failure\n"); 3368 skb_dump(KERN_ERR, skb, true); 3369 dump_stack(); 3370 } 3371 3372 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3373 { 3374 DO_ONCE_LITE(do_netdev_rx_csum_fault, dev, skb); 3375 } 3376 EXPORT_SYMBOL(netdev_rx_csum_fault); 3377 #endif 3378 3379 /* XXX: check that highmem exists at all on the given machine. */ 3380 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) 3381 { 3382 #ifdef CONFIG_HIGHMEM 3383 int i; 3384 3385 if (!(dev->features & NETIF_F_HIGHDMA)) { 3386 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 3387 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 3388 3389 if (PageHighMem(skb_frag_page(frag))) 3390 return 1; 3391 } 3392 } 3393 #endif 3394 return 0; 3395 } 3396 3397 /* If MPLS offload request, verify we are testing hardware MPLS features 3398 * instead of standard features for the netdev. 3399 */ 3400 #if IS_ENABLED(CONFIG_NET_MPLS_GSO) 3401 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3402 netdev_features_t features, 3403 __be16 type) 3404 { 3405 if (eth_p_mpls(type)) 3406 features &= skb->dev->mpls_features; 3407 3408 return features; 3409 } 3410 #else 3411 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3412 netdev_features_t features, 3413 __be16 type) 3414 { 3415 return features; 3416 } 3417 #endif 3418 3419 static netdev_features_t harmonize_features(struct sk_buff *skb, 3420 netdev_features_t features) 3421 { 3422 __be16 type; 3423 3424 type = skb_network_protocol(skb, NULL); 3425 features = net_mpls_features(skb, features, type); 3426 3427 if (skb->ip_summed != CHECKSUM_NONE && 3428 !can_checksum_protocol(features, type)) { 3429 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); 3430 } 3431 if (illegal_highdma(skb->dev, skb)) 3432 features &= ~NETIF_F_SG; 3433 3434 return features; 3435 } 3436 3437 netdev_features_t passthru_features_check(struct sk_buff *skb, 3438 struct net_device *dev, 3439 netdev_features_t features) 3440 { 3441 return features; 3442 } 3443 EXPORT_SYMBOL(passthru_features_check); 3444 3445 static netdev_features_t dflt_features_check(struct sk_buff *skb, 3446 struct net_device *dev, 3447 netdev_features_t features) 3448 { 3449 return vlan_features_check(skb, features); 3450 } 3451 3452 static netdev_features_t gso_features_check(const struct sk_buff *skb, 3453 struct net_device *dev, 3454 netdev_features_t features) 3455 { 3456 u16 gso_segs = skb_shinfo(skb)->gso_segs; 3457 3458 if (gso_segs > READ_ONCE(dev->gso_max_segs)) 3459 return features & ~NETIF_F_GSO_MASK; 3460 3461 if (!skb_shinfo(skb)->gso_type) { 3462 skb_warn_bad_offload(skb); 3463 return features & ~NETIF_F_GSO_MASK; 3464 } 3465 3466 /* Support for GSO partial features requires software 3467 * intervention before we can actually process the packets 3468 * so we need to strip support for any partial features now 3469 * and we can pull them back in after we have partially 3470 * segmented the frame. 3471 */ 3472 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)) 3473 features &= ~dev->gso_partial_features; 3474 3475 /* Make sure to clear the IPv4 ID mangling feature if the 3476 * IPv4 header has the potential to be fragmented. 3477 */ 3478 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { 3479 struct iphdr *iph = skb->encapsulation ? 3480 inner_ip_hdr(skb) : ip_hdr(skb); 3481 3482 if (!(iph->frag_off & htons(IP_DF))) 3483 features &= ~NETIF_F_TSO_MANGLEID; 3484 } 3485 3486 return features; 3487 } 3488 3489 netdev_features_t netif_skb_features(struct sk_buff *skb) 3490 { 3491 struct net_device *dev = skb->dev; 3492 netdev_features_t features = dev->features; 3493 3494 if (skb_is_gso(skb)) 3495 features = gso_features_check(skb, dev, features); 3496 3497 /* If encapsulation offload request, verify we are testing 3498 * hardware encapsulation features instead of standard 3499 * features for the netdev 3500 */ 3501 if (skb->encapsulation) 3502 features &= dev->hw_enc_features; 3503 3504 if (skb_vlan_tagged(skb)) 3505 features = netdev_intersect_features(features, 3506 dev->vlan_features | 3507 NETIF_F_HW_VLAN_CTAG_TX | 3508 NETIF_F_HW_VLAN_STAG_TX); 3509 3510 if (dev->netdev_ops->ndo_features_check) 3511 features &= dev->netdev_ops->ndo_features_check(skb, dev, 3512 features); 3513 else 3514 features &= dflt_features_check(skb, dev, features); 3515 3516 return harmonize_features(skb, features); 3517 } 3518 EXPORT_SYMBOL(netif_skb_features); 3519 3520 static int xmit_one(struct sk_buff *skb, struct net_device *dev, 3521 struct netdev_queue *txq, bool more) 3522 { 3523 unsigned int len; 3524 int rc; 3525 3526 if (dev_nit_active(dev)) 3527 dev_queue_xmit_nit(skb, dev); 3528 3529 len = skb->len; 3530 PRANDOM_ADD_NOISE(skb, dev, txq, len + jiffies); 3531 trace_net_dev_start_xmit(skb, dev); 3532 rc = netdev_start_xmit(skb, dev, txq, more); 3533 trace_net_dev_xmit(skb, rc, dev, len); 3534 3535 return rc; 3536 } 3537 3538 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, 3539 struct netdev_queue *txq, int *ret) 3540 { 3541 struct sk_buff *skb = first; 3542 int rc = NETDEV_TX_OK; 3543 3544 while (skb) { 3545 struct sk_buff *next = skb->next; 3546 3547 skb_mark_not_on_list(skb); 3548 rc = xmit_one(skb, dev, txq, next != NULL); 3549 if (unlikely(!dev_xmit_complete(rc))) { 3550 skb->next = next; 3551 goto out; 3552 } 3553 3554 skb = next; 3555 if (netif_tx_queue_stopped(txq) && skb) { 3556 rc = NETDEV_TX_BUSY; 3557 break; 3558 } 3559 } 3560 3561 out: 3562 *ret = rc; 3563 return skb; 3564 } 3565 3566 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, 3567 netdev_features_t features) 3568 { 3569 if (skb_vlan_tag_present(skb) && 3570 !vlan_hw_offload_capable(features, skb->vlan_proto)) 3571 skb = __vlan_hwaccel_push_inside(skb); 3572 return skb; 3573 } 3574 3575 int skb_csum_hwoffload_help(struct sk_buff *skb, 3576 const netdev_features_t features) 3577 { 3578 if (unlikely(skb_csum_is_sctp(skb))) 3579 return !!(features & NETIF_F_SCTP_CRC) ? 0 : 3580 skb_crc32c_csum_help(skb); 3581 3582 if (features & NETIF_F_HW_CSUM) 3583 return 0; 3584 3585 if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) { 3586 switch (skb->csum_offset) { 3587 case offsetof(struct tcphdr, check): 3588 case offsetof(struct udphdr, check): 3589 return 0; 3590 } 3591 } 3592 3593 return skb_checksum_help(skb); 3594 } 3595 EXPORT_SYMBOL(skb_csum_hwoffload_help); 3596 3597 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again) 3598 { 3599 netdev_features_t features; 3600 3601 features = netif_skb_features(skb); 3602 skb = validate_xmit_vlan(skb, features); 3603 if (unlikely(!skb)) 3604 goto out_null; 3605 3606 skb = sk_validate_xmit_skb(skb, dev); 3607 if (unlikely(!skb)) 3608 goto out_null; 3609 3610 if (netif_needs_gso(skb, features)) { 3611 struct sk_buff *segs; 3612 3613 segs = skb_gso_segment(skb, features); 3614 if (IS_ERR(segs)) { 3615 goto out_kfree_skb; 3616 } else if (segs) { 3617 consume_skb(skb); 3618 skb = segs; 3619 } 3620 } else { 3621 if (skb_needs_linearize(skb, features) && 3622 __skb_linearize(skb)) 3623 goto out_kfree_skb; 3624 3625 /* If packet is not checksummed and device does not 3626 * support checksumming for this protocol, complete 3627 * checksumming here. 3628 */ 3629 if (skb->ip_summed == CHECKSUM_PARTIAL) { 3630 if (skb->encapsulation) 3631 skb_set_inner_transport_header(skb, 3632 skb_checksum_start_offset(skb)); 3633 else 3634 skb_set_transport_header(skb, 3635 skb_checksum_start_offset(skb)); 3636 if (skb_csum_hwoffload_help(skb, features)) 3637 goto out_kfree_skb; 3638 } 3639 } 3640 3641 skb = validate_xmit_xfrm(skb, features, again); 3642 3643 return skb; 3644 3645 out_kfree_skb: 3646 kfree_skb(skb); 3647 out_null: 3648 dev_core_stats_tx_dropped_inc(dev); 3649 return NULL; 3650 } 3651 3652 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again) 3653 { 3654 struct sk_buff *next, *head = NULL, *tail; 3655 3656 for (; skb != NULL; skb = next) { 3657 next = skb->next; 3658 skb_mark_not_on_list(skb); 3659 3660 /* in case skb wont be segmented, point to itself */ 3661 skb->prev = skb; 3662 3663 skb = validate_xmit_skb(skb, dev, again); 3664 if (!skb) 3665 continue; 3666 3667 if (!head) 3668 head = skb; 3669 else 3670 tail->next = skb; 3671 /* If skb was segmented, skb->prev points to 3672 * the last segment. If not, it still contains skb. 3673 */ 3674 tail = skb->prev; 3675 } 3676 return head; 3677 } 3678 EXPORT_SYMBOL_GPL(validate_xmit_skb_list); 3679 3680 static void qdisc_pkt_len_init(struct sk_buff *skb) 3681 { 3682 const struct skb_shared_info *shinfo = skb_shinfo(skb); 3683 3684 qdisc_skb_cb(skb)->pkt_len = skb->len; 3685 3686 /* To get more precise estimation of bytes sent on wire, 3687 * we add to pkt_len the headers size of all segments 3688 */ 3689 if (shinfo->gso_size && skb_transport_header_was_set(skb)) { 3690 unsigned int hdr_len; 3691 u16 gso_segs = shinfo->gso_segs; 3692 3693 /* mac layer + network layer */ 3694 hdr_len = skb_transport_header(skb) - skb_mac_header(skb); 3695 3696 /* + transport layer */ 3697 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { 3698 const struct tcphdr *th; 3699 struct tcphdr _tcphdr; 3700 3701 th = skb_header_pointer(skb, skb_transport_offset(skb), 3702 sizeof(_tcphdr), &_tcphdr); 3703 if (likely(th)) 3704 hdr_len += __tcp_hdrlen(th); 3705 } else { 3706 struct udphdr _udphdr; 3707 3708 if (skb_header_pointer(skb, skb_transport_offset(skb), 3709 sizeof(_udphdr), &_udphdr)) 3710 hdr_len += sizeof(struct udphdr); 3711 } 3712 3713 if (shinfo->gso_type & SKB_GSO_DODGY) 3714 gso_segs = DIV_ROUND_UP(skb->len - hdr_len, 3715 shinfo->gso_size); 3716 3717 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; 3718 } 3719 } 3720 3721 static int dev_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *q, 3722 struct sk_buff **to_free, 3723 struct netdev_queue *txq) 3724 { 3725 int rc; 3726 3727 rc = q->enqueue(skb, q, to_free) & NET_XMIT_MASK; 3728 if (rc == NET_XMIT_SUCCESS) 3729 trace_qdisc_enqueue(q, txq, skb); 3730 return rc; 3731 } 3732 3733 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, 3734 struct net_device *dev, 3735 struct netdev_queue *txq) 3736 { 3737 spinlock_t *root_lock = qdisc_lock(q); 3738 struct sk_buff *to_free = NULL; 3739 bool contended; 3740 int rc; 3741 3742 qdisc_calculate_pkt_len(skb, q); 3743 3744 if (q->flags & TCQ_F_NOLOCK) { 3745 if (q->flags & TCQ_F_CAN_BYPASS && nolock_qdisc_is_empty(q) && 3746 qdisc_run_begin(q)) { 3747 /* Retest nolock_qdisc_is_empty() within the protection 3748 * of q->seqlock to protect from racing with requeuing. 3749 */ 3750 if (unlikely(!nolock_qdisc_is_empty(q))) { 3751 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3752 __qdisc_run(q); 3753 qdisc_run_end(q); 3754 3755 goto no_lock_out; 3756 } 3757 3758 qdisc_bstats_cpu_update(q, skb); 3759 if (sch_direct_xmit(skb, q, dev, txq, NULL, true) && 3760 !nolock_qdisc_is_empty(q)) 3761 __qdisc_run(q); 3762 3763 qdisc_run_end(q); 3764 return NET_XMIT_SUCCESS; 3765 } 3766 3767 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3768 qdisc_run(q); 3769 3770 no_lock_out: 3771 if (unlikely(to_free)) 3772 kfree_skb_list_reason(to_free, 3773 SKB_DROP_REASON_QDISC_DROP); 3774 return rc; 3775 } 3776 3777 /* 3778 * Heuristic to force contended enqueues to serialize on a 3779 * separate lock before trying to get qdisc main lock. 3780 * This permits qdisc->running owner to get the lock more 3781 * often and dequeue packets faster. 3782 * On PREEMPT_RT it is possible to preempt the qdisc owner during xmit 3783 * and then other tasks will only enqueue packets. The packets will be 3784 * sent after the qdisc owner is scheduled again. To prevent this 3785 * scenario the task always serialize on the lock. 3786 */ 3787 contended = qdisc_is_running(q) || IS_ENABLED(CONFIG_PREEMPT_RT); 3788 if (unlikely(contended)) 3789 spin_lock(&q->busylock); 3790 3791 spin_lock(root_lock); 3792 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { 3793 __qdisc_drop(skb, &to_free); 3794 rc = NET_XMIT_DROP; 3795 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && 3796 qdisc_run_begin(q)) { 3797 /* 3798 * This is a work-conserving queue; there are no old skbs 3799 * waiting to be sent out; and the qdisc is not running - 3800 * xmit the skb directly. 3801 */ 3802 3803 qdisc_bstats_update(q, skb); 3804 3805 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { 3806 if (unlikely(contended)) { 3807 spin_unlock(&q->busylock); 3808 contended = false; 3809 } 3810 __qdisc_run(q); 3811 } 3812 3813 qdisc_run_end(q); 3814 rc = NET_XMIT_SUCCESS; 3815 } else { 3816 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3817 if (qdisc_run_begin(q)) { 3818 if (unlikely(contended)) { 3819 spin_unlock(&q->busylock); 3820 contended = false; 3821 } 3822 __qdisc_run(q); 3823 qdisc_run_end(q); 3824 } 3825 } 3826 spin_unlock(root_lock); 3827 if (unlikely(to_free)) 3828 kfree_skb_list_reason(to_free, SKB_DROP_REASON_QDISC_DROP); 3829 if (unlikely(contended)) 3830 spin_unlock(&q->busylock); 3831 return rc; 3832 } 3833 3834 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) 3835 static void skb_update_prio(struct sk_buff *skb) 3836 { 3837 const struct netprio_map *map; 3838 const struct sock *sk; 3839 unsigned int prioidx; 3840 3841 if (skb->priority) 3842 return; 3843 map = rcu_dereference_bh(skb->dev->priomap); 3844 if (!map) 3845 return; 3846 sk = skb_to_full_sk(skb); 3847 if (!sk) 3848 return; 3849 3850 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data); 3851 3852 if (prioidx < map->priomap_len) 3853 skb->priority = map->priomap[prioidx]; 3854 } 3855 #else 3856 #define skb_update_prio(skb) 3857 #endif 3858 3859 /** 3860 * dev_loopback_xmit - loop back @skb 3861 * @net: network namespace this loopback is happening in 3862 * @sk: sk needed to be a netfilter okfn 3863 * @skb: buffer to transmit 3864 */ 3865 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) 3866 { 3867 skb_reset_mac_header(skb); 3868 __skb_pull(skb, skb_network_offset(skb)); 3869 skb->pkt_type = PACKET_LOOPBACK; 3870 if (skb->ip_summed == CHECKSUM_NONE) 3871 skb->ip_summed = CHECKSUM_UNNECESSARY; 3872 WARN_ON(!skb_dst(skb)); 3873 skb_dst_force(skb); 3874 netif_rx(skb); 3875 return 0; 3876 } 3877 EXPORT_SYMBOL(dev_loopback_xmit); 3878 3879 #ifdef CONFIG_NET_EGRESS 3880 static struct sk_buff * 3881 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 3882 { 3883 #ifdef CONFIG_NET_CLS_ACT 3884 struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress); 3885 struct tcf_result cl_res; 3886 3887 if (!miniq) 3888 return skb; 3889 3890 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */ 3891 tc_skb_cb(skb)->mru = 0; 3892 tc_skb_cb(skb)->post_ct = false; 3893 mini_qdisc_bstats_cpu_update(miniq, skb); 3894 3895 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) { 3896 case TC_ACT_OK: 3897 case TC_ACT_RECLASSIFY: 3898 skb->tc_index = TC_H_MIN(cl_res.classid); 3899 break; 3900 case TC_ACT_SHOT: 3901 mini_qdisc_qstats_cpu_drop(miniq); 3902 *ret = NET_XMIT_DROP; 3903 kfree_skb_reason(skb, SKB_DROP_REASON_TC_EGRESS); 3904 return NULL; 3905 case TC_ACT_STOLEN: 3906 case TC_ACT_QUEUED: 3907 case TC_ACT_TRAP: 3908 *ret = NET_XMIT_SUCCESS; 3909 consume_skb(skb); 3910 return NULL; 3911 case TC_ACT_REDIRECT: 3912 /* No need to push/pop skb's mac_header here on egress! */ 3913 skb_do_redirect(skb); 3914 *ret = NET_XMIT_SUCCESS; 3915 return NULL; 3916 default: 3917 break; 3918 } 3919 #endif /* CONFIG_NET_CLS_ACT */ 3920 3921 return skb; 3922 } 3923 #endif /* CONFIG_NET_EGRESS */ 3924 3925 #ifdef CONFIG_XPS 3926 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb, 3927 struct xps_dev_maps *dev_maps, unsigned int tci) 3928 { 3929 int tc = netdev_get_prio_tc_map(dev, skb->priority); 3930 struct xps_map *map; 3931 int queue_index = -1; 3932 3933 if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids) 3934 return queue_index; 3935 3936 tci *= dev_maps->num_tc; 3937 tci += tc; 3938 3939 map = rcu_dereference(dev_maps->attr_map[tci]); 3940 if (map) { 3941 if (map->len == 1) 3942 queue_index = map->queues[0]; 3943 else 3944 queue_index = map->queues[reciprocal_scale( 3945 skb_get_hash(skb), map->len)]; 3946 if (unlikely(queue_index >= dev->real_num_tx_queues)) 3947 queue_index = -1; 3948 } 3949 return queue_index; 3950 } 3951 #endif 3952 3953 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev, 3954 struct sk_buff *skb) 3955 { 3956 #ifdef CONFIG_XPS 3957 struct xps_dev_maps *dev_maps; 3958 struct sock *sk = skb->sk; 3959 int queue_index = -1; 3960 3961 if (!static_key_false(&xps_needed)) 3962 return -1; 3963 3964 rcu_read_lock(); 3965 if (!static_key_false(&xps_rxqs_needed)) 3966 goto get_cpus_map; 3967 3968 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]); 3969 if (dev_maps) { 3970 int tci = sk_rx_queue_get(sk); 3971 3972 if (tci >= 0) 3973 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 3974 tci); 3975 } 3976 3977 get_cpus_map: 3978 if (queue_index < 0) { 3979 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]); 3980 if (dev_maps) { 3981 unsigned int tci = skb->sender_cpu - 1; 3982 3983 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 3984 tci); 3985 } 3986 } 3987 rcu_read_unlock(); 3988 3989 return queue_index; 3990 #else 3991 return -1; 3992 #endif 3993 } 3994 3995 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb, 3996 struct net_device *sb_dev) 3997 { 3998 return 0; 3999 } 4000 EXPORT_SYMBOL(dev_pick_tx_zero); 4001 4002 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb, 4003 struct net_device *sb_dev) 4004 { 4005 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues; 4006 } 4007 EXPORT_SYMBOL(dev_pick_tx_cpu_id); 4008 4009 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb, 4010 struct net_device *sb_dev) 4011 { 4012 struct sock *sk = skb->sk; 4013 int queue_index = sk_tx_queue_get(sk); 4014 4015 sb_dev = sb_dev ? : dev; 4016 4017 if (queue_index < 0 || skb->ooo_okay || 4018 queue_index >= dev->real_num_tx_queues) { 4019 int new_index = get_xps_queue(dev, sb_dev, skb); 4020 4021 if (new_index < 0) 4022 new_index = skb_tx_hash(dev, sb_dev, skb); 4023 4024 if (queue_index != new_index && sk && 4025 sk_fullsock(sk) && 4026 rcu_access_pointer(sk->sk_dst_cache)) 4027 sk_tx_queue_set(sk, new_index); 4028 4029 queue_index = new_index; 4030 } 4031 4032 return queue_index; 4033 } 4034 EXPORT_SYMBOL(netdev_pick_tx); 4035 4036 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev, 4037 struct sk_buff *skb, 4038 struct net_device *sb_dev) 4039 { 4040 int queue_index = 0; 4041 4042 #ifdef CONFIG_XPS 4043 u32 sender_cpu = skb->sender_cpu - 1; 4044 4045 if (sender_cpu >= (u32)NR_CPUS) 4046 skb->sender_cpu = raw_smp_processor_id() + 1; 4047 #endif 4048 4049 if (dev->real_num_tx_queues != 1) { 4050 const struct net_device_ops *ops = dev->netdev_ops; 4051 4052 if (ops->ndo_select_queue) 4053 queue_index = ops->ndo_select_queue(dev, skb, sb_dev); 4054 else 4055 queue_index = netdev_pick_tx(dev, skb, sb_dev); 4056 4057 queue_index = netdev_cap_txqueue(dev, queue_index); 4058 } 4059 4060 skb_set_queue_mapping(skb, queue_index); 4061 return netdev_get_tx_queue(dev, queue_index); 4062 } 4063 4064 /** 4065 * __dev_queue_xmit - transmit a buffer 4066 * @skb: buffer to transmit 4067 * @sb_dev: suboordinate device used for L2 forwarding offload 4068 * 4069 * Queue a buffer for transmission to a network device. The caller must 4070 * have set the device and priority and built the buffer before calling 4071 * this function. The function can be called from an interrupt. 4072 * 4073 * A negative errno code is returned on a failure. A success does not 4074 * guarantee the frame will be transmitted as it may be dropped due 4075 * to congestion or traffic shaping. 4076 * 4077 * ----------------------------------------------------------------------------------- 4078 * I notice this method can also return errors from the queue disciplines, 4079 * including NET_XMIT_DROP, which is a positive value. So, errors can also 4080 * be positive. 4081 * 4082 * Regardless of the return value, the skb is consumed, so it is currently 4083 * difficult to retry a send to this method. (You can bump the ref count 4084 * before sending to hold a reference for retry if you are careful.) 4085 * 4086 * When calling this method, interrupts MUST be enabled. This is because 4087 * the BH enable code must have IRQs enabled so that it will not deadlock. 4088 * --BLG 4089 */ 4090 static int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev) 4091 { 4092 struct net_device *dev = skb->dev; 4093 struct netdev_queue *txq; 4094 struct Qdisc *q; 4095 int rc = -ENOMEM; 4096 bool again = false; 4097 4098 skb_reset_mac_header(skb); 4099 4100 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) 4101 __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED); 4102 4103 /* Disable soft irqs for various locks below. Also 4104 * stops preemption for RCU. 4105 */ 4106 rcu_read_lock_bh(); 4107 4108 skb_update_prio(skb); 4109 4110 qdisc_pkt_len_init(skb); 4111 #ifdef CONFIG_NET_CLS_ACT 4112 skb->tc_at_ingress = 0; 4113 #endif 4114 #ifdef CONFIG_NET_EGRESS 4115 if (static_branch_unlikely(&egress_needed_key)) { 4116 if (nf_hook_egress_active()) { 4117 skb = nf_hook_egress(skb, &rc, dev); 4118 if (!skb) 4119 goto out; 4120 } 4121 nf_skip_egress(skb, true); 4122 skb = sch_handle_egress(skb, &rc, dev); 4123 if (!skb) 4124 goto out; 4125 nf_skip_egress(skb, false); 4126 } 4127 #endif 4128 /* If device/qdisc don't need skb->dst, release it right now while 4129 * its hot in this cpu cache. 4130 */ 4131 if (dev->priv_flags & IFF_XMIT_DST_RELEASE) 4132 skb_dst_drop(skb); 4133 else 4134 skb_dst_force(skb); 4135 4136 txq = netdev_core_pick_tx(dev, skb, sb_dev); 4137 q = rcu_dereference_bh(txq->qdisc); 4138 4139 trace_net_dev_queue(skb); 4140 if (q->enqueue) { 4141 rc = __dev_xmit_skb(skb, q, dev, txq); 4142 goto out; 4143 } 4144 4145 /* The device has no queue. Common case for software devices: 4146 * loopback, all the sorts of tunnels... 4147 4148 * Really, it is unlikely that netif_tx_lock protection is necessary 4149 * here. (f.e. loopback and IP tunnels are clean ignoring statistics 4150 * counters.) 4151 * However, it is possible, that they rely on protection 4152 * made by us here. 4153 4154 * Check this and shot the lock. It is not prone from deadlocks. 4155 *Either shot noqueue qdisc, it is even simpler 8) 4156 */ 4157 if (dev->flags & IFF_UP) { 4158 int cpu = smp_processor_id(); /* ok because BHs are off */ 4159 4160 /* Other cpus might concurrently change txq->xmit_lock_owner 4161 * to -1 or to their cpu id, but not to our id. 4162 */ 4163 if (READ_ONCE(txq->xmit_lock_owner) != cpu) { 4164 if (dev_xmit_recursion()) 4165 goto recursion_alert; 4166 4167 skb = validate_xmit_skb(skb, dev, &again); 4168 if (!skb) 4169 goto out; 4170 4171 PRANDOM_ADD_NOISE(skb, dev, txq, jiffies); 4172 HARD_TX_LOCK(dev, txq, cpu); 4173 4174 if (!netif_xmit_stopped(txq)) { 4175 dev_xmit_recursion_inc(); 4176 skb = dev_hard_start_xmit(skb, dev, txq, &rc); 4177 dev_xmit_recursion_dec(); 4178 if (dev_xmit_complete(rc)) { 4179 HARD_TX_UNLOCK(dev, txq); 4180 goto out; 4181 } 4182 } 4183 HARD_TX_UNLOCK(dev, txq); 4184 net_crit_ratelimited("Virtual device %s asks to queue packet!\n", 4185 dev->name); 4186 } else { 4187 /* Recursion is detected! It is possible, 4188 * unfortunately 4189 */ 4190 recursion_alert: 4191 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", 4192 dev->name); 4193 } 4194 } 4195 4196 rc = -ENETDOWN; 4197 rcu_read_unlock_bh(); 4198 4199 dev_core_stats_tx_dropped_inc(dev); 4200 kfree_skb_list(skb); 4201 return rc; 4202 out: 4203 rcu_read_unlock_bh(); 4204 return rc; 4205 } 4206 4207 int dev_queue_xmit(struct sk_buff *skb) 4208 { 4209 return __dev_queue_xmit(skb, NULL); 4210 } 4211 EXPORT_SYMBOL(dev_queue_xmit); 4212 4213 int dev_queue_xmit_accel(struct sk_buff *skb, struct net_device *sb_dev) 4214 { 4215 return __dev_queue_xmit(skb, sb_dev); 4216 } 4217 EXPORT_SYMBOL(dev_queue_xmit_accel); 4218 4219 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id) 4220 { 4221 struct net_device *dev = skb->dev; 4222 struct sk_buff *orig_skb = skb; 4223 struct netdev_queue *txq; 4224 int ret = NETDEV_TX_BUSY; 4225 bool again = false; 4226 4227 if (unlikely(!netif_running(dev) || 4228 !netif_carrier_ok(dev))) 4229 goto drop; 4230 4231 skb = validate_xmit_skb_list(skb, dev, &again); 4232 if (skb != orig_skb) 4233 goto drop; 4234 4235 skb_set_queue_mapping(skb, queue_id); 4236 txq = skb_get_tx_queue(dev, skb); 4237 PRANDOM_ADD_NOISE(skb, dev, txq, jiffies); 4238 4239 local_bh_disable(); 4240 4241 dev_xmit_recursion_inc(); 4242 HARD_TX_LOCK(dev, txq, smp_processor_id()); 4243 if (!netif_xmit_frozen_or_drv_stopped(txq)) 4244 ret = netdev_start_xmit(skb, dev, txq, false); 4245 HARD_TX_UNLOCK(dev, txq); 4246 dev_xmit_recursion_dec(); 4247 4248 local_bh_enable(); 4249 return ret; 4250 drop: 4251 dev_core_stats_tx_dropped_inc(dev); 4252 kfree_skb_list(skb); 4253 return NET_XMIT_DROP; 4254 } 4255 EXPORT_SYMBOL(__dev_direct_xmit); 4256 4257 /************************************************************************* 4258 * Receiver routines 4259 *************************************************************************/ 4260 4261 int netdev_max_backlog __read_mostly = 1000; 4262 EXPORT_SYMBOL(netdev_max_backlog); 4263 4264 int netdev_tstamp_prequeue __read_mostly = 1; 4265 int netdev_budget __read_mostly = 300; 4266 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */ 4267 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ; 4268 int weight_p __read_mostly = 64; /* old backlog weight */ 4269 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ 4270 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ 4271 int dev_rx_weight __read_mostly = 64; 4272 int dev_tx_weight __read_mostly = 64; 4273 4274 /* Called with irq disabled */ 4275 static inline void ____napi_schedule(struct softnet_data *sd, 4276 struct napi_struct *napi) 4277 { 4278 struct task_struct *thread; 4279 4280 lockdep_assert_irqs_disabled(); 4281 4282 if (test_bit(NAPI_STATE_THREADED, &napi->state)) { 4283 /* Paired with smp_mb__before_atomic() in 4284 * napi_enable()/dev_set_threaded(). 4285 * Use READ_ONCE() to guarantee a complete 4286 * read on napi->thread. Only call 4287 * wake_up_process() when it's not NULL. 4288 */ 4289 thread = READ_ONCE(napi->thread); 4290 if (thread) { 4291 /* Avoid doing set_bit() if the thread is in 4292 * INTERRUPTIBLE state, cause napi_thread_wait() 4293 * makes sure to proceed with napi polling 4294 * if the thread is explicitly woken from here. 4295 */ 4296 if (READ_ONCE(thread->__state) != TASK_INTERRUPTIBLE) 4297 set_bit(NAPI_STATE_SCHED_THREADED, &napi->state); 4298 wake_up_process(thread); 4299 return; 4300 } 4301 } 4302 4303 list_add_tail(&napi->poll_list, &sd->poll_list); 4304 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4305 } 4306 4307 #ifdef CONFIG_RPS 4308 4309 /* One global table that all flow-based protocols share. */ 4310 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; 4311 EXPORT_SYMBOL(rps_sock_flow_table); 4312 u32 rps_cpu_mask __read_mostly; 4313 EXPORT_SYMBOL(rps_cpu_mask); 4314 4315 struct static_key_false rps_needed __read_mostly; 4316 EXPORT_SYMBOL(rps_needed); 4317 struct static_key_false rfs_needed __read_mostly; 4318 EXPORT_SYMBOL(rfs_needed); 4319 4320 static struct rps_dev_flow * 4321 set_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4322 struct rps_dev_flow *rflow, u16 next_cpu) 4323 { 4324 if (next_cpu < nr_cpu_ids) { 4325 #ifdef CONFIG_RFS_ACCEL 4326 struct netdev_rx_queue *rxqueue; 4327 struct rps_dev_flow_table *flow_table; 4328 struct rps_dev_flow *old_rflow; 4329 u32 flow_id; 4330 u16 rxq_index; 4331 int rc; 4332 4333 /* Should we steer this flow to a different hardware queue? */ 4334 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || 4335 !(dev->features & NETIF_F_NTUPLE)) 4336 goto out; 4337 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); 4338 if (rxq_index == skb_get_rx_queue(skb)) 4339 goto out; 4340 4341 rxqueue = dev->_rx + rxq_index; 4342 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4343 if (!flow_table) 4344 goto out; 4345 flow_id = skb_get_hash(skb) & flow_table->mask; 4346 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, 4347 rxq_index, flow_id); 4348 if (rc < 0) 4349 goto out; 4350 old_rflow = rflow; 4351 rflow = &flow_table->flows[flow_id]; 4352 rflow->filter = rc; 4353 if (old_rflow->filter == rflow->filter) 4354 old_rflow->filter = RPS_NO_FILTER; 4355 out: 4356 #endif 4357 rflow->last_qtail = 4358 per_cpu(softnet_data, next_cpu).input_queue_head; 4359 } 4360 4361 rflow->cpu = next_cpu; 4362 return rflow; 4363 } 4364 4365 /* 4366 * get_rps_cpu is called from netif_receive_skb and returns the target 4367 * CPU from the RPS map of the receiving queue for a given skb. 4368 * rcu_read_lock must be held on entry. 4369 */ 4370 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4371 struct rps_dev_flow **rflowp) 4372 { 4373 const struct rps_sock_flow_table *sock_flow_table; 4374 struct netdev_rx_queue *rxqueue = dev->_rx; 4375 struct rps_dev_flow_table *flow_table; 4376 struct rps_map *map; 4377 int cpu = -1; 4378 u32 tcpu; 4379 u32 hash; 4380 4381 if (skb_rx_queue_recorded(skb)) { 4382 u16 index = skb_get_rx_queue(skb); 4383 4384 if (unlikely(index >= dev->real_num_rx_queues)) { 4385 WARN_ONCE(dev->real_num_rx_queues > 1, 4386 "%s received packet on queue %u, but number " 4387 "of RX queues is %u\n", 4388 dev->name, index, dev->real_num_rx_queues); 4389 goto done; 4390 } 4391 rxqueue += index; 4392 } 4393 4394 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ 4395 4396 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4397 map = rcu_dereference(rxqueue->rps_map); 4398 if (!flow_table && !map) 4399 goto done; 4400 4401 skb_reset_network_header(skb); 4402 hash = skb_get_hash(skb); 4403 if (!hash) 4404 goto done; 4405 4406 sock_flow_table = rcu_dereference(rps_sock_flow_table); 4407 if (flow_table && sock_flow_table) { 4408 struct rps_dev_flow *rflow; 4409 u32 next_cpu; 4410 u32 ident; 4411 4412 /* First check into global flow table if there is a match */ 4413 ident = sock_flow_table->ents[hash & sock_flow_table->mask]; 4414 if ((ident ^ hash) & ~rps_cpu_mask) 4415 goto try_rps; 4416 4417 next_cpu = ident & rps_cpu_mask; 4418 4419 /* OK, now we know there is a match, 4420 * we can look at the local (per receive queue) flow table 4421 */ 4422 rflow = &flow_table->flows[hash & flow_table->mask]; 4423 tcpu = rflow->cpu; 4424 4425 /* 4426 * If the desired CPU (where last recvmsg was done) is 4427 * different from current CPU (one in the rx-queue flow 4428 * table entry), switch if one of the following holds: 4429 * - Current CPU is unset (>= nr_cpu_ids). 4430 * - Current CPU is offline. 4431 * - The current CPU's queue tail has advanced beyond the 4432 * last packet that was enqueued using this table entry. 4433 * This guarantees that all previous packets for the flow 4434 * have been dequeued, thus preserving in order delivery. 4435 */ 4436 if (unlikely(tcpu != next_cpu) && 4437 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || 4438 ((int)(per_cpu(softnet_data, tcpu).input_queue_head - 4439 rflow->last_qtail)) >= 0)) { 4440 tcpu = next_cpu; 4441 rflow = set_rps_cpu(dev, skb, rflow, next_cpu); 4442 } 4443 4444 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { 4445 *rflowp = rflow; 4446 cpu = tcpu; 4447 goto done; 4448 } 4449 } 4450 4451 try_rps: 4452 4453 if (map) { 4454 tcpu = map->cpus[reciprocal_scale(hash, map->len)]; 4455 if (cpu_online(tcpu)) { 4456 cpu = tcpu; 4457 goto done; 4458 } 4459 } 4460 4461 done: 4462 return cpu; 4463 } 4464 4465 #ifdef CONFIG_RFS_ACCEL 4466 4467 /** 4468 * rps_may_expire_flow - check whether an RFS hardware filter may be removed 4469 * @dev: Device on which the filter was set 4470 * @rxq_index: RX queue index 4471 * @flow_id: Flow ID passed to ndo_rx_flow_steer() 4472 * @filter_id: Filter ID returned by ndo_rx_flow_steer() 4473 * 4474 * Drivers that implement ndo_rx_flow_steer() should periodically call 4475 * this function for each installed filter and remove the filters for 4476 * which it returns %true. 4477 */ 4478 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, 4479 u32 flow_id, u16 filter_id) 4480 { 4481 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; 4482 struct rps_dev_flow_table *flow_table; 4483 struct rps_dev_flow *rflow; 4484 bool expire = true; 4485 unsigned int cpu; 4486 4487 rcu_read_lock(); 4488 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4489 if (flow_table && flow_id <= flow_table->mask) { 4490 rflow = &flow_table->flows[flow_id]; 4491 cpu = READ_ONCE(rflow->cpu); 4492 if (rflow->filter == filter_id && cpu < nr_cpu_ids && 4493 ((int)(per_cpu(softnet_data, cpu).input_queue_head - 4494 rflow->last_qtail) < 4495 (int)(10 * flow_table->mask))) 4496 expire = false; 4497 } 4498 rcu_read_unlock(); 4499 return expire; 4500 } 4501 EXPORT_SYMBOL(rps_may_expire_flow); 4502 4503 #endif /* CONFIG_RFS_ACCEL */ 4504 4505 /* Called from hardirq (IPI) context */ 4506 static void rps_trigger_softirq(void *data) 4507 { 4508 struct softnet_data *sd = data; 4509 4510 ____napi_schedule(sd, &sd->backlog); 4511 sd->received_rps++; 4512 } 4513 4514 #endif /* CONFIG_RPS */ 4515 4516 /* 4517 * Check if this softnet_data structure is another cpu one 4518 * If yes, queue it to our IPI list and return 1 4519 * If no, return 0 4520 */ 4521 static int napi_schedule_rps(struct softnet_data *sd) 4522 { 4523 struct softnet_data *mysd = this_cpu_ptr(&softnet_data); 4524 4525 #ifdef CONFIG_RPS 4526 if (sd != mysd) { 4527 sd->rps_ipi_next = mysd->rps_ipi_list; 4528 mysd->rps_ipi_list = sd; 4529 4530 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4531 return 1; 4532 } 4533 #endif /* CONFIG_RPS */ 4534 __napi_schedule_irqoff(&mysd->backlog); 4535 return 0; 4536 } 4537 4538 #ifdef CONFIG_NET_FLOW_LIMIT 4539 int netdev_flow_limit_table_len __read_mostly = (1 << 12); 4540 #endif 4541 4542 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) 4543 { 4544 #ifdef CONFIG_NET_FLOW_LIMIT 4545 struct sd_flow_limit *fl; 4546 struct softnet_data *sd; 4547 unsigned int old_flow, new_flow; 4548 4549 if (qlen < (netdev_max_backlog >> 1)) 4550 return false; 4551 4552 sd = this_cpu_ptr(&softnet_data); 4553 4554 rcu_read_lock(); 4555 fl = rcu_dereference(sd->flow_limit); 4556 if (fl) { 4557 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); 4558 old_flow = fl->history[fl->history_head]; 4559 fl->history[fl->history_head] = new_flow; 4560 4561 fl->history_head++; 4562 fl->history_head &= FLOW_LIMIT_HISTORY - 1; 4563 4564 if (likely(fl->buckets[old_flow])) 4565 fl->buckets[old_flow]--; 4566 4567 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { 4568 fl->count++; 4569 rcu_read_unlock(); 4570 return true; 4571 } 4572 } 4573 rcu_read_unlock(); 4574 #endif 4575 return false; 4576 } 4577 4578 /* 4579 * enqueue_to_backlog is called to queue an skb to a per CPU backlog 4580 * queue (may be a remote CPU queue). 4581 */ 4582 static int enqueue_to_backlog(struct sk_buff *skb, int cpu, 4583 unsigned int *qtail) 4584 { 4585 enum skb_drop_reason reason; 4586 struct softnet_data *sd; 4587 unsigned long flags; 4588 unsigned int qlen; 4589 4590 reason = SKB_DROP_REASON_NOT_SPECIFIED; 4591 sd = &per_cpu(softnet_data, cpu); 4592 4593 rps_lock_irqsave(sd, &flags); 4594 if (!netif_running(skb->dev)) 4595 goto drop; 4596 qlen = skb_queue_len(&sd->input_pkt_queue); 4597 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) { 4598 if (qlen) { 4599 enqueue: 4600 __skb_queue_tail(&sd->input_pkt_queue, skb); 4601 input_queue_tail_incr_save(sd, qtail); 4602 rps_unlock_irq_restore(sd, &flags); 4603 return NET_RX_SUCCESS; 4604 } 4605 4606 /* Schedule NAPI for backlog device 4607 * We can use non atomic operation since we own the queue lock 4608 */ 4609 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) 4610 napi_schedule_rps(sd); 4611 goto enqueue; 4612 } 4613 reason = SKB_DROP_REASON_CPU_BACKLOG; 4614 4615 drop: 4616 sd->dropped++; 4617 rps_unlock_irq_restore(sd, &flags); 4618 4619 dev_core_stats_rx_dropped_inc(skb->dev); 4620 kfree_skb_reason(skb, reason); 4621 return NET_RX_DROP; 4622 } 4623 4624 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb) 4625 { 4626 struct net_device *dev = skb->dev; 4627 struct netdev_rx_queue *rxqueue; 4628 4629 rxqueue = dev->_rx; 4630 4631 if (skb_rx_queue_recorded(skb)) { 4632 u16 index = skb_get_rx_queue(skb); 4633 4634 if (unlikely(index >= dev->real_num_rx_queues)) { 4635 WARN_ONCE(dev->real_num_rx_queues > 1, 4636 "%s received packet on queue %u, but number " 4637 "of RX queues is %u\n", 4638 dev->name, index, dev->real_num_rx_queues); 4639 4640 return rxqueue; /* Return first rxqueue */ 4641 } 4642 rxqueue += index; 4643 } 4644 return rxqueue; 4645 } 4646 4647 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp, 4648 struct bpf_prog *xdp_prog) 4649 { 4650 void *orig_data, *orig_data_end, *hard_start; 4651 struct netdev_rx_queue *rxqueue; 4652 bool orig_bcast, orig_host; 4653 u32 mac_len, frame_sz; 4654 __be16 orig_eth_type; 4655 struct ethhdr *eth; 4656 u32 metalen, act; 4657 int off; 4658 4659 /* The XDP program wants to see the packet starting at the MAC 4660 * header. 4661 */ 4662 mac_len = skb->data - skb_mac_header(skb); 4663 hard_start = skb->data - skb_headroom(skb); 4664 4665 /* SKB "head" area always have tailroom for skb_shared_info */ 4666 frame_sz = (void *)skb_end_pointer(skb) - hard_start; 4667 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 4668 4669 rxqueue = netif_get_rxqueue(skb); 4670 xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq); 4671 xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len, 4672 skb_headlen(skb) + mac_len, true); 4673 4674 orig_data_end = xdp->data_end; 4675 orig_data = xdp->data; 4676 eth = (struct ethhdr *)xdp->data; 4677 orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr); 4678 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest); 4679 orig_eth_type = eth->h_proto; 4680 4681 act = bpf_prog_run_xdp(xdp_prog, xdp); 4682 4683 /* check if bpf_xdp_adjust_head was used */ 4684 off = xdp->data - orig_data; 4685 if (off) { 4686 if (off > 0) 4687 __skb_pull(skb, off); 4688 else if (off < 0) 4689 __skb_push(skb, -off); 4690 4691 skb->mac_header += off; 4692 skb_reset_network_header(skb); 4693 } 4694 4695 /* check if bpf_xdp_adjust_tail was used */ 4696 off = xdp->data_end - orig_data_end; 4697 if (off != 0) { 4698 skb_set_tail_pointer(skb, xdp->data_end - xdp->data); 4699 skb->len += off; /* positive on grow, negative on shrink */ 4700 } 4701 4702 /* check if XDP changed eth hdr such SKB needs update */ 4703 eth = (struct ethhdr *)xdp->data; 4704 if ((orig_eth_type != eth->h_proto) || 4705 (orig_host != ether_addr_equal_64bits(eth->h_dest, 4706 skb->dev->dev_addr)) || 4707 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) { 4708 __skb_push(skb, ETH_HLEN); 4709 skb->pkt_type = PACKET_HOST; 4710 skb->protocol = eth_type_trans(skb, skb->dev); 4711 } 4712 4713 /* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull 4714 * before calling us again on redirect path. We do not call do_redirect 4715 * as we leave that up to the caller. 4716 * 4717 * Caller is responsible for managing lifetime of skb (i.e. calling 4718 * kfree_skb in response to actions it cannot handle/XDP_DROP). 4719 */ 4720 switch (act) { 4721 case XDP_REDIRECT: 4722 case XDP_TX: 4723 __skb_push(skb, mac_len); 4724 break; 4725 case XDP_PASS: 4726 metalen = xdp->data - xdp->data_meta; 4727 if (metalen) 4728 skb_metadata_set(skb, metalen); 4729 break; 4730 } 4731 4732 return act; 4733 } 4734 4735 static u32 netif_receive_generic_xdp(struct sk_buff *skb, 4736 struct xdp_buff *xdp, 4737 struct bpf_prog *xdp_prog) 4738 { 4739 u32 act = XDP_DROP; 4740 4741 /* Reinjected packets coming from act_mirred or similar should 4742 * not get XDP generic processing. 4743 */ 4744 if (skb_is_redirected(skb)) 4745 return XDP_PASS; 4746 4747 /* XDP packets must be linear and must have sufficient headroom 4748 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also 4749 * native XDP provides, thus we need to do it here as well. 4750 */ 4751 if (skb_cloned(skb) || skb_is_nonlinear(skb) || 4752 skb_headroom(skb) < XDP_PACKET_HEADROOM) { 4753 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb); 4754 int troom = skb->tail + skb->data_len - skb->end; 4755 4756 /* In case we have to go down the path and also linearize, 4757 * then lets do the pskb_expand_head() work just once here. 4758 */ 4759 if (pskb_expand_head(skb, 4760 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0, 4761 troom > 0 ? troom + 128 : 0, GFP_ATOMIC)) 4762 goto do_drop; 4763 if (skb_linearize(skb)) 4764 goto do_drop; 4765 } 4766 4767 act = bpf_prog_run_generic_xdp(skb, xdp, xdp_prog); 4768 switch (act) { 4769 case XDP_REDIRECT: 4770 case XDP_TX: 4771 case XDP_PASS: 4772 break; 4773 default: 4774 bpf_warn_invalid_xdp_action(skb->dev, xdp_prog, act); 4775 fallthrough; 4776 case XDP_ABORTED: 4777 trace_xdp_exception(skb->dev, xdp_prog, act); 4778 fallthrough; 4779 case XDP_DROP: 4780 do_drop: 4781 kfree_skb(skb); 4782 break; 4783 } 4784 4785 return act; 4786 } 4787 4788 /* When doing generic XDP we have to bypass the qdisc layer and the 4789 * network taps in order to match in-driver-XDP behavior. 4790 */ 4791 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog) 4792 { 4793 struct net_device *dev = skb->dev; 4794 struct netdev_queue *txq; 4795 bool free_skb = true; 4796 int cpu, rc; 4797 4798 txq = netdev_core_pick_tx(dev, skb, NULL); 4799 cpu = smp_processor_id(); 4800 HARD_TX_LOCK(dev, txq, cpu); 4801 if (!netif_xmit_stopped(txq)) { 4802 rc = netdev_start_xmit(skb, dev, txq, 0); 4803 if (dev_xmit_complete(rc)) 4804 free_skb = false; 4805 } 4806 HARD_TX_UNLOCK(dev, txq); 4807 if (free_skb) { 4808 trace_xdp_exception(dev, xdp_prog, XDP_TX); 4809 kfree_skb(skb); 4810 } 4811 } 4812 4813 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key); 4814 4815 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb) 4816 { 4817 if (xdp_prog) { 4818 struct xdp_buff xdp; 4819 u32 act; 4820 int err; 4821 4822 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog); 4823 if (act != XDP_PASS) { 4824 switch (act) { 4825 case XDP_REDIRECT: 4826 err = xdp_do_generic_redirect(skb->dev, skb, 4827 &xdp, xdp_prog); 4828 if (err) 4829 goto out_redir; 4830 break; 4831 case XDP_TX: 4832 generic_xdp_tx(skb, xdp_prog); 4833 break; 4834 } 4835 return XDP_DROP; 4836 } 4837 } 4838 return XDP_PASS; 4839 out_redir: 4840 kfree_skb_reason(skb, SKB_DROP_REASON_XDP); 4841 return XDP_DROP; 4842 } 4843 EXPORT_SYMBOL_GPL(do_xdp_generic); 4844 4845 static int netif_rx_internal(struct sk_buff *skb) 4846 { 4847 int ret; 4848 4849 net_timestamp_check(netdev_tstamp_prequeue, skb); 4850 4851 trace_netif_rx(skb); 4852 4853 #ifdef CONFIG_RPS 4854 if (static_branch_unlikely(&rps_needed)) { 4855 struct rps_dev_flow voidflow, *rflow = &voidflow; 4856 int cpu; 4857 4858 rcu_read_lock(); 4859 4860 cpu = get_rps_cpu(skb->dev, skb, &rflow); 4861 if (cpu < 0) 4862 cpu = smp_processor_id(); 4863 4864 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 4865 4866 rcu_read_unlock(); 4867 } else 4868 #endif 4869 { 4870 unsigned int qtail; 4871 4872 ret = enqueue_to_backlog(skb, smp_processor_id(), &qtail); 4873 } 4874 return ret; 4875 } 4876 4877 /** 4878 * __netif_rx - Slightly optimized version of netif_rx 4879 * @skb: buffer to post 4880 * 4881 * This behaves as netif_rx except that it does not disable bottom halves. 4882 * As a result this function may only be invoked from the interrupt context 4883 * (either hard or soft interrupt). 4884 */ 4885 int __netif_rx(struct sk_buff *skb) 4886 { 4887 int ret; 4888 4889 lockdep_assert_once(hardirq_count() | softirq_count()); 4890 4891 trace_netif_rx_entry(skb); 4892 ret = netif_rx_internal(skb); 4893 trace_netif_rx_exit(ret); 4894 return ret; 4895 } 4896 EXPORT_SYMBOL(__netif_rx); 4897 4898 /** 4899 * netif_rx - post buffer to the network code 4900 * @skb: buffer to post 4901 * 4902 * This function receives a packet from a device driver and queues it for 4903 * the upper (protocol) levels to process via the backlog NAPI device. It 4904 * always succeeds. The buffer may be dropped during processing for 4905 * congestion control or by the protocol layers. 4906 * The network buffer is passed via the backlog NAPI device. Modern NIC 4907 * driver should use NAPI and GRO. 4908 * This function can used from interrupt and from process context. The 4909 * caller from process context must not disable interrupts before invoking 4910 * this function. 4911 * 4912 * return values: 4913 * NET_RX_SUCCESS (no congestion) 4914 * NET_RX_DROP (packet was dropped) 4915 * 4916 */ 4917 int netif_rx(struct sk_buff *skb) 4918 { 4919 bool need_bh_off = !(hardirq_count() | softirq_count()); 4920 int ret; 4921 4922 if (need_bh_off) 4923 local_bh_disable(); 4924 trace_netif_rx_entry(skb); 4925 ret = netif_rx_internal(skb); 4926 trace_netif_rx_exit(ret); 4927 if (need_bh_off) 4928 local_bh_enable(); 4929 return ret; 4930 } 4931 EXPORT_SYMBOL(netif_rx); 4932 4933 static __latent_entropy void net_tx_action(struct softirq_action *h) 4934 { 4935 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 4936 4937 if (sd->completion_queue) { 4938 struct sk_buff *clist; 4939 4940 local_irq_disable(); 4941 clist = sd->completion_queue; 4942 sd->completion_queue = NULL; 4943 local_irq_enable(); 4944 4945 while (clist) { 4946 struct sk_buff *skb = clist; 4947 4948 clist = clist->next; 4949 4950 WARN_ON(refcount_read(&skb->users)); 4951 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED)) 4952 trace_consume_skb(skb); 4953 else 4954 trace_kfree_skb(skb, net_tx_action, 4955 SKB_DROP_REASON_NOT_SPECIFIED); 4956 4957 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) 4958 __kfree_skb(skb); 4959 else 4960 __kfree_skb_defer(skb); 4961 } 4962 } 4963 4964 if (sd->output_queue) { 4965 struct Qdisc *head; 4966 4967 local_irq_disable(); 4968 head = sd->output_queue; 4969 sd->output_queue = NULL; 4970 sd->output_queue_tailp = &sd->output_queue; 4971 local_irq_enable(); 4972 4973 rcu_read_lock(); 4974 4975 while (head) { 4976 struct Qdisc *q = head; 4977 spinlock_t *root_lock = NULL; 4978 4979 head = head->next_sched; 4980 4981 /* We need to make sure head->next_sched is read 4982 * before clearing __QDISC_STATE_SCHED 4983 */ 4984 smp_mb__before_atomic(); 4985 4986 if (!(q->flags & TCQ_F_NOLOCK)) { 4987 root_lock = qdisc_lock(q); 4988 spin_lock(root_lock); 4989 } else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, 4990 &q->state))) { 4991 /* There is a synchronize_net() between 4992 * STATE_DEACTIVATED flag being set and 4993 * qdisc_reset()/some_qdisc_is_busy() in 4994 * dev_deactivate(), so we can safely bail out 4995 * early here to avoid data race between 4996 * qdisc_deactivate() and some_qdisc_is_busy() 4997 * for lockless qdisc. 4998 */ 4999 clear_bit(__QDISC_STATE_SCHED, &q->state); 5000 continue; 5001 } 5002 5003 clear_bit(__QDISC_STATE_SCHED, &q->state); 5004 qdisc_run(q); 5005 if (root_lock) 5006 spin_unlock(root_lock); 5007 } 5008 5009 rcu_read_unlock(); 5010 } 5011 5012 xfrm_dev_backlog(sd); 5013 } 5014 5015 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) 5016 /* This hook is defined here for ATM LANE */ 5017 int (*br_fdb_test_addr_hook)(struct net_device *dev, 5018 unsigned char *addr) __read_mostly; 5019 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); 5020 #endif 5021 5022 static inline struct sk_buff * 5023 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 5024 struct net_device *orig_dev, bool *another) 5025 { 5026 #ifdef CONFIG_NET_CLS_ACT 5027 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress); 5028 struct tcf_result cl_res; 5029 5030 /* If there's at least one ingress present somewhere (so 5031 * we get here via enabled static key), remaining devices 5032 * that are not configured with an ingress qdisc will bail 5033 * out here. 5034 */ 5035 if (!miniq) 5036 return skb; 5037 5038 if (*pt_prev) { 5039 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5040 *pt_prev = NULL; 5041 } 5042 5043 qdisc_skb_cb(skb)->pkt_len = skb->len; 5044 tc_skb_cb(skb)->mru = 0; 5045 tc_skb_cb(skb)->post_ct = false; 5046 skb->tc_at_ingress = 1; 5047 mini_qdisc_bstats_cpu_update(miniq, skb); 5048 5049 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) { 5050 case TC_ACT_OK: 5051 case TC_ACT_RECLASSIFY: 5052 skb->tc_index = TC_H_MIN(cl_res.classid); 5053 break; 5054 case TC_ACT_SHOT: 5055 mini_qdisc_qstats_cpu_drop(miniq); 5056 kfree_skb_reason(skb, SKB_DROP_REASON_TC_INGRESS); 5057 return NULL; 5058 case TC_ACT_STOLEN: 5059 case TC_ACT_QUEUED: 5060 case TC_ACT_TRAP: 5061 consume_skb(skb); 5062 return NULL; 5063 case TC_ACT_REDIRECT: 5064 /* skb_mac_header check was done by cls/act_bpf, so 5065 * we can safely push the L2 header back before 5066 * redirecting to another netdev 5067 */ 5068 __skb_push(skb, skb->mac_len); 5069 if (skb_do_redirect(skb) == -EAGAIN) { 5070 __skb_pull(skb, skb->mac_len); 5071 *another = true; 5072 break; 5073 } 5074 return NULL; 5075 case TC_ACT_CONSUMED: 5076 return NULL; 5077 default: 5078 break; 5079 } 5080 #endif /* CONFIG_NET_CLS_ACT */ 5081 return skb; 5082 } 5083 5084 /** 5085 * netdev_is_rx_handler_busy - check if receive handler is registered 5086 * @dev: device to check 5087 * 5088 * Check if a receive handler is already registered for a given device. 5089 * Return true if there one. 5090 * 5091 * The caller must hold the rtnl_mutex. 5092 */ 5093 bool netdev_is_rx_handler_busy(struct net_device *dev) 5094 { 5095 ASSERT_RTNL(); 5096 return dev && rtnl_dereference(dev->rx_handler); 5097 } 5098 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); 5099 5100 /** 5101 * netdev_rx_handler_register - register receive handler 5102 * @dev: device to register a handler for 5103 * @rx_handler: receive handler to register 5104 * @rx_handler_data: data pointer that is used by rx handler 5105 * 5106 * Register a receive handler for a device. This handler will then be 5107 * called from __netif_receive_skb. A negative errno code is returned 5108 * on a failure. 5109 * 5110 * The caller must hold the rtnl_mutex. 5111 * 5112 * For a general description of rx_handler, see enum rx_handler_result. 5113 */ 5114 int netdev_rx_handler_register(struct net_device *dev, 5115 rx_handler_func_t *rx_handler, 5116 void *rx_handler_data) 5117 { 5118 if (netdev_is_rx_handler_busy(dev)) 5119 return -EBUSY; 5120 5121 if (dev->priv_flags & IFF_NO_RX_HANDLER) 5122 return -EINVAL; 5123 5124 /* Note: rx_handler_data must be set before rx_handler */ 5125 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); 5126 rcu_assign_pointer(dev->rx_handler, rx_handler); 5127 5128 return 0; 5129 } 5130 EXPORT_SYMBOL_GPL(netdev_rx_handler_register); 5131 5132 /** 5133 * netdev_rx_handler_unregister - unregister receive handler 5134 * @dev: device to unregister a handler from 5135 * 5136 * Unregister a receive handler from a device. 5137 * 5138 * The caller must hold the rtnl_mutex. 5139 */ 5140 void netdev_rx_handler_unregister(struct net_device *dev) 5141 { 5142 5143 ASSERT_RTNL(); 5144 RCU_INIT_POINTER(dev->rx_handler, NULL); 5145 /* a reader seeing a non NULL rx_handler in a rcu_read_lock() 5146 * section has a guarantee to see a non NULL rx_handler_data 5147 * as well. 5148 */ 5149 synchronize_net(); 5150 RCU_INIT_POINTER(dev->rx_handler_data, NULL); 5151 } 5152 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); 5153 5154 /* 5155 * Limit the use of PFMEMALLOC reserves to those protocols that implement 5156 * the special handling of PFMEMALLOC skbs. 5157 */ 5158 static bool skb_pfmemalloc_protocol(struct sk_buff *skb) 5159 { 5160 switch (skb->protocol) { 5161 case htons(ETH_P_ARP): 5162 case htons(ETH_P_IP): 5163 case htons(ETH_P_IPV6): 5164 case htons(ETH_P_8021Q): 5165 case htons(ETH_P_8021AD): 5166 return true; 5167 default: 5168 return false; 5169 } 5170 } 5171 5172 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, 5173 int *ret, struct net_device *orig_dev) 5174 { 5175 if (nf_hook_ingress_active(skb)) { 5176 int ingress_retval; 5177 5178 if (*pt_prev) { 5179 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5180 *pt_prev = NULL; 5181 } 5182 5183 rcu_read_lock(); 5184 ingress_retval = nf_hook_ingress(skb); 5185 rcu_read_unlock(); 5186 return ingress_retval; 5187 } 5188 return 0; 5189 } 5190 5191 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc, 5192 struct packet_type **ppt_prev) 5193 { 5194 struct packet_type *ptype, *pt_prev; 5195 rx_handler_func_t *rx_handler; 5196 struct sk_buff *skb = *pskb; 5197 struct net_device *orig_dev; 5198 bool deliver_exact = false; 5199 int ret = NET_RX_DROP; 5200 __be16 type; 5201 5202 net_timestamp_check(!netdev_tstamp_prequeue, skb); 5203 5204 trace_netif_receive_skb(skb); 5205 5206 orig_dev = skb->dev; 5207 5208 skb_reset_network_header(skb); 5209 if (!skb_transport_header_was_set(skb)) 5210 skb_reset_transport_header(skb); 5211 skb_reset_mac_len(skb); 5212 5213 pt_prev = NULL; 5214 5215 another_round: 5216 skb->skb_iif = skb->dev->ifindex; 5217 5218 __this_cpu_inc(softnet_data.processed); 5219 5220 if (static_branch_unlikely(&generic_xdp_needed_key)) { 5221 int ret2; 5222 5223 migrate_disable(); 5224 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb); 5225 migrate_enable(); 5226 5227 if (ret2 != XDP_PASS) { 5228 ret = NET_RX_DROP; 5229 goto out; 5230 } 5231 } 5232 5233 if (eth_type_vlan(skb->protocol)) { 5234 skb = skb_vlan_untag(skb); 5235 if (unlikely(!skb)) 5236 goto out; 5237 } 5238 5239 if (skb_skip_tc_classify(skb)) 5240 goto skip_classify; 5241 5242 if (pfmemalloc) 5243 goto skip_taps; 5244 5245 list_for_each_entry_rcu(ptype, &ptype_all, list) { 5246 if (pt_prev) 5247 ret = deliver_skb(skb, pt_prev, orig_dev); 5248 pt_prev = ptype; 5249 } 5250 5251 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { 5252 if (pt_prev) 5253 ret = deliver_skb(skb, pt_prev, orig_dev); 5254 pt_prev = ptype; 5255 } 5256 5257 skip_taps: 5258 #ifdef CONFIG_NET_INGRESS 5259 if (static_branch_unlikely(&ingress_needed_key)) { 5260 bool another = false; 5261 5262 nf_skip_egress(skb, true); 5263 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev, 5264 &another); 5265 if (another) 5266 goto another_round; 5267 if (!skb) 5268 goto out; 5269 5270 nf_skip_egress(skb, false); 5271 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) 5272 goto out; 5273 } 5274 #endif 5275 skb_reset_redirect(skb); 5276 skip_classify: 5277 if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) 5278 goto drop; 5279 5280 if (skb_vlan_tag_present(skb)) { 5281 if (pt_prev) { 5282 ret = deliver_skb(skb, pt_prev, orig_dev); 5283 pt_prev = NULL; 5284 } 5285 if (vlan_do_receive(&skb)) 5286 goto another_round; 5287 else if (unlikely(!skb)) 5288 goto out; 5289 } 5290 5291 rx_handler = rcu_dereference(skb->dev->rx_handler); 5292 if (rx_handler) { 5293 if (pt_prev) { 5294 ret = deliver_skb(skb, pt_prev, orig_dev); 5295 pt_prev = NULL; 5296 } 5297 switch (rx_handler(&skb)) { 5298 case RX_HANDLER_CONSUMED: 5299 ret = NET_RX_SUCCESS; 5300 goto out; 5301 case RX_HANDLER_ANOTHER: 5302 goto another_round; 5303 case RX_HANDLER_EXACT: 5304 deliver_exact = true; 5305 break; 5306 case RX_HANDLER_PASS: 5307 break; 5308 default: 5309 BUG(); 5310 } 5311 } 5312 5313 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) { 5314 check_vlan_id: 5315 if (skb_vlan_tag_get_id(skb)) { 5316 /* Vlan id is non 0 and vlan_do_receive() above couldn't 5317 * find vlan device. 5318 */ 5319 skb->pkt_type = PACKET_OTHERHOST; 5320 } else if (eth_type_vlan(skb->protocol)) { 5321 /* Outer header is 802.1P with vlan 0, inner header is 5322 * 802.1Q or 802.1AD and vlan_do_receive() above could 5323 * not find vlan dev for vlan id 0. 5324 */ 5325 __vlan_hwaccel_clear_tag(skb); 5326 skb = skb_vlan_untag(skb); 5327 if (unlikely(!skb)) 5328 goto out; 5329 if (vlan_do_receive(&skb)) 5330 /* After stripping off 802.1P header with vlan 0 5331 * vlan dev is found for inner header. 5332 */ 5333 goto another_round; 5334 else if (unlikely(!skb)) 5335 goto out; 5336 else 5337 /* We have stripped outer 802.1P vlan 0 header. 5338 * But could not find vlan dev. 5339 * check again for vlan id to set OTHERHOST. 5340 */ 5341 goto check_vlan_id; 5342 } 5343 /* Note: we might in the future use prio bits 5344 * and set skb->priority like in vlan_do_receive() 5345 * For the time being, just ignore Priority Code Point 5346 */ 5347 __vlan_hwaccel_clear_tag(skb); 5348 } 5349 5350 type = skb->protocol; 5351 5352 /* deliver only exact match when indicated */ 5353 if (likely(!deliver_exact)) { 5354 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5355 &ptype_base[ntohs(type) & 5356 PTYPE_HASH_MASK]); 5357 } 5358 5359 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5360 &orig_dev->ptype_specific); 5361 5362 if (unlikely(skb->dev != orig_dev)) { 5363 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5364 &skb->dev->ptype_specific); 5365 } 5366 5367 if (pt_prev) { 5368 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 5369 goto drop; 5370 *ppt_prev = pt_prev; 5371 } else { 5372 drop: 5373 if (!deliver_exact) { 5374 dev_core_stats_rx_dropped_inc(skb->dev); 5375 kfree_skb_reason(skb, SKB_DROP_REASON_PTYPE_ABSENT); 5376 } else { 5377 dev_core_stats_rx_nohandler_inc(skb->dev); 5378 kfree_skb(skb); 5379 } 5380 /* Jamal, now you will not able to escape explaining 5381 * me how you were going to use this. :-) 5382 */ 5383 ret = NET_RX_DROP; 5384 } 5385 5386 out: 5387 /* The invariant here is that if *ppt_prev is not NULL 5388 * then skb should also be non-NULL. 5389 * 5390 * Apparently *ppt_prev assignment above holds this invariant due to 5391 * skb dereferencing near it. 5392 */ 5393 *pskb = skb; 5394 return ret; 5395 } 5396 5397 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc) 5398 { 5399 struct net_device *orig_dev = skb->dev; 5400 struct packet_type *pt_prev = NULL; 5401 int ret; 5402 5403 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5404 if (pt_prev) 5405 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb, 5406 skb->dev, pt_prev, orig_dev); 5407 return ret; 5408 } 5409 5410 /** 5411 * netif_receive_skb_core - special purpose version of netif_receive_skb 5412 * @skb: buffer to process 5413 * 5414 * More direct receive version of netif_receive_skb(). It should 5415 * only be used by callers that have a need to skip RPS and Generic XDP. 5416 * Caller must also take care of handling if ``(page_is_)pfmemalloc``. 5417 * 5418 * This function may only be called from softirq context and interrupts 5419 * should be enabled. 5420 * 5421 * Return values (usually ignored): 5422 * NET_RX_SUCCESS: no congestion 5423 * NET_RX_DROP: packet was dropped 5424 */ 5425 int netif_receive_skb_core(struct sk_buff *skb) 5426 { 5427 int ret; 5428 5429 rcu_read_lock(); 5430 ret = __netif_receive_skb_one_core(skb, false); 5431 rcu_read_unlock(); 5432 5433 return ret; 5434 } 5435 EXPORT_SYMBOL(netif_receive_skb_core); 5436 5437 static inline void __netif_receive_skb_list_ptype(struct list_head *head, 5438 struct packet_type *pt_prev, 5439 struct net_device *orig_dev) 5440 { 5441 struct sk_buff *skb, *next; 5442 5443 if (!pt_prev) 5444 return; 5445 if (list_empty(head)) 5446 return; 5447 if (pt_prev->list_func != NULL) 5448 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv, 5449 ip_list_rcv, head, pt_prev, orig_dev); 5450 else 5451 list_for_each_entry_safe(skb, next, head, list) { 5452 skb_list_del_init(skb); 5453 pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 5454 } 5455 } 5456 5457 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc) 5458 { 5459 /* Fast-path assumptions: 5460 * - There is no RX handler. 5461 * - Only one packet_type matches. 5462 * If either of these fails, we will end up doing some per-packet 5463 * processing in-line, then handling the 'last ptype' for the whole 5464 * sublist. This can't cause out-of-order delivery to any single ptype, 5465 * because the 'last ptype' must be constant across the sublist, and all 5466 * other ptypes are handled per-packet. 5467 */ 5468 /* Current (common) ptype of sublist */ 5469 struct packet_type *pt_curr = NULL; 5470 /* Current (common) orig_dev of sublist */ 5471 struct net_device *od_curr = NULL; 5472 struct list_head sublist; 5473 struct sk_buff *skb, *next; 5474 5475 INIT_LIST_HEAD(&sublist); 5476 list_for_each_entry_safe(skb, next, head, list) { 5477 struct net_device *orig_dev = skb->dev; 5478 struct packet_type *pt_prev = NULL; 5479 5480 skb_list_del_init(skb); 5481 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5482 if (!pt_prev) 5483 continue; 5484 if (pt_curr != pt_prev || od_curr != orig_dev) { 5485 /* dispatch old sublist */ 5486 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5487 /* start new sublist */ 5488 INIT_LIST_HEAD(&sublist); 5489 pt_curr = pt_prev; 5490 od_curr = orig_dev; 5491 } 5492 list_add_tail(&skb->list, &sublist); 5493 } 5494 5495 /* dispatch final sublist */ 5496 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5497 } 5498 5499 static int __netif_receive_skb(struct sk_buff *skb) 5500 { 5501 int ret; 5502 5503 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { 5504 unsigned int noreclaim_flag; 5505 5506 /* 5507 * PFMEMALLOC skbs are special, they should 5508 * - be delivered to SOCK_MEMALLOC sockets only 5509 * - stay away from userspace 5510 * - have bounded memory usage 5511 * 5512 * Use PF_MEMALLOC as this saves us from propagating the allocation 5513 * context down to all allocation sites. 5514 */ 5515 noreclaim_flag = memalloc_noreclaim_save(); 5516 ret = __netif_receive_skb_one_core(skb, true); 5517 memalloc_noreclaim_restore(noreclaim_flag); 5518 } else 5519 ret = __netif_receive_skb_one_core(skb, false); 5520 5521 return ret; 5522 } 5523 5524 static void __netif_receive_skb_list(struct list_head *head) 5525 { 5526 unsigned long noreclaim_flag = 0; 5527 struct sk_buff *skb, *next; 5528 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */ 5529 5530 list_for_each_entry_safe(skb, next, head, list) { 5531 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) { 5532 struct list_head sublist; 5533 5534 /* Handle the previous sublist */ 5535 list_cut_before(&sublist, head, &skb->list); 5536 if (!list_empty(&sublist)) 5537 __netif_receive_skb_list_core(&sublist, pfmemalloc); 5538 pfmemalloc = !pfmemalloc; 5539 /* See comments in __netif_receive_skb */ 5540 if (pfmemalloc) 5541 noreclaim_flag = memalloc_noreclaim_save(); 5542 else 5543 memalloc_noreclaim_restore(noreclaim_flag); 5544 } 5545 } 5546 /* Handle the remaining sublist */ 5547 if (!list_empty(head)) 5548 __netif_receive_skb_list_core(head, pfmemalloc); 5549 /* Restore pflags */ 5550 if (pfmemalloc) 5551 memalloc_noreclaim_restore(noreclaim_flag); 5552 } 5553 5554 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp) 5555 { 5556 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog); 5557 struct bpf_prog *new = xdp->prog; 5558 int ret = 0; 5559 5560 switch (xdp->command) { 5561 case XDP_SETUP_PROG: 5562 rcu_assign_pointer(dev->xdp_prog, new); 5563 if (old) 5564 bpf_prog_put(old); 5565 5566 if (old && !new) { 5567 static_branch_dec(&generic_xdp_needed_key); 5568 } else if (new && !old) { 5569 static_branch_inc(&generic_xdp_needed_key); 5570 dev_disable_lro(dev); 5571 dev_disable_gro_hw(dev); 5572 } 5573 break; 5574 5575 default: 5576 ret = -EINVAL; 5577 break; 5578 } 5579 5580 return ret; 5581 } 5582 5583 static int netif_receive_skb_internal(struct sk_buff *skb) 5584 { 5585 int ret; 5586 5587 net_timestamp_check(netdev_tstamp_prequeue, skb); 5588 5589 if (skb_defer_rx_timestamp(skb)) 5590 return NET_RX_SUCCESS; 5591 5592 rcu_read_lock(); 5593 #ifdef CONFIG_RPS 5594 if (static_branch_unlikely(&rps_needed)) { 5595 struct rps_dev_flow voidflow, *rflow = &voidflow; 5596 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5597 5598 if (cpu >= 0) { 5599 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5600 rcu_read_unlock(); 5601 return ret; 5602 } 5603 } 5604 #endif 5605 ret = __netif_receive_skb(skb); 5606 rcu_read_unlock(); 5607 return ret; 5608 } 5609 5610 void netif_receive_skb_list_internal(struct list_head *head) 5611 { 5612 struct sk_buff *skb, *next; 5613 struct list_head sublist; 5614 5615 INIT_LIST_HEAD(&sublist); 5616 list_for_each_entry_safe(skb, next, head, list) { 5617 net_timestamp_check(netdev_tstamp_prequeue, skb); 5618 skb_list_del_init(skb); 5619 if (!skb_defer_rx_timestamp(skb)) 5620 list_add_tail(&skb->list, &sublist); 5621 } 5622 list_splice_init(&sublist, head); 5623 5624 rcu_read_lock(); 5625 #ifdef CONFIG_RPS 5626 if (static_branch_unlikely(&rps_needed)) { 5627 list_for_each_entry_safe(skb, next, head, list) { 5628 struct rps_dev_flow voidflow, *rflow = &voidflow; 5629 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5630 5631 if (cpu >= 0) { 5632 /* Will be handled, remove from list */ 5633 skb_list_del_init(skb); 5634 enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5635 } 5636 } 5637 } 5638 #endif 5639 __netif_receive_skb_list(head); 5640 rcu_read_unlock(); 5641 } 5642 5643 /** 5644 * netif_receive_skb - process receive buffer from network 5645 * @skb: buffer to process 5646 * 5647 * netif_receive_skb() is the main receive data processing function. 5648 * It always succeeds. The buffer may be dropped during processing 5649 * for congestion control or by the protocol layers. 5650 * 5651 * This function may only be called from softirq context and interrupts 5652 * should be enabled. 5653 * 5654 * Return values (usually ignored): 5655 * NET_RX_SUCCESS: no congestion 5656 * NET_RX_DROP: packet was dropped 5657 */ 5658 int netif_receive_skb(struct sk_buff *skb) 5659 { 5660 int ret; 5661 5662 trace_netif_receive_skb_entry(skb); 5663 5664 ret = netif_receive_skb_internal(skb); 5665 trace_netif_receive_skb_exit(ret); 5666 5667 return ret; 5668 } 5669 EXPORT_SYMBOL(netif_receive_skb); 5670 5671 /** 5672 * netif_receive_skb_list - process many receive buffers from network 5673 * @head: list of skbs to process. 5674 * 5675 * Since return value of netif_receive_skb() is normally ignored, and 5676 * wouldn't be meaningful for a list, this function returns void. 5677 * 5678 * This function may only be called from softirq context and interrupts 5679 * should be enabled. 5680 */ 5681 void netif_receive_skb_list(struct list_head *head) 5682 { 5683 struct sk_buff *skb; 5684 5685 if (list_empty(head)) 5686 return; 5687 if (trace_netif_receive_skb_list_entry_enabled()) { 5688 list_for_each_entry(skb, head, list) 5689 trace_netif_receive_skb_list_entry(skb); 5690 } 5691 netif_receive_skb_list_internal(head); 5692 trace_netif_receive_skb_list_exit(0); 5693 } 5694 EXPORT_SYMBOL(netif_receive_skb_list); 5695 5696 static DEFINE_PER_CPU(struct work_struct, flush_works); 5697 5698 /* Network device is going away, flush any packets still pending */ 5699 static void flush_backlog(struct work_struct *work) 5700 { 5701 struct sk_buff *skb, *tmp; 5702 struct softnet_data *sd; 5703 5704 local_bh_disable(); 5705 sd = this_cpu_ptr(&softnet_data); 5706 5707 rps_lock_irq_disable(sd); 5708 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { 5709 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5710 __skb_unlink(skb, &sd->input_pkt_queue); 5711 dev_kfree_skb_irq(skb); 5712 input_queue_head_incr(sd); 5713 } 5714 } 5715 rps_unlock_irq_enable(sd); 5716 5717 skb_queue_walk_safe(&sd->process_queue, skb, tmp) { 5718 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5719 __skb_unlink(skb, &sd->process_queue); 5720 kfree_skb(skb); 5721 input_queue_head_incr(sd); 5722 } 5723 } 5724 local_bh_enable(); 5725 } 5726 5727 static bool flush_required(int cpu) 5728 { 5729 #if IS_ENABLED(CONFIG_RPS) 5730 struct softnet_data *sd = &per_cpu(softnet_data, cpu); 5731 bool do_flush; 5732 5733 rps_lock_irq_disable(sd); 5734 5735 /* as insertion into process_queue happens with the rps lock held, 5736 * process_queue access may race only with dequeue 5737 */ 5738 do_flush = !skb_queue_empty(&sd->input_pkt_queue) || 5739 !skb_queue_empty_lockless(&sd->process_queue); 5740 rps_unlock_irq_enable(sd); 5741 5742 return do_flush; 5743 #endif 5744 /* without RPS we can't safely check input_pkt_queue: during a 5745 * concurrent remote skb_queue_splice() we can detect as empty both 5746 * input_pkt_queue and process_queue even if the latter could end-up 5747 * containing a lot of packets. 5748 */ 5749 return true; 5750 } 5751 5752 static void flush_all_backlogs(void) 5753 { 5754 static cpumask_t flush_cpus; 5755 unsigned int cpu; 5756 5757 /* since we are under rtnl lock protection we can use static data 5758 * for the cpumask and avoid allocating on stack the possibly 5759 * large mask 5760 */ 5761 ASSERT_RTNL(); 5762 5763 cpus_read_lock(); 5764 5765 cpumask_clear(&flush_cpus); 5766 for_each_online_cpu(cpu) { 5767 if (flush_required(cpu)) { 5768 queue_work_on(cpu, system_highpri_wq, 5769 per_cpu_ptr(&flush_works, cpu)); 5770 cpumask_set_cpu(cpu, &flush_cpus); 5771 } 5772 } 5773 5774 /* we can have in flight packet[s] on the cpus we are not flushing, 5775 * synchronize_net() in unregister_netdevice_many() will take care of 5776 * them 5777 */ 5778 for_each_cpu(cpu, &flush_cpus) 5779 flush_work(per_cpu_ptr(&flush_works, cpu)); 5780 5781 cpus_read_unlock(); 5782 } 5783 5784 static void net_rps_send_ipi(struct softnet_data *remsd) 5785 { 5786 #ifdef CONFIG_RPS 5787 while (remsd) { 5788 struct softnet_data *next = remsd->rps_ipi_next; 5789 5790 if (cpu_online(remsd->cpu)) 5791 smp_call_function_single_async(remsd->cpu, &remsd->csd); 5792 remsd = next; 5793 } 5794 #endif 5795 } 5796 5797 /* 5798 * net_rps_action_and_irq_enable sends any pending IPI's for rps. 5799 * Note: called with local irq disabled, but exits with local irq enabled. 5800 */ 5801 static void net_rps_action_and_irq_enable(struct softnet_data *sd) 5802 { 5803 #ifdef CONFIG_RPS 5804 struct softnet_data *remsd = sd->rps_ipi_list; 5805 5806 if (remsd) { 5807 sd->rps_ipi_list = NULL; 5808 5809 local_irq_enable(); 5810 5811 /* Send pending IPI's to kick RPS processing on remote cpus. */ 5812 net_rps_send_ipi(remsd); 5813 } else 5814 #endif 5815 local_irq_enable(); 5816 } 5817 5818 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) 5819 { 5820 #ifdef CONFIG_RPS 5821 return sd->rps_ipi_list != NULL; 5822 #else 5823 return false; 5824 #endif 5825 } 5826 5827 static int process_backlog(struct napi_struct *napi, int quota) 5828 { 5829 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); 5830 bool again = true; 5831 int work = 0; 5832 5833 /* Check if we have pending ipi, its better to send them now, 5834 * not waiting net_rx_action() end. 5835 */ 5836 if (sd_has_rps_ipi_waiting(sd)) { 5837 local_irq_disable(); 5838 net_rps_action_and_irq_enable(sd); 5839 } 5840 5841 napi->weight = dev_rx_weight; 5842 while (again) { 5843 struct sk_buff *skb; 5844 5845 while ((skb = __skb_dequeue(&sd->process_queue))) { 5846 rcu_read_lock(); 5847 __netif_receive_skb(skb); 5848 rcu_read_unlock(); 5849 input_queue_head_incr(sd); 5850 if (++work >= quota) 5851 return work; 5852 5853 } 5854 5855 rps_lock_irq_disable(sd); 5856 if (skb_queue_empty(&sd->input_pkt_queue)) { 5857 /* 5858 * Inline a custom version of __napi_complete(). 5859 * only current cpu owns and manipulates this napi, 5860 * and NAPI_STATE_SCHED is the only possible flag set 5861 * on backlog. 5862 * We can use a plain write instead of clear_bit(), 5863 * and we dont need an smp_mb() memory barrier. 5864 */ 5865 napi->state = 0; 5866 again = false; 5867 } else { 5868 skb_queue_splice_tail_init(&sd->input_pkt_queue, 5869 &sd->process_queue); 5870 } 5871 rps_unlock_irq_enable(sd); 5872 } 5873 5874 return work; 5875 } 5876 5877 /** 5878 * __napi_schedule - schedule for receive 5879 * @n: entry to schedule 5880 * 5881 * The entry's receive function will be scheduled to run. 5882 * Consider using __napi_schedule_irqoff() if hard irqs are masked. 5883 */ 5884 void __napi_schedule(struct napi_struct *n) 5885 { 5886 unsigned long flags; 5887 5888 local_irq_save(flags); 5889 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 5890 local_irq_restore(flags); 5891 } 5892 EXPORT_SYMBOL(__napi_schedule); 5893 5894 /** 5895 * napi_schedule_prep - check if napi can be scheduled 5896 * @n: napi context 5897 * 5898 * Test if NAPI routine is already running, and if not mark 5899 * it as running. This is used as a condition variable to 5900 * insure only one NAPI poll instance runs. We also make 5901 * sure there is no pending NAPI disable. 5902 */ 5903 bool napi_schedule_prep(struct napi_struct *n) 5904 { 5905 unsigned long val, new; 5906 5907 do { 5908 val = READ_ONCE(n->state); 5909 if (unlikely(val & NAPIF_STATE_DISABLE)) 5910 return false; 5911 new = val | NAPIF_STATE_SCHED; 5912 5913 /* Sets STATE_MISSED bit if STATE_SCHED was already set 5914 * This was suggested by Alexander Duyck, as compiler 5915 * emits better code than : 5916 * if (val & NAPIF_STATE_SCHED) 5917 * new |= NAPIF_STATE_MISSED; 5918 */ 5919 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * 5920 NAPIF_STATE_MISSED; 5921 } while (cmpxchg(&n->state, val, new) != val); 5922 5923 return !(val & NAPIF_STATE_SCHED); 5924 } 5925 EXPORT_SYMBOL(napi_schedule_prep); 5926 5927 /** 5928 * __napi_schedule_irqoff - schedule for receive 5929 * @n: entry to schedule 5930 * 5931 * Variant of __napi_schedule() assuming hard irqs are masked. 5932 * 5933 * On PREEMPT_RT enabled kernels this maps to __napi_schedule() 5934 * because the interrupt disabled assumption might not be true 5935 * due to force-threaded interrupts and spinlock substitution. 5936 */ 5937 void __napi_schedule_irqoff(struct napi_struct *n) 5938 { 5939 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 5940 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 5941 else 5942 __napi_schedule(n); 5943 } 5944 EXPORT_SYMBOL(__napi_schedule_irqoff); 5945 5946 bool napi_complete_done(struct napi_struct *n, int work_done) 5947 { 5948 unsigned long flags, val, new, timeout = 0; 5949 bool ret = true; 5950 5951 /* 5952 * 1) Don't let napi dequeue from the cpu poll list 5953 * just in case its running on a different cpu. 5954 * 2) If we are busy polling, do nothing here, we have 5955 * the guarantee we will be called later. 5956 */ 5957 if (unlikely(n->state & (NAPIF_STATE_NPSVC | 5958 NAPIF_STATE_IN_BUSY_POLL))) 5959 return false; 5960 5961 if (work_done) { 5962 if (n->gro_bitmask) 5963 timeout = READ_ONCE(n->dev->gro_flush_timeout); 5964 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs); 5965 } 5966 if (n->defer_hard_irqs_count > 0) { 5967 n->defer_hard_irqs_count--; 5968 timeout = READ_ONCE(n->dev->gro_flush_timeout); 5969 if (timeout) 5970 ret = false; 5971 } 5972 if (n->gro_bitmask) { 5973 /* When the NAPI instance uses a timeout and keeps postponing 5974 * it, we need to bound somehow the time packets are kept in 5975 * the GRO layer 5976 */ 5977 napi_gro_flush(n, !!timeout); 5978 } 5979 5980 gro_normal_list(n); 5981 5982 if (unlikely(!list_empty(&n->poll_list))) { 5983 /* If n->poll_list is not empty, we need to mask irqs */ 5984 local_irq_save(flags); 5985 list_del_init(&n->poll_list); 5986 local_irq_restore(flags); 5987 } 5988 5989 do { 5990 val = READ_ONCE(n->state); 5991 5992 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); 5993 5994 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED | 5995 NAPIF_STATE_SCHED_THREADED | 5996 NAPIF_STATE_PREFER_BUSY_POLL); 5997 5998 /* If STATE_MISSED was set, leave STATE_SCHED set, 5999 * because we will call napi->poll() one more time. 6000 * This C code was suggested by Alexander Duyck to help gcc. 6001 */ 6002 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * 6003 NAPIF_STATE_SCHED; 6004 } while (cmpxchg(&n->state, val, new) != val); 6005 6006 if (unlikely(val & NAPIF_STATE_MISSED)) { 6007 __napi_schedule(n); 6008 return false; 6009 } 6010 6011 if (timeout) 6012 hrtimer_start(&n->timer, ns_to_ktime(timeout), 6013 HRTIMER_MODE_REL_PINNED); 6014 return ret; 6015 } 6016 EXPORT_SYMBOL(napi_complete_done); 6017 6018 /* must be called under rcu_read_lock(), as we dont take a reference */ 6019 static struct napi_struct *napi_by_id(unsigned int napi_id) 6020 { 6021 unsigned int hash = napi_id % HASH_SIZE(napi_hash); 6022 struct napi_struct *napi; 6023 6024 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) 6025 if (napi->napi_id == napi_id) 6026 return napi; 6027 6028 return NULL; 6029 } 6030 6031 #if defined(CONFIG_NET_RX_BUSY_POLL) 6032 6033 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule) 6034 { 6035 if (!skip_schedule) { 6036 gro_normal_list(napi); 6037 __napi_schedule(napi); 6038 return; 6039 } 6040 6041 if (napi->gro_bitmask) { 6042 /* flush too old packets 6043 * If HZ < 1000, flush all packets. 6044 */ 6045 napi_gro_flush(napi, HZ >= 1000); 6046 } 6047 6048 gro_normal_list(napi); 6049 clear_bit(NAPI_STATE_SCHED, &napi->state); 6050 } 6051 6052 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll, 6053 u16 budget) 6054 { 6055 bool skip_schedule = false; 6056 unsigned long timeout; 6057 int rc; 6058 6059 /* Busy polling means there is a high chance device driver hard irq 6060 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was 6061 * set in napi_schedule_prep(). 6062 * Since we are about to call napi->poll() once more, we can safely 6063 * clear NAPI_STATE_MISSED. 6064 * 6065 * Note: x86 could use a single "lock and ..." instruction 6066 * to perform these two clear_bit() 6067 */ 6068 clear_bit(NAPI_STATE_MISSED, &napi->state); 6069 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); 6070 6071 local_bh_disable(); 6072 6073 if (prefer_busy_poll) { 6074 napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs); 6075 timeout = READ_ONCE(napi->dev->gro_flush_timeout); 6076 if (napi->defer_hard_irqs_count && timeout) { 6077 hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED); 6078 skip_schedule = true; 6079 } 6080 } 6081 6082 /* All we really want here is to re-enable device interrupts. 6083 * Ideally, a new ndo_busy_poll_stop() could avoid another round. 6084 */ 6085 rc = napi->poll(napi, budget); 6086 /* We can't gro_normal_list() here, because napi->poll() might have 6087 * rearmed the napi (napi_complete_done()) in which case it could 6088 * already be running on another CPU. 6089 */ 6090 trace_napi_poll(napi, rc, budget); 6091 netpoll_poll_unlock(have_poll_lock); 6092 if (rc == budget) 6093 __busy_poll_stop(napi, skip_schedule); 6094 local_bh_enable(); 6095 } 6096 6097 void napi_busy_loop(unsigned int napi_id, 6098 bool (*loop_end)(void *, unsigned long), 6099 void *loop_end_arg, bool prefer_busy_poll, u16 budget) 6100 { 6101 unsigned long start_time = loop_end ? busy_loop_current_time() : 0; 6102 int (*napi_poll)(struct napi_struct *napi, int budget); 6103 void *have_poll_lock = NULL; 6104 struct napi_struct *napi; 6105 6106 restart: 6107 napi_poll = NULL; 6108 6109 rcu_read_lock(); 6110 6111 napi = napi_by_id(napi_id); 6112 if (!napi) 6113 goto out; 6114 6115 preempt_disable(); 6116 for (;;) { 6117 int work = 0; 6118 6119 local_bh_disable(); 6120 if (!napi_poll) { 6121 unsigned long val = READ_ONCE(napi->state); 6122 6123 /* If multiple threads are competing for this napi, 6124 * we avoid dirtying napi->state as much as we can. 6125 */ 6126 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | 6127 NAPIF_STATE_IN_BUSY_POLL)) { 6128 if (prefer_busy_poll) 6129 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6130 goto count; 6131 } 6132 if (cmpxchg(&napi->state, val, 6133 val | NAPIF_STATE_IN_BUSY_POLL | 6134 NAPIF_STATE_SCHED) != val) { 6135 if (prefer_busy_poll) 6136 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6137 goto count; 6138 } 6139 have_poll_lock = netpoll_poll_lock(napi); 6140 napi_poll = napi->poll; 6141 } 6142 work = napi_poll(napi, budget); 6143 trace_napi_poll(napi, work, budget); 6144 gro_normal_list(napi); 6145 count: 6146 if (work > 0) 6147 __NET_ADD_STATS(dev_net(napi->dev), 6148 LINUX_MIB_BUSYPOLLRXPACKETS, work); 6149 local_bh_enable(); 6150 6151 if (!loop_end || loop_end(loop_end_arg, start_time)) 6152 break; 6153 6154 if (unlikely(need_resched())) { 6155 if (napi_poll) 6156 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6157 preempt_enable(); 6158 rcu_read_unlock(); 6159 cond_resched(); 6160 if (loop_end(loop_end_arg, start_time)) 6161 return; 6162 goto restart; 6163 } 6164 cpu_relax(); 6165 } 6166 if (napi_poll) 6167 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6168 preempt_enable(); 6169 out: 6170 rcu_read_unlock(); 6171 } 6172 EXPORT_SYMBOL(napi_busy_loop); 6173 6174 #endif /* CONFIG_NET_RX_BUSY_POLL */ 6175 6176 static void napi_hash_add(struct napi_struct *napi) 6177 { 6178 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state)) 6179 return; 6180 6181 spin_lock(&napi_hash_lock); 6182 6183 /* 0..NR_CPUS range is reserved for sender_cpu use */ 6184 do { 6185 if (unlikely(++napi_gen_id < MIN_NAPI_ID)) 6186 napi_gen_id = MIN_NAPI_ID; 6187 } while (napi_by_id(napi_gen_id)); 6188 napi->napi_id = napi_gen_id; 6189 6190 hlist_add_head_rcu(&napi->napi_hash_node, 6191 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); 6192 6193 spin_unlock(&napi_hash_lock); 6194 } 6195 6196 /* Warning : caller is responsible to make sure rcu grace period 6197 * is respected before freeing memory containing @napi 6198 */ 6199 static void napi_hash_del(struct napi_struct *napi) 6200 { 6201 spin_lock(&napi_hash_lock); 6202 6203 hlist_del_init_rcu(&napi->napi_hash_node); 6204 6205 spin_unlock(&napi_hash_lock); 6206 } 6207 6208 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) 6209 { 6210 struct napi_struct *napi; 6211 6212 napi = container_of(timer, struct napi_struct, timer); 6213 6214 /* Note : we use a relaxed variant of napi_schedule_prep() not setting 6215 * NAPI_STATE_MISSED, since we do not react to a device IRQ. 6216 */ 6217 if (!napi_disable_pending(napi) && 6218 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) { 6219 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6220 __napi_schedule_irqoff(napi); 6221 } 6222 6223 return HRTIMER_NORESTART; 6224 } 6225 6226 static void init_gro_hash(struct napi_struct *napi) 6227 { 6228 int i; 6229 6230 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6231 INIT_LIST_HEAD(&napi->gro_hash[i].list); 6232 napi->gro_hash[i].count = 0; 6233 } 6234 napi->gro_bitmask = 0; 6235 } 6236 6237 int dev_set_threaded(struct net_device *dev, bool threaded) 6238 { 6239 struct napi_struct *napi; 6240 int err = 0; 6241 6242 if (dev->threaded == threaded) 6243 return 0; 6244 6245 if (threaded) { 6246 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6247 if (!napi->thread) { 6248 err = napi_kthread_create(napi); 6249 if (err) { 6250 threaded = false; 6251 break; 6252 } 6253 } 6254 } 6255 } 6256 6257 dev->threaded = threaded; 6258 6259 /* Make sure kthread is created before THREADED bit 6260 * is set. 6261 */ 6262 smp_mb__before_atomic(); 6263 6264 /* Setting/unsetting threaded mode on a napi might not immediately 6265 * take effect, if the current napi instance is actively being 6266 * polled. In this case, the switch between threaded mode and 6267 * softirq mode will happen in the next round of napi_schedule(). 6268 * This should not cause hiccups/stalls to the live traffic. 6269 */ 6270 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6271 if (threaded) 6272 set_bit(NAPI_STATE_THREADED, &napi->state); 6273 else 6274 clear_bit(NAPI_STATE_THREADED, &napi->state); 6275 } 6276 6277 return err; 6278 } 6279 EXPORT_SYMBOL(dev_set_threaded); 6280 6281 void netif_napi_add(struct net_device *dev, struct napi_struct *napi, 6282 int (*poll)(struct napi_struct *, int), int weight) 6283 { 6284 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state))) 6285 return; 6286 6287 INIT_LIST_HEAD(&napi->poll_list); 6288 INIT_HLIST_NODE(&napi->napi_hash_node); 6289 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); 6290 napi->timer.function = napi_watchdog; 6291 init_gro_hash(napi); 6292 napi->skb = NULL; 6293 INIT_LIST_HEAD(&napi->rx_list); 6294 napi->rx_count = 0; 6295 napi->poll = poll; 6296 if (weight > NAPI_POLL_WEIGHT) 6297 netdev_err_once(dev, "%s() called with weight %d\n", __func__, 6298 weight); 6299 napi->weight = weight; 6300 napi->dev = dev; 6301 #ifdef CONFIG_NETPOLL 6302 napi->poll_owner = -1; 6303 #endif 6304 set_bit(NAPI_STATE_SCHED, &napi->state); 6305 set_bit(NAPI_STATE_NPSVC, &napi->state); 6306 list_add_rcu(&napi->dev_list, &dev->napi_list); 6307 napi_hash_add(napi); 6308 /* Create kthread for this napi if dev->threaded is set. 6309 * Clear dev->threaded if kthread creation failed so that 6310 * threaded mode will not be enabled in napi_enable(). 6311 */ 6312 if (dev->threaded && napi_kthread_create(napi)) 6313 dev->threaded = 0; 6314 } 6315 EXPORT_SYMBOL(netif_napi_add); 6316 6317 void napi_disable(struct napi_struct *n) 6318 { 6319 unsigned long val, new; 6320 6321 might_sleep(); 6322 set_bit(NAPI_STATE_DISABLE, &n->state); 6323 6324 for ( ; ; ) { 6325 val = READ_ONCE(n->state); 6326 if (val & (NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC)) { 6327 usleep_range(20, 200); 6328 continue; 6329 } 6330 6331 new = val | NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC; 6332 new &= ~(NAPIF_STATE_THREADED | NAPIF_STATE_PREFER_BUSY_POLL); 6333 6334 if (cmpxchg(&n->state, val, new) == val) 6335 break; 6336 } 6337 6338 hrtimer_cancel(&n->timer); 6339 6340 clear_bit(NAPI_STATE_DISABLE, &n->state); 6341 } 6342 EXPORT_SYMBOL(napi_disable); 6343 6344 /** 6345 * napi_enable - enable NAPI scheduling 6346 * @n: NAPI context 6347 * 6348 * Resume NAPI from being scheduled on this context. 6349 * Must be paired with napi_disable. 6350 */ 6351 void napi_enable(struct napi_struct *n) 6352 { 6353 unsigned long val, new; 6354 6355 do { 6356 val = READ_ONCE(n->state); 6357 BUG_ON(!test_bit(NAPI_STATE_SCHED, &val)); 6358 6359 new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC); 6360 if (n->dev->threaded && n->thread) 6361 new |= NAPIF_STATE_THREADED; 6362 } while (cmpxchg(&n->state, val, new) != val); 6363 } 6364 EXPORT_SYMBOL(napi_enable); 6365 6366 static void flush_gro_hash(struct napi_struct *napi) 6367 { 6368 int i; 6369 6370 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6371 struct sk_buff *skb, *n; 6372 6373 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list) 6374 kfree_skb(skb); 6375 napi->gro_hash[i].count = 0; 6376 } 6377 } 6378 6379 /* Must be called in process context */ 6380 void __netif_napi_del(struct napi_struct *napi) 6381 { 6382 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state)) 6383 return; 6384 6385 napi_hash_del(napi); 6386 list_del_rcu(&napi->dev_list); 6387 napi_free_frags(napi); 6388 6389 flush_gro_hash(napi); 6390 napi->gro_bitmask = 0; 6391 6392 if (napi->thread) { 6393 kthread_stop(napi->thread); 6394 napi->thread = NULL; 6395 } 6396 } 6397 EXPORT_SYMBOL(__netif_napi_del); 6398 6399 static int __napi_poll(struct napi_struct *n, bool *repoll) 6400 { 6401 int work, weight; 6402 6403 weight = n->weight; 6404 6405 /* This NAPI_STATE_SCHED test is for avoiding a race 6406 * with netpoll's poll_napi(). Only the entity which 6407 * obtains the lock and sees NAPI_STATE_SCHED set will 6408 * actually make the ->poll() call. Therefore we avoid 6409 * accidentally calling ->poll() when NAPI is not scheduled. 6410 */ 6411 work = 0; 6412 if (test_bit(NAPI_STATE_SCHED, &n->state)) { 6413 work = n->poll(n, weight); 6414 trace_napi_poll(n, work, weight); 6415 } 6416 6417 if (unlikely(work > weight)) 6418 netdev_err_once(n->dev, "NAPI poll function %pS returned %d, exceeding its budget of %d.\n", 6419 n->poll, work, weight); 6420 6421 if (likely(work < weight)) 6422 return work; 6423 6424 /* Drivers must not modify the NAPI state if they 6425 * consume the entire weight. In such cases this code 6426 * still "owns" the NAPI instance and therefore can 6427 * move the instance around on the list at-will. 6428 */ 6429 if (unlikely(napi_disable_pending(n))) { 6430 napi_complete(n); 6431 return work; 6432 } 6433 6434 /* The NAPI context has more processing work, but busy-polling 6435 * is preferred. Exit early. 6436 */ 6437 if (napi_prefer_busy_poll(n)) { 6438 if (napi_complete_done(n, work)) { 6439 /* If timeout is not set, we need to make sure 6440 * that the NAPI is re-scheduled. 6441 */ 6442 napi_schedule(n); 6443 } 6444 return work; 6445 } 6446 6447 if (n->gro_bitmask) { 6448 /* flush too old packets 6449 * If HZ < 1000, flush all packets. 6450 */ 6451 napi_gro_flush(n, HZ >= 1000); 6452 } 6453 6454 gro_normal_list(n); 6455 6456 /* Some drivers may have called napi_schedule 6457 * prior to exhausting their budget. 6458 */ 6459 if (unlikely(!list_empty(&n->poll_list))) { 6460 pr_warn_once("%s: Budget exhausted after napi rescheduled\n", 6461 n->dev ? n->dev->name : "backlog"); 6462 return work; 6463 } 6464 6465 *repoll = true; 6466 6467 return work; 6468 } 6469 6470 static int napi_poll(struct napi_struct *n, struct list_head *repoll) 6471 { 6472 bool do_repoll = false; 6473 void *have; 6474 int work; 6475 6476 list_del_init(&n->poll_list); 6477 6478 have = netpoll_poll_lock(n); 6479 6480 work = __napi_poll(n, &do_repoll); 6481 6482 if (do_repoll) 6483 list_add_tail(&n->poll_list, repoll); 6484 6485 netpoll_poll_unlock(have); 6486 6487 return work; 6488 } 6489 6490 static int napi_thread_wait(struct napi_struct *napi) 6491 { 6492 bool woken = false; 6493 6494 set_current_state(TASK_INTERRUPTIBLE); 6495 6496 while (!kthread_should_stop()) { 6497 /* Testing SCHED_THREADED bit here to make sure the current 6498 * kthread owns this napi and could poll on this napi. 6499 * Testing SCHED bit is not enough because SCHED bit might be 6500 * set by some other busy poll thread or by napi_disable(). 6501 */ 6502 if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state) || woken) { 6503 WARN_ON(!list_empty(&napi->poll_list)); 6504 __set_current_state(TASK_RUNNING); 6505 return 0; 6506 } 6507 6508 schedule(); 6509 /* woken being true indicates this thread owns this napi. */ 6510 woken = true; 6511 set_current_state(TASK_INTERRUPTIBLE); 6512 } 6513 __set_current_state(TASK_RUNNING); 6514 6515 return -1; 6516 } 6517 6518 static int napi_threaded_poll(void *data) 6519 { 6520 struct napi_struct *napi = data; 6521 void *have; 6522 6523 while (!napi_thread_wait(napi)) { 6524 for (;;) { 6525 bool repoll = false; 6526 6527 local_bh_disable(); 6528 6529 have = netpoll_poll_lock(napi); 6530 __napi_poll(napi, &repoll); 6531 netpoll_poll_unlock(have); 6532 6533 local_bh_enable(); 6534 6535 if (!repoll) 6536 break; 6537 6538 cond_resched(); 6539 } 6540 } 6541 return 0; 6542 } 6543 6544 static __latent_entropy void net_rx_action(struct softirq_action *h) 6545 { 6546 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 6547 unsigned long time_limit = jiffies + 6548 usecs_to_jiffies(netdev_budget_usecs); 6549 int budget = netdev_budget; 6550 LIST_HEAD(list); 6551 LIST_HEAD(repoll); 6552 6553 local_irq_disable(); 6554 list_splice_init(&sd->poll_list, &list); 6555 local_irq_enable(); 6556 6557 for (;;) { 6558 struct napi_struct *n; 6559 6560 if (list_empty(&list)) { 6561 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll)) 6562 return; 6563 break; 6564 } 6565 6566 n = list_first_entry(&list, struct napi_struct, poll_list); 6567 budget -= napi_poll(n, &repoll); 6568 6569 /* If softirq window is exhausted then punt. 6570 * Allow this to run for 2 jiffies since which will allow 6571 * an average latency of 1.5/HZ. 6572 */ 6573 if (unlikely(budget <= 0 || 6574 time_after_eq(jiffies, time_limit))) { 6575 sd->time_squeeze++; 6576 break; 6577 } 6578 } 6579 6580 local_irq_disable(); 6581 6582 list_splice_tail_init(&sd->poll_list, &list); 6583 list_splice_tail(&repoll, &list); 6584 list_splice(&list, &sd->poll_list); 6585 if (!list_empty(&sd->poll_list)) 6586 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 6587 6588 net_rps_action_and_irq_enable(sd); 6589 } 6590 6591 struct netdev_adjacent { 6592 struct net_device *dev; 6593 netdevice_tracker dev_tracker; 6594 6595 /* upper master flag, there can only be one master device per list */ 6596 bool master; 6597 6598 /* lookup ignore flag */ 6599 bool ignore; 6600 6601 /* counter for the number of times this device was added to us */ 6602 u16 ref_nr; 6603 6604 /* private field for the users */ 6605 void *private; 6606 6607 struct list_head list; 6608 struct rcu_head rcu; 6609 }; 6610 6611 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, 6612 struct list_head *adj_list) 6613 { 6614 struct netdev_adjacent *adj; 6615 6616 list_for_each_entry(adj, adj_list, list) { 6617 if (adj->dev == adj_dev) 6618 return adj; 6619 } 6620 return NULL; 6621 } 6622 6623 static int ____netdev_has_upper_dev(struct net_device *upper_dev, 6624 struct netdev_nested_priv *priv) 6625 { 6626 struct net_device *dev = (struct net_device *)priv->data; 6627 6628 return upper_dev == dev; 6629 } 6630 6631 /** 6632 * netdev_has_upper_dev - Check if device is linked to an upper device 6633 * @dev: device 6634 * @upper_dev: upper device to check 6635 * 6636 * Find out if a device is linked to specified upper device and return true 6637 * in case it is. Note that this checks only immediate upper device, 6638 * not through a complete stack of devices. The caller must hold the RTNL lock. 6639 */ 6640 bool netdev_has_upper_dev(struct net_device *dev, 6641 struct net_device *upper_dev) 6642 { 6643 struct netdev_nested_priv priv = { 6644 .data = (void *)upper_dev, 6645 }; 6646 6647 ASSERT_RTNL(); 6648 6649 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 6650 &priv); 6651 } 6652 EXPORT_SYMBOL(netdev_has_upper_dev); 6653 6654 /** 6655 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device 6656 * @dev: device 6657 * @upper_dev: upper device to check 6658 * 6659 * Find out if a device is linked to specified upper device and return true 6660 * in case it is. Note that this checks the entire upper device chain. 6661 * The caller must hold rcu lock. 6662 */ 6663 6664 bool netdev_has_upper_dev_all_rcu(struct net_device *dev, 6665 struct net_device *upper_dev) 6666 { 6667 struct netdev_nested_priv priv = { 6668 .data = (void *)upper_dev, 6669 }; 6670 6671 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 6672 &priv); 6673 } 6674 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); 6675 6676 /** 6677 * netdev_has_any_upper_dev - Check if device is linked to some device 6678 * @dev: device 6679 * 6680 * Find out if a device is linked to an upper device and return true in case 6681 * it is. The caller must hold the RTNL lock. 6682 */ 6683 bool netdev_has_any_upper_dev(struct net_device *dev) 6684 { 6685 ASSERT_RTNL(); 6686 6687 return !list_empty(&dev->adj_list.upper); 6688 } 6689 EXPORT_SYMBOL(netdev_has_any_upper_dev); 6690 6691 /** 6692 * netdev_master_upper_dev_get - Get master upper device 6693 * @dev: device 6694 * 6695 * Find a master upper device and return pointer to it or NULL in case 6696 * it's not there. The caller must hold the RTNL lock. 6697 */ 6698 struct net_device *netdev_master_upper_dev_get(struct net_device *dev) 6699 { 6700 struct netdev_adjacent *upper; 6701 6702 ASSERT_RTNL(); 6703 6704 if (list_empty(&dev->adj_list.upper)) 6705 return NULL; 6706 6707 upper = list_first_entry(&dev->adj_list.upper, 6708 struct netdev_adjacent, list); 6709 if (likely(upper->master)) 6710 return upper->dev; 6711 return NULL; 6712 } 6713 EXPORT_SYMBOL(netdev_master_upper_dev_get); 6714 6715 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev) 6716 { 6717 struct netdev_adjacent *upper; 6718 6719 ASSERT_RTNL(); 6720 6721 if (list_empty(&dev->adj_list.upper)) 6722 return NULL; 6723 6724 upper = list_first_entry(&dev->adj_list.upper, 6725 struct netdev_adjacent, list); 6726 if (likely(upper->master) && !upper->ignore) 6727 return upper->dev; 6728 return NULL; 6729 } 6730 6731 /** 6732 * netdev_has_any_lower_dev - Check if device is linked to some device 6733 * @dev: device 6734 * 6735 * Find out if a device is linked to a lower device and return true in case 6736 * it is. The caller must hold the RTNL lock. 6737 */ 6738 static bool netdev_has_any_lower_dev(struct net_device *dev) 6739 { 6740 ASSERT_RTNL(); 6741 6742 return !list_empty(&dev->adj_list.lower); 6743 } 6744 6745 void *netdev_adjacent_get_private(struct list_head *adj_list) 6746 { 6747 struct netdev_adjacent *adj; 6748 6749 adj = list_entry(adj_list, struct netdev_adjacent, list); 6750 6751 return adj->private; 6752 } 6753 EXPORT_SYMBOL(netdev_adjacent_get_private); 6754 6755 /** 6756 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list 6757 * @dev: device 6758 * @iter: list_head ** of the current position 6759 * 6760 * Gets the next device from the dev's upper list, starting from iter 6761 * position. The caller must hold RCU read lock. 6762 */ 6763 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, 6764 struct list_head **iter) 6765 { 6766 struct netdev_adjacent *upper; 6767 6768 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 6769 6770 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 6771 6772 if (&upper->list == &dev->adj_list.upper) 6773 return NULL; 6774 6775 *iter = &upper->list; 6776 6777 return upper->dev; 6778 } 6779 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); 6780 6781 static struct net_device *__netdev_next_upper_dev(struct net_device *dev, 6782 struct list_head **iter, 6783 bool *ignore) 6784 { 6785 struct netdev_adjacent *upper; 6786 6787 upper = list_entry((*iter)->next, struct netdev_adjacent, list); 6788 6789 if (&upper->list == &dev->adj_list.upper) 6790 return NULL; 6791 6792 *iter = &upper->list; 6793 *ignore = upper->ignore; 6794 6795 return upper->dev; 6796 } 6797 6798 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, 6799 struct list_head **iter) 6800 { 6801 struct netdev_adjacent *upper; 6802 6803 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 6804 6805 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 6806 6807 if (&upper->list == &dev->adj_list.upper) 6808 return NULL; 6809 6810 *iter = &upper->list; 6811 6812 return upper->dev; 6813 } 6814 6815 static int __netdev_walk_all_upper_dev(struct net_device *dev, 6816 int (*fn)(struct net_device *dev, 6817 struct netdev_nested_priv *priv), 6818 struct netdev_nested_priv *priv) 6819 { 6820 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 6821 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 6822 int ret, cur = 0; 6823 bool ignore; 6824 6825 now = dev; 6826 iter = &dev->adj_list.upper; 6827 6828 while (1) { 6829 if (now != dev) { 6830 ret = fn(now, priv); 6831 if (ret) 6832 return ret; 6833 } 6834 6835 next = NULL; 6836 while (1) { 6837 udev = __netdev_next_upper_dev(now, &iter, &ignore); 6838 if (!udev) 6839 break; 6840 if (ignore) 6841 continue; 6842 6843 next = udev; 6844 niter = &udev->adj_list.upper; 6845 dev_stack[cur] = now; 6846 iter_stack[cur++] = iter; 6847 break; 6848 } 6849 6850 if (!next) { 6851 if (!cur) 6852 return 0; 6853 next = dev_stack[--cur]; 6854 niter = iter_stack[cur]; 6855 } 6856 6857 now = next; 6858 iter = niter; 6859 } 6860 6861 return 0; 6862 } 6863 6864 int netdev_walk_all_upper_dev_rcu(struct net_device *dev, 6865 int (*fn)(struct net_device *dev, 6866 struct netdev_nested_priv *priv), 6867 struct netdev_nested_priv *priv) 6868 { 6869 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 6870 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 6871 int ret, cur = 0; 6872 6873 now = dev; 6874 iter = &dev->adj_list.upper; 6875 6876 while (1) { 6877 if (now != dev) { 6878 ret = fn(now, priv); 6879 if (ret) 6880 return ret; 6881 } 6882 6883 next = NULL; 6884 while (1) { 6885 udev = netdev_next_upper_dev_rcu(now, &iter); 6886 if (!udev) 6887 break; 6888 6889 next = udev; 6890 niter = &udev->adj_list.upper; 6891 dev_stack[cur] = now; 6892 iter_stack[cur++] = iter; 6893 break; 6894 } 6895 6896 if (!next) { 6897 if (!cur) 6898 return 0; 6899 next = dev_stack[--cur]; 6900 niter = iter_stack[cur]; 6901 } 6902 6903 now = next; 6904 iter = niter; 6905 } 6906 6907 return 0; 6908 } 6909 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); 6910 6911 static bool __netdev_has_upper_dev(struct net_device *dev, 6912 struct net_device *upper_dev) 6913 { 6914 struct netdev_nested_priv priv = { 6915 .flags = 0, 6916 .data = (void *)upper_dev, 6917 }; 6918 6919 ASSERT_RTNL(); 6920 6921 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev, 6922 &priv); 6923 } 6924 6925 /** 6926 * netdev_lower_get_next_private - Get the next ->private from the 6927 * lower neighbour list 6928 * @dev: device 6929 * @iter: list_head ** of the current position 6930 * 6931 * Gets the next netdev_adjacent->private from the dev's lower neighbour 6932 * list, starting from iter position. The caller must hold either hold the 6933 * RTNL lock or its own locking that guarantees that the neighbour lower 6934 * list will remain unchanged. 6935 */ 6936 void *netdev_lower_get_next_private(struct net_device *dev, 6937 struct list_head **iter) 6938 { 6939 struct netdev_adjacent *lower; 6940 6941 lower = list_entry(*iter, struct netdev_adjacent, list); 6942 6943 if (&lower->list == &dev->adj_list.lower) 6944 return NULL; 6945 6946 *iter = lower->list.next; 6947 6948 return lower->private; 6949 } 6950 EXPORT_SYMBOL(netdev_lower_get_next_private); 6951 6952 /** 6953 * netdev_lower_get_next_private_rcu - Get the next ->private from the 6954 * lower neighbour list, RCU 6955 * variant 6956 * @dev: device 6957 * @iter: list_head ** of the current position 6958 * 6959 * Gets the next netdev_adjacent->private from the dev's lower neighbour 6960 * list, starting from iter position. The caller must hold RCU read lock. 6961 */ 6962 void *netdev_lower_get_next_private_rcu(struct net_device *dev, 6963 struct list_head **iter) 6964 { 6965 struct netdev_adjacent *lower; 6966 6967 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 6968 6969 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 6970 6971 if (&lower->list == &dev->adj_list.lower) 6972 return NULL; 6973 6974 *iter = &lower->list; 6975 6976 return lower->private; 6977 } 6978 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); 6979 6980 /** 6981 * netdev_lower_get_next - Get the next device from the lower neighbour 6982 * list 6983 * @dev: device 6984 * @iter: list_head ** of the current position 6985 * 6986 * Gets the next netdev_adjacent from the dev's lower neighbour 6987 * list, starting from iter position. The caller must hold RTNL lock or 6988 * its own locking that guarantees that the neighbour lower 6989 * list will remain unchanged. 6990 */ 6991 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) 6992 { 6993 struct netdev_adjacent *lower; 6994 6995 lower = list_entry(*iter, struct netdev_adjacent, list); 6996 6997 if (&lower->list == &dev->adj_list.lower) 6998 return NULL; 6999 7000 *iter = lower->list.next; 7001 7002 return lower->dev; 7003 } 7004 EXPORT_SYMBOL(netdev_lower_get_next); 7005 7006 static struct net_device *netdev_next_lower_dev(struct net_device *dev, 7007 struct list_head **iter) 7008 { 7009 struct netdev_adjacent *lower; 7010 7011 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7012 7013 if (&lower->list == &dev->adj_list.lower) 7014 return NULL; 7015 7016 *iter = &lower->list; 7017 7018 return lower->dev; 7019 } 7020 7021 static struct net_device *__netdev_next_lower_dev(struct net_device *dev, 7022 struct list_head **iter, 7023 bool *ignore) 7024 { 7025 struct netdev_adjacent *lower; 7026 7027 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7028 7029 if (&lower->list == &dev->adj_list.lower) 7030 return NULL; 7031 7032 *iter = &lower->list; 7033 *ignore = lower->ignore; 7034 7035 return lower->dev; 7036 } 7037 7038 int netdev_walk_all_lower_dev(struct net_device *dev, 7039 int (*fn)(struct net_device *dev, 7040 struct netdev_nested_priv *priv), 7041 struct netdev_nested_priv *priv) 7042 { 7043 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7044 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7045 int ret, cur = 0; 7046 7047 now = dev; 7048 iter = &dev->adj_list.lower; 7049 7050 while (1) { 7051 if (now != dev) { 7052 ret = fn(now, priv); 7053 if (ret) 7054 return ret; 7055 } 7056 7057 next = NULL; 7058 while (1) { 7059 ldev = netdev_next_lower_dev(now, &iter); 7060 if (!ldev) 7061 break; 7062 7063 next = ldev; 7064 niter = &ldev->adj_list.lower; 7065 dev_stack[cur] = now; 7066 iter_stack[cur++] = iter; 7067 break; 7068 } 7069 7070 if (!next) { 7071 if (!cur) 7072 return 0; 7073 next = dev_stack[--cur]; 7074 niter = iter_stack[cur]; 7075 } 7076 7077 now = next; 7078 iter = niter; 7079 } 7080 7081 return 0; 7082 } 7083 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); 7084 7085 static int __netdev_walk_all_lower_dev(struct net_device *dev, 7086 int (*fn)(struct net_device *dev, 7087 struct netdev_nested_priv *priv), 7088 struct netdev_nested_priv *priv) 7089 { 7090 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7091 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7092 int ret, cur = 0; 7093 bool ignore; 7094 7095 now = dev; 7096 iter = &dev->adj_list.lower; 7097 7098 while (1) { 7099 if (now != dev) { 7100 ret = fn(now, priv); 7101 if (ret) 7102 return ret; 7103 } 7104 7105 next = NULL; 7106 while (1) { 7107 ldev = __netdev_next_lower_dev(now, &iter, &ignore); 7108 if (!ldev) 7109 break; 7110 if (ignore) 7111 continue; 7112 7113 next = ldev; 7114 niter = &ldev->adj_list.lower; 7115 dev_stack[cur] = now; 7116 iter_stack[cur++] = iter; 7117 break; 7118 } 7119 7120 if (!next) { 7121 if (!cur) 7122 return 0; 7123 next = dev_stack[--cur]; 7124 niter = iter_stack[cur]; 7125 } 7126 7127 now = next; 7128 iter = niter; 7129 } 7130 7131 return 0; 7132 } 7133 7134 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, 7135 struct list_head **iter) 7136 { 7137 struct netdev_adjacent *lower; 7138 7139 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7140 if (&lower->list == &dev->adj_list.lower) 7141 return NULL; 7142 7143 *iter = &lower->list; 7144 7145 return lower->dev; 7146 } 7147 EXPORT_SYMBOL(netdev_next_lower_dev_rcu); 7148 7149 static u8 __netdev_upper_depth(struct net_device *dev) 7150 { 7151 struct net_device *udev; 7152 struct list_head *iter; 7153 u8 max_depth = 0; 7154 bool ignore; 7155 7156 for (iter = &dev->adj_list.upper, 7157 udev = __netdev_next_upper_dev(dev, &iter, &ignore); 7158 udev; 7159 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) { 7160 if (ignore) 7161 continue; 7162 if (max_depth < udev->upper_level) 7163 max_depth = udev->upper_level; 7164 } 7165 7166 return max_depth; 7167 } 7168 7169 static u8 __netdev_lower_depth(struct net_device *dev) 7170 { 7171 struct net_device *ldev; 7172 struct list_head *iter; 7173 u8 max_depth = 0; 7174 bool ignore; 7175 7176 for (iter = &dev->adj_list.lower, 7177 ldev = __netdev_next_lower_dev(dev, &iter, &ignore); 7178 ldev; 7179 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) { 7180 if (ignore) 7181 continue; 7182 if (max_depth < ldev->lower_level) 7183 max_depth = ldev->lower_level; 7184 } 7185 7186 return max_depth; 7187 } 7188 7189 static int __netdev_update_upper_level(struct net_device *dev, 7190 struct netdev_nested_priv *__unused) 7191 { 7192 dev->upper_level = __netdev_upper_depth(dev) + 1; 7193 return 0; 7194 } 7195 7196 #ifdef CONFIG_LOCKDEP 7197 static LIST_HEAD(net_unlink_list); 7198 7199 static void net_unlink_todo(struct net_device *dev) 7200 { 7201 if (list_empty(&dev->unlink_list)) 7202 list_add_tail(&dev->unlink_list, &net_unlink_list); 7203 } 7204 #endif 7205 7206 static int __netdev_update_lower_level(struct net_device *dev, 7207 struct netdev_nested_priv *priv) 7208 { 7209 dev->lower_level = __netdev_lower_depth(dev) + 1; 7210 7211 #ifdef CONFIG_LOCKDEP 7212 if (!priv) 7213 return 0; 7214 7215 if (priv->flags & NESTED_SYNC_IMM) 7216 dev->nested_level = dev->lower_level - 1; 7217 if (priv->flags & NESTED_SYNC_TODO) 7218 net_unlink_todo(dev); 7219 #endif 7220 return 0; 7221 } 7222 7223 int netdev_walk_all_lower_dev_rcu(struct net_device *dev, 7224 int (*fn)(struct net_device *dev, 7225 struct netdev_nested_priv *priv), 7226 struct netdev_nested_priv *priv) 7227 { 7228 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7229 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7230 int ret, cur = 0; 7231 7232 now = dev; 7233 iter = &dev->adj_list.lower; 7234 7235 while (1) { 7236 if (now != dev) { 7237 ret = fn(now, priv); 7238 if (ret) 7239 return ret; 7240 } 7241 7242 next = NULL; 7243 while (1) { 7244 ldev = netdev_next_lower_dev_rcu(now, &iter); 7245 if (!ldev) 7246 break; 7247 7248 next = ldev; 7249 niter = &ldev->adj_list.lower; 7250 dev_stack[cur] = now; 7251 iter_stack[cur++] = iter; 7252 break; 7253 } 7254 7255 if (!next) { 7256 if (!cur) 7257 return 0; 7258 next = dev_stack[--cur]; 7259 niter = iter_stack[cur]; 7260 } 7261 7262 now = next; 7263 iter = niter; 7264 } 7265 7266 return 0; 7267 } 7268 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); 7269 7270 /** 7271 * netdev_lower_get_first_private_rcu - Get the first ->private from the 7272 * lower neighbour list, RCU 7273 * variant 7274 * @dev: device 7275 * 7276 * Gets the first netdev_adjacent->private from the dev's lower neighbour 7277 * list. The caller must hold RCU read lock. 7278 */ 7279 void *netdev_lower_get_first_private_rcu(struct net_device *dev) 7280 { 7281 struct netdev_adjacent *lower; 7282 7283 lower = list_first_or_null_rcu(&dev->adj_list.lower, 7284 struct netdev_adjacent, list); 7285 if (lower) 7286 return lower->private; 7287 return NULL; 7288 } 7289 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); 7290 7291 /** 7292 * netdev_master_upper_dev_get_rcu - Get master upper device 7293 * @dev: device 7294 * 7295 * Find a master upper device and return pointer to it or NULL in case 7296 * it's not there. The caller must hold the RCU read lock. 7297 */ 7298 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) 7299 { 7300 struct netdev_adjacent *upper; 7301 7302 upper = list_first_or_null_rcu(&dev->adj_list.upper, 7303 struct netdev_adjacent, list); 7304 if (upper && likely(upper->master)) 7305 return upper->dev; 7306 return NULL; 7307 } 7308 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); 7309 7310 static int netdev_adjacent_sysfs_add(struct net_device *dev, 7311 struct net_device *adj_dev, 7312 struct list_head *dev_list) 7313 { 7314 char linkname[IFNAMSIZ+7]; 7315 7316 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7317 "upper_%s" : "lower_%s", adj_dev->name); 7318 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), 7319 linkname); 7320 } 7321 static void netdev_adjacent_sysfs_del(struct net_device *dev, 7322 char *name, 7323 struct list_head *dev_list) 7324 { 7325 char linkname[IFNAMSIZ+7]; 7326 7327 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7328 "upper_%s" : "lower_%s", name); 7329 sysfs_remove_link(&(dev->dev.kobj), linkname); 7330 } 7331 7332 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, 7333 struct net_device *adj_dev, 7334 struct list_head *dev_list) 7335 { 7336 return (dev_list == &dev->adj_list.upper || 7337 dev_list == &dev->adj_list.lower) && 7338 net_eq(dev_net(dev), dev_net(adj_dev)); 7339 } 7340 7341 static int __netdev_adjacent_dev_insert(struct net_device *dev, 7342 struct net_device *adj_dev, 7343 struct list_head *dev_list, 7344 void *private, bool master) 7345 { 7346 struct netdev_adjacent *adj; 7347 int ret; 7348 7349 adj = __netdev_find_adj(adj_dev, dev_list); 7350 7351 if (adj) { 7352 adj->ref_nr += 1; 7353 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", 7354 dev->name, adj_dev->name, adj->ref_nr); 7355 7356 return 0; 7357 } 7358 7359 adj = kmalloc(sizeof(*adj), GFP_KERNEL); 7360 if (!adj) 7361 return -ENOMEM; 7362 7363 adj->dev = adj_dev; 7364 adj->master = master; 7365 adj->ref_nr = 1; 7366 adj->private = private; 7367 adj->ignore = false; 7368 dev_hold_track(adj_dev, &adj->dev_tracker, GFP_KERNEL); 7369 7370 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", 7371 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); 7372 7373 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { 7374 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); 7375 if (ret) 7376 goto free_adj; 7377 } 7378 7379 /* Ensure that master link is always the first item in list. */ 7380 if (master) { 7381 ret = sysfs_create_link(&(dev->dev.kobj), 7382 &(adj_dev->dev.kobj), "master"); 7383 if (ret) 7384 goto remove_symlinks; 7385 7386 list_add_rcu(&adj->list, dev_list); 7387 } else { 7388 list_add_tail_rcu(&adj->list, dev_list); 7389 } 7390 7391 return 0; 7392 7393 remove_symlinks: 7394 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7395 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7396 free_adj: 7397 dev_put_track(adj_dev, &adj->dev_tracker); 7398 kfree(adj); 7399 7400 return ret; 7401 } 7402 7403 static void __netdev_adjacent_dev_remove(struct net_device *dev, 7404 struct net_device *adj_dev, 7405 u16 ref_nr, 7406 struct list_head *dev_list) 7407 { 7408 struct netdev_adjacent *adj; 7409 7410 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", 7411 dev->name, adj_dev->name, ref_nr); 7412 7413 adj = __netdev_find_adj(adj_dev, dev_list); 7414 7415 if (!adj) { 7416 pr_err("Adjacency does not exist for device %s from %s\n", 7417 dev->name, adj_dev->name); 7418 WARN_ON(1); 7419 return; 7420 } 7421 7422 if (adj->ref_nr > ref_nr) { 7423 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", 7424 dev->name, adj_dev->name, ref_nr, 7425 adj->ref_nr - ref_nr); 7426 adj->ref_nr -= ref_nr; 7427 return; 7428 } 7429 7430 if (adj->master) 7431 sysfs_remove_link(&(dev->dev.kobj), "master"); 7432 7433 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7434 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7435 7436 list_del_rcu(&adj->list); 7437 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", 7438 adj_dev->name, dev->name, adj_dev->name); 7439 dev_put_track(adj_dev, &adj->dev_tracker); 7440 kfree_rcu(adj, rcu); 7441 } 7442 7443 static int __netdev_adjacent_dev_link_lists(struct net_device *dev, 7444 struct net_device *upper_dev, 7445 struct list_head *up_list, 7446 struct list_head *down_list, 7447 void *private, bool master) 7448 { 7449 int ret; 7450 7451 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, 7452 private, master); 7453 if (ret) 7454 return ret; 7455 7456 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, 7457 private, false); 7458 if (ret) { 7459 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); 7460 return ret; 7461 } 7462 7463 return 0; 7464 } 7465 7466 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, 7467 struct net_device *upper_dev, 7468 u16 ref_nr, 7469 struct list_head *up_list, 7470 struct list_head *down_list) 7471 { 7472 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); 7473 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); 7474 } 7475 7476 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, 7477 struct net_device *upper_dev, 7478 void *private, bool master) 7479 { 7480 return __netdev_adjacent_dev_link_lists(dev, upper_dev, 7481 &dev->adj_list.upper, 7482 &upper_dev->adj_list.lower, 7483 private, master); 7484 } 7485 7486 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, 7487 struct net_device *upper_dev) 7488 { 7489 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, 7490 &dev->adj_list.upper, 7491 &upper_dev->adj_list.lower); 7492 } 7493 7494 static int __netdev_upper_dev_link(struct net_device *dev, 7495 struct net_device *upper_dev, bool master, 7496 void *upper_priv, void *upper_info, 7497 struct netdev_nested_priv *priv, 7498 struct netlink_ext_ack *extack) 7499 { 7500 struct netdev_notifier_changeupper_info changeupper_info = { 7501 .info = { 7502 .dev = dev, 7503 .extack = extack, 7504 }, 7505 .upper_dev = upper_dev, 7506 .master = master, 7507 .linking = true, 7508 .upper_info = upper_info, 7509 }; 7510 struct net_device *master_dev; 7511 int ret = 0; 7512 7513 ASSERT_RTNL(); 7514 7515 if (dev == upper_dev) 7516 return -EBUSY; 7517 7518 /* To prevent loops, check if dev is not upper device to upper_dev. */ 7519 if (__netdev_has_upper_dev(upper_dev, dev)) 7520 return -EBUSY; 7521 7522 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV) 7523 return -EMLINK; 7524 7525 if (!master) { 7526 if (__netdev_has_upper_dev(dev, upper_dev)) 7527 return -EEXIST; 7528 } else { 7529 master_dev = __netdev_master_upper_dev_get(dev); 7530 if (master_dev) 7531 return master_dev == upper_dev ? -EEXIST : -EBUSY; 7532 } 7533 7534 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7535 &changeupper_info.info); 7536 ret = notifier_to_errno(ret); 7537 if (ret) 7538 return ret; 7539 7540 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, 7541 master); 7542 if (ret) 7543 return ret; 7544 7545 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 7546 &changeupper_info.info); 7547 ret = notifier_to_errno(ret); 7548 if (ret) 7549 goto rollback; 7550 7551 __netdev_update_upper_level(dev, NULL); 7552 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 7553 7554 __netdev_update_lower_level(upper_dev, priv); 7555 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 7556 priv); 7557 7558 return 0; 7559 7560 rollback: 7561 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 7562 7563 return ret; 7564 } 7565 7566 /** 7567 * netdev_upper_dev_link - Add a link to the upper device 7568 * @dev: device 7569 * @upper_dev: new upper device 7570 * @extack: netlink extended ack 7571 * 7572 * Adds a link to device which is upper to this one. The caller must hold 7573 * the RTNL lock. On a failure a negative errno code is returned. 7574 * On success the reference counts are adjusted and the function 7575 * returns zero. 7576 */ 7577 int netdev_upper_dev_link(struct net_device *dev, 7578 struct net_device *upper_dev, 7579 struct netlink_ext_ack *extack) 7580 { 7581 struct netdev_nested_priv priv = { 7582 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7583 .data = NULL, 7584 }; 7585 7586 return __netdev_upper_dev_link(dev, upper_dev, false, 7587 NULL, NULL, &priv, extack); 7588 } 7589 EXPORT_SYMBOL(netdev_upper_dev_link); 7590 7591 /** 7592 * netdev_master_upper_dev_link - Add a master link to the upper device 7593 * @dev: device 7594 * @upper_dev: new upper device 7595 * @upper_priv: upper device private 7596 * @upper_info: upper info to be passed down via notifier 7597 * @extack: netlink extended ack 7598 * 7599 * Adds a link to device which is upper to this one. In this case, only 7600 * one master upper device can be linked, although other non-master devices 7601 * might be linked as well. The caller must hold the RTNL lock. 7602 * On a failure a negative errno code is returned. On success the reference 7603 * counts are adjusted and the function returns zero. 7604 */ 7605 int netdev_master_upper_dev_link(struct net_device *dev, 7606 struct net_device *upper_dev, 7607 void *upper_priv, void *upper_info, 7608 struct netlink_ext_ack *extack) 7609 { 7610 struct netdev_nested_priv priv = { 7611 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7612 .data = NULL, 7613 }; 7614 7615 return __netdev_upper_dev_link(dev, upper_dev, true, 7616 upper_priv, upper_info, &priv, extack); 7617 } 7618 EXPORT_SYMBOL(netdev_master_upper_dev_link); 7619 7620 static void __netdev_upper_dev_unlink(struct net_device *dev, 7621 struct net_device *upper_dev, 7622 struct netdev_nested_priv *priv) 7623 { 7624 struct netdev_notifier_changeupper_info changeupper_info = { 7625 .info = { 7626 .dev = dev, 7627 }, 7628 .upper_dev = upper_dev, 7629 .linking = false, 7630 }; 7631 7632 ASSERT_RTNL(); 7633 7634 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; 7635 7636 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7637 &changeupper_info.info); 7638 7639 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 7640 7641 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 7642 &changeupper_info.info); 7643 7644 __netdev_update_upper_level(dev, NULL); 7645 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 7646 7647 __netdev_update_lower_level(upper_dev, priv); 7648 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 7649 priv); 7650 } 7651 7652 /** 7653 * netdev_upper_dev_unlink - Removes a link to upper device 7654 * @dev: device 7655 * @upper_dev: new upper device 7656 * 7657 * Removes a link to device which is upper to this one. The caller must hold 7658 * the RTNL lock. 7659 */ 7660 void netdev_upper_dev_unlink(struct net_device *dev, 7661 struct net_device *upper_dev) 7662 { 7663 struct netdev_nested_priv priv = { 7664 .flags = NESTED_SYNC_TODO, 7665 .data = NULL, 7666 }; 7667 7668 __netdev_upper_dev_unlink(dev, upper_dev, &priv); 7669 } 7670 EXPORT_SYMBOL(netdev_upper_dev_unlink); 7671 7672 static void __netdev_adjacent_dev_set(struct net_device *upper_dev, 7673 struct net_device *lower_dev, 7674 bool val) 7675 { 7676 struct netdev_adjacent *adj; 7677 7678 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower); 7679 if (adj) 7680 adj->ignore = val; 7681 7682 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper); 7683 if (adj) 7684 adj->ignore = val; 7685 } 7686 7687 static void netdev_adjacent_dev_disable(struct net_device *upper_dev, 7688 struct net_device *lower_dev) 7689 { 7690 __netdev_adjacent_dev_set(upper_dev, lower_dev, true); 7691 } 7692 7693 static void netdev_adjacent_dev_enable(struct net_device *upper_dev, 7694 struct net_device *lower_dev) 7695 { 7696 __netdev_adjacent_dev_set(upper_dev, lower_dev, false); 7697 } 7698 7699 int netdev_adjacent_change_prepare(struct net_device *old_dev, 7700 struct net_device *new_dev, 7701 struct net_device *dev, 7702 struct netlink_ext_ack *extack) 7703 { 7704 struct netdev_nested_priv priv = { 7705 .flags = 0, 7706 .data = NULL, 7707 }; 7708 int err; 7709 7710 if (!new_dev) 7711 return 0; 7712 7713 if (old_dev && new_dev != old_dev) 7714 netdev_adjacent_dev_disable(dev, old_dev); 7715 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv, 7716 extack); 7717 if (err) { 7718 if (old_dev && new_dev != old_dev) 7719 netdev_adjacent_dev_enable(dev, old_dev); 7720 return err; 7721 } 7722 7723 return 0; 7724 } 7725 EXPORT_SYMBOL(netdev_adjacent_change_prepare); 7726 7727 void netdev_adjacent_change_commit(struct net_device *old_dev, 7728 struct net_device *new_dev, 7729 struct net_device *dev) 7730 { 7731 struct netdev_nested_priv priv = { 7732 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7733 .data = NULL, 7734 }; 7735 7736 if (!new_dev || !old_dev) 7737 return; 7738 7739 if (new_dev == old_dev) 7740 return; 7741 7742 netdev_adjacent_dev_enable(dev, old_dev); 7743 __netdev_upper_dev_unlink(old_dev, dev, &priv); 7744 } 7745 EXPORT_SYMBOL(netdev_adjacent_change_commit); 7746 7747 void netdev_adjacent_change_abort(struct net_device *old_dev, 7748 struct net_device *new_dev, 7749 struct net_device *dev) 7750 { 7751 struct netdev_nested_priv priv = { 7752 .flags = 0, 7753 .data = NULL, 7754 }; 7755 7756 if (!new_dev) 7757 return; 7758 7759 if (old_dev && new_dev != old_dev) 7760 netdev_adjacent_dev_enable(dev, old_dev); 7761 7762 __netdev_upper_dev_unlink(new_dev, dev, &priv); 7763 } 7764 EXPORT_SYMBOL(netdev_adjacent_change_abort); 7765 7766 /** 7767 * netdev_bonding_info_change - Dispatch event about slave change 7768 * @dev: device 7769 * @bonding_info: info to dispatch 7770 * 7771 * Send NETDEV_BONDING_INFO to netdev notifiers with info. 7772 * The caller must hold the RTNL lock. 7773 */ 7774 void netdev_bonding_info_change(struct net_device *dev, 7775 struct netdev_bonding_info *bonding_info) 7776 { 7777 struct netdev_notifier_bonding_info info = { 7778 .info.dev = dev, 7779 }; 7780 7781 memcpy(&info.bonding_info, bonding_info, 7782 sizeof(struct netdev_bonding_info)); 7783 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, 7784 &info.info); 7785 } 7786 EXPORT_SYMBOL(netdev_bonding_info_change); 7787 7788 static int netdev_offload_xstats_enable_l3(struct net_device *dev, 7789 struct netlink_ext_ack *extack) 7790 { 7791 struct netdev_notifier_offload_xstats_info info = { 7792 .info.dev = dev, 7793 .info.extack = extack, 7794 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, 7795 }; 7796 int err; 7797 int rc; 7798 7799 dev->offload_xstats_l3 = kzalloc(sizeof(*dev->offload_xstats_l3), 7800 GFP_KERNEL); 7801 if (!dev->offload_xstats_l3) 7802 return -ENOMEM; 7803 7804 rc = call_netdevice_notifiers_info_robust(NETDEV_OFFLOAD_XSTATS_ENABLE, 7805 NETDEV_OFFLOAD_XSTATS_DISABLE, 7806 &info.info); 7807 err = notifier_to_errno(rc); 7808 if (err) 7809 goto free_stats; 7810 7811 return 0; 7812 7813 free_stats: 7814 kfree(dev->offload_xstats_l3); 7815 dev->offload_xstats_l3 = NULL; 7816 return err; 7817 } 7818 7819 int netdev_offload_xstats_enable(struct net_device *dev, 7820 enum netdev_offload_xstats_type type, 7821 struct netlink_ext_ack *extack) 7822 { 7823 ASSERT_RTNL(); 7824 7825 if (netdev_offload_xstats_enabled(dev, type)) 7826 return -EALREADY; 7827 7828 switch (type) { 7829 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 7830 return netdev_offload_xstats_enable_l3(dev, extack); 7831 } 7832 7833 WARN_ON(1); 7834 return -EINVAL; 7835 } 7836 EXPORT_SYMBOL(netdev_offload_xstats_enable); 7837 7838 static void netdev_offload_xstats_disable_l3(struct net_device *dev) 7839 { 7840 struct netdev_notifier_offload_xstats_info info = { 7841 .info.dev = dev, 7842 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, 7843 }; 7844 7845 call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_DISABLE, 7846 &info.info); 7847 kfree(dev->offload_xstats_l3); 7848 dev->offload_xstats_l3 = NULL; 7849 } 7850 7851 int netdev_offload_xstats_disable(struct net_device *dev, 7852 enum netdev_offload_xstats_type type) 7853 { 7854 ASSERT_RTNL(); 7855 7856 if (!netdev_offload_xstats_enabled(dev, type)) 7857 return -EALREADY; 7858 7859 switch (type) { 7860 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 7861 netdev_offload_xstats_disable_l3(dev); 7862 return 0; 7863 } 7864 7865 WARN_ON(1); 7866 return -EINVAL; 7867 } 7868 EXPORT_SYMBOL(netdev_offload_xstats_disable); 7869 7870 static void netdev_offload_xstats_disable_all(struct net_device *dev) 7871 { 7872 netdev_offload_xstats_disable(dev, NETDEV_OFFLOAD_XSTATS_TYPE_L3); 7873 } 7874 7875 static struct rtnl_hw_stats64 * 7876 netdev_offload_xstats_get_ptr(const struct net_device *dev, 7877 enum netdev_offload_xstats_type type) 7878 { 7879 switch (type) { 7880 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 7881 return dev->offload_xstats_l3; 7882 } 7883 7884 WARN_ON(1); 7885 return NULL; 7886 } 7887 7888 bool netdev_offload_xstats_enabled(const struct net_device *dev, 7889 enum netdev_offload_xstats_type type) 7890 { 7891 ASSERT_RTNL(); 7892 7893 return netdev_offload_xstats_get_ptr(dev, type); 7894 } 7895 EXPORT_SYMBOL(netdev_offload_xstats_enabled); 7896 7897 struct netdev_notifier_offload_xstats_ru { 7898 bool used; 7899 }; 7900 7901 struct netdev_notifier_offload_xstats_rd { 7902 struct rtnl_hw_stats64 stats; 7903 bool used; 7904 }; 7905 7906 static void netdev_hw_stats64_add(struct rtnl_hw_stats64 *dest, 7907 const struct rtnl_hw_stats64 *src) 7908 { 7909 dest->rx_packets += src->rx_packets; 7910 dest->tx_packets += src->tx_packets; 7911 dest->rx_bytes += src->rx_bytes; 7912 dest->tx_bytes += src->tx_bytes; 7913 dest->rx_errors += src->rx_errors; 7914 dest->tx_errors += src->tx_errors; 7915 dest->rx_dropped += src->rx_dropped; 7916 dest->tx_dropped += src->tx_dropped; 7917 dest->multicast += src->multicast; 7918 } 7919 7920 static int netdev_offload_xstats_get_used(struct net_device *dev, 7921 enum netdev_offload_xstats_type type, 7922 bool *p_used, 7923 struct netlink_ext_ack *extack) 7924 { 7925 struct netdev_notifier_offload_xstats_ru report_used = {}; 7926 struct netdev_notifier_offload_xstats_info info = { 7927 .info.dev = dev, 7928 .info.extack = extack, 7929 .type = type, 7930 .report_used = &report_used, 7931 }; 7932 int rc; 7933 7934 WARN_ON(!netdev_offload_xstats_enabled(dev, type)); 7935 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_USED, 7936 &info.info); 7937 *p_used = report_used.used; 7938 return notifier_to_errno(rc); 7939 } 7940 7941 static int netdev_offload_xstats_get_stats(struct net_device *dev, 7942 enum netdev_offload_xstats_type type, 7943 struct rtnl_hw_stats64 *p_stats, 7944 bool *p_used, 7945 struct netlink_ext_ack *extack) 7946 { 7947 struct netdev_notifier_offload_xstats_rd report_delta = {}; 7948 struct netdev_notifier_offload_xstats_info info = { 7949 .info.dev = dev, 7950 .info.extack = extack, 7951 .type = type, 7952 .report_delta = &report_delta, 7953 }; 7954 struct rtnl_hw_stats64 *stats; 7955 int rc; 7956 7957 stats = netdev_offload_xstats_get_ptr(dev, type); 7958 if (WARN_ON(!stats)) 7959 return -EINVAL; 7960 7961 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_DELTA, 7962 &info.info); 7963 7964 /* Cache whatever we got, even if there was an error, otherwise the 7965 * successful stats retrievals would get lost. 7966 */ 7967 netdev_hw_stats64_add(stats, &report_delta.stats); 7968 7969 if (p_stats) 7970 *p_stats = *stats; 7971 *p_used = report_delta.used; 7972 7973 return notifier_to_errno(rc); 7974 } 7975 7976 int netdev_offload_xstats_get(struct net_device *dev, 7977 enum netdev_offload_xstats_type type, 7978 struct rtnl_hw_stats64 *p_stats, bool *p_used, 7979 struct netlink_ext_ack *extack) 7980 { 7981 ASSERT_RTNL(); 7982 7983 if (p_stats) 7984 return netdev_offload_xstats_get_stats(dev, type, p_stats, 7985 p_used, extack); 7986 else 7987 return netdev_offload_xstats_get_used(dev, type, p_used, 7988 extack); 7989 } 7990 EXPORT_SYMBOL(netdev_offload_xstats_get); 7991 7992 void 7993 netdev_offload_xstats_report_delta(struct netdev_notifier_offload_xstats_rd *report_delta, 7994 const struct rtnl_hw_stats64 *stats) 7995 { 7996 report_delta->used = true; 7997 netdev_hw_stats64_add(&report_delta->stats, stats); 7998 } 7999 EXPORT_SYMBOL(netdev_offload_xstats_report_delta); 8000 8001 void 8002 netdev_offload_xstats_report_used(struct netdev_notifier_offload_xstats_ru *report_used) 8003 { 8004 report_used->used = true; 8005 } 8006 EXPORT_SYMBOL(netdev_offload_xstats_report_used); 8007 8008 void netdev_offload_xstats_push_delta(struct net_device *dev, 8009 enum netdev_offload_xstats_type type, 8010 const struct rtnl_hw_stats64 *p_stats) 8011 { 8012 struct rtnl_hw_stats64 *stats; 8013 8014 ASSERT_RTNL(); 8015 8016 stats = netdev_offload_xstats_get_ptr(dev, type); 8017 if (WARN_ON(!stats)) 8018 return; 8019 8020 netdev_hw_stats64_add(stats, p_stats); 8021 } 8022 EXPORT_SYMBOL(netdev_offload_xstats_push_delta); 8023 8024 /** 8025 * netdev_get_xmit_slave - Get the xmit slave of master device 8026 * @dev: device 8027 * @skb: The packet 8028 * @all_slaves: assume all the slaves are active 8029 * 8030 * The reference counters are not incremented so the caller must be 8031 * careful with locks. The caller must hold RCU lock. 8032 * %NULL is returned if no slave is found. 8033 */ 8034 8035 struct net_device *netdev_get_xmit_slave(struct net_device *dev, 8036 struct sk_buff *skb, 8037 bool all_slaves) 8038 { 8039 const struct net_device_ops *ops = dev->netdev_ops; 8040 8041 if (!ops->ndo_get_xmit_slave) 8042 return NULL; 8043 return ops->ndo_get_xmit_slave(dev, skb, all_slaves); 8044 } 8045 EXPORT_SYMBOL(netdev_get_xmit_slave); 8046 8047 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev, 8048 struct sock *sk) 8049 { 8050 const struct net_device_ops *ops = dev->netdev_ops; 8051 8052 if (!ops->ndo_sk_get_lower_dev) 8053 return NULL; 8054 return ops->ndo_sk_get_lower_dev(dev, sk); 8055 } 8056 8057 /** 8058 * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket 8059 * @dev: device 8060 * @sk: the socket 8061 * 8062 * %NULL is returned if no lower device is found. 8063 */ 8064 8065 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev, 8066 struct sock *sk) 8067 { 8068 struct net_device *lower; 8069 8070 lower = netdev_sk_get_lower_dev(dev, sk); 8071 while (lower) { 8072 dev = lower; 8073 lower = netdev_sk_get_lower_dev(dev, sk); 8074 } 8075 8076 return dev; 8077 } 8078 EXPORT_SYMBOL(netdev_sk_get_lowest_dev); 8079 8080 static void netdev_adjacent_add_links(struct net_device *dev) 8081 { 8082 struct netdev_adjacent *iter; 8083 8084 struct net *net = dev_net(dev); 8085 8086 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8087 if (!net_eq(net, dev_net(iter->dev))) 8088 continue; 8089 netdev_adjacent_sysfs_add(iter->dev, dev, 8090 &iter->dev->adj_list.lower); 8091 netdev_adjacent_sysfs_add(dev, iter->dev, 8092 &dev->adj_list.upper); 8093 } 8094 8095 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8096 if (!net_eq(net, dev_net(iter->dev))) 8097 continue; 8098 netdev_adjacent_sysfs_add(iter->dev, dev, 8099 &iter->dev->adj_list.upper); 8100 netdev_adjacent_sysfs_add(dev, iter->dev, 8101 &dev->adj_list.lower); 8102 } 8103 } 8104 8105 static void netdev_adjacent_del_links(struct net_device *dev) 8106 { 8107 struct netdev_adjacent *iter; 8108 8109 struct net *net = dev_net(dev); 8110 8111 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8112 if (!net_eq(net, dev_net(iter->dev))) 8113 continue; 8114 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8115 &iter->dev->adj_list.lower); 8116 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8117 &dev->adj_list.upper); 8118 } 8119 8120 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8121 if (!net_eq(net, dev_net(iter->dev))) 8122 continue; 8123 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8124 &iter->dev->adj_list.upper); 8125 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8126 &dev->adj_list.lower); 8127 } 8128 } 8129 8130 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) 8131 { 8132 struct netdev_adjacent *iter; 8133 8134 struct net *net = dev_net(dev); 8135 8136 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8137 if (!net_eq(net, dev_net(iter->dev))) 8138 continue; 8139 netdev_adjacent_sysfs_del(iter->dev, oldname, 8140 &iter->dev->adj_list.lower); 8141 netdev_adjacent_sysfs_add(iter->dev, dev, 8142 &iter->dev->adj_list.lower); 8143 } 8144 8145 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8146 if (!net_eq(net, dev_net(iter->dev))) 8147 continue; 8148 netdev_adjacent_sysfs_del(iter->dev, oldname, 8149 &iter->dev->adj_list.upper); 8150 netdev_adjacent_sysfs_add(iter->dev, dev, 8151 &iter->dev->adj_list.upper); 8152 } 8153 } 8154 8155 void *netdev_lower_dev_get_private(struct net_device *dev, 8156 struct net_device *lower_dev) 8157 { 8158 struct netdev_adjacent *lower; 8159 8160 if (!lower_dev) 8161 return NULL; 8162 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); 8163 if (!lower) 8164 return NULL; 8165 8166 return lower->private; 8167 } 8168 EXPORT_SYMBOL(netdev_lower_dev_get_private); 8169 8170 8171 /** 8172 * netdev_lower_state_changed - Dispatch event about lower device state change 8173 * @lower_dev: device 8174 * @lower_state_info: state to dispatch 8175 * 8176 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. 8177 * The caller must hold the RTNL lock. 8178 */ 8179 void netdev_lower_state_changed(struct net_device *lower_dev, 8180 void *lower_state_info) 8181 { 8182 struct netdev_notifier_changelowerstate_info changelowerstate_info = { 8183 .info.dev = lower_dev, 8184 }; 8185 8186 ASSERT_RTNL(); 8187 changelowerstate_info.lower_state_info = lower_state_info; 8188 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, 8189 &changelowerstate_info.info); 8190 } 8191 EXPORT_SYMBOL(netdev_lower_state_changed); 8192 8193 static void dev_change_rx_flags(struct net_device *dev, int flags) 8194 { 8195 const struct net_device_ops *ops = dev->netdev_ops; 8196 8197 if (ops->ndo_change_rx_flags) 8198 ops->ndo_change_rx_flags(dev, flags); 8199 } 8200 8201 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) 8202 { 8203 unsigned int old_flags = dev->flags; 8204 kuid_t uid; 8205 kgid_t gid; 8206 8207 ASSERT_RTNL(); 8208 8209 dev->flags |= IFF_PROMISC; 8210 dev->promiscuity += inc; 8211 if (dev->promiscuity == 0) { 8212 /* 8213 * Avoid overflow. 8214 * If inc causes overflow, untouch promisc and return error. 8215 */ 8216 if (inc < 0) 8217 dev->flags &= ~IFF_PROMISC; 8218 else { 8219 dev->promiscuity -= inc; 8220 netdev_warn(dev, "promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n"); 8221 return -EOVERFLOW; 8222 } 8223 } 8224 if (dev->flags != old_flags) { 8225 pr_info("device %s %s promiscuous mode\n", 8226 dev->name, 8227 dev->flags & IFF_PROMISC ? "entered" : "left"); 8228 if (audit_enabled) { 8229 current_uid_gid(&uid, &gid); 8230 audit_log(audit_context(), GFP_ATOMIC, 8231 AUDIT_ANOM_PROMISCUOUS, 8232 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", 8233 dev->name, (dev->flags & IFF_PROMISC), 8234 (old_flags & IFF_PROMISC), 8235 from_kuid(&init_user_ns, audit_get_loginuid(current)), 8236 from_kuid(&init_user_ns, uid), 8237 from_kgid(&init_user_ns, gid), 8238 audit_get_sessionid(current)); 8239 } 8240 8241 dev_change_rx_flags(dev, IFF_PROMISC); 8242 } 8243 if (notify) 8244 __dev_notify_flags(dev, old_flags, IFF_PROMISC); 8245 return 0; 8246 } 8247 8248 /** 8249 * dev_set_promiscuity - update promiscuity count on a device 8250 * @dev: device 8251 * @inc: modifier 8252 * 8253 * Add or remove promiscuity from a device. While the count in the device 8254 * remains above zero the interface remains promiscuous. Once it hits zero 8255 * the device reverts back to normal filtering operation. A negative inc 8256 * value is used to drop promiscuity on the device. 8257 * Return 0 if successful or a negative errno code on error. 8258 */ 8259 int dev_set_promiscuity(struct net_device *dev, int inc) 8260 { 8261 unsigned int old_flags = dev->flags; 8262 int err; 8263 8264 err = __dev_set_promiscuity(dev, inc, true); 8265 if (err < 0) 8266 return err; 8267 if (dev->flags != old_flags) 8268 dev_set_rx_mode(dev); 8269 return err; 8270 } 8271 EXPORT_SYMBOL(dev_set_promiscuity); 8272 8273 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) 8274 { 8275 unsigned int old_flags = dev->flags, old_gflags = dev->gflags; 8276 8277 ASSERT_RTNL(); 8278 8279 dev->flags |= IFF_ALLMULTI; 8280 dev->allmulti += inc; 8281 if (dev->allmulti == 0) { 8282 /* 8283 * Avoid overflow. 8284 * If inc causes overflow, untouch allmulti and return error. 8285 */ 8286 if (inc < 0) 8287 dev->flags &= ~IFF_ALLMULTI; 8288 else { 8289 dev->allmulti -= inc; 8290 netdev_warn(dev, "allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n"); 8291 return -EOVERFLOW; 8292 } 8293 } 8294 if (dev->flags ^ old_flags) { 8295 dev_change_rx_flags(dev, IFF_ALLMULTI); 8296 dev_set_rx_mode(dev); 8297 if (notify) 8298 __dev_notify_flags(dev, old_flags, 8299 dev->gflags ^ old_gflags); 8300 } 8301 return 0; 8302 } 8303 8304 /** 8305 * dev_set_allmulti - update allmulti count on a device 8306 * @dev: device 8307 * @inc: modifier 8308 * 8309 * Add or remove reception of all multicast frames to a device. While the 8310 * count in the device remains above zero the interface remains listening 8311 * to all interfaces. Once it hits zero the device reverts back to normal 8312 * filtering operation. A negative @inc value is used to drop the counter 8313 * when releasing a resource needing all multicasts. 8314 * Return 0 if successful or a negative errno code on error. 8315 */ 8316 8317 int dev_set_allmulti(struct net_device *dev, int inc) 8318 { 8319 return __dev_set_allmulti(dev, inc, true); 8320 } 8321 EXPORT_SYMBOL(dev_set_allmulti); 8322 8323 /* 8324 * Upload unicast and multicast address lists to device and 8325 * configure RX filtering. When the device doesn't support unicast 8326 * filtering it is put in promiscuous mode while unicast addresses 8327 * are present. 8328 */ 8329 void __dev_set_rx_mode(struct net_device *dev) 8330 { 8331 const struct net_device_ops *ops = dev->netdev_ops; 8332 8333 /* dev_open will call this function so the list will stay sane. */ 8334 if (!(dev->flags&IFF_UP)) 8335 return; 8336 8337 if (!netif_device_present(dev)) 8338 return; 8339 8340 if (!(dev->priv_flags & IFF_UNICAST_FLT)) { 8341 /* Unicast addresses changes may only happen under the rtnl, 8342 * therefore calling __dev_set_promiscuity here is safe. 8343 */ 8344 if (!netdev_uc_empty(dev) && !dev->uc_promisc) { 8345 __dev_set_promiscuity(dev, 1, false); 8346 dev->uc_promisc = true; 8347 } else if (netdev_uc_empty(dev) && dev->uc_promisc) { 8348 __dev_set_promiscuity(dev, -1, false); 8349 dev->uc_promisc = false; 8350 } 8351 } 8352 8353 if (ops->ndo_set_rx_mode) 8354 ops->ndo_set_rx_mode(dev); 8355 } 8356 8357 void dev_set_rx_mode(struct net_device *dev) 8358 { 8359 netif_addr_lock_bh(dev); 8360 __dev_set_rx_mode(dev); 8361 netif_addr_unlock_bh(dev); 8362 } 8363 8364 /** 8365 * dev_get_flags - get flags reported to userspace 8366 * @dev: device 8367 * 8368 * Get the combination of flag bits exported through APIs to userspace. 8369 */ 8370 unsigned int dev_get_flags(const struct net_device *dev) 8371 { 8372 unsigned int flags; 8373 8374 flags = (dev->flags & ~(IFF_PROMISC | 8375 IFF_ALLMULTI | 8376 IFF_RUNNING | 8377 IFF_LOWER_UP | 8378 IFF_DORMANT)) | 8379 (dev->gflags & (IFF_PROMISC | 8380 IFF_ALLMULTI)); 8381 8382 if (netif_running(dev)) { 8383 if (netif_oper_up(dev)) 8384 flags |= IFF_RUNNING; 8385 if (netif_carrier_ok(dev)) 8386 flags |= IFF_LOWER_UP; 8387 if (netif_dormant(dev)) 8388 flags |= IFF_DORMANT; 8389 } 8390 8391 return flags; 8392 } 8393 EXPORT_SYMBOL(dev_get_flags); 8394 8395 int __dev_change_flags(struct net_device *dev, unsigned int flags, 8396 struct netlink_ext_ack *extack) 8397 { 8398 unsigned int old_flags = dev->flags; 8399 int ret; 8400 8401 ASSERT_RTNL(); 8402 8403 /* 8404 * Set the flags on our device. 8405 */ 8406 8407 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | 8408 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | 8409 IFF_AUTOMEDIA)) | 8410 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | 8411 IFF_ALLMULTI)); 8412 8413 /* 8414 * Load in the correct multicast list now the flags have changed. 8415 */ 8416 8417 if ((old_flags ^ flags) & IFF_MULTICAST) 8418 dev_change_rx_flags(dev, IFF_MULTICAST); 8419 8420 dev_set_rx_mode(dev); 8421 8422 /* 8423 * Have we downed the interface. We handle IFF_UP ourselves 8424 * according to user attempts to set it, rather than blindly 8425 * setting it. 8426 */ 8427 8428 ret = 0; 8429 if ((old_flags ^ flags) & IFF_UP) { 8430 if (old_flags & IFF_UP) 8431 __dev_close(dev); 8432 else 8433 ret = __dev_open(dev, extack); 8434 } 8435 8436 if ((flags ^ dev->gflags) & IFF_PROMISC) { 8437 int inc = (flags & IFF_PROMISC) ? 1 : -1; 8438 unsigned int old_flags = dev->flags; 8439 8440 dev->gflags ^= IFF_PROMISC; 8441 8442 if (__dev_set_promiscuity(dev, inc, false) >= 0) 8443 if (dev->flags != old_flags) 8444 dev_set_rx_mode(dev); 8445 } 8446 8447 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI 8448 * is important. Some (broken) drivers set IFF_PROMISC, when 8449 * IFF_ALLMULTI is requested not asking us and not reporting. 8450 */ 8451 if ((flags ^ dev->gflags) & IFF_ALLMULTI) { 8452 int inc = (flags & IFF_ALLMULTI) ? 1 : -1; 8453 8454 dev->gflags ^= IFF_ALLMULTI; 8455 __dev_set_allmulti(dev, inc, false); 8456 } 8457 8458 return ret; 8459 } 8460 8461 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, 8462 unsigned int gchanges) 8463 { 8464 unsigned int changes = dev->flags ^ old_flags; 8465 8466 if (gchanges) 8467 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC); 8468 8469 if (changes & IFF_UP) { 8470 if (dev->flags & IFF_UP) 8471 call_netdevice_notifiers(NETDEV_UP, dev); 8472 else 8473 call_netdevice_notifiers(NETDEV_DOWN, dev); 8474 } 8475 8476 if (dev->flags & IFF_UP && 8477 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { 8478 struct netdev_notifier_change_info change_info = { 8479 .info = { 8480 .dev = dev, 8481 }, 8482 .flags_changed = changes, 8483 }; 8484 8485 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info); 8486 } 8487 } 8488 8489 /** 8490 * dev_change_flags - change device settings 8491 * @dev: device 8492 * @flags: device state flags 8493 * @extack: netlink extended ack 8494 * 8495 * Change settings on device based state flags. The flags are 8496 * in the userspace exported format. 8497 */ 8498 int dev_change_flags(struct net_device *dev, unsigned int flags, 8499 struct netlink_ext_ack *extack) 8500 { 8501 int ret; 8502 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; 8503 8504 ret = __dev_change_flags(dev, flags, extack); 8505 if (ret < 0) 8506 return ret; 8507 8508 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); 8509 __dev_notify_flags(dev, old_flags, changes); 8510 return ret; 8511 } 8512 EXPORT_SYMBOL(dev_change_flags); 8513 8514 int __dev_set_mtu(struct net_device *dev, int new_mtu) 8515 { 8516 const struct net_device_ops *ops = dev->netdev_ops; 8517 8518 if (ops->ndo_change_mtu) 8519 return ops->ndo_change_mtu(dev, new_mtu); 8520 8521 /* Pairs with all the lockless reads of dev->mtu in the stack */ 8522 WRITE_ONCE(dev->mtu, new_mtu); 8523 return 0; 8524 } 8525 EXPORT_SYMBOL(__dev_set_mtu); 8526 8527 int dev_validate_mtu(struct net_device *dev, int new_mtu, 8528 struct netlink_ext_ack *extack) 8529 { 8530 /* MTU must be positive, and in range */ 8531 if (new_mtu < 0 || new_mtu < dev->min_mtu) { 8532 NL_SET_ERR_MSG(extack, "mtu less than device minimum"); 8533 return -EINVAL; 8534 } 8535 8536 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { 8537 NL_SET_ERR_MSG(extack, "mtu greater than device maximum"); 8538 return -EINVAL; 8539 } 8540 return 0; 8541 } 8542 8543 /** 8544 * dev_set_mtu_ext - Change maximum transfer unit 8545 * @dev: device 8546 * @new_mtu: new transfer unit 8547 * @extack: netlink extended ack 8548 * 8549 * Change the maximum transfer size of the network device. 8550 */ 8551 int dev_set_mtu_ext(struct net_device *dev, int new_mtu, 8552 struct netlink_ext_ack *extack) 8553 { 8554 int err, orig_mtu; 8555 8556 if (new_mtu == dev->mtu) 8557 return 0; 8558 8559 err = dev_validate_mtu(dev, new_mtu, extack); 8560 if (err) 8561 return err; 8562 8563 if (!netif_device_present(dev)) 8564 return -ENODEV; 8565 8566 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); 8567 err = notifier_to_errno(err); 8568 if (err) 8569 return err; 8570 8571 orig_mtu = dev->mtu; 8572 err = __dev_set_mtu(dev, new_mtu); 8573 8574 if (!err) { 8575 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8576 orig_mtu); 8577 err = notifier_to_errno(err); 8578 if (err) { 8579 /* setting mtu back and notifying everyone again, 8580 * so that they have a chance to revert changes. 8581 */ 8582 __dev_set_mtu(dev, orig_mtu); 8583 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8584 new_mtu); 8585 } 8586 } 8587 return err; 8588 } 8589 8590 int dev_set_mtu(struct net_device *dev, int new_mtu) 8591 { 8592 struct netlink_ext_ack extack; 8593 int err; 8594 8595 memset(&extack, 0, sizeof(extack)); 8596 err = dev_set_mtu_ext(dev, new_mtu, &extack); 8597 if (err && extack._msg) 8598 net_err_ratelimited("%s: %s\n", dev->name, extack._msg); 8599 return err; 8600 } 8601 EXPORT_SYMBOL(dev_set_mtu); 8602 8603 /** 8604 * dev_change_tx_queue_len - Change TX queue length of a netdevice 8605 * @dev: device 8606 * @new_len: new tx queue length 8607 */ 8608 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len) 8609 { 8610 unsigned int orig_len = dev->tx_queue_len; 8611 int res; 8612 8613 if (new_len != (unsigned int)new_len) 8614 return -ERANGE; 8615 8616 if (new_len != orig_len) { 8617 dev->tx_queue_len = new_len; 8618 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev); 8619 res = notifier_to_errno(res); 8620 if (res) 8621 goto err_rollback; 8622 res = dev_qdisc_change_tx_queue_len(dev); 8623 if (res) 8624 goto err_rollback; 8625 } 8626 8627 return 0; 8628 8629 err_rollback: 8630 netdev_err(dev, "refused to change device tx_queue_len\n"); 8631 dev->tx_queue_len = orig_len; 8632 return res; 8633 } 8634 8635 /** 8636 * dev_set_group - Change group this device belongs to 8637 * @dev: device 8638 * @new_group: group this device should belong to 8639 */ 8640 void dev_set_group(struct net_device *dev, int new_group) 8641 { 8642 dev->group = new_group; 8643 } 8644 EXPORT_SYMBOL(dev_set_group); 8645 8646 /** 8647 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR. 8648 * @dev: device 8649 * @addr: new address 8650 * @extack: netlink extended ack 8651 */ 8652 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr, 8653 struct netlink_ext_ack *extack) 8654 { 8655 struct netdev_notifier_pre_changeaddr_info info = { 8656 .info.dev = dev, 8657 .info.extack = extack, 8658 .dev_addr = addr, 8659 }; 8660 int rc; 8661 8662 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info); 8663 return notifier_to_errno(rc); 8664 } 8665 EXPORT_SYMBOL(dev_pre_changeaddr_notify); 8666 8667 /** 8668 * dev_set_mac_address - Change Media Access Control Address 8669 * @dev: device 8670 * @sa: new address 8671 * @extack: netlink extended ack 8672 * 8673 * Change the hardware (MAC) address of the device 8674 */ 8675 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa, 8676 struct netlink_ext_ack *extack) 8677 { 8678 const struct net_device_ops *ops = dev->netdev_ops; 8679 int err; 8680 8681 if (!ops->ndo_set_mac_address) 8682 return -EOPNOTSUPP; 8683 if (sa->sa_family != dev->type) 8684 return -EINVAL; 8685 if (!netif_device_present(dev)) 8686 return -ENODEV; 8687 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack); 8688 if (err) 8689 return err; 8690 err = ops->ndo_set_mac_address(dev, sa); 8691 if (err) 8692 return err; 8693 dev->addr_assign_type = NET_ADDR_SET; 8694 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); 8695 add_device_randomness(dev->dev_addr, dev->addr_len); 8696 return 0; 8697 } 8698 EXPORT_SYMBOL(dev_set_mac_address); 8699 8700 static DECLARE_RWSEM(dev_addr_sem); 8701 8702 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa, 8703 struct netlink_ext_ack *extack) 8704 { 8705 int ret; 8706 8707 down_write(&dev_addr_sem); 8708 ret = dev_set_mac_address(dev, sa, extack); 8709 up_write(&dev_addr_sem); 8710 return ret; 8711 } 8712 EXPORT_SYMBOL(dev_set_mac_address_user); 8713 8714 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name) 8715 { 8716 size_t size = sizeof(sa->sa_data); 8717 struct net_device *dev; 8718 int ret = 0; 8719 8720 down_read(&dev_addr_sem); 8721 rcu_read_lock(); 8722 8723 dev = dev_get_by_name_rcu(net, dev_name); 8724 if (!dev) { 8725 ret = -ENODEV; 8726 goto unlock; 8727 } 8728 if (!dev->addr_len) 8729 memset(sa->sa_data, 0, size); 8730 else 8731 memcpy(sa->sa_data, dev->dev_addr, 8732 min_t(size_t, size, dev->addr_len)); 8733 sa->sa_family = dev->type; 8734 8735 unlock: 8736 rcu_read_unlock(); 8737 up_read(&dev_addr_sem); 8738 return ret; 8739 } 8740 EXPORT_SYMBOL(dev_get_mac_address); 8741 8742 /** 8743 * dev_change_carrier - Change device carrier 8744 * @dev: device 8745 * @new_carrier: new value 8746 * 8747 * Change device carrier 8748 */ 8749 int dev_change_carrier(struct net_device *dev, bool new_carrier) 8750 { 8751 const struct net_device_ops *ops = dev->netdev_ops; 8752 8753 if (!ops->ndo_change_carrier) 8754 return -EOPNOTSUPP; 8755 if (!netif_device_present(dev)) 8756 return -ENODEV; 8757 return ops->ndo_change_carrier(dev, new_carrier); 8758 } 8759 EXPORT_SYMBOL(dev_change_carrier); 8760 8761 /** 8762 * dev_get_phys_port_id - Get device physical port ID 8763 * @dev: device 8764 * @ppid: port ID 8765 * 8766 * Get device physical port ID 8767 */ 8768 int dev_get_phys_port_id(struct net_device *dev, 8769 struct netdev_phys_item_id *ppid) 8770 { 8771 const struct net_device_ops *ops = dev->netdev_ops; 8772 8773 if (!ops->ndo_get_phys_port_id) 8774 return -EOPNOTSUPP; 8775 return ops->ndo_get_phys_port_id(dev, ppid); 8776 } 8777 EXPORT_SYMBOL(dev_get_phys_port_id); 8778 8779 /** 8780 * dev_get_phys_port_name - Get device physical port name 8781 * @dev: device 8782 * @name: port name 8783 * @len: limit of bytes to copy to name 8784 * 8785 * Get device physical port name 8786 */ 8787 int dev_get_phys_port_name(struct net_device *dev, 8788 char *name, size_t len) 8789 { 8790 const struct net_device_ops *ops = dev->netdev_ops; 8791 int err; 8792 8793 if (ops->ndo_get_phys_port_name) { 8794 err = ops->ndo_get_phys_port_name(dev, name, len); 8795 if (err != -EOPNOTSUPP) 8796 return err; 8797 } 8798 return devlink_compat_phys_port_name_get(dev, name, len); 8799 } 8800 EXPORT_SYMBOL(dev_get_phys_port_name); 8801 8802 /** 8803 * dev_get_port_parent_id - Get the device's port parent identifier 8804 * @dev: network device 8805 * @ppid: pointer to a storage for the port's parent identifier 8806 * @recurse: allow/disallow recursion to lower devices 8807 * 8808 * Get the devices's port parent identifier 8809 */ 8810 int dev_get_port_parent_id(struct net_device *dev, 8811 struct netdev_phys_item_id *ppid, 8812 bool recurse) 8813 { 8814 const struct net_device_ops *ops = dev->netdev_ops; 8815 struct netdev_phys_item_id first = { }; 8816 struct net_device *lower_dev; 8817 struct list_head *iter; 8818 int err; 8819 8820 if (ops->ndo_get_port_parent_id) { 8821 err = ops->ndo_get_port_parent_id(dev, ppid); 8822 if (err != -EOPNOTSUPP) 8823 return err; 8824 } 8825 8826 err = devlink_compat_switch_id_get(dev, ppid); 8827 if (!recurse || err != -EOPNOTSUPP) 8828 return err; 8829 8830 netdev_for_each_lower_dev(dev, lower_dev, iter) { 8831 err = dev_get_port_parent_id(lower_dev, ppid, true); 8832 if (err) 8833 break; 8834 if (!first.id_len) 8835 first = *ppid; 8836 else if (memcmp(&first, ppid, sizeof(*ppid))) 8837 return -EOPNOTSUPP; 8838 } 8839 8840 return err; 8841 } 8842 EXPORT_SYMBOL(dev_get_port_parent_id); 8843 8844 /** 8845 * netdev_port_same_parent_id - Indicate if two network devices have 8846 * the same port parent identifier 8847 * @a: first network device 8848 * @b: second network device 8849 */ 8850 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b) 8851 { 8852 struct netdev_phys_item_id a_id = { }; 8853 struct netdev_phys_item_id b_id = { }; 8854 8855 if (dev_get_port_parent_id(a, &a_id, true) || 8856 dev_get_port_parent_id(b, &b_id, true)) 8857 return false; 8858 8859 return netdev_phys_item_id_same(&a_id, &b_id); 8860 } 8861 EXPORT_SYMBOL(netdev_port_same_parent_id); 8862 8863 /** 8864 * dev_change_proto_down - set carrier according to proto_down. 8865 * 8866 * @dev: device 8867 * @proto_down: new value 8868 */ 8869 int dev_change_proto_down(struct net_device *dev, bool proto_down) 8870 { 8871 if (!(dev->priv_flags & IFF_CHANGE_PROTO_DOWN)) 8872 return -EOPNOTSUPP; 8873 if (!netif_device_present(dev)) 8874 return -ENODEV; 8875 if (proto_down) 8876 netif_carrier_off(dev); 8877 else 8878 netif_carrier_on(dev); 8879 dev->proto_down = proto_down; 8880 return 0; 8881 } 8882 EXPORT_SYMBOL(dev_change_proto_down); 8883 8884 /** 8885 * dev_change_proto_down_reason - proto down reason 8886 * 8887 * @dev: device 8888 * @mask: proto down mask 8889 * @value: proto down value 8890 */ 8891 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask, 8892 u32 value) 8893 { 8894 int b; 8895 8896 if (!mask) { 8897 dev->proto_down_reason = value; 8898 } else { 8899 for_each_set_bit(b, &mask, 32) { 8900 if (value & (1 << b)) 8901 dev->proto_down_reason |= BIT(b); 8902 else 8903 dev->proto_down_reason &= ~BIT(b); 8904 } 8905 } 8906 } 8907 EXPORT_SYMBOL(dev_change_proto_down_reason); 8908 8909 struct bpf_xdp_link { 8910 struct bpf_link link; 8911 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */ 8912 int flags; 8913 }; 8914 8915 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags) 8916 { 8917 if (flags & XDP_FLAGS_HW_MODE) 8918 return XDP_MODE_HW; 8919 if (flags & XDP_FLAGS_DRV_MODE) 8920 return XDP_MODE_DRV; 8921 if (flags & XDP_FLAGS_SKB_MODE) 8922 return XDP_MODE_SKB; 8923 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB; 8924 } 8925 8926 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode) 8927 { 8928 switch (mode) { 8929 case XDP_MODE_SKB: 8930 return generic_xdp_install; 8931 case XDP_MODE_DRV: 8932 case XDP_MODE_HW: 8933 return dev->netdev_ops->ndo_bpf; 8934 default: 8935 return NULL; 8936 } 8937 } 8938 8939 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev, 8940 enum bpf_xdp_mode mode) 8941 { 8942 return dev->xdp_state[mode].link; 8943 } 8944 8945 static struct bpf_prog *dev_xdp_prog(struct net_device *dev, 8946 enum bpf_xdp_mode mode) 8947 { 8948 struct bpf_xdp_link *link = dev_xdp_link(dev, mode); 8949 8950 if (link) 8951 return link->link.prog; 8952 return dev->xdp_state[mode].prog; 8953 } 8954 8955 u8 dev_xdp_prog_count(struct net_device *dev) 8956 { 8957 u8 count = 0; 8958 int i; 8959 8960 for (i = 0; i < __MAX_XDP_MODE; i++) 8961 if (dev->xdp_state[i].prog || dev->xdp_state[i].link) 8962 count++; 8963 return count; 8964 } 8965 EXPORT_SYMBOL_GPL(dev_xdp_prog_count); 8966 8967 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode) 8968 { 8969 struct bpf_prog *prog = dev_xdp_prog(dev, mode); 8970 8971 return prog ? prog->aux->id : 0; 8972 } 8973 8974 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode, 8975 struct bpf_xdp_link *link) 8976 { 8977 dev->xdp_state[mode].link = link; 8978 dev->xdp_state[mode].prog = NULL; 8979 } 8980 8981 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode, 8982 struct bpf_prog *prog) 8983 { 8984 dev->xdp_state[mode].link = NULL; 8985 dev->xdp_state[mode].prog = prog; 8986 } 8987 8988 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode, 8989 bpf_op_t bpf_op, struct netlink_ext_ack *extack, 8990 u32 flags, struct bpf_prog *prog) 8991 { 8992 struct netdev_bpf xdp; 8993 int err; 8994 8995 memset(&xdp, 0, sizeof(xdp)); 8996 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG; 8997 xdp.extack = extack; 8998 xdp.flags = flags; 8999 xdp.prog = prog; 9000 9001 /* Drivers assume refcnt is already incremented (i.e, prog pointer is 9002 * "moved" into driver), so they don't increment it on their own, but 9003 * they do decrement refcnt when program is detached or replaced. 9004 * Given net_device also owns link/prog, we need to bump refcnt here 9005 * to prevent drivers from underflowing it. 9006 */ 9007 if (prog) 9008 bpf_prog_inc(prog); 9009 err = bpf_op(dev, &xdp); 9010 if (err) { 9011 if (prog) 9012 bpf_prog_put(prog); 9013 return err; 9014 } 9015 9016 if (mode != XDP_MODE_HW) 9017 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog); 9018 9019 return 0; 9020 } 9021 9022 static void dev_xdp_uninstall(struct net_device *dev) 9023 { 9024 struct bpf_xdp_link *link; 9025 struct bpf_prog *prog; 9026 enum bpf_xdp_mode mode; 9027 bpf_op_t bpf_op; 9028 9029 ASSERT_RTNL(); 9030 9031 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) { 9032 prog = dev_xdp_prog(dev, mode); 9033 if (!prog) 9034 continue; 9035 9036 bpf_op = dev_xdp_bpf_op(dev, mode); 9037 if (!bpf_op) 9038 continue; 9039 9040 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9041 9042 /* auto-detach link from net device */ 9043 link = dev_xdp_link(dev, mode); 9044 if (link) 9045 link->dev = NULL; 9046 else 9047 bpf_prog_put(prog); 9048 9049 dev_xdp_set_link(dev, mode, NULL); 9050 } 9051 } 9052 9053 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack, 9054 struct bpf_xdp_link *link, struct bpf_prog *new_prog, 9055 struct bpf_prog *old_prog, u32 flags) 9056 { 9057 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES); 9058 struct bpf_prog *cur_prog; 9059 struct net_device *upper; 9060 struct list_head *iter; 9061 enum bpf_xdp_mode mode; 9062 bpf_op_t bpf_op; 9063 int err; 9064 9065 ASSERT_RTNL(); 9066 9067 /* either link or prog attachment, never both */ 9068 if (link && (new_prog || old_prog)) 9069 return -EINVAL; 9070 /* link supports only XDP mode flags */ 9071 if (link && (flags & ~XDP_FLAGS_MODES)) { 9072 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment"); 9073 return -EINVAL; 9074 } 9075 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */ 9076 if (num_modes > 1) { 9077 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set"); 9078 return -EINVAL; 9079 } 9080 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */ 9081 if (!num_modes && dev_xdp_prog_count(dev) > 1) { 9082 NL_SET_ERR_MSG(extack, 9083 "More than one program loaded, unset mode is ambiguous"); 9084 return -EINVAL; 9085 } 9086 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */ 9087 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) { 9088 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified"); 9089 return -EINVAL; 9090 } 9091 9092 mode = dev_xdp_mode(dev, flags); 9093 /* can't replace attached link */ 9094 if (dev_xdp_link(dev, mode)) { 9095 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link"); 9096 return -EBUSY; 9097 } 9098 9099 /* don't allow if an upper device already has a program */ 9100 netdev_for_each_upper_dev_rcu(dev, upper, iter) { 9101 if (dev_xdp_prog_count(upper) > 0) { 9102 NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program"); 9103 return -EEXIST; 9104 } 9105 } 9106 9107 cur_prog = dev_xdp_prog(dev, mode); 9108 /* can't replace attached prog with link */ 9109 if (link && cur_prog) { 9110 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link"); 9111 return -EBUSY; 9112 } 9113 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) { 9114 NL_SET_ERR_MSG(extack, "Active program does not match expected"); 9115 return -EEXIST; 9116 } 9117 9118 /* put effective new program into new_prog */ 9119 if (link) 9120 new_prog = link->link.prog; 9121 9122 if (new_prog) { 9123 bool offload = mode == XDP_MODE_HW; 9124 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB 9125 ? XDP_MODE_DRV : XDP_MODE_SKB; 9126 9127 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) { 9128 NL_SET_ERR_MSG(extack, "XDP program already attached"); 9129 return -EBUSY; 9130 } 9131 if (!offload && dev_xdp_prog(dev, other_mode)) { 9132 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time"); 9133 return -EEXIST; 9134 } 9135 if (!offload && bpf_prog_is_dev_bound(new_prog->aux)) { 9136 NL_SET_ERR_MSG(extack, "Using device-bound program without HW_MODE flag is not supported"); 9137 return -EINVAL; 9138 } 9139 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) { 9140 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device"); 9141 return -EINVAL; 9142 } 9143 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) { 9144 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device"); 9145 return -EINVAL; 9146 } 9147 } 9148 9149 /* don't call drivers if the effective program didn't change */ 9150 if (new_prog != cur_prog) { 9151 bpf_op = dev_xdp_bpf_op(dev, mode); 9152 if (!bpf_op) { 9153 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode"); 9154 return -EOPNOTSUPP; 9155 } 9156 9157 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog); 9158 if (err) 9159 return err; 9160 } 9161 9162 if (link) 9163 dev_xdp_set_link(dev, mode, link); 9164 else 9165 dev_xdp_set_prog(dev, mode, new_prog); 9166 if (cur_prog) 9167 bpf_prog_put(cur_prog); 9168 9169 return 0; 9170 } 9171 9172 static int dev_xdp_attach_link(struct net_device *dev, 9173 struct netlink_ext_ack *extack, 9174 struct bpf_xdp_link *link) 9175 { 9176 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags); 9177 } 9178 9179 static int dev_xdp_detach_link(struct net_device *dev, 9180 struct netlink_ext_ack *extack, 9181 struct bpf_xdp_link *link) 9182 { 9183 enum bpf_xdp_mode mode; 9184 bpf_op_t bpf_op; 9185 9186 ASSERT_RTNL(); 9187 9188 mode = dev_xdp_mode(dev, link->flags); 9189 if (dev_xdp_link(dev, mode) != link) 9190 return -EINVAL; 9191 9192 bpf_op = dev_xdp_bpf_op(dev, mode); 9193 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9194 dev_xdp_set_link(dev, mode, NULL); 9195 return 0; 9196 } 9197 9198 static void bpf_xdp_link_release(struct bpf_link *link) 9199 { 9200 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9201 9202 rtnl_lock(); 9203 9204 /* if racing with net_device's tear down, xdp_link->dev might be 9205 * already NULL, in which case link was already auto-detached 9206 */ 9207 if (xdp_link->dev) { 9208 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link)); 9209 xdp_link->dev = NULL; 9210 } 9211 9212 rtnl_unlock(); 9213 } 9214 9215 static int bpf_xdp_link_detach(struct bpf_link *link) 9216 { 9217 bpf_xdp_link_release(link); 9218 return 0; 9219 } 9220 9221 static void bpf_xdp_link_dealloc(struct bpf_link *link) 9222 { 9223 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9224 9225 kfree(xdp_link); 9226 } 9227 9228 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link, 9229 struct seq_file *seq) 9230 { 9231 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9232 u32 ifindex = 0; 9233 9234 rtnl_lock(); 9235 if (xdp_link->dev) 9236 ifindex = xdp_link->dev->ifindex; 9237 rtnl_unlock(); 9238 9239 seq_printf(seq, "ifindex:\t%u\n", ifindex); 9240 } 9241 9242 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link, 9243 struct bpf_link_info *info) 9244 { 9245 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9246 u32 ifindex = 0; 9247 9248 rtnl_lock(); 9249 if (xdp_link->dev) 9250 ifindex = xdp_link->dev->ifindex; 9251 rtnl_unlock(); 9252 9253 info->xdp.ifindex = ifindex; 9254 return 0; 9255 } 9256 9257 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog, 9258 struct bpf_prog *old_prog) 9259 { 9260 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9261 enum bpf_xdp_mode mode; 9262 bpf_op_t bpf_op; 9263 int err = 0; 9264 9265 rtnl_lock(); 9266 9267 /* link might have been auto-released already, so fail */ 9268 if (!xdp_link->dev) { 9269 err = -ENOLINK; 9270 goto out_unlock; 9271 } 9272 9273 if (old_prog && link->prog != old_prog) { 9274 err = -EPERM; 9275 goto out_unlock; 9276 } 9277 old_prog = link->prog; 9278 if (old_prog->type != new_prog->type || 9279 old_prog->expected_attach_type != new_prog->expected_attach_type) { 9280 err = -EINVAL; 9281 goto out_unlock; 9282 } 9283 9284 if (old_prog == new_prog) { 9285 /* no-op, don't disturb drivers */ 9286 bpf_prog_put(new_prog); 9287 goto out_unlock; 9288 } 9289 9290 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags); 9291 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode); 9292 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL, 9293 xdp_link->flags, new_prog); 9294 if (err) 9295 goto out_unlock; 9296 9297 old_prog = xchg(&link->prog, new_prog); 9298 bpf_prog_put(old_prog); 9299 9300 out_unlock: 9301 rtnl_unlock(); 9302 return err; 9303 } 9304 9305 static const struct bpf_link_ops bpf_xdp_link_lops = { 9306 .release = bpf_xdp_link_release, 9307 .dealloc = bpf_xdp_link_dealloc, 9308 .detach = bpf_xdp_link_detach, 9309 .show_fdinfo = bpf_xdp_link_show_fdinfo, 9310 .fill_link_info = bpf_xdp_link_fill_link_info, 9311 .update_prog = bpf_xdp_link_update, 9312 }; 9313 9314 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog) 9315 { 9316 struct net *net = current->nsproxy->net_ns; 9317 struct bpf_link_primer link_primer; 9318 struct bpf_xdp_link *link; 9319 struct net_device *dev; 9320 int err, fd; 9321 9322 rtnl_lock(); 9323 dev = dev_get_by_index(net, attr->link_create.target_ifindex); 9324 if (!dev) { 9325 rtnl_unlock(); 9326 return -EINVAL; 9327 } 9328 9329 link = kzalloc(sizeof(*link), GFP_USER); 9330 if (!link) { 9331 err = -ENOMEM; 9332 goto unlock; 9333 } 9334 9335 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog); 9336 link->dev = dev; 9337 link->flags = attr->link_create.flags; 9338 9339 err = bpf_link_prime(&link->link, &link_primer); 9340 if (err) { 9341 kfree(link); 9342 goto unlock; 9343 } 9344 9345 err = dev_xdp_attach_link(dev, NULL, link); 9346 rtnl_unlock(); 9347 9348 if (err) { 9349 link->dev = NULL; 9350 bpf_link_cleanup(&link_primer); 9351 goto out_put_dev; 9352 } 9353 9354 fd = bpf_link_settle(&link_primer); 9355 /* link itself doesn't hold dev's refcnt to not complicate shutdown */ 9356 dev_put(dev); 9357 return fd; 9358 9359 unlock: 9360 rtnl_unlock(); 9361 9362 out_put_dev: 9363 dev_put(dev); 9364 return err; 9365 } 9366 9367 /** 9368 * dev_change_xdp_fd - set or clear a bpf program for a device rx path 9369 * @dev: device 9370 * @extack: netlink extended ack 9371 * @fd: new program fd or negative value to clear 9372 * @expected_fd: old program fd that userspace expects to replace or clear 9373 * @flags: xdp-related flags 9374 * 9375 * Set or clear a bpf program for a device 9376 */ 9377 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, 9378 int fd, int expected_fd, u32 flags) 9379 { 9380 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags); 9381 struct bpf_prog *new_prog = NULL, *old_prog = NULL; 9382 int err; 9383 9384 ASSERT_RTNL(); 9385 9386 if (fd >= 0) { 9387 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP, 9388 mode != XDP_MODE_SKB); 9389 if (IS_ERR(new_prog)) 9390 return PTR_ERR(new_prog); 9391 } 9392 9393 if (expected_fd >= 0) { 9394 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP, 9395 mode != XDP_MODE_SKB); 9396 if (IS_ERR(old_prog)) { 9397 err = PTR_ERR(old_prog); 9398 old_prog = NULL; 9399 goto err_out; 9400 } 9401 } 9402 9403 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags); 9404 9405 err_out: 9406 if (err && new_prog) 9407 bpf_prog_put(new_prog); 9408 if (old_prog) 9409 bpf_prog_put(old_prog); 9410 return err; 9411 } 9412 9413 /** 9414 * dev_new_index - allocate an ifindex 9415 * @net: the applicable net namespace 9416 * 9417 * Returns a suitable unique value for a new device interface 9418 * number. The caller must hold the rtnl semaphore or the 9419 * dev_base_lock to be sure it remains unique. 9420 */ 9421 static int dev_new_index(struct net *net) 9422 { 9423 int ifindex = net->ifindex; 9424 9425 for (;;) { 9426 if (++ifindex <= 0) 9427 ifindex = 1; 9428 if (!__dev_get_by_index(net, ifindex)) 9429 return net->ifindex = ifindex; 9430 } 9431 } 9432 9433 /* Delayed registration/unregisteration */ 9434 static LIST_HEAD(net_todo_list); 9435 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); 9436 9437 static void net_set_todo(struct net_device *dev) 9438 { 9439 list_add_tail(&dev->todo_list, &net_todo_list); 9440 atomic_inc(&dev_net(dev)->dev_unreg_count); 9441 } 9442 9443 static netdev_features_t netdev_sync_upper_features(struct net_device *lower, 9444 struct net_device *upper, netdev_features_t features) 9445 { 9446 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9447 netdev_features_t feature; 9448 int feature_bit; 9449 9450 for_each_netdev_feature(upper_disables, feature_bit) { 9451 feature = __NETIF_F_BIT(feature_bit); 9452 if (!(upper->wanted_features & feature) 9453 && (features & feature)) { 9454 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", 9455 &feature, upper->name); 9456 features &= ~feature; 9457 } 9458 } 9459 9460 return features; 9461 } 9462 9463 static void netdev_sync_lower_features(struct net_device *upper, 9464 struct net_device *lower, netdev_features_t features) 9465 { 9466 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9467 netdev_features_t feature; 9468 int feature_bit; 9469 9470 for_each_netdev_feature(upper_disables, feature_bit) { 9471 feature = __NETIF_F_BIT(feature_bit); 9472 if (!(features & feature) && (lower->features & feature)) { 9473 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", 9474 &feature, lower->name); 9475 lower->wanted_features &= ~feature; 9476 __netdev_update_features(lower); 9477 9478 if (unlikely(lower->features & feature)) 9479 netdev_WARN(upper, "failed to disable %pNF on %s!\n", 9480 &feature, lower->name); 9481 else 9482 netdev_features_change(lower); 9483 } 9484 } 9485 } 9486 9487 static netdev_features_t netdev_fix_features(struct net_device *dev, 9488 netdev_features_t features) 9489 { 9490 /* Fix illegal checksum combinations */ 9491 if ((features & NETIF_F_HW_CSUM) && 9492 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { 9493 netdev_warn(dev, "mixed HW and IP checksum settings.\n"); 9494 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); 9495 } 9496 9497 /* TSO requires that SG is present as well. */ 9498 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { 9499 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); 9500 features &= ~NETIF_F_ALL_TSO; 9501 } 9502 9503 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && 9504 !(features & NETIF_F_IP_CSUM)) { 9505 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); 9506 features &= ~NETIF_F_TSO; 9507 features &= ~NETIF_F_TSO_ECN; 9508 } 9509 9510 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && 9511 !(features & NETIF_F_IPV6_CSUM)) { 9512 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); 9513 features &= ~NETIF_F_TSO6; 9514 } 9515 9516 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ 9517 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) 9518 features &= ~NETIF_F_TSO_MANGLEID; 9519 9520 /* TSO ECN requires that TSO is present as well. */ 9521 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) 9522 features &= ~NETIF_F_TSO_ECN; 9523 9524 /* Software GSO depends on SG. */ 9525 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { 9526 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); 9527 features &= ~NETIF_F_GSO; 9528 } 9529 9530 /* GSO partial features require GSO partial be set */ 9531 if ((features & dev->gso_partial_features) && 9532 !(features & NETIF_F_GSO_PARTIAL)) { 9533 netdev_dbg(dev, 9534 "Dropping partially supported GSO features since no GSO partial.\n"); 9535 features &= ~dev->gso_partial_features; 9536 } 9537 9538 if (!(features & NETIF_F_RXCSUM)) { 9539 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet 9540 * successfully merged by hardware must also have the 9541 * checksum verified by hardware. If the user does not 9542 * want to enable RXCSUM, logically, we should disable GRO_HW. 9543 */ 9544 if (features & NETIF_F_GRO_HW) { 9545 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n"); 9546 features &= ~NETIF_F_GRO_HW; 9547 } 9548 } 9549 9550 /* LRO/HW-GRO features cannot be combined with RX-FCS */ 9551 if (features & NETIF_F_RXFCS) { 9552 if (features & NETIF_F_LRO) { 9553 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n"); 9554 features &= ~NETIF_F_LRO; 9555 } 9556 9557 if (features & NETIF_F_GRO_HW) { 9558 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n"); 9559 features &= ~NETIF_F_GRO_HW; 9560 } 9561 } 9562 9563 if ((features & NETIF_F_GRO_HW) && (features & NETIF_F_LRO)) { 9564 netdev_dbg(dev, "Dropping LRO feature since HW-GRO is requested.\n"); 9565 features &= ~NETIF_F_LRO; 9566 } 9567 9568 if (features & NETIF_F_HW_TLS_TX) { 9569 bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) == 9570 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM); 9571 bool hw_csum = features & NETIF_F_HW_CSUM; 9572 9573 if (!ip_csum && !hw_csum) { 9574 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n"); 9575 features &= ~NETIF_F_HW_TLS_TX; 9576 } 9577 } 9578 9579 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) { 9580 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n"); 9581 features &= ~NETIF_F_HW_TLS_RX; 9582 } 9583 9584 return features; 9585 } 9586 9587 int __netdev_update_features(struct net_device *dev) 9588 { 9589 struct net_device *upper, *lower; 9590 netdev_features_t features; 9591 struct list_head *iter; 9592 int err = -1; 9593 9594 ASSERT_RTNL(); 9595 9596 features = netdev_get_wanted_features(dev); 9597 9598 if (dev->netdev_ops->ndo_fix_features) 9599 features = dev->netdev_ops->ndo_fix_features(dev, features); 9600 9601 /* driver might be less strict about feature dependencies */ 9602 features = netdev_fix_features(dev, features); 9603 9604 /* some features can't be enabled if they're off on an upper device */ 9605 netdev_for_each_upper_dev_rcu(dev, upper, iter) 9606 features = netdev_sync_upper_features(dev, upper, features); 9607 9608 if (dev->features == features) 9609 goto sync_lower; 9610 9611 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", 9612 &dev->features, &features); 9613 9614 if (dev->netdev_ops->ndo_set_features) 9615 err = dev->netdev_ops->ndo_set_features(dev, features); 9616 else 9617 err = 0; 9618 9619 if (unlikely(err < 0)) { 9620 netdev_err(dev, 9621 "set_features() failed (%d); wanted %pNF, left %pNF\n", 9622 err, &features, &dev->features); 9623 /* return non-0 since some features might have changed and 9624 * it's better to fire a spurious notification than miss it 9625 */ 9626 return -1; 9627 } 9628 9629 sync_lower: 9630 /* some features must be disabled on lower devices when disabled 9631 * on an upper device (think: bonding master or bridge) 9632 */ 9633 netdev_for_each_lower_dev(dev, lower, iter) 9634 netdev_sync_lower_features(dev, lower, features); 9635 9636 if (!err) { 9637 netdev_features_t diff = features ^ dev->features; 9638 9639 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) { 9640 /* udp_tunnel_{get,drop}_rx_info both need 9641 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the 9642 * device, or they won't do anything. 9643 * Thus we need to update dev->features 9644 * *before* calling udp_tunnel_get_rx_info, 9645 * but *after* calling udp_tunnel_drop_rx_info. 9646 */ 9647 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) { 9648 dev->features = features; 9649 udp_tunnel_get_rx_info(dev); 9650 } else { 9651 udp_tunnel_drop_rx_info(dev); 9652 } 9653 } 9654 9655 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) { 9656 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) { 9657 dev->features = features; 9658 err |= vlan_get_rx_ctag_filter_info(dev); 9659 } else { 9660 vlan_drop_rx_ctag_filter_info(dev); 9661 } 9662 } 9663 9664 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) { 9665 if (features & NETIF_F_HW_VLAN_STAG_FILTER) { 9666 dev->features = features; 9667 err |= vlan_get_rx_stag_filter_info(dev); 9668 } else { 9669 vlan_drop_rx_stag_filter_info(dev); 9670 } 9671 } 9672 9673 dev->features = features; 9674 } 9675 9676 return err < 0 ? 0 : 1; 9677 } 9678 9679 /** 9680 * netdev_update_features - recalculate device features 9681 * @dev: the device to check 9682 * 9683 * Recalculate dev->features set and send notifications if it 9684 * has changed. Should be called after driver or hardware dependent 9685 * conditions might have changed that influence the features. 9686 */ 9687 void netdev_update_features(struct net_device *dev) 9688 { 9689 if (__netdev_update_features(dev)) 9690 netdev_features_change(dev); 9691 } 9692 EXPORT_SYMBOL(netdev_update_features); 9693 9694 /** 9695 * netdev_change_features - recalculate device features 9696 * @dev: the device to check 9697 * 9698 * Recalculate dev->features set and send notifications even 9699 * if they have not changed. Should be called instead of 9700 * netdev_update_features() if also dev->vlan_features might 9701 * have changed to allow the changes to be propagated to stacked 9702 * VLAN devices. 9703 */ 9704 void netdev_change_features(struct net_device *dev) 9705 { 9706 __netdev_update_features(dev); 9707 netdev_features_change(dev); 9708 } 9709 EXPORT_SYMBOL(netdev_change_features); 9710 9711 /** 9712 * netif_stacked_transfer_operstate - transfer operstate 9713 * @rootdev: the root or lower level device to transfer state from 9714 * @dev: the device to transfer operstate to 9715 * 9716 * Transfer operational state from root to device. This is normally 9717 * called when a stacking relationship exists between the root 9718 * device and the device(a leaf device). 9719 */ 9720 void netif_stacked_transfer_operstate(const struct net_device *rootdev, 9721 struct net_device *dev) 9722 { 9723 if (rootdev->operstate == IF_OPER_DORMANT) 9724 netif_dormant_on(dev); 9725 else 9726 netif_dormant_off(dev); 9727 9728 if (rootdev->operstate == IF_OPER_TESTING) 9729 netif_testing_on(dev); 9730 else 9731 netif_testing_off(dev); 9732 9733 if (netif_carrier_ok(rootdev)) 9734 netif_carrier_on(dev); 9735 else 9736 netif_carrier_off(dev); 9737 } 9738 EXPORT_SYMBOL(netif_stacked_transfer_operstate); 9739 9740 static int netif_alloc_rx_queues(struct net_device *dev) 9741 { 9742 unsigned int i, count = dev->num_rx_queues; 9743 struct netdev_rx_queue *rx; 9744 size_t sz = count * sizeof(*rx); 9745 int err = 0; 9746 9747 BUG_ON(count < 1); 9748 9749 rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 9750 if (!rx) 9751 return -ENOMEM; 9752 9753 dev->_rx = rx; 9754 9755 for (i = 0; i < count; i++) { 9756 rx[i].dev = dev; 9757 9758 /* XDP RX-queue setup */ 9759 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0); 9760 if (err < 0) 9761 goto err_rxq_info; 9762 } 9763 return 0; 9764 9765 err_rxq_info: 9766 /* Rollback successful reg's and free other resources */ 9767 while (i--) 9768 xdp_rxq_info_unreg(&rx[i].xdp_rxq); 9769 kvfree(dev->_rx); 9770 dev->_rx = NULL; 9771 return err; 9772 } 9773 9774 static void netif_free_rx_queues(struct net_device *dev) 9775 { 9776 unsigned int i, count = dev->num_rx_queues; 9777 9778 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */ 9779 if (!dev->_rx) 9780 return; 9781 9782 for (i = 0; i < count; i++) 9783 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq); 9784 9785 kvfree(dev->_rx); 9786 } 9787 9788 static void netdev_init_one_queue(struct net_device *dev, 9789 struct netdev_queue *queue, void *_unused) 9790 { 9791 /* Initialize queue lock */ 9792 spin_lock_init(&queue->_xmit_lock); 9793 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); 9794 queue->xmit_lock_owner = -1; 9795 netdev_queue_numa_node_write(queue, NUMA_NO_NODE); 9796 queue->dev = dev; 9797 #ifdef CONFIG_BQL 9798 dql_init(&queue->dql, HZ); 9799 #endif 9800 } 9801 9802 static void netif_free_tx_queues(struct net_device *dev) 9803 { 9804 kvfree(dev->_tx); 9805 } 9806 9807 static int netif_alloc_netdev_queues(struct net_device *dev) 9808 { 9809 unsigned int count = dev->num_tx_queues; 9810 struct netdev_queue *tx; 9811 size_t sz = count * sizeof(*tx); 9812 9813 if (count < 1 || count > 0xffff) 9814 return -EINVAL; 9815 9816 tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 9817 if (!tx) 9818 return -ENOMEM; 9819 9820 dev->_tx = tx; 9821 9822 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); 9823 spin_lock_init(&dev->tx_global_lock); 9824 9825 return 0; 9826 } 9827 9828 void netif_tx_stop_all_queues(struct net_device *dev) 9829 { 9830 unsigned int i; 9831 9832 for (i = 0; i < dev->num_tx_queues; i++) { 9833 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 9834 9835 netif_tx_stop_queue(txq); 9836 } 9837 } 9838 EXPORT_SYMBOL(netif_tx_stop_all_queues); 9839 9840 /** 9841 * register_netdevice - register a network device 9842 * @dev: device to register 9843 * 9844 * Take a completed network device structure and add it to the kernel 9845 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 9846 * chain. 0 is returned on success. A negative errno code is returned 9847 * on a failure to set up the device, or if the name is a duplicate. 9848 * 9849 * Callers must hold the rtnl semaphore. You may want 9850 * register_netdev() instead of this. 9851 * 9852 * BUGS: 9853 * The locking appears insufficient to guarantee two parallel registers 9854 * will not get the same name. 9855 */ 9856 9857 int register_netdevice(struct net_device *dev) 9858 { 9859 int ret; 9860 struct net *net = dev_net(dev); 9861 9862 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE < 9863 NETDEV_FEATURE_COUNT); 9864 BUG_ON(dev_boot_phase); 9865 ASSERT_RTNL(); 9866 9867 might_sleep(); 9868 9869 /* When net_device's are persistent, this will be fatal. */ 9870 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); 9871 BUG_ON(!net); 9872 9873 ret = ethtool_check_ops(dev->ethtool_ops); 9874 if (ret) 9875 return ret; 9876 9877 spin_lock_init(&dev->addr_list_lock); 9878 netdev_set_addr_lockdep_class(dev); 9879 9880 ret = dev_get_valid_name(net, dev, dev->name); 9881 if (ret < 0) 9882 goto out; 9883 9884 ret = -ENOMEM; 9885 dev->name_node = netdev_name_node_head_alloc(dev); 9886 if (!dev->name_node) 9887 goto out; 9888 9889 /* Init, if this function is available */ 9890 if (dev->netdev_ops->ndo_init) { 9891 ret = dev->netdev_ops->ndo_init(dev); 9892 if (ret) { 9893 if (ret > 0) 9894 ret = -EIO; 9895 goto err_free_name; 9896 } 9897 } 9898 9899 if (((dev->hw_features | dev->features) & 9900 NETIF_F_HW_VLAN_CTAG_FILTER) && 9901 (!dev->netdev_ops->ndo_vlan_rx_add_vid || 9902 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { 9903 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); 9904 ret = -EINVAL; 9905 goto err_uninit; 9906 } 9907 9908 ret = -EBUSY; 9909 if (!dev->ifindex) 9910 dev->ifindex = dev_new_index(net); 9911 else if (__dev_get_by_index(net, dev->ifindex)) 9912 goto err_uninit; 9913 9914 /* Transfer changeable features to wanted_features and enable 9915 * software offloads (GSO and GRO). 9916 */ 9917 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF); 9918 dev->features |= NETIF_F_SOFT_FEATURES; 9919 9920 if (dev->udp_tunnel_nic_info) { 9921 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT; 9922 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT; 9923 } 9924 9925 dev->wanted_features = dev->features & dev->hw_features; 9926 9927 if (!(dev->flags & IFF_LOOPBACK)) 9928 dev->hw_features |= NETIF_F_NOCACHE_COPY; 9929 9930 /* If IPv4 TCP segmentation offload is supported we should also 9931 * allow the device to enable segmenting the frame with the option 9932 * of ignoring a static IP ID value. This doesn't enable the 9933 * feature itself but allows the user to enable it later. 9934 */ 9935 if (dev->hw_features & NETIF_F_TSO) 9936 dev->hw_features |= NETIF_F_TSO_MANGLEID; 9937 if (dev->vlan_features & NETIF_F_TSO) 9938 dev->vlan_features |= NETIF_F_TSO_MANGLEID; 9939 if (dev->mpls_features & NETIF_F_TSO) 9940 dev->mpls_features |= NETIF_F_TSO_MANGLEID; 9941 if (dev->hw_enc_features & NETIF_F_TSO) 9942 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; 9943 9944 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. 9945 */ 9946 dev->vlan_features |= NETIF_F_HIGHDMA; 9947 9948 /* Make NETIF_F_SG inheritable to tunnel devices. 9949 */ 9950 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; 9951 9952 /* Make NETIF_F_SG inheritable to MPLS. 9953 */ 9954 dev->mpls_features |= NETIF_F_SG; 9955 9956 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); 9957 ret = notifier_to_errno(ret); 9958 if (ret) 9959 goto err_uninit; 9960 9961 ret = netdev_register_kobject(dev); 9962 if (ret) { 9963 dev->reg_state = NETREG_UNREGISTERED; 9964 goto err_uninit; 9965 } 9966 dev->reg_state = NETREG_REGISTERED; 9967 9968 __netdev_update_features(dev); 9969 9970 /* 9971 * Default initial state at registry is that the 9972 * device is present. 9973 */ 9974 9975 set_bit(__LINK_STATE_PRESENT, &dev->state); 9976 9977 linkwatch_init_dev(dev); 9978 9979 dev_init_scheduler(dev); 9980 9981 dev_hold_track(dev, &dev->dev_registered_tracker, GFP_KERNEL); 9982 list_netdevice(dev); 9983 9984 add_device_randomness(dev->dev_addr, dev->addr_len); 9985 9986 /* If the device has permanent device address, driver should 9987 * set dev_addr and also addr_assign_type should be set to 9988 * NET_ADDR_PERM (default value). 9989 */ 9990 if (dev->addr_assign_type == NET_ADDR_PERM) 9991 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); 9992 9993 /* Notify protocols, that a new device appeared. */ 9994 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); 9995 ret = notifier_to_errno(ret); 9996 if (ret) { 9997 /* Expect explicit free_netdev() on failure */ 9998 dev->needs_free_netdev = false; 9999 unregister_netdevice_queue(dev, NULL); 10000 goto out; 10001 } 10002 /* 10003 * Prevent userspace races by waiting until the network 10004 * device is fully setup before sending notifications. 10005 */ 10006 if (!dev->rtnl_link_ops || 10007 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 10008 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); 10009 10010 out: 10011 return ret; 10012 10013 err_uninit: 10014 if (dev->netdev_ops->ndo_uninit) 10015 dev->netdev_ops->ndo_uninit(dev); 10016 if (dev->priv_destructor) 10017 dev->priv_destructor(dev); 10018 err_free_name: 10019 netdev_name_node_free(dev->name_node); 10020 goto out; 10021 } 10022 EXPORT_SYMBOL(register_netdevice); 10023 10024 /** 10025 * init_dummy_netdev - init a dummy network device for NAPI 10026 * @dev: device to init 10027 * 10028 * This takes a network device structure and initialize the minimum 10029 * amount of fields so it can be used to schedule NAPI polls without 10030 * registering a full blown interface. This is to be used by drivers 10031 * that need to tie several hardware interfaces to a single NAPI 10032 * poll scheduler due to HW limitations. 10033 */ 10034 int init_dummy_netdev(struct net_device *dev) 10035 { 10036 /* Clear everything. Note we don't initialize spinlocks 10037 * are they aren't supposed to be taken by any of the 10038 * NAPI code and this dummy netdev is supposed to be 10039 * only ever used for NAPI polls 10040 */ 10041 memset(dev, 0, sizeof(struct net_device)); 10042 10043 /* make sure we BUG if trying to hit standard 10044 * register/unregister code path 10045 */ 10046 dev->reg_state = NETREG_DUMMY; 10047 10048 /* NAPI wants this */ 10049 INIT_LIST_HEAD(&dev->napi_list); 10050 10051 /* a dummy interface is started by default */ 10052 set_bit(__LINK_STATE_PRESENT, &dev->state); 10053 set_bit(__LINK_STATE_START, &dev->state); 10054 10055 /* napi_busy_loop stats accounting wants this */ 10056 dev_net_set(dev, &init_net); 10057 10058 /* Note : We dont allocate pcpu_refcnt for dummy devices, 10059 * because users of this 'device' dont need to change 10060 * its refcount. 10061 */ 10062 10063 return 0; 10064 } 10065 EXPORT_SYMBOL_GPL(init_dummy_netdev); 10066 10067 10068 /** 10069 * register_netdev - register a network device 10070 * @dev: device to register 10071 * 10072 * Take a completed network device structure and add it to the kernel 10073 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 10074 * chain. 0 is returned on success. A negative errno code is returned 10075 * on a failure to set up the device, or if the name is a duplicate. 10076 * 10077 * This is a wrapper around register_netdevice that takes the rtnl semaphore 10078 * and expands the device name if you passed a format string to 10079 * alloc_netdev. 10080 */ 10081 int register_netdev(struct net_device *dev) 10082 { 10083 int err; 10084 10085 if (rtnl_lock_killable()) 10086 return -EINTR; 10087 err = register_netdevice(dev); 10088 rtnl_unlock(); 10089 return err; 10090 } 10091 EXPORT_SYMBOL(register_netdev); 10092 10093 int netdev_refcnt_read(const struct net_device *dev) 10094 { 10095 #ifdef CONFIG_PCPU_DEV_REFCNT 10096 int i, refcnt = 0; 10097 10098 for_each_possible_cpu(i) 10099 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); 10100 return refcnt; 10101 #else 10102 return refcount_read(&dev->dev_refcnt); 10103 #endif 10104 } 10105 EXPORT_SYMBOL(netdev_refcnt_read); 10106 10107 int netdev_unregister_timeout_secs __read_mostly = 10; 10108 10109 #define WAIT_REFS_MIN_MSECS 1 10110 #define WAIT_REFS_MAX_MSECS 250 10111 /** 10112 * netdev_wait_allrefs_any - wait until all references are gone. 10113 * @list: list of net_devices to wait on 10114 * 10115 * This is called when unregistering network devices. 10116 * 10117 * Any protocol or device that holds a reference should register 10118 * for netdevice notification, and cleanup and put back the 10119 * reference if they receive an UNREGISTER event. 10120 * We can get stuck here if buggy protocols don't correctly 10121 * call dev_put. 10122 */ 10123 static struct net_device *netdev_wait_allrefs_any(struct list_head *list) 10124 { 10125 unsigned long rebroadcast_time, warning_time; 10126 struct net_device *dev; 10127 int wait = 0; 10128 10129 rebroadcast_time = warning_time = jiffies; 10130 10131 list_for_each_entry(dev, list, todo_list) 10132 if (netdev_refcnt_read(dev) == 1) 10133 return dev; 10134 10135 while (true) { 10136 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { 10137 rtnl_lock(); 10138 10139 /* Rebroadcast unregister notification */ 10140 list_for_each_entry(dev, list, todo_list) 10141 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10142 10143 __rtnl_unlock(); 10144 rcu_barrier(); 10145 rtnl_lock(); 10146 10147 list_for_each_entry(dev, list, todo_list) 10148 if (test_bit(__LINK_STATE_LINKWATCH_PENDING, 10149 &dev->state)) { 10150 /* We must not have linkwatch events 10151 * pending on unregister. If this 10152 * happens, we simply run the queue 10153 * unscheduled, resulting in a noop 10154 * for this device. 10155 */ 10156 linkwatch_run_queue(); 10157 break; 10158 } 10159 10160 __rtnl_unlock(); 10161 10162 rebroadcast_time = jiffies; 10163 } 10164 10165 if (!wait) { 10166 rcu_barrier(); 10167 wait = WAIT_REFS_MIN_MSECS; 10168 } else { 10169 msleep(wait); 10170 wait = min(wait << 1, WAIT_REFS_MAX_MSECS); 10171 } 10172 10173 list_for_each_entry(dev, list, todo_list) 10174 if (netdev_refcnt_read(dev) == 1) 10175 return dev; 10176 10177 if (time_after(jiffies, warning_time + 10178 netdev_unregister_timeout_secs * HZ)) { 10179 list_for_each_entry(dev, list, todo_list) { 10180 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", 10181 dev->name, netdev_refcnt_read(dev)); 10182 ref_tracker_dir_print(&dev->refcnt_tracker, 10); 10183 } 10184 10185 warning_time = jiffies; 10186 } 10187 } 10188 } 10189 10190 /* The sequence is: 10191 * 10192 * rtnl_lock(); 10193 * ... 10194 * register_netdevice(x1); 10195 * register_netdevice(x2); 10196 * ... 10197 * unregister_netdevice(y1); 10198 * unregister_netdevice(y2); 10199 * ... 10200 * rtnl_unlock(); 10201 * free_netdev(y1); 10202 * free_netdev(y2); 10203 * 10204 * We are invoked by rtnl_unlock(). 10205 * This allows us to deal with problems: 10206 * 1) We can delete sysfs objects which invoke hotplug 10207 * without deadlocking with linkwatch via keventd. 10208 * 2) Since we run with the RTNL semaphore not held, we can sleep 10209 * safely in order to wait for the netdev refcnt to drop to zero. 10210 * 10211 * We must not return until all unregister events added during 10212 * the interval the lock was held have been completed. 10213 */ 10214 void netdev_run_todo(void) 10215 { 10216 struct net_device *dev, *tmp; 10217 struct list_head list; 10218 #ifdef CONFIG_LOCKDEP 10219 struct list_head unlink_list; 10220 10221 list_replace_init(&net_unlink_list, &unlink_list); 10222 10223 while (!list_empty(&unlink_list)) { 10224 struct net_device *dev = list_first_entry(&unlink_list, 10225 struct net_device, 10226 unlink_list); 10227 list_del_init(&dev->unlink_list); 10228 dev->nested_level = dev->lower_level - 1; 10229 } 10230 #endif 10231 10232 /* Snapshot list, allow later requests */ 10233 list_replace_init(&net_todo_list, &list); 10234 10235 __rtnl_unlock(); 10236 10237 /* Wait for rcu callbacks to finish before next phase */ 10238 if (!list_empty(&list)) 10239 rcu_barrier(); 10240 10241 list_for_each_entry_safe(dev, tmp, &list, todo_list) { 10242 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { 10243 netdev_WARN(dev, "run_todo but not unregistering\n"); 10244 list_del(&dev->todo_list); 10245 continue; 10246 } 10247 10248 dev->reg_state = NETREG_UNREGISTERED; 10249 linkwatch_forget_dev(dev); 10250 } 10251 10252 while (!list_empty(&list)) { 10253 dev = netdev_wait_allrefs_any(&list); 10254 list_del(&dev->todo_list); 10255 10256 /* paranoia */ 10257 BUG_ON(netdev_refcnt_read(dev) != 1); 10258 BUG_ON(!list_empty(&dev->ptype_all)); 10259 BUG_ON(!list_empty(&dev->ptype_specific)); 10260 WARN_ON(rcu_access_pointer(dev->ip_ptr)); 10261 WARN_ON(rcu_access_pointer(dev->ip6_ptr)); 10262 #if IS_ENABLED(CONFIG_DECNET) 10263 WARN_ON(dev->dn_ptr); 10264 #endif 10265 if (dev->priv_destructor) 10266 dev->priv_destructor(dev); 10267 if (dev->needs_free_netdev) 10268 free_netdev(dev); 10269 10270 if (atomic_dec_and_test(&dev_net(dev)->dev_unreg_count)) 10271 wake_up(&netdev_unregistering_wq); 10272 10273 /* Free network device */ 10274 kobject_put(&dev->dev.kobj); 10275 } 10276 } 10277 10278 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has 10279 * all the same fields in the same order as net_device_stats, with only 10280 * the type differing, but rtnl_link_stats64 may have additional fields 10281 * at the end for newer counters. 10282 */ 10283 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, 10284 const struct net_device_stats *netdev_stats) 10285 { 10286 #if BITS_PER_LONG == 64 10287 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats)); 10288 memcpy(stats64, netdev_stats, sizeof(*netdev_stats)); 10289 /* zero out counters that only exist in rtnl_link_stats64 */ 10290 memset((char *)stats64 + sizeof(*netdev_stats), 0, 10291 sizeof(*stats64) - sizeof(*netdev_stats)); 10292 #else 10293 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long); 10294 const unsigned long *src = (const unsigned long *)netdev_stats; 10295 u64 *dst = (u64 *)stats64; 10296 10297 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); 10298 for (i = 0; i < n; i++) 10299 dst[i] = src[i]; 10300 /* zero out counters that only exist in rtnl_link_stats64 */ 10301 memset((char *)stats64 + n * sizeof(u64), 0, 10302 sizeof(*stats64) - n * sizeof(u64)); 10303 #endif 10304 } 10305 EXPORT_SYMBOL(netdev_stats_to_stats64); 10306 10307 struct net_device_core_stats __percpu *netdev_core_stats_alloc(struct net_device *dev) 10308 { 10309 struct net_device_core_stats __percpu *p; 10310 10311 p = alloc_percpu_gfp(struct net_device_core_stats, 10312 GFP_ATOMIC | __GFP_NOWARN); 10313 10314 if (p && cmpxchg(&dev->core_stats, NULL, p)) 10315 free_percpu(p); 10316 10317 /* This READ_ONCE() pairs with the cmpxchg() above */ 10318 return READ_ONCE(dev->core_stats); 10319 } 10320 EXPORT_SYMBOL(netdev_core_stats_alloc); 10321 10322 /** 10323 * dev_get_stats - get network device statistics 10324 * @dev: device to get statistics from 10325 * @storage: place to store stats 10326 * 10327 * Get network statistics from device. Return @storage. 10328 * The device driver may provide its own method by setting 10329 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; 10330 * otherwise the internal statistics structure is used. 10331 */ 10332 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, 10333 struct rtnl_link_stats64 *storage) 10334 { 10335 const struct net_device_ops *ops = dev->netdev_ops; 10336 const struct net_device_core_stats __percpu *p; 10337 10338 if (ops->ndo_get_stats64) { 10339 memset(storage, 0, sizeof(*storage)); 10340 ops->ndo_get_stats64(dev, storage); 10341 } else if (ops->ndo_get_stats) { 10342 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); 10343 } else { 10344 netdev_stats_to_stats64(storage, &dev->stats); 10345 } 10346 10347 /* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */ 10348 p = READ_ONCE(dev->core_stats); 10349 if (p) { 10350 const struct net_device_core_stats *core_stats; 10351 int i; 10352 10353 for_each_possible_cpu(i) { 10354 core_stats = per_cpu_ptr(p, i); 10355 storage->rx_dropped += READ_ONCE(core_stats->rx_dropped); 10356 storage->tx_dropped += READ_ONCE(core_stats->tx_dropped); 10357 storage->rx_nohandler += READ_ONCE(core_stats->rx_nohandler); 10358 } 10359 } 10360 return storage; 10361 } 10362 EXPORT_SYMBOL(dev_get_stats); 10363 10364 /** 10365 * dev_fetch_sw_netstats - get per-cpu network device statistics 10366 * @s: place to store stats 10367 * @netstats: per-cpu network stats to read from 10368 * 10369 * Read per-cpu network statistics and populate the related fields in @s. 10370 */ 10371 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s, 10372 const struct pcpu_sw_netstats __percpu *netstats) 10373 { 10374 int cpu; 10375 10376 for_each_possible_cpu(cpu) { 10377 const struct pcpu_sw_netstats *stats; 10378 struct pcpu_sw_netstats tmp; 10379 unsigned int start; 10380 10381 stats = per_cpu_ptr(netstats, cpu); 10382 do { 10383 start = u64_stats_fetch_begin_irq(&stats->syncp); 10384 tmp.rx_packets = stats->rx_packets; 10385 tmp.rx_bytes = stats->rx_bytes; 10386 tmp.tx_packets = stats->tx_packets; 10387 tmp.tx_bytes = stats->tx_bytes; 10388 } while (u64_stats_fetch_retry_irq(&stats->syncp, start)); 10389 10390 s->rx_packets += tmp.rx_packets; 10391 s->rx_bytes += tmp.rx_bytes; 10392 s->tx_packets += tmp.tx_packets; 10393 s->tx_bytes += tmp.tx_bytes; 10394 } 10395 } 10396 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats); 10397 10398 /** 10399 * dev_get_tstats64 - ndo_get_stats64 implementation 10400 * @dev: device to get statistics from 10401 * @s: place to store stats 10402 * 10403 * Populate @s from dev->stats and dev->tstats. Can be used as 10404 * ndo_get_stats64() callback. 10405 */ 10406 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s) 10407 { 10408 netdev_stats_to_stats64(s, &dev->stats); 10409 dev_fetch_sw_netstats(s, dev->tstats); 10410 } 10411 EXPORT_SYMBOL_GPL(dev_get_tstats64); 10412 10413 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) 10414 { 10415 struct netdev_queue *queue = dev_ingress_queue(dev); 10416 10417 #ifdef CONFIG_NET_CLS_ACT 10418 if (queue) 10419 return queue; 10420 queue = kzalloc(sizeof(*queue), GFP_KERNEL); 10421 if (!queue) 10422 return NULL; 10423 netdev_init_one_queue(dev, queue, NULL); 10424 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); 10425 queue->qdisc_sleeping = &noop_qdisc; 10426 rcu_assign_pointer(dev->ingress_queue, queue); 10427 #endif 10428 return queue; 10429 } 10430 10431 static const struct ethtool_ops default_ethtool_ops; 10432 10433 void netdev_set_default_ethtool_ops(struct net_device *dev, 10434 const struct ethtool_ops *ops) 10435 { 10436 if (dev->ethtool_ops == &default_ethtool_ops) 10437 dev->ethtool_ops = ops; 10438 } 10439 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); 10440 10441 void netdev_freemem(struct net_device *dev) 10442 { 10443 char *addr = (char *)dev - dev->padded; 10444 10445 kvfree(addr); 10446 } 10447 10448 /** 10449 * alloc_netdev_mqs - allocate network device 10450 * @sizeof_priv: size of private data to allocate space for 10451 * @name: device name format string 10452 * @name_assign_type: origin of device name 10453 * @setup: callback to initialize device 10454 * @txqs: the number of TX subqueues to allocate 10455 * @rxqs: the number of RX subqueues to allocate 10456 * 10457 * Allocates a struct net_device with private data area for driver use 10458 * and performs basic initialization. Also allocates subqueue structs 10459 * for each queue on the device. 10460 */ 10461 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, 10462 unsigned char name_assign_type, 10463 void (*setup)(struct net_device *), 10464 unsigned int txqs, unsigned int rxqs) 10465 { 10466 struct net_device *dev; 10467 unsigned int alloc_size; 10468 struct net_device *p; 10469 10470 BUG_ON(strlen(name) >= sizeof(dev->name)); 10471 10472 if (txqs < 1) { 10473 pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); 10474 return NULL; 10475 } 10476 10477 if (rxqs < 1) { 10478 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); 10479 return NULL; 10480 } 10481 10482 alloc_size = sizeof(struct net_device); 10483 if (sizeof_priv) { 10484 /* ensure 32-byte alignment of private area */ 10485 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); 10486 alloc_size += sizeof_priv; 10487 } 10488 /* ensure 32-byte alignment of whole construct */ 10489 alloc_size += NETDEV_ALIGN - 1; 10490 10491 p = kvzalloc(alloc_size, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10492 if (!p) 10493 return NULL; 10494 10495 dev = PTR_ALIGN(p, NETDEV_ALIGN); 10496 dev->padded = (char *)dev - (char *)p; 10497 10498 ref_tracker_dir_init(&dev->refcnt_tracker, 128); 10499 #ifdef CONFIG_PCPU_DEV_REFCNT 10500 dev->pcpu_refcnt = alloc_percpu(int); 10501 if (!dev->pcpu_refcnt) 10502 goto free_dev; 10503 __dev_hold(dev); 10504 #else 10505 refcount_set(&dev->dev_refcnt, 1); 10506 #endif 10507 10508 if (dev_addr_init(dev)) 10509 goto free_pcpu; 10510 10511 dev_mc_init(dev); 10512 dev_uc_init(dev); 10513 10514 dev_net_set(dev, &init_net); 10515 10516 dev->gso_max_size = GSO_MAX_SIZE; 10517 dev->gso_max_segs = GSO_MAX_SEGS; 10518 dev->gro_max_size = GRO_MAX_SIZE; 10519 dev->upper_level = 1; 10520 dev->lower_level = 1; 10521 #ifdef CONFIG_LOCKDEP 10522 dev->nested_level = 0; 10523 INIT_LIST_HEAD(&dev->unlink_list); 10524 #endif 10525 10526 INIT_LIST_HEAD(&dev->napi_list); 10527 INIT_LIST_HEAD(&dev->unreg_list); 10528 INIT_LIST_HEAD(&dev->close_list); 10529 INIT_LIST_HEAD(&dev->link_watch_list); 10530 INIT_LIST_HEAD(&dev->adj_list.upper); 10531 INIT_LIST_HEAD(&dev->adj_list.lower); 10532 INIT_LIST_HEAD(&dev->ptype_all); 10533 INIT_LIST_HEAD(&dev->ptype_specific); 10534 INIT_LIST_HEAD(&dev->net_notifier_list); 10535 #ifdef CONFIG_NET_SCHED 10536 hash_init(dev->qdisc_hash); 10537 #endif 10538 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; 10539 setup(dev); 10540 10541 if (!dev->tx_queue_len) { 10542 dev->priv_flags |= IFF_NO_QUEUE; 10543 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; 10544 } 10545 10546 dev->num_tx_queues = txqs; 10547 dev->real_num_tx_queues = txqs; 10548 if (netif_alloc_netdev_queues(dev)) 10549 goto free_all; 10550 10551 dev->num_rx_queues = rxqs; 10552 dev->real_num_rx_queues = rxqs; 10553 if (netif_alloc_rx_queues(dev)) 10554 goto free_all; 10555 10556 strcpy(dev->name, name); 10557 dev->name_assign_type = name_assign_type; 10558 dev->group = INIT_NETDEV_GROUP; 10559 if (!dev->ethtool_ops) 10560 dev->ethtool_ops = &default_ethtool_ops; 10561 10562 nf_hook_netdev_init(dev); 10563 10564 return dev; 10565 10566 free_all: 10567 free_netdev(dev); 10568 return NULL; 10569 10570 free_pcpu: 10571 #ifdef CONFIG_PCPU_DEV_REFCNT 10572 free_percpu(dev->pcpu_refcnt); 10573 free_dev: 10574 #endif 10575 netdev_freemem(dev); 10576 return NULL; 10577 } 10578 EXPORT_SYMBOL(alloc_netdev_mqs); 10579 10580 /** 10581 * free_netdev - free network device 10582 * @dev: device 10583 * 10584 * This function does the last stage of destroying an allocated device 10585 * interface. The reference to the device object is released. If this 10586 * is the last reference then it will be freed.Must be called in process 10587 * context. 10588 */ 10589 void free_netdev(struct net_device *dev) 10590 { 10591 struct napi_struct *p, *n; 10592 10593 might_sleep(); 10594 10595 /* When called immediately after register_netdevice() failed the unwind 10596 * handling may still be dismantling the device. Handle that case by 10597 * deferring the free. 10598 */ 10599 if (dev->reg_state == NETREG_UNREGISTERING) { 10600 ASSERT_RTNL(); 10601 dev->needs_free_netdev = true; 10602 return; 10603 } 10604 10605 netif_free_tx_queues(dev); 10606 netif_free_rx_queues(dev); 10607 10608 kfree(rcu_dereference_protected(dev->ingress_queue, 1)); 10609 10610 /* Flush device addresses */ 10611 dev_addr_flush(dev); 10612 10613 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) 10614 netif_napi_del(p); 10615 10616 ref_tracker_dir_exit(&dev->refcnt_tracker); 10617 #ifdef CONFIG_PCPU_DEV_REFCNT 10618 free_percpu(dev->pcpu_refcnt); 10619 dev->pcpu_refcnt = NULL; 10620 #endif 10621 free_percpu(dev->core_stats); 10622 dev->core_stats = NULL; 10623 free_percpu(dev->xdp_bulkq); 10624 dev->xdp_bulkq = NULL; 10625 10626 /* Compatibility with error handling in drivers */ 10627 if (dev->reg_state == NETREG_UNINITIALIZED) { 10628 netdev_freemem(dev); 10629 return; 10630 } 10631 10632 BUG_ON(dev->reg_state != NETREG_UNREGISTERED); 10633 dev->reg_state = NETREG_RELEASED; 10634 10635 /* will free via device release */ 10636 put_device(&dev->dev); 10637 } 10638 EXPORT_SYMBOL(free_netdev); 10639 10640 /** 10641 * synchronize_net - Synchronize with packet receive processing 10642 * 10643 * Wait for packets currently being received to be done. 10644 * Does not block later packets from starting. 10645 */ 10646 void synchronize_net(void) 10647 { 10648 might_sleep(); 10649 if (rtnl_is_locked()) 10650 synchronize_rcu_expedited(); 10651 else 10652 synchronize_rcu(); 10653 } 10654 EXPORT_SYMBOL(synchronize_net); 10655 10656 /** 10657 * unregister_netdevice_queue - remove device from the kernel 10658 * @dev: device 10659 * @head: list 10660 * 10661 * This function shuts down a device interface and removes it 10662 * from the kernel tables. 10663 * If head not NULL, device is queued to be unregistered later. 10664 * 10665 * Callers must hold the rtnl semaphore. You may want 10666 * unregister_netdev() instead of this. 10667 */ 10668 10669 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) 10670 { 10671 ASSERT_RTNL(); 10672 10673 if (head) { 10674 list_move_tail(&dev->unreg_list, head); 10675 } else { 10676 LIST_HEAD(single); 10677 10678 list_add(&dev->unreg_list, &single); 10679 unregister_netdevice_many(&single); 10680 } 10681 } 10682 EXPORT_SYMBOL(unregister_netdevice_queue); 10683 10684 /** 10685 * unregister_netdevice_many - unregister many devices 10686 * @head: list of devices 10687 * 10688 * Note: As most callers use a stack allocated list_head, 10689 * we force a list_del() to make sure stack wont be corrupted later. 10690 */ 10691 void unregister_netdevice_many(struct list_head *head) 10692 { 10693 struct net_device *dev, *tmp; 10694 LIST_HEAD(close_head); 10695 10696 BUG_ON(dev_boot_phase); 10697 ASSERT_RTNL(); 10698 10699 if (list_empty(head)) 10700 return; 10701 10702 list_for_each_entry_safe(dev, tmp, head, unreg_list) { 10703 /* Some devices call without registering 10704 * for initialization unwind. Remove those 10705 * devices and proceed with the remaining. 10706 */ 10707 if (dev->reg_state == NETREG_UNINITIALIZED) { 10708 pr_debug("unregister_netdevice: device %s/%p never was registered\n", 10709 dev->name, dev); 10710 10711 WARN_ON(1); 10712 list_del(&dev->unreg_list); 10713 continue; 10714 } 10715 dev->dismantle = true; 10716 BUG_ON(dev->reg_state != NETREG_REGISTERED); 10717 } 10718 10719 /* If device is running, close it first. */ 10720 list_for_each_entry(dev, head, unreg_list) 10721 list_add_tail(&dev->close_list, &close_head); 10722 dev_close_many(&close_head, true); 10723 10724 list_for_each_entry(dev, head, unreg_list) { 10725 /* And unlink it from device chain. */ 10726 unlist_netdevice(dev); 10727 10728 dev->reg_state = NETREG_UNREGISTERING; 10729 } 10730 flush_all_backlogs(); 10731 10732 synchronize_net(); 10733 10734 list_for_each_entry(dev, head, unreg_list) { 10735 struct sk_buff *skb = NULL; 10736 10737 /* Shutdown queueing discipline. */ 10738 dev_shutdown(dev); 10739 10740 dev_xdp_uninstall(dev); 10741 10742 netdev_offload_xstats_disable_all(dev); 10743 10744 /* Notify protocols, that we are about to destroy 10745 * this device. They should clean all the things. 10746 */ 10747 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10748 10749 if (!dev->rtnl_link_ops || 10750 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 10751 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0, 10752 GFP_KERNEL, NULL, 0); 10753 10754 /* 10755 * Flush the unicast and multicast chains 10756 */ 10757 dev_uc_flush(dev); 10758 dev_mc_flush(dev); 10759 10760 netdev_name_node_alt_flush(dev); 10761 netdev_name_node_free(dev->name_node); 10762 10763 if (dev->netdev_ops->ndo_uninit) 10764 dev->netdev_ops->ndo_uninit(dev); 10765 10766 if (skb) 10767 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL); 10768 10769 /* Notifier chain MUST detach us all upper devices. */ 10770 WARN_ON(netdev_has_any_upper_dev(dev)); 10771 WARN_ON(netdev_has_any_lower_dev(dev)); 10772 10773 /* Remove entries from kobject tree */ 10774 netdev_unregister_kobject(dev); 10775 #ifdef CONFIG_XPS 10776 /* Remove XPS queueing entries */ 10777 netif_reset_xps_queues_gt(dev, 0); 10778 #endif 10779 } 10780 10781 synchronize_net(); 10782 10783 list_for_each_entry(dev, head, unreg_list) { 10784 dev_put_track(dev, &dev->dev_registered_tracker); 10785 net_set_todo(dev); 10786 } 10787 10788 list_del(head); 10789 } 10790 EXPORT_SYMBOL(unregister_netdevice_many); 10791 10792 /** 10793 * unregister_netdev - remove device from the kernel 10794 * @dev: device 10795 * 10796 * This function shuts down a device interface and removes it 10797 * from the kernel tables. 10798 * 10799 * This is just a wrapper for unregister_netdevice that takes 10800 * the rtnl semaphore. In general you want to use this and not 10801 * unregister_netdevice. 10802 */ 10803 void unregister_netdev(struct net_device *dev) 10804 { 10805 rtnl_lock(); 10806 unregister_netdevice(dev); 10807 rtnl_unlock(); 10808 } 10809 EXPORT_SYMBOL(unregister_netdev); 10810 10811 /** 10812 * __dev_change_net_namespace - move device to different nethost namespace 10813 * @dev: device 10814 * @net: network namespace 10815 * @pat: If not NULL name pattern to try if the current device name 10816 * is already taken in the destination network namespace. 10817 * @new_ifindex: If not zero, specifies device index in the target 10818 * namespace. 10819 * 10820 * This function shuts down a device interface and moves it 10821 * to a new network namespace. On success 0 is returned, on 10822 * a failure a netagive errno code is returned. 10823 * 10824 * Callers must hold the rtnl semaphore. 10825 */ 10826 10827 int __dev_change_net_namespace(struct net_device *dev, struct net *net, 10828 const char *pat, int new_ifindex) 10829 { 10830 struct net *net_old = dev_net(dev); 10831 int err, new_nsid; 10832 10833 ASSERT_RTNL(); 10834 10835 /* Don't allow namespace local devices to be moved. */ 10836 err = -EINVAL; 10837 if (dev->features & NETIF_F_NETNS_LOCAL) 10838 goto out; 10839 10840 /* Ensure the device has been registrered */ 10841 if (dev->reg_state != NETREG_REGISTERED) 10842 goto out; 10843 10844 /* Get out if there is nothing todo */ 10845 err = 0; 10846 if (net_eq(net_old, net)) 10847 goto out; 10848 10849 /* Pick the destination device name, and ensure 10850 * we can use it in the destination network namespace. 10851 */ 10852 err = -EEXIST; 10853 if (netdev_name_in_use(net, dev->name)) { 10854 /* We get here if we can't use the current device name */ 10855 if (!pat) 10856 goto out; 10857 err = dev_get_valid_name(net, dev, pat); 10858 if (err < 0) 10859 goto out; 10860 } 10861 10862 /* Check that new_ifindex isn't used yet. */ 10863 err = -EBUSY; 10864 if (new_ifindex && __dev_get_by_index(net, new_ifindex)) 10865 goto out; 10866 10867 /* 10868 * And now a mini version of register_netdevice unregister_netdevice. 10869 */ 10870 10871 /* If device is running close it first. */ 10872 dev_close(dev); 10873 10874 /* And unlink it from device chain */ 10875 unlist_netdevice(dev); 10876 10877 synchronize_net(); 10878 10879 /* Shutdown queueing discipline. */ 10880 dev_shutdown(dev); 10881 10882 /* Notify protocols, that we are about to destroy 10883 * this device. They should clean all the things. 10884 * 10885 * Note that dev->reg_state stays at NETREG_REGISTERED. 10886 * This is wanted because this way 8021q and macvlan know 10887 * the device is just moving and can keep their slaves up. 10888 */ 10889 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10890 rcu_barrier(); 10891 10892 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL); 10893 /* If there is an ifindex conflict assign a new one */ 10894 if (!new_ifindex) { 10895 if (__dev_get_by_index(net, dev->ifindex)) 10896 new_ifindex = dev_new_index(net); 10897 else 10898 new_ifindex = dev->ifindex; 10899 } 10900 10901 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid, 10902 new_ifindex); 10903 10904 /* 10905 * Flush the unicast and multicast chains 10906 */ 10907 dev_uc_flush(dev); 10908 dev_mc_flush(dev); 10909 10910 /* Send a netdev-removed uevent to the old namespace */ 10911 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); 10912 netdev_adjacent_del_links(dev); 10913 10914 /* Move per-net netdevice notifiers that are following the netdevice */ 10915 move_netdevice_notifiers_dev_net(dev, net); 10916 10917 /* Actually switch the network namespace */ 10918 dev_net_set(dev, net); 10919 dev->ifindex = new_ifindex; 10920 10921 /* Send a netdev-add uevent to the new namespace */ 10922 kobject_uevent(&dev->dev.kobj, KOBJ_ADD); 10923 netdev_adjacent_add_links(dev); 10924 10925 /* Fixup kobjects */ 10926 err = device_rename(&dev->dev, dev->name); 10927 WARN_ON(err); 10928 10929 /* Adapt owner in case owning user namespace of target network 10930 * namespace is different from the original one. 10931 */ 10932 err = netdev_change_owner(dev, net_old, net); 10933 WARN_ON(err); 10934 10935 /* Add the device back in the hashes */ 10936 list_netdevice(dev); 10937 10938 /* Notify protocols, that a new device appeared. */ 10939 call_netdevice_notifiers(NETDEV_REGISTER, dev); 10940 10941 /* 10942 * Prevent userspace races by waiting until the network 10943 * device is fully setup before sending notifications. 10944 */ 10945 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); 10946 10947 synchronize_net(); 10948 err = 0; 10949 out: 10950 return err; 10951 } 10952 EXPORT_SYMBOL_GPL(__dev_change_net_namespace); 10953 10954 static int dev_cpu_dead(unsigned int oldcpu) 10955 { 10956 struct sk_buff **list_skb; 10957 struct sk_buff *skb; 10958 unsigned int cpu; 10959 struct softnet_data *sd, *oldsd, *remsd = NULL; 10960 10961 local_irq_disable(); 10962 cpu = smp_processor_id(); 10963 sd = &per_cpu(softnet_data, cpu); 10964 oldsd = &per_cpu(softnet_data, oldcpu); 10965 10966 /* Find end of our completion_queue. */ 10967 list_skb = &sd->completion_queue; 10968 while (*list_skb) 10969 list_skb = &(*list_skb)->next; 10970 /* Append completion queue from offline CPU. */ 10971 *list_skb = oldsd->completion_queue; 10972 oldsd->completion_queue = NULL; 10973 10974 /* Append output queue from offline CPU. */ 10975 if (oldsd->output_queue) { 10976 *sd->output_queue_tailp = oldsd->output_queue; 10977 sd->output_queue_tailp = oldsd->output_queue_tailp; 10978 oldsd->output_queue = NULL; 10979 oldsd->output_queue_tailp = &oldsd->output_queue; 10980 } 10981 /* Append NAPI poll list from offline CPU, with one exception : 10982 * process_backlog() must be called by cpu owning percpu backlog. 10983 * We properly handle process_queue & input_pkt_queue later. 10984 */ 10985 while (!list_empty(&oldsd->poll_list)) { 10986 struct napi_struct *napi = list_first_entry(&oldsd->poll_list, 10987 struct napi_struct, 10988 poll_list); 10989 10990 list_del_init(&napi->poll_list); 10991 if (napi->poll == process_backlog) 10992 napi->state = 0; 10993 else 10994 ____napi_schedule(sd, napi); 10995 } 10996 10997 raise_softirq_irqoff(NET_TX_SOFTIRQ); 10998 local_irq_enable(); 10999 11000 #ifdef CONFIG_RPS 11001 remsd = oldsd->rps_ipi_list; 11002 oldsd->rps_ipi_list = NULL; 11003 #endif 11004 /* send out pending IPI's on offline CPU */ 11005 net_rps_send_ipi(remsd); 11006 11007 /* Process offline CPU's input_pkt_queue */ 11008 while ((skb = __skb_dequeue(&oldsd->process_queue))) { 11009 netif_rx(skb); 11010 input_queue_head_incr(oldsd); 11011 } 11012 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { 11013 netif_rx(skb); 11014 input_queue_head_incr(oldsd); 11015 } 11016 11017 return 0; 11018 } 11019 11020 /** 11021 * netdev_increment_features - increment feature set by one 11022 * @all: current feature set 11023 * @one: new feature set 11024 * @mask: mask feature set 11025 * 11026 * Computes a new feature set after adding a device with feature set 11027 * @one to the master device with current feature set @all. Will not 11028 * enable anything that is off in @mask. Returns the new feature set. 11029 */ 11030 netdev_features_t netdev_increment_features(netdev_features_t all, 11031 netdev_features_t one, netdev_features_t mask) 11032 { 11033 if (mask & NETIF_F_HW_CSUM) 11034 mask |= NETIF_F_CSUM_MASK; 11035 mask |= NETIF_F_VLAN_CHALLENGED; 11036 11037 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; 11038 all &= one | ~NETIF_F_ALL_FOR_ALL; 11039 11040 /* If one device supports hw checksumming, set for all. */ 11041 if (all & NETIF_F_HW_CSUM) 11042 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); 11043 11044 return all; 11045 } 11046 EXPORT_SYMBOL(netdev_increment_features); 11047 11048 static struct hlist_head * __net_init netdev_create_hash(void) 11049 { 11050 int i; 11051 struct hlist_head *hash; 11052 11053 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL); 11054 if (hash != NULL) 11055 for (i = 0; i < NETDEV_HASHENTRIES; i++) 11056 INIT_HLIST_HEAD(&hash[i]); 11057 11058 return hash; 11059 } 11060 11061 /* Initialize per network namespace state */ 11062 static int __net_init netdev_init(struct net *net) 11063 { 11064 BUILD_BUG_ON(GRO_HASH_BUCKETS > 11065 8 * sizeof_field(struct napi_struct, gro_bitmask)); 11066 11067 INIT_LIST_HEAD(&net->dev_base_head); 11068 11069 net->dev_name_head = netdev_create_hash(); 11070 if (net->dev_name_head == NULL) 11071 goto err_name; 11072 11073 net->dev_index_head = netdev_create_hash(); 11074 if (net->dev_index_head == NULL) 11075 goto err_idx; 11076 11077 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain); 11078 11079 return 0; 11080 11081 err_idx: 11082 kfree(net->dev_name_head); 11083 err_name: 11084 return -ENOMEM; 11085 } 11086 11087 /** 11088 * netdev_drivername - network driver for the device 11089 * @dev: network device 11090 * 11091 * Determine network driver for device. 11092 */ 11093 const char *netdev_drivername(const struct net_device *dev) 11094 { 11095 const struct device_driver *driver; 11096 const struct device *parent; 11097 const char *empty = ""; 11098 11099 parent = dev->dev.parent; 11100 if (!parent) 11101 return empty; 11102 11103 driver = parent->driver; 11104 if (driver && driver->name) 11105 return driver->name; 11106 return empty; 11107 } 11108 11109 static void __netdev_printk(const char *level, const struct net_device *dev, 11110 struct va_format *vaf) 11111 { 11112 if (dev && dev->dev.parent) { 11113 dev_printk_emit(level[1] - '0', 11114 dev->dev.parent, 11115 "%s %s %s%s: %pV", 11116 dev_driver_string(dev->dev.parent), 11117 dev_name(dev->dev.parent), 11118 netdev_name(dev), netdev_reg_state(dev), 11119 vaf); 11120 } else if (dev) { 11121 printk("%s%s%s: %pV", 11122 level, netdev_name(dev), netdev_reg_state(dev), vaf); 11123 } else { 11124 printk("%s(NULL net_device): %pV", level, vaf); 11125 } 11126 } 11127 11128 void netdev_printk(const char *level, const struct net_device *dev, 11129 const char *format, ...) 11130 { 11131 struct va_format vaf; 11132 va_list args; 11133 11134 va_start(args, format); 11135 11136 vaf.fmt = format; 11137 vaf.va = &args; 11138 11139 __netdev_printk(level, dev, &vaf); 11140 11141 va_end(args); 11142 } 11143 EXPORT_SYMBOL(netdev_printk); 11144 11145 #define define_netdev_printk_level(func, level) \ 11146 void func(const struct net_device *dev, const char *fmt, ...) \ 11147 { \ 11148 struct va_format vaf; \ 11149 va_list args; \ 11150 \ 11151 va_start(args, fmt); \ 11152 \ 11153 vaf.fmt = fmt; \ 11154 vaf.va = &args; \ 11155 \ 11156 __netdev_printk(level, dev, &vaf); \ 11157 \ 11158 va_end(args); \ 11159 } \ 11160 EXPORT_SYMBOL(func); 11161 11162 define_netdev_printk_level(netdev_emerg, KERN_EMERG); 11163 define_netdev_printk_level(netdev_alert, KERN_ALERT); 11164 define_netdev_printk_level(netdev_crit, KERN_CRIT); 11165 define_netdev_printk_level(netdev_err, KERN_ERR); 11166 define_netdev_printk_level(netdev_warn, KERN_WARNING); 11167 define_netdev_printk_level(netdev_notice, KERN_NOTICE); 11168 define_netdev_printk_level(netdev_info, KERN_INFO); 11169 11170 static void __net_exit netdev_exit(struct net *net) 11171 { 11172 kfree(net->dev_name_head); 11173 kfree(net->dev_index_head); 11174 if (net != &init_net) 11175 WARN_ON_ONCE(!list_empty(&net->dev_base_head)); 11176 } 11177 11178 static struct pernet_operations __net_initdata netdev_net_ops = { 11179 .init = netdev_init, 11180 .exit = netdev_exit, 11181 }; 11182 11183 static void __net_exit default_device_exit_net(struct net *net) 11184 { 11185 struct net_device *dev, *aux; 11186 /* 11187 * Push all migratable network devices back to the 11188 * initial network namespace 11189 */ 11190 ASSERT_RTNL(); 11191 for_each_netdev_safe(net, dev, aux) { 11192 int err; 11193 char fb_name[IFNAMSIZ]; 11194 11195 /* Ignore unmoveable devices (i.e. loopback) */ 11196 if (dev->features & NETIF_F_NETNS_LOCAL) 11197 continue; 11198 11199 /* Leave virtual devices for the generic cleanup */ 11200 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund) 11201 continue; 11202 11203 /* Push remaining network devices to init_net */ 11204 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); 11205 if (netdev_name_in_use(&init_net, fb_name)) 11206 snprintf(fb_name, IFNAMSIZ, "dev%%d"); 11207 err = dev_change_net_namespace(dev, &init_net, fb_name); 11208 if (err) { 11209 pr_emerg("%s: failed to move %s to init_net: %d\n", 11210 __func__, dev->name, err); 11211 BUG(); 11212 } 11213 } 11214 } 11215 11216 static void __net_exit default_device_exit_batch(struct list_head *net_list) 11217 { 11218 /* At exit all network devices most be removed from a network 11219 * namespace. Do this in the reverse order of registration. 11220 * Do this across as many network namespaces as possible to 11221 * improve batching efficiency. 11222 */ 11223 struct net_device *dev; 11224 struct net *net; 11225 LIST_HEAD(dev_kill_list); 11226 11227 rtnl_lock(); 11228 list_for_each_entry(net, net_list, exit_list) { 11229 default_device_exit_net(net); 11230 cond_resched(); 11231 } 11232 11233 list_for_each_entry(net, net_list, exit_list) { 11234 for_each_netdev_reverse(net, dev) { 11235 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) 11236 dev->rtnl_link_ops->dellink(dev, &dev_kill_list); 11237 else 11238 unregister_netdevice_queue(dev, &dev_kill_list); 11239 } 11240 } 11241 unregister_netdevice_many(&dev_kill_list); 11242 rtnl_unlock(); 11243 } 11244 11245 static struct pernet_operations __net_initdata default_device_ops = { 11246 .exit_batch = default_device_exit_batch, 11247 }; 11248 11249 /* 11250 * Initialize the DEV module. At boot time this walks the device list and 11251 * unhooks any devices that fail to initialise (normally hardware not 11252 * present) and leaves us with a valid list of present and active devices. 11253 * 11254 */ 11255 11256 /* 11257 * This is called single threaded during boot, so no need 11258 * to take the rtnl semaphore. 11259 */ 11260 static int __init net_dev_init(void) 11261 { 11262 int i, rc = -ENOMEM; 11263 11264 BUG_ON(!dev_boot_phase); 11265 11266 if (dev_proc_init()) 11267 goto out; 11268 11269 if (netdev_kobject_init()) 11270 goto out; 11271 11272 INIT_LIST_HEAD(&ptype_all); 11273 for (i = 0; i < PTYPE_HASH_SIZE; i++) 11274 INIT_LIST_HEAD(&ptype_base[i]); 11275 11276 if (register_pernet_subsys(&netdev_net_ops)) 11277 goto out; 11278 11279 /* 11280 * Initialise the packet receive queues. 11281 */ 11282 11283 for_each_possible_cpu(i) { 11284 struct work_struct *flush = per_cpu_ptr(&flush_works, i); 11285 struct softnet_data *sd = &per_cpu(softnet_data, i); 11286 11287 INIT_WORK(flush, flush_backlog); 11288 11289 skb_queue_head_init(&sd->input_pkt_queue); 11290 skb_queue_head_init(&sd->process_queue); 11291 #ifdef CONFIG_XFRM_OFFLOAD 11292 skb_queue_head_init(&sd->xfrm_backlog); 11293 #endif 11294 INIT_LIST_HEAD(&sd->poll_list); 11295 sd->output_queue_tailp = &sd->output_queue; 11296 #ifdef CONFIG_RPS 11297 INIT_CSD(&sd->csd, rps_trigger_softirq, sd); 11298 sd->cpu = i; 11299 #endif 11300 11301 init_gro_hash(&sd->backlog); 11302 sd->backlog.poll = process_backlog; 11303 sd->backlog.weight = weight_p; 11304 } 11305 11306 dev_boot_phase = 0; 11307 11308 /* The loopback device is special if any other network devices 11309 * is present in a network namespace the loopback device must 11310 * be present. Since we now dynamically allocate and free the 11311 * loopback device ensure this invariant is maintained by 11312 * keeping the loopback device as the first device on the 11313 * list of network devices. Ensuring the loopback devices 11314 * is the first device that appears and the last network device 11315 * that disappears. 11316 */ 11317 if (register_pernet_device(&loopback_net_ops)) 11318 goto out; 11319 11320 if (register_pernet_device(&default_device_ops)) 11321 goto out; 11322 11323 open_softirq(NET_TX_SOFTIRQ, net_tx_action); 11324 open_softirq(NET_RX_SOFTIRQ, net_rx_action); 11325 11326 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", 11327 NULL, dev_cpu_dead); 11328 WARN_ON(rc < 0); 11329 rc = 0; 11330 out: 11331 return rc; 11332 } 11333 11334 subsys_initcall(net_dev_init); 11335