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/bitmap.h> 73 #include <linux/capability.h> 74 #include <linux/cpu.h> 75 #include <linux/types.h> 76 #include <linux/kernel.h> 77 #include <linux/hash.h> 78 #include <linux/slab.h> 79 #include <linux/sched.h> 80 #include <linux/sched/mm.h> 81 #include <linux/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 <net/tcx.h> 111 #include <linux/highmem.h> 112 #include <linux/init.h> 113 #include <linux/module.h> 114 #include <linux/netpoll.h> 115 #include <linux/rcupdate.h> 116 #include <linux/delay.h> 117 #include <net/iw_handler.h> 118 #include <asm/current.h> 119 #include <linux/audit.h> 120 #include <linux/dmaengine.h> 121 #include <linux/err.h> 122 #include <linux/ctype.h> 123 #include <linux/if_arp.h> 124 #include <linux/if_vlan.h> 125 #include <linux/ip.h> 126 #include <net/ip.h> 127 #include <net/mpls.h> 128 #include <linux/ipv6.h> 129 #include <linux/in.h> 130 #include <linux/jhash.h> 131 #include <linux/random.h> 132 #include <trace/events/napi.h> 133 #include <trace/events/net.h> 134 #include <trace/events/skb.h> 135 #include <trace/events/qdisc.h> 136 #include <trace/events/xdp.h> 137 #include <linux/inetdevice.h> 138 #include <linux/cpu_rmap.h> 139 #include <linux/static_key.h> 140 #include <linux/hashtable.h> 141 #include <linux/vmalloc.h> 142 #include <linux/if_macvlan.h> 143 #include <linux/errqueue.h> 144 #include <linux/hrtimer.h> 145 #include <linux/netfilter_netdev.h> 146 #include <linux/crash_dump.h> 147 #include <linux/sctp.h> 148 #include <net/udp_tunnel.h> 149 #include <linux/net_namespace.h> 150 #include <linux/indirect_call_wrapper.h> 151 #include <net/devlink.h> 152 #include <linux/pm_runtime.h> 153 #include <linux/prandom.h> 154 #include <linux/once_lite.h> 155 #include <net/netdev_rx_queue.h> 156 157 #include "dev.h" 158 #include "net-sysfs.h" 159 160 static DEFINE_SPINLOCK(ptype_lock); 161 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; 162 struct list_head ptype_all __read_mostly; /* Taps */ 163 164 static int netif_rx_internal(struct sk_buff *skb); 165 static int call_netdevice_notifiers_extack(unsigned long val, 166 struct net_device *dev, 167 struct netlink_ext_ack *extack); 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 kfree(name_node->name); 348 netdev_name_node_free(name_node); 349 } 350 351 int netdev_name_node_alt_destroy(struct net_device *dev, const char *name) 352 { 353 struct netdev_name_node *name_node; 354 struct net *net = dev_net(dev); 355 356 name_node = netdev_name_node_lookup(net, name); 357 if (!name_node) 358 return -ENOENT; 359 /* lookup might have found our primary name or a name belonging 360 * to another device. 361 */ 362 if (name_node == dev->name_node || name_node->dev != dev) 363 return -EINVAL; 364 365 netdev_name_node_del(name_node); 366 synchronize_rcu(); 367 __netdev_name_node_alt_destroy(name_node); 368 369 return 0; 370 } 371 372 static void netdev_name_node_alt_flush(struct net_device *dev) 373 { 374 struct netdev_name_node *name_node, *tmp; 375 376 list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list) 377 __netdev_name_node_alt_destroy(name_node); 378 } 379 380 /* Device list insertion */ 381 static void list_netdevice(struct net_device *dev) 382 { 383 struct netdev_name_node *name_node; 384 struct net *net = dev_net(dev); 385 386 ASSERT_RTNL(); 387 388 write_lock(&dev_base_lock); 389 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); 390 netdev_name_node_add(net, dev->name_node); 391 hlist_add_head_rcu(&dev->index_hlist, 392 dev_index_hash(net, dev->ifindex)); 393 write_unlock(&dev_base_lock); 394 395 netdev_for_each_altname(dev, name_node) 396 netdev_name_node_add(net, name_node); 397 398 /* We reserved the ifindex, this can't fail */ 399 WARN_ON(xa_store(&net->dev_by_index, dev->ifindex, dev, GFP_KERNEL)); 400 401 dev_base_seq_inc(net); 402 } 403 404 /* Device list removal 405 * caller must respect a RCU grace period before freeing/reusing dev 406 */ 407 static void unlist_netdevice(struct net_device *dev, bool lock) 408 { 409 struct netdev_name_node *name_node; 410 struct net *net = dev_net(dev); 411 412 ASSERT_RTNL(); 413 414 xa_erase(&net->dev_by_index, dev->ifindex); 415 416 netdev_for_each_altname(dev, name_node) 417 netdev_name_node_del(name_node); 418 419 /* Unlink dev from the device chain */ 420 if (lock) 421 write_lock(&dev_base_lock); 422 list_del_rcu(&dev->dev_list); 423 netdev_name_node_del(dev->name_node); 424 hlist_del_rcu(&dev->index_hlist); 425 if (lock) 426 write_unlock(&dev_base_lock); 427 428 dev_base_seq_inc(dev_net(dev)); 429 } 430 431 /* 432 * Our notifier list 433 */ 434 435 static RAW_NOTIFIER_HEAD(netdev_chain); 436 437 /* 438 * Device drivers call our routines to queue packets here. We empty the 439 * queue in the local softnet handler. 440 */ 441 442 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); 443 EXPORT_PER_CPU_SYMBOL(softnet_data); 444 445 #ifdef CONFIG_LOCKDEP 446 /* 447 * register_netdevice() inits txq->_xmit_lock and sets lockdep class 448 * according to dev->type 449 */ 450 static const unsigned short netdev_lock_type[] = { 451 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, 452 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, 453 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, 454 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, 455 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, 456 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, 457 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, 458 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, 459 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, 460 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, 461 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, 462 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, 463 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, 464 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, 465 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; 466 467 static const char *const netdev_lock_name[] = { 468 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", 469 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", 470 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", 471 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", 472 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", 473 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", 474 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", 475 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", 476 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", 477 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", 478 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", 479 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", 480 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", 481 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", 482 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; 483 484 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; 485 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; 486 487 static inline unsigned short netdev_lock_pos(unsigned short dev_type) 488 { 489 int i; 490 491 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) 492 if (netdev_lock_type[i] == dev_type) 493 return i; 494 /* the last key is used by default */ 495 return ARRAY_SIZE(netdev_lock_type) - 1; 496 } 497 498 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 499 unsigned short dev_type) 500 { 501 int i; 502 503 i = netdev_lock_pos(dev_type); 504 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], 505 netdev_lock_name[i]); 506 } 507 508 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 509 { 510 int i; 511 512 i = netdev_lock_pos(dev->type); 513 lockdep_set_class_and_name(&dev->addr_list_lock, 514 &netdev_addr_lock_key[i], 515 netdev_lock_name[i]); 516 } 517 #else 518 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 519 unsigned short dev_type) 520 { 521 } 522 523 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 524 { 525 } 526 #endif 527 528 /******************************************************************************* 529 * 530 * Protocol management and registration routines 531 * 532 *******************************************************************************/ 533 534 535 /* 536 * Add a protocol ID to the list. Now that the input handler is 537 * smarter we can dispense with all the messy stuff that used to be 538 * here. 539 * 540 * BEWARE!!! Protocol handlers, mangling input packets, 541 * MUST BE last in hash buckets and checking protocol handlers 542 * MUST start from promiscuous ptype_all chain in net_bh. 543 * It is true now, do not change it. 544 * Explanation follows: if protocol handler, mangling packet, will 545 * be the first on list, it is not able to sense, that packet 546 * is cloned and should be copied-on-write, so that it will 547 * change it and subsequent readers will get broken packet. 548 * --ANK (980803) 549 */ 550 551 static inline struct list_head *ptype_head(const struct packet_type *pt) 552 { 553 if (pt->type == htons(ETH_P_ALL)) 554 return pt->dev ? &pt->dev->ptype_all : &ptype_all; 555 else 556 return pt->dev ? &pt->dev->ptype_specific : 557 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; 558 } 559 560 /** 561 * dev_add_pack - add packet handler 562 * @pt: packet type declaration 563 * 564 * Add a protocol handler to the networking stack. The passed &packet_type 565 * is linked into kernel lists and may not be freed until it has been 566 * removed from the kernel lists. 567 * 568 * This call does not sleep therefore it can not 569 * guarantee all CPU's that are in middle of receiving packets 570 * will see the new packet type (until the next received packet). 571 */ 572 573 void dev_add_pack(struct packet_type *pt) 574 { 575 struct list_head *head = ptype_head(pt); 576 577 spin_lock(&ptype_lock); 578 list_add_rcu(&pt->list, head); 579 spin_unlock(&ptype_lock); 580 } 581 EXPORT_SYMBOL(dev_add_pack); 582 583 /** 584 * __dev_remove_pack - remove packet handler 585 * @pt: packet type declaration 586 * 587 * Remove a protocol handler that was previously added to the kernel 588 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 589 * from the kernel lists and can be freed or reused once this function 590 * returns. 591 * 592 * The packet type might still be in use by receivers 593 * and must not be freed until after all the CPU's have gone 594 * through a quiescent state. 595 */ 596 void __dev_remove_pack(struct packet_type *pt) 597 { 598 struct list_head *head = ptype_head(pt); 599 struct packet_type *pt1; 600 601 spin_lock(&ptype_lock); 602 603 list_for_each_entry(pt1, head, list) { 604 if (pt == pt1) { 605 list_del_rcu(&pt->list); 606 goto out; 607 } 608 } 609 610 pr_warn("dev_remove_pack: %p not found\n", pt); 611 out: 612 spin_unlock(&ptype_lock); 613 } 614 EXPORT_SYMBOL(__dev_remove_pack); 615 616 /** 617 * dev_remove_pack - remove packet handler 618 * @pt: packet type declaration 619 * 620 * Remove a protocol handler that was previously added to the kernel 621 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 622 * from the kernel lists and can be freed or reused once this function 623 * returns. 624 * 625 * This call sleeps to guarantee that no CPU is looking at the packet 626 * type after return. 627 */ 628 void dev_remove_pack(struct packet_type *pt) 629 { 630 __dev_remove_pack(pt); 631 632 synchronize_net(); 633 } 634 EXPORT_SYMBOL(dev_remove_pack); 635 636 637 /******************************************************************************* 638 * 639 * Device Interface Subroutines 640 * 641 *******************************************************************************/ 642 643 /** 644 * dev_get_iflink - get 'iflink' value of a interface 645 * @dev: targeted interface 646 * 647 * Indicates the ifindex the interface is linked to. 648 * Physical interfaces have the same 'ifindex' and 'iflink' values. 649 */ 650 651 int dev_get_iflink(const struct net_device *dev) 652 { 653 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink) 654 return dev->netdev_ops->ndo_get_iflink(dev); 655 656 return dev->ifindex; 657 } 658 EXPORT_SYMBOL(dev_get_iflink); 659 660 /** 661 * dev_fill_metadata_dst - Retrieve tunnel egress information. 662 * @dev: targeted interface 663 * @skb: The packet. 664 * 665 * For better visibility of tunnel traffic OVS needs to retrieve 666 * egress tunnel information for a packet. Following API allows 667 * user to get this info. 668 */ 669 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) 670 { 671 struct ip_tunnel_info *info; 672 673 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst) 674 return -EINVAL; 675 676 info = skb_tunnel_info_unclone(skb); 677 if (!info) 678 return -ENOMEM; 679 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX))) 680 return -EINVAL; 681 682 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb); 683 } 684 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst); 685 686 static struct net_device_path *dev_fwd_path(struct net_device_path_stack *stack) 687 { 688 int k = stack->num_paths++; 689 690 if (WARN_ON_ONCE(k >= NET_DEVICE_PATH_STACK_MAX)) 691 return NULL; 692 693 return &stack->path[k]; 694 } 695 696 int dev_fill_forward_path(const struct net_device *dev, const u8 *daddr, 697 struct net_device_path_stack *stack) 698 { 699 const struct net_device *last_dev; 700 struct net_device_path_ctx ctx = { 701 .dev = dev, 702 }; 703 struct net_device_path *path; 704 int ret = 0; 705 706 memcpy(ctx.daddr, daddr, sizeof(ctx.daddr)); 707 stack->num_paths = 0; 708 while (ctx.dev && ctx.dev->netdev_ops->ndo_fill_forward_path) { 709 last_dev = ctx.dev; 710 path = dev_fwd_path(stack); 711 if (!path) 712 return -1; 713 714 memset(path, 0, sizeof(struct net_device_path)); 715 ret = ctx.dev->netdev_ops->ndo_fill_forward_path(&ctx, path); 716 if (ret < 0) 717 return -1; 718 719 if (WARN_ON_ONCE(last_dev == ctx.dev)) 720 return -1; 721 } 722 723 if (!ctx.dev) 724 return ret; 725 726 path = dev_fwd_path(stack); 727 if (!path) 728 return -1; 729 path->type = DEV_PATH_ETHERNET; 730 path->dev = ctx.dev; 731 732 return ret; 733 } 734 EXPORT_SYMBOL_GPL(dev_fill_forward_path); 735 736 /** 737 * __dev_get_by_name - find a device by its name 738 * @net: the applicable net namespace 739 * @name: name to find 740 * 741 * Find an interface by name. Must be called under RTNL semaphore 742 * or @dev_base_lock. If the name is found a pointer to the device 743 * is returned. If the name is not found then %NULL is returned. The 744 * reference counters are not incremented so the caller must be 745 * careful with locks. 746 */ 747 748 struct net_device *__dev_get_by_name(struct net *net, const char *name) 749 { 750 struct netdev_name_node *node_name; 751 752 node_name = netdev_name_node_lookup(net, name); 753 return node_name ? node_name->dev : NULL; 754 } 755 EXPORT_SYMBOL(__dev_get_by_name); 756 757 /** 758 * dev_get_by_name_rcu - find a device by its name 759 * @net: the applicable net namespace 760 * @name: name to find 761 * 762 * Find an interface by name. 763 * If the name is found a pointer to the device is returned. 764 * If the name is not found then %NULL is returned. 765 * The reference counters are not incremented so the caller must be 766 * careful with locks. The caller must hold RCU lock. 767 */ 768 769 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) 770 { 771 struct netdev_name_node *node_name; 772 773 node_name = netdev_name_node_lookup_rcu(net, name); 774 return node_name ? node_name->dev : NULL; 775 } 776 EXPORT_SYMBOL(dev_get_by_name_rcu); 777 778 /* Deprecated for new users, call netdev_get_by_name() instead */ 779 struct net_device *dev_get_by_name(struct net *net, const char *name) 780 { 781 struct net_device *dev; 782 783 rcu_read_lock(); 784 dev = dev_get_by_name_rcu(net, name); 785 dev_hold(dev); 786 rcu_read_unlock(); 787 return dev; 788 } 789 EXPORT_SYMBOL(dev_get_by_name); 790 791 /** 792 * netdev_get_by_name() - find a device by its name 793 * @net: the applicable net namespace 794 * @name: name to find 795 * @tracker: tracking object for the acquired reference 796 * @gfp: allocation flags for the tracker 797 * 798 * Find an interface by name. This can be called from any 799 * context and does its own locking. The returned handle has 800 * the usage count incremented and the caller must use netdev_put() to 801 * release it when it is no longer needed. %NULL is returned if no 802 * matching device is found. 803 */ 804 struct net_device *netdev_get_by_name(struct net *net, const char *name, 805 netdevice_tracker *tracker, gfp_t gfp) 806 { 807 struct net_device *dev; 808 809 dev = dev_get_by_name(net, name); 810 if (dev) 811 netdev_tracker_alloc(dev, tracker, gfp); 812 return dev; 813 } 814 EXPORT_SYMBOL(netdev_get_by_name); 815 816 /** 817 * __dev_get_by_index - find a device by its ifindex 818 * @net: the applicable net namespace 819 * @ifindex: index of device 820 * 821 * Search for an interface by index. Returns %NULL if the device 822 * is not found or a pointer to the device. The device has not 823 * had its reference counter increased so the caller must be careful 824 * about locking. The caller must hold either the RTNL semaphore 825 * or @dev_base_lock. 826 */ 827 828 struct net_device *__dev_get_by_index(struct net *net, int ifindex) 829 { 830 struct net_device *dev; 831 struct hlist_head *head = dev_index_hash(net, ifindex); 832 833 hlist_for_each_entry(dev, head, index_hlist) 834 if (dev->ifindex == ifindex) 835 return dev; 836 837 return NULL; 838 } 839 EXPORT_SYMBOL(__dev_get_by_index); 840 841 /** 842 * dev_get_by_index_rcu - find a device by its ifindex 843 * @net: the applicable net namespace 844 * @ifindex: index of device 845 * 846 * Search for an interface by index. Returns %NULL if the device 847 * is not found or a pointer to the device. The device has not 848 * had its reference counter increased so the caller must be careful 849 * about locking. The caller must hold RCU lock. 850 */ 851 852 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) 853 { 854 struct net_device *dev; 855 struct hlist_head *head = dev_index_hash(net, ifindex); 856 857 hlist_for_each_entry_rcu(dev, head, index_hlist) 858 if (dev->ifindex == ifindex) 859 return dev; 860 861 return NULL; 862 } 863 EXPORT_SYMBOL(dev_get_by_index_rcu); 864 865 /* Deprecated for new users, call netdev_get_by_index() instead */ 866 struct net_device *dev_get_by_index(struct net *net, int ifindex) 867 { 868 struct net_device *dev; 869 870 rcu_read_lock(); 871 dev = dev_get_by_index_rcu(net, ifindex); 872 dev_hold(dev); 873 rcu_read_unlock(); 874 return dev; 875 } 876 EXPORT_SYMBOL(dev_get_by_index); 877 878 /** 879 * netdev_get_by_index() - find a device by its ifindex 880 * @net: the applicable net namespace 881 * @ifindex: index of device 882 * @tracker: tracking object for the acquired reference 883 * @gfp: allocation flags for the tracker 884 * 885 * Search for an interface by index. Returns NULL if the device 886 * is not found or a pointer to the device. The device returned has 887 * had a reference added and the pointer is safe until the user calls 888 * netdev_put() to indicate they have finished with it. 889 */ 890 struct net_device *netdev_get_by_index(struct net *net, int ifindex, 891 netdevice_tracker *tracker, gfp_t gfp) 892 { 893 struct net_device *dev; 894 895 dev = dev_get_by_index(net, ifindex); 896 if (dev) 897 netdev_tracker_alloc(dev, tracker, gfp); 898 return dev; 899 } 900 EXPORT_SYMBOL(netdev_get_by_index); 901 902 /** 903 * dev_get_by_napi_id - find a device by napi_id 904 * @napi_id: ID of the NAPI struct 905 * 906 * Search for an interface by NAPI ID. Returns %NULL if the device 907 * is not found or a pointer to the device. The device has not had 908 * its reference counter increased so the caller must be careful 909 * about locking. The caller must hold RCU lock. 910 */ 911 912 struct net_device *dev_get_by_napi_id(unsigned int napi_id) 913 { 914 struct napi_struct *napi; 915 916 WARN_ON_ONCE(!rcu_read_lock_held()); 917 918 if (napi_id < MIN_NAPI_ID) 919 return NULL; 920 921 napi = napi_by_id(napi_id); 922 923 return napi ? napi->dev : NULL; 924 } 925 EXPORT_SYMBOL(dev_get_by_napi_id); 926 927 /** 928 * netdev_get_name - get a netdevice name, knowing its ifindex. 929 * @net: network namespace 930 * @name: a pointer to the buffer where the name will be stored. 931 * @ifindex: the ifindex of the interface to get the name from. 932 */ 933 int netdev_get_name(struct net *net, char *name, int ifindex) 934 { 935 struct net_device *dev; 936 int ret; 937 938 down_read(&devnet_rename_sem); 939 rcu_read_lock(); 940 941 dev = dev_get_by_index_rcu(net, ifindex); 942 if (!dev) { 943 ret = -ENODEV; 944 goto out; 945 } 946 947 strcpy(name, dev->name); 948 949 ret = 0; 950 out: 951 rcu_read_unlock(); 952 up_read(&devnet_rename_sem); 953 return ret; 954 } 955 956 /** 957 * dev_getbyhwaddr_rcu - find a device by its hardware address 958 * @net: the applicable net namespace 959 * @type: media type of device 960 * @ha: hardware address 961 * 962 * Search for an interface by MAC address. Returns NULL if the device 963 * is not found or a pointer to the device. 964 * The caller must hold RCU or RTNL. 965 * The returned device has not had its ref count increased 966 * and the caller must therefore be careful about locking 967 * 968 */ 969 970 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, 971 const char *ha) 972 { 973 struct net_device *dev; 974 975 for_each_netdev_rcu(net, dev) 976 if (dev->type == type && 977 !memcmp(dev->dev_addr, ha, dev->addr_len)) 978 return dev; 979 980 return NULL; 981 } 982 EXPORT_SYMBOL(dev_getbyhwaddr_rcu); 983 984 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) 985 { 986 struct net_device *dev, *ret = NULL; 987 988 rcu_read_lock(); 989 for_each_netdev_rcu(net, dev) 990 if (dev->type == type) { 991 dev_hold(dev); 992 ret = dev; 993 break; 994 } 995 rcu_read_unlock(); 996 return ret; 997 } 998 EXPORT_SYMBOL(dev_getfirstbyhwtype); 999 1000 /** 1001 * __dev_get_by_flags - find any device with given flags 1002 * @net: the applicable net namespace 1003 * @if_flags: IFF_* values 1004 * @mask: bitmask of bits in if_flags to check 1005 * 1006 * Search for any interface with the given flags. Returns NULL if a device 1007 * is not found or a pointer to the device. Must be called inside 1008 * rtnl_lock(), and result refcount is unchanged. 1009 */ 1010 1011 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags, 1012 unsigned short mask) 1013 { 1014 struct net_device *dev, *ret; 1015 1016 ASSERT_RTNL(); 1017 1018 ret = NULL; 1019 for_each_netdev(net, dev) { 1020 if (((dev->flags ^ if_flags) & mask) == 0) { 1021 ret = dev; 1022 break; 1023 } 1024 } 1025 return ret; 1026 } 1027 EXPORT_SYMBOL(__dev_get_by_flags); 1028 1029 /** 1030 * dev_valid_name - check if name is okay for network device 1031 * @name: name string 1032 * 1033 * Network device names need to be valid file names to 1034 * allow sysfs to work. We also disallow any kind of 1035 * whitespace. 1036 */ 1037 bool dev_valid_name(const char *name) 1038 { 1039 if (*name == '\0') 1040 return false; 1041 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ) 1042 return false; 1043 if (!strcmp(name, ".") || !strcmp(name, "..")) 1044 return false; 1045 1046 while (*name) { 1047 if (*name == '/' || *name == ':' || isspace(*name)) 1048 return false; 1049 name++; 1050 } 1051 return true; 1052 } 1053 EXPORT_SYMBOL(dev_valid_name); 1054 1055 /** 1056 * __dev_alloc_name - allocate a name for a device 1057 * @net: network namespace to allocate the device name in 1058 * @name: name format string 1059 * @res: result name string 1060 * 1061 * Passed a format string - eg "lt%d" it will try and find a suitable 1062 * id. It scans list of devices to build up a free map, then chooses 1063 * the first empty slot. The caller must hold the dev_base or rtnl lock 1064 * while allocating the name and adding the device in order to avoid 1065 * duplicates. 1066 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1067 * Returns the number of the unit assigned or a negative errno code. 1068 */ 1069 1070 static int __dev_alloc_name(struct net *net, const char *name, char *res) 1071 { 1072 int i = 0; 1073 const char *p; 1074 const int max_netdevices = 8*PAGE_SIZE; 1075 unsigned long *inuse; 1076 struct net_device *d; 1077 char buf[IFNAMSIZ]; 1078 1079 /* Verify the string as this thing may have come from the user. 1080 * There must be one "%d" and no other "%" characters. 1081 */ 1082 p = strchr(name, '%'); 1083 if (!p || p[1] != 'd' || strchr(p + 2, '%')) 1084 return -EINVAL; 1085 1086 /* Use one page as a bit array of possible slots */ 1087 inuse = bitmap_zalloc(max_netdevices, GFP_ATOMIC); 1088 if (!inuse) 1089 return -ENOMEM; 1090 1091 for_each_netdev(net, d) { 1092 struct netdev_name_node *name_node; 1093 1094 netdev_for_each_altname(d, name_node) { 1095 if (!sscanf(name_node->name, name, &i)) 1096 continue; 1097 if (i < 0 || i >= max_netdevices) 1098 continue; 1099 1100 /* avoid cases where sscanf is not exact inverse of printf */ 1101 snprintf(buf, IFNAMSIZ, name, i); 1102 if (!strncmp(buf, name_node->name, IFNAMSIZ)) 1103 __set_bit(i, inuse); 1104 } 1105 if (!sscanf(d->name, name, &i)) 1106 continue; 1107 if (i < 0 || i >= max_netdevices) 1108 continue; 1109 1110 /* avoid cases where sscanf is not exact inverse of printf */ 1111 snprintf(buf, IFNAMSIZ, name, i); 1112 if (!strncmp(buf, d->name, IFNAMSIZ)) 1113 __set_bit(i, inuse); 1114 } 1115 1116 i = find_first_zero_bit(inuse, max_netdevices); 1117 bitmap_free(inuse); 1118 if (i == max_netdevices) 1119 return -ENFILE; 1120 1121 /* 'res' and 'name' could overlap, use 'buf' as an intermediate buffer */ 1122 strscpy(buf, name, IFNAMSIZ); 1123 snprintf(res, IFNAMSIZ, buf, i); 1124 return i; 1125 } 1126 1127 /* Returns negative errno or allocated unit id (see __dev_alloc_name()) */ 1128 static int dev_prep_valid_name(struct net *net, struct net_device *dev, 1129 const char *want_name, char *out_name, 1130 int dup_errno) 1131 { 1132 if (!dev_valid_name(want_name)) 1133 return -EINVAL; 1134 1135 if (strchr(want_name, '%')) 1136 return __dev_alloc_name(net, want_name, out_name); 1137 1138 if (netdev_name_in_use(net, want_name)) 1139 return -dup_errno; 1140 if (out_name != want_name) 1141 strscpy(out_name, want_name, IFNAMSIZ); 1142 return 0; 1143 } 1144 1145 /** 1146 * dev_alloc_name - allocate a name for a device 1147 * @dev: device 1148 * @name: name format string 1149 * 1150 * Passed a format string - eg "lt%d" it will try and find a suitable 1151 * id. It scans list of devices to build up a free map, then chooses 1152 * the first empty slot. The caller must hold the dev_base or rtnl lock 1153 * while allocating the name and adding the device in order to avoid 1154 * duplicates. 1155 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1156 * Returns the number of the unit assigned or a negative errno code. 1157 */ 1158 1159 int dev_alloc_name(struct net_device *dev, const char *name) 1160 { 1161 return dev_prep_valid_name(dev_net(dev), dev, name, dev->name, ENFILE); 1162 } 1163 EXPORT_SYMBOL(dev_alloc_name); 1164 1165 static int dev_get_valid_name(struct net *net, struct net_device *dev, 1166 const char *name) 1167 { 1168 int ret; 1169 1170 ret = dev_prep_valid_name(net, dev, name, dev->name, EEXIST); 1171 return ret < 0 ? ret : 0; 1172 } 1173 1174 /** 1175 * dev_change_name - change name of a device 1176 * @dev: device 1177 * @newname: name (or format string) must be at least IFNAMSIZ 1178 * 1179 * Change name of a device, can pass format strings "eth%d". 1180 * for wildcarding. 1181 */ 1182 int dev_change_name(struct net_device *dev, const char *newname) 1183 { 1184 unsigned char old_assign_type; 1185 char oldname[IFNAMSIZ]; 1186 int err = 0; 1187 int ret; 1188 struct net *net; 1189 1190 ASSERT_RTNL(); 1191 BUG_ON(!dev_net(dev)); 1192 1193 net = dev_net(dev); 1194 1195 down_write(&devnet_rename_sem); 1196 1197 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { 1198 up_write(&devnet_rename_sem); 1199 return 0; 1200 } 1201 1202 memcpy(oldname, dev->name, IFNAMSIZ); 1203 1204 err = dev_get_valid_name(net, dev, newname); 1205 if (err < 0) { 1206 up_write(&devnet_rename_sem); 1207 return err; 1208 } 1209 1210 if (oldname[0] && !strchr(oldname, '%')) 1211 netdev_info(dev, "renamed from %s%s\n", oldname, 1212 dev->flags & IFF_UP ? " (while UP)" : ""); 1213 1214 old_assign_type = dev->name_assign_type; 1215 dev->name_assign_type = NET_NAME_RENAMED; 1216 1217 rollback: 1218 ret = device_rename(&dev->dev, dev->name); 1219 if (ret) { 1220 memcpy(dev->name, oldname, IFNAMSIZ); 1221 dev->name_assign_type = old_assign_type; 1222 up_write(&devnet_rename_sem); 1223 return ret; 1224 } 1225 1226 up_write(&devnet_rename_sem); 1227 1228 netdev_adjacent_rename_links(dev, oldname); 1229 1230 write_lock(&dev_base_lock); 1231 netdev_name_node_del(dev->name_node); 1232 write_unlock(&dev_base_lock); 1233 1234 synchronize_rcu(); 1235 1236 write_lock(&dev_base_lock); 1237 netdev_name_node_add(net, dev->name_node); 1238 write_unlock(&dev_base_lock); 1239 1240 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); 1241 ret = notifier_to_errno(ret); 1242 1243 if (ret) { 1244 /* err >= 0 after dev_alloc_name() or stores the first errno */ 1245 if (err >= 0) { 1246 err = ret; 1247 down_write(&devnet_rename_sem); 1248 memcpy(dev->name, oldname, IFNAMSIZ); 1249 memcpy(oldname, newname, IFNAMSIZ); 1250 dev->name_assign_type = old_assign_type; 1251 old_assign_type = NET_NAME_RENAMED; 1252 goto rollback; 1253 } else { 1254 netdev_err(dev, "name change rollback failed: %d\n", 1255 ret); 1256 } 1257 } 1258 1259 return err; 1260 } 1261 1262 /** 1263 * dev_set_alias - change ifalias of a device 1264 * @dev: device 1265 * @alias: name up to IFALIASZ 1266 * @len: limit of bytes to copy from info 1267 * 1268 * Set ifalias for a device, 1269 */ 1270 int dev_set_alias(struct net_device *dev, const char *alias, size_t len) 1271 { 1272 struct dev_ifalias *new_alias = NULL; 1273 1274 if (len >= IFALIASZ) 1275 return -EINVAL; 1276 1277 if (len) { 1278 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL); 1279 if (!new_alias) 1280 return -ENOMEM; 1281 1282 memcpy(new_alias->ifalias, alias, len); 1283 new_alias->ifalias[len] = 0; 1284 } 1285 1286 mutex_lock(&ifalias_mutex); 1287 new_alias = rcu_replace_pointer(dev->ifalias, new_alias, 1288 mutex_is_locked(&ifalias_mutex)); 1289 mutex_unlock(&ifalias_mutex); 1290 1291 if (new_alias) 1292 kfree_rcu(new_alias, rcuhead); 1293 1294 return len; 1295 } 1296 EXPORT_SYMBOL(dev_set_alias); 1297 1298 /** 1299 * dev_get_alias - get ifalias of a device 1300 * @dev: device 1301 * @name: buffer to store name of ifalias 1302 * @len: size of buffer 1303 * 1304 * get ifalias for a device. Caller must make sure dev cannot go 1305 * away, e.g. rcu read lock or own a reference count to device. 1306 */ 1307 int dev_get_alias(const struct net_device *dev, char *name, size_t len) 1308 { 1309 const struct dev_ifalias *alias; 1310 int ret = 0; 1311 1312 rcu_read_lock(); 1313 alias = rcu_dereference(dev->ifalias); 1314 if (alias) 1315 ret = snprintf(name, len, "%s", alias->ifalias); 1316 rcu_read_unlock(); 1317 1318 return ret; 1319 } 1320 1321 /** 1322 * netdev_features_change - device changes features 1323 * @dev: device to cause notification 1324 * 1325 * Called to indicate a device has changed features. 1326 */ 1327 void netdev_features_change(struct net_device *dev) 1328 { 1329 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); 1330 } 1331 EXPORT_SYMBOL(netdev_features_change); 1332 1333 /** 1334 * netdev_state_change - device changes state 1335 * @dev: device to cause notification 1336 * 1337 * Called to indicate a device has changed state. This function calls 1338 * the notifier chains for netdev_chain and sends a NEWLINK message 1339 * to the routing socket. 1340 */ 1341 void netdev_state_change(struct net_device *dev) 1342 { 1343 if (dev->flags & IFF_UP) { 1344 struct netdev_notifier_change_info change_info = { 1345 .info.dev = dev, 1346 }; 1347 1348 call_netdevice_notifiers_info(NETDEV_CHANGE, 1349 &change_info.info); 1350 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL, 0, NULL); 1351 } 1352 } 1353 EXPORT_SYMBOL(netdev_state_change); 1354 1355 /** 1356 * __netdev_notify_peers - notify network peers about existence of @dev, 1357 * to be called when rtnl lock is already held. 1358 * @dev: network device 1359 * 1360 * Generate traffic such that interested network peers are aware of 1361 * @dev, such as by generating a gratuitous ARP. This may be used when 1362 * a device wants to inform the rest of the network about some sort of 1363 * reconfiguration such as a failover event or virtual machine 1364 * migration. 1365 */ 1366 void __netdev_notify_peers(struct net_device *dev) 1367 { 1368 ASSERT_RTNL(); 1369 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); 1370 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev); 1371 } 1372 EXPORT_SYMBOL(__netdev_notify_peers); 1373 1374 /** 1375 * netdev_notify_peers - notify network peers about existence of @dev 1376 * @dev: network device 1377 * 1378 * Generate traffic such that interested network peers are aware of 1379 * @dev, such as by generating a gratuitous ARP. This may be used when 1380 * a device wants to inform the rest of the network about some sort of 1381 * reconfiguration such as a failover event or virtual machine 1382 * migration. 1383 */ 1384 void netdev_notify_peers(struct net_device *dev) 1385 { 1386 rtnl_lock(); 1387 __netdev_notify_peers(dev); 1388 rtnl_unlock(); 1389 } 1390 EXPORT_SYMBOL(netdev_notify_peers); 1391 1392 static int napi_threaded_poll(void *data); 1393 1394 static int napi_kthread_create(struct napi_struct *n) 1395 { 1396 int err = 0; 1397 1398 /* Create and wake up the kthread once to put it in 1399 * TASK_INTERRUPTIBLE mode to avoid the blocked task 1400 * warning and work with loadavg. 1401 */ 1402 n->thread = kthread_run(napi_threaded_poll, n, "napi/%s-%d", 1403 n->dev->name, n->napi_id); 1404 if (IS_ERR(n->thread)) { 1405 err = PTR_ERR(n->thread); 1406 pr_err("kthread_run failed with err %d\n", err); 1407 n->thread = NULL; 1408 } 1409 1410 return err; 1411 } 1412 1413 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1414 { 1415 const struct net_device_ops *ops = dev->netdev_ops; 1416 int ret; 1417 1418 ASSERT_RTNL(); 1419 dev_addr_check(dev); 1420 1421 if (!netif_device_present(dev)) { 1422 /* may be detached because parent is runtime-suspended */ 1423 if (dev->dev.parent) 1424 pm_runtime_resume(dev->dev.parent); 1425 if (!netif_device_present(dev)) 1426 return -ENODEV; 1427 } 1428 1429 /* Block netpoll from trying to do any rx path servicing. 1430 * If we don't do this there is a chance ndo_poll_controller 1431 * or ndo_poll may be running while we open the device 1432 */ 1433 netpoll_poll_disable(dev); 1434 1435 ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack); 1436 ret = notifier_to_errno(ret); 1437 if (ret) 1438 return ret; 1439 1440 set_bit(__LINK_STATE_START, &dev->state); 1441 1442 if (ops->ndo_validate_addr) 1443 ret = ops->ndo_validate_addr(dev); 1444 1445 if (!ret && ops->ndo_open) 1446 ret = ops->ndo_open(dev); 1447 1448 netpoll_poll_enable(dev); 1449 1450 if (ret) 1451 clear_bit(__LINK_STATE_START, &dev->state); 1452 else { 1453 dev->flags |= IFF_UP; 1454 dev_set_rx_mode(dev); 1455 dev_activate(dev); 1456 add_device_randomness(dev->dev_addr, dev->addr_len); 1457 } 1458 1459 return ret; 1460 } 1461 1462 /** 1463 * dev_open - prepare an interface for use. 1464 * @dev: device to open 1465 * @extack: netlink extended ack 1466 * 1467 * Takes a device from down to up state. The device's private open 1468 * function is invoked and then the multicast lists are loaded. Finally 1469 * the device is moved into the up state and a %NETDEV_UP message is 1470 * sent to the netdev notifier chain. 1471 * 1472 * Calling this function on an active interface is a nop. On a failure 1473 * a negative errno code is returned. 1474 */ 1475 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1476 { 1477 int ret; 1478 1479 if (dev->flags & IFF_UP) 1480 return 0; 1481 1482 ret = __dev_open(dev, extack); 1483 if (ret < 0) 1484 return ret; 1485 1486 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL); 1487 call_netdevice_notifiers(NETDEV_UP, dev); 1488 1489 return ret; 1490 } 1491 EXPORT_SYMBOL(dev_open); 1492 1493 static void __dev_close_many(struct list_head *head) 1494 { 1495 struct net_device *dev; 1496 1497 ASSERT_RTNL(); 1498 might_sleep(); 1499 1500 list_for_each_entry(dev, head, close_list) { 1501 /* Temporarily disable netpoll until the interface is down */ 1502 netpoll_poll_disable(dev); 1503 1504 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); 1505 1506 clear_bit(__LINK_STATE_START, &dev->state); 1507 1508 /* Synchronize to scheduled poll. We cannot touch poll list, it 1509 * can be even on different cpu. So just clear netif_running(). 1510 * 1511 * dev->stop() will invoke napi_disable() on all of it's 1512 * napi_struct instances on this device. 1513 */ 1514 smp_mb__after_atomic(); /* Commit netif_running(). */ 1515 } 1516 1517 dev_deactivate_many(head); 1518 1519 list_for_each_entry(dev, head, close_list) { 1520 const struct net_device_ops *ops = dev->netdev_ops; 1521 1522 /* 1523 * Call the device specific close. This cannot fail. 1524 * Only if device is UP 1525 * 1526 * We allow it to be called even after a DETACH hot-plug 1527 * event. 1528 */ 1529 if (ops->ndo_stop) 1530 ops->ndo_stop(dev); 1531 1532 dev->flags &= ~IFF_UP; 1533 netpoll_poll_enable(dev); 1534 } 1535 } 1536 1537 static void __dev_close(struct net_device *dev) 1538 { 1539 LIST_HEAD(single); 1540 1541 list_add(&dev->close_list, &single); 1542 __dev_close_many(&single); 1543 list_del(&single); 1544 } 1545 1546 void dev_close_many(struct list_head *head, bool unlink) 1547 { 1548 struct net_device *dev, *tmp; 1549 1550 /* Remove the devices that don't need to be closed */ 1551 list_for_each_entry_safe(dev, tmp, head, close_list) 1552 if (!(dev->flags & IFF_UP)) 1553 list_del_init(&dev->close_list); 1554 1555 __dev_close_many(head); 1556 1557 list_for_each_entry_safe(dev, tmp, head, close_list) { 1558 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL); 1559 call_netdevice_notifiers(NETDEV_DOWN, dev); 1560 if (unlink) 1561 list_del_init(&dev->close_list); 1562 } 1563 } 1564 EXPORT_SYMBOL(dev_close_many); 1565 1566 /** 1567 * dev_close - shutdown an interface. 1568 * @dev: device to shutdown 1569 * 1570 * This function moves an active device into down state. A 1571 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device 1572 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier 1573 * chain. 1574 */ 1575 void dev_close(struct net_device *dev) 1576 { 1577 if (dev->flags & IFF_UP) { 1578 LIST_HEAD(single); 1579 1580 list_add(&dev->close_list, &single); 1581 dev_close_many(&single, true); 1582 list_del(&single); 1583 } 1584 } 1585 EXPORT_SYMBOL(dev_close); 1586 1587 1588 /** 1589 * dev_disable_lro - disable Large Receive Offload on a device 1590 * @dev: device 1591 * 1592 * Disable Large Receive Offload (LRO) on a net device. Must be 1593 * called under RTNL. This is needed if received packets may be 1594 * forwarded to another interface. 1595 */ 1596 void dev_disable_lro(struct net_device *dev) 1597 { 1598 struct net_device *lower_dev; 1599 struct list_head *iter; 1600 1601 dev->wanted_features &= ~NETIF_F_LRO; 1602 netdev_update_features(dev); 1603 1604 if (unlikely(dev->features & NETIF_F_LRO)) 1605 netdev_WARN(dev, "failed to disable LRO!\n"); 1606 1607 netdev_for_each_lower_dev(dev, lower_dev, iter) 1608 dev_disable_lro(lower_dev); 1609 } 1610 EXPORT_SYMBOL(dev_disable_lro); 1611 1612 /** 1613 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device 1614 * @dev: device 1615 * 1616 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be 1617 * called under RTNL. This is needed if Generic XDP is installed on 1618 * the device. 1619 */ 1620 static void dev_disable_gro_hw(struct net_device *dev) 1621 { 1622 dev->wanted_features &= ~NETIF_F_GRO_HW; 1623 netdev_update_features(dev); 1624 1625 if (unlikely(dev->features & NETIF_F_GRO_HW)) 1626 netdev_WARN(dev, "failed to disable GRO_HW!\n"); 1627 } 1628 1629 const char *netdev_cmd_to_name(enum netdev_cmd cmd) 1630 { 1631 #define N(val) \ 1632 case NETDEV_##val: \ 1633 return "NETDEV_" __stringify(val); 1634 switch (cmd) { 1635 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER) 1636 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE) 1637 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE) 1638 N(POST_INIT) N(PRE_UNINIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) 1639 N(CHANGEUPPER) N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) 1640 N(BONDING_INFO) N(PRECHANGEUPPER) N(CHANGELOWERSTATE) 1641 N(UDP_TUNNEL_PUSH_INFO) N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN) 1642 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO) 1643 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO) 1644 N(PRE_CHANGEADDR) N(OFFLOAD_XSTATS_ENABLE) N(OFFLOAD_XSTATS_DISABLE) 1645 N(OFFLOAD_XSTATS_REPORT_USED) N(OFFLOAD_XSTATS_REPORT_DELTA) 1646 N(XDP_FEAT_CHANGE) 1647 } 1648 #undef N 1649 return "UNKNOWN_NETDEV_EVENT"; 1650 } 1651 EXPORT_SYMBOL_GPL(netdev_cmd_to_name); 1652 1653 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, 1654 struct net_device *dev) 1655 { 1656 struct netdev_notifier_info info = { 1657 .dev = dev, 1658 }; 1659 1660 return nb->notifier_call(nb, val, &info); 1661 } 1662 1663 static int call_netdevice_register_notifiers(struct notifier_block *nb, 1664 struct net_device *dev) 1665 { 1666 int err; 1667 1668 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); 1669 err = notifier_to_errno(err); 1670 if (err) 1671 return err; 1672 1673 if (!(dev->flags & IFF_UP)) 1674 return 0; 1675 1676 call_netdevice_notifier(nb, NETDEV_UP, dev); 1677 return 0; 1678 } 1679 1680 static void call_netdevice_unregister_notifiers(struct notifier_block *nb, 1681 struct net_device *dev) 1682 { 1683 if (dev->flags & IFF_UP) { 1684 call_netdevice_notifier(nb, NETDEV_GOING_DOWN, 1685 dev); 1686 call_netdevice_notifier(nb, NETDEV_DOWN, dev); 1687 } 1688 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); 1689 } 1690 1691 static int call_netdevice_register_net_notifiers(struct notifier_block *nb, 1692 struct net *net) 1693 { 1694 struct net_device *dev; 1695 int err; 1696 1697 for_each_netdev(net, dev) { 1698 err = call_netdevice_register_notifiers(nb, dev); 1699 if (err) 1700 goto rollback; 1701 } 1702 return 0; 1703 1704 rollback: 1705 for_each_netdev_continue_reverse(net, dev) 1706 call_netdevice_unregister_notifiers(nb, dev); 1707 return err; 1708 } 1709 1710 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb, 1711 struct net *net) 1712 { 1713 struct net_device *dev; 1714 1715 for_each_netdev(net, dev) 1716 call_netdevice_unregister_notifiers(nb, dev); 1717 } 1718 1719 static int dev_boot_phase = 1; 1720 1721 /** 1722 * register_netdevice_notifier - register a network notifier block 1723 * @nb: notifier 1724 * 1725 * Register a notifier to be called when network device events occur. 1726 * The notifier passed is linked into the kernel structures and must 1727 * not be reused until it has been unregistered. A negative errno code 1728 * is returned on a failure. 1729 * 1730 * When registered all registration and up events are replayed 1731 * to the new notifier to allow device to have a race free 1732 * view of the network device list. 1733 */ 1734 1735 int register_netdevice_notifier(struct notifier_block *nb) 1736 { 1737 struct net *net; 1738 int err; 1739 1740 /* Close race with setup_net() and cleanup_net() */ 1741 down_write(&pernet_ops_rwsem); 1742 rtnl_lock(); 1743 err = raw_notifier_chain_register(&netdev_chain, nb); 1744 if (err) 1745 goto unlock; 1746 if (dev_boot_phase) 1747 goto unlock; 1748 for_each_net(net) { 1749 err = call_netdevice_register_net_notifiers(nb, net); 1750 if (err) 1751 goto rollback; 1752 } 1753 1754 unlock: 1755 rtnl_unlock(); 1756 up_write(&pernet_ops_rwsem); 1757 return err; 1758 1759 rollback: 1760 for_each_net_continue_reverse(net) 1761 call_netdevice_unregister_net_notifiers(nb, net); 1762 1763 raw_notifier_chain_unregister(&netdev_chain, nb); 1764 goto unlock; 1765 } 1766 EXPORT_SYMBOL(register_netdevice_notifier); 1767 1768 /** 1769 * unregister_netdevice_notifier - unregister a network notifier block 1770 * @nb: notifier 1771 * 1772 * Unregister a notifier previously registered by 1773 * register_netdevice_notifier(). The notifier is unlinked into the 1774 * kernel structures and may then be reused. A negative errno code 1775 * is returned on a failure. 1776 * 1777 * After unregistering unregister and down device events are synthesized 1778 * for all devices on the device list to the removed notifier to remove 1779 * the need for special case cleanup code. 1780 */ 1781 1782 int unregister_netdevice_notifier(struct notifier_block *nb) 1783 { 1784 struct net *net; 1785 int err; 1786 1787 /* Close race with setup_net() and cleanup_net() */ 1788 down_write(&pernet_ops_rwsem); 1789 rtnl_lock(); 1790 err = raw_notifier_chain_unregister(&netdev_chain, nb); 1791 if (err) 1792 goto unlock; 1793 1794 for_each_net(net) 1795 call_netdevice_unregister_net_notifiers(nb, net); 1796 1797 unlock: 1798 rtnl_unlock(); 1799 up_write(&pernet_ops_rwsem); 1800 return err; 1801 } 1802 EXPORT_SYMBOL(unregister_netdevice_notifier); 1803 1804 static int __register_netdevice_notifier_net(struct net *net, 1805 struct notifier_block *nb, 1806 bool ignore_call_fail) 1807 { 1808 int err; 1809 1810 err = raw_notifier_chain_register(&net->netdev_chain, nb); 1811 if (err) 1812 return err; 1813 if (dev_boot_phase) 1814 return 0; 1815 1816 err = call_netdevice_register_net_notifiers(nb, net); 1817 if (err && !ignore_call_fail) 1818 goto chain_unregister; 1819 1820 return 0; 1821 1822 chain_unregister: 1823 raw_notifier_chain_unregister(&net->netdev_chain, nb); 1824 return err; 1825 } 1826 1827 static int __unregister_netdevice_notifier_net(struct net *net, 1828 struct notifier_block *nb) 1829 { 1830 int err; 1831 1832 err = raw_notifier_chain_unregister(&net->netdev_chain, nb); 1833 if (err) 1834 return err; 1835 1836 call_netdevice_unregister_net_notifiers(nb, net); 1837 return 0; 1838 } 1839 1840 /** 1841 * register_netdevice_notifier_net - register a per-netns network notifier block 1842 * @net: network namespace 1843 * @nb: notifier 1844 * 1845 * Register a notifier to be called when network device events occur. 1846 * The notifier passed is linked into the kernel structures and must 1847 * not be reused until it has been unregistered. A negative errno code 1848 * is returned on a failure. 1849 * 1850 * When registered all registration and up events are replayed 1851 * to the new notifier to allow device to have a race free 1852 * view of the network device list. 1853 */ 1854 1855 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb) 1856 { 1857 int err; 1858 1859 rtnl_lock(); 1860 err = __register_netdevice_notifier_net(net, nb, false); 1861 rtnl_unlock(); 1862 return err; 1863 } 1864 EXPORT_SYMBOL(register_netdevice_notifier_net); 1865 1866 /** 1867 * unregister_netdevice_notifier_net - unregister a per-netns 1868 * network notifier block 1869 * @net: network namespace 1870 * @nb: notifier 1871 * 1872 * Unregister a notifier previously registered by 1873 * register_netdevice_notifier_net(). The notifier is unlinked from the 1874 * kernel structures and may then be reused. A negative errno code 1875 * is returned on a failure. 1876 * 1877 * After unregistering unregister and down device events are synthesized 1878 * for all devices on the device list to the removed notifier to remove 1879 * the need for special case cleanup code. 1880 */ 1881 1882 int unregister_netdevice_notifier_net(struct net *net, 1883 struct notifier_block *nb) 1884 { 1885 int err; 1886 1887 rtnl_lock(); 1888 err = __unregister_netdevice_notifier_net(net, nb); 1889 rtnl_unlock(); 1890 return err; 1891 } 1892 EXPORT_SYMBOL(unregister_netdevice_notifier_net); 1893 1894 static void __move_netdevice_notifier_net(struct net *src_net, 1895 struct net *dst_net, 1896 struct notifier_block *nb) 1897 { 1898 __unregister_netdevice_notifier_net(src_net, nb); 1899 __register_netdevice_notifier_net(dst_net, nb, true); 1900 } 1901 1902 int register_netdevice_notifier_dev_net(struct net_device *dev, 1903 struct notifier_block *nb, 1904 struct netdev_net_notifier *nn) 1905 { 1906 int err; 1907 1908 rtnl_lock(); 1909 err = __register_netdevice_notifier_net(dev_net(dev), nb, false); 1910 if (!err) { 1911 nn->nb = nb; 1912 list_add(&nn->list, &dev->net_notifier_list); 1913 } 1914 rtnl_unlock(); 1915 return err; 1916 } 1917 EXPORT_SYMBOL(register_netdevice_notifier_dev_net); 1918 1919 int unregister_netdevice_notifier_dev_net(struct net_device *dev, 1920 struct notifier_block *nb, 1921 struct netdev_net_notifier *nn) 1922 { 1923 int err; 1924 1925 rtnl_lock(); 1926 list_del(&nn->list); 1927 err = __unregister_netdevice_notifier_net(dev_net(dev), nb); 1928 rtnl_unlock(); 1929 return err; 1930 } 1931 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net); 1932 1933 static void move_netdevice_notifiers_dev_net(struct net_device *dev, 1934 struct net *net) 1935 { 1936 struct netdev_net_notifier *nn; 1937 1938 list_for_each_entry(nn, &dev->net_notifier_list, list) 1939 __move_netdevice_notifier_net(dev_net(dev), net, nn->nb); 1940 } 1941 1942 /** 1943 * call_netdevice_notifiers_info - call all network notifier blocks 1944 * @val: value passed unmodified to notifier function 1945 * @info: notifier information data 1946 * 1947 * Call all network notifier blocks. Parameters and return value 1948 * are as for raw_notifier_call_chain(). 1949 */ 1950 1951 int call_netdevice_notifiers_info(unsigned long val, 1952 struct netdev_notifier_info *info) 1953 { 1954 struct net *net = dev_net(info->dev); 1955 int ret; 1956 1957 ASSERT_RTNL(); 1958 1959 /* Run per-netns notifier block chain first, then run the global one. 1960 * Hopefully, one day, the global one is going to be removed after 1961 * all notifier block registrators get converted to be per-netns. 1962 */ 1963 ret = raw_notifier_call_chain(&net->netdev_chain, val, info); 1964 if (ret & NOTIFY_STOP_MASK) 1965 return ret; 1966 return raw_notifier_call_chain(&netdev_chain, val, info); 1967 } 1968 1969 /** 1970 * call_netdevice_notifiers_info_robust - call per-netns notifier blocks 1971 * for and rollback on error 1972 * @val_up: value passed unmodified to notifier function 1973 * @val_down: value passed unmodified to the notifier function when 1974 * recovering from an error on @val_up 1975 * @info: notifier information data 1976 * 1977 * Call all per-netns network notifier blocks, but not notifier blocks on 1978 * the global notifier chain. Parameters and return value are as for 1979 * raw_notifier_call_chain_robust(). 1980 */ 1981 1982 static int 1983 call_netdevice_notifiers_info_robust(unsigned long val_up, 1984 unsigned long val_down, 1985 struct netdev_notifier_info *info) 1986 { 1987 struct net *net = dev_net(info->dev); 1988 1989 ASSERT_RTNL(); 1990 1991 return raw_notifier_call_chain_robust(&net->netdev_chain, 1992 val_up, val_down, info); 1993 } 1994 1995 static int call_netdevice_notifiers_extack(unsigned long val, 1996 struct net_device *dev, 1997 struct netlink_ext_ack *extack) 1998 { 1999 struct netdev_notifier_info info = { 2000 .dev = dev, 2001 .extack = extack, 2002 }; 2003 2004 return call_netdevice_notifiers_info(val, &info); 2005 } 2006 2007 /** 2008 * call_netdevice_notifiers - call all network notifier blocks 2009 * @val: value passed unmodified to notifier function 2010 * @dev: net_device pointer passed unmodified to notifier function 2011 * 2012 * Call all network notifier blocks. Parameters and return value 2013 * are as for raw_notifier_call_chain(). 2014 */ 2015 2016 int call_netdevice_notifiers(unsigned long val, struct net_device *dev) 2017 { 2018 return call_netdevice_notifiers_extack(val, dev, NULL); 2019 } 2020 EXPORT_SYMBOL(call_netdevice_notifiers); 2021 2022 /** 2023 * call_netdevice_notifiers_mtu - call all network notifier blocks 2024 * @val: value passed unmodified to notifier function 2025 * @dev: net_device pointer passed unmodified to notifier function 2026 * @arg: additional u32 argument passed to the notifier function 2027 * 2028 * Call all network notifier blocks. Parameters and return value 2029 * are as for raw_notifier_call_chain(). 2030 */ 2031 static int call_netdevice_notifiers_mtu(unsigned long val, 2032 struct net_device *dev, u32 arg) 2033 { 2034 struct netdev_notifier_info_ext info = { 2035 .info.dev = dev, 2036 .ext.mtu = arg, 2037 }; 2038 2039 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0); 2040 2041 return call_netdevice_notifiers_info(val, &info.info); 2042 } 2043 2044 #ifdef CONFIG_NET_INGRESS 2045 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key); 2046 2047 void net_inc_ingress_queue(void) 2048 { 2049 static_branch_inc(&ingress_needed_key); 2050 } 2051 EXPORT_SYMBOL_GPL(net_inc_ingress_queue); 2052 2053 void net_dec_ingress_queue(void) 2054 { 2055 static_branch_dec(&ingress_needed_key); 2056 } 2057 EXPORT_SYMBOL_GPL(net_dec_ingress_queue); 2058 #endif 2059 2060 #ifdef CONFIG_NET_EGRESS 2061 static DEFINE_STATIC_KEY_FALSE(egress_needed_key); 2062 2063 void net_inc_egress_queue(void) 2064 { 2065 static_branch_inc(&egress_needed_key); 2066 } 2067 EXPORT_SYMBOL_GPL(net_inc_egress_queue); 2068 2069 void net_dec_egress_queue(void) 2070 { 2071 static_branch_dec(&egress_needed_key); 2072 } 2073 EXPORT_SYMBOL_GPL(net_dec_egress_queue); 2074 #endif 2075 2076 DEFINE_STATIC_KEY_FALSE(netstamp_needed_key); 2077 EXPORT_SYMBOL(netstamp_needed_key); 2078 #ifdef CONFIG_JUMP_LABEL 2079 static atomic_t netstamp_needed_deferred; 2080 static atomic_t netstamp_wanted; 2081 static void netstamp_clear(struct work_struct *work) 2082 { 2083 int deferred = atomic_xchg(&netstamp_needed_deferred, 0); 2084 int wanted; 2085 2086 wanted = atomic_add_return(deferred, &netstamp_wanted); 2087 if (wanted > 0) 2088 static_branch_enable(&netstamp_needed_key); 2089 else 2090 static_branch_disable(&netstamp_needed_key); 2091 } 2092 static DECLARE_WORK(netstamp_work, netstamp_clear); 2093 #endif 2094 2095 void net_enable_timestamp(void) 2096 { 2097 #ifdef CONFIG_JUMP_LABEL 2098 int wanted = atomic_read(&netstamp_wanted); 2099 2100 while (wanted > 0) { 2101 if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted + 1)) 2102 return; 2103 } 2104 atomic_inc(&netstamp_needed_deferred); 2105 schedule_work(&netstamp_work); 2106 #else 2107 static_branch_inc(&netstamp_needed_key); 2108 #endif 2109 } 2110 EXPORT_SYMBOL(net_enable_timestamp); 2111 2112 void net_disable_timestamp(void) 2113 { 2114 #ifdef CONFIG_JUMP_LABEL 2115 int wanted = atomic_read(&netstamp_wanted); 2116 2117 while (wanted > 1) { 2118 if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted - 1)) 2119 return; 2120 } 2121 atomic_dec(&netstamp_needed_deferred); 2122 schedule_work(&netstamp_work); 2123 #else 2124 static_branch_dec(&netstamp_needed_key); 2125 #endif 2126 } 2127 EXPORT_SYMBOL(net_disable_timestamp); 2128 2129 static inline void net_timestamp_set(struct sk_buff *skb) 2130 { 2131 skb->tstamp = 0; 2132 skb->mono_delivery_time = 0; 2133 if (static_branch_unlikely(&netstamp_needed_key)) 2134 skb->tstamp = ktime_get_real(); 2135 } 2136 2137 #define net_timestamp_check(COND, SKB) \ 2138 if (static_branch_unlikely(&netstamp_needed_key)) { \ 2139 if ((COND) && !(SKB)->tstamp) \ 2140 (SKB)->tstamp = ktime_get_real(); \ 2141 } \ 2142 2143 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb) 2144 { 2145 return __is_skb_forwardable(dev, skb, true); 2146 } 2147 EXPORT_SYMBOL_GPL(is_skb_forwardable); 2148 2149 static int __dev_forward_skb2(struct net_device *dev, struct sk_buff *skb, 2150 bool check_mtu) 2151 { 2152 int ret = ____dev_forward_skb(dev, skb, check_mtu); 2153 2154 if (likely(!ret)) { 2155 skb->protocol = eth_type_trans(skb, dev); 2156 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); 2157 } 2158 2159 return ret; 2160 } 2161 2162 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2163 { 2164 return __dev_forward_skb2(dev, skb, true); 2165 } 2166 EXPORT_SYMBOL_GPL(__dev_forward_skb); 2167 2168 /** 2169 * dev_forward_skb - loopback an skb to another netif 2170 * 2171 * @dev: destination network device 2172 * @skb: buffer to forward 2173 * 2174 * return values: 2175 * NET_RX_SUCCESS (no congestion) 2176 * NET_RX_DROP (packet was dropped, but freed) 2177 * 2178 * dev_forward_skb can be used for injecting an skb from the 2179 * start_xmit function of one device into the receive queue 2180 * of another device. 2181 * 2182 * The receiving device may be in another namespace, so 2183 * we have to clear all information in the skb that could 2184 * impact namespace isolation. 2185 */ 2186 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2187 { 2188 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); 2189 } 2190 EXPORT_SYMBOL_GPL(dev_forward_skb); 2191 2192 int dev_forward_skb_nomtu(struct net_device *dev, struct sk_buff *skb) 2193 { 2194 return __dev_forward_skb2(dev, skb, false) ?: netif_rx_internal(skb); 2195 } 2196 2197 static inline int deliver_skb(struct sk_buff *skb, 2198 struct packet_type *pt_prev, 2199 struct net_device *orig_dev) 2200 { 2201 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 2202 return -ENOMEM; 2203 refcount_inc(&skb->users); 2204 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 2205 } 2206 2207 static inline void deliver_ptype_list_skb(struct sk_buff *skb, 2208 struct packet_type **pt, 2209 struct net_device *orig_dev, 2210 __be16 type, 2211 struct list_head *ptype_list) 2212 { 2213 struct packet_type *ptype, *pt_prev = *pt; 2214 2215 list_for_each_entry_rcu(ptype, ptype_list, list) { 2216 if (ptype->type != type) 2217 continue; 2218 if (pt_prev) 2219 deliver_skb(skb, pt_prev, orig_dev); 2220 pt_prev = ptype; 2221 } 2222 *pt = pt_prev; 2223 } 2224 2225 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) 2226 { 2227 if (!ptype->af_packet_priv || !skb->sk) 2228 return false; 2229 2230 if (ptype->id_match) 2231 return ptype->id_match(ptype, skb->sk); 2232 else if ((struct sock *)ptype->af_packet_priv == skb->sk) 2233 return true; 2234 2235 return false; 2236 } 2237 2238 /** 2239 * dev_nit_active - return true if any network interface taps are in use 2240 * 2241 * @dev: network device to check for the presence of taps 2242 */ 2243 bool dev_nit_active(struct net_device *dev) 2244 { 2245 return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all); 2246 } 2247 EXPORT_SYMBOL_GPL(dev_nit_active); 2248 2249 /* 2250 * Support routine. Sends outgoing frames to any network 2251 * taps currently in use. 2252 */ 2253 2254 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) 2255 { 2256 struct packet_type *ptype; 2257 struct sk_buff *skb2 = NULL; 2258 struct packet_type *pt_prev = NULL; 2259 struct list_head *ptype_list = &ptype_all; 2260 2261 rcu_read_lock(); 2262 again: 2263 list_for_each_entry_rcu(ptype, ptype_list, list) { 2264 if (ptype->ignore_outgoing) 2265 continue; 2266 2267 /* Never send packets back to the socket 2268 * they originated from - MvS (miquels@drinkel.ow.org) 2269 */ 2270 if (skb_loop_sk(ptype, skb)) 2271 continue; 2272 2273 if (pt_prev) { 2274 deliver_skb(skb2, pt_prev, skb->dev); 2275 pt_prev = ptype; 2276 continue; 2277 } 2278 2279 /* need to clone skb, done only once */ 2280 skb2 = skb_clone(skb, GFP_ATOMIC); 2281 if (!skb2) 2282 goto out_unlock; 2283 2284 net_timestamp_set(skb2); 2285 2286 /* skb->nh should be correctly 2287 * set by sender, so that the second statement is 2288 * just protection against buggy protocols. 2289 */ 2290 skb_reset_mac_header(skb2); 2291 2292 if (skb_network_header(skb2) < skb2->data || 2293 skb_network_header(skb2) > skb_tail_pointer(skb2)) { 2294 net_crit_ratelimited("protocol %04x is buggy, dev %s\n", 2295 ntohs(skb2->protocol), 2296 dev->name); 2297 skb_reset_network_header(skb2); 2298 } 2299 2300 skb2->transport_header = skb2->network_header; 2301 skb2->pkt_type = PACKET_OUTGOING; 2302 pt_prev = ptype; 2303 } 2304 2305 if (ptype_list == &ptype_all) { 2306 ptype_list = &dev->ptype_all; 2307 goto again; 2308 } 2309 out_unlock: 2310 if (pt_prev) { 2311 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC)) 2312 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); 2313 else 2314 kfree_skb(skb2); 2315 } 2316 rcu_read_unlock(); 2317 } 2318 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit); 2319 2320 /** 2321 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change 2322 * @dev: Network device 2323 * @txq: number of queues available 2324 * 2325 * If real_num_tx_queues is changed the tc mappings may no longer be 2326 * valid. To resolve this verify the tc mapping remains valid and if 2327 * not NULL the mapping. With no priorities mapping to this 2328 * offset/count pair it will no longer be used. In the worst case TC0 2329 * is invalid nothing can be done so disable priority mappings. If is 2330 * expected that drivers will fix this mapping if they can before 2331 * calling netif_set_real_num_tx_queues. 2332 */ 2333 static void netif_setup_tc(struct net_device *dev, unsigned int txq) 2334 { 2335 int i; 2336 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2337 2338 /* If TC0 is invalidated disable TC mapping */ 2339 if (tc->offset + tc->count > txq) { 2340 netdev_warn(dev, "Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); 2341 dev->num_tc = 0; 2342 return; 2343 } 2344 2345 /* Invalidated prio to tc mappings set to TC0 */ 2346 for (i = 1; i < TC_BITMASK + 1; i++) { 2347 int q = netdev_get_prio_tc_map(dev, i); 2348 2349 tc = &dev->tc_to_txq[q]; 2350 if (tc->offset + tc->count > txq) { 2351 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", 2352 i, q); 2353 netdev_set_prio_tc_map(dev, i, 0); 2354 } 2355 } 2356 } 2357 2358 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq) 2359 { 2360 if (dev->num_tc) { 2361 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2362 int i; 2363 2364 /* walk through the TCs and see if it falls into any of them */ 2365 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) { 2366 if ((txq - tc->offset) < tc->count) 2367 return i; 2368 } 2369 2370 /* didn't find it, just return -1 to indicate no match */ 2371 return -1; 2372 } 2373 2374 return 0; 2375 } 2376 EXPORT_SYMBOL(netdev_txq_to_tc); 2377 2378 #ifdef CONFIG_XPS 2379 static struct static_key xps_needed __read_mostly; 2380 static struct static_key xps_rxqs_needed __read_mostly; 2381 static DEFINE_MUTEX(xps_map_mutex); 2382 #define xmap_dereference(P) \ 2383 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) 2384 2385 static bool remove_xps_queue(struct xps_dev_maps *dev_maps, 2386 struct xps_dev_maps *old_maps, int tci, u16 index) 2387 { 2388 struct xps_map *map = NULL; 2389 int pos; 2390 2391 map = xmap_dereference(dev_maps->attr_map[tci]); 2392 if (!map) 2393 return false; 2394 2395 for (pos = map->len; pos--;) { 2396 if (map->queues[pos] != index) 2397 continue; 2398 2399 if (map->len > 1) { 2400 map->queues[pos] = map->queues[--map->len]; 2401 break; 2402 } 2403 2404 if (old_maps) 2405 RCU_INIT_POINTER(old_maps->attr_map[tci], NULL); 2406 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2407 kfree_rcu(map, rcu); 2408 return false; 2409 } 2410 2411 return true; 2412 } 2413 2414 static bool remove_xps_queue_cpu(struct net_device *dev, 2415 struct xps_dev_maps *dev_maps, 2416 int cpu, u16 offset, u16 count) 2417 { 2418 int num_tc = dev_maps->num_tc; 2419 bool active = false; 2420 int tci; 2421 2422 for (tci = cpu * num_tc; num_tc--; tci++) { 2423 int i, j; 2424 2425 for (i = count, j = offset; i--; j++) { 2426 if (!remove_xps_queue(dev_maps, NULL, tci, j)) 2427 break; 2428 } 2429 2430 active |= i < 0; 2431 } 2432 2433 return active; 2434 } 2435 2436 static void reset_xps_maps(struct net_device *dev, 2437 struct xps_dev_maps *dev_maps, 2438 enum xps_map_type type) 2439 { 2440 static_key_slow_dec_cpuslocked(&xps_needed); 2441 if (type == XPS_RXQS) 2442 static_key_slow_dec_cpuslocked(&xps_rxqs_needed); 2443 2444 RCU_INIT_POINTER(dev->xps_maps[type], NULL); 2445 2446 kfree_rcu(dev_maps, rcu); 2447 } 2448 2449 static void clean_xps_maps(struct net_device *dev, enum xps_map_type type, 2450 u16 offset, u16 count) 2451 { 2452 struct xps_dev_maps *dev_maps; 2453 bool active = false; 2454 int i, j; 2455 2456 dev_maps = xmap_dereference(dev->xps_maps[type]); 2457 if (!dev_maps) 2458 return; 2459 2460 for (j = 0; j < dev_maps->nr_ids; j++) 2461 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, count); 2462 if (!active) 2463 reset_xps_maps(dev, dev_maps, type); 2464 2465 if (type == XPS_CPUS) { 2466 for (i = offset + (count - 1); count--; i--) 2467 netdev_queue_numa_node_write( 2468 netdev_get_tx_queue(dev, i), NUMA_NO_NODE); 2469 } 2470 } 2471 2472 static void netif_reset_xps_queues(struct net_device *dev, u16 offset, 2473 u16 count) 2474 { 2475 if (!static_key_false(&xps_needed)) 2476 return; 2477 2478 cpus_read_lock(); 2479 mutex_lock(&xps_map_mutex); 2480 2481 if (static_key_false(&xps_rxqs_needed)) 2482 clean_xps_maps(dev, XPS_RXQS, offset, count); 2483 2484 clean_xps_maps(dev, XPS_CPUS, offset, count); 2485 2486 mutex_unlock(&xps_map_mutex); 2487 cpus_read_unlock(); 2488 } 2489 2490 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) 2491 { 2492 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index); 2493 } 2494 2495 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index, 2496 u16 index, bool is_rxqs_map) 2497 { 2498 struct xps_map *new_map; 2499 int alloc_len = XPS_MIN_MAP_ALLOC; 2500 int i, pos; 2501 2502 for (pos = 0; map && pos < map->len; pos++) { 2503 if (map->queues[pos] != index) 2504 continue; 2505 return map; 2506 } 2507 2508 /* Need to add tx-queue to this CPU's/rx-queue's existing map */ 2509 if (map) { 2510 if (pos < map->alloc_len) 2511 return map; 2512 2513 alloc_len = map->alloc_len * 2; 2514 } 2515 2516 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's 2517 * map 2518 */ 2519 if (is_rxqs_map) 2520 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL); 2521 else 2522 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, 2523 cpu_to_node(attr_index)); 2524 if (!new_map) 2525 return NULL; 2526 2527 for (i = 0; i < pos; i++) 2528 new_map->queues[i] = map->queues[i]; 2529 new_map->alloc_len = alloc_len; 2530 new_map->len = pos; 2531 2532 return new_map; 2533 } 2534 2535 /* Copy xps maps at a given index */ 2536 static void xps_copy_dev_maps(struct xps_dev_maps *dev_maps, 2537 struct xps_dev_maps *new_dev_maps, int index, 2538 int tc, bool skip_tc) 2539 { 2540 int i, tci = index * dev_maps->num_tc; 2541 struct xps_map *map; 2542 2543 /* copy maps belonging to foreign traffic classes */ 2544 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2545 if (i == tc && skip_tc) 2546 continue; 2547 2548 /* fill in the new device map from the old device map */ 2549 map = xmap_dereference(dev_maps->attr_map[tci]); 2550 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2551 } 2552 } 2553 2554 /* Must be called under cpus_read_lock */ 2555 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask, 2556 u16 index, enum xps_map_type type) 2557 { 2558 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL, *old_dev_maps = NULL; 2559 const unsigned long *online_mask = NULL; 2560 bool active = false, copy = false; 2561 int i, j, tci, numa_node_id = -2; 2562 int maps_sz, num_tc = 1, tc = 0; 2563 struct xps_map *map, *new_map; 2564 unsigned int nr_ids; 2565 2566 WARN_ON_ONCE(index >= dev->num_tx_queues); 2567 2568 if (dev->num_tc) { 2569 /* Do not allow XPS on subordinate device directly */ 2570 num_tc = dev->num_tc; 2571 if (num_tc < 0) 2572 return -EINVAL; 2573 2574 /* If queue belongs to subordinate dev use its map */ 2575 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev; 2576 2577 tc = netdev_txq_to_tc(dev, index); 2578 if (tc < 0) 2579 return -EINVAL; 2580 } 2581 2582 mutex_lock(&xps_map_mutex); 2583 2584 dev_maps = xmap_dereference(dev->xps_maps[type]); 2585 if (type == XPS_RXQS) { 2586 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues); 2587 nr_ids = dev->num_rx_queues; 2588 } else { 2589 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc); 2590 if (num_possible_cpus() > 1) 2591 online_mask = cpumask_bits(cpu_online_mask); 2592 nr_ids = nr_cpu_ids; 2593 } 2594 2595 if (maps_sz < L1_CACHE_BYTES) 2596 maps_sz = L1_CACHE_BYTES; 2597 2598 /* The old dev_maps could be larger or smaller than the one we're 2599 * setting up now, as dev->num_tc or nr_ids could have been updated in 2600 * between. We could try to be smart, but let's be safe instead and only 2601 * copy foreign traffic classes if the two map sizes match. 2602 */ 2603 if (dev_maps && 2604 dev_maps->num_tc == num_tc && dev_maps->nr_ids == nr_ids) 2605 copy = true; 2606 2607 /* allocate memory for queue storage */ 2608 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids), 2609 j < nr_ids;) { 2610 if (!new_dev_maps) { 2611 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); 2612 if (!new_dev_maps) { 2613 mutex_unlock(&xps_map_mutex); 2614 return -ENOMEM; 2615 } 2616 2617 new_dev_maps->nr_ids = nr_ids; 2618 new_dev_maps->num_tc = num_tc; 2619 } 2620 2621 tci = j * num_tc + tc; 2622 map = copy ? xmap_dereference(dev_maps->attr_map[tci]) : NULL; 2623 2624 map = expand_xps_map(map, j, index, type == XPS_RXQS); 2625 if (!map) 2626 goto error; 2627 2628 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2629 } 2630 2631 if (!new_dev_maps) 2632 goto out_no_new_maps; 2633 2634 if (!dev_maps) { 2635 /* Increment static keys at most once per type */ 2636 static_key_slow_inc_cpuslocked(&xps_needed); 2637 if (type == XPS_RXQS) 2638 static_key_slow_inc_cpuslocked(&xps_rxqs_needed); 2639 } 2640 2641 for (j = 0; j < nr_ids; j++) { 2642 bool skip_tc = false; 2643 2644 tci = j * num_tc + tc; 2645 if (netif_attr_test_mask(j, mask, nr_ids) && 2646 netif_attr_test_online(j, online_mask, nr_ids)) { 2647 /* add tx-queue to CPU/rx-queue maps */ 2648 int pos = 0; 2649 2650 skip_tc = true; 2651 2652 map = xmap_dereference(new_dev_maps->attr_map[tci]); 2653 while ((pos < map->len) && (map->queues[pos] != index)) 2654 pos++; 2655 2656 if (pos == map->len) 2657 map->queues[map->len++] = index; 2658 #ifdef CONFIG_NUMA 2659 if (type == XPS_CPUS) { 2660 if (numa_node_id == -2) 2661 numa_node_id = cpu_to_node(j); 2662 else if (numa_node_id != cpu_to_node(j)) 2663 numa_node_id = -1; 2664 } 2665 #endif 2666 } 2667 2668 if (copy) 2669 xps_copy_dev_maps(dev_maps, new_dev_maps, j, tc, 2670 skip_tc); 2671 } 2672 2673 rcu_assign_pointer(dev->xps_maps[type], new_dev_maps); 2674 2675 /* Cleanup old maps */ 2676 if (!dev_maps) 2677 goto out_no_old_maps; 2678 2679 for (j = 0; j < dev_maps->nr_ids; j++) { 2680 for (i = num_tc, tci = j * dev_maps->num_tc; i--; tci++) { 2681 map = xmap_dereference(dev_maps->attr_map[tci]); 2682 if (!map) 2683 continue; 2684 2685 if (copy) { 2686 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2687 if (map == new_map) 2688 continue; 2689 } 2690 2691 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2692 kfree_rcu(map, rcu); 2693 } 2694 } 2695 2696 old_dev_maps = dev_maps; 2697 2698 out_no_old_maps: 2699 dev_maps = new_dev_maps; 2700 active = true; 2701 2702 out_no_new_maps: 2703 if (type == XPS_CPUS) 2704 /* update Tx queue numa node */ 2705 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), 2706 (numa_node_id >= 0) ? 2707 numa_node_id : NUMA_NO_NODE); 2708 2709 if (!dev_maps) 2710 goto out_no_maps; 2711 2712 /* removes tx-queue from unused CPUs/rx-queues */ 2713 for (j = 0; j < dev_maps->nr_ids; j++) { 2714 tci = j * dev_maps->num_tc; 2715 2716 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2717 if (i == tc && 2718 netif_attr_test_mask(j, mask, dev_maps->nr_ids) && 2719 netif_attr_test_online(j, online_mask, dev_maps->nr_ids)) 2720 continue; 2721 2722 active |= remove_xps_queue(dev_maps, 2723 copy ? old_dev_maps : NULL, 2724 tci, index); 2725 } 2726 } 2727 2728 if (old_dev_maps) 2729 kfree_rcu(old_dev_maps, rcu); 2730 2731 /* free map if not active */ 2732 if (!active) 2733 reset_xps_maps(dev, dev_maps, type); 2734 2735 out_no_maps: 2736 mutex_unlock(&xps_map_mutex); 2737 2738 return 0; 2739 error: 2740 /* remove any maps that we added */ 2741 for (j = 0; j < nr_ids; j++) { 2742 for (i = num_tc, tci = j * num_tc; i--; tci++) { 2743 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2744 map = copy ? 2745 xmap_dereference(dev_maps->attr_map[tci]) : 2746 NULL; 2747 if (new_map && new_map != map) 2748 kfree(new_map); 2749 } 2750 } 2751 2752 mutex_unlock(&xps_map_mutex); 2753 2754 kfree(new_dev_maps); 2755 return -ENOMEM; 2756 } 2757 EXPORT_SYMBOL_GPL(__netif_set_xps_queue); 2758 2759 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, 2760 u16 index) 2761 { 2762 int ret; 2763 2764 cpus_read_lock(); 2765 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, XPS_CPUS); 2766 cpus_read_unlock(); 2767 2768 return ret; 2769 } 2770 EXPORT_SYMBOL(netif_set_xps_queue); 2771 2772 #endif 2773 static void netdev_unbind_all_sb_channels(struct net_device *dev) 2774 { 2775 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2776 2777 /* Unbind any subordinate channels */ 2778 while (txq-- != &dev->_tx[0]) { 2779 if (txq->sb_dev) 2780 netdev_unbind_sb_channel(dev, txq->sb_dev); 2781 } 2782 } 2783 2784 void netdev_reset_tc(struct net_device *dev) 2785 { 2786 #ifdef CONFIG_XPS 2787 netif_reset_xps_queues_gt(dev, 0); 2788 #endif 2789 netdev_unbind_all_sb_channels(dev); 2790 2791 /* Reset TC configuration of device */ 2792 dev->num_tc = 0; 2793 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq)); 2794 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map)); 2795 } 2796 EXPORT_SYMBOL(netdev_reset_tc); 2797 2798 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset) 2799 { 2800 if (tc >= dev->num_tc) 2801 return -EINVAL; 2802 2803 #ifdef CONFIG_XPS 2804 netif_reset_xps_queues(dev, offset, count); 2805 #endif 2806 dev->tc_to_txq[tc].count = count; 2807 dev->tc_to_txq[tc].offset = offset; 2808 return 0; 2809 } 2810 EXPORT_SYMBOL(netdev_set_tc_queue); 2811 2812 int netdev_set_num_tc(struct net_device *dev, u8 num_tc) 2813 { 2814 if (num_tc > TC_MAX_QUEUE) 2815 return -EINVAL; 2816 2817 #ifdef CONFIG_XPS 2818 netif_reset_xps_queues_gt(dev, 0); 2819 #endif 2820 netdev_unbind_all_sb_channels(dev); 2821 2822 dev->num_tc = num_tc; 2823 return 0; 2824 } 2825 EXPORT_SYMBOL(netdev_set_num_tc); 2826 2827 void netdev_unbind_sb_channel(struct net_device *dev, 2828 struct net_device *sb_dev) 2829 { 2830 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2831 2832 #ifdef CONFIG_XPS 2833 netif_reset_xps_queues_gt(sb_dev, 0); 2834 #endif 2835 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq)); 2836 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map)); 2837 2838 while (txq-- != &dev->_tx[0]) { 2839 if (txq->sb_dev == sb_dev) 2840 txq->sb_dev = NULL; 2841 } 2842 } 2843 EXPORT_SYMBOL(netdev_unbind_sb_channel); 2844 2845 int netdev_bind_sb_channel_queue(struct net_device *dev, 2846 struct net_device *sb_dev, 2847 u8 tc, u16 count, u16 offset) 2848 { 2849 /* Make certain the sb_dev and dev are already configured */ 2850 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc) 2851 return -EINVAL; 2852 2853 /* We cannot hand out queues we don't have */ 2854 if ((offset + count) > dev->real_num_tx_queues) 2855 return -EINVAL; 2856 2857 /* Record the mapping */ 2858 sb_dev->tc_to_txq[tc].count = count; 2859 sb_dev->tc_to_txq[tc].offset = offset; 2860 2861 /* Provide a way for Tx queue to find the tc_to_txq map or 2862 * XPS map for itself. 2863 */ 2864 while (count--) 2865 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev; 2866 2867 return 0; 2868 } 2869 EXPORT_SYMBOL(netdev_bind_sb_channel_queue); 2870 2871 int netdev_set_sb_channel(struct net_device *dev, u16 channel) 2872 { 2873 /* Do not use a multiqueue device to represent a subordinate channel */ 2874 if (netif_is_multiqueue(dev)) 2875 return -ENODEV; 2876 2877 /* We allow channels 1 - 32767 to be used for subordinate channels. 2878 * Channel 0 is meant to be "native" mode and used only to represent 2879 * the main root device. We allow writing 0 to reset the device back 2880 * to normal mode after being used as a subordinate channel. 2881 */ 2882 if (channel > S16_MAX) 2883 return -EINVAL; 2884 2885 dev->num_tc = -channel; 2886 2887 return 0; 2888 } 2889 EXPORT_SYMBOL(netdev_set_sb_channel); 2890 2891 /* 2892 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues 2893 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed. 2894 */ 2895 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) 2896 { 2897 bool disabling; 2898 int rc; 2899 2900 disabling = txq < dev->real_num_tx_queues; 2901 2902 if (txq < 1 || txq > dev->num_tx_queues) 2903 return -EINVAL; 2904 2905 if (dev->reg_state == NETREG_REGISTERED || 2906 dev->reg_state == NETREG_UNREGISTERING) { 2907 ASSERT_RTNL(); 2908 2909 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, 2910 txq); 2911 if (rc) 2912 return rc; 2913 2914 if (dev->num_tc) 2915 netif_setup_tc(dev, txq); 2916 2917 dev_qdisc_change_real_num_tx(dev, txq); 2918 2919 dev->real_num_tx_queues = txq; 2920 2921 if (disabling) { 2922 synchronize_net(); 2923 qdisc_reset_all_tx_gt(dev, txq); 2924 #ifdef CONFIG_XPS 2925 netif_reset_xps_queues_gt(dev, txq); 2926 #endif 2927 } 2928 } else { 2929 dev->real_num_tx_queues = txq; 2930 } 2931 2932 return 0; 2933 } 2934 EXPORT_SYMBOL(netif_set_real_num_tx_queues); 2935 2936 #ifdef CONFIG_SYSFS 2937 /** 2938 * netif_set_real_num_rx_queues - set actual number of RX queues used 2939 * @dev: Network device 2940 * @rxq: Actual number of RX queues 2941 * 2942 * This must be called either with the rtnl_lock held or before 2943 * registration of the net device. Returns 0 on success, or a 2944 * negative error code. If called before registration, it always 2945 * succeeds. 2946 */ 2947 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) 2948 { 2949 int rc; 2950 2951 if (rxq < 1 || rxq > dev->num_rx_queues) 2952 return -EINVAL; 2953 2954 if (dev->reg_state == NETREG_REGISTERED) { 2955 ASSERT_RTNL(); 2956 2957 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, 2958 rxq); 2959 if (rc) 2960 return rc; 2961 } 2962 2963 dev->real_num_rx_queues = rxq; 2964 return 0; 2965 } 2966 EXPORT_SYMBOL(netif_set_real_num_rx_queues); 2967 #endif 2968 2969 /** 2970 * netif_set_real_num_queues - set actual number of RX and TX queues used 2971 * @dev: Network device 2972 * @txq: Actual number of TX queues 2973 * @rxq: Actual number of RX queues 2974 * 2975 * Set the real number of both TX and RX queues. 2976 * Does nothing if the number of queues is already correct. 2977 */ 2978 int netif_set_real_num_queues(struct net_device *dev, 2979 unsigned int txq, unsigned int rxq) 2980 { 2981 unsigned int old_rxq = dev->real_num_rx_queues; 2982 int err; 2983 2984 if (txq < 1 || txq > dev->num_tx_queues || 2985 rxq < 1 || rxq > dev->num_rx_queues) 2986 return -EINVAL; 2987 2988 /* Start from increases, so the error path only does decreases - 2989 * decreases can't fail. 2990 */ 2991 if (rxq > dev->real_num_rx_queues) { 2992 err = netif_set_real_num_rx_queues(dev, rxq); 2993 if (err) 2994 return err; 2995 } 2996 if (txq > dev->real_num_tx_queues) { 2997 err = netif_set_real_num_tx_queues(dev, txq); 2998 if (err) 2999 goto undo_rx; 3000 } 3001 if (rxq < dev->real_num_rx_queues) 3002 WARN_ON(netif_set_real_num_rx_queues(dev, rxq)); 3003 if (txq < dev->real_num_tx_queues) 3004 WARN_ON(netif_set_real_num_tx_queues(dev, txq)); 3005 3006 return 0; 3007 undo_rx: 3008 WARN_ON(netif_set_real_num_rx_queues(dev, old_rxq)); 3009 return err; 3010 } 3011 EXPORT_SYMBOL(netif_set_real_num_queues); 3012 3013 /** 3014 * netif_set_tso_max_size() - set the max size of TSO frames supported 3015 * @dev: netdev to update 3016 * @size: max skb->len of a TSO frame 3017 * 3018 * Set the limit on the size of TSO super-frames the device can handle. 3019 * Unless explicitly set the stack will assume the value of 3020 * %GSO_LEGACY_MAX_SIZE. 3021 */ 3022 void netif_set_tso_max_size(struct net_device *dev, unsigned int size) 3023 { 3024 dev->tso_max_size = min(GSO_MAX_SIZE, size); 3025 if (size < READ_ONCE(dev->gso_max_size)) 3026 netif_set_gso_max_size(dev, size); 3027 if (size < READ_ONCE(dev->gso_ipv4_max_size)) 3028 netif_set_gso_ipv4_max_size(dev, size); 3029 } 3030 EXPORT_SYMBOL(netif_set_tso_max_size); 3031 3032 /** 3033 * netif_set_tso_max_segs() - set the max number of segs supported for TSO 3034 * @dev: netdev to update 3035 * @segs: max number of TCP segments 3036 * 3037 * Set the limit on the number of TCP segments the device can generate from 3038 * a single TSO super-frame. 3039 * Unless explicitly set the stack will assume the value of %GSO_MAX_SEGS. 3040 */ 3041 void netif_set_tso_max_segs(struct net_device *dev, unsigned int segs) 3042 { 3043 dev->tso_max_segs = segs; 3044 if (segs < READ_ONCE(dev->gso_max_segs)) 3045 netif_set_gso_max_segs(dev, segs); 3046 } 3047 EXPORT_SYMBOL(netif_set_tso_max_segs); 3048 3049 /** 3050 * netif_inherit_tso_max() - copy all TSO limits from a lower device to an upper 3051 * @to: netdev to update 3052 * @from: netdev from which to copy the limits 3053 */ 3054 void netif_inherit_tso_max(struct net_device *to, const struct net_device *from) 3055 { 3056 netif_set_tso_max_size(to, from->tso_max_size); 3057 netif_set_tso_max_segs(to, from->tso_max_segs); 3058 } 3059 EXPORT_SYMBOL(netif_inherit_tso_max); 3060 3061 /** 3062 * netif_get_num_default_rss_queues - default number of RSS queues 3063 * 3064 * Default value is the number of physical cores if there are only 1 or 2, or 3065 * divided by 2 if there are more. 3066 */ 3067 int netif_get_num_default_rss_queues(void) 3068 { 3069 cpumask_var_t cpus; 3070 int cpu, count = 0; 3071 3072 if (unlikely(is_kdump_kernel() || !zalloc_cpumask_var(&cpus, GFP_KERNEL))) 3073 return 1; 3074 3075 cpumask_copy(cpus, cpu_online_mask); 3076 for_each_cpu(cpu, cpus) { 3077 ++count; 3078 cpumask_andnot(cpus, cpus, topology_sibling_cpumask(cpu)); 3079 } 3080 free_cpumask_var(cpus); 3081 3082 return count > 2 ? DIV_ROUND_UP(count, 2) : count; 3083 } 3084 EXPORT_SYMBOL(netif_get_num_default_rss_queues); 3085 3086 static void __netif_reschedule(struct Qdisc *q) 3087 { 3088 struct softnet_data *sd; 3089 unsigned long flags; 3090 3091 local_irq_save(flags); 3092 sd = this_cpu_ptr(&softnet_data); 3093 q->next_sched = NULL; 3094 *sd->output_queue_tailp = q; 3095 sd->output_queue_tailp = &q->next_sched; 3096 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3097 local_irq_restore(flags); 3098 } 3099 3100 void __netif_schedule(struct Qdisc *q) 3101 { 3102 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) 3103 __netif_reschedule(q); 3104 } 3105 EXPORT_SYMBOL(__netif_schedule); 3106 3107 struct dev_kfree_skb_cb { 3108 enum skb_drop_reason reason; 3109 }; 3110 3111 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) 3112 { 3113 return (struct dev_kfree_skb_cb *)skb->cb; 3114 } 3115 3116 void netif_schedule_queue(struct netdev_queue *txq) 3117 { 3118 rcu_read_lock(); 3119 if (!netif_xmit_stopped(txq)) { 3120 struct Qdisc *q = rcu_dereference(txq->qdisc); 3121 3122 __netif_schedule(q); 3123 } 3124 rcu_read_unlock(); 3125 } 3126 EXPORT_SYMBOL(netif_schedule_queue); 3127 3128 void netif_tx_wake_queue(struct netdev_queue *dev_queue) 3129 { 3130 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { 3131 struct Qdisc *q; 3132 3133 rcu_read_lock(); 3134 q = rcu_dereference(dev_queue->qdisc); 3135 __netif_schedule(q); 3136 rcu_read_unlock(); 3137 } 3138 } 3139 EXPORT_SYMBOL(netif_tx_wake_queue); 3140 3141 void dev_kfree_skb_irq_reason(struct sk_buff *skb, enum skb_drop_reason reason) 3142 { 3143 unsigned long flags; 3144 3145 if (unlikely(!skb)) 3146 return; 3147 3148 if (likely(refcount_read(&skb->users) == 1)) { 3149 smp_rmb(); 3150 refcount_set(&skb->users, 0); 3151 } else if (likely(!refcount_dec_and_test(&skb->users))) { 3152 return; 3153 } 3154 get_kfree_skb_cb(skb)->reason = reason; 3155 local_irq_save(flags); 3156 skb->next = __this_cpu_read(softnet_data.completion_queue); 3157 __this_cpu_write(softnet_data.completion_queue, skb); 3158 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3159 local_irq_restore(flags); 3160 } 3161 EXPORT_SYMBOL(dev_kfree_skb_irq_reason); 3162 3163 void dev_kfree_skb_any_reason(struct sk_buff *skb, enum skb_drop_reason reason) 3164 { 3165 if (in_hardirq() || irqs_disabled()) 3166 dev_kfree_skb_irq_reason(skb, reason); 3167 else 3168 kfree_skb_reason(skb, reason); 3169 } 3170 EXPORT_SYMBOL(dev_kfree_skb_any_reason); 3171 3172 3173 /** 3174 * netif_device_detach - mark device as removed 3175 * @dev: network device 3176 * 3177 * Mark device as removed from system and therefore no longer available. 3178 */ 3179 void netif_device_detach(struct net_device *dev) 3180 { 3181 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && 3182 netif_running(dev)) { 3183 netif_tx_stop_all_queues(dev); 3184 } 3185 } 3186 EXPORT_SYMBOL(netif_device_detach); 3187 3188 /** 3189 * netif_device_attach - mark device as attached 3190 * @dev: network device 3191 * 3192 * Mark device as attached from system and restart if needed. 3193 */ 3194 void netif_device_attach(struct net_device *dev) 3195 { 3196 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && 3197 netif_running(dev)) { 3198 netif_tx_wake_all_queues(dev); 3199 __netdev_watchdog_up(dev); 3200 } 3201 } 3202 EXPORT_SYMBOL(netif_device_attach); 3203 3204 /* 3205 * Returns a Tx hash based on the given packet descriptor a Tx queues' number 3206 * to be used as a distribution range. 3207 */ 3208 static u16 skb_tx_hash(const struct net_device *dev, 3209 const struct net_device *sb_dev, 3210 struct sk_buff *skb) 3211 { 3212 u32 hash; 3213 u16 qoffset = 0; 3214 u16 qcount = dev->real_num_tx_queues; 3215 3216 if (dev->num_tc) { 3217 u8 tc = netdev_get_prio_tc_map(dev, skb->priority); 3218 3219 qoffset = sb_dev->tc_to_txq[tc].offset; 3220 qcount = sb_dev->tc_to_txq[tc].count; 3221 if (unlikely(!qcount)) { 3222 net_warn_ratelimited("%s: invalid qcount, qoffset %u for tc %u\n", 3223 sb_dev->name, qoffset, tc); 3224 qoffset = 0; 3225 qcount = dev->real_num_tx_queues; 3226 } 3227 } 3228 3229 if (skb_rx_queue_recorded(skb)) { 3230 DEBUG_NET_WARN_ON_ONCE(qcount == 0); 3231 hash = skb_get_rx_queue(skb); 3232 if (hash >= qoffset) 3233 hash -= qoffset; 3234 while (unlikely(hash >= qcount)) 3235 hash -= qcount; 3236 return hash + qoffset; 3237 } 3238 3239 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; 3240 } 3241 3242 void skb_warn_bad_offload(const struct sk_buff *skb) 3243 { 3244 static const netdev_features_t null_features; 3245 struct net_device *dev = skb->dev; 3246 const char *name = ""; 3247 3248 if (!net_ratelimit()) 3249 return; 3250 3251 if (dev) { 3252 if (dev->dev.parent) 3253 name = dev_driver_string(dev->dev.parent); 3254 else 3255 name = netdev_name(dev); 3256 } 3257 skb_dump(KERN_WARNING, skb, false); 3258 WARN(1, "%s: caps=(%pNF, %pNF)\n", 3259 name, dev ? &dev->features : &null_features, 3260 skb->sk ? &skb->sk->sk_route_caps : &null_features); 3261 } 3262 3263 /* 3264 * Invalidate hardware checksum when packet is to be mangled, and 3265 * complete checksum manually on outgoing path. 3266 */ 3267 int skb_checksum_help(struct sk_buff *skb) 3268 { 3269 __wsum csum; 3270 int ret = 0, offset; 3271 3272 if (skb->ip_summed == CHECKSUM_COMPLETE) 3273 goto out_set_summed; 3274 3275 if (unlikely(skb_is_gso(skb))) { 3276 skb_warn_bad_offload(skb); 3277 return -EINVAL; 3278 } 3279 3280 /* Before computing a checksum, we should make sure no frag could 3281 * be modified by an external entity : checksum could be wrong. 3282 */ 3283 if (skb_has_shared_frag(skb)) { 3284 ret = __skb_linearize(skb); 3285 if (ret) 3286 goto out; 3287 } 3288 3289 offset = skb_checksum_start_offset(skb); 3290 ret = -EINVAL; 3291 if (unlikely(offset >= skb_headlen(skb))) { 3292 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false); 3293 WARN_ONCE(true, "offset (%d) >= skb_headlen() (%u)\n", 3294 offset, skb_headlen(skb)); 3295 goto out; 3296 } 3297 csum = skb_checksum(skb, offset, skb->len - offset, 0); 3298 3299 offset += skb->csum_offset; 3300 if (unlikely(offset + sizeof(__sum16) > skb_headlen(skb))) { 3301 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false); 3302 WARN_ONCE(true, "offset+2 (%zu) > skb_headlen() (%u)\n", 3303 offset + sizeof(__sum16), skb_headlen(skb)); 3304 goto out; 3305 } 3306 ret = skb_ensure_writable(skb, offset + sizeof(__sum16)); 3307 if (ret) 3308 goto out; 3309 3310 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0; 3311 out_set_summed: 3312 skb->ip_summed = CHECKSUM_NONE; 3313 out: 3314 return ret; 3315 } 3316 EXPORT_SYMBOL(skb_checksum_help); 3317 3318 int skb_crc32c_csum_help(struct sk_buff *skb) 3319 { 3320 __le32 crc32c_csum; 3321 int ret = 0, offset, start; 3322 3323 if (skb->ip_summed != CHECKSUM_PARTIAL) 3324 goto out; 3325 3326 if (unlikely(skb_is_gso(skb))) 3327 goto out; 3328 3329 /* Before computing a checksum, we should make sure no frag could 3330 * be modified by an external entity : checksum could be wrong. 3331 */ 3332 if (unlikely(skb_has_shared_frag(skb))) { 3333 ret = __skb_linearize(skb); 3334 if (ret) 3335 goto out; 3336 } 3337 start = skb_checksum_start_offset(skb); 3338 offset = start + offsetof(struct sctphdr, checksum); 3339 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { 3340 ret = -EINVAL; 3341 goto out; 3342 } 3343 3344 ret = skb_ensure_writable(skb, offset + sizeof(__le32)); 3345 if (ret) 3346 goto out; 3347 3348 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start, 3349 skb->len - start, ~(__u32)0, 3350 crc32c_csum_stub)); 3351 *(__le32 *)(skb->data + offset) = crc32c_csum; 3352 skb_reset_csum_not_inet(skb); 3353 out: 3354 return ret; 3355 } 3356 3357 __be16 skb_network_protocol(struct sk_buff *skb, int *depth) 3358 { 3359 __be16 type = skb->protocol; 3360 3361 /* Tunnel gso handlers can set protocol to ethernet. */ 3362 if (type == htons(ETH_P_TEB)) { 3363 struct ethhdr *eth; 3364 3365 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) 3366 return 0; 3367 3368 eth = (struct ethhdr *)skb->data; 3369 type = eth->h_proto; 3370 } 3371 3372 return vlan_get_protocol_and_depth(skb, type, depth); 3373 } 3374 3375 3376 /* Take action when hardware reception checksum errors are detected. */ 3377 #ifdef CONFIG_BUG 3378 static void do_netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3379 { 3380 netdev_err(dev, "hw csum failure\n"); 3381 skb_dump(KERN_ERR, skb, true); 3382 dump_stack(); 3383 } 3384 3385 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3386 { 3387 DO_ONCE_LITE(do_netdev_rx_csum_fault, dev, skb); 3388 } 3389 EXPORT_SYMBOL(netdev_rx_csum_fault); 3390 #endif 3391 3392 /* XXX: check that highmem exists at all on the given machine. */ 3393 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) 3394 { 3395 #ifdef CONFIG_HIGHMEM 3396 int i; 3397 3398 if (!(dev->features & NETIF_F_HIGHDMA)) { 3399 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 3400 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 3401 3402 if (PageHighMem(skb_frag_page(frag))) 3403 return 1; 3404 } 3405 } 3406 #endif 3407 return 0; 3408 } 3409 3410 /* If MPLS offload request, verify we are testing hardware MPLS features 3411 * instead of standard features for the netdev. 3412 */ 3413 #if IS_ENABLED(CONFIG_NET_MPLS_GSO) 3414 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3415 netdev_features_t features, 3416 __be16 type) 3417 { 3418 if (eth_p_mpls(type)) 3419 features &= skb->dev->mpls_features; 3420 3421 return features; 3422 } 3423 #else 3424 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3425 netdev_features_t features, 3426 __be16 type) 3427 { 3428 return features; 3429 } 3430 #endif 3431 3432 static netdev_features_t harmonize_features(struct sk_buff *skb, 3433 netdev_features_t features) 3434 { 3435 __be16 type; 3436 3437 type = skb_network_protocol(skb, NULL); 3438 features = net_mpls_features(skb, features, type); 3439 3440 if (skb->ip_summed != CHECKSUM_NONE && 3441 !can_checksum_protocol(features, type)) { 3442 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); 3443 } 3444 if (illegal_highdma(skb->dev, skb)) 3445 features &= ~NETIF_F_SG; 3446 3447 return features; 3448 } 3449 3450 netdev_features_t passthru_features_check(struct sk_buff *skb, 3451 struct net_device *dev, 3452 netdev_features_t features) 3453 { 3454 return features; 3455 } 3456 EXPORT_SYMBOL(passthru_features_check); 3457 3458 static netdev_features_t dflt_features_check(struct sk_buff *skb, 3459 struct net_device *dev, 3460 netdev_features_t features) 3461 { 3462 return vlan_features_check(skb, features); 3463 } 3464 3465 static netdev_features_t gso_features_check(const struct sk_buff *skb, 3466 struct net_device *dev, 3467 netdev_features_t features) 3468 { 3469 u16 gso_segs = skb_shinfo(skb)->gso_segs; 3470 3471 if (gso_segs > READ_ONCE(dev->gso_max_segs)) 3472 return features & ~NETIF_F_GSO_MASK; 3473 3474 if (unlikely(skb->len >= READ_ONCE(dev->gso_max_size))) 3475 return features & ~NETIF_F_GSO_MASK; 3476 3477 if (!skb_shinfo(skb)->gso_type) { 3478 skb_warn_bad_offload(skb); 3479 return features & ~NETIF_F_GSO_MASK; 3480 } 3481 3482 /* Support for GSO partial features requires software 3483 * intervention before we can actually process the packets 3484 * so we need to strip support for any partial features now 3485 * and we can pull them back in after we have partially 3486 * segmented the frame. 3487 */ 3488 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)) 3489 features &= ~dev->gso_partial_features; 3490 3491 /* Make sure to clear the IPv4 ID mangling feature if the 3492 * IPv4 header has the potential to be fragmented. 3493 */ 3494 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { 3495 struct iphdr *iph = skb->encapsulation ? 3496 inner_ip_hdr(skb) : ip_hdr(skb); 3497 3498 if (!(iph->frag_off & htons(IP_DF))) 3499 features &= ~NETIF_F_TSO_MANGLEID; 3500 } 3501 3502 return features; 3503 } 3504 3505 netdev_features_t netif_skb_features(struct sk_buff *skb) 3506 { 3507 struct net_device *dev = skb->dev; 3508 netdev_features_t features = dev->features; 3509 3510 if (skb_is_gso(skb)) 3511 features = gso_features_check(skb, dev, features); 3512 3513 /* If encapsulation offload request, verify we are testing 3514 * hardware encapsulation features instead of standard 3515 * features for the netdev 3516 */ 3517 if (skb->encapsulation) 3518 features &= dev->hw_enc_features; 3519 3520 if (skb_vlan_tagged(skb)) 3521 features = netdev_intersect_features(features, 3522 dev->vlan_features | 3523 NETIF_F_HW_VLAN_CTAG_TX | 3524 NETIF_F_HW_VLAN_STAG_TX); 3525 3526 if (dev->netdev_ops->ndo_features_check) 3527 features &= dev->netdev_ops->ndo_features_check(skb, dev, 3528 features); 3529 else 3530 features &= dflt_features_check(skb, dev, features); 3531 3532 return harmonize_features(skb, features); 3533 } 3534 EXPORT_SYMBOL(netif_skb_features); 3535 3536 static int xmit_one(struct sk_buff *skb, struct net_device *dev, 3537 struct netdev_queue *txq, bool more) 3538 { 3539 unsigned int len; 3540 int rc; 3541 3542 if (dev_nit_active(dev)) 3543 dev_queue_xmit_nit(skb, dev); 3544 3545 len = skb->len; 3546 trace_net_dev_start_xmit(skb, dev); 3547 rc = netdev_start_xmit(skb, dev, txq, more); 3548 trace_net_dev_xmit(skb, rc, dev, len); 3549 3550 return rc; 3551 } 3552 3553 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, 3554 struct netdev_queue *txq, int *ret) 3555 { 3556 struct sk_buff *skb = first; 3557 int rc = NETDEV_TX_OK; 3558 3559 while (skb) { 3560 struct sk_buff *next = skb->next; 3561 3562 skb_mark_not_on_list(skb); 3563 rc = xmit_one(skb, dev, txq, next != NULL); 3564 if (unlikely(!dev_xmit_complete(rc))) { 3565 skb->next = next; 3566 goto out; 3567 } 3568 3569 skb = next; 3570 if (netif_tx_queue_stopped(txq) && skb) { 3571 rc = NETDEV_TX_BUSY; 3572 break; 3573 } 3574 } 3575 3576 out: 3577 *ret = rc; 3578 return skb; 3579 } 3580 3581 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, 3582 netdev_features_t features) 3583 { 3584 if (skb_vlan_tag_present(skb) && 3585 !vlan_hw_offload_capable(features, skb->vlan_proto)) 3586 skb = __vlan_hwaccel_push_inside(skb); 3587 return skb; 3588 } 3589 3590 int skb_csum_hwoffload_help(struct sk_buff *skb, 3591 const netdev_features_t features) 3592 { 3593 if (unlikely(skb_csum_is_sctp(skb))) 3594 return !!(features & NETIF_F_SCTP_CRC) ? 0 : 3595 skb_crc32c_csum_help(skb); 3596 3597 if (features & NETIF_F_HW_CSUM) 3598 return 0; 3599 3600 if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) { 3601 switch (skb->csum_offset) { 3602 case offsetof(struct tcphdr, check): 3603 case offsetof(struct udphdr, check): 3604 return 0; 3605 } 3606 } 3607 3608 return skb_checksum_help(skb); 3609 } 3610 EXPORT_SYMBOL(skb_csum_hwoffload_help); 3611 3612 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again) 3613 { 3614 netdev_features_t features; 3615 3616 features = netif_skb_features(skb); 3617 skb = validate_xmit_vlan(skb, features); 3618 if (unlikely(!skb)) 3619 goto out_null; 3620 3621 skb = sk_validate_xmit_skb(skb, dev); 3622 if (unlikely(!skb)) 3623 goto out_null; 3624 3625 if (netif_needs_gso(skb, features)) { 3626 struct sk_buff *segs; 3627 3628 segs = skb_gso_segment(skb, features); 3629 if (IS_ERR(segs)) { 3630 goto out_kfree_skb; 3631 } else if (segs) { 3632 consume_skb(skb); 3633 skb = segs; 3634 } 3635 } else { 3636 if (skb_needs_linearize(skb, features) && 3637 __skb_linearize(skb)) 3638 goto out_kfree_skb; 3639 3640 /* If packet is not checksummed and device does not 3641 * support checksumming for this protocol, complete 3642 * checksumming here. 3643 */ 3644 if (skb->ip_summed == CHECKSUM_PARTIAL) { 3645 if (skb->encapsulation) 3646 skb_set_inner_transport_header(skb, 3647 skb_checksum_start_offset(skb)); 3648 else 3649 skb_set_transport_header(skb, 3650 skb_checksum_start_offset(skb)); 3651 if (skb_csum_hwoffload_help(skb, features)) 3652 goto out_kfree_skb; 3653 } 3654 } 3655 3656 skb = validate_xmit_xfrm(skb, features, again); 3657 3658 return skb; 3659 3660 out_kfree_skb: 3661 kfree_skb(skb); 3662 out_null: 3663 dev_core_stats_tx_dropped_inc(dev); 3664 return NULL; 3665 } 3666 3667 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again) 3668 { 3669 struct sk_buff *next, *head = NULL, *tail; 3670 3671 for (; skb != NULL; skb = next) { 3672 next = skb->next; 3673 skb_mark_not_on_list(skb); 3674 3675 /* in case skb wont be segmented, point to itself */ 3676 skb->prev = skb; 3677 3678 skb = validate_xmit_skb(skb, dev, again); 3679 if (!skb) 3680 continue; 3681 3682 if (!head) 3683 head = skb; 3684 else 3685 tail->next = skb; 3686 /* If skb was segmented, skb->prev points to 3687 * the last segment. If not, it still contains skb. 3688 */ 3689 tail = skb->prev; 3690 } 3691 return head; 3692 } 3693 EXPORT_SYMBOL_GPL(validate_xmit_skb_list); 3694 3695 static void qdisc_pkt_len_init(struct sk_buff *skb) 3696 { 3697 const struct skb_shared_info *shinfo = skb_shinfo(skb); 3698 3699 qdisc_skb_cb(skb)->pkt_len = skb->len; 3700 3701 /* To get more precise estimation of bytes sent on wire, 3702 * we add to pkt_len the headers size of all segments 3703 */ 3704 if (shinfo->gso_size && skb_transport_header_was_set(skb)) { 3705 u16 gso_segs = shinfo->gso_segs; 3706 unsigned int hdr_len; 3707 3708 /* mac layer + network layer */ 3709 hdr_len = skb_transport_offset(skb); 3710 3711 /* + transport layer */ 3712 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { 3713 const struct tcphdr *th; 3714 struct tcphdr _tcphdr; 3715 3716 th = skb_header_pointer(skb, hdr_len, 3717 sizeof(_tcphdr), &_tcphdr); 3718 if (likely(th)) 3719 hdr_len += __tcp_hdrlen(th); 3720 } else { 3721 struct udphdr _udphdr; 3722 3723 if (skb_header_pointer(skb, hdr_len, 3724 sizeof(_udphdr), &_udphdr)) 3725 hdr_len += sizeof(struct udphdr); 3726 } 3727 3728 if (shinfo->gso_type & SKB_GSO_DODGY) 3729 gso_segs = DIV_ROUND_UP(skb->len - hdr_len, 3730 shinfo->gso_size); 3731 3732 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; 3733 } 3734 } 3735 3736 static int dev_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *q, 3737 struct sk_buff **to_free, 3738 struct netdev_queue *txq) 3739 { 3740 int rc; 3741 3742 rc = q->enqueue(skb, q, to_free) & NET_XMIT_MASK; 3743 if (rc == NET_XMIT_SUCCESS) 3744 trace_qdisc_enqueue(q, txq, skb); 3745 return rc; 3746 } 3747 3748 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, 3749 struct net_device *dev, 3750 struct netdev_queue *txq) 3751 { 3752 spinlock_t *root_lock = qdisc_lock(q); 3753 struct sk_buff *to_free = NULL; 3754 bool contended; 3755 int rc; 3756 3757 qdisc_calculate_pkt_len(skb, q); 3758 3759 tcf_set_drop_reason(skb, SKB_DROP_REASON_QDISC_DROP); 3760 3761 if (q->flags & TCQ_F_NOLOCK) { 3762 if (q->flags & TCQ_F_CAN_BYPASS && nolock_qdisc_is_empty(q) && 3763 qdisc_run_begin(q)) { 3764 /* Retest nolock_qdisc_is_empty() within the protection 3765 * of q->seqlock to protect from racing with requeuing. 3766 */ 3767 if (unlikely(!nolock_qdisc_is_empty(q))) { 3768 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3769 __qdisc_run(q); 3770 qdisc_run_end(q); 3771 3772 goto no_lock_out; 3773 } 3774 3775 qdisc_bstats_cpu_update(q, skb); 3776 if (sch_direct_xmit(skb, q, dev, txq, NULL, true) && 3777 !nolock_qdisc_is_empty(q)) 3778 __qdisc_run(q); 3779 3780 qdisc_run_end(q); 3781 return NET_XMIT_SUCCESS; 3782 } 3783 3784 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3785 qdisc_run(q); 3786 3787 no_lock_out: 3788 if (unlikely(to_free)) 3789 kfree_skb_list_reason(to_free, 3790 tcf_get_drop_reason(to_free)); 3791 return rc; 3792 } 3793 3794 /* 3795 * Heuristic to force contended enqueues to serialize on a 3796 * separate lock before trying to get qdisc main lock. 3797 * This permits qdisc->running owner to get the lock more 3798 * often and dequeue packets faster. 3799 * On PREEMPT_RT it is possible to preempt the qdisc owner during xmit 3800 * and then other tasks will only enqueue packets. The packets will be 3801 * sent after the qdisc owner is scheduled again. To prevent this 3802 * scenario the task always serialize on the lock. 3803 */ 3804 contended = qdisc_is_running(q) || IS_ENABLED(CONFIG_PREEMPT_RT); 3805 if (unlikely(contended)) 3806 spin_lock(&q->busylock); 3807 3808 spin_lock(root_lock); 3809 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { 3810 __qdisc_drop(skb, &to_free); 3811 rc = NET_XMIT_DROP; 3812 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && 3813 qdisc_run_begin(q)) { 3814 /* 3815 * This is a work-conserving queue; there are no old skbs 3816 * waiting to be sent out; and the qdisc is not running - 3817 * xmit the skb directly. 3818 */ 3819 3820 qdisc_bstats_update(q, skb); 3821 3822 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { 3823 if (unlikely(contended)) { 3824 spin_unlock(&q->busylock); 3825 contended = false; 3826 } 3827 __qdisc_run(q); 3828 } 3829 3830 qdisc_run_end(q); 3831 rc = NET_XMIT_SUCCESS; 3832 } else { 3833 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3834 if (qdisc_run_begin(q)) { 3835 if (unlikely(contended)) { 3836 spin_unlock(&q->busylock); 3837 contended = false; 3838 } 3839 __qdisc_run(q); 3840 qdisc_run_end(q); 3841 } 3842 } 3843 spin_unlock(root_lock); 3844 if (unlikely(to_free)) 3845 kfree_skb_list_reason(to_free, 3846 tcf_get_drop_reason(to_free)); 3847 if (unlikely(contended)) 3848 spin_unlock(&q->busylock); 3849 return rc; 3850 } 3851 3852 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) 3853 static void skb_update_prio(struct sk_buff *skb) 3854 { 3855 const struct netprio_map *map; 3856 const struct sock *sk; 3857 unsigned int prioidx; 3858 3859 if (skb->priority) 3860 return; 3861 map = rcu_dereference_bh(skb->dev->priomap); 3862 if (!map) 3863 return; 3864 sk = skb_to_full_sk(skb); 3865 if (!sk) 3866 return; 3867 3868 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data); 3869 3870 if (prioidx < map->priomap_len) 3871 skb->priority = map->priomap[prioidx]; 3872 } 3873 #else 3874 #define skb_update_prio(skb) 3875 #endif 3876 3877 /** 3878 * dev_loopback_xmit - loop back @skb 3879 * @net: network namespace this loopback is happening in 3880 * @sk: sk needed to be a netfilter okfn 3881 * @skb: buffer to transmit 3882 */ 3883 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) 3884 { 3885 skb_reset_mac_header(skb); 3886 __skb_pull(skb, skb_network_offset(skb)); 3887 skb->pkt_type = PACKET_LOOPBACK; 3888 if (skb->ip_summed == CHECKSUM_NONE) 3889 skb->ip_summed = CHECKSUM_UNNECESSARY; 3890 DEBUG_NET_WARN_ON_ONCE(!skb_dst(skb)); 3891 skb_dst_force(skb); 3892 netif_rx(skb); 3893 return 0; 3894 } 3895 EXPORT_SYMBOL(dev_loopback_xmit); 3896 3897 #ifdef CONFIG_NET_EGRESS 3898 static struct netdev_queue * 3899 netdev_tx_queue_mapping(struct net_device *dev, struct sk_buff *skb) 3900 { 3901 int qm = skb_get_queue_mapping(skb); 3902 3903 return netdev_get_tx_queue(dev, netdev_cap_txqueue(dev, qm)); 3904 } 3905 3906 static bool netdev_xmit_txqueue_skipped(void) 3907 { 3908 return __this_cpu_read(softnet_data.xmit.skip_txqueue); 3909 } 3910 3911 void netdev_xmit_skip_txqueue(bool skip) 3912 { 3913 __this_cpu_write(softnet_data.xmit.skip_txqueue, skip); 3914 } 3915 EXPORT_SYMBOL_GPL(netdev_xmit_skip_txqueue); 3916 #endif /* CONFIG_NET_EGRESS */ 3917 3918 #ifdef CONFIG_NET_XGRESS 3919 static int tc_run(struct tcx_entry *entry, struct sk_buff *skb, 3920 enum skb_drop_reason *drop_reason) 3921 { 3922 int ret = TC_ACT_UNSPEC; 3923 #ifdef CONFIG_NET_CLS_ACT 3924 struct mini_Qdisc *miniq = rcu_dereference_bh(entry->miniq); 3925 struct tcf_result res; 3926 3927 if (!miniq) 3928 return ret; 3929 3930 tc_skb_cb(skb)->mru = 0; 3931 tc_skb_cb(skb)->post_ct = false; 3932 tcf_set_drop_reason(skb, *drop_reason); 3933 3934 mini_qdisc_bstats_cpu_update(miniq, skb); 3935 ret = tcf_classify(skb, miniq->block, miniq->filter_list, &res, false); 3936 /* Only tcf related quirks below. */ 3937 switch (ret) { 3938 case TC_ACT_SHOT: 3939 *drop_reason = tcf_get_drop_reason(skb); 3940 mini_qdisc_qstats_cpu_drop(miniq); 3941 break; 3942 case TC_ACT_OK: 3943 case TC_ACT_RECLASSIFY: 3944 skb->tc_index = TC_H_MIN(res.classid); 3945 break; 3946 } 3947 #endif /* CONFIG_NET_CLS_ACT */ 3948 return ret; 3949 } 3950 3951 static DEFINE_STATIC_KEY_FALSE(tcx_needed_key); 3952 3953 void tcx_inc(void) 3954 { 3955 static_branch_inc(&tcx_needed_key); 3956 } 3957 3958 void tcx_dec(void) 3959 { 3960 static_branch_dec(&tcx_needed_key); 3961 } 3962 3963 static __always_inline enum tcx_action_base 3964 tcx_run(const struct bpf_mprog_entry *entry, struct sk_buff *skb, 3965 const bool needs_mac) 3966 { 3967 const struct bpf_mprog_fp *fp; 3968 const struct bpf_prog *prog; 3969 int ret = TCX_NEXT; 3970 3971 if (needs_mac) 3972 __skb_push(skb, skb->mac_len); 3973 bpf_mprog_foreach_prog(entry, fp, prog) { 3974 bpf_compute_data_pointers(skb); 3975 ret = bpf_prog_run(prog, skb); 3976 if (ret != TCX_NEXT) 3977 break; 3978 } 3979 if (needs_mac) 3980 __skb_pull(skb, skb->mac_len); 3981 return tcx_action_code(skb, ret); 3982 } 3983 3984 static __always_inline struct sk_buff * 3985 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 3986 struct net_device *orig_dev, bool *another) 3987 { 3988 struct bpf_mprog_entry *entry = rcu_dereference_bh(skb->dev->tcx_ingress); 3989 enum skb_drop_reason drop_reason = SKB_DROP_REASON_TC_INGRESS; 3990 int sch_ret; 3991 3992 if (!entry) 3993 return skb; 3994 if (*pt_prev) { 3995 *ret = deliver_skb(skb, *pt_prev, orig_dev); 3996 *pt_prev = NULL; 3997 } 3998 3999 qdisc_skb_cb(skb)->pkt_len = skb->len; 4000 tcx_set_ingress(skb, true); 4001 4002 if (static_branch_unlikely(&tcx_needed_key)) { 4003 sch_ret = tcx_run(entry, skb, true); 4004 if (sch_ret != TC_ACT_UNSPEC) 4005 goto ingress_verdict; 4006 } 4007 sch_ret = tc_run(tcx_entry(entry), skb, &drop_reason); 4008 ingress_verdict: 4009 switch (sch_ret) { 4010 case TC_ACT_REDIRECT: 4011 /* skb_mac_header check was done by BPF, so we can safely 4012 * push the L2 header back before redirecting to another 4013 * netdev. 4014 */ 4015 __skb_push(skb, skb->mac_len); 4016 if (skb_do_redirect(skb) == -EAGAIN) { 4017 __skb_pull(skb, skb->mac_len); 4018 *another = true; 4019 break; 4020 } 4021 *ret = NET_RX_SUCCESS; 4022 return NULL; 4023 case TC_ACT_SHOT: 4024 kfree_skb_reason(skb, drop_reason); 4025 *ret = NET_RX_DROP; 4026 return NULL; 4027 /* used by tc_run */ 4028 case TC_ACT_STOLEN: 4029 case TC_ACT_QUEUED: 4030 case TC_ACT_TRAP: 4031 consume_skb(skb); 4032 fallthrough; 4033 case TC_ACT_CONSUMED: 4034 *ret = NET_RX_SUCCESS; 4035 return NULL; 4036 } 4037 4038 return skb; 4039 } 4040 4041 static __always_inline struct sk_buff * 4042 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 4043 { 4044 struct bpf_mprog_entry *entry = rcu_dereference_bh(dev->tcx_egress); 4045 enum skb_drop_reason drop_reason = SKB_DROP_REASON_TC_EGRESS; 4046 int sch_ret; 4047 4048 if (!entry) 4049 return skb; 4050 4051 /* qdisc_skb_cb(skb)->pkt_len & tcx_set_ingress() was 4052 * already set by the caller. 4053 */ 4054 if (static_branch_unlikely(&tcx_needed_key)) { 4055 sch_ret = tcx_run(entry, skb, false); 4056 if (sch_ret != TC_ACT_UNSPEC) 4057 goto egress_verdict; 4058 } 4059 sch_ret = tc_run(tcx_entry(entry), skb, &drop_reason); 4060 egress_verdict: 4061 switch (sch_ret) { 4062 case TC_ACT_REDIRECT: 4063 /* No need to push/pop skb's mac_header here on egress! */ 4064 skb_do_redirect(skb); 4065 *ret = NET_XMIT_SUCCESS; 4066 return NULL; 4067 case TC_ACT_SHOT: 4068 kfree_skb_reason(skb, drop_reason); 4069 *ret = NET_XMIT_DROP; 4070 return NULL; 4071 /* used by tc_run */ 4072 case TC_ACT_STOLEN: 4073 case TC_ACT_QUEUED: 4074 case TC_ACT_TRAP: 4075 consume_skb(skb); 4076 fallthrough; 4077 case TC_ACT_CONSUMED: 4078 *ret = NET_XMIT_SUCCESS; 4079 return NULL; 4080 } 4081 4082 return skb; 4083 } 4084 #else 4085 static __always_inline struct sk_buff * 4086 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 4087 struct net_device *orig_dev, bool *another) 4088 { 4089 return skb; 4090 } 4091 4092 static __always_inline struct sk_buff * 4093 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 4094 { 4095 return skb; 4096 } 4097 #endif /* CONFIG_NET_XGRESS */ 4098 4099 #ifdef CONFIG_XPS 4100 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb, 4101 struct xps_dev_maps *dev_maps, unsigned int tci) 4102 { 4103 int tc = netdev_get_prio_tc_map(dev, skb->priority); 4104 struct xps_map *map; 4105 int queue_index = -1; 4106 4107 if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids) 4108 return queue_index; 4109 4110 tci *= dev_maps->num_tc; 4111 tci += tc; 4112 4113 map = rcu_dereference(dev_maps->attr_map[tci]); 4114 if (map) { 4115 if (map->len == 1) 4116 queue_index = map->queues[0]; 4117 else 4118 queue_index = map->queues[reciprocal_scale( 4119 skb_get_hash(skb), map->len)]; 4120 if (unlikely(queue_index >= dev->real_num_tx_queues)) 4121 queue_index = -1; 4122 } 4123 return queue_index; 4124 } 4125 #endif 4126 4127 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev, 4128 struct sk_buff *skb) 4129 { 4130 #ifdef CONFIG_XPS 4131 struct xps_dev_maps *dev_maps; 4132 struct sock *sk = skb->sk; 4133 int queue_index = -1; 4134 4135 if (!static_key_false(&xps_needed)) 4136 return -1; 4137 4138 rcu_read_lock(); 4139 if (!static_key_false(&xps_rxqs_needed)) 4140 goto get_cpus_map; 4141 4142 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]); 4143 if (dev_maps) { 4144 int tci = sk_rx_queue_get(sk); 4145 4146 if (tci >= 0) 4147 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4148 tci); 4149 } 4150 4151 get_cpus_map: 4152 if (queue_index < 0) { 4153 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]); 4154 if (dev_maps) { 4155 unsigned int tci = skb->sender_cpu - 1; 4156 4157 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4158 tci); 4159 } 4160 } 4161 rcu_read_unlock(); 4162 4163 return queue_index; 4164 #else 4165 return -1; 4166 #endif 4167 } 4168 4169 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb, 4170 struct net_device *sb_dev) 4171 { 4172 return 0; 4173 } 4174 EXPORT_SYMBOL(dev_pick_tx_zero); 4175 4176 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb, 4177 struct net_device *sb_dev) 4178 { 4179 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues; 4180 } 4181 EXPORT_SYMBOL(dev_pick_tx_cpu_id); 4182 4183 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb, 4184 struct net_device *sb_dev) 4185 { 4186 struct sock *sk = skb->sk; 4187 int queue_index = sk_tx_queue_get(sk); 4188 4189 sb_dev = sb_dev ? : dev; 4190 4191 if (queue_index < 0 || skb->ooo_okay || 4192 queue_index >= dev->real_num_tx_queues) { 4193 int new_index = get_xps_queue(dev, sb_dev, skb); 4194 4195 if (new_index < 0) 4196 new_index = skb_tx_hash(dev, sb_dev, skb); 4197 4198 if (queue_index != new_index && sk && 4199 sk_fullsock(sk) && 4200 rcu_access_pointer(sk->sk_dst_cache)) 4201 sk_tx_queue_set(sk, new_index); 4202 4203 queue_index = new_index; 4204 } 4205 4206 return queue_index; 4207 } 4208 EXPORT_SYMBOL(netdev_pick_tx); 4209 4210 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev, 4211 struct sk_buff *skb, 4212 struct net_device *sb_dev) 4213 { 4214 int queue_index = 0; 4215 4216 #ifdef CONFIG_XPS 4217 u32 sender_cpu = skb->sender_cpu - 1; 4218 4219 if (sender_cpu >= (u32)NR_CPUS) 4220 skb->sender_cpu = raw_smp_processor_id() + 1; 4221 #endif 4222 4223 if (dev->real_num_tx_queues != 1) { 4224 const struct net_device_ops *ops = dev->netdev_ops; 4225 4226 if (ops->ndo_select_queue) 4227 queue_index = ops->ndo_select_queue(dev, skb, sb_dev); 4228 else 4229 queue_index = netdev_pick_tx(dev, skb, sb_dev); 4230 4231 queue_index = netdev_cap_txqueue(dev, queue_index); 4232 } 4233 4234 skb_set_queue_mapping(skb, queue_index); 4235 return netdev_get_tx_queue(dev, queue_index); 4236 } 4237 4238 /** 4239 * __dev_queue_xmit() - transmit a buffer 4240 * @skb: buffer to transmit 4241 * @sb_dev: suboordinate device used for L2 forwarding offload 4242 * 4243 * Queue a buffer for transmission to a network device. The caller must 4244 * have set the device and priority and built the buffer before calling 4245 * this function. The function can be called from an interrupt. 4246 * 4247 * When calling this method, interrupts MUST be enabled. This is because 4248 * the BH enable code must have IRQs enabled so that it will not deadlock. 4249 * 4250 * Regardless of the return value, the skb is consumed, so it is currently 4251 * difficult to retry a send to this method. (You can bump the ref count 4252 * before sending to hold a reference for retry if you are careful.) 4253 * 4254 * Return: 4255 * * 0 - buffer successfully transmitted 4256 * * positive qdisc return code - NET_XMIT_DROP etc. 4257 * * negative errno - other errors 4258 */ 4259 int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev) 4260 { 4261 struct net_device *dev = skb->dev; 4262 struct netdev_queue *txq = NULL; 4263 struct Qdisc *q; 4264 int rc = -ENOMEM; 4265 bool again = false; 4266 4267 skb_reset_mac_header(skb); 4268 skb_assert_len(skb); 4269 4270 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) 4271 __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED); 4272 4273 /* Disable soft irqs for various locks below. Also 4274 * stops preemption for RCU. 4275 */ 4276 rcu_read_lock_bh(); 4277 4278 skb_update_prio(skb); 4279 4280 qdisc_pkt_len_init(skb); 4281 tcx_set_ingress(skb, false); 4282 #ifdef CONFIG_NET_EGRESS 4283 if (static_branch_unlikely(&egress_needed_key)) { 4284 if (nf_hook_egress_active()) { 4285 skb = nf_hook_egress(skb, &rc, dev); 4286 if (!skb) 4287 goto out; 4288 } 4289 4290 netdev_xmit_skip_txqueue(false); 4291 4292 nf_skip_egress(skb, true); 4293 skb = sch_handle_egress(skb, &rc, dev); 4294 if (!skb) 4295 goto out; 4296 nf_skip_egress(skb, false); 4297 4298 if (netdev_xmit_txqueue_skipped()) 4299 txq = netdev_tx_queue_mapping(dev, skb); 4300 } 4301 #endif 4302 /* If device/qdisc don't need skb->dst, release it right now while 4303 * its hot in this cpu cache. 4304 */ 4305 if (dev->priv_flags & IFF_XMIT_DST_RELEASE) 4306 skb_dst_drop(skb); 4307 else 4308 skb_dst_force(skb); 4309 4310 if (!txq) 4311 txq = netdev_core_pick_tx(dev, skb, sb_dev); 4312 4313 q = rcu_dereference_bh(txq->qdisc); 4314 4315 trace_net_dev_queue(skb); 4316 if (q->enqueue) { 4317 rc = __dev_xmit_skb(skb, q, dev, txq); 4318 goto out; 4319 } 4320 4321 /* The device has no queue. Common case for software devices: 4322 * loopback, all the sorts of tunnels... 4323 4324 * Really, it is unlikely that netif_tx_lock protection is necessary 4325 * here. (f.e. loopback and IP tunnels are clean ignoring statistics 4326 * counters.) 4327 * However, it is possible, that they rely on protection 4328 * made by us here. 4329 4330 * Check this and shot the lock. It is not prone from deadlocks. 4331 *Either shot noqueue qdisc, it is even simpler 8) 4332 */ 4333 if (dev->flags & IFF_UP) { 4334 int cpu = smp_processor_id(); /* ok because BHs are off */ 4335 4336 /* Other cpus might concurrently change txq->xmit_lock_owner 4337 * to -1 or to their cpu id, but not to our id. 4338 */ 4339 if (READ_ONCE(txq->xmit_lock_owner) != cpu) { 4340 if (dev_xmit_recursion()) 4341 goto recursion_alert; 4342 4343 skb = validate_xmit_skb(skb, dev, &again); 4344 if (!skb) 4345 goto out; 4346 4347 HARD_TX_LOCK(dev, txq, cpu); 4348 4349 if (!netif_xmit_stopped(txq)) { 4350 dev_xmit_recursion_inc(); 4351 skb = dev_hard_start_xmit(skb, dev, txq, &rc); 4352 dev_xmit_recursion_dec(); 4353 if (dev_xmit_complete(rc)) { 4354 HARD_TX_UNLOCK(dev, txq); 4355 goto out; 4356 } 4357 } 4358 HARD_TX_UNLOCK(dev, txq); 4359 net_crit_ratelimited("Virtual device %s asks to queue packet!\n", 4360 dev->name); 4361 } else { 4362 /* Recursion is detected! It is possible, 4363 * unfortunately 4364 */ 4365 recursion_alert: 4366 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", 4367 dev->name); 4368 } 4369 } 4370 4371 rc = -ENETDOWN; 4372 rcu_read_unlock_bh(); 4373 4374 dev_core_stats_tx_dropped_inc(dev); 4375 kfree_skb_list(skb); 4376 return rc; 4377 out: 4378 rcu_read_unlock_bh(); 4379 return rc; 4380 } 4381 EXPORT_SYMBOL(__dev_queue_xmit); 4382 4383 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id) 4384 { 4385 struct net_device *dev = skb->dev; 4386 struct sk_buff *orig_skb = skb; 4387 struct netdev_queue *txq; 4388 int ret = NETDEV_TX_BUSY; 4389 bool again = false; 4390 4391 if (unlikely(!netif_running(dev) || 4392 !netif_carrier_ok(dev))) 4393 goto drop; 4394 4395 skb = validate_xmit_skb_list(skb, dev, &again); 4396 if (skb != orig_skb) 4397 goto drop; 4398 4399 skb_set_queue_mapping(skb, queue_id); 4400 txq = skb_get_tx_queue(dev, skb); 4401 4402 local_bh_disable(); 4403 4404 dev_xmit_recursion_inc(); 4405 HARD_TX_LOCK(dev, txq, smp_processor_id()); 4406 if (!netif_xmit_frozen_or_drv_stopped(txq)) 4407 ret = netdev_start_xmit(skb, dev, txq, false); 4408 HARD_TX_UNLOCK(dev, txq); 4409 dev_xmit_recursion_dec(); 4410 4411 local_bh_enable(); 4412 return ret; 4413 drop: 4414 dev_core_stats_tx_dropped_inc(dev); 4415 kfree_skb_list(skb); 4416 return NET_XMIT_DROP; 4417 } 4418 EXPORT_SYMBOL(__dev_direct_xmit); 4419 4420 /************************************************************************* 4421 * Receiver routines 4422 *************************************************************************/ 4423 4424 int netdev_max_backlog __read_mostly = 1000; 4425 EXPORT_SYMBOL(netdev_max_backlog); 4426 4427 int netdev_tstamp_prequeue __read_mostly = 1; 4428 unsigned int sysctl_skb_defer_max __read_mostly = 64; 4429 int netdev_budget __read_mostly = 300; 4430 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */ 4431 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ; 4432 int weight_p __read_mostly = 64; /* old backlog weight */ 4433 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ 4434 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ 4435 int dev_rx_weight __read_mostly = 64; 4436 int dev_tx_weight __read_mostly = 64; 4437 4438 /* Called with irq disabled */ 4439 static inline void ____napi_schedule(struct softnet_data *sd, 4440 struct napi_struct *napi) 4441 { 4442 struct task_struct *thread; 4443 4444 lockdep_assert_irqs_disabled(); 4445 4446 if (test_bit(NAPI_STATE_THREADED, &napi->state)) { 4447 /* Paired with smp_mb__before_atomic() in 4448 * napi_enable()/dev_set_threaded(). 4449 * Use READ_ONCE() to guarantee a complete 4450 * read on napi->thread. Only call 4451 * wake_up_process() when it's not NULL. 4452 */ 4453 thread = READ_ONCE(napi->thread); 4454 if (thread) { 4455 /* Avoid doing set_bit() if the thread is in 4456 * INTERRUPTIBLE state, cause napi_thread_wait() 4457 * makes sure to proceed with napi polling 4458 * if the thread is explicitly woken from here. 4459 */ 4460 if (READ_ONCE(thread->__state) != TASK_INTERRUPTIBLE) 4461 set_bit(NAPI_STATE_SCHED_THREADED, &napi->state); 4462 wake_up_process(thread); 4463 return; 4464 } 4465 } 4466 4467 list_add_tail(&napi->poll_list, &sd->poll_list); 4468 WRITE_ONCE(napi->list_owner, smp_processor_id()); 4469 /* If not called from net_rx_action() 4470 * we have to raise NET_RX_SOFTIRQ. 4471 */ 4472 if (!sd->in_net_rx_action) 4473 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4474 } 4475 4476 #ifdef CONFIG_RPS 4477 4478 /* One global table that all flow-based protocols share. */ 4479 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; 4480 EXPORT_SYMBOL(rps_sock_flow_table); 4481 u32 rps_cpu_mask __read_mostly; 4482 EXPORT_SYMBOL(rps_cpu_mask); 4483 4484 struct static_key_false rps_needed __read_mostly; 4485 EXPORT_SYMBOL(rps_needed); 4486 struct static_key_false rfs_needed __read_mostly; 4487 EXPORT_SYMBOL(rfs_needed); 4488 4489 static struct rps_dev_flow * 4490 set_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4491 struct rps_dev_flow *rflow, u16 next_cpu) 4492 { 4493 if (next_cpu < nr_cpu_ids) { 4494 #ifdef CONFIG_RFS_ACCEL 4495 struct netdev_rx_queue *rxqueue; 4496 struct rps_dev_flow_table *flow_table; 4497 struct rps_dev_flow *old_rflow; 4498 u32 flow_id; 4499 u16 rxq_index; 4500 int rc; 4501 4502 /* Should we steer this flow to a different hardware queue? */ 4503 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || 4504 !(dev->features & NETIF_F_NTUPLE)) 4505 goto out; 4506 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); 4507 if (rxq_index == skb_get_rx_queue(skb)) 4508 goto out; 4509 4510 rxqueue = dev->_rx + rxq_index; 4511 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4512 if (!flow_table) 4513 goto out; 4514 flow_id = skb_get_hash(skb) & flow_table->mask; 4515 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, 4516 rxq_index, flow_id); 4517 if (rc < 0) 4518 goto out; 4519 old_rflow = rflow; 4520 rflow = &flow_table->flows[flow_id]; 4521 rflow->filter = rc; 4522 if (old_rflow->filter == rflow->filter) 4523 old_rflow->filter = RPS_NO_FILTER; 4524 out: 4525 #endif 4526 rflow->last_qtail = 4527 per_cpu(softnet_data, next_cpu).input_queue_head; 4528 } 4529 4530 rflow->cpu = next_cpu; 4531 return rflow; 4532 } 4533 4534 /* 4535 * get_rps_cpu is called from netif_receive_skb and returns the target 4536 * CPU from the RPS map of the receiving queue for a given skb. 4537 * rcu_read_lock must be held on entry. 4538 */ 4539 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4540 struct rps_dev_flow **rflowp) 4541 { 4542 const struct rps_sock_flow_table *sock_flow_table; 4543 struct netdev_rx_queue *rxqueue = dev->_rx; 4544 struct rps_dev_flow_table *flow_table; 4545 struct rps_map *map; 4546 int cpu = -1; 4547 u32 tcpu; 4548 u32 hash; 4549 4550 if (skb_rx_queue_recorded(skb)) { 4551 u16 index = skb_get_rx_queue(skb); 4552 4553 if (unlikely(index >= dev->real_num_rx_queues)) { 4554 WARN_ONCE(dev->real_num_rx_queues > 1, 4555 "%s received packet on queue %u, but number " 4556 "of RX queues is %u\n", 4557 dev->name, index, dev->real_num_rx_queues); 4558 goto done; 4559 } 4560 rxqueue += index; 4561 } 4562 4563 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ 4564 4565 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4566 map = rcu_dereference(rxqueue->rps_map); 4567 if (!flow_table && !map) 4568 goto done; 4569 4570 skb_reset_network_header(skb); 4571 hash = skb_get_hash(skb); 4572 if (!hash) 4573 goto done; 4574 4575 sock_flow_table = rcu_dereference(rps_sock_flow_table); 4576 if (flow_table && sock_flow_table) { 4577 struct rps_dev_flow *rflow; 4578 u32 next_cpu; 4579 u32 ident; 4580 4581 /* First check into global flow table if there is a match. 4582 * This READ_ONCE() pairs with WRITE_ONCE() from rps_record_sock_flow(). 4583 */ 4584 ident = READ_ONCE(sock_flow_table->ents[hash & sock_flow_table->mask]); 4585 if ((ident ^ hash) & ~rps_cpu_mask) 4586 goto try_rps; 4587 4588 next_cpu = ident & rps_cpu_mask; 4589 4590 /* OK, now we know there is a match, 4591 * we can look at the local (per receive queue) flow table 4592 */ 4593 rflow = &flow_table->flows[hash & flow_table->mask]; 4594 tcpu = rflow->cpu; 4595 4596 /* 4597 * If the desired CPU (where last recvmsg was done) is 4598 * different from current CPU (one in the rx-queue flow 4599 * table entry), switch if one of the following holds: 4600 * - Current CPU is unset (>= nr_cpu_ids). 4601 * - Current CPU is offline. 4602 * - The current CPU's queue tail has advanced beyond the 4603 * last packet that was enqueued using this table entry. 4604 * This guarantees that all previous packets for the flow 4605 * have been dequeued, thus preserving in order delivery. 4606 */ 4607 if (unlikely(tcpu != next_cpu) && 4608 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || 4609 ((int)(per_cpu(softnet_data, tcpu).input_queue_head - 4610 rflow->last_qtail)) >= 0)) { 4611 tcpu = next_cpu; 4612 rflow = set_rps_cpu(dev, skb, rflow, next_cpu); 4613 } 4614 4615 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { 4616 *rflowp = rflow; 4617 cpu = tcpu; 4618 goto done; 4619 } 4620 } 4621 4622 try_rps: 4623 4624 if (map) { 4625 tcpu = map->cpus[reciprocal_scale(hash, map->len)]; 4626 if (cpu_online(tcpu)) { 4627 cpu = tcpu; 4628 goto done; 4629 } 4630 } 4631 4632 done: 4633 return cpu; 4634 } 4635 4636 #ifdef CONFIG_RFS_ACCEL 4637 4638 /** 4639 * rps_may_expire_flow - check whether an RFS hardware filter may be removed 4640 * @dev: Device on which the filter was set 4641 * @rxq_index: RX queue index 4642 * @flow_id: Flow ID passed to ndo_rx_flow_steer() 4643 * @filter_id: Filter ID returned by ndo_rx_flow_steer() 4644 * 4645 * Drivers that implement ndo_rx_flow_steer() should periodically call 4646 * this function for each installed filter and remove the filters for 4647 * which it returns %true. 4648 */ 4649 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, 4650 u32 flow_id, u16 filter_id) 4651 { 4652 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; 4653 struct rps_dev_flow_table *flow_table; 4654 struct rps_dev_flow *rflow; 4655 bool expire = true; 4656 unsigned int cpu; 4657 4658 rcu_read_lock(); 4659 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4660 if (flow_table && flow_id <= flow_table->mask) { 4661 rflow = &flow_table->flows[flow_id]; 4662 cpu = READ_ONCE(rflow->cpu); 4663 if (rflow->filter == filter_id && cpu < nr_cpu_ids && 4664 ((int)(per_cpu(softnet_data, cpu).input_queue_head - 4665 rflow->last_qtail) < 4666 (int)(10 * flow_table->mask))) 4667 expire = false; 4668 } 4669 rcu_read_unlock(); 4670 return expire; 4671 } 4672 EXPORT_SYMBOL(rps_may_expire_flow); 4673 4674 #endif /* CONFIG_RFS_ACCEL */ 4675 4676 /* Called from hardirq (IPI) context */ 4677 static void rps_trigger_softirq(void *data) 4678 { 4679 struct softnet_data *sd = data; 4680 4681 ____napi_schedule(sd, &sd->backlog); 4682 sd->received_rps++; 4683 } 4684 4685 #endif /* CONFIG_RPS */ 4686 4687 /* Called from hardirq (IPI) context */ 4688 static void trigger_rx_softirq(void *data) 4689 { 4690 struct softnet_data *sd = data; 4691 4692 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4693 smp_store_release(&sd->defer_ipi_scheduled, 0); 4694 } 4695 4696 /* 4697 * After we queued a packet into sd->input_pkt_queue, 4698 * we need to make sure this queue is serviced soon. 4699 * 4700 * - If this is another cpu queue, link it to our rps_ipi_list, 4701 * and make sure we will process rps_ipi_list from net_rx_action(). 4702 * 4703 * - If this is our own queue, NAPI schedule our backlog. 4704 * Note that this also raises NET_RX_SOFTIRQ. 4705 */ 4706 static void napi_schedule_rps(struct softnet_data *sd) 4707 { 4708 struct softnet_data *mysd = this_cpu_ptr(&softnet_data); 4709 4710 #ifdef CONFIG_RPS 4711 if (sd != mysd) { 4712 sd->rps_ipi_next = mysd->rps_ipi_list; 4713 mysd->rps_ipi_list = sd; 4714 4715 /* If not called from net_rx_action() or napi_threaded_poll() 4716 * we have to raise NET_RX_SOFTIRQ. 4717 */ 4718 if (!mysd->in_net_rx_action && !mysd->in_napi_threaded_poll) 4719 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4720 return; 4721 } 4722 #endif /* CONFIG_RPS */ 4723 __napi_schedule_irqoff(&mysd->backlog); 4724 } 4725 4726 #ifdef CONFIG_NET_FLOW_LIMIT 4727 int netdev_flow_limit_table_len __read_mostly = (1 << 12); 4728 #endif 4729 4730 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) 4731 { 4732 #ifdef CONFIG_NET_FLOW_LIMIT 4733 struct sd_flow_limit *fl; 4734 struct softnet_data *sd; 4735 unsigned int old_flow, new_flow; 4736 4737 if (qlen < (READ_ONCE(netdev_max_backlog) >> 1)) 4738 return false; 4739 4740 sd = this_cpu_ptr(&softnet_data); 4741 4742 rcu_read_lock(); 4743 fl = rcu_dereference(sd->flow_limit); 4744 if (fl) { 4745 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); 4746 old_flow = fl->history[fl->history_head]; 4747 fl->history[fl->history_head] = new_flow; 4748 4749 fl->history_head++; 4750 fl->history_head &= FLOW_LIMIT_HISTORY - 1; 4751 4752 if (likely(fl->buckets[old_flow])) 4753 fl->buckets[old_flow]--; 4754 4755 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { 4756 fl->count++; 4757 rcu_read_unlock(); 4758 return true; 4759 } 4760 } 4761 rcu_read_unlock(); 4762 #endif 4763 return false; 4764 } 4765 4766 /* 4767 * enqueue_to_backlog is called to queue an skb to a per CPU backlog 4768 * queue (may be a remote CPU queue). 4769 */ 4770 static int enqueue_to_backlog(struct sk_buff *skb, int cpu, 4771 unsigned int *qtail) 4772 { 4773 enum skb_drop_reason reason; 4774 struct softnet_data *sd; 4775 unsigned long flags; 4776 unsigned int qlen; 4777 4778 reason = SKB_DROP_REASON_NOT_SPECIFIED; 4779 sd = &per_cpu(softnet_data, cpu); 4780 4781 rps_lock_irqsave(sd, &flags); 4782 if (!netif_running(skb->dev)) 4783 goto drop; 4784 qlen = skb_queue_len(&sd->input_pkt_queue); 4785 if (qlen <= READ_ONCE(netdev_max_backlog) && !skb_flow_limit(skb, qlen)) { 4786 if (qlen) { 4787 enqueue: 4788 __skb_queue_tail(&sd->input_pkt_queue, skb); 4789 input_queue_tail_incr_save(sd, qtail); 4790 rps_unlock_irq_restore(sd, &flags); 4791 return NET_RX_SUCCESS; 4792 } 4793 4794 /* Schedule NAPI for backlog device 4795 * We can use non atomic operation since we own the queue lock 4796 */ 4797 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) 4798 napi_schedule_rps(sd); 4799 goto enqueue; 4800 } 4801 reason = SKB_DROP_REASON_CPU_BACKLOG; 4802 4803 drop: 4804 sd->dropped++; 4805 rps_unlock_irq_restore(sd, &flags); 4806 4807 dev_core_stats_rx_dropped_inc(skb->dev); 4808 kfree_skb_reason(skb, reason); 4809 return NET_RX_DROP; 4810 } 4811 4812 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb) 4813 { 4814 struct net_device *dev = skb->dev; 4815 struct netdev_rx_queue *rxqueue; 4816 4817 rxqueue = dev->_rx; 4818 4819 if (skb_rx_queue_recorded(skb)) { 4820 u16 index = skb_get_rx_queue(skb); 4821 4822 if (unlikely(index >= dev->real_num_rx_queues)) { 4823 WARN_ONCE(dev->real_num_rx_queues > 1, 4824 "%s received packet on queue %u, but number " 4825 "of RX queues is %u\n", 4826 dev->name, index, dev->real_num_rx_queues); 4827 4828 return rxqueue; /* Return first rxqueue */ 4829 } 4830 rxqueue += index; 4831 } 4832 return rxqueue; 4833 } 4834 4835 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp, 4836 struct bpf_prog *xdp_prog) 4837 { 4838 void *orig_data, *orig_data_end, *hard_start; 4839 struct netdev_rx_queue *rxqueue; 4840 bool orig_bcast, orig_host; 4841 u32 mac_len, frame_sz; 4842 __be16 orig_eth_type; 4843 struct ethhdr *eth; 4844 u32 metalen, act; 4845 int off; 4846 4847 /* The XDP program wants to see the packet starting at the MAC 4848 * header. 4849 */ 4850 mac_len = skb->data - skb_mac_header(skb); 4851 hard_start = skb->data - skb_headroom(skb); 4852 4853 /* SKB "head" area always have tailroom for skb_shared_info */ 4854 frame_sz = (void *)skb_end_pointer(skb) - hard_start; 4855 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 4856 4857 rxqueue = netif_get_rxqueue(skb); 4858 xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq); 4859 xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len, 4860 skb_headlen(skb) + mac_len, true); 4861 4862 orig_data_end = xdp->data_end; 4863 orig_data = xdp->data; 4864 eth = (struct ethhdr *)xdp->data; 4865 orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr); 4866 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest); 4867 orig_eth_type = eth->h_proto; 4868 4869 act = bpf_prog_run_xdp(xdp_prog, xdp); 4870 4871 /* check if bpf_xdp_adjust_head was used */ 4872 off = xdp->data - orig_data; 4873 if (off) { 4874 if (off > 0) 4875 __skb_pull(skb, off); 4876 else if (off < 0) 4877 __skb_push(skb, -off); 4878 4879 skb->mac_header += off; 4880 skb_reset_network_header(skb); 4881 } 4882 4883 /* check if bpf_xdp_adjust_tail was used */ 4884 off = xdp->data_end - orig_data_end; 4885 if (off != 0) { 4886 skb_set_tail_pointer(skb, xdp->data_end - xdp->data); 4887 skb->len += off; /* positive on grow, negative on shrink */ 4888 } 4889 4890 /* check if XDP changed eth hdr such SKB needs update */ 4891 eth = (struct ethhdr *)xdp->data; 4892 if ((orig_eth_type != eth->h_proto) || 4893 (orig_host != ether_addr_equal_64bits(eth->h_dest, 4894 skb->dev->dev_addr)) || 4895 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) { 4896 __skb_push(skb, ETH_HLEN); 4897 skb->pkt_type = PACKET_HOST; 4898 skb->protocol = eth_type_trans(skb, skb->dev); 4899 } 4900 4901 /* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull 4902 * before calling us again on redirect path. We do not call do_redirect 4903 * as we leave that up to the caller. 4904 * 4905 * Caller is responsible for managing lifetime of skb (i.e. calling 4906 * kfree_skb in response to actions it cannot handle/XDP_DROP). 4907 */ 4908 switch (act) { 4909 case XDP_REDIRECT: 4910 case XDP_TX: 4911 __skb_push(skb, mac_len); 4912 break; 4913 case XDP_PASS: 4914 metalen = xdp->data - xdp->data_meta; 4915 if (metalen) 4916 skb_metadata_set(skb, metalen); 4917 break; 4918 } 4919 4920 return act; 4921 } 4922 4923 static u32 netif_receive_generic_xdp(struct sk_buff *skb, 4924 struct xdp_buff *xdp, 4925 struct bpf_prog *xdp_prog) 4926 { 4927 u32 act = XDP_DROP; 4928 4929 /* Reinjected packets coming from act_mirred or similar should 4930 * not get XDP generic processing. 4931 */ 4932 if (skb_is_redirected(skb)) 4933 return XDP_PASS; 4934 4935 /* XDP packets must be linear and must have sufficient headroom 4936 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also 4937 * native XDP provides, thus we need to do it here as well. 4938 */ 4939 if (skb_cloned(skb) || skb_is_nonlinear(skb) || 4940 skb_headroom(skb) < XDP_PACKET_HEADROOM) { 4941 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb); 4942 int troom = skb->tail + skb->data_len - skb->end; 4943 4944 /* In case we have to go down the path and also linearize, 4945 * then lets do the pskb_expand_head() work just once here. 4946 */ 4947 if (pskb_expand_head(skb, 4948 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0, 4949 troom > 0 ? troom + 128 : 0, GFP_ATOMIC)) 4950 goto do_drop; 4951 if (skb_linearize(skb)) 4952 goto do_drop; 4953 } 4954 4955 act = bpf_prog_run_generic_xdp(skb, xdp, xdp_prog); 4956 switch (act) { 4957 case XDP_REDIRECT: 4958 case XDP_TX: 4959 case XDP_PASS: 4960 break; 4961 default: 4962 bpf_warn_invalid_xdp_action(skb->dev, xdp_prog, act); 4963 fallthrough; 4964 case XDP_ABORTED: 4965 trace_xdp_exception(skb->dev, xdp_prog, act); 4966 fallthrough; 4967 case XDP_DROP: 4968 do_drop: 4969 kfree_skb(skb); 4970 break; 4971 } 4972 4973 return act; 4974 } 4975 4976 /* When doing generic XDP we have to bypass the qdisc layer and the 4977 * network taps in order to match in-driver-XDP behavior. This also means 4978 * that XDP packets are able to starve other packets going through a qdisc, 4979 * and DDOS attacks will be more effective. In-driver-XDP use dedicated TX 4980 * queues, so they do not have this starvation issue. 4981 */ 4982 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog) 4983 { 4984 struct net_device *dev = skb->dev; 4985 struct netdev_queue *txq; 4986 bool free_skb = true; 4987 int cpu, rc; 4988 4989 txq = netdev_core_pick_tx(dev, skb, NULL); 4990 cpu = smp_processor_id(); 4991 HARD_TX_LOCK(dev, txq, cpu); 4992 if (!netif_xmit_frozen_or_drv_stopped(txq)) { 4993 rc = netdev_start_xmit(skb, dev, txq, 0); 4994 if (dev_xmit_complete(rc)) 4995 free_skb = false; 4996 } 4997 HARD_TX_UNLOCK(dev, txq); 4998 if (free_skb) { 4999 trace_xdp_exception(dev, xdp_prog, XDP_TX); 5000 dev_core_stats_tx_dropped_inc(dev); 5001 kfree_skb(skb); 5002 } 5003 } 5004 5005 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key); 5006 5007 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb) 5008 { 5009 if (xdp_prog) { 5010 struct xdp_buff xdp; 5011 u32 act; 5012 int err; 5013 5014 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog); 5015 if (act != XDP_PASS) { 5016 switch (act) { 5017 case XDP_REDIRECT: 5018 err = xdp_do_generic_redirect(skb->dev, skb, 5019 &xdp, xdp_prog); 5020 if (err) 5021 goto out_redir; 5022 break; 5023 case XDP_TX: 5024 generic_xdp_tx(skb, xdp_prog); 5025 break; 5026 } 5027 return XDP_DROP; 5028 } 5029 } 5030 return XDP_PASS; 5031 out_redir: 5032 kfree_skb_reason(skb, SKB_DROP_REASON_XDP); 5033 return XDP_DROP; 5034 } 5035 EXPORT_SYMBOL_GPL(do_xdp_generic); 5036 5037 static int netif_rx_internal(struct sk_buff *skb) 5038 { 5039 int ret; 5040 5041 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb); 5042 5043 trace_netif_rx(skb); 5044 5045 #ifdef CONFIG_RPS 5046 if (static_branch_unlikely(&rps_needed)) { 5047 struct rps_dev_flow voidflow, *rflow = &voidflow; 5048 int cpu; 5049 5050 rcu_read_lock(); 5051 5052 cpu = get_rps_cpu(skb->dev, skb, &rflow); 5053 if (cpu < 0) 5054 cpu = smp_processor_id(); 5055 5056 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5057 5058 rcu_read_unlock(); 5059 } else 5060 #endif 5061 { 5062 unsigned int qtail; 5063 5064 ret = enqueue_to_backlog(skb, smp_processor_id(), &qtail); 5065 } 5066 return ret; 5067 } 5068 5069 /** 5070 * __netif_rx - Slightly optimized version of netif_rx 5071 * @skb: buffer to post 5072 * 5073 * This behaves as netif_rx except that it does not disable bottom halves. 5074 * As a result this function may only be invoked from the interrupt context 5075 * (either hard or soft interrupt). 5076 */ 5077 int __netif_rx(struct sk_buff *skb) 5078 { 5079 int ret; 5080 5081 lockdep_assert_once(hardirq_count() | softirq_count()); 5082 5083 trace_netif_rx_entry(skb); 5084 ret = netif_rx_internal(skb); 5085 trace_netif_rx_exit(ret); 5086 return ret; 5087 } 5088 EXPORT_SYMBOL(__netif_rx); 5089 5090 /** 5091 * netif_rx - post buffer to the network code 5092 * @skb: buffer to post 5093 * 5094 * This function receives a packet from a device driver and queues it for 5095 * the upper (protocol) levels to process via the backlog NAPI device. It 5096 * always succeeds. The buffer may be dropped during processing for 5097 * congestion control or by the protocol layers. 5098 * The network buffer is passed via the backlog NAPI device. Modern NIC 5099 * driver should use NAPI and GRO. 5100 * This function can used from interrupt and from process context. The 5101 * caller from process context must not disable interrupts before invoking 5102 * this function. 5103 * 5104 * return values: 5105 * NET_RX_SUCCESS (no congestion) 5106 * NET_RX_DROP (packet was dropped) 5107 * 5108 */ 5109 int netif_rx(struct sk_buff *skb) 5110 { 5111 bool need_bh_off = !(hardirq_count() | softirq_count()); 5112 int ret; 5113 5114 if (need_bh_off) 5115 local_bh_disable(); 5116 trace_netif_rx_entry(skb); 5117 ret = netif_rx_internal(skb); 5118 trace_netif_rx_exit(ret); 5119 if (need_bh_off) 5120 local_bh_enable(); 5121 return ret; 5122 } 5123 EXPORT_SYMBOL(netif_rx); 5124 5125 static __latent_entropy void net_tx_action(struct softirq_action *h) 5126 { 5127 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 5128 5129 if (sd->completion_queue) { 5130 struct sk_buff *clist; 5131 5132 local_irq_disable(); 5133 clist = sd->completion_queue; 5134 sd->completion_queue = NULL; 5135 local_irq_enable(); 5136 5137 while (clist) { 5138 struct sk_buff *skb = clist; 5139 5140 clist = clist->next; 5141 5142 WARN_ON(refcount_read(&skb->users)); 5143 if (likely(get_kfree_skb_cb(skb)->reason == SKB_CONSUMED)) 5144 trace_consume_skb(skb, net_tx_action); 5145 else 5146 trace_kfree_skb(skb, net_tx_action, 5147 get_kfree_skb_cb(skb)->reason); 5148 5149 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) 5150 __kfree_skb(skb); 5151 else 5152 __napi_kfree_skb(skb, 5153 get_kfree_skb_cb(skb)->reason); 5154 } 5155 } 5156 5157 if (sd->output_queue) { 5158 struct Qdisc *head; 5159 5160 local_irq_disable(); 5161 head = sd->output_queue; 5162 sd->output_queue = NULL; 5163 sd->output_queue_tailp = &sd->output_queue; 5164 local_irq_enable(); 5165 5166 rcu_read_lock(); 5167 5168 while (head) { 5169 struct Qdisc *q = head; 5170 spinlock_t *root_lock = NULL; 5171 5172 head = head->next_sched; 5173 5174 /* We need to make sure head->next_sched is read 5175 * before clearing __QDISC_STATE_SCHED 5176 */ 5177 smp_mb__before_atomic(); 5178 5179 if (!(q->flags & TCQ_F_NOLOCK)) { 5180 root_lock = qdisc_lock(q); 5181 spin_lock(root_lock); 5182 } else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, 5183 &q->state))) { 5184 /* There is a synchronize_net() between 5185 * STATE_DEACTIVATED flag being set and 5186 * qdisc_reset()/some_qdisc_is_busy() in 5187 * dev_deactivate(), so we can safely bail out 5188 * early here to avoid data race between 5189 * qdisc_deactivate() and some_qdisc_is_busy() 5190 * for lockless qdisc. 5191 */ 5192 clear_bit(__QDISC_STATE_SCHED, &q->state); 5193 continue; 5194 } 5195 5196 clear_bit(__QDISC_STATE_SCHED, &q->state); 5197 qdisc_run(q); 5198 if (root_lock) 5199 spin_unlock(root_lock); 5200 } 5201 5202 rcu_read_unlock(); 5203 } 5204 5205 xfrm_dev_backlog(sd); 5206 } 5207 5208 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) 5209 /* This hook is defined here for ATM LANE */ 5210 int (*br_fdb_test_addr_hook)(struct net_device *dev, 5211 unsigned char *addr) __read_mostly; 5212 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); 5213 #endif 5214 5215 /** 5216 * netdev_is_rx_handler_busy - check if receive handler is registered 5217 * @dev: device to check 5218 * 5219 * Check if a receive handler is already registered for a given device. 5220 * Return true if there one. 5221 * 5222 * The caller must hold the rtnl_mutex. 5223 */ 5224 bool netdev_is_rx_handler_busy(struct net_device *dev) 5225 { 5226 ASSERT_RTNL(); 5227 return dev && rtnl_dereference(dev->rx_handler); 5228 } 5229 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); 5230 5231 /** 5232 * netdev_rx_handler_register - register receive handler 5233 * @dev: device to register a handler for 5234 * @rx_handler: receive handler to register 5235 * @rx_handler_data: data pointer that is used by rx handler 5236 * 5237 * Register a receive handler for a device. This handler will then be 5238 * called from __netif_receive_skb. A negative errno code is returned 5239 * on a failure. 5240 * 5241 * The caller must hold the rtnl_mutex. 5242 * 5243 * For a general description of rx_handler, see enum rx_handler_result. 5244 */ 5245 int netdev_rx_handler_register(struct net_device *dev, 5246 rx_handler_func_t *rx_handler, 5247 void *rx_handler_data) 5248 { 5249 if (netdev_is_rx_handler_busy(dev)) 5250 return -EBUSY; 5251 5252 if (dev->priv_flags & IFF_NO_RX_HANDLER) 5253 return -EINVAL; 5254 5255 /* Note: rx_handler_data must be set before rx_handler */ 5256 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); 5257 rcu_assign_pointer(dev->rx_handler, rx_handler); 5258 5259 return 0; 5260 } 5261 EXPORT_SYMBOL_GPL(netdev_rx_handler_register); 5262 5263 /** 5264 * netdev_rx_handler_unregister - unregister receive handler 5265 * @dev: device to unregister a handler from 5266 * 5267 * Unregister a receive handler from a device. 5268 * 5269 * The caller must hold the rtnl_mutex. 5270 */ 5271 void netdev_rx_handler_unregister(struct net_device *dev) 5272 { 5273 5274 ASSERT_RTNL(); 5275 RCU_INIT_POINTER(dev->rx_handler, NULL); 5276 /* a reader seeing a non NULL rx_handler in a rcu_read_lock() 5277 * section has a guarantee to see a non NULL rx_handler_data 5278 * as well. 5279 */ 5280 synchronize_net(); 5281 RCU_INIT_POINTER(dev->rx_handler_data, NULL); 5282 } 5283 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); 5284 5285 /* 5286 * Limit the use of PFMEMALLOC reserves to those protocols that implement 5287 * the special handling of PFMEMALLOC skbs. 5288 */ 5289 static bool skb_pfmemalloc_protocol(struct sk_buff *skb) 5290 { 5291 switch (skb->protocol) { 5292 case htons(ETH_P_ARP): 5293 case htons(ETH_P_IP): 5294 case htons(ETH_P_IPV6): 5295 case htons(ETH_P_8021Q): 5296 case htons(ETH_P_8021AD): 5297 return true; 5298 default: 5299 return false; 5300 } 5301 } 5302 5303 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, 5304 int *ret, struct net_device *orig_dev) 5305 { 5306 if (nf_hook_ingress_active(skb)) { 5307 int ingress_retval; 5308 5309 if (*pt_prev) { 5310 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5311 *pt_prev = NULL; 5312 } 5313 5314 rcu_read_lock(); 5315 ingress_retval = nf_hook_ingress(skb); 5316 rcu_read_unlock(); 5317 return ingress_retval; 5318 } 5319 return 0; 5320 } 5321 5322 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc, 5323 struct packet_type **ppt_prev) 5324 { 5325 struct packet_type *ptype, *pt_prev; 5326 rx_handler_func_t *rx_handler; 5327 struct sk_buff *skb = *pskb; 5328 struct net_device *orig_dev; 5329 bool deliver_exact = false; 5330 int ret = NET_RX_DROP; 5331 __be16 type; 5332 5333 net_timestamp_check(!READ_ONCE(netdev_tstamp_prequeue), skb); 5334 5335 trace_netif_receive_skb(skb); 5336 5337 orig_dev = skb->dev; 5338 5339 skb_reset_network_header(skb); 5340 if (!skb_transport_header_was_set(skb)) 5341 skb_reset_transport_header(skb); 5342 skb_reset_mac_len(skb); 5343 5344 pt_prev = NULL; 5345 5346 another_round: 5347 skb->skb_iif = skb->dev->ifindex; 5348 5349 __this_cpu_inc(softnet_data.processed); 5350 5351 if (static_branch_unlikely(&generic_xdp_needed_key)) { 5352 int ret2; 5353 5354 migrate_disable(); 5355 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb); 5356 migrate_enable(); 5357 5358 if (ret2 != XDP_PASS) { 5359 ret = NET_RX_DROP; 5360 goto out; 5361 } 5362 } 5363 5364 if (eth_type_vlan(skb->protocol)) { 5365 skb = skb_vlan_untag(skb); 5366 if (unlikely(!skb)) 5367 goto out; 5368 } 5369 5370 if (skb_skip_tc_classify(skb)) 5371 goto skip_classify; 5372 5373 if (pfmemalloc) 5374 goto skip_taps; 5375 5376 list_for_each_entry_rcu(ptype, &ptype_all, list) { 5377 if (pt_prev) 5378 ret = deliver_skb(skb, pt_prev, orig_dev); 5379 pt_prev = ptype; 5380 } 5381 5382 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { 5383 if (pt_prev) 5384 ret = deliver_skb(skb, pt_prev, orig_dev); 5385 pt_prev = ptype; 5386 } 5387 5388 skip_taps: 5389 #ifdef CONFIG_NET_INGRESS 5390 if (static_branch_unlikely(&ingress_needed_key)) { 5391 bool another = false; 5392 5393 nf_skip_egress(skb, true); 5394 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev, 5395 &another); 5396 if (another) 5397 goto another_round; 5398 if (!skb) 5399 goto out; 5400 5401 nf_skip_egress(skb, false); 5402 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) 5403 goto out; 5404 } 5405 #endif 5406 skb_reset_redirect(skb); 5407 skip_classify: 5408 if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) 5409 goto drop; 5410 5411 if (skb_vlan_tag_present(skb)) { 5412 if (pt_prev) { 5413 ret = deliver_skb(skb, pt_prev, orig_dev); 5414 pt_prev = NULL; 5415 } 5416 if (vlan_do_receive(&skb)) 5417 goto another_round; 5418 else if (unlikely(!skb)) 5419 goto out; 5420 } 5421 5422 rx_handler = rcu_dereference(skb->dev->rx_handler); 5423 if (rx_handler) { 5424 if (pt_prev) { 5425 ret = deliver_skb(skb, pt_prev, orig_dev); 5426 pt_prev = NULL; 5427 } 5428 switch (rx_handler(&skb)) { 5429 case RX_HANDLER_CONSUMED: 5430 ret = NET_RX_SUCCESS; 5431 goto out; 5432 case RX_HANDLER_ANOTHER: 5433 goto another_round; 5434 case RX_HANDLER_EXACT: 5435 deliver_exact = true; 5436 break; 5437 case RX_HANDLER_PASS: 5438 break; 5439 default: 5440 BUG(); 5441 } 5442 } 5443 5444 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) { 5445 check_vlan_id: 5446 if (skb_vlan_tag_get_id(skb)) { 5447 /* Vlan id is non 0 and vlan_do_receive() above couldn't 5448 * find vlan device. 5449 */ 5450 skb->pkt_type = PACKET_OTHERHOST; 5451 } else if (eth_type_vlan(skb->protocol)) { 5452 /* Outer header is 802.1P with vlan 0, inner header is 5453 * 802.1Q or 802.1AD and vlan_do_receive() above could 5454 * not find vlan dev for vlan id 0. 5455 */ 5456 __vlan_hwaccel_clear_tag(skb); 5457 skb = skb_vlan_untag(skb); 5458 if (unlikely(!skb)) 5459 goto out; 5460 if (vlan_do_receive(&skb)) 5461 /* After stripping off 802.1P header with vlan 0 5462 * vlan dev is found for inner header. 5463 */ 5464 goto another_round; 5465 else if (unlikely(!skb)) 5466 goto out; 5467 else 5468 /* We have stripped outer 802.1P vlan 0 header. 5469 * But could not find vlan dev. 5470 * check again for vlan id to set OTHERHOST. 5471 */ 5472 goto check_vlan_id; 5473 } 5474 /* Note: we might in the future use prio bits 5475 * and set skb->priority like in vlan_do_receive() 5476 * For the time being, just ignore Priority Code Point 5477 */ 5478 __vlan_hwaccel_clear_tag(skb); 5479 } 5480 5481 type = skb->protocol; 5482 5483 /* deliver only exact match when indicated */ 5484 if (likely(!deliver_exact)) { 5485 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5486 &ptype_base[ntohs(type) & 5487 PTYPE_HASH_MASK]); 5488 } 5489 5490 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5491 &orig_dev->ptype_specific); 5492 5493 if (unlikely(skb->dev != orig_dev)) { 5494 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5495 &skb->dev->ptype_specific); 5496 } 5497 5498 if (pt_prev) { 5499 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 5500 goto drop; 5501 *ppt_prev = pt_prev; 5502 } else { 5503 drop: 5504 if (!deliver_exact) 5505 dev_core_stats_rx_dropped_inc(skb->dev); 5506 else 5507 dev_core_stats_rx_nohandler_inc(skb->dev); 5508 kfree_skb_reason(skb, SKB_DROP_REASON_UNHANDLED_PROTO); 5509 /* Jamal, now you will not able to escape explaining 5510 * me how you were going to use this. :-) 5511 */ 5512 ret = NET_RX_DROP; 5513 } 5514 5515 out: 5516 /* The invariant here is that if *ppt_prev is not NULL 5517 * then skb should also be non-NULL. 5518 * 5519 * Apparently *ppt_prev assignment above holds this invariant due to 5520 * skb dereferencing near it. 5521 */ 5522 *pskb = skb; 5523 return ret; 5524 } 5525 5526 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc) 5527 { 5528 struct net_device *orig_dev = skb->dev; 5529 struct packet_type *pt_prev = NULL; 5530 int ret; 5531 5532 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5533 if (pt_prev) 5534 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb, 5535 skb->dev, pt_prev, orig_dev); 5536 return ret; 5537 } 5538 5539 /** 5540 * netif_receive_skb_core - special purpose version of netif_receive_skb 5541 * @skb: buffer to process 5542 * 5543 * More direct receive version of netif_receive_skb(). It should 5544 * only be used by callers that have a need to skip RPS and Generic XDP. 5545 * Caller must also take care of handling if ``(page_is_)pfmemalloc``. 5546 * 5547 * This function may only be called from softirq context and interrupts 5548 * should be enabled. 5549 * 5550 * Return values (usually ignored): 5551 * NET_RX_SUCCESS: no congestion 5552 * NET_RX_DROP: packet was dropped 5553 */ 5554 int netif_receive_skb_core(struct sk_buff *skb) 5555 { 5556 int ret; 5557 5558 rcu_read_lock(); 5559 ret = __netif_receive_skb_one_core(skb, false); 5560 rcu_read_unlock(); 5561 5562 return ret; 5563 } 5564 EXPORT_SYMBOL(netif_receive_skb_core); 5565 5566 static inline void __netif_receive_skb_list_ptype(struct list_head *head, 5567 struct packet_type *pt_prev, 5568 struct net_device *orig_dev) 5569 { 5570 struct sk_buff *skb, *next; 5571 5572 if (!pt_prev) 5573 return; 5574 if (list_empty(head)) 5575 return; 5576 if (pt_prev->list_func != NULL) 5577 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv, 5578 ip_list_rcv, head, pt_prev, orig_dev); 5579 else 5580 list_for_each_entry_safe(skb, next, head, list) { 5581 skb_list_del_init(skb); 5582 pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 5583 } 5584 } 5585 5586 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc) 5587 { 5588 /* Fast-path assumptions: 5589 * - There is no RX handler. 5590 * - Only one packet_type matches. 5591 * If either of these fails, we will end up doing some per-packet 5592 * processing in-line, then handling the 'last ptype' for the whole 5593 * sublist. This can't cause out-of-order delivery to any single ptype, 5594 * because the 'last ptype' must be constant across the sublist, and all 5595 * other ptypes are handled per-packet. 5596 */ 5597 /* Current (common) ptype of sublist */ 5598 struct packet_type *pt_curr = NULL; 5599 /* Current (common) orig_dev of sublist */ 5600 struct net_device *od_curr = NULL; 5601 struct list_head sublist; 5602 struct sk_buff *skb, *next; 5603 5604 INIT_LIST_HEAD(&sublist); 5605 list_for_each_entry_safe(skb, next, head, list) { 5606 struct net_device *orig_dev = skb->dev; 5607 struct packet_type *pt_prev = NULL; 5608 5609 skb_list_del_init(skb); 5610 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5611 if (!pt_prev) 5612 continue; 5613 if (pt_curr != pt_prev || od_curr != orig_dev) { 5614 /* dispatch old sublist */ 5615 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5616 /* start new sublist */ 5617 INIT_LIST_HEAD(&sublist); 5618 pt_curr = pt_prev; 5619 od_curr = orig_dev; 5620 } 5621 list_add_tail(&skb->list, &sublist); 5622 } 5623 5624 /* dispatch final sublist */ 5625 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5626 } 5627 5628 static int __netif_receive_skb(struct sk_buff *skb) 5629 { 5630 int ret; 5631 5632 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { 5633 unsigned int noreclaim_flag; 5634 5635 /* 5636 * PFMEMALLOC skbs are special, they should 5637 * - be delivered to SOCK_MEMALLOC sockets only 5638 * - stay away from userspace 5639 * - have bounded memory usage 5640 * 5641 * Use PF_MEMALLOC as this saves us from propagating the allocation 5642 * context down to all allocation sites. 5643 */ 5644 noreclaim_flag = memalloc_noreclaim_save(); 5645 ret = __netif_receive_skb_one_core(skb, true); 5646 memalloc_noreclaim_restore(noreclaim_flag); 5647 } else 5648 ret = __netif_receive_skb_one_core(skb, false); 5649 5650 return ret; 5651 } 5652 5653 static void __netif_receive_skb_list(struct list_head *head) 5654 { 5655 unsigned long noreclaim_flag = 0; 5656 struct sk_buff *skb, *next; 5657 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */ 5658 5659 list_for_each_entry_safe(skb, next, head, list) { 5660 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) { 5661 struct list_head sublist; 5662 5663 /* Handle the previous sublist */ 5664 list_cut_before(&sublist, head, &skb->list); 5665 if (!list_empty(&sublist)) 5666 __netif_receive_skb_list_core(&sublist, pfmemalloc); 5667 pfmemalloc = !pfmemalloc; 5668 /* See comments in __netif_receive_skb */ 5669 if (pfmemalloc) 5670 noreclaim_flag = memalloc_noreclaim_save(); 5671 else 5672 memalloc_noreclaim_restore(noreclaim_flag); 5673 } 5674 } 5675 /* Handle the remaining sublist */ 5676 if (!list_empty(head)) 5677 __netif_receive_skb_list_core(head, pfmemalloc); 5678 /* Restore pflags */ 5679 if (pfmemalloc) 5680 memalloc_noreclaim_restore(noreclaim_flag); 5681 } 5682 5683 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp) 5684 { 5685 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog); 5686 struct bpf_prog *new = xdp->prog; 5687 int ret = 0; 5688 5689 switch (xdp->command) { 5690 case XDP_SETUP_PROG: 5691 rcu_assign_pointer(dev->xdp_prog, new); 5692 if (old) 5693 bpf_prog_put(old); 5694 5695 if (old && !new) { 5696 static_branch_dec(&generic_xdp_needed_key); 5697 } else if (new && !old) { 5698 static_branch_inc(&generic_xdp_needed_key); 5699 dev_disable_lro(dev); 5700 dev_disable_gro_hw(dev); 5701 } 5702 break; 5703 5704 default: 5705 ret = -EINVAL; 5706 break; 5707 } 5708 5709 return ret; 5710 } 5711 5712 static int netif_receive_skb_internal(struct sk_buff *skb) 5713 { 5714 int ret; 5715 5716 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb); 5717 5718 if (skb_defer_rx_timestamp(skb)) 5719 return NET_RX_SUCCESS; 5720 5721 rcu_read_lock(); 5722 #ifdef CONFIG_RPS 5723 if (static_branch_unlikely(&rps_needed)) { 5724 struct rps_dev_flow voidflow, *rflow = &voidflow; 5725 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5726 5727 if (cpu >= 0) { 5728 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5729 rcu_read_unlock(); 5730 return ret; 5731 } 5732 } 5733 #endif 5734 ret = __netif_receive_skb(skb); 5735 rcu_read_unlock(); 5736 return ret; 5737 } 5738 5739 void netif_receive_skb_list_internal(struct list_head *head) 5740 { 5741 struct sk_buff *skb, *next; 5742 struct list_head sublist; 5743 5744 INIT_LIST_HEAD(&sublist); 5745 list_for_each_entry_safe(skb, next, head, list) { 5746 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb); 5747 skb_list_del_init(skb); 5748 if (!skb_defer_rx_timestamp(skb)) 5749 list_add_tail(&skb->list, &sublist); 5750 } 5751 list_splice_init(&sublist, head); 5752 5753 rcu_read_lock(); 5754 #ifdef CONFIG_RPS 5755 if (static_branch_unlikely(&rps_needed)) { 5756 list_for_each_entry_safe(skb, next, head, list) { 5757 struct rps_dev_flow voidflow, *rflow = &voidflow; 5758 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5759 5760 if (cpu >= 0) { 5761 /* Will be handled, remove from list */ 5762 skb_list_del_init(skb); 5763 enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5764 } 5765 } 5766 } 5767 #endif 5768 __netif_receive_skb_list(head); 5769 rcu_read_unlock(); 5770 } 5771 5772 /** 5773 * netif_receive_skb - process receive buffer from network 5774 * @skb: buffer to process 5775 * 5776 * netif_receive_skb() is the main receive data processing function. 5777 * It always succeeds. The buffer may be dropped during processing 5778 * for congestion control or by the protocol layers. 5779 * 5780 * This function may only be called from softirq context and interrupts 5781 * should be enabled. 5782 * 5783 * Return values (usually ignored): 5784 * NET_RX_SUCCESS: no congestion 5785 * NET_RX_DROP: packet was dropped 5786 */ 5787 int netif_receive_skb(struct sk_buff *skb) 5788 { 5789 int ret; 5790 5791 trace_netif_receive_skb_entry(skb); 5792 5793 ret = netif_receive_skb_internal(skb); 5794 trace_netif_receive_skb_exit(ret); 5795 5796 return ret; 5797 } 5798 EXPORT_SYMBOL(netif_receive_skb); 5799 5800 /** 5801 * netif_receive_skb_list - process many receive buffers from network 5802 * @head: list of skbs to process. 5803 * 5804 * Since return value of netif_receive_skb() is normally ignored, and 5805 * wouldn't be meaningful for a list, this function returns void. 5806 * 5807 * This function may only be called from softirq context and interrupts 5808 * should be enabled. 5809 */ 5810 void netif_receive_skb_list(struct list_head *head) 5811 { 5812 struct sk_buff *skb; 5813 5814 if (list_empty(head)) 5815 return; 5816 if (trace_netif_receive_skb_list_entry_enabled()) { 5817 list_for_each_entry(skb, head, list) 5818 trace_netif_receive_skb_list_entry(skb); 5819 } 5820 netif_receive_skb_list_internal(head); 5821 trace_netif_receive_skb_list_exit(0); 5822 } 5823 EXPORT_SYMBOL(netif_receive_skb_list); 5824 5825 static DEFINE_PER_CPU(struct work_struct, flush_works); 5826 5827 /* Network device is going away, flush any packets still pending */ 5828 static void flush_backlog(struct work_struct *work) 5829 { 5830 struct sk_buff *skb, *tmp; 5831 struct softnet_data *sd; 5832 5833 local_bh_disable(); 5834 sd = this_cpu_ptr(&softnet_data); 5835 5836 rps_lock_irq_disable(sd); 5837 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { 5838 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5839 __skb_unlink(skb, &sd->input_pkt_queue); 5840 dev_kfree_skb_irq(skb); 5841 input_queue_head_incr(sd); 5842 } 5843 } 5844 rps_unlock_irq_enable(sd); 5845 5846 skb_queue_walk_safe(&sd->process_queue, skb, tmp) { 5847 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5848 __skb_unlink(skb, &sd->process_queue); 5849 kfree_skb(skb); 5850 input_queue_head_incr(sd); 5851 } 5852 } 5853 local_bh_enable(); 5854 } 5855 5856 static bool flush_required(int cpu) 5857 { 5858 #if IS_ENABLED(CONFIG_RPS) 5859 struct softnet_data *sd = &per_cpu(softnet_data, cpu); 5860 bool do_flush; 5861 5862 rps_lock_irq_disable(sd); 5863 5864 /* as insertion into process_queue happens with the rps lock held, 5865 * process_queue access may race only with dequeue 5866 */ 5867 do_flush = !skb_queue_empty(&sd->input_pkt_queue) || 5868 !skb_queue_empty_lockless(&sd->process_queue); 5869 rps_unlock_irq_enable(sd); 5870 5871 return do_flush; 5872 #endif 5873 /* without RPS we can't safely check input_pkt_queue: during a 5874 * concurrent remote skb_queue_splice() we can detect as empty both 5875 * input_pkt_queue and process_queue even if the latter could end-up 5876 * containing a lot of packets. 5877 */ 5878 return true; 5879 } 5880 5881 static void flush_all_backlogs(void) 5882 { 5883 static cpumask_t flush_cpus; 5884 unsigned int cpu; 5885 5886 /* since we are under rtnl lock protection we can use static data 5887 * for the cpumask and avoid allocating on stack the possibly 5888 * large mask 5889 */ 5890 ASSERT_RTNL(); 5891 5892 cpus_read_lock(); 5893 5894 cpumask_clear(&flush_cpus); 5895 for_each_online_cpu(cpu) { 5896 if (flush_required(cpu)) { 5897 queue_work_on(cpu, system_highpri_wq, 5898 per_cpu_ptr(&flush_works, cpu)); 5899 cpumask_set_cpu(cpu, &flush_cpus); 5900 } 5901 } 5902 5903 /* we can have in flight packet[s] on the cpus we are not flushing, 5904 * synchronize_net() in unregister_netdevice_many() will take care of 5905 * them 5906 */ 5907 for_each_cpu(cpu, &flush_cpus) 5908 flush_work(per_cpu_ptr(&flush_works, cpu)); 5909 5910 cpus_read_unlock(); 5911 } 5912 5913 static void net_rps_send_ipi(struct softnet_data *remsd) 5914 { 5915 #ifdef CONFIG_RPS 5916 while (remsd) { 5917 struct softnet_data *next = remsd->rps_ipi_next; 5918 5919 if (cpu_online(remsd->cpu)) 5920 smp_call_function_single_async(remsd->cpu, &remsd->csd); 5921 remsd = next; 5922 } 5923 #endif 5924 } 5925 5926 /* 5927 * net_rps_action_and_irq_enable sends any pending IPI's for rps. 5928 * Note: called with local irq disabled, but exits with local irq enabled. 5929 */ 5930 static void net_rps_action_and_irq_enable(struct softnet_data *sd) 5931 { 5932 #ifdef CONFIG_RPS 5933 struct softnet_data *remsd = sd->rps_ipi_list; 5934 5935 if (remsd) { 5936 sd->rps_ipi_list = NULL; 5937 5938 local_irq_enable(); 5939 5940 /* Send pending IPI's to kick RPS processing on remote cpus. */ 5941 net_rps_send_ipi(remsd); 5942 } else 5943 #endif 5944 local_irq_enable(); 5945 } 5946 5947 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) 5948 { 5949 #ifdef CONFIG_RPS 5950 return sd->rps_ipi_list != NULL; 5951 #else 5952 return false; 5953 #endif 5954 } 5955 5956 static int process_backlog(struct napi_struct *napi, int quota) 5957 { 5958 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); 5959 bool again = true; 5960 int work = 0; 5961 5962 /* Check if we have pending ipi, its better to send them now, 5963 * not waiting net_rx_action() end. 5964 */ 5965 if (sd_has_rps_ipi_waiting(sd)) { 5966 local_irq_disable(); 5967 net_rps_action_and_irq_enable(sd); 5968 } 5969 5970 napi->weight = READ_ONCE(dev_rx_weight); 5971 while (again) { 5972 struct sk_buff *skb; 5973 5974 while ((skb = __skb_dequeue(&sd->process_queue))) { 5975 rcu_read_lock(); 5976 __netif_receive_skb(skb); 5977 rcu_read_unlock(); 5978 input_queue_head_incr(sd); 5979 if (++work >= quota) 5980 return work; 5981 5982 } 5983 5984 rps_lock_irq_disable(sd); 5985 if (skb_queue_empty(&sd->input_pkt_queue)) { 5986 /* 5987 * Inline a custom version of __napi_complete(). 5988 * only current cpu owns and manipulates this napi, 5989 * and NAPI_STATE_SCHED is the only possible flag set 5990 * on backlog. 5991 * We can use a plain write instead of clear_bit(), 5992 * and we dont need an smp_mb() memory barrier. 5993 */ 5994 napi->state = 0; 5995 again = false; 5996 } else { 5997 skb_queue_splice_tail_init(&sd->input_pkt_queue, 5998 &sd->process_queue); 5999 } 6000 rps_unlock_irq_enable(sd); 6001 } 6002 6003 return work; 6004 } 6005 6006 /** 6007 * __napi_schedule - schedule for receive 6008 * @n: entry to schedule 6009 * 6010 * The entry's receive function will be scheduled to run. 6011 * Consider using __napi_schedule_irqoff() if hard irqs are masked. 6012 */ 6013 void __napi_schedule(struct napi_struct *n) 6014 { 6015 unsigned long flags; 6016 6017 local_irq_save(flags); 6018 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6019 local_irq_restore(flags); 6020 } 6021 EXPORT_SYMBOL(__napi_schedule); 6022 6023 /** 6024 * napi_schedule_prep - check if napi can be scheduled 6025 * @n: napi context 6026 * 6027 * Test if NAPI routine is already running, and if not mark 6028 * it as running. This is used as a condition variable to 6029 * insure only one NAPI poll instance runs. We also make 6030 * sure there is no pending NAPI disable. 6031 */ 6032 bool napi_schedule_prep(struct napi_struct *n) 6033 { 6034 unsigned long new, val = READ_ONCE(n->state); 6035 6036 do { 6037 if (unlikely(val & NAPIF_STATE_DISABLE)) 6038 return false; 6039 new = val | NAPIF_STATE_SCHED; 6040 6041 /* Sets STATE_MISSED bit if STATE_SCHED was already set 6042 * This was suggested by Alexander Duyck, as compiler 6043 * emits better code than : 6044 * if (val & NAPIF_STATE_SCHED) 6045 * new |= NAPIF_STATE_MISSED; 6046 */ 6047 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * 6048 NAPIF_STATE_MISSED; 6049 } while (!try_cmpxchg(&n->state, &val, new)); 6050 6051 return !(val & NAPIF_STATE_SCHED); 6052 } 6053 EXPORT_SYMBOL(napi_schedule_prep); 6054 6055 /** 6056 * __napi_schedule_irqoff - schedule for receive 6057 * @n: entry to schedule 6058 * 6059 * Variant of __napi_schedule() assuming hard irqs are masked. 6060 * 6061 * On PREEMPT_RT enabled kernels this maps to __napi_schedule() 6062 * because the interrupt disabled assumption might not be true 6063 * due to force-threaded interrupts and spinlock substitution. 6064 */ 6065 void __napi_schedule_irqoff(struct napi_struct *n) 6066 { 6067 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6068 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6069 else 6070 __napi_schedule(n); 6071 } 6072 EXPORT_SYMBOL(__napi_schedule_irqoff); 6073 6074 bool napi_complete_done(struct napi_struct *n, int work_done) 6075 { 6076 unsigned long flags, val, new, timeout = 0; 6077 bool ret = true; 6078 6079 /* 6080 * 1) Don't let napi dequeue from the cpu poll list 6081 * just in case its running on a different cpu. 6082 * 2) If we are busy polling, do nothing here, we have 6083 * the guarantee we will be called later. 6084 */ 6085 if (unlikely(n->state & (NAPIF_STATE_NPSVC | 6086 NAPIF_STATE_IN_BUSY_POLL))) 6087 return false; 6088 6089 if (work_done) { 6090 if (n->gro_bitmask) 6091 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6092 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs); 6093 } 6094 if (n->defer_hard_irqs_count > 0) { 6095 n->defer_hard_irqs_count--; 6096 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6097 if (timeout) 6098 ret = false; 6099 } 6100 if (n->gro_bitmask) { 6101 /* When the NAPI instance uses a timeout and keeps postponing 6102 * it, we need to bound somehow the time packets are kept in 6103 * the GRO layer 6104 */ 6105 napi_gro_flush(n, !!timeout); 6106 } 6107 6108 gro_normal_list(n); 6109 6110 if (unlikely(!list_empty(&n->poll_list))) { 6111 /* If n->poll_list is not empty, we need to mask irqs */ 6112 local_irq_save(flags); 6113 list_del_init(&n->poll_list); 6114 local_irq_restore(flags); 6115 } 6116 WRITE_ONCE(n->list_owner, -1); 6117 6118 val = READ_ONCE(n->state); 6119 do { 6120 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); 6121 6122 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED | 6123 NAPIF_STATE_SCHED_THREADED | 6124 NAPIF_STATE_PREFER_BUSY_POLL); 6125 6126 /* If STATE_MISSED was set, leave STATE_SCHED set, 6127 * because we will call napi->poll() one more time. 6128 * This C code was suggested by Alexander Duyck to help gcc. 6129 */ 6130 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * 6131 NAPIF_STATE_SCHED; 6132 } while (!try_cmpxchg(&n->state, &val, new)); 6133 6134 if (unlikely(val & NAPIF_STATE_MISSED)) { 6135 __napi_schedule(n); 6136 return false; 6137 } 6138 6139 if (timeout) 6140 hrtimer_start(&n->timer, ns_to_ktime(timeout), 6141 HRTIMER_MODE_REL_PINNED); 6142 return ret; 6143 } 6144 EXPORT_SYMBOL(napi_complete_done); 6145 6146 /* must be called under rcu_read_lock(), as we dont take a reference */ 6147 struct napi_struct *napi_by_id(unsigned int napi_id) 6148 { 6149 unsigned int hash = napi_id % HASH_SIZE(napi_hash); 6150 struct napi_struct *napi; 6151 6152 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) 6153 if (napi->napi_id == napi_id) 6154 return napi; 6155 6156 return NULL; 6157 } 6158 6159 #if defined(CONFIG_NET_RX_BUSY_POLL) 6160 6161 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule) 6162 { 6163 if (!skip_schedule) { 6164 gro_normal_list(napi); 6165 __napi_schedule(napi); 6166 return; 6167 } 6168 6169 if (napi->gro_bitmask) { 6170 /* flush too old packets 6171 * If HZ < 1000, flush all packets. 6172 */ 6173 napi_gro_flush(napi, HZ >= 1000); 6174 } 6175 6176 gro_normal_list(napi); 6177 clear_bit(NAPI_STATE_SCHED, &napi->state); 6178 } 6179 6180 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll, 6181 u16 budget) 6182 { 6183 bool skip_schedule = false; 6184 unsigned long timeout; 6185 int rc; 6186 6187 /* Busy polling means there is a high chance device driver hard irq 6188 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was 6189 * set in napi_schedule_prep(). 6190 * Since we are about to call napi->poll() once more, we can safely 6191 * clear NAPI_STATE_MISSED. 6192 * 6193 * Note: x86 could use a single "lock and ..." instruction 6194 * to perform these two clear_bit() 6195 */ 6196 clear_bit(NAPI_STATE_MISSED, &napi->state); 6197 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); 6198 6199 local_bh_disable(); 6200 6201 if (prefer_busy_poll) { 6202 napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs); 6203 timeout = READ_ONCE(napi->dev->gro_flush_timeout); 6204 if (napi->defer_hard_irqs_count && timeout) { 6205 hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED); 6206 skip_schedule = true; 6207 } 6208 } 6209 6210 /* All we really want here is to re-enable device interrupts. 6211 * Ideally, a new ndo_busy_poll_stop() could avoid another round. 6212 */ 6213 rc = napi->poll(napi, budget); 6214 /* We can't gro_normal_list() here, because napi->poll() might have 6215 * rearmed the napi (napi_complete_done()) in which case it could 6216 * already be running on another CPU. 6217 */ 6218 trace_napi_poll(napi, rc, budget); 6219 netpoll_poll_unlock(have_poll_lock); 6220 if (rc == budget) 6221 __busy_poll_stop(napi, skip_schedule); 6222 local_bh_enable(); 6223 } 6224 6225 void napi_busy_loop(unsigned int napi_id, 6226 bool (*loop_end)(void *, unsigned long), 6227 void *loop_end_arg, bool prefer_busy_poll, u16 budget) 6228 { 6229 unsigned long start_time = loop_end ? busy_loop_current_time() : 0; 6230 int (*napi_poll)(struct napi_struct *napi, int budget); 6231 void *have_poll_lock = NULL; 6232 struct napi_struct *napi; 6233 6234 restart: 6235 napi_poll = NULL; 6236 6237 rcu_read_lock(); 6238 6239 napi = napi_by_id(napi_id); 6240 if (!napi) 6241 goto out; 6242 6243 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6244 preempt_disable(); 6245 for (;;) { 6246 int work = 0; 6247 6248 local_bh_disable(); 6249 if (!napi_poll) { 6250 unsigned long val = READ_ONCE(napi->state); 6251 6252 /* If multiple threads are competing for this napi, 6253 * we avoid dirtying napi->state as much as we can. 6254 */ 6255 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | 6256 NAPIF_STATE_IN_BUSY_POLL)) { 6257 if (prefer_busy_poll) 6258 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6259 goto count; 6260 } 6261 if (cmpxchg(&napi->state, val, 6262 val | NAPIF_STATE_IN_BUSY_POLL | 6263 NAPIF_STATE_SCHED) != val) { 6264 if (prefer_busy_poll) 6265 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6266 goto count; 6267 } 6268 have_poll_lock = netpoll_poll_lock(napi); 6269 napi_poll = napi->poll; 6270 } 6271 work = napi_poll(napi, budget); 6272 trace_napi_poll(napi, work, budget); 6273 gro_normal_list(napi); 6274 count: 6275 if (work > 0) 6276 __NET_ADD_STATS(dev_net(napi->dev), 6277 LINUX_MIB_BUSYPOLLRXPACKETS, work); 6278 local_bh_enable(); 6279 6280 if (!loop_end || loop_end(loop_end_arg, start_time)) 6281 break; 6282 6283 if (unlikely(need_resched())) { 6284 if (napi_poll) 6285 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6286 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6287 preempt_enable(); 6288 rcu_read_unlock(); 6289 cond_resched(); 6290 if (loop_end(loop_end_arg, start_time)) 6291 return; 6292 goto restart; 6293 } 6294 cpu_relax(); 6295 } 6296 if (napi_poll) 6297 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6298 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6299 preempt_enable(); 6300 out: 6301 rcu_read_unlock(); 6302 } 6303 EXPORT_SYMBOL(napi_busy_loop); 6304 6305 #endif /* CONFIG_NET_RX_BUSY_POLL */ 6306 6307 static void napi_hash_add(struct napi_struct *napi) 6308 { 6309 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state)) 6310 return; 6311 6312 spin_lock(&napi_hash_lock); 6313 6314 /* 0..NR_CPUS range is reserved for sender_cpu use */ 6315 do { 6316 if (unlikely(++napi_gen_id < MIN_NAPI_ID)) 6317 napi_gen_id = MIN_NAPI_ID; 6318 } while (napi_by_id(napi_gen_id)); 6319 napi->napi_id = napi_gen_id; 6320 6321 hlist_add_head_rcu(&napi->napi_hash_node, 6322 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); 6323 6324 spin_unlock(&napi_hash_lock); 6325 } 6326 6327 /* Warning : caller is responsible to make sure rcu grace period 6328 * is respected before freeing memory containing @napi 6329 */ 6330 static void napi_hash_del(struct napi_struct *napi) 6331 { 6332 spin_lock(&napi_hash_lock); 6333 6334 hlist_del_init_rcu(&napi->napi_hash_node); 6335 6336 spin_unlock(&napi_hash_lock); 6337 } 6338 6339 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) 6340 { 6341 struct napi_struct *napi; 6342 6343 napi = container_of(timer, struct napi_struct, timer); 6344 6345 /* Note : we use a relaxed variant of napi_schedule_prep() not setting 6346 * NAPI_STATE_MISSED, since we do not react to a device IRQ. 6347 */ 6348 if (!napi_disable_pending(napi) && 6349 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) { 6350 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6351 __napi_schedule_irqoff(napi); 6352 } 6353 6354 return HRTIMER_NORESTART; 6355 } 6356 6357 static void init_gro_hash(struct napi_struct *napi) 6358 { 6359 int i; 6360 6361 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6362 INIT_LIST_HEAD(&napi->gro_hash[i].list); 6363 napi->gro_hash[i].count = 0; 6364 } 6365 napi->gro_bitmask = 0; 6366 } 6367 6368 int dev_set_threaded(struct net_device *dev, bool threaded) 6369 { 6370 struct napi_struct *napi; 6371 int err = 0; 6372 6373 if (dev->threaded == threaded) 6374 return 0; 6375 6376 if (threaded) { 6377 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6378 if (!napi->thread) { 6379 err = napi_kthread_create(napi); 6380 if (err) { 6381 threaded = false; 6382 break; 6383 } 6384 } 6385 } 6386 } 6387 6388 dev->threaded = threaded; 6389 6390 /* Make sure kthread is created before THREADED bit 6391 * is set. 6392 */ 6393 smp_mb__before_atomic(); 6394 6395 /* Setting/unsetting threaded mode on a napi might not immediately 6396 * take effect, if the current napi instance is actively being 6397 * polled. In this case, the switch between threaded mode and 6398 * softirq mode will happen in the next round of napi_schedule(). 6399 * This should not cause hiccups/stalls to the live traffic. 6400 */ 6401 list_for_each_entry(napi, &dev->napi_list, dev_list) 6402 assign_bit(NAPI_STATE_THREADED, &napi->state, threaded); 6403 6404 return err; 6405 } 6406 EXPORT_SYMBOL(dev_set_threaded); 6407 6408 /** 6409 * netif_queue_set_napi - Associate queue with the napi 6410 * @dev: device to which NAPI and queue belong 6411 * @queue_index: Index of queue 6412 * @type: queue type as RX or TX 6413 * @napi: NAPI context, pass NULL to clear previously set NAPI 6414 * 6415 * Set queue with its corresponding napi context. This should be done after 6416 * registering the NAPI handler for the queue-vector and the queues have been 6417 * mapped to the corresponding interrupt vector. 6418 */ 6419 void netif_queue_set_napi(struct net_device *dev, unsigned int queue_index, 6420 enum netdev_queue_type type, struct napi_struct *napi) 6421 { 6422 struct netdev_rx_queue *rxq; 6423 struct netdev_queue *txq; 6424 6425 if (WARN_ON_ONCE(napi && !napi->dev)) 6426 return; 6427 if (dev->reg_state >= NETREG_REGISTERED) 6428 ASSERT_RTNL(); 6429 6430 switch (type) { 6431 case NETDEV_QUEUE_TYPE_RX: 6432 rxq = __netif_get_rx_queue(dev, queue_index); 6433 rxq->napi = napi; 6434 return; 6435 case NETDEV_QUEUE_TYPE_TX: 6436 txq = netdev_get_tx_queue(dev, queue_index); 6437 txq->napi = napi; 6438 return; 6439 default: 6440 return; 6441 } 6442 } 6443 EXPORT_SYMBOL(netif_queue_set_napi); 6444 6445 void netif_napi_add_weight(struct net_device *dev, struct napi_struct *napi, 6446 int (*poll)(struct napi_struct *, int), int weight) 6447 { 6448 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state))) 6449 return; 6450 6451 INIT_LIST_HEAD(&napi->poll_list); 6452 INIT_HLIST_NODE(&napi->napi_hash_node); 6453 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); 6454 napi->timer.function = napi_watchdog; 6455 init_gro_hash(napi); 6456 napi->skb = NULL; 6457 INIT_LIST_HEAD(&napi->rx_list); 6458 napi->rx_count = 0; 6459 napi->poll = poll; 6460 if (weight > NAPI_POLL_WEIGHT) 6461 netdev_err_once(dev, "%s() called with weight %d\n", __func__, 6462 weight); 6463 napi->weight = weight; 6464 napi->dev = dev; 6465 #ifdef CONFIG_NETPOLL 6466 napi->poll_owner = -1; 6467 #endif 6468 napi->list_owner = -1; 6469 set_bit(NAPI_STATE_SCHED, &napi->state); 6470 set_bit(NAPI_STATE_NPSVC, &napi->state); 6471 list_add_rcu(&napi->dev_list, &dev->napi_list); 6472 napi_hash_add(napi); 6473 napi_get_frags_check(napi); 6474 /* Create kthread for this napi if dev->threaded is set. 6475 * Clear dev->threaded if kthread creation failed so that 6476 * threaded mode will not be enabled in napi_enable(). 6477 */ 6478 if (dev->threaded && napi_kthread_create(napi)) 6479 dev->threaded = 0; 6480 netif_napi_set_irq(napi, -1); 6481 } 6482 EXPORT_SYMBOL(netif_napi_add_weight); 6483 6484 void napi_disable(struct napi_struct *n) 6485 { 6486 unsigned long val, new; 6487 6488 might_sleep(); 6489 set_bit(NAPI_STATE_DISABLE, &n->state); 6490 6491 val = READ_ONCE(n->state); 6492 do { 6493 while (val & (NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC)) { 6494 usleep_range(20, 200); 6495 val = READ_ONCE(n->state); 6496 } 6497 6498 new = val | NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC; 6499 new &= ~(NAPIF_STATE_THREADED | NAPIF_STATE_PREFER_BUSY_POLL); 6500 } while (!try_cmpxchg(&n->state, &val, new)); 6501 6502 hrtimer_cancel(&n->timer); 6503 6504 clear_bit(NAPI_STATE_DISABLE, &n->state); 6505 } 6506 EXPORT_SYMBOL(napi_disable); 6507 6508 /** 6509 * napi_enable - enable NAPI scheduling 6510 * @n: NAPI context 6511 * 6512 * Resume NAPI from being scheduled on this context. 6513 * Must be paired with napi_disable. 6514 */ 6515 void napi_enable(struct napi_struct *n) 6516 { 6517 unsigned long new, val = READ_ONCE(n->state); 6518 6519 do { 6520 BUG_ON(!test_bit(NAPI_STATE_SCHED, &val)); 6521 6522 new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC); 6523 if (n->dev->threaded && n->thread) 6524 new |= NAPIF_STATE_THREADED; 6525 } while (!try_cmpxchg(&n->state, &val, new)); 6526 } 6527 EXPORT_SYMBOL(napi_enable); 6528 6529 static void flush_gro_hash(struct napi_struct *napi) 6530 { 6531 int i; 6532 6533 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6534 struct sk_buff *skb, *n; 6535 6536 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list) 6537 kfree_skb(skb); 6538 napi->gro_hash[i].count = 0; 6539 } 6540 } 6541 6542 /* Must be called in process context */ 6543 void __netif_napi_del(struct napi_struct *napi) 6544 { 6545 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state)) 6546 return; 6547 6548 napi_hash_del(napi); 6549 list_del_rcu(&napi->dev_list); 6550 napi_free_frags(napi); 6551 6552 flush_gro_hash(napi); 6553 napi->gro_bitmask = 0; 6554 6555 if (napi->thread) { 6556 kthread_stop(napi->thread); 6557 napi->thread = NULL; 6558 } 6559 } 6560 EXPORT_SYMBOL(__netif_napi_del); 6561 6562 static int __napi_poll(struct napi_struct *n, bool *repoll) 6563 { 6564 int work, weight; 6565 6566 weight = n->weight; 6567 6568 /* This NAPI_STATE_SCHED test is for avoiding a race 6569 * with netpoll's poll_napi(). Only the entity which 6570 * obtains the lock and sees NAPI_STATE_SCHED set will 6571 * actually make the ->poll() call. Therefore we avoid 6572 * accidentally calling ->poll() when NAPI is not scheduled. 6573 */ 6574 work = 0; 6575 if (napi_is_scheduled(n)) { 6576 work = n->poll(n, weight); 6577 trace_napi_poll(n, work, weight); 6578 6579 xdp_do_check_flushed(n); 6580 } 6581 6582 if (unlikely(work > weight)) 6583 netdev_err_once(n->dev, "NAPI poll function %pS returned %d, exceeding its budget of %d.\n", 6584 n->poll, work, weight); 6585 6586 if (likely(work < weight)) 6587 return work; 6588 6589 /* Drivers must not modify the NAPI state if they 6590 * consume the entire weight. In such cases this code 6591 * still "owns" the NAPI instance and therefore can 6592 * move the instance around on the list at-will. 6593 */ 6594 if (unlikely(napi_disable_pending(n))) { 6595 napi_complete(n); 6596 return work; 6597 } 6598 6599 /* The NAPI context has more processing work, but busy-polling 6600 * is preferred. Exit early. 6601 */ 6602 if (napi_prefer_busy_poll(n)) { 6603 if (napi_complete_done(n, work)) { 6604 /* If timeout is not set, we need to make sure 6605 * that the NAPI is re-scheduled. 6606 */ 6607 napi_schedule(n); 6608 } 6609 return work; 6610 } 6611 6612 if (n->gro_bitmask) { 6613 /* flush too old packets 6614 * If HZ < 1000, flush all packets. 6615 */ 6616 napi_gro_flush(n, HZ >= 1000); 6617 } 6618 6619 gro_normal_list(n); 6620 6621 /* Some drivers may have called napi_schedule 6622 * prior to exhausting their budget. 6623 */ 6624 if (unlikely(!list_empty(&n->poll_list))) { 6625 pr_warn_once("%s: Budget exhausted after napi rescheduled\n", 6626 n->dev ? n->dev->name : "backlog"); 6627 return work; 6628 } 6629 6630 *repoll = true; 6631 6632 return work; 6633 } 6634 6635 static int napi_poll(struct napi_struct *n, struct list_head *repoll) 6636 { 6637 bool do_repoll = false; 6638 void *have; 6639 int work; 6640 6641 list_del_init(&n->poll_list); 6642 6643 have = netpoll_poll_lock(n); 6644 6645 work = __napi_poll(n, &do_repoll); 6646 6647 if (do_repoll) 6648 list_add_tail(&n->poll_list, repoll); 6649 6650 netpoll_poll_unlock(have); 6651 6652 return work; 6653 } 6654 6655 static int napi_thread_wait(struct napi_struct *napi) 6656 { 6657 bool woken = false; 6658 6659 set_current_state(TASK_INTERRUPTIBLE); 6660 6661 while (!kthread_should_stop()) { 6662 /* Testing SCHED_THREADED bit here to make sure the current 6663 * kthread owns this napi and could poll on this napi. 6664 * Testing SCHED bit is not enough because SCHED bit might be 6665 * set by some other busy poll thread or by napi_disable(). 6666 */ 6667 if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state) || woken) { 6668 WARN_ON(!list_empty(&napi->poll_list)); 6669 __set_current_state(TASK_RUNNING); 6670 return 0; 6671 } 6672 6673 schedule(); 6674 /* woken being true indicates this thread owns this napi. */ 6675 woken = true; 6676 set_current_state(TASK_INTERRUPTIBLE); 6677 } 6678 __set_current_state(TASK_RUNNING); 6679 6680 return -1; 6681 } 6682 6683 static void skb_defer_free_flush(struct softnet_data *sd) 6684 { 6685 struct sk_buff *skb, *next; 6686 6687 /* Paired with WRITE_ONCE() in skb_attempt_defer_free() */ 6688 if (!READ_ONCE(sd->defer_list)) 6689 return; 6690 6691 spin_lock(&sd->defer_lock); 6692 skb = sd->defer_list; 6693 sd->defer_list = NULL; 6694 sd->defer_count = 0; 6695 spin_unlock(&sd->defer_lock); 6696 6697 while (skb != NULL) { 6698 next = skb->next; 6699 napi_consume_skb(skb, 1); 6700 skb = next; 6701 } 6702 } 6703 6704 static int napi_threaded_poll(void *data) 6705 { 6706 struct napi_struct *napi = data; 6707 struct softnet_data *sd; 6708 void *have; 6709 6710 while (!napi_thread_wait(napi)) { 6711 for (;;) { 6712 bool repoll = false; 6713 6714 local_bh_disable(); 6715 sd = this_cpu_ptr(&softnet_data); 6716 sd->in_napi_threaded_poll = true; 6717 6718 have = netpoll_poll_lock(napi); 6719 __napi_poll(napi, &repoll); 6720 netpoll_poll_unlock(have); 6721 6722 sd->in_napi_threaded_poll = false; 6723 barrier(); 6724 6725 if (sd_has_rps_ipi_waiting(sd)) { 6726 local_irq_disable(); 6727 net_rps_action_and_irq_enable(sd); 6728 } 6729 skb_defer_free_flush(sd); 6730 local_bh_enable(); 6731 6732 if (!repoll) 6733 break; 6734 6735 cond_resched(); 6736 } 6737 } 6738 return 0; 6739 } 6740 6741 static __latent_entropy void net_rx_action(struct softirq_action *h) 6742 { 6743 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 6744 unsigned long time_limit = jiffies + 6745 usecs_to_jiffies(READ_ONCE(netdev_budget_usecs)); 6746 int budget = READ_ONCE(netdev_budget); 6747 LIST_HEAD(list); 6748 LIST_HEAD(repoll); 6749 6750 start: 6751 sd->in_net_rx_action = true; 6752 local_irq_disable(); 6753 list_splice_init(&sd->poll_list, &list); 6754 local_irq_enable(); 6755 6756 for (;;) { 6757 struct napi_struct *n; 6758 6759 skb_defer_free_flush(sd); 6760 6761 if (list_empty(&list)) { 6762 if (list_empty(&repoll)) { 6763 sd->in_net_rx_action = false; 6764 barrier(); 6765 /* We need to check if ____napi_schedule() 6766 * had refilled poll_list while 6767 * sd->in_net_rx_action was true. 6768 */ 6769 if (!list_empty(&sd->poll_list)) 6770 goto start; 6771 if (!sd_has_rps_ipi_waiting(sd)) 6772 goto end; 6773 } 6774 break; 6775 } 6776 6777 n = list_first_entry(&list, struct napi_struct, poll_list); 6778 budget -= napi_poll(n, &repoll); 6779 6780 /* If softirq window is exhausted then punt. 6781 * Allow this to run for 2 jiffies since which will allow 6782 * an average latency of 1.5/HZ. 6783 */ 6784 if (unlikely(budget <= 0 || 6785 time_after_eq(jiffies, time_limit))) { 6786 sd->time_squeeze++; 6787 break; 6788 } 6789 } 6790 6791 local_irq_disable(); 6792 6793 list_splice_tail_init(&sd->poll_list, &list); 6794 list_splice_tail(&repoll, &list); 6795 list_splice(&list, &sd->poll_list); 6796 if (!list_empty(&sd->poll_list)) 6797 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 6798 else 6799 sd->in_net_rx_action = false; 6800 6801 net_rps_action_and_irq_enable(sd); 6802 end:; 6803 } 6804 6805 struct netdev_adjacent { 6806 struct net_device *dev; 6807 netdevice_tracker dev_tracker; 6808 6809 /* upper master flag, there can only be one master device per list */ 6810 bool master; 6811 6812 /* lookup ignore flag */ 6813 bool ignore; 6814 6815 /* counter for the number of times this device was added to us */ 6816 u16 ref_nr; 6817 6818 /* private field for the users */ 6819 void *private; 6820 6821 struct list_head list; 6822 struct rcu_head rcu; 6823 }; 6824 6825 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, 6826 struct list_head *adj_list) 6827 { 6828 struct netdev_adjacent *adj; 6829 6830 list_for_each_entry(adj, adj_list, list) { 6831 if (adj->dev == adj_dev) 6832 return adj; 6833 } 6834 return NULL; 6835 } 6836 6837 static int ____netdev_has_upper_dev(struct net_device *upper_dev, 6838 struct netdev_nested_priv *priv) 6839 { 6840 struct net_device *dev = (struct net_device *)priv->data; 6841 6842 return upper_dev == dev; 6843 } 6844 6845 /** 6846 * netdev_has_upper_dev - Check if device is linked to an upper device 6847 * @dev: device 6848 * @upper_dev: upper device to check 6849 * 6850 * Find out if a device is linked to specified upper device and return true 6851 * in case it is. Note that this checks only immediate upper device, 6852 * not through a complete stack of devices. The caller must hold the RTNL lock. 6853 */ 6854 bool netdev_has_upper_dev(struct net_device *dev, 6855 struct net_device *upper_dev) 6856 { 6857 struct netdev_nested_priv priv = { 6858 .data = (void *)upper_dev, 6859 }; 6860 6861 ASSERT_RTNL(); 6862 6863 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 6864 &priv); 6865 } 6866 EXPORT_SYMBOL(netdev_has_upper_dev); 6867 6868 /** 6869 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device 6870 * @dev: device 6871 * @upper_dev: upper device to check 6872 * 6873 * Find out if a device is linked to specified upper device and return true 6874 * in case it is. Note that this checks the entire upper device chain. 6875 * The caller must hold rcu lock. 6876 */ 6877 6878 bool netdev_has_upper_dev_all_rcu(struct net_device *dev, 6879 struct net_device *upper_dev) 6880 { 6881 struct netdev_nested_priv priv = { 6882 .data = (void *)upper_dev, 6883 }; 6884 6885 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 6886 &priv); 6887 } 6888 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); 6889 6890 /** 6891 * netdev_has_any_upper_dev - Check if device is linked to some device 6892 * @dev: device 6893 * 6894 * Find out if a device is linked to an upper device and return true in case 6895 * it is. The caller must hold the RTNL lock. 6896 */ 6897 bool netdev_has_any_upper_dev(struct net_device *dev) 6898 { 6899 ASSERT_RTNL(); 6900 6901 return !list_empty(&dev->adj_list.upper); 6902 } 6903 EXPORT_SYMBOL(netdev_has_any_upper_dev); 6904 6905 /** 6906 * netdev_master_upper_dev_get - Get master upper device 6907 * @dev: device 6908 * 6909 * Find a master upper device and return pointer to it or NULL in case 6910 * it's not there. The caller must hold the RTNL lock. 6911 */ 6912 struct net_device *netdev_master_upper_dev_get(struct net_device *dev) 6913 { 6914 struct netdev_adjacent *upper; 6915 6916 ASSERT_RTNL(); 6917 6918 if (list_empty(&dev->adj_list.upper)) 6919 return NULL; 6920 6921 upper = list_first_entry(&dev->adj_list.upper, 6922 struct netdev_adjacent, list); 6923 if (likely(upper->master)) 6924 return upper->dev; 6925 return NULL; 6926 } 6927 EXPORT_SYMBOL(netdev_master_upper_dev_get); 6928 6929 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev) 6930 { 6931 struct netdev_adjacent *upper; 6932 6933 ASSERT_RTNL(); 6934 6935 if (list_empty(&dev->adj_list.upper)) 6936 return NULL; 6937 6938 upper = list_first_entry(&dev->adj_list.upper, 6939 struct netdev_adjacent, list); 6940 if (likely(upper->master) && !upper->ignore) 6941 return upper->dev; 6942 return NULL; 6943 } 6944 6945 /** 6946 * netdev_has_any_lower_dev - Check if device is linked to some device 6947 * @dev: device 6948 * 6949 * Find out if a device is linked to a lower device and return true in case 6950 * it is. The caller must hold the RTNL lock. 6951 */ 6952 static bool netdev_has_any_lower_dev(struct net_device *dev) 6953 { 6954 ASSERT_RTNL(); 6955 6956 return !list_empty(&dev->adj_list.lower); 6957 } 6958 6959 void *netdev_adjacent_get_private(struct list_head *adj_list) 6960 { 6961 struct netdev_adjacent *adj; 6962 6963 adj = list_entry(adj_list, struct netdev_adjacent, list); 6964 6965 return adj->private; 6966 } 6967 EXPORT_SYMBOL(netdev_adjacent_get_private); 6968 6969 /** 6970 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list 6971 * @dev: device 6972 * @iter: list_head ** of the current position 6973 * 6974 * Gets the next device from the dev's upper list, starting from iter 6975 * position. The caller must hold RCU read lock. 6976 */ 6977 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, 6978 struct list_head **iter) 6979 { 6980 struct netdev_adjacent *upper; 6981 6982 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 6983 6984 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 6985 6986 if (&upper->list == &dev->adj_list.upper) 6987 return NULL; 6988 6989 *iter = &upper->list; 6990 6991 return upper->dev; 6992 } 6993 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); 6994 6995 static struct net_device *__netdev_next_upper_dev(struct net_device *dev, 6996 struct list_head **iter, 6997 bool *ignore) 6998 { 6999 struct netdev_adjacent *upper; 7000 7001 upper = list_entry((*iter)->next, struct netdev_adjacent, list); 7002 7003 if (&upper->list == &dev->adj_list.upper) 7004 return NULL; 7005 7006 *iter = &upper->list; 7007 *ignore = upper->ignore; 7008 7009 return upper->dev; 7010 } 7011 7012 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, 7013 struct list_head **iter) 7014 { 7015 struct netdev_adjacent *upper; 7016 7017 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 7018 7019 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7020 7021 if (&upper->list == &dev->adj_list.upper) 7022 return NULL; 7023 7024 *iter = &upper->list; 7025 7026 return upper->dev; 7027 } 7028 7029 static int __netdev_walk_all_upper_dev(struct net_device *dev, 7030 int (*fn)(struct net_device *dev, 7031 struct netdev_nested_priv *priv), 7032 struct netdev_nested_priv *priv) 7033 { 7034 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7035 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7036 int ret, cur = 0; 7037 bool ignore; 7038 7039 now = dev; 7040 iter = &dev->adj_list.upper; 7041 7042 while (1) { 7043 if (now != dev) { 7044 ret = fn(now, priv); 7045 if (ret) 7046 return ret; 7047 } 7048 7049 next = NULL; 7050 while (1) { 7051 udev = __netdev_next_upper_dev(now, &iter, &ignore); 7052 if (!udev) 7053 break; 7054 if (ignore) 7055 continue; 7056 7057 next = udev; 7058 niter = &udev->adj_list.upper; 7059 dev_stack[cur] = now; 7060 iter_stack[cur++] = iter; 7061 break; 7062 } 7063 7064 if (!next) { 7065 if (!cur) 7066 return 0; 7067 next = dev_stack[--cur]; 7068 niter = iter_stack[cur]; 7069 } 7070 7071 now = next; 7072 iter = niter; 7073 } 7074 7075 return 0; 7076 } 7077 7078 int netdev_walk_all_upper_dev_rcu(struct net_device *dev, 7079 int (*fn)(struct net_device *dev, 7080 struct netdev_nested_priv *priv), 7081 struct netdev_nested_priv *priv) 7082 { 7083 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7084 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7085 int ret, cur = 0; 7086 7087 now = dev; 7088 iter = &dev->adj_list.upper; 7089 7090 while (1) { 7091 if (now != dev) { 7092 ret = fn(now, priv); 7093 if (ret) 7094 return ret; 7095 } 7096 7097 next = NULL; 7098 while (1) { 7099 udev = netdev_next_upper_dev_rcu(now, &iter); 7100 if (!udev) 7101 break; 7102 7103 next = udev; 7104 niter = &udev->adj_list.upper; 7105 dev_stack[cur] = now; 7106 iter_stack[cur++] = iter; 7107 break; 7108 } 7109 7110 if (!next) { 7111 if (!cur) 7112 return 0; 7113 next = dev_stack[--cur]; 7114 niter = iter_stack[cur]; 7115 } 7116 7117 now = next; 7118 iter = niter; 7119 } 7120 7121 return 0; 7122 } 7123 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); 7124 7125 static bool __netdev_has_upper_dev(struct net_device *dev, 7126 struct net_device *upper_dev) 7127 { 7128 struct netdev_nested_priv priv = { 7129 .flags = 0, 7130 .data = (void *)upper_dev, 7131 }; 7132 7133 ASSERT_RTNL(); 7134 7135 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev, 7136 &priv); 7137 } 7138 7139 /** 7140 * netdev_lower_get_next_private - Get the next ->private from the 7141 * lower neighbour list 7142 * @dev: device 7143 * @iter: list_head ** of the current position 7144 * 7145 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7146 * list, starting from iter position. The caller must hold either hold the 7147 * RTNL lock or its own locking that guarantees that the neighbour lower 7148 * list will remain unchanged. 7149 */ 7150 void *netdev_lower_get_next_private(struct net_device *dev, 7151 struct list_head **iter) 7152 { 7153 struct netdev_adjacent *lower; 7154 7155 lower = list_entry(*iter, struct netdev_adjacent, list); 7156 7157 if (&lower->list == &dev->adj_list.lower) 7158 return NULL; 7159 7160 *iter = lower->list.next; 7161 7162 return lower->private; 7163 } 7164 EXPORT_SYMBOL(netdev_lower_get_next_private); 7165 7166 /** 7167 * netdev_lower_get_next_private_rcu - Get the next ->private from the 7168 * lower neighbour list, RCU 7169 * variant 7170 * @dev: device 7171 * @iter: list_head ** of the current position 7172 * 7173 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7174 * list, starting from iter position. The caller must hold RCU read lock. 7175 */ 7176 void *netdev_lower_get_next_private_rcu(struct net_device *dev, 7177 struct list_head **iter) 7178 { 7179 struct netdev_adjacent *lower; 7180 7181 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 7182 7183 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7184 7185 if (&lower->list == &dev->adj_list.lower) 7186 return NULL; 7187 7188 *iter = &lower->list; 7189 7190 return lower->private; 7191 } 7192 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); 7193 7194 /** 7195 * netdev_lower_get_next - Get the next device from the lower neighbour 7196 * list 7197 * @dev: device 7198 * @iter: list_head ** of the current position 7199 * 7200 * Gets the next netdev_adjacent from the dev's lower neighbour 7201 * list, starting from iter position. The caller must hold RTNL lock or 7202 * its own locking that guarantees that the neighbour lower 7203 * list will remain unchanged. 7204 */ 7205 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) 7206 { 7207 struct netdev_adjacent *lower; 7208 7209 lower = list_entry(*iter, struct netdev_adjacent, list); 7210 7211 if (&lower->list == &dev->adj_list.lower) 7212 return NULL; 7213 7214 *iter = lower->list.next; 7215 7216 return lower->dev; 7217 } 7218 EXPORT_SYMBOL(netdev_lower_get_next); 7219 7220 static struct net_device *netdev_next_lower_dev(struct net_device *dev, 7221 struct list_head **iter) 7222 { 7223 struct netdev_adjacent *lower; 7224 7225 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7226 7227 if (&lower->list == &dev->adj_list.lower) 7228 return NULL; 7229 7230 *iter = &lower->list; 7231 7232 return lower->dev; 7233 } 7234 7235 static struct net_device *__netdev_next_lower_dev(struct net_device *dev, 7236 struct list_head **iter, 7237 bool *ignore) 7238 { 7239 struct netdev_adjacent *lower; 7240 7241 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7242 7243 if (&lower->list == &dev->adj_list.lower) 7244 return NULL; 7245 7246 *iter = &lower->list; 7247 *ignore = lower->ignore; 7248 7249 return lower->dev; 7250 } 7251 7252 int netdev_walk_all_lower_dev(struct net_device *dev, 7253 int (*fn)(struct net_device *dev, 7254 struct netdev_nested_priv *priv), 7255 struct netdev_nested_priv *priv) 7256 { 7257 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7258 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7259 int ret, cur = 0; 7260 7261 now = dev; 7262 iter = &dev->adj_list.lower; 7263 7264 while (1) { 7265 if (now != dev) { 7266 ret = fn(now, priv); 7267 if (ret) 7268 return ret; 7269 } 7270 7271 next = NULL; 7272 while (1) { 7273 ldev = netdev_next_lower_dev(now, &iter); 7274 if (!ldev) 7275 break; 7276 7277 next = ldev; 7278 niter = &ldev->adj_list.lower; 7279 dev_stack[cur] = now; 7280 iter_stack[cur++] = iter; 7281 break; 7282 } 7283 7284 if (!next) { 7285 if (!cur) 7286 return 0; 7287 next = dev_stack[--cur]; 7288 niter = iter_stack[cur]; 7289 } 7290 7291 now = next; 7292 iter = niter; 7293 } 7294 7295 return 0; 7296 } 7297 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); 7298 7299 static int __netdev_walk_all_lower_dev(struct net_device *dev, 7300 int (*fn)(struct net_device *dev, 7301 struct netdev_nested_priv *priv), 7302 struct netdev_nested_priv *priv) 7303 { 7304 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7305 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7306 int ret, cur = 0; 7307 bool ignore; 7308 7309 now = dev; 7310 iter = &dev->adj_list.lower; 7311 7312 while (1) { 7313 if (now != dev) { 7314 ret = fn(now, priv); 7315 if (ret) 7316 return ret; 7317 } 7318 7319 next = NULL; 7320 while (1) { 7321 ldev = __netdev_next_lower_dev(now, &iter, &ignore); 7322 if (!ldev) 7323 break; 7324 if (ignore) 7325 continue; 7326 7327 next = ldev; 7328 niter = &ldev->adj_list.lower; 7329 dev_stack[cur] = now; 7330 iter_stack[cur++] = iter; 7331 break; 7332 } 7333 7334 if (!next) { 7335 if (!cur) 7336 return 0; 7337 next = dev_stack[--cur]; 7338 niter = iter_stack[cur]; 7339 } 7340 7341 now = next; 7342 iter = niter; 7343 } 7344 7345 return 0; 7346 } 7347 7348 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, 7349 struct list_head **iter) 7350 { 7351 struct netdev_adjacent *lower; 7352 7353 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7354 if (&lower->list == &dev->adj_list.lower) 7355 return NULL; 7356 7357 *iter = &lower->list; 7358 7359 return lower->dev; 7360 } 7361 EXPORT_SYMBOL(netdev_next_lower_dev_rcu); 7362 7363 static u8 __netdev_upper_depth(struct net_device *dev) 7364 { 7365 struct net_device *udev; 7366 struct list_head *iter; 7367 u8 max_depth = 0; 7368 bool ignore; 7369 7370 for (iter = &dev->adj_list.upper, 7371 udev = __netdev_next_upper_dev(dev, &iter, &ignore); 7372 udev; 7373 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) { 7374 if (ignore) 7375 continue; 7376 if (max_depth < udev->upper_level) 7377 max_depth = udev->upper_level; 7378 } 7379 7380 return max_depth; 7381 } 7382 7383 static u8 __netdev_lower_depth(struct net_device *dev) 7384 { 7385 struct net_device *ldev; 7386 struct list_head *iter; 7387 u8 max_depth = 0; 7388 bool ignore; 7389 7390 for (iter = &dev->adj_list.lower, 7391 ldev = __netdev_next_lower_dev(dev, &iter, &ignore); 7392 ldev; 7393 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) { 7394 if (ignore) 7395 continue; 7396 if (max_depth < ldev->lower_level) 7397 max_depth = ldev->lower_level; 7398 } 7399 7400 return max_depth; 7401 } 7402 7403 static int __netdev_update_upper_level(struct net_device *dev, 7404 struct netdev_nested_priv *__unused) 7405 { 7406 dev->upper_level = __netdev_upper_depth(dev) + 1; 7407 return 0; 7408 } 7409 7410 #ifdef CONFIG_LOCKDEP 7411 static LIST_HEAD(net_unlink_list); 7412 7413 static void net_unlink_todo(struct net_device *dev) 7414 { 7415 if (list_empty(&dev->unlink_list)) 7416 list_add_tail(&dev->unlink_list, &net_unlink_list); 7417 } 7418 #endif 7419 7420 static int __netdev_update_lower_level(struct net_device *dev, 7421 struct netdev_nested_priv *priv) 7422 { 7423 dev->lower_level = __netdev_lower_depth(dev) + 1; 7424 7425 #ifdef CONFIG_LOCKDEP 7426 if (!priv) 7427 return 0; 7428 7429 if (priv->flags & NESTED_SYNC_IMM) 7430 dev->nested_level = dev->lower_level - 1; 7431 if (priv->flags & NESTED_SYNC_TODO) 7432 net_unlink_todo(dev); 7433 #endif 7434 return 0; 7435 } 7436 7437 int netdev_walk_all_lower_dev_rcu(struct net_device *dev, 7438 int (*fn)(struct net_device *dev, 7439 struct netdev_nested_priv *priv), 7440 struct netdev_nested_priv *priv) 7441 { 7442 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7443 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7444 int ret, cur = 0; 7445 7446 now = dev; 7447 iter = &dev->adj_list.lower; 7448 7449 while (1) { 7450 if (now != dev) { 7451 ret = fn(now, priv); 7452 if (ret) 7453 return ret; 7454 } 7455 7456 next = NULL; 7457 while (1) { 7458 ldev = netdev_next_lower_dev_rcu(now, &iter); 7459 if (!ldev) 7460 break; 7461 7462 next = ldev; 7463 niter = &ldev->adj_list.lower; 7464 dev_stack[cur] = now; 7465 iter_stack[cur++] = iter; 7466 break; 7467 } 7468 7469 if (!next) { 7470 if (!cur) 7471 return 0; 7472 next = dev_stack[--cur]; 7473 niter = iter_stack[cur]; 7474 } 7475 7476 now = next; 7477 iter = niter; 7478 } 7479 7480 return 0; 7481 } 7482 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); 7483 7484 /** 7485 * netdev_lower_get_first_private_rcu - Get the first ->private from the 7486 * lower neighbour list, RCU 7487 * variant 7488 * @dev: device 7489 * 7490 * Gets the first netdev_adjacent->private from the dev's lower neighbour 7491 * list. The caller must hold RCU read lock. 7492 */ 7493 void *netdev_lower_get_first_private_rcu(struct net_device *dev) 7494 { 7495 struct netdev_adjacent *lower; 7496 7497 lower = list_first_or_null_rcu(&dev->adj_list.lower, 7498 struct netdev_adjacent, list); 7499 if (lower) 7500 return lower->private; 7501 return NULL; 7502 } 7503 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); 7504 7505 /** 7506 * netdev_master_upper_dev_get_rcu - Get master upper device 7507 * @dev: device 7508 * 7509 * Find a master upper device and return pointer to it or NULL in case 7510 * it's not there. The caller must hold the RCU read lock. 7511 */ 7512 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) 7513 { 7514 struct netdev_adjacent *upper; 7515 7516 upper = list_first_or_null_rcu(&dev->adj_list.upper, 7517 struct netdev_adjacent, list); 7518 if (upper && likely(upper->master)) 7519 return upper->dev; 7520 return NULL; 7521 } 7522 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); 7523 7524 static int netdev_adjacent_sysfs_add(struct net_device *dev, 7525 struct net_device *adj_dev, 7526 struct list_head *dev_list) 7527 { 7528 char linkname[IFNAMSIZ+7]; 7529 7530 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7531 "upper_%s" : "lower_%s", adj_dev->name); 7532 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), 7533 linkname); 7534 } 7535 static void netdev_adjacent_sysfs_del(struct net_device *dev, 7536 char *name, 7537 struct list_head *dev_list) 7538 { 7539 char linkname[IFNAMSIZ+7]; 7540 7541 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7542 "upper_%s" : "lower_%s", name); 7543 sysfs_remove_link(&(dev->dev.kobj), linkname); 7544 } 7545 7546 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, 7547 struct net_device *adj_dev, 7548 struct list_head *dev_list) 7549 { 7550 return (dev_list == &dev->adj_list.upper || 7551 dev_list == &dev->adj_list.lower) && 7552 net_eq(dev_net(dev), dev_net(adj_dev)); 7553 } 7554 7555 static int __netdev_adjacent_dev_insert(struct net_device *dev, 7556 struct net_device *adj_dev, 7557 struct list_head *dev_list, 7558 void *private, bool master) 7559 { 7560 struct netdev_adjacent *adj; 7561 int ret; 7562 7563 adj = __netdev_find_adj(adj_dev, dev_list); 7564 7565 if (adj) { 7566 adj->ref_nr += 1; 7567 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", 7568 dev->name, adj_dev->name, adj->ref_nr); 7569 7570 return 0; 7571 } 7572 7573 adj = kmalloc(sizeof(*adj), GFP_KERNEL); 7574 if (!adj) 7575 return -ENOMEM; 7576 7577 adj->dev = adj_dev; 7578 adj->master = master; 7579 adj->ref_nr = 1; 7580 adj->private = private; 7581 adj->ignore = false; 7582 netdev_hold(adj_dev, &adj->dev_tracker, GFP_KERNEL); 7583 7584 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", 7585 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); 7586 7587 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { 7588 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); 7589 if (ret) 7590 goto free_adj; 7591 } 7592 7593 /* Ensure that master link is always the first item in list. */ 7594 if (master) { 7595 ret = sysfs_create_link(&(dev->dev.kobj), 7596 &(adj_dev->dev.kobj), "master"); 7597 if (ret) 7598 goto remove_symlinks; 7599 7600 list_add_rcu(&adj->list, dev_list); 7601 } else { 7602 list_add_tail_rcu(&adj->list, dev_list); 7603 } 7604 7605 return 0; 7606 7607 remove_symlinks: 7608 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7609 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7610 free_adj: 7611 netdev_put(adj_dev, &adj->dev_tracker); 7612 kfree(adj); 7613 7614 return ret; 7615 } 7616 7617 static void __netdev_adjacent_dev_remove(struct net_device *dev, 7618 struct net_device *adj_dev, 7619 u16 ref_nr, 7620 struct list_head *dev_list) 7621 { 7622 struct netdev_adjacent *adj; 7623 7624 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", 7625 dev->name, adj_dev->name, ref_nr); 7626 7627 adj = __netdev_find_adj(adj_dev, dev_list); 7628 7629 if (!adj) { 7630 pr_err("Adjacency does not exist for device %s from %s\n", 7631 dev->name, adj_dev->name); 7632 WARN_ON(1); 7633 return; 7634 } 7635 7636 if (adj->ref_nr > ref_nr) { 7637 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", 7638 dev->name, adj_dev->name, ref_nr, 7639 adj->ref_nr - ref_nr); 7640 adj->ref_nr -= ref_nr; 7641 return; 7642 } 7643 7644 if (adj->master) 7645 sysfs_remove_link(&(dev->dev.kobj), "master"); 7646 7647 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7648 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7649 7650 list_del_rcu(&adj->list); 7651 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", 7652 adj_dev->name, dev->name, adj_dev->name); 7653 netdev_put(adj_dev, &adj->dev_tracker); 7654 kfree_rcu(adj, rcu); 7655 } 7656 7657 static int __netdev_adjacent_dev_link_lists(struct net_device *dev, 7658 struct net_device *upper_dev, 7659 struct list_head *up_list, 7660 struct list_head *down_list, 7661 void *private, bool master) 7662 { 7663 int ret; 7664 7665 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, 7666 private, master); 7667 if (ret) 7668 return ret; 7669 7670 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, 7671 private, false); 7672 if (ret) { 7673 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); 7674 return ret; 7675 } 7676 7677 return 0; 7678 } 7679 7680 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, 7681 struct net_device *upper_dev, 7682 u16 ref_nr, 7683 struct list_head *up_list, 7684 struct list_head *down_list) 7685 { 7686 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); 7687 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); 7688 } 7689 7690 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, 7691 struct net_device *upper_dev, 7692 void *private, bool master) 7693 { 7694 return __netdev_adjacent_dev_link_lists(dev, upper_dev, 7695 &dev->adj_list.upper, 7696 &upper_dev->adj_list.lower, 7697 private, master); 7698 } 7699 7700 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, 7701 struct net_device *upper_dev) 7702 { 7703 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, 7704 &dev->adj_list.upper, 7705 &upper_dev->adj_list.lower); 7706 } 7707 7708 static int __netdev_upper_dev_link(struct net_device *dev, 7709 struct net_device *upper_dev, bool master, 7710 void *upper_priv, void *upper_info, 7711 struct netdev_nested_priv *priv, 7712 struct netlink_ext_ack *extack) 7713 { 7714 struct netdev_notifier_changeupper_info changeupper_info = { 7715 .info = { 7716 .dev = dev, 7717 .extack = extack, 7718 }, 7719 .upper_dev = upper_dev, 7720 .master = master, 7721 .linking = true, 7722 .upper_info = upper_info, 7723 }; 7724 struct net_device *master_dev; 7725 int ret = 0; 7726 7727 ASSERT_RTNL(); 7728 7729 if (dev == upper_dev) 7730 return -EBUSY; 7731 7732 /* To prevent loops, check if dev is not upper device to upper_dev. */ 7733 if (__netdev_has_upper_dev(upper_dev, dev)) 7734 return -EBUSY; 7735 7736 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV) 7737 return -EMLINK; 7738 7739 if (!master) { 7740 if (__netdev_has_upper_dev(dev, upper_dev)) 7741 return -EEXIST; 7742 } else { 7743 master_dev = __netdev_master_upper_dev_get(dev); 7744 if (master_dev) 7745 return master_dev == upper_dev ? -EEXIST : -EBUSY; 7746 } 7747 7748 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7749 &changeupper_info.info); 7750 ret = notifier_to_errno(ret); 7751 if (ret) 7752 return ret; 7753 7754 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, 7755 master); 7756 if (ret) 7757 return ret; 7758 7759 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 7760 &changeupper_info.info); 7761 ret = notifier_to_errno(ret); 7762 if (ret) 7763 goto rollback; 7764 7765 __netdev_update_upper_level(dev, NULL); 7766 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 7767 7768 __netdev_update_lower_level(upper_dev, priv); 7769 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 7770 priv); 7771 7772 return 0; 7773 7774 rollback: 7775 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 7776 7777 return ret; 7778 } 7779 7780 /** 7781 * netdev_upper_dev_link - Add a link to the upper device 7782 * @dev: device 7783 * @upper_dev: new upper device 7784 * @extack: netlink extended ack 7785 * 7786 * Adds a link to device which is upper to this one. The caller must hold 7787 * the RTNL lock. On a failure a negative errno code is returned. 7788 * On success the reference counts are adjusted and the function 7789 * returns zero. 7790 */ 7791 int netdev_upper_dev_link(struct net_device *dev, 7792 struct net_device *upper_dev, 7793 struct netlink_ext_ack *extack) 7794 { 7795 struct netdev_nested_priv priv = { 7796 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7797 .data = NULL, 7798 }; 7799 7800 return __netdev_upper_dev_link(dev, upper_dev, false, 7801 NULL, NULL, &priv, extack); 7802 } 7803 EXPORT_SYMBOL(netdev_upper_dev_link); 7804 7805 /** 7806 * netdev_master_upper_dev_link - Add a master link to the upper device 7807 * @dev: device 7808 * @upper_dev: new upper device 7809 * @upper_priv: upper device private 7810 * @upper_info: upper info to be passed down via notifier 7811 * @extack: netlink extended ack 7812 * 7813 * Adds a link to device which is upper to this one. In this case, only 7814 * one master upper device can be linked, although other non-master devices 7815 * might be linked as well. The caller must hold the RTNL lock. 7816 * On a failure a negative errno code is returned. On success the reference 7817 * counts are adjusted and the function returns zero. 7818 */ 7819 int netdev_master_upper_dev_link(struct net_device *dev, 7820 struct net_device *upper_dev, 7821 void *upper_priv, void *upper_info, 7822 struct netlink_ext_ack *extack) 7823 { 7824 struct netdev_nested_priv priv = { 7825 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7826 .data = NULL, 7827 }; 7828 7829 return __netdev_upper_dev_link(dev, upper_dev, true, 7830 upper_priv, upper_info, &priv, extack); 7831 } 7832 EXPORT_SYMBOL(netdev_master_upper_dev_link); 7833 7834 static void __netdev_upper_dev_unlink(struct net_device *dev, 7835 struct net_device *upper_dev, 7836 struct netdev_nested_priv *priv) 7837 { 7838 struct netdev_notifier_changeupper_info changeupper_info = { 7839 .info = { 7840 .dev = dev, 7841 }, 7842 .upper_dev = upper_dev, 7843 .linking = false, 7844 }; 7845 7846 ASSERT_RTNL(); 7847 7848 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; 7849 7850 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7851 &changeupper_info.info); 7852 7853 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 7854 7855 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 7856 &changeupper_info.info); 7857 7858 __netdev_update_upper_level(dev, NULL); 7859 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 7860 7861 __netdev_update_lower_level(upper_dev, priv); 7862 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 7863 priv); 7864 } 7865 7866 /** 7867 * netdev_upper_dev_unlink - Removes a link to upper device 7868 * @dev: device 7869 * @upper_dev: new upper device 7870 * 7871 * Removes a link to device which is upper to this one. The caller must hold 7872 * the RTNL lock. 7873 */ 7874 void netdev_upper_dev_unlink(struct net_device *dev, 7875 struct net_device *upper_dev) 7876 { 7877 struct netdev_nested_priv priv = { 7878 .flags = NESTED_SYNC_TODO, 7879 .data = NULL, 7880 }; 7881 7882 __netdev_upper_dev_unlink(dev, upper_dev, &priv); 7883 } 7884 EXPORT_SYMBOL(netdev_upper_dev_unlink); 7885 7886 static void __netdev_adjacent_dev_set(struct net_device *upper_dev, 7887 struct net_device *lower_dev, 7888 bool val) 7889 { 7890 struct netdev_adjacent *adj; 7891 7892 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower); 7893 if (adj) 7894 adj->ignore = val; 7895 7896 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper); 7897 if (adj) 7898 adj->ignore = val; 7899 } 7900 7901 static void netdev_adjacent_dev_disable(struct net_device *upper_dev, 7902 struct net_device *lower_dev) 7903 { 7904 __netdev_adjacent_dev_set(upper_dev, lower_dev, true); 7905 } 7906 7907 static void netdev_adjacent_dev_enable(struct net_device *upper_dev, 7908 struct net_device *lower_dev) 7909 { 7910 __netdev_adjacent_dev_set(upper_dev, lower_dev, false); 7911 } 7912 7913 int netdev_adjacent_change_prepare(struct net_device *old_dev, 7914 struct net_device *new_dev, 7915 struct net_device *dev, 7916 struct netlink_ext_ack *extack) 7917 { 7918 struct netdev_nested_priv priv = { 7919 .flags = 0, 7920 .data = NULL, 7921 }; 7922 int err; 7923 7924 if (!new_dev) 7925 return 0; 7926 7927 if (old_dev && new_dev != old_dev) 7928 netdev_adjacent_dev_disable(dev, old_dev); 7929 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv, 7930 extack); 7931 if (err) { 7932 if (old_dev && new_dev != old_dev) 7933 netdev_adjacent_dev_enable(dev, old_dev); 7934 return err; 7935 } 7936 7937 return 0; 7938 } 7939 EXPORT_SYMBOL(netdev_adjacent_change_prepare); 7940 7941 void netdev_adjacent_change_commit(struct net_device *old_dev, 7942 struct net_device *new_dev, 7943 struct net_device *dev) 7944 { 7945 struct netdev_nested_priv priv = { 7946 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7947 .data = NULL, 7948 }; 7949 7950 if (!new_dev || !old_dev) 7951 return; 7952 7953 if (new_dev == old_dev) 7954 return; 7955 7956 netdev_adjacent_dev_enable(dev, old_dev); 7957 __netdev_upper_dev_unlink(old_dev, dev, &priv); 7958 } 7959 EXPORT_SYMBOL(netdev_adjacent_change_commit); 7960 7961 void netdev_adjacent_change_abort(struct net_device *old_dev, 7962 struct net_device *new_dev, 7963 struct net_device *dev) 7964 { 7965 struct netdev_nested_priv priv = { 7966 .flags = 0, 7967 .data = NULL, 7968 }; 7969 7970 if (!new_dev) 7971 return; 7972 7973 if (old_dev && new_dev != old_dev) 7974 netdev_adjacent_dev_enable(dev, old_dev); 7975 7976 __netdev_upper_dev_unlink(new_dev, dev, &priv); 7977 } 7978 EXPORT_SYMBOL(netdev_adjacent_change_abort); 7979 7980 /** 7981 * netdev_bonding_info_change - Dispatch event about slave change 7982 * @dev: device 7983 * @bonding_info: info to dispatch 7984 * 7985 * Send NETDEV_BONDING_INFO to netdev notifiers with info. 7986 * The caller must hold the RTNL lock. 7987 */ 7988 void netdev_bonding_info_change(struct net_device *dev, 7989 struct netdev_bonding_info *bonding_info) 7990 { 7991 struct netdev_notifier_bonding_info info = { 7992 .info.dev = dev, 7993 }; 7994 7995 memcpy(&info.bonding_info, bonding_info, 7996 sizeof(struct netdev_bonding_info)); 7997 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, 7998 &info.info); 7999 } 8000 EXPORT_SYMBOL(netdev_bonding_info_change); 8001 8002 static int netdev_offload_xstats_enable_l3(struct net_device *dev, 8003 struct netlink_ext_ack *extack) 8004 { 8005 struct netdev_notifier_offload_xstats_info info = { 8006 .info.dev = dev, 8007 .info.extack = extack, 8008 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, 8009 }; 8010 int err; 8011 int rc; 8012 8013 dev->offload_xstats_l3 = kzalloc(sizeof(*dev->offload_xstats_l3), 8014 GFP_KERNEL); 8015 if (!dev->offload_xstats_l3) 8016 return -ENOMEM; 8017 8018 rc = call_netdevice_notifiers_info_robust(NETDEV_OFFLOAD_XSTATS_ENABLE, 8019 NETDEV_OFFLOAD_XSTATS_DISABLE, 8020 &info.info); 8021 err = notifier_to_errno(rc); 8022 if (err) 8023 goto free_stats; 8024 8025 return 0; 8026 8027 free_stats: 8028 kfree(dev->offload_xstats_l3); 8029 dev->offload_xstats_l3 = NULL; 8030 return err; 8031 } 8032 8033 int netdev_offload_xstats_enable(struct net_device *dev, 8034 enum netdev_offload_xstats_type type, 8035 struct netlink_ext_ack *extack) 8036 { 8037 ASSERT_RTNL(); 8038 8039 if (netdev_offload_xstats_enabled(dev, type)) 8040 return -EALREADY; 8041 8042 switch (type) { 8043 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 8044 return netdev_offload_xstats_enable_l3(dev, extack); 8045 } 8046 8047 WARN_ON(1); 8048 return -EINVAL; 8049 } 8050 EXPORT_SYMBOL(netdev_offload_xstats_enable); 8051 8052 static void netdev_offload_xstats_disable_l3(struct net_device *dev) 8053 { 8054 struct netdev_notifier_offload_xstats_info info = { 8055 .info.dev = dev, 8056 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, 8057 }; 8058 8059 call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_DISABLE, 8060 &info.info); 8061 kfree(dev->offload_xstats_l3); 8062 dev->offload_xstats_l3 = NULL; 8063 } 8064 8065 int netdev_offload_xstats_disable(struct net_device *dev, 8066 enum netdev_offload_xstats_type type) 8067 { 8068 ASSERT_RTNL(); 8069 8070 if (!netdev_offload_xstats_enabled(dev, type)) 8071 return -EALREADY; 8072 8073 switch (type) { 8074 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 8075 netdev_offload_xstats_disable_l3(dev); 8076 return 0; 8077 } 8078 8079 WARN_ON(1); 8080 return -EINVAL; 8081 } 8082 EXPORT_SYMBOL(netdev_offload_xstats_disable); 8083 8084 static void netdev_offload_xstats_disable_all(struct net_device *dev) 8085 { 8086 netdev_offload_xstats_disable(dev, NETDEV_OFFLOAD_XSTATS_TYPE_L3); 8087 } 8088 8089 static struct rtnl_hw_stats64 * 8090 netdev_offload_xstats_get_ptr(const struct net_device *dev, 8091 enum netdev_offload_xstats_type type) 8092 { 8093 switch (type) { 8094 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 8095 return dev->offload_xstats_l3; 8096 } 8097 8098 WARN_ON(1); 8099 return NULL; 8100 } 8101 8102 bool netdev_offload_xstats_enabled(const struct net_device *dev, 8103 enum netdev_offload_xstats_type type) 8104 { 8105 ASSERT_RTNL(); 8106 8107 return netdev_offload_xstats_get_ptr(dev, type); 8108 } 8109 EXPORT_SYMBOL(netdev_offload_xstats_enabled); 8110 8111 struct netdev_notifier_offload_xstats_ru { 8112 bool used; 8113 }; 8114 8115 struct netdev_notifier_offload_xstats_rd { 8116 struct rtnl_hw_stats64 stats; 8117 bool used; 8118 }; 8119 8120 static void netdev_hw_stats64_add(struct rtnl_hw_stats64 *dest, 8121 const struct rtnl_hw_stats64 *src) 8122 { 8123 dest->rx_packets += src->rx_packets; 8124 dest->tx_packets += src->tx_packets; 8125 dest->rx_bytes += src->rx_bytes; 8126 dest->tx_bytes += src->tx_bytes; 8127 dest->rx_errors += src->rx_errors; 8128 dest->tx_errors += src->tx_errors; 8129 dest->rx_dropped += src->rx_dropped; 8130 dest->tx_dropped += src->tx_dropped; 8131 dest->multicast += src->multicast; 8132 } 8133 8134 static int netdev_offload_xstats_get_used(struct net_device *dev, 8135 enum netdev_offload_xstats_type type, 8136 bool *p_used, 8137 struct netlink_ext_ack *extack) 8138 { 8139 struct netdev_notifier_offload_xstats_ru report_used = {}; 8140 struct netdev_notifier_offload_xstats_info info = { 8141 .info.dev = dev, 8142 .info.extack = extack, 8143 .type = type, 8144 .report_used = &report_used, 8145 }; 8146 int rc; 8147 8148 WARN_ON(!netdev_offload_xstats_enabled(dev, type)); 8149 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_USED, 8150 &info.info); 8151 *p_used = report_used.used; 8152 return notifier_to_errno(rc); 8153 } 8154 8155 static int netdev_offload_xstats_get_stats(struct net_device *dev, 8156 enum netdev_offload_xstats_type type, 8157 struct rtnl_hw_stats64 *p_stats, 8158 bool *p_used, 8159 struct netlink_ext_ack *extack) 8160 { 8161 struct netdev_notifier_offload_xstats_rd report_delta = {}; 8162 struct netdev_notifier_offload_xstats_info info = { 8163 .info.dev = dev, 8164 .info.extack = extack, 8165 .type = type, 8166 .report_delta = &report_delta, 8167 }; 8168 struct rtnl_hw_stats64 *stats; 8169 int rc; 8170 8171 stats = netdev_offload_xstats_get_ptr(dev, type); 8172 if (WARN_ON(!stats)) 8173 return -EINVAL; 8174 8175 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_DELTA, 8176 &info.info); 8177 8178 /* Cache whatever we got, even if there was an error, otherwise the 8179 * successful stats retrievals would get lost. 8180 */ 8181 netdev_hw_stats64_add(stats, &report_delta.stats); 8182 8183 if (p_stats) 8184 *p_stats = *stats; 8185 *p_used = report_delta.used; 8186 8187 return notifier_to_errno(rc); 8188 } 8189 8190 int netdev_offload_xstats_get(struct net_device *dev, 8191 enum netdev_offload_xstats_type type, 8192 struct rtnl_hw_stats64 *p_stats, bool *p_used, 8193 struct netlink_ext_ack *extack) 8194 { 8195 ASSERT_RTNL(); 8196 8197 if (p_stats) 8198 return netdev_offload_xstats_get_stats(dev, type, p_stats, 8199 p_used, extack); 8200 else 8201 return netdev_offload_xstats_get_used(dev, type, p_used, 8202 extack); 8203 } 8204 EXPORT_SYMBOL(netdev_offload_xstats_get); 8205 8206 void 8207 netdev_offload_xstats_report_delta(struct netdev_notifier_offload_xstats_rd *report_delta, 8208 const struct rtnl_hw_stats64 *stats) 8209 { 8210 report_delta->used = true; 8211 netdev_hw_stats64_add(&report_delta->stats, stats); 8212 } 8213 EXPORT_SYMBOL(netdev_offload_xstats_report_delta); 8214 8215 void 8216 netdev_offload_xstats_report_used(struct netdev_notifier_offload_xstats_ru *report_used) 8217 { 8218 report_used->used = true; 8219 } 8220 EXPORT_SYMBOL(netdev_offload_xstats_report_used); 8221 8222 void netdev_offload_xstats_push_delta(struct net_device *dev, 8223 enum netdev_offload_xstats_type type, 8224 const struct rtnl_hw_stats64 *p_stats) 8225 { 8226 struct rtnl_hw_stats64 *stats; 8227 8228 ASSERT_RTNL(); 8229 8230 stats = netdev_offload_xstats_get_ptr(dev, type); 8231 if (WARN_ON(!stats)) 8232 return; 8233 8234 netdev_hw_stats64_add(stats, p_stats); 8235 } 8236 EXPORT_SYMBOL(netdev_offload_xstats_push_delta); 8237 8238 /** 8239 * netdev_get_xmit_slave - Get the xmit slave of master device 8240 * @dev: device 8241 * @skb: The packet 8242 * @all_slaves: assume all the slaves are active 8243 * 8244 * The reference counters are not incremented so the caller must be 8245 * careful with locks. The caller must hold RCU lock. 8246 * %NULL is returned if no slave is found. 8247 */ 8248 8249 struct net_device *netdev_get_xmit_slave(struct net_device *dev, 8250 struct sk_buff *skb, 8251 bool all_slaves) 8252 { 8253 const struct net_device_ops *ops = dev->netdev_ops; 8254 8255 if (!ops->ndo_get_xmit_slave) 8256 return NULL; 8257 return ops->ndo_get_xmit_slave(dev, skb, all_slaves); 8258 } 8259 EXPORT_SYMBOL(netdev_get_xmit_slave); 8260 8261 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev, 8262 struct sock *sk) 8263 { 8264 const struct net_device_ops *ops = dev->netdev_ops; 8265 8266 if (!ops->ndo_sk_get_lower_dev) 8267 return NULL; 8268 return ops->ndo_sk_get_lower_dev(dev, sk); 8269 } 8270 8271 /** 8272 * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket 8273 * @dev: device 8274 * @sk: the socket 8275 * 8276 * %NULL is returned if no lower device is found. 8277 */ 8278 8279 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev, 8280 struct sock *sk) 8281 { 8282 struct net_device *lower; 8283 8284 lower = netdev_sk_get_lower_dev(dev, sk); 8285 while (lower) { 8286 dev = lower; 8287 lower = netdev_sk_get_lower_dev(dev, sk); 8288 } 8289 8290 return dev; 8291 } 8292 EXPORT_SYMBOL(netdev_sk_get_lowest_dev); 8293 8294 static void netdev_adjacent_add_links(struct net_device *dev) 8295 { 8296 struct netdev_adjacent *iter; 8297 8298 struct net *net = dev_net(dev); 8299 8300 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8301 if (!net_eq(net, dev_net(iter->dev))) 8302 continue; 8303 netdev_adjacent_sysfs_add(iter->dev, dev, 8304 &iter->dev->adj_list.lower); 8305 netdev_adjacent_sysfs_add(dev, iter->dev, 8306 &dev->adj_list.upper); 8307 } 8308 8309 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8310 if (!net_eq(net, dev_net(iter->dev))) 8311 continue; 8312 netdev_adjacent_sysfs_add(iter->dev, dev, 8313 &iter->dev->adj_list.upper); 8314 netdev_adjacent_sysfs_add(dev, iter->dev, 8315 &dev->adj_list.lower); 8316 } 8317 } 8318 8319 static void netdev_adjacent_del_links(struct net_device *dev) 8320 { 8321 struct netdev_adjacent *iter; 8322 8323 struct net *net = dev_net(dev); 8324 8325 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8326 if (!net_eq(net, dev_net(iter->dev))) 8327 continue; 8328 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8329 &iter->dev->adj_list.lower); 8330 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8331 &dev->adj_list.upper); 8332 } 8333 8334 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8335 if (!net_eq(net, dev_net(iter->dev))) 8336 continue; 8337 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8338 &iter->dev->adj_list.upper); 8339 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8340 &dev->adj_list.lower); 8341 } 8342 } 8343 8344 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) 8345 { 8346 struct netdev_adjacent *iter; 8347 8348 struct net *net = dev_net(dev); 8349 8350 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8351 if (!net_eq(net, dev_net(iter->dev))) 8352 continue; 8353 netdev_adjacent_sysfs_del(iter->dev, oldname, 8354 &iter->dev->adj_list.lower); 8355 netdev_adjacent_sysfs_add(iter->dev, dev, 8356 &iter->dev->adj_list.lower); 8357 } 8358 8359 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8360 if (!net_eq(net, dev_net(iter->dev))) 8361 continue; 8362 netdev_adjacent_sysfs_del(iter->dev, oldname, 8363 &iter->dev->adj_list.upper); 8364 netdev_adjacent_sysfs_add(iter->dev, dev, 8365 &iter->dev->adj_list.upper); 8366 } 8367 } 8368 8369 void *netdev_lower_dev_get_private(struct net_device *dev, 8370 struct net_device *lower_dev) 8371 { 8372 struct netdev_adjacent *lower; 8373 8374 if (!lower_dev) 8375 return NULL; 8376 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); 8377 if (!lower) 8378 return NULL; 8379 8380 return lower->private; 8381 } 8382 EXPORT_SYMBOL(netdev_lower_dev_get_private); 8383 8384 8385 /** 8386 * netdev_lower_state_changed - Dispatch event about lower device state change 8387 * @lower_dev: device 8388 * @lower_state_info: state to dispatch 8389 * 8390 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. 8391 * The caller must hold the RTNL lock. 8392 */ 8393 void netdev_lower_state_changed(struct net_device *lower_dev, 8394 void *lower_state_info) 8395 { 8396 struct netdev_notifier_changelowerstate_info changelowerstate_info = { 8397 .info.dev = lower_dev, 8398 }; 8399 8400 ASSERT_RTNL(); 8401 changelowerstate_info.lower_state_info = lower_state_info; 8402 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, 8403 &changelowerstate_info.info); 8404 } 8405 EXPORT_SYMBOL(netdev_lower_state_changed); 8406 8407 static void dev_change_rx_flags(struct net_device *dev, int flags) 8408 { 8409 const struct net_device_ops *ops = dev->netdev_ops; 8410 8411 if (ops->ndo_change_rx_flags) 8412 ops->ndo_change_rx_flags(dev, flags); 8413 } 8414 8415 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) 8416 { 8417 unsigned int old_flags = dev->flags; 8418 kuid_t uid; 8419 kgid_t gid; 8420 8421 ASSERT_RTNL(); 8422 8423 dev->flags |= IFF_PROMISC; 8424 dev->promiscuity += inc; 8425 if (dev->promiscuity == 0) { 8426 /* 8427 * Avoid overflow. 8428 * If inc causes overflow, untouch promisc and return error. 8429 */ 8430 if (inc < 0) 8431 dev->flags &= ~IFF_PROMISC; 8432 else { 8433 dev->promiscuity -= inc; 8434 netdev_warn(dev, "promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n"); 8435 return -EOVERFLOW; 8436 } 8437 } 8438 if (dev->flags != old_flags) { 8439 netdev_info(dev, "%s promiscuous mode\n", 8440 dev->flags & IFF_PROMISC ? "entered" : "left"); 8441 if (audit_enabled) { 8442 current_uid_gid(&uid, &gid); 8443 audit_log(audit_context(), GFP_ATOMIC, 8444 AUDIT_ANOM_PROMISCUOUS, 8445 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", 8446 dev->name, (dev->flags & IFF_PROMISC), 8447 (old_flags & IFF_PROMISC), 8448 from_kuid(&init_user_ns, audit_get_loginuid(current)), 8449 from_kuid(&init_user_ns, uid), 8450 from_kgid(&init_user_ns, gid), 8451 audit_get_sessionid(current)); 8452 } 8453 8454 dev_change_rx_flags(dev, IFF_PROMISC); 8455 } 8456 if (notify) 8457 __dev_notify_flags(dev, old_flags, IFF_PROMISC, 0, NULL); 8458 return 0; 8459 } 8460 8461 /** 8462 * dev_set_promiscuity - update promiscuity count on a device 8463 * @dev: device 8464 * @inc: modifier 8465 * 8466 * Add or remove promiscuity from a device. While the count in the device 8467 * remains above zero the interface remains promiscuous. Once it hits zero 8468 * the device reverts back to normal filtering operation. A negative inc 8469 * value is used to drop promiscuity on the device. 8470 * Return 0 if successful or a negative errno code on error. 8471 */ 8472 int dev_set_promiscuity(struct net_device *dev, int inc) 8473 { 8474 unsigned int old_flags = dev->flags; 8475 int err; 8476 8477 err = __dev_set_promiscuity(dev, inc, true); 8478 if (err < 0) 8479 return err; 8480 if (dev->flags != old_flags) 8481 dev_set_rx_mode(dev); 8482 return err; 8483 } 8484 EXPORT_SYMBOL(dev_set_promiscuity); 8485 8486 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) 8487 { 8488 unsigned int old_flags = dev->flags, old_gflags = dev->gflags; 8489 8490 ASSERT_RTNL(); 8491 8492 dev->flags |= IFF_ALLMULTI; 8493 dev->allmulti += inc; 8494 if (dev->allmulti == 0) { 8495 /* 8496 * Avoid overflow. 8497 * If inc causes overflow, untouch allmulti and return error. 8498 */ 8499 if (inc < 0) 8500 dev->flags &= ~IFF_ALLMULTI; 8501 else { 8502 dev->allmulti -= inc; 8503 netdev_warn(dev, "allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n"); 8504 return -EOVERFLOW; 8505 } 8506 } 8507 if (dev->flags ^ old_flags) { 8508 netdev_info(dev, "%s allmulticast mode\n", 8509 dev->flags & IFF_ALLMULTI ? "entered" : "left"); 8510 dev_change_rx_flags(dev, IFF_ALLMULTI); 8511 dev_set_rx_mode(dev); 8512 if (notify) 8513 __dev_notify_flags(dev, old_flags, 8514 dev->gflags ^ old_gflags, 0, NULL); 8515 } 8516 return 0; 8517 } 8518 8519 /** 8520 * dev_set_allmulti - update allmulti count on a device 8521 * @dev: device 8522 * @inc: modifier 8523 * 8524 * Add or remove reception of all multicast frames to a device. While the 8525 * count in the device remains above zero the interface remains listening 8526 * to all interfaces. Once it hits zero the device reverts back to normal 8527 * filtering operation. A negative @inc value is used to drop the counter 8528 * when releasing a resource needing all multicasts. 8529 * Return 0 if successful or a negative errno code on error. 8530 */ 8531 8532 int dev_set_allmulti(struct net_device *dev, int inc) 8533 { 8534 return __dev_set_allmulti(dev, inc, true); 8535 } 8536 EXPORT_SYMBOL(dev_set_allmulti); 8537 8538 /* 8539 * Upload unicast and multicast address lists to device and 8540 * configure RX filtering. When the device doesn't support unicast 8541 * filtering it is put in promiscuous mode while unicast addresses 8542 * are present. 8543 */ 8544 void __dev_set_rx_mode(struct net_device *dev) 8545 { 8546 const struct net_device_ops *ops = dev->netdev_ops; 8547 8548 /* dev_open will call this function so the list will stay sane. */ 8549 if (!(dev->flags&IFF_UP)) 8550 return; 8551 8552 if (!netif_device_present(dev)) 8553 return; 8554 8555 if (!(dev->priv_flags & IFF_UNICAST_FLT)) { 8556 /* Unicast addresses changes may only happen under the rtnl, 8557 * therefore calling __dev_set_promiscuity here is safe. 8558 */ 8559 if (!netdev_uc_empty(dev) && !dev->uc_promisc) { 8560 __dev_set_promiscuity(dev, 1, false); 8561 dev->uc_promisc = true; 8562 } else if (netdev_uc_empty(dev) && dev->uc_promisc) { 8563 __dev_set_promiscuity(dev, -1, false); 8564 dev->uc_promisc = false; 8565 } 8566 } 8567 8568 if (ops->ndo_set_rx_mode) 8569 ops->ndo_set_rx_mode(dev); 8570 } 8571 8572 void dev_set_rx_mode(struct net_device *dev) 8573 { 8574 netif_addr_lock_bh(dev); 8575 __dev_set_rx_mode(dev); 8576 netif_addr_unlock_bh(dev); 8577 } 8578 8579 /** 8580 * dev_get_flags - get flags reported to userspace 8581 * @dev: device 8582 * 8583 * Get the combination of flag bits exported through APIs to userspace. 8584 */ 8585 unsigned int dev_get_flags(const struct net_device *dev) 8586 { 8587 unsigned int flags; 8588 8589 flags = (dev->flags & ~(IFF_PROMISC | 8590 IFF_ALLMULTI | 8591 IFF_RUNNING | 8592 IFF_LOWER_UP | 8593 IFF_DORMANT)) | 8594 (dev->gflags & (IFF_PROMISC | 8595 IFF_ALLMULTI)); 8596 8597 if (netif_running(dev)) { 8598 if (netif_oper_up(dev)) 8599 flags |= IFF_RUNNING; 8600 if (netif_carrier_ok(dev)) 8601 flags |= IFF_LOWER_UP; 8602 if (netif_dormant(dev)) 8603 flags |= IFF_DORMANT; 8604 } 8605 8606 return flags; 8607 } 8608 EXPORT_SYMBOL(dev_get_flags); 8609 8610 int __dev_change_flags(struct net_device *dev, unsigned int flags, 8611 struct netlink_ext_ack *extack) 8612 { 8613 unsigned int old_flags = dev->flags; 8614 int ret; 8615 8616 ASSERT_RTNL(); 8617 8618 /* 8619 * Set the flags on our device. 8620 */ 8621 8622 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | 8623 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | 8624 IFF_AUTOMEDIA)) | 8625 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | 8626 IFF_ALLMULTI)); 8627 8628 /* 8629 * Load in the correct multicast list now the flags have changed. 8630 */ 8631 8632 if ((old_flags ^ flags) & IFF_MULTICAST) 8633 dev_change_rx_flags(dev, IFF_MULTICAST); 8634 8635 dev_set_rx_mode(dev); 8636 8637 /* 8638 * Have we downed the interface. We handle IFF_UP ourselves 8639 * according to user attempts to set it, rather than blindly 8640 * setting it. 8641 */ 8642 8643 ret = 0; 8644 if ((old_flags ^ flags) & IFF_UP) { 8645 if (old_flags & IFF_UP) 8646 __dev_close(dev); 8647 else 8648 ret = __dev_open(dev, extack); 8649 } 8650 8651 if ((flags ^ dev->gflags) & IFF_PROMISC) { 8652 int inc = (flags & IFF_PROMISC) ? 1 : -1; 8653 unsigned int old_flags = dev->flags; 8654 8655 dev->gflags ^= IFF_PROMISC; 8656 8657 if (__dev_set_promiscuity(dev, inc, false) >= 0) 8658 if (dev->flags != old_flags) 8659 dev_set_rx_mode(dev); 8660 } 8661 8662 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI 8663 * is important. Some (broken) drivers set IFF_PROMISC, when 8664 * IFF_ALLMULTI is requested not asking us and not reporting. 8665 */ 8666 if ((flags ^ dev->gflags) & IFF_ALLMULTI) { 8667 int inc = (flags & IFF_ALLMULTI) ? 1 : -1; 8668 8669 dev->gflags ^= IFF_ALLMULTI; 8670 __dev_set_allmulti(dev, inc, false); 8671 } 8672 8673 return ret; 8674 } 8675 8676 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, 8677 unsigned int gchanges, u32 portid, 8678 const struct nlmsghdr *nlh) 8679 { 8680 unsigned int changes = dev->flags ^ old_flags; 8681 8682 if (gchanges) 8683 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC, portid, nlh); 8684 8685 if (changes & IFF_UP) { 8686 if (dev->flags & IFF_UP) 8687 call_netdevice_notifiers(NETDEV_UP, dev); 8688 else 8689 call_netdevice_notifiers(NETDEV_DOWN, dev); 8690 } 8691 8692 if (dev->flags & IFF_UP && 8693 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { 8694 struct netdev_notifier_change_info change_info = { 8695 .info = { 8696 .dev = dev, 8697 }, 8698 .flags_changed = changes, 8699 }; 8700 8701 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info); 8702 } 8703 } 8704 8705 /** 8706 * dev_change_flags - change device settings 8707 * @dev: device 8708 * @flags: device state flags 8709 * @extack: netlink extended ack 8710 * 8711 * Change settings on device based state flags. The flags are 8712 * in the userspace exported format. 8713 */ 8714 int dev_change_flags(struct net_device *dev, unsigned int flags, 8715 struct netlink_ext_ack *extack) 8716 { 8717 int ret; 8718 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; 8719 8720 ret = __dev_change_flags(dev, flags, extack); 8721 if (ret < 0) 8722 return ret; 8723 8724 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); 8725 __dev_notify_flags(dev, old_flags, changes, 0, NULL); 8726 return ret; 8727 } 8728 EXPORT_SYMBOL(dev_change_flags); 8729 8730 int __dev_set_mtu(struct net_device *dev, int new_mtu) 8731 { 8732 const struct net_device_ops *ops = dev->netdev_ops; 8733 8734 if (ops->ndo_change_mtu) 8735 return ops->ndo_change_mtu(dev, new_mtu); 8736 8737 /* Pairs with all the lockless reads of dev->mtu in the stack */ 8738 WRITE_ONCE(dev->mtu, new_mtu); 8739 return 0; 8740 } 8741 EXPORT_SYMBOL(__dev_set_mtu); 8742 8743 int dev_validate_mtu(struct net_device *dev, int new_mtu, 8744 struct netlink_ext_ack *extack) 8745 { 8746 /* MTU must be positive, and in range */ 8747 if (new_mtu < 0 || new_mtu < dev->min_mtu) { 8748 NL_SET_ERR_MSG(extack, "mtu less than device minimum"); 8749 return -EINVAL; 8750 } 8751 8752 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { 8753 NL_SET_ERR_MSG(extack, "mtu greater than device maximum"); 8754 return -EINVAL; 8755 } 8756 return 0; 8757 } 8758 8759 /** 8760 * dev_set_mtu_ext - Change maximum transfer unit 8761 * @dev: device 8762 * @new_mtu: new transfer unit 8763 * @extack: netlink extended ack 8764 * 8765 * Change the maximum transfer size of the network device. 8766 */ 8767 int dev_set_mtu_ext(struct net_device *dev, int new_mtu, 8768 struct netlink_ext_ack *extack) 8769 { 8770 int err, orig_mtu; 8771 8772 if (new_mtu == dev->mtu) 8773 return 0; 8774 8775 err = dev_validate_mtu(dev, new_mtu, extack); 8776 if (err) 8777 return err; 8778 8779 if (!netif_device_present(dev)) 8780 return -ENODEV; 8781 8782 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); 8783 err = notifier_to_errno(err); 8784 if (err) 8785 return err; 8786 8787 orig_mtu = dev->mtu; 8788 err = __dev_set_mtu(dev, new_mtu); 8789 8790 if (!err) { 8791 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8792 orig_mtu); 8793 err = notifier_to_errno(err); 8794 if (err) { 8795 /* setting mtu back and notifying everyone again, 8796 * so that they have a chance to revert changes. 8797 */ 8798 __dev_set_mtu(dev, orig_mtu); 8799 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8800 new_mtu); 8801 } 8802 } 8803 return err; 8804 } 8805 8806 int dev_set_mtu(struct net_device *dev, int new_mtu) 8807 { 8808 struct netlink_ext_ack extack; 8809 int err; 8810 8811 memset(&extack, 0, sizeof(extack)); 8812 err = dev_set_mtu_ext(dev, new_mtu, &extack); 8813 if (err && extack._msg) 8814 net_err_ratelimited("%s: %s\n", dev->name, extack._msg); 8815 return err; 8816 } 8817 EXPORT_SYMBOL(dev_set_mtu); 8818 8819 /** 8820 * dev_change_tx_queue_len - Change TX queue length of a netdevice 8821 * @dev: device 8822 * @new_len: new tx queue length 8823 */ 8824 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len) 8825 { 8826 unsigned int orig_len = dev->tx_queue_len; 8827 int res; 8828 8829 if (new_len != (unsigned int)new_len) 8830 return -ERANGE; 8831 8832 if (new_len != orig_len) { 8833 dev->tx_queue_len = new_len; 8834 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev); 8835 res = notifier_to_errno(res); 8836 if (res) 8837 goto err_rollback; 8838 res = dev_qdisc_change_tx_queue_len(dev); 8839 if (res) 8840 goto err_rollback; 8841 } 8842 8843 return 0; 8844 8845 err_rollback: 8846 netdev_err(dev, "refused to change device tx_queue_len\n"); 8847 dev->tx_queue_len = orig_len; 8848 return res; 8849 } 8850 8851 /** 8852 * dev_set_group - Change group this device belongs to 8853 * @dev: device 8854 * @new_group: group this device should belong to 8855 */ 8856 void dev_set_group(struct net_device *dev, int new_group) 8857 { 8858 dev->group = new_group; 8859 } 8860 8861 /** 8862 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR. 8863 * @dev: device 8864 * @addr: new address 8865 * @extack: netlink extended ack 8866 */ 8867 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr, 8868 struct netlink_ext_ack *extack) 8869 { 8870 struct netdev_notifier_pre_changeaddr_info info = { 8871 .info.dev = dev, 8872 .info.extack = extack, 8873 .dev_addr = addr, 8874 }; 8875 int rc; 8876 8877 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info); 8878 return notifier_to_errno(rc); 8879 } 8880 EXPORT_SYMBOL(dev_pre_changeaddr_notify); 8881 8882 /** 8883 * dev_set_mac_address - Change Media Access Control Address 8884 * @dev: device 8885 * @sa: new address 8886 * @extack: netlink extended ack 8887 * 8888 * Change the hardware (MAC) address of the device 8889 */ 8890 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa, 8891 struct netlink_ext_ack *extack) 8892 { 8893 const struct net_device_ops *ops = dev->netdev_ops; 8894 int err; 8895 8896 if (!ops->ndo_set_mac_address) 8897 return -EOPNOTSUPP; 8898 if (sa->sa_family != dev->type) 8899 return -EINVAL; 8900 if (!netif_device_present(dev)) 8901 return -ENODEV; 8902 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack); 8903 if (err) 8904 return err; 8905 if (memcmp(dev->dev_addr, sa->sa_data, dev->addr_len)) { 8906 err = ops->ndo_set_mac_address(dev, sa); 8907 if (err) 8908 return err; 8909 } 8910 dev->addr_assign_type = NET_ADDR_SET; 8911 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); 8912 add_device_randomness(dev->dev_addr, dev->addr_len); 8913 return 0; 8914 } 8915 EXPORT_SYMBOL(dev_set_mac_address); 8916 8917 static DECLARE_RWSEM(dev_addr_sem); 8918 8919 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa, 8920 struct netlink_ext_ack *extack) 8921 { 8922 int ret; 8923 8924 down_write(&dev_addr_sem); 8925 ret = dev_set_mac_address(dev, sa, extack); 8926 up_write(&dev_addr_sem); 8927 return ret; 8928 } 8929 EXPORT_SYMBOL(dev_set_mac_address_user); 8930 8931 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name) 8932 { 8933 size_t size = sizeof(sa->sa_data_min); 8934 struct net_device *dev; 8935 int ret = 0; 8936 8937 down_read(&dev_addr_sem); 8938 rcu_read_lock(); 8939 8940 dev = dev_get_by_name_rcu(net, dev_name); 8941 if (!dev) { 8942 ret = -ENODEV; 8943 goto unlock; 8944 } 8945 if (!dev->addr_len) 8946 memset(sa->sa_data, 0, size); 8947 else 8948 memcpy(sa->sa_data, dev->dev_addr, 8949 min_t(size_t, size, dev->addr_len)); 8950 sa->sa_family = dev->type; 8951 8952 unlock: 8953 rcu_read_unlock(); 8954 up_read(&dev_addr_sem); 8955 return ret; 8956 } 8957 EXPORT_SYMBOL(dev_get_mac_address); 8958 8959 /** 8960 * dev_change_carrier - Change device carrier 8961 * @dev: device 8962 * @new_carrier: new value 8963 * 8964 * Change device carrier 8965 */ 8966 int dev_change_carrier(struct net_device *dev, bool new_carrier) 8967 { 8968 const struct net_device_ops *ops = dev->netdev_ops; 8969 8970 if (!ops->ndo_change_carrier) 8971 return -EOPNOTSUPP; 8972 if (!netif_device_present(dev)) 8973 return -ENODEV; 8974 return ops->ndo_change_carrier(dev, new_carrier); 8975 } 8976 8977 /** 8978 * dev_get_phys_port_id - Get device physical port ID 8979 * @dev: device 8980 * @ppid: port ID 8981 * 8982 * Get device physical port ID 8983 */ 8984 int dev_get_phys_port_id(struct net_device *dev, 8985 struct netdev_phys_item_id *ppid) 8986 { 8987 const struct net_device_ops *ops = dev->netdev_ops; 8988 8989 if (!ops->ndo_get_phys_port_id) 8990 return -EOPNOTSUPP; 8991 return ops->ndo_get_phys_port_id(dev, ppid); 8992 } 8993 8994 /** 8995 * dev_get_phys_port_name - Get device physical port name 8996 * @dev: device 8997 * @name: port name 8998 * @len: limit of bytes to copy to name 8999 * 9000 * Get device physical port name 9001 */ 9002 int dev_get_phys_port_name(struct net_device *dev, 9003 char *name, size_t len) 9004 { 9005 const struct net_device_ops *ops = dev->netdev_ops; 9006 int err; 9007 9008 if (ops->ndo_get_phys_port_name) { 9009 err = ops->ndo_get_phys_port_name(dev, name, len); 9010 if (err != -EOPNOTSUPP) 9011 return err; 9012 } 9013 return devlink_compat_phys_port_name_get(dev, name, len); 9014 } 9015 9016 /** 9017 * dev_get_port_parent_id - Get the device's port parent identifier 9018 * @dev: network device 9019 * @ppid: pointer to a storage for the port's parent identifier 9020 * @recurse: allow/disallow recursion to lower devices 9021 * 9022 * Get the devices's port parent identifier 9023 */ 9024 int dev_get_port_parent_id(struct net_device *dev, 9025 struct netdev_phys_item_id *ppid, 9026 bool recurse) 9027 { 9028 const struct net_device_ops *ops = dev->netdev_ops; 9029 struct netdev_phys_item_id first = { }; 9030 struct net_device *lower_dev; 9031 struct list_head *iter; 9032 int err; 9033 9034 if (ops->ndo_get_port_parent_id) { 9035 err = ops->ndo_get_port_parent_id(dev, ppid); 9036 if (err != -EOPNOTSUPP) 9037 return err; 9038 } 9039 9040 err = devlink_compat_switch_id_get(dev, ppid); 9041 if (!recurse || err != -EOPNOTSUPP) 9042 return err; 9043 9044 netdev_for_each_lower_dev(dev, lower_dev, iter) { 9045 err = dev_get_port_parent_id(lower_dev, ppid, true); 9046 if (err) 9047 break; 9048 if (!first.id_len) 9049 first = *ppid; 9050 else if (memcmp(&first, ppid, sizeof(*ppid))) 9051 return -EOPNOTSUPP; 9052 } 9053 9054 return err; 9055 } 9056 EXPORT_SYMBOL(dev_get_port_parent_id); 9057 9058 /** 9059 * netdev_port_same_parent_id - Indicate if two network devices have 9060 * the same port parent identifier 9061 * @a: first network device 9062 * @b: second network device 9063 */ 9064 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b) 9065 { 9066 struct netdev_phys_item_id a_id = { }; 9067 struct netdev_phys_item_id b_id = { }; 9068 9069 if (dev_get_port_parent_id(a, &a_id, true) || 9070 dev_get_port_parent_id(b, &b_id, true)) 9071 return false; 9072 9073 return netdev_phys_item_id_same(&a_id, &b_id); 9074 } 9075 EXPORT_SYMBOL(netdev_port_same_parent_id); 9076 9077 static void netdev_dpll_pin_assign(struct net_device *dev, struct dpll_pin *dpll_pin) 9078 { 9079 #if IS_ENABLED(CONFIG_DPLL) 9080 rtnl_lock(); 9081 dev->dpll_pin = dpll_pin; 9082 rtnl_unlock(); 9083 #endif 9084 } 9085 9086 void netdev_dpll_pin_set(struct net_device *dev, struct dpll_pin *dpll_pin) 9087 { 9088 WARN_ON(!dpll_pin); 9089 netdev_dpll_pin_assign(dev, dpll_pin); 9090 } 9091 EXPORT_SYMBOL(netdev_dpll_pin_set); 9092 9093 void netdev_dpll_pin_clear(struct net_device *dev) 9094 { 9095 netdev_dpll_pin_assign(dev, NULL); 9096 } 9097 EXPORT_SYMBOL(netdev_dpll_pin_clear); 9098 9099 /** 9100 * dev_change_proto_down - set carrier according to proto_down. 9101 * 9102 * @dev: device 9103 * @proto_down: new value 9104 */ 9105 int dev_change_proto_down(struct net_device *dev, bool proto_down) 9106 { 9107 if (!(dev->priv_flags & IFF_CHANGE_PROTO_DOWN)) 9108 return -EOPNOTSUPP; 9109 if (!netif_device_present(dev)) 9110 return -ENODEV; 9111 if (proto_down) 9112 netif_carrier_off(dev); 9113 else 9114 netif_carrier_on(dev); 9115 dev->proto_down = proto_down; 9116 return 0; 9117 } 9118 9119 /** 9120 * dev_change_proto_down_reason - proto down reason 9121 * 9122 * @dev: device 9123 * @mask: proto down mask 9124 * @value: proto down value 9125 */ 9126 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask, 9127 u32 value) 9128 { 9129 int b; 9130 9131 if (!mask) { 9132 dev->proto_down_reason = value; 9133 } else { 9134 for_each_set_bit(b, &mask, 32) { 9135 if (value & (1 << b)) 9136 dev->proto_down_reason |= BIT(b); 9137 else 9138 dev->proto_down_reason &= ~BIT(b); 9139 } 9140 } 9141 } 9142 9143 struct bpf_xdp_link { 9144 struct bpf_link link; 9145 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */ 9146 int flags; 9147 }; 9148 9149 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags) 9150 { 9151 if (flags & XDP_FLAGS_HW_MODE) 9152 return XDP_MODE_HW; 9153 if (flags & XDP_FLAGS_DRV_MODE) 9154 return XDP_MODE_DRV; 9155 if (flags & XDP_FLAGS_SKB_MODE) 9156 return XDP_MODE_SKB; 9157 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB; 9158 } 9159 9160 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode) 9161 { 9162 switch (mode) { 9163 case XDP_MODE_SKB: 9164 return generic_xdp_install; 9165 case XDP_MODE_DRV: 9166 case XDP_MODE_HW: 9167 return dev->netdev_ops->ndo_bpf; 9168 default: 9169 return NULL; 9170 } 9171 } 9172 9173 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev, 9174 enum bpf_xdp_mode mode) 9175 { 9176 return dev->xdp_state[mode].link; 9177 } 9178 9179 static struct bpf_prog *dev_xdp_prog(struct net_device *dev, 9180 enum bpf_xdp_mode mode) 9181 { 9182 struct bpf_xdp_link *link = dev_xdp_link(dev, mode); 9183 9184 if (link) 9185 return link->link.prog; 9186 return dev->xdp_state[mode].prog; 9187 } 9188 9189 u8 dev_xdp_prog_count(struct net_device *dev) 9190 { 9191 u8 count = 0; 9192 int i; 9193 9194 for (i = 0; i < __MAX_XDP_MODE; i++) 9195 if (dev->xdp_state[i].prog || dev->xdp_state[i].link) 9196 count++; 9197 return count; 9198 } 9199 EXPORT_SYMBOL_GPL(dev_xdp_prog_count); 9200 9201 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode) 9202 { 9203 struct bpf_prog *prog = dev_xdp_prog(dev, mode); 9204 9205 return prog ? prog->aux->id : 0; 9206 } 9207 9208 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode, 9209 struct bpf_xdp_link *link) 9210 { 9211 dev->xdp_state[mode].link = link; 9212 dev->xdp_state[mode].prog = NULL; 9213 } 9214 9215 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode, 9216 struct bpf_prog *prog) 9217 { 9218 dev->xdp_state[mode].link = NULL; 9219 dev->xdp_state[mode].prog = prog; 9220 } 9221 9222 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode, 9223 bpf_op_t bpf_op, struct netlink_ext_ack *extack, 9224 u32 flags, struct bpf_prog *prog) 9225 { 9226 struct netdev_bpf xdp; 9227 int err; 9228 9229 memset(&xdp, 0, sizeof(xdp)); 9230 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG; 9231 xdp.extack = extack; 9232 xdp.flags = flags; 9233 xdp.prog = prog; 9234 9235 /* Drivers assume refcnt is already incremented (i.e, prog pointer is 9236 * "moved" into driver), so they don't increment it on their own, but 9237 * they do decrement refcnt when program is detached or replaced. 9238 * Given net_device also owns link/prog, we need to bump refcnt here 9239 * to prevent drivers from underflowing it. 9240 */ 9241 if (prog) 9242 bpf_prog_inc(prog); 9243 err = bpf_op(dev, &xdp); 9244 if (err) { 9245 if (prog) 9246 bpf_prog_put(prog); 9247 return err; 9248 } 9249 9250 if (mode != XDP_MODE_HW) 9251 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog); 9252 9253 return 0; 9254 } 9255 9256 static void dev_xdp_uninstall(struct net_device *dev) 9257 { 9258 struct bpf_xdp_link *link; 9259 struct bpf_prog *prog; 9260 enum bpf_xdp_mode mode; 9261 bpf_op_t bpf_op; 9262 9263 ASSERT_RTNL(); 9264 9265 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) { 9266 prog = dev_xdp_prog(dev, mode); 9267 if (!prog) 9268 continue; 9269 9270 bpf_op = dev_xdp_bpf_op(dev, mode); 9271 if (!bpf_op) 9272 continue; 9273 9274 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9275 9276 /* auto-detach link from net device */ 9277 link = dev_xdp_link(dev, mode); 9278 if (link) 9279 link->dev = NULL; 9280 else 9281 bpf_prog_put(prog); 9282 9283 dev_xdp_set_link(dev, mode, NULL); 9284 } 9285 } 9286 9287 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack, 9288 struct bpf_xdp_link *link, struct bpf_prog *new_prog, 9289 struct bpf_prog *old_prog, u32 flags) 9290 { 9291 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES); 9292 struct bpf_prog *cur_prog; 9293 struct net_device *upper; 9294 struct list_head *iter; 9295 enum bpf_xdp_mode mode; 9296 bpf_op_t bpf_op; 9297 int err; 9298 9299 ASSERT_RTNL(); 9300 9301 /* either link or prog attachment, never both */ 9302 if (link && (new_prog || old_prog)) 9303 return -EINVAL; 9304 /* link supports only XDP mode flags */ 9305 if (link && (flags & ~XDP_FLAGS_MODES)) { 9306 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment"); 9307 return -EINVAL; 9308 } 9309 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */ 9310 if (num_modes > 1) { 9311 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set"); 9312 return -EINVAL; 9313 } 9314 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */ 9315 if (!num_modes && dev_xdp_prog_count(dev) > 1) { 9316 NL_SET_ERR_MSG(extack, 9317 "More than one program loaded, unset mode is ambiguous"); 9318 return -EINVAL; 9319 } 9320 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */ 9321 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) { 9322 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified"); 9323 return -EINVAL; 9324 } 9325 9326 mode = dev_xdp_mode(dev, flags); 9327 /* can't replace attached link */ 9328 if (dev_xdp_link(dev, mode)) { 9329 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link"); 9330 return -EBUSY; 9331 } 9332 9333 /* don't allow if an upper device already has a program */ 9334 netdev_for_each_upper_dev_rcu(dev, upper, iter) { 9335 if (dev_xdp_prog_count(upper) > 0) { 9336 NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program"); 9337 return -EEXIST; 9338 } 9339 } 9340 9341 cur_prog = dev_xdp_prog(dev, mode); 9342 /* can't replace attached prog with link */ 9343 if (link && cur_prog) { 9344 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link"); 9345 return -EBUSY; 9346 } 9347 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) { 9348 NL_SET_ERR_MSG(extack, "Active program does not match expected"); 9349 return -EEXIST; 9350 } 9351 9352 /* put effective new program into new_prog */ 9353 if (link) 9354 new_prog = link->link.prog; 9355 9356 if (new_prog) { 9357 bool offload = mode == XDP_MODE_HW; 9358 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB 9359 ? XDP_MODE_DRV : XDP_MODE_SKB; 9360 9361 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) { 9362 NL_SET_ERR_MSG(extack, "XDP program already attached"); 9363 return -EBUSY; 9364 } 9365 if (!offload && dev_xdp_prog(dev, other_mode)) { 9366 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time"); 9367 return -EEXIST; 9368 } 9369 if (!offload && bpf_prog_is_offloaded(new_prog->aux)) { 9370 NL_SET_ERR_MSG(extack, "Using offloaded program without HW_MODE flag is not supported"); 9371 return -EINVAL; 9372 } 9373 if (bpf_prog_is_dev_bound(new_prog->aux) && !bpf_offload_dev_match(new_prog, dev)) { 9374 NL_SET_ERR_MSG(extack, "Program bound to different device"); 9375 return -EINVAL; 9376 } 9377 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) { 9378 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device"); 9379 return -EINVAL; 9380 } 9381 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) { 9382 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device"); 9383 return -EINVAL; 9384 } 9385 } 9386 9387 /* don't call drivers if the effective program didn't change */ 9388 if (new_prog != cur_prog) { 9389 bpf_op = dev_xdp_bpf_op(dev, mode); 9390 if (!bpf_op) { 9391 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode"); 9392 return -EOPNOTSUPP; 9393 } 9394 9395 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog); 9396 if (err) 9397 return err; 9398 } 9399 9400 if (link) 9401 dev_xdp_set_link(dev, mode, link); 9402 else 9403 dev_xdp_set_prog(dev, mode, new_prog); 9404 if (cur_prog) 9405 bpf_prog_put(cur_prog); 9406 9407 return 0; 9408 } 9409 9410 static int dev_xdp_attach_link(struct net_device *dev, 9411 struct netlink_ext_ack *extack, 9412 struct bpf_xdp_link *link) 9413 { 9414 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags); 9415 } 9416 9417 static int dev_xdp_detach_link(struct net_device *dev, 9418 struct netlink_ext_ack *extack, 9419 struct bpf_xdp_link *link) 9420 { 9421 enum bpf_xdp_mode mode; 9422 bpf_op_t bpf_op; 9423 9424 ASSERT_RTNL(); 9425 9426 mode = dev_xdp_mode(dev, link->flags); 9427 if (dev_xdp_link(dev, mode) != link) 9428 return -EINVAL; 9429 9430 bpf_op = dev_xdp_bpf_op(dev, mode); 9431 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9432 dev_xdp_set_link(dev, mode, NULL); 9433 return 0; 9434 } 9435 9436 static void bpf_xdp_link_release(struct bpf_link *link) 9437 { 9438 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9439 9440 rtnl_lock(); 9441 9442 /* if racing with net_device's tear down, xdp_link->dev might be 9443 * already NULL, in which case link was already auto-detached 9444 */ 9445 if (xdp_link->dev) { 9446 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link)); 9447 xdp_link->dev = NULL; 9448 } 9449 9450 rtnl_unlock(); 9451 } 9452 9453 static int bpf_xdp_link_detach(struct bpf_link *link) 9454 { 9455 bpf_xdp_link_release(link); 9456 return 0; 9457 } 9458 9459 static void bpf_xdp_link_dealloc(struct bpf_link *link) 9460 { 9461 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9462 9463 kfree(xdp_link); 9464 } 9465 9466 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link, 9467 struct seq_file *seq) 9468 { 9469 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9470 u32 ifindex = 0; 9471 9472 rtnl_lock(); 9473 if (xdp_link->dev) 9474 ifindex = xdp_link->dev->ifindex; 9475 rtnl_unlock(); 9476 9477 seq_printf(seq, "ifindex:\t%u\n", ifindex); 9478 } 9479 9480 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link, 9481 struct bpf_link_info *info) 9482 { 9483 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9484 u32 ifindex = 0; 9485 9486 rtnl_lock(); 9487 if (xdp_link->dev) 9488 ifindex = xdp_link->dev->ifindex; 9489 rtnl_unlock(); 9490 9491 info->xdp.ifindex = ifindex; 9492 return 0; 9493 } 9494 9495 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog, 9496 struct bpf_prog *old_prog) 9497 { 9498 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9499 enum bpf_xdp_mode mode; 9500 bpf_op_t bpf_op; 9501 int err = 0; 9502 9503 rtnl_lock(); 9504 9505 /* link might have been auto-released already, so fail */ 9506 if (!xdp_link->dev) { 9507 err = -ENOLINK; 9508 goto out_unlock; 9509 } 9510 9511 if (old_prog && link->prog != old_prog) { 9512 err = -EPERM; 9513 goto out_unlock; 9514 } 9515 old_prog = link->prog; 9516 if (old_prog->type != new_prog->type || 9517 old_prog->expected_attach_type != new_prog->expected_attach_type) { 9518 err = -EINVAL; 9519 goto out_unlock; 9520 } 9521 9522 if (old_prog == new_prog) { 9523 /* no-op, don't disturb drivers */ 9524 bpf_prog_put(new_prog); 9525 goto out_unlock; 9526 } 9527 9528 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags); 9529 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode); 9530 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL, 9531 xdp_link->flags, new_prog); 9532 if (err) 9533 goto out_unlock; 9534 9535 old_prog = xchg(&link->prog, new_prog); 9536 bpf_prog_put(old_prog); 9537 9538 out_unlock: 9539 rtnl_unlock(); 9540 return err; 9541 } 9542 9543 static const struct bpf_link_ops bpf_xdp_link_lops = { 9544 .release = bpf_xdp_link_release, 9545 .dealloc = bpf_xdp_link_dealloc, 9546 .detach = bpf_xdp_link_detach, 9547 .show_fdinfo = bpf_xdp_link_show_fdinfo, 9548 .fill_link_info = bpf_xdp_link_fill_link_info, 9549 .update_prog = bpf_xdp_link_update, 9550 }; 9551 9552 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog) 9553 { 9554 struct net *net = current->nsproxy->net_ns; 9555 struct bpf_link_primer link_primer; 9556 struct netlink_ext_ack extack = {}; 9557 struct bpf_xdp_link *link; 9558 struct net_device *dev; 9559 int err, fd; 9560 9561 rtnl_lock(); 9562 dev = dev_get_by_index(net, attr->link_create.target_ifindex); 9563 if (!dev) { 9564 rtnl_unlock(); 9565 return -EINVAL; 9566 } 9567 9568 link = kzalloc(sizeof(*link), GFP_USER); 9569 if (!link) { 9570 err = -ENOMEM; 9571 goto unlock; 9572 } 9573 9574 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog); 9575 link->dev = dev; 9576 link->flags = attr->link_create.flags; 9577 9578 err = bpf_link_prime(&link->link, &link_primer); 9579 if (err) { 9580 kfree(link); 9581 goto unlock; 9582 } 9583 9584 err = dev_xdp_attach_link(dev, &extack, link); 9585 rtnl_unlock(); 9586 9587 if (err) { 9588 link->dev = NULL; 9589 bpf_link_cleanup(&link_primer); 9590 trace_bpf_xdp_link_attach_failed(extack._msg); 9591 goto out_put_dev; 9592 } 9593 9594 fd = bpf_link_settle(&link_primer); 9595 /* link itself doesn't hold dev's refcnt to not complicate shutdown */ 9596 dev_put(dev); 9597 return fd; 9598 9599 unlock: 9600 rtnl_unlock(); 9601 9602 out_put_dev: 9603 dev_put(dev); 9604 return err; 9605 } 9606 9607 /** 9608 * dev_change_xdp_fd - set or clear a bpf program for a device rx path 9609 * @dev: device 9610 * @extack: netlink extended ack 9611 * @fd: new program fd or negative value to clear 9612 * @expected_fd: old program fd that userspace expects to replace or clear 9613 * @flags: xdp-related flags 9614 * 9615 * Set or clear a bpf program for a device 9616 */ 9617 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, 9618 int fd, int expected_fd, u32 flags) 9619 { 9620 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags); 9621 struct bpf_prog *new_prog = NULL, *old_prog = NULL; 9622 int err; 9623 9624 ASSERT_RTNL(); 9625 9626 if (fd >= 0) { 9627 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP, 9628 mode != XDP_MODE_SKB); 9629 if (IS_ERR(new_prog)) 9630 return PTR_ERR(new_prog); 9631 } 9632 9633 if (expected_fd >= 0) { 9634 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP, 9635 mode != XDP_MODE_SKB); 9636 if (IS_ERR(old_prog)) { 9637 err = PTR_ERR(old_prog); 9638 old_prog = NULL; 9639 goto err_out; 9640 } 9641 } 9642 9643 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags); 9644 9645 err_out: 9646 if (err && new_prog) 9647 bpf_prog_put(new_prog); 9648 if (old_prog) 9649 bpf_prog_put(old_prog); 9650 return err; 9651 } 9652 9653 /** 9654 * dev_index_reserve() - allocate an ifindex in a namespace 9655 * @net: the applicable net namespace 9656 * @ifindex: requested ifindex, pass %0 to get one allocated 9657 * 9658 * Allocate a ifindex for a new device. Caller must either use the ifindex 9659 * to store the device (via list_netdevice()) or call dev_index_release() 9660 * to give the index up. 9661 * 9662 * Return: a suitable unique value for a new device interface number or -errno. 9663 */ 9664 static int dev_index_reserve(struct net *net, u32 ifindex) 9665 { 9666 int err; 9667 9668 if (ifindex > INT_MAX) { 9669 DEBUG_NET_WARN_ON_ONCE(1); 9670 return -EINVAL; 9671 } 9672 9673 if (!ifindex) 9674 err = xa_alloc_cyclic(&net->dev_by_index, &ifindex, NULL, 9675 xa_limit_31b, &net->ifindex, GFP_KERNEL); 9676 else 9677 err = xa_insert(&net->dev_by_index, ifindex, NULL, GFP_KERNEL); 9678 if (err < 0) 9679 return err; 9680 9681 return ifindex; 9682 } 9683 9684 static void dev_index_release(struct net *net, int ifindex) 9685 { 9686 /* Expect only unused indexes, unlist_netdevice() removes the used */ 9687 WARN_ON(xa_erase(&net->dev_by_index, ifindex)); 9688 } 9689 9690 /* Delayed registration/unregisteration */ 9691 LIST_HEAD(net_todo_list); 9692 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); 9693 9694 static void net_set_todo(struct net_device *dev) 9695 { 9696 list_add_tail(&dev->todo_list, &net_todo_list); 9697 atomic_inc(&dev_net(dev)->dev_unreg_count); 9698 } 9699 9700 static netdev_features_t netdev_sync_upper_features(struct net_device *lower, 9701 struct net_device *upper, netdev_features_t features) 9702 { 9703 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9704 netdev_features_t feature; 9705 int feature_bit; 9706 9707 for_each_netdev_feature(upper_disables, feature_bit) { 9708 feature = __NETIF_F_BIT(feature_bit); 9709 if (!(upper->wanted_features & feature) 9710 && (features & feature)) { 9711 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", 9712 &feature, upper->name); 9713 features &= ~feature; 9714 } 9715 } 9716 9717 return features; 9718 } 9719 9720 static void netdev_sync_lower_features(struct net_device *upper, 9721 struct net_device *lower, netdev_features_t features) 9722 { 9723 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9724 netdev_features_t feature; 9725 int feature_bit; 9726 9727 for_each_netdev_feature(upper_disables, feature_bit) { 9728 feature = __NETIF_F_BIT(feature_bit); 9729 if (!(features & feature) && (lower->features & feature)) { 9730 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", 9731 &feature, lower->name); 9732 lower->wanted_features &= ~feature; 9733 __netdev_update_features(lower); 9734 9735 if (unlikely(lower->features & feature)) 9736 netdev_WARN(upper, "failed to disable %pNF on %s!\n", 9737 &feature, lower->name); 9738 else 9739 netdev_features_change(lower); 9740 } 9741 } 9742 } 9743 9744 static netdev_features_t netdev_fix_features(struct net_device *dev, 9745 netdev_features_t features) 9746 { 9747 /* Fix illegal checksum combinations */ 9748 if ((features & NETIF_F_HW_CSUM) && 9749 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { 9750 netdev_warn(dev, "mixed HW and IP checksum settings.\n"); 9751 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); 9752 } 9753 9754 /* TSO requires that SG is present as well. */ 9755 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { 9756 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); 9757 features &= ~NETIF_F_ALL_TSO; 9758 } 9759 9760 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && 9761 !(features & NETIF_F_IP_CSUM)) { 9762 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); 9763 features &= ~NETIF_F_TSO; 9764 features &= ~NETIF_F_TSO_ECN; 9765 } 9766 9767 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && 9768 !(features & NETIF_F_IPV6_CSUM)) { 9769 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); 9770 features &= ~NETIF_F_TSO6; 9771 } 9772 9773 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ 9774 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) 9775 features &= ~NETIF_F_TSO_MANGLEID; 9776 9777 /* TSO ECN requires that TSO is present as well. */ 9778 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) 9779 features &= ~NETIF_F_TSO_ECN; 9780 9781 /* Software GSO depends on SG. */ 9782 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { 9783 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); 9784 features &= ~NETIF_F_GSO; 9785 } 9786 9787 /* GSO partial features require GSO partial be set */ 9788 if ((features & dev->gso_partial_features) && 9789 !(features & NETIF_F_GSO_PARTIAL)) { 9790 netdev_dbg(dev, 9791 "Dropping partially supported GSO features since no GSO partial.\n"); 9792 features &= ~dev->gso_partial_features; 9793 } 9794 9795 if (!(features & NETIF_F_RXCSUM)) { 9796 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet 9797 * successfully merged by hardware must also have the 9798 * checksum verified by hardware. If the user does not 9799 * want to enable RXCSUM, logically, we should disable GRO_HW. 9800 */ 9801 if (features & NETIF_F_GRO_HW) { 9802 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n"); 9803 features &= ~NETIF_F_GRO_HW; 9804 } 9805 } 9806 9807 /* LRO/HW-GRO features cannot be combined with RX-FCS */ 9808 if (features & NETIF_F_RXFCS) { 9809 if (features & NETIF_F_LRO) { 9810 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n"); 9811 features &= ~NETIF_F_LRO; 9812 } 9813 9814 if (features & NETIF_F_GRO_HW) { 9815 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n"); 9816 features &= ~NETIF_F_GRO_HW; 9817 } 9818 } 9819 9820 if ((features & NETIF_F_GRO_HW) && (features & NETIF_F_LRO)) { 9821 netdev_dbg(dev, "Dropping LRO feature since HW-GRO is requested.\n"); 9822 features &= ~NETIF_F_LRO; 9823 } 9824 9825 if (features & NETIF_F_HW_TLS_TX) { 9826 bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) == 9827 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM); 9828 bool hw_csum = features & NETIF_F_HW_CSUM; 9829 9830 if (!ip_csum && !hw_csum) { 9831 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n"); 9832 features &= ~NETIF_F_HW_TLS_TX; 9833 } 9834 } 9835 9836 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) { 9837 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n"); 9838 features &= ~NETIF_F_HW_TLS_RX; 9839 } 9840 9841 return features; 9842 } 9843 9844 int __netdev_update_features(struct net_device *dev) 9845 { 9846 struct net_device *upper, *lower; 9847 netdev_features_t features; 9848 struct list_head *iter; 9849 int err = -1; 9850 9851 ASSERT_RTNL(); 9852 9853 features = netdev_get_wanted_features(dev); 9854 9855 if (dev->netdev_ops->ndo_fix_features) 9856 features = dev->netdev_ops->ndo_fix_features(dev, features); 9857 9858 /* driver might be less strict about feature dependencies */ 9859 features = netdev_fix_features(dev, features); 9860 9861 /* some features can't be enabled if they're off on an upper device */ 9862 netdev_for_each_upper_dev_rcu(dev, upper, iter) 9863 features = netdev_sync_upper_features(dev, upper, features); 9864 9865 if (dev->features == features) 9866 goto sync_lower; 9867 9868 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", 9869 &dev->features, &features); 9870 9871 if (dev->netdev_ops->ndo_set_features) 9872 err = dev->netdev_ops->ndo_set_features(dev, features); 9873 else 9874 err = 0; 9875 9876 if (unlikely(err < 0)) { 9877 netdev_err(dev, 9878 "set_features() failed (%d); wanted %pNF, left %pNF\n", 9879 err, &features, &dev->features); 9880 /* return non-0 since some features might have changed and 9881 * it's better to fire a spurious notification than miss it 9882 */ 9883 return -1; 9884 } 9885 9886 sync_lower: 9887 /* some features must be disabled on lower devices when disabled 9888 * on an upper device (think: bonding master or bridge) 9889 */ 9890 netdev_for_each_lower_dev(dev, lower, iter) 9891 netdev_sync_lower_features(dev, lower, features); 9892 9893 if (!err) { 9894 netdev_features_t diff = features ^ dev->features; 9895 9896 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) { 9897 /* udp_tunnel_{get,drop}_rx_info both need 9898 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the 9899 * device, or they won't do anything. 9900 * Thus we need to update dev->features 9901 * *before* calling udp_tunnel_get_rx_info, 9902 * but *after* calling udp_tunnel_drop_rx_info. 9903 */ 9904 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) { 9905 dev->features = features; 9906 udp_tunnel_get_rx_info(dev); 9907 } else { 9908 udp_tunnel_drop_rx_info(dev); 9909 } 9910 } 9911 9912 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) { 9913 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) { 9914 dev->features = features; 9915 err |= vlan_get_rx_ctag_filter_info(dev); 9916 } else { 9917 vlan_drop_rx_ctag_filter_info(dev); 9918 } 9919 } 9920 9921 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) { 9922 if (features & NETIF_F_HW_VLAN_STAG_FILTER) { 9923 dev->features = features; 9924 err |= vlan_get_rx_stag_filter_info(dev); 9925 } else { 9926 vlan_drop_rx_stag_filter_info(dev); 9927 } 9928 } 9929 9930 dev->features = features; 9931 } 9932 9933 return err < 0 ? 0 : 1; 9934 } 9935 9936 /** 9937 * netdev_update_features - recalculate device features 9938 * @dev: the device to check 9939 * 9940 * Recalculate dev->features set and send notifications if it 9941 * has changed. Should be called after driver or hardware dependent 9942 * conditions might have changed that influence the features. 9943 */ 9944 void netdev_update_features(struct net_device *dev) 9945 { 9946 if (__netdev_update_features(dev)) 9947 netdev_features_change(dev); 9948 } 9949 EXPORT_SYMBOL(netdev_update_features); 9950 9951 /** 9952 * netdev_change_features - recalculate device features 9953 * @dev: the device to check 9954 * 9955 * Recalculate dev->features set and send notifications even 9956 * if they have not changed. Should be called instead of 9957 * netdev_update_features() if also dev->vlan_features might 9958 * have changed to allow the changes to be propagated to stacked 9959 * VLAN devices. 9960 */ 9961 void netdev_change_features(struct net_device *dev) 9962 { 9963 __netdev_update_features(dev); 9964 netdev_features_change(dev); 9965 } 9966 EXPORT_SYMBOL(netdev_change_features); 9967 9968 /** 9969 * netif_stacked_transfer_operstate - transfer operstate 9970 * @rootdev: the root or lower level device to transfer state from 9971 * @dev: the device to transfer operstate to 9972 * 9973 * Transfer operational state from root to device. This is normally 9974 * called when a stacking relationship exists between the root 9975 * device and the device(a leaf device). 9976 */ 9977 void netif_stacked_transfer_operstate(const struct net_device *rootdev, 9978 struct net_device *dev) 9979 { 9980 if (rootdev->operstate == IF_OPER_DORMANT) 9981 netif_dormant_on(dev); 9982 else 9983 netif_dormant_off(dev); 9984 9985 if (rootdev->operstate == IF_OPER_TESTING) 9986 netif_testing_on(dev); 9987 else 9988 netif_testing_off(dev); 9989 9990 if (netif_carrier_ok(rootdev)) 9991 netif_carrier_on(dev); 9992 else 9993 netif_carrier_off(dev); 9994 } 9995 EXPORT_SYMBOL(netif_stacked_transfer_operstate); 9996 9997 static int netif_alloc_rx_queues(struct net_device *dev) 9998 { 9999 unsigned int i, count = dev->num_rx_queues; 10000 struct netdev_rx_queue *rx; 10001 size_t sz = count * sizeof(*rx); 10002 int err = 0; 10003 10004 BUG_ON(count < 1); 10005 10006 rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10007 if (!rx) 10008 return -ENOMEM; 10009 10010 dev->_rx = rx; 10011 10012 for (i = 0; i < count; i++) { 10013 rx[i].dev = dev; 10014 10015 /* XDP RX-queue setup */ 10016 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0); 10017 if (err < 0) 10018 goto err_rxq_info; 10019 } 10020 return 0; 10021 10022 err_rxq_info: 10023 /* Rollback successful reg's and free other resources */ 10024 while (i--) 10025 xdp_rxq_info_unreg(&rx[i].xdp_rxq); 10026 kvfree(dev->_rx); 10027 dev->_rx = NULL; 10028 return err; 10029 } 10030 10031 static void netif_free_rx_queues(struct net_device *dev) 10032 { 10033 unsigned int i, count = dev->num_rx_queues; 10034 10035 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */ 10036 if (!dev->_rx) 10037 return; 10038 10039 for (i = 0; i < count; i++) 10040 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq); 10041 10042 kvfree(dev->_rx); 10043 } 10044 10045 static void netdev_init_one_queue(struct net_device *dev, 10046 struct netdev_queue *queue, void *_unused) 10047 { 10048 /* Initialize queue lock */ 10049 spin_lock_init(&queue->_xmit_lock); 10050 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); 10051 queue->xmit_lock_owner = -1; 10052 netdev_queue_numa_node_write(queue, NUMA_NO_NODE); 10053 queue->dev = dev; 10054 #ifdef CONFIG_BQL 10055 dql_init(&queue->dql, HZ); 10056 #endif 10057 } 10058 10059 static void netif_free_tx_queues(struct net_device *dev) 10060 { 10061 kvfree(dev->_tx); 10062 } 10063 10064 static int netif_alloc_netdev_queues(struct net_device *dev) 10065 { 10066 unsigned int count = dev->num_tx_queues; 10067 struct netdev_queue *tx; 10068 size_t sz = count * sizeof(*tx); 10069 10070 if (count < 1 || count > 0xffff) 10071 return -EINVAL; 10072 10073 tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10074 if (!tx) 10075 return -ENOMEM; 10076 10077 dev->_tx = tx; 10078 10079 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); 10080 spin_lock_init(&dev->tx_global_lock); 10081 10082 return 0; 10083 } 10084 10085 void netif_tx_stop_all_queues(struct net_device *dev) 10086 { 10087 unsigned int i; 10088 10089 for (i = 0; i < dev->num_tx_queues; i++) { 10090 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 10091 10092 netif_tx_stop_queue(txq); 10093 } 10094 } 10095 EXPORT_SYMBOL(netif_tx_stop_all_queues); 10096 10097 static int netdev_do_alloc_pcpu_stats(struct net_device *dev) 10098 { 10099 void __percpu *v; 10100 10101 /* Drivers implementing ndo_get_peer_dev must support tstat 10102 * accounting, so that skb_do_redirect() can bump the dev's 10103 * RX stats upon network namespace switch. 10104 */ 10105 if (dev->netdev_ops->ndo_get_peer_dev && 10106 dev->pcpu_stat_type != NETDEV_PCPU_STAT_TSTATS) 10107 return -EOPNOTSUPP; 10108 10109 switch (dev->pcpu_stat_type) { 10110 case NETDEV_PCPU_STAT_NONE: 10111 return 0; 10112 case NETDEV_PCPU_STAT_LSTATS: 10113 v = dev->lstats = netdev_alloc_pcpu_stats(struct pcpu_lstats); 10114 break; 10115 case NETDEV_PCPU_STAT_TSTATS: 10116 v = dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats); 10117 break; 10118 case NETDEV_PCPU_STAT_DSTATS: 10119 v = dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats); 10120 break; 10121 default: 10122 return -EINVAL; 10123 } 10124 10125 return v ? 0 : -ENOMEM; 10126 } 10127 10128 static void netdev_do_free_pcpu_stats(struct net_device *dev) 10129 { 10130 switch (dev->pcpu_stat_type) { 10131 case NETDEV_PCPU_STAT_NONE: 10132 return; 10133 case NETDEV_PCPU_STAT_LSTATS: 10134 free_percpu(dev->lstats); 10135 break; 10136 case NETDEV_PCPU_STAT_TSTATS: 10137 free_percpu(dev->tstats); 10138 break; 10139 case NETDEV_PCPU_STAT_DSTATS: 10140 free_percpu(dev->dstats); 10141 break; 10142 } 10143 } 10144 10145 /** 10146 * register_netdevice() - register a network device 10147 * @dev: device to register 10148 * 10149 * Take a prepared network device structure and make it externally accessible. 10150 * A %NETDEV_REGISTER message is sent to the netdev notifier chain. 10151 * Callers must hold the rtnl lock - you may want register_netdev() 10152 * instead of this. 10153 */ 10154 int register_netdevice(struct net_device *dev) 10155 { 10156 int ret; 10157 struct net *net = dev_net(dev); 10158 10159 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE < 10160 NETDEV_FEATURE_COUNT); 10161 BUG_ON(dev_boot_phase); 10162 ASSERT_RTNL(); 10163 10164 might_sleep(); 10165 10166 /* When net_device's are persistent, this will be fatal. */ 10167 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); 10168 BUG_ON(!net); 10169 10170 ret = ethtool_check_ops(dev->ethtool_ops); 10171 if (ret) 10172 return ret; 10173 10174 spin_lock_init(&dev->addr_list_lock); 10175 netdev_set_addr_lockdep_class(dev); 10176 10177 ret = dev_get_valid_name(net, dev, dev->name); 10178 if (ret < 0) 10179 goto out; 10180 10181 ret = -ENOMEM; 10182 dev->name_node = netdev_name_node_head_alloc(dev); 10183 if (!dev->name_node) 10184 goto out; 10185 10186 /* Init, if this function is available */ 10187 if (dev->netdev_ops->ndo_init) { 10188 ret = dev->netdev_ops->ndo_init(dev); 10189 if (ret) { 10190 if (ret > 0) 10191 ret = -EIO; 10192 goto err_free_name; 10193 } 10194 } 10195 10196 if (((dev->hw_features | dev->features) & 10197 NETIF_F_HW_VLAN_CTAG_FILTER) && 10198 (!dev->netdev_ops->ndo_vlan_rx_add_vid || 10199 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { 10200 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); 10201 ret = -EINVAL; 10202 goto err_uninit; 10203 } 10204 10205 ret = netdev_do_alloc_pcpu_stats(dev); 10206 if (ret) 10207 goto err_uninit; 10208 10209 ret = dev_index_reserve(net, dev->ifindex); 10210 if (ret < 0) 10211 goto err_free_pcpu; 10212 dev->ifindex = ret; 10213 10214 /* Transfer changeable features to wanted_features and enable 10215 * software offloads (GSO and GRO). 10216 */ 10217 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF); 10218 dev->features |= NETIF_F_SOFT_FEATURES; 10219 10220 if (dev->udp_tunnel_nic_info) { 10221 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10222 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10223 } 10224 10225 dev->wanted_features = dev->features & dev->hw_features; 10226 10227 if (!(dev->flags & IFF_LOOPBACK)) 10228 dev->hw_features |= NETIF_F_NOCACHE_COPY; 10229 10230 /* If IPv4 TCP segmentation offload is supported we should also 10231 * allow the device to enable segmenting the frame with the option 10232 * of ignoring a static IP ID value. This doesn't enable the 10233 * feature itself but allows the user to enable it later. 10234 */ 10235 if (dev->hw_features & NETIF_F_TSO) 10236 dev->hw_features |= NETIF_F_TSO_MANGLEID; 10237 if (dev->vlan_features & NETIF_F_TSO) 10238 dev->vlan_features |= NETIF_F_TSO_MANGLEID; 10239 if (dev->mpls_features & NETIF_F_TSO) 10240 dev->mpls_features |= NETIF_F_TSO_MANGLEID; 10241 if (dev->hw_enc_features & NETIF_F_TSO) 10242 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; 10243 10244 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. 10245 */ 10246 dev->vlan_features |= NETIF_F_HIGHDMA; 10247 10248 /* Make NETIF_F_SG inheritable to tunnel devices. 10249 */ 10250 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; 10251 10252 /* Make NETIF_F_SG inheritable to MPLS. 10253 */ 10254 dev->mpls_features |= NETIF_F_SG; 10255 10256 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); 10257 ret = notifier_to_errno(ret); 10258 if (ret) 10259 goto err_ifindex_release; 10260 10261 ret = netdev_register_kobject(dev); 10262 write_lock(&dev_base_lock); 10263 dev->reg_state = ret ? NETREG_UNREGISTERED : NETREG_REGISTERED; 10264 write_unlock(&dev_base_lock); 10265 if (ret) 10266 goto err_uninit_notify; 10267 10268 __netdev_update_features(dev); 10269 10270 /* 10271 * Default initial state at registry is that the 10272 * device is present. 10273 */ 10274 10275 set_bit(__LINK_STATE_PRESENT, &dev->state); 10276 10277 linkwatch_init_dev(dev); 10278 10279 dev_init_scheduler(dev); 10280 10281 netdev_hold(dev, &dev->dev_registered_tracker, GFP_KERNEL); 10282 list_netdevice(dev); 10283 10284 add_device_randomness(dev->dev_addr, dev->addr_len); 10285 10286 /* If the device has permanent device address, driver should 10287 * set dev_addr and also addr_assign_type should be set to 10288 * NET_ADDR_PERM (default value). 10289 */ 10290 if (dev->addr_assign_type == NET_ADDR_PERM) 10291 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); 10292 10293 /* Notify protocols, that a new device appeared. */ 10294 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); 10295 ret = notifier_to_errno(ret); 10296 if (ret) { 10297 /* Expect explicit free_netdev() on failure */ 10298 dev->needs_free_netdev = false; 10299 unregister_netdevice_queue(dev, NULL); 10300 goto out; 10301 } 10302 /* 10303 * Prevent userspace races by waiting until the network 10304 * device is fully setup before sending notifications. 10305 */ 10306 if (!dev->rtnl_link_ops || 10307 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 10308 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL); 10309 10310 out: 10311 return ret; 10312 10313 err_uninit_notify: 10314 call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev); 10315 err_ifindex_release: 10316 dev_index_release(net, dev->ifindex); 10317 err_free_pcpu: 10318 netdev_do_free_pcpu_stats(dev); 10319 err_uninit: 10320 if (dev->netdev_ops->ndo_uninit) 10321 dev->netdev_ops->ndo_uninit(dev); 10322 if (dev->priv_destructor) 10323 dev->priv_destructor(dev); 10324 err_free_name: 10325 netdev_name_node_free(dev->name_node); 10326 goto out; 10327 } 10328 EXPORT_SYMBOL(register_netdevice); 10329 10330 /** 10331 * init_dummy_netdev - init a dummy network device for NAPI 10332 * @dev: device to init 10333 * 10334 * This takes a network device structure and initialize the minimum 10335 * amount of fields so it can be used to schedule NAPI polls without 10336 * registering a full blown interface. This is to be used by drivers 10337 * that need to tie several hardware interfaces to a single NAPI 10338 * poll scheduler due to HW limitations. 10339 */ 10340 int init_dummy_netdev(struct net_device *dev) 10341 { 10342 /* Clear everything. Note we don't initialize spinlocks 10343 * are they aren't supposed to be taken by any of the 10344 * NAPI code and this dummy netdev is supposed to be 10345 * only ever used for NAPI polls 10346 */ 10347 memset(dev, 0, sizeof(struct net_device)); 10348 10349 /* make sure we BUG if trying to hit standard 10350 * register/unregister code path 10351 */ 10352 dev->reg_state = NETREG_DUMMY; 10353 10354 /* NAPI wants this */ 10355 INIT_LIST_HEAD(&dev->napi_list); 10356 10357 /* a dummy interface is started by default */ 10358 set_bit(__LINK_STATE_PRESENT, &dev->state); 10359 set_bit(__LINK_STATE_START, &dev->state); 10360 10361 /* napi_busy_loop stats accounting wants this */ 10362 dev_net_set(dev, &init_net); 10363 10364 /* Note : We dont allocate pcpu_refcnt for dummy devices, 10365 * because users of this 'device' dont need to change 10366 * its refcount. 10367 */ 10368 10369 return 0; 10370 } 10371 EXPORT_SYMBOL_GPL(init_dummy_netdev); 10372 10373 10374 /** 10375 * register_netdev - register a network device 10376 * @dev: device to register 10377 * 10378 * Take a completed network device structure and add it to the kernel 10379 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 10380 * chain. 0 is returned on success. A negative errno code is returned 10381 * on a failure to set up the device, or if the name is a duplicate. 10382 * 10383 * This is a wrapper around register_netdevice that takes the rtnl semaphore 10384 * and expands the device name if you passed a format string to 10385 * alloc_netdev. 10386 */ 10387 int register_netdev(struct net_device *dev) 10388 { 10389 int err; 10390 10391 if (rtnl_lock_killable()) 10392 return -EINTR; 10393 err = register_netdevice(dev); 10394 rtnl_unlock(); 10395 return err; 10396 } 10397 EXPORT_SYMBOL(register_netdev); 10398 10399 int netdev_refcnt_read(const struct net_device *dev) 10400 { 10401 #ifdef CONFIG_PCPU_DEV_REFCNT 10402 int i, refcnt = 0; 10403 10404 for_each_possible_cpu(i) 10405 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); 10406 return refcnt; 10407 #else 10408 return refcount_read(&dev->dev_refcnt); 10409 #endif 10410 } 10411 EXPORT_SYMBOL(netdev_refcnt_read); 10412 10413 int netdev_unregister_timeout_secs __read_mostly = 10; 10414 10415 #define WAIT_REFS_MIN_MSECS 1 10416 #define WAIT_REFS_MAX_MSECS 250 10417 /** 10418 * netdev_wait_allrefs_any - wait until all references are gone. 10419 * @list: list of net_devices to wait on 10420 * 10421 * This is called when unregistering network devices. 10422 * 10423 * Any protocol or device that holds a reference should register 10424 * for netdevice notification, and cleanup and put back the 10425 * reference if they receive an UNREGISTER event. 10426 * We can get stuck here if buggy protocols don't correctly 10427 * call dev_put. 10428 */ 10429 static struct net_device *netdev_wait_allrefs_any(struct list_head *list) 10430 { 10431 unsigned long rebroadcast_time, warning_time; 10432 struct net_device *dev; 10433 int wait = 0; 10434 10435 rebroadcast_time = warning_time = jiffies; 10436 10437 list_for_each_entry(dev, list, todo_list) 10438 if (netdev_refcnt_read(dev) == 1) 10439 return dev; 10440 10441 while (true) { 10442 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { 10443 rtnl_lock(); 10444 10445 /* Rebroadcast unregister notification */ 10446 list_for_each_entry(dev, list, todo_list) 10447 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10448 10449 __rtnl_unlock(); 10450 rcu_barrier(); 10451 rtnl_lock(); 10452 10453 list_for_each_entry(dev, list, todo_list) 10454 if (test_bit(__LINK_STATE_LINKWATCH_PENDING, 10455 &dev->state)) { 10456 /* We must not have linkwatch events 10457 * pending on unregister. If this 10458 * happens, we simply run the queue 10459 * unscheduled, resulting in a noop 10460 * for this device. 10461 */ 10462 linkwatch_run_queue(); 10463 break; 10464 } 10465 10466 __rtnl_unlock(); 10467 10468 rebroadcast_time = jiffies; 10469 } 10470 10471 if (!wait) { 10472 rcu_barrier(); 10473 wait = WAIT_REFS_MIN_MSECS; 10474 } else { 10475 msleep(wait); 10476 wait = min(wait << 1, WAIT_REFS_MAX_MSECS); 10477 } 10478 10479 list_for_each_entry(dev, list, todo_list) 10480 if (netdev_refcnt_read(dev) == 1) 10481 return dev; 10482 10483 if (time_after(jiffies, warning_time + 10484 READ_ONCE(netdev_unregister_timeout_secs) * HZ)) { 10485 list_for_each_entry(dev, list, todo_list) { 10486 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", 10487 dev->name, netdev_refcnt_read(dev)); 10488 ref_tracker_dir_print(&dev->refcnt_tracker, 10); 10489 } 10490 10491 warning_time = jiffies; 10492 } 10493 } 10494 } 10495 10496 /* The sequence is: 10497 * 10498 * rtnl_lock(); 10499 * ... 10500 * register_netdevice(x1); 10501 * register_netdevice(x2); 10502 * ... 10503 * unregister_netdevice(y1); 10504 * unregister_netdevice(y2); 10505 * ... 10506 * rtnl_unlock(); 10507 * free_netdev(y1); 10508 * free_netdev(y2); 10509 * 10510 * We are invoked by rtnl_unlock(). 10511 * This allows us to deal with problems: 10512 * 1) We can delete sysfs objects which invoke hotplug 10513 * without deadlocking with linkwatch via keventd. 10514 * 2) Since we run with the RTNL semaphore not held, we can sleep 10515 * safely in order to wait for the netdev refcnt to drop to zero. 10516 * 10517 * We must not return until all unregister events added during 10518 * the interval the lock was held have been completed. 10519 */ 10520 void netdev_run_todo(void) 10521 { 10522 struct net_device *dev, *tmp; 10523 struct list_head list; 10524 #ifdef CONFIG_LOCKDEP 10525 struct list_head unlink_list; 10526 10527 list_replace_init(&net_unlink_list, &unlink_list); 10528 10529 while (!list_empty(&unlink_list)) { 10530 struct net_device *dev = list_first_entry(&unlink_list, 10531 struct net_device, 10532 unlink_list); 10533 list_del_init(&dev->unlink_list); 10534 dev->nested_level = dev->lower_level - 1; 10535 } 10536 #endif 10537 10538 /* Snapshot list, allow later requests */ 10539 list_replace_init(&net_todo_list, &list); 10540 10541 __rtnl_unlock(); 10542 10543 /* Wait for rcu callbacks to finish before next phase */ 10544 if (!list_empty(&list)) 10545 rcu_barrier(); 10546 10547 list_for_each_entry_safe(dev, tmp, &list, todo_list) { 10548 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { 10549 netdev_WARN(dev, "run_todo but not unregistering\n"); 10550 list_del(&dev->todo_list); 10551 continue; 10552 } 10553 10554 write_lock(&dev_base_lock); 10555 dev->reg_state = NETREG_UNREGISTERED; 10556 write_unlock(&dev_base_lock); 10557 linkwatch_sync_dev(dev); 10558 } 10559 10560 while (!list_empty(&list)) { 10561 dev = netdev_wait_allrefs_any(&list); 10562 list_del(&dev->todo_list); 10563 10564 /* paranoia */ 10565 BUG_ON(netdev_refcnt_read(dev) != 1); 10566 BUG_ON(!list_empty(&dev->ptype_all)); 10567 BUG_ON(!list_empty(&dev->ptype_specific)); 10568 WARN_ON(rcu_access_pointer(dev->ip_ptr)); 10569 WARN_ON(rcu_access_pointer(dev->ip6_ptr)); 10570 10571 netdev_do_free_pcpu_stats(dev); 10572 if (dev->priv_destructor) 10573 dev->priv_destructor(dev); 10574 if (dev->needs_free_netdev) 10575 free_netdev(dev); 10576 10577 if (atomic_dec_and_test(&dev_net(dev)->dev_unreg_count)) 10578 wake_up(&netdev_unregistering_wq); 10579 10580 /* Free network device */ 10581 kobject_put(&dev->dev.kobj); 10582 } 10583 } 10584 10585 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has 10586 * all the same fields in the same order as net_device_stats, with only 10587 * the type differing, but rtnl_link_stats64 may have additional fields 10588 * at the end for newer counters. 10589 */ 10590 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, 10591 const struct net_device_stats *netdev_stats) 10592 { 10593 size_t i, n = sizeof(*netdev_stats) / sizeof(atomic_long_t); 10594 const atomic_long_t *src = (atomic_long_t *)netdev_stats; 10595 u64 *dst = (u64 *)stats64; 10596 10597 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); 10598 for (i = 0; i < n; i++) 10599 dst[i] = (unsigned long)atomic_long_read(&src[i]); 10600 /* zero out counters that only exist in rtnl_link_stats64 */ 10601 memset((char *)stats64 + n * sizeof(u64), 0, 10602 sizeof(*stats64) - n * sizeof(u64)); 10603 } 10604 EXPORT_SYMBOL(netdev_stats_to_stats64); 10605 10606 static __cold struct net_device_core_stats __percpu *netdev_core_stats_alloc( 10607 struct net_device *dev) 10608 { 10609 struct net_device_core_stats __percpu *p; 10610 10611 p = alloc_percpu_gfp(struct net_device_core_stats, 10612 GFP_ATOMIC | __GFP_NOWARN); 10613 10614 if (p && cmpxchg(&dev->core_stats, NULL, p)) 10615 free_percpu(p); 10616 10617 /* This READ_ONCE() pairs with the cmpxchg() above */ 10618 return READ_ONCE(dev->core_stats); 10619 } 10620 10621 noinline void netdev_core_stats_inc(struct net_device *dev, u32 offset) 10622 { 10623 /* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */ 10624 struct net_device_core_stats __percpu *p = READ_ONCE(dev->core_stats); 10625 unsigned long __percpu *field; 10626 10627 if (unlikely(!p)) { 10628 p = netdev_core_stats_alloc(dev); 10629 if (!p) 10630 return; 10631 } 10632 10633 field = (__force unsigned long __percpu *)((__force void *)p + offset); 10634 this_cpu_inc(*field); 10635 } 10636 EXPORT_SYMBOL_GPL(netdev_core_stats_inc); 10637 10638 /** 10639 * dev_get_stats - get network device statistics 10640 * @dev: device to get statistics from 10641 * @storage: place to store stats 10642 * 10643 * Get network statistics from device. Return @storage. 10644 * The device driver may provide its own method by setting 10645 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; 10646 * otherwise the internal statistics structure is used. 10647 */ 10648 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, 10649 struct rtnl_link_stats64 *storage) 10650 { 10651 const struct net_device_ops *ops = dev->netdev_ops; 10652 const struct net_device_core_stats __percpu *p; 10653 10654 if (ops->ndo_get_stats64) { 10655 memset(storage, 0, sizeof(*storage)); 10656 ops->ndo_get_stats64(dev, storage); 10657 } else if (ops->ndo_get_stats) { 10658 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); 10659 } else { 10660 netdev_stats_to_stats64(storage, &dev->stats); 10661 } 10662 10663 /* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */ 10664 p = READ_ONCE(dev->core_stats); 10665 if (p) { 10666 const struct net_device_core_stats *core_stats; 10667 int i; 10668 10669 for_each_possible_cpu(i) { 10670 core_stats = per_cpu_ptr(p, i); 10671 storage->rx_dropped += READ_ONCE(core_stats->rx_dropped); 10672 storage->tx_dropped += READ_ONCE(core_stats->tx_dropped); 10673 storage->rx_nohandler += READ_ONCE(core_stats->rx_nohandler); 10674 storage->rx_otherhost_dropped += READ_ONCE(core_stats->rx_otherhost_dropped); 10675 } 10676 } 10677 return storage; 10678 } 10679 EXPORT_SYMBOL(dev_get_stats); 10680 10681 /** 10682 * dev_fetch_sw_netstats - get per-cpu network device statistics 10683 * @s: place to store stats 10684 * @netstats: per-cpu network stats to read from 10685 * 10686 * Read per-cpu network statistics and populate the related fields in @s. 10687 */ 10688 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s, 10689 const struct pcpu_sw_netstats __percpu *netstats) 10690 { 10691 int cpu; 10692 10693 for_each_possible_cpu(cpu) { 10694 u64 rx_packets, rx_bytes, tx_packets, tx_bytes; 10695 const struct pcpu_sw_netstats *stats; 10696 unsigned int start; 10697 10698 stats = per_cpu_ptr(netstats, cpu); 10699 do { 10700 start = u64_stats_fetch_begin(&stats->syncp); 10701 rx_packets = u64_stats_read(&stats->rx_packets); 10702 rx_bytes = u64_stats_read(&stats->rx_bytes); 10703 tx_packets = u64_stats_read(&stats->tx_packets); 10704 tx_bytes = u64_stats_read(&stats->tx_bytes); 10705 } while (u64_stats_fetch_retry(&stats->syncp, start)); 10706 10707 s->rx_packets += rx_packets; 10708 s->rx_bytes += rx_bytes; 10709 s->tx_packets += tx_packets; 10710 s->tx_bytes += tx_bytes; 10711 } 10712 } 10713 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats); 10714 10715 /** 10716 * dev_get_tstats64 - ndo_get_stats64 implementation 10717 * @dev: device to get statistics from 10718 * @s: place to store stats 10719 * 10720 * Populate @s from dev->stats and dev->tstats. Can be used as 10721 * ndo_get_stats64() callback. 10722 */ 10723 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s) 10724 { 10725 netdev_stats_to_stats64(s, &dev->stats); 10726 dev_fetch_sw_netstats(s, dev->tstats); 10727 } 10728 EXPORT_SYMBOL_GPL(dev_get_tstats64); 10729 10730 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) 10731 { 10732 struct netdev_queue *queue = dev_ingress_queue(dev); 10733 10734 #ifdef CONFIG_NET_CLS_ACT 10735 if (queue) 10736 return queue; 10737 queue = kzalloc(sizeof(*queue), GFP_KERNEL); 10738 if (!queue) 10739 return NULL; 10740 netdev_init_one_queue(dev, queue, NULL); 10741 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); 10742 RCU_INIT_POINTER(queue->qdisc_sleeping, &noop_qdisc); 10743 rcu_assign_pointer(dev->ingress_queue, queue); 10744 #endif 10745 return queue; 10746 } 10747 10748 static const struct ethtool_ops default_ethtool_ops; 10749 10750 void netdev_set_default_ethtool_ops(struct net_device *dev, 10751 const struct ethtool_ops *ops) 10752 { 10753 if (dev->ethtool_ops == &default_ethtool_ops) 10754 dev->ethtool_ops = ops; 10755 } 10756 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); 10757 10758 /** 10759 * netdev_sw_irq_coalesce_default_on() - enable SW IRQ coalescing by default 10760 * @dev: netdev to enable the IRQ coalescing on 10761 * 10762 * Sets a conservative default for SW IRQ coalescing. Users can use 10763 * sysfs attributes to override the default values. 10764 */ 10765 void netdev_sw_irq_coalesce_default_on(struct net_device *dev) 10766 { 10767 WARN_ON(dev->reg_state == NETREG_REGISTERED); 10768 10769 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) { 10770 dev->gro_flush_timeout = 20000; 10771 dev->napi_defer_hard_irqs = 1; 10772 } 10773 } 10774 EXPORT_SYMBOL_GPL(netdev_sw_irq_coalesce_default_on); 10775 10776 void netdev_freemem(struct net_device *dev) 10777 { 10778 char *addr = (char *)dev - dev->padded; 10779 10780 kvfree(addr); 10781 } 10782 10783 /** 10784 * alloc_netdev_mqs - allocate network device 10785 * @sizeof_priv: size of private data to allocate space for 10786 * @name: device name format string 10787 * @name_assign_type: origin of device name 10788 * @setup: callback to initialize device 10789 * @txqs: the number of TX subqueues to allocate 10790 * @rxqs: the number of RX subqueues to allocate 10791 * 10792 * Allocates a struct net_device with private data area for driver use 10793 * and performs basic initialization. Also allocates subqueue structs 10794 * for each queue on the device. 10795 */ 10796 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, 10797 unsigned char name_assign_type, 10798 void (*setup)(struct net_device *), 10799 unsigned int txqs, unsigned int rxqs) 10800 { 10801 struct net_device *dev; 10802 unsigned int alloc_size; 10803 struct net_device *p; 10804 10805 BUG_ON(strlen(name) >= sizeof(dev->name)); 10806 10807 if (txqs < 1) { 10808 pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); 10809 return NULL; 10810 } 10811 10812 if (rxqs < 1) { 10813 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); 10814 return NULL; 10815 } 10816 10817 alloc_size = sizeof(struct net_device); 10818 if (sizeof_priv) { 10819 /* ensure 32-byte alignment of private area */ 10820 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); 10821 alloc_size += sizeof_priv; 10822 } 10823 /* ensure 32-byte alignment of whole construct */ 10824 alloc_size += NETDEV_ALIGN - 1; 10825 10826 p = kvzalloc(alloc_size, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10827 if (!p) 10828 return NULL; 10829 10830 dev = PTR_ALIGN(p, NETDEV_ALIGN); 10831 dev->padded = (char *)dev - (char *)p; 10832 10833 ref_tracker_dir_init(&dev->refcnt_tracker, 128, name); 10834 #ifdef CONFIG_PCPU_DEV_REFCNT 10835 dev->pcpu_refcnt = alloc_percpu(int); 10836 if (!dev->pcpu_refcnt) 10837 goto free_dev; 10838 __dev_hold(dev); 10839 #else 10840 refcount_set(&dev->dev_refcnt, 1); 10841 #endif 10842 10843 if (dev_addr_init(dev)) 10844 goto free_pcpu; 10845 10846 dev_mc_init(dev); 10847 dev_uc_init(dev); 10848 10849 dev_net_set(dev, &init_net); 10850 10851 dev->gso_max_size = GSO_LEGACY_MAX_SIZE; 10852 dev->xdp_zc_max_segs = 1; 10853 dev->gso_max_segs = GSO_MAX_SEGS; 10854 dev->gro_max_size = GRO_LEGACY_MAX_SIZE; 10855 dev->gso_ipv4_max_size = GSO_LEGACY_MAX_SIZE; 10856 dev->gro_ipv4_max_size = GRO_LEGACY_MAX_SIZE; 10857 dev->tso_max_size = TSO_LEGACY_MAX_SIZE; 10858 dev->tso_max_segs = TSO_MAX_SEGS; 10859 dev->upper_level = 1; 10860 dev->lower_level = 1; 10861 #ifdef CONFIG_LOCKDEP 10862 dev->nested_level = 0; 10863 INIT_LIST_HEAD(&dev->unlink_list); 10864 #endif 10865 10866 INIT_LIST_HEAD(&dev->napi_list); 10867 INIT_LIST_HEAD(&dev->unreg_list); 10868 INIT_LIST_HEAD(&dev->close_list); 10869 INIT_LIST_HEAD(&dev->link_watch_list); 10870 INIT_LIST_HEAD(&dev->adj_list.upper); 10871 INIT_LIST_HEAD(&dev->adj_list.lower); 10872 INIT_LIST_HEAD(&dev->ptype_all); 10873 INIT_LIST_HEAD(&dev->ptype_specific); 10874 INIT_LIST_HEAD(&dev->net_notifier_list); 10875 #ifdef CONFIG_NET_SCHED 10876 hash_init(dev->qdisc_hash); 10877 #endif 10878 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; 10879 setup(dev); 10880 10881 if (!dev->tx_queue_len) { 10882 dev->priv_flags |= IFF_NO_QUEUE; 10883 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; 10884 } 10885 10886 dev->num_tx_queues = txqs; 10887 dev->real_num_tx_queues = txqs; 10888 if (netif_alloc_netdev_queues(dev)) 10889 goto free_all; 10890 10891 dev->num_rx_queues = rxqs; 10892 dev->real_num_rx_queues = rxqs; 10893 if (netif_alloc_rx_queues(dev)) 10894 goto free_all; 10895 10896 strcpy(dev->name, name); 10897 dev->name_assign_type = name_assign_type; 10898 dev->group = INIT_NETDEV_GROUP; 10899 if (!dev->ethtool_ops) 10900 dev->ethtool_ops = &default_ethtool_ops; 10901 10902 nf_hook_netdev_init(dev); 10903 10904 return dev; 10905 10906 free_all: 10907 free_netdev(dev); 10908 return NULL; 10909 10910 free_pcpu: 10911 #ifdef CONFIG_PCPU_DEV_REFCNT 10912 free_percpu(dev->pcpu_refcnt); 10913 free_dev: 10914 #endif 10915 netdev_freemem(dev); 10916 return NULL; 10917 } 10918 EXPORT_SYMBOL(alloc_netdev_mqs); 10919 10920 /** 10921 * free_netdev - free network device 10922 * @dev: device 10923 * 10924 * This function does the last stage of destroying an allocated device 10925 * interface. The reference to the device object is released. If this 10926 * is the last reference then it will be freed.Must be called in process 10927 * context. 10928 */ 10929 void free_netdev(struct net_device *dev) 10930 { 10931 struct napi_struct *p, *n; 10932 10933 might_sleep(); 10934 10935 /* When called immediately after register_netdevice() failed the unwind 10936 * handling may still be dismantling the device. Handle that case by 10937 * deferring the free. 10938 */ 10939 if (dev->reg_state == NETREG_UNREGISTERING) { 10940 ASSERT_RTNL(); 10941 dev->needs_free_netdev = true; 10942 return; 10943 } 10944 10945 netif_free_tx_queues(dev); 10946 netif_free_rx_queues(dev); 10947 10948 kfree(rcu_dereference_protected(dev->ingress_queue, 1)); 10949 10950 /* Flush device addresses */ 10951 dev_addr_flush(dev); 10952 10953 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) 10954 netif_napi_del(p); 10955 10956 ref_tracker_dir_exit(&dev->refcnt_tracker); 10957 #ifdef CONFIG_PCPU_DEV_REFCNT 10958 free_percpu(dev->pcpu_refcnt); 10959 dev->pcpu_refcnt = NULL; 10960 #endif 10961 free_percpu(dev->core_stats); 10962 dev->core_stats = NULL; 10963 free_percpu(dev->xdp_bulkq); 10964 dev->xdp_bulkq = NULL; 10965 10966 /* Compatibility with error handling in drivers */ 10967 if (dev->reg_state == NETREG_UNINITIALIZED) { 10968 netdev_freemem(dev); 10969 return; 10970 } 10971 10972 BUG_ON(dev->reg_state != NETREG_UNREGISTERED); 10973 dev->reg_state = NETREG_RELEASED; 10974 10975 /* will free via device release */ 10976 put_device(&dev->dev); 10977 } 10978 EXPORT_SYMBOL(free_netdev); 10979 10980 /** 10981 * synchronize_net - Synchronize with packet receive processing 10982 * 10983 * Wait for packets currently being received to be done. 10984 * Does not block later packets from starting. 10985 */ 10986 void synchronize_net(void) 10987 { 10988 might_sleep(); 10989 if (rtnl_is_locked()) 10990 synchronize_rcu_expedited(); 10991 else 10992 synchronize_rcu(); 10993 } 10994 EXPORT_SYMBOL(synchronize_net); 10995 10996 /** 10997 * unregister_netdevice_queue - remove device from the kernel 10998 * @dev: device 10999 * @head: list 11000 * 11001 * This function shuts down a device interface and removes it 11002 * from the kernel tables. 11003 * If head not NULL, device is queued to be unregistered later. 11004 * 11005 * Callers must hold the rtnl semaphore. You may want 11006 * unregister_netdev() instead of this. 11007 */ 11008 11009 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) 11010 { 11011 ASSERT_RTNL(); 11012 11013 if (head) { 11014 list_move_tail(&dev->unreg_list, head); 11015 } else { 11016 LIST_HEAD(single); 11017 11018 list_add(&dev->unreg_list, &single); 11019 unregister_netdevice_many(&single); 11020 } 11021 } 11022 EXPORT_SYMBOL(unregister_netdevice_queue); 11023 11024 void unregister_netdevice_many_notify(struct list_head *head, 11025 u32 portid, const struct nlmsghdr *nlh) 11026 { 11027 struct net_device *dev, *tmp; 11028 LIST_HEAD(close_head); 11029 11030 BUG_ON(dev_boot_phase); 11031 ASSERT_RTNL(); 11032 11033 if (list_empty(head)) 11034 return; 11035 11036 list_for_each_entry_safe(dev, tmp, head, unreg_list) { 11037 /* Some devices call without registering 11038 * for initialization unwind. Remove those 11039 * devices and proceed with the remaining. 11040 */ 11041 if (dev->reg_state == NETREG_UNINITIALIZED) { 11042 pr_debug("unregister_netdevice: device %s/%p never was registered\n", 11043 dev->name, dev); 11044 11045 WARN_ON(1); 11046 list_del(&dev->unreg_list); 11047 continue; 11048 } 11049 dev->dismantle = true; 11050 BUG_ON(dev->reg_state != NETREG_REGISTERED); 11051 } 11052 11053 /* If device is running, close it first. */ 11054 list_for_each_entry(dev, head, unreg_list) 11055 list_add_tail(&dev->close_list, &close_head); 11056 dev_close_many(&close_head, true); 11057 11058 list_for_each_entry(dev, head, unreg_list) { 11059 /* And unlink it from device chain. */ 11060 write_lock(&dev_base_lock); 11061 unlist_netdevice(dev, false); 11062 dev->reg_state = NETREG_UNREGISTERING; 11063 write_unlock(&dev_base_lock); 11064 } 11065 flush_all_backlogs(); 11066 11067 synchronize_net(); 11068 11069 list_for_each_entry(dev, head, unreg_list) { 11070 struct sk_buff *skb = NULL; 11071 11072 /* Shutdown queueing discipline. */ 11073 dev_shutdown(dev); 11074 dev_tcx_uninstall(dev); 11075 dev_xdp_uninstall(dev); 11076 bpf_dev_bound_netdev_unregister(dev); 11077 11078 netdev_offload_xstats_disable_all(dev); 11079 11080 /* Notify protocols, that we are about to destroy 11081 * this device. They should clean all the things. 11082 */ 11083 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 11084 11085 if (!dev->rtnl_link_ops || 11086 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 11087 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0, 11088 GFP_KERNEL, NULL, 0, 11089 portid, nlh); 11090 11091 /* 11092 * Flush the unicast and multicast chains 11093 */ 11094 dev_uc_flush(dev); 11095 dev_mc_flush(dev); 11096 11097 netdev_name_node_alt_flush(dev); 11098 netdev_name_node_free(dev->name_node); 11099 11100 call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev); 11101 11102 if (dev->netdev_ops->ndo_uninit) 11103 dev->netdev_ops->ndo_uninit(dev); 11104 11105 if (skb) 11106 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL, portid, nlh); 11107 11108 /* Notifier chain MUST detach us all upper devices. */ 11109 WARN_ON(netdev_has_any_upper_dev(dev)); 11110 WARN_ON(netdev_has_any_lower_dev(dev)); 11111 11112 /* Remove entries from kobject tree */ 11113 netdev_unregister_kobject(dev); 11114 #ifdef CONFIG_XPS 11115 /* Remove XPS queueing entries */ 11116 netif_reset_xps_queues_gt(dev, 0); 11117 #endif 11118 } 11119 11120 synchronize_net(); 11121 11122 list_for_each_entry(dev, head, unreg_list) { 11123 netdev_put(dev, &dev->dev_registered_tracker); 11124 net_set_todo(dev); 11125 } 11126 11127 list_del(head); 11128 } 11129 11130 /** 11131 * unregister_netdevice_many - unregister many devices 11132 * @head: list of devices 11133 * 11134 * Note: As most callers use a stack allocated list_head, 11135 * we force a list_del() to make sure stack wont be corrupted later. 11136 */ 11137 void unregister_netdevice_many(struct list_head *head) 11138 { 11139 unregister_netdevice_many_notify(head, 0, NULL); 11140 } 11141 EXPORT_SYMBOL(unregister_netdevice_many); 11142 11143 /** 11144 * unregister_netdev - remove device from the kernel 11145 * @dev: device 11146 * 11147 * This function shuts down a device interface and removes it 11148 * from the kernel tables. 11149 * 11150 * This is just a wrapper for unregister_netdevice that takes 11151 * the rtnl semaphore. In general you want to use this and not 11152 * unregister_netdevice. 11153 */ 11154 void unregister_netdev(struct net_device *dev) 11155 { 11156 rtnl_lock(); 11157 unregister_netdevice(dev); 11158 rtnl_unlock(); 11159 } 11160 EXPORT_SYMBOL(unregister_netdev); 11161 11162 /** 11163 * __dev_change_net_namespace - move device to different nethost namespace 11164 * @dev: device 11165 * @net: network namespace 11166 * @pat: If not NULL name pattern to try if the current device name 11167 * is already taken in the destination network namespace. 11168 * @new_ifindex: If not zero, specifies device index in the target 11169 * namespace. 11170 * 11171 * This function shuts down a device interface and moves it 11172 * to a new network namespace. On success 0 is returned, on 11173 * a failure a netagive errno code is returned. 11174 * 11175 * Callers must hold the rtnl semaphore. 11176 */ 11177 11178 int __dev_change_net_namespace(struct net_device *dev, struct net *net, 11179 const char *pat, int new_ifindex) 11180 { 11181 struct netdev_name_node *name_node; 11182 struct net *net_old = dev_net(dev); 11183 char new_name[IFNAMSIZ] = {}; 11184 int err, new_nsid; 11185 11186 ASSERT_RTNL(); 11187 11188 /* Don't allow namespace local devices to be moved. */ 11189 err = -EINVAL; 11190 if (dev->features & NETIF_F_NETNS_LOCAL) 11191 goto out; 11192 11193 /* Ensure the device has been registrered */ 11194 if (dev->reg_state != NETREG_REGISTERED) 11195 goto out; 11196 11197 /* Get out if there is nothing todo */ 11198 err = 0; 11199 if (net_eq(net_old, net)) 11200 goto out; 11201 11202 /* Pick the destination device name, and ensure 11203 * we can use it in the destination network namespace. 11204 */ 11205 err = -EEXIST; 11206 if (netdev_name_in_use(net, dev->name)) { 11207 /* We get here if we can't use the current device name */ 11208 if (!pat) 11209 goto out; 11210 err = dev_prep_valid_name(net, dev, pat, new_name, EEXIST); 11211 if (err < 0) 11212 goto out; 11213 } 11214 /* Check that none of the altnames conflicts. */ 11215 err = -EEXIST; 11216 netdev_for_each_altname(dev, name_node) 11217 if (netdev_name_in_use(net, name_node->name)) 11218 goto out; 11219 11220 /* Check that new_ifindex isn't used yet. */ 11221 if (new_ifindex) { 11222 err = dev_index_reserve(net, new_ifindex); 11223 if (err < 0) 11224 goto out; 11225 } else { 11226 /* If there is an ifindex conflict assign a new one */ 11227 err = dev_index_reserve(net, dev->ifindex); 11228 if (err == -EBUSY) 11229 err = dev_index_reserve(net, 0); 11230 if (err < 0) 11231 goto out; 11232 new_ifindex = err; 11233 } 11234 11235 /* 11236 * And now a mini version of register_netdevice unregister_netdevice. 11237 */ 11238 11239 /* If device is running close it first. */ 11240 dev_close(dev); 11241 11242 /* And unlink it from device chain */ 11243 unlist_netdevice(dev, true); 11244 11245 synchronize_net(); 11246 11247 /* Shutdown queueing discipline. */ 11248 dev_shutdown(dev); 11249 11250 /* Notify protocols, that we are about to destroy 11251 * this device. They should clean all the things. 11252 * 11253 * Note that dev->reg_state stays at NETREG_REGISTERED. 11254 * This is wanted because this way 8021q and macvlan know 11255 * the device is just moving and can keep their slaves up. 11256 */ 11257 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 11258 rcu_barrier(); 11259 11260 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL); 11261 11262 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid, 11263 new_ifindex); 11264 11265 /* 11266 * Flush the unicast and multicast chains 11267 */ 11268 dev_uc_flush(dev); 11269 dev_mc_flush(dev); 11270 11271 /* Send a netdev-removed uevent to the old namespace */ 11272 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); 11273 netdev_adjacent_del_links(dev); 11274 11275 /* Move per-net netdevice notifiers that are following the netdevice */ 11276 move_netdevice_notifiers_dev_net(dev, net); 11277 11278 /* Actually switch the network namespace */ 11279 dev_net_set(dev, net); 11280 dev->ifindex = new_ifindex; 11281 11282 if (new_name[0]) /* Rename the netdev to prepared name */ 11283 strscpy(dev->name, new_name, IFNAMSIZ); 11284 11285 /* Fixup kobjects */ 11286 dev_set_uevent_suppress(&dev->dev, 1); 11287 err = device_rename(&dev->dev, dev->name); 11288 dev_set_uevent_suppress(&dev->dev, 0); 11289 WARN_ON(err); 11290 11291 /* Send a netdev-add uevent to the new namespace */ 11292 kobject_uevent(&dev->dev.kobj, KOBJ_ADD); 11293 netdev_adjacent_add_links(dev); 11294 11295 /* Adapt owner in case owning user namespace of target network 11296 * namespace is different from the original one. 11297 */ 11298 err = netdev_change_owner(dev, net_old, net); 11299 WARN_ON(err); 11300 11301 /* Add the device back in the hashes */ 11302 list_netdevice(dev); 11303 11304 /* Notify protocols, that a new device appeared. */ 11305 call_netdevice_notifiers(NETDEV_REGISTER, dev); 11306 11307 /* 11308 * Prevent userspace races by waiting until the network 11309 * device is fully setup before sending notifications. 11310 */ 11311 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL); 11312 11313 synchronize_net(); 11314 err = 0; 11315 out: 11316 return err; 11317 } 11318 EXPORT_SYMBOL_GPL(__dev_change_net_namespace); 11319 11320 static int dev_cpu_dead(unsigned int oldcpu) 11321 { 11322 struct sk_buff **list_skb; 11323 struct sk_buff *skb; 11324 unsigned int cpu; 11325 struct softnet_data *sd, *oldsd, *remsd = NULL; 11326 11327 local_irq_disable(); 11328 cpu = smp_processor_id(); 11329 sd = &per_cpu(softnet_data, cpu); 11330 oldsd = &per_cpu(softnet_data, oldcpu); 11331 11332 /* Find end of our completion_queue. */ 11333 list_skb = &sd->completion_queue; 11334 while (*list_skb) 11335 list_skb = &(*list_skb)->next; 11336 /* Append completion queue from offline CPU. */ 11337 *list_skb = oldsd->completion_queue; 11338 oldsd->completion_queue = NULL; 11339 11340 /* Append output queue from offline CPU. */ 11341 if (oldsd->output_queue) { 11342 *sd->output_queue_tailp = oldsd->output_queue; 11343 sd->output_queue_tailp = oldsd->output_queue_tailp; 11344 oldsd->output_queue = NULL; 11345 oldsd->output_queue_tailp = &oldsd->output_queue; 11346 } 11347 /* Append NAPI poll list from offline CPU, with one exception : 11348 * process_backlog() must be called by cpu owning percpu backlog. 11349 * We properly handle process_queue & input_pkt_queue later. 11350 */ 11351 while (!list_empty(&oldsd->poll_list)) { 11352 struct napi_struct *napi = list_first_entry(&oldsd->poll_list, 11353 struct napi_struct, 11354 poll_list); 11355 11356 list_del_init(&napi->poll_list); 11357 if (napi->poll == process_backlog) 11358 napi->state = 0; 11359 else 11360 ____napi_schedule(sd, napi); 11361 } 11362 11363 raise_softirq_irqoff(NET_TX_SOFTIRQ); 11364 local_irq_enable(); 11365 11366 #ifdef CONFIG_RPS 11367 remsd = oldsd->rps_ipi_list; 11368 oldsd->rps_ipi_list = NULL; 11369 #endif 11370 /* send out pending IPI's on offline CPU */ 11371 net_rps_send_ipi(remsd); 11372 11373 /* Process offline CPU's input_pkt_queue */ 11374 while ((skb = __skb_dequeue(&oldsd->process_queue))) { 11375 netif_rx(skb); 11376 input_queue_head_incr(oldsd); 11377 } 11378 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { 11379 netif_rx(skb); 11380 input_queue_head_incr(oldsd); 11381 } 11382 11383 return 0; 11384 } 11385 11386 /** 11387 * netdev_increment_features - increment feature set by one 11388 * @all: current feature set 11389 * @one: new feature set 11390 * @mask: mask feature set 11391 * 11392 * Computes a new feature set after adding a device with feature set 11393 * @one to the master device with current feature set @all. Will not 11394 * enable anything that is off in @mask. Returns the new feature set. 11395 */ 11396 netdev_features_t netdev_increment_features(netdev_features_t all, 11397 netdev_features_t one, netdev_features_t mask) 11398 { 11399 if (mask & NETIF_F_HW_CSUM) 11400 mask |= NETIF_F_CSUM_MASK; 11401 mask |= NETIF_F_VLAN_CHALLENGED; 11402 11403 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; 11404 all &= one | ~NETIF_F_ALL_FOR_ALL; 11405 11406 /* If one device supports hw checksumming, set for all. */ 11407 if (all & NETIF_F_HW_CSUM) 11408 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); 11409 11410 return all; 11411 } 11412 EXPORT_SYMBOL(netdev_increment_features); 11413 11414 static struct hlist_head * __net_init netdev_create_hash(void) 11415 { 11416 int i; 11417 struct hlist_head *hash; 11418 11419 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL); 11420 if (hash != NULL) 11421 for (i = 0; i < NETDEV_HASHENTRIES; i++) 11422 INIT_HLIST_HEAD(&hash[i]); 11423 11424 return hash; 11425 } 11426 11427 /* Initialize per network namespace state */ 11428 static int __net_init netdev_init(struct net *net) 11429 { 11430 BUILD_BUG_ON(GRO_HASH_BUCKETS > 11431 8 * sizeof_field(struct napi_struct, gro_bitmask)); 11432 11433 INIT_LIST_HEAD(&net->dev_base_head); 11434 11435 net->dev_name_head = netdev_create_hash(); 11436 if (net->dev_name_head == NULL) 11437 goto err_name; 11438 11439 net->dev_index_head = netdev_create_hash(); 11440 if (net->dev_index_head == NULL) 11441 goto err_idx; 11442 11443 xa_init_flags(&net->dev_by_index, XA_FLAGS_ALLOC1); 11444 11445 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain); 11446 11447 return 0; 11448 11449 err_idx: 11450 kfree(net->dev_name_head); 11451 err_name: 11452 return -ENOMEM; 11453 } 11454 11455 /** 11456 * netdev_drivername - network driver for the device 11457 * @dev: network device 11458 * 11459 * Determine network driver for device. 11460 */ 11461 const char *netdev_drivername(const struct net_device *dev) 11462 { 11463 const struct device_driver *driver; 11464 const struct device *parent; 11465 const char *empty = ""; 11466 11467 parent = dev->dev.parent; 11468 if (!parent) 11469 return empty; 11470 11471 driver = parent->driver; 11472 if (driver && driver->name) 11473 return driver->name; 11474 return empty; 11475 } 11476 11477 static void __netdev_printk(const char *level, const struct net_device *dev, 11478 struct va_format *vaf) 11479 { 11480 if (dev && dev->dev.parent) { 11481 dev_printk_emit(level[1] - '0', 11482 dev->dev.parent, 11483 "%s %s %s%s: %pV", 11484 dev_driver_string(dev->dev.parent), 11485 dev_name(dev->dev.parent), 11486 netdev_name(dev), netdev_reg_state(dev), 11487 vaf); 11488 } else if (dev) { 11489 printk("%s%s%s: %pV", 11490 level, netdev_name(dev), netdev_reg_state(dev), vaf); 11491 } else { 11492 printk("%s(NULL net_device): %pV", level, vaf); 11493 } 11494 } 11495 11496 void netdev_printk(const char *level, const struct net_device *dev, 11497 const char *format, ...) 11498 { 11499 struct va_format vaf; 11500 va_list args; 11501 11502 va_start(args, format); 11503 11504 vaf.fmt = format; 11505 vaf.va = &args; 11506 11507 __netdev_printk(level, dev, &vaf); 11508 11509 va_end(args); 11510 } 11511 EXPORT_SYMBOL(netdev_printk); 11512 11513 #define define_netdev_printk_level(func, level) \ 11514 void func(const struct net_device *dev, const char *fmt, ...) \ 11515 { \ 11516 struct va_format vaf; \ 11517 va_list args; \ 11518 \ 11519 va_start(args, fmt); \ 11520 \ 11521 vaf.fmt = fmt; \ 11522 vaf.va = &args; \ 11523 \ 11524 __netdev_printk(level, dev, &vaf); \ 11525 \ 11526 va_end(args); \ 11527 } \ 11528 EXPORT_SYMBOL(func); 11529 11530 define_netdev_printk_level(netdev_emerg, KERN_EMERG); 11531 define_netdev_printk_level(netdev_alert, KERN_ALERT); 11532 define_netdev_printk_level(netdev_crit, KERN_CRIT); 11533 define_netdev_printk_level(netdev_err, KERN_ERR); 11534 define_netdev_printk_level(netdev_warn, KERN_WARNING); 11535 define_netdev_printk_level(netdev_notice, KERN_NOTICE); 11536 define_netdev_printk_level(netdev_info, KERN_INFO); 11537 11538 static void __net_exit netdev_exit(struct net *net) 11539 { 11540 kfree(net->dev_name_head); 11541 kfree(net->dev_index_head); 11542 xa_destroy(&net->dev_by_index); 11543 if (net != &init_net) 11544 WARN_ON_ONCE(!list_empty(&net->dev_base_head)); 11545 } 11546 11547 static struct pernet_operations __net_initdata netdev_net_ops = { 11548 .init = netdev_init, 11549 .exit = netdev_exit, 11550 }; 11551 11552 static void __net_exit default_device_exit_net(struct net *net) 11553 { 11554 struct net_device *dev, *aux; 11555 /* 11556 * Push all migratable network devices back to the 11557 * initial network namespace 11558 */ 11559 ASSERT_RTNL(); 11560 for_each_netdev_safe(net, dev, aux) { 11561 int err; 11562 char fb_name[IFNAMSIZ]; 11563 11564 /* Ignore unmoveable devices (i.e. loopback) */ 11565 if (dev->features & NETIF_F_NETNS_LOCAL) 11566 continue; 11567 11568 /* Leave virtual devices for the generic cleanup */ 11569 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund) 11570 continue; 11571 11572 /* Push remaining network devices to init_net */ 11573 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); 11574 if (netdev_name_in_use(&init_net, fb_name)) 11575 snprintf(fb_name, IFNAMSIZ, "dev%%d"); 11576 err = dev_change_net_namespace(dev, &init_net, fb_name); 11577 if (err) { 11578 pr_emerg("%s: failed to move %s to init_net: %d\n", 11579 __func__, dev->name, err); 11580 BUG(); 11581 } 11582 } 11583 } 11584 11585 static void __net_exit default_device_exit_batch(struct list_head *net_list) 11586 { 11587 /* At exit all network devices most be removed from a network 11588 * namespace. Do this in the reverse order of registration. 11589 * Do this across as many network namespaces as possible to 11590 * improve batching efficiency. 11591 */ 11592 struct net_device *dev; 11593 struct net *net; 11594 LIST_HEAD(dev_kill_list); 11595 11596 rtnl_lock(); 11597 list_for_each_entry(net, net_list, exit_list) { 11598 default_device_exit_net(net); 11599 cond_resched(); 11600 } 11601 11602 list_for_each_entry(net, net_list, exit_list) { 11603 for_each_netdev_reverse(net, dev) { 11604 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) 11605 dev->rtnl_link_ops->dellink(dev, &dev_kill_list); 11606 else 11607 unregister_netdevice_queue(dev, &dev_kill_list); 11608 } 11609 } 11610 unregister_netdevice_many(&dev_kill_list); 11611 rtnl_unlock(); 11612 } 11613 11614 static struct pernet_operations __net_initdata default_device_ops = { 11615 .exit_batch = default_device_exit_batch, 11616 }; 11617 11618 static void __init net_dev_struct_check(void) 11619 { 11620 /* TX read-mostly hotpath */ 11621 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, priv_flags); 11622 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, netdev_ops); 11623 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, header_ops); 11624 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, _tx); 11625 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, real_num_tx_queues); 11626 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_max_size); 11627 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_ipv4_max_size); 11628 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_max_segs); 11629 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_partial_features); 11630 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, num_tc); 11631 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, mtu); 11632 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, needed_headroom); 11633 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, tc_to_txq); 11634 #ifdef CONFIG_XPS 11635 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, xps_maps); 11636 #endif 11637 #ifdef CONFIG_NETFILTER_EGRESS 11638 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, nf_hooks_egress); 11639 #endif 11640 #ifdef CONFIG_NET_XGRESS 11641 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, tcx_egress); 11642 #endif 11643 CACHELINE_ASSERT_GROUP_SIZE(struct net_device, net_device_read_tx, 160); 11644 11645 /* TXRX read-mostly hotpath */ 11646 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, flags); 11647 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, hard_header_len); 11648 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, features); 11649 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, ip6_ptr); 11650 CACHELINE_ASSERT_GROUP_SIZE(struct net_device, net_device_read_txrx, 30); 11651 11652 /* RX read-mostly hotpath */ 11653 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, ptype_specific); 11654 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, ifindex); 11655 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, real_num_rx_queues); 11656 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, _rx); 11657 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, gro_flush_timeout); 11658 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, napi_defer_hard_irqs); 11659 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, gro_max_size); 11660 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, gro_ipv4_max_size); 11661 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, rx_handler); 11662 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, rx_handler_data); 11663 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, nd_net); 11664 #ifdef CONFIG_NETPOLL 11665 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, npinfo); 11666 #endif 11667 #ifdef CONFIG_NET_XGRESS 11668 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, tcx_ingress); 11669 #endif 11670 CACHELINE_ASSERT_GROUP_SIZE(struct net_device, net_device_read_rx, 104); 11671 } 11672 11673 /* 11674 * Initialize the DEV module. At boot time this walks the device list and 11675 * unhooks any devices that fail to initialise (normally hardware not 11676 * present) and leaves us with a valid list of present and active devices. 11677 * 11678 */ 11679 11680 /* 11681 * This is called single threaded during boot, so no need 11682 * to take the rtnl semaphore. 11683 */ 11684 static int __init net_dev_init(void) 11685 { 11686 int i, rc = -ENOMEM; 11687 11688 BUG_ON(!dev_boot_phase); 11689 11690 net_dev_struct_check(); 11691 11692 if (dev_proc_init()) 11693 goto out; 11694 11695 if (netdev_kobject_init()) 11696 goto out; 11697 11698 INIT_LIST_HEAD(&ptype_all); 11699 for (i = 0; i < PTYPE_HASH_SIZE; i++) 11700 INIT_LIST_HEAD(&ptype_base[i]); 11701 11702 if (register_pernet_subsys(&netdev_net_ops)) 11703 goto out; 11704 11705 /* 11706 * Initialise the packet receive queues. 11707 */ 11708 11709 for_each_possible_cpu(i) { 11710 struct work_struct *flush = per_cpu_ptr(&flush_works, i); 11711 struct softnet_data *sd = &per_cpu(softnet_data, i); 11712 11713 INIT_WORK(flush, flush_backlog); 11714 11715 skb_queue_head_init(&sd->input_pkt_queue); 11716 skb_queue_head_init(&sd->process_queue); 11717 #ifdef CONFIG_XFRM_OFFLOAD 11718 skb_queue_head_init(&sd->xfrm_backlog); 11719 #endif 11720 INIT_LIST_HEAD(&sd->poll_list); 11721 sd->output_queue_tailp = &sd->output_queue; 11722 #ifdef CONFIG_RPS 11723 INIT_CSD(&sd->csd, rps_trigger_softirq, sd); 11724 sd->cpu = i; 11725 #endif 11726 INIT_CSD(&sd->defer_csd, trigger_rx_softirq, sd); 11727 spin_lock_init(&sd->defer_lock); 11728 11729 init_gro_hash(&sd->backlog); 11730 sd->backlog.poll = process_backlog; 11731 sd->backlog.weight = weight_p; 11732 } 11733 11734 dev_boot_phase = 0; 11735 11736 /* The loopback device is special if any other network devices 11737 * is present in a network namespace the loopback device must 11738 * be present. Since we now dynamically allocate and free the 11739 * loopback device ensure this invariant is maintained by 11740 * keeping the loopback device as the first device on the 11741 * list of network devices. Ensuring the loopback devices 11742 * is the first device that appears and the last network device 11743 * that disappears. 11744 */ 11745 if (register_pernet_device(&loopback_net_ops)) 11746 goto out; 11747 11748 if (register_pernet_device(&default_device_ops)) 11749 goto out; 11750 11751 open_softirq(NET_TX_SOFTIRQ, net_tx_action); 11752 open_softirq(NET_RX_SOFTIRQ, net_rx_action); 11753 11754 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", 11755 NULL, dev_cpu_dead); 11756 WARN_ON(rc < 0); 11757 rc = 0; 11758 out: 11759 return rc; 11760 } 11761 11762 subsys_initcall(net_dev_init); 11763