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 (!skb_shinfo(skb)->gso_type) { 3475 skb_warn_bad_offload(skb); 3476 return features & ~NETIF_F_GSO_MASK; 3477 } 3478 3479 /* Support for GSO partial features requires software 3480 * intervention before we can actually process the packets 3481 * so we need to strip support for any partial features now 3482 * and we can pull them back in after we have partially 3483 * segmented the frame. 3484 */ 3485 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)) 3486 features &= ~dev->gso_partial_features; 3487 3488 /* Make sure to clear the IPv4 ID mangling feature if the 3489 * IPv4 header has the potential to be fragmented. 3490 */ 3491 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { 3492 struct iphdr *iph = skb->encapsulation ? 3493 inner_ip_hdr(skb) : ip_hdr(skb); 3494 3495 if (!(iph->frag_off & htons(IP_DF))) 3496 features &= ~NETIF_F_TSO_MANGLEID; 3497 } 3498 3499 return features; 3500 } 3501 3502 netdev_features_t netif_skb_features(struct sk_buff *skb) 3503 { 3504 struct net_device *dev = skb->dev; 3505 netdev_features_t features = dev->features; 3506 3507 if (skb_is_gso(skb)) 3508 features = gso_features_check(skb, dev, features); 3509 3510 /* If encapsulation offload request, verify we are testing 3511 * hardware encapsulation features instead of standard 3512 * features for the netdev 3513 */ 3514 if (skb->encapsulation) 3515 features &= dev->hw_enc_features; 3516 3517 if (skb_vlan_tagged(skb)) 3518 features = netdev_intersect_features(features, 3519 dev->vlan_features | 3520 NETIF_F_HW_VLAN_CTAG_TX | 3521 NETIF_F_HW_VLAN_STAG_TX); 3522 3523 if (dev->netdev_ops->ndo_features_check) 3524 features &= dev->netdev_ops->ndo_features_check(skb, dev, 3525 features); 3526 else 3527 features &= dflt_features_check(skb, dev, features); 3528 3529 return harmonize_features(skb, features); 3530 } 3531 EXPORT_SYMBOL(netif_skb_features); 3532 3533 static int xmit_one(struct sk_buff *skb, struct net_device *dev, 3534 struct netdev_queue *txq, bool more) 3535 { 3536 unsigned int len; 3537 int rc; 3538 3539 if (dev_nit_active(dev)) 3540 dev_queue_xmit_nit(skb, dev); 3541 3542 len = skb->len; 3543 trace_net_dev_start_xmit(skb, dev); 3544 rc = netdev_start_xmit(skb, dev, txq, more); 3545 trace_net_dev_xmit(skb, rc, dev, len); 3546 3547 return rc; 3548 } 3549 3550 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, 3551 struct netdev_queue *txq, int *ret) 3552 { 3553 struct sk_buff *skb = first; 3554 int rc = NETDEV_TX_OK; 3555 3556 while (skb) { 3557 struct sk_buff *next = skb->next; 3558 3559 skb_mark_not_on_list(skb); 3560 rc = xmit_one(skb, dev, txq, next != NULL); 3561 if (unlikely(!dev_xmit_complete(rc))) { 3562 skb->next = next; 3563 goto out; 3564 } 3565 3566 skb = next; 3567 if (netif_tx_queue_stopped(txq) && skb) { 3568 rc = NETDEV_TX_BUSY; 3569 break; 3570 } 3571 } 3572 3573 out: 3574 *ret = rc; 3575 return skb; 3576 } 3577 3578 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, 3579 netdev_features_t features) 3580 { 3581 if (skb_vlan_tag_present(skb) && 3582 !vlan_hw_offload_capable(features, skb->vlan_proto)) 3583 skb = __vlan_hwaccel_push_inside(skb); 3584 return skb; 3585 } 3586 3587 int skb_csum_hwoffload_help(struct sk_buff *skb, 3588 const netdev_features_t features) 3589 { 3590 if (unlikely(skb_csum_is_sctp(skb))) 3591 return !!(features & NETIF_F_SCTP_CRC) ? 0 : 3592 skb_crc32c_csum_help(skb); 3593 3594 if (features & NETIF_F_HW_CSUM) 3595 return 0; 3596 3597 if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) { 3598 switch (skb->csum_offset) { 3599 case offsetof(struct tcphdr, check): 3600 case offsetof(struct udphdr, check): 3601 return 0; 3602 } 3603 } 3604 3605 return skb_checksum_help(skb); 3606 } 3607 EXPORT_SYMBOL(skb_csum_hwoffload_help); 3608 3609 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again) 3610 { 3611 netdev_features_t features; 3612 3613 features = netif_skb_features(skb); 3614 skb = validate_xmit_vlan(skb, features); 3615 if (unlikely(!skb)) 3616 goto out_null; 3617 3618 skb = sk_validate_xmit_skb(skb, dev); 3619 if (unlikely(!skb)) 3620 goto out_null; 3621 3622 if (netif_needs_gso(skb, features)) { 3623 struct sk_buff *segs; 3624 3625 segs = skb_gso_segment(skb, features); 3626 if (IS_ERR(segs)) { 3627 goto out_kfree_skb; 3628 } else if (segs) { 3629 consume_skb(skb); 3630 skb = segs; 3631 } 3632 } else { 3633 if (skb_needs_linearize(skb, features) && 3634 __skb_linearize(skb)) 3635 goto out_kfree_skb; 3636 3637 /* If packet is not checksummed and device does not 3638 * support checksumming for this protocol, complete 3639 * checksumming here. 3640 */ 3641 if (skb->ip_summed == CHECKSUM_PARTIAL) { 3642 if (skb->encapsulation) 3643 skb_set_inner_transport_header(skb, 3644 skb_checksum_start_offset(skb)); 3645 else 3646 skb_set_transport_header(skb, 3647 skb_checksum_start_offset(skb)); 3648 if (skb_csum_hwoffload_help(skb, features)) 3649 goto out_kfree_skb; 3650 } 3651 } 3652 3653 skb = validate_xmit_xfrm(skb, features, again); 3654 3655 return skb; 3656 3657 out_kfree_skb: 3658 kfree_skb(skb); 3659 out_null: 3660 dev_core_stats_tx_dropped_inc(dev); 3661 return NULL; 3662 } 3663 3664 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again) 3665 { 3666 struct sk_buff *next, *head = NULL, *tail; 3667 3668 for (; skb != NULL; skb = next) { 3669 next = skb->next; 3670 skb_mark_not_on_list(skb); 3671 3672 /* in case skb wont be segmented, point to itself */ 3673 skb->prev = skb; 3674 3675 skb = validate_xmit_skb(skb, dev, again); 3676 if (!skb) 3677 continue; 3678 3679 if (!head) 3680 head = skb; 3681 else 3682 tail->next = skb; 3683 /* If skb was segmented, skb->prev points to 3684 * the last segment. If not, it still contains skb. 3685 */ 3686 tail = skb->prev; 3687 } 3688 return head; 3689 } 3690 EXPORT_SYMBOL_GPL(validate_xmit_skb_list); 3691 3692 static void qdisc_pkt_len_init(struct sk_buff *skb) 3693 { 3694 const struct skb_shared_info *shinfo = skb_shinfo(skb); 3695 3696 qdisc_skb_cb(skb)->pkt_len = skb->len; 3697 3698 /* To get more precise estimation of bytes sent on wire, 3699 * we add to pkt_len the headers size of all segments 3700 */ 3701 if (shinfo->gso_size && skb_transport_header_was_set(skb)) { 3702 u16 gso_segs = shinfo->gso_segs; 3703 unsigned int hdr_len; 3704 3705 /* mac layer + network layer */ 3706 hdr_len = skb_transport_offset(skb); 3707 3708 /* + transport layer */ 3709 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { 3710 const struct tcphdr *th; 3711 struct tcphdr _tcphdr; 3712 3713 th = skb_header_pointer(skb, hdr_len, 3714 sizeof(_tcphdr), &_tcphdr); 3715 if (likely(th)) 3716 hdr_len += __tcp_hdrlen(th); 3717 } else { 3718 struct udphdr _udphdr; 3719 3720 if (skb_header_pointer(skb, hdr_len, 3721 sizeof(_udphdr), &_udphdr)) 3722 hdr_len += sizeof(struct udphdr); 3723 } 3724 3725 if (shinfo->gso_type & SKB_GSO_DODGY) 3726 gso_segs = DIV_ROUND_UP(skb->len - hdr_len, 3727 shinfo->gso_size); 3728 3729 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; 3730 } 3731 } 3732 3733 static int dev_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *q, 3734 struct sk_buff **to_free, 3735 struct netdev_queue *txq) 3736 { 3737 int rc; 3738 3739 rc = q->enqueue(skb, q, to_free) & NET_XMIT_MASK; 3740 if (rc == NET_XMIT_SUCCESS) 3741 trace_qdisc_enqueue(q, txq, skb); 3742 return rc; 3743 } 3744 3745 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, 3746 struct net_device *dev, 3747 struct netdev_queue *txq) 3748 { 3749 spinlock_t *root_lock = qdisc_lock(q); 3750 struct sk_buff *to_free = NULL; 3751 bool contended; 3752 int rc; 3753 3754 qdisc_calculate_pkt_len(skb, q); 3755 3756 if (q->flags & TCQ_F_NOLOCK) { 3757 if (q->flags & TCQ_F_CAN_BYPASS && nolock_qdisc_is_empty(q) && 3758 qdisc_run_begin(q)) { 3759 /* Retest nolock_qdisc_is_empty() within the protection 3760 * of q->seqlock to protect from racing with requeuing. 3761 */ 3762 if (unlikely(!nolock_qdisc_is_empty(q))) { 3763 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3764 __qdisc_run(q); 3765 qdisc_run_end(q); 3766 3767 goto no_lock_out; 3768 } 3769 3770 qdisc_bstats_cpu_update(q, skb); 3771 if (sch_direct_xmit(skb, q, dev, txq, NULL, true) && 3772 !nolock_qdisc_is_empty(q)) 3773 __qdisc_run(q); 3774 3775 qdisc_run_end(q); 3776 return NET_XMIT_SUCCESS; 3777 } 3778 3779 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3780 qdisc_run(q); 3781 3782 no_lock_out: 3783 if (unlikely(to_free)) 3784 kfree_skb_list_reason(to_free, 3785 SKB_DROP_REASON_QDISC_DROP); 3786 return rc; 3787 } 3788 3789 /* 3790 * Heuristic to force contended enqueues to serialize on a 3791 * separate lock before trying to get qdisc main lock. 3792 * This permits qdisc->running owner to get the lock more 3793 * often and dequeue packets faster. 3794 * On PREEMPT_RT it is possible to preempt the qdisc owner during xmit 3795 * and then other tasks will only enqueue packets. The packets will be 3796 * sent after the qdisc owner is scheduled again. To prevent this 3797 * scenario the task always serialize on the lock. 3798 */ 3799 contended = qdisc_is_running(q) || IS_ENABLED(CONFIG_PREEMPT_RT); 3800 if (unlikely(contended)) 3801 spin_lock(&q->busylock); 3802 3803 spin_lock(root_lock); 3804 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { 3805 __qdisc_drop(skb, &to_free); 3806 rc = NET_XMIT_DROP; 3807 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && 3808 qdisc_run_begin(q)) { 3809 /* 3810 * This is a work-conserving queue; there are no old skbs 3811 * waiting to be sent out; and the qdisc is not running - 3812 * xmit the skb directly. 3813 */ 3814 3815 qdisc_bstats_update(q, skb); 3816 3817 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { 3818 if (unlikely(contended)) { 3819 spin_unlock(&q->busylock); 3820 contended = false; 3821 } 3822 __qdisc_run(q); 3823 } 3824 3825 qdisc_run_end(q); 3826 rc = NET_XMIT_SUCCESS; 3827 } else { 3828 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3829 if (qdisc_run_begin(q)) { 3830 if (unlikely(contended)) { 3831 spin_unlock(&q->busylock); 3832 contended = false; 3833 } 3834 __qdisc_run(q); 3835 qdisc_run_end(q); 3836 } 3837 } 3838 spin_unlock(root_lock); 3839 if (unlikely(to_free)) 3840 kfree_skb_list_reason(to_free, SKB_DROP_REASON_QDISC_DROP); 3841 if (unlikely(contended)) 3842 spin_unlock(&q->busylock); 3843 return rc; 3844 } 3845 3846 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) 3847 static void skb_update_prio(struct sk_buff *skb) 3848 { 3849 const struct netprio_map *map; 3850 const struct sock *sk; 3851 unsigned int prioidx; 3852 3853 if (skb->priority) 3854 return; 3855 map = rcu_dereference_bh(skb->dev->priomap); 3856 if (!map) 3857 return; 3858 sk = skb_to_full_sk(skb); 3859 if (!sk) 3860 return; 3861 3862 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data); 3863 3864 if (prioidx < map->priomap_len) 3865 skb->priority = map->priomap[prioidx]; 3866 } 3867 #else 3868 #define skb_update_prio(skb) 3869 #endif 3870 3871 /** 3872 * dev_loopback_xmit - loop back @skb 3873 * @net: network namespace this loopback is happening in 3874 * @sk: sk needed to be a netfilter okfn 3875 * @skb: buffer to transmit 3876 */ 3877 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) 3878 { 3879 skb_reset_mac_header(skb); 3880 __skb_pull(skb, skb_network_offset(skb)); 3881 skb->pkt_type = PACKET_LOOPBACK; 3882 if (skb->ip_summed == CHECKSUM_NONE) 3883 skb->ip_summed = CHECKSUM_UNNECESSARY; 3884 DEBUG_NET_WARN_ON_ONCE(!skb_dst(skb)); 3885 skb_dst_force(skb); 3886 netif_rx(skb); 3887 return 0; 3888 } 3889 EXPORT_SYMBOL(dev_loopback_xmit); 3890 3891 #ifdef CONFIG_NET_EGRESS 3892 static struct netdev_queue * 3893 netdev_tx_queue_mapping(struct net_device *dev, struct sk_buff *skb) 3894 { 3895 int qm = skb_get_queue_mapping(skb); 3896 3897 return netdev_get_tx_queue(dev, netdev_cap_txqueue(dev, qm)); 3898 } 3899 3900 static bool netdev_xmit_txqueue_skipped(void) 3901 { 3902 return __this_cpu_read(softnet_data.xmit.skip_txqueue); 3903 } 3904 3905 void netdev_xmit_skip_txqueue(bool skip) 3906 { 3907 __this_cpu_write(softnet_data.xmit.skip_txqueue, skip); 3908 } 3909 EXPORT_SYMBOL_GPL(netdev_xmit_skip_txqueue); 3910 #endif /* CONFIG_NET_EGRESS */ 3911 3912 #ifdef CONFIG_NET_XGRESS 3913 static int tc_run(struct tcx_entry *entry, struct sk_buff *skb, 3914 enum skb_drop_reason *drop_reason) 3915 { 3916 int ret = TC_ACT_UNSPEC; 3917 #ifdef CONFIG_NET_CLS_ACT 3918 struct mini_Qdisc *miniq = rcu_dereference_bh(entry->miniq); 3919 struct tcf_result res; 3920 3921 if (!miniq) 3922 return ret; 3923 3924 tc_skb_cb(skb)->mru = 0; 3925 tc_skb_cb(skb)->post_ct = false; 3926 res.drop_reason = *drop_reason; 3927 3928 mini_qdisc_bstats_cpu_update(miniq, skb); 3929 ret = tcf_classify(skb, miniq->block, miniq->filter_list, &res, false); 3930 /* Only tcf related quirks below. */ 3931 switch (ret) { 3932 case TC_ACT_SHOT: 3933 *drop_reason = res.drop_reason; 3934 mini_qdisc_qstats_cpu_drop(miniq); 3935 break; 3936 case TC_ACT_OK: 3937 case TC_ACT_RECLASSIFY: 3938 skb->tc_index = TC_H_MIN(res.classid); 3939 break; 3940 } 3941 #endif /* CONFIG_NET_CLS_ACT */ 3942 return ret; 3943 } 3944 3945 static DEFINE_STATIC_KEY_FALSE(tcx_needed_key); 3946 3947 void tcx_inc(void) 3948 { 3949 static_branch_inc(&tcx_needed_key); 3950 } 3951 3952 void tcx_dec(void) 3953 { 3954 static_branch_dec(&tcx_needed_key); 3955 } 3956 3957 static __always_inline enum tcx_action_base 3958 tcx_run(const struct bpf_mprog_entry *entry, struct sk_buff *skb, 3959 const bool needs_mac) 3960 { 3961 const struct bpf_mprog_fp *fp; 3962 const struct bpf_prog *prog; 3963 int ret = TCX_NEXT; 3964 3965 if (needs_mac) 3966 __skb_push(skb, skb->mac_len); 3967 bpf_mprog_foreach_prog(entry, fp, prog) { 3968 bpf_compute_data_pointers(skb); 3969 ret = bpf_prog_run(prog, skb); 3970 if (ret != TCX_NEXT) 3971 break; 3972 } 3973 if (needs_mac) 3974 __skb_pull(skb, skb->mac_len); 3975 return tcx_action_code(skb, ret); 3976 } 3977 3978 static __always_inline struct sk_buff * 3979 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 3980 struct net_device *orig_dev, bool *another) 3981 { 3982 struct bpf_mprog_entry *entry = rcu_dereference_bh(skb->dev->tcx_ingress); 3983 enum skb_drop_reason drop_reason = SKB_DROP_REASON_TC_INGRESS; 3984 int sch_ret; 3985 3986 if (!entry) 3987 return skb; 3988 if (*pt_prev) { 3989 *ret = deliver_skb(skb, *pt_prev, orig_dev); 3990 *pt_prev = NULL; 3991 } 3992 3993 qdisc_skb_cb(skb)->pkt_len = skb->len; 3994 tcx_set_ingress(skb, true); 3995 3996 if (static_branch_unlikely(&tcx_needed_key)) { 3997 sch_ret = tcx_run(entry, skb, true); 3998 if (sch_ret != TC_ACT_UNSPEC) 3999 goto ingress_verdict; 4000 } 4001 sch_ret = tc_run(tcx_entry(entry), skb, &drop_reason); 4002 ingress_verdict: 4003 switch (sch_ret) { 4004 case TC_ACT_REDIRECT: 4005 /* skb_mac_header check was done by BPF, so we can safely 4006 * push the L2 header back before redirecting to another 4007 * netdev. 4008 */ 4009 __skb_push(skb, skb->mac_len); 4010 if (skb_do_redirect(skb) == -EAGAIN) { 4011 __skb_pull(skb, skb->mac_len); 4012 *another = true; 4013 break; 4014 } 4015 *ret = NET_RX_SUCCESS; 4016 return NULL; 4017 case TC_ACT_SHOT: 4018 kfree_skb_reason(skb, drop_reason); 4019 *ret = NET_RX_DROP; 4020 return NULL; 4021 /* used by tc_run */ 4022 case TC_ACT_STOLEN: 4023 case TC_ACT_QUEUED: 4024 case TC_ACT_TRAP: 4025 consume_skb(skb); 4026 fallthrough; 4027 case TC_ACT_CONSUMED: 4028 *ret = NET_RX_SUCCESS; 4029 return NULL; 4030 } 4031 4032 return skb; 4033 } 4034 4035 static __always_inline struct sk_buff * 4036 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 4037 { 4038 struct bpf_mprog_entry *entry = rcu_dereference_bh(dev->tcx_egress); 4039 enum skb_drop_reason drop_reason = SKB_DROP_REASON_TC_EGRESS; 4040 int sch_ret; 4041 4042 if (!entry) 4043 return skb; 4044 4045 /* qdisc_skb_cb(skb)->pkt_len & tcx_set_ingress() was 4046 * already set by the caller. 4047 */ 4048 if (static_branch_unlikely(&tcx_needed_key)) { 4049 sch_ret = tcx_run(entry, skb, false); 4050 if (sch_ret != TC_ACT_UNSPEC) 4051 goto egress_verdict; 4052 } 4053 sch_ret = tc_run(tcx_entry(entry), skb, &drop_reason); 4054 egress_verdict: 4055 switch (sch_ret) { 4056 case TC_ACT_REDIRECT: 4057 /* No need to push/pop skb's mac_header here on egress! */ 4058 skb_do_redirect(skb); 4059 *ret = NET_XMIT_SUCCESS; 4060 return NULL; 4061 case TC_ACT_SHOT: 4062 kfree_skb_reason(skb, drop_reason); 4063 *ret = NET_XMIT_DROP; 4064 return NULL; 4065 /* used by tc_run */ 4066 case TC_ACT_STOLEN: 4067 case TC_ACT_QUEUED: 4068 case TC_ACT_TRAP: 4069 consume_skb(skb); 4070 fallthrough; 4071 case TC_ACT_CONSUMED: 4072 *ret = NET_XMIT_SUCCESS; 4073 return NULL; 4074 } 4075 4076 return skb; 4077 } 4078 #else 4079 static __always_inline struct sk_buff * 4080 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 4081 struct net_device *orig_dev, bool *another) 4082 { 4083 return skb; 4084 } 4085 4086 static __always_inline struct sk_buff * 4087 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 4088 { 4089 return skb; 4090 } 4091 #endif /* CONFIG_NET_XGRESS */ 4092 4093 #ifdef CONFIG_XPS 4094 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb, 4095 struct xps_dev_maps *dev_maps, unsigned int tci) 4096 { 4097 int tc = netdev_get_prio_tc_map(dev, skb->priority); 4098 struct xps_map *map; 4099 int queue_index = -1; 4100 4101 if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids) 4102 return queue_index; 4103 4104 tci *= dev_maps->num_tc; 4105 tci += tc; 4106 4107 map = rcu_dereference(dev_maps->attr_map[tci]); 4108 if (map) { 4109 if (map->len == 1) 4110 queue_index = map->queues[0]; 4111 else 4112 queue_index = map->queues[reciprocal_scale( 4113 skb_get_hash(skb), map->len)]; 4114 if (unlikely(queue_index >= dev->real_num_tx_queues)) 4115 queue_index = -1; 4116 } 4117 return queue_index; 4118 } 4119 #endif 4120 4121 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev, 4122 struct sk_buff *skb) 4123 { 4124 #ifdef CONFIG_XPS 4125 struct xps_dev_maps *dev_maps; 4126 struct sock *sk = skb->sk; 4127 int queue_index = -1; 4128 4129 if (!static_key_false(&xps_needed)) 4130 return -1; 4131 4132 rcu_read_lock(); 4133 if (!static_key_false(&xps_rxqs_needed)) 4134 goto get_cpus_map; 4135 4136 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]); 4137 if (dev_maps) { 4138 int tci = sk_rx_queue_get(sk); 4139 4140 if (tci >= 0) 4141 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4142 tci); 4143 } 4144 4145 get_cpus_map: 4146 if (queue_index < 0) { 4147 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]); 4148 if (dev_maps) { 4149 unsigned int tci = skb->sender_cpu - 1; 4150 4151 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4152 tci); 4153 } 4154 } 4155 rcu_read_unlock(); 4156 4157 return queue_index; 4158 #else 4159 return -1; 4160 #endif 4161 } 4162 4163 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb, 4164 struct net_device *sb_dev) 4165 { 4166 return 0; 4167 } 4168 EXPORT_SYMBOL(dev_pick_tx_zero); 4169 4170 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb, 4171 struct net_device *sb_dev) 4172 { 4173 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues; 4174 } 4175 EXPORT_SYMBOL(dev_pick_tx_cpu_id); 4176 4177 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb, 4178 struct net_device *sb_dev) 4179 { 4180 struct sock *sk = skb->sk; 4181 int queue_index = sk_tx_queue_get(sk); 4182 4183 sb_dev = sb_dev ? : dev; 4184 4185 if (queue_index < 0 || skb->ooo_okay || 4186 queue_index >= dev->real_num_tx_queues) { 4187 int new_index = get_xps_queue(dev, sb_dev, skb); 4188 4189 if (new_index < 0) 4190 new_index = skb_tx_hash(dev, sb_dev, skb); 4191 4192 if (queue_index != new_index && sk && 4193 sk_fullsock(sk) && 4194 rcu_access_pointer(sk->sk_dst_cache)) 4195 sk_tx_queue_set(sk, new_index); 4196 4197 queue_index = new_index; 4198 } 4199 4200 return queue_index; 4201 } 4202 EXPORT_SYMBOL(netdev_pick_tx); 4203 4204 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev, 4205 struct sk_buff *skb, 4206 struct net_device *sb_dev) 4207 { 4208 int queue_index = 0; 4209 4210 #ifdef CONFIG_XPS 4211 u32 sender_cpu = skb->sender_cpu - 1; 4212 4213 if (sender_cpu >= (u32)NR_CPUS) 4214 skb->sender_cpu = raw_smp_processor_id() + 1; 4215 #endif 4216 4217 if (dev->real_num_tx_queues != 1) { 4218 const struct net_device_ops *ops = dev->netdev_ops; 4219 4220 if (ops->ndo_select_queue) 4221 queue_index = ops->ndo_select_queue(dev, skb, sb_dev); 4222 else 4223 queue_index = netdev_pick_tx(dev, skb, sb_dev); 4224 4225 queue_index = netdev_cap_txqueue(dev, queue_index); 4226 } 4227 4228 skb_set_queue_mapping(skb, queue_index); 4229 return netdev_get_tx_queue(dev, queue_index); 4230 } 4231 4232 /** 4233 * __dev_queue_xmit() - transmit a buffer 4234 * @skb: buffer to transmit 4235 * @sb_dev: suboordinate device used for L2 forwarding offload 4236 * 4237 * Queue a buffer for transmission to a network device. The caller must 4238 * have set the device and priority and built the buffer before calling 4239 * this function. The function can be called from an interrupt. 4240 * 4241 * When calling this method, interrupts MUST be enabled. This is because 4242 * the BH enable code must have IRQs enabled so that it will not deadlock. 4243 * 4244 * Regardless of the return value, the skb is consumed, so it is currently 4245 * difficult to retry a send to this method. (You can bump the ref count 4246 * before sending to hold a reference for retry if you are careful.) 4247 * 4248 * Return: 4249 * * 0 - buffer successfully transmitted 4250 * * positive qdisc return code - NET_XMIT_DROP etc. 4251 * * negative errno - other errors 4252 */ 4253 int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev) 4254 { 4255 struct net_device *dev = skb->dev; 4256 struct netdev_queue *txq = NULL; 4257 struct Qdisc *q; 4258 int rc = -ENOMEM; 4259 bool again = false; 4260 4261 skb_reset_mac_header(skb); 4262 skb_assert_len(skb); 4263 4264 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) 4265 __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED); 4266 4267 /* Disable soft irqs for various locks below. Also 4268 * stops preemption for RCU. 4269 */ 4270 rcu_read_lock_bh(); 4271 4272 skb_update_prio(skb); 4273 4274 qdisc_pkt_len_init(skb); 4275 tcx_set_ingress(skb, false); 4276 #ifdef CONFIG_NET_EGRESS 4277 if (static_branch_unlikely(&egress_needed_key)) { 4278 if (nf_hook_egress_active()) { 4279 skb = nf_hook_egress(skb, &rc, dev); 4280 if (!skb) 4281 goto out; 4282 } 4283 4284 netdev_xmit_skip_txqueue(false); 4285 4286 nf_skip_egress(skb, true); 4287 skb = sch_handle_egress(skb, &rc, dev); 4288 if (!skb) 4289 goto out; 4290 nf_skip_egress(skb, false); 4291 4292 if (netdev_xmit_txqueue_skipped()) 4293 txq = netdev_tx_queue_mapping(dev, skb); 4294 } 4295 #endif 4296 /* If device/qdisc don't need skb->dst, release it right now while 4297 * its hot in this cpu cache. 4298 */ 4299 if (dev->priv_flags & IFF_XMIT_DST_RELEASE) 4300 skb_dst_drop(skb); 4301 else 4302 skb_dst_force(skb); 4303 4304 if (!txq) 4305 txq = netdev_core_pick_tx(dev, skb, sb_dev); 4306 4307 q = rcu_dereference_bh(txq->qdisc); 4308 4309 trace_net_dev_queue(skb); 4310 if (q->enqueue) { 4311 rc = __dev_xmit_skb(skb, q, dev, txq); 4312 goto out; 4313 } 4314 4315 /* The device has no queue. Common case for software devices: 4316 * loopback, all the sorts of tunnels... 4317 4318 * Really, it is unlikely that netif_tx_lock protection is necessary 4319 * here. (f.e. loopback and IP tunnels are clean ignoring statistics 4320 * counters.) 4321 * However, it is possible, that they rely on protection 4322 * made by us here. 4323 4324 * Check this and shot the lock. It is not prone from deadlocks. 4325 *Either shot noqueue qdisc, it is even simpler 8) 4326 */ 4327 if (dev->flags & IFF_UP) { 4328 int cpu = smp_processor_id(); /* ok because BHs are off */ 4329 4330 /* Other cpus might concurrently change txq->xmit_lock_owner 4331 * to -1 or to their cpu id, but not to our id. 4332 */ 4333 if (READ_ONCE(txq->xmit_lock_owner) != cpu) { 4334 if (dev_xmit_recursion()) 4335 goto recursion_alert; 4336 4337 skb = validate_xmit_skb(skb, dev, &again); 4338 if (!skb) 4339 goto out; 4340 4341 HARD_TX_LOCK(dev, txq, cpu); 4342 4343 if (!netif_xmit_stopped(txq)) { 4344 dev_xmit_recursion_inc(); 4345 skb = dev_hard_start_xmit(skb, dev, txq, &rc); 4346 dev_xmit_recursion_dec(); 4347 if (dev_xmit_complete(rc)) { 4348 HARD_TX_UNLOCK(dev, txq); 4349 goto out; 4350 } 4351 } 4352 HARD_TX_UNLOCK(dev, txq); 4353 net_crit_ratelimited("Virtual device %s asks to queue packet!\n", 4354 dev->name); 4355 } else { 4356 /* Recursion is detected! It is possible, 4357 * unfortunately 4358 */ 4359 recursion_alert: 4360 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", 4361 dev->name); 4362 } 4363 } 4364 4365 rc = -ENETDOWN; 4366 rcu_read_unlock_bh(); 4367 4368 dev_core_stats_tx_dropped_inc(dev); 4369 kfree_skb_list(skb); 4370 return rc; 4371 out: 4372 rcu_read_unlock_bh(); 4373 return rc; 4374 } 4375 EXPORT_SYMBOL(__dev_queue_xmit); 4376 4377 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id) 4378 { 4379 struct net_device *dev = skb->dev; 4380 struct sk_buff *orig_skb = skb; 4381 struct netdev_queue *txq; 4382 int ret = NETDEV_TX_BUSY; 4383 bool again = false; 4384 4385 if (unlikely(!netif_running(dev) || 4386 !netif_carrier_ok(dev))) 4387 goto drop; 4388 4389 skb = validate_xmit_skb_list(skb, dev, &again); 4390 if (skb != orig_skb) 4391 goto drop; 4392 4393 skb_set_queue_mapping(skb, queue_id); 4394 txq = skb_get_tx_queue(dev, skb); 4395 4396 local_bh_disable(); 4397 4398 dev_xmit_recursion_inc(); 4399 HARD_TX_LOCK(dev, txq, smp_processor_id()); 4400 if (!netif_xmit_frozen_or_drv_stopped(txq)) 4401 ret = netdev_start_xmit(skb, dev, txq, false); 4402 HARD_TX_UNLOCK(dev, txq); 4403 dev_xmit_recursion_dec(); 4404 4405 local_bh_enable(); 4406 return ret; 4407 drop: 4408 dev_core_stats_tx_dropped_inc(dev); 4409 kfree_skb_list(skb); 4410 return NET_XMIT_DROP; 4411 } 4412 EXPORT_SYMBOL(__dev_direct_xmit); 4413 4414 /************************************************************************* 4415 * Receiver routines 4416 *************************************************************************/ 4417 4418 int netdev_max_backlog __read_mostly = 1000; 4419 EXPORT_SYMBOL(netdev_max_backlog); 4420 4421 int netdev_tstamp_prequeue __read_mostly = 1; 4422 unsigned int sysctl_skb_defer_max __read_mostly = 64; 4423 int netdev_budget __read_mostly = 300; 4424 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */ 4425 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ; 4426 int weight_p __read_mostly = 64; /* old backlog weight */ 4427 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ 4428 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ 4429 int dev_rx_weight __read_mostly = 64; 4430 int dev_tx_weight __read_mostly = 64; 4431 4432 /* Called with irq disabled */ 4433 static inline void ____napi_schedule(struct softnet_data *sd, 4434 struct napi_struct *napi) 4435 { 4436 struct task_struct *thread; 4437 4438 lockdep_assert_irqs_disabled(); 4439 4440 if (test_bit(NAPI_STATE_THREADED, &napi->state)) { 4441 /* Paired with smp_mb__before_atomic() in 4442 * napi_enable()/dev_set_threaded(). 4443 * Use READ_ONCE() to guarantee a complete 4444 * read on napi->thread. Only call 4445 * wake_up_process() when it's not NULL. 4446 */ 4447 thread = READ_ONCE(napi->thread); 4448 if (thread) { 4449 /* Avoid doing set_bit() if the thread is in 4450 * INTERRUPTIBLE state, cause napi_thread_wait() 4451 * makes sure to proceed with napi polling 4452 * if the thread is explicitly woken from here. 4453 */ 4454 if (READ_ONCE(thread->__state) != TASK_INTERRUPTIBLE) 4455 set_bit(NAPI_STATE_SCHED_THREADED, &napi->state); 4456 wake_up_process(thread); 4457 return; 4458 } 4459 } 4460 4461 list_add_tail(&napi->poll_list, &sd->poll_list); 4462 WRITE_ONCE(napi->list_owner, smp_processor_id()); 4463 /* If not called from net_rx_action() 4464 * we have to raise NET_RX_SOFTIRQ. 4465 */ 4466 if (!sd->in_net_rx_action) 4467 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4468 } 4469 4470 #ifdef CONFIG_RPS 4471 4472 /* One global table that all flow-based protocols share. */ 4473 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; 4474 EXPORT_SYMBOL(rps_sock_flow_table); 4475 u32 rps_cpu_mask __read_mostly; 4476 EXPORT_SYMBOL(rps_cpu_mask); 4477 4478 struct static_key_false rps_needed __read_mostly; 4479 EXPORT_SYMBOL(rps_needed); 4480 struct static_key_false rfs_needed __read_mostly; 4481 EXPORT_SYMBOL(rfs_needed); 4482 4483 static struct rps_dev_flow * 4484 set_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4485 struct rps_dev_flow *rflow, u16 next_cpu) 4486 { 4487 if (next_cpu < nr_cpu_ids) { 4488 #ifdef CONFIG_RFS_ACCEL 4489 struct netdev_rx_queue *rxqueue; 4490 struct rps_dev_flow_table *flow_table; 4491 struct rps_dev_flow *old_rflow; 4492 u32 flow_id; 4493 u16 rxq_index; 4494 int rc; 4495 4496 /* Should we steer this flow to a different hardware queue? */ 4497 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || 4498 !(dev->features & NETIF_F_NTUPLE)) 4499 goto out; 4500 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); 4501 if (rxq_index == skb_get_rx_queue(skb)) 4502 goto out; 4503 4504 rxqueue = dev->_rx + rxq_index; 4505 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4506 if (!flow_table) 4507 goto out; 4508 flow_id = skb_get_hash(skb) & flow_table->mask; 4509 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, 4510 rxq_index, flow_id); 4511 if (rc < 0) 4512 goto out; 4513 old_rflow = rflow; 4514 rflow = &flow_table->flows[flow_id]; 4515 rflow->filter = rc; 4516 if (old_rflow->filter == rflow->filter) 4517 old_rflow->filter = RPS_NO_FILTER; 4518 out: 4519 #endif 4520 rflow->last_qtail = 4521 per_cpu(softnet_data, next_cpu).input_queue_head; 4522 } 4523 4524 rflow->cpu = next_cpu; 4525 return rflow; 4526 } 4527 4528 /* 4529 * get_rps_cpu is called from netif_receive_skb and returns the target 4530 * CPU from the RPS map of the receiving queue for a given skb. 4531 * rcu_read_lock must be held on entry. 4532 */ 4533 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4534 struct rps_dev_flow **rflowp) 4535 { 4536 const struct rps_sock_flow_table *sock_flow_table; 4537 struct netdev_rx_queue *rxqueue = dev->_rx; 4538 struct rps_dev_flow_table *flow_table; 4539 struct rps_map *map; 4540 int cpu = -1; 4541 u32 tcpu; 4542 u32 hash; 4543 4544 if (skb_rx_queue_recorded(skb)) { 4545 u16 index = skb_get_rx_queue(skb); 4546 4547 if (unlikely(index >= dev->real_num_rx_queues)) { 4548 WARN_ONCE(dev->real_num_rx_queues > 1, 4549 "%s received packet on queue %u, but number " 4550 "of RX queues is %u\n", 4551 dev->name, index, dev->real_num_rx_queues); 4552 goto done; 4553 } 4554 rxqueue += index; 4555 } 4556 4557 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ 4558 4559 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4560 map = rcu_dereference(rxqueue->rps_map); 4561 if (!flow_table && !map) 4562 goto done; 4563 4564 skb_reset_network_header(skb); 4565 hash = skb_get_hash(skb); 4566 if (!hash) 4567 goto done; 4568 4569 sock_flow_table = rcu_dereference(rps_sock_flow_table); 4570 if (flow_table && sock_flow_table) { 4571 struct rps_dev_flow *rflow; 4572 u32 next_cpu; 4573 u32 ident; 4574 4575 /* First check into global flow table if there is a match. 4576 * This READ_ONCE() pairs with WRITE_ONCE() from rps_record_sock_flow(). 4577 */ 4578 ident = READ_ONCE(sock_flow_table->ents[hash & sock_flow_table->mask]); 4579 if ((ident ^ hash) & ~rps_cpu_mask) 4580 goto try_rps; 4581 4582 next_cpu = ident & rps_cpu_mask; 4583 4584 /* OK, now we know there is a match, 4585 * we can look at the local (per receive queue) flow table 4586 */ 4587 rflow = &flow_table->flows[hash & flow_table->mask]; 4588 tcpu = rflow->cpu; 4589 4590 /* 4591 * If the desired CPU (where last recvmsg was done) is 4592 * different from current CPU (one in the rx-queue flow 4593 * table entry), switch if one of the following holds: 4594 * - Current CPU is unset (>= nr_cpu_ids). 4595 * - Current CPU is offline. 4596 * - The current CPU's queue tail has advanced beyond the 4597 * last packet that was enqueued using this table entry. 4598 * This guarantees that all previous packets for the flow 4599 * have been dequeued, thus preserving in order delivery. 4600 */ 4601 if (unlikely(tcpu != next_cpu) && 4602 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || 4603 ((int)(per_cpu(softnet_data, tcpu).input_queue_head - 4604 rflow->last_qtail)) >= 0)) { 4605 tcpu = next_cpu; 4606 rflow = set_rps_cpu(dev, skb, rflow, next_cpu); 4607 } 4608 4609 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { 4610 *rflowp = rflow; 4611 cpu = tcpu; 4612 goto done; 4613 } 4614 } 4615 4616 try_rps: 4617 4618 if (map) { 4619 tcpu = map->cpus[reciprocal_scale(hash, map->len)]; 4620 if (cpu_online(tcpu)) { 4621 cpu = tcpu; 4622 goto done; 4623 } 4624 } 4625 4626 done: 4627 return cpu; 4628 } 4629 4630 #ifdef CONFIG_RFS_ACCEL 4631 4632 /** 4633 * rps_may_expire_flow - check whether an RFS hardware filter may be removed 4634 * @dev: Device on which the filter was set 4635 * @rxq_index: RX queue index 4636 * @flow_id: Flow ID passed to ndo_rx_flow_steer() 4637 * @filter_id: Filter ID returned by ndo_rx_flow_steer() 4638 * 4639 * Drivers that implement ndo_rx_flow_steer() should periodically call 4640 * this function for each installed filter and remove the filters for 4641 * which it returns %true. 4642 */ 4643 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, 4644 u32 flow_id, u16 filter_id) 4645 { 4646 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; 4647 struct rps_dev_flow_table *flow_table; 4648 struct rps_dev_flow *rflow; 4649 bool expire = true; 4650 unsigned int cpu; 4651 4652 rcu_read_lock(); 4653 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4654 if (flow_table && flow_id <= flow_table->mask) { 4655 rflow = &flow_table->flows[flow_id]; 4656 cpu = READ_ONCE(rflow->cpu); 4657 if (rflow->filter == filter_id && cpu < nr_cpu_ids && 4658 ((int)(per_cpu(softnet_data, cpu).input_queue_head - 4659 rflow->last_qtail) < 4660 (int)(10 * flow_table->mask))) 4661 expire = false; 4662 } 4663 rcu_read_unlock(); 4664 return expire; 4665 } 4666 EXPORT_SYMBOL(rps_may_expire_flow); 4667 4668 #endif /* CONFIG_RFS_ACCEL */ 4669 4670 /* Called from hardirq (IPI) context */ 4671 static void rps_trigger_softirq(void *data) 4672 { 4673 struct softnet_data *sd = data; 4674 4675 ____napi_schedule(sd, &sd->backlog); 4676 sd->received_rps++; 4677 } 4678 4679 #endif /* CONFIG_RPS */ 4680 4681 /* Called from hardirq (IPI) context */ 4682 static void trigger_rx_softirq(void *data) 4683 { 4684 struct softnet_data *sd = data; 4685 4686 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4687 smp_store_release(&sd->defer_ipi_scheduled, 0); 4688 } 4689 4690 /* 4691 * After we queued a packet into sd->input_pkt_queue, 4692 * we need to make sure this queue is serviced soon. 4693 * 4694 * - If this is another cpu queue, link it to our rps_ipi_list, 4695 * and make sure we will process rps_ipi_list from net_rx_action(). 4696 * 4697 * - If this is our own queue, NAPI schedule our backlog. 4698 * Note that this also raises NET_RX_SOFTIRQ. 4699 */ 4700 static void napi_schedule_rps(struct softnet_data *sd) 4701 { 4702 struct softnet_data *mysd = this_cpu_ptr(&softnet_data); 4703 4704 #ifdef CONFIG_RPS 4705 if (sd != mysd) { 4706 sd->rps_ipi_next = mysd->rps_ipi_list; 4707 mysd->rps_ipi_list = sd; 4708 4709 /* If not called from net_rx_action() or napi_threaded_poll() 4710 * we have to raise NET_RX_SOFTIRQ. 4711 */ 4712 if (!mysd->in_net_rx_action && !mysd->in_napi_threaded_poll) 4713 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4714 return; 4715 } 4716 #endif /* CONFIG_RPS */ 4717 __napi_schedule_irqoff(&mysd->backlog); 4718 } 4719 4720 #ifdef CONFIG_NET_FLOW_LIMIT 4721 int netdev_flow_limit_table_len __read_mostly = (1 << 12); 4722 #endif 4723 4724 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) 4725 { 4726 #ifdef CONFIG_NET_FLOW_LIMIT 4727 struct sd_flow_limit *fl; 4728 struct softnet_data *sd; 4729 unsigned int old_flow, new_flow; 4730 4731 if (qlen < (READ_ONCE(netdev_max_backlog) >> 1)) 4732 return false; 4733 4734 sd = this_cpu_ptr(&softnet_data); 4735 4736 rcu_read_lock(); 4737 fl = rcu_dereference(sd->flow_limit); 4738 if (fl) { 4739 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); 4740 old_flow = fl->history[fl->history_head]; 4741 fl->history[fl->history_head] = new_flow; 4742 4743 fl->history_head++; 4744 fl->history_head &= FLOW_LIMIT_HISTORY - 1; 4745 4746 if (likely(fl->buckets[old_flow])) 4747 fl->buckets[old_flow]--; 4748 4749 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { 4750 fl->count++; 4751 rcu_read_unlock(); 4752 return true; 4753 } 4754 } 4755 rcu_read_unlock(); 4756 #endif 4757 return false; 4758 } 4759 4760 /* 4761 * enqueue_to_backlog is called to queue an skb to a per CPU backlog 4762 * queue (may be a remote CPU queue). 4763 */ 4764 static int enqueue_to_backlog(struct sk_buff *skb, int cpu, 4765 unsigned int *qtail) 4766 { 4767 enum skb_drop_reason reason; 4768 struct softnet_data *sd; 4769 unsigned long flags; 4770 unsigned int qlen; 4771 4772 reason = SKB_DROP_REASON_NOT_SPECIFIED; 4773 sd = &per_cpu(softnet_data, cpu); 4774 4775 rps_lock_irqsave(sd, &flags); 4776 if (!netif_running(skb->dev)) 4777 goto drop; 4778 qlen = skb_queue_len(&sd->input_pkt_queue); 4779 if (qlen <= READ_ONCE(netdev_max_backlog) && !skb_flow_limit(skb, qlen)) { 4780 if (qlen) { 4781 enqueue: 4782 __skb_queue_tail(&sd->input_pkt_queue, skb); 4783 input_queue_tail_incr_save(sd, qtail); 4784 rps_unlock_irq_restore(sd, &flags); 4785 return NET_RX_SUCCESS; 4786 } 4787 4788 /* Schedule NAPI for backlog device 4789 * We can use non atomic operation since we own the queue lock 4790 */ 4791 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) 4792 napi_schedule_rps(sd); 4793 goto enqueue; 4794 } 4795 reason = SKB_DROP_REASON_CPU_BACKLOG; 4796 4797 drop: 4798 sd->dropped++; 4799 rps_unlock_irq_restore(sd, &flags); 4800 4801 dev_core_stats_rx_dropped_inc(skb->dev); 4802 kfree_skb_reason(skb, reason); 4803 return NET_RX_DROP; 4804 } 4805 4806 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb) 4807 { 4808 struct net_device *dev = skb->dev; 4809 struct netdev_rx_queue *rxqueue; 4810 4811 rxqueue = dev->_rx; 4812 4813 if (skb_rx_queue_recorded(skb)) { 4814 u16 index = skb_get_rx_queue(skb); 4815 4816 if (unlikely(index >= dev->real_num_rx_queues)) { 4817 WARN_ONCE(dev->real_num_rx_queues > 1, 4818 "%s received packet on queue %u, but number " 4819 "of RX queues is %u\n", 4820 dev->name, index, dev->real_num_rx_queues); 4821 4822 return rxqueue; /* Return first rxqueue */ 4823 } 4824 rxqueue += index; 4825 } 4826 return rxqueue; 4827 } 4828 4829 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp, 4830 struct bpf_prog *xdp_prog) 4831 { 4832 void *orig_data, *orig_data_end, *hard_start; 4833 struct netdev_rx_queue *rxqueue; 4834 bool orig_bcast, orig_host; 4835 u32 mac_len, frame_sz; 4836 __be16 orig_eth_type; 4837 struct ethhdr *eth; 4838 u32 metalen, act; 4839 int off; 4840 4841 /* The XDP program wants to see the packet starting at the MAC 4842 * header. 4843 */ 4844 mac_len = skb->data - skb_mac_header(skb); 4845 hard_start = skb->data - skb_headroom(skb); 4846 4847 /* SKB "head" area always have tailroom for skb_shared_info */ 4848 frame_sz = (void *)skb_end_pointer(skb) - hard_start; 4849 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 4850 4851 rxqueue = netif_get_rxqueue(skb); 4852 xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq); 4853 xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len, 4854 skb_headlen(skb) + mac_len, true); 4855 4856 orig_data_end = xdp->data_end; 4857 orig_data = xdp->data; 4858 eth = (struct ethhdr *)xdp->data; 4859 orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr); 4860 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest); 4861 orig_eth_type = eth->h_proto; 4862 4863 act = bpf_prog_run_xdp(xdp_prog, xdp); 4864 4865 /* check if bpf_xdp_adjust_head was used */ 4866 off = xdp->data - orig_data; 4867 if (off) { 4868 if (off > 0) 4869 __skb_pull(skb, off); 4870 else if (off < 0) 4871 __skb_push(skb, -off); 4872 4873 skb->mac_header += off; 4874 skb_reset_network_header(skb); 4875 } 4876 4877 /* check if bpf_xdp_adjust_tail was used */ 4878 off = xdp->data_end - orig_data_end; 4879 if (off != 0) { 4880 skb_set_tail_pointer(skb, xdp->data_end - xdp->data); 4881 skb->len += off; /* positive on grow, negative on shrink */ 4882 } 4883 4884 /* check if XDP changed eth hdr such SKB needs update */ 4885 eth = (struct ethhdr *)xdp->data; 4886 if ((orig_eth_type != eth->h_proto) || 4887 (orig_host != ether_addr_equal_64bits(eth->h_dest, 4888 skb->dev->dev_addr)) || 4889 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) { 4890 __skb_push(skb, ETH_HLEN); 4891 skb->pkt_type = PACKET_HOST; 4892 skb->protocol = eth_type_trans(skb, skb->dev); 4893 } 4894 4895 /* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull 4896 * before calling us again on redirect path. We do not call do_redirect 4897 * as we leave that up to the caller. 4898 * 4899 * Caller is responsible for managing lifetime of skb (i.e. calling 4900 * kfree_skb in response to actions it cannot handle/XDP_DROP). 4901 */ 4902 switch (act) { 4903 case XDP_REDIRECT: 4904 case XDP_TX: 4905 __skb_push(skb, mac_len); 4906 break; 4907 case XDP_PASS: 4908 metalen = xdp->data - xdp->data_meta; 4909 if (metalen) 4910 skb_metadata_set(skb, metalen); 4911 break; 4912 } 4913 4914 return act; 4915 } 4916 4917 static u32 netif_receive_generic_xdp(struct sk_buff *skb, 4918 struct xdp_buff *xdp, 4919 struct bpf_prog *xdp_prog) 4920 { 4921 u32 act = XDP_DROP; 4922 4923 /* Reinjected packets coming from act_mirred or similar should 4924 * not get XDP generic processing. 4925 */ 4926 if (skb_is_redirected(skb)) 4927 return XDP_PASS; 4928 4929 /* XDP packets must be linear and must have sufficient headroom 4930 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also 4931 * native XDP provides, thus we need to do it here as well. 4932 */ 4933 if (skb_cloned(skb) || skb_is_nonlinear(skb) || 4934 skb_headroom(skb) < XDP_PACKET_HEADROOM) { 4935 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb); 4936 int troom = skb->tail + skb->data_len - skb->end; 4937 4938 /* In case we have to go down the path and also linearize, 4939 * then lets do the pskb_expand_head() work just once here. 4940 */ 4941 if (pskb_expand_head(skb, 4942 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0, 4943 troom > 0 ? troom + 128 : 0, GFP_ATOMIC)) 4944 goto do_drop; 4945 if (skb_linearize(skb)) 4946 goto do_drop; 4947 } 4948 4949 act = bpf_prog_run_generic_xdp(skb, xdp, xdp_prog); 4950 switch (act) { 4951 case XDP_REDIRECT: 4952 case XDP_TX: 4953 case XDP_PASS: 4954 break; 4955 default: 4956 bpf_warn_invalid_xdp_action(skb->dev, xdp_prog, act); 4957 fallthrough; 4958 case XDP_ABORTED: 4959 trace_xdp_exception(skb->dev, xdp_prog, act); 4960 fallthrough; 4961 case XDP_DROP: 4962 do_drop: 4963 kfree_skb(skb); 4964 break; 4965 } 4966 4967 return act; 4968 } 4969 4970 /* When doing generic XDP we have to bypass the qdisc layer and the 4971 * network taps in order to match in-driver-XDP behavior. This also means 4972 * that XDP packets are able to starve other packets going through a qdisc, 4973 * and DDOS attacks will be more effective. In-driver-XDP use dedicated TX 4974 * queues, so they do not have this starvation issue. 4975 */ 4976 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog) 4977 { 4978 struct net_device *dev = skb->dev; 4979 struct netdev_queue *txq; 4980 bool free_skb = true; 4981 int cpu, rc; 4982 4983 txq = netdev_core_pick_tx(dev, skb, NULL); 4984 cpu = smp_processor_id(); 4985 HARD_TX_LOCK(dev, txq, cpu); 4986 if (!netif_xmit_frozen_or_drv_stopped(txq)) { 4987 rc = netdev_start_xmit(skb, dev, txq, 0); 4988 if (dev_xmit_complete(rc)) 4989 free_skb = false; 4990 } 4991 HARD_TX_UNLOCK(dev, txq); 4992 if (free_skb) { 4993 trace_xdp_exception(dev, xdp_prog, XDP_TX); 4994 dev_core_stats_tx_dropped_inc(dev); 4995 kfree_skb(skb); 4996 } 4997 } 4998 4999 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key); 5000 5001 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb) 5002 { 5003 if (xdp_prog) { 5004 struct xdp_buff xdp; 5005 u32 act; 5006 int err; 5007 5008 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog); 5009 if (act != XDP_PASS) { 5010 switch (act) { 5011 case XDP_REDIRECT: 5012 err = xdp_do_generic_redirect(skb->dev, skb, 5013 &xdp, xdp_prog); 5014 if (err) 5015 goto out_redir; 5016 break; 5017 case XDP_TX: 5018 generic_xdp_tx(skb, xdp_prog); 5019 break; 5020 } 5021 return XDP_DROP; 5022 } 5023 } 5024 return XDP_PASS; 5025 out_redir: 5026 kfree_skb_reason(skb, SKB_DROP_REASON_XDP); 5027 return XDP_DROP; 5028 } 5029 EXPORT_SYMBOL_GPL(do_xdp_generic); 5030 5031 static int netif_rx_internal(struct sk_buff *skb) 5032 { 5033 int ret; 5034 5035 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb); 5036 5037 trace_netif_rx(skb); 5038 5039 #ifdef CONFIG_RPS 5040 if (static_branch_unlikely(&rps_needed)) { 5041 struct rps_dev_flow voidflow, *rflow = &voidflow; 5042 int cpu; 5043 5044 rcu_read_lock(); 5045 5046 cpu = get_rps_cpu(skb->dev, skb, &rflow); 5047 if (cpu < 0) 5048 cpu = smp_processor_id(); 5049 5050 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5051 5052 rcu_read_unlock(); 5053 } else 5054 #endif 5055 { 5056 unsigned int qtail; 5057 5058 ret = enqueue_to_backlog(skb, smp_processor_id(), &qtail); 5059 } 5060 return ret; 5061 } 5062 5063 /** 5064 * __netif_rx - Slightly optimized version of netif_rx 5065 * @skb: buffer to post 5066 * 5067 * This behaves as netif_rx except that it does not disable bottom halves. 5068 * As a result this function may only be invoked from the interrupt context 5069 * (either hard or soft interrupt). 5070 */ 5071 int __netif_rx(struct sk_buff *skb) 5072 { 5073 int ret; 5074 5075 lockdep_assert_once(hardirq_count() | softirq_count()); 5076 5077 trace_netif_rx_entry(skb); 5078 ret = netif_rx_internal(skb); 5079 trace_netif_rx_exit(ret); 5080 return ret; 5081 } 5082 EXPORT_SYMBOL(__netif_rx); 5083 5084 /** 5085 * netif_rx - post buffer to the network code 5086 * @skb: buffer to post 5087 * 5088 * This function receives a packet from a device driver and queues it for 5089 * the upper (protocol) levels to process via the backlog NAPI device. It 5090 * always succeeds. The buffer may be dropped during processing for 5091 * congestion control or by the protocol layers. 5092 * The network buffer is passed via the backlog NAPI device. Modern NIC 5093 * driver should use NAPI and GRO. 5094 * This function can used from interrupt and from process context. The 5095 * caller from process context must not disable interrupts before invoking 5096 * this function. 5097 * 5098 * return values: 5099 * NET_RX_SUCCESS (no congestion) 5100 * NET_RX_DROP (packet was dropped) 5101 * 5102 */ 5103 int netif_rx(struct sk_buff *skb) 5104 { 5105 bool need_bh_off = !(hardirq_count() | softirq_count()); 5106 int ret; 5107 5108 if (need_bh_off) 5109 local_bh_disable(); 5110 trace_netif_rx_entry(skb); 5111 ret = netif_rx_internal(skb); 5112 trace_netif_rx_exit(ret); 5113 if (need_bh_off) 5114 local_bh_enable(); 5115 return ret; 5116 } 5117 EXPORT_SYMBOL(netif_rx); 5118 5119 static __latent_entropy void net_tx_action(struct softirq_action *h) 5120 { 5121 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 5122 5123 if (sd->completion_queue) { 5124 struct sk_buff *clist; 5125 5126 local_irq_disable(); 5127 clist = sd->completion_queue; 5128 sd->completion_queue = NULL; 5129 local_irq_enable(); 5130 5131 while (clist) { 5132 struct sk_buff *skb = clist; 5133 5134 clist = clist->next; 5135 5136 WARN_ON(refcount_read(&skb->users)); 5137 if (likely(get_kfree_skb_cb(skb)->reason == SKB_CONSUMED)) 5138 trace_consume_skb(skb, net_tx_action); 5139 else 5140 trace_kfree_skb(skb, net_tx_action, 5141 get_kfree_skb_cb(skb)->reason); 5142 5143 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) 5144 __kfree_skb(skb); 5145 else 5146 __napi_kfree_skb(skb, 5147 get_kfree_skb_cb(skb)->reason); 5148 } 5149 } 5150 5151 if (sd->output_queue) { 5152 struct Qdisc *head; 5153 5154 local_irq_disable(); 5155 head = sd->output_queue; 5156 sd->output_queue = NULL; 5157 sd->output_queue_tailp = &sd->output_queue; 5158 local_irq_enable(); 5159 5160 rcu_read_lock(); 5161 5162 while (head) { 5163 struct Qdisc *q = head; 5164 spinlock_t *root_lock = NULL; 5165 5166 head = head->next_sched; 5167 5168 /* We need to make sure head->next_sched is read 5169 * before clearing __QDISC_STATE_SCHED 5170 */ 5171 smp_mb__before_atomic(); 5172 5173 if (!(q->flags & TCQ_F_NOLOCK)) { 5174 root_lock = qdisc_lock(q); 5175 spin_lock(root_lock); 5176 } else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, 5177 &q->state))) { 5178 /* There is a synchronize_net() between 5179 * STATE_DEACTIVATED flag being set and 5180 * qdisc_reset()/some_qdisc_is_busy() in 5181 * dev_deactivate(), so we can safely bail out 5182 * early here to avoid data race between 5183 * qdisc_deactivate() and some_qdisc_is_busy() 5184 * for lockless qdisc. 5185 */ 5186 clear_bit(__QDISC_STATE_SCHED, &q->state); 5187 continue; 5188 } 5189 5190 clear_bit(__QDISC_STATE_SCHED, &q->state); 5191 qdisc_run(q); 5192 if (root_lock) 5193 spin_unlock(root_lock); 5194 } 5195 5196 rcu_read_unlock(); 5197 } 5198 5199 xfrm_dev_backlog(sd); 5200 } 5201 5202 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) 5203 /* This hook is defined here for ATM LANE */ 5204 int (*br_fdb_test_addr_hook)(struct net_device *dev, 5205 unsigned char *addr) __read_mostly; 5206 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); 5207 #endif 5208 5209 /** 5210 * netdev_is_rx_handler_busy - check if receive handler is registered 5211 * @dev: device to check 5212 * 5213 * Check if a receive handler is already registered for a given device. 5214 * Return true if there one. 5215 * 5216 * The caller must hold the rtnl_mutex. 5217 */ 5218 bool netdev_is_rx_handler_busy(struct net_device *dev) 5219 { 5220 ASSERT_RTNL(); 5221 return dev && rtnl_dereference(dev->rx_handler); 5222 } 5223 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); 5224 5225 /** 5226 * netdev_rx_handler_register - register receive handler 5227 * @dev: device to register a handler for 5228 * @rx_handler: receive handler to register 5229 * @rx_handler_data: data pointer that is used by rx handler 5230 * 5231 * Register a receive handler for a device. This handler will then be 5232 * called from __netif_receive_skb. A negative errno code is returned 5233 * on a failure. 5234 * 5235 * The caller must hold the rtnl_mutex. 5236 * 5237 * For a general description of rx_handler, see enum rx_handler_result. 5238 */ 5239 int netdev_rx_handler_register(struct net_device *dev, 5240 rx_handler_func_t *rx_handler, 5241 void *rx_handler_data) 5242 { 5243 if (netdev_is_rx_handler_busy(dev)) 5244 return -EBUSY; 5245 5246 if (dev->priv_flags & IFF_NO_RX_HANDLER) 5247 return -EINVAL; 5248 5249 /* Note: rx_handler_data must be set before rx_handler */ 5250 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); 5251 rcu_assign_pointer(dev->rx_handler, rx_handler); 5252 5253 return 0; 5254 } 5255 EXPORT_SYMBOL_GPL(netdev_rx_handler_register); 5256 5257 /** 5258 * netdev_rx_handler_unregister - unregister receive handler 5259 * @dev: device to unregister a handler from 5260 * 5261 * Unregister a receive handler from a device. 5262 * 5263 * The caller must hold the rtnl_mutex. 5264 */ 5265 void netdev_rx_handler_unregister(struct net_device *dev) 5266 { 5267 5268 ASSERT_RTNL(); 5269 RCU_INIT_POINTER(dev->rx_handler, NULL); 5270 /* a reader seeing a non NULL rx_handler in a rcu_read_lock() 5271 * section has a guarantee to see a non NULL rx_handler_data 5272 * as well. 5273 */ 5274 synchronize_net(); 5275 RCU_INIT_POINTER(dev->rx_handler_data, NULL); 5276 } 5277 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); 5278 5279 /* 5280 * Limit the use of PFMEMALLOC reserves to those protocols that implement 5281 * the special handling of PFMEMALLOC skbs. 5282 */ 5283 static bool skb_pfmemalloc_protocol(struct sk_buff *skb) 5284 { 5285 switch (skb->protocol) { 5286 case htons(ETH_P_ARP): 5287 case htons(ETH_P_IP): 5288 case htons(ETH_P_IPV6): 5289 case htons(ETH_P_8021Q): 5290 case htons(ETH_P_8021AD): 5291 return true; 5292 default: 5293 return false; 5294 } 5295 } 5296 5297 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, 5298 int *ret, struct net_device *orig_dev) 5299 { 5300 if (nf_hook_ingress_active(skb)) { 5301 int ingress_retval; 5302 5303 if (*pt_prev) { 5304 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5305 *pt_prev = NULL; 5306 } 5307 5308 rcu_read_lock(); 5309 ingress_retval = nf_hook_ingress(skb); 5310 rcu_read_unlock(); 5311 return ingress_retval; 5312 } 5313 return 0; 5314 } 5315 5316 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc, 5317 struct packet_type **ppt_prev) 5318 { 5319 struct packet_type *ptype, *pt_prev; 5320 rx_handler_func_t *rx_handler; 5321 struct sk_buff *skb = *pskb; 5322 struct net_device *orig_dev; 5323 bool deliver_exact = false; 5324 int ret = NET_RX_DROP; 5325 __be16 type; 5326 5327 net_timestamp_check(!READ_ONCE(netdev_tstamp_prequeue), skb); 5328 5329 trace_netif_receive_skb(skb); 5330 5331 orig_dev = skb->dev; 5332 5333 skb_reset_network_header(skb); 5334 if (!skb_transport_header_was_set(skb)) 5335 skb_reset_transport_header(skb); 5336 skb_reset_mac_len(skb); 5337 5338 pt_prev = NULL; 5339 5340 another_round: 5341 skb->skb_iif = skb->dev->ifindex; 5342 5343 __this_cpu_inc(softnet_data.processed); 5344 5345 if (static_branch_unlikely(&generic_xdp_needed_key)) { 5346 int ret2; 5347 5348 migrate_disable(); 5349 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb); 5350 migrate_enable(); 5351 5352 if (ret2 != XDP_PASS) { 5353 ret = NET_RX_DROP; 5354 goto out; 5355 } 5356 } 5357 5358 if (eth_type_vlan(skb->protocol)) { 5359 skb = skb_vlan_untag(skb); 5360 if (unlikely(!skb)) 5361 goto out; 5362 } 5363 5364 if (skb_skip_tc_classify(skb)) 5365 goto skip_classify; 5366 5367 if (pfmemalloc) 5368 goto skip_taps; 5369 5370 list_for_each_entry_rcu(ptype, &ptype_all, list) { 5371 if (pt_prev) 5372 ret = deliver_skb(skb, pt_prev, orig_dev); 5373 pt_prev = ptype; 5374 } 5375 5376 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { 5377 if (pt_prev) 5378 ret = deliver_skb(skb, pt_prev, orig_dev); 5379 pt_prev = ptype; 5380 } 5381 5382 skip_taps: 5383 #ifdef CONFIG_NET_INGRESS 5384 if (static_branch_unlikely(&ingress_needed_key)) { 5385 bool another = false; 5386 5387 nf_skip_egress(skb, true); 5388 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev, 5389 &another); 5390 if (another) 5391 goto another_round; 5392 if (!skb) 5393 goto out; 5394 5395 nf_skip_egress(skb, false); 5396 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) 5397 goto out; 5398 } 5399 #endif 5400 skb_reset_redirect(skb); 5401 skip_classify: 5402 if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) 5403 goto drop; 5404 5405 if (skb_vlan_tag_present(skb)) { 5406 if (pt_prev) { 5407 ret = deliver_skb(skb, pt_prev, orig_dev); 5408 pt_prev = NULL; 5409 } 5410 if (vlan_do_receive(&skb)) 5411 goto another_round; 5412 else if (unlikely(!skb)) 5413 goto out; 5414 } 5415 5416 rx_handler = rcu_dereference(skb->dev->rx_handler); 5417 if (rx_handler) { 5418 if (pt_prev) { 5419 ret = deliver_skb(skb, pt_prev, orig_dev); 5420 pt_prev = NULL; 5421 } 5422 switch (rx_handler(&skb)) { 5423 case RX_HANDLER_CONSUMED: 5424 ret = NET_RX_SUCCESS; 5425 goto out; 5426 case RX_HANDLER_ANOTHER: 5427 goto another_round; 5428 case RX_HANDLER_EXACT: 5429 deliver_exact = true; 5430 break; 5431 case RX_HANDLER_PASS: 5432 break; 5433 default: 5434 BUG(); 5435 } 5436 } 5437 5438 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) { 5439 check_vlan_id: 5440 if (skb_vlan_tag_get_id(skb)) { 5441 /* Vlan id is non 0 and vlan_do_receive() above couldn't 5442 * find vlan device. 5443 */ 5444 skb->pkt_type = PACKET_OTHERHOST; 5445 } else if (eth_type_vlan(skb->protocol)) { 5446 /* Outer header is 802.1P with vlan 0, inner header is 5447 * 802.1Q or 802.1AD and vlan_do_receive() above could 5448 * not find vlan dev for vlan id 0. 5449 */ 5450 __vlan_hwaccel_clear_tag(skb); 5451 skb = skb_vlan_untag(skb); 5452 if (unlikely(!skb)) 5453 goto out; 5454 if (vlan_do_receive(&skb)) 5455 /* After stripping off 802.1P header with vlan 0 5456 * vlan dev is found for inner header. 5457 */ 5458 goto another_round; 5459 else if (unlikely(!skb)) 5460 goto out; 5461 else 5462 /* We have stripped outer 802.1P vlan 0 header. 5463 * But could not find vlan dev. 5464 * check again for vlan id to set OTHERHOST. 5465 */ 5466 goto check_vlan_id; 5467 } 5468 /* Note: we might in the future use prio bits 5469 * and set skb->priority like in vlan_do_receive() 5470 * For the time being, just ignore Priority Code Point 5471 */ 5472 __vlan_hwaccel_clear_tag(skb); 5473 } 5474 5475 type = skb->protocol; 5476 5477 /* deliver only exact match when indicated */ 5478 if (likely(!deliver_exact)) { 5479 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5480 &ptype_base[ntohs(type) & 5481 PTYPE_HASH_MASK]); 5482 } 5483 5484 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5485 &orig_dev->ptype_specific); 5486 5487 if (unlikely(skb->dev != orig_dev)) { 5488 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5489 &skb->dev->ptype_specific); 5490 } 5491 5492 if (pt_prev) { 5493 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 5494 goto drop; 5495 *ppt_prev = pt_prev; 5496 } else { 5497 drop: 5498 if (!deliver_exact) 5499 dev_core_stats_rx_dropped_inc(skb->dev); 5500 else 5501 dev_core_stats_rx_nohandler_inc(skb->dev); 5502 kfree_skb_reason(skb, SKB_DROP_REASON_UNHANDLED_PROTO); 5503 /* Jamal, now you will not able to escape explaining 5504 * me how you were going to use this. :-) 5505 */ 5506 ret = NET_RX_DROP; 5507 } 5508 5509 out: 5510 /* The invariant here is that if *ppt_prev is not NULL 5511 * then skb should also be non-NULL. 5512 * 5513 * Apparently *ppt_prev assignment above holds this invariant due to 5514 * skb dereferencing near it. 5515 */ 5516 *pskb = skb; 5517 return ret; 5518 } 5519 5520 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc) 5521 { 5522 struct net_device *orig_dev = skb->dev; 5523 struct packet_type *pt_prev = NULL; 5524 int ret; 5525 5526 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5527 if (pt_prev) 5528 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb, 5529 skb->dev, pt_prev, orig_dev); 5530 return ret; 5531 } 5532 5533 /** 5534 * netif_receive_skb_core - special purpose version of netif_receive_skb 5535 * @skb: buffer to process 5536 * 5537 * More direct receive version of netif_receive_skb(). It should 5538 * only be used by callers that have a need to skip RPS and Generic XDP. 5539 * Caller must also take care of handling if ``(page_is_)pfmemalloc``. 5540 * 5541 * This function may only be called from softirq context and interrupts 5542 * should be enabled. 5543 * 5544 * Return values (usually ignored): 5545 * NET_RX_SUCCESS: no congestion 5546 * NET_RX_DROP: packet was dropped 5547 */ 5548 int netif_receive_skb_core(struct sk_buff *skb) 5549 { 5550 int ret; 5551 5552 rcu_read_lock(); 5553 ret = __netif_receive_skb_one_core(skb, false); 5554 rcu_read_unlock(); 5555 5556 return ret; 5557 } 5558 EXPORT_SYMBOL(netif_receive_skb_core); 5559 5560 static inline void __netif_receive_skb_list_ptype(struct list_head *head, 5561 struct packet_type *pt_prev, 5562 struct net_device *orig_dev) 5563 { 5564 struct sk_buff *skb, *next; 5565 5566 if (!pt_prev) 5567 return; 5568 if (list_empty(head)) 5569 return; 5570 if (pt_prev->list_func != NULL) 5571 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv, 5572 ip_list_rcv, head, pt_prev, orig_dev); 5573 else 5574 list_for_each_entry_safe(skb, next, head, list) { 5575 skb_list_del_init(skb); 5576 pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 5577 } 5578 } 5579 5580 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc) 5581 { 5582 /* Fast-path assumptions: 5583 * - There is no RX handler. 5584 * - Only one packet_type matches. 5585 * If either of these fails, we will end up doing some per-packet 5586 * processing in-line, then handling the 'last ptype' for the whole 5587 * sublist. This can't cause out-of-order delivery to any single ptype, 5588 * because the 'last ptype' must be constant across the sublist, and all 5589 * other ptypes are handled per-packet. 5590 */ 5591 /* Current (common) ptype of sublist */ 5592 struct packet_type *pt_curr = NULL; 5593 /* Current (common) orig_dev of sublist */ 5594 struct net_device *od_curr = NULL; 5595 struct list_head sublist; 5596 struct sk_buff *skb, *next; 5597 5598 INIT_LIST_HEAD(&sublist); 5599 list_for_each_entry_safe(skb, next, head, list) { 5600 struct net_device *orig_dev = skb->dev; 5601 struct packet_type *pt_prev = NULL; 5602 5603 skb_list_del_init(skb); 5604 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5605 if (!pt_prev) 5606 continue; 5607 if (pt_curr != pt_prev || od_curr != orig_dev) { 5608 /* dispatch old sublist */ 5609 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5610 /* start new sublist */ 5611 INIT_LIST_HEAD(&sublist); 5612 pt_curr = pt_prev; 5613 od_curr = orig_dev; 5614 } 5615 list_add_tail(&skb->list, &sublist); 5616 } 5617 5618 /* dispatch final sublist */ 5619 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5620 } 5621 5622 static int __netif_receive_skb(struct sk_buff *skb) 5623 { 5624 int ret; 5625 5626 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { 5627 unsigned int noreclaim_flag; 5628 5629 /* 5630 * PFMEMALLOC skbs are special, they should 5631 * - be delivered to SOCK_MEMALLOC sockets only 5632 * - stay away from userspace 5633 * - have bounded memory usage 5634 * 5635 * Use PF_MEMALLOC as this saves us from propagating the allocation 5636 * context down to all allocation sites. 5637 */ 5638 noreclaim_flag = memalloc_noreclaim_save(); 5639 ret = __netif_receive_skb_one_core(skb, true); 5640 memalloc_noreclaim_restore(noreclaim_flag); 5641 } else 5642 ret = __netif_receive_skb_one_core(skb, false); 5643 5644 return ret; 5645 } 5646 5647 static void __netif_receive_skb_list(struct list_head *head) 5648 { 5649 unsigned long noreclaim_flag = 0; 5650 struct sk_buff *skb, *next; 5651 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */ 5652 5653 list_for_each_entry_safe(skb, next, head, list) { 5654 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) { 5655 struct list_head sublist; 5656 5657 /* Handle the previous sublist */ 5658 list_cut_before(&sublist, head, &skb->list); 5659 if (!list_empty(&sublist)) 5660 __netif_receive_skb_list_core(&sublist, pfmemalloc); 5661 pfmemalloc = !pfmemalloc; 5662 /* See comments in __netif_receive_skb */ 5663 if (pfmemalloc) 5664 noreclaim_flag = memalloc_noreclaim_save(); 5665 else 5666 memalloc_noreclaim_restore(noreclaim_flag); 5667 } 5668 } 5669 /* Handle the remaining sublist */ 5670 if (!list_empty(head)) 5671 __netif_receive_skb_list_core(head, pfmemalloc); 5672 /* Restore pflags */ 5673 if (pfmemalloc) 5674 memalloc_noreclaim_restore(noreclaim_flag); 5675 } 5676 5677 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp) 5678 { 5679 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog); 5680 struct bpf_prog *new = xdp->prog; 5681 int ret = 0; 5682 5683 switch (xdp->command) { 5684 case XDP_SETUP_PROG: 5685 rcu_assign_pointer(dev->xdp_prog, new); 5686 if (old) 5687 bpf_prog_put(old); 5688 5689 if (old && !new) { 5690 static_branch_dec(&generic_xdp_needed_key); 5691 } else if (new && !old) { 5692 static_branch_inc(&generic_xdp_needed_key); 5693 dev_disable_lro(dev); 5694 dev_disable_gro_hw(dev); 5695 } 5696 break; 5697 5698 default: 5699 ret = -EINVAL; 5700 break; 5701 } 5702 5703 return ret; 5704 } 5705 5706 static int netif_receive_skb_internal(struct sk_buff *skb) 5707 { 5708 int ret; 5709 5710 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb); 5711 5712 if (skb_defer_rx_timestamp(skb)) 5713 return NET_RX_SUCCESS; 5714 5715 rcu_read_lock(); 5716 #ifdef CONFIG_RPS 5717 if (static_branch_unlikely(&rps_needed)) { 5718 struct rps_dev_flow voidflow, *rflow = &voidflow; 5719 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5720 5721 if (cpu >= 0) { 5722 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5723 rcu_read_unlock(); 5724 return ret; 5725 } 5726 } 5727 #endif 5728 ret = __netif_receive_skb(skb); 5729 rcu_read_unlock(); 5730 return ret; 5731 } 5732 5733 void netif_receive_skb_list_internal(struct list_head *head) 5734 { 5735 struct sk_buff *skb, *next; 5736 struct list_head sublist; 5737 5738 INIT_LIST_HEAD(&sublist); 5739 list_for_each_entry_safe(skb, next, head, list) { 5740 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb); 5741 skb_list_del_init(skb); 5742 if (!skb_defer_rx_timestamp(skb)) 5743 list_add_tail(&skb->list, &sublist); 5744 } 5745 list_splice_init(&sublist, head); 5746 5747 rcu_read_lock(); 5748 #ifdef CONFIG_RPS 5749 if (static_branch_unlikely(&rps_needed)) { 5750 list_for_each_entry_safe(skb, next, head, list) { 5751 struct rps_dev_flow voidflow, *rflow = &voidflow; 5752 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5753 5754 if (cpu >= 0) { 5755 /* Will be handled, remove from list */ 5756 skb_list_del_init(skb); 5757 enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5758 } 5759 } 5760 } 5761 #endif 5762 __netif_receive_skb_list(head); 5763 rcu_read_unlock(); 5764 } 5765 5766 /** 5767 * netif_receive_skb - process receive buffer from network 5768 * @skb: buffer to process 5769 * 5770 * netif_receive_skb() is the main receive data processing function. 5771 * It always succeeds. The buffer may be dropped during processing 5772 * for congestion control or by the protocol layers. 5773 * 5774 * This function may only be called from softirq context and interrupts 5775 * should be enabled. 5776 * 5777 * Return values (usually ignored): 5778 * NET_RX_SUCCESS: no congestion 5779 * NET_RX_DROP: packet was dropped 5780 */ 5781 int netif_receive_skb(struct sk_buff *skb) 5782 { 5783 int ret; 5784 5785 trace_netif_receive_skb_entry(skb); 5786 5787 ret = netif_receive_skb_internal(skb); 5788 trace_netif_receive_skb_exit(ret); 5789 5790 return ret; 5791 } 5792 EXPORT_SYMBOL(netif_receive_skb); 5793 5794 /** 5795 * netif_receive_skb_list - process many receive buffers from network 5796 * @head: list of skbs to process. 5797 * 5798 * Since return value of netif_receive_skb() is normally ignored, and 5799 * wouldn't be meaningful for a list, this function returns void. 5800 * 5801 * This function may only be called from softirq context and interrupts 5802 * should be enabled. 5803 */ 5804 void netif_receive_skb_list(struct list_head *head) 5805 { 5806 struct sk_buff *skb; 5807 5808 if (list_empty(head)) 5809 return; 5810 if (trace_netif_receive_skb_list_entry_enabled()) { 5811 list_for_each_entry(skb, head, list) 5812 trace_netif_receive_skb_list_entry(skb); 5813 } 5814 netif_receive_skb_list_internal(head); 5815 trace_netif_receive_skb_list_exit(0); 5816 } 5817 EXPORT_SYMBOL(netif_receive_skb_list); 5818 5819 static DEFINE_PER_CPU(struct work_struct, flush_works); 5820 5821 /* Network device is going away, flush any packets still pending */ 5822 static void flush_backlog(struct work_struct *work) 5823 { 5824 struct sk_buff *skb, *tmp; 5825 struct softnet_data *sd; 5826 5827 local_bh_disable(); 5828 sd = this_cpu_ptr(&softnet_data); 5829 5830 rps_lock_irq_disable(sd); 5831 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { 5832 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5833 __skb_unlink(skb, &sd->input_pkt_queue); 5834 dev_kfree_skb_irq(skb); 5835 input_queue_head_incr(sd); 5836 } 5837 } 5838 rps_unlock_irq_enable(sd); 5839 5840 skb_queue_walk_safe(&sd->process_queue, skb, tmp) { 5841 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5842 __skb_unlink(skb, &sd->process_queue); 5843 kfree_skb(skb); 5844 input_queue_head_incr(sd); 5845 } 5846 } 5847 local_bh_enable(); 5848 } 5849 5850 static bool flush_required(int cpu) 5851 { 5852 #if IS_ENABLED(CONFIG_RPS) 5853 struct softnet_data *sd = &per_cpu(softnet_data, cpu); 5854 bool do_flush; 5855 5856 rps_lock_irq_disable(sd); 5857 5858 /* as insertion into process_queue happens with the rps lock held, 5859 * process_queue access may race only with dequeue 5860 */ 5861 do_flush = !skb_queue_empty(&sd->input_pkt_queue) || 5862 !skb_queue_empty_lockless(&sd->process_queue); 5863 rps_unlock_irq_enable(sd); 5864 5865 return do_flush; 5866 #endif 5867 /* without RPS we can't safely check input_pkt_queue: during a 5868 * concurrent remote skb_queue_splice() we can detect as empty both 5869 * input_pkt_queue and process_queue even if the latter could end-up 5870 * containing a lot of packets. 5871 */ 5872 return true; 5873 } 5874 5875 static void flush_all_backlogs(void) 5876 { 5877 static cpumask_t flush_cpus; 5878 unsigned int cpu; 5879 5880 /* since we are under rtnl lock protection we can use static data 5881 * for the cpumask and avoid allocating on stack the possibly 5882 * large mask 5883 */ 5884 ASSERT_RTNL(); 5885 5886 cpus_read_lock(); 5887 5888 cpumask_clear(&flush_cpus); 5889 for_each_online_cpu(cpu) { 5890 if (flush_required(cpu)) { 5891 queue_work_on(cpu, system_highpri_wq, 5892 per_cpu_ptr(&flush_works, cpu)); 5893 cpumask_set_cpu(cpu, &flush_cpus); 5894 } 5895 } 5896 5897 /* we can have in flight packet[s] on the cpus we are not flushing, 5898 * synchronize_net() in unregister_netdevice_many() will take care of 5899 * them 5900 */ 5901 for_each_cpu(cpu, &flush_cpus) 5902 flush_work(per_cpu_ptr(&flush_works, cpu)); 5903 5904 cpus_read_unlock(); 5905 } 5906 5907 static void net_rps_send_ipi(struct softnet_data *remsd) 5908 { 5909 #ifdef CONFIG_RPS 5910 while (remsd) { 5911 struct softnet_data *next = remsd->rps_ipi_next; 5912 5913 if (cpu_online(remsd->cpu)) 5914 smp_call_function_single_async(remsd->cpu, &remsd->csd); 5915 remsd = next; 5916 } 5917 #endif 5918 } 5919 5920 /* 5921 * net_rps_action_and_irq_enable sends any pending IPI's for rps. 5922 * Note: called with local irq disabled, but exits with local irq enabled. 5923 */ 5924 static void net_rps_action_and_irq_enable(struct softnet_data *sd) 5925 { 5926 #ifdef CONFIG_RPS 5927 struct softnet_data *remsd = sd->rps_ipi_list; 5928 5929 if (remsd) { 5930 sd->rps_ipi_list = NULL; 5931 5932 local_irq_enable(); 5933 5934 /* Send pending IPI's to kick RPS processing on remote cpus. */ 5935 net_rps_send_ipi(remsd); 5936 } else 5937 #endif 5938 local_irq_enable(); 5939 } 5940 5941 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) 5942 { 5943 #ifdef CONFIG_RPS 5944 return sd->rps_ipi_list != NULL; 5945 #else 5946 return false; 5947 #endif 5948 } 5949 5950 static int process_backlog(struct napi_struct *napi, int quota) 5951 { 5952 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); 5953 bool again = true; 5954 int work = 0; 5955 5956 /* Check if we have pending ipi, its better to send them now, 5957 * not waiting net_rx_action() end. 5958 */ 5959 if (sd_has_rps_ipi_waiting(sd)) { 5960 local_irq_disable(); 5961 net_rps_action_and_irq_enable(sd); 5962 } 5963 5964 napi->weight = READ_ONCE(dev_rx_weight); 5965 while (again) { 5966 struct sk_buff *skb; 5967 5968 while ((skb = __skb_dequeue(&sd->process_queue))) { 5969 rcu_read_lock(); 5970 __netif_receive_skb(skb); 5971 rcu_read_unlock(); 5972 input_queue_head_incr(sd); 5973 if (++work >= quota) 5974 return work; 5975 5976 } 5977 5978 rps_lock_irq_disable(sd); 5979 if (skb_queue_empty(&sd->input_pkt_queue)) { 5980 /* 5981 * Inline a custom version of __napi_complete(). 5982 * only current cpu owns and manipulates this napi, 5983 * and NAPI_STATE_SCHED is the only possible flag set 5984 * on backlog. 5985 * We can use a plain write instead of clear_bit(), 5986 * and we dont need an smp_mb() memory barrier. 5987 */ 5988 napi->state = 0; 5989 again = false; 5990 } else { 5991 skb_queue_splice_tail_init(&sd->input_pkt_queue, 5992 &sd->process_queue); 5993 } 5994 rps_unlock_irq_enable(sd); 5995 } 5996 5997 return work; 5998 } 5999 6000 /** 6001 * __napi_schedule - schedule for receive 6002 * @n: entry to schedule 6003 * 6004 * The entry's receive function will be scheduled to run. 6005 * Consider using __napi_schedule_irqoff() if hard irqs are masked. 6006 */ 6007 void __napi_schedule(struct napi_struct *n) 6008 { 6009 unsigned long flags; 6010 6011 local_irq_save(flags); 6012 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6013 local_irq_restore(flags); 6014 } 6015 EXPORT_SYMBOL(__napi_schedule); 6016 6017 /** 6018 * napi_schedule_prep - check if napi can be scheduled 6019 * @n: napi context 6020 * 6021 * Test if NAPI routine is already running, and if not mark 6022 * it as running. This is used as a condition variable to 6023 * insure only one NAPI poll instance runs. We also make 6024 * sure there is no pending NAPI disable. 6025 */ 6026 bool napi_schedule_prep(struct napi_struct *n) 6027 { 6028 unsigned long new, val = READ_ONCE(n->state); 6029 6030 do { 6031 if (unlikely(val & NAPIF_STATE_DISABLE)) 6032 return false; 6033 new = val | NAPIF_STATE_SCHED; 6034 6035 /* Sets STATE_MISSED bit if STATE_SCHED was already set 6036 * This was suggested by Alexander Duyck, as compiler 6037 * emits better code than : 6038 * if (val & NAPIF_STATE_SCHED) 6039 * new |= NAPIF_STATE_MISSED; 6040 */ 6041 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * 6042 NAPIF_STATE_MISSED; 6043 } while (!try_cmpxchg(&n->state, &val, new)); 6044 6045 return !(val & NAPIF_STATE_SCHED); 6046 } 6047 EXPORT_SYMBOL(napi_schedule_prep); 6048 6049 /** 6050 * __napi_schedule_irqoff - schedule for receive 6051 * @n: entry to schedule 6052 * 6053 * Variant of __napi_schedule() assuming hard irqs are masked. 6054 * 6055 * On PREEMPT_RT enabled kernels this maps to __napi_schedule() 6056 * because the interrupt disabled assumption might not be true 6057 * due to force-threaded interrupts and spinlock substitution. 6058 */ 6059 void __napi_schedule_irqoff(struct napi_struct *n) 6060 { 6061 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6062 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6063 else 6064 __napi_schedule(n); 6065 } 6066 EXPORT_SYMBOL(__napi_schedule_irqoff); 6067 6068 bool napi_complete_done(struct napi_struct *n, int work_done) 6069 { 6070 unsigned long flags, val, new, timeout = 0; 6071 bool ret = true; 6072 6073 /* 6074 * 1) Don't let napi dequeue from the cpu poll list 6075 * just in case its running on a different cpu. 6076 * 2) If we are busy polling, do nothing here, we have 6077 * the guarantee we will be called later. 6078 */ 6079 if (unlikely(n->state & (NAPIF_STATE_NPSVC | 6080 NAPIF_STATE_IN_BUSY_POLL))) 6081 return false; 6082 6083 if (work_done) { 6084 if (n->gro_bitmask) 6085 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6086 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs); 6087 } 6088 if (n->defer_hard_irqs_count > 0) { 6089 n->defer_hard_irqs_count--; 6090 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6091 if (timeout) 6092 ret = false; 6093 } 6094 if (n->gro_bitmask) { 6095 /* When the NAPI instance uses a timeout and keeps postponing 6096 * it, we need to bound somehow the time packets are kept in 6097 * the GRO layer 6098 */ 6099 napi_gro_flush(n, !!timeout); 6100 } 6101 6102 gro_normal_list(n); 6103 6104 if (unlikely(!list_empty(&n->poll_list))) { 6105 /* If n->poll_list is not empty, we need to mask irqs */ 6106 local_irq_save(flags); 6107 list_del_init(&n->poll_list); 6108 local_irq_restore(flags); 6109 } 6110 WRITE_ONCE(n->list_owner, -1); 6111 6112 val = READ_ONCE(n->state); 6113 do { 6114 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); 6115 6116 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED | 6117 NAPIF_STATE_SCHED_THREADED | 6118 NAPIF_STATE_PREFER_BUSY_POLL); 6119 6120 /* If STATE_MISSED was set, leave STATE_SCHED set, 6121 * because we will call napi->poll() one more time. 6122 * This C code was suggested by Alexander Duyck to help gcc. 6123 */ 6124 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * 6125 NAPIF_STATE_SCHED; 6126 } while (!try_cmpxchg(&n->state, &val, new)); 6127 6128 if (unlikely(val & NAPIF_STATE_MISSED)) { 6129 __napi_schedule(n); 6130 return false; 6131 } 6132 6133 if (timeout) 6134 hrtimer_start(&n->timer, ns_to_ktime(timeout), 6135 HRTIMER_MODE_REL_PINNED); 6136 return ret; 6137 } 6138 EXPORT_SYMBOL(napi_complete_done); 6139 6140 /* must be called under rcu_read_lock(), as we dont take a reference */ 6141 struct napi_struct *napi_by_id(unsigned int napi_id) 6142 { 6143 unsigned int hash = napi_id % HASH_SIZE(napi_hash); 6144 struct napi_struct *napi; 6145 6146 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) 6147 if (napi->napi_id == napi_id) 6148 return napi; 6149 6150 return NULL; 6151 } 6152 6153 #if defined(CONFIG_NET_RX_BUSY_POLL) 6154 6155 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule) 6156 { 6157 if (!skip_schedule) { 6158 gro_normal_list(napi); 6159 __napi_schedule(napi); 6160 return; 6161 } 6162 6163 if (napi->gro_bitmask) { 6164 /* flush too old packets 6165 * If HZ < 1000, flush all packets. 6166 */ 6167 napi_gro_flush(napi, HZ >= 1000); 6168 } 6169 6170 gro_normal_list(napi); 6171 clear_bit(NAPI_STATE_SCHED, &napi->state); 6172 } 6173 6174 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll, 6175 u16 budget) 6176 { 6177 bool skip_schedule = false; 6178 unsigned long timeout; 6179 int rc; 6180 6181 /* Busy polling means there is a high chance device driver hard irq 6182 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was 6183 * set in napi_schedule_prep(). 6184 * Since we are about to call napi->poll() once more, we can safely 6185 * clear NAPI_STATE_MISSED. 6186 * 6187 * Note: x86 could use a single "lock and ..." instruction 6188 * to perform these two clear_bit() 6189 */ 6190 clear_bit(NAPI_STATE_MISSED, &napi->state); 6191 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); 6192 6193 local_bh_disable(); 6194 6195 if (prefer_busy_poll) { 6196 napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs); 6197 timeout = READ_ONCE(napi->dev->gro_flush_timeout); 6198 if (napi->defer_hard_irqs_count && timeout) { 6199 hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED); 6200 skip_schedule = true; 6201 } 6202 } 6203 6204 /* All we really want here is to re-enable device interrupts. 6205 * Ideally, a new ndo_busy_poll_stop() could avoid another round. 6206 */ 6207 rc = napi->poll(napi, budget); 6208 /* We can't gro_normal_list() here, because napi->poll() might have 6209 * rearmed the napi (napi_complete_done()) in which case it could 6210 * already be running on another CPU. 6211 */ 6212 trace_napi_poll(napi, rc, budget); 6213 netpoll_poll_unlock(have_poll_lock); 6214 if (rc == budget) 6215 __busy_poll_stop(napi, skip_schedule); 6216 local_bh_enable(); 6217 } 6218 6219 void napi_busy_loop(unsigned int napi_id, 6220 bool (*loop_end)(void *, unsigned long), 6221 void *loop_end_arg, bool prefer_busy_poll, u16 budget) 6222 { 6223 unsigned long start_time = loop_end ? busy_loop_current_time() : 0; 6224 int (*napi_poll)(struct napi_struct *napi, int budget); 6225 void *have_poll_lock = NULL; 6226 struct napi_struct *napi; 6227 6228 restart: 6229 napi_poll = NULL; 6230 6231 rcu_read_lock(); 6232 6233 napi = napi_by_id(napi_id); 6234 if (!napi) 6235 goto out; 6236 6237 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6238 preempt_disable(); 6239 for (;;) { 6240 int work = 0; 6241 6242 local_bh_disable(); 6243 if (!napi_poll) { 6244 unsigned long val = READ_ONCE(napi->state); 6245 6246 /* If multiple threads are competing for this napi, 6247 * we avoid dirtying napi->state as much as we can. 6248 */ 6249 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | 6250 NAPIF_STATE_IN_BUSY_POLL)) { 6251 if (prefer_busy_poll) 6252 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6253 goto count; 6254 } 6255 if (cmpxchg(&napi->state, val, 6256 val | NAPIF_STATE_IN_BUSY_POLL | 6257 NAPIF_STATE_SCHED) != val) { 6258 if (prefer_busy_poll) 6259 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6260 goto count; 6261 } 6262 have_poll_lock = netpoll_poll_lock(napi); 6263 napi_poll = napi->poll; 6264 } 6265 work = napi_poll(napi, budget); 6266 trace_napi_poll(napi, work, budget); 6267 gro_normal_list(napi); 6268 count: 6269 if (work > 0) 6270 __NET_ADD_STATS(dev_net(napi->dev), 6271 LINUX_MIB_BUSYPOLLRXPACKETS, work); 6272 local_bh_enable(); 6273 6274 if (!loop_end || loop_end(loop_end_arg, start_time)) 6275 break; 6276 6277 if (unlikely(need_resched())) { 6278 if (napi_poll) 6279 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6280 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6281 preempt_enable(); 6282 rcu_read_unlock(); 6283 cond_resched(); 6284 if (loop_end(loop_end_arg, start_time)) 6285 return; 6286 goto restart; 6287 } 6288 cpu_relax(); 6289 } 6290 if (napi_poll) 6291 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6292 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6293 preempt_enable(); 6294 out: 6295 rcu_read_unlock(); 6296 } 6297 EXPORT_SYMBOL(napi_busy_loop); 6298 6299 #endif /* CONFIG_NET_RX_BUSY_POLL */ 6300 6301 static void napi_hash_add(struct napi_struct *napi) 6302 { 6303 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state)) 6304 return; 6305 6306 spin_lock(&napi_hash_lock); 6307 6308 /* 0..NR_CPUS range is reserved for sender_cpu use */ 6309 do { 6310 if (unlikely(++napi_gen_id < MIN_NAPI_ID)) 6311 napi_gen_id = MIN_NAPI_ID; 6312 } while (napi_by_id(napi_gen_id)); 6313 napi->napi_id = napi_gen_id; 6314 6315 hlist_add_head_rcu(&napi->napi_hash_node, 6316 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); 6317 6318 spin_unlock(&napi_hash_lock); 6319 } 6320 6321 /* Warning : caller is responsible to make sure rcu grace period 6322 * is respected before freeing memory containing @napi 6323 */ 6324 static void napi_hash_del(struct napi_struct *napi) 6325 { 6326 spin_lock(&napi_hash_lock); 6327 6328 hlist_del_init_rcu(&napi->napi_hash_node); 6329 6330 spin_unlock(&napi_hash_lock); 6331 } 6332 6333 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) 6334 { 6335 struct napi_struct *napi; 6336 6337 napi = container_of(timer, struct napi_struct, timer); 6338 6339 /* Note : we use a relaxed variant of napi_schedule_prep() not setting 6340 * NAPI_STATE_MISSED, since we do not react to a device IRQ. 6341 */ 6342 if (!napi_disable_pending(napi) && 6343 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) { 6344 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6345 __napi_schedule_irqoff(napi); 6346 } 6347 6348 return HRTIMER_NORESTART; 6349 } 6350 6351 static void init_gro_hash(struct napi_struct *napi) 6352 { 6353 int i; 6354 6355 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6356 INIT_LIST_HEAD(&napi->gro_hash[i].list); 6357 napi->gro_hash[i].count = 0; 6358 } 6359 napi->gro_bitmask = 0; 6360 } 6361 6362 int dev_set_threaded(struct net_device *dev, bool threaded) 6363 { 6364 struct napi_struct *napi; 6365 int err = 0; 6366 6367 if (dev->threaded == threaded) 6368 return 0; 6369 6370 if (threaded) { 6371 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6372 if (!napi->thread) { 6373 err = napi_kthread_create(napi); 6374 if (err) { 6375 threaded = false; 6376 break; 6377 } 6378 } 6379 } 6380 } 6381 6382 dev->threaded = threaded; 6383 6384 /* Make sure kthread is created before THREADED bit 6385 * is set. 6386 */ 6387 smp_mb__before_atomic(); 6388 6389 /* Setting/unsetting threaded mode on a napi might not immediately 6390 * take effect, if the current napi instance is actively being 6391 * polled. In this case, the switch between threaded mode and 6392 * softirq mode will happen in the next round of napi_schedule(). 6393 * This should not cause hiccups/stalls to the live traffic. 6394 */ 6395 list_for_each_entry(napi, &dev->napi_list, dev_list) 6396 assign_bit(NAPI_STATE_THREADED, &napi->state, threaded); 6397 6398 return err; 6399 } 6400 EXPORT_SYMBOL(dev_set_threaded); 6401 6402 /** 6403 * netif_queue_set_napi - Associate queue with the napi 6404 * @dev: device to which NAPI and queue belong 6405 * @queue_index: Index of queue 6406 * @type: queue type as RX or TX 6407 * @napi: NAPI context, pass NULL to clear previously set NAPI 6408 * 6409 * Set queue with its corresponding napi context. This should be done after 6410 * registering the NAPI handler for the queue-vector and the queues have been 6411 * mapped to the corresponding interrupt vector. 6412 */ 6413 void netif_queue_set_napi(struct net_device *dev, unsigned int queue_index, 6414 enum netdev_queue_type type, struct napi_struct *napi) 6415 { 6416 struct netdev_rx_queue *rxq; 6417 struct netdev_queue *txq; 6418 6419 if (WARN_ON_ONCE(napi && !napi->dev)) 6420 return; 6421 if (dev->reg_state >= NETREG_REGISTERED) 6422 ASSERT_RTNL(); 6423 6424 switch (type) { 6425 case NETDEV_QUEUE_TYPE_RX: 6426 rxq = __netif_get_rx_queue(dev, queue_index); 6427 rxq->napi = napi; 6428 return; 6429 case NETDEV_QUEUE_TYPE_TX: 6430 txq = netdev_get_tx_queue(dev, queue_index); 6431 txq->napi = napi; 6432 return; 6433 default: 6434 return; 6435 } 6436 } 6437 EXPORT_SYMBOL(netif_queue_set_napi); 6438 6439 void netif_napi_add_weight(struct net_device *dev, struct napi_struct *napi, 6440 int (*poll)(struct napi_struct *, int), int weight) 6441 { 6442 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state))) 6443 return; 6444 6445 INIT_LIST_HEAD(&napi->poll_list); 6446 INIT_HLIST_NODE(&napi->napi_hash_node); 6447 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); 6448 napi->timer.function = napi_watchdog; 6449 init_gro_hash(napi); 6450 napi->skb = NULL; 6451 INIT_LIST_HEAD(&napi->rx_list); 6452 napi->rx_count = 0; 6453 napi->poll = poll; 6454 if (weight > NAPI_POLL_WEIGHT) 6455 netdev_err_once(dev, "%s() called with weight %d\n", __func__, 6456 weight); 6457 napi->weight = weight; 6458 napi->dev = dev; 6459 #ifdef CONFIG_NETPOLL 6460 napi->poll_owner = -1; 6461 #endif 6462 napi->list_owner = -1; 6463 set_bit(NAPI_STATE_SCHED, &napi->state); 6464 set_bit(NAPI_STATE_NPSVC, &napi->state); 6465 list_add_rcu(&napi->dev_list, &dev->napi_list); 6466 napi_hash_add(napi); 6467 napi_get_frags_check(napi); 6468 /* Create kthread for this napi if dev->threaded is set. 6469 * Clear dev->threaded if kthread creation failed so that 6470 * threaded mode will not be enabled in napi_enable(). 6471 */ 6472 if (dev->threaded && napi_kthread_create(napi)) 6473 dev->threaded = 0; 6474 netif_napi_set_irq(napi, -1); 6475 } 6476 EXPORT_SYMBOL(netif_napi_add_weight); 6477 6478 void napi_disable(struct napi_struct *n) 6479 { 6480 unsigned long val, new; 6481 6482 might_sleep(); 6483 set_bit(NAPI_STATE_DISABLE, &n->state); 6484 6485 val = READ_ONCE(n->state); 6486 do { 6487 while (val & (NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC)) { 6488 usleep_range(20, 200); 6489 val = READ_ONCE(n->state); 6490 } 6491 6492 new = val | NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC; 6493 new &= ~(NAPIF_STATE_THREADED | NAPIF_STATE_PREFER_BUSY_POLL); 6494 } while (!try_cmpxchg(&n->state, &val, new)); 6495 6496 hrtimer_cancel(&n->timer); 6497 6498 clear_bit(NAPI_STATE_DISABLE, &n->state); 6499 } 6500 EXPORT_SYMBOL(napi_disable); 6501 6502 /** 6503 * napi_enable - enable NAPI scheduling 6504 * @n: NAPI context 6505 * 6506 * Resume NAPI from being scheduled on this context. 6507 * Must be paired with napi_disable. 6508 */ 6509 void napi_enable(struct napi_struct *n) 6510 { 6511 unsigned long new, val = READ_ONCE(n->state); 6512 6513 do { 6514 BUG_ON(!test_bit(NAPI_STATE_SCHED, &val)); 6515 6516 new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC); 6517 if (n->dev->threaded && n->thread) 6518 new |= NAPIF_STATE_THREADED; 6519 } while (!try_cmpxchg(&n->state, &val, new)); 6520 } 6521 EXPORT_SYMBOL(napi_enable); 6522 6523 static void flush_gro_hash(struct napi_struct *napi) 6524 { 6525 int i; 6526 6527 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6528 struct sk_buff *skb, *n; 6529 6530 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list) 6531 kfree_skb(skb); 6532 napi->gro_hash[i].count = 0; 6533 } 6534 } 6535 6536 /* Must be called in process context */ 6537 void __netif_napi_del(struct napi_struct *napi) 6538 { 6539 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state)) 6540 return; 6541 6542 napi_hash_del(napi); 6543 list_del_rcu(&napi->dev_list); 6544 napi_free_frags(napi); 6545 6546 flush_gro_hash(napi); 6547 napi->gro_bitmask = 0; 6548 6549 if (napi->thread) { 6550 kthread_stop(napi->thread); 6551 napi->thread = NULL; 6552 } 6553 } 6554 EXPORT_SYMBOL(__netif_napi_del); 6555 6556 static int __napi_poll(struct napi_struct *n, bool *repoll) 6557 { 6558 int work, weight; 6559 6560 weight = n->weight; 6561 6562 /* This NAPI_STATE_SCHED test is for avoiding a race 6563 * with netpoll's poll_napi(). Only the entity which 6564 * obtains the lock and sees NAPI_STATE_SCHED set will 6565 * actually make the ->poll() call. Therefore we avoid 6566 * accidentally calling ->poll() when NAPI is not scheduled. 6567 */ 6568 work = 0; 6569 if (napi_is_scheduled(n)) { 6570 work = n->poll(n, weight); 6571 trace_napi_poll(n, work, weight); 6572 6573 xdp_do_check_flushed(n); 6574 } 6575 6576 if (unlikely(work > weight)) 6577 netdev_err_once(n->dev, "NAPI poll function %pS returned %d, exceeding its budget of %d.\n", 6578 n->poll, work, weight); 6579 6580 if (likely(work < weight)) 6581 return work; 6582 6583 /* Drivers must not modify the NAPI state if they 6584 * consume the entire weight. In such cases this code 6585 * still "owns" the NAPI instance and therefore can 6586 * move the instance around on the list at-will. 6587 */ 6588 if (unlikely(napi_disable_pending(n))) { 6589 napi_complete(n); 6590 return work; 6591 } 6592 6593 /* The NAPI context has more processing work, but busy-polling 6594 * is preferred. Exit early. 6595 */ 6596 if (napi_prefer_busy_poll(n)) { 6597 if (napi_complete_done(n, work)) { 6598 /* If timeout is not set, we need to make sure 6599 * that the NAPI is re-scheduled. 6600 */ 6601 napi_schedule(n); 6602 } 6603 return work; 6604 } 6605 6606 if (n->gro_bitmask) { 6607 /* flush too old packets 6608 * If HZ < 1000, flush all packets. 6609 */ 6610 napi_gro_flush(n, HZ >= 1000); 6611 } 6612 6613 gro_normal_list(n); 6614 6615 /* Some drivers may have called napi_schedule 6616 * prior to exhausting their budget. 6617 */ 6618 if (unlikely(!list_empty(&n->poll_list))) { 6619 pr_warn_once("%s: Budget exhausted after napi rescheduled\n", 6620 n->dev ? n->dev->name : "backlog"); 6621 return work; 6622 } 6623 6624 *repoll = true; 6625 6626 return work; 6627 } 6628 6629 static int napi_poll(struct napi_struct *n, struct list_head *repoll) 6630 { 6631 bool do_repoll = false; 6632 void *have; 6633 int work; 6634 6635 list_del_init(&n->poll_list); 6636 6637 have = netpoll_poll_lock(n); 6638 6639 work = __napi_poll(n, &do_repoll); 6640 6641 if (do_repoll) 6642 list_add_tail(&n->poll_list, repoll); 6643 6644 netpoll_poll_unlock(have); 6645 6646 return work; 6647 } 6648 6649 static int napi_thread_wait(struct napi_struct *napi) 6650 { 6651 bool woken = false; 6652 6653 set_current_state(TASK_INTERRUPTIBLE); 6654 6655 while (!kthread_should_stop()) { 6656 /* Testing SCHED_THREADED bit here to make sure the current 6657 * kthread owns this napi and could poll on this napi. 6658 * Testing SCHED bit is not enough because SCHED bit might be 6659 * set by some other busy poll thread or by napi_disable(). 6660 */ 6661 if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state) || woken) { 6662 WARN_ON(!list_empty(&napi->poll_list)); 6663 __set_current_state(TASK_RUNNING); 6664 return 0; 6665 } 6666 6667 schedule(); 6668 /* woken being true indicates this thread owns this napi. */ 6669 woken = true; 6670 set_current_state(TASK_INTERRUPTIBLE); 6671 } 6672 __set_current_state(TASK_RUNNING); 6673 6674 return -1; 6675 } 6676 6677 static void skb_defer_free_flush(struct softnet_data *sd) 6678 { 6679 struct sk_buff *skb, *next; 6680 6681 /* Paired with WRITE_ONCE() in skb_attempt_defer_free() */ 6682 if (!READ_ONCE(sd->defer_list)) 6683 return; 6684 6685 spin_lock(&sd->defer_lock); 6686 skb = sd->defer_list; 6687 sd->defer_list = NULL; 6688 sd->defer_count = 0; 6689 spin_unlock(&sd->defer_lock); 6690 6691 while (skb != NULL) { 6692 next = skb->next; 6693 napi_consume_skb(skb, 1); 6694 skb = next; 6695 } 6696 } 6697 6698 static int napi_threaded_poll(void *data) 6699 { 6700 struct napi_struct *napi = data; 6701 struct softnet_data *sd; 6702 void *have; 6703 6704 while (!napi_thread_wait(napi)) { 6705 for (;;) { 6706 bool repoll = false; 6707 6708 local_bh_disable(); 6709 sd = this_cpu_ptr(&softnet_data); 6710 sd->in_napi_threaded_poll = true; 6711 6712 have = netpoll_poll_lock(napi); 6713 __napi_poll(napi, &repoll); 6714 netpoll_poll_unlock(have); 6715 6716 sd->in_napi_threaded_poll = false; 6717 barrier(); 6718 6719 if (sd_has_rps_ipi_waiting(sd)) { 6720 local_irq_disable(); 6721 net_rps_action_and_irq_enable(sd); 6722 } 6723 skb_defer_free_flush(sd); 6724 local_bh_enable(); 6725 6726 if (!repoll) 6727 break; 6728 6729 cond_resched(); 6730 } 6731 } 6732 return 0; 6733 } 6734 6735 static __latent_entropy void net_rx_action(struct softirq_action *h) 6736 { 6737 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 6738 unsigned long time_limit = jiffies + 6739 usecs_to_jiffies(READ_ONCE(netdev_budget_usecs)); 6740 int budget = READ_ONCE(netdev_budget); 6741 LIST_HEAD(list); 6742 LIST_HEAD(repoll); 6743 6744 start: 6745 sd->in_net_rx_action = true; 6746 local_irq_disable(); 6747 list_splice_init(&sd->poll_list, &list); 6748 local_irq_enable(); 6749 6750 for (;;) { 6751 struct napi_struct *n; 6752 6753 skb_defer_free_flush(sd); 6754 6755 if (list_empty(&list)) { 6756 if (list_empty(&repoll)) { 6757 sd->in_net_rx_action = false; 6758 barrier(); 6759 /* We need to check if ____napi_schedule() 6760 * had refilled poll_list while 6761 * sd->in_net_rx_action was true. 6762 */ 6763 if (!list_empty(&sd->poll_list)) 6764 goto start; 6765 if (!sd_has_rps_ipi_waiting(sd)) 6766 goto end; 6767 } 6768 break; 6769 } 6770 6771 n = list_first_entry(&list, struct napi_struct, poll_list); 6772 budget -= napi_poll(n, &repoll); 6773 6774 /* If softirq window is exhausted then punt. 6775 * Allow this to run for 2 jiffies since which will allow 6776 * an average latency of 1.5/HZ. 6777 */ 6778 if (unlikely(budget <= 0 || 6779 time_after_eq(jiffies, time_limit))) { 6780 sd->time_squeeze++; 6781 break; 6782 } 6783 } 6784 6785 local_irq_disable(); 6786 6787 list_splice_tail_init(&sd->poll_list, &list); 6788 list_splice_tail(&repoll, &list); 6789 list_splice(&list, &sd->poll_list); 6790 if (!list_empty(&sd->poll_list)) 6791 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 6792 else 6793 sd->in_net_rx_action = false; 6794 6795 net_rps_action_and_irq_enable(sd); 6796 end:; 6797 } 6798 6799 struct netdev_adjacent { 6800 struct net_device *dev; 6801 netdevice_tracker dev_tracker; 6802 6803 /* upper master flag, there can only be one master device per list */ 6804 bool master; 6805 6806 /* lookup ignore flag */ 6807 bool ignore; 6808 6809 /* counter for the number of times this device was added to us */ 6810 u16 ref_nr; 6811 6812 /* private field for the users */ 6813 void *private; 6814 6815 struct list_head list; 6816 struct rcu_head rcu; 6817 }; 6818 6819 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, 6820 struct list_head *adj_list) 6821 { 6822 struct netdev_adjacent *adj; 6823 6824 list_for_each_entry(adj, adj_list, list) { 6825 if (adj->dev == adj_dev) 6826 return adj; 6827 } 6828 return NULL; 6829 } 6830 6831 static int ____netdev_has_upper_dev(struct net_device *upper_dev, 6832 struct netdev_nested_priv *priv) 6833 { 6834 struct net_device *dev = (struct net_device *)priv->data; 6835 6836 return upper_dev == dev; 6837 } 6838 6839 /** 6840 * netdev_has_upper_dev - Check if device is linked to an upper device 6841 * @dev: device 6842 * @upper_dev: upper device to check 6843 * 6844 * Find out if a device is linked to specified upper device and return true 6845 * in case it is. Note that this checks only immediate upper device, 6846 * not through a complete stack of devices. The caller must hold the RTNL lock. 6847 */ 6848 bool netdev_has_upper_dev(struct net_device *dev, 6849 struct net_device *upper_dev) 6850 { 6851 struct netdev_nested_priv priv = { 6852 .data = (void *)upper_dev, 6853 }; 6854 6855 ASSERT_RTNL(); 6856 6857 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 6858 &priv); 6859 } 6860 EXPORT_SYMBOL(netdev_has_upper_dev); 6861 6862 /** 6863 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device 6864 * @dev: device 6865 * @upper_dev: upper device to check 6866 * 6867 * Find out if a device is linked to specified upper device and return true 6868 * in case it is. Note that this checks the entire upper device chain. 6869 * The caller must hold rcu lock. 6870 */ 6871 6872 bool netdev_has_upper_dev_all_rcu(struct net_device *dev, 6873 struct net_device *upper_dev) 6874 { 6875 struct netdev_nested_priv priv = { 6876 .data = (void *)upper_dev, 6877 }; 6878 6879 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 6880 &priv); 6881 } 6882 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); 6883 6884 /** 6885 * netdev_has_any_upper_dev - Check if device is linked to some device 6886 * @dev: device 6887 * 6888 * Find out if a device is linked to an upper device and return true in case 6889 * it is. The caller must hold the RTNL lock. 6890 */ 6891 bool netdev_has_any_upper_dev(struct net_device *dev) 6892 { 6893 ASSERT_RTNL(); 6894 6895 return !list_empty(&dev->adj_list.upper); 6896 } 6897 EXPORT_SYMBOL(netdev_has_any_upper_dev); 6898 6899 /** 6900 * netdev_master_upper_dev_get - Get master upper device 6901 * @dev: device 6902 * 6903 * Find a master upper device and return pointer to it or NULL in case 6904 * it's not there. The caller must hold the RTNL lock. 6905 */ 6906 struct net_device *netdev_master_upper_dev_get(struct net_device *dev) 6907 { 6908 struct netdev_adjacent *upper; 6909 6910 ASSERT_RTNL(); 6911 6912 if (list_empty(&dev->adj_list.upper)) 6913 return NULL; 6914 6915 upper = list_first_entry(&dev->adj_list.upper, 6916 struct netdev_adjacent, list); 6917 if (likely(upper->master)) 6918 return upper->dev; 6919 return NULL; 6920 } 6921 EXPORT_SYMBOL(netdev_master_upper_dev_get); 6922 6923 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev) 6924 { 6925 struct netdev_adjacent *upper; 6926 6927 ASSERT_RTNL(); 6928 6929 if (list_empty(&dev->adj_list.upper)) 6930 return NULL; 6931 6932 upper = list_first_entry(&dev->adj_list.upper, 6933 struct netdev_adjacent, list); 6934 if (likely(upper->master) && !upper->ignore) 6935 return upper->dev; 6936 return NULL; 6937 } 6938 6939 /** 6940 * netdev_has_any_lower_dev - Check if device is linked to some device 6941 * @dev: device 6942 * 6943 * Find out if a device is linked to a lower device and return true in case 6944 * it is. The caller must hold the RTNL lock. 6945 */ 6946 static bool netdev_has_any_lower_dev(struct net_device *dev) 6947 { 6948 ASSERT_RTNL(); 6949 6950 return !list_empty(&dev->adj_list.lower); 6951 } 6952 6953 void *netdev_adjacent_get_private(struct list_head *adj_list) 6954 { 6955 struct netdev_adjacent *adj; 6956 6957 adj = list_entry(adj_list, struct netdev_adjacent, list); 6958 6959 return adj->private; 6960 } 6961 EXPORT_SYMBOL(netdev_adjacent_get_private); 6962 6963 /** 6964 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list 6965 * @dev: device 6966 * @iter: list_head ** of the current position 6967 * 6968 * Gets the next device from the dev's upper list, starting from iter 6969 * position. The caller must hold RCU read lock. 6970 */ 6971 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, 6972 struct list_head **iter) 6973 { 6974 struct netdev_adjacent *upper; 6975 6976 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 6977 6978 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 6979 6980 if (&upper->list == &dev->adj_list.upper) 6981 return NULL; 6982 6983 *iter = &upper->list; 6984 6985 return upper->dev; 6986 } 6987 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); 6988 6989 static struct net_device *__netdev_next_upper_dev(struct net_device *dev, 6990 struct list_head **iter, 6991 bool *ignore) 6992 { 6993 struct netdev_adjacent *upper; 6994 6995 upper = list_entry((*iter)->next, struct netdev_adjacent, list); 6996 6997 if (&upper->list == &dev->adj_list.upper) 6998 return NULL; 6999 7000 *iter = &upper->list; 7001 *ignore = upper->ignore; 7002 7003 return upper->dev; 7004 } 7005 7006 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, 7007 struct list_head **iter) 7008 { 7009 struct netdev_adjacent *upper; 7010 7011 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 7012 7013 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7014 7015 if (&upper->list == &dev->adj_list.upper) 7016 return NULL; 7017 7018 *iter = &upper->list; 7019 7020 return upper->dev; 7021 } 7022 7023 static int __netdev_walk_all_upper_dev(struct net_device *dev, 7024 int (*fn)(struct net_device *dev, 7025 struct netdev_nested_priv *priv), 7026 struct netdev_nested_priv *priv) 7027 { 7028 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7029 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7030 int ret, cur = 0; 7031 bool ignore; 7032 7033 now = dev; 7034 iter = &dev->adj_list.upper; 7035 7036 while (1) { 7037 if (now != dev) { 7038 ret = fn(now, priv); 7039 if (ret) 7040 return ret; 7041 } 7042 7043 next = NULL; 7044 while (1) { 7045 udev = __netdev_next_upper_dev(now, &iter, &ignore); 7046 if (!udev) 7047 break; 7048 if (ignore) 7049 continue; 7050 7051 next = udev; 7052 niter = &udev->adj_list.upper; 7053 dev_stack[cur] = now; 7054 iter_stack[cur++] = iter; 7055 break; 7056 } 7057 7058 if (!next) { 7059 if (!cur) 7060 return 0; 7061 next = dev_stack[--cur]; 7062 niter = iter_stack[cur]; 7063 } 7064 7065 now = next; 7066 iter = niter; 7067 } 7068 7069 return 0; 7070 } 7071 7072 int netdev_walk_all_upper_dev_rcu(struct net_device *dev, 7073 int (*fn)(struct net_device *dev, 7074 struct netdev_nested_priv *priv), 7075 struct netdev_nested_priv *priv) 7076 { 7077 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7078 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7079 int ret, cur = 0; 7080 7081 now = dev; 7082 iter = &dev->adj_list.upper; 7083 7084 while (1) { 7085 if (now != dev) { 7086 ret = fn(now, priv); 7087 if (ret) 7088 return ret; 7089 } 7090 7091 next = NULL; 7092 while (1) { 7093 udev = netdev_next_upper_dev_rcu(now, &iter); 7094 if (!udev) 7095 break; 7096 7097 next = udev; 7098 niter = &udev->adj_list.upper; 7099 dev_stack[cur] = now; 7100 iter_stack[cur++] = iter; 7101 break; 7102 } 7103 7104 if (!next) { 7105 if (!cur) 7106 return 0; 7107 next = dev_stack[--cur]; 7108 niter = iter_stack[cur]; 7109 } 7110 7111 now = next; 7112 iter = niter; 7113 } 7114 7115 return 0; 7116 } 7117 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); 7118 7119 static bool __netdev_has_upper_dev(struct net_device *dev, 7120 struct net_device *upper_dev) 7121 { 7122 struct netdev_nested_priv priv = { 7123 .flags = 0, 7124 .data = (void *)upper_dev, 7125 }; 7126 7127 ASSERT_RTNL(); 7128 7129 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev, 7130 &priv); 7131 } 7132 7133 /** 7134 * netdev_lower_get_next_private - Get the next ->private from the 7135 * lower neighbour list 7136 * @dev: device 7137 * @iter: list_head ** of the current position 7138 * 7139 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7140 * list, starting from iter position. The caller must hold either hold the 7141 * RTNL lock or its own locking that guarantees that the neighbour lower 7142 * list will remain unchanged. 7143 */ 7144 void *netdev_lower_get_next_private(struct net_device *dev, 7145 struct list_head **iter) 7146 { 7147 struct netdev_adjacent *lower; 7148 7149 lower = list_entry(*iter, struct netdev_adjacent, list); 7150 7151 if (&lower->list == &dev->adj_list.lower) 7152 return NULL; 7153 7154 *iter = lower->list.next; 7155 7156 return lower->private; 7157 } 7158 EXPORT_SYMBOL(netdev_lower_get_next_private); 7159 7160 /** 7161 * netdev_lower_get_next_private_rcu - Get the next ->private from the 7162 * lower neighbour list, RCU 7163 * variant 7164 * @dev: device 7165 * @iter: list_head ** of the current position 7166 * 7167 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7168 * list, starting from iter position. The caller must hold RCU read lock. 7169 */ 7170 void *netdev_lower_get_next_private_rcu(struct net_device *dev, 7171 struct list_head **iter) 7172 { 7173 struct netdev_adjacent *lower; 7174 7175 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 7176 7177 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7178 7179 if (&lower->list == &dev->adj_list.lower) 7180 return NULL; 7181 7182 *iter = &lower->list; 7183 7184 return lower->private; 7185 } 7186 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); 7187 7188 /** 7189 * netdev_lower_get_next - Get the next device from the lower neighbour 7190 * list 7191 * @dev: device 7192 * @iter: list_head ** of the current position 7193 * 7194 * Gets the next netdev_adjacent from the dev's lower neighbour 7195 * list, starting from iter position. The caller must hold RTNL lock or 7196 * its own locking that guarantees that the neighbour lower 7197 * list will remain unchanged. 7198 */ 7199 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) 7200 { 7201 struct netdev_adjacent *lower; 7202 7203 lower = list_entry(*iter, struct netdev_adjacent, list); 7204 7205 if (&lower->list == &dev->adj_list.lower) 7206 return NULL; 7207 7208 *iter = lower->list.next; 7209 7210 return lower->dev; 7211 } 7212 EXPORT_SYMBOL(netdev_lower_get_next); 7213 7214 static struct net_device *netdev_next_lower_dev(struct net_device *dev, 7215 struct list_head **iter) 7216 { 7217 struct netdev_adjacent *lower; 7218 7219 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7220 7221 if (&lower->list == &dev->adj_list.lower) 7222 return NULL; 7223 7224 *iter = &lower->list; 7225 7226 return lower->dev; 7227 } 7228 7229 static struct net_device *__netdev_next_lower_dev(struct net_device *dev, 7230 struct list_head **iter, 7231 bool *ignore) 7232 { 7233 struct netdev_adjacent *lower; 7234 7235 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7236 7237 if (&lower->list == &dev->adj_list.lower) 7238 return NULL; 7239 7240 *iter = &lower->list; 7241 *ignore = lower->ignore; 7242 7243 return lower->dev; 7244 } 7245 7246 int netdev_walk_all_lower_dev(struct net_device *dev, 7247 int (*fn)(struct net_device *dev, 7248 struct netdev_nested_priv *priv), 7249 struct netdev_nested_priv *priv) 7250 { 7251 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7252 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7253 int ret, cur = 0; 7254 7255 now = dev; 7256 iter = &dev->adj_list.lower; 7257 7258 while (1) { 7259 if (now != dev) { 7260 ret = fn(now, priv); 7261 if (ret) 7262 return ret; 7263 } 7264 7265 next = NULL; 7266 while (1) { 7267 ldev = netdev_next_lower_dev(now, &iter); 7268 if (!ldev) 7269 break; 7270 7271 next = ldev; 7272 niter = &ldev->adj_list.lower; 7273 dev_stack[cur] = now; 7274 iter_stack[cur++] = iter; 7275 break; 7276 } 7277 7278 if (!next) { 7279 if (!cur) 7280 return 0; 7281 next = dev_stack[--cur]; 7282 niter = iter_stack[cur]; 7283 } 7284 7285 now = next; 7286 iter = niter; 7287 } 7288 7289 return 0; 7290 } 7291 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); 7292 7293 static int __netdev_walk_all_lower_dev(struct net_device *dev, 7294 int (*fn)(struct net_device *dev, 7295 struct netdev_nested_priv *priv), 7296 struct netdev_nested_priv *priv) 7297 { 7298 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7299 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7300 int ret, cur = 0; 7301 bool ignore; 7302 7303 now = dev; 7304 iter = &dev->adj_list.lower; 7305 7306 while (1) { 7307 if (now != dev) { 7308 ret = fn(now, priv); 7309 if (ret) 7310 return ret; 7311 } 7312 7313 next = NULL; 7314 while (1) { 7315 ldev = __netdev_next_lower_dev(now, &iter, &ignore); 7316 if (!ldev) 7317 break; 7318 if (ignore) 7319 continue; 7320 7321 next = ldev; 7322 niter = &ldev->adj_list.lower; 7323 dev_stack[cur] = now; 7324 iter_stack[cur++] = iter; 7325 break; 7326 } 7327 7328 if (!next) { 7329 if (!cur) 7330 return 0; 7331 next = dev_stack[--cur]; 7332 niter = iter_stack[cur]; 7333 } 7334 7335 now = next; 7336 iter = niter; 7337 } 7338 7339 return 0; 7340 } 7341 7342 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, 7343 struct list_head **iter) 7344 { 7345 struct netdev_adjacent *lower; 7346 7347 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7348 if (&lower->list == &dev->adj_list.lower) 7349 return NULL; 7350 7351 *iter = &lower->list; 7352 7353 return lower->dev; 7354 } 7355 EXPORT_SYMBOL(netdev_next_lower_dev_rcu); 7356 7357 static u8 __netdev_upper_depth(struct net_device *dev) 7358 { 7359 struct net_device *udev; 7360 struct list_head *iter; 7361 u8 max_depth = 0; 7362 bool ignore; 7363 7364 for (iter = &dev->adj_list.upper, 7365 udev = __netdev_next_upper_dev(dev, &iter, &ignore); 7366 udev; 7367 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) { 7368 if (ignore) 7369 continue; 7370 if (max_depth < udev->upper_level) 7371 max_depth = udev->upper_level; 7372 } 7373 7374 return max_depth; 7375 } 7376 7377 static u8 __netdev_lower_depth(struct net_device *dev) 7378 { 7379 struct net_device *ldev; 7380 struct list_head *iter; 7381 u8 max_depth = 0; 7382 bool ignore; 7383 7384 for (iter = &dev->adj_list.lower, 7385 ldev = __netdev_next_lower_dev(dev, &iter, &ignore); 7386 ldev; 7387 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) { 7388 if (ignore) 7389 continue; 7390 if (max_depth < ldev->lower_level) 7391 max_depth = ldev->lower_level; 7392 } 7393 7394 return max_depth; 7395 } 7396 7397 static int __netdev_update_upper_level(struct net_device *dev, 7398 struct netdev_nested_priv *__unused) 7399 { 7400 dev->upper_level = __netdev_upper_depth(dev) + 1; 7401 return 0; 7402 } 7403 7404 #ifdef CONFIG_LOCKDEP 7405 static LIST_HEAD(net_unlink_list); 7406 7407 static void net_unlink_todo(struct net_device *dev) 7408 { 7409 if (list_empty(&dev->unlink_list)) 7410 list_add_tail(&dev->unlink_list, &net_unlink_list); 7411 } 7412 #endif 7413 7414 static int __netdev_update_lower_level(struct net_device *dev, 7415 struct netdev_nested_priv *priv) 7416 { 7417 dev->lower_level = __netdev_lower_depth(dev) + 1; 7418 7419 #ifdef CONFIG_LOCKDEP 7420 if (!priv) 7421 return 0; 7422 7423 if (priv->flags & NESTED_SYNC_IMM) 7424 dev->nested_level = dev->lower_level - 1; 7425 if (priv->flags & NESTED_SYNC_TODO) 7426 net_unlink_todo(dev); 7427 #endif 7428 return 0; 7429 } 7430 7431 int netdev_walk_all_lower_dev_rcu(struct net_device *dev, 7432 int (*fn)(struct net_device *dev, 7433 struct netdev_nested_priv *priv), 7434 struct netdev_nested_priv *priv) 7435 { 7436 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7437 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7438 int ret, cur = 0; 7439 7440 now = dev; 7441 iter = &dev->adj_list.lower; 7442 7443 while (1) { 7444 if (now != dev) { 7445 ret = fn(now, priv); 7446 if (ret) 7447 return ret; 7448 } 7449 7450 next = NULL; 7451 while (1) { 7452 ldev = netdev_next_lower_dev_rcu(now, &iter); 7453 if (!ldev) 7454 break; 7455 7456 next = ldev; 7457 niter = &ldev->adj_list.lower; 7458 dev_stack[cur] = now; 7459 iter_stack[cur++] = iter; 7460 break; 7461 } 7462 7463 if (!next) { 7464 if (!cur) 7465 return 0; 7466 next = dev_stack[--cur]; 7467 niter = iter_stack[cur]; 7468 } 7469 7470 now = next; 7471 iter = niter; 7472 } 7473 7474 return 0; 7475 } 7476 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); 7477 7478 /** 7479 * netdev_lower_get_first_private_rcu - Get the first ->private from the 7480 * lower neighbour list, RCU 7481 * variant 7482 * @dev: device 7483 * 7484 * Gets the first netdev_adjacent->private from the dev's lower neighbour 7485 * list. The caller must hold RCU read lock. 7486 */ 7487 void *netdev_lower_get_first_private_rcu(struct net_device *dev) 7488 { 7489 struct netdev_adjacent *lower; 7490 7491 lower = list_first_or_null_rcu(&dev->adj_list.lower, 7492 struct netdev_adjacent, list); 7493 if (lower) 7494 return lower->private; 7495 return NULL; 7496 } 7497 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); 7498 7499 /** 7500 * netdev_master_upper_dev_get_rcu - Get master upper device 7501 * @dev: device 7502 * 7503 * Find a master upper device and return pointer to it or NULL in case 7504 * it's not there. The caller must hold the RCU read lock. 7505 */ 7506 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) 7507 { 7508 struct netdev_adjacent *upper; 7509 7510 upper = list_first_or_null_rcu(&dev->adj_list.upper, 7511 struct netdev_adjacent, list); 7512 if (upper && likely(upper->master)) 7513 return upper->dev; 7514 return NULL; 7515 } 7516 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); 7517 7518 static int netdev_adjacent_sysfs_add(struct net_device *dev, 7519 struct net_device *adj_dev, 7520 struct list_head *dev_list) 7521 { 7522 char linkname[IFNAMSIZ+7]; 7523 7524 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7525 "upper_%s" : "lower_%s", adj_dev->name); 7526 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), 7527 linkname); 7528 } 7529 static void netdev_adjacent_sysfs_del(struct net_device *dev, 7530 char *name, 7531 struct list_head *dev_list) 7532 { 7533 char linkname[IFNAMSIZ+7]; 7534 7535 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7536 "upper_%s" : "lower_%s", name); 7537 sysfs_remove_link(&(dev->dev.kobj), linkname); 7538 } 7539 7540 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, 7541 struct net_device *adj_dev, 7542 struct list_head *dev_list) 7543 { 7544 return (dev_list == &dev->adj_list.upper || 7545 dev_list == &dev->adj_list.lower) && 7546 net_eq(dev_net(dev), dev_net(adj_dev)); 7547 } 7548 7549 static int __netdev_adjacent_dev_insert(struct net_device *dev, 7550 struct net_device *adj_dev, 7551 struct list_head *dev_list, 7552 void *private, bool master) 7553 { 7554 struct netdev_adjacent *adj; 7555 int ret; 7556 7557 adj = __netdev_find_adj(adj_dev, dev_list); 7558 7559 if (adj) { 7560 adj->ref_nr += 1; 7561 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", 7562 dev->name, adj_dev->name, adj->ref_nr); 7563 7564 return 0; 7565 } 7566 7567 adj = kmalloc(sizeof(*adj), GFP_KERNEL); 7568 if (!adj) 7569 return -ENOMEM; 7570 7571 adj->dev = adj_dev; 7572 adj->master = master; 7573 adj->ref_nr = 1; 7574 adj->private = private; 7575 adj->ignore = false; 7576 netdev_hold(adj_dev, &adj->dev_tracker, GFP_KERNEL); 7577 7578 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", 7579 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); 7580 7581 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { 7582 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); 7583 if (ret) 7584 goto free_adj; 7585 } 7586 7587 /* Ensure that master link is always the first item in list. */ 7588 if (master) { 7589 ret = sysfs_create_link(&(dev->dev.kobj), 7590 &(adj_dev->dev.kobj), "master"); 7591 if (ret) 7592 goto remove_symlinks; 7593 7594 list_add_rcu(&adj->list, dev_list); 7595 } else { 7596 list_add_tail_rcu(&adj->list, dev_list); 7597 } 7598 7599 return 0; 7600 7601 remove_symlinks: 7602 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7603 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7604 free_adj: 7605 netdev_put(adj_dev, &adj->dev_tracker); 7606 kfree(adj); 7607 7608 return ret; 7609 } 7610 7611 static void __netdev_adjacent_dev_remove(struct net_device *dev, 7612 struct net_device *adj_dev, 7613 u16 ref_nr, 7614 struct list_head *dev_list) 7615 { 7616 struct netdev_adjacent *adj; 7617 7618 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", 7619 dev->name, adj_dev->name, ref_nr); 7620 7621 adj = __netdev_find_adj(adj_dev, dev_list); 7622 7623 if (!adj) { 7624 pr_err("Adjacency does not exist for device %s from %s\n", 7625 dev->name, adj_dev->name); 7626 WARN_ON(1); 7627 return; 7628 } 7629 7630 if (adj->ref_nr > ref_nr) { 7631 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", 7632 dev->name, adj_dev->name, ref_nr, 7633 adj->ref_nr - ref_nr); 7634 adj->ref_nr -= ref_nr; 7635 return; 7636 } 7637 7638 if (adj->master) 7639 sysfs_remove_link(&(dev->dev.kobj), "master"); 7640 7641 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7642 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7643 7644 list_del_rcu(&adj->list); 7645 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", 7646 adj_dev->name, dev->name, adj_dev->name); 7647 netdev_put(adj_dev, &adj->dev_tracker); 7648 kfree_rcu(adj, rcu); 7649 } 7650 7651 static int __netdev_adjacent_dev_link_lists(struct net_device *dev, 7652 struct net_device *upper_dev, 7653 struct list_head *up_list, 7654 struct list_head *down_list, 7655 void *private, bool master) 7656 { 7657 int ret; 7658 7659 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, 7660 private, master); 7661 if (ret) 7662 return ret; 7663 7664 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, 7665 private, false); 7666 if (ret) { 7667 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); 7668 return ret; 7669 } 7670 7671 return 0; 7672 } 7673 7674 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, 7675 struct net_device *upper_dev, 7676 u16 ref_nr, 7677 struct list_head *up_list, 7678 struct list_head *down_list) 7679 { 7680 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); 7681 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); 7682 } 7683 7684 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, 7685 struct net_device *upper_dev, 7686 void *private, bool master) 7687 { 7688 return __netdev_adjacent_dev_link_lists(dev, upper_dev, 7689 &dev->adj_list.upper, 7690 &upper_dev->adj_list.lower, 7691 private, master); 7692 } 7693 7694 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, 7695 struct net_device *upper_dev) 7696 { 7697 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, 7698 &dev->adj_list.upper, 7699 &upper_dev->adj_list.lower); 7700 } 7701 7702 static int __netdev_upper_dev_link(struct net_device *dev, 7703 struct net_device *upper_dev, bool master, 7704 void *upper_priv, void *upper_info, 7705 struct netdev_nested_priv *priv, 7706 struct netlink_ext_ack *extack) 7707 { 7708 struct netdev_notifier_changeupper_info changeupper_info = { 7709 .info = { 7710 .dev = dev, 7711 .extack = extack, 7712 }, 7713 .upper_dev = upper_dev, 7714 .master = master, 7715 .linking = true, 7716 .upper_info = upper_info, 7717 }; 7718 struct net_device *master_dev; 7719 int ret = 0; 7720 7721 ASSERT_RTNL(); 7722 7723 if (dev == upper_dev) 7724 return -EBUSY; 7725 7726 /* To prevent loops, check if dev is not upper device to upper_dev. */ 7727 if (__netdev_has_upper_dev(upper_dev, dev)) 7728 return -EBUSY; 7729 7730 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV) 7731 return -EMLINK; 7732 7733 if (!master) { 7734 if (__netdev_has_upper_dev(dev, upper_dev)) 7735 return -EEXIST; 7736 } else { 7737 master_dev = __netdev_master_upper_dev_get(dev); 7738 if (master_dev) 7739 return master_dev == upper_dev ? -EEXIST : -EBUSY; 7740 } 7741 7742 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7743 &changeupper_info.info); 7744 ret = notifier_to_errno(ret); 7745 if (ret) 7746 return ret; 7747 7748 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, 7749 master); 7750 if (ret) 7751 return ret; 7752 7753 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 7754 &changeupper_info.info); 7755 ret = notifier_to_errno(ret); 7756 if (ret) 7757 goto rollback; 7758 7759 __netdev_update_upper_level(dev, NULL); 7760 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 7761 7762 __netdev_update_lower_level(upper_dev, priv); 7763 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 7764 priv); 7765 7766 return 0; 7767 7768 rollback: 7769 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 7770 7771 return ret; 7772 } 7773 7774 /** 7775 * netdev_upper_dev_link - Add a link to the upper device 7776 * @dev: device 7777 * @upper_dev: new upper device 7778 * @extack: netlink extended ack 7779 * 7780 * Adds a link to device which is upper to this one. The caller must hold 7781 * the RTNL lock. On a failure a negative errno code is returned. 7782 * On success the reference counts are adjusted and the function 7783 * returns zero. 7784 */ 7785 int netdev_upper_dev_link(struct net_device *dev, 7786 struct net_device *upper_dev, 7787 struct netlink_ext_ack *extack) 7788 { 7789 struct netdev_nested_priv priv = { 7790 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7791 .data = NULL, 7792 }; 7793 7794 return __netdev_upper_dev_link(dev, upper_dev, false, 7795 NULL, NULL, &priv, extack); 7796 } 7797 EXPORT_SYMBOL(netdev_upper_dev_link); 7798 7799 /** 7800 * netdev_master_upper_dev_link - Add a master link to the upper device 7801 * @dev: device 7802 * @upper_dev: new upper device 7803 * @upper_priv: upper device private 7804 * @upper_info: upper info to be passed down via notifier 7805 * @extack: netlink extended ack 7806 * 7807 * Adds a link to device which is upper to this one. In this case, only 7808 * one master upper device can be linked, although other non-master devices 7809 * might be linked as well. The caller must hold the RTNL lock. 7810 * On a failure a negative errno code is returned. On success the reference 7811 * counts are adjusted and the function returns zero. 7812 */ 7813 int netdev_master_upper_dev_link(struct net_device *dev, 7814 struct net_device *upper_dev, 7815 void *upper_priv, void *upper_info, 7816 struct netlink_ext_ack *extack) 7817 { 7818 struct netdev_nested_priv priv = { 7819 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7820 .data = NULL, 7821 }; 7822 7823 return __netdev_upper_dev_link(dev, upper_dev, true, 7824 upper_priv, upper_info, &priv, extack); 7825 } 7826 EXPORT_SYMBOL(netdev_master_upper_dev_link); 7827 7828 static void __netdev_upper_dev_unlink(struct net_device *dev, 7829 struct net_device *upper_dev, 7830 struct netdev_nested_priv *priv) 7831 { 7832 struct netdev_notifier_changeupper_info changeupper_info = { 7833 .info = { 7834 .dev = dev, 7835 }, 7836 .upper_dev = upper_dev, 7837 .linking = false, 7838 }; 7839 7840 ASSERT_RTNL(); 7841 7842 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; 7843 7844 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7845 &changeupper_info.info); 7846 7847 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 7848 7849 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 7850 &changeupper_info.info); 7851 7852 __netdev_update_upper_level(dev, NULL); 7853 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 7854 7855 __netdev_update_lower_level(upper_dev, priv); 7856 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 7857 priv); 7858 } 7859 7860 /** 7861 * netdev_upper_dev_unlink - Removes a link to upper device 7862 * @dev: device 7863 * @upper_dev: new upper device 7864 * 7865 * Removes a link to device which is upper to this one. The caller must hold 7866 * the RTNL lock. 7867 */ 7868 void netdev_upper_dev_unlink(struct net_device *dev, 7869 struct net_device *upper_dev) 7870 { 7871 struct netdev_nested_priv priv = { 7872 .flags = NESTED_SYNC_TODO, 7873 .data = NULL, 7874 }; 7875 7876 __netdev_upper_dev_unlink(dev, upper_dev, &priv); 7877 } 7878 EXPORT_SYMBOL(netdev_upper_dev_unlink); 7879 7880 static void __netdev_adjacent_dev_set(struct net_device *upper_dev, 7881 struct net_device *lower_dev, 7882 bool val) 7883 { 7884 struct netdev_adjacent *adj; 7885 7886 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower); 7887 if (adj) 7888 adj->ignore = val; 7889 7890 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper); 7891 if (adj) 7892 adj->ignore = val; 7893 } 7894 7895 static void netdev_adjacent_dev_disable(struct net_device *upper_dev, 7896 struct net_device *lower_dev) 7897 { 7898 __netdev_adjacent_dev_set(upper_dev, lower_dev, true); 7899 } 7900 7901 static void netdev_adjacent_dev_enable(struct net_device *upper_dev, 7902 struct net_device *lower_dev) 7903 { 7904 __netdev_adjacent_dev_set(upper_dev, lower_dev, false); 7905 } 7906 7907 int netdev_adjacent_change_prepare(struct net_device *old_dev, 7908 struct net_device *new_dev, 7909 struct net_device *dev, 7910 struct netlink_ext_ack *extack) 7911 { 7912 struct netdev_nested_priv priv = { 7913 .flags = 0, 7914 .data = NULL, 7915 }; 7916 int err; 7917 7918 if (!new_dev) 7919 return 0; 7920 7921 if (old_dev && new_dev != old_dev) 7922 netdev_adjacent_dev_disable(dev, old_dev); 7923 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv, 7924 extack); 7925 if (err) { 7926 if (old_dev && new_dev != old_dev) 7927 netdev_adjacent_dev_enable(dev, old_dev); 7928 return err; 7929 } 7930 7931 return 0; 7932 } 7933 EXPORT_SYMBOL(netdev_adjacent_change_prepare); 7934 7935 void netdev_adjacent_change_commit(struct net_device *old_dev, 7936 struct net_device *new_dev, 7937 struct net_device *dev) 7938 { 7939 struct netdev_nested_priv priv = { 7940 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7941 .data = NULL, 7942 }; 7943 7944 if (!new_dev || !old_dev) 7945 return; 7946 7947 if (new_dev == old_dev) 7948 return; 7949 7950 netdev_adjacent_dev_enable(dev, old_dev); 7951 __netdev_upper_dev_unlink(old_dev, dev, &priv); 7952 } 7953 EXPORT_SYMBOL(netdev_adjacent_change_commit); 7954 7955 void netdev_adjacent_change_abort(struct net_device *old_dev, 7956 struct net_device *new_dev, 7957 struct net_device *dev) 7958 { 7959 struct netdev_nested_priv priv = { 7960 .flags = 0, 7961 .data = NULL, 7962 }; 7963 7964 if (!new_dev) 7965 return; 7966 7967 if (old_dev && new_dev != old_dev) 7968 netdev_adjacent_dev_enable(dev, old_dev); 7969 7970 __netdev_upper_dev_unlink(new_dev, dev, &priv); 7971 } 7972 EXPORT_SYMBOL(netdev_adjacent_change_abort); 7973 7974 /** 7975 * netdev_bonding_info_change - Dispatch event about slave change 7976 * @dev: device 7977 * @bonding_info: info to dispatch 7978 * 7979 * Send NETDEV_BONDING_INFO to netdev notifiers with info. 7980 * The caller must hold the RTNL lock. 7981 */ 7982 void netdev_bonding_info_change(struct net_device *dev, 7983 struct netdev_bonding_info *bonding_info) 7984 { 7985 struct netdev_notifier_bonding_info info = { 7986 .info.dev = dev, 7987 }; 7988 7989 memcpy(&info.bonding_info, bonding_info, 7990 sizeof(struct netdev_bonding_info)); 7991 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, 7992 &info.info); 7993 } 7994 EXPORT_SYMBOL(netdev_bonding_info_change); 7995 7996 static int netdev_offload_xstats_enable_l3(struct net_device *dev, 7997 struct netlink_ext_ack *extack) 7998 { 7999 struct netdev_notifier_offload_xstats_info info = { 8000 .info.dev = dev, 8001 .info.extack = extack, 8002 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, 8003 }; 8004 int err; 8005 int rc; 8006 8007 dev->offload_xstats_l3 = kzalloc(sizeof(*dev->offload_xstats_l3), 8008 GFP_KERNEL); 8009 if (!dev->offload_xstats_l3) 8010 return -ENOMEM; 8011 8012 rc = call_netdevice_notifiers_info_robust(NETDEV_OFFLOAD_XSTATS_ENABLE, 8013 NETDEV_OFFLOAD_XSTATS_DISABLE, 8014 &info.info); 8015 err = notifier_to_errno(rc); 8016 if (err) 8017 goto free_stats; 8018 8019 return 0; 8020 8021 free_stats: 8022 kfree(dev->offload_xstats_l3); 8023 dev->offload_xstats_l3 = NULL; 8024 return err; 8025 } 8026 8027 int netdev_offload_xstats_enable(struct net_device *dev, 8028 enum netdev_offload_xstats_type type, 8029 struct netlink_ext_ack *extack) 8030 { 8031 ASSERT_RTNL(); 8032 8033 if (netdev_offload_xstats_enabled(dev, type)) 8034 return -EALREADY; 8035 8036 switch (type) { 8037 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 8038 return netdev_offload_xstats_enable_l3(dev, extack); 8039 } 8040 8041 WARN_ON(1); 8042 return -EINVAL; 8043 } 8044 EXPORT_SYMBOL(netdev_offload_xstats_enable); 8045 8046 static void netdev_offload_xstats_disable_l3(struct net_device *dev) 8047 { 8048 struct netdev_notifier_offload_xstats_info info = { 8049 .info.dev = dev, 8050 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, 8051 }; 8052 8053 call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_DISABLE, 8054 &info.info); 8055 kfree(dev->offload_xstats_l3); 8056 dev->offload_xstats_l3 = NULL; 8057 } 8058 8059 int netdev_offload_xstats_disable(struct net_device *dev, 8060 enum netdev_offload_xstats_type type) 8061 { 8062 ASSERT_RTNL(); 8063 8064 if (!netdev_offload_xstats_enabled(dev, type)) 8065 return -EALREADY; 8066 8067 switch (type) { 8068 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 8069 netdev_offload_xstats_disable_l3(dev); 8070 return 0; 8071 } 8072 8073 WARN_ON(1); 8074 return -EINVAL; 8075 } 8076 EXPORT_SYMBOL(netdev_offload_xstats_disable); 8077 8078 static void netdev_offload_xstats_disable_all(struct net_device *dev) 8079 { 8080 netdev_offload_xstats_disable(dev, NETDEV_OFFLOAD_XSTATS_TYPE_L3); 8081 } 8082 8083 static struct rtnl_hw_stats64 * 8084 netdev_offload_xstats_get_ptr(const struct net_device *dev, 8085 enum netdev_offload_xstats_type type) 8086 { 8087 switch (type) { 8088 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 8089 return dev->offload_xstats_l3; 8090 } 8091 8092 WARN_ON(1); 8093 return NULL; 8094 } 8095 8096 bool netdev_offload_xstats_enabled(const struct net_device *dev, 8097 enum netdev_offload_xstats_type type) 8098 { 8099 ASSERT_RTNL(); 8100 8101 return netdev_offload_xstats_get_ptr(dev, type); 8102 } 8103 EXPORT_SYMBOL(netdev_offload_xstats_enabled); 8104 8105 struct netdev_notifier_offload_xstats_ru { 8106 bool used; 8107 }; 8108 8109 struct netdev_notifier_offload_xstats_rd { 8110 struct rtnl_hw_stats64 stats; 8111 bool used; 8112 }; 8113 8114 static void netdev_hw_stats64_add(struct rtnl_hw_stats64 *dest, 8115 const struct rtnl_hw_stats64 *src) 8116 { 8117 dest->rx_packets += src->rx_packets; 8118 dest->tx_packets += src->tx_packets; 8119 dest->rx_bytes += src->rx_bytes; 8120 dest->tx_bytes += src->tx_bytes; 8121 dest->rx_errors += src->rx_errors; 8122 dest->tx_errors += src->tx_errors; 8123 dest->rx_dropped += src->rx_dropped; 8124 dest->tx_dropped += src->tx_dropped; 8125 dest->multicast += src->multicast; 8126 } 8127 8128 static int netdev_offload_xstats_get_used(struct net_device *dev, 8129 enum netdev_offload_xstats_type type, 8130 bool *p_used, 8131 struct netlink_ext_ack *extack) 8132 { 8133 struct netdev_notifier_offload_xstats_ru report_used = {}; 8134 struct netdev_notifier_offload_xstats_info info = { 8135 .info.dev = dev, 8136 .info.extack = extack, 8137 .type = type, 8138 .report_used = &report_used, 8139 }; 8140 int rc; 8141 8142 WARN_ON(!netdev_offload_xstats_enabled(dev, type)); 8143 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_USED, 8144 &info.info); 8145 *p_used = report_used.used; 8146 return notifier_to_errno(rc); 8147 } 8148 8149 static int netdev_offload_xstats_get_stats(struct net_device *dev, 8150 enum netdev_offload_xstats_type type, 8151 struct rtnl_hw_stats64 *p_stats, 8152 bool *p_used, 8153 struct netlink_ext_ack *extack) 8154 { 8155 struct netdev_notifier_offload_xstats_rd report_delta = {}; 8156 struct netdev_notifier_offload_xstats_info info = { 8157 .info.dev = dev, 8158 .info.extack = extack, 8159 .type = type, 8160 .report_delta = &report_delta, 8161 }; 8162 struct rtnl_hw_stats64 *stats; 8163 int rc; 8164 8165 stats = netdev_offload_xstats_get_ptr(dev, type); 8166 if (WARN_ON(!stats)) 8167 return -EINVAL; 8168 8169 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_DELTA, 8170 &info.info); 8171 8172 /* Cache whatever we got, even if there was an error, otherwise the 8173 * successful stats retrievals would get lost. 8174 */ 8175 netdev_hw_stats64_add(stats, &report_delta.stats); 8176 8177 if (p_stats) 8178 *p_stats = *stats; 8179 *p_used = report_delta.used; 8180 8181 return notifier_to_errno(rc); 8182 } 8183 8184 int netdev_offload_xstats_get(struct net_device *dev, 8185 enum netdev_offload_xstats_type type, 8186 struct rtnl_hw_stats64 *p_stats, bool *p_used, 8187 struct netlink_ext_ack *extack) 8188 { 8189 ASSERT_RTNL(); 8190 8191 if (p_stats) 8192 return netdev_offload_xstats_get_stats(dev, type, p_stats, 8193 p_used, extack); 8194 else 8195 return netdev_offload_xstats_get_used(dev, type, p_used, 8196 extack); 8197 } 8198 EXPORT_SYMBOL(netdev_offload_xstats_get); 8199 8200 void 8201 netdev_offload_xstats_report_delta(struct netdev_notifier_offload_xstats_rd *report_delta, 8202 const struct rtnl_hw_stats64 *stats) 8203 { 8204 report_delta->used = true; 8205 netdev_hw_stats64_add(&report_delta->stats, stats); 8206 } 8207 EXPORT_SYMBOL(netdev_offload_xstats_report_delta); 8208 8209 void 8210 netdev_offload_xstats_report_used(struct netdev_notifier_offload_xstats_ru *report_used) 8211 { 8212 report_used->used = true; 8213 } 8214 EXPORT_SYMBOL(netdev_offload_xstats_report_used); 8215 8216 void netdev_offload_xstats_push_delta(struct net_device *dev, 8217 enum netdev_offload_xstats_type type, 8218 const struct rtnl_hw_stats64 *p_stats) 8219 { 8220 struct rtnl_hw_stats64 *stats; 8221 8222 ASSERT_RTNL(); 8223 8224 stats = netdev_offload_xstats_get_ptr(dev, type); 8225 if (WARN_ON(!stats)) 8226 return; 8227 8228 netdev_hw_stats64_add(stats, p_stats); 8229 } 8230 EXPORT_SYMBOL(netdev_offload_xstats_push_delta); 8231 8232 /** 8233 * netdev_get_xmit_slave - Get the xmit slave of master device 8234 * @dev: device 8235 * @skb: The packet 8236 * @all_slaves: assume all the slaves are active 8237 * 8238 * The reference counters are not incremented so the caller must be 8239 * careful with locks. The caller must hold RCU lock. 8240 * %NULL is returned if no slave is found. 8241 */ 8242 8243 struct net_device *netdev_get_xmit_slave(struct net_device *dev, 8244 struct sk_buff *skb, 8245 bool all_slaves) 8246 { 8247 const struct net_device_ops *ops = dev->netdev_ops; 8248 8249 if (!ops->ndo_get_xmit_slave) 8250 return NULL; 8251 return ops->ndo_get_xmit_slave(dev, skb, all_slaves); 8252 } 8253 EXPORT_SYMBOL(netdev_get_xmit_slave); 8254 8255 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev, 8256 struct sock *sk) 8257 { 8258 const struct net_device_ops *ops = dev->netdev_ops; 8259 8260 if (!ops->ndo_sk_get_lower_dev) 8261 return NULL; 8262 return ops->ndo_sk_get_lower_dev(dev, sk); 8263 } 8264 8265 /** 8266 * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket 8267 * @dev: device 8268 * @sk: the socket 8269 * 8270 * %NULL is returned if no lower device is found. 8271 */ 8272 8273 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev, 8274 struct sock *sk) 8275 { 8276 struct net_device *lower; 8277 8278 lower = netdev_sk_get_lower_dev(dev, sk); 8279 while (lower) { 8280 dev = lower; 8281 lower = netdev_sk_get_lower_dev(dev, sk); 8282 } 8283 8284 return dev; 8285 } 8286 EXPORT_SYMBOL(netdev_sk_get_lowest_dev); 8287 8288 static void netdev_adjacent_add_links(struct net_device *dev) 8289 { 8290 struct netdev_adjacent *iter; 8291 8292 struct net *net = dev_net(dev); 8293 8294 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8295 if (!net_eq(net, dev_net(iter->dev))) 8296 continue; 8297 netdev_adjacent_sysfs_add(iter->dev, dev, 8298 &iter->dev->adj_list.lower); 8299 netdev_adjacent_sysfs_add(dev, iter->dev, 8300 &dev->adj_list.upper); 8301 } 8302 8303 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8304 if (!net_eq(net, dev_net(iter->dev))) 8305 continue; 8306 netdev_adjacent_sysfs_add(iter->dev, dev, 8307 &iter->dev->adj_list.upper); 8308 netdev_adjacent_sysfs_add(dev, iter->dev, 8309 &dev->adj_list.lower); 8310 } 8311 } 8312 8313 static void netdev_adjacent_del_links(struct net_device *dev) 8314 { 8315 struct netdev_adjacent *iter; 8316 8317 struct net *net = dev_net(dev); 8318 8319 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8320 if (!net_eq(net, dev_net(iter->dev))) 8321 continue; 8322 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8323 &iter->dev->adj_list.lower); 8324 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8325 &dev->adj_list.upper); 8326 } 8327 8328 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8329 if (!net_eq(net, dev_net(iter->dev))) 8330 continue; 8331 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8332 &iter->dev->adj_list.upper); 8333 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8334 &dev->adj_list.lower); 8335 } 8336 } 8337 8338 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) 8339 { 8340 struct netdev_adjacent *iter; 8341 8342 struct net *net = dev_net(dev); 8343 8344 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8345 if (!net_eq(net, dev_net(iter->dev))) 8346 continue; 8347 netdev_adjacent_sysfs_del(iter->dev, oldname, 8348 &iter->dev->adj_list.lower); 8349 netdev_adjacent_sysfs_add(iter->dev, dev, 8350 &iter->dev->adj_list.lower); 8351 } 8352 8353 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8354 if (!net_eq(net, dev_net(iter->dev))) 8355 continue; 8356 netdev_adjacent_sysfs_del(iter->dev, oldname, 8357 &iter->dev->adj_list.upper); 8358 netdev_adjacent_sysfs_add(iter->dev, dev, 8359 &iter->dev->adj_list.upper); 8360 } 8361 } 8362 8363 void *netdev_lower_dev_get_private(struct net_device *dev, 8364 struct net_device *lower_dev) 8365 { 8366 struct netdev_adjacent *lower; 8367 8368 if (!lower_dev) 8369 return NULL; 8370 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); 8371 if (!lower) 8372 return NULL; 8373 8374 return lower->private; 8375 } 8376 EXPORT_SYMBOL(netdev_lower_dev_get_private); 8377 8378 8379 /** 8380 * netdev_lower_state_changed - Dispatch event about lower device state change 8381 * @lower_dev: device 8382 * @lower_state_info: state to dispatch 8383 * 8384 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. 8385 * The caller must hold the RTNL lock. 8386 */ 8387 void netdev_lower_state_changed(struct net_device *lower_dev, 8388 void *lower_state_info) 8389 { 8390 struct netdev_notifier_changelowerstate_info changelowerstate_info = { 8391 .info.dev = lower_dev, 8392 }; 8393 8394 ASSERT_RTNL(); 8395 changelowerstate_info.lower_state_info = lower_state_info; 8396 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, 8397 &changelowerstate_info.info); 8398 } 8399 EXPORT_SYMBOL(netdev_lower_state_changed); 8400 8401 static void dev_change_rx_flags(struct net_device *dev, int flags) 8402 { 8403 const struct net_device_ops *ops = dev->netdev_ops; 8404 8405 if (ops->ndo_change_rx_flags) 8406 ops->ndo_change_rx_flags(dev, flags); 8407 } 8408 8409 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) 8410 { 8411 unsigned int old_flags = dev->flags; 8412 kuid_t uid; 8413 kgid_t gid; 8414 8415 ASSERT_RTNL(); 8416 8417 dev->flags |= IFF_PROMISC; 8418 dev->promiscuity += inc; 8419 if (dev->promiscuity == 0) { 8420 /* 8421 * Avoid overflow. 8422 * If inc causes overflow, untouch promisc and return error. 8423 */ 8424 if (inc < 0) 8425 dev->flags &= ~IFF_PROMISC; 8426 else { 8427 dev->promiscuity -= inc; 8428 netdev_warn(dev, "promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n"); 8429 return -EOVERFLOW; 8430 } 8431 } 8432 if (dev->flags != old_flags) { 8433 netdev_info(dev, "%s promiscuous mode\n", 8434 dev->flags & IFF_PROMISC ? "entered" : "left"); 8435 if (audit_enabled) { 8436 current_uid_gid(&uid, &gid); 8437 audit_log(audit_context(), GFP_ATOMIC, 8438 AUDIT_ANOM_PROMISCUOUS, 8439 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", 8440 dev->name, (dev->flags & IFF_PROMISC), 8441 (old_flags & IFF_PROMISC), 8442 from_kuid(&init_user_ns, audit_get_loginuid(current)), 8443 from_kuid(&init_user_ns, uid), 8444 from_kgid(&init_user_ns, gid), 8445 audit_get_sessionid(current)); 8446 } 8447 8448 dev_change_rx_flags(dev, IFF_PROMISC); 8449 } 8450 if (notify) 8451 __dev_notify_flags(dev, old_flags, IFF_PROMISC, 0, NULL); 8452 return 0; 8453 } 8454 8455 /** 8456 * dev_set_promiscuity - update promiscuity count on a device 8457 * @dev: device 8458 * @inc: modifier 8459 * 8460 * Add or remove promiscuity from a device. While the count in the device 8461 * remains above zero the interface remains promiscuous. Once it hits zero 8462 * the device reverts back to normal filtering operation. A negative inc 8463 * value is used to drop promiscuity on the device. 8464 * Return 0 if successful or a negative errno code on error. 8465 */ 8466 int dev_set_promiscuity(struct net_device *dev, int inc) 8467 { 8468 unsigned int old_flags = dev->flags; 8469 int err; 8470 8471 err = __dev_set_promiscuity(dev, inc, true); 8472 if (err < 0) 8473 return err; 8474 if (dev->flags != old_flags) 8475 dev_set_rx_mode(dev); 8476 return err; 8477 } 8478 EXPORT_SYMBOL(dev_set_promiscuity); 8479 8480 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) 8481 { 8482 unsigned int old_flags = dev->flags, old_gflags = dev->gflags; 8483 8484 ASSERT_RTNL(); 8485 8486 dev->flags |= IFF_ALLMULTI; 8487 dev->allmulti += inc; 8488 if (dev->allmulti == 0) { 8489 /* 8490 * Avoid overflow. 8491 * If inc causes overflow, untouch allmulti and return error. 8492 */ 8493 if (inc < 0) 8494 dev->flags &= ~IFF_ALLMULTI; 8495 else { 8496 dev->allmulti -= inc; 8497 netdev_warn(dev, "allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n"); 8498 return -EOVERFLOW; 8499 } 8500 } 8501 if (dev->flags ^ old_flags) { 8502 netdev_info(dev, "%s allmulticast mode\n", 8503 dev->flags & IFF_ALLMULTI ? "entered" : "left"); 8504 dev_change_rx_flags(dev, IFF_ALLMULTI); 8505 dev_set_rx_mode(dev); 8506 if (notify) 8507 __dev_notify_flags(dev, old_flags, 8508 dev->gflags ^ old_gflags, 0, NULL); 8509 } 8510 return 0; 8511 } 8512 8513 /** 8514 * dev_set_allmulti - update allmulti count on a device 8515 * @dev: device 8516 * @inc: modifier 8517 * 8518 * Add or remove reception of all multicast frames to a device. While the 8519 * count in the device remains above zero the interface remains listening 8520 * to all interfaces. Once it hits zero the device reverts back to normal 8521 * filtering operation. A negative @inc value is used to drop the counter 8522 * when releasing a resource needing all multicasts. 8523 * Return 0 if successful or a negative errno code on error. 8524 */ 8525 8526 int dev_set_allmulti(struct net_device *dev, int inc) 8527 { 8528 return __dev_set_allmulti(dev, inc, true); 8529 } 8530 EXPORT_SYMBOL(dev_set_allmulti); 8531 8532 /* 8533 * Upload unicast and multicast address lists to device and 8534 * configure RX filtering. When the device doesn't support unicast 8535 * filtering it is put in promiscuous mode while unicast addresses 8536 * are present. 8537 */ 8538 void __dev_set_rx_mode(struct net_device *dev) 8539 { 8540 const struct net_device_ops *ops = dev->netdev_ops; 8541 8542 /* dev_open will call this function so the list will stay sane. */ 8543 if (!(dev->flags&IFF_UP)) 8544 return; 8545 8546 if (!netif_device_present(dev)) 8547 return; 8548 8549 if (!(dev->priv_flags & IFF_UNICAST_FLT)) { 8550 /* Unicast addresses changes may only happen under the rtnl, 8551 * therefore calling __dev_set_promiscuity here is safe. 8552 */ 8553 if (!netdev_uc_empty(dev) && !dev->uc_promisc) { 8554 __dev_set_promiscuity(dev, 1, false); 8555 dev->uc_promisc = true; 8556 } else if (netdev_uc_empty(dev) && dev->uc_promisc) { 8557 __dev_set_promiscuity(dev, -1, false); 8558 dev->uc_promisc = false; 8559 } 8560 } 8561 8562 if (ops->ndo_set_rx_mode) 8563 ops->ndo_set_rx_mode(dev); 8564 } 8565 8566 void dev_set_rx_mode(struct net_device *dev) 8567 { 8568 netif_addr_lock_bh(dev); 8569 __dev_set_rx_mode(dev); 8570 netif_addr_unlock_bh(dev); 8571 } 8572 8573 /** 8574 * dev_get_flags - get flags reported to userspace 8575 * @dev: device 8576 * 8577 * Get the combination of flag bits exported through APIs to userspace. 8578 */ 8579 unsigned int dev_get_flags(const struct net_device *dev) 8580 { 8581 unsigned int flags; 8582 8583 flags = (dev->flags & ~(IFF_PROMISC | 8584 IFF_ALLMULTI | 8585 IFF_RUNNING | 8586 IFF_LOWER_UP | 8587 IFF_DORMANT)) | 8588 (dev->gflags & (IFF_PROMISC | 8589 IFF_ALLMULTI)); 8590 8591 if (netif_running(dev)) { 8592 if (netif_oper_up(dev)) 8593 flags |= IFF_RUNNING; 8594 if (netif_carrier_ok(dev)) 8595 flags |= IFF_LOWER_UP; 8596 if (netif_dormant(dev)) 8597 flags |= IFF_DORMANT; 8598 } 8599 8600 return flags; 8601 } 8602 EXPORT_SYMBOL(dev_get_flags); 8603 8604 int __dev_change_flags(struct net_device *dev, unsigned int flags, 8605 struct netlink_ext_ack *extack) 8606 { 8607 unsigned int old_flags = dev->flags; 8608 int ret; 8609 8610 ASSERT_RTNL(); 8611 8612 /* 8613 * Set the flags on our device. 8614 */ 8615 8616 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | 8617 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | 8618 IFF_AUTOMEDIA)) | 8619 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | 8620 IFF_ALLMULTI)); 8621 8622 /* 8623 * Load in the correct multicast list now the flags have changed. 8624 */ 8625 8626 if ((old_flags ^ flags) & IFF_MULTICAST) 8627 dev_change_rx_flags(dev, IFF_MULTICAST); 8628 8629 dev_set_rx_mode(dev); 8630 8631 /* 8632 * Have we downed the interface. We handle IFF_UP ourselves 8633 * according to user attempts to set it, rather than blindly 8634 * setting it. 8635 */ 8636 8637 ret = 0; 8638 if ((old_flags ^ flags) & IFF_UP) { 8639 if (old_flags & IFF_UP) 8640 __dev_close(dev); 8641 else 8642 ret = __dev_open(dev, extack); 8643 } 8644 8645 if ((flags ^ dev->gflags) & IFF_PROMISC) { 8646 int inc = (flags & IFF_PROMISC) ? 1 : -1; 8647 unsigned int old_flags = dev->flags; 8648 8649 dev->gflags ^= IFF_PROMISC; 8650 8651 if (__dev_set_promiscuity(dev, inc, false) >= 0) 8652 if (dev->flags != old_flags) 8653 dev_set_rx_mode(dev); 8654 } 8655 8656 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI 8657 * is important. Some (broken) drivers set IFF_PROMISC, when 8658 * IFF_ALLMULTI is requested not asking us and not reporting. 8659 */ 8660 if ((flags ^ dev->gflags) & IFF_ALLMULTI) { 8661 int inc = (flags & IFF_ALLMULTI) ? 1 : -1; 8662 8663 dev->gflags ^= IFF_ALLMULTI; 8664 __dev_set_allmulti(dev, inc, false); 8665 } 8666 8667 return ret; 8668 } 8669 8670 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, 8671 unsigned int gchanges, u32 portid, 8672 const struct nlmsghdr *nlh) 8673 { 8674 unsigned int changes = dev->flags ^ old_flags; 8675 8676 if (gchanges) 8677 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC, portid, nlh); 8678 8679 if (changes & IFF_UP) { 8680 if (dev->flags & IFF_UP) 8681 call_netdevice_notifiers(NETDEV_UP, dev); 8682 else 8683 call_netdevice_notifiers(NETDEV_DOWN, dev); 8684 } 8685 8686 if (dev->flags & IFF_UP && 8687 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { 8688 struct netdev_notifier_change_info change_info = { 8689 .info = { 8690 .dev = dev, 8691 }, 8692 .flags_changed = changes, 8693 }; 8694 8695 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info); 8696 } 8697 } 8698 8699 /** 8700 * dev_change_flags - change device settings 8701 * @dev: device 8702 * @flags: device state flags 8703 * @extack: netlink extended ack 8704 * 8705 * Change settings on device based state flags. The flags are 8706 * in the userspace exported format. 8707 */ 8708 int dev_change_flags(struct net_device *dev, unsigned int flags, 8709 struct netlink_ext_ack *extack) 8710 { 8711 int ret; 8712 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; 8713 8714 ret = __dev_change_flags(dev, flags, extack); 8715 if (ret < 0) 8716 return ret; 8717 8718 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); 8719 __dev_notify_flags(dev, old_flags, changes, 0, NULL); 8720 return ret; 8721 } 8722 EXPORT_SYMBOL(dev_change_flags); 8723 8724 int __dev_set_mtu(struct net_device *dev, int new_mtu) 8725 { 8726 const struct net_device_ops *ops = dev->netdev_ops; 8727 8728 if (ops->ndo_change_mtu) 8729 return ops->ndo_change_mtu(dev, new_mtu); 8730 8731 /* Pairs with all the lockless reads of dev->mtu in the stack */ 8732 WRITE_ONCE(dev->mtu, new_mtu); 8733 return 0; 8734 } 8735 EXPORT_SYMBOL(__dev_set_mtu); 8736 8737 int dev_validate_mtu(struct net_device *dev, int new_mtu, 8738 struct netlink_ext_ack *extack) 8739 { 8740 /* MTU must be positive, and in range */ 8741 if (new_mtu < 0 || new_mtu < dev->min_mtu) { 8742 NL_SET_ERR_MSG(extack, "mtu less than device minimum"); 8743 return -EINVAL; 8744 } 8745 8746 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { 8747 NL_SET_ERR_MSG(extack, "mtu greater than device maximum"); 8748 return -EINVAL; 8749 } 8750 return 0; 8751 } 8752 8753 /** 8754 * dev_set_mtu_ext - Change maximum transfer unit 8755 * @dev: device 8756 * @new_mtu: new transfer unit 8757 * @extack: netlink extended ack 8758 * 8759 * Change the maximum transfer size of the network device. 8760 */ 8761 int dev_set_mtu_ext(struct net_device *dev, int new_mtu, 8762 struct netlink_ext_ack *extack) 8763 { 8764 int err, orig_mtu; 8765 8766 if (new_mtu == dev->mtu) 8767 return 0; 8768 8769 err = dev_validate_mtu(dev, new_mtu, extack); 8770 if (err) 8771 return err; 8772 8773 if (!netif_device_present(dev)) 8774 return -ENODEV; 8775 8776 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); 8777 err = notifier_to_errno(err); 8778 if (err) 8779 return err; 8780 8781 orig_mtu = dev->mtu; 8782 err = __dev_set_mtu(dev, new_mtu); 8783 8784 if (!err) { 8785 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8786 orig_mtu); 8787 err = notifier_to_errno(err); 8788 if (err) { 8789 /* setting mtu back and notifying everyone again, 8790 * so that they have a chance to revert changes. 8791 */ 8792 __dev_set_mtu(dev, orig_mtu); 8793 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8794 new_mtu); 8795 } 8796 } 8797 return err; 8798 } 8799 8800 int dev_set_mtu(struct net_device *dev, int new_mtu) 8801 { 8802 struct netlink_ext_ack extack; 8803 int err; 8804 8805 memset(&extack, 0, sizeof(extack)); 8806 err = dev_set_mtu_ext(dev, new_mtu, &extack); 8807 if (err && extack._msg) 8808 net_err_ratelimited("%s: %s\n", dev->name, extack._msg); 8809 return err; 8810 } 8811 EXPORT_SYMBOL(dev_set_mtu); 8812 8813 /** 8814 * dev_change_tx_queue_len - Change TX queue length of a netdevice 8815 * @dev: device 8816 * @new_len: new tx queue length 8817 */ 8818 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len) 8819 { 8820 unsigned int orig_len = dev->tx_queue_len; 8821 int res; 8822 8823 if (new_len != (unsigned int)new_len) 8824 return -ERANGE; 8825 8826 if (new_len != orig_len) { 8827 dev->tx_queue_len = new_len; 8828 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev); 8829 res = notifier_to_errno(res); 8830 if (res) 8831 goto err_rollback; 8832 res = dev_qdisc_change_tx_queue_len(dev); 8833 if (res) 8834 goto err_rollback; 8835 } 8836 8837 return 0; 8838 8839 err_rollback: 8840 netdev_err(dev, "refused to change device tx_queue_len\n"); 8841 dev->tx_queue_len = orig_len; 8842 return res; 8843 } 8844 8845 /** 8846 * dev_set_group - Change group this device belongs to 8847 * @dev: device 8848 * @new_group: group this device should belong to 8849 */ 8850 void dev_set_group(struct net_device *dev, int new_group) 8851 { 8852 dev->group = new_group; 8853 } 8854 8855 /** 8856 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR. 8857 * @dev: device 8858 * @addr: new address 8859 * @extack: netlink extended ack 8860 */ 8861 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr, 8862 struct netlink_ext_ack *extack) 8863 { 8864 struct netdev_notifier_pre_changeaddr_info info = { 8865 .info.dev = dev, 8866 .info.extack = extack, 8867 .dev_addr = addr, 8868 }; 8869 int rc; 8870 8871 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info); 8872 return notifier_to_errno(rc); 8873 } 8874 EXPORT_SYMBOL(dev_pre_changeaddr_notify); 8875 8876 /** 8877 * dev_set_mac_address - Change Media Access Control Address 8878 * @dev: device 8879 * @sa: new address 8880 * @extack: netlink extended ack 8881 * 8882 * Change the hardware (MAC) address of the device 8883 */ 8884 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa, 8885 struct netlink_ext_ack *extack) 8886 { 8887 const struct net_device_ops *ops = dev->netdev_ops; 8888 int err; 8889 8890 if (!ops->ndo_set_mac_address) 8891 return -EOPNOTSUPP; 8892 if (sa->sa_family != dev->type) 8893 return -EINVAL; 8894 if (!netif_device_present(dev)) 8895 return -ENODEV; 8896 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack); 8897 if (err) 8898 return err; 8899 if (memcmp(dev->dev_addr, sa->sa_data, dev->addr_len)) { 8900 err = ops->ndo_set_mac_address(dev, sa); 8901 if (err) 8902 return err; 8903 } 8904 dev->addr_assign_type = NET_ADDR_SET; 8905 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); 8906 add_device_randomness(dev->dev_addr, dev->addr_len); 8907 return 0; 8908 } 8909 EXPORT_SYMBOL(dev_set_mac_address); 8910 8911 static DECLARE_RWSEM(dev_addr_sem); 8912 8913 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa, 8914 struct netlink_ext_ack *extack) 8915 { 8916 int ret; 8917 8918 down_write(&dev_addr_sem); 8919 ret = dev_set_mac_address(dev, sa, extack); 8920 up_write(&dev_addr_sem); 8921 return ret; 8922 } 8923 EXPORT_SYMBOL(dev_set_mac_address_user); 8924 8925 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name) 8926 { 8927 size_t size = sizeof(sa->sa_data_min); 8928 struct net_device *dev; 8929 int ret = 0; 8930 8931 down_read(&dev_addr_sem); 8932 rcu_read_lock(); 8933 8934 dev = dev_get_by_name_rcu(net, dev_name); 8935 if (!dev) { 8936 ret = -ENODEV; 8937 goto unlock; 8938 } 8939 if (!dev->addr_len) 8940 memset(sa->sa_data, 0, size); 8941 else 8942 memcpy(sa->sa_data, dev->dev_addr, 8943 min_t(size_t, size, dev->addr_len)); 8944 sa->sa_family = dev->type; 8945 8946 unlock: 8947 rcu_read_unlock(); 8948 up_read(&dev_addr_sem); 8949 return ret; 8950 } 8951 EXPORT_SYMBOL(dev_get_mac_address); 8952 8953 /** 8954 * dev_change_carrier - Change device carrier 8955 * @dev: device 8956 * @new_carrier: new value 8957 * 8958 * Change device carrier 8959 */ 8960 int dev_change_carrier(struct net_device *dev, bool new_carrier) 8961 { 8962 const struct net_device_ops *ops = dev->netdev_ops; 8963 8964 if (!ops->ndo_change_carrier) 8965 return -EOPNOTSUPP; 8966 if (!netif_device_present(dev)) 8967 return -ENODEV; 8968 return ops->ndo_change_carrier(dev, new_carrier); 8969 } 8970 8971 /** 8972 * dev_get_phys_port_id - Get device physical port ID 8973 * @dev: device 8974 * @ppid: port ID 8975 * 8976 * Get device physical port ID 8977 */ 8978 int dev_get_phys_port_id(struct net_device *dev, 8979 struct netdev_phys_item_id *ppid) 8980 { 8981 const struct net_device_ops *ops = dev->netdev_ops; 8982 8983 if (!ops->ndo_get_phys_port_id) 8984 return -EOPNOTSUPP; 8985 return ops->ndo_get_phys_port_id(dev, ppid); 8986 } 8987 8988 /** 8989 * dev_get_phys_port_name - Get device physical port name 8990 * @dev: device 8991 * @name: port name 8992 * @len: limit of bytes to copy to name 8993 * 8994 * Get device physical port name 8995 */ 8996 int dev_get_phys_port_name(struct net_device *dev, 8997 char *name, size_t len) 8998 { 8999 const struct net_device_ops *ops = dev->netdev_ops; 9000 int err; 9001 9002 if (ops->ndo_get_phys_port_name) { 9003 err = ops->ndo_get_phys_port_name(dev, name, len); 9004 if (err != -EOPNOTSUPP) 9005 return err; 9006 } 9007 return devlink_compat_phys_port_name_get(dev, name, len); 9008 } 9009 9010 /** 9011 * dev_get_port_parent_id - Get the device's port parent identifier 9012 * @dev: network device 9013 * @ppid: pointer to a storage for the port's parent identifier 9014 * @recurse: allow/disallow recursion to lower devices 9015 * 9016 * Get the devices's port parent identifier 9017 */ 9018 int dev_get_port_parent_id(struct net_device *dev, 9019 struct netdev_phys_item_id *ppid, 9020 bool recurse) 9021 { 9022 const struct net_device_ops *ops = dev->netdev_ops; 9023 struct netdev_phys_item_id first = { }; 9024 struct net_device *lower_dev; 9025 struct list_head *iter; 9026 int err; 9027 9028 if (ops->ndo_get_port_parent_id) { 9029 err = ops->ndo_get_port_parent_id(dev, ppid); 9030 if (err != -EOPNOTSUPP) 9031 return err; 9032 } 9033 9034 err = devlink_compat_switch_id_get(dev, ppid); 9035 if (!recurse || err != -EOPNOTSUPP) 9036 return err; 9037 9038 netdev_for_each_lower_dev(dev, lower_dev, iter) { 9039 err = dev_get_port_parent_id(lower_dev, ppid, true); 9040 if (err) 9041 break; 9042 if (!first.id_len) 9043 first = *ppid; 9044 else if (memcmp(&first, ppid, sizeof(*ppid))) 9045 return -EOPNOTSUPP; 9046 } 9047 9048 return err; 9049 } 9050 EXPORT_SYMBOL(dev_get_port_parent_id); 9051 9052 /** 9053 * netdev_port_same_parent_id - Indicate if two network devices have 9054 * the same port parent identifier 9055 * @a: first network device 9056 * @b: second network device 9057 */ 9058 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b) 9059 { 9060 struct netdev_phys_item_id a_id = { }; 9061 struct netdev_phys_item_id b_id = { }; 9062 9063 if (dev_get_port_parent_id(a, &a_id, true) || 9064 dev_get_port_parent_id(b, &b_id, true)) 9065 return false; 9066 9067 return netdev_phys_item_id_same(&a_id, &b_id); 9068 } 9069 EXPORT_SYMBOL(netdev_port_same_parent_id); 9070 9071 static void netdev_dpll_pin_assign(struct net_device *dev, struct dpll_pin *dpll_pin) 9072 { 9073 #if IS_ENABLED(CONFIG_DPLL) 9074 rtnl_lock(); 9075 dev->dpll_pin = dpll_pin; 9076 rtnl_unlock(); 9077 #endif 9078 } 9079 9080 void netdev_dpll_pin_set(struct net_device *dev, struct dpll_pin *dpll_pin) 9081 { 9082 WARN_ON(!dpll_pin); 9083 netdev_dpll_pin_assign(dev, dpll_pin); 9084 } 9085 EXPORT_SYMBOL(netdev_dpll_pin_set); 9086 9087 void netdev_dpll_pin_clear(struct net_device *dev) 9088 { 9089 netdev_dpll_pin_assign(dev, NULL); 9090 } 9091 EXPORT_SYMBOL(netdev_dpll_pin_clear); 9092 9093 /** 9094 * dev_change_proto_down - set carrier according to proto_down. 9095 * 9096 * @dev: device 9097 * @proto_down: new value 9098 */ 9099 int dev_change_proto_down(struct net_device *dev, bool proto_down) 9100 { 9101 if (!(dev->priv_flags & IFF_CHANGE_PROTO_DOWN)) 9102 return -EOPNOTSUPP; 9103 if (!netif_device_present(dev)) 9104 return -ENODEV; 9105 if (proto_down) 9106 netif_carrier_off(dev); 9107 else 9108 netif_carrier_on(dev); 9109 dev->proto_down = proto_down; 9110 return 0; 9111 } 9112 9113 /** 9114 * dev_change_proto_down_reason - proto down reason 9115 * 9116 * @dev: device 9117 * @mask: proto down mask 9118 * @value: proto down value 9119 */ 9120 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask, 9121 u32 value) 9122 { 9123 int b; 9124 9125 if (!mask) { 9126 dev->proto_down_reason = value; 9127 } else { 9128 for_each_set_bit(b, &mask, 32) { 9129 if (value & (1 << b)) 9130 dev->proto_down_reason |= BIT(b); 9131 else 9132 dev->proto_down_reason &= ~BIT(b); 9133 } 9134 } 9135 } 9136 9137 struct bpf_xdp_link { 9138 struct bpf_link link; 9139 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */ 9140 int flags; 9141 }; 9142 9143 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags) 9144 { 9145 if (flags & XDP_FLAGS_HW_MODE) 9146 return XDP_MODE_HW; 9147 if (flags & XDP_FLAGS_DRV_MODE) 9148 return XDP_MODE_DRV; 9149 if (flags & XDP_FLAGS_SKB_MODE) 9150 return XDP_MODE_SKB; 9151 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB; 9152 } 9153 9154 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode) 9155 { 9156 switch (mode) { 9157 case XDP_MODE_SKB: 9158 return generic_xdp_install; 9159 case XDP_MODE_DRV: 9160 case XDP_MODE_HW: 9161 return dev->netdev_ops->ndo_bpf; 9162 default: 9163 return NULL; 9164 } 9165 } 9166 9167 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev, 9168 enum bpf_xdp_mode mode) 9169 { 9170 return dev->xdp_state[mode].link; 9171 } 9172 9173 static struct bpf_prog *dev_xdp_prog(struct net_device *dev, 9174 enum bpf_xdp_mode mode) 9175 { 9176 struct bpf_xdp_link *link = dev_xdp_link(dev, mode); 9177 9178 if (link) 9179 return link->link.prog; 9180 return dev->xdp_state[mode].prog; 9181 } 9182 9183 u8 dev_xdp_prog_count(struct net_device *dev) 9184 { 9185 u8 count = 0; 9186 int i; 9187 9188 for (i = 0; i < __MAX_XDP_MODE; i++) 9189 if (dev->xdp_state[i].prog || dev->xdp_state[i].link) 9190 count++; 9191 return count; 9192 } 9193 EXPORT_SYMBOL_GPL(dev_xdp_prog_count); 9194 9195 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode) 9196 { 9197 struct bpf_prog *prog = dev_xdp_prog(dev, mode); 9198 9199 return prog ? prog->aux->id : 0; 9200 } 9201 9202 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode, 9203 struct bpf_xdp_link *link) 9204 { 9205 dev->xdp_state[mode].link = link; 9206 dev->xdp_state[mode].prog = NULL; 9207 } 9208 9209 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode, 9210 struct bpf_prog *prog) 9211 { 9212 dev->xdp_state[mode].link = NULL; 9213 dev->xdp_state[mode].prog = prog; 9214 } 9215 9216 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode, 9217 bpf_op_t bpf_op, struct netlink_ext_ack *extack, 9218 u32 flags, struct bpf_prog *prog) 9219 { 9220 struct netdev_bpf xdp; 9221 int err; 9222 9223 memset(&xdp, 0, sizeof(xdp)); 9224 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG; 9225 xdp.extack = extack; 9226 xdp.flags = flags; 9227 xdp.prog = prog; 9228 9229 /* Drivers assume refcnt is already incremented (i.e, prog pointer is 9230 * "moved" into driver), so they don't increment it on their own, but 9231 * they do decrement refcnt when program is detached or replaced. 9232 * Given net_device also owns link/prog, we need to bump refcnt here 9233 * to prevent drivers from underflowing it. 9234 */ 9235 if (prog) 9236 bpf_prog_inc(prog); 9237 err = bpf_op(dev, &xdp); 9238 if (err) { 9239 if (prog) 9240 bpf_prog_put(prog); 9241 return err; 9242 } 9243 9244 if (mode != XDP_MODE_HW) 9245 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog); 9246 9247 return 0; 9248 } 9249 9250 static void dev_xdp_uninstall(struct net_device *dev) 9251 { 9252 struct bpf_xdp_link *link; 9253 struct bpf_prog *prog; 9254 enum bpf_xdp_mode mode; 9255 bpf_op_t bpf_op; 9256 9257 ASSERT_RTNL(); 9258 9259 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) { 9260 prog = dev_xdp_prog(dev, mode); 9261 if (!prog) 9262 continue; 9263 9264 bpf_op = dev_xdp_bpf_op(dev, mode); 9265 if (!bpf_op) 9266 continue; 9267 9268 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9269 9270 /* auto-detach link from net device */ 9271 link = dev_xdp_link(dev, mode); 9272 if (link) 9273 link->dev = NULL; 9274 else 9275 bpf_prog_put(prog); 9276 9277 dev_xdp_set_link(dev, mode, NULL); 9278 } 9279 } 9280 9281 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack, 9282 struct bpf_xdp_link *link, struct bpf_prog *new_prog, 9283 struct bpf_prog *old_prog, u32 flags) 9284 { 9285 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES); 9286 struct bpf_prog *cur_prog; 9287 struct net_device *upper; 9288 struct list_head *iter; 9289 enum bpf_xdp_mode mode; 9290 bpf_op_t bpf_op; 9291 int err; 9292 9293 ASSERT_RTNL(); 9294 9295 /* either link or prog attachment, never both */ 9296 if (link && (new_prog || old_prog)) 9297 return -EINVAL; 9298 /* link supports only XDP mode flags */ 9299 if (link && (flags & ~XDP_FLAGS_MODES)) { 9300 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment"); 9301 return -EINVAL; 9302 } 9303 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */ 9304 if (num_modes > 1) { 9305 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set"); 9306 return -EINVAL; 9307 } 9308 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */ 9309 if (!num_modes && dev_xdp_prog_count(dev) > 1) { 9310 NL_SET_ERR_MSG(extack, 9311 "More than one program loaded, unset mode is ambiguous"); 9312 return -EINVAL; 9313 } 9314 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */ 9315 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) { 9316 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified"); 9317 return -EINVAL; 9318 } 9319 9320 mode = dev_xdp_mode(dev, flags); 9321 /* can't replace attached link */ 9322 if (dev_xdp_link(dev, mode)) { 9323 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link"); 9324 return -EBUSY; 9325 } 9326 9327 /* don't allow if an upper device already has a program */ 9328 netdev_for_each_upper_dev_rcu(dev, upper, iter) { 9329 if (dev_xdp_prog_count(upper) > 0) { 9330 NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program"); 9331 return -EEXIST; 9332 } 9333 } 9334 9335 cur_prog = dev_xdp_prog(dev, mode); 9336 /* can't replace attached prog with link */ 9337 if (link && cur_prog) { 9338 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link"); 9339 return -EBUSY; 9340 } 9341 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) { 9342 NL_SET_ERR_MSG(extack, "Active program does not match expected"); 9343 return -EEXIST; 9344 } 9345 9346 /* put effective new program into new_prog */ 9347 if (link) 9348 new_prog = link->link.prog; 9349 9350 if (new_prog) { 9351 bool offload = mode == XDP_MODE_HW; 9352 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB 9353 ? XDP_MODE_DRV : XDP_MODE_SKB; 9354 9355 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) { 9356 NL_SET_ERR_MSG(extack, "XDP program already attached"); 9357 return -EBUSY; 9358 } 9359 if (!offload && dev_xdp_prog(dev, other_mode)) { 9360 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time"); 9361 return -EEXIST; 9362 } 9363 if (!offload && bpf_prog_is_offloaded(new_prog->aux)) { 9364 NL_SET_ERR_MSG(extack, "Using offloaded program without HW_MODE flag is not supported"); 9365 return -EINVAL; 9366 } 9367 if (bpf_prog_is_dev_bound(new_prog->aux) && !bpf_offload_dev_match(new_prog, dev)) { 9368 NL_SET_ERR_MSG(extack, "Program bound to different device"); 9369 return -EINVAL; 9370 } 9371 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) { 9372 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device"); 9373 return -EINVAL; 9374 } 9375 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) { 9376 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device"); 9377 return -EINVAL; 9378 } 9379 } 9380 9381 /* don't call drivers if the effective program didn't change */ 9382 if (new_prog != cur_prog) { 9383 bpf_op = dev_xdp_bpf_op(dev, mode); 9384 if (!bpf_op) { 9385 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode"); 9386 return -EOPNOTSUPP; 9387 } 9388 9389 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog); 9390 if (err) 9391 return err; 9392 } 9393 9394 if (link) 9395 dev_xdp_set_link(dev, mode, link); 9396 else 9397 dev_xdp_set_prog(dev, mode, new_prog); 9398 if (cur_prog) 9399 bpf_prog_put(cur_prog); 9400 9401 return 0; 9402 } 9403 9404 static int dev_xdp_attach_link(struct net_device *dev, 9405 struct netlink_ext_ack *extack, 9406 struct bpf_xdp_link *link) 9407 { 9408 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags); 9409 } 9410 9411 static int dev_xdp_detach_link(struct net_device *dev, 9412 struct netlink_ext_ack *extack, 9413 struct bpf_xdp_link *link) 9414 { 9415 enum bpf_xdp_mode mode; 9416 bpf_op_t bpf_op; 9417 9418 ASSERT_RTNL(); 9419 9420 mode = dev_xdp_mode(dev, link->flags); 9421 if (dev_xdp_link(dev, mode) != link) 9422 return -EINVAL; 9423 9424 bpf_op = dev_xdp_bpf_op(dev, mode); 9425 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9426 dev_xdp_set_link(dev, mode, NULL); 9427 return 0; 9428 } 9429 9430 static void bpf_xdp_link_release(struct bpf_link *link) 9431 { 9432 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9433 9434 rtnl_lock(); 9435 9436 /* if racing with net_device's tear down, xdp_link->dev might be 9437 * already NULL, in which case link was already auto-detached 9438 */ 9439 if (xdp_link->dev) { 9440 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link)); 9441 xdp_link->dev = NULL; 9442 } 9443 9444 rtnl_unlock(); 9445 } 9446 9447 static int bpf_xdp_link_detach(struct bpf_link *link) 9448 { 9449 bpf_xdp_link_release(link); 9450 return 0; 9451 } 9452 9453 static void bpf_xdp_link_dealloc(struct bpf_link *link) 9454 { 9455 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9456 9457 kfree(xdp_link); 9458 } 9459 9460 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link, 9461 struct seq_file *seq) 9462 { 9463 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9464 u32 ifindex = 0; 9465 9466 rtnl_lock(); 9467 if (xdp_link->dev) 9468 ifindex = xdp_link->dev->ifindex; 9469 rtnl_unlock(); 9470 9471 seq_printf(seq, "ifindex:\t%u\n", ifindex); 9472 } 9473 9474 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link, 9475 struct bpf_link_info *info) 9476 { 9477 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9478 u32 ifindex = 0; 9479 9480 rtnl_lock(); 9481 if (xdp_link->dev) 9482 ifindex = xdp_link->dev->ifindex; 9483 rtnl_unlock(); 9484 9485 info->xdp.ifindex = ifindex; 9486 return 0; 9487 } 9488 9489 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog, 9490 struct bpf_prog *old_prog) 9491 { 9492 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9493 enum bpf_xdp_mode mode; 9494 bpf_op_t bpf_op; 9495 int err = 0; 9496 9497 rtnl_lock(); 9498 9499 /* link might have been auto-released already, so fail */ 9500 if (!xdp_link->dev) { 9501 err = -ENOLINK; 9502 goto out_unlock; 9503 } 9504 9505 if (old_prog && link->prog != old_prog) { 9506 err = -EPERM; 9507 goto out_unlock; 9508 } 9509 old_prog = link->prog; 9510 if (old_prog->type != new_prog->type || 9511 old_prog->expected_attach_type != new_prog->expected_attach_type) { 9512 err = -EINVAL; 9513 goto out_unlock; 9514 } 9515 9516 if (old_prog == new_prog) { 9517 /* no-op, don't disturb drivers */ 9518 bpf_prog_put(new_prog); 9519 goto out_unlock; 9520 } 9521 9522 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags); 9523 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode); 9524 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL, 9525 xdp_link->flags, new_prog); 9526 if (err) 9527 goto out_unlock; 9528 9529 old_prog = xchg(&link->prog, new_prog); 9530 bpf_prog_put(old_prog); 9531 9532 out_unlock: 9533 rtnl_unlock(); 9534 return err; 9535 } 9536 9537 static const struct bpf_link_ops bpf_xdp_link_lops = { 9538 .release = bpf_xdp_link_release, 9539 .dealloc = bpf_xdp_link_dealloc, 9540 .detach = bpf_xdp_link_detach, 9541 .show_fdinfo = bpf_xdp_link_show_fdinfo, 9542 .fill_link_info = bpf_xdp_link_fill_link_info, 9543 .update_prog = bpf_xdp_link_update, 9544 }; 9545 9546 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog) 9547 { 9548 struct net *net = current->nsproxy->net_ns; 9549 struct bpf_link_primer link_primer; 9550 struct netlink_ext_ack extack = {}; 9551 struct bpf_xdp_link *link; 9552 struct net_device *dev; 9553 int err, fd; 9554 9555 rtnl_lock(); 9556 dev = dev_get_by_index(net, attr->link_create.target_ifindex); 9557 if (!dev) { 9558 rtnl_unlock(); 9559 return -EINVAL; 9560 } 9561 9562 link = kzalloc(sizeof(*link), GFP_USER); 9563 if (!link) { 9564 err = -ENOMEM; 9565 goto unlock; 9566 } 9567 9568 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog); 9569 link->dev = dev; 9570 link->flags = attr->link_create.flags; 9571 9572 err = bpf_link_prime(&link->link, &link_primer); 9573 if (err) { 9574 kfree(link); 9575 goto unlock; 9576 } 9577 9578 err = dev_xdp_attach_link(dev, &extack, link); 9579 rtnl_unlock(); 9580 9581 if (err) { 9582 link->dev = NULL; 9583 bpf_link_cleanup(&link_primer); 9584 trace_bpf_xdp_link_attach_failed(extack._msg); 9585 goto out_put_dev; 9586 } 9587 9588 fd = bpf_link_settle(&link_primer); 9589 /* link itself doesn't hold dev's refcnt to not complicate shutdown */ 9590 dev_put(dev); 9591 return fd; 9592 9593 unlock: 9594 rtnl_unlock(); 9595 9596 out_put_dev: 9597 dev_put(dev); 9598 return err; 9599 } 9600 9601 /** 9602 * dev_change_xdp_fd - set or clear a bpf program for a device rx path 9603 * @dev: device 9604 * @extack: netlink extended ack 9605 * @fd: new program fd or negative value to clear 9606 * @expected_fd: old program fd that userspace expects to replace or clear 9607 * @flags: xdp-related flags 9608 * 9609 * Set or clear a bpf program for a device 9610 */ 9611 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, 9612 int fd, int expected_fd, u32 flags) 9613 { 9614 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags); 9615 struct bpf_prog *new_prog = NULL, *old_prog = NULL; 9616 int err; 9617 9618 ASSERT_RTNL(); 9619 9620 if (fd >= 0) { 9621 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP, 9622 mode != XDP_MODE_SKB); 9623 if (IS_ERR(new_prog)) 9624 return PTR_ERR(new_prog); 9625 } 9626 9627 if (expected_fd >= 0) { 9628 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP, 9629 mode != XDP_MODE_SKB); 9630 if (IS_ERR(old_prog)) { 9631 err = PTR_ERR(old_prog); 9632 old_prog = NULL; 9633 goto err_out; 9634 } 9635 } 9636 9637 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags); 9638 9639 err_out: 9640 if (err && new_prog) 9641 bpf_prog_put(new_prog); 9642 if (old_prog) 9643 bpf_prog_put(old_prog); 9644 return err; 9645 } 9646 9647 /** 9648 * dev_index_reserve() - allocate an ifindex in a namespace 9649 * @net: the applicable net namespace 9650 * @ifindex: requested ifindex, pass %0 to get one allocated 9651 * 9652 * Allocate a ifindex for a new device. Caller must either use the ifindex 9653 * to store the device (via list_netdevice()) or call dev_index_release() 9654 * to give the index up. 9655 * 9656 * Return: a suitable unique value for a new device interface number or -errno. 9657 */ 9658 static int dev_index_reserve(struct net *net, u32 ifindex) 9659 { 9660 int err; 9661 9662 if (ifindex > INT_MAX) { 9663 DEBUG_NET_WARN_ON_ONCE(1); 9664 return -EINVAL; 9665 } 9666 9667 if (!ifindex) 9668 err = xa_alloc_cyclic(&net->dev_by_index, &ifindex, NULL, 9669 xa_limit_31b, &net->ifindex, GFP_KERNEL); 9670 else 9671 err = xa_insert(&net->dev_by_index, ifindex, NULL, GFP_KERNEL); 9672 if (err < 0) 9673 return err; 9674 9675 return ifindex; 9676 } 9677 9678 static void dev_index_release(struct net *net, int ifindex) 9679 { 9680 /* Expect only unused indexes, unlist_netdevice() removes the used */ 9681 WARN_ON(xa_erase(&net->dev_by_index, ifindex)); 9682 } 9683 9684 /* Delayed registration/unregisteration */ 9685 LIST_HEAD(net_todo_list); 9686 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); 9687 9688 static void net_set_todo(struct net_device *dev) 9689 { 9690 list_add_tail(&dev->todo_list, &net_todo_list); 9691 atomic_inc(&dev_net(dev)->dev_unreg_count); 9692 } 9693 9694 static netdev_features_t netdev_sync_upper_features(struct net_device *lower, 9695 struct net_device *upper, netdev_features_t features) 9696 { 9697 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9698 netdev_features_t feature; 9699 int feature_bit; 9700 9701 for_each_netdev_feature(upper_disables, feature_bit) { 9702 feature = __NETIF_F_BIT(feature_bit); 9703 if (!(upper->wanted_features & feature) 9704 && (features & feature)) { 9705 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", 9706 &feature, upper->name); 9707 features &= ~feature; 9708 } 9709 } 9710 9711 return features; 9712 } 9713 9714 static void netdev_sync_lower_features(struct net_device *upper, 9715 struct net_device *lower, netdev_features_t features) 9716 { 9717 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9718 netdev_features_t feature; 9719 int feature_bit; 9720 9721 for_each_netdev_feature(upper_disables, feature_bit) { 9722 feature = __NETIF_F_BIT(feature_bit); 9723 if (!(features & feature) && (lower->features & feature)) { 9724 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", 9725 &feature, lower->name); 9726 lower->wanted_features &= ~feature; 9727 __netdev_update_features(lower); 9728 9729 if (unlikely(lower->features & feature)) 9730 netdev_WARN(upper, "failed to disable %pNF on %s!\n", 9731 &feature, lower->name); 9732 else 9733 netdev_features_change(lower); 9734 } 9735 } 9736 } 9737 9738 static netdev_features_t netdev_fix_features(struct net_device *dev, 9739 netdev_features_t features) 9740 { 9741 /* Fix illegal checksum combinations */ 9742 if ((features & NETIF_F_HW_CSUM) && 9743 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { 9744 netdev_warn(dev, "mixed HW and IP checksum settings.\n"); 9745 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); 9746 } 9747 9748 /* TSO requires that SG is present as well. */ 9749 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { 9750 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); 9751 features &= ~NETIF_F_ALL_TSO; 9752 } 9753 9754 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && 9755 !(features & NETIF_F_IP_CSUM)) { 9756 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); 9757 features &= ~NETIF_F_TSO; 9758 features &= ~NETIF_F_TSO_ECN; 9759 } 9760 9761 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && 9762 !(features & NETIF_F_IPV6_CSUM)) { 9763 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); 9764 features &= ~NETIF_F_TSO6; 9765 } 9766 9767 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ 9768 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) 9769 features &= ~NETIF_F_TSO_MANGLEID; 9770 9771 /* TSO ECN requires that TSO is present as well. */ 9772 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) 9773 features &= ~NETIF_F_TSO_ECN; 9774 9775 /* Software GSO depends on SG. */ 9776 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { 9777 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); 9778 features &= ~NETIF_F_GSO; 9779 } 9780 9781 /* GSO partial features require GSO partial be set */ 9782 if ((features & dev->gso_partial_features) && 9783 !(features & NETIF_F_GSO_PARTIAL)) { 9784 netdev_dbg(dev, 9785 "Dropping partially supported GSO features since no GSO partial.\n"); 9786 features &= ~dev->gso_partial_features; 9787 } 9788 9789 if (!(features & NETIF_F_RXCSUM)) { 9790 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet 9791 * successfully merged by hardware must also have the 9792 * checksum verified by hardware. If the user does not 9793 * want to enable RXCSUM, logically, we should disable GRO_HW. 9794 */ 9795 if (features & NETIF_F_GRO_HW) { 9796 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n"); 9797 features &= ~NETIF_F_GRO_HW; 9798 } 9799 } 9800 9801 /* LRO/HW-GRO features cannot be combined with RX-FCS */ 9802 if (features & NETIF_F_RXFCS) { 9803 if (features & NETIF_F_LRO) { 9804 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n"); 9805 features &= ~NETIF_F_LRO; 9806 } 9807 9808 if (features & NETIF_F_GRO_HW) { 9809 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n"); 9810 features &= ~NETIF_F_GRO_HW; 9811 } 9812 } 9813 9814 if ((features & NETIF_F_GRO_HW) && (features & NETIF_F_LRO)) { 9815 netdev_dbg(dev, "Dropping LRO feature since HW-GRO is requested.\n"); 9816 features &= ~NETIF_F_LRO; 9817 } 9818 9819 if (features & NETIF_F_HW_TLS_TX) { 9820 bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) == 9821 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM); 9822 bool hw_csum = features & NETIF_F_HW_CSUM; 9823 9824 if (!ip_csum && !hw_csum) { 9825 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n"); 9826 features &= ~NETIF_F_HW_TLS_TX; 9827 } 9828 } 9829 9830 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) { 9831 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n"); 9832 features &= ~NETIF_F_HW_TLS_RX; 9833 } 9834 9835 return features; 9836 } 9837 9838 int __netdev_update_features(struct net_device *dev) 9839 { 9840 struct net_device *upper, *lower; 9841 netdev_features_t features; 9842 struct list_head *iter; 9843 int err = -1; 9844 9845 ASSERT_RTNL(); 9846 9847 features = netdev_get_wanted_features(dev); 9848 9849 if (dev->netdev_ops->ndo_fix_features) 9850 features = dev->netdev_ops->ndo_fix_features(dev, features); 9851 9852 /* driver might be less strict about feature dependencies */ 9853 features = netdev_fix_features(dev, features); 9854 9855 /* some features can't be enabled if they're off on an upper device */ 9856 netdev_for_each_upper_dev_rcu(dev, upper, iter) 9857 features = netdev_sync_upper_features(dev, upper, features); 9858 9859 if (dev->features == features) 9860 goto sync_lower; 9861 9862 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", 9863 &dev->features, &features); 9864 9865 if (dev->netdev_ops->ndo_set_features) 9866 err = dev->netdev_ops->ndo_set_features(dev, features); 9867 else 9868 err = 0; 9869 9870 if (unlikely(err < 0)) { 9871 netdev_err(dev, 9872 "set_features() failed (%d); wanted %pNF, left %pNF\n", 9873 err, &features, &dev->features); 9874 /* return non-0 since some features might have changed and 9875 * it's better to fire a spurious notification than miss it 9876 */ 9877 return -1; 9878 } 9879 9880 sync_lower: 9881 /* some features must be disabled on lower devices when disabled 9882 * on an upper device (think: bonding master or bridge) 9883 */ 9884 netdev_for_each_lower_dev(dev, lower, iter) 9885 netdev_sync_lower_features(dev, lower, features); 9886 9887 if (!err) { 9888 netdev_features_t diff = features ^ dev->features; 9889 9890 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) { 9891 /* udp_tunnel_{get,drop}_rx_info both need 9892 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the 9893 * device, or they won't do anything. 9894 * Thus we need to update dev->features 9895 * *before* calling udp_tunnel_get_rx_info, 9896 * but *after* calling udp_tunnel_drop_rx_info. 9897 */ 9898 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) { 9899 dev->features = features; 9900 udp_tunnel_get_rx_info(dev); 9901 } else { 9902 udp_tunnel_drop_rx_info(dev); 9903 } 9904 } 9905 9906 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) { 9907 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) { 9908 dev->features = features; 9909 err |= vlan_get_rx_ctag_filter_info(dev); 9910 } else { 9911 vlan_drop_rx_ctag_filter_info(dev); 9912 } 9913 } 9914 9915 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) { 9916 if (features & NETIF_F_HW_VLAN_STAG_FILTER) { 9917 dev->features = features; 9918 err |= vlan_get_rx_stag_filter_info(dev); 9919 } else { 9920 vlan_drop_rx_stag_filter_info(dev); 9921 } 9922 } 9923 9924 dev->features = features; 9925 } 9926 9927 return err < 0 ? 0 : 1; 9928 } 9929 9930 /** 9931 * netdev_update_features - recalculate device features 9932 * @dev: the device to check 9933 * 9934 * Recalculate dev->features set and send notifications if it 9935 * has changed. Should be called after driver or hardware dependent 9936 * conditions might have changed that influence the features. 9937 */ 9938 void netdev_update_features(struct net_device *dev) 9939 { 9940 if (__netdev_update_features(dev)) 9941 netdev_features_change(dev); 9942 } 9943 EXPORT_SYMBOL(netdev_update_features); 9944 9945 /** 9946 * netdev_change_features - recalculate device features 9947 * @dev: the device to check 9948 * 9949 * Recalculate dev->features set and send notifications even 9950 * if they have not changed. Should be called instead of 9951 * netdev_update_features() if also dev->vlan_features might 9952 * have changed to allow the changes to be propagated to stacked 9953 * VLAN devices. 9954 */ 9955 void netdev_change_features(struct net_device *dev) 9956 { 9957 __netdev_update_features(dev); 9958 netdev_features_change(dev); 9959 } 9960 EXPORT_SYMBOL(netdev_change_features); 9961 9962 /** 9963 * netif_stacked_transfer_operstate - transfer operstate 9964 * @rootdev: the root or lower level device to transfer state from 9965 * @dev: the device to transfer operstate to 9966 * 9967 * Transfer operational state from root to device. This is normally 9968 * called when a stacking relationship exists between the root 9969 * device and the device(a leaf device). 9970 */ 9971 void netif_stacked_transfer_operstate(const struct net_device *rootdev, 9972 struct net_device *dev) 9973 { 9974 if (rootdev->operstate == IF_OPER_DORMANT) 9975 netif_dormant_on(dev); 9976 else 9977 netif_dormant_off(dev); 9978 9979 if (rootdev->operstate == IF_OPER_TESTING) 9980 netif_testing_on(dev); 9981 else 9982 netif_testing_off(dev); 9983 9984 if (netif_carrier_ok(rootdev)) 9985 netif_carrier_on(dev); 9986 else 9987 netif_carrier_off(dev); 9988 } 9989 EXPORT_SYMBOL(netif_stacked_transfer_operstate); 9990 9991 static int netif_alloc_rx_queues(struct net_device *dev) 9992 { 9993 unsigned int i, count = dev->num_rx_queues; 9994 struct netdev_rx_queue *rx; 9995 size_t sz = count * sizeof(*rx); 9996 int err = 0; 9997 9998 BUG_ON(count < 1); 9999 10000 rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10001 if (!rx) 10002 return -ENOMEM; 10003 10004 dev->_rx = rx; 10005 10006 for (i = 0; i < count; i++) { 10007 rx[i].dev = dev; 10008 10009 /* XDP RX-queue setup */ 10010 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0); 10011 if (err < 0) 10012 goto err_rxq_info; 10013 } 10014 return 0; 10015 10016 err_rxq_info: 10017 /* Rollback successful reg's and free other resources */ 10018 while (i--) 10019 xdp_rxq_info_unreg(&rx[i].xdp_rxq); 10020 kvfree(dev->_rx); 10021 dev->_rx = NULL; 10022 return err; 10023 } 10024 10025 static void netif_free_rx_queues(struct net_device *dev) 10026 { 10027 unsigned int i, count = dev->num_rx_queues; 10028 10029 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */ 10030 if (!dev->_rx) 10031 return; 10032 10033 for (i = 0; i < count; i++) 10034 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq); 10035 10036 kvfree(dev->_rx); 10037 } 10038 10039 static void netdev_init_one_queue(struct net_device *dev, 10040 struct netdev_queue *queue, void *_unused) 10041 { 10042 /* Initialize queue lock */ 10043 spin_lock_init(&queue->_xmit_lock); 10044 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); 10045 queue->xmit_lock_owner = -1; 10046 netdev_queue_numa_node_write(queue, NUMA_NO_NODE); 10047 queue->dev = dev; 10048 #ifdef CONFIG_BQL 10049 dql_init(&queue->dql, HZ); 10050 #endif 10051 } 10052 10053 static void netif_free_tx_queues(struct net_device *dev) 10054 { 10055 kvfree(dev->_tx); 10056 } 10057 10058 static int netif_alloc_netdev_queues(struct net_device *dev) 10059 { 10060 unsigned int count = dev->num_tx_queues; 10061 struct netdev_queue *tx; 10062 size_t sz = count * sizeof(*tx); 10063 10064 if (count < 1 || count > 0xffff) 10065 return -EINVAL; 10066 10067 tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10068 if (!tx) 10069 return -ENOMEM; 10070 10071 dev->_tx = tx; 10072 10073 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); 10074 spin_lock_init(&dev->tx_global_lock); 10075 10076 return 0; 10077 } 10078 10079 void netif_tx_stop_all_queues(struct net_device *dev) 10080 { 10081 unsigned int i; 10082 10083 for (i = 0; i < dev->num_tx_queues; i++) { 10084 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 10085 10086 netif_tx_stop_queue(txq); 10087 } 10088 } 10089 EXPORT_SYMBOL(netif_tx_stop_all_queues); 10090 10091 static int netdev_do_alloc_pcpu_stats(struct net_device *dev) 10092 { 10093 void __percpu *v; 10094 10095 /* Drivers implementing ndo_get_peer_dev must support tstat 10096 * accounting, so that skb_do_redirect() can bump the dev's 10097 * RX stats upon network namespace switch. 10098 */ 10099 if (dev->netdev_ops->ndo_get_peer_dev && 10100 dev->pcpu_stat_type != NETDEV_PCPU_STAT_TSTATS) 10101 return -EOPNOTSUPP; 10102 10103 switch (dev->pcpu_stat_type) { 10104 case NETDEV_PCPU_STAT_NONE: 10105 return 0; 10106 case NETDEV_PCPU_STAT_LSTATS: 10107 v = dev->lstats = netdev_alloc_pcpu_stats(struct pcpu_lstats); 10108 break; 10109 case NETDEV_PCPU_STAT_TSTATS: 10110 v = dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats); 10111 break; 10112 case NETDEV_PCPU_STAT_DSTATS: 10113 v = dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats); 10114 break; 10115 default: 10116 return -EINVAL; 10117 } 10118 10119 return v ? 0 : -ENOMEM; 10120 } 10121 10122 static void netdev_do_free_pcpu_stats(struct net_device *dev) 10123 { 10124 switch (dev->pcpu_stat_type) { 10125 case NETDEV_PCPU_STAT_NONE: 10126 return; 10127 case NETDEV_PCPU_STAT_LSTATS: 10128 free_percpu(dev->lstats); 10129 break; 10130 case NETDEV_PCPU_STAT_TSTATS: 10131 free_percpu(dev->tstats); 10132 break; 10133 case NETDEV_PCPU_STAT_DSTATS: 10134 free_percpu(dev->dstats); 10135 break; 10136 } 10137 } 10138 10139 /** 10140 * register_netdevice() - register a network device 10141 * @dev: device to register 10142 * 10143 * Take a prepared network device structure and make it externally accessible. 10144 * A %NETDEV_REGISTER message is sent to the netdev notifier chain. 10145 * Callers must hold the rtnl lock - you may want register_netdev() 10146 * instead of this. 10147 */ 10148 int register_netdevice(struct net_device *dev) 10149 { 10150 int ret; 10151 struct net *net = dev_net(dev); 10152 10153 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE < 10154 NETDEV_FEATURE_COUNT); 10155 BUG_ON(dev_boot_phase); 10156 ASSERT_RTNL(); 10157 10158 might_sleep(); 10159 10160 /* When net_device's are persistent, this will be fatal. */ 10161 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); 10162 BUG_ON(!net); 10163 10164 ret = ethtool_check_ops(dev->ethtool_ops); 10165 if (ret) 10166 return ret; 10167 10168 spin_lock_init(&dev->addr_list_lock); 10169 netdev_set_addr_lockdep_class(dev); 10170 10171 ret = dev_get_valid_name(net, dev, dev->name); 10172 if (ret < 0) 10173 goto out; 10174 10175 ret = -ENOMEM; 10176 dev->name_node = netdev_name_node_head_alloc(dev); 10177 if (!dev->name_node) 10178 goto out; 10179 10180 /* Init, if this function is available */ 10181 if (dev->netdev_ops->ndo_init) { 10182 ret = dev->netdev_ops->ndo_init(dev); 10183 if (ret) { 10184 if (ret > 0) 10185 ret = -EIO; 10186 goto err_free_name; 10187 } 10188 } 10189 10190 if (((dev->hw_features | dev->features) & 10191 NETIF_F_HW_VLAN_CTAG_FILTER) && 10192 (!dev->netdev_ops->ndo_vlan_rx_add_vid || 10193 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { 10194 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); 10195 ret = -EINVAL; 10196 goto err_uninit; 10197 } 10198 10199 ret = netdev_do_alloc_pcpu_stats(dev); 10200 if (ret) 10201 goto err_uninit; 10202 10203 ret = dev_index_reserve(net, dev->ifindex); 10204 if (ret < 0) 10205 goto err_free_pcpu; 10206 dev->ifindex = ret; 10207 10208 /* Transfer changeable features to wanted_features and enable 10209 * software offloads (GSO and GRO). 10210 */ 10211 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF); 10212 dev->features |= NETIF_F_SOFT_FEATURES; 10213 10214 if (dev->udp_tunnel_nic_info) { 10215 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10216 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10217 } 10218 10219 dev->wanted_features = dev->features & dev->hw_features; 10220 10221 if (!(dev->flags & IFF_LOOPBACK)) 10222 dev->hw_features |= NETIF_F_NOCACHE_COPY; 10223 10224 /* If IPv4 TCP segmentation offload is supported we should also 10225 * allow the device to enable segmenting the frame with the option 10226 * of ignoring a static IP ID value. This doesn't enable the 10227 * feature itself but allows the user to enable it later. 10228 */ 10229 if (dev->hw_features & NETIF_F_TSO) 10230 dev->hw_features |= NETIF_F_TSO_MANGLEID; 10231 if (dev->vlan_features & NETIF_F_TSO) 10232 dev->vlan_features |= NETIF_F_TSO_MANGLEID; 10233 if (dev->mpls_features & NETIF_F_TSO) 10234 dev->mpls_features |= NETIF_F_TSO_MANGLEID; 10235 if (dev->hw_enc_features & NETIF_F_TSO) 10236 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; 10237 10238 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. 10239 */ 10240 dev->vlan_features |= NETIF_F_HIGHDMA; 10241 10242 /* Make NETIF_F_SG inheritable to tunnel devices. 10243 */ 10244 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; 10245 10246 /* Make NETIF_F_SG inheritable to MPLS. 10247 */ 10248 dev->mpls_features |= NETIF_F_SG; 10249 10250 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); 10251 ret = notifier_to_errno(ret); 10252 if (ret) 10253 goto err_ifindex_release; 10254 10255 ret = netdev_register_kobject(dev); 10256 write_lock(&dev_base_lock); 10257 dev->reg_state = ret ? NETREG_UNREGISTERED : NETREG_REGISTERED; 10258 write_unlock(&dev_base_lock); 10259 if (ret) 10260 goto err_uninit_notify; 10261 10262 __netdev_update_features(dev); 10263 10264 /* 10265 * Default initial state at registry is that the 10266 * device is present. 10267 */ 10268 10269 set_bit(__LINK_STATE_PRESENT, &dev->state); 10270 10271 linkwatch_init_dev(dev); 10272 10273 dev_init_scheduler(dev); 10274 10275 netdev_hold(dev, &dev->dev_registered_tracker, GFP_KERNEL); 10276 list_netdevice(dev); 10277 10278 add_device_randomness(dev->dev_addr, dev->addr_len); 10279 10280 /* If the device has permanent device address, driver should 10281 * set dev_addr and also addr_assign_type should be set to 10282 * NET_ADDR_PERM (default value). 10283 */ 10284 if (dev->addr_assign_type == NET_ADDR_PERM) 10285 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); 10286 10287 /* Notify protocols, that a new device appeared. */ 10288 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); 10289 ret = notifier_to_errno(ret); 10290 if (ret) { 10291 /* Expect explicit free_netdev() on failure */ 10292 dev->needs_free_netdev = false; 10293 unregister_netdevice_queue(dev, NULL); 10294 goto out; 10295 } 10296 /* 10297 * Prevent userspace races by waiting until the network 10298 * device is fully setup before sending notifications. 10299 */ 10300 if (!dev->rtnl_link_ops || 10301 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 10302 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL); 10303 10304 out: 10305 return ret; 10306 10307 err_uninit_notify: 10308 call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev); 10309 err_ifindex_release: 10310 dev_index_release(net, dev->ifindex); 10311 err_free_pcpu: 10312 netdev_do_free_pcpu_stats(dev); 10313 err_uninit: 10314 if (dev->netdev_ops->ndo_uninit) 10315 dev->netdev_ops->ndo_uninit(dev); 10316 if (dev->priv_destructor) 10317 dev->priv_destructor(dev); 10318 err_free_name: 10319 netdev_name_node_free(dev->name_node); 10320 goto out; 10321 } 10322 EXPORT_SYMBOL(register_netdevice); 10323 10324 /** 10325 * init_dummy_netdev - init a dummy network device for NAPI 10326 * @dev: device to init 10327 * 10328 * This takes a network device structure and initialize the minimum 10329 * amount of fields so it can be used to schedule NAPI polls without 10330 * registering a full blown interface. This is to be used by drivers 10331 * that need to tie several hardware interfaces to a single NAPI 10332 * poll scheduler due to HW limitations. 10333 */ 10334 int init_dummy_netdev(struct net_device *dev) 10335 { 10336 /* Clear everything. Note we don't initialize spinlocks 10337 * are they aren't supposed to be taken by any of the 10338 * NAPI code and this dummy netdev is supposed to be 10339 * only ever used for NAPI polls 10340 */ 10341 memset(dev, 0, sizeof(struct net_device)); 10342 10343 /* make sure we BUG if trying to hit standard 10344 * register/unregister code path 10345 */ 10346 dev->reg_state = NETREG_DUMMY; 10347 10348 /* NAPI wants this */ 10349 INIT_LIST_HEAD(&dev->napi_list); 10350 10351 /* a dummy interface is started by default */ 10352 set_bit(__LINK_STATE_PRESENT, &dev->state); 10353 set_bit(__LINK_STATE_START, &dev->state); 10354 10355 /* napi_busy_loop stats accounting wants this */ 10356 dev_net_set(dev, &init_net); 10357 10358 /* Note : We dont allocate pcpu_refcnt for dummy devices, 10359 * because users of this 'device' dont need to change 10360 * its refcount. 10361 */ 10362 10363 return 0; 10364 } 10365 EXPORT_SYMBOL_GPL(init_dummy_netdev); 10366 10367 10368 /** 10369 * register_netdev - register a network device 10370 * @dev: device to register 10371 * 10372 * Take a completed network device structure and add it to the kernel 10373 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 10374 * chain. 0 is returned on success. A negative errno code is returned 10375 * on a failure to set up the device, or if the name is a duplicate. 10376 * 10377 * This is a wrapper around register_netdevice that takes the rtnl semaphore 10378 * and expands the device name if you passed a format string to 10379 * alloc_netdev. 10380 */ 10381 int register_netdev(struct net_device *dev) 10382 { 10383 int err; 10384 10385 if (rtnl_lock_killable()) 10386 return -EINTR; 10387 err = register_netdevice(dev); 10388 rtnl_unlock(); 10389 return err; 10390 } 10391 EXPORT_SYMBOL(register_netdev); 10392 10393 int netdev_refcnt_read(const struct net_device *dev) 10394 { 10395 #ifdef CONFIG_PCPU_DEV_REFCNT 10396 int i, refcnt = 0; 10397 10398 for_each_possible_cpu(i) 10399 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); 10400 return refcnt; 10401 #else 10402 return refcount_read(&dev->dev_refcnt); 10403 #endif 10404 } 10405 EXPORT_SYMBOL(netdev_refcnt_read); 10406 10407 int netdev_unregister_timeout_secs __read_mostly = 10; 10408 10409 #define WAIT_REFS_MIN_MSECS 1 10410 #define WAIT_REFS_MAX_MSECS 250 10411 /** 10412 * netdev_wait_allrefs_any - wait until all references are gone. 10413 * @list: list of net_devices to wait on 10414 * 10415 * This is called when unregistering network devices. 10416 * 10417 * Any protocol or device that holds a reference should register 10418 * for netdevice notification, and cleanup and put back the 10419 * reference if they receive an UNREGISTER event. 10420 * We can get stuck here if buggy protocols don't correctly 10421 * call dev_put. 10422 */ 10423 static struct net_device *netdev_wait_allrefs_any(struct list_head *list) 10424 { 10425 unsigned long rebroadcast_time, warning_time; 10426 struct net_device *dev; 10427 int wait = 0; 10428 10429 rebroadcast_time = warning_time = jiffies; 10430 10431 list_for_each_entry(dev, list, todo_list) 10432 if (netdev_refcnt_read(dev) == 1) 10433 return dev; 10434 10435 while (true) { 10436 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { 10437 rtnl_lock(); 10438 10439 /* Rebroadcast unregister notification */ 10440 list_for_each_entry(dev, list, todo_list) 10441 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10442 10443 __rtnl_unlock(); 10444 rcu_barrier(); 10445 rtnl_lock(); 10446 10447 list_for_each_entry(dev, list, todo_list) 10448 if (test_bit(__LINK_STATE_LINKWATCH_PENDING, 10449 &dev->state)) { 10450 /* We must not have linkwatch events 10451 * pending on unregister. If this 10452 * happens, we simply run the queue 10453 * unscheduled, resulting in a noop 10454 * for this device. 10455 */ 10456 linkwatch_run_queue(); 10457 break; 10458 } 10459 10460 __rtnl_unlock(); 10461 10462 rebroadcast_time = jiffies; 10463 } 10464 10465 if (!wait) { 10466 rcu_barrier(); 10467 wait = WAIT_REFS_MIN_MSECS; 10468 } else { 10469 msleep(wait); 10470 wait = min(wait << 1, WAIT_REFS_MAX_MSECS); 10471 } 10472 10473 list_for_each_entry(dev, list, todo_list) 10474 if (netdev_refcnt_read(dev) == 1) 10475 return dev; 10476 10477 if (time_after(jiffies, warning_time + 10478 READ_ONCE(netdev_unregister_timeout_secs) * HZ)) { 10479 list_for_each_entry(dev, list, todo_list) { 10480 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", 10481 dev->name, netdev_refcnt_read(dev)); 10482 ref_tracker_dir_print(&dev->refcnt_tracker, 10); 10483 } 10484 10485 warning_time = jiffies; 10486 } 10487 } 10488 } 10489 10490 /* The sequence is: 10491 * 10492 * rtnl_lock(); 10493 * ... 10494 * register_netdevice(x1); 10495 * register_netdevice(x2); 10496 * ... 10497 * unregister_netdevice(y1); 10498 * unregister_netdevice(y2); 10499 * ... 10500 * rtnl_unlock(); 10501 * free_netdev(y1); 10502 * free_netdev(y2); 10503 * 10504 * We are invoked by rtnl_unlock(). 10505 * This allows us to deal with problems: 10506 * 1) We can delete sysfs objects which invoke hotplug 10507 * without deadlocking with linkwatch via keventd. 10508 * 2) Since we run with the RTNL semaphore not held, we can sleep 10509 * safely in order to wait for the netdev refcnt to drop to zero. 10510 * 10511 * We must not return until all unregister events added during 10512 * the interval the lock was held have been completed. 10513 */ 10514 void netdev_run_todo(void) 10515 { 10516 struct net_device *dev, *tmp; 10517 struct list_head list; 10518 #ifdef CONFIG_LOCKDEP 10519 struct list_head unlink_list; 10520 10521 list_replace_init(&net_unlink_list, &unlink_list); 10522 10523 while (!list_empty(&unlink_list)) { 10524 struct net_device *dev = list_first_entry(&unlink_list, 10525 struct net_device, 10526 unlink_list); 10527 list_del_init(&dev->unlink_list); 10528 dev->nested_level = dev->lower_level - 1; 10529 } 10530 #endif 10531 10532 /* Snapshot list, allow later requests */ 10533 list_replace_init(&net_todo_list, &list); 10534 10535 __rtnl_unlock(); 10536 10537 /* Wait for rcu callbacks to finish before next phase */ 10538 if (!list_empty(&list)) 10539 rcu_barrier(); 10540 10541 list_for_each_entry_safe(dev, tmp, &list, todo_list) { 10542 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { 10543 netdev_WARN(dev, "run_todo but not unregistering\n"); 10544 list_del(&dev->todo_list); 10545 continue; 10546 } 10547 10548 write_lock(&dev_base_lock); 10549 dev->reg_state = NETREG_UNREGISTERED; 10550 write_unlock(&dev_base_lock); 10551 linkwatch_sync_dev(dev); 10552 } 10553 10554 while (!list_empty(&list)) { 10555 dev = netdev_wait_allrefs_any(&list); 10556 list_del(&dev->todo_list); 10557 10558 /* paranoia */ 10559 BUG_ON(netdev_refcnt_read(dev) != 1); 10560 BUG_ON(!list_empty(&dev->ptype_all)); 10561 BUG_ON(!list_empty(&dev->ptype_specific)); 10562 WARN_ON(rcu_access_pointer(dev->ip_ptr)); 10563 WARN_ON(rcu_access_pointer(dev->ip6_ptr)); 10564 10565 netdev_do_free_pcpu_stats(dev); 10566 if (dev->priv_destructor) 10567 dev->priv_destructor(dev); 10568 if (dev->needs_free_netdev) 10569 free_netdev(dev); 10570 10571 if (atomic_dec_and_test(&dev_net(dev)->dev_unreg_count)) 10572 wake_up(&netdev_unregistering_wq); 10573 10574 /* Free network device */ 10575 kobject_put(&dev->dev.kobj); 10576 } 10577 } 10578 10579 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has 10580 * all the same fields in the same order as net_device_stats, with only 10581 * the type differing, but rtnl_link_stats64 may have additional fields 10582 * at the end for newer counters. 10583 */ 10584 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, 10585 const struct net_device_stats *netdev_stats) 10586 { 10587 size_t i, n = sizeof(*netdev_stats) / sizeof(atomic_long_t); 10588 const atomic_long_t *src = (atomic_long_t *)netdev_stats; 10589 u64 *dst = (u64 *)stats64; 10590 10591 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); 10592 for (i = 0; i < n; i++) 10593 dst[i] = (unsigned long)atomic_long_read(&src[i]); 10594 /* zero out counters that only exist in rtnl_link_stats64 */ 10595 memset((char *)stats64 + n * sizeof(u64), 0, 10596 sizeof(*stats64) - n * sizeof(u64)); 10597 } 10598 EXPORT_SYMBOL(netdev_stats_to_stats64); 10599 10600 static __cold struct net_device_core_stats __percpu *netdev_core_stats_alloc( 10601 struct net_device *dev) 10602 { 10603 struct net_device_core_stats __percpu *p; 10604 10605 p = alloc_percpu_gfp(struct net_device_core_stats, 10606 GFP_ATOMIC | __GFP_NOWARN); 10607 10608 if (p && cmpxchg(&dev->core_stats, NULL, p)) 10609 free_percpu(p); 10610 10611 /* This READ_ONCE() pairs with the cmpxchg() above */ 10612 return READ_ONCE(dev->core_stats); 10613 } 10614 10615 noinline void netdev_core_stats_inc(struct net_device *dev, u32 offset) 10616 { 10617 /* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */ 10618 struct net_device_core_stats __percpu *p = READ_ONCE(dev->core_stats); 10619 unsigned long __percpu *field; 10620 10621 if (unlikely(!p)) { 10622 p = netdev_core_stats_alloc(dev); 10623 if (!p) 10624 return; 10625 } 10626 10627 field = (__force unsigned long __percpu *)((__force void *)p + offset); 10628 this_cpu_inc(*field); 10629 } 10630 EXPORT_SYMBOL_GPL(netdev_core_stats_inc); 10631 10632 /** 10633 * dev_get_stats - get network device statistics 10634 * @dev: device to get statistics from 10635 * @storage: place to store stats 10636 * 10637 * Get network statistics from device. Return @storage. 10638 * The device driver may provide its own method by setting 10639 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; 10640 * otherwise the internal statistics structure is used. 10641 */ 10642 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, 10643 struct rtnl_link_stats64 *storage) 10644 { 10645 const struct net_device_ops *ops = dev->netdev_ops; 10646 const struct net_device_core_stats __percpu *p; 10647 10648 if (ops->ndo_get_stats64) { 10649 memset(storage, 0, sizeof(*storage)); 10650 ops->ndo_get_stats64(dev, storage); 10651 } else if (ops->ndo_get_stats) { 10652 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); 10653 } else { 10654 netdev_stats_to_stats64(storage, &dev->stats); 10655 } 10656 10657 /* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */ 10658 p = READ_ONCE(dev->core_stats); 10659 if (p) { 10660 const struct net_device_core_stats *core_stats; 10661 int i; 10662 10663 for_each_possible_cpu(i) { 10664 core_stats = per_cpu_ptr(p, i); 10665 storage->rx_dropped += READ_ONCE(core_stats->rx_dropped); 10666 storage->tx_dropped += READ_ONCE(core_stats->tx_dropped); 10667 storage->rx_nohandler += READ_ONCE(core_stats->rx_nohandler); 10668 storage->rx_otherhost_dropped += READ_ONCE(core_stats->rx_otherhost_dropped); 10669 } 10670 } 10671 return storage; 10672 } 10673 EXPORT_SYMBOL(dev_get_stats); 10674 10675 /** 10676 * dev_fetch_sw_netstats - get per-cpu network device statistics 10677 * @s: place to store stats 10678 * @netstats: per-cpu network stats to read from 10679 * 10680 * Read per-cpu network statistics and populate the related fields in @s. 10681 */ 10682 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s, 10683 const struct pcpu_sw_netstats __percpu *netstats) 10684 { 10685 int cpu; 10686 10687 for_each_possible_cpu(cpu) { 10688 u64 rx_packets, rx_bytes, tx_packets, tx_bytes; 10689 const struct pcpu_sw_netstats *stats; 10690 unsigned int start; 10691 10692 stats = per_cpu_ptr(netstats, cpu); 10693 do { 10694 start = u64_stats_fetch_begin(&stats->syncp); 10695 rx_packets = u64_stats_read(&stats->rx_packets); 10696 rx_bytes = u64_stats_read(&stats->rx_bytes); 10697 tx_packets = u64_stats_read(&stats->tx_packets); 10698 tx_bytes = u64_stats_read(&stats->tx_bytes); 10699 } while (u64_stats_fetch_retry(&stats->syncp, start)); 10700 10701 s->rx_packets += rx_packets; 10702 s->rx_bytes += rx_bytes; 10703 s->tx_packets += tx_packets; 10704 s->tx_bytes += tx_bytes; 10705 } 10706 } 10707 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats); 10708 10709 /** 10710 * dev_get_tstats64 - ndo_get_stats64 implementation 10711 * @dev: device to get statistics from 10712 * @s: place to store stats 10713 * 10714 * Populate @s from dev->stats and dev->tstats. Can be used as 10715 * ndo_get_stats64() callback. 10716 */ 10717 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s) 10718 { 10719 netdev_stats_to_stats64(s, &dev->stats); 10720 dev_fetch_sw_netstats(s, dev->tstats); 10721 } 10722 EXPORT_SYMBOL_GPL(dev_get_tstats64); 10723 10724 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) 10725 { 10726 struct netdev_queue *queue = dev_ingress_queue(dev); 10727 10728 #ifdef CONFIG_NET_CLS_ACT 10729 if (queue) 10730 return queue; 10731 queue = kzalloc(sizeof(*queue), GFP_KERNEL); 10732 if (!queue) 10733 return NULL; 10734 netdev_init_one_queue(dev, queue, NULL); 10735 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); 10736 RCU_INIT_POINTER(queue->qdisc_sleeping, &noop_qdisc); 10737 rcu_assign_pointer(dev->ingress_queue, queue); 10738 #endif 10739 return queue; 10740 } 10741 10742 static const struct ethtool_ops default_ethtool_ops; 10743 10744 void netdev_set_default_ethtool_ops(struct net_device *dev, 10745 const struct ethtool_ops *ops) 10746 { 10747 if (dev->ethtool_ops == &default_ethtool_ops) 10748 dev->ethtool_ops = ops; 10749 } 10750 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); 10751 10752 /** 10753 * netdev_sw_irq_coalesce_default_on() - enable SW IRQ coalescing by default 10754 * @dev: netdev to enable the IRQ coalescing on 10755 * 10756 * Sets a conservative default for SW IRQ coalescing. Users can use 10757 * sysfs attributes to override the default values. 10758 */ 10759 void netdev_sw_irq_coalesce_default_on(struct net_device *dev) 10760 { 10761 WARN_ON(dev->reg_state == NETREG_REGISTERED); 10762 10763 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) { 10764 dev->gro_flush_timeout = 20000; 10765 dev->napi_defer_hard_irqs = 1; 10766 } 10767 } 10768 EXPORT_SYMBOL_GPL(netdev_sw_irq_coalesce_default_on); 10769 10770 void netdev_freemem(struct net_device *dev) 10771 { 10772 char *addr = (char *)dev - dev->padded; 10773 10774 kvfree(addr); 10775 } 10776 10777 /** 10778 * alloc_netdev_mqs - allocate network device 10779 * @sizeof_priv: size of private data to allocate space for 10780 * @name: device name format string 10781 * @name_assign_type: origin of device name 10782 * @setup: callback to initialize device 10783 * @txqs: the number of TX subqueues to allocate 10784 * @rxqs: the number of RX subqueues to allocate 10785 * 10786 * Allocates a struct net_device with private data area for driver use 10787 * and performs basic initialization. Also allocates subqueue structs 10788 * for each queue on the device. 10789 */ 10790 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, 10791 unsigned char name_assign_type, 10792 void (*setup)(struct net_device *), 10793 unsigned int txqs, unsigned int rxqs) 10794 { 10795 struct net_device *dev; 10796 unsigned int alloc_size; 10797 struct net_device *p; 10798 10799 BUG_ON(strlen(name) >= sizeof(dev->name)); 10800 10801 if (txqs < 1) { 10802 pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); 10803 return NULL; 10804 } 10805 10806 if (rxqs < 1) { 10807 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); 10808 return NULL; 10809 } 10810 10811 alloc_size = sizeof(struct net_device); 10812 if (sizeof_priv) { 10813 /* ensure 32-byte alignment of private area */ 10814 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); 10815 alloc_size += sizeof_priv; 10816 } 10817 /* ensure 32-byte alignment of whole construct */ 10818 alloc_size += NETDEV_ALIGN - 1; 10819 10820 p = kvzalloc(alloc_size, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10821 if (!p) 10822 return NULL; 10823 10824 dev = PTR_ALIGN(p, NETDEV_ALIGN); 10825 dev->padded = (char *)dev - (char *)p; 10826 10827 ref_tracker_dir_init(&dev->refcnt_tracker, 128, name); 10828 #ifdef CONFIG_PCPU_DEV_REFCNT 10829 dev->pcpu_refcnt = alloc_percpu(int); 10830 if (!dev->pcpu_refcnt) 10831 goto free_dev; 10832 __dev_hold(dev); 10833 #else 10834 refcount_set(&dev->dev_refcnt, 1); 10835 #endif 10836 10837 if (dev_addr_init(dev)) 10838 goto free_pcpu; 10839 10840 dev_mc_init(dev); 10841 dev_uc_init(dev); 10842 10843 dev_net_set(dev, &init_net); 10844 10845 dev->gso_max_size = GSO_LEGACY_MAX_SIZE; 10846 dev->xdp_zc_max_segs = 1; 10847 dev->gso_max_segs = GSO_MAX_SEGS; 10848 dev->gro_max_size = GRO_LEGACY_MAX_SIZE; 10849 dev->gso_ipv4_max_size = GSO_LEGACY_MAX_SIZE; 10850 dev->gro_ipv4_max_size = GRO_LEGACY_MAX_SIZE; 10851 dev->tso_max_size = TSO_LEGACY_MAX_SIZE; 10852 dev->tso_max_segs = TSO_MAX_SEGS; 10853 dev->upper_level = 1; 10854 dev->lower_level = 1; 10855 #ifdef CONFIG_LOCKDEP 10856 dev->nested_level = 0; 10857 INIT_LIST_HEAD(&dev->unlink_list); 10858 #endif 10859 10860 INIT_LIST_HEAD(&dev->napi_list); 10861 INIT_LIST_HEAD(&dev->unreg_list); 10862 INIT_LIST_HEAD(&dev->close_list); 10863 INIT_LIST_HEAD(&dev->link_watch_list); 10864 INIT_LIST_HEAD(&dev->adj_list.upper); 10865 INIT_LIST_HEAD(&dev->adj_list.lower); 10866 INIT_LIST_HEAD(&dev->ptype_all); 10867 INIT_LIST_HEAD(&dev->ptype_specific); 10868 INIT_LIST_HEAD(&dev->net_notifier_list); 10869 #ifdef CONFIG_NET_SCHED 10870 hash_init(dev->qdisc_hash); 10871 #endif 10872 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; 10873 setup(dev); 10874 10875 if (!dev->tx_queue_len) { 10876 dev->priv_flags |= IFF_NO_QUEUE; 10877 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; 10878 } 10879 10880 dev->num_tx_queues = txqs; 10881 dev->real_num_tx_queues = txqs; 10882 if (netif_alloc_netdev_queues(dev)) 10883 goto free_all; 10884 10885 dev->num_rx_queues = rxqs; 10886 dev->real_num_rx_queues = rxqs; 10887 if (netif_alloc_rx_queues(dev)) 10888 goto free_all; 10889 10890 strcpy(dev->name, name); 10891 dev->name_assign_type = name_assign_type; 10892 dev->group = INIT_NETDEV_GROUP; 10893 if (!dev->ethtool_ops) 10894 dev->ethtool_ops = &default_ethtool_ops; 10895 10896 nf_hook_netdev_init(dev); 10897 10898 return dev; 10899 10900 free_all: 10901 free_netdev(dev); 10902 return NULL; 10903 10904 free_pcpu: 10905 #ifdef CONFIG_PCPU_DEV_REFCNT 10906 free_percpu(dev->pcpu_refcnt); 10907 free_dev: 10908 #endif 10909 netdev_freemem(dev); 10910 return NULL; 10911 } 10912 EXPORT_SYMBOL(alloc_netdev_mqs); 10913 10914 /** 10915 * free_netdev - free network device 10916 * @dev: device 10917 * 10918 * This function does the last stage of destroying an allocated device 10919 * interface. The reference to the device object is released. If this 10920 * is the last reference then it will be freed.Must be called in process 10921 * context. 10922 */ 10923 void free_netdev(struct net_device *dev) 10924 { 10925 struct napi_struct *p, *n; 10926 10927 might_sleep(); 10928 10929 /* When called immediately after register_netdevice() failed the unwind 10930 * handling may still be dismantling the device. Handle that case by 10931 * deferring the free. 10932 */ 10933 if (dev->reg_state == NETREG_UNREGISTERING) { 10934 ASSERT_RTNL(); 10935 dev->needs_free_netdev = true; 10936 return; 10937 } 10938 10939 netif_free_tx_queues(dev); 10940 netif_free_rx_queues(dev); 10941 10942 kfree(rcu_dereference_protected(dev->ingress_queue, 1)); 10943 10944 /* Flush device addresses */ 10945 dev_addr_flush(dev); 10946 10947 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) 10948 netif_napi_del(p); 10949 10950 ref_tracker_dir_exit(&dev->refcnt_tracker); 10951 #ifdef CONFIG_PCPU_DEV_REFCNT 10952 free_percpu(dev->pcpu_refcnt); 10953 dev->pcpu_refcnt = NULL; 10954 #endif 10955 free_percpu(dev->core_stats); 10956 dev->core_stats = NULL; 10957 free_percpu(dev->xdp_bulkq); 10958 dev->xdp_bulkq = NULL; 10959 10960 /* Compatibility with error handling in drivers */ 10961 if (dev->reg_state == NETREG_UNINITIALIZED) { 10962 netdev_freemem(dev); 10963 return; 10964 } 10965 10966 BUG_ON(dev->reg_state != NETREG_UNREGISTERED); 10967 dev->reg_state = NETREG_RELEASED; 10968 10969 /* will free via device release */ 10970 put_device(&dev->dev); 10971 } 10972 EXPORT_SYMBOL(free_netdev); 10973 10974 /** 10975 * synchronize_net - Synchronize with packet receive processing 10976 * 10977 * Wait for packets currently being received to be done. 10978 * Does not block later packets from starting. 10979 */ 10980 void synchronize_net(void) 10981 { 10982 might_sleep(); 10983 if (rtnl_is_locked()) 10984 synchronize_rcu_expedited(); 10985 else 10986 synchronize_rcu(); 10987 } 10988 EXPORT_SYMBOL(synchronize_net); 10989 10990 /** 10991 * unregister_netdevice_queue - remove device from the kernel 10992 * @dev: device 10993 * @head: list 10994 * 10995 * This function shuts down a device interface and removes it 10996 * from the kernel tables. 10997 * If head not NULL, device is queued to be unregistered later. 10998 * 10999 * Callers must hold the rtnl semaphore. You may want 11000 * unregister_netdev() instead of this. 11001 */ 11002 11003 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) 11004 { 11005 ASSERT_RTNL(); 11006 11007 if (head) { 11008 list_move_tail(&dev->unreg_list, head); 11009 } else { 11010 LIST_HEAD(single); 11011 11012 list_add(&dev->unreg_list, &single); 11013 unregister_netdevice_many(&single); 11014 } 11015 } 11016 EXPORT_SYMBOL(unregister_netdevice_queue); 11017 11018 void unregister_netdevice_many_notify(struct list_head *head, 11019 u32 portid, const struct nlmsghdr *nlh) 11020 { 11021 struct net_device *dev, *tmp; 11022 LIST_HEAD(close_head); 11023 11024 BUG_ON(dev_boot_phase); 11025 ASSERT_RTNL(); 11026 11027 if (list_empty(head)) 11028 return; 11029 11030 list_for_each_entry_safe(dev, tmp, head, unreg_list) { 11031 /* Some devices call without registering 11032 * for initialization unwind. Remove those 11033 * devices and proceed with the remaining. 11034 */ 11035 if (dev->reg_state == NETREG_UNINITIALIZED) { 11036 pr_debug("unregister_netdevice: device %s/%p never was registered\n", 11037 dev->name, dev); 11038 11039 WARN_ON(1); 11040 list_del(&dev->unreg_list); 11041 continue; 11042 } 11043 dev->dismantle = true; 11044 BUG_ON(dev->reg_state != NETREG_REGISTERED); 11045 } 11046 11047 /* If device is running, close it first. */ 11048 list_for_each_entry(dev, head, unreg_list) 11049 list_add_tail(&dev->close_list, &close_head); 11050 dev_close_many(&close_head, true); 11051 11052 list_for_each_entry(dev, head, unreg_list) { 11053 /* And unlink it from device chain. */ 11054 write_lock(&dev_base_lock); 11055 unlist_netdevice(dev, false); 11056 dev->reg_state = NETREG_UNREGISTERING; 11057 write_unlock(&dev_base_lock); 11058 } 11059 flush_all_backlogs(); 11060 11061 synchronize_net(); 11062 11063 list_for_each_entry(dev, head, unreg_list) { 11064 struct sk_buff *skb = NULL; 11065 11066 /* Shutdown queueing discipline. */ 11067 dev_shutdown(dev); 11068 dev_tcx_uninstall(dev); 11069 dev_xdp_uninstall(dev); 11070 bpf_dev_bound_netdev_unregister(dev); 11071 11072 netdev_offload_xstats_disable_all(dev); 11073 11074 /* Notify protocols, that we are about to destroy 11075 * this device. They should clean all the things. 11076 */ 11077 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 11078 11079 if (!dev->rtnl_link_ops || 11080 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 11081 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0, 11082 GFP_KERNEL, NULL, 0, 11083 portid, nlh); 11084 11085 /* 11086 * Flush the unicast and multicast chains 11087 */ 11088 dev_uc_flush(dev); 11089 dev_mc_flush(dev); 11090 11091 netdev_name_node_alt_flush(dev); 11092 netdev_name_node_free(dev->name_node); 11093 11094 call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev); 11095 11096 if (dev->netdev_ops->ndo_uninit) 11097 dev->netdev_ops->ndo_uninit(dev); 11098 11099 if (skb) 11100 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL, portid, nlh); 11101 11102 /* Notifier chain MUST detach us all upper devices. */ 11103 WARN_ON(netdev_has_any_upper_dev(dev)); 11104 WARN_ON(netdev_has_any_lower_dev(dev)); 11105 11106 /* Remove entries from kobject tree */ 11107 netdev_unregister_kobject(dev); 11108 #ifdef CONFIG_XPS 11109 /* Remove XPS queueing entries */ 11110 netif_reset_xps_queues_gt(dev, 0); 11111 #endif 11112 } 11113 11114 synchronize_net(); 11115 11116 list_for_each_entry(dev, head, unreg_list) { 11117 netdev_put(dev, &dev->dev_registered_tracker); 11118 net_set_todo(dev); 11119 } 11120 11121 list_del(head); 11122 } 11123 11124 /** 11125 * unregister_netdevice_many - unregister many devices 11126 * @head: list of devices 11127 * 11128 * Note: As most callers use a stack allocated list_head, 11129 * we force a list_del() to make sure stack wont be corrupted later. 11130 */ 11131 void unregister_netdevice_many(struct list_head *head) 11132 { 11133 unregister_netdevice_many_notify(head, 0, NULL); 11134 } 11135 EXPORT_SYMBOL(unregister_netdevice_many); 11136 11137 /** 11138 * unregister_netdev - remove device from the kernel 11139 * @dev: device 11140 * 11141 * This function shuts down a device interface and removes it 11142 * from the kernel tables. 11143 * 11144 * This is just a wrapper for unregister_netdevice that takes 11145 * the rtnl semaphore. In general you want to use this and not 11146 * unregister_netdevice. 11147 */ 11148 void unregister_netdev(struct net_device *dev) 11149 { 11150 rtnl_lock(); 11151 unregister_netdevice(dev); 11152 rtnl_unlock(); 11153 } 11154 EXPORT_SYMBOL(unregister_netdev); 11155 11156 /** 11157 * __dev_change_net_namespace - move device to different nethost namespace 11158 * @dev: device 11159 * @net: network namespace 11160 * @pat: If not NULL name pattern to try if the current device name 11161 * is already taken in the destination network namespace. 11162 * @new_ifindex: If not zero, specifies device index in the target 11163 * namespace. 11164 * 11165 * This function shuts down a device interface and moves it 11166 * to a new network namespace. On success 0 is returned, on 11167 * a failure a netagive errno code is returned. 11168 * 11169 * Callers must hold the rtnl semaphore. 11170 */ 11171 11172 int __dev_change_net_namespace(struct net_device *dev, struct net *net, 11173 const char *pat, int new_ifindex) 11174 { 11175 struct netdev_name_node *name_node; 11176 struct net *net_old = dev_net(dev); 11177 char new_name[IFNAMSIZ] = {}; 11178 int err, new_nsid; 11179 11180 ASSERT_RTNL(); 11181 11182 /* Don't allow namespace local devices to be moved. */ 11183 err = -EINVAL; 11184 if (dev->features & NETIF_F_NETNS_LOCAL) 11185 goto out; 11186 11187 /* Ensure the device has been registrered */ 11188 if (dev->reg_state != NETREG_REGISTERED) 11189 goto out; 11190 11191 /* Get out if there is nothing todo */ 11192 err = 0; 11193 if (net_eq(net_old, net)) 11194 goto out; 11195 11196 /* Pick the destination device name, and ensure 11197 * we can use it in the destination network namespace. 11198 */ 11199 err = -EEXIST; 11200 if (netdev_name_in_use(net, dev->name)) { 11201 /* We get here if we can't use the current device name */ 11202 if (!pat) 11203 goto out; 11204 err = dev_prep_valid_name(net, dev, pat, new_name, EEXIST); 11205 if (err < 0) 11206 goto out; 11207 } 11208 /* Check that none of the altnames conflicts. */ 11209 err = -EEXIST; 11210 netdev_for_each_altname(dev, name_node) 11211 if (netdev_name_in_use(net, name_node->name)) 11212 goto out; 11213 11214 /* Check that new_ifindex isn't used yet. */ 11215 if (new_ifindex) { 11216 err = dev_index_reserve(net, new_ifindex); 11217 if (err < 0) 11218 goto out; 11219 } else { 11220 /* If there is an ifindex conflict assign a new one */ 11221 err = dev_index_reserve(net, dev->ifindex); 11222 if (err == -EBUSY) 11223 err = dev_index_reserve(net, 0); 11224 if (err < 0) 11225 goto out; 11226 new_ifindex = err; 11227 } 11228 11229 /* 11230 * And now a mini version of register_netdevice unregister_netdevice. 11231 */ 11232 11233 /* If device is running close it first. */ 11234 dev_close(dev); 11235 11236 /* And unlink it from device chain */ 11237 unlist_netdevice(dev, true); 11238 11239 synchronize_net(); 11240 11241 /* Shutdown queueing discipline. */ 11242 dev_shutdown(dev); 11243 11244 /* Notify protocols, that we are about to destroy 11245 * this device. They should clean all the things. 11246 * 11247 * Note that dev->reg_state stays at NETREG_REGISTERED. 11248 * This is wanted because this way 8021q and macvlan know 11249 * the device is just moving and can keep their slaves up. 11250 */ 11251 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 11252 rcu_barrier(); 11253 11254 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL); 11255 11256 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid, 11257 new_ifindex); 11258 11259 /* 11260 * Flush the unicast and multicast chains 11261 */ 11262 dev_uc_flush(dev); 11263 dev_mc_flush(dev); 11264 11265 /* Send a netdev-removed uevent to the old namespace */ 11266 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); 11267 netdev_adjacent_del_links(dev); 11268 11269 /* Move per-net netdevice notifiers that are following the netdevice */ 11270 move_netdevice_notifiers_dev_net(dev, net); 11271 11272 /* Actually switch the network namespace */ 11273 dev_net_set(dev, net); 11274 dev->ifindex = new_ifindex; 11275 11276 if (new_name[0]) /* Rename the netdev to prepared name */ 11277 strscpy(dev->name, new_name, IFNAMSIZ); 11278 11279 /* Fixup kobjects */ 11280 dev_set_uevent_suppress(&dev->dev, 1); 11281 err = device_rename(&dev->dev, dev->name); 11282 dev_set_uevent_suppress(&dev->dev, 0); 11283 WARN_ON(err); 11284 11285 /* Send a netdev-add uevent to the new namespace */ 11286 kobject_uevent(&dev->dev.kobj, KOBJ_ADD); 11287 netdev_adjacent_add_links(dev); 11288 11289 /* Adapt owner in case owning user namespace of target network 11290 * namespace is different from the original one. 11291 */ 11292 err = netdev_change_owner(dev, net_old, net); 11293 WARN_ON(err); 11294 11295 /* Add the device back in the hashes */ 11296 list_netdevice(dev); 11297 11298 /* Notify protocols, that a new device appeared. */ 11299 call_netdevice_notifiers(NETDEV_REGISTER, dev); 11300 11301 /* 11302 * Prevent userspace races by waiting until the network 11303 * device is fully setup before sending notifications. 11304 */ 11305 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL); 11306 11307 synchronize_net(); 11308 err = 0; 11309 out: 11310 return err; 11311 } 11312 EXPORT_SYMBOL_GPL(__dev_change_net_namespace); 11313 11314 static int dev_cpu_dead(unsigned int oldcpu) 11315 { 11316 struct sk_buff **list_skb; 11317 struct sk_buff *skb; 11318 unsigned int cpu; 11319 struct softnet_data *sd, *oldsd, *remsd = NULL; 11320 11321 local_irq_disable(); 11322 cpu = smp_processor_id(); 11323 sd = &per_cpu(softnet_data, cpu); 11324 oldsd = &per_cpu(softnet_data, oldcpu); 11325 11326 /* Find end of our completion_queue. */ 11327 list_skb = &sd->completion_queue; 11328 while (*list_skb) 11329 list_skb = &(*list_skb)->next; 11330 /* Append completion queue from offline CPU. */ 11331 *list_skb = oldsd->completion_queue; 11332 oldsd->completion_queue = NULL; 11333 11334 /* Append output queue from offline CPU. */ 11335 if (oldsd->output_queue) { 11336 *sd->output_queue_tailp = oldsd->output_queue; 11337 sd->output_queue_tailp = oldsd->output_queue_tailp; 11338 oldsd->output_queue = NULL; 11339 oldsd->output_queue_tailp = &oldsd->output_queue; 11340 } 11341 /* Append NAPI poll list from offline CPU, with one exception : 11342 * process_backlog() must be called by cpu owning percpu backlog. 11343 * We properly handle process_queue & input_pkt_queue later. 11344 */ 11345 while (!list_empty(&oldsd->poll_list)) { 11346 struct napi_struct *napi = list_first_entry(&oldsd->poll_list, 11347 struct napi_struct, 11348 poll_list); 11349 11350 list_del_init(&napi->poll_list); 11351 if (napi->poll == process_backlog) 11352 napi->state = 0; 11353 else 11354 ____napi_schedule(sd, napi); 11355 } 11356 11357 raise_softirq_irqoff(NET_TX_SOFTIRQ); 11358 local_irq_enable(); 11359 11360 #ifdef CONFIG_RPS 11361 remsd = oldsd->rps_ipi_list; 11362 oldsd->rps_ipi_list = NULL; 11363 #endif 11364 /* send out pending IPI's on offline CPU */ 11365 net_rps_send_ipi(remsd); 11366 11367 /* Process offline CPU's input_pkt_queue */ 11368 while ((skb = __skb_dequeue(&oldsd->process_queue))) { 11369 netif_rx(skb); 11370 input_queue_head_incr(oldsd); 11371 } 11372 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { 11373 netif_rx(skb); 11374 input_queue_head_incr(oldsd); 11375 } 11376 11377 return 0; 11378 } 11379 11380 /** 11381 * netdev_increment_features - increment feature set by one 11382 * @all: current feature set 11383 * @one: new feature set 11384 * @mask: mask feature set 11385 * 11386 * Computes a new feature set after adding a device with feature set 11387 * @one to the master device with current feature set @all. Will not 11388 * enable anything that is off in @mask. Returns the new feature set. 11389 */ 11390 netdev_features_t netdev_increment_features(netdev_features_t all, 11391 netdev_features_t one, netdev_features_t mask) 11392 { 11393 if (mask & NETIF_F_HW_CSUM) 11394 mask |= NETIF_F_CSUM_MASK; 11395 mask |= NETIF_F_VLAN_CHALLENGED; 11396 11397 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; 11398 all &= one | ~NETIF_F_ALL_FOR_ALL; 11399 11400 /* If one device supports hw checksumming, set for all. */ 11401 if (all & NETIF_F_HW_CSUM) 11402 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); 11403 11404 return all; 11405 } 11406 EXPORT_SYMBOL(netdev_increment_features); 11407 11408 static struct hlist_head * __net_init netdev_create_hash(void) 11409 { 11410 int i; 11411 struct hlist_head *hash; 11412 11413 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL); 11414 if (hash != NULL) 11415 for (i = 0; i < NETDEV_HASHENTRIES; i++) 11416 INIT_HLIST_HEAD(&hash[i]); 11417 11418 return hash; 11419 } 11420 11421 /* Initialize per network namespace state */ 11422 static int __net_init netdev_init(struct net *net) 11423 { 11424 BUILD_BUG_ON(GRO_HASH_BUCKETS > 11425 8 * sizeof_field(struct napi_struct, gro_bitmask)); 11426 11427 INIT_LIST_HEAD(&net->dev_base_head); 11428 11429 net->dev_name_head = netdev_create_hash(); 11430 if (net->dev_name_head == NULL) 11431 goto err_name; 11432 11433 net->dev_index_head = netdev_create_hash(); 11434 if (net->dev_index_head == NULL) 11435 goto err_idx; 11436 11437 xa_init_flags(&net->dev_by_index, XA_FLAGS_ALLOC1); 11438 11439 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain); 11440 11441 return 0; 11442 11443 err_idx: 11444 kfree(net->dev_name_head); 11445 err_name: 11446 return -ENOMEM; 11447 } 11448 11449 /** 11450 * netdev_drivername - network driver for the device 11451 * @dev: network device 11452 * 11453 * Determine network driver for device. 11454 */ 11455 const char *netdev_drivername(const struct net_device *dev) 11456 { 11457 const struct device_driver *driver; 11458 const struct device *parent; 11459 const char *empty = ""; 11460 11461 parent = dev->dev.parent; 11462 if (!parent) 11463 return empty; 11464 11465 driver = parent->driver; 11466 if (driver && driver->name) 11467 return driver->name; 11468 return empty; 11469 } 11470 11471 static void __netdev_printk(const char *level, const struct net_device *dev, 11472 struct va_format *vaf) 11473 { 11474 if (dev && dev->dev.parent) { 11475 dev_printk_emit(level[1] - '0', 11476 dev->dev.parent, 11477 "%s %s %s%s: %pV", 11478 dev_driver_string(dev->dev.parent), 11479 dev_name(dev->dev.parent), 11480 netdev_name(dev), netdev_reg_state(dev), 11481 vaf); 11482 } else if (dev) { 11483 printk("%s%s%s: %pV", 11484 level, netdev_name(dev), netdev_reg_state(dev), vaf); 11485 } else { 11486 printk("%s(NULL net_device): %pV", level, vaf); 11487 } 11488 } 11489 11490 void netdev_printk(const char *level, const struct net_device *dev, 11491 const char *format, ...) 11492 { 11493 struct va_format vaf; 11494 va_list args; 11495 11496 va_start(args, format); 11497 11498 vaf.fmt = format; 11499 vaf.va = &args; 11500 11501 __netdev_printk(level, dev, &vaf); 11502 11503 va_end(args); 11504 } 11505 EXPORT_SYMBOL(netdev_printk); 11506 11507 #define define_netdev_printk_level(func, level) \ 11508 void func(const struct net_device *dev, const char *fmt, ...) \ 11509 { \ 11510 struct va_format vaf; \ 11511 va_list args; \ 11512 \ 11513 va_start(args, fmt); \ 11514 \ 11515 vaf.fmt = fmt; \ 11516 vaf.va = &args; \ 11517 \ 11518 __netdev_printk(level, dev, &vaf); \ 11519 \ 11520 va_end(args); \ 11521 } \ 11522 EXPORT_SYMBOL(func); 11523 11524 define_netdev_printk_level(netdev_emerg, KERN_EMERG); 11525 define_netdev_printk_level(netdev_alert, KERN_ALERT); 11526 define_netdev_printk_level(netdev_crit, KERN_CRIT); 11527 define_netdev_printk_level(netdev_err, KERN_ERR); 11528 define_netdev_printk_level(netdev_warn, KERN_WARNING); 11529 define_netdev_printk_level(netdev_notice, KERN_NOTICE); 11530 define_netdev_printk_level(netdev_info, KERN_INFO); 11531 11532 static void __net_exit netdev_exit(struct net *net) 11533 { 11534 kfree(net->dev_name_head); 11535 kfree(net->dev_index_head); 11536 xa_destroy(&net->dev_by_index); 11537 if (net != &init_net) 11538 WARN_ON_ONCE(!list_empty(&net->dev_base_head)); 11539 } 11540 11541 static struct pernet_operations __net_initdata netdev_net_ops = { 11542 .init = netdev_init, 11543 .exit = netdev_exit, 11544 }; 11545 11546 static void __net_exit default_device_exit_net(struct net *net) 11547 { 11548 struct net_device *dev, *aux; 11549 /* 11550 * Push all migratable network devices back to the 11551 * initial network namespace 11552 */ 11553 ASSERT_RTNL(); 11554 for_each_netdev_safe(net, dev, aux) { 11555 int err; 11556 char fb_name[IFNAMSIZ]; 11557 11558 /* Ignore unmoveable devices (i.e. loopback) */ 11559 if (dev->features & NETIF_F_NETNS_LOCAL) 11560 continue; 11561 11562 /* Leave virtual devices for the generic cleanup */ 11563 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund) 11564 continue; 11565 11566 /* Push remaining network devices to init_net */ 11567 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); 11568 if (netdev_name_in_use(&init_net, fb_name)) 11569 snprintf(fb_name, IFNAMSIZ, "dev%%d"); 11570 err = dev_change_net_namespace(dev, &init_net, fb_name); 11571 if (err) { 11572 pr_emerg("%s: failed to move %s to init_net: %d\n", 11573 __func__, dev->name, err); 11574 BUG(); 11575 } 11576 } 11577 } 11578 11579 static void __net_exit default_device_exit_batch(struct list_head *net_list) 11580 { 11581 /* At exit all network devices most be removed from a network 11582 * namespace. Do this in the reverse order of registration. 11583 * Do this across as many network namespaces as possible to 11584 * improve batching efficiency. 11585 */ 11586 struct net_device *dev; 11587 struct net *net; 11588 LIST_HEAD(dev_kill_list); 11589 11590 rtnl_lock(); 11591 list_for_each_entry(net, net_list, exit_list) { 11592 default_device_exit_net(net); 11593 cond_resched(); 11594 } 11595 11596 list_for_each_entry(net, net_list, exit_list) { 11597 for_each_netdev_reverse(net, dev) { 11598 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) 11599 dev->rtnl_link_ops->dellink(dev, &dev_kill_list); 11600 else 11601 unregister_netdevice_queue(dev, &dev_kill_list); 11602 } 11603 } 11604 unregister_netdevice_many(&dev_kill_list); 11605 rtnl_unlock(); 11606 } 11607 11608 static struct pernet_operations __net_initdata default_device_ops = { 11609 .exit_batch = default_device_exit_batch, 11610 }; 11611 11612 static void __init net_dev_struct_check(void) 11613 { 11614 /* TX read-mostly hotpath */ 11615 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, priv_flags); 11616 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, netdev_ops); 11617 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, header_ops); 11618 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, _tx); 11619 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, real_num_tx_queues); 11620 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_max_size); 11621 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_ipv4_max_size); 11622 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_max_segs); 11623 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, num_tc); 11624 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, mtu); 11625 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, needed_headroom); 11626 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, tc_to_txq); 11627 #ifdef CONFIG_XPS 11628 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, xps_maps); 11629 #endif 11630 #ifdef CONFIG_NETFILTER_EGRESS 11631 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, nf_hooks_egress); 11632 #endif 11633 #ifdef CONFIG_NET_XGRESS 11634 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, tcx_egress); 11635 #endif 11636 CACHELINE_ASSERT_GROUP_SIZE(struct net_device, net_device_read_tx, 152); 11637 11638 /* TXRX read-mostly hotpath */ 11639 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, flags); 11640 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, hard_header_len); 11641 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, features); 11642 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, ip6_ptr); 11643 CACHELINE_ASSERT_GROUP_SIZE(struct net_device, net_device_read_txrx, 30); 11644 11645 /* RX read-mostly hotpath */ 11646 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, ptype_specific); 11647 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, ifindex); 11648 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, real_num_rx_queues); 11649 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, _rx); 11650 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, gro_flush_timeout); 11651 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, napi_defer_hard_irqs); 11652 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, gro_max_size); 11653 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, gro_ipv4_max_size); 11654 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, rx_handler); 11655 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, rx_handler_data); 11656 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, nd_net); 11657 #ifdef CONFIG_NETPOLL 11658 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, npinfo); 11659 #endif 11660 #ifdef CONFIG_NET_XGRESS 11661 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, tcx_ingress); 11662 #endif 11663 CACHELINE_ASSERT_GROUP_SIZE(struct net_device, net_device_read_rx, 96); 11664 } 11665 11666 /* 11667 * Initialize the DEV module. At boot time this walks the device list and 11668 * unhooks any devices that fail to initialise (normally hardware not 11669 * present) and leaves us with a valid list of present and active devices. 11670 * 11671 */ 11672 11673 /* 11674 * This is called single threaded during boot, so no need 11675 * to take the rtnl semaphore. 11676 */ 11677 static int __init net_dev_init(void) 11678 { 11679 int i, rc = -ENOMEM; 11680 11681 BUG_ON(!dev_boot_phase); 11682 11683 net_dev_struct_check(); 11684 11685 if (dev_proc_init()) 11686 goto out; 11687 11688 if (netdev_kobject_init()) 11689 goto out; 11690 11691 INIT_LIST_HEAD(&ptype_all); 11692 for (i = 0; i < PTYPE_HASH_SIZE; i++) 11693 INIT_LIST_HEAD(&ptype_base[i]); 11694 11695 if (register_pernet_subsys(&netdev_net_ops)) 11696 goto out; 11697 11698 /* 11699 * Initialise the packet receive queues. 11700 */ 11701 11702 for_each_possible_cpu(i) { 11703 struct work_struct *flush = per_cpu_ptr(&flush_works, i); 11704 struct softnet_data *sd = &per_cpu(softnet_data, i); 11705 11706 INIT_WORK(flush, flush_backlog); 11707 11708 skb_queue_head_init(&sd->input_pkt_queue); 11709 skb_queue_head_init(&sd->process_queue); 11710 #ifdef CONFIG_XFRM_OFFLOAD 11711 skb_queue_head_init(&sd->xfrm_backlog); 11712 #endif 11713 INIT_LIST_HEAD(&sd->poll_list); 11714 sd->output_queue_tailp = &sd->output_queue; 11715 #ifdef CONFIG_RPS 11716 INIT_CSD(&sd->csd, rps_trigger_softirq, sd); 11717 sd->cpu = i; 11718 #endif 11719 INIT_CSD(&sd->defer_csd, trigger_rx_softirq, sd); 11720 spin_lock_init(&sd->defer_lock); 11721 11722 init_gro_hash(&sd->backlog); 11723 sd->backlog.poll = process_backlog; 11724 sd->backlog.weight = weight_p; 11725 } 11726 11727 dev_boot_phase = 0; 11728 11729 /* The loopback device is special if any other network devices 11730 * is present in a network namespace the loopback device must 11731 * be present. Since we now dynamically allocate and free the 11732 * loopback device ensure this invariant is maintained by 11733 * keeping the loopback device as the first device on the 11734 * list of network devices. Ensuring the loopback devices 11735 * is the first device that appears and the last network device 11736 * that disappears. 11737 */ 11738 if (register_pernet_device(&loopback_net_ops)) 11739 goto out; 11740 11741 if (register_pernet_device(&default_device_ops)) 11742 goto out; 11743 11744 open_softirq(NET_TX_SOFTIRQ, net_tx_action); 11745 open_softirq(NET_RX_SOFTIRQ, net_rx_action); 11746 11747 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", 11748 NULL, dev_cpu_dead); 11749 WARN_ON(rc < 0); 11750 rc = 0; 11751 out: 11752 return rc; 11753 } 11754 11755 subsys_initcall(net_dev_init); 11756