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 static struct napi_struct *napi_by_id(unsigned int napi_id); 169 170 /* 171 * The @dev_base_head list is protected by @dev_base_lock and the rtnl 172 * semaphore. 173 * 174 * Pure readers hold dev_base_lock for reading, or rcu_read_lock() 175 * 176 * Writers must hold the rtnl semaphore while they loop through the 177 * dev_base_head list, and hold dev_base_lock for writing when they do the 178 * actual updates. This allows pure readers to access the list even 179 * while a writer is preparing to update it. 180 * 181 * To put it another way, dev_base_lock is held for writing only to 182 * protect against pure readers; the rtnl semaphore provides the 183 * protection against other writers. 184 * 185 * See, for example usages, register_netdevice() and 186 * unregister_netdevice(), which must be called with the rtnl 187 * semaphore held. 188 */ 189 DEFINE_RWLOCK(dev_base_lock); 190 EXPORT_SYMBOL(dev_base_lock); 191 192 static DEFINE_MUTEX(ifalias_mutex); 193 194 /* protects napi_hash addition/deletion and napi_gen_id */ 195 static DEFINE_SPINLOCK(napi_hash_lock); 196 197 static unsigned int napi_gen_id = NR_CPUS; 198 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8); 199 200 static DECLARE_RWSEM(devnet_rename_sem); 201 202 static inline void dev_base_seq_inc(struct net *net) 203 { 204 while (++net->dev_base_seq == 0) 205 ; 206 } 207 208 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) 209 { 210 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ)); 211 212 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; 213 } 214 215 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) 216 { 217 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; 218 } 219 220 static inline void rps_lock_irqsave(struct softnet_data *sd, 221 unsigned long *flags) 222 { 223 if (IS_ENABLED(CONFIG_RPS)) 224 spin_lock_irqsave(&sd->input_pkt_queue.lock, *flags); 225 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 226 local_irq_save(*flags); 227 } 228 229 static inline void rps_lock_irq_disable(struct softnet_data *sd) 230 { 231 if (IS_ENABLED(CONFIG_RPS)) 232 spin_lock_irq(&sd->input_pkt_queue.lock); 233 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 234 local_irq_disable(); 235 } 236 237 static inline void rps_unlock_irq_restore(struct softnet_data *sd, 238 unsigned long *flags) 239 { 240 if (IS_ENABLED(CONFIG_RPS)) 241 spin_unlock_irqrestore(&sd->input_pkt_queue.lock, *flags); 242 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 243 local_irq_restore(*flags); 244 } 245 246 static inline void rps_unlock_irq_enable(struct softnet_data *sd) 247 { 248 if (IS_ENABLED(CONFIG_RPS)) 249 spin_unlock_irq(&sd->input_pkt_queue.lock); 250 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 251 local_irq_enable(); 252 } 253 254 static struct netdev_name_node *netdev_name_node_alloc(struct net_device *dev, 255 const char *name) 256 { 257 struct netdev_name_node *name_node; 258 259 name_node = kmalloc(sizeof(*name_node), GFP_KERNEL); 260 if (!name_node) 261 return NULL; 262 INIT_HLIST_NODE(&name_node->hlist); 263 name_node->dev = dev; 264 name_node->name = name; 265 return name_node; 266 } 267 268 static struct netdev_name_node * 269 netdev_name_node_head_alloc(struct net_device *dev) 270 { 271 struct netdev_name_node *name_node; 272 273 name_node = netdev_name_node_alloc(dev, dev->name); 274 if (!name_node) 275 return NULL; 276 INIT_LIST_HEAD(&name_node->list); 277 return name_node; 278 } 279 280 static void netdev_name_node_free(struct netdev_name_node *name_node) 281 { 282 kfree(name_node); 283 } 284 285 static void netdev_name_node_add(struct net *net, 286 struct netdev_name_node *name_node) 287 { 288 hlist_add_head_rcu(&name_node->hlist, 289 dev_name_hash(net, name_node->name)); 290 } 291 292 static void netdev_name_node_del(struct netdev_name_node *name_node) 293 { 294 hlist_del_rcu(&name_node->hlist); 295 } 296 297 static struct netdev_name_node *netdev_name_node_lookup(struct net *net, 298 const char *name) 299 { 300 struct hlist_head *head = dev_name_hash(net, name); 301 struct netdev_name_node *name_node; 302 303 hlist_for_each_entry(name_node, head, hlist) 304 if (!strcmp(name_node->name, name)) 305 return name_node; 306 return NULL; 307 } 308 309 static struct netdev_name_node *netdev_name_node_lookup_rcu(struct net *net, 310 const char *name) 311 { 312 struct hlist_head *head = dev_name_hash(net, name); 313 struct netdev_name_node *name_node; 314 315 hlist_for_each_entry_rcu(name_node, head, hlist) 316 if (!strcmp(name_node->name, name)) 317 return name_node; 318 return NULL; 319 } 320 321 bool netdev_name_in_use(struct net *net, const char *name) 322 { 323 return netdev_name_node_lookup(net, name); 324 } 325 EXPORT_SYMBOL(netdev_name_in_use); 326 327 int netdev_name_node_alt_create(struct net_device *dev, const char *name) 328 { 329 struct netdev_name_node *name_node; 330 struct net *net = dev_net(dev); 331 332 name_node = netdev_name_node_lookup(net, name); 333 if (name_node) 334 return -EEXIST; 335 name_node = netdev_name_node_alloc(dev, name); 336 if (!name_node) 337 return -ENOMEM; 338 netdev_name_node_add(net, name_node); 339 /* The node that holds dev->name acts as a head of per-device list. */ 340 list_add_tail(&name_node->list, &dev->name_node->list); 341 342 return 0; 343 } 344 345 static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node) 346 { 347 list_del(&name_node->list); 348 kfree(name_node->name); 349 netdev_name_node_free(name_node); 350 } 351 352 int netdev_name_node_alt_destroy(struct net_device *dev, const char *name) 353 { 354 struct netdev_name_node *name_node; 355 struct net *net = dev_net(dev); 356 357 name_node = netdev_name_node_lookup(net, name); 358 if (!name_node) 359 return -ENOENT; 360 /* lookup might have found our primary name or a name belonging 361 * to another device. 362 */ 363 if (name_node == dev->name_node || name_node->dev != dev) 364 return -EINVAL; 365 366 netdev_name_node_del(name_node); 367 synchronize_rcu(); 368 __netdev_name_node_alt_destroy(name_node); 369 370 return 0; 371 } 372 373 static void netdev_name_node_alt_flush(struct net_device *dev) 374 { 375 struct netdev_name_node *name_node, *tmp; 376 377 list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list) 378 __netdev_name_node_alt_destroy(name_node); 379 } 380 381 /* Device list insertion */ 382 static void list_netdevice(struct net_device *dev) 383 { 384 struct netdev_name_node *name_node; 385 struct net *net = dev_net(dev); 386 387 ASSERT_RTNL(); 388 389 write_lock(&dev_base_lock); 390 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); 391 netdev_name_node_add(net, dev->name_node); 392 hlist_add_head_rcu(&dev->index_hlist, 393 dev_index_hash(net, dev->ifindex)); 394 write_unlock(&dev_base_lock); 395 396 netdev_for_each_altname(dev, name_node) 397 netdev_name_node_add(net, name_node); 398 399 /* We reserved the ifindex, this can't fail */ 400 WARN_ON(xa_store(&net->dev_by_index, dev->ifindex, dev, GFP_KERNEL)); 401 402 dev_base_seq_inc(net); 403 } 404 405 /* Device list removal 406 * caller must respect a RCU grace period before freeing/reusing dev 407 */ 408 static void unlist_netdevice(struct net_device *dev, bool lock) 409 { 410 struct netdev_name_node *name_node; 411 struct net *net = dev_net(dev); 412 413 ASSERT_RTNL(); 414 415 xa_erase(&net->dev_by_index, dev->ifindex); 416 417 netdev_for_each_altname(dev, name_node) 418 netdev_name_node_del(name_node); 419 420 /* Unlink dev from the device chain */ 421 if (lock) 422 write_lock(&dev_base_lock); 423 list_del_rcu(&dev->dev_list); 424 netdev_name_node_del(dev->name_node); 425 hlist_del_rcu(&dev->index_hlist); 426 if (lock) 427 write_unlock(&dev_base_lock); 428 429 dev_base_seq_inc(dev_net(dev)); 430 } 431 432 /* 433 * Our notifier list 434 */ 435 436 static RAW_NOTIFIER_HEAD(netdev_chain); 437 438 /* 439 * Device drivers call our routines to queue packets here. We empty the 440 * queue in the local softnet handler. 441 */ 442 443 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); 444 EXPORT_PER_CPU_SYMBOL(softnet_data); 445 446 #ifdef CONFIG_LOCKDEP 447 /* 448 * register_netdevice() inits txq->_xmit_lock and sets lockdep class 449 * according to dev->type 450 */ 451 static const unsigned short netdev_lock_type[] = { 452 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, 453 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, 454 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, 455 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, 456 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, 457 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, 458 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, 459 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, 460 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, 461 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, 462 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, 463 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, 464 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, 465 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, 466 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; 467 468 static const char *const netdev_lock_name[] = { 469 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", 470 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", 471 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", 472 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", 473 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", 474 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", 475 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", 476 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", 477 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", 478 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", 479 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", 480 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", 481 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", 482 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", 483 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; 484 485 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; 486 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; 487 488 static inline unsigned short netdev_lock_pos(unsigned short dev_type) 489 { 490 int i; 491 492 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) 493 if (netdev_lock_type[i] == dev_type) 494 return i; 495 /* the last key is used by default */ 496 return ARRAY_SIZE(netdev_lock_type) - 1; 497 } 498 499 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 500 unsigned short dev_type) 501 { 502 int i; 503 504 i = netdev_lock_pos(dev_type); 505 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], 506 netdev_lock_name[i]); 507 } 508 509 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 510 { 511 int i; 512 513 i = netdev_lock_pos(dev->type); 514 lockdep_set_class_and_name(&dev->addr_list_lock, 515 &netdev_addr_lock_key[i], 516 netdev_lock_name[i]); 517 } 518 #else 519 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 520 unsigned short dev_type) 521 { 522 } 523 524 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 525 { 526 } 527 #endif 528 529 /******************************************************************************* 530 * 531 * Protocol management and registration routines 532 * 533 *******************************************************************************/ 534 535 536 /* 537 * Add a protocol ID to the list. Now that the input handler is 538 * smarter we can dispense with all the messy stuff that used to be 539 * here. 540 * 541 * BEWARE!!! Protocol handlers, mangling input packets, 542 * MUST BE last in hash buckets and checking protocol handlers 543 * MUST start from promiscuous ptype_all chain in net_bh. 544 * It is true now, do not change it. 545 * Explanation follows: if protocol handler, mangling packet, will 546 * be the first on list, it is not able to sense, that packet 547 * is cloned and should be copied-on-write, so that it will 548 * change it and subsequent readers will get broken packet. 549 * --ANK (980803) 550 */ 551 552 static inline struct list_head *ptype_head(const struct packet_type *pt) 553 { 554 if (pt->type == htons(ETH_P_ALL)) 555 return pt->dev ? &pt->dev->ptype_all : &ptype_all; 556 else 557 return pt->dev ? &pt->dev->ptype_specific : 558 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; 559 } 560 561 /** 562 * dev_add_pack - add packet handler 563 * @pt: packet type declaration 564 * 565 * Add a protocol handler to the networking stack. The passed &packet_type 566 * is linked into kernel lists and may not be freed until it has been 567 * removed from the kernel lists. 568 * 569 * This call does not sleep therefore it can not 570 * guarantee all CPU's that are in middle of receiving packets 571 * will see the new packet type (until the next received packet). 572 */ 573 574 void dev_add_pack(struct packet_type *pt) 575 { 576 struct list_head *head = ptype_head(pt); 577 578 spin_lock(&ptype_lock); 579 list_add_rcu(&pt->list, head); 580 spin_unlock(&ptype_lock); 581 } 582 EXPORT_SYMBOL(dev_add_pack); 583 584 /** 585 * __dev_remove_pack - remove packet handler 586 * @pt: packet type declaration 587 * 588 * Remove a protocol handler that was previously added to the kernel 589 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 590 * from the kernel lists and can be freed or reused once this function 591 * returns. 592 * 593 * The packet type might still be in use by receivers 594 * and must not be freed until after all the CPU's have gone 595 * through a quiescent state. 596 */ 597 void __dev_remove_pack(struct packet_type *pt) 598 { 599 struct list_head *head = ptype_head(pt); 600 struct packet_type *pt1; 601 602 spin_lock(&ptype_lock); 603 604 list_for_each_entry(pt1, head, list) { 605 if (pt == pt1) { 606 list_del_rcu(&pt->list); 607 goto out; 608 } 609 } 610 611 pr_warn("dev_remove_pack: %p not found\n", pt); 612 out: 613 spin_unlock(&ptype_lock); 614 } 615 EXPORT_SYMBOL(__dev_remove_pack); 616 617 /** 618 * dev_remove_pack - remove packet handler 619 * @pt: packet type declaration 620 * 621 * Remove a protocol handler that was previously added to the kernel 622 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 623 * from the kernel lists and can be freed or reused once this function 624 * returns. 625 * 626 * This call sleeps to guarantee that no CPU is looking at the packet 627 * type after return. 628 */ 629 void dev_remove_pack(struct packet_type *pt) 630 { 631 __dev_remove_pack(pt); 632 633 synchronize_net(); 634 } 635 EXPORT_SYMBOL(dev_remove_pack); 636 637 638 /******************************************************************************* 639 * 640 * Device Interface Subroutines 641 * 642 *******************************************************************************/ 643 644 /** 645 * dev_get_iflink - get 'iflink' value of a interface 646 * @dev: targeted interface 647 * 648 * Indicates the ifindex the interface is linked to. 649 * Physical interfaces have the same 'ifindex' and 'iflink' values. 650 */ 651 652 int dev_get_iflink(const struct net_device *dev) 653 { 654 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink) 655 return dev->netdev_ops->ndo_get_iflink(dev); 656 657 return dev->ifindex; 658 } 659 EXPORT_SYMBOL(dev_get_iflink); 660 661 /** 662 * dev_fill_metadata_dst - Retrieve tunnel egress information. 663 * @dev: targeted interface 664 * @skb: The packet. 665 * 666 * For better visibility of tunnel traffic OVS needs to retrieve 667 * egress tunnel information for a packet. Following API allows 668 * user to get this info. 669 */ 670 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) 671 { 672 struct ip_tunnel_info *info; 673 674 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst) 675 return -EINVAL; 676 677 info = skb_tunnel_info_unclone(skb); 678 if (!info) 679 return -ENOMEM; 680 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX))) 681 return -EINVAL; 682 683 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb); 684 } 685 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst); 686 687 static struct net_device_path *dev_fwd_path(struct net_device_path_stack *stack) 688 { 689 int k = stack->num_paths++; 690 691 if (WARN_ON_ONCE(k >= NET_DEVICE_PATH_STACK_MAX)) 692 return NULL; 693 694 return &stack->path[k]; 695 } 696 697 int dev_fill_forward_path(const struct net_device *dev, const u8 *daddr, 698 struct net_device_path_stack *stack) 699 { 700 const struct net_device *last_dev; 701 struct net_device_path_ctx ctx = { 702 .dev = dev, 703 }; 704 struct net_device_path *path; 705 int ret = 0; 706 707 memcpy(ctx.daddr, daddr, sizeof(ctx.daddr)); 708 stack->num_paths = 0; 709 while (ctx.dev && ctx.dev->netdev_ops->ndo_fill_forward_path) { 710 last_dev = ctx.dev; 711 path = dev_fwd_path(stack); 712 if (!path) 713 return -1; 714 715 memset(path, 0, sizeof(struct net_device_path)); 716 ret = ctx.dev->netdev_ops->ndo_fill_forward_path(&ctx, path); 717 if (ret < 0) 718 return -1; 719 720 if (WARN_ON_ONCE(last_dev == ctx.dev)) 721 return -1; 722 } 723 724 if (!ctx.dev) 725 return ret; 726 727 path = dev_fwd_path(stack); 728 if (!path) 729 return -1; 730 path->type = DEV_PATH_ETHERNET; 731 path->dev = ctx.dev; 732 733 return ret; 734 } 735 EXPORT_SYMBOL_GPL(dev_fill_forward_path); 736 737 /** 738 * __dev_get_by_name - find a device by its name 739 * @net: the applicable net namespace 740 * @name: name to find 741 * 742 * Find an interface by name. Must be called under RTNL semaphore 743 * or @dev_base_lock. If the name is found a pointer to the device 744 * is returned. If the name is not found then %NULL is returned. The 745 * reference counters are not incremented so the caller must be 746 * careful with locks. 747 */ 748 749 struct net_device *__dev_get_by_name(struct net *net, const char *name) 750 { 751 struct netdev_name_node *node_name; 752 753 node_name = netdev_name_node_lookup(net, name); 754 return node_name ? node_name->dev : NULL; 755 } 756 EXPORT_SYMBOL(__dev_get_by_name); 757 758 /** 759 * dev_get_by_name_rcu - find a device by its name 760 * @net: the applicable net namespace 761 * @name: name to find 762 * 763 * Find an interface by name. 764 * If the name is found a pointer to the device is returned. 765 * If the name is not found then %NULL is returned. 766 * The reference counters are not incremented so the caller must be 767 * careful with locks. The caller must hold RCU lock. 768 */ 769 770 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) 771 { 772 struct netdev_name_node *node_name; 773 774 node_name = netdev_name_node_lookup_rcu(net, name); 775 return node_name ? node_name->dev : NULL; 776 } 777 EXPORT_SYMBOL(dev_get_by_name_rcu); 778 779 /* Deprecated for new users, call netdev_get_by_name() instead */ 780 struct net_device *dev_get_by_name(struct net *net, const char *name) 781 { 782 struct net_device *dev; 783 784 rcu_read_lock(); 785 dev = dev_get_by_name_rcu(net, name); 786 dev_hold(dev); 787 rcu_read_unlock(); 788 return dev; 789 } 790 EXPORT_SYMBOL(dev_get_by_name); 791 792 /** 793 * netdev_get_by_name() - find a device by its name 794 * @net: the applicable net namespace 795 * @name: name to find 796 * @tracker: tracking object for the acquired reference 797 * @gfp: allocation flags for the tracker 798 * 799 * Find an interface by name. This can be called from any 800 * context and does its own locking. The returned handle has 801 * the usage count incremented and the caller must use netdev_put() to 802 * release it when it is no longer needed. %NULL is returned if no 803 * matching device is found. 804 */ 805 struct net_device *netdev_get_by_name(struct net *net, const char *name, 806 netdevice_tracker *tracker, gfp_t gfp) 807 { 808 struct net_device *dev; 809 810 dev = dev_get_by_name(net, name); 811 if (dev) 812 netdev_tracker_alloc(dev, tracker, gfp); 813 return dev; 814 } 815 EXPORT_SYMBOL(netdev_get_by_name); 816 817 /** 818 * __dev_get_by_index - find a device by its ifindex 819 * @net: the applicable net namespace 820 * @ifindex: index of device 821 * 822 * Search for an interface by index. Returns %NULL if the device 823 * is not found or a pointer to the device. The device has not 824 * had its reference counter increased so the caller must be careful 825 * about locking. The caller must hold either the RTNL semaphore 826 * or @dev_base_lock. 827 */ 828 829 struct net_device *__dev_get_by_index(struct net *net, int ifindex) 830 { 831 struct net_device *dev; 832 struct hlist_head *head = dev_index_hash(net, ifindex); 833 834 hlist_for_each_entry(dev, head, index_hlist) 835 if (dev->ifindex == ifindex) 836 return dev; 837 838 return NULL; 839 } 840 EXPORT_SYMBOL(__dev_get_by_index); 841 842 /** 843 * dev_get_by_index_rcu - find a device by its ifindex 844 * @net: the applicable net namespace 845 * @ifindex: index of device 846 * 847 * Search for an interface by index. Returns %NULL if the device 848 * is not found or a pointer to the device. The device has not 849 * had its reference counter increased so the caller must be careful 850 * about locking. The caller must hold RCU lock. 851 */ 852 853 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) 854 { 855 struct net_device *dev; 856 struct hlist_head *head = dev_index_hash(net, ifindex); 857 858 hlist_for_each_entry_rcu(dev, head, index_hlist) 859 if (dev->ifindex == ifindex) 860 return dev; 861 862 return NULL; 863 } 864 EXPORT_SYMBOL(dev_get_by_index_rcu); 865 866 /* Deprecated for new users, call netdev_get_by_index() instead */ 867 struct net_device *dev_get_by_index(struct net *net, int ifindex) 868 { 869 struct net_device *dev; 870 871 rcu_read_lock(); 872 dev = dev_get_by_index_rcu(net, ifindex); 873 dev_hold(dev); 874 rcu_read_unlock(); 875 return dev; 876 } 877 EXPORT_SYMBOL(dev_get_by_index); 878 879 /** 880 * netdev_get_by_index() - find a device by its ifindex 881 * @net: the applicable net namespace 882 * @ifindex: index of device 883 * @tracker: tracking object for the acquired reference 884 * @gfp: allocation flags for the tracker 885 * 886 * Search for an interface by index. Returns NULL if the device 887 * is not found or a pointer to the device. The device returned has 888 * had a reference added and the pointer is safe until the user calls 889 * netdev_put() to indicate they have finished with it. 890 */ 891 struct net_device *netdev_get_by_index(struct net *net, int ifindex, 892 netdevice_tracker *tracker, gfp_t gfp) 893 { 894 struct net_device *dev; 895 896 dev = dev_get_by_index(net, ifindex); 897 if (dev) 898 netdev_tracker_alloc(dev, tracker, gfp); 899 return dev; 900 } 901 EXPORT_SYMBOL(netdev_get_by_index); 902 903 /** 904 * dev_get_by_napi_id - find a device by napi_id 905 * @napi_id: ID of the NAPI struct 906 * 907 * Search for an interface by NAPI ID. Returns %NULL if the device 908 * is not found or a pointer to the device. The device has not had 909 * its reference counter increased so the caller must be careful 910 * about locking. The caller must hold RCU lock. 911 */ 912 913 struct net_device *dev_get_by_napi_id(unsigned int napi_id) 914 { 915 struct napi_struct *napi; 916 917 WARN_ON_ONCE(!rcu_read_lock_held()); 918 919 if (napi_id < MIN_NAPI_ID) 920 return NULL; 921 922 napi = napi_by_id(napi_id); 923 924 return napi ? napi->dev : NULL; 925 } 926 EXPORT_SYMBOL(dev_get_by_napi_id); 927 928 /** 929 * netdev_get_name - get a netdevice name, knowing its ifindex. 930 * @net: network namespace 931 * @name: a pointer to the buffer where the name will be stored. 932 * @ifindex: the ifindex of the interface to get the name from. 933 */ 934 int netdev_get_name(struct net *net, char *name, int ifindex) 935 { 936 struct net_device *dev; 937 int ret; 938 939 down_read(&devnet_rename_sem); 940 rcu_read_lock(); 941 942 dev = dev_get_by_index_rcu(net, ifindex); 943 if (!dev) { 944 ret = -ENODEV; 945 goto out; 946 } 947 948 strcpy(name, dev->name); 949 950 ret = 0; 951 out: 952 rcu_read_unlock(); 953 up_read(&devnet_rename_sem); 954 return ret; 955 } 956 957 /** 958 * dev_getbyhwaddr_rcu - find a device by its hardware address 959 * @net: the applicable net namespace 960 * @type: media type of device 961 * @ha: hardware address 962 * 963 * Search for an interface by MAC address. Returns NULL if the device 964 * is not found or a pointer to the device. 965 * The caller must hold RCU or RTNL. 966 * The returned device has not had its ref count increased 967 * and the caller must therefore be careful about locking 968 * 969 */ 970 971 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, 972 const char *ha) 973 { 974 struct net_device *dev; 975 976 for_each_netdev_rcu(net, dev) 977 if (dev->type == type && 978 !memcmp(dev->dev_addr, ha, dev->addr_len)) 979 return dev; 980 981 return NULL; 982 } 983 EXPORT_SYMBOL(dev_getbyhwaddr_rcu); 984 985 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) 986 { 987 struct net_device *dev, *ret = NULL; 988 989 rcu_read_lock(); 990 for_each_netdev_rcu(net, dev) 991 if (dev->type == type) { 992 dev_hold(dev); 993 ret = dev; 994 break; 995 } 996 rcu_read_unlock(); 997 return ret; 998 } 999 EXPORT_SYMBOL(dev_getfirstbyhwtype); 1000 1001 /** 1002 * __dev_get_by_flags - find any device with given flags 1003 * @net: the applicable net namespace 1004 * @if_flags: IFF_* values 1005 * @mask: bitmask of bits in if_flags to check 1006 * 1007 * Search for any interface with the given flags. Returns NULL if a device 1008 * is not found or a pointer to the device. Must be called inside 1009 * rtnl_lock(), and result refcount is unchanged. 1010 */ 1011 1012 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags, 1013 unsigned short mask) 1014 { 1015 struct net_device *dev, *ret; 1016 1017 ASSERT_RTNL(); 1018 1019 ret = NULL; 1020 for_each_netdev(net, dev) { 1021 if (((dev->flags ^ if_flags) & mask) == 0) { 1022 ret = dev; 1023 break; 1024 } 1025 } 1026 return ret; 1027 } 1028 EXPORT_SYMBOL(__dev_get_by_flags); 1029 1030 /** 1031 * dev_valid_name - check if name is okay for network device 1032 * @name: name string 1033 * 1034 * Network device names need to be valid file names to 1035 * allow sysfs to work. We also disallow any kind of 1036 * whitespace. 1037 */ 1038 bool dev_valid_name(const char *name) 1039 { 1040 if (*name == '\0') 1041 return false; 1042 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ) 1043 return false; 1044 if (!strcmp(name, ".") || !strcmp(name, "..")) 1045 return false; 1046 1047 while (*name) { 1048 if (*name == '/' || *name == ':' || isspace(*name)) 1049 return false; 1050 name++; 1051 } 1052 return true; 1053 } 1054 EXPORT_SYMBOL(dev_valid_name); 1055 1056 /** 1057 * __dev_alloc_name - allocate a name for a device 1058 * @net: network namespace to allocate the device name in 1059 * @name: name format string 1060 * @buf: scratch buffer and result name string 1061 * 1062 * Passed a format string - eg "lt%d" it will try and find a suitable 1063 * id. It scans list of devices to build up a free map, then chooses 1064 * the first empty slot. The caller must hold the dev_base or rtnl lock 1065 * while allocating the name and adding the device in order to avoid 1066 * duplicates. 1067 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1068 * Returns the number of the unit assigned or a negative errno code. 1069 */ 1070 1071 static int __dev_alloc_name(struct net *net, const char *name, char *buf) 1072 { 1073 int i = 0; 1074 const char *p; 1075 const int max_netdevices = 8*PAGE_SIZE; 1076 unsigned long *inuse; 1077 struct net_device *d; 1078 1079 if (!dev_valid_name(name)) 1080 return -EINVAL; 1081 1082 p = strchr(name, '%'); 1083 if (p) { 1084 /* 1085 * Verify the string as this thing may have come from 1086 * the user. There must be either one "%d" and no other "%" 1087 * characters. 1088 */ 1089 if (p[1] != 'd' || strchr(p + 2, '%')) 1090 return -EINVAL; 1091 1092 /* Use one page as a bit array of possible slots */ 1093 inuse = bitmap_zalloc(max_netdevices, GFP_ATOMIC); 1094 if (!inuse) 1095 return -ENOMEM; 1096 1097 for_each_netdev(net, d) { 1098 struct netdev_name_node *name_node; 1099 1100 netdev_for_each_altname(d, name_node) { 1101 if (!sscanf(name_node->name, name, &i)) 1102 continue; 1103 if (i < 0 || i >= max_netdevices) 1104 continue; 1105 1106 /* avoid cases where sscanf is not exact inverse of printf */ 1107 snprintf(buf, IFNAMSIZ, name, i); 1108 if (!strncmp(buf, name_node->name, IFNAMSIZ)) 1109 __set_bit(i, inuse); 1110 } 1111 if (!sscanf(d->name, name, &i)) 1112 continue; 1113 if (i < 0 || i >= max_netdevices) 1114 continue; 1115 1116 /* avoid cases where sscanf is not exact inverse of printf */ 1117 snprintf(buf, IFNAMSIZ, name, i); 1118 if (!strncmp(buf, d->name, IFNAMSIZ)) 1119 __set_bit(i, inuse); 1120 } 1121 1122 i = find_first_zero_bit(inuse, max_netdevices); 1123 bitmap_free(inuse); 1124 } 1125 1126 snprintf(buf, IFNAMSIZ, name, i); 1127 if (!netdev_name_in_use(net, buf)) 1128 return i; 1129 1130 /* It is possible to run out of possible slots 1131 * when the name is long and there isn't enough space left 1132 * for the digits, or if all bits are used. 1133 */ 1134 return -ENFILE; 1135 } 1136 1137 static int dev_prep_valid_name(struct net *net, struct net_device *dev, 1138 const char *want_name, char *out_name) 1139 { 1140 int ret; 1141 1142 if (!dev_valid_name(want_name)) 1143 return -EINVAL; 1144 1145 if (strchr(want_name, '%')) { 1146 ret = __dev_alloc_name(net, want_name, out_name); 1147 return ret < 0 ? ret : 0; 1148 } else if (netdev_name_in_use(net, want_name)) { 1149 return -EEXIST; 1150 } else if (out_name != want_name) { 1151 strscpy(out_name, want_name, IFNAMSIZ); 1152 } 1153 1154 return 0; 1155 } 1156 1157 static int dev_alloc_name_ns(struct net *net, 1158 struct net_device *dev, 1159 const char *name) 1160 { 1161 char buf[IFNAMSIZ]; 1162 int ret; 1163 1164 BUG_ON(!net); 1165 ret = __dev_alloc_name(net, name, buf); 1166 if (ret >= 0) 1167 strscpy(dev->name, buf, IFNAMSIZ); 1168 return ret; 1169 } 1170 1171 /** 1172 * dev_alloc_name - allocate a name for a device 1173 * @dev: device 1174 * @name: name format string 1175 * 1176 * Passed a format string - eg "lt%d" it will try and find a suitable 1177 * id. It scans list of devices to build up a free map, then chooses 1178 * the first empty slot. The caller must hold the dev_base or rtnl lock 1179 * while allocating the name and adding the device in order to avoid 1180 * duplicates. 1181 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1182 * Returns the number of the unit assigned or a negative errno code. 1183 */ 1184 1185 int dev_alloc_name(struct net_device *dev, const char *name) 1186 { 1187 return dev_alloc_name_ns(dev_net(dev), dev, name); 1188 } 1189 EXPORT_SYMBOL(dev_alloc_name); 1190 1191 static int dev_get_valid_name(struct net *net, struct net_device *dev, 1192 const char *name) 1193 { 1194 char buf[IFNAMSIZ]; 1195 int ret; 1196 1197 ret = dev_prep_valid_name(net, dev, name, buf); 1198 if (ret >= 0) 1199 strscpy(dev->name, buf, IFNAMSIZ); 1200 return ret; 1201 } 1202 1203 /** 1204 * dev_change_name - change name of a device 1205 * @dev: device 1206 * @newname: name (or format string) must be at least IFNAMSIZ 1207 * 1208 * Change name of a device, can pass format strings "eth%d". 1209 * for wildcarding. 1210 */ 1211 int dev_change_name(struct net_device *dev, const char *newname) 1212 { 1213 unsigned char old_assign_type; 1214 char oldname[IFNAMSIZ]; 1215 int err = 0; 1216 int ret; 1217 struct net *net; 1218 1219 ASSERT_RTNL(); 1220 BUG_ON(!dev_net(dev)); 1221 1222 net = dev_net(dev); 1223 1224 down_write(&devnet_rename_sem); 1225 1226 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { 1227 up_write(&devnet_rename_sem); 1228 return 0; 1229 } 1230 1231 memcpy(oldname, dev->name, IFNAMSIZ); 1232 1233 err = dev_get_valid_name(net, dev, newname); 1234 if (err < 0) { 1235 up_write(&devnet_rename_sem); 1236 return err; 1237 } 1238 1239 if (oldname[0] && !strchr(oldname, '%')) 1240 netdev_info(dev, "renamed from %s%s\n", oldname, 1241 dev->flags & IFF_UP ? " (while UP)" : ""); 1242 1243 old_assign_type = dev->name_assign_type; 1244 dev->name_assign_type = NET_NAME_RENAMED; 1245 1246 rollback: 1247 ret = device_rename(&dev->dev, dev->name); 1248 if (ret) { 1249 memcpy(dev->name, oldname, IFNAMSIZ); 1250 dev->name_assign_type = old_assign_type; 1251 up_write(&devnet_rename_sem); 1252 return ret; 1253 } 1254 1255 up_write(&devnet_rename_sem); 1256 1257 netdev_adjacent_rename_links(dev, oldname); 1258 1259 write_lock(&dev_base_lock); 1260 netdev_name_node_del(dev->name_node); 1261 write_unlock(&dev_base_lock); 1262 1263 synchronize_rcu(); 1264 1265 write_lock(&dev_base_lock); 1266 netdev_name_node_add(net, dev->name_node); 1267 write_unlock(&dev_base_lock); 1268 1269 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); 1270 ret = notifier_to_errno(ret); 1271 1272 if (ret) { 1273 /* err >= 0 after dev_alloc_name() or stores the first errno */ 1274 if (err >= 0) { 1275 err = ret; 1276 down_write(&devnet_rename_sem); 1277 memcpy(dev->name, oldname, IFNAMSIZ); 1278 memcpy(oldname, newname, IFNAMSIZ); 1279 dev->name_assign_type = old_assign_type; 1280 old_assign_type = NET_NAME_RENAMED; 1281 goto rollback; 1282 } else { 1283 netdev_err(dev, "name change rollback failed: %d\n", 1284 ret); 1285 } 1286 } 1287 1288 return err; 1289 } 1290 1291 /** 1292 * dev_set_alias - change ifalias of a device 1293 * @dev: device 1294 * @alias: name up to IFALIASZ 1295 * @len: limit of bytes to copy from info 1296 * 1297 * Set ifalias for a device, 1298 */ 1299 int dev_set_alias(struct net_device *dev, const char *alias, size_t len) 1300 { 1301 struct dev_ifalias *new_alias = NULL; 1302 1303 if (len >= IFALIASZ) 1304 return -EINVAL; 1305 1306 if (len) { 1307 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL); 1308 if (!new_alias) 1309 return -ENOMEM; 1310 1311 memcpy(new_alias->ifalias, alias, len); 1312 new_alias->ifalias[len] = 0; 1313 } 1314 1315 mutex_lock(&ifalias_mutex); 1316 new_alias = rcu_replace_pointer(dev->ifalias, new_alias, 1317 mutex_is_locked(&ifalias_mutex)); 1318 mutex_unlock(&ifalias_mutex); 1319 1320 if (new_alias) 1321 kfree_rcu(new_alias, rcuhead); 1322 1323 return len; 1324 } 1325 EXPORT_SYMBOL(dev_set_alias); 1326 1327 /** 1328 * dev_get_alias - get ifalias of a device 1329 * @dev: device 1330 * @name: buffer to store name of ifalias 1331 * @len: size of buffer 1332 * 1333 * get ifalias for a device. Caller must make sure dev cannot go 1334 * away, e.g. rcu read lock or own a reference count to device. 1335 */ 1336 int dev_get_alias(const struct net_device *dev, char *name, size_t len) 1337 { 1338 const struct dev_ifalias *alias; 1339 int ret = 0; 1340 1341 rcu_read_lock(); 1342 alias = rcu_dereference(dev->ifalias); 1343 if (alias) 1344 ret = snprintf(name, len, "%s", alias->ifalias); 1345 rcu_read_unlock(); 1346 1347 return ret; 1348 } 1349 1350 /** 1351 * netdev_features_change - device changes features 1352 * @dev: device to cause notification 1353 * 1354 * Called to indicate a device has changed features. 1355 */ 1356 void netdev_features_change(struct net_device *dev) 1357 { 1358 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); 1359 } 1360 EXPORT_SYMBOL(netdev_features_change); 1361 1362 /** 1363 * netdev_state_change - device changes state 1364 * @dev: device to cause notification 1365 * 1366 * Called to indicate a device has changed state. This function calls 1367 * the notifier chains for netdev_chain and sends a NEWLINK message 1368 * to the routing socket. 1369 */ 1370 void netdev_state_change(struct net_device *dev) 1371 { 1372 if (dev->flags & IFF_UP) { 1373 struct netdev_notifier_change_info change_info = { 1374 .info.dev = dev, 1375 }; 1376 1377 call_netdevice_notifiers_info(NETDEV_CHANGE, 1378 &change_info.info); 1379 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL, 0, NULL); 1380 } 1381 } 1382 EXPORT_SYMBOL(netdev_state_change); 1383 1384 /** 1385 * __netdev_notify_peers - notify network peers about existence of @dev, 1386 * to be called when rtnl lock is already held. 1387 * @dev: network device 1388 * 1389 * Generate traffic such that interested network peers are aware of 1390 * @dev, such as by generating a gratuitous ARP. This may be used when 1391 * a device wants to inform the rest of the network about some sort of 1392 * reconfiguration such as a failover event or virtual machine 1393 * migration. 1394 */ 1395 void __netdev_notify_peers(struct net_device *dev) 1396 { 1397 ASSERT_RTNL(); 1398 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); 1399 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev); 1400 } 1401 EXPORT_SYMBOL(__netdev_notify_peers); 1402 1403 /** 1404 * netdev_notify_peers - notify network peers about existence of @dev 1405 * @dev: network device 1406 * 1407 * Generate traffic such that interested network peers are aware of 1408 * @dev, such as by generating a gratuitous ARP. This may be used when 1409 * a device wants to inform the rest of the network about some sort of 1410 * reconfiguration such as a failover event or virtual machine 1411 * migration. 1412 */ 1413 void netdev_notify_peers(struct net_device *dev) 1414 { 1415 rtnl_lock(); 1416 __netdev_notify_peers(dev); 1417 rtnl_unlock(); 1418 } 1419 EXPORT_SYMBOL(netdev_notify_peers); 1420 1421 static int napi_threaded_poll(void *data); 1422 1423 static int napi_kthread_create(struct napi_struct *n) 1424 { 1425 int err = 0; 1426 1427 /* Create and wake up the kthread once to put it in 1428 * TASK_INTERRUPTIBLE mode to avoid the blocked task 1429 * warning and work with loadavg. 1430 */ 1431 n->thread = kthread_run(napi_threaded_poll, n, "napi/%s-%d", 1432 n->dev->name, n->napi_id); 1433 if (IS_ERR(n->thread)) { 1434 err = PTR_ERR(n->thread); 1435 pr_err("kthread_run failed with err %d\n", err); 1436 n->thread = NULL; 1437 } 1438 1439 return err; 1440 } 1441 1442 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1443 { 1444 const struct net_device_ops *ops = dev->netdev_ops; 1445 int ret; 1446 1447 ASSERT_RTNL(); 1448 dev_addr_check(dev); 1449 1450 if (!netif_device_present(dev)) { 1451 /* may be detached because parent is runtime-suspended */ 1452 if (dev->dev.parent) 1453 pm_runtime_resume(dev->dev.parent); 1454 if (!netif_device_present(dev)) 1455 return -ENODEV; 1456 } 1457 1458 /* Block netpoll from trying to do any rx path servicing. 1459 * If we don't do this there is a chance ndo_poll_controller 1460 * or ndo_poll may be running while we open the device 1461 */ 1462 netpoll_poll_disable(dev); 1463 1464 ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack); 1465 ret = notifier_to_errno(ret); 1466 if (ret) 1467 return ret; 1468 1469 set_bit(__LINK_STATE_START, &dev->state); 1470 1471 if (ops->ndo_validate_addr) 1472 ret = ops->ndo_validate_addr(dev); 1473 1474 if (!ret && ops->ndo_open) 1475 ret = ops->ndo_open(dev); 1476 1477 netpoll_poll_enable(dev); 1478 1479 if (ret) 1480 clear_bit(__LINK_STATE_START, &dev->state); 1481 else { 1482 dev->flags |= IFF_UP; 1483 dev_set_rx_mode(dev); 1484 dev_activate(dev); 1485 add_device_randomness(dev->dev_addr, dev->addr_len); 1486 } 1487 1488 return ret; 1489 } 1490 1491 /** 1492 * dev_open - prepare an interface for use. 1493 * @dev: device to open 1494 * @extack: netlink extended ack 1495 * 1496 * Takes a device from down to up state. The device's private open 1497 * function is invoked and then the multicast lists are loaded. Finally 1498 * the device is moved into the up state and a %NETDEV_UP message is 1499 * sent to the netdev notifier chain. 1500 * 1501 * Calling this function on an active interface is a nop. On a failure 1502 * a negative errno code is returned. 1503 */ 1504 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1505 { 1506 int ret; 1507 1508 if (dev->flags & IFF_UP) 1509 return 0; 1510 1511 ret = __dev_open(dev, extack); 1512 if (ret < 0) 1513 return ret; 1514 1515 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL); 1516 call_netdevice_notifiers(NETDEV_UP, dev); 1517 1518 return ret; 1519 } 1520 EXPORT_SYMBOL(dev_open); 1521 1522 static void __dev_close_many(struct list_head *head) 1523 { 1524 struct net_device *dev; 1525 1526 ASSERT_RTNL(); 1527 might_sleep(); 1528 1529 list_for_each_entry(dev, head, close_list) { 1530 /* Temporarily disable netpoll until the interface is down */ 1531 netpoll_poll_disable(dev); 1532 1533 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); 1534 1535 clear_bit(__LINK_STATE_START, &dev->state); 1536 1537 /* Synchronize to scheduled poll. We cannot touch poll list, it 1538 * can be even on different cpu. So just clear netif_running(). 1539 * 1540 * dev->stop() will invoke napi_disable() on all of it's 1541 * napi_struct instances on this device. 1542 */ 1543 smp_mb__after_atomic(); /* Commit netif_running(). */ 1544 } 1545 1546 dev_deactivate_many(head); 1547 1548 list_for_each_entry(dev, head, close_list) { 1549 const struct net_device_ops *ops = dev->netdev_ops; 1550 1551 /* 1552 * Call the device specific close. This cannot fail. 1553 * Only if device is UP 1554 * 1555 * We allow it to be called even after a DETACH hot-plug 1556 * event. 1557 */ 1558 if (ops->ndo_stop) 1559 ops->ndo_stop(dev); 1560 1561 dev->flags &= ~IFF_UP; 1562 netpoll_poll_enable(dev); 1563 } 1564 } 1565 1566 static void __dev_close(struct net_device *dev) 1567 { 1568 LIST_HEAD(single); 1569 1570 list_add(&dev->close_list, &single); 1571 __dev_close_many(&single); 1572 list_del(&single); 1573 } 1574 1575 void dev_close_many(struct list_head *head, bool unlink) 1576 { 1577 struct net_device *dev, *tmp; 1578 1579 /* Remove the devices that don't need to be closed */ 1580 list_for_each_entry_safe(dev, tmp, head, close_list) 1581 if (!(dev->flags & IFF_UP)) 1582 list_del_init(&dev->close_list); 1583 1584 __dev_close_many(head); 1585 1586 list_for_each_entry_safe(dev, tmp, head, close_list) { 1587 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL); 1588 call_netdevice_notifiers(NETDEV_DOWN, dev); 1589 if (unlink) 1590 list_del_init(&dev->close_list); 1591 } 1592 } 1593 EXPORT_SYMBOL(dev_close_many); 1594 1595 /** 1596 * dev_close - shutdown an interface. 1597 * @dev: device to shutdown 1598 * 1599 * This function moves an active device into down state. A 1600 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device 1601 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier 1602 * chain. 1603 */ 1604 void dev_close(struct net_device *dev) 1605 { 1606 if (dev->flags & IFF_UP) { 1607 LIST_HEAD(single); 1608 1609 list_add(&dev->close_list, &single); 1610 dev_close_many(&single, true); 1611 list_del(&single); 1612 } 1613 } 1614 EXPORT_SYMBOL(dev_close); 1615 1616 1617 /** 1618 * dev_disable_lro - disable Large Receive Offload on a device 1619 * @dev: device 1620 * 1621 * Disable Large Receive Offload (LRO) on a net device. Must be 1622 * called under RTNL. This is needed if received packets may be 1623 * forwarded to another interface. 1624 */ 1625 void dev_disable_lro(struct net_device *dev) 1626 { 1627 struct net_device *lower_dev; 1628 struct list_head *iter; 1629 1630 dev->wanted_features &= ~NETIF_F_LRO; 1631 netdev_update_features(dev); 1632 1633 if (unlikely(dev->features & NETIF_F_LRO)) 1634 netdev_WARN(dev, "failed to disable LRO!\n"); 1635 1636 netdev_for_each_lower_dev(dev, lower_dev, iter) 1637 dev_disable_lro(lower_dev); 1638 } 1639 EXPORT_SYMBOL(dev_disable_lro); 1640 1641 /** 1642 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device 1643 * @dev: device 1644 * 1645 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be 1646 * called under RTNL. This is needed if Generic XDP is installed on 1647 * the device. 1648 */ 1649 static void dev_disable_gro_hw(struct net_device *dev) 1650 { 1651 dev->wanted_features &= ~NETIF_F_GRO_HW; 1652 netdev_update_features(dev); 1653 1654 if (unlikely(dev->features & NETIF_F_GRO_HW)) 1655 netdev_WARN(dev, "failed to disable GRO_HW!\n"); 1656 } 1657 1658 const char *netdev_cmd_to_name(enum netdev_cmd cmd) 1659 { 1660 #define N(val) \ 1661 case NETDEV_##val: \ 1662 return "NETDEV_" __stringify(val); 1663 switch (cmd) { 1664 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER) 1665 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE) 1666 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE) 1667 N(POST_INIT) N(PRE_UNINIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) 1668 N(CHANGEUPPER) N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) 1669 N(BONDING_INFO) N(PRECHANGEUPPER) N(CHANGELOWERSTATE) 1670 N(UDP_TUNNEL_PUSH_INFO) N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN) 1671 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO) 1672 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO) 1673 N(PRE_CHANGEADDR) N(OFFLOAD_XSTATS_ENABLE) N(OFFLOAD_XSTATS_DISABLE) 1674 N(OFFLOAD_XSTATS_REPORT_USED) N(OFFLOAD_XSTATS_REPORT_DELTA) 1675 N(XDP_FEAT_CHANGE) 1676 } 1677 #undef N 1678 return "UNKNOWN_NETDEV_EVENT"; 1679 } 1680 EXPORT_SYMBOL_GPL(netdev_cmd_to_name); 1681 1682 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, 1683 struct net_device *dev) 1684 { 1685 struct netdev_notifier_info info = { 1686 .dev = dev, 1687 }; 1688 1689 return nb->notifier_call(nb, val, &info); 1690 } 1691 1692 static int call_netdevice_register_notifiers(struct notifier_block *nb, 1693 struct net_device *dev) 1694 { 1695 int err; 1696 1697 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); 1698 err = notifier_to_errno(err); 1699 if (err) 1700 return err; 1701 1702 if (!(dev->flags & IFF_UP)) 1703 return 0; 1704 1705 call_netdevice_notifier(nb, NETDEV_UP, dev); 1706 return 0; 1707 } 1708 1709 static void call_netdevice_unregister_notifiers(struct notifier_block *nb, 1710 struct net_device *dev) 1711 { 1712 if (dev->flags & IFF_UP) { 1713 call_netdevice_notifier(nb, NETDEV_GOING_DOWN, 1714 dev); 1715 call_netdevice_notifier(nb, NETDEV_DOWN, dev); 1716 } 1717 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); 1718 } 1719 1720 static int call_netdevice_register_net_notifiers(struct notifier_block *nb, 1721 struct net *net) 1722 { 1723 struct net_device *dev; 1724 int err; 1725 1726 for_each_netdev(net, dev) { 1727 err = call_netdevice_register_notifiers(nb, dev); 1728 if (err) 1729 goto rollback; 1730 } 1731 return 0; 1732 1733 rollback: 1734 for_each_netdev_continue_reverse(net, dev) 1735 call_netdevice_unregister_notifiers(nb, dev); 1736 return err; 1737 } 1738 1739 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb, 1740 struct net *net) 1741 { 1742 struct net_device *dev; 1743 1744 for_each_netdev(net, dev) 1745 call_netdevice_unregister_notifiers(nb, dev); 1746 } 1747 1748 static int dev_boot_phase = 1; 1749 1750 /** 1751 * register_netdevice_notifier - register a network notifier block 1752 * @nb: notifier 1753 * 1754 * Register a notifier to be called when network device events occur. 1755 * The notifier passed is linked into the kernel structures and must 1756 * not be reused until it has been unregistered. A negative errno code 1757 * is returned on a failure. 1758 * 1759 * When registered all registration and up events are replayed 1760 * to the new notifier to allow device to have a race free 1761 * view of the network device list. 1762 */ 1763 1764 int register_netdevice_notifier(struct notifier_block *nb) 1765 { 1766 struct net *net; 1767 int err; 1768 1769 /* Close race with setup_net() and cleanup_net() */ 1770 down_write(&pernet_ops_rwsem); 1771 rtnl_lock(); 1772 err = raw_notifier_chain_register(&netdev_chain, nb); 1773 if (err) 1774 goto unlock; 1775 if (dev_boot_phase) 1776 goto unlock; 1777 for_each_net(net) { 1778 err = call_netdevice_register_net_notifiers(nb, net); 1779 if (err) 1780 goto rollback; 1781 } 1782 1783 unlock: 1784 rtnl_unlock(); 1785 up_write(&pernet_ops_rwsem); 1786 return err; 1787 1788 rollback: 1789 for_each_net_continue_reverse(net) 1790 call_netdevice_unregister_net_notifiers(nb, net); 1791 1792 raw_notifier_chain_unregister(&netdev_chain, nb); 1793 goto unlock; 1794 } 1795 EXPORT_SYMBOL(register_netdevice_notifier); 1796 1797 /** 1798 * unregister_netdevice_notifier - unregister a network notifier block 1799 * @nb: notifier 1800 * 1801 * Unregister a notifier previously registered by 1802 * register_netdevice_notifier(). The notifier is unlinked into the 1803 * kernel structures and may then be reused. A negative errno code 1804 * is returned on a failure. 1805 * 1806 * After unregistering unregister and down device events are synthesized 1807 * for all devices on the device list to the removed notifier to remove 1808 * the need for special case cleanup code. 1809 */ 1810 1811 int unregister_netdevice_notifier(struct notifier_block *nb) 1812 { 1813 struct net *net; 1814 int err; 1815 1816 /* Close race with setup_net() and cleanup_net() */ 1817 down_write(&pernet_ops_rwsem); 1818 rtnl_lock(); 1819 err = raw_notifier_chain_unregister(&netdev_chain, nb); 1820 if (err) 1821 goto unlock; 1822 1823 for_each_net(net) 1824 call_netdevice_unregister_net_notifiers(nb, net); 1825 1826 unlock: 1827 rtnl_unlock(); 1828 up_write(&pernet_ops_rwsem); 1829 return err; 1830 } 1831 EXPORT_SYMBOL(unregister_netdevice_notifier); 1832 1833 static int __register_netdevice_notifier_net(struct net *net, 1834 struct notifier_block *nb, 1835 bool ignore_call_fail) 1836 { 1837 int err; 1838 1839 err = raw_notifier_chain_register(&net->netdev_chain, nb); 1840 if (err) 1841 return err; 1842 if (dev_boot_phase) 1843 return 0; 1844 1845 err = call_netdevice_register_net_notifiers(nb, net); 1846 if (err && !ignore_call_fail) 1847 goto chain_unregister; 1848 1849 return 0; 1850 1851 chain_unregister: 1852 raw_notifier_chain_unregister(&net->netdev_chain, nb); 1853 return err; 1854 } 1855 1856 static int __unregister_netdevice_notifier_net(struct net *net, 1857 struct notifier_block *nb) 1858 { 1859 int err; 1860 1861 err = raw_notifier_chain_unregister(&net->netdev_chain, nb); 1862 if (err) 1863 return err; 1864 1865 call_netdevice_unregister_net_notifiers(nb, net); 1866 return 0; 1867 } 1868 1869 /** 1870 * register_netdevice_notifier_net - register a per-netns network notifier block 1871 * @net: network namespace 1872 * @nb: notifier 1873 * 1874 * Register a notifier to be called when network device events occur. 1875 * The notifier passed is linked into the kernel structures and must 1876 * not be reused until it has been unregistered. A negative errno code 1877 * is returned on a failure. 1878 * 1879 * When registered all registration and up events are replayed 1880 * to the new notifier to allow device to have a race free 1881 * view of the network device list. 1882 */ 1883 1884 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb) 1885 { 1886 int err; 1887 1888 rtnl_lock(); 1889 err = __register_netdevice_notifier_net(net, nb, false); 1890 rtnl_unlock(); 1891 return err; 1892 } 1893 EXPORT_SYMBOL(register_netdevice_notifier_net); 1894 1895 /** 1896 * unregister_netdevice_notifier_net - unregister a per-netns 1897 * network notifier block 1898 * @net: network namespace 1899 * @nb: notifier 1900 * 1901 * Unregister a notifier previously registered by 1902 * register_netdevice_notifier_net(). The notifier is unlinked from the 1903 * kernel structures and may then be reused. A negative errno code 1904 * is returned on a failure. 1905 * 1906 * After unregistering unregister and down device events are synthesized 1907 * for all devices on the device list to the removed notifier to remove 1908 * the need for special case cleanup code. 1909 */ 1910 1911 int unregister_netdevice_notifier_net(struct net *net, 1912 struct notifier_block *nb) 1913 { 1914 int err; 1915 1916 rtnl_lock(); 1917 err = __unregister_netdevice_notifier_net(net, nb); 1918 rtnl_unlock(); 1919 return err; 1920 } 1921 EXPORT_SYMBOL(unregister_netdevice_notifier_net); 1922 1923 static void __move_netdevice_notifier_net(struct net *src_net, 1924 struct net *dst_net, 1925 struct notifier_block *nb) 1926 { 1927 __unregister_netdevice_notifier_net(src_net, nb); 1928 __register_netdevice_notifier_net(dst_net, nb, true); 1929 } 1930 1931 int register_netdevice_notifier_dev_net(struct net_device *dev, 1932 struct notifier_block *nb, 1933 struct netdev_net_notifier *nn) 1934 { 1935 int err; 1936 1937 rtnl_lock(); 1938 err = __register_netdevice_notifier_net(dev_net(dev), nb, false); 1939 if (!err) { 1940 nn->nb = nb; 1941 list_add(&nn->list, &dev->net_notifier_list); 1942 } 1943 rtnl_unlock(); 1944 return err; 1945 } 1946 EXPORT_SYMBOL(register_netdevice_notifier_dev_net); 1947 1948 int unregister_netdevice_notifier_dev_net(struct net_device *dev, 1949 struct notifier_block *nb, 1950 struct netdev_net_notifier *nn) 1951 { 1952 int err; 1953 1954 rtnl_lock(); 1955 list_del(&nn->list); 1956 err = __unregister_netdevice_notifier_net(dev_net(dev), nb); 1957 rtnl_unlock(); 1958 return err; 1959 } 1960 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net); 1961 1962 static void move_netdevice_notifiers_dev_net(struct net_device *dev, 1963 struct net *net) 1964 { 1965 struct netdev_net_notifier *nn; 1966 1967 list_for_each_entry(nn, &dev->net_notifier_list, list) 1968 __move_netdevice_notifier_net(dev_net(dev), net, nn->nb); 1969 } 1970 1971 /** 1972 * call_netdevice_notifiers_info - call all network notifier blocks 1973 * @val: value passed unmodified to notifier function 1974 * @info: notifier information data 1975 * 1976 * Call all network notifier blocks. Parameters and return value 1977 * are as for raw_notifier_call_chain(). 1978 */ 1979 1980 int call_netdevice_notifiers_info(unsigned long val, 1981 struct netdev_notifier_info *info) 1982 { 1983 struct net *net = dev_net(info->dev); 1984 int ret; 1985 1986 ASSERT_RTNL(); 1987 1988 /* Run per-netns notifier block chain first, then run the global one. 1989 * Hopefully, one day, the global one is going to be removed after 1990 * all notifier block registrators get converted to be per-netns. 1991 */ 1992 ret = raw_notifier_call_chain(&net->netdev_chain, val, info); 1993 if (ret & NOTIFY_STOP_MASK) 1994 return ret; 1995 return raw_notifier_call_chain(&netdev_chain, val, info); 1996 } 1997 1998 /** 1999 * call_netdevice_notifiers_info_robust - call per-netns notifier blocks 2000 * for and rollback on error 2001 * @val_up: value passed unmodified to notifier function 2002 * @val_down: value passed unmodified to the notifier function when 2003 * recovering from an error on @val_up 2004 * @info: notifier information data 2005 * 2006 * Call all per-netns network notifier blocks, but not notifier blocks on 2007 * the global notifier chain. Parameters and return value are as for 2008 * raw_notifier_call_chain_robust(). 2009 */ 2010 2011 static int 2012 call_netdevice_notifiers_info_robust(unsigned long val_up, 2013 unsigned long val_down, 2014 struct netdev_notifier_info *info) 2015 { 2016 struct net *net = dev_net(info->dev); 2017 2018 ASSERT_RTNL(); 2019 2020 return raw_notifier_call_chain_robust(&net->netdev_chain, 2021 val_up, val_down, info); 2022 } 2023 2024 static int call_netdevice_notifiers_extack(unsigned long val, 2025 struct net_device *dev, 2026 struct netlink_ext_ack *extack) 2027 { 2028 struct netdev_notifier_info info = { 2029 .dev = dev, 2030 .extack = extack, 2031 }; 2032 2033 return call_netdevice_notifiers_info(val, &info); 2034 } 2035 2036 /** 2037 * call_netdevice_notifiers - call all network notifier blocks 2038 * @val: value passed unmodified to notifier function 2039 * @dev: net_device pointer passed unmodified to notifier function 2040 * 2041 * Call all network notifier blocks. Parameters and return value 2042 * are as for raw_notifier_call_chain(). 2043 */ 2044 2045 int call_netdevice_notifiers(unsigned long val, struct net_device *dev) 2046 { 2047 return call_netdevice_notifiers_extack(val, dev, NULL); 2048 } 2049 EXPORT_SYMBOL(call_netdevice_notifiers); 2050 2051 /** 2052 * call_netdevice_notifiers_mtu - call all network notifier blocks 2053 * @val: value passed unmodified to notifier function 2054 * @dev: net_device pointer passed unmodified to notifier function 2055 * @arg: additional u32 argument passed to the notifier function 2056 * 2057 * Call all network notifier blocks. Parameters and return value 2058 * are as for raw_notifier_call_chain(). 2059 */ 2060 static int call_netdevice_notifiers_mtu(unsigned long val, 2061 struct net_device *dev, u32 arg) 2062 { 2063 struct netdev_notifier_info_ext info = { 2064 .info.dev = dev, 2065 .ext.mtu = arg, 2066 }; 2067 2068 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0); 2069 2070 return call_netdevice_notifiers_info(val, &info.info); 2071 } 2072 2073 #ifdef CONFIG_NET_INGRESS 2074 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key); 2075 2076 void net_inc_ingress_queue(void) 2077 { 2078 static_branch_inc(&ingress_needed_key); 2079 } 2080 EXPORT_SYMBOL_GPL(net_inc_ingress_queue); 2081 2082 void net_dec_ingress_queue(void) 2083 { 2084 static_branch_dec(&ingress_needed_key); 2085 } 2086 EXPORT_SYMBOL_GPL(net_dec_ingress_queue); 2087 #endif 2088 2089 #ifdef CONFIG_NET_EGRESS 2090 static DEFINE_STATIC_KEY_FALSE(egress_needed_key); 2091 2092 void net_inc_egress_queue(void) 2093 { 2094 static_branch_inc(&egress_needed_key); 2095 } 2096 EXPORT_SYMBOL_GPL(net_inc_egress_queue); 2097 2098 void net_dec_egress_queue(void) 2099 { 2100 static_branch_dec(&egress_needed_key); 2101 } 2102 EXPORT_SYMBOL_GPL(net_dec_egress_queue); 2103 #endif 2104 2105 DEFINE_STATIC_KEY_FALSE(netstamp_needed_key); 2106 EXPORT_SYMBOL(netstamp_needed_key); 2107 #ifdef CONFIG_JUMP_LABEL 2108 static atomic_t netstamp_needed_deferred; 2109 static atomic_t netstamp_wanted; 2110 static void netstamp_clear(struct work_struct *work) 2111 { 2112 int deferred = atomic_xchg(&netstamp_needed_deferred, 0); 2113 int wanted; 2114 2115 wanted = atomic_add_return(deferred, &netstamp_wanted); 2116 if (wanted > 0) 2117 static_branch_enable(&netstamp_needed_key); 2118 else 2119 static_branch_disable(&netstamp_needed_key); 2120 } 2121 static DECLARE_WORK(netstamp_work, netstamp_clear); 2122 #endif 2123 2124 void net_enable_timestamp(void) 2125 { 2126 #ifdef CONFIG_JUMP_LABEL 2127 int wanted = atomic_read(&netstamp_wanted); 2128 2129 while (wanted > 0) { 2130 if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted + 1)) 2131 return; 2132 } 2133 atomic_inc(&netstamp_needed_deferred); 2134 schedule_work(&netstamp_work); 2135 #else 2136 static_branch_inc(&netstamp_needed_key); 2137 #endif 2138 } 2139 EXPORT_SYMBOL(net_enable_timestamp); 2140 2141 void net_disable_timestamp(void) 2142 { 2143 #ifdef CONFIG_JUMP_LABEL 2144 int wanted = atomic_read(&netstamp_wanted); 2145 2146 while (wanted > 1) { 2147 if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted - 1)) 2148 return; 2149 } 2150 atomic_dec(&netstamp_needed_deferred); 2151 schedule_work(&netstamp_work); 2152 #else 2153 static_branch_dec(&netstamp_needed_key); 2154 #endif 2155 } 2156 EXPORT_SYMBOL(net_disable_timestamp); 2157 2158 static inline void net_timestamp_set(struct sk_buff *skb) 2159 { 2160 skb->tstamp = 0; 2161 skb->mono_delivery_time = 0; 2162 if (static_branch_unlikely(&netstamp_needed_key)) 2163 skb->tstamp = ktime_get_real(); 2164 } 2165 2166 #define net_timestamp_check(COND, SKB) \ 2167 if (static_branch_unlikely(&netstamp_needed_key)) { \ 2168 if ((COND) && !(SKB)->tstamp) \ 2169 (SKB)->tstamp = ktime_get_real(); \ 2170 } \ 2171 2172 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb) 2173 { 2174 return __is_skb_forwardable(dev, skb, true); 2175 } 2176 EXPORT_SYMBOL_GPL(is_skb_forwardable); 2177 2178 static int __dev_forward_skb2(struct net_device *dev, struct sk_buff *skb, 2179 bool check_mtu) 2180 { 2181 int ret = ____dev_forward_skb(dev, skb, check_mtu); 2182 2183 if (likely(!ret)) { 2184 skb->protocol = eth_type_trans(skb, dev); 2185 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); 2186 } 2187 2188 return ret; 2189 } 2190 2191 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2192 { 2193 return __dev_forward_skb2(dev, skb, true); 2194 } 2195 EXPORT_SYMBOL_GPL(__dev_forward_skb); 2196 2197 /** 2198 * dev_forward_skb - loopback an skb to another netif 2199 * 2200 * @dev: destination network device 2201 * @skb: buffer to forward 2202 * 2203 * return values: 2204 * NET_RX_SUCCESS (no congestion) 2205 * NET_RX_DROP (packet was dropped, but freed) 2206 * 2207 * dev_forward_skb can be used for injecting an skb from the 2208 * start_xmit function of one device into the receive queue 2209 * of another device. 2210 * 2211 * The receiving device may be in another namespace, so 2212 * we have to clear all information in the skb that could 2213 * impact namespace isolation. 2214 */ 2215 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2216 { 2217 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); 2218 } 2219 EXPORT_SYMBOL_GPL(dev_forward_skb); 2220 2221 int dev_forward_skb_nomtu(struct net_device *dev, struct sk_buff *skb) 2222 { 2223 return __dev_forward_skb2(dev, skb, false) ?: netif_rx_internal(skb); 2224 } 2225 2226 static inline int deliver_skb(struct sk_buff *skb, 2227 struct packet_type *pt_prev, 2228 struct net_device *orig_dev) 2229 { 2230 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 2231 return -ENOMEM; 2232 refcount_inc(&skb->users); 2233 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 2234 } 2235 2236 static inline void deliver_ptype_list_skb(struct sk_buff *skb, 2237 struct packet_type **pt, 2238 struct net_device *orig_dev, 2239 __be16 type, 2240 struct list_head *ptype_list) 2241 { 2242 struct packet_type *ptype, *pt_prev = *pt; 2243 2244 list_for_each_entry_rcu(ptype, ptype_list, list) { 2245 if (ptype->type != type) 2246 continue; 2247 if (pt_prev) 2248 deliver_skb(skb, pt_prev, orig_dev); 2249 pt_prev = ptype; 2250 } 2251 *pt = pt_prev; 2252 } 2253 2254 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) 2255 { 2256 if (!ptype->af_packet_priv || !skb->sk) 2257 return false; 2258 2259 if (ptype->id_match) 2260 return ptype->id_match(ptype, skb->sk); 2261 else if ((struct sock *)ptype->af_packet_priv == skb->sk) 2262 return true; 2263 2264 return false; 2265 } 2266 2267 /** 2268 * dev_nit_active - return true if any network interface taps are in use 2269 * 2270 * @dev: network device to check for the presence of taps 2271 */ 2272 bool dev_nit_active(struct net_device *dev) 2273 { 2274 return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all); 2275 } 2276 EXPORT_SYMBOL_GPL(dev_nit_active); 2277 2278 /* 2279 * Support routine. Sends outgoing frames to any network 2280 * taps currently in use. 2281 */ 2282 2283 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) 2284 { 2285 struct packet_type *ptype; 2286 struct sk_buff *skb2 = NULL; 2287 struct packet_type *pt_prev = NULL; 2288 struct list_head *ptype_list = &ptype_all; 2289 2290 rcu_read_lock(); 2291 again: 2292 list_for_each_entry_rcu(ptype, ptype_list, list) { 2293 if (READ_ONCE(ptype->ignore_outgoing)) 2294 continue; 2295 2296 /* Never send packets back to the socket 2297 * they originated from - MvS (miquels@drinkel.ow.org) 2298 */ 2299 if (skb_loop_sk(ptype, skb)) 2300 continue; 2301 2302 if (pt_prev) { 2303 deliver_skb(skb2, pt_prev, skb->dev); 2304 pt_prev = ptype; 2305 continue; 2306 } 2307 2308 /* need to clone skb, done only once */ 2309 skb2 = skb_clone(skb, GFP_ATOMIC); 2310 if (!skb2) 2311 goto out_unlock; 2312 2313 net_timestamp_set(skb2); 2314 2315 /* skb->nh should be correctly 2316 * set by sender, so that the second statement is 2317 * just protection against buggy protocols. 2318 */ 2319 skb_reset_mac_header(skb2); 2320 2321 if (skb_network_header(skb2) < skb2->data || 2322 skb_network_header(skb2) > skb_tail_pointer(skb2)) { 2323 net_crit_ratelimited("protocol %04x is buggy, dev %s\n", 2324 ntohs(skb2->protocol), 2325 dev->name); 2326 skb_reset_network_header(skb2); 2327 } 2328 2329 skb2->transport_header = skb2->network_header; 2330 skb2->pkt_type = PACKET_OUTGOING; 2331 pt_prev = ptype; 2332 } 2333 2334 if (ptype_list == &ptype_all) { 2335 ptype_list = &dev->ptype_all; 2336 goto again; 2337 } 2338 out_unlock: 2339 if (pt_prev) { 2340 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC)) 2341 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); 2342 else 2343 kfree_skb(skb2); 2344 } 2345 rcu_read_unlock(); 2346 } 2347 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit); 2348 2349 /** 2350 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change 2351 * @dev: Network device 2352 * @txq: number of queues available 2353 * 2354 * If real_num_tx_queues is changed the tc mappings may no longer be 2355 * valid. To resolve this verify the tc mapping remains valid and if 2356 * not NULL the mapping. With no priorities mapping to this 2357 * offset/count pair it will no longer be used. In the worst case TC0 2358 * is invalid nothing can be done so disable priority mappings. If is 2359 * expected that drivers will fix this mapping if they can before 2360 * calling netif_set_real_num_tx_queues. 2361 */ 2362 static void netif_setup_tc(struct net_device *dev, unsigned int txq) 2363 { 2364 int i; 2365 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2366 2367 /* If TC0 is invalidated disable TC mapping */ 2368 if (tc->offset + tc->count > txq) { 2369 netdev_warn(dev, "Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); 2370 dev->num_tc = 0; 2371 return; 2372 } 2373 2374 /* Invalidated prio to tc mappings set to TC0 */ 2375 for (i = 1; i < TC_BITMASK + 1; i++) { 2376 int q = netdev_get_prio_tc_map(dev, i); 2377 2378 tc = &dev->tc_to_txq[q]; 2379 if (tc->offset + tc->count > txq) { 2380 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", 2381 i, q); 2382 netdev_set_prio_tc_map(dev, i, 0); 2383 } 2384 } 2385 } 2386 2387 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq) 2388 { 2389 if (dev->num_tc) { 2390 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2391 int i; 2392 2393 /* walk through the TCs and see if it falls into any of them */ 2394 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) { 2395 if ((txq - tc->offset) < tc->count) 2396 return i; 2397 } 2398 2399 /* didn't find it, just return -1 to indicate no match */ 2400 return -1; 2401 } 2402 2403 return 0; 2404 } 2405 EXPORT_SYMBOL(netdev_txq_to_tc); 2406 2407 #ifdef CONFIG_XPS 2408 static struct static_key xps_needed __read_mostly; 2409 static struct static_key xps_rxqs_needed __read_mostly; 2410 static DEFINE_MUTEX(xps_map_mutex); 2411 #define xmap_dereference(P) \ 2412 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) 2413 2414 static bool remove_xps_queue(struct xps_dev_maps *dev_maps, 2415 struct xps_dev_maps *old_maps, int tci, u16 index) 2416 { 2417 struct xps_map *map = NULL; 2418 int pos; 2419 2420 map = xmap_dereference(dev_maps->attr_map[tci]); 2421 if (!map) 2422 return false; 2423 2424 for (pos = map->len; pos--;) { 2425 if (map->queues[pos] != index) 2426 continue; 2427 2428 if (map->len > 1) { 2429 map->queues[pos] = map->queues[--map->len]; 2430 break; 2431 } 2432 2433 if (old_maps) 2434 RCU_INIT_POINTER(old_maps->attr_map[tci], NULL); 2435 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2436 kfree_rcu(map, rcu); 2437 return false; 2438 } 2439 2440 return true; 2441 } 2442 2443 static bool remove_xps_queue_cpu(struct net_device *dev, 2444 struct xps_dev_maps *dev_maps, 2445 int cpu, u16 offset, u16 count) 2446 { 2447 int num_tc = dev_maps->num_tc; 2448 bool active = false; 2449 int tci; 2450 2451 for (tci = cpu * num_tc; num_tc--; tci++) { 2452 int i, j; 2453 2454 for (i = count, j = offset; i--; j++) { 2455 if (!remove_xps_queue(dev_maps, NULL, tci, j)) 2456 break; 2457 } 2458 2459 active |= i < 0; 2460 } 2461 2462 return active; 2463 } 2464 2465 static void reset_xps_maps(struct net_device *dev, 2466 struct xps_dev_maps *dev_maps, 2467 enum xps_map_type type) 2468 { 2469 static_key_slow_dec_cpuslocked(&xps_needed); 2470 if (type == XPS_RXQS) 2471 static_key_slow_dec_cpuslocked(&xps_rxqs_needed); 2472 2473 RCU_INIT_POINTER(dev->xps_maps[type], NULL); 2474 2475 kfree_rcu(dev_maps, rcu); 2476 } 2477 2478 static void clean_xps_maps(struct net_device *dev, enum xps_map_type type, 2479 u16 offset, u16 count) 2480 { 2481 struct xps_dev_maps *dev_maps; 2482 bool active = false; 2483 int i, j; 2484 2485 dev_maps = xmap_dereference(dev->xps_maps[type]); 2486 if (!dev_maps) 2487 return; 2488 2489 for (j = 0; j < dev_maps->nr_ids; j++) 2490 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, count); 2491 if (!active) 2492 reset_xps_maps(dev, dev_maps, type); 2493 2494 if (type == XPS_CPUS) { 2495 for (i = offset + (count - 1); count--; i--) 2496 netdev_queue_numa_node_write( 2497 netdev_get_tx_queue(dev, i), NUMA_NO_NODE); 2498 } 2499 } 2500 2501 static void netif_reset_xps_queues(struct net_device *dev, u16 offset, 2502 u16 count) 2503 { 2504 if (!static_key_false(&xps_needed)) 2505 return; 2506 2507 cpus_read_lock(); 2508 mutex_lock(&xps_map_mutex); 2509 2510 if (static_key_false(&xps_rxqs_needed)) 2511 clean_xps_maps(dev, XPS_RXQS, offset, count); 2512 2513 clean_xps_maps(dev, XPS_CPUS, offset, count); 2514 2515 mutex_unlock(&xps_map_mutex); 2516 cpus_read_unlock(); 2517 } 2518 2519 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) 2520 { 2521 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index); 2522 } 2523 2524 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index, 2525 u16 index, bool is_rxqs_map) 2526 { 2527 struct xps_map *new_map; 2528 int alloc_len = XPS_MIN_MAP_ALLOC; 2529 int i, pos; 2530 2531 for (pos = 0; map && pos < map->len; pos++) { 2532 if (map->queues[pos] != index) 2533 continue; 2534 return map; 2535 } 2536 2537 /* Need to add tx-queue to this CPU's/rx-queue's existing map */ 2538 if (map) { 2539 if (pos < map->alloc_len) 2540 return map; 2541 2542 alloc_len = map->alloc_len * 2; 2543 } 2544 2545 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's 2546 * map 2547 */ 2548 if (is_rxqs_map) 2549 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL); 2550 else 2551 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, 2552 cpu_to_node(attr_index)); 2553 if (!new_map) 2554 return NULL; 2555 2556 for (i = 0; i < pos; i++) 2557 new_map->queues[i] = map->queues[i]; 2558 new_map->alloc_len = alloc_len; 2559 new_map->len = pos; 2560 2561 return new_map; 2562 } 2563 2564 /* Copy xps maps at a given index */ 2565 static void xps_copy_dev_maps(struct xps_dev_maps *dev_maps, 2566 struct xps_dev_maps *new_dev_maps, int index, 2567 int tc, bool skip_tc) 2568 { 2569 int i, tci = index * dev_maps->num_tc; 2570 struct xps_map *map; 2571 2572 /* copy maps belonging to foreign traffic classes */ 2573 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2574 if (i == tc && skip_tc) 2575 continue; 2576 2577 /* fill in the new device map from the old device map */ 2578 map = xmap_dereference(dev_maps->attr_map[tci]); 2579 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2580 } 2581 } 2582 2583 /* Must be called under cpus_read_lock */ 2584 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask, 2585 u16 index, enum xps_map_type type) 2586 { 2587 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL, *old_dev_maps = NULL; 2588 const unsigned long *online_mask = NULL; 2589 bool active = false, copy = false; 2590 int i, j, tci, numa_node_id = -2; 2591 int maps_sz, num_tc = 1, tc = 0; 2592 struct xps_map *map, *new_map; 2593 unsigned int nr_ids; 2594 2595 WARN_ON_ONCE(index >= dev->num_tx_queues); 2596 2597 if (dev->num_tc) { 2598 /* Do not allow XPS on subordinate device directly */ 2599 num_tc = dev->num_tc; 2600 if (num_tc < 0) 2601 return -EINVAL; 2602 2603 /* If queue belongs to subordinate dev use its map */ 2604 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev; 2605 2606 tc = netdev_txq_to_tc(dev, index); 2607 if (tc < 0) 2608 return -EINVAL; 2609 } 2610 2611 mutex_lock(&xps_map_mutex); 2612 2613 dev_maps = xmap_dereference(dev->xps_maps[type]); 2614 if (type == XPS_RXQS) { 2615 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues); 2616 nr_ids = dev->num_rx_queues; 2617 } else { 2618 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc); 2619 if (num_possible_cpus() > 1) 2620 online_mask = cpumask_bits(cpu_online_mask); 2621 nr_ids = nr_cpu_ids; 2622 } 2623 2624 if (maps_sz < L1_CACHE_BYTES) 2625 maps_sz = L1_CACHE_BYTES; 2626 2627 /* The old dev_maps could be larger or smaller than the one we're 2628 * setting up now, as dev->num_tc or nr_ids could have been updated in 2629 * between. We could try to be smart, but let's be safe instead and only 2630 * copy foreign traffic classes if the two map sizes match. 2631 */ 2632 if (dev_maps && 2633 dev_maps->num_tc == num_tc && dev_maps->nr_ids == nr_ids) 2634 copy = true; 2635 2636 /* allocate memory for queue storage */ 2637 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids), 2638 j < nr_ids;) { 2639 if (!new_dev_maps) { 2640 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); 2641 if (!new_dev_maps) { 2642 mutex_unlock(&xps_map_mutex); 2643 return -ENOMEM; 2644 } 2645 2646 new_dev_maps->nr_ids = nr_ids; 2647 new_dev_maps->num_tc = num_tc; 2648 } 2649 2650 tci = j * num_tc + tc; 2651 map = copy ? xmap_dereference(dev_maps->attr_map[tci]) : NULL; 2652 2653 map = expand_xps_map(map, j, index, type == XPS_RXQS); 2654 if (!map) 2655 goto error; 2656 2657 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2658 } 2659 2660 if (!new_dev_maps) 2661 goto out_no_new_maps; 2662 2663 if (!dev_maps) { 2664 /* Increment static keys at most once per type */ 2665 static_key_slow_inc_cpuslocked(&xps_needed); 2666 if (type == XPS_RXQS) 2667 static_key_slow_inc_cpuslocked(&xps_rxqs_needed); 2668 } 2669 2670 for (j = 0; j < nr_ids; j++) { 2671 bool skip_tc = false; 2672 2673 tci = j * num_tc + tc; 2674 if (netif_attr_test_mask(j, mask, nr_ids) && 2675 netif_attr_test_online(j, online_mask, nr_ids)) { 2676 /* add tx-queue to CPU/rx-queue maps */ 2677 int pos = 0; 2678 2679 skip_tc = true; 2680 2681 map = xmap_dereference(new_dev_maps->attr_map[tci]); 2682 while ((pos < map->len) && (map->queues[pos] != index)) 2683 pos++; 2684 2685 if (pos == map->len) 2686 map->queues[map->len++] = index; 2687 #ifdef CONFIG_NUMA 2688 if (type == XPS_CPUS) { 2689 if (numa_node_id == -2) 2690 numa_node_id = cpu_to_node(j); 2691 else if (numa_node_id != cpu_to_node(j)) 2692 numa_node_id = -1; 2693 } 2694 #endif 2695 } 2696 2697 if (copy) 2698 xps_copy_dev_maps(dev_maps, new_dev_maps, j, tc, 2699 skip_tc); 2700 } 2701 2702 rcu_assign_pointer(dev->xps_maps[type], new_dev_maps); 2703 2704 /* Cleanup old maps */ 2705 if (!dev_maps) 2706 goto out_no_old_maps; 2707 2708 for (j = 0; j < dev_maps->nr_ids; j++) { 2709 for (i = num_tc, tci = j * dev_maps->num_tc; i--; tci++) { 2710 map = xmap_dereference(dev_maps->attr_map[tci]); 2711 if (!map) 2712 continue; 2713 2714 if (copy) { 2715 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2716 if (map == new_map) 2717 continue; 2718 } 2719 2720 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2721 kfree_rcu(map, rcu); 2722 } 2723 } 2724 2725 old_dev_maps = dev_maps; 2726 2727 out_no_old_maps: 2728 dev_maps = new_dev_maps; 2729 active = true; 2730 2731 out_no_new_maps: 2732 if (type == XPS_CPUS) 2733 /* update Tx queue numa node */ 2734 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), 2735 (numa_node_id >= 0) ? 2736 numa_node_id : NUMA_NO_NODE); 2737 2738 if (!dev_maps) 2739 goto out_no_maps; 2740 2741 /* removes tx-queue from unused CPUs/rx-queues */ 2742 for (j = 0; j < dev_maps->nr_ids; j++) { 2743 tci = j * dev_maps->num_tc; 2744 2745 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2746 if (i == tc && 2747 netif_attr_test_mask(j, mask, dev_maps->nr_ids) && 2748 netif_attr_test_online(j, online_mask, dev_maps->nr_ids)) 2749 continue; 2750 2751 active |= remove_xps_queue(dev_maps, 2752 copy ? old_dev_maps : NULL, 2753 tci, index); 2754 } 2755 } 2756 2757 if (old_dev_maps) 2758 kfree_rcu(old_dev_maps, rcu); 2759 2760 /* free map if not active */ 2761 if (!active) 2762 reset_xps_maps(dev, dev_maps, type); 2763 2764 out_no_maps: 2765 mutex_unlock(&xps_map_mutex); 2766 2767 return 0; 2768 error: 2769 /* remove any maps that we added */ 2770 for (j = 0; j < nr_ids; j++) { 2771 for (i = num_tc, tci = j * num_tc; i--; tci++) { 2772 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2773 map = copy ? 2774 xmap_dereference(dev_maps->attr_map[tci]) : 2775 NULL; 2776 if (new_map && new_map != map) 2777 kfree(new_map); 2778 } 2779 } 2780 2781 mutex_unlock(&xps_map_mutex); 2782 2783 kfree(new_dev_maps); 2784 return -ENOMEM; 2785 } 2786 EXPORT_SYMBOL_GPL(__netif_set_xps_queue); 2787 2788 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, 2789 u16 index) 2790 { 2791 int ret; 2792 2793 cpus_read_lock(); 2794 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, XPS_CPUS); 2795 cpus_read_unlock(); 2796 2797 return ret; 2798 } 2799 EXPORT_SYMBOL(netif_set_xps_queue); 2800 2801 #endif 2802 static void netdev_unbind_all_sb_channels(struct net_device *dev) 2803 { 2804 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2805 2806 /* Unbind any subordinate channels */ 2807 while (txq-- != &dev->_tx[0]) { 2808 if (txq->sb_dev) 2809 netdev_unbind_sb_channel(dev, txq->sb_dev); 2810 } 2811 } 2812 2813 void netdev_reset_tc(struct net_device *dev) 2814 { 2815 #ifdef CONFIG_XPS 2816 netif_reset_xps_queues_gt(dev, 0); 2817 #endif 2818 netdev_unbind_all_sb_channels(dev); 2819 2820 /* Reset TC configuration of device */ 2821 dev->num_tc = 0; 2822 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq)); 2823 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map)); 2824 } 2825 EXPORT_SYMBOL(netdev_reset_tc); 2826 2827 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset) 2828 { 2829 if (tc >= dev->num_tc) 2830 return -EINVAL; 2831 2832 #ifdef CONFIG_XPS 2833 netif_reset_xps_queues(dev, offset, count); 2834 #endif 2835 dev->tc_to_txq[tc].count = count; 2836 dev->tc_to_txq[tc].offset = offset; 2837 return 0; 2838 } 2839 EXPORT_SYMBOL(netdev_set_tc_queue); 2840 2841 int netdev_set_num_tc(struct net_device *dev, u8 num_tc) 2842 { 2843 if (num_tc > TC_MAX_QUEUE) 2844 return -EINVAL; 2845 2846 #ifdef CONFIG_XPS 2847 netif_reset_xps_queues_gt(dev, 0); 2848 #endif 2849 netdev_unbind_all_sb_channels(dev); 2850 2851 dev->num_tc = num_tc; 2852 return 0; 2853 } 2854 EXPORT_SYMBOL(netdev_set_num_tc); 2855 2856 void netdev_unbind_sb_channel(struct net_device *dev, 2857 struct net_device *sb_dev) 2858 { 2859 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2860 2861 #ifdef CONFIG_XPS 2862 netif_reset_xps_queues_gt(sb_dev, 0); 2863 #endif 2864 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq)); 2865 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map)); 2866 2867 while (txq-- != &dev->_tx[0]) { 2868 if (txq->sb_dev == sb_dev) 2869 txq->sb_dev = NULL; 2870 } 2871 } 2872 EXPORT_SYMBOL(netdev_unbind_sb_channel); 2873 2874 int netdev_bind_sb_channel_queue(struct net_device *dev, 2875 struct net_device *sb_dev, 2876 u8 tc, u16 count, u16 offset) 2877 { 2878 /* Make certain the sb_dev and dev are already configured */ 2879 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc) 2880 return -EINVAL; 2881 2882 /* We cannot hand out queues we don't have */ 2883 if ((offset + count) > dev->real_num_tx_queues) 2884 return -EINVAL; 2885 2886 /* Record the mapping */ 2887 sb_dev->tc_to_txq[tc].count = count; 2888 sb_dev->tc_to_txq[tc].offset = offset; 2889 2890 /* Provide a way for Tx queue to find the tc_to_txq map or 2891 * XPS map for itself. 2892 */ 2893 while (count--) 2894 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev; 2895 2896 return 0; 2897 } 2898 EXPORT_SYMBOL(netdev_bind_sb_channel_queue); 2899 2900 int netdev_set_sb_channel(struct net_device *dev, u16 channel) 2901 { 2902 /* Do not use a multiqueue device to represent a subordinate channel */ 2903 if (netif_is_multiqueue(dev)) 2904 return -ENODEV; 2905 2906 /* We allow channels 1 - 32767 to be used for subordinate channels. 2907 * Channel 0 is meant to be "native" mode and used only to represent 2908 * the main root device. We allow writing 0 to reset the device back 2909 * to normal mode after being used as a subordinate channel. 2910 */ 2911 if (channel > S16_MAX) 2912 return -EINVAL; 2913 2914 dev->num_tc = -channel; 2915 2916 return 0; 2917 } 2918 EXPORT_SYMBOL(netdev_set_sb_channel); 2919 2920 /* 2921 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues 2922 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed. 2923 */ 2924 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) 2925 { 2926 bool disabling; 2927 int rc; 2928 2929 disabling = txq < dev->real_num_tx_queues; 2930 2931 if (txq < 1 || txq > dev->num_tx_queues) 2932 return -EINVAL; 2933 2934 if (dev->reg_state == NETREG_REGISTERED || 2935 dev->reg_state == NETREG_UNREGISTERING) { 2936 ASSERT_RTNL(); 2937 2938 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, 2939 txq); 2940 if (rc) 2941 return rc; 2942 2943 if (dev->num_tc) 2944 netif_setup_tc(dev, txq); 2945 2946 dev_qdisc_change_real_num_tx(dev, txq); 2947 2948 dev->real_num_tx_queues = txq; 2949 2950 if (disabling) { 2951 synchronize_net(); 2952 qdisc_reset_all_tx_gt(dev, txq); 2953 #ifdef CONFIG_XPS 2954 netif_reset_xps_queues_gt(dev, txq); 2955 #endif 2956 } 2957 } else { 2958 dev->real_num_tx_queues = txq; 2959 } 2960 2961 return 0; 2962 } 2963 EXPORT_SYMBOL(netif_set_real_num_tx_queues); 2964 2965 #ifdef CONFIG_SYSFS 2966 /** 2967 * netif_set_real_num_rx_queues - set actual number of RX queues used 2968 * @dev: Network device 2969 * @rxq: Actual number of RX queues 2970 * 2971 * This must be called either with the rtnl_lock held or before 2972 * registration of the net device. Returns 0 on success, or a 2973 * negative error code. If called before registration, it always 2974 * succeeds. 2975 */ 2976 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) 2977 { 2978 int rc; 2979 2980 if (rxq < 1 || rxq > dev->num_rx_queues) 2981 return -EINVAL; 2982 2983 if (dev->reg_state == NETREG_REGISTERED) { 2984 ASSERT_RTNL(); 2985 2986 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, 2987 rxq); 2988 if (rc) 2989 return rc; 2990 } 2991 2992 dev->real_num_rx_queues = rxq; 2993 return 0; 2994 } 2995 EXPORT_SYMBOL(netif_set_real_num_rx_queues); 2996 #endif 2997 2998 /** 2999 * netif_set_real_num_queues - set actual number of RX and TX queues used 3000 * @dev: Network device 3001 * @txq: Actual number of TX queues 3002 * @rxq: Actual number of RX queues 3003 * 3004 * Set the real number of both TX and RX queues. 3005 * Does nothing if the number of queues is already correct. 3006 */ 3007 int netif_set_real_num_queues(struct net_device *dev, 3008 unsigned int txq, unsigned int rxq) 3009 { 3010 unsigned int old_rxq = dev->real_num_rx_queues; 3011 int err; 3012 3013 if (txq < 1 || txq > dev->num_tx_queues || 3014 rxq < 1 || rxq > dev->num_rx_queues) 3015 return -EINVAL; 3016 3017 /* Start from increases, so the error path only does decreases - 3018 * decreases can't fail. 3019 */ 3020 if (rxq > dev->real_num_rx_queues) { 3021 err = netif_set_real_num_rx_queues(dev, rxq); 3022 if (err) 3023 return err; 3024 } 3025 if (txq > dev->real_num_tx_queues) { 3026 err = netif_set_real_num_tx_queues(dev, txq); 3027 if (err) 3028 goto undo_rx; 3029 } 3030 if (rxq < dev->real_num_rx_queues) 3031 WARN_ON(netif_set_real_num_rx_queues(dev, rxq)); 3032 if (txq < dev->real_num_tx_queues) 3033 WARN_ON(netif_set_real_num_tx_queues(dev, txq)); 3034 3035 return 0; 3036 undo_rx: 3037 WARN_ON(netif_set_real_num_rx_queues(dev, old_rxq)); 3038 return err; 3039 } 3040 EXPORT_SYMBOL(netif_set_real_num_queues); 3041 3042 /** 3043 * netif_set_tso_max_size() - set the max size of TSO frames supported 3044 * @dev: netdev to update 3045 * @size: max skb->len of a TSO frame 3046 * 3047 * Set the limit on the size of TSO super-frames the device can handle. 3048 * Unless explicitly set the stack will assume the value of 3049 * %GSO_LEGACY_MAX_SIZE. 3050 */ 3051 void netif_set_tso_max_size(struct net_device *dev, unsigned int size) 3052 { 3053 dev->tso_max_size = min(GSO_MAX_SIZE, size); 3054 if (size < READ_ONCE(dev->gso_max_size)) 3055 netif_set_gso_max_size(dev, size); 3056 if (size < READ_ONCE(dev->gso_ipv4_max_size)) 3057 netif_set_gso_ipv4_max_size(dev, size); 3058 } 3059 EXPORT_SYMBOL(netif_set_tso_max_size); 3060 3061 /** 3062 * netif_set_tso_max_segs() - set the max number of segs supported for TSO 3063 * @dev: netdev to update 3064 * @segs: max number of TCP segments 3065 * 3066 * Set the limit on the number of TCP segments the device can generate from 3067 * a single TSO super-frame. 3068 * Unless explicitly set the stack will assume the value of %GSO_MAX_SEGS. 3069 */ 3070 void netif_set_tso_max_segs(struct net_device *dev, unsigned int segs) 3071 { 3072 dev->tso_max_segs = segs; 3073 if (segs < READ_ONCE(dev->gso_max_segs)) 3074 netif_set_gso_max_segs(dev, segs); 3075 } 3076 EXPORT_SYMBOL(netif_set_tso_max_segs); 3077 3078 /** 3079 * netif_inherit_tso_max() - copy all TSO limits from a lower device to an upper 3080 * @to: netdev to update 3081 * @from: netdev from which to copy the limits 3082 */ 3083 void netif_inherit_tso_max(struct net_device *to, const struct net_device *from) 3084 { 3085 netif_set_tso_max_size(to, from->tso_max_size); 3086 netif_set_tso_max_segs(to, from->tso_max_segs); 3087 } 3088 EXPORT_SYMBOL(netif_inherit_tso_max); 3089 3090 /** 3091 * netif_get_num_default_rss_queues - default number of RSS queues 3092 * 3093 * Default value is the number of physical cores if there are only 1 or 2, or 3094 * divided by 2 if there are more. 3095 */ 3096 int netif_get_num_default_rss_queues(void) 3097 { 3098 cpumask_var_t cpus; 3099 int cpu, count = 0; 3100 3101 if (unlikely(is_kdump_kernel() || !zalloc_cpumask_var(&cpus, GFP_KERNEL))) 3102 return 1; 3103 3104 cpumask_copy(cpus, cpu_online_mask); 3105 for_each_cpu(cpu, cpus) { 3106 ++count; 3107 cpumask_andnot(cpus, cpus, topology_sibling_cpumask(cpu)); 3108 } 3109 free_cpumask_var(cpus); 3110 3111 return count > 2 ? DIV_ROUND_UP(count, 2) : count; 3112 } 3113 EXPORT_SYMBOL(netif_get_num_default_rss_queues); 3114 3115 static void __netif_reschedule(struct Qdisc *q) 3116 { 3117 struct softnet_data *sd; 3118 unsigned long flags; 3119 3120 local_irq_save(flags); 3121 sd = this_cpu_ptr(&softnet_data); 3122 q->next_sched = NULL; 3123 *sd->output_queue_tailp = q; 3124 sd->output_queue_tailp = &q->next_sched; 3125 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3126 local_irq_restore(flags); 3127 } 3128 3129 void __netif_schedule(struct Qdisc *q) 3130 { 3131 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) 3132 __netif_reschedule(q); 3133 } 3134 EXPORT_SYMBOL(__netif_schedule); 3135 3136 struct dev_kfree_skb_cb { 3137 enum skb_drop_reason reason; 3138 }; 3139 3140 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) 3141 { 3142 return (struct dev_kfree_skb_cb *)skb->cb; 3143 } 3144 3145 void netif_schedule_queue(struct netdev_queue *txq) 3146 { 3147 rcu_read_lock(); 3148 if (!netif_xmit_stopped(txq)) { 3149 struct Qdisc *q = rcu_dereference(txq->qdisc); 3150 3151 __netif_schedule(q); 3152 } 3153 rcu_read_unlock(); 3154 } 3155 EXPORT_SYMBOL(netif_schedule_queue); 3156 3157 void netif_tx_wake_queue(struct netdev_queue *dev_queue) 3158 { 3159 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { 3160 struct Qdisc *q; 3161 3162 rcu_read_lock(); 3163 q = rcu_dereference(dev_queue->qdisc); 3164 __netif_schedule(q); 3165 rcu_read_unlock(); 3166 } 3167 } 3168 EXPORT_SYMBOL(netif_tx_wake_queue); 3169 3170 void dev_kfree_skb_irq_reason(struct sk_buff *skb, enum skb_drop_reason reason) 3171 { 3172 unsigned long flags; 3173 3174 if (unlikely(!skb)) 3175 return; 3176 3177 if (likely(refcount_read(&skb->users) == 1)) { 3178 smp_rmb(); 3179 refcount_set(&skb->users, 0); 3180 } else if (likely(!refcount_dec_and_test(&skb->users))) { 3181 return; 3182 } 3183 get_kfree_skb_cb(skb)->reason = reason; 3184 local_irq_save(flags); 3185 skb->next = __this_cpu_read(softnet_data.completion_queue); 3186 __this_cpu_write(softnet_data.completion_queue, skb); 3187 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3188 local_irq_restore(flags); 3189 } 3190 EXPORT_SYMBOL(dev_kfree_skb_irq_reason); 3191 3192 void dev_kfree_skb_any_reason(struct sk_buff *skb, enum skb_drop_reason reason) 3193 { 3194 if (in_hardirq() || irqs_disabled()) 3195 dev_kfree_skb_irq_reason(skb, reason); 3196 else 3197 kfree_skb_reason(skb, reason); 3198 } 3199 EXPORT_SYMBOL(dev_kfree_skb_any_reason); 3200 3201 3202 /** 3203 * netif_device_detach - mark device as removed 3204 * @dev: network device 3205 * 3206 * Mark device as removed from system and therefore no longer available. 3207 */ 3208 void netif_device_detach(struct net_device *dev) 3209 { 3210 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && 3211 netif_running(dev)) { 3212 netif_tx_stop_all_queues(dev); 3213 } 3214 } 3215 EXPORT_SYMBOL(netif_device_detach); 3216 3217 /** 3218 * netif_device_attach - mark device as attached 3219 * @dev: network device 3220 * 3221 * Mark device as attached from system and restart if needed. 3222 */ 3223 void netif_device_attach(struct net_device *dev) 3224 { 3225 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && 3226 netif_running(dev)) { 3227 netif_tx_wake_all_queues(dev); 3228 __netdev_watchdog_up(dev); 3229 } 3230 } 3231 EXPORT_SYMBOL(netif_device_attach); 3232 3233 /* 3234 * Returns a Tx hash based on the given packet descriptor a Tx queues' number 3235 * to be used as a distribution range. 3236 */ 3237 static u16 skb_tx_hash(const struct net_device *dev, 3238 const struct net_device *sb_dev, 3239 struct sk_buff *skb) 3240 { 3241 u32 hash; 3242 u16 qoffset = 0; 3243 u16 qcount = dev->real_num_tx_queues; 3244 3245 if (dev->num_tc) { 3246 u8 tc = netdev_get_prio_tc_map(dev, skb->priority); 3247 3248 qoffset = sb_dev->tc_to_txq[tc].offset; 3249 qcount = sb_dev->tc_to_txq[tc].count; 3250 if (unlikely(!qcount)) { 3251 net_warn_ratelimited("%s: invalid qcount, qoffset %u for tc %u\n", 3252 sb_dev->name, qoffset, tc); 3253 qoffset = 0; 3254 qcount = dev->real_num_tx_queues; 3255 } 3256 } 3257 3258 if (skb_rx_queue_recorded(skb)) { 3259 DEBUG_NET_WARN_ON_ONCE(qcount == 0); 3260 hash = skb_get_rx_queue(skb); 3261 if (hash >= qoffset) 3262 hash -= qoffset; 3263 while (unlikely(hash >= qcount)) 3264 hash -= qcount; 3265 return hash + qoffset; 3266 } 3267 3268 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; 3269 } 3270 3271 void skb_warn_bad_offload(const struct sk_buff *skb) 3272 { 3273 static const netdev_features_t null_features; 3274 struct net_device *dev = skb->dev; 3275 const char *name = ""; 3276 3277 if (!net_ratelimit()) 3278 return; 3279 3280 if (dev) { 3281 if (dev->dev.parent) 3282 name = dev_driver_string(dev->dev.parent); 3283 else 3284 name = netdev_name(dev); 3285 } 3286 skb_dump(KERN_WARNING, skb, false); 3287 WARN(1, "%s: caps=(%pNF, %pNF)\n", 3288 name, dev ? &dev->features : &null_features, 3289 skb->sk ? &skb->sk->sk_route_caps : &null_features); 3290 } 3291 3292 /* 3293 * Invalidate hardware checksum when packet is to be mangled, and 3294 * complete checksum manually on outgoing path. 3295 */ 3296 int skb_checksum_help(struct sk_buff *skb) 3297 { 3298 __wsum csum; 3299 int ret = 0, offset; 3300 3301 if (skb->ip_summed == CHECKSUM_COMPLETE) 3302 goto out_set_summed; 3303 3304 if (unlikely(skb_is_gso(skb))) { 3305 skb_warn_bad_offload(skb); 3306 return -EINVAL; 3307 } 3308 3309 /* Before computing a checksum, we should make sure no frag could 3310 * be modified by an external entity : checksum could be wrong. 3311 */ 3312 if (skb_has_shared_frag(skb)) { 3313 ret = __skb_linearize(skb); 3314 if (ret) 3315 goto out; 3316 } 3317 3318 offset = skb_checksum_start_offset(skb); 3319 ret = -EINVAL; 3320 if (unlikely(offset >= skb_headlen(skb))) { 3321 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false); 3322 WARN_ONCE(true, "offset (%d) >= skb_headlen() (%u)\n", 3323 offset, skb_headlen(skb)); 3324 goto out; 3325 } 3326 csum = skb_checksum(skb, offset, skb->len - offset, 0); 3327 3328 offset += skb->csum_offset; 3329 if (unlikely(offset + sizeof(__sum16) > skb_headlen(skb))) { 3330 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false); 3331 WARN_ONCE(true, "offset+2 (%zu) > skb_headlen() (%u)\n", 3332 offset + sizeof(__sum16), skb_headlen(skb)); 3333 goto out; 3334 } 3335 ret = skb_ensure_writable(skb, offset + sizeof(__sum16)); 3336 if (ret) 3337 goto out; 3338 3339 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0; 3340 out_set_summed: 3341 skb->ip_summed = CHECKSUM_NONE; 3342 out: 3343 return ret; 3344 } 3345 EXPORT_SYMBOL(skb_checksum_help); 3346 3347 int skb_crc32c_csum_help(struct sk_buff *skb) 3348 { 3349 __le32 crc32c_csum; 3350 int ret = 0, offset, start; 3351 3352 if (skb->ip_summed != CHECKSUM_PARTIAL) 3353 goto out; 3354 3355 if (unlikely(skb_is_gso(skb))) 3356 goto out; 3357 3358 /* Before computing a checksum, we should make sure no frag could 3359 * be modified by an external entity : checksum could be wrong. 3360 */ 3361 if (unlikely(skb_has_shared_frag(skb))) { 3362 ret = __skb_linearize(skb); 3363 if (ret) 3364 goto out; 3365 } 3366 start = skb_checksum_start_offset(skb); 3367 offset = start + offsetof(struct sctphdr, checksum); 3368 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { 3369 ret = -EINVAL; 3370 goto out; 3371 } 3372 3373 ret = skb_ensure_writable(skb, offset + sizeof(__le32)); 3374 if (ret) 3375 goto out; 3376 3377 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start, 3378 skb->len - start, ~(__u32)0, 3379 crc32c_csum_stub)); 3380 *(__le32 *)(skb->data + offset) = crc32c_csum; 3381 skb_reset_csum_not_inet(skb); 3382 out: 3383 return ret; 3384 } 3385 3386 __be16 skb_network_protocol(struct sk_buff *skb, int *depth) 3387 { 3388 __be16 type = skb->protocol; 3389 3390 /* Tunnel gso handlers can set protocol to ethernet. */ 3391 if (type == htons(ETH_P_TEB)) { 3392 struct ethhdr *eth; 3393 3394 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) 3395 return 0; 3396 3397 eth = (struct ethhdr *)skb->data; 3398 type = eth->h_proto; 3399 } 3400 3401 return vlan_get_protocol_and_depth(skb, type, depth); 3402 } 3403 3404 3405 /* Take action when hardware reception checksum errors are detected. */ 3406 #ifdef CONFIG_BUG 3407 static void do_netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3408 { 3409 netdev_err(dev, "hw csum failure\n"); 3410 skb_dump(KERN_ERR, skb, true); 3411 dump_stack(); 3412 } 3413 3414 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3415 { 3416 DO_ONCE_LITE(do_netdev_rx_csum_fault, dev, skb); 3417 } 3418 EXPORT_SYMBOL(netdev_rx_csum_fault); 3419 #endif 3420 3421 /* XXX: check that highmem exists at all on the given machine. */ 3422 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) 3423 { 3424 #ifdef CONFIG_HIGHMEM 3425 int i; 3426 3427 if (!(dev->features & NETIF_F_HIGHDMA)) { 3428 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 3429 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 3430 3431 if (PageHighMem(skb_frag_page(frag))) 3432 return 1; 3433 } 3434 } 3435 #endif 3436 return 0; 3437 } 3438 3439 /* If MPLS offload request, verify we are testing hardware MPLS features 3440 * instead of standard features for the netdev. 3441 */ 3442 #if IS_ENABLED(CONFIG_NET_MPLS_GSO) 3443 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3444 netdev_features_t features, 3445 __be16 type) 3446 { 3447 if (eth_p_mpls(type)) 3448 features &= skb->dev->mpls_features; 3449 3450 return features; 3451 } 3452 #else 3453 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3454 netdev_features_t features, 3455 __be16 type) 3456 { 3457 return features; 3458 } 3459 #endif 3460 3461 static netdev_features_t harmonize_features(struct sk_buff *skb, 3462 netdev_features_t features) 3463 { 3464 __be16 type; 3465 3466 type = skb_network_protocol(skb, NULL); 3467 features = net_mpls_features(skb, features, type); 3468 3469 if (skb->ip_summed != CHECKSUM_NONE && 3470 !can_checksum_protocol(features, type)) { 3471 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); 3472 } 3473 if (illegal_highdma(skb->dev, skb)) 3474 features &= ~NETIF_F_SG; 3475 3476 return features; 3477 } 3478 3479 netdev_features_t passthru_features_check(struct sk_buff *skb, 3480 struct net_device *dev, 3481 netdev_features_t features) 3482 { 3483 return features; 3484 } 3485 EXPORT_SYMBOL(passthru_features_check); 3486 3487 static netdev_features_t dflt_features_check(struct sk_buff *skb, 3488 struct net_device *dev, 3489 netdev_features_t features) 3490 { 3491 return vlan_features_check(skb, features); 3492 } 3493 3494 static netdev_features_t gso_features_check(const struct sk_buff *skb, 3495 struct net_device *dev, 3496 netdev_features_t features) 3497 { 3498 u16 gso_segs = skb_shinfo(skb)->gso_segs; 3499 3500 if (gso_segs > READ_ONCE(dev->gso_max_segs)) 3501 return features & ~NETIF_F_GSO_MASK; 3502 3503 if (unlikely(skb->len >= netif_get_gso_max_size(dev, skb))) 3504 return features & ~NETIF_F_GSO_MASK; 3505 3506 if (!skb_shinfo(skb)->gso_type) { 3507 skb_warn_bad_offload(skb); 3508 return features & ~NETIF_F_GSO_MASK; 3509 } 3510 3511 /* Support for GSO partial features requires software 3512 * intervention before we can actually process the packets 3513 * so we need to strip support for any partial features now 3514 * and we can pull them back in after we have partially 3515 * segmented the frame. 3516 */ 3517 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)) 3518 features &= ~dev->gso_partial_features; 3519 3520 /* Make sure to clear the IPv4 ID mangling feature if the 3521 * IPv4 header has the potential to be fragmented. 3522 */ 3523 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { 3524 struct iphdr *iph = skb->encapsulation ? 3525 inner_ip_hdr(skb) : ip_hdr(skb); 3526 3527 if (!(iph->frag_off & htons(IP_DF))) 3528 features &= ~NETIF_F_TSO_MANGLEID; 3529 } 3530 3531 return features; 3532 } 3533 3534 netdev_features_t netif_skb_features(struct sk_buff *skb) 3535 { 3536 struct net_device *dev = skb->dev; 3537 netdev_features_t features = dev->features; 3538 3539 if (skb_is_gso(skb)) 3540 features = gso_features_check(skb, dev, features); 3541 3542 /* If encapsulation offload request, verify we are testing 3543 * hardware encapsulation features instead of standard 3544 * features for the netdev 3545 */ 3546 if (skb->encapsulation) 3547 features &= dev->hw_enc_features; 3548 3549 if (skb_vlan_tagged(skb)) 3550 features = netdev_intersect_features(features, 3551 dev->vlan_features | 3552 NETIF_F_HW_VLAN_CTAG_TX | 3553 NETIF_F_HW_VLAN_STAG_TX); 3554 3555 if (dev->netdev_ops->ndo_features_check) 3556 features &= dev->netdev_ops->ndo_features_check(skb, dev, 3557 features); 3558 else 3559 features &= dflt_features_check(skb, dev, features); 3560 3561 return harmonize_features(skb, features); 3562 } 3563 EXPORT_SYMBOL(netif_skb_features); 3564 3565 static int xmit_one(struct sk_buff *skb, struct net_device *dev, 3566 struct netdev_queue *txq, bool more) 3567 { 3568 unsigned int len; 3569 int rc; 3570 3571 if (dev_nit_active(dev)) 3572 dev_queue_xmit_nit(skb, dev); 3573 3574 len = skb->len; 3575 trace_net_dev_start_xmit(skb, dev); 3576 rc = netdev_start_xmit(skb, dev, txq, more); 3577 trace_net_dev_xmit(skb, rc, dev, len); 3578 3579 return rc; 3580 } 3581 3582 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, 3583 struct netdev_queue *txq, int *ret) 3584 { 3585 struct sk_buff *skb = first; 3586 int rc = NETDEV_TX_OK; 3587 3588 while (skb) { 3589 struct sk_buff *next = skb->next; 3590 3591 skb_mark_not_on_list(skb); 3592 rc = xmit_one(skb, dev, txq, next != NULL); 3593 if (unlikely(!dev_xmit_complete(rc))) { 3594 skb->next = next; 3595 goto out; 3596 } 3597 3598 skb = next; 3599 if (netif_tx_queue_stopped(txq) && skb) { 3600 rc = NETDEV_TX_BUSY; 3601 break; 3602 } 3603 } 3604 3605 out: 3606 *ret = rc; 3607 return skb; 3608 } 3609 3610 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, 3611 netdev_features_t features) 3612 { 3613 if (skb_vlan_tag_present(skb) && 3614 !vlan_hw_offload_capable(features, skb->vlan_proto)) 3615 skb = __vlan_hwaccel_push_inside(skb); 3616 return skb; 3617 } 3618 3619 int skb_csum_hwoffload_help(struct sk_buff *skb, 3620 const netdev_features_t features) 3621 { 3622 if (unlikely(skb_csum_is_sctp(skb))) 3623 return !!(features & NETIF_F_SCTP_CRC) ? 0 : 3624 skb_crc32c_csum_help(skb); 3625 3626 if (features & NETIF_F_HW_CSUM) 3627 return 0; 3628 3629 if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) { 3630 if (vlan_get_protocol(skb) == htons(ETH_P_IPV6) && 3631 skb_network_header_len(skb) != sizeof(struct ipv6hdr)) 3632 goto sw_checksum; 3633 switch (skb->csum_offset) { 3634 case offsetof(struct tcphdr, check): 3635 case offsetof(struct udphdr, check): 3636 return 0; 3637 } 3638 } 3639 3640 sw_checksum: 3641 return skb_checksum_help(skb); 3642 } 3643 EXPORT_SYMBOL(skb_csum_hwoffload_help); 3644 3645 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again) 3646 { 3647 netdev_features_t features; 3648 3649 features = netif_skb_features(skb); 3650 skb = validate_xmit_vlan(skb, features); 3651 if (unlikely(!skb)) 3652 goto out_null; 3653 3654 skb = sk_validate_xmit_skb(skb, dev); 3655 if (unlikely(!skb)) 3656 goto out_null; 3657 3658 if (netif_needs_gso(skb, features)) { 3659 struct sk_buff *segs; 3660 3661 segs = skb_gso_segment(skb, features); 3662 if (IS_ERR(segs)) { 3663 goto out_kfree_skb; 3664 } else if (segs) { 3665 consume_skb(skb); 3666 skb = segs; 3667 } 3668 } else { 3669 if (skb_needs_linearize(skb, features) && 3670 __skb_linearize(skb)) 3671 goto out_kfree_skb; 3672 3673 /* If packet is not checksummed and device does not 3674 * support checksumming for this protocol, complete 3675 * checksumming here. 3676 */ 3677 if (skb->ip_summed == CHECKSUM_PARTIAL) { 3678 if (skb->encapsulation) 3679 skb_set_inner_transport_header(skb, 3680 skb_checksum_start_offset(skb)); 3681 else 3682 skb_set_transport_header(skb, 3683 skb_checksum_start_offset(skb)); 3684 if (skb_csum_hwoffload_help(skb, features)) 3685 goto out_kfree_skb; 3686 } 3687 } 3688 3689 skb = validate_xmit_xfrm(skb, features, again); 3690 3691 return skb; 3692 3693 out_kfree_skb: 3694 kfree_skb(skb); 3695 out_null: 3696 dev_core_stats_tx_dropped_inc(dev); 3697 return NULL; 3698 } 3699 3700 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again) 3701 { 3702 struct sk_buff *next, *head = NULL, *tail; 3703 3704 for (; skb != NULL; skb = next) { 3705 next = skb->next; 3706 skb_mark_not_on_list(skb); 3707 3708 /* in case skb wont be segmented, point to itself */ 3709 skb->prev = skb; 3710 3711 skb = validate_xmit_skb(skb, dev, again); 3712 if (!skb) 3713 continue; 3714 3715 if (!head) 3716 head = skb; 3717 else 3718 tail->next = skb; 3719 /* If skb was segmented, skb->prev points to 3720 * the last segment. If not, it still contains skb. 3721 */ 3722 tail = skb->prev; 3723 } 3724 return head; 3725 } 3726 EXPORT_SYMBOL_GPL(validate_xmit_skb_list); 3727 3728 static void qdisc_pkt_len_init(struct sk_buff *skb) 3729 { 3730 const struct skb_shared_info *shinfo = skb_shinfo(skb); 3731 3732 qdisc_skb_cb(skb)->pkt_len = skb->len; 3733 3734 /* To get more precise estimation of bytes sent on wire, 3735 * we add to pkt_len the headers size of all segments 3736 */ 3737 if (shinfo->gso_size && skb_transport_header_was_set(skb)) { 3738 u16 gso_segs = shinfo->gso_segs; 3739 unsigned int hdr_len; 3740 3741 /* mac layer + network layer */ 3742 hdr_len = skb_transport_offset(skb); 3743 3744 /* + transport layer */ 3745 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { 3746 const struct tcphdr *th; 3747 struct tcphdr _tcphdr; 3748 3749 th = skb_header_pointer(skb, hdr_len, 3750 sizeof(_tcphdr), &_tcphdr); 3751 if (likely(th)) 3752 hdr_len += __tcp_hdrlen(th); 3753 } else if (shinfo->gso_type & SKB_GSO_UDP_L4) { 3754 struct udphdr _udphdr; 3755 3756 if (skb_header_pointer(skb, hdr_len, 3757 sizeof(_udphdr), &_udphdr)) 3758 hdr_len += sizeof(struct udphdr); 3759 } 3760 3761 if (unlikely(shinfo->gso_type & SKB_GSO_DODGY)) { 3762 int payload = skb->len - hdr_len; 3763 3764 /* Malicious packet. */ 3765 if (payload <= 0) 3766 return; 3767 gso_segs = DIV_ROUND_UP(payload, shinfo->gso_size); 3768 } 3769 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; 3770 } 3771 } 3772 3773 static int dev_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *q, 3774 struct sk_buff **to_free, 3775 struct netdev_queue *txq) 3776 { 3777 int rc; 3778 3779 rc = q->enqueue(skb, q, to_free) & NET_XMIT_MASK; 3780 if (rc == NET_XMIT_SUCCESS) 3781 trace_qdisc_enqueue(q, txq, skb); 3782 return rc; 3783 } 3784 3785 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, 3786 struct net_device *dev, 3787 struct netdev_queue *txq) 3788 { 3789 spinlock_t *root_lock = qdisc_lock(q); 3790 struct sk_buff *to_free = NULL; 3791 bool contended; 3792 int rc; 3793 3794 qdisc_calculate_pkt_len(skb, q); 3795 3796 if (q->flags & TCQ_F_NOLOCK) { 3797 if (q->flags & TCQ_F_CAN_BYPASS && nolock_qdisc_is_empty(q) && 3798 qdisc_run_begin(q)) { 3799 /* Retest nolock_qdisc_is_empty() within the protection 3800 * of q->seqlock to protect from racing with requeuing. 3801 */ 3802 if (unlikely(!nolock_qdisc_is_empty(q))) { 3803 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3804 __qdisc_run(q); 3805 qdisc_run_end(q); 3806 3807 goto no_lock_out; 3808 } 3809 3810 qdisc_bstats_cpu_update(q, skb); 3811 if (sch_direct_xmit(skb, q, dev, txq, NULL, true) && 3812 !nolock_qdisc_is_empty(q)) 3813 __qdisc_run(q); 3814 3815 qdisc_run_end(q); 3816 return NET_XMIT_SUCCESS; 3817 } 3818 3819 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3820 qdisc_run(q); 3821 3822 no_lock_out: 3823 if (unlikely(to_free)) 3824 kfree_skb_list_reason(to_free, 3825 SKB_DROP_REASON_QDISC_DROP); 3826 return rc; 3827 } 3828 3829 /* 3830 * Heuristic to force contended enqueues to serialize on a 3831 * separate lock before trying to get qdisc main lock. 3832 * This permits qdisc->running owner to get the lock more 3833 * often and dequeue packets faster. 3834 * On PREEMPT_RT it is possible to preempt the qdisc owner during xmit 3835 * and then other tasks will only enqueue packets. The packets will be 3836 * sent after the qdisc owner is scheduled again. To prevent this 3837 * scenario the task always serialize on the lock. 3838 */ 3839 contended = qdisc_is_running(q) || IS_ENABLED(CONFIG_PREEMPT_RT); 3840 if (unlikely(contended)) 3841 spin_lock(&q->busylock); 3842 3843 spin_lock(root_lock); 3844 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { 3845 __qdisc_drop(skb, &to_free); 3846 rc = NET_XMIT_DROP; 3847 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && 3848 qdisc_run_begin(q)) { 3849 /* 3850 * This is a work-conserving queue; there are no old skbs 3851 * waiting to be sent out; and the qdisc is not running - 3852 * xmit the skb directly. 3853 */ 3854 3855 qdisc_bstats_update(q, skb); 3856 3857 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { 3858 if (unlikely(contended)) { 3859 spin_unlock(&q->busylock); 3860 contended = false; 3861 } 3862 __qdisc_run(q); 3863 } 3864 3865 qdisc_run_end(q); 3866 rc = NET_XMIT_SUCCESS; 3867 } else { 3868 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3869 if (qdisc_run_begin(q)) { 3870 if (unlikely(contended)) { 3871 spin_unlock(&q->busylock); 3872 contended = false; 3873 } 3874 __qdisc_run(q); 3875 qdisc_run_end(q); 3876 } 3877 } 3878 spin_unlock(root_lock); 3879 if (unlikely(to_free)) 3880 kfree_skb_list_reason(to_free, SKB_DROP_REASON_QDISC_DROP); 3881 if (unlikely(contended)) 3882 spin_unlock(&q->busylock); 3883 return rc; 3884 } 3885 3886 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) 3887 static void skb_update_prio(struct sk_buff *skb) 3888 { 3889 const struct netprio_map *map; 3890 const struct sock *sk; 3891 unsigned int prioidx; 3892 3893 if (skb->priority) 3894 return; 3895 map = rcu_dereference_bh(skb->dev->priomap); 3896 if (!map) 3897 return; 3898 sk = skb_to_full_sk(skb); 3899 if (!sk) 3900 return; 3901 3902 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data); 3903 3904 if (prioidx < map->priomap_len) 3905 skb->priority = map->priomap[prioidx]; 3906 } 3907 #else 3908 #define skb_update_prio(skb) 3909 #endif 3910 3911 /** 3912 * dev_loopback_xmit - loop back @skb 3913 * @net: network namespace this loopback is happening in 3914 * @sk: sk needed to be a netfilter okfn 3915 * @skb: buffer to transmit 3916 */ 3917 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) 3918 { 3919 skb_reset_mac_header(skb); 3920 __skb_pull(skb, skb_network_offset(skb)); 3921 skb->pkt_type = PACKET_LOOPBACK; 3922 if (skb->ip_summed == CHECKSUM_NONE) 3923 skb->ip_summed = CHECKSUM_UNNECESSARY; 3924 DEBUG_NET_WARN_ON_ONCE(!skb_dst(skb)); 3925 skb_dst_force(skb); 3926 netif_rx(skb); 3927 return 0; 3928 } 3929 EXPORT_SYMBOL(dev_loopback_xmit); 3930 3931 #ifdef CONFIG_NET_EGRESS 3932 static struct netdev_queue * 3933 netdev_tx_queue_mapping(struct net_device *dev, struct sk_buff *skb) 3934 { 3935 int qm = skb_get_queue_mapping(skb); 3936 3937 return netdev_get_tx_queue(dev, netdev_cap_txqueue(dev, qm)); 3938 } 3939 3940 static bool netdev_xmit_txqueue_skipped(void) 3941 { 3942 return __this_cpu_read(softnet_data.xmit.skip_txqueue); 3943 } 3944 3945 void netdev_xmit_skip_txqueue(bool skip) 3946 { 3947 __this_cpu_write(softnet_data.xmit.skip_txqueue, skip); 3948 } 3949 EXPORT_SYMBOL_GPL(netdev_xmit_skip_txqueue); 3950 #endif /* CONFIG_NET_EGRESS */ 3951 3952 #ifdef CONFIG_NET_XGRESS 3953 static int tc_run(struct tcx_entry *entry, struct sk_buff *skb) 3954 { 3955 int ret = TC_ACT_UNSPEC; 3956 #ifdef CONFIG_NET_CLS_ACT 3957 struct mini_Qdisc *miniq = rcu_dereference_bh(entry->miniq); 3958 struct tcf_result res; 3959 3960 if (!miniq) 3961 return ret; 3962 3963 tc_skb_cb(skb)->mru = 0; 3964 tc_skb_cb(skb)->post_ct = false; 3965 3966 mini_qdisc_bstats_cpu_update(miniq, skb); 3967 ret = tcf_classify(skb, miniq->block, miniq->filter_list, &res, false); 3968 /* Only tcf related quirks below. */ 3969 switch (ret) { 3970 case TC_ACT_SHOT: 3971 mini_qdisc_qstats_cpu_drop(miniq); 3972 break; 3973 case TC_ACT_OK: 3974 case TC_ACT_RECLASSIFY: 3975 skb->tc_index = TC_H_MIN(res.classid); 3976 break; 3977 } 3978 #endif /* CONFIG_NET_CLS_ACT */ 3979 return ret; 3980 } 3981 3982 static DEFINE_STATIC_KEY_FALSE(tcx_needed_key); 3983 3984 void tcx_inc(void) 3985 { 3986 static_branch_inc(&tcx_needed_key); 3987 } 3988 3989 void tcx_dec(void) 3990 { 3991 static_branch_dec(&tcx_needed_key); 3992 } 3993 3994 static __always_inline enum tcx_action_base 3995 tcx_run(const struct bpf_mprog_entry *entry, struct sk_buff *skb, 3996 const bool needs_mac) 3997 { 3998 const struct bpf_mprog_fp *fp; 3999 const struct bpf_prog *prog; 4000 int ret = TCX_NEXT; 4001 4002 if (needs_mac) 4003 __skb_push(skb, skb->mac_len); 4004 bpf_mprog_foreach_prog(entry, fp, prog) { 4005 bpf_compute_data_pointers(skb); 4006 ret = bpf_prog_run(prog, skb); 4007 if (ret != TCX_NEXT) 4008 break; 4009 } 4010 if (needs_mac) 4011 __skb_pull(skb, skb->mac_len); 4012 return tcx_action_code(skb, ret); 4013 } 4014 4015 static __always_inline struct sk_buff * 4016 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 4017 struct net_device *orig_dev, bool *another) 4018 { 4019 struct bpf_mprog_entry *entry = rcu_dereference_bh(skb->dev->tcx_ingress); 4020 int sch_ret; 4021 4022 if (!entry) 4023 return skb; 4024 if (*pt_prev) { 4025 *ret = deliver_skb(skb, *pt_prev, orig_dev); 4026 *pt_prev = NULL; 4027 } 4028 4029 qdisc_skb_cb(skb)->pkt_len = skb->len; 4030 tcx_set_ingress(skb, true); 4031 4032 if (static_branch_unlikely(&tcx_needed_key)) { 4033 sch_ret = tcx_run(entry, skb, true); 4034 if (sch_ret != TC_ACT_UNSPEC) 4035 goto ingress_verdict; 4036 } 4037 sch_ret = tc_run(tcx_entry(entry), skb); 4038 ingress_verdict: 4039 switch (sch_ret) { 4040 case TC_ACT_REDIRECT: 4041 /* skb_mac_header check was done by BPF, so we can safely 4042 * push the L2 header back before redirecting to another 4043 * netdev. 4044 */ 4045 __skb_push(skb, skb->mac_len); 4046 if (skb_do_redirect(skb) == -EAGAIN) { 4047 __skb_pull(skb, skb->mac_len); 4048 *another = true; 4049 break; 4050 } 4051 *ret = NET_RX_SUCCESS; 4052 return NULL; 4053 case TC_ACT_SHOT: 4054 kfree_skb_reason(skb, SKB_DROP_REASON_TC_INGRESS); 4055 *ret = NET_RX_DROP; 4056 return NULL; 4057 /* used by tc_run */ 4058 case TC_ACT_STOLEN: 4059 case TC_ACT_QUEUED: 4060 case TC_ACT_TRAP: 4061 consume_skb(skb); 4062 fallthrough; 4063 case TC_ACT_CONSUMED: 4064 *ret = NET_RX_SUCCESS; 4065 return NULL; 4066 } 4067 4068 return skb; 4069 } 4070 4071 static __always_inline struct sk_buff * 4072 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 4073 { 4074 struct bpf_mprog_entry *entry = rcu_dereference_bh(dev->tcx_egress); 4075 int sch_ret; 4076 4077 if (!entry) 4078 return skb; 4079 4080 /* qdisc_skb_cb(skb)->pkt_len & tcx_set_ingress() was 4081 * already set by the caller. 4082 */ 4083 if (static_branch_unlikely(&tcx_needed_key)) { 4084 sch_ret = tcx_run(entry, skb, false); 4085 if (sch_ret != TC_ACT_UNSPEC) 4086 goto egress_verdict; 4087 } 4088 sch_ret = tc_run(tcx_entry(entry), skb); 4089 egress_verdict: 4090 switch (sch_ret) { 4091 case TC_ACT_REDIRECT: 4092 /* No need to push/pop skb's mac_header here on egress! */ 4093 skb_do_redirect(skb); 4094 *ret = NET_XMIT_SUCCESS; 4095 return NULL; 4096 case TC_ACT_SHOT: 4097 kfree_skb_reason(skb, SKB_DROP_REASON_TC_EGRESS); 4098 *ret = NET_XMIT_DROP; 4099 return NULL; 4100 /* used by tc_run */ 4101 case TC_ACT_STOLEN: 4102 case TC_ACT_QUEUED: 4103 case TC_ACT_TRAP: 4104 consume_skb(skb); 4105 fallthrough; 4106 case TC_ACT_CONSUMED: 4107 *ret = NET_XMIT_SUCCESS; 4108 return NULL; 4109 } 4110 4111 return skb; 4112 } 4113 #else 4114 static __always_inline struct sk_buff * 4115 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 4116 struct net_device *orig_dev, bool *another) 4117 { 4118 return skb; 4119 } 4120 4121 static __always_inline struct sk_buff * 4122 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 4123 { 4124 return skb; 4125 } 4126 #endif /* CONFIG_NET_XGRESS */ 4127 4128 #ifdef CONFIG_XPS 4129 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb, 4130 struct xps_dev_maps *dev_maps, unsigned int tci) 4131 { 4132 int tc = netdev_get_prio_tc_map(dev, skb->priority); 4133 struct xps_map *map; 4134 int queue_index = -1; 4135 4136 if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids) 4137 return queue_index; 4138 4139 tci *= dev_maps->num_tc; 4140 tci += tc; 4141 4142 map = rcu_dereference(dev_maps->attr_map[tci]); 4143 if (map) { 4144 if (map->len == 1) 4145 queue_index = map->queues[0]; 4146 else 4147 queue_index = map->queues[reciprocal_scale( 4148 skb_get_hash(skb), map->len)]; 4149 if (unlikely(queue_index >= dev->real_num_tx_queues)) 4150 queue_index = -1; 4151 } 4152 return queue_index; 4153 } 4154 #endif 4155 4156 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev, 4157 struct sk_buff *skb) 4158 { 4159 #ifdef CONFIG_XPS 4160 struct xps_dev_maps *dev_maps; 4161 struct sock *sk = skb->sk; 4162 int queue_index = -1; 4163 4164 if (!static_key_false(&xps_needed)) 4165 return -1; 4166 4167 rcu_read_lock(); 4168 if (!static_key_false(&xps_rxqs_needed)) 4169 goto get_cpus_map; 4170 4171 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]); 4172 if (dev_maps) { 4173 int tci = sk_rx_queue_get(sk); 4174 4175 if (tci >= 0) 4176 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4177 tci); 4178 } 4179 4180 get_cpus_map: 4181 if (queue_index < 0) { 4182 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]); 4183 if (dev_maps) { 4184 unsigned int tci = skb->sender_cpu - 1; 4185 4186 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4187 tci); 4188 } 4189 } 4190 rcu_read_unlock(); 4191 4192 return queue_index; 4193 #else 4194 return -1; 4195 #endif 4196 } 4197 4198 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb, 4199 struct net_device *sb_dev) 4200 { 4201 return 0; 4202 } 4203 EXPORT_SYMBOL(dev_pick_tx_zero); 4204 4205 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb, 4206 struct net_device *sb_dev) 4207 { 4208 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues; 4209 } 4210 EXPORT_SYMBOL(dev_pick_tx_cpu_id); 4211 4212 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb, 4213 struct net_device *sb_dev) 4214 { 4215 struct sock *sk = skb->sk; 4216 int queue_index = sk_tx_queue_get(sk); 4217 4218 sb_dev = sb_dev ? : dev; 4219 4220 if (queue_index < 0 || skb->ooo_okay || 4221 queue_index >= dev->real_num_tx_queues) { 4222 int new_index = get_xps_queue(dev, sb_dev, skb); 4223 4224 if (new_index < 0) 4225 new_index = skb_tx_hash(dev, sb_dev, skb); 4226 4227 if (queue_index != new_index && sk && 4228 sk_fullsock(sk) && 4229 rcu_access_pointer(sk->sk_dst_cache)) 4230 sk_tx_queue_set(sk, new_index); 4231 4232 queue_index = new_index; 4233 } 4234 4235 return queue_index; 4236 } 4237 EXPORT_SYMBOL(netdev_pick_tx); 4238 4239 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev, 4240 struct sk_buff *skb, 4241 struct net_device *sb_dev) 4242 { 4243 int queue_index = 0; 4244 4245 #ifdef CONFIG_XPS 4246 u32 sender_cpu = skb->sender_cpu - 1; 4247 4248 if (sender_cpu >= (u32)NR_CPUS) 4249 skb->sender_cpu = raw_smp_processor_id() + 1; 4250 #endif 4251 4252 if (dev->real_num_tx_queues != 1) { 4253 const struct net_device_ops *ops = dev->netdev_ops; 4254 4255 if (ops->ndo_select_queue) 4256 queue_index = ops->ndo_select_queue(dev, skb, sb_dev); 4257 else 4258 queue_index = netdev_pick_tx(dev, skb, sb_dev); 4259 4260 queue_index = netdev_cap_txqueue(dev, queue_index); 4261 } 4262 4263 skb_set_queue_mapping(skb, queue_index); 4264 return netdev_get_tx_queue(dev, queue_index); 4265 } 4266 4267 /** 4268 * __dev_queue_xmit() - transmit a buffer 4269 * @skb: buffer to transmit 4270 * @sb_dev: suboordinate device used for L2 forwarding offload 4271 * 4272 * Queue a buffer for transmission to a network device. The caller must 4273 * have set the device and priority and built the buffer before calling 4274 * this function. The function can be called from an interrupt. 4275 * 4276 * When calling this method, interrupts MUST be enabled. This is because 4277 * the BH enable code must have IRQs enabled so that it will not deadlock. 4278 * 4279 * Regardless of the return value, the skb is consumed, so it is currently 4280 * difficult to retry a send to this method. (You can bump the ref count 4281 * before sending to hold a reference for retry if you are careful.) 4282 * 4283 * Return: 4284 * * 0 - buffer successfully transmitted 4285 * * positive qdisc return code - NET_XMIT_DROP etc. 4286 * * negative errno - other errors 4287 */ 4288 int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev) 4289 { 4290 struct net_device *dev = skb->dev; 4291 struct netdev_queue *txq = NULL; 4292 struct Qdisc *q; 4293 int rc = -ENOMEM; 4294 bool again = false; 4295 4296 skb_reset_mac_header(skb); 4297 skb_assert_len(skb); 4298 4299 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) 4300 __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED); 4301 4302 /* Disable soft irqs for various locks below. Also 4303 * stops preemption for RCU. 4304 */ 4305 rcu_read_lock_bh(); 4306 4307 skb_update_prio(skb); 4308 4309 qdisc_pkt_len_init(skb); 4310 tcx_set_ingress(skb, false); 4311 #ifdef CONFIG_NET_EGRESS 4312 if (static_branch_unlikely(&egress_needed_key)) { 4313 if (nf_hook_egress_active()) { 4314 skb = nf_hook_egress(skb, &rc, dev); 4315 if (!skb) 4316 goto out; 4317 } 4318 4319 netdev_xmit_skip_txqueue(false); 4320 4321 nf_skip_egress(skb, true); 4322 skb = sch_handle_egress(skb, &rc, dev); 4323 if (!skb) 4324 goto out; 4325 nf_skip_egress(skb, false); 4326 4327 if (netdev_xmit_txqueue_skipped()) 4328 txq = netdev_tx_queue_mapping(dev, skb); 4329 } 4330 #endif 4331 /* If device/qdisc don't need skb->dst, release it right now while 4332 * its hot in this cpu cache. 4333 */ 4334 if (dev->priv_flags & IFF_XMIT_DST_RELEASE) 4335 skb_dst_drop(skb); 4336 else 4337 skb_dst_force(skb); 4338 4339 if (!txq) 4340 txq = netdev_core_pick_tx(dev, skb, sb_dev); 4341 4342 q = rcu_dereference_bh(txq->qdisc); 4343 4344 trace_net_dev_queue(skb); 4345 if (q->enqueue) { 4346 rc = __dev_xmit_skb(skb, q, dev, txq); 4347 goto out; 4348 } 4349 4350 /* The device has no queue. Common case for software devices: 4351 * loopback, all the sorts of tunnels... 4352 4353 * Really, it is unlikely that netif_tx_lock protection is necessary 4354 * here. (f.e. loopback and IP tunnels are clean ignoring statistics 4355 * counters.) 4356 * However, it is possible, that they rely on protection 4357 * made by us here. 4358 4359 * Check this and shot the lock. It is not prone from deadlocks. 4360 *Either shot noqueue qdisc, it is even simpler 8) 4361 */ 4362 if (dev->flags & IFF_UP) { 4363 int cpu = smp_processor_id(); /* ok because BHs are off */ 4364 4365 /* Other cpus might concurrently change txq->xmit_lock_owner 4366 * to -1 or to their cpu id, but not to our id. 4367 */ 4368 if (READ_ONCE(txq->xmit_lock_owner) != cpu) { 4369 if (dev_xmit_recursion()) 4370 goto recursion_alert; 4371 4372 skb = validate_xmit_skb(skb, dev, &again); 4373 if (!skb) 4374 goto out; 4375 4376 HARD_TX_LOCK(dev, txq, cpu); 4377 4378 if (!netif_xmit_stopped(txq)) { 4379 dev_xmit_recursion_inc(); 4380 skb = dev_hard_start_xmit(skb, dev, txq, &rc); 4381 dev_xmit_recursion_dec(); 4382 if (dev_xmit_complete(rc)) { 4383 HARD_TX_UNLOCK(dev, txq); 4384 goto out; 4385 } 4386 } 4387 HARD_TX_UNLOCK(dev, txq); 4388 net_crit_ratelimited("Virtual device %s asks to queue packet!\n", 4389 dev->name); 4390 } else { 4391 /* Recursion is detected! It is possible, 4392 * unfortunately 4393 */ 4394 recursion_alert: 4395 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", 4396 dev->name); 4397 } 4398 } 4399 4400 rc = -ENETDOWN; 4401 rcu_read_unlock_bh(); 4402 4403 dev_core_stats_tx_dropped_inc(dev); 4404 kfree_skb_list(skb); 4405 return rc; 4406 out: 4407 rcu_read_unlock_bh(); 4408 return rc; 4409 } 4410 EXPORT_SYMBOL(__dev_queue_xmit); 4411 4412 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id) 4413 { 4414 struct net_device *dev = skb->dev; 4415 struct sk_buff *orig_skb = skb; 4416 struct netdev_queue *txq; 4417 int ret = NETDEV_TX_BUSY; 4418 bool again = false; 4419 4420 if (unlikely(!netif_running(dev) || 4421 !netif_carrier_ok(dev))) 4422 goto drop; 4423 4424 skb = validate_xmit_skb_list(skb, dev, &again); 4425 if (skb != orig_skb) 4426 goto drop; 4427 4428 skb_set_queue_mapping(skb, queue_id); 4429 txq = skb_get_tx_queue(dev, skb); 4430 4431 local_bh_disable(); 4432 4433 dev_xmit_recursion_inc(); 4434 HARD_TX_LOCK(dev, txq, smp_processor_id()); 4435 if (!netif_xmit_frozen_or_drv_stopped(txq)) 4436 ret = netdev_start_xmit(skb, dev, txq, false); 4437 HARD_TX_UNLOCK(dev, txq); 4438 dev_xmit_recursion_dec(); 4439 4440 local_bh_enable(); 4441 return ret; 4442 drop: 4443 dev_core_stats_tx_dropped_inc(dev); 4444 kfree_skb_list(skb); 4445 return NET_XMIT_DROP; 4446 } 4447 EXPORT_SYMBOL(__dev_direct_xmit); 4448 4449 /************************************************************************* 4450 * Receiver routines 4451 *************************************************************************/ 4452 4453 int netdev_max_backlog __read_mostly = 1000; 4454 EXPORT_SYMBOL(netdev_max_backlog); 4455 4456 int netdev_tstamp_prequeue __read_mostly = 1; 4457 unsigned int sysctl_skb_defer_max __read_mostly = 64; 4458 int netdev_budget __read_mostly = 300; 4459 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */ 4460 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ; 4461 int weight_p __read_mostly = 64; /* old backlog weight */ 4462 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ 4463 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ 4464 int dev_rx_weight __read_mostly = 64; 4465 int dev_tx_weight __read_mostly = 64; 4466 4467 /* Called with irq disabled */ 4468 static inline void ____napi_schedule(struct softnet_data *sd, 4469 struct napi_struct *napi) 4470 { 4471 struct task_struct *thread; 4472 4473 lockdep_assert_irqs_disabled(); 4474 4475 if (test_bit(NAPI_STATE_THREADED, &napi->state)) { 4476 /* Paired with smp_mb__before_atomic() in 4477 * napi_enable()/dev_set_threaded(). 4478 * Use READ_ONCE() to guarantee a complete 4479 * read on napi->thread. Only call 4480 * wake_up_process() when it's not NULL. 4481 */ 4482 thread = READ_ONCE(napi->thread); 4483 if (thread) { 4484 /* Avoid doing set_bit() if the thread is in 4485 * INTERRUPTIBLE state, cause napi_thread_wait() 4486 * makes sure to proceed with napi polling 4487 * if the thread is explicitly woken from here. 4488 */ 4489 if (READ_ONCE(thread->__state) != TASK_INTERRUPTIBLE) 4490 set_bit(NAPI_STATE_SCHED_THREADED, &napi->state); 4491 wake_up_process(thread); 4492 return; 4493 } 4494 } 4495 4496 list_add_tail(&napi->poll_list, &sd->poll_list); 4497 WRITE_ONCE(napi->list_owner, smp_processor_id()); 4498 /* If not called from net_rx_action() 4499 * we have to raise NET_RX_SOFTIRQ. 4500 */ 4501 if (!sd->in_net_rx_action) 4502 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4503 } 4504 4505 #ifdef CONFIG_RPS 4506 4507 /* One global table that all flow-based protocols share. */ 4508 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; 4509 EXPORT_SYMBOL(rps_sock_flow_table); 4510 u32 rps_cpu_mask __read_mostly; 4511 EXPORT_SYMBOL(rps_cpu_mask); 4512 4513 struct static_key_false rps_needed __read_mostly; 4514 EXPORT_SYMBOL(rps_needed); 4515 struct static_key_false rfs_needed __read_mostly; 4516 EXPORT_SYMBOL(rfs_needed); 4517 4518 static struct rps_dev_flow * 4519 set_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4520 struct rps_dev_flow *rflow, u16 next_cpu) 4521 { 4522 if (next_cpu < nr_cpu_ids) { 4523 #ifdef CONFIG_RFS_ACCEL 4524 struct netdev_rx_queue *rxqueue; 4525 struct rps_dev_flow_table *flow_table; 4526 struct rps_dev_flow *old_rflow; 4527 u32 flow_id; 4528 u16 rxq_index; 4529 int rc; 4530 4531 /* Should we steer this flow to a different hardware queue? */ 4532 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || 4533 !(dev->features & NETIF_F_NTUPLE)) 4534 goto out; 4535 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); 4536 if (rxq_index == skb_get_rx_queue(skb)) 4537 goto out; 4538 4539 rxqueue = dev->_rx + rxq_index; 4540 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4541 if (!flow_table) 4542 goto out; 4543 flow_id = skb_get_hash(skb) & flow_table->mask; 4544 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, 4545 rxq_index, flow_id); 4546 if (rc < 0) 4547 goto out; 4548 old_rflow = rflow; 4549 rflow = &flow_table->flows[flow_id]; 4550 rflow->filter = rc; 4551 if (old_rflow->filter == rflow->filter) 4552 old_rflow->filter = RPS_NO_FILTER; 4553 out: 4554 #endif 4555 rflow->last_qtail = 4556 per_cpu(softnet_data, next_cpu).input_queue_head; 4557 } 4558 4559 rflow->cpu = next_cpu; 4560 return rflow; 4561 } 4562 4563 /* 4564 * get_rps_cpu is called from netif_receive_skb and returns the target 4565 * CPU from the RPS map of the receiving queue for a given skb. 4566 * rcu_read_lock must be held on entry. 4567 */ 4568 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4569 struct rps_dev_flow **rflowp) 4570 { 4571 const struct rps_sock_flow_table *sock_flow_table; 4572 struct netdev_rx_queue *rxqueue = dev->_rx; 4573 struct rps_dev_flow_table *flow_table; 4574 struct rps_map *map; 4575 int cpu = -1; 4576 u32 tcpu; 4577 u32 hash; 4578 4579 if (skb_rx_queue_recorded(skb)) { 4580 u16 index = skb_get_rx_queue(skb); 4581 4582 if (unlikely(index >= dev->real_num_rx_queues)) { 4583 WARN_ONCE(dev->real_num_rx_queues > 1, 4584 "%s received packet on queue %u, but number " 4585 "of RX queues is %u\n", 4586 dev->name, index, dev->real_num_rx_queues); 4587 goto done; 4588 } 4589 rxqueue += index; 4590 } 4591 4592 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ 4593 4594 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4595 map = rcu_dereference(rxqueue->rps_map); 4596 if (!flow_table && !map) 4597 goto done; 4598 4599 skb_reset_network_header(skb); 4600 hash = skb_get_hash(skb); 4601 if (!hash) 4602 goto done; 4603 4604 sock_flow_table = rcu_dereference(rps_sock_flow_table); 4605 if (flow_table && sock_flow_table) { 4606 struct rps_dev_flow *rflow; 4607 u32 next_cpu; 4608 u32 ident; 4609 4610 /* First check into global flow table if there is a match. 4611 * This READ_ONCE() pairs with WRITE_ONCE() from rps_record_sock_flow(). 4612 */ 4613 ident = READ_ONCE(sock_flow_table->ents[hash & sock_flow_table->mask]); 4614 if ((ident ^ hash) & ~rps_cpu_mask) 4615 goto try_rps; 4616 4617 next_cpu = ident & rps_cpu_mask; 4618 4619 /* OK, now we know there is a match, 4620 * we can look at the local (per receive queue) flow table 4621 */ 4622 rflow = &flow_table->flows[hash & flow_table->mask]; 4623 tcpu = rflow->cpu; 4624 4625 /* 4626 * If the desired CPU (where last recvmsg was done) is 4627 * different from current CPU (one in the rx-queue flow 4628 * table entry), switch if one of the following holds: 4629 * - Current CPU is unset (>= nr_cpu_ids). 4630 * - Current CPU is offline. 4631 * - The current CPU's queue tail has advanced beyond the 4632 * last packet that was enqueued using this table entry. 4633 * This guarantees that all previous packets for the flow 4634 * have been dequeued, thus preserving in order delivery. 4635 */ 4636 if (unlikely(tcpu != next_cpu) && 4637 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || 4638 ((int)(per_cpu(softnet_data, tcpu).input_queue_head - 4639 rflow->last_qtail)) >= 0)) { 4640 tcpu = next_cpu; 4641 rflow = set_rps_cpu(dev, skb, rflow, next_cpu); 4642 } 4643 4644 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { 4645 *rflowp = rflow; 4646 cpu = tcpu; 4647 goto done; 4648 } 4649 } 4650 4651 try_rps: 4652 4653 if (map) { 4654 tcpu = map->cpus[reciprocal_scale(hash, map->len)]; 4655 if (cpu_online(tcpu)) { 4656 cpu = tcpu; 4657 goto done; 4658 } 4659 } 4660 4661 done: 4662 return cpu; 4663 } 4664 4665 #ifdef CONFIG_RFS_ACCEL 4666 4667 /** 4668 * rps_may_expire_flow - check whether an RFS hardware filter may be removed 4669 * @dev: Device on which the filter was set 4670 * @rxq_index: RX queue index 4671 * @flow_id: Flow ID passed to ndo_rx_flow_steer() 4672 * @filter_id: Filter ID returned by ndo_rx_flow_steer() 4673 * 4674 * Drivers that implement ndo_rx_flow_steer() should periodically call 4675 * this function for each installed filter and remove the filters for 4676 * which it returns %true. 4677 */ 4678 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, 4679 u32 flow_id, u16 filter_id) 4680 { 4681 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; 4682 struct rps_dev_flow_table *flow_table; 4683 struct rps_dev_flow *rflow; 4684 bool expire = true; 4685 unsigned int cpu; 4686 4687 rcu_read_lock(); 4688 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4689 if (flow_table && flow_id <= flow_table->mask) { 4690 rflow = &flow_table->flows[flow_id]; 4691 cpu = READ_ONCE(rflow->cpu); 4692 if (rflow->filter == filter_id && cpu < nr_cpu_ids && 4693 ((int)(per_cpu(softnet_data, cpu).input_queue_head - 4694 rflow->last_qtail) < 4695 (int)(10 * flow_table->mask))) 4696 expire = false; 4697 } 4698 rcu_read_unlock(); 4699 return expire; 4700 } 4701 EXPORT_SYMBOL(rps_may_expire_flow); 4702 4703 #endif /* CONFIG_RFS_ACCEL */ 4704 4705 /* Called from hardirq (IPI) context */ 4706 static void rps_trigger_softirq(void *data) 4707 { 4708 struct softnet_data *sd = data; 4709 4710 ____napi_schedule(sd, &sd->backlog); 4711 sd->received_rps++; 4712 } 4713 4714 #endif /* CONFIG_RPS */ 4715 4716 /* Called from hardirq (IPI) context */ 4717 static void trigger_rx_softirq(void *data) 4718 { 4719 struct softnet_data *sd = data; 4720 4721 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4722 smp_store_release(&sd->defer_ipi_scheduled, 0); 4723 } 4724 4725 /* 4726 * After we queued a packet into sd->input_pkt_queue, 4727 * we need to make sure this queue is serviced soon. 4728 * 4729 * - If this is another cpu queue, link it to our rps_ipi_list, 4730 * and make sure we will process rps_ipi_list from net_rx_action(). 4731 * 4732 * - If this is our own queue, NAPI schedule our backlog. 4733 * Note that this also raises NET_RX_SOFTIRQ. 4734 */ 4735 static void napi_schedule_rps(struct softnet_data *sd) 4736 { 4737 struct softnet_data *mysd = this_cpu_ptr(&softnet_data); 4738 4739 #ifdef CONFIG_RPS 4740 if (sd != mysd) { 4741 sd->rps_ipi_next = mysd->rps_ipi_list; 4742 mysd->rps_ipi_list = sd; 4743 4744 /* If not called from net_rx_action() or napi_threaded_poll() 4745 * we have to raise NET_RX_SOFTIRQ. 4746 */ 4747 if (!mysd->in_net_rx_action && !mysd->in_napi_threaded_poll) 4748 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4749 return; 4750 } 4751 #endif /* CONFIG_RPS */ 4752 __napi_schedule_irqoff(&mysd->backlog); 4753 } 4754 4755 #ifdef CONFIG_NET_FLOW_LIMIT 4756 int netdev_flow_limit_table_len __read_mostly = (1 << 12); 4757 #endif 4758 4759 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) 4760 { 4761 #ifdef CONFIG_NET_FLOW_LIMIT 4762 struct sd_flow_limit *fl; 4763 struct softnet_data *sd; 4764 unsigned int old_flow, new_flow; 4765 4766 if (qlen < (READ_ONCE(netdev_max_backlog) >> 1)) 4767 return false; 4768 4769 sd = this_cpu_ptr(&softnet_data); 4770 4771 rcu_read_lock(); 4772 fl = rcu_dereference(sd->flow_limit); 4773 if (fl) { 4774 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); 4775 old_flow = fl->history[fl->history_head]; 4776 fl->history[fl->history_head] = new_flow; 4777 4778 fl->history_head++; 4779 fl->history_head &= FLOW_LIMIT_HISTORY - 1; 4780 4781 if (likely(fl->buckets[old_flow])) 4782 fl->buckets[old_flow]--; 4783 4784 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { 4785 fl->count++; 4786 rcu_read_unlock(); 4787 return true; 4788 } 4789 } 4790 rcu_read_unlock(); 4791 #endif 4792 return false; 4793 } 4794 4795 /* 4796 * enqueue_to_backlog is called to queue an skb to a per CPU backlog 4797 * queue (may be a remote CPU queue). 4798 */ 4799 static int enqueue_to_backlog(struct sk_buff *skb, int cpu, 4800 unsigned int *qtail) 4801 { 4802 enum skb_drop_reason reason; 4803 struct softnet_data *sd; 4804 unsigned long flags; 4805 unsigned int qlen; 4806 4807 reason = SKB_DROP_REASON_NOT_SPECIFIED; 4808 sd = &per_cpu(softnet_data, cpu); 4809 4810 rps_lock_irqsave(sd, &flags); 4811 if (!netif_running(skb->dev)) 4812 goto drop; 4813 qlen = skb_queue_len(&sd->input_pkt_queue); 4814 if (qlen <= READ_ONCE(netdev_max_backlog) && !skb_flow_limit(skb, qlen)) { 4815 if (qlen) { 4816 enqueue: 4817 __skb_queue_tail(&sd->input_pkt_queue, skb); 4818 input_queue_tail_incr_save(sd, qtail); 4819 rps_unlock_irq_restore(sd, &flags); 4820 return NET_RX_SUCCESS; 4821 } 4822 4823 /* Schedule NAPI for backlog device 4824 * We can use non atomic operation since we own the queue lock 4825 */ 4826 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) 4827 napi_schedule_rps(sd); 4828 goto enqueue; 4829 } 4830 reason = SKB_DROP_REASON_CPU_BACKLOG; 4831 4832 drop: 4833 sd->dropped++; 4834 rps_unlock_irq_restore(sd, &flags); 4835 4836 dev_core_stats_rx_dropped_inc(skb->dev); 4837 kfree_skb_reason(skb, reason); 4838 return NET_RX_DROP; 4839 } 4840 4841 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb) 4842 { 4843 struct net_device *dev = skb->dev; 4844 struct netdev_rx_queue *rxqueue; 4845 4846 rxqueue = dev->_rx; 4847 4848 if (skb_rx_queue_recorded(skb)) { 4849 u16 index = skb_get_rx_queue(skb); 4850 4851 if (unlikely(index >= dev->real_num_rx_queues)) { 4852 WARN_ONCE(dev->real_num_rx_queues > 1, 4853 "%s received packet on queue %u, but number " 4854 "of RX queues is %u\n", 4855 dev->name, index, dev->real_num_rx_queues); 4856 4857 return rxqueue; /* Return first rxqueue */ 4858 } 4859 rxqueue += index; 4860 } 4861 return rxqueue; 4862 } 4863 4864 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp, 4865 struct bpf_prog *xdp_prog) 4866 { 4867 void *orig_data, *orig_data_end, *hard_start; 4868 struct netdev_rx_queue *rxqueue; 4869 bool orig_bcast, orig_host; 4870 u32 mac_len, frame_sz; 4871 __be16 orig_eth_type; 4872 struct ethhdr *eth; 4873 u32 metalen, act; 4874 int off; 4875 4876 /* The XDP program wants to see the packet starting at the MAC 4877 * header. 4878 */ 4879 mac_len = skb->data - skb_mac_header(skb); 4880 hard_start = skb->data - skb_headroom(skb); 4881 4882 /* SKB "head" area always have tailroom for skb_shared_info */ 4883 frame_sz = (void *)skb_end_pointer(skb) - hard_start; 4884 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 4885 4886 rxqueue = netif_get_rxqueue(skb); 4887 xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq); 4888 xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len, 4889 skb_headlen(skb) + mac_len, true); 4890 4891 orig_data_end = xdp->data_end; 4892 orig_data = xdp->data; 4893 eth = (struct ethhdr *)xdp->data; 4894 orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr); 4895 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest); 4896 orig_eth_type = eth->h_proto; 4897 4898 act = bpf_prog_run_xdp(xdp_prog, xdp); 4899 4900 /* check if bpf_xdp_adjust_head was used */ 4901 off = xdp->data - orig_data; 4902 if (off) { 4903 if (off > 0) 4904 __skb_pull(skb, off); 4905 else if (off < 0) 4906 __skb_push(skb, -off); 4907 4908 skb->mac_header += off; 4909 skb_reset_network_header(skb); 4910 } 4911 4912 /* check if bpf_xdp_adjust_tail was used */ 4913 off = xdp->data_end - orig_data_end; 4914 if (off != 0) { 4915 skb_set_tail_pointer(skb, xdp->data_end - xdp->data); 4916 skb->len += off; /* positive on grow, negative on shrink */ 4917 } 4918 4919 /* check if XDP changed eth hdr such SKB needs update */ 4920 eth = (struct ethhdr *)xdp->data; 4921 if ((orig_eth_type != eth->h_proto) || 4922 (orig_host != ether_addr_equal_64bits(eth->h_dest, 4923 skb->dev->dev_addr)) || 4924 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) { 4925 __skb_push(skb, ETH_HLEN); 4926 skb->pkt_type = PACKET_HOST; 4927 skb->protocol = eth_type_trans(skb, skb->dev); 4928 } 4929 4930 /* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull 4931 * before calling us again on redirect path. We do not call do_redirect 4932 * as we leave that up to the caller. 4933 * 4934 * Caller is responsible for managing lifetime of skb (i.e. calling 4935 * kfree_skb in response to actions it cannot handle/XDP_DROP). 4936 */ 4937 switch (act) { 4938 case XDP_REDIRECT: 4939 case XDP_TX: 4940 __skb_push(skb, mac_len); 4941 break; 4942 case XDP_PASS: 4943 metalen = xdp->data - xdp->data_meta; 4944 if (metalen) 4945 skb_metadata_set(skb, metalen); 4946 break; 4947 } 4948 4949 return act; 4950 } 4951 4952 static u32 netif_receive_generic_xdp(struct sk_buff *skb, 4953 struct xdp_buff *xdp, 4954 struct bpf_prog *xdp_prog) 4955 { 4956 u32 act = XDP_DROP; 4957 4958 /* Reinjected packets coming from act_mirred or similar should 4959 * not get XDP generic processing. 4960 */ 4961 if (skb_is_redirected(skb)) 4962 return XDP_PASS; 4963 4964 /* XDP packets must be linear and must have sufficient headroom 4965 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also 4966 * native XDP provides, thus we need to do it here as well. 4967 */ 4968 if (skb_cloned(skb) || skb_is_nonlinear(skb) || 4969 skb_headroom(skb) < XDP_PACKET_HEADROOM) { 4970 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb); 4971 int troom = skb->tail + skb->data_len - skb->end; 4972 4973 /* In case we have to go down the path and also linearize, 4974 * then lets do the pskb_expand_head() work just once here. 4975 */ 4976 if (pskb_expand_head(skb, 4977 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0, 4978 troom > 0 ? troom + 128 : 0, GFP_ATOMIC)) 4979 goto do_drop; 4980 if (skb_linearize(skb)) 4981 goto do_drop; 4982 } 4983 4984 act = bpf_prog_run_generic_xdp(skb, xdp, xdp_prog); 4985 switch (act) { 4986 case XDP_REDIRECT: 4987 case XDP_TX: 4988 case XDP_PASS: 4989 break; 4990 default: 4991 bpf_warn_invalid_xdp_action(skb->dev, xdp_prog, act); 4992 fallthrough; 4993 case XDP_ABORTED: 4994 trace_xdp_exception(skb->dev, xdp_prog, act); 4995 fallthrough; 4996 case XDP_DROP: 4997 do_drop: 4998 kfree_skb(skb); 4999 break; 5000 } 5001 5002 return act; 5003 } 5004 5005 /* When doing generic XDP we have to bypass the qdisc layer and the 5006 * network taps in order to match in-driver-XDP behavior. This also means 5007 * that XDP packets are able to starve other packets going through a qdisc, 5008 * and DDOS attacks will be more effective. In-driver-XDP use dedicated TX 5009 * queues, so they do not have this starvation issue. 5010 */ 5011 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog) 5012 { 5013 struct net_device *dev = skb->dev; 5014 struct netdev_queue *txq; 5015 bool free_skb = true; 5016 int cpu, rc; 5017 5018 txq = netdev_core_pick_tx(dev, skb, NULL); 5019 cpu = smp_processor_id(); 5020 HARD_TX_LOCK(dev, txq, cpu); 5021 if (!netif_xmit_frozen_or_drv_stopped(txq)) { 5022 rc = netdev_start_xmit(skb, dev, txq, 0); 5023 if (dev_xmit_complete(rc)) 5024 free_skb = false; 5025 } 5026 HARD_TX_UNLOCK(dev, txq); 5027 if (free_skb) { 5028 trace_xdp_exception(dev, xdp_prog, XDP_TX); 5029 dev_core_stats_tx_dropped_inc(dev); 5030 kfree_skb(skb); 5031 } 5032 } 5033 5034 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key); 5035 5036 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb) 5037 { 5038 if (xdp_prog) { 5039 struct xdp_buff xdp; 5040 u32 act; 5041 int err; 5042 5043 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog); 5044 if (act != XDP_PASS) { 5045 switch (act) { 5046 case XDP_REDIRECT: 5047 err = xdp_do_generic_redirect(skb->dev, skb, 5048 &xdp, xdp_prog); 5049 if (err) 5050 goto out_redir; 5051 break; 5052 case XDP_TX: 5053 generic_xdp_tx(skb, xdp_prog); 5054 break; 5055 } 5056 return XDP_DROP; 5057 } 5058 } 5059 return XDP_PASS; 5060 out_redir: 5061 kfree_skb_reason(skb, SKB_DROP_REASON_XDP); 5062 return XDP_DROP; 5063 } 5064 EXPORT_SYMBOL_GPL(do_xdp_generic); 5065 5066 static int netif_rx_internal(struct sk_buff *skb) 5067 { 5068 int ret; 5069 5070 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb); 5071 5072 trace_netif_rx(skb); 5073 5074 #ifdef CONFIG_RPS 5075 if (static_branch_unlikely(&rps_needed)) { 5076 struct rps_dev_flow voidflow, *rflow = &voidflow; 5077 int cpu; 5078 5079 rcu_read_lock(); 5080 5081 cpu = get_rps_cpu(skb->dev, skb, &rflow); 5082 if (cpu < 0) 5083 cpu = smp_processor_id(); 5084 5085 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5086 5087 rcu_read_unlock(); 5088 } else 5089 #endif 5090 { 5091 unsigned int qtail; 5092 5093 ret = enqueue_to_backlog(skb, smp_processor_id(), &qtail); 5094 } 5095 return ret; 5096 } 5097 5098 /** 5099 * __netif_rx - Slightly optimized version of netif_rx 5100 * @skb: buffer to post 5101 * 5102 * This behaves as netif_rx except that it does not disable bottom halves. 5103 * As a result this function may only be invoked from the interrupt context 5104 * (either hard or soft interrupt). 5105 */ 5106 int __netif_rx(struct sk_buff *skb) 5107 { 5108 int ret; 5109 5110 lockdep_assert_once(hardirq_count() | softirq_count()); 5111 5112 trace_netif_rx_entry(skb); 5113 ret = netif_rx_internal(skb); 5114 trace_netif_rx_exit(ret); 5115 return ret; 5116 } 5117 EXPORT_SYMBOL(__netif_rx); 5118 5119 /** 5120 * netif_rx - post buffer to the network code 5121 * @skb: buffer to post 5122 * 5123 * This function receives a packet from a device driver and queues it for 5124 * the upper (protocol) levels to process via the backlog NAPI device. It 5125 * always succeeds. The buffer may be dropped during processing for 5126 * congestion control or by the protocol layers. 5127 * The network buffer is passed via the backlog NAPI device. Modern NIC 5128 * driver should use NAPI and GRO. 5129 * This function can used from interrupt and from process context. The 5130 * caller from process context must not disable interrupts before invoking 5131 * this function. 5132 * 5133 * return values: 5134 * NET_RX_SUCCESS (no congestion) 5135 * NET_RX_DROP (packet was dropped) 5136 * 5137 */ 5138 int netif_rx(struct sk_buff *skb) 5139 { 5140 bool need_bh_off = !(hardirq_count() | softirq_count()); 5141 int ret; 5142 5143 if (need_bh_off) 5144 local_bh_disable(); 5145 trace_netif_rx_entry(skb); 5146 ret = netif_rx_internal(skb); 5147 trace_netif_rx_exit(ret); 5148 if (need_bh_off) 5149 local_bh_enable(); 5150 return ret; 5151 } 5152 EXPORT_SYMBOL(netif_rx); 5153 5154 static __latent_entropy void net_tx_action(struct softirq_action *h) 5155 { 5156 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 5157 5158 if (sd->completion_queue) { 5159 struct sk_buff *clist; 5160 5161 local_irq_disable(); 5162 clist = sd->completion_queue; 5163 sd->completion_queue = NULL; 5164 local_irq_enable(); 5165 5166 while (clist) { 5167 struct sk_buff *skb = clist; 5168 5169 clist = clist->next; 5170 5171 WARN_ON(refcount_read(&skb->users)); 5172 if (likely(get_kfree_skb_cb(skb)->reason == SKB_CONSUMED)) 5173 trace_consume_skb(skb, net_tx_action); 5174 else 5175 trace_kfree_skb(skb, net_tx_action, 5176 get_kfree_skb_cb(skb)->reason); 5177 5178 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) 5179 __kfree_skb(skb); 5180 else 5181 __napi_kfree_skb(skb, 5182 get_kfree_skb_cb(skb)->reason); 5183 } 5184 } 5185 5186 if (sd->output_queue) { 5187 struct Qdisc *head; 5188 5189 local_irq_disable(); 5190 head = sd->output_queue; 5191 sd->output_queue = NULL; 5192 sd->output_queue_tailp = &sd->output_queue; 5193 local_irq_enable(); 5194 5195 rcu_read_lock(); 5196 5197 while (head) { 5198 struct Qdisc *q = head; 5199 spinlock_t *root_lock = NULL; 5200 5201 head = head->next_sched; 5202 5203 /* We need to make sure head->next_sched is read 5204 * before clearing __QDISC_STATE_SCHED 5205 */ 5206 smp_mb__before_atomic(); 5207 5208 if (!(q->flags & TCQ_F_NOLOCK)) { 5209 root_lock = qdisc_lock(q); 5210 spin_lock(root_lock); 5211 } else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, 5212 &q->state))) { 5213 /* There is a synchronize_net() between 5214 * STATE_DEACTIVATED flag being set and 5215 * qdisc_reset()/some_qdisc_is_busy() in 5216 * dev_deactivate(), so we can safely bail out 5217 * early here to avoid data race between 5218 * qdisc_deactivate() and some_qdisc_is_busy() 5219 * for lockless qdisc. 5220 */ 5221 clear_bit(__QDISC_STATE_SCHED, &q->state); 5222 continue; 5223 } 5224 5225 clear_bit(__QDISC_STATE_SCHED, &q->state); 5226 qdisc_run(q); 5227 if (root_lock) 5228 spin_unlock(root_lock); 5229 } 5230 5231 rcu_read_unlock(); 5232 } 5233 5234 xfrm_dev_backlog(sd); 5235 } 5236 5237 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) 5238 /* This hook is defined here for ATM LANE */ 5239 int (*br_fdb_test_addr_hook)(struct net_device *dev, 5240 unsigned char *addr) __read_mostly; 5241 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); 5242 #endif 5243 5244 /** 5245 * netdev_is_rx_handler_busy - check if receive handler is registered 5246 * @dev: device to check 5247 * 5248 * Check if a receive handler is already registered for a given device. 5249 * Return true if there one. 5250 * 5251 * The caller must hold the rtnl_mutex. 5252 */ 5253 bool netdev_is_rx_handler_busy(struct net_device *dev) 5254 { 5255 ASSERT_RTNL(); 5256 return dev && rtnl_dereference(dev->rx_handler); 5257 } 5258 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); 5259 5260 /** 5261 * netdev_rx_handler_register - register receive handler 5262 * @dev: device to register a handler for 5263 * @rx_handler: receive handler to register 5264 * @rx_handler_data: data pointer that is used by rx handler 5265 * 5266 * Register a receive handler for a device. This handler will then be 5267 * called from __netif_receive_skb. A negative errno code is returned 5268 * on a failure. 5269 * 5270 * The caller must hold the rtnl_mutex. 5271 * 5272 * For a general description of rx_handler, see enum rx_handler_result. 5273 */ 5274 int netdev_rx_handler_register(struct net_device *dev, 5275 rx_handler_func_t *rx_handler, 5276 void *rx_handler_data) 5277 { 5278 if (netdev_is_rx_handler_busy(dev)) 5279 return -EBUSY; 5280 5281 if (dev->priv_flags & IFF_NO_RX_HANDLER) 5282 return -EINVAL; 5283 5284 /* Note: rx_handler_data must be set before rx_handler */ 5285 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); 5286 rcu_assign_pointer(dev->rx_handler, rx_handler); 5287 5288 return 0; 5289 } 5290 EXPORT_SYMBOL_GPL(netdev_rx_handler_register); 5291 5292 /** 5293 * netdev_rx_handler_unregister - unregister receive handler 5294 * @dev: device to unregister a handler from 5295 * 5296 * Unregister a receive handler from a device. 5297 * 5298 * The caller must hold the rtnl_mutex. 5299 */ 5300 void netdev_rx_handler_unregister(struct net_device *dev) 5301 { 5302 5303 ASSERT_RTNL(); 5304 RCU_INIT_POINTER(dev->rx_handler, NULL); 5305 /* a reader seeing a non NULL rx_handler in a rcu_read_lock() 5306 * section has a guarantee to see a non NULL rx_handler_data 5307 * as well. 5308 */ 5309 synchronize_net(); 5310 RCU_INIT_POINTER(dev->rx_handler_data, NULL); 5311 } 5312 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); 5313 5314 /* 5315 * Limit the use of PFMEMALLOC reserves to those protocols that implement 5316 * the special handling of PFMEMALLOC skbs. 5317 */ 5318 static bool skb_pfmemalloc_protocol(struct sk_buff *skb) 5319 { 5320 switch (skb->protocol) { 5321 case htons(ETH_P_ARP): 5322 case htons(ETH_P_IP): 5323 case htons(ETH_P_IPV6): 5324 case htons(ETH_P_8021Q): 5325 case htons(ETH_P_8021AD): 5326 return true; 5327 default: 5328 return false; 5329 } 5330 } 5331 5332 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, 5333 int *ret, struct net_device *orig_dev) 5334 { 5335 if (nf_hook_ingress_active(skb)) { 5336 int ingress_retval; 5337 5338 if (*pt_prev) { 5339 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5340 *pt_prev = NULL; 5341 } 5342 5343 rcu_read_lock(); 5344 ingress_retval = nf_hook_ingress(skb); 5345 rcu_read_unlock(); 5346 return ingress_retval; 5347 } 5348 return 0; 5349 } 5350 5351 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc, 5352 struct packet_type **ppt_prev) 5353 { 5354 struct packet_type *ptype, *pt_prev; 5355 rx_handler_func_t *rx_handler; 5356 struct sk_buff *skb = *pskb; 5357 struct net_device *orig_dev; 5358 bool deliver_exact = false; 5359 int ret = NET_RX_DROP; 5360 __be16 type; 5361 5362 net_timestamp_check(!READ_ONCE(netdev_tstamp_prequeue), skb); 5363 5364 trace_netif_receive_skb(skb); 5365 5366 orig_dev = skb->dev; 5367 5368 skb_reset_network_header(skb); 5369 if (!skb_transport_header_was_set(skb)) 5370 skb_reset_transport_header(skb); 5371 skb_reset_mac_len(skb); 5372 5373 pt_prev = NULL; 5374 5375 another_round: 5376 skb->skb_iif = skb->dev->ifindex; 5377 5378 __this_cpu_inc(softnet_data.processed); 5379 5380 if (static_branch_unlikely(&generic_xdp_needed_key)) { 5381 int ret2; 5382 5383 migrate_disable(); 5384 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb); 5385 migrate_enable(); 5386 5387 if (ret2 != XDP_PASS) { 5388 ret = NET_RX_DROP; 5389 goto out; 5390 } 5391 } 5392 5393 if (eth_type_vlan(skb->protocol)) { 5394 skb = skb_vlan_untag(skb); 5395 if (unlikely(!skb)) 5396 goto out; 5397 } 5398 5399 if (skb_skip_tc_classify(skb)) 5400 goto skip_classify; 5401 5402 if (pfmemalloc) 5403 goto skip_taps; 5404 5405 list_for_each_entry_rcu(ptype, &ptype_all, list) { 5406 if (pt_prev) 5407 ret = deliver_skb(skb, pt_prev, orig_dev); 5408 pt_prev = ptype; 5409 } 5410 5411 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { 5412 if (pt_prev) 5413 ret = deliver_skb(skb, pt_prev, orig_dev); 5414 pt_prev = ptype; 5415 } 5416 5417 skip_taps: 5418 #ifdef CONFIG_NET_INGRESS 5419 if (static_branch_unlikely(&ingress_needed_key)) { 5420 bool another = false; 5421 5422 nf_skip_egress(skb, true); 5423 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev, 5424 &another); 5425 if (another) 5426 goto another_round; 5427 if (!skb) 5428 goto out; 5429 5430 nf_skip_egress(skb, false); 5431 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) 5432 goto out; 5433 } 5434 #endif 5435 skb_reset_redirect(skb); 5436 skip_classify: 5437 if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) 5438 goto drop; 5439 5440 if (skb_vlan_tag_present(skb)) { 5441 if (pt_prev) { 5442 ret = deliver_skb(skb, pt_prev, orig_dev); 5443 pt_prev = NULL; 5444 } 5445 if (vlan_do_receive(&skb)) 5446 goto another_round; 5447 else if (unlikely(!skb)) 5448 goto out; 5449 } 5450 5451 rx_handler = rcu_dereference(skb->dev->rx_handler); 5452 if (rx_handler) { 5453 if (pt_prev) { 5454 ret = deliver_skb(skb, pt_prev, orig_dev); 5455 pt_prev = NULL; 5456 } 5457 switch (rx_handler(&skb)) { 5458 case RX_HANDLER_CONSUMED: 5459 ret = NET_RX_SUCCESS; 5460 goto out; 5461 case RX_HANDLER_ANOTHER: 5462 goto another_round; 5463 case RX_HANDLER_EXACT: 5464 deliver_exact = true; 5465 break; 5466 case RX_HANDLER_PASS: 5467 break; 5468 default: 5469 BUG(); 5470 } 5471 } 5472 5473 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) { 5474 check_vlan_id: 5475 if (skb_vlan_tag_get_id(skb)) { 5476 /* Vlan id is non 0 and vlan_do_receive() above couldn't 5477 * find vlan device. 5478 */ 5479 skb->pkt_type = PACKET_OTHERHOST; 5480 } else if (eth_type_vlan(skb->protocol)) { 5481 /* Outer header is 802.1P with vlan 0, inner header is 5482 * 802.1Q or 802.1AD and vlan_do_receive() above could 5483 * not find vlan dev for vlan id 0. 5484 */ 5485 __vlan_hwaccel_clear_tag(skb); 5486 skb = skb_vlan_untag(skb); 5487 if (unlikely(!skb)) 5488 goto out; 5489 if (vlan_do_receive(&skb)) 5490 /* After stripping off 802.1P header with vlan 0 5491 * vlan dev is found for inner header. 5492 */ 5493 goto another_round; 5494 else if (unlikely(!skb)) 5495 goto out; 5496 else 5497 /* We have stripped outer 802.1P vlan 0 header. 5498 * But could not find vlan dev. 5499 * check again for vlan id to set OTHERHOST. 5500 */ 5501 goto check_vlan_id; 5502 } 5503 /* Note: we might in the future use prio bits 5504 * and set skb->priority like in vlan_do_receive() 5505 * For the time being, just ignore Priority Code Point 5506 */ 5507 __vlan_hwaccel_clear_tag(skb); 5508 } 5509 5510 type = skb->protocol; 5511 5512 /* deliver only exact match when indicated */ 5513 if (likely(!deliver_exact)) { 5514 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5515 &ptype_base[ntohs(type) & 5516 PTYPE_HASH_MASK]); 5517 } 5518 5519 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5520 &orig_dev->ptype_specific); 5521 5522 if (unlikely(skb->dev != orig_dev)) { 5523 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5524 &skb->dev->ptype_specific); 5525 } 5526 5527 if (pt_prev) { 5528 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 5529 goto drop; 5530 *ppt_prev = pt_prev; 5531 } else { 5532 drop: 5533 if (!deliver_exact) 5534 dev_core_stats_rx_dropped_inc(skb->dev); 5535 else 5536 dev_core_stats_rx_nohandler_inc(skb->dev); 5537 kfree_skb_reason(skb, SKB_DROP_REASON_UNHANDLED_PROTO); 5538 /* Jamal, now you will not able to escape explaining 5539 * me how you were going to use this. :-) 5540 */ 5541 ret = NET_RX_DROP; 5542 } 5543 5544 out: 5545 /* The invariant here is that if *ppt_prev is not NULL 5546 * then skb should also be non-NULL. 5547 * 5548 * Apparently *ppt_prev assignment above holds this invariant due to 5549 * skb dereferencing near it. 5550 */ 5551 *pskb = skb; 5552 return ret; 5553 } 5554 5555 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc) 5556 { 5557 struct net_device *orig_dev = skb->dev; 5558 struct packet_type *pt_prev = NULL; 5559 int ret; 5560 5561 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5562 if (pt_prev) 5563 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb, 5564 skb->dev, pt_prev, orig_dev); 5565 return ret; 5566 } 5567 5568 /** 5569 * netif_receive_skb_core - special purpose version of netif_receive_skb 5570 * @skb: buffer to process 5571 * 5572 * More direct receive version of netif_receive_skb(). It should 5573 * only be used by callers that have a need to skip RPS and Generic XDP. 5574 * Caller must also take care of handling if ``(page_is_)pfmemalloc``. 5575 * 5576 * This function may only be called from softirq context and interrupts 5577 * should be enabled. 5578 * 5579 * Return values (usually ignored): 5580 * NET_RX_SUCCESS: no congestion 5581 * NET_RX_DROP: packet was dropped 5582 */ 5583 int netif_receive_skb_core(struct sk_buff *skb) 5584 { 5585 int ret; 5586 5587 rcu_read_lock(); 5588 ret = __netif_receive_skb_one_core(skb, false); 5589 rcu_read_unlock(); 5590 5591 return ret; 5592 } 5593 EXPORT_SYMBOL(netif_receive_skb_core); 5594 5595 static inline void __netif_receive_skb_list_ptype(struct list_head *head, 5596 struct packet_type *pt_prev, 5597 struct net_device *orig_dev) 5598 { 5599 struct sk_buff *skb, *next; 5600 5601 if (!pt_prev) 5602 return; 5603 if (list_empty(head)) 5604 return; 5605 if (pt_prev->list_func != NULL) 5606 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv, 5607 ip_list_rcv, head, pt_prev, orig_dev); 5608 else 5609 list_for_each_entry_safe(skb, next, head, list) { 5610 skb_list_del_init(skb); 5611 pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 5612 } 5613 } 5614 5615 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc) 5616 { 5617 /* Fast-path assumptions: 5618 * - There is no RX handler. 5619 * - Only one packet_type matches. 5620 * If either of these fails, we will end up doing some per-packet 5621 * processing in-line, then handling the 'last ptype' for the whole 5622 * sublist. This can't cause out-of-order delivery to any single ptype, 5623 * because the 'last ptype' must be constant across the sublist, and all 5624 * other ptypes are handled per-packet. 5625 */ 5626 /* Current (common) ptype of sublist */ 5627 struct packet_type *pt_curr = NULL; 5628 /* Current (common) orig_dev of sublist */ 5629 struct net_device *od_curr = NULL; 5630 struct list_head sublist; 5631 struct sk_buff *skb, *next; 5632 5633 INIT_LIST_HEAD(&sublist); 5634 list_for_each_entry_safe(skb, next, head, list) { 5635 struct net_device *orig_dev = skb->dev; 5636 struct packet_type *pt_prev = NULL; 5637 5638 skb_list_del_init(skb); 5639 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5640 if (!pt_prev) 5641 continue; 5642 if (pt_curr != pt_prev || od_curr != orig_dev) { 5643 /* dispatch old sublist */ 5644 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5645 /* start new sublist */ 5646 INIT_LIST_HEAD(&sublist); 5647 pt_curr = pt_prev; 5648 od_curr = orig_dev; 5649 } 5650 list_add_tail(&skb->list, &sublist); 5651 } 5652 5653 /* dispatch final sublist */ 5654 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5655 } 5656 5657 static int __netif_receive_skb(struct sk_buff *skb) 5658 { 5659 int ret; 5660 5661 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { 5662 unsigned int noreclaim_flag; 5663 5664 /* 5665 * PFMEMALLOC skbs are special, they should 5666 * - be delivered to SOCK_MEMALLOC sockets only 5667 * - stay away from userspace 5668 * - have bounded memory usage 5669 * 5670 * Use PF_MEMALLOC as this saves us from propagating the allocation 5671 * context down to all allocation sites. 5672 */ 5673 noreclaim_flag = memalloc_noreclaim_save(); 5674 ret = __netif_receive_skb_one_core(skb, true); 5675 memalloc_noreclaim_restore(noreclaim_flag); 5676 } else 5677 ret = __netif_receive_skb_one_core(skb, false); 5678 5679 return ret; 5680 } 5681 5682 static void __netif_receive_skb_list(struct list_head *head) 5683 { 5684 unsigned long noreclaim_flag = 0; 5685 struct sk_buff *skb, *next; 5686 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */ 5687 5688 list_for_each_entry_safe(skb, next, head, list) { 5689 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) { 5690 struct list_head sublist; 5691 5692 /* Handle the previous sublist */ 5693 list_cut_before(&sublist, head, &skb->list); 5694 if (!list_empty(&sublist)) 5695 __netif_receive_skb_list_core(&sublist, pfmemalloc); 5696 pfmemalloc = !pfmemalloc; 5697 /* See comments in __netif_receive_skb */ 5698 if (pfmemalloc) 5699 noreclaim_flag = memalloc_noreclaim_save(); 5700 else 5701 memalloc_noreclaim_restore(noreclaim_flag); 5702 } 5703 } 5704 /* Handle the remaining sublist */ 5705 if (!list_empty(head)) 5706 __netif_receive_skb_list_core(head, pfmemalloc); 5707 /* Restore pflags */ 5708 if (pfmemalloc) 5709 memalloc_noreclaim_restore(noreclaim_flag); 5710 } 5711 5712 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp) 5713 { 5714 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog); 5715 struct bpf_prog *new = xdp->prog; 5716 int ret = 0; 5717 5718 switch (xdp->command) { 5719 case XDP_SETUP_PROG: 5720 rcu_assign_pointer(dev->xdp_prog, new); 5721 if (old) 5722 bpf_prog_put(old); 5723 5724 if (old && !new) { 5725 static_branch_dec(&generic_xdp_needed_key); 5726 } else if (new && !old) { 5727 static_branch_inc(&generic_xdp_needed_key); 5728 dev_disable_lro(dev); 5729 dev_disable_gro_hw(dev); 5730 } 5731 break; 5732 5733 default: 5734 ret = -EINVAL; 5735 break; 5736 } 5737 5738 return ret; 5739 } 5740 5741 static int netif_receive_skb_internal(struct sk_buff *skb) 5742 { 5743 int ret; 5744 5745 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb); 5746 5747 if (skb_defer_rx_timestamp(skb)) 5748 return NET_RX_SUCCESS; 5749 5750 rcu_read_lock(); 5751 #ifdef CONFIG_RPS 5752 if (static_branch_unlikely(&rps_needed)) { 5753 struct rps_dev_flow voidflow, *rflow = &voidflow; 5754 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5755 5756 if (cpu >= 0) { 5757 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5758 rcu_read_unlock(); 5759 return ret; 5760 } 5761 } 5762 #endif 5763 ret = __netif_receive_skb(skb); 5764 rcu_read_unlock(); 5765 return ret; 5766 } 5767 5768 void netif_receive_skb_list_internal(struct list_head *head) 5769 { 5770 struct sk_buff *skb, *next; 5771 struct list_head sublist; 5772 5773 INIT_LIST_HEAD(&sublist); 5774 list_for_each_entry_safe(skb, next, head, list) { 5775 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb); 5776 skb_list_del_init(skb); 5777 if (!skb_defer_rx_timestamp(skb)) 5778 list_add_tail(&skb->list, &sublist); 5779 } 5780 list_splice_init(&sublist, head); 5781 5782 rcu_read_lock(); 5783 #ifdef CONFIG_RPS 5784 if (static_branch_unlikely(&rps_needed)) { 5785 list_for_each_entry_safe(skb, next, head, list) { 5786 struct rps_dev_flow voidflow, *rflow = &voidflow; 5787 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5788 5789 if (cpu >= 0) { 5790 /* Will be handled, remove from list */ 5791 skb_list_del_init(skb); 5792 enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5793 } 5794 } 5795 } 5796 #endif 5797 __netif_receive_skb_list(head); 5798 rcu_read_unlock(); 5799 } 5800 5801 /** 5802 * netif_receive_skb - process receive buffer from network 5803 * @skb: buffer to process 5804 * 5805 * netif_receive_skb() is the main receive data processing function. 5806 * It always succeeds. The buffer may be dropped during processing 5807 * for congestion control or by the protocol layers. 5808 * 5809 * This function may only be called from softirq context and interrupts 5810 * should be enabled. 5811 * 5812 * Return values (usually ignored): 5813 * NET_RX_SUCCESS: no congestion 5814 * NET_RX_DROP: packet was dropped 5815 */ 5816 int netif_receive_skb(struct sk_buff *skb) 5817 { 5818 int ret; 5819 5820 trace_netif_receive_skb_entry(skb); 5821 5822 ret = netif_receive_skb_internal(skb); 5823 trace_netif_receive_skb_exit(ret); 5824 5825 return ret; 5826 } 5827 EXPORT_SYMBOL(netif_receive_skb); 5828 5829 /** 5830 * netif_receive_skb_list - process many receive buffers from network 5831 * @head: list of skbs to process. 5832 * 5833 * Since return value of netif_receive_skb() is normally ignored, and 5834 * wouldn't be meaningful for a list, this function returns void. 5835 * 5836 * This function may only be called from softirq context and interrupts 5837 * should be enabled. 5838 */ 5839 void netif_receive_skb_list(struct list_head *head) 5840 { 5841 struct sk_buff *skb; 5842 5843 if (list_empty(head)) 5844 return; 5845 if (trace_netif_receive_skb_list_entry_enabled()) { 5846 list_for_each_entry(skb, head, list) 5847 trace_netif_receive_skb_list_entry(skb); 5848 } 5849 netif_receive_skb_list_internal(head); 5850 trace_netif_receive_skb_list_exit(0); 5851 } 5852 EXPORT_SYMBOL(netif_receive_skb_list); 5853 5854 static DEFINE_PER_CPU(struct work_struct, flush_works); 5855 5856 /* Network device is going away, flush any packets still pending */ 5857 static void flush_backlog(struct work_struct *work) 5858 { 5859 struct sk_buff *skb, *tmp; 5860 struct softnet_data *sd; 5861 5862 local_bh_disable(); 5863 sd = this_cpu_ptr(&softnet_data); 5864 5865 rps_lock_irq_disable(sd); 5866 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { 5867 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5868 __skb_unlink(skb, &sd->input_pkt_queue); 5869 dev_kfree_skb_irq(skb); 5870 input_queue_head_incr(sd); 5871 } 5872 } 5873 rps_unlock_irq_enable(sd); 5874 5875 skb_queue_walk_safe(&sd->process_queue, skb, tmp) { 5876 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5877 __skb_unlink(skb, &sd->process_queue); 5878 kfree_skb(skb); 5879 input_queue_head_incr(sd); 5880 } 5881 } 5882 local_bh_enable(); 5883 } 5884 5885 static bool flush_required(int cpu) 5886 { 5887 #if IS_ENABLED(CONFIG_RPS) 5888 struct softnet_data *sd = &per_cpu(softnet_data, cpu); 5889 bool do_flush; 5890 5891 rps_lock_irq_disable(sd); 5892 5893 /* as insertion into process_queue happens with the rps lock held, 5894 * process_queue access may race only with dequeue 5895 */ 5896 do_flush = !skb_queue_empty(&sd->input_pkt_queue) || 5897 !skb_queue_empty_lockless(&sd->process_queue); 5898 rps_unlock_irq_enable(sd); 5899 5900 return do_flush; 5901 #endif 5902 /* without RPS we can't safely check input_pkt_queue: during a 5903 * concurrent remote skb_queue_splice() we can detect as empty both 5904 * input_pkt_queue and process_queue even if the latter could end-up 5905 * containing a lot of packets. 5906 */ 5907 return true; 5908 } 5909 5910 static void flush_all_backlogs(void) 5911 { 5912 static cpumask_t flush_cpus; 5913 unsigned int cpu; 5914 5915 /* since we are under rtnl lock protection we can use static data 5916 * for the cpumask and avoid allocating on stack the possibly 5917 * large mask 5918 */ 5919 ASSERT_RTNL(); 5920 5921 cpus_read_lock(); 5922 5923 cpumask_clear(&flush_cpus); 5924 for_each_online_cpu(cpu) { 5925 if (flush_required(cpu)) { 5926 queue_work_on(cpu, system_highpri_wq, 5927 per_cpu_ptr(&flush_works, cpu)); 5928 cpumask_set_cpu(cpu, &flush_cpus); 5929 } 5930 } 5931 5932 /* we can have in flight packet[s] on the cpus we are not flushing, 5933 * synchronize_net() in unregister_netdevice_many() will take care of 5934 * them 5935 */ 5936 for_each_cpu(cpu, &flush_cpus) 5937 flush_work(per_cpu_ptr(&flush_works, cpu)); 5938 5939 cpus_read_unlock(); 5940 } 5941 5942 static void net_rps_send_ipi(struct softnet_data *remsd) 5943 { 5944 #ifdef CONFIG_RPS 5945 while (remsd) { 5946 struct softnet_data *next = remsd->rps_ipi_next; 5947 5948 if (cpu_online(remsd->cpu)) 5949 smp_call_function_single_async(remsd->cpu, &remsd->csd); 5950 remsd = next; 5951 } 5952 #endif 5953 } 5954 5955 /* 5956 * net_rps_action_and_irq_enable sends any pending IPI's for rps. 5957 * Note: called with local irq disabled, but exits with local irq enabled. 5958 */ 5959 static void net_rps_action_and_irq_enable(struct softnet_data *sd) 5960 { 5961 #ifdef CONFIG_RPS 5962 struct softnet_data *remsd = sd->rps_ipi_list; 5963 5964 if (remsd) { 5965 sd->rps_ipi_list = NULL; 5966 5967 local_irq_enable(); 5968 5969 /* Send pending IPI's to kick RPS processing on remote cpus. */ 5970 net_rps_send_ipi(remsd); 5971 } else 5972 #endif 5973 local_irq_enable(); 5974 } 5975 5976 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) 5977 { 5978 #ifdef CONFIG_RPS 5979 return sd->rps_ipi_list != NULL; 5980 #else 5981 return false; 5982 #endif 5983 } 5984 5985 static int process_backlog(struct napi_struct *napi, int quota) 5986 { 5987 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); 5988 bool again = true; 5989 int work = 0; 5990 5991 /* Check if we have pending ipi, its better to send them now, 5992 * not waiting net_rx_action() end. 5993 */ 5994 if (sd_has_rps_ipi_waiting(sd)) { 5995 local_irq_disable(); 5996 net_rps_action_and_irq_enable(sd); 5997 } 5998 5999 napi->weight = READ_ONCE(dev_rx_weight); 6000 while (again) { 6001 struct sk_buff *skb; 6002 6003 while ((skb = __skb_dequeue(&sd->process_queue))) { 6004 rcu_read_lock(); 6005 __netif_receive_skb(skb); 6006 rcu_read_unlock(); 6007 input_queue_head_incr(sd); 6008 if (++work >= quota) 6009 return work; 6010 6011 } 6012 6013 rps_lock_irq_disable(sd); 6014 if (skb_queue_empty(&sd->input_pkt_queue)) { 6015 /* 6016 * Inline a custom version of __napi_complete(). 6017 * only current cpu owns and manipulates this napi, 6018 * and NAPI_STATE_SCHED is the only possible flag set 6019 * on backlog. 6020 * We can use a plain write instead of clear_bit(), 6021 * and we dont need an smp_mb() memory barrier. 6022 */ 6023 napi->state = 0; 6024 again = false; 6025 } else { 6026 skb_queue_splice_tail_init(&sd->input_pkt_queue, 6027 &sd->process_queue); 6028 } 6029 rps_unlock_irq_enable(sd); 6030 } 6031 6032 return work; 6033 } 6034 6035 /** 6036 * __napi_schedule - schedule for receive 6037 * @n: entry to schedule 6038 * 6039 * The entry's receive function will be scheduled to run. 6040 * Consider using __napi_schedule_irqoff() if hard irqs are masked. 6041 */ 6042 void __napi_schedule(struct napi_struct *n) 6043 { 6044 unsigned long flags; 6045 6046 local_irq_save(flags); 6047 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6048 local_irq_restore(flags); 6049 } 6050 EXPORT_SYMBOL(__napi_schedule); 6051 6052 /** 6053 * napi_schedule_prep - check if napi can be scheduled 6054 * @n: napi context 6055 * 6056 * Test if NAPI routine is already running, and if not mark 6057 * it as running. This is used as a condition variable to 6058 * insure only one NAPI poll instance runs. We also make 6059 * sure there is no pending NAPI disable. 6060 */ 6061 bool napi_schedule_prep(struct napi_struct *n) 6062 { 6063 unsigned long new, val = READ_ONCE(n->state); 6064 6065 do { 6066 if (unlikely(val & NAPIF_STATE_DISABLE)) 6067 return false; 6068 new = val | NAPIF_STATE_SCHED; 6069 6070 /* Sets STATE_MISSED bit if STATE_SCHED was already set 6071 * This was suggested by Alexander Duyck, as compiler 6072 * emits better code than : 6073 * if (val & NAPIF_STATE_SCHED) 6074 * new |= NAPIF_STATE_MISSED; 6075 */ 6076 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * 6077 NAPIF_STATE_MISSED; 6078 } while (!try_cmpxchg(&n->state, &val, new)); 6079 6080 return !(val & NAPIF_STATE_SCHED); 6081 } 6082 EXPORT_SYMBOL(napi_schedule_prep); 6083 6084 /** 6085 * __napi_schedule_irqoff - schedule for receive 6086 * @n: entry to schedule 6087 * 6088 * Variant of __napi_schedule() assuming hard irqs are masked. 6089 * 6090 * On PREEMPT_RT enabled kernels this maps to __napi_schedule() 6091 * because the interrupt disabled assumption might not be true 6092 * due to force-threaded interrupts and spinlock substitution. 6093 */ 6094 void __napi_schedule_irqoff(struct napi_struct *n) 6095 { 6096 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6097 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6098 else 6099 __napi_schedule(n); 6100 } 6101 EXPORT_SYMBOL(__napi_schedule_irqoff); 6102 6103 bool napi_complete_done(struct napi_struct *n, int work_done) 6104 { 6105 unsigned long flags, val, new, timeout = 0; 6106 bool ret = true; 6107 6108 /* 6109 * 1) Don't let napi dequeue from the cpu poll list 6110 * just in case its running on a different cpu. 6111 * 2) If we are busy polling, do nothing here, we have 6112 * the guarantee we will be called later. 6113 */ 6114 if (unlikely(n->state & (NAPIF_STATE_NPSVC | 6115 NAPIF_STATE_IN_BUSY_POLL))) 6116 return false; 6117 6118 if (work_done) { 6119 if (n->gro_bitmask) 6120 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6121 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs); 6122 } 6123 if (n->defer_hard_irqs_count > 0) { 6124 n->defer_hard_irqs_count--; 6125 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6126 if (timeout) 6127 ret = false; 6128 } 6129 if (n->gro_bitmask) { 6130 /* When the NAPI instance uses a timeout and keeps postponing 6131 * it, we need to bound somehow the time packets are kept in 6132 * the GRO layer 6133 */ 6134 napi_gro_flush(n, !!timeout); 6135 } 6136 6137 gro_normal_list(n); 6138 6139 if (unlikely(!list_empty(&n->poll_list))) { 6140 /* If n->poll_list is not empty, we need to mask irqs */ 6141 local_irq_save(flags); 6142 list_del_init(&n->poll_list); 6143 local_irq_restore(flags); 6144 } 6145 WRITE_ONCE(n->list_owner, -1); 6146 6147 val = READ_ONCE(n->state); 6148 do { 6149 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); 6150 6151 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED | 6152 NAPIF_STATE_SCHED_THREADED | 6153 NAPIF_STATE_PREFER_BUSY_POLL); 6154 6155 /* If STATE_MISSED was set, leave STATE_SCHED set, 6156 * because we will call napi->poll() one more time. 6157 * This C code was suggested by Alexander Duyck to help gcc. 6158 */ 6159 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * 6160 NAPIF_STATE_SCHED; 6161 } while (!try_cmpxchg(&n->state, &val, new)); 6162 6163 if (unlikely(val & NAPIF_STATE_MISSED)) { 6164 __napi_schedule(n); 6165 return false; 6166 } 6167 6168 if (timeout) 6169 hrtimer_start(&n->timer, ns_to_ktime(timeout), 6170 HRTIMER_MODE_REL_PINNED); 6171 return ret; 6172 } 6173 EXPORT_SYMBOL(napi_complete_done); 6174 6175 /* must be called under rcu_read_lock(), as we dont take a reference */ 6176 static struct napi_struct *napi_by_id(unsigned int napi_id) 6177 { 6178 unsigned int hash = napi_id % HASH_SIZE(napi_hash); 6179 struct napi_struct *napi; 6180 6181 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) 6182 if (napi->napi_id == napi_id) 6183 return napi; 6184 6185 return NULL; 6186 } 6187 6188 #if defined(CONFIG_NET_RX_BUSY_POLL) 6189 6190 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule) 6191 { 6192 if (!skip_schedule) { 6193 gro_normal_list(napi); 6194 __napi_schedule(napi); 6195 return; 6196 } 6197 6198 if (napi->gro_bitmask) { 6199 /* flush too old packets 6200 * If HZ < 1000, flush all packets. 6201 */ 6202 napi_gro_flush(napi, HZ >= 1000); 6203 } 6204 6205 gro_normal_list(napi); 6206 clear_bit(NAPI_STATE_SCHED, &napi->state); 6207 } 6208 6209 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll, 6210 u16 budget) 6211 { 6212 bool skip_schedule = false; 6213 unsigned long timeout; 6214 int rc; 6215 6216 /* Busy polling means there is a high chance device driver hard irq 6217 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was 6218 * set in napi_schedule_prep(). 6219 * Since we are about to call napi->poll() once more, we can safely 6220 * clear NAPI_STATE_MISSED. 6221 * 6222 * Note: x86 could use a single "lock and ..." instruction 6223 * to perform these two clear_bit() 6224 */ 6225 clear_bit(NAPI_STATE_MISSED, &napi->state); 6226 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); 6227 6228 local_bh_disable(); 6229 6230 if (prefer_busy_poll) { 6231 napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs); 6232 timeout = READ_ONCE(napi->dev->gro_flush_timeout); 6233 if (napi->defer_hard_irqs_count && timeout) { 6234 hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED); 6235 skip_schedule = true; 6236 } 6237 } 6238 6239 /* All we really want here is to re-enable device interrupts. 6240 * Ideally, a new ndo_busy_poll_stop() could avoid another round. 6241 */ 6242 rc = napi->poll(napi, budget); 6243 /* We can't gro_normal_list() here, because napi->poll() might have 6244 * rearmed the napi (napi_complete_done()) in which case it could 6245 * already be running on another CPU. 6246 */ 6247 trace_napi_poll(napi, rc, budget); 6248 netpoll_poll_unlock(have_poll_lock); 6249 if (rc == budget) 6250 __busy_poll_stop(napi, skip_schedule); 6251 local_bh_enable(); 6252 } 6253 6254 void napi_busy_loop(unsigned int napi_id, 6255 bool (*loop_end)(void *, unsigned long), 6256 void *loop_end_arg, bool prefer_busy_poll, u16 budget) 6257 { 6258 unsigned long start_time = loop_end ? busy_loop_current_time() : 0; 6259 int (*napi_poll)(struct napi_struct *napi, int budget); 6260 void *have_poll_lock = NULL; 6261 struct napi_struct *napi; 6262 6263 restart: 6264 napi_poll = NULL; 6265 6266 rcu_read_lock(); 6267 6268 napi = napi_by_id(napi_id); 6269 if (!napi) 6270 goto out; 6271 6272 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6273 preempt_disable(); 6274 for (;;) { 6275 int work = 0; 6276 6277 local_bh_disable(); 6278 if (!napi_poll) { 6279 unsigned long val = READ_ONCE(napi->state); 6280 6281 /* If multiple threads are competing for this napi, 6282 * we avoid dirtying napi->state as much as we can. 6283 */ 6284 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | 6285 NAPIF_STATE_IN_BUSY_POLL)) { 6286 if (prefer_busy_poll) 6287 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6288 goto count; 6289 } 6290 if (cmpxchg(&napi->state, val, 6291 val | NAPIF_STATE_IN_BUSY_POLL | 6292 NAPIF_STATE_SCHED) != val) { 6293 if (prefer_busy_poll) 6294 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6295 goto count; 6296 } 6297 have_poll_lock = netpoll_poll_lock(napi); 6298 napi_poll = napi->poll; 6299 } 6300 work = napi_poll(napi, budget); 6301 trace_napi_poll(napi, work, budget); 6302 gro_normal_list(napi); 6303 count: 6304 if (work > 0) 6305 __NET_ADD_STATS(dev_net(napi->dev), 6306 LINUX_MIB_BUSYPOLLRXPACKETS, work); 6307 local_bh_enable(); 6308 6309 if (!loop_end || loop_end(loop_end_arg, start_time)) 6310 break; 6311 6312 if (unlikely(need_resched())) { 6313 if (napi_poll) 6314 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6315 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6316 preempt_enable(); 6317 rcu_read_unlock(); 6318 cond_resched(); 6319 if (loop_end(loop_end_arg, start_time)) 6320 return; 6321 goto restart; 6322 } 6323 cpu_relax(); 6324 } 6325 if (napi_poll) 6326 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6327 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6328 preempt_enable(); 6329 out: 6330 rcu_read_unlock(); 6331 } 6332 EXPORT_SYMBOL(napi_busy_loop); 6333 6334 #endif /* CONFIG_NET_RX_BUSY_POLL */ 6335 6336 static void napi_hash_add(struct napi_struct *napi) 6337 { 6338 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state)) 6339 return; 6340 6341 spin_lock(&napi_hash_lock); 6342 6343 /* 0..NR_CPUS range is reserved for sender_cpu use */ 6344 do { 6345 if (unlikely(++napi_gen_id < MIN_NAPI_ID)) 6346 napi_gen_id = MIN_NAPI_ID; 6347 } while (napi_by_id(napi_gen_id)); 6348 napi->napi_id = napi_gen_id; 6349 6350 hlist_add_head_rcu(&napi->napi_hash_node, 6351 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); 6352 6353 spin_unlock(&napi_hash_lock); 6354 } 6355 6356 /* Warning : caller is responsible to make sure rcu grace period 6357 * is respected before freeing memory containing @napi 6358 */ 6359 static void napi_hash_del(struct napi_struct *napi) 6360 { 6361 spin_lock(&napi_hash_lock); 6362 6363 hlist_del_init_rcu(&napi->napi_hash_node); 6364 6365 spin_unlock(&napi_hash_lock); 6366 } 6367 6368 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) 6369 { 6370 struct napi_struct *napi; 6371 6372 napi = container_of(timer, struct napi_struct, timer); 6373 6374 /* Note : we use a relaxed variant of napi_schedule_prep() not setting 6375 * NAPI_STATE_MISSED, since we do not react to a device IRQ. 6376 */ 6377 if (!napi_disable_pending(napi) && 6378 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) { 6379 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6380 __napi_schedule_irqoff(napi); 6381 } 6382 6383 return HRTIMER_NORESTART; 6384 } 6385 6386 static void init_gro_hash(struct napi_struct *napi) 6387 { 6388 int i; 6389 6390 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6391 INIT_LIST_HEAD(&napi->gro_hash[i].list); 6392 napi->gro_hash[i].count = 0; 6393 } 6394 napi->gro_bitmask = 0; 6395 } 6396 6397 int dev_set_threaded(struct net_device *dev, bool threaded) 6398 { 6399 struct napi_struct *napi; 6400 int err = 0; 6401 6402 if (dev->threaded == threaded) 6403 return 0; 6404 6405 if (threaded) { 6406 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6407 if (!napi->thread) { 6408 err = napi_kthread_create(napi); 6409 if (err) { 6410 threaded = false; 6411 break; 6412 } 6413 } 6414 } 6415 } 6416 6417 dev->threaded = threaded; 6418 6419 /* Make sure kthread is created before THREADED bit 6420 * is set. 6421 */ 6422 smp_mb__before_atomic(); 6423 6424 /* Setting/unsetting threaded mode on a napi might not immediately 6425 * take effect, if the current napi instance is actively being 6426 * polled. In this case, the switch between threaded mode and 6427 * softirq mode will happen in the next round of napi_schedule(). 6428 * This should not cause hiccups/stalls to the live traffic. 6429 */ 6430 list_for_each_entry(napi, &dev->napi_list, dev_list) 6431 assign_bit(NAPI_STATE_THREADED, &napi->state, threaded); 6432 6433 return err; 6434 } 6435 EXPORT_SYMBOL(dev_set_threaded); 6436 6437 void netif_napi_add_weight(struct net_device *dev, struct napi_struct *napi, 6438 int (*poll)(struct napi_struct *, int), int weight) 6439 { 6440 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state))) 6441 return; 6442 6443 INIT_LIST_HEAD(&napi->poll_list); 6444 INIT_HLIST_NODE(&napi->napi_hash_node); 6445 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); 6446 napi->timer.function = napi_watchdog; 6447 init_gro_hash(napi); 6448 napi->skb = NULL; 6449 INIT_LIST_HEAD(&napi->rx_list); 6450 napi->rx_count = 0; 6451 napi->poll = poll; 6452 if (weight > NAPI_POLL_WEIGHT) 6453 netdev_err_once(dev, "%s() called with weight %d\n", __func__, 6454 weight); 6455 napi->weight = weight; 6456 napi->dev = dev; 6457 #ifdef CONFIG_NETPOLL 6458 napi->poll_owner = -1; 6459 #endif 6460 napi->list_owner = -1; 6461 set_bit(NAPI_STATE_SCHED, &napi->state); 6462 set_bit(NAPI_STATE_NPSVC, &napi->state); 6463 list_add_rcu(&napi->dev_list, &dev->napi_list); 6464 napi_hash_add(napi); 6465 napi_get_frags_check(napi); 6466 /* Create kthread for this napi if dev->threaded is set. 6467 * Clear dev->threaded if kthread creation failed so that 6468 * threaded mode will not be enabled in napi_enable(). 6469 */ 6470 if (dev->threaded && napi_kthread_create(napi)) 6471 dev->threaded = 0; 6472 } 6473 EXPORT_SYMBOL(netif_napi_add_weight); 6474 6475 void napi_disable(struct napi_struct *n) 6476 { 6477 unsigned long val, new; 6478 6479 might_sleep(); 6480 set_bit(NAPI_STATE_DISABLE, &n->state); 6481 6482 val = READ_ONCE(n->state); 6483 do { 6484 while (val & (NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC)) { 6485 usleep_range(20, 200); 6486 val = READ_ONCE(n->state); 6487 } 6488 6489 new = val | NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC; 6490 new &= ~(NAPIF_STATE_THREADED | NAPIF_STATE_PREFER_BUSY_POLL); 6491 } while (!try_cmpxchg(&n->state, &val, new)); 6492 6493 hrtimer_cancel(&n->timer); 6494 6495 clear_bit(NAPI_STATE_DISABLE, &n->state); 6496 } 6497 EXPORT_SYMBOL(napi_disable); 6498 6499 /** 6500 * napi_enable - enable NAPI scheduling 6501 * @n: NAPI context 6502 * 6503 * Resume NAPI from being scheduled on this context. 6504 * Must be paired with napi_disable. 6505 */ 6506 void napi_enable(struct napi_struct *n) 6507 { 6508 unsigned long new, val = READ_ONCE(n->state); 6509 6510 do { 6511 BUG_ON(!test_bit(NAPI_STATE_SCHED, &val)); 6512 6513 new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC); 6514 if (n->dev->threaded && n->thread) 6515 new |= NAPIF_STATE_THREADED; 6516 } while (!try_cmpxchg(&n->state, &val, new)); 6517 } 6518 EXPORT_SYMBOL(napi_enable); 6519 6520 static void flush_gro_hash(struct napi_struct *napi) 6521 { 6522 int i; 6523 6524 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6525 struct sk_buff *skb, *n; 6526 6527 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list) 6528 kfree_skb(skb); 6529 napi->gro_hash[i].count = 0; 6530 } 6531 } 6532 6533 /* Must be called in process context */ 6534 void __netif_napi_del(struct napi_struct *napi) 6535 { 6536 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state)) 6537 return; 6538 6539 napi_hash_del(napi); 6540 list_del_rcu(&napi->dev_list); 6541 napi_free_frags(napi); 6542 6543 flush_gro_hash(napi); 6544 napi->gro_bitmask = 0; 6545 6546 if (napi->thread) { 6547 kthread_stop(napi->thread); 6548 napi->thread = NULL; 6549 } 6550 } 6551 EXPORT_SYMBOL(__netif_napi_del); 6552 6553 static int __napi_poll(struct napi_struct *n, bool *repoll) 6554 { 6555 int work, weight; 6556 6557 weight = n->weight; 6558 6559 /* This NAPI_STATE_SCHED test is for avoiding a race 6560 * with netpoll's poll_napi(). Only the entity which 6561 * obtains the lock and sees NAPI_STATE_SCHED set will 6562 * actually make the ->poll() call. Therefore we avoid 6563 * accidentally calling ->poll() when NAPI is not scheduled. 6564 */ 6565 work = 0; 6566 if (test_bit(NAPI_STATE_SCHED, &n->state)) { 6567 work = n->poll(n, weight); 6568 trace_napi_poll(n, work, weight); 6569 } 6570 6571 if (unlikely(work > weight)) 6572 netdev_err_once(n->dev, "NAPI poll function %pS returned %d, exceeding its budget of %d.\n", 6573 n->poll, work, weight); 6574 6575 if (likely(work < weight)) 6576 return work; 6577 6578 /* Drivers must not modify the NAPI state if they 6579 * consume the entire weight. In such cases this code 6580 * still "owns" the NAPI instance and therefore can 6581 * move the instance around on the list at-will. 6582 */ 6583 if (unlikely(napi_disable_pending(n))) { 6584 napi_complete(n); 6585 return work; 6586 } 6587 6588 /* The NAPI context has more processing work, but busy-polling 6589 * is preferred. Exit early. 6590 */ 6591 if (napi_prefer_busy_poll(n)) { 6592 if (napi_complete_done(n, work)) { 6593 /* If timeout is not set, we need to make sure 6594 * that the NAPI is re-scheduled. 6595 */ 6596 napi_schedule(n); 6597 } 6598 return work; 6599 } 6600 6601 if (n->gro_bitmask) { 6602 /* flush too old packets 6603 * If HZ < 1000, flush all packets. 6604 */ 6605 napi_gro_flush(n, HZ >= 1000); 6606 } 6607 6608 gro_normal_list(n); 6609 6610 /* Some drivers may have called napi_schedule 6611 * prior to exhausting their budget. 6612 */ 6613 if (unlikely(!list_empty(&n->poll_list))) { 6614 pr_warn_once("%s: Budget exhausted after napi rescheduled\n", 6615 n->dev ? n->dev->name : "backlog"); 6616 return work; 6617 } 6618 6619 *repoll = true; 6620 6621 return work; 6622 } 6623 6624 static int napi_poll(struct napi_struct *n, struct list_head *repoll) 6625 { 6626 bool do_repoll = false; 6627 void *have; 6628 int work; 6629 6630 list_del_init(&n->poll_list); 6631 6632 have = netpoll_poll_lock(n); 6633 6634 work = __napi_poll(n, &do_repoll); 6635 6636 if (do_repoll) 6637 list_add_tail(&n->poll_list, repoll); 6638 6639 netpoll_poll_unlock(have); 6640 6641 return work; 6642 } 6643 6644 static int napi_thread_wait(struct napi_struct *napi) 6645 { 6646 bool woken = false; 6647 6648 set_current_state(TASK_INTERRUPTIBLE); 6649 6650 while (!kthread_should_stop()) { 6651 /* Testing SCHED_THREADED bit here to make sure the current 6652 * kthread owns this napi and could poll on this napi. 6653 * Testing SCHED bit is not enough because SCHED bit might be 6654 * set by some other busy poll thread or by napi_disable(). 6655 */ 6656 if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state) || woken) { 6657 WARN_ON(!list_empty(&napi->poll_list)); 6658 __set_current_state(TASK_RUNNING); 6659 return 0; 6660 } 6661 6662 schedule(); 6663 /* woken being true indicates this thread owns this napi. */ 6664 woken = true; 6665 set_current_state(TASK_INTERRUPTIBLE); 6666 } 6667 __set_current_state(TASK_RUNNING); 6668 6669 return -1; 6670 } 6671 6672 static void skb_defer_free_flush(struct softnet_data *sd) 6673 { 6674 struct sk_buff *skb, *next; 6675 6676 /* Paired with WRITE_ONCE() in skb_attempt_defer_free() */ 6677 if (!READ_ONCE(sd->defer_list)) 6678 return; 6679 6680 spin_lock(&sd->defer_lock); 6681 skb = sd->defer_list; 6682 sd->defer_list = NULL; 6683 sd->defer_count = 0; 6684 spin_unlock(&sd->defer_lock); 6685 6686 while (skb != NULL) { 6687 next = skb->next; 6688 napi_consume_skb(skb, 1); 6689 skb = next; 6690 } 6691 } 6692 6693 static int napi_threaded_poll(void *data) 6694 { 6695 struct napi_struct *napi = data; 6696 struct softnet_data *sd; 6697 void *have; 6698 6699 while (!napi_thread_wait(napi)) { 6700 unsigned long last_qs = jiffies; 6701 6702 for (;;) { 6703 bool repoll = false; 6704 6705 local_bh_disable(); 6706 sd = this_cpu_ptr(&softnet_data); 6707 sd->in_napi_threaded_poll = true; 6708 6709 have = netpoll_poll_lock(napi); 6710 __napi_poll(napi, &repoll); 6711 netpoll_poll_unlock(have); 6712 6713 sd->in_napi_threaded_poll = false; 6714 barrier(); 6715 6716 if (sd_has_rps_ipi_waiting(sd)) { 6717 local_irq_disable(); 6718 net_rps_action_and_irq_enable(sd); 6719 } 6720 skb_defer_free_flush(sd); 6721 local_bh_enable(); 6722 6723 if (!repoll) 6724 break; 6725 6726 rcu_softirq_qs_periodic(last_qs); 6727 cond_resched(); 6728 } 6729 } 6730 return 0; 6731 } 6732 6733 static __latent_entropy void net_rx_action(struct softirq_action *h) 6734 { 6735 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 6736 unsigned long time_limit = jiffies + 6737 usecs_to_jiffies(READ_ONCE(netdev_budget_usecs)); 6738 int budget = READ_ONCE(netdev_budget); 6739 LIST_HEAD(list); 6740 LIST_HEAD(repoll); 6741 6742 start: 6743 sd->in_net_rx_action = true; 6744 local_irq_disable(); 6745 list_splice_init(&sd->poll_list, &list); 6746 local_irq_enable(); 6747 6748 for (;;) { 6749 struct napi_struct *n; 6750 6751 skb_defer_free_flush(sd); 6752 6753 if (list_empty(&list)) { 6754 if (list_empty(&repoll)) { 6755 sd->in_net_rx_action = false; 6756 barrier(); 6757 /* We need to check if ____napi_schedule() 6758 * had refilled poll_list while 6759 * sd->in_net_rx_action was true. 6760 */ 6761 if (!list_empty(&sd->poll_list)) 6762 goto start; 6763 if (!sd_has_rps_ipi_waiting(sd)) 6764 goto end; 6765 } 6766 break; 6767 } 6768 6769 n = list_first_entry(&list, struct napi_struct, poll_list); 6770 budget -= napi_poll(n, &repoll); 6771 6772 /* If softirq window is exhausted then punt. 6773 * Allow this to run for 2 jiffies since which will allow 6774 * an average latency of 1.5/HZ. 6775 */ 6776 if (unlikely(budget <= 0 || 6777 time_after_eq(jiffies, time_limit))) { 6778 sd->time_squeeze++; 6779 break; 6780 } 6781 } 6782 6783 local_irq_disable(); 6784 6785 list_splice_tail_init(&sd->poll_list, &list); 6786 list_splice_tail(&repoll, &list); 6787 list_splice(&list, &sd->poll_list); 6788 if (!list_empty(&sd->poll_list)) 6789 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 6790 else 6791 sd->in_net_rx_action = false; 6792 6793 net_rps_action_and_irq_enable(sd); 6794 end:; 6795 } 6796 6797 struct netdev_adjacent { 6798 struct net_device *dev; 6799 netdevice_tracker dev_tracker; 6800 6801 /* upper master flag, there can only be one master device per list */ 6802 bool master; 6803 6804 /* lookup ignore flag */ 6805 bool ignore; 6806 6807 /* counter for the number of times this device was added to us */ 6808 u16 ref_nr; 6809 6810 /* private field for the users */ 6811 void *private; 6812 6813 struct list_head list; 6814 struct rcu_head rcu; 6815 }; 6816 6817 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, 6818 struct list_head *adj_list) 6819 { 6820 struct netdev_adjacent *adj; 6821 6822 list_for_each_entry(adj, adj_list, list) { 6823 if (adj->dev == adj_dev) 6824 return adj; 6825 } 6826 return NULL; 6827 } 6828 6829 static int ____netdev_has_upper_dev(struct net_device *upper_dev, 6830 struct netdev_nested_priv *priv) 6831 { 6832 struct net_device *dev = (struct net_device *)priv->data; 6833 6834 return upper_dev == dev; 6835 } 6836 6837 /** 6838 * netdev_has_upper_dev - Check if device is linked to an upper device 6839 * @dev: device 6840 * @upper_dev: upper device to check 6841 * 6842 * Find out if a device is linked to specified upper device and return true 6843 * in case it is. Note that this checks only immediate upper device, 6844 * not through a complete stack of devices. The caller must hold the RTNL lock. 6845 */ 6846 bool netdev_has_upper_dev(struct net_device *dev, 6847 struct net_device *upper_dev) 6848 { 6849 struct netdev_nested_priv priv = { 6850 .data = (void *)upper_dev, 6851 }; 6852 6853 ASSERT_RTNL(); 6854 6855 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 6856 &priv); 6857 } 6858 EXPORT_SYMBOL(netdev_has_upper_dev); 6859 6860 /** 6861 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device 6862 * @dev: device 6863 * @upper_dev: upper device to check 6864 * 6865 * Find out if a device is linked to specified upper device and return true 6866 * in case it is. Note that this checks the entire upper device chain. 6867 * The caller must hold rcu lock. 6868 */ 6869 6870 bool netdev_has_upper_dev_all_rcu(struct net_device *dev, 6871 struct net_device *upper_dev) 6872 { 6873 struct netdev_nested_priv priv = { 6874 .data = (void *)upper_dev, 6875 }; 6876 6877 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 6878 &priv); 6879 } 6880 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); 6881 6882 /** 6883 * netdev_has_any_upper_dev - Check if device is linked to some device 6884 * @dev: device 6885 * 6886 * Find out if a device is linked to an upper device and return true in case 6887 * it is. The caller must hold the RTNL lock. 6888 */ 6889 bool netdev_has_any_upper_dev(struct net_device *dev) 6890 { 6891 ASSERT_RTNL(); 6892 6893 return !list_empty(&dev->adj_list.upper); 6894 } 6895 EXPORT_SYMBOL(netdev_has_any_upper_dev); 6896 6897 /** 6898 * netdev_master_upper_dev_get - Get master upper device 6899 * @dev: device 6900 * 6901 * Find a master upper device and return pointer to it or NULL in case 6902 * it's not there. The caller must hold the RTNL lock. 6903 */ 6904 struct net_device *netdev_master_upper_dev_get(struct net_device *dev) 6905 { 6906 struct netdev_adjacent *upper; 6907 6908 ASSERT_RTNL(); 6909 6910 if (list_empty(&dev->adj_list.upper)) 6911 return NULL; 6912 6913 upper = list_first_entry(&dev->adj_list.upper, 6914 struct netdev_adjacent, list); 6915 if (likely(upper->master)) 6916 return upper->dev; 6917 return NULL; 6918 } 6919 EXPORT_SYMBOL(netdev_master_upper_dev_get); 6920 6921 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev) 6922 { 6923 struct netdev_adjacent *upper; 6924 6925 ASSERT_RTNL(); 6926 6927 if (list_empty(&dev->adj_list.upper)) 6928 return NULL; 6929 6930 upper = list_first_entry(&dev->adj_list.upper, 6931 struct netdev_adjacent, list); 6932 if (likely(upper->master) && !upper->ignore) 6933 return upper->dev; 6934 return NULL; 6935 } 6936 6937 /** 6938 * netdev_has_any_lower_dev - Check if device is linked to some device 6939 * @dev: device 6940 * 6941 * Find out if a device is linked to a lower device and return true in case 6942 * it is. The caller must hold the RTNL lock. 6943 */ 6944 static bool netdev_has_any_lower_dev(struct net_device *dev) 6945 { 6946 ASSERT_RTNL(); 6947 6948 return !list_empty(&dev->adj_list.lower); 6949 } 6950 6951 void *netdev_adjacent_get_private(struct list_head *adj_list) 6952 { 6953 struct netdev_adjacent *adj; 6954 6955 adj = list_entry(adj_list, struct netdev_adjacent, list); 6956 6957 return adj->private; 6958 } 6959 EXPORT_SYMBOL(netdev_adjacent_get_private); 6960 6961 /** 6962 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list 6963 * @dev: device 6964 * @iter: list_head ** of the current position 6965 * 6966 * Gets the next device from the dev's upper list, starting from iter 6967 * position. The caller must hold RCU read lock. 6968 */ 6969 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, 6970 struct list_head **iter) 6971 { 6972 struct netdev_adjacent *upper; 6973 6974 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 6975 6976 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 6977 6978 if (&upper->list == &dev->adj_list.upper) 6979 return NULL; 6980 6981 *iter = &upper->list; 6982 6983 return upper->dev; 6984 } 6985 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); 6986 6987 static struct net_device *__netdev_next_upper_dev(struct net_device *dev, 6988 struct list_head **iter, 6989 bool *ignore) 6990 { 6991 struct netdev_adjacent *upper; 6992 6993 upper = list_entry((*iter)->next, struct netdev_adjacent, list); 6994 6995 if (&upper->list == &dev->adj_list.upper) 6996 return NULL; 6997 6998 *iter = &upper->list; 6999 *ignore = upper->ignore; 7000 7001 return upper->dev; 7002 } 7003 7004 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, 7005 struct list_head **iter) 7006 { 7007 struct netdev_adjacent *upper; 7008 7009 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 7010 7011 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7012 7013 if (&upper->list == &dev->adj_list.upper) 7014 return NULL; 7015 7016 *iter = &upper->list; 7017 7018 return upper->dev; 7019 } 7020 7021 static int __netdev_walk_all_upper_dev(struct net_device *dev, 7022 int (*fn)(struct net_device *dev, 7023 struct netdev_nested_priv *priv), 7024 struct netdev_nested_priv *priv) 7025 { 7026 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7027 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7028 int ret, cur = 0; 7029 bool ignore; 7030 7031 now = dev; 7032 iter = &dev->adj_list.upper; 7033 7034 while (1) { 7035 if (now != dev) { 7036 ret = fn(now, priv); 7037 if (ret) 7038 return ret; 7039 } 7040 7041 next = NULL; 7042 while (1) { 7043 udev = __netdev_next_upper_dev(now, &iter, &ignore); 7044 if (!udev) 7045 break; 7046 if (ignore) 7047 continue; 7048 7049 next = udev; 7050 niter = &udev->adj_list.upper; 7051 dev_stack[cur] = now; 7052 iter_stack[cur++] = iter; 7053 break; 7054 } 7055 7056 if (!next) { 7057 if (!cur) 7058 return 0; 7059 next = dev_stack[--cur]; 7060 niter = iter_stack[cur]; 7061 } 7062 7063 now = next; 7064 iter = niter; 7065 } 7066 7067 return 0; 7068 } 7069 7070 int netdev_walk_all_upper_dev_rcu(struct net_device *dev, 7071 int (*fn)(struct net_device *dev, 7072 struct netdev_nested_priv *priv), 7073 struct netdev_nested_priv *priv) 7074 { 7075 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7076 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7077 int ret, cur = 0; 7078 7079 now = dev; 7080 iter = &dev->adj_list.upper; 7081 7082 while (1) { 7083 if (now != dev) { 7084 ret = fn(now, priv); 7085 if (ret) 7086 return ret; 7087 } 7088 7089 next = NULL; 7090 while (1) { 7091 udev = netdev_next_upper_dev_rcu(now, &iter); 7092 if (!udev) 7093 break; 7094 7095 next = udev; 7096 niter = &udev->adj_list.upper; 7097 dev_stack[cur] = now; 7098 iter_stack[cur++] = iter; 7099 break; 7100 } 7101 7102 if (!next) { 7103 if (!cur) 7104 return 0; 7105 next = dev_stack[--cur]; 7106 niter = iter_stack[cur]; 7107 } 7108 7109 now = next; 7110 iter = niter; 7111 } 7112 7113 return 0; 7114 } 7115 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); 7116 7117 static bool __netdev_has_upper_dev(struct net_device *dev, 7118 struct net_device *upper_dev) 7119 { 7120 struct netdev_nested_priv priv = { 7121 .flags = 0, 7122 .data = (void *)upper_dev, 7123 }; 7124 7125 ASSERT_RTNL(); 7126 7127 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev, 7128 &priv); 7129 } 7130 7131 /** 7132 * netdev_lower_get_next_private - Get the next ->private from the 7133 * lower neighbour list 7134 * @dev: device 7135 * @iter: list_head ** of the current position 7136 * 7137 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7138 * list, starting from iter position. The caller must hold either hold the 7139 * RTNL lock or its own locking that guarantees that the neighbour lower 7140 * list will remain unchanged. 7141 */ 7142 void *netdev_lower_get_next_private(struct net_device *dev, 7143 struct list_head **iter) 7144 { 7145 struct netdev_adjacent *lower; 7146 7147 lower = list_entry(*iter, struct netdev_adjacent, list); 7148 7149 if (&lower->list == &dev->adj_list.lower) 7150 return NULL; 7151 7152 *iter = lower->list.next; 7153 7154 return lower->private; 7155 } 7156 EXPORT_SYMBOL(netdev_lower_get_next_private); 7157 7158 /** 7159 * netdev_lower_get_next_private_rcu - Get the next ->private from the 7160 * lower neighbour list, RCU 7161 * variant 7162 * @dev: device 7163 * @iter: list_head ** of the current position 7164 * 7165 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7166 * list, starting from iter position. The caller must hold RCU read lock. 7167 */ 7168 void *netdev_lower_get_next_private_rcu(struct net_device *dev, 7169 struct list_head **iter) 7170 { 7171 struct netdev_adjacent *lower; 7172 7173 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 7174 7175 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7176 7177 if (&lower->list == &dev->adj_list.lower) 7178 return NULL; 7179 7180 *iter = &lower->list; 7181 7182 return lower->private; 7183 } 7184 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); 7185 7186 /** 7187 * netdev_lower_get_next - Get the next device from the lower neighbour 7188 * list 7189 * @dev: device 7190 * @iter: list_head ** of the current position 7191 * 7192 * Gets the next netdev_adjacent from the dev's lower neighbour 7193 * list, starting from iter position. The caller must hold RTNL lock or 7194 * its own locking that guarantees that the neighbour lower 7195 * list will remain unchanged. 7196 */ 7197 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) 7198 { 7199 struct netdev_adjacent *lower; 7200 7201 lower = list_entry(*iter, struct netdev_adjacent, list); 7202 7203 if (&lower->list == &dev->adj_list.lower) 7204 return NULL; 7205 7206 *iter = lower->list.next; 7207 7208 return lower->dev; 7209 } 7210 EXPORT_SYMBOL(netdev_lower_get_next); 7211 7212 static struct net_device *netdev_next_lower_dev(struct net_device *dev, 7213 struct list_head **iter) 7214 { 7215 struct netdev_adjacent *lower; 7216 7217 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7218 7219 if (&lower->list == &dev->adj_list.lower) 7220 return NULL; 7221 7222 *iter = &lower->list; 7223 7224 return lower->dev; 7225 } 7226 7227 static struct net_device *__netdev_next_lower_dev(struct net_device *dev, 7228 struct list_head **iter, 7229 bool *ignore) 7230 { 7231 struct netdev_adjacent *lower; 7232 7233 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7234 7235 if (&lower->list == &dev->adj_list.lower) 7236 return NULL; 7237 7238 *iter = &lower->list; 7239 *ignore = lower->ignore; 7240 7241 return lower->dev; 7242 } 7243 7244 int netdev_walk_all_lower_dev(struct net_device *dev, 7245 int (*fn)(struct net_device *dev, 7246 struct netdev_nested_priv *priv), 7247 struct netdev_nested_priv *priv) 7248 { 7249 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7250 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7251 int ret, cur = 0; 7252 7253 now = dev; 7254 iter = &dev->adj_list.lower; 7255 7256 while (1) { 7257 if (now != dev) { 7258 ret = fn(now, priv); 7259 if (ret) 7260 return ret; 7261 } 7262 7263 next = NULL; 7264 while (1) { 7265 ldev = netdev_next_lower_dev(now, &iter); 7266 if (!ldev) 7267 break; 7268 7269 next = ldev; 7270 niter = &ldev->adj_list.lower; 7271 dev_stack[cur] = now; 7272 iter_stack[cur++] = iter; 7273 break; 7274 } 7275 7276 if (!next) { 7277 if (!cur) 7278 return 0; 7279 next = dev_stack[--cur]; 7280 niter = iter_stack[cur]; 7281 } 7282 7283 now = next; 7284 iter = niter; 7285 } 7286 7287 return 0; 7288 } 7289 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); 7290 7291 static int __netdev_walk_all_lower_dev(struct net_device *dev, 7292 int (*fn)(struct net_device *dev, 7293 struct netdev_nested_priv *priv), 7294 struct netdev_nested_priv *priv) 7295 { 7296 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7297 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7298 int ret, cur = 0; 7299 bool ignore; 7300 7301 now = dev; 7302 iter = &dev->adj_list.lower; 7303 7304 while (1) { 7305 if (now != dev) { 7306 ret = fn(now, priv); 7307 if (ret) 7308 return ret; 7309 } 7310 7311 next = NULL; 7312 while (1) { 7313 ldev = __netdev_next_lower_dev(now, &iter, &ignore); 7314 if (!ldev) 7315 break; 7316 if (ignore) 7317 continue; 7318 7319 next = ldev; 7320 niter = &ldev->adj_list.lower; 7321 dev_stack[cur] = now; 7322 iter_stack[cur++] = iter; 7323 break; 7324 } 7325 7326 if (!next) { 7327 if (!cur) 7328 return 0; 7329 next = dev_stack[--cur]; 7330 niter = iter_stack[cur]; 7331 } 7332 7333 now = next; 7334 iter = niter; 7335 } 7336 7337 return 0; 7338 } 7339 7340 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, 7341 struct list_head **iter) 7342 { 7343 struct netdev_adjacent *lower; 7344 7345 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7346 if (&lower->list == &dev->adj_list.lower) 7347 return NULL; 7348 7349 *iter = &lower->list; 7350 7351 return lower->dev; 7352 } 7353 EXPORT_SYMBOL(netdev_next_lower_dev_rcu); 7354 7355 static u8 __netdev_upper_depth(struct net_device *dev) 7356 { 7357 struct net_device *udev; 7358 struct list_head *iter; 7359 u8 max_depth = 0; 7360 bool ignore; 7361 7362 for (iter = &dev->adj_list.upper, 7363 udev = __netdev_next_upper_dev(dev, &iter, &ignore); 7364 udev; 7365 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) { 7366 if (ignore) 7367 continue; 7368 if (max_depth < udev->upper_level) 7369 max_depth = udev->upper_level; 7370 } 7371 7372 return max_depth; 7373 } 7374 7375 static u8 __netdev_lower_depth(struct net_device *dev) 7376 { 7377 struct net_device *ldev; 7378 struct list_head *iter; 7379 u8 max_depth = 0; 7380 bool ignore; 7381 7382 for (iter = &dev->adj_list.lower, 7383 ldev = __netdev_next_lower_dev(dev, &iter, &ignore); 7384 ldev; 7385 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) { 7386 if (ignore) 7387 continue; 7388 if (max_depth < ldev->lower_level) 7389 max_depth = ldev->lower_level; 7390 } 7391 7392 return max_depth; 7393 } 7394 7395 static int __netdev_update_upper_level(struct net_device *dev, 7396 struct netdev_nested_priv *__unused) 7397 { 7398 dev->upper_level = __netdev_upper_depth(dev) + 1; 7399 return 0; 7400 } 7401 7402 #ifdef CONFIG_LOCKDEP 7403 static LIST_HEAD(net_unlink_list); 7404 7405 static void net_unlink_todo(struct net_device *dev) 7406 { 7407 if (list_empty(&dev->unlink_list)) 7408 list_add_tail(&dev->unlink_list, &net_unlink_list); 7409 } 7410 #endif 7411 7412 static int __netdev_update_lower_level(struct net_device *dev, 7413 struct netdev_nested_priv *priv) 7414 { 7415 dev->lower_level = __netdev_lower_depth(dev) + 1; 7416 7417 #ifdef CONFIG_LOCKDEP 7418 if (!priv) 7419 return 0; 7420 7421 if (priv->flags & NESTED_SYNC_IMM) 7422 dev->nested_level = dev->lower_level - 1; 7423 if (priv->flags & NESTED_SYNC_TODO) 7424 net_unlink_todo(dev); 7425 #endif 7426 return 0; 7427 } 7428 7429 int netdev_walk_all_lower_dev_rcu(struct net_device *dev, 7430 int (*fn)(struct net_device *dev, 7431 struct netdev_nested_priv *priv), 7432 struct netdev_nested_priv *priv) 7433 { 7434 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7435 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7436 int ret, cur = 0; 7437 7438 now = dev; 7439 iter = &dev->adj_list.lower; 7440 7441 while (1) { 7442 if (now != dev) { 7443 ret = fn(now, priv); 7444 if (ret) 7445 return ret; 7446 } 7447 7448 next = NULL; 7449 while (1) { 7450 ldev = netdev_next_lower_dev_rcu(now, &iter); 7451 if (!ldev) 7452 break; 7453 7454 next = ldev; 7455 niter = &ldev->adj_list.lower; 7456 dev_stack[cur] = now; 7457 iter_stack[cur++] = iter; 7458 break; 7459 } 7460 7461 if (!next) { 7462 if (!cur) 7463 return 0; 7464 next = dev_stack[--cur]; 7465 niter = iter_stack[cur]; 7466 } 7467 7468 now = next; 7469 iter = niter; 7470 } 7471 7472 return 0; 7473 } 7474 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); 7475 7476 /** 7477 * netdev_lower_get_first_private_rcu - Get the first ->private from the 7478 * lower neighbour list, RCU 7479 * variant 7480 * @dev: device 7481 * 7482 * Gets the first netdev_adjacent->private from the dev's lower neighbour 7483 * list. The caller must hold RCU read lock. 7484 */ 7485 void *netdev_lower_get_first_private_rcu(struct net_device *dev) 7486 { 7487 struct netdev_adjacent *lower; 7488 7489 lower = list_first_or_null_rcu(&dev->adj_list.lower, 7490 struct netdev_adjacent, list); 7491 if (lower) 7492 return lower->private; 7493 return NULL; 7494 } 7495 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); 7496 7497 /** 7498 * netdev_master_upper_dev_get_rcu - Get master upper device 7499 * @dev: device 7500 * 7501 * Find a master upper device and return pointer to it or NULL in case 7502 * it's not there. The caller must hold the RCU read lock. 7503 */ 7504 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) 7505 { 7506 struct netdev_adjacent *upper; 7507 7508 upper = list_first_or_null_rcu(&dev->adj_list.upper, 7509 struct netdev_adjacent, list); 7510 if (upper && likely(upper->master)) 7511 return upper->dev; 7512 return NULL; 7513 } 7514 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); 7515 7516 static int netdev_adjacent_sysfs_add(struct net_device *dev, 7517 struct net_device *adj_dev, 7518 struct list_head *dev_list) 7519 { 7520 char linkname[IFNAMSIZ+7]; 7521 7522 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7523 "upper_%s" : "lower_%s", adj_dev->name); 7524 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), 7525 linkname); 7526 } 7527 static void netdev_adjacent_sysfs_del(struct net_device *dev, 7528 char *name, 7529 struct list_head *dev_list) 7530 { 7531 char linkname[IFNAMSIZ+7]; 7532 7533 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7534 "upper_%s" : "lower_%s", name); 7535 sysfs_remove_link(&(dev->dev.kobj), linkname); 7536 } 7537 7538 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, 7539 struct net_device *adj_dev, 7540 struct list_head *dev_list) 7541 { 7542 return (dev_list == &dev->adj_list.upper || 7543 dev_list == &dev->adj_list.lower) && 7544 net_eq(dev_net(dev), dev_net(adj_dev)); 7545 } 7546 7547 static int __netdev_adjacent_dev_insert(struct net_device *dev, 7548 struct net_device *adj_dev, 7549 struct list_head *dev_list, 7550 void *private, bool master) 7551 { 7552 struct netdev_adjacent *adj; 7553 int ret; 7554 7555 adj = __netdev_find_adj(adj_dev, dev_list); 7556 7557 if (adj) { 7558 adj->ref_nr += 1; 7559 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", 7560 dev->name, adj_dev->name, adj->ref_nr); 7561 7562 return 0; 7563 } 7564 7565 adj = kmalloc(sizeof(*adj), GFP_KERNEL); 7566 if (!adj) 7567 return -ENOMEM; 7568 7569 adj->dev = adj_dev; 7570 adj->master = master; 7571 adj->ref_nr = 1; 7572 adj->private = private; 7573 adj->ignore = false; 7574 netdev_hold(adj_dev, &adj->dev_tracker, GFP_KERNEL); 7575 7576 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", 7577 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); 7578 7579 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { 7580 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); 7581 if (ret) 7582 goto free_adj; 7583 } 7584 7585 /* Ensure that master link is always the first item in list. */ 7586 if (master) { 7587 ret = sysfs_create_link(&(dev->dev.kobj), 7588 &(adj_dev->dev.kobj), "master"); 7589 if (ret) 7590 goto remove_symlinks; 7591 7592 list_add_rcu(&adj->list, dev_list); 7593 } else { 7594 list_add_tail_rcu(&adj->list, dev_list); 7595 } 7596 7597 return 0; 7598 7599 remove_symlinks: 7600 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7601 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7602 free_adj: 7603 netdev_put(adj_dev, &adj->dev_tracker); 7604 kfree(adj); 7605 7606 return ret; 7607 } 7608 7609 static void __netdev_adjacent_dev_remove(struct net_device *dev, 7610 struct net_device *adj_dev, 7611 u16 ref_nr, 7612 struct list_head *dev_list) 7613 { 7614 struct netdev_adjacent *adj; 7615 7616 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", 7617 dev->name, adj_dev->name, ref_nr); 7618 7619 adj = __netdev_find_adj(adj_dev, dev_list); 7620 7621 if (!adj) { 7622 pr_err("Adjacency does not exist for device %s from %s\n", 7623 dev->name, adj_dev->name); 7624 WARN_ON(1); 7625 return; 7626 } 7627 7628 if (adj->ref_nr > ref_nr) { 7629 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", 7630 dev->name, adj_dev->name, ref_nr, 7631 adj->ref_nr - ref_nr); 7632 adj->ref_nr -= ref_nr; 7633 return; 7634 } 7635 7636 if (adj->master) 7637 sysfs_remove_link(&(dev->dev.kobj), "master"); 7638 7639 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7640 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7641 7642 list_del_rcu(&adj->list); 7643 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", 7644 adj_dev->name, dev->name, adj_dev->name); 7645 netdev_put(adj_dev, &adj->dev_tracker); 7646 kfree_rcu(adj, rcu); 7647 } 7648 7649 static int __netdev_adjacent_dev_link_lists(struct net_device *dev, 7650 struct net_device *upper_dev, 7651 struct list_head *up_list, 7652 struct list_head *down_list, 7653 void *private, bool master) 7654 { 7655 int ret; 7656 7657 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, 7658 private, master); 7659 if (ret) 7660 return ret; 7661 7662 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, 7663 private, false); 7664 if (ret) { 7665 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); 7666 return ret; 7667 } 7668 7669 return 0; 7670 } 7671 7672 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, 7673 struct net_device *upper_dev, 7674 u16 ref_nr, 7675 struct list_head *up_list, 7676 struct list_head *down_list) 7677 { 7678 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); 7679 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); 7680 } 7681 7682 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, 7683 struct net_device *upper_dev, 7684 void *private, bool master) 7685 { 7686 return __netdev_adjacent_dev_link_lists(dev, upper_dev, 7687 &dev->adj_list.upper, 7688 &upper_dev->adj_list.lower, 7689 private, master); 7690 } 7691 7692 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, 7693 struct net_device *upper_dev) 7694 { 7695 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, 7696 &dev->adj_list.upper, 7697 &upper_dev->adj_list.lower); 7698 } 7699 7700 static int __netdev_upper_dev_link(struct net_device *dev, 7701 struct net_device *upper_dev, bool master, 7702 void *upper_priv, void *upper_info, 7703 struct netdev_nested_priv *priv, 7704 struct netlink_ext_ack *extack) 7705 { 7706 struct netdev_notifier_changeupper_info changeupper_info = { 7707 .info = { 7708 .dev = dev, 7709 .extack = extack, 7710 }, 7711 .upper_dev = upper_dev, 7712 .master = master, 7713 .linking = true, 7714 .upper_info = upper_info, 7715 }; 7716 struct net_device *master_dev; 7717 int ret = 0; 7718 7719 ASSERT_RTNL(); 7720 7721 if (dev == upper_dev) 7722 return -EBUSY; 7723 7724 /* To prevent loops, check if dev is not upper device to upper_dev. */ 7725 if (__netdev_has_upper_dev(upper_dev, dev)) 7726 return -EBUSY; 7727 7728 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV) 7729 return -EMLINK; 7730 7731 if (!master) { 7732 if (__netdev_has_upper_dev(dev, upper_dev)) 7733 return -EEXIST; 7734 } else { 7735 master_dev = __netdev_master_upper_dev_get(dev); 7736 if (master_dev) 7737 return master_dev == upper_dev ? -EEXIST : -EBUSY; 7738 } 7739 7740 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7741 &changeupper_info.info); 7742 ret = notifier_to_errno(ret); 7743 if (ret) 7744 return ret; 7745 7746 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, 7747 master); 7748 if (ret) 7749 return ret; 7750 7751 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 7752 &changeupper_info.info); 7753 ret = notifier_to_errno(ret); 7754 if (ret) 7755 goto rollback; 7756 7757 __netdev_update_upper_level(dev, NULL); 7758 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 7759 7760 __netdev_update_lower_level(upper_dev, priv); 7761 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 7762 priv); 7763 7764 return 0; 7765 7766 rollback: 7767 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 7768 7769 return ret; 7770 } 7771 7772 /** 7773 * netdev_upper_dev_link - Add a link to the upper device 7774 * @dev: device 7775 * @upper_dev: new upper device 7776 * @extack: netlink extended ack 7777 * 7778 * Adds a link to device which is upper to this one. The caller must hold 7779 * the RTNL lock. On a failure a negative errno code is returned. 7780 * On success the reference counts are adjusted and the function 7781 * returns zero. 7782 */ 7783 int netdev_upper_dev_link(struct net_device *dev, 7784 struct net_device *upper_dev, 7785 struct netlink_ext_ack *extack) 7786 { 7787 struct netdev_nested_priv priv = { 7788 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7789 .data = NULL, 7790 }; 7791 7792 return __netdev_upper_dev_link(dev, upper_dev, false, 7793 NULL, NULL, &priv, extack); 7794 } 7795 EXPORT_SYMBOL(netdev_upper_dev_link); 7796 7797 /** 7798 * netdev_master_upper_dev_link - Add a master link to the upper device 7799 * @dev: device 7800 * @upper_dev: new upper device 7801 * @upper_priv: upper device private 7802 * @upper_info: upper info to be passed down via notifier 7803 * @extack: netlink extended ack 7804 * 7805 * Adds a link to device which is upper to this one. In this case, only 7806 * one master upper device can be linked, although other non-master devices 7807 * might be linked as well. The caller must hold the RTNL lock. 7808 * On a failure a negative errno code is returned. On success the reference 7809 * counts are adjusted and the function returns zero. 7810 */ 7811 int netdev_master_upper_dev_link(struct net_device *dev, 7812 struct net_device *upper_dev, 7813 void *upper_priv, void *upper_info, 7814 struct netlink_ext_ack *extack) 7815 { 7816 struct netdev_nested_priv priv = { 7817 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7818 .data = NULL, 7819 }; 7820 7821 return __netdev_upper_dev_link(dev, upper_dev, true, 7822 upper_priv, upper_info, &priv, extack); 7823 } 7824 EXPORT_SYMBOL(netdev_master_upper_dev_link); 7825 7826 static void __netdev_upper_dev_unlink(struct net_device *dev, 7827 struct net_device *upper_dev, 7828 struct netdev_nested_priv *priv) 7829 { 7830 struct netdev_notifier_changeupper_info changeupper_info = { 7831 .info = { 7832 .dev = dev, 7833 }, 7834 .upper_dev = upper_dev, 7835 .linking = false, 7836 }; 7837 7838 ASSERT_RTNL(); 7839 7840 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; 7841 7842 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7843 &changeupper_info.info); 7844 7845 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 7846 7847 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 7848 &changeupper_info.info); 7849 7850 __netdev_update_upper_level(dev, NULL); 7851 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 7852 7853 __netdev_update_lower_level(upper_dev, priv); 7854 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 7855 priv); 7856 } 7857 7858 /** 7859 * netdev_upper_dev_unlink - Removes a link to upper device 7860 * @dev: device 7861 * @upper_dev: new upper device 7862 * 7863 * Removes a link to device which is upper to this one. The caller must hold 7864 * the RTNL lock. 7865 */ 7866 void netdev_upper_dev_unlink(struct net_device *dev, 7867 struct net_device *upper_dev) 7868 { 7869 struct netdev_nested_priv priv = { 7870 .flags = NESTED_SYNC_TODO, 7871 .data = NULL, 7872 }; 7873 7874 __netdev_upper_dev_unlink(dev, upper_dev, &priv); 7875 } 7876 EXPORT_SYMBOL(netdev_upper_dev_unlink); 7877 7878 static void __netdev_adjacent_dev_set(struct net_device *upper_dev, 7879 struct net_device *lower_dev, 7880 bool val) 7881 { 7882 struct netdev_adjacent *adj; 7883 7884 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower); 7885 if (adj) 7886 adj->ignore = val; 7887 7888 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper); 7889 if (adj) 7890 adj->ignore = val; 7891 } 7892 7893 static void netdev_adjacent_dev_disable(struct net_device *upper_dev, 7894 struct net_device *lower_dev) 7895 { 7896 __netdev_adjacent_dev_set(upper_dev, lower_dev, true); 7897 } 7898 7899 static void netdev_adjacent_dev_enable(struct net_device *upper_dev, 7900 struct net_device *lower_dev) 7901 { 7902 __netdev_adjacent_dev_set(upper_dev, lower_dev, false); 7903 } 7904 7905 int netdev_adjacent_change_prepare(struct net_device *old_dev, 7906 struct net_device *new_dev, 7907 struct net_device *dev, 7908 struct netlink_ext_ack *extack) 7909 { 7910 struct netdev_nested_priv priv = { 7911 .flags = 0, 7912 .data = NULL, 7913 }; 7914 int err; 7915 7916 if (!new_dev) 7917 return 0; 7918 7919 if (old_dev && new_dev != old_dev) 7920 netdev_adjacent_dev_disable(dev, old_dev); 7921 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv, 7922 extack); 7923 if (err) { 7924 if (old_dev && new_dev != old_dev) 7925 netdev_adjacent_dev_enable(dev, old_dev); 7926 return err; 7927 } 7928 7929 return 0; 7930 } 7931 EXPORT_SYMBOL(netdev_adjacent_change_prepare); 7932 7933 void netdev_adjacent_change_commit(struct net_device *old_dev, 7934 struct net_device *new_dev, 7935 struct net_device *dev) 7936 { 7937 struct netdev_nested_priv priv = { 7938 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7939 .data = NULL, 7940 }; 7941 7942 if (!new_dev || !old_dev) 7943 return; 7944 7945 if (new_dev == old_dev) 7946 return; 7947 7948 netdev_adjacent_dev_enable(dev, old_dev); 7949 __netdev_upper_dev_unlink(old_dev, dev, &priv); 7950 } 7951 EXPORT_SYMBOL(netdev_adjacent_change_commit); 7952 7953 void netdev_adjacent_change_abort(struct net_device *old_dev, 7954 struct net_device *new_dev, 7955 struct net_device *dev) 7956 { 7957 struct netdev_nested_priv priv = { 7958 .flags = 0, 7959 .data = NULL, 7960 }; 7961 7962 if (!new_dev) 7963 return; 7964 7965 if (old_dev && new_dev != old_dev) 7966 netdev_adjacent_dev_enable(dev, old_dev); 7967 7968 __netdev_upper_dev_unlink(new_dev, dev, &priv); 7969 } 7970 EXPORT_SYMBOL(netdev_adjacent_change_abort); 7971 7972 /** 7973 * netdev_bonding_info_change - Dispatch event about slave change 7974 * @dev: device 7975 * @bonding_info: info to dispatch 7976 * 7977 * Send NETDEV_BONDING_INFO to netdev notifiers with info. 7978 * The caller must hold the RTNL lock. 7979 */ 7980 void netdev_bonding_info_change(struct net_device *dev, 7981 struct netdev_bonding_info *bonding_info) 7982 { 7983 struct netdev_notifier_bonding_info info = { 7984 .info.dev = dev, 7985 }; 7986 7987 memcpy(&info.bonding_info, bonding_info, 7988 sizeof(struct netdev_bonding_info)); 7989 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, 7990 &info.info); 7991 } 7992 EXPORT_SYMBOL(netdev_bonding_info_change); 7993 7994 static int netdev_offload_xstats_enable_l3(struct net_device *dev, 7995 struct netlink_ext_ack *extack) 7996 { 7997 struct netdev_notifier_offload_xstats_info info = { 7998 .info.dev = dev, 7999 .info.extack = extack, 8000 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, 8001 }; 8002 int err; 8003 int rc; 8004 8005 dev->offload_xstats_l3 = kzalloc(sizeof(*dev->offload_xstats_l3), 8006 GFP_KERNEL); 8007 if (!dev->offload_xstats_l3) 8008 return -ENOMEM; 8009 8010 rc = call_netdevice_notifiers_info_robust(NETDEV_OFFLOAD_XSTATS_ENABLE, 8011 NETDEV_OFFLOAD_XSTATS_DISABLE, 8012 &info.info); 8013 err = notifier_to_errno(rc); 8014 if (err) 8015 goto free_stats; 8016 8017 return 0; 8018 8019 free_stats: 8020 kfree(dev->offload_xstats_l3); 8021 dev->offload_xstats_l3 = NULL; 8022 return err; 8023 } 8024 8025 int netdev_offload_xstats_enable(struct net_device *dev, 8026 enum netdev_offload_xstats_type type, 8027 struct netlink_ext_ack *extack) 8028 { 8029 ASSERT_RTNL(); 8030 8031 if (netdev_offload_xstats_enabled(dev, type)) 8032 return -EALREADY; 8033 8034 switch (type) { 8035 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 8036 return netdev_offload_xstats_enable_l3(dev, extack); 8037 } 8038 8039 WARN_ON(1); 8040 return -EINVAL; 8041 } 8042 EXPORT_SYMBOL(netdev_offload_xstats_enable); 8043 8044 static void netdev_offload_xstats_disable_l3(struct net_device *dev) 8045 { 8046 struct netdev_notifier_offload_xstats_info info = { 8047 .info.dev = dev, 8048 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, 8049 }; 8050 8051 call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_DISABLE, 8052 &info.info); 8053 kfree(dev->offload_xstats_l3); 8054 dev->offload_xstats_l3 = NULL; 8055 } 8056 8057 int netdev_offload_xstats_disable(struct net_device *dev, 8058 enum netdev_offload_xstats_type type) 8059 { 8060 ASSERT_RTNL(); 8061 8062 if (!netdev_offload_xstats_enabled(dev, type)) 8063 return -EALREADY; 8064 8065 switch (type) { 8066 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 8067 netdev_offload_xstats_disable_l3(dev); 8068 return 0; 8069 } 8070 8071 WARN_ON(1); 8072 return -EINVAL; 8073 } 8074 EXPORT_SYMBOL(netdev_offload_xstats_disable); 8075 8076 static void netdev_offload_xstats_disable_all(struct net_device *dev) 8077 { 8078 netdev_offload_xstats_disable(dev, NETDEV_OFFLOAD_XSTATS_TYPE_L3); 8079 } 8080 8081 static struct rtnl_hw_stats64 * 8082 netdev_offload_xstats_get_ptr(const struct net_device *dev, 8083 enum netdev_offload_xstats_type type) 8084 { 8085 switch (type) { 8086 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 8087 return dev->offload_xstats_l3; 8088 } 8089 8090 WARN_ON(1); 8091 return NULL; 8092 } 8093 8094 bool netdev_offload_xstats_enabled(const struct net_device *dev, 8095 enum netdev_offload_xstats_type type) 8096 { 8097 ASSERT_RTNL(); 8098 8099 return netdev_offload_xstats_get_ptr(dev, type); 8100 } 8101 EXPORT_SYMBOL(netdev_offload_xstats_enabled); 8102 8103 struct netdev_notifier_offload_xstats_ru { 8104 bool used; 8105 }; 8106 8107 struct netdev_notifier_offload_xstats_rd { 8108 struct rtnl_hw_stats64 stats; 8109 bool used; 8110 }; 8111 8112 static void netdev_hw_stats64_add(struct rtnl_hw_stats64 *dest, 8113 const struct rtnl_hw_stats64 *src) 8114 { 8115 dest->rx_packets += src->rx_packets; 8116 dest->tx_packets += src->tx_packets; 8117 dest->rx_bytes += src->rx_bytes; 8118 dest->tx_bytes += src->tx_bytes; 8119 dest->rx_errors += src->rx_errors; 8120 dest->tx_errors += src->tx_errors; 8121 dest->rx_dropped += src->rx_dropped; 8122 dest->tx_dropped += src->tx_dropped; 8123 dest->multicast += src->multicast; 8124 } 8125 8126 static int netdev_offload_xstats_get_used(struct net_device *dev, 8127 enum netdev_offload_xstats_type type, 8128 bool *p_used, 8129 struct netlink_ext_ack *extack) 8130 { 8131 struct netdev_notifier_offload_xstats_ru report_used = {}; 8132 struct netdev_notifier_offload_xstats_info info = { 8133 .info.dev = dev, 8134 .info.extack = extack, 8135 .type = type, 8136 .report_used = &report_used, 8137 }; 8138 int rc; 8139 8140 WARN_ON(!netdev_offload_xstats_enabled(dev, type)); 8141 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_USED, 8142 &info.info); 8143 *p_used = report_used.used; 8144 return notifier_to_errno(rc); 8145 } 8146 8147 static int netdev_offload_xstats_get_stats(struct net_device *dev, 8148 enum netdev_offload_xstats_type type, 8149 struct rtnl_hw_stats64 *p_stats, 8150 bool *p_used, 8151 struct netlink_ext_ack *extack) 8152 { 8153 struct netdev_notifier_offload_xstats_rd report_delta = {}; 8154 struct netdev_notifier_offload_xstats_info info = { 8155 .info.dev = dev, 8156 .info.extack = extack, 8157 .type = type, 8158 .report_delta = &report_delta, 8159 }; 8160 struct rtnl_hw_stats64 *stats; 8161 int rc; 8162 8163 stats = netdev_offload_xstats_get_ptr(dev, type); 8164 if (WARN_ON(!stats)) 8165 return -EINVAL; 8166 8167 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_DELTA, 8168 &info.info); 8169 8170 /* Cache whatever we got, even if there was an error, otherwise the 8171 * successful stats retrievals would get lost. 8172 */ 8173 netdev_hw_stats64_add(stats, &report_delta.stats); 8174 8175 if (p_stats) 8176 *p_stats = *stats; 8177 *p_used = report_delta.used; 8178 8179 return notifier_to_errno(rc); 8180 } 8181 8182 int netdev_offload_xstats_get(struct net_device *dev, 8183 enum netdev_offload_xstats_type type, 8184 struct rtnl_hw_stats64 *p_stats, bool *p_used, 8185 struct netlink_ext_ack *extack) 8186 { 8187 ASSERT_RTNL(); 8188 8189 if (p_stats) 8190 return netdev_offload_xstats_get_stats(dev, type, p_stats, 8191 p_used, extack); 8192 else 8193 return netdev_offload_xstats_get_used(dev, type, p_used, 8194 extack); 8195 } 8196 EXPORT_SYMBOL(netdev_offload_xstats_get); 8197 8198 void 8199 netdev_offload_xstats_report_delta(struct netdev_notifier_offload_xstats_rd *report_delta, 8200 const struct rtnl_hw_stats64 *stats) 8201 { 8202 report_delta->used = true; 8203 netdev_hw_stats64_add(&report_delta->stats, stats); 8204 } 8205 EXPORT_SYMBOL(netdev_offload_xstats_report_delta); 8206 8207 void 8208 netdev_offload_xstats_report_used(struct netdev_notifier_offload_xstats_ru *report_used) 8209 { 8210 report_used->used = true; 8211 } 8212 EXPORT_SYMBOL(netdev_offload_xstats_report_used); 8213 8214 void netdev_offload_xstats_push_delta(struct net_device *dev, 8215 enum netdev_offload_xstats_type type, 8216 const struct rtnl_hw_stats64 *p_stats) 8217 { 8218 struct rtnl_hw_stats64 *stats; 8219 8220 ASSERT_RTNL(); 8221 8222 stats = netdev_offload_xstats_get_ptr(dev, type); 8223 if (WARN_ON(!stats)) 8224 return; 8225 8226 netdev_hw_stats64_add(stats, p_stats); 8227 } 8228 EXPORT_SYMBOL(netdev_offload_xstats_push_delta); 8229 8230 /** 8231 * netdev_get_xmit_slave - Get the xmit slave of master device 8232 * @dev: device 8233 * @skb: The packet 8234 * @all_slaves: assume all the slaves are active 8235 * 8236 * The reference counters are not incremented so the caller must be 8237 * careful with locks. The caller must hold RCU lock. 8238 * %NULL is returned if no slave is found. 8239 */ 8240 8241 struct net_device *netdev_get_xmit_slave(struct net_device *dev, 8242 struct sk_buff *skb, 8243 bool all_slaves) 8244 { 8245 const struct net_device_ops *ops = dev->netdev_ops; 8246 8247 if (!ops->ndo_get_xmit_slave) 8248 return NULL; 8249 return ops->ndo_get_xmit_slave(dev, skb, all_slaves); 8250 } 8251 EXPORT_SYMBOL(netdev_get_xmit_slave); 8252 8253 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev, 8254 struct sock *sk) 8255 { 8256 const struct net_device_ops *ops = dev->netdev_ops; 8257 8258 if (!ops->ndo_sk_get_lower_dev) 8259 return NULL; 8260 return ops->ndo_sk_get_lower_dev(dev, sk); 8261 } 8262 8263 /** 8264 * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket 8265 * @dev: device 8266 * @sk: the socket 8267 * 8268 * %NULL is returned if no lower device is found. 8269 */ 8270 8271 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev, 8272 struct sock *sk) 8273 { 8274 struct net_device *lower; 8275 8276 lower = netdev_sk_get_lower_dev(dev, sk); 8277 while (lower) { 8278 dev = lower; 8279 lower = netdev_sk_get_lower_dev(dev, sk); 8280 } 8281 8282 return dev; 8283 } 8284 EXPORT_SYMBOL(netdev_sk_get_lowest_dev); 8285 8286 static void netdev_adjacent_add_links(struct net_device *dev) 8287 { 8288 struct netdev_adjacent *iter; 8289 8290 struct net *net = dev_net(dev); 8291 8292 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8293 if (!net_eq(net, dev_net(iter->dev))) 8294 continue; 8295 netdev_adjacent_sysfs_add(iter->dev, dev, 8296 &iter->dev->adj_list.lower); 8297 netdev_adjacent_sysfs_add(dev, iter->dev, 8298 &dev->adj_list.upper); 8299 } 8300 8301 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8302 if (!net_eq(net, dev_net(iter->dev))) 8303 continue; 8304 netdev_adjacent_sysfs_add(iter->dev, dev, 8305 &iter->dev->adj_list.upper); 8306 netdev_adjacent_sysfs_add(dev, iter->dev, 8307 &dev->adj_list.lower); 8308 } 8309 } 8310 8311 static void netdev_adjacent_del_links(struct net_device *dev) 8312 { 8313 struct netdev_adjacent *iter; 8314 8315 struct net *net = dev_net(dev); 8316 8317 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8318 if (!net_eq(net, dev_net(iter->dev))) 8319 continue; 8320 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8321 &iter->dev->adj_list.lower); 8322 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8323 &dev->adj_list.upper); 8324 } 8325 8326 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8327 if (!net_eq(net, dev_net(iter->dev))) 8328 continue; 8329 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8330 &iter->dev->adj_list.upper); 8331 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8332 &dev->adj_list.lower); 8333 } 8334 } 8335 8336 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) 8337 { 8338 struct netdev_adjacent *iter; 8339 8340 struct net *net = dev_net(dev); 8341 8342 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8343 if (!net_eq(net, dev_net(iter->dev))) 8344 continue; 8345 netdev_adjacent_sysfs_del(iter->dev, oldname, 8346 &iter->dev->adj_list.lower); 8347 netdev_adjacent_sysfs_add(iter->dev, dev, 8348 &iter->dev->adj_list.lower); 8349 } 8350 8351 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8352 if (!net_eq(net, dev_net(iter->dev))) 8353 continue; 8354 netdev_adjacent_sysfs_del(iter->dev, oldname, 8355 &iter->dev->adj_list.upper); 8356 netdev_adjacent_sysfs_add(iter->dev, dev, 8357 &iter->dev->adj_list.upper); 8358 } 8359 } 8360 8361 void *netdev_lower_dev_get_private(struct net_device *dev, 8362 struct net_device *lower_dev) 8363 { 8364 struct netdev_adjacent *lower; 8365 8366 if (!lower_dev) 8367 return NULL; 8368 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); 8369 if (!lower) 8370 return NULL; 8371 8372 return lower->private; 8373 } 8374 EXPORT_SYMBOL(netdev_lower_dev_get_private); 8375 8376 8377 /** 8378 * netdev_lower_state_changed - Dispatch event about lower device state change 8379 * @lower_dev: device 8380 * @lower_state_info: state to dispatch 8381 * 8382 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. 8383 * The caller must hold the RTNL lock. 8384 */ 8385 void netdev_lower_state_changed(struct net_device *lower_dev, 8386 void *lower_state_info) 8387 { 8388 struct netdev_notifier_changelowerstate_info changelowerstate_info = { 8389 .info.dev = lower_dev, 8390 }; 8391 8392 ASSERT_RTNL(); 8393 changelowerstate_info.lower_state_info = lower_state_info; 8394 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, 8395 &changelowerstate_info.info); 8396 } 8397 EXPORT_SYMBOL(netdev_lower_state_changed); 8398 8399 static void dev_change_rx_flags(struct net_device *dev, int flags) 8400 { 8401 const struct net_device_ops *ops = dev->netdev_ops; 8402 8403 if (ops->ndo_change_rx_flags) 8404 ops->ndo_change_rx_flags(dev, flags); 8405 } 8406 8407 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) 8408 { 8409 unsigned int old_flags = dev->flags; 8410 kuid_t uid; 8411 kgid_t gid; 8412 8413 ASSERT_RTNL(); 8414 8415 dev->flags |= IFF_PROMISC; 8416 dev->promiscuity += inc; 8417 if (dev->promiscuity == 0) { 8418 /* 8419 * Avoid overflow. 8420 * If inc causes overflow, untouch promisc and return error. 8421 */ 8422 if (inc < 0) 8423 dev->flags &= ~IFF_PROMISC; 8424 else { 8425 dev->promiscuity -= inc; 8426 netdev_warn(dev, "promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n"); 8427 return -EOVERFLOW; 8428 } 8429 } 8430 if (dev->flags != old_flags) { 8431 netdev_info(dev, "%s promiscuous mode\n", 8432 dev->flags & IFF_PROMISC ? "entered" : "left"); 8433 if (audit_enabled) { 8434 current_uid_gid(&uid, &gid); 8435 audit_log(audit_context(), GFP_ATOMIC, 8436 AUDIT_ANOM_PROMISCUOUS, 8437 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", 8438 dev->name, (dev->flags & IFF_PROMISC), 8439 (old_flags & IFF_PROMISC), 8440 from_kuid(&init_user_ns, audit_get_loginuid(current)), 8441 from_kuid(&init_user_ns, uid), 8442 from_kgid(&init_user_ns, gid), 8443 audit_get_sessionid(current)); 8444 } 8445 8446 dev_change_rx_flags(dev, IFF_PROMISC); 8447 } 8448 if (notify) 8449 __dev_notify_flags(dev, old_flags, IFF_PROMISC, 0, NULL); 8450 return 0; 8451 } 8452 8453 /** 8454 * dev_set_promiscuity - update promiscuity count on a device 8455 * @dev: device 8456 * @inc: modifier 8457 * 8458 * Add or remove promiscuity from a device. While the count in the device 8459 * remains above zero the interface remains promiscuous. Once it hits zero 8460 * the device reverts back to normal filtering operation. A negative inc 8461 * value is used to drop promiscuity on the device. 8462 * Return 0 if successful or a negative errno code on error. 8463 */ 8464 int dev_set_promiscuity(struct net_device *dev, int inc) 8465 { 8466 unsigned int old_flags = dev->flags; 8467 int err; 8468 8469 err = __dev_set_promiscuity(dev, inc, true); 8470 if (err < 0) 8471 return err; 8472 if (dev->flags != old_flags) 8473 dev_set_rx_mode(dev); 8474 return err; 8475 } 8476 EXPORT_SYMBOL(dev_set_promiscuity); 8477 8478 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) 8479 { 8480 unsigned int old_flags = dev->flags, old_gflags = dev->gflags; 8481 8482 ASSERT_RTNL(); 8483 8484 dev->flags |= IFF_ALLMULTI; 8485 dev->allmulti += inc; 8486 if (dev->allmulti == 0) { 8487 /* 8488 * Avoid overflow. 8489 * If inc causes overflow, untouch allmulti and return error. 8490 */ 8491 if (inc < 0) 8492 dev->flags &= ~IFF_ALLMULTI; 8493 else { 8494 dev->allmulti -= inc; 8495 netdev_warn(dev, "allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n"); 8496 return -EOVERFLOW; 8497 } 8498 } 8499 if (dev->flags ^ old_flags) { 8500 netdev_info(dev, "%s allmulticast mode\n", 8501 dev->flags & IFF_ALLMULTI ? "entered" : "left"); 8502 dev_change_rx_flags(dev, IFF_ALLMULTI); 8503 dev_set_rx_mode(dev); 8504 if (notify) 8505 __dev_notify_flags(dev, old_flags, 8506 dev->gflags ^ old_gflags, 0, NULL); 8507 } 8508 return 0; 8509 } 8510 8511 /** 8512 * dev_set_allmulti - update allmulti count on a device 8513 * @dev: device 8514 * @inc: modifier 8515 * 8516 * Add or remove reception of all multicast frames to a device. While the 8517 * count in the device remains above zero the interface remains listening 8518 * to all interfaces. Once it hits zero the device reverts back to normal 8519 * filtering operation. A negative @inc value is used to drop the counter 8520 * when releasing a resource needing all multicasts. 8521 * Return 0 if successful or a negative errno code on error. 8522 */ 8523 8524 int dev_set_allmulti(struct net_device *dev, int inc) 8525 { 8526 return __dev_set_allmulti(dev, inc, true); 8527 } 8528 EXPORT_SYMBOL(dev_set_allmulti); 8529 8530 /* 8531 * Upload unicast and multicast address lists to device and 8532 * configure RX filtering. When the device doesn't support unicast 8533 * filtering it is put in promiscuous mode while unicast addresses 8534 * are present. 8535 */ 8536 void __dev_set_rx_mode(struct net_device *dev) 8537 { 8538 const struct net_device_ops *ops = dev->netdev_ops; 8539 8540 /* dev_open will call this function so the list will stay sane. */ 8541 if (!(dev->flags&IFF_UP)) 8542 return; 8543 8544 if (!netif_device_present(dev)) 8545 return; 8546 8547 if (!(dev->priv_flags & IFF_UNICAST_FLT)) { 8548 /* Unicast addresses changes may only happen under the rtnl, 8549 * therefore calling __dev_set_promiscuity here is safe. 8550 */ 8551 if (!netdev_uc_empty(dev) && !dev->uc_promisc) { 8552 __dev_set_promiscuity(dev, 1, false); 8553 dev->uc_promisc = true; 8554 } else if (netdev_uc_empty(dev) && dev->uc_promisc) { 8555 __dev_set_promiscuity(dev, -1, false); 8556 dev->uc_promisc = false; 8557 } 8558 } 8559 8560 if (ops->ndo_set_rx_mode) 8561 ops->ndo_set_rx_mode(dev); 8562 } 8563 8564 void dev_set_rx_mode(struct net_device *dev) 8565 { 8566 netif_addr_lock_bh(dev); 8567 __dev_set_rx_mode(dev); 8568 netif_addr_unlock_bh(dev); 8569 } 8570 8571 /** 8572 * dev_get_flags - get flags reported to userspace 8573 * @dev: device 8574 * 8575 * Get the combination of flag bits exported through APIs to userspace. 8576 */ 8577 unsigned int dev_get_flags(const struct net_device *dev) 8578 { 8579 unsigned int flags; 8580 8581 flags = (dev->flags & ~(IFF_PROMISC | 8582 IFF_ALLMULTI | 8583 IFF_RUNNING | 8584 IFF_LOWER_UP | 8585 IFF_DORMANT)) | 8586 (dev->gflags & (IFF_PROMISC | 8587 IFF_ALLMULTI)); 8588 8589 if (netif_running(dev)) { 8590 if (netif_oper_up(dev)) 8591 flags |= IFF_RUNNING; 8592 if (netif_carrier_ok(dev)) 8593 flags |= IFF_LOWER_UP; 8594 if (netif_dormant(dev)) 8595 flags |= IFF_DORMANT; 8596 } 8597 8598 return flags; 8599 } 8600 EXPORT_SYMBOL(dev_get_flags); 8601 8602 int __dev_change_flags(struct net_device *dev, unsigned int flags, 8603 struct netlink_ext_ack *extack) 8604 { 8605 unsigned int old_flags = dev->flags; 8606 int ret; 8607 8608 ASSERT_RTNL(); 8609 8610 /* 8611 * Set the flags on our device. 8612 */ 8613 8614 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | 8615 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | 8616 IFF_AUTOMEDIA)) | 8617 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | 8618 IFF_ALLMULTI)); 8619 8620 /* 8621 * Load in the correct multicast list now the flags have changed. 8622 */ 8623 8624 if ((old_flags ^ flags) & IFF_MULTICAST) 8625 dev_change_rx_flags(dev, IFF_MULTICAST); 8626 8627 dev_set_rx_mode(dev); 8628 8629 /* 8630 * Have we downed the interface. We handle IFF_UP ourselves 8631 * according to user attempts to set it, rather than blindly 8632 * setting it. 8633 */ 8634 8635 ret = 0; 8636 if ((old_flags ^ flags) & IFF_UP) { 8637 if (old_flags & IFF_UP) 8638 __dev_close(dev); 8639 else 8640 ret = __dev_open(dev, extack); 8641 } 8642 8643 if ((flags ^ dev->gflags) & IFF_PROMISC) { 8644 int inc = (flags & IFF_PROMISC) ? 1 : -1; 8645 unsigned int old_flags = dev->flags; 8646 8647 dev->gflags ^= IFF_PROMISC; 8648 8649 if (__dev_set_promiscuity(dev, inc, false) >= 0) 8650 if (dev->flags != old_flags) 8651 dev_set_rx_mode(dev); 8652 } 8653 8654 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI 8655 * is important. Some (broken) drivers set IFF_PROMISC, when 8656 * IFF_ALLMULTI is requested not asking us and not reporting. 8657 */ 8658 if ((flags ^ dev->gflags) & IFF_ALLMULTI) { 8659 int inc = (flags & IFF_ALLMULTI) ? 1 : -1; 8660 8661 dev->gflags ^= IFF_ALLMULTI; 8662 __dev_set_allmulti(dev, inc, false); 8663 } 8664 8665 return ret; 8666 } 8667 8668 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, 8669 unsigned int gchanges, u32 portid, 8670 const struct nlmsghdr *nlh) 8671 { 8672 unsigned int changes = dev->flags ^ old_flags; 8673 8674 if (gchanges) 8675 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC, portid, nlh); 8676 8677 if (changes & IFF_UP) { 8678 if (dev->flags & IFF_UP) 8679 call_netdevice_notifiers(NETDEV_UP, dev); 8680 else 8681 call_netdevice_notifiers(NETDEV_DOWN, dev); 8682 } 8683 8684 if (dev->flags & IFF_UP && 8685 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { 8686 struct netdev_notifier_change_info change_info = { 8687 .info = { 8688 .dev = dev, 8689 }, 8690 .flags_changed = changes, 8691 }; 8692 8693 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info); 8694 } 8695 } 8696 8697 /** 8698 * dev_change_flags - change device settings 8699 * @dev: device 8700 * @flags: device state flags 8701 * @extack: netlink extended ack 8702 * 8703 * Change settings on device based state flags. The flags are 8704 * in the userspace exported format. 8705 */ 8706 int dev_change_flags(struct net_device *dev, unsigned int flags, 8707 struct netlink_ext_ack *extack) 8708 { 8709 int ret; 8710 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; 8711 8712 ret = __dev_change_flags(dev, flags, extack); 8713 if (ret < 0) 8714 return ret; 8715 8716 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); 8717 __dev_notify_flags(dev, old_flags, changes, 0, NULL); 8718 return ret; 8719 } 8720 EXPORT_SYMBOL(dev_change_flags); 8721 8722 int __dev_set_mtu(struct net_device *dev, int new_mtu) 8723 { 8724 const struct net_device_ops *ops = dev->netdev_ops; 8725 8726 if (ops->ndo_change_mtu) 8727 return ops->ndo_change_mtu(dev, new_mtu); 8728 8729 /* Pairs with all the lockless reads of dev->mtu in the stack */ 8730 WRITE_ONCE(dev->mtu, new_mtu); 8731 return 0; 8732 } 8733 EXPORT_SYMBOL(__dev_set_mtu); 8734 8735 int dev_validate_mtu(struct net_device *dev, int new_mtu, 8736 struct netlink_ext_ack *extack) 8737 { 8738 /* MTU must be positive, and in range */ 8739 if (new_mtu < 0 || new_mtu < dev->min_mtu) { 8740 NL_SET_ERR_MSG(extack, "mtu less than device minimum"); 8741 return -EINVAL; 8742 } 8743 8744 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { 8745 NL_SET_ERR_MSG(extack, "mtu greater than device maximum"); 8746 return -EINVAL; 8747 } 8748 return 0; 8749 } 8750 8751 /** 8752 * dev_set_mtu_ext - Change maximum transfer unit 8753 * @dev: device 8754 * @new_mtu: new transfer unit 8755 * @extack: netlink extended ack 8756 * 8757 * Change the maximum transfer size of the network device. 8758 */ 8759 int dev_set_mtu_ext(struct net_device *dev, int new_mtu, 8760 struct netlink_ext_ack *extack) 8761 { 8762 int err, orig_mtu; 8763 8764 if (new_mtu == dev->mtu) 8765 return 0; 8766 8767 err = dev_validate_mtu(dev, new_mtu, extack); 8768 if (err) 8769 return err; 8770 8771 if (!netif_device_present(dev)) 8772 return -ENODEV; 8773 8774 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); 8775 err = notifier_to_errno(err); 8776 if (err) 8777 return err; 8778 8779 orig_mtu = dev->mtu; 8780 err = __dev_set_mtu(dev, new_mtu); 8781 8782 if (!err) { 8783 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8784 orig_mtu); 8785 err = notifier_to_errno(err); 8786 if (err) { 8787 /* setting mtu back and notifying everyone again, 8788 * so that they have a chance to revert changes. 8789 */ 8790 __dev_set_mtu(dev, orig_mtu); 8791 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8792 new_mtu); 8793 } 8794 } 8795 return err; 8796 } 8797 8798 int dev_set_mtu(struct net_device *dev, int new_mtu) 8799 { 8800 struct netlink_ext_ack extack; 8801 int err; 8802 8803 memset(&extack, 0, sizeof(extack)); 8804 err = dev_set_mtu_ext(dev, new_mtu, &extack); 8805 if (err && extack._msg) 8806 net_err_ratelimited("%s: %s\n", dev->name, extack._msg); 8807 return err; 8808 } 8809 EXPORT_SYMBOL(dev_set_mtu); 8810 8811 /** 8812 * dev_change_tx_queue_len - Change TX queue length of a netdevice 8813 * @dev: device 8814 * @new_len: new tx queue length 8815 */ 8816 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len) 8817 { 8818 unsigned int orig_len = dev->tx_queue_len; 8819 int res; 8820 8821 if (new_len != (unsigned int)new_len) 8822 return -ERANGE; 8823 8824 if (new_len != orig_len) { 8825 dev->tx_queue_len = new_len; 8826 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev); 8827 res = notifier_to_errno(res); 8828 if (res) 8829 goto err_rollback; 8830 res = dev_qdisc_change_tx_queue_len(dev); 8831 if (res) 8832 goto err_rollback; 8833 } 8834 8835 return 0; 8836 8837 err_rollback: 8838 netdev_err(dev, "refused to change device tx_queue_len\n"); 8839 dev->tx_queue_len = orig_len; 8840 return res; 8841 } 8842 8843 /** 8844 * dev_set_group - Change group this device belongs to 8845 * @dev: device 8846 * @new_group: group this device should belong to 8847 */ 8848 void dev_set_group(struct net_device *dev, int new_group) 8849 { 8850 dev->group = new_group; 8851 } 8852 8853 /** 8854 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR. 8855 * @dev: device 8856 * @addr: new address 8857 * @extack: netlink extended ack 8858 */ 8859 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr, 8860 struct netlink_ext_ack *extack) 8861 { 8862 struct netdev_notifier_pre_changeaddr_info info = { 8863 .info.dev = dev, 8864 .info.extack = extack, 8865 .dev_addr = addr, 8866 }; 8867 int rc; 8868 8869 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info); 8870 return notifier_to_errno(rc); 8871 } 8872 EXPORT_SYMBOL(dev_pre_changeaddr_notify); 8873 8874 /** 8875 * dev_set_mac_address - Change Media Access Control Address 8876 * @dev: device 8877 * @sa: new address 8878 * @extack: netlink extended ack 8879 * 8880 * Change the hardware (MAC) address of the device 8881 */ 8882 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa, 8883 struct netlink_ext_ack *extack) 8884 { 8885 const struct net_device_ops *ops = dev->netdev_ops; 8886 int err; 8887 8888 if (!ops->ndo_set_mac_address) 8889 return -EOPNOTSUPP; 8890 if (sa->sa_family != dev->type) 8891 return -EINVAL; 8892 if (!netif_device_present(dev)) 8893 return -ENODEV; 8894 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack); 8895 if (err) 8896 return err; 8897 if (memcmp(dev->dev_addr, sa->sa_data, dev->addr_len)) { 8898 err = ops->ndo_set_mac_address(dev, sa); 8899 if (err) 8900 return err; 8901 } 8902 dev->addr_assign_type = NET_ADDR_SET; 8903 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); 8904 add_device_randomness(dev->dev_addr, dev->addr_len); 8905 return 0; 8906 } 8907 EXPORT_SYMBOL(dev_set_mac_address); 8908 8909 static DECLARE_RWSEM(dev_addr_sem); 8910 8911 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa, 8912 struct netlink_ext_ack *extack) 8913 { 8914 int ret; 8915 8916 down_write(&dev_addr_sem); 8917 ret = dev_set_mac_address(dev, sa, extack); 8918 up_write(&dev_addr_sem); 8919 return ret; 8920 } 8921 EXPORT_SYMBOL(dev_set_mac_address_user); 8922 8923 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name) 8924 { 8925 size_t size = sizeof(sa->sa_data_min); 8926 struct net_device *dev; 8927 int ret = 0; 8928 8929 down_read(&dev_addr_sem); 8930 rcu_read_lock(); 8931 8932 dev = dev_get_by_name_rcu(net, dev_name); 8933 if (!dev) { 8934 ret = -ENODEV; 8935 goto unlock; 8936 } 8937 if (!dev->addr_len) 8938 memset(sa->sa_data, 0, size); 8939 else 8940 memcpy(sa->sa_data, dev->dev_addr, 8941 min_t(size_t, size, dev->addr_len)); 8942 sa->sa_family = dev->type; 8943 8944 unlock: 8945 rcu_read_unlock(); 8946 up_read(&dev_addr_sem); 8947 return ret; 8948 } 8949 EXPORT_SYMBOL(dev_get_mac_address); 8950 8951 /** 8952 * dev_change_carrier - Change device carrier 8953 * @dev: device 8954 * @new_carrier: new value 8955 * 8956 * Change device carrier 8957 */ 8958 int dev_change_carrier(struct net_device *dev, bool new_carrier) 8959 { 8960 const struct net_device_ops *ops = dev->netdev_ops; 8961 8962 if (!ops->ndo_change_carrier) 8963 return -EOPNOTSUPP; 8964 if (!netif_device_present(dev)) 8965 return -ENODEV; 8966 return ops->ndo_change_carrier(dev, new_carrier); 8967 } 8968 8969 /** 8970 * dev_get_phys_port_id - Get device physical port ID 8971 * @dev: device 8972 * @ppid: port ID 8973 * 8974 * Get device physical port ID 8975 */ 8976 int dev_get_phys_port_id(struct net_device *dev, 8977 struct netdev_phys_item_id *ppid) 8978 { 8979 const struct net_device_ops *ops = dev->netdev_ops; 8980 8981 if (!ops->ndo_get_phys_port_id) 8982 return -EOPNOTSUPP; 8983 return ops->ndo_get_phys_port_id(dev, ppid); 8984 } 8985 8986 /** 8987 * dev_get_phys_port_name - Get device physical port name 8988 * @dev: device 8989 * @name: port name 8990 * @len: limit of bytes to copy to name 8991 * 8992 * Get device physical port name 8993 */ 8994 int dev_get_phys_port_name(struct net_device *dev, 8995 char *name, size_t len) 8996 { 8997 const struct net_device_ops *ops = dev->netdev_ops; 8998 int err; 8999 9000 if (ops->ndo_get_phys_port_name) { 9001 err = ops->ndo_get_phys_port_name(dev, name, len); 9002 if (err != -EOPNOTSUPP) 9003 return err; 9004 } 9005 return devlink_compat_phys_port_name_get(dev, name, len); 9006 } 9007 9008 /** 9009 * dev_get_port_parent_id - Get the device's port parent identifier 9010 * @dev: network device 9011 * @ppid: pointer to a storage for the port's parent identifier 9012 * @recurse: allow/disallow recursion to lower devices 9013 * 9014 * Get the devices's port parent identifier 9015 */ 9016 int dev_get_port_parent_id(struct net_device *dev, 9017 struct netdev_phys_item_id *ppid, 9018 bool recurse) 9019 { 9020 const struct net_device_ops *ops = dev->netdev_ops; 9021 struct netdev_phys_item_id first = { }; 9022 struct net_device *lower_dev; 9023 struct list_head *iter; 9024 int err; 9025 9026 if (ops->ndo_get_port_parent_id) { 9027 err = ops->ndo_get_port_parent_id(dev, ppid); 9028 if (err != -EOPNOTSUPP) 9029 return err; 9030 } 9031 9032 err = devlink_compat_switch_id_get(dev, ppid); 9033 if (!recurse || err != -EOPNOTSUPP) 9034 return err; 9035 9036 netdev_for_each_lower_dev(dev, lower_dev, iter) { 9037 err = dev_get_port_parent_id(lower_dev, ppid, true); 9038 if (err) 9039 break; 9040 if (!first.id_len) 9041 first = *ppid; 9042 else if (memcmp(&first, ppid, sizeof(*ppid))) 9043 return -EOPNOTSUPP; 9044 } 9045 9046 return err; 9047 } 9048 EXPORT_SYMBOL(dev_get_port_parent_id); 9049 9050 /** 9051 * netdev_port_same_parent_id - Indicate if two network devices have 9052 * the same port parent identifier 9053 * @a: first network device 9054 * @b: second network device 9055 */ 9056 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b) 9057 { 9058 struct netdev_phys_item_id a_id = { }; 9059 struct netdev_phys_item_id b_id = { }; 9060 9061 if (dev_get_port_parent_id(a, &a_id, true) || 9062 dev_get_port_parent_id(b, &b_id, true)) 9063 return false; 9064 9065 return netdev_phys_item_id_same(&a_id, &b_id); 9066 } 9067 EXPORT_SYMBOL(netdev_port_same_parent_id); 9068 9069 /** 9070 * dev_change_proto_down - set carrier according to proto_down. 9071 * 9072 * @dev: device 9073 * @proto_down: new value 9074 */ 9075 int dev_change_proto_down(struct net_device *dev, bool proto_down) 9076 { 9077 if (!(dev->priv_flags & IFF_CHANGE_PROTO_DOWN)) 9078 return -EOPNOTSUPP; 9079 if (!netif_device_present(dev)) 9080 return -ENODEV; 9081 if (proto_down) 9082 netif_carrier_off(dev); 9083 else 9084 netif_carrier_on(dev); 9085 dev->proto_down = proto_down; 9086 return 0; 9087 } 9088 9089 /** 9090 * dev_change_proto_down_reason - proto down reason 9091 * 9092 * @dev: device 9093 * @mask: proto down mask 9094 * @value: proto down value 9095 */ 9096 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask, 9097 u32 value) 9098 { 9099 int b; 9100 9101 if (!mask) { 9102 dev->proto_down_reason = value; 9103 } else { 9104 for_each_set_bit(b, &mask, 32) { 9105 if (value & (1 << b)) 9106 dev->proto_down_reason |= BIT(b); 9107 else 9108 dev->proto_down_reason &= ~BIT(b); 9109 } 9110 } 9111 } 9112 9113 struct bpf_xdp_link { 9114 struct bpf_link link; 9115 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */ 9116 int flags; 9117 }; 9118 9119 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags) 9120 { 9121 if (flags & XDP_FLAGS_HW_MODE) 9122 return XDP_MODE_HW; 9123 if (flags & XDP_FLAGS_DRV_MODE) 9124 return XDP_MODE_DRV; 9125 if (flags & XDP_FLAGS_SKB_MODE) 9126 return XDP_MODE_SKB; 9127 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB; 9128 } 9129 9130 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode) 9131 { 9132 switch (mode) { 9133 case XDP_MODE_SKB: 9134 return generic_xdp_install; 9135 case XDP_MODE_DRV: 9136 case XDP_MODE_HW: 9137 return dev->netdev_ops->ndo_bpf; 9138 default: 9139 return NULL; 9140 } 9141 } 9142 9143 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev, 9144 enum bpf_xdp_mode mode) 9145 { 9146 return dev->xdp_state[mode].link; 9147 } 9148 9149 static struct bpf_prog *dev_xdp_prog(struct net_device *dev, 9150 enum bpf_xdp_mode mode) 9151 { 9152 struct bpf_xdp_link *link = dev_xdp_link(dev, mode); 9153 9154 if (link) 9155 return link->link.prog; 9156 return dev->xdp_state[mode].prog; 9157 } 9158 9159 u8 dev_xdp_prog_count(struct net_device *dev) 9160 { 9161 u8 count = 0; 9162 int i; 9163 9164 for (i = 0; i < __MAX_XDP_MODE; i++) 9165 if (dev->xdp_state[i].prog || dev->xdp_state[i].link) 9166 count++; 9167 return count; 9168 } 9169 EXPORT_SYMBOL_GPL(dev_xdp_prog_count); 9170 9171 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode) 9172 { 9173 struct bpf_prog *prog = dev_xdp_prog(dev, mode); 9174 9175 return prog ? prog->aux->id : 0; 9176 } 9177 9178 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode, 9179 struct bpf_xdp_link *link) 9180 { 9181 dev->xdp_state[mode].link = link; 9182 dev->xdp_state[mode].prog = NULL; 9183 } 9184 9185 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode, 9186 struct bpf_prog *prog) 9187 { 9188 dev->xdp_state[mode].link = NULL; 9189 dev->xdp_state[mode].prog = prog; 9190 } 9191 9192 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode, 9193 bpf_op_t bpf_op, struct netlink_ext_ack *extack, 9194 u32 flags, struct bpf_prog *prog) 9195 { 9196 struct netdev_bpf xdp; 9197 int err; 9198 9199 memset(&xdp, 0, sizeof(xdp)); 9200 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG; 9201 xdp.extack = extack; 9202 xdp.flags = flags; 9203 xdp.prog = prog; 9204 9205 /* Drivers assume refcnt is already incremented (i.e, prog pointer is 9206 * "moved" into driver), so they don't increment it on their own, but 9207 * they do decrement refcnt when program is detached or replaced. 9208 * Given net_device also owns link/prog, we need to bump refcnt here 9209 * to prevent drivers from underflowing it. 9210 */ 9211 if (prog) 9212 bpf_prog_inc(prog); 9213 err = bpf_op(dev, &xdp); 9214 if (err) { 9215 if (prog) 9216 bpf_prog_put(prog); 9217 return err; 9218 } 9219 9220 if (mode != XDP_MODE_HW) 9221 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog); 9222 9223 return 0; 9224 } 9225 9226 static void dev_xdp_uninstall(struct net_device *dev) 9227 { 9228 struct bpf_xdp_link *link; 9229 struct bpf_prog *prog; 9230 enum bpf_xdp_mode mode; 9231 bpf_op_t bpf_op; 9232 9233 ASSERT_RTNL(); 9234 9235 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) { 9236 prog = dev_xdp_prog(dev, mode); 9237 if (!prog) 9238 continue; 9239 9240 bpf_op = dev_xdp_bpf_op(dev, mode); 9241 if (!bpf_op) 9242 continue; 9243 9244 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9245 9246 /* auto-detach link from net device */ 9247 link = dev_xdp_link(dev, mode); 9248 if (link) 9249 link->dev = NULL; 9250 else 9251 bpf_prog_put(prog); 9252 9253 dev_xdp_set_link(dev, mode, NULL); 9254 } 9255 } 9256 9257 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack, 9258 struct bpf_xdp_link *link, struct bpf_prog *new_prog, 9259 struct bpf_prog *old_prog, u32 flags) 9260 { 9261 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES); 9262 struct bpf_prog *cur_prog; 9263 struct net_device *upper; 9264 struct list_head *iter; 9265 enum bpf_xdp_mode mode; 9266 bpf_op_t bpf_op; 9267 int err; 9268 9269 ASSERT_RTNL(); 9270 9271 /* either link or prog attachment, never both */ 9272 if (link && (new_prog || old_prog)) 9273 return -EINVAL; 9274 /* link supports only XDP mode flags */ 9275 if (link && (flags & ~XDP_FLAGS_MODES)) { 9276 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment"); 9277 return -EINVAL; 9278 } 9279 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */ 9280 if (num_modes > 1) { 9281 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set"); 9282 return -EINVAL; 9283 } 9284 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */ 9285 if (!num_modes && dev_xdp_prog_count(dev) > 1) { 9286 NL_SET_ERR_MSG(extack, 9287 "More than one program loaded, unset mode is ambiguous"); 9288 return -EINVAL; 9289 } 9290 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */ 9291 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) { 9292 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified"); 9293 return -EINVAL; 9294 } 9295 9296 mode = dev_xdp_mode(dev, flags); 9297 /* can't replace attached link */ 9298 if (dev_xdp_link(dev, mode)) { 9299 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link"); 9300 return -EBUSY; 9301 } 9302 9303 /* don't allow if an upper device already has a program */ 9304 netdev_for_each_upper_dev_rcu(dev, upper, iter) { 9305 if (dev_xdp_prog_count(upper) > 0) { 9306 NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program"); 9307 return -EEXIST; 9308 } 9309 } 9310 9311 cur_prog = dev_xdp_prog(dev, mode); 9312 /* can't replace attached prog with link */ 9313 if (link && cur_prog) { 9314 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link"); 9315 return -EBUSY; 9316 } 9317 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) { 9318 NL_SET_ERR_MSG(extack, "Active program does not match expected"); 9319 return -EEXIST; 9320 } 9321 9322 /* put effective new program into new_prog */ 9323 if (link) 9324 new_prog = link->link.prog; 9325 9326 if (new_prog) { 9327 bool offload = mode == XDP_MODE_HW; 9328 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB 9329 ? XDP_MODE_DRV : XDP_MODE_SKB; 9330 9331 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) { 9332 NL_SET_ERR_MSG(extack, "XDP program already attached"); 9333 return -EBUSY; 9334 } 9335 if (!offload && dev_xdp_prog(dev, other_mode)) { 9336 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time"); 9337 return -EEXIST; 9338 } 9339 if (!offload && bpf_prog_is_offloaded(new_prog->aux)) { 9340 NL_SET_ERR_MSG(extack, "Using offloaded program without HW_MODE flag is not supported"); 9341 return -EINVAL; 9342 } 9343 if (bpf_prog_is_dev_bound(new_prog->aux) && !bpf_offload_dev_match(new_prog, dev)) { 9344 NL_SET_ERR_MSG(extack, "Program bound to different device"); 9345 return -EINVAL; 9346 } 9347 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) { 9348 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device"); 9349 return -EINVAL; 9350 } 9351 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) { 9352 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device"); 9353 return -EINVAL; 9354 } 9355 } 9356 9357 /* don't call drivers if the effective program didn't change */ 9358 if (new_prog != cur_prog) { 9359 bpf_op = dev_xdp_bpf_op(dev, mode); 9360 if (!bpf_op) { 9361 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode"); 9362 return -EOPNOTSUPP; 9363 } 9364 9365 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog); 9366 if (err) 9367 return err; 9368 } 9369 9370 if (link) 9371 dev_xdp_set_link(dev, mode, link); 9372 else 9373 dev_xdp_set_prog(dev, mode, new_prog); 9374 if (cur_prog) 9375 bpf_prog_put(cur_prog); 9376 9377 return 0; 9378 } 9379 9380 static int dev_xdp_attach_link(struct net_device *dev, 9381 struct netlink_ext_ack *extack, 9382 struct bpf_xdp_link *link) 9383 { 9384 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags); 9385 } 9386 9387 static int dev_xdp_detach_link(struct net_device *dev, 9388 struct netlink_ext_ack *extack, 9389 struct bpf_xdp_link *link) 9390 { 9391 enum bpf_xdp_mode mode; 9392 bpf_op_t bpf_op; 9393 9394 ASSERT_RTNL(); 9395 9396 mode = dev_xdp_mode(dev, link->flags); 9397 if (dev_xdp_link(dev, mode) != link) 9398 return -EINVAL; 9399 9400 bpf_op = dev_xdp_bpf_op(dev, mode); 9401 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9402 dev_xdp_set_link(dev, mode, NULL); 9403 return 0; 9404 } 9405 9406 static void bpf_xdp_link_release(struct bpf_link *link) 9407 { 9408 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9409 9410 rtnl_lock(); 9411 9412 /* if racing with net_device's tear down, xdp_link->dev might be 9413 * already NULL, in which case link was already auto-detached 9414 */ 9415 if (xdp_link->dev) { 9416 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link)); 9417 xdp_link->dev = NULL; 9418 } 9419 9420 rtnl_unlock(); 9421 } 9422 9423 static int bpf_xdp_link_detach(struct bpf_link *link) 9424 { 9425 bpf_xdp_link_release(link); 9426 return 0; 9427 } 9428 9429 static void bpf_xdp_link_dealloc(struct bpf_link *link) 9430 { 9431 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9432 9433 kfree(xdp_link); 9434 } 9435 9436 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link, 9437 struct seq_file *seq) 9438 { 9439 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9440 u32 ifindex = 0; 9441 9442 rtnl_lock(); 9443 if (xdp_link->dev) 9444 ifindex = xdp_link->dev->ifindex; 9445 rtnl_unlock(); 9446 9447 seq_printf(seq, "ifindex:\t%u\n", ifindex); 9448 } 9449 9450 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link, 9451 struct bpf_link_info *info) 9452 { 9453 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9454 u32 ifindex = 0; 9455 9456 rtnl_lock(); 9457 if (xdp_link->dev) 9458 ifindex = xdp_link->dev->ifindex; 9459 rtnl_unlock(); 9460 9461 info->xdp.ifindex = ifindex; 9462 return 0; 9463 } 9464 9465 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog, 9466 struct bpf_prog *old_prog) 9467 { 9468 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9469 enum bpf_xdp_mode mode; 9470 bpf_op_t bpf_op; 9471 int err = 0; 9472 9473 rtnl_lock(); 9474 9475 /* link might have been auto-released already, so fail */ 9476 if (!xdp_link->dev) { 9477 err = -ENOLINK; 9478 goto out_unlock; 9479 } 9480 9481 if (old_prog && link->prog != old_prog) { 9482 err = -EPERM; 9483 goto out_unlock; 9484 } 9485 old_prog = link->prog; 9486 if (old_prog->type != new_prog->type || 9487 old_prog->expected_attach_type != new_prog->expected_attach_type) { 9488 err = -EINVAL; 9489 goto out_unlock; 9490 } 9491 9492 if (old_prog == new_prog) { 9493 /* no-op, don't disturb drivers */ 9494 bpf_prog_put(new_prog); 9495 goto out_unlock; 9496 } 9497 9498 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags); 9499 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode); 9500 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL, 9501 xdp_link->flags, new_prog); 9502 if (err) 9503 goto out_unlock; 9504 9505 old_prog = xchg(&link->prog, new_prog); 9506 bpf_prog_put(old_prog); 9507 9508 out_unlock: 9509 rtnl_unlock(); 9510 return err; 9511 } 9512 9513 static const struct bpf_link_ops bpf_xdp_link_lops = { 9514 .release = bpf_xdp_link_release, 9515 .dealloc = bpf_xdp_link_dealloc, 9516 .detach = bpf_xdp_link_detach, 9517 .show_fdinfo = bpf_xdp_link_show_fdinfo, 9518 .fill_link_info = bpf_xdp_link_fill_link_info, 9519 .update_prog = bpf_xdp_link_update, 9520 }; 9521 9522 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog) 9523 { 9524 struct net *net = current->nsproxy->net_ns; 9525 struct bpf_link_primer link_primer; 9526 struct netlink_ext_ack extack = {}; 9527 struct bpf_xdp_link *link; 9528 struct net_device *dev; 9529 int err, fd; 9530 9531 rtnl_lock(); 9532 dev = dev_get_by_index(net, attr->link_create.target_ifindex); 9533 if (!dev) { 9534 rtnl_unlock(); 9535 return -EINVAL; 9536 } 9537 9538 link = kzalloc(sizeof(*link), GFP_USER); 9539 if (!link) { 9540 err = -ENOMEM; 9541 goto unlock; 9542 } 9543 9544 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog); 9545 link->dev = dev; 9546 link->flags = attr->link_create.flags; 9547 9548 err = bpf_link_prime(&link->link, &link_primer); 9549 if (err) { 9550 kfree(link); 9551 goto unlock; 9552 } 9553 9554 err = dev_xdp_attach_link(dev, &extack, link); 9555 rtnl_unlock(); 9556 9557 if (err) { 9558 link->dev = NULL; 9559 bpf_link_cleanup(&link_primer); 9560 trace_bpf_xdp_link_attach_failed(extack._msg); 9561 goto out_put_dev; 9562 } 9563 9564 fd = bpf_link_settle(&link_primer); 9565 /* link itself doesn't hold dev's refcnt to not complicate shutdown */ 9566 dev_put(dev); 9567 return fd; 9568 9569 unlock: 9570 rtnl_unlock(); 9571 9572 out_put_dev: 9573 dev_put(dev); 9574 return err; 9575 } 9576 9577 /** 9578 * dev_change_xdp_fd - set or clear a bpf program for a device rx path 9579 * @dev: device 9580 * @extack: netlink extended ack 9581 * @fd: new program fd or negative value to clear 9582 * @expected_fd: old program fd that userspace expects to replace or clear 9583 * @flags: xdp-related flags 9584 * 9585 * Set or clear a bpf program for a device 9586 */ 9587 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, 9588 int fd, int expected_fd, u32 flags) 9589 { 9590 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags); 9591 struct bpf_prog *new_prog = NULL, *old_prog = NULL; 9592 int err; 9593 9594 ASSERT_RTNL(); 9595 9596 if (fd >= 0) { 9597 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP, 9598 mode != XDP_MODE_SKB); 9599 if (IS_ERR(new_prog)) 9600 return PTR_ERR(new_prog); 9601 } 9602 9603 if (expected_fd >= 0) { 9604 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP, 9605 mode != XDP_MODE_SKB); 9606 if (IS_ERR(old_prog)) { 9607 err = PTR_ERR(old_prog); 9608 old_prog = NULL; 9609 goto err_out; 9610 } 9611 } 9612 9613 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags); 9614 9615 err_out: 9616 if (err && new_prog) 9617 bpf_prog_put(new_prog); 9618 if (old_prog) 9619 bpf_prog_put(old_prog); 9620 return err; 9621 } 9622 9623 /** 9624 * dev_index_reserve() - allocate an ifindex in a namespace 9625 * @net: the applicable net namespace 9626 * @ifindex: requested ifindex, pass %0 to get one allocated 9627 * 9628 * Allocate a ifindex for a new device. Caller must either use the ifindex 9629 * to store the device (via list_netdevice()) or call dev_index_release() 9630 * to give the index up. 9631 * 9632 * Return: a suitable unique value for a new device interface number or -errno. 9633 */ 9634 static int dev_index_reserve(struct net *net, u32 ifindex) 9635 { 9636 int err; 9637 9638 if (ifindex > INT_MAX) { 9639 DEBUG_NET_WARN_ON_ONCE(1); 9640 return -EINVAL; 9641 } 9642 9643 if (!ifindex) 9644 err = xa_alloc_cyclic(&net->dev_by_index, &ifindex, NULL, 9645 xa_limit_31b, &net->ifindex, GFP_KERNEL); 9646 else 9647 err = xa_insert(&net->dev_by_index, ifindex, NULL, GFP_KERNEL); 9648 if (err < 0) 9649 return err; 9650 9651 return ifindex; 9652 } 9653 9654 static void dev_index_release(struct net *net, int ifindex) 9655 { 9656 /* Expect only unused indexes, unlist_netdevice() removes the used */ 9657 WARN_ON(xa_erase(&net->dev_by_index, ifindex)); 9658 } 9659 9660 /* Delayed registration/unregisteration */ 9661 LIST_HEAD(net_todo_list); 9662 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); 9663 9664 static void net_set_todo(struct net_device *dev) 9665 { 9666 list_add_tail(&dev->todo_list, &net_todo_list); 9667 atomic_inc(&dev_net(dev)->dev_unreg_count); 9668 } 9669 9670 static netdev_features_t netdev_sync_upper_features(struct net_device *lower, 9671 struct net_device *upper, netdev_features_t features) 9672 { 9673 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9674 netdev_features_t feature; 9675 int feature_bit; 9676 9677 for_each_netdev_feature(upper_disables, feature_bit) { 9678 feature = __NETIF_F_BIT(feature_bit); 9679 if (!(upper->wanted_features & feature) 9680 && (features & feature)) { 9681 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", 9682 &feature, upper->name); 9683 features &= ~feature; 9684 } 9685 } 9686 9687 return features; 9688 } 9689 9690 static void netdev_sync_lower_features(struct net_device *upper, 9691 struct net_device *lower, netdev_features_t features) 9692 { 9693 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9694 netdev_features_t feature; 9695 int feature_bit; 9696 9697 for_each_netdev_feature(upper_disables, feature_bit) { 9698 feature = __NETIF_F_BIT(feature_bit); 9699 if (!(features & feature) && (lower->features & feature)) { 9700 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", 9701 &feature, lower->name); 9702 lower->wanted_features &= ~feature; 9703 __netdev_update_features(lower); 9704 9705 if (unlikely(lower->features & feature)) 9706 netdev_WARN(upper, "failed to disable %pNF on %s!\n", 9707 &feature, lower->name); 9708 else 9709 netdev_features_change(lower); 9710 } 9711 } 9712 } 9713 9714 static netdev_features_t netdev_fix_features(struct net_device *dev, 9715 netdev_features_t features) 9716 { 9717 /* Fix illegal checksum combinations */ 9718 if ((features & NETIF_F_HW_CSUM) && 9719 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { 9720 netdev_warn(dev, "mixed HW and IP checksum settings.\n"); 9721 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); 9722 } 9723 9724 /* TSO requires that SG is present as well. */ 9725 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { 9726 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); 9727 features &= ~NETIF_F_ALL_TSO; 9728 } 9729 9730 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && 9731 !(features & NETIF_F_IP_CSUM)) { 9732 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); 9733 features &= ~NETIF_F_TSO; 9734 features &= ~NETIF_F_TSO_ECN; 9735 } 9736 9737 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && 9738 !(features & NETIF_F_IPV6_CSUM)) { 9739 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); 9740 features &= ~NETIF_F_TSO6; 9741 } 9742 9743 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ 9744 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) 9745 features &= ~NETIF_F_TSO_MANGLEID; 9746 9747 /* TSO ECN requires that TSO is present as well. */ 9748 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) 9749 features &= ~NETIF_F_TSO_ECN; 9750 9751 /* Software GSO depends on SG. */ 9752 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { 9753 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); 9754 features &= ~NETIF_F_GSO; 9755 } 9756 9757 /* GSO partial features require GSO partial be set */ 9758 if ((features & dev->gso_partial_features) && 9759 !(features & NETIF_F_GSO_PARTIAL)) { 9760 netdev_dbg(dev, 9761 "Dropping partially supported GSO features since no GSO partial.\n"); 9762 features &= ~dev->gso_partial_features; 9763 } 9764 9765 if (!(features & NETIF_F_RXCSUM)) { 9766 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet 9767 * successfully merged by hardware must also have the 9768 * checksum verified by hardware. If the user does not 9769 * want to enable RXCSUM, logically, we should disable GRO_HW. 9770 */ 9771 if (features & NETIF_F_GRO_HW) { 9772 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n"); 9773 features &= ~NETIF_F_GRO_HW; 9774 } 9775 } 9776 9777 /* LRO/HW-GRO features cannot be combined with RX-FCS */ 9778 if (features & NETIF_F_RXFCS) { 9779 if (features & NETIF_F_LRO) { 9780 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n"); 9781 features &= ~NETIF_F_LRO; 9782 } 9783 9784 if (features & NETIF_F_GRO_HW) { 9785 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n"); 9786 features &= ~NETIF_F_GRO_HW; 9787 } 9788 } 9789 9790 if ((features & NETIF_F_GRO_HW) && (features & NETIF_F_LRO)) { 9791 netdev_dbg(dev, "Dropping LRO feature since HW-GRO is requested.\n"); 9792 features &= ~NETIF_F_LRO; 9793 } 9794 9795 if (features & NETIF_F_HW_TLS_TX) { 9796 bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) == 9797 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM); 9798 bool hw_csum = features & NETIF_F_HW_CSUM; 9799 9800 if (!ip_csum && !hw_csum) { 9801 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n"); 9802 features &= ~NETIF_F_HW_TLS_TX; 9803 } 9804 } 9805 9806 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) { 9807 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n"); 9808 features &= ~NETIF_F_HW_TLS_RX; 9809 } 9810 9811 return features; 9812 } 9813 9814 int __netdev_update_features(struct net_device *dev) 9815 { 9816 struct net_device *upper, *lower; 9817 netdev_features_t features; 9818 struct list_head *iter; 9819 int err = -1; 9820 9821 ASSERT_RTNL(); 9822 9823 features = netdev_get_wanted_features(dev); 9824 9825 if (dev->netdev_ops->ndo_fix_features) 9826 features = dev->netdev_ops->ndo_fix_features(dev, features); 9827 9828 /* driver might be less strict about feature dependencies */ 9829 features = netdev_fix_features(dev, features); 9830 9831 /* some features can't be enabled if they're off on an upper device */ 9832 netdev_for_each_upper_dev_rcu(dev, upper, iter) 9833 features = netdev_sync_upper_features(dev, upper, features); 9834 9835 if (dev->features == features) 9836 goto sync_lower; 9837 9838 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", 9839 &dev->features, &features); 9840 9841 if (dev->netdev_ops->ndo_set_features) 9842 err = dev->netdev_ops->ndo_set_features(dev, features); 9843 else 9844 err = 0; 9845 9846 if (unlikely(err < 0)) { 9847 netdev_err(dev, 9848 "set_features() failed (%d); wanted %pNF, left %pNF\n", 9849 err, &features, &dev->features); 9850 /* return non-0 since some features might have changed and 9851 * it's better to fire a spurious notification than miss it 9852 */ 9853 return -1; 9854 } 9855 9856 sync_lower: 9857 /* some features must be disabled on lower devices when disabled 9858 * on an upper device (think: bonding master or bridge) 9859 */ 9860 netdev_for_each_lower_dev(dev, lower, iter) 9861 netdev_sync_lower_features(dev, lower, features); 9862 9863 if (!err) { 9864 netdev_features_t diff = features ^ dev->features; 9865 9866 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) { 9867 /* udp_tunnel_{get,drop}_rx_info both need 9868 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the 9869 * device, or they won't do anything. 9870 * Thus we need to update dev->features 9871 * *before* calling udp_tunnel_get_rx_info, 9872 * but *after* calling udp_tunnel_drop_rx_info. 9873 */ 9874 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) { 9875 dev->features = features; 9876 udp_tunnel_get_rx_info(dev); 9877 } else { 9878 udp_tunnel_drop_rx_info(dev); 9879 } 9880 } 9881 9882 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) { 9883 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) { 9884 dev->features = features; 9885 err |= vlan_get_rx_ctag_filter_info(dev); 9886 } else { 9887 vlan_drop_rx_ctag_filter_info(dev); 9888 } 9889 } 9890 9891 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) { 9892 if (features & NETIF_F_HW_VLAN_STAG_FILTER) { 9893 dev->features = features; 9894 err |= vlan_get_rx_stag_filter_info(dev); 9895 } else { 9896 vlan_drop_rx_stag_filter_info(dev); 9897 } 9898 } 9899 9900 dev->features = features; 9901 } 9902 9903 return err < 0 ? 0 : 1; 9904 } 9905 9906 /** 9907 * netdev_update_features - recalculate device features 9908 * @dev: the device to check 9909 * 9910 * Recalculate dev->features set and send notifications if it 9911 * has changed. Should be called after driver or hardware dependent 9912 * conditions might have changed that influence the features. 9913 */ 9914 void netdev_update_features(struct net_device *dev) 9915 { 9916 if (__netdev_update_features(dev)) 9917 netdev_features_change(dev); 9918 } 9919 EXPORT_SYMBOL(netdev_update_features); 9920 9921 /** 9922 * netdev_change_features - recalculate device features 9923 * @dev: the device to check 9924 * 9925 * Recalculate dev->features set and send notifications even 9926 * if they have not changed. Should be called instead of 9927 * netdev_update_features() if also dev->vlan_features might 9928 * have changed to allow the changes to be propagated to stacked 9929 * VLAN devices. 9930 */ 9931 void netdev_change_features(struct net_device *dev) 9932 { 9933 __netdev_update_features(dev); 9934 netdev_features_change(dev); 9935 } 9936 EXPORT_SYMBOL(netdev_change_features); 9937 9938 /** 9939 * netif_stacked_transfer_operstate - transfer operstate 9940 * @rootdev: the root or lower level device to transfer state from 9941 * @dev: the device to transfer operstate to 9942 * 9943 * Transfer operational state from root to device. This is normally 9944 * called when a stacking relationship exists between the root 9945 * device and the device(a leaf device). 9946 */ 9947 void netif_stacked_transfer_operstate(const struct net_device *rootdev, 9948 struct net_device *dev) 9949 { 9950 if (rootdev->operstate == IF_OPER_DORMANT) 9951 netif_dormant_on(dev); 9952 else 9953 netif_dormant_off(dev); 9954 9955 if (rootdev->operstate == IF_OPER_TESTING) 9956 netif_testing_on(dev); 9957 else 9958 netif_testing_off(dev); 9959 9960 if (netif_carrier_ok(rootdev)) 9961 netif_carrier_on(dev); 9962 else 9963 netif_carrier_off(dev); 9964 } 9965 EXPORT_SYMBOL(netif_stacked_transfer_operstate); 9966 9967 static int netif_alloc_rx_queues(struct net_device *dev) 9968 { 9969 unsigned int i, count = dev->num_rx_queues; 9970 struct netdev_rx_queue *rx; 9971 size_t sz = count * sizeof(*rx); 9972 int err = 0; 9973 9974 BUG_ON(count < 1); 9975 9976 rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 9977 if (!rx) 9978 return -ENOMEM; 9979 9980 dev->_rx = rx; 9981 9982 for (i = 0; i < count; i++) { 9983 rx[i].dev = dev; 9984 9985 /* XDP RX-queue setup */ 9986 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0); 9987 if (err < 0) 9988 goto err_rxq_info; 9989 } 9990 return 0; 9991 9992 err_rxq_info: 9993 /* Rollback successful reg's and free other resources */ 9994 while (i--) 9995 xdp_rxq_info_unreg(&rx[i].xdp_rxq); 9996 kvfree(dev->_rx); 9997 dev->_rx = NULL; 9998 return err; 9999 } 10000 10001 static void netif_free_rx_queues(struct net_device *dev) 10002 { 10003 unsigned int i, count = dev->num_rx_queues; 10004 10005 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */ 10006 if (!dev->_rx) 10007 return; 10008 10009 for (i = 0; i < count; i++) 10010 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq); 10011 10012 kvfree(dev->_rx); 10013 } 10014 10015 static void netdev_init_one_queue(struct net_device *dev, 10016 struct netdev_queue *queue, void *_unused) 10017 { 10018 /* Initialize queue lock */ 10019 spin_lock_init(&queue->_xmit_lock); 10020 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); 10021 queue->xmit_lock_owner = -1; 10022 netdev_queue_numa_node_write(queue, NUMA_NO_NODE); 10023 queue->dev = dev; 10024 #ifdef CONFIG_BQL 10025 dql_init(&queue->dql, HZ); 10026 #endif 10027 } 10028 10029 static void netif_free_tx_queues(struct net_device *dev) 10030 { 10031 kvfree(dev->_tx); 10032 } 10033 10034 static int netif_alloc_netdev_queues(struct net_device *dev) 10035 { 10036 unsigned int count = dev->num_tx_queues; 10037 struct netdev_queue *tx; 10038 size_t sz = count * sizeof(*tx); 10039 10040 if (count < 1 || count > 0xffff) 10041 return -EINVAL; 10042 10043 tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10044 if (!tx) 10045 return -ENOMEM; 10046 10047 dev->_tx = tx; 10048 10049 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); 10050 spin_lock_init(&dev->tx_global_lock); 10051 10052 return 0; 10053 } 10054 10055 void netif_tx_stop_all_queues(struct net_device *dev) 10056 { 10057 unsigned int i; 10058 10059 for (i = 0; i < dev->num_tx_queues; i++) { 10060 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 10061 10062 netif_tx_stop_queue(txq); 10063 } 10064 } 10065 EXPORT_SYMBOL(netif_tx_stop_all_queues); 10066 10067 static int netdev_do_alloc_pcpu_stats(struct net_device *dev) 10068 { 10069 void __percpu *v; 10070 10071 /* Drivers implementing ndo_get_peer_dev must support tstat 10072 * accounting, so that skb_do_redirect() can bump the dev's 10073 * RX stats upon network namespace switch. 10074 */ 10075 if (dev->netdev_ops->ndo_get_peer_dev && 10076 dev->pcpu_stat_type != NETDEV_PCPU_STAT_TSTATS) 10077 return -EOPNOTSUPP; 10078 10079 switch (dev->pcpu_stat_type) { 10080 case NETDEV_PCPU_STAT_NONE: 10081 return 0; 10082 case NETDEV_PCPU_STAT_LSTATS: 10083 v = dev->lstats = netdev_alloc_pcpu_stats(struct pcpu_lstats); 10084 break; 10085 case NETDEV_PCPU_STAT_TSTATS: 10086 v = dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats); 10087 break; 10088 case NETDEV_PCPU_STAT_DSTATS: 10089 v = dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats); 10090 break; 10091 default: 10092 return -EINVAL; 10093 } 10094 10095 return v ? 0 : -ENOMEM; 10096 } 10097 10098 static void netdev_do_free_pcpu_stats(struct net_device *dev) 10099 { 10100 switch (dev->pcpu_stat_type) { 10101 case NETDEV_PCPU_STAT_NONE: 10102 return; 10103 case NETDEV_PCPU_STAT_LSTATS: 10104 free_percpu(dev->lstats); 10105 break; 10106 case NETDEV_PCPU_STAT_TSTATS: 10107 free_percpu(dev->tstats); 10108 break; 10109 case NETDEV_PCPU_STAT_DSTATS: 10110 free_percpu(dev->dstats); 10111 break; 10112 } 10113 } 10114 10115 /** 10116 * register_netdevice() - register a network device 10117 * @dev: device to register 10118 * 10119 * Take a prepared network device structure and make it externally accessible. 10120 * A %NETDEV_REGISTER message is sent to the netdev notifier chain. 10121 * Callers must hold the rtnl lock - you may want register_netdev() 10122 * instead of this. 10123 */ 10124 int register_netdevice(struct net_device *dev) 10125 { 10126 int ret; 10127 struct net *net = dev_net(dev); 10128 10129 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE < 10130 NETDEV_FEATURE_COUNT); 10131 BUG_ON(dev_boot_phase); 10132 ASSERT_RTNL(); 10133 10134 might_sleep(); 10135 10136 /* When net_device's are persistent, this will be fatal. */ 10137 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); 10138 BUG_ON(!net); 10139 10140 ret = ethtool_check_ops(dev->ethtool_ops); 10141 if (ret) 10142 return ret; 10143 10144 spin_lock_init(&dev->addr_list_lock); 10145 netdev_set_addr_lockdep_class(dev); 10146 10147 ret = dev_get_valid_name(net, dev, dev->name); 10148 if (ret < 0) 10149 goto out; 10150 10151 ret = -ENOMEM; 10152 dev->name_node = netdev_name_node_head_alloc(dev); 10153 if (!dev->name_node) 10154 goto out; 10155 10156 /* Init, if this function is available */ 10157 if (dev->netdev_ops->ndo_init) { 10158 ret = dev->netdev_ops->ndo_init(dev); 10159 if (ret) { 10160 if (ret > 0) 10161 ret = -EIO; 10162 goto err_free_name; 10163 } 10164 } 10165 10166 if (((dev->hw_features | dev->features) & 10167 NETIF_F_HW_VLAN_CTAG_FILTER) && 10168 (!dev->netdev_ops->ndo_vlan_rx_add_vid || 10169 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { 10170 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); 10171 ret = -EINVAL; 10172 goto err_uninit; 10173 } 10174 10175 ret = netdev_do_alloc_pcpu_stats(dev); 10176 if (ret) 10177 goto err_uninit; 10178 10179 ret = dev_index_reserve(net, dev->ifindex); 10180 if (ret < 0) 10181 goto err_free_pcpu; 10182 dev->ifindex = ret; 10183 10184 /* Transfer changeable features to wanted_features and enable 10185 * software offloads (GSO and GRO). 10186 */ 10187 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF); 10188 dev->features |= NETIF_F_SOFT_FEATURES; 10189 10190 if (dev->udp_tunnel_nic_info) { 10191 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10192 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10193 } 10194 10195 dev->wanted_features = dev->features & dev->hw_features; 10196 10197 if (!(dev->flags & IFF_LOOPBACK)) 10198 dev->hw_features |= NETIF_F_NOCACHE_COPY; 10199 10200 /* If IPv4 TCP segmentation offload is supported we should also 10201 * allow the device to enable segmenting the frame with the option 10202 * of ignoring a static IP ID value. This doesn't enable the 10203 * feature itself but allows the user to enable it later. 10204 */ 10205 if (dev->hw_features & NETIF_F_TSO) 10206 dev->hw_features |= NETIF_F_TSO_MANGLEID; 10207 if (dev->vlan_features & NETIF_F_TSO) 10208 dev->vlan_features |= NETIF_F_TSO_MANGLEID; 10209 if (dev->mpls_features & NETIF_F_TSO) 10210 dev->mpls_features |= NETIF_F_TSO_MANGLEID; 10211 if (dev->hw_enc_features & NETIF_F_TSO) 10212 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; 10213 10214 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. 10215 */ 10216 dev->vlan_features |= NETIF_F_HIGHDMA; 10217 10218 /* Make NETIF_F_SG inheritable to tunnel devices. 10219 */ 10220 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; 10221 10222 /* Make NETIF_F_SG inheritable to MPLS. 10223 */ 10224 dev->mpls_features |= NETIF_F_SG; 10225 10226 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); 10227 ret = notifier_to_errno(ret); 10228 if (ret) 10229 goto err_ifindex_release; 10230 10231 ret = netdev_register_kobject(dev); 10232 write_lock(&dev_base_lock); 10233 dev->reg_state = ret ? NETREG_UNREGISTERED : NETREG_REGISTERED; 10234 write_unlock(&dev_base_lock); 10235 if (ret) 10236 goto err_uninit_notify; 10237 10238 __netdev_update_features(dev); 10239 10240 /* 10241 * Default initial state at registry is that the 10242 * device is present. 10243 */ 10244 10245 set_bit(__LINK_STATE_PRESENT, &dev->state); 10246 10247 linkwatch_init_dev(dev); 10248 10249 dev_init_scheduler(dev); 10250 10251 netdev_hold(dev, &dev->dev_registered_tracker, GFP_KERNEL); 10252 list_netdevice(dev); 10253 10254 add_device_randomness(dev->dev_addr, dev->addr_len); 10255 10256 /* If the device has permanent device address, driver should 10257 * set dev_addr and also addr_assign_type should be set to 10258 * NET_ADDR_PERM (default value). 10259 */ 10260 if (dev->addr_assign_type == NET_ADDR_PERM) 10261 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); 10262 10263 /* Notify protocols, that a new device appeared. */ 10264 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); 10265 ret = notifier_to_errno(ret); 10266 if (ret) { 10267 /* Expect explicit free_netdev() on failure */ 10268 dev->needs_free_netdev = false; 10269 unregister_netdevice_queue(dev, NULL); 10270 goto out; 10271 } 10272 /* 10273 * Prevent userspace races by waiting until the network 10274 * device is fully setup before sending notifications. 10275 */ 10276 if (!dev->rtnl_link_ops || 10277 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 10278 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL); 10279 10280 out: 10281 return ret; 10282 10283 err_uninit_notify: 10284 call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev); 10285 err_ifindex_release: 10286 dev_index_release(net, dev->ifindex); 10287 err_free_pcpu: 10288 netdev_do_free_pcpu_stats(dev); 10289 err_uninit: 10290 if (dev->netdev_ops->ndo_uninit) 10291 dev->netdev_ops->ndo_uninit(dev); 10292 if (dev->priv_destructor) 10293 dev->priv_destructor(dev); 10294 err_free_name: 10295 netdev_name_node_free(dev->name_node); 10296 goto out; 10297 } 10298 EXPORT_SYMBOL(register_netdevice); 10299 10300 /** 10301 * init_dummy_netdev - init a dummy network device for NAPI 10302 * @dev: device to init 10303 * 10304 * This takes a network device structure and initialize the minimum 10305 * amount of fields so it can be used to schedule NAPI polls without 10306 * registering a full blown interface. This is to be used by drivers 10307 * that need to tie several hardware interfaces to a single NAPI 10308 * poll scheduler due to HW limitations. 10309 */ 10310 int init_dummy_netdev(struct net_device *dev) 10311 { 10312 /* Clear everything. Note we don't initialize spinlocks 10313 * are they aren't supposed to be taken by any of the 10314 * NAPI code and this dummy netdev is supposed to be 10315 * only ever used for NAPI polls 10316 */ 10317 memset(dev, 0, sizeof(struct net_device)); 10318 10319 /* make sure we BUG if trying to hit standard 10320 * register/unregister code path 10321 */ 10322 dev->reg_state = NETREG_DUMMY; 10323 10324 /* NAPI wants this */ 10325 INIT_LIST_HEAD(&dev->napi_list); 10326 10327 /* a dummy interface is started by default */ 10328 set_bit(__LINK_STATE_PRESENT, &dev->state); 10329 set_bit(__LINK_STATE_START, &dev->state); 10330 10331 /* napi_busy_loop stats accounting wants this */ 10332 dev_net_set(dev, &init_net); 10333 10334 /* Note : We dont allocate pcpu_refcnt for dummy devices, 10335 * because users of this 'device' dont need to change 10336 * its refcount. 10337 */ 10338 10339 return 0; 10340 } 10341 EXPORT_SYMBOL_GPL(init_dummy_netdev); 10342 10343 10344 /** 10345 * register_netdev - register a network device 10346 * @dev: device to register 10347 * 10348 * Take a completed network device structure and add it to the kernel 10349 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 10350 * chain. 0 is returned on success. A negative errno code is returned 10351 * on a failure to set up the device, or if the name is a duplicate. 10352 * 10353 * This is a wrapper around register_netdevice that takes the rtnl semaphore 10354 * and expands the device name if you passed a format string to 10355 * alloc_netdev. 10356 */ 10357 int register_netdev(struct net_device *dev) 10358 { 10359 int err; 10360 10361 if (rtnl_lock_killable()) 10362 return -EINTR; 10363 err = register_netdevice(dev); 10364 rtnl_unlock(); 10365 return err; 10366 } 10367 EXPORT_SYMBOL(register_netdev); 10368 10369 int netdev_refcnt_read(const struct net_device *dev) 10370 { 10371 #ifdef CONFIG_PCPU_DEV_REFCNT 10372 int i, refcnt = 0; 10373 10374 for_each_possible_cpu(i) 10375 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); 10376 return refcnt; 10377 #else 10378 return refcount_read(&dev->dev_refcnt); 10379 #endif 10380 } 10381 EXPORT_SYMBOL(netdev_refcnt_read); 10382 10383 int netdev_unregister_timeout_secs __read_mostly = 10; 10384 10385 #define WAIT_REFS_MIN_MSECS 1 10386 #define WAIT_REFS_MAX_MSECS 250 10387 /** 10388 * netdev_wait_allrefs_any - wait until all references are gone. 10389 * @list: list of net_devices to wait on 10390 * 10391 * This is called when unregistering network devices. 10392 * 10393 * Any protocol or device that holds a reference should register 10394 * for netdevice notification, and cleanup and put back the 10395 * reference if they receive an UNREGISTER event. 10396 * We can get stuck here if buggy protocols don't correctly 10397 * call dev_put. 10398 */ 10399 static struct net_device *netdev_wait_allrefs_any(struct list_head *list) 10400 { 10401 unsigned long rebroadcast_time, warning_time; 10402 struct net_device *dev; 10403 int wait = 0; 10404 10405 rebroadcast_time = warning_time = jiffies; 10406 10407 list_for_each_entry(dev, list, todo_list) 10408 if (netdev_refcnt_read(dev) == 1) 10409 return dev; 10410 10411 while (true) { 10412 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { 10413 rtnl_lock(); 10414 10415 /* Rebroadcast unregister notification */ 10416 list_for_each_entry(dev, list, todo_list) 10417 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10418 10419 __rtnl_unlock(); 10420 rcu_barrier(); 10421 rtnl_lock(); 10422 10423 list_for_each_entry(dev, list, todo_list) 10424 if (test_bit(__LINK_STATE_LINKWATCH_PENDING, 10425 &dev->state)) { 10426 /* We must not have linkwatch events 10427 * pending on unregister. If this 10428 * happens, we simply run the queue 10429 * unscheduled, resulting in a noop 10430 * for this device. 10431 */ 10432 linkwatch_run_queue(); 10433 break; 10434 } 10435 10436 __rtnl_unlock(); 10437 10438 rebroadcast_time = jiffies; 10439 } 10440 10441 rcu_barrier(); 10442 10443 if (!wait) { 10444 wait = WAIT_REFS_MIN_MSECS; 10445 } else { 10446 msleep(wait); 10447 wait = min(wait << 1, WAIT_REFS_MAX_MSECS); 10448 } 10449 10450 list_for_each_entry(dev, list, todo_list) 10451 if (netdev_refcnt_read(dev) == 1) 10452 return dev; 10453 10454 if (time_after(jiffies, warning_time + 10455 READ_ONCE(netdev_unregister_timeout_secs) * HZ)) { 10456 list_for_each_entry(dev, list, todo_list) { 10457 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", 10458 dev->name, netdev_refcnt_read(dev)); 10459 ref_tracker_dir_print(&dev->refcnt_tracker, 10); 10460 } 10461 10462 warning_time = jiffies; 10463 } 10464 } 10465 } 10466 10467 /* The sequence is: 10468 * 10469 * rtnl_lock(); 10470 * ... 10471 * register_netdevice(x1); 10472 * register_netdevice(x2); 10473 * ... 10474 * unregister_netdevice(y1); 10475 * unregister_netdevice(y2); 10476 * ... 10477 * rtnl_unlock(); 10478 * free_netdev(y1); 10479 * free_netdev(y2); 10480 * 10481 * We are invoked by rtnl_unlock(). 10482 * This allows us to deal with problems: 10483 * 1) We can delete sysfs objects which invoke hotplug 10484 * without deadlocking with linkwatch via keventd. 10485 * 2) Since we run with the RTNL semaphore not held, we can sleep 10486 * safely in order to wait for the netdev refcnt to drop to zero. 10487 * 10488 * We must not return until all unregister events added during 10489 * the interval the lock was held have been completed. 10490 */ 10491 void netdev_run_todo(void) 10492 { 10493 struct net_device *dev, *tmp; 10494 struct list_head list; 10495 #ifdef CONFIG_LOCKDEP 10496 struct list_head unlink_list; 10497 10498 list_replace_init(&net_unlink_list, &unlink_list); 10499 10500 while (!list_empty(&unlink_list)) { 10501 struct net_device *dev = list_first_entry(&unlink_list, 10502 struct net_device, 10503 unlink_list); 10504 list_del_init(&dev->unlink_list); 10505 dev->nested_level = dev->lower_level - 1; 10506 } 10507 #endif 10508 10509 /* Snapshot list, allow later requests */ 10510 list_replace_init(&net_todo_list, &list); 10511 10512 __rtnl_unlock(); 10513 10514 /* Wait for rcu callbacks to finish before next phase */ 10515 if (!list_empty(&list)) 10516 rcu_barrier(); 10517 10518 list_for_each_entry_safe(dev, tmp, &list, todo_list) { 10519 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { 10520 netdev_WARN(dev, "run_todo but not unregistering\n"); 10521 list_del(&dev->todo_list); 10522 continue; 10523 } 10524 10525 write_lock(&dev_base_lock); 10526 dev->reg_state = NETREG_UNREGISTERED; 10527 write_unlock(&dev_base_lock); 10528 linkwatch_forget_dev(dev); 10529 } 10530 10531 while (!list_empty(&list)) { 10532 dev = netdev_wait_allrefs_any(&list); 10533 list_del(&dev->todo_list); 10534 10535 /* paranoia */ 10536 BUG_ON(netdev_refcnt_read(dev) != 1); 10537 BUG_ON(!list_empty(&dev->ptype_all)); 10538 BUG_ON(!list_empty(&dev->ptype_specific)); 10539 WARN_ON(rcu_access_pointer(dev->ip_ptr)); 10540 WARN_ON(rcu_access_pointer(dev->ip6_ptr)); 10541 10542 netdev_do_free_pcpu_stats(dev); 10543 if (dev->priv_destructor) 10544 dev->priv_destructor(dev); 10545 if (dev->needs_free_netdev) 10546 free_netdev(dev); 10547 10548 if (atomic_dec_and_test(&dev_net(dev)->dev_unreg_count)) 10549 wake_up(&netdev_unregistering_wq); 10550 10551 /* Free network device */ 10552 kobject_put(&dev->dev.kobj); 10553 } 10554 } 10555 10556 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has 10557 * all the same fields in the same order as net_device_stats, with only 10558 * the type differing, but rtnl_link_stats64 may have additional fields 10559 * at the end for newer counters. 10560 */ 10561 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, 10562 const struct net_device_stats *netdev_stats) 10563 { 10564 size_t i, n = sizeof(*netdev_stats) / sizeof(atomic_long_t); 10565 const atomic_long_t *src = (atomic_long_t *)netdev_stats; 10566 u64 *dst = (u64 *)stats64; 10567 10568 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); 10569 for (i = 0; i < n; i++) 10570 dst[i] = (unsigned long)atomic_long_read(&src[i]); 10571 /* zero out counters that only exist in rtnl_link_stats64 */ 10572 memset((char *)stats64 + n * sizeof(u64), 0, 10573 sizeof(*stats64) - n * sizeof(u64)); 10574 } 10575 EXPORT_SYMBOL(netdev_stats_to_stats64); 10576 10577 struct net_device_core_stats __percpu *netdev_core_stats_alloc(struct net_device *dev) 10578 { 10579 struct net_device_core_stats __percpu *p; 10580 10581 p = alloc_percpu_gfp(struct net_device_core_stats, 10582 GFP_ATOMIC | __GFP_NOWARN); 10583 10584 if (p && cmpxchg(&dev->core_stats, NULL, p)) 10585 free_percpu(p); 10586 10587 /* This READ_ONCE() pairs with the cmpxchg() above */ 10588 return READ_ONCE(dev->core_stats); 10589 } 10590 EXPORT_SYMBOL(netdev_core_stats_alloc); 10591 10592 /** 10593 * dev_get_stats - get network device statistics 10594 * @dev: device to get statistics from 10595 * @storage: place to store stats 10596 * 10597 * Get network statistics from device. Return @storage. 10598 * The device driver may provide its own method by setting 10599 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; 10600 * otherwise the internal statistics structure is used. 10601 */ 10602 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, 10603 struct rtnl_link_stats64 *storage) 10604 { 10605 const struct net_device_ops *ops = dev->netdev_ops; 10606 const struct net_device_core_stats __percpu *p; 10607 10608 if (ops->ndo_get_stats64) { 10609 memset(storage, 0, sizeof(*storage)); 10610 ops->ndo_get_stats64(dev, storage); 10611 } else if (ops->ndo_get_stats) { 10612 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); 10613 } else { 10614 netdev_stats_to_stats64(storage, &dev->stats); 10615 } 10616 10617 /* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */ 10618 p = READ_ONCE(dev->core_stats); 10619 if (p) { 10620 const struct net_device_core_stats *core_stats; 10621 int i; 10622 10623 for_each_possible_cpu(i) { 10624 core_stats = per_cpu_ptr(p, i); 10625 storage->rx_dropped += READ_ONCE(core_stats->rx_dropped); 10626 storage->tx_dropped += READ_ONCE(core_stats->tx_dropped); 10627 storage->rx_nohandler += READ_ONCE(core_stats->rx_nohandler); 10628 storage->rx_otherhost_dropped += READ_ONCE(core_stats->rx_otherhost_dropped); 10629 } 10630 } 10631 return storage; 10632 } 10633 EXPORT_SYMBOL(dev_get_stats); 10634 10635 /** 10636 * dev_fetch_sw_netstats - get per-cpu network device statistics 10637 * @s: place to store stats 10638 * @netstats: per-cpu network stats to read from 10639 * 10640 * Read per-cpu network statistics and populate the related fields in @s. 10641 */ 10642 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s, 10643 const struct pcpu_sw_netstats __percpu *netstats) 10644 { 10645 int cpu; 10646 10647 for_each_possible_cpu(cpu) { 10648 u64 rx_packets, rx_bytes, tx_packets, tx_bytes; 10649 const struct pcpu_sw_netstats *stats; 10650 unsigned int start; 10651 10652 stats = per_cpu_ptr(netstats, cpu); 10653 do { 10654 start = u64_stats_fetch_begin(&stats->syncp); 10655 rx_packets = u64_stats_read(&stats->rx_packets); 10656 rx_bytes = u64_stats_read(&stats->rx_bytes); 10657 tx_packets = u64_stats_read(&stats->tx_packets); 10658 tx_bytes = u64_stats_read(&stats->tx_bytes); 10659 } while (u64_stats_fetch_retry(&stats->syncp, start)); 10660 10661 s->rx_packets += rx_packets; 10662 s->rx_bytes += rx_bytes; 10663 s->tx_packets += tx_packets; 10664 s->tx_bytes += tx_bytes; 10665 } 10666 } 10667 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats); 10668 10669 /** 10670 * dev_get_tstats64 - ndo_get_stats64 implementation 10671 * @dev: device to get statistics from 10672 * @s: place to store stats 10673 * 10674 * Populate @s from dev->stats and dev->tstats. Can be used as 10675 * ndo_get_stats64() callback. 10676 */ 10677 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s) 10678 { 10679 netdev_stats_to_stats64(s, &dev->stats); 10680 dev_fetch_sw_netstats(s, dev->tstats); 10681 } 10682 EXPORT_SYMBOL_GPL(dev_get_tstats64); 10683 10684 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) 10685 { 10686 struct netdev_queue *queue = dev_ingress_queue(dev); 10687 10688 #ifdef CONFIG_NET_CLS_ACT 10689 if (queue) 10690 return queue; 10691 queue = kzalloc(sizeof(*queue), GFP_KERNEL); 10692 if (!queue) 10693 return NULL; 10694 netdev_init_one_queue(dev, queue, NULL); 10695 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); 10696 RCU_INIT_POINTER(queue->qdisc_sleeping, &noop_qdisc); 10697 rcu_assign_pointer(dev->ingress_queue, queue); 10698 #endif 10699 return queue; 10700 } 10701 10702 static const struct ethtool_ops default_ethtool_ops; 10703 10704 void netdev_set_default_ethtool_ops(struct net_device *dev, 10705 const struct ethtool_ops *ops) 10706 { 10707 if (dev->ethtool_ops == &default_ethtool_ops) 10708 dev->ethtool_ops = ops; 10709 } 10710 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); 10711 10712 /** 10713 * netdev_sw_irq_coalesce_default_on() - enable SW IRQ coalescing by default 10714 * @dev: netdev to enable the IRQ coalescing on 10715 * 10716 * Sets a conservative default for SW IRQ coalescing. Users can use 10717 * sysfs attributes to override the default values. 10718 */ 10719 void netdev_sw_irq_coalesce_default_on(struct net_device *dev) 10720 { 10721 WARN_ON(dev->reg_state == NETREG_REGISTERED); 10722 10723 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) { 10724 dev->gro_flush_timeout = 20000; 10725 dev->napi_defer_hard_irqs = 1; 10726 } 10727 } 10728 EXPORT_SYMBOL_GPL(netdev_sw_irq_coalesce_default_on); 10729 10730 void netdev_freemem(struct net_device *dev) 10731 { 10732 char *addr = (char *)dev - dev->padded; 10733 10734 kvfree(addr); 10735 } 10736 10737 /** 10738 * alloc_netdev_mqs - allocate network device 10739 * @sizeof_priv: size of private data to allocate space for 10740 * @name: device name format string 10741 * @name_assign_type: origin of device name 10742 * @setup: callback to initialize device 10743 * @txqs: the number of TX subqueues to allocate 10744 * @rxqs: the number of RX subqueues to allocate 10745 * 10746 * Allocates a struct net_device with private data area for driver use 10747 * and performs basic initialization. Also allocates subqueue structs 10748 * for each queue on the device. 10749 */ 10750 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, 10751 unsigned char name_assign_type, 10752 void (*setup)(struct net_device *), 10753 unsigned int txqs, unsigned int rxqs) 10754 { 10755 struct net_device *dev; 10756 unsigned int alloc_size; 10757 struct net_device *p; 10758 10759 BUG_ON(strlen(name) >= sizeof(dev->name)); 10760 10761 if (txqs < 1) { 10762 pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); 10763 return NULL; 10764 } 10765 10766 if (rxqs < 1) { 10767 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); 10768 return NULL; 10769 } 10770 10771 alloc_size = sizeof(struct net_device); 10772 if (sizeof_priv) { 10773 /* ensure 32-byte alignment of private area */ 10774 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); 10775 alloc_size += sizeof_priv; 10776 } 10777 /* ensure 32-byte alignment of whole construct */ 10778 alloc_size += NETDEV_ALIGN - 1; 10779 10780 p = kvzalloc(alloc_size, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10781 if (!p) 10782 return NULL; 10783 10784 dev = PTR_ALIGN(p, NETDEV_ALIGN); 10785 dev->padded = (char *)dev - (char *)p; 10786 10787 ref_tracker_dir_init(&dev->refcnt_tracker, 128, name); 10788 #ifdef CONFIG_PCPU_DEV_REFCNT 10789 dev->pcpu_refcnt = alloc_percpu(int); 10790 if (!dev->pcpu_refcnt) 10791 goto free_dev; 10792 __dev_hold(dev); 10793 #else 10794 refcount_set(&dev->dev_refcnt, 1); 10795 #endif 10796 10797 if (dev_addr_init(dev)) 10798 goto free_pcpu; 10799 10800 dev_mc_init(dev); 10801 dev_uc_init(dev); 10802 10803 dev_net_set(dev, &init_net); 10804 10805 dev->gso_max_size = GSO_LEGACY_MAX_SIZE; 10806 dev->xdp_zc_max_segs = 1; 10807 dev->gso_max_segs = GSO_MAX_SEGS; 10808 dev->gro_max_size = GRO_LEGACY_MAX_SIZE; 10809 dev->gso_ipv4_max_size = GSO_LEGACY_MAX_SIZE; 10810 dev->gro_ipv4_max_size = GRO_LEGACY_MAX_SIZE; 10811 dev->tso_max_size = TSO_LEGACY_MAX_SIZE; 10812 dev->tso_max_segs = TSO_MAX_SEGS; 10813 dev->upper_level = 1; 10814 dev->lower_level = 1; 10815 #ifdef CONFIG_LOCKDEP 10816 dev->nested_level = 0; 10817 INIT_LIST_HEAD(&dev->unlink_list); 10818 #endif 10819 10820 INIT_LIST_HEAD(&dev->napi_list); 10821 INIT_LIST_HEAD(&dev->unreg_list); 10822 INIT_LIST_HEAD(&dev->close_list); 10823 INIT_LIST_HEAD(&dev->link_watch_list); 10824 INIT_LIST_HEAD(&dev->adj_list.upper); 10825 INIT_LIST_HEAD(&dev->adj_list.lower); 10826 INIT_LIST_HEAD(&dev->ptype_all); 10827 INIT_LIST_HEAD(&dev->ptype_specific); 10828 INIT_LIST_HEAD(&dev->net_notifier_list); 10829 #ifdef CONFIG_NET_SCHED 10830 hash_init(dev->qdisc_hash); 10831 #endif 10832 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; 10833 setup(dev); 10834 10835 if (!dev->tx_queue_len) { 10836 dev->priv_flags |= IFF_NO_QUEUE; 10837 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; 10838 } 10839 10840 dev->num_tx_queues = txqs; 10841 dev->real_num_tx_queues = txqs; 10842 if (netif_alloc_netdev_queues(dev)) 10843 goto free_all; 10844 10845 dev->num_rx_queues = rxqs; 10846 dev->real_num_rx_queues = rxqs; 10847 if (netif_alloc_rx_queues(dev)) 10848 goto free_all; 10849 10850 strcpy(dev->name, name); 10851 dev->name_assign_type = name_assign_type; 10852 dev->group = INIT_NETDEV_GROUP; 10853 if (!dev->ethtool_ops) 10854 dev->ethtool_ops = &default_ethtool_ops; 10855 10856 nf_hook_netdev_init(dev); 10857 10858 return dev; 10859 10860 free_all: 10861 free_netdev(dev); 10862 return NULL; 10863 10864 free_pcpu: 10865 #ifdef CONFIG_PCPU_DEV_REFCNT 10866 free_percpu(dev->pcpu_refcnt); 10867 free_dev: 10868 #endif 10869 netdev_freemem(dev); 10870 return NULL; 10871 } 10872 EXPORT_SYMBOL(alloc_netdev_mqs); 10873 10874 /** 10875 * free_netdev - free network device 10876 * @dev: device 10877 * 10878 * This function does the last stage of destroying an allocated device 10879 * interface. The reference to the device object is released. If this 10880 * is the last reference then it will be freed.Must be called in process 10881 * context. 10882 */ 10883 void free_netdev(struct net_device *dev) 10884 { 10885 struct napi_struct *p, *n; 10886 10887 might_sleep(); 10888 10889 /* When called immediately after register_netdevice() failed the unwind 10890 * handling may still be dismantling the device. Handle that case by 10891 * deferring the free. 10892 */ 10893 if (dev->reg_state == NETREG_UNREGISTERING) { 10894 ASSERT_RTNL(); 10895 dev->needs_free_netdev = true; 10896 return; 10897 } 10898 10899 netif_free_tx_queues(dev); 10900 netif_free_rx_queues(dev); 10901 10902 kfree(rcu_dereference_protected(dev->ingress_queue, 1)); 10903 10904 /* Flush device addresses */ 10905 dev_addr_flush(dev); 10906 10907 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) 10908 netif_napi_del(p); 10909 10910 ref_tracker_dir_exit(&dev->refcnt_tracker); 10911 #ifdef CONFIG_PCPU_DEV_REFCNT 10912 free_percpu(dev->pcpu_refcnt); 10913 dev->pcpu_refcnt = NULL; 10914 #endif 10915 free_percpu(dev->core_stats); 10916 dev->core_stats = NULL; 10917 free_percpu(dev->xdp_bulkq); 10918 dev->xdp_bulkq = NULL; 10919 10920 /* Compatibility with error handling in drivers */ 10921 if (dev->reg_state == NETREG_UNINITIALIZED) { 10922 netdev_freemem(dev); 10923 return; 10924 } 10925 10926 BUG_ON(dev->reg_state != NETREG_UNREGISTERED); 10927 dev->reg_state = NETREG_RELEASED; 10928 10929 /* will free via device release */ 10930 put_device(&dev->dev); 10931 } 10932 EXPORT_SYMBOL(free_netdev); 10933 10934 /** 10935 * synchronize_net - Synchronize with packet receive processing 10936 * 10937 * Wait for packets currently being received to be done. 10938 * Does not block later packets from starting. 10939 */ 10940 void synchronize_net(void) 10941 { 10942 might_sleep(); 10943 if (rtnl_is_locked()) 10944 synchronize_rcu_expedited(); 10945 else 10946 synchronize_rcu(); 10947 } 10948 EXPORT_SYMBOL(synchronize_net); 10949 10950 /** 10951 * unregister_netdevice_queue - remove device from the kernel 10952 * @dev: device 10953 * @head: list 10954 * 10955 * This function shuts down a device interface and removes it 10956 * from the kernel tables. 10957 * If head not NULL, device is queued to be unregistered later. 10958 * 10959 * Callers must hold the rtnl semaphore. You may want 10960 * unregister_netdev() instead of this. 10961 */ 10962 10963 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) 10964 { 10965 ASSERT_RTNL(); 10966 10967 if (head) { 10968 list_move_tail(&dev->unreg_list, head); 10969 } else { 10970 LIST_HEAD(single); 10971 10972 list_add(&dev->unreg_list, &single); 10973 unregister_netdevice_many(&single); 10974 } 10975 } 10976 EXPORT_SYMBOL(unregister_netdevice_queue); 10977 10978 void unregister_netdevice_many_notify(struct list_head *head, 10979 u32 portid, const struct nlmsghdr *nlh) 10980 { 10981 struct net_device *dev, *tmp; 10982 LIST_HEAD(close_head); 10983 10984 BUG_ON(dev_boot_phase); 10985 ASSERT_RTNL(); 10986 10987 if (list_empty(head)) 10988 return; 10989 10990 list_for_each_entry_safe(dev, tmp, head, unreg_list) { 10991 /* Some devices call without registering 10992 * for initialization unwind. Remove those 10993 * devices and proceed with the remaining. 10994 */ 10995 if (dev->reg_state == NETREG_UNINITIALIZED) { 10996 pr_debug("unregister_netdevice: device %s/%p never was registered\n", 10997 dev->name, dev); 10998 10999 WARN_ON(1); 11000 list_del(&dev->unreg_list); 11001 continue; 11002 } 11003 dev->dismantle = true; 11004 BUG_ON(dev->reg_state != NETREG_REGISTERED); 11005 } 11006 11007 /* If device is running, close it first. */ 11008 list_for_each_entry(dev, head, unreg_list) 11009 list_add_tail(&dev->close_list, &close_head); 11010 dev_close_many(&close_head, true); 11011 11012 list_for_each_entry(dev, head, unreg_list) { 11013 /* And unlink it from device chain. */ 11014 write_lock(&dev_base_lock); 11015 unlist_netdevice(dev, false); 11016 dev->reg_state = NETREG_UNREGISTERING; 11017 write_unlock(&dev_base_lock); 11018 } 11019 flush_all_backlogs(); 11020 11021 synchronize_net(); 11022 11023 list_for_each_entry(dev, head, unreg_list) { 11024 struct sk_buff *skb = NULL; 11025 11026 /* Shutdown queueing discipline. */ 11027 dev_shutdown(dev); 11028 dev_tcx_uninstall(dev); 11029 dev_xdp_uninstall(dev); 11030 bpf_dev_bound_netdev_unregister(dev); 11031 11032 netdev_offload_xstats_disable_all(dev); 11033 11034 /* Notify protocols, that we are about to destroy 11035 * this device. They should clean all the things. 11036 */ 11037 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 11038 11039 if (!dev->rtnl_link_ops || 11040 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 11041 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0, 11042 GFP_KERNEL, NULL, 0, 11043 portid, nlh); 11044 11045 /* 11046 * Flush the unicast and multicast chains 11047 */ 11048 dev_uc_flush(dev); 11049 dev_mc_flush(dev); 11050 11051 netdev_name_node_alt_flush(dev); 11052 netdev_name_node_free(dev->name_node); 11053 11054 call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev); 11055 11056 if (dev->netdev_ops->ndo_uninit) 11057 dev->netdev_ops->ndo_uninit(dev); 11058 11059 if (skb) 11060 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL, portid, nlh); 11061 11062 /* Notifier chain MUST detach us all upper devices. */ 11063 WARN_ON(netdev_has_any_upper_dev(dev)); 11064 WARN_ON(netdev_has_any_lower_dev(dev)); 11065 11066 /* Remove entries from kobject tree */ 11067 netdev_unregister_kobject(dev); 11068 #ifdef CONFIG_XPS 11069 /* Remove XPS queueing entries */ 11070 netif_reset_xps_queues_gt(dev, 0); 11071 #endif 11072 } 11073 11074 synchronize_net(); 11075 11076 list_for_each_entry(dev, head, unreg_list) { 11077 netdev_put(dev, &dev->dev_registered_tracker); 11078 net_set_todo(dev); 11079 } 11080 11081 list_del(head); 11082 } 11083 11084 /** 11085 * unregister_netdevice_many - unregister many devices 11086 * @head: list of devices 11087 * 11088 * Note: As most callers use a stack allocated list_head, 11089 * we force a list_del() to make sure stack wont be corrupted later. 11090 */ 11091 void unregister_netdevice_many(struct list_head *head) 11092 { 11093 unregister_netdevice_many_notify(head, 0, NULL); 11094 } 11095 EXPORT_SYMBOL(unregister_netdevice_many); 11096 11097 /** 11098 * unregister_netdev - remove device from the kernel 11099 * @dev: device 11100 * 11101 * This function shuts down a device interface and removes it 11102 * from the kernel tables. 11103 * 11104 * This is just a wrapper for unregister_netdevice that takes 11105 * the rtnl semaphore. In general you want to use this and not 11106 * unregister_netdevice. 11107 */ 11108 void unregister_netdev(struct net_device *dev) 11109 { 11110 rtnl_lock(); 11111 unregister_netdevice(dev); 11112 rtnl_unlock(); 11113 } 11114 EXPORT_SYMBOL(unregister_netdev); 11115 11116 /** 11117 * __dev_change_net_namespace - move device to different nethost namespace 11118 * @dev: device 11119 * @net: network namespace 11120 * @pat: If not NULL name pattern to try if the current device name 11121 * is already taken in the destination network namespace. 11122 * @new_ifindex: If not zero, specifies device index in the target 11123 * namespace. 11124 * 11125 * This function shuts down a device interface and moves it 11126 * to a new network namespace. On success 0 is returned, on 11127 * a failure a netagive errno code is returned. 11128 * 11129 * Callers must hold the rtnl semaphore. 11130 */ 11131 11132 int __dev_change_net_namespace(struct net_device *dev, struct net *net, 11133 const char *pat, int new_ifindex) 11134 { 11135 struct netdev_name_node *name_node; 11136 struct net *net_old = dev_net(dev); 11137 char new_name[IFNAMSIZ] = {}; 11138 int err, new_nsid; 11139 11140 ASSERT_RTNL(); 11141 11142 /* Don't allow namespace local devices to be moved. */ 11143 err = -EINVAL; 11144 if (dev->features & NETIF_F_NETNS_LOCAL) 11145 goto out; 11146 11147 /* Ensure the device has been registrered */ 11148 if (dev->reg_state != NETREG_REGISTERED) 11149 goto out; 11150 11151 /* Get out if there is nothing todo */ 11152 err = 0; 11153 if (net_eq(net_old, net)) 11154 goto out; 11155 11156 /* Pick the destination device name, and ensure 11157 * we can use it in the destination network namespace. 11158 */ 11159 err = -EEXIST; 11160 if (netdev_name_in_use(net, dev->name)) { 11161 /* We get here if we can't use the current device name */ 11162 if (!pat) 11163 goto out; 11164 err = dev_prep_valid_name(net, dev, pat, new_name); 11165 if (err < 0) 11166 goto out; 11167 } 11168 /* Check that none of the altnames conflicts. */ 11169 err = -EEXIST; 11170 netdev_for_each_altname(dev, name_node) 11171 if (netdev_name_in_use(net, name_node->name)) 11172 goto out; 11173 11174 /* Check that new_ifindex isn't used yet. */ 11175 if (new_ifindex) { 11176 err = dev_index_reserve(net, new_ifindex); 11177 if (err < 0) 11178 goto out; 11179 } else { 11180 /* If there is an ifindex conflict assign a new one */ 11181 err = dev_index_reserve(net, dev->ifindex); 11182 if (err == -EBUSY) 11183 err = dev_index_reserve(net, 0); 11184 if (err < 0) 11185 goto out; 11186 new_ifindex = err; 11187 } 11188 11189 /* 11190 * And now a mini version of register_netdevice unregister_netdevice. 11191 */ 11192 11193 /* If device is running close it first. */ 11194 dev_close(dev); 11195 11196 /* And unlink it from device chain */ 11197 unlist_netdevice(dev, true); 11198 11199 synchronize_net(); 11200 11201 /* Shutdown queueing discipline. */ 11202 dev_shutdown(dev); 11203 11204 /* Notify protocols, that we are about to destroy 11205 * this device. They should clean all the things. 11206 * 11207 * Note that dev->reg_state stays at NETREG_REGISTERED. 11208 * This is wanted because this way 8021q and macvlan know 11209 * the device is just moving and can keep their slaves up. 11210 */ 11211 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 11212 rcu_barrier(); 11213 11214 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL); 11215 11216 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid, 11217 new_ifindex); 11218 11219 /* 11220 * Flush the unicast and multicast chains 11221 */ 11222 dev_uc_flush(dev); 11223 dev_mc_flush(dev); 11224 11225 /* Send a netdev-removed uevent to the old namespace */ 11226 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); 11227 netdev_adjacent_del_links(dev); 11228 11229 /* Move per-net netdevice notifiers that are following the netdevice */ 11230 move_netdevice_notifiers_dev_net(dev, net); 11231 11232 /* Actually switch the network namespace */ 11233 dev_net_set(dev, net); 11234 dev->ifindex = new_ifindex; 11235 11236 /* Send a netdev-add uevent to the new namespace */ 11237 kobject_uevent(&dev->dev.kobj, KOBJ_ADD); 11238 netdev_adjacent_add_links(dev); 11239 11240 if (new_name[0]) /* Rename the netdev to prepared name */ 11241 strscpy(dev->name, new_name, IFNAMSIZ); 11242 11243 /* Fixup kobjects */ 11244 err = device_rename(&dev->dev, dev->name); 11245 WARN_ON(err); 11246 11247 /* Adapt owner in case owning user namespace of target network 11248 * namespace is different from the original one. 11249 */ 11250 err = netdev_change_owner(dev, net_old, net); 11251 WARN_ON(err); 11252 11253 /* Add the device back in the hashes */ 11254 list_netdevice(dev); 11255 11256 /* Notify protocols, that a new device appeared. */ 11257 call_netdevice_notifiers(NETDEV_REGISTER, dev); 11258 11259 /* 11260 * Prevent userspace races by waiting until the network 11261 * device is fully setup before sending notifications. 11262 */ 11263 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL); 11264 11265 synchronize_net(); 11266 err = 0; 11267 out: 11268 return err; 11269 } 11270 EXPORT_SYMBOL_GPL(__dev_change_net_namespace); 11271 11272 static int dev_cpu_dead(unsigned int oldcpu) 11273 { 11274 struct sk_buff **list_skb; 11275 struct sk_buff *skb; 11276 unsigned int cpu; 11277 struct softnet_data *sd, *oldsd, *remsd = NULL; 11278 11279 local_irq_disable(); 11280 cpu = smp_processor_id(); 11281 sd = &per_cpu(softnet_data, cpu); 11282 oldsd = &per_cpu(softnet_data, oldcpu); 11283 11284 /* Find end of our completion_queue. */ 11285 list_skb = &sd->completion_queue; 11286 while (*list_skb) 11287 list_skb = &(*list_skb)->next; 11288 /* Append completion queue from offline CPU. */ 11289 *list_skb = oldsd->completion_queue; 11290 oldsd->completion_queue = NULL; 11291 11292 /* Append output queue from offline CPU. */ 11293 if (oldsd->output_queue) { 11294 *sd->output_queue_tailp = oldsd->output_queue; 11295 sd->output_queue_tailp = oldsd->output_queue_tailp; 11296 oldsd->output_queue = NULL; 11297 oldsd->output_queue_tailp = &oldsd->output_queue; 11298 } 11299 /* Append NAPI poll list from offline CPU, with one exception : 11300 * process_backlog() must be called by cpu owning percpu backlog. 11301 * We properly handle process_queue & input_pkt_queue later. 11302 */ 11303 while (!list_empty(&oldsd->poll_list)) { 11304 struct napi_struct *napi = list_first_entry(&oldsd->poll_list, 11305 struct napi_struct, 11306 poll_list); 11307 11308 list_del_init(&napi->poll_list); 11309 if (napi->poll == process_backlog) 11310 napi->state = 0; 11311 else 11312 ____napi_schedule(sd, napi); 11313 } 11314 11315 raise_softirq_irqoff(NET_TX_SOFTIRQ); 11316 local_irq_enable(); 11317 11318 #ifdef CONFIG_RPS 11319 remsd = oldsd->rps_ipi_list; 11320 oldsd->rps_ipi_list = NULL; 11321 #endif 11322 /* send out pending IPI's on offline CPU */ 11323 net_rps_send_ipi(remsd); 11324 11325 /* Process offline CPU's input_pkt_queue */ 11326 while ((skb = __skb_dequeue(&oldsd->process_queue))) { 11327 netif_rx(skb); 11328 input_queue_head_incr(oldsd); 11329 } 11330 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { 11331 netif_rx(skb); 11332 input_queue_head_incr(oldsd); 11333 } 11334 11335 return 0; 11336 } 11337 11338 /** 11339 * netdev_increment_features - increment feature set by one 11340 * @all: current feature set 11341 * @one: new feature set 11342 * @mask: mask feature set 11343 * 11344 * Computes a new feature set after adding a device with feature set 11345 * @one to the master device with current feature set @all. Will not 11346 * enable anything that is off in @mask. Returns the new feature set. 11347 */ 11348 netdev_features_t netdev_increment_features(netdev_features_t all, 11349 netdev_features_t one, netdev_features_t mask) 11350 { 11351 if (mask & NETIF_F_HW_CSUM) 11352 mask |= NETIF_F_CSUM_MASK; 11353 mask |= NETIF_F_VLAN_CHALLENGED; 11354 11355 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; 11356 all &= one | ~NETIF_F_ALL_FOR_ALL; 11357 11358 /* If one device supports hw checksumming, set for all. */ 11359 if (all & NETIF_F_HW_CSUM) 11360 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); 11361 11362 return all; 11363 } 11364 EXPORT_SYMBOL(netdev_increment_features); 11365 11366 static struct hlist_head * __net_init netdev_create_hash(void) 11367 { 11368 int i; 11369 struct hlist_head *hash; 11370 11371 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL); 11372 if (hash != NULL) 11373 for (i = 0; i < NETDEV_HASHENTRIES; i++) 11374 INIT_HLIST_HEAD(&hash[i]); 11375 11376 return hash; 11377 } 11378 11379 /* Initialize per network namespace state */ 11380 static int __net_init netdev_init(struct net *net) 11381 { 11382 BUILD_BUG_ON(GRO_HASH_BUCKETS > 11383 8 * sizeof_field(struct napi_struct, gro_bitmask)); 11384 11385 INIT_LIST_HEAD(&net->dev_base_head); 11386 11387 net->dev_name_head = netdev_create_hash(); 11388 if (net->dev_name_head == NULL) 11389 goto err_name; 11390 11391 net->dev_index_head = netdev_create_hash(); 11392 if (net->dev_index_head == NULL) 11393 goto err_idx; 11394 11395 xa_init_flags(&net->dev_by_index, XA_FLAGS_ALLOC1); 11396 11397 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain); 11398 11399 return 0; 11400 11401 err_idx: 11402 kfree(net->dev_name_head); 11403 err_name: 11404 return -ENOMEM; 11405 } 11406 11407 /** 11408 * netdev_drivername - network driver for the device 11409 * @dev: network device 11410 * 11411 * Determine network driver for device. 11412 */ 11413 const char *netdev_drivername(const struct net_device *dev) 11414 { 11415 const struct device_driver *driver; 11416 const struct device *parent; 11417 const char *empty = ""; 11418 11419 parent = dev->dev.parent; 11420 if (!parent) 11421 return empty; 11422 11423 driver = parent->driver; 11424 if (driver && driver->name) 11425 return driver->name; 11426 return empty; 11427 } 11428 11429 static void __netdev_printk(const char *level, const struct net_device *dev, 11430 struct va_format *vaf) 11431 { 11432 if (dev && dev->dev.parent) { 11433 dev_printk_emit(level[1] - '0', 11434 dev->dev.parent, 11435 "%s %s %s%s: %pV", 11436 dev_driver_string(dev->dev.parent), 11437 dev_name(dev->dev.parent), 11438 netdev_name(dev), netdev_reg_state(dev), 11439 vaf); 11440 } else if (dev) { 11441 printk("%s%s%s: %pV", 11442 level, netdev_name(dev), netdev_reg_state(dev), vaf); 11443 } else { 11444 printk("%s(NULL net_device): %pV", level, vaf); 11445 } 11446 } 11447 11448 void netdev_printk(const char *level, const struct net_device *dev, 11449 const char *format, ...) 11450 { 11451 struct va_format vaf; 11452 va_list args; 11453 11454 va_start(args, format); 11455 11456 vaf.fmt = format; 11457 vaf.va = &args; 11458 11459 __netdev_printk(level, dev, &vaf); 11460 11461 va_end(args); 11462 } 11463 EXPORT_SYMBOL(netdev_printk); 11464 11465 #define define_netdev_printk_level(func, level) \ 11466 void func(const struct net_device *dev, const char *fmt, ...) \ 11467 { \ 11468 struct va_format vaf; \ 11469 va_list args; \ 11470 \ 11471 va_start(args, fmt); \ 11472 \ 11473 vaf.fmt = fmt; \ 11474 vaf.va = &args; \ 11475 \ 11476 __netdev_printk(level, dev, &vaf); \ 11477 \ 11478 va_end(args); \ 11479 } \ 11480 EXPORT_SYMBOL(func); 11481 11482 define_netdev_printk_level(netdev_emerg, KERN_EMERG); 11483 define_netdev_printk_level(netdev_alert, KERN_ALERT); 11484 define_netdev_printk_level(netdev_crit, KERN_CRIT); 11485 define_netdev_printk_level(netdev_err, KERN_ERR); 11486 define_netdev_printk_level(netdev_warn, KERN_WARNING); 11487 define_netdev_printk_level(netdev_notice, KERN_NOTICE); 11488 define_netdev_printk_level(netdev_info, KERN_INFO); 11489 11490 static void __net_exit netdev_exit(struct net *net) 11491 { 11492 kfree(net->dev_name_head); 11493 kfree(net->dev_index_head); 11494 xa_destroy(&net->dev_by_index); 11495 if (net != &init_net) 11496 WARN_ON_ONCE(!list_empty(&net->dev_base_head)); 11497 } 11498 11499 static struct pernet_operations __net_initdata netdev_net_ops = { 11500 .init = netdev_init, 11501 .exit = netdev_exit, 11502 }; 11503 11504 static void __net_exit default_device_exit_net(struct net *net) 11505 { 11506 struct netdev_name_node *name_node, *tmp; 11507 struct net_device *dev, *aux; 11508 /* 11509 * Push all migratable network devices back to the 11510 * initial network namespace 11511 */ 11512 ASSERT_RTNL(); 11513 for_each_netdev_safe(net, dev, aux) { 11514 int err; 11515 char fb_name[IFNAMSIZ]; 11516 11517 /* Ignore unmoveable devices (i.e. loopback) */ 11518 if (dev->features & NETIF_F_NETNS_LOCAL) 11519 continue; 11520 11521 /* Leave virtual devices for the generic cleanup */ 11522 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund) 11523 continue; 11524 11525 /* Push remaining network devices to init_net */ 11526 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); 11527 if (netdev_name_in_use(&init_net, fb_name)) 11528 snprintf(fb_name, IFNAMSIZ, "dev%%d"); 11529 11530 netdev_for_each_altname_safe(dev, name_node, tmp) 11531 if (netdev_name_in_use(&init_net, name_node->name)) { 11532 netdev_name_node_del(name_node); 11533 synchronize_rcu(); 11534 __netdev_name_node_alt_destroy(name_node); 11535 } 11536 11537 err = dev_change_net_namespace(dev, &init_net, fb_name); 11538 if (err) { 11539 pr_emerg("%s: failed to move %s to init_net: %d\n", 11540 __func__, dev->name, err); 11541 BUG(); 11542 } 11543 } 11544 } 11545 11546 static void __net_exit default_device_exit_batch(struct list_head *net_list) 11547 { 11548 /* At exit all network devices most be removed from a network 11549 * namespace. Do this in the reverse order of registration. 11550 * Do this across as many network namespaces as possible to 11551 * improve batching efficiency. 11552 */ 11553 struct net_device *dev; 11554 struct net *net; 11555 LIST_HEAD(dev_kill_list); 11556 11557 rtnl_lock(); 11558 list_for_each_entry(net, net_list, exit_list) { 11559 default_device_exit_net(net); 11560 cond_resched(); 11561 } 11562 11563 list_for_each_entry(net, net_list, exit_list) { 11564 for_each_netdev_reverse(net, dev) { 11565 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) 11566 dev->rtnl_link_ops->dellink(dev, &dev_kill_list); 11567 else 11568 unregister_netdevice_queue(dev, &dev_kill_list); 11569 } 11570 } 11571 unregister_netdevice_many(&dev_kill_list); 11572 rtnl_unlock(); 11573 } 11574 11575 static struct pernet_operations __net_initdata default_device_ops = { 11576 .exit_batch = default_device_exit_batch, 11577 }; 11578 11579 /* 11580 * Initialize the DEV module. At boot time this walks the device list and 11581 * unhooks any devices that fail to initialise (normally hardware not 11582 * present) and leaves us with a valid list of present and active devices. 11583 * 11584 */ 11585 11586 /* 11587 * This is called single threaded during boot, so no need 11588 * to take the rtnl semaphore. 11589 */ 11590 static int __init net_dev_init(void) 11591 { 11592 int i, rc = -ENOMEM; 11593 11594 BUG_ON(!dev_boot_phase); 11595 11596 if (dev_proc_init()) 11597 goto out; 11598 11599 if (netdev_kobject_init()) 11600 goto out; 11601 11602 INIT_LIST_HEAD(&ptype_all); 11603 for (i = 0; i < PTYPE_HASH_SIZE; i++) 11604 INIT_LIST_HEAD(&ptype_base[i]); 11605 11606 if (register_pernet_subsys(&netdev_net_ops)) 11607 goto out; 11608 11609 /* 11610 * Initialise the packet receive queues. 11611 */ 11612 11613 for_each_possible_cpu(i) { 11614 struct work_struct *flush = per_cpu_ptr(&flush_works, i); 11615 struct softnet_data *sd = &per_cpu(softnet_data, i); 11616 11617 INIT_WORK(flush, flush_backlog); 11618 11619 skb_queue_head_init(&sd->input_pkt_queue); 11620 skb_queue_head_init(&sd->process_queue); 11621 #ifdef CONFIG_XFRM_OFFLOAD 11622 skb_queue_head_init(&sd->xfrm_backlog); 11623 #endif 11624 INIT_LIST_HEAD(&sd->poll_list); 11625 sd->output_queue_tailp = &sd->output_queue; 11626 #ifdef CONFIG_RPS 11627 INIT_CSD(&sd->csd, rps_trigger_softirq, sd); 11628 sd->cpu = i; 11629 #endif 11630 INIT_CSD(&sd->defer_csd, trigger_rx_softirq, sd); 11631 spin_lock_init(&sd->defer_lock); 11632 11633 init_gro_hash(&sd->backlog); 11634 sd->backlog.poll = process_backlog; 11635 sd->backlog.weight = weight_p; 11636 } 11637 11638 dev_boot_phase = 0; 11639 11640 /* The loopback device is special if any other network devices 11641 * is present in a network namespace the loopback device must 11642 * be present. Since we now dynamically allocate and free the 11643 * loopback device ensure this invariant is maintained by 11644 * keeping the loopback device as the first device on the 11645 * list of network devices. Ensuring the loopback devices 11646 * is the first device that appears and the last network device 11647 * that disappears. 11648 */ 11649 if (register_pernet_device(&loopback_net_ops)) 11650 goto out; 11651 11652 if (register_pernet_device(&default_device_ops)) 11653 goto out; 11654 11655 open_softirq(NET_TX_SOFTIRQ, net_tx_action); 11656 open_softirq(NET_RX_SOFTIRQ, net_rx_action); 11657 11658 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", 11659 NULL, dev_cpu_dead); 11660 WARN_ON(rc < 0); 11661 rc = 0; 11662 out: 11663 return rc; 11664 } 11665 11666 subsys_initcall(net_dev_init); 11667