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