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 #include <linux/phy_link_topology_core.h> 157 158 #include "dev.h" 159 #include "net-sysfs.h" 160 161 static DEFINE_SPINLOCK(ptype_lock); 162 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; 163 struct list_head ptype_all __read_mostly; /* Taps */ 164 165 static int netif_rx_internal(struct sk_buff *skb); 166 static int call_netdevice_notifiers_extack(unsigned long val, 167 struct net_device *dev, 168 struct netlink_ext_ack *extack); 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 tcf_set_drop_reason(skb, SKB_DROP_REASON_QDISC_DROP); 3761 3762 if (q->flags & TCQ_F_NOLOCK) { 3763 if (q->flags & TCQ_F_CAN_BYPASS && nolock_qdisc_is_empty(q) && 3764 qdisc_run_begin(q)) { 3765 /* Retest nolock_qdisc_is_empty() within the protection 3766 * of q->seqlock to protect from racing with requeuing. 3767 */ 3768 if (unlikely(!nolock_qdisc_is_empty(q))) { 3769 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3770 __qdisc_run(q); 3771 qdisc_run_end(q); 3772 3773 goto no_lock_out; 3774 } 3775 3776 qdisc_bstats_cpu_update(q, skb); 3777 if (sch_direct_xmit(skb, q, dev, txq, NULL, true) && 3778 !nolock_qdisc_is_empty(q)) 3779 __qdisc_run(q); 3780 3781 qdisc_run_end(q); 3782 return NET_XMIT_SUCCESS; 3783 } 3784 3785 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3786 qdisc_run(q); 3787 3788 no_lock_out: 3789 if (unlikely(to_free)) 3790 kfree_skb_list_reason(to_free, 3791 tcf_get_drop_reason(to_free)); 3792 return rc; 3793 } 3794 3795 /* 3796 * Heuristic to force contended enqueues to serialize on a 3797 * separate lock before trying to get qdisc main lock. 3798 * This permits qdisc->running owner to get the lock more 3799 * often and dequeue packets faster. 3800 * On PREEMPT_RT it is possible to preempt the qdisc owner during xmit 3801 * and then other tasks will only enqueue packets. The packets will be 3802 * sent after the qdisc owner is scheduled again. To prevent this 3803 * scenario the task always serialize on the lock. 3804 */ 3805 contended = qdisc_is_running(q) || IS_ENABLED(CONFIG_PREEMPT_RT); 3806 if (unlikely(contended)) 3807 spin_lock(&q->busylock); 3808 3809 spin_lock(root_lock); 3810 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { 3811 __qdisc_drop(skb, &to_free); 3812 rc = NET_XMIT_DROP; 3813 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && 3814 qdisc_run_begin(q)) { 3815 /* 3816 * This is a work-conserving queue; there are no old skbs 3817 * waiting to be sent out; and the qdisc is not running - 3818 * xmit the skb directly. 3819 */ 3820 3821 qdisc_bstats_update(q, skb); 3822 3823 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { 3824 if (unlikely(contended)) { 3825 spin_unlock(&q->busylock); 3826 contended = false; 3827 } 3828 __qdisc_run(q); 3829 } 3830 3831 qdisc_run_end(q); 3832 rc = NET_XMIT_SUCCESS; 3833 } else { 3834 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3835 if (qdisc_run_begin(q)) { 3836 if (unlikely(contended)) { 3837 spin_unlock(&q->busylock); 3838 contended = false; 3839 } 3840 __qdisc_run(q); 3841 qdisc_run_end(q); 3842 } 3843 } 3844 spin_unlock(root_lock); 3845 if (unlikely(to_free)) 3846 kfree_skb_list_reason(to_free, 3847 tcf_get_drop_reason(to_free)); 3848 if (unlikely(contended)) 3849 spin_unlock(&q->busylock); 3850 return rc; 3851 } 3852 3853 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) 3854 static void skb_update_prio(struct sk_buff *skb) 3855 { 3856 const struct netprio_map *map; 3857 const struct sock *sk; 3858 unsigned int prioidx; 3859 3860 if (skb->priority) 3861 return; 3862 map = rcu_dereference_bh(skb->dev->priomap); 3863 if (!map) 3864 return; 3865 sk = skb_to_full_sk(skb); 3866 if (!sk) 3867 return; 3868 3869 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data); 3870 3871 if (prioidx < map->priomap_len) 3872 skb->priority = map->priomap[prioidx]; 3873 } 3874 #else 3875 #define skb_update_prio(skb) 3876 #endif 3877 3878 /** 3879 * dev_loopback_xmit - loop back @skb 3880 * @net: network namespace this loopback is happening in 3881 * @sk: sk needed to be a netfilter okfn 3882 * @skb: buffer to transmit 3883 */ 3884 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) 3885 { 3886 skb_reset_mac_header(skb); 3887 __skb_pull(skb, skb_network_offset(skb)); 3888 skb->pkt_type = PACKET_LOOPBACK; 3889 if (skb->ip_summed == CHECKSUM_NONE) 3890 skb->ip_summed = CHECKSUM_UNNECESSARY; 3891 DEBUG_NET_WARN_ON_ONCE(!skb_dst(skb)); 3892 skb_dst_force(skb); 3893 netif_rx(skb); 3894 return 0; 3895 } 3896 EXPORT_SYMBOL(dev_loopback_xmit); 3897 3898 #ifdef CONFIG_NET_EGRESS 3899 static struct netdev_queue * 3900 netdev_tx_queue_mapping(struct net_device *dev, struct sk_buff *skb) 3901 { 3902 int qm = skb_get_queue_mapping(skb); 3903 3904 return netdev_get_tx_queue(dev, netdev_cap_txqueue(dev, qm)); 3905 } 3906 3907 static bool netdev_xmit_txqueue_skipped(void) 3908 { 3909 return __this_cpu_read(softnet_data.xmit.skip_txqueue); 3910 } 3911 3912 void netdev_xmit_skip_txqueue(bool skip) 3913 { 3914 __this_cpu_write(softnet_data.xmit.skip_txqueue, skip); 3915 } 3916 EXPORT_SYMBOL_GPL(netdev_xmit_skip_txqueue); 3917 #endif /* CONFIG_NET_EGRESS */ 3918 3919 #ifdef CONFIG_NET_XGRESS 3920 static int tc_run(struct tcx_entry *entry, struct sk_buff *skb, 3921 enum skb_drop_reason *drop_reason) 3922 { 3923 int ret = TC_ACT_UNSPEC; 3924 #ifdef CONFIG_NET_CLS_ACT 3925 struct mini_Qdisc *miniq = rcu_dereference_bh(entry->miniq); 3926 struct tcf_result res; 3927 3928 if (!miniq) 3929 return ret; 3930 3931 tc_skb_cb(skb)->mru = 0; 3932 tc_skb_cb(skb)->post_ct = false; 3933 tcf_set_drop_reason(skb, *drop_reason); 3934 3935 mini_qdisc_bstats_cpu_update(miniq, skb); 3936 ret = tcf_classify(skb, miniq->block, miniq->filter_list, &res, false); 3937 /* Only tcf related quirks below. */ 3938 switch (ret) { 3939 case TC_ACT_SHOT: 3940 *drop_reason = tcf_get_drop_reason(skb); 3941 mini_qdisc_qstats_cpu_drop(miniq); 3942 break; 3943 case TC_ACT_OK: 3944 case TC_ACT_RECLASSIFY: 3945 skb->tc_index = TC_H_MIN(res.classid); 3946 break; 3947 } 3948 #endif /* CONFIG_NET_CLS_ACT */ 3949 return ret; 3950 } 3951 3952 static DEFINE_STATIC_KEY_FALSE(tcx_needed_key); 3953 3954 void tcx_inc(void) 3955 { 3956 static_branch_inc(&tcx_needed_key); 3957 } 3958 3959 void tcx_dec(void) 3960 { 3961 static_branch_dec(&tcx_needed_key); 3962 } 3963 3964 static __always_inline enum tcx_action_base 3965 tcx_run(const struct bpf_mprog_entry *entry, struct sk_buff *skb, 3966 const bool needs_mac) 3967 { 3968 const struct bpf_mprog_fp *fp; 3969 const struct bpf_prog *prog; 3970 int ret = TCX_NEXT; 3971 3972 if (needs_mac) 3973 __skb_push(skb, skb->mac_len); 3974 bpf_mprog_foreach_prog(entry, fp, prog) { 3975 bpf_compute_data_pointers(skb); 3976 ret = bpf_prog_run(prog, skb); 3977 if (ret != TCX_NEXT) 3978 break; 3979 } 3980 if (needs_mac) 3981 __skb_pull(skb, skb->mac_len); 3982 return tcx_action_code(skb, ret); 3983 } 3984 3985 static __always_inline struct sk_buff * 3986 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 3987 struct net_device *orig_dev, bool *another) 3988 { 3989 struct bpf_mprog_entry *entry = rcu_dereference_bh(skb->dev->tcx_ingress); 3990 enum skb_drop_reason drop_reason = SKB_DROP_REASON_TC_INGRESS; 3991 int sch_ret; 3992 3993 if (!entry) 3994 return skb; 3995 if (*pt_prev) { 3996 *ret = deliver_skb(skb, *pt_prev, orig_dev); 3997 *pt_prev = NULL; 3998 } 3999 4000 qdisc_skb_cb(skb)->pkt_len = skb->len; 4001 tcx_set_ingress(skb, true); 4002 4003 if (static_branch_unlikely(&tcx_needed_key)) { 4004 sch_ret = tcx_run(entry, skb, true); 4005 if (sch_ret != TC_ACT_UNSPEC) 4006 goto ingress_verdict; 4007 } 4008 sch_ret = tc_run(tcx_entry(entry), skb, &drop_reason); 4009 ingress_verdict: 4010 switch (sch_ret) { 4011 case TC_ACT_REDIRECT: 4012 /* skb_mac_header check was done by BPF, so we can safely 4013 * push the L2 header back before redirecting to another 4014 * netdev. 4015 */ 4016 __skb_push(skb, skb->mac_len); 4017 if (skb_do_redirect(skb) == -EAGAIN) { 4018 __skb_pull(skb, skb->mac_len); 4019 *another = true; 4020 break; 4021 } 4022 *ret = NET_RX_SUCCESS; 4023 return NULL; 4024 case TC_ACT_SHOT: 4025 kfree_skb_reason(skb, drop_reason); 4026 *ret = NET_RX_DROP; 4027 return NULL; 4028 /* used by tc_run */ 4029 case TC_ACT_STOLEN: 4030 case TC_ACT_QUEUED: 4031 case TC_ACT_TRAP: 4032 consume_skb(skb); 4033 fallthrough; 4034 case TC_ACT_CONSUMED: 4035 *ret = NET_RX_SUCCESS; 4036 return NULL; 4037 } 4038 4039 return skb; 4040 } 4041 4042 static __always_inline struct sk_buff * 4043 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 4044 { 4045 struct bpf_mprog_entry *entry = rcu_dereference_bh(dev->tcx_egress); 4046 enum skb_drop_reason drop_reason = SKB_DROP_REASON_TC_EGRESS; 4047 int sch_ret; 4048 4049 if (!entry) 4050 return skb; 4051 4052 /* qdisc_skb_cb(skb)->pkt_len & tcx_set_ingress() was 4053 * already set by the caller. 4054 */ 4055 if (static_branch_unlikely(&tcx_needed_key)) { 4056 sch_ret = tcx_run(entry, skb, false); 4057 if (sch_ret != TC_ACT_UNSPEC) 4058 goto egress_verdict; 4059 } 4060 sch_ret = tc_run(tcx_entry(entry), skb, &drop_reason); 4061 egress_verdict: 4062 switch (sch_ret) { 4063 case TC_ACT_REDIRECT: 4064 /* No need to push/pop skb's mac_header here on egress! */ 4065 skb_do_redirect(skb); 4066 *ret = NET_XMIT_SUCCESS; 4067 return NULL; 4068 case TC_ACT_SHOT: 4069 kfree_skb_reason(skb, drop_reason); 4070 *ret = NET_XMIT_DROP; 4071 return NULL; 4072 /* used by tc_run */ 4073 case TC_ACT_STOLEN: 4074 case TC_ACT_QUEUED: 4075 case TC_ACT_TRAP: 4076 consume_skb(skb); 4077 fallthrough; 4078 case TC_ACT_CONSUMED: 4079 *ret = NET_XMIT_SUCCESS; 4080 return NULL; 4081 } 4082 4083 return skb; 4084 } 4085 #else 4086 static __always_inline struct sk_buff * 4087 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 4088 struct net_device *orig_dev, bool *another) 4089 { 4090 return skb; 4091 } 4092 4093 static __always_inline struct sk_buff * 4094 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 4095 { 4096 return skb; 4097 } 4098 #endif /* CONFIG_NET_XGRESS */ 4099 4100 #ifdef CONFIG_XPS 4101 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb, 4102 struct xps_dev_maps *dev_maps, unsigned int tci) 4103 { 4104 int tc = netdev_get_prio_tc_map(dev, skb->priority); 4105 struct xps_map *map; 4106 int queue_index = -1; 4107 4108 if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids) 4109 return queue_index; 4110 4111 tci *= dev_maps->num_tc; 4112 tci += tc; 4113 4114 map = rcu_dereference(dev_maps->attr_map[tci]); 4115 if (map) { 4116 if (map->len == 1) 4117 queue_index = map->queues[0]; 4118 else 4119 queue_index = map->queues[reciprocal_scale( 4120 skb_get_hash(skb), map->len)]; 4121 if (unlikely(queue_index >= dev->real_num_tx_queues)) 4122 queue_index = -1; 4123 } 4124 return queue_index; 4125 } 4126 #endif 4127 4128 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev, 4129 struct sk_buff *skb) 4130 { 4131 #ifdef CONFIG_XPS 4132 struct xps_dev_maps *dev_maps; 4133 struct sock *sk = skb->sk; 4134 int queue_index = -1; 4135 4136 if (!static_key_false(&xps_needed)) 4137 return -1; 4138 4139 rcu_read_lock(); 4140 if (!static_key_false(&xps_rxqs_needed)) 4141 goto get_cpus_map; 4142 4143 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]); 4144 if (dev_maps) { 4145 int tci = sk_rx_queue_get(sk); 4146 4147 if (tci >= 0) 4148 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4149 tci); 4150 } 4151 4152 get_cpus_map: 4153 if (queue_index < 0) { 4154 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]); 4155 if (dev_maps) { 4156 unsigned int tci = skb->sender_cpu - 1; 4157 4158 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4159 tci); 4160 } 4161 } 4162 rcu_read_unlock(); 4163 4164 return queue_index; 4165 #else 4166 return -1; 4167 #endif 4168 } 4169 4170 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb, 4171 struct net_device *sb_dev) 4172 { 4173 return 0; 4174 } 4175 EXPORT_SYMBOL(dev_pick_tx_zero); 4176 4177 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb, 4178 struct net_device *sb_dev) 4179 { 4180 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues; 4181 } 4182 EXPORT_SYMBOL(dev_pick_tx_cpu_id); 4183 4184 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb, 4185 struct net_device *sb_dev) 4186 { 4187 struct sock *sk = skb->sk; 4188 int queue_index = sk_tx_queue_get(sk); 4189 4190 sb_dev = sb_dev ? : dev; 4191 4192 if (queue_index < 0 || skb->ooo_okay || 4193 queue_index >= dev->real_num_tx_queues) { 4194 int new_index = get_xps_queue(dev, sb_dev, skb); 4195 4196 if (new_index < 0) 4197 new_index = skb_tx_hash(dev, sb_dev, skb); 4198 4199 if (queue_index != new_index && sk && 4200 sk_fullsock(sk) && 4201 rcu_access_pointer(sk->sk_dst_cache)) 4202 sk_tx_queue_set(sk, new_index); 4203 4204 queue_index = new_index; 4205 } 4206 4207 return queue_index; 4208 } 4209 EXPORT_SYMBOL(netdev_pick_tx); 4210 4211 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev, 4212 struct sk_buff *skb, 4213 struct net_device *sb_dev) 4214 { 4215 int queue_index = 0; 4216 4217 #ifdef CONFIG_XPS 4218 u32 sender_cpu = skb->sender_cpu - 1; 4219 4220 if (sender_cpu >= (u32)NR_CPUS) 4221 skb->sender_cpu = raw_smp_processor_id() + 1; 4222 #endif 4223 4224 if (dev->real_num_tx_queues != 1) { 4225 const struct net_device_ops *ops = dev->netdev_ops; 4226 4227 if (ops->ndo_select_queue) 4228 queue_index = ops->ndo_select_queue(dev, skb, sb_dev); 4229 else 4230 queue_index = netdev_pick_tx(dev, skb, sb_dev); 4231 4232 queue_index = netdev_cap_txqueue(dev, queue_index); 4233 } 4234 4235 skb_set_queue_mapping(skb, queue_index); 4236 return netdev_get_tx_queue(dev, queue_index); 4237 } 4238 4239 /** 4240 * __dev_queue_xmit() - transmit a buffer 4241 * @skb: buffer to transmit 4242 * @sb_dev: suboordinate device used for L2 forwarding offload 4243 * 4244 * Queue a buffer for transmission to a network device. The caller must 4245 * have set the device and priority and built the buffer before calling 4246 * this function. The function can be called from an interrupt. 4247 * 4248 * When calling this method, interrupts MUST be enabled. This is because 4249 * the BH enable code must have IRQs enabled so that it will not deadlock. 4250 * 4251 * Regardless of the return value, the skb is consumed, so it is currently 4252 * difficult to retry a send to this method. (You can bump the ref count 4253 * before sending to hold a reference for retry if you are careful.) 4254 * 4255 * Return: 4256 * * 0 - buffer successfully transmitted 4257 * * positive qdisc return code - NET_XMIT_DROP etc. 4258 * * negative errno - other errors 4259 */ 4260 int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev) 4261 { 4262 struct net_device *dev = skb->dev; 4263 struct netdev_queue *txq = NULL; 4264 struct Qdisc *q; 4265 int rc = -ENOMEM; 4266 bool again = false; 4267 4268 skb_reset_mac_header(skb); 4269 skb_assert_len(skb); 4270 4271 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) 4272 __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED); 4273 4274 /* Disable soft irqs for various locks below. Also 4275 * stops preemption for RCU. 4276 */ 4277 rcu_read_lock_bh(); 4278 4279 skb_update_prio(skb); 4280 4281 qdisc_pkt_len_init(skb); 4282 tcx_set_ingress(skb, false); 4283 #ifdef CONFIG_NET_EGRESS 4284 if (static_branch_unlikely(&egress_needed_key)) { 4285 if (nf_hook_egress_active()) { 4286 skb = nf_hook_egress(skb, &rc, dev); 4287 if (!skb) 4288 goto out; 4289 } 4290 4291 netdev_xmit_skip_txqueue(false); 4292 4293 nf_skip_egress(skb, true); 4294 skb = sch_handle_egress(skb, &rc, dev); 4295 if (!skb) 4296 goto out; 4297 nf_skip_egress(skb, false); 4298 4299 if (netdev_xmit_txqueue_skipped()) 4300 txq = netdev_tx_queue_mapping(dev, skb); 4301 } 4302 #endif 4303 /* If device/qdisc don't need skb->dst, release it right now while 4304 * its hot in this cpu cache. 4305 */ 4306 if (dev->priv_flags & IFF_XMIT_DST_RELEASE) 4307 skb_dst_drop(skb); 4308 else 4309 skb_dst_force(skb); 4310 4311 if (!txq) 4312 txq = netdev_core_pick_tx(dev, skb, sb_dev); 4313 4314 q = rcu_dereference_bh(txq->qdisc); 4315 4316 trace_net_dev_queue(skb); 4317 if (q->enqueue) { 4318 rc = __dev_xmit_skb(skb, q, dev, txq); 4319 goto out; 4320 } 4321 4322 /* The device has no queue. Common case for software devices: 4323 * loopback, all the sorts of tunnels... 4324 4325 * Really, it is unlikely that netif_tx_lock protection is necessary 4326 * here. (f.e. loopback and IP tunnels are clean ignoring statistics 4327 * counters.) 4328 * However, it is possible, that they rely on protection 4329 * made by us here. 4330 4331 * Check this and shot the lock. It is not prone from deadlocks. 4332 *Either shot noqueue qdisc, it is even simpler 8) 4333 */ 4334 if (dev->flags & IFF_UP) { 4335 int cpu = smp_processor_id(); /* ok because BHs are off */ 4336 4337 /* Other cpus might concurrently change txq->xmit_lock_owner 4338 * to -1 or to their cpu id, but not to our id. 4339 */ 4340 if (READ_ONCE(txq->xmit_lock_owner) != cpu) { 4341 if (dev_xmit_recursion()) 4342 goto recursion_alert; 4343 4344 skb = validate_xmit_skb(skb, dev, &again); 4345 if (!skb) 4346 goto out; 4347 4348 HARD_TX_LOCK(dev, txq, cpu); 4349 4350 if (!netif_xmit_stopped(txq)) { 4351 dev_xmit_recursion_inc(); 4352 skb = dev_hard_start_xmit(skb, dev, txq, &rc); 4353 dev_xmit_recursion_dec(); 4354 if (dev_xmit_complete(rc)) { 4355 HARD_TX_UNLOCK(dev, txq); 4356 goto out; 4357 } 4358 } 4359 HARD_TX_UNLOCK(dev, txq); 4360 net_crit_ratelimited("Virtual device %s asks to queue packet!\n", 4361 dev->name); 4362 } else { 4363 /* Recursion is detected! It is possible, 4364 * unfortunately 4365 */ 4366 recursion_alert: 4367 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", 4368 dev->name); 4369 } 4370 } 4371 4372 rc = -ENETDOWN; 4373 rcu_read_unlock_bh(); 4374 4375 dev_core_stats_tx_dropped_inc(dev); 4376 kfree_skb_list(skb); 4377 return rc; 4378 out: 4379 rcu_read_unlock_bh(); 4380 return rc; 4381 } 4382 EXPORT_SYMBOL(__dev_queue_xmit); 4383 4384 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id) 4385 { 4386 struct net_device *dev = skb->dev; 4387 struct sk_buff *orig_skb = skb; 4388 struct netdev_queue *txq; 4389 int ret = NETDEV_TX_BUSY; 4390 bool again = false; 4391 4392 if (unlikely(!netif_running(dev) || 4393 !netif_carrier_ok(dev))) 4394 goto drop; 4395 4396 skb = validate_xmit_skb_list(skb, dev, &again); 4397 if (skb != orig_skb) 4398 goto drop; 4399 4400 skb_set_queue_mapping(skb, queue_id); 4401 txq = skb_get_tx_queue(dev, skb); 4402 4403 local_bh_disable(); 4404 4405 dev_xmit_recursion_inc(); 4406 HARD_TX_LOCK(dev, txq, smp_processor_id()); 4407 if (!netif_xmit_frozen_or_drv_stopped(txq)) 4408 ret = netdev_start_xmit(skb, dev, txq, false); 4409 HARD_TX_UNLOCK(dev, txq); 4410 dev_xmit_recursion_dec(); 4411 4412 local_bh_enable(); 4413 return ret; 4414 drop: 4415 dev_core_stats_tx_dropped_inc(dev); 4416 kfree_skb_list(skb); 4417 return NET_XMIT_DROP; 4418 } 4419 EXPORT_SYMBOL(__dev_direct_xmit); 4420 4421 /************************************************************************* 4422 * Receiver routines 4423 *************************************************************************/ 4424 4425 int netdev_max_backlog __read_mostly = 1000; 4426 EXPORT_SYMBOL(netdev_max_backlog); 4427 4428 int netdev_tstamp_prequeue __read_mostly = 1; 4429 unsigned int sysctl_skb_defer_max __read_mostly = 64; 4430 int netdev_budget __read_mostly = 300; 4431 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */ 4432 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ; 4433 int weight_p __read_mostly = 64; /* old backlog weight */ 4434 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ 4435 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ 4436 int dev_rx_weight __read_mostly = 64; 4437 int dev_tx_weight __read_mostly = 64; 4438 4439 /* Called with irq disabled */ 4440 static inline void ____napi_schedule(struct softnet_data *sd, 4441 struct napi_struct *napi) 4442 { 4443 struct task_struct *thread; 4444 4445 lockdep_assert_irqs_disabled(); 4446 4447 if (test_bit(NAPI_STATE_THREADED, &napi->state)) { 4448 /* Paired with smp_mb__before_atomic() in 4449 * napi_enable()/dev_set_threaded(). 4450 * Use READ_ONCE() to guarantee a complete 4451 * read on napi->thread. Only call 4452 * wake_up_process() when it's not NULL. 4453 */ 4454 thread = READ_ONCE(napi->thread); 4455 if (thread) { 4456 /* Avoid doing set_bit() if the thread is in 4457 * INTERRUPTIBLE state, cause napi_thread_wait() 4458 * makes sure to proceed with napi polling 4459 * if the thread is explicitly woken from here. 4460 */ 4461 if (READ_ONCE(thread->__state) != TASK_INTERRUPTIBLE) 4462 set_bit(NAPI_STATE_SCHED_THREADED, &napi->state); 4463 wake_up_process(thread); 4464 return; 4465 } 4466 } 4467 4468 list_add_tail(&napi->poll_list, &sd->poll_list); 4469 WRITE_ONCE(napi->list_owner, smp_processor_id()); 4470 /* If not called from net_rx_action() 4471 * we have to raise NET_RX_SOFTIRQ. 4472 */ 4473 if (!sd->in_net_rx_action) 4474 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4475 } 4476 4477 #ifdef CONFIG_RPS 4478 4479 /* One global table that all flow-based protocols share. */ 4480 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; 4481 EXPORT_SYMBOL(rps_sock_flow_table); 4482 u32 rps_cpu_mask __read_mostly; 4483 EXPORT_SYMBOL(rps_cpu_mask); 4484 4485 struct static_key_false rps_needed __read_mostly; 4486 EXPORT_SYMBOL(rps_needed); 4487 struct static_key_false rfs_needed __read_mostly; 4488 EXPORT_SYMBOL(rfs_needed); 4489 4490 static struct rps_dev_flow * 4491 set_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4492 struct rps_dev_flow *rflow, u16 next_cpu) 4493 { 4494 if (next_cpu < nr_cpu_ids) { 4495 #ifdef CONFIG_RFS_ACCEL 4496 struct netdev_rx_queue *rxqueue; 4497 struct rps_dev_flow_table *flow_table; 4498 struct rps_dev_flow *old_rflow; 4499 u32 flow_id; 4500 u16 rxq_index; 4501 int rc; 4502 4503 /* Should we steer this flow to a different hardware queue? */ 4504 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || 4505 !(dev->features & NETIF_F_NTUPLE)) 4506 goto out; 4507 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); 4508 if (rxq_index == skb_get_rx_queue(skb)) 4509 goto out; 4510 4511 rxqueue = dev->_rx + rxq_index; 4512 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4513 if (!flow_table) 4514 goto out; 4515 flow_id = skb_get_hash(skb) & flow_table->mask; 4516 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, 4517 rxq_index, flow_id); 4518 if (rc < 0) 4519 goto out; 4520 old_rflow = rflow; 4521 rflow = &flow_table->flows[flow_id]; 4522 rflow->filter = rc; 4523 if (old_rflow->filter == rflow->filter) 4524 old_rflow->filter = RPS_NO_FILTER; 4525 out: 4526 #endif 4527 rflow->last_qtail = 4528 per_cpu(softnet_data, next_cpu).input_queue_head; 4529 } 4530 4531 rflow->cpu = next_cpu; 4532 return rflow; 4533 } 4534 4535 /* 4536 * get_rps_cpu is called from netif_receive_skb and returns the target 4537 * CPU from the RPS map of the receiving queue for a given skb. 4538 * rcu_read_lock must be held on entry. 4539 */ 4540 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4541 struct rps_dev_flow **rflowp) 4542 { 4543 const struct rps_sock_flow_table *sock_flow_table; 4544 struct netdev_rx_queue *rxqueue = dev->_rx; 4545 struct rps_dev_flow_table *flow_table; 4546 struct rps_map *map; 4547 int cpu = -1; 4548 u32 tcpu; 4549 u32 hash; 4550 4551 if (skb_rx_queue_recorded(skb)) { 4552 u16 index = skb_get_rx_queue(skb); 4553 4554 if (unlikely(index >= dev->real_num_rx_queues)) { 4555 WARN_ONCE(dev->real_num_rx_queues > 1, 4556 "%s received packet on queue %u, but number " 4557 "of RX queues is %u\n", 4558 dev->name, index, dev->real_num_rx_queues); 4559 goto done; 4560 } 4561 rxqueue += index; 4562 } 4563 4564 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ 4565 4566 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4567 map = rcu_dereference(rxqueue->rps_map); 4568 if (!flow_table && !map) 4569 goto done; 4570 4571 skb_reset_network_header(skb); 4572 hash = skb_get_hash(skb); 4573 if (!hash) 4574 goto done; 4575 4576 sock_flow_table = rcu_dereference(rps_sock_flow_table); 4577 if (flow_table && sock_flow_table) { 4578 struct rps_dev_flow *rflow; 4579 u32 next_cpu; 4580 u32 ident; 4581 4582 /* First check into global flow table if there is a match. 4583 * This READ_ONCE() pairs with WRITE_ONCE() from rps_record_sock_flow(). 4584 */ 4585 ident = READ_ONCE(sock_flow_table->ents[hash & sock_flow_table->mask]); 4586 if ((ident ^ hash) & ~rps_cpu_mask) 4587 goto try_rps; 4588 4589 next_cpu = ident & rps_cpu_mask; 4590 4591 /* OK, now we know there is a match, 4592 * we can look at the local (per receive queue) flow table 4593 */ 4594 rflow = &flow_table->flows[hash & flow_table->mask]; 4595 tcpu = rflow->cpu; 4596 4597 /* 4598 * If the desired CPU (where last recvmsg was done) is 4599 * different from current CPU (one in the rx-queue flow 4600 * table entry), switch if one of the following holds: 4601 * - Current CPU is unset (>= nr_cpu_ids). 4602 * - Current CPU is offline. 4603 * - The current CPU's queue tail has advanced beyond the 4604 * last packet that was enqueued using this table entry. 4605 * This guarantees that all previous packets for the flow 4606 * have been dequeued, thus preserving in order delivery. 4607 */ 4608 if (unlikely(tcpu != next_cpu) && 4609 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || 4610 ((int)(per_cpu(softnet_data, tcpu).input_queue_head - 4611 rflow->last_qtail)) >= 0)) { 4612 tcpu = next_cpu; 4613 rflow = set_rps_cpu(dev, skb, rflow, next_cpu); 4614 } 4615 4616 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { 4617 *rflowp = rflow; 4618 cpu = tcpu; 4619 goto done; 4620 } 4621 } 4622 4623 try_rps: 4624 4625 if (map) { 4626 tcpu = map->cpus[reciprocal_scale(hash, map->len)]; 4627 if (cpu_online(tcpu)) { 4628 cpu = tcpu; 4629 goto done; 4630 } 4631 } 4632 4633 done: 4634 return cpu; 4635 } 4636 4637 #ifdef CONFIG_RFS_ACCEL 4638 4639 /** 4640 * rps_may_expire_flow - check whether an RFS hardware filter may be removed 4641 * @dev: Device on which the filter was set 4642 * @rxq_index: RX queue index 4643 * @flow_id: Flow ID passed to ndo_rx_flow_steer() 4644 * @filter_id: Filter ID returned by ndo_rx_flow_steer() 4645 * 4646 * Drivers that implement ndo_rx_flow_steer() should periodically call 4647 * this function for each installed filter and remove the filters for 4648 * which it returns %true. 4649 */ 4650 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, 4651 u32 flow_id, u16 filter_id) 4652 { 4653 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; 4654 struct rps_dev_flow_table *flow_table; 4655 struct rps_dev_flow *rflow; 4656 bool expire = true; 4657 unsigned int cpu; 4658 4659 rcu_read_lock(); 4660 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4661 if (flow_table && flow_id <= flow_table->mask) { 4662 rflow = &flow_table->flows[flow_id]; 4663 cpu = READ_ONCE(rflow->cpu); 4664 if (rflow->filter == filter_id && cpu < nr_cpu_ids && 4665 ((int)(per_cpu(softnet_data, cpu).input_queue_head - 4666 rflow->last_qtail) < 4667 (int)(10 * flow_table->mask))) 4668 expire = false; 4669 } 4670 rcu_read_unlock(); 4671 return expire; 4672 } 4673 EXPORT_SYMBOL(rps_may_expire_flow); 4674 4675 #endif /* CONFIG_RFS_ACCEL */ 4676 4677 /* Called from hardirq (IPI) context */ 4678 static void rps_trigger_softirq(void *data) 4679 { 4680 struct softnet_data *sd = data; 4681 4682 ____napi_schedule(sd, &sd->backlog); 4683 sd->received_rps++; 4684 } 4685 4686 #endif /* CONFIG_RPS */ 4687 4688 /* Called from hardirq (IPI) context */ 4689 static void trigger_rx_softirq(void *data) 4690 { 4691 struct softnet_data *sd = data; 4692 4693 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4694 smp_store_release(&sd->defer_ipi_scheduled, 0); 4695 } 4696 4697 /* 4698 * After we queued a packet into sd->input_pkt_queue, 4699 * we need to make sure this queue is serviced soon. 4700 * 4701 * - If this is another cpu queue, link it to our rps_ipi_list, 4702 * and make sure we will process rps_ipi_list from net_rx_action(). 4703 * 4704 * - If this is our own queue, NAPI schedule our backlog. 4705 * Note that this also raises NET_RX_SOFTIRQ. 4706 */ 4707 static void napi_schedule_rps(struct softnet_data *sd) 4708 { 4709 struct softnet_data *mysd = this_cpu_ptr(&softnet_data); 4710 4711 #ifdef CONFIG_RPS 4712 if (sd != mysd) { 4713 sd->rps_ipi_next = mysd->rps_ipi_list; 4714 mysd->rps_ipi_list = sd; 4715 4716 /* If not called from net_rx_action() or napi_threaded_poll() 4717 * we have to raise NET_RX_SOFTIRQ. 4718 */ 4719 if (!mysd->in_net_rx_action && !mysd->in_napi_threaded_poll) 4720 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4721 return; 4722 } 4723 #endif /* CONFIG_RPS */ 4724 __napi_schedule_irqoff(&mysd->backlog); 4725 } 4726 4727 #ifdef CONFIG_NET_FLOW_LIMIT 4728 int netdev_flow_limit_table_len __read_mostly = (1 << 12); 4729 #endif 4730 4731 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) 4732 { 4733 #ifdef CONFIG_NET_FLOW_LIMIT 4734 struct sd_flow_limit *fl; 4735 struct softnet_data *sd; 4736 unsigned int old_flow, new_flow; 4737 4738 if (qlen < (READ_ONCE(netdev_max_backlog) >> 1)) 4739 return false; 4740 4741 sd = this_cpu_ptr(&softnet_data); 4742 4743 rcu_read_lock(); 4744 fl = rcu_dereference(sd->flow_limit); 4745 if (fl) { 4746 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); 4747 old_flow = fl->history[fl->history_head]; 4748 fl->history[fl->history_head] = new_flow; 4749 4750 fl->history_head++; 4751 fl->history_head &= FLOW_LIMIT_HISTORY - 1; 4752 4753 if (likely(fl->buckets[old_flow])) 4754 fl->buckets[old_flow]--; 4755 4756 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { 4757 fl->count++; 4758 rcu_read_unlock(); 4759 return true; 4760 } 4761 } 4762 rcu_read_unlock(); 4763 #endif 4764 return false; 4765 } 4766 4767 /* 4768 * enqueue_to_backlog is called to queue an skb to a per CPU backlog 4769 * queue (may be a remote CPU queue). 4770 */ 4771 static int enqueue_to_backlog(struct sk_buff *skb, int cpu, 4772 unsigned int *qtail) 4773 { 4774 enum skb_drop_reason reason; 4775 struct softnet_data *sd; 4776 unsigned long flags; 4777 unsigned int qlen; 4778 4779 reason = SKB_DROP_REASON_NOT_SPECIFIED; 4780 sd = &per_cpu(softnet_data, cpu); 4781 4782 rps_lock_irqsave(sd, &flags); 4783 if (!netif_running(skb->dev)) 4784 goto drop; 4785 qlen = skb_queue_len(&sd->input_pkt_queue); 4786 if (qlen <= READ_ONCE(netdev_max_backlog) && !skb_flow_limit(skb, qlen)) { 4787 if (qlen) { 4788 enqueue: 4789 __skb_queue_tail(&sd->input_pkt_queue, skb); 4790 input_queue_tail_incr_save(sd, qtail); 4791 rps_unlock_irq_restore(sd, &flags); 4792 return NET_RX_SUCCESS; 4793 } 4794 4795 /* Schedule NAPI for backlog device 4796 * We can use non atomic operation since we own the queue lock 4797 */ 4798 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) 4799 napi_schedule_rps(sd); 4800 goto enqueue; 4801 } 4802 reason = SKB_DROP_REASON_CPU_BACKLOG; 4803 4804 drop: 4805 sd->dropped++; 4806 rps_unlock_irq_restore(sd, &flags); 4807 4808 dev_core_stats_rx_dropped_inc(skb->dev); 4809 kfree_skb_reason(skb, reason); 4810 return NET_RX_DROP; 4811 } 4812 4813 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb) 4814 { 4815 struct net_device *dev = skb->dev; 4816 struct netdev_rx_queue *rxqueue; 4817 4818 rxqueue = dev->_rx; 4819 4820 if (skb_rx_queue_recorded(skb)) { 4821 u16 index = skb_get_rx_queue(skb); 4822 4823 if (unlikely(index >= dev->real_num_rx_queues)) { 4824 WARN_ONCE(dev->real_num_rx_queues > 1, 4825 "%s received packet on queue %u, but number " 4826 "of RX queues is %u\n", 4827 dev->name, index, dev->real_num_rx_queues); 4828 4829 return rxqueue; /* Return first rxqueue */ 4830 } 4831 rxqueue += index; 4832 } 4833 return rxqueue; 4834 } 4835 4836 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp, 4837 struct bpf_prog *xdp_prog) 4838 { 4839 void *orig_data, *orig_data_end, *hard_start; 4840 struct netdev_rx_queue *rxqueue; 4841 bool orig_bcast, orig_host; 4842 u32 mac_len, frame_sz; 4843 __be16 orig_eth_type; 4844 struct ethhdr *eth; 4845 u32 metalen, act; 4846 int off; 4847 4848 /* The XDP program wants to see the packet starting at the MAC 4849 * header. 4850 */ 4851 mac_len = skb->data - skb_mac_header(skb); 4852 hard_start = skb->data - skb_headroom(skb); 4853 4854 /* SKB "head" area always have tailroom for skb_shared_info */ 4855 frame_sz = (void *)skb_end_pointer(skb) - hard_start; 4856 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 4857 4858 rxqueue = netif_get_rxqueue(skb); 4859 xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq); 4860 xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len, 4861 skb_headlen(skb) + mac_len, true); 4862 4863 orig_data_end = xdp->data_end; 4864 orig_data = xdp->data; 4865 eth = (struct ethhdr *)xdp->data; 4866 orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr); 4867 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest); 4868 orig_eth_type = eth->h_proto; 4869 4870 act = bpf_prog_run_xdp(xdp_prog, xdp); 4871 4872 /* check if bpf_xdp_adjust_head was used */ 4873 off = xdp->data - orig_data; 4874 if (off) { 4875 if (off > 0) 4876 __skb_pull(skb, off); 4877 else if (off < 0) 4878 __skb_push(skb, -off); 4879 4880 skb->mac_header += off; 4881 skb_reset_network_header(skb); 4882 } 4883 4884 /* check if bpf_xdp_adjust_tail was used */ 4885 off = xdp->data_end - orig_data_end; 4886 if (off != 0) { 4887 skb_set_tail_pointer(skb, xdp->data_end - xdp->data); 4888 skb->len += off; /* positive on grow, negative on shrink */ 4889 } 4890 4891 /* check if XDP changed eth hdr such SKB needs update */ 4892 eth = (struct ethhdr *)xdp->data; 4893 if ((orig_eth_type != eth->h_proto) || 4894 (orig_host != ether_addr_equal_64bits(eth->h_dest, 4895 skb->dev->dev_addr)) || 4896 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) { 4897 __skb_push(skb, ETH_HLEN); 4898 skb->pkt_type = PACKET_HOST; 4899 skb->protocol = eth_type_trans(skb, skb->dev); 4900 } 4901 4902 /* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull 4903 * before calling us again on redirect path. We do not call do_redirect 4904 * as we leave that up to the caller. 4905 * 4906 * Caller is responsible for managing lifetime of skb (i.e. calling 4907 * kfree_skb in response to actions it cannot handle/XDP_DROP). 4908 */ 4909 switch (act) { 4910 case XDP_REDIRECT: 4911 case XDP_TX: 4912 __skb_push(skb, mac_len); 4913 break; 4914 case XDP_PASS: 4915 metalen = xdp->data - xdp->data_meta; 4916 if (metalen) 4917 skb_metadata_set(skb, metalen); 4918 break; 4919 } 4920 4921 return act; 4922 } 4923 4924 static u32 netif_receive_generic_xdp(struct sk_buff *skb, 4925 struct xdp_buff *xdp, 4926 struct bpf_prog *xdp_prog) 4927 { 4928 u32 act = XDP_DROP; 4929 4930 /* Reinjected packets coming from act_mirred or similar should 4931 * not get XDP generic processing. 4932 */ 4933 if (skb_is_redirected(skb)) 4934 return XDP_PASS; 4935 4936 /* XDP packets must be linear and must have sufficient headroom 4937 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also 4938 * native XDP provides, thus we need to do it here as well. 4939 */ 4940 if (skb_cloned(skb) || skb_is_nonlinear(skb) || 4941 skb_headroom(skb) < XDP_PACKET_HEADROOM) { 4942 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb); 4943 int troom = skb->tail + skb->data_len - skb->end; 4944 4945 /* In case we have to go down the path and also linearize, 4946 * then lets do the pskb_expand_head() work just once here. 4947 */ 4948 if (pskb_expand_head(skb, 4949 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0, 4950 troom > 0 ? troom + 128 : 0, GFP_ATOMIC)) 4951 goto do_drop; 4952 if (skb_linearize(skb)) 4953 goto do_drop; 4954 } 4955 4956 act = bpf_prog_run_generic_xdp(skb, xdp, xdp_prog); 4957 switch (act) { 4958 case XDP_REDIRECT: 4959 case XDP_TX: 4960 case XDP_PASS: 4961 break; 4962 default: 4963 bpf_warn_invalid_xdp_action(skb->dev, xdp_prog, act); 4964 fallthrough; 4965 case XDP_ABORTED: 4966 trace_xdp_exception(skb->dev, xdp_prog, act); 4967 fallthrough; 4968 case XDP_DROP: 4969 do_drop: 4970 kfree_skb(skb); 4971 break; 4972 } 4973 4974 return act; 4975 } 4976 4977 /* When doing generic XDP we have to bypass the qdisc layer and the 4978 * network taps in order to match in-driver-XDP behavior. This also means 4979 * that XDP packets are able to starve other packets going through a qdisc, 4980 * and DDOS attacks will be more effective. In-driver-XDP use dedicated TX 4981 * queues, so they do not have this starvation issue. 4982 */ 4983 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog) 4984 { 4985 struct net_device *dev = skb->dev; 4986 struct netdev_queue *txq; 4987 bool free_skb = true; 4988 int cpu, rc; 4989 4990 txq = netdev_core_pick_tx(dev, skb, NULL); 4991 cpu = smp_processor_id(); 4992 HARD_TX_LOCK(dev, txq, cpu); 4993 if (!netif_xmit_frozen_or_drv_stopped(txq)) { 4994 rc = netdev_start_xmit(skb, dev, txq, 0); 4995 if (dev_xmit_complete(rc)) 4996 free_skb = false; 4997 } 4998 HARD_TX_UNLOCK(dev, txq); 4999 if (free_skb) { 5000 trace_xdp_exception(dev, xdp_prog, XDP_TX); 5001 dev_core_stats_tx_dropped_inc(dev); 5002 kfree_skb(skb); 5003 } 5004 } 5005 5006 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key); 5007 5008 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb) 5009 { 5010 if (xdp_prog) { 5011 struct xdp_buff xdp; 5012 u32 act; 5013 int err; 5014 5015 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog); 5016 if (act != XDP_PASS) { 5017 switch (act) { 5018 case XDP_REDIRECT: 5019 err = xdp_do_generic_redirect(skb->dev, skb, 5020 &xdp, xdp_prog); 5021 if (err) 5022 goto out_redir; 5023 break; 5024 case XDP_TX: 5025 generic_xdp_tx(skb, xdp_prog); 5026 break; 5027 } 5028 return XDP_DROP; 5029 } 5030 } 5031 return XDP_PASS; 5032 out_redir: 5033 kfree_skb_reason(skb, SKB_DROP_REASON_XDP); 5034 return XDP_DROP; 5035 } 5036 EXPORT_SYMBOL_GPL(do_xdp_generic); 5037 5038 static int netif_rx_internal(struct sk_buff *skb) 5039 { 5040 int ret; 5041 5042 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb); 5043 5044 trace_netif_rx(skb); 5045 5046 #ifdef CONFIG_RPS 5047 if (static_branch_unlikely(&rps_needed)) { 5048 struct rps_dev_flow voidflow, *rflow = &voidflow; 5049 int cpu; 5050 5051 rcu_read_lock(); 5052 5053 cpu = get_rps_cpu(skb->dev, skb, &rflow); 5054 if (cpu < 0) 5055 cpu = smp_processor_id(); 5056 5057 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5058 5059 rcu_read_unlock(); 5060 } else 5061 #endif 5062 { 5063 unsigned int qtail; 5064 5065 ret = enqueue_to_backlog(skb, smp_processor_id(), &qtail); 5066 } 5067 return ret; 5068 } 5069 5070 /** 5071 * __netif_rx - Slightly optimized version of netif_rx 5072 * @skb: buffer to post 5073 * 5074 * This behaves as netif_rx except that it does not disable bottom halves. 5075 * As a result this function may only be invoked from the interrupt context 5076 * (either hard or soft interrupt). 5077 */ 5078 int __netif_rx(struct sk_buff *skb) 5079 { 5080 int ret; 5081 5082 lockdep_assert_once(hardirq_count() | softirq_count()); 5083 5084 trace_netif_rx_entry(skb); 5085 ret = netif_rx_internal(skb); 5086 trace_netif_rx_exit(ret); 5087 return ret; 5088 } 5089 EXPORT_SYMBOL(__netif_rx); 5090 5091 /** 5092 * netif_rx - post buffer to the network code 5093 * @skb: buffer to post 5094 * 5095 * This function receives a packet from a device driver and queues it for 5096 * the upper (protocol) levels to process via the backlog NAPI device. It 5097 * always succeeds. The buffer may be dropped during processing for 5098 * congestion control or by the protocol layers. 5099 * The network buffer is passed via the backlog NAPI device. Modern NIC 5100 * driver should use NAPI and GRO. 5101 * This function can used from interrupt and from process context. The 5102 * caller from process context must not disable interrupts before invoking 5103 * this function. 5104 * 5105 * return values: 5106 * NET_RX_SUCCESS (no congestion) 5107 * NET_RX_DROP (packet was dropped) 5108 * 5109 */ 5110 int netif_rx(struct sk_buff *skb) 5111 { 5112 bool need_bh_off = !(hardirq_count() | softirq_count()); 5113 int ret; 5114 5115 if (need_bh_off) 5116 local_bh_disable(); 5117 trace_netif_rx_entry(skb); 5118 ret = netif_rx_internal(skb); 5119 trace_netif_rx_exit(ret); 5120 if (need_bh_off) 5121 local_bh_enable(); 5122 return ret; 5123 } 5124 EXPORT_SYMBOL(netif_rx); 5125 5126 static __latent_entropy void net_tx_action(struct softirq_action *h) 5127 { 5128 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 5129 5130 if (sd->completion_queue) { 5131 struct sk_buff *clist; 5132 5133 local_irq_disable(); 5134 clist = sd->completion_queue; 5135 sd->completion_queue = NULL; 5136 local_irq_enable(); 5137 5138 while (clist) { 5139 struct sk_buff *skb = clist; 5140 5141 clist = clist->next; 5142 5143 WARN_ON(refcount_read(&skb->users)); 5144 if (likely(get_kfree_skb_cb(skb)->reason == SKB_CONSUMED)) 5145 trace_consume_skb(skb, net_tx_action); 5146 else 5147 trace_kfree_skb(skb, net_tx_action, 5148 get_kfree_skb_cb(skb)->reason); 5149 5150 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) 5151 __kfree_skb(skb); 5152 else 5153 __napi_kfree_skb(skb, 5154 get_kfree_skb_cb(skb)->reason); 5155 } 5156 } 5157 5158 if (sd->output_queue) { 5159 struct Qdisc *head; 5160 5161 local_irq_disable(); 5162 head = sd->output_queue; 5163 sd->output_queue = NULL; 5164 sd->output_queue_tailp = &sd->output_queue; 5165 local_irq_enable(); 5166 5167 rcu_read_lock(); 5168 5169 while (head) { 5170 struct Qdisc *q = head; 5171 spinlock_t *root_lock = NULL; 5172 5173 head = head->next_sched; 5174 5175 /* We need to make sure head->next_sched is read 5176 * before clearing __QDISC_STATE_SCHED 5177 */ 5178 smp_mb__before_atomic(); 5179 5180 if (!(q->flags & TCQ_F_NOLOCK)) { 5181 root_lock = qdisc_lock(q); 5182 spin_lock(root_lock); 5183 } else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, 5184 &q->state))) { 5185 /* There is a synchronize_net() between 5186 * STATE_DEACTIVATED flag being set and 5187 * qdisc_reset()/some_qdisc_is_busy() in 5188 * dev_deactivate(), so we can safely bail out 5189 * early here to avoid data race between 5190 * qdisc_deactivate() and some_qdisc_is_busy() 5191 * for lockless qdisc. 5192 */ 5193 clear_bit(__QDISC_STATE_SCHED, &q->state); 5194 continue; 5195 } 5196 5197 clear_bit(__QDISC_STATE_SCHED, &q->state); 5198 qdisc_run(q); 5199 if (root_lock) 5200 spin_unlock(root_lock); 5201 } 5202 5203 rcu_read_unlock(); 5204 } 5205 5206 xfrm_dev_backlog(sd); 5207 } 5208 5209 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) 5210 /* This hook is defined here for ATM LANE */ 5211 int (*br_fdb_test_addr_hook)(struct net_device *dev, 5212 unsigned char *addr) __read_mostly; 5213 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); 5214 #endif 5215 5216 /** 5217 * netdev_is_rx_handler_busy - check if receive handler is registered 5218 * @dev: device to check 5219 * 5220 * Check if a receive handler is already registered for a given device. 5221 * Return true if there one. 5222 * 5223 * The caller must hold the rtnl_mutex. 5224 */ 5225 bool netdev_is_rx_handler_busy(struct net_device *dev) 5226 { 5227 ASSERT_RTNL(); 5228 return dev && rtnl_dereference(dev->rx_handler); 5229 } 5230 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); 5231 5232 /** 5233 * netdev_rx_handler_register - register receive handler 5234 * @dev: device to register a handler for 5235 * @rx_handler: receive handler to register 5236 * @rx_handler_data: data pointer that is used by rx handler 5237 * 5238 * Register a receive handler for a device. This handler will then be 5239 * called from __netif_receive_skb. A negative errno code is returned 5240 * on a failure. 5241 * 5242 * The caller must hold the rtnl_mutex. 5243 * 5244 * For a general description of rx_handler, see enum rx_handler_result. 5245 */ 5246 int netdev_rx_handler_register(struct net_device *dev, 5247 rx_handler_func_t *rx_handler, 5248 void *rx_handler_data) 5249 { 5250 if (netdev_is_rx_handler_busy(dev)) 5251 return -EBUSY; 5252 5253 if (dev->priv_flags & IFF_NO_RX_HANDLER) 5254 return -EINVAL; 5255 5256 /* Note: rx_handler_data must be set before rx_handler */ 5257 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); 5258 rcu_assign_pointer(dev->rx_handler, rx_handler); 5259 5260 return 0; 5261 } 5262 EXPORT_SYMBOL_GPL(netdev_rx_handler_register); 5263 5264 /** 5265 * netdev_rx_handler_unregister - unregister receive handler 5266 * @dev: device to unregister a handler from 5267 * 5268 * Unregister a receive handler from a device. 5269 * 5270 * The caller must hold the rtnl_mutex. 5271 */ 5272 void netdev_rx_handler_unregister(struct net_device *dev) 5273 { 5274 5275 ASSERT_RTNL(); 5276 RCU_INIT_POINTER(dev->rx_handler, NULL); 5277 /* a reader seeing a non NULL rx_handler in a rcu_read_lock() 5278 * section has a guarantee to see a non NULL rx_handler_data 5279 * as well. 5280 */ 5281 synchronize_net(); 5282 RCU_INIT_POINTER(dev->rx_handler_data, NULL); 5283 } 5284 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); 5285 5286 /* 5287 * Limit the use of PFMEMALLOC reserves to those protocols that implement 5288 * the special handling of PFMEMALLOC skbs. 5289 */ 5290 static bool skb_pfmemalloc_protocol(struct sk_buff *skb) 5291 { 5292 switch (skb->protocol) { 5293 case htons(ETH_P_ARP): 5294 case htons(ETH_P_IP): 5295 case htons(ETH_P_IPV6): 5296 case htons(ETH_P_8021Q): 5297 case htons(ETH_P_8021AD): 5298 return true; 5299 default: 5300 return false; 5301 } 5302 } 5303 5304 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, 5305 int *ret, struct net_device *orig_dev) 5306 { 5307 if (nf_hook_ingress_active(skb)) { 5308 int ingress_retval; 5309 5310 if (*pt_prev) { 5311 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5312 *pt_prev = NULL; 5313 } 5314 5315 rcu_read_lock(); 5316 ingress_retval = nf_hook_ingress(skb); 5317 rcu_read_unlock(); 5318 return ingress_retval; 5319 } 5320 return 0; 5321 } 5322 5323 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc, 5324 struct packet_type **ppt_prev) 5325 { 5326 struct packet_type *ptype, *pt_prev; 5327 rx_handler_func_t *rx_handler; 5328 struct sk_buff *skb = *pskb; 5329 struct net_device *orig_dev; 5330 bool deliver_exact = false; 5331 int ret = NET_RX_DROP; 5332 __be16 type; 5333 5334 net_timestamp_check(!READ_ONCE(netdev_tstamp_prequeue), skb); 5335 5336 trace_netif_receive_skb(skb); 5337 5338 orig_dev = skb->dev; 5339 5340 skb_reset_network_header(skb); 5341 if (!skb_transport_header_was_set(skb)) 5342 skb_reset_transport_header(skb); 5343 skb_reset_mac_len(skb); 5344 5345 pt_prev = NULL; 5346 5347 another_round: 5348 skb->skb_iif = skb->dev->ifindex; 5349 5350 __this_cpu_inc(softnet_data.processed); 5351 5352 if (static_branch_unlikely(&generic_xdp_needed_key)) { 5353 int ret2; 5354 5355 migrate_disable(); 5356 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb); 5357 migrate_enable(); 5358 5359 if (ret2 != XDP_PASS) { 5360 ret = NET_RX_DROP; 5361 goto out; 5362 } 5363 } 5364 5365 if (eth_type_vlan(skb->protocol)) { 5366 skb = skb_vlan_untag(skb); 5367 if (unlikely(!skb)) 5368 goto out; 5369 } 5370 5371 if (skb_skip_tc_classify(skb)) 5372 goto skip_classify; 5373 5374 if (pfmemalloc) 5375 goto skip_taps; 5376 5377 list_for_each_entry_rcu(ptype, &ptype_all, list) { 5378 if (pt_prev) 5379 ret = deliver_skb(skb, pt_prev, orig_dev); 5380 pt_prev = ptype; 5381 } 5382 5383 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { 5384 if (pt_prev) 5385 ret = deliver_skb(skb, pt_prev, orig_dev); 5386 pt_prev = ptype; 5387 } 5388 5389 skip_taps: 5390 #ifdef CONFIG_NET_INGRESS 5391 if (static_branch_unlikely(&ingress_needed_key)) { 5392 bool another = false; 5393 5394 nf_skip_egress(skb, true); 5395 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev, 5396 &another); 5397 if (another) 5398 goto another_round; 5399 if (!skb) 5400 goto out; 5401 5402 nf_skip_egress(skb, false); 5403 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) 5404 goto out; 5405 } 5406 #endif 5407 skb_reset_redirect(skb); 5408 skip_classify: 5409 if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) 5410 goto drop; 5411 5412 if (skb_vlan_tag_present(skb)) { 5413 if (pt_prev) { 5414 ret = deliver_skb(skb, pt_prev, orig_dev); 5415 pt_prev = NULL; 5416 } 5417 if (vlan_do_receive(&skb)) 5418 goto another_round; 5419 else if (unlikely(!skb)) 5420 goto out; 5421 } 5422 5423 rx_handler = rcu_dereference(skb->dev->rx_handler); 5424 if (rx_handler) { 5425 if (pt_prev) { 5426 ret = deliver_skb(skb, pt_prev, orig_dev); 5427 pt_prev = NULL; 5428 } 5429 switch (rx_handler(&skb)) { 5430 case RX_HANDLER_CONSUMED: 5431 ret = NET_RX_SUCCESS; 5432 goto out; 5433 case RX_HANDLER_ANOTHER: 5434 goto another_round; 5435 case RX_HANDLER_EXACT: 5436 deliver_exact = true; 5437 break; 5438 case RX_HANDLER_PASS: 5439 break; 5440 default: 5441 BUG(); 5442 } 5443 } 5444 5445 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) { 5446 check_vlan_id: 5447 if (skb_vlan_tag_get_id(skb)) { 5448 /* Vlan id is non 0 and vlan_do_receive() above couldn't 5449 * find vlan device. 5450 */ 5451 skb->pkt_type = PACKET_OTHERHOST; 5452 } else if (eth_type_vlan(skb->protocol)) { 5453 /* Outer header is 802.1P with vlan 0, inner header is 5454 * 802.1Q or 802.1AD and vlan_do_receive() above could 5455 * not find vlan dev for vlan id 0. 5456 */ 5457 __vlan_hwaccel_clear_tag(skb); 5458 skb = skb_vlan_untag(skb); 5459 if (unlikely(!skb)) 5460 goto out; 5461 if (vlan_do_receive(&skb)) 5462 /* After stripping off 802.1P header with vlan 0 5463 * vlan dev is found for inner header. 5464 */ 5465 goto another_round; 5466 else if (unlikely(!skb)) 5467 goto out; 5468 else 5469 /* We have stripped outer 802.1P vlan 0 header. 5470 * But could not find vlan dev. 5471 * check again for vlan id to set OTHERHOST. 5472 */ 5473 goto check_vlan_id; 5474 } 5475 /* Note: we might in the future use prio bits 5476 * and set skb->priority like in vlan_do_receive() 5477 * For the time being, just ignore Priority Code Point 5478 */ 5479 __vlan_hwaccel_clear_tag(skb); 5480 } 5481 5482 type = skb->protocol; 5483 5484 /* deliver only exact match when indicated */ 5485 if (likely(!deliver_exact)) { 5486 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5487 &ptype_base[ntohs(type) & 5488 PTYPE_HASH_MASK]); 5489 } 5490 5491 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5492 &orig_dev->ptype_specific); 5493 5494 if (unlikely(skb->dev != orig_dev)) { 5495 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5496 &skb->dev->ptype_specific); 5497 } 5498 5499 if (pt_prev) { 5500 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 5501 goto drop; 5502 *ppt_prev = pt_prev; 5503 } else { 5504 drop: 5505 if (!deliver_exact) 5506 dev_core_stats_rx_dropped_inc(skb->dev); 5507 else 5508 dev_core_stats_rx_nohandler_inc(skb->dev); 5509 kfree_skb_reason(skb, SKB_DROP_REASON_UNHANDLED_PROTO); 5510 /* Jamal, now you will not able to escape explaining 5511 * me how you were going to use this. :-) 5512 */ 5513 ret = NET_RX_DROP; 5514 } 5515 5516 out: 5517 /* The invariant here is that if *ppt_prev is not NULL 5518 * then skb should also be non-NULL. 5519 * 5520 * Apparently *ppt_prev assignment above holds this invariant due to 5521 * skb dereferencing near it. 5522 */ 5523 *pskb = skb; 5524 return ret; 5525 } 5526 5527 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc) 5528 { 5529 struct net_device *orig_dev = skb->dev; 5530 struct packet_type *pt_prev = NULL; 5531 int ret; 5532 5533 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5534 if (pt_prev) 5535 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb, 5536 skb->dev, pt_prev, orig_dev); 5537 return ret; 5538 } 5539 5540 /** 5541 * netif_receive_skb_core - special purpose version of netif_receive_skb 5542 * @skb: buffer to process 5543 * 5544 * More direct receive version of netif_receive_skb(). It should 5545 * only be used by callers that have a need to skip RPS and Generic XDP. 5546 * Caller must also take care of handling if ``(page_is_)pfmemalloc``. 5547 * 5548 * This function may only be called from softirq context and interrupts 5549 * should be enabled. 5550 * 5551 * Return values (usually ignored): 5552 * NET_RX_SUCCESS: no congestion 5553 * NET_RX_DROP: packet was dropped 5554 */ 5555 int netif_receive_skb_core(struct sk_buff *skb) 5556 { 5557 int ret; 5558 5559 rcu_read_lock(); 5560 ret = __netif_receive_skb_one_core(skb, false); 5561 rcu_read_unlock(); 5562 5563 return ret; 5564 } 5565 EXPORT_SYMBOL(netif_receive_skb_core); 5566 5567 static inline void __netif_receive_skb_list_ptype(struct list_head *head, 5568 struct packet_type *pt_prev, 5569 struct net_device *orig_dev) 5570 { 5571 struct sk_buff *skb, *next; 5572 5573 if (!pt_prev) 5574 return; 5575 if (list_empty(head)) 5576 return; 5577 if (pt_prev->list_func != NULL) 5578 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv, 5579 ip_list_rcv, head, pt_prev, orig_dev); 5580 else 5581 list_for_each_entry_safe(skb, next, head, list) { 5582 skb_list_del_init(skb); 5583 pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 5584 } 5585 } 5586 5587 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc) 5588 { 5589 /* Fast-path assumptions: 5590 * - There is no RX handler. 5591 * - Only one packet_type matches. 5592 * If either of these fails, we will end up doing some per-packet 5593 * processing in-line, then handling the 'last ptype' for the whole 5594 * sublist. This can't cause out-of-order delivery to any single ptype, 5595 * because the 'last ptype' must be constant across the sublist, and all 5596 * other ptypes are handled per-packet. 5597 */ 5598 /* Current (common) ptype of sublist */ 5599 struct packet_type *pt_curr = NULL; 5600 /* Current (common) orig_dev of sublist */ 5601 struct net_device *od_curr = NULL; 5602 struct list_head sublist; 5603 struct sk_buff *skb, *next; 5604 5605 INIT_LIST_HEAD(&sublist); 5606 list_for_each_entry_safe(skb, next, head, list) { 5607 struct net_device *orig_dev = skb->dev; 5608 struct packet_type *pt_prev = NULL; 5609 5610 skb_list_del_init(skb); 5611 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5612 if (!pt_prev) 5613 continue; 5614 if (pt_curr != pt_prev || od_curr != orig_dev) { 5615 /* dispatch old sublist */ 5616 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5617 /* start new sublist */ 5618 INIT_LIST_HEAD(&sublist); 5619 pt_curr = pt_prev; 5620 od_curr = orig_dev; 5621 } 5622 list_add_tail(&skb->list, &sublist); 5623 } 5624 5625 /* dispatch final sublist */ 5626 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5627 } 5628 5629 static int __netif_receive_skb(struct sk_buff *skb) 5630 { 5631 int ret; 5632 5633 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { 5634 unsigned int noreclaim_flag; 5635 5636 /* 5637 * PFMEMALLOC skbs are special, they should 5638 * - be delivered to SOCK_MEMALLOC sockets only 5639 * - stay away from userspace 5640 * - have bounded memory usage 5641 * 5642 * Use PF_MEMALLOC as this saves us from propagating the allocation 5643 * context down to all allocation sites. 5644 */ 5645 noreclaim_flag = memalloc_noreclaim_save(); 5646 ret = __netif_receive_skb_one_core(skb, true); 5647 memalloc_noreclaim_restore(noreclaim_flag); 5648 } else 5649 ret = __netif_receive_skb_one_core(skb, false); 5650 5651 return ret; 5652 } 5653 5654 static void __netif_receive_skb_list(struct list_head *head) 5655 { 5656 unsigned long noreclaim_flag = 0; 5657 struct sk_buff *skb, *next; 5658 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */ 5659 5660 list_for_each_entry_safe(skb, next, head, list) { 5661 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) { 5662 struct list_head sublist; 5663 5664 /* Handle the previous sublist */ 5665 list_cut_before(&sublist, head, &skb->list); 5666 if (!list_empty(&sublist)) 5667 __netif_receive_skb_list_core(&sublist, pfmemalloc); 5668 pfmemalloc = !pfmemalloc; 5669 /* See comments in __netif_receive_skb */ 5670 if (pfmemalloc) 5671 noreclaim_flag = memalloc_noreclaim_save(); 5672 else 5673 memalloc_noreclaim_restore(noreclaim_flag); 5674 } 5675 } 5676 /* Handle the remaining sublist */ 5677 if (!list_empty(head)) 5678 __netif_receive_skb_list_core(head, pfmemalloc); 5679 /* Restore pflags */ 5680 if (pfmemalloc) 5681 memalloc_noreclaim_restore(noreclaim_flag); 5682 } 5683 5684 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp) 5685 { 5686 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog); 5687 struct bpf_prog *new = xdp->prog; 5688 int ret = 0; 5689 5690 switch (xdp->command) { 5691 case XDP_SETUP_PROG: 5692 rcu_assign_pointer(dev->xdp_prog, new); 5693 if (old) 5694 bpf_prog_put(old); 5695 5696 if (old && !new) { 5697 static_branch_dec(&generic_xdp_needed_key); 5698 } else if (new && !old) { 5699 static_branch_inc(&generic_xdp_needed_key); 5700 dev_disable_lro(dev); 5701 dev_disable_gro_hw(dev); 5702 } 5703 break; 5704 5705 default: 5706 ret = -EINVAL; 5707 break; 5708 } 5709 5710 return ret; 5711 } 5712 5713 static int netif_receive_skb_internal(struct sk_buff *skb) 5714 { 5715 int ret; 5716 5717 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb); 5718 5719 if (skb_defer_rx_timestamp(skb)) 5720 return NET_RX_SUCCESS; 5721 5722 rcu_read_lock(); 5723 #ifdef CONFIG_RPS 5724 if (static_branch_unlikely(&rps_needed)) { 5725 struct rps_dev_flow voidflow, *rflow = &voidflow; 5726 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5727 5728 if (cpu >= 0) { 5729 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5730 rcu_read_unlock(); 5731 return ret; 5732 } 5733 } 5734 #endif 5735 ret = __netif_receive_skb(skb); 5736 rcu_read_unlock(); 5737 return ret; 5738 } 5739 5740 void netif_receive_skb_list_internal(struct list_head *head) 5741 { 5742 struct sk_buff *skb, *next; 5743 struct list_head sublist; 5744 5745 INIT_LIST_HEAD(&sublist); 5746 list_for_each_entry_safe(skb, next, head, list) { 5747 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb); 5748 skb_list_del_init(skb); 5749 if (!skb_defer_rx_timestamp(skb)) 5750 list_add_tail(&skb->list, &sublist); 5751 } 5752 list_splice_init(&sublist, head); 5753 5754 rcu_read_lock(); 5755 #ifdef CONFIG_RPS 5756 if (static_branch_unlikely(&rps_needed)) { 5757 list_for_each_entry_safe(skb, next, head, list) { 5758 struct rps_dev_flow voidflow, *rflow = &voidflow; 5759 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5760 5761 if (cpu >= 0) { 5762 /* Will be handled, remove from list */ 5763 skb_list_del_init(skb); 5764 enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5765 } 5766 } 5767 } 5768 #endif 5769 __netif_receive_skb_list(head); 5770 rcu_read_unlock(); 5771 } 5772 5773 /** 5774 * netif_receive_skb - process receive buffer from network 5775 * @skb: buffer to process 5776 * 5777 * netif_receive_skb() is the main receive data processing function. 5778 * It always succeeds. The buffer may be dropped during processing 5779 * for congestion control or by the protocol layers. 5780 * 5781 * This function may only be called from softirq context and interrupts 5782 * should be enabled. 5783 * 5784 * Return values (usually ignored): 5785 * NET_RX_SUCCESS: no congestion 5786 * NET_RX_DROP: packet was dropped 5787 */ 5788 int netif_receive_skb(struct sk_buff *skb) 5789 { 5790 int ret; 5791 5792 trace_netif_receive_skb_entry(skb); 5793 5794 ret = netif_receive_skb_internal(skb); 5795 trace_netif_receive_skb_exit(ret); 5796 5797 return ret; 5798 } 5799 EXPORT_SYMBOL(netif_receive_skb); 5800 5801 /** 5802 * netif_receive_skb_list - process many receive buffers from network 5803 * @head: list of skbs to process. 5804 * 5805 * Since return value of netif_receive_skb() is normally ignored, and 5806 * wouldn't be meaningful for a list, this function returns void. 5807 * 5808 * This function may only be called from softirq context and interrupts 5809 * should be enabled. 5810 */ 5811 void netif_receive_skb_list(struct list_head *head) 5812 { 5813 struct sk_buff *skb; 5814 5815 if (list_empty(head)) 5816 return; 5817 if (trace_netif_receive_skb_list_entry_enabled()) { 5818 list_for_each_entry(skb, head, list) 5819 trace_netif_receive_skb_list_entry(skb); 5820 } 5821 netif_receive_skb_list_internal(head); 5822 trace_netif_receive_skb_list_exit(0); 5823 } 5824 EXPORT_SYMBOL(netif_receive_skb_list); 5825 5826 static DEFINE_PER_CPU(struct work_struct, flush_works); 5827 5828 /* Network device is going away, flush any packets still pending */ 5829 static void flush_backlog(struct work_struct *work) 5830 { 5831 struct sk_buff *skb, *tmp; 5832 struct softnet_data *sd; 5833 5834 local_bh_disable(); 5835 sd = this_cpu_ptr(&softnet_data); 5836 5837 rps_lock_irq_disable(sd); 5838 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { 5839 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5840 __skb_unlink(skb, &sd->input_pkt_queue); 5841 dev_kfree_skb_irq(skb); 5842 input_queue_head_incr(sd); 5843 } 5844 } 5845 rps_unlock_irq_enable(sd); 5846 5847 skb_queue_walk_safe(&sd->process_queue, skb, tmp) { 5848 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5849 __skb_unlink(skb, &sd->process_queue); 5850 kfree_skb(skb); 5851 input_queue_head_incr(sd); 5852 } 5853 } 5854 local_bh_enable(); 5855 } 5856 5857 static bool flush_required(int cpu) 5858 { 5859 #if IS_ENABLED(CONFIG_RPS) 5860 struct softnet_data *sd = &per_cpu(softnet_data, cpu); 5861 bool do_flush; 5862 5863 rps_lock_irq_disable(sd); 5864 5865 /* as insertion into process_queue happens with the rps lock held, 5866 * process_queue access may race only with dequeue 5867 */ 5868 do_flush = !skb_queue_empty(&sd->input_pkt_queue) || 5869 !skb_queue_empty_lockless(&sd->process_queue); 5870 rps_unlock_irq_enable(sd); 5871 5872 return do_flush; 5873 #endif 5874 /* without RPS we can't safely check input_pkt_queue: during a 5875 * concurrent remote skb_queue_splice() we can detect as empty both 5876 * input_pkt_queue and process_queue even if the latter could end-up 5877 * containing a lot of packets. 5878 */ 5879 return true; 5880 } 5881 5882 static void flush_all_backlogs(void) 5883 { 5884 static cpumask_t flush_cpus; 5885 unsigned int cpu; 5886 5887 /* since we are under rtnl lock protection we can use static data 5888 * for the cpumask and avoid allocating on stack the possibly 5889 * large mask 5890 */ 5891 ASSERT_RTNL(); 5892 5893 cpus_read_lock(); 5894 5895 cpumask_clear(&flush_cpus); 5896 for_each_online_cpu(cpu) { 5897 if (flush_required(cpu)) { 5898 queue_work_on(cpu, system_highpri_wq, 5899 per_cpu_ptr(&flush_works, cpu)); 5900 cpumask_set_cpu(cpu, &flush_cpus); 5901 } 5902 } 5903 5904 /* we can have in flight packet[s] on the cpus we are not flushing, 5905 * synchronize_net() in unregister_netdevice_many() will take care of 5906 * them 5907 */ 5908 for_each_cpu(cpu, &flush_cpus) 5909 flush_work(per_cpu_ptr(&flush_works, cpu)); 5910 5911 cpus_read_unlock(); 5912 } 5913 5914 static void net_rps_send_ipi(struct softnet_data *remsd) 5915 { 5916 #ifdef CONFIG_RPS 5917 while (remsd) { 5918 struct softnet_data *next = remsd->rps_ipi_next; 5919 5920 if (cpu_online(remsd->cpu)) 5921 smp_call_function_single_async(remsd->cpu, &remsd->csd); 5922 remsd = next; 5923 } 5924 #endif 5925 } 5926 5927 /* 5928 * net_rps_action_and_irq_enable sends any pending IPI's for rps. 5929 * Note: called with local irq disabled, but exits with local irq enabled. 5930 */ 5931 static void net_rps_action_and_irq_enable(struct softnet_data *sd) 5932 { 5933 #ifdef CONFIG_RPS 5934 struct softnet_data *remsd = sd->rps_ipi_list; 5935 5936 if (remsd) { 5937 sd->rps_ipi_list = NULL; 5938 5939 local_irq_enable(); 5940 5941 /* Send pending IPI's to kick RPS processing on remote cpus. */ 5942 net_rps_send_ipi(remsd); 5943 } else 5944 #endif 5945 local_irq_enable(); 5946 } 5947 5948 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) 5949 { 5950 #ifdef CONFIG_RPS 5951 return sd->rps_ipi_list != NULL; 5952 #else 5953 return false; 5954 #endif 5955 } 5956 5957 static int process_backlog(struct napi_struct *napi, int quota) 5958 { 5959 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); 5960 bool again = true; 5961 int work = 0; 5962 5963 /* Check if we have pending ipi, its better to send them now, 5964 * not waiting net_rx_action() end. 5965 */ 5966 if (sd_has_rps_ipi_waiting(sd)) { 5967 local_irq_disable(); 5968 net_rps_action_and_irq_enable(sd); 5969 } 5970 5971 napi->weight = READ_ONCE(dev_rx_weight); 5972 while (again) { 5973 struct sk_buff *skb; 5974 5975 while ((skb = __skb_dequeue(&sd->process_queue))) { 5976 rcu_read_lock(); 5977 __netif_receive_skb(skb); 5978 rcu_read_unlock(); 5979 input_queue_head_incr(sd); 5980 if (++work >= quota) 5981 return work; 5982 5983 } 5984 5985 rps_lock_irq_disable(sd); 5986 if (skb_queue_empty(&sd->input_pkt_queue)) { 5987 /* 5988 * Inline a custom version of __napi_complete(). 5989 * only current cpu owns and manipulates this napi, 5990 * and NAPI_STATE_SCHED is the only possible flag set 5991 * on backlog. 5992 * We can use a plain write instead of clear_bit(), 5993 * and we dont need an smp_mb() memory barrier. 5994 */ 5995 napi->state = 0; 5996 again = false; 5997 } else { 5998 skb_queue_splice_tail_init(&sd->input_pkt_queue, 5999 &sd->process_queue); 6000 } 6001 rps_unlock_irq_enable(sd); 6002 } 6003 6004 return work; 6005 } 6006 6007 /** 6008 * __napi_schedule - schedule for receive 6009 * @n: entry to schedule 6010 * 6011 * The entry's receive function will be scheduled to run. 6012 * Consider using __napi_schedule_irqoff() if hard irqs are masked. 6013 */ 6014 void __napi_schedule(struct napi_struct *n) 6015 { 6016 unsigned long flags; 6017 6018 local_irq_save(flags); 6019 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6020 local_irq_restore(flags); 6021 } 6022 EXPORT_SYMBOL(__napi_schedule); 6023 6024 /** 6025 * napi_schedule_prep - check if napi can be scheduled 6026 * @n: napi context 6027 * 6028 * Test if NAPI routine is already running, and if not mark 6029 * it as running. This is used as a condition variable to 6030 * insure only one NAPI poll instance runs. We also make 6031 * sure there is no pending NAPI disable. 6032 */ 6033 bool napi_schedule_prep(struct napi_struct *n) 6034 { 6035 unsigned long new, val = READ_ONCE(n->state); 6036 6037 do { 6038 if (unlikely(val & NAPIF_STATE_DISABLE)) 6039 return false; 6040 new = val | NAPIF_STATE_SCHED; 6041 6042 /* Sets STATE_MISSED bit if STATE_SCHED was already set 6043 * This was suggested by Alexander Duyck, as compiler 6044 * emits better code than : 6045 * if (val & NAPIF_STATE_SCHED) 6046 * new |= NAPIF_STATE_MISSED; 6047 */ 6048 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * 6049 NAPIF_STATE_MISSED; 6050 } while (!try_cmpxchg(&n->state, &val, new)); 6051 6052 return !(val & NAPIF_STATE_SCHED); 6053 } 6054 EXPORT_SYMBOL(napi_schedule_prep); 6055 6056 /** 6057 * __napi_schedule_irqoff - schedule for receive 6058 * @n: entry to schedule 6059 * 6060 * Variant of __napi_schedule() assuming hard irqs are masked. 6061 * 6062 * On PREEMPT_RT enabled kernels this maps to __napi_schedule() 6063 * because the interrupt disabled assumption might not be true 6064 * due to force-threaded interrupts and spinlock substitution. 6065 */ 6066 void __napi_schedule_irqoff(struct napi_struct *n) 6067 { 6068 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6069 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6070 else 6071 __napi_schedule(n); 6072 } 6073 EXPORT_SYMBOL(__napi_schedule_irqoff); 6074 6075 bool napi_complete_done(struct napi_struct *n, int work_done) 6076 { 6077 unsigned long flags, val, new, timeout = 0; 6078 bool ret = true; 6079 6080 /* 6081 * 1) Don't let napi dequeue from the cpu poll list 6082 * just in case its running on a different cpu. 6083 * 2) If we are busy polling, do nothing here, we have 6084 * the guarantee we will be called later. 6085 */ 6086 if (unlikely(n->state & (NAPIF_STATE_NPSVC | 6087 NAPIF_STATE_IN_BUSY_POLL))) 6088 return false; 6089 6090 if (work_done) { 6091 if (n->gro_bitmask) 6092 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6093 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs); 6094 } 6095 if (n->defer_hard_irqs_count > 0) { 6096 n->defer_hard_irqs_count--; 6097 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6098 if (timeout) 6099 ret = false; 6100 } 6101 if (n->gro_bitmask) { 6102 /* When the NAPI instance uses a timeout and keeps postponing 6103 * it, we need to bound somehow the time packets are kept in 6104 * the GRO layer 6105 */ 6106 napi_gro_flush(n, !!timeout); 6107 } 6108 6109 gro_normal_list(n); 6110 6111 if (unlikely(!list_empty(&n->poll_list))) { 6112 /* If n->poll_list is not empty, we need to mask irqs */ 6113 local_irq_save(flags); 6114 list_del_init(&n->poll_list); 6115 local_irq_restore(flags); 6116 } 6117 WRITE_ONCE(n->list_owner, -1); 6118 6119 val = READ_ONCE(n->state); 6120 do { 6121 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); 6122 6123 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED | 6124 NAPIF_STATE_SCHED_THREADED | 6125 NAPIF_STATE_PREFER_BUSY_POLL); 6126 6127 /* If STATE_MISSED was set, leave STATE_SCHED set, 6128 * because we will call napi->poll() one more time. 6129 * This C code was suggested by Alexander Duyck to help gcc. 6130 */ 6131 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * 6132 NAPIF_STATE_SCHED; 6133 } while (!try_cmpxchg(&n->state, &val, new)); 6134 6135 if (unlikely(val & NAPIF_STATE_MISSED)) { 6136 __napi_schedule(n); 6137 return false; 6138 } 6139 6140 if (timeout) 6141 hrtimer_start(&n->timer, ns_to_ktime(timeout), 6142 HRTIMER_MODE_REL_PINNED); 6143 return ret; 6144 } 6145 EXPORT_SYMBOL(napi_complete_done); 6146 6147 /* must be called under rcu_read_lock(), as we dont take a reference */ 6148 struct napi_struct *napi_by_id(unsigned int napi_id) 6149 { 6150 unsigned int hash = napi_id % HASH_SIZE(napi_hash); 6151 struct napi_struct *napi; 6152 6153 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) 6154 if (napi->napi_id == napi_id) 6155 return napi; 6156 6157 return NULL; 6158 } 6159 6160 #if defined(CONFIG_NET_RX_BUSY_POLL) 6161 6162 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule) 6163 { 6164 if (!skip_schedule) { 6165 gro_normal_list(napi); 6166 __napi_schedule(napi); 6167 return; 6168 } 6169 6170 if (napi->gro_bitmask) { 6171 /* flush too old packets 6172 * If HZ < 1000, flush all packets. 6173 */ 6174 napi_gro_flush(napi, HZ >= 1000); 6175 } 6176 6177 gro_normal_list(napi); 6178 clear_bit(NAPI_STATE_SCHED, &napi->state); 6179 } 6180 6181 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll, 6182 u16 budget) 6183 { 6184 bool skip_schedule = false; 6185 unsigned long timeout; 6186 int rc; 6187 6188 /* Busy polling means there is a high chance device driver hard irq 6189 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was 6190 * set in napi_schedule_prep(). 6191 * Since we are about to call napi->poll() once more, we can safely 6192 * clear NAPI_STATE_MISSED. 6193 * 6194 * Note: x86 could use a single "lock and ..." instruction 6195 * to perform these two clear_bit() 6196 */ 6197 clear_bit(NAPI_STATE_MISSED, &napi->state); 6198 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); 6199 6200 local_bh_disable(); 6201 6202 if (prefer_busy_poll) { 6203 napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs); 6204 timeout = READ_ONCE(napi->dev->gro_flush_timeout); 6205 if (napi->defer_hard_irqs_count && timeout) { 6206 hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED); 6207 skip_schedule = true; 6208 } 6209 } 6210 6211 /* All we really want here is to re-enable device interrupts. 6212 * Ideally, a new ndo_busy_poll_stop() could avoid another round. 6213 */ 6214 rc = napi->poll(napi, budget); 6215 /* We can't gro_normal_list() here, because napi->poll() might have 6216 * rearmed the napi (napi_complete_done()) in which case it could 6217 * already be running on another CPU. 6218 */ 6219 trace_napi_poll(napi, rc, budget); 6220 netpoll_poll_unlock(have_poll_lock); 6221 if (rc == budget) 6222 __busy_poll_stop(napi, skip_schedule); 6223 local_bh_enable(); 6224 } 6225 6226 void napi_busy_loop(unsigned int napi_id, 6227 bool (*loop_end)(void *, unsigned long), 6228 void *loop_end_arg, bool prefer_busy_poll, u16 budget) 6229 { 6230 unsigned long start_time = loop_end ? busy_loop_current_time() : 0; 6231 int (*napi_poll)(struct napi_struct *napi, int budget); 6232 void *have_poll_lock = NULL; 6233 struct napi_struct *napi; 6234 6235 restart: 6236 napi_poll = NULL; 6237 6238 rcu_read_lock(); 6239 6240 napi = napi_by_id(napi_id); 6241 if (!napi) 6242 goto out; 6243 6244 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6245 preempt_disable(); 6246 for (;;) { 6247 int work = 0; 6248 6249 local_bh_disable(); 6250 if (!napi_poll) { 6251 unsigned long val = READ_ONCE(napi->state); 6252 6253 /* If multiple threads are competing for this napi, 6254 * we avoid dirtying napi->state as much as we can. 6255 */ 6256 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | 6257 NAPIF_STATE_IN_BUSY_POLL)) { 6258 if (prefer_busy_poll) 6259 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6260 goto count; 6261 } 6262 if (cmpxchg(&napi->state, val, 6263 val | NAPIF_STATE_IN_BUSY_POLL | 6264 NAPIF_STATE_SCHED) != val) { 6265 if (prefer_busy_poll) 6266 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6267 goto count; 6268 } 6269 have_poll_lock = netpoll_poll_lock(napi); 6270 napi_poll = napi->poll; 6271 } 6272 work = napi_poll(napi, budget); 6273 trace_napi_poll(napi, work, budget); 6274 gro_normal_list(napi); 6275 count: 6276 if (work > 0) 6277 __NET_ADD_STATS(dev_net(napi->dev), 6278 LINUX_MIB_BUSYPOLLRXPACKETS, work); 6279 local_bh_enable(); 6280 6281 if (!loop_end || loop_end(loop_end_arg, start_time)) 6282 break; 6283 6284 if (unlikely(need_resched())) { 6285 if (napi_poll) 6286 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6287 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6288 preempt_enable(); 6289 rcu_read_unlock(); 6290 cond_resched(); 6291 if (loop_end(loop_end_arg, start_time)) 6292 return; 6293 goto restart; 6294 } 6295 cpu_relax(); 6296 } 6297 if (napi_poll) 6298 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6299 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6300 preempt_enable(); 6301 out: 6302 rcu_read_unlock(); 6303 } 6304 EXPORT_SYMBOL(napi_busy_loop); 6305 6306 #endif /* CONFIG_NET_RX_BUSY_POLL */ 6307 6308 static void napi_hash_add(struct napi_struct *napi) 6309 { 6310 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state)) 6311 return; 6312 6313 spin_lock(&napi_hash_lock); 6314 6315 /* 0..NR_CPUS range is reserved for sender_cpu use */ 6316 do { 6317 if (unlikely(++napi_gen_id < MIN_NAPI_ID)) 6318 napi_gen_id = MIN_NAPI_ID; 6319 } while (napi_by_id(napi_gen_id)); 6320 napi->napi_id = napi_gen_id; 6321 6322 hlist_add_head_rcu(&napi->napi_hash_node, 6323 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); 6324 6325 spin_unlock(&napi_hash_lock); 6326 } 6327 6328 /* Warning : caller is responsible to make sure rcu grace period 6329 * is respected before freeing memory containing @napi 6330 */ 6331 static void napi_hash_del(struct napi_struct *napi) 6332 { 6333 spin_lock(&napi_hash_lock); 6334 6335 hlist_del_init_rcu(&napi->napi_hash_node); 6336 6337 spin_unlock(&napi_hash_lock); 6338 } 6339 6340 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) 6341 { 6342 struct napi_struct *napi; 6343 6344 napi = container_of(timer, struct napi_struct, timer); 6345 6346 /* Note : we use a relaxed variant of napi_schedule_prep() not setting 6347 * NAPI_STATE_MISSED, since we do not react to a device IRQ. 6348 */ 6349 if (!napi_disable_pending(napi) && 6350 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) { 6351 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6352 __napi_schedule_irqoff(napi); 6353 } 6354 6355 return HRTIMER_NORESTART; 6356 } 6357 6358 static void init_gro_hash(struct napi_struct *napi) 6359 { 6360 int i; 6361 6362 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6363 INIT_LIST_HEAD(&napi->gro_hash[i].list); 6364 napi->gro_hash[i].count = 0; 6365 } 6366 napi->gro_bitmask = 0; 6367 } 6368 6369 int dev_set_threaded(struct net_device *dev, bool threaded) 6370 { 6371 struct napi_struct *napi; 6372 int err = 0; 6373 6374 if (dev->threaded == threaded) 6375 return 0; 6376 6377 if (threaded) { 6378 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6379 if (!napi->thread) { 6380 err = napi_kthread_create(napi); 6381 if (err) { 6382 threaded = false; 6383 break; 6384 } 6385 } 6386 } 6387 } 6388 6389 dev->threaded = threaded; 6390 6391 /* Make sure kthread is created before THREADED bit 6392 * is set. 6393 */ 6394 smp_mb__before_atomic(); 6395 6396 /* Setting/unsetting threaded mode on a napi might not immediately 6397 * take effect, if the current napi instance is actively being 6398 * polled. In this case, the switch between threaded mode and 6399 * softirq mode will happen in the next round of napi_schedule(). 6400 * This should not cause hiccups/stalls to the live traffic. 6401 */ 6402 list_for_each_entry(napi, &dev->napi_list, dev_list) 6403 assign_bit(NAPI_STATE_THREADED, &napi->state, threaded); 6404 6405 return err; 6406 } 6407 EXPORT_SYMBOL(dev_set_threaded); 6408 6409 /** 6410 * netif_queue_set_napi - Associate queue with the napi 6411 * @dev: device to which NAPI and queue belong 6412 * @queue_index: Index of queue 6413 * @type: queue type as RX or TX 6414 * @napi: NAPI context, pass NULL to clear previously set NAPI 6415 * 6416 * Set queue with its corresponding napi context. This should be done after 6417 * registering the NAPI handler for the queue-vector and the queues have been 6418 * mapped to the corresponding interrupt vector. 6419 */ 6420 void netif_queue_set_napi(struct net_device *dev, unsigned int queue_index, 6421 enum netdev_queue_type type, struct napi_struct *napi) 6422 { 6423 struct netdev_rx_queue *rxq; 6424 struct netdev_queue *txq; 6425 6426 if (WARN_ON_ONCE(napi && !napi->dev)) 6427 return; 6428 if (dev->reg_state >= NETREG_REGISTERED) 6429 ASSERT_RTNL(); 6430 6431 switch (type) { 6432 case NETDEV_QUEUE_TYPE_RX: 6433 rxq = __netif_get_rx_queue(dev, queue_index); 6434 rxq->napi = napi; 6435 return; 6436 case NETDEV_QUEUE_TYPE_TX: 6437 txq = netdev_get_tx_queue(dev, queue_index); 6438 txq->napi = napi; 6439 return; 6440 default: 6441 return; 6442 } 6443 } 6444 EXPORT_SYMBOL(netif_queue_set_napi); 6445 6446 void netif_napi_add_weight(struct net_device *dev, struct napi_struct *napi, 6447 int (*poll)(struct napi_struct *, int), int weight) 6448 { 6449 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state))) 6450 return; 6451 6452 INIT_LIST_HEAD(&napi->poll_list); 6453 INIT_HLIST_NODE(&napi->napi_hash_node); 6454 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); 6455 napi->timer.function = napi_watchdog; 6456 init_gro_hash(napi); 6457 napi->skb = NULL; 6458 INIT_LIST_HEAD(&napi->rx_list); 6459 napi->rx_count = 0; 6460 napi->poll = poll; 6461 if (weight > NAPI_POLL_WEIGHT) 6462 netdev_err_once(dev, "%s() called with weight %d\n", __func__, 6463 weight); 6464 napi->weight = weight; 6465 napi->dev = dev; 6466 #ifdef CONFIG_NETPOLL 6467 napi->poll_owner = -1; 6468 #endif 6469 napi->list_owner = -1; 6470 set_bit(NAPI_STATE_SCHED, &napi->state); 6471 set_bit(NAPI_STATE_NPSVC, &napi->state); 6472 list_add_rcu(&napi->dev_list, &dev->napi_list); 6473 napi_hash_add(napi); 6474 napi_get_frags_check(napi); 6475 /* Create kthread for this napi if dev->threaded is set. 6476 * Clear dev->threaded if kthread creation failed so that 6477 * threaded mode will not be enabled in napi_enable(). 6478 */ 6479 if (dev->threaded && napi_kthread_create(napi)) 6480 dev->threaded = 0; 6481 netif_napi_set_irq(napi, -1); 6482 } 6483 EXPORT_SYMBOL(netif_napi_add_weight); 6484 6485 void napi_disable(struct napi_struct *n) 6486 { 6487 unsigned long val, new; 6488 6489 might_sleep(); 6490 set_bit(NAPI_STATE_DISABLE, &n->state); 6491 6492 val = READ_ONCE(n->state); 6493 do { 6494 while (val & (NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC)) { 6495 usleep_range(20, 200); 6496 val = READ_ONCE(n->state); 6497 } 6498 6499 new = val | NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC; 6500 new &= ~(NAPIF_STATE_THREADED | NAPIF_STATE_PREFER_BUSY_POLL); 6501 } while (!try_cmpxchg(&n->state, &val, new)); 6502 6503 hrtimer_cancel(&n->timer); 6504 6505 clear_bit(NAPI_STATE_DISABLE, &n->state); 6506 } 6507 EXPORT_SYMBOL(napi_disable); 6508 6509 /** 6510 * napi_enable - enable NAPI scheduling 6511 * @n: NAPI context 6512 * 6513 * Resume NAPI from being scheduled on this context. 6514 * Must be paired with napi_disable. 6515 */ 6516 void napi_enable(struct napi_struct *n) 6517 { 6518 unsigned long new, val = READ_ONCE(n->state); 6519 6520 do { 6521 BUG_ON(!test_bit(NAPI_STATE_SCHED, &val)); 6522 6523 new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC); 6524 if (n->dev->threaded && n->thread) 6525 new |= NAPIF_STATE_THREADED; 6526 } while (!try_cmpxchg(&n->state, &val, new)); 6527 } 6528 EXPORT_SYMBOL(napi_enable); 6529 6530 static void flush_gro_hash(struct napi_struct *napi) 6531 { 6532 int i; 6533 6534 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6535 struct sk_buff *skb, *n; 6536 6537 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list) 6538 kfree_skb(skb); 6539 napi->gro_hash[i].count = 0; 6540 } 6541 } 6542 6543 /* Must be called in process context */ 6544 void __netif_napi_del(struct napi_struct *napi) 6545 { 6546 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state)) 6547 return; 6548 6549 napi_hash_del(napi); 6550 list_del_rcu(&napi->dev_list); 6551 napi_free_frags(napi); 6552 6553 flush_gro_hash(napi); 6554 napi->gro_bitmask = 0; 6555 6556 if (napi->thread) { 6557 kthread_stop(napi->thread); 6558 napi->thread = NULL; 6559 } 6560 } 6561 EXPORT_SYMBOL(__netif_napi_del); 6562 6563 static int __napi_poll(struct napi_struct *n, bool *repoll) 6564 { 6565 int work, weight; 6566 6567 weight = n->weight; 6568 6569 /* This NAPI_STATE_SCHED test is for avoiding a race 6570 * with netpoll's poll_napi(). Only the entity which 6571 * obtains the lock and sees NAPI_STATE_SCHED set will 6572 * actually make the ->poll() call. Therefore we avoid 6573 * accidentally calling ->poll() when NAPI is not scheduled. 6574 */ 6575 work = 0; 6576 if (napi_is_scheduled(n)) { 6577 work = n->poll(n, weight); 6578 trace_napi_poll(n, work, weight); 6579 6580 xdp_do_check_flushed(n); 6581 } 6582 6583 if (unlikely(work > weight)) 6584 netdev_err_once(n->dev, "NAPI poll function %pS returned %d, exceeding its budget of %d.\n", 6585 n->poll, work, weight); 6586 6587 if (likely(work < weight)) 6588 return work; 6589 6590 /* Drivers must not modify the NAPI state if they 6591 * consume the entire weight. In such cases this code 6592 * still "owns" the NAPI instance and therefore can 6593 * move the instance around on the list at-will. 6594 */ 6595 if (unlikely(napi_disable_pending(n))) { 6596 napi_complete(n); 6597 return work; 6598 } 6599 6600 /* The NAPI context has more processing work, but busy-polling 6601 * is preferred. Exit early. 6602 */ 6603 if (napi_prefer_busy_poll(n)) { 6604 if (napi_complete_done(n, work)) { 6605 /* If timeout is not set, we need to make sure 6606 * that the NAPI is re-scheduled. 6607 */ 6608 napi_schedule(n); 6609 } 6610 return work; 6611 } 6612 6613 if (n->gro_bitmask) { 6614 /* flush too old packets 6615 * If HZ < 1000, flush all packets. 6616 */ 6617 napi_gro_flush(n, HZ >= 1000); 6618 } 6619 6620 gro_normal_list(n); 6621 6622 /* Some drivers may have called napi_schedule 6623 * prior to exhausting their budget. 6624 */ 6625 if (unlikely(!list_empty(&n->poll_list))) { 6626 pr_warn_once("%s: Budget exhausted after napi rescheduled\n", 6627 n->dev ? n->dev->name : "backlog"); 6628 return work; 6629 } 6630 6631 *repoll = true; 6632 6633 return work; 6634 } 6635 6636 static int napi_poll(struct napi_struct *n, struct list_head *repoll) 6637 { 6638 bool do_repoll = false; 6639 void *have; 6640 int work; 6641 6642 list_del_init(&n->poll_list); 6643 6644 have = netpoll_poll_lock(n); 6645 6646 work = __napi_poll(n, &do_repoll); 6647 6648 if (do_repoll) 6649 list_add_tail(&n->poll_list, repoll); 6650 6651 netpoll_poll_unlock(have); 6652 6653 return work; 6654 } 6655 6656 static int napi_thread_wait(struct napi_struct *napi) 6657 { 6658 bool woken = false; 6659 6660 set_current_state(TASK_INTERRUPTIBLE); 6661 6662 while (!kthread_should_stop()) { 6663 /* Testing SCHED_THREADED bit here to make sure the current 6664 * kthread owns this napi and could poll on this napi. 6665 * Testing SCHED bit is not enough because SCHED bit might be 6666 * set by some other busy poll thread or by napi_disable(). 6667 */ 6668 if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state) || woken) { 6669 WARN_ON(!list_empty(&napi->poll_list)); 6670 __set_current_state(TASK_RUNNING); 6671 return 0; 6672 } 6673 6674 schedule(); 6675 /* woken being true indicates this thread owns this napi. */ 6676 woken = true; 6677 set_current_state(TASK_INTERRUPTIBLE); 6678 } 6679 __set_current_state(TASK_RUNNING); 6680 6681 return -1; 6682 } 6683 6684 static void skb_defer_free_flush(struct softnet_data *sd) 6685 { 6686 struct sk_buff *skb, *next; 6687 6688 /* Paired with WRITE_ONCE() in skb_attempt_defer_free() */ 6689 if (!READ_ONCE(sd->defer_list)) 6690 return; 6691 6692 spin_lock(&sd->defer_lock); 6693 skb = sd->defer_list; 6694 sd->defer_list = NULL; 6695 sd->defer_count = 0; 6696 spin_unlock(&sd->defer_lock); 6697 6698 while (skb != NULL) { 6699 next = skb->next; 6700 napi_consume_skb(skb, 1); 6701 skb = next; 6702 } 6703 } 6704 6705 static int napi_threaded_poll(void *data) 6706 { 6707 struct napi_struct *napi = data; 6708 struct softnet_data *sd; 6709 void *have; 6710 6711 while (!napi_thread_wait(napi)) { 6712 for (;;) { 6713 bool repoll = false; 6714 6715 local_bh_disable(); 6716 sd = this_cpu_ptr(&softnet_data); 6717 sd->in_napi_threaded_poll = true; 6718 6719 have = netpoll_poll_lock(napi); 6720 __napi_poll(napi, &repoll); 6721 netpoll_poll_unlock(have); 6722 6723 sd->in_napi_threaded_poll = false; 6724 barrier(); 6725 6726 if (sd_has_rps_ipi_waiting(sd)) { 6727 local_irq_disable(); 6728 net_rps_action_and_irq_enable(sd); 6729 } 6730 skb_defer_free_flush(sd); 6731 local_bh_enable(); 6732 6733 if (!repoll) 6734 break; 6735 6736 cond_resched(); 6737 } 6738 } 6739 return 0; 6740 } 6741 6742 static __latent_entropy void net_rx_action(struct softirq_action *h) 6743 { 6744 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 6745 unsigned long time_limit = jiffies + 6746 usecs_to_jiffies(READ_ONCE(netdev_budget_usecs)); 6747 int budget = READ_ONCE(netdev_budget); 6748 LIST_HEAD(list); 6749 LIST_HEAD(repoll); 6750 6751 start: 6752 sd->in_net_rx_action = true; 6753 local_irq_disable(); 6754 list_splice_init(&sd->poll_list, &list); 6755 local_irq_enable(); 6756 6757 for (;;) { 6758 struct napi_struct *n; 6759 6760 skb_defer_free_flush(sd); 6761 6762 if (list_empty(&list)) { 6763 if (list_empty(&repoll)) { 6764 sd->in_net_rx_action = false; 6765 barrier(); 6766 /* We need to check if ____napi_schedule() 6767 * had refilled poll_list while 6768 * sd->in_net_rx_action was true. 6769 */ 6770 if (!list_empty(&sd->poll_list)) 6771 goto start; 6772 if (!sd_has_rps_ipi_waiting(sd)) 6773 goto end; 6774 } 6775 break; 6776 } 6777 6778 n = list_first_entry(&list, struct napi_struct, poll_list); 6779 budget -= napi_poll(n, &repoll); 6780 6781 /* If softirq window is exhausted then punt. 6782 * Allow this to run for 2 jiffies since which will allow 6783 * an average latency of 1.5/HZ. 6784 */ 6785 if (unlikely(budget <= 0 || 6786 time_after_eq(jiffies, time_limit))) { 6787 sd->time_squeeze++; 6788 break; 6789 } 6790 } 6791 6792 local_irq_disable(); 6793 6794 list_splice_tail_init(&sd->poll_list, &list); 6795 list_splice_tail(&repoll, &list); 6796 list_splice(&list, &sd->poll_list); 6797 if (!list_empty(&sd->poll_list)) 6798 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 6799 else 6800 sd->in_net_rx_action = false; 6801 6802 net_rps_action_and_irq_enable(sd); 6803 end:; 6804 } 6805 6806 struct netdev_adjacent { 6807 struct net_device *dev; 6808 netdevice_tracker dev_tracker; 6809 6810 /* upper master flag, there can only be one master device per list */ 6811 bool master; 6812 6813 /* lookup ignore flag */ 6814 bool ignore; 6815 6816 /* counter for the number of times this device was added to us */ 6817 u16 ref_nr; 6818 6819 /* private field for the users */ 6820 void *private; 6821 6822 struct list_head list; 6823 struct rcu_head rcu; 6824 }; 6825 6826 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, 6827 struct list_head *adj_list) 6828 { 6829 struct netdev_adjacent *adj; 6830 6831 list_for_each_entry(adj, adj_list, list) { 6832 if (adj->dev == adj_dev) 6833 return adj; 6834 } 6835 return NULL; 6836 } 6837 6838 static int ____netdev_has_upper_dev(struct net_device *upper_dev, 6839 struct netdev_nested_priv *priv) 6840 { 6841 struct net_device *dev = (struct net_device *)priv->data; 6842 6843 return upper_dev == dev; 6844 } 6845 6846 /** 6847 * netdev_has_upper_dev - Check if device is linked to an upper device 6848 * @dev: device 6849 * @upper_dev: upper device to check 6850 * 6851 * Find out if a device is linked to specified upper device and return true 6852 * in case it is. Note that this checks only immediate upper device, 6853 * not through a complete stack of devices. The caller must hold the RTNL lock. 6854 */ 6855 bool netdev_has_upper_dev(struct net_device *dev, 6856 struct net_device *upper_dev) 6857 { 6858 struct netdev_nested_priv priv = { 6859 .data = (void *)upper_dev, 6860 }; 6861 6862 ASSERT_RTNL(); 6863 6864 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 6865 &priv); 6866 } 6867 EXPORT_SYMBOL(netdev_has_upper_dev); 6868 6869 /** 6870 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device 6871 * @dev: device 6872 * @upper_dev: upper device to check 6873 * 6874 * Find out if a device is linked to specified upper device and return true 6875 * in case it is. Note that this checks the entire upper device chain. 6876 * The caller must hold rcu lock. 6877 */ 6878 6879 bool netdev_has_upper_dev_all_rcu(struct net_device *dev, 6880 struct net_device *upper_dev) 6881 { 6882 struct netdev_nested_priv priv = { 6883 .data = (void *)upper_dev, 6884 }; 6885 6886 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 6887 &priv); 6888 } 6889 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); 6890 6891 /** 6892 * netdev_has_any_upper_dev - Check if device is linked to some device 6893 * @dev: device 6894 * 6895 * Find out if a device is linked to an upper device and return true in case 6896 * it is. The caller must hold the RTNL lock. 6897 */ 6898 bool netdev_has_any_upper_dev(struct net_device *dev) 6899 { 6900 ASSERT_RTNL(); 6901 6902 return !list_empty(&dev->adj_list.upper); 6903 } 6904 EXPORT_SYMBOL(netdev_has_any_upper_dev); 6905 6906 /** 6907 * netdev_master_upper_dev_get - Get master upper device 6908 * @dev: device 6909 * 6910 * Find a master upper device and return pointer to it or NULL in case 6911 * it's not there. The caller must hold the RTNL lock. 6912 */ 6913 struct net_device *netdev_master_upper_dev_get(struct net_device *dev) 6914 { 6915 struct netdev_adjacent *upper; 6916 6917 ASSERT_RTNL(); 6918 6919 if (list_empty(&dev->adj_list.upper)) 6920 return NULL; 6921 6922 upper = list_first_entry(&dev->adj_list.upper, 6923 struct netdev_adjacent, list); 6924 if (likely(upper->master)) 6925 return upper->dev; 6926 return NULL; 6927 } 6928 EXPORT_SYMBOL(netdev_master_upper_dev_get); 6929 6930 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev) 6931 { 6932 struct netdev_adjacent *upper; 6933 6934 ASSERT_RTNL(); 6935 6936 if (list_empty(&dev->adj_list.upper)) 6937 return NULL; 6938 6939 upper = list_first_entry(&dev->adj_list.upper, 6940 struct netdev_adjacent, list); 6941 if (likely(upper->master) && !upper->ignore) 6942 return upper->dev; 6943 return NULL; 6944 } 6945 6946 /** 6947 * netdev_has_any_lower_dev - Check if device is linked to some device 6948 * @dev: device 6949 * 6950 * Find out if a device is linked to a lower device and return true in case 6951 * it is. The caller must hold the RTNL lock. 6952 */ 6953 static bool netdev_has_any_lower_dev(struct net_device *dev) 6954 { 6955 ASSERT_RTNL(); 6956 6957 return !list_empty(&dev->adj_list.lower); 6958 } 6959 6960 void *netdev_adjacent_get_private(struct list_head *adj_list) 6961 { 6962 struct netdev_adjacent *adj; 6963 6964 adj = list_entry(adj_list, struct netdev_adjacent, list); 6965 6966 return adj->private; 6967 } 6968 EXPORT_SYMBOL(netdev_adjacent_get_private); 6969 6970 /** 6971 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list 6972 * @dev: device 6973 * @iter: list_head ** of the current position 6974 * 6975 * Gets the next device from the dev's upper list, starting from iter 6976 * position. The caller must hold RCU read lock. 6977 */ 6978 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, 6979 struct list_head **iter) 6980 { 6981 struct netdev_adjacent *upper; 6982 6983 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 6984 6985 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 6986 6987 if (&upper->list == &dev->adj_list.upper) 6988 return NULL; 6989 6990 *iter = &upper->list; 6991 6992 return upper->dev; 6993 } 6994 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); 6995 6996 static struct net_device *__netdev_next_upper_dev(struct net_device *dev, 6997 struct list_head **iter, 6998 bool *ignore) 6999 { 7000 struct netdev_adjacent *upper; 7001 7002 upper = list_entry((*iter)->next, struct netdev_adjacent, list); 7003 7004 if (&upper->list == &dev->adj_list.upper) 7005 return NULL; 7006 7007 *iter = &upper->list; 7008 *ignore = upper->ignore; 7009 7010 return upper->dev; 7011 } 7012 7013 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, 7014 struct list_head **iter) 7015 { 7016 struct netdev_adjacent *upper; 7017 7018 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 7019 7020 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7021 7022 if (&upper->list == &dev->adj_list.upper) 7023 return NULL; 7024 7025 *iter = &upper->list; 7026 7027 return upper->dev; 7028 } 7029 7030 static int __netdev_walk_all_upper_dev(struct net_device *dev, 7031 int (*fn)(struct net_device *dev, 7032 struct netdev_nested_priv *priv), 7033 struct netdev_nested_priv *priv) 7034 { 7035 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7036 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7037 int ret, cur = 0; 7038 bool ignore; 7039 7040 now = dev; 7041 iter = &dev->adj_list.upper; 7042 7043 while (1) { 7044 if (now != dev) { 7045 ret = fn(now, priv); 7046 if (ret) 7047 return ret; 7048 } 7049 7050 next = NULL; 7051 while (1) { 7052 udev = __netdev_next_upper_dev(now, &iter, &ignore); 7053 if (!udev) 7054 break; 7055 if (ignore) 7056 continue; 7057 7058 next = udev; 7059 niter = &udev->adj_list.upper; 7060 dev_stack[cur] = now; 7061 iter_stack[cur++] = iter; 7062 break; 7063 } 7064 7065 if (!next) { 7066 if (!cur) 7067 return 0; 7068 next = dev_stack[--cur]; 7069 niter = iter_stack[cur]; 7070 } 7071 7072 now = next; 7073 iter = niter; 7074 } 7075 7076 return 0; 7077 } 7078 7079 int netdev_walk_all_upper_dev_rcu(struct net_device *dev, 7080 int (*fn)(struct net_device *dev, 7081 struct netdev_nested_priv *priv), 7082 struct netdev_nested_priv *priv) 7083 { 7084 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7085 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7086 int ret, cur = 0; 7087 7088 now = dev; 7089 iter = &dev->adj_list.upper; 7090 7091 while (1) { 7092 if (now != dev) { 7093 ret = fn(now, priv); 7094 if (ret) 7095 return ret; 7096 } 7097 7098 next = NULL; 7099 while (1) { 7100 udev = netdev_next_upper_dev_rcu(now, &iter); 7101 if (!udev) 7102 break; 7103 7104 next = udev; 7105 niter = &udev->adj_list.upper; 7106 dev_stack[cur] = now; 7107 iter_stack[cur++] = iter; 7108 break; 7109 } 7110 7111 if (!next) { 7112 if (!cur) 7113 return 0; 7114 next = dev_stack[--cur]; 7115 niter = iter_stack[cur]; 7116 } 7117 7118 now = next; 7119 iter = niter; 7120 } 7121 7122 return 0; 7123 } 7124 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); 7125 7126 static bool __netdev_has_upper_dev(struct net_device *dev, 7127 struct net_device *upper_dev) 7128 { 7129 struct netdev_nested_priv priv = { 7130 .flags = 0, 7131 .data = (void *)upper_dev, 7132 }; 7133 7134 ASSERT_RTNL(); 7135 7136 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev, 7137 &priv); 7138 } 7139 7140 /** 7141 * netdev_lower_get_next_private - Get the next ->private from the 7142 * lower neighbour list 7143 * @dev: device 7144 * @iter: list_head ** of the current position 7145 * 7146 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7147 * list, starting from iter position. The caller must hold either hold the 7148 * RTNL lock or its own locking that guarantees that the neighbour lower 7149 * list will remain unchanged. 7150 */ 7151 void *netdev_lower_get_next_private(struct net_device *dev, 7152 struct list_head **iter) 7153 { 7154 struct netdev_adjacent *lower; 7155 7156 lower = list_entry(*iter, struct netdev_adjacent, list); 7157 7158 if (&lower->list == &dev->adj_list.lower) 7159 return NULL; 7160 7161 *iter = lower->list.next; 7162 7163 return lower->private; 7164 } 7165 EXPORT_SYMBOL(netdev_lower_get_next_private); 7166 7167 /** 7168 * netdev_lower_get_next_private_rcu - Get the next ->private from the 7169 * lower neighbour list, RCU 7170 * variant 7171 * @dev: device 7172 * @iter: list_head ** of the current position 7173 * 7174 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7175 * list, starting from iter position. The caller must hold RCU read lock. 7176 */ 7177 void *netdev_lower_get_next_private_rcu(struct net_device *dev, 7178 struct list_head **iter) 7179 { 7180 struct netdev_adjacent *lower; 7181 7182 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 7183 7184 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7185 7186 if (&lower->list == &dev->adj_list.lower) 7187 return NULL; 7188 7189 *iter = &lower->list; 7190 7191 return lower->private; 7192 } 7193 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); 7194 7195 /** 7196 * netdev_lower_get_next - Get the next device from the lower neighbour 7197 * list 7198 * @dev: device 7199 * @iter: list_head ** of the current position 7200 * 7201 * Gets the next netdev_adjacent from the dev's lower neighbour 7202 * list, starting from iter position. The caller must hold RTNL lock or 7203 * its own locking that guarantees that the neighbour lower 7204 * list will remain unchanged. 7205 */ 7206 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) 7207 { 7208 struct netdev_adjacent *lower; 7209 7210 lower = list_entry(*iter, struct netdev_adjacent, list); 7211 7212 if (&lower->list == &dev->adj_list.lower) 7213 return NULL; 7214 7215 *iter = lower->list.next; 7216 7217 return lower->dev; 7218 } 7219 EXPORT_SYMBOL(netdev_lower_get_next); 7220 7221 static struct net_device *netdev_next_lower_dev(struct net_device *dev, 7222 struct list_head **iter) 7223 { 7224 struct netdev_adjacent *lower; 7225 7226 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7227 7228 if (&lower->list == &dev->adj_list.lower) 7229 return NULL; 7230 7231 *iter = &lower->list; 7232 7233 return lower->dev; 7234 } 7235 7236 static struct net_device *__netdev_next_lower_dev(struct net_device *dev, 7237 struct list_head **iter, 7238 bool *ignore) 7239 { 7240 struct netdev_adjacent *lower; 7241 7242 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7243 7244 if (&lower->list == &dev->adj_list.lower) 7245 return NULL; 7246 7247 *iter = &lower->list; 7248 *ignore = lower->ignore; 7249 7250 return lower->dev; 7251 } 7252 7253 int netdev_walk_all_lower_dev(struct net_device *dev, 7254 int (*fn)(struct net_device *dev, 7255 struct netdev_nested_priv *priv), 7256 struct netdev_nested_priv *priv) 7257 { 7258 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7259 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7260 int ret, cur = 0; 7261 7262 now = dev; 7263 iter = &dev->adj_list.lower; 7264 7265 while (1) { 7266 if (now != dev) { 7267 ret = fn(now, priv); 7268 if (ret) 7269 return ret; 7270 } 7271 7272 next = NULL; 7273 while (1) { 7274 ldev = netdev_next_lower_dev(now, &iter); 7275 if (!ldev) 7276 break; 7277 7278 next = ldev; 7279 niter = &ldev->adj_list.lower; 7280 dev_stack[cur] = now; 7281 iter_stack[cur++] = iter; 7282 break; 7283 } 7284 7285 if (!next) { 7286 if (!cur) 7287 return 0; 7288 next = dev_stack[--cur]; 7289 niter = iter_stack[cur]; 7290 } 7291 7292 now = next; 7293 iter = niter; 7294 } 7295 7296 return 0; 7297 } 7298 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); 7299 7300 static int __netdev_walk_all_lower_dev(struct net_device *dev, 7301 int (*fn)(struct net_device *dev, 7302 struct netdev_nested_priv *priv), 7303 struct netdev_nested_priv *priv) 7304 { 7305 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7306 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7307 int ret, cur = 0; 7308 bool ignore; 7309 7310 now = dev; 7311 iter = &dev->adj_list.lower; 7312 7313 while (1) { 7314 if (now != dev) { 7315 ret = fn(now, priv); 7316 if (ret) 7317 return ret; 7318 } 7319 7320 next = NULL; 7321 while (1) { 7322 ldev = __netdev_next_lower_dev(now, &iter, &ignore); 7323 if (!ldev) 7324 break; 7325 if (ignore) 7326 continue; 7327 7328 next = ldev; 7329 niter = &ldev->adj_list.lower; 7330 dev_stack[cur] = now; 7331 iter_stack[cur++] = iter; 7332 break; 7333 } 7334 7335 if (!next) { 7336 if (!cur) 7337 return 0; 7338 next = dev_stack[--cur]; 7339 niter = iter_stack[cur]; 7340 } 7341 7342 now = next; 7343 iter = niter; 7344 } 7345 7346 return 0; 7347 } 7348 7349 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, 7350 struct list_head **iter) 7351 { 7352 struct netdev_adjacent *lower; 7353 7354 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7355 if (&lower->list == &dev->adj_list.lower) 7356 return NULL; 7357 7358 *iter = &lower->list; 7359 7360 return lower->dev; 7361 } 7362 EXPORT_SYMBOL(netdev_next_lower_dev_rcu); 7363 7364 static u8 __netdev_upper_depth(struct net_device *dev) 7365 { 7366 struct net_device *udev; 7367 struct list_head *iter; 7368 u8 max_depth = 0; 7369 bool ignore; 7370 7371 for (iter = &dev->adj_list.upper, 7372 udev = __netdev_next_upper_dev(dev, &iter, &ignore); 7373 udev; 7374 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) { 7375 if (ignore) 7376 continue; 7377 if (max_depth < udev->upper_level) 7378 max_depth = udev->upper_level; 7379 } 7380 7381 return max_depth; 7382 } 7383 7384 static u8 __netdev_lower_depth(struct net_device *dev) 7385 { 7386 struct net_device *ldev; 7387 struct list_head *iter; 7388 u8 max_depth = 0; 7389 bool ignore; 7390 7391 for (iter = &dev->adj_list.lower, 7392 ldev = __netdev_next_lower_dev(dev, &iter, &ignore); 7393 ldev; 7394 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) { 7395 if (ignore) 7396 continue; 7397 if (max_depth < ldev->lower_level) 7398 max_depth = ldev->lower_level; 7399 } 7400 7401 return max_depth; 7402 } 7403 7404 static int __netdev_update_upper_level(struct net_device *dev, 7405 struct netdev_nested_priv *__unused) 7406 { 7407 dev->upper_level = __netdev_upper_depth(dev) + 1; 7408 return 0; 7409 } 7410 7411 #ifdef CONFIG_LOCKDEP 7412 static LIST_HEAD(net_unlink_list); 7413 7414 static void net_unlink_todo(struct net_device *dev) 7415 { 7416 if (list_empty(&dev->unlink_list)) 7417 list_add_tail(&dev->unlink_list, &net_unlink_list); 7418 } 7419 #endif 7420 7421 static int __netdev_update_lower_level(struct net_device *dev, 7422 struct netdev_nested_priv *priv) 7423 { 7424 dev->lower_level = __netdev_lower_depth(dev) + 1; 7425 7426 #ifdef CONFIG_LOCKDEP 7427 if (!priv) 7428 return 0; 7429 7430 if (priv->flags & NESTED_SYNC_IMM) 7431 dev->nested_level = dev->lower_level - 1; 7432 if (priv->flags & NESTED_SYNC_TODO) 7433 net_unlink_todo(dev); 7434 #endif 7435 return 0; 7436 } 7437 7438 int netdev_walk_all_lower_dev_rcu(struct net_device *dev, 7439 int (*fn)(struct net_device *dev, 7440 struct netdev_nested_priv *priv), 7441 struct netdev_nested_priv *priv) 7442 { 7443 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7444 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7445 int ret, cur = 0; 7446 7447 now = dev; 7448 iter = &dev->adj_list.lower; 7449 7450 while (1) { 7451 if (now != dev) { 7452 ret = fn(now, priv); 7453 if (ret) 7454 return ret; 7455 } 7456 7457 next = NULL; 7458 while (1) { 7459 ldev = netdev_next_lower_dev_rcu(now, &iter); 7460 if (!ldev) 7461 break; 7462 7463 next = ldev; 7464 niter = &ldev->adj_list.lower; 7465 dev_stack[cur] = now; 7466 iter_stack[cur++] = iter; 7467 break; 7468 } 7469 7470 if (!next) { 7471 if (!cur) 7472 return 0; 7473 next = dev_stack[--cur]; 7474 niter = iter_stack[cur]; 7475 } 7476 7477 now = next; 7478 iter = niter; 7479 } 7480 7481 return 0; 7482 } 7483 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); 7484 7485 /** 7486 * netdev_lower_get_first_private_rcu - Get the first ->private from the 7487 * lower neighbour list, RCU 7488 * variant 7489 * @dev: device 7490 * 7491 * Gets the first netdev_adjacent->private from the dev's lower neighbour 7492 * list. The caller must hold RCU read lock. 7493 */ 7494 void *netdev_lower_get_first_private_rcu(struct net_device *dev) 7495 { 7496 struct netdev_adjacent *lower; 7497 7498 lower = list_first_or_null_rcu(&dev->adj_list.lower, 7499 struct netdev_adjacent, list); 7500 if (lower) 7501 return lower->private; 7502 return NULL; 7503 } 7504 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); 7505 7506 /** 7507 * netdev_master_upper_dev_get_rcu - Get master upper device 7508 * @dev: device 7509 * 7510 * Find a master upper device and return pointer to it or NULL in case 7511 * it's not there. The caller must hold the RCU read lock. 7512 */ 7513 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) 7514 { 7515 struct netdev_adjacent *upper; 7516 7517 upper = list_first_or_null_rcu(&dev->adj_list.upper, 7518 struct netdev_adjacent, list); 7519 if (upper && likely(upper->master)) 7520 return upper->dev; 7521 return NULL; 7522 } 7523 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); 7524 7525 static int netdev_adjacent_sysfs_add(struct net_device *dev, 7526 struct net_device *adj_dev, 7527 struct list_head *dev_list) 7528 { 7529 char linkname[IFNAMSIZ+7]; 7530 7531 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7532 "upper_%s" : "lower_%s", adj_dev->name); 7533 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), 7534 linkname); 7535 } 7536 static void netdev_adjacent_sysfs_del(struct net_device *dev, 7537 char *name, 7538 struct list_head *dev_list) 7539 { 7540 char linkname[IFNAMSIZ+7]; 7541 7542 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7543 "upper_%s" : "lower_%s", name); 7544 sysfs_remove_link(&(dev->dev.kobj), linkname); 7545 } 7546 7547 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, 7548 struct net_device *adj_dev, 7549 struct list_head *dev_list) 7550 { 7551 return (dev_list == &dev->adj_list.upper || 7552 dev_list == &dev->adj_list.lower) && 7553 net_eq(dev_net(dev), dev_net(adj_dev)); 7554 } 7555 7556 static int __netdev_adjacent_dev_insert(struct net_device *dev, 7557 struct net_device *adj_dev, 7558 struct list_head *dev_list, 7559 void *private, bool master) 7560 { 7561 struct netdev_adjacent *adj; 7562 int ret; 7563 7564 adj = __netdev_find_adj(adj_dev, dev_list); 7565 7566 if (adj) { 7567 adj->ref_nr += 1; 7568 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", 7569 dev->name, adj_dev->name, adj->ref_nr); 7570 7571 return 0; 7572 } 7573 7574 adj = kmalloc(sizeof(*adj), GFP_KERNEL); 7575 if (!adj) 7576 return -ENOMEM; 7577 7578 adj->dev = adj_dev; 7579 adj->master = master; 7580 adj->ref_nr = 1; 7581 adj->private = private; 7582 adj->ignore = false; 7583 netdev_hold(adj_dev, &adj->dev_tracker, GFP_KERNEL); 7584 7585 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", 7586 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); 7587 7588 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { 7589 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); 7590 if (ret) 7591 goto free_adj; 7592 } 7593 7594 /* Ensure that master link is always the first item in list. */ 7595 if (master) { 7596 ret = sysfs_create_link(&(dev->dev.kobj), 7597 &(adj_dev->dev.kobj), "master"); 7598 if (ret) 7599 goto remove_symlinks; 7600 7601 list_add_rcu(&adj->list, dev_list); 7602 } else { 7603 list_add_tail_rcu(&adj->list, dev_list); 7604 } 7605 7606 return 0; 7607 7608 remove_symlinks: 7609 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7610 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7611 free_adj: 7612 netdev_put(adj_dev, &adj->dev_tracker); 7613 kfree(adj); 7614 7615 return ret; 7616 } 7617 7618 static void __netdev_adjacent_dev_remove(struct net_device *dev, 7619 struct net_device *adj_dev, 7620 u16 ref_nr, 7621 struct list_head *dev_list) 7622 { 7623 struct netdev_adjacent *adj; 7624 7625 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", 7626 dev->name, adj_dev->name, ref_nr); 7627 7628 adj = __netdev_find_adj(adj_dev, dev_list); 7629 7630 if (!adj) { 7631 pr_err("Adjacency does not exist for device %s from %s\n", 7632 dev->name, adj_dev->name); 7633 WARN_ON(1); 7634 return; 7635 } 7636 7637 if (adj->ref_nr > ref_nr) { 7638 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", 7639 dev->name, adj_dev->name, ref_nr, 7640 adj->ref_nr - ref_nr); 7641 adj->ref_nr -= ref_nr; 7642 return; 7643 } 7644 7645 if (adj->master) 7646 sysfs_remove_link(&(dev->dev.kobj), "master"); 7647 7648 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7649 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7650 7651 list_del_rcu(&adj->list); 7652 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", 7653 adj_dev->name, dev->name, adj_dev->name); 7654 netdev_put(adj_dev, &adj->dev_tracker); 7655 kfree_rcu(adj, rcu); 7656 } 7657 7658 static int __netdev_adjacent_dev_link_lists(struct net_device *dev, 7659 struct net_device *upper_dev, 7660 struct list_head *up_list, 7661 struct list_head *down_list, 7662 void *private, bool master) 7663 { 7664 int ret; 7665 7666 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, 7667 private, master); 7668 if (ret) 7669 return ret; 7670 7671 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, 7672 private, false); 7673 if (ret) { 7674 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); 7675 return ret; 7676 } 7677 7678 return 0; 7679 } 7680 7681 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, 7682 struct net_device *upper_dev, 7683 u16 ref_nr, 7684 struct list_head *up_list, 7685 struct list_head *down_list) 7686 { 7687 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); 7688 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); 7689 } 7690 7691 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, 7692 struct net_device *upper_dev, 7693 void *private, bool master) 7694 { 7695 return __netdev_adjacent_dev_link_lists(dev, upper_dev, 7696 &dev->adj_list.upper, 7697 &upper_dev->adj_list.lower, 7698 private, master); 7699 } 7700 7701 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, 7702 struct net_device *upper_dev) 7703 { 7704 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, 7705 &dev->adj_list.upper, 7706 &upper_dev->adj_list.lower); 7707 } 7708 7709 static int __netdev_upper_dev_link(struct net_device *dev, 7710 struct net_device *upper_dev, bool master, 7711 void *upper_priv, void *upper_info, 7712 struct netdev_nested_priv *priv, 7713 struct netlink_ext_ack *extack) 7714 { 7715 struct netdev_notifier_changeupper_info changeupper_info = { 7716 .info = { 7717 .dev = dev, 7718 .extack = extack, 7719 }, 7720 .upper_dev = upper_dev, 7721 .master = master, 7722 .linking = true, 7723 .upper_info = upper_info, 7724 }; 7725 struct net_device *master_dev; 7726 int ret = 0; 7727 7728 ASSERT_RTNL(); 7729 7730 if (dev == upper_dev) 7731 return -EBUSY; 7732 7733 /* To prevent loops, check if dev is not upper device to upper_dev. */ 7734 if (__netdev_has_upper_dev(upper_dev, dev)) 7735 return -EBUSY; 7736 7737 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV) 7738 return -EMLINK; 7739 7740 if (!master) { 7741 if (__netdev_has_upper_dev(dev, upper_dev)) 7742 return -EEXIST; 7743 } else { 7744 master_dev = __netdev_master_upper_dev_get(dev); 7745 if (master_dev) 7746 return master_dev == upper_dev ? -EEXIST : -EBUSY; 7747 } 7748 7749 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7750 &changeupper_info.info); 7751 ret = notifier_to_errno(ret); 7752 if (ret) 7753 return ret; 7754 7755 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, 7756 master); 7757 if (ret) 7758 return ret; 7759 7760 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 7761 &changeupper_info.info); 7762 ret = notifier_to_errno(ret); 7763 if (ret) 7764 goto rollback; 7765 7766 __netdev_update_upper_level(dev, NULL); 7767 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 7768 7769 __netdev_update_lower_level(upper_dev, priv); 7770 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 7771 priv); 7772 7773 return 0; 7774 7775 rollback: 7776 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 7777 7778 return ret; 7779 } 7780 7781 /** 7782 * netdev_upper_dev_link - Add a link to the upper device 7783 * @dev: device 7784 * @upper_dev: new upper device 7785 * @extack: netlink extended ack 7786 * 7787 * Adds a link to device which is upper to this one. The caller must hold 7788 * the RTNL lock. On a failure a negative errno code is returned. 7789 * On success the reference counts are adjusted and the function 7790 * returns zero. 7791 */ 7792 int netdev_upper_dev_link(struct net_device *dev, 7793 struct net_device *upper_dev, 7794 struct netlink_ext_ack *extack) 7795 { 7796 struct netdev_nested_priv priv = { 7797 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7798 .data = NULL, 7799 }; 7800 7801 return __netdev_upper_dev_link(dev, upper_dev, false, 7802 NULL, NULL, &priv, extack); 7803 } 7804 EXPORT_SYMBOL(netdev_upper_dev_link); 7805 7806 /** 7807 * netdev_master_upper_dev_link - Add a master link to the upper device 7808 * @dev: device 7809 * @upper_dev: new upper device 7810 * @upper_priv: upper device private 7811 * @upper_info: upper info to be passed down via notifier 7812 * @extack: netlink extended ack 7813 * 7814 * Adds a link to device which is upper to this one. In this case, only 7815 * one master upper device can be linked, although other non-master devices 7816 * might be linked as well. The caller must hold the RTNL lock. 7817 * On a failure a negative errno code is returned. On success the reference 7818 * counts are adjusted and the function returns zero. 7819 */ 7820 int netdev_master_upper_dev_link(struct net_device *dev, 7821 struct net_device *upper_dev, 7822 void *upper_priv, void *upper_info, 7823 struct netlink_ext_ack *extack) 7824 { 7825 struct netdev_nested_priv priv = { 7826 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7827 .data = NULL, 7828 }; 7829 7830 return __netdev_upper_dev_link(dev, upper_dev, true, 7831 upper_priv, upper_info, &priv, extack); 7832 } 7833 EXPORT_SYMBOL(netdev_master_upper_dev_link); 7834 7835 static void __netdev_upper_dev_unlink(struct net_device *dev, 7836 struct net_device *upper_dev, 7837 struct netdev_nested_priv *priv) 7838 { 7839 struct netdev_notifier_changeupper_info changeupper_info = { 7840 .info = { 7841 .dev = dev, 7842 }, 7843 .upper_dev = upper_dev, 7844 .linking = false, 7845 }; 7846 7847 ASSERT_RTNL(); 7848 7849 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; 7850 7851 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7852 &changeupper_info.info); 7853 7854 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 7855 7856 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 7857 &changeupper_info.info); 7858 7859 __netdev_update_upper_level(dev, NULL); 7860 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 7861 7862 __netdev_update_lower_level(upper_dev, priv); 7863 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 7864 priv); 7865 } 7866 7867 /** 7868 * netdev_upper_dev_unlink - Removes a link to upper device 7869 * @dev: device 7870 * @upper_dev: new upper device 7871 * 7872 * Removes a link to device which is upper to this one. The caller must hold 7873 * the RTNL lock. 7874 */ 7875 void netdev_upper_dev_unlink(struct net_device *dev, 7876 struct net_device *upper_dev) 7877 { 7878 struct netdev_nested_priv priv = { 7879 .flags = NESTED_SYNC_TODO, 7880 .data = NULL, 7881 }; 7882 7883 __netdev_upper_dev_unlink(dev, upper_dev, &priv); 7884 } 7885 EXPORT_SYMBOL(netdev_upper_dev_unlink); 7886 7887 static void __netdev_adjacent_dev_set(struct net_device *upper_dev, 7888 struct net_device *lower_dev, 7889 bool val) 7890 { 7891 struct netdev_adjacent *adj; 7892 7893 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower); 7894 if (adj) 7895 adj->ignore = val; 7896 7897 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper); 7898 if (adj) 7899 adj->ignore = val; 7900 } 7901 7902 static void netdev_adjacent_dev_disable(struct net_device *upper_dev, 7903 struct net_device *lower_dev) 7904 { 7905 __netdev_adjacent_dev_set(upper_dev, lower_dev, true); 7906 } 7907 7908 static void netdev_adjacent_dev_enable(struct net_device *upper_dev, 7909 struct net_device *lower_dev) 7910 { 7911 __netdev_adjacent_dev_set(upper_dev, lower_dev, false); 7912 } 7913 7914 int netdev_adjacent_change_prepare(struct net_device *old_dev, 7915 struct net_device *new_dev, 7916 struct net_device *dev, 7917 struct netlink_ext_ack *extack) 7918 { 7919 struct netdev_nested_priv priv = { 7920 .flags = 0, 7921 .data = NULL, 7922 }; 7923 int err; 7924 7925 if (!new_dev) 7926 return 0; 7927 7928 if (old_dev && new_dev != old_dev) 7929 netdev_adjacent_dev_disable(dev, old_dev); 7930 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv, 7931 extack); 7932 if (err) { 7933 if (old_dev && new_dev != old_dev) 7934 netdev_adjacent_dev_enable(dev, old_dev); 7935 return err; 7936 } 7937 7938 return 0; 7939 } 7940 EXPORT_SYMBOL(netdev_adjacent_change_prepare); 7941 7942 void netdev_adjacent_change_commit(struct net_device *old_dev, 7943 struct net_device *new_dev, 7944 struct net_device *dev) 7945 { 7946 struct netdev_nested_priv priv = { 7947 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7948 .data = NULL, 7949 }; 7950 7951 if (!new_dev || !old_dev) 7952 return; 7953 7954 if (new_dev == old_dev) 7955 return; 7956 7957 netdev_adjacent_dev_enable(dev, old_dev); 7958 __netdev_upper_dev_unlink(old_dev, dev, &priv); 7959 } 7960 EXPORT_SYMBOL(netdev_adjacent_change_commit); 7961 7962 void netdev_adjacent_change_abort(struct net_device *old_dev, 7963 struct net_device *new_dev, 7964 struct net_device *dev) 7965 { 7966 struct netdev_nested_priv priv = { 7967 .flags = 0, 7968 .data = NULL, 7969 }; 7970 7971 if (!new_dev) 7972 return; 7973 7974 if (old_dev && new_dev != old_dev) 7975 netdev_adjacent_dev_enable(dev, old_dev); 7976 7977 __netdev_upper_dev_unlink(new_dev, dev, &priv); 7978 } 7979 EXPORT_SYMBOL(netdev_adjacent_change_abort); 7980 7981 /** 7982 * netdev_bonding_info_change - Dispatch event about slave change 7983 * @dev: device 7984 * @bonding_info: info to dispatch 7985 * 7986 * Send NETDEV_BONDING_INFO to netdev notifiers with info. 7987 * The caller must hold the RTNL lock. 7988 */ 7989 void netdev_bonding_info_change(struct net_device *dev, 7990 struct netdev_bonding_info *bonding_info) 7991 { 7992 struct netdev_notifier_bonding_info info = { 7993 .info.dev = dev, 7994 }; 7995 7996 memcpy(&info.bonding_info, bonding_info, 7997 sizeof(struct netdev_bonding_info)); 7998 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, 7999 &info.info); 8000 } 8001 EXPORT_SYMBOL(netdev_bonding_info_change); 8002 8003 static int netdev_offload_xstats_enable_l3(struct net_device *dev, 8004 struct netlink_ext_ack *extack) 8005 { 8006 struct netdev_notifier_offload_xstats_info info = { 8007 .info.dev = dev, 8008 .info.extack = extack, 8009 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, 8010 }; 8011 int err; 8012 int rc; 8013 8014 dev->offload_xstats_l3 = kzalloc(sizeof(*dev->offload_xstats_l3), 8015 GFP_KERNEL); 8016 if (!dev->offload_xstats_l3) 8017 return -ENOMEM; 8018 8019 rc = call_netdevice_notifiers_info_robust(NETDEV_OFFLOAD_XSTATS_ENABLE, 8020 NETDEV_OFFLOAD_XSTATS_DISABLE, 8021 &info.info); 8022 err = notifier_to_errno(rc); 8023 if (err) 8024 goto free_stats; 8025 8026 return 0; 8027 8028 free_stats: 8029 kfree(dev->offload_xstats_l3); 8030 dev->offload_xstats_l3 = NULL; 8031 return err; 8032 } 8033 8034 int netdev_offload_xstats_enable(struct net_device *dev, 8035 enum netdev_offload_xstats_type type, 8036 struct netlink_ext_ack *extack) 8037 { 8038 ASSERT_RTNL(); 8039 8040 if (netdev_offload_xstats_enabled(dev, type)) 8041 return -EALREADY; 8042 8043 switch (type) { 8044 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 8045 return netdev_offload_xstats_enable_l3(dev, extack); 8046 } 8047 8048 WARN_ON(1); 8049 return -EINVAL; 8050 } 8051 EXPORT_SYMBOL(netdev_offload_xstats_enable); 8052 8053 static void netdev_offload_xstats_disable_l3(struct net_device *dev) 8054 { 8055 struct netdev_notifier_offload_xstats_info info = { 8056 .info.dev = dev, 8057 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, 8058 }; 8059 8060 call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_DISABLE, 8061 &info.info); 8062 kfree(dev->offload_xstats_l3); 8063 dev->offload_xstats_l3 = NULL; 8064 } 8065 8066 int netdev_offload_xstats_disable(struct net_device *dev, 8067 enum netdev_offload_xstats_type type) 8068 { 8069 ASSERT_RTNL(); 8070 8071 if (!netdev_offload_xstats_enabled(dev, type)) 8072 return -EALREADY; 8073 8074 switch (type) { 8075 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 8076 netdev_offload_xstats_disable_l3(dev); 8077 return 0; 8078 } 8079 8080 WARN_ON(1); 8081 return -EINVAL; 8082 } 8083 EXPORT_SYMBOL(netdev_offload_xstats_disable); 8084 8085 static void netdev_offload_xstats_disable_all(struct net_device *dev) 8086 { 8087 netdev_offload_xstats_disable(dev, NETDEV_OFFLOAD_XSTATS_TYPE_L3); 8088 } 8089 8090 static struct rtnl_hw_stats64 * 8091 netdev_offload_xstats_get_ptr(const struct net_device *dev, 8092 enum netdev_offload_xstats_type type) 8093 { 8094 switch (type) { 8095 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 8096 return dev->offload_xstats_l3; 8097 } 8098 8099 WARN_ON(1); 8100 return NULL; 8101 } 8102 8103 bool netdev_offload_xstats_enabled(const struct net_device *dev, 8104 enum netdev_offload_xstats_type type) 8105 { 8106 ASSERT_RTNL(); 8107 8108 return netdev_offload_xstats_get_ptr(dev, type); 8109 } 8110 EXPORT_SYMBOL(netdev_offload_xstats_enabled); 8111 8112 struct netdev_notifier_offload_xstats_ru { 8113 bool used; 8114 }; 8115 8116 struct netdev_notifier_offload_xstats_rd { 8117 struct rtnl_hw_stats64 stats; 8118 bool used; 8119 }; 8120 8121 static void netdev_hw_stats64_add(struct rtnl_hw_stats64 *dest, 8122 const struct rtnl_hw_stats64 *src) 8123 { 8124 dest->rx_packets += src->rx_packets; 8125 dest->tx_packets += src->tx_packets; 8126 dest->rx_bytes += src->rx_bytes; 8127 dest->tx_bytes += src->tx_bytes; 8128 dest->rx_errors += src->rx_errors; 8129 dest->tx_errors += src->tx_errors; 8130 dest->rx_dropped += src->rx_dropped; 8131 dest->tx_dropped += src->tx_dropped; 8132 dest->multicast += src->multicast; 8133 } 8134 8135 static int netdev_offload_xstats_get_used(struct net_device *dev, 8136 enum netdev_offload_xstats_type type, 8137 bool *p_used, 8138 struct netlink_ext_ack *extack) 8139 { 8140 struct netdev_notifier_offload_xstats_ru report_used = {}; 8141 struct netdev_notifier_offload_xstats_info info = { 8142 .info.dev = dev, 8143 .info.extack = extack, 8144 .type = type, 8145 .report_used = &report_used, 8146 }; 8147 int rc; 8148 8149 WARN_ON(!netdev_offload_xstats_enabled(dev, type)); 8150 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_USED, 8151 &info.info); 8152 *p_used = report_used.used; 8153 return notifier_to_errno(rc); 8154 } 8155 8156 static int netdev_offload_xstats_get_stats(struct net_device *dev, 8157 enum netdev_offload_xstats_type type, 8158 struct rtnl_hw_stats64 *p_stats, 8159 bool *p_used, 8160 struct netlink_ext_ack *extack) 8161 { 8162 struct netdev_notifier_offload_xstats_rd report_delta = {}; 8163 struct netdev_notifier_offload_xstats_info info = { 8164 .info.dev = dev, 8165 .info.extack = extack, 8166 .type = type, 8167 .report_delta = &report_delta, 8168 }; 8169 struct rtnl_hw_stats64 *stats; 8170 int rc; 8171 8172 stats = netdev_offload_xstats_get_ptr(dev, type); 8173 if (WARN_ON(!stats)) 8174 return -EINVAL; 8175 8176 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_DELTA, 8177 &info.info); 8178 8179 /* Cache whatever we got, even if there was an error, otherwise the 8180 * successful stats retrievals would get lost. 8181 */ 8182 netdev_hw_stats64_add(stats, &report_delta.stats); 8183 8184 if (p_stats) 8185 *p_stats = *stats; 8186 *p_used = report_delta.used; 8187 8188 return notifier_to_errno(rc); 8189 } 8190 8191 int netdev_offload_xstats_get(struct net_device *dev, 8192 enum netdev_offload_xstats_type type, 8193 struct rtnl_hw_stats64 *p_stats, bool *p_used, 8194 struct netlink_ext_ack *extack) 8195 { 8196 ASSERT_RTNL(); 8197 8198 if (p_stats) 8199 return netdev_offload_xstats_get_stats(dev, type, p_stats, 8200 p_used, extack); 8201 else 8202 return netdev_offload_xstats_get_used(dev, type, p_used, 8203 extack); 8204 } 8205 EXPORT_SYMBOL(netdev_offload_xstats_get); 8206 8207 void 8208 netdev_offload_xstats_report_delta(struct netdev_notifier_offload_xstats_rd *report_delta, 8209 const struct rtnl_hw_stats64 *stats) 8210 { 8211 report_delta->used = true; 8212 netdev_hw_stats64_add(&report_delta->stats, stats); 8213 } 8214 EXPORT_SYMBOL(netdev_offload_xstats_report_delta); 8215 8216 void 8217 netdev_offload_xstats_report_used(struct netdev_notifier_offload_xstats_ru *report_used) 8218 { 8219 report_used->used = true; 8220 } 8221 EXPORT_SYMBOL(netdev_offload_xstats_report_used); 8222 8223 void netdev_offload_xstats_push_delta(struct net_device *dev, 8224 enum netdev_offload_xstats_type type, 8225 const struct rtnl_hw_stats64 *p_stats) 8226 { 8227 struct rtnl_hw_stats64 *stats; 8228 8229 ASSERT_RTNL(); 8230 8231 stats = netdev_offload_xstats_get_ptr(dev, type); 8232 if (WARN_ON(!stats)) 8233 return; 8234 8235 netdev_hw_stats64_add(stats, p_stats); 8236 } 8237 EXPORT_SYMBOL(netdev_offload_xstats_push_delta); 8238 8239 /** 8240 * netdev_get_xmit_slave - Get the xmit slave of master device 8241 * @dev: device 8242 * @skb: The packet 8243 * @all_slaves: assume all the slaves are active 8244 * 8245 * The reference counters are not incremented so the caller must be 8246 * careful with locks. The caller must hold RCU lock. 8247 * %NULL is returned if no slave is found. 8248 */ 8249 8250 struct net_device *netdev_get_xmit_slave(struct net_device *dev, 8251 struct sk_buff *skb, 8252 bool all_slaves) 8253 { 8254 const struct net_device_ops *ops = dev->netdev_ops; 8255 8256 if (!ops->ndo_get_xmit_slave) 8257 return NULL; 8258 return ops->ndo_get_xmit_slave(dev, skb, all_slaves); 8259 } 8260 EXPORT_SYMBOL(netdev_get_xmit_slave); 8261 8262 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev, 8263 struct sock *sk) 8264 { 8265 const struct net_device_ops *ops = dev->netdev_ops; 8266 8267 if (!ops->ndo_sk_get_lower_dev) 8268 return NULL; 8269 return ops->ndo_sk_get_lower_dev(dev, sk); 8270 } 8271 8272 /** 8273 * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket 8274 * @dev: device 8275 * @sk: the socket 8276 * 8277 * %NULL is returned if no lower device is found. 8278 */ 8279 8280 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev, 8281 struct sock *sk) 8282 { 8283 struct net_device *lower; 8284 8285 lower = netdev_sk_get_lower_dev(dev, sk); 8286 while (lower) { 8287 dev = lower; 8288 lower = netdev_sk_get_lower_dev(dev, sk); 8289 } 8290 8291 return dev; 8292 } 8293 EXPORT_SYMBOL(netdev_sk_get_lowest_dev); 8294 8295 static void netdev_adjacent_add_links(struct net_device *dev) 8296 { 8297 struct netdev_adjacent *iter; 8298 8299 struct net *net = dev_net(dev); 8300 8301 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8302 if (!net_eq(net, dev_net(iter->dev))) 8303 continue; 8304 netdev_adjacent_sysfs_add(iter->dev, dev, 8305 &iter->dev->adj_list.lower); 8306 netdev_adjacent_sysfs_add(dev, iter->dev, 8307 &dev->adj_list.upper); 8308 } 8309 8310 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8311 if (!net_eq(net, dev_net(iter->dev))) 8312 continue; 8313 netdev_adjacent_sysfs_add(iter->dev, dev, 8314 &iter->dev->adj_list.upper); 8315 netdev_adjacent_sysfs_add(dev, iter->dev, 8316 &dev->adj_list.lower); 8317 } 8318 } 8319 8320 static void netdev_adjacent_del_links(struct net_device *dev) 8321 { 8322 struct netdev_adjacent *iter; 8323 8324 struct net *net = dev_net(dev); 8325 8326 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8327 if (!net_eq(net, dev_net(iter->dev))) 8328 continue; 8329 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8330 &iter->dev->adj_list.lower); 8331 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8332 &dev->adj_list.upper); 8333 } 8334 8335 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8336 if (!net_eq(net, dev_net(iter->dev))) 8337 continue; 8338 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8339 &iter->dev->adj_list.upper); 8340 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8341 &dev->adj_list.lower); 8342 } 8343 } 8344 8345 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) 8346 { 8347 struct netdev_adjacent *iter; 8348 8349 struct net *net = dev_net(dev); 8350 8351 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8352 if (!net_eq(net, dev_net(iter->dev))) 8353 continue; 8354 netdev_adjacent_sysfs_del(iter->dev, oldname, 8355 &iter->dev->adj_list.lower); 8356 netdev_adjacent_sysfs_add(iter->dev, dev, 8357 &iter->dev->adj_list.lower); 8358 } 8359 8360 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8361 if (!net_eq(net, dev_net(iter->dev))) 8362 continue; 8363 netdev_adjacent_sysfs_del(iter->dev, oldname, 8364 &iter->dev->adj_list.upper); 8365 netdev_adjacent_sysfs_add(iter->dev, dev, 8366 &iter->dev->adj_list.upper); 8367 } 8368 } 8369 8370 void *netdev_lower_dev_get_private(struct net_device *dev, 8371 struct net_device *lower_dev) 8372 { 8373 struct netdev_adjacent *lower; 8374 8375 if (!lower_dev) 8376 return NULL; 8377 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); 8378 if (!lower) 8379 return NULL; 8380 8381 return lower->private; 8382 } 8383 EXPORT_SYMBOL(netdev_lower_dev_get_private); 8384 8385 8386 /** 8387 * netdev_lower_state_changed - Dispatch event about lower device state change 8388 * @lower_dev: device 8389 * @lower_state_info: state to dispatch 8390 * 8391 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. 8392 * The caller must hold the RTNL lock. 8393 */ 8394 void netdev_lower_state_changed(struct net_device *lower_dev, 8395 void *lower_state_info) 8396 { 8397 struct netdev_notifier_changelowerstate_info changelowerstate_info = { 8398 .info.dev = lower_dev, 8399 }; 8400 8401 ASSERT_RTNL(); 8402 changelowerstate_info.lower_state_info = lower_state_info; 8403 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, 8404 &changelowerstate_info.info); 8405 } 8406 EXPORT_SYMBOL(netdev_lower_state_changed); 8407 8408 static void dev_change_rx_flags(struct net_device *dev, int flags) 8409 { 8410 const struct net_device_ops *ops = dev->netdev_ops; 8411 8412 if (ops->ndo_change_rx_flags) 8413 ops->ndo_change_rx_flags(dev, flags); 8414 } 8415 8416 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) 8417 { 8418 unsigned int old_flags = dev->flags; 8419 kuid_t uid; 8420 kgid_t gid; 8421 8422 ASSERT_RTNL(); 8423 8424 dev->flags |= IFF_PROMISC; 8425 dev->promiscuity += inc; 8426 if (dev->promiscuity == 0) { 8427 /* 8428 * Avoid overflow. 8429 * If inc causes overflow, untouch promisc and return error. 8430 */ 8431 if (inc < 0) 8432 dev->flags &= ~IFF_PROMISC; 8433 else { 8434 dev->promiscuity -= inc; 8435 netdev_warn(dev, "promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n"); 8436 return -EOVERFLOW; 8437 } 8438 } 8439 if (dev->flags != old_flags) { 8440 netdev_info(dev, "%s promiscuous mode\n", 8441 dev->flags & IFF_PROMISC ? "entered" : "left"); 8442 if (audit_enabled) { 8443 current_uid_gid(&uid, &gid); 8444 audit_log(audit_context(), GFP_ATOMIC, 8445 AUDIT_ANOM_PROMISCUOUS, 8446 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", 8447 dev->name, (dev->flags & IFF_PROMISC), 8448 (old_flags & IFF_PROMISC), 8449 from_kuid(&init_user_ns, audit_get_loginuid(current)), 8450 from_kuid(&init_user_ns, uid), 8451 from_kgid(&init_user_ns, gid), 8452 audit_get_sessionid(current)); 8453 } 8454 8455 dev_change_rx_flags(dev, IFF_PROMISC); 8456 } 8457 if (notify) 8458 __dev_notify_flags(dev, old_flags, IFF_PROMISC, 0, NULL); 8459 return 0; 8460 } 8461 8462 /** 8463 * dev_set_promiscuity - update promiscuity count on a device 8464 * @dev: device 8465 * @inc: modifier 8466 * 8467 * Add or remove promiscuity from a device. While the count in the device 8468 * remains above zero the interface remains promiscuous. Once it hits zero 8469 * the device reverts back to normal filtering operation. A negative inc 8470 * value is used to drop promiscuity on the device. 8471 * Return 0 if successful or a negative errno code on error. 8472 */ 8473 int dev_set_promiscuity(struct net_device *dev, int inc) 8474 { 8475 unsigned int old_flags = dev->flags; 8476 int err; 8477 8478 err = __dev_set_promiscuity(dev, inc, true); 8479 if (err < 0) 8480 return err; 8481 if (dev->flags != old_flags) 8482 dev_set_rx_mode(dev); 8483 return err; 8484 } 8485 EXPORT_SYMBOL(dev_set_promiscuity); 8486 8487 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) 8488 { 8489 unsigned int old_flags = dev->flags, old_gflags = dev->gflags; 8490 8491 ASSERT_RTNL(); 8492 8493 dev->flags |= IFF_ALLMULTI; 8494 dev->allmulti += inc; 8495 if (dev->allmulti == 0) { 8496 /* 8497 * Avoid overflow. 8498 * If inc causes overflow, untouch allmulti and return error. 8499 */ 8500 if (inc < 0) 8501 dev->flags &= ~IFF_ALLMULTI; 8502 else { 8503 dev->allmulti -= inc; 8504 netdev_warn(dev, "allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n"); 8505 return -EOVERFLOW; 8506 } 8507 } 8508 if (dev->flags ^ old_flags) { 8509 netdev_info(dev, "%s allmulticast mode\n", 8510 dev->flags & IFF_ALLMULTI ? "entered" : "left"); 8511 dev_change_rx_flags(dev, IFF_ALLMULTI); 8512 dev_set_rx_mode(dev); 8513 if (notify) 8514 __dev_notify_flags(dev, old_flags, 8515 dev->gflags ^ old_gflags, 0, NULL); 8516 } 8517 return 0; 8518 } 8519 8520 /** 8521 * dev_set_allmulti - update allmulti count on a device 8522 * @dev: device 8523 * @inc: modifier 8524 * 8525 * Add or remove reception of all multicast frames to a device. While the 8526 * count in the device remains above zero the interface remains listening 8527 * to all interfaces. Once it hits zero the device reverts back to normal 8528 * filtering operation. A negative @inc value is used to drop the counter 8529 * when releasing a resource needing all multicasts. 8530 * Return 0 if successful or a negative errno code on error. 8531 */ 8532 8533 int dev_set_allmulti(struct net_device *dev, int inc) 8534 { 8535 return __dev_set_allmulti(dev, inc, true); 8536 } 8537 EXPORT_SYMBOL(dev_set_allmulti); 8538 8539 /* 8540 * Upload unicast and multicast address lists to device and 8541 * configure RX filtering. When the device doesn't support unicast 8542 * filtering it is put in promiscuous mode while unicast addresses 8543 * are present. 8544 */ 8545 void __dev_set_rx_mode(struct net_device *dev) 8546 { 8547 const struct net_device_ops *ops = dev->netdev_ops; 8548 8549 /* dev_open will call this function so the list will stay sane. */ 8550 if (!(dev->flags&IFF_UP)) 8551 return; 8552 8553 if (!netif_device_present(dev)) 8554 return; 8555 8556 if (!(dev->priv_flags & IFF_UNICAST_FLT)) { 8557 /* Unicast addresses changes may only happen under the rtnl, 8558 * therefore calling __dev_set_promiscuity here is safe. 8559 */ 8560 if (!netdev_uc_empty(dev) && !dev->uc_promisc) { 8561 __dev_set_promiscuity(dev, 1, false); 8562 dev->uc_promisc = true; 8563 } else if (netdev_uc_empty(dev) && dev->uc_promisc) { 8564 __dev_set_promiscuity(dev, -1, false); 8565 dev->uc_promisc = false; 8566 } 8567 } 8568 8569 if (ops->ndo_set_rx_mode) 8570 ops->ndo_set_rx_mode(dev); 8571 } 8572 8573 void dev_set_rx_mode(struct net_device *dev) 8574 { 8575 netif_addr_lock_bh(dev); 8576 __dev_set_rx_mode(dev); 8577 netif_addr_unlock_bh(dev); 8578 } 8579 8580 /** 8581 * dev_get_flags - get flags reported to userspace 8582 * @dev: device 8583 * 8584 * Get the combination of flag bits exported through APIs to userspace. 8585 */ 8586 unsigned int dev_get_flags(const struct net_device *dev) 8587 { 8588 unsigned int flags; 8589 8590 flags = (dev->flags & ~(IFF_PROMISC | 8591 IFF_ALLMULTI | 8592 IFF_RUNNING | 8593 IFF_LOWER_UP | 8594 IFF_DORMANT)) | 8595 (dev->gflags & (IFF_PROMISC | 8596 IFF_ALLMULTI)); 8597 8598 if (netif_running(dev)) { 8599 if (netif_oper_up(dev)) 8600 flags |= IFF_RUNNING; 8601 if (netif_carrier_ok(dev)) 8602 flags |= IFF_LOWER_UP; 8603 if (netif_dormant(dev)) 8604 flags |= IFF_DORMANT; 8605 } 8606 8607 return flags; 8608 } 8609 EXPORT_SYMBOL(dev_get_flags); 8610 8611 int __dev_change_flags(struct net_device *dev, unsigned int flags, 8612 struct netlink_ext_ack *extack) 8613 { 8614 unsigned int old_flags = dev->flags; 8615 int ret; 8616 8617 ASSERT_RTNL(); 8618 8619 /* 8620 * Set the flags on our device. 8621 */ 8622 8623 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | 8624 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | 8625 IFF_AUTOMEDIA)) | 8626 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | 8627 IFF_ALLMULTI)); 8628 8629 /* 8630 * Load in the correct multicast list now the flags have changed. 8631 */ 8632 8633 if ((old_flags ^ flags) & IFF_MULTICAST) 8634 dev_change_rx_flags(dev, IFF_MULTICAST); 8635 8636 dev_set_rx_mode(dev); 8637 8638 /* 8639 * Have we downed the interface. We handle IFF_UP ourselves 8640 * according to user attempts to set it, rather than blindly 8641 * setting it. 8642 */ 8643 8644 ret = 0; 8645 if ((old_flags ^ flags) & IFF_UP) { 8646 if (old_flags & IFF_UP) 8647 __dev_close(dev); 8648 else 8649 ret = __dev_open(dev, extack); 8650 } 8651 8652 if ((flags ^ dev->gflags) & IFF_PROMISC) { 8653 int inc = (flags & IFF_PROMISC) ? 1 : -1; 8654 unsigned int old_flags = dev->flags; 8655 8656 dev->gflags ^= IFF_PROMISC; 8657 8658 if (__dev_set_promiscuity(dev, inc, false) >= 0) 8659 if (dev->flags != old_flags) 8660 dev_set_rx_mode(dev); 8661 } 8662 8663 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI 8664 * is important. Some (broken) drivers set IFF_PROMISC, when 8665 * IFF_ALLMULTI is requested not asking us and not reporting. 8666 */ 8667 if ((flags ^ dev->gflags) & IFF_ALLMULTI) { 8668 int inc = (flags & IFF_ALLMULTI) ? 1 : -1; 8669 8670 dev->gflags ^= IFF_ALLMULTI; 8671 __dev_set_allmulti(dev, inc, false); 8672 } 8673 8674 return ret; 8675 } 8676 8677 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, 8678 unsigned int gchanges, u32 portid, 8679 const struct nlmsghdr *nlh) 8680 { 8681 unsigned int changes = dev->flags ^ old_flags; 8682 8683 if (gchanges) 8684 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC, portid, nlh); 8685 8686 if (changes & IFF_UP) { 8687 if (dev->flags & IFF_UP) 8688 call_netdevice_notifiers(NETDEV_UP, dev); 8689 else 8690 call_netdevice_notifiers(NETDEV_DOWN, dev); 8691 } 8692 8693 if (dev->flags & IFF_UP && 8694 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { 8695 struct netdev_notifier_change_info change_info = { 8696 .info = { 8697 .dev = dev, 8698 }, 8699 .flags_changed = changes, 8700 }; 8701 8702 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info); 8703 } 8704 } 8705 8706 /** 8707 * dev_change_flags - change device settings 8708 * @dev: device 8709 * @flags: device state flags 8710 * @extack: netlink extended ack 8711 * 8712 * Change settings on device based state flags. The flags are 8713 * in the userspace exported format. 8714 */ 8715 int dev_change_flags(struct net_device *dev, unsigned int flags, 8716 struct netlink_ext_ack *extack) 8717 { 8718 int ret; 8719 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; 8720 8721 ret = __dev_change_flags(dev, flags, extack); 8722 if (ret < 0) 8723 return ret; 8724 8725 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); 8726 __dev_notify_flags(dev, old_flags, changes, 0, NULL); 8727 return ret; 8728 } 8729 EXPORT_SYMBOL(dev_change_flags); 8730 8731 int __dev_set_mtu(struct net_device *dev, int new_mtu) 8732 { 8733 const struct net_device_ops *ops = dev->netdev_ops; 8734 8735 if (ops->ndo_change_mtu) 8736 return ops->ndo_change_mtu(dev, new_mtu); 8737 8738 /* Pairs with all the lockless reads of dev->mtu in the stack */ 8739 WRITE_ONCE(dev->mtu, new_mtu); 8740 return 0; 8741 } 8742 EXPORT_SYMBOL(__dev_set_mtu); 8743 8744 int dev_validate_mtu(struct net_device *dev, int new_mtu, 8745 struct netlink_ext_ack *extack) 8746 { 8747 /* MTU must be positive, and in range */ 8748 if (new_mtu < 0 || new_mtu < dev->min_mtu) { 8749 NL_SET_ERR_MSG(extack, "mtu less than device minimum"); 8750 return -EINVAL; 8751 } 8752 8753 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { 8754 NL_SET_ERR_MSG(extack, "mtu greater than device maximum"); 8755 return -EINVAL; 8756 } 8757 return 0; 8758 } 8759 8760 /** 8761 * dev_set_mtu_ext - Change maximum transfer unit 8762 * @dev: device 8763 * @new_mtu: new transfer unit 8764 * @extack: netlink extended ack 8765 * 8766 * Change the maximum transfer size of the network device. 8767 */ 8768 int dev_set_mtu_ext(struct net_device *dev, int new_mtu, 8769 struct netlink_ext_ack *extack) 8770 { 8771 int err, orig_mtu; 8772 8773 if (new_mtu == dev->mtu) 8774 return 0; 8775 8776 err = dev_validate_mtu(dev, new_mtu, extack); 8777 if (err) 8778 return err; 8779 8780 if (!netif_device_present(dev)) 8781 return -ENODEV; 8782 8783 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); 8784 err = notifier_to_errno(err); 8785 if (err) 8786 return err; 8787 8788 orig_mtu = dev->mtu; 8789 err = __dev_set_mtu(dev, new_mtu); 8790 8791 if (!err) { 8792 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8793 orig_mtu); 8794 err = notifier_to_errno(err); 8795 if (err) { 8796 /* setting mtu back and notifying everyone again, 8797 * so that they have a chance to revert changes. 8798 */ 8799 __dev_set_mtu(dev, orig_mtu); 8800 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8801 new_mtu); 8802 } 8803 } 8804 return err; 8805 } 8806 8807 int dev_set_mtu(struct net_device *dev, int new_mtu) 8808 { 8809 struct netlink_ext_ack extack; 8810 int err; 8811 8812 memset(&extack, 0, sizeof(extack)); 8813 err = dev_set_mtu_ext(dev, new_mtu, &extack); 8814 if (err && extack._msg) 8815 net_err_ratelimited("%s: %s\n", dev->name, extack._msg); 8816 return err; 8817 } 8818 EXPORT_SYMBOL(dev_set_mtu); 8819 8820 /** 8821 * dev_change_tx_queue_len - Change TX queue length of a netdevice 8822 * @dev: device 8823 * @new_len: new tx queue length 8824 */ 8825 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len) 8826 { 8827 unsigned int orig_len = dev->tx_queue_len; 8828 int res; 8829 8830 if (new_len != (unsigned int)new_len) 8831 return -ERANGE; 8832 8833 if (new_len != orig_len) { 8834 dev->tx_queue_len = new_len; 8835 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev); 8836 res = notifier_to_errno(res); 8837 if (res) 8838 goto err_rollback; 8839 res = dev_qdisc_change_tx_queue_len(dev); 8840 if (res) 8841 goto err_rollback; 8842 } 8843 8844 return 0; 8845 8846 err_rollback: 8847 netdev_err(dev, "refused to change device tx_queue_len\n"); 8848 dev->tx_queue_len = orig_len; 8849 return res; 8850 } 8851 8852 /** 8853 * dev_set_group - Change group this device belongs to 8854 * @dev: device 8855 * @new_group: group this device should belong to 8856 */ 8857 void dev_set_group(struct net_device *dev, int new_group) 8858 { 8859 dev->group = new_group; 8860 } 8861 8862 /** 8863 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR. 8864 * @dev: device 8865 * @addr: new address 8866 * @extack: netlink extended ack 8867 */ 8868 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr, 8869 struct netlink_ext_ack *extack) 8870 { 8871 struct netdev_notifier_pre_changeaddr_info info = { 8872 .info.dev = dev, 8873 .info.extack = extack, 8874 .dev_addr = addr, 8875 }; 8876 int rc; 8877 8878 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info); 8879 return notifier_to_errno(rc); 8880 } 8881 EXPORT_SYMBOL(dev_pre_changeaddr_notify); 8882 8883 /** 8884 * dev_set_mac_address - Change Media Access Control Address 8885 * @dev: device 8886 * @sa: new address 8887 * @extack: netlink extended ack 8888 * 8889 * Change the hardware (MAC) address of the device 8890 */ 8891 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa, 8892 struct netlink_ext_ack *extack) 8893 { 8894 const struct net_device_ops *ops = dev->netdev_ops; 8895 int err; 8896 8897 if (!ops->ndo_set_mac_address) 8898 return -EOPNOTSUPP; 8899 if (sa->sa_family != dev->type) 8900 return -EINVAL; 8901 if (!netif_device_present(dev)) 8902 return -ENODEV; 8903 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack); 8904 if (err) 8905 return err; 8906 if (memcmp(dev->dev_addr, sa->sa_data, dev->addr_len)) { 8907 err = ops->ndo_set_mac_address(dev, sa); 8908 if (err) 8909 return err; 8910 } 8911 dev->addr_assign_type = NET_ADDR_SET; 8912 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); 8913 add_device_randomness(dev->dev_addr, dev->addr_len); 8914 return 0; 8915 } 8916 EXPORT_SYMBOL(dev_set_mac_address); 8917 8918 static DECLARE_RWSEM(dev_addr_sem); 8919 8920 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa, 8921 struct netlink_ext_ack *extack) 8922 { 8923 int ret; 8924 8925 down_write(&dev_addr_sem); 8926 ret = dev_set_mac_address(dev, sa, extack); 8927 up_write(&dev_addr_sem); 8928 return ret; 8929 } 8930 EXPORT_SYMBOL(dev_set_mac_address_user); 8931 8932 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name) 8933 { 8934 size_t size = sizeof(sa->sa_data_min); 8935 struct net_device *dev; 8936 int ret = 0; 8937 8938 down_read(&dev_addr_sem); 8939 rcu_read_lock(); 8940 8941 dev = dev_get_by_name_rcu(net, dev_name); 8942 if (!dev) { 8943 ret = -ENODEV; 8944 goto unlock; 8945 } 8946 if (!dev->addr_len) 8947 memset(sa->sa_data, 0, size); 8948 else 8949 memcpy(sa->sa_data, dev->dev_addr, 8950 min_t(size_t, size, dev->addr_len)); 8951 sa->sa_family = dev->type; 8952 8953 unlock: 8954 rcu_read_unlock(); 8955 up_read(&dev_addr_sem); 8956 return ret; 8957 } 8958 EXPORT_SYMBOL(dev_get_mac_address); 8959 8960 /** 8961 * dev_change_carrier - Change device carrier 8962 * @dev: device 8963 * @new_carrier: new value 8964 * 8965 * Change device carrier 8966 */ 8967 int dev_change_carrier(struct net_device *dev, bool new_carrier) 8968 { 8969 const struct net_device_ops *ops = dev->netdev_ops; 8970 8971 if (!ops->ndo_change_carrier) 8972 return -EOPNOTSUPP; 8973 if (!netif_device_present(dev)) 8974 return -ENODEV; 8975 return ops->ndo_change_carrier(dev, new_carrier); 8976 } 8977 8978 /** 8979 * dev_get_phys_port_id - Get device physical port ID 8980 * @dev: device 8981 * @ppid: port ID 8982 * 8983 * Get device physical port ID 8984 */ 8985 int dev_get_phys_port_id(struct net_device *dev, 8986 struct netdev_phys_item_id *ppid) 8987 { 8988 const struct net_device_ops *ops = dev->netdev_ops; 8989 8990 if (!ops->ndo_get_phys_port_id) 8991 return -EOPNOTSUPP; 8992 return ops->ndo_get_phys_port_id(dev, ppid); 8993 } 8994 8995 /** 8996 * dev_get_phys_port_name - Get device physical port name 8997 * @dev: device 8998 * @name: port name 8999 * @len: limit of bytes to copy to name 9000 * 9001 * Get device physical port name 9002 */ 9003 int dev_get_phys_port_name(struct net_device *dev, 9004 char *name, size_t len) 9005 { 9006 const struct net_device_ops *ops = dev->netdev_ops; 9007 int err; 9008 9009 if (ops->ndo_get_phys_port_name) { 9010 err = ops->ndo_get_phys_port_name(dev, name, len); 9011 if (err != -EOPNOTSUPP) 9012 return err; 9013 } 9014 return devlink_compat_phys_port_name_get(dev, name, len); 9015 } 9016 9017 /** 9018 * dev_get_port_parent_id - Get the device's port parent identifier 9019 * @dev: network device 9020 * @ppid: pointer to a storage for the port's parent identifier 9021 * @recurse: allow/disallow recursion to lower devices 9022 * 9023 * Get the devices's port parent identifier 9024 */ 9025 int dev_get_port_parent_id(struct net_device *dev, 9026 struct netdev_phys_item_id *ppid, 9027 bool recurse) 9028 { 9029 const struct net_device_ops *ops = dev->netdev_ops; 9030 struct netdev_phys_item_id first = { }; 9031 struct net_device *lower_dev; 9032 struct list_head *iter; 9033 int err; 9034 9035 if (ops->ndo_get_port_parent_id) { 9036 err = ops->ndo_get_port_parent_id(dev, ppid); 9037 if (err != -EOPNOTSUPP) 9038 return err; 9039 } 9040 9041 err = devlink_compat_switch_id_get(dev, ppid); 9042 if (!recurse || err != -EOPNOTSUPP) 9043 return err; 9044 9045 netdev_for_each_lower_dev(dev, lower_dev, iter) { 9046 err = dev_get_port_parent_id(lower_dev, ppid, true); 9047 if (err) 9048 break; 9049 if (!first.id_len) 9050 first = *ppid; 9051 else if (memcmp(&first, ppid, sizeof(*ppid))) 9052 return -EOPNOTSUPP; 9053 } 9054 9055 return err; 9056 } 9057 EXPORT_SYMBOL(dev_get_port_parent_id); 9058 9059 /** 9060 * netdev_port_same_parent_id - Indicate if two network devices have 9061 * the same port parent identifier 9062 * @a: first network device 9063 * @b: second network device 9064 */ 9065 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b) 9066 { 9067 struct netdev_phys_item_id a_id = { }; 9068 struct netdev_phys_item_id b_id = { }; 9069 9070 if (dev_get_port_parent_id(a, &a_id, true) || 9071 dev_get_port_parent_id(b, &b_id, true)) 9072 return false; 9073 9074 return netdev_phys_item_id_same(&a_id, &b_id); 9075 } 9076 EXPORT_SYMBOL(netdev_port_same_parent_id); 9077 9078 static void netdev_dpll_pin_assign(struct net_device *dev, struct dpll_pin *dpll_pin) 9079 { 9080 #if IS_ENABLED(CONFIG_DPLL) 9081 rtnl_lock(); 9082 dev->dpll_pin = dpll_pin; 9083 rtnl_unlock(); 9084 #endif 9085 } 9086 9087 void netdev_dpll_pin_set(struct net_device *dev, struct dpll_pin *dpll_pin) 9088 { 9089 WARN_ON(!dpll_pin); 9090 netdev_dpll_pin_assign(dev, dpll_pin); 9091 } 9092 EXPORT_SYMBOL(netdev_dpll_pin_set); 9093 9094 void netdev_dpll_pin_clear(struct net_device *dev) 9095 { 9096 netdev_dpll_pin_assign(dev, NULL); 9097 } 9098 EXPORT_SYMBOL(netdev_dpll_pin_clear); 9099 9100 /** 9101 * dev_change_proto_down - set carrier according to proto_down. 9102 * 9103 * @dev: device 9104 * @proto_down: new value 9105 */ 9106 int dev_change_proto_down(struct net_device *dev, bool proto_down) 9107 { 9108 if (!(dev->priv_flags & IFF_CHANGE_PROTO_DOWN)) 9109 return -EOPNOTSUPP; 9110 if (!netif_device_present(dev)) 9111 return -ENODEV; 9112 if (proto_down) 9113 netif_carrier_off(dev); 9114 else 9115 netif_carrier_on(dev); 9116 dev->proto_down = proto_down; 9117 return 0; 9118 } 9119 9120 /** 9121 * dev_change_proto_down_reason - proto down reason 9122 * 9123 * @dev: device 9124 * @mask: proto down mask 9125 * @value: proto down value 9126 */ 9127 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask, 9128 u32 value) 9129 { 9130 int b; 9131 9132 if (!mask) { 9133 dev->proto_down_reason = value; 9134 } else { 9135 for_each_set_bit(b, &mask, 32) { 9136 if (value & (1 << b)) 9137 dev->proto_down_reason |= BIT(b); 9138 else 9139 dev->proto_down_reason &= ~BIT(b); 9140 } 9141 } 9142 } 9143 9144 struct bpf_xdp_link { 9145 struct bpf_link link; 9146 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */ 9147 int flags; 9148 }; 9149 9150 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags) 9151 { 9152 if (flags & XDP_FLAGS_HW_MODE) 9153 return XDP_MODE_HW; 9154 if (flags & XDP_FLAGS_DRV_MODE) 9155 return XDP_MODE_DRV; 9156 if (flags & XDP_FLAGS_SKB_MODE) 9157 return XDP_MODE_SKB; 9158 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB; 9159 } 9160 9161 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode) 9162 { 9163 switch (mode) { 9164 case XDP_MODE_SKB: 9165 return generic_xdp_install; 9166 case XDP_MODE_DRV: 9167 case XDP_MODE_HW: 9168 return dev->netdev_ops->ndo_bpf; 9169 default: 9170 return NULL; 9171 } 9172 } 9173 9174 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev, 9175 enum bpf_xdp_mode mode) 9176 { 9177 return dev->xdp_state[mode].link; 9178 } 9179 9180 static struct bpf_prog *dev_xdp_prog(struct net_device *dev, 9181 enum bpf_xdp_mode mode) 9182 { 9183 struct bpf_xdp_link *link = dev_xdp_link(dev, mode); 9184 9185 if (link) 9186 return link->link.prog; 9187 return dev->xdp_state[mode].prog; 9188 } 9189 9190 u8 dev_xdp_prog_count(struct net_device *dev) 9191 { 9192 u8 count = 0; 9193 int i; 9194 9195 for (i = 0; i < __MAX_XDP_MODE; i++) 9196 if (dev->xdp_state[i].prog || dev->xdp_state[i].link) 9197 count++; 9198 return count; 9199 } 9200 EXPORT_SYMBOL_GPL(dev_xdp_prog_count); 9201 9202 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode) 9203 { 9204 struct bpf_prog *prog = dev_xdp_prog(dev, mode); 9205 9206 return prog ? prog->aux->id : 0; 9207 } 9208 9209 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode, 9210 struct bpf_xdp_link *link) 9211 { 9212 dev->xdp_state[mode].link = link; 9213 dev->xdp_state[mode].prog = NULL; 9214 } 9215 9216 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode, 9217 struct bpf_prog *prog) 9218 { 9219 dev->xdp_state[mode].link = NULL; 9220 dev->xdp_state[mode].prog = prog; 9221 } 9222 9223 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode, 9224 bpf_op_t bpf_op, struct netlink_ext_ack *extack, 9225 u32 flags, struct bpf_prog *prog) 9226 { 9227 struct netdev_bpf xdp; 9228 int err; 9229 9230 memset(&xdp, 0, sizeof(xdp)); 9231 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG; 9232 xdp.extack = extack; 9233 xdp.flags = flags; 9234 xdp.prog = prog; 9235 9236 /* Drivers assume refcnt is already incremented (i.e, prog pointer is 9237 * "moved" into driver), so they don't increment it on their own, but 9238 * they do decrement refcnt when program is detached or replaced. 9239 * Given net_device also owns link/prog, we need to bump refcnt here 9240 * to prevent drivers from underflowing it. 9241 */ 9242 if (prog) 9243 bpf_prog_inc(prog); 9244 err = bpf_op(dev, &xdp); 9245 if (err) { 9246 if (prog) 9247 bpf_prog_put(prog); 9248 return err; 9249 } 9250 9251 if (mode != XDP_MODE_HW) 9252 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog); 9253 9254 return 0; 9255 } 9256 9257 static void dev_xdp_uninstall(struct net_device *dev) 9258 { 9259 struct bpf_xdp_link *link; 9260 struct bpf_prog *prog; 9261 enum bpf_xdp_mode mode; 9262 bpf_op_t bpf_op; 9263 9264 ASSERT_RTNL(); 9265 9266 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) { 9267 prog = dev_xdp_prog(dev, mode); 9268 if (!prog) 9269 continue; 9270 9271 bpf_op = dev_xdp_bpf_op(dev, mode); 9272 if (!bpf_op) 9273 continue; 9274 9275 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9276 9277 /* auto-detach link from net device */ 9278 link = dev_xdp_link(dev, mode); 9279 if (link) 9280 link->dev = NULL; 9281 else 9282 bpf_prog_put(prog); 9283 9284 dev_xdp_set_link(dev, mode, NULL); 9285 } 9286 } 9287 9288 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack, 9289 struct bpf_xdp_link *link, struct bpf_prog *new_prog, 9290 struct bpf_prog *old_prog, u32 flags) 9291 { 9292 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES); 9293 struct bpf_prog *cur_prog; 9294 struct net_device *upper; 9295 struct list_head *iter; 9296 enum bpf_xdp_mode mode; 9297 bpf_op_t bpf_op; 9298 int err; 9299 9300 ASSERT_RTNL(); 9301 9302 /* either link or prog attachment, never both */ 9303 if (link && (new_prog || old_prog)) 9304 return -EINVAL; 9305 /* link supports only XDP mode flags */ 9306 if (link && (flags & ~XDP_FLAGS_MODES)) { 9307 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment"); 9308 return -EINVAL; 9309 } 9310 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */ 9311 if (num_modes > 1) { 9312 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set"); 9313 return -EINVAL; 9314 } 9315 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */ 9316 if (!num_modes && dev_xdp_prog_count(dev) > 1) { 9317 NL_SET_ERR_MSG(extack, 9318 "More than one program loaded, unset mode is ambiguous"); 9319 return -EINVAL; 9320 } 9321 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */ 9322 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) { 9323 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified"); 9324 return -EINVAL; 9325 } 9326 9327 mode = dev_xdp_mode(dev, flags); 9328 /* can't replace attached link */ 9329 if (dev_xdp_link(dev, mode)) { 9330 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link"); 9331 return -EBUSY; 9332 } 9333 9334 /* don't allow if an upper device already has a program */ 9335 netdev_for_each_upper_dev_rcu(dev, upper, iter) { 9336 if (dev_xdp_prog_count(upper) > 0) { 9337 NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program"); 9338 return -EEXIST; 9339 } 9340 } 9341 9342 cur_prog = dev_xdp_prog(dev, mode); 9343 /* can't replace attached prog with link */ 9344 if (link && cur_prog) { 9345 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link"); 9346 return -EBUSY; 9347 } 9348 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) { 9349 NL_SET_ERR_MSG(extack, "Active program does not match expected"); 9350 return -EEXIST; 9351 } 9352 9353 /* put effective new program into new_prog */ 9354 if (link) 9355 new_prog = link->link.prog; 9356 9357 if (new_prog) { 9358 bool offload = mode == XDP_MODE_HW; 9359 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB 9360 ? XDP_MODE_DRV : XDP_MODE_SKB; 9361 9362 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) { 9363 NL_SET_ERR_MSG(extack, "XDP program already attached"); 9364 return -EBUSY; 9365 } 9366 if (!offload && dev_xdp_prog(dev, other_mode)) { 9367 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time"); 9368 return -EEXIST; 9369 } 9370 if (!offload && bpf_prog_is_offloaded(new_prog->aux)) { 9371 NL_SET_ERR_MSG(extack, "Using offloaded program without HW_MODE flag is not supported"); 9372 return -EINVAL; 9373 } 9374 if (bpf_prog_is_dev_bound(new_prog->aux) && !bpf_offload_dev_match(new_prog, dev)) { 9375 NL_SET_ERR_MSG(extack, "Program bound to different device"); 9376 return -EINVAL; 9377 } 9378 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) { 9379 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device"); 9380 return -EINVAL; 9381 } 9382 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) { 9383 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device"); 9384 return -EINVAL; 9385 } 9386 } 9387 9388 /* don't call drivers if the effective program didn't change */ 9389 if (new_prog != cur_prog) { 9390 bpf_op = dev_xdp_bpf_op(dev, mode); 9391 if (!bpf_op) { 9392 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode"); 9393 return -EOPNOTSUPP; 9394 } 9395 9396 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog); 9397 if (err) 9398 return err; 9399 } 9400 9401 if (link) 9402 dev_xdp_set_link(dev, mode, link); 9403 else 9404 dev_xdp_set_prog(dev, mode, new_prog); 9405 if (cur_prog) 9406 bpf_prog_put(cur_prog); 9407 9408 return 0; 9409 } 9410 9411 static int dev_xdp_attach_link(struct net_device *dev, 9412 struct netlink_ext_ack *extack, 9413 struct bpf_xdp_link *link) 9414 { 9415 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags); 9416 } 9417 9418 static int dev_xdp_detach_link(struct net_device *dev, 9419 struct netlink_ext_ack *extack, 9420 struct bpf_xdp_link *link) 9421 { 9422 enum bpf_xdp_mode mode; 9423 bpf_op_t bpf_op; 9424 9425 ASSERT_RTNL(); 9426 9427 mode = dev_xdp_mode(dev, link->flags); 9428 if (dev_xdp_link(dev, mode) != link) 9429 return -EINVAL; 9430 9431 bpf_op = dev_xdp_bpf_op(dev, mode); 9432 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9433 dev_xdp_set_link(dev, mode, NULL); 9434 return 0; 9435 } 9436 9437 static void bpf_xdp_link_release(struct bpf_link *link) 9438 { 9439 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9440 9441 rtnl_lock(); 9442 9443 /* if racing with net_device's tear down, xdp_link->dev might be 9444 * already NULL, in which case link was already auto-detached 9445 */ 9446 if (xdp_link->dev) { 9447 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link)); 9448 xdp_link->dev = NULL; 9449 } 9450 9451 rtnl_unlock(); 9452 } 9453 9454 static int bpf_xdp_link_detach(struct bpf_link *link) 9455 { 9456 bpf_xdp_link_release(link); 9457 return 0; 9458 } 9459 9460 static void bpf_xdp_link_dealloc(struct bpf_link *link) 9461 { 9462 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9463 9464 kfree(xdp_link); 9465 } 9466 9467 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link, 9468 struct seq_file *seq) 9469 { 9470 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9471 u32 ifindex = 0; 9472 9473 rtnl_lock(); 9474 if (xdp_link->dev) 9475 ifindex = xdp_link->dev->ifindex; 9476 rtnl_unlock(); 9477 9478 seq_printf(seq, "ifindex:\t%u\n", ifindex); 9479 } 9480 9481 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link, 9482 struct bpf_link_info *info) 9483 { 9484 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9485 u32 ifindex = 0; 9486 9487 rtnl_lock(); 9488 if (xdp_link->dev) 9489 ifindex = xdp_link->dev->ifindex; 9490 rtnl_unlock(); 9491 9492 info->xdp.ifindex = ifindex; 9493 return 0; 9494 } 9495 9496 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog, 9497 struct bpf_prog *old_prog) 9498 { 9499 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9500 enum bpf_xdp_mode mode; 9501 bpf_op_t bpf_op; 9502 int err = 0; 9503 9504 rtnl_lock(); 9505 9506 /* link might have been auto-released already, so fail */ 9507 if (!xdp_link->dev) { 9508 err = -ENOLINK; 9509 goto out_unlock; 9510 } 9511 9512 if (old_prog && link->prog != old_prog) { 9513 err = -EPERM; 9514 goto out_unlock; 9515 } 9516 old_prog = link->prog; 9517 if (old_prog->type != new_prog->type || 9518 old_prog->expected_attach_type != new_prog->expected_attach_type) { 9519 err = -EINVAL; 9520 goto out_unlock; 9521 } 9522 9523 if (old_prog == new_prog) { 9524 /* no-op, don't disturb drivers */ 9525 bpf_prog_put(new_prog); 9526 goto out_unlock; 9527 } 9528 9529 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags); 9530 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode); 9531 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL, 9532 xdp_link->flags, new_prog); 9533 if (err) 9534 goto out_unlock; 9535 9536 old_prog = xchg(&link->prog, new_prog); 9537 bpf_prog_put(old_prog); 9538 9539 out_unlock: 9540 rtnl_unlock(); 9541 return err; 9542 } 9543 9544 static const struct bpf_link_ops bpf_xdp_link_lops = { 9545 .release = bpf_xdp_link_release, 9546 .dealloc = bpf_xdp_link_dealloc, 9547 .detach = bpf_xdp_link_detach, 9548 .show_fdinfo = bpf_xdp_link_show_fdinfo, 9549 .fill_link_info = bpf_xdp_link_fill_link_info, 9550 .update_prog = bpf_xdp_link_update, 9551 }; 9552 9553 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog) 9554 { 9555 struct net *net = current->nsproxy->net_ns; 9556 struct bpf_link_primer link_primer; 9557 struct netlink_ext_ack extack = {}; 9558 struct bpf_xdp_link *link; 9559 struct net_device *dev; 9560 int err, fd; 9561 9562 rtnl_lock(); 9563 dev = dev_get_by_index(net, attr->link_create.target_ifindex); 9564 if (!dev) { 9565 rtnl_unlock(); 9566 return -EINVAL; 9567 } 9568 9569 link = kzalloc(sizeof(*link), GFP_USER); 9570 if (!link) { 9571 err = -ENOMEM; 9572 goto unlock; 9573 } 9574 9575 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog); 9576 link->dev = dev; 9577 link->flags = attr->link_create.flags; 9578 9579 err = bpf_link_prime(&link->link, &link_primer); 9580 if (err) { 9581 kfree(link); 9582 goto unlock; 9583 } 9584 9585 err = dev_xdp_attach_link(dev, &extack, link); 9586 rtnl_unlock(); 9587 9588 if (err) { 9589 link->dev = NULL; 9590 bpf_link_cleanup(&link_primer); 9591 trace_bpf_xdp_link_attach_failed(extack._msg); 9592 goto out_put_dev; 9593 } 9594 9595 fd = bpf_link_settle(&link_primer); 9596 /* link itself doesn't hold dev's refcnt to not complicate shutdown */ 9597 dev_put(dev); 9598 return fd; 9599 9600 unlock: 9601 rtnl_unlock(); 9602 9603 out_put_dev: 9604 dev_put(dev); 9605 return err; 9606 } 9607 9608 /** 9609 * dev_change_xdp_fd - set or clear a bpf program for a device rx path 9610 * @dev: device 9611 * @extack: netlink extended ack 9612 * @fd: new program fd or negative value to clear 9613 * @expected_fd: old program fd that userspace expects to replace or clear 9614 * @flags: xdp-related flags 9615 * 9616 * Set or clear a bpf program for a device 9617 */ 9618 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, 9619 int fd, int expected_fd, u32 flags) 9620 { 9621 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags); 9622 struct bpf_prog *new_prog = NULL, *old_prog = NULL; 9623 int err; 9624 9625 ASSERT_RTNL(); 9626 9627 if (fd >= 0) { 9628 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP, 9629 mode != XDP_MODE_SKB); 9630 if (IS_ERR(new_prog)) 9631 return PTR_ERR(new_prog); 9632 } 9633 9634 if (expected_fd >= 0) { 9635 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP, 9636 mode != XDP_MODE_SKB); 9637 if (IS_ERR(old_prog)) { 9638 err = PTR_ERR(old_prog); 9639 old_prog = NULL; 9640 goto err_out; 9641 } 9642 } 9643 9644 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags); 9645 9646 err_out: 9647 if (err && new_prog) 9648 bpf_prog_put(new_prog); 9649 if (old_prog) 9650 bpf_prog_put(old_prog); 9651 return err; 9652 } 9653 9654 /** 9655 * dev_index_reserve() - allocate an ifindex in a namespace 9656 * @net: the applicable net namespace 9657 * @ifindex: requested ifindex, pass %0 to get one allocated 9658 * 9659 * Allocate a ifindex for a new device. Caller must either use the ifindex 9660 * to store the device (via list_netdevice()) or call dev_index_release() 9661 * to give the index up. 9662 * 9663 * Return: a suitable unique value for a new device interface number or -errno. 9664 */ 9665 static int dev_index_reserve(struct net *net, u32 ifindex) 9666 { 9667 int err; 9668 9669 if (ifindex > INT_MAX) { 9670 DEBUG_NET_WARN_ON_ONCE(1); 9671 return -EINVAL; 9672 } 9673 9674 if (!ifindex) 9675 err = xa_alloc_cyclic(&net->dev_by_index, &ifindex, NULL, 9676 xa_limit_31b, &net->ifindex, GFP_KERNEL); 9677 else 9678 err = xa_insert(&net->dev_by_index, ifindex, NULL, GFP_KERNEL); 9679 if (err < 0) 9680 return err; 9681 9682 return ifindex; 9683 } 9684 9685 static void dev_index_release(struct net *net, int ifindex) 9686 { 9687 /* Expect only unused indexes, unlist_netdevice() removes the used */ 9688 WARN_ON(xa_erase(&net->dev_by_index, ifindex)); 9689 } 9690 9691 /* Delayed registration/unregisteration */ 9692 LIST_HEAD(net_todo_list); 9693 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); 9694 9695 static void net_set_todo(struct net_device *dev) 9696 { 9697 list_add_tail(&dev->todo_list, &net_todo_list); 9698 atomic_inc(&dev_net(dev)->dev_unreg_count); 9699 } 9700 9701 static netdev_features_t netdev_sync_upper_features(struct net_device *lower, 9702 struct net_device *upper, netdev_features_t features) 9703 { 9704 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9705 netdev_features_t feature; 9706 int feature_bit; 9707 9708 for_each_netdev_feature(upper_disables, feature_bit) { 9709 feature = __NETIF_F_BIT(feature_bit); 9710 if (!(upper->wanted_features & feature) 9711 && (features & feature)) { 9712 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", 9713 &feature, upper->name); 9714 features &= ~feature; 9715 } 9716 } 9717 9718 return features; 9719 } 9720 9721 static void netdev_sync_lower_features(struct net_device *upper, 9722 struct net_device *lower, netdev_features_t features) 9723 { 9724 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9725 netdev_features_t feature; 9726 int feature_bit; 9727 9728 for_each_netdev_feature(upper_disables, feature_bit) { 9729 feature = __NETIF_F_BIT(feature_bit); 9730 if (!(features & feature) && (lower->features & feature)) { 9731 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", 9732 &feature, lower->name); 9733 lower->wanted_features &= ~feature; 9734 __netdev_update_features(lower); 9735 9736 if (unlikely(lower->features & feature)) 9737 netdev_WARN(upper, "failed to disable %pNF on %s!\n", 9738 &feature, lower->name); 9739 else 9740 netdev_features_change(lower); 9741 } 9742 } 9743 } 9744 9745 static netdev_features_t netdev_fix_features(struct net_device *dev, 9746 netdev_features_t features) 9747 { 9748 /* Fix illegal checksum combinations */ 9749 if ((features & NETIF_F_HW_CSUM) && 9750 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { 9751 netdev_warn(dev, "mixed HW and IP checksum settings.\n"); 9752 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); 9753 } 9754 9755 /* TSO requires that SG is present as well. */ 9756 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { 9757 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); 9758 features &= ~NETIF_F_ALL_TSO; 9759 } 9760 9761 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && 9762 !(features & NETIF_F_IP_CSUM)) { 9763 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); 9764 features &= ~NETIF_F_TSO; 9765 features &= ~NETIF_F_TSO_ECN; 9766 } 9767 9768 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && 9769 !(features & NETIF_F_IPV6_CSUM)) { 9770 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); 9771 features &= ~NETIF_F_TSO6; 9772 } 9773 9774 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ 9775 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) 9776 features &= ~NETIF_F_TSO_MANGLEID; 9777 9778 /* TSO ECN requires that TSO is present as well. */ 9779 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) 9780 features &= ~NETIF_F_TSO_ECN; 9781 9782 /* Software GSO depends on SG. */ 9783 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { 9784 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); 9785 features &= ~NETIF_F_GSO; 9786 } 9787 9788 /* GSO partial features require GSO partial be set */ 9789 if ((features & dev->gso_partial_features) && 9790 !(features & NETIF_F_GSO_PARTIAL)) { 9791 netdev_dbg(dev, 9792 "Dropping partially supported GSO features since no GSO partial.\n"); 9793 features &= ~dev->gso_partial_features; 9794 } 9795 9796 if (!(features & NETIF_F_RXCSUM)) { 9797 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet 9798 * successfully merged by hardware must also have the 9799 * checksum verified by hardware. If the user does not 9800 * want to enable RXCSUM, logically, we should disable GRO_HW. 9801 */ 9802 if (features & NETIF_F_GRO_HW) { 9803 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n"); 9804 features &= ~NETIF_F_GRO_HW; 9805 } 9806 } 9807 9808 /* LRO/HW-GRO features cannot be combined with RX-FCS */ 9809 if (features & NETIF_F_RXFCS) { 9810 if (features & NETIF_F_LRO) { 9811 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n"); 9812 features &= ~NETIF_F_LRO; 9813 } 9814 9815 if (features & NETIF_F_GRO_HW) { 9816 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n"); 9817 features &= ~NETIF_F_GRO_HW; 9818 } 9819 } 9820 9821 if ((features & NETIF_F_GRO_HW) && (features & NETIF_F_LRO)) { 9822 netdev_dbg(dev, "Dropping LRO feature since HW-GRO is requested.\n"); 9823 features &= ~NETIF_F_LRO; 9824 } 9825 9826 if (features & NETIF_F_HW_TLS_TX) { 9827 bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) == 9828 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM); 9829 bool hw_csum = features & NETIF_F_HW_CSUM; 9830 9831 if (!ip_csum && !hw_csum) { 9832 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n"); 9833 features &= ~NETIF_F_HW_TLS_TX; 9834 } 9835 } 9836 9837 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) { 9838 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n"); 9839 features &= ~NETIF_F_HW_TLS_RX; 9840 } 9841 9842 return features; 9843 } 9844 9845 int __netdev_update_features(struct net_device *dev) 9846 { 9847 struct net_device *upper, *lower; 9848 netdev_features_t features; 9849 struct list_head *iter; 9850 int err = -1; 9851 9852 ASSERT_RTNL(); 9853 9854 features = netdev_get_wanted_features(dev); 9855 9856 if (dev->netdev_ops->ndo_fix_features) 9857 features = dev->netdev_ops->ndo_fix_features(dev, features); 9858 9859 /* driver might be less strict about feature dependencies */ 9860 features = netdev_fix_features(dev, features); 9861 9862 /* some features can't be enabled if they're off on an upper device */ 9863 netdev_for_each_upper_dev_rcu(dev, upper, iter) 9864 features = netdev_sync_upper_features(dev, upper, features); 9865 9866 if (dev->features == features) 9867 goto sync_lower; 9868 9869 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", 9870 &dev->features, &features); 9871 9872 if (dev->netdev_ops->ndo_set_features) 9873 err = dev->netdev_ops->ndo_set_features(dev, features); 9874 else 9875 err = 0; 9876 9877 if (unlikely(err < 0)) { 9878 netdev_err(dev, 9879 "set_features() failed (%d); wanted %pNF, left %pNF\n", 9880 err, &features, &dev->features); 9881 /* return non-0 since some features might have changed and 9882 * it's better to fire a spurious notification than miss it 9883 */ 9884 return -1; 9885 } 9886 9887 sync_lower: 9888 /* some features must be disabled on lower devices when disabled 9889 * on an upper device (think: bonding master or bridge) 9890 */ 9891 netdev_for_each_lower_dev(dev, lower, iter) 9892 netdev_sync_lower_features(dev, lower, features); 9893 9894 if (!err) { 9895 netdev_features_t diff = features ^ dev->features; 9896 9897 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) { 9898 /* udp_tunnel_{get,drop}_rx_info both need 9899 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the 9900 * device, or they won't do anything. 9901 * Thus we need to update dev->features 9902 * *before* calling udp_tunnel_get_rx_info, 9903 * but *after* calling udp_tunnel_drop_rx_info. 9904 */ 9905 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) { 9906 dev->features = features; 9907 udp_tunnel_get_rx_info(dev); 9908 } else { 9909 udp_tunnel_drop_rx_info(dev); 9910 } 9911 } 9912 9913 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) { 9914 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) { 9915 dev->features = features; 9916 err |= vlan_get_rx_ctag_filter_info(dev); 9917 } else { 9918 vlan_drop_rx_ctag_filter_info(dev); 9919 } 9920 } 9921 9922 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) { 9923 if (features & NETIF_F_HW_VLAN_STAG_FILTER) { 9924 dev->features = features; 9925 err |= vlan_get_rx_stag_filter_info(dev); 9926 } else { 9927 vlan_drop_rx_stag_filter_info(dev); 9928 } 9929 } 9930 9931 dev->features = features; 9932 } 9933 9934 return err < 0 ? 0 : 1; 9935 } 9936 9937 /** 9938 * netdev_update_features - recalculate device features 9939 * @dev: the device to check 9940 * 9941 * Recalculate dev->features set and send notifications if it 9942 * has changed. Should be called after driver or hardware dependent 9943 * conditions might have changed that influence the features. 9944 */ 9945 void netdev_update_features(struct net_device *dev) 9946 { 9947 if (__netdev_update_features(dev)) 9948 netdev_features_change(dev); 9949 } 9950 EXPORT_SYMBOL(netdev_update_features); 9951 9952 /** 9953 * netdev_change_features - recalculate device features 9954 * @dev: the device to check 9955 * 9956 * Recalculate dev->features set and send notifications even 9957 * if they have not changed. Should be called instead of 9958 * netdev_update_features() if also dev->vlan_features might 9959 * have changed to allow the changes to be propagated to stacked 9960 * VLAN devices. 9961 */ 9962 void netdev_change_features(struct net_device *dev) 9963 { 9964 __netdev_update_features(dev); 9965 netdev_features_change(dev); 9966 } 9967 EXPORT_SYMBOL(netdev_change_features); 9968 9969 /** 9970 * netif_stacked_transfer_operstate - transfer operstate 9971 * @rootdev: the root or lower level device to transfer state from 9972 * @dev: the device to transfer operstate to 9973 * 9974 * Transfer operational state from root to device. This is normally 9975 * called when a stacking relationship exists between the root 9976 * device and the device(a leaf device). 9977 */ 9978 void netif_stacked_transfer_operstate(const struct net_device *rootdev, 9979 struct net_device *dev) 9980 { 9981 if (rootdev->operstate == IF_OPER_DORMANT) 9982 netif_dormant_on(dev); 9983 else 9984 netif_dormant_off(dev); 9985 9986 if (rootdev->operstate == IF_OPER_TESTING) 9987 netif_testing_on(dev); 9988 else 9989 netif_testing_off(dev); 9990 9991 if (netif_carrier_ok(rootdev)) 9992 netif_carrier_on(dev); 9993 else 9994 netif_carrier_off(dev); 9995 } 9996 EXPORT_SYMBOL(netif_stacked_transfer_operstate); 9997 9998 static int netif_alloc_rx_queues(struct net_device *dev) 9999 { 10000 unsigned int i, count = dev->num_rx_queues; 10001 struct netdev_rx_queue *rx; 10002 size_t sz = count * sizeof(*rx); 10003 int err = 0; 10004 10005 BUG_ON(count < 1); 10006 10007 rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10008 if (!rx) 10009 return -ENOMEM; 10010 10011 dev->_rx = rx; 10012 10013 for (i = 0; i < count; i++) { 10014 rx[i].dev = dev; 10015 10016 /* XDP RX-queue setup */ 10017 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0); 10018 if (err < 0) 10019 goto err_rxq_info; 10020 } 10021 return 0; 10022 10023 err_rxq_info: 10024 /* Rollback successful reg's and free other resources */ 10025 while (i--) 10026 xdp_rxq_info_unreg(&rx[i].xdp_rxq); 10027 kvfree(dev->_rx); 10028 dev->_rx = NULL; 10029 return err; 10030 } 10031 10032 static void netif_free_rx_queues(struct net_device *dev) 10033 { 10034 unsigned int i, count = dev->num_rx_queues; 10035 10036 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */ 10037 if (!dev->_rx) 10038 return; 10039 10040 for (i = 0; i < count; i++) 10041 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq); 10042 10043 kvfree(dev->_rx); 10044 } 10045 10046 static void netdev_init_one_queue(struct net_device *dev, 10047 struct netdev_queue *queue, void *_unused) 10048 { 10049 /* Initialize queue lock */ 10050 spin_lock_init(&queue->_xmit_lock); 10051 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); 10052 queue->xmit_lock_owner = -1; 10053 netdev_queue_numa_node_write(queue, NUMA_NO_NODE); 10054 queue->dev = dev; 10055 #ifdef CONFIG_BQL 10056 dql_init(&queue->dql, HZ); 10057 #endif 10058 } 10059 10060 static void netif_free_tx_queues(struct net_device *dev) 10061 { 10062 kvfree(dev->_tx); 10063 } 10064 10065 static int netif_alloc_netdev_queues(struct net_device *dev) 10066 { 10067 unsigned int count = dev->num_tx_queues; 10068 struct netdev_queue *tx; 10069 size_t sz = count * sizeof(*tx); 10070 10071 if (count < 1 || count > 0xffff) 10072 return -EINVAL; 10073 10074 tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10075 if (!tx) 10076 return -ENOMEM; 10077 10078 dev->_tx = tx; 10079 10080 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); 10081 spin_lock_init(&dev->tx_global_lock); 10082 10083 return 0; 10084 } 10085 10086 void netif_tx_stop_all_queues(struct net_device *dev) 10087 { 10088 unsigned int i; 10089 10090 for (i = 0; i < dev->num_tx_queues; i++) { 10091 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 10092 10093 netif_tx_stop_queue(txq); 10094 } 10095 } 10096 EXPORT_SYMBOL(netif_tx_stop_all_queues); 10097 10098 static int netdev_do_alloc_pcpu_stats(struct net_device *dev) 10099 { 10100 void __percpu *v; 10101 10102 /* Drivers implementing ndo_get_peer_dev must support tstat 10103 * accounting, so that skb_do_redirect() can bump the dev's 10104 * RX stats upon network namespace switch. 10105 */ 10106 if (dev->netdev_ops->ndo_get_peer_dev && 10107 dev->pcpu_stat_type != NETDEV_PCPU_STAT_TSTATS) 10108 return -EOPNOTSUPP; 10109 10110 switch (dev->pcpu_stat_type) { 10111 case NETDEV_PCPU_STAT_NONE: 10112 return 0; 10113 case NETDEV_PCPU_STAT_LSTATS: 10114 v = dev->lstats = netdev_alloc_pcpu_stats(struct pcpu_lstats); 10115 break; 10116 case NETDEV_PCPU_STAT_TSTATS: 10117 v = dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats); 10118 break; 10119 case NETDEV_PCPU_STAT_DSTATS: 10120 v = dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats); 10121 break; 10122 default: 10123 return -EINVAL; 10124 } 10125 10126 return v ? 0 : -ENOMEM; 10127 } 10128 10129 static void netdev_do_free_pcpu_stats(struct net_device *dev) 10130 { 10131 switch (dev->pcpu_stat_type) { 10132 case NETDEV_PCPU_STAT_NONE: 10133 return; 10134 case NETDEV_PCPU_STAT_LSTATS: 10135 free_percpu(dev->lstats); 10136 break; 10137 case NETDEV_PCPU_STAT_TSTATS: 10138 free_percpu(dev->tstats); 10139 break; 10140 case NETDEV_PCPU_STAT_DSTATS: 10141 free_percpu(dev->dstats); 10142 break; 10143 } 10144 } 10145 10146 /** 10147 * register_netdevice() - register a network device 10148 * @dev: device to register 10149 * 10150 * Take a prepared network device structure and make it externally accessible. 10151 * A %NETDEV_REGISTER message is sent to the netdev notifier chain. 10152 * Callers must hold the rtnl lock - you may want register_netdev() 10153 * instead of this. 10154 */ 10155 int register_netdevice(struct net_device *dev) 10156 { 10157 int ret; 10158 struct net *net = dev_net(dev); 10159 10160 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE < 10161 NETDEV_FEATURE_COUNT); 10162 BUG_ON(dev_boot_phase); 10163 ASSERT_RTNL(); 10164 10165 might_sleep(); 10166 10167 /* When net_device's are persistent, this will be fatal. */ 10168 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); 10169 BUG_ON(!net); 10170 10171 ret = ethtool_check_ops(dev->ethtool_ops); 10172 if (ret) 10173 return ret; 10174 10175 spin_lock_init(&dev->addr_list_lock); 10176 netdev_set_addr_lockdep_class(dev); 10177 10178 ret = dev_get_valid_name(net, dev, dev->name); 10179 if (ret < 0) 10180 goto out; 10181 10182 ret = -ENOMEM; 10183 dev->name_node = netdev_name_node_head_alloc(dev); 10184 if (!dev->name_node) 10185 goto out; 10186 10187 /* Init, if this function is available */ 10188 if (dev->netdev_ops->ndo_init) { 10189 ret = dev->netdev_ops->ndo_init(dev); 10190 if (ret) { 10191 if (ret > 0) 10192 ret = -EIO; 10193 goto err_free_name; 10194 } 10195 } 10196 10197 if (((dev->hw_features | dev->features) & 10198 NETIF_F_HW_VLAN_CTAG_FILTER) && 10199 (!dev->netdev_ops->ndo_vlan_rx_add_vid || 10200 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { 10201 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); 10202 ret = -EINVAL; 10203 goto err_uninit; 10204 } 10205 10206 ret = netdev_do_alloc_pcpu_stats(dev); 10207 if (ret) 10208 goto err_uninit; 10209 10210 ret = dev_index_reserve(net, dev->ifindex); 10211 if (ret < 0) 10212 goto err_free_pcpu; 10213 dev->ifindex = ret; 10214 10215 /* Transfer changeable features to wanted_features and enable 10216 * software offloads (GSO and GRO). 10217 */ 10218 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF); 10219 dev->features |= NETIF_F_SOFT_FEATURES; 10220 10221 if (dev->udp_tunnel_nic_info) { 10222 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10223 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10224 } 10225 10226 dev->wanted_features = dev->features & dev->hw_features; 10227 10228 if (!(dev->flags & IFF_LOOPBACK)) 10229 dev->hw_features |= NETIF_F_NOCACHE_COPY; 10230 10231 /* If IPv4 TCP segmentation offload is supported we should also 10232 * allow the device to enable segmenting the frame with the option 10233 * of ignoring a static IP ID value. This doesn't enable the 10234 * feature itself but allows the user to enable it later. 10235 */ 10236 if (dev->hw_features & NETIF_F_TSO) 10237 dev->hw_features |= NETIF_F_TSO_MANGLEID; 10238 if (dev->vlan_features & NETIF_F_TSO) 10239 dev->vlan_features |= NETIF_F_TSO_MANGLEID; 10240 if (dev->mpls_features & NETIF_F_TSO) 10241 dev->mpls_features |= NETIF_F_TSO_MANGLEID; 10242 if (dev->hw_enc_features & NETIF_F_TSO) 10243 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; 10244 10245 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. 10246 */ 10247 dev->vlan_features |= NETIF_F_HIGHDMA; 10248 10249 /* Make NETIF_F_SG inheritable to tunnel devices. 10250 */ 10251 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; 10252 10253 /* Make NETIF_F_SG inheritable to MPLS. 10254 */ 10255 dev->mpls_features |= NETIF_F_SG; 10256 10257 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); 10258 ret = notifier_to_errno(ret); 10259 if (ret) 10260 goto err_ifindex_release; 10261 10262 ret = netdev_register_kobject(dev); 10263 write_lock(&dev_base_lock); 10264 dev->reg_state = ret ? NETREG_UNREGISTERED : NETREG_REGISTERED; 10265 write_unlock(&dev_base_lock); 10266 if (ret) 10267 goto err_uninit_notify; 10268 10269 __netdev_update_features(dev); 10270 10271 /* 10272 * Default initial state at registry is that the 10273 * device is present. 10274 */ 10275 10276 set_bit(__LINK_STATE_PRESENT, &dev->state); 10277 10278 linkwatch_init_dev(dev); 10279 10280 dev_init_scheduler(dev); 10281 10282 netdev_hold(dev, &dev->dev_registered_tracker, GFP_KERNEL); 10283 list_netdevice(dev); 10284 10285 add_device_randomness(dev->dev_addr, dev->addr_len); 10286 10287 /* If the device has permanent device address, driver should 10288 * set dev_addr and also addr_assign_type should be set to 10289 * NET_ADDR_PERM (default value). 10290 */ 10291 if (dev->addr_assign_type == NET_ADDR_PERM) 10292 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); 10293 10294 /* Notify protocols, that a new device appeared. */ 10295 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); 10296 ret = notifier_to_errno(ret); 10297 if (ret) { 10298 /* Expect explicit free_netdev() on failure */ 10299 dev->needs_free_netdev = false; 10300 unregister_netdevice_queue(dev, NULL); 10301 goto out; 10302 } 10303 /* 10304 * Prevent userspace races by waiting until the network 10305 * device is fully setup before sending notifications. 10306 */ 10307 if (!dev->rtnl_link_ops || 10308 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 10309 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL); 10310 10311 out: 10312 return ret; 10313 10314 err_uninit_notify: 10315 call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev); 10316 err_ifindex_release: 10317 dev_index_release(net, dev->ifindex); 10318 err_free_pcpu: 10319 netdev_do_free_pcpu_stats(dev); 10320 err_uninit: 10321 if (dev->netdev_ops->ndo_uninit) 10322 dev->netdev_ops->ndo_uninit(dev); 10323 if (dev->priv_destructor) 10324 dev->priv_destructor(dev); 10325 err_free_name: 10326 netdev_name_node_free(dev->name_node); 10327 goto out; 10328 } 10329 EXPORT_SYMBOL(register_netdevice); 10330 10331 /** 10332 * init_dummy_netdev - init a dummy network device for NAPI 10333 * @dev: device to init 10334 * 10335 * This takes a network device structure and initialize the minimum 10336 * amount of fields so it can be used to schedule NAPI polls without 10337 * registering a full blown interface. This is to be used by drivers 10338 * that need to tie several hardware interfaces to a single NAPI 10339 * poll scheduler due to HW limitations. 10340 */ 10341 int init_dummy_netdev(struct net_device *dev) 10342 { 10343 /* Clear everything. Note we don't initialize spinlocks 10344 * are they aren't supposed to be taken by any of the 10345 * NAPI code and this dummy netdev is supposed to be 10346 * only ever used for NAPI polls 10347 */ 10348 memset(dev, 0, sizeof(struct net_device)); 10349 10350 /* make sure we BUG if trying to hit standard 10351 * register/unregister code path 10352 */ 10353 dev->reg_state = NETREG_DUMMY; 10354 10355 /* NAPI wants this */ 10356 INIT_LIST_HEAD(&dev->napi_list); 10357 10358 /* a dummy interface is started by default */ 10359 set_bit(__LINK_STATE_PRESENT, &dev->state); 10360 set_bit(__LINK_STATE_START, &dev->state); 10361 10362 /* napi_busy_loop stats accounting wants this */ 10363 dev_net_set(dev, &init_net); 10364 10365 /* Note : We dont allocate pcpu_refcnt for dummy devices, 10366 * because users of this 'device' dont need to change 10367 * its refcount. 10368 */ 10369 10370 return 0; 10371 } 10372 EXPORT_SYMBOL_GPL(init_dummy_netdev); 10373 10374 10375 /** 10376 * register_netdev - register a network device 10377 * @dev: device to register 10378 * 10379 * Take a completed network device structure and add it to the kernel 10380 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 10381 * chain. 0 is returned on success. A negative errno code is returned 10382 * on a failure to set up the device, or if the name is a duplicate. 10383 * 10384 * This is a wrapper around register_netdevice that takes the rtnl semaphore 10385 * and expands the device name if you passed a format string to 10386 * alloc_netdev. 10387 */ 10388 int register_netdev(struct net_device *dev) 10389 { 10390 int err; 10391 10392 if (rtnl_lock_killable()) 10393 return -EINTR; 10394 err = register_netdevice(dev); 10395 rtnl_unlock(); 10396 return err; 10397 } 10398 EXPORT_SYMBOL(register_netdev); 10399 10400 int netdev_refcnt_read(const struct net_device *dev) 10401 { 10402 #ifdef CONFIG_PCPU_DEV_REFCNT 10403 int i, refcnt = 0; 10404 10405 for_each_possible_cpu(i) 10406 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); 10407 return refcnt; 10408 #else 10409 return refcount_read(&dev->dev_refcnt); 10410 #endif 10411 } 10412 EXPORT_SYMBOL(netdev_refcnt_read); 10413 10414 int netdev_unregister_timeout_secs __read_mostly = 10; 10415 10416 #define WAIT_REFS_MIN_MSECS 1 10417 #define WAIT_REFS_MAX_MSECS 250 10418 /** 10419 * netdev_wait_allrefs_any - wait until all references are gone. 10420 * @list: list of net_devices to wait on 10421 * 10422 * This is called when unregistering network devices. 10423 * 10424 * Any protocol or device that holds a reference should register 10425 * for netdevice notification, and cleanup and put back the 10426 * reference if they receive an UNREGISTER event. 10427 * We can get stuck here if buggy protocols don't correctly 10428 * call dev_put. 10429 */ 10430 static struct net_device *netdev_wait_allrefs_any(struct list_head *list) 10431 { 10432 unsigned long rebroadcast_time, warning_time; 10433 struct net_device *dev; 10434 int wait = 0; 10435 10436 rebroadcast_time = warning_time = jiffies; 10437 10438 list_for_each_entry(dev, list, todo_list) 10439 if (netdev_refcnt_read(dev) == 1) 10440 return dev; 10441 10442 while (true) { 10443 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { 10444 rtnl_lock(); 10445 10446 /* Rebroadcast unregister notification */ 10447 list_for_each_entry(dev, list, todo_list) 10448 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10449 10450 __rtnl_unlock(); 10451 rcu_barrier(); 10452 rtnl_lock(); 10453 10454 list_for_each_entry(dev, list, todo_list) 10455 if (test_bit(__LINK_STATE_LINKWATCH_PENDING, 10456 &dev->state)) { 10457 /* We must not have linkwatch events 10458 * pending on unregister. If this 10459 * happens, we simply run the queue 10460 * unscheduled, resulting in a noop 10461 * for this device. 10462 */ 10463 linkwatch_run_queue(); 10464 break; 10465 } 10466 10467 __rtnl_unlock(); 10468 10469 rebroadcast_time = jiffies; 10470 } 10471 10472 if (!wait) { 10473 rcu_barrier(); 10474 wait = WAIT_REFS_MIN_MSECS; 10475 } else { 10476 msleep(wait); 10477 wait = min(wait << 1, WAIT_REFS_MAX_MSECS); 10478 } 10479 10480 list_for_each_entry(dev, list, todo_list) 10481 if (netdev_refcnt_read(dev) == 1) 10482 return dev; 10483 10484 if (time_after(jiffies, warning_time + 10485 READ_ONCE(netdev_unregister_timeout_secs) * HZ)) { 10486 list_for_each_entry(dev, list, todo_list) { 10487 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", 10488 dev->name, netdev_refcnt_read(dev)); 10489 ref_tracker_dir_print(&dev->refcnt_tracker, 10); 10490 } 10491 10492 warning_time = jiffies; 10493 } 10494 } 10495 } 10496 10497 /* The sequence is: 10498 * 10499 * rtnl_lock(); 10500 * ... 10501 * register_netdevice(x1); 10502 * register_netdevice(x2); 10503 * ... 10504 * unregister_netdevice(y1); 10505 * unregister_netdevice(y2); 10506 * ... 10507 * rtnl_unlock(); 10508 * free_netdev(y1); 10509 * free_netdev(y2); 10510 * 10511 * We are invoked by rtnl_unlock(). 10512 * This allows us to deal with problems: 10513 * 1) We can delete sysfs objects which invoke hotplug 10514 * without deadlocking with linkwatch via keventd. 10515 * 2) Since we run with the RTNL semaphore not held, we can sleep 10516 * safely in order to wait for the netdev refcnt to drop to zero. 10517 * 10518 * We must not return until all unregister events added during 10519 * the interval the lock was held have been completed. 10520 */ 10521 void netdev_run_todo(void) 10522 { 10523 struct net_device *dev, *tmp; 10524 struct list_head list; 10525 #ifdef CONFIG_LOCKDEP 10526 struct list_head unlink_list; 10527 10528 list_replace_init(&net_unlink_list, &unlink_list); 10529 10530 while (!list_empty(&unlink_list)) { 10531 struct net_device *dev = list_first_entry(&unlink_list, 10532 struct net_device, 10533 unlink_list); 10534 list_del_init(&dev->unlink_list); 10535 dev->nested_level = dev->lower_level - 1; 10536 } 10537 #endif 10538 10539 /* Snapshot list, allow later requests */ 10540 list_replace_init(&net_todo_list, &list); 10541 10542 __rtnl_unlock(); 10543 10544 /* Wait for rcu callbacks to finish before next phase */ 10545 if (!list_empty(&list)) 10546 rcu_barrier(); 10547 10548 list_for_each_entry_safe(dev, tmp, &list, todo_list) { 10549 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { 10550 netdev_WARN(dev, "run_todo but not unregistering\n"); 10551 list_del(&dev->todo_list); 10552 continue; 10553 } 10554 10555 write_lock(&dev_base_lock); 10556 dev->reg_state = NETREG_UNREGISTERED; 10557 write_unlock(&dev_base_lock); 10558 linkwatch_sync_dev(dev); 10559 } 10560 10561 while (!list_empty(&list)) { 10562 dev = netdev_wait_allrefs_any(&list); 10563 list_del(&dev->todo_list); 10564 10565 /* paranoia */ 10566 BUG_ON(netdev_refcnt_read(dev) != 1); 10567 BUG_ON(!list_empty(&dev->ptype_all)); 10568 BUG_ON(!list_empty(&dev->ptype_specific)); 10569 WARN_ON(rcu_access_pointer(dev->ip_ptr)); 10570 WARN_ON(rcu_access_pointer(dev->ip6_ptr)); 10571 10572 netdev_do_free_pcpu_stats(dev); 10573 if (dev->priv_destructor) 10574 dev->priv_destructor(dev); 10575 if (dev->needs_free_netdev) 10576 free_netdev(dev); 10577 10578 if (atomic_dec_and_test(&dev_net(dev)->dev_unreg_count)) 10579 wake_up(&netdev_unregistering_wq); 10580 10581 /* Free network device */ 10582 kobject_put(&dev->dev.kobj); 10583 } 10584 } 10585 10586 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has 10587 * all the same fields in the same order as net_device_stats, with only 10588 * the type differing, but rtnl_link_stats64 may have additional fields 10589 * at the end for newer counters. 10590 */ 10591 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, 10592 const struct net_device_stats *netdev_stats) 10593 { 10594 size_t i, n = sizeof(*netdev_stats) / sizeof(atomic_long_t); 10595 const atomic_long_t *src = (atomic_long_t *)netdev_stats; 10596 u64 *dst = (u64 *)stats64; 10597 10598 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); 10599 for (i = 0; i < n; i++) 10600 dst[i] = (unsigned long)atomic_long_read(&src[i]); 10601 /* zero out counters that only exist in rtnl_link_stats64 */ 10602 memset((char *)stats64 + n * sizeof(u64), 0, 10603 sizeof(*stats64) - n * sizeof(u64)); 10604 } 10605 EXPORT_SYMBOL(netdev_stats_to_stats64); 10606 10607 static __cold struct net_device_core_stats __percpu *netdev_core_stats_alloc( 10608 struct net_device *dev) 10609 { 10610 struct net_device_core_stats __percpu *p; 10611 10612 p = alloc_percpu_gfp(struct net_device_core_stats, 10613 GFP_ATOMIC | __GFP_NOWARN); 10614 10615 if (p && cmpxchg(&dev->core_stats, NULL, p)) 10616 free_percpu(p); 10617 10618 /* This READ_ONCE() pairs with the cmpxchg() above */ 10619 return READ_ONCE(dev->core_stats); 10620 } 10621 10622 noinline void netdev_core_stats_inc(struct net_device *dev, u32 offset) 10623 { 10624 /* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */ 10625 struct net_device_core_stats __percpu *p = READ_ONCE(dev->core_stats); 10626 unsigned long __percpu *field; 10627 10628 if (unlikely(!p)) { 10629 p = netdev_core_stats_alloc(dev); 10630 if (!p) 10631 return; 10632 } 10633 10634 field = (__force unsigned long __percpu *)((__force void *)p + offset); 10635 this_cpu_inc(*field); 10636 } 10637 EXPORT_SYMBOL_GPL(netdev_core_stats_inc); 10638 10639 /** 10640 * dev_get_stats - get network device statistics 10641 * @dev: device to get statistics from 10642 * @storage: place to store stats 10643 * 10644 * Get network statistics from device. Return @storage. 10645 * The device driver may provide its own method by setting 10646 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; 10647 * otherwise the internal statistics structure is used. 10648 */ 10649 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, 10650 struct rtnl_link_stats64 *storage) 10651 { 10652 const struct net_device_ops *ops = dev->netdev_ops; 10653 const struct net_device_core_stats __percpu *p; 10654 10655 if (ops->ndo_get_stats64) { 10656 memset(storage, 0, sizeof(*storage)); 10657 ops->ndo_get_stats64(dev, storage); 10658 } else if (ops->ndo_get_stats) { 10659 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); 10660 } else { 10661 netdev_stats_to_stats64(storage, &dev->stats); 10662 } 10663 10664 /* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */ 10665 p = READ_ONCE(dev->core_stats); 10666 if (p) { 10667 const struct net_device_core_stats *core_stats; 10668 int i; 10669 10670 for_each_possible_cpu(i) { 10671 core_stats = per_cpu_ptr(p, i); 10672 storage->rx_dropped += READ_ONCE(core_stats->rx_dropped); 10673 storage->tx_dropped += READ_ONCE(core_stats->tx_dropped); 10674 storage->rx_nohandler += READ_ONCE(core_stats->rx_nohandler); 10675 storage->rx_otherhost_dropped += READ_ONCE(core_stats->rx_otherhost_dropped); 10676 } 10677 } 10678 return storage; 10679 } 10680 EXPORT_SYMBOL(dev_get_stats); 10681 10682 /** 10683 * dev_fetch_sw_netstats - get per-cpu network device statistics 10684 * @s: place to store stats 10685 * @netstats: per-cpu network stats to read from 10686 * 10687 * Read per-cpu network statistics and populate the related fields in @s. 10688 */ 10689 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s, 10690 const struct pcpu_sw_netstats __percpu *netstats) 10691 { 10692 int cpu; 10693 10694 for_each_possible_cpu(cpu) { 10695 u64 rx_packets, rx_bytes, tx_packets, tx_bytes; 10696 const struct pcpu_sw_netstats *stats; 10697 unsigned int start; 10698 10699 stats = per_cpu_ptr(netstats, cpu); 10700 do { 10701 start = u64_stats_fetch_begin(&stats->syncp); 10702 rx_packets = u64_stats_read(&stats->rx_packets); 10703 rx_bytes = u64_stats_read(&stats->rx_bytes); 10704 tx_packets = u64_stats_read(&stats->tx_packets); 10705 tx_bytes = u64_stats_read(&stats->tx_bytes); 10706 } while (u64_stats_fetch_retry(&stats->syncp, start)); 10707 10708 s->rx_packets += rx_packets; 10709 s->rx_bytes += rx_bytes; 10710 s->tx_packets += tx_packets; 10711 s->tx_bytes += tx_bytes; 10712 } 10713 } 10714 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats); 10715 10716 /** 10717 * dev_get_tstats64 - ndo_get_stats64 implementation 10718 * @dev: device to get statistics from 10719 * @s: place to store stats 10720 * 10721 * Populate @s from dev->stats and dev->tstats. Can be used as 10722 * ndo_get_stats64() callback. 10723 */ 10724 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s) 10725 { 10726 netdev_stats_to_stats64(s, &dev->stats); 10727 dev_fetch_sw_netstats(s, dev->tstats); 10728 } 10729 EXPORT_SYMBOL_GPL(dev_get_tstats64); 10730 10731 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) 10732 { 10733 struct netdev_queue *queue = dev_ingress_queue(dev); 10734 10735 #ifdef CONFIG_NET_CLS_ACT 10736 if (queue) 10737 return queue; 10738 queue = kzalloc(sizeof(*queue), GFP_KERNEL); 10739 if (!queue) 10740 return NULL; 10741 netdev_init_one_queue(dev, queue, NULL); 10742 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); 10743 RCU_INIT_POINTER(queue->qdisc_sleeping, &noop_qdisc); 10744 rcu_assign_pointer(dev->ingress_queue, queue); 10745 #endif 10746 return queue; 10747 } 10748 10749 static const struct ethtool_ops default_ethtool_ops; 10750 10751 void netdev_set_default_ethtool_ops(struct net_device *dev, 10752 const struct ethtool_ops *ops) 10753 { 10754 if (dev->ethtool_ops == &default_ethtool_ops) 10755 dev->ethtool_ops = ops; 10756 } 10757 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); 10758 10759 /** 10760 * netdev_sw_irq_coalesce_default_on() - enable SW IRQ coalescing by default 10761 * @dev: netdev to enable the IRQ coalescing on 10762 * 10763 * Sets a conservative default for SW IRQ coalescing. Users can use 10764 * sysfs attributes to override the default values. 10765 */ 10766 void netdev_sw_irq_coalesce_default_on(struct net_device *dev) 10767 { 10768 WARN_ON(dev->reg_state == NETREG_REGISTERED); 10769 10770 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) { 10771 dev->gro_flush_timeout = 20000; 10772 dev->napi_defer_hard_irqs = 1; 10773 } 10774 } 10775 EXPORT_SYMBOL_GPL(netdev_sw_irq_coalesce_default_on); 10776 10777 void netdev_freemem(struct net_device *dev) 10778 { 10779 char *addr = (char *)dev - dev->padded; 10780 10781 kvfree(addr); 10782 } 10783 10784 /** 10785 * alloc_netdev_mqs - allocate network device 10786 * @sizeof_priv: size of private data to allocate space for 10787 * @name: device name format string 10788 * @name_assign_type: origin of device name 10789 * @setup: callback to initialize device 10790 * @txqs: the number of TX subqueues to allocate 10791 * @rxqs: the number of RX subqueues to allocate 10792 * 10793 * Allocates a struct net_device with private data area for driver use 10794 * and performs basic initialization. Also allocates subqueue structs 10795 * for each queue on the device. 10796 */ 10797 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, 10798 unsigned char name_assign_type, 10799 void (*setup)(struct net_device *), 10800 unsigned int txqs, unsigned int rxqs) 10801 { 10802 struct net_device *dev; 10803 unsigned int alloc_size; 10804 struct net_device *p; 10805 10806 BUG_ON(strlen(name) >= sizeof(dev->name)); 10807 10808 if (txqs < 1) { 10809 pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); 10810 return NULL; 10811 } 10812 10813 if (rxqs < 1) { 10814 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); 10815 return NULL; 10816 } 10817 10818 alloc_size = sizeof(struct net_device); 10819 if (sizeof_priv) { 10820 /* ensure 32-byte alignment of private area */ 10821 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); 10822 alloc_size += sizeof_priv; 10823 } 10824 /* ensure 32-byte alignment of whole construct */ 10825 alloc_size += NETDEV_ALIGN - 1; 10826 10827 p = kvzalloc(alloc_size, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10828 if (!p) 10829 return NULL; 10830 10831 dev = PTR_ALIGN(p, NETDEV_ALIGN); 10832 dev->padded = (char *)dev - (char *)p; 10833 10834 ref_tracker_dir_init(&dev->refcnt_tracker, 128, name); 10835 #ifdef CONFIG_PCPU_DEV_REFCNT 10836 dev->pcpu_refcnt = alloc_percpu(int); 10837 if (!dev->pcpu_refcnt) 10838 goto free_dev; 10839 __dev_hold(dev); 10840 #else 10841 refcount_set(&dev->dev_refcnt, 1); 10842 #endif 10843 10844 if (dev_addr_init(dev)) 10845 goto free_pcpu; 10846 10847 dev_mc_init(dev); 10848 dev_uc_init(dev); 10849 10850 dev_net_set(dev, &init_net); 10851 10852 dev->gso_max_size = GSO_LEGACY_MAX_SIZE; 10853 dev->xdp_zc_max_segs = 1; 10854 dev->gso_max_segs = GSO_MAX_SEGS; 10855 dev->gro_max_size = GRO_LEGACY_MAX_SIZE; 10856 dev->gso_ipv4_max_size = GSO_LEGACY_MAX_SIZE; 10857 dev->gro_ipv4_max_size = GRO_LEGACY_MAX_SIZE; 10858 dev->tso_max_size = TSO_LEGACY_MAX_SIZE; 10859 dev->tso_max_segs = TSO_MAX_SEGS; 10860 dev->upper_level = 1; 10861 dev->lower_level = 1; 10862 #ifdef CONFIG_LOCKDEP 10863 dev->nested_level = 0; 10864 INIT_LIST_HEAD(&dev->unlink_list); 10865 #endif 10866 10867 INIT_LIST_HEAD(&dev->napi_list); 10868 INIT_LIST_HEAD(&dev->unreg_list); 10869 INIT_LIST_HEAD(&dev->close_list); 10870 INIT_LIST_HEAD(&dev->link_watch_list); 10871 INIT_LIST_HEAD(&dev->adj_list.upper); 10872 INIT_LIST_HEAD(&dev->adj_list.lower); 10873 INIT_LIST_HEAD(&dev->ptype_all); 10874 INIT_LIST_HEAD(&dev->ptype_specific); 10875 INIT_LIST_HEAD(&dev->net_notifier_list); 10876 #ifdef CONFIG_NET_SCHED 10877 hash_init(dev->qdisc_hash); 10878 #endif 10879 phy_link_topo_init(&dev->link_topo); 10880 10881 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; 10882 setup(dev); 10883 10884 if (!dev->tx_queue_len) { 10885 dev->priv_flags |= IFF_NO_QUEUE; 10886 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; 10887 } 10888 10889 dev->num_tx_queues = txqs; 10890 dev->real_num_tx_queues = txqs; 10891 if (netif_alloc_netdev_queues(dev)) 10892 goto free_all; 10893 10894 dev->num_rx_queues = rxqs; 10895 dev->real_num_rx_queues = rxqs; 10896 if (netif_alloc_rx_queues(dev)) 10897 goto free_all; 10898 10899 strcpy(dev->name, name); 10900 dev->name_assign_type = name_assign_type; 10901 dev->group = INIT_NETDEV_GROUP; 10902 if (!dev->ethtool_ops) 10903 dev->ethtool_ops = &default_ethtool_ops; 10904 10905 nf_hook_netdev_init(dev); 10906 10907 return dev; 10908 10909 free_all: 10910 free_netdev(dev); 10911 return NULL; 10912 10913 free_pcpu: 10914 #ifdef CONFIG_PCPU_DEV_REFCNT 10915 free_percpu(dev->pcpu_refcnt); 10916 free_dev: 10917 #endif 10918 netdev_freemem(dev); 10919 return NULL; 10920 } 10921 EXPORT_SYMBOL(alloc_netdev_mqs); 10922 10923 /** 10924 * free_netdev - free network device 10925 * @dev: device 10926 * 10927 * This function does the last stage of destroying an allocated device 10928 * interface. The reference to the device object is released. If this 10929 * is the last reference then it will be freed.Must be called in process 10930 * context. 10931 */ 10932 void free_netdev(struct net_device *dev) 10933 { 10934 struct napi_struct *p, *n; 10935 10936 might_sleep(); 10937 10938 /* When called immediately after register_netdevice() failed the unwind 10939 * handling may still be dismantling the device. Handle that case by 10940 * deferring the free. 10941 */ 10942 if (dev->reg_state == NETREG_UNREGISTERING) { 10943 ASSERT_RTNL(); 10944 dev->needs_free_netdev = true; 10945 return; 10946 } 10947 10948 netif_free_tx_queues(dev); 10949 netif_free_rx_queues(dev); 10950 10951 kfree(rcu_dereference_protected(dev->ingress_queue, 1)); 10952 10953 /* Flush device addresses */ 10954 dev_addr_flush(dev); 10955 10956 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) 10957 netif_napi_del(p); 10958 10959 ref_tracker_dir_exit(&dev->refcnt_tracker); 10960 #ifdef CONFIG_PCPU_DEV_REFCNT 10961 free_percpu(dev->pcpu_refcnt); 10962 dev->pcpu_refcnt = NULL; 10963 #endif 10964 free_percpu(dev->core_stats); 10965 dev->core_stats = NULL; 10966 free_percpu(dev->xdp_bulkq); 10967 dev->xdp_bulkq = NULL; 10968 10969 /* Compatibility with error handling in drivers */ 10970 if (dev->reg_state == NETREG_UNINITIALIZED) { 10971 netdev_freemem(dev); 10972 return; 10973 } 10974 10975 BUG_ON(dev->reg_state != NETREG_UNREGISTERED); 10976 dev->reg_state = NETREG_RELEASED; 10977 10978 /* will free via device release */ 10979 put_device(&dev->dev); 10980 } 10981 EXPORT_SYMBOL(free_netdev); 10982 10983 /** 10984 * synchronize_net - Synchronize with packet receive processing 10985 * 10986 * Wait for packets currently being received to be done. 10987 * Does not block later packets from starting. 10988 */ 10989 void synchronize_net(void) 10990 { 10991 might_sleep(); 10992 if (rtnl_is_locked()) 10993 synchronize_rcu_expedited(); 10994 else 10995 synchronize_rcu(); 10996 } 10997 EXPORT_SYMBOL(synchronize_net); 10998 10999 /** 11000 * unregister_netdevice_queue - remove device from the kernel 11001 * @dev: device 11002 * @head: list 11003 * 11004 * This function shuts down a device interface and removes it 11005 * from the kernel tables. 11006 * If head not NULL, device is queued to be unregistered later. 11007 * 11008 * Callers must hold the rtnl semaphore. You may want 11009 * unregister_netdev() instead of this. 11010 */ 11011 11012 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) 11013 { 11014 ASSERT_RTNL(); 11015 11016 if (head) { 11017 list_move_tail(&dev->unreg_list, head); 11018 } else { 11019 LIST_HEAD(single); 11020 11021 list_add(&dev->unreg_list, &single); 11022 unregister_netdevice_many(&single); 11023 } 11024 } 11025 EXPORT_SYMBOL(unregister_netdevice_queue); 11026 11027 void unregister_netdevice_many_notify(struct list_head *head, 11028 u32 portid, const struct nlmsghdr *nlh) 11029 { 11030 struct net_device *dev, *tmp; 11031 LIST_HEAD(close_head); 11032 11033 BUG_ON(dev_boot_phase); 11034 ASSERT_RTNL(); 11035 11036 if (list_empty(head)) 11037 return; 11038 11039 list_for_each_entry_safe(dev, tmp, head, unreg_list) { 11040 /* Some devices call without registering 11041 * for initialization unwind. Remove those 11042 * devices and proceed with the remaining. 11043 */ 11044 if (dev->reg_state == NETREG_UNINITIALIZED) { 11045 pr_debug("unregister_netdevice: device %s/%p never was registered\n", 11046 dev->name, dev); 11047 11048 WARN_ON(1); 11049 list_del(&dev->unreg_list); 11050 continue; 11051 } 11052 dev->dismantle = true; 11053 BUG_ON(dev->reg_state != NETREG_REGISTERED); 11054 } 11055 11056 /* If device is running, close it first. */ 11057 list_for_each_entry(dev, head, unreg_list) 11058 list_add_tail(&dev->close_list, &close_head); 11059 dev_close_many(&close_head, true); 11060 11061 list_for_each_entry(dev, head, unreg_list) { 11062 /* And unlink it from device chain. */ 11063 write_lock(&dev_base_lock); 11064 unlist_netdevice(dev, false); 11065 dev->reg_state = NETREG_UNREGISTERING; 11066 write_unlock(&dev_base_lock); 11067 } 11068 flush_all_backlogs(); 11069 11070 synchronize_net(); 11071 11072 list_for_each_entry(dev, head, unreg_list) { 11073 struct sk_buff *skb = NULL; 11074 11075 /* Shutdown queueing discipline. */ 11076 dev_shutdown(dev); 11077 dev_tcx_uninstall(dev); 11078 dev_xdp_uninstall(dev); 11079 bpf_dev_bound_netdev_unregister(dev); 11080 11081 netdev_offload_xstats_disable_all(dev); 11082 11083 /* Notify protocols, that we are about to destroy 11084 * this device. They should clean all the things. 11085 */ 11086 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 11087 11088 if (!dev->rtnl_link_ops || 11089 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 11090 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0, 11091 GFP_KERNEL, NULL, 0, 11092 portid, nlh); 11093 11094 /* 11095 * Flush the unicast and multicast chains 11096 */ 11097 dev_uc_flush(dev); 11098 dev_mc_flush(dev); 11099 11100 netdev_name_node_alt_flush(dev); 11101 netdev_name_node_free(dev->name_node); 11102 11103 call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev); 11104 11105 if (dev->netdev_ops->ndo_uninit) 11106 dev->netdev_ops->ndo_uninit(dev); 11107 11108 if (skb) 11109 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL, portid, nlh); 11110 11111 /* Notifier chain MUST detach us all upper devices. */ 11112 WARN_ON(netdev_has_any_upper_dev(dev)); 11113 WARN_ON(netdev_has_any_lower_dev(dev)); 11114 11115 /* Remove entries from kobject tree */ 11116 netdev_unregister_kobject(dev); 11117 #ifdef CONFIG_XPS 11118 /* Remove XPS queueing entries */ 11119 netif_reset_xps_queues_gt(dev, 0); 11120 #endif 11121 } 11122 11123 synchronize_net(); 11124 11125 list_for_each_entry(dev, head, unreg_list) { 11126 netdev_put(dev, &dev->dev_registered_tracker); 11127 net_set_todo(dev); 11128 } 11129 11130 list_del(head); 11131 } 11132 11133 /** 11134 * unregister_netdevice_many - unregister many devices 11135 * @head: list of devices 11136 * 11137 * Note: As most callers use a stack allocated list_head, 11138 * we force a list_del() to make sure stack wont be corrupted later. 11139 */ 11140 void unregister_netdevice_many(struct list_head *head) 11141 { 11142 unregister_netdevice_many_notify(head, 0, NULL); 11143 } 11144 EXPORT_SYMBOL(unregister_netdevice_many); 11145 11146 /** 11147 * unregister_netdev - remove device from the kernel 11148 * @dev: device 11149 * 11150 * This function shuts down a device interface and removes it 11151 * from the kernel tables. 11152 * 11153 * This is just a wrapper for unregister_netdevice that takes 11154 * the rtnl semaphore. In general you want to use this and not 11155 * unregister_netdevice. 11156 */ 11157 void unregister_netdev(struct net_device *dev) 11158 { 11159 rtnl_lock(); 11160 unregister_netdevice(dev); 11161 rtnl_unlock(); 11162 } 11163 EXPORT_SYMBOL(unregister_netdev); 11164 11165 /** 11166 * __dev_change_net_namespace - move device to different nethost namespace 11167 * @dev: device 11168 * @net: network namespace 11169 * @pat: If not NULL name pattern to try if the current device name 11170 * is already taken in the destination network namespace. 11171 * @new_ifindex: If not zero, specifies device index in the target 11172 * namespace. 11173 * 11174 * This function shuts down a device interface and moves it 11175 * to a new network namespace. On success 0 is returned, on 11176 * a failure a netagive errno code is returned. 11177 * 11178 * Callers must hold the rtnl semaphore. 11179 */ 11180 11181 int __dev_change_net_namespace(struct net_device *dev, struct net *net, 11182 const char *pat, int new_ifindex) 11183 { 11184 struct netdev_name_node *name_node; 11185 struct net *net_old = dev_net(dev); 11186 char new_name[IFNAMSIZ] = {}; 11187 int err, new_nsid; 11188 11189 ASSERT_RTNL(); 11190 11191 /* Don't allow namespace local devices to be moved. */ 11192 err = -EINVAL; 11193 if (dev->features & NETIF_F_NETNS_LOCAL) 11194 goto out; 11195 11196 /* Ensure the device has been registrered */ 11197 if (dev->reg_state != NETREG_REGISTERED) 11198 goto out; 11199 11200 /* Get out if there is nothing todo */ 11201 err = 0; 11202 if (net_eq(net_old, net)) 11203 goto out; 11204 11205 /* Pick the destination device name, and ensure 11206 * we can use it in the destination network namespace. 11207 */ 11208 err = -EEXIST; 11209 if (netdev_name_in_use(net, dev->name)) { 11210 /* We get here if we can't use the current device name */ 11211 if (!pat) 11212 goto out; 11213 err = dev_prep_valid_name(net, dev, pat, new_name, EEXIST); 11214 if (err < 0) 11215 goto out; 11216 } 11217 /* Check that none of the altnames conflicts. */ 11218 err = -EEXIST; 11219 netdev_for_each_altname(dev, name_node) 11220 if (netdev_name_in_use(net, name_node->name)) 11221 goto out; 11222 11223 /* Check that new_ifindex isn't used yet. */ 11224 if (new_ifindex) { 11225 err = dev_index_reserve(net, new_ifindex); 11226 if (err < 0) 11227 goto out; 11228 } else { 11229 /* If there is an ifindex conflict assign a new one */ 11230 err = dev_index_reserve(net, dev->ifindex); 11231 if (err == -EBUSY) 11232 err = dev_index_reserve(net, 0); 11233 if (err < 0) 11234 goto out; 11235 new_ifindex = err; 11236 } 11237 11238 /* 11239 * And now a mini version of register_netdevice unregister_netdevice. 11240 */ 11241 11242 /* If device is running close it first. */ 11243 dev_close(dev); 11244 11245 /* And unlink it from device chain */ 11246 unlist_netdevice(dev, true); 11247 11248 synchronize_net(); 11249 11250 /* Shutdown queueing discipline. */ 11251 dev_shutdown(dev); 11252 11253 /* Notify protocols, that we are about to destroy 11254 * this device. They should clean all the things. 11255 * 11256 * Note that dev->reg_state stays at NETREG_REGISTERED. 11257 * This is wanted because this way 8021q and macvlan know 11258 * the device is just moving and can keep their slaves up. 11259 */ 11260 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 11261 rcu_barrier(); 11262 11263 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL); 11264 11265 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid, 11266 new_ifindex); 11267 11268 /* 11269 * Flush the unicast and multicast chains 11270 */ 11271 dev_uc_flush(dev); 11272 dev_mc_flush(dev); 11273 11274 /* Send a netdev-removed uevent to the old namespace */ 11275 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); 11276 netdev_adjacent_del_links(dev); 11277 11278 /* Move per-net netdevice notifiers that are following the netdevice */ 11279 move_netdevice_notifiers_dev_net(dev, net); 11280 11281 /* Actually switch the network namespace */ 11282 dev_net_set(dev, net); 11283 dev->ifindex = new_ifindex; 11284 11285 if (new_name[0]) /* Rename the netdev to prepared name */ 11286 strscpy(dev->name, new_name, IFNAMSIZ); 11287 11288 /* Fixup kobjects */ 11289 dev_set_uevent_suppress(&dev->dev, 1); 11290 err = device_rename(&dev->dev, dev->name); 11291 dev_set_uevent_suppress(&dev->dev, 0); 11292 WARN_ON(err); 11293 11294 /* Send a netdev-add uevent to the new namespace */ 11295 kobject_uevent(&dev->dev.kobj, KOBJ_ADD); 11296 netdev_adjacent_add_links(dev); 11297 11298 /* Adapt owner in case owning user namespace of target network 11299 * namespace is different from the original one. 11300 */ 11301 err = netdev_change_owner(dev, net_old, net); 11302 WARN_ON(err); 11303 11304 /* Add the device back in the hashes */ 11305 list_netdevice(dev); 11306 11307 /* Notify protocols, that a new device appeared. */ 11308 call_netdevice_notifiers(NETDEV_REGISTER, dev); 11309 11310 /* 11311 * Prevent userspace races by waiting until the network 11312 * device is fully setup before sending notifications. 11313 */ 11314 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL); 11315 11316 synchronize_net(); 11317 err = 0; 11318 out: 11319 return err; 11320 } 11321 EXPORT_SYMBOL_GPL(__dev_change_net_namespace); 11322 11323 static int dev_cpu_dead(unsigned int oldcpu) 11324 { 11325 struct sk_buff **list_skb; 11326 struct sk_buff *skb; 11327 unsigned int cpu; 11328 struct softnet_data *sd, *oldsd, *remsd = NULL; 11329 11330 local_irq_disable(); 11331 cpu = smp_processor_id(); 11332 sd = &per_cpu(softnet_data, cpu); 11333 oldsd = &per_cpu(softnet_data, oldcpu); 11334 11335 /* Find end of our completion_queue. */ 11336 list_skb = &sd->completion_queue; 11337 while (*list_skb) 11338 list_skb = &(*list_skb)->next; 11339 /* Append completion queue from offline CPU. */ 11340 *list_skb = oldsd->completion_queue; 11341 oldsd->completion_queue = NULL; 11342 11343 /* Append output queue from offline CPU. */ 11344 if (oldsd->output_queue) { 11345 *sd->output_queue_tailp = oldsd->output_queue; 11346 sd->output_queue_tailp = oldsd->output_queue_tailp; 11347 oldsd->output_queue = NULL; 11348 oldsd->output_queue_tailp = &oldsd->output_queue; 11349 } 11350 /* Append NAPI poll list from offline CPU, with one exception : 11351 * process_backlog() must be called by cpu owning percpu backlog. 11352 * We properly handle process_queue & input_pkt_queue later. 11353 */ 11354 while (!list_empty(&oldsd->poll_list)) { 11355 struct napi_struct *napi = list_first_entry(&oldsd->poll_list, 11356 struct napi_struct, 11357 poll_list); 11358 11359 list_del_init(&napi->poll_list); 11360 if (napi->poll == process_backlog) 11361 napi->state = 0; 11362 else 11363 ____napi_schedule(sd, napi); 11364 } 11365 11366 raise_softirq_irqoff(NET_TX_SOFTIRQ); 11367 local_irq_enable(); 11368 11369 #ifdef CONFIG_RPS 11370 remsd = oldsd->rps_ipi_list; 11371 oldsd->rps_ipi_list = NULL; 11372 #endif 11373 /* send out pending IPI's on offline CPU */ 11374 net_rps_send_ipi(remsd); 11375 11376 /* Process offline CPU's input_pkt_queue */ 11377 while ((skb = __skb_dequeue(&oldsd->process_queue))) { 11378 netif_rx(skb); 11379 input_queue_head_incr(oldsd); 11380 } 11381 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { 11382 netif_rx(skb); 11383 input_queue_head_incr(oldsd); 11384 } 11385 11386 return 0; 11387 } 11388 11389 /** 11390 * netdev_increment_features - increment feature set by one 11391 * @all: current feature set 11392 * @one: new feature set 11393 * @mask: mask feature set 11394 * 11395 * Computes a new feature set after adding a device with feature set 11396 * @one to the master device with current feature set @all. Will not 11397 * enable anything that is off in @mask. Returns the new feature set. 11398 */ 11399 netdev_features_t netdev_increment_features(netdev_features_t all, 11400 netdev_features_t one, netdev_features_t mask) 11401 { 11402 if (mask & NETIF_F_HW_CSUM) 11403 mask |= NETIF_F_CSUM_MASK; 11404 mask |= NETIF_F_VLAN_CHALLENGED; 11405 11406 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; 11407 all &= one | ~NETIF_F_ALL_FOR_ALL; 11408 11409 /* If one device supports hw checksumming, set for all. */ 11410 if (all & NETIF_F_HW_CSUM) 11411 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); 11412 11413 return all; 11414 } 11415 EXPORT_SYMBOL(netdev_increment_features); 11416 11417 static struct hlist_head * __net_init netdev_create_hash(void) 11418 { 11419 int i; 11420 struct hlist_head *hash; 11421 11422 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL); 11423 if (hash != NULL) 11424 for (i = 0; i < NETDEV_HASHENTRIES; i++) 11425 INIT_HLIST_HEAD(&hash[i]); 11426 11427 return hash; 11428 } 11429 11430 /* Initialize per network namespace state */ 11431 static int __net_init netdev_init(struct net *net) 11432 { 11433 BUILD_BUG_ON(GRO_HASH_BUCKETS > 11434 8 * sizeof_field(struct napi_struct, gro_bitmask)); 11435 11436 INIT_LIST_HEAD(&net->dev_base_head); 11437 11438 net->dev_name_head = netdev_create_hash(); 11439 if (net->dev_name_head == NULL) 11440 goto err_name; 11441 11442 net->dev_index_head = netdev_create_hash(); 11443 if (net->dev_index_head == NULL) 11444 goto err_idx; 11445 11446 xa_init_flags(&net->dev_by_index, XA_FLAGS_ALLOC1); 11447 11448 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain); 11449 11450 return 0; 11451 11452 err_idx: 11453 kfree(net->dev_name_head); 11454 err_name: 11455 return -ENOMEM; 11456 } 11457 11458 /** 11459 * netdev_drivername - network driver for the device 11460 * @dev: network device 11461 * 11462 * Determine network driver for device. 11463 */ 11464 const char *netdev_drivername(const struct net_device *dev) 11465 { 11466 const struct device_driver *driver; 11467 const struct device *parent; 11468 const char *empty = ""; 11469 11470 parent = dev->dev.parent; 11471 if (!parent) 11472 return empty; 11473 11474 driver = parent->driver; 11475 if (driver && driver->name) 11476 return driver->name; 11477 return empty; 11478 } 11479 11480 static void __netdev_printk(const char *level, const struct net_device *dev, 11481 struct va_format *vaf) 11482 { 11483 if (dev && dev->dev.parent) { 11484 dev_printk_emit(level[1] - '0', 11485 dev->dev.parent, 11486 "%s %s %s%s: %pV", 11487 dev_driver_string(dev->dev.parent), 11488 dev_name(dev->dev.parent), 11489 netdev_name(dev), netdev_reg_state(dev), 11490 vaf); 11491 } else if (dev) { 11492 printk("%s%s%s: %pV", 11493 level, netdev_name(dev), netdev_reg_state(dev), vaf); 11494 } else { 11495 printk("%s(NULL net_device): %pV", level, vaf); 11496 } 11497 } 11498 11499 void netdev_printk(const char *level, const struct net_device *dev, 11500 const char *format, ...) 11501 { 11502 struct va_format vaf; 11503 va_list args; 11504 11505 va_start(args, format); 11506 11507 vaf.fmt = format; 11508 vaf.va = &args; 11509 11510 __netdev_printk(level, dev, &vaf); 11511 11512 va_end(args); 11513 } 11514 EXPORT_SYMBOL(netdev_printk); 11515 11516 #define define_netdev_printk_level(func, level) \ 11517 void func(const struct net_device *dev, const char *fmt, ...) \ 11518 { \ 11519 struct va_format vaf; \ 11520 va_list args; \ 11521 \ 11522 va_start(args, fmt); \ 11523 \ 11524 vaf.fmt = fmt; \ 11525 vaf.va = &args; \ 11526 \ 11527 __netdev_printk(level, dev, &vaf); \ 11528 \ 11529 va_end(args); \ 11530 } \ 11531 EXPORT_SYMBOL(func); 11532 11533 define_netdev_printk_level(netdev_emerg, KERN_EMERG); 11534 define_netdev_printk_level(netdev_alert, KERN_ALERT); 11535 define_netdev_printk_level(netdev_crit, KERN_CRIT); 11536 define_netdev_printk_level(netdev_err, KERN_ERR); 11537 define_netdev_printk_level(netdev_warn, KERN_WARNING); 11538 define_netdev_printk_level(netdev_notice, KERN_NOTICE); 11539 define_netdev_printk_level(netdev_info, KERN_INFO); 11540 11541 static void __net_exit netdev_exit(struct net *net) 11542 { 11543 kfree(net->dev_name_head); 11544 kfree(net->dev_index_head); 11545 xa_destroy(&net->dev_by_index); 11546 if (net != &init_net) 11547 WARN_ON_ONCE(!list_empty(&net->dev_base_head)); 11548 } 11549 11550 static struct pernet_operations __net_initdata netdev_net_ops = { 11551 .init = netdev_init, 11552 .exit = netdev_exit, 11553 }; 11554 11555 static void __net_exit default_device_exit_net(struct net *net) 11556 { 11557 struct net_device *dev, *aux; 11558 /* 11559 * Push all migratable network devices back to the 11560 * initial network namespace 11561 */ 11562 ASSERT_RTNL(); 11563 for_each_netdev_safe(net, dev, aux) { 11564 int err; 11565 char fb_name[IFNAMSIZ]; 11566 11567 /* Ignore unmoveable devices (i.e. loopback) */ 11568 if (dev->features & NETIF_F_NETNS_LOCAL) 11569 continue; 11570 11571 /* Leave virtual devices for the generic cleanup */ 11572 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund) 11573 continue; 11574 11575 /* Push remaining network devices to init_net */ 11576 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); 11577 if (netdev_name_in_use(&init_net, fb_name)) 11578 snprintf(fb_name, IFNAMSIZ, "dev%%d"); 11579 err = dev_change_net_namespace(dev, &init_net, fb_name); 11580 if (err) { 11581 pr_emerg("%s: failed to move %s to init_net: %d\n", 11582 __func__, dev->name, err); 11583 BUG(); 11584 } 11585 } 11586 } 11587 11588 static void __net_exit default_device_exit_batch(struct list_head *net_list) 11589 { 11590 /* At exit all network devices most be removed from a network 11591 * namespace. Do this in the reverse order of registration. 11592 * Do this across as many network namespaces as possible to 11593 * improve batching efficiency. 11594 */ 11595 struct net_device *dev; 11596 struct net *net; 11597 LIST_HEAD(dev_kill_list); 11598 11599 rtnl_lock(); 11600 list_for_each_entry(net, net_list, exit_list) { 11601 default_device_exit_net(net); 11602 cond_resched(); 11603 } 11604 11605 list_for_each_entry(net, net_list, exit_list) { 11606 for_each_netdev_reverse(net, dev) { 11607 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) 11608 dev->rtnl_link_ops->dellink(dev, &dev_kill_list); 11609 else 11610 unregister_netdevice_queue(dev, &dev_kill_list); 11611 } 11612 } 11613 unregister_netdevice_many(&dev_kill_list); 11614 rtnl_unlock(); 11615 } 11616 11617 static struct pernet_operations __net_initdata default_device_ops = { 11618 .exit_batch = default_device_exit_batch, 11619 }; 11620 11621 static void __init net_dev_struct_check(void) 11622 { 11623 /* TX read-mostly hotpath */ 11624 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, priv_flags); 11625 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, netdev_ops); 11626 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, header_ops); 11627 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, _tx); 11628 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, real_num_tx_queues); 11629 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_max_size); 11630 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_ipv4_max_size); 11631 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_max_segs); 11632 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_partial_features); 11633 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, num_tc); 11634 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, mtu); 11635 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, needed_headroom); 11636 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, tc_to_txq); 11637 #ifdef CONFIG_XPS 11638 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, xps_maps); 11639 #endif 11640 #ifdef CONFIG_NETFILTER_EGRESS 11641 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, nf_hooks_egress); 11642 #endif 11643 #ifdef CONFIG_NET_XGRESS 11644 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, tcx_egress); 11645 #endif 11646 CACHELINE_ASSERT_GROUP_SIZE(struct net_device, net_device_read_tx, 160); 11647 11648 /* TXRX read-mostly hotpath */ 11649 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, flags); 11650 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, hard_header_len); 11651 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, features); 11652 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, ip6_ptr); 11653 CACHELINE_ASSERT_GROUP_SIZE(struct net_device, net_device_read_txrx, 30); 11654 11655 /* RX read-mostly hotpath */ 11656 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, ptype_specific); 11657 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, ifindex); 11658 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, real_num_rx_queues); 11659 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, _rx); 11660 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, gro_flush_timeout); 11661 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, napi_defer_hard_irqs); 11662 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, gro_max_size); 11663 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, gro_ipv4_max_size); 11664 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, rx_handler); 11665 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, rx_handler_data); 11666 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, nd_net); 11667 #ifdef CONFIG_NETPOLL 11668 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, npinfo); 11669 #endif 11670 #ifdef CONFIG_NET_XGRESS 11671 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, tcx_ingress); 11672 #endif 11673 CACHELINE_ASSERT_GROUP_SIZE(struct net_device, net_device_read_rx, 104); 11674 } 11675 11676 /* 11677 * Initialize the DEV module. At boot time this walks the device list and 11678 * unhooks any devices that fail to initialise (normally hardware not 11679 * present) and leaves us with a valid list of present and active devices. 11680 * 11681 */ 11682 11683 /* 11684 * This is called single threaded during boot, so no need 11685 * to take the rtnl semaphore. 11686 */ 11687 static int __init net_dev_init(void) 11688 { 11689 int i, rc = -ENOMEM; 11690 11691 BUG_ON(!dev_boot_phase); 11692 11693 net_dev_struct_check(); 11694 11695 if (dev_proc_init()) 11696 goto out; 11697 11698 if (netdev_kobject_init()) 11699 goto out; 11700 11701 INIT_LIST_HEAD(&ptype_all); 11702 for (i = 0; i < PTYPE_HASH_SIZE; i++) 11703 INIT_LIST_HEAD(&ptype_base[i]); 11704 11705 if (register_pernet_subsys(&netdev_net_ops)) 11706 goto out; 11707 11708 /* 11709 * Initialise the packet receive queues. 11710 */ 11711 11712 for_each_possible_cpu(i) { 11713 struct work_struct *flush = per_cpu_ptr(&flush_works, i); 11714 struct softnet_data *sd = &per_cpu(softnet_data, i); 11715 11716 INIT_WORK(flush, flush_backlog); 11717 11718 skb_queue_head_init(&sd->input_pkt_queue); 11719 skb_queue_head_init(&sd->process_queue); 11720 #ifdef CONFIG_XFRM_OFFLOAD 11721 skb_queue_head_init(&sd->xfrm_backlog); 11722 #endif 11723 INIT_LIST_HEAD(&sd->poll_list); 11724 sd->output_queue_tailp = &sd->output_queue; 11725 #ifdef CONFIG_RPS 11726 INIT_CSD(&sd->csd, rps_trigger_softirq, sd); 11727 sd->cpu = i; 11728 #endif 11729 INIT_CSD(&sd->defer_csd, trigger_rx_softirq, sd); 11730 spin_lock_init(&sd->defer_lock); 11731 11732 init_gro_hash(&sd->backlog); 11733 sd->backlog.poll = process_backlog; 11734 sd->backlog.weight = weight_p; 11735 } 11736 11737 dev_boot_phase = 0; 11738 11739 /* The loopback device is special if any other network devices 11740 * is present in a network namespace the loopback device must 11741 * be present. Since we now dynamically allocate and free the 11742 * loopback device ensure this invariant is maintained by 11743 * keeping the loopback device as the first device on the 11744 * list of network devices. Ensuring the loopback devices 11745 * is the first device that appears and the last network device 11746 * that disappears. 11747 */ 11748 if (register_pernet_device(&loopback_net_ops)) 11749 goto out; 11750 11751 if (register_pernet_device(&default_device_ops)) 11752 goto out; 11753 11754 open_softirq(NET_TX_SOFTIRQ, net_tx_action); 11755 open_softirq(NET_RX_SOFTIRQ, net_rx_action); 11756 11757 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", 11758 NULL, dev_cpu_dead); 11759 WARN_ON(rc < 0); 11760 rc = 0; 11761 out: 11762 return rc; 11763 } 11764 11765 subsys_initcall(net_dev_init); 11766