1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * NET3 Protocol independent device support routines. 4 * 5 * Derived from the non IP parts of dev.c 1.0.19 6 * Authors: Ross Biro 7 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 8 * Mark Evans, <evansmp@uhura.aston.ac.uk> 9 * 10 * Additional Authors: 11 * Florian la Roche <rzsfl@rz.uni-sb.de> 12 * Alan Cox <gw4pts@gw4pts.ampr.org> 13 * David Hinds <dahinds@users.sourceforge.net> 14 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru> 15 * Adam Sulmicki <adam@cfar.umd.edu> 16 * Pekka Riikonen <priikone@poesidon.pspt.fi> 17 * 18 * Changes: 19 * D.J. Barrow : Fixed bug where dev->refcnt gets set 20 * to 2 if register_netdev gets called 21 * before net_dev_init & also removed a 22 * few lines of code in the process. 23 * Alan Cox : device private ioctl copies fields back. 24 * Alan Cox : Transmit queue code does relevant 25 * stunts to keep the queue safe. 26 * Alan Cox : Fixed double lock. 27 * Alan Cox : Fixed promisc NULL pointer trap 28 * ???????? : Support the full private ioctl range 29 * Alan Cox : Moved ioctl permission check into 30 * drivers 31 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI 32 * Alan Cox : 100 backlog just doesn't cut it when 33 * you start doing multicast video 8) 34 * Alan Cox : Rewrote net_bh and list manager. 35 * Alan Cox : Fix ETH_P_ALL echoback lengths. 36 * Alan Cox : Took out transmit every packet pass 37 * Saved a few bytes in the ioctl handler 38 * Alan Cox : Network driver sets packet type before 39 * calling netif_rx. Saves a function 40 * call a packet. 41 * Alan Cox : Hashed net_bh() 42 * Richard Kooijman: Timestamp fixes. 43 * Alan Cox : Wrong field in SIOCGIFDSTADDR 44 * Alan Cox : Device lock protection. 45 * Alan Cox : Fixed nasty side effect of device close 46 * changes. 47 * Rudi Cilibrasi : Pass the right thing to 48 * set_mac_address() 49 * Dave Miller : 32bit quantity for the device lock to 50 * make it work out on a Sparc. 51 * Bjorn Ekwall : Added KERNELD hack. 52 * Alan Cox : Cleaned up the backlog initialise. 53 * Craig Metz : SIOCGIFCONF fix if space for under 54 * 1 device. 55 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there 56 * is no device open function. 57 * Andi Kleen : Fix error reporting for SIOCGIFCONF 58 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF 59 * Cyrus Durgin : Cleaned for KMOD 60 * Adam Sulmicki : Bug Fix : Network Device Unload 61 * A network device unload needs to purge 62 * the backlog queue. 63 * Paul Rusty Russell : SIOCSIFNAME 64 * Pekka Riikonen : Netdev boot-time settings code 65 * Andrew Morton : Make unregister_netdevice wait 66 * indefinitely on dev->refcnt 67 * J Hadi Salim : - Backlog queue sampling 68 * - netif_rx() feedback 69 */ 70 71 #include <linux/uaccess.h> 72 #include <linux/bitmap.h> 73 #include <linux/capability.h> 74 #include <linux/cpu.h> 75 #include <linux/types.h> 76 #include <linux/kernel.h> 77 #include <linux/hash.h> 78 #include <linux/slab.h> 79 #include <linux/sched.h> 80 #include <linux/sched/mm.h> 81 #include <linux/mutex.h> 82 #include <linux/rwsem.h> 83 #include <linux/string.h> 84 #include <linux/mm.h> 85 #include <linux/socket.h> 86 #include <linux/sockios.h> 87 #include <linux/errno.h> 88 #include <linux/interrupt.h> 89 #include <linux/if_ether.h> 90 #include <linux/netdevice.h> 91 #include <linux/etherdevice.h> 92 #include <linux/ethtool.h> 93 #include <linux/skbuff.h> 94 #include <linux/kthread.h> 95 #include <linux/bpf.h> 96 #include <linux/bpf_trace.h> 97 #include <net/net_namespace.h> 98 #include <net/sock.h> 99 #include <net/busy_poll.h> 100 #include <linux/rtnetlink.h> 101 #include <linux/stat.h> 102 #include <net/dsa.h> 103 #include <net/dst.h> 104 #include <net/dst_metadata.h> 105 #include <net/gro.h> 106 #include <net/pkt_sched.h> 107 #include <net/pkt_cls.h> 108 #include <net/checksum.h> 109 #include <net/xfrm.h> 110 #include <net/tcx.h> 111 #include <linux/highmem.h> 112 #include <linux/init.h> 113 #include <linux/module.h> 114 #include <linux/netpoll.h> 115 #include <linux/rcupdate.h> 116 #include <linux/delay.h> 117 #include <net/iw_handler.h> 118 #include <asm/current.h> 119 #include <linux/audit.h> 120 #include <linux/dmaengine.h> 121 #include <linux/err.h> 122 #include <linux/ctype.h> 123 #include <linux/if_arp.h> 124 #include <linux/if_vlan.h> 125 #include <linux/ip.h> 126 #include <net/ip.h> 127 #include <net/mpls.h> 128 #include <linux/ipv6.h> 129 #include <linux/in.h> 130 #include <linux/jhash.h> 131 #include <linux/random.h> 132 #include <trace/events/napi.h> 133 #include <trace/events/net.h> 134 #include <trace/events/skb.h> 135 #include <trace/events/qdisc.h> 136 #include <trace/events/xdp.h> 137 #include <linux/inetdevice.h> 138 #include <linux/cpu_rmap.h> 139 #include <linux/static_key.h> 140 #include <linux/hashtable.h> 141 #include <linux/vmalloc.h> 142 #include <linux/if_macvlan.h> 143 #include <linux/errqueue.h> 144 #include <linux/hrtimer.h> 145 #include <linux/netfilter_netdev.h> 146 #include <linux/crash_dump.h> 147 #include <linux/sctp.h> 148 #include <net/udp_tunnel.h> 149 #include <linux/net_namespace.h> 150 #include <linux/indirect_call_wrapper.h> 151 #include <net/devlink.h> 152 #include <linux/pm_runtime.h> 153 #include <linux/prandom.h> 154 #include <linux/once_lite.h> 155 #include <net/netdev_rx_queue.h> 156 157 #include "dev.h" 158 #include "net-sysfs.h" 159 160 static DEFINE_SPINLOCK(ptype_lock); 161 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; 162 struct list_head ptype_all __read_mostly; /* Taps */ 163 164 static int netif_rx_internal(struct sk_buff *skb); 165 static int call_netdevice_notifiers_extack(unsigned long val, 166 struct net_device *dev, 167 struct netlink_ext_ack *extack); 168 static struct napi_struct *napi_by_id(unsigned int napi_id); 169 170 /* 171 * The @dev_base_head list is protected by @dev_base_lock and the rtnl 172 * semaphore. 173 * 174 * Pure readers hold dev_base_lock for reading, or rcu_read_lock() 175 * 176 * Writers must hold the rtnl semaphore while they loop through the 177 * dev_base_head list, and hold dev_base_lock for writing when they do the 178 * actual updates. This allows pure readers to access the list even 179 * while a writer is preparing to update it. 180 * 181 * To put it another way, dev_base_lock is held for writing only to 182 * protect against pure readers; the rtnl semaphore provides the 183 * protection against other writers. 184 * 185 * See, for example usages, register_netdevice() and 186 * unregister_netdevice(), which must be called with the rtnl 187 * semaphore held. 188 */ 189 DEFINE_RWLOCK(dev_base_lock); 190 EXPORT_SYMBOL(dev_base_lock); 191 192 static DEFINE_MUTEX(ifalias_mutex); 193 194 /* protects napi_hash addition/deletion and napi_gen_id */ 195 static DEFINE_SPINLOCK(napi_hash_lock); 196 197 static unsigned int napi_gen_id = NR_CPUS; 198 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8); 199 200 static DECLARE_RWSEM(devnet_rename_sem); 201 202 static inline void dev_base_seq_inc(struct net *net) 203 { 204 while (++net->dev_base_seq == 0) 205 ; 206 } 207 208 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) 209 { 210 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ)); 211 212 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; 213 } 214 215 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) 216 { 217 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; 218 } 219 220 static inline void rps_lock_irqsave(struct softnet_data *sd, 221 unsigned long *flags) 222 { 223 if (IS_ENABLED(CONFIG_RPS)) 224 spin_lock_irqsave(&sd->input_pkt_queue.lock, *flags); 225 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 226 local_irq_save(*flags); 227 } 228 229 static inline void rps_lock_irq_disable(struct softnet_data *sd) 230 { 231 if (IS_ENABLED(CONFIG_RPS)) 232 spin_lock_irq(&sd->input_pkt_queue.lock); 233 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 234 local_irq_disable(); 235 } 236 237 static inline void rps_unlock_irq_restore(struct softnet_data *sd, 238 unsigned long *flags) 239 { 240 if (IS_ENABLED(CONFIG_RPS)) 241 spin_unlock_irqrestore(&sd->input_pkt_queue.lock, *flags); 242 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 243 local_irq_restore(*flags); 244 } 245 246 static inline void rps_unlock_irq_enable(struct softnet_data *sd) 247 { 248 if (IS_ENABLED(CONFIG_RPS)) 249 spin_unlock_irq(&sd->input_pkt_queue.lock); 250 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 251 local_irq_enable(); 252 } 253 254 static struct netdev_name_node *netdev_name_node_alloc(struct net_device *dev, 255 const char *name) 256 { 257 struct netdev_name_node *name_node; 258 259 name_node = kmalloc(sizeof(*name_node), GFP_KERNEL); 260 if (!name_node) 261 return NULL; 262 INIT_HLIST_NODE(&name_node->hlist); 263 name_node->dev = dev; 264 name_node->name = name; 265 return name_node; 266 } 267 268 static struct netdev_name_node * 269 netdev_name_node_head_alloc(struct net_device *dev) 270 { 271 struct netdev_name_node *name_node; 272 273 name_node = netdev_name_node_alloc(dev, dev->name); 274 if (!name_node) 275 return NULL; 276 INIT_LIST_HEAD(&name_node->list); 277 return name_node; 278 } 279 280 static void netdev_name_node_free(struct netdev_name_node *name_node) 281 { 282 kfree(name_node); 283 } 284 285 static void netdev_name_node_add(struct net *net, 286 struct netdev_name_node *name_node) 287 { 288 hlist_add_head_rcu(&name_node->hlist, 289 dev_name_hash(net, name_node->name)); 290 } 291 292 static void netdev_name_node_del(struct netdev_name_node *name_node) 293 { 294 hlist_del_rcu(&name_node->hlist); 295 } 296 297 static struct netdev_name_node *netdev_name_node_lookup(struct net *net, 298 const char *name) 299 { 300 struct hlist_head *head = dev_name_hash(net, name); 301 struct netdev_name_node *name_node; 302 303 hlist_for_each_entry(name_node, head, hlist) 304 if (!strcmp(name_node->name, name)) 305 return name_node; 306 return NULL; 307 } 308 309 static struct netdev_name_node *netdev_name_node_lookup_rcu(struct net *net, 310 const char *name) 311 { 312 struct hlist_head *head = dev_name_hash(net, name); 313 struct netdev_name_node *name_node; 314 315 hlist_for_each_entry_rcu(name_node, head, hlist) 316 if (!strcmp(name_node->name, name)) 317 return name_node; 318 return NULL; 319 } 320 321 bool netdev_name_in_use(struct net *net, const char *name) 322 { 323 return netdev_name_node_lookup(net, name); 324 } 325 EXPORT_SYMBOL(netdev_name_in_use); 326 327 int netdev_name_node_alt_create(struct net_device *dev, const char *name) 328 { 329 struct netdev_name_node *name_node; 330 struct net *net = dev_net(dev); 331 332 name_node = netdev_name_node_lookup(net, name); 333 if (name_node) 334 return -EEXIST; 335 name_node = netdev_name_node_alloc(dev, name); 336 if (!name_node) 337 return -ENOMEM; 338 netdev_name_node_add(net, name_node); 339 /* The node that holds dev->name acts as a head of per-device list. */ 340 list_add_tail(&name_node->list, &dev->name_node->list); 341 342 return 0; 343 } 344 345 static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node) 346 { 347 list_del(&name_node->list); 348 kfree(name_node->name); 349 netdev_name_node_free(name_node); 350 } 351 352 int netdev_name_node_alt_destroy(struct net_device *dev, const char *name) 353 { 354 struct netdev_name_node *name_node; 355 struct net *net = dev_net(dev); 356 357 name_node = netdev_name_node_lookup(net, name); 358 if (!name_node) 359 return -ENOENT; 360 /* lookup might have found our primary name or a name belonging 361 * to another device. 362 */ 363 if (name_node == dev->name_node || name_node->dev != dev) 364 return -EINVAL; 365 366 netdev_name_node_del(name_node); 367 synchronize_rcu(); 368 __netdev_name_node_alt_destroy(name_node); 369 370 return 0; 371 } 372 373 static void netdev_name_node_alt_flush(struct net_device *dev) 374 { 375 struct netdev_name_node *name_node, *tmp; 376 377 list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list) 378 __netdev_name_node_alt_destroy(name_node); 379 } 380 381 /* Device list insertion */ 382 static void list_netdevice(struct net_device *dev) 383 { 384 struct netdev_name_node *name_node; 385 struct net *net = dev_net(dev); 386 387 ASSERT_RTNL(); 388 389 write_lock(&dev_base_lock); 390 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); 391 netdev_name_node_add(net, dev->name_node); 392 hlist_add_head_rcu(&dev->index_hlist, 393 dev_index_hash(net, dev->ifindex)); 394 write_unlock(&dev_base_lock); 395 396 netdev_for_each_altname(dev, name_node) 397 netdev_name_node_add(net, name_node); 398 399 /* We reserved the ifindex, this can't fail */ 400 WARN_ON(xa_store(&net->dev_by_index, dev->ifindex, dev, GFP_KERNEL)); 401 402 dev_base_seq_inc(net); 403 } 404 405 /* Device list removal 406 * caller must respect a RCU grace period before freeing/reusing dev 407 */ 408 static void unlist_netdevice(struct net_device *dev, bool lock) 409 { 410 struct netdev_name_node *name_node; 411 struct net *net = dev_net(dev); 412 413 ASSERT_RTNL(); 414 415 xa_erase(&net->dev_by_index, dev->ifindex); 416 417 netdev_for_each_altname(dev, name_node) 418 netdev_name_node_del(name_node); 419 420 /* Unlink dev from the device chain */ 421 if (lock) 422 write_lock(&dev_base_lock); 423 list_del_rcu(&dev->dev_list); 424 netdev_name_node_del(dev->name_node); 425 hlist_del_rcu(&dev->index_hlist); 426 if (lock) 427 write_unlock(&dev_base_lock); 428 429 dev_base_seq_inc(dev_net(dev)); 430 } 431 432 /* 433 * Our notifier list 434 */ 435 436 static RAW_NOTIFIER_HEAD(netdev_chain); 437 438 /* 439 * Device drivers call our routines to queue packets here. We empty the 440 * queue in the local softnet handler. 441 */ 442 443 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); 444 EXPORT_PER_CPU_SYMBOL(softnet_data); 445 446 #ifdef CONFIG_LOCKDEP 447 /* 448 * register_netdevice() inits txq->_xmit_lock and sets lockdep class 449 * according to dev->type 450 */ 451 static const unsigned short netdev_lock_type[] = { 452 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, 453 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, 454 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, 455 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, 456 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, 457 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, 458 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, 459 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, 460 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, 461 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, 462 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, 463 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, 464 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, 465 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, 466 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; 467 468 static const char *const netdev_lock_name[] = { 469 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", 470 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", 471 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", 472 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", 473 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", 474 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", 475 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", 476 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", 477 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", 478 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", 479 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", 480 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", 481 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", 482 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", 483 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; 484 485 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; 486 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; 487 488 static inline unsigned short netdev_lock_pos(unsigned short dev_type) 489 { 490 int i; 491 492 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) 493 if (netdev_lock_type[i] == dev_type) 494 return i; 495 /* the last key is used by default */ 496 return ARRAY_SIZE(netdev_lock_type) - 1; 497 } 498 499 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 500 unsigned short dev_type) 501 { 502 int i; 503 504 i = netdev_lock_pos(dev_type); 505 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], 506 netdev_lock_name[i]); 507 } 508 509 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 510 { 511 int i; 512 513 i = netdev_lock_pos(dev->type); 514 lockdep_set_class_and_name(&dev->addr_list_lock, 515 &netdev_addr_lock_key[i], 516 netdev_lock_name[i]); 517 } 518 #else 519 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 520 unsigned short dev_type) 521 { 522 } 523 524 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 525 { 526 } 527 #endif 528 529 /******************************************************************************* 530 * 531 * Protocol management and registration routines 532 * 533 *******************************************************************************/ 534 535 536 /* 537 * Add a protocol ID to the list. Now that the input handler is 538 * smarter we can dispense with all the messy stuff that used to be 539 * here. 540 * 541 * BEWARE!!! Protocol handlers, mangling input packets, 542 * MUST BE last in hash buckets and checking protocol handlers 543 * MUST start from promiscuous ptype_all chain in net_bh. 544 * It is true now, do not change it. 545 * Explanation follows: if protocol handler, mangling packet, will 546 * be the first on list, it is not able to sense, that packet 547 * is cloned and should be copied-on-write, so that it will 548 * change it and subsequent readers will get broken packet. 549 * --ANK (980803) 550 */ 551 552 static inline struct list_head *ptype_head(const struct packet_type *pt) 553 { 554 if (pt->type == htons(ETH_P_ALL)) 555 return pt->dev ? &pt->dev->ptype_all : &ptype_all; 556 else 557 return pt->dev ? &pt->dev->ptype_specific : 558 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; 559 } 560 561 /** 562 * dev_add_pack - add packet handler 563 * @pt: packet type declaration 564 * 565 * Add a protocol handler to the networking stack. The passed &packet_type 566 * is linked into kernel lists and may not be freed until it has been 567 * removed from the kernel lists. 568 * 569 * This call does not sleep therefore it can not 570 * guarantee all CPU's that are in middle of receiving packets 571 * will see the new packet type (until the next received packet). 572 */ 573 574 void dev_add_pack(struct packet_type *pt) 575 { 576 struct list_head *head = ptype_head(pt); 577 578 spin_lock(&ptype_lock); 579 list_add_rcu(&pt->list, head); 580 spin_unlock(&ptype_lock); 581 } 582 EXPORT_SYMBOL(dev_add_pack); 583 584 /** 585 * __dev_remove_pack - remove packet handler 586 * @pt: packet type declaration 587 * 588 * Remove a protocol handler that was previously added to the kernel 589 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 590 * from the kernel lists and can be freed or reused once this function 591 * returns. 592 * 593 * The packet type might still be in use by receivers 594 * and must not be freed until after all the CPU's have gone 595 * through a quiescent state. 596 */ 597 void __dev_remove_pack(struct packet_type *pt) 598 { 599 struct list_head *head = ptype_head(pt); 600 struct packet_type *pt1; 601 602 spin_lock(&ptype_lock); 603 604 list_for_each_entry(pt1, head, list) { 605 if (pt == pt1) { 606 list_del_rcu(&pt->list); 607 goto out; 608 } 609 } 610 611 pr_warn("dev_remove_pack: %p not found\n", pt); 612 out: 613 spin_unlock(&ptype_lock); 614 } 615 EXPORT_SYMBOL(__dev_remove_pack); 616 617 /** 618 * dev_remove_pack - remove packet handler 619 * @pt: packet type declaration 620 * 621 * Remove a protocol handler that was previously added to the kernel 622 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 623 * from the kernel lists and can be freed or reused once this function 624 * returns. 625 * 626 * This call sleeps to guarantee that no CPU is looking at the packet 627 * type after return. 628 */ 629 void dev_remove_pack(struct packet_type *pt) 630 { 631 __dev_remove_pack(pt); 632 633 synchronize_net(); 634 } 635 EXPORT_SYMBOL(dev_remove_pack); 636 637 638 /******************************************************************************* 639 * 640 * Device Interface Subroutines 641 * 642 *******************************************************************************/ 643 644 /** 645 * dev_get_iflink - get 'iflink' value of a interface 646 * @dev: targeted interface 647 * 648 * Indicates the ifindex the interface is linked to. 649 * Physical interfaces have the same 'ifindex' and 'iflink' values. 650 */ 651 652 int dev_get_iflink(const struct net_device *dev) 653 { 654 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink) 655 return dev->netdev_ops->ndo_get_iflink(dev); 656 657 return dev->ifindex; 658 } 659 EXPORT_SYMBOL(dev_get_iflink); 660 661 /** 662 * dev_fill_metadata_dst - Retrieve tunnel egress information. 663 * @dev: targeted interface 664 * @skb: The packet. 665 * 666 * For better visibility of tunnel traffic OVS needs to retrieve 667 * egress tunnel information for a packet. Following API allows 668 * user to get this info. 669 */ 670 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) 671 { 672 struct ip_tunnel_info *info; 673 674 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst) 675 return -EINVAL; 676 677 info = skb_tunnel_info_unclone(skb); 678 if (!info) 679 return -ENOMEM; 680 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX))) 681 return -EINVAL; 682 683 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb); 684 } 685 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst); 686 687 static struct net_device_path *dev_fwd_path(struct net_device_path_stack *stack) 688 { 689 int k = stack->num_paths++; 690 691 if (WARN_ON_ONCE(k >= NET_DEVICE_PATH_STACK_MAX)) 692 return NULL; 693 694 return &stack->path[k]; 695 } 696 697 int dev_fill_forward_path(const struct net_device *dev, const u8 *daddr, 698 struct net_device_path_stack *stack) 699 { 700 const struct net_device *last_dev; 701 struct net_device_path_ctx ctx = { 702 .dev = dev, 703 }; 704 struct net_device_path *path; 705 int ret = 0; 706 707 memcpy(ctx.daddr, daddr, sizeof(ctx.daddr)); 708 stack->num_paths = 0; 709 while (ctx.dev && ctx.dev->netdev_ops->ndo_fill_forward_path) { 710 last_dev = ctx.dev; 711 path = dev_fwd_path(stack); 712 if (!path) 713 return -1; 714 715 memset(path, 0, sizeof(struct net_device_path)); 716 ret = ctx.dev->netdev_ops->ndo_fill_forward_path(&ctx, path); 717 if (ret < 0) 718 return -1; 719 720 if (WARN_ON_ONCE(last_dev == ctx.dev)) 721 return -1; 722 } 723 724 if (!ctx.dev) 725 return ret; 726 727 path = dev_fwd_path(stack); 728 if (!path) 729 return -1; 730 path->type = DEV_PATH_ETHERNET; 731 path->dev = ctx.dev; 732 733 return ret; 734 } 735 EXPORT_SYMBOL_GPL(dev_fill_forward_path); 736 737 /** 738 * __dev_get_by_name - find a device by its name 739 * @net: the applicable net namespace 740 * @name: name to find 741 * 742 * Find an interface by name. Must be called under RTNL semaphore 743 * or @dev_base_lock. If the name is found a pointer to the device 744 * is returned. If the name is not found then %NULL is returned. The 745 * reference counters are not incremented so the caller must be 746 * careful with locks. 747 */ 748 749 struct net_device *__dev_get_by_name(struct net *net, const char *name) 750 { 751 struct netdev_name_node *node_name; 752 753 node_name = netdev_name_node_lookup(net, name); 754 return node_name ? node_name->dev : NULL; 755 } 756 EXPORT_SYMBOL(__dev_get_by_name); 757 758 /** 759 * dev_get_by_name_rcu - find a device by its name 760 * @net: the applicable net namespace 761 * @name: name to find 762 * 763 * Find an interface by name. 764 * If the name is found a pointer to the device is returned. 765 * If the name is not found then %NULL is returned. 766 * The reference counters are not incremented so the caller must be 767 * careful with locks. The caller must hold RCU lock. 768 */ 769 770 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) 771 { 772 struct netdev_name_node *node_name; 773 774 node_name = netdev_name_node_lookup_rcu(net, name); 775 return node_name ? node_name->dev : NULL; 776 } 777 EXPORT_SYMBOL(dev_get_by_name_rcu); 778 779 /* Deprecated for new users, call netdev_get_by_name() instead */ 780 struct net_device *dev_get_by_name(struct net *net, const char *name) 781 { 782 struct net_device *dev; 783 784 rcu_read_lock(); 785 dev = dev_get_by_name_rcu(net, name); 786 dev_hold(dev); 787 rcu_read_unlock(); 788 return dev; 789 } 790 EXPORT_SYMBOL(dev_get_by_name); 791 792 /** 793 * netdev_get_by_name() - find a device by its name 794 * @net: the applicable net namespace 795 * @name: name to find 796 * @tracker: tracking object for the acquired reference 797 * @gfp: allocation flags for the tracker 798 * 799 * Find an interface by name. This can be called from any 800 * context and does its own locking. The returned handle has 801 * the usage count incremented and the caller must use netdev_put() to 802 * release it when it is no longer needed. %NULL is returned if no 803 * matching device is found. 804 */ 805 struct net_device *netdev_get_by_name(struct net *net, const char *name, 806 netdevice_tracker *tracker, gfp_t gfp) 807 { 808 struct net_device *dev; 809 810 dev = dev_get_by_name(net, name); 811 if (dev) 812 netdev_tracker_alloc(dev, tracker, gfp); 813 return dev; 814 } 815 EXPORT_SYMBOL(netdev_get_by_name); 816 817 /** 818 * __dev_get_by_index - find a device by its ifindex 819 * @net: the applicable net namespace 820 * @ifindex: index of device 821 * 822 * Search for an interface by index. Returns %NULL if the device 823 * is not found or a pointer to the device. The device has not 824 * had its reference counter increased so the caller must be careful 825 * about locking. The caller must hold either the RTNL semaphore 826 * or @dev_base_lock. 827 */ 828 829 struct net_device *__dev_get_by_index(struct net *net, int ifindex) 830 { 831 struct net_device *dev; 832 struct hlist_head *head = dev_index_hash(net, ifindex); 833 834 hlist_for_each_entry(dev, head, index_hlist) 835 if (dev->ifindex == ifindex) 836 return dev; 837 838 return NULL; 839 } 840 EXPORT_SYMBOL(__dev_get_by_index); 841 842 /** 843 * dev_get_by_index_rcu - find a device by its ifindex 844 * @net: the applicable net namespace 845 * @ifindex: index of device 846 * 847 * Search for an interface by index. Returns %NULL if the device 848 * is not found or a pointer to the device. The device has not 849 * had its reference counter increased so the caller must be careful 850 * about locking. The caller must hold RCU lock. 851 */ 852 853 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) 854 { 855 struct net_device *dev; 856 struct hlist_head *head = dev_index_hash(net, ifindex); 857 858 hlist_for_each_entry_rcu(dev, head, index_hlist) 859 if (dev->ifindex == ifindex) 860 return dev; 861 862 return NULL; 863 } 864 EXPORT_SYMBOL(dev_get_by_index_rcu); 865 866 /* Deprecated for new users, call netdev_get_by_index() instead */ 867 struct net_device *dev_get_by_index(struct net *net, int ifindex) 868 { 869 struct net_device *dev; 870 871 rcu_read_lock(); 872 dev = dev_get_by_index_rcu(net, ifindex); 873 dev_hold(dev); 874 rcu_read_unlock(); 875 return dev; 876 } 877 EXPORT_SYMBOL(dev_get_by_index); 878 879 /** 880 * netdev_get_by_index() - find a device by its ifindex 881 * @net: the applicable net namespace 882 * @ifindex: index of device 883 * @tracker: tracking object for the acquired reference 884 * @gfp: allocation flags for the tracker 885 * 886 * Search for an interface by index. Returns NULL if the device 887 * is not found or a pointer to the device. The device returned has 888 * had a reference added and the pointer is safe until the user calls 889 * netdev_put() to indicate they have finished with it. 890 */ 891 struct net_device *netdev_get_by_index(struct net *net, int ifindex, 892 netdevice_tracker *tracker, gfp_t gfp) 893 { 894 struct net_device *dev; 895 896 dev = dev_get_by_index(net, ifindex); 897 if (dev) 898 netdev_tracker_alloc(dev, tracker, gfp); 899 return dev; 900 } 901 EXPORT_SYMBOL(netdev_get_by_index); 902 903 /** 904 * dev_get_by_napi_id - find a device by napi_id 905 * @napi_id: ID of the NAPI struct 906 * 907 * Search for an interface by NAPI ID. Returns %NULL if the device 908 * is not found or a pointer to the device. The device has not had 909 * its reference counter increased so the caller must be careful 910 * about locking. The caller must hold RCU lock. 911 */ 912 913 struct net_device *dev_get_by_napi_id(unsigned int napi_id) 914 { 915 struct napi_struct *napi; 916 917 WARN_ON_ONCE(!rcu_read_lock_held()); 918 919 if (napi_id < MIN_NAPI_ID) 920 return NULL; 921 922 napi = napi_by_id(napi_id); 923 924 return napi ? napi->dev : NULL; 925 } 926 EXPORT_SYMBOL(dev_get_by_napi_id); 927 928 /** 929 * netdev_get_name - get a netdevice name, knowing its ifindex. 930 * @net: network namespace 931 * @name: a pointer to the buffer where the name will be stored. 932 * @ifindex: the ifindex of the interface to get the name from. 933 */ 934 int netdev_get_name(struct net *net, char *name, int ifindex) 935 { 936 struct net_device *dev; 937 int ret; 938 939 down_read(&devnet_rename_sem); 940 rcu_read_lock(); 941 942 dev = dev_get_by_index_rcu(net, ifindex); 943 if (!dev) { 944 ret = -ENODEV; 945 goto out; 946 } 947 948 strcpy(name, dev->name); 949 950 ret = 0; 951 out: 952 rcu_read_unlock(); 953 up_read(&devnet_rename_sem); 954 return ret; 955 } 956 957 /** 958 * dev_getbyhwaddr_rcu - find a device by its hardware address 959 * @net: the applicable net namespace 960 * @type: media type of device 961 * @ha: hardware address 962 * 963 * Search for an interface by MAC address. Returns NULL if the device 964 * is not found or a pointer to the device. 965 * The caller must hold RCU or RTNL. 966 * The returned device has not had its ref count increased 967 * and the caller must therefore be careful about locking 968 * 969 */ 970 971 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, 972 const char *ha) 973 { 974 struct net_device *dev; 975 976 for_each_netdev_rcu(net, dev) 977 if (dev->type == type && 978 !memcmp(dev->dev_addr, ha, dev->addr_len)) 979 return dev; 980 981 return NULL; 982 } 983 EXPORT_SYMBOL(dev_getbyhwaddr_rcu); 984 985 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) 986 { 987 struct net_device *dev, *ret = NULL; 988 989 rcu_read_lock(); 990 for_each_netdev_rcu(net, dev) 991 if (dev->type == type) { 992 dev_hold(dev); 993 ret = dev; 994 break; 995 } 996 rcu_read_unlock(); 997 return ret; 998 } 999 EXPORT_SYMBOL(dev_getfirstbyhwtype); 1000 1001 /** 1002 * __dev_get_by_flags - find any device with given flags 1003 * @net: the applicable net namespace 1004 * @if_flags: IFF_* values 1005 * @mask: bitmask of bits in if_flags to check 1006 * 1007 * Search for any interface with the given flags. Returns NULL if a device 1008 * is not found or a pointer to the device. Must be called inside 1009 * rtnl_lock(), and result refcount is unchanged. 1010 */ 1011 1012 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags, 1013 unsigned short mask) 1014 { 1015 struct net_device *dev, *ret; 1016 1017 ASSERT_RTNL(); 1018 1019 ret = NULL; 1020 for_each_netdev(net, dev) { 1021 if (((dev->flags ^ if_flags) & mask) == 0) { 1022 ret = dev; 1023 break; 1024 } 1025 } 1026 return ret; 1027 } 1028 EXPORT_SYMBOL(__dev_get_by_flags); 1029 1030 /** 1031 * dev_valid_name - check if name is okay for network device 1032 * @name: name string 1033 * 1034 * Network device names need to be valid file names to 1035 * allow sysfs to work. We also disallow any kind of 1036 * whitespace. 1037 */ 1038 bool dev_valid_name(const char *name) 1039 { 1040 if (*name == '\0') 1041 return false; 1042 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ) 1043 return false; 1044 if (!strcmp(name, ".") || !strcmp(name, "..")) 1045 return false; 1046 1047 while (*name) { 1048 if (*name == '/' || *name == ':' || isspace(*name)) 1049 return false; 1050 name++; 1051 } 1052 return true; 1053 } 1054 EXPORT_SYMBOL(dev_valid_name); 1055 1056 /** 1057 * __dev_alloc_name - allocate a name for a device 1058 * @net: network namespace to allocate the device name in 1059 * @name: name format string 1060 * @buf: scratch buffer and result name string 1061 * 1062 * Passed a format string - eg "lt%d" it will try and find a suitable 1063 * id. It scans list of devices to build up a free map, then chooses 1064 * the first empty slot. The caller must hold the dev_base or rtnl lock 1065 * while allocating the name and adding the device in order to avoid 1066 * duplicates. 1067 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1068 * Returns the number of the unit assigned or a negative errno code. 1069 */ 1070 1071 static int __dev_alloc_name(struct net *net, const char *name, char *buf) 1072 { 1073 int i = 0; 1074 const char *p; 1075 const int max_netdevices = 8*PAGE_SIZE; 1076 unsigned long *inuse; 1077 struct net_device *d; 1078 1079 if (!dev_valid_name(name)) 1080 return -EINVAL; 1081 1082 p = strchr(name, '%'); 1083 if (p) { 1084 /* 1085 * Verify the string as this thing may have come from 1086 * the user. There must be either one "%d" and no other "%" 1087 * characters. 1088 */ 1089 if (p[1] != 'd' || strchr(p + 2, '%')) 1090 return -EINVAL; 1091 1092 /* Use one page as a bit array of possible slots */ 1093 inuse = bitmap_zalloc(max_netdevices, GFP_ATOMIC); 1094 if (!inuse) 1095 return -ENOMEM; 1096 1097 for_each_netdev(net, d) { 1098 struct netdev_name_node *name_node; 1099 1100 netdev_for_each_altname(d, name_node) { 1101 if (!sscanf(name_node->name, name, &i)) 1102 continue; 1103 if (i < 0 || i >= max_netdevices) 1104 continue; 1105 1106 /* avoid cases where sscanf is not exact inverse of printf */ 1107 snprintf(buf, IFNAMSIZ, name, i); 1108 if (!strncmp(buf, name_node->name, IFNAMSIZ)) 1109 __set_bit(i, inuse); 1110 } 1111 if (!sscanf(d->name, name, &i)) 1112 continue; 1113 if (i < 0 || i >= max_netdevices) 1114 continue; 1115 1116 /* avoid cases where sscanf is not exact inverse of printf */ 1117 snprintf(buf, IFNAMSIZ, name, i); 1118 if (!strncmp(buf, d->name, IFNAMSIZ)) 1119 __set_bit(i, inuse); 1120 } 1121 1122 i = find_first_zero_bit(inuse, max_netdevices); 1123 bitmap_free(inuse); 1124 } 1125 1126 snprintf(buf, IFNAMSIZ, name, i); 1127 if (!netdev_name_in_use(net, buf)) 1128 return i; 1129 1130 /* It is possible to run out of possible slots 1131 * when the name is long and there isn't enough space left 1132 * for the digits, or if all bits are used. 1133 */ 1134 return -ENFILE; 1135 } 1136 1137 static int dev_prep_valid_name(struct net *net, struct net_device *dev, 1138 const char *want_name, char *out_name) 1139 { 1140 int ret; 1141 1142 if (!dev_valid_name(want_name)) 1143 return -EINVAL; 1144 1145 if (strchr(want_name, '%')) { 1146 ret = __dev_alloc_name(net, want_name, out_name); 1147 return ret < 0 ? ret : 0; 1148 } else if (netdev_name_in_use(net, want_name)) { 1149 return -EEXIST; 1150 } else if (out_name != want_name) { 1151 strscpy(out_name, want_name, IFNAMSIZ); 1152 } 1153 1154 return 0; 1155 } 1156 1157 static int dev_alloc_name_ns(struct net *net, 1158 struct net_device *dev, 1159 const char *name) 1160 { 1161 char buf[IFNAMSIZ]; 1162 int ret; 1163 1164 BUG_ON(!net); 1165 ret = __dev_alloc_name(net, name, buf); 1166 if (ret >= 0) 1167 strscpy(dev->name, buf, IFNAMSIZ); 1168 return ret; 1169 } 1170 1171 /** 1172 * dev_alloc_name - allocate a name for a device 1173 * @dev: device 1174 * @name: name format string 1175 * 1176 * Passed a format string - eg "lt%d" it will try and find a suitable 1177 * id. It scans list of devices to build up a free map, then chooses 1178 * the first empty slot. The caller must hold the dev_base or rtnl lock 1179 * while allocating the name and adding the device in order to avoid 1180 * duplicates. 1181 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1182 * Returns the number of the unit assigned or a negative errno code. 1183 */ 1184 1185 int dev_alloc_name(struct net_device *dev, const char *name) 1186 { 1187 return dev_alloc_name_ns(dev_net(dev), dev, name); 1188 } 1189 EXPORT_SYMBOL(dev_alloc_name); 1190 1191 static int dev_get_valid_name(struct net *net, struct net_device *dev, 1192 const char *name) 1193 { 1194 char buf[IFNAMSIZ]; 1195 int ret; 1196 1197 ret = dev_prep_valid_name(net, dev, name, buf); 1198 if (ret >= 0) 1199 strscpy(dev->name, buf, IFNAMSIZ); 1200 return ret; 1201 } 1202 1203 /** 1204 * dev_change_name - change name of a device 1205 * @dev: device 1206 * @newname: name (or format string) must be at least IFNAMSIZ 1207 * 1208 * Change name of a device, can pass format strings "eth%d". 1209 * for wildcarding. 1210 */ 1211 int dev_change_name(struct net_device *dev, const char *newname) 1212 { 1213 unsigned char old_assign_type; 1214 char oldname[IFNAMSIZ]; 1215 int err = 0; 1216 int ret; 1217 struct net *net; 1218 1219 ASSERT_RTNL(); 1220 BUG_ON(!dev_net(dev)); 1221 1222 net = dev_net(dev); 1223 1224 down_write(&devnet_rename_sem); 1225 1226 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { 1227 up_write(&devnet_rename_sem); 1228 return 0; 1229 } 1230 1231 memcpy(oldname, dev->name, IFNAMSIZ); 1232 1233 err = dev_get_valid_name(net, dev, newname); 1234 if (err < 0) { 1235 up_write(&devnet_rename_sem); 1236 return err; 1237 } 1238 1239 if (oldname[0] && !strchr(oldname, '%')) 1240 netdev_info(dev, "renamed from %s%s\n", oldname, 1241 dev->flags & IFF_UP ? " (while UP)" : ""); 1242 1243 old_assign_type = dev->name_assign_type; 1244 dev->name_assign_type = NET_NAME_RENAMED; 1245 1246 rollback: 1247 ret = device_rename(&dev->dev, dev->name); 1248 if (ret) { 1249 memcpy(dev->name, oldname, IFNAMSIZ); 1250 dev->name_assign_type = old_assign_type; 1251 up_write(&devnet_rename_sem); 1252 return ret; 1253 } 1254 1255 up_write(&devnet_rename_sem); 1256 1257 netdev_adjacent_rename_links(dev, oldname); 1258 1259 write_lock(&dev_base_lock); 1260 netdev_name_node_del(dev->name_node); 1261 write_unlock(&dev_base_lock); 1262 1263 synchronize_rcu(); 1264 1265 write_lock(&dev_base_lock); 1266 netdev_name_node_add(net, dev->name_node); 1267 write_unlock(&dev_base_lock); 1268 1269 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); 1270 ret = notifier_to_errno(ret); 1271 1272 if (ret) { 1273 /* err >= 0 after dev_alloc_name() or stores the first errno */ 1274 if (err >= 0) { 1275 err = ret; 1276 down_write(&devnet_rename_sem); 1277 memcpy(dev->name, oldname, IFNAMSIZ); 1278 memcpy(oldname, newname, IFNAMSIZ); 1279 dev->name_assign_type = old_assign_type; 1280 old_assign_type = NET_NAME_RENAMED; 1281 goto rollback; 1282 } else { 1283 netdev_err(dev, "name change rollback failed: %d\n", 1284 ret); 1285 } 1286 } 1287 1288 return err; 1289 } 1290 1291 /** 1292 * dev_set_alias - change ifalias of a device 1293 * @dev: device 1294 * @alias: name up to IFALIASZ 1295 * @len: limit of bytes to copy from info 1296 * 1297 * Set ifalias for a device, 1298 */ 1299 int dev_set_alias(struct net_device *dev, const char *alias, size_t len) 1300 { 1301 struct dev_ifalias *new_alias = NULL; 1302 1303 if (len >= IFALIASZ) 1304 return -EINVAL; 1305 1306 if (len) { 1307 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL); 1308 if (!new_alias) 1309 return -ENOMEM; 1310 1311 memcpy(new_alias->ifalias, alias, len); 1312 new_alias->ifalias[len] = 0; 1313 } 1314 1315 mutex_lock(&ifalias_mutex); 1316 new_alias = rcu_replace_pointer(dev->ifalias, new_alias, 1317 mutex_is_locked(&ifalias_mutex)); 1318 mutex_unlock(&ifalias_mutex); 1319 1320 if (new_alias) 1321 kfree_rcu(new_alias, rcuhead); 1322 1323 return len; 1324 } 1325 EXPORT_SYMBOL(dev_set_alias); 1326 1327 /** 1328 * dev_get_alias - get ifalias of a device 1329 * @dev: device 1330 * @name: buffer to store name of ifalias 1331 * @len: size of buffer 1332 * 1333 * get ifalias for a device. Caller must make sure dev cannot go 1334 * away, e.g. rcu read lock or own a reference count to device. 1335 */ 1336 int dev_get_alias(const struct net_device *dev, char *name, size_t len) 1337 { 1338 const struct dev_ifalias *alias; 1339 int ret = 0; 1340 1341 rcu_read_lock(); 1342 alias = rcu_dereference(dev->ifalias); 1343 if (alias) 1344 ret = snprintf(name, len, "%s", alias->ifalias); 1345 rcu_read_unlock(); 1346 1347 return ret; 1348 } 1349 1350 /** 1351 * netdev_features_change - device changes features 1352 * @dev: device to cause notification 1353 * 1354 * Called to indicate a device has changed features. 1355 */ 1356 void netdev_features_change(struct net_device *dev) 1357 { 1358 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); 1359 } 1360 EXPORT_SYMBOL(netdev_features_change); 1361 1362 /** 1363 * netdev_state_change - device changes state 1364 * @dev: device to cause notification 1365 * 1366 * Called to indicate a device has changed state. This function calls 1367 * the notifier chains for netdev_chain and sends a NEWLINK message 1368 * to the routing socket. 1369 */ 1370 void netdev_state_change(struct net_device *dev) 1371 { 1372 if (dev->flags & IFF_UP) { 1373 struct netdev_notifier_change_info change_info = { 1374 .info.dev = dev, 1375 }; 1376 1377 call_netdevice_notifiers_info(NETDEV_CHANGE, 1378 &change_info.info); 1379 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL, 0, NULL); 1380 } 1381 } 1382 EXPORT_SYMBOL(netdev_state_change); 1383 1384 /** 1385 * __netdev_notify_peers - notify network peers about existence of @dev, 1386 * to be called when rtnl lock is already held. 1387 * @dev: network device 1388 * 1389 * Generate traffic such that interested network peers are aware of 1390 * @dev, such as by generating a gratuitous ARP. This may be used when 1391 * a device wants to inform the rest of the network about some sort of 1392 * reconfiguration such as a failover event or virtual machine 1393 * migration. 1394 */ 1395 void __netdev_notify_peers(struct net_device *dev) 1396 { 1397 ASSERT_RTNL(); 1398 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); 1399 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev); 1400 } 1401 EXPORT_SYMBOL(__netdev_notify_peers); 1402 1403 /** 1404 * netdev_notify_peers - notify network peers about existence of @dev 1405 * @dev: network device 1406 * 1407 * Generate traffic such that interested network peers are aware of 1408 * @dev, such as by generating a gratuitous ARP. This may be used when 1409 * a device wants to inform the rest of the network about some sort of 1410 * reconfiguration such as a failover event or virtual machine 1411 * migration. 1412 */ 1413 void netdev_notify_peers(struct net_device *dev) 1414 { 1415 rtnl_lock(); 1416 __netdev_notify_peers(dev); 1417 rtnl_unlock(); 1418 } 1419 EXPORT_SYMBOL(netdev_notify_peers); 1420 1421 static int napi_threaded_poll(void *data); 1422 1423 static int napi_kthread_create(struct napi_struct *n) 1424 { 1425 int err = 0; 1426 1427 /* Create and wake up the kthread once to put it in 1428 * TASK_INTERRUPTIBLE mode to avoid the blocked task 1429 * warning and work with loadavg. 1430 */ 1431 n->thread = kthread_run(napi_threaded_poll, n, "napi/%s-%d", 1432 n->dev->name, n->napi_id); 1433 if (IS_ERR(n->thread)) { 1434 err = PTR_ERR(n->thread); 1435 pr_err("kthread_run failed with err %d\n", err); 1436 n->thread = NULL; 1437 } 1438 1439 return err; 1440 } 1441 1442 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1443 { 1444 const struct net_device_ops *ops = dev->netdev_ops; 1445 int ret; 1446 1447 ASSERT_RTNL(); 1448 dev_addr_check(dev); 1449 1450 if (!netif_device_present(dev)) { 1451 /* may be detached because parent is runtime-suspended */ 1452 if (dev->dev.parent) 1453 pm_runtime_resume(dev->dev.parent); 1454 if (!netif_device_present(dev)) 1455 return -ENODEV; 1456 } 1457 1458 /* Block netpoll from trying to do any rx path servicing. 1459 * If we don't do this there is a chance ndo_poll_controller 1460 * or ndo_poll may be running while we open the device 1461 */ 1462 netpoll_poll_disable(dev); 1463 1464 ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack); 1465 ret = notifier_to_errno(ret); 1466 if (ret) 1467 return ret; 1468 1469 set_bit(__LINK_STATE_START, &dev->state); 1470 1471 if (ops->ndo_validate_addr) 1472 ret = ops->ndo_validate_addr(dev); 1473 1474 if (!ret && ops->ndo_open) 1475 ret = ops->ndo_open(dev); 1476 1477 netpoll_poll_enable(dev); 1478 1479 if (ret) 1480 clear_bit(__LINK_STATE_START, &dev->state); 1481 else { 1482 dev->flags |= IFF_UP; 1483 dev_set_rx_mode(dev); 1484 dev_activate(dev); 1485 add_device_randomness(dev->dev_addr, dev->addr_len); 1486 } 1487 1488 return ret; 1489 } 1490 1491 /** 1492 * dev_open - prepare an interface for use. 1493 * @dev: device to open 1494 * @extack: netlink extended ack 1495 * 1496 * Takes a device from down to up state. The device's private open 1497 * function is invoked and then the multicast lists are loaded. Finally 1498 * the device is moved into the up state and a %NETDEV_UP message is 1499 * sent to the netdev notifier chain. 1500 * 1501 * Calling this function on an active interface is a nop. On a failure 1502 * a negative errno code is returned. 1503 */ 1504 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1505 { 1506 int ret; 1507 1508 if (dev->flags & IFF_UP) 1509 return 0; 1510 1511 ret = __dev_open(dev, extack); 1512 if (ret < 0) 1513 return ret; 1514 1515 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL); 1516 call_netdevice_notifiers(NETDEV_UP, dev); 1517 1518 return ret; 1519 } 1520 EXPORT_SYMBOL(dev_open); 1521 1522 static void __dev_close_many(struct list_head *head) 1523 { 1524 struct net_device *dev; 1525 1526 ASSERT_RTNL(); 1527 might_sleep(); 1528 1529 list_for_each_entry(dev, head, close_list) { 1530 /* Temporarily disable netpoll until the interface is down */ 1531 netpoll_poll_disable(dev); 1532 1533 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); 1534 1535 clear_bit(__LINK_STATE_START, &dev->state); 1536 1537 /* Synchronize to scheduled poll. We cannot touch poll list, it 1538 * can be even on different cpu. So just clear netif_running(). 1539 * 1540 * dev->stop() will invoke napi_disable() on all of it's 1541 * napi_struct instances on this device. 1542 */ 1543 smp_mb__after_atomic(); /* Commit netif_running(). */ 1544 } 1545 1546 dev_deactivate_many(head); 1547 1548 list_for_each_entry(dev, head, close_list) { 1549 const struct net_device_ops *ops = dev->netdev_ops; 1550 1551 /* 1552 * Call the device specific close. This cannot fail. 1553 * Only if device is UP 1554 * 1555 * We allow it to be called even after a DETACH hot-plug 1556 * event. 1557 */ 1558 if (ops->ndo_stop) 1559 ops->ndo_stop(dev); 1560 1561 dev->flags &= ~IFF_UP; 1562 netpoll_poll_enable(dev); 1563 } 1564 } 1565 1566 static void __dev_close(struct net_device *dev) 1567 { 1568 LIST_HEAD(single); 1569 1570 list_add(&dev->close_list, &single); 1571 __dev_close_many(&single); 1572 list_del(&single); 1573 } 1574 1575 void dev_close_many(struct list_head *head, bool unlink) 1576 { 1577 struct net_device *dev, *tmp; 1578 1579 /* Remove the devices that don't need to be closed */ 1580 list_for_each_entry_safe(dev, tmp, head, close_list) 1581 if (!(dev->flags & IFF_UP)) 1582 list_del_init(&dev->close_list); 1583 1584 __dev_close_many(head); 1585 1586 list_for_each_entry_safe(dev, tmp, head, close_list) { 1587 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL); 1588 call_netdevice_notifiers(NETDEV_DOWN, dev); 1589 if (unlink) 1590 list_del_init(&dev->close_list); 1591 } 1592 } 1593 EXPORT_SYMBOL(dev_close_many); 1594 1595 /** 1596 * dev_close - shutdown an interface. 1597 * @dev: device to shutdown 1598 * 1599 * This function moves an active device into down state. A 1600 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device 1601 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier 1602 * chain. 1603 */ 1604 void dev_close(struct net_device *dev) 1605 { 1606 if (dev->flags & IFF_UP) { 1607 LIST_HEAD(single); 1608 1609 list_add(&dev->close_list, &single); 1610 dev_close_many(&single, true); 1611 list_del(&single); 1612 } 1613 } 1614 EXPORT_SYMBOL(dev_close); 1615 1616 1617 /** 1618 * dev_disable_lro - disable Large Receive Offload on a device 1619 * @dev: device 1620 * 1621 * Disable Large Receive Offload (LRO) on a net device. Must be 1622 * called under RTNL. This is needed if received packets may be 1623 * forwarded to another interface. 1624 */ 1625 void dev_disable_lro(struct net_device *dev) 1626 { 1627 struct net_device *lower_dev; 1628 struct list_head *iter; 1629 1630 dev->wanted_features &= ~NETIF_F_LRO; 1631 netdev_update_features(dev); 1632 1633 if (unlikely(dev->features & NETIF_F_LRO)) 1634 netdev_WARN(dev, "failed to disable LRO!\n"); 1635 1636 netdev_for_each_lower_dev(dev, lower_dev, iter) 1637 dev_disable_lro(lower_dev); 1638 } 1639 EXPORT_SYMBOL(dev_disable_lro); 1640 1641 /** 1642 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device 1643 * @dev: device 1644 * 1645 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be 1646 * called under RTNL. This is needed if Generic XDP is installed on 1647 * the device. 1648 */ 1649 static void dev_disable_gro_hw(struct net_device *dev) 1650 { 1651 dev->wanted_features &= ~NETIF_F_GRO_HW; 1652 netdev_update_features(dev); 1653 1654 if (unlikely(dev->features & NETIF_F_GRO_HW)) 1655 netdev_WARN(dev, "failed to disable GRO_HW!\n"); 1656 } 1657 1658 const char *netdev_cmd_to_name(enum netdev_cmd cmd) 1659 { 1660 #define N(val) \ 1661 case NETDEV_##val: \ 1662 return "NETDEV_" __stringify(val); 1663 switch (cmd) { 1664 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER) 1665 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE) 1666 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE) 1667 N(POST_INIT) N(PRE_UNINIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) 1668 N(CHANGEUPPER) N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) 1669 N(BONDING_INFO) N(PRECHANGEUPPER) N(CHANGELOWERSTATE) 1670 N(UDP_TUNNEL_PUSH_INFO) N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN) 1671 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO) 1672 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO) 1673 N(PRE_CHANGEADDR) N(OFFLOAD_XSTATS_ENABLE) N(OFFLOAD_XSTATS_DISABLE) 1674 N(OFFLOAD_XSTATS_REPORT_USED) N(OFFLOAD_XSTATS_REPORT_DELTA) 1675 N(XDP_FEAT_CHANGE) 1676 } 1677 #undef N 1678 return "UNKNOWN_NETDEV_EVENT"; 1679 } 1680 EXPORT_SYMBOL_GPL(netdev_cmd_to_name); 1681 1682 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, 1683 struct net_device *dev) 1684 { 1685 struct netdev_notifier_info info = { 1686 .dev = dev, 1687 }; 1688 1689 return nb->notifier_call(nb, val, &info); 1690 } 1691 1692 static int call_netdevice_register_notifiers(struct notifier_block *nb, 1693 struct net_device *dev) 1694 { 1695 int err; 1696 1697 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); 1698 err = notifier_to_errno(err); 1699 if (err) 1700 return err; 1701 1702 if (!(dev->flags & IFF_UP)) 1703 return 0; 1704 1705 call_netdevice_notifier(nb, NETDEV_UP, dev); 1706 return 0; 1707 } 1708 1709 static void call_netdevice_unregister_notifiers(struct notifier_block *nb, 1710 struct net_device *dev) 1711 { 1712 if (dev->flags & IFF_UP) { 1713 call_netdevice_notifier(nb, NETDEV_GOING_DOWN, 1714 dev); 1715 call_netdevice_notifier(nb, NETDEV_DOWN, dev); 1716 } 1717 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); 1718 } 1719 1720 static int call_netdevice_register_net_notifiers(struct notifier_block *nb, 1721 struct net *net) 1722 { 1723 struct net_device *dev; 1724 int err; 1725 1726 for_each_netdev(net, dev) { 1727 err = call_netdevice_register_notifiers(nb, dev); 1728 if (err) 1729 goto rollback; 1730 } 1731 return 0; 1732 1733 rollback: 1734 for_each_netdev_continue_reverse(net, dev) 1735 call_netdevice_unregister_notifiers(nb, dev); 1736 return err; 1737 } 1738 1739 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb, 1740 struct net *net) 1741 { 1742 struct net_device *dev; 1743 1744 for_each_netdev(net, dev) 1745 call_netdevice_unregister_notifiers(nb, dev); 1746 } 1747 1748 static int dev_boot_phase = 1; 1749 1750 /** 1751 * register_netdevice_notifier - register a network notifier block 1752 * @nb: notifier 1753 * 1754 * Register a notifier to be called when network device events occur. 1755 * The notifier passed is linked into the kernel structures and must 1756 * not be reused until it has been unregistered. A negative errno code 1757 * is returned on a failure. 1758 * 1759 * When registered all registration and up events are replayed 1760 * to the new notifier to allow device to have a race free 1761 * view of the network device list. 1762 */ 1763 1764 int register_netdevice_notifier(struct notifier_block *nb) 1765 { 1766 struct net *net; 1767 int err; 1768 1769 /* Close race with setup_net() and cleanup_net() */ 1770 down_write(&pernet_ops_rwsem); 1771 rtnl_lock(); 1772 err = raw_notifier_chain_register(&netdev_chain, nb); 1773 if (err) 1774 goto unlock; 1775 if (dev_boot_phase) 1776 goto unlock; 1777 for_each_net(net) { 1778 err = call_netdevice_register_net_notifiers(nb, net); 1779 if (err) 1780 goto rollback; 1781 } 1782 1783 unlock: 1784 rtnl_unlock(); 1785 up_write(&pernet_ops_rwsem); 1786 return err; 1787 1788 rollback: 1789 for_each_net_continue_reverse(net) 1790 call_netdevice_unregister_net_notifiers(nb, net); 1791 1792 raw_notifier_chain_unregister(&netdev_chain, nb); 1793 goto unlock; 1794 } 1795 EXPORT_SYMBOL(register_netdevice_notifier); 1796 1797 /** 1798 * unregister_netdevice_notifier - unregister a network notifier block 1799 * @nb: notifier 1800 * 1801 * Unregister a notifier previously registered by 1802 * register_netdevice_notifier(). The notifier is unlinked into the 1803 * kernel structures and may then be reused. A negative errno code 1804 * is returned on a failure. 1805 * 1806 * After unregistering unregister and down device events are synthesized 1807 * for all devices on the device list to the removed notifier to remove 1808 * the need for special case cleanup code. 1809 */ 1810 1811 int unregister_netdevice_notifier(struct notifier_block *nb) 1812 { 1813 struct net *net; 1814 int err; 1815 1816 /* Close race with setup_net() and cleanup_net() */ 1817 down_write(&pernet_ops_rwsem); 1818 rtnl_lock(); 1819 err = raw_notifier_chain_unregister(&netdev_chain, nb); 1820 if (err) 1821 goto unlock; 1822 1823 for_each_net(net) 1824 call_netdevice_unregister_net_notifiers(nb, net); 1825 1826 unlock: 1827 rtnl_unlock(); 1828 up_write(&pernet_ops_rwsem); 1829 return err; 1830 } 1831 EXPORT_SYMBOL(unregister_netdevice_notifier); 1832 1833 static int __register_netdevice_notifier_net(struct net *net, 1834 struct notifier_block *nb, 1835 bool ignore_call_fail) 1836 { 1837 int err; 1838 1839 err = raw_notifier_chain_register(&net->netdev_chain, nb); 1840 if (err) 1841 return err; 1842 if (dev_boot_phase) 1843 return 0; 1844 1845 err = call_netdevice_register_net_notifiers(nb, net); 1846 if (err && !ignore_call_fail) 1847 goto chain_unregister; 1848 1849 return 0; 1850 1851 chain_unregister: 1852 raw_notifier_chain_unregister(&net->netdev_chain, nb); 1853 return err; 1854 } 1855 1856 static int __unregister_netdevice_notifier_net(struct net *net, 1857 struct notifier_block *nb) 1858 { 1859 int err; 1860 1861 err = raw_notifier_chain_unregister(&net->netdev_chain, nb); 1862 if (err) 1863 return err; 1864 1865 call_netdevice_unregister_net_notifiers(nb, net); 1866 return 0; 1867 } 1868 1869 /** 1870 * register_netdevice_notifier_net - register a per-netns network notifier block 1871 * @net: network namespace 1872 * @nb: notifier 1873 * 1874 * Register a notifier to be called when network device events occur. 1875 * The notifier passed is linked into the kernel structures and must 1876 * not be reused until it has been unregistered. A negative errno code 1877 * is returned on a failure. 1878 * 1879 * When registered all registration and up events are replayed 1880 * to the new notifier to allow device to have a race free 1881 * view of the network device list. 1882 */ 1883 1884 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb) 1885 { 1886 int err; 1887 1888 rtnl_lock(); 1889 err = __register_netdevice_notifier_net(net, nb, false); 1890 rtnl_unlock(); 1891 return err; 1892 } 1893 EXPORT_SYMBOL(register_netdevice_notifier_net); 1894 1895 /** 1896 * unregister_netdevice_notifier_net - unregister a per-netns 1897 * network notifier block 1898 * @net: network namespace 1899 * @nb: notifier 1900 * 1901 * Unregister a notifier previously registered by 1902 * register_netdevice_notifier_net(). The notifier is unlinked from the 1903 * kernel structures and may then be reused. A negative errno code 1904 * is returned on a failure. 1905 * 1906 * After unregistering unregister and down device events are synthesized 1907 * for all devices on the device list to the removed notifier to remove 1908 * the need for special case cleanup code. 1909 */ 1910 1911 int unregister_netdevice_notifier_net(struct net *net, 1912 struct notifier_block *nb) 1913 { 1914 int err; 1915 1916 rtnl_lock(); 1917 err = __unregister_netdevice_notifier_net(net, nb); 1918 rtnl_unlock(); 1919 return err; 1920 } 1921 EXPORT_SYMBOL(unregister_netdevice_notifier_net); 1922 1923 static void __move_netdevice_notifier_net(struct net *src_net, 1924 struct net *dst_net, 1925 struct notifier_block *nb) 1926 { 1927 __unregister_netdevice_notifier_net(src_net, nb); 1928 __register_netdevice_notifier_net(dst_net, nb, true); 1929 } 1930 1931 int register_netdevice_notifier_dev_net(struct net_device *dev, 1932 struct notifier_block *nb, 1933 struct netdev_net_notifier *nn) 1934 { 1935 int err; 1936 1937 rtnl_lock(); 1938 err = __register_netdevice_notifier_net(dev_net(dev), nb, false); 1939 if (!err) { 1940 nn->nb = nb; 1941 list_add(&nn->list, &dev->net_notifier_list); 1942 } 1943 rtnl_unlock(); 1944 return err; 1945 } 1946 EXPORT_SYMBOL(register_netdevice_notifier_dev_net); 1947 1948 int unregister_netdevice_notifier_dev_net(struct net_device *dev, 1949 struct notifier_block *nb, 1950 struct netdev_net_notifier *nn) 1951 { 1952 int err; 1953 1954 rtnl_lock(); 1955 list_del(&nn->list); 1956 err = __unregister_netdevice_notifier_net(dev_net(dev), nb); 1957 rtnl_unlock(); 1958 return err; 1959 } 1960 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net); 1961 1962 static void move_netdevice_notifiers_dev_net(struct net_device *dev, 1963 struct net *net) 1964 { 1965 struct netdev_net_notifier *nn; 1966 1967 list_for_each_entry(nn, &dev->net_notifier_list, list) 1968 __move_netdevice_notifier_net(dev_net(dev), net, nn->nb); 1969 } 1970 1971 /** 1972 * call_netdevice_notifiers_info - call all network notifier blocks 1973 * @val: value passed unmodified to notifier function 1974 * @info: notifier information data 1975 * 1976 * Call all network notifier blocks. Parameters and return value 1977 * are as for raw_notifier_call_chain(). 1978 */ 1979 1980 int call_netdevice_notifiers_info(unsigned long val, 1981 struct netdev_notifier_info *info) 1982 { 1983 struct net *net = dev_net(info->dev); 1984 int ret; 1985 1986 ASSERT_RTNL(); 1987 1988 /* Run per-netns notifier block chain first, then run the global one. 1989 * Hopefully, one day, the global one is going to be removed after 1990 * all notifier block registrators get converted to be per-netns. 1991 */ 1992 ret = raw_notifier_call_chain(&net->netdev_chain, val, info); 1993 if (ret & NOTIFY_STOP_MASK) 1994 return ret; 1995 return raw_notifier_call_chain(&netdev_chain, val, info); 1996 } 1997 1998 /** 1999 * call_netdevice_notifiers_info_robust - call per-netns notifier blocks 2000 * for and rollback on error 2001 * @val_up: value passed unmodified to notifier function 2002 * @val_down: value passed unmodified to the notifier function when 2003 * recovering from an error on @val_up 2004 * @info: notifier information data 2005 * 2006 * Call all per-netns network notifier blocks, but not notifier blocks on 2007 * the global notifier chain. Parameters and return value are as for 2008 * raw_notifier_call_chain_robust(). 2009 */ 2010 2011 static int 2012 call_netdevice_notifiers_info_robust(unsigned long val_up, 2013 unsigned long val_down, 2014 struct netdev_notifier_info *info) 2015 { 2016 struct net *net = dev_net(info->dev); 2017 2018 ASSERT_RTNL(); 2019 2020 return raw_notifier_call_chain_robust(&net->netdev_chain, 2021 val_up, val_down, info); 2022 } 2023 2024 static int call_netdevice_notifiers_extack(unsigned long val, 2025 struct net_device *dev, 2026 struct netlink_ext_ack *extack) 2027 { 2028 struct netdev_notifier_info info = { 2029 .dev = dev, 2030 .extack = extack, 2031 }; 2032 2033 return call_netdevice_notifiers_info(val, &info); 2034 } 2035 2036 /** 2037 * call_netdevice_notifiers - call all network notifier blocks 2038 * @val: value passed unmodified to notifier function 2039 * @dev: net_device pointer passed unmodified to notifier function 2040 * 2041 * Call all network notifier blocks. Parameters and return value 2042 * are as for raw_notifier_call_chain(). 2043 */ 2044 2045 int call_netdevice_notifiers(unsigned long val, struct net_device *dev) 2046 { 2047 return call_netdevice_notifiers_extack(val, dev, NULL); 2048 } 2049 EXPORT_SYMBOL(call_netdevice_notifiers); 2050 2051 /** 2052 * call_netdevice_notifiers_mtu - call all network notifier blocks 2053 * @val: value passed unmodified to notifier function 2054 * @dev: net_device pointer passed unmodified to notifier function 2055 * @arg: additional u32 argument passed to the notifier function 2056 * 2057 * Call all network notifier blocks. Parameters and return value 2058 * are as for raw_notifier_call_chain(). 2059 */ 2060 static int call_netdevice_notifiers_mtu(unsigned long val, 2061 struct net_device *dev, u32 arg) 2062 { 2063 struct netdev_notifier_info_ext info = { 2064 .info.dev = dev, 2065 .ext.mtu = arg, 2066 }; 2067 2068 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0); 2069 2070 return call_netdevice_notifiers_info(val, &info.info); 2071 } 2072 2073 #ifdef CONFIG_NET_INGRESS 2074 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key); 2075 2076 void net_inc_ingress_queue(void) 2077 { 2078 static_branch_inc(&ingress_needed_key); 2079 } 2080 EXPORT_SYMBOL_GPL(net_inc_ingress_queue); 2081 2082 void net_dec_ingress_queue(void) 2083 { 2084 static_branch_dec(&ingress_needed_key); 2085 } 2086 EXPORT_SYMBOL_GPL(net_dec_ingress_queue); 2087 #endif 2088 2089 #ifdef CONFIG_NET_EGRESS 2090 static DEFINE_STATIC_KEY_FALSE(egress_needed_key); 2091 2092 void net_inc_egress_queue(void) 2093 { 2094 static_branch_inc(&egress_needed_key); 2095 } 2096 EXPORT_SYMBOL_GPL(net_inc_egress_queue); 2097 2098 void net_dec_egress_queue(void) 2099 { 2100 static_branch_dec(&egress_needed_key); 2101 } 2102 EXPORT_SYMBOL_GPL(net_dec_egress_queue); 2103 #endif 2104 2105 DEFINE_STATIC_KEY_FALSE(netstamp_needed_key); 2106 EXPORT_SYMBOL(netstamp_needed_key); 2107 #ifdef CONFIG_JUMP_LABEL 2108 static atomic_t netstamp_needed_deferred; 2109 static atomic_t netstamp_wanted; 2110 static void netstamp_clear(struct work_struct *work) 2111 { 2112 int deferred = atomic_xchg(&netstamp_needed_deferred, 0); 2113 int wanted; 2114 2115 wanted = atomic_add_return(deferred, &netstamp_wanted); 2116 if (wanted > 0) 2117 static_branch_enable(&netstamp_needed_key); 2118 else 2119 static_branch_disable(&netstamp_needed_key); 2120 } 2121 static DECLARE_WORK(netstamp_work, netstamp_clear); 2122 #endif 2123 2124 void net_enable_timestamp(void) 2125 { 2126 #ifdef CONFIG_JUMP_LABEL 2127 int wanted = atomic_read(&netstamp_wanted); 2128 2129 while (wanted > 0) { 2130 if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted + 1)) 2131 return; 2132 } 2133 atomic_inc(&netstamp_needed_deferred); 2134 schedule_work(&netstamp_work); 2135 #else 2136 static_branch_inc(&netstamp_needed_key); 2137 #endif 2138 } 2139 EXPORT_SYMBOL(net_enable_timestamp); 2140 2141 void net_disable_timestamp(void) 2142 { 2143 #ifdef CONFIG_JUMP_LABEL 2144 int wanted = atomic_read(&netstamp_wanted); 2145 2146 while (wanted > 1) { 2147 if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted - 1)) 2148 return; 2149 } 2150 atomic_dec(&netstamp_needed_deferred); 2151 schedule_work(&netstamp_work); 2152 #else 2153 static_branch_dec(&netstamp_needed_key); 2154 #endif 2155 } 2156 EXPORT_SYMBOL(net_disable_timestamp); 2157 2158 static inline void net_timestamp_set(struct sk_buff *skb) 2159 { 2160 skb->tstamp = 0; 2161 skb->mono_delivery_time = 0; 2162 if (static_branch_unlikely(&netstamp_needed_key)) 2163 skb->tstamp = ktime_get_real(); 2164 } 2165 2166 #define net_timestamp_check(COND, SKB) \ 2167 if (static_branch_unlikely(&netstamp_needed_key)) { \ 2168 if ((COND) && !(SKB)->tstamp) \ 2169 (SKB)->tstamp = ktime_get_real(); \ 2170 } \ 2171 2172 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb) 2173 { 2174 return __is_skb_forwardable(dev, skb, true); 2175 } 2176 EXPORT_SYMBOL_GPL(is_skb_forwardable); 2177 2178 static int __dev_forward_skb2(struct net_device *dev, struct sk_buff *skb, 2179 bool check_mtu) 2180 { 2181 int ret = ____dev_forward_skb(dev, skb, check_mtu); 2182 2183 if (likely(!ret)) { 2184 skb->protocol = eth_type_trans(skb, dev); 2185 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); 2186 } 2187 2188 return ret; 2189 } 2190 2191 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2192 { 2193 return __dev_forward_skb2(dev, skb, true); 2194 } 2195 EXPORT_SYMBOL_GPL(__dev_forward_skb); 2196 2197 /** 2198 * dev_forward_skb - loopback an skb to another netif 2199 * 2200 * @dev: destination network device 2201 * @skb: buffer to forward 2202 * 2203 * return values: 2204 * NET_RX_SUCCESS (no congestion) 2205 * NET_RX_DROP (packet was dropped, but freed) 2206 * 2207 * dev_forward_skb can be used for injecting an skb from the 2208 * start_xmit function of one device into the receive queue 2209 * of another device. 2210 * 2211 * The receiving device may be in another namespace, so 2212 * we have to clear all information in the skb that could 2213 * impact namespace isolation. 2214 */ 2215 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2216 { 2217 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); 2218 } 2219 EXPORT_SYMBOL_GPL(dev_forward_skb); 2220 2221 int dev_forward_skb_nomtu(struct net_device *dev, struct sk_buff *skb) 2222 { 2223 return __dev_forward_skb2(dev, skb, false) ?: netif_rx_internal(skb); 2224 } 2225 2226 static inline int deliver_skb(struct sk_buff *skb, 2227 struct packet_type *pt_prev, 2228 struct net_device *orig_dev) 2229 { 2230 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 2231 return -ENOMEM; 2232 refcount_inc(&skb->users); 2233 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 2234 } 2235 2236 static inline void deliver_ptype_list_skb(struct sk_buff *skb, 2237 struct packet_type **pt, 2238 struct net_device *orig_dev, 2239 __be16 type, 2240 struct list_head *ptype_list) 2241 { 2242 struct packet_type *ptype, *pt_prev = *pt; 2243 2244 list_for_each_entry_rcu(ptype, ptype_list, list) { 2245 if (ptype->type != type) 2246 continue; 2247 if (pt_prev) 2248 deliver_skb(skb, pt_prev, orig_dev); 2249 pt_prev = ptype; 2250 } 2251 *pt = pt_prev; 2252 } 2253 2254 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) 2255 { 2256 if (!ptype->af_packet_priv || !skb->sk) 2257 return false; 2258 2259 if (ptype->id_match) 2260 return ptype->id_match(ptype, skb->sk); 2261 else if ((struct sock *)ptype->af_packet_priv == skb->sk) 2262 return true; 2263 2264 return false; 2265 } 2266 2267 /** 2268 * dev_nit_active - return true if any network interface taps are in use 2269 * 2270 * @dev: network device to check for the presence of taps 2271 */ 2272 bool dev_nit_active(struct net_device *dev) 2273 { 2274 return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all); 2275 } 2276 EXPORT_SYMBOL_GPL(dev_nit_active); 2277 2278 /* 2279 * Support routine. Sends outgoing frames to any network 2280 * taps currently in use. 2281 */ 2282 2283 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) 2284 { 2285 struct packet_type *ptype; 2286 struct sk_buff *skb2 = NULL; 2287 struct packet_type *pt_prev = NULL; 2288 struct list_head *ptype_list = &ptype_all; 2289 2290 rcu_read_lock(); 2291 again: 2292 list_for_each_entry_rcu(ptype, ptype_list, list) { 2293 if (ptype->ignore_outgoing) 2294 continue; 2295 2296 /* Never send packets back to the socket 2297 * they originated from - MvS (miquels@drinkel.ow.org) 2298 */ 2299 if (skb_loop_sk(ptype, skb)) 2300 continue; 2301 2302 if (pt_prev) { 2303 deliver_skb(skb2, pt_prev, skb->dev); 2304 pt_prev = ptype; 2305 continue; 2306 } 2307 2308 /* need to clone skb, done only once */ 2309 skb2 = skb_clone(skb, GFP_ATOMIC); 2310 if (!skb2) 2311 goto out_unlock; 2312 2313 net_timestamp_set(skb2); 2314 2315 /* skb->nh should be correctly 2316 * set by sender, so that the second statement is 2317 * just protection against buggy protocols. 2318 */ 2319 skb_reset_mac_header(skb2); 2320 2321 if (skb_network_header(skb2) < skb2->data || 2322 skb_network_header(skb2) > skb_tail_pointer(skb2)) { 2323 net_crit_ratelimited("protocol %04x is buggy, dev %s\n", 2324 ntohs(skb2->protocol), 2325 dev->name); 2326 skb_reset_network_header(skb2); 2327 } 2328 2329 skb2->transport_header = skb2->network_header; 2330 skb2->pkt_type = PACKET_OUTGOING; 2331 pt_prev = ptype; 2332 } 2333 2334 if (ptype_list == &ptype_all) { 2335 ptype_list = &dev->ptype_all; 2336 goto again; 2337 } 2338 out_unlock: 2339 if (pt_prev) { 2340 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC)) 2341 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); 2342 else 2343 kfree_skb(skb2); 2344 } 2345 rcu_read_unlock(); 2346 } 2347 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit); 2348 2349 /** 2350 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change 2351 * @dev: Network device 2352 * @txq: number of queues available 2353 * 2354 * If real_num_tx_queues is changed the tc mappings may no longer be 2355 * valid. To resolve this verify the tc mapping remains valid and if 2356 * not NULL the mapping. With no priorities mapping to this 2357 * offset/count pair it will no longer be used. In the worst case TC0 2358 * is invalid nothing can be done so disable priority mappings. If is 2359 * expected that drivers will fix this mapping if they can before 2360 * calling netif_set_real_num_tx_queues. 2361 */ 2362 static void netif_setup_tc(struct net_device *dev, unsigned int txq) 2363 { 2364 int i; 2365 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2366 2367 /* If TC0 is invalidated disable TC mapping */ 2368 if (tc->offset + tc->count > txq) { 2369 netdev_warn(dev, "Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); 2370 dev->num_tc = 0; 2371 return; 2372 } 2373 2374 /* Invalidated prio to tc mappings set to TC0 */ 2375 for (i = 1; i < TC_BITMASK + 1; i++) { 2376 int q = netdev_get_prio_tc_map(dev, i); 2377 2378 tc = &dev->tc_to_txq[q]; 2379 if (tc->offset + tc->count > txq) { 2380 netdev_warn(dev, "Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n", 2381 i, q); 2382 netdev_set_prio_tc_map(dev, i, 0); 2383 } 2384 } 2385 } 2386 2387 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq) 2388 { 2389 if (dev->num_tc) { 2390 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2391 int i; 2392 2393 /* walk through the TCs and see if it falls into any of them */ 2394 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) { 2395 if ((txq - tc->offset) < tc->count) 2396 return i; 2397 } 2398 2399 /* didn't find it, just return -1 to indicate no match */ 2400 return -1; 2401 } 2402 2403 return 0; 2404 } 2405 EXPORT_SYMBOL(netdev_txq_to_tc); 2406 2407 #ifdef CONFIG_XPS 2408 static struct static_key xps_needed __read_mostly; 2409 static struct static_key xps_rxqs_needed __read_mostly; 2410 static DEFINE_MUTEX(xps_map_mutex); 2411 #define xmap_dereference(P) \ 2412 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) 2413 2414 static bool remove_xps_queue(struct xps_dev_maps *dev_maps, 2415 struct xps_dev_maps *old_maps, int tci, u16 index) 2416 { 2417 struct xps_map *map = NULL; 2418 int pos; 2419 2420 map = xmap_dereference(dev_maps->attr_map[tci]); 2421 if (!map) 2422 return false; 2423 2424 for (pos = map->len; pos--;) { 2425 if (map->queues[pos] != index) 2426 continue; 2427 2428 if (map->len > 1) { 2429 map->queues[pos] = map->queues[--map->len]; 2430 break; 2431 } 2432 2433 if (old_maps) 2434 RCU_INIT_POINTER(old_maps->attr_map[tci], NULL); 2435 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2436 kfree_rcu(map, rcu); 2437 return false; 2438 } 2439 2440 return true; 2441 } 2442 2443 static bool remove_xps_queue_cpu(struct net_device *dev, 2444 struct xps_dev_maps *dev_maps, 2445 int cpu, u16 offset, u16 count) 2446 { 2447 int num_tc = dev_maps->num_tc; 2448 bool active = false; 2449 int tci; 2450 2451 for (tci = cpu * num_tc; num_tc--; tci++) { 2452 int i, j; 2453 2454 for (i = count, j = offset; i--; j++) { 2455 if (!remove_xps_queue(dev_maps, NULL, tci, j)) 2456 break; 2457 } 2458 2459 active |= i < 0; 2460 } 2461 2462 return active; 2463 } 2464 2465 static void reset_xps_maps(struct net_device *dev, 2466 struct xps_dev_maps *dev_maps, 2467 enum xps_map_type type) 2468 { 2469 static_key_slow_dec_cpuslocked(&xps_needed); 2470 if (type == XPS_RXQS) 2471 static_key_slow_dec_cpuslocked(&xps_rxqs_needed); 2472 2473 RCU_INIT_POINTER(dev->xps_maps[type], NULL); 2474 2475 kfree_rcu(dev_maps, rcu); 2476 } 2477 2478 static void clean_xps_maps(struct net_device *dev, enum xps_map_type type, 2479 u16 offset, u16 count) 2480 { 2481 struct xps_dev_maps *dev_maps; 2482 bool active = false; 2483 int i, j; 2484 2485 dev_maps = xmap_dereference(dev->xps_maps[type]); 2486 if (!dev_maps) 2487 return; 2488 2489 for (j = 0; j < dev_maps->nr_ids; j++) 2490 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, count); 2491 if (!active) 2492 reset_xps_maps(dev, dev_maps, type); 2493 2494 if (type == XPS_CPUS) { 2495 for (i = offset + (count - 1); count--; i--) 2496 netdev_queue_numa_node_write( 2497 netdev_get_tx_queue(dev, i), NUMA_NO_NODE); 2498 } 2499 } 2500 2501 static void netif_reset_xps_queues(struct net_device *dev, u16 offset, 2502 u16 count) 2503 { 2504 if (!static_key_false(&xps_needed)) 2505 return; 2506 2507 cpus_read_lock(); 2508 mutex_lock(&xps_map_mutex); 2509 2510 if (static_key_false(&xps_rxqs_needed)) 2511 clean_xps_maps(dev, XPS_RXQS, offset, count); 2512 2513 clean_xps_maps(dev, XPS_CPUS, offset, count); 2514 2515 mutex_unlock(&xps_map_mutex); 2516 cpus_read_unlock(); 2517 } 2518 2519 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) 2520 { 2521 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index); 2522 } 2523 2524 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index, 2525 u16 index, bool is_rxqs_map) 2526 { 2527 struct xps_map *new_map; 2528 int alloc_len = XPS_MIN_MAP_ALLOC; 2529 int i, pos; 2530 2531 for (pos = 0; map && pos < map->len; pos++) { 2532 if (map->queues[pos] != index) 2533 continue; 2534 return map; 2535 } 2536 2537 /* Need to add tx-queue to this CPU's/rx-queue's existing map */ 2538 if (map) { 2539 if (pos < map->alloc_len) 2540 return map; 2541 2542 alloc_len = map->alloc_len * 2; 2543 } 2544 2545 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's 2546 * map 2547 */ 2548 if (is_rxqs_map) 2549 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL); 2550 else 2551 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, 2552 cpu_to_node(attr_index)); 2553 if (!new_map) 2554 return NULL; 2555 2556 for (i = 0; i < pos; i++) 2557 new_map->queues[i] = map->queues[i]; 2558 new_map->alloc_len = alloc_len; 2559 new_map->len = pos; 2560 2561 return new_map; 2562 } 2563 2564 /* Copy xps maps at a given index */ 2565 static void xps_copy_dev_maps(struct xps_dev_maps *dev_maps, 2566 struct xps_dev_maps *new_dev_maps, int index, 2567 int tc, bool skip_tc) 2568 { 2569 int i, tci = index * dev_maps->num_tc; 2570 struct xps_map *map; 2571 2572 /* copy maps belonging to foreign traffic classes */ 2573 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2574 if (i == tc && skip_tc) 2575 continue; 2576 2577 /* fill in the new device map from the old device map */ 2578 map = xmap_dereference(dev_maps->attr_map[tci]); 2579 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2580 } 2581 } 2582 2583 /* Must be called under cpus_read_lock */ 2584 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask, 2585 u16 index, enum xps_map_type type) 2586 { 2587 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL, *old_dev_maps = NULL; 2588 const unsigned long *online_mask = NULL; 2589 bool active = false, copy = false; 2590 int i, j, tci, numa_node_id = -2; 2591 int maps_sz, num_tc = 1, tc = 0; 2592 struct xps_map *map, *new_map; 2593 unsigned int nr_ids; 2594 2595 WARN_ON_ONCE(index >= dev->num_tx_queues); 2596 2597 if (dev->num_tc) { 2598 /* Do not allow XPS on subordinate device directly */ 2599 num_tc = dev->num_tc; 2600 if (num_tc < 0) 2601 return -EINVAL; 2602 2603 /* If queue belongs to subordinate dev use its map */ 2604 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev; 2605 2606 tc = netdev_txq_to_tc(dev, index); 2607 if (tc < 0) 2608 return -EINVAL; 2609 } 2610 2611 mutex_lock(&xps_map_mutex); 2612 2613 dev_maps = xmap_dereference(dev->xps_maps[type]); 2614 if (type == XPS_RXQS) { 2615 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues); 2616 nr_ids = dev->num_rx_queues; 2617 } else { 2618 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc); 2619 if (num_possible_cpus() > 1) 2620 online_mask = cpumask_bits(cpu_online_mask); 2621 nr_ids = nr_cpu_ids; 2622 } 2623 2624 if (maps_sz < L1_CACHE_BYTES) 2625 maps_sz = L1_CACHE_BYTES; 2626 2627 /* The old dev_maps could be larger or smaller than the one we're 2628 * setting up now, as dev->num_tc or nr_ids could have been updated in 2629 * between. We could try to be smart, but let's be safe instead and only 2630 * copy foreign traffic classes if the two map sizes match. 2631 */ 2632 if (dev_maps && 2633 dev_maps->num_tc == num_tc && dev_maps->nr_ids == nr_ids) 2634 copy = true; 2635 2636 /* allocate memory for queue storage */ 2637 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids), 2638 j < nr_ids;) { 2639 if (!new_dev_maps) { 2640 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); 2641 if (!new_dev_maps) { 2642 mutex_unlock(&xps_map_mutex); 2643 return -ENOMEM; 2644 } 2645 2646 new_dev_maps->nr_ids = nr_ids; 2647 new_dev_maps->num_tc = num_tc; 2648 } 2649 2650 tci = j * num_tc + tc; 2651 map = copy ? xmap_dereference(dev_maps->attr_map[tci]) : NULL; 2652 2653 map = expand_xps_map(map, j, index, type == XPS_RXQS); 2654 if (!map) 2655 goto error; 2656 2657 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2658 } 2659 2660 if (!new_dev_maps) 2661 goto out_no_new_maps; 2662 2663 if (!dev_maps) { 2664 /* Increment static keys at most once per type */ 2665 static_key_slow_inc_cpuslocked(&xps_needed); 2666 if (type == XPS_RXQS) 2667 static_key_slow_inc_cpuslocked(&xps_rxqs_needed); 2668 } 2669 2670 for (j = 0; j < nr_ids; j++) { 2671 bool skip_tc = false; 2672 2673 tci = j * num_tc + tc; 2674 if (netif_attr_test_mask(j, mask, nr_ids) && 2675 netif_attr_test_online(j, online_mask, nr_ids)) { 2676 /* add tx-queue to CPU/rx-queue maps */ 2677 int pos = 0; 2678 2679 skip_tc = true; 2680 2681 map = xmap_dereference(new_dev_maps->attr_map[tci]); 2682 while ((pos < map->len) && (map->queues[pos] != index)) 2683 pos++; 2684 2685 if (pos == map->len) 2686 map->queues[map->len++] = index; 2687 #ifdef CONFIG_NUMA 2688 if (type == XPS_CPUS) { 2689 if (numa_node_id == -2) 2690 numa_node_id = cpu_to_node(j); 2691 else if (numa_node_id != cpu_to_node(j)) 2692 numa_node_id = -1; 2693 } 2694 #endif 2695 } 2696 2697 if (copy) 2698 xps_copy_dev_maps(dev_maps, new_dev_maps, j, tc, 2699 skip_tc); 2700 } 2701 2702 rcu_assign_pointer(dev->xps_maps[type], new_dev_maps); 2703 2704 /* Cleanup old maps */ 2705 if (!dev_maps) 2706 goto out_no_old_maps; 2707 2708 for (j = 0; j < dev_maps->nr_ids; j++) { 2709 for (i = num_tc, tci = j * dev_maps->num_tc; i--; tci++) { 2710 map = xmap_dereference(dev_maps->attr_map[tci]); 2711 if (!map) 2712 continue; 2713 2714 if (copy) { 2715 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2716 if (map == new_map) 2717 continue; 2718 } 2719 2720 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2721 kfree_rcu(map, rcu); 2722 } 2723 } 2724 2725 old_dev_maps = dev_maps; 2726 2727 out_no_old_maps: 2728 dev_maps = new_dev_maps; 2729 active = true; 2730 2731 out_no_new_maps: 2732 if (type == XPS_CPUS) 2733 /* update Tx queue numa node */ 2734 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), 2735 (numa_node_id >= 0) ? 2736 numa_node_id : NUMA_NO_NODE); 2737 2738 if (!dev_maps) 2739 goto out_no_maps; 2740 2741 /* removes tx-queue from unused CPUs/rx-queues */ 2742 for (j = 0; j < dev_maps->nr_ids; j++) { 2743 tci = j * dev_maps->num_tc; 2744 2745 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2746 if (i == tc && 2747 netif_attr_test_mask(j, mask, dev_maps->nr_ids) && 2748 netif_attr_test_online(j, online_mask, dev_maps->nr_ids)) 2749 continue; 2750 2751 active |= remove_xps_queue(dev_maps, 2752 copy ? old_dev_maps : NULL, 2753 tci, index); 2754 } 2755 } 2756 2757 if (old_dev_maps) 2758 kfree_rcu(old_dev_maps, rcu); 2759 2760 /* free map if not active */ 2761 if (!active) 2762 reset_xps_maps(dev, dev_maps, type); 2763 2764 out_no_maps: 2765 mutex_unlock(&xps_map_mutex); 2766 2767 return 0; 2768 error: 2769 /* remove any maps that we added */ 2770 for (j = 0; j < nr_ids; j++) { 2771 for (i = num_tc, tci = j * num_tc; i--; tci++) { 2772 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2773 map = copy ? 2774 xmap_dereference(dev_maps->attr_map[tci]) : 2775 NULL; 2776 if (new_map && new_map != map) 2777 kfree(new_map); 2778 } 2779 } 2780 2781 mutex_unlock(&xps_map_mutex); 2782 2783 kfree(new_dev_maps); 2784 return -ENOMEM; 2785 } 2786 EXPORT_SYMBOL_GPL(__netif_set_xps_queue); 2787 2788 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, 2789 u16 index) 2790 { 2791 int ret; 2792 2793 cpus_read_lock(); 2794 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, XPS_CPUS); 2795 cpus_read_unlock(); 2796 2797 return ret; 2798 } 2799 EXPORT_SYMBOL(netif_set_xps_queue); 2800 2801 #endif 2802 static void netdev_unbind_all_sb_channels(struct net_device *dev) 2803 { 2804 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2805 2806 /* Unbind any subordinate channels */ 2807 while (txq-- != &dev->_tx[0]) { 2808 if (txq->sb_dev) 2809 netdev_unbind_sb_channel(dev, txq->sb_dev); 2810 } 2811 } 2812 2813 void netdev_reset_tc(struct net_device *dev) 2814 { 2815 #ifdef CONFIG_XPS 2816 netif_reset_xps_queues_gt(dev, 0); 2817 #endif 2818 netdev_unbind_all_sb_channels(dev); 2819 2820 /* Reset TC configuration of device */ 2821 dev->num_tc = 0; 2822 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq)); 2823 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map)); 2824 } 2825 EXPORT_SYMBOL(netdev_reset_tc); 2826 2827 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset) 2828 { 2829 if (tc >= dev->num_tc) 2830 return -EINVAL; 2831 2832 #ifdef CONFIG_XPS 2833 netif_reset_xps_queues(dev, offset, count); 2834 #endif 2835 dev->tc_to_txq[tc].count = count; 2836 dev->tc_to_txq[tc].offset = offset; 2837 return 0; 2838 } 2839 EXPORT_SYMBOL(netdev_set_tc_queue); 2840 2841 int netdev_set_num_tc(struct net_device *dev, u8 num_tc) 2842 { 2843 if (num_tc > TC_MAX_QUEUE) 2844 return -EINVAL; 2845 2846 #ifdef CONFIG_XPS 2847 netif_reset_xps_queues_gt(dev, 0); 2848 #endif 2849 netdev_unbind_all_sb_channels(dev); 2850 2851 dev->num_tc = num_tc; 2852 return 0; 2853 } 2854 EXPORT_SYMBOL(netdev_set_num_tc); 2855 2856 void netdev_unbind_sb_channel(struct net_device *dev, 2857 struct net_device *sb_dev) 2858 { 2859 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2860 2861 #ifdef CONFIG_XPS 2862 netif_reset_xps_queues_gt(sb_dev, 0); 2863 #endif 2864 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq)); 2865 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map)); 2866 2867 while (txq-- != &dev->_tx[0]) { 2868 if (txq->sb_dev == sb_dev) 2869 txq->sb_dev = NULL; 2870 } 2871 } 2872 EXPORT_SYMBOL(netdev_unbind_sb_channel); 2873 2874 int netdev_bind_sb_channel_queue(struct net_device *dev, 2875 struct net_device *sb_dev, 2876 u8 tc, u16 count, u16 offset) 2877 { 2878 /* Make certain the sb_dev and dev are already configured */ 2879 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc) 2880 return -EINVAL; 2881 2882 /* We cannot hand out queues we don't have */ 2883 if ((offset + count) > dev->real_num_tx_queues) 2884 return -EINVAL; 2885 2886 /* Record the mapping */ 2887 sb_dev->tc_to_txq[tc].count = count; 2888 sb_dev->tc_to_txq[tc].offset = offset; 2889 2890 /* Provide a way for Tx queue to find the tc_to_txq map or 2891 * XPS map for itself. 2892 */ 2893 while (count--) 2894 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev; 2895 2896 return 0; 2897 } 2898 EXPORT_SYMBOL(netdev_bind_sb_channel_queue); 2899 2900 int netdev_set_sb_channel(struct net_device *dev, u16 channel) 2901 { 2902 /* Do not use a multiqueue device to represent a subordinate channel */ 2903 if (netif_is_multiqueue(dev)) 2904 return -ENODEV; 2905 2906 /* We allow channels 1 - 32767 to be used for subordinate channels. 2907 * Channel 0 is meant to be "native" mode and used only to represent 2908 * the main root device. We allow writing 0 to reset the device back 2909 * to normal mode after being used as a subordinate channel. 2910 */ 2911 if (channel > S16_MAX) 2912 return -EINVAL; 2913 2914 dev->num_tc = -channel; 2915 2916 return 0; 2917 } 2918 EXPORT_SYMBOL(netdev_set_sb_channel); 2919 2920 /* 2921 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues 2922 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed. 2923 */ 2924 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) 2925 { 2926 bool disabling; 2927 int rc; 2928 2929 disabling = txq < dev->real_num_tx_queues; 2930 2931 if (txq < 1 || txq > dev->num_tx_queues) 2932 return -EINVAL; 2933 2934 if (dev->reg_state == NETREG_REGISTERED || 2935 dev->reg_state == NETREG_UNREGISTERING) { 2936 ASSERT_RTNL(); 2937 2938 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, 2939 txq); 2940 if (rc) 2941 return rc; 2942 2943 if (dev->num_tc) 2944 netif_setup_tc(dev, txq); 2945 2946 dev_qdisc_change_real_num_tx(dev, txq); 2947 2948 dev->real_num_tx_queues = txq; 2949 2950 if (disabling) { 2951 synchronize_net(); 2952 qdisc_reset_all_tx_gt(dev, txq); 2953 #ifdef CONFIG_XPS 2954 netif_reset_xps_queues_gt(dev, txq); 2955 #endif 2956 } 2957 } else { 2958 dev->real_num_tx_queues = txq; 2959 } 2960 2961 return 0; 2962 } 2963 EXPORT_SYMBOL(netif_set_real_num_tx_queues); 2964 2965 #ifdef CONFIG_SYSFS 2966 /** 2967 * netif_set_real_num_rx_queues - set actual number of RX queues used 2968 * @dev: Network device 2969 * @rxq: Actual number of RX queues 2970 * 2971 * This must be called either with the rtnl_lock held or before 2972 * registration of the net device. Returns 0 on success, or a 2973 * negative error code. If called before registration, it always 2974 * succeeds. 2975 */ 2976 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) 2977 { 2978 int rc; 2979 2980 if (rxq < 1 || rxq > dev->num_rx_queues) 2981 return -EINVAL; 2982 2983 if (dev->reg_state == NETREG_REGISTERED) { 2984 ASSERT_RTNL(); 2985 2986 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, 2987 rxq); 2988 if (rc) 2989 return rc; 2990 } 2991 2992 dev->real_num_rx_queues = rxq; 2993 return 0; 2994 } 2995 EXPORT_SYMBOL(netif_set_real_num_rx_queues); 2996 #endif 2997 2998 /** 2999 * netif_set_real_num_queues - set actual number of RX and TX queues used 3000 * @dev: Network device 3001 * @txq: Actual number of TX queues 3002 * @rxq: Actual number of RX queues 3003 * 3004 * Set the real number of both TX and RX queues. 3005 * Does nothing if the number of queues is already correct. 3006 */ 3007 int netif_set_real_num_queues(struct net_device *dev, 3008 unsigned int txq, unsigned int rxq) 3009 { 3010 unsigned int old_rxq = dev->real_num_rx_queues; 3011 int err; 3012 3013 if (txq < 1 || txq > dev->num_tx_queues || 3014 rxq < 1 || rxq > dev->num_rx_queues) 3015 return -EINVAL; 3016 3017 /* Start from increases, so the error path only does decreases - 3018 * decreases can't fail. 3019 */ 3020 if (rxq > dev->real_num_rx_queues) { 3021 err = netif_set_real_num_rx_queues(dev, rxq); 3022 if (err) 3023 return err; 3024 } 3025 if (txq > dev->real_num_tx_queues) { 3026 err = netif_set_real_num_tx_queues(dev, txq); 3027 if (err) 3028 goto undo_rx; 3029 } 3030 if (rxq < dev->real_num_rx_queues) 3031 WARN_ON(netif_set_real_num_rx_queues(dev, rxq)); 3032 if (txq < dev->real_num_tx_queues) 3033 WARN_ON(netif_set_real_num_tx_queues(dev, txq)); 3034 3035 return 0; 3036 undo_rx: 3037 WARN_ON(netif_set_real_num_rx_queues(dev, old_rxq)); 3038 return err; 3039 } 3040 EXPORT_SYMBOL(netif_set_real_num_queues); 3041 3042 /** 3043 * netif_set_tso_max_size() - set the max size of TSO frames supported 3044 * @dev: netdev to update 3045 * @size: max skb->len of a TSO frame 3046 * 3047 * Set the limit on the size of TSO super-frames the device can handle. 3048 * Unless explicitly set the stack will assume the value of 3049 * %GSO_LEGACY_MAX_SIZE. 3050 */ 3051 void netif_set_tso_max_size(struct net_device *dev, unsigned int size) 3052 { 3053 dev->tso_max_size = min(GSO_MAX_SIZE, size); 3054 if (size < READ_ONCE(dev->gso_max_size)) 3055 netif_set_gso_max_size(dev, size); 3056 if (size < READ_ONCE(dev->gso_ipv4_max_size)) 3057 netif_set_gso_ipv4_max_size(dev, size); 3058 } 3059 EXPORT_SYMBOL(netif_set_tso_max_size); 3060 3061 /** 3062 * netif_set_tso_max_segs() - set the max number of segs supported for TSO 3063 * @dev: netdev to update 3064 * @segs: max number of TCP segments 3065 * 3066 * Set the limit on the number of TCP segments the device can generate from 3067 * a single TSO super-frame. 3068 * Unless explicitly set the stack will assume the value of %GSO_MAX_SEGS. 3069 */ 3070 void netif_set_tso_max_segs(struct net_device *dev, unsigned int segs) 3071 { 3072 dev->tso_max_segs = segs; 3073 if (segs < READ_ONCE(dev->gso_max_segs)) 3074 netif_set_gso_max_segs(dev, segs); 3075 } 3076 EXPORT_SYMBOL(netif_set_tso_max_segs); 3077 3078 /** 3079 * netif_inherit_tso_max() - copy all TSO limits from a lower device to an upper 3080 * @to: netdev to update 3081 * @from: netdev from which to copy the limits 3082 */ 3083 void netif_inherit_tso_max(struct net_device *to, const struct net_device *from) 3084 { 3085 netif_set_tso_max_size(to, from->tso_max_size); 3086 netif_set_tso_max_segs(to, from->tso_max_segs); 3087 } 3088 EXPORT_SYMBOL(netif_inherit_tso_max); 3089 3090 /** 3091 * netif_get_num_default_rss_queues - default number of RSS queues 3092 * 3093 * Default value is the number of physical cores if there are only 1 or 2, or 3094 * divided by 2 if there are more. 3095 */ 3096 int netif_get_num_default_rss_queues(void) 3097 { 3098 cpumask_var_t cpus; 3099 int cpu, count = 0; 3100 3101 if (unlikely(is_kdump_kernel() || !zalloc_cpumask_var(&cpus, GFP_KERNEL))) 3102 return 1; 3103 3104 cpumask_copy(cpus, cpu_online_mask); 3105 for_each_cpu(cpu, cpus) { 3106 ++count; 3107 cpumask_andnot(cpus, cpus, topology_sibling_cpumask(cpu)); 3108 } 3109 free_cpumask_var(cpus); 3110 3111 return count > 2 ? DIV_ROUND_UP(count, 2) : count; 3112 } 3113 EXPORT_SYMBOL(netif_get_num_default_rss_queues); 3114 3115 static void __netif_reschedule(struct Qdisc *q) 3116 { 3117 struct softnet_data *sd; 3118 unsigned long flags; 3119 3120 local_irq_save(flags); 3121 sd = this_cpu_ptr(&softnet_data); 3122 q->next_sched = NULL; 3123 *sd->output_queue_tailp = q; 3124 sd->output_queue_tailp = &q->next_sched; 3125 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3126 local_irq_restore(flags); 3127 } 3128 3129 void __netif_schedule(struct Qdisc *q) 3130 { 3131 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) 3132 __netif_reschedule(q); 3133 } 3134 EXPORT_SYMBOL(__netif_schedule); 3135 3136 struct dev_kfree_skb_cb { 3137 enum skb_drop_reason reason; 3138 }; 3139 3140 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) 3141 { 3142 return (struct dev_kfree_skb_cb *)skb->cb; 3143 } 3144 3145 void netif_schedule_queue(struct netdev_queue *txq) 3146 { 3147 rcu_read_lock(); 3148 if (!netif_xmit_stopped(txq)) { 3149 struct Qdisc *q = rcu_dereference(txq->qdisc); 3150 3151 __netif_schedule(q); 3152 } 3153 rcu_read_unlock(); 3154 } 3155 EXPORT_SYMBOL(netif_schedule_queue); 3156 3157 void netif_tx_wake_queue(struct netdev_queue *dev_queue) 3158 { 3159 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { 3160 struct Qdisc *q; 3161 3162 rcu_read_lock(); 3163 q = rcu_dereference(dev_queue->qdisc); 3164 __netif_schedule(q); 3165 rcu_read_unlock(); 3166 } 3167 } 3168 EXPORT_SYMBOL(netif_tx_wake_queue); 3169 3170 void dev_kfree_skb_irq_reason(struct sk_buff *skb, enum skb_drop_reason reason) 3171 { 3172 unsigned long flags; 3173 3174 if (unlikely(!skb)) 3175 return; 3176 3177 if (likely(refcount_read(&skb->users) == 1)) { 3178 smp_rmb(); 3179 refcount_set(&skb->users, 0); 3180 } else if (likely(!refcount_dec_and_test(&skb->users))) { 3181 return; 3182 } 3183 get_kfree_skb_cb(skb)->reason = reason; 3184 local_irq_save(flags); 3185 skb->next = __this_cpu_read(softnet_data.completion_queue); 3186 __this_cpu_write(softnet_data.completion_queue, skb); 3187 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3188 local_irq_restore(flags); 3189 } 3190 EXPORT_SYMBOL(dev_kfree_skb_irq_reason); 3191 3192 void dev_kfree_skb_any_reason(struct sk_buff *skb, enum skb_drop_reason reason) 3193 { 3194 if (in_hardirq() || irqs_disabled()) 3195 dev_kfree_skb_irq_reason(skb, reason); 3196 else 3197 kfree_skb_reason(skb, reason); 3198 } 3199 EXPORT_SYMBOL(dev_kfree_skb_any_reason); 3200 3201 3202 /** 3203 * netif_device_detach - mark device as removed 3204 * @dev: network device 3205 * 3206 * Mark device as removed from system and therefore no longer available. 3207 */ 3208 void netif_device_detach(struct net_device *dev) 3209 { 3210 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && 3211 netif_running(dev)) { 3212 netif_tx_stop_all_queues(dev); 3213 } 3214 } 3215 EXPORT_SYMBOL(netif_device_detach); 3216 3217 /** 3218 * netif_device_attach - mark device as attached 3219 * @dev: network device 3220 * 3221 * Mark device as attached from system and restart if needed. 3222 */ 3223 void netif_device_attach(struct net_device *dev) 3224 { 3225 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && 3226 netif_running(dev)) { 3227 netif_tx_wake_all_queues(dev); 3228 __netdev_watchdog_up(dev); 3229 } 3230 } 3231 EXPORT_SYMBOL(netif_device_attach); 3232 3233 /* 3234 * Returns a Tx hash based on the given packet descriptor a Tx queues' number 3235 * to be used as a distribution range. 3236 */ 3237 static u16 skb_tx_hash(const struct net_device *dev, 3238 const struct net_device *sb_dev, 3239 struct sk_buff *skb) 3240 { 3241 u32 hash; 3242 u16 qoffset = 0; 3243 u16 qcount = dev->real_num_tx_queues; 3244 3245 if (dev->num_tc) { 3246 u8 tc = netdev_get_prio_tc_map(dev, skb->priority); 3247 3248 qoffset = sb_dev->tc_to_txq[tc].offset; 3249 qcount = sb_dev->tc_to_txq[tc].count; 3250 if (unlikely(!qcount)) { 3251 net_warn_ratelimited("%s: invalid qcount, qoffset %u for tc %u\n", 3252 sb_dev->name, qoffset, tc); 3253 qoffset = 0; 3254 qcount = dev->real_num_tx_queues; 3255 } 3256 } 3257 3258 if (skb_rx_queue_recorded(skb)) { 3259 DEBUG_NET_WARN_ON_ONCE(qcount == 0); 3260 hash = skb_get_rx_queue(skb); 3261 if (hash >= qoffset) 3262 hash -= qoffset; 3263 while (unlikely(hash >= qcount)) 3264 hash -= qcount; 3265 return hash + qoffset; 3266 } 3267 3268 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; 3269 } 3270 3271 void skb_warn_bad_offload(const struct sk_buff *skb) 3272 { 3273 static const netdev_features_t null_features; 3274 struct net_device *dev = skb->dev; 3275 const char *name = ""; 3276 3277 if (!net_ratelimit()) 3278 return; 3279 3280 if (dev) { 3281 if (dev->dev.parent) 3282 name = dev_driver_string(dev->dev.parent); 3283 else 3284 name = netdev_name(dev); 3285 } 3286 skb_dump(KERN_WARNING, skb, false); 3287 WARN(1, "%s: caps=(%pNF, %pNF)\n", 3288 name, dev ? &dev->features : &null_features, 3289 skb->sk ? &skb->sk->sk_route_caps : &null_features); 3290 } 3291 3292 /* 3293 * Invalidate hardware checksum when packet is to be mangled, and 3294 * complete checksum manually on outgoing path. 3295 */ 3296 int skb_checksum_help(struct sk_buff *skb) 3297 { 3298 __wsum csum; 3299 int ret = 0, offset; 3300 3301 if (skb->ip_summed == CHECKSUM_COMPLETE) 3302 goto out_set_summed; 3303 3304 if (unlikely(skb_is_gso(skb))) { 3305 skb_warn_bad_offload(skb); 3306 return -EINVAL; 3307 } 3308 3309 /* Before computing a checksum, we should make sure no frag could 3310 * be modified by an external entity : checksum could be wrong. 3311 */ 3312 if (skb_has_shared_frag(skb)) { 3313 ret = __skb_linearize(skb); 3314 if (ret) 3315 goto out; 3316 } 3317 3318 offset = skb_checksum_start_offset(skb); 3319 ret = -EINVAL; 3320 if (unlikely(offset >= skb_headlen(skb))) { 3321 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false); 3322 WARN_ONCE(true, "offset (%d) >= skb_headlen() (%u)\n", 3323 offset, skb_headlen(skb)); 3324 goto out; 3325 } 3326 csum = skb_checksum(skb, offset, skb->len - offset, 0); 3327 3328 offset += skb->csum_offset; 3329 if (unlikely(offset + sizeof(__sum16) > skb_headlen(skb))) { 3330 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false); 3331 WARN_ONCE(true, "offset+2 (%zu) > skb_headlen() (%u)\n", 3332 offset + sizeof(__sum16), skb_headlen(skb)); 3333 goto out; 3334 } 3335 ret = skb_ensure_writable(skb, offset + sizeof(__sum16)); 3336 if (ret) 3337 goto out; 3338 3339 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0; 3340 out_set_summed: 3341 skb->ip_summed = CHECKSUM_NONE; 3342 out: 3343 return ret; 3344 } 3345 EXPORT_SYMBOL(skb_checksum_help); 3346 3347 int skb_crc32c_csum_help(struct sk_buff *skb) 3348 { 3349 __le32 crc32c_csum; 3350 int ret = 0, offset, start; 3351 3352 if (skb->ip_summed != CHECKSUM_PARTIAL) 3353 goto out; 3354 3355 if (unlikely(skb_is_gso(skb))) 3356 goto out; 3357 3358 /* Before computing a checksum, we should make sure no frag could 3359 * be modified by an external entity : checksum could be wrong. 3360 */ 3361 if (unlikely(skb_has_shared_frag(skb))) { 3362 ret = __skb_linearize(skb); 3363 if (ret) 3364 goto out; 3365 } 3366 start = skb_checksum_start_offset(skb); 3367 offset = start + offsetof(struct sctphdr, checksum); 3368 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { 3369 ret = -EINVAL; 3370 goto out; 3371 } 3372 3373 ret = skb_ensure_writable(skb, offset + sizeof(__le32)); 3374 if (ret) 3375 goto out; 3376 3377 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start, 3378 skb->len - start, ~(__u32)0, 3379 crc32c_csum_stub)); 3380 *(__le32 *)(skb->data + offset) = crc32c_csum; 3381 skb_reset_csum_not_inet(skb); 3382 out: 3383 return ret; 3384 } 3385 3386 __be16 skb_network_protocol(struct sk_buff *skb, int *depth) 3387 { 3388 __be16 type = skb->protocol; 3389 3390 /* Tunnel gso handlers can set protocol to ethernet. */ 3391 if (type == htons(ETH_P_TEB)) { 3392 struct ethhdr *eth; 3393 3394 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) 3395 return 0; 3396 3397 eth = (struct ethhdr *)skb->data; 3398 type = eth->h_proto; 3399 } 3400 3401 return vlan_get_protocol_and_depth(skb, type, depth); 3402 } 3403 3404 3405 /* Take action when hardware reception checksum errors are detected. */ 3406 #ifdef CONFIG_BUG 3407 static void do_netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3408 { 3409 netdev_err(dev, "hw csum failure\n"); 3410 skb_dump(KERN_ERR, skb, true); 3411 dump_stack(); 3412 } 3413 3414 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3415 { 3416 DO_ONCE_LITE(do_netdev_rx_csum_fault, dev, skb); 3417 } 3418 EXPORT_SYMBOL(netdev_rx_csum_fault); 3419 #endif 3420 3421 /* XXX: check that highmem exists at all on the given machine. */ 3422 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) 3423 { 3424 #ifdef CONFIG_HIGHMEM 3425 int i; 3426 3427 if (!(dev->features & NETIF_F_HIGHDMA)) { 3428 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 3429 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 3430 3431 if (PageHighMem(skb_frag_page(frag))) 3432 return 1; 3433 } 3434 } 3435 #endif 3436 return 0; 3437 } 3438 3439 /* If MPLS offload request, verify we are testing hardware MPLS features 3440 * instead of standard features for the netdev. 3441 */ 3442 #if IS_ENABLED(CONFIG_NET_MPLS_GSO) 3443 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3444 netdev_features_t features, 3445 __be16 type) 3446 { 3447 if (eth_p_mpls(type)) 3448 features &= skb->dev->mpls_features; 3449 3450 return features; 3451 } 3452 #else 3453 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3454 netdev_features_t features, 3455 __be16 type) 3456 { 3457 return features; 3458 } 3459 #endif 3460 3461 static netdev_features_t harmonize_features(struct sk_buff *skb, 3462 netdev_features_t features) 3463 { 3464 __be16 type; 3465 3466 type = skb_network_protocol(skb, NULL); 3467 features = net_mpls_features(skb, features, type); 3468 3469 if (skb->ip_summed != CHECKSUM_NONE && 3470 !can_checksum_protocol(features, type)) { 3471 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); 3472 } 3473 if (illegal_highdma(skb->dev, skb)) 3474 features &= ~NETIF_F_SG; 3475 3476 return features; 3477 } 3478 3479 netdev_features_t passthru_features_check(struct sk_buff *skb, 3480 struct net_device *dev, 3481 netdev_features_t features) 3482 { 3483 return features; 3484 } 3485 EXPORT_SYMBOL(passthru_features_check); 3486 3487 static netdev_features_t dflt_features_check(struct sk_buff *skb, 3488 struct net_device *dev, 3489 netdev_features_t features) 3490 { 3491 return vlan_features_check(skb, features); 3492 } 3493 3494 static netdev_features_t gso_features_check(const struct sk_buff *skb, 3495 struct net_device *dev, 3496 netdev_features_t features) 3497 { 3498 u16 gso_segs = skb_shinfo(skb)->gso_segs; 3499 3500 if (gso_segs > READ_ONCE(dev->gso_max_segs)) 3501 return features & ~NETIF_F_GSO_MASK; 3502 3503 if (unlikely(skb->len >= READ_ONCE(dev->gso_max_size))) 3504 return features & ~NETIF_F_GSO_MASK; 3505 3506 if (!skb_shinfo(skb)->gso_type) { 3507 skb_warn_bad_offload(skb); 3508 return features & ~NETIF_F_GSO_MASK; 3509 } 3510 3511 /* Support for GSO partial features requires software 3512 * intervention before we can actually process the packets 3513 * so we need to strip support for any partial features now 3514 * and we can pull them back in after we have partially 3515 * segmented the frame. 3516 */ 3517 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)) 3518 features &= ~dev->gso_partial_features; 3519 3520 /* Make sure to clear the IPv4 ID mangling feature if the 3521 * IPv4 header has the potential to be fragmented. 3522 */ 3523 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { 3524 struct iphdr *iph = skb->encapsulation ? 3525 inner_ip_hdr(skb) : ip_hdr(skb); 3526 3527 if (!(iph->frag_off & htons(IP_DF))) 3528 features &= ~NETIF_F_TSO_MANGLEID; 3529 } 3530 3531 return features; 3532 } 3533 3534 netdev_features_t netif_skb_features(struct sk_buff *skb) 3535 { 3536 struct net_device *dev = skb->dev; 3537 netdev_features_t features = dev->features; 3538 3539 if (skb_is_gso(skb)) 3540 features = gso_features_check(skb, dev, features); 3541 3542 /* If encapsulation offload request, verify we are testing 3543 * hardware encapsulation features instead of standard 3544 * features for the netdev 3545 */ 3546 if (skb->encapsulation) 3547 features &= dev->hw_enc_features; 3548 3549 if (skb_vlan_tagged(skb)) 3550 features = netdev_intersect_features(features, 3551 dev->vlan_features | 3552 NETIF_F_HW_VLAN_CTAG_TX | 3553 NETIF_F_HW_VLAN_STAG_TX); 3554 3555 if (dev->netdev_ops->ndo_features_check) 3556 features &= dev->netdev_ops->ndo_features_check(skb, dev, 3557 features); 3558 else 3559 features &= dflt_features_check(skb, dev, features); 3560 3561 return harmonize_features(skb, features); 3562 } 3563 EXPORT_SYMBOL(netif_skb_features); 3564 3565 static int xmit_one(struct sk_buff *skb, struct net_device *dev, 3566 struct netdev_queue *txq, bool more) 3567 { 3568 unsigned int len; 3569 int rc; 3570 3571 if (dev_nit_active(dev)) 3572 dev_queue_xmit_nit(skb, dev); 3573 3574 len = skb->len; 3575 trace_net_dev_start_xmit(skb, dev); 3576 rc = netdev_start_xmit(skb, dev, txq, more); 3577 trace_net_dev_xmit(skb, rc, dev, len); 3578 3579 return rc; 3580 } 3581 3582 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, 3583 struct netdev_queue *txq, int *ret) 3584 { 3585 struct sk_buff *skb = first; 3586 int rc = NETDEV_TX_OK; 3587 3588 while (skb) { 3589 struct sk_buff *next = skb->next; 3590 3591 skb_mark_not_on_list(skb); 3592 rc = xmit_one(skb, dev, txq, next != NULL); 3593 if (unlikely(!dev_xmit_complete(rc))) { 3594 skb->next = next; 3595 goto out; 3596 } 3597 3598 skb = next; 3599 if (netif_tx_queue_stopped(txq) && skb) { 3600 rc = NETDEV_TX_BUSY; 3601 break; 3602 } 3603 } 3604 3605 out: 3606 *ret = rc; 3607 return skb; 3608 } 3609 3610 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, 3611 netdev_features_t features) 3612 { 3613 if (skb_vlan_tag_present(skb) && 3614 !vlan_hw_offload_capable(features, skb->vlan_proto)) 3615 skb = __vlan_hwaccel_push_inside(skb); 3616 return skb; 3617 } 3618 3619 int skb_csum_hwoffload_help(struct sk_buff *skb, 3620 const netdev_features_t features) 3621 { 3622 if (unlikely(skb_csum_is_sctp(skb))) 3623 return !!(features & NETIF_F_SCTP_CRC) ? 0 : 3624 skb_crc32c_csum_help(skb); 3625 3626 if (features & NETIF_F_HW_CSUM) 3627 return 0; 3628 3629 if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) { 3630 switch (skb->csum_offset) { 3631 case offsetof(struct tcphdr, check): 3632 case offsetof(struct udphdr, check): 3633 return 0; 3634 } 3635 } 3636 3637 return skb_checksum_help(skb); 3638 } 3639 EXPORT_SYMBOL(skb_csum_hwoffload_help); 3640 3641 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again) 3642 { 3643 netdev_features_t features; 3644 3645 features = netif_skb_features(skb); 3646 skb = validate_xmit_vlan(skb, features); 3647 if (unlikely(!skb)) 3648 goto out_null; 3649 3650 skb = sk_validate_xmit_skb(skb, dev); 3651 if (unlikely(!skb)) 3652 goto out_null; 3653 3654 if (netif_needs_gso(skb, features)) { 3655 struct sk_buff *segs; 3656 3657 segs = skb_gso_segment(skb, features); 3658 if (IS_ERR(segs)) { 3659 goto out_kfree_skb; 3660 } else if (segs) { 3661 consume_skb(skb); 3662 skb = segs; 3663 } 3664 } else { 3665 if (skb_needs_linearize(skb, features) && 3666 __skb_linearize(skb)) 3667 goto out_kfree_skb; 3668 3669 /* If packet is not checksummed and device does not 3670 * support checksumming for this protocol, complete 3671 * checksumming here. 3672 */ 3673 if (skb->ip_summed == CHECKSUM_PARTIAL) { 3674 if (skb->encapsulation) 3675 skb_set_inner_transport_header(skb, 3676 skb_checksum_start_offset(skb)); 3677 else 3678 skb_set_transport_header(skb, 3679 skb_checksum_start_offset(skb)); 3680 if (skb_csum_hwoffload_help(skb, features)) 3681 goto out_kfree_skb; 3682 } 3683 } 3684 3685 skb = validate_xmit_xfrm(skb, features, again); 3686 3687 return skb; 3688 3689 out_kfree_skb: 3690 kfree_skb(skb); 3691 out_null: 3692 dev_core_stats_tx_dropped_inc(dev); 3693 return NULL; 3694 } 3695 3696 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again) 3697 { 3698 struct sk_buff *next, *head = NULL, *tail; 3699 3700 for (; skb != NULL; skb = next) { 3701 next = skb->next; 3702 skb_mark_not_on_list(skb); 3703 3704 /* in case skb wont be segmented, point to itself */ 3705 skb->prev = skb; 3706 3707 skb = validate_xmit_skb(skb, dev, again); 3708 if (!skb) 3709 continue; 3710 3711 if (!head) 3712 head = skb; 3713 else 3714 tail->next = skb; 3715 /* If skb was segmented, skb->prev points to 3716 * the last segment. If not, it still contains skb. 3717 */ 3718 tail = skb->prev; 3719 } 3720 return head; 3721 } 3722 EXPORT_SYMBOL_GPL(validate_xmit_skb_list); 3723 3724 static void qdisc_pkt_len_init(struct sk_buff *skb) 3725 { 3726 const struct skb_shared_info *shinfo = skb_shinfo(skb); 3727 3728 qdisc_skb_cb(skb)->pkt_len = skb->len; 3729 3730 /* To get more precise estimation of bytes sent on wire, 3731 * we add to pkt_len the headers size of all segments 3732 */ 3733 if (shinfo->gso_size && skb_transport_header_was_set(skb)) { 3734 u16 gso_segs = shinfo->gso_segs; 3735 unsigned int hdr_len; 3736 3737 /* mac layer + network layer */ 3738 hdr_len = skb_transport_offset(skb); 3739 3740 /* + transport layer */ 3741 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { 3742 const struct tcphdr *th; 3743 struct tcphdr _tcphdr; 3744 3745 th = skb_header_pointer(skb, hdr_len, 3746 sizeof(_tcphdr), &_tcphdr); 3747 if (likely(th)) 3748 hdr_len += __tcp_hdrlen(th); 3749 } else { 3750 struct udphdr _udphdr; 3751 3752 if (skb_header_pointer(skb, hdr_len, 3753 sizeof(_udphdr), &_udphdr)) 3754 hdr_len += sizeof(struct udphdr); 3755 } 3756 3757 if (shinfo->gso_type & SKB_GSO_DODGY) 3758 gso_segs = DIV_ROUND_UP(skb->len - hdr_len, 3759 shinfo->gso_size); 3760 3761 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; 3762 } 3763 } 3764 3765 static int dev_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *q, 3766 struct sk_buff **to_free, 3767 struct netdev_queue *txq) 3768 { 3769 int rc; 3770 3771 rc = q->enqueue(skb, q, to_free) & NET_XMIT_MASK; 3772 if (rc == NET_XMIT_SUCCESS) 3773 trace_qdisc_enqueue(q, txq, skb); 3774 return rc; 3775 } 3776 3777 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, 3778 struct net_device *dev, 3779 struct netdev_queue *txq) 3780 { 3781 spinlock_t *root_lock = qdisc_lock(q); 3782 struct sk_buff *to_free = NULL; 3783 bool contended; 3784 int rc; 3785 3786 qdisc_calculate_pkt_len(skb, q); 3787 3788 if (q->flags & TCQ_F_NOLOCK) { 3789 if (q->flags & TCQ_F_CAN_BYPASS && nolock_qdisc_is_empty(q) && 3790 qdisc_run_begin(q)) { 3791 /* Retest nolock_qdisc_is_empty() within the protection 3792 * of q->seqlock to protect from racing with requeuing. 3793 */ 3794 if (unlikely(!nolock_qdisc_is_empty(q))) { 3795 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3796 __qdisc_run(q); 3797 qdisc_run_end(q); 3798 3799 goto no_lock_out; 3800 } 3801 3802 qdisc_bstats_cpu_update(q, skb); 3803 if (sch_direct_xmit(skb, q, dev, txq, NULL, true) && 3804 !nolock_qdisc_is_empty(q)) 3805 __qdisc_run(q); 3806 3807 qdisc_run_end(q); 3808 return NET_XMIT_SUCCESS; 3809 } 3810 3811 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3812 qdisc_run(q); 3813 3814 no_lock_out: 3815 if (unlikely(to_free)) 3816 kfree_skb_list_reason(to_free, 3817 SKB_DROP_REASON_QDISC_DROP); 3818 return rc; 3819 } 3820 3821 /* 3822 * Heuristic to force contended enqueues to serialize on a 3823 * separate lock before trying to get qdisc main lock. 3824 * This permits qdisc->running owner to get the lock more 3825 * often and dequeue packets faster. 3826 * On PREEMPT_RT it is possible to preempt the qdisc owner during xmit 3827 * and then other tasks will only enqueue packets. The packets will be 3828 * sent after the qdisc owner is scheduled again. To prevent this 3829 * scenario the task always serialize on the lock. 3830 */ 3831 contended = qdisc_is_running(q) || IS_ENABLED(CONFIG_PREEMPT_RT); 3832 if (unlikely(contended)) 3833 spin_lock(&q->busylock); 3834 3835 spin_lock(root_lock); 3836 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { 3837 __qdisc_drop(skb, &to_free); 3838 rc = NET_XMIT_DROP; 3839 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && 3840 qdisc_run_begin(q)) { 3841 /* 3842 * This is a work-conserving queue; there are no old skbs 3843 * waiting to be sent out; and the qdisc is not running - 3844 * xmit the skb directly. 3845 */ 3846 3847 qdisc_bstats_update(q, skb); 3848 3849 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { 3850 if (unlikely(contended)) { 3851 spin_unlock(&q->busylock); 3852 contended = false; 3853 } 3854 __qdisc_run(q); 3855 } 3856 3857 qdisc_run_end(q); 3858 rc = NET_XMIT_SUCCESS; 3859 } else { 3860 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3861 if (qdisc_run_begin(q)) { 3862 if (unlikely(contended)) { 3863 spin_unlock(&q->busylock); 3864 contended = false; 3865 } 3866 __qdisc_run(q); 3867 qdisc_run_end(q); 3868 } 3869 } 3870 spin_unlock(root_lock); 3871 if (unlikely(to_free)) 3872 kfree_skb_list_reason(to_free, SKB_DROP_REASON_QDISC_DROP); 3873 if (unlikely(contended)) 3874 spin_unlock(&q->busylock); 3875 return rc; 3876 } 3877 3878 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) 3879 static void skb_update_prio(struct sk_buff *skb) 3880 { 3881 const struct netprio_map *map; 3882 const struct sock *sk; 3883 unsigned int prioidx; 3884 3885 if (skb->priority) 3886 return; 3887 map = rcu_dereference_bh(skb->dev->priomap); 3888 if (!map) 3889 return; 3890 sk = skb_to_full_sk(skb); 3891 if (!sk) 3892 return; 3893 3894 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data); 3895 3896 if (prioidx < map->priomap_len) 3897 skb->priority = map->priomap[prioidx]; 3898 } 3899 #else 3900 #define skb_update_prio(skb) 3901 #endif 3902 3903 /** 3904 * dev_loopback_xmit - loop back @skb 3905 * @net: network namespace this loopback is happening in 3906 * @sk: sk needed to be a netfilter okfn 3907 * @skb: buffer to transmit 3908 */ 3909 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) 3910 { 3911 skb_reset_mac_header(skb); 3912 __skb_pull(skb, skb_network_offset(skb)); 3913 skb->pkt_type = PACKET_LOOPBACK; 3914 if (skb->ip_summed == CHECKSUM_NONE) 3915 skb->ip_summed = CHECKSUM_UNNECESSARY; 3916 DEBUG_NET_WARN_ON_ONCE(!skb_dst(skb)); 3917 skb_dst_force(skb); 3918 netif_rx(skb); 3919 return 0; 3920 } 3921 EXPORT_SYMBOL(dev_loopback_xmit); 3922 3923 #ifdef CONFIG_NET_EGRESS 3924 static struct netdev_queue * 3925 netdev_tx_queue_mapping(struct net_device *dev, struct sk_buff *skb) 3926 { 3927 int qm = skb_get_queue_mapping(skb); 3928 3929 return netdev_get_tx_queue(dev, netdev_cap_txqueue(dev, qm)); 3930 } 3931 3932 static bool netdev_xmit_txqueue_skipped(void) 3933 { 3934 return __this_cpu_read(softnet_data.xmit.skip_txqueue); 3935 } 3936 3937 void netdev_xmit_skip_txqueue(bool skip) 3938 { 3939 __this_cpu_write(softnet_data.xmit.skip_txqueue, skip); 3940 } 3941 EXPORT_SYMBOL_GPL(netdev_xmit_skip_txqueue); 3942 #endif /* CONFIG_NET_EGRESS */ 3943 3944 #ifdef CONFIG_NET_XGRESS 3945 static int tc_run(struct tcx_entry *entry, struct sk_buff *skb) 3946 { 3947 int ret = TC_ACT_UNSPEC; 3948 #ifdef CONFIG_NET_CLS_ACT 3949 struct mini_Qdisc *miniq = rcu_dereference_bh(entry->miniq); 3950 struct tcf_result res; 3951 3952 if (!miniq) 3953 return ret; 3954 3955 tc_skb_cb(skb)->mru = 0; 3956 tc_skb_cb(skb)->post_ct = false; 3957 3958 mini_qdisc_bstats_cpu_update(miniq, skb); 3959 ret = tcf_classify(skb, miniq->block, miniq->filter_list, &res, false); 3960 /* Only tcf related quirks below. */ 3961 switch (ret) { 3962 case TC_ACT_SHOT: 3963 mini_qdisc_qstats_cpu_drop(miniq); 3964 break; 3965 case TC_ACT_OK: 3966 case TC_ACT_RECLASSIFY: 3967 skb->tc_index = TC_H_MIN(res.classid); 3968 break; 3969 } 3970 #endif /* CONFIG_NET_CLS_ACT */ 3971 return ret; 3972 } 3973 3974 static DEFINE_STATIC_KEY_FALSE(tcx_needed_key); 3975 3976 void tcx_inc(void) 3977 { 3978 static_branch_inc(&tcx_needed_key); 3979 } 3980 3981 void tcx_dec(void) 3982 { 3983 static_branch_dec(&tcx_needed_key); 3984 } 3985 3986 static __always_inline enum tcx_action_base 3987 tcx_run(const struct bpf_mprog_entry *entry, struct sk_buff *skb, 3988 const bool needs_mac) 3989 { 3990 const struct bpf_mprog_fp *fp; 3991 const struct bpf_prog *prog; 3992 int ret = TCX_NEXT; 3993 3994 if (needs_mac) 3995 __skb_push(skb, skb->mac_len); 3996 bpf_mprog_foreach_prog(entry, fp, prog) { 3997 bpf_compute_data_pointers(skb); 3998 ret = bpf_prog_run(prog, skb); 3999 if (ret != TCX_NEXT) 4000 break; 4001 } 4002 if (needs_mac) 4003 __skb_pull(skb, skb->mac_len); 4004 return tcx_action_code(skb, ret); 4005 } 4006 4007 static __always_inline struct sk_buff * 4008 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 4009 struct net_device *orig_dev, bool *another) 4010 { 4011 struct bpf_mprog_entry *entry = rcu_dereference_bh(skb->dev->tcx_ingress); 4012 int sch_ret; 4013 4014 if (!entry) 4015 return skb; 4016 if (*pt_prev) { 4017 *ret = deliver_skb(skb, *pt_prev, orig_dev); 4018 *pt_prev = NULL; 4019 } 4020 4021 qdisc_skb_cb(skb)->pkt_len = skb->len; 4022 tcx_set_ingress(skb, true); 4023 4024 if (static_branch_unlikely(&tcx_needed_key)) { 4025 sch_ret = tcx_run(entry, skb, true); 4026 if (sch_ret != TC_ACT_UNSPEC) 4027 goto ingress_verdict; 4028 } 4029 sch_ret = tc_run(tcx_entry(entry), skb); 4030 ingress_verdict: 4031 switch (sch_ret) { 4032 case TC_ACT_REDIRECT: 4033 /* skb_mac_header check was done by BPF, so we can safely 4034 * push the L2 header back before redirecting to another 4035 * netdev. 4036 */ 4037 __skb_push(skb, skb->mac_len); 4038 if (skb_do_redirect(skb) == -EAGAIN) { 4039 __skb_pull(skb, skb->mac_len); 4040 *another = true; 4041 break; 4042 } 4043 *ret = NET_RX_SUCCESS; 4044 return NULL; 4045 case TC_ACT_SHOT: 4046 kfree_skb_reason(skb, SKB_DROP_REASON_TC_INGRESS); 4047 *ret = NET_RX_DROP; 4048 return NULL; 4049 /* used by tc_run */ 4050 case TC_ACT_STOLEN: 4051 case TC_ACT_QUEUED: 4052 case TC_ACT_TRAP: 4053 consume_skb(skb); 4054 fallthrough; 4055 case TC_ACT_CONSUMED: 4056 *ret = NET_RX_SUCCESS; 4057 return NULL; 4058 } 4059 4060 return skb; 4061 } 4062 4063 static __always_inline struct sk_buff * 4064 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 4065 { 4066 struct bpf_mprog_entry *entry = rcu_dereference_bh(dev->tcx_egress); 4067 int sch_ret; 4068 4069 if (!entry) 4070 return skb; 4071 4072 /* qdisc_skb_cb(skb)->pkt_len & tcx_set_ingress() was 4073 * already set by the caller. 4074 */ 4075 if (static_branch_unlikely(&tcx_needed_key)) { 4076 sch_ret = tcx_run(entry, skb, false); 4077 if (sch_ret != TC_ACT_UNSPEC) 4078 goto egress_verdict; 4079 } 4080 sch_ret = tc_run(tcx_entry(entry), skb); 4081 egress_verdict: 4082 switch (sch_ret) { 4083 case TC_ACT_REDIRECT: 4084 /* No need to push/pop skb's mac_header here on egress! */ 4085 skb_do_redirect(skb); 4086 *ret = NET_XMIT_SUCCESS; 4087 return NULL; 4088 case TC_ACT_SHOT: 4089 kfree_skb_reason(skb, SKB_DROP_REASON_TC_EGRESS); 4090 *ret = NET_XMIT_DROP; 4091 return NULL; 4092 /* used by tc_run */ 4093 case TC_ACT_STOLEN: 4094 case TC_ACT_QUEUED: 4095 case TC_ACT_TRAP: 4096 consume_skb(skb); 4097 fallthrough; 4098 case TC_ACT_CONSUMED: 4099 *ret = NET_XMIT_SUCCESS; 4100 return NULL; 4101 } 4102 4103 return skb; 4104 } 4105 #else 4106 static __always_inline struct sk_buff * 4107 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 4108 struct net_device *orig_dev, bool *another) 4109 { 4110 return skb; 4111 } 4112 4113 static __always_inline struct sk_buff * 4114 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 4115 { 4116 return skb; 4117 } 4118 #endif /* CONFIG_NET_XGRESS */ 4119 4120 #ifdef CONFIG_XPS 4121 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb, 4122 struct xps_dev_maps *dev_maps, unsigned int tci) 4123 { 4124 int tc = netdev_get_prio_tc_map(dev, skb->priority); 4125 struct xps_map *map; 4126 int queue_index = -1; 4127 4128 if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids) 4129 return queue_index; 4130 4131 tci *= dev_maps->num_tc; 4132 tci += tc; 4133 4134 map = rcu_dereference(dev_maps->attr_map[tci]); 4135 if (map) { 4136 if (map->len == 1) 4137 queue_index = map->queues[0]; 4138 else 4139 queue_index = map->queues[reciprocal_scale( 4140 skb_get_hash(skb), map->len)]; 4141 if (unlikely(queue_index >= dev->real_num_tx_queues)) 4142 queue_index = -1; 4143 } 4144 return queue_index; 4145 } 4146 #endif 4147 4148 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev, 4149 struct sk_buff *skb) 4150 { 4151 #ifdef CONFIG_XPS 4152 struct xps_dev_maps *dev_maps; 4153 struct sock *sk = skb->sk; 4154 int queue_index = -1; 4155 4156 if (!static_key_false(&xps_needed)) 4157 return -1; 4158 4159 rcu_read_lock(); 4160 if (!static_key_false(&xps_rxqs_needed)) 4161 goto get_cpus_map; 4162 4163 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]); 4164 if (dev_maps) { 4165 int tci = sk_rx_queue_get(sk); 4166 4167 if (tci >= 0) 4168 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4169 tci); 4170 } 4171 4172 get_cpus_map: 4173 if (queue_index < 0) { 4174 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]); 4175 if (dev_maps) { 4176 unsigned int tci = skb->sender_cpu - 1; 4177 4178 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4179 tci); 4180 } 4181 } 4182 rcu_read_unlock(); 4183 4184 return queue_index; 4185 #else 4186 return -1; 4187 #endif 4188 } 4189 4190 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb, 4191 struct net_device *sb_dev) 4192 { 4193 return 0; 4194 } 4195 EXPORT_SYMBOL(dev_pick_tx_zero); 4196 4197 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb, 4198 struct net_device *sb_dev) 4199 { 4200 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues; 4201 } 4202 EXPORT_SYMBOL(dev_pick_tx_cpu_id); 4203 4204 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb, 4205 struct net_device *sb_dev) 4206 { 4207 struct sock *sk = skb->sk; 4208 int queue_index = sk_tx_queue_get(sk); 4209 4210 sb_dev = sb_dev ? : dev; 4211 4212 if (queue_index < 0 || skb->ooo_okay || 4213 queue_index >= dev->real_num_tx_queues) { 4214 int new_index = get_xps_queue(dev, sb_dev, skb); 4215 4216 if (new_index < 0) 4217 new_index = skb_tx_hash(dev, sb_dev, skb); 4218 4219 if (queue_index != new_index && sk && 4220 sk_fullsock(sk) && 4221 rcu_access_pointer(sk->sk_dst_cache)) 4222 sk_tx_queue_set(sk, new_index); 4223 4224 queue_index = new_index; 4225 } 4226 4227 return queue_index; 4228 } 4229 EXPORT_SYMBOL(netdev_pick_tx); 4230 4231 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev, 4232 struct sk_buff *skb, 4233 struct net_device *sb_dev) 4234 { 4235 int queue_index = 0; 4236 4237 #ifdef CONFIG_XPS 4238 u32 sender_cpu = skb->sender_cpu - 1; 4239 4240 if (sender_cpu >= (u32)NR_CPUS) 4241 skb->sender_cpu = raw_smp_processor_id() + 1; 4242 #endif 4243 4244 if (dev->real_num_tx_queues != 1) { 4245 const struct net_device_ops *ops = dev->netdev_ops; 4246 4247 if (ops->ndo_select_queue) 4248 queue_index = ops->ndo_select_queue(dev, skb, sb_dev); 4249 else 4250 queue_index = netdev_pick_tx(dev, skb, sb_dev); 4251 4252 queue_index = netdev_cap_txqueue(dev, queue_index); 4253 } 4254 4255 skb_set_queue_mapping(skb, queue_index); 4256 return netdev_get_tx_queue(dev, queue_index); 4257 } 4258 4259 /** 4260 * __dev_queue_xmit() - transmit a buffer 4261 * @skb: buffer to transmit 4262 * @sb_dev: suboordinate device used for L2 forwarding offload 4263 * 4264 * Queue a buffer for transmission to a network device. The caller must 4265 * have set the device and priority and built the buffer before calling 4266 * this function. The function can be called from an interrupt. 4267 * 4268 * When calling this method, interrupts MUST be enabled. This is because 4269 * the BH enable code must have IRQs enabled so that it will not deadlock. 4270 * 4271 * Regardless of the return value, the skb is consumed, so it is currently 4272 * difficult to retry a send to this method. (You can bump the ref count 4273 * before sending to hold a reference for retry if you are careful.) 4274 * 4275 * Return: 4276 * * 0 - buffer successfully transmitted 4277 * * positive qdisc return code - NET_XMIT_DROP etc. 4278 * * negative errno - other errors 4279 */ 4280 int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev) 4281 { 4282 struct net_device *dev = skb->dev; 4283 struct netdev_queue *txq = NULL; 4284 struct Qdisc *q; 4285 int rc = -ENOMEM; 4286 bool again = false; 4287 4288 skb_reset_mac_header(skb); 4289 skb_assert_len(skb); 4290 4291 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) 4292 __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED); 4293 4294 /* Disable soft irqs for various locks below. Also 4295 * stops preemption for RCU. 4296 */ 4297 rcu_read_lock_bh(); 4298 4299 skb_update_prio(skb); 4300 4301 qdisc_pkt_len_init(skb); 4302 tcx_set_ingress(skb, false); 4303 #ifdef CONFIG_NET_EGRESS 4304 if (static_branch_unlikely(&egress_needed_key)) { 4305 if (nf_hook_egress_active()) { 4306 skb = nf_hook_egress(skb, &rc, dev); 4307 if (!skb) 4308 goto out; 4309 } 4310 4311 netdev_xmit_skip_txqueue(false); 4312 4313 nf_skip_egress(skb, true); 4314 skb = sch_handle_egress(skb, &rc, dev); 4315 if (!skb) 4316 goto out; 4317 nf_skip_egress(skb, false); 4318 4319 if (netdev_xmit_txqueue_skipped()) 4320 txq = netdev_tx_queue_mapping(dev, skb); 4321 } 4322 #endif 4323 /* If device/qdisc don't need skb->dst, release it right now while 4324 * its hot in this cpu cache. 4325 */ 4326 if (dev->priv_flags & IFF_XMIT_DST_RELEASE) 4327 skb_dst_drop(skb); 4328 else 4329 skb_dst_force(skb); 4330 4331 if (!txq) 4332 txq = netdev_core_pick_tx(dev, skb, sb_dev); 4333 4334 q = rcu_dereference_bh(txq->qdisc); 4335 4336 trace_net_dev_queue(skb); 4337 if (q->enqueue) { 4338 rc = __dev_xmit_skb(skb, q, dev, txq); 4339 goto out; 4340 } 4341 4342 /* The device has no queue. Common case for software devices: 4343 * loopback, all the sorts of tunnels... 4344 4345 * Really, it is unlikely that netif_tx_lock protection is necessary 4346 * here. (f.e. loopback and IP tunnels are clean ignoring statistics 4347 * counters.) 4348 * However, it is possible, that they rely on protection 4349 * made by us here. 4350 4351 * Check this and shot the lock. It is not prone from deadlocks. 4352 *Either shot noqueue qdisc, it is even simpler 8) 4353 */ 4354 if (dev->flags & IFF_UP) { 4355 int cpu = smp_processor_id(); /* ok because BHs are off */ 4356 4357 /* Other cpus might concurrently change txq->xmit_lock_owner 4358 * to -1 or to their cpu id, but not to our id. 4359 */ 4360 if (READ_ONCE(txq->xmit_lock_owner) != cpu) { 4361 if (dev_xmit_recursion()) 4362 goto recursion_alert; 4363 4364 skb = validate_xmit_skb(skb, dev, &again); 4365 if (!skb) 4366 goto out; 4367 4368 HARD_TX_LOCK(dev, txq, cpu); 4369 4370 if (!netif_xmit_stopped(txq)) { 4371 dev_xmit_recursion_inc(); 4372 skb = dev_hard_start_xmit(skb, dev, txq, &rc); 4373 dev_xmit_recursion_dec(); 4374 if (dev_xmit_complete(rc)) { 4375 HARD_TX_UNLOCK(dev, txq); 4376 goto out; 4377 } 4378 } 4379 HARD_TX_UNLOCK(dev, txq); 4380 net_crit_ratelimited("Virtual device %s asks to queue packet!\n", 4381 dev->name); 4382 } else { 4383 /* Recursion is detected! It is possible, 4384 * unfortunately 4385 */ 4386 recursion_alert: 4387 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", 4388 dev->name); 4389 } 4390 } 4391 4392 rc = -ENETDOWN; 4393 rcu_read_unlock_bh(); 4394 4395 dev_core_stats_tx_dropped_inc(dev); 4396 kfree_skb_list(skb); 4397 return rc; 4398 out: 4399 rcu_read_unlock_bh(); 4400 return rc; 4401 } 4402 EXPORT_SYMBOL(__dev_queue_xmit); 4403 4404 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id) 4405 { 4406 struct net_device *dev = skb->dev; 4407 struct sk_buff *orig_skb = skb; 4408 struct netdev_queue *txq; 4409 int ret = NETDEV_TX_BUSY; 4410 bool again = false; 4411 4412 if (unlikely(!netif_running(dev) || 4413 !netif_carrier_ok(dev))) 4414 goto drop; 4415 4416 skb = validate_xmit_skb_list(skb, dev, &again); 4417 if (skb != orig_skb) 4418 goto drop; 4419 4420 skb_set_queue_mapping(skb, queue_id); 4421 txq = skb_get_tx_queue(dev, skb); 4422 4423 local_bh_disable(); 4424 4425 dev_xmit_recursion_inc(); 4426 HARD_TX_LOCK(dev, txq, smp_processor_id()); 4427 if (!netif_xmit_frozen_or_drv_stopped(txq)) 4428 ret = netdev_start_xmit(skb, dev, txq, false); 4429 HARD_TX_UNLOCK(dev, txq); 4430 dev_xmit_recursion_dec(); 4431 4432 local_bh_enable(); 4433 return ret; 4434 drop: 4435 dev_core_stats_tx_dropped_inc(dev); 4436 kfree_skb_list(skb); 4437 return NET_XMIT_DROP; 4438 } 4439 EXPORT_SYMBOL(__dev_direct_xmit); 4440 4441 /************************************************************************* 4442 * Receiver routines 4443 *************************************************************************/ 4444 4445 int netdev_max_backlog __read_mostly = 1000; 4446 EXPORT_SYMBOL(netdev_max_backlog); 4447 4448 int netdev_tstamp_prequeue __read_mostly = 1; 4449 unsigned int sysctl_skb_defer_max __read_mostly = 64; 4450 int netdev_budget __read_mostly = 300; 4451 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */ 4452 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ; 4453 int weight_p __read_mostly = 64; /* old backlog weight */ 4454 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ 4455 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ 4456 int dev_rx_weight __read_mostly = 64; 4457 int dev_tx_weight __read_mostly = 64; 4458 4459 /* Called with irq disabled */ 4460 static inline void ____napi_schedule(struct softnet_data *sd, 4461 struct napi_struct *napi) 4462 { 4463 struct task_struct *thread; 4464 4465 lockdep_assert_irqs_disabled(); 4466 4467 if (test_bit(NAPI_STATE_THREADED, &napi->state)) { 4468 /* Paired with smp_mb__before_atomic() in 4469 * napi_enable()/dev_set_threaded(). 4470 * Use READ_ONCE() to guarantee a complete 4471 * read on napi->thread. Only call 4472 * wake_up_process() when it's not NULL. 4473 */ 4474 thread = READ_ONCE(napi->thread); 4475 if (thread) { 4476 /* Avoid doing set_bit() if the thread is in 4477 * INTERRUPTIBLE state, cause napi_thread_wait() 4478 * makes sure to proceed with napi polling 4479 * if the thread is explicitly woken from here. 4480 */ 4481 if (READ_ONCE(thread->__state) != TASK_INTERRUPTIBLE) 4482 set_bit(NAPI_STATE_SCHED_THREADED, &napi->state); 4483 wake_up_process(thread); 4484 return; 4485 } 4486 } 4487 4488 list_add_tail(&napi->poll_list, &sd->poll_list); 4489 WRITE_ONCE(napi->list_owner, smp_processor_id()); 4490 /* If not called from net_rx_action() 4491 * we have to raise NET_RX_SOFTIRQ. 4492 */ 4493 if (!sd->in_net_rx_action) 4494 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4495 } 4496 4497 #ifdef CONFIG_RPS 4498 4499 /* One global table that all flow-based protocols share. */ 4500 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; 4501 EXPORT_SYMBOL(rps_sock_flow_table); 4502 u32 rps_cpu_mask __read_mostly; 4503 EXPORT_SYMBOL(rps_cpu_mask); 4504 4505 struct static_key_false rps_needed __read_mostly; 4506 EXPORT_SYMBOL(rps_needed); 4507 struct static_key_false rfs_needed __read_mostly; 4508 EXPORT_SYMBOL(rfs_needed); 4509 4510 static struct rps_dev_flow * 4511 set_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4512 struct rps_dev_flow *rflow, u16 next_cpu) 4513 { 4514 if (next_cpu < nr_cpu_ids) { 4515 #ifdef CONFIG_RFS_ACCEL 4516 struct netdev_rx_queue *rxqueue; 4517 struct rps_dev_flow_table *flow_table; 4518 struct rps_dev_flow *old_rflow; 4519 u32 flow_id; 4520 u16 rxq_index; 4521 int rc; 4522 4523 /* Should we steer this flow to a different hardware queue? */ 4524 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || 4525 !(dev->features & NETIF_F_NTUPLE)) 4526 goto out; 4527 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); 4528 if (rxq_index == skb_get_rx_queue(skb)) 4529 goto out; 4530 4531 rxqueue = dev->_rx + rxq_index; 4532 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4533 if (!flow_table) 4534 goto out; 4535 flow_id = skb_get_hash(skb) & flow_table->mask; 4536 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, 4537 rxq_index, flow_id); 4538 if (rc < 0) 4539 goto out; 4540 old_rflow = rflow; 4541 rflow = &flow_table->flows[flow_id]; 4542 rflow->filter = rc; 4543 if (old_rflow->filter == rflow->filter) 4544 old_rflow->filter = RPS_NO_FILTER; 4545 out: 4546 #endif 4547 rflow->last_qtail = 4548 per_cpu(softnet_data, next_cpu).input_queue_head; 4549 } 4550 4551 rflow->cpu = next_cpu; 4552 return rflow; 4553 } 4554 4555 /* 4556 * get_rps_cpu is called from netif_receive_skb and returns the target 4557 * CPU from the RPS map of the receiving queue for a given skb. 4558 * rcu_read_lock must be held on entry. 4559 */ 4560 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4561 struct rps_dev_flow **rflowp) 4562 { 4563 const struct rps_sock_flow_table *sock_flow_table; 4564 struct netdev_rx_queue *rxqueue = dev->_rx; 4565 struct rps_dev_flow_table *flow_table; 4566 struct rps_map *map; 4567 int cpu = -1; 4568 u32 tcpu; 4569 u32 hash; 4570 4571 if (skb_rx_queue_recorded(skb)) { 4572 u16 index = skb_get_rx_queue(skb); 4573 4574 if (unlikely(index >= dev->real_num_rx_queues)) { 4575 WARN_ONCE(dev->real_num_rx_queues > 1, 4576 "%s received packet on queue %u, but number " 4577 "of RX queues is %u\n", 4578 dev->name, index, dev->real_num_rx_queues); 4579 goto done; 4580 } 4581 rxqueue += index; 4582 } 4583 4584 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ 4585 4586 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4587 map = rcu_dereference(rxqueue->rps_map); 4588 if (!flow_table && !map) 4589 goto done; 4590 4591 skb_reset_network_header(skb); 4592 hash = skb_get_hash(skb); 4593 if (!hash) 4594 goto done; 4595 4596 sock_flow_table = rcu_dereference(rps_sock_flow_table); 4597 if (flow_table && sock_flow_table) { 4598 struct rps_dev_flow *rflow; 4599 u32 next_cpu; 4600 u32 ident; 4601 4602 /* First check into global flow table if there is a match. 4603 * This READ_ONCE() pairs with WRITE_ONCE() from rps_record_sock_flow(). 4604 */ 4605 ident = READ_ONCE(sock_flow_table->ents[hash & sock_flow_table->mask]); 4606 if ((ident ^ hash) & ~rps_cpu_mask) 4607 goto try_rps; 4608 4609 next_cpu = ident & rps_cpu_mask; 4610 4611 /* OK, now we know there is a match, 4612 * we can look at the local (per receive queue) flow table 4613 */ 4614 rflow = &flow_table->flows[hash & flow_table->mask]; 4615 tcpu = rflow->cpu; 4616 4617 /* 4618 * If the desired CPU (where last recvmsg was done) is 4619 * different from current CPU (one in the rx-queue flow 4620 * table entry), switch if one of the following holds: 4621 * - Current CPU is unset (>= nr_cpu_ids). 4622 * - Current CPU is offline. 4623 * - The current CPU's queue tail has advanced beyond the 4624 * last packet that was enqueued using this table entry. 4625 * This guarantees that all previous packets for the flow 4626 * have been dequeued, thus preserving in order delivery. 4627 */ 4628 if (unlikely(tcpu != next_cpu) && 4629 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || 4630 ((int)(per_cpu(softnet_data, tcpu).input_queue_head - 4631 rflow->last_qtail)) >= 0)) { 4632 tcpu = next_cpu; 4633 rflow = set_rps_cpu(dev, skb, rflow, next_cpu); 4634 } 4635 4636 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { 4637 *rflowp = rflow; 4638 cpu = tcpu; 4639 goto done; 4640 } 4641 } 4642 4643 try_rps: 4644 4645 if (map) { 4646 tcpu = map->cpus[reciprocal_scale(hash, map->len)]; 4647 if (cpu_online(tcpu)) { 4648 cpu = tcpu; 4649 goto done; 4650 } 4651 } 4652 4653 done: 4654 return cpu; 4655 } 4656 4657 #ifdef CONFIG_RFS_ACCEL 4658 4659 /** 4660 * rps_may_expire_flow - check whether an RFS hardware filter may be removed 4661 * @dev: Device on which the filter was set 4662 * @rxq_index: RX queue index 4663 * @flow_id: Flow ID passed to ndo_rx_flow_steer() 4664 * @filter_id: Filter ID returned by ndo_rx_flow_steer() 4665 * 4666 * Drivers that implement ndo_rx_flow_steer() should periodically call 4667 * this function for each installed filter and remove the filters for 4668 * which it returns %true. 4669 */ 4670 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, 4671 u32 flow_id, u16 filter_id) 4672 { 4673 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; 4674 struct rps_dev_flow_table *flow_table; 4675 struct rps_dev_flow *rflow; 4676 bool expire = true; 4677 unsigned int cpu; 4678 4679 rcu_read_lock(); 4680 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4681 if (flow_table && flow_id <= flow_table->mask) { 4682 rflow = &flow_table->flows[flow_id]; 4683 cpu = READ_ONCE(rflow->cpu); 4684 if (rflow->filter == filter_id && cpu < nr_cpu_ids && 4685 ((int)(per_cpu(softnet_data, cpu).input_queue_head - 4686 rflow->last_qtail) < 4687 (int)(10 * flow_table->mask))) 4688 expire = false; 4689 } 4690 rcu_read_unlock(); 4691 return expire; 4692 } 4693 EXPORT_SYMBOL(rps_may_expire_flow); 4694 4695 #endif /* CONFIG_RFS_ACCEL */ 4696 4697 /* Called from hardirq (IPI) context */ 4698 static void rps_trigger_softirq(void *data) 4699 { 4700 struct softnet_data *sd = data; 4701 4702 ____napi_schedule(sd, &sd->backlog); 4703 sd->received_rps++; 4704 } 4705 4706 #endif /* CONFIG_RPS */ 4707 4708 /* Called from hardirq (IPI) context */ 4709 static void trigger_rx_softirq(void *data) 4710 { 4711 struct softnet_data *sd = data; 4712 4713 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4714 smp_store_release(&sd->defer_ipi_scheduled, 0); 4715 } 4716 4717 /* 4718 * After we queued a packet into sd->input_pkt_queue, 4719 * we need to make sure this queue is serviced soon. 4720 * 4721 * - If this is another cpu queue, link it to our rps_ipi_list, 4722 * and make sure we will process rps_ipi_list from net_rx_action(). 4723 * 4724 * - If this is our own queue, NAPI schedule our backlog. 4725 * Note that this also raises NET_RX_SOFTIRQ. 4726 */ 4727 static void napi_schedule_rps(struct softnet_data *sd) 4728 { 4729 struct softnet_data *mysd = this_cpu_ptr(&softnet_data); 4730 4731 #ifdef CONFIG_RPS 4732 if (sd != mysd) { 4733 sd->rps_ipi_next = mysd->rps_ipi_list; 4734 mysd->rps_ipi_list = sd; 4735 4736 /* If not called from net_rx_action() or napi_threaded_poll() 4737 * we have to raise NET_RX_SOFTIRQ. 4738 */ 4739 if (!mysd->in_net_rx_action && !mysd->in_napi_threaded_poll) 4740 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4741 return; 4742 } 4743 #endif /* CONFIG_RPS */ 4744 __napi_schedule_irqoff(&mysd->backlog); 4745 } 4746 4747 #ifdef CONFIG_NET_FLOW_LIMIT 4748 int netdev_flow_limit_table_len __read_mostly = (1 << 12); 4749 #endif 4750 4751 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) 4752 { 4753 #ifdef CONFIG_NET_FLOW_LIMIT 4754 struct sd_flow_limit *fl; 4755 struct softnet_data *sd; 4756 unsigned int old_flow, new_flow; 4757 4758 if (qlen < (READ_ONCE(netdev_max_backlog) >> 1)) 4759 return false; 4760 4761 sd = this_cpu_ptr(&softnet_data); 4762 4763 rcu_read_lock(); 4764 fl = rcu_dereference(sd->flow_limit); 4765 if (fl) { 4766 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); 4767 old_flow = fl->history[fl->history_head]; 4768 fl->history[fl->history_head] = new_flow; 4769 4770 fl->history_head++; 4771 fl->history_head &= FLOW_LIMIT_HISTORY - 1; 4772 4773 if (likely(fl->buckets[old_flow])) 4774 fl->buckets[old_flow]--; 4775 4776 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { 4777 fl->count++; 4778 rcu_read_unlock(); 4779 return true; 4780 } 4781 } 4782 rcu_read_unlock(); 4783 #endif 4784 return false; 4785 } 4786 4787 /* 4788 * enqueue_to_backlog is called to queue an skb to a per CPU backlog 4789 * queue (may be a remote CPU queue). 4790 */ 4791 static int enqueue_to_backlog(struct sk_buff *skb, int cpu, 4792 unsigned int *qtail) 4793 { 4794 enum skb_drop_reason reason; 4795 struct softnet_data *sd; 4796 unsigned long flags; 4797 unsigned int qlen; 4798 4799 reason = SKB_DROP_REASON_NOT_SPECIFIED; 4800 sd = &per_cpu(softnet_data, cpu); 4801 4802 rps_lock_irqsave(sd, &flags); 4803 if (!netif_running(skb->dev)) 4804 goto drop; 4805 qlen = skb_queue_len(&sd->input_pkt_queue); 4806 if (qlen <= READ_ONCE(netdev_max_backlog) && !skb_flow_limit(skb, qlen)) { 4807 if (qlen) { 4808 enqueue: 4809 __skb_queue_tail(&sd->input_pkt_queue, skb); 4810 input_queue_tail_incr_save(sd, qtail); 4811 rps_unlock_irq_restore(sd, &flags); 4812 return NET_RX_SUCCESS; 4813 } 4814 4815 /* Schedule NAPI for backlog device 4816 * We can use non atomic operation since we own the queue lock 4817 */ 4818 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) 4819 napi_schedule_rps(sd); 4820 goto enqueue; 4821 } 4822 reason = SKB_DROP_REASON_CPU_BACKLOG; 4823 4824 drop: 4825 sd->dropped++; 4826 rps_unlock_irq_restore(sd, &flags); 4827 4828 dev_core_stats_rx_dropped_inc(skb->dev); 4829 kfree_skb_reason(skb, reason); 4830 return NET_RX_DROP; 4831 } 4832 4833 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb) 4834 { 4835 struct net_device *dev = skb->dev; 4836 struct netdev_rx_queue *rxqueue; 4837 4838 rxqueue = dev->_rx; 4839 4840 if (skb_rx_queue_recorded(skb)) { 4841 u16 index = skb_get_rx_queue(skb); 4842 4843 if (unlikely(index >= dev->real_num_rx_queues)) { 4844 WARN_ONCE(dev->real_num_rx_queues > 1, 4845 "%s received packet on queue %u, but number " 4846 "of RX queues is %u\n", 4847 dev->name, index, dev->real_num_rx_queues); 4848 4849 return rxqueue; /* Return first rxqueue */ 4850 } 4851 rxqueue += index; 4852 } 4853 return rxqueue; 4854 } 4855 4856 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp, 4857 struct bpf_prog *xdp_prog) 4858 { 4859 void *orig_data, *orig_data_end, *hard_start; 4860 struct netdev_rx_queue *rxqueue; 4861 bool orig_bcast, orig_host; 4862 u32 mac_len, frame_sz; 4863 __be16 orig_eth_type; 4864 struct ethhdr *eth; 4865 u32 metalen, act; 4866 int off; 4867 4868 /* The XDP program wants to see the packet starting at the MAC 4869 * header. 4870 */ 4871 mac_len = skb->data - skb_mac_header(skb); 4872 hard_start = skb->data - skb_headroom(skb); 4873 4874 /* SKB "head" area always have tailroom for skb_shared_info */ 4875 frame_sz = (void *)skb_end_pointer(skb) - hard_start; 4876 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 4877 4878 rxqueue = netif_get_rxqueue(skb); 4879 xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq); 4880 xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len, 4881 skb_headlen(skb) + mac_len, true); 4882 4883 orig_data_end = xdp->data_end; 4884 orig_data = xdp->data; 4885 eth = (struct ethhdr *)xdp->data; 4886 orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr); 4887 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest); 4888 orig_eth_type = eth->h_proto; 4889 4890 act = bpf_prog_run_xdp(xdp_prog, xdp); 4891 4892 /* check if bpf_xdp_adjust_head was used */ 4893 off = xdp->data - orig_data; 4894 if (off) { 4895 if (off > 0) 4896 __skb_pull(skb, off); 4897 else if (off < 0) 4898 __skb_push(skb, -off); 4899 4900 skb->mac_header += off; 4901 skb_reset_network_header(skb); 4902 } 4903 4904 /* check if bpf_xdp_adjust_tail was used */ 4905 off = xdp->data_end - orig_data_end; 4906 if (off != 0) { 4907 skb_set_tail_pointer(skb, xdp->data_end - xdp->data); 4908 skb->len += off; /* positive on grow, negative on shrink */ 4909 } 4910 4911 /* check if XDP changed eth hdr such SKB needs update */ 4912 eth = (struct ethhdr *)xdp->data; 4913 if ((orig_eth_type != eth->h_proto) || 4914 (orig_host != ether_addr_equal_64bits(eth->h_dest, 4915 skb->dev->dev_addr)) || 4916 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) { 4917 __skb_push(skb, ETH_HLEN); 4918 skb->pkt_type = PACKET_HOST; 4919 skb->protocol = eth_type_trans(skb, skb->dev); 4920 } 4921 4922 /* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull 4923 * before calling us again on redirect path. We do not call do_redirect 4924 * as we leave that up to the caller. 4925 * 4926 * Caller is responsible for managing lifetime of skb (i.e. calling 4927 * kfree_skb in response to actions it cannot handle/XDP_DROP). 4928 */ 4929 switch (act) { 4930 case XDP_REDIRECT: 4931 case XDP_TX: 4932 __skb_push(skb, mac_len); 4933 break; 4934 case XDP_PASS: 4935 metalen = xdp->data - xdp->data_meta; 4936 if (metalen) 4937 skb_metadata_set(skb, metalen); 4938 break; 4939 } 4940 4941 return act; 4942 } 4943 4944 static u32 netif_receive_generic_xdp(struct sk_buff *skb, 4945 struct xdp_buff *xdp, 4946 struct bpf_prog *xdp_prog) 4947 { 4948 u32 act = XDP_DROP; 4949 4950 /* Reinjected packets coming from act_mirred or similar should 4951 * not get XDP generic processing. 4952 */ 4953 if (skb_is_redirected(skb)) 4954 return XDP_PASS; 4955 4956 /* XDP packets must be linear and must have sufficient headroom 4957 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also 4958 * native XDP provides, thus we need to do it here as well. 4959 */ 4960 if (skb_cloned(skb) || skb_is_nonlinear(skb) || 4961 skb_headroom(skb) < XDP_PACKET_HEADROOM) { 4962 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb); 4963 int troom = skb->tail + skb->data_len - skb->end; 4964 4965 /* In case we have to go down the path and also linearize, 4966 * then lets do the pskb_expand_head() work just once here. 4967 */ 4968 if (pskb_expand_head(skb, 4969 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0, 4970 troom > 0 ? troom + 128 : 0, GFP_ATOMIC)) 4971 goto do_drop; 4972 if (skb_linearize(skb)) 4973 goto do_drop; 4974 } 4975 4976 act = bpf_prog_run_generic_xdp(skb, xdp, xdp_prog); 4977 switch (act) { 4978 case XDP_REDIRECT: 4979 case XDP_TX: 4980 case XDP_PASS: 4981 break; 4982 default: 4983 bpf_warn_invalid_xdp_action(skb->dev, xdp_prog, act); 4984 fallthrough; 4985 case XDP_ABORTED: 4986 trace_xdp_exception(skb->dev, xdp_prog, act); 4987 fallthrough; 4988 case XDP_DROP: 4989 do_drop: 4990 kfree_skb(skb); 4991 break; 4992 } 4993 4994 return act; 4995 } 4996 4997 /* When doing generic XDP we have to bypass the qdisc layer and the 4998 * network taps in order to match in-driver-XDP behavior. This also means 4999 * that XDP packets are able to starve other packets going through a qdisc, 5000 * and DDOS attacks will be more effective. In-driver-XDP use dedicated TX 5001 * queues, so they do not have this starvation issue. 5002 */ 5003 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog) 5004 { 5005 struct net_device *dev = skb->dev; 5006 struct netdev_queue *txq; 5007 bool free_skb = true; 5008 int cpu, rc; 5009 5010 txq = netdev_core_pick_tx(dev, skb, NULL); 5011 cpu = smp_processor_id(); 5012 HARD_TX_LOCK(dev, txq, cpu); 5013 if (!netif_xmit_frozen_or_drv_stopped(txq)) { 5014 rc = netdev_start_xmit(skb, dev, txq, 0); 5015 if (dev_xmit_complete(rc)) 5016 free_skb = false; 5017 } 5018 HARD_TX_UNLOCK(dev, txq); 5019 if (free_skb) { 5020 trace_xdp_exception(dev, xdp_prog, XDP_TX); 5021 dev_core_stats_tx_dropped_inc(dev); 5022 kfree_skb(skb); 5023 } 5024 } 5025 5026 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key); 5027 5028 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb) 5029 { 5030 if (xdp_prog) { 5031 struct xdp_buff xdp; 5032 u32 act; 5033 int err; 5034 5035 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog); 5036 if (act != XDP_PASS) { 5037 switch (act) { 5038 case XDP_REDIRECT: 5039 err = xdp_do_generic_redirect(skb->dev, skb, 5040 &xdp, xdp_prog); 5041 if (err) 5042 goto out_redir; 5043 break; 5044 case XDP_TX: 5045 generic_xdp_tx(skb, xdp_prog); 5046 break; 5047 } 5048 return XDP_DROP; 5049 } 5050 } 5051 return XDP_PASS; 5052 out_redir: 5053 kfree_skb_reason(skb, SKB_DROP_REASON_XDP); 5054 return XDP_DROP; 5055 } 5056 EXPORT_SYMBOL_GPL(do_xdp_generic); 5057 5058 static int netif_rx_internal(struct sk_buff *skb) 5059 { 5060 int ret; 5061 5062 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb); 5063 5064 trace_netif_rx(skb); 5065 5066 #ifdef CONFIG_RPS 5067 if (static_branch_unlikely(&rps_needed)) { 5068 struct rps_dev_flow voidflow, *rflow = &voidflow; 5069 int cpu; 5070 5071 rcu_read_lock(); 5072 5073 cpu = get_rps_cpu(skb->dev, skb, &rflow); 5074 if (cpu < 0) 5075 cpu = smp_processor_id(); 5076 5077 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5078 5079 rcu_read_unlock(); 5080 } else 5081 #endif 5082 { 5083 unsigned int qtail; 5084 5085 ret = enqueue_to_backlog(skb, smp_processor_id(), &qtail); 5086 } 5087 return ret; 5088 } 5089 5090 /** 5091 * __netif_rx - Slightly optimized version of netif_rx 5092 * @skb: buffer to post 5093 * 5094 * This behaves as netif_rx except that it does not disable bottom halves. 5095 * As a result this function may only be invoked from the interrupt context 5096 * (either hard or soft interrupt). 5097 */ 5098 int __netif_rx(struct sk_buff *skb) 5099 { 5100 int ret; 5101 5102 lockdep_assert_once(hardirq_count() | softirq_count()); 5103 5104 trace_netif_rx_entry(skb); 5105 ret = netif_rx_internal(skb); 5106 trace_netif_rx_exit(ret); 5107 return ret; 5108 } 5109 EXPORT_SYMBOL(__netif_rx); 5110 5111 /** 5112 * netif_rx - post buffer to the network code 5113 * @skb: buffer to post 5114 * 5115 * This function receives a packet from a device driver and queues it for 5116 * the upper (protocol) levels to process via the backlog NAPI device. It 5117 * always succeeds. The buffer may be dropped during processing for 5118 * congestion control or by the protocol layers. 5119 * The network buffer is passed via the backlog NAPI device. Modern NIC 5120 * driver should use NAPI and GRO. 5121 * This function can used from interrupt and from process context. The 5122 * caller from process context must not disable interrupts before invoking 5123 * this function. 5124 * 5125 * return values: 5126 * NET_RX_SUCCESS (no congestion) 5127 * NET_RX_DROP (packet was dropped) 5128 * 5129 */ 5130 int netif_rx(struct sk_buff *skb) 5131 { 5132 bool need_bh_off = !(hardirq_count() | softirq_count()); 5133 int ret; 5134 5135 if (need_bh_off) 5136 local_bh_disable(); 5137 trace_netif_rx_entry(skb); 5138 ret = netif_rx_internal(skb); 5139 trace_netif_rx_exit(ret); 5140 if (need_bh_off) 5141 local_bh_enable(); 5142 return ret; 5143 } 5144 EXPORT_SYMBOL(netif_rx); 5145 5146 static __latent_entropy void net_tx_action(struct softirq_action *h) 5147 { 5148 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 5149 5150 if (sd->completion_queue) { 5151 struct sk_buff *clist; 5152 5153 local_irq_disable(); 5154 clist = sd->completion_queue; 5155 sd->completion_queue = NULL; 5156 local_irq_enable(); 5157 5158 while (clist) { 5159 struct sk_buff *skb = clist; 5160 5161 clist = clist->next; 5162 5163 WARN_ON(refcount_read(&skb->users)); 5164 if (likely(get_kfree_skb_cb(skb)->reason == SKB_CONSUMED)) 5165 trace_consume_skb(skb, net_tx_action); 5166 else 5167 trace_kfree_skb(skb, net_tx_action, 5168 get_kfree_skb_cb(skb)->reason); 5169 5170 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) 5171 __kfree_skb(skb); 5172 else 5173 __napi_kfree_skb(skb, 5174 get_kfree_skb_cb(skb)->reason); 5175 } 5176 } 5177 5178 if (sd->output_queue) { 5179 struct Qdisc *head; 5180 5181 local_irq_disable(); 5182 head = sd->output_queue; 5183 sd->output_queue = NULL; 5184 sd->output_queue_tailp = &sd->output_queue; 5185 local_irq_enable(); 5186 5187 rcu_read_lock(); 5188 5189 while (head) { 5190 struct Qdisc *q = head; 5191 spinlock_t *root_lock = NULL; 5192 5193 head = head->next_sched; 5194 5195 /* We need to make sure head->next_sched is read 5196 * before clearing __QDISC_STATE_SCHED 5197 */ 5198 smp_mb__before_atomic(); 5199 5200 if (!(q->flags & TCQ_F_NOLOCK)) { 5201 root_lock = qdisc_lock(q); 5202 spin_lock(root_lock); 5203 } else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, 5204 &q->state))) { 5205 /* There is a synchronize_net() between 5206 * STATE_DEACTIVATED flag being set and 5207 * qdisc_reset()/some_qdisc_is_busy() in 5208 * dev_deactivate(), so we can safely bail out 5209 * early here to avoid data race between 5210 * qdisc_deactivate() and some_qdisc_is_busy() 5211 * for lockless qdisc. 5212 */ 5213 clear_bit(__QDISC_STATE_SCHED, &q->state); 5214 continue; 5215 } 5216 5217 clear_bit(__QDISC_STATE_SCHED, &q->state); 5218 qdisc_run(q); 5219 if (root_lock) 5220 spin_unlock(root_lock); 5221 } 5222 5223 rcu_read_unlock(); 5224 } 5225 5226 xfrm_dev_backlog(sd); 5227 } 5228 5229 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) 5230 /* This hook is defined here for ATM LANE */ 5231 int (*br_fdb_test_addr_hook)(struct net_device *dev, 5232 unsigned char *addr) __read_mostly; 5233 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); 5234 #endif 5235 5236 /** 5237 * netdev_is_rx_handler_busy - check if receive handler is registered 5238 * @dev: device to check 5239 * 5240 * Check if a receive handler is already registered for a given device. 5241 * Return true if there one. 5242 * 5243 * The caller must hold the rtnl_mutex. 5244 */ 5245 bool netdev_is_rx_handler_busy(struct net_device *dev) 5246 { 5247 ASSERT_RTNL(); 5248 return dev && rtnl_dereference(dev->rx_handler); 5249 } 5250 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); 5251 5252 /** 5253 * netdev_rx_handler_register - register receive handler 5254 * @dev: device to register a handler for 5255 * @rx_handler: receive handler to register 5256 * @rx_handler_data: data pointer that is used by rx handler 5257 * 5258 * Register a receive handler for a device. This handler will then be 5259 * called from __netif_receive_skb. A negative errno code is returned 5260 * on a failure. 5261 * 5262 * The caller must hold the rtnl_mutex. 5263 * 5264 * For a general description of rx_handler, see enum rx_handler_result. 5265 */ 5266 int netdev_rx_handler_register(struct net_device *dev, 5267 rx_handler_func_t *rx_handler, 5268 void *rx_handler_data) 5269 { 5270 if (netdev_is_rx_handler_busy(dev)) 5271 return -EBUSY; 5272 5273 if (dev->priv_flags & IFF_NO_RX_HANDLER) 5274 return -EINVAL; 5275 5276 /* Note: rx_handler_data must be set before rx_handler */ 5277 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); 5278 rcu_assign_pointer(dev->rx_handler, rx_handler); 5279 5280 return 0; 5281 } 5282 EXPORT_SYMBOL_GPL(netdev_rx_handler_register); 5283 5284 /** 5285 * netdev_rx_handler_unregister - unregister receive handler 5286 * @dev: device to unregister a handler from 5287 * 5288 * Unregister a receive handler from a device. 5289 * 5290 * The caller must hold the rtnl_mutex. 5291 */ 5292 void netdev_rx_handler_unregister(struct net_device *dev) 5293 { 5294 5295 ASSERT_RTNL(); 5296 RCU_INIT_POINTER(dev->rx_handler, NULL); 5297 /* a reader seeing a non NULL rx_handler in a rcu_read_lock() 5298 * section has a guarantee to see a non NULL rx_handler_data 5299 * as well. 5300 */ 5301 synchronize_net(); 5302 RCU_INIT_POINTER(dev->rx_handler_data, NULL); 5303 } 5304 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); 5305 5306 /* 5307 * Limit the use of PFMEMALLOC reserves to those protocols that implement 5308 * the special handling of PFMEMALLOC skbs. 5309 */ 5310 static bool skb_pfmemalloc_protocol(struct sk_buff *skb) 5311 { 5312 switch (skb->protocol) { 5313 case htons(ETH_P_ARP): 5314 case htons(ETH_P_IP): 5315 case htons(ETH_P_IPV6): 5316 case htons(ETH_P_8021Q): 5317 case htons(ETH_P_8021AD): 5318 return true; 5319 default: 5320 return false; 5321 } 5322 } 5323 5324 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, 5325 int *ret, struct net_device *orig_dev) 5326 { 5327 if (nf_hook_ingress_active(skb)) { 5328 int ingress_retval; 5329 5330 if (*pt_prev) { 5331 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5332 *pt_prev = NULL; 5333 } 5334 5335 rcu_read_lock(); 5336 ingress_retval = nf_hook_ingress(skb); 5337 rcu_read_unlock(); 5338 return ingress_retval; 5339 } 5340 return 0; 5341 } 5342 5343 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc, 5344 struct packet_type **ppt_prev) 5345 { 5346 struct packet_type *ptype, *pt_prev; 5347 rx_handler_func_t *rx_handler; 5348 struct sk_buff *skb = *pskb; 5349 struct net_device *orig_dev; 5350 bool deliver_exact = false; 5351 int ret = NET_RX_DROP; 5352 __be16 type; 5353 5354 net_timestamp_check(!READ_ONCE(netdev_tstamp_prequeue), skb); 5355 5356 trace_netif_receive_skb(skb); 5357 5358 orig_dev = skb->dev; 5359 5360 skb_reset_network_header(skb); 5361 if (!skb_transport_header_was_set(skb)) 5362 skb_reset_transport_header(skb); 5363 skb_reset_mac_len(skb); 5364 5365 pt_prev = NULL; 5366 5367 another_round: 5368 skb->skb_iif = skb->dev->ifindex; 5369 5370 __this_cpu_inc(softnet_data.processed); 5371 5372 if (static_branch_unlikely(&generic_xdp_needed_key)) { 5373 int ret2; 5374 5375 migrate_disable(); 5376 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb); 5377 migrate_enable(); 5378 5379 if (ret2 != XDP_PASS) { 5380 ret = NET_RX_DROP; 5381 goto out; 5382 } 5383 } 5384 5385 if (eth_type_vlan(skb->protocol)) { 5386 skb = skb_vlan_untag(skb); 5387 if (unlikely(!skb)) 5388 goto out; 5389 } 5390 5391 if (skb_skip_tc_classify(skb)) 5392 goto skip_classify; 5393 5394 if (pfmemalloc) 5395 goto skip_taps; 5396 5397 list_for_each_entry_rcu(ptype, &ptype_all, list) { 5398 if (pt_prev) 5399 ret = deliver_skb(skb, pt_prev, orig_dev); 5400 pt_prev = ptype; 5401 } 5402 5403 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { 5404 if (pt_prev) 5405 ret = deliver_skb(skb, pt_prev, orig_dev); 5406 pt_prev = ptype; 5407 } 5408 5409 skip_taps: 5410 #ifdef CONFIG_NET_INGRESS 5411 if (static_branch_unlikely(&ingress_needed_key)) { 5412 bool another = false; 5413 5414 nf_skip_egress(skb, true); 5415 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev, 5416 &another); 5417 if (another) 5418 goto another_round; 5419 if (!skb) 5420 goto out; 5421 5422 nf_skip_egress(skb, false); 5423 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) 5424 goto out; 5425 } 5426 #endif 5427 skb_reset_redirect(skb); 5428 skip_classify: 5429 if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) 5430 goto drop; 5431 5432 if (skb_vlan_tag_present(skb)) { 5433 if (pt_prev) { 5434 ret = deliver_skb(skb, pt_prev, orig_dev); 5435 pt_prev = NULL; 5436 } 5437 if (vlan_do_receive(&skb)) 5438 goto another_round; 5439 else if (unlikely(!skb)) 5440 goto out; 5441 } 5442 5443 rx_handler = rcu_dereference(skb->dev->rx_handler); 5444 if (rx_handler) { 5445 if (pt_prev) { 5446 ret = deliver_skb(skb, pt_prev, orig_dev); 5447 pt_prev = NULL; 5448 } 5449 switch (rx_handler(&skb)) { 5450 case RX_HANDLER_CONSUMED: 5451 ret = NET_RX_SUCCESS; 5452 goto out; 5453 case RX_HANDLER_ANOTHER: 5454 goto another_round; 5455 case RX_HANDLER_EXACT: 5456 deliver_exact = true; 5457 break; 5458 case RX_HANDLER_PASS: 5459 break; 5460 default: 5461 BUG(); 5462 } 5463 } 5464 5465 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) { 5466 check_vlan_id: 5467 if (skb_vlan_tag_get_id(skb)) { 5468 /* Vlan id is non 0 and vlan_do_receive() above couldn't 5469 * find vlan device. 5470 */ 5471 skb->pkt_type = PACKET_OTHERHOST; 5472 } else if (eth_type_vlan(skb->protocol)) { 5473 /* Outer header is 802.1P with vlan 0, inner header is 5474 * 802.1Q or 802.1AD and vlan_do_receive() above could 5475 * not find vlan dev for vlan id 0. 5476 */ 5477 __vlan_hwaccel_clear_tag(skb); 5478 skb = skb_vlan_untag(skb); 5479 if (unlikely(!skb)) 5480 goto out; 5481 if (vlan_do_receive(&skb)) 5482 /* After stripping off 802.1P header with vlan 0 5483 * vlan dev is found for inner header. 5484 */ 5485 goto another_round; 5486 else if (unlikely(!skb)) 5487 goto out; 5488 else 5489 /* We have stripped outer 802.1P vlan 0 header. 5490 * But could not find vlan dev. 5491 * check again for vlan id to set OTHERHOST. 5492 */ 5493 goto check_vlan_id; 5494 } 5495 /* Note: we might in the future use prio bits 5496 * and set skb->priority like in vlan_do_receive() 5497 * For the time being, just ignore Priority Code Point 5498 */ 5499 __vlan_hwaccel_clear_tag(skb); 5500 } 5501 5502 type = skb->protocol; 5503 5504 /* deliver only exact match when indicated */ 5505 if (likely(!deliver_exact)) { 5506 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5507 &ptype_base[ntohs(type) & 5508 PTYPE_HASH_MASK]); 5509 } 5510 5511 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5512 &orig_dev->ptype_specific); 5513 5514 if (unlikely(skb->dev != orig_dev)) { 5515 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5516 &skb->dev->ptype_specific); 5517 } 5518 5519 if (pt_prev) { 5520 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 5521 goto drop; 5522 *ppt_prev = pt_prev; 5523 } else { 5524 drop: 5525 if (!deliver_exact) 5526 dev_core_stats_rx_dropped_inc(skb->dev); 5527 else 5528 dev_core_stats_rx_nohandler_inc(skb->dev); 5529 kfree_skb_reason(skb, SKB_DROP_REASON_UNHANDLED_PROTO); 5530 /* Jamal, now you will not able to escape explaining 5531 * me how you were going to use this. :-) 5532 */ 5533 ret = NET_RX_DROP; 5534 } 5535 5536 out: 5537 /* The invariant here is that if *ppt_prev is not NULL 5538 * then skb should also be non-NULL. 5539 * 5540 * Apparently *ppt_prev assignment above holds this invariant due to 5541 * skb dereferencing near it. 5542 */ 5543 *pskb = skb; 5544 return ret; 5545 } 5546 5547 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc) 5548 { 5549 struct net_device *orig_dev = skb->dev; 5550 struct packet_type *pt_prev = NULL; 5551 int ret; 5552 5553 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5554 if (pt_prev) 5555 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb, 5556 skb->dev, pt_prev, orig_dev); 5557 return ret; 5558 } 5559 5560 /** 5561 * netif_receive_skb_core - special purpose version of netif_receive_skb 5562 * @skb: buffer to process 5563 * 5564 * More direct receive version of netif_receive_skb(). It should 5565 * only be used by callers that have a need to skip RPS and Generic XDP. 5566 * Caller must also take care of handling if ``(page_is_)pfmemalloc``. 5567 * 5568 * This function may only be called from softirq context and interrupts 5569 * should be enabled. 5570 * 5571 * Return values (usually ignored): 5572 * NET_RX_SUCCESS: no congestion 5573 * NET_RX_DROP: packet was dropped 5574 */ 5575 int netif_receive_skb_core(struct sk_buff *skb) 5576 { 5577 int ret; 5578 5579 rcu_read_lock(); 5580 ret = __netif_receive_skb_one_core(skb, false); 5581 rcu_read_unlock(); 5582 5583 return ret; 5584 } 5585 EXPORT_SYMBOL(netif_receive_skb_core); 5586 5587 static inline void __netif_receive_skb_list_ptype(struct list_head *head, 5588 struct packet_type *pt_prev, 5589 struct net_device *orig_dev) 5590 { 5591 struct sk_buff *skb, *next; 5592 5593 if (!pt_prev) 5594 return; 5595 if (list_empty(head)) 5596 return; 5597 if (pt_prev->list_func != NULL) 5598 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv, 5599 ip_list_rcv, head, pt_prev, orig_dev); 5600 else 5601 list_for_each_entry_safe(skb, next, head, list) { 5602 skb_list_del_init(skb); 5603 pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 5604 } 5605 } 5606 5607 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc) 5608 { 5609 /* Fast-path assumptions: 5610 * - There is no RX handler. 5611 * - Only one packet_type matches. 5612 * If either of these fails, we will end up doing some per-packet 5613 * processing in-line, then handling the 'last ptype' for the whole 5614 * sublist. This can't cause out-of-order delivery to any single ptype, 5615 * because the 'last ptype' must be constant across the sublist, and all 5616 * other ptypes are handled per-packet. 5617 */ 5618 /* Current (common) ptype of sublist */ 5619 struct packet_type *pt_curr = NULL; 5620 /* Current (common) orig_dev of sublist */ 5621 struct net_device *od_curr = NULL; 5622 struct list_head sublist; 5623 struct sk_buff *skb, *next; 5624 5625 INIT_LIST_HEAD(&sublist); 5626 list_for_each_entry_safe(skb, next, head, list) { 5627 struct net_device *orig_dev = skb->dev; 5628 struct packet_type *pt_prev = NULL; 5629 5630 skb_list_del_init(skb); 5631 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5632 if (!pt_prev) 5633 continue; 5634 if (pt_curr != pt_prev || od_curr != orig_dev) { 5635 /* dispatch old sublist */ 5636 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5637 /* start new sublist */ 5638 INIT_LIST_HEAD(&sublist); 5639 pt_curr = pt_prev; 5640 od_curr = orig_dev; 5641 } 5642 list_add_tail(&skb->list, &sublist); 5643 } 5644 5645 /* dispatch final sublist */ 5646 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5647 } 5648 5649 static int __netif_receive_skb(struct sk_buff *skb) 5650 { 5651 int ret; 5652 5653 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { 5654 unsigned int noreclaim_flag; 5655 5656 /* 5657 * PFMEMALLOC skbs are special, they should 5658 * - be delivered to SOCK_MEMALLOC sockets only 5659 * - stay away from userspace 5660 * - have bounded memory usage 5661 * 5662 * Use PF_MEMALLOC as this saves us from propagating the allocation 5663 * context down to all allocation sites. 5664 */ 5665 noreclaim_flag = memalloc_noreclaim_save(); 5666 ret = __netif_receive_skb_one_core(skb, true); 5667 memalloc_noreclaim_restore(noreclaim_flag); 5668 } else 5669 ret = __netif_receive_skb_one_core(skb, false); 5670 5671 return ret; 5672 } 5673 5674 static void __netif_receive_skb_list(struct list_head *head) 5675 { 5676 unsigned long noreclaim_flag = 0; 5677 struct sk_buff *skb, *next; 5678 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */ 5679 5680 list_for_each_entry_safe(skb, next, head, list) { 5681 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) { 5682 struct list_head sublist; 5683 5684 /* Handle the previous sublist */ 5685 list_cut_before(&sublist, head, &skb->list); 5686 if (!list_empty(&sublist)) 5687 __netif_receive_skb_list_core(&sublist, pfmemalloc); 5688 pfmemalloc = !pfmemalloc; 5689 /* See comments in __netif_receive_skb */ 5690 if (pfmemalloc) 5691 noreclaim_flag = memalloc_noreclaim_save(); 5692 else 5693 memalloc_noreclaim_restore(noreclaim_flag); 5694 } 5695 } 5696 /* Handle the remaining sublist */ 5697 if (!list_empty(head)) 5698 __netif_receive_skb_list_core(head, pfmemalloc); 5699 /* Restore pflags */ 5700 if (pfmemalloc) 5701 memalloc_noreclaim_restore(noreclaim_flag); 5702 } 5703 5704 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp) 5705 { 5706 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog); 5707 struct bpf_prog *new = xdp->prog; 5708 int ret = 0; 5709 5710 switch (xdp->command) { 5711 case XDP_SETUP_PROG: 5712 rcu_assign_pointer(dev->xdp_prog, new); 5713 if (old) 5714 bpf_prog_put(old); 5715 5716 if (old && !new) { 5717 static_branch_dec(&generic_xdp_needed_key); 5718 } else if (new && !old) { 5719 static_branch_inc(&generic_xdp_needed_key); 5720 dev_disable_lro(dev); 5721 dev_disable_gro_hw(dev); 5722 } 5723 break; 5724 5725 default: 5726 ret = -EINVAL; 5727 break; 5728 } 5729 5730 return ret; 5731 } 5732 5733 static int netif_receive_skb_internal(struct sk_buff *skb) 5734 { 5735 int ret; 5736 5737 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb); 5738 5739 if (skb_defer_rx_timestamp(skb)) 5740 return NET_RX_SUCCESS; 5741 5742 rcu_read_lock(); 5743 #ifdef CONFIG_RPS 5744 if (static_branch_unlikely(&rps_needed)) { 5745 struct rps_dev_flow voidflow, *rflow = &voidflow; 5746 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5747 5748 if (cpu >= 0) { 5749 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5750 rcu_read_unlock(); 5751 return ret; 5752 } 5753 } 5754 #endif 5755 ret = __netif_receive_skb(skb); 5756 rcu_read_unlock(); 5757 return ret; 5758 } 5759 5760 void netif_receive_skb_list_internal(struct list_head *head) 5761 { 5762 struct sk_buff *skb, *next; 5763 struct list_head sublist; 5764 5765 INIT_LIST_HEAD(&sublist); 5766 list_for_each_entry_safe(skb, next, head, list) { 5767 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb); 5768 skb_list_del_init(skb); 5769 if (!skb_defer_rx_timestamp(skb)) 5770 list_add_tail(&skb->list, &sublist); 5771 } 5772 list_splice_init(&sublist, head); 5773 5774 rcu_read_lock(); 5775 #ifdef CONFIG_RPS 5776 if (static_branch_unlikely(&rps_needed)) { 5777 list_for_each_entry_safe(skb, next, head, list) { 5778 struct rps_dev_flow voidflow, *rflow = &voidflow; 5779 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5780 5781 if (cpu >= 0) { 5782 /* Will be handled, remove from list */ 5783 skb_list_del_init(skb); 5784 enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5785 } 5786 } 5787 } 5788 #endif 5789 __netif_receive_skb_list(head); 5790 rcu_read_unlock(); 5791 } 5792 5793 /** 5794 * netif_receive_skb - process receive buffer from network 5795 * @skb: buffer to process 5796 * 5797 * netif_receive_skb() is the main receive data processing function. 5798 * It always succeeds. The buffer may be dropped during processing 5799 * for congestion control or by the protocol layers. 5800 * 5801 * This function may only be called from softirq context and interrupts 5802 * should be enabled. 5803 * 5804 * Return values (usually ignored): 5805 * NET_RX_SUCCESS: no congestion 5806 * NET_RX_DROP: packet was dropped 5807 */ 5808 int netif_receive_skb(struct sk_buff *skb) 5809 { 5810 int ret; 5811 5812 trace_netif_receive_skb_entry(skb); 5813 5814 ret = netif_receive_skb_internal(skb); 5815 trace_netif_receive_skb_exit(ret); 5816 5817 return ret; 5818 } 5819 EXPORT_SYMBOL(netif_receive_skb); 5820 5821 /** 5822 * netif_receive_skb_list - process many receive buffers from network 5823 * @head: list of skbs to process. 5824 * 5825 * Since return value of netif_receive_skb() is normally ignored, and 5826 * wouldn't be meaningful for a list, this function returns void. 5827 * 5828 * This function may only be called from softirq context and interrupts 5829 * should be enabled. 5830 */ 5831 void netif_receive_skb_list(struct list_head *head) 5832 { 5833 struct sk_buff *skb; 5834 5835 if (list_empty(head)) 5836 return; 5837 if (trace_netif_receive_skb_list_entry_enabled()) { 5838 list_for_each_entry(skb, head, list) 5839 trace_netif_receive_skb_list_entry(skb); 5840 } 5841 netif_receive_skb_list_internal(head); 5842 trace_netif_receive_skb_list_exit(0); 5843 } 5844 EXPORT_SYMBOL(netif_receive_skb_list); 5845 5846 static DEFINE_PER_CPU(struct work_struct, flush_works); 5847 5848 /* Network device is going away, flush any packets still pending */ 5849 static void flush_backlog(struct work_struct *work) 5850 { 5851 struct sk_buff *skb, *tmp; 5852 struct softnet_data *sd; 5853 5854 local_bh_disable(); 5855 sd = this_cpu_ptr(&softnet_data); 5856 5857 rps_lock_irq_disable(sd); 5858 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { 5859 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5860 __skb_unlink(skb, &sd->input_pkt_queue); 5861 dev_kfree_skb_irq(skb); 5862 input_queue_head_incr(sd); 5863 } 5864 } 5865 rps_unlock_irq_enable(sd); 5866 5867 skb_queue_walk_safe(&sd->process_queue, skb, tmp) { 5868 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5869 __skb_unlink(skb, &sd->process_queue); 5870 kfree_skb(skb); 5871 input_queue_head_incr(sd); 5872 } 5873 } 5874 local_bh_enable(); 5875 } 5876 5877 static bool flush_required(int cpu) 5878 { 5879 #if IS_ENABLED(CONFIG_RPS) 5880 struct softnet_data *sd = &per_cpu(softnet_data, cpu); 5881 bool do_flush; 5882 5883 rps_lock_irq_disable(sd); 5884 5885 /* as insertion into process_queue happens with the rps lock held, 5886 * process_queue access may race only with dequeue 5887 */ 5888 do_flush = !skb_queue_empty(&sd->input_pkt_queue) || 5889 !skb_queue_empty_lockless(&sd->process_queue); 5890 rps_unlock_irq_enable(sd); 5891 5892 return do_flush; 5893 #endif 5894 /* without RPS we can't safely check input_pkt_queue: during a 5895 * concurrent remote skb_queue_splice() we can detect as empty both 5896 * input_pkt_queue and process_queue even if the latter could end-up 5897 * containing a lot of packets. 5898 */ 5899 return true; 5900 } 5901 5902 static void flush_all_backlogs(void) 5903 { 5904 static cpumask_t flush_cpus; 5905 unsigned int cpu; 5906 5907 /* since we are under rtnl lock protection we can use static data 5908 * for the cpumask and avoid allocating on stack the possibly 5909 * large mask 5910 */ 5911 ASSERT_RTNL(); 5912 5913 cpus_read_lock(); 5914 5915 cpumask_clear(&flush_cpus); 5916 for_each_online_cpu(cpu) { 5917 if (flush_required(cpu)) { 5918 queue_work_on(cpu, system_highpri_wq, 5919 per_cpu_ptr(&flush_works, cpu)); 5920 cpumask_set_cpu(cpu, &flush_cpus); 5921 } 5922 } 5923 5924 /* we can have in flight packet[s] on the cpus we are not flushing, 5925 * synchronize_net() in unregister_netdevice_many() will take care of 5926 * them 5927 */ 5928 for_each_cpu(cpu, &flush_cpus) 5929 flush_work(per_cpu_ptr(&flush_works, cpu)); 5930 5931 cpus_read_unlock(); 5932 } 5933 5934 static void net_rps_send_ipi(struct softnet_data *remsd) 5935 { 5936 #ifdef CONFIG_RPS 5937 while (remsd) { 5938 struct softnet_data *next = remsd->rps_ipi_next; 5939 5940 if (cpu_online(remsd->cpu)) 5941 smp_call_function_single_async(remsd->cpu, &remsd->csd); 5942 remsd = next; 5943 } 5944 #endif 5945 } 5946 5947 /* 5948 * net_rps_action_and_irq_enable sends any pending IPI's for rps. 5949 * Note: called with local irq disabled, but exits with local irq enabled. 5950 */ 5951 static void net_rps_action_and_irq_enable(struct softnet_data *sd) 5952 { 5953 #ifdef CONFIG_RPS 5954 struct softnet_data *remsd = sd->rps_ipi_list; 5955 5956 if (remsd) { 5957 sd->rps_ipi_list = NULL; 5958 5959 local_irq_enable(); 5960 5961 /* Send pending IPI's to kick RPS processing on remote cpus. */ 5962 net_rps_send_ipi(remsd); 5963 } else 5964 #endif 5965 local_irq_enable(); 5966 } 5967 5968 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) 5969 { 5970 #ifdef CONFIG_RPS 5971 return sd->rps_ipi_list != NULL; 5972 #else 5973 return false; 5974 #endif 5975 } 5976 5977 static int process_backlog(struct napi_struct *napi, int quota) 5978 { 5979 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); 5980 bool again = true; 5981 int work = 0; 5982 5983 /* Check if we have pending ipi, its better to send them now, 5984 * not waiting net_rx_action() end. 5985 */ 5986 if (sd_has_rps_ipi_waiting(sd)) { 5987 local_irq_disable(); 5988 net_rps_action_and_irq_enable(sd); 5989 } 5990 5991 napi->weight = READ_ONCE(dev_rx_weight); 5992 while (again) { 5993 struct sk_buff *skb; 5994 5995 while ((skb = __skb_dequeue(&sd->process_queue))) { 5996 rcu_read_lock(); 5997 __netif_receive_skb(skb); 5998 rcu_read_unlock(); 5999 input_queue_head_incr(sd); 6000 if (++work >= quota) 6001 return work; 6002 6003 } 6004 6005 rps_lock_irq_disable(sd); 6006 if (skb_queue_empty(&sd->input_pkt_queue)) { 6007 /* 6008 * Inline a custom version of __napi_complete(). 6009 * only current cpu owns and manipulates this napi, 6010 * and NAPI_STATE_SCHED is the only possible flag set 6011 * on backlog. 6012 * We can use a plain write instead of clear_bit(), 6013 * and we dont need an smp_mb() memory barrier. 6014 */ 6015 napi->state = 0; 6016 again = false; 6017 } else { 6018 skb_queue_splice_tail_init(&sd->input_pkt_queue, 6019 &sd->process_queue); 6020 } 6021 rps_unlock_irq_enable(sd); 6022 } 6023 6024 return work; 6025 } 6026 6027 /** 6028 * __napi_schedule - schedule for receive 6029 * @n: entry to schedule 6030 * 6031 * The entry's receive function will be scheduled to run. 6032 * Consider using __napi_schedule_irqoff() if hard irqs are masked. 6033 */ 6034 void __napi_schedule(struct napi_struct *n) 6035 { 6036 unsigned long flags; 6037 6038 local_irq_save(flags); 6039 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6040 local_irq_restore(flags); 6041 } 6042 EXPORT_SYMBOL(__napi_schedule); 6043 6044 /** 6045 * napi_schedule_prep - check if napi can be scheduled 6046 * @n: napi context 6047 * 6048 * Test if NAPI routine is already running, and if not mark 6049 * it as running. This is used as a condition variable to 6050 * insure only one NAPI poll instance runs. We also make 6051 * sure there is no pending NAPI disable. 6052 */ 6053 bool napi_schedule_prep(struct napi_struct *n) 6054 { 6055 unsigned long new, val = READ_ONCE(n->state); 6056 6057 do { 6058 if (unlikely(val & NAPIF_STATE_DISABLE)) 6059 return false; 6060 new = val | NAPIF_STATE_SCHED; 6061 6062 /* Sets STATE_MISSED bit if STATE_SCHED was already set 6063 * This was suggested by Alexander Duyck, as compiler 6064 * emits better code than : 6065 * if (val & NAPIF_STATE_SCHED) 6066 * new |= NAPIF_STATE_MISSED; 6067 */ 6068 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * 6069 NAPIF_STATE_MISSED; 6070 } while (!try_cmpxchg(&n->state, &val, new)); 6071 6072 return !(val & NAPIF_STATE_SCHED); 6073 } 6074 EXPORT_SYMBOL(napi_schedule_prep); 6075 6076 /** 6077 * __napi_schedule_irqoff - schedule for receive 6078 * @n: entry to schedule 6079 * 6080 * Variant of __napi_schedule() assuming hard irqs are masked. 6081 * 6082 * On PREEMPT_RT enabled kernels this maps to __napi_schedule() 6083 * because the interrupt disabled assumption might not be true 6084 * due to force-threaded interrupts and spinlock substitution. 6085 */ 6086 void __napi_schedule_irqoff(struct napi_struct *n) 6087 { 6088 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6089 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6090 else 6091 __napi_schedule(n); 6092 } 6093 EXPORT_SYMBOL(__napi_schedule_irqoff); 6094 6095 bool napi_complete_done(struct napi_struct *n, int work_done) 6096 { 6097 unsigned long flags, val, new, timeout = 0; 6098 bool ret = true; 6099 6100 /* 6101 * 1) Don't let napi dequeue from the cpu poll list 6102 * just in case its running on a different cpu. 6103 * 2) If we are busy polling, do nothing here, we have 6104 * the guarantee we will be called later. 6105 */ 6106 if (unlikely(n->state & (NAPIF_STATE_NPSVC | 6107 NAPIF_STATE_IN_BUSY_POLL))) 6108 return false; 6109 6110 if (work_done) { 6111 if (n->gro_bitmask) 6112 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6113 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs); 6114 } 6115 if (n->defer_hard_irqs_count > 0) { 6116 n->defer_hard_irqs_count--; 6117 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6118 if (timeout) 6119 ret = false; 6120 } 6121 if (n->gro_bitmask) { 6122 /* When the NAPI instance uses a timeout and keeps postponing 6123 * it, we need to bound somehow the time packets are kept in 6124 * the GRO layer 6125 */ 6126 napi_gro_flush(n, !!timeout); 6127 } 6128 6129 gro_normal_list(n); 6130 6131 if (unlikely(!list_empty(&n->poll_list))) { 6132 /* If n->poll_list is not empty, we need to mask irqs */ 6133 local_irq_save(flags); 6134 list_del_init(&n->poll_list); 6135 local_irq_restore(flags); 6136 } 6137 WRITE_ONCE(n->list_owner, -1); 6138 6139 val = READ_ONCE(n->state); 6140 do { 6141 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); 6142 6143 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED | 6144 NAPIF_STATE_SCHED_THREADED | 6145 NAPIF_STATE_PREFER_BUSY_POLL); 6146 6147 /* If STATE_MISSED was set, leave STATE_SCHED set, 6148 * because we will call napi->poll() one more time. 6149 * This C code was suggested by Alexander Duyck to help gcc. 6150 */ 6151 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * 6152 NAPIF_STATE_SCHED; 6153 } while (!try_cmpxchg(&n->state, &val, new)); 6154 6155 if (unlikely(val & NAPIF_STATE_MISSED)) { 6156 __napi_schedule(n); 6157 return false; 6158 } 6159 6160 if (timeout) 6161 hrtimer_start(&n->timer, ns_to_ktime(timeout), 6162 HRTIMER_MODE_REL_PINNED); 6163 return ret; 6164 } 6165 EXPORT_SYMBOL(napi_complete_done); 6166 6167 /* must be called under rcu_read_lock(), as we dont take a reference */ 6168 static struct napi_struct *napi_by_id(unsigned int napi_id) 6169 { 6170 unsigned int hash = napi_id % HASH_SIZE(napi_hash); 6171 struct napi_struct *napi; 6172 6173 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) 6174 if (napi->napi_id == napi_id) 6175 return napi; 6176 6177 return NULL; 6178 } 6179 6180 #if defined(CONFIG_NET_RX_BUSY_POLL) 6181 6182 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule) 6183 { 6184 if (!skip_schedule) { 6185 gro_normal_list(napi); 6186 __napi_schedule(napi); 6187 return; 6188 } 6189 6190 if (napi->gro_bitmask) { 6191 /* flush too old packets 6192 * If HZ < 1000, flush all packets. 6193 */ 6194 napi_gro_flush(napi, HZ >= 1000); 6195 } 6196 6197 gro_normal_list(napi); 6198 clear_bit(NAPI_STATE_SCHED, &napi->state); 6199 } 6200 6201 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll, 6202 u16 budget) 6203 { 6204 bool skip_schedule = false; 6205 unsigned long timeout; 6206 int rc; 6207 6208 /* Busy polling means there is a high chance device driver hard irq 6209 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was 6210 * set in napi_schedule_prep(). 6211 * Since we are about to call napi->poll() once more, we can safely 6212 * clear NAPI_STATE_MISSED. 6213 * 6214 * Note: x86 could use a single "lock and ..." instruction 6215 * to perform these two clear_bit() 6216 */ 6217 clear_bit(NAPI_STATE_MISSED, &napi->state); 6218 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); 6219 6220 local_bh_disable(); 6221 6222 if (prefer_busy_poll) { 6223 napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs); 6224 timeout = READ_ONCE(napi->dev->gro_flush_timeout); 6225 if (napi->defer_hard_irqs_count && timeout) { 6226 hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED); 6227 skip_schedule = true; 6228 } 6229 } 6230 6231 /* All we really want here is to re-enable device interrupts. 6232 * Ideally, a new ndo_busy_poll_stop() could avoid another round. 6233 */ 6234 rc = napi->poll(napi, budget); 6235 /* We can't gro_normal_list() here, because napi->poll() might have 6236 * rearmed the napi (napi_complete_done()) in which case it could 6237 * already be running on another CPU. 6238 */ 6239 trace_napi_poll(napi, rc, budget); 6240 netpoll_poll_unlock(have_poll_lock); 6241 if (rc == budget) 6242 __busy_poll_stop(napi, skip_schedule); 6243 local_bh_enable(); 6244 } 6245 6246 void napi_busy_loop(unsigned int napi_id, 6247 bool (*loop_end)(void *, unsigned long), 6248 void *loop_end_arg, bool prefer_busy_poll, u16 budget) 6249 { 6250 unsigned long start_time = loop_end ? busy_loop_current_time() : 0; 6251 int (*napi_poll)(struct napi_struct *napi, int budget); 6252 void *have_poll_lock = NULL; 6253 struct napi_struct *napi; 6254 6255 restart: 6256 napi_poll = NULL; 6257 6258 rcu_read_lock(); 6259 6260 napi = napi_by_id(napi_id); 6261 if (!napi) 6262 goto out; 6263 6264 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6265 preempt_disable(); 6266 for (;;) { 6267 int work = 0; 6268 6269 local_bh_disable(); 6270 if (!napi_poll) { 6271 unsigned long val = READ_ONCE(napi->state); 6272 6273 /* If multiple threads are competing for this napi, 6274 * we avoid dirtying napi->state as much as we can. 6275 */ 6276 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | 6277 NAPIF_STATE_IN_BUSY_POLL)) { 6278 if (prefer_busy_poll) 6279 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6280 goto count; 6281 } 6282 if (cmpxchg(&napi->state, val, 6283 val | NAPIF_STATE_IN_BUSY_POLL | 6284 NAPIF_STATE_SCHED) != val) { 6285 if (prefer_busy_poll) 6286 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6287 goto count; 6288 } 6289 have_poll_lock = netpoll_poll_lock(napi); 6290 napi_poll = napi->poll; 6291 } 6292 work = napi_poll(napi, budget); 6293 trace_napi_poll(napi, work, budget); 6294 gro_normal_list(napi); 6295 count: 6296 if (work > 0) 6297 __NET_ADD_STATS(dev_net(napi->dev), 6298 LINUX_MIB_BUSYPOLLRXPACKETS, work); 6299 local_bh_enable(); 6300 6301 if (!loop_end || loop_end(loop_end_arg, start_time)) 6302 break; 6303 6304 if (unlikely(need_resched())) { 6305 if (napi_poll) 6306 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6307 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6308 preempt_enable(); 6309 rcu_read_unlock(); 6310 cond_resched(); 6311 if (loop_end(loop_end_arg, start_time)) 6312 return; 6313 goto restart; 6314 } 6315 cpu_relax(); 6316 } 6317 if (napi_poll) 6318 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6319 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6320 preempt_enable(); 6321 out: 6322 rcu_read_unlock(); 6323 } 6324 EXPORT_SYMBOL(napi_busy_loop); 6325 6326 #endif /* CONFIG_NET_RX_BUSY_POLL */ 6327 6328 static void napi_hash_add(struct napi_struct *napi) 6329 { 6330 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state)) 6331 return; 6332 6333 spin_lock(&napi_hash_lock); 6334 6335 /* 0..NR_CPUS range is reserved for sender_cpu use */ 6336 do { 6337 if (unlikely(++napi_gen_id < MIN_NAPI_ID)) 6338 napi_gen_id = MIN_NAPI_ID; 6339 } while (napi_by_id(napi_gen_id)); 6340 napi->napi_id = napi_gen_id; 6341 6342 hlist_add_head_rcu(&napi->napi_hash_node, 6343 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); 6344 6345 spin_unlock(&napi_hash_lock); 6346 } 6347 6348 /* Warning : caller is responsible to make sure rcu grace period 6349 * is respected before freeing memory containing @napi 6350 */ 6351 static void napi_hash_del(struct napi_struct *napi) 6352 { 6353 spin_lock(&napi_hash_lock); 6354 6355 hlist_del_init_rcu(&napi->napi_hash_node); 6356 6357 spin_unlock(&napi_hash_lock); 6358 } 6359 6360 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) 6361 { 6362 struct napi_struct *napi; 6363 6364 napi = container_of(timer, struct napi_struct, timer); 6365 6366 /* Note : we use a relaxed variant of napi_schedule_prep() not setting 6367 * NAPI_STATE_MISSED, since we do not react to a device IRQ. 6368 */ 6369 if (!napi_disable_pending(napi) && 6370 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) { 6371 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6372 __napi_schedule_irqoff(napi); 6373 } 6374 6375 return HRTIMER_NORESTART; 6376 } 6377 6378 static void init_gro_hash(struct napi_struct *napi) 6379 { 6380 int i; 6381 6382 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6383 INIT_LIST_HEAD(&napi->gro_hash[i].list); 6384 napi->gro_hash[i].count = 0; 6385 } 6386 napi->gro_bitmask = 0; 6387 } 6388 6389 int dev_set_threaded(struct net_device *dev, bool threaded) 6390 { 6391 struct napi_struct *napi; 6392 int err = 0; 6393 6394 if (dev->threaded == threaded) 6395 return 0; 6396 6397 if (threaded) { 6398 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6399 if (!napi->thread) { 6400 err = napi_kthread_create(napi); 6401 if (err) { 6402 threaded = false; 6403 break; 6404 } 6405 } 6406 } 6407 } 6408 6409 dev->threaded = threaded; 6410 6411 /* Make sure kthread is created before THREADED bit 6412 * is set. 6413 */ 6414 smp_mb__before_atomic(); 6415 6416 /* Setting/unsetting threaded mode on a napi might not immediately 6417 * take effect, if the current napi instance is actively being 6418 * polled. In this case, the switch between threaded mode and 6419 * softirq mode will happen in the next round of napi_schedule(). 6420 * This should not cause hiccups/stalls to the live traffic. 6421 */ 6422 list_for_each_entry(napi, &dev->napi_list, dev_list) 6423 assign_bit(NAPI_STATE_THREADED, &napi->state, threaded); 6424 6425 return err; 6426 } 6427 EXPORT_SYMBOL(dev_set_threaded); 6428 6429 void netif_napi_add_weight(struct net_device *dev, struct napi_struct *napi, 6430 int (*poll)(struct napi_struct *, int), int weight) 6431 { 6432 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state))) 6433 return; 6434 6435 INIT_LIST_HEAD(&napi->poll_list); 6436 INIT_HLIST_NODE(&napi->napi_hash_node); 6437 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); 6438 napi->timer.function = napi_watchdog; 6439 init_gro_hash(napi); 6440 napi->skb = NULL; 6441 INIT_LIST_HEAD(&napi->rx_list); 6442 napi->rx_count = 0; 6443 napi->poll = poll; 6444 if (weight > NAPI_POLL_WEIGHT) 6445 netdev_err_once(dev, "%s() called with weight %d\n", __func__, 6446 weight); 6447 napi->weight = weight; 6448 napi->dev = dev; 6449 #ifdef CONFIG_NETPOLL 6450 napi->poll_owner = -1; 6451 #endif 6452 napi->list_owner = -1; 6453 set_bit(NAPI_STATE_SCHED, &napi->state); 6454 set_bit(NAPI_STATE_NPSVC, &napi->state); 6455 list_add_rcu(&napi->dev_list, &dev->napi_list); 6456 napi_hash_add(napi); 6457 napi_get_frags_check(napi); 6458 /* Create kthread for this napi if dev->threaded is set. 6459 * Clear dev->threaded if kthread creation failed so that 6460 * threaded mode will not be enabled in napi_enable(). 6461 */ 6462 if (dev->threaded && napi_kthread_create(napi)) 6463 dev->threaded = 0; 6464 } 6465 EXPORT_SYMBOL(netif_napi_add_weight); 6466 6467 void napi_disable(struct napi_struct *n) 6468 { 6469 unsigned long val, new; 6470 6471 might_sleep(); 6472 set_bit(NAPI_STATE_DISABLE, &n->state); 6473 6474 val = READ_ONCE(n->state); 6475 do { 6476 while (val & (NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC)) { 6477 usleep_range(20, 200); 6478 val = READ_ONCE(n->state); 6479 } 6480 6481 new = val | NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC; 6482 new &= ~(NAPIF_STATE_THREADED | NAPIF_STATE_PREFER_BUSY_POLL); 6483 } while (!try_cmpxchg(&n->state, &val, new)); 6484 6485 hrtimer_cancel(&n->timer); 6486 6487 clear_bit(NAPI_STATE_DISABLE, &n->state); 6488 } 6489 EXPORT_SYMBOL(napi_disable); 6490 6491 /** 6492 * napi_enable - enable NAPI scheduling 6493 * @n: NAPI context 6494 * 6495 * Resume NAPI from being scheduled on this context. 6496 * Must be paired with napi_disable. 6497 */ 6498 void napi_enable(struct napi_struct *n) 6499 { 6500 unsigned long new, val = READ_ONCE(n->state); 6501 6502 do { 6503 BUG_ON(!test_bit(NAPI_STATE_SCHED, &val)); 6504 6505 new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC); 6506 if (n->dev->threaded && n->thread) 6507 new |= NAPIF_STATE_THREADED; 6508 } while (!try_cmpxchg(&n->state, &val, new)); 6509 } 6510 EXPORT_SYMBOL(napi_enable); 6511 6512 static void flush_gro_hash(struct napi_struct *napi) 6513 { 6514 int i; 6515 6516 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6517 struct sk_buff *skb, *n; 6518 6519 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list) 6520 kfree_skb(skb); 6521 napi->gro_hash[i].count = 0; 6522 } 6523 } 6524 6525 /* Must be called in process context */ 6526 void __netif_napi_del(struct napi_struct *napi) 6527 { 6528 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state)) 6529 return; 6530 6531 napi_hash_del(napi); 6532 list_del_rcu(&napi->dev_list); 6533 napi_free_frags(napi); 6534 6535 flush_gro_hash(napi); 6536 napi->gro_bitmask = 0; 6537 6538 if (napi->thread) { 6539 kthread_stop(napi->thread); 6540 napi->thread = NULL; 6541 } 6542 } 6543 EXPORT_SYMBOL(__netif_napi_del); 6544 6545 static int __napi_poll(struct napi_struct *n, bool *repoll) 6546 { 6547 int work, weight; 6548 6549 weight = n->weight; 6550 6551 /* This NAPI_STATE_SCHED test is for avoiding a race 6552 * with netpoll's poll_napi(). Only the entity which 6553 * obtains the lock and sees NAPI_STATE_SCHED set will 6554 * actually make the ->poll() call. Therefore we avoid 6555 * accidentally calling ->poll() when NAPI is not scheduled. 6556 */ 6557 work = 0; 6558 if (test_bit(NAPI_STATE_SCHED, &n->state)) { 6559 work = n->poll(n, weight); 6560 trace_napi_poll(n, work, weight); 6561 } 6562 6563 if (unlikely(work > weight)) 6564 netdev_err_once(n->dev, "NAPI poll function %pS returned %d, exceeding its budget of %d.\n", 6565 n->poll, work, weight); 6566 6567 if (likely(work < weight)) 6568 return work; 6569 6570 /* Drivers must not modify the NAPI state if they 6571 * consume the entire weight. In such cases this code 6572 * still "owns" the NAPI instance and therefore can 6573 * move the instance around on the list at-will. 6574 */ 6575 if (unlikely(napi_disable_pending(n))) { 6576 napi_complete(n); 6577 return work; 6578 } 6579 6580 /* The NAPI context has more processing work, but busy-polling 6581 * is preferred. Exit early. 6582 */ 6583 if (napi_prefer_busy_poll(n)) { 6584 if (napi_complete_done(n, work)) { 6585 /* If timeout is not set, we need to make sure 6586 * that the NAPI is re-scheduled. 6587 */ 6588 napi_schedule(n); 6589 } 6590 return work; 6591 } 6592 6593 if (n->gro_bitmask) { 6594 /* flush too old packets 6595 * If HZ < 1000, flush all packets. 6596 */ 6597 napi_gro_flush(n, HZ >= 1000); 6598 } 6599 6600 gro_normal_list(n); 6601 6602 /* Some drivers may have called napi_schedule 6603 * prior to exhausting their budget. 6604 */ 6605 if (unlikely(!list_empty(&n->poll_list))) { 6606 pr_warn_once("%s: Budget exhausted after napi rescheduled\n", 6607 n->dev ? n->dev->name : "backlog"); 6608 return work; 6609 } 6610 6611 *repoll = true; 6612 6613 return work; 6614 } 6615 6616 static int napi_poll(struct napi_struct *n, struct list_head *repoll) 6617 { 6618 bool do_repoll = false; 6619 void *have; 6620 int work; 6621 6622 list_del_init(&n->poll_list); 6623 6624 have = netpoll_poll_lock(n); 6625 6626 work = __napi_poll(n, &do_repoll); 6627 6628 if (do_repoll) 6629 list_add_tail(&n->poll_list, repoll); 6630 6631 netpoll_poll_unlock(have); 6632 6633 return work; 6634 } 6635 6636 static int napi_thread_wait(struct napi_struct *napi) 6637 { 6638 bool woken = false; 6639 6640 set_current_state(TASK_INTERRUPTIBLE); 6641 6642 while (!kthread_should_stop()) { 6643 /* Testing SCHED_THREADED bit here to make sure the current 6644 * kthread owns this napi and could poll on this napi. 6645 * Testing SCHED bit is not enough because SCHED bit might be 6646 * set by some other busy poll thread or by napi_disable(). 6647 */ 6648 if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state) || woken) { 6649 WARN_ON(!list_empty(&napi->poll_list)); 6650 __set_current_state(TASK_RUNNING); 6651 return 0; 6652 } 6653 6654 schedule(); 6655 /* woken being true indicates this thread owns this napi. */ 6656 woken = true; 6657 set_current_state(TASK_INTERRUPTIBLE); 6658 } 6659 __set_current_state(TASK_RUNNING); 6660 6661 return -1; 6662 } 6663 6664 static void skb_defer_free_flush(struct softnet_data *sd) 6665 { 6666 struct sk_buff *skb, *next; 6667 6668 /* Paired with WRITE_ONCE() in skb_attempt_defer_free() */ 6669 if (!READ_ONCE(sd->defer_list)) 6670 return; 6671 6672 spin_lock(&sd->defer_lock); 6673 skb = sd->defer_list; 6674 sd->defer_list = NULL; 6675 sd->defer_count = 0; 6676 spin_unlock(&sd->defer_lock); 6677 6678 while (skb != NULL) { 6679 next = skb->next; 6680 napi_consume_skb(skb, 1); 6681 skb = next; 6682 } 6683 } 6684 6685 static int napi_threaded_poll(void *data) 6686 { 6687 struct napi_struct *napi = data; 6688 struct softnet_data *sd; 6689 void *have; 6690 6691 while (!napi_thread_wait(napi)) { 6692 for (;;) { 6693 bool repoll = false; 6694 6695 local_bh_disable(); 6696 sd = this_cpu_ptr(&softnet_data); 6697 sd->in_napi_threaded_poll = true; 6698 6699 have = netpoll_poll_lock(napi); 6700 __napi_poll(napi, &repoll); 6701 netpoll_poll_unlock(have); 6702 6703 sd->in_napi_threaded_poll = false; 6704 barrier(); 6705 6706 if (sd_has_rps_ipi_waiting(sd)) { 6707 local_irq_disable(); 6708 net_rps_action_and_irq_enable(sd); 6709 } 6710 skb_defer_free_flush(sd); 6711 local_bh_enable(); 6712 6713 if (!repoll) 6714 break; 6715 6716 cond_resched(); 6717 } 6718 } 6719 return 0; 6720 } 6721 6722 static __latent_entropy void net_rx_action(struct softirq_action *h) 6723 { 6724 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 6725 unsigned long time_limit = jiffies + 6726 usecs_to_jiffies(READ_ONCE(netdev_budget_usecs)); 6727 int budget = READ_ONCE(netdev_budget); 6728 LIST_HEAD(list); 6729 LIST_HEAD(repoll); 6730 6731 start: 6732 sd->in_net_rx_action = true; 6733 local_irq_disable(); 6734 list_splice_init(&sd->poll_list, &list); 6735 local_irq_enable(); 6736 6737 for (;;) { 6738 struct napi_struct *n; 6739 6740 skb_defer_free_flush(sd); 6741 6742 if (list_empty(&list)) { 6743 if (list_empty(&repoll)) { 6744 sd->in_net_rx_action = false; 6745 barrier(); 6746 /* We need to check if ____napi_schedule() 6747 * had refilled poll_list while 6748 * sd->in_net_rx_action was true. 6749 */ 6750 if (!list_empty(&sd->poll_list)) 6751 goto start; 6752 if (!sd_has_rps_ipi_waiting(sd)) 6753 goto end; 6754 } 6755 break; 6756 } 6757 6758 n = list_first_entry(&list, struct napi_struct, poll_list); 6759 budget -= napi_poll(n, &repoll); 6760 6761 /* If softirq window is exhausted then punt. 6762 * Allow this to run for 2 jiffies since which will allow 6763 * an average latency of 1.5/HZ. 6764 */ 6765 if (unlikely(budget <= 0 || 6766 time_after_eq(jiffies, time_limit))) { 6767 sd->time_squeeze++; 6768 break; 6769 } 6770 } 6771 6772 local_irq_disable(); 6773 6774 list_splice_tail_init(&sd->poll_list, &list); 6775 list_splice_tail(&repoll, &list); 6776 list_splice(&list, &sd->poll_list); 6777 if (!list_empty(&sd->poll_list)) 6778 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 6779 else 6780 sd->in_net_rx_action = false; 6781 6782 net_rps_action_and_irq_enable(sd); 6783 end:; 6784 } 6785 6786 struct netdev_adjacent { 6787 struct net_device *dev; 6788 netdevice_tracker dev_tracker; 6789 6790 /* upper master flag, there can only be one master device per list */ 6791 bool master; 6792 6793 /* lookup ignore flag */ 6794 bool ignore; 6795 6796 /* counter for the number of times this device was added to us */ 6797 u16 ref_nr; 6798 6799 /* private field for the users */ 6800 void *private; 6801 6802 struct list_head list; 6803 struct rcu_head rcu; 6804 }; 6805 6806 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, 6807 struct list_head *adj_list) 6808 { 6809 struct netdev_adjacent *adj; 6810 6811 list_for_each_entry(adj, adj_list, list) { 6812 if (adj->dev == adj_dev) 6813 return adj; 6814 } 6815 return NULL; 6816 } 6817 6818 static int ____netdev_has_upper_dev(struct net_device *upper_dev, 6819 struct netdev_nested_priv *priv) 6820 { 6821 struct net_device *dev = (struct net_device *)priv->data; 6822 6823 return upper_dev == dev; 6824 } 6825 6826 /** 6827 * netdev_has_upper_dev - Check if device is linked to an upper device 6828 * @dev: device 6829 * @upper_dev: upper device to check 6830 * 6831 * Find out if a device is linked to specified upper device and return true 6832 * in case it is. Note that this checks only immediate upper device, 6833 * not through a complete stack of devices. The caller must hold the RTNL lock. 6834 */ 6835 bool netdev_has_upper_dev(struct net_device *dev, 6836 struct net_device *upper_dev) 6837 { 6838 struct netdev_nested_priv priv = { 6839 .data = (void *)upper_dev, 6840 }; 6841 6842 ASSERT_RTNL(); 6843 6844 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 6845 &priv); 6846 } 6847 EXPORT_SYMBOL(netdev_has_upper_dev); 6848 6849 /** 6850 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device 6851 * @dev: device 6852 * @upper_dev: upper device to check 6853 * 6854 * Find out if a device is linked to specified upper device and return true 6855 * in case it is. Note that this checks the entire upper device chain. 6856 * The caller must hold rcu lock. 6857 */ 6858 6859 bool netdev_has_upper_dev_all_rcu(struct net_device *dev, 6860 struct net_device *upper_dev) 6861 { 6862 struct netdev_nested_priv priv = { 6863 .data = (void *)upper_dev, 6864 }; 6865 6866 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 6867 &priv); 6868 } 6869 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); 6870 6871 /** 6872 * netdev_has_any_upper_dev - Check if device is linked to some device 6873 * @dev: device 6874 * 6875 * Find out if a device is linked to an upper device and return true in case 6876 * it is. The caller must hold the RTNL lock. 6877 */ 6878 bool netdev_has_any_upper_dev(struct net_device *dev) 6879 { 6880 ASSERT_RTNL(); 6881 6882 return !list_empty(&dev->adj_list.upper); 6883 } 6884 EXPORT_SYMBOL(netdev_has_any_upper_dev); 6885 6886 /** 6887 * netdev_master_upper_dev_get - Get master upper device 6888 * @dev: device 6889 * 6890 * Find a master upper device and return pointer to it or NULL in case 6891 * it's not there. The caller must hold the RTNL lock. 6892 */ 6893 struct net_device *netdev_master_upper_dev_get(struct net_device *dev) 6894 { 6895 struct netdev_adjacent *upper; 6896 6897 ASSERT_RTNL(); 6898 6899 if (list_empty(&dev->adj_list.upper)) 6900 return NULL; 6901 6902 upper = list_first_entry(&dev->adj_list.upper, 6903 struct netdev_adjacent, list); 6904 if (likely(upper->master)) 6905 return upper->dev; 6906 return NULL; 6907 } 6908 EXPORT_SYMBOL(netdev_master_upper_dev_get); 6909 6910 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev) 6911 { 6912 struct netdev_adjacent *upper; 6913 6914 ASSERT_RTNL(); 6915 6916 if (list_empty(&dev->adj_list.upper)) 6917 return NULL; 6918 6919 upper = list_first_entry(&dev->adj_list.upper, 6920 struct netdev_adjacent, list); 6921 if (likely(upper->master) && !upper->ignore) 6922 return upper->dev; 6923 return NULL; 6924 } 6925 6926 /** 6927 * netdev_has_any_lower_dev - Check if device is linked to some device 6928 * @dev: device 6929 * 6930 * Find out if a device is linked to a lower device and return true in case 6931 * it is. The caller must hold the RTNL lock. 6932 */ 6933 static bool netdev_has_any_lower_dev(struct net_device *dev) 6934 { 6935 ASSERT_RTNL(); 6936 6937 return !list_empty(&dev->adj_list.lower); 6938 } 6939 6940 void *netdev_adjacent_get_private(struct list_head *adj_list) 6941 { 6942 struct netdev_adjacent *adj; 6943 6944 adj = list_entry(adj_list, struct netdev_adjacent, list); 6945 6946 return adj->private; 6947 } 6948 EXPORT_SYMBOL(netdev_adjacent_get_private); 6949 6950 /** 6951 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list 6952 * @dev: device 6953 * @iter: list_head ** of the current position 6954 * 6955 * Gets the next device from the dev's upper list, starting from iter 6956 * position. The caller must hold RCU read lock. 6957 */ 6958 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, 6959 struct list_head **iter) 6960 { 6961 struct netdev_adjacent *upper; 6962 6963 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 6964 6965 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 6966 6967 if (&upper->list == &dev->adj_list.upper) 6968 return NULL; 6969 6970 *iter = &upper->list; 6971 6972 return upper->dev; 6973 } 6974 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); 6975 6976 static struct net_device *__netdev_next_upper_dev(struct net_device *dev, 6977 struct list_head **iter, 6978 bool *ignore) 6979 { 6980 struct netdev_adjacent *upper; 6981 6982 upper = list_entry((*iter)->next, struct netdev_adjacent, list); 6983 6984 if (&upper->list == &dev->adj_list.upper) 6985 return NULL; 6986 6987 *iter = &upper->list; 6988 *ignore = upper->ignore; 6989 6990 return upper->dev; 6991 } 6992 6993 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, 6994 struct list_head **iter) 6995 { 6996 struct netdev_adjacent *upper; 6997 6998 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 6999 7000 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7001 7002 if (&upper->list == &dev->adj_list.upper) 7003 return NULL; 7004 7005 *iter = &upper->list; 7006 7007 return upper->dev; 7008 } 7009 7010 static int __netdev_walk_all_upper_dev(struct net_device *dev, 7011 int (*fn)(struct net_device *dev, 7012 struct netdev_nested_priv *priv), 7013 struct netdev_nested_priv *priv) 7014 { 7015 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7016 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7017 int ret, cur = 0; 7018 bool ignore; 7019 7020 now = dev; 7021 iter = &dev->adj_list.upper; 7022 7023 while (1) { 7024 if (now != dev) { 7025 ret = fn(now, priv); 7026 if (ret) 7027 return ret; 7028 } 7029 7030 next = NULL; 7031 while (1) { 7032 udev = __netdev_next_upper_dev(now, &iter, &ignore); 7033 if (!udev) 7034 break; 7035 if (ignore) 7036 continue; 7037 7038 next = udev; 7039 niter = &udev->adj_list.upper; 7040 dev_stack[cur] = now; 7041 iter_stack[cur++] = iter; 7042 break; 7043 } 7044 7045 if (!next) { 7046 if (!cur) 7047 return 0; 7048 next = dev_stack[--cur]; 7049 niter = iter_stack[cur]; 7050 } 7051 7052 now = next; 7053 iter = niter; 7054 } 7055 7056 return 0; 7057 } 7058 7059 int netdev_walk_all_upper_dev_rcu(struct net_device *dev, 7060 int (*fn)(struct net_device *dev, 7061 struct netdev_nested_priv *priv), 7062 struct netdev_nested_priv *priv) 7063 { 7064 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7065 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7066 int ret, cur = 0; 7067 7068 now = dev; 7069 iter = &dev->adj_list.upper; 7070 7071 while (1) { 7072 if (now != dev) { 7073 ret = fn(now, priv); 7074 if (ret) 7075 return ret; 7076 } 7077 7078 next = NULL; 7079 while (1) { 7080 udev = netdev_next_upper_dev_rcu(now, &iter); 7081 if (!udev) 7082 break; 7083 7084 next = udev; 7085 niter = &udev->adj_list.upper; 7086 dev_stack[cur] = now; 7087 iter_stack[cur++] = iter; 7088 break; 7089 } 7090 7091 if (!next) { 7092 if (!cur) 7093 return 0; 7094 next = dev_stack[--cur]; 7095 niter = iter_stack[cur]; 7096 } 7097 7098 now = next; 7099 iter = niter; 7100 } 7101 7102 return 0; 7103 } 7104 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); 7105 7106 static bool __netdev_has_upper_dev(struct net_device *dev, 7107 struct net_device *upper_dev) 7108 { 7109 struct netdev_nested_priv priv = { 7110 .flags = 0, 7111 .data = (void *)upper_dev, 7112 }; 7113 7114 ASSERT_RTNL(); 7115 7116 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev, 7117 &priv); 7118 } 7119 7120 /** 7121 * netdev_lower_get_next_private - Get the next ->private from the 7122 * lower neighbour list 7123 * @dev: device 7124 * @iter: list_head ** of the current position 7125 * 7126 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7127 * list, starting from iter position. The caller must hold either hold the 7128 * RTNL lock or its own locking that guarantees that the neighbour lower 7129 * list will remain unchanged. 7130 */ 7131 void *netdev_lower_get_next_private(struct net_device *dev, 7132 struct list_head **iter) 7133 { 7134 struct netdev_adjacent *lower; 7135 7136 lower = list_entry(*iter, struct netdev_adjacent, list); 7137 7138 if (&lower->list == &dev->adj_list.lower) 7139 return NULL; 7140 7141 *iter = lower->list.next; 7142 7143 return lower->private; 7144 } 7145 EXPORT_SYMBOL(netdev_lower_get_next_private); 7146 7147 /** 7148 * netdev_lower_get_next_private_rcu - Get the next ->private from the 7149 * lower neighbour list, RCU 7150 * variant 7151 * @dev: device 7152 * @iter: list_head ** of the current position 7153 * 7154 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7155 * list, starting from iter position. The caller must hold RCU read lock. 7156 */ 7157 void *netdev_lower_get_next_private_rcu(struct net_device *dev, 7158 struct list_head **iter) 7159 { 7160 struct netdev_adjacent *lower; 7161 7162 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 7163 7164 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7165 7166 if (&lower->list == &dev->adj_list.lower) 7167 return NULL; 7168 7169 *iter = &lower->list; 7170 7171 return lower->private; 7172 } 7173 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); 7174 7175 /** 7176 * netdev_lower_get_next - Get the next device from the lower neighbour 7177 * list 7178 * @dev: device 7179 * @iter: list_head ** of the current position 7180 * 7181 * Gets the next netdev_adjacent from the dev's lower neighbour 7182 * list, starting from iter position. The caller must hold RTNL lock or 7183 * its own locking that guarantees that the neighbour lower 7184 * list will remain unchanged. 7185 */ 7186 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) 7187 { 7188 struct netdev_adjacent *lower; 7189 7190 lower = list_entry(*iter, struct netdev_adjacent, list); 7191 7192 if (&lower->list == &dev->adj_list.lower) 7193 return NULL; 7194 7195 *iter = lower->list.next; 7196 7197 return lower->dev; 7198 } 7199 EXPORT_SYMBOL(netdev_lower_get_next); 7200 7201 static struct net_device *netdev_next_lower_dev(struct net_device *dev, 7202 struct list_head **iter) 7203 { 7204 struct netdev_adjacent *lower; 7205 7206 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7207 7208 if (&lower->list == &dev->adj_list.lower) 7209 return NULL; 7210 7211 *iter = &lower->list; 7212 7213 return lower->dev; 7214 } 7215 7216 static struct net_device *__netdev_next_lower_dev(struct net_device *dev, 7217 struct list_head **iter, 7218 bool *ignore) 7219 { 7220 struct netdev_adjacent *lower; 7221 7222 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7223 7224 if (&lower->list == &dev->adj_list.lower) 7225 return NULL; 7226 7227 *iter = &lower->list; 7228 *ignore = lower->ignore; 7229 7230 return lower->dev; 7231 } 7232 7233 int netdev_walk_all_lower_dev(struct net_device *dev, 7234 int (*fn)(struct net_device *dev, 7235 struct netdev_nested_priv *priv), 7236 struct netdev_nested_priv *priv) 7237 { 7238 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7239 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7240 int ret, cur = 0; 7241 7242 now = dev; 7243 iter = &dev->adj_list.lower; 7244 7245 while (1) { 7246 if (now != dev) { 7247 ret = fn(now, priv); 7248 if (ret) 7249 return ret; 7250 } 7251 7252 next = NULL; 7253 while (1) { 7254 ldev = netdev_next_lower_dev(now, &iter); 7255 if (!ldev) 7256 break; 7257 7258 next = ldev; 7259 niter = &ldev->adj_list.lower; 7260 dev_stack[cur] = now; 7261 iter_stack[cur++] = iter; 7262 break; 7263 } 7264 7265 if (!next) { 7266 if (!cur) 7267 return 0; 7268 next = dev_stack[--cur]; 7269 niter = iter_stack[cur]; 7270 } 7271 7272 now = next; 7273 iter = niter; 7274 } 7275 7276 return 0; 7277 } 7278 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); 7279 7280 static int __netdev_walk_all_lower_dev(struct net_device *dev, 7281 int (*fn)(struct net_device *dev, 7282 struct netdev_nested_priv *priv), 7283 struct netdev_nested_priv *priv) 7284 { 7285 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7286 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7287 int ret, cur = 0; 7288 bool ignore; 7289 7290 now = dev; 7291 iter = &dev->adj_list.lower; 7292 7293 while (1) { 7294 if (now != dev) { 7295 ret = fn(now, priv); 7296 if (ret) 7297 return ret; 7298 } 7299 7300 next = NULL; 7301 while (1) { 7302 ldev = __netdev_next_lower_dev(now, &iter, &ignore); 7303 if (!ldev) 7304 break; 7305 if (ignore) 7306 continue; 7307 7308 next = ldev; 7309 niter = &ldev->adj_list.lower; 7310 dev_stack[cur] = now; 7311 iter_stack[cur++] = iter; 7312 break; 7313 } 7314 7315 if (!next) { 7316 if (!cur) 7317 return 0; 7318 next = dev_stack[--cur]; 7319 niter = iter_stack[cur]; 7320 } 7321 7322 now = next; 7323 iter = niter; 7324 } 7325 7326 return 0; 7327 } 7328 7329 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, 7330 struct list_head **iter) 7331 { 7332 struct netdev_adjacent *lower; 7333 7334 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7335 if (&lower->list == &dev->adj_list.lower) 7336 return NULL; 7337 7338 *iter = &lower->list; 7339 7340 return lower->dev; 7341 } 7342 EXPORT_SYMBOL(netdev_next_lower_dev_rcu); 7343 7344 static u8 __netdev_upper_depth(struct net_device *dev) 7345 { 7346 struct net_device *udev; 7347 struct list_head *iter; 7348 u8 max_depth = 0; 7349 bool ignore; 7350 7351 for (iter = &dev->adj_list.upper, 7352 udev = __netdev_next_upper_dev(dev, &iter, &ignore); 7353 udev; 7354 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) { 7355 if (ignore) 7356 continue; 7357 if (max_depth < udev->upper_level) 7358 max_depth = udev->upper_level; 7359 } 7360 7361 return max_depth; 7362 } 7363 7364 static u8 __netdev_lower_depth(struct net_device *dev) 7365 { 7366 struct net_device *ldev; 7367 struct list_head *iter; 7368 u8 max_depth = 0; 7369 bool ignore; 7370 7371 for (iter = &dev->adj_list.lower, 7372 ldev = __netdev_next_lower_dev(dev, &iter, &ignore); 7373 ldev; 7374 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) { 7375 if (ignore) 7376 continue; 7377 if (max_depth < ldev->lower_level) 7378 max_depth = ldev->lower_level; 7379 } 7380 7381 return max_depth; 7382 } 7383 7384 static int __netdev_update_upper_level(struct net_device *dev, 7385 struct netdev_nested_priv *__unused) 7386 { 7387 dev->upper_level = __netdev_upper_depth(dev) + 1; 7388 return 0; 7389 } 7390 7391 #ifdef CONFIG_LOCKDEP 7392 static LIST_HEAD(net_unlink_list); 7393 7394 static void net_unlink_todo(struct net_device *dev) 7395 { 7396 if (list_empty(&dev->unlink_list)) 7397 list_add_tail(&dev->unlink_list, &net_unlink_list); 7398 } 7399 #endif 7400 7401 static int __netdev_update_lower_level(struct net_device *dev, 7402 struct netdev_nested_priv *priv) 7403 { 7404 dev->lower_level = __netdev_lower_depth(dev) + 1; 7405 7406 #ifdef CONFIG_LOCKDEP 7407 if (!priv) 7408 return 0; 7409 7410 if (priv->flags & NESTED_SYNC_IMM) 7411 dev->nested_level = dev->lower_level - 1; 7412 if (priv->flags & NESTED_SYNC_TODO) 7413 net_unlink_todo(dev); 7414 #endif 7415 return 0; 7416 } 7417 7418 int netdev_walk_all_lower_dev_rcu(struct net_device *dev, 7419 int (*fn)(struct net_device *dev, 7420 struct netdev_nested_priv *priv), 7421 struct netdev_nested_priv *priv) 7422 { 7423 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7424 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7425 int ret, cur = 0; 7426 7427 now = dev; 7428 iter = &dev->adj_list.lower; 7429 7430 while (1) { 7431 if (now != dev) { 7432 ret = fn(now, priv); 7433 if (ret) 7434 return ret; 7435 } 7436 7437 next = NULL; 7438 while (1) { 7439 ldev = netdev_next_lower_dev_rcu(now, &iter); 7440 if (!ldev) 7441 break; 7442 7443 next = ldev; 7444 niter = &ldev->adj_list.lower; 7445 dev_stack[cur] = now; 7446 iter_stack[cur++] = iter; 7447 break; 7448 } 7449 7450 if (!next) { 7451 if (!cur) 7452 return 0; 7453 next = dev_stack[--cur]; 7454 niter = iter_stack[cur]; 7455 } 7456 7457 now = next; 7458 iter = niter; 7459 } 7460 7461 return 0; 7462 } 7463 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); 7464 7465 /** 7466 * netdev_lower_get_first_private_rcu - Get the first ->private from the 7467 * lower neighbour list, RCU 7468 * variant 7469 * @dev: device 7470 * 7471 * Gets the first netdev_adjacent->private from the dev's lower neighbour 7472 * list. The caller must hold RCU read lock. 7473 */ 7474 void *netdev_lower_get_first_private_rcu(struct net_device *dev) 7475 { 7476 struct netdev_adjacent *lower; 7477 7478 lower = list_first_or_null_rcu(&dev->adj_list.lower, 7479 struct netdev_adjacent, list); 7480 if (lower) 7481 return lower->private; 7482 return NULL; 7483 } 7484 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); 7485 7486 /** 7487 * netdev_master_upper_dev_get_rcu - Get master upper device 7488 * @dev: device 7489 * 7490 * Find a master upper device and return pointer to it or NULL in case 7491 * it's not there. The caller must hold the RCU read lock. 7492 */ 7493 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) 7494 { 7495 struct netdev_adjacent *upper; 7496 7497 upper = list_first_or_null_rcu(&dev->adj_list.upper, 7498 struct netdev_adjacent, list); 7499 if (upper && likely(upper->master)) 7500 return upper->dev; 7501 return NULL; 7502 } 7503 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); 7504 7505 static int netdev_adjacent_sysfs_add(struct net_device *dev, 7506 struct net_device *adj_dev, 7507 struct list_head *dev_list) 7508 { 7509 char linkname[IFNAMSIZ+7]; 7510 7511 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7512 "upper_%s" : "lower_%s", adj_dev->name); 7513 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), 7514 linkname); 7515 } 7516 static void netdev_adjacent_sysfs_del(struct net_device *dev, 7517 char *name, 7518 struct list_head *dev_list) 7519 { 7520 char linkname[IFNAMSIZ+7]; 7521 7522 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7523 "upper_%s" : "lower_%s", name); 7524 sysfs_remove_link(&(dev->dev.kobj), linkname); 7525 } 7526 7527 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, 7528 struct net_device *adj_dev, 7529 struct list_head *dev_list) 7530 { 7531 return (dev_list == &dev->adj_list.upper || 7532 dev_list == &dev->adj_list.lower) && 7533 net_eq(dev_net(dev), dev_net(adj_dev)); 7534 } 7535 7536 static int __netdev_adjacent_dev_insert(struct net_device *dev, 7537 struct net_device *adj_dev, 7538 struct list_head *dev_list, 7539 void *private, bool master) 7540 { 7541 struct netdev_adjacent *adj; 7542 int ret; 7543 7544 adj = __netdev_find_adj(adj_dev, dev_list); 7545 7546 if (adj) { 7547 adj->ref_nr += 1; 7548 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", 7549 dev->name, adj_dev->name, adj->ref_nr); 7550 7551 return 0; 7552 } 7553 7554 adj = kmalloc(sizeof(*adj), GFP_KERNEL); 7555 if (!adj) 7556 return -ENOMEM; 7557 7558 adj->dev = adj_dev; 7559 adj->master = master; 7560 adj->ref_nr = 1; 7561 adj->private = private; 7562 adj->ignore = false; 7563 netdev_hold(adj_dev, &adj->dev_tracker, GFP_KERNEL); 7564 7565 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", 7566 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); 7567 7568 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { 7569 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); 7570 if (ret) 7571 goto free_adj; 7572 } 7573 7574 /* Ensure that master link is always the first item in list. */ 7575 if (master) { 7576 ret = sysfs_create_link(&(dev->dev.kobj), 7577 &(adj_dev->dev.kobj), "master"); 7578 if (ret) 7579 goto remove_symlinks; 7580 7581 list_add_rcu(&adj->list, dev_list); 7582 } else { 7583 list_add_tail_rcu(&adj->list, dev_list); 7584 } 7585 7586 return 0; 7587 7588 remove_symlinks: 7589 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7590 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7591 free_adj: 7592 netdev_put(adj_dev, &adj->dev_tracker); 7593 kfree(adj); 7594 7595 return ret; 7596 } 7597 7598 static void __netdev_adjacent_dev_remove(struct net_device *dev, 7599 struct net_device *adj_dev, 7600 u16 ref_nr, 7601 struct list_head *dev_list) 7602 { 7603 struct netdev_adjacent *adj; 7604 7605 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", 7606 dev->name, adj_dev->name, ref_nr); 7607 7608 adj = __netdev_find_adj(adj_dev, dev_list); 7609 7610 if (!adj) { 7611 pr_err("Adjacency does not exist for device %s from %s\n", 7612 dev->name, adj_dev->name); 7613 WARN_ON(1); 7614 return; 7615 } 7616 7617 if (adj->ref_nr > ref_nr) { 7618 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", 7619 dev->name, adj_dev->name, ref_nr, 7620 adj->ref_nr - ref_nr); 7621 adj->ref_nr -= ref_nr; 7622 return; 7623 } 7624 7625 if (adj->master) 7626 sysfs_remove_link(&(dev->dev.kobj), "master"); 7627 7628 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7629 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7630 7631 list_del_rcu(&adj->list); 7632 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", 7633 adj_dev->name, dev->name, adj_dev->name); 7634 netdev_put(adj_dev, &adj->dev_tracker); 7635 kfree_rcu(adj, rcu); 7636 } 7637 7638 static int __netdev_adjacent_dev_link_lists(struct net_device *dev, 7639 struct net_device *upper_dev, 7640 struct list_head *up_list, 7641 struct list_head *down_list, 7642 void *private, bool master) 7643 { 7644 int ret; 7645 7646 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, 7647 private, master); 7648 if (ret) 7649 return ret; 7650 7651 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, 7652 private, false); 7653 if (ret) { 7654 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); 7655 return ret; 7656 } 7657 7658 return 0; 7659 } 7660 7661 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, 7662 struct net_device *upper_dev, 7663 u16 ref_nr, 7664 struct list_head *up_list, 7665 struct list_head *down_list) 7666 { 7667 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); 7668 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); 7669 } 7670 7671 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, 7672 struct net_device *upper_dev, 7673 void *private, bool master) 7674 { 7675 return __netdev_adjacent_dev_link_lists(dev, upper_dev, 7676 &dev->adj_list.upper, 7677 &upper_dev->adj_list.lower, 7678 private, master); 7679 } 7680 7681 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, 7682 struct net_device *upper_dev) 7683 { 7684 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, 7685 &dev->adj_list.upper, 7686 &upper_dev->adj_list.lower); 7687 } 7688 7689 static int __netdev_upper_dev_link(struct net_device *dev, 7690 struct net_device *upper_dev, bool master, 7691 void *upper_priv, void *upper_info, 7692 struct netdev_nested_priv *priv, 7693 struct netlink_ext_ack *extack) 7694 { 7695 struct netdev_notifier_changeupper_info changeupper_info = { 7696 .info = { 7697 .dev = dev, 7698 .extack = extack, 7699 }, 7700 .upper_dev = upper_dev, 7701 .master = master, 7702 .linking = true, 7703 .upper_info = upper_info, 7704 }; 7705 struct net_device *master_dev; 7706 int ret = 0; 7707 7708 ASSERT_RTNL(); 7709 7710 if (dev == upper_dev) 7711 return -EBUSY; 7712 7713 /* To prevent loops, check if dev is not upper device to upper_dev. */ 7714 if (__netdev_has_upper_dev(upper_dev, dev)) 7715 return -EBUSY; 7716 7717 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV) 7718 return -EMLINK; 7719 7720 if (!master) { 7721 if (__netdev_has_upper_dev(dev, upper_dev)) 7722 return -EEXIST; 7723 } else { 7724 master_dev = __netdev_master_upper_dev_get(dev); 7725 if (master_dev) 7726 return master_dev == upper_dev ? -EEXIST : -EBUSY; 7727 } 7728 7729 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7730 &changeupper_info.info); 7731 ret = notifier_to_errno(ret); 7732 if (ret) 7733 return ret; 7734 7735 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, 7736 master); 7737 if (ret) 7738 return ret; 7739 7740 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 7741 &changeupper_info.info); 7742 ret = notifier_to_errno(ret); 7743 if (ret) 7744 goto rollback; 7745 7746 __netdev_update_upper_level(dev, NULL); 7747 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 7748 7749 __netdev_update_lower_level(upper_dev, priv); 7750 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 7751 priv); 7752 7753 return 0; 7754 7755 rollback: 7756 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 7757 7758 return ret; 7759 } 7760 7761 /** 7762 * netdev_upper_dev_link - Add a link to the upper device 7763 * @dev: device 7764 * @upper_dev: new upper device 7765 * @extack: netlink extended ack 7766 * 7767 * Adds a link to device which is upper to this one. The caller must hold 7768 * the RTNL lock. On a failure a negative errno code is returned. 7769 * On success the reference counts are adjusted and the function 7770 * returns zero. 7771 */ 7772 int netdev_upper_dev_link(struct net_device *dev, 7773 struct net_device *upper_dev, 7774 struct netlink_ext_ack *extack) 7775 { 7776 struct netdev_nested_priv priv = { 7777 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7778 .data = NULL, 7779 }; 7780 7781 return __netdev_upper_dev_link(dev, upper_dev, false, 7782 NULL, NULL, &priv, extack); 7783 } 7784 EXPORT_SYMBOL(netdev_upper_dev_link); 7785 7786 /** 7787 * netdev_master_upper_dev_link - Add a master link to the upper device 7788 * @dev: device 7789 * @upper_dev: new upper device 7790 * @upper_priv: upper device private 7791 * @upper_info: upper info to be passed down via notifier 7792 * @extack: netlink extended ack 7793 * 7794 * Adds a link to device which is upper to this one. In this case, only 7795 * one master upper device can be linked, although other non-master devices 7796 * might be linked as well. The caller must hold the RTNL lock. 7797 * On a failure a negative errno code is returned. On success the reference 7798 * counts are adjusted and the function returns zero. 7799 */ 7800 int netdev_master_upper_dev_link(struct net_device *dev, 7801 struct net_device *upper_dev, 7802 void *upper_priv, void *upper_info, 7803 struct netlink_ext_ack *extack) 7804 { 7805 struct netdev_nested_priv priv = { 7806 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7807 .data = NULL, 7808 }; 7809 7810 return __netdev_upper_dev_link(dev, upper_dev, true, 7811 upper_priv, upper_info, &priv, extack); 7812 } 7813 EXPORT_SYMBOL(netdev_master_upper_dev_link); 7814 7815 static void __netdev_upper_dev_unlink(struct net_device *dev, 7816 struct net_device *upper_dev, 7817 struct netdev_nested_priv *priv) 7818 { 7819 struct netdev_notifier_changeupper_info changeupper_info = { 7820 .info = { 7821 .dev = dev, 7822 }, 7823 .upper_dev = upper_dev, 7824 .linking = false, 7825 }; 7826 7827 ASSERT_RTNL(); 7828 7829 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; 7830 7831 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7832 &changeupper_info.info); 7833 7834 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 7835 7836 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 7837 &changeupper_info.info); 7838 7839 __netdev_update_upper_level(dev, NULL); 7840 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 7841 7842 __netdev_update_lower_level(upper_dev, priv); 7843 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 7844 priv); 7845 } 7846 7847 /** 7848 * netdev_upper_dev_unlink - Removes a link to upper device 7849 * @dev: device 7850 * @upper_dev: new upper device 7851 * 7852 * Removes a link to device which is upper to this one. The caller must hold 7853 * the RTNL lock. 7854 */ 7855 void netdev_upper_dev_unlink(struct net_device *dev, 7856 struct net_device *upper_dev) 7857 { 7858 struct netdev_nested_priv priv = { 7859 .flags = NESTED_SYNC_TODO, 7860 .data = NULL, 7861 }; 7862 7863 __netdev_upper_dev_unlink(dev, upper_dev, &priv); 7864 } 7865 EXPORT_SYMBOL(netdev_upper_dev_unlink); 7866 7867 static void __netdev_adjacent_dev_set(struct net_device *upper_dev, 7868 struct net_device *lower_dev, 7869 bool val) 7870 { 7871 struct netdev_adjacent *adj; 7872 7873 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower); 7874 if (adj) 7875 adj->ignore = val; 7876 7877 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper); 7878 if (adj) 7879 adj->ignore = val; 7880 } 7881 7882 static void netdev_adjacent_dev_disable(struct net_device *upper_dev, 7883 struct net_device *lower_dev) 7884 { 7885 __netdev_adjacent_dev_set(upper_dev, lower_dev, true); 7886 } 7887 7888 static void netdev_adjacent_dev_enable(struct net_device *upper_dev, 7889 struct net_device *lower_dev) 7890 { 7891 __netdev_adjacent_dev_set(upper_dev, lower_dev, false); 7892 } 7893 7894 int netdev_adjacent_change_prepare(struct net_device *old_dev, 7895 struct net_device *new_dev, 7896 struct net_device *dev, 7897 struct netlink_ext_ack *extack) 7898 { 7899 struct netdev_nested_priv priv = { 7900 .flags = 0, 7901 .data = NULL, 7902 }; 7903 int err; 7904 7905 if (!new_dev) 7906 return 0; 7907 7908 if (old_dev && new_dev != old_dev) 7909 netdev_adjacent_dev_disable(dev, old_dev); 7910 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv, 7911 extack); 7912 if (err) { 7913 if (old_dev && new_dev != old_dev) 7914 netdev_adjacent_dev_enable(dev, old_dev); 7915 return err; 7916 } 7917 7918 return 0; 7919 } 7920 EXPORT_SYMBOL(netdev_adjacent_change_prepare); 7921 7922 void netdev_adjacent_change_commit(struct net_device *old_dev, 7923 struct net_device *new_dev, 7924 struct net_device *dev) 7925 { 7926 struct netdev_nested_priv priv = { 7927 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7928 .data = NULL, 7929 }; 7930 7931 if (!new_dev || !old_dev) 7932 return; 7933 7934 if (new_dev == old_dev) 7935 return; 7936 7937 netdev_adjacent_dev_enable(dev, old_dev); 7938 __netdev_upper_dev_unlink(old_dev, dev, &priv); 7939 } 7940 EXPORT_SYMBOL(netdev_adjacent_change_commit); 7941 7942 void netdev_adjacent_change_abort(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 = 0, 7948 .data = NULL, 7949 }; 7950 7951 if (!new_dev) 7952 return; 7953 7954 if (old_dev && new_dev != old_dev) 7955 netdev_adjacent_dev_enable(dev, old_dev); 7956 7957 __netdev_upper_dev_unlink(new_dev, dev, &priv); 7958 } 7959 EXPORT_SYMBOL(netdev_adjacent_change_abort); 7960 7961 /** 7962 * netdev_bonding_info_change - Dispatch event about slave change 7963 * @dev: device 7964 * @bonding_info: info to dispatch 7965 * 7966 * Send NETDEV_BONDING_INFO to netdev notifiers with info. 7967 * The caller must hold the RTNL lock. 7968 */ 7969 void netdev_bonding_info_change(struct net_device *dev, 7970 struct netdev_bonding_info *bonding_info) 7971 { 7972 struct netdev_notifier_bonding_info info = { 7973 .info.dev = dev, 7974 }; 7975 7976 memcpy(&info.bonding_info, bonding_info, 7977 sizeof(struct netdev_bonding_info)); 7978 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, 7979 &info.info); 7980 } 7981 EXPORT_SYMBOL(netdev_bonding_info_change); 7982 7983 static int netdev_offload_xstats_enable_l3(struct net_device *dev, 7984 struct netlink_ext_ack *extack) 7985 { 7986 struct netdev_notifier_offload_xstats_info info = { 7987 .info.dev = dev, 7988 .info.extack = extack, 7989 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, 7990 }; 7991 int err; 7992 int rc; 7993 7994 dev->offload_xstats_l3 = kzalloc(sizeof(*dev->offload_xstats_l3), 7995 GFP_KERNEL); 7996 if (!dev->offload_xstats_l3) 7997 return -ENOMEM; 7998 7999 rc = call_netdevice_notifiers_info_robust(NETDEV_OFFLOAD_XSTATS_ENABLE, 8000 NETDEV_OFFLOAD_XSTATS_DISABLE, 8001 &info.info); 8002 err = notifier_to_errno(rc); 8003 if (err) 8004 goto free_stats; 8005 8006 return 0; 8007 8008 free_stats: 8009 kfree(dev->offload_xstats_l3); 8010 dev->offload_xstats_l3 = NULL; 8011 return err; 8012 } 8013 8014 int netdev_offload_xstats_enable(struct net_device *dev, 8015 enum netdev_offload_xstats_type type, 8016 struct netlink_ext_ack *extack) 8017 { 8018 ASSERT_RTNL(); 8019 8020 if (netdev_offload_xstats_enabled(dev, type)) 8021 return -EALREADY; 8022 8023 switch (type) { 8024 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 8025 return netdev_offload_xstats_enable_l3(dev, extack); 8026 } 8027 8028 WARN_ON(1); 8029 return -EINVAL; 8030 } 8031 EXPORT_SYMBOL(netdev_offload_xstats_enable); 8032 8033 static void netdev_offload_xstats_disable_l3(struct net_device *dev) 8034 { 8035 struct netdev_notifier_offload_xstats_info info = { 8036 .info.dev = dev, 8037 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, 8038 }; 8039 8040 call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_DISABLE, 8041 &info.info); 8042 kfree(dev->offload_xstats_l3); 8043 dev->offload_xstats_l3 = NULL; 8044 } 8045 8046 int netdev_offload_xstats_disable(struct net_device *dev, 8047 enum netdev_offload_xstats_type type) 8048 { 8049 ASSERT_RTNL(); 8050 8051 if (!netdev_offload_xstats_enabled(dev, type)) 8052 return -EALREADY; 8053 8054 switch (type) { 8055 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 8056 netdev_offload_xstats_disable_l3(dev); 8057 return 0; 8058 } 8059 8060 WARN_ON(1); 8061 return -EINVAL; 8062 } 8063 EXPORT_SYMBOL(netdev_offload_xstats_disable); 8064 8065 static void netdev_offload_xstats_disable_all(struct net_device *dev) 8066 { 8067 netdev_offload_xstats_disable(dev, NETDEV_OFFLOAD_XSTATS_TYPE_L3); 8068 } 8069 8070 static struct rtnl_hw_stats64 * 8071 netdev_offload_xstats_get_ptr(const struct net_device *dev, 8072 enum netdev_offload_xstats_type type) 8073 { 8074 switch (type) { 8075 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 8076 return dev->offload_xstats_l3; 8077 } 8078 8079 WARN_ON(1); 8080 return NULL; 8081 } 8082 8083 bool netdev_offload_xstats_enabled(const struct net_device *dev, 8084 enum netdev_offload_xstats_type type) 8085 { 8086 ASSERT_RTNL(); 8087 8088 return netdev_offload_xstats_get_ptr(dev, type); 8089 } 8090 EXPORT_SYMBOL(netdev_offload_xstats_enabled); 8091 8092 struct netdev_notifier_offload_xstats_ru { 8093 bool used; 8094 }; 8095 8096 struct netdev_notifier_offload_xstats_rd { 8097 struct rtnl_hw_stats64 stats; 8098 bool used; 8099 }; 8100 8101 static void netdev_hw_stats64_add(struct rtnl_hw_stats64 *dest, 8102 const struct rtnl_hw_stats64 *src) 8103 { 8104 dest->rx_packets += src->rx_packets; 8105 dest->tx_packets += src->tx_packets; 8106 dest->rx_bytes += src->rx_bytes; 8107 dest->tx_bytes += src->tx_bytes; 8108 dest->rx_errors += src->rx_errors; 8109 dest->tx_errors += src->tx_errors; 8110 dest->rx_dropped += src->rx_dropped; 8111 dest->tx_dropped += src->tx_dropped; 8112 dest->multicast += src->multicast; 8113 } 8114 8115 static int netdev_offload_xstats_get_used(struct net_device *dev, 8116 enum netdev_offload_xstats_type type, 8117 bool *p_used, 8118 struct netlink_ext_ack *extack) 8119 { 8120 struct netdev_notifier_offload_xstats_ru report_used = {}; 8121 struct netdev_notifier_offload_xstats_info info = { 8122 .info.dev = dev, 8123 .info.extack = extack, 8124 .type = type, 8125 .report_used = &report_used, 8126 }; 8127 int rc; 8128 8129 WARN_ON(!netdev_offload_xstats_enabled(dev, type)); 8130 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_USED, 8131 &info.info); 8132 *p_used = report_used.used; 8133 return notifier_to_errno(rc); 8134 } 8135 8136 static int netdev_offload_xstats_get_stats(struct net_device *dev, 8137 enum netdev_offload_xstats_type type, 8138 struct rtnl_hw_stats64 *p_stats, 8139 bool *p_used, 8140 struct netlink_ext_ack *extack) 8141 { 8142 struct netdev_notifier_offload_xstats_rd report_delta = {}; 8143 struct netdev_notifier_offload_xstats_info info = { 8144 .info.dev = dev, 8145 .info.extack = extack, 8146 .type = type, 8147 .report_delta = &report_delta, 8148 }; 8149 struct rtnl_hw_stats64 *stats; 8150 int rc; 8151 8152 stats = netdev_offload_xstats_get_ptr(dev, type); 8153 if (WARN_ON(!stats)) 8154 return -EINVAL; 8155 8156 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_DELTA, 8157 &info.info); 8158 8159 /* Cache whatever we got, even if there was an error, otherwise the 8160 * successful stats retrievals would get lost. 8161 */ 8162 netdev_hw_stats64_add(stats, &report_delta.stats); 8163 8164 if (p_stats) 8165 *p_stats = *stats; 8166 *p_used = report_delta.used; 8167 8168 return notifier_to_errno(rc); 8169 } 8170 8171 int netdev_offload_xstats_get(struct net_device *dev, 8172 enum netdev_offload_xstats_type type, 8173 struct rtnl_hw_stats64 *p_stats, bool *p_used, 8174 struct netlink_ext_ack *extack) 8175 { 8176 ASSERT_RTNL(); 8177 8178 if (p_stats) 8179 return netdev_offload_xstats_get_stats(dev, type, p_stats, 8180 p_used, extack); 8181 else 8182 return netdev_offload_xstats_get_used(dev, type, p_used, 8183 extack); 8184 } 8185 EXPORT_SYMBOL(netdev_offload_xstats_get); 8186 8187 void 8188 netdev_offload_xstats_report_delta(struct netdev_notifier_offload_xstats_rd *report_delta, 8189 const struct rtnl_hw_stats64 *stats) 8190 { 8191 report_delta->used = true; 8192 netdev_hw_stats64_add(&report_delta->stats, stats); 8193 } 8194 EXPORT_SYMBOL(netdev_offload_xstats_report_delta); 8195 8196 void 8197 netdev_offload_xstats_report_used(struct netdev_notifier_offload_xstats_ru *report_used) 8198 { 8199 report_used->used = true; 8200 } 8201 EXPORT_SYMBOL(netdev_offload_xstats_report_used); 8202 8203 void netdev_offload_xstats_push_delta(struct net_device *dev, 8204 enum netdev_offload_xstats_type type, 8205 const struct rtnl_hw_stats64 *p_stats) 8206 { 8207 struct rtnl_hw_stats64 *stats; 8208 8209 ASSERT_RTNL(); 8210 8211 stats = netdev_offload_xstats_get_ptr(dev, type); 8212 if (WARN_ON(!stats)) 8213 return; 8214 8215 netdev_hw_stats64_add(stats, p_stats); 8216 } 8217 EXPORT_SYMBOL(netdev_offload_xstats_push_delta); 8218 8219 /** 8220 * netdev_get_xmit_slave - Get the xmit slave of master device 8221 * @dev: device 8222 * @skb: The packet 8223 * @all_slaves: assume all the slaves are active 8224 * 8225 * The reference counters are not incremented so the caller must be 8226 * careful with locks. The caller must hold RCU lock. 8227 * %NULL is returned if no slave is found. 8228 */ 8229 8230 struct net_device *netdev_get_xmit_slave(struct net_device *dev, 8231 struct sk_buff *skb, 8232 bool all_slaves) 8233 { 8234 const struct net_device_ops *ops = dev->netdev_ops; 8235 8236 if (!ops->ndo_get_xmit_slave) 8237 return NULL; 8238 return ops->ndo_get_xmit_slave(dev, skb, all_slaves); 8239 } 8240 EXPORT_SYMBOL(netdev_get_xmit_slave); 8241 8242 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev, 8243 struct sock *sk) 8244 { 8245 const struct net_device_ops *ops = dev->netdev_ops; 8246 8247 if (!ops->ndo_sk_get_lower_dev) 8248 return NULL; 8249 return ops->ndo_sk_get_lower_dev(dev, sk); 8250 } 8251 8252 /** 8253 * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket 8254 * @dev: device 8255 * @sk: the socket 8256 * 8257 * %NULL is returned if no lower device is found. 8258 */ 8259 8260 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev, 8261 struct sock *sk) 8262 { 8263 struct net_device *lower; 8264 8265 lower = netdev_sk_get_lower_dev(dev, sk); 8266 while (lower) { 8267 dev = lower; 8268 lower = netdev_sk_get_lower_dev(dev, sk); 8269 } 8270 8271 return dev; 8272 } 8273 EXPORT_SYMBOL(netdev_sk_get_lowest_dev); 8274 8275 static void netdev_adjacent_add_links(struct net_device *dev) 8276 { 8277 struct netdev_adjacent *iter; 8278 8279 struct net *net = dev_net(dev); 8280 8281 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8282 if (!net_eq(net, dev_net(iter->dev))) 8283 continue; 8284 netdev_adjacent_sysfs_add(iter->dev, dev, 8285 &iter->dev->adj_list.lower); 8286 netdev_adjacent_sysfs_add(dev, iter->dev, 8287 &dev->adj_list.upper); 8288 } 8289 8290 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8291 if (!net_eq(net, dev_net(iter->dev))) 8292 continue; 8293 netdev_adjacent_sysfs_add(iter->dev, dev, 8294 &iter->dev->adj_list.upper); 8295 netdev_adjacent_sysfs_add(dev, iter->dev, 8296 &dev->adj_list.lower); 8297 } 8298 } 8299 8300 static void netdev_adjacent_del_links(struct net_device *dev) 8301 { 8302 struct netdev_adjacent *iter; 8303 8304 struct net *net = dev_net(dev); 8305 8306 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8307 if (!net_eq(net, dev_net(iter->dev))) 8308 continue; 8309 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8310 &iter->dev->adj_list.lower); 8311 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8312 &dev->adj_list.upper); 8313 } 8314 8315 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8316 if (!net_eq(net, dev_net(iter->dev))) 8317 continue; 8318 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8319 &iter->dev->adj_list.upper); 8320 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8321 &dev->adj_list.lower); 8322 } 8323 } 8324 8325 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) 8326 { 8327 struct netdev_adjacent *iter; 8328 8329 struct net *net = dev_net(dev); 8330 8331 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8332 if (!net_eq(net, dev_net(iter->dev))) 8333 continue; 8334 netdev_adjacent_sysfs_del(iter->dev, oldname, 8335 &iter->dev->adj_list.lower); 8336 netdev_adjacent_sysfs_add(iter->dev, dev, 8337 &iter->dev->adj_list.lower); 8338 } 8339 8340 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8341 if (!net_eq(net, dev_net(iter->dev))) 8342 continue; 8343 netdev_adjacent_sysfs_del(iter->dev, oldname, 8344 &iter->dev->adj_list.upper); 8345 netdev_adjacent_sysfs_add(iter->dev, dev, 8346 &iter->dev->adj_list.upper); 8347 } 8348 } 8349 8350 void *netdev_lower_dev_get_private(struct net_device *dev, 8351 struct net_device *lower_dev) 8352 { 8353 struct netdev_adjacent *lower; 8354 8355 if (!lower_dev) 8356 return NULL; 8357 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); 8358 if (!lower) 8359 return NULL; 8360 8361 return lower->private; 8362 } 8363 EXPORT_SYMBOL(netdev_lower_dev_get_private); 8364 8365 8366 /** 8367 * netdev_lower_state_changed - Dispatch event about lower device state change 8368 * @lower_dev: device 8369 * @lower_state_info: state to dispatch 8370 * 8371 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. 8372 * The caller must hold the RTNL lock. 8373 */ 8374 void netdev_lower_state_changed(struct net_device *lower_dev, 8375 void *lower_state_info) 8376 { 8377 struct netdev_notifier_changelowerstate_info changelowerstate_info = { 8378 .info.dev = lower_dev, 8379 }; 8380 8381 ASSERT_RTNL(); 8382 changelowerstate_info.lower_state_info = lower_state_info; 8383 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, 8384 &changelowerstate_info.info); 8385 } 8386 EXPORT_SYMBOL(netdev_lower_state_changed); 8387 8388 static void dev_change_rx_flags(struct net_device *dev, int flags) 8389 { 8390 const struct net_device_ops *ops = dev->netdev_ops; 8391 8392 if (ops->ndo_change_rx_flags) 8393 ops->ndo_change_rx_flags(dev, flags); 8394 } 8395 8396 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) 8397 { 8398 unsigned int old_flags = dev->flags; 8399 kuid_t uid; 8400 kgid_t gid; 8401 8402 ASSERT_RTNL(); 8403 8404 dev->flags |= IFF_PROMISC; 8405 dev->promiscuity += inc; 8406 if (dev->promiscuity == 0) { 8407 /* 8408 * Avoid overflow. 8409 * If inc causes overflow, untouch promisc and return error. 8410 */ 8411 if (inc < 0) 8412 dev->flags &= ~IFF_PROMISC; 8413 else { 8414 dev->promiscuity -= inc; 8415 netdev_warn(dev, "promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n"); 8416 return -EOVERFLOW; 8417 } 8418 } 8419 if (dev->flags != old_flags) { 8420 netdev_info(dev, "%s promiscuous mode\n", 8421 dev->flags & IFF_PROMISC ? "entered" : "left"); 8422 if (audit_enabled) { 8423 current_uid_gid(&uid, &gid); 8424 audit_log(audit_context(), GFP_ATOMIC, 8425 AUDIT_ANOM_PROMISCUOUS, 8426 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", 8427 dev->name, (dev->flags & IFF_PROMISC), 8428 (old_flags & IFF_PROMISC), 8429 from_kuid(&init_user_ns, audit_get_loginuid(current)), 8430 from_kuid(&init_user_ns, uid), 8431 from_kgid(&init_user_ns, gid), 8432 audit_get_sessionid(current)); 8433 } 8434 8435 dev_change_rx_flags(dev, IFF_PROMISC); 8436 } 8437 if (notify) 8438 __dev_notify_flags(dev, old_flags, IFF_PROMISC, 0, NULL); 8439 return 0; 8440 } 8441 8442 /** 8443 * dev_set_promiscuity - update promiscuity count on a device 8444 * @dev: device 8445 * @inc: modifier 8446 * 8447 * Add or remove promiscuity from a device. While the count in the device 8448 * remains above zero the interface remains promiscuous. Once it hits zero 8449 * the device reverts back to normal filtering operation. A negative inc 8450 * value is used to drop promiscuity on the device. 8451 * Return 0 if successful or a negative errno code on error. 8452 */ 8453 int dev_set_promiscuity(struct net_device *dev, int inc) 8454 { 8455 unsigned int old_flags = dev->flags; 8456 int err; 8457 8458 err = __dev_set_promiscuity(dev, inc, true); 8459 if (err < 0) 8460 return err; 8461 if (dev->flags != old_flags) 8462 dev_set_rx_mode(dev); 8463 return err; 8464 } 8465 EXPORT_SYMBOL(dev_set_promiscuity); 8466 8467 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) 8468 { 8469 unsigned int old_flags = dev->flags, old_gflags = dev->gflags; 8470 8471 ASSERT_RTNL(); 8472 8473 dev->flags |= IFF_ALLMULTI; 8474 dev->allmulti += inc; 8475 if (dev->allmulti == 0) { 8476 /* 8477 * Avoid overflow. 8478 * If inc causes overflow, untouch allmulti and return error. 8479 */ 8480 if (inc < 0) 8481 dev->flags &= ~IFF_ALLMULTI; 8482 else { 8483 dev->allmulti -= inc; 8484 netdev_warn(dev, "allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n"); 8485 return -EOVERFLOW; 8486 } 8487 } 8488 if (dev->flags ^ old_flags) { 8489 netdev_info(dev, "%s allmulticast mode\n", 8490 dev->flags & IFF_ALLMULTI ? "entered" : "left"); 8491 dev_change_rx_flags(dev, IFF_ALLMULTI); 8492 dev_set_rx_mode(dev); 8493 if (notify) 8494 __dev_notify_flags(dev, old_flags, 8495 dev->gflags ^ old_gflags, 0, NULL); 8496 } 8497 return 0; 8498 } 8499 8500 /** 8501 * dev_set_allmulti - update allmulti count on a device 8502 * @dev: device 8503 * @inc: modifier 8504 * 8505 * Add or remove reception of all multicast frames to a device. While the 8506 * count in the device remains above zero the interface remains listening 8507 * to all interfaces. Once it hits zero the device reverts back to normal 8508 * filtering operation. A negative @inc value is used to drop the counter 8509 * when releasing a resource needing all multicasts. 8510 * Return 0 if successful or a negative errno code on error. 8511 */ 8512 8513 int dev_set_allmulti(struct net_device *dev, int inc) 8514 { 8515 return __dev_set_allmulti(dev, inc, true); 8516 } 8517 EXPORT_SYMBOL(dev_set_allmulti); 8518 8519 /* 8520 * Upload unicast and multicast address lists to device and 8521 * configure RX filtering. When the device doesn't support unicast 8522 * filtering it is put in promiscuous mode while unicast addresses 8523 * are present. 8524 */ 8525 void __dev_set_rx_mode(struct net_device *dev) 8526 { 8527 const struct net_device_ops *ops = dev->netdev_ops; 8528 8529 /* dev_open will call this function so the list will stay sane. */ 8530 if (!(dev->flags&IFF_UP)) 8531 return; 8532 8533 if (!netif_device_present(dev)) 8534 return; 8535 8536 if (!(dev->priv_flags & IFF_UNICAST_FLT)) { 8537 /* Unicast addresses changes may only happen under the rtnl, 8538 * therefore calling __dev_set_promiscuity here is safe. 8539 */ 8540 if (!netdev_uc_empty(dev) && !dev->uc_promisc) { 8541 __dev_set_promiscuity(dev, 1, false); 8542 dev->uc_promisc = true; 8543 } else if (netdev_uc_empty(dev) && dev->uc_promisc) { 8544 __dev_set_promiscuity(dev, -1, false); 8545 dev->uc_promisc = false; 8546 } 8547 } 8548 8549 if (ops->ndo_set_rx_mode) 8550 ops->ndo_set_rx_mode(dev); 8551 } 8552 8553 void dev_set_rx_mode(struct net_device *dev) 8554 { 8555 netif_addr_lock_bh(dev); 8556 __dev_set_rx_mode(dev); 8557 netif_addr_unlock_bh(dev); 8558 } 8559 8560 /** 8561 * dev_get_flags - get flags reported to userspace 8562 * @dev: device 8563 * 8564 * Get the combination of flag bits exported through APIs to userspace. 8565 */ 8566 unsigned int dev_get_flags(const struct net_device *dev) 8567 { 8568 unsigned int flags; 8569 8570 flags = (dev->flags & ~(IFF_PROMISC | 8571 IFF_ALLMULTI | 8572 IFF_RUNNING | 8573 IFF_LOWER_UP | 8574 IFF_DORMANT)) | 8575 (dev->gflags & (IFF_PROMISC | 8576 IFF_ALLMULTI)); 8577 8578 if (netif_running(dev)) { 8579 if (netif_oper_up(dev)) 8580 flags |= IFF_RUNNING; 8581 if (netif_carrier_ok(dev)) 8582 flags |= IFF_LOWER_UP; 8583 if (netif_dormant(dev)) 8584 flags |= IFF_DORMANT; 8585 } 8586 8587 return flags; 8588 } 8589 EXPORT_SYMBOL(dev_get_flags); 8590 8591 int __dev_change_flags(struct net_device *dev, unsigned int flags, 8592 struct netlink_ext_ack *extack) 8593 { 8594 unsigned int old_flags = dev->flags; 8595 int ret; 8596 8597 ASSERT_RTNL(); 8598 8599 /* 8600 * Set the flags on our device. 8601 */ 8602 8603 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | 8604 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | 8605 IFF_AUTOMEDIA)) | 8606 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | 8607 IFF_ALLMULTI)); 8608 8609 /* 8610 * Load in the correct multicast list now the flags have changed. 8611 */ 8612 8613 if ((old_flags ^ flags) & IFF_MULTICAST) 8614 dev_change_rx_flags(dev, IFF_MULTICAST); 8615 8616 dev_set_rx_mode(dev); 8617 8618 /* 8619 * Have we downed the interface. We handle IFF_UP ourselves 8620 * according to user attempts to set it, rather than blindly 8621 * setting it. 8622 */ 8623 8624 ret = 0; 8625 if ((old_flags ^ flags) & IFF_UP) { 8626 if (old_flags & IFF_UP) 8627 __dev_close(dev); 8628 else 8629 ret = __dev_open(dev, extack); 8630 } 8631 8632 if ((flags ^ dev->gflags) & IFF_PROMISC) { 8633 int inc = (flags & IFF_PROMISC) ? 1 : -1; 8634 unsigned int old_flags = dev->flags; 8635 8636 dev->gflags ^= IFF_PROMISC; 8637 8638 if (__dev_set_promiscuity(dev, inc, false) >= 0) 8639 if (dev->flags != old_flags) 8640 dev_set_rx_mode(dev); 8641 } 8642 8643 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI 8644 * is important. Some (broken) drivers set IFF_PROMISC, when 8645 * IFF_ALLMULTI is requested not asking us and not reporting. 8646 */ 8647 if ((flags ^ dev->gflags) & IFF_ALLMULTI) { 8648 int inc = (flags & IFF_ALLMULTI) ? 1 : -1; 8649 8650 dev->gflags ^= IFF_ALLMULTI; 8651 __dev_set_allmulti(dev, inc, false); 8652 } 8653 8654 return ret; 8655 } 8656 8657 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, 8658 unsigned int gchanges, u32 portid, 8659 const struct nlmsghdr *nlh) 8660 { 8661 unsigned int changes = dev->flags ^ old_flags; 8662 8663 if (gchanges) 8664 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC, portid, nlh); 8665 8666 if (changes & IFF_UP) { 8667 if (dev->flags & IFF_UP) 8668 call_netdevice_notifiers(NETDEV_UP, dev); 8669 else 8670 call_netdevice_notifiers(NETDEV_DOWN, dev); 8671 } 8672 8673 if (dev->flags & IFF_UP && 8674 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { 8675 struct netdev_notifier_change_info change_info = { 8676 .info = { 8677 .dev = dev, 8678 }, 8679 .flags_changed = changes, 8680 }; 8681 8682 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info); 8683 } 8684 } 8685 8686 /** 8687 * dev_change_flags - change device settings 8688 * @dev: device 8689 * @flags: device state flags 8690 * @extack: netlink extended ack 8691 * 8692 * Change settings on device based state flags. The flags are 8693 * in the userspace exported format. 8694 */ 8695 int dev_change_flags(struct net_device *dev, unsigned int flags, 8696 struct netlink_ext_ack *extack) 8697 { 8698 int ret; 8699 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; 8700 8701 ret = __dev_change_flags(dev, flags, extack); 8702 if (ret < 0) 8703 return ret; 8704 8705 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); 8706 __dev_notify_flags(dev, old_flags, changes, 0, NULL); 8707 return ret; 8708 } 8709 EXPORT_SYMBOL(dev_change_flags); 8710 8711 int __dev_set_mtu(struct net_device *dev, int new_mtu) 8712 { 8713 const struct net_device_ops *ops = dev->netdev_ops; 8714 8715 if (ops->ndo_change_mtu) 8716 return ops->ndo_change_mtu(dev, new_mtu); 8717 8718 /* Pairs with all the lockless reads of dev->mtu in the stack */ 8719 WRITE_ONCE(dev->mtu, new_mtu); 8720 return 0; 8721 } 8722 EXPORT_SYMBOL(__dev_set_mtu); 8723 8724 int dev_validate_mtu(struct net_device *dev, int new_mtu, 8725 struct netlink_ext_ack *extack) 8726 { 8727 /* MTU must be positive, and in range */ 8728 if (new_mtu < 0 || new_mtu < dev->min_mtu) { 8729 NL_SET_ERR_MSG(extack, "mtu less than device minimum"); 8730 return -EINVAL; 8731 } 8732 8733 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { 8734 NL_SET_ERR_MSG(extack, "mtu greater than device maximum"); 8735 return -EINVAL; 8736 } 8737 return 0; 8738 } 8739 8740 /** 8741 * dev_set_mtu_ext - Change maximum transfer unit 8742 * @dev: device 8743 * @new_mtu: new transfer unit 8744 * @extack: netlink extended ack 8745 * 8746 * Change the maximum transfer size of the network device. 8747 */ 8748 int dev_set_mtu_ext(struct net_device *dev, int new_mtu, 8749 struct netlink_ext_ack *extack) 8750 { 8751 int err, orig_mtu; 8752 8753 if (new_mtu == dev->mtu) 8754 return 0; 8755 8756 err = dev_validate_mtu(dev, new_mtu, extack); 8757 if (err) 8758 return err; 8759 8760 if (!netif_device_present(dev)) 8761 return -ENODEV; 8762 8763 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); 8764 err = notifier_to_errno(err); 8765 if (err) 8766 return err; 8767 8768 orig_mtu = dev->mtu; 8769 err = __dev_set_mtu(dev, new_mtu); 8770 8771 if (!err) { 8772 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8773 orig_mtu); 8774 err = notifier_to_errno(err); 8775 if (err) { 8776 /* setting mtu back and notifying everyone again, 8777 * so that they have a chance to revert changes. 8778 */ 8779 __dev_set_mtu(dev, orig_mtu); 8780 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8781 new_mtu); 8782 } 8783 } 8784 return err; 8785 } 8786 8787 int dev_set_mtu(struct net_device *dev, int new_mtu) 8788 { 8789 struct netlink_ext_ack extack; 8790 int err; 8791 8792 memset(&extack, 0, sizeof(extack)); 8793 err = dev_set_mtu_ext(dev, new_mtu, &extack); 8794 if (err && extack._msg) 8795 net_err_ratelimited("%s: %s\n", dev->name, extack._msg); 8796 return err; 8797 } 8798 EXPORT_SYMBOL(dev_set_mtu); 8799 8800 /** 8801 * dev_change_tx_queue_len - Change TX queue length of a netdevice 8802 * @dev: device 8803 * @new_len: new tx queue length 8804 */ 8805 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len) 8806 { 8807 unsigned int orig_len = dev->tx_queue_len; 8808 int res; 8809 8810 if (new_len != (unsigned int)new_len) 8811 return -ERANGE; 8812 8813 if (new_len != orig_len) { 8814 dev->tx_queue_len = new_len; 8815 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev); 8816 res = notifier_to_errno(res); 8817 if (res) 8818 goto err_rollback; 8819 res = dev_qdisc_change_tx_queue_len(dev); 8820 if (res) 8821 goto err_rollback; 8822 } 8823 8824 return 0; 8825 8826 err_rollback: 8827 netdev_err(dev, "refused to change device tx_queue_len\n"); 8828 dev->tx_queue_len = orig_len; 8829 return res; 8830 } 8831 8832 /** 8833 * dev_set_group - Change group this device belongs to 8834 * @dev: device 8835 * @new_group: group this device should belong to 8836 */ 8837 void dev_set_group(struct net_device *dev, int new_group) 8838 { 8839 dev->group = new_group; 8840 } 8841 8842 /** 8843 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR. 8844 * @dev: device 8845 * @addr: new address 8846 * @extack: netlink extended ack 8847 */ 8848 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr, 8849 struct netlink_ext_ack *extack) 8850 { 8851 struct netdev_notifier_pre_changeaddr_info info = { 8852 .info.dev = dev, 8853 .info.extack = extack, 8854 .dev_addr = addr, 8855 }; 8856 int rc; 8857 8858 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info); 8859 return notifier_to_errno(rc); 8860 } 8861 EXPORT_SYMBOL(dev_pre_changeaddr_notify); 8862 8863 /** 8864 * dev_set_mac_address - Change Media Access Control Address 8865 * @dev: device 8866 * @sa: new address 8867 * @extack: netlink extended ack 8868 * 8869 * Change the hardware (MAC) address of the device 8870 */ 8871 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa, 8872 struct netlink_ext_ack *extack) 8873 { 8874 const struct net_device_ops *ops = dev->netdev_ops; 8875 int err; 8876 8877 if (!ops->ndo_set_mac_address) 8878 return -EOPNOTSUPP; 8879 if (sa->sa_family != dev->type) 8880 return -EINVAL; 8881 if (!netif_device_present(dev)) 8882 return -ENODEV; 8883 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack); 8884 if (err) 8885 return err; 8886 if (memcmp(dev->dev_addr, sa->sa_data, dev->addr_len)) { 8887 err = ops->ndo_set_mac_address(dev, sa); 8888 if (err) 8889 return err; 8890 } 8891 dev->addr_assign_type = NET_ADDR_SET; 8892 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); 8893 add_device_randomness(dev->dev_addr, dev->addr_len); 8894 return 0; 8895 } 8896 EXPORT_SYMBOL(dev_set_mac_address); 8897 8898 static DECLARE_RWSEM(dev_addr_sem); 8899 8900 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa, 8901 struct netlink_ext_ack *extack) 8902 { 8903 int ret; 8904 8905 down_write(&dev_addr_sem); 8906 ret = dev_set_mac_address(dev, sa, extack); 8907 up_write(&dev_addr_sem); 8908 return ret; 8909 } 8910 EXPORT_SYMBOL(dev_set_mac_address_user); 8911 8912 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name) 8913 { 8914 size_t size = sizeof(sa->sa_data_min); 8915 struct net_device *dev; 8916 int ret = 0; 8917 8918 down_read(&dev_addr_sem); 8919 rcu_read_lock(); 8920 8921 dev = dev_get_by_name_rcu(net, dev_name); 8922 if (!dev) { 8923 ret = -ENODEV; 8924 goto unlock; 8925 } 8926 if (!dev->addr_len) 8927 memset(sa->sa_data, 0, size); 8928 else 8929 memcpy(sa->sa_data, dev->dev_addr, 8930 min_t(size_t, size, dev->addr_len)); 8931 sa->sa_family = dev->type; 8932 8933 unlock: 8934 rcu_read_unlock(); 8935 up_read(&dev_addr_sem); 8936 return ret; 8937 } 8938 EXPORT_SYMBOL(dev_get_mac_address); 8939 8940 /** 8941 * dev_change_carrier - Change device carrier 8942 * @dev: device 8943 * @new_carrier: new value 8944 * 8945 * Change device carrier 8946 */ 8947 int dev_change_carrier(struct net_device *dev, bool new_carrier) 8948 { 8949 const struct net_device_ops *ops = dev->netdev_ops; 8950 8951 if (!ops->ndo_change_carrier) 8952 return -EOPNOTSUPP; 8953 if (!netif_device_present(dev)) 8954 return -ENODEV; 8955 return ops->ndo_change_carrier(dev, new_carrier); 8956 } 8957 8958 /** 8959 * dev_get_phys_port_id - Get device physical port ID 8960 * @dev: device 8961 * @ppid: port ID 8962 * 8963 * Get device physical port ID 8964 */ 8965 int dev_get_phys_port_id(struct net_device *dev, 8966 struct netdev_phys_item_id *ppid) 8967 { 8968 const struct net_device_ops *ops = dev->netdev_ops; 8969 8970 if (!ops->ndo_get_phys_port_id) 8971 return -EOPNOTSUPP; 8972 return ops->ndo_get_phys_port_id(dev, ppid); 8973 } 8974 8975 /** 8976 * dev_get_phys_port_name - Get device physical port name 8977 * @dev: device 8978 * @name: port name 8979 * @len: limit of bytes to copy to name 8980 * 8981 * Get device physical port name 8982 */ 8983 int dev_get_phys_port_name(struct net_device *dev, 8984 char *name, size_t len) 8985 { 8986 const struct net_device_ops *ops = dev->netdev_ops; 8987 int err; 8988 8989 if (ops->ndo_get_phys_port_name) { 8990 err = ops->ndo_get_phys_port_name(dev, name, len); 8991 if (err != -EOPNOTSUPP) 8992 return err; 8993 } 8994 return devlink_compat_phys_port_name_get(dev, name, len); 8995 } 8996 8997 /** 8998 * dev_get_port_parent_id - Get the device's port parent identifier 8999 * @dev: network device 9000 * @ppid: pointer to a storage for the port's parent identifier 9001 * @recurse: allow/disallow recursion to lower devices 9002 * 9003 * Get the devices's port parent identifier 9004 */ 9005 int dev_get_port_parent_id(struct net_device *dev, 9006 struct netdev_phys_item_id *ppid, 9007 bool recurse) 9008 { 9009 const struct net_device_ops *ops = dev->netdev_ops; 9010 struct netdev_phys_item_id first = { }; 9011 struct net_device *lower_dev; 9012 struct list_head *iter; 9013 int err; 9014 9015 if (ops->ndo_get_port_parent_id) { 9016 err = ops->ndo_get_port_parent_id(dev, ppid); 9017 if (err != -EOPNOTSUPP) 9018 return err; 9019 } 9020 9021 err = devlink_compat_switch_id_get(dev, ppid); 9022 if (!recurse || err != -EOPNOTSUPP) 9023 return err; 9024 9025 netdev_for_each_lower_dev(dev, lower_dev, iter) { 9026 err = dev_get_port_parent_id(lower_dev, ppid, true); 9027 if (err) 9028 break; 9029 if (!first.id_len) 9030 first = *ppid; 9031 else if (memcmp(&first, ppid, sizeof(*ppid))) 9032 return -EOPNOTSUPP; 9033 } 9034 9035 return err; 9036 } 9037 EXPORT_SYMBOL(dev_get_port_parent_id); 9038 9039 /** 9040 * netdev_port_same_parent_id - Indicate if two network devices have 9041 * the same port parent identifier 9042 * @a: first network device 9043 * @b: second network device 9044 */ 9045 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b) 9046 { 9047 struct netdev_phys_item_id a_id = { }; 9048 struct netdev_phys_item_id b_id = { }; 9049 9050 if (dev_get_port_parent_id(a, &a_id, true) || 9051 dev_get_port_parent_id(b, &b_id, true)) 9052 return false; 9053 9054 return netdev_phys_item_id_same(&a_id, &b_id); 9055 } 9056 EXPORT_SYMBOL(netdev_port_same_parent_id); 9057 9058 /** 9059 * dev_change_proto_down - set carrier according to proto_down. 9060 * 9061 * @dev: device 9062 * @proto_down: new value 9063 */ 9064 int dev_change_proto_down(struct net_device *dev, bool proto_down) 9065 { 9066 if (!(dev->priv_flags & IFF_CHANGE_PROTO_DOWN)) 9067 return -EOPNOTSUPP; 9068 if (!netif_device_present(dev)) 9069 return -ENODEV; 9070 if (proto_down) 9071 netif_carrier_off(dev); 9072 else 9073 netif_carrier_on(dev); 9074 dev->proto_down = proto_down; 9075 return 0; 9076 } 9077 9078 /** 9079 * dev_change_proto_down_reason - proto down reason 9080 * 9081 * @dev: device 9082 * @mask: proto down mask 9083 * @value: proto down value 9084 */ 9085 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask, 9086 u32 value) 9087 { 9088 int b; 9089 9090 if (!mask) { 9091 dev->proto_down_reason = value; 9092 } else { 9093 for_each_set_bit(b, &mask, 32) { 9094 if (value & (1 << b)) 9095 dev->proto_down_reason |= BIT(b); 9096 else 9097 dev->proto_down_reason &= ~BIT(b); 9098 } 9099 } 9100 } 9101 9102 struct bpf_xdp_link { 9103 struct bpf_link link; 9104 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */ 9105 int flags; 9106 }; 9107 9108 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags) 9109 { 9110 if (flags & XDP_FLAGS_HW_MODE) 9111 return XDP_MODE_HW; 9112 if (flags & XDP_FLAGS_DRV_MODE) 9113 return XDP_MODE_DRV; 9114 if (flags & XDP_FLAGS_SKB_MODE) 9115 return XDP_MODE_SKB; 9116 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB; 9117 } 9118 9119 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode) 9120 { 9121 switch (mode) { 9122 case XDP_MODE_SKB: 9123 return generic_xdp_install; 9124 case XDP_MODE_DRV: 9125 case XDP_MODE_HW: 9126 return dev->netdev_ops->ndo_bpf; 9127 default: 9128 return NULL; 9129 } 9130 } 9131 9132 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev, 9133 enum bpf_xdp_mode mode) 9134 { 9135 return dev->xdp_state[mode].link; 9136 } 9137 9138 static struct bpf_prog *dev_xdp_prog(struct net_device *dev, 9139 enum bpf_xdp_mode mode) 9140 { 9141 struct bpf_xdp_link *link = dev_xdp_link(dev, mode); 9142 9143 if (link) 9144 return link->link.prog; 9145 return dev->xdp_state[mode].prog; 9146 } 9147 9148 u8 dev_xdp_prog_count(struct net_device *dev) 9149 { 9150 u8 count = 0; 9151 int i; 9152 9153 for (i = 0; i < __MAX_XDP_MODE; i++) 9154 if (dev->xdp_state[i].prog || dev->xdp_state[i].link) 9155 count++; 9156 return count; 9157 } 9158 EXPORT_SYMBOL_GPL(dev_xdp_prog_count); 9159 9160 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode) 9161 { 9162 struct bpf_prog *prog = dev_xdp_prog(dev, mode); 9163 9164 return prog ? prog->aux->id : 0; 9165 } 9166 9167 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode, 9168 struct bpf_xdp_link *link) 9169 { 9170 dev->xdp_state[mode].link = link; 9171 dev->xdp_state[mode].prog = NULL; 9172 } 9173 9174 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode, 9175 struct bpf_prog *prog) 9176 { 9177 dev->xdp_state[mode].link = NULL; 9178 dev->xdp_state[mode].prog = prog; 9179 } 9180 9181 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode, 9182 bpf_op_t bpf_op, struct netlink_ext_ack *extack, 9183 u32 flags, struct bpf_prog *prog) 9184 { 9185 struct netdev_bpf xdp; 9186 int err; 9187 9188 memset(&xdp, 0, sizeof(xdp)); 9189 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG; 9190 xdp.extack = extack; 9191 xdp.flags = flags; 9192 xdp.prog = prog; 9193 9194 /* Drivers assume refcnt is already incremented (i.e, prog pointer is 9195 * "moved" into driver), so they don't increment it on their own, but 9196 * they do decrement refcnt when program is detached or replaced. 9197 * Given net_device also owns link/prog, we need to bump refcnt here 9198 * to prevent drivers from underflowing it. 9199 */ 9200 if (prog) 9201 bpf_prog_inc(prog); 9202 err = bpf_op(dev, &xdp); 9203 if (err) { 9204 if (prog) 9205 bpf_prog_put(prog); 9206 return err; 9207 } 9208 9209 if (mode != XDP_MODE_HW) 9210 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog); 9211 9212 return 0; 9213 } 9214 9215 static void dev_xdp_uninstall(struct net_device *dev) 9216 { 9217 struct bpf_xdp_link *link; 9218 struct bpf_prog *prog; 9219 enum bpf_xdp_mode mode; 9220 bpf_op_t bpf_op; 9221 9222 ASSERT_RTNL(); 9223 9224 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) { 9225 prog = dev_xdp_prog(dev, mode); 9226 if (!prog) 9227 continue; 9228 9229 bpf_op = dev_xdp_bpf_op(dev, mode); 9230 if (!bpf_op) 9231 continue; 9232 9233 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9234 9235 /* auto-detach link from net device */ 9236 link = dev_xdp_link(dev, mode); 9237 if (link) 9238 link->dev = NULL; 9239 else 9240 bpf_prog_put(prog); 9241 9242 dev_xdp_set_link(dev, mode, NULL); 9243 } 9244 } 9245 9246 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack, 9247 struct bpf_xdp_link *link, struct bpf_prog *new_prog, 9248 struct bpf_prog *old_prog, u32 flags) 9249 { 9250 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES); 9251 struct bpf_prog *cur_prog; 9252 struct net_device *upper; 9253 struct list_head *iter; 9254 enum bpf_xdp_mode mode; 9255 bpf_op_t bpf_op; 9256 int err; 9257 9258 ASSERT_RTNL(); 9259 9260 /* either link or prog attachment, never both */ 9261 if (link && (new_prog || old_prog)) 9262 return -EINVAL; 9263 /* link supports only XDP mode flags */ 9264 if (link && (flags & ~XDP_FLAGS_MODES)) { 9265 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment"); 9266 return -EINVAL; 9267 } 9268 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */ 9269 if (num_modes > 1) { 9270 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set"); 9271 return -EINVAL; 9272 } 9273 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */ 9274 if (!num_modes && dev_xdp_prog_count(dev) > 1) { 9275 NL_SET_ERR_MSG(extack, 9276 "More than one program loaded, unset mode is ambiguous"); 9277 return -EINVAL; 9278 } 9279 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */ 9280 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) { 9281 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified"); 9282 return -EINVAL; 9283 } 9284 9285 mode = dev_xdp_mode(dev, flags); 9286 /* can't replace attached link */ 9287 if (dev_xdp_link(dev, mode)) { 9288 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link"); 9289 return -EBUSY; 9290 } 9291 9292 /* don't allow if an upper device already has a program */ 9293 netdev_for_each_upper_dev_rcu(dev, upper, iter) { 9294 if (dev_xdp_prog_count(upper) > 0) { 9295 NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program"); 9296 return -EEXIST; 9297 } 9298 } 9299 9300 cur_prog = dev_xdp_prog(dev, mode); 9301 /* can't replace attached prog with link */ 9302 if (link && cur_prog) { 9303 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link"); 9304 return -EBUSY; 9305 } 9306 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) { 9307 NL_SET_ERR_MSG(extack, "Active program does not match expected"); 9308 return -EEXIST; 9309 } 9310 9311 /* put effective new program into new_prog */ 9312 if (link) 9313 new_prog = link->link.prog; 9314 9315 if (new_prog) { 9316 bool offload = mode == XDP_MODE_HW; 9317 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB 9318 ? XDP_MODE_DRV : XDP_MODE_SKB; 9319 9320 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) { 9321 NL_SET_ERR_MSG(extack, "XDP program already attached"); 9322 return -EBUSY; 9323 } 9324 if (!offload && dev_xdp_prog(dev, other_mode)) { 9325 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time"); 9326 return -EEXIST; 9327 } 9328 if (!offload && bpf_prog_is_offloaded(new_prog->aux)) { 9329 NL_SET_ERR_MSG(extack, "Using offloaded program without HW_MODE flag is not supported"); 9330 return -EINVAL; 9331 } 9332 if (bpf_prog_is_dev_bound(new_prog->aux) && !bpf_offload_dev_match(new_prog, dev)) { 9333 NL_SET_ERR_MSG(extack, "Program bound to different device"); 9334 return -EINVAL; 9335 } 9336 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) { 9337 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device"); 9338 return -EINVAL; 9339 } 9340 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) { 9341 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device"); 9342 return -EINVAL; 9343 } 9344 } 9345 9346 /* don't call drivers if the effective program didn't change */ 9347 if (new_prog != cur_prog) { 9348 bpf_op = dev_xdp_bpf_op(dev, mode); 9349 if (!bpf_op) { 9350 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode"); 9351 return -EOPNOTSUPP; 9352 } 9353 9354 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog); 9355 if (err) 9356 return err; 9357 } 9358 9359 if (link) 9360 dev_xdp_set_link(dev, mode, link); 9361 else 9362 dev_xdp_set_prog(dev, mode, new_prog); 9363 if (cur_prog) 9364 bpf_prog_put(cur_prog); 9365 9366 return 0; 9367 } 9368 9369 static int dev_xdp_attach_link(struct net_device *dev, 9370 struct netlink_ext_ack *extack, 9371 struct bpf_xdp_link *link) 9372 { 9373 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags); 9374 } 9375 9376 static int dev_xdp_detach_link(struct net_device *dev, 9377 struct netlink_ext_ack *extack, 9378 struct bpf_xdp_link *link) 9379 { 9380 enum bpf_xdp_mode mode; 9381 bpf_op_t bpf_op; 9382 9383 ASSERT_RTNL(); 9384 9385 mode = dev_xdp_mode(dev, link->flags); 9386 if (dev_xdp_link(dev, mode) != link) 9387 return -EINVAL; 9388 9389 bpf_op = dev_xdp_bpf_op(dev, mode); 9390 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9391 dev_xdp_set_link(dev, mode, NULL); 9392 return 0; 9393 } 9394 9395 static void bpf_xdp_link_release(struct bpf_link *link) 9396 { 9397 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9398 9399 rtnl_lock(); 9400 9401 /* if racing with net_device's tear down, xdp_link->dev might be 9402 * already NULL, in which case link was already auto-detached 9403 */ 9404 if (xdp_link->dev) { 9405 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link)); 9406 xdp_link->dev = NULL; 9407 } 9408 9409 rtnl_unlock(); 9410 } 9411 9412 static int bpf_xdp_link_detach(struct bpf_link *link) 9413 { 9414 bpf_xdp_link_release(link); 9415 return 0; 9416 } 9417 9418 static void bpf_xdp_link_dealloc(struct bpf_link *link) 9419 { 9420 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9421 9422 kfree(xdp_link); 9423 } 9424 9425 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link, 9426 struct seq_file *seq) 9427 { 9428 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9429 u32 ifindex = 0; 9430 9431 rtnl_lock(); 9432 if (xdp_link->dev) 9433 ifindex = xdp_link->dev->ifindex; 9434 rtnl_unlock(); 9435 9436 seq_printf(seq, "ifindex:\t%u\n", ifindex); 9437 } 9438 9439 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link, 9440 struct bpf_link_info *info) 9441 { 9442 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9443 u32 ifindex = 0; 9444 9445 rtnl_lock(); 9446 if (xdp_link->dev) 9447 ifindex = xdp_link->dev->ifindex; 9448 rtnl_unlock(); 9449 9450 info->xdp.ifindex = ifindex; 9451 return 0; 9452 } 9453 9454 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog, 9455 struct bpf_prog *old_prog) 9456 { 9457 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9458 enum bpf_xdp_mode mode; 9459 bpf_op_t bpf_op; 9460 int err = 0; 9461 9462 rtnl_lock(); 9463 9464 /* link might have been auto-released already, so fail */ 9465 if (!xdp_link->dev) { 9466 err = -ENOLINK; 9467 goto out_unlock; 9468 } 9469 9470 if (old_prog && link->prog != old_prog) { 9471 err = -EPERM; 9472 goto out_unlock; 9473 } 9474 old_prog = link->prog; 9475 if (old_prog->type != new_prog->type || 9476 old_prog->expected_attach_type != new_prog->expected_attach_type) { 9477 err = -EINVAL; 9478 goto out_unlock; 9479 } 9480 9481 if (old_prog == new_prog) { 9482 /* no-op, don't disturb drivers */ 9483 bpf_prog_put(new_prog); 9484 goto out_unlock; 9485 } 9486 9487 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags); 9488 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode); 9489 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL, 9490 xdp_link->flags, new_prog); 9491 if (err) 9492 goto out_unlock; 9493 9494 old_prog = xchg(&link->prog, new_prog); 9495 bpf_prog_put(old_prog); 9496 9497 out_unlock: 9498 rtnl_unlock(); 9499 return err; 9500 } 9501 9502 static const struct bpf_link_ops bpf_xdp_link_lops = { 9503 .release = bpf_xdp_link_release, 9504 .dealloc = bpf_xdp_link_dealloc, 9505 .detach = bpf_xdp_link_detach, 9506 .show_fdinfo = bpf_xdp_link_show_fdinfo, 9507 .fill_link_info = bpf_xdp_link_fill_link_info, 9508 .update_prog = bpf_xdp_link_update, 9509 }; 9510 9511 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog) 9512 { 9513 struct net *net = current->nsproxy->net_ns; 9514 struct bpf_link_primer link_primer; 9515 struct netlink_ext_ack extack = {}; 9516 struct bpf_xdp_link *link; 9517 struct net_device *dev; 9518 int err, fd; 9519 9520 rtnl_lock(); 9521 dev = dev_get_by_index(net, attr->link_create.target_ifindex); 9522 if (!dev) { 9523 rtnl_unlock(); 9524 return -EINVAL; 9525 } 9526 9527 link = kzalloc(sizeof(*link), GFP_USER); 9528 if (!link) { 9529 err = -ENOMEM; 9530 goto unlock; 9531 } 9532 9533 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog); 9534 link->dev = dev; 9535 link->flags = attr->link_create.flags; 9536 9537 err = bpf_link_prime(&link->link, &link_primer); 9538 if (err) { 9539 kfree(link); 9540 goto unlock; 9541 } 9542 9543 err = dev_xdp_attach_link(dev, &extack, link); 9544 rtnl_unlock(); 9545 9546 if (err) { 9547 link->dev = NULL; 9548 bpf_link_cleanup(&link_primer); 9549 trace_bpf_xdp_link_attach_failed(extack._msg); 9550 goto out_put_dev; 9551 } 9552 9553 fd = bpf_link_settle(&link_primer); 9554 /* link itself doesn't hold dev's refcnt to not complicate shutdown */ 9555 dev_put(dev); 9556 return fd; 9557 9558 unlock: 9559 rtnl_unlock(); 9560 9561 out_put_dev: 9562 dev_put(dev); 9563 return err; 9564 } 9565 9566 /** 9567 * dev_change_xdp_fd - set or clear a bpf program for a device rx path 9568 * @dev: device 9569 * @extack: netlink extended ack 9570 * @fd: new program fd or negative value to clear 9571 * @expected_fd: old program fd that userspace expects to replace or clear 9572 * @flags: xdp-related flags 9573 * 9574 * Set or clear a bpf program for a device 9575 */ 9576 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, 9577 int fd, int expected_fd, u32 flags) 9578 { 9579 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags); 9580 struct bpf_prog *new_prog = NULL, *old_prog = NULL; 9581 int err; 9582 9583 ASSERT_RTNL(); 9584 9585 if (fd >= 0) { 9586 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP, 9587 mode != XDP_MODE_SKB); 9588 if (IS_ERR(new_prog)) 9589 return PTR_ERR(new_prog); 9590 } 9591 9592 if (expected_fd >= 0) { 9593 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP, 9594 mode != XDP_MODE_SKB); 9595 if (IS_ERR(old_prog)) { 9596 err = PTR_ERR(old_prog); 9597 old_prog = NULL; 9598 goto err_out; 9599 } 9600 } 9601 9602 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags); 9603 9604 err_out: 9605 if (err && new_prog) 9606 bpf_prog_put(new_prog); 9607 if (old_prog) 9608 bpf_prog_put(old_prog); 9609 return err; 9610 } 9611 9612 /** 9613 * dev_index_reserve() - allocate an ifindex in a namespace 9614 * @net: the applicable net namespace 9615 * @ifindex: requested ifindex, pass %0 to get one allocated 9616 * 9617 * Allocate a ifindex for a new device. Caller must either use the ifindex 9618 * to store the device (via list_netdevice()) or call dev_index_release() 9619 * to give the index up. 9620 * 9621 * Return: a suitable unique value for a new device interface number or -errno. 9622 */ 9623 static int dev_index_reserve(struct net *net, u32 ifindex) 9624 { 9625 int err; 9626 9627 if (ifindex > INT_MAX) { 9628 DEBUG_NET_WARN_ON_ONCE(1); 9629 return -EINVAL; 9630 } 9631 9632 if (!ifindex) 9633 err = xa_alloc_cyclic(&net->dev_by_index, &ifindex, NULL, 9634 xa_limit_31b, &net->ifindex, GFP_KERNEL); 9635 else 9636 err = xa_insert(&net->dev_by_index, ifindex, NULL, GFP_KERNEL); 9637 if (err < 0) 9638 return err; 9639 9640 return ifindex; 9641 } 9642 9643 static void dev_index_release(struct net *net, int ifindex) 9644 { 9645 /* Expect only unused indexes, unlist_netdevice() removes the used */ 9646 WARN_ON(xa_erase(&net->dev_by_index, ifindex)); 9647 } 9648 9649 /* Delayed registration/unregisteration */ 9650 LIST_HEAD(net_todo_list); 9651 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); 9652 9653 static void net_set_todo(struct net_device *dev) 9654 { 9655 list_add_tail(&dev->todo_list, &net_todo_list); 9656 atomic_inc(&dev_net(dev)->dev_unreg_count); 9657 } 9658 9659 static netdev_features_t netdev_sync_upper_features(struct net_device *lower, 9660 struct net_device *upper, netdev_features_t features) 9661 { 9662 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9663 netdev_features_t feature; 9664 int feature_bit; 9665 9666 for_each_netdev_feature(upper_disables, feature_bit) { 9667 feature = __NETIF_F_BIT(feature_bit); 9668 if (!(upper->wanted_features & feature) 9669 && (features & feature)) { 9670 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", 9671 &feature, upper->name); 9672 features &= ~feature; 9673 } 9674 } 9675 9676 return features; 9677 } 9678 9679 static void netdev_sync_lower_features(struct net_device *upper, 9680 struct net_device *lower, netdev_features_t features) 9681 { 9682 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9683 netdev_features_t feature; 9684 int feature_bit; 9685 9686 for_each_netdev_feature(upper_disables, feature_bit) { 9687 feature = __NETIF_F_BIT(feature_bit); 9688 if (!(features & feature) && (lower->features & feature)) { 9689 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", 9690 &feature, lower->name); 9691 lower->wanted_features &= ~feature; 9692 __netdev_update_features(lower); 9693 9694 if (unlikely(lower->features & feature)) 9695 netdev_WARN(upper, "failed to disable %pNF on %s!\n", 9696 &feature, lower->name); 9697 else 9698 netdev_features_change(lower); 9699 } 9700 } 9701 } 9702 9703 static netdev_features_t netdev_fix_features(struct net_device *dev, 9704 netdev_features_t features) 9705 { 9706 /* Fix illegal checksum combinations */ 9707 if ((features & NETIF_F_HW_CSUM) && 9708 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { 9709 netdev_warn(dev, "mixed HW and IP checksum settings.\n"); 9710 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); 9711 } 9712 9713 /* TSO requires that SG is present as well. */ 9714 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { 9715 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); 9716 features &= ~NETIF_F_ALL_TSO; 9717 } 9718 9719 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && 9720 !(features & NETIF_F_IP_CSUM)) { 9721 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); 9722 features &= ~NETIF_F_TSO; 9723 features &= ~NETIF_F_TSO_ECN; 9724 } 9725 9726 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && 9727 !(features & NETIF_F_IPV6_CSUM)) { 9728 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); 9729 features &= ~NETIF_F_TSO6; 9730 } 9731 9732 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ 9733 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) 9734 features &= ~NETIF_F_TSO_MANGLEID; 9735 9736 /* TSO ECN requires that TSO is present as well. */ 9737 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) 9738 features &= ~NETIF_F_TSO_ECN; 9739 9740 /* Software GSO depends on SG. */ 9741 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { 9742 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); 9743 features &= ~NETIF_F_GSO; 9744 } 9745 9746 /* GSO partial features require GSO partial be set */ 9747 if ((features & dev->gso_partial_features) && 9748 !(features & NETIF_F_GSO_PARTIAL)) { 9749 netdev_dbg(dev, 9750 "Dropping partially supported GSO features since no GSO partial.\n"); 9751 features &= ~dev->gso_partial_features; 9752 } 9753 9754 if (!(features & NETIF_F_RXCSUM)) { 9755 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet 9756 * successfully merged by hardware must also have the 9757 * checksum verified by hardware. If the user does not 9758 * want to enable RXCSUM, logically, we should disable GRO_HW. 9759 */ 9760 if (features & NETIF_F_GRO_HW) { 9761 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n"); 9762 features &= ~NETIF_F_GRO_HW; 9763 } 9764 } 9765 9766 /* LRO/HW-GRO features cannot be combined with RX-FCS */ 9767 if (features & NETIF_F_RXFCS) { 9768 if (features & NETIF_F_LRO) { 9769 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n"); 9770 features &= ~NETIF_F_LRO; 9771 } 9772 9773 if (features & NETIF_F_GRO_HW) { 9774 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n"); 9775 features &= ~NETIF_F_GRO_HW; 9776 } 9777 } 9778 9779 if ((features & NETIF_F_GRO_HW) && (features & NETIF_F_LRO)) { 9780 netdev_dbg(dev, "Dropping LRO feature since HW-GRO is requested.\n"); 9781 features &= ~NETIF_F_LRO; 9782 } 9783 9784 if (features & NETIF_F_HW_TLS_TX) { 9785 bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) == 9786 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM); 9787 bool hw_csum = features & NETIF_F_HW_CSUM; 9788 9789 if (!ip_csum && !hw_csum) { 9790 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n"); 9791 features &= ~NETIF_F_HW_TLS_TX; 9792 } 9793 } 9794 9795 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) { 9796 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n"); 9797 features &= ~NETIF_F_HW_TLS_RX; 9798 } 9799 9800 return features; 9801 } 9802 9803 int __netdev_update_features(struct net_device *dev) 9804 { 9805 struct net_device *upper, *lower; 9806 netdev_features_t features; 9807 struct list_head *iter; 9808 int err = -1; 9809 9810 ASSERT_RTNL(); 9811 9812 features = netdev_get_wanted_features(dev); 9813 9814 if (dev->netdev_ops->ndo_fix_features) 9815 features = dev->netdev_ops->ndo_fix_features(dev, features); 9816 9817 /* driver might be less strict about feature dependencies */ 9818 features = netdev_fix_features(dev, features); 9819 9820 /* some features can't be enabled if they're off on an upper device */ 9821 netdev_for_each_upper_dev_rcu(dev, upper, iter) 9822 features = netdev_sync_upper_features(dev, upper, features); 9823 9824 if (dev->features == features) 9825 goto sync_lower; 9826 9827 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", 9828 &dev->features, &features); 9829 9830 if (dev->netdev_ops->ndo_set_features) 9831 err = dev->netdev_ops->ndo_set_features(dev, features); 9832 else 9833 err = 0; 9834 9835 if (unlikely(err < 0)) { 9836 netdev_err(dev, 9837 "set_features() failed (%d); wanted %pNF, left %pNF\n", 9838 err, &features, &dev->features); 9839 /* return non-0 since some features might have changed and 9840 * it's better to fire a spurious notification than miss it 9841 */ 9842 return -1; 9843 } 9844 9845 sync_lower: 9846 /* some features must be disabled on lower devices when disabled 9847 * on an upper device (think: bonding master or bridge) 9848 */ 9849 netdev_for_each_lower_dev(dev, lower, iter) 9850 netdev_sync_lower_features(dev, lower, features); 9851 9852 if (!err) { 9853 netdev_features_t diff = features ^ dev->features; 9854 9855 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) { 9856 /* udp_tunnel_{get,drop}_rx_info both need 9857 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the 9858 * device, or they won't do anything. 9859 * Thus we need to update dev->features 9860 * *before* calling udp_tunnel_get_rx_info, 9861 * but *after* calling udp_tunnel_drop_rx_info. 9862 */ 9863 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) { 9864 dev->features = features; 9865 udp_tunnel_get_rx_info(dev); 9866 } else { 9867 udp_tunnel_drop_rx_info(dev); 9868 } 9869 } 9870 9871 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) { 9872 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) { 9873 dev->features = features; 9874 err |= vlan_get_rx_ctag_filter_info(dev); 9875 } else { 9876 vlan_drop_rx_ctag_filter_info(dev); 9877 } 9878 } 9879 9880 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) { 9881 if (features & NETIF_F_HW_VLAN_STAG_FILTER) { 9882 dev->features = features; 9883 err |= vlan_get_rx_stag_filter_info(dev); 9884 } else { 9885 vlan_drop_rx_stag_filter_info(dev); 9886 } 9887 } 9888 9889 dev->features = features; 9890 } 9891 9892 return err < 0 ? 0 : 1; 9893 } 9894 9895 /** 9896 * netdev_update_features - recalculate device features 9897 * @dev: the device to check 9898 * 9899 * Recalculate dev->features set and send notifications if it 9900 * has changed. Should be called after driver or hardware dependent 9901 * conditions might have changed that influence the features. 9902 */ 9903 void netdev_update_features(struct net_device *dev) 9904 { 9905 if (__netdev_update_features(dev)) 9906 netdev_features_change(dev); 9907 } 9908 EXPORT_SYMBOL(netdev_update_features); 9909 9910 /** 9911 * netdev_change_features - recalculate device features 9912 * @dev: the device to check 9913 * 9914 * Recalculate dev->features set and send notifications even 9915 * if they have not changed. Should be called instead of 9916 * netdev_update_features() if also dev->vlan_features might 9917 * have changed to allow the changes to be propagated to stacked 9918 * VLAN devices. 9919 */ 9920 void netdev_change_features(struct net_device *dev) 9921 { 9922 __netdev_update_features(dev); 9923 netdev_features_change(dev); 9924 } 9925 EXPORT_SYMBOL(netdev_change_features); 9926 9927 /** 9928 * netif_stacked_transfer_operstate - transfer operstate 9929 * @rootdev: the root or lower level device to transfer state from 9930 * @dev: the device to transfer operstate to 9931 * 9932 * Transfer operational state from root to device. This is normally 9933 * called when a stacking relationship exists between the root 9934 * device and the device(a leaf device). 9935 */ 9936 void netif_stacked_transfer_operstate(const struct net_device *rootdev, 9937 struct net_device *dev) 9938 { 9939 if (rootdev->operstate == IF_OPER_DORMANT) 9940 netif_dormant_on(dev); 9941 else 9942 netif_dormant_off(dev); 9943 9944 if (rootdev->operstate == IF_OPER_TESTING) 9945 netif_testing_on(dev); 9946 else 9947 netif_testing_off(dev); 9948 9949 if (netif_carrier_ok(rootdev)) 9950 netif_carrier_on(dev); 9951 else 9952 netif_carrier_off(dev); 9953 } 9954 EXPORT_SYMBOL(netif_stacked_transfer_operstate); 9955 9956 static int netif_alloc_rx_queues(struct net_device *dev) 9957 { 9958 unsigned int i, count = dev->num_rx_queues; 9959 struct netdev_rx_queue *rx; 9960 size_t sz = count * sizeof(*rx); 9961 int err = 0; 9962 9963 BUG_ON(count < 1); 9964 9965 rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 9966 if (!rx) 9967 return -ENOMEM; 9968 9969 dev->_rx = rx; 9970 9971 for (i = 0; i < count; i++) { 9972 rx[i].dev = dev; 9973 9974 /* XDP RX-queue setup */ 9975 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0); 9976 if (err < 0) 9977 goto err_rxq_info; 9978 } 9979 return 0; 9980 9981 err_rxq_info: 9982 /* Rollback successful reg's and free other resources */ 9983 while (i--) 9984 xdp_rxq_info_unreg(&rx[i].xdp_rxq); 9985 kvfree(dev->_rx); 9986 dev->_rx = NULL; 9987 return err; 9988 } 9989 9990 static void netif_free_rx_queues(struct net_device *dev) 9991 { 9992 unsigned int i, count = dev->num_rx_queues; 9993 9994 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */ 9995 if (!dev->_rx) 9996 return; 9997 9998 for (i = 0; i < count; i++) 9999 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq); 10000 10001 kvfree(dev->_rx); 10002 } 10003 10004 static void netdev_init_one_queue(struct net_device *dev, 10005 struct netdev_queue *queue, void *_unused) 10006 { 10007 /* Initialize queue lock */ 10008 spin_lock_init(&queue->_xmit_lock); 10009 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); 10010 queue->xmit_lock_owner = -1; 10011 netdev_queue_numa_node_write(queue, NUMA_NO_NODE); 10012 queue->dev = dev; 10013 #ifdef CONFIG_BQL 10014 dql_init(&queue->dql, HZ); 10015 #endif 10016 } 10017 10018 static void netif_free_tx_queues(struct net_device *dev) 10019 { 10020 kvfree(dev->_tx); 10021 } 10022 10023 static int netif_alloc_netdev_queues(struct net_device *dev) 10024 { 10025 unsigned int count = dev->num_tx_queues; 10026 struct netdev_queue *tx; 10027 size_t sz = count * sizeof(*tx); 10028 10029 if (count < 1 || count > 0xffff) 10030 return -EINVAL; 10031 10032 tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10033 if (!tx) 10034 return -ENOMEM; 10035 10036 dev->_tx = tx; 10037 10038 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); 10039 spin_lock_init(&dev->tx_global_lock); 10040 10041 return 0; 10042 } 10043 10044 void netif_tx_stop_all_queues(struct net_device *dev) 10045 { 10046 unsigned int i; 10047 10048 for (i = 0; i < dev->num_tx_queues; i++) { 10049 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 10050 10051 netif_tx_stop_queue(txq); 10052 } 10053 } 10054 EXPORT_SYMBOL(netif_tx_stop_all_queues); 10055 10056 static int netdev_do_alloc_pcpu_stats(struct net_device *dev) 10057 { 10058 void __percpu *v; 10059 10060 /* Drivers implementing ndo_get_peer_dev must support tstat 10061 * accounting, so that skb_do_redirect() can bump the dev's 10062 * RX stats upon network namespace switch. 10063 */ 10064 if (dev->netdev_ops->ndo_get_peer_dev && 10065 dev->pcpu_stat_type != NETDEV_PCPU_STAT_TSTATS) 10066 return -EOPNOTSUPP; 10067 10068 switch (dev->pcpu_stat_type) { 10069 case NETDEV_PCPU_STAT_NONE: 10070 return 0; 10071 case NETDEV_PCPU_STAT_LSTATS: 10072 v = dev->lstats = netdev_alloc_pcpu_stats(struct pcpu_lstats); 10073 break; 10074 case NETDEV_PCPU_STAT_TSTATS: 10075 v = dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats); 10076 break; 10077 case NETDEV_PCPU_STAT_DSTATS: 10078 v = dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats); 10079 break; 10080 default: 10081 return -EINVAL; 10082 } 10083 10084 return v ? 0 : -ENOMEM; 10085 } 10086 10087 static void netdev_do_free_pcpu_stats(struct net_device *dev) 10088 { 10089 switch (dev->pcpu_stat_type) { 10090 case NETDEV_PCPU_STAT_NONE: 10091 return; 10092 case NETDEV_PCPU_STAT_LSTATS: 10093 free_percpu(dev->lstats); 10094 break; 10095 case NETDEV_PCPU_STAT_TSTATS: 10096 free_percpu(dev->tstats); 10097 break; 10098 case NETDEV_PCPU_STAT_DSTATS: 10099 free_percpu(dev->dstats); 10100 break; 10101 } 10102 } 10103 10104 /** 10105 * register_netdevice() - register a network device 10106 * @dev: device to register 10107 * 10108 * Take a prepared network device structure and make it externally accessible. 10109 * A %NETDEV_REGISTER message is sent to the netdev notifier chain. 10110 * Callers must hold the rtnl lock - you may want register_netdev() 10111 * instead of this. 10112 */ 10113 int register_netdevice(struct net_device *dev) 10114 { 10115 int ret; 10116 struct net *net = dev_net(dev); 10117 10118 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE < 10119 NETDEV_FEATURE_COUNT); 10120 BUG_ON(dev_boot_phase); 10121 ASSERT_RTNL(); 10122 10123 might_sleep(); 10124 10125 /* When net_device's are persistent, this will be fatal. */ 10126 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); 10127 BUG_ON(!net); 10128 10129 ret = ethtool_check_ops(dev->ethtool_ops); 10130 if (ret) 10131 return ret; 10132 10133 spin_lock_init(&dev->addr_list_lock); 10134 netdev_set_addr_lockdep_class(dev); 10135 10136 ret = dev_get_valid_name(net, dev, dev->name); 10137 if (ret < 0) 10138 goto out; 10139 10140 ret = -ENOMEM; 10141 dev->name_node = netdev_name_node_head_alloc(dev); 10142 if (!dev->name_node) 10143 goto out; 10144 10145 /* Init, if this function is available */ 10146 if (dev->netdev_ops->ndo_init) { 10147 ret = dev->netdev_ops->ndo_init(dev); 10148 if (ret) { 10149 if (ret > 0) 10150 ret = -EIO; 10151 goto err_free_name; 10152 } 10153 } 10154 10155 if (((dev->hw_features | dev->features) & 10156 NETIF_F_HW_VLAN_CTAG_FILTER) && 10157 (!dev->netdev_ops->ndo_vlan_rx_add_vid || 10158 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { 10159 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); 10160 ret = -EINVAL; 10161 goto err_uninit; 10162 } 10163 10164 ret = netdev_do_alloc_pcpu_stats(dev); 10165 if (ret) 10166 goto err_uninit; 10167 10168 ret = dev_index_reserve(net, dev->ifindex); 10169 if (ret < 0) 10170 goto err_free_pcpu; 10171 dev->ifindex = ret; 10172 10173 /* Transfer changeable features to wanted_features and enable 10174 * software offloads (GSO and GRO). 10175 */ 10176 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF); 10177 dev->features |= NETIF_F_SOFT_FEATURES; 10178 10179 if (dev->udp_tunnel_nic_info) { 10180 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10181 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10182 } 10183 10184 dev->wanted_features = dev->features & dev->hw_features; 10185 10186 if (!(dev->flags & IFF_LOOPBACK)) 10187 dev->hw_features |= NETIF_F_NOCACHE_COPY; 10188 10189 /* If IPv4 TCP segmentation offload is supported we should also 10190 * allow the device to enable segmenting the frame with the option 10191 * of ignoring a static IP ID value. This doesn't enable the 10192 * feature itself but allows the user to enable it later. 10193 */ 10194 if (dev->hw_features & NETIF_F_TSO) 10195 dev->hw_features |= NETIF_F_TSO_MANGLEID; 10196 if (dev->vlan_features & NETIF_F_TSO) 10197 dev->vlan_features |= NETIF_F_TSO_MANGLEID; 10198 if (dev->mpls_features & NETIF_F_TSO) 10199 dev->mpls_features |= NETIF_F_TSO_MANGLEID; 10200 if (dev->hw_enc_features & NETIF_F_TSO) 10201 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; 10202 10203 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. 10204 */ 10205 dev->vlan_features |= NETIF_F_HIGHDMA; 10206 10207 /* Make NETIF_F_SG inheritable to tunnel devices. 10208 */ 10209 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; 10210 10211 /* Make NETIF_F_SG inheritable to MPLS. 10212 */ 10213 dev->mpls_features |= NETIF_F_SG; 10214 10215 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); 10216 ret = notifier_to_errno(ret); 10217 if (ret) 10218 goto err_ifindex_release; 10219 10220 ret = netdev_register_kobject(dev); 10221 write_lock(&dev_base_lock); 10222 dev->reg_state = ret ? NETREG_UNREGISTERED : NETREG_REGISTERED; 10223 write_unlock(&dev_base_lock); 10224 if (ret) 10225 goto err_uninit_notify; 10226 10227 __netdev_update_features(dev); 10228 10229 /* 10230 * Default initial state at registry is that the 10231 * device is present. 10232 */ 10233 10234 set_bit(__LINK_STATE_PRESENT, &dev->state); 10235 10236 linkwatch_init_dev(dev); 10237 10238 dev_init_scheduler(dev); 10239 10240 netdev_hold(dev, &dev->dev_registered_tracker, GFP_KERNEL); 10241 list_netdevice(dev); 10242 10243 add_device_randomness(dev->dev_addr, dev->addr_len); 10244 10245 /* If the device has permanent device address, driver should 10246 * set dev_addr and also addr_assign_type should be set to 10247 * NET_ADDR_PERM (default value). 10248 */ 10249 if (dev->addr_assign_type == NET_ADDR_PERM) 10250 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); 10251 10252 /* Notify protocols, that a new device appeared. */ 10253 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); 10254 ret = notifier_to_errno(ret); 10255 if (ret) { 10256 /* Expect explicit free_netdev() on failure */ 10257 dev->needs_free_netdev = false; 10258 unregister_netdevice_queue(dev, NULL); 10259 goto out; 10260 } 10261 /* 10262 * Prevent userspace races by waiting until the network 10263 * device is fully setup before sending notifications. 10264 */ 10265 if (!dev->rtnl_link_ops || 10266 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 10267 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL); 10268 10269 out: 10270 return ret; 10271 10272 err_uninit_notify: 10273 call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev); 10274 err_ifindex_release: 10275 dev_index_release(net, dev->ifindex); 10276 err_free_pcpu: 10277 netdev_do_free_pcpu_stats(dev); 10278 err_uninit: 10279 if (dev->netdev_ops->ndo_uninit) 10280 dev->netdev_ops->ndo_uninit(dev); 10281 if (dev->priv_destructor) 10282 dev->priv_destructor(dev); 10283 err_free_name: 10284 netdev_name_node_free(dev->name_node); 10285 goto out; 10286 } 10287 EXPORT_SYMBOL(register_netdevice); 10288 10289 /** 10290 * init_dummy_netdev - init a dummy network device for NAPI 10291 * @dev: device to init 10292 * 10293 * This takes a network device structure and initialize the minimum 10294 * amount of fields so it can be used to schedule NAPI polls without 10295 * registering a full blown interface. This is to be used by drivers 10296 * that need to tie several hardware interfaces to a single NAPI 10297 * poll scheduler due to HW limitations. 10298 */ 10299 int init_dummy_netdev(struct net_device *dev) 10300 { 10301 /* Clear everything. Note we don't initialize spinlocks 10302 * are they aren't supposed to be taken by any of the 10303 * NAPI code and this dummy netdev is supposed to be 10304 * only ever used for NAPI polls 10305 */ 10306 memset(dev, 0, sizeof(struct net_device)); 10307 10308 /* make sure we BUG if trying to hit standard 10309 * register/unregister code path 10310 */ 10311 dev->reg_state = NETREG_DUMMY; 10312 10313 /* NAPI wants this */ 10314 INIT_LIST_HEAD(&dev->napi_list); 10315 10316 /* a dummy interface is started by default */ 10317 set_bit(__LINK_STATE_PRESENT, &dev->state); 10318 set_bit(__LINK_STATE_START, &dev->state); 10319 10320 /* napi_busy_loop stats accounting wants this */ 10321 dev_net_set(dev, &init_net); 10322 10323 /* Note : We dont allocate pcpu_refcnt for dummy devices, 10324 * because users of this 'device' dont need to change 10325 * its refcount. 10326 */ 10327 10328 return 0; 10329 } 10330 EXPORT_SYMBOL_GPL(init_dummy_netdev); 10331 10332 10333 /** 10334 * register_netdev - register a network device 10335 * @dev: device to register 10336 * 10337 * Take a completed network device structure and add it to the kernel 10338 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 10339 * chain. 0 is returned on success. A negative errno code is returned 10340 * on a failure to set up the device, or if the name is a duplicate. 10341 * 10342 * This is a wrapper around register_netdevice that takes the rtnl semaphore 10343 * and expands the device name if you passed a format string to 10344 * alloc_netdev. 10345 */ 10346 int register_netdev(struct net_device *dev) 10347 { 10348 int err; 10349 10350 if (rtnl_lock_killable()) 10351 return -EINTR; 10352 err = register_netdevice(dev); 10353 rtnl_unlock(); 10354 return err; 10355 } 10356 EXPORT_SYMBOL(register_netdev); 10357 10358 int netdev_refcnt_read(const struct net_device *dev) 10359 { 10360 #ifdef CONFIG_PCPU_DEV_REFCNT 10361 int i, refcnt = 0; 10362 10363 for_each_possible_cpu(i) 10364 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); 10365 return refcnt; 10366 #else 10367 return refcount_read(&dev->dev_refcnt); 10368 #endif 10369 } 10370 EXPORT_SYMBOL(netdev_refcnt_read); 10371 10372 int netdev_unregister_timeout_secs __read_mostly = 10; 10373 10374 #define WAIT_REFS_MIN_MSECS 1 10375 #define WAIT_REFS_MAX_MSECS 250 10376 /** 10377 * netdev_wait_allrefs_any - wait until all references are gone. 10378 * @list: list of net_devices to wait on 10379 * 10380 * This is called when unregistering network devices. 10381 * 10382 * Any protocol or device that holds a reference should register 10383 * for netdevice notification, and cleanup and put back the 10384 * reference if they receive an UNREGISTER event. 10385 * We can get stuck here if buggy protocols don't correctly 10386 * call dev_put. 10387 */ 10388 static struct net_device *netdev_wait_allrefs_any(struct list_head *list) 10389 { 10390 unsigned long rebroadcast_time, warning_time; 10391 struct net_device *dev; 10392 int wait = 0; 10393 10394 rebroadcast_time = warning_time = jiffies; 10395 10396 list_for_each_entry(dev, list, todo_list) 10397 if (netdev_refcnt_read(dev) == 1) 10398 return dev; 10399 10400 while (true) { 10401 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { 10402 rtnl_lock(); 10403 10404 /* Rebroadcast unregister notification */ 10405 list_for_each_entry(dev, list, todo_list) 10406 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10407 10408 __rtnl_unlock(); 10409 rcu_barrier(); 10410 rtnl_lock(); 10411 10412 list_for_each_entry(dev, list, todo_list) 10413 if (test_bit(__LINK_STATE_LINKWATCH_PENDING, 10414 &dev->state)) { 10415 /* We must not have linkwatch events 10416 * pending on unregister. If this 10417 * happens, we simply run the queue 10418 * unscheduled, resulting in a noop 10419 * for this device. 10420 */ 10421 linkwatch_run_queue(); 10422 break; 10423 } 10424 10425 __rtnl_unlock(); 10426 10427 rebroadcast_time = jiffies; 10428 } 10429 10430 if (!wait) { 10431 rcu_barrier(); 10432 wait = WAIT_REFS_MIN_MSECS; 10433 } else { 10434 msleep(wait); 10435 wait = min(wait << 1, WAIT_REFS_MAX_MSECS); 10436 } 10437 10438 list_for_each_entry(dev, list, todo_list) 10439 if (netdev_refcnt_read(dev) == 1) 10440 return dev; 10441 10442 if (time_after(jiffies, warning_time + 10443 READ_ONCE(netdev_unregister_timeout_secs) * HZ)) { 10444 list_for_each_entry(dev, list, todo_list) { 10445 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", 10446 dev->name, netdev_refcnt_read(dev)); 10447 ref_tracker_dir_print(&dev->refcnt_tracker, 10); 10448 } 10449 10450 warning_time = jiffies; 10451 } 10452 } 10453 } 10454 10455 /* The sequence is: 10456 * 10457 * rtnl_lock(); 10458 * ... 10459 * register_netdevice(x1); 10460 * register_netdevice(x2); 10461 * ... 10462 * unregister_netdevice(y1); 10463 * unregister_netdevice(y2); 10464 * ... 10465 * rtnl_unlock(); 10466 * free_netdev(y1); 10467 * free_netdev(y2); 10468 * 10469 * We are invoked by rtnl_unlock(). 10470 * This allows us to deal with problems: 10471 * 1) We can delete sysfs objects which invoke hotplug 10472 * without deadlocking with linkwatch via keventd. 10473 * 2) Since we run with the RTNL semaphore not held, we can sleep 10474 * safely in order to wait for the netdev refcnt to drop to zero. 10475 * 10476 * We must not return until all unregister events added during 10477 * the interval the lock was held have been completed. 10478 */ 10479 void netdev_run_todo(void) 10480 { 10481 struct net_device *dev, *tmp; 10482 struct list_head list; 10483 #ifdef CONFIG_LOCKDEP 10484 struct list_head unlink_list; 10485 10486 list_replace_init(&net_unlink_list, &unlink_list); 10487 10488 while (!list_empty(&unlink_list)) { 10489 struct net_device *dev = list_first_entry(&unlink_list, 10490 struct net_device, 10491 unlink_list); 10492 list_del_init(&dev->unlink_list); 10493 dev->nested_level = dev->lower_level - 1; 10494 } 10495 #endif 10496 10497 /* Snapshot list, allow later requests */ 10498 list_replace_init(&net_todo_list, &list); 10499 10500 __rtnl_unlock(); 10501 10502 /* Wait for rcu callbacks to finish before next phase */ 10503 if (!list_empty(&list)) 10504 rcu_barrier(); 10505 10506 list_for_each_entry_safe(dev, tmp, &list, todo_list) { 10507 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { 10508 netdev_WARN(dev, "run_todo but not unregistering\n"); 10509 list_del(&dev->todo_list); 10510 continue; 10511 } 10512 10513 write_lock(&dev_base_lock); 10514 dev->reg_state = NETREG_UNREGISTERED; 10515 write_unlock(&dev_base_lock); 10516 linkwatch_forget_dev(dev); 10517 } 10518 10519 while (!list_empty(&list)) { 10520 dev = netdev_wait_allrefs_any(&list); 10521 list_del(&dev->todo_list); 10522 10523 /* paranoia */ 10524 BUG_ON(netdev_refcnt_read(dev) != 1); 10525 BUG_ON(!list_empty(&dev->ptype_all)); 10526 BUG_ON(!list_empty(&dev->ptype_specific)); 10527 WARN_ON(rcu_access_pointer(dev->ip_ptr)); 10528 WARN_ON(rcu_access_pointer(dev->ip6_ptr)); 10529 10530 netdev_do_free_pcpu_stats(dev); 10531 if (dev->priv_destructor) 10532 dev->priv_destructor(dev); 10533 if (dev->needs_free_netdev) 10534 free_netdev(dev); 10535 10536 if (atomic_dec_and_test(&dev_net(dev)->dev_unreg_count)) 10537 wake_up(&netdev_unregistering_wq); 10538 10539 /* Free network device */ 10540 kobject_put(&dev->dev.kobj); 10541 } 10542 } 10543 10544 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has 10545 * all the same fields in the same order as net_device_stats, with only 10546 * the type differing, but rtnl_link_stats64 may have additional fields 10547 * at the end for newer counters. 10548 */ 10549 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, 10550 const struct net_device_stats *netdev_stats) 10551 { 10552 size_t i, n = sizeof(*netdev_stats) / sizeof(atomic_long_t); 10553 const atomic_long_t *src = (atomic_long_t *)netdev_stats; 10554 u64 *dst = (u64 *)stats64; 10555 10556 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); 10557 for (i = 0; i < n; i++) 10558 dst[i] = (unsigned long)atomic_long_read(&src[i]); 10559 /* zero out counters that only exist in rtnl_link_stats64 */ 10560 memset((char *)stats64 + n * sizeof(u64), 0, 10561 sizeof(*stats64) - n * sizeof(u64)); 10562 } 10563 EXPORT_SYMBOL(netdev_stats_to_stats64); 10564 10565 struct net_device_core_stats __percpu *netdev_core_stats_alloc(struct net_device *dev) 10566 { 10567 struct net_device_core_stats __percpu *p; 10568 10569 p = alloc_percpu_gfp(struct net_device_core_stats, 10570 GFP_ATOMIC | __GFP_NOWARN); 10571 10572 if (p && cmpxchg(&dev->core_stats, NULL, p)) 10573 free_percpu(p); 10574 10575 /* This READ_ONCE() pairs with the cmpxchg() above */ 10576 return READ_ONCE(dev->core_stats); 10577 } 10578 EXPORT_SYMBOL(netdev_core_stats_alloc); 10579 10580 /** 10581 * dev_get_stats - get network device statistics 10582 * @dev: device to get statistics from 10583 * @storage: place to store stats 10584 * 10585 * Get network statistics from device. Return @storage. 10586 * The device driver may provide its own method by setting 10587 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; 10588 * otherwise the internal statistics structure is used. 10589 */ 10590 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, 10591 struct rtnl_link_stats64 *storage) 10592 { 10593 const struct net_device_ops *ops = dev->netdev_ops; 10594 const struct net_device_core_stats __percpu *p; 10595 10596 if (ops->ndo_get_stats64) { 10597 memset(storage, 0, sizeof(*storage)); 10598 ops->ndo_get_stats64(dev, storage); 10599 } else if (ops->ndo_get_stats) { 10600 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); 10601 } else { 10602 netdev_stats_to_stats64(storage, &dev->stats); 10603 } 10604 10605 /* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */ 10606 p = READ_ONCE(dev->core_stats); 10607 if (p) { 10608 const struct net_device_core_stats *core_stats; 10609 int i; 10610 10611 for_each_possible_cpu(i) { 10612 core_stats = per_cpu_ptr(p, i); 10613 storage->rx_dropped += READ_ONCE(core_stats->rx_dropped); 10614 storage->tx_dropped += READ_ONCE(core_stats->tx_dropped); 10615 storage->rx_nohandler += READ_ONCE(core_stats->rx_nohandler); 10616 storage->rx_otherhost_dropped += READ_ONCE(core_stats->rx_otherhost_dropped); 10617 } 10618 } 10619 return storage; 10620 } 10621 EXPORT_SYMBOL(dev_get_stats); 10622 10623 /** 10624 * dev_fetch_sw_netstats - get per-cpu network device statistics 10625 * @s: place to store stats 10626 * @netstats: per-cpu network stats to read from 10627 * 10628 * Read per-cpu network statistics and populate the related fields in @s. 10629 */ 10630 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s, 10631 const struct pcpu_sw_netstats __percpu *netstats) 10632 { 10633 int cpu; 10634 10635 for_each_possible_cpu(cpu) { 10636 u64 rx_packets, rx_bytes, tx_packets, tx_bytes; 10637 const struct pcpu_sw_netstats *stats; 10638 unsigned int start; 10639 10640 stats = per_cpu_ptr(netstats, cpu); 10641 do { 10642 start = u64_stats_fetch_begin(&stats->syncp); 10643 rx_packets = u64_stats_read(&stats->rx_packets); 10644 rx_bytes = u64_stats_read(&stats->rx_bytes); 10645 tx_packets = u64_stats_read(&stats->tx_packets); 10646 tx_bytes = u64_stats_read(&stats->tx_bytes); 10647 } while (u64_stats_fetch_retry(&stats->syncp, start)); 10648 10649 s->rx_packets += rx_packets; 10650 s->rx_bytes += rx_bytes; 10651 s->tx_packets += tx_packets; 10652 s->tx_bytes += tx_bytes; 10653 } 10654 } 10655 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats); 10656 10657 /** 10658 * dev_get_tstats64 - ndo_get_stats64 implementation 10659 * @dev: device to get statistics from 10660 * @s: place to store stats 10661 * 10662 * Populate @s from dev->stats and dev->tstats. Can be used as 10663 * ndo_get_stats64() callback. 10664 */ 10665 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s) 10666 { 10667 netdev_stats_to_stats64(s, &dev->stats); 10668 dev_fetch_sw_netstats(s, dev->tstats); 10669 } 10670 EXPORT_SYMBOL_GPL(dev_get_tstats64); 10671 10672 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) 10673 { 10674 struct netdev_queue *queue = dev_ingress_queue(dev); 10675 10676 #ifdef CONFIG_NET_CLS_ACT 10677 if (queue) 10678 return queue; 10679 queue = kzalloc(sizeof(*queue), GFP_KERNEL); 10680 if (!queue) 10681 return NULL; 10682 netdev_init_one_queue(dev, queue, NULL); 10683 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); 10684 RCU_INIT_POINTER(queue->qdisc_sleeping, &noop_qdisc); 10685 rcu_assign_pointer(dev->ingress_queue, queue); 10686 #endif 10687 return queue; 10688 } 10689 10690 static const struct ethtool_ops default_ethtool_ops; 10691 10692 void netdev_set_default_ethtool_ops(struct net_device *dev, 10693 const struct ethtool_ops *ops) 10694 { 10695 if (dev->ethtool_ops == &default_ethtool_ops) 10696 dev->ethtool_ops = ops; 10697 } 10698 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); 10699 10700 /** 10701 * netdev_sw_irq_coalesce_default_on() - enable SW IRQ coalescing by default 10702 * @dev: netdev to enable the IRQ coalescing on 10703 * 10704 * Sets a conservative default for SW IRQ coalescing. Users can use 10705 * sysfs attributes to override the default values. 10706 */ 10707 void netdev_sw_irq_coalesce_default_on(struct net_device *dev) 10708 { 10709 WARN_ON(dev->reg_state == NETREG_REGISTERED); 10710 10711 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) { 10712 dev->gro_flush_timeout = 20000; 10713 dev->napi_defer_hard_irqs = 1; 10714 } 10715 } 10716 EXPORT_SYMBOL_GPL(netdev_sw_irq_coalesce_default_on); 10717 10718 void netdev_freemem(struct net_device *dev) 10719 { 10720 char *addr = (char *)dev - dev->padded; 10721 10722 kvfree(addr); 10723 } 10724 10725 /** 10726 * alloc_netdev_mqs - allocate network device 10727 * @sizeof_priv: size of private data to allocate space for 10728 * @name: device name format string 10729 * @name_assign_type: origin of device name 10730 * @setup: callback to initialize device 10731 * @txqs: the number of TX subqueues to allocate 10732 * @rxqs: the number of RX subqueues to allocate 10733 * 10734 * Allocates a struct net_device with private data area for driver use 10735 * and performs basic initialization. Also allocates subqueue structs 10736 * for each queue on the device. 10737 */ 10738 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, 10739 unsigned char name_assign_type, 10740 void (*setup)(struct net_device *), 10741 unsigned int txqs, unsigned int rxqs) 10742 { 10743 struct net_device *dev; 10744 unsigned int alloc_size; 10745 struct net_device *p; 10746 10747 BUG_ON(strlen(name) >= sizeof(dev->name)); 10748 10749 if (txqs < 1) { 10750 pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); 10751 return NULL; 10752 } 10753 10754 if (rxqs < 1) { 10755 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); 10756 return NULL; 10757 } 10758 10759 alloc_size = sizeof(struct net_device); 10760 if (sizeof_priv) { 10761 /* ensure 32-byte alignment of private area */ 10762 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); 10763 alloc_size += sizeof_priv; 10764 } 10765 /* ensure 32-byte alignment of whole construct */ 10766 alloc_size += NETDEV_ALIGN - 1; 10767 10768 p = kvzalloc(alloc_size, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10769 if (!p) 10770 return NULL; 10771 10772 dev = PTR_ALIGN(p, NETDEV_ALIGN); 10773 dev->padded = (char *)dev - (char *)p; 10774 10775 ref_tracker_dir_init(&dev->refcnt_tracker, 128, name); 10776 #ifdef CONFIG_PCPU_DEV_REFCNT 10777 dev->pcpu_refcnt = alloc_percpu(int); 10778 if (!dev->pcpu_refcnt) 10779 goto free_dev; 10780 __dev_hold(dev); 10781 #else 10782 refcount_set(&dev->dev_refcnt, 1); 10783 #endif 10784 10785 if (dev_addr_init(dev)) 10786 goto free_pcpu; 10787 10788 dev_mc_init(dev); 10789 dev_uc_init(dev); 10790 10791 dev_net_set(dev, &init_net); 10792 10793 dev->gso_max_size = GSO_LEGACY_MAX_SIZE; 10794 dev->xdp_zc_max_segs = 1; 10795 dev->gso_max_segs = GSO_MAX_SEGS; 10796 dev->gro_max_size = GRO_LEGACY_MAX_SIZE; 10797 dev->gso_ipv4_max_size = GSO_LEGACY_MAX_SIZE; 10798 dev->gro_ipv4_max_size = GRO_LEGACY_MAX_SIZE; 10799 dev->tso_max_size = TSO_LEGACY_MAX_SIZE; 10800 dev->tso_max_segs = TSO_MAX_SEGS; 10801 dev->upper_level = 1; 10802 dev->lower_level = 1; 10803 #ifdef CONFIG_LOCKDEP 10804 dev->nested_level = 0; 10805 INIT_LIST_HEAD(&dev->unlink_list); 10806 #endif 10807 10808 INIT_LIST_HEAD(&dev->napi_list); 10809 INIT_LIST_HEAD(&dev->unreg_list); 10810 INIT_LIST_HEAD(&dev->close_list); 10811 INIT_LIST_HEAD(&dev->link_watch_list); 10812 INIT_LIST_HEAD(&dev->adj_list.upper); 10813 INIT_LIST_HEAD(&dev->adj_list.lower); 10814 INIT_LIST_HEAD(&dev->ptype_all); 10815 INIT_LIST_HEAD(&dev->ptype_specific); 10816 INIT_LIST_HEAD(&dev->net_notifier_list); 10817 #ifdef CONFIG_NET_SCHED 10818 hash_init(dev->qdisc_hash); 10819 #endif 10820 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; 10821 setup(dev); 10822 10823 if (!dev->tx_queue_len) { 10824 dev->priv_flags |= IFF_NO_QUEUE; 10825 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; 10826 } 10827 10828 dev->num_tx_queues = txqs; 10829 dev->real_num_tx_queues = txqs; 10830 if (netif_alloc_netdev_queues(dev)) 10831 goto free_all; 10832 10833 dev->num_rx_queues = rxqs; 10834 dev->real_num_rx_queues = rxqs; 10835 if (netif_alloc_rx_queues(dev)) 10836 goto free_all; 10837 10838 strcpy(dev->name, name); 10839 dev->name_assign_type = name_assign_type; 10840 dev->group = INIT_NETDEV_GROUP; 10841 if (!dev->ethtool_ops) 10842 dev->ethtool_ops = &default_ethtool_ops; 10843 10844 nf_hook_netdev_init(dev); 10845 10846 return dev; 10847 10848 free_all: 10849 free_netdev(dev); 10850 return NULL; 10851 10852 free_pcpu: 10853 #ifdef CONFIG_PCPU_DEV_REFCNT 10854 free_percpu(dev->pcpu_refcnt); 10855 free_dev: 10856 #endif 10857 netdev_freemem(dev); 10858 return NULL; 10859 } 10860 EXPORT_SYMBOL(alloc_netdev_mqs); 10861 10862 /** 10863 * free_netdev - free network device 10864 * @dev: device 10865 * 10866 * This function does the last stage of destroying an allocated device 10867 * interface. The reference to the device object is released. If this 10868 * is the last reference then it will be freed.Must be called in process 10869 * context. 10870 */ 10871 void free_netdev(struct net_device *dev) 10872 { 10873 struct napi_struct *p, *n; 10874 10875 might_sleep(); 10876 10877 /* When called immediately after register_netdevice() failed the unwind 10878 * handling may still be dismantling the device. Handle that case by 10879 * deferring the free. 10880 */ 10881 if (dev->reg_state == NETREG_UNREGISTERING) { 10882 ASSERT_RTNL(); 10883 dev->needs_free_netdev = true; 10884 return; 10885 } 10886 10887 netif_free_tx_queues(dev); 10888 netif_free_rx_queues(dev); 10889 10890 kfree(rcu_dereference_protected(dev->ingress_queue, 1)); 10891 10892 /* Flush device addresses */ 10893 dev_addr_flush(dev); 10894 10895 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) 10896 netif_napi_del(p); 10897 10898 ref_tracker_dir_exit(&dev->refcnt_tracker); 10899 #ifdef CONFIG_PCPU_DEV_REFCNT 10900 free_percpu(dev->pcpu_refcnt); 10901 dev->pcpu_refcnt = NULL; 10902 #endif 10903 free_percpu(dev->core_stats); 10904 dev->core_stats = NULL; 10905 free_percpu(dev->xdp_bulkq); 10906 dev->xdp_bulkq = NULL; 10907 10908 /* Compatibility with error handling in drivers */ 10909 if (dev->reg_state == NETREG_UNINITIALIZED) { 10910 netdev_freemem(dev); 10911 return; 10912 } 10913 10914 BUG_ON(dev->reg_state != NETREG_UNREGISTERED); 10915 dev->reg_state = NETREG_RELEASED; 10916 10917 /* will free via device release */ 10918 put_device(&dev->dev); 10919 } 10920 EXPORT_SYMBOL(free_netdev); 10921 10922 /** 10923 * synchronize_net - Synchronize with packet receive processing 10924 * 10925 * Wait for packets currently being received to be done. 10926 * Does not block later packets from starting. 10927 */ 10928 void synchronize_net(void) 10929 { 10930 might_sleep(); 10931 if (rtnl_is_locked()) 10932 synchronize_rcu_expedited(); 10933 else 10934 synchronize_rcu(); 10935 } 10936 EXPORT_SYMBOL(synchronize_net); 10937 10938 /** 10939 * unregister_netdevice_queue - remove device from the kernel 10940 * @dev: device 10941 * @head: list 10942 * 10943 * This function shuts down a device interface and removes it 10944 * from the kernel tables. 10945 * If head not NULL, device is queued to be unregistered later. 10946 * 10947 * Callers must hold the rtnl semaphore. You may want 10948 * unregister_netdev() instead of this. 10949 */ 10950 10951 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) 10952 { 10953 ASSERT_RTNL(); 10954 10955 if (head) { 10956 list_move_tail(&dev->unreg_list, head); 10957 } else { 10958 LIST_HEAD(single); 10959 10960 list_add(&dev->unreg_list, &single); 10961 unregister_netdevice_many(&single); 10962 } 10963 } 10964 EXPORT_SYMBOL(unregister_netdevice_queue); 10965 10966 void unregister_netdevice_many_notify(struct list_head *head, 10967 u32 portid, const struct nlmsghdr *nlh) 10968 { 10969 struct net_device *dev, *tmp; 10970 LIST_HEAD(close_head); 10971 10972 BUG_ON(dev_boot_phase); 10973 ASSERT_RTNL(); 10974 10975 if (list_empty(head)) 10976 return; 10977 10978 list_for_each_entry_safe(dev, tmp, head, unreg_list) { 10979 /* Some devices call without registering 10980 * for initialization unwind. Remove those 10981 * devices and proceed with the remaining. 10982 */ 10983 if (dev->reg_state == NETREG_UNINITIALIZED) { 10984 pr_debug("unregister_netdevice: device %s/%p never was registered\n", 10985 dev->name, dev); 10986 10987 WARN_ON(1); 10988 list_del(&dev->unreg_list); 10989 continue; 10990 } 10991 dev->dismantle = true; 10992 BUG_ON(dev->reg_state != NETREG_REGISTERED); 10993 } 10994 10995 /* If device is running, close it first. */ 10996 list_for_each_entry(dev, head, unreg_list) 10997 list_add_tail(&dev->close_list, &close_head); 10998 dev_close_many(&close_head, true); 10999 11000 list_for_each_entry(dev, head, unreg_list) { 11001 /* And unlink it from device chain. */ 11002 write_lock(&dev_base_lock); 11003 unlist_netdevice(dev, false); 11004 dev->reg_state = NETREG_UNREGISTERING; 11005 write_unlock(&dev_base_lock); 11006 } 11007 flush_all_backlogs(); 11008 11009 synchronize_net(); 11010 11011 list_for_each_entry(dev, head, unreg_list) { 11012 struct sk_buff *skb = NULL; 11013 11014 /* Shutdown queueing discipline. */ 11015 dev_shutdown(dev); 11016 dev_tcx_uninstall(dev); 11017 dev_xdp_uninstall(dev); 11018 bpf_dev_bound_netdev_unregister(dev); 11019 11020 netdev_offload_xstats_disable_all(dev); 11021 11022 /* Notify protocols, that we are about to destroy 11023 * this device. They should clean all the things. 11024 */ 11025 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 11026 11027 if (!dev->rtnl_link_ops || 11028 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 11029 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0, 11030 GFP_KERNEL, NULL, 0, 11031 portid, nlh); 11032 11033 /* 11034 * Flush the unicast and multicast chains 11035 */ 11036 dev_uc_flush(dev); 11037 dev_mc_flush(dev); 11038 11039 netdev_name_node_alt_flush(dev); 11040 netdev_name_node_free(dev->name_node); 11041 11042 call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev); 11043 11044 if (dev->netdev_ops->ndo_uninit) 11045 dev->netdev_ops->ndo_uninit(dev); 11046 11047 if (skb) 11048 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL, portid, nlh); 11049 11050 /* Notifier chain MUST detach us all upper devices. */ 11051 WARN_ON(netdev_has_any_upper_dev(dev)); 11052 WARN_ON(netdev_has_any_lower_dev(dev)); 11053 11054 /* Remove entries from kobject tree */ 11055 netdev_unregister_kobject(dev); 11056 #ifdef CONFIG_XPS 11057 /* Remove XPS queueing entries */ 11058 netif_reset_xps_queues_gt(dev, 0); 11059 #endif 11060 } 11061 11062 synchronize_net(); 11063 11064 list_for_each_entry(dev, head, unreg_list) { 11065 netdev_put(dev, &dev->dev_registered_tracker); 11066 net_set_todo(dev); 11067 } 11068 11069 list_del(head); 11070 } 11071 11072 /** 11073 * unregister_netdevice_many - unregister many devices 11074 * @head: list of devices 11075 * 11076 * Note: As most callers use a stack allocated list_head, 11077 * we force a list_del() to make sure stack wont be corrupted later. 11078 */ 11079 void unregister_netdevice_many(struct list_head *head) 11080 { 11081 unregister_netdevice_many_notify(head, 0, NULL); 11082 } 11083 EXPORT_SYMBOL(unregister_netdevice_many); 11084 11085 /** 11086 * unregister_netdev - remove device from the kernel 11087 * @dev: device 11088 * 11089 * This function shuts down a device interface and removes it 11090 * from the kernel tables. 11091 * 11092 * This is just a wrapper for unregister_netdevice that takes 11093 * the rtnl semaphore. In general you want to use this and not 11094 * unregister_netdevice. 11095 */ 11096 void unregister_netdev(struct net_device *dev) 11097 { 11098 rtnl_lock(); 11099 unregister_netdevice(dev); 11100 rtnl_unlock(); 11101 } 11102 EXPORT_SYMBOL(unregister_netdev); 11103 11104 /** 11105 * __dev_change_net_namespace - move device to different nethost namespace 11106 * @dev: device 11107 * @net: network namespace 11108 * @pat: If not NULL name pattern to try if the current device name 11109 * is already taken in the destination network namespace. 11110 * @new_ifindex: If not zero, specifies device index in the target 11111 * namespace. 11112 * 11113 * This function shuts down a device interface and moves it 11114 * to a new network namespace. On success 0 is returned, on 11115 * a failure a netagive errno code is returned. 11116 * 11117 * Callers must hold the rtnl semaphore. 11118 */ 11119 11120 int __dev_change_net_namespace(struct net_device *dev, struct net *net, 11121 const char *pat, int new_ifindex) 11122 { 11123 struct netdev_name_node *name_node; 11124 struct net *net_old = dev_net(dev); 11125 char new_name[IFNAMSIZ] = {}; 11126 int err, new_nsid; 11127 11128 ASSERT_RTNL(); 11129 11130 /* Don't allow namespace local devices to be moved. */ 11131 err = -EINVAL; 11132 if (dev->features & NETIF_F_NETNS_LOCAL) 11133 goto out; 11134 11135 /* Ensure the device has been registrered */ 11136 if (dev->reg_state != NETREG_REGISTERED) 11137 goto out; 11138 11139 /* Get out if there is nothing todo */ 11140 err = 0; 11141 if (net_eq(net_old, net)) 11142 goto out; 11143 11144 /* Pick the destination device name, and ensure 11145 * we can use it in the destination network namespace. 11146 */ 11147 err = -EEXIST; 11148 if (netdev_name_in_use(net, dev->name)) { 11149 /* We get here if we can't use the current device name */ 11150 if (!pat) 11151 goto out; 11152 err = dev_prep_valid_name(net, dev, pat, new_name); 11153 if (err < 0) 11154 goto out; 11155 } 11156 /* Check that none of the altnames conflicts. */ 11157 err = -EEXIST; 11158 netdev_for_each_altname(dev, name_node) 11159 if (netdev_name_in_use(net, name_node->name)) 11160 goto out; 11161 11162 /* Check that new_ifindex isn't used yet. */ 11163 if (new_ifindex) { 11164 err = dev_index_reserve(net, new_ifindex); 11165 if (err < 0) 11166 goto out; 11167 } else { 11168 /* If there is an ifindex conflict assign a new one */ 11169 err = dev_index_reserve(net, dev->ifindex); 11170 if (err == -EBUSY) 11171 err = dev_index_reserve(net, 0); 11172 if (err < 0) 11173 goto out; 11174 new_ifindex = err; 11175 } 11176 11177 /* 11178 * And now a mini version of register_netdevice unregister_netdevice. 11179 */ 11180 11181 /* If device is running close it first. */ 11182 dev_close(dev); 11183 11184 /* And unlink it from device chain */ 11185 unlist_netdevice(dev, true); 11186 11187 synchronize_net(); 11188 11189 /* Shutdown queueing discipline. */ 11190 dev_shutdown(dev); 11191 11192 /* Notify protocols, that we are about to destroy 11193 * this device. They should clean all the things. 11194 * 11195 * Note that dev->reg_state stays at NETREG_REGISTERED. 11196 * This is wanted because this way 8021q and macvlan know 11197 * the device is just moving and can keep their slaves up. 11198 */ 11199 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 11200 rcu_barrier(); 11201 11202 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL); 11203 11204 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid, 11205 new_ifindex); 11206 11207 /* 11208 * Flush the unicast and multicast chains 11209 */ 11210 dev_uc_flush(dev); 11211 dev_mc_flush(dev); 11212 11213 /* Send a netdev-removed uevent to the old namespace */ 11214 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); 11215 netdev_adjacent_del_links(dev); 11216 11217 /* Move per-net netdevice notifiers that are following the netdevice */ 11218 move_netdevice_notifiers_dev_net(dev, net); 11219 11220 /* Actually switch the network namespace */ 11221 dev_net_set(dev, net); 11222 dev->ifindex = new_ifindex; 11223 11224 /* Send a netdev-add uevent to the new namespace */ 11225 kobject_uevent(&dev->dev.kobj, KOBJ_ADD); 11226 netdev_adjacent_add_links(dev); 11227 11228 if (new_name[0]) /* Rename the netdev to prepared name */ 11229 strscpy(dev->name, new_name, IFNAMSIZ); 11230 11231 /* Fixup kobjects */ 11232 err = device_rename(&dev->dev, dev->name); 11233 WARN_ON(err); 11234 11235 /* Adapt owner in case owning user namespace of target network 11236 * namespace is different from the original one. 11237 */ 11238 err = netdev_change_owner(dev, net_old, net); 11239 WARN_ON(err); 11240 11241 /* Add the device back in the hashes */ 11242 list_netdevice(dev); 11243 11244 /* Notify protocols, that a new device appeared. */ 11245 call_netdevice_notifiers(NETDEV_REGISTER, dev); 11246 11247 /* 11248 * Prevent userspace races by waiting until the network 11249 * device is fully setup before sending notifications. 11250 */ 11251 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL); 11252 11253 synchronize_net(); 11254 err = 0; 11255 out: 11256 return err; 11257 } 11258 EXPORT_SYMBOL_GPL(__dev_change_net_namespace); 11259 11260 static int dev_cpu_dead(unsigned int oldcpu) 11261 { 11262 struct sk_buff **list_skb; 11263 struct sk_buff *skb; 11264 unsigned int cpu; 11265 struct softnet_data *sd, *oldsd, *remsd = NULL; 11266 11267 local_irq_disable(); 11268 cpu = smp_processor_id(); 11269 sd = &per_cpu(softnet_data, cpu); 11270 oldsd = &per_cpu(softnet_data, oldcpu); 11271 11272 /* Find end of our completion_queue. */ 11273 list_skb = &sd->completion_queue; 11274 while (*list_skb) 11275 list_skb = &(*list_skb)->next; 11276 /* Append completion queue from offline CPU. */ 11277 *list_skb = oldsd->completion_queue; 11278 oldsd->completion_queue = NULL; 11279 11280 /* Append output queue from offline CPU. */ 11281 if (oldsd->output_queue) { 11282 *sd->output_queue_tailp = oldsd->output_queue; 11283 sd->output_queue_tailp = oldsd->output_queue_tailp; 11284 oldsd->output_queue = NULL; 11285 oldsd->output_queue_tailp = &oldsd->output_queue; 11286 } 11287 /* Append NAPI poll list from offline CPU, with one exception : 11288 * process_backlog() must be called by cpu owning percpu backlog. 11289 * We properly handle process_queue & input_pkt_queue later. 11290 */ 11291 while (!list_empty(&oldsd->poll_list)) { 11292 struct napi_struct *napi = list_first_entry(&oldsd->poll_list, 11293 struct napi_struct, 11294 poll_list); 11295 11296 list_del_init(&napi->poll_list); 11297 if (napi->poll == process_backlog) 11298 napi->state = 0; 11299 else 11300 ____napi_schedule(sd, napi); 11301 } 11302 11303 raise_softirq_irqoff(NET_TX_SOFTIRQ); 11304 local_irq_enable(); 11305 11306 #ifdef CONFIG_RPS 11307 remsd = oldsd->rps_ipi_list; 11308 oldsd->rps_ipi_list = NULL; 11309 #endif 11310 /* send out pending IPI's on offline CPU */ 11311 net_rps_send_ipi(remsd); 11312 11313 /* Process offline CPU's input_pkt_queue */ 11314 while ((skb = __skb_dequeue(&oldsd->process_queue))) { 11315 netif_rx(skb); 11316 input_queue_head_incr(oldsd); 11317 } 11318 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { 11319 netif_rx(skb); 11320 input_queue_head_incr(oldsd); 11321 } 11322 11323 return 0; 11324 } 11325 11326 /** 11327 * netdev_increment_features - increment feature set by one 11328 * @all: current feature set 11329 * @one: new feature set 11330 * @mask: mask feature set 11331 * 11332 * Computes a new feature set after adding a device with feature set 11333 * @one to the master device with current feature set @all. Will not 11334 * enable anything that is off in @mask. Returns the new feature set. 11335 */ 11336 netdev_features_t netdev_increment_features(netdev_features_t all, 11337 netdev_features_t one, netdev_features_t mask) 11338 { 11339 if (mask & NETIF_F_HW_CSUM) 11340 mask |= NETIF_F_CSUM_MASK; 11341 mask |= NETIF_F_VLAN_CHALLENGED; 11342 11343 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; 11344 all &= one | ~NETIF_F_ALL_FOR_ALL; 11345 11346 /* If one device supports hw checksumming, set for all. */ 11347 if (all & NETIF_F_HW_CSUM) 11348 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); 11349 11350 return all; 11351 } 11352 EXPORT_SYMBOL(netdev_increment_features); 11353 11354 static struct hlist_head * __net_init netdev_create_hash(void) 11355 { 11356 int i; 11357 struct hlist_head *hash; 11358 11359 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL); 11360 if (hash != NULL) 11361 for (i = 0; i < NETDEV_HASHENTRIES; i++) 11362 INIT_HLIST_HEAD(&hash[i]); 11363 11364 return hash; 11365 } 11366 11367 /* Initialize per network namespace state */ 11368 static int __net_init netdev_init(struct net *net) 11369 { 11370 BUILD_BUG_ON(GRO_HASH_BUCKETS > 11371 8 * sizeof_field(struct napi_struct, gro_bitmask)); 11372 11373 INIT_LIST_HEAD(&net->dev_base_head); 11374 11375 net->dev_name_head = netdev_create_hash(); 11376 if (net->dev_name_head == NULL) 11377 goto err_name; 11378 11379 net->dev_index_head = netdev_create_hash(); 11380 if (net->dev_index_head == NULL) 11381 goto err_idx; 11382 11383 xa_init_flags(&net->dev_by_index, XA_FLAGS_ALLOC1); 11384 11385 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain); 11386 11387 return 0; 11388 11389 err_idx: 11390 kfree(net->dev_name_head); 11391 err_name: 11392 return -ENOMEM; 11393 } 11394 11395 /** 11396 * netdev_drivername - network driver for the device 11397 * @dev: network device 11398 * 11399 * Determine network driver for device. 11400 */ 11401 const char *netdev_drivername(const struct net_device *dev) 11402 { 11403 const struct device_driver *driver; 11404 const struct device *parent; 11405 const char *empty = ""; 11406 11407 parent = dev->dev.parent; 11408 if (!parent) 11409 return empty; 11410 11411 driver = parent->driver; 11412 if (driver && driver->name) 11413 return driver->name; 11414 return empty; 11415 } 11416 11417 static void __netdev_printk(const char *level, const struct net_device *dev, 11418 struct va_format *vaf) 11419 { 11420 if (dev && dev->dev.parent) { 11421 dev_printk_emit(level[1] - '0', 11422 dev->dev.parent, 11423 "%s %s %s%s: %pV", 11424 dev_driver_string(dev->dev.parent), 11425 dev_name(dev->dev.parent), 11426 netdev_name(dev), netdev_reg_state(dev), 11427 vaf); 11428 } else if (dev) { 11429 printk("%s%s%s: %pV", 11430 level, netdev_name(dev), netdev_reg_state(dev), vaf); 11431 } else { 11432 printk("%s(NULL net_device): %pV", level, vaf); 11433 } 11434 } 11435 11436 void netdev_printk(const char *level, const struct net_device *dev, 11437 const char *format, ...) 11438 { 11439 struct va_format vaf; 11440 va_list args; 11441 11442 va_start(args, format); 11443 11444 vaf.fmt = format; 11445 vaf.va = &args; 11446 11447 __netdev_printk(level, dev, &vaf); 11448 11449 va_end(args); 11450 } 11451 EXPORT_SYMBOL(netdev_printk); 11452 11453 #define define_netdev_printk_level(func, level) \ 11454 void func(const struct net_device *dev, const char *fmt, ...) \ 11455 { \ 11456 struct va_format vaf; \ 11457 va_list args; \ 11458 \ 11459 va_start(args, fmt); \ 11460 \ 11461 vaf.fmt = fmt; \ 11462 vaf.va = &args; \ 11463 \ 11464 __netdev_printk(level, dev, &vaf); \ 11465 \ 11466 va_end(args); \ 11467 } \ 11468 EXPORT_SYMBOL(func); 11469 11470 define_netdev_printk_level(netdev_emerg, KERN_EMERG); 11471 define_netdev_printk_level(netdev_alert, KERN_ALERT); 11472 define_netdev_printk_level(netdev_crit, KERN_CRIT); 11473 define_netdev_printk_level(netdev_err, KERN_ERR); 11474 define_netdev_printk_level(netdev_warn, KERN_WARNING); 11475 define_netdev_printk_level(netdev_notice, KERN_NOTICE); 11476 define_netdev_printk_level(netdev_info, KERN_INFO); 11477 11478 static void __net_exit netdev_exit(struct net *net) 11479 { 11480 kfree(net->dev_name_head); 11481 kfree(net->dev_index_head); 11482 xa_destroy(&net->dev_by_index); 11483 if (net != &init_net) 11484 WARN_ON_ONCE(!list_empty(&net->dev_base_head)); 11485 } 11486 11487 static struct pernet_operations __net_initdata netdev_net_ops = { 11488 .init = netdev_init, 11489 .exit = netdev_exit, 11490 }; 11491 11492 static void __net_exit default_device_exit_net(struct net *net) 11493 { 11494 struct netdev_name_node *name_node, *tmp; 11495 struct net_device *dev, *aux; 11496 /* 11497 * Push all migratable network devices back to the 11498 * initial network namespace 11499 */ 11500 ASSERT_RTNL(); 11501 for_each_netdev_safe(net, dev, aux) { 11502 int err; 11503 char fb_name[IFNAMSIZ]; 11504 11505 /* Ignore unmoveable devices (i.e. loopback) */ 11506 if (dev->features & NETIF_F_NETNS_LOCAL) 11507 continue; 11508 11509 /* Leave virtual devices for the generic cleanup */ 11510 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund) 11511 continue; 11512 11513 /* Push remaining network devices to init_net */ 11514 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); 11515 if (netdev_name_in_use(&init_net, fb_name)) 11516 snprintf(fb_name, IFNAMSIZ, "dev%%d"); 11517 11518 netdev_for_each_altname_safe(dev, name_node, tmp) 11519 if (netdev_name_in_use(&init_net, name_node->name)) { 11520 netdev_name_node_del(name_node); 11521 synchronize_rcu(); 11522 __netdev_name_node_alt_destroy(name_node); 11523 } 11524 11525 err = dev_change_net_namespace(dev, &init_net, fb_name); 11526 if (err) { 11527 pr_emerg("%s: failed to move %s to init_net: %d\n", 11528 __func__, dev->name, err); 11529 BUG(); 11530 } 11531 } 11532 } 11533 11534 static void __net_exit default_device_exit_batch(struct list_head *net_list) 11535 { 11536 /* At exit all network devices most be removed from a network 11537 * namespace. Do this in the reverse order of registration. 11538 * Do this across as many network namespaces as possible to 11539 * improve batching efficiency. 11540 */ 11541 struct net_device *dev; 11542 struct net *net; 11543 LIST_HEAD(dev_kill_list); 11544 11545 rtnl_lock(); 11546 list_for_each_entry(net, net_list, exit_list) { 11547 default_device_exit_net(net); 11548 cond_resched(); 11549 } 11550 11551 list_for_each_entry(net, net_list, exit_list) { 11552 for_each_netdev_reverse(net, dev) { 11553 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) 11554 dev->rtnl_link_ops->dellink(dev, &dev_kill_list); 11555 else 11556 unregister_netdevice_queue(dev, &dev_kill_list); 11557 } 11558 } 11559 unregister_netdevice_many(&dev_kill_list); 11560 rtnl_unlock(); 11561 } 11562 11563 static struct pernet_operations __net_initdata default_device_ops = { 11564 .exit_batch = default_device_exit_batch, 11565 }; 11566 11567 /* 11568 * Initialize the DEV module. At boot time this walks the device list and 11569 * unhooks any devices that fail to initialise (normally hardware not 11570 * present) and leaves us with a valid list of present and active devices. 11571 * 11572 */ 11573 11574 /* 11575 * This is called single threaded during boot, so no need 11576 * to take the rtnl semaphore. 11577 */ 11578 static int __init net_dev_init(void) 11579 { 11580 int i, rc = -ENOMEM; 11581 11582 BUG_ON(!dev_boot_phase); 11583 11584 if (dev_proc_init()) 11585 goto out; 11586 11587 if (netdev_kobject_init()) 11588 goto out; 11589 11590 INIT_LIST_HEAD(&ptype_all); 11591 for (i = 0; i < PTYPE_HASH_SIZE; i++) 11592 INIT_LIST_HEAD(&ptype_base[i]); 11593 11594 if (register_pernet_subsys(&netdev_net_ops)) 11595 goto out; 11596 11597 /* 11598 * Initialise the packet receive queues. 11599 */ 11600 11601 for_each_possible_cpu(i) { 11602 struct work_struct *flush = per_cpu_ptr(&flush_works, i); 11603 struct softnet_data *sd = &per_cpu(softnet_data, i); 11604 11605 INIT_WORK(flush, flush_backlog); 11606 11607 skb_queue_head_init(&sd->input_pkt_queue); 11608 skb_queue_head_init(&sd->process_queue); 11609 #ifdef CONFIG_XFRM_OFFLOAD 11610 skb_queue_head_init(&sd->xfrm_backlog); 11611 #endif 11612 INIT_LIST_HEAD(&sd->poll_list); 11613 sd->output_queue_tailp = &sd->output_queue; 11614 #ifdef CONFIG_RPS 11615 INIT_CSD(&sd->csd, rps_trigger_softirq, sd); 11616 sd->cpu = i; 11617 #endif 11618 INIT_CSD(&sd->defer_csd, trigger_rx_softirq, sd); 11619 spin_lock_init(&sd->defer_lock); 11620 11621 init_gro_hash(&sd->backlog); 11622 sd->backlog.poll = process_backlog; 11623 sd->backlog.weight = weight_p; 11624 } 11625 11626 dev_boot_phase = 0; 11627 11628 /* The loopback device is special if any other network devices 11629 * is present in a network namespace the loopback device must 11630 * be present. Since we now dynamically allocate and free the 11631 * loopback device ensure this invariant is maintained by 11632 * keeping the loopback device as the first device on the 11633 * list of network devices. Ensuring the loopback devices 11634 * is the first device that appears and the last network device 11635 * that disappears. 11636 */ 11637 if (register_pernet_device(&loopback_net_ops)) 11638 goto out; 11639 11640 if (register_pernet_device(&default_device_ops)) 11641 goto out; 11642 11643 open_softirq(NET_TX_SOFTIRQ, net_tx_action); 11644 open_softirq(NET_RX_SOFTIRQ, net_rx_action); 11645 11646 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", 11647 NULL, dev_cpu_dead); 11648 WARN_ON(rc < 0); 11649 rc = 0; 11650 out: 11651 return rc; 11652 } 11653 11654 subsys_initcall(net_dev_init); 11655