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