1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * NET3 Protocol independent device support routines. 4 * 5 * Derived from the non IP parts of dev.c 1.0.19 6 * Authors: Ross Biro 7 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 8 * Mark Evans, <evansmp@uhura.aston.ac.uk> 9 * 10 * Additional Authors: 11 * Florian la Roche <rzsfl@rz.uni-sb.de> 12 * Alan Cox <gw4pts@gw4pts.ampr.org> 13 * David Hinds <dahinds@users.sourceforge.net> 14 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru> 15 * Adam Sulmicki <adam@cfar.umd.edu> 16 * Pekka Riikonen <priikone@poesidon.pspt.fi> 17 * 18 * Changes: 19 * D.J. Barrow : Fixed bug where dev->refcnt gets set 20 * to 2 if register_netdev gets called 21 * before net_dev_init & also removed a 22 * few lines of code in the process. 23 * Alan Cox : device private ioctl copies fields back. 24 * Alan Cox : Transmit queue code does relevant 25 * stunts to keep the queue safe. 26 * Alan Cox : Fixed double lock. 27 * Alan Cox : Fixed promisc NULL pointer trap 28 * ???????? : Support the full private ioctl range 29 * Alan Cox : Moved ioctl permission check into 30 * drivers 31 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI 32 * Alan Cox : 100 backlog just doesn't cut it when 33 * you start doing multicast video 8) 34 * Alan Cox : Rewrote net_bh and list manager. 35 * Alan Cox : Fix ETH_P_ALL echoback lengths. 36 * Alan Cox : Took out transmit every packet pass 37 * Saved a few bytes in the ioctl handler 38 * Alan Cox : Network driver sets packet type before 39 * calling netif_rx. Saves a function 40 * call a packet. 41 * Alan Cox : Hashed net_bh() 42 * Richard Kooijman: Timestamp fixes. 43 * Alan Cox : Wrong field in SIOCGIFDSTADDR 44 * Alan Cox : Device lock protection. 45 * Alan Cox : Fixed nasty side effect of device close 46 * changes. 47 * Rudi Cilibrasi : Pass the right thing to 48 * set_mac_address() 49 * Dave Miller : 32bit quantity for the device lock to 50 * make it work out on a Sparc. 51 * Bjorn Ekwall : Added KERNELD hack. 52 * Alan Cox : Cleaned up the backlog initialise. 53 * Craig Metz : SIOCGIFCONF fix if space for under 54 * 1 device. 55 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there 56 * is no device open function. 57 * Andi Kleen : Fix error reporting for SIOCGIFCONF 58 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF 59 * Cyrus Durgin : Cleaned for KMOD 60 * Adam Sulmicki : Bug Fix : Network Device Unload 61 * A network device unload needs to purge 62 * the backlog queue. 63 * Paul Rusty Russell : SIOCSIFNAME 64 * Pekka Riikonen : Netdev boot-time settings code 65 * Andrew Morton : Make unregister_netdevice wait 66 * indefinitely on dev->refcnt 67 * J Hadi Salim : - Backlog queue sampling 68 * - netif_rx() feedback 69 */ 70 71 #include <linux/uaccess.h> 72 #include <linux/bitmap.h> 73 #include <linux/capability.h> 74 #include <linux/cpu.h> 75 #include <linux/types.h> 76 #include <linux/kernel.h> 77 #include <linux/hash.h> 78 #include <linux/slab.h> 79 #include <linux/sched.h> 80 #include <linux/sched/mm.h> 81 #include <linux/mutex.h> 82 #include <linux/rwsem.h> 83 #include <linux/string.h> 84 #include <linux/mm.h> 85 #include <linux/socket.h> 86 #include <linux/sockios.h> 87 #include <linux/errno.h> 88 #include <linux/interrupt.h> 89 #include <linux/if_ether.h> 90 #include <linux/netdevice.h> 91 #include <linux/etherdevice.h> 92 #include <linux/ethtool.h> 93 #include <linux/skbuff.h> 94 #include <linux/kthread.h> 95 #include <linux/bpf.h> 96 #include <linux/bpf_trace.h> 97 #include <net/net_namespace.h> 98 #include <net/sock.h> 99 #include <net/busy_poll.h> 100 #include <linux/rtnetlink.h> 101 #include <linux/stat.h> 102 #include <net/dsa.h> 103 #include <net/dst.h> 104 #include <net/dst_metadata.h> 105 #include <net/gro.h> 106 #include <net/pkt_sched.h> 107 #include <net/pkt_cls.h> 108 #include <net/checksum.h> 109 #include <net/xfrm.h> 110 #include <net/tcx.h> 111 #include <linux/highmem.h> 112 #include <linux/init.h> 113 #include <linux/module.h> 114 #include <linux/netpoll.h> 115 #include <linux/rcupdate.h> 116 #include <linux/delay.h> 117 #include <net/iw_handler.h> 118 #include <asm/current.h> 119 #include <linux/audit.h> 120 #include <linux/dmaengine.h> 121 #include <linux/err.h> 122 #include <linux/ctype.h> 123 #include <linux/if_arp.h> 124 #include <linux/if_vlan.h> 125 #include <linux/ip.h> 126 #include <net/ip.h> 127 #include <net/mpls.h> 128 #include <linux/ipv6.h> 129 #include <linux/in.h> 130 #include <linux/jhash.h> 131 #include <linux/random.h> 132 #include <trace/events/napi.h> 133 #include <trace/events/net.h> 134 #include <trace/events/skb.h> 135 #include <trace/events/qdisc.h> 136 #include <trace/events/xdp.h> 137 #include <linux/inetdevice.h> 138 #include <linux/cpu_rmap.h> 139 #include <linux/static_key.h> 140 #include <linux/hashtable.h> 141 #include <linux/vmalloc.h> 142 #include <linux/if_macvlan.h> 143 #include <linux/errqueue.h> 144 #include <linux/hrtimer.h> 145 #include <linux/netfilter_netdev.h> 146 #include <linux/crash_dump.h> 147 #include <linux/sctp.h> 148 #include <net/udp_tunnel.h> 149 #include <linux/net_namespace.h> 150 #include <linux/indirect_call_wrapper.h> 151 #include <net/devlink.h> 152 #include <linux/pm_runtime.h> 153 #include <linux/prandom.h> 154 #include <linux/once_lite.h> 155 #include <net/netdev_rx_queue.h> 156 157 #include "dev.h" 158 #include "net-sysfs.h" 159 160 static DEFINE_SPINLOCK(ptype_lock); 161 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; 162 struct list_head ptype_all __read_mostly; /* Taps */ 163 164 static int netif_rx_internal(struct sk_buff *skb); 165 static int call_netdevice_notifiers_extack(unsigned long val, 166 struct net_device *dev, 167 struct netlink_ext_ack *extack); 168 static struct napi_struct *napi_by_id(unsigned int napi_id); 169 170 /* 171 * The @dev_base_head list is protected by @dev_base_lock and the rtnl 172 * semaphore. 173 * 174 * Pure readers hold dev_base_lock for reading, or rcu_read_lock() 175 * 176 * Writers must hold the rtnl semaphore while they loop through the 177 * dev_base_head list, and hold dev_base_lock for writing when they do the 178 * actual updates. This allows pure readers to access the list even 179 * while a writer is preparing to update it. 180 * 181 * To put it another way, dev_base_lock is held for writing only to 182 * protect against pure readers; the rtnl semaphore provides the 183 * protection against other writers. 184 * 185 * See, for example usages, register_netdevice() and 186 * unregister_netdevice(), which must be called with the rtnl 187 * semaphore held. 188 */ 189 DEFINE_RWLOCK(dev_base_lock); 190 EXPORT_SYMBOL(dev_base_lock); 191 192 static DEFINE_MUTEX(ifalias_mutex); 193 194 /* protects napi_hash addition/deletion and napi_gen_id */ 195 static DEFINE_SPINLOCK(napi_hash_lock); 196 197 static unsigned int napi_gen_id = NR_CPUS; 198 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8); 199 200 static DECLARE_RWSEM(devnet_rename_sem); 201 202 static inline void dev_base_seq_inc(struct net *net) 203 { 204 while (++net->dev_base_seq == 0) 205 ; 206 } 207 208 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) 209 { 210 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ)); 211 212 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; 213 } 214 215 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) 216 { 217 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; 218 } 219 220 static inline void rps_lock_irqsave(struct softnet_data *sd, 221 unsigned long *flags) 222 { 223 if (IS_ENABLED(CONFIG_RPS)) 224 spin_lock_irqsave(&sd->input_pkt_queue.lock, *flags); 225 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 226 local_irq_save(*flags); 227 } 228 229 static inline void rps_lock_irq_disable(struct softnet_data *sd) 230 { 231 if (IS_ENABLED(CONFIG_RPS)) 232 spin_lock_irq(&sd->input_pkt_queue.lock); 233 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 234 local_irq_disable(); 235 } 236 237 static inline void rps_unlock_irq_restore(struct softnet_data *sd, 238 unsigned long *flags) 239 { 240 if (IS_ENABLED(CONFIG_RPS)) 241 spin_unlock_irqrestore(&sd->input_pkt_queue.lock, *flags); 242 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 243 local_irq_restore(*flags); 244 } 245 246 static inline void rps_unlock_irq_enable(struct softnet_data *sd) 247 { 248 if (IS_ENABLED(CONFIG_RPS)) 249 spin_unlock_irq(&sd->input_pkt_queue.lock); 250 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 251 local_irq_enable(); 252 } 253 254 static struct netdev_name_node *netdev_name_node_alloc(struct net_device *dev, 255 const char *name) 256 { 257 struct netdev_name_node *name_node; 258 259 name_node = kmalloc(sizeof(*name_node), GFP_KERNEL); 260 if (!name_node) 261 return NULL; 262 INIT_HLIST_NODE(&name_node->hlist); 263 name_node->dev = dev; 264 name_node->name = name; 265 return name_node; 266 } 267 268 static struct netdev_name_node * 269 netdev_name_node_head_alloc(struct net_device *dev) 270 { 271 struct netdev_name_node *name_node; 272 273 name_node = netdev_name_node_alloc(dev, dev->name); 274 if (!name_node) 275 return NULL; 276 INIT_LIST_HEAD(&name_node->list); 277 return name_node; 278 } 279 280 static void netdev_name_node_free(struct netdev_name_node *name_node) 281 { 282 kfree(name_node); 283 } 284 285 static void netdev_name_node_add(struct net *net, 286 struct netdev_name_node *name_node) 287 { 288 hlist_add_head_rcu(&name_node->hlist, 289 dev_name_hash(net, name_node->name)); 290 } 291 292 static void netdev_name_node_del(struct netdev_name_node *name_node) 293 { 294 hlist_del_rcu(&name_node->hlist); 295 } 296 297 static struct netdev_name_node *netdev_name_node_lookup(struct net *net, 298 const char *name) 299 { 300 struct hlist_head *head = dev_name_hash(net, name); 301 struct netdev_name_node *name_node; 302 303 hlist_for_each_entry(name_node, head, hlist) 304 if (!strcmp(name_node->name, name)) 305 return name_node; 306 return NULL; 307 } 308 309 static struct netdev_name_node *netdev_name_node_lookup_rcu(struct net *net, 310 const char *name) 311 { 312 struct hlist_head *head = dev_name_hash(net, name); 313 struct netdev_name_node *name_node; 314 315 hlist_for_each_entry_rcu(name_node, head, hlist) 316 if (!strcmp(name_node->name, name)) 317 return name_node; 318 return NULL; 319 } 320 321 bool netdev_name_in_use(struct net *net, const char *name) 322 { 323 return netdev_name_node_lookup(net, name); 324 } 325 EXPORT_SYMBOL(netdev_name_in_use); 326 327 int netdev_name_node_alt_create(struct net_device *dev, const char *name) 328 { 329 struct netdev_name_node *name_node; 330 struct net *net = dev_net(dev); 331 332 name_node = netdev_name_node_lookup(net, name); 333 if (name_node) 334 return -EEXIST; 335 name_node = netdev_name_node_alloc(dev, name); 336 if (!name_node) 337 return -ENOMEM; 338 netdev_name_node_add(net, name_node); 339 /* The node that holds dev->name acts as a head of per-device list. */ 340 list_add_tail(&name_node->list, &dev->name_node->list); 341 342 return 0; 343 } 344 345 static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node) 346 { 347 list_del(&name_node->list); 348 kfree(name_node->name); 349 netdev_name_node_free(name_node); 350 } 351 352 int netdev_name_node_alt_destroy(struct net_device *dev, const char *name) 353 { 354 struct netdev_name_node *name_node; 355 struct net *net = dev_net(dev); 356 357 name_node = netdev_name_node_lookup(net, name); 358 if (!name_node) 359 return -ENOENT; 360 /* lookup might have found our primary name or a name belonging 361 * to another device. 362 */ 363 if (name_node == dev->name_node || name_node->dev != dev) 364 return -EINVAL; 365 366 netdev_name_node_del(name_node); 367 synchronize_rcu(); 368 __netdev_name_node_alt_destroy(name_node); 369 370 return 0; 371 } 372 373 static void netdev_name_node_alt_flush(struct net_device *dev) 374 { 375 struct netdev_name_node *name_node, *tmp; 376 377 list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list) 378 __netdev_name_node_alt_destroy(name_node); 379 } 380 381 /* Device list insertion */ 382 static void list_netdevice(struct net_device *dev) 383 { 384 struct netdev_name_node *name_node; 385 struct net *net = dev_net(dev); 386 387 ASSERT_RTNL(); 388 389 write_lock(&dev_base_lock); 390 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); 391 netdev_name_node_add(net, dev->name_node); 392 hlist_add_head_rcu(&dev->index_hlist, 393 dev_index_hash(net, dev->ifindex)); 394 write_unlock(&dev_base_lock); 395 396 netdev_for_each_altname(dev, name_node) 397 netdev_name_node_add(net, name_node); 398 399 /* We reserved the ifindex, this can't fail */ 400 WARN_ON(xa_store(&net->dev_by_index, dev->ifindex, dev, GFP_KERNEL)); 401 402 dev_base_seq_inc(net); 403 } 404 405 /* Device list removal 406 * caller must respect a RCU grace period before freeing/reusing dev 407 */ 408 static void unlist_netdevice(struct net_device *dev, bool lock) 409 { 410 struct netdev_name_node *name_node; 411 struct net *net = dev_net(dev); 412 413 ASSERT_RTNL(); 414 415 xa_erase(&net->dev_by_index, dev->ifindex); 416 417 netdev_for_each_altname(dev, name_node) 418 netdev_name_node_del(name_node); 419 420 /* Unlink dev from the device chain */ 421 if (lock) 422 write_lock(&dev_base_lock); 423 list_del_rcu(&dev->dev_list); 424 netdev_name_node_del(dev->name_node); 425 hlist_del_rcu(&dev->index_hlist); 426 if (lock) 427 write_unlock(&dev_base_lock); 428 429 dev_base_seq_inc(dev_net(dev)); 430 } 431 432 /* 433 * Our notifier list 434 */ 435 436 static RAW_NOTIFIER_HEAD(netdev_chain); 437 438 /* 439 * Device drivers call our routines to queue packets here. We empty the 440 * queue in the local softnet handler. 441 */ 442 443 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); 444 EXPORT_PER_CPU_SYMBOL(softnet_data); 445 446 #ifdef CONFIG_LOCKDEP 447 /* 448 * register_netdevice() inits txq->_xmit_lock and sets lockdep class 449 * according to dev->type 450 */ 451 static const unsigned short netdev_lock_type[] = { 452 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, 453 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, 454 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, 455 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, 456 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, 457 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, 458 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, 459 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, 460 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, 461 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, 462 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, 463 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, 464 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, 465 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, 466 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; 467 468 static const char *const netdev_lock_name[] = { 469 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", 470 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", 471 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", 472 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", 473 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", 474 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", 475 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", 476 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", 477 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", 478 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", 479 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", 480 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", 481 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", 482 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", 483 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; 484 485 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; 486 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; 487 488 static inline unsigned short netdev_lock_pos(unsigned short dev_type) 489 { 490 int i; 491 492 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) 493 if (netdev_lock_type[i] == dev_type) 494 return i; 495 /* the last key is used by default */ 496 return ARRAY_SIZE(netdev_lock_type) - 1; 497 } 498 499 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 500 unsigned short dev_type) 501 { 502 int i; 503 504 i = netdev_lock_pos(dev_type); 505 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], 506 netdev_lock_name[i]); 507 } 508 509 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 510 { 511 int i; 512 513 i = netdev_lock_pos(dev->type); 514 lockdep_set_class_and_name(&dev->addr_list_lock, 515 &netdev_addr_lock_key[i], 516 netdev_lock_name[i]); 517 } 518 #else 519 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 520 unsigned short dev_type) 521 { 522 } 523 524 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 525 { 526 } 527 #endif 528 529 /******************************************************************************* 530 * 531 * Protocol management and registration routines 532 * 533 *******************************************************************************/ 534 535 536 /* 537 * Add a protocol ID to the list. Now that the input handler is 538 * smarter we can dispense with all the messy stuff that used to be 539 * here. 540 * 541 * BEWARE!!! Protocol handlers, mangling input packets, 542 * MUST BE last in hash buckets and checking protocol handlers 543 * MUST start from promiscuous ptype_all chain in net_bh. 544 * It is true now, do not change it. 545 * Explanation follows: if protocol handler, mangling packet, will 546 * be the first on list, it is not able to sense, that packet 547 * is cloned and should be copied-on-write, so that it will 548 * change it and subsequent readers will get broken packet. 549 * --ANK (980803) 550 */ 551 552 static inline struct list_head *ptype_head(const struct packet_type *pt) 553 { 554 if (pt->type == htons(ETH_P_ALL)) 555 return pt->dev ? &pt->dev->ptype_all : &ptype_all; 556 else 557 return pt->dev ? &pt->dev->ptype_specific : 558 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; 559 } 560 561 /** 562 * dev_add_pack - add packet handler 563 * @pt: packet type declaration 564 * 565 * Add a protocol handler to the networking stack. The passed &packet_type 566 * is linked into kernel lists and may not be freed until it has been 567 * removed from the kernel lists. 568 * 569 * This call does not sleep therefore it can not 570 * guarantee all CPU's that are in middle of receiving packets 571 * will see the new packet type (until the next received packet). 572 */ 573 574 void dev_add_pack(struct packet_type *pt) 575 { 576 struct list_head *head = ptype_head(pt); 577 578 spin_lock(&ptype_lock); 579 list_add_rcu(&pt->list, head); 580 spin_unlock(&ptype_lock); 581 } 582 EXPORT_SYMBOL(dev_add_pack); 583 584 /** 585 * __dev_remove_pack - remove packet handler 586 * @pt: packet type declaration 587 * 588 * Remove a protocol handler that was previously added to the kernel 589 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 590 * from the kernel lists and can be freed or reused once this function 591 * returns. 592 * 593 * The packet type might still be in use by receivers 594 * and must not be freed until after all the CPU's have gone 595 * through a quiescent state. 596 */ 597 void __dev_remove_pack(struct packet_type *pt) 598 { 599 struct list_head *head = ptype_head(pt); 600 struct packet_type *pt1; 601 602 spin_lock(&ptype_lock); 603 604 list_for_each_entry(pt1, head, list) { 605 if (pt == pt1) { 606 list_del_rcu(&pt->list); 607 goto out; 608 } 609 } 610 611 pr_warn("dev_remove_pack: %p not found\n", pt); 612 out: 613 spin_unlock(&ptype_lock); 614 } 615 EXPORT_SYMBOL(__dev_remove_pack); 616 617 /** 618 * dev_remove_pack - remove packet handler 619 * @pt: packet type declaration 620 * 621 * Remove a protocol handler that was previously added to the kernel 622 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 623 * from the kernel lists and can be freed or reused once this function 624 * returns. 625 * 626 * This call sleeps to guarantee that no CPU is looking at the packet 627 * type after return. 628 */ 629 void dev_remove_pack(struct packet_type *pt) 630 { 631 __dev_remove_pack(pt); 632 633 synchronize_net(); 634 } 635 EXPORT_SYMBOL(dev_remove_pack); 636 637 638 /******************************************************************************* 639 * 640 * Device Interface Subroutines 641 * 642 *******************************************************************************/ 643 644 /** 645 * dev_get_iflink - get 'iflink' value of a interface 646 * @dev: targeted interface 647 * 648 * Indicates the ifindex the interface is linked to. 649 * Physical interfaces have the same 'ifindex' and 'iflink' values. 650 */ 651 652 int dev_get_iflink(const struct net_device *dev) 653 { 654 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink) 655 return dev->netdev_ops->ndo_get_iflink(dev); 656 657 return dev->ifindex; 658 } 659 EXPORT_SYMBOL(dev_get_iflink); 660 661 /** 662 * dev_fill_metadata_dst - Retrieve tunnel egress information. 663 * @dev: targeted interface 664 * @skb: The packet. 665 * 666 * For better visibility of tunnel traffic OVS needs to retrieve 667 * egress tunnel information for a packet. Following API allows 668 * user to get this info. 669 */ 670 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) 671 { 672 struct ip_tunnel_info *info; 673 674 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst) 675 return -EINVAL; 676 677 info = skb_tunnel_info_unclone(skb); 678 if (!info) 679 return -ENOMEM; 680 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX))) 681 return -EINVAL; 682 683 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb); 684 } 685 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst); 686 687 static struct net_device_path *dev_fwd_path(struct net_device_path_stack *stack) 688 { 689 int k = stack->num_paths++; 690 691 if (WARN_ON_ONCE(k >= NET_DEVICE_PATH_STACK_MAX)) 692 return NULL; 693 694 return &stack->path[k]; 695 } 696 697 int dev_fill_forward_path(const struct net_device *dev, const u8 *daddr, 698 struct net_device_path_stack *stack) 699 { 700 const struct net_device *last_dev; 701 struct net_device_path_ctx ctx = { 702 .dev = dev, 703 }; 704 struct net_device_path *path; 705 int ret = 0; 706 707 memcpy(ctx.daddr, daddr, sizeof(ctx.daddr)); 708 stack->num_paths = 0; 709 while (ctx.dev && ctx.dev->netdev_ops->ndo_fill_forward_path) { 710 last_dev = ctx.dev; 711 path = dev_fwd_path(stack); 712 if (!path) 713 return -1; 714 715 memset(path, 0, sizeof(struct net_device_path)); 716 ret = ctx.dev->netdev_ops->ndo_fill_forward_path(&ctx, path); 717 if (ret < 0) 718 return -1; 719 720 if (WARN_ON_ONCE(last_dev == ctx.dev)) 721 return -1; 722 } 723 724 if (!ctx.dev) 725 return ret; 726 727 path = dev_fwd_path(stack); 728 if (!path) 729 return -1; 730 path->type = DEV_PATH_ETHERNET; 731 path->dev = ctx.dev; 732 733 return ret; 734 } 735 EXPORT_SYMBOL_GPL(dev_fill_forward_path); 736 737 /** 738 * __dev_get_by_name - find a device by its name 739 * @net: the applicable net namespace 740 * @name: name to find 741 * 742 * Find an interface by name. Must be called under RTNL semaphore 743 * or @dev_base_lock. If the name is found a pointer to the device 744 * is returned. If the name is not found then %NULL is returned. The 745 * reference counters are not incremented so the caller must be 746 * careful with locks. 747 */ 748 749 struct net_device *__dev_get_by_name(struct net *net, const char *name) 750 { 751 struct netdev_name_node *node_name; 752 753 node_name = netdev_name_node_lookup(net, name); 754 return node_name ? node_name->dev : NULL; 755 } 756 EXPORT_SYMBOL(__dev_get_by_name); 757 758 /** 759 * dev_get_by_name_rcu - find a device by its name 760 * @net: the applicable net namespace 761 * @name: name to find 762 * 763 * Find an interface by name. 764 * If the name is found a pointer to the device is returned. 765 * If the name is not found then %NULL is returned. 766 * The reference counters are not incremented so the caller must be 767 * careful with locks. The caller must hold RCU lock. 768 */ 769 770 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) 771 { 772 struct netdev_name_node *node_name; 773 774 node_name = netdev_name_node_lookup_rcu(net, name); 775 return node_name ? node_name->dev : NULL; 776 } 777 EXPORT_SYMBOL(dev_get_by_name_rcu); 778 779 /* Deprecated for new users, call netdev_get_by_name() instead */ 780 struct net_device *dev_get_by_name(struct net *net, const char *name) 781 { 782 struct net_device *dev; 783 784 rcu_read_lock(); 785 dev = dev_get_by_name_rcu(net, name); 786 dev_hold(dev); 787 rcu_read_unlock(); 788 return dev; 789 } 790 EXPORT_SYMBOL(dev_get_by_name); 791 792 /** 793 * netdev_get_by_name() - find a device by its name 794 * @net: the applicable net namespace 795 * @name: name to find 796 * @tracker: tracking object for the acquired reference 797 * @gfp: allocation flags for the tracker 798 * 799 * Find an interface by name. This can be called from any 800 * context and does its own locking. The returned handle has 801 * the usage count incremented and the caller must use netdev_put() to 802 * release it when it is no longer needed. %NULL is returned if no 803 * matching device is found. 804 */ 805 struct net_device *netdev_get_by_name(struct net *net, const char *name, 806 netdevice_tracker *tracker, gfp_t gfp) 807 { 808 struct net_device *dev; 809 810 dev = dev_get_by_name(net, name); 811 if (dev) 812 netdev_tracker_alloc(dev, tracker, gfp); 813 return dev; 814 } 815 EXPORT_SYMBOL(netdev_get_by_name); 816 817 /** 818 * __dev_get_by_index - find a device by its ifindex 819 * @net: the applicable net namespace 820 * @ifindex: index of device 821 * 822 * Search for an interface by index. Returns %NULL if the device 823 * is not found or a pointer to the device. The device has not 824 * had its reference counter increased so the caller must be careful 825 * about locking. The caller must hold either the RTNL semaphore 826 * or @dev_base_lock. 827 */ 828 829 struct net_device *__dev_get_by_index(struct net *net, int ifindex) 830 { 831 struct net_device *dev; 832 struct hlist_head *head = dev_index_hash(net, ifindex); 833 834 hlist_for_each_entry(dev, head, index_hlist) 835 if (dev->ifindex == ifindex) 836 return dev; 837 838 return NULL; 839 } 840 EXPORT_SYMBOL(__dev_get_by_index); 841 842 /** 843 * dev_get_by_index_rcu - find a device by its ifindex 844 * @net: the applicable net namespace 845 * @ifindex: index of device 846 * 847 * Search for an interface by index. Returns %NULL if the device 848 * is not found or a pointer to the device. The device has not 849 * had its reference counter increased so the caller must be careful 850 * about locking. The caller must hold RCU lock. 851 */ 852 853 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) 854 { 855 struct net_device *dev; 856 struct hlist_head *head = dev_index_hash(net, ifindex); 857 858 hlist_for_each_entry_rcu(dev, head, index_hlist) 859 if (dev->ifindex == ifindex) 860 return dev; 861 862 return NULL; 863 } 864 EXPORT_SYMBOL(dev_get_by_index_rcu); 865 866 /* Deprecated for new users, call netdev_get_by_index() instead */ 867 struct net_device *dev_get_by_index(struct net *net, int ifindex) 868 { 869 struct net_device *dev; 870 871 rcu_read_lock(); 872 dev = dev_get_by_index_rcu(net, ifindex); 873 dev_hold(dev); 874 rcu_read_unlock(); 875 return dev; 876 } 877 EXPORT_SYMBOL(dev_get_by_index); 878 879 /** 880 * netdev_get_by_index() - find a device by its ifindex 881 * @net: the applicable net namespace 882 * @ifindex: index of device 883 * @tracker: tracking object for the acquired reference 884 * @gfp: allocation flags for the tracker 885 * 886 * Search for an interface by index. Returns NULL if the device 887 * is not found or a pointer to the device. The device returned has 888 * had a reference added and the pointer is safe until the user calls 889 * netdev_put() to indicate they have finished with it. 890 */ 891 struct net_device *netdev_get_by_index(struct net *net, int ifindex, 892 netdevice_tracker *tracker, gfp_t gfp) 893 { 894 struct net_device *dev; 895 896 dev = dev_get_by_index(net, ifindex); 897 if (dev) 898 netdev_tracker_alloc(dev, tracker, gfp); 899 return dev; 900 } 901 EXPORT_SYMBOL(netdev_get_by_index); 902 903 /** 904 * dev_get_by_napi_id - find a device by napi_id 905 * @napi_id: ID of the NAPI struct 906 * 907 * Search for an interface by NAPI ID. Returns %NULL if the device 908 * is not found or a pointer to the device. The device has not had 909 * its reference counter increased so the caller must be careful 910 * about locking. The caller must hold RCU lock. 911 */ 912 913 struct net_device *dev_get_by_napi_id(unsigned int napi_id) 914 { 915 struct napi_struct *napi; 916 917 WARN_ON_ONCE(!rcu_read_lock_held()); 918 919 if (napi_id < MIN_NAPI_ID) 920 return NULL; 921 922 napi = napi_by_id(napi_id); 923 924 return napi ? napi->dev : NULL; 925 } 926 EXPORT_SYMBOL(dev_get_by_napi_id); 927 928 /** 929 * netdev_get_name - get a netdevice name, knowing its ifindex. 930 * @net: network namespace 931 * @name: a pointer to the buffer where the name will be stored. 932 * @ifindex: the ifindex of the interface to get the name from. 933 */ 934 int netdev_get_name(struct net *net, char *name, int ifindex) 935 { 936 struct net_device *dev; 937 int ret; 938 939 down_read(&devnet_rename_sem); 940 rcu_read_lock(); 941 942 dev = dev_get_by_index_rcu(net, ifindex); 943 if (!dev) { 944 ret = -ENODEV; 945 goto out; 946 } 947 948 strcpy(name, dev->name); 949 950 ret = 0; 951 out: 952 rcu_read_unlock(); 953 up_read(&devnet_rename_sem); 954 return ret; 955 } 956 957 /** 958 * dev_getbyhwaddr_rcu - find a device by its hardware address 959 * @net: the applicable net namespace 960 * @type: media type of device 961 * @ha: hardware address 962 * 963 * Search for an interface by MAC address. Returns NULL if the device 964 * is not found or a pointer to the device. 965 * The caller must hold RCU or RTNL. 966 * The returned device has not had its ref count increased 967 * and the caller must therefore be careful about locking 968 * 969 */ 970 971 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, 972 const char *ha) 973 { 974 struct net_device *dev; 975 976 for_each_netdev_rcu(net, dev) 977 if (dev->type == type && 978 !memcmp(dev->dev_addr, ha, dev->addr_len)) 979 return dev; 980 981 return NULL; 982 } 983 EXPORT_SYMBOL(dev_getbyhwaddr_rcu); 984 985 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) 986 { 987 struct net_device *dev, *ret = NULL; 988 989 rcu_read_lock(); 990 for_each_netdev_rcu(net, dev) 991 if (dev->type == type) { 992 dev_hold(dev); 993 ret = dev; 994 break; 995 } 996 rcu_read_unlock(); 997 return ret; 998 } 999 EXPORT_SYMBOL(dev_getfirstbyhwtype); 1000 1001 /** 1002 * __dev_get_by_flags - find any device with given flags 1003 * @net: the applicable net namespace 1004 * @if_flags: IFF_* values 1005 * @mask: bitmask of bits in if_flags to check 1006 * 1007 * Search for any interface with the given flags. Returns NULL if a device 1008 * is not found or a pointer to the device. Must be called inside 1009 * rtnl_lock(), and result refcount is unchanged. 1010 */ 1011 1012 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags, 1013 unsigned short mask) 1014 { 1015 struct net_device *dev, *ret; 1016 1017 ASSERT_RTNL(); 1018 1019 ret = NULL; 1020 for_each_netdev(net, dev) { 1021 if (((dev->flags ^ if_flags) & mask) == 0) { 1022 ret = dev; 1023 break; 1024 } 1025 } 1026 return ret; 1027 } 1028 EXPORT_SYMBOL(__dev_get_by_flags); 1029 1030 /** 1031 * dev_valid_name - check if name is okay for network device 1032 * @name: name string 1033 * 1034 * Network device names need to be valid file names to 1035 * allow sysfs to work. We also disallow any kind of 1036 * whitespace. 1037 */ 1038 bool dev_valid_name(const char *name) 1039 { 1040 if (*name == '\0') 1041 return false; 1042 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ) 1043 return false; 1044 if (!strcmp(name, ".") || !strcmp(name, "..")) 1045 return false; 1046 1047 while (*name) { 1048 if (*name == '/' || *name == ':' || isspace(*name)) 1049 return false; 1050 name++; 1051 } 1052 return true; 1053 } 1054 EXPORT_SYMBOL(dev_valid_name); 1055 1056 /** 1057 * __dev_alloc_name - allocate a name for a device 1058 * @net: network namespace to allocate the device name in 1059 * @name: name format string 1060 * @buf: scratch buffer and result name string 1061 * 1062 * Passed a format string - eg "lt%d" it will try and find a suitable 1063 * id. It scans list of devices to build up a free map, then chooses 1064 * the first empty slot. The caller must hold the dev_base or rtnl lock 1065 * while allocating the name and adding the device in order to avoid 1066 * duplicates. 1067 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1068 * Returns the number of the unit assigned or a negative errno code. 1069 */ 1070 1071 static int __dev_alloc_name(struct net *net, const char *name, char *buf) 1072 { 1073 int i = 0; 1074 const char *p; 1075 const int max_netdevices = 8*PAGE_SIZE; 1076 unsigned long *inuse; 1077 struct net_device *d; 1078 1079 if (!dev_valid_name(name)) 1080 return -EINVAL; 1081 1082 p = strchr(name, '%'); 1083 if (p) { 1084 /* 1085 * Verify the string as this thing may have come from 1086 * the user. There must be either one "%d" and no other "%" 1087 * characters. 1088 */ 1089 if (p[1] != 'd' || strchr(p + 2, '%')) 1090 return -EINVAL; 1091 1092 /* Use one page as a bit array of possible slots */ 1093 inuse = bitmap_zalloc(max_netdevices, GFP_ATOMIC); 1094 if (!inuse) 1095 return -ENOMEM; 1096 1097 for_each_netdev(net, d) { 1098 struct netdev_name_node *name_node; 1099 1100 netdev_for_each_altname(d, name_node) { 1101 if (!sscanf(name_node->name, name, &i)) 1102 continue; 1103 if (i < 0 || i >= max_netdevices) 1104 continue; 1105 1106 /* avoid cases where sscanf is not exact inverse of printf */ 1107 snprintf(buf, IFNAMSIZ, name, i); 1108 if (!strncmp(buf, name_node->name, IFNAMSIZ)) 1109 __set_bit(i, inuse); 1110 } 1111 if (!sscanf(d->name, name, &i)) 1112 continue; 1113 if (i < 0 || i >= max_netdevices) 1114 continue; 1115 1116 /* avoid cases where sscanf is not exact inverse of printf */ 1117 snprintf(buf, IFNAMSIZ, name, i); 1118 if (!strncmp(buf, d->name, IFNAMSIZ)) 1119 __set_bit(i, inuse); 1120 } 1121 1122 i = find_first_zero_bit(inuse, max_netdevices); 1123 bitmap_free(inuse); 1124 } 1125 1126 snprintf(buf, IFNAMSIZ, name, i); 1127 if (!netdev_name_in_use(net, buf)) 1128 return i; 1129 1130 /* It is possible to run out of possible slots 1131 * when the name is long and there isn't enough space left 1132 * for the digits, or if all bits are used. 1133 */ 1134 return -ENFILE; 1135 } 1136 1137 static int dev_prep_valid_name(struct net *net, struct net_device *dev, 1138 const char *want_name, char *out_name) 1139 { 1140 int ret; 1141 1142 if (!dev_valid_name(want_name)) 1143 return -EINVAL; 1144 1145 if (strchr(want_name, '%')) { 1146 ret = __dev_alloc_name(net, want_name, out_name); 1147 return ret < 0 ? ret : 0; 1148 } else if (netdev_name_in_use(net, want_name)) { 1149 return -EEXIST; 1150 } else if (out_name != want_name) { 1151 strscpy(out_name, want_name, IFNAMSIZ); 1152 } 1153 1154 return 0; 1155 } 1156 1157 static int dev_alloc_name_ns(struct net *net, 1158 struct net_device *dev, 1159 const char *name) 1160 { 1161 char buf[IFNAMSIZ]; 1162 int ret; 1163 1164 BUG_ON(!net); 1165 ret = __dev_alloc_name(net, name, buf); 1166 if (ret >= 0) 1167 strscpy(dev->name, buf, IFNAMSIZ); 1168 return ret; 1169 } 1170 1171 /** 1172 * dev_alloc_name - allocate a name for a device 1173 * @dev: device 1174 * @name: name format string 1175 * 1176 * Passed a format string - eg "lt%d" it will try and find a suitable 1177 * id. It scans list of devices to build up a free map, then chooses 1178 * the first empty slot. The caller must hold the dev_base or rtnl lock 1179 * while allocating the name and adding the device in order to avoid 1180 * duplicates. 1181 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1182 * Returns the number of the unit assigned or a negative errno code. 1183 */ 1184 1185 int dev_alloc_name(struct net_device *dev, const char *name) 1186 { 1187 return dev_alloc_name_ns(dev_net(dev), dev, name); 1188 } 1189 EXPORT_SYMBOL(dev_alloc_name); 1190 1191 static int dev_get_valid_name(struct net *net, struct net_device *dev, 1192 const char *name) 1193 { 1194 char buf[IFNAMSIZ]; 1195 int ret; 1196 1197 ret = dev_prep_valid_name(net, dev, name, buf); 1198 if (ret >= 0) 1199 strscpy(dev->name, buf, IFNAMSIZ); 1200 return ret; 1201 } 1202 1203 /** 1204 * dev_change_name - change name of a device 1205 * @dev: device 1206 * @newname: name (or format string) must be at least IFNAMSIZ 1207 * 1208 * Change name of a device, can pass format strings "eth%d". 1209 * for wildcarding. 1210 */ 1211 int dev_change_name(struct net_device *dev, const char *newname) 1212 { 1213 unsigned char old_assign_type; 1214 char oldname[IFNAMSIZ]; 1215 int err = 0; 1216 int ret; 1217 struct net *net; 1218 1219 ASSERT_RTNL(); 1220 BUG_ON(!dev_net(dev)); 1221 1222 net = dev_net(dev); 1223 1224 down_write(&devnet_rename_sem); 1225 1226 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { 1227 up_write(&devnet_rename_sem); 1228 return 0; 1229 } 1230 1231 memcpy(oldname, dev->name, IFNAMSIZ); 1232 1233 err = dev_get_valid_name(net, dev, newname); 1234 if (err < 0) { 1235 up_write(&devnet_rename_sem); 1236 return err; 1237 } 1238 1239 if (oldname[0] && !strchr(oldname, '%')) 1240 netdev_info(dev, "renamed from %s%s\n", oldname, 1241 dev->flags & IFF_UP ? " (while UP)" : ""); 1242 1243 old_assign_type = dev->name_assign_type; 1244 dev->name_assign_type = NET_NAME_RENAMED; 1245 1246 rollback: 1247 ret = device_rename(&dev->dev, dev->name); 1248 if (ret) { 1249 memcpy(dev->name, oldname, IFNAMSIZ); 1250 dev->name_assign_type = old_assign_type; 1251 up_write(&devnet_rename_sem); 1252 return ret; 1253 } 1254 1255 up_write(&devnet_rename_sem); 1256 1257 netdev_adjacent_rename_links(dev, oldname); 1258 1259 write_lock(&dev_base_lock); 1260 netdev_name_node_del(dev->name_node); 1261 write_unlock(&dev_base_lock); 1262 1263 synchronize_rcu(); 1264 1265 write_lock(&dev_base_lock); 1266 netdev_name_node_add(net, dev->name_node); 1267 write_unlock(&dev_base_lock); 1268 1269 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); 1270 ret = notifier_to_errno(ret); 1271 1272 if (ret) { 1273 /* err >= 0 after dev_alloc_name() or stores the first errno */ 1274 if (err >= 0) { 1275 err = ret; 1276 down_write(&devnet_rename_sem); 1277 memcpy(dev->name, oldname, IFNAMSIZ); 1278 memcpy(oldname, newname, IFNAMSIZ); 1279 dev->name_assign_type = old_assign_type; 1280 old_assign_type = NET_NAME_RENAMED; 1281 goto rollback; 1282 } else { 1283 netdev_err(dev, "name change rollback failed: %d\n", 1284 ret); 1285 } 1286 } 1287 1288 return err; 1289 } 1290 1291 /** 1292 * dev_set_alias - change ifalias of a device 1293 * @dev: device 1294 * @alias: name up to IFALIASZ 1295 * @len: limit of bytes to copy from info 1296 * 1297 * Set ifalias for a device, 1298 */ 1299 int dev_set_alias(struct net_device *dev, const char *alias, size_t len) 1300 { 1301 struct dev_ifalias *new_alias = NULL; 1302 1303 if (len >= IFALIASZ) 1304 return -EINVAL; 1305 1306 if (len) { 1307 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL); 1308 if (!new_alias) 1309 return -ENOMEM; 1310 1311 memcpy(new_alias->ifalias, alias, len); 1312 new_alias->ifalias[len] = 0; 1313 } 1314 1315 mutex_lock(&ifalias_mutex); 1316 new_alias = rcu_replace_pointer(dev->ifalias, new_alias, 1317 mutex_is_locked(&ifalias_mutex)); 1318 mutex_unlock(&ifalias_mutex); 1319 1320 if (new_alias) 1321 kfree_rcu(new_alias, rcuhead); 1322 1323 return len; 1324 } 1325 EXPORT_SYMBOL(dev_set_alias); 1326 1327 /** 1328 * dev_get_alias - get ifalias of a device 1329 * @dev: device 1330 * @name: buffer to store name of ifalias 1331 * @len: size of buffer 1332 * 1333 * get ifalias for a device. Caller must make sure dev cannot go 1334 * away, e.g. rcu read lock or own a reference count to device. 1335 */ 1336 int dev_get_alias(const struct net_device *dev, char *name, size_t len) 1337 { 1338 const struct dev_ifalias *alias; 1339 int ret = 0; 1340 1341 rcu_read_lock(); 1342 alias = rcu_dereference(dev->ifalias); 1343 if (alias) 1344 ret = snprintf(name, len, "%s", alias->ifalias); 1345 rcu_read_unlock(); 1346 1347 return ret; 1348 } 1349 1350 /** 1351 * netdev_features_change - device changes features 1352 * @dev: device to cause notification 1353 * 1354 * Called to indicate a device has changed features. 1355 */ 1356 void netdev_features_change(struct net_device *dev) 1357 { 1358 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); 1359 } 1360 EXPORT_SYMBOL(netdev_features_change); 1361 1362 /** 1363 * netdev_state_change - device changes state 1364 * @dev: device to cause notification 1365 * 1366 * Called to indicate a device has changed state. This function calls 1367 * the notifier chains for netdev_chain and sends a NEWLINK message 1368 * to the routing socket. 1369 */ 1370 void netdev_state_change(struct net_device *dev) 1371 { 1372 if (dev->flags & IFF_UP) { 1373 struct netdev_notifier_change_info change_info = { 1374 .info.dev = dev, 1375 }; 1376 1377 call_netdevice_notifiers_info(NETDEV_CHANGE, 1378 &change_info.info); 1379 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL, 0, NULL); 1380 } 1381 } 1382 EXPORT_SYMBOL(netdev_state_change); 1383 1384 /** 1385 * __netdev_notify_peers - notify network peers about existence of @dev, 1386 * to be called when rtnl lock is already held. 1387 * @dev: network device 1388 * 1389 * Generate traffic such that interested network peers are aware of 1390 * @dev, such as by generating a gratuitous ARP. This may be used when 1391 * a device wants to inform the rest of the network about some sort of 1392 * reconfiguration such as a failover event or virtual machine 1393 * migration. 1394 */ 1395 void __netdev_notify_peers(struct net_device *dev) 1396 { 1397 ASSERT_RTNL(); 1398 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); 1399 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev); 1400 } 1401 EXPORT_SYMBOL(__netdev_notify_peers); 1402 1403 /** 1404 * netdev_notify_peers - notify network peers about existence of @dev 1405 * @dev: network device 1406 * 1407 * Generate traffic such that interested network peers are aware of 1408 * @dev, such as by generating a gratuitous ARP. This may be used when 1409 * a device wants to inform the rest of the network about some sort of 1410 * reconfiguration such as a failover event or virtual machine 1411 * migration. 1412 */ 1413 void netdev_notify_peers(struct net_device *dev) 1414 { 1415 rtnl_lock(); 1416 __netdev_notify_peers(dev); 1417 rtnl_unlock(); 1418 } 1419 EXPORT_SYMBOL(netdev_notify_peers); 1420 1421 static int napi_threaded_poll(void *data); 1422 1423 static int napi_kthread_create(struct napi_struct *n) 1424 { 1425 int err = 0; 1426 1427 /* Create and wake up the kthread once to put it in 1428 * TASK_INTERRUPTIBLE mode to avoid the blocked task 1429 * warning and work with loadavg. 1430 */ 1431 n->thread = kthread_run(napi_threaded_poll, n, "napi/%s-%d", 1432 n->dev->name, n->napi_id); 1433 if (IS_ERR(n->thread)) { 1434 err = PTR_ERR(n->thread); 1435 pr_err("kthread_run failed with err %d\n", err); 1436 n->thread = NULL; 1437 } 1438 1439 return err; 1440 } 1441 1442 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1443 { 1444 const struct net_device_ops *ops = dev->netdev_ops; 1445 int ret; 1446 1447 ASSERT_RTNL(); 1448 dev_addr_check(dev); 1449 1450 if (!netif_device_present(dev)) { 1451 /* may be detached because parent is runtime-suspended */ 1452 if (dev->dev.parent) 1453 pm_runtime_resume(dev->dev.parent); 1454 if (!netif_device_present(dev)) 1455 return -ENODEV; 1456 } 1457 1458 /* Block netpoll from trying to do any rx path servicing. 1459 * If we don't do this there is a chance ndo_poll_controller 1460 * or ndo_poll may be running while we open the device 1461 */ 1462 netpoll_poll_disable(dev); 1463 1464 ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack); 1465 ret = notifier_to_errno(ret); 1466 if (ret) 1467 return ret; 1468 1469 set_bit(__LINK_STATE_START, &dev->state); 1470 1471 if (ops->ndo_validate_addr) 1472 ret = ops->ndo_validate_addr(dev); 1473 1474 if (!ret && ops->ndo_open) 1475 ret = ops->ndo_open(dev); 1476 1477 netpoll_poll_enable(dev); 1478 1479 if (ret) 1480 clear_bit(__LINK_STATE_START, &dev->state); 1481 else { 1482 dev->flags |= IFF_UP; 1483 dev_set_rx_mode(dev); 1484 dev_activate(dev); 1485 add_device_randomness(dev->dev_addr, dev->addr_len); 1486 } 1487 1488 return ret; 1489 } 1490 1491 /** 1492 * dev_open - prepare an interface for use. 1493 * @dev: device to open 1494 * @extack: netlink extended ack 1495 * 1496 * Takes a device from down to up state. The device's private open 1497 * function is invoked and then the multicast lists are loaded. Finally 1498 * the device is moved into the up state and a %NETDEV_UP message is 1499 * sent to the netdev notifier chain. 1500 * 1501 * Calling this function on an active interface is a nop. On a failure 1502 * a negative errno code is returned. 1503 */ 1504 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1505 { 1506 int ret; 1507 1508 if (dev->flags & IFF_UP) 1509 return 0; 1510 1511 ret = __dev_open(dev, extack); 1512 if (ret < 0) 1513 return ret; 1514 1515 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL); 1516 call_netdevice_notifiers(NETDEV_UP, dev); 1517 1518 return ret; 1519 } 1520 EXPORT_SYMBOL(dev_open); 1521 1522 static void __dev_close_many(struct list_head *head) 1523 { 1524 struct net_device *dev; 1525 1526 ASSERT_RTNL(); 1527 might_sleep(); 1528 1529 list_for_each_entry(dev, head, close_list) { 1530 /* Temporarily disable netpoll until the interface is down */ 1531 netpoll_poll_disable(dev); 1532 1533 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); 1534 1535 clear_bit(__LINK_STATE_START, &dev->state); 1536 1537 /* Synchronize to scheduled poll. We cannot touch poll list, it 1538 * can be even on different cpu. So just clear netif_running(). 1539 * 1540 * dev->stop() will invoke napi_disable() on all of it's 1541 * napi_struct instances on this device. 1542 */ 1543 smp_mb__after_atomic(); /* Commit netif_running(). */ 1544 } 1545 1546 dev_deactivate_many(head); 1547 1548 list_for_each_entry(dev, head, close_list) { 1549 const struct net_device_ops *ops = dev->netdev_ops; 1550 1551 /* 1552 * Call the device specific close. This cannot fail. 1553 * Only if device is UP 1554 * 1555 * We allow it to be called even after a DETACH hot-plug 1556 * event. 1557 */ 1558 if (ops->ndo_stop) 1559 ops->ndo_stop(dev); 1560 1561 dev->flags &= ~IFF_UP; 1562 netpoll_poll_enable(dev); 1563 } 1564 } 1565 1566 static void __dev_close(struct net_device *dev) 1567 { 1568 LIST_HEAD(single); 1569 1570 list_add(&dev->close_list, &single); 1571 __dev_close_many(&single); 1572 list_del(&single); 1573 } 1574 1575 void dev_close_many(struct list_head *head, bool unlink) 1576 { 1577 struct net_device *dev, *tmp; 1578 1579 /* Remove the devices that don't need to be closed */ 1580 list_for_each_entry_safe(dev, tmp, head, close_list) 1581 if (!(dev->flags & IFF_UP)) 1582 list_del_init(&dev->close_list); 1583 1584 __dev_close_many(head); 1585 1586 list_for_each_entry_safe(dev, tmp, head, close_list) { 1587 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL); 1588 call_netdevice_notifiers(NETDEV_DOWN, dev); 1589 if (unlink) 1590 list_del_init(&dev->close_list); 1591 } 1592 } 1593 EXPORT_SYMBOL(dev_close_many); 1594 1595 /** 1596 * dev_close - shutdown an interface. 1597 * @dev: device to shutdown 1598 * 1599 * This function moves an active device into down state. A 1600 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device 1601 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier 1602 * chain. 1603 */ 1604 void dev_close(struct net_device *dev) 1605 { 1606 if (dev->flags & IFF_UP) { 1607 LIST_HEAD(single); 1608 1609 list_add(&dev->close_list, &single); 1610 dev_close_many(&single, true); 1611 list_del(&single); 1612 } 1613 } 1614 EXPORT_SYMBOL(dev_close); 1615 1616 1617 /** 1618 * dev_disable_lro - disable Large Receive Offload on a device 1619 * @dev: device 1620 * 1621 * Disable Large Receive Offload (LRO) on a net device. Must be 1622 * called under RTNL. This is needed if received packets may be 1623 * forwarded to another interface. 1624 */ 1625 void dev_disable_lro(struct net_device *dev) 1626 { 1627 struct net_device *lower_dev; 1628 struct list_head *iter; 1629 1630 dev->wanted_features &= ~NETIF_F_LRO; 1631 netdev_update_features(dev); 1632 1633 if (unlikely(dev->features & NETIF_F_LRO)) 1634 netdev_WARN(dev, "failed to disable LRO!\n"); 1635 1636 netdev_for_each_lower_dev(dev, lower_dev, iter) 1637 dev_disable_lro(lower_dev); 1638 } 1639 EXPORT_SYMBOL(dev_disable_lro); 1640 1641 /** 1642 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device 1643 * @dev: device 1644 * 1645 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be 1646 * called under RTNL. This is needed if Generic XDP is installed on 1647 * the device. 1648 */ 1649 static void dev_disable_gro_hw(struct net_device *dev) 1650 { 1651 dev->wanted_features &= ~NETIF_F_GRO_HW; 1652 netdev_update_features(dev); 1653 1654 if (unlikely(dev->features & NETIF_F_GRO_HW)) 1655 netdev_WARN(dev, "failed to disable GRO_HW!\n"); 1656 } 1657 1658 const char *netdev_cmd_to_name(enum netdev_cmd cmd) 1659 { 1660 #define N(val) \ 1661 case NETDEV_##val: \ 1662 return "NETDEV_" __stringify(val); 1663 switch (cmd) { 1664 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER) 1665 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE) 1666 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE) 1667 N(POST_INIT) N(PRE_UNINIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) 1668 N(CHANGEUPPER) N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) 1669 N(BONDING_INFO) N(PRECHANGEUPPER) N(CHANGELOWERSTATE) 1670 N(UDP_TUNNEL_PUSH_INFO) N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN) 1671 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO) 1672 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO) 1673 N(PRE_CHANGEADDR) N(OFFLOAD_XSTATS_ENABLE) N(OFFLOAD_XSTATS_DISABLE) 1674 N(OFFLOAD_XSTATS_REPORT_USED) N(OFFLOAD_XSTATS_REPORT_DELTA) 1675 N(XDP_FEAT_CHANGE) 1676 } 1677 #undef N 1678 return "UNKNOWN_NETDEV_EVENT"; 1679 } 1680 EXPORT_SYMBOL_GPL(netdev_cmd_to_name); 1681 1682 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, 1683 struct net_device *dev) 1684 { 1685 struct netdev_notifier_info info = { 1686 .dev = dev, 1687 }; 1688 1689 return nb->notifier_call(nb, val, &info); 1690 } 1691 1692 static int call_netdevice_register_notifiers(struct notifier_block *nb, 1693 struct net_device *dev) 1694 { 1695 int err; 1696 1697 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); 1698 err = notifier_to_errno(err); 1699 if (err) 1700 return err; 1701 1702 if (!(dev->flags & IFF_UP)) 1703 return 0; 1704 1705 call_netdevice_notifier(nb, NETDEV_UP, dev); 1706 return 0; 1707 } 1708 1709 static void call_netdevice_unregister_notifiers(struct notifier_block *nb, 1710 struct net_device *dev) 1711 { 1712 if (dev->flags & IFF_UP) { 1713 call_netdevice_notifier(nb, NETDEV_GOING_DOWN, 1714 dev); 1715 call_netdevice_notifier(nb, NETDEV_DOWN, dev); 1716 } 1717 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); 1718 } 1719 1720 static int call_netdevice_register_net_notifiers(struct notifier_block *nb, 1721 struct net *net) 1722 { 1723 struct net_device *dev; 1724 int err; 1725 1726 for_each_netdev(net, dev) { 1727 err = call_netdevice_register_notifiers(nb, dev); 1728 if (err) 1729 goto rollback; 1730 } 1731 return 0; 1732 1733 rollback: 1734 for_each_netdev_continue_reverse(net, dev) 1735 call_netdevice_unregister_notifiers(nb, dev); 1736 return err; 1737 } 1738 1739 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb, 1740 struct net *net) 1741 { 1742 struct net_device *dev; 1743 1744 for_each_netdev(net, dev) 1745 call_netdevice_unregister_notifiers(nb, dev); 1746 } 1747 1748 static int dev_boot_phase = 1; 1749 1750 /** 1751 * register_netdevice_notifier - register a network notifier block 1752 * @nb: notifier 1753 * 1754 * Register a notifier to be called when network device events occur. 1755 * The notifier passed is linked into the kernel structures and must 1756 * not be reused until it has been unregistered. A negative errno code 1757 * is returned on a failure. 1758 * 1759 * When registered all registration and up events are replayed 1760 * to the new notifier to allow device to have a race free 1761 * view of the network device list. 1762 */ 1763 1764 int register_netdevice_notifier(struct notifier_block *nb) 1765 { 1766 struct net *net; 1767 int err; 1768 1769 /* Close race with setup_net() and cleanup_net() */ 1770 down_write(&pernet_ops_rwsem); 1771 rtnl_lock(); 1772 err = raw_notifier_chain_register(&netdev_chain, nb); 1773 if (err) 1774 goto unlock; 1775 if (dev_boot_phase) 1776 goto unlock; 1777 for_each_net(net) { 1778 err = call_netdevice_register_net_notifiers(nb, net); 1779 if (err) 1780 goto rollback; 1781 } 1782 1783 unlock: 1784 rtnl_unlock(); 1785 up_write(&pernet_ops_rwsem); 1786 return err; 1787 1788 rollback: 1789 for_each_net_continue_reverse(net) 1790 call_netdevice_unregister_net_notifiers(nb, net); 1791 1792 raw_notifier_chain_unregister(&netdev_chain, nb); 1793 goto unlock; 1794 } 1795 EXPORT_SYMBOL(register_netdevice_notifier); 1796 1797 /** 1798 * unregister_netdevice_notifier - unregister a network notifier block 1799 * @nb: notifier 1800 * 1801 * Unregister a notifier previously registered by 1802 * register_netdevice_notifier(). The notifier is unlinked into the 1803 * kernel structures and may then be reused. A negative errno code 1804 * is returned on a failure. 1805 * 1806 * After unregistering unregister and down device events are synthesized 1807 * for all devices on the device list to the removed notifier to remove 1808 * the need for special case cleanup code. 1809 */ 1810 1811 int unregister_netdevice_notifier(struct notifier_block *nb) 1812 { 1813 struct net *net; 1814 int err; 1815 1816 /* Close race with setup_net() and cleanup_net() */ 1817 down_write(&pernet_ops_rwsem); 1818 rtnl_lock(); 1819 err = raw_notifier_chain_unregister(&netdev_chain, nb); 1820 if (err) 1821 goto unlock; 1822 1823 for_each_net(net) 1824 call_netdevice_unregister_net_notifiers(nb, net); 1825 1826 unlock: 1827 rtnl_unlock(); 1828 up_write(&pernet_ops_rwsem); 1829 return err; 1830 } 1831 EXPORT_SYMBOL(unregister_netdevice_notifier); 1832 1833 static int __register_netdevice_notifier_net(struct net *net, 1834 struct notifier_block *nb, 1835 bool ignore_call_fail) 1836 { 1837 int err; 1838 1839 err = raw_notifier_chain_register(&net->netdev_chain, nb); 1840 if (err) 1841 return err; 1842 if (dev_boot_phase) 1843 return 0; 1844 1845 err = call_netdevice_register_net_notifiers(nb, net); 1846 if (err && !ignore_call_fail) 1847 goto chain_unregister; 1848 1849 return 0; 1850 1851 chain_unregister: 1852 raw_notifier_chain_unregister(&net->netdev_chain, nb); 1853 return err; 1854 } 1855 1856 static int __unregister_netdevice_notifier_net(struct net *net, 1857 struct notifier_block *nb) 1858 { 1859 int err; 1860 1861 err = raw_notifier_chain_unregister(&net->netdev_chain, nb); 1862 if (err) 1863 return err; 1864 1865 call_netdevice_unregister_net_notifiers(nb, net); 1866 return 0; 1867 } 1868 1869 /** 1870 * register_netdevice_notifier_net - register a per-netns network notifier block 1871 * @net: network namespace 1872 * @nb: notifier 1873 * 1874 * Register a notifier to be called when network device events occur. 1875 * The notifier passed is linked into the kernel structures and must 1876 * not be reused until it has been unregistered. A negative errno code 1877 * is returned on a failure. 1878 * 1879 * When registered all registration and up events are replayed 1880 * to the new notifier to allow device to have a race free 1881 * view of the network device list. 1882 */ 1883 1884 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb) 1885 { 1886 int err; 1887 1888 rtnl_lock(); 1889 err = __register_netdevice_notifier_net(net, nb, false); 1890 rtnl_unlock(); 1891 return err; 1892 } 1893 EXPORT_SYMBOL(register_netdevice_notifier_net); 1894 1895 /** 1896 * unregister_netdevice_notifier_net - unregister a per-netns 1897 * network notifier block 1898 * @net: network namespace 1899 * @nb: notifier 1900 * 1901 * Unregister a notifier previously registered by 1902 * register_netdevice_notifier_net(). The notifier is unlinked from the 1903 * kernel structures and may then be reused. A negative errno code 1904 * is returned on a failure. 1905 * 1906 * After unregistering unregister and down device events are synthesized 1907 * for all devices on the device list to the removed notifier to remove 1908 * the need for special case cleanup code. 1909 */ 1910 1911 int unregister_netdevice_notifier_net(struct net *net, 1912 struct notifier_block *nb) 1913 { 1914 int err; 1915 1916 rtnl_lock(); 1917 err = __unregister_netdevice_notifier_net(net, nb); 1918 rtnl_unlock(); 1919 return err; 1920 } 1921 EXPORT_SYMBOL(unregister_netdevice_notifier_net); 1922 1923 static void __move_netdevice_notifier_net(struct net *src_net, 1924 struct net *dst_net, 1925 struct notifier_block *nb) 1926 { 1927 __unregister_netdevice_notifier_net(src_net, nb); 1928 __register_netdevice_notifier_net(dst_net, nb, true); 1929 } 1930 1931 int register_netdevice_notifier_dev_net(struct net_device *dev, 1932 struct notifier_block *nb, 1933 struct netdev_net_notifier *nn) 1934 { 1935 int err; 1936 1937 rtnl_lock(); 1938 err = __register_netdevice_notifier_net(dev_net(dev), nb, false); 1939 if (!err) { 1940 nn->nb = nb; 1941 list_add(&nn->list, &dev->net_notifier_list); 1942 } 1943 rtnl_unlock(); 1944 return err; 1945 } 1946 EXPORT_SYMBOL(register_netdevice_notifier_dev_net); 1947 1948 int unregister_netdevice_notifier_dev_net(struct net_device *dev, 1949 struct notifier_block *nb, 1950 struct netdev_net_notifier *nn) 1951 { 1952 int err; 1953 1954 rtnl_lock(); 1955 list_del(&nn->list); 1956 err = __unregister_netdevice_notifier_net(dev_net(dev), nb); 1957 rtnl_unlock(); 1958 return err; 1959 } 1960 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net); 1961 1962 static void move_netdevice_notifiers_dev_net(struct net_device *dev, 1963 struct net *net) 1964 { 1965 struct netdev_net_notifier *nn; 1966 1967 list_for_each_entry(nn, &dev->net_notifier_list, list) 1968 __move_netdevice_notifier_net(dev_net(dev), net, nn->nb); 1969 } 1970 1971 /** 1972 * call_netdevice_notifiers_info - call all network notifier blocks 1973 * @val: value passed unmodified to notifier function 1974 * @info: notifier information data 1975 * 1976 * Call all network notifier blocks. Parameters and return value 1977 * are as for raw_notifier_call_chain(). 1978 */ 1979 1980 int call_netdevice_notifiers_info(unsigned long val, 1981 struct netdev_notifier_info *info) 1982 { 1983 struct net *net = dev_net(info->dev); 1984 int ret; 1985 1986 ASSERT_RTNL(); 1987 1988 /* Run per-netns notifier block chain first, then run the global one. 1989 * Hopefully, one day, the global one is going to be removed after 1990 * all notifier block registrators get converted to be per-netns. 1991 */ 1992 ret = raw_notifier_call_chain(&net->netdev_chain, val, info); 1993 if (ret & NOTIFY_STOP_MASK) 1994 return ret; 1995 return raw_notifier_call_chain(&netdev_chain, val, info); 1996 } 1997 1998 /** 1999 * call_netdevice_notifiers_info_robust - call per-netns notifier blocks 2000 * for and rollback on error 2001 * @val_up: value passed unmodified to notifier function 2002 * @val_down: value passed unmodified to the notifier function when 2003 * recovering from an error on @val_up 2004 * @info: notifier information data 2005 * 2006 * Call all per-netns network notifier blocks, but not notifier blocks on 2007 * the global notifier chain. Parameters and return value are as for 2008 * raw_notifier_call_chain_robust(). 2009 */ 2010 2011 static int 2012 call_netdevice_notifiers_info_robust(unsigned long val_up, 2013 unsigned long val_down, 2014 struct netdev_notifier_info *info) 2015 { 2016 struct net *net = dev_net(info->dev); 2017 2018 ASSERT_RTNL(); 2019 2020 return raw_notifier_call_chain_robust(&net->netdev_chain, 2021 val_up, val_down, info); 2022 } 2023 2024 static int call_netdevice_notifiers_extack(unsigned long val, 2025 struct net_device *dev, 2026 struct netlink_ext_ack *extack) 2027 { 2028 struct netdev_notifier_info info = { 2029 .dev = dev, 2030 .extack = extack, 2031 }; 2032 2033 return call_netdevice_notifiers_info(val, &info); 2034 } 2035 2036 /** 2037 * call_netdevice_notifiers - call all network notifier blocks 2038 * @val: value passed unmodified to notifier function 2039 * @dev: net_device pointer passed unmodified to notifier function 2040 * 2041 * Call all network notifier blocks. Parameters and return value 2042 * are as for raw_notifier_call_chain(). 2043 */ 2044 2045 int call_netdevice_notifiers(unsigned long val, struct net_device *dev) 2046 { 2047 return call_netdevice_notifiers_extack(val, dev, NULL); 2048 } 2049 EXPORT_SYMBOL(call_netdevice_notifiers); 2050 2051 /** 2052 * call_netdevice_notifiers_mtu - call all network notifier blocks 2053 * @val: value passed unmodified to notifier function 2054 * @dev: net_device pointer passed unmodified to notifier function 2055 * @arg: additional u32 argument passed to the notifier function 2056 * 2057 * Call all network notifier blocks. Parameters and return value 2058 * are as for raw_notifier_call_chain(). 2059 */ 2060 static int call_netdevice_notifiers_mtu(unsigned long val, 2061 struct net_device *dev, u32 arg) 2062 { 2063 struct netdev_notifier_info_ext info = { 2064 .info.dev = dev, 2065 .ext.mtu = arg, 2066 }; 2067 2068 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0); 2069 2070 return call_netdevice_notifiers_info(val, &info.info); 2071 } 2072 2073 #ifdef CONFIG_NET_INGRESS 2074 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key); 2075 2076 void net_inc_ingress_queue(void) 2077 { 2078 static_branch_inc(&ingress_needed_key); 2079 } 2080 EXPORT_SYMBOL_GPL(net_inc_ingress_queue); 2081 2082 void net_dec_ingress_queue(void) 2083 { 2084 static_branch_dec(&ingress_needed_key); 2085 } 2086 EXPORT_SYMBOL_GPL(net_dec_ingress_queue); 2087 #endif 2088 2089 #ifdef CONFIG_NET_EGRESS 2090 static DEFINE_STATIC_KEY_FALSE(egress_needed_key); 2091 2092 void net_inc_egress_queue(void) 2093 { 2094 static_branch_inc(&egress_needed_key); 2095 } 2096 EXPORT_SYMBOL_GPL(net_inc_egress_queue); 2097 2098 void net_dec_egress_queue(void) 2099 { 2100 static_branch_dec(&egress_needed_key); 2101 } 2102 EXPORT_SYMBOL_GPL(net_dec_egress_queue); 2103 #endif 2104 2105 DEFINE_STATIC_KEY_FALSE(netstamp_needed_key); 2106 EXPORT_SYMBOL(netstamp_needed_key); 2107 #ifdef CONFIG_JUMP_LABEL 2108 static atomic_t netstamp_needed_deferred; 2109 static atomic_t netstamp_wanted; 2110 static void netstamp_clear(struct work_struct *work) 2111 { 2112 int deferred = atomic_xchg(&netstamp_needed_deferred, 0); 2113 int wanted; 2114 2115 wanted = atomic_add_return(deferred, &netstamp_wanted); 2116 if (wanted > 0) 2117 static_branch_enable(&netstamp_needed_key); 2118 else 2119 static_branch_disable(&netstamp_needed_key); 2120 } 2121 static DECLARE_WORK(netstamp_work, netstamp_clear); 2122 #endif 2123 2124 void net_enable_timestamp(void) 2125 { 2126 #ifdef CONFIG_JUMP_LABEL 2127 int wanted = atomic_read(&netstamp_wanted); 2128 2129 while (wanted > 0) { 2130 if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted + 1)) 2131 return; 2132 } 2133 atomic_inc(&netstamp_needed_deferred); 2134 schedule_work(&netstamp_work); 2135 #else 2136 static_branch_inc(&netstamp_needed_key); 2137 #endif 2138 } 2139 EXPORT_SYMBOL(net_enable_timestamp); 2140 2141 void net_disable_timestamp(void) 2142 { 2143 #ifdef CONFIG_JUMP_LABEL 2144 int wanted = atomic_read(&netstamp_wanted); 2145 2146 while (wanted > 1) { 2147 if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted - 1)) 2148 return; 2149 } 2150 atomic_dec(&netstamp_needed_deferred); 2151 schedule_work(&netstamp_work); 2152 #else 2153 static_branch_dec(&netstamp_needed_key); 2154 #endif 2155 } 2156 EXPORT_SYMBOL(net_disable_timestamp); 2157 2158 static inline void net_timestamp_set(struct sk_buff *skb) 2159 { 2160 skb->tstamp = 0; 2161 skb->mono_delivery_time = 0; 2162 if (static_branch_unlikely(&netstamp_needed_key)) 2163 skb->tstamp = ktime_get_real(); 2164 } 2165 2166 #define net_timestamp_check(COND, SKB) \ 2167 if (static_branch_unlikely(&netstamp_needed_key)) { \ 2168 if ((COND) && !(SKB)->tstamp) \ 2169 (SKB)->tstamp = ktime_get_real(); \ 2170 } \ 2171 2172 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb) 2173 { 2174 return __is_skb_forwardable(dev, skb, true); 2175 } 2176 EXPORT_SYMBOL_GPL(is_skb_forwardable); 2177 2178 static int __dev_forward_skb2(struct net_device *dev, struct sk_buff *skb, 2179 bool check_mtu) 2180 { 2181 int ret = ____dev_forward_skb(dev, skb, check_mtu); 2182 2183 if (likely(!ret)) { 2184 skb->protocol = eth_type_trans(skb, dev); 2185 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); 2186 } 2187 2188 return ret; 2189 } 2190 2191 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2192 { 2193 return __dev_forward_skb2(dev, skb, true); 2194 } 2195 EXPORT_SYMBOL_GPL(__dev_forward_skb); 2196 2197 /** 2198 * dev_forward_skb - loopback an skb to another netif 2199 * 2200 * @dev: destination network device 2201 * @skb: buffer to forward 2202 * 2203 * return values: 2204 * NET_RX_SUCCESS (no congestion) 2205 * NET_RX_DROP (packet was dropped, but freed) 2206 * 2207 * dev_forward_skb can be used for injecting an skb from the 2208 * start_xmit function of one device into the receive queue 2209 * of another device. 2210 * 2211 * The receiving device may be in another namespace, so 2212 * we have to clear all information in the skb that could 2213 * impact namespace isolation. 2214 */ 2215 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2216 { 2217 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); 2218 } 2219 EXPORT_SYMBOL_GPL(dev_forward_skb); 2220 2221 int dev_forward_skb_nomtu(struct net_device *dev, struct sk_buff *skb) 2222 { 2223 return __dev_forward_skb2(dev, skb, false) ?: netif_rx_internal(skb); 2224 } 2225 2226 static inline int deliver_skb(struct sk_buff *skb, 2227 struct packet_type *pt_prev, 2228 struct net_device *orig_dev) 2229 { 2230 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 2231 return -ENOMEM; 2232 refcount_inc(&skb->users); 2233 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 2234 } 2235 2236 static inline void deliver_ptype_list_skb(struct sk_buff *skb, 2237 struct packet_type **pt, 2238 struct net_device *orig_dev, 2239 __be16 type, 2240 struct list_head *ptype_list) 2241 { 2242 struct packet_type *ptype, *pt_prev = *pt; 2243 2244 list_for_each_entry_rcu(ptype, ptype_list, list) { 2245 if (ptype->type != type) 2246 continue; 2247 if (pt_prev) 2248 deliver_skb(skb, pt_prev, orig_dev); 2249 pt_prev = ptype; 2250 } 2251 *pt = pt_prev; 2252 } 2253 2254 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) 2255 { 2256 if (!ptype->af_packet_priv || !skb->sk) 2257 return false; 2258 2259 if (ptype->id_match) 2260 return ptype->id_match(ptype, skb->sk); 2261 else if ((struct sock *)ptype->af_packet_priv == skb->sk) 2262 return true; 2263 2264 return false; 2265 } 2266 2267 /** 2268 * dev_nit_active - return true if any network interface taps are in use 2269 * 2270 * @dev: network device to check for the presence of taps 2271 */ 2272 bool dev_nit_active(struct net_device *dev) 2273 { 2274 return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all); 2275 } 2276 EXPORT_SYMBOL_GPL(dev_nit_active); 2277 2278 /* 2279 * Support routine. Sends outgoing frames to any network 2280 * taps currently in use. 2281 */ 2282 2283 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) 2284 { 2285 struct packet_type *ptype; 2286 struct sk_buff *skb2 = NULL; 2287 struct packet_type *pt_prev = NULL; 2288 struct list_head *ptype_list = &ptype_all; 2289 2290 rcu_read_lock(); 2291 again: 2292 list_for_each_entry_rcu(ptype, ptype_list, list) { 2293 if (READ_ONCE(ptype->ignore_outgoing)) 2294 continue; 2295 2296 /* Never send packets back to the socket 2297 * they originated from - MvS (miquels@drinkel.ow.org) 2298 */ 2299 if (skb_loop_sk(ptype, skb)) 2300 continue; 2301 2302 if (pt_prev) { 2303 deliver_skb(skb2, pt_prev, skb->dev); 2304 pt_prev = ptype; 2305 continue; 2306 } 2307 2308 /* need to clone skb, done only once */ 2309 skb2 = skb_clone(skb, GFP_ATOMIC); 2310 if (!skb2) 2311 goto out_unlock; 2312 2313 net_timestamp_set(skb2); 2314 2315 /* skb->nh should be correctly 2316 * set by sender, so that the second statement is 2317 * just protection against buggy protocols. 2318 */ 2319 skb_reset_mac_header(skb2); 2320 2321 if (skb_network_header(skb2) < skb2->data || 2322 skb_network_header(skb2) > skb_tail_pointer(skb2)) { 2323 net_crit_ratelimited("protocol %04x is buggy, dev %s\n", 2324 ntohs(skb2->protocol), 2325 dev->name); 2326 skb_reset_network_header(skb2); 2327 } 2328 2329 skb2->transport_header = skb2->network_header; 2330 skb2->pkt_type = PACKET_OUTGOING; 2331 pt_prev = ptype; 2332 } 2333 2334 if (ptype_list == &ptype_all) { 2335 ptype_list = &dev->ptype_all; 2336 goto again; 2337 } 2338 out_unlock: 2339 if (pt_prev) { 2340 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC)) 2341 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); 2342 else 2343 kfree_skb(skb2); 2344 } 2345 rcu_read_unlock(); 2346 } 2347 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit); 2348 2349 /** 2350 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change 2351 * @dev: Network device 2352 * @txq: number of queues available 2353 * 2354 * If real_num_tx_queues is changed the tc mappings may no longer be 2355 * valid. To resolve this verify the tc mapping remains valid and if 2356 * not NULL the mapping. With no priorities mapping to this 2357 * offset/count pair it will no longer be used. In the worst case TC0 2358 * is invalid nothing can be done so disable priority mappings. If is 2359 * expected that drivers will fix this mapping if they can before 2360 * calling netif_set_real_num_tx_queues. 2361 */ 2362 static void netif_setup_tc(struct net_device *dev, unsigned int txq) 2363 { 2364 int i; 2365 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2366 2367 /* If TC0 is invalidated disable TC mapping */ 2368 if (tc->offset + tc->count > txq) { 2369 netdev_warn(dev, "Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); 2370 dev->num_tc = 0; 2371 return; 2372 } 2373 2374 /* Invalidated prio to tc mappings set to TC0 */ 2375 for (i = 1; i < TC_BITMASK + 1; i++) { 2376 int q = netdev_get_prio_tc_map(dev, i); 2377 2378 tc = &dev->tc_to_txq[q]; 2379 if (tc->offset + tc->count > txq) { 2380 netdev_warn(dev, "Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n", 2381 i, q); 2382 netdev_set_prio_tc_map(dev, i, 0); 2383 } 2384 } 2385 } 2386 2387 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq) 2388 { 2389 if (dev->num_tc) { 2390 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2391 int i; 2392 2393 /* walk through the TCs and see if it falls into any of them */ 2394 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) { 2395 if ((txq - tc->offset) < tc->count) 2396 return i; 2397 } 2398 2399 /* didn't find it, just return -1 to indicate no match */ 2400 return -1; 2401 } 2402 2403 return 0; 2404 } 2405 EXPORT_SYMBOL(netdev_txq_to_tc); 2406 2407 #ifdef CONFIG_XPS 2408 static struct static_key xps_needed __read_mostly; 2409 static struct static_key xps_rxqs_needed __read_mostly; 2410 static DEFINE_MUTEX(xps_map_mutex); 2411 #define xmap_dereference(P) \ 2412 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) 2413 2414 static bool remove_xps_queue(struct xps_dev_maps *dev_maps, 2415 struct xps_dev_maps *old_maps, int tci, u16 index) 2416 { 2417 struct xps_map *map = NULL; 2418 int pos; 2419 2420 map = xmap_dereference(dev_maps->attr_map[tci]); 2421 if (!map) 2422 return false; 2423 2424 for (pos = map->len; pos--;) { 2425 if (map->queues[pos] != index) 2426 continue; 2427 2428 if (map->len > 1) { 2429 map->queues[pos] = map->queues[--map->len]; 2430 break; 2431 } 2432 2433 if (old_maps) 2434 RCU_INIT_POINTER(old_maps->attr_map[tci], NULL); 2435 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2436 kfree_rcu(map, rcu); 2437 return false; 2438 } 2439 2440 return true; 2441 } 2442 2443 static bool remove_xps_queue_cpu(struct net_device *dev, 2444 struct xps_dev_maps *dev_maps, 2445 int cpu, u16 offset, u16 count) 2446 { 2447 int num_tc = dev_maps->num_tc; 2448 bool active = false; 2449 int tci; 2450 2451 for (tci = cpu * num_tc; num_tc--; tci++) { 2452 int i, j; 2453 2454 for (i = count, j = offset; i--; j++) { 2455 if (!remove_xps_queue(dev_maps, NULL, tci, j)) 2456 break; 2457 } 2458 2459 active |= i < 0; 2460 } 2461 2462 return active; 2463 } 2464 2465 static void reset_xps_maps(struct net_device *dev, 2466 struct xps_dev_maps *dev_maps, 2467 enum xps_map_type type) 2468 { 2469 static_key_slow_dec_cpuslocked(&xps_needed); 2470 if (type == XPS_RXQS) 2471 static_key_slow_dec_cpuslocked(&xps_rxqs_needed); 2472 2473 RCU_INIT_POINTER(dev->xps_maps[type], NULL); 2474 2475 kfree_rcu(dev_maps, rcu); 2476 } 2477 2478 static void clean_xps_maps(struct net_device *dev, enum xps_map_type type, 2479 u16 offset, u16 count) 2480 { 2481 struct xps_dev_maps *dev_maps; 2482 bool active = false; 2483 int i, j; 2484 2485 dev_maps = xmap_dereference(dev->xps_maps[type]); 2486 if (!dev_maps) 2487 return; 2488 2489 for (j = 0; j < dev_maps->nr_ids; j++) 2490 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, count); 2491 if (!active) 2492 reset_xps_maps(dev, dev_maps, type); 2493 2494 if (type == XPS_CPUS) { 2495 for (i = offset + (count - 1); count--; i--) 2496 netdev_queue_numa_node_write( 2497 netdev_get_tx_queue(dev, i), NUMA_NO_NODE); 2498 } 2499 } 2500 2501 static void netif_reset_xps_queues(struct net_device *dev, u16 offset, 2502 u16 count) 2503 { 2504 if (!static_key_false(&xps_needed)) 2505 return; 2506 2507 cpus_read_lock(); 2508 mutex_lock(&xps_map_mutex); 2509 2510 if (static_key_false(&xps_rxqs_needed)) 2511 clean_xps_maps(dev, XPS_RXQS, offset, count); 2512 2513 clean_xps_maps(dev, XPS_CPUS, offset, count); 2514 2515 mutex_unlock(&xps_map_mutex); 2516 cpus_read_unlock(); 2517 } 2518 2519 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) 2520 { 2521 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index); 2522 } 2523 2524 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index, 2525 u16 index, bool is_rxqs_map) 2526 { 2527 struct xps_map *new_map; 2528 int alloc_len = XPS_MIN_MAP_ALLOC; 2529 int i, pos; 2530 2531 for (pos = 0; map && pos < map->len; pos++) { 2532 if (map->queues[pos] != index) 2533 continue; 2534 return map; 2535 } 2536 2537 /* Need to add tx-queue to this CPU's/rx-queue's existing map */ 2538 if (map) { 2539 if (pos < map->alloc_len) 2540 return map; 2541 2542 alloc_len = map->alloc_len * 2; 2543 } 2544 2545 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's 2546 * map 2547 */ 2548 if (is_rxqs_map) 2549 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL); 2550 else 2551 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, 2552 cpu_to_node(attr_index)); 2553 if (!new_map) 2554 return NULL; 2555 2556 for (i = 0; i < pos; i++) 2557 new_map->queues[i] = map->queues[i]; 2558 new_map->alloc_len = alloc_len; 2559 new_map->len = pos; 2560 2561 return new_map; 2562 } 2563 2564 /* Copy xps maps at a given index */ 2565 static void xps_copy_dev_maps(struct xps_dev_maps *dev_maps, 2566 struct xps_dev_maps *new_dev_maps, int index, 2567 int tc, bool skip_tc) 2568 { 2569 int i, tci = index * dev_maps->num_tc; 2570 struct xps_map *map; 2571 2572 /* copy maps belonging to foreign traffic classes */ 2573 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2574 if (i == tc && skip_tc) 2575 continue; 2576 2577 /* fill in the new device map from the old device map */ 2578 map = xmap_dereference(dev_maps->attr_map[tci]); 2579 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2580 } 2581 } 2582 2583 /* Must be called under cpus_read_lock */ 2584 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask, 2585 u16 index, enum xps_map_type type) 2586 { 2587 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL, *old_dev_maps = NULL; 2588 const unsigned long *online_mask = NULL; 2589 bool active = false, copy = false; 2590 int i, j, tci, numa_node_id = -2; 2591 int maps_sz, num_tc = 1, tc = 0; 2592 struct xps_map *map, *new_map; 2593 unsigned int nr_ids; 2594 2595 WARN_ON_ONCE(index >= dev->num_tx_queues); 2596 2597 if (dev->num_tc) { 2598 /* Do not allow XPS on subordinate device directly */ 2599 num_tc = dev->num_tc; 2600 if (num_tc < 0) 2601 return -EINVAL; 2602 2603 /* If queue belongs to subordinate dev use its map */ 2604 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev; 2605 2606 tc = netdev_txq_to_tc(dev, index); 2607 if (tc < 0) 2608 return -EINVAL; 2609 } 2610 2611 mutex_lock(&xps_map_mutex); 2612 2613 dev_maps = xmap_dereference(dev->xps_maps[type]); 2614 if (type == XPS_RXQS) { 2615 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues); 2616 nr_ids = dev->num_rx_queues; 2617 } else { 2618 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc); 2619 if (num_possible_cpus() > 1) 2620 online_mask = cpumask_bits(cpu_online_mask); 2621 nr_ids = nr_cpu_ids; 2622 } 2623 2624 if (maps_sz < L1_CACHE_BYTES) 2625 maps_sz = L1_CACHE_BYTES; 2626 2627 /* The old dev_maps could be larger or smaller than the one we're 2628 * setting up now, as dev->num_tc or nr_ids could have been updated in 2629 * between. We could try to be smart, but let's be safe instead and only 2630 * copy foreign traffic classes if the two map sizes match. 2631 */ 2632 if (dev_maps && 2633 dev_maps->num_tc == num_tc && dev_maps->nr_ids == nr_ids) 2634 copy = true; 2635 2636 /* allocate memory for queue storage */ 2637 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids), 2638 j < nr_ids;) { 2639 if (!new_dev_maps) { 2640 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); 2641 if (!new_dev_maps) { 2642 mutex_unlock(&xps_map_mutex); 2643 return -ENOMEM; 2644 } 2645 2646 new_dev_maps->nr_ids = nr_ids; 2647 new_dev_maps->num_tc = num_tc; 2648 } 2649 2650 tci = j * num_tc + tc; 2651 map = copy ? xmap_dereference(dev_maps->attr_map[tci]) : NULL; 2652 2653 map = expand_xps_map(map, j, index, type == XPS_RXQS); 2654 if (!map) 2655 goto error; 2656 2657 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2658 } 2659 2660 if (!new_dev_maps) 2661 goto out_no_new_maps; 2662 2663 if (!dev_maps) { 2664 /* Increment static keys at most once per type */ 2665 static_key_slow_inc_cpuslocked(&xps_needed); 2666 if (type == XPS_RXQS) 2667 static_key_slow_inc_cpuslocked(&xps_rxqs_needed); 2668 } 2669 2670 for (j = 0; j < nr_ids; j++) { 2671 bool skip_tc = false; 2672 2673 tci = j * num_tc + tc; 2674 if (netif_attr_test_mask(j, mask, nr_ids) && 2675 netif_attr_test_online(j, online_mask, nr_ids)) { 2676 /* add tx-queue to CPU/rx-queue maps */ 2677 int pos = 0; 2678 2679 skip_tc = true; 2680 2681 map = xmap_dereference(new_dev_maps->attr_map[tci]); 2682 while ((pos < map->len) && (map->queues[pos] != index)) 2683 pos++; 2684 2685 if (pos == map->len) 2686 map->queues[map->len++] = index; 2687 #ifdef CONFIG_NUMA 2688 if (type == XPS_CPUS) { 2689 if (numa_node_id == -2) 2690 numa_node_id = cpu_to_node(j); 2691 else if (numa_node_id != cpu_to_node(j)) 2692 numa_node_id = -1; 2693 } 2694 #endif 2695 } 2696 2697 if (copy) 2698 xps_copy_dev_maps(dev_maps, new_dev_maps, j, tc, 2699 skip_tc); 2700 } 2701 2702 rcu_assign_pointer(dev->xps_maps[type], new_dev_maps); 2703 2704 /* Cleanup old maps */ 2705 if (!dev_maps) 2706 goto out_no_old_maps; 2707 2708 for (j = 0; j < dev_maps->nr_ids; j++) { 2709 for (i = num_tc, tci = j * dev_maps->num_tc; i--; tci++) { 2710 map = xmap_dereference(dev_maps->attr_map[tci]); 2711 if (!map) 2712 continue; 2713 2714 if (copy) { 2715 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2716 if (map == new_map) 2717 continue; 2718 } 2719 2720 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2721 kfree_rcu(map, rcu); 2722 } 2723 } 2724 2725 old_dev_maps = dev_maps; 2726 2727 out_no_old_maps: 2728 dev_maps = new_dev_maps; 2729 active = true; 2730 2731 out_no_new_maps: 2732 if (type == XPS_CPUS) 2733 /* update Tx queue numa node */ 2734 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), 2735 (numa_node_id >= 0) ? 2736 numa_node_id : NUMA_NO_NODE); 2737 2738 if (!dev_maps) 2739 goto out_no_maps; 2740 2741 /* removes tx-queue from unused CPUs/rx-queues */ 2742 for (j = 0; j < dev_maps->nr_ids; j++) { 2743 tci = j * dev_maps->num_tc; 2744 2745 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2746 if (i == tc && 2747 netif_attr_test_mask(j, mask, dev_maps->nr_ids) && 2748 netif_attr_test_online(j, online_mask, dev_maps->nr_ids)) 2749 continue; 2750 2751 active |= remove_xps_queue(dev_maps, 2752 copy ? old_dev_maps : NULL, 2753 tci, index); 2754 } 2755 } 2756 2757 if (old_dev_maps) 2758 kfree_rcu(old_dev_maps, rcu); 2759 2760 /* free map if not active */ 2761 if (!active) 2762 reset_xps_maps(dev, dev_maps, type); 2763 2764 out_no_maps: 2765 mutex_unlock(&xps_map_mutex); 2766 2767 return 0; 2768 error: 2769 /* remove any maps that we added */ 2770 for (j = 0; j < nr_ids; j++) { 2771 for (i = num_tc, tci = j * num_tc; i--; tci++) { 2772 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2773 map = copy ? 2774 xmap_dereference(dev_maps->attr_map[tci]) : 2775 NULL; 2776 if (new_map && new_map != map) 2777 kfree(new_map); 2778 } 2779 } 2780 2781 mutex_unlock(&xps_map_mutex); 2782 2783 kfree(new_dev_maps); 2784 return -ENOMEM; 2785 } 2786 EXPORT_SYMBOL_GPL(__netif_set_xps_queue); 2787 2788 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, 2789 u16 index) 2790 { 2791 int ret; 2792 2793 cpus_read_lock(); 2794 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, XPS_CPUS); 2795 cpus_read_unlock(); 2796 2797 return ret; 2798 } 2799 EXPORT_SYMBOL(netif_set_xps_queue); 2800 2801 #endif 2802 static void netdev_unbind_all_sb_channels(struct net_device *dev) 2803 { 2804 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2805 2806 /* Unbind any subordinate channels */ 2807 while (txq-- != &dev->_tx[0]) { 2808 if (txq->sb_dev) 2809 netdev_unbind_sb_channel(dev, txq->sb_dev); 2810 } 2811 } 2812 2813 void netdev_reset_tc(struct net_device *dev) 2814 { 2815 #ifdef CONFIG_XPS 2816 netif_reset_xps_queues_gt(dev, 0); 2817 #endif 2818 netdev_unbind_all_sb_channels(dev); 2819 2820 /* Reset TC configuration of device */ 2821 dev->num_tc = 0; 2822 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq)); 2823 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map)); 2824 } 2825 EXPORT_SYMBOL(netdev_reset_tc); 2826 2827 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset) 2828 { 2829 if (tc >= dev->num_tc) 2830 return -EINVAL; 2831 2832 #ifdef CONFIG_XPS 2833 netif_reset_xps_queues(dev, offset, count); 2834 #endif 2835 dev->tc_to_txq[tc].count = count; 2836 dev->tc_to_txq[tc].offset = offset; 2837 return 0; 2838 } 2839 EXPORT_SYMBOL(netdev_set_tc_queue); 2840 2841 int netdev_set_num_tc(struct net_device *dev, u8 num_tc) 2842 { 2843 if (num_tc > TC_MAX_QUEUE) 2844 return -EINVAL; 2845 2846 #ifdef CONFIG_XPS 2847 netif_reset_xps_queues_gt(dev, 0); 2848 #endif 2849 netdev_unbind_all_sb_channels(dev); 2850 2851 dev->num_tc = num_tc; 2852 return 0; 2853 } 2854 EXPORT_SYMBOL(netdev_set_num_tc); 2855 2856 void netdev_unbind_sb_channel(struct net_device *dev, 2857 struct net_device *sb_dev) 2858 { 2859 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2860 2861 #ifdef CONFIG_XPS 2862 netif_reset_xps_queues_gt(sb_dev, 0); 2863 #endif 2864 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq)); 2865 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map)); 2866 2867 while (txq-- != &dev->_tx[0]) { 2868 if (txq->sb_dev == sb_dev) 2869 txq->sb_dev = NULL; 2870 } 2871 } 2872 EXPORT_SYMBOL(netdev_unbind_sb_channel); 2873 2874 int netdev_bind_sb_channel_queue(struct net_device *dev, 2875 struct net_device *sb_dev, 2876 u8 tc, u16 count, u16 offset) 2877 { 2878 /* Make certain the sb_dev and dev are already configured */ 2879 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc) 2880 return -EINVAL; 2881 2882 /* We cannot hand out queues we don't have */ 2883 if ((offset + count) > dev->real_num_tx_queues) 2884 return -EINVAL; 2885 2886 /* Record the mapping */ 2887 sb_dev->tc_to_txq[tc].count = count; 2888 sb_dev->tc_to_txq[tc].offset = offset; 2889 2890 /* Provide a way for Tx queue to find the tc_to_txq map or 2891 * XPS map for itself. 2892 */ 2893 while (count--) 2894 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev; 2895 2896 return 0; 2897 } 2898 EXPORT_SYMBOL(netdev_bind_sb_channel_queue); 2899 2900 int netdev_set_sb_channel(struct net_device *dev, u16 channel) 2901 { 2902 /* Do not use a multiqueue device to represent a subordinate channel */ 2903 if (netif_is_multiqueue(dev)) 2904 return -ENODEV; 2905 2906 /* We allow channels 1 - 32767 to be used for subordinate channels. 2907 * Channel 0 is meant to be "native" mode and used only to represent 2908 * the main root device. We allow writing 0 to reset the device back 2909 * to normal mode after being used as a subordinate channel. 2910 */ 2911 if (channel > S16_MAX) 2912 return -EINVAL; 2913 2914 dev->num_tc = -channel; 2915 2916 return 0; 2917 } 2918 EXPORT_SYMBOL(netdev_set_sb_channel); 2919 2920 /* 2921 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues 2922 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed. 2923 */ 2924 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) 2925 { 2926 bool disabling; 2927 int rc; 2928 2929 disabling = txq < dev->real_num_tx_queues; 2930 2931 if (txq < 1 || txq > dev->num_tx_queues) 2932 return -EINVAL; 2933 2934 if (dev->reg_state == NETREG_REGISTERED || 2935 dev->reg_state == NETREG_UNREGISTERING) { 2936 ASSERT_RTNL(); 2937 2938 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, 2939 txq); 2940 if (rc) 2941 return rc; 2942 2943 if (dev->num_tc) 2944 netif_setup_tc(dev, txq); 2945 2946 dev_qdisc_change_real_num_tx(dev, txq); 2947 2948 dev->real_num_tx_queues = txq; 2949 2950 if (disabling) { 2951 synchronize_net(); 2952 qdisc_reset_all_tx_gt(dev, txq); 2953 #ifdef CONFIG_XPS 2954 netif_reset_xps_queues_gt(dev, txq); 2955 #endif 2956 } 2957 } else { 2958 dev->real_num_tx_queues = txq; 2959 } 2960 2961 return 0; 2962 } 2963 EXPORT_SYMBOL(netif_set_real_num_tx_queues); 2964 2965 #ifdef CONFIG_SYSFS 2966 /** 2967 * netif_set_real_num_rx_queues - set actual number of RX queues used 2968 * @dev: Network device 2969 * @rxq: Actual number of RX queues 2970 * 2971 * This must be called either with the rtnl_lock held or before 2972 * registration of the net device. Returns 0 on success, or a 2973 * negative error code. If called before registration, it always 2974 * succeeds. 2975 */ 2976 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) 2977 { 2978 int rc; 2979 2980 if (rxq < 1 || rxq > dev->num_rx_queues) 2981 return -EINVAL; 2982 2983 if (dev->reg_state == NETREG_REGISTERED) { 2984 ASSERT_RTNL(); 2985 2986 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, 2987 rxq); 2988 if (rc) 2989 return rc; 2990 } 2991 2992 dev->real_num_rx_queues = rxq; 2993 return 0; 2994 } 2995 EXPORT_SYMBOL(netif_set_real_num_rx_queues); 2996 #endif 2997 2998 /** 2999 * netif_set_real_num_queues - set actual number of RX and TX queues used 3000 * @dev: Network device 3001 * @txq: Actual number of TX queues 3002 * @rxq: Actual number of RX queues 3003 * 3004 * Set the real number of both TX and RX queues. 3005 * Does nothing if the number of queues is already correct. 3006 */ 3007 int netif_set_real_num_queues(struct net_device *dev, 3008 unsigned int txq, unsigned int rxq) 3009 { 3010 unsigned int old_rxq = dev->real_num_rx_queues; 3011 int err; 3012 3013 if (txq < 1 || txq > dev->num_tx_queues || 3014 rxq < 1 || rxq > dev->num_rx_queues) 3015 return -EINVAL; 3016 3017 /* Start from increases, so the error path only does decreases - 3018 * decreases can't fail. 3019 */ 3020 if (rxq > dev->real_num_rx_queues) { 3021 err = netif_set_real_num_rx_queues(dev, rxq); 3022 if (err) 3023 return err; 3024 } 3025 if (txq > dev->real_num_tx_queues) { 3026 err = netif_set_real_num_tx_queues(dev, txq); 3027 if (err) 3028 goto undo_rx; 3029 } 3030 if (rxq < dev->real_num_rx_queues) 3031 WARN_ON(netif_set_real_num_rx_queues(dev, rxq)); 3032 if (txq < dev->real_num_tx_queues) 3033 WARN_ON(netif_set_real_num_tx_queues(dev, txq)); 3034 3035 return 0; 3036 undo_rx: 3037 WARN_ON(netif_set_real_num_rx_queues(dev, old_rxq)); 3038 return err; 3039 } 3040 EXPORT_SYMBOL(netif_set_real_num_queues); 3041 3042 /** 3043 * netif_set_tso_max_size() - set the max size of TSO frames supported 3044 * @dev: netdev to update 3045 * @size: max skb->len of a TSO frame 3046 * 3047 * Set the limit on the size of TSO super-frames the device can handle. 3048 * Unless explicitly set the stack will assume the value of 3049 * %GSO_LEGACY_MAX_SIZE. 3050 */ 3051 void netif_set_tso_max_size(struct net_device *dev, unsigned int size) 3052 { 3053 dev->tso_max_size = min(GSO_MAX_SIZE, size); 3054 if (size < READ_ONCE(dev->gso_max_size)) 3055 netif_set_gso_max_size(dev, size); 3056 if (size < READ_ONCE(dev->gso_ipv4_max_size)) 3057 netif_set_gso_ipv4_max_size(dev, size); 3058 } 3059 EXPORT_SYMBOL(netif_set_tso_max_size); 3060 3061 /** 3062 * netif_set_tso_max_segs() - set the max number of segs supported for TSO 3063 * @dev: netdev to update 3064 * @segs: max number of TCP segments 3065 * 3066 * Set the limit on the number of TCP segments the device can generate from 3067 * a single TSO super-frame. 3068 * Unless explicitly set the stack will assume the value of %GSO_MAX_SEGS. 3069 */ 3070 void netif_set_tso_max_segs(struct net_device *dev, unsigned int segs) 3071 { 3072 dev->tso_max_segs = segs; 3073 if (segs < READ_ONCE(dev->gso_max_segs)) 3074 netif_set_gso_max_segs(dev, segs); 3075 } 3076 EXPORT_SYMBOL(netif_set_tso_max_segs); 3077 3078 /** 3079 * netif_inherit_tso_max() - copy all TSO limits from a lower device to an upper 3080 * @to: netdev to update 3081 * @from: netdev from which to copy the limits 3082 */ 3083 void netif_inherit_tso_max(struct net_device *to, const struct net_device *from) 3084 { 3085 netif_set_tso_max_size(to, from->tso_max_size); 3086 netif_set_tso_max_segs(to, from->tso_max_segs); 3087 } 3088 EXPORT_SYMBOL(netif_inherit_tso_max); 3089 3090 /** 3091 * netif_get_num_default_rss_queues - default number of RSS queues 3092 * 3093 * Default value is the number of physical cores if there are only 1 or 2, or 3094 * divided by 2 if there are more. 3095 */ 3096 int netif_get_num_default_rss_queues(void) 3097 { 3098 cpumask_var_t cpus; 3099 int cpu, count = 0; 3100 3101 if (unlikely(is_kdump_kernel() || !zalloc_cpumask_var(&cpus, GFP_KERNEL))) 3102 return 1; 3103 3104 cpumask_copy(cpus, cpu_online_mask); 3105 for_each_cpu(cpu, cpus) { 3106 ++count; 3107 cpumask_andnot(cpus, cpus, topology_sibling_cpumask(cpu)); 3108 } 3109 free_cpumask_var(cpus); 3110 3111 return count > 2 ? DIV_ROUND_UP(count, 2) : count; 3112 } 3113 EXPORT_SYMBOL(netif_get_num_default_rss_queues); 3114 3115 static void __netif_reschedule(struct Qdisc *q) 3116 { 3117 struct softnet_data *sd; 3118 unsigned long flags; 3119 3120 local_irq_save(flags); 3121 sd = this_cpu_ptr(&softnet_data); 3122 q->next_sched = NULL; 3123 *sd->output_queue_tailp = q; 3124 sd->output_queue_tailp = &q->next_sched; 3125 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3126 local_irq_restore(flags); 3127 } 3128 3129 void __netif_schedule(struct Qdisc *q) 3130 { 3131 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) 3132 __netif_reschedule(q); 3133 } 3134 EXPORT_SYMBOL(__netif_schedule); 3135 3136 struct dev_kfree_skb_cb { 3137 enum skb_drop_reason reason; 3138 }; 3139 3140 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) 3141 { 3142 return (struct dev_kfree_skb_cb *)skb->cb; 3143 } 3144 3145 void netif_schedule_queue(struct netdev_queue *txq) 3146 { 3147 rcu_read_lock(); 3148 if (!netif_xmit_stopped(txq)) { 3149 struct Qdisc *q = rcu_dereference(txq->qdisc); 3150 3151 __netif_schedule(q); 3152 } 3153 rcu_read_unlock(); 3154 } 3155 EXPORT_SYMBOL(netif_schedule_queue); 3156 3157 void netif_tx_wake_queue(struct netdev_queue *dev_queue) 3158 { 3159 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { 3160 struct Qdisc *q; 3161 3162 rcu_read_lock(); 3163 q = rcu_dereference(dev_queue->qdisc); 3164 __netif_schedule(q); 3165 rcu_read_unlock(); 3166 } 3167 } 3168 EXPORT_SYMBOL(netif_tx_wake_queue); 3169 3170 void dev_kfree_skb_irq_reason(struct sk_buff *skb, enum skb_drop_reason reason) 3171 { 3172 unsigned long flags; 3173 3174 if (unlikely(!skb)) 3175 return; 3176 3177 if (likely(refcount_read(&skb->users) == 1)) { 3178 smp_rmb(); 3179 refcount_set(&skb->users, 0); 3180 } else if (likely(!refcount_dec_and_test(&skb->users))) { 3181 return; 3182 } 3183 get_kfree_skb_cb(skb)->reason = reason; 3184 local_irq_save(flags); 3185 skb->next = __this_cpu_read(softnet_data.completion_queue); 3186 __this_cpu_write(softnet_data.completion_queue, skb); 3187 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3188 local_irq_restore(flags); 3189 } 3190 EXPORT_SYMBOL(dev_kfree_skb_irq_reason); 3191 3192 void dev_kfree_skb_any_reason(struct sk_buff *skb, enum skb_drop_reason reason) 3193 { 3194 if (in_hardirq() || irqs_disabled()) 3195 dev_kfree_skb_irq_reason(skb, reason); 3196 else 3197 kfree_skb_reason(skb, reason); 3198 } 3199 EXPORT_SYMBOL(dev_kfree_skb_any_reason); 3200 3201 3202 /** 3203 * netif_device_detach - mark device as removed 3204 * @dev: network device 3205 * 3206 * Mark device as removed from system and therefore no longer available. 3207 */ 3208 void netif_device_detach(struct net_device *dev) 3209 { 3210 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && 3211 netif_running(dev)) { 3212 netif_tx_stop_all_queues(dev); 3213 } 3214 } 3215 EXPORT_SYMBOL(netif_device_detach); 3216 3217 /** 3218 * netif_device_attach - mark device as attached 3219 * @dev: network device 3220 * 3221 * Mark device as attached from system and restart if needed. 3222 */ 3223 void netif_device_attach(struct net_device *dev) 3224 { 3225 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && 3226 netif_running(dev)) { 3227 netif_tx_wake_all_queues(dev); 3228 __netdev_watchdog_up(dev); 3229 } 3230 } 3231 EXPORT_SYMBOL(netif_device_attach); 3232 3233 /* 3234 * Returns a Tx hash based on the given packet descriptor a Tx queues' number 3235 * to be used as a distribution range. 3236 */ 3237 static u16 skb_tx_hash(const struct net_device *dev, 3238 const struct net_device *sb_dev, 3239 struct sk_buff *skb) 3240 { 3241 u32 hash; 3242 u16 qoffset = 0; 3243 u16 qcount = dev->real_num_tx_queues; 3244 3245 if (dev->num_tc) { 3246 u8 tc = netdev_get_prio_tc_map(dev, skb->priority); 3247 3248 qoffset = sb_dev->tc_to_txq[tc].offset; 3249 qcount = sb_dev->tc_to_txq[tc].count; 3250 if (unlikely(!qcount)) { 3251 net_warn_ratelimited("%s: invalid qcount, qoffset %u for tc %u\n", 3252 sb_dev->name, qoffset, tc); 3253 qoffset = 0; 3254 qcount = dev->real_num_tx_queues; 3255 } 3256 } 3257 3258 if (skb_rx_queue_recorded(skb)) { 3259 DEBUG_NET_WARN_ON_ONCE(qcount == 0); 3260 hash = skb_get_rx_queue(skb); 3261 if (hash >= qoffset) 3262 hash -= qoffset; 3263 while (unlikely(hash >= qcount)) 3264 hash -= qcount; 3265 return hash + qoffset; 3266 } 3267 3268 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; 3269 } 3270 3271 void skb_warn_bad_offload(const struct sk_buff *skb) 3272 { 3273 static const netdev_features_t null_features; 3274 struct net_device *dev = skb->dev; 3275 const char *name = ""; 3276 3277 if (!net_ratelimit()) 3278 return; 3279 3280 if (dev) { 3281 if (dev->dev.parent) 3282 name = dev_driver_string(dev->dev.parent); 3283 else 3284 name = netdev_name(dev); 3285 } 3286 skb_dump(KERN_WARNING, skb, false); 3287 WARN(1, "%s: caps=(%pNF, %pNF)\n", 3288 name, dev ? &dev->features : &null_features, 3289 skb->sk ? &skb->sk->sk_route_caps : &null_features); 3290 } 3291 3292 /* 3293 * Invalidate hardware checksum when packet is to be mangled, and 3294 * complete checksum manually on outgoing path. 3295 */ 3296 int skb_checksum_help(struct sk_buff *skb) 3297 { 3298 __wsum csum; 3299 int ret = 0, offset; 3300 3301 if (skb->ip_summed == CHECKSUM_COMPLETE) 3302 goto out_set_summed; 3303 3304 if (unlikely(skb_is_gso(skb))) { 3305 skb_warn_bad_offload(skb); 3306 return -EINVAL; 3307 } 3308 3309 /* Before computing a checksum, we should make sure no frag could 3310 * be modified by an external entity : checksum could be wrong. 3311 */ 3312 if (skb_has_shared_frag(skb)) { 3313 ret = __skb_linearize(skb); 3314 if (ret) 3315 goto out; 3316 } 3317 3318 offset = skb_checksum_start_offset(skb); 3319 ret = -EINVAL; 3320 if (unlikely(offset >= skb_headlen(skb))) { 3321 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false); 3322 WARN_ONCE(true, "offset (%d) >= skb_headlen() (%u)\n", 3323 offset, skb_headlen(skb)); 3324 goto out; 3325 } 3326 csum = skb_checksum(skb, offset, skb->len - offset, 0); 3327 3328 offset += skb->csum_offset; 3329 if (unlikely(offset + sizeof(__sum16) > skb_headlen(skb))) { 3330 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false); 3331 WARN_ONCE(true, "offset+2 (%zu) > skb_headlen() (%u)\n", 3332 offset + sizeof(__sum16), skb_headlen(skb)); 3333 goto out; 3334 } 3335 ret = skb_ensure_writable(skb, offset + sizeof(__sum16)); 3336 if (ret) 3337 goto out; 3338 3339 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0; 3340 out_set_summed: 3341 skb->ip_summed = CHECKSUM_NONE; 3342 out: 3343 return ret; 3344 } 3345 EXPORT_SYMBOL(skb_checksum_help); 3346 3347 int skb_crc32c_csum_help(struct sk_buff *skb) 3348 { 3349 __le32 crc32c_csum; 3350 int ret = 0, offset, start; 3351 3352 if (skb->ip_summed != CHECKSUM_PARTIAL) 3353 goto out; 3354 3355 if (unlikely(skb_is_gso(skb))) 3356 goto out; 3357 3358 /* Before computing a checksum, we should make sure no frag could 3359 * be modified by an external entity : checksum could be wrong. 3360 */ 3361 if (unlikely(skb_has_shared_frag(skb))) { 3362 ret = __skb_linearize(skb); 3363 if (ret) 3364 goto out; 3365 } 3366 start = skb_checksum_start_offset(skb); 3367 offset = start + offsetof(struct sctphdr, checksum); 3368 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { 3369 ret = -EINVAL; 3370 goto out; 3371 } 3372 3373 ret = skb_ensure_writable(skb, offset + sizeof(__le32)); 3374 if (ret) 3375 goto out; 3376 3377 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start, 3378 skb->len - start, ~(__u32)0, 3379 crc32c_csum_stub)); 3380 *(__le32 *)(skb->data + offset) = crc32c_csum; 3381 skb_reset_csum_not_inet(skb); 3382 out: 3383 return ret; 3384 } 3385 3386 __be16 skb_network_protocol(struct sk_buff *skb, int *depth) 3387 { 3388 __be16 type = skb->protocol; 3389 3390 /* Tunnel gso handlers can set protocol to ethernet. */ 3391 if (type == htons(ETH_P_TEB)) { 3392 struct ethhdr *eth; 3393 3394 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) 3395 return 0; 3396 3397 eth = (struct ethhdr *)skb->data; 3398 type = eth->h_proto; 3399 } 3400 3401 return vlan_get_protocol_and_depth(skb, type, depth); 3402 } 3403 3404 3405 /* Take action when hardware reception checksum errors are detected. */ 3406 #ifdef CONFIG_BUG 3407 static void do_netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3408 { 3409 netdev_err(dev, "hw csum failure\n"); 3410 skb_dump(KERN_ERR, skb, true); 3411 dump_stack(); 3412 } 3413 3414 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3415 { 3416 DO_ONCE_LITE(do_netdev_rx_csum_fault, dev, skb); 3417 } 3418 EXPORT_SYMBOL(netdev_rx_csum_fault); 3419 #endif 3420 3421 /* XXX: check that highmem exists at all on the given machine. */ 3422 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) 3423 { 3424 #ifdef CONFIG_HIGHMEM 3425 int i; 3426 3427 if (!(dev->features & NETIF_F_HIGHDMA)) { 3428 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 3429 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 3430 3431 if (PageHighMem(skb_frag_page(frag))) 3432 return 1; 3433 } 3434 } 3435 #endif 3436 return 0; 3437 } 3438 3439 /* If MPLS offload request, verify we are testing hardware MPLS features 3440 * instead of standard features for the netdev. 3441 */ 3442 #if IS_ENABLED(CONFIG_NET_MPLS_GSO) 3443 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3444 netdev_features_t features, 3445 __be16 type) 3446 { 3447 if (eth_p_mpls(type)) 3448 features &= skb->dev->mpls_features; 3449 3450 return features; 3451 } 3452 #else 3453 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3454 netdev_features_t features, 3455 __be16 type) 3456 { 3457 return features; 3458 } 3459 #endif 3460 3461 static netdev_features_t harmonize_features(struct sk_buff *skb, 3462 netdev_features_t features) 3463 { 3464 __be16 type; 3465 3466 type = skb_network_protocol(skb, NULL); 3467 features = net_mpls_features(skb, features, type); 3468 3469 if (skb->ip_summed != CHECKSUM_NONE && 3470 !can_checksum_protocol(features, type)) { 3471 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); 3472 } 3473 if (illegal_highdma(skb->dev, skb)) 3474 features &= ~NETIF_F_SG; 3475 3476 return features; 3477 } 3478 3479 netdev_features_t passthru_features_check(struct sk_buff *skb, 3480 struct net_device *dev, 3481 netdev_features_t features) 3482 { 3483 return features; 3484 } 3485 EXPORT_SYMBOL(passthru_features_check); 3486 3487 static netdev_features_t dflt_features_check(struct sk_buff *skb, 3488 struct net_device *dev, 3489 netdev_features_t features) 3490 { 3491 return vlan_features_check(skb, features); 3492 } 3493 3494 static netdev_features_t gso_features_check(const struct sk_buff *skb, 3495 struct net_device *dev, 3496 netdev_features_t features) 3497 { 3498 u16 gso_segs = skb_shinfo(skb)->gso_segs; 3499 3500 if (gso_segs > READ_ONCE(dev->gso_max_segs)) 3501 return features & ~NETIF_F_GSO_MASK; 3502 3503 if (unlikely(skb->len >= netif_get_gso_max_size(dev, skb))) 3504 return features & ~NETIF_F_GSO_MASK; 3505 3506 if (!skb_shinfo(skb)->gso_type) { 3507 skb_warn_bad_offload(skb); 3508 return features & ~NETIF_F_GSO_MASK; 3509 } 3510 3511 /* Support for GSO partial features requires software 3512 * intervention before we can actually process the packets 3513 * so we need to strip support for any partial features now 3514 * and we can pull them back in after we have partially 3515 * segmented the frame. 3516 */ 3517 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)) 3518 features &= ~dev->gso_partial_features; 3519 3520 /* Make sure to clear the IPv4 ID mangling feature if the 3521 * IPv4 header has the potential to be fragmented. 3522 */ 3523 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { 3524 struct iphdr *iph = skb->encapsulation ? 3525 inner_ip_hdr(skb) : ip_hdr(skb); 3526 3527 if (!(iph->frag_off & htons(IP_DF))) 3528 features &= ~NETIF_F_TSO_MANGLEID; 3529 } 3530 3531 return features; 3532 } 3533 3534 netdev_features_t netif_skb_features(struct sk_buff *skb) 3535 { 3536 struct net_device *dev = skb->dev; 3537 netdev_features_t features = dev->features; 3538 3539 if (skb_is_gso(skb)) 3540 features = gso_features_check(skb, dev, features); 3541 3542 /* If encapsulation offload request, verify we are testing 3543 * hardware encapsulation features instead of standard 3544 * features for the netdev 3545 */ 3546 if (skb->encapsulation) 3547 features &= dev->hw_enc_features; 3548 3549 if (skb_vlan_tagged(skb)) 3550 features = netdev_intersect_features(features, 3551 dev->vlan_features | 3552 NETIF_F_HW_VLAN_CTAG_TX | 3553 NETIF_F_HW_VLAN_STAG_TX); 3554 3555 if (dev->netdev_ops->ndo_features_check) 3556 features &= dev->netdev_ops->ndo_features_check(skb, dev, 3557 features); 3558 else 3559 features &= dflt_features_check(skb, dev, features); 3560 3561 return harmonize_features(skb, features); 3562 } 3563 EXPORT_SYMBOL(netif_skb_features); 3564 3565 static int xmit_one(struct sk_buff *skb, struct net_device *dev, 3566 struct netdev_queue *txq, bool more) 3567 { 3568 unsigned int len; 3569 int rc; 3570 3571 if (dev_nit_active(dev)) 3572 dev_queue_xmit_nit(skb, dev); 3573 3574 len = skb->len; 3575 trace_net_dev_start_xmit(skb, dev); 3576 rc = netdev_start_xmit(skb, dev, txq, more); 3577 trace_net_dev_xmit(skb, rc, dev, len); 3578 3579 return rc; 3580 } 3581 3582 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, 3583 struct netdev_queue *txq, int *ret) 3584 { 3585 struct sk_buff *skb = first; 3586 int rc = NETDEV_TX_OK; 3587 3588 while (skb) { 3589 struct sk_buff *next = skb->next; 3590 3591 skb_mark_not_on_list(skb); 3592 rc = xmit_one(skb, dev, txq, next != NULL); 3593 if (unlikely(!dev_xmit_complete(rc))) { 3594 skb->next = next; 3595 goto out; 3596 } 3597 3598 skb = next; 3599 if (netif_tx_queue_stopped(txq) && skb) { 3600 rc = NETDEV_TX_BUSY; 3601 break; 3602 } 3603 } 3604 3605 out: 3606 *ret = rc; 3607 return skb; 3608 } 3609 3610 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, 3611 netdev_features_t features) 3612 { 3613 if (skb_vlan_tag_present(skb) && 3614 !vlan_hw_offload_capable(features, skb->vlan_proto)) 3615 skb = __vlan_hwaccel_push_inside(skb); 3616 return skb; 3617 } 3618 3619 int skb_csum_hwoffload_help(struct sk_buff *skb, 3620 const netdev_features_t features) 3621 { 3622 if (unlikely(skb_csum_is_sctp(skb))) 3623 return !!(features & NETIF_F_SCTP_CRC) ? 0 : 3624 skb_crc32c_csum_help(skb); 3625 3626 if (features & NETIF_F_HW_CSUM) 3627 return 0; 3628 3629 if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) { 3630 if (vlan_get_protocol(skb) == htons(ETH_P_IPV6) && 3631 skb_network_header_len(skb) != sizeof(struct ipv6hdr) && 3632 !ipv6_has_hopopt_jumbo(skb)) 3633 goto sw_checksum; 3634 3635 switch (skb->csum_offset) { 3636 case offsetof(struct tcphdr, check): 3637 case offsetof(struct udphdr, check): 3638 return 0; 3639 } 3640 } 3641 3642 sw_checksum: 3643 return skb_checksum_help(skb); 3644 } 3645 EXPORT_SYMBOL(skb_csum_hwoffload_help); 3646 3647 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again) 3648 { 3649 netdev_features_t features; 3650 3651 features = netif_skb_features(skb); 3652 skb = validate_xmit_vlan(skb, features); 3653 if (unlikely(!skb)) 3654 goto out_null; 3655 3656 skb = sk_validate_xmit_skb(skb, dev); 3657 if (unlikely(!skb)) 3658 goto out_null; 3659 3660 if (netif_needs_gso(skb, features)) { 3661 struct sk_buff *segs; 3662 3663 segs = skb_gso_segment(skb, features); 3664 if (IS_ERR(segs)) { 3665 goto out_kfree_skb; 3666 } else if (segs) { 3667 consume_skb(skb); 3668 skb = segs; 3669 } 3670 } else { 3671 if (skb_needs_linearize(skb, features) && 3672 __skb_linearize(skb)) 3673 goto out_kfree_skb; 3674 3675 /* If packet is not checksummed and device does not 3676 * support checksumming for this protocol, complete 3677 * checksumming here. 3678 */ 3679 if (skb->ip_summed == CHECKSUM_PARTIAL) { 3680 if (skb->encapsulation) 3681 skb_set_inner_transport_header(skb, 3682 skb_checksum_start_offset(skb)); 3683 else 3684 skb_set_transport_header(skb, 3685 skb_checksum_start_offset(skb)); 3686 if (skb_csum_hwoffload_help(skb, features)) 3687 goto out_kfree_skb; 3688 } 3689 } 3690 3691 skb = validate_xmit_xfrm(skb, features, again); 3692 3693 return skb; 3694 3695 out_kfree_skb: 3696 kfree_skb(skb); 3697 out_null: 3698 dev_core_stats_tx_dropped_inc(dev); 3699 return NULL; 3700 } 3701 3702 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again) 3703 { 3704 struct sk_buff *next, *head = NULL, *tail; 3705 3706 for (; skb != NULL; skb = next) { 3707 next = skb->next; 3708 skb_mark_not_on_list(skb); 3709 3710 /* in case skb wont be segmented, point to itself */ 3711 skb->prev = skb; 3712 3713 skb = validate_xmit_skb(skb, dev, again); 3714 if (!skb) 3715 continue; 3716 3717 if (!head) 3718 head = skb; 3719 else 3720 tail->next = skb; 3721 /* If skb was segmented, skb->prev points to 3722 * the last segment. If not, it still contains skb. 3723 */ 3724 tail = skb->prev; 3725 } 3726 return head; 3727 } 3728 EXPORT_SYMBOL_GPL(validate_xmit_skb_list); 3729 3730 static void qdisc_pkt_len_init(struct sk_buff *skb) 3731 { 3732 const struct skb_shared_info *shinfo = skb_shinfo(skb); 3733 3734 qdisc_skb_cb(skb)->pkt_len = skb->len; 3735 3736 /* To get more precise estimation of bytes sent on wire, 3737 * we add to pkt_len the headers size of all segments 3738 */ 3739 if (shinfo->gso_size && skb_transport_header_was_set(skb)) { 3740 u16 gso_segs = shinfo->gso_segs; 3741 unsigned int hdr_len; 3742 3743 /* mac layer + network layer */ 3744 hdr_len = skb_transport_offset(skb); 3745 3746 /* + transport layer */ 3747 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { 3748 const struct tcphdr *th; 3749 struct tcphdr _tcphdr; 3750 3751 th = skb_header_pointer(skb, hdr_len, 3752 sizeof(_tcphdr), &_tcphdr); 3753 if (likely(th)) 3754 hdr_len += __tcp_hdrlen(th); 3755 } else if (shinfo->gso_type & SKB_GSO_UDP_L4) { 3756 struct udphdr _udphdr; 3757 3758 if (skb_header_pointer(skb, hdr_len, 3759 sizeof(_udphdr), &_udphdr)) 3760 hdr_len += sizeof(struct udphdr); 3761 } 3762 3763 if (unlikely(shinfo->gso_type & SKB_GSO_DODGY)) { 3764 int payload = skb->len - hdr_len; 3765 3766 /* Malicious packet. */ 3767 if (payload <= 0) 3768 return; 3769 gso_segs = DIV_ROUND_UP(payload, shinfo->gso_size); 3770 } 3771 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; 3772 } 3773 } 3774 3775 static int dev_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *q, 3776 struct sk_buff **to_free, 3777 struct netdev_queue *txq) 3778 { 3779 int rc; 3780 3781 rc = q->enqueue(skb, q, to_free) & NET_XMIT_MASK; 3782 if (rc == NET_XMIT_SUCCESS) 3783 trace_qdisc_enqueue(q, txq, skb); 3784 return rc; 3785 } 3786 3787 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, 3788 struct net_device *dev, 3789 struct netdev_queue *txq) 3790 { 3791 spinlock_t *root_lock = qdisc_lock(q); 3792 struct sk_buff *to_free = NULL; 3793 bool contended; 3794 int rc; 3795 3796 qdisc_calculate_pkt_len(skb, q); 3797 3798 if (q->flags & TCQ_F_NOLOCK) { 3799 if (q->flags & TCQ_F_CAN_BYPASS && nolock_qdisc_is_empty(q) && 3800 qdisc_run_begin(q)) { 3801 /* Retest nolock_qdisc_is_empty() within the protection 3802 * of q->seqlock to protect from racing with requeuing. 3803 */ 3804 if (unlikely(!nolock_qdisc_is_empty(q))) { 3805 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3806 __qdisc_run(q); 3807 qdisc_run_end(q); 3808 3809 goto no_lock_out; 3810 } 3811 3812 qdisc_bstats_cpu_update(q, skb); 3813 if (sch_direct_xmit(skb, q, dev, txq, NULL, true) && 3814 !nolock_qdisc_is_empty(q)) 3815 __qdisc_run(q); 3816 3817 qdisc_run_end(q); 3818 return NET_XMIT_SUCCESS; 3819 } 3820 3821 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3822 qdisc_run(q); 3823 3824 no_lock_out: 3825 if (unlikely(to_free)) 3826 kfree_skb_list_reason(to_free, 3827 SKB_DROP_REASON_QDISC_DROP); 3828 return rc; 3829 } 3830 3831 /* 3832 * Heuristic to force contended enqueues to serialize on a 3833 * separate lock before trying to get qdisc main lock. 3834 * This permits qdisc->running owner to get the lock more 3835 * often and dequeue packets faster. 3836 * On PREEMPT_RT it is possible to preempt the qdisc owner during xmit 3837 * and then other tasks will only enqueue packets. The packets will be 3838 * sent after the qdisc owner is scheduled again. To prevent this 3839 * scenario the task always serialize on the lock. 3840 */ 3841 contended = qdisc_is_running(q) || IS_ENABLED(CONFIG_PREEMPT_RT); 3842 if (unlikely(contended)) 3843 spin_lock(&q->busylock); 3844 3845 spin_lock(root_lock); 3846 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { 3847 __qdisc_drop(skb, &to_free); 3848 rc = NET_XMIT_DROP; 3849 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && 3850 qdisc_run_begin(q)) { 3851 /* 3852 * This is a work-conserving queue; there are no old skbs 3853 * waiting to be sent out; and the qdisc is not running - 3854 * xmit the skb directly. 3855 */ 3856 3857 qdisc_bstats_update(q, skb); 3858 3859 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { 3860 if (unlikely(contended)) { 3861 spin_unlock(&q->busylock); 3862 contended = false; 3863 } 3864 __qdisc_run(q); 3865 } 3866 3867 qdisc_run_end(q); 3868 rc = NET_XMIT_SUCCESS; 3869 } else { 3870 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3871 if (qdisc_run_begin(q)) { 3872 if (unlikely(contended)) { 3873 spin_unlock(&q->busylock); 3874 contended = false; 3875 } 3876 __qdisc_run(q); 3877 qdisc_run_end(q); 3878 } 3879 } 3880 spin_unlock(root_lock); 3881 if (unlikely(to_free)) 3882 kfree_skb_list_reason(to_free, SKB_DROP_REASON_QDISC_DROP); 3883 if (unlikely(contended)) 3884 spin_unlock(&q->busylock); 3885 return rc; 3886 } 3887 3888 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) 3889 static void skb_update_prio(struct sk_buff *skb) 3890 { 3891 const struct netprio_map *map; 3892 const struct sock *sk; 3893 unsigned int prioidx; 3894 3895 if (skb->priority) 3896 return; 3897 map = rcu_dereference_bh(skb->dev->priomap); 3898 if (!map) 3899 return; 3900 sk = skb_to_full_sk(skb); 3901 if (!sk) 3902 return; 3903 3904 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data); 3905 3906 if (prioidx < map->priomap_len) 3907 skb->priority = map->priomap[prioidx]; 3908 } 3909 #else 3910 #define skb_update_prio(skb) 3911 #endif 3912 3913 /** 3914 * dev_loopback_xmit - loop back @skb 3915 * @net: network namespace this loopback is happening in 3916 * @sk: sk needed to be a netfilter okfn 3917 * @skb: buffer to transmit 3918 */ 3919 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) 3920 { 3921 skb_reset_mac_header(skb); 3922 __skb_pull(skb, skb_network_offset(skb)); 3923 skb->pkt_type = PACKET_LOOPBACK; 3924 if (skb->ip_summed == CHECKSUM_NONE) 3925 skb->ip_summed = CHECKSUM_UNNECESSARY; 3926 DEBUG_NET_WARN_ON_ONCE(!skb_dst(skb)); 3927 skb_dst_force(skb); 3928 netif_rx(skb); 3929 return 0; 3930 } 3931 EXPORT_SYMBOL(dev_loopback_xmit); 3932 3933 #ifdef CONFIG_NET_EGRESS 3934 static struct netdev_queue * 3935 netdev_tx_queue_mapping(struct net_device *dev, struct sk_buff *skb) 3936 { 3937 int qm = skb_get_queue_mapping(skb); 3938 3939 return netdev_get_tx_queue(dev, netdev_cap_txqueue(dev, qm)); 3940 } 3941 3942 static bool netdev_xmit_txqueue_skipped(void) 3943 { 3944 return __this_cpu_read(softnet_data.xmit.skip_txqueue); 3945 } 3946 3947 void netdev_xmit_skip_txqueue(bool skip) 3948 { 3949 __this_cpu_write(softnet_data.xmit.skip_txqueue, skip); 3950 } 3951 EXPORT_SYMBOL_GPL(netdev_xmit_skip_txqueue); 3952 #endif /* CONFIG_NET_EGRESS */ 3953 3954 #ifdef CONFIG_NET_XGRESS 3955 static int tc_run(struct tcx_entry *entry, struct sk_buff *skb) 3956 { 3957 int ret = TC_ACT_UNSPEC; 3958 #ifdef CONFIG_NET_CLS_ACT 3959 struct mini_Qdisc *miniq = rcu_dereference_bh(entry->miniq); 3960 struct tcf_result res; 3961 3962 if (!miniq) 3963 return ret; 3964 3965 tc_skb_cb(skb)->mru = 0; 3966 tc_skb_cb(skb)->post_ct = false; 3967 3968 mini_qdisc_bstats_cpu_update(miniq, skb); 3969 ret = tcf_classify(skb, miniq->block, miniq->filter_list, &res, false); 3970 /* Only tcf related quirks below. */ 3971 switch (ret) { 3972 case TC_ACT_SHOT: 3973 mini_qdisc_qstats_cpu_drop(miniq); 3974 break; 3975 case TC_ACT_OK: 3976 case TC_ACT_RECLASSIFY: 3977 skb->tc_index = TC_H_MIN(res.classid); 3978 break; 3979 } 3980 #endif /* CONFIG_NET_CLS_ACT */ 3981 return ret; 3982 } 3983 3984 static DEFINE_STATIC_KEY_FALSE(tcx_needed_key); 3985 3986 void tcx_inc(void) 3987 { 3988 static_branch_inc(&tcx_needed_key); 3989 } 3990 3991 void tcx_dec(void) 3992 { 3993 static_branch_dec(&tcx_needed_key); 3994 } 3995 3996 static __always_inline enum tcx_action_base 3997 tcx_run(const struct bpf_mprog_entry *entry, struct sk_buff *skb, 3998 const bool needs_mac) 3999 { 4000 const struct bpf_mprog_fp *fp; 4001 const struct bpf_prog *prog; 4002 int ret = TCX_NEXT; 4003 4004 if (needs_mac) 4005 __skb_push(skb, skb->mac_len); 4006 bpf_mprog_foreach_prog(entry, fp, prog) { 4007 bpf_compute_data_pointers(skb); 4008 ret = bpf_prog_run(prog, skb); 4009 if (ret != TCX_NEXT) 4010 break; 4011 } 4012 if (needs_mac) 4013 __skb_pull(skb, skb->mac_len); 4014 return tcx_action_code(skb, ret); 4015 } 4016 4017 static __always_inline struct sk_buff * 4018 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 4019 struct net_device *orig_dev, bool *another) 4020 { 4021 struct bpf_mprog_entry *entry = rcu_dereference_bh(skb->dev->tcx_ingress); 4022 int sch_ret; 4023 4024 if (!entry) 4025 return skb; 4026 if (*pt_prev) { 4027 *ret = deliver_skb(skb, *pt_prev, orig_dev); 4028 *pt_prev = NULL; 4029 } 4030 4031 qdisc_skb_cb(skb)->pkt_len = skb->len; 4032 tcx_set_ingress(skb, true); 4033 4034 if (static_branch_unlikely(&tcx_needed_key)) { 4035 sch_ret = tcx_run(entry, skb, true); 4036 if (sch_ret != TC_ACT_UNSPEC) 4037 goto ingress_verdict; 4038 } 4039 sch_ret = tc_run(tcx_entry(entry), skb); 4040 ingress_verdict: 4041 switch (sch_ret) { 4042 case TC_ACT_REDIRECT: 4043 /* skb_mac_header check was done by BPF, so we can safely 4044 * push the L2 header back before redirecting to another 4045 * netdev. 4046 */ 4047 __skb_push(skb, skb->mac_len); 4048 if (skb_do_redirect(skb) == -EAGAIN) { 4049 __skb_pull(skb, skb->mac_len); 4050 *another = true; 4051 break; 4052 } 4053 *ret = NET_RX_SUCCESS; 4054 return NULL; 4055 case TC_ACT_SHOT: 4056 kfree_skb_reason(skb, SKB_DROP_REASON_TC_INGRESS); 4057 *ret = NET_RX_DROP; 4058 return NULL; 4059 /* used by tc_run */ 4060 case TC_ACT_STOLEN: 4061 case TC_ACT_QUEUED: 4062 case TC_ACT_TRAP: 4063 consume_skb(skb); 4064 fallthrough; 4065 case TC_ACT_CONSUMED: 4066 *ret = NET_RX_SUCCESS; 4067 return NULL; 4068 } 4069 4070 return skb; 4071 } 4072 4073 static __always_inline struct sk_buff * 4074 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 4075 { 4076 struct bpf_mprog_entry *entry = rcu_dereference_bh(dev->tcx_egress); 4077 int sch_ret; 4078 4079 if (!entry) 4080 return skb; 4081 4082 /* qdisc_skb_cb(skb)->pkt_len & tcx_set_ingress() was 4083 * already set by the caller. 4084 */ 4085 if (static_branch_unlikely(&tcx_needed_key)) { 4086 sch_ret = tcx_run(entry, skb, false); 4087 if (sch_ret != TC_ACT_UNSPEC) 4088 goto egress_verdict; 4089 } 4090 sch_ret = tc_run(tcx_entry(entry), skb); 4091 egress_verdict: 4092 switch (sch_ret) { 4093 case TC_ACT_REDIRECT: 4094 /* No need to push/pop skb's mac_header here on egress! */ 4095 skb_do_redirect(skb); 4096 *ret = NET_XMIT_SUCCESS; 4097 return NULL; 4098 case TC_ACT_SHOT: 4099 kfree_skb_reason(skb, SKB_DROP_REASON_TC_EGRESS); 4100 *ret = NET_XMIT_DROP; 4101 return NULL; 4102 /* used by tc_run */ 4103 case TC_ACT_STOLEN: 4104 case TC_ACT_QUEUED: 4105 case TC_ACT_TRAP: 4106 consume_skb(skb); 4107 fallthrough; 4108 case TC_ACT_CONSUMED: 4109 *ret = NET_XMIT_SUCCESS; 4110 return NULL; 4111 } 4112 4113 return skb; 4114 } 4115 #else 4116 static __always_inline struct sk_buff * 4117 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 4118 struct net_device *orig_dev, bool *another) 4119 { 4120 return skb; 4121 } 4122 4123 static __always_inline struct sk_buff * 4124 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 4125 { 4126 return skb; 4127 } 4128 #endif /* CONFIG_NET_XGRESS */ 4129 4130 #ifdef CONFIG_XPS 4131 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb, 4132 struct xps_dev_maps *dev_maps, unsigned int tci) 4133 { 4134 int tc = netdev_get_prio_tc_map(dev, skb->priority); 4135 struct xps_map *map; 4136 int queue_index = -1; 4137 4138 if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids) 4139 return queue_index; 4140 4141 tci *= dev_maps->num_tc; 4142 tci += tc; 4143 4144 map = rcu_dereference(dev_maps->attr_map[tci]); 4145 if (map) { 4146 if (map->len == 1) 4147 queue_index = map->queues[0]; 4148 else 4149 queue_index = map->queues[reciprocal_scale( 4150 skb_get_hash(skb), map->len)]; 4151 if (unlikely(queue_index >= dev->real_num_tx_queues)) 4152 queue_index = -1; 4153 } 4154 return queue_index; 4155 } 4156 #endif 4157 4158 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev, 4159 struct sk_buff *skb) 4160 { 4161 #ifdef CONFIG_XPS 4162 struct xps_dev_maps *dev_maps; 4163 struct sock *sk = skb->sk; 4164 int queue_index = -1; 4165 4166 if (!static_key_false(&xps_needed)) 4167 return -1; 4168 4169 rcu_read_lock(); 4170 if (!static_key_false(&xps_rxqs_needed)) 4171 goto get_cpus_map; 4172 4173 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]); 4174 if (dev_maps) { 4175 int tci = sk_rx_queue_get(sk); 4176 4177 if (tci >= 0) 4178 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4179 tci); 4180 } 4181 4182 get_cpus_map: 4183 if (queue_index < 0) { 4184 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]); 4185 if (dev_maps) { 4186 unsigned int tci = skb->sender_cpu - 1; 4187 4188 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4189 tci); 4190 } 4191 } 4192 rcu_read_unlock(); 4193 4194 return queue_index; 4195 #else 4196 return -1; 4197 #endif 4198 } 4199 4200 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb, 4201 struct net_device *sb_dev) 4202 { 4203 return 0; 4204 } 4205 EXPORT_SYMBOL(dev_pick_tx_zero); 4206 4207 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb, 4208 struct net_device *sb_dev) 4209 { 4210 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues; 4211 } 4212 EXPORT_SYMBOL(dev_pick_tx_cpu_id); 4213 4214 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb, 4215 struct net_device *sb_dev) 4216 { 4217 struct sock *sk = skb->sk; 4218 int queue_index = sk_tx_queue_get(sk); 4219 4220 sb_dev = sb_dev ? : dev; 4221 4222 if (queue_index < 0 || skb->ooo_okay || 4223 queue_index >= dev->real_num_tx_queues) { 4224 int new_index = get_xps_queue(dev, sb_dev, skb); 4225 4226 if (new_index < 0) 4227 new_index = skb_tx_hash(dev, sb_dev, skb); 4228 4229 if (queue_index != new_index && sk && 4230 sk_fullsock(sk) && 4231 rcu_access_pointer(sk->sk_dst_cache)) 4232 sk_tx_queue_set(sk, new_index); 4233 4234 queue_index = new_index; 4235 } 4236 4237 return queue_index; 4238 } 4239 EXPORT_SYMBOL(netdev_pick_tx); 4240 4241 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev, 4242 struct sk_buff *skb, 4243 struct net_device *sb_dev) 4244 { 4245 int queue_index = 0; 4246 4247 #ifdef CONFIG_XPS 4248 u32 sender_cpu = skb->sender_cpu - 1; 4249 4250 if (sender_cpu >= (u32)NR_CPUS) 4251 skb->sender_cpu = raw_smp_processor_id() + 1; 4252 #endif 4253 4254 if (dev->real_num_tx_queues != 1) { 4255 const struct net_device_ops *ops = dev->netdev_ops; 4256 4257 if (ops->ndo_select_queue) 4258 queue_index = ops->ndo_select_queue(dev, skb, sb_dev); 4259 else 4260 queue_index = netdev_pick_tx(dev, skb, sb_dev); 4261 4262 queue_index = netdev_cap_txqueue(dev, queue_index); 4263 } 4264 4265 skb_set_queue_mapping(skb, queue_index); 4266 return netdev_get_tx_queue(dev, queue_index); 4267 } 4268 4269 /** 4270 * __dev_queue_xmit() - transmit a buffer 4271 * @skb: buffer to transmit 4272 * @sb_dev: suboordinate device used for L2 forwarding offload 4273 * 4274 * Queue a buffer for transmission to a network device. The caller must 4275 * have set the device and priority and built the buffer before calling 4276 * this function. The function can be called from an interrupt. 4277 * 4278 * When calling this method, interrupts MUST be enabled. This is because 4279 * the BH enable code must have IRQs enabled so that it will not deadlock. 4280 * 4281 * Regardless of the return value, the skb is consumed, so it is currently 4282 * difficult to retry a send to this method. (You can bump the ref count 4283 * before sending to hold a reference for retry if you are careful.) 4284 * 4285 * Return: 4286 * * 0 - buffer successfully transmitted 4287 * * positive qdisc return code - NET_XMIT_DROP etc. 4288 * * negative errno - other errors 4289 */ 4290 int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev) 4291 { 4292 struct net_device *dev = skb->dev; 4293 struct netdev_queue *txq = NULL; 4294 struct Qdisc *q; 4295 int rc = -ENOMEM; 4296 bool again = false; 4297 4298 skb_reset_mac_header(skb); 4299 skb_assert_len(skb); 4300 4301 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) 4302 __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED); 4303 4304 /* Disable soft irqs for various locks below. Also 4305 * stops preemption for RCU. 4306 */ 4307 rcu_read_lock_bh(); 4308 4309 skb_update_prio(skb); 4310 4311 qdisc_pkt_len_init(skb); 4312 tcx_set_ingress(skb, false); 4313 #ifdef CONFIG_NET_EGRESS 4314 if (static_branch_unlikely(&egress_needed_key)) { 4315 if (nf_hook_egress_active()) { 4316 skb = nf_hook_egress(skb, &rc, dev); 4317 if (!skb) 4318 goto out; 4319 } 4320 4321 netdev_xmit_skip_txqueue(false); 4322 4323 nf_skip_egress(skb, true); 4324 skb = sch_handle_egress(skb, &rc, dev); 4325 if (!skb) 4326 goto out; 4327 nf_skip_egress(skb, false); 4328 4329 if (netdev_xmit_txqueue_skipped()) 4330 txq = netdev_tx_queue_mapping(dev, skb); 4331 } 4332 #endif 4333 /* If device/qdisc don't need skb->dst, release it right now while 4334 * its hot in this cpu cache. 4335 */ 4336 if (dev->priv_flags & IFF_XMIT_DST_RELEASE) 4337 skb_dst_drop(skb); 4338 else 4339 skb_dst_force(skb); 4340 4341 if (!txq) 4342 txq = netdev_core_pick_tx(dev, skb, sb_dev); 4343 4344 q = rcu_dereference_bh(txq->qdisc); 4345 4346 trace_net_dev_queue(skb); 4347 if (q->enqueue) { 4348 rc = __dev_xmit_skb(skb, q, dev, txq); 4349 goto out; 4350 } 4351 4352 /* The device has no queue. Common case for software devices: 4353 * loopback, all the sorts of tunnels... 4354 4355 * Really, it is unlikely that netif_tx_lock protection is necessary 4356 * here. (f.e. loopback and IP tunnels are clean ignoring statistics 4357 * counters.) 4358 * However, it is possible, that they rely on protection 4359 * made by us here. 4360 4361 * Check this and shot the lock. It is not prone from deadlocks. 4362 *Either shot noqueue qdisc, it is even simpler 8) 4363 */ 4364 if (dev->flags & IFF_UP) { 4365 int cpu = smp_processor_id(); /* ok because BHs are off */ 4366 4367 /* Other cpus might concurrently change txq->xmit_lock_owner 4368 * to -1 or to their cpu id, but not to our id. 4369 */ 4370 if (READ_ONCE(txq->xmit_lock_owner) != cpu) { 4371 if (dev_xmit_recursion()) 4372 goto recursion_alert; 4373 4374 skb = validate_xmit_skb(skb, dev, &again); 4375 if (!skb) 4376 goto out; 4377 4378 HARD_TX_LOCK(dev, txq, cpu); 4379 4380 if (!netif_xmit_stopped(txq)) { 4381 dev_xmit_recursion_inc(); 4382 skb = dev_hard_start_xmit(skb, dev, txq, &rc); 4383 dev_xmit_recursion_dec(); 4384 if (dev_xmit_complete(rc)) { 4385 HARD_TX_UNLOCK(dev, txq); 4386 goto out; 4387 } 4388 } 4389 HARD_TX_UNLOCK(dev, txq); 4390 net_crit_ratelimited("Virtual device %s asks to queue packet!\n", 4391 dev->name); 4392 } else { 4393 /* Recursion is detected! It is possible, 4394 * unfortunately 4395 */ 4396 recursion_alert: 4397 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", 4398 dev->name); 4399 } 4400 } 4401 4402 rc = -ENETDOWN; 4403 rcu_read_unlock_bh(); 4404 4405 dev_core_stats_tx_dropped_inc(dev); 4406 kfree_skb_list(skb); 4407 return rc; 4408 out: 4409 rcu_read_unlock_bh(); 4410 return rc; 4411 } 4412 EXPORT_SYMBOL(__dev_queue_xmit); 4413 4414 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id) 4415 { 4416 struct net_device *dev = skb->dev; 4417 struct sk_buff *orig_skb = skb; 4418 struct netdev_queue *txq; 4419 int ret = NETDEV_TX_BUSY; 4420 bool again = false; 4421 4422 if (unlikely(!netif_running(dev) || 4423 !netif_carrier_ok(dev))) 4424 goto drop; 4425 4426 skb = validate_xmit_skb_list(skb, dev, &again); 4427 if (skb != orig_skb) 4428 goto drop; 4429 4430 skb_set_queue_mapping(skb, queue_id); 4431 txq = skb_get_tx_queue(dev, skb); 4432 4433 local_bh_disable(); 4434 4435 dev_xmit_recursion_inc(); 4436 HARD_TX_LOCK(dev, txq, smp_processor_id()); 4437 if (!netif_xmit_frozen_or_drv_stopped(txq)) 4438 ret = netdev_start_xmit(skb, dev, txq, false); 4439 HARD_TX_UNLOCK(dev, txq); 4440 dev_xmit_recursion_dec(); 4441 4442 local_bh_enable(); 4443 return ret; 4444 drop: 4445 dev_core_stats_tx_dropped_inc(dev); 4446 kfree_skb_list(skb); 4447 return NET_XMIT_DROP; 4448 } 4449 EXPORT_SYMBOL(__dev_direct_xmit); 4450 4451 /************************************************************************* 4452 * Receiver routines 4453 *************************************************************************/ 4454 4455 int netdev_max_backlog __read_mostly = 1000; 4456 EXPORT_SYMBOL(netdev_max_backlog); 4457 4458 int netdev_tstamp_prequeue __read_mostly = 1; 4459 unsigned int sysctl_skb_defer_max __read_mostly = 64; 4460 int netdev_budget __read_mostly = 300; 4461 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */ 4462 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ; 4463 int weight_p __read_mostly = 64; /* old backlog weight */ 4464 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ 4465 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ 4466 int dev_rx_weight __read_mostly = 64; 4467 int dev_tx_weight __read_mostly = 64; 4468 4469 /* Called with irq disabled */ 4470 static inline void ____napi_schedule(struct softnet_data *sd, 4471 struct napi_struct *napi) 4472 { 4473 struct task_struct *thread; 4474 4475 lockdep_assert_irqs_disabled(); 4476 4477 if (test_bit(NAPI_STATE_THREADED, &napi->state)) { 4478 /* Paired with smp_mb__before_atomic() in 4479 * napi_enable()/dev_set_threaded(). 4480 * Use READ_ONCE() to guarantee a complete 4481 * read on napi->thread. Only call 4482 * wake_up_process() when it's not NULL. 4483 */ 4484 thread = READ_ONCE(napi->thread); 4485 if (thread) { 4486 /* Avoid doing set_bit() if the thread is in 4487 * INTERRUPTIBLE state, cause napi_thread_wait() 4488 * makes sure to proceed with napi polling 4489 * if the thread is explicitly woken from here. 4490 */ 4491 if (READ_ONCE(thread->__state) != TASK_INTERRUPTIBLE) 4492 set_bit(NAPI_STATE_SCHED_THREADED, &napi->state); 4493 wake_up_process(thread); 4494 return; 4495 } 4496 } 4497 4498 list_add_tail(&napi->poll_list, &sd->poll_list); 4499 WRITE_ONCE(napi->list_owner, smp_processor_id()); 4500 /* If not called from net_rx_action() 4501 * we have to raise NET_RX_SOFTIRQ. 4502 */ 4503 if (!sd->in_net_rx_action) 4504 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4505 } 4506 4507 #ifdef CONFIG_RPS 4508 4509 /* One global table that all flow-based protocols share. */ 4510 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; 4511 EXPORT_SYMBOL(rps_sock_flow_table); 4512 u32 rps_cpu_mask __read_mostly; 4513 EXPORT_SYMBOL(rps_cpu_mask); 4514 4515 struct static_key_false rps_needed __read_mostly; 4516 EXPORT_SYMBOL(rps_needed); 4517 struct static_key_false rfs_needed __read_mostly; 4518 EXPORT_SYMBOL(rfs_needed); 4519 4520 static struct rps_dev_flow * 4521 set_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4522 struct rps_dev_flow *rflow, u16 next_cpu) 4523 { 4524 if (next_cpu < nr_cpu_ids) { 4525 #ifdef CONFIG_RFS_ACCEL 4526 struct netdev_rx_queue *rxqueue; 4527 struct rps_dev_flow_table *flow_table; 4528 struct rps_dev_flow *old_rflow; 4529 u32 flow_id; 4530 u16 rxq_index; 4531 int rc; 4532 4533 /* Should we steer this flow to a different hardware queue? */ 4534 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || 4535 !(dev->features & NETIF_F_NTUPLE)) 4536 goto out; 4537 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); 4538 if (rxq_index == skb_get_rx_queue(skb)) 4539 goto out; 4540 4541 rxqueue = dev->_rx + rxq_index; 4542 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4543 if (!flow_table) 4544 goto out; 4545 flow_id = skb_get_hash(skb) & flow_table->mask; 4546 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, 4547 rxq_index, flow_id); 4548 if (rc < 0) 4549 goto out; 4550 old_rflow = rflow; 4551 rflow = &flow_table->flows[flow_id]; 4552 rflow->filter = rc; 4553 if (old_rflow->filter == rflow->filter) 4554 old_rflow->filter = RPS_NO_FILTER; 4555 out: 4556 #endif 4557 rflow->last_qtail = 4558 per_cpu(softnet_data, next_cpu).input_queue_head; 4559 } 4560 4561 rflow->cpu = next_cpu; 4562 return rflow; 4563 } 4564 4565 /* 4566 * get_rps_cpu is called from netif_receive_skb and returns the target 4567 * CPU from the RPS map of the receiving queue for a given skb. 4568 * rcu_read_lock must be held on entry. 4569 */ 4570 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4571 struct rps_dev_flow **rflowp) 4572 { 4573 const struct rps_sock_flow_table *sock_flow_table; 4574 struct netdev_rx_queue *rxqueue = dev->_rx; 4575 struct rps_dev_flow_table *flow_table; 4576 struct rps_map *map; 4577 int cpu = -1; 4578 u32 tcpu; 4579 u32 hash; 4580 4581 if (skb_rx_queue_recorded(skb)) { 4582 u16 index = skb_get_rx_queue(skb); 4583 4584 if (unlikely(index >= dev->real_num_rx_queues)) { 4585 WARN_ONCE(dev->real_num_rx_queues > 1, 4586 "%s received packet on queue %u, but number " 4587 "of RX queues is %u\n", 4588 dev->name, index, dev->real_num_rx_queues); 4589 goto done; 4590 } 4591 rxqueue += index; 4592 } 4593 4594 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ 4595 4596 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4597 map = rcu_dereference(rxqueue->rps_map); 4598 if (!flow_table && !map) 4599 goto done; 4600 4601 skb_reset_network_header(skb); 4602 hash = skb_get_hash(skb); 4603 if (!hash) 4604 goto done; 4605 4606 sock_flow_table = rcu_dereference(rps_sock_flow_table); 4607 if (flow_table && sock_flow_table) { 4608 struct rps_dev_flow *rflow; 4609 u32 next_cpu; 4610 u32 ident; 4611 4612 /* First check into global flow table if there is a match. 4613 * This READ_ONCE() pairs with WRITE_ONCE() from rps_record_sock_flow(). 4614 */ 4615 ident = READ_ONCE(sock_flow_table->ents[hash & sock_flow_table->mask]); 4616 if ((ident ^ hash) & ~rps_cpu_mask) 4617 goto try_rps; 4618 4619 next_cpu = ident & rps_cpu_mask; 4620 4621 /* OK, now we know there is a match, 4622 * we can look at the local (per receive queue) flow table 4623 */ 4624 rflow = &flow_table->flows[hash & flow_table->mask]; 4625 tcpu = rflow->cpu; 4626 4627 /* 4628 * If the desired CPU (where last recvmsg was done) is 4629 * different from current CPU (one in the rx-queue flow 4630 * table entry), switch if one of the following holds: 4631 * - Current CPU is unset (>= nr_cpu_ids). 4632 * - Current CPU is offline. 4633 * - The current CPU's queue tail has advanced beyond the 4634 * last packet that was enqueued using this table entry. 4635 * This guarantees that all previous packets for the flow 4636 * have been dequeued, thus preserving in order delivery. 4637 */ 4638 if (unlikely(tcpu != next_cpu) && 4639 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || 4640 ((int)(per_cpu(softnet_data, tcpu).input_queue_head - 4641 rflow->last_qtail)) >= 0)) { 4642 tcpu = next_cpu; 4643 rflow = set_rps_cpu(dev, skb, rflow, next_cpu); 4644 } 4645 4646 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { 4647 *rflowp = rflow; 4648 cpu = tcpu; 4649 goto done; 4650 } 4651 } 4652 4653 try_rps: 4654 4655 if (map) { 4656 tcpu = map->cpus[reciprocal_scale(hash, map->len)]; 4657 if (cpu_online(tcpu)) { 4658 cpu = tcpu; 4659 goto done; 4660 } 4661 } 4662 4663 done: 4664 return cpu; 4665 } 4666 4667 #ifdef CONFIG_RFS_ACCEL 4668 4669 /** 4670 * rps_may_expire_flow - check whether an RFS hardware filter may be removed 4671 * @dev: Device on which the filter was set 4672 * @rxq_index: RX queue index 4673 * @flow_id: Flow ID passed to ndo_rx_flow_steer() 4674 * @filter_id: Filter ID returned by ndo_rx_flow_steer() 4675 * 4676 * Drivers that implement ndo_rx_flow_steer() should periodically call 4677 * this function for each installed filter and remove the filters for 4678 * which it returns %true. 4679 */ 4680 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, 4681 u32 flow_id, u16 filter_id) 4682 { 4683 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; 4684 struct rps_dev_flow_table *flow_table; 4685 struct rps_dev_flow *rflow; 4686 bool expire = true; 4687 unsigned int cpu; 4688 4689 rcu_read_lock(); 4690 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4691 if (flow_table && flow_id <= flow_table->mask) { 4692 rflow = &flow_table->flows[flow_id]; 4693 cpu = READ_ONCE(rflow->cpu); 4694 if (rflow->filter == filter_id && cpu < nr_cpu_ids && 4695 ((int)(per_cpu(softnet_data, cpu).input_queue_head - 4696 rflow->last_qtail) < 4697 (int)(10 * flow_table->mask))) 4698 expire = false; 4699 } 4700 rcu_read_unlock(); 4701 return expire; 4702 } 4703 EXPORT_SYMBOL(rps_may_expire_flow); 4704 4705 #endif /* CONFIG_RFS_ACCEL */ 4706 4707 /* Called from hardirq (IPI) context */ 4708 static void rps_trigger_softirq(void *data) 4709 { 4710 struct softnet_data *sd = data; 4711 4712 ____napi_schedule(sd, &sd->backlog); 4713 sd->received_rps++; 4714 } 4715 4716 #endif /* CONFIG_RPS */ 4717 4718 /* Called from hardirq (IPI) context */ 4719 static void trigger_rx_softirq(void *data) 4720 { 4721 struct softnet_data *sd = data; 4722 4723 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4724 smp_store_release(&sd->defer_ipi_scheduled, 0); 4725 } 4726 4727 /* 4728 * After we queued a packet into sd->input_pkt_queue, 4729 * we need to make sure this queue is serviced soon. 4730 * 4731 * - If this is another cpu queue, link it to our rps_ipi_list, 4732 * and make sure we will process rps_ipi_list from net_rx_action(). 4733 * 4734 * - If this is our own queue, NAPI schedule our backlog. 4735 * Note that this also raises NET_RX_SOFTIRQ. 4736 */ 4737 static void napi_schedule_rps(struct softnet_data *sd) 4738 { 4739 struct softnet_data *mysd = this_cpu_ptr(&softnet_data); 4740 4741 #ifdef CONFIG_RPS 4742 if (sd != mysd) { 4743 sd->rps_ipi_next = mysd->rps_ipi_list; 4744 mysd->rps_ipi_list = sd; 4745 4746 /* If not called from net_rx_action() or napi_threaded_poll() 4747 * we have to raise NET_RX_SOFTIRQ. 4748 */ 4749 if (!mysd->in_net_rx_action && !mysd->in_napi_threaded_poll) 4750 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4751 return; 4752 } 4753 #endif /* CONFIG_RPS */ 4754 __napi_schedule_irqoff(&mysd->backlog); 4755 } 4756 4757 #ifdef CONFIG_NET_FLOW_LIMIT 4758 int netdev_flow_limit_table_len __read_mostly = (1 << 12); 4759 #endif 4760 4761 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) 4762 { 4763 #ifdef CONFIG_NET_FLOW_LIMIT 4764 struct sd_flow_limit *fl; 4765 struct softnet_data *sd; 4766 unsigned int old_flow, new_flow; 4767 4768 if (qlen < (READ_ONCE(netdev_max_backlog) >> 1)) 4769 return false; 4770 4771 sd = this_cpu_ptr(&softnet_data); 4772 4773 rcu_read_lock(); 4774 fl = rcu_dereference(sd->flow_limit); 4775 if (fl) { 4776 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); 4777 old_flow = fl->history[fl->history_head]; 4778 fl->history[fl->history_head] = new_flow; 4779 4780 fl->history_head++; 4781 fl->history_head &= FLOW_LIMIT_HISTORY - 1; 4782 4783 if (likely(fl->buckets[old_flow])) 4784 fl->buckets[old_flow]--; 4785 4786 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { 4787 fl->count++; 4788 rcu_read_unlock(); 4789 return true; 4790 } 4791 } 4792 rcu_read_unlock(); 4793 #endif 4794 return false; 4795 } 4796 4797 /* 4798 * enqueue_to_backlog is called to queue an skb to a per CPU backlog 4799 * queue (may be a remote CPU queue). 4800 */ 4801 static int enqueue_to_backlog(struct sk_buff *skb, int cpu, 4802 unsigned int *qtail) 4803 { 4804 enum skb_drop_reason reason; 4805 struct softnet_data *sd; 4806 unsigned long flags; 4807 unsigned int qlen; 4808 4809 reason = SKB_DROP_REASON_NOT_SPECIFIED; 4810 sd = &per_cpu(softnet_data, cpu); 4811 4812 rps_lock_irqsave(sd, &flags); 4813 if (!netif_running(skb->dev)) 4814 goto drop; 4815 qlen = skb_queue_len(&sd->input_pkt_queue); 4816 if (qlen <= READ_ONCE(netdev_max_backlog) && !skb_flow_limit(skb, qlen)) { 4817 if (qlen) { 4818 enqueue: 4819 __skb_queue_tail(&sd->input_pkt_queue, skb); 4820 input_queue_tail_incr_save(sd, qtail); 4821 rps_unlock_irq_restore(sd, &flags); 4822 return NET_RX_SUCCESS; 4823 } 4824 4825 /* Schedule NAPI for backlog device 4826 * We can use non atomic operation since we own the queue lock 4827 */ 4828 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) 4829 napi_schedule_rps(sd); 4830 goto enqueue; 4831 } 4832 reason = SKB_DROP_REASON_CPU_BACKLOG; 4833 4834 drop: 4835 sd->dropped++; 4836 rps_unlock_irq_restore(sd, &flags); 4837 4838 dev_core_stats_rx_dropped_inc(skb->dev); 4839 kfree_skb_reason(skb, reason); 4840 return NET_RX_DROP; 4841 } 4842 4843 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb) 4844 { 4845 struct net_device *dev = skb->dev; 4846 struct netdev_rx_queue *rxqueue; 4847 4848 rxqueue = dev->_rx; 4849 4850 if (skb_rx_queue_recorded(skb)) { 4851 u16 index = skb_get_rx_queue(skb); 4852 4853 if (unlikely(index >= dev->real_num_rx_queues)) { 4854 WARN_ONCE(dev->real_num_rx_queues > 1, 4855 "%s received packet on queue %u, but number " 4856 "of RX queues is %u\n", 4857 dev->name, index, dev->real_num_rx_queues); 4858 4859 return rxqueue; /* Return first rxqueue */ 4860 } 4861 rxqueue += index; 4862 } 4863 return rxqueue; 4864 } 4865 4866 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp, 4867 struct bpf_prog *xdp_prog) 4868 { 4869 void *orig_data, *orig_data_end, *hard_start; 4870 struct netdev_rx_queue *rxqueue; 4871 bool orig_bcast, orig_host; 4872 u32 mac_len, frame_sz; 4873 __be16 orig_eth_type; 4874 struct ethhdr *eth; 4875 u32 metalen, act; 4876 int off; 4877 4878 /* The XDP program wants to see the packet starting at the MAC 4879 * header. 4880 */ 4881 mac_len = skb->data - skb_mac_header(skb); 4882 hard_start = skb->data - skb_headroom(skb); 4883 4884 /* SKB "head" area always have tailroom for skb_shared_info */ 4885 frame_sz = (void *)skb_end_pointer(skb) - hard_start; 4886 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 4887 4888 rxqueue = netif_get_rxqueue(skb); 4889 xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq); 4890 xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len, 4891 skb_headlen(skb) + mac_len, true); 4892 4893 orig_data_end = xdp->data_end; 4894 orig_data = xdp->data; 4895 eth = (struct ethhdr *)xdp->data; 4896 orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr); 4897 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest); 4898 orig_eth_type = eth->h_proto; 4899 4900 act = bpf_prog_run_xdp(xdp_prog, xdp); 4901 4902 /* check if bpf_xdp_adjust_head was used */ 4903 off = xdp->data - orig_data; 4904 if (off) { 4905 if (off > 0) 4906 __skb_pull(skb, off); 4907 else if (off < 0) 4908 __skb_push(skb, -off); 4909 4910 skb->mac_header += off; 4911 skb_reset_network_header(skb); 4912 } 4913 4914 /* check if bpf_xdp_adjust_tail was used */ 4915 off = xdp->data_end - orig_data_end; 4916 if (off != 0) { 4917 skb_set_tail_pointer(skb, xdp->data_end - xdp->data); 4918 skb->len += off; /* positive on grow, negative on shrink */ 4919 } 4920 4921 /* check if XDP changed eth hdr such SKB needs update */ 4922 eth = (struct ethhdr *)xdp->data; 4923 if ((orig_eth_type != eth->h_proto) || 4924 (orig_host != ether_addr_equal_64bits(eth->h_dest, 4925 skb->dev->dev_addr)) || 4926 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) { 4927 __skb_push(skb, ETH_HLEN); 4928 skb->pkt_type = PACKET_HOST; 4929 skb->protocol = eth_type_trans(skb, skb->dev); 4930 } 4931 4932 /* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull 4933 * before calling us again on redirect path. We do not call do_redirect 4934 * as we leave that up to the caller. 4935 * 4936 * Caller is responsible for managing lifetime of skb (i.e. calling 4937 * kfree_skb in response to actions it cannot handle/XDP_DROP). 4938 */ 4939 switch (act) { 4940 case XDP_REDIRECT: 4941 case XDP_TX: 4942 __skb_push(skb, mac_len); 4943 break; 4944 case XDP_PASS: 4945 metalen = xdp->data - xdp->data_meta; 4946 if (metalen) 4947 skb_metadata_set(skb, metalen); 4948 break; 4949 } 4950 4951 return act; 4952 } 4953 4954 static u32 netif_receive_generic_xdp(struct sk_buff *skb, 4955 struct xdp_buff *xdp, 4956 struct bpf_prog *xdp_prog) 4957 { 4958 u32 act = XDP_DROP; 4959 4960 /* Reinjected packets coming from act_mirred or similar should 4961 * not get XDP generic processing. 4962 */ 4963 if (skb_is_redirected(skb)) 4964 return XDP_PASS; 4965 4966 /* XDP packets must be linear and must have sufficient headroom 4967 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also 4968 * native XDP provides, thus we need to do it here as well. 4969 */ 4970 if (skb_cloned(skb) || skb_is_nonlinear(skb) || 4971 skb_headroom(skb) < XDP_PACKET_HEADROOM) { 4972 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb); 4973 int troom = skb->tail + skb->data_len - skb->end; 4974 4975 /* In case we have to go down the path and also linearize, 4976 * then lets do the pskb_expand_head() work just once here. 4977 */ 4978 if (pskb_expand_head(skb, 4979 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0, 4980 troom > 0 ? troom + 128 : 0, GFP_ATOMIC)) 4981 goto do_drop; 4982 if (skb_linearize(skb)) 4983 goto do_drop; 4984 } 4985 4986 act = bpf_prog_run_generic_xdp(skb, xdp, xdp_prog); 4987 switch (act) { 4988 case XDP_REDIRECT: 4989 case XDP_TX: 4990 case XDP_PASS: 4991 break; 4992 default: 4993 bpf_warn_invalid_xdp_action(skb->dev, xdp_prog, act); 4994 fallthrough; 4995 case XDP_ABORTED: 4996 trace_xdp_exception(skb->dev, xdp_prog, act); 4997 fallthrough; 4998 case XDP_DROP: 4999 do_drop: 5000 kfree_skb(skb); 5001 break; 5002 } 5003 5004 return act; 5005 } 5006 5007 /* When doing generic XDP we have to bypass the qdisc layer and the 5008 * network taps in order to match in-driver-XDP behavior. This also means 5009 * that XDP packets are able to starve other packets going through a qdisc, 5010 * and DDOS attacks will be more effective. In-driver-XDP use dedicated TX 5011 * queues, so they do not have this starvation issue. 5012 */ 5013 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog) 5014 { 5015 struct net_device *dev = skb->dev; 5016 struct netdev_queue *txq; 5017 bool free_skb = true; 5018 int cpu, rc; 5019 5020 txq = netdev_core_pick_tx(dev, skb, NULL); 5021 cpu = smp_processor_id(); 5022 HARD_TX_LOCK(dev, txq, cpu); 5023 if (!netif_xmit_frozen_or_drv_stopped(txq)) { 5024 rc = netdev_start_xmit(skb, dev, txq, 0); 5025 if (dev_xmit_complete(rc)) 5026 free_skb = false; 5027 } 5028 HARD_TX_UNLOCK(dev, txq); 5029 if (free_skb) { 5030 trace_xdp_exception(dev, xdp_prog, XDP_TX); 5031 dev_core_stats_tx_dropped_inc(dev); 5032 kfree_skb(skb); 5033 } 5034 } 5035 5036 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key); 5037 5038 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb) 5039 { 5040 if (xdp_prog) { 5041 struct xdp_buff xdp; 5042 u32 act; 5043 int err; 5044 5045 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog); 5046 if (act != XDP_PASS) { 5047 switch (act) { 5048 case XDP_REDIRECT: 5049 err = xdp_do_generic_redirect(skb->dev, skb, 5050 &xdp, xdp_prog); 5051 if (err) 5052 goto out_redir; 5053 break; 5054 case XDP_TX: 5055 generic_xdp_tx(skb, xdp_prog); 5056 break; 5057 } 5058 return XDP_DROP; 5059 } 5060 } 5061 return XDP_PASS; 5062 out_redir: 5063 kfree_skb_reason(skb, SKB_DROP_REASON_XDP); 5064 return XDP_DROP; 5065 } 5066 EXPORT_SYMBOL_GPL(do_xdp_generic); 5067 5068 static int netif_rx_internal(struct sk_buff *skb) 5069 { 5070 int ret; 5071 5072 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb); 5073 5074 trace_netif_rx(skb); 5075 5076 #ifdef CONFIG_RPS 5077 if (static_branch_unlikely(&rps_needed)) { 5078 struct rps_dev_flow voidflow, *rflow = &voidflow; 5079 int cpu; 5080 5081 rcu_read_lock(); 5082 5083 cpu = get_rps_cpu(skb->dev, skb, &rflow); 5084 if (cpu < 0) 5085 cpu = smp_processor_id(); 5086 5087 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5088 5089 rcu_read_unlock(); 5090 } else 5091 #endif 5092 { 5093 unsigned int qtail; 5094 5095 ret = enqueue_to_backlog(skb, smp_processor_id(), &qtail); 5096 } 5097 return ret; 5098 } 5099 5100 /** 5101 * __netif_rx - Slightly optimized version of netif_rx 5102 * @skb: buffer to post 5103 * 5104 * This behaves as netif_rx except that it does not disable bottom halves. 5105 * As a result this function may only be invoked from the interrupt context 5106 * (either hard or soft interrupt). 5107 */ 5108 int __netif_rx(struct sk_buff *skb) 5109 { 5110 int ret; 5111 5112 lockdep_assert_once(hardirq_count() | softirq_count()); 5113 5114 trace_netif_rx_entry(skb); 5115 ret = netif_rx_internal(skb); 5116 trace_netif_rx_exit(ret); 5117 return ret; 5118 } 5119 EXPORT_SYMBOL(__netif_rx); 5120 5121 /** 5122 * netif_rx - post buffer to the network code 5123 * @skb: buffer to post 5124 * 5125 * This function receives a packet from a device driver and queues it for 5126 * the upper (protocol) levels to process via the backlog NAPI device. It 5127 * always succeeds. The buffer may be dropped during processing for 5128 * congestion control or by the protocol layers. 5129 * The network buffer is passed via the backlog NAPI device. Modern NIC 5130 * driver should use NAPI and GRO. 5131 * This function can used from interrupt and from process context. The 5132 * caller from process context must not disable interrupts before invoking 5133 * this function. 5134 * 5135 * return values: 5136 * NET_RX_SUCCESS (no congestion) 5137 * NET_RX_DROP (packet was dropped) 5138 * 5139 */ 5140 int netif_rx(struct sk_buff *skb) 5141 { 5142 bool need_bh_off = !(hardirq_count() | softirq_count()); 5143 int ret; 5144 5145 if (need_bh_off) 5146 local_bh_disable(); 5147 trace_netif_rx_entry(skb); 5148 ret = netif_rx_internal(skb); 5149 trace_netif_rx_exit(ret); 5150 if (need_bh_off) 5151 local_bh_enable(); 5152 return ret; 5153 } 5154 EXPORT_SYMBOL(netif_rx); 5155 5156 static __latent_entropy void net_tx_action(struct softirq_action *h) 5157 { 5158 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 5159 5160 if (sd->completion_queue) { 5161 struct sk_buff *clist; 5162 5163 local_irq_disable(); 5164 clist = sd->completion_queue; 5165 sd->completion_queue = NULL; 5166 local_irq_enable(); 5167 5168 while (clist) { 5169 struct sk_buff *skb = clist; 5170 5171 clist = clist->next; 5172 5173 WARN_ON(refcount_read(&skb->users)); 5174 if (likely(get_kfree_skb_cb(skb)->reason == SKB_CONSUMED)) 5175 trace_consume_skb(skb, net_tx_action); 5176 else 5177 trace_kfree_skb(skb, net_tx_action, 5178 get_kfree_skb_cb(skb)->reason); 5179 5180 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) 5181 __kfree_skb(skb); 5182 else 5183 __napi_kfree_skb(skb, 5184 get_kfree_skb_cb(skb)->reason); 5185 } 5186 } 5187 5188 if (sd->output_queue) { 5189 struct Qdisc *head; 5190 5191 local_irq_disable(); 5192 head = sd->output_queue; 5193 sd->output_queue = NULL; 5194 sd->output_queue_tailp = &sd->output_queue; 5195 local_irq_enable(); 5196 5197 rcu_read_lock(); 5198 5199 while (head) { 5200 struct Qdisc *q = head; 5201 spinlock_t *root_lock = NULL; 5202 5203 head = head->next_sched; 5204 5205 /* We need to make sure head->next_sched is read 5206 * before clearing __QDISC_STATE_SCHED 5207 */ 5208 smp_mb__before_atomic(); 5209 5210 if (!(q->flags & TCQ_F_NOLOCK)) { 5211 root_lock = qdisc_lock(q); 5212 spin_lock(root_lock); 5213 } else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, 5214 &q->state))) { 5215 /* There is a synchronize_net() between 5216 * STATE_DEACTIVATED flag being set and 5217 * qdisc_reset()/some_qdisc_is_busy() in 5218 * dev_deactivate(), so we can safely bail out 5219 * early here to avoid data race between 5220 * qdisc_deactivate() and some_qdisc_is_busy() 5221 * for lockless qdisc. 5222 */ 5223 clear_bit(__QDISC_STATE_SCHED, &q->state); 5224 continue; 5225 } 5226 5227 clear_bit(__QDISC_STATE_SCHED, &q->state); 5228 qdisc_run(q); 5229 if (root_lock) 5230 spin_unlock(root_lock); 5231 } 5232 5233 rcu_read_unlock(); 5234 } 5235 5236 xfrm_dev_backlog(sd); 5237 } 5238 5239 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) 5240 /* This hook is defined here for ATM LANE */ 5241 int (*br_fdb_test_addr_hook)(struct net_device *dev, 5242 unsigned char *addr) __read_mostly; 5243 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); 5244 #endif 5245 5246 /** 5247 * netdev_is_rx_handler_busy - check if receive handler is registered 5248 * @dev: device to check 5249 * 5250 * Check if a receive handler is already registered for a given device. 5251 * Return true if there one. 5252 * 5253 * The caller must hold the rtnl_mutex. 5254 */ 5255 bool netdev_is_rx_handler_busy(struct net_device *dev) 5256 { 5257 ASSERT_RTNL(); 5258 return dev && rtnl_dereference(dev->rx_handler); 5259 } 5260 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); 5261 5262 /** 5263 * netdev_rx_handler_register - register receive handler 5264 * @dev: device to register a handler for 5265 * @rx_handler: receive handler to register 5266 * @rx_handler_data: data pointer that is used by rx handler 5267 * 5268 * Register a receive handler for a device. This handler will then be 5269 * called from __netif_receive_skb. A negative errno code is returned 5270 * on a failure. 5271 * 5272 * The caller must hold the rtnl_mutex. 5273 * 5274 * For a general description of rx_handler, see enum rx_handler_result. 5275 */ 5276 int netdev_rx_handler_register(struct net_device *dev, 5277 rx_handler_func_t *rx_handler, 5278 void *rx_handler_data) 5279 { 5280 if (netdev_is_rx_handler_busy(dev)) 5281 return -EBUSY; 5282 5283 if (dev->priv_flags & IFF_NO_RX_HANDLER) 5284 return -EINVAL; 5285 5286 /* Note: rx_handler_data must be set before rx_handler */ 5287 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); 5288 rcu_assign_pointer(dev->rx_handler, rx_handler); 5289 5290 return 0; 5291 } 5292 EXPORT_SYMBOL_GPL(netdev_rx_handler_register); 5293 5294 /** 5295 * netdev_rx_handler_unregister - unregister receive handler 5296 * @dev: device to unregister a handler from 5297 * 5298 * Unregister a receive handler from a device. 5299 * 5300 * The caller must hold the rtnl_mutex. 5301 */ 5302 void netdev_rx_handler_unregister(struct net_device *dev) 5303 { 5304 5305 ASSERT_RTNL(); 5306 RCU_INIT_POINTER(dev->rx_handler, NULL); 5307 /* a reader seeing a non NULL rx_handler in a rcu_read_lock() 5308 * section has a guarantee to see a non NULL rx_handler_data 5309 * as well. 5310 */ 5311 synchronize_net(); 5312 RCU_INIT_POINTER(dev->rx_handler_data, NULL); 5313 } 5314 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); 5315 5316 /* 5317 * Limit the use of PFMEMALLOC reserves to those protocols that implement 5318 * the special handling of PFMEMALLOC skbs. 5319 */ 5320 static bool skb_pfmemalloc_protocol(struct sk_buff *skb) 5321 { 5322 switch (skb->protocol) { 5323 case htons(ETH_P_ARP): 5324 case htons(ETH_P_IP): 5325 case htons(ETH_P_IPV6): 5326 case htons(ETH_P_8021Q): 5327 case htons(ETH_P_8021AD): 5328 return true; 5329 default: 5330 return false; 5331 } 5332 } 5333 5334 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, 5335 int *ret, struct net_device *orig_dev) 5336 { 5337 if (nf_hook_ingress_active(skb)) { 5338 int ingress_retval; 5339 5340 if (*pt_prev) { 5341 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5342 *pt_prev = NULL; 5343 } 5344 5345 rcu_read_lock(); 5346 ingress_retval = nf_hook_ingress(skb); 5347 rcu_read_unlock(); 5348 return ingress_retval; 5349 } 5350 return 0; 5351 } 5352 5353 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc, 5354 struct packet_type **ppt_prev) 5355 { 5356 struct packet_type *ptype, *pt_prev; 5357 rx_handler_func_t *rx_handler; 5358 struct sk_buff *skb = *pskb; 5359 struct net_device *orig_dev; 5360 bool deliver_exact = false; 5361 int ret = NET_RX_DROP; 5362 __be16 type; 5363 5364 net_timestamp_check(!READ_ONCE(netdev_tstamp_prequeue), skb); 5365 5366 trace_netif_receive_skb(skb); 5367 5368 orig_dev = skb->dev; 5369 5370 skb_reset_network_header(skb); 5371 if (!skb_transport_header_was_set(skb)) 5372 skb_reset_transport_header(skb); 5373 skb_reset_mac_len(skb); 5374 5375 pt_prev = NULL; 5376 5377 another_round: 5378 skb->skb_iif = skb->dev->ifindex; 5379 5380 __this_cpu_inc(softnet_data.processed); 5381 5382 if (static_branch_unlikely(&generic_xdp_needed_key)) { 5383 int ret2; 5384 5385 migrate_disable(); 5386 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb); 5387 migrate_enable(); 5388 5389 if (ret2 != XDP_PASS) { 5390 ret = NET_RX_DROP; 5391 goto out; 5392 } 5393 } 5394 5395 if (eth_type_vlan(skb->protocol)) { 5396 skb = skb_vlan_untag(skb); 5397 if (unlikely(!skb)) 5398 goto out; 5399 } 5400 5401 if (skb_skip_tc_classify(skb)) 5402 goto skip_classify; 5403 5404 if (pfmemalloc) 5405 goto skip_taps; 5406 5407 list_for_each_entry_rcu(ptype, &ptype_all, list) { 5408 if (pt_prev) 5409 ret = deliver_skb(skb, pt_prev, orig_dev); 5410 pt_prev = ptype; 5411 } 5412 5413 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { 5414 if (pt_prev) 5415 ret = deliver_skb(skb, pt_prev, orig_dev); 5416 pt_prev = ptype; 5417 } 5418 5419 skip_taps: 5420 #ifdef CONFIG_NET_INGRESS 5421 if (static_branch_unlikely(&ingress_needed_key)) { 5422 bool another = false; 5423 5424 nf_skip_egress(skb, true); 5425 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev, 5426 &another); 5427 if (another) 5428 goto another_round; 5429 if (!skb) 5430 goto out; 5431 5432 nf_skip_egress(skb, false); 5433 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) 5434 goto out; 5435 } 5436 #endif 5437 skb_reset_redirect(skb); 5438 skip_classify: 5439 if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) 5440 goto drop; 5441 5442 if (skb_vlan_tag_present(skb)) { 5443 if (pt_prev) { 5444 ret = deliver_skb(skb, pt_prev, orig_dev); 5445 pt_prev = NULL; 5446 } 5447 if (vlan_do_receive(&skb)) 5448 goto another_round; 5449 else if (unlikely(!skb)) 5450 goto out; 5451 } 5452 5453 rx_handler = rcu_dereference(skb->dev->rx_handler); 5454 if (rx_handler) { 5455 if (pt_prev) { 5456 ret = deliver_skb(skb, pt_prev, orig_dev); 5457 pt_prev = NULL; 5458 } 5459 switch (rx_handler(&skb)) { 5460 case RX_HANDLER_CONSUMED: 5461 ret = NET_RX_SUCCESS; 5462 goto out; 5463 case RX_HANDLER_ANOTHER: 5464 goto another_round; 5465 case RX_HANDLER_EXACT: 5466 deliver_exact = true; 5467 break; 5468 case RX_HANDLER_PASS: 5469 break; 5470 default: 5471 BUG(); 5472 } 5473 } 5474 5475 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) { 5476 check_vlan_id: 5477 if (skb_vlan_tag_get_id(skb)) { 5478 /* Vlan id is non 0 and vlan_do_receive() above couldn't 5479 * find vlan device. 5480 */ 5481 skb->pkt_type = PACKET_OTHERHOST; 5482 } else if (eth_type_vlan(skb->protocol)) { 5483 /* Outer header is 802.1P with vlan 0, inner header is 5484 * 802.1Q or 802.1AD and vlan_do_receive() above could 5485 * not find vlan dev for vlan id 0. 5486 */ 5487 __vlan_hwaccel_clear_tag(skb); 5488 skb = skb_vlan_untag(skb); 5489 if (unlikely(!skb)) 5490 goto out; 5491 if (vlan_do_receive(&skb)) 5492 /* After stripping off 802.1P header with vlan 0 5493 * vlan dev is found for inner header. 5494 */ 5495 goto another_round; 5496 else if (unlikely(!skb)) 5497 goto out; 5498 else 5499 /* We have stripped outer 802.1P vlan 0 header. 5500 * But could not find vlan dev. 5501 * check again for vlan id to set OTHERHOST. 5502 */ 5503 goto check_vlan_id; 5504 } 5505 /* Note: we might in the future use prio bits 5506 * and set skb->priority like in vlan_do_receive() 5507 * For the time being, just ignore Priority Code Point 5508 */ 5509 __vlan_hwaccel_clear_tag(skb); 5510 } 5511 5512 type = skb->protocol; 5513 5514 /* deliver only exact match when indicated */ 5515 if (likely(!deliver_exact)) { 5516 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5517 &ptype_base[ntohs(type) & 5518 PTYPE_HASH_MASK]); 5519 } 5520 5521 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5522 &orig_dev->ptype_specific); 5523 5524 if (unlikely(skb->dev != orig_dev)) { 5525 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5526 &skb->dev->ptype_specific); 5527 } 5528 5529 if (pt_prev) { 5530 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 5531 goto drop; 5532 *ppt_prev = pt_prev; 5533 } else { 5534 drop: 5535 if (!deliver_exact) 5536 dev_core_stats_rx_dropped_inc(skb->dev); 5537 else 5538 dev_core_stats_rx_nohandler_inc(skb->dev); 5539 kfree_skb_reason(skb, SKB_DROP_REASON_UNHANDLED_PROTO); 5540 /* Jamal, now you will not able to escape explaining 5541 * me how you were going to use this. :-) 5542 */ 5543 ret = NET_RX_DROP; 5544 } 5545 5546 out: 5547 /* The invariant here is that if *ppt_prev is not NULL 5548 * then skb should also be non-NULL. 5549 * 5550 * Apparently *ppt_prev assignment above holds this invariant due to 5551 * skb dereferencing near it. 5552 */ 5553 *pskb = skb; 5554 return ret; 5555 } 5556 5557 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc) 5558 { 5559 struct net_device *orig_dev = skb->dev; 5560 struct packet_type *pt_prev = NULL; 5561 int ret; 5562 5563 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5564 if (pt_prev) 5565 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb, 5566 skb->dev, pt_prev, orig_dev); 5567 return ret; 5568 } 5569 5570 /** 5571 * netif_receive_skb_core - special purpose version of netif_receive_skb 5572 * @skb: buffer to process 5573 * 5574 * More direct receive version of netif_receive_skb(). It should 5575 * only be used by callers that have a need to skip RPS and Generic XDP. 5576 * Caller must also take care of handling if ``(page_is_)pfmemalloc``. 5577 * 5578 * This function may only be called from softirq context and interrupts 5579 * should be enabled. 5580 * 5581 * Return values (usually ignored): 5582 * NET_RX_SUCCESS: no congestion 5583 * NET_RX_DROP: packet was dropped 5584 */ 5585 int netif_receive_skb_core(struct sk_buff *skb) 5586 { 5587 int ret; 5588 5589 rcu_read_lock(); 5590 ret = __netif_receive_skb_one_core(skb, false); 5591 rcu_read_unlock(); 5592 5593 return ret; 5594 } 5595 EXPORT_SYMBOL(netif_receive_skb_core); 5596 5597 static inline void __netif_receive_skb_list_ptype(struct list_head *head, 5598 struct packet_type *pt_prev, 5599 struct net_device *orig_dev) 5600 { 5601 struct sk_buff *skb, *next; 5602 5603 if (!pt_prev) 5604 return; 5605 if (list_empty(head)) 5606 return; 5607 if (pt_prev->list_func != NULL) 5608 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv, 5609 ip_list_rcv, head, pt_prev, orig_dev); 5610 else 5611 list_for_each_entry_safe(skb, next, head, list) { 5612 skb_list_del_init(skb); 5613 pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 5614 } 5615 } 5616 5617 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc) 5618 { 5619 /* Fast-path assumptions: 5620 * - There is no RX handler. 5621 * - Only one packet_type matches. 5622 * If either of these fails, we will end up doing some per-packet 5623 * processing in-line, then handling the 'last ptype' for the whole 5624 * sublist. This can't cause out-of-order delivery to any single ptype, 5625 * because the 'last ptype' must be constant across the sublist, and all 5626 * other ptypes are handled per-packet. 5627 */ 5628 /* Current (common) ptype of sublist */ 5629 struct packet_type *pt_curr = NULL; 5630 /* Current (common) orig_dev of sublist */ 5631 struct net_device *od_curr = NULL; 5632 struct list_head sublist; 5633 struct sk_buff *skb, *next; 5634 5635 INIT_LIST_HEAD(&sublist); 5636 list_for_each_entry_safe(skb, next, head, list) { 5637 struct net_device *orig_dev = skb->dev; 5638 struct packet_type *pt_prev = NULL; 5639 5640 skb_list_del_init(skb); 5641 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5642 if (!pt_prev) 5643 continue; 5644 if (pt_curr != pt_prev || od_curr != orig_dev) { 5645 /* dispatch old sublist */ 5646 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5647 /* start new sublist */ 5648 INIT_LIST_HEAD(&sublist); 5649 pt_curr = pt_prev; 5650 od_curr = orig_dev; 5651 } 5652 list_add_tail(&skb->list, &sublist); 5653 } 5654 5655 /* dispatch final sublist */ 5656 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5657 } 5658 5659 static int __netif_receive_skb(struct sk_buff *skb) 5660 { 5661 int ret; 5662 5663 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { 5664 unsigned int noreclaim_flag; 5665 5666 /* 5667 * PFMEMALLOC skbs are special, they should 5668 * - be delivered to SOCK_MEMALLOC sockets only 5669 * - stay away from userspace 5670 * - have bounded memory usage 5671 * 5672 * Use PF_MEMALLOC as this saves us from propagating the allocation 5673 * context down to all allocation sites. 5674 */ 5675 noreclaim_flag = memalloc_noreclaim_save(); 5676 ret = __netif_receive_skb_one_core(skb, true); 5677 memalloc_noreclaim_restore(noreclaim_flag); 5678 } else 5679 ret = __netif_receive_skb_one_core(skb, false); 5680 5681 return ret; 5682 } 5683 5684 static void __netif_receive_skb_list(struct list_head *head) 5685 { 5686 unsigned long noreclaim_flag = 0; 5687 struct sk_buff *skb, *next; 5688 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */ 5689 5690 list_for_each_entry_safe(skb, next, head, list) { 5691 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) { 5692 struct list_head sublist; 5693 5694 /* Handle the previous sublist */ 5695 list_cut_before(&sublist, head, &skb->list); 5696 if (!list_empty(&sublist)) 5697 __netif_receive_skb_list_core(&sublist, pfmemalloc); 5698 pfmemalloc = !pfmemalloc; 5699 /* See comments in __netif_receive_skb */ 5700 if (pfmemalloc) 5701 noreclaim_flag = memalloc_noreclaim_save(); 5702 else 5703 memalloc_noreclaim_restore(noreclaim_flag); 5704 } 5705 } 5706 /* Handle the remaining sublist */ 5707 if (!list_empty(head)) 5708 __netif_receive_skb_list_core(head, pfmemalloc); 5709 /* Restore pflags */ 5710 if (pfmemalloc) 5711 memalloc_noreclaim_restore(noreclaim_flag); 5712 } 5713 5714 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp) 5715 { 5716 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog); 5717 struct bpf_prog *new = xdp->prog; 5718 int ret = 0; 5719 5720 switch (xdp->command) { 5721 case XDP_SETUP_PROG: 5722 rcu_assign_pointer(dev->xdp_prog, new); 5723 if (old) 5724 bpf_prog_put(old); 5725 5726 if (old && !new) { 5727 static_branch_dec(&generic_xdp_needed_key); 5728 } else if (new && !old) { 5729 static_branch_inc(&generic_xdp_needed_key); 5730 dev_disable_lro(dev); 5731 dev_disable_gro_hw(dev); 5732 } 5733 break; 5734 5735 default: 5736 ret = -EINVAL; 5737 break; 5738 } 5739 5740 return ret; 5741 } 5742 5743 static int netif_receive_skb_internal(struct sk_buff *skb) 5744 { 5745 int ret; 5746 5747 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb); 5748 5749 if (skb_defer_rx_timestamp(skb)) 5750 return NET_RX_SUCCESS; 5751 5752 rcu_read_lock(); 5753 #ifdef CONFIG_RPS 5754 if (static_branch_unlikely(&rps_needed)) { 5755 struct rps_dev_flow voidflow, *rflow = &voidflow; 5756 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5757 5758 if (cpu >= 0) { 5759 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5760 rcu_read_unlock(); 5761 return ret; 5762 } 5763 } 5764 #endif 5765 ret = __netif_receive_skb(skb); 5766 rcu_read_unlock(); 5767 return ret; 5768 } 5769 5770 void netif_receive_skb_list_internal(struct list_head *head) 5771 { 5772 struct sk_buff *skb, *next; 5773 struct list_head sublist; 5774 5775 INIT_LIST_HEAD(&sublist); 5776 list_for_each_entry_safe(skb, next, head, list) { 5777 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb); 5778 skb_list_del_init(skb); 5779 if (!skb_defer_rx_timestamp(skb)) 5780 list_add_tail(&skb->list, &sublist); 5781 } 5782 list_splice_init(&sublist, head); 5783 5784 rcu_read_lock(); 5785 #ifdef CONFIG_RPS 5786 if (static_branch_unlikely(&rps_needed)) { 5787 list_for_each_entry_safe(skb, next, head, list) { 5788 struct rps_dev_flow voidflow, *rflow = &voidflow; 5789 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5790 5791 if (cpu >= 0) { 5792 /* Will be handled, remove from list */ 5793 skb_list_del_init(skb); 5794 enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5795 } 5796 } 5797 } 5798 #endif 5799 __netif_receive_skb_list(head); 5800 rcu_read_unlock(); 5801 } 5802 5803 /** 5804 * netif_receive_skb - process receive buffer from network 5805 * @skb: buffer to process 5806 * 5807 * netif_receive_skb() is the main receive data processing function. 5808 * It always succeeds. The buffer may be dropped during processing 5809 * for congestion control or by the protocol layers. 5810 * 5811 * This function may only be called from softirq context and interrupts 5812 * should be enabled. 5813 * 5814 * Return values (usually ignored): 5815 * NET_RX_SUCCESS: no congestion 5816 * NET_RX_DROP: packet was dropped 5817 */ 5818 int netif_receive_skb(struct sk_buff *skb) 5819 { 5820 int ret; 5821 5822 trace_netif_receive_skb_entry(skb); 5823 5824 ret = netif_receive_skb_internal(skb); 5825 trace_netif_receive_skb_exit(ret); 5826 5827 return ret; 5828 } 5829 EXPORT_SYMBOL(netif_receive_skb); 5830 5831 /** 5832 * netif_receive_skb_list - process many receive buffers from network 5833 * @head: list of skbs to process. 5834 * 5835 * Since return value of netif_receive_skb() is normally ignored, and 5836 * wouldn't be meaningful for a list, this function returns void. 5837 * 5838 * This function may only be called from softirq context and interrupts 5839 * should be enabled. 5840 */ 5841 void netif_receive_skb_list(struct list_head *head) 5842 { 5843 struct sk_buff *skb; 5844 5845 if (list_empty(head)) 5846 return; 5847 if (trace_netif_receive_skb_list_entry_enabled()) { 5848 list_for_each_entry(skb, head, list) 5849 trace_netif_receive_skb_list_entry(skb); 5850 } 5851 netif_receive_skb_list_internal(head); 5852 trace_netif_receive_skb_list_exit(0); 5853 } 5854 EXPORT_SYMBOL(netif_receive_skb_list); 5855 5856 static DEFINE_PER_CPU(struct work_struct, flush_works); 5857 5858 /* Network device is going away, flush any packets still pending */ 5859 static void flush_backlog(struct work_struct *work) 5860 { 5861 struct sk_buff *skb, *tmp; 5862 struct softnet_data *sd; 5863 5864 local_bh_disable(); 5865 sd = this_cpu_ptr(&softnet_data); 5866 5867 rps_lock_irq_disable(sd); 5868 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { 5869 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5870 __skb_unlink(skb, &sd->input_pkt_queue); 5871 dev_kfree_skb_irq(skb); 5872 input_queue_head_incr(sd); 5873 } 5874 } 5875 rps_unlock_irq_enable(sd); 5876 5877 skb_queue_walk_safe(&sd->process_queue, skb, tmp) { 5878 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5879 __skb_unlink(skb, &sd->process_queue); 5880 kfree_skb(skb); 5881 input_queue_head_incr(sd); 5882 } 5883 } 5884 local_bh_enable(); 5885 } 5886 5887 static bool flush_required(int cpu) 5888 { 5889 #if IS_ENABLED(CONFIG_RPS) 5890 struct softnet_data *sd = &per_cpu(softnet_data, cpu); 5891 bool do_flush; 5892 5893 rps_lock_irq_disable(sd); 5894 5895 /* as insertion into process_queue happens with the rps lock held, 5896 * process_queue access may race only with dequeue 5897 */ 5898 do_flush = !skb_queue_empty(&sd->input_pkt_queue) || 5899 !skb_queue_empty_lockless(&sd->process_queue); 5900 rps_unlock_irq_enable(sd); 5901 5902 return do_flush; 5903 #endif 5904 /* without RPS we can't safely check input_pkt_queue: during a 5905 * concurrent remote skb_queue_splice() we can detect as empty both 5906 * input_pkt_queue and process_queue even if the latter could end-up 5907 * containing a lot of packets. 5908 */ 5909 return true; 5910 } 5911 5912 static void flush_all_backlogs(void) 5913 { 5914 static cpumask_t flush_cpus; 5915 unsigned int cpu; 5916 5917 /* since we are under rtnl lock protection we can use static data 5918 * for the cpumask and avoid allocating on stack the possibly 5919 * large mask 5920 */ 5921 ASSERT_RTNL(); 5922 5923 cpus_read_lock(); 5924 5925 cpumask_clear(&flush_cpus); 5926 for_each_online_cpu(cpu) { 5927 if (flush_required(cpu)) { 5928 queue_work_on(cpu, system_highpri_wq, 5929 per_cpu_ptr(&flush_works, cpu)); 5930 cpumask_set_cpu(cpu, &flush_cpus); 5931 } 5932 } 5933 5934 /* we can have in flight packet[s] on the cpus we are not flushing, 5935 * synchronize_net() in unregister_netdevice_many() will take care of 5936 * them 5937 */ 5938 for_each_cpu(cpu, &flush_cpus) 5939 flush_work(per_cpu_ptr(&flush_works, cpu)); 5940 5941 cpus_read_unlock(); 5942 } 5943 5944 static void net_rps_send_ipi(struct softnet_data *remsd) 5945 { 5946 #ifdef CONFIG_RPS 5947 while (remsd) { 5948 struct softnet_data *next = remsd->rps_ipi_next; 5949 5950 if (cpu_online(remsd->cpu)) 5951 smp_call_function_single_async(remsd->cpu, &remsd->csd); 5952 remsd = next; 5953 } 5954 #endif 5955 } 5956 5957 /* 5958 * net_rps_action_and_irq_enable sends any pending IPI's for rps. 5959 * Note: called with local irq disabled, but exits with local irq enabled. 5960 */ 5961 static void net_rps_action_and_irq_enable(struct softnet_data *sd) 5962 { 5963 #ifdef CONFIG_RPS 5964 struct softnet_data *remsd = sd->rps_ipi_list; 5965 5966 if (remsd) { 5967 sd->rps_ipi_list = NULL; 5968 5969 local_irq_enable(); 5970 5971 /* Send pending IPI's to kick RPS processing on remote cpus. */ 5972 net_rps_send_ipi(remsd); 5973 } else 5974 #endif 5975 local_irq_enable(); 5976 } 5977 5978 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) 5979 { 5980 #ifdef CONFIG_RPS 5981 return sd->rps_ipi_list != NULL; 5982 #else 5983 return false; 5984 #endif 5985 } 5986 5987 static int process_backlog(struct napi_struct *napi, int quota) 5988 { 5989 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); 5990 bool again = true; 5991 int work = 0; 5992 5993 /* Check if we have pending ipi, its better to send them now, 5994 * not waiting net_rx_action() end. 5995 */ 5996 if (sd_has_rps_ipi_waiting(sd)) { 5997 local_irq_disable(); 5998 net_rps_action_and_irq_enable(sd); 5999 } 6000 6001 napi->weight = READ_ONCE(dev_rx_weight); 6002 while (again) { 6003 struct sk_buff *skb; 6004 6005 while ((skb = __skb_dequeue(&sd->process_queue))) { 6006 rcu_read_lock(); 6007 __netif_receive_skb(skb); 6008 rcu_read_unlock(); 6009 input_queue_head_incr(sd); 6010 if (++work >= quota) 6011 return work; 6012 6013 } 6014 6015 rps_lock_irq_disable(sd); 6016 if (skb_queue_empty(&sd->input_pkt_queue)) { 6017 /* 6018 * Inline a custom version of __napi_complete(). 6019 * only current cpu owns and manipulates this napi, 6020 * and NAPI_STATE_SCHED is the only possible flag set 6021 * on backlog. 6022 * We can use a plain write instead of clear_bit(), 6023 * and we dont need an smp_mb() memory barrier. 6024 */ 6025 napi->state = 0; 6026 again = false; 6027 } else { 6028 skb_queue_splice_tail_init(&sd->input_pkt_queue, 6029 &sd->process_queue); 6030 } 6031 rps_unlock_irq_enable(sd); 6032 } 6033 6034 return work; 6035 } 6036 6037 /** 6038 * __napi_schedule - schedule for receive 6039 * @n: entry to schedule 6040 * 6041 * The entry's receive function will be scheduled to run. 6042 * Consider using __napi_schedule_irqoff() if hard irqs are masked. 6043 */ 6044 void __napi_schedule(struct napi_struct *n) 6045 { 6046 unsigned long flags; 6047 6048 local_irq_save(flags); 6049 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6050 local_irq_restore(flags); 6051 } 6052 EXPORT_SYMBOL(__napi_schedule); 6053 6054 /** 6055 * napi_schedule_prep - check if napi can be scheduled 6056 * @n: napi context 6057 * 6058 * Test if NAPI routine is already running, and if not mark 6059 * it as running. This is used as a condition variable to 6060 * insure only one NAPI poll instance runs. We also make 6061 * sure there is no pending NAPI disable. 6062 */ 6063 bool napi_schedule_prep(struct napi_struct *n) 6064 { 6065 unsigned long new, val = READ_ONCE(n->state); 6066 6067 do { 6068 if (unlikely(val & NAPIF_STATE_DISABLE)) 6069 return false; 6070 new = val | NAPIF_STATE_SCHED; 6071 6072 /* Sets STATE_MISSED bit if STATE_SCHED was already set 6073 * This was suggested by Alexander Duyck, as compiler 6074 * emits better code than : 6075 * if (val & NAPIF_STATE_SCHED) 6076 * new |= NAPIF_STATE_MISSED; 6077 */ 6078 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * 6079 NAPIF_STATE_MISSED; 6080 } while (!try_cmpxchg(&n->state, &val, new)); 6081 6082 return !(val & NAPIF_STATE_SCHED); 6083 } 6084 EXPORT_SYMBOL(napi_schedule_prep); 6085 6086 /** 6087 * __napi_schedule_irqoff - schedule for receive 6088 * @n: entry to schedule 6089 * 6090 * Variant of __napi_schedule() assuming hard irqs are masked. 6091 * 6092 * On PREEMPT_RT enabled kernels this maps to __napi_schedule() 6093 * because the interrupt disabled assumption might not be true 6094 * due to force-threaded interrupts and spinlock substitution. 6095 */ 6096 void __napi_schedule_irqoff(struct napi_struct *n) 6097 { 6098 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6099 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6100 else 6101 __napi_schedule(n); 6102 } 6103 EXPORT_SYMBOL(__napi_schedule_irqoff); 6104 6105 bool napi_complete_done(struct napi_struct *n, int work_done) 6106 { 6107 unsigned long flags, val, new, timeout = 0; 6108 bool ret = true; 6109 6110 /* 6111 * 1) Don't let napi dequeue from the cpu poll list 6112 * just in case its running on a different cpu. 6113 * 2) If we are busy polling, do nothing here, we have 6114 * the guarantee we will be called later. 6115 */ 6116 if (unlikely(n->state & (NAPIF_STATE_NPSVC | 6117 NAPIF_STATE_IN_BUSY_POLL))) 6118 return false; 6119 6120 if (work_done) { 6121 if (n->gro_bitmask) 6122 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6123 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs); 6124 } 6125 if (n->defer_hard_irqs_count > 0) { 6126 n->defer_hard_irqs_count--; 6127 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6128 if (timeout) 6129 ret = false; 6130 } 6131 if (n->gro_bitmask) { 6132 /* When the NAPI instance uses a timeout and keeps postponing 6133 * it, we need to bound somehow the time packets are kept in 6134 * the GRO layer 6135 */ 6136 napi_gro_flush(n, !!timeout); 6137 } 6138 6139 gro_normal_list(n); 6140 6141 if (unlikely(!list_empty(&n->poll_list))) { 6142 /* If n->poll_list is not empty, we need to mask irqs */ 6143 local_irq_save(flags); 6144 list_del_init(&n->poll_list); 6145 local_irq_restore(flags); 6146 } 6147 WRITE_ONCE(n->list_owner, -1); 6148 6149 val = READ_ONCE(n->state); 6150 do { 6151 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); 6152 6153 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED | 6154 NAPIF_STATE_SCHED_THREADED | 6155 NAPIF_STATE_PREFER_BUSY_POLL); 6156 6157 /* If STATE_MISSED was set, leave STATE_SCHED set, 6158 * because we will call napi->poll() one more time. 6159 * This C code was suggested by Alexander Duyck to help gcc. 6160 */ 6161 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * 6162 NAPIF_STATE_SCHED; 6163 } while (!try_cmpxchg(&n->state, &val, new)); 6164 6165 if (unlikely(val & NAPIF_STATE_MISSED)) { 6166 __napi_schedule(n); 6167 return false; 6168 } 6169 6170 if (timeout) 6171 hrtimer_start(&n->timer, ns_to_ktime(timeout), 6172 HRTIMER_MODE_REL_PINNED); 6173 return ret; 6174 } 6175 EXPORT_SYMBOL(napi_complete_done); 6176 6177 /* must be called under rcu_read_lock(), as we dont take a reference */ 6178 static struct napi_struct *napi_by_id(unsigned int napi_id) 6179 { 6180 unsigned int hash = napi_id % HASH_SIZE(napi_hash); 6181 struct napi_struct *napi; 6182 6183 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) 6184 if (napi->napi_id == napi_id) 6185 return napi; 6186 6187 return NULL; 6188 } 6189 6190 #if defined(CONFIG_NET_RX_BUSY_POLL) 6191 6192 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule) 6193 { 6194 if (!skip_schedule) { 6195 gro_normal_list(napi); 6196 __napi_schedule(napi); 6197 return; 6198 } 6199 6200 if (napi->gro_bitmask) { 6201 /* flush too old packets 6202 * If HZ < 1000, flush all packets. 6203 */ 6204 napi_gro_flush(napi, HZ >= 1000); 6205 } 6206 6207 gro_normal_list(napi); 6208 clear_bit(NAPI_STATE_SCHED, &napi->state); 6209 } 6210 6211 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll, 6212 u16 budget) 6213 { 6214 bool skip_schedule = false; 6215 unsigned long timeout; 6216 int rc; 6217 6218 /* Busy polling means there is a high chance device driver hard irq 6219 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was 6220 * set in napi_schedule_prep(). 6221 * Since we are about to call napi->poll() once more, we can safely 6222 * clear NAPI_STATE_MISSED. 6223 * 6224 * Note: x86 could use a single "lock and ..." instruction 6225 * to perform these two clear_bit() 6226 */ 6227 clear_bit(NAPI_STATE_MISSED, &napi->state); 6228 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); 6229 6230 local_bh_disable(); 6231 6232 if (prefer_busy_poll) { 6233 napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs); 6234 timeout = READ_ONCE(napi->dev->gro_flush_timeout); 6235 if (napi->defer_hard_irqs_count && timeout) { 6236 hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED); 6237 skip_schedule = true; 6238 } 6239 } 6240 6241 /* All we really want here is to re-enable device interrupts. 6242 * Ideally, a new ndo_busy_poll_stop() could avoid another round. 6243 */ 6244 rc = napi->poll(napi, budget); 6245 /* We can't gro_normal_list() here, because napi->poll() might have 6246 * rearmed the napi (napi_complete_done()) in which case it could 6247 * already be running on another CPU. 6248 */ 6249 trace_napi_poll(napi, rc, budget); 6250 netpoll_poll_unlock(have_poll_lock); 6251 if (rc == budget) 6252 __busy_poll_stop(napi, skip_schedule); 6253 local_bh_enable(); 6254 } 6255 6256 void napi_busy_loop(unsigned int napi_id, 6257 bool (*loop_end)(void *, unsigned long), 6258 void *loop_end_arg, bool prefer_busy_poll, u16 budget) 6259 { 6260 unsigned long start_time = loop_end ? busy_loop_current_time() : 0; 6261 int (*napi_poll)(struct napi_struct *napi, int budget); 6262 void *have_poll_lock = NULL; 6263 struct napi_struct *napi; 6264 6265 restart: 6266 napi_poll = NULL; 6267 6268 rcu_read_lock(); 6269 6270 napi = napi_by_id(napi_id); 6271 if (!napi) 6272 goto out; 6273 6274 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6275 preempt_disable(); 6276 for (;;) { 6277 int work = 0; 6278 6279 local_bh_disable(); 6280 if (!napi_poll) { 6281 unsigned long val = READ_ONCE(napi->state); 6282 6283 /* If multiple threads are competing for this napi, 6284 * we avoid dirtying napi->state as much as we can. 6285 */ 6286 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | 6287 NAPIF_STATE_IN_BUSY_POLL)) { 6288 if (prefer_busy_poll) 6289 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6290 goto count; 6291 } 6292 if (cmpxchg(&napi->state, val, 6293 val | NAPIF_STATE_IN_BUSY_POLL | 6294 NAPIF_STATE_SCHED) != val) { 6295 if (prefer_busy_poll) 6296 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6297 goto count; 6298 } 6299 have_poll_lock = netpoll_poll_lock(napi); 6300 napi_poll = napi->poll; 6301 } 6302 work = napi_poll(napi, budget); 6303 trace_napi_poll(napi, work, budget); 6304 gro_normal_list(napi); 6305 count: 6306 if (work > 0) 6307 __NET_ADD_STATS(dev_net(napi->dev), 6308 LINUX_MIB_BUSYPOLLRXPACKETS, work); 6309 local_bh_enable(); 6310 6311 if (!loop_end || loop_end(loop_end_arg, start_time)) 6312 break; 6313 6314 if (unlikely(need_resched())) { 6315 if (napi_poll) 6316 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6317 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6318 preempt_enable(); 6319 rcu_read_unlock(); 6320 cond_resched(); 6321 if (loop_end(loop_end_arg, start_time)) 6322 return; 6323 goto restart; 6324 } 6325 cpu_relax(); 6326 } 6327 if (napi_poll) 6328 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6329 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6330 preempt_enable(); 6331 out: 6332 rcu_read_unlock(); 6333 } 6334 EXPORT_SYMBOL(napi_busy_loop); 6335 6336 #endif /* CONFIG_NET_RX_BUSY_POLL */ 6337 6338 static void napi_hash_add(struct napi_struct *napi) 6339 { 6340 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state)) 6341 return; 6342 6343 spin_lock(&napi_hash_lock); 6344 6345 /* 0..NR_CPUS range is reserved for sender_cpu use */ 6346 do { 6347 if (unlikely(++napi_gen_id < MIN_NAPI_ID)) 6348 napi_gen_id = MIN_NAPI_ID; 6349 } while (napi_by_id(napi_gen_id)); 6350 napi->napi_id = napi_gen_id; 6351 6352 hlist_add_head_rcu(&napi->napi_hash_node, 6353 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); 6354 6355 spin_unlock(&napi_hash_lock); 6356 } 6357 6358 /* Warning : caller is responsible to make sure rcu grace period 6359 * is respected before freeing memory containing @napi 6360 */ 6361 static void napi_hash_del(struct napi_struct *napi) 6362 { 6363 spin_lock(&napi_hash_lock); 6364 6365 hlist_del_init_rcu(&napi->napi_hash_node); 6366 6367 spin_unlock(&napi_hash_lock); 6368 } 6369 6370 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) 6371 { 6372 struct napi_struct *napi; 6373 6374 napi = container_of(timer, struct napi_struct, timer); 6375 6376 /* Note : we use a relaxed variant of napi_schedule_prep() not setting 6377 * NAPI_STATE_MISSED, since we do not react to a device IRQ. 6378 */ 6379 if (!napi_disable_pending(napi) && 6380 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) { 6381 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6382 __napi_schedule_irqoff(napi); 6383 } 6384 6385 return HRTIMER_NORESTART; 6386 } 6387 6388 static void init_gro_hash(struct napi_struct *napi) 6389 { 6390 int i; 6391 6392 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6393 INIT_LIST_HEAD(&napi->gro_hash[i].list); 6394 napi->gro_hash[i].count = 0; 6395 } 6396 napi->gro_bitmask = 0; 6397 } 6398 6399 int dev_set_threaded(struct net_device *dev, bool threaded) 6400 { 6401 struct napi_struct *napi; 6402 int err = 0; 6403 6404 if (dev->threaded == threaded) 6405 return 0; 6406 6407 if (threaded) { 6408 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6409 if (!napi->thread) { 6410 err = napi_kthread_create(napi); 6411 if (err) { 6412 threaded = false; 6413 break; 6414 } 6415 } 6416 } 6417 } 6418 6419 dev->threaded = threaded; 6420 6421 /* Make sure kthread is created before THREADED bit 6422 * is set. 6423 */ 6424 smp_mb__before_atomic(); 6425 6426 /* Setting/unsetting threaded mode on a napi might not immediately 6427 * take effect, if the current napi instance is actively being 6428 * polled. In this case, the switch between threaded mode and 6429 * softirq mode will happen in the next round of napi_schedule(). 6430 * This should not cause hiccups/stalls to the live traffic. 6431 */ 6432 list_for_each_entry(napi, &dev->napi_list, dev_list) 6433 assign_bit(NAPI_STATE_THREADED, &napi->state, threaded); 6434 6435 return err; 6436 } 6437 EXPORT_SYMBOL(dev_set_threaded); 6438 6439 void netif_napi_add_weight(struct net_device *dev, struct napi_struct *napi, 6440 int (*poll)(struct napi_struct *, int), int weight) 6441 { 6442 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state))) 6443 return; 6444 6445 INIT_LIST_HEAD(&napi->poll_list); 6446 INIT_HLIST_NODE(&napi->napi_hash_node); 6447 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); 6448 napi->timer.function = napi_watchdog; 6449 init_gro_hash(napi); 6450 napi->skb = NULL; 6451 INIT_LIST_HEAD(&napi->rx_list); 6452 napi->rx_count = 0; 6453 napi->poll = poll; 6454 if (weight > NAPI_POLL_WEIGHT) 6455 netdev_err_once(dev, "%s() called with weight %d\n", __func__, 6456 weight); 6457 napi->weight = weight; 6458 napi->dev = dev; 6459 #ifdef CONFIG_NETPOLL 6460 napi->poll_owner = -1; 6461 #endif 6462 napi->list_owner = -1; 6463 set_bit(NAPI_STATE_SCHED, &napi->state); 6464 set_bit(NAPI_STATE_NPSVC, &napi->state); 6465 list_add_rcu(&napi->dev_list, &dev->napi_list); 6466 napi_hash_add(napi); 6467 napi_get_frags_check(napi); 6468 /* Create kthread for this napi if dev->threaded is set. 6469 * Clear dev->threaded if kthread creation failed so that 6470 * threaded mode will not be enabled in napi_enable(). 6471 */ 6472 if (dev->threaded && napi_kthread_create(napi)) 6473 dev->threaded = 0; 6474 } 6475 EXPORT_SYMBOL(netif_napi_add_weight); 6476 6477 void napi_disable(struct napi_struct *n) 6478 { 6479 unsigned long val, new; 6480 6481 might_sleep(); 6482 set_bit(NAPI_STATE_DISABLE, &n->state); 6483 6484 val = READ_ONCE(n->state); 6485 do { 6486 while (val & (NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC)) { 6487 usleep_range(20, 200); 6488 val = READ_ONCE(n->state); 6489 } 6490 6491 new = val | NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC; 6492 new &= ~(NAPIF_STATE_THREADED | NAPIF_STATE_PREFER_BUSY_POLL); 6493 } while (!try_cmpxchg(&n->state, &val, new)); 6494 6495 hrtimer_cancel(&n->timer); 6496 6497 clear_bit(NAPI_STATE_DISABLE, &n->state); 6498 } 6499 EXPORT_SYMBOL(napi_disable); 6500 6501 /** 6502 * napi_enable - enable NAPI scheduling 6503 * @n: NAPI context 6504 * 6505 * Resume NAPI from being scheduled on this context. 6506 * Must be paired with napi_disable. 6507 */ 6508 void napi_enable(struct napi_struct *n) 6509 { 6510 unsigned long new, val = READ_ONCE(n->state); 6511 6512 do { 6513 BUG_ON(!test_bit(NAPI_STATE_SCHED, &val)); 6514 6515 new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC); 6516 if (n->dev->threaded && n->thread) 6517 new |= NAPIF_STATE_THREADED; 6518 } while (!try_cmpxchg(&n->state, &val, new)); 6519 } 6520 EXPORT_SYMBOL(napi_enable); 6521 6522 static void flush_gro_hash(struct napi_struct *napi) 6523 { 6524 int i; 6525 6526 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6527 struct sk_buff *skb, *n; 6528 6529 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list) 6530 kfree_skb(skb); 6531 napi->gro_hash[i].count = 0; 6532 } 6533 } 6534 6535 /* Must be called in process context */ 6536 void __netif_napi_del(struct napi_struct *napi) 6537 { 6538 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state)) 6539 return; 6540 6541 napi_hash_del(napi); 6542 list_del_rcu(&napi->dev_list); 6543 napi_free_frags(napi); 6544 6545 flush_gro_hash(napi); 6546 napi->gro_bitmask = 0; 6547 6548 if (napi->thread) { 6549 kthread_stop(napi->thread); 6550 napi->thread = NULL; 6551 } 6552 } 6553 EXPORT_SYMBOL(__netif_napi_del); 6554 6555 static int __napi_poll(struct napi_struct *n, bool *repoll) 6556 { 6557 int work, weight; 6558 6559 weight = n->weight; 6560 6561 /* This NAPI_STATE_SCHED test is for avoiding a race 6562 * with netpoll's poll_napi(). Only the entity which 6563 * obtains the lock and sees NAPI_STATE_SCHED set will 6564 * actually make the ->poll() call. Therefore we avoid 6565 * accidentally calling ->poll() when NAPI is not scheduled. 6566 */ 6567 work = 0; 6568 if (test_bit(NAPI_STATE_SCHED, &n->state)) { 6569 work = n->poll(n, weight); 6570 trace_napi_poll(n, work, weight); 6571 } 6572 6573 if (unlikely(work > weight)) 6574 netdev_err_once(n->dev, "NAPI poll function %pS returned %d, exceeding its budget of %d.\n", 6575 n->poll, work, weight); 6576 6577 if (likely(work < weight)) 6578 return work; 6579 6580 /* Drivers must not modify the NAPI state if they 6581 * consume the entire weight. In such cases this code 6582 * still "owns" the NAPI instance and therefore can 6583 * move the instance around on the list at-will. 6584 */ 6585 if (unlikely(napi_disable_pending(n))) { 6586 napi_complete(n); 6587 return work; 6588 } 6589 6590 /* The NAPI context has more processing work, but busy-polling 6591 * is preferred. Exit early. 6592 */ 6593 if (napi_prefer_busy_poll(n)) { 6594 if (napi_complete_done(n, work)) { 6595 /* If timeout is not set, we need to make sure 6596 * that the NAPI is re-scheduled. 6597 */ 6598 napi_schedule(n); 6599 } 6600 return work; 6601 } 6602 6603 if (n->gro_bitmask) { 6604 /* flush too old packets 6605 * If HZ < 1000, flush all packets. 6606 */ 6607 napi_gro_flush(n, HZ >= 1000); 6608 } 6609 6610 gro_normal_list(n); 6611 6612 /* Some drivers may have called napi_schedule 6613 * prior to exhausting their budget. 6614 */ 6615 if (unlikely(!list_empty(&n->poll_list))) { 6616 pr_warn_once("%s: Budget exhausted after napi rescheduled\n", 6617 n->dev ? n->dev->name : "backlog"); 6618 return work; 6619 } 6620 6621 *repoll = true; 6622 6623 return work; 6624 } 6625 6626 static int napi_poll(struct napi_struct *n, struct list_head *repoll) 6627 { 6628 bool do_repoll = false; 6629 void *have; 6630 int work; 6631 6632 list_del_init(&n->poll_list); 6633 6634 have = netpoll_poll_lock(n); 6635 6636 work = __napi_poll(n, &do_repoll); 6637 6638 if (do_repoll) 6639 list_add_tail(&n->poll_list, repoll); 6640 6641 netpoll_poll_unlock(have); 6642 6643 return work; 6644 } 6645 6646 static int napi_thread_wait(struct napi_struct *napi) 6647 { 6648 bool woken = false; 6649 6650 set_current_state(TASK_INTERRUPTIBLE); 6651 6652 while (!kthread_should_stop()) { 6653 /* Testing SCHED_THREADED bit here to make sure the current 6654 * kthread owns this napi and could poll on this napi. 6655 * Testing SCHED bit is not enough because SCHED bit might be 6656 * set by some other busy poll thread or by napi_disable(). 6657 */ 6658 if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state) || woken) { 6659 WARN_ON(!list_empty(&napi->poll_list)); 6660 __set_current_state(TASK_RUNNING); 6661 return 0; 6662 } 6663 6664 schedule(); 6665 /* woken being true indicates this thread owns this napi. */ 6666 woken = true; 6667 set_current_state(TASK_INTERRUPTIBLE); 6668 } 6669 __set_current_state(TASK_RUNNING); 6670 6671 return -1; 6672 } 6673 6674 static void skb_defer_free_flush(struct softnet_data *sd) 6675 { 6676 struct sk_buff *skb, *next; 6677 6678 /* Paired with WRITE_ONCE() in skb_attempt_defer_free() */ 6679 if (!READ_ONCE(sd->defer_list)) 6680 return; 6681 6682 spin_lock(&sd->defer_lock); 6683 skb = sd->defer_list; 6684 sd->defer_list = NULL; 6685 sd->defer_count = 0; 6686 spin_unlock(&sd->defer_lock); 6687 6688 while (skb != NULL) { 6689 next = skb->next; 6690 napi_consume_skb(skb, 1); 6691 skb = next; 6692 } 6693 } 6694 6695 static int napi_threaded_poll(void *data) 6696 { 6697 struct napi_struct *napi = data; 6698 struct softnet_data *sd; 6699 void *have; 6700 6701 while (!napi_thread_wait(napi)) { 6702 unsigned long last_qs = jiffies; 6703 6704 for (;;) { 6705 bool repoll = false; 6706 6707 local_bh_disable(); 6708 sd = this_cpu_ptr(&softnet_data); 6709 sd->in_napi_threaded_poll = true; 6710 6711 have = netpoll_poll_lock(napi); 6712 __napi_poll(napi, &repoll); 6713 netpoll_poll_unlock(have); 6714 6715 sd->in_napi_threaded_poll = false; 6716 barrier(); 6717 6718 if (sd_has_rps_ipi_waiting(sd)) { 6719 local_irq_disable(); 6720 net_rps_action_and_irq_enable(sd); 6721 } 6722 skb_defer_free_flush(sd); 6723 local_bh_enable(); 6724 6725 if (!repoll) 6726 break; 6727 6728 rcu_softirq_qs_periodic(last_qs); 6729 cond_resched(); 6730 } 6731 } 6732 return 0; 6733 } 6734 6735 static __latent_entropy void net_rx_action(struct softirq_action *h) 6736 { 6737 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 6738 unsigned long time_limit = jiffies + 6739 usecs_to_jiffies(READ_ONCE(netdev_budget_usecs)); 6740 int budget = READ_ONCE(netdev_budget); 6741 LIST_HEAD(list); 6742 LIST_HEAD(repoll); 6743 6744 start: 6745 sd->in_net_rx_action = true; 6746 local_irq_disable(); 6747 list_splice_init(&sd->poll_list, &list); 6748 local_irq_enable(); 6749 6750 for (;;) { 6751 struct napi_struct *n; 6752 6753 skb_defer_free_flush(sd); 6754 6755 if (list_empty(&list)) { 6756 if (list_empty(&repoll)) { 6757 sd->in_net_rx_action = false; 6758 barrier(); 6759 /* We need to check if ____napi_schedule() 6760 * had refilled poll_list while 6761 * sd->in_net_rx_action was true. 6762 */ 6763 if (!list_empty(&sd->poll_list)) 6764 goto start; 6765 if (!sd_has_rps_ipi_waiting(sd)) 6766 goto end; 6767 } 6768 break; 6769 } 6770 6771 n = list_first_entry(&list, struct napi_struct, poll_list); 6772 budget -= napi_poll(n, &repoll); 6773 6774 /* If softirq window is exhausted then punt. 6775 * Allow this to run for 2 jiffies since which will allow 6776 * an average latency of 1.5/HZ. 6777 */ 6778 if (unlikely(budget <= 0 || 6779 time_after_eq(jiffies, time_limit))) { 6780 sd->time_squeeze++; 6781 break; 6782 } 6783 } 6784 6785 local_irq_disable(); 6786 6787 list_splice_tail_init(&sd->poll_list, &list); 6788 list_splice_tail(&repoll, &list); 6789 list_splice(&list, &sd->poll_list); 6790 if (!list_empty(&sd->poll_list)) 6791 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 6792 else 6793 sd->in_net_rx_action = false; 6794 6795 net_rps_action_and_irq_enable(sd); 6796 end:; 6797 } 6798 6799 struct netdev_adjacent { 6800 struct net_device *dev; 6801 netdevice_tracker dev_tracker; 6802 6803 /* upper master flag, there can only be one master device per list */ 6804 bool master; 6805 6806 /* lookup ignore flag */ 6807 bool ignore; 6808 6809 /* counter for the number of times this device was added to us */ 6810 u16 ref_nr; 6811 6812 /* private field for the users */ 6813 void *private; 6814 6815 struct list_head list; 6816 struct rcu_head rcu; 6817 }; 6818 6819 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, 6820 struct list_head *adj_list) 6821 { 6822 struct netdev_adjacent *adj; 6823 6824 list_for_each_entry(adj, adj_list, list) { 6825 if (adj->dev == adj_dev) 6826 return adj; 6827 } 6828 return NULL; 6829 } 6830 6831 static int ____netdev_has_upper_dev(struct net_device *upper_dev, 6832 struct netdev_nested_priv *priv) 6833 { 6834 struct net_device *dev = (struct net_device *)priv->data; 6835 6836 return upper_dev == dev; 6837 } 6838 6839 /** 6840 * netdev_has_upper_dev - Check if device is linked to an upper device 6841 * @dev: device 6842 * @upper_dev: upper device to check 6843 * 6844 * Find out if a device is linked to specified upper device and return true 6845 * in case it is. Note that this checks only immediate upper device, 6846 * not through a complete stack of devices. The caller must hold the RTNL lock. 6847 */ 6848 bool netdev_has_upper_dev(struct net_device *dev, 6849 struct net_device *upper_dev) 6850 { 6851 struct netdev_nested_priv priv = { 6852 .data = (void *)upper_dev, 6853 }; 6854 6855 ASSERT_RTNL(); 6856 6857 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 6858 &priv); 6859 } 6860 EXPORT_SYMBOL(netdev_has_upper_dev); 6861 6862 /** 6863 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device 6864 * @dev: device 6865 * @upper_dev: upper device to check 6866 * 6867 * Find out if a device is linked to specified upper device and return true 6868 * in case it is. Note that this checks the entire upper device chain. 6869 * The caller must hold rcu lock. 6870 */ 6871 6872 bool netdev_has_upper_dev_all_rcu(struct net_device *dev, 6873 struct net_device *upper_dev) 6874 { 6875 struct netdev_nested_priv priv = { 6876 .data = (void *)upper_dev, 6877 }; 6878 6879 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 6880 &priv); 6881 } 6882 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); 6883 6884 /** 6885 * netdev_has_any_upper_dev - Check if device is linked to some device 6886 * @dev: device 6887 * 6888 * Find out if a device is linked to an upper device and return true in case 6889 * it is. The caller must hold the RTNL lock. 6890 */ 6891 bool netdev_has_any_upper_dev(struct net_device *dev) 6892 { 6893 ASSERT_RTNL(); 6894 6895 return !list_empty(&dev->adj_list.upper); 6896 } 6897 EXPORT_SYMBOL(netdev_has_any_upper_dev); 6898 6899 /** 6900 * netdev_master_upper_dev_get - Get master upper device 6901 * @dev: device 6902 * 6903 * Find a master upper device and return pointer to it or NULL in case 6904 * it's not there. The caller must hold the RTNL lock. 6905 */ 6906 struct net_device *netdev_master_upper_dev_get(struct net_device *dev) 6907 { 6908 struct netdev_adjacent *upper; 6909 6910 ASSERT_RTNL(); 6911 6912 if (list_empty(&dev->adj_list.upper)) 6913 return NULL; 6914 6915 upper = list_first_entry(&dev->adj_list.upper, 6916 struct netdev_adjacent, list); 6917 if (likely(upper->master)) 6918 return upper->dev; 6919 return NULL; 6920 } 6921 EXPORT_SYMBOL(netdev_master_upper_dev_get); 6922 6923 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev) 6924 { 6925 struct netdev_adjacent *upper; 6926 6927 ASSERT_RTNL(); 6928 6929 if (list_empty(&dev->adj_list.upper)) 6930 return NULL; 6931 6932 upper = list_first_entry(&dev->adj_list.upper, 6933 struct netdev_adjacent, list); 6934 if (likely(upper->master) && !upper->ignore) 6935 return upper->dev; 6936 return NULL; 6937 } 6938 6939 /** 6940 * netdev_has_any_lower_dev - Check if device is linked to some device 6941 * @dev: device 6942 * 6943 * Find out if a device is linked to a lower device and return true in case 6944 * it is. The caller must hold the RTNL lock. 6945 */ 6946 static bool netdev_has_any_lower_dev(struct net_device *dev) 6947 { 6948 ASSERT_RTNL(); 6949 6950 return !list_empty(&dev->adj_list.lower); 6951 } 6952 6953 void *netdev_adjacent_get_private(struct list_head *adj_list) 6954 { 6955 struct netdev_adjacent *adj; 6956 6957 adj = list_entry(adj_list, struct netdev_adjacent, list); 6958 6959 return adj->private; 6960 } 6961 EXPORT_SYMBOL(netdev_adjacent_get_private); 6962 6963 /** 6964 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list 6965 * @dev: device 6966 * @iter: list_head ** of the current position 6967 * 6968 * Gets the next device from the dev's upper list, starting from iter 6969 * position. The caller must hold RCU read lock. 6970 */ 6971 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, 6972 struct list_head **iter) 6973 { 6974 struct netdev_adjacent *upper; 6975 6976 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 6977 6978 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 6979 6980 if (&upper->list == &dev->adj_list.upper) 6981 return NULL; 6982 6983 *iter = &upper->list; 6984 6985 return upper->dev; 6986 } 6987 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); 6988 6989 static struct net_device *__netdev_next_upper_dev(struct net_device *dev, 6990 struct list_head **iter, 6991 bool *ignore) 6992 { 6993 struct netdev_adjacent *upper; 6994 6995 upper = list_entry((*iter)->next, struct netdev_adjacent, list); 6996 6997 if (&upper->list == &dev->adj_list.upper) 6998 return NULL; 6999 7000 *iter = &upper->list; 7001 *ignore = upper->ignore; 7002 7003 return upper->dev; 7004 } 7005 7006 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, 7007 struct list_head **iter) 7008 { 7009 struct netdev_adjacent *upper; 7010 7011 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 7012 7013 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7014 7015 if (&upper->list == &dev->adj_list.upper) 7016 return NULL; 7017 7018 *iter = &upper->list; 7019 7020 return upper->dev; 7021 } 7022 7023 static int __netdev_walk_all_upper_dev(struct net_device *dev, 7024 int (*fn)(struct net_device *dev, 7025 struct netdev_nested_priv *priv), 7026 struct netdev_nested_priv *priv) 7027 { 7028 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7029 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7030 int ret, cur = 0; 7031 bool ignore; 7032 7033 now = dev; 7034 iter = &dev->adj_list.upper; 7035 7036 while (1) { 7037 if (now != dev) { 7038 ret = fn(now, priv); 7039 if (ret) 7040 return ret; 7041 } 7042 7043 next = NULL; 7044 while (1) { 7045 udev = __netdev_next_upper_dev(now, &iter, &ignore); 7046 if (!udev) 7047 break; 7048 if (ignore) 7049 continue; 7050 7051 next = udev; 7052 niter = &udev->adj_list.upper; 7053 dev_stack[cur] = now; 7054 iter_stack[cur++] = iter; 7055 break; 7056 } 7057 7058 if (!next) { 7059 if (!cur) 7060 return 0; 7061 next = dev_stack[--cur]; 7062 niter = iter_stack[cur]; 7063 } 7064 7065 now = next; 7066 iter = niter; 7067 } 7068 7069 return 0; 7070 } 7071 7072 int netdev_walk_all_upper_dev_rcu(struct net_device *dev, 7073 int (*fn)(struct net_device *dev, 7074 struct netdev_nested_priv *priv), 7075 struct netdev_nested_priv *priv) 7076 { 7077 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7078 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7079 int ret, cur = 0; 7080 7081 now = dev; 7082 iter = &dev->adj_list.upper; 7083 7084 while (1) { 7085 if (now != dev) { 7086 ret = fn(now, priv); 7087 if (ret) 7088 return ret; 7089 } 7090 7091 next = NULL; 7092 while (1) { 7093 udev = netdev_next_upper_dev_rcu(now, &iter); 7094 if (!udev) 7095 break; 7096 7097 next = udev; 7098 niter = &udev->adj_list.upper; 7099 dev_stack[cur] = now; 7100 iter_stack[cur++] = iter; 7101 break; 7102 } 7103 7104 if (!next) { 7105 if (!cur) 7106 return 0; 7107 next = dev_stack[--cur]; 7108 niter = iter_stack[cur]; 7109 } 7110 7111 now = next; 7112 iter = niter; 7113 } 7114 7115 return 0; 7116 } 7117 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); 7118 7119 static bool __netdev_has_upper_dev(struct net_device *dev, 7120 struct net_device *upper_dev) 7121 { 7122 struct netdev_nested_priv priv = { 7123 .flags = 0, 7124 .data = (void *)upper_dev, 7125 }; 7126 7127 ASSERT_RTNL(); 7128 7129 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev, 7130 &priv); 7131 } 7132 7133 /** 7134 * netdev_lower_get_next_private - Get the next ->private from the 7135 * lower neighbour list 7136 * @dev: device 7137 * @iter: list_head ** of the current position 7138 * 7139 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7140 * list, starting from iter position. The caller must hold either hold the 7141 * RTNL lock or its own locking that guarantees that the neighbour lower 7142 * list will remain unchanged. 7143 */ 7144 void *netdev_lower_get_next_private(struct net_device *dev, 7145 struct list_head **iter) 7146 { 7147 struct netdev_adjacent *lower; 7148 7149 lower = list_entry(*iter, struct netdev_adjacent, list); 7150 7151 if (&lower->list == &dev->adj_list.lower) 7152 return NULL; 7153 7154 *iter = lower->list.next; 7155 7156 return lower->private; 7157 } 7158 EXPORT_SYMBOL(netdev_lower_get_next_private); 7159 7160 /** 7161 * netdev_lower_get_next_private_rcu - Get the next ->private from the 7162 * lower neighbour list, RCU 7163 * variant 7164 * @dev: device 7165 * @iter: list_head ** of the current position 7166 * 7167 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7168 * list, starting from iter position. The caller must hold RCU read lock. 7169 */ 7170 void *netdev_lower_get_next_private_rcu(struct net_device *dev, 7171 struct list_head **iter) 7172 { 7173 struct netdev_adjacent *lower; 7174 7175 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 7176 7177 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7178 7179 if (&lower->list == &dev->adj_list.lower) 7180 return NULL; 7181 7182 *iter = &lower->list; 7183 7184 return lower->private; 7185 } 7186 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); 7187 7188 /** 7189 * netdev_lower_get_next - Get the next device from the lower neighbour 7190 * list 7191 * @dev: device 7192 * @iter: list_head ** of the current position 7193 * 7194 * Gets the next netdev_adjacent from the dev's lower neighbour 7195 * list, starting from iter position. The caller must hold RTNL lock or 7196 * its own locking that guarantees that the neighbour lower 7197 * list will remain unchanged. 7198 */ 7199 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) 7200 { 7201 struct netdev_adjacent *lower; 7202 7203 lower = list_entry(*iter, struct netdev_adjacent, list); 7204 7205 if (&lower->list == &dev->adj_list.lower) 7206 return NULL; 7207 7208 *iter = lower->list.next; 7209 7210 return lower->dev; 7211 } 7212 EXPORT_SYMBOL(netdev_lower_get_next); 7213 7214 static struct net_device *netdev_next_lower_dev(struct net_device *dev, 7215 struct list_head **iter) 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 7226 return lower->dev; 7227 } 7228 7229 static struct net_device *__netdev_next_lower_dev(struct net_device *dev, 7230 struct list_head **iter, 7231 bool *ignore) 7232 { 7233 struct netdev_adjacent *lower; 7234 7235 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7236 7237 if (&lower->list == &dev->adj_list.lower) 7238 return NULL; 7239 7240 *iter = &lower->list; 7241 *ignore = lower->ignore; 7242 7243 return lower->dev; 7244 } 7245 7246 int netdev_walk_all_lower_dev(struct net_device *dev, 7247 int (*fn)(struct net_device *dev, 7248 struct netdev_nested_priv *priv), 7249 struct netdev_nested_priv *priv) 7250 { 7251 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7252 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7253 int ret, cur = 0; 7254 7255 now = dev; 7256 iter = &dev->adj_list.lower; 7257 7258 while (1) { 7259 if (now != dev) { 7260 ret = fn(now, priv); 7261 if (ret) 7262 return ret; 7263 } 7264 7265 next = NULL; 7266 while (1) { 7267 ldev = netdev_next_lower_dev(now, &iter); 7268 if (!ldev) 7269 break; 7270 7271 next = ldev; 7272 niter = &ldev->adj_list.lower; 7273 dev_stack[cur] = now; 7274 iter_stack[cur++] = iter; 7275 break; 7276 } 7277 7278 if (!next) { 7279 if (!cur) 7280 return 0; 7281 next = dev_stack[--cur]; 7282 niter = iter_stack[cur]; 7283 } 7284 7285 now = next; 7286 iter = niter; 7287 } 7288 7289 return 0; 7290 } 7291 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); 7292 7293 static int __netdev_walk_all_lower_dev(struct net_device *dev, 7294 int (*fn)(struct net_device *dev, 7295 struct netdev_nested_priv *priv), 7296 struct netdev_nested_priv *priv) 7297 { 7298 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7299 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7300 int ret, cur = 0; 7301 bool ignore; 7302 7303 now = dev; 7304 iter = &dev->adj_list.lower; 7305 7306 while (1) { 7307 if (now != dev) { 7308 ret = fn(now, priv); 7309 if (ret) 7310 return ret; 7311 } 7312 7313 next = NULL; 7314 while (1) { 7315 ldev = __netdev_next_lower_dev(now, &iter, &ignore); 7316 if (!ldev) 7317 break; 7318 if (ignore) 7319 continue; 7320 7321 next = ldev; 7322 niter = &ldev->adj_list.lower; 7323 dev_stack[cur] = now; 7324 iter_stack[cur++] = iter; 7325 break; 7326 } 7327 7328 if (!next) { 7329 if (!cur) 7330 return 0; 7331 next = dev_stack[--cur]; 7332 niter = iter_stack[cur]; 7333 } 7334 7335 now = next; 7336 iter = niter; 7337 } 7338 7339 return 0; 7340 } 7341 7342 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, 7343 struct list_head **iter) 7344 { 7345 struct netdev_adjacent *lower; 7346 7347 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7348 if (&lower->list == &dev->adj_list.lower) 7349 return NULL; 7350 7351 *iter = &lower->list; 7352 7353 return lower->dev; 7354 } 7355 EXPORT_SYMBOL(netdev_next_lower_dev_rcu); 7356 7357 static u8 __netdev_upper_depth(struct net_device *dev) 7358 { 7359 struct net_device *udev; 7360 struct list_head *iter; 7361 u8 max_depth = 0; 7362 bool ignore; 7363 7364 for (iter = &dev->adj_list.upper, 7365 udev = __netdev_next_upper_dev(dev, &iter, &ignore); 7366 udev; 7367 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) { 7368 if (ignore) 7369 continue; 7370 if (max_depth < udev->upper_level) 7371 max_depth = udev->upper_level; 7372 } 7373 7374 return max_depth; 7375 } 7376 7377 static u8 __netdev_lower_depth(struct net_device *dev) 7378 { 7379 struct net_device *ldev; 7380 struct list_head *iter; 7381 u8 max_depth = 0; 7382 bool ignore; 7383 7384 for (iter = &dev->adj_list.lower, 7385 ldev = __netdev_next_lower_dev(dev, &iter, &ignore); 7386 ldev; 7387 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) { 7388 if (ignore) 7389 continue; 7390 if (max_depth < ldev->lower_level) 7391 max_depth = ldev->lower_level; 7392 } 7393 7394 return max_depth; 7395 } 7396 7397 static int __netdev_update_upper_level(struct net_device *dev, 7398 struct netdev_nested_priv *__unused) 7399 { 7400 dev->upper_level = __netdev_upper_depth(dev) + 1; 7401 return 0; 7402 } 7403 7404 #ifdef CONFIG_LOCKDEP 7405 static LIST_HEAD(net_unlink_list); 7406 7407 static void net_unlink_todo(struct net_device *dev) 7408 { 7409 if (list_empty(&dev->unlink_list)) 7410 list_add_tail(&dev->unlink_list, &net_unlink_list); 7411 } 7412 #endif 7413 7414 static int __netdev_update_lower_level(struct net_device *dev, 7415 struct netdev_nested_priv *priv) 7416 { 7417 dev->lower_level = __netdev_lower_depth(dev) + 1; 7418 7419 #ifdef CONFIG_LOCKDEP 7420 if (!priv) 7421 return 0; 7422 7423 if (priv->flags & NESTED_SYNC_IMM) 7424 dev->nested_level = dev->lower_level - 1; 7425 if (priv->flags & NESTED_SYNC_TODO) 7426 net_unlink_todo(dev); 7427 #endif 7428 return 0; 7429 } 7430 7431 int netdev_walk_all_lower_dev_rcu(struct net_device *dev, 7432 int (*fn)(struct net_device *dev, 7433 struct netdev_nested_priv *priv), 7434 struct netdev_nested_priv *priv) 7435 { 7436 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7437 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7438 int ret, cur = 0; 7439 7440 now = dev; 7441 iter = &dev->adj_list.lower; 7442 7443 while (1) { 7444 if (now != dev) { 7445 ret = fn(now, priv); 7446 if (ret) 7447 return ret; 7448 } 7449 7450 next = NULL; 7451 while (1) { 7452 ldev = netdev_next_lower_dev_rcu(now, &iter); 7453 if (!ldev) 7454 break; 7455 7456 next = ldev; 7457 niter = &ldev->adj_list.lower; 7458 dev_stack[cur] = now; 7459 iter_stack[cur++] = iter; 7460 break; 7461 } 7462 7463 if (!next) { 7464 if (!cur) 7465 return 0; 7466 next = dev_stack[--cur]; 7467 niter = iter_stack[cur]; 7468 } 7469 7470 now = next; 7471 iter = niter; 7472 } 7473 7474 return 0; 7475 } 7476 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); 7477 7478 /** 7479 * netdev_lower_get_first_private_rcu - Get the first ->private from the 7480 * lower neighbour list, RCU 7481 * variant 7482 * @dev: device 7483 * 7484 * Gets the first netdev_adjacent->private from the dev's lower neighbour 7485 * list. The caller must hold RCU read lock. 7486 */ 7487 void *netdev_lower_get_first_private_rcu(struct net_device *dev) 7488 { 7489 struct netdev_adjacent *lower; 7490 7491 lower = list_first_or_null_rcu(&dev->adj_list.lower, 7492 struct netdev_adjacent, list); 7493 if (lower) 7494 return lower->private; 7495 return NULL; 7496 } 7497 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); 7498 7499 /** 7500 * netdev_master_upper_dev_get_rcu - Get master upper device 7501 * @dev: device 7502 * 7503 * Find a master upper device and return pointer to it or NULL in case 7504 * it's not there. The caller must hold the RCU read lock. 7505 */ 7506 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) 7507 { 7508 struct netdev_adjacent *upper; 7509 7510 upper = list_first_or_null_rcu(&dev->adj_list.upper, 7511 struct netdev_adjacent, list); 7512 if (upper && likely(upper->master)) 7513 return upper->dev; 7514 return NULL; 7515 } 7516 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); 7517 7518 static int netdev_adjacent_sysfs_add(struct net_device *dev, 7519 struct net_device *adj_dev, 7520 struct list_head *dev_list) 7521 { 7522 char linkname[IFNAMSIZ+7]; 7523 7524 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7525 "upper_%s" : "lower_%s", adj_dev->name); 7526 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), 7527 linkname); 7528 } 7529 static void netdev_adjacent_sysfs_del(struct net_device *dev, 7530 char *name, 7531 struct list_head *dev_list) 7532 { 7533 char linkname[IFNAMSIZ+7]; 7534 7535 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7536 "upper_%s" : "lower_%s", name); 7537 sysfs_remove_link(&(dev->dev.kobj), linkname); 7538 } 7539 7540 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, 7541 struct net_device *adj_dev, 7542 struct list_head *dev_list) 7543 { 7544 return (dev_list == &dev->adj_list.upper || 7545 dev_list == &dev->adj_list.lower) && 7546 net_eq(dev_net(dev), dev_net(adj_dev)); 7547 } 7548 7549 static int __netdev_adjacent_dev_insert(struct net_device *dev, 7550 struct net_device *adj_dev, 7551 struct list_head *dev_list, 7552 void *private, bool master) 7553 { 7554 struct netdev_adjacent *adj; 7555 int ret; 7556 7557 adj = __netdev_find_adj(adj_dev, dev_list); 7558 7559 if (adj) { 7560 adj->ref_nr += 1; 7561 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", 7562 dev->name, adj_dev->name, adj->ref_nr); 7563 7564 return 0; 7565 } 7566 7567 adj = kmalloc(sizeof(*adj), GFP_KERNEL); 7568 if (!adj) 7569 return -ENOMEM; 7570 7571 adj->dev = adj_dev; 7572 adj->master = master; 7573 adj->ref_nr = 1; 7574 adj->private = private; 7575 adj->ignore = false; 7576 netdev_hold(adj_dev, &adj->dev_tracker, GFP_KERNEL); 7577 7578 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", 7579 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); 7580 7581 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { 7582 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); 7583 if (ret) 7584 goto free_adj; 7585 } 7586 7587 /* Ensure that master link is always the first item in list. */ 7588 if (master) { 7589 ret = sysfs_create_link(&(dev->dev.kobj), 7590 &(adj_dev->dev.kobj), "master"); 7591 if (ret) 7592 goto remove_symlinks; 7593 7594 list_add_rcu(&adj->list, dev_list); 7595 } else { 7596 list_add_tail_rcu(&adj->list, dev_list); 7597 } 7598 7599 return 0; 7600 7601 remove_symlinks: 7602 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7603 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7604 free_adj: 7605 netdev_put(adj_dev, &adj->dev_tracker); 7606 kfree(adj); 7607 7608 return ret; 7609 } 7610 7611 static void __netdev_adjacent_dev_remove(struct net_device *dev, 7612 struct net_device *adj_dev, 7613 u16 ref_nr, 7614 struct list_head *dev_list) 7615 { 7616 struct netdev_adjacent *adj; 7617 7618 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", 7619 dev->name, adj_dev->name, ref_nr); 7620 7621 adj = __netdev_find_adj(adj_dev, dev_list); 7622 7623 if (!adj) { 7624 pr_err("Adjacency does not exist for device %s from %s\n", 7625 dev->name, adj_dev->name); 7626 WARN_ON(1); 7627 return; 7628 } 7629 7630 if (adj->ref_nr > ref_nr) { 7631 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", 7632 dev->name, adj_dev->name, ref_nr, 7633 adj->ref_nr - ref_nr); 7634 adj->ref_nr -= ref_nr; 7635 return; 7636 } 7637 7638 if (adj->master) 7639 sysfs_remove_link(&(dev->dev.kobj), "master"); 7640 7641 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7642 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7643 7644 list_del_rcu(&adj->list); 7645 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", 7646 adj_dev->name, dev->name, adj_dev->name); 7647 netdev_put(adj_dev, &adj->dev_tracker); 7648 kfree_rcu(adj, rcu); 7649 } 7650 7651 static int __netdev_adjacent_dev_link_lists(struct net_device *dev, 7652 struct net_device *upper_dev, 7653 struct list_head *up_list, 7654 struct list_head *down_list, 7655 void *private, bool master) 7656 { 7657 int ret; 7658 7659 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, 7660 private, master); 7661 if (ret) 7662 return ret; 7663 7664 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, 7665 private, false); 7666 if (ret) { 7667 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); 7668 return ret; 7669 } 7670 7671 return 0; 7672 } 7673 7674 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, 7675 struct net_device *upper_dev, 7676 u16 ref_nr, 7677 struct list_head *up_list, 7678 struct list_head *down_list) 7679 { 7680 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); 7681 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); 7682 } 7683 7684 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, 7685 struct net_device *upper_dev, 7686 void *private, bool master) 7687 { 7688 return __netdev_adjacent_dev_link_lists(dev, upper_dev, 7689 &dev->adj_list.upper, 7690 &upper_dev->adj_list.lower, 7691 private, master); 7692 } 7693 7694 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, 7695 struct net_device *upper_dev) 7696 { 7697 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, 7698 &dev->adj_list.upper, 7699 &upper_dev->adj_list.lower); 7700 } 7701 7702 static int __netdev_upper_dev_link(struct net_device *dev, 7703 struct net_device *upper_dev, bool master, 7704 void *upper_priv, void *upper_info, 7705 struct netdev_nested_priv *priv, 7706 struct netlink_ext_ack *extack) 7707 { 7708 struct netdev_notifier_changeupper_info changeupper_info = { 7709 .info = { 7710 .dev = dev, 7711 .extack = extack, 7712 }, 7713 .upper_dev = upper_dev, 7714 .master = master, 7715 .linking = true, 7716 .upper_info = upper_info, 7717 }; 7718 struct net_device *master_dev; 7719 int ret = 0; 7720 7721 ASSERT_RTNL(); 7722 7723 if (dev == upper_dev) 7724 return -EBUSY; 7725 7726 /* To prevent loops, check if dev is not upper device to upper_dev. */ 7727 if (__netdev_has_upper_dev(upper_dev, dev)) 7728 return -EBUSY; 7729 7730 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV) 7731 return -EMLINK; 7732 7733 if (!master) { 7734 if (__netdev_has_upper_dev(dev, upper_dev)) 7735 return -EEXIST; 7736 } else { 7737 master_dev = __netdev_master_upper_dev_get(dev); 7738 if (master_dev) 7739 return master_dev == upper_dev ? -EEXIST : -EBUSY; 7740 } 7741 7742 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7743 &changeupper_info.info); 7744 ret = notifier_to_errno(ret); 7745 if (ret) 7746 return ret; 7747 7748 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, 7749 master); 7750 if (ret) 7751 return ret; 7752 7753 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 7754 &changeupper_info.info); 7755 ret = notifier_to_errno(ret); 7756 if (ret) 7757 goto rollback; 7758 7759 __netdev_update_upper_level(dev, NULL); 7760 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 7761 7762 __netdev_update_lower_level(upper_dev, priv); 7763 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 7764 priv); 7765 7766 return 0; 7767 7768 rollback: 7769 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 7770 7771 return ret; 7772 } 7773 7774 /** 7775 * netdev_upper_dev_link - Add a link to the upper device 7776 * @dev: device 7777 * @upper_dev: new upper device 7778 * @extack: netlink extended ack 7779 * 7780 * Adds a link to device which is upper to this one. The caller must hold 7781 * the RTNL lock. On a failure a negative errno code is returned. 7782 * On success the reference counts are adjusted and the function 7783 * returns zero. 7784 */ 7785 int netdev_upper_dev_link(struct net_device *dev, 7786 struct net_device *upper_dev, 7787 struct netlink_ext_ack *extack) 7788 { 7789 struct netdev_nested_priv priv = { 7790 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7791 .data = NULL, 7792 }; 7793 7794 return __netdev_upper_dev_link(dev, upper_dev, false, 7795 NULL, NULL, &priv, extack); 7796 } 7797 EXPORT_SYMBOL(netdev_upper_dev_link); 7798 7799 /** 7800 * netdev_master_upper_dev_link - Add a master link to the upper device 7801 * @dev: device 7802 * @upper_dev: new upper device 7803 * @upper_priv: upper device private 7804 * @upper_info: upper info to be passed down via notifier 7805 * @extack: netlink extended ack 7806 * 7807 * Adds a link to device which is upper to this one. In this case, only 7808 * one master upper device can be linked, although other non-master devices 7809 * might be linked as well. The caller must hold the RTNL lock. 7810 * On a failure a negative errno code is returned. On success the reference 7811 * counts are adjusted and the function returns zero. 7812 */ 7813 int netdev_master_upper_dev_link(struct net_device *dev, 7814 struct net_device *upper_dev, 7815 void *upper_priv, void *upper_info, 7816 struct netlink_ext_ack *extack) 7817 { 7818 struct netdev_nested_priv priv = { 7819 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7820 .data = NULL, 7821 }; 7822 7823 return __netdev_upper_dev_link(dev, upper_dev, true, 7824 upper_priv, upper_info, &priv, extack); 7825 } 7826 EXPORT_SYMBOL(netdev_master_upper_dev_link); 7827 7828 static void __netdev_upper_dev_unlink(struct net_device *dev, 7829 struct net_device *upper_dev, 7830 struct netdev_nested_priv *priv) 7831 { 7832 struct netdev_notifier_changeupper_info changeupper_info = { 7833 .info = { 7834 .dev = dev, 7835 }, 7836 .upper_dev = upper_dev, 7837 .linking = false, 7838 }; 7839 7840 ASSERT_RTNL(); 7841 7842 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; 7843 7844 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7845 &changeupper_info.info); 7846 7847 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 7848 7849 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 7850 &changeupper_info.info); 7851 7852 __netdev_update_upper_level(dev, NULL); 7853 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 7854 7855 __netdev_update_lower_level(upper_dev, priv); 7856 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 7857 priv); 7858 } 7859 7860 /** 7861 * netdev_upper_dev_unlink - Removes a link to upper device 7862 * @dev: device 7863 * @upper_dev: new upper device 7864 * 7865 * Removes a link to device which is upper to this one. The caller must hold 7866 * the RTNL lock. 7867 */ 7868 void netdev_upper_dev_unlink(struct net_device *dev, 7869 struct net_device *upper_dev) 7870 { 7871 struct netdev_nested_priv priv = { 7872 .flags = NESTED_SYNC_TODO, 7873 .data = NULL, 7874 }; 7875 7876 __netdev_upper_dev_unlink(dev, upper_dev, &priv); 7877 } 7878 EXPORT_SYMBOL(netdev_upper_dev_unlink); 7879 7880 static void __netdev_adjacent_dev_set(struct net_device *upper_dev, 7881 struct net_device *lower_dev, 7882 bool val) 7883 { 7884 struct netdev_adjacent *adj; 7885 7886 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower); 7887 if (adj) 7888 adj->ignore = val; 7889 7890 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper); 7891 if (adj) 7892 adj->ignore = val; 7893 } 7894 7895 static void netdev_adjacent_dev_disable(struct net_device *upper_dev, 7896 struct net_device *lower_dev) 7897 { 7898 __netdev_adjacent_dev_set(upper_dev, lower_dev, true); 7899 } 7900 7901 static void netdev_adjacent_dev_enable(struct net_device *upper_dev, 7902 struct net_device *lower_dev) 7903 { 7904 __netdev_adjacent_dev_set(upper_dev, lower_dev, false); 7905 } 7906 7907 int netdev_adjacent_change_prepare(struct net_device *old_dev, 7908 struct net_device *new_dev, 7909 struct net_device *dev, 7910 struct netlink_ext_ack *extack) 7911 { 7912 struct netdev_nested_priv priv = { 7913 .flags = 0, 7914 .data = NULL, 7915 }; 7916 int err; 7917 7918 if (!new_dev) 7919 return 0; 7920 7921 if (old_dev && new_dev != old_dev) 7922 netdev_adjacent_dev_disable(dev, old_dev); 7923 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv, 7924 extack); 7925 if (err) { 7926 if (old_dev && new_dev != old_dev) 7927 netdev_adjacent_dev_enable(dev, old_dev); 7928 return err; 7929 } 7930 7931 return 0; 7932 } 7933 EXPORT_SYMBOL(netdev_adjacent_change_prepare); 7934 7935 void netdev_adjacent_change_commit(struct net_device *old_dev, 7936 struct net_device *new_dev, 7937 struct net_device *dev) 7938 { 7939 struct netdev_nested_priv priv = { 7940 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7941 .data = NULL, 7942 }; 7943 7944 if (!new_dev || !old_dev) 7945 return; 7946 7947 if (new_dev == old_dev) 7948 return; 7949 7950 netdev_adjacent_dev_enable(dev, old_dev); 7951 __netdev_upper_dev_unlink(old_dev, dev, &priv); 7952 } 7953 EXPORT_SYMBOL(netdev_adjacent_change_commit); 7954 7955 void netdev_adjacent_change_abort(struct net_device *old_dev, 7956 struct net_device *new_dev, 7957 struct net_device *dev) 7958 { 7959 struct netdev_nested_priv priv = { 7960 .flags = 0, 7961 .data = NULL, 7962 }; 7963 7964 if (!new_dev) 7965 return; 7966 7967 if (old_dev && new_dev != old_dev) 7968 netdev_adjacent_dev_enable(dev, old_dev); 7969 7970 __netdev_upper_dev_unlink(new_dev, dev, &priv); 7971 } 7972 EXPORT_SYMBOL(netdev_adjacent_change_abort); 7973 7974 /** 7975 * netdev_bonding_info_change - Dispatch event about slave change 7976 * @dev: device 7977 * @bonding_info: info to dispatch 7978 * 7979 * Send NETDEV_BONDING_INFO to netdev notifiers with info. 7980 * The caller must hold the RTNL lock. 7981 */ 7982 void netdev_bonding_info_change(struct net_device *dev, 7983 struct netdev_bonding_info *bonding_info) 7984 { 7985 struct netdev_notifier_bonding_info info = { 7986 .info.dev = dev, 7987 }; 7988 7989 memcpy(&info.bonding_info, bonding_info, 7990 sizeof(struct netdev_bonding_info)); 7991 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, 7992 &info.info); 7993 } 7994 EXPORT_SYMBOL(netdev_bonding_info_change); 7995 7996 static int netdev_offload_xstats_enable_l3(struct net_device *dev, 7997 struct netlink_ext_ack *extack) 7998 { 7999 struct netdev_notifier_offload_xstats_info info = { 8000 .info.dev = dev, 8001 .info.extack = extack, 8002 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, 8003 }; 8004 int err; 8005 int rc; 8006 8007 dev->offload_xstats_l3 = kzalloc(sizeof(*dev->offload_xstats_l3), 8008 GFP_KERNEL); 8009 if (!dev->offload_xstats_l3) 8010 return -ENOMEM; 8011 8012 rc = call_netdevice_notifiers_info_robust(NETDEV_OFFLOAD_XSTATS_ENABLE, 8013 NETDEV_OFFLOAD_XSTATS_DISABLE, 8014 &info.info); 8015 err = notifier_to_errno(rc); 8016 if (err) 8017 goto free_stats; 8018 8019 return 0; 8020 8021 free_stats: 8022 kfree(dev->offload_xstats_l3); 8023 dev->offload_xstats_l3 = NULL; 8024 return err; 8025 } 8026 8027 int netdev_offload_xstats_enable(struct net_device *dev, 8028 enum netdev_offload_xstats_type type, 8029 struct netlink_ext_ack *extack) 8030 { 8031 ASSERT_RTNL(); 8032 8033 if (netdev_offload_xstats_enabled(dev, type)) 8034 return -EALREADY; 8035 8036 switch (type) { 8037 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 8038 return netdev_offload_xstats_enable_l3(dev, extack); 8039 } 8040 8041 WARN_ON(1); 8042 return -EINVAL; 8043 } 8044 EXPORT_SYMBOL(netdev_offload_xstats_enable); 8045 8046 static void netdev_offload_xstats_disable_l3(struct net_device *dev) 8047 { 8048 struct netdev_notifier_offload_xstats_info info = { 8049 .info.dev = dev, 8050 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, 8051 }; 8052 8053 call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_DISABLE, 8054 &info.info); 8055 kfree(dev->offload_xstats_l3); 8056 dev->offload_xstats_l3 = NULL; 8057 } 8058 8059 int netdev_offload_xstats_disable(struct net_device *dev, 8060 enum netdev_offload_xstats_type type) 8061 { 8062 ASSERT_RTNL(); 8063 8064 if (!netdev_offload_xstats_enabled(dev, type)) 8065 return -EALREADY; 8066 8067 switch (type) { 8068 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 8069 netdev_offload_xstats_disable_l3(dev); 8070 return 0; 8071 } 8072 8073 WARN_ON(1); 8074 return -EINVAL; 8075 } 8076 EXPORT_SYMBOL(netdev_offload_xstats_disable); 8077 8078 static void netdev_offload_xstats_disable_all(struct net_device *dev) 8079 { 8080 netdev_offload_xstats_disable(dev, NETDEV_OFFLOAD_XSTATS_TYPE_L3); 8081 } 8082 8083 static struct rtnl_hw_stats64 * 8084 netdev_offload_xstats_get_ptr(const struct net_device *dev, 8085 enum netdev_offload_xstats_type type) 8086 { 8087 switch (type) { 8088 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 8089 return dev->offload_xstats_l3; 8090 } 8091 8092 WARN_ON(1); 8093 return NULL; 8094 } 8095 8096 bool netdev_offload_xstats_enabled(const struct net_device *dev, 8097 enum netdev_offload_xstats_type type) 8098 { 8099 ASSERT_RTNL(); 8100 8101 return netdev_offload_xstats_get_ptr(dev, type); 8102 } 8103 EXPORT_SYMBOL(netdev_offload_xstats_enabled); 8104 8105 struct netdev_notifier_offload_xstats_ru { 8106 bool used; 8107 }; 8108 8109 struct netdev_notifier_offload_xstats_rd { 8110 struct rtnl_hw_stats64 stats; 8111 bool used; 8112 }; 8113 8114 static void netdev_hw_stats64_add(struct rtnl_hw_stats64 *dest, 8115 const struct rtnl_hw_stats64 *src) 8116 { 8117 dest->rx_packets += src->rx_packets; 8118 dest->tx_packets += src->tx_packets; 8119 dest->rx_bytes += src->rx_bytes; 8120 dest->tx_bytes += src->tx_bytes; 8121 dest->rx_errors += src->rx_errors; 8122 dest->tx_errors += src->tx_errors; 8123 dest->rx_dropped += src->rx_dropped; 8124 dest->tx_dropped += src->tx_dropped; 8125 dest->multicast += src->multicast; 8126 } 8127 8128 static int netdev_offload_xstats_get_used(struct net_device *dev, 8129 enum netdev_offload_xstats_type type, 8130 bool *p_used, 8131 struct netlink_ext_ack *extack) 8132 { 8133 struct netdev_notifier_offload_xstats_ru report_used = {}; 8134 struct netdev_notifier_offload_xstats_info info = { 8135 .info.dev = dev, 8136 .info.extack = extack, 8137 .type = type, 8138 .report_used = &report_used, 8139 }; 8140 int rc; 8141 8142 WARN_ON(!netdev_offload_xstats_enabled(dev, type)); 8143 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_USED, 8144 &info.info); 8145 *p_used = report_used.used; 8146 return notifier_to_errno(rc); 8147 } 8148 8149 static int netdev_offload_xstats_get_stats(struct net_device *dev, 8150 enum netdev_offload_xstats_type type, 8151 struct rtnl_hw_stats64 *p_stats, 8152 bool *p_used, 8153 struct netlink_ext_ack *extack) 8154 { 8155 struct netdev_notifier_offload_xstats_rd report_delta = {}; 8156 struct netdev_notifier_offload_xstats_info info = { 8157 .info.dev = dev, 8158 .info.extack = extack, 8159 .type = type, 8160 .report_delta = &report_delta, 8161 }; 8162 struct rtnl_hw_stats64 *stats; 8163 int rc; 8164 8165 stats = netdev_offload_xstats_get_ptr(dev, type); 8166 if (WARN_ON(!stats)) 8167 return -EINVAL; 8168 8169 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_DELTA, 8170 &info.info); 8171 8172 /* Cache whatever we got, even if there was an error, otherwise the 8173 * successful stats retrievals would get lost. 8174 */ 8175 netdev_hw_stats64_add(stats, &report_delta.stats); 8176 8177 if (p_stats) 8178 *p_stats = *stats; 8179 *p_used = report_delta.used; 8180 8181 return notifier_to_errno(rc); 8182 } 8183 8184 int netdev_offload_xstats_get(struct net_device *dev, 8185 enum netdev_offload_xstats_type type, 8186 struct rtnl_hw_stats64 *p_stats, bool *p_used, 8187 struct netlink_ext_ack *extack) 8188 { 8189 ASSERT_RTNL(); 8190 8191 if (p_stats) 8192 return netdev_offload_xstats_get_stats(dev, type, p_stats, 8193 p_used, extack); 8194 else 8195 return netdev_offload_xstats_get_used(dev, type, p_used, 8196 extack); 8197 } 8198 EXPORT_SYMBOL(netdev_offload_xstats_get); 8199 8200 void 8201 netdev_offload_xstats_report_delta(struct netdev_notifier_offload_xstats_rd *report_delta, 8202 const struct rtnl_hw_stats64 *stats) 8203 { 8204 report_delta->used = true; 8205 netdev_hw_stats64_add(&report_delta->stats, stats); 8206 } 8207 EXPORT_SYMBOL(netdev_offload_xstats_report_delta); 8208 8209 void 8210 netdev_offload_xstats_report_used(struct netdev_notifier_offload_xstats_ru *report_used) 8211 { 8212 report_used->used = true; 8213 } 8214 EXPORT_SYMBOL(netdev_offload_xstats_report_used); 8215 8216 void netdev_offload_xstats_push_delta(struct net_device *dev, 8217 enum netdev_offload_xstats_type type, 8218 const struct rtnl_hw_stats64 *p_stats) 8219 { 8220 struct rtnl_hw_stats64 *stats; 8221 8222 ASSERT_RTNL(); 8223 8224 stats = netdev_offload_xstats_get_ptr(dev, type); 8225 if (WARN_ON(!stats)) 8226 return; 8227 8228 netdev_hw_stats64_add(stats, p_stats); 8229 } 8230 EXPORT_SYMBOL(netdev_offload_xstats_push_delta); 8231 8232 /** 8233 * netdev_get_xmit_slave - Get the xmit slave of master device 8234 * @dev: device 8235 * @skb: The packet 8236 * @all_slaves: assume all the slaves are active 8237 * 8238 * The reference counters are not incremented so the caller must be 8239 * careful with locks. The caller must hold RCU lock. 8240 * %NULL is returned if no slave is found. 8241 */ 8242 8243 struct net_device *netdev_get_xmit_slave(struct net_device *dev, 8244 struct sk_buff *skb, 8245 bool all_slaves) 8246 { 8247 const struct net_device_ops *ops = dev->netdev_ops; 8248 8249 if (!ops->ndo_get_xmit_slave) 8250 return NULL; 8251 return ops->ndo_get_xmit_slave(dev, skb, all_slaves); 8252 } 8253 EXPORT_SYMBOL(netdev_get_xmit_slave); 8254 8255 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev, 8256 struct sock *sk) 8257 { 8258 const struct net_device_ops *ops = dev->netdev_ops; 8259 8260 if (!ops->ndo_sk_get_lower_dev) 8261 return NULL; 8262 return ops->ndo_sk_get_lower_dev(dev, sk); 8263 } 8264 8265 /** 8266 * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket 8267 * @dev: device 8268 * @sk: the socket 8269 * 8270 * %NULL is returned if no lower device is found. 8271 */ 8272 8273 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev, 8274 struct sock *sk) 8275 { 8276 struct net_device *lower; 8277 8278 lower = netdev_sk_get_lower_dev(dev, sk); 8279 while (lower) { 8280 dev = lower; 8281 lower = netdev_sk_get_lower_dev(dev, sk); 8282 } 8283 8284 return dev; 8285 } 8286 EXPORT_SYMBOL(netdev_sk_get_lowest_dev); 8287 8288 static void netdev_adjacent_add_links(struct net_device *dev) 8289 { 8290 struct netdev_adjacent *iter; 8291 8292 struct net *net = dev_net(dev); 8293 8294 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8295 if (!net_eq(net, dev_net(iter->dev))) 8296 continue; 8297 netdev_adjacent_sysfs_add(iter->dev, dev, 8298 &iter->dev->adj_list.lower); 8299 netdev_adjacent_sysfs_add(dev, iter->dev, 8300 &dev->adj_list.upper); 8301 } 8302 8303 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8304 if (!net_eq(net, dev_net(iter->dev))) 8305 continue; 8306 netdev_adjacent_sysfs_add(iter->dev, dev, 8307 &iter->dev->adj_list.upper); 8308 netdev_adjacent_sysfs_add(dev, iter->dev, 8309 &dev->adj_list.lower); 8310 } 8311 } 8312 8313 static void netdev_adjacent_del_links(struct net_device *dev) 8314 { 8315 struct netdev_adjacent *iter; 8316 8317 struct net *net = dev_net(dev); 8318 8319 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8320 if (!net_eq(net, dev_net(iter->dev))) 8321 continue; 8322 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8323 &iter->dev->adj_list.lower); 8324 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8325 &dev->adj_list.upper); 8326 } 8327 8328 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8329 if (!net_eq(net, dev_net(iter->dev))) 8330 continue; 8331 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8332 &iter->dev->adj_list.upper); 8333 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8334 &dev->adj_list.lower); 8335 } 8336 } 8337 8338 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) 8339 { 8340 struct netdev_adjacent *iter; 8341 8342 struct net *net = dev_net(dev); 8343 8344 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8345 if (!net_eq(net, dev_net(iter->dev))) 8346 continue; 8347 netdev_adjacent_sysfs_del(iter->dev, oldname, 8348 &iter->dev->adj_list.lower); 8349 netdev_adjacent_sysfs_add(iter->dev, dev, 8350 &iter->dev->adj_list.lower); 8351 } 8352 8353 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8354 if (!net_eq(net, dev_net(iter->dev))) 8355 continue; 8356 netdev_adjacent_sysfs_del(iter->dev, oldname, 8357 &iter->dev->adj_list.upper); 8358 netdev_adjacent_sysfs_add(iter->dev, dev, 8359 &iter->dev->adj_list.upper); 8360 } 8361 } 8362 8363 void *netdev_lower_dev_get_private(struct net_device *dev, 8364 struct net_device *lower_dev) 8365 { 8366 struct netdev_adjacent *lower; 8367 8368 if (!lower_dev) 8369 return NULL; 8370 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); 8371 if (!lower) 8372 return NULL; 8373 8374 return lower->private; 8375 } 8376 EXPORT_SYMBOL(netdev_lower_dev_get_private); 8377 8378 8379 /** 8380 * netdev_lower_state_changed - Dispatch event about lower device state change 8381 * @lower_dev: device 8382 * @lower_state_info: state to dispatch 8383 * 8384 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. 8385 * The caller must hold the RTNL lock. 8386 */ 8387 void netdev_lower_state_changed(struct net_device *lower_dev, 8388 void *lower_state_info) 8389 { 8390 struct netdev_notifier_changelowerstate_info changelowerstate_info = { 8391 .info.dev = lower_dev, 8392 }; 8393 8394 ASSERT_RTNL(); 8395 changelowerstate_info.lower_state_info = lower_state_info; 8396 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, 8397 &changelowerstate_info.info); 8398 } 8399 EXPORT_SYMBOL(netdev_lower_state_changed); 8400 8401 static void dev_change_rx_flags(struct net_device *dev, int flags) 8402 { 8403 const struct net_device_ops *ops = dev->netdev_ops; 8404 8405 if (ops->ndo_change_rx_flags) 8406 ops->ndo_change_rx_flags(dev, flags); 8407 } 8408 8409 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) 8410 { 8411 unsigned int old_flags = dev->flags; 8412 kuid_t uid; 8413 kgid_t gid; 8414 8415 ASSERT_RTNL(); 8416 8417 dev->flags |= IFF_PROMISC; 8418 dev->promiscuity += inc; 8419 if (dev->promiscuity == 0) { 8420 /* 8421 * Avoid overflow. 8422 * If inc causes overflow, untouch promisc and return error. 8423 */ 8424 if (inc < 0) 8425 dev->flags &= ~IFF_PROMISC; 8426 else { 8427 dev->promiscuity -= inc; 8428 netdev_warn(dev, "promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n"); 8429 return -EOVERFLOW; 8430 } 8431 } 8432 if (dev->flags != old_flags) { 8433 netdev_info(dev, "%s promiscuous mode\n", 8434 dev->flags & IFF_PROMISC ? "entered" : "left"); 8435 if (audit_enabled) { 8436 current_uid_gid(&uid, &gid); 8437 audit_log(audit_context(), GFP_ATOMIC, 8438 AUDIT_ANOM_PROMISCUOUS, 8439 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", 8440 dev->name, (dev->flags & IFF_PROMISC), 8441 (old_flags & IFF_PROMISC), 8442 from_kuid(&init_user_ns, audit_get_loginuid(current)), 8443 from_kuid(&init_user_ns, uid), 8444 from_kgid(&init_user_ns, gid), 8445 audit_get_sessionid(current)); 8446 } 8447 8448 dev_change_rx_flags(dev, IFF_PROMISC); 8449 } 8450 if (notify) 8451 __dev_notify_flags(dev, old_flags, IFF_PROMISC, 0, NULL); 8452 return 0; 8453 } 8454 8455 /** 8456 * dev_set_promiscuity - update promiscuity count on a device 8457 * @dev: device 8458 * @inc: modifier 8459 * 8460 * Add or remove promiscuity from a device. While the count in the device 8461 * remains above zero the interface remains promiscuous. Once it hits zero 8462 * the device reverts back to normal filtering operation. A negative inc 8463 * value is used to drop promiscuity on the device. 8464 * Return 0 if successful or a negative errno code on error. 8465 */ 8466 int dev_set_promiscuity(struct net_device *dev, int inc) 8467 { 8468 unsigned int old_flags = dev->flags; 8469 int err; 8470 8471 err = __dev_set_promiscuity(dev, inc, true); 8472 if (err < 0) 8473 return err; 8474 if (dev->flags != old_flags) 8475 dev_set_rx_mode(dev); 8476 return err; 8477 } 8478 EXPORT_SYMBOL(dev_set_promiscuity); 8479 8480 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) 8481 { 8482 unsigned int old_flags = dev->flags, old_gflags = dev->gflags; 8483 8484 ASSERT_RTNL(); 8485 8486 dev->flags |= IFF_ALLMULTI; 8487 dev->allmulti += inc; 8488 if (dev->allmulti == 0) { 8489 /* 8490 * Avoid overflow. 8491 * If inc causes overflow, untouch allmulti and return error. 8492 */ 8493 if (inc < 0) 8494 dev->flags &= ~IFF_ALLMULTI; 8495 else { 8496 dev->allmulti -= inc; 8497 netdev_warn(dev, "allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n"); 8498 return -EOVERFLOW; 8499 } 8500 } 8501 if (dev->flags ^ old_flags) { 8502 netdev_info(dev, "%s allmulticast mode\n", 8503 dev->flags & IFF_ALLMULTI ? "entered" : "left"); 8504 dev_change_rx_flags(dev, IFF_ALLMULTI); 8505 dev_set_rx_mode(dev); 8506 if (notify) 8507 __dev_notify_flags(dev, old_flags, 8508 dev->gflags ^ old_gflags, 0, NULL); 8509 } 8510 return 0; 8511 } 8512 8513 /** 8514 * dev_set_allmulti - update allmulti count on a device 8515 * @dev: device 8516 * @inc: modifier 8517 * 8518 * Add or remove reception of all multicast frames to a device. While the 8519 * count in the device remains above zero the interface remains listening 8520 * to all interfaces. Once it hits zero the device reverts back to normal 8521 * filtering operation. A negative @inc value is used to drop the counter 8522 * when releasing a resource needing all multicasts. 8523 * Return 0 if successful or a negative errno code on error. 8524 */ 8525 8526 int dev_set_allmulti(struct net_device *dev, int inc) 8527 { 8528 return __dev_set_allmulti(dev, inc, true); 8529 } 8530 EXPORT_SYMBOL(dev_set_allmulti); 8531 8532 /* 8533 * Upload unicast and multicast address lists to device and 8534 * configure RX filtering. When the device doesn't support unicast 8535 * filtering it is put in promiscuous mode while unicast addresses 8536 * are present. 8537 */ 8538 void __dev_set_rx_mode(struct net_device *dev) 8539 { 8540 const struct net_device_ops *ops = dev->netdev_ops; 8541 8542 /* dev_open will call this function so the list will stay sane. */ 8543 if (!(dev->flags&IFF_UP)) 8544 return; 8545 8546 if (!netif_device_present(dev)) 8547 return; 8548 8549 if (!(dev->priv_flags & IFF_UNICAST_FLT)) { 8550 /* Unicast addresses changes may only happen under the rtnl, 8551 * therefore calling __dev_set_promiscuity here is safe. 8552 */ 8553 if (!netdev_uc_empty(dev) && !dev->uc_promisc) { 8554 __dev_set_promiscuity(dev, 1, false); 8555 dev->uc_promisc = true; 8556 } else if (netdev_uc_empty(dev) && dev->uc_promisc) { 8557 __dev_set_promiscuity(dev, -1, false); 8558 dev->uc_promisc = false; 8559 } 8560 } 8561 8562 if (ops->ndo_set_rx_mode) 8563 ops->ndo_set_rx_mode(dev); 8564 } 8565 8566 void dev_set_rx_mode(struct net_device *dev) 8567 { 8568 netif_addr_lock_bh(dev); 8569 __dev_set_rx_mode(dev); 8570 netif_addr_unlock_bh(dev); 8571 } 8572 8573 /** 8574 * dev_get_flags - get flags reported to userspace 8575 * @dev: device 8576 * 8577 * Get the combination of flag bits exported through APIs to userspace. 8578 */ 8579 unsigned int dev_get_flags(const struct net_device *dev) 8580 { 8581 unsigned int flags; 8582 8583 flags = (dev->flags & ~(IFF_PROMISC | 8584 IFF_ALLMULTI | 8585 IFF_RUNNING | 8586 IFF_LOWER_UP | 8587 IFF_DORMANT)) | 8588 (dev->gflags & (IFF_PROMISC | 8589 IFF_ALLMULTI)); 8590 8591 if (netif_running(dev)) { 8592 if (netif_oper_up(dev)) 8593 flags |= IFF_RUNNING; 8594 if (netif_carrier_ok(dev)) 8595 flags |= IFF_LOWER_UP; 8596 if (netif_dormant(dev)) 8597 flags |= IFF_DORMANT; 8598 } 8599 8600 return flags; 8601 } 8602 EXPORT_SYMBOL(dev_get_flags); 8603 8604 int __dev_change_flags(struct net_device *dev, unsigned int flags, 8605 struct netlink_ext_ack *extack) 8606 { 8607 unsigned int old_flags = dev->flags; 8608 int ret; 8609 8610 ASSERT_RTNL(); 8611 8612 /* 8613 * Set the flags on our device. 8614 */ 8615 8616 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | 8617 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | 8618 IFF_AUTOMEDIA)) | 8619 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | 8620 IFF_ALLMULTI)); 8621 8622 /* 8623 * Load in the correct multicast list now the flags have changed. 8624 */ 8625 8626 if ((old_flags ^ flags) & IFF_MULTICAST) 8627 dev_change_rx_flags(dev, IFF_MULTICAST); 8628 8629 dev_set_rx_mode(dev); 8630 8631 /* 8632 * Have we downed the interface. We handle IFF_UP ourselves 8633 * according to user attempts to set it, rather than blindly 8634 * setting it. 8635 */ 8636 8637 ret = 0; 8638 if ((old_flags ^ flags) & IFF_UP) { 8639 if (old_flags & IFF_UP) 8640 __dev_close(dev); 8641 else 8642 ret = __dev_open(dev, extack); 8643 } 8644 8645 if ((flags ^ dev->gflags) & IFF_PROMISC) { 8646 int inc = (flags & IFF_PROMISC) ? 1 : -1; 8647 unsigned int old_flags = dev->flags; 8648 8649 dev->gflags ^= IFF_PROMISC; 8650 8651 if (__dev_set_promiscuity(dev, inc, false) >= 0) 8652 if (dev->flags != old_flags) 8653 dev_set_rx_mode(dev); 8654 } 8655 8656 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI 8657 * is important. Some (broken) drivers set IFF_PROMISC, when 8658 * IFF_ALLMULTI is requested not asking us and not reporting. 8659 */ 8660 if ((flags ^ dev->gflags) & IFF_ALLMULTI) { 8661 int inc = (flags & IFF_ALLMULTI) ? 1 : -1; 8662 8663 dev->gflags ^= IFF_ALLMULTI; 8664 __dev_set_allmulti(dev, inc, false); 8665 } 8666 8667 return ret; 8668 } 8669 8670 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, 8671 unsigned int gchanges, u32 portid, 8672 const struct nlmsghdr *nlh) 8673 { 8674 unsigned int changes = dev->flags ^ old_flags; 8675 8676 if (gchanges) 8677 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC, portid, nlh); 8678 8679 if (changes & IFF_UP) { 8680 if (dev->flags & IFF_UP) 8681 call_netdevice_notifiers(NETDEV_UP, dev); 8682 else 8683 call_netdevice_notifiers(NETDEV_DOWN, dev); 8684 } 8685 8686 if (dev->flags & IFF_UP && 8687 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { 8688 struct netdev_notifier_change_info change_info = { 8689 .info = { 8690 .dev = dev, 8691 }, 8692 .flags_changed = changes, 8693 }; 8694 8695 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info); 8696 } 8697 } 8698 8699 /** 8700 * dev_change_flags - change device settings 8701 * @dev: device 8702 * @flags: device state flags 8703 * @extack: netlink extended ack 8704 * 8705 * Change settings on device based state flags. The flags are 8706 * in the userspace exported format. 8707 */ 8708 int dev_change_flags(struct net_device *dev, unsigned int flags, 8709 struct netlink_ext_ack *extack) 8710 { 8711 int ret; 8712 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; 8713 8714 ret = __dev_change_flags(dev, flags, extack); 8715 if (ret < 0) 8716 return ret; 8717 8718 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); 8719 __dev_notify_flags(dev, old_flags, changes, 0, NULL); 8720 return ret; 8721 } 8722 EXPORT_SYMBOL(dev_change_flags); 8723 8724 int __dev_set_mtu(struct net_device *dev, int new_mtu) 8725 { 8726 const struct net_device_ops *ops = dev->netdev_ops; 8727 8728 if (ops->ndo_change_mtu) 8729 return ops->ndo_change_mtu(dev, new_mtu); 8730 8731 /* Pairs with all the lockless reads of dev->mtu in the stack */ 8732 WRITE_ONCE(dev->mtu, new_mtu); 8733 return 0; 8734 } 8735 EXPORT_SYMBOL(__dev_set_mtu); 8736 8737 int dev_validate_mtu(struct net_device *dev, int new_mtu, 8738 struct netlink_ext_ack *extack) 8739 { 8740 /* MTU must be positive, and in range */ 8741 if (new_mtu < 0 || new_mtu < dev->min_mtu) { 8742 NL_SET_ERR_MSG(extack, "mtu less than device minimum"); 8743 return -EINVAL; 8744 } 8745 8746 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { 8747 NL_SET_ERR_MSG(extack, "mtu greater than device maximum"); 8748 return -EINVAL; 8749 } 8750 return 0; 8751 } 8752 8753 /** 8754 * dev_set_mtu_ext - Change maximum transfer unit 8755 * @dev: device 8756 * @new_mtu: new transfer unit 8757 * @extack: netlink extended ack 8758 * 8759 * Change the maximum transfer size of the network device. 8760 */ 8761 int dev_set_mtu_ext(struct net_device *dev, int new_mtu, 8762 struct netlink_ext_ack *extack) 8763 { 8764 int err, orig_mtu; 8765 8766 if (new_mtu == dev->mtu) 8767 return 0; 8768 8769 err = dev_validate_mtu(dev, new_mtu, extack); 8770 if (err) 8771 return err; 8772 8773 if (!netif_device_present(dev)) 8774 return -ENODEV; 8775 8776 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); 8777 err = notifier_to_errno(err); 8778 if (err) 8779 return err; 8780 8781 orig_mtu = dev->mtu; 8782 err = __dev_set_mtu(dev, new_mtu); 8783 8784 if (!err) { 8785 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8786 orig_mtu); 8787 err = notifier_to_errno(err); 8788 if (err) { 8789 /* setting mtu back and notifying everyone again, 8790 * so that they have a chance to revert changes. 8791 */ 8792 __dev_set_mtu(dev, orig_mtu); 8793 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8794 new_mtu); 8795 } 8796 } 8797 return err; 8798 } 8799 8800 int dev_set_mtu(struct net_device *dev, int new_mtu) 8801 { 8802 struct netlink_ext_ack extack; 8803 int err; 8804 8805 memset(&extack, 0, sizeof(extack)); 8806 err = dev_set_mtu_ext(dev, new_mtu, &extack); 8807 if (err && extack._msg) 8808 net_err_ratelimited("%s: %s\n", dev->name, extack._msg); 8809 return err; 8810 } 8811 EXPORT_SYMBOL(dev_set_mtu); 8812 8813 /** 8814 * dev_change_tx_queue_len - Change TX queue length of a netdevice 8815 * @dev: device 8816 * @new_len: new tx queue length 8817 */ 8818 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len) 8819 { 8820 unsigned int orig_len = dev->tx_queue_len; 8821 int res; 8822 8823 if (new_len != (unsigned int)new_len) 8824 return -ERANGE; 8825 8826 if (new_len != orig_len) { 8827 dev->tx_queue_len = new_len; 8828 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev); 8829 res = notifier_to_errno(res); 8830 if (res) 8831 goto err_rollback; 8832 res = dev_qdisc_change_tx_queue_len(dev); 8833 if (res) 8834 goto err_rollback; 8835 } 8836 8837 return 0; 8838 8839 err_rollback: 8840 netdev_err(dev, "refused to change device tx_queue_len\n"); 8841 dev->tx_queue_len = orig_len; 8842 return res; 8843 } 8844 8845 /** 8846 * dev_set_group - Change group this device belongs to 8847 * @dev: device 8848 * @new_group: group this device should belong to 8849 */ 8850 void dev_set_group(struct net_device *dev, int new_group) 8851 { 8852 dev->group = new_group; 8853 } 8854 8855 /** 8856 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR. 8857 * @dev: device 8858 * @addr: new address 8859 * @extack: netlink extended ack 8860 */ 8861 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr, 8862 struct netlink_ext_ack *extack) 8863 { 8864 struct netdev_notifier_pre_changeaddr_info info = { 8865 .info.dev = dev, 8866 .info.extack = extack, 8867 .dev_addr = addr, 8868 }; 8869 int rc; 8870 8871 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info); 8872 return notifier_to_errno(rc); 8873 } 8874 EXPORT_SYMBOL(dev_pre_changeaddr_notify); 8875 8876 /** 8877 * dev_set_mac_address - Change Media Access Control Address 8878 * @dev: device 8879 * @sa: new address 8880 * @extack: netlink extended ack 8881 * 8882 * Change the hardware (MAC) address of the device 8883 */ 8884 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa, 8885 struct netlink_ext_ack *extack) 8886 { 8887 const struct net_device_ops *ops = dev->netdev_ops; 8888 int err; 8889 8890 if (!ops->ndo_set_mac_address) 8891 return -EOPNOTSUPP; 8892 if (sa->sa_family != dev->type) 8893 return -EINVAL; 8894 if (!netif_device_present(dev)) 8895 return -ENODEV; 8896 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack); 8897 if (err) 8898 return err; 8899 if (memcmp(dev->dev_addr, sa->sa_data, dev->addr_len)) { 8900 err = ops->ndo_set_mac_address(dev, sa); 8901 if (err) 8902 return err; 8903 } 8904 dev->addr_assign_type = NET_ADDR_SET; 8905 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); 8906 add_device_randomness(dev->dev_addr, dev->addr_len); 8907 return 0; 8908 } 8909 EXPORT_SYMBOL(dev_set_mac_address); 8910 8911 static DECLARE_RWSEM(dev_addr_sem); 8912 8913 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa, 8914 struct netlink_ext_ack *extack) 8915 { 8916 int ret; 8917 8918 down_write(&dev_addr_sem); 8919 ret = dev_set_mac_address(dev, sa, extack); 8920 up_write(&dev_addr_sem); 8921 return ret; 8922 } 8923 EXPORT_SYMBOL(dev_set_mac_address_user); 8924 8925 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name) 8926 { 8927 size_t size = sizeof(sa->sa_data_min); 8928 struct net_device *dev; 8929 int ret = 0; 8930 8931 down_read(&dev_addr_sem); 8932 rcu_read_lock(); 8933 8934 dev = dev_get_by_name_rcu(net, dev_name); 8935 if (!dev) { 8936 ret = -ENODEV; 8937 goto unlock; 8938 } 8939 if (!dev->addr_len) 8940 memset(sa->sa_data, 0, size); 8941 else 8942 memcpy(sa->sa_data, dev->dev_addr, 8943 min_t(size_t, size, dev->addr_len)); 8944 sa->sa_family = dev->type; 8945 8946 unlock: 8947 rcu_read_unlock(); 8948 up_read(&dev_addr_sem); 8949 return ret; 8950 } 8951 EXPORT_SYMBOL(dev_get_mac_address); 8952 8953 /** 8954 * dev_change_carrier - Change device carrier 8955 * @dev: device 8956 * @new_carrier: new value 8957 * 8958 * Change device carrier 8959 */ 8960 int dev_change_carrier(struct net_device *dev, bool new_carrier) 8961 { 8962 const struct net_device_ops *ops = dev->netdev_ops; 8963 8964 if (!ops->ndo_change_carrier) 8965 return -EOPNOTSUPP; 8966 if (!netif_device_present(dev)) 8967 return -ENODEV; 8968 return ops->ndo_change_carrier(dev, new_carrier); 8969 } 8970 8971 /** 8972 * dev_get_phys_port_id - Get device physical port ID 8973 * @dev: device 8974 * @ppid: port ID 8975 * 8976 * Get device physical port ID 8977 */ 8978 int dev_get_phys_port_id(struct net_device *dev, 8979 struct netdev_phys_item_id *ppid) 8980 { 8981 const struct net_device_ops *ops = dev->netdev_ops; 8982 8983 if (!ops->ndo_get_phys_port_id) 8984 return -EOPNOTSUPP; 8985 return ops->ndo_get_phys_port_id(dev, ppid); 8986 } 8987 8988 /** 8989 * dev_get_phys_port_name - Get device physical port name 8990 * @dev: device 8991 * @name: port name 8992 * @len: limit of bytes to copy to name 8993 * 8994 * Get device physical port name 8995 */ 8996 int dev_get_phys_port_name(struct net_device *dev, 8997 char *name, size_t len) 8998 { 8999 const struct net_device_ops *ops = dev->netdev_ops; 9000 int err; 9001 9002 if (ops->ndo_get_phys_port_name) { 9003 err = ops->ndo_get_phys_port_name(dev, name, len); 9004 if (err != -EOPNOTSUPP) 9005 return err; 9006 } 9007 return devlink_compat_phys_port_name_get(dev, name, len); 9008 } 9009 9010 /** 9011 * dev_get_port_parent_id - Get the device's port parent identifier 9012 * @dev: network device 9013 * @ppid: pointer to a storage for the port's parent identifier 9014 * @recurse: allow/disallow recursion to lower devices 9015 * 9016 * Get the devices's port parent identifier 9017 */ 9018 int dev_get_port_parent_id(struct net_device *dev, 9019 struct netdev_phys_item_id *ppid, 9020 bool recurse) 9021 { 9022 const struct net_device_ops *ops = dev->netdev_ops; 9023 struct netdev_phys_item_id first = { }; 9024 struct net_device *lower_dev; 9025 struct list_head *iter; 9026 int err; 9027 9028 if (ops->ndo_get_port_parent_id) { 9029 err = ops->ndo_get_port_parent_id(dev, ppid); 9030 if (err != -EOPNOTSUPP) 9031 return err; 9032 } 9033 9034 err = devlink_compat_switch_id_get(dev, ppid); 9035 if (!recurse || err != -EOPNOTSUPP) 9036 return err; 9037 9038 netdev_for_each_lower_dev(dev, lower_dev, iter) { 9039 err = dev_get_port_parent_id(lower_dev, ppid, true); 9040 if (err) 9041 break; 9042 if (!first.id_len) 9043 first = *ppid; 9044 else if (memcmp(&first, ppid, sizeof(*ppid))) 9045 return -EOPNOTSUPP; 9046 } 9047 9048 return err; 9049 } 9050 EXPORT_SYMBOL(dev_get_port_parent_id); 9051 9052 /** 9053 * netdev_port_same_parent_id - Indicate if two network devices have 9054 * the same port parent identifier 9055 * @a: first network device 9056 * @b: second network device 9057 */ 9058 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b) 9059 { 9060 struct netdev_phys_item_id a_id = { }; 9061 struct netdev_phys_item_id b_id = { }; 9062 9063 if (dev_get_port_parent_id(a, &a_id, true) || 9064 dev_get_port_parent_id(b, &b_id, true)) 9065 return false; 9066 9067 return netdev_phys_item_id_same(&a_id, &b_id); 9068 } 9069 EXPORT_SYMBOL(netdev_port_same_parent_id); 9070 9071 /** 9072 * dev_change_proto_down - set carrier according to proto_down. 9073 * 9074 * @dev: device 9075 * @proto_down: new value 9076 */ 9077 int dev_change_proto_down(struct net_device *dev, bool proto_down) 9078 { 9079 if (!(dev->priv_flags & IFF_CHANGE_PROTO_DOWN)) 9080 return -EOPNOTSUPP; 9081 if (!netif_device_present(dev)) 9082 return -ENODEV; 9083 if (proto_down) 9084 netif_carrier_off(dev); 9085 else 9086 netif_carrier_on(dev); 9087 dev->proto_down = proto_down; 9088 return 0; 9089 } 9090 9091 /** 9092 * dev_change_proto_down_reason - proto down reason 9093 * 9094 * @dev: device 9095 * @mask: proto down mask 9096 * @value: proto down value 9097 */ 9098 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask, 9099 u32 value) 9100 { 9101 int b; 9102 9103 if (!mask) { 9104 dev->proto_down_reason = value; 9105 } else { 9106 for_each_set_bit(b, &mask, 32) { 9107 if (value & (1 << b)) 9108 dev->proto_down_reason |= BIT(b); 9109 else 9110 dev->proto_down_reason &= ~BIT(b); 9111 } 9112 } 9113 } 9114 9115 struct bpf_xdp_link { 9116 struct bpf_link link; 9117 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */ 9118 int flags; 9119 }; 9120 9121 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags) 9122 { 9123 if (flags & XDP_FLAGS_HW_MODE) 9124 return XDP_MODE_HW; 9125 if (flags & XDP_FLAGS_DRV_MODE) 9126 return XDP_MODE_DRV; 9127 if (flags & XDP_FLAGS_SKB_MODE) 9128 return XDP_MODE_SKB; 9129 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB; 9130 } 9131 9132 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode) 9133 { 9134 switch (mode) { 9135 case XDP_MODE_SKB: 9136 return generic_xdp_install; 9137 case XDP_MODE_DRV: 9138 case XDP_MODE_HW: 9139 return dev->netdev_ops->ndo_bpf; 9140 default: 9141 return NULL; 9142 } 9143 } 9144 9145 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev, 9146 enum bpf_xdp_mode mode) 9147 { 9148 return dev->xdp_state[mode].link; 9149 } 9150 9151 static struct bpf_prog *dev_xdp_prog(struct net_device *dev, 9152 enum bpf_xdp_mode mode) 9153 { 9154 struct bpf_xdp_link *link = dev_xdp_link(dev, mode); 9155 9156 if (link) 9157 return link->link.prog; 9158 return dev->xdp_state[mode].prog; 9159 } 9160 9161 u8 dev_xdp_prog_count(struct net_device *dev) 9162 { 9163 u8 count = 0; 9164 int i; 9165 9166 for (i = 0; i < __MAX_XDP_MODE; i++) 9167 if (dev->xdp_state[i].prog || dev->xdp_state[i].link) 9168 count++; 9169 return count; 9170 } 9171 EXPORT_SYMBOL_GPL(dev_xdp_prog_count); 9172 9173 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode) 9174 { 9175 struct bpf_prog *prog = dev_xdp_prog(dev, mode); 9176 9177 return prog ? prog->aux->id : 0; 9178 } 9179 9180 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode, 9181 struct bpf_xdp_link *link) 9182 { 9183 dev->xdp_state[mode].link = link; 9184 dev->xdp_state[mode].prog = NULL; 9185 } 9186 9187 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode, 9188 struct bpf_prog *prog) 9189 { 9190 dev->xdp_state[mode].link = NULL; 9191 dev->xdp_state[mode].prog = prog; 9192 } 9193 9194 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode, 9195 bpf_op_t bpf_op, struct netlink_ext_ack *extack, 9196 u32 flags, struct bpf_prog *prog) 9197 { 9198 struct netdev_bpf xdp; 9199 int err; 9200 9201 memset(&xdp, 0, sizeof(xdp)); 9202 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG; 9203 xdp.extack = extack; 9204 xdp.flags = flags; 9205 xdp.prog = prog; 9206 9207 /* Drivers assume refcnt is already incremented (i.e, prog pointer is 9208 * "moved" into driver), so they don't increment it on their own, but 9209 * they do decrement refcnt when program is detached or replaced. 9210 * Given net_device also owns link/prog, we need to bump refcnt here 9211 * to prevent drivers from underflowing it. 9212 */ 9213 if (prog) 9214 bpf_prog_inc(prog); 9215 err = bpf_op(dev, &xdp); 9216 if (err) { 9217 if (prog) 9218 bpf_prog_put(prog); 9219 return err; 9220 } 9221 9222 if (mode != XDP_MODE_HW) 9223 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog); 9224 9225 return 0; 9226 } 9227 9228 static void dev_xdp_uninstall(struct net_device *dev) 9229 { 9230 struct bpf_xdp_link *link; 9231 struct bpf_prog *prog; 9232 enum bpf_xdp_mode mode; 9233 bpf_op_t bpf_op; 9234 9235 ASSERT_RTNL(); 9236 9237 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) { 9238 prog = dev_xdp_prog(dev, mode); 9239 if (!prog) 9240 continue; 9241 9242 bpf_op = dev_xdp_bpf_op(dev, mode); 9243 if (!bpf_op) 9244 continue; 9245 9246 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9247 9248 /* auto-detach link from net device */ 9249 link = dev_xdp_link(dev, mode); 9250 if (link) 9251 link->dev = NULL; 9252 else 9253 bpf_prog_put(prog); 9254 9255 dev_xdp_set_link(dev, mode, NULL); 9256 } 9257 } 9258 9259 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack, 9260 struct bpf_xdp_link *link, struct bpf_prog *new_prog, 9261 struct bpf_prog *old_prog, u32 flags) 9262 { 9263 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES); 9264 struct bpf_prog *cur_prog; 9265 struct net_device *upper; 9266 struct list_head *iter; 9267 enum bpf_xdp_mode mode; 9268 bpf_op_t bpf_op; 9269 int err; 9270 9271 ASSERT_RTNL(); 9272 9273 /* either link or prog attachment, never both */ 9274 if (link && (new_prog || old_prog)) 9275 return -EINVAL; 9276 /* link supports only XDP mode flags */ 9277 if (link && (flags & ~XDP_FLAGS_MODES)) { 9278 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment"); 9279 return -EINVAL; 9280 } 9281 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */ 9282 if (num_modes > 1) { 9283 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set"); 9284 return -EINVAL; 9285 } 9286 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */ 9287 if (!num_modes && dev_xdp_prog_count(dev) > 1) { 9288 NL_SET_ERR_MSG(extack, 9289 "More than one program loaded, unset mode is ambiguous"); 9290 return -EINVAL; 9291 } 9292 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */ 9293 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) { 9294 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified"); 9295 return -EINVAL; 9296 } 9297 9298 mode = dev_xdp_mode(dev, flags); 9299 /* can't replace attached link */ 9300 if (dev_xdp_link(dev, mode)) { 9301 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link"); 9302 return -EBUSY; 9303 } 9304 9305 /* don't allow if an upper device already has a program */ 9306 netdev_for_each_upper_dev_rcu(dev, upper, iter) { 9307 if (dev_xdp_prog_count(upper) > 0) { 9308 NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program"); 9309 return -EEXIST; 9310 } 9311 } 9312 9313 cur_prog = dev_xdp_prog(dev, mode); 9314 /* can't replace attached prog with link */ 9315 if (link && cur_prog) { 9316 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link"); 9317 return -EBUSY; 9318 } 9319 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) { 9320 NL_SET_ERR_MSG(extack, "Active program does not match expected"); 9321 return -EEXIST; 9322 } 9323 9324 /* put effective new program into new_prog */ 9325 if (link) 9326 new_prog = link->link.prog; 9327 9328 if (new_prog) { 9329 bool offload = mode == XDP_MODE_HW; 9330 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB 9331 ? XDP_MODE_DRV : XDP_MODE_SKB; 9332 9333 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) { 9334 NL_SET_ERR_MSG(extack, "XDP program already attached"); 9335 return -EBUSY; 9336 } 9337 if (!offload && dev_xdp_prog(dev, other_mode)) { 9338 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time"); 9339 return -EEXIST; 9340 } 9341 if (!offload && bpf_prog_is_offloaded(new_prog->aux)) { 9342 NL_SET_ERR_MSG(extack, "Using offloaded program without HW_MODE flag is not supported"); 9343 return -EINVAL; 9344 } 9345 if (bpf_prog_is_dev_bound(new_prog->aux) && !bpf_offload_dev_match(new_prog, dev)) { 9346 NL_SET_ERR_MSG(extack, "Program bound to different device"); 9347 return -EINVAL; 9348 } 9349 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) { 9350 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device"); 9351 return -EINVAL; 9352 } 9353 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) { 9354 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device"); 9355 return -EINVAL; 9356 } 9357 } 9358 9359 /* don't call drivers if the effective program didn't change */ 9360 if (new_prog != cur_prog) { 9361 bpf_op = dev_xdp_bpf_op(dev, mode); 9362 if (!bpf_op) { 9363 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode"); 9364 return -EOPNOTSUPP; 9365 } 9366 9367 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog); 9368 if (err) 9369 return err; 9370 } 9371 9372 if (link) 9373 dev_xdp_set_link(dev, mode, link); 9374 else 9375 dev_xdp_set_prog(dev, mode, new_prog); 9376 if (cur_prog) 9377 bpf_prog_put(cur_prog); 9378 9379 return 0; 9380 } 9381 9382 static int dev_xdp_attach_link(struct net_device *dev, 9383 struct netlink_ext_ack *extack, 9384 struct bpf_xdp_link *link) 9385 { 9386 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags); 9387 } 9388 9389 static int dev_xdp_detach_link(struct net_device *dev, 9390 struct netlink_ext_ack *extack, 9391 struct bpf_xdp_link *link) 9392 { 9393 enum bpf_xdp_mode mode; 9394 bpf_op_t bpf_op; 9395 9396 ASSERT_RTNL(); 9397 9398 mode = dev_xdp_mode(dev, link->flags); 9399 if (dev_xdp_link(dev, mode) != link) 9400 return -EINVAL; 9401 9402 bpf_op = dev_xdp_bpf_op(dev, mode); 9403 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9404 dev_xdp_set_link(dev, mode, NULL); 9405 return 0; 9406 } 9407 9408 static void bpf_xdp_link_release(struct bpf_link *link) 9409 { 9410 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9411 9412 rtnl_lock(); 9413 9414 /* if racing with net_device's tear down, xdp_link->dev might be 9415 * already NULL, in which case link was already auto-detached 9416 */ 9417 if (xdp_link->dev) { 9418 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link)); 9419 xdp_link->dev = NULL; 9420 } 9421 9422 rtnl_unlock(); 9423 } 9424 9425 static int bpf_xdp_link_detach(struct bpf_link *link) 9426 { 9427 bpf_xdp_link_release(link); 9428 return 0; 9429 } 9430 9431 static void bpf_xdp_link_dealloc(struct bpf_link *link) 9432 { 9433 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9434 9435 kfree(xdp_link); 9436 } 9437 9438 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link, 9439 struct seq_file *seq) 9440 { 9441 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9442 u32 ifindex = 0; 9443 9444 rtnl_lock(); 9445 if (xdp_link->dev) 9446 ifindex = xdp_link->dev->ifindex; 9447 rtnl_unlock(); 9448 9449 seq_printf(seq, "ifindex:\t%u\n", ifindex); 9450 } 9451 9452 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link, 9453 struct bpf_link_info *info) 9454 { 9455 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9456 u32 ifindex = 0; 9457 9458 rtnl_lock(); 9459 if (xdp_link->dev) 9460 ifindex = xdp_link->dev->ifindex; 9461 rtnl_unlock(); 9462 9463 info->xdp.ifindex = ifindex; 9464 return 0; 9465 } 9466 9467 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog, 9468 struct bpf_prog *old_prog) 9469 { 9470 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9471 enum bpf_xdp_mode mode; 9472 bpf_op_t bpf_op; 9473 int err = 0; 9474 9475 rtnl_lock(); 9476 9477 /* link might have been auto-released already, so fail */ 9478 if (!xdp_link->dev) { 9479 err = -ENOLINK; 9480 goto out_unlock; 9481 } 9482 9483 if (old_prog && link->prog != old_prog) { 9484 err = -EPERM; 9485 goto out_unlock; 9486 } 9487 old_prog = link->prog; 9488 if (old_prog->type != new_prog->type || 9489 old_prog->expected_attach_type != new_prog->expected_attach_type) { 9490 err = -EINVAL; 9491 goto out_unlock; 9492 } 9493 9494 if (old_prog == new_prog) { 9495 /* no-op, don't disturb drivers */ 9496 bpf_prog_put(new_prog); 9497 goto out_unlock; 9498 } 9499 9500 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags); 9501 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode); 9502 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL, 9503 xdp_link->flags, new_prog); 9504 if (err) 9505 goto out_unlock; 9506 9507 old_prog = xchg(&link->prog, new_prog); 9508 bpf_prog_put(old_prog); 9509 9510 out_unlock: 9511 rtnl_unlock(); 9512 return err; 9513 } 9514 9515 static const struct bpf_link_ops bpf_xdp_link_lops = { 9516 .release = bpf_xdp_link_release, 9517 .dealloc = bpf_xdp_link_dealloc, 9518 .detach = bpf_xdp_link_detach, 9519 .show_fdinfo = bpf_xdp_link_show_fdinfo, 9520 .fill_link_info = bpf_xdp_link_fill_link_info, 9521 .update_prog = bpf_xdp_link_update, 9522 }; 9523 9524 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog) 9525 { 9526 struct net *net = current->nsproxy->net_ns; 9527 struct bpf_link_primer link_primer; 9528 struct netlink_ext_ack extack = {}; 9529 struct bpf_xdp_link *link; 9530 struct net_device *dev; 9531 int err, fd; 9532 9533 rtnl_lock(); 9534 dev = dev_get_by_index(net, attr->link_create.target_ifindex); 9535 if (!dev) { 9536 rtnl_unlock(); 9537 return -EINVAL; 9538 } 9539 9540 link = kzalloc(sizeof(*link), GFP_USER); 9541 if (!link) { 9542 err = -ENOMEM; 9543 goto unlock; 9544 } 9545 9546 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog); 9547 link->dev = dev; 9548 link->flags = attr->link_create.flags; 9549 9550 err = bpf_link_prime(&link->link, &link_primer); 9551 if (err) { 9552 kfree(link); 9553 goto unlock; 9554 } 9555 9556 err = dev_xdp_attach_link(dev, &extack, link); 9557 rtnl_unlock(); 9558 9559 if (err) { 9560 link->dev = NULL; 9561 bpf_link_cleanup(&link_primer); 9562 trace_bpf_xdp_link_attach_failed(extack._msg); 9563 goto out_put_dev; 9564 } 9565 9566 fd = bpf_link_settle(&link_primer); 9567 /* link itself doesn't hold dev's refcnt to not complicate shutdown */ 9568 dev_put(dev); 9569 return fd; 9570 9571 unlock: 9572 rtnl_unlock(); 9573 9574 out_put_dev: 9575 dev_put(dev); 9576 return err; 9577 } 9578 9579 /** 9580 * dev_change_xdp_fd - set or clear a bpf program for a device rx path 9581 * @dev: device 9582 * @extack: netlink extended ack 9583 * @fd: new program fd or negative value to clear 9584 * @expected_fd: old program fd that userspace expects to replace or clear 9585 * @flags: xdp-related flags 9586 * 9587 * Set or clear a bpf program for a device 9588 */ 9589 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, 9590 int fd, int expected_fd, u32 flags) 9591 { 9592 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags); 9593 struct bpf_prog *new_prog = NULL, *old_prog = NULL; 9594 int err; 9595 9596 ASSERT_RTNL(); 9597 9598 if (fd >= 0) { 9599 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP, 9600 mode != XDP_MODE_SKB); 9601 if (IS_ERR(new_prog)) 9602 return PTR_ERR(new_prog); 9603 } 9604 9605 if (expected_fd >= 0) { 9606 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP, 9607 mode != XDP_MODE_SKB); 9608 if (IS_ERR(old_prog)) { 9609 err = PTR_ERR(old_prog); 9610 old_prog = NULL; 9611 goto err_out; 9612 } 9613 } 9614 9615 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags); 9616 9617 err_out: 9618 if (err && new_prog) 9619 bpf_prog_put(new_prog); 9620 if (old_prog) 9621 bpf_prog_put(old_prog); 9622 return err; 9623 } 9624 9625 /** 9626 * dev_index_reserve() - allocate an ifindex in a namespace 9627 * @net: the applicable net namespace 9628 * @ifindex: requested ifindex, pass %0 to get one allocated 9629 * 9630 * Allocate a ifindex for a new device. Caller must either use the ifindex 9631 * to store the device (via list_netdevice()) or call dev_index_release() 9632 * to give the index up. 9633 * 9634 * Return: a suitable unique value for a new device interface number or -errno. 9635 */ 9636 static int dev_index_reserve(struct net *net, u32 ifindex) 9637 { 9638 int err; 9639 9640 if (ifindex > INT_MAX) { 9641 DEBUG_NET_WARN_ON_ONCE(1); 9642 return -EINVAL; 9643 } 9644 9645 if (!ifindex) 9646 err = xa_alloc_cyclic(&net->dev_by_index, &ifindex, NULL, 9647 xa_limit_31b, &net->ifindex, GFP_KERNEL); 9648 else 9649 err = xa_insert(&net->dev_by_index, ifindex, NULL, GFP_KERNEL); 9650 if (err < 0) 9651 return err; 9652 9653 return ifindex; 9654 } 9655 9656 static void dev_index_release(struct net *net, int ifindex) 9657 { 9658 /* Expect only unused indexes, unlist_netdevice() removes the used */ 9659 WARN_ON(xa_erase(&net->dev_by_index, ifindex)); 9660 } 9661 9662 /* Delayed registration/unregisteration */ 9663 LIST_HEAD(net_todo_list); 9664 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); 9665 9666 static void net_set_todo(struct net_device *dev) 9667 { 9668 list_add_tail(&dev->todo_list, &net_todo_list); 9669 atomic_inc(&dev_net(dev)->dev_unreg_count); 9670 } 9671 9672 static netdev_features_t netdev_sync_upper_features(struct net_device *lower, 9673 struct net_device *upper, netdev_features_t features) 9674 { 9675 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9676 netdev_features_t feature; 9677 int feature_bit; 9678 9679 for_each_netdev_feature(upper_disables, feature_bit) { 9680 feature = __NETIF_F_BIT(feature_bit); 9681 if (!(upper->wanted_features & feature) 9682 && (features & feature)) { 9683 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", 9684 &feature, upper->name); 9685 features &= ~feature; 9686 } 9687 } 9688 9689 return features; 9690 } 9691 9692 static void netdev_sync_lower_features(struct net_device *upper, 9693 struct net_device *lower, netdev_features_t features) 9694 { 9695 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9696 netdev_features_t feature; 9697 int feature_bit; 9698 9699 for_each_netdev_feature(upper_disables, feature_bit) { 9700 feature = __NETIF_F_BIT(feature_bit); 9701 if (!(features & feature) && (lower->features & feature)) { 9702 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", 9703 &feature, lower->name); 9704 lower->wanted_features &= ~feature; 9705 __netdev_update_features(lower); 9706 9707 if (unlikely(lower->features & feature)) 9708 netdev_WARN(upper, "failed to disable %pNF on %s!\n", 9709 &feature, lower->name); 9710 else 9711 netdev_features_change(lower); 9712 } 9713 } 9714 } 9715 9716 static netdev_features_t netdev_fix_features(struct net_device *dev, 9717 netdev_features_t features) 9718 { 9719 /* Fix illegal checksum combinations */ 9720 if ((features & NETIF_F_HW_CSUM) && 9721 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { 9722 netdev_warn(dev, "mixed HW and IP checksum settings.\n"); 9723 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); 9724 } 9725 9726 /* TSO requires that SG is present as well. */ 9727 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { 9728 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); 9729 features &= ~NETIF_F_ALL_TSO; 9730 } 9731 9732 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && 9733 !(features & NETIF_F_IP_CSUM)) { 9734 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); 9735 features &= ~NETIF_F_TSO; 9736 features &= ~NETIF_F_TSO_ECN; 9737 } 9738 9739 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && 9740 !(features & NETIF_F_IPV6_CSUM)) { 9741 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); 9742 features &= ~NETIF_F_TSO6; 9743 } 9744 9745 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ 9746 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) 9747 features &= ~NETIF_F_TSO_MANGLEID; 9748 9749 /* TSO ECN requires that TSO is present as well. */ 9750 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) 9751 features &= ~NETIF_F_TSO_ECN; 9752 9753 /* Software GSO depends on SG. */ 9754 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { 9755 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); 9756 features &= ~NETIF_F_GSO; 9757 } 9758 9759 /* GSO partial features require GSO partial be set */ 9760 if ((features & dev->gso_partial_features) && 9761 !(features & NETIF_F_GSO_PARTIAL)) { 9762 netdev_dbg(dev, 9763 "Dropping partially supported GSO features since no GSO partial.\n"); 9764 features &= ~dev->gso_partial_features; 9765 } 9766 9767 if (!(features & NETIF_F_RXCSUM)) { 9768 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet 9769 * successfully merged by hardware must also have the 9770 * checksum verified by hardware. If the user does not 9771 * want to enable RXCSUM, logically, we should disable GRO_HW. 9772 */ 9773 if (features & NETIF_F_GRO_HW) { 9774 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n"); 9775 features &= ~NETIF_F_GRO_HW; 9776 } 9777 } 9778 9779 /* LRO/HW-GRO features cannot be combined with RX-FCS */ 9780 if (features & NETIF_F_RXFCS) { 9781 if (features & NETIF_F_LRO) { 9782 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n"); 9783 features &= ~NETIF_F_LRO; 9784 } 9785 9786 if (features & NETIF_F_GRO_HW) { 9787 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n"); 9788 features &= ~NETIF_F_GRO_HW; 9789 } 9790 } 9791 9792 if ((features & NETIF_F_GRO_HW) && (features & NETIF_F_LRO)) { 9793 netdev_dbg(dev, "Dropping LRO feature since HW-GRO is requested.\n"); 9794 features &= ~NETIF_F_LRO; 9795 } 9796 9797 if (features & NETIF_F_HW_TLS_TX) { 9798 bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) == 9799 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM); 9800 bool hw_csum = features & NETIF_F_HW_CSUM; 9801 9802 if (!ip_csum && !hw_csum) { 9803 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n"); 9804 features &= ~NETIF_F_HW_TLS_TX; 9805 } 9806 } 9807 9808 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) { 9809 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n"); 9810 features &= ~NETIF_F_HW_TLS_RX; 9811 } 9812 9813 return features; 9814 } 9815 9816 int __netdev_update_features(struct net_device *dev) 9817 { 9818 struct net_device *upper, *lower; 9819 netdev_features_t features; 9820 struct list_head *iter; 9821 int err = -1; 9822 9823 ASSERT_RTNL(); 9824 9825 features = netdev_get_wanted_features(dev); 9826 9827 if (dev->netdev_ops->ndo_fix_features) 9828 features = dev->netdev_ops->ndo_fix_features(dev, features); 9829 9830 /* driver might be less strict about feature dependencies */ 9831 features = netdev_fix_features(dev, features); 9832 9833 /* some features can't be enabled if they're off on an upper device */ 9834 netdev_for_each_upper_dev_rcu(dev, upper, iter) 9835 features = netdev_sync_upper_features(dev, upper, features); 9836 9837 if (dev->features == features) 9838 goto sync_lower; 9839 9840 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", 9841 &dev->features, &features); 9842 9843 if (dev->netdev_ops->ndo_set_features) 9844 err = dev->netdev_ops->ndo_set_features(dev, features); 9845 else 9846 err = 0; 9847 9848 if (unlikely(err < 0)) { 9849 netdev_err(dev, 9850 "set_features() failed (%d); wanted %pNF, left %pNF\n", 9851 err, &features, &dev->features); 9852 /* return non-0 since some features might have changed and 9853 * it's better to fire a spurious notification than miss it 9854 */ 9855 return -1; 9856 } 9857 9858 sync_lower: 9859 /* some features must be disabled on lower devices when disabled 9860 * on an upper device (think: bonding master or bridge) 9861 */ 9862 netdev_for_each_lower_dev(dev, lower, iter) 9863 netdev_sync_lower_features(dev, lower, features); 9864 9865 if (!err) { 9866 netdev_features_t diff = features ^ dev->features; 9867 9868 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) { 9869 /* udp_tunnel_{get,drop}_rx_info both need 9870 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the 9871 * device, or they won't do anything. 9872 * Thus we need to update dev->features 9873 * *before* calling udp_tunnel_get_rx_info, 9874 * but *after* calling udp_tunnel_drop_rx_info. 9875 */ 9876 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) { 9877 dev->features = features; 9878 udp_tunnel_get_rx_info(dev); 9879 } else { 9880 udp_tunnel_drop_rx_info(dev); 9881 } 9882 } 9883 9884 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) { 9885 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) { 9886 dev->features = features; 9887 err |= vlan_get_rx_ctag_filter_info(dev); 9888 } else { 9889 vlan_drop_rx_ctag_filter_info(dev); 9890 } 9891 } 9892 9893 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) { 9894 if (features & NETIF_F_HW_VLAN_STAG_FILTER) { 9895 dev->features = features; 9896 err |= vlan_get_rx_stag_filter_info(dev); 9897 } else { 9898 vlan_drop_rx_stag_filter_info(dev); 9899 } 9900 } 9901 9902 dev->features = features; 9903 } 9904 9905 return err < 0 ? 0 : 1; 9906 } 9907 9908 /** 9909 * netdev_update_features - recalculate device features 9910 * @dev: the device to check 9911 * 9912 * Recalculate dev->features set and send notifications if it 9913 * has changed. Should be called after driver or hardware dependent 9914 * conditions might have changed that influence the features. 9915 */ 9916 void netdev_update_features(struct net_device *dev) 9917 { 9918 if (__netdev_update_features(dev)) 9919 netdev_features_change(dev); 9920 } 9921 EXPORT_SYMBOL(netdev_update_features); 9922 9923 /** 9924 * netdev_change_features - recalculate device features 9925 * @dev: the device to check 9926 * 9927 * Recalculate dev->features set and send notifications even 9928 * if they have not changed. Should be called instead of 9929 * netdev_update_features() if also dev->vlan_features might 9930 * have changed to allow the changes to be propagated to stacked 9931 * VLAN devices. 9932 */ 9933 void netdev_change_features(struct net_device *dev) 9934 { 9935 __netdev_update_features(dev); 9936 netdev_features_change(dev); 9937 } 9938 EXPORT_SYMBOL(netdev_change_features); 9939 9940 /** 9941 * netif_stacked_transfer_operstate - transfer operstate 9942 * @rootdev: the root or lower level device to transfer state from 9943 * @dev: the device to transfer operstate to 9944 * 9945 * Transfer operational state from root to device. This is normally 9946 * called when a stacking relationship exists between the root 9947 * device and the device(a leaf device). 9948 */ 9949 void netif_stacked_transfer_operstate(const struct net_device *rootdev, 9950 struct net_device *dev) 9951 { 9952 if (rootdev->operstate == IF_OPER_DORMANT) 9953 netif_dormant_on(dev); 9954 else 9955 netif_dormant_off(dev); 9956 9957 if (rootdev->operstate == IF_OPER_TESTING) 9958 netif_testing_on(dev); 9959 else 9960 netif_testing_off(dev); 9961 9962 if (netif_carrier_ok(rootdev)) 9963 netif_carrier_on(dev); 9964 else 9965 netif_carrier_off(dev); 9966 } 9967 EXPORT_SYMBOL(netif_stacked_transfer_operstate); 9968 9969 static int netif_alloc_rx_queues(struct net_device *dev) 9970 { 9971 unsigned int i, count = dev->num_rx_queues; 9972 struct netdev_rx_queue *rx; 9973 size_t sz = count * sizeof(*rx); 9974 int err = 0; 9975 9976 BUG_ON(count < 1); 9977 9978 rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 9979 if (!rx) 9980 return -ENOMEM; 9981 9982 dev->_rx = rx; 9983 9984 for (i = 0; i < count; i++) { 9985 rx[i].dev = dev; 9986 9987 /* XDP RX-queue setup */ 9988 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0); 9989 if (err < 0) 9990 goto err_rxq_info; 9991 } 9992 return 0; 9993 9994 err_rxq_info: 9995 /* Rollback successful reg's and free other resources */ 9996 while (i--) 9997 xdp_rxq_info_unreg(&rx[i].xdp_rxq); 9998 kvfree(dev->_rx); 9999 dev->_rx = NULL; 10000 return err; 10001 } 10002 10003 static void netif_free_rx_queues(struct net_device *dev) 10004 { 10005 unsigned int i, count = dev->num_rx_queues; 10006 10007 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */ 10008 if (!dev->_rx) 10009 return; 10010 10011 for (i = 0; i < count; i++) 10012 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq); 10013 10014 kvfree(dev->_rx); 10015 } 10016 10017 static void netdev_init_one_queue(struct net_device *dev, 10018 struct netdev_queue *queue, void *_unused) 10019 { 10020 /* Initialize queue lock */ 10021 spin_lock_init(&queue->_xmit_lock); 10022 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); 10023 queue->xmit_lock_owner = -1; 10024 netdev_queue_numa_node_write(queue, NUMA_NO_NODE); 10025 queue->dev = dev; 10026 #ifdef CONFIG_BQL 10027 dql_init(&queue->dql, HZ); 10028 #endif 10029 } 10030 10031 static void netif_free_tx_queues(struct net_device *dev) 10032 { 10033 kvfree(dev->_tx); 10034 } 10035 10036 static int netif_alloc_netdev_queues(struct net_device *dev) 10037 { 10038 unsigned int count = dev->num_tx_queues; 10039 struct netdev_queue *tx; 10040 size_t sz = count * sizeof(*tx); 10041 10042 if (count < 1 || count > 0xffff) 10043 return -EINVAL; 10044 10045 tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10046 if (!tx) 10047 return -ENOMEM; 10048 10049 dev->_tx = tx; 10050 10051 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); 10052 spin_lock_init(&dev->tx_global_lock); 10053 10054 return 0; 10055 } 10056 10057 void netif_tx_stop_all_queues(struct net_device *dev) 10058 { 10059 unsigned int i; 10060 10061 for (i = 0; i < dev->num_tx_queues; i++) { 10062 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 10063 10064 netif_tx_stop_queue(txq); 10065 } 10066 } 10067 EXPORT_SYMBOL(netif_tx_stop_all_queues); 10068 10069 static int netdev_do_alloc_pcpu_stats(struct net_device *dev) 10070 { 10071 void __percpu *v; 10072 10073 /* Drivers implementing ndo_get_peer_dev must support tstat 10074 * accounting, so that skb_do_redirect() can bump the dev's 10075 * RX stats upon network namespace switch. 10076 */ 10077 if (dev->netdev_ops->ndo_get_peer_dev && 10078 dev->pcpu_stat_type != NETDEV_PCPU_STAT_TSTATS) 10079 return -EOPNOTSUPP; 10080 10081 switch (dev->pcpu_stat_type) { 10082 case NETDEV_PCPU_STAT_NONE: 10083 return 0; 10084 case NETDEV_PCPU_STAT_LSTATS: 10085 v = dev->lstats = netdev_alloc_pcpu_stats(struct pcpu_lstats); 10086 break; 10087 case NETDEV_PCPU_STAT_TSTATS: 10088 v = dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats); 10089 break; 10090 case NETDEV_PCPU_STAT_DSTATS: 10091 v = dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats); 10092 break; 10093 default: 10094 return -EINVAL; 10095 } 10096 10097 return v ? 0 : -ENOMEM; 10098 } 10099 10100 static void netdev_do_free_pcpu_stats(struct net_device *dev) 10101 { 10102 switch (dev->pcpu_stat_type) { 10103 case NETDEV_PCPU_STAT_NONE: 10104 return; 10105 case NETDEV_PCPU_STAT_LSTATS: 10106 free_percpu(dev->lstats); 10107 break; 10108 case NETDEV_PCPU_STAT_TSTATS: 10109 free_percpu(dev->tstats); 10110 break; 10111 case NETDEV_PCPU_STAT_DSTATS: 10112 free_percpu(dev->dstats); 10113 break; 10114 } 10115 } 10116 10117 /** 10118 * register_netdevice() - register a network device 10119 * @dev: device to register 10120 * 10121 * Take a prepared network device structure and make it externally accessible. 10122 * A %NETDEV_REGISTER message is sent to the netdev notifier chain. 10123 * Callers must hold the rtnl lock - you may want register_netdev() 10124 * instead of this. 10125 */ 10126 int register_netdevice(struct net_device *dev) 10127 { 10128 int ret; 10129 struct net *net = dev_net(dev); 10130 10131 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE < 10132 NETDEV_FEATURE_COUNT); 10133 BUG_ON(dev_boot_phase); 10134 ASSERT_RTNL(); 10135 10136 might_sleep(); 10137 10138 /* When net_device's are persistent, this will be fatal. */ 10139 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); 10140 BUG_ON(!net); 10141 10142 ret = ethtool_check_ops(dev->ethtool_ops); 10143 if (ret) 10144 return ret; 10145 10146 spin_lock_init(&dev->addr_list_lock); 10147 netdev_set_addr_lockdep_class(dev); 10148 10149 ret = dev_get_valid_name(net, dev, dev->name); 10150 if (ret < 0) 10151 goto out; 10152 10153 ret = -ENOMEM; 10154 dev->name_node = netdev_name_node_head_alloc(dev); 10155 if (!dev->name_node) 10156 goto out; 10157 10158 /* Init, if this function is available */ 10159 if (dev->netdev_ops->ndo_init) { 10160 ret = dev->netdev_ops->ndo_init(dev); 10161 if (ret) { 10162 if (ret > 0) 10163 ret = -EIO; 10164 goto err_free_name; 10165 } 10166 } 10167 10168 if (((dev->hw_features | dev->features) & 10169 NETIF_F_HW_VLAN_CTAG_FILTER) && 10170 (!dev->netdev_ops->ndo_vlan_rx_add_vid || 10171 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { 10172 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); 10173 ret = -EINVAL; 10174 goto err_uninit; 10175 } 10176 10177 ret = netdev_do_alloc_pcpu_stats(dev); 10178 if (ret) 10179 goto err_uninit; 10180 10181 ret = dev_index_reserve(net, dev->ifindex); 10182 if (ret < 0) 10183 goto err_free_pcpu; 10184 dev->ifindex = ret; 10185 10186 /* Transfer changeable features to wanted_features and enable 10187 * software offloads (GSO and GRO). 10188 */ 10189 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF); 10190 dev->features |= NETIF_F_SOFT_FEATURES; 10191 10192 if (dev->udp_tunnel_nic_info) { 10193 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10194 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10195 } 10196 10197 dev->wanted_features = dev->features & dev->hw_features; 10198 10199 if (!(dev->flags & IFF_LOOPBACK)) 10200 dev->hw_features |= NETIF_F_NOCACHE_COPY; 10201 10202 /* If IPv4 TCP segmentation offload is supported we should also 10203 * allow the device to enable segmenting the frame with the option 10204 * of ignoring a static IP ID value. This doesn't enable the 10205 * feature itself but allows the user to enable it later. 10206 */ 10207 if (dev->hw_features & NETIF_F_TSO) 10208 dev->hw_features |= NETIF_F_TSO_MANGLEID; 10209 if (dev->vlan_features & NETIF_F_TSO) 10210 dev->vlan_features |= NETIF_F_TSO_MANGLEID; 10211 if (dev->mpls_features & NETIF_F_TSO) 10212 dev->mpls_features |= NETIF_F_TSO_MANGLEID; 10213 if (dev->hw_enc_features & NETIF_F_TSO) 10214 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; 10215 10216 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. 10217 */ 10218 dev->vlan_features |= NETIF_F_HIGHDMA; 10219 10220 /* Make NETIF_F_SG inheritable to tunnel devices. 10221 */ 10222 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; 10223 10224 /* Make NETIF_F_SG inheritable to MPLS. 10225 */ 10226 dev->mpls_features |= NETIF_F_SG; 10227 10228 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); 10229 ret = notifier_to_errno(ret); 10230 if (ret) 10231 goto err_ifindex_release; 10232 10233 ret = netdev_register_kobject(dev); 10234 write_lock(&dev_base_lock); 10235 dev->reg_state = ret ? NETREG_UNREGISTERED : NETREG_REGISTERED; 10236 write_unlock(&dev_base_lock); 10237 if (ret) 10238 goto err_uninit_notify; 10239 10240 __netdev_update_features(dev); 10241 10242 /* 10243 * Default initial state at registry is that the 10244 * device is present. 10245 */ 10246 10247 set_bit(__LINK_STATE_PRESENT, &dev->state); 10248 10249 linkwatch_init_dev(dev); 10250 10251 dev_init_scheduler(dev); 10252 10253 netdev_hold(dev, &dev->dev_registered_tracker, GFP_KERNEL); 10254 list_netdevice(dev); 10255 10256 add_device_randomness(dev->dev_addr, dev->addr_len); 10257 10258 /* If the device has permanent device address, driver should 10259 * set dev_addr and also addr_assign_type should be set to 10260 * NET_ADDR_PERM (default value). 10261 */ 10262 if (dev->addr_assign_type == NET_ADDR_PERM) 10263 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); 10264 10265 /* Notify protocols, that a new device appeared. */ 10266 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); 10267 ret = notifier_to_errno(ret); 10268 if (ret) { 10269 /* Expect explicit free_netdev() on failure */ 10270 dev->needs_free_netdev = false; 10271 unregister_netdevice_queue(dev, NULL); 10272 goto out; 10273 } 10274 /* 10275 * Prevent userspace races by waiting until the network 10276 * device is fully setup before sending notifications. 10277 */ 10278 if (!dev->rtnl_link_ops || 10279 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 10280 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL); 10281 10282 out: 10283 return ret; 10284 10285 err_uninit_notify: 10286 call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev); 10287 err_ifindex_release: 10288 dev_index_release(net, dev->ifindex); 10289 err_free_pcpu: 10290 netdev_do_free_pcpu_stats(dev); 10291 err_uninit: 10292 if (dev->netdev_ops->ndo_uninit) 10293 dev->netdev_ops->ndo_uninit(dev); 10294 if (dev->priv_destructor) 10295 dev->priv_destructor(dev); 10296 err_free_name: 10297 netdev_name_node_free(dev->name_node); 10298 goto out; 10299 } 10300 EXPORT_SYMBOL(register_netdevice); 10301 10302 /** 10303 * init_dummy_netdev - init a dummy network device for NAPI 10304 * @dev: device to init 10305 * 10306 * This takes a network device structure and initialize the minimum 10307 * amount of fields so it can be used to schedule NAPI polls without 10308 * registering a full blown interface. This is to be used by drivers 10309 * that need to tie several hardware interfaces to a single NAPI 10310 * poll scheduler due to HW limitations. 10311 */ 10312 int init_dummy_netdev(struct net_device *dev) 10313 { 10314 /* Clear everything. Note we don't initialize spinlocks 10315 * are they aren't supposed to be taken by any of the 10316 * NAPI code and this dummy netdev is supposed to be 10317 * only ever used for NAPI polls 10318 */ 10319 memset(dev, 0, sizeof(struct net_device)); 10320 10321 /* make sure we BUG if trying to hit standard 10322 * register/unregister code path 10323 */ 10324 dev->reg_state = NETREG_DUMMY; 10325 10326 /* NAPI wants this */ 10327 INIT_LIST_HEAD(&dev->napi_list); 10328 10329 /* a dummy interface is started by default */ 10330 set_bit(__LINK_STATE_PRESENT, &dev->state); 10331 set_bit(__LINK_STATE_START, &dev->state); 10332 10333 /* napi_busy_loop stats accounting wants this */ 10334 dev_net_set(dev, &init_net); 10335 10336 /* Note : We dont allocate pcpu_refcnt for dummy devices, 10337 * because users of this 'device' dont need to change 10338 * its refcount. 10339 */ 10340 10341 return 0; 10342 } 10343 EXPORT_SYMBOL_GPL(init_dummy_netdev); 10344 10345 10346 /** 10347 * register_netdev - register a network device 10348 * @dev: device to register 10349 * 10350 * Take a completed network device structure and add it to the kernel 10351 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 10352 * chain. 0 is returned on success. A negative errno code is returned 10353 * on a failure to set up the device, or if the name is a duplicate. 10354 * 10355 * This is a wrapper around register_netdevice that takes the rtnl semaphore 10356 * and expands the device name if you passed a format string to 10357 * alloc_netdev. 10358 */ 10359 int register_netdev(struct net_device *dev) 10360 { 10361 int err; 10362 10363 if (rtnl_lock_killable()) 10364 return -EINTR; 10365 err = register_netdevice(dev); 10366 rtnl_unlock(); 10367 return err; 10368 } 10369 EXPORT_SYMBOL(register_netdev); 10370 10371 int netdev_refcnt_read(const struct net_device *dev) 10372 { 10373 #ifdef CONFIG_PCPU_DEV_REFCNT 10374 int i, refcnt = 0; 10375 10376 for_each_possible_cpu(i) 10377 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); 10378 return refcnt; 10379 #else 10380 return refcount_read(&dev->dev_refcnt); 10381 #endif 10382 } 10383 EXPORT_SYMBOL(netdev_refcnt_read); 10384 10385 int netdev_unregister_timeout_secs __read_mostly = 10; 10386 10387 #define WAIT_REFS_MIN_MSECS 1 10388 #define WAIT_REFS_MAX_MSECS 250 10389 /** 10390 * netdev_wait_allrefs_any - wait until all references are gone. 10391 * @list: list of net_devices to wait on 10392 * 10393 * This is called when unregistering network devices. 10394 * 10395 * Any protocol or device that holds a reference should register 10396 * for netdevice notification, and cleanup and put back the 10397 * reference if they receive an UNREGISTER event. 10398 * We can get stuck here if buggy protocols don't correctly 10399 * call dev_put. 10400 */ 10401 static struct net_device *netdev_wait_allrefs_any(struct list_head *list) 10402 { 10403 unsigned long rebroadcast_time, warning_time; 10404 struct net_device *dev; 10405 int wait = 0; 10406 10407 rebroadcast_time = warning_time = jiffies; 10408 10409 list_for_each_entry(dev, list, todo_list) 10410 if (netdev_refcnt_read(dev) == 1) 10411 return dev; 10412 10413 while (true) { 10414 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { 10415 rtnl_lock(); 10416 10417 /* Rebroadcast unregister notification */ 10418 list_for_each_entry(dev, list, todo_list) 10419 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10420 10421 __rtnl_unlock(); 10422 rcu_barrier(); 10423 rtnl_lock(); 10424 10425 list_for_each_entry(dev, list, todo_list) 10426 if (test_bit(__LINK_STATE_LINKWATCH_PENDING, 10427 &dev->state)) { 10428 /* We must not have linkwatch events 10429 * pending on unregister. If this 10430 * happens, we simply run the queue 10431 * unscheduled, resulting in a noop 10432 * for this device. 10433 */ 10434 linkwatch_run_queue(); 10435 break; 10436 } 10437 10438 __rtnl_unlock(); 10439 10440 rebroadcast_time = jiffies; 10441 } 10442 10443 rcu_barrier(); 10444 10445 if (!wait) { 10446 wait = WAIT_REFS_MIN_MSECS; 10447 } else { 10448 msleep(wait); 10449 wait = min(wait << 1, WAIT_REFS_MAX_MSECS); 10450 } 10451 10452 list_for_each_entry(dev, list, todo_list) 10453 if (netdev_refcnt_read(dev) == 1) 10454 return dev; 10455 10456 if (time_after(jiffies, warning_time + 10457 READ_ONCE(netdev_unregister_timeout_secs) * HZ)) { 10458 list_for_each_entry(dev, list, todo_list) { 10459 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", 10460 dev->name, netdev_refcnt_read(dev)); 10461 ref_tracker_dir_print(&dev->refcnt_tracker, 10); 10462 } 10463 10464 warning_time = jiffies; 10465 } 10466 } 10467 } 10468 10469 /* The sequence is: 10470 * 10471 * rtnl_lock(); 10472 * ... 10473 * register_netdevice(x1); 10474 * register_netdevice(x2); 10475 * ... 10476 * unregister_netdevice(y1); 10477 * unregister_netdevice(y2); 10478 * ... 10479 * rtnl_unlock(); 10480 * free_netdev(y1); 10481 * free_netdev(y2); 10482 * 10483 * We are invoked by rtnl_unlock(). 10484 * This allows us to deal with problems: 10485 * 1) We can delete sysfs objects which invoke hotplug 10486 * without deadlocking with linkwatch via keventd. 10487 * 2) Since we run with the RTNL semaphore not held, we can sleep 10488 * safely in order to wait for the netdev refcnt to drop to zero. 10489 * 10490 * We must not return until all unregister events added during 10491 * the interval the lock was held have been completed. 10492 */ 10493 void netdev_run_todo(void) 10494 { 10495 struct net_device *dev, *tmp; 10496 struct list_head list; 10497 #ifdef CONFIG_LOCKDEP 10498 struct list_head unlink_list; 10499 10500 list_replace_init(&net_unlink_list, &unlink_list); 10501 10502 while (!list_empty(&unlink_list)) { 10503 struct net_device *dev = list_first_entry(&unlink_list, 10504 struct net_device, 10505 unlink_list); 10506 list_del_init(&dev->unlink_list); 10507 dev->nested_level = dev->lower_level - 1; 10508 } 10509 #endif 10510 10511 /* Snapshot list, allow later requests */ 10512 list_replace_init(&net_todo_list, &list); 10513 10514 __rtnl_unlock(); 10515 10516 /* Wait for rcu callbacks to finish before next phase */ 10517 if (!list_empty(&list)) 10518 rcu_barrier(); 10519 10520 list_for_each_entry_safe(dev, tmp, &list, todo_list) { 10521 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { 10522 netdev_WARN(dev, "run_todo but not unregistering\n"); 10523 list_del(&dev->todo_list); 10524 continue; 10525 } 10526 10527 write_lock(&dev_base_lock); 10528 dev->reg_state = NETREG_UNREGISTERED; 10529 write_unlock(&dev_base_lock); 10530 linkwatch_forget_dev(dev); 10531 } 10532 10533 while (!list_empty(&list)) { 10534 dev = netdev_wait_allrefs_any(&list); 10535 list_del(&dev->todo_list); 10536 10537 /* paranoia */ 10538 BUG_ON(netdev_refcnt_read(dev) != 1); 10539 BUG_ON(!list_empty(&dev->ptype_all)); 10540 BUG_ON(!list_empty(&dev->ptype_specific)); 10541 WARN_ON(rcu_access_pointer(dev->ip_ptr)); 10542 WARN_ON(rcu_access_pointer(dev->ip6_ptr)); 10543 10544 netdev_do_free_pcpu_stats(dev); 10545 if (dev->priv_destructor) 10546 dev->priv_destructor(dev); 10547 if (dev->needs_free_netdev) 10548 free_netdev(dev); 10549 10550 if (atomic_dec_and_test(&dev_net(dev)->dev_unreg_count)) 10551 wake_up(&netdev_unregistering_wq); 10552 10553 /* Free network device */ 10554 kobject_put(&dev->dev.kobj); 10555 } 10556 } 10557 10558 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has 10559 * all the same fields in the same order as net_device_stats, with only 10560 * the type differing, but rtnl_link_stats64 may have additional fields 10561 * at the end for newer counters. 10562 */ 10563 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, 10564 const struct net_device_stats *netdev_stats) 10565 { 10566 size_t i, n = sizeof(*netdev_stats) / sizeof(atomic_long_t); 10567 const atomic_long_t *src = (atomic_long_t *)netdev_stats; 10568 u64 *dst = (u64 *)stats64; 10569 10570 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); 10571 for (i = 0; i < n; i++) 10572 dst[i] = (unsigned long)atomic_long_read(&src[i]); 10573 /* zero out counters that only exist in rtnl_link_stats64 */ 10574 memset((char *)stats64 + n * sizeof(u64), 0, 10575 sizeof(*stats64) - n * sizeof(u64)); 10576 } 10577 EXPORT_SYMBOL(netdev_stats_to_stats64); 10578 10579 struct net_device_core_stats __percpu *netdev_core_stats_alloc(struct net_device *dev) 10580 { 10581 struct net_device_core_stats __percpu *p; 10582 10583 p = alloc_percpu_gfp(struct net_device_core_stats, 10584 GFP_ATOMIC | __GFP_NOWARN); 10585 10586 if (p && cmpxchg(&dev->core_stats, NULL, p)) 10587 free_percpu(p); 10588 10589 /* This READ_ONCE() pairs with the cmpxchg() above */ 10590 return READ_ONCE(dev->core_stats); 10591 } 10592 EXPORT_SYMBOL(netdev_core_stats_alloc); 10593 10594 /** 10595 * dev_get_stats - get network device statistics 10596 * @dev: device to get statistics from 10597 * @storage: place to store stats 10598 * 10599 * Get network statistics from device. Return @storage. 10600 * The device driver may provide its own method by setting 10601 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; 10602 * otherwise the internal statistics structure is used. 10603 */ 10604 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, 10605 struct rtnl_link_stats64 *storage) 10606 { 10607 const struct net_device_ops *ops = dev->netdev_ops; 10608 const struct net_device_core_stats __percpu *p; 10609 10610 if (ops->ndo_get_stats64) { 10611 memset(storage, 0, sizeof(*storage)); 10612 ops->ndo_get_stats64(dev, storage); 10613 } else if (ops->ndo_get_stats) { 10614 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); 10615 } else { 10616 netdev_stats_to_stats64(storage, &dev->stats); 10617 } 10618 10619 /* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */ 10620 p = READ_ONCE(dev->core_stats); 10621 if (p) { 10622 const struct net_device_core_stats *core_stats; 10623 int i; 10624 10625 for_each_possible_cpu(i) { 10626 core_stats = per_cpu_ptr(p, i); 10627 storage->rx_dropped += READ_ONCE(core_stats->rx_dropped); 10628 storage->tx_dropped += READ_ONCE(core_stats->tx_dropped); 10629 storage->rx_nohandler += READ_ONCE(core_stats->rx_nohandler); 10630 storage->rx_otherhost_dropped += READ_ONCE(core_stats->rx_otherhost_dropped); 10631 } 10632 } 10633 return storage; 10634 } 10635 EXPORT_SYMBOL(dev_get_stats); 10636 10637 /** 10638 * dev_fetch_sw_netstats - get per-cpu network device statistics 10639 * @s: place to store stats 10640 * @netstats: per-cpu network stats to read from 10641 * 10642 * Read per-cpu network statistics and populate the related fields in @s. 10643 */ 10644 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s, 10645 const struct pcpu_sw_netstats __percpu *netstats) 10646 { 10647 int cpu; 10648 10649 for_each_possible_cpu(cpu) { 10650 u64 rx_packets, rx_bytes, tx_packets, tx_bytes; 10651 const struct pcpu_sw_netstats *stats; 10652 unsigned int start; 10653 10654 stats = per_cpu_ptr(netstats, cpu); 10655 do { 10656 start = u64_stats_fetch_begin(&stats->syncp); 10657 rx_packets = u64_stats_read(&stats->rx_packets); 10658 rx_bytes = u64_stats_read(&stats->rx_bytes); 10659 tx_packets = u64_stats_read(&stats->tx_packets); 10660 tx_bytes = u64_stats_read(&stats->tx_bytes); 10661 } while (u64_stats_fetch_retry(&stats->syncp, start)); 10662 10663 s->rx_packets += rx_packets; 10664 s->rx_bytes += rx_bytes; 10665 s->tx_packets += tx_packets; 10666 s->tx_bytes += tx_bytes; 10667 } 10668 } 10669 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats); 10670 10671 /** 10672 * dev_get_tstats64 - ndo_get_stats64 implementation 10673 * @dev: device to get statistics from 10674 * @s: place to store stats 10675 * 10676 * Populate @s from dev->stats and dev->tstats. Can be used as 10677 * ndo_get_stats64() callback. 10678 */ 10679 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s) 10680 { 10681 netdev_stats_to_stats64(s, &dev->stats); 10682 dev_fetch_sw_netstats(s, dev->tstats); 10683 } 10684 EXPORT_SYMBOL_GPL(dev_get_tstats64); 10685 10686 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) 10687 { 10688 struct netdev_queue *queue = dev_ingress_queue(dev); 10689 10690 #ifdef CONFIG_NET_CLS_ACT 10691 if (queue) 10692 return queue; 10693 queue = kzalloc(sizeof(*queue), GFP_KERNEL); 10694 if (!queue) 10695 return NULL; 10696 netdev_init_one_queue(dev, queue, NULL); 10697 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); 10698 RCU_INIT_POINTER(queue->qdisc_sleeping, &noop_qdisc); 10699 rcu_assign_pointer(dev->ingress_queue, queue); 10700 #endif 10701 return queue; 10702 } 10703 10704 static const struct ethtool_ops default_ethtool_ops; 10705 10706 void netdev_set_default_ethtool_ops(struct net_device *dev, 10707 const struct ethtool_ops *ops) 10708 { 10709 if (dev->ethtool_ops == &default_ethtool_ops) 10710 dev->ethtool_ops = ops; 10711 } 10712 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); 10713 10714 /** 10715 * netdev_sw_irq_coalesce_default_on() - enable SW IRQ coalescing by default 10716 * @dev: netdev to enable the IRQ coalescing on 10717 * 10718 * Sets a conservative default for SW IRQ coalescing. Users can use 10719 * sysfs attributes to override the default values. 10720 */ 10721 void netdev_sw_irq_coalesce_default_on(struct net_device *dev) 10722 { 10723 WARN_ON(dev->reg_state == NETREG_REGISTERED); 10724 10725 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) { 10726 dev->gro_flush_timeout = 20000; 10727 dev->napi_defer_hard_irqs = 1; 10728 } 10729 } 10730 EXPORT_SYMBOL_GPL(netdev_sw_irq_coalesce_default_on); 10731 10732 void netdev_freemem(struct net_device *dev) 10733 { 10734 char *addr = (char *)dev - dev->padded; 10735 10736 kvfree(addr); 10737 } 10738 10739 /** 10740 * alloc_netdev_mqs - allocate network device 10741 * @sizeof_priv: size of private data to allocate space for 10742 * @name: device name format string 10743 * @name_assign_type: origin of device name 10744 * @setup: callback to initialize device 10745 * @txqs: the number of TX subqueues to allocate 10746 * @rxqs: the number of RX subqueues to allocate 10747 * 10748 * Allocates a struct net_device with private data area for driver use 10749 * and performs basic initialization. Also allocates subqueue structs 10750 * for each queue on the device. 10751 */ 10752 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, 10753 unsigned char name_assign_type, 10754 void (*setup)(struct net_device *), 10755 unsigned int txqs, unsigned int rxqs) 10756 { 10757 struct net_device *dev; 10758 unsigned int alloc_size; 10759 struct net_device *p; 10760 10761 BUG_ON(strlen(name) >= sizeof(dev->name)); 10762 10763 if (txqs < 1) { 10764 pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); 10765 return NULL; 10766 } 10767 10768 if (rxqs < 1) { 10769 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); 10770 return NULL; 10771 } 10772 10773 alloc_size = sizeof(struct net_device); 10774 if (sizeof_priv) { 10775 /* ensure 32-byte alignment of private area */ 10776 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); 10777 alloc_size += sizeof_priv; 10778 } 10779 /* ensure 32-byte alignment of whole construct */ 10780 alloc_size += NETDEV_ALIGN - 1; 10781 10782 p = kvzalloc(alloc_size, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10783 if (!p) 10784 return NULL; 10785 10786 dev = PTR_ALIGN(p, NETDEV_ALIGN); 10787 dev->padded = (char *)dev - (char *)p; 10788 10789 ref_tracker_dir_init(&dev->refcnt_tracker, 128, name); 10790 #ifdef CONFIG_PCPU_DEV_REFCNT 10791 dev->pcpu_refcnt = alloc_percpu(int); 10792 if (!dev->pcpu_refcnt) 10793 goto free_dev; 10794 __dev_hold(dev); 10795 #else 10796 refcount_set(&dev->dev_refcnt, 1); 10797 #endif 10798 10799 if (dev_addr_init(dev)) 10800 goto free_pcpu; 10801 10802 dev_mc_init(dev); 10803 dev_uc_init(dev); 10804 10805 dev_net_set(dev, &init_net); 10806 10807 dev->gso_max_size = GSO_LEGACY_MAX_SIZE; 10808 dev->xdp_zc_max_segs = 1; 10809 dev->gso_max_segs = GSO_MAX_SEGS; 10810 dev->gro_max_size = GRO_LEGACY_MAX_SIZE; 10811 dev->gso_ipv4_max_size = GSO_LEGACY_MAX_SIZE; 10812 dev->gro_ipv4_max_size = GRO_LEGACY_MAX_SIZE; 10813 dev->tso_max_size = TSO_LEGACY_MAX_SIZE; 10814 dev->tso_max_segs = TSO_MAX_SEGS; 10815 dev->upper_level = 1; 10816 dev->lower_level = 1; 10817 #ifdef CONFIG_LOCKDEP 10818 dev->nested_level = 0; 10819 INIT_LIST_HEAD(&dev->unlink_list); 10820 #endif 10821 10822 INIT_LIST_HEAD(&dev->napi_list); 10823 INIT_LIST_HEAD(&dev->unreg_list); 10824 INIT_LIST_HEAD(&dev->close_list); 10825 INIT_LIST_HEAD(&dev->link_watch_list); 10826 INIT_LIST_HEAD(&dev->adj_list.upper); 10827 INIT_LIST_HEAD(&dev->adj_list.lower); 10828 INIT_LIST_HEAD(&dev->ptype_all); 10829 INIT_LIST_HEAD(&dev->ptype_specific); 10830 INIT_LIST_HEAD(&dev->net_notifier_list); 10831 #ifdef CONFIG_NET_SCHED 10832 hash_init(dev->qdisc_hash); 10833 #endif 10834 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; 10835 setup(dev); 10836 10837 if (!dev->tx_queue_len) { 10838 dev->priv_flags |= IFF_NO_QUEUE; 10839 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; 10840 } 10841 10842 dev->num_tx_queues = txqs; 10843 dev->real_num_tx_queues = txqs; 10844 if (netif_alloc_netdev_queues(dev)) 10845 goto free_all; 10846 10847 dev->num_rx_queues = rxqs; 10848 dev->real_num_rx_queues = rxqs; 10849 if (netif_alloc_rx_queues(dev)) 10850 goto free_all; 10851 10852 strcpy(dev->name, name); 10853 dev->name_assign_type = name_assign_type; 10854 dev->group = INIT_NETDEV_GROUP; 10855 if (!dev->ethtool_ops) 10856 dev->ethtool_ops = &default_ethtool_ops; 10857 10858 nf_hook_netdev_init(dev); 10859 10860 return dev; 10861 10862 free_all: 10863 free_netdev(dev); 10864 return NULL; 10865 10866 free_pcpu: 10867 #ifdef CONFIG_PCPU_DEV_REFCNT 10868 free_percpu(dev->pcpu_refcnt); 10869 free_dev: 10870 #endif 10871 netdev_freemem(dev); 10872 return NULL; 10873 } 10874 EXPORT_SYMBOL(alloc_netdev_mqs); 10875 10876 /** 10877 * free_netdev - free network device 10878 * @dev: device 10879 * 10880 * This function does the last stage of destroying an allocated device 10881 * interface. The reference to the device object is released. If this 10882 * is the last reference then it will be freed.Must be called in process 10883 * context. 10884 */ 10885 void free_netdev(struct net_device *dev) 10886 { 10887 struct napi_struct *p, *n; 10888 10889 might_sleep(); 10890 10891 /* When called immediately after register_netdevice() failed the unwind 10892 * handling may still be dismantling the device. Handle that case by 10893 * deferring the free. 10894 */ 10895 if (dev->reg_state == NETREG_UNREGISTERING) { 10896 ASSERT_RTNL(); 10897 dev->needs_free_netdev = true; 10898 return; 10899 } 10900 10901 netif_free_tx_queues(dev); 10902 netif_free_rx_queues(dev); 10903 10904 kfree(rcu_dereference_protected(dev->ingress_queue, 1)); 10905 10906 /* Flush device addresses */ 10907 dev_addr_flush(dev); 10908 10909 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) 10910 netif_napi_del(p); 10911 10912 ref_tracker_dir_exit(&dev->refcnt_tracker); 10913 #ifdef CONFIG_PCPU_DEV_REFCNT 10914 free_percpu(dev->pcpu_refcnt); 10915 dev->pcpu_refcnt = NULL; 10916 #endif 10917 free_percpu(dev->core_stats); 10918 dev->core_stats = NULL; 10919 free_percpu(dev->xdp_bulkq); 10920 dev->xdp_bulkq = NULL; 10921 10922 /* Compatibility with error handling in drivers */ 10923 if (dev->reg_state == NETREG_UNINITIALIZED) { 10924 netdev_freemem(dev); 10925 return; 10926 } 10927 10928 BUG_ON(dev->reg_state != NETREG_UNREGISTERED); 10929 dev->reg_state = NETREG_RELEASED; 10930 10931 /* will free via device release */ 10932 put_device(&dev->dev); 10933 } 10934 EXPORT_SYMBOL(free_netdev); 10935 10936 /** 10937 * synchronize_net - Synchronize with packet receive processing 10938 * 10939 * Wait for packets currently being received to be done. 10940 * Does not block later packets from starting. 10941 */ 10942 void synchronize_net(void) 10943 { 10944 might_sleep(); 10945 if (rtnl_is_locked()) 10946 synchronize_rcu_expedited(); 10947 else 10948 synchronize_rcu(); 10949 } 10950 EXPORT_SYMBOL(synchronize_net); 10951 10952 /** 10953 * unregister_netdevice_queue - remove device from the kernel 10954 * @dev: device 10955 * @head: list 10956 * 10957 * This function shuts down a device interface and removes it 10958 * from the kernel tables. 10959 * If head not NULL, device is queued to be unregistered later. 10960 * 10961 * Callers must hold the rtnl semaphore. You may want 10962 * unregister_netdev() instead of this. 10963 */ 10964 10965 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) 10966 { 10967 ASSERT_RTNL(); 10968 10969 if (head) { 10970 list_move_tail(&dev->unreg_list, head); 10971 } else { 10972 LIST_HEAD(single); 10973 10974 list_add(&dev->unreg_list, &single); 10975 unregister_netdevice_many(&single); 10976 } 10977 } 10978 EXPORT_SYMBOL(unregister_netdevice_queue); 10979 10980 void unregister_netdevice_many_notify(struct list_head *head, 10981 u32 portid, const struct nlmsghdr *nlh) 10982 { 10983 struct net_device *dev, *tmp; 10984 LIST_HEAD(close_head); 10985 10986 BUG_ON(dev_boot_phase); 10987 ASSERT_RTNL(); 10988 10989 if (list_empty(head)) 10990 return; 10991 10992 list_for_each_entry_safe(dev, tmp, head, unreg_list) { 10993 /* Some devices call without registering 10994 * for initialization unwind. Remove those 10995 * devices and proceed with the remaining. 10996 */ 10997 if (dev->reg_state == NETREG_UNINITIALIZED) { 10998 pr_debug("unregister_netdevice: device %s/%p never was registered\n", 10999 dev->name, dev); 11000 11001 WARN_ON(1); 11002 list_del(&dev->unreg_list); 11003 continue; 11004 } 11005 dev->dismantle = true; 11006 BUG_ON(dev->reg_state != NETREG_REGISTERED); 11007 } 11008 11009 /* If device is running, close it first. */ 11010 list_for_each_entry(dev, head, unreg_list) 11011 list_add_tail(&dev->close_list, &close_head); 11012 dev_close_many(&close_head, true); 11013 11014 list_for_each_entry(dev, head, unreg_list) { 11015 /* And unlink it from device chain. */ 11016 write_lock(&dev_base_lock); 11017 unlist_netdevice(dev, false); 11018 dev->reg_state = NETREG_UNREGISTERING; 11019 write_unlock(&dev_base_lock); 11020 } 11021 flush_all_backlogs(); 11022 11023 synchronize_net(); 11024 11025 list_for_each_entry(dev, head, unreg_list) { 11026 struct sk_buff *skb = NULL; 11027 11028 /* Shutdown queueing discipline. */ 11029 dev_shutdown(dev); 11030 dev_tcx_uninstall(dev); 11031 dev_xdp_uninstall(dev); 11032 bpf_dev_bound_netdev_unregister(dev); 11033 11034 netdev_offload_xstats_disable_all(dev); 11035 11036 /* Notify protocols, that we are about to destroy 11037 * this device. They should clean all the things. 11038 */ 11039 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 11040 11041 if (!dev->rtnl_link_ops || 11042 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 11043 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0, 11044 GFP_KERNEL, NULL, 0, 11045 portid, nlh); 11046 11047 /* 11048 * Flush the unicast and multicast chains 11049 */ 11050 dev_uc_flush(dev); 11051 dev_mc_flush(dev); 11052 11053 netdev_name_node_alt_flush(dev); 11054 netdev_name_node_free(dev->name_node); 11055 11056 call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev); 11057 11058 if (dev->netdev_ops->ndo_uninit) 11059 dev->netdev_ops->ndo_uninit(dev); 11060 11061 if (skb) 11062 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL, portid, nlh); 11063 11064 /* Notifier chain MUST detach us all upper devices. */ 11065 WARN_ON(netdev_has_any_upper_dev(dev)); 11066 WARN_ON(netdev_has_any_lower_dev(dev)); 11067 11068 /* Remove entries from kobject tree */ 11069 netdev_unregister_kobject(dev); 11070 #ifdef CONFIG_XPS 11071 /* Remove XPS queueing entries */ 11072 netif_reset_xps_queues_gt(dev, 0); 11073 #endif 11074 } 11075 11076 synchronize_net(); 11077 11078 list_for_each_entry(dev, head, unreg_list) { 11079 netdev_put(dev, &dev->dev_registered_tracker); 11080 net_set_todo(dev); 11081 } 11082 11083 list_del(head); 11084 } 11085 11086 /** 11087 * unregister_netdevice_many - unregister many devices 11088 * @head: list of devices 11089 * 11090 * Note: As most callers use a stack allocated list_head, 11091 * we force a list_del() to make sure stack wont be corrupted later. 11092 */ 11093 void unregister_netdevice_many(struct list_head *head) 11094 { 11095 unregister_netdevice_many_notify(head, 0, NULL); 11096 } 11097 EXPORT_SYMBOL(unregister_netdevice_many); 11098 11099 /** 11100 * unregister_netdev - remove device from the kernel 11101 * @dev: device 11102 * 11103 * This function shuts down a device interface and removes it 11104 * from the kernel tables. 11105 * 11106 * This is just a wrapper for unregister_netdevice that takes 11107 * the rtnl semaphore. In general you want to use this and not 11108 * unregister_netdevice. 11109 */ 11110 void unregister_netdev(struct net_device *dev) 11111 { 11112 rtnl_lock(); 11113 unregister_netdevice(dev); 11114 rtnl_unlock(); 11115 } 11116 EXPORT_SYMBOL(unregister_netdev); 11117 11118 /** 11119 * __dev_change_net_namespace - move device to different nethost namespace 11120 * @dev: device 11121 * @net: network namespace 11122 * @pat: If not NULL name pattern to try if the current device name 11123 * is already taken in the destination network namespace. 11124 * @new_ifindex: If not zero, specifies device index in the target 11125 * namespace. 11126 * 11127 * This function shuts down a device interface and moves it 11128 * to a new network namespace. On success 0 is returned, on 11129 * a failure a netagive errno code is returned. 11130 * 11131 * Callers must hold the rtnl semaphore. 11132 */ 11133 11134 int __dev_change_net_namespace(struct net_device *dev, struct net *net, 11135 const char *pat, int new_ifindex) 11136 { 11137 struct netdev_name_node *name_node; 11138 struct net *net_old = dev_net(dev); 11139 char new_name[IFNAMSIZ] = {}; 11140 int err, new_nsid; 11141 11142 ASSERT_RTNL(); 11143 11144 /* Don't allow namespace local devices to be moved. */ 11145 err = -EINVAL; 11146 if (dev->features & NETIF_F_NETNS_LOCAL) 11147 goto out; 11148 11149 /* Ensure the device has been registrered */ 11150 if (dev->reg_state != NETREG_REGISTERED) 11151 goto out; 11152 11153 /* Get out if there is nothing todo */ 11154 err = 0; 11155 if (net_eq(net_old, net)) 11156 goto out; 11157 11158 /* Pick the destination device name, and ensure 11159 * we can use it in the destination network namespace. 11160 */ 11161 err = -EEXIST; 11162 if (netdev_name_in_use(net, dev->name)) { 11163 /* We get here if we can't use the current device name */ 11164 if (!pat) 11165 goto out; 11166 err = dev_prep_valid_name(net, dev, pat, new_name); 11167 if (err < 0) 11168 goto out; 11169 } 11170 /* Check that none of the altnames conflicts. */ 11171 err = -EEXIST; 11172 netdev_for_each_altname(dev, name_node) 11173 if (netdev_name_in_use(net, name_node->name)) 11174 goto out; 11175 11176 /* Check that new_ifindex isn't used yet. */ 11177 if (new_ifindex) { 11178 err = dev_index_reserve(net, new_ifindex); 11179 if (err < 0) 11180 goto out; 11181 } else { 11182 /* If there is an ifindex conflict assign a new one */ 11183 err = dev_index_reserve(net, dev->ifindex); 11184 if (err == -EBUSY) 11185 err = dev_index_reserve(net, 0); 11186 if (err < 0) 11187 goto out; 11188 new_ifindex = err; 11189 } 11190 11191 /* 11192 * And now a mini version of register_netdevice unregister_netdevice. 11193 */ 11194 11195 /* If device is running close it first. */ 11196 dev_close(dev); 11197 11198 /* And unlink it from device chain */ 11199 unlist_netdevice(dev, true); 11200 11201 synchronize_net(); 11202 11203 /* Shutdown queueing discipline. */ 11204 dev_shutdown(dev); 11205 11206 /* Notify protocols, that we are about to destroy 11207 * this device. They should clean all the things. 11208 * 11209 * Note that dev->reg_state stays at NETREG_REGISTERED. 11210 * This is wanted because this way 8021q and macvlan know 11211 * the device is just moving and can keep their slaves up. 11212 */ 11213 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 11214 rcu_barrier(); 11215 11216 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL); 11217 11218 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid, 11219 new_ifindex); 11220 11221 /* 11222 * Flush the unicast and multicast chains 11223 */ 11224 dev_uc_flush(dev); 11225 dev_mc_flush(dev); 11226 11227 /* Send a netdev-removed uevent to the old namespace */ 11228 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); 11229 netdev_adjacent_del_links(dev); 11230 11231 /* Move per-net netdevice notifiers that are following the netdevice */ 11232 move_netdevice_notifiers_dev_net(dev, net); 11233 11234 /* Actually switch the network namespace */ 11235 dev_net_set(dev, net); 11236 dev->ifindex = new_ifindex; 11237 11238 /* Send a netdev-add uevent to the new namespace */ 11239 kobject_uevent(&dev->dev.kobj, KOBJ_ADD); 11240 netdev_adjacent_add_links(dev); 11241 11242 if (new_name[0]) /* Rename the netdev to prepared name */ 11243 strscpy(dev->name, new_name, IFNAMSIZ); 11244 11245 /* Fixup kobjects */ 11246 err = device_rename(&dev->dev, dev->name); 11247 WARN_ON(err); 11248 11249 /* Adapt owner in case owning user namespace of target network 11250 * namespace is different from the original one. 11251 */ 11252 err = netdev_change_owner(dev, net_old, net); 11253 WARN_ON(err); 11254 11255 /* Add the device back in the hashes */ 11256 list_netdevice(dev); 11257 11258 /* Notify protocols, that a new device appeared. */ 11259 call_netdevice_notifiers(NETDEV_REGISTER, dev); 11260 11261 /* 11262 * Prevent userspace races by waiting until the network 11263 * device is fully setup before sending notifications. 11264 */ 11265 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL); 11266 11267 synchronize_net(); 11268 err = 0; 11269 out: 11270 return err; 11271 } 11272 EXPORT_SYMBOL_GPL(__dev_change_net_namespace); 11273 11274 static int dev_cpu_dead(unsigned int oldcpu) 11275 { 11276 struct sk_buff **list_skb; 11277 struct sk_buff *skb; 11278 unsigned int cpu; 11279 struct softnet_data *sd, *oldsd, *remsd = NULL; 11280 11281 local_irq_disable(); 11282 cpu = smp_processor_id(); 11283 sd = &per_cpu(softnet_data, cpu); 11284 oldsd = &per_cpu(softnet_data, oldcpu); 11285 11286 /* Find end of our completion_queue. */ 11287 list_skb = &sd->completion_queue; 11288 while (*list_skb) 11289 list_skb = &(*list_skb)->next; 11290 /* Append completion queue from offline CPU. */ 11291 *list_skb = oldsd->completion_queue; 11292 oldsd->completion_queue = NULL; 11293 11294 /* Append output queue from offline CPU. */ 11295 if (oldsd->output_queue) { 11296 *sd->output_queue_tailp = oldsd->output_queue; 11297 sd->output_queue_tailp = oldsd->output_queue_tailp; 11298 oldsd->output_queue = NULL; 11299 oldsd->output_queue_tailp = &oldsd->output_queue; 11300 } 11301 /* Append NAPI poll list from offline CPU, with one exception : 11302 * process_backlog() must be called by cpu owning percpu backlog. 11303 * We properly handle process_queue & input_pkt_queue later. 11304 */ 11305 while (!list_empty(&oldsd->poll_list)) { 11306 struct napi_struct *napi = list_first_entry(&oldsd->poll_list, 11307 struct napi_struct, 11308 poll_list); 11309 11310 list_del_init(&napi->poll_list); 11311 if (napi->poll == process_backlog) 11312 napi->state = 0; 11313 else 11314 ____napi_schedule(sd, napi); 11315 } 11316 11317 raise_softirq_irqoff(NET_TX_SOFTIRQ); 11318 local_irq_enable(); 11319 11320 #ifdef CONFIG_RPS 11321 remsd = oldsd->rps_ipi_list; 11322 oldsd->rps_ipi_list = NULL; 11323 #endif 11324 /* send out pending IPI's on offline CPU */ 11325 net_rps_send_ipi(remsd); 11326 11327 /* Process offline CPU's input_pkt_queue */ 11328 while ((skb = __skb_dequeue(&oldsd->process_queue))) { 11329 netif_rx(skb); 11330 input_queue_head_incr(oldsd); 11331 } 11332 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { 11333 netif_rx(skb); 11334 input_queue_head_incr(oldsd); 11335 } 11336 11337 return 0; 11338 } 11339 11340 /** 11341 * netdev_increment_features - increment feature set by one 11342 * @all: current feature set 11343 * @one: new feature set 11344 * @mask: mask feature set 11345 * 11346 * Computes a new feature set after adding a device with feature set 11347 * @one to the master device with current feature set @all. Will not 11348 * enable anything that is off in @mask. Returns the new feature set. 11349 */ 11350 netdev_features_t netdev_increment_features(netdev_features_t all, 11351 netdev_features_t one, netdev_features_t mask) 11352 { 11353 if (mask & NETIF_F_HW_CSUM) 11354 mask |= NETIF_F_CSUM_MASK; 11355 mask |= NETIF_F_VLAN_CHALLENGED; 11356 11357 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; 11358 all &= one | ~NETIF_F_ALL_FOR_ALL; 11359 11360 /* If one device supports hw checksumming, set for all. */ 11361 if (all & NETIF_F_HW_CSUM) 11362 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); 11363 11364 return all; 11365 } 11366 EXPORT_SYMBOL(netdev_increment_features); 11367 11368 static struct hlist_head * __net_init netdev_create_hash(void) 11369 { 11370 int i; 11371 struct hlist_head *hash; 11372 11373 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL); 11374 if (hash != NULL) 11375 for (i = 0; i < NETDEV_HASHENTRIES; i++) 11376 INIT_HLIST_HEAD(&hash[i]); 11377 11378 return hash; 11379 } 11380 11381 /* Initialize per network namespace state */ 11382 static int __net_init netdev_init(struct net *net) 11383 { 11384 BUILD_BUG_ON(GRO_HASH_BUCKETS > 11385 8 * sizeof_field(struct napi_struct, gro_bitmask)); 11386 11387 INIT_LIST_HEAD(&net->dev_base_head); 11388 11389 net->dev_name_head = netdev_create_hash(); 11390 if (net->dev_name_head == NULL) 11391 goto err_name; 11392 11393 net->dev_index_head = netdev_create_hash(); 11394 if (net->dev_index_head == NULL) 11395 goto err_idx; 11396 11397 xa_init_flags(&net->dev_by_index, XA_FLAGS_ALLOC1); 11398 11399 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain); 11400 11401 return 0; 11402 11403 err_idx: 11404 kfree(net->dev_name_head); 11405 err_name: 11406 return -ENOMEM; 11407 } 11408 11409 /** 11410 * netdev_drivername - network driver for the device 11411 * @dev: network device 11412 * 11413 * Determine network driver for device. 11414 */ 11415 const char *netdev_drivername(const struct net_device *dev) 11416 { 11417 const struct device_driver *driver; 11418 const struct device *parent; 11419 const char *empty = ""; 11420 11421 parent = dev->dev.parent; 11422 if (!parent) 11423 return empty; 11424 11425 driver = parent->driver; 11426 if (driver && driver->name) 11427 return driver->name; 11428 return empty; 11429 } 11430 11431 static void __netdev_printk(const char *level, const struct net_device *dev, 11432 struct va_format *vaf) 11433 { 11434 if (dev && dev->dev.parent) { 11435 dev_printk_emit(level[1] - '0', 11436 dev->dev.parent, 11437 "%s %s %s%s: %pV", 11438 dev_driver_string(dev->dev.parent), 11439 dev_name(dev->dev.parent), 11440 netdev_name(dev), netdev_reg_state(dev), 11441 vaf); 11442 } else if (dev) { 11443 printk("%s%s%s: %pV", 11444 level, netdev_name(dev), netdev_reg_state(dev), vaf); 11445 } else { 11446 printk("%s(NULL net_device): %pV", level, vaf); 11447 } 11448 } 11449 11450 void netdev_printk(const char *level, const struct net_device *dev, 11451 const char *format, ...) 11452 { 11453 struct va_format vaf; 11454 va_list args; 11455 11456 va_start(args, format); 11457 11458 vaf.fmt = format; 11459 vaf.va = &args; 11460 11461 __netdev_printk(level, dev, &vaf); 11462 11463 va_end(args); 11464 } 11465 EXPORT_SYMBOL(netdev_printk); 11466 11467 #define define_netdev_printk_level(func, level) \ 11468 void func(const struct net_device *dev, const char *fmt, ...) \ 11469 { \ 11470 struct va_format vaf; \ 11471 va_list args; \ 11472 \ 11473 va_start(args, fmt); \ 11474 \ 11475 vaf.fmt = fmt; \ 11476 vaf.va = &args; \ 11477 \ 11478 __netdev_printk(level, dev, &vaf); \ 11479 \ 11480 va_end(args); \ 11481 } \ 11482 EXPORT_SYMBOL(func); 11483 11484 define_netdev_printk_level(netdev_emerg, KERN_EMERG); 11485 define_netdev_printk_level(netdev_alert, KERN_ALERT); 11486 define_netdev_printk_level(netdev_crit, KERN_CRIT); 11487 define_netdev_printk_level(netdev_err, KERN_ERR); 11488 define_netdev_printk_level(netdev_warn, KERN_WARNING); 11489 define_netdev_printk_level(netdev_notice, KERN_NOTICE); 11490 define_netdev_printk_level(netdev_info, KERN_INFO); 11491 11492 static void __net_exit netdev_exit(struct net *net) 11493 { 11494 kfree(net->dev_name_head); 11495 kfree(net->dev_index_head); 11496 xa_destroy(&net->dev_by_index); 11497 if (net != &init_net) 11498 WARN_ON_ONCE(!list_empty(&net->dev_base_head)); 11499 } 11500 11501 static struct pernet_operations __net_initdata netdev_net_ops = { 11502 .init = netdev_init, 11503 .exit = netdev_exit, 11504 }; 11505 11506 static void __net_exit default_device_exit_net(struct net *net) 11507 { 11508 struct netdev_name_node *name_node, *tmp; 11509 struct net_device *dev, *aux; 11510 /* 11511 * Push all migratable network devices back to the 11512 * initial network namespace 11513 */ 11514 ASSERT_RTNL(); 11515 for_each_netdev_safe(net, dev, aux) { 11516 int err; 11517 char fb_name[IFNAMSIZ]; 11518 11519 /* Ignore unmoveable devices (i.e. loopback) */ 11520 if (dev->features & NETIF_F_NETNS_LOCAL) 11521 continue; 11522 11523 /* Leave virtual devices for the generic cleanup */ 11524 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund) 11525 continue; 11526 11527 /* Push remaining network devices to init_net */ 11528 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); 11529 if (netdev_name_in_use(&init_net, fb_name)) 11530 snprintf(fb_name, IFNAMSIZ, "dev%%d"); 11531 11532 netdev_for_each_altname_safe(dev, name_node, tmp) 11533 if (netdev_name_in_use(&init_net, name_node->name)) { 11534 netdev_name_node_del(name_node); 11535 synchronize_rcu(); 11536 __netdev_name_node_alt_destroy(name_node); 11537 } 11538 11539 err = dev_change_net_namespace(dev, &init_net, fb_name); 11540 if (err) { 11541 pr_emerg("%s: failed to move %s to init_net: %d\n", 11542 __func__, dev->name, err); 11543 BUG(); 11544 } 11545 } 11546 } 11547 11548 static void __net_exit default_device_exit_batch(struct list_head *net_list) 11549 { 11550 /* At exit all network devices most be removed from a network 11551 * namespace. Do this in the reverse order of registration. 11552 * Do this across as many network namespaces as possible to 11553 * improve batching efficiency. 11554 */ 11555 struct net_device *dev; 11556 struct net *net; 11557 LIST_HEAD(dev_kill_list); 11558 11559 rtnl_lock(); 11560 list_for_each_entry(net, net_list, exit_list) { 11561 default_device_exit_net(net); 11562 cond_resched(); 11563 } 11564 11565 list_for_each_entry(net, net_list, exit_list) { 11566 for_each_netdev_reverse(net, dev) { 11567 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) 11568 dev->rtnl_link_ops->dellink(dev, &dev_kill_list); 11569 else 11570 unregister_netdevice_queue(dev, &dev_kill_list); 11571 } 11572 } 11573 unregister_netdevice_many(&dev_kill_list); 11574 rtnl_unlock(); 11575 } 11576 11577 static struct pernet_operations __net_initdata default_device_ops = { 11578 .exit_batch = default_device_exit_batch, 11579 }; 11580 11581 /* 11582 * Initialize the DEV module. At boot time this walks the device list and 11583 * unhooks any devices that fail to initialise (normally hardware not 11584 * present) and leaves us with a valid list of present and active devices. 11585 * 11586 */ 11587 11588 /* 11589 * This is called single threaded during boot, so no need 11590 * to take the rtnl semaphore. 11591 */ 11592 static int __init net_dev_init(void) 11593 { 11594 int i, rc = -ENOMEM; 11595 11596 BUG_ON(!dev_boot_phase); 11597 11598 if (dev_proc_init()) 11599 goto out; 11600 11601 if (netdev_kobject_init()) 11602 goto out; 11603 11604 INIT_LIST_HEAD(&ptype_all); 11605 for (i = 0; i < PTYPE_HASH_SIZE; i++) 11606 INIT_LIST_HEAD(&ptype_base[i]); 11607 11608 if (register_pernet_subsys(&netdev_net_ops)) 11609 goto out; 11610 11611 /* 11612 * Initialise the packet receive queues. 11613 */ 11614 11615 for_each_possible_cpu(i) { 11616 struct work_struct *flush = per_cpu_ptr(&flush_works, i); 11617 struct softnet_data *sd = &per_cpu(softnet_data, i); 11618 11619 INIT_WORK(flush, flush_backlog); 11620 11621 skb_queue_head_init(&sd->input_pkt_queue); 11622 skb_queue_head_init(&sd->process_queue); 11623 #ifdef CONFIG_XFRM_OFFLOAD 11624 skb_queue_head_init(&sd->xfrm_backlog); 11625 #endif 11626 INIT_LIST_HEAD(&sd->poll_list); 11627 sd->output_queue_tailp = &sd->output_queue; 11628 #ifdef CONFIG_RPS 11629 INIT_CSD(&sd->csd, rps_trigger_softirq, sd); 11630 sd->cpu = i; 11631 #endif 11632 INIT_CSD(&sd->defer_csd, trigger_rx_softirq, sd); 11633 spin_lock_init(&sd->defer_lock); 11634 11635 init_gro_hash(&sd->backlog); 11636 sd->backlog.poll = process_backlog; 11637 sd->backlog.weight = weight_p; 11638 } 11639 11640 dev_boot_phase = 0; 11641 11642 /* The loopback device is special if any other network devices 11643 * is present in a network namespace the loopback device must 11644 * be present. Since we now dynamically allocate and free the 11645 * loopback device ensure this invariant is maintained by 11646 * keeping the loopback device as the first device on the 11647 * list of network devices. Ensuring the loopback devices 11648 * is the first device that appears and the last network device 11649 * that disappears. 11650 */ 11651 if (register_pernet_device(&loopback_net_ops)) 11652 goto out; 11653 11654 if (register_pernet_device(&default_device_ops)) 11655 goto out; 11656 11657 open_softirq(NET_TX_SOFTIRQ, net_tx_action); 11658 open_softirq(NET_RX_SOFTIRQ, net_rx_action); 11659 11660 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", 11661 NULL, dev_cpu_dead); 11662 WARN_ON(rc < 0); 11663 rc = 0; 11664 out: 11665 return rc; 11666 } 11667 11668 subsys_initcall(net_dev_init); 11669