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/bitops.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 <linux/highmem.h> 111 #include <linux/init.h> 112 #include <linux/module.h> 113 #include <linux/netpoll.h> 114 #include <linux/rcupdate.h> 115 #include <linux/delay.h> 116 #include <net/iw_handler.h> 117 #include <asm/current.h> 118 #include <linux/audit.h> 119 #include <linux/dmaengine.h> 120 #include <linux/err.h> 121 #include <linux/ctype.h> 122 #include <linux/if_arp.h> 123 #include <linux/if_vlan.h> 124 #include <linux/ip.h> 125 #include <net/ip.h> 126 #include <net/mpls.h> 127 #include <linux/ipv6.h> 128 #include <linux/in.h> 129 #include <linux/jhash.h> 130 #include <linux/random.h> 131 #include <trace/events/napi.h> 132 #include <trace/events/net.h> 133 #include <trace/events/skb.h> 134 #include <linux/inetdevice.h> 135 #include <linux/cpu_rmap.h> 136 #include <linux/static_key.h> 137 #include <linux/hashtable.h> 138 #include <linux/vmalloc.h> 139 #include <linux/if_macvlan.h> 140 #include <linux/errqueue.h> 141 #include <linux/hrtimer.h> 142 #include <linux/netfilter_ingress.h> 143 #include <linux/crash_dump.h> 144 #include <linux/sctp.h> 145 #include <net/udp_tunnel.h> 146 #include <linux/net_namespace.h> 147 #include <linux/indirect_call_wrapper.h> 148 #include <net/devlink.h> 149 #include <linux/pm_runtime.h> 150 #include <linux/prandom.h> 151 152 #include "net-sysfs.h" 153 154 #define MAX_GRO_SKBS 8 155 156 /* This should be increased if a protocol with a bigger head is added. */ 157 #define GRO_MAX_HEAD (MAX_HEADER + 128) 158 159 static DEFINE_SPINLOCK(ptype_lock); 160 static DEFINE_SPINLOCK(offload_lock); 161 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; 162 struct list_head ptype_all __read_mostly; /* Taps */ 163 static struct list_head offload_base __read_mostly; 164 165 static int netif_rx_internal(struct sk_buff *skb); 166 static int call_netdevice_notifiers_info(unsigned long val, 167 struct netdev_notifier_info *info); 168 static int call_netdevice_notifiers_extack(unsigned long val, 169 struct net_device *dev, 170 struct netlink_ext_ack *extack); 171 static struct napi_struct *napi_by_id(unsigned int napi_id); 172 173 /* 174 * The @dev_base_head list is protected by @dev_base_lock and the rtnl 175 * semaphore. 176 * 177 * Pure readers hold dev_base_lock for reading, or rcu_read_lock() 178 * 179 * Writers must hold the rtnl semaphore while they loop through the 180 * dev_base_head list, and hold dev_base_lock for writing when they do the 181 * actual updates. This allows pure readers to access the list even 182 * while a writer is preparing to update it. 183 * 184 * To put it another way, dev_base_lock is held for writing only to 185 * protect against pure readers; the rtnl semaphore provides the 186 * protection against other writers. 187 * 188 * See, for example usages, register_netdevice() and 189 * unregister_netdevice(), which must be called with the rtnl 190 * semaphore held. 191 */ 192 DEFINE_RWLOCK(dev_base_lock); 193 EXPORT_SYMBOL(dev_base_lock); 194 195 static DEFINE_MUTEX(ifalias_mutex); 196 197 /* protects napi_hash addition/deletion and napi_gen_id */ 198 static DEFINE_SPINLOCK(napi_hash_lock); 199 200 static unsigned int napi_gen_id = NR_CPUS; 201 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8); 202 203 static DECLARE_RWSEM(devnet_rename_sem); 204 205 static inline void dev_base_seq_inc(struct net *net) 206 { 207 while (++net->dev_base_seq == 0) 208 ; 209 } 210 211 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) 212 { 213 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ)); 214 215 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; 216 } 217 218 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) 219 { 220 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; 221 } 222 223 static inline void rps_lock(struct softnet_data *sd) 224 { 225 #ifdef CONFIG_RPS 226 spin_lock(&sd->input_pkt_queue.lock); 227 #endif 228 } 229 230 static inline void rps_unlock(struct softnet_data *sd) 231 { 232 #ifdef CONFIG_RPS 233 spin_unlock(&sd->input_pkt_queue.lock); 234 #endif 235 } 236 237 static struct netdev_name_node *netdev_name_node_alloc(struct net_device *dev, 238 const char *name) 239 { 240 struct netdev_name_node *name_node; 241 242 name_node = kmalloc(sizeof(*name_node), GFP_KERNEL); 243 if (!name_node) 244 return NULL; 245 INIT_HLIST_NODE(&name_node->hlist); 246 name_node->dev = dev; 247 name_node->name = name; 248 return name_node; 249 } 250 251 static struct netdev_name_node * 252 netdev_name_node_head_alloc(struct net_device *dev) 253 { 254 struct netdev_name_node *name_node; 255 256 name_node = netdev_name_node_alloc(dev, dev->name); 257 if (!name_node) 258 return NULL; 259 INIT_LIST_HEAD(&name_node->list); 260 return name_node; 261 } 262 263 static void netdev_name_node_free(struct netdev_name_node *name_node) 264 { 265 kfree(name_node); 266 } 267 268 static void netdev_name_node_add(struct net *net, 269 struct netdev_name_node *name_node) 270 { 271 hlist_add_head_rcu(&name_node->hlist, 272 dev_name_hash(net, name_node->name)); 273 } 274 275 static void netdev_name_node_del(struct netdev_name_node *name_node) 276 { 277 hlist_del_rcu(&name_node->hlist); 278 } 279 280 static struct netdev_name_node *netdev_name_node_lookup(struct net *net, 281 const char *name) 282 { 283 struct hlist_head *head = dev_name_hash(net, name); 284 struct netdev_name_node *name_node; 285 286 hlist_for_each_entry(name_node, head, hlist) 287 if (!strcmp(name_node->name, name)) 288 return name_node; 289 return NULL; 290 } 291 292 static struct netdev_name_node *netdev_name_node_lookup_rcu(struct net *net, 293 const char *name) 294 { 295 struct hlist_head *head = dev_name_hash(net, name); 296 struct netdev_name_node *name_node; 297 298 hlist_for_each_entry_rcu(name_node, head, hlist) 299 if (!strcmp(name_node->name, name)) 300 return name_node; 301 return NULL; 302 } 303 304 int netdev_name_node_alt_create(struct net_device *dev, const char *name) 305 { 306 struct netdev_name_node *name_node; 307 struct net *net = dev_net(dev); 308 309 name_node = netdev_name_node_lookup(net, name); 310 if (name_node) 311 return -EEXIST; 312 name_node = netdev_name_node_alloc(dev, name); 313 if (!name_node) 314 return -ENOMEM; 315 netdev_name_node_add(net, name_node); 316 /* The node that holds dev->name acts as a head of per-device list. */ 317 list_add_tail(&name_node->list, &dev->name_node->list); 318 319 return 0; 320 } 321 EXPORT_SYMBOL(netdev_name_node_alt_create); 322 323 static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node) 324 { 325 list_del(&name_node->list); 326 netdev_name_node_del(name_node); 327 kfree(name_node->name); 328 netdev_name_node_free(name_node); 329 } 330 331 int netdev_name_node_alt_destroy(struct net_device *dev, const char *name) 332 { 333 struct netdev_name_node *name_node; 334 struct net *net = dev_net(dev); 335 336 name_node = netdev_name_node_lookup(net, name); 337 if (!name_node) 338 return -ENOENT; 339 /* lookup might have found our primary name or a name belonging 340 * to another device. 341 */ 342 if (name_node == dev->name_node || name_node->dev != dev) 343 return -EINVAL; 344 345 __netdev_name_node_alt_destroy(name_node); 346 347 return 0; 348 } 349 EXPORT_SYMBOL(netdev_name_node_alt_destroy); 350 351 static void netdev_name_node_alt_flush(struct net_device *dev) 352 { 353 struct netdev_name_node *name_node, *tmp; 354 355 list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list) 356 __netdev_name_node_alt_destroy(name_node); 357 } 358 359 /* Device list insertion */ 360 static void list_netdevice(struct net_device *dev) 361 { 362 struct net *net = dev_net(dev); 363 364 ASSERT_RTNL(); 365 366 write_lock_bh(&dev_base_lock); 367 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); 368 netdev_name_node_add(net, dev->name_node); 369 hlist_add_head_rcu(&dev->index_hlist, 370 dev_index_hash(net, dev->ifindex)); 371 write_unlock_bh(&dev_base_lock); 372 373 dev_base_seq_inc(net); 374 } 375 376 /* Device list removal 377 * caller must respect a RCU grace period before freeing/reusing dev 378 */ 379 static void unlist_netdevice(struct net_device *dev) 380 { 381 ASSERT_RTNL(); 382 383 /* Unlink dev from the device chain */ 384 write_lock_bh(&dev_base_lock); 385 list_del_rcu(&dev->dev_list); 386 netdev_name_node_del(dev->name_node); 387 hlist_del_rcu(&dev->index_hlist); 388 write_unlock_bh(&dev_base_lock); 389 390 dev_base_seq_inc(dev_net(dev)); 391 } 392 393 /* 394 * Our notifier list 395 */ 396 397 static RAW_NOTIFIER_HEAD(netdev_chain); 398 399 /* 400 * Device drivers call our routines to queue packets here. We empty the 401 * queue in the local softnet handler. 402 */ 403 404 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); 405 EXPORT_PER_CPU_SYMBOL(softnet_data); 406 407 #ifdef CONFIG_LOCKDEP 408 /* 409 * register_netdevice() inits txq->_xmit_lock and sets lockdep class 410 * according to dev->type 411 */ 412 static const unsigned short netdev_lock_type[] = { 413 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, 414 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, 415 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, 416 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, 417 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, 418 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, 419 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, 420 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, 421 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, 422 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, 423 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, 424 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, 425 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, 426 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, 427 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; 428 429 static const char *const netdev_lock_name[] = { 430 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", 431 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", 432 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", 433 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", 434 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", 435 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", 436 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", 437 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", 438 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", 439 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", 440 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", 441 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", 442 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", 443 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", 444 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; 445 446 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; 447 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; 448 449 static inline unsigned short netdev_lock_pos(unsigned short dev_type) 450 { 451 int i; 452 453 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) 454 if (netdev_lock_type[i] == dev_type) 455 return i; 456 /* the last key is used by default */ 457 return ARRAY_SIZE(netdev_lock_type) - 1; 458 } 459 460 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 461 unsigned short dev_type) 462 { 463 int i; 464 465 i = netdev_lock_pos(dev_type); 466 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], 467 netdev_lock_name[i]); 468 } 469 470 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 471 { 472 int i; 473 474 i = netdev_lock_pos(dev->type); 475 lockdep_set_class_and_name(&dev->addr_list_lock, 476 &netdev_addr_lock_key[i], 477 netdev_lock_name[i]); 478 } 479 #else 480 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 481 unsigned short dev_type) 482 { 483 } 484 485 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 486 { 487 } 488 #endif 489 490 /******************************************************************************* 491 * 492 * Protocol management and registration routines 493 * 494 *******************************************************************************/ 495 496 497 /* 498 * Add a protocol ID to the list. Now that the input handler is 499 * smarter we can dispense with all the messy stuff that used to be 500 * here. 501 * 502 * BEWARE!!! Protocol handlers, mangling input packets, 503 * MUST BE last in hash buckets and checking protocol handlers 504 * MUST start from promiscuous ptype_all chain in net_bh. 505 * It is true now, do not change it. 506 * Explanation follows: if protocol handler, mangling packet, will 507 * be the first on list, it is not able to sense, that packet 508 * is cloned and should be copied-on-write, so that it will 509 * change it and subsequent readers will get broken packet. 510 * --ANK (980803) 511 */ 512 513 static inline struct list_head *ptype_head(const struct packet_type *pt) 514 { 515 if (pt->type == htons(ETH_P_ALL)) 516 return pt->dev ? &pt->dev->ptype_all : &ptype_all; 517 else 518 return pt->dev ? &pt->dev->ptype_specific : 519 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; 520 } 521 522 /** 523 * dev_add_pack - add packet handler 524 * @pt: packet type declaration 525 * 526 * Add a protocol handler to the networking stack. The passed &packet_type 527 * is linked into kernel lists and may not be freed until it has been 528 * removed from the kernel lists. 529 * 530 * This call does not sleep therefore it can not 531 * guarantee all CPU's that are in middle of receiving packets 532 * will see the new packet type (until the next received packet). 533 */ 534 535 void dev_add_pack(struct packet_type *pt) 536 { 537 struct list_head *head = ptype_head(pt); 538 539 spin_lock(&ptype_lock); 540 list_add_rcu(&pt->list, head); 541 spin_unlock(&ptype_lock); 542 } 543 EXPORT_SYMBOL(dev_add_pack); 544 545 /** 546 * __dev_remove_pack - remove packet handler 547 * @pt: packet type declaration 548 * 549 * Remove a protocol handler that was previously added to the kernel 550 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 551 * from the kernel lists and can be freed or reused once this function 552 * returns. 553 * 554 * The packet type might still be in use by receivers 555 * and must not be freed until after all the CPU's have gone 556 * through a quiescent state. 557 */ 558 void __dev_remove_pack(struct packet_type *pt) 559 { 560 struct list_head *head = ptype_head(pt); 561 struct packet_type *pt1; 562 563 spin_lock(&ptype_lock); 564 565 list_for_each_entry(pt1, head, list) { 566 if (pt == pt1) { 567 list_del_rcu(&pt->list); 568 goto out; 569 } 570 } 571 572 pr_warn("dev_remove_pack: %p not found\n", pt); 573 out: 574 spin_unlock(&ptype_lock); 575 } 576 EXPORT_SYMBOL(__dev_remove_pack); 577 578 /** 579 * dev_remove_pack - remove packet handler 580 * @pt: packet type declaration 581 * 582 * Remove a protocol handler that was previously added to the kernel 583 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 584 * from the kernel lists and can be freed or reused once this function 585 * returns. 586 * 587 * This call sleeps to guarantee that no CPU is looking at the packet 588 * type after return. 589 */ 590 void dev_remove_pack(struct packet_type *pt) 591 { 592 __dev_remove_pack(pt); 593 594 synchronize_net(); 595 } 596 EXPORT_SYMBOL(dev_remove_pack); 597 598 599 /** 600 * dev_add_offload - register offload handlers 601 * @po: protocol offload declaration 602 * 603 * Add protocol offload handlers to the networking stack. The passed 604 * &proto_offload is linked into kernel lists and may not be freed until 605 * it has been removed from the kernel lists. 606 * 607 * This call does not sleep therefore it can not 608 * guarantee all CPU's that are in middle of receiving packets 609 * will see the new offload handlers (until the next received packet). 610 */ 611 void dev_add_offload(struct packet_offload *po) 612 { 613 struct packet_offload *elem; 614 615 spin_lock(&offload_lock); 616 list_for_each_entry(elem, &offload_base, list) { 617 if (po->priority < elem->priority) 618 break; 619 } 620 list_add_rcu(&po->list, elem->list.prev); 621 spin_unlock(&offload_lock); 622 } 623 EXPORT_SYMBOL(dev_add_offload); 624 625 /** 626 * __dev_remove_offload - remove offload handler 627 * @po: packet offload declaration 628 * 629 * Remove a protocol offload handler that was previously added to the 630 * kernel offload handlers by dev_add_offload(). The passed &offload_type 631 * is removed from the kernel lists and can be freed or reused once this 632 * function returns. 633 * 634 * The packet type might still be in use by receivers 635 * and must not be freed until after all the CPU's have gone 636 * through a quiescent state. 637 */ 638 static void __dev_remove_offload(struct packet_offload *po) 639 { 640 struct list_head *head = &offload_base; 641 struct packet_offload *po1; 642 643 spin_lock(&offload_lock); 644 645 list_for_each_entry(po1, head, list) { 646 if (po == po1) { 647 list_del_rcu(&po->list); 648 goto out; 649 } 650 } 651 652 pr_warn("dev_remove_offload: %p not found\n", po); 653 out: 654 spin_unlock(&offload_lock); 655 } 656 657 /** 658 * dev_remove_offload - remove packet offload handler 659 * @po: packet offload declaration 660 * 661 * Remove a packet offload handler that was previously added to the kernel 662 * offload handlers by dev_add_offload(). The passed &offload_type is 663 * removed from the kernel lists and can be freed or reused once this 664 * function returns. 665 * 666 * This call sleeps to guarantee that no CPU is looking at the packet 667 * type after return. 668 */ 669 void dev_remove_offload(struct packet_offload *po) 670 { 671 __dev_remove_offload(po); 672 673 synchronize_net(); 674 } 675 EXPORT_SYMBOL(dev_remove_offload); 676 677 /****************************************************************************** 678 * 679 * Device Boot-time Settings Routines 680 * 681 ******************************************************************************/ 682 683 /* Boot time configuration table */ 684 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX]; 685 686 /** 687 * netdev_boot_setup_add - add new setup entry 688 * @name: name of the device 689 * @map: configured settings for the device 690 * 691 * Adds new setup entry to the dev_boot_setup list. The function 692 * returns 0 on error and 1 on success. This is a generic routine to 693 * all netdevices. 694 */ 695 static int netdev_boot_setup_add(char *name, struct ifmap *map) 696 { 697 struct netdev_boot_setup *s; 698 int i; 699 700 s = dev_boot_setup; 701 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { 702 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') { 703 memset(s[i].name, 0, sizeof(s[i].name)); 704 strlcpy(s[i].name, name, IFNAMSIZ); 705 memcpy(&s[i].map, map, sizeof(s[i].map)); 706 break; 707 } 708 } 709 710 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1; 711 } 712 713 /** 714 * netdev_boot_setup_check - check boot time settings 715 * @dev: the netdevice 716 * 717 * Check boot time settings for the device. 718 * The found settings are set for the device to be used 719 * later in the device probing. 720 * Returns 0 if no settings found, 1 if they are. 721 */ 722 int netdev_boot_setup_check(struct net_device *dev) 723 { 724 struct netdev_boot_setup *s = dev_boot_setup; 725 int i; 726 727 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { 728 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' && 729 !strcmp(dev->name, s[i].name)) { 730 dev->irq = s[i].map.irq; 731 dev->base_addr = s[i].map.base_addr; 732 dev->mem_start = s[i].map.mem_start; 733 dev->mem_end = s[i].map.mem_end; 734 return 1; 735 } 736 } 737 return 0; 738 } 739 EXPORT_SYMBOL(netdev_boot_setup_check); 740 741 742 /** 743 * netdev_boot_base - get address from boot time settings 744 * @prefix: prefix for network device 745 * @unit: id for network device 746 * 747 * Check boot time settings for the base address of device. 748 * The found settings are set for the device to be used 749 * later in the device probing. 750 * Returns 0 if no settings found. 751 */ 752 unsigned long netdev_boot_base(const char *prefix, int unit) 753 { 754 const struct netdev_boot_setup *s = dev_boot_setup; 755 char name[IFNAMSIZ]; 756 int i; 757 758 sprintf(name, "%s%d", prefix, unit); 759 760 /* 761 * If device already registered then return base of 1 762 * to indicate not to probe for this interface 763 */ 764 if (__dev_get_by_name(&init_net, name)) 765 return 1; 766 767 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) 768 if (!strcmp(name, s[i].name)) 769 return s[i].map.base_addr; 770 return 0; 771 } 772 773 /* 774 * Saves at boot time configured settings for any netdevice. 775 */ 776 int __init netdev_boot_setup(char *str) 777 { 778 int ints[5]; 779 struct ifmap map; 780 781 str = get_options(str, ARRAY_SIZE(ints), ints); 782 if (!str || !*str) 783 return 0; 784 785 /* Save settings */ 786 memset(&map, 0, sizeof(map)); 787 if (ints[0] > 0) 788 map.irq = ints[1]; 789 if (ints[0] > 1) 790 map.base_addr = ints[2]; 791 if (ints[0] > 2) 792 map.mem_start = ints[3]; 793 if (ints[0] > 3) 794 map.mem_end = ints[4]; 795 796 /* Add new entry to the list */ 797 return netdev_boot_setup_add(str, &map); 798 } 799 800 __setup("netdev=", netdev_boot_setup); 801 802 /******************************************************************************* 803 * 804 * Device Interface Subroutines 805 * 806 *******************************************************************************/ 807 808 /** 809 * dev_get_iflink - get 'iflink' value of a interface 810 * @dev: targeted interface 811 * 812 * Indicates the ifindex the interface is linked to. 813 * Physical interfaces have the same 'ifindex' and 'iflink' values. 814 */ 815 816 int dev_get_iflink(const struct net_device *dev) 817 { 818 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink) 819 return dev->netdev_ops->ndo_get_iflink(dev); 820 821 return dev->ifindex; 822 } 823 EXPORT_SYMBOL(dev_get_iflink); 824 825 /** 826 * dev_fill_metadata_dst - Retrieve tunnel egress information. 827 * @dev: targeted interface 828 * @skb: The packet. 829 * 830 * For better visibility of tunnel traffic OVS needs to retrieve 831 * egress tunnel information for a packet. Following API allows 832 * user to get this info. 833 */ 834 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) 835 { 836 struct ip_tunnel_info *info; 837 838 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst) 839 return -EINVAL; 840 841 info = skb_tunnel_info_unclone(skb); 842 if (!info) 843 return -ENOMEM; 844 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX))) 845 return -EINVAL; 846 847 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb); 848 } 849 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst); 850 851 /** 852 * __dev_get_by_name - find a device by its name 853 * @net: the applicable net namespace 854 * @name: name to find 855 * 856 * Find an interface by name. Must be called under RTNL semaphore 857 * or @dev_base_lock. If the name is found a pointer to the device 858 * is returned. If the name is not found then %NULL is returned. The 859 * reference counters are not incremented so the caller must be 860 * careful with locks. 861 */ 862 863 struct net_device *__dev_get_by_name(struct net *net, const char *name) 864 { 865 struct netdev_name_node *node_name; 866 867 node_name = netdev_name_node_lookup(net, name); 868 return node_name ? node_name->dev : NULL; 869 } 870 EXPORT_SYMBOL(__dev_get_by_name); 871 872 /** 873 * dev_get_by_name_rcu - find a device by its name 874 * @net: the applicable net namespace 875 * @name: name to find 876 * 877 * Find an interface by name. 878 * If the name is found a pointer to the device is returned. 879 * If the name is not found then %NULL is returned. 880 * The reference counters are not incremented so the caller must be 881 * careful with locks. The caller must hold RCU lock. 882 */ 883 884 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) 885 { 886 struct netdev_name_node *node_name; 887 888 node_name = netdev_name_node_lookup_rcu(net, name); 889 return node_name ? node_name->dev : NULL; 890 } 891 EXPORT_SYMBOL(dev_get_by_name_rcu); 892 893 /** 894 * dev_get_by_name - find a device by its name 895 * @net: the applicable net namespace 896 * @name: name to find 897 * 898 * Find an interface by name. This can be called from any 899 * context and does its own locking. The returned handle has 900 * the usage count incremented and the caller must use dev_put() to 901 * release it when it is no longer needed. %NULL is returned if no 902 * matching device is found. 903 */ 904 905 struct net_device *dev_get_by_name(struct net *net, const char *name) 906 { 907 struct net_device *dev; 908 909 rcu_read_lock(); 910 dev = dev_get_by_name_rcu(net, name); 911 if (dev) 912 dev_hold(dev); 913 rcu_read_unlock(); 914 return dev; 915 } 916 EXPORT_SYMBOL(dev_get_by_name); 917 918 /** 919 * __dev_get_by_index - find a device by its ifindex 920 * @net: the applicable net namespace 921 * @ifindex: index of device 922 * 923 * Search for an interface by index. Returns %NULL if the device 924 * is not found or a pointer to the device. The device has not 925 * had its reference counter increased so the caller must be careful 926 * about locking. The caller must hold either the RTNL semaphore 927 * or @dev_base_lock. 928 */ 929 930 struct net_device *__dev_get_by_index(struct net *net, int ifindex) 931 { 932 struct net_device *dev; 933 struct hlist_head *head = dev_index_hash(net, ifindex); 934 935 hlist_for_each_entry(dev, head, index_hlist) 936 if (dev->ifindex == ifindex) 937 return dev; 938 939 return NULL; 940 } 941 EXPORT_SYMBOL(__dev_get_by_index); 942 943 /** 944 * dev_get_by_index_rcu - find a device by its ifindex 945 * @net: the applicable net namespace 946 * @ifindex: index of device 947 * 948 * Search for an interface by index. Returns %NULL if the device 949 * is not found or a pointer to the device. The device has not 950 * had its reference counter increased so the caller must be careful 951 * about locking. The caller must hold RCU lock. 952 */ 953 954 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) 955 { 956 struct net_device *dev; 957 struct hlist_head *head = dev_index_hash(net, ifindex); 958 959 hlist_for_each_entry_rcu(dev, head, index_hlist) 960 if (dev->ifindex == ifindex) 961 return dev; 962 963 return NULL; 964 } 965 EXPORT_SYMBOL(dev_get_by_index_rcu); 966 967 968 /** 969 * dev_get_by_index - find a device by its ifindex 970 * @net: the applicable net namespace 971 * @ifindex: index of device 972 * 973 * Search for an interface by index. Returns NULL if the device 974 * is not found or a pointer to the device. The device returned has 975 * had a reference added and the pointer is safe until the user calls 976 * dev_put to indicate they have finished with it. 977 */ 978 979 struct net_device *dev_get_by_index(struct net *net, int ifindex) 980 { 981 struct net_device *dev; 982 983 rcu_read_lock(); 984 dev = dev_get_by_index_rcu(net, ifindex); 985 if (dev) 986 dev_hold(dev); 987 rcu_read_unlock(); 988 return dev; 989 } 990 EXPORT_SYMBOL(dev_get_by_index); 991 992 /** 993 * dev_get_by_napi_id - find a device by napi_id 994 * @napi_id: ID of the NAPI struct 995 * 996 * Search for an interface by NAPI ID. Returns %NULL if the device 997 * is not found or a pointer to the device. The device has not had 998 * its reference counter increased so the caller must be careful 999 * about locking. The caller must hold RCU lock. 1000 */ 1001 1002 struct net_device *dev_get_by_napi_id(unsigned int napi_id) 1003 { 1004 struct napi_struct *napi; 1005 1006 WARN_ON_ONCE(!rcu_read_lock_held()); 1007 1008 if (napi_id < MIN_NAPI_ID) 1009 return NULL; 1010 1011 napi = napi_by_id(napi_id); 1012 1013 return napi ? napi->dev : NULL; 1014 } 1015 EXPORT_SYMBOL(dev_get_by_napi_id); 1016 1017 /** 1018 * netdev_get_name - get a netdevice name, knowing its ifindex. 1019 * @net: network namespace 1020 * @name: a pointer to the buffer where the name will be stored. 1021 * @ifindex: the ifindex of the interface to get the name from. 1022 */ 1023 int netdev_get_name(struct net *net, char *name, int ifindex) 1024 { 1025 struct net_device *dev; 1026 int ret; 1027 1028 down_read(&devnet_rename_sem); 1029 rcu_read_lock(); 1030 1031 dev = dev_get_by_index_rcu(net, ifindex); 1032 if (!dev) { 1033 ret = -ENODEV; 1034 goto out; 1035 } 1036 1037 strcpy(name, dev->name); 1038 1039 ret = 0; 1040 out: 1041 rcu_read_unlock(); 1042 up_read(&devnet_rename_sem); 1043 return ret; 1044 } 1045 1046 /** 1047 * dev_getbyhwaddr_rcu - find a device by its hardware address 1048 * @net: the applicable net namespace 1049 * @type: media type of device 1050 * @ha: hardware address 1051 * 1052 * Search for an interface by MAC address. Returns NULL if the device 1053 * is not found or a pointer to the device. 1054 * The caller must hold RCU or RTNL. 1055 * The returned device has not had its ref count increased 1056 * and the caller must therefore be careful about locking 1057 * 1058 */ 1059 1060 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, 1061 const char *ha) 1062 { 1063 struct net_device *dev; 1064 1065 for_each_netdev_rcu(net, dev) 1066 if (dev->type == type && 1067 !memcmp(dev->dev_addr, ha, dev->addr_len)) 1068 return dev; 1069 1070 return NULL; 1071 } 1072 EXPORT_SYMBOL(dev_getbyhwaddr_rcu); 1073 1074 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) 1075 { 1076 struct net_device *dev, *ret = NULL; 1077 1078 rcu_read_lock(); 1079 for_each_netdev_rcu(net, dev) 1080 if (dev->type == type) { 1081 dev_hold(dev); 1082 ret = dev; 1083 break; 1084 } 1085 rcu_read_unlock(); 1086 return ret; 1087 } 1088 EXPORT_SYMBOL(dev_getfirstbyhwtype); 1089 1090 /** 1091 * __dev_get_by_flags - find any device with given flags 1092 * @net: the applicable net namespace 1093 * @if_flags: IFF_* values 1094 * @mask: bitmask of bits in if_flags to check 1095 * 1096 * Search for any interface with the given flags. Returns NULL if a device 1097 * is not found or a pointer to the device. Must be called inside 1098 * rtnl_lock(), and result refcount is unchanged. 1099 */ 1100 1101 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags, 1102 unsigned short mask) 1103 { 1104 struct net_device *dev, *ret; 1105 1106 ASSERT_RTNL(); 1107 1108 ret = NULL; 1109 for_each_netdev(net, dev) { 1110 if (((dev->flags ^ if_flags) & mask) == 0) { 1111 ret = dev; 1112 break; 1113 } 1114 } 1115 return ret; 1116 } 1117 EXPORT_SYMBOL(__dev_get_by_flags); 1118 1119 /** 1120 * dev_valid_name - check if name is okay for network device 1121 * @name: name string 1122 * 1123 * Network device names need to be valid file names to 1124 * allow sysfs to work. We also disallow any kind of 1125 * whitespace. 1126 */ 1127 bool dev_valid_name(const char *name) 1128 { 1129 if (*name == '\0') 1130 return false; 1131 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ) 1132 return false; 1133 if (!strcmp(name, ".") || !strcmp(name, "..")) 1134 return false; 1135 1136 while (*name) { 1137 if (*name == '/' || *name == ':' || isspace(*name)) 1138 return false; 1139 name++; 1140 } 1141 return true; 1142 } 1143 EXPORT_SYMBOL(dev_valid_name); 1144 1145 /** 1146 * __dev_alloc_name - allocate a name for a device 1147 * @net: network namespace to allocate the device name in 1148 * @name: name format string 1149 * @buf: scratch buffer and result name string 1150 * 1151 * Passed a format string - eg "lt%d" it will try and find a suitable 1152 * id. It scans list of devices to build up a free map, then chooses 1153 * the first empty slot. The caller must hold the dev_base or rtnl lock 1154 * while allocating the name and adding the device in order to avoid 1155 * duplicates. 1156 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1157 * Returns the number of the unit assigned or a negative errno code. 1158 */ 1159 1160 static int __dev_alloc_name(struct net *net, const char *name, char *buf) 1161 { 1162 int i = 0; 1163 const char *p; 1164 const int max_netdevices = 8*PAGE_SIZE; 1165 unsigned long *inuse; 1166 struct net_device *d; 1167 1168 if (!dev_valid_name(name)) 1169 return -EINVAL; 1170 1171 p = strchr(name, '%'); 1172 if (p) { 1173 /* 1174 * Verify the string as this thing may have come from 1175 * the user. There must be either one "%d" and no other "%" 1176 * characters. 1177 */ 1178 if (p[1] != 'd' || strchr(p + 2, '%')) 1179 return -EINVAL; 1180 1181 /* Use one page as a bit array of possible slots */ 1182 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); 1183 if (!inuse) 1184 return -ENOMEM; 1185 1186 for_each_netdev(net, d) { 1187 if (!sscanf(d->name, name, &i)) 1188 continue; 1189 if (i < 0 || i >= max_netdevices) 1190 continue; 1191 1192 /* avoid cases where sscanf is not exact inverse of printf */ 1193 snprintf(buf, IFNAMSIZ, name, i); 1194 if (!strncmp(buf, d->name, IFNAMSIZ)) 1195 set_bit(i, inuse); 1196 } 1197 1198 i = find_first_zero_bit(inuse, max_netdevices); 1199 free_page((unsigned long) inuse); 1200 } 1201 1202 snprintf(buf, IFNAMSIZ, name, i); 1203 if (!__dev_get_by_name(net, buf)) 1204 return i; 1205 1206 /* It is possible to run out of possible slots 1207 * when the name is long and there isn't enough space left 1208 * for the digits, or if all bits are used. 1209 */ 1210 return -ENFILE; 1211 } 1212 1213 static int dev_alloc_name_ns(struct net *net, 1214 struct net_device *dev, 1215 const char *name) 1216 { 1217 char buf[IFNAMSIZ]; 1218 int ret; 1219 1220 BUG_ON(!net); 1221 ret = __dev_alloc_name(net, name, buf); 1222 if (ret >= 0) 1223 strlcpy(dev->name, buf, IFNAMSIZ); 1224 return ret; 1225 } 1226 1227 /** 1228 * dev_alloc_name - allocate a name for a device 1229 * @dev: device 1230 * @name: name format string 1231 * 1232 * Passed a format string - eg "lt%d" it will try and find a suitable 1233 * id. It scans list of devices to build up a free map, then chooses 1234 * the first empty slot. The caller must hold the dev_base or rtnl lock 1235 * while allocating the name and adding the device in order to avoid 1236 * duplicates. 1237 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1238 * Returns the number of the unit assigned or a negative errno code. 1239 */ 1240 1241 int dev_alloc_name(struct net_device *dev, const char *name) 1242 { 1243 return dev_alloc_name_ns(dev_net(dev), dev, name); 1244 } 1245 EXPORT_SYMBOL(dev_alloc_name); 1246 1247 static int dev_get_valid_name(struct net *net, struct net_device *dev, 1248 const char *name) 1249 { 1250 BUG_ON(!net); 1251 1252 if (!dev_valid_name(name)) 1253 return -EINVAL; 1254 1255 if (strchr(name, '%')) 1256 return dev_alloc_name_ns(net, dev, name); 1257 else if (__dev_get_by_name(net, name)) 1258 return -EEXIST; 1259 else if (dev->name != name) 1260 strlcpy(dev->name, name, IFNAMSIZ); 1261 1262 return 0; 1263 } 1264 1265 /** 1266 * dev_change_name - change name of a device 1267 * @dev: device 1268 * @newname: name (or format string) must be at least IFNAMSIZ 1269 * 1270 * Change name of a device, can pass format strings "eth%d". 1271 * for wildcarding. 1272 */ 1273 int dev_change_name(struct net_device *dev, const char *newname) 1274 { 1275 unsigned char old_assign_type; 1276 char oldname[IFNAMSIZ]; 1277 int err = 0; 1278 int ret; 1279 struct net *net; 1280 1281 ASSERT_RTNL(); 1282 BUG_ON(!dev_net(dev)); 1283 1284 net = dev_net(dev); 1285 1286 /* Some auto-enslaved devices e.g. failover slaves are 1287 * special, as userspace might rename the device after 1288 * the interface had been brought up and running since 1289 * the point kernel initiated auto-enslavement. Allow 1290 * live name change even when these slave devices are 1291 * up and running. 1292 * 1293 * Typically, users of these auto-enslaving devices 1294 * don't actually care about slave name change, as 1295 * they are supposed to operate on master interface 1296 * directly. 1297 */ 1298 if (dev->flags & IFF_UP && 1299 likely(!(dev->priv_flags & IFF_LIVE_RENAME_OK))) 1300 return -EBUSY; 1301 1302 down_write(&devnet_rename_sem); 1303 1304 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { 1305 up_write(&devnet_rename_sem); 1306 return 0; 1307 } 1308 1309 memcpy(oldname, dev->name, IFNAMSIZ); 1310 1311 err = dev_get_valid_name(net, dev, newname); 1312 if (err < 0) { 1313 up_write(&devnet_rename_sem); 1314 return err; 1315 } 1316 1317 if (oldname[0] && !strchr(oldname, '%')) 1318 netdev_info(dev, "renamed from %s\n", oldname); 1319 1320 old_assign_type = dev->name_assign_type; 1321 dev->name_assign_type = NET_NAME_RENAMED; 1322 1323 rollback: 1324 ret = device_rename(&dev->dev, dev->name); 1325 if (ret) { 1326 memcpy(dev->name, oldname, IFNAMSIZ); 1327 dev->name_assign_type = old_assign_type; 1328 up_write(&devnet_rename_sem); 1329 return ret; 1330 } 1331 1332 up_write(&devnet_rename_sem); 1333 1334 netdev_adjacent_rename_links(dev, oldname); 1335 1336 write_lock_bh(&dev_base_lock); 1337 netdev_name_node_del(dev->name_node); 1338 write_unlock_bh(&dev_base_lock); 1339 1340 synchronize_rcu(); 1341 1342 write_lock_bh(&dev_base_lock); 1343 netdev_name_node_add(net, dev->name_node); 1344 write_unlock_bh(&dev_base_lock); 1345 1346 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); 1347 ret = notifier_to_errno(ret); 1348 1349 if (ret) { 1350 /* err >= 0 after dev_alloc_name() or stores the first errno */ 1351 if (err >= 0) { 1352 err = ret; 1353 down_write(&devnet_rename_sem); 1354 memcpy(dev->name, oldname, IFNAMSIZ); 1355 memcpy(oldname, newname, IFNAMSIZ); 1356 dev->name_assign_type = old_assign_type; 1357 old_assign_type = NET_NAME_RENAMED; 1358 goto rollback; 1359 } else { 1360 pr_err("%s: name change rollback failed: %d\n", 1361 dev->name, ret); 1362 } 1363 } 1364 1365 return err; 1366 } 1367 1368 /** 1369 * dev_set_alias - change ifalias of a device 1370 * @dev: device 1371 * @alias: name up to IFALIASZ 1372 * @len: limit of bytes to copy from info 1373 * 1374 * Set ifalias for a device, 1375 */ 1376 int dev_set_alias(struct net_device *dev, const char *alias, size_t len) 1377 { 1378 struct dev_ifalias *new_alias = NULL; 1379 1380 if (len >= IFALIASZ) 1381 return -EINVAL; 1382 1383 if (len) { 1384 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL); 1385 if (!new_alias) 1386 return -ENOMEM; 1387 1388 memcpy(new_alias->ifalias, alias, len); 1389 new_alias->ifalias[len] = 0; 1390 } 1391 1392 mutex_lock(&ifalias_mutex); 1393 new_alias = rcu_replace_pointer(dev->ifalias, new_alias, 1394 mutex_is_locked(&ifalias_mutex)); 1395 mutex_unlock(&ifalias_mutex); 1396 1397 if (new_alias) 1398 kfree_rcu(new_alias, rcuhead); 1399 1400 return len; 1401 } 1402 EXPORT_SYMBOL(dev_set_alias); 1403 1404 /** 1405 * dev_get_alias - get ifalias of a device 1406 * @dev: device 1407 * @name: buffer to store name of ifalias 1408 * @len: size of buffer 1409 * 1410 * get ifalias for a device. Caller must make sure dev cannot go 1411 * away, e.g. rcu read lock or own a reference count to device. 1412 */ 1413 int dev_get_alias(const struct net_device *dev, char *name, size_t len) 1414 { 1415 const struct dev_ifalias *alias; 1416 int ret = 0; 1417 1418 rcu_read_lock(); 1419 alias = rcu_dereference(dev->ifalias); 1420 if (alias) 1421 ret = snprintf(name, len, "%s", alias->ifalias); 1422 rcu_read_unlock(); 1423 1424 return ret; 1425 } 1426 1427 /** 1428 * netdev_features_change - device changes features 1429 * @dev: device to cause notification 1430 * 1431 * Called to indicate a device has changed features. 1432 */ 1433 void netdev_features_change(struct net_device *dev) 1434 { 1435 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); 1436 } 1437 EXPORT_SYMBOL(netdev_features_change); 1438 1439 /** 1440 * netdev_state_change - device changes state 1441 * @dev: device to cause notification 1442 * 1443 * Called to indicate a device has changed state. This function calls 1444 * the notifier chains for netdev_chain and sends a NEWLINK message 1445 * to the routing socket. 1446 */ 1447 void netdev_state_change(struct net_device *dev) 1448 { 1449 if (dev->flags & IFF_UP) { 1450 struct netdev_notifier_change_info change_info = { 1451 .info.dev = dev, 1452 }; 1453 1454 call_netdevice_notifiers_info(NETDEV_CHANGE, 1455 &change_info.info); 1456 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL); 1457 } 1458 } 1459 EXPORT_SYMBOL(netdev_state_change); 1460 1461 /** 1462 * __netdev_notify_peers - notify network peers about existence of @dev, 1463 * to be called when rtnl lock is already held. 1464 * @dev: network device 1465 * 1466 * Generate traffic such that interested network peers are aware of 1467 * @dev, such as by generating a gratuitous ARP. This may be used when 1468 * a device wants to inform the rest of the network about some sort of 1469 * reconfiguration such as a failover event or virtual machine 1470 * migration. 1471 */ 1472 void __netdev_notify_peers(struct net_device *dev) 1473 { 1474 ASSERT_RTNL(); 1475 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); 1476 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev); 1477 } 1478 EXPORT_SYMBOL(__netdev_notify_peers); 1479 1480 /** 1481 * netdev_notify_peers - notify network peers about existence of @dev 1482 * @dev: network device 1483 * 1484 * Generate traffic such that interested network peers are aware of 1485 * @dev, such as by generating a gratuitous ARP. This may be used when 1486 * a device wants to inform the rest of the network about some sort of 1487 * reconfiguration such as a failover event or virtual machine 1488 * migration. 1489 */ 1490 void netdev_notify_peers(struct net_device *dev) 1491 { 1492 rtnl_lock(); 1493 __netdev_notify_peers(dev); 1494 rtnl_unlock(); 1495 } 1496 EXPORT_SYMBOL(netdev_notify_peers); 1497 1498 static int napi_threaded_poll(void *data); 1499 1500 static int napi_kthread_create(struct napi_struct *n) 1501 { 1502 int err = 0; 1503 1504 /* Create and wake up the kthread once to put it in 1505 * TASK_INTERRUPTIBLE mode to avoid the blocked task 1506 * warning and work with loadavg. 1507 */ 1508 n->thread = kthread_run(napi_threaded_poll, n, "napi/%s-%d", 1509 n->dev->name, n->napi_id); 1510 if (IS_ERR(n->thread)) { 1511 err = PTR_ERR(n->thread); 1512 pr_err("kthread_run failed with err %d\n", err); 1513 n->thread = NULL; 1514 } 1515 1516 return err; 1517 } 1518 1519 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1520 { 1521 const struct net_device_ops *ops = dev->netdev_ops; 1522 int ret; 1523 1524 ASSERT_RTNL(); 1525 1526 if (!netif_device_present(dev)) { 1527 /* may be detached because parent is runtime-suspended */ 1528 if (dev->dev.parent) 1529 pm_runtime_resume(dev->dev.parent); 1530 if (!netif_device_present(dev)) 1531 return -ENODEV; 1532 } 1533 1534 /* Block netpoll from trying to do any rx path servicing. 1535 * If we don't do this there is a chance ndo_poll_controller 1536 * or ndo_poll may be running while we open the device 1537 */ 1538 netpoll_poll_disable(dev); 1539 1540 ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack); 1541 ret = notifier_to_errno(ret); 1542 if (ret) 1543 return ret; 1544 1545 set_bit(__LINK_STATE_START, &dev->state); 1546 1547 if (ops->ndo_validate_addr) 1548 ret = ops->ndo_validate_addr(dev); 1549 1550 if (!ret && ops->ndo_open) 1551 ret = ops->ndo_open(dev); 1552 1553 netpoll_poll_enable(dev); 1554 1555 if (ret) 1556 clear_bit(__LINK_STATE_START, &dev->state); 1557 else { 1558 dev->flags |= IFF_UP; 1559 dev_set_rx_mode(dev); 1560 dev_activate(dev); 1561 add_device_randomness(dev->dev_addr, dev->addr_len); 1562 } 1563 1564 return ret; 1565 } 1566 1567 /** 1568 * dev_open - prepare an interface for use. 1569 * @dev: device to open 1570 * @extack: netlink extended ack 1571 * 1572 * Takes a device from down to up state. The device's private open 1573 * function is invoked and then the multicast lists are loaded. Finally 1574 * the device is moved into the up state and a %NETDEV_UP message is 1575 * sent to the netdev notifier chain. 1576 * 1577 * Calling this function on an active interface is a nop. On a failure 1578 * a negative errno code is returned. 1579 */ 1580 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1581 { 1582 int ret; 1583 1584 if (dev->flags & IFF_UP) 1585 return 0; 1586 1587 ret = __dev_open(dev, extack); 1588 if (ret < 0) 1589 return ret; 1590 1591 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); 1592 call_netdevice_notifiers(NETDEV_UP, dev); 1593 1594 return ret; 1595 } 1596 EXPORT_SYMBOL(dev_open); 1597 1598 static void __dev_close_many(struct list_head *head) 1599 { 1600 struct net_device *dev; 1601 1602 ASSERT_RTNL(); 1603 might_sleep(); 1604 1605 list_for_each_entry(dev, head, close_list) { 1606 /* Temporarily disable netpoll until the interface is down */ 1607 netpoll_poll_disable(dev); 1608 1609 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); 1610 1611 clear_bit(__LINK_STATE_START, &dev->state); 1612 1613 /* Synchronize to scheduled poll. We cannot touch poll list, it 1614 * can be even on different cpu. So just clear netif_running(). 1615 * 1616 * dev->stop() will invoke napi_disable() on all of it's 1617 * napi_struct instances on this device. 1618 */ 1619 smp_mb__after_atomic(); /* Commit netif_running(). */ 1620 } 1621 1622 dev_deactivate_many(head); 1623 1624 list_for_each_entry(dev, head, close_list) { 1625 const struct net_device_ops *ops = dev->netdev_ops; 1626 1627 /* 1628 * Call the device specific close. This cannot fail. 1629 * Only if device is UP 1630 * 1631 * We allow it to be called even after a DETACH hot-plug 1632 * event. 1633 */ 1634 if (ops->ndo_stop) 1635 ops->ndo_stop(dev); 1636 1637 dev->flags &= ~IFF_UP; 1638 netpoll_poll_enable(dev); 1639 } 1640 } 1641 1642 static void __dev_close(struct net_device *dev) 1643 { 1644 LIST_HEAD(single); 1645 1646 list_add(&dev->close_list, &single); 1647 __dev_close_many(&single); 1648 list_del(&single); 1649 } 1650 1651 void dev_close_many(struct list_head *head, bool unlink) 1652 { 1653 struct net_device *dev, *tmp; 1654 1655 /* Remove the devices that don't need to be closed */ 1656 list_for_each_entry_safe(dev, tmp, head, close_list) 1657 if (!(dev->flags & IFF_UP)) 1658 list_del_init(&dev->close_list); 1659 1660 __dev_close_many(head); 1661 1662 list_for_each_entry_safe(dev, tmp, head, close_list) { 1663 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); 1664 call_netdevice_notifiers(NETDEV_DOWN, dev); 1665 if (unlink) 1666 list_del_init(&dev->close_list); 1667 } 1668 } 1669 EXPORT_SYMBOL(dev_close_many); 1670 1671 /** 1672 * dev_close - shutdown an interface. 1673 * @dev: device to shutdown 1674 * 1675 * This function moves an active device into down state. A 1676 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device 1677 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier 1678 * chain. 1679 */ 1680 void dev_close(struct net_device *dev) 1681 { 1682 if (dev->flags & IFF_UP) { 1683 LIST_HEAD(single); 1684 1685 list_add(&dev->close_list, &single); 1686 dev_close_many(&single, true); 1687 list_del(&single); 1688 } 1689 } 1690 EXPORT_SYMBOL(dev_close); 1691 1692 1693 /** 1694 * dev_disable_lro - disable Large Receive Offload on a device 1695 * @dev: device 1696 * 1697 * Disable Large Receive Offload (LRO) on a net device. Must be 1698 * called under RTNL. This is needed if received packets may be 1699 * forwarded to another interface. 1700 */ 1701 void dev_disable_lro(struct net_device *dev) 1702 { 1703 struct net_device *lower_dev; 1704 struct list_head *iter; 1705 1706 dev->wanted_features &= ~NETIF_F_LRO; 1707 netdev_update_features(dev); 1708 1709 if (unlikely(dev->features & NETIF_F_LRO)) 1710 netdev_WARN(dev, "failed to disable LRO!\n"); 1711 1712 netdev_for_each_lower_dev(dev, lower_dev, iter) 1713 dev_disable_lro(lower_dev); 1714 } 1715 EXPORT_SYMBOL(dev_disable_lro); 1716 1717 /** 1718 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device 1719 * @dev: device 1720 * 1721 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be 1722 * called under RTNL. This is needed if Generic XDP is installed on 1723 * the device. 1724 */ 1725 static void dev_disable_gro_hw(struct net_device *dev) 1726 { 1727 dev->wanted_features &= ~NETIF_F_GRO_HW; 1728 netdev_update_features(dev); 1729 1730 if (unlikely(dev->features & NETIF_F_GRO_HW)) 1731 netdev_WARN(dev, "failed to disable GRO_HW!\n"); 1732 } 1733 1734 const char *netdev_cmd_to_name(enum netdev_cmd cmd) 1735 { 1736 #define N(val) \ 1737 case NETDEV_##val: \ 1738 return "NETDEV_" __stringify(val); 1739 switch (cmd) { 1740 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER) 1741 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE) 1742 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE) 1743 N(POST_INIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) N(CHANGEUPPER) 1744 N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) N(BONDING_INFO) 1745 N(PRECHANGEUPPER) N(CHANGELOWERSTATE) N(UDP_TUNNEL_PUSH_INFO) 1746 N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN) 1747 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO) 1748 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO) 1749 N(PRE_CHANGEADDR) 1750 } 1751 #undef N 1752 return "UNKNOWN_NETDEV_EVENT"; 1753 } 1754 EXPORT_SYMBOL_GPL(netdev_cmd_to_name); 1755 1756 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, 1757 struct net_device *dev) 1758 { 1759 struct netdev_notifier_info info = { 1760 .dev = dev, 1761 }; 1762 1763 return nb->notifier_call(nb, val, &info); 1764 } 1765 1766 static int call_netdevice_register_notifiers(struct notifier_block *nb, 1767 struct net_device *dev) 1768 { 1769 int err; 1770 1771 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); 1772 err = notifier_to_errno(err); 1773 if (err) 1774 return err; 1775 1776 if (!(dev->flags & IFF_UP)) 1777 return 0; 1778 1779 call_netdevice_notifier(nb, NETDEV_UP, dev); 1780 return 0; 1781 } 1782 1783 static void call_netdevice_unregister_notifiers(struct notifier_block *nb, 1784 struct net_device *dev) 1785 { 1786 if (dev->flags & IFF_UP) { 1787 call_netdevice_notifier(nb, NETDEV_GOING_DOWN, 1788 dev); 1789 call_netdevice_notifier(nb, NETDEV_DOWN, dev); 1790 } 1791 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); 1792 } 1793 1794 static int call_netdevice_register_net_notifiers(struct notifier_block *nb, 1795 struct net *net) 1796 { 1797 struct net_device *dev; 1798 int err; 1799 1800 for_each_netdev(net, dev) { 1801 err = call_netdevice_register_notifiers(nb, dev); 1802 if (err) 1803 goto rollback; 1804 } 1805 return 0; 1806 1807 rollback: 1808 for_each_netdev_continue_reverse(net, dev) 1809 call_netdevice_unregister_notifiers(nb, dev); 1810 return err; 1811 } 1812 1813 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb, 1814 struct net *net) 1815 { 1816 struct net_device *dev; 1817 1818 for_each_netdev(net, dev) 1819 call_netdevice_unregister_notifiers(nb, dev); 1820 } 1821 1822 static int dev_boot_phase = 1; 1823 1824 /** 1825 * register_netdevice_notifier - register a network notifier block 1826 * @nb: notifier 1827 * 1828 * Register a notifier to be called when network device events occur. 1829 * The notifier passed is linked into the kernel structures and must 1830 * not be reused until it has been unregistered. A negative errno code 1831 * is returned on a failure. 1832 * 1833 * When registered all registration and up events are replayed 1834 * to the new notifier to allow device to have a race free 1835 * view of the network device list. 1836 */ 1837 1838 int register_netdevice_notifier(struct notifier_block *nb) 1839 { 1840 struct net *net; 1841 int err; 1842 1843 /* Close race with setup_net() and cleanup_net() */ 1844 down_write(&pernet_ops_rwsem); 1845 rtnl_lock(); 1846 err = raw_notifier_chain_register(&netdev_chain, nb); 1847 if (err) 1848 goto unlock; 1849 if (dev_boot_phase) 1850 goto unlock; 1851 for_each_net(net) { 1852 err = call_netdevice_register_net_notifiers(nb, net); 1853 if (err) 1854 goto rollback; 1855 } 1856 1857 unlock: 1858 rtnl_unlock(); 1859 up_write(&pernet_ops_rwsem); 1860 return err; 1861 1862 rollback: 1863 for_each_net_continue_reverse(net) 1864 call_netdevice_unregister_net_notifiers(nb, net); 1865 1866 raw_notifier_chain_unregister(&netdev_chain, nb); 1867 goto unlock; 1868 } 1869 EXPORT_SYMBOL(register_netdevice_notifier); 1870 1871 /** 1872 * unregister_netdevice_notifier - unregister a network notifier block 1873 * @nb: notifier 1874 * 1875 * Unregister a notifier previously registered by 1876 * register_netdevice_notifier(). The notifier is unlinked into the 1877 * kernel structures and may then be reused. A negative errno code 1878 * is returned on a failure. 1879 * 1880 * After unregistering unregister and down device events are synthesized 1881 * for all devices on the device list to the removed notifier to remove 1882 * the need for special case cleanup code. 1883 */ 1884 1885 int unregister_netdevice_notifier(struct notifier_block *nb) 1886 { 1887 struct net *net; 1888 int err; 1889 1890 /* Close race with setup_net() and cleanup_net() */ 1891 down_write(&pernet_ops_rwsem); 1892 rtnl_lock(); 1893 err = raw_notifier_chain_unregister(&netdev_chain, nb); 1894 if (err) 1895 goto unlock; 1896 1897 for_each_net(net) 1898 call_netdevice_unregister_net_notifiers(nb, net); 1899 1900 unlock: 1901 rtnl_unlock(); 1902 up_write(&pernet_ops_rwsem); 1903 return err; 1904 } 1905 EXPORT_SYMBOL(unregister_netdevice_notifier); 1906 1907 static int __register_netdevice_notifier_net(struct net *net, 1908 struct notifier_block *nb, 1909 bool ignore_call_fail) 1910 { 1911 int err; 1912 1913 err = raw_notifier_chain_register(&net->netdev_chain, nb); 1914 if (err) 1915 return err; 1916 if (dev_boot_phase) 1917 return 0; 1918 1919 err = call_netdevice_register_net_notifiers(nb, net); 1920 if (err && !ignore_call_fail) 1921 goto chain_unregister; 1922 1923 return 0; 1924 1925 chain_unregister: 1926 raw_notifier_chain_unregister(&net->netdev_chain, nb); 1927 return err; 1928 } 1929 1930 static int __unregister_netdevice_notifier_net(struct net *net, 1931 struct notifier_block *nb) 1932 { 1933 int err; 1934 1935 err = raw_notifier_chain_unregister(&net->netdev_chain, nb); 1936 if (err) 1937 return err; 1938 1939 call_netdevice_unregister_net_notifiers(nb, net); 1940 return 0; 1941 } 1942 1943 /** 1944 * register_netdevice_notifier_net - register a per-netns network notifier block 1945 * @net: network namespace 1946 * @nb: notifier 1947 * 1948 * Register a notifier to be called when network device events occur. 1949 * The notifier passed is linked into the kernel structures and must 1950 * not be reused until it has been unregistered. A negative errno code 1951 * is returned on a failure. 1952 * 1953 * When registered all registration and up events are replayed 1954 * to the new notifier to allow device to have a race free 1955 * view of the network device list. 1956 */ 1957 1958 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb) 1959 { 1960 int err; 1961 1962 rtnl_lock(); 1963 err = __register_netdevice_notifier_net(net, nb, false); 1964 rtnl_unlock(); 1965 return err; 1966 } 1967 EXPORT_SYMBOL(register_netdevice_notifier_net); 1968 1969 /** 1970 * unregister_netdevice_notifier_net - unregister a per-netns 1971 * network notifier block 1972 * @net: network namespace 1973 * @nb: notifier 1974 * 1975 * Unregister a notifier previously registered by 1976 * register_netdevice_notifier(). The notifier is unlinked into the 1977 * kernel structures and may then be reused. A negative errno code 1978 * is returned on a failure. 1979 * 1980 * After unregistering unregister and down device events are synthesized 1981 * for all devices on the device list to the removed notifier to remove 1982 * the need for special case cleanup code. 1983 */ 1984 1985 int unregister_netdevice_notifier_net(struct net *net, 1986 struct notifier_block *nb) 1987 { 1988 int err; 1989 1990 rtnl_lock(); 1991 err = __unregister_netdevice_notifier_net(net, nb); 1992 rtnl_unlock(); 1993 return err; 1994 } 1995 EXPORT_SYMBOL(unregister_netdevice_notifier_net); 1996 1997 int register_netdevice_notifier_dev_net(struct net_device *dev, 1998 struct notifier_block *nb, 1999 struct netdev_net_notifier *nn) 2000 { 2001 int err; 2002 2003 rtnl_lock(); 2004 err = __register_netdevice_notifier_net(dev_net(dev), nb, false); 2005 if (!err) { 2006 nn->nb = nb; 2007 list_add(&nn->list, &dev->net_notifier_list); 2008 } 2009 rtnl_unlock(); 2010 return err; 2011 } 2012 EXPORT_SYMBOL(register_netdevice_notifier_dev_net); 2013 2014 int unregister_netdevice_notifier_dev_net(struct net_device *dev, 2015 struct notifier_block *nb, 2016 struct netdev_net_notifier *nn) 2017 { 2018 int err; 2019 2020 rtnl_lock(); 2021 list_del(&nn->list); 2022 err = __unregister_netdevice_notifier_net(dev_net(dev), nb); 2023 rtnl_unlock(); 2024 return err; 2025 } 2026 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net); 2027 2028 static void move_netdevice_notifiers_dev_net(struct net_device *dev, 2029 struct net *net) 2030 { 2031 struct netdev_net_notifier *nn; 2032 2033 list_for_each_entry(nn, &dev->net_notifier_list, list) { 2034 __unregister_netdevice_notifier_net(dev_net(dev), nn->nb); 2035 __register_netdevice_notifier_net(net, nn->nb, true); 2036 } 2037 } 2038 2039 /** 2040 * call_netdevice_notifiers_info - call all network notifier blocks 2041 * @val: value passed unmodified to notifier function 2042 * @info: notifier information data 2043 * 2044 * Call all network notifier blocks. Parameters and return value 2045 * are as for raw_notifier_call_chain(). 2046 */ 2047 2048 static int call_netdevice_notifiers_info(unsigned long val, 2049 struct netdev_notifier_info *info) 2050 { 2051 struct net *net = dev_net(info->dev); 2052 int ret; 2053 2054 ASSERT_RTNL(); 2055 2056 /* Run per-netns notifier block chain first, then run the global one. 2057 * Hopefully, one day, the global one is going to be removed after 2058 * all notifier block registrators get converted to be per-netns. 2059 */ 2060 ret = raw_notifier_call_chain(&net->netdev_chain, val, info); 2061 if (ret & NOTIFY_STOP_MASK) 2062 return ret; 2063 return raw_notifier_call_chain(&netdev_chain, val, info); 2064 } 2065 2066 static int call_netdevice_notifiers_extack(unsigned long val, 2067 struct net_device *dev, 2068 struct netlink_ext_ack *extack) 2069 { 2070 struct netdev_notifier_info info = { 2071 .dev = dev, 2072 .extack = extack, 2073 }; 2074 2075 return call_netdevice_notifiers_info(val, &info); 2076 } 2077 2078 /** 2079 * call_netdevice_notifiers - call all network notifier blocks 2080 * @val: value passed unmodified to notifier function 2081 * @dev: net_device pointer passed unmodified to notifier function 2082 * 2083 * Call all network notifier blocks. Parameters and return value 2084 * are as for raw_notifier_call_chain(). 2085 */ 2086 2087 int call_netdevice_notifiers(unsigned long val, struct net_device *dev) 2088 { 2089 return call_netdevice_notifiers_extack(val, dev, NULL); 2090 } 2091 EXPORT_SYMBOL(call_netdevice_notifiers); 2092 2093 /** 2094 * call_netdevice_notifiers_mtu - call all network notifier blocks 2095 * @val: value passed unmodified to notifier function 2096 * @dev: net_device pointer passed unmodified to notifier function 2097 * @arg: additional u32 argument passed to the notifier function 2098 * 2099 * Call all network notifier blocks. Parameters and return value 2100 * are as for raw_notifier_call_chain(). 2101 */ 2102 static int call_netdevice_notifiers_mtu(unsigned long val, 2103 struct net_device *dev, u32 arg) 2104 { 2105 struct netdev_notifier_info_ext info = { 2106 .info.dev = dev, 2107 .ext.mtu = arg, 2108 }; 2109 2110 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0); 2111 2112 return call_netdevice_notifiers_info(val, &info.info); 2113 } 2114 2115 #ifdef CONFIG_NET_INGRESS 2116 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key); 2117 2118 void net_inc_ingress_queue(void) 2119 { 2120 static_branch_inc(&ingress_needed_key); 2121 } 2122 EXPORT_SYMBOL_GPL(net_inc_ingress_queue); 2123 2124 void net_dec_ingress_queue(void) 2125 { 2126 static_branch_dec(&ingress_needed_key); 2127 } 2128 EXPORT_SYMBOL_GPL(net_dec_ingress_queue); 2129 #endif 2130 2131 #ifdef CONFIG_NET_EGRESS 2132 static DEFINE_STATIC_KEY_FALSE(egress_needed_key); 2133 2134 void net_inc_egress_queue(void) 2135 { 2136 static_branch_inc(&egress_needed_key); 2137 } 2138 EXPORT_SYMBOL_GPL(net_inc_egress_queue); 2139 2140 void net_dec_egress_queue(void) 2141 { 2142 static_branch_dec(&egress_needed_key); 2143 } 2144 EXPORT_SYMBOL_GPL(net_dec_egress_queue); 2145 #endif 2146 2147 static DEFINE_STATIC_KEY_FALSE(netstamp_needed_key); 2148 #ifdef CONFIG_JUMP_LABEL 2149 static atomic_t netstamp_needed_deferred; 2150 static atomic_t netstamp_wanted; 2151 static void netstamp_clear(struct work_struct *work) 2152 { 2153 int deferred = atomic_xchg(&netstamp_needed_deferred, 0); 2154 int wanted; 2155 2156 wanted = atomic_add_return(deferred, &netstamp_wanted); 2157 if (wanted > 0) 2158 static_branch_enable(&netstamp_needed_key); 2159 else 2160 static_branch_disable(&netstamp_needed_key); 2161 } 2162 static DECLARE_WORK(netstamp_work, netstamp_clear); 2163 #endif 2164 2165 void net_enable_timestamp(void) 2166 { 2167 #ifdef CONFIG_JUMP_LABEL 2168 int wanted; 2169 2170 while (1) { 2171 wanted = atomic_read(&netstamp_wanted); 2172 if (wanted <= 0) 2173 break; 2174 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted) 2175 return; 2176 } 2177 atomic_inc(&netstamp_needed_deferred); 2178 schedule_work(&netstamp_work); 2179 #else 2180 static_branch_inc(&netstamp_needed_key); 2181 #endif 2182 } 2183 EXPORT_SYMBOL(net_enable_timestamp); 2184 2185 void net_disable_timestamp(void) 2186 { 2187 #ifdef CONFIG_JUMP_LABEL 2188 int wanted; 2189 2190 while (1) { 2191 wanted = atomic_read(&netstamp_wanted); 2192 if (wanted <= 1) 2193 break; 2194 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted) 2195 return; 2196 } 2197 atomic_dec(&netstamp_needed_deferred); 2198 schedule_work(&netstamp_work); 2199 #else 2200 static_branch_dec(&netstamp_needed_key); 2201 #endif 2202 } 2203 EXPORT_SYMBOL(net_disable_timestamp); 2204 2205 static inline void net_timestamp_set(struct sk_buff *skb) 2206 { 2207 skb->tstamp = 0; 2208 if (static_branch_unlikely(&netstamp_needed_key)) 2209 __net_timestamp(skb); 2210 } 2211 2212 #define net_timestamp_check(COND, SKB) \ 2213 if (static_branch_unlikely(&netstamp_needed_key)) { \ 2214 if ((COND) && !(SKB)->tstamp) \ 2215 __net_timestamp(SKB); \ 2216 } \ 2217 2218 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb) 2219 { 2220 return __is_skb_forwardable(dev, skb, true); 2221 } 2222 EXPORT_SYMBOL_GPL(is_skb_forwardable); 2223 2224 static int __dev_forward_skb2(struct net_device *dev, struct sk_buff *skb, 2225 bool check_mtu) 2226 { 2227 int ret = ____dev_forward_skb(dev, skb, check_mtu); 2228 2229 if (likely(!ret)) { 2230 skb->protocol = eth_type_trans(skb, dev); 2231 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); 2232 } 2233 2234 return ret; 2235 } 2236 2237 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2238 { 2239 return __dev_forward_skb2(dev, skb, true); 2240 } 2241 EXPORT_SYMBOL_GPL(__dev_forward_skb); 2242 2243 /** 2244 * dev_forward_skb - loopback an skb to another netif 2245 * 2246 * @dev: destination network device 2247 * @skb: buffer to forward 2248 * 2249 * return values: 2250 * NET_RX_SUCCESS (no congestion) 2251 * NET_RX_DROP (packet was dropped, but freed) 2252 * 2253 * dev_forward_skb can be used for injecting an skb from the 2254 * start_xmit function of one device into the receive queue 2255 * of another device. 2256 * 2257 * The receiving device may be in another namespace, so 2258 * we have to clear all information in the skb that could 2259 * impact namespace isolation. 2260 */ 2261 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2262 { 2263 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); 2264 } 2265 EXPORT_SYMBOL_GPL(dev_forward_skb); 2266 2267 int dev_forward_skb_nomtu(struct net_device *dev, struct sk_buff *skb) 2268 { 2269 return __dev_forward_skb2(dev, skb, false) ?: netif_rx_internal(skb); 2270 } 2271 2272 static inline int deliver_skb(struct sk_buff *skb, 2273 struct packet_type *pt_prev, 2274 struct net_device *orig_dev) 2275 { 2276 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 2277 return -ENOMEM; 2278 refcount_inc(&skb->users); 2279 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 2280 } 2281 2282 static inline void deliver_ptype_list_skb(struct sk_buff *skb, 2283 struct packet_type **pt, 2284 struct net_device *orig_dev, 2285 __be16 type, 2286 struct list_head *ptype_list) 2287 { 2288 struct packet_type *ptype, *pt_prev = *pt; 2289 2290 list_for_each_entry_rcu(ptype, ptype_list, list) { 2291 if (ptype->type != type) 2292 continue; 2293 if (pt_prev) 2294 deliver_skb(skb, pt_prev, orig_dev); 2295 pt_prev = ptype; 2296 } 2297 *pt = pt_prev; 2298 } 2299 2300 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) 2301 { 2302 if (!ptype->af_packet_priv || !skb->sk) 2303 return false; 2304 2305 if (ptype->id_match) 2306 return ptype->id_match(ptype, skb->sk); 2307 else if ((struct sock *)ptype->af_packet_priv == skb->sk) 2308 return true; 2309 2310 return false; 2311 } 2312 2313 /** 2314 * dev_nit_active - return true if any network interface taps are in use 2315 * 2316 * @dev: network device to check for the presence of taps 2317 */ 2318 bool dev_nit_active(struct net_device *dev) 2319 { 2320 return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all); 2321 } 2322 EXPORT_SYMBOL_GPL(dev_nit_active); 2323 2324 /* 2325 * Support routine. Sends outgoing frames to any network 2326 * taps currently in use. 2327 */ 2328 2329 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) 2330 { 2331 struct packet_type *ptype; 2332 struct sk_buff *skb2 = NULL; 2333 struct packet_type *pt_prev = NULL; 2334 struct list_head *ptype_list = &ptype_all; 2335 2336 rcu_read_lock(); 2337 again: 2338 list_for_each_entry_rcu(ptype, ptype_list, list) { 2339 if (ptype->ignore_outgoing) 2340 continue; 2341 2342 /* Never send packets back to the socket 2343 * they originated from - MvS (miquels@drinkel.ow.org) 2344 */ 2345 if (skb_loop_sk(ptype, skb)) 2346 continue; 2347 2348 if (pt_prev) { 2349 deliver_skb(skb2, pt_prev, skb->dev); 2350 pt_prev = ptype; 2351 continue; 2352 } 2353 2354 /* need to clone skb, done only once */ 2355 skb2 = skb_clone(skb, GFP_ATOMIC); 2356 if (!skb2) 2357 goto out_unlock; 2358 2359 net_timestamp_set(skb2); 2360 2361 /* skb->nh should be correctly 2362 * set by sender, so that the second statement is 2363 * just protection against buggy protocols. 2364 */ 2365 skb_reset_mac_header(skb2); 2366 2367 if (skb_network_header(skb2) < skb2->data || 2368 skb_network_header(skb2) > skb_tail_pointer(skb2)) { 2369 net_crit_ratelimited("protocol %04x is buggy, dev %s\n", 2370 ntohs(skb2->protocol), 2371 dev->name); 2372 skb_reset_network_header(skb2); 2373 } 2374 2375 skb2->transport_header = skb2->network_header; 2376 skb2->pkt_type = PACKET_OUTGOING; 2377 pt_prev = ptype; 2378 } 2379 2380 if (ptype_list == &ptype_all) { 2381 ptype_list = &dev->ptype_all; 2382 goto again; 2383 } 2384 out_unlock: 2385 if (pt_prev) { 2386 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC)) 2387 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); 2388 else 2389 kfree_skb(skb2); 2390 } 2391 rcu_read_unlock(); 2392 } 2393 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit); 2394 2395 /** 2396 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change 2397 * @dev: Network device 2398 * @txq: number of queues available 2399 * 2400 * If real_num_tx_queues is changed the tc mappings may no longer be 2401 * valid. To resolve this verify the tc mapping remains valid and if 2402 * not NULL the mapping. With no priorities mapping to this 2403 * offset/count pair it will no longer be used. In the worst case TC0 2404 * is invalid nothing can be done so disable priority mappings. If is 2405 * expected that drivers will fix this mapping if they can before 2406 * calling netif_set_real_num_tx_queues. 2407 */ 2408 static void netif_setup_tc(struct net_device *dev, unsigned int txq) 2409 { 2410 int i; 2411 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2412 2413 /* If TC0 is invalidated disable TC mapping */ 2414 if (tc->offset + tc->count > txq) { 2415 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); 2416 dev->num_tc = 0; 2417 return; 2418 } 2419 2420 /* Invalidated prio to tc mappings set to TC0 */ 2421 for (i = 1; i < TC_BITMASK + 1; i++) { 2422 int q = netdev_get_prio_tc_map(dev, i); 2423 2424 tc = &dev->tc_to_txq[q]; 2425 if (tc->offset + tc->count > txq) { 2426 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n", 2427 i, q); 2428 netdev_set_prio_tc_map(dev, i, 0); 2429 } 2430 } 2431 } 2432 2433 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq) 2434 { 2435 if (dev->num_tc) { 2436 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2437 int i; 2438 2439 /* walk through the TCs and see if it falls into any of them */ 2440 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) { 2441 if ((txq - tc->offset) < tc->count) 2442 return i; 2443 } 2444 2445 /* didn't find it, just return -1 to indicate no match */ 2446 return -1; 2447 } 2448 2449 return 0; 2450 } 2451 EXPORT_SYMBOL(netdev_txq_to_tc); 2452 2453 #ifdef CONFIG_XPS 2454 static struct static_key xps_needed __read_mostly; 2455 static struct static_key xps_rxqs_needed __read_mostly; 2456 static DEFINE_MUTEX(xps_map_mutex); 2457 #define xmap_dereference(P) \ 2458 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) 2459 2460 static bool remove_xps_queue(struct xps_dev_maps *dev_maps, 2461 struct xps_dev_maps *old_maps, int tci, u16 index) 2462 { 2463 struct xps_map *map = NULL; 2464 int pos; 2465 2466 if (dev_maps) 2467 map = xmap_dereference(dev_maps->attr_map[tci]); 2468 if (!map) 2469 return false; 2470 2471 for (pos = map->len; pos--;) { 2472 if (map->queues[pos] != index) 2473 continue; 2474 2475 if (map->len > 1) { 2476 map->queues[pos] = map->queues[--map->len]; 2477 break; 2478 } 2479 2480 if (old_maps) 2481 RCU_INIT_POINTER(old_maps->attr_map[tci], NULL); 2482 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2483 kfree_rcu(map, rcu); 2484 return false; 2485 } 2486 2487 return true; 2488 } 2489 2490 static bool remove_xps_queue_cpu(struct net_device *dev, 2491 struct xps_dev_maps *dev_maps, 2492 int cpu, u16 offset, u16 count) 2493 { 2494 int num_tc = dev_maps->num_tc; 2495 bool active = false; 2496 int tci; 2497 2498 for (tci = cpu * num_tc; num_tc--; tci++) { 2499 int i, j; 2500 2501 for (i = count, j = offset; i--; j++) { 2502 if (!remove_xps_queue(dev_maps, NULL, tci, j)) 2503 break; 2504 } 2505 2506 active |= i < 0; 2507 } 2508 2509 return active; 2510 } 2511 2512 static void reset_xps_maps(struct net_device *dev, 2513 struct xps_dev_maps *dev_maps, 2514 enum xps_map_type type) 2515 { 2516 static_key_slow_dec_cpuslocked(&xps_needed); 2517 if (type == XPS_RXQS) 2518 static_key_slow_dec_cpuslocked(&xps_rxqs_needed); 2519 2520 RCU_INIT_POINTER(dev->xps_maps[type], NULL); 2521 2522 kfree_rcu(dev_maps, rcu); 2523 } 2524 2525 static void clean_xps_maps(struct net_device *dev, enum xps_map_type type, 2526 u16 offset, u16 count) 2527 { 2528 struct xps_dev_maps *dev_maps; 2529 bool active = false; 2530 int i, j; 2531 2532 dev_maps = xmap_dereference(dev->xps_maps[type]); 2533 if (!dev_maps) 2534 return; 2535 2536 for (j = 0; j < dev_maps->nr_ids; j++) 2537 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, count); 2538 if (!active) 2539 reset_xps_maps(dev, dev_maps, type); 2540 2541 if (type == XPS_CPUS) { 2542 for (i = offset + (count - 1); count--; i--) 2543 netdev_queue_numa_node_write( 2544 netdev_get_tx_queue(dev, i), NUMA_NO_NODE); 2545 } 2546 } 2547 2548 static void netif_reset_xps_queues(struct net_device *dev, u16 offset, 2549 u16 count) 2550 { 2551 if (!static_key_false(&xps_needed)) 2552 return; 2553 2554 cpus_read_lock(); 2555 mutex_lock(&xps_map_mutex); 2556 2557 if (static_key_false(&xps_rxqs_needed)) 2558 clean_xps_maps(dev, XPS_RXQS, offset, count); 2559 2560 clean_xps_maps(dev, XPS_CPUS, offset, count); 2561 2562 mutex_unlock(&xps_map_mutex); 2563 cpus_read_unlock(); 2564 } 2565 2566 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) 2567 { 2568 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index); 2569 } 2570 2571 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index, 2572 u16 index, bool is_rxqs_map) 2573 { 2574 struct xps_map *new_map; 2575 int alloc_len = XPS_MIN_MAP_ALLOC; 2576 int i, pos; 2577 2578 for (pos = 0; map && pos < map->len; pos++) { 2579 if (map->queues[pos] != index) 2580 continue; 2581 return map; 2582 } 2583 2584 /* Need to add tx-queue to this CPU's/rx-queue's existing map */ 2585 if (map) { 2586 if (pos < map->alloc_len) 2587 return map; 2588 2589 alloc_len = map->alloc_len * 2; 2590 } 2591 2592 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's 2593 * map 2594 */ 2595 if (is_rxqs_map) 2596 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL); 2597 else 2598 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, 2599 cpu_to_node(attr_index)); 2600 if (!new_map) 2601 return NULL; 2602 2603 for (i = 0; i < pos; i++) 2604 new_map->queues[i] = map->queues[i]; 2605 new_map->alloc_len = alloc_len; 2606 new_map->len = pos; 2607 2608 return new_map; 2609 } 2610 2611 /* Copy xps maps at a given index */ 2612 static void xps_copy_dev_maps(struct xps_dev_maps *dev_maps, 2613 struct xps_dev_maps *new_dev_maps, int index, 2614 int tc, bool skip_tc) 2615 { 2616 int i, tci = index * dev_maps->num_tc; 2617 struct xps_map *map; 2618 2619 /* copy maps belonging to foreign traffic classes */ 2620 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2621 if (i == tc && skip_tc) 2622 continue; 2623 2624 /* fill in the new device map from the old device map */ 2625 map = xmap_dereference(dev_maps->attr_map[tci]); 2626 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2627 } 2628 } 2629 2630 /* Must be called under cpus_read_lock */ 2631 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask, 2632 u16 index, enum xps_map_type type) 2633 { 2634 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL, *old_dev_maps = NULL; 2635 const unsigned long *online_mask = NULL; 2636 bool active = false, copy = false; 2637 int i, j, tci, numa_node_id = -2; 2638 int maps_sz, num_tc = 1, tc = 0; 2639 struct xps_map *map, *new_map; 2640 unsigned int nr_ids; 2641 2642 if (dev->num_tc) { 2643 /* Do not allow XPS on subordinate device directly */ 2644 num_tc = dev->num_tc; 2645 if (num_tc < 0) 2646 return -EINVAL; 2647 2648 /* If queue belongs to subordinate dev use its map */ 2649 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev; 2650 2651 tc = netdev_txq_to_tc(dev, index); 2652 if (tc < 0) 2653 return -EINVAL; 2654 } 2655 2656 mutex_lock(&xps_map_mutex); 2657 2658 dev_maps = xmap_dereference(dev->xps_maps[type]); 2659 if (type == XPS_RXQS) { 2660 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues); 2661 nr_ids = dev->num_rx_queues; 2662 } else { 2663 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc); 2664 if (num_possible_cpus() > 1) 2665 online_mask = cpumask_bits(cpu_online_mask); 2666 nr_ids = nr_cpu_ids; 2667 } 2668 2669 if (maps_sz < L1_CACHE_BYTES) 2670 maps_sz = L1_CACHE_BYTES; 2671 2672 /* The old dev_maps could be larger or smaller than the one we're 2673 * setting up now, as dev->num_tc or nr_ids could have been updated in 2674 * between. We could try to be smart, but let's be safe instead and only 2675 * copy foreign traffic classes if the two map sizes match. 2676 */ 2677 if (dev_maps && 2678 dev_maps->num_tc == num_tc && dev_maps->nr_ids == nr_ids) 2679 copy = true; 2680 2681 /* allocate memory for queue storage */ 2682 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids), 2683 j < nr_ids;) { 2684 if (!new_dev_maps) { 2685 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); 2686 if (!new_dev_maps) { 2687 mutex_unlock(&xps_map_mutex); 2688 return -ENOMEM; 2689 } 2690 2691 new_dev_maps->nr_ids = nr_ids; 2692 new_dev_maps->num_tc = num_tc; 2693 } 2694 2695 tci = j * num_tc + tc; 2696 map = copy ? xmap_dereference(dev_maps->attr_map[tci]) : NULL; 2697 2698 map = expand_xps_map(map, j, index, type == XPS_RXQS); 2699 if (!map) 2700 goto error; 2701 2702 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2703 } 2704 2705 if (!new_dev_maps) 2706 goto out_no_new_maps; 2707 2708 if (!dev_maps) { 2709 /* Increment static keys at most once per type */ 2710 static_key_slow_inc_cpuslocked(&xps_needed); 2711 if (type == XPS_RXQS) 2712 static_key_slow_inc_cpuslocked(&xps_rxqs_needed); 2713 } 2714 2715 for (j = 0; j < nr_ids; j++) { 2716 bool skip_tc = false; 2717 2718 tci = j * num_tc + tc; 2719 if (netif_attr_test_mask(j, mask, nr_ids) && 2720 netif_attr_test_online(j, online_mask, nr_ids)) { 2721 /* add tx-queue to CPU/rx-queue maps */ 2722 int pos = 0; 2723 2724 skip_tc = true; 2725 2726 map = xmap_dereference(new_dev_maps->attr_map[tci]); 2727 while ((pos < map->len) && (map->queues[pos] != index)) 2728 pos++; 2729 2730 if (pos == map->len) 2731 map->queues[map->len++] = index; 2732 #ifdef CONFIG_NUMA 2733 if (type == XPS_CPUS) { 2734 if (numa_node_id == -2) 2735 numa_node_id = cpu_to_node(j); 2736 else if (numa_node_id != cpu_to_node(j)) 2737 numa_node_id = -1; 2738 } 2739 #endif 2740 } 2741 2742 if (copy) 2743 xps_copy_dev_maps(dev_maps, new_dev_maps, j, tc, 2744 skip_tc); 2745 } 2746 2747 rcu_assign_pointer(dev->xps_maps[type], new_dev_maps); 2748 2749 /* Cleanup old maps */ 2750 if (!dev_maps) 2751 goto out_no_old_maps; 2752 2753 for (j = 0; j < dev_maps->nr_ids; j++) { 2754 for (i = num_tc, tci = j * dev_maps->num_tc; i--; tci++) { 2755 map = xmap_dereference(dev_maps->attr_map[tci]); 2756 if (!map) 2757 continue; 2758 2759 if (copy) { 2760 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2761 if (map == new_map) 2762 continue; 2763 } 2764 2765 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2766 kfree_rcu(map, rcu); 2767 } 2768 } 2769 2770 old_dev_maps = dev_maps; 2771 2772 out_no_old_maps: 2773 dev_maps = new_dev_maps; 2774 active = true; 2775 2776 out_no_new_maps: 2777 if (type == XPS_CPUS) 2778 /* update Tx queue numa node */ 2779 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), 2780 (numa_node_id >= 0) ? 2781 numa_node_id : NUMA_NO_NODE); 2782 2783 if (!dev_maps) 2784 goto out_no_maps; 2785 2786 /* removes tx-queue from unused CPUs/rx-queues */ 2787 for (j = 0; j < dev_maps->nr_ids; j++) { 2788 tci = j * dev_maps->num_tc; 2789 2790 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2791 if (i == tc && 2792 netif_attr_test_mask(j, mask, dev_maps->nr_ids) && 2793 netif_attr_test_online(j, online_mask, dev_maps->nr_ids)) 2794 continue; 2795 2796 active |= remove_xps_queue(dev_maps, 2797 copy ? old_dev_maps : NULL, 2798 tci, index); 2799 } 2800 } 2801 2802 if (old_dev_maps) 2803 kfree_rcu(old_dev_maps, rcu); 2804 2805 /* free map if not active */ 2806 if (!active) 2807 reset_xps_maps(dev, dev_maps, type); 2808 2809 out_no_maps: 2810 mutex_unlock(&xps_map_mutex); 2811 2812 return 0; 2813 error: 2814 /* remove any maps that we added */ 2815 for (j = 0; j < nr_ids; j++) { 2816 for (i = num_tc, tci = j * num_tc; i--; tci++) { 2817 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2818 map = copy ? 2819 xmap_dereference(dev_maps->attr_map[tci]) : 2820 NULL; 2821 if (new_map && new_map != map) 2822 kfree(new_map); 2823 } 2824 } 2825 2826 mutex_unlock(&xps_map_mutex); 2827 2828 kfree(new_dev_maps); 2829 return -ENOMEM; 2830 } 2831 EXPORT_SYMBOL_GPL(__netif_set_xps_queue); 2832 2833 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, 2834 u16 index) 2835 { 2836 int ret; 2837 2838 cpus_read_lock(); 2839 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, XPS_CPUS); 2840 cpus_read_unlock(); 2841 2842 return ret; 2843 } 2844 EXPORT_SYMBOL(netif_set_xps_queue); 2845 2846 #endif 2847 static void netdev_unbind_all_sb_channels(struct net_device *dev) 2848 { 2849 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2850 2851 /* Unbind any subordinate channels */ 2852 while (txq-- != &dev->_tx[0]) { 2853 if (txq->sb_dev) 2854 netdev_unbind_sb_channel(dev, txq->sb_dev); 2855 } 2856 } 2857 2858 void netdev_reset_tc(struct net_device *dev) 2859 { 2860 #ifdef CONFIG_XPS 2861 netif_reset_xps_queues_gt(dev, 0); 2862 #endif 2863 netdev_unbind_all_sb_channels(dev); 2864 2865 /* Reset TC configuration of device */ 2866 dev->num_tc = 0; 2867 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq)); 2868 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map)); 2869 } 2870 EXPORT_SYMBOL(netdev_reset_tc); 2871 2872 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset) 2873 { 2874 if (tc >= dev->num_tc) 2875 return -EINVAL; 2876 2877 #ifdef CONFIG_XPS 2878 netif_reset_xps_queues(dev, offset, count); 2879 #endif 2880 dev->tc_to_txq[tc].count = count; 2881 dev->tc_to_txq[tc].offset = offset; 2882 return 0; 2883 } 2884 EXPORT_SYMBOL(netdev_set_tc_queue); 2885 2886 int netdev_set_num_tc(struct net_device *dev, u8 num_tc) 2887 { 2888 if (num_tc > TC_MAX_QUEUE) 2889 return -EINVAL; 2890 2891 #ifdef CONFIG_XPS 2892 netif_reset_xps_queues_gt(dev, 0); 2893 #endif 2894 netdev_unbind_all_sb_channels(dev); 2895 2896 dev->num_tc = num_tc; 2897 return 0; 2898 } 2899 EXPORT_SYMBOL(netdev_set_num_tc); 2900 2901 void netdev_unbind_sb_channel(struct net_device *dev, 2902 struct net_device *sb_dev) 2903 { 2904 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2905 2906 #ifdef CONFIG_XPS 2907 netif_reset_xps_queues_gt(sb_dev, 0); 2908 #endif 2909 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq)); 2910 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map)); 2911 2912 while (txq-- != &dev->_tx[0]) { 2913 if (txq->sb_dev == sb_dev) 2914 txq->sb_dev = NULL; 2915 } 2916 } 2917 EXPORT_SYMBOL(netdev_unbind_sb_channel); 2918 2919 int netdev_bind_sb_channel_queue(struct net_device *dev, 2920 struct net_device *sb_dev, 2921 u8 tc, u16 count, u16 offset) 2922 { 2923 /* Make certain the sb_dev and dev are already configured */ 2924 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc) 2925 return -EINVAL; 2926 2927 /* We cannot hand out queues we don't have */ 2928 if ((offset + count) > dev->real_num_tx_queues) 2929 return -EINVAL; 2930 2931 /* Record the mapping */ 2932 sb_dev->tc_to_txq[tc].count = count; 2933 sb_dev->tc_to_txq[tc].offset = offset; 2934 2935 /* Provide a way for Tx queue to find the tc_to_txq map or 2936 * XPS map for itself. 2937 */ 2938 while (count--) 2939 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev; 2940 2941 return 0; 2942 } 2943 EXPORT_SYMBOL(netdev_bind_sb_channel_queue); 2944 2945 int netdev_set_sb_channel(struct net_device *dev, u16 channel) 2946 { 2947 /* Do not use a multiqueue device to represent a subordinate channel */ 2948 if (netif_is_multiqueue(dev)) 2949 return -ENODEV; 2950 2951 /* We allow channels 1 - 32767 to be used for subordinate channels. 2952 * Channel 0 is meant to be "native" mode and used only to represent 2953 * the main root device. We allow writing 0 to reset the device back 2954 * to normal mode after being used as a subordinate channel. 2955 */ 2956 if (channel > S16_MAX) 2957 return -EINVAL; 2958 2959 dev->num_tc = -channel; 2960 2961 return 0; 2962 } 2963 EXPORT_SYMBOL(netdev_set_sb_channel); 2964 2965 /* 2966 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues 2967 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed. 2968 */ 2969 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) 2970 { 2971 bool disabling; 2972 int rc; 2973 2974 disabling = txq < dev->real_num_tx_queues; 2975 2976 if (txq < 1 || txq > dev->num_tx_queues) 2977 return -EINVAL; 2978 2979 if (dev->reg_state == NETREG_REGISTERED || 2980 dev->reg_state == NETREG_UNREGISTERING) { 2981 ASSERT_RTNL(); 2982 2983 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, 2984 txq); 2985 if (rc) 2986 return rc; 2987 2988 if (dev->num_tc) 2989 netif_setup_tc(dev, txq); 2990 2991 dev->real_num_tx_queues = txq; 2992 2993 if (disabling) { 2994 synchronize_net(); 2995 qdisc_reset_all_tx_gt(dev, txq); 2996 #ifdef CONFIG_XPS 2997 netif_reset_xps_queues_gt(dev, txq); 2998 #endif 2999 } 3000 } else { 3001 dev->real_num_tx_queues = txq; 3002 } 3003 3004 return 0; 3005 } 3006 EXPORT_SYMBOL(netif_set_real_num_tx_queues); 3007 3008 #ifdef CONFIG_SYSFS 3009 /** 3010 * netif_set_real_num_rx_queues - set actual number of RX queues used 3011 * @dev: Network device 3012 * @rxq: Actual number of RX queues 3013 * 3014 * This must be called either with the rtnl_lock held or before 3015 * registration of the net device. Returns 0 on success, or a 3016 * negative error code. If called before registration, it always 3017 * succeeds. 3018 */ 3019 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) 3020 { 3021 int rc; 3022 3023 if (rxq < 1 || rxq > dev->num_rx_queues) 3024 return -EINVAL; 3025 3026 if (dev->reg_state == NETREG_REGISTERED) { 3027 ASSERT_RTNL(); 3028 3029 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, 3030 rxq); 3031 if (rc) 3032 return rc; 3033 } 3034 3035 dev->real_num_rx_queues = rxq; 3036 return 0; 3037 } 3038 EXPORT_SYMBOL(netif_set_real_num_rx_queues); 3039 #endif 3040 3041 /** 3042 * netif_get_num_default_rss_queues - default number of RSS queues 3043 * 3044 * This routine should set an upper limit on the number of RSS queues 3045 * used by default by multiqueue devices. 3046 */ 3047 int netif_get_num_default_rss_queues(void) 3048 { 3049 return is_kdump_kernel() ? 3050 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus()); 3051 } 3052 EXPORT_SYMBOL(netif_get_num_default_rss_queues); 3053 3054 static void __netif_reschedule(struct Qdisc *q) 3055 { 3056 struct softnet_data *sd; 3057 unsigned long flags; 3058 3059 local_irq_save(flags); 3060 sd = this_cpu_ptr(&softnet_data); 3061 q->next_sched = NULL; 3062 *sd->output_queue_tailp = q; 3063 sd->output_queue_tailp = &q->next_sched; 3064 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3065 local_irq_restore(flags); 3066 } 3067 3068 void __netif_schedule(struct Qdisc *q) 3069 { 3070 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) 3071 __netif_reschedule(q); 3072 } 3073 EXPORT_SYMBOL(__netif_schedule); 3074 3075 struct dev_kfree_skb_cb { 3076 enum skb_free_reason reason; 3077 }; 3078 3079 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) 3080 { 3081 return (struct dev_kfree_skb_cb *)skb->cb; 3082 } 3083 3084 void netif_schedule_queue(struct netdev_queue *txq) 3085 { 3086 rcu_read_lock(); 3087 if (!netif_xmit_stopped(txq)) { 3088 struct Qdisc *q = rcu_dereference(txq->qdisc); 3089 3090 __netif_schedule(q); 3091 } 3092 rcu_read_unlock(); 3093 } 3094 EXPORT_SYMBOL(netif_schedule_queue); 3095 3096 void netif_tx_wake_queue(struct netdev_queue *dev_queue) 3097 { 3098 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { 3099 struct Qdisc *q; 3100 3101 rcu_read_lock(); 3102 q = rcu_dereference(dev_queue->qdisc); 3103 __netif_schedule(q); 3104 rcu_read_unlock(); 3105 } 3106 } 3107 EXPORT_SYMBOL(netif_tx_wake_queue); 3108 3109 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason) 3110 { 3111 unsigned long flags; 3112 3113 if (unlikely(!skb)) 3114 return; 3115 3116 if (likely(refcount_read(&skb->users) == 1)) { 3117 smp_rmb(); 3118 refcount_set(&skb->users, 0); 3119 } else if (likely(!refcount_dec_and_test(&skb->users))) { 3120 return; 3121 } 3122 get_kfree_skb_cb(skb)->reason = reason; 3123 local_irq_save(flags); 3124 skb->next = __this_cpu_read(softnet_data.completion_queue); 3125 __this_cpu_write(softnet_data.completion_queue, skb); 3126 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3127 local_irq_restore(flags); 3128 } 3129 EXPORT_SYMBOL(__dev_kfree_skb_irq); 3130 3131 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason) 3132 { 3133 if (in_irq() || irqs_disabled()) 3134 __dev_kfree_skb_irq(skb, reason); 3135 else 3136 dev_kfree_skb(skb); 3137 } 3138 EXPORT_SYMBOL(__dev_kfree_skb_any); 3139 3140 3141 /** 3142 * netif_device_detach - mark device as removed 3143 * @dev: network device 3144 * 3145 * Mark device as removed from system and therefore no longer available. 3146 */ 3147 void netif_device_detach(struct net_device *dev) 3148 { 3149 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && 3150 netif_running(dev)) { 3151 netif_tx_stop_all_queues(dev); 3152 } 3153 } 3154 EXPORT_SYMBOL(netif_device_detach); 3155 3156 /** 3157 * netif_device_attach - mark device as attached 3158 * @dev: network device 3159 * 3160 * Mark device as attached from system and restart if needed. 3161 */ 3162 void netif_device_attach(struct net_device *dev) 3163 { 3164 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && 3165 netif_running(dev)) { 3166 netif_tx_wake_all_queues(dev); 3167 __netdev_watchdog_up(dev); 3168 } 3169 } 3170 EXPORT_SYMBOL(netif_device_attach); 3171 3172 /* 3173 * Returns a Tx hash based on the given packet descriptor a Tx queues' number 3174 * to be used as a distribution range. 3175 */ 3176 static u16 skb_tx_hash(const struct net_device *dev, 3177 const struct net_device *sb_dev, 3178 struct sk_buff *skb) 3179 { 3180 u32 hash; 3181 u16 qoffset = 0; 3182 u16 qcount = dev->real_num_tx_queues; 3183 3184 if (dev->num_tc) { 3185 u8 tc = netdev_get_prio_tc_map(dev, skb->priority); 3186 3187 qoffset = sb_dev->tc_to_txq[tc].offset; 3188 qcount = sb_dev->tc_to_txq[tc].count; 3189 } 3190 3191 if (skb_rx_queue_recorded(skb)) { 3192 hash = skb_get_rx_queue(skb); 3193 if (hash >= qoffset) 3194 hash -= qoffset; 3195 while (unlikely(hash >= qcount)) 3196 hash -= qcount; 3197 return hash + qoffset; 3198 } 3199 3200 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; 3201 } 3202 3203 static void skb_warn_bad_offload(const struct sk_buff *skb) 3204 { 3205 static const netdev_features_t null_features; 3206 struct net_device *dev = skb->dev; 3207 const char *name = ""; 3208 3209 if (!net_ratelimit()) 3210 return; 3211 3212 if (dev) { 3213 if (dev->dev.parent) 3214 name = dev_driver_string(dev->dev.parent); 3215 else 3216 name = netdev_name(dev); 3217 } 3218 skb_dump(KERN_WARNING, skb, false); 3219 WARN(1, "%s: caps=(%pNF, %pNF)\n", 3220 name, dev ? &dev->features : &null_features, 3221 skb->sk ? &skb->sk->sk_route_caps : &null_features); 3222 } 3223 3224 /* 3225 * Invalidate hardware checksum when packet is to be mangled, and 3226 * complete checksum manually on outgoing path. 3227 */ 3228 int skb_checksum_help(struct sk_buff *skb) 3229 { 3230 __wsum csum; 3231 int ret = 0, offset; 3232 3233 if (skb->ip_summed == CHECKSUM_COMPLETE) 3234 goto out_set_summed; 3235 3236 if (unlikely(skb_is_gso(skb))) { 3237 skb_warn_bad_offload(skb); 3238 return -EINVAL; 3239 } 3240 3241 /* Before computing a checksum, we should make sure no frag could 3242 * be modified by an external entity : checksum could be wrong. 3243 */ 3244 if (skb_has_shared_frag(skb)) { 3245 ret = __skb_linearize(skb); 3246 if (ret) 3247 goto out; 3248 } 3249 3250 offset = skb_checksum_start_offset(skb); 3251 BUG_ON(offset >= skb_headlen(skb)); 3252 csum = skb_checksum(skb, offset, skb->len - offset, 0); 3253 3254 offset += skb->csum_offset; 3255 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); 3256 3257 ret = skb_ensure_writable(skb, offset + sizeof(__sum16)); 3258 if (ret) 3259 goto out; 3260 3261 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0; 3262 out_set_summed: 3263 skb->ip_summed = CHECKSUM_NONE; 3264 out: 3265 return ret; 3266 } 3267 EXPORT_SYMBOL(skb_checksum_help); 3268 3269 int skb_crc32c_csum_help(struct sk_buff *skb) 3270 { 3271 __le32 crc32c_csum; 3272 int ret = 0, offset, start; 3273 3274 if (skb->ip_summed != CHECKSUM_PARTIAL) 3275 goto out; 3276 3277 if (unlikely(skb_is_gso(skb))) 3278 goto out; 3279 3280 /* Before computing a checksum, we should make sure no frag could 3281 * be modified by an external entity : checksum could be wrong. 3282 */ 3283 if (unlikely(skb_has_shared_frag(skb))) { 3284 ret = __skb_linearize(skb); 3285 if (ret) 3286 goto out; 3287 } 3288 start = skb_checksum_start_offset(skb); 3289 offset = start + offsetof(struct sctphdr, checksum); 3290 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { 3291 ret = -EINVAL; 3292 goto out; 3293 } 3294 3295 ret = skb_ensure_writable(skb, offset + sizeof(__le32)); 3296 if (ret) 3297 goto out; 3298 3299 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start, 3300 skb->len - start, ~(__u32)0, 3301 crc32c_csum_stub)); 3302 *(__le32 *)(skb->data + offset) = crc32c_csum; 3303 skb->ip_summed = CHECKSUM_NONE; 3304 skb->csum_not_inet = 0; 3305 out: 3306 return ret; 3307 } 3308 3309 __be16 skb_network_protocol(struct sk_buff *skb, int *depth) 3310 { 3311 __be16 type = skb->protocol; 3312 3313 /* Tunnel gso handlers can set protocol to ethernet. */ 3314 if (type == htons(ETH_P_TEB)) { 3315 struct ethhdr *eth; 3316 3317 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) 3318 return 0; 3319 3320 eth = (struct ethhdr *)skb->data; 3321 type = eth->h_proto; 3322 } 3323 3324 return __vlan_get_protocol(skb, type, depth); 3325 } 3326 3327 /** 3328 * skb_mac_gso_segment - mac layer segmentation handler. 3329 * @skb: buffer to segment 3330 * @features: features for the output path (see dev->features) 3331 */ 3332 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb, 3333 netdev_features_t features) 3334 { 3335 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); 3336 struct packet_offload *ptype; 3337 int vlan_depth = skb->mac_len; 3338 __be16 type = skb_network_protocol(skb, &vlan_depth); 3339 3340 if (unlikely(!type)) 3341 return ERR_PTR(-EINVAL); 3342 3343 __skb_pull(skb, vlan_depth); 3344 3345 rcu_read_lock(); 3346 list_for_each_entry_rcu(ptype, &offload_base, list) { 3347 if (ptype->type == type && ptype->callbacks.gso_segment) { 3348 segs = ptype->callbacks.gso_segment(skb, features); 3349 break; 3350 } 3351 } 3352 rcu_read_unlock(); 3353 3354 __skb_push(skb, skb->data - skb_mac_header(skb)); 3355 3356 return segs; 3357 } 3358 EXPORT_SYMBOL(skb_mac_gso_segment); 3359 3360 3361 /* openvswitch calls this on rx path, so we need a different check. 3362 */ 3363 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path) 3364 { 3365 if (tx_path) 3366 return skb->ip_summed != CHECKSUM_PARTIAL && 3367 skb->ip_summed != CHECKSUM_UNNECESSARY; 3368 3369 return skb->ip_summed == CHECKSUM_NONE; 3370 } 3371 3372 /** 3373 * __skb_gso_segment - Perform segmentation on skb. 3374 * @skb: buffer to segment 3375 * @features: features for the output path (see dev->features) 3376 * @tx_path: whether it is called in TX path 3377 * 3378 * This function segments the given skb and returns a list of segments. 3379 * 3380 * It may return NULL if the skb requires no segmentation. This is 3381 * only possible when GSO is used for verifying header integrity. 3382 * 3383 * Segmentation preserves SKB_GSO_CB_OFFSET bytes of previous skb cb. 3384 */ 3385 struct sk_buff *__skb_gso_segment(struct sk_buff *skb, 3386 netdev_features_t features, bool tx_path) 3387 { 3388 struct sk_buff *segs; 3389 3390 if (unlikely(skb_needs_check(skb, tx_path))) { 3391 int err; 3392 3393 /* We're going to init ->check field in TCP or UDP header */ 3394 err = skb_cow_head(skb, 0); 3395 if (err < 0) 3396 return ERR_PTR(err); 3397 } 3398 3399 /* Only report GSO partial support if it will enable us to 3400 * support segmentation on this frame without needing additional 3401 * work. 3402 */ 3403 if (features & NETIF_F_GSO_PARTIAL) { 3404 netdev_features_t partial_features = NETIF_F_GSO_ROBUST; 3405 struct net_device *dev = skb->dev; 3406 3407 partial_features |= dev->features & dev->gso_partial_features; 3408 if (!skb_gso_ok(skb, features | partial_features)) 3409 features &= ~NETIF_F_GSO_PARTIAL; 3410 } 3411 3412 BUILD_BUG_ON(SKB_GSO_CB_OFFSET + 3413 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb)); 3414 3415 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb); 3416 SKB_GSO_CB(skb)->encap_level = 0; 3417 3418 skb_reset_mac_header(skb); 3419 skb_reset_mac_len(skb); 3420 3421 segs = skb_mac_gso_segment(skb, features); 3422 3423 if (segs != skb && unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs))) 3424 skb_warn_bad_offload(skb); 3425 3426 return segs; 3427 } 3428 EXPORT_SYMBOL(__skb_gso_segment); 3429 3430 /* Take action when hardware reception checksum errors are detected. */ 3431 #ifdef CONFIG_BUG 3432 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3433 { 3434 if (net_ratelimit()) { 3435 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>"); 3436 skb_dump(KERN_ERR, skb, true); 3437 dump_stack(); 3438 } 3439 } 3440 EXPORT_SYMBOL(netdev_rx_csum_fault); 3441 #endif 3442 3443 /* XXX: check that highmem exists at all on the given machine. */ 3444 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) 3445 { 3446 #ifdef CONFIG_HIGHMEM 3447 int i; 3448 3449 if (!(dev->features & NETIF_F_HIGHDMA)) { 3450 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 3451 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 3452 3453 if (PageHighMem(skb_frag_page(frag))) 3454 return 1; 3455 } 3456 } 3457 #endif 3458 return 0; 3459 } 3460 3461 /* If MPLS offload request, verify we are testing hardware MPLS features 3462 * instead of standard features for the netdev. 3463 */ 3464 #if IS_ENABLED(CONFIG_NET_MPLS_GSO) 3465 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3466 netdev_features_t features, 3467 __be16 type) 3468 { 3469 if (eth_p_mpls(type)) 3470 features &= skb->dev->mpls_features; 3471 3472 return features; 3473 } 3474 #else 3475 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3476 netdev_features_t features, 3477 __be16 type) 3478 { 3479 return features; 3480 } 3481 #endif 3482 3483 static netdev_features_t harmonize_features(struct sk_buff *skb, 3484 netdev_features_t features) 3485 { 3486 __be16 type; 3487 3488 type = skb_network_protocol(skb, NULL); 3489 features = net_mpls_features(skb, features, type); 3490 3491 if (skb->ip_summed != CHECKSUM_NONE && 3492 !can_checksum_protocol(features, type)) { 3493 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); 3494 } 3495 if (illegal_highdma(skb->dev, skb)) 3496 features &= ~NETIF_F_SG; 3497 3498 return features; 3499 } 3500 3501 netdev_features_t passthru_features_check(struct sk_buff *skb, 3502 struct net_device *dev, 3503 netdev_features_t features) 3504 { 3505 return features; 3506 } 3507 EXPORT_SYMBOL(passthru_features_check); 3508 3509 static netdev_features_t dflt_features_check(struct sk_buff *skb, 3510 struct net_device *dev, 3511 netdev_features_t features) 3512 { 3513 return vlan_features_check(skb, features); 3514 } 3515 3516 static netdev_features_t gso_features_check(const struct sk_buff *skb, 3517 struct net_device *dev, 3518 netdev_features_t features) 3519 { 3520 u16 gso_segs = skb_shinfo(skb)->gso_segs; 3521 3522 if (gso_segs > dev->gso_max_segs) 3523 return features & ~NETIF_F_GSO_MASK; 3524 3525 if (!skb_shinfo(skb)->gso_type) { 3526 skb_warn_bad_offload(skb); 3527 return features & ~NETIF_F_GSO_MASK; 3528 } 3529 3530 /* Support for GSO partial features requires software 3531 * intervention before we can actually process the packets 3532 * so we need to strip support for any partial features now 3533 * and we can pull them back in after we have partially 3534 * segmented the frame. 3535 */ 3536 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)) 3537 features &= ~dev->gso_partial_features; 3538 3539 /* Make sure to clear the IPv4 ID mangling feature if the 3540 * IPv4 header has the potential to be fragmented. 3541 */ 3542 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { 3543 struct iphdr *iph = skb->encapsulation ? 3544 inner_ip_hdr(skb) : ip_hdr(skb); 3545 3546 if (!(iph->frag_off & htons(IP_DF))) 3547 features &= ~NETIF_F_TSO_MANGLEID; 3548 } 3549 3550 return features; 3551 } 3552 3553 netdev_features_t netif_skb_features(struct sk_buff *skb) 3554 { 3555 struct net_device *dev = skb->dev; 3556 netdev_features_t features = dev->features; 3557 3558 if (skb_is_gso(skb)) 3559 features = gso_features_check(skb, dev, features); 3560 3561 /* If encapsulation offload request, verify we are testing 3562 * hardware encapsulation features instead of standard 3563 * features for the netdev 3564 */ 3565 if (skb->encapsulation) 3566 features &= dev->hw_enc_features; 3567 3568 if (skb_vlan_tagged(skb)) 3569 features = netdev_intersect_features(features, 3570 dev->vlan_features | 3571 NETIF_F_HW_VLAN_CTAG_TX | 3572 NETIF_F_HW_VLAN_STAG_TX); 3573 3574 if (dev->netdev_ops->ndo_features_check) 3575 features &= dev->netdev_ops->ndo_features_check(skb, dev, 3576 features); 3577 else 3578 features &= dflt_features_check(skb, dev, features); 3579 3580 return harmonize_features(skb, features); 3581 } 3582 EXPORT_SYMBOL(netif_skb_features); 3583 3584 static int xmit_one(struct sk_buff *skb, struct net_device *dev, 3585 struct netdev_queue *txq, bool more) 3586 { 3587 unsigned int len; 3588 int rc; 3589 3590 if (dev_nit_active(dev)) 3591 dev_queue_xmit_nit(skb, dev); 3592 3593 len = skb->len; 3594 PRANDOM_ADD_NOISE(skb, dev, txq, len + jiffies); 3595 trace_net_dev_start_xmit(skb, dev); 3596 rc = netdev_start_xmit(skb, dev, txq, more); 3597 trace_net_dev_xmit(skb, rc, dev, len); 3598 3599 return rc; 3600 } 3601 3602 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, 3603 struct netdev_queue *txq, int *ret) 3604 { 3605 struct sk_buff *skb = first; 3606 int rc = NETDEV_TX_OK; 3607 3608 while (skb) { 3609 struct sk_buff *next = skb->next; 3610 3611 skb_mark_not_on_list(skb); 3612 rc = xmit_one(skb, dev, txq, next != NULL); 3613 if (unlikely(!dev_xmit_complete(rc))) { 3614 skb->next = next; 3615 goto out; 3616 } 3617 3618 skb = next; 3619 if (netif_tx_queue_stopped(txq) && skb) { 3620 rc = NETDEV_TX_BUSY; 3621 break; 3622 } 3623 } 3624 3625 out: 3626 *ret = rc; 3627 return skb; 3628 } 3629 3630 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, 3631 netdev_features_t features) 3632 { 3633 if (skb_vlan_tag_present(skb) && 3634 !vlan_hw_offload_capable(features, skb->vlan_proto)) 3635 skb = __vlan_hwaccel_push_inside(skb); 3636 return skb; 3637 } 3638 3639 int skb_csum_hwoffload_help(struct sk_buff *skb, 3640 const netdev_features_t features) 3641 { 3642 if (unlikely(skb_csum_is_sctp(skb))) 3643 return !!(features & NETIF_F_SCTP_CRC) ? 0 : 3644 skb_crc32c_csum_help(skb); 3645 3646 if (features & NETIF_F_HW_CSUM) 3647 return 0; 3648 3649 if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) { 3650 switch (skb->csum_offset) { 3651 case offsetof(struct tcphdr, check): 3652 case offsetof(struct udphdr, check): 3653 return 0; 3654 } 3655 } 3656 3657 return skb_checksum_help(skb); 3658 } 3659 EXPORT_SYMBOL(skb_csum_hwoffload_help); 3660 3661 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again) 3662 { 3663 netdev_features_t features; 3664 3665 features = netif_skb_features(skb); 3666 skb = validate_xmit_vlan(skb, features); 3667 if (unlikely(!skb)) 3668 goto out_null; 3669 3670 skb = sk_validate_xmit_skb(skb, dev); 3671 if (unlikely(!skb)) 3672 goto out_null; 3673 3674 if (netif_needs_gso(skb, features)) { 3675 struct sk_buff *segs; 3676 3677 segs = skb_gso_segment(skb, features); 3678 if (IS_ERR(segs)) { 3679 goto out_kfree_skb; 3680 } else if (segs) { 3681 consume_skb(skb); 3682 skb = segs; 3683 } 3684 } else { 3685 if (skb_needs_linearize(skb, features) && 3686 __skb_linearize(skb)) 3687 goto out_kfree_skb; 3688 3689 /* If packet is not checksummed and device does not 3690 * support checksumming for this protocol, complete 3691 * checksumming here. 3692 */ 3693 if (skb->ip_summed == CHECKSUM_PARTIAL) { 3694 if (skb->encapsulation) 3695 skb_set_inner_transport_header(skb, 3696 skb_checksum_start_offset(skb)); 3697 else 3698 skb_set_transport_header(skb, 3699 skb_checksum_start_offset(skb)); 3700 if (skb_csum_hwoffload_help(skb, features)) 3701 goto out_kfree_skb; 3702 } 3703 } 3704 3705 skb = validate_xmit_xfrm(skb, features, again); 3706 3707 return skb; 3708 3709 out_kfree_skb: 3710 kfree_skb(skb); 3711 out_null: 3712 atomic_long_inc(&dev->tx_dropped); 3713 return NULL; 3714 } 3715 3716 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again) 3717 { 3718 struct sk_buff *next, *head = NULL, *tail; 3719 3720 for (; skb != NULL; skb = next) { 3721 next = skb->next; 3722 skb_mark_not_on_list(skb); 3723 3724 /* in case skb wont be segmented, point to itself */ 3725 skb->prev = skb; 3726 3727 skb = validate_xmit_skb(skb, dev, again); 3728 if (!skb) 3729 continue; 3730 3731 if (!head) 3732 head = skb; 3733 else 3734 tail->next = skb; 3735 /* If skb was segmented, skb->prev points to 3736 * the last segment. If not, it still contains skb. 3737 */ 3738 tail = skb->prev; 3739 } 3740 return head; 3741 } 3742 EXPORT_SYMBOL_GPL(validate_xmit_skb_list); 3743 3744 static void qdisc_pkt_len_init(struct sk_buff *skb) 3745 { 3746 const struct skb_shared_info *shinfo = skb_shinfo(skb); 3747 3748 qdisc_skb_cb(skb)->pkt_len = skb->len; 3749 3750 /* To get more precise estimation of bytes sent on wire, 3751 * we add to pkt_len the headers size of all segments 3752 */ 3753 if (shinfo->gso_size && skb_transport_header_was_set(skb)) { 3754 unsigned int hdr_len; 3755 u16 gso_segs = shinfo->gso_segs; 3756 3757 /* mac layer + network layer */ 3758 hdr_len = skb_transport_header(skb) - skb_mac_header(skb); 3759 3760 /* + transport layer */ 3761 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { 3762 const struct tcphdr *th; 3763 struct tcphdr _tcphdr; 3764 3765 th = skb_header_pointer(skb, skb_transport_offset(skb), 3766 sizeof(_tcphdr), &_tcphdr); 3767 if (likely(th)) 3768 hdr_len += __tcp_hdrlen(th); 3769 } else { 3770 struct udphdr _udphdr; 3771 3772 if (skb_header_pointer(skb, skb_transport_offset(skb), 3773 sizeof(_udphdr), &_udphdr)) 3774 hdr_len += sizeof(struct udphdr); 3775 } 3776 3777 if (shinfo->gso_type & SKB_GSO_DODGY) 3778 gso_segs = DIV_ROUND_UP(skb->len - hdr_len, 3779 shinfo->gso_size); 3780 3781 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; 3782 } 3783 } 3784 3785 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, 3786 struct net_device *dev, 3787 struct netdev_queue *txq) 3788 { 3789 spinlock_t *root_lock = qdisc_lock(q); 3790 struct sk_buff *to_free = NULL; 3791 bool contended; 3792 int rc; 3793 3794 qdisc_calculate_pkt_len(skb, q); 3795 3796 if (q->flags & TCQ_F_NOLOCK) { 3797 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK; 3798 qdisc_run(q); 3799 3800 if (unlikely(to_free)) 3801 kfree_skb_list(to_free); 3802 return rc; 3803 } 3804 3805 /* 3806 * Heuristic to force contended enqueues to serialize on a 3807 * separate lock before trying to get qdisc main lock. 3808 * This permits qdisc->running owner to get the lock more 3809 * often and dequeue packets faster. 3810 */ 3811 contended = qdisc_is_running(q); 3812 if (unlikely(contended)) 3813 spin_lock(&q->busylock); 3814 3815 spin_lock(root_lock); 3816 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { 3817 __qdisc_drop(skb, &to_free); 3818 rc = NET_XMIT_DROP; 3819 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && 3820 qdisc_run_begin(q)) { 3821 /* 3822 * This is a work-conserving queue; there are no old skbs 3823 * waiting to be sent out; and the qdisc is not running - 3824 * xmit the skb directly. 3825 */ 3826 3827 qdisc_bstats_update(q, skb); 3828 3829 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { 3830 if (unlikely(contended)) { 3831 spin_unlock(&q->busylock); 3832 contended = false; 3833 } 3834 __qdisc_run(q); 3835 } 3836 3837 qdisc_run_end(q); 3838 rc = NET_XMIT_SUCCESS; 3839 } else { 3840 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK; 3841 if (qdisc_run_begin(q)) { 3842 if (unlikely(contended)) { 3843 spin_unlock(&q->busylock); 3844 contended = false; 3845 } 3846 __qdisc_run(q); 3847 qdisc_run_end(q); 3848 } 3849 } 3850 spin_unlock(root_lock); 3851 if (unlikely(to_free)) 3852 kfree_skb_list(to_free); 3853 if (unlikely(contended)) 3854 spin_unlock(&q->busylock); 3855 return rc; 3856 } 3857 3858 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) 3859 static void skb_update_prio(struct sk_buff *skb) 3860 { 3861 const struct netprio_map *map; 3862 const struct sock *sk; 3863 unsigned int prioidx; 3864 3865 if (skb->priority) 3866 return; 3867 map = rcu_dereference_bh(skb->dev->priomap); 3868 if (!map) 3869 return; 3870 sk = skb_to_full_sk(skb); 3871 if (!sk) 3872 return; 3873 3874 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data); 3875 3876 if (prioidx < map->priomap_len) 3877 skb->priority = map->priomap[prioidx]; 3878 } 3879 #else 3880 #define skb_update_prio(skb) 3881 #endif 3882 3883 /** 3884 * dev_loopback_xmit - loop back @skb 3885 * @net: network namespace this loopback is happening in 3886 * @sk: sk needed to be a netfilter okfn 3887 * @skb: buffer to transmit 3888 */ 3889 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) 3890 { 3891 skb_reset_mac_header(skb); 3892 __skb_pull(skb, skb_network_offset(skb)); 3893 skb->pkt_type = PACKET_LOOPBACK; 3894 skb->ip_summed = CHECKSUM_UNNECESSARY; 3895 WARN_ON(!skb_dst(skb)); 3896 skb_dst_force(skb); 3897 netif_rx_ni(skb); 3898 return 0; 3899 } 3900 EXPORT_SYMBOL(dev_loopback_xmit); 3901 3902 #ifdef CONFIG_NET_EGRESS 3903 static struct sk_buff * 3904 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 3905 { 3906 struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress); 3907 struct tcf_result cl_res; 3908 3909 if (!miniq) 3910 return skb; 3911 3912 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */ 3913 qdisc_skb_cb(skb)->mru = 0; 3914 qdisc_skb_cb(skb)->post_ct = false; 3915 mini_qdisc_bstats_cpu_update(miniq, skb); 3916 3917 switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) { 3918 case TC_ACT_OK: 3919 case TC_ACT_RECLASSIFY: 3920 skb->tc_index = TC_H_MIN(cl_res.classid); 3921 break; 3922 case TC_ACT_SHOT: 3923 mini_qdisc_qstats_cpu_drop(miniq); 3924 *ret = NET_XMIT_DROP; 3925 kfree_skb(skb); 3926 return NULL; 3927 case TC_ACT_STOLEN: 3928 case TC_ACT_QUEUED: 3929 case TC_ACT_TRAP: 3930 *ret = NET_XMIT_SUCCESS; 3931 consume_skb(skb); 3932 return NULL; 3933 case TC_ACT_REDIRECT: 3934 /* No need to push/pop skb's mac_header here on egress! */ 3935 skb_do_redirect(skb); 3936 *ret = NET_XMIT_SUCCESS; 3937 return NULL; 3938 default: 3939 break; 3940 } 3941 3942 return skb; 3943 } 3944 #endif /* CONFIG_NET_EGRESS */ 3945 3946 #ifdef CONFIG_XPS 3947 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb, 3948 struct xps_dev_maps *dev_maps, unsigned int tci) 3949 { 3950 int tc = netdev_get_prio_tc_map(dev, skb->priority); 3951 struct xps_map *map; 3952 int queue_index = -1; 3953 3954 if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids) 3955 return queue_index; 3956 3957 tci *= dev_maps->num_tc; 3958 tci += tc; 3959 3960 map = rcu_dereference(dev_maps->attr_map[tci]); 3961 if (map) { 3962 if (map->len == 1) 3963 queue_index = map->queues[0]; 3964 else 3965 queue_index = map->queues[reciprocal_scale( 3966 skb_get_hash(skb), map->len)]; 3967 if (unlikely(queue_index >= dev->real_num_tx_queues)) 3968 queue_index = -1; 3969 } 3970 return queue_index; 3971 } 3972 #endif 3973 3974 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev, 3975 struct sk_buff *skb) 3976 { 3977 #ifdef CONFIG_XPS 3978 struct xps_dev_maps *dev_maps; 3979 struct sock *sk = skb->sk; 3980 int queue_index = -1; 3981 3982 if (!static_key_false(&xps_needed)) 3983 return -1; 3984 3985 rcu_read_lock(); 3986 if (!static_key_false(&xps_rxqs_needed)) 3987 goto get_cpus_map; 3988 3989 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]); 3990 if (dev_maps) { 3991 int tci = sk_rx_queue_get(sk); 3992 3993 if (tci >= 0) 3994 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 3995 tci); 3996 } 3997 3998 get_cpus_map: 3999 if (queue_index < 0) { 4000 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]); 4001 if (dev_maps) { 4002 unsigned int tci = skb->sender_cpu - 1; 4003 4004 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4005 tci); 4006 } 4007 } 4008 rcu_read_unlock(); 4009 4010 return queue_index; 4011 #else 4012 return -1; 4013 #endif 4014 } 4015 4016 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb, 4017 struct net_device *sb_dev) 4018 { 4019 return 0; 4020 } 4021 EXPORT_SYMBOL(dev_pick_tx_zero); 4022 4023 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb, 4024 struct net_device *sb_dev) 4025 { 4026 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues; 4027 } 4028 EXPORT_SYMBOL(dev_pick_tx_cpu_id); 4029 4030 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb, 4031 struct net_device *sb_dev) 4032 { 4033 struct sock *sk = skb->sk; 4034 int queue_index = sk_tx_queue_get(sk); 4035 4036 sb_dev = sb_dev ? : dev; 4037 4038 if (queue_index < 0 || skb->ooo_okay || 4039 queue_index >= dev->real_num_tx_queues) { 4040 int new_index = get_xps_queue(dev, sb_dev, skb); 4041 4042 if (new_index < 0) 4043 new_index = skb_tx_hash(dev, sb_dev, skb); 4044 4045 if (queue_index != new_index && sk && 4046 sk_fullsock(sk) && 4047 rcu_access_pointer(sk->sk_dst_cache)) 4048 sk_tx_queue_set(sk, new_index); 4049 4050 queue_index = new_index; 4051 } 4052 4053 return queue_index; 4054 } 4055 EXPORT_SYMBOL(netdev_pick_tx); 4056 4057 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev, 4058 struct sk_buff *skb, 4059 struct net_device *sb_dev) 4060 { 4061 int queue_index = 0; 4062 4063 #ifdef CONFIG_XPS 4064 u32 sender_cpu = skb->sender_cpu - 1; 4065 4066 if (sender_cpu >= (u32)NR_CPUS) 4067 skb->sender_cpu = raw_smp_processor_id() + 1; 4068 #endif 4069 4070 if (dev->real_num_tx_queues != 1) { 4071 const struct net_device_ops *ops = dev->netdev_ops; 4072 4073 if (ops->ndo_select_queue) 4074 queue_index = ops->ndo_select_queue(dev, skb, sb_dev); 4075 else 4076 queue_index = netdev_pick_tx(dev, skb, sb_dev); 4077 4078 queue_index = netdev_cap_txqueue(dev, queue_index); 4079 } 4080 4081 skb_set_queue_mapping(skb, queue_index); 4082 return netdev_get_tx_queue(dev, queue_index); 4083 } 4084 4085 /** 4086 * __dev_queue_xmit - transmit a buffer 4087 * @skb: buffer to transmit 4088 * @sb_dev: suboordinate device used for L2 forwarding offload 4089 * 4090 * Queue a buffer for transmission to a network device. The caller must 4091 * have set the device and priority and built the buffer before calling 4092 * this function. The function can be called from an interrupt. 4093 * 4094 * A negative errno code is returned on a failure. A success does not 4095 * guarantee the frame will be transmitted as it may be dropped due 4096 * to congestion or traffic shaping. 4097 * 4098 * ----------------------------------------------------------------------------------- 4099 * I notice this method can also return errors from the queue disciplines, 4100 * including NET_XMIT_DROP, which is a positive value. So, errors can also 4101 * be positive. 4102 * 4103 * Regardless of the return value, the skb is consumed, so it is currently 4104 * difficult to retry a send to this method. (You can bump the ref count 4105 * before sending to hold a reference for retry if you are careful.) 4106 * 4107 * When calling this method, interrupts MUST be enabled. This is because 4108 * the BH enable code must have IRQs enabled so that it will not deadlock. 4109 * --BLG 4110 */ 4111 static int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev) 4112 { 4113 struct net_device *dev = skb->dev; 4114 struct netdev_queue *txq; 4115 struct Qdisc *q; 4116 int rc = -ENOMEM; 4117 bool again = false; 4118 4119 skb_reset_mac_header(skb); 4120 4121 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) 4122 __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED); 4123 4124 /* Disable soft irqs for various locks below. Also 4125 * stops preemption for RCU. 4126 */ 4127 rcu_read_lock_bh(); 4128 4129 skb_update_prio(skb); 4130 4131 qdisc_pkt_len_init(skb); 4132 #ifdef CONFIG_NET_CLS_ACT 4133 skb->tc_at_ingress = 0; 4134 # ifdef CONFIG_NET_EGRESS 4135 if (static_branch_unlikely(&egress_needed_key)) { 4136 skb = sch_handle_egress(skb, &rc, dev); 4137 if (!skb) 4138 goto out; 4139 } 4140 # endif 4141 #endif 4142 /* If device/qdisc don't need skb->dst, release it right now while 4143 * its hot in this cpu cache. 4144 */ 4145 if (dev->priv_flags & IFF_XMIT_DST_RELEASE) 4146 skb_dst_drop(skb); 4147 else 4148 skb_dst_force(skb); 4149 4150 txq = netdev_core_pick_tx(dev, skb, sb_dev); 4151 q = rcu_dereference_bh(txq->qdisc); 4152 4153 trace_net_dev_queue(skb); 4154 if (q->enqueue) { 4155 rc = __dev_xmit_skb(skb, q, dev, txq); 4156 goto out; 4157 } 4158 4159 /* The device has no queue. Common case for software devices: 4160 * loopback, all the sorts of tunnels... 4161 4162 * Really, it is unlikely that netif_tx_lock protection is necessary 4163 * here. (f.e. loopback and IP tunnels are clean ignoring statistics 4164 * counters.) 4165 * However, it is possible, that they rely on protection 4166 * made by us here. 4167 4168 * Check this and shot the lock. It is not prone from deadlocks. 4169 *Either shot noqueue qdisc, it is even simpler 8) 4170 */ 4171 if (dev->flags & IFF_UP) { 4172 int cpu = smp_processor_id(); /* ok because BHs are off */ 4173 4174 if (txq->xmit_lock_owner != cpu) { 4175 if (dev_xmit_recursion()) 4176 goto recursion_alert; 4177 4178 skb = validate_xmit_skb(skb, dev, &again); 4179 if (!skb) 4180 goto out; 4181 4182 PRANDOM_ADD_NOISE(skb, dev, txq, jiffies); 4183 HARD_TX_LOCK(dev, txq, cpu); 4184 4185 if (!netif_xmit_stopped(txq)) { 4186 dev_xmit_recursion_inc(); 4187 skb = dev_hard_start_xmit(skb, dev, txq, &rc); 4188 dev_xmit_recursion_dec(); 4189 if (dev_xmit_complete(rc)) { 4190 HARD_TX_UNLOCK(dev, txq); 4191 goto out; 4192 } 4193 } 4194 HARD_TX_UNLOCK(dev, txq); 4195 net_crit_ratelimited("Virtual device %s asks to queue packet!\n", 4196 dev->name); 4197 } else { 4198 /* Recursion is detected! It is possible, 4199 * unfortunately 4200 */ 4201 recursion_alert: 4202 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", 4203 dev->name); 4204 } 4205 } 4206 4207 rc = -ENETDOWN; 4208 rcu_read_unlock_bh(); 4209 4210 atomic_long_inc(&dev->tx_dropped); 4211 kfree_skb_list(skb); 4212 return rc; 4213 out: 4214 rcu_read_unlock_bh(); 4215 return rc; 4216 } 4217 4218 int dev_queue_xmit(struct sk_buff *skb) 4219 { 4220 return __dev_queue_xmit(skb, NULL); 4221 } 4222 EXPORT_SYMBOL(dev_queue_xmit); 4223 4224 int dev_queue_xmit_accel(struct sk_buff *skb, struct net_device *sb_dev) 4225 { 4226 return __dev_queue_xmit(skb, sb_dev); 4227 } 4228 EXPORT_SYMBOL(dev_queue_xmit_accel); 4229 4230 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id) 4231 { 4232 struct net_device *dev = skb->dev; 4233 struct sk_buff *orig_skb = skb; 4234 struct netdev_queue *txq; 4235 int ret = NETDEV_TX_BUSY; 4236 bool again = false; 4237 4238 if (unlikely(!netif_running(dev) || 4239 !netif_carrier_ok(dev))) 4240 goto drop; 4241 4242 skb = validate_xmit_skb_list(skb, dev, &again); 4243 if (skb != orig_skb) 4244 goto drop; 4245 4246 skb_set_queue_mapping(skb, queue_id); 4247 txq = skb_get_tx_queue(dev, skb); 4248 PRANDOM_ADD_NOISE(skb, dev, txq, jiffies); 4249 4250 local_bh_disable(); 4251 4252 dev_xmit_recursion_inc(); 4253 HARD_TX_LOCK(dev, txq, smp_processor_id()); 4254 if (!netif_xmit_frozen_or_drv_stopped(txq)) 4255 ret = netdev_start_xmit(skb, dev, txq, false); 4256 HARD_TX_UNLOCK(dev, txq); 4257 dev_xmit_recursion_dec(); 4258 4259 local_bh_enable(); 4260 return ret; 4261 drop: 4262 atomic_long_inc(&dev->tx_dropped); 4263 kfree_skb_list(skb); 4264 return NET_XMIT_DROP; 4265 } 4266 EXPORT_SYMBOL(__dev_direct_xmit); 4267 4268 /************************************************************************* 4269 * Receiver routines 4270 *************************************************************************/ 4271 4272 int netdev_max_backlog __read_mostly = 1000; 4273 EXPORT_SYMBOL(netdev_max_backlog); 4274 4275 int netdev_tstamp_prequeue __read_mostly = 1; 4276 int netdev_budget __read_mostly = 300; 4277 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */ 4278 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ; 4279 int weight_p __read_mostly = 64; /* old backlog weight */ 4280 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ 4281 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ 4282 int dev_rx_weight __read_mostly = 64; 4283 int dev_tx_weight __read_mostly = 64; 4284 /* Maximum number of GRO_NORMAL skbs to batch up for list-RX */ 4285 int gro_normal_batch __read_mostly = 8; 4286 4287 /* Called with irq disabled */ 4288 static inline void ____napi_schedule(struct softnet_data *sd, 4289 struct napi_struct *napi) 4290 { 4291 struct task_struct *thread; 4292 4293 if (test_bit(NAPI_STATE_THREADED, &napi->state)) { 4294 /* Paired with smp_mb__before_atomic() in 4295 * napi_enable()/dev_set_threaded(). 4296 * Use READ_ONCE() to guarantee a complete 4297 * read on napi->thread. Only call 4298 * wake_up_process() when it's not NULL. 4299 */ 4300 thread = READ_ONCE(napi->thread); 4301 if (thread) { 4302 wake_up_process(thread); 4303 return; 4304 } 4305 } 4306 4307 list_add_tail(&napi->poll_list, &sd->poll_list); 4308 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4309 } 4310 4311 #ifdef CONFIG_RPS 4312 4313 /* One global table that all flow-based protocols share. */ 4314 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; 4315 EXPORT_SYMBOL(rps_sock_flow_table); 4316 u32 rps_cpu_mask __read_mostly; 4317 EXPORT_SYMBOL(rps_cpu_mask); 4318 4319 struct static_key_false rps_needed __read_mostly; 4320 EXPORT_SYMBOL(rps_needed); 4321 struct static_key_false rfs_needed __read_mostly; 4322 EXPORT_SYMBOL(rfs_needed); 4323 4324 static struct rps_dev_flow * 4325 set_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4326 struct rps_dev_flow *rflow, u16 next_cpu) 4327 { 4328 if (next_cpu < nr_cpu_ids) { 4329 #ifdef CONFIG_RFS_ACCEL 4330 struct netdev_rx_queue *rxqueue; 4331 struct rps_dev_flow_table *flow_table; 4332 struct rps_dev_flow *old_rflow; 4333 u32 flow_id; 4334 u16 rxq_index; 4335 int rc; 4336 4337 /* Should we steer this flow to a different hardware queue? */ 4338 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || 4339 !(dev->features & NETIF_F_NTUPLE)) 4340 goto out; 4341 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); 4342 if (rxq_index == skb_get_rx_queue(skb)) 4343 goto out; 4344 4345 rxqueue = dev->_rx + rxq_index; 4346 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4347 if (!flow_table) 4348 goto out; 4349 flow_id = skb_get_hash(skb) & flow_table->mask; 4350 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, 4351 rxq_index, flow_id); 4352 if (rc < 0) 4353 goto out; 4354 old_rflow = rflow; 4355 rflow = &flow_table->flows[flow_id]; 4356 rflow->filter = rc; 4357 if (old_rflow->filter == rflow->filter) 4358 old_rflow->filter = RPS_NO_FILTER; 4359 out: 4360 #endif 4361 rflow->last_qtail = 4362 per_cpu(softnet_data, next_cpu).input_queue_head; 4363 } 4364 4365 rflow->cpu = next_cpu; 4366 return rflow; 4367 } 4368 4369 /* 4370 * get_rps_cpu is called from netif_receive_skb and returns the target 4371 * CPU from the RPS map of the receiving queue for a given skb. 4372 * rcu_read_lock must be held on entry. 4373 */ 4374 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4375 struct rps_dev_flow **rflowp) 4376 { 4377 const struct rps_sock_flow_table *sock_flow_table; 4378 struct netdev_rx_queue *rxqueue = dev->_rx; 4379 struct rps_dev_flow_table *flow_table; 4380 struct rps_map *map; 4381 int cpu = -1; 4382 u32 tcpu; 4383 u32 hash; 4384 4385 if (skb_rx_queue_recorded(skb)) { 4386 u16 index = skb_get_rx_queue(skb); 4387 4388 if (unlikely(index >= dev->real_num_rx_queues)) { 4389 WARN_ONCE(dev->real_num_rx_queues > 1, 4390 "%s received packet on queue %u, but number " 4391 "of RX queues is %u\n", 4392 dev->name, index, dev->real_num_rx_queues); 4393 goto done; 4394 } 4395 rxqueue += index; 4396 } 4397 4398 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ 4399 4400 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4401 map = rcu_dereference(rxqueue->rps_map); 4402 if (!flow_table && !map) 4403 goto done; 4404 4405 skb_reset_network_header(skb); 4406 hash = skb_get_hash(skb); 4407 if (!hash) 4408 goto done; 4409 4410 sock_flow_table = rcu_dereference(rps_sock_flow_table); 4411 if (flow_table && sock_flow_table) { 4412 struct rps_dev_flow *rflow; 4413 u32 next_cpu; 4414 u32 ident; 4415 4416 /* First check into global flow table if there is a match */ 4417 ident = sock_flow_table->ents[hash & sock_flow_table->mask]; 4418 if ((ident ^ hash) & ~rps_cpu_mask) 4419 goto try_rps; 4420 4421 next_cpu = ident & rps_cpu_mask; 4422 4423 /* OK, now we know there is a match, 4424 * we can look at the local (per receive queue) flow table 4425 */ 4426 rflow = &flow_table->flows[hash & flow_table->mask]; 4427 tcpu = rflow->cpu; 4428 4429 /* 4430 * If the desired CPU (where last recvmsg was done) is 4431 * different from current CPU (one in the rx-queue flow 4432 * table entry), switch if one of the following holds: 4433 * - Current CPU is unset (>= nr_cpu_ids). 4434 * - Current CPU is offline. 4435 * - The current CPU's queue tail has advanced beyond the 4436 * last packet that was enqueued using this table entry. 4437 * This guarantees that all previous packets for the flow 4438 * have been dequeued, thus preserving in order delivery. 4439 */ 4440 if (unlikely(tcpu != next_cpu) && 4441 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || 4442 ((int)(per_cpu(softnet_data, tcpu).input_queue_head - 4443 rflow->last_qtail)) >= 0)) { 4444 tcpu = next_cpu; 4445 rflow = set_rps_cpu(dev, skb, rflow, next_cpu); 4446 } 4447 4448 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { 4449 *rflowp = rflow; 4450 cpu = tcpu; 4451 goto done; 4452 } 4453 } 4454 4455 try_rps: 4456 4457 if (map) { 4458 tcpu = map->cpus[reciprocal_scale(hash, map->len)]; 4459 if (cpu_online(tcpu)) { 4460 cpu = tcpu; 4461 goto done; 4462 } 4463 } 4464 4465 done: 4466 return cpu; 4467 } 4468 4469 #ifdef CONFIG_RFS_ACCEL 4470 4471 /** 4472 * rps_may_expire_flow - check whether an RFS hardware filter may be removed 4473 * @dev: Device on which the filter was set 4474 * @rxq_index: RX queue index 4475 * @flow_id: Flow ID passed to ndo_rx_flow_steer() 4476 * @filter_id: Filter ID returned by ndo_rx_flow_steer() 4477 * 4478 * Drivers that implement ndo_rx_flow_steer() should periodically call 4479 * this function for each installed filter and remove the filters for 4480 * which it returns %true. 4481 */ 4482 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, 4483 u32 flow_id, u16 filter_id) 4484 { 4485 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; 4486 struct rps_dev_flow_table *flow_table; 4487 struct rps_dev_flow *rflow; 4488 bool expire = true; 4489 unsigned int cpu; 4490 4491 rcu_read_lock(); 4492 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4493 if (flow_table && flow_id <= flow_table->mask) { 4494 rflow = &flow_table->flows[flow_id]; 4495 cpu = READ_ONCE(rflow->cpu); 4496 if (rflow->filter == filter_id && cpu < nr_cpu_ids && 4497 ((int)(per_cpu(softnet_data, cpu).input_queue_head - 4498 rflow->last_qtail) < 4499 (int)(10 * flow_table->mask))) 4500 expire = false; 4501 } 4502 rcu_read_unlock(); 4503 return expire; 4504 } 4505 EXPORT_SYMBOL(rps_may_expire_flow); 4506 4507 #endif /* CONFIG_RFS_ACCEL */ 4508 4509 /* Called from hardirq (IPI) context */ 4510 static void rps_trigger_softirq(void *data) 4511 { 4512 struct softnet_data *sd = data; 4513 4514 ____napi_schedule(sd, &sd->backlog); 4515 sd->received_rps++; 4516 } 4517 4518 #endif /* CONFIG_RPS */ 4519 4520 /* 4521 * Check if this softnet_data structure is another cpu one 4522 * If yes, queue it to our IPI list and return 1 4523 * If no, return 0 4524 */ 4525 static int rps_ipi_queued(struct softnet_data *sd) 4526 { 4527 #ifdef CONFIG_RPS 4528 struct softnet_data *mysd = this_cpu_ptr(&softnet_data); 4529 4530 if (sd != mysd) { 4531 sd->rps_ipi_next = mysd->rps_ipi_list; 4532 mysd->rps_ipi_list = sd; 4533 4534 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4535 return 1; 4536 } 4537 #endif /* CONFIG_RPS */ 4538 return 0; 4539 } 4540 4541 #ifdef CONFIG_NET_FLOW_LIMIT 4542 int netdev_flow_limit_table_len __read_mostly = (1 << 12); 4543 #endif 4544 4545 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) 4546 { 4547 #ifdef CONFIG_NET_FLOW_LIMIT 4548 struct sd_flow_limit *fl; 4549 struct softnet_data *sd; 4550 unsigned int old_flow, new_flow; 4551 4552 if (qlen < (netdev_max_backlog >> 1)) 4553 return false; 4554 4555 sd = this_cpu_ptr(&softnet_data); 4556 4557 rcu_read_lock(); 4558 fl = rcu_dereference(sd->flow_limit); 4559 if (fl) { 4560 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); 4561 old_flow = fl->history[fl->history_head]; 4562 fl->history[fl->history_head] = new_flow; 4563 4564 fl->history_head++; 4565 fl->history_head &= FLOW_LIMIT_HISTORY - 1; 4566 4567 if (likely(fl->buckets[old_flow])) 4568 fl->buckets[old_flow]--; 4569 4570 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { 4571 fl->count++; 4572 rcu_read_unlock(); 4573 return true; 4574 } 4575 } 4576 rcu_read_unlock(); 4577 #endif 4578 return false; 4579 } 4580 4581 /* 4582 * enqueue_to_backlog is called to queue an skb to a per CPU backlog 4583 * queue (may be a remote CPU queue). 4584 */ 4585 static int enqueue_to_backlog(struct sk_buff *skb, int cpu, 4586 unsigned int *qtail) 4587 { 4588 struct softnet_data *sd; 4589 unsigned long flags; 4590 unsigned int qlen; 4591 4592 sd = &per_cpu(softnet_data, cpu); 4593 4594 local_irq_save(flags); 4595 4596 rps_lock(sd); 4597 if (!netif_running(skb->dev)) 4598 goto drop; 4599 qlen = skb_queue_len(&sd->input_pkt_queue); 4600 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) { 4601 if (qlen) { 4602 enqueue: 4603 __skb_queue_tail(&sd->input_pkt_queue, skb); 4604 input_queue_tail_incr_save(sd, qtail); 4605 rps_unlock(sd); 4606 local_irq_restore(flags); 4607 return NET_RX_SUCCESS; 4608 } 4609 4610 /* Schedule NAPI for backlog device 4611 * We can use non atomic operation since we own the queue lock 4612 */ 4613 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) { 4614 if (!rps_ipi_queued(sd)) 4615 ____napi_schedule(sd, &sd->backlog); 4616 } 4617 goto enqueue; 4618 } 4619 4620 drop: 4621 sd->dropped++; 4622 rps_unlock(sd); 4623 4624 local_irq_restore(flags); 4625 4626 atomic_long_inc(&skb->dev->rx_dropped); 4627 kfree_skb(skb); 4628 return NET_RX_DROP; 4629 } 4630 4631 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb) 4632 { 4633 struct net_device *dev = skb->dev; 4634 struct netdev_rx_queue *rxqueue; 4635 4636 rxqueue = dev->_rx; 4637 4638 if (skb_rx_queue_recorded(skb)) { 4639 u16 index = skb_get_rx_queue(skb); 4640 4641 if (unlikely(index >= dev->real_num_rx_queues)) { 4642 WARN_ONCE(dev->real_num_rx_queues > 1, 4643 "%s received packet on queue %u, but number " 4644 "of RX queues is %u\n", 4645 dev->name, index, dev->real_num_rx_queues); 4646 4647 return rxqueue; /* Return first rxqueue */ 4648 } 4649 rxqueue += index; 4650 } 4651 return rxqueue; 4652 } 4653 4654 static u32 netif_receive_generic_xdp(struct sk_buff *skb, 4655 struct xdp_buff *xdp, 4656 struct bpf_prog *xdp_prog) 4657 { 4658 void *orig_data, *orig_data_end, *hard_start; 4659 struct netdev_rx_queue *rxqueue; 4660 u32 metalen, act = XDP_DROP; 4661 u32 mac_len, frame_sz; 4662 __be16 orig_eth_type; 4663 struct ethhdr *eth; 4664 bool orig_bcast; 4665 int off; 4666 4667 /* Reinjected packets coming from act_mirred or similar should 4668 * not get XDP generic processing. 4669 */ 4670 if (skb_is_redirected(skb)) 4671 return XDP_PASS; 4672 4673 /* XDP packets must be linear and must have sufficient headroom 4674 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also 4675 * native XDP provides, thus we need to do it here as well. 4676 */ 4677 if (skb_cloned(skb) || skb_is_nonlinear(skb) || 4678 skb_headroom(skb) < XDP_PACKET_HEADROOM) { 4679 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb); 4680 int troom = skb->tail + skb->data_len - skb->end; 4681 4682 /* In case we have to go down the path and also linearize, 4683 * then lets do the pskb_expand_head() work just once here. 4684 */ 4685 if (pskb_expand_head(skb, 4686 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0, 4687 troom > 0 ? troom + 128 : 0, GFP_ATOMIC)) 4688 goto do_drop; 4689 if (skb_linearize(skb)) 4690 goto do_drop; 4691 } 4692 4693 /* The XDP program wants to see the packet starting at the MAC 4694 * header. 4695 */ 4696 mac_len = skb->data - skb_mac_header(skb); 4697 hard_start = skb->data - skb_headroom(skb); 4698 4699 /* SKB "head" area always have tailroom for skb_shared_info */ 4700 frame_sz = (void *)skb_end_pointer(skb) - hard_start; 4701 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 4702 4703 rxqueue = netif_get_rxqueue(skb); 4704 xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq); 4705 xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len, 4706 skb_headlen(skb) + mac_len, true); 4707 4708 orig_data_end = xdp->data_end; 4709 orig_data = xdp->data; 4710 eth = (struct ethhdr *)xdp->data; 4711 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest); 4712 orig_eth_type = eth->h_proto; 4713 4714 act = bpf_prog_run_xdp(xdp_prog, xdp); 4715 4716 /* check if bpf_xdp_adjust_head was used */ 4717 off = xdp->data - orig_data; 4718 if (off) { 4719 if (off > 0) 4720 __skb_pull(skb, off); 4721 else if (off < 0) 4722 __skb_push(skb, -off); 4723 4724 skb->mac_header += off; 4725 skb_reset_network_header(skb); 4726 } 4727 4728 /* check if bpf_xdp_adjust_tail was used */ 4729 off = xdp->data_end - orig_data_end; 4730 if (off != 0) { 4731 skb_set_tail_pointer(skb, xdp->data_end - xdp->data); 4732 skb->len += off; /* positive on grow, negative on shrink */ 4733 } 4734 4735 /* check if XDP changed eth hdr such SKB needs update */ 4736 eth = (struct ethhdr *)xdp->data; 4737 if ((orig_eth_type != eth->h_proto) || 4738 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) { 4739 __skb_push(skb, ETH_HLEN); 4740 skb->protocol = eth_type_trans(skb, skb->dev); 4741 } 4742 4743 switch (act) { 4744 case XDP_REDIRECT: 4745 case XDP_TX: 4746 __skb_push(skb, mac_len); 4747 break; 4748 case XDP_PASS: 4749 metalen = xdp->data - xdp->data_meta; 4750 if (metalen) 4751 skb_metadata_set(skb, metalen); 4752 break; 4753 default: 4754 bpf_warn_invalid_xdp_action(act); 4755 fallthrough; 4756 case XDP_ABORTED: 4757 trace_xdp_exception(skb->dev, xdp_prog, act); 4758 fallthrough; 4759 case XDP_DROP: 4760 do_drop: 4761 kfree_skb(skb); 4762 break; 4763 } 4764 4765 return act; 4766 } 4767 4768 /* When doing generic XDP we have to bypass the qdisc layer and the 4769 * network taps in order to match in-driver-XDP behavior. 4770 */ 4771 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog) 4772 { 4773 struct net_device *dev = skb->dev; 4774 struct netdev_queue *txq; 4775 bool free_skb = true; 4776 int cpu, rc; 4777 4778 txq = netdev_core_pick_tx(dev, skb, NULL); 4779 cpu = smp_processor_id(); 4780 HARD_TX_LOCK(dev, txq, cpu); 4781 if (!netif_xmit_stopped(txq)) { 4782 rc = netdev_start_xmit(skb, dev, txq, 0); 4783 if (dev_xmit_complete(rc)) 4784 free_skb = false; 4785 } 4786 HARD_TX_UNLOCK(dev, txq); 4787 if (free_skb) { 4788 trace_xdp_exception(dev, xdp_prog, XDP_TX); 4789 kfree_skb(skb); 4790 } 4791 } 4792 4793 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key); 4794 4795 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb) 4796 { 4797 if (xdp_prog) { 4798 struct xdp_buff xdp; 4799 u32 act; 4800 int err; 4801 4802 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog); 4803 if (act != XDP_PASS) { 4804 switch (act) { 4805 case XDP_REDIRECT: 4806 err = xdp_do_generic_redirect(skb->dev, skb, 4807 &xdp, xdp_prog); 4808 if (err) 4809 goto out_redir; 4810 break; 4811 case XDP_TX: 4812 generic_xdp_tx(skb, xdp_prog); 4813 break; 4814 } 4815 return XDP_DROP; 4816 } 4817 } 4818 return XDP_PASS; 4819 out_redir: 4820 kfree_skb(skb); 4821 return XDP_DROP; 4822 } 4823 EXPORT_SYMBOL_GPL(do_xdp_generic); 4824 4825 static int netif_rx_internal(struct sk_buff *skb) 4826 { 4827 int ret; 4828 4829 net_timestamp_check(netdev_tstamp_prequeue, skb); 4830 4831 trace_netif_rx(skb); 4832 4833 #ifdef CONFIG_RPS 4834 if (static_branch_unlikely(&rps_needed)) { 4835 struct rps_dev_flow voidflow, *rflow = &voidflow; 4836 int cpu; 4837 4838 preempt_disable(); 4839 rcu_read_lock(); 4840 4841 cpu = get_rps_cpu(skb->dev, skb, &rflow); 4842 if (cpu < 0) 4843 cpu = smp_processor_id(); 4844 4845 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 4846 4847 rcu_read_unlock(); 4848 preempt_enable(); 4849 } else 4850 #endif 4851 { 4852 unsigned int qtail; 4853 4854 ret = enqueue_to_backlog(skb, get_cpu(), &qtail); 4855 put_cpu(); 4856 } 4857 return ret; 4858 } 4859 4860 /** 4861 * netif_rx - post buffer to the network code 4862 * @skb: buffer to post 4863 * 4864 * This function receives a packet from a device driver and queues it for 4865 * the upper (protocol) levels to process. It always succeeds. The buffer 4866 * may be dropped during processing for congestion control or by the 4867 * protocol layers. 4868 * 4869 * return values: 4870 * NET_RX_SUCCESS (no congestion) 4871 * NET_RX_DROP (packet was dropped) 4872 * 4873 */ 4874 4875 int netif_rx(struct sk_buff *skb) 4876 { 4877 int ret; 4878 4879 trace_netif_rx_entry(skb); 4880 4881 ret = netif_rx_internal(skb); 4882 trace_netif_rx_exit(ret); 4883 4884 return ret; 4885 } 4886 EXPORT_SYMBOL(netif_rx); 4887 4888 int netif_rx_ni(struct sk_buff *skb) 4889 { 4890 int err; 4891 4892 trace_netif_rx_ni_entry(skb); 4893 4894 preempt_disable(); 4895 err = netif_rx_internal(skb); 4896 if (local_softirq_pending()) 4897 do_softirq(); 4898 preempt_enable(); 4899 trace_netif_rx_ni_exit(err); 4900 4901 return err; 4902 } 4903 EXPORT_SYMBOL(netif_rx_ni); 4904 4905 int netif_rx_any_context(struct sk_buff *skb) 4906 { 4907 /* 4908 * If invoked from contexts which do not invoke bottom half 4909 * processing either at return from interrupt or when softrqs are 4910 * reenabled, use netif_rx_ni() which invokes bottomhalf processing 4911 * directly. 4912 */ 4913 if (in_interrupt()) 4914 return netif_rx(skb); 4915 else 4916 return netif_rx_ni(skb); 4917 } 4918 EXPORT_SYMBOL(netif_rx_any_context); 4919 4920 static __latent_entropy void net_tx_action(struct softirq_action *h) 4921 { 4922 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 4923 4924 if (sd->completion_queue) { 4925 struct sk_buff *clist; 4926 4927 local_irq_disable(); 4928 clist = sd->completion_queue; 4929 sd->completion_queue = NULL; 4930 local_irq_enable(); 4931 4932 while (clist) { 4933 struct sk_buff *skb = clist; 4934 4935 clist = clist->next; 4936 4937 WARN_ON(refcount_read(&skb->users)); 4938 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED)) 4939 trace_consume_skb(skb); 4940 else 4941 trace_kfree_skb(skb, net_tx_action); 4942 4943 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) 4944 __kfree_skb(skb); 4945 else 4946 __kfree_skb_defer(skb); 4947 } 4948 } 4949 4950 if (sd->output_queue) { 4951 struct Qdisc *head; 4952 4953 local_irq_disable(); 4954 head = sd->output_queue; 4955 sd->output_queue = NULL; 4956 sd->output_queue_tailp = &sd->output_queue; 4957 local_irq_enable(); 4958 4959 while (head) { 4960 struct Qdisc *q = head; 4961 spinlock_t *root_lock = NULL; 4962 4963 head = head->next_sched; 4964 4965 if (!(q->flags & TCQ_F_NOLOCK)) { 4966 root_lock = qdisc_lock(q); 4967 spin_lock(root_lock); 4968 } 4969 /* We need to make sure head->next_sched is read 4970 * before clearing __QDISC_STATE_SCHED 4971 */ 4972 smp_mb__before_atomic(); 4973 clear_bit(__QDISC_STATE_SCHED, &q->state); 4974 qdisc_run(q); 4975 if (root_lock) 4976 spin_unlock(root_lock); 4977 } 4978 } 4979 4980 xfrm_dev_backlog(sd); 4981 } 4982 4983 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) 4984 /* This hook is defined here for ATM LANE */ 4985 int (*br_fdb_test_addr_hook)(struct net_device *dev, 4986 unsigned char *addr) __read_mostly; 4987 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); 4988 #endif 4989 4990 static inline struct sk_buff * 4991 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 4992 struct net_device *orig_dev, bool *another) 4993 { 4994 #ifdef CONFIG_NET_CLS_ACT 4995 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress); 4996 struct tcf_result cl_res; 4997 4998 /* If there's at least one ingress present somewhere (so 4999 * we get here via enabled static key), remaining devices 5000 * that are not configured with an ingress qdisc will bail 5001 * out here. 5002 */ 5003 if (!miniq) 5004 return skb; 5005 5006 if (*pt_prev) { 5007 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5008 *pt_prev = NULL; 5009 } 5010 5011 qdisc_skb_cb(skb)->pkt_len = skb->len; 5012 qdisc_skb_cb(skb)->mru = 0; 5013 qdisc_skb_cb(skb)->post_ct = false; 5014 skb->tc_at_ingress = 1; 5015 mini_qdisc_bstats_cpu_update(miniq, skb); 5016 5017 switch (tcf_classify_ingress(skb, miniq->block, miniq->filter_list, 5018 &cl_res, false)) { 5019 case TC_ACT_OK: 5020 case TC_ACT_RECLASSIFY: 5021 skb->tc_index = TC_H_MIN(cl_res.classid); 5022 break; 5023 case TC_ACT_SHOT: 5024 mini_qdisc_qstats_cpu_drop(miniq); 5025 kfree_skb(skb); 5026 return NULL; 5027 case TC_ACT_STOLEN: 5028 case TC_ACT_QUEUED: 5029 case TC_ACT_TRAP: 5030 consume_skb(skb); 5031 return NULL; 5032 case TC_ACT_REDIRECT: 5033 /* skb_mac_header check was done by cls/act_bpf, so 5034 * we can safely push the L2 header back before 5035 * redirecting to another netdev 5036 */ 5037 __skb_push(skb, skb->mac_len); 5038 if (skb_do_redirect(skb) == -EAGAIN) { 5039 __skb_pull(skb, skb->mac_len); 5040 *another = true; 5041 break; 5042 } 5043 return NULL; 5044 case TC_ACT_CONSUMED: 5045 return NULL; 5046 default: 5047 break; 5048 } 5049 #endif /* CONFIG_NET_CLS_ACT */ 5050 return skb; 5051 } 5052 5053 /** 5054 * netdev_is_rx_handler_busy - check if receive handler is registered 5055 * @dev: device to check 5056 * 5057 * Check if a receive handler is already registered for a given device. 5058 * Return true if there one. 5059 * 5060 * The caller must hold the rtnl_mutex. 5061 */ 5062 bool netdev_is_rx_handler_busy(struct net_device *dev) 5063 { 5064 ASSERT_RTNL(); 5065 return dev && rtnl_dereference(dev->rx_handler); 5066 } 5067 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); 5068 5069 /** 5070 * netdev_rx_handler_register - register receive handler 5071 * @dev: device to register a handler for 5072 * @rx_handler: receive handler to register 5073 * @rx_handler_data: data pointer that is used by rx handler 5074 * 5075 * Register a receive handler for a device. This handler will then be 5076 * called from __netif_receive_skb. A negative errno code is returned 5077 * on a failure. 5078 * 5079 * The caller must hold the rtnl_mutex. 5080 * 5081 * For a general description of rx_handler, see enum rx_handler_result. 5082 */ 5083 int netdev_rx_handler_register(struct net_device *dev, 5084 rx_handler_func_t *rx_handler, 5085 void *rx_handler_data) 5086 { 5087 if (netdev_is_rx_handler_busy(dev)) 5088 return -EBUSY; 5089 5090 if (dev->priv_flags & IFF_NO_RX_HANDLER) 5091 return -EINVAL; 5092 5093 /* Note: rx_handler_data must be set before rx_handler */ 5094 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); 5095 rcu_assign_pointer(dev->rx_handler, rx_handler); 5096 5097 return 0; 5098 } 5099 EXPORT_SYMBOL_GPL(netdev_rx_handler_register); 5100 5101 /** 5102 * netdev_rx_handler_unregister - unregister receive handler 5103 * @dev: device to unregister a handler from 5104 * 5105 * Unregister a receive handler from a device. 5106 * 5107 * The caller must hold the rtnl_mutex. 5108 */ 5109 void netdev_rx_handler_unregister(struct net_device *dev) 5110 { 5111 5112 ASSERT_RTNL(); 5113 RCU_INIT_POINTER(dev->rx_handler, NULL); 5114 /* a reader seeing a non NULL rx_handler in a rcu_read_lock() 5115 * section has a guarantee to see a non NULL rx_handler_data 5116 * as well. 5117 */ 5118 synchronize_net(); 5119 RCU_INIT_POINTER(dev->rx_handler_data, NULL); 5120 } 5121 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); 5122 5123 /* 5124 * Limit the use of PFMEMALLOC reserves to those protocols that implement 5125 * the special handling of PFMEMALLOC skbs. 5126 */ 5127 static bool skb_pfmemalloc_protocol(struct sk_buff *skb) 5128 { 5129 switch (skb->protocol) { 5130 case htons(ETH_P_ARP): 5131 case htons(ETH_P_IP): 5132 case htons(ETH_P_IPV6): 5133 case htons(ETH_P_8021Q): 5134 case htons(ETH_P_8021AD): 5135 return true; 5136 default: 5137 return false; 5138 } 5139 } 5140 5141 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, 5142 int *ret, struct net_device *orig_dev) 5143 { 5144 if (nf_hook_ingress_active(skb)) { 5145 int ingress_retval; 5146 5147 if (*pt_prev) { 5148 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5149 *pt_prev = NULL; 5150 } 5151 5152 rcu_read_lock(); 5153 ingress_retval = nf_hook_ingress(skb); 5154 rcu_read_unlock(); 5155 return ingress_retval; 5156 } 5157 return 0; 5158 } 5159 5160 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc, 5161 struct packet_type **ppt_prev) 5162 { 5163 struct packet_type *ptype, *pt_prev; 5164 rx_handler_func_t *rx_handler; 5165 struct sk_buff *skb = *pskb; 5166 struct net_device *orig_dev; 5167 bool deliver_exact = false; 5168 int ret = NET_RX_DROP; 5169 __be16 type; 5170 5171 net_timestamp_check(!netdev_tstamp_prequeue, skb); 5172 5173 trace_netif_receive_skb(skb); 5174 5175 orig_dev = skb->dev; 5176 5177 skb_reset_network_header(skb); 5178 if (!skb_transport_header_was_set(skb)) 5179 skb_reset_transport_header(skb); 5180 skb_reset_mac_len(skb); 5181 5182 pt_prev = NULL; 5183 5184 another_round: 5185 skb->skb_iif = skb->dev->ifindex; 5186 5187 __this_cpu_inc(softnet_data.processed); 5188 5189 if (static_branch_unlikely(&generic_xdp_needed_key)) { 5190 int ret2; 5191 5192 preempt_disable(); 5193 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb); 5194 preempt_enable(); 5195 5196 if (ret2 != XDP_PASS) { 5197 ret = NET_RX_DROP; 5198 goto out; 5199 } 5200 skb_reset_mac_len(skb); 5201 } 5202 5203 if (eth_type_vlan(skb->protocol)) { 5204 skb = skb_vlan_untag(skb); 5205 if (unlikely(!skb)) 5206 goto out; 5207 } 5208 5209 if (skb_skip_tc_classify(skb)) 5210 goto skip_classify; 5211 5212 if (pfmemalloc) 5213 goto skip_taps; 5214 5215 list_for_each_entry_rcu(ptype, &ptype_all, list) { 5216 if (pt_prev) 5217 ret = deliver_skb(skb, pt_prev, orig_dev); 5218 pt_prev = ptype; 5219 } 5220 5221 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { 5222 if (pt_prev) 5223 ret = deliver_skb(skb, pt_prev, orig_dev); 5224 pt_prev = ptype; 5225 } 5226 5227 skip_taps: 5228 #ifdef CONFIG_NET_INGRESS 5229 if (static_branch_unlikely(&ingress_needed_key)) { 5230 bool another = false; 5231 5232 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev, 5233 &another); 5234 if (another) 5235 goto another_round; 5236 if (!skb) 5237 goto out; 5238 5239 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) 5240 goto out; 5241 } 5242 #endif 5243 skb_reset_redirect(skb); 5244 skip_classify: 5245 if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) 5246 goto drop; 5247 5248 if (skb_vlan_tag_present(skb)) { 5249 if (pt_prev) { 5250 ret = deliver_skb(skb, pt_prev, orig_dev); 5251 pt_prev = NULL; 5252 } 5253 if (vlan_do_receive(&skb)) 5254 goto another_round; 5255 else if (unlikely(!skb)) 5256 goto out; 5257 } 5258 5259 rx_handler = rcu_dereference(skb->dev->rx_handler); 5260 if (rx_handler) { 5261 if (pt_prev) { 5262 ret = deliver_skb(skb, pt_prev, orig_dev); 5263 pt_prev = NULL; 5264 } 5265 switch (rx_handler(&skb)) { 5266 case RX_HANDLER_CONSUMED: 5267 ret = NET_RX_SUCCESS; 5268 goto out; 5269 case RX_HANDLER_ANOTHER: 5270 goto another_round; 5271 case RX_HANDLER_EXACT: 5272 deliver_exact = true; 5273 break; 5274 case RX_HANDLER_PASS: 5275 break; 5276 default: 5277 BUG(); 5278 } 5279 } 5280 5281 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) { 5282 check_vlan_id: 5283 if (skb_vlan_tag_get_id(skb)) { 5284 /* Vlan id is non 0 and vlan_do_receive() above couldn't 5285 * find vlan device. 5286 */ 5287 skb->pkt_type = PACKET_OTHERHOST; 5288 } else if (eth_type_vlan(skb->protocol)) { 5289 /* Outer header is 802.1P with vlan 0, inner header is 5290 * 802.1Q or 802.1AD and vlan_do_receive() above could 5291 * not find vlan dev for vlan id 0. 5292 */ 5293 __vlan_hwaccel_clear_tag(skb); 5294 skb = skb_vlan_untag(skb); 5295 if (unlikely(!skb)) 5296 goto out; 5297 if (vlan_do_receive(&skb)) 5298 /* After stripping off 802.1P header with vlan 0 5299 * vlan dev is found for inner header. 5300 */ 5301 goto another_round; 5302 else if (unlikely(!skb)) 5303 goto out; 5304 else 5305 /* We have stripped outer 802.1P vlan 0 header. 5306 * But could not find vlan dev. 5307 * check again for vlan id to set OTHERHOST. 5308 */ 5309 goto check_vlan_id; 5310 } 5311 /* Note: we might in the future use prio bits 5312 * and set skb->priority like in vlan_do_receive() 5313 * For the time being, just ignore Priority Code Point 5314 */ 5315 __vlan_hwaccel_clear_tag(skb); 5316 } 5317 5318 type = skb->protocol; 5319 5320 /* deliver only exact match when indicated */ 5321 if (likely(!deliver_exact)) { 5322 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5323 &ptype_base[ntohs(type) & 5324 PTYPE_HASH_MASK]); 5325 } 5326 5327 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5328 &orig_dev->ptype_specific); 5329 5330 if (unlikely(skb->dev != orig_dev)) { 5331 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5332 &skb->dev->ptype_specific); 5333 } 5334 5335 if (pt_prev) { 5336 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 5337 goto drop; 5338 *ppt_prev = pt_prev; 5339 } else { 5340 drop: 5341 if (!deliver_exact) 5342 atomic_long_inc(&skb->dev->rx_dropped); 5343 else 5344 atomic_long_inc(&skb->dev->rx_nohandler); 5345 kfree_skb(skb); 5346 /* Jamal, now you will not able to escape explaining 5347 * me how you were going to use this. :-) 5348 */ 5349 ret = NET_RX_DROP; 5350 } 5351 5352 out: 5353 /* The invariant here is that if *ppt_prev is not NULL 5354 * then skb should also be non-NULL. 5355 * 5356 * Apparently *ppt_prev assignment above holds this invariant due to 5357 * skb dereferencing near it. 5358 */ 5359 *pskb = skb; 5360 return ret; 5361 } 5362 5363 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc) 5364 { 5365 struct net_device *orig_dev = skb->dev; 5366 struct packet_type *pt_prev = NULL; 5367 int ret; 5368 5369 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5370 if (pt_prev) 5371 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb, 5372 skb->dev, pt_prev, orig_dev); 5373 return ret; 5374 } 5375 5376 /** 5377 * netif_receive_skb_core - special purpose version of netif_receive_skb 5378 * @skb: buffer to process 5379 * 5380 * More direct receive version of netif_receive_skb(). It should 5381 * only be used by callers that have a need to skip RPS and Generic XDP. 5382 * Caller must also take care of handling if ``(page_is_)pfmemalloc``. 5383 * 5384 * This function may only be called from softirq context and interrupts 5385 * should be enabled. 5386 * 5387 * Return values (usually ignored): 5388 * NET_RX_SUCCESS: no congestion 5389 * NET_RX_DROP: packet was dropped 5390 */ 5391 int netif_receive_skb_core(struct sk_buff *skb) 5392 { 5393 int ret; 5394 5395 rcu_read_lock(); 5396 ret = __netif_receive_skb_one_core(skb, false); 5397 rcu_read_unlock(); 5398 5399 return ret; 5400 } 5401 EXPORT_SYMBOL(netif_receive_skb_core); 5402 5403 static inline void __netif_receive_skb_list_ptype(struct list_head *head, 5404 struct packet_type *pt_prev, 5405 struct net_device *orig_dev) 5406 { 5407 struct sk_buff *skb, *next; 5408 5409 if (!pt_prev) 5410 return; 5411 if (list_empty(head)) 5412 return; 5413 if (pt_prev->list_func != NULL) 5414 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv, 5415 ip_list_rcv, head, pt_prev, orig_dev); 5416 else 5417 list_for_each_entry_safe(skb, next, head, list) { 5418 skb_list_del_init(skb); 5419 pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 5420 } 5421 } 5422 5423 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc) 5424 { 5425 /* Fast-path assumptions: 5426 * - There is no RX handler. 5427 * - Only one packet_type matches. 5428 * If either of these fails, we will end up doing some per-packet 5429 * processing in-line, then handling the 'last ptype' for the whole 5430 * sublist. This can't cause out-of-order delivery to any single ptype, 5431 * because the 'last ptype' must be constant across the sublist, and all 5432 * other ptypes are handled per-packet. 5433 */ 5434 /* Current (common) ptype of sublist */ 5435 struct packet_type *pt_curr = NULL; 5436 /* Current (common) orig_dev of sublist */ 5437 struct net_device *od_curr = NULL; 5438 struct list_head sublist; 5439 struct sk_buff *skb, *next; 5440 5441 INIT_LIST_HEAD(&sublist); 5442 list_for_each_entry_safe(skb, next, head, list) { 5443 struct net_device *orig_dev = skb->dev; 5444 struct packet_type *pt_prev = NULL; 5445 5446 skb_list_del_init(skb); 5447 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5448 if (!pt_prev) 5449 continue; 5450 if (pt_curr != pt_prev || od_curr != orig_dev) { 5451 /* dispatch old sublist */ 5452 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5453 /* start new sublist */ 5454 INIT_LIST_HEAD(&sublist); 5455 pt_curr = pt_prev; 5456 od_curr = orig_dev; 5457 } 5458 list_add_tail(&skb->list, &sublist); 5459 } 5460 5461 /* dispatch final sublist */ 5462 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5463 } 5464 5465 static int __netif_receive_skb(struct sk_buff *skb) 5466 { 5467 int ret; 5468 5469 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { 5470 unsigned int noreclaim_flag; 5471 5472 /* 5473 * PFMEMALLOC skbs are special, they should 5474 * - be delivered to SOCK_MEMALLOC sockets only 5475 * - stay away from userspace 5476 * - have bounded memory usage 5477 * 5478 * Use PF_MEMALLOC as this saves us from propagating the allocation 5479 * context down to all allocation sites. 5480 */ 5481 noreclaim_flag = memalloc_noreclaim_save(); 5482 ret = __netif_receive_skb_one_core(skb, true); 5483 memalloc_noreclaim_restore(noreclaim_flag); 5484 } else 5485 ret = __netif_receive_skb_one_core(skb, false); 5486 5487 return ret; 5488 } 5489 5490 static void __netif_receive_skb_list(struct list_head *head) 5491 { 5492 unsigned long noreclaim_flag = 0; 5493 struct sk_buff *skb, *next; 5494 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */ 5495 5496 list_for_each_entry_safe(skb, next, head, list) { 5497 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) { 5498 struct list_head sublist; 5499 5500 /* Handle the previous sublist */ 5501 list_cut_before(&sublist, head, &skb->list); 5502 if (!list_empty(&sublist)) 5503 __netif_receive_skb_list_core(&sublist, pfmemalloc); 5504 pfmemalloc = !pfmemalloc; 5505 /* See comments in __netif_receive_skb */ 5506 if (pfmemalloc) 5507 noreclaim_flag = memalloc_noreclaim_save(); 5508 else 5509 memalloc_noreclaim_restore(noreclaim_flag); 5510 } 5511 } 5512 /* Handle the remaining sublist */ 5513 if (!list_empty(head)) 5514 __netif_receive_skb_list_core(head, pfmemalloc); 5515 /* Restore pflags */ 5516 if (pfmemalloc) 5517 memalloc_noreclaim_restore(noreclaim_flag); 5518 } 5519 5520 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp) 5521 { 5522 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog); 5523 struct bpf_prog *new = xdp->prog; 5524 int ret = 0; 5525 5526 if (new) { 5527 u32 i; 5528 5529 mutex_lock(&new->aux->used_maps_mutex); 5530 5531 /* generic XDP does not work with DEVMAPs that can 5532 * have a bpf_prog installed on an entry 5533 */ 5534 for (i = 0; i < new->aux->used_map_cnt; i++) { 5535 if (dev_map_can_have_prog(new->aux->used_maps[i]) || 5536 cpu_map_prog_allowed(new->aux->used_maps[i])) { 5537 mutex_unlock(&new->aux->used_maps_mutex); 5538 return -EINVAL; 5539 } 5540 } 5541 5542 mutex_unlock(&new->aux->used_maps_mutex); 5543 } 5544 5545 switch (xdp->command) { 5546 case XDP_SETUP_PROG: 5547 rcu_assign_pointer(dev->xdp_prog, new); 5548 if (old) 5549 bpf_prog_put(old); 5550 5551 if (old && !new) { 5552 static_branch_dec(&generic_xdp_needed_key); 5553 } else if (new && !old) { 5554 static_branch_inc(&generic_xdp_needed_key); 5555 dev_disable_lro(dev); 5556 dev_disable_gro_hw(dev); 5557 } 5558 break; 5559 5560 default: 5561 ret = -EINVAL; 5562 break; 5563 } 5564 5565 return ret; 5566 } 5567 5568 static int netif_receive_skb_internal(struct sk_buff *skb) 5569 { 5570 int ret; 5571 5572 net_timestamp_check(netdev_tstamp_prequeue, skb); 5573 5574 if (skb_defer_rx_timestamp(skb)) 5575 return NET_RX_SUCCESS; 5576 5577 rcu_read_lock(); 5578 #ifdef CONFIG_RPS 5579 if (static_branch_unlikely(&rps_needed)) { 5580 struct rps_dev_flow voidflow, *rflow = &voidflow; 5581 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5582 5583 if (cpu >= 0) { 5584 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5585 rcu_read_unlock(); 5586 return ret; 5587 } 5588 } 5589 #endif 5590 ret = __netif_receive_skb(skb); 5591 rcu_read_unlock(); 5592 return ret; 5593 } 5594 5595 static void netif_receive_skb_list_internal(struct list_head *head) 5596 { 5597 struct sk_buff *skb, *next; 5598 struct list_head sublist; 5599 5600 INIT_LIST_HEAD(&sublist); 5601 list_for_each_entry_safe(skb, next, head, list) { 5602 net_timestamp_check(netdev_tstamp_prequeue, skb); 5603 skb_list_del_init(skb); 5604 if (!skb_defer_rx_timestamp(skb)) 5605 list_add_tail(&skb->list, &sublist); 5606 } 5607 list_splice_init(&sublist, head); 5608 5609 rcu_read_lock(); 5610 #ifdef CONFIG_RPS 5611 if (static_branch_unlikely(&rps_needed)) { 5612 list_for_each_entry_safe(skb, next, head, list) { 5613 struct rps_dev_flow voidflow, *rflow = &voidflow; 5614 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5615 5616 if (cpu >= 0) { 5617 /* Will be handled, remove from list */ 5618 skb_list_del_init(skb); 5619 enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5620 } 5621 } 5622 } 5623 #endif 5624 __netif_receive_skb_list(head); 5625 rcu_read_unlock(); 5626 } 5627 5628 /** 5629 * netif_receive_skb - process receive buffer from network 5630 * @skb: buffer to process 5631 * 5632 * netif_receive_skb() is the main receive data processing function. 5633 * It always succeeds. The buffer may be dropped during processing 5634 * for congestion control or by the protocol layers. 5635 * 5636 * This function may only be called from softirq context and interrupts 5637 * should be enabled. 5638 * 5639 * Return values (usually ignored): 5640 * NET_RX_SUCCESS: no congestion 5641 * NET_RX_DROP: packet was dropped 5642 */ 5643 int netif_receive_skb(struct sk_buff *skb) 5644 { 5645 int ret; 5646 5647 trace_netif_receive_skb_entry(skb); 5648 5649 ret = netif_receive_skb_internal(skb); 5650 trace_netif_receive_skb_exit(ret); 5651 5652 return ret; 5653 } 5654 EXPORT_SYMBOL(netif_receive_skb); 5655 5656 /** 5657 * netif_receive_skb_list - process many receive buffers from network 5658 * @head: list of skbs to process. 5659 * 5660 * Since return value of netif_receive_skb() is normally ignored, and 5661 * wouldn't be meaningful for a list, this function returns void. 5662 * 5663 * This function may only be called from softirq context and interrupts 5664 * should be enabled. 5665 */ 5666 void netif_receive_skb_list(struct list_head *head) 5667 { 5668 struct sk_buff *skb; 5669 5670 if (list_empty(head)) 5671 return; 5672 if (trace_netif_receive_skb_list_entry_enabled()) { 5673 list_for_each_entry(skb, head, list) 5674 trace_netif_receive_skb_list_entry(skb); 5675 } 5676 netif_receive_skb_list_internal(head); 5677 trace_netif_receive_skb_list_exit(0); 5678 } 5679 EXPORT_SYMBOL(netif_receive_skb_list); 5680 5681 static DEFINE_PER_CPU(struct work_struct, flush_works); 5682 5683 /* Network device is going away, flush any packets still pending */ 5684 static void flush_backlog(struct work_struct *work) 5685 { 5686 struct sk_buff *skb, *tmp; 5687 struct softnet_data *sd; 5688 5689 local_bh_disable(); 5690 sd = this_cpu_ptr(&softnet_data); 5691 5692 local_irq_disable(); 5693 rps_lock(sd); 5694 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { 5695 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5696 __skb_unlink(skb, &sd->input_pkt_queue); 5697 dev_kfree_skb_irq(skb); 5698 input_queue_head_incr(sd); 5699 } 5700 } 5701 rps_unlock(sd); 5702 local_irq_enable(); 5703 5704 skb_queue_walk_safe(&sd->process_queue, skb, tmp) { 5705 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5706 __skb_unlink(skb, &sd->process_queue); 5707 kfree_skb(skb); 5708 input_queue_head_incr(sd); 5709 } 5710 } 5711 local_bh_enable(); 5712 } 5713 5714 static bool flush_required(int cpu) 5715 { 5716 #if IS_ENABLED(CONFIG_RPS) 5717 struct softnet_data *sd = &per_cpu(softnet_data, cpu); 5718 bool do_flush; 5719 5720 local_irq_disable(); 5721 rps_lock(sd); 5722 5723 /* as insertion into process_queue happens with the rps lock held, 5724 * process_queue access may race only with dequeue 5725 */ 5726 do_flush = !skb_queue_empty(&sd->input_pkt_queue) || 5727 !skb_queue_empty_lockless(&sd->process_queue); 5728 rps_unlock(sd); 5729 local_irq_enable(); 5730 5731 return do_flush; 5732 #endif 5733 /* without RPS we can't safely check input_pkt_queue: during a 5734 * concurrent remote skb_queue_splice() we can detect as empty both 5735 * input_pkt_queue and process_queue even if the latter could end-up 5736 * containing a lot of packets. 5737 */ 5738 return true; 5739 } 5740 5741 static void flush_all_backlogs(void) 5742 { 5743 static cpumask_t flush_cpus; 5744 unsigned int cpu; 5745 5746 /* since we are under rtnl lock protection we can use static data 5747 * for the cpumask and avoid allocating on stack the possibly 5748 * large mask 5749 */ 5750 ASSERT_RTNL(); 5751 5752 get_online_cpus(); 5753 5754 cpumask_clear(&flush_cpus); 5755 for_each_online_cpu(cpu) { 5756 if (flush_required(cpu)) { 5757 queue_work_on(cpu, system_highpri_wq, 5758 per_cpu_ptr(&flush_works, cpu)); 5759 cpumask_set_cpu(cpu, &flush_cpus); 5760 } 5761 } 5762 5763 /* we can have in flight packet[s] on the cpus we are not flushing, 5764 * synchronize_net() in unregister_netdevice_many() will take care of 5765 * them 5766 */ 5767 for_each_cpu(cpu, &flush_cpus) 5768 flush_work(per_cpu_ptr(&flush_works, cpu)); 5769 5770 put_online_cpus(); 5771 } 5772 5773 /* Pass the currently batched GRO_NORMAL SKBs up to the stack. */ 5774 static void gro_normal_list(struct napi_struct *napi) 5775 { 5776 if (!napi->rx_count) 5777 return; 5778 netif_receive_skb_list_internal(&napi->rx_list); 5779 INIT_LIST_HEAD(&napi->rx_list); 5780 napi->rx_count = 0; 5781 } 5782 5783 /* Queue one GRO_NORMAL SKB up for list processing. If batch size exceeded, 5784 * pass the whole batch up to the stack. 5785 */ 5786 static void gro_normal_one(struct napi_struct *napi, struct sk_buff *skb, int segs) 5787 { 5788 list_add_tail(&skb->list, &napi->rx_list); 5789 napi->rx_count += segs; 5790 if (napi->rx_count >= gro_normal_batch) 5791 gro_normal_list(napi); 5792 } 5793 5794 static int napi_gro_complete(struct napi_struct *napi, struct sk_buff *skb) 5795 { 5796 struct packet_offload *ptype; 5797 __be16 type = skb->protocol; 5798 struct list_head *head = &offload_base; 5799 int err = -ENOENT; 5800 5801 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb)); 5802 5803 if (NAPI_GRO_CB(skb)->count == 1) { 5804 skb_shinfo(skb)->gso_size = 0; 5805 goto out; 5806 } 5807 5808 rcu_read_lock(); 5809 list_for_each_entry_rcu(ptype, head, list) { 5810 if (ptype->type != type || !ptype->callbacks.gro_complete) 5811 continue; 5812 5813 err = INDIRECT_CALL_INET(ptype->callbacks.gro_complete, 5814 ipv6_gro_complete, inet_gro_complete, 5815 skb, 0); 5816 break; 5817 } 5818 rcu_read_unlock(); 5819 5820 if (err) { 5821 WARN_ON(&ptype->list == head); 5822 kfree_skb(skb); 5823 return NET_RX_SUCCESS; 5824 } 5825 5826 out: 5827 gro_normal_one(napi, skb, NAPI_GRO_CB(skb)->count); 5828 return NET_RX_SUCCESS; 5829 } 5830 5831 static void __napi_gro_flush_chain(struct napi_struct *napi, u32 index, 5832 bool flush_old) 5833 { 5834 struct list_head *head = &napi->gro_hash[index].list; 5835 struct sk_buff *skb, *p; 5836 5837 list_for_each_entry_safe_reverse(skb, p, head, list) { 5838 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies) 5839 return; 5840 skb_list_del_init(skb); 5841 napi_gro_complete(napi, skb); 5842 napi->gro_hash[index].count--; 5843 } 5844 5845 if (!napi->gro_hash[index].count) 5846 __clear_bit(index, &napi->gro_bitmask); 5847 } 5848 5849 /* napi->gro_hash[].list contains packets ordered by age. 5850 * youngest packets at the head of it. 5851 * Complete skbs in reverse order to reduce latencies. 5852 */ 5853 void napi_gro_flush(struct napi_struct *napi, bool flush_old) 5854 { 5855 unsigned long bitmask = napi->gro_bitmask; 5856 unsigned int i, base = ~0U; 5857 5858 while ((i = ffs(bitmask)) != 0) { 5859 bitmask >>= i; 5860 base += i; 5861 __napi_gro_flush_chain(napi, base, flush_old); 5862 } 5863 } 5864 EXPORT_SYMBOL(napi_gro_flush); 5865 5866 static void gro_list_prepare(const struct list_head *head, 5867 const struct sk_buff *skb) 5868 { 5869 unsigned int maclen = skb->dev->hard_header_len; 5870 u32 hash = skb_get_hash_raw(skb); 5871 struct sk_buff *p; 5872 5873 list_for_each_entry(p, head, list) { 5874 unsigned long diffs; 5875 5876 NAPI_GRO_CB(p)->flush = 0; 5877 5878 if (hash != skb_get_hash_raw(p)) { 5879 NAPI_GRO_CB(p)->same_flow = 0; 5880 continue; 5881 } 5882 5883 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev; 5884 diffs |= skb_vlan_tag_present(p) ^ skb_vlan_tag_present(skb); 5885 if (skb_vlan_tag_present(p)) 5886 diffs |= skb_vlan_tag_get(p) ^ skb_vlan_tag_get(skb); 5887 diffs |= skb_metadata_dst_cmp(p, skb); 5888 diffs |= skb_metadata_differs(p, skb); 5889 if (maclen == ETH_HLEN) 5890 diffs |= compare_ether_header(skb_mac_header(p), 5891 skb_mac_header(skb)); 5892 else if (!diffs) 5893 diffs = memcmp(skb_mac_header(p), 5894 skb_mac_header(skb), 5895 maclen); 5896 NAPI_GRO_CB(p)->same_flow = !diffs; 5897 } 5898 } 5899 5900 static void skb_gro_reset_offset(struct sk_buff *skb) 5901 { 5902 const struct skb_shared_info *pinfo = skb_shinfo(skb); 5903 const skb_frag_t *frag0 = &pinfo->frags[0]; 5904 5905 NAPI_GRO_CB(skb)->data_offset = 0; 5906 NAPI_GRO_CB(skb)->frag0 = NULL; 5907 NAPI_GRO_CB(skb)->frag0_len = 0; 5908 5909 if (!skb_headlen(skb) && pinfo->nr_frags && 5910 !PageHighMem(skb_frag_page(frag0))) { 5911 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0); 5912 NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int, 5913 skb_frag_size(frag0), 5914 skb->end - skb->tail); 5915 } 5916 } 5917 5918 static void gro_pull_from_frag0(struct sk_buff *skb, int grow) 5919 { 5920 struct skb_shared_info *pinfo = skb_shinfo(skb); 5921 5922 BUG_ON(skb->end - skb->tail < grow); 5923 5924 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); 5925 5926 skb->data_len -= grow; 5927 skb->tail += grow; 5928 5929 skb_frag_off_add(&pinfo->frags[0], grow); 5930 skb_frag_size_sub(&pinfo->frags[0], grow); 5931 5932 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) { 5933 skb_frag_unref(skb, 0); 5934 memmove(pinfo->frags, pinfo->frags + 1, 5935 --pinfo->nr_frags * sizeof(pinfo->frags[0])); 5936 } 5937 } 5938 5939 static void gro_flush_oldest(struct napi_struct *napi, struct list_head *head) 5940 { 5941 struct sk_buff *oldest; 5942 5943 oldest = list_last_entry(head, struct sk_buff, list); 5944 5945 /* We are called with head length >= MAX_GRO_SKBS, so this is 5946 * impossible. 5947 */ 5948 if (WARN_ON_ONCE(!oldest)) 5949 return; 5950 5951 /* Do not adjust napi->gro_hash[].count, caller is adding a new 5952 * SKB to the chain. 5953 */ 5954 skb_list_del_init(oldest); 5955 napi_gro_complete(napi, oldest); 5956 } 5957 5958 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) 5959 { 5960 u32 bucket = skb_get_hash_raw(skb) & (GRO_HASH_BUCKETS - 1); 5961 struct gro_list *gro_list = &napi->gro_hash[bucket]; 5962 struct list_head *head = &offload_base; 5963 struct packet_offload *ptype; 5964 __be16 type = skb->protocol; 5965 struct sk_buff *pp = NULL; 5966 enum gro_result ret; 5967 int same_flow; 5968 int grow; 5969 5970 if (netif_elide_gro(skb->dev)) 5971 goto normal; 5972 5973 gro_list_prepare(&gro_list->list, skb); 5974 5975 rcu_read_lock(); 5976 list_for_each_entry_rcu(ptype, head, list) { 5977 if (ptype->type != type || !ptype->callbacks.gro_receive) 5978 continue; 5979 5980 skb_set_network_header(skb, skb_gro_offset(skb)); 5981 skb_reset_mac_len(skb); 5982 NAPI_GRO_CB(skb)->same_flow = 0; 5983 NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb); 5984 NAPI_GRO_CB(skb)->free = 0; 5985 NAPI_GRO_CB(skb)->encap_mark = 0; 5986 NAPI_GRO_CB(skb)->recursion_counter = 0; 5987 NAPI_GRO_CB(skb)->is_fou = 0; 5988 NAPI_GRO_CB(skb)->is_atomic = 1; 5989 NAPI_GRO_CB(skb)->gro_remcsum_start = 0; 5990 5991 /* Setup for GRO checksum validation */ 5992 switch (skb->ip_summed) { 5993 case CHECKSUM_COMPLETE: 5994 NAPI_GRO_CB(skb)->csum = skb->csum; 5995 NAPI_GRO_CB(skb)->csum_valid = 1; 5996 NAPI_GRO_CB(skb)->csum_cnt = 0; 5997 break; 5998 case CHECKSUM_UNNECESSARY: 5999 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1; 6000 NAPI_GRO_CB(skb)->csum_valid = 0; 6001 break; 6002 default: 6003 NAPI_GRO_CB(skb)->csum_cnt = 0; 6004 NAPI_GRO_CB(skb)->csum_valid = 0; 6005 } 6006 6007 pp = INDIRECT_CALL_INET(ptype->callbacks.gro_receive, 6008 ipv6_gro_receive, inet_gro_receive, 6009 &gro_list->list, skb); 6010 break; 6011 } 6012 rcu_read_unlock(); 6013 6014 if (&ptype->list == head) 6015 goto normal; 6016 6017 if (PTR_ERR(pp) == -EINPROGRESS) { 6018 ret = GRO_CONSUMED; 6019 goto ok; 6020 } 6021 6022 same_flow = NAPI_GRO_CB(skb)->same_flow; 6023 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; 6024 6025 if (pp) { 6026 skb_list_del_init(pp); 6027 napi_gro_complete(napi, pp); 6028 gro_list->count--; 6029 } 6030 6031 if (same_flow) 6032 goto ok; 6033 6034 if (NAPI_GRO_CB(skb)->flush) 6035 goto normal; 6036 6037 if (unlikely(gro_list->count >= MAX_GRO_SKBS)) 6038 gro_flush_oldest(napi, &gro_list->list); 6039 else 6040 gro_list->count++; 6041 6042 NAPI_GRO_CB(skb)->count = 1; 6043 NAPI_GRO_CB(skb)->age = jiffies; 6044 NAPI_GRO_CB(skb)->last = skb; 6045 skb_shinfo(skb)->gso_size = skb_gro_len(skb); 6046 list_add(&skb->list, &gro_list->list); 6047 ret = GRO_HELD; 6048 6049 pull: 6050 grow = skb_gro_offset(skb) - skb_headlen(skb); 6051 if (grow > 0) 6052 gro_pull_from_frag0(skb, grow); 6053 ok: 6054 if (gro_list->count) { 6055 if (!test_bit(bucket, &napi->gro_bitmask)) 6056 __set_bit(bucket, &napi->gro_bitmask); 6057 } else if (test_bit(bucket, &napi->gro_bitmask)) { 6058 __clear_bit(bucket, &napi->gro_bitmask); 6059 } 6060 6061 return ret; 6062 6063 normal: 6064 ret = GRO_NORMAL; 6065 goto pull; 6066 } 6067 6068 struct packet_offload *gro_find_receive_by_type(__be16 type) 6069 { 6070 struct list_head *offload_head = &offload_base; 6071 struct packet_offload *ptype; 6072 6073 list_for_each_entry_rcu(ptype, offload_head, list) { 6074 if (ptype->type != type || !ptype->callbacks.gro_receive) 6075 continue; 6076 return ptype; 6077 } 6078 return NULL; 6079 } 6080 EXPORT_SYMBOL(gro_find_receive_by_type); 6081 6082 struct packet_offload *gro_find_complete_by_type(__be16 type) 6083 { 6084 struct list_head *offload_head = &offload_base; 6085 struct packet_offload *ptype; 6086 6087 list_for_each_entry_rcu(ptype, offload_head, list) { 6088 if (ptype->type != type || !ptype->callbacks.gro_complete) 6089 continue; 6090 return ptype; 6091 } 6092 return NULL; 6093 } 6094 EXPORT_SYMBOL(gro_find_complete_by_type); 6095 6096 static gro_result_t napi_skb_finish(struct napi_struct *napi, 6097 struct sk_buff *skb, 6098 gro_result_t ret) 6099 { 6100 switch (ret) { 6101 case GRO_NORMAL: 6102 gro_normal_one(napi, skb, 1); 6103 break; 6104 6105 case GRO_MERGED_FREE: 6106 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) 6107 napi_skb_free_stolen_head(skb); 6108 else 6109 __kfree_skb_defer(skb); 6110 break; 6111 6112 case GRO_HELD: 6113 case GRO_MERGED: 6114 case GRO_CONSUMED: 6115 break; 6116 } 6117 6118 return ret; 6119 } 6120 6121 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) 6122 { 6123 gro_result_t ret; 6124 6125 skb_mark_napi_id(skb, napi); 6126 trace_napi_gro_receive_entry(skb); 6127 6128 skb_gro_reset_offset(skb); 6129 6130 ret = napi_skb_finish(napi, skb, dev_gro_receive(napi, skb)); 6131 trace_napi_gro_receive_exit(ret); 6132 6133 return ret; 6134 } 6135 EXPORT_SYMBOL(napi_gro_receive); 6136 6137 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) 6138 { 6139 if (unlikely(skb->pfmemalloc)) { 6140 consume_skb(skb); 6141 return; 6142 } 6143 __skb_pull(skb, skb_headlen(skb)); 6144 /* restore the reserve we had after netdev_alloc_skb_ip_align() */ 6145 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb)); 6146 __vlan_hwaccel_clear_tag(skb); 6147 skb->dev = napi->dev; 6148 skb->skb_iif = 0; 6149 6150 /* eth_type_trans() assumes pkt_type is PACKET_HOST */ 6151 skb->pkt_type = PACKET_HOST; 6152 6153 skb->encapsulation = 0; 6154 skb_shinfo(skb)->gso_type = 0; 6155 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); 6156 skb_ext_reset(skb); 6157 6158 napi->skb = skb; 6159 } 6160 6161 struct sk_buff *napi_get_frags(struct napi_struct *napi) 6162 { 6163 struct sk_buff *skb = napi->skb; 6164 6165 if (!skb) { 6166 skb = napi_alloc_skb(napi, GRO_MAX_HEAD); 6167 if (skb) { 6168 napi->skb = skb; 6169 skb_mark_napi_id(skb, napi); 6170 } 6171 } 6172 return skb; 6173 } 6174 EXPORT_SYMBOL(napi_get_frags); 6175 6176 static gro_result_t napi_frags_finish(struct napi_struct *napi, 6177 struct sk_buff *skb, 6178 gro_result_t ret) 6179 { 6180 switch (ret) { 6181 case GRO_NORMAL: 6182 case GRO_HELD: 6183 __skb_push(skb, ETH_HLEN); 6184 skb->protocol = eth_type_trans(skb, skb->dev); 6185 if (ret == GRO_NORMAL) 6186 gro_normal_one(napi, skb, 1); 6187 break; 6188 6189 case GRO_MERGED_FREE: 6190 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) 6191 napi_skb_free_stolen_head(skb); 6192 else 6193 napi_reuse_skb(napi, skb); 6194 break; 6195 6196 case GRO_MERGED: 6197 case GRO_CONSUMED: 6198 break; 6199 } 6200 6201 return ret; 6202 } 6203 6204 /* Upper GRO stack assumes network header starts at gro_offset=0 6205 * Drivers could call both napi_gro_frags() and napi_gro_receive() 6206 * We copy ethernet header into skb->data to have a common layout. 6207 */ 6208 static struct sk_buff *napi_frags_skb(struct napi_struct *napi) 6209 { 6210 struct sk_buff *skb = napi->skb; 6211 const struct ethhdr *eth; 6212 unsigned int hlen = sizeof(*eth); 6213 6214 napi->skb = NULL; 6215 6216 skb_reset_mac_header(skb); 6217 skb_gro_reset_offset(skb); 6218 6219 if (unlikely(skb_gro_header_hard(skb, hlen))) { 6220 eth = skb_gro_header_slow(skb, hlen, 0); 6221 if (unlikely(!eth)) { 6222 net_warn_ratelimited("%s: dropping impossible skb from %s\n", 6223 __func__, napi->dev->name); 6224 napi_reuse_skb(napi, skb); 6225 return NULL; 6226 } 6227 } else { 6228 eth = (const struct ethhdr *)skb->data; 6229 gro_pull_from_frag0(skb, hlen); 6230 NAPI_GRO_CB(skb)->frag0 += hlen; 6231 NAPI_GRO_CB(skb)->frag0_len -= hlen; 6232 } 6233 __skb_pull(skb, hlen); 6234 6235 /* 6236 * This works because the only protocols we care about don't require 6237 * special handling. 6238 * We'll fix it up properly in napi_frags_finish() 6239 */ 6240 skb->protocol = eth->h_proto; 6241 6242 return skb; 6243 } 6244 6245 gro_result_t napi_gro_frags(struct napi_struct *napi) 6246 { 6247 gro_result_t ret; 6248 struct sk_buff *skb = napi_frags_skb(napi); 6249 6250 trace_napi_gro_frags_entry(skb); 6251 6252 ret = napi_frags_finish(napi, skb, dev_gro_receive(napi, skb)); 6253 trace_napi_gro_frags_exit(ret); 6254 6255 return ret; 6256 } 6257 EXPORT_SYMBOL(napi_gro_frags); 6258 6259 /* Compute the checksum from gro_offset and return the folded value 6260 * after adding in any pseudo checksum. 6261 */ 6262 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb) 6263 { 6264 __wsum wsum; 6265 __sum16 sum; 6266 6267 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0); 6268 6269 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */ 6270 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum)); 6271 /* See comments in __skb_checksum_complete(). */ 6272 if (likely(!sum)) { 6273 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && 6274 !skb->csum_complete_sw) 6275 netdev_rx_csum_fault(skb->dev, skb); 6276 } 6277 6278 NAPI_GRO_CB(skb)->csum = wsum; 6279 NAPI_GRO_CB(skb)->csum_valid = 1; 6280 6281 return sum; 6282 } 6283 EXPORT_SYMBOL(__skb_gro_checksum_complete); 6284 6285 static void net_rps_send_ipi(struct softnet_data *remsd) 6286 { 6287 #ifdef CONFIG_RPS 6288 while (remsd) { 6289 struct softnet_data *next = remsd->rps_ipi_next; 6290 6291 if (cpu_online(remsd->cpu)) 6292 smp_call_function_single_async(remsd->cpu, &remsd->csd); 6293 remsd = next; 6294 } 6295 #endif 6296 } 6297 6298 /* 6299 * net_rps_action_and_irq_enable sends any pending IPI's for rps. 6300 * Note: called with local irq disabled, but exits with local irq enabled. 6301 */ 6302 static void net_rps_action_and_irq_enable(struct softnet_data *sd) 6303 { 6304 #ifdef CONFIG_RPS 6305 struct softnet_data *remsd = sd->rps_ipi_list; 6306 6307 if (remsd) { 6308 sd->rps_ipi_list = NULL; 6309 6310 local_irq_enable(); 6311 6312 /* Send pending IPI's to kick RPS processing on remote cpus. */ 6313 net_rps_send_ipi(remsd); 6314 } else 6315 #endif 6316 local_irq_enable(); 6317 } 6318 6319 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) 6320 { 6321 #ifdef CONFIG_RPS 6322 return sd->rps_ipi_list != NULL; 6323 #else 6324 return false; 6325 #endif 6326 } 6327 6328 static int process_backlog(struct napi_struct *napi, int quota) 6329 { 6330 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); 6331 bool again = true; 6332 int work = 0; 6333 6334 /* Check if we have pending ipi, its better to send them now, 6335 * not waiting net_rx_action() end. 6336 */ 6337 if (sd_has_rps_ipi_waiting(sd)) { 6338 local_irq_disable(); 6339 net_rps_action_and_irq_enable(sd); 6340 } 6341 6342 napi->weight = dev_rx_weight; 6343 while (again) { 6344 struct sk_buff *skb; 6345 6346 while ((skb = __skb_dequeue(&sd->process_queue))) { 6347 rcu_read_lock(); 6348 __netif_receive_skb(skb); 6349 rcu_read_unlock(); 6350 input_queue_head_incr(sd); 6351 if (++work >= quota) 6352 return work; 6353 6354 } 6355 6356 local_irq_disable(); 6357 rps_lock(sd); 6358 if (skb_queue_empty(&sd->input_pkt_queue)) { 6359 /* 6360 * Inline a custom version of __napi_complete(). 6361 * only current cpu owns and manipulates this napi, 6362 * and NAPI_STATE_SCHED is the only possible flag set 6363 * on backlog. 6364 * We can use a plain write instead of clear_bit(), 6365 * and we dont need an smp_mb() memory barrier. 6366 */ 6367 napi->state = 0; 6368 again = false; 6369 } else { 6370 skb_queue_splice_tail_init(&sd->input_pkt_queue, 6371 &sd->process_queue); 6372 } 6373 rps_unlock(sd); 6374 local_irq_enable(); 6375 } 6376 6377 return work; 6378 } 6379 6380 /** 6381 * __napi_schedule - schedule for receive 6382 * @n: entry to schedule 6383 * 6384 * The entry's receive function will be scheduled to run. 6385 * Consider using __napi_schedule_irqoff() if hard irqs are masked. 6386 */ 6387 void __napi_schedule(struct napi_struct *n) 6388 { 6389 unsigned long flags; 6390 6391 local_irq_save(flags); 6392 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6393 local_irq_restore(flags); 6394 } 6395 EXPORT_SYMBOL(__napi_schedule); 6396 6397 /** 6398 * napi_schedule_prep - check if napi can be scheduled 6399 * @n: napi context 6400 * 6401 * Test if NAPI routine is already running, and if not mark 6402 * it as running. This is used as a condition variable to 6403 * insure only one NAPI poll instance runs. We also make 6404 * sure there is no pending NAPI disable. 6405 */ 6406 bool napi_schedule_prep(struct napi_struct *n) 6407 { 6408 unsigned long val, new; 6409 6410 do { 6411 val = READ_ONCE(n->state); 6412 if (unlikely(val & NAPIF_STATE_DISABLE)) 6413 return false; 6414 new = val | NAPIF_STATE_SCHED; 6415 6416 /* Sets STATE_MISSED bit if STATE_SCHED was already set 6417 * This was suggested by Alexander Duyck, as compiler 6418 * emits better code than : 6419 * if (val & NAPIF_STATE_SCHED) 6420 * new |= NAPIF_STATE_MISSED; 6421 */ 6422 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * 6423 NAPIF_STATE_MISSED; 6424 } while (cmpxchg(&n->state, val, new) != val); 6425 6426 return !(val & NAPIF_STATE_SCHED); 6427 } 6428 EXPORT_SYMBOL(napi_schedule_prep); 6429 6430 /** 6431 * __napi_schedule_irqoff - schedule for receive 6432 * @n: entry to schedule 6433 * 6434 * Variant of __napi_schedule() assuming hard irqs are masked 6435 */ 6436 void __napi_schedule_irqoff(struct napi_struct *n) 6437 { 6438 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6439 } 6440 EXPORT_SYMBOL(__napi_schedule_irqoff); 6441 6442 bool napi_complete_done(struct napi_struct *n, int work_done) 6443 { 6444 unsigned long flags, val, new, timeout = 0; 6445 bool ret = true; 6446 6447 /* 6448 * 1) Don't let napi dequeue from the cpu poll list 6449 * just in case its running on a different cpu. 6450 * 2) If we are busy polling, do nothing here, we have 6451 * the guarantee we will be called later. 6452 */ 6453 if (unlikely(n->state & (NAPIF_STATE_NPSVC | 6454 NAPIF_STATE_IN_BUSY_POLL))) 6455 return false; 6456 6457 if (work_done) { 6458 if (n->gro_bitmask) 6459 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6460 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs); 6461 } 6462 if (n->defer_hard_irqs_count > 0) { 6463 n->defer_hard_irqs_count--; 6464 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6465 if (timeout) 6466 ret = false; 6467 } 6468 if (n->gro_bitmask) { 6469 /* When the NAPI instance uses a timeout and keeps postponing 6470 * it, we need to bound somehow the time packets are kept in 6471 * the GRO layer 6472 */ 6473 napi_gro_flush(n, !!timeout); 6474 } 6475 6476 gro_normal_list(n); 6477 6478 if (unlikely(!list_empty(&n->poll_list))) { 6479 /* If n->poll_list is not empty, we need to mask irqs */ 6480 local_irq_save(flags); 6481 list_del_init(&n->poll_list); 6482 local_irq_restore(flags); 6483 } 6484 6485 do { 6486 val = READ_ONCE(n->state); 6487 6488 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); 6489 6490 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED | 6491 NAPIF_STATE_PREFER_BUSY_POLL); 6492 6493 /* If STATE_MISSED was set, leave STATE_SCHED set, 6494 * because we will call napi->poll() one more time. 6495 * This C code was suggested by Alexander Duyck to help gcc. 6496 */ 6497 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * 6498 NAPIF_STATE_SCHED; 6499 } while (cmpxchg(&n->state, val, new) != val); 6500 6501 if (unlikely(val & NAPIF_STATE_MISSED)) { 6502 __napi_schedule(n); 6503 return false; 6504 } 6505 6506 if (timeout) 6507 hrtimer_start(&n->timer, ns_to_ktime(timeout), 6508 HRTIMER_MODE_REL_PINNED); 6509 return ret; 6510 } 6511 EXPORT_SYMBOL(napi_complete_done); 6512 6513 /* must be called under rcu_read_lock(), as we dont take a reference */ 6514 static struct napi_struct *napi_by_id(unsigned int napi_id) 6515 { 6516 unsigned int hash = napi_id % HASH_SIZE(napi_hash); 6517 struct napi_struct *napi; 6518 6519 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) 6520 if (napi->napi_id == napi_id) 6521 return napi; 6522 6523 return NULL; 6524 } 6525 6526 #if defined(CONFIG_NET_RX_BUSY_POLL) 6527 6528 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule) 6529 { 6530 if (!skip_schedule) { 6531 gro_normal_list(napi); 6532 __napi_schedule(napi); 6533 return; 6534 } 6535 6536 if (napi->gro_bitmask) { 6537 /* flush too old packets 6538 * If HZ < 1000, flush all packets. 6539 */ 6540 napi_gro_flush(napi, HZ >= 1000); 6541 } 6542 6543 gro_normal_list(napi); 6544 clear_bit(NAPI_STATE_SCHED, &napi->state); 6545 } 6546 6547 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll, 6548 u16 budget) 6549 { 6550 bool skip_schedule = false; 6551 unsigned long timeout; 6552 int rc; 6553 6554 /* Busy polling means there is a high chance device driver hard irq 6555 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was 6556 * set in napi_schedule_prep(). 6557 * Since we are about to call napi->poll() once more, we can safely 6558 * clear NAPI_STATE_MISSED. 6559 * 6560 * Note: x86 could use a single "lock and ..." instruction 6561 * to perform these two clear_bit() 6562 */ 6563 clear_bit(NAPI_STATE_MISSED, &napi->state); 6564 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); 6565 6566 local_bh_disable(); 6567 6568 if (prefer_busy_poll) { 6569 napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs); 6570 timeout = READ_ONCE(napi->dev->gro_flush_timeout); 6571 if (napi->defer_hard_irqs_count && timeout) { 6572 hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED); 6573 skip_schedule = true; 6574 } 6575 } 6576 6577 /* All we really want here is to re-enable device interrupts. 6578 * Ideally, a new ndo_busy_poll_stop() could avoid another round. 6579 */ 6580 rc = napi->poll(napi, budget); 6581 /* We can't gro_normal_list() here, because napi->poll() might have 6582 * rearmed the napi (napi_complete_done()) in which case it could 6583 * already be running on another CPU. 6584 */ 6585 trace_napi_poll(napi, rc, budget); 6586 netpoll_poll_unlock(have_poll_lock); 6587 if (rc == budget) 6588 __busy_poll_stop(napi, skip_schedule); 6589 local_bh_enable(); 6590 } 6591 6592 void napi_busy_loop(unsigned int napi_id, 6593 bool (*loop_end)(void *, unsigned long), 6594 void *loop_end_arg, bool prefer_busy_poll, u16 budget) 6595 { 6596 unsigned long start_time = loop_end ? busy_loop_current_time() : 0; 6597 int (*napi_poll)(struct napi_struct *napi, int budget); 6598 void *have_poll_lock = NULL; 6599 struct napi_struct *napi; 6600 6601 restart: 6602 napi_poll = NULL; 6603 6604 rcu_read_lock(); 6605 6606 napi = napi_by_id(napi_id); 6607 if (!napi) 6608 goto out; 6609 6610 preempt_disable(); 6611 for (;;) { 6612 int work = 0; 6613 6614 local_bh_disable(); 6615 if (!napi_poll) { 6616 unsigned long val = READ_ONCE(napi->state); 6617 6618 /* If multiple threads are competing for this napi, 6619 * we avoid dirtying napi->state as much as we can. 6620 */ 6621 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | 6622 NAPIF_STATE_IN_BUSY_POLL)) { 6623 if (prefer_busy_poll) 6624 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6625 goto count; 6626 } 6627 if (cmpxchg(&napi->state, val, 6628 val | NAPIF_STATE_IN_BUSY_POLL | 6629 NAPIF_STATE_SCHED) != val) { 6630 if (prefer_busy_poll) 6631 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6632 goto count; 6633 } 6634 have_poll_lock = netpoll_poll_lock(napi); 6635 napi_poll = napi->poll; 6636 } 6637 work = napi_poll(napi, budget); 6638 trace_napi_poll(napi, work, budget); 6639 gro_normal_list(napi); 6640 count: 6641 if (work > 0) 6642 __NET_ADD_STATS(dev_net(napi->dev), 6643 LINUX_MIB_BUSYPOLLRXPACKETS, work); 6644 local_bh_enable(); 6645 6646 if (!loop_end || loop_end(loop_end_arg, start_time)) 6647 break; 6648 6649 if (unlikely(need_resched())) { 6650 if (napi_poll) 6651 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6652 preempt_enable(); 6653 rcu_read_unlock(); 6654 cond_resched(); 6655 if (loop_end(loop_end_arg, start_time)) 6656 return; 6657 goto restart; 6658 } 6659 cpu_relax(); 6660 } 6661 if (napi_poll) 6662 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6663 preempt_enable(); 6664 out: 6665 rcu_read_unlock(); 6666 } 6667 EXPORT_SYMBOL(napi_busy_loop); 6668 6669 #endif /* CONFIG_NET_RX_BUSY_POLL */ 6670 6671 static void napi_hash_add(struct napi_struct *napi) 6672 { 6673 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state)) 6674 return; 6675 6676 spin_lock(&napi_hash_lock); 6677 6678 /* 0..NR_CPUS range is reserved for sender_cpu use */ 6679 do { 6680 if (unlikely(++napi_gen_id < MIN_NAPI_ID)) 6681 napi_gen_id = MIN_NAPI_ID; 6682 } while (napi_by_id(napi_gen_id)); 6683 napi->napi_id = napi_gen_id; 6684 6685 hlist_add_head_rcu(&napi->napi_hash_node, 6686 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); 6687 6688 spin_unlock(&napi_hash_lock); 6689 } 6690 6691 /* Warning : caller is responsible to make sure rcu grace period 6692 * is respected before freeing memory containing @napi 6693 */ 6694 static void napi_hash_del(struct napi_struct *napi) 6695 { 6696 spin_lock(&napi_hash_lock); 6697 6698 hlist_del_init_rcu(&napi->napi_hash_node); 6699 6700 spin_unlock(&napi_hash_lock); 6701 } 6702 6703 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) 6704 { 6705 struct napi_struct *napi; 6706 6707 napi = container_of(timer, struct napi_struct, timer); 6708 6709 /* Note : we use a relaxed variant of napi_schedule_prep() not setting 6710 * NAPI_STATE_MISSED, since we do not react to a device IRQ. 6711 */ 6712 if (!napi_disable_pending(napi) && 6713 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) { 6714 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6715 __napi_schedule_irqoff(napi); 6716 } 6717 6718 return HRTIMER_NORESTART; 6719 } 6720 6721 static void init_gro_hash(struct napi_struct *napi) 6722 { 6723 int i; 6724 6725 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6726 INIT_LIST_HEAD(&napi->gro_hash[i].list); 6727 napi->gro_hash[i].count = 0; 6728 } 6729 napi->gro_bitmask = 0; 6730 } 6731 6732 int dev_set_threaded(struct net_device *dev, bool threaded) 6733 { 6734 struct napi_struct *napi; 6735 int err = 0; 6736 6737 if (dev->threaded == threaded) 6738 return 0; 6739 6740 if (threaded) { 6741 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6742 if (!napi->thread) { 6743 err = napi_kthread_create(napi); 6744 if (err) { 6745 threaded = false; 6746 break; 6747 } 6748 } 6749 } 6750 } 6751 6752 dev->threaded = threaded; 6753 6754 /* Make sure kthread is created before THREADED bit 6755 * is set. 6756 */ 6757 smp_mb__before_atomic(); 6758 6759 /* Setting/unsetting threaded mode on a napi might not immediately 6760 * take effect, if the current napi instance is actively being 6761 * polled. In this case, the switch between threaded mode and 6762 * softirq mode will happen in the next round of napi_schedule(). 6763 * This should not cause hiccups/stalls to the live traffic. 6764 */ 6765 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6766 if (threaded) 6767 set_bit(NAPI_STATE_THREADED, &napi->state); 6768 else 6769 clear_bit(NAPI_STATE_THREADED, &napi->state); 6770 } 6771 6772 return err; 6773 } 6774 EXPORT_SYMBOL(dev_set_threaded); 6775 6776 void netif_napi_add(struct net_device *dev, struct napi_struct *napi, 6777 int (*poll)(struct napi_struct *, int), int weight) 6778 { 6779 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state))) 6780 return; 6781 6782 INIT_LIST_HEAD(&napi->poll_list); 6783 INIT_HLIST_NODE(&napi->napi_hash_node); 6784 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); 6785 napi->timer.function = napi_watchdog; 6786 init_gro_hash(napi); 6787 napi->skb = NULL; 6788 INIT_LIST_HEAD(&napi->rx_list); 6789 napi->rx_count = 0; 6790 napi->poll = poll; 6791 if (weight > NAPI_POLL_WEIGHT) 6792 netdev_err_once(dev, "%s() called with weight %d\n", __func__, 6793 weight); 6794 napi->weight = weight; 6795 napi->dev = dev; 6796 #ifdef CONFIG_NETPOLL 6797 napi->poll_owner = -1; 6798 #endif 6799 set_bit(NAPI_STATE_SCHED, &napi->state); 6800 set_bit(NAPI_STATE_NPSVC, &napi->state); 6801 list_add_rcu(&napi->dev_list, &dev->napi_list); 6802 napi_hash_add(napi); 6803 /* Create kthread for this napi if dev->threaded is set. 6804 * Clear dev->threaded if kthread creation failed so that 6805 * threaded mode will not be enabled in napi_enable(). 6806 */ 6807 if (dev->threaded && napi_kthread_create(napi)) 6808 dev->threaded = 0; 6809 } 6810 EXPORT_SYMBOL(netif_napi_add); 6811 6812 void napi_disable(struct napi_struct *n) 6813 { 6814 might_sleep(); 6815 set_bit(NAPI_STATE_DISABLE, &n->state); 6816 6817 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state)) 6818 msleep(1); 6819 while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state)) 6820 msleep(1); 6821 6822 hrtimer_cancel(&n->timer); 6823 6824 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &n->state); 6825 clear_bit(NAPI_STATE_DISABLE, &n->state); 6826 clear_bit(NAPI_STATE_THREADED, &n->state); 6827 } 6828 EXPORT_SYMBOL(napi_disable); 6829 6830 /** 6831 * napi_enable - enable NAPI scheduling 6832 * @n: NAPI context 6833 * 6834 * Resume NAPI from being scheduled on this context. 6835 * Must be paired with napi_disable. 6836 */ 6837 void napi_enable(struct napi_struct *n) 6838 { 6839 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state)); 6840 smp_mb__before_atomic(); 6841 clear_bit(NAPI_STATE_SCHED, &n->state); 6842 clear_bit(NAPI_STATE_NPSVC, &n->state); 6843 if (n->dev->threaded && n->thread) 6844 set_bit(NAPI_STATE_THREADED, &n->state); 6845 } 6846 EXPORT_SYMBOL(napi_enable); 6847 6848 static void flush_gro_hash(struct napi_struct *napi) 6849 { 6850 int i; 6851 6852 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6853 struct sk_buff *skb, *n; 6854 6855 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list) 6856 kfree_skb(skb); 6857 napi->gro_hash[i].count = 0; 6858 } 6859 } 6860 6861 /* Must be called in process context */ 6862 void __netif_napi_del(struct napi_struct *napi) 6863 { 6864 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state)) 6865 return; 6866 6867 napi_hash_del(napi); 6868 list_del_rcu(&napi->dev_list); 6869 napi_free_frags(napi); 6870 6871 flush_gro_hash(napi); 6872 napi->gro_bitmask = 0; 6873 6874 if (napi->thread) { 6875 kthread_stop(napi->thread); 6876 napi->thread = NULL; 6877 } 6878 } 6879 EXPORT_SYMBOL(__netif_napi_del); 6880 6881 static int __napi_poll(struct napi_struct *n, bool *repoll) 6882 { 6883 int work, weight; 6884 6885 weight = n->weight; 6886 6887 /* This NAPI_STATE_SCHED test is for avoiding a race 6888 * with netpoll's poll_napi(). Only the entity which 6889 * obtains the lock and sees NAPI_STATE_SCHED set will 6890 * actually make the ->poll() call. Therefore we avoid 6891 * accidentally calling ->poll() when NAPI is not scheduled. 6892 */ 6893 work = 0; 6894 if (test_bit(NAPI_STATE_SCHED, &n->state)) { 6895 work = n->poll(n, weight); 6896 trace_napi_poll(n, work, weight); 6897 } 6898 6899 if (unlikely(work > weight)) 6900 pr_err_once("NAPI poll function %pS returned %d, exceeding its budget of %d.\n", 6901 n->poll, work, weight); 6902 6903 if (likely(work < weight)) 6904 return work; 6905 6906 /* Drivers must not modify the NAPI state if they 6907 * consume the entire weight. In such cases this code 6908 * still "owns" the NAPI instance and therefore can 6909 * move the instance around on the list at-will. 6910 */ 6911 if (unlikely(napi_disable_pending(n))) { 6912 napi_complete(n); 6913 return work; 6914 } 6915 6916 /* The NAPI context has more processing work, but busy-polling 6917 * is preferred. Exit early. 6918 */ 6919 if (napi_prefer_busy_poll(n)) { 6920 if (napi_complete_done(n, work)) { 6921 /* If timeout is not set, we need to make sure 6922 * that the NAPI is re-scheduled. 6923 */ 6924 napi_schedule(n); 6925 } 6926 return work; 6927 } 6928 6929 if (n->gro_bitmask) { 6930 /* flush too old packets 6931 * If HZ < 1000, flush all packets. 6932 */ 6933 napi_gro_flush(n, HZ >= 1000); 6934 } 6935 6936 gro_normal_list(n); 6937 6938 /* Some drivers may have called napi_schedule 6939 * prior to exhausting their budget. 6940 */ 6941 if (unlikely(!list_empty(&n->poll_list))) { 6942 pr_warn_once("%s: Budget exhausted after napi rescheduled\n", 6943 n->dev ? n->dev->name : "backlog"); 6944 return work; 6945 } 6946 6947 *repoll = true; 6948 6949 return work; 6950 } 6951 6952 static int napi_poll(struct napi_struct *n, struct list_head *repoll) 6953 { 6954 bool do_repoll = false; 6955 void *have; 6956 int work; 6957 6958 list_del_init(&n->poll_list); 6959 6960 have = netpoll_poll_lock(n); 6961 6962 work = __napi_poll(n, &do_repoll); 6963 6964 if (do_repoll) 6965 list_add_tail(&n->poll_list, repoll); 6966 6967 netpoll_poll_unlock(have); 6968 6969 return work; 6970 } 6971 6972 static int napi_thread_wait(struct napi_struct *napi) 6973 { 6974 set_current_state(TASK_INTERRUPTIBLE); 6975 6976 while (!kthread_should_stop() && !napi_disable_pending(napi)) { 6977 if (test_bit(NAPI_STATE_SCHED, &napi->state)) { 6978 WARN_ON(!list_empty(&napi->poll_list)); 6979 __set_current_state(TASK_RUNNING); 6980 return 0; 6981 } 6982 6983 schedule(); 6984 set_current_state(TASK_INTERRUPTIBLE); 6985 } 6986 __set_current_state(TASK_RUNNING); 6987 return -1; 6988 } 6989 6990 static int napi_threaded_poll(void *data) 6991 { 6992 struct napi_struct *napi = data; 6993 void *have; 6994 6995 while (!napi_thread_wait(napi)) { 6996 for (;;) { 6997 bool repoll = false; 6998 6999 local_bh_disable(); 7000 7001 have = netpoll_poll_lock(napi); 7002 __napi_poll(napi, &repoll); 7003 netpoll_poll_unlock(have); 7004 7005 local_bh_enable(); 7006 7007 if (!repoll) 7008 break; 7009 7010 cond_resched(); 7011 } 7012 } 7013 return 0; 7014 } 7015 7016 static __latent_entropy void net_rx_action(struct softirq_action *h) 7017 { 7018 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 7019 unsigned long time_limit = jiffies + 7020 usecs_to_jiffies(netdev_budget_usecs); 7021 int budget = netdev_budget; 7022 LIST_HEAD(list); 7023 LIST_HEAD(repoll); 7024 7025 local_irq_disable(); 7026 list_splice_init(&sd->poll_list, &list); 7027 local_irq_enable(); 7028 7029 for (;;) { 7030 struct napi_struct *n; 7031 7032 if (list_empty(&list)) { 7033 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll)) 7034 return; 7035 break; 7036 } 7037 7038 n = list_first_entry(&list, struct napi_struct, poll_list); 7039 budget -= napi_poll(n, &repoll); 7040 7041 /* If softirq window is exhausted then punt. 7042 * Allow this to run for 2 jiffies since which will allow 7043 * an average latency of 1.5/HZ. 7044 */ 7045 if (unlikely(budget <= 0 || 7046 time_after_eq(jiffies, time_limit))) { 7047 sd->time_squeeze++; 7048 break; 7049 } 7050 } 7051 7052 local_irq_disable(); 7053 7054 list_splice_tail_init(&sd->poll_list, &list); 7055 list_splice_tail(&repoll, &list); 7056 list_splice(&list, &sd->poll_list); 7057 if (!list_empty(&sd->poll_list)) 7058 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 7059 7060 net_rps_action_and_irq_enable(sd); 7061 } 7062 7063 struct netdev_adjacent { 7064 struct net_device *dev; 7065 7066 /* upper master flag, there can only be one master device per list */ 7067 bool master; 7068 7069 /* lookup ignore flag */ 7070 bool ignore; 7071 7072 /* counter for the number of times this device was added to us */ 7073 u16 ref_nr; 7074 7075 /* private field for the users */ 7076 void *private; 7077 7078 struct list_head list; 7079 struct rcu_head rcu; 7080 }; 7081 7082 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, 7083 struct list_head *adj_list) 7084 { 7085 struct netdev_adjacent *adj; 7086 7087 list_for_each_entry(adj, adj_list, list) { 7088 if (adj->dev == adj_dev) 7089 return adj; 7090 } 7091 return NULL; 7092 } 7093 7094 static int ____netdev_has_upper_dev(struct net_device *upper_dev, 7095 struct netdev_nested_priv *priv) 7096 { 7097 struct net_device *dev = (struct net_device *)priv->data; 7098 7099 return upper_dev == dev; 7100 } 7101 7102 /** 7103 * netdev_has_upper_dev - Check if device is linked to an upper device 7104 * @dev: device 7105 * @upper_dev: upper device to check 7106 * 7107 * Find out if a device is linked to specified upper device and return true 7108 * in case it is. Note that this checks only immediate upper device, 7109 * not through a complete stack of devices. The caller must hold the RTNL lock. 7110 */ 7111 bool netdev_has_upper_dev(struct net_device *dev, 7112 struct net_device *upper_dev) 7113 { 7114 struct netdev_nested_priv priv = { 7115 .data = (void *)upper_dev, 7116 }; 7117 7118 ASSERT_RTNL(); 7119 7120 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 7121 &priv); 7122 } 7123 EXPORT_SYMBOL(netdev_has_upper_dev); 7124 7125 /** 7126 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device 7127 * @dev: device 7128 * @upper_dev: upper device to check 7129 * 7130 * Find out if a device is linked to specified upper device and return true 7131 * in case it is. Note that this checks the entire upper device chain. 7132 * The caller must hold rcu lock. 7133 */ 7134 7135 bool netdev_has_upper_dev_all_rcu(struct net_device *dev, 7136 struct net_device *upper_dev) 7137 { 7138 struct netdev_nested_priv priv = { 7139 .data = (void *)upper_dev, 7140 }; 7141 7142 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 7143 &priv); 7144 } 7145 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); 7146 7147 /** 7148 * netdev_has_any_upper_dev - Check if device is linked to some device 7149 * @dev: device 7150 * 7151 * Find out if a device is linked to an upper device and return true in case 7152 * it is. The caller must hold the RTNL lock. 7153 */ 7154 bool netdev_has_any_upper_dev(struct net_device *dev) 7155 { 7156 ASSERT_RTNL(); 7157 7158 return !list_empty(&dev->adj_list.upper); 7159 } 7160 EXPORT_SYMBOL(netdev_has_any_upper_dev); 7161 7162 /** 7163 * netdev_master_upper_dev_get - Get master upper device 7164 * @dev: device 7165 * 7166 * Find a master upper device and return pointer to it or NULL in case 7167 * it's not there. The caller must hold the RTNL lock. 7168 */ 7169 struct net_device *netdev_master_upper_dev_get(struct net_device *dev) 7170 { 7171 struct netdev_adjacent *upper; 7172 7173 ASSERT_RTNL(); 7174 7175 if (list_empty(&dev->adj_list.upper)) 7176 return NULL; 7177 7178 upper = list_first_entry(&dev->adj_list.upper, 7179 struct netdev_adjacent, list); 7180 if (likely(upper->master)) 7181 return upper->dev; 7182 return NULL; 7183 } 7184 EXPORT_SYMBOL(netdev_master_upper_dev_get); 7185 7186 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev) 7187 { 7188 struct netdev_adjacent *upper; 7189 7190 ASSERT_RTNL(); 7191 7192 if (list_empty(&dev->adj_list.upper)) 7193 return NULL; 7194 7195 upper = list_first_entry(&dev->adj_list.upper, 7196 struct netdev_adjacent, list); 7197 if (likely(upper->master) && !upper->ignore) 7198 return upper->dev; 7199 return NULL; 7200 } 7201 7202 /** 7203 * netdev_has_any_lower_dev - Check if device is linked to some device 7204 * @dev: device 7205 * 7206 * Find out if a device is linked to a lower device and return true in case 7207 * it is. The caller must hold the RTNL lock. 7208 */ 7209 static bool netdev_has_any_lower_dev(struct net_device *dev) 7210 { 7211 ASSERT_RTNL(); 7212 7213 return !list_empty(&dev->adj_list.lower); 7214 } 7215 7216 void *netdev_adjacent_get_private(struct list_head *adj_list) 7217 { 7218 struct netdev_adjacent *adj; 7219 7220 adj = list_entry(adj_list, struct netdev_adjacent, list); 7221 7222 return adj->private; 7223 } 7224 EXPORT_SYMBOL(netdev_adjacent_get_private); 7225 7226 /** 7227 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list 7228 * @dev: device 7229 * @iter: list_head ** of the current position 7230 * 7231 * Gets the next device from the dev's upper list, starting from iter 7232 * position. The caller must hold RCU read lock. 7233 */ 7234 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, 7235 struct list_head **iter) 7236 { 7237 struct netdev_adjacent *upper; 7238 7239 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 7240 7241 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7242 7243 if (&upper->list == &dev->adj_list.upper) 7244 return NULL; 7245 7246 *iter = &upper->list; 7247 7248 return upper->dev; 7249 } 7250 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); 7251 7252 static struct net_device *__netdev_next_upper_dev(struct net_device *dev, 7253 struct list_head **iter, 7254 bool *ignore) 7255 { 7256 struct netdev_adjacent *upper; 7257 7258 upper = list_entry((*iter)->next, struct netdev_adjacent, list); 7259 7260 if (&upper->list == &dev->adj_list.upper) 7261 return NULL; 7262 7263 *iter = &upper->list; 7264 *ignore = upper->ignore; 7265 7266 return upper->dev; 7267 } 7268 7269 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, 7270 struct list_head **iter) 7271 { 7272 struct netdev_adjacent *upper; 7273 7274 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 7275 7276 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7277 7278 if (&upper->list == &dev->adj_list.upper) 7279 return NULL; 7280 7281 *iter = &upper->list; 7282 7283 return upper->dev; 7284 } 7285 7286 static int __netdev_walk_all_upper_dev(struct net_device *dev, 7287 int (*fn)(struct net_device *dev, 7288 struct netdev_nested_priv *priv), 7289 struct netdev_nested_priv *priv) 7290 { 7291 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7292 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7293 int ret, cur = 0; 7294 bool ignore; 7295 7296 now = dev; 7297 iter = &dev->adj_list.upper; 7298 7299 while (1) { 7300 if (now != dev) { 7301 ret = fn(now, priv); 7302 if (ret) 7303 return ret; 7304 } 7305 7306 next = NULL; 7307 while (1) { 7308 udev = __netdev_next_upper_dev(now, &iter, &ignore); 7309 if (!udev) 7310 break; 7311 if (ignore) 7312 continue; 7313 7314 next = udev; 7315 niter = &udev->adj_list.upper; 7316 dev_stack[cur] = now; 7317 iter_stack[cur++] = iter; 7318 break; 7319 } 7320 7321 if (!next) { 7322 if (!cur) 7323 return 0; 7324 next = dev_stack[--cur]; 7325 niter = iter_stack[cur]; 7326 } 7327 7328 now = next; 7329 iter = niter; 7330 } 7331 7332 return 0; 7333 } 7334 7335 int netdev_walk_all_upper_dev_rcu(struct net_device *dev, 7336 int (*fn)(struct net_device *dev, 7337 struct netdev_nested_priv *priv), 7338 struct netdev_nested_priv *priv) 7339 { 7340 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7341 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7342 int ret, cur = 0; 7343 7344 now = dev; 7345 iter = &dev->adj_list.upper; 7346 7347 while (1) { 7348 if (now != dev) { 7349 ret = fn(now, priv); 7350 if (ret) 7351 return ret; 7352 } 7353 7354 next = NULL; 7355 while (1) { 7356 udev = netdev_next_upper_dev_rcu(now, &iter); 7357 if (!udev) 7358 break; 7359 7360 next = udev; 7361 niter = &udev->adj_list.upper; 7362 dev_stack[cur] = now; 7363 iter_stack[cur++] = iter; 7364 break; 7365 } 7366 7367 if (!next) { 7368 if (!cur) 7369 return 0; 7370 next = dev_stack[--cur]; 7371 niter = iter_stack[cur]; 7372 } 7373 7374 now = next; 7375 iter = niter; 7376 } 7377 7378 return 0; 7379 } 7380 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); 7381 7382 static bool __netdev_has_upper_dev(struct net_device *dev, 7383 struct net_device *upper_dev) 7384 { 7385 struct netdev_nested_priv priv = { 7386 .flags = 0, 7387 .data = (void *)upper_dev, 7388 }; 7389 7390 ASSERT_RTNL(); 7391 7392 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev, 7393 &priv); 7394 } 7395 7396 /** 7397 * netdev_lower_get_next_private - Get the next ->private from the 7398 * lower neighbour list 7399 * @dev: device 7400 * @iter: list_head ** of the current position 7401 * 7402 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7403 * list, starting from iter position. The caller must hold either hold the 7404 * RTNL lock or its own locking that guarantees that the neighbour lower 7405 * list will remain unchanged. 7406 */ 7407 void *netdev_lower_get_next_private(struct net_device *dev, 7408 struct list_head **iter) 7409 { 7410 struct netdev_adjacent *lower; 7411 7412 lower = list_entry(*iter, struct netdev_adjacent, list); 7413 7414 if (&lower->list == &dev->adj_list.lower) 7415 return NULL; 7416 7417 *iter = lower->list.next; 7418 7419 return lower->private; 7420 } 7421 EXPORT_SYMBOL(netdev_lower_get_next_private); 7422 7423 /** 7424 * netdev_lower_get_next_private_rcu - Get the next ->private from the 7425 * lower neighbour list, RCU 7426 * variant 7427 * @dev: device 7428 * @iter: list_head ** of the current position 7429 * 7430 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7431 * list, starting from iter position. The caller must hold RCU read lock. 7432 */ 7433 void *netdev_lower_get_next_private_rcu(struct net_device *dev, 7434 struct list_head **iter) 7435 { 7436 struct netdev_adjacent *lower; 7437 7438 WARN_ON_ONCE(!rcu_read_lock_held()); 7439 7440 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7441 7442 if (&lower->list == &dev->adj_list.lower) 7443 return NULL; 7444 7445 *iter = &lower->list; 7446 7447 return lower->private; 7448 } 7449 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); 7450 7451 /** 7452 * netdev_lower_get_next - Get the next device from the lower neighbour 7453 * list 7454 * @dev: device 7455 * @iter: list_head ** of the current position 7456 * 7457 * Gets the next netdev_adjacent from the dev's lower neighbour 7458 * list, starting from iter position. The caller must hold RTNL lock or 7459 * its own locking that guarantees that the neighbour lower 7460 * list will remain unchanged. 7461 */ 7462 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) 7463 { 7464 struct netdev_adjacent *lower; 7465 7466 lower = list_entry(*iter, struct netdev_adjacent, list); 7467 7468 if (&lower->list == &dev->adj_list.lower) 7469 return NULL; 7470 7471 *iter = lower->list.next; 7472 7473 return lower->dev; 7474 } 7475 EXPORT_SYMBOL(netdev_lower_get_next); 7476 7477 static struct net_device *netdev_next_lower_dev(struct net_device *dev, 7478 struct list_head **iter) 7479 { 7480 struct netdev_adjacent *lower; 7481 7482 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7483 7484 if (&lower->list == &dev->adj_list.lower) 7485 return NULL; 7486 7487 *iter = &lower->list; 7488 7489 return lower->dev; 7490 } 7491 7492 static struct net_device *__netdev_next_lower_dev(struct net_device *dev, 7493 struct list_head **iter, 7494 bool *ignore) 7495 { 7496 struct netdev_adjacent *lower; 7497 7498 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7499 7500 if (&lower->list == &dev->adj_list.lower) 7501 return NULL; 7502 7503 *iter = &lower->list; 7504 *ignore = lower->ignore; 7505 7506 return lower->dev; 7507 } 7508 7509 int netdev_walk_all_lower_dev(struct net_device *dev, 7510 int (*fn)(struct net_device *dev, 7511 struct netdev_nested_priv *priv), 7512 struct netdev_nested_priv *priv) 7513 { 7514 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7515 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7516 int ret, cur = 0; 7517 7518 now = dev; 7519 iter = &dev->adj_list.lower; 7520 7521 while (1) { 7522 if (now != dev) { 7523 ret = fn(now, priv); 7524 if (ret) 7525 return ret; 7526 } 7527 7528 next = NULL; 7529 while (1) { 7530 ldev = netdev_next_lower_dev(now, &iter); 7531 if (!ldev) 7532 break; 7533 7534 next = ldev; 7535 niter = &ldev->adj_list.lower; 7536 dev_stack[cur] = now; 7537 iter_stack[cur++] = iter; 7538 break; 7539 } 7540 7541 if (!next) { 7542 if (!cur) 7543 return 0; 7544 next = dev_stack[--cur]; 7545 niter = iter_stack[cur]; 7546 } 7547 7548 now = next; 7549 iter = niter; 7550 } 7551 7552 return 0; 7553 } 7554 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); 7555 7556 static int __netdev_walk_all_lower_dev(struct net_device *dev, 7557 int (*fn)(struct net_device *dev, 7558 struct netdev_nested_priv *priv), 7559 struct netdev_nested_priv *priv) 7560 { 7561 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7562 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7563 int ret, cur = 0; 7564 bool ignore; 7565 7566 now = dev; 7567 iter = &dev->adj_list.lower; 7568 7569 while (1) { 7570 if (now != dev) { 7571 ret = fn(now, priv); 7572 if (ret) 7573 return ret; 7574 } 7575 7576 next = NULL; 7577 while (1) { 7578 ldev = __netdev_next_lower_dev(now, &iter, &ignore); 7579 if (!ldev) 7580 break; 7581 if (ignore) 7582 continue; 7583 7584 next = ldev; 7585 niter = &ldev->adj_list.lower; 7586 dev_stack[cur] = now; 7587 iter_stack[cur++] = iter; 7588 break; 7589 } 7590 7591 if (!next) { 7592 if (!cur) 7593 return 0; 7594 next = dev_stack[--cur]; 7595 niter = iter_stack[cur]; 7596 } 7597 7598 now = next; 7599 iter = niter; 7600 } 7601 7602 return 0; 7603 } 7604 7605 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, 7606 struct list_head **iter) 7607 { 7608 struct netdev_adjacent *lower; 7609 7610 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7611 if (&lower->list == &dev->adj_list.lower) 7612 return NULL; 7613 7614 *iter = &lower->list; 7615 7616 return lower->dev; 7617 } 7618 EXPORT_SYMBOL(netdev_next_lower_dev_rcu); 7619 7620 static u8 __netdev_upper_depth(struct net_device *dev) 7621 { 7622 struct net_device *udev; 7623 struct list_head *iter; 7624 u8 max_depth = 0; 7625 bool ignore; 7626 7627 for (iter = &dev->adj_list.upper, 7628 udev = __netdev_next_upper_dev(dev, &iter, &ignore); 7629 udev; 7630 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) { 7631 if (ignore) 7632 continue; 7633 if (max_depth < udev->upper_level) 7634 max_depth = udev->upper_level; 7635 } 7636 7637 return max_depth; 7638 } 7639 7640 static u8 __netdev_lower_depth(struct net_device *dev) 7641 { 7642 struct net_device *ldev; 7643 struct list_head *iter; 7644 u8 max_depth = 0; 7645 bool ignore; 7646 7647 for (iter = &dev->adj_list.lower, 7648 ldev = __netdev_next_lower_dev(dev, &iter, &ignore); 7649 ldev; 7650 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) { 7651 if (ignore) 7652 continue; 7653 if (max_depth < ldev->lower_level) 7654 max_depth = ldev->lower_level; 7655 } 7656 7657 return max_depth; 7658 } 7659 7660 static int __netdev_update_upper_level(struct net_device *dev, 7661 struct netdev_nested_priv *__unused) 7662 { 7663 dev->upper_level = __netdev_upper_depth(dev) + 1; 7664 return 0; 7665 } 7666 7667 static int __netdev_update_lower_level(struct net_device *dev, 7668 struct netdev_nested_priv *priv) 7669 { 7670 dev->lower_level = __netdev_lower_depth(dev) + 1; 7671 7672 #ifdef CONFIG_LOCKDEP 7673 if (!priv) 7674 return 0; 7675 7676 if (priv->flags & NESTED_SYNC_IMM) 7677 dev->nested_level = dev->lower_level - 1; 7678 if (priv->flags & NESTED_SYNC_TODO) 7679 net_unlink_todo(dev); 7680 #endif 7681 return 0; 7682 } 7683 7684 int netdev_walk_all_lower_dev_rcu(struct net_device *dev, 7685 int (*fn)(struct net_device *dev, 7686 struct netdev_nested_priv *priv), 7687 struct netdev_nested_priv *priv) 7688 { 7689 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7690 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7691 int ret, cur = 0; 7692 7693 now = dev; 7694 iter = &dev->adj_list.lower; 7695 7696 while (1) { 7697 if (now != dev) { 7698 ret = fn(now, priv); 7699 if (ret) 7700 return ret; 7701 } 7702 7703 next = NULL; 7704 while (1) { 7705 ldev = netdev_next_lower_dev_rcu(now, &iter); 7706 if (!ldev) 7707 break; 7708 7709 next = ldev; 7710 niter = &ldev->adj_list.lower; 7711 dev_stack[cur] = now; 7712 iter_stack[cur++] = iter; 7713 break; 7714 } 7715 7716 if (!next) { 7717 if (!cur) 7718 return 0; 7719 next = dev_stack[--cur]; 7720 niter = iter_stack[cur]; 7721 } 7722 7723 now = next; 7724 iter = niter; 7725 } 7726 7727 return 0; 7728 } 7729 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); 7730 7731 /** 7732 * netdev_lower_get_first_private_rcu - Get the first ->private from the 7733 * lower neighbour list, RCU 7734 * variant 7735 * @dev: device 7736 * 7737 * Gets the first netdev_adjacent->private from the dev's lower neighbour 7738 * list. The caller must hold RCU read lock. 7739 */ 7740 void *netdev_lower_get_first_private_rcu(struct net_device *dev) 7741 { 7742 struct netdev_adjacent *lower; 7743 7744 lower = list_first_or_null_rcu(&dev->adj_list.lower, 7745 struct netdev_adjacent, list); 7746 if (lower) 7747 return lower->private; 7748 return NULL; 7749 } 7750 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); 7751 7752 /** 7753 * netdev_master_upper_dev_get_rcu - Get master upper device 7754 * @dev: device 7755 * 7756 * Find a master upper device and return pointer to it or NULL in case 7757 * it's not there. The caller must hold the RCU read lock. 7758 */ 7759 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) 7760 { 7761 struct netdev_adjacent *upper; 7762 7763 upper = list_first_or_null_rcu(&dev->adj_list.upper, 7764 struct netdev_adjacent, list); 7765 if (upper && likely(upper->master)) 7766 return upper->dev; 7767 return NULL; 7768 } 7769 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); 7770 7771 static int netdev_adjacent_sysfs_add(struct net_device *dev, 7772 struct net_device *adj_dev, 7773 struct list_head *dev_list) 7774 { 7775 char linkname[IFNAMSIZ+7]; 7776 7777 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7778 "upper_%s" : "lower_%s", adj_dev->name); 7779 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), 7780 linkname); 7781 } 7782 static void netdev_adjacent_sysfs_del(struct net_device *dev, 7783 char *name, 7784 struct list_head *dev_list) 7785 { 7786 char linkname[IFNAMSIZ+7]; 7787 7788 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7789 "upper_%s" : "lower_%s", name); 7790 sysfs_remove_link(&(dev->dev.kobj), linkname); 7791 } 7792 7793 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, 7794 struct net_device *adj_dev, 7795 struct list_head *dev_list) 7796 { 7797 return (dev_list == &dev->adj_list.upper || 7798 dev_list == &dev->adj_list.lower) && 7799 net_eq(dev_net(dev), dev_net(adj_dev)); 7800 } 7801 7802 static int __netdev_adjacent_dev_insert(struct net_device *dev, 7803 struct net_device *adj_dev, 7804 struct list_head *dev_list, 7805 void *private, bool master) 7806 { 7807 struct netdev_adjacent *adj; 7808 int ret; 7809 7810 adj = __netdev_find_adj(adj_dev, dev_list); 7811 7812 if (adj) { 7813 adj->ref_nr += 1; 7814 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", 7815 dev->name, adj_dev->name, adj->ref_nr); 7816 7817 return 0; 7818 } 7819 7820 adj = kmalloc(sizeof(*adj), GFP_KERNEL); 7821 if (!adj) 7822 return -ENOMEM; 7823 7824 adj->dev = adj_dev; 7825 adj->master = master; 7826 adj->ref_nr = 1; 7827 adj->private = private; 7828 adj->ignore = false; 7829 dev_hold(adj_dev); 7830 7831 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", 7832 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); 7833 7834 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { 7835 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); 7836 if (ret) 7837 goto free_adj; 7838 } 7839 7840 /* Ensure that master link is always the first item in list. */ 7841 if (master) { 7842 ret = sysfs_create_link(&(dev->dev.kobj), 7843 &(adj_dev->dev.kobj), "master"); 7844 if (ret) 7845 goto remove_symlinks; 7846 7847 list_add_rcu(&adj->list, dev_list); 7848 } else { 7849 list_add_tail_rcu(&adj->list, dev_list); 7850 } 7851 7852 return 0; 7853 7854 remove_symlinks: 7855 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7856 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7857 free_adj: 7858 kfree(adj); 7859 dev_put(adj_dev); 7860 7861 return ret; 7862 } 7863 7864 static void __netdev_adjacent_dev_remove(struct net_device *dev, 7865 struct net_device *adj_dev, 7866 u16 ref_nr, 7867 struct list_head *dev_list) 7868 { 7869 struct netdev_adjacent *adj; 7870 7871 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", 7872 dev->name, adj_dev->name, ref_nr); 7873 7874 adj = __netdev_find_adj(adj_dev, dev_list); 7875 7876 if (!adj) { 7877 pr_err("Adjacency does not exist for device %s from %s\n", 7878 dev->name, adj_dev->name); 7879 WARN_ON(1); 7880 return; 7881 } 7882 7883 if (adj->ref_nr > ref_nr) { 7884 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", 7885 dev->name, adj_dev->name, ref_nr, 7886 adj->ref_nr - ref_nr); 7887 adj->ref_nr -= ref_nr; 7888 return; 7889 } 7890 7891 if (adj->master) 7892 sysfs_remove_link(&(dev->dev.kobj), "master"); 7893 7894 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7895 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7896 7897 list_del_rcu(&adj->list); 7898 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", 7899 adj_dev->name, dev->name, adj_dev->name); 7900 dev_put(adj_dev); 7901 kfree_rcu(adj, rcu); 7902 } 7903 7904 static int __netdev_adjacent_dev_link_lists(struct net_device *dev, 7905 struct net_device *upper_dev, 7906 struct list_head *up_list, 7907 struct list_head *down_list, 7908 void *private, bool master) 7909 { 7910 int ret; 7911 7912 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, 7913 private, master); 7914 if (ret) 7915 return ret; 7916 7917 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, 7918 private, false); 7919 if (ret) { 7920 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); 7921 return ret; 7922 } 7923 7924 return 0; 7925 } 7926 7927 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, 7928 struct net_device *upper_dev, 7929 u16 ref_nr, 7930 struct list_head *up_list, 7931 struct list_head *down_list) 7932 { 7933 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); 7934 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); 7935 } 7936 7937 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, 7938 struct net_device *upper_dev, 7939 void *private, bool master) 7940 { 7941 return __netdev_adjacent_dev_link_lists(dev, upper_dev, 7942 &dev->adj_list.upper, 7943 &upper_dev->adj_list.lower, 7944 private, master); 7945 } 7946 7947 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, 7948 struct net_device *upper_dev) 7949 { 7950 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, 7951 &dev->adj_list.upper, 7952 &upper_dev->adj_list.lower); 7953 } 7954 7955 static int __netdev_upper_dev_link(struct net_device *dev, 7956 struct net_device *upper_dev, bool master, 7957 void *upper_priv, void *upper_info, 7958 struct netdev_nested_priv *priv, 7959 struct netlink_ext_ack *extack) 7960 { 7961 struct netdev_notifier_changeupper_info changeupper_info = { 7962 .info = { 7963 .dev = dev, 7964 .extack = extack, 7965 }, 7966 .upper_dev = upper_dev, 7967 .master = master, 7968 .linking = true, 7969 .upper_info = upper_info, 7970 }; 7971 struct net_device *master_dev; 7972 int ret = 0; 7973 7974 ASSERT_RTNL(); 7975 7976 if (dev == upper_dev) 7977 return -EBUSY; 7978 7979 /* To prevent loops, check if dev is not upper device to upper_dev. */ 7980 if (__netdev_has_upper_dev(upper_dev, dev)) 7981 return -EBUSY; 7982 7983 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV) 7984 return -EMLINK; 7985 7986 if (!master) { 7987 if (__netdev_has_upper_dev(dev, upper_dev)) 7988 return -EEXIST; 7989 } else { 7990 master_dev = __netdev_master_upper_dev_get(dev); 7991 if (master_dev) 7992 return master_dev == upper_dev ? -EEXIST : -EBUSY; 7993 } 7994 7995 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7996 &changeupper_info.info); 7997 ret = notifier_to_errno(ret); 7998 if (ret) 7999 return ret; 8000 8001 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, 8002 master); 8003 if (ret) 8004 return ret; 8005 8006 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 8007 &changeupper_info.info); 8008 ret = notifier_to_errno(ret); 8009 if (ret) 8010 goto rollback; 8011 8012 __netdev_update_upper_level(dev, NULL); 8013 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 8014 8015 __netdev_update_lower_level(upper_dev, priv); 8016 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 8017 priv); 8018 8019 return 0; 8020 8021 rollback: 8022 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 8023 8024 return ret; 8025 } 8026 8027 /** 8028 * netdev_upper_dev_link - Add a link to the upper device 8029 * @dev: device 8030 * @upper_dev: new upper device 8031 * @extack: netlink extended ack 8032 * 8033 * Adds a link to device which is upper to this one. The caller must hold 8034 * the RTNL lock. On a failure a negative errno code is returned. 8035 * On success the reference counts are adjusted and the function 8036 * returns zero. 8037 */ 8038 int netdev_upper_dev_link(struct net_device *dev, 8039 struct net_device *upper_dev, 8040 struct netlink_ext_ack *extack) 8041 { 8042 struct netdev_nested_priv priv = { 8043 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 8044 .data = NULL, 8045 }; 8046 8047 return __netdev_upper_dev_link(dev, upper_dev, false, 8048 NULL, NULL, &priv, extack); 8049 } 8050 EXPORT_SYMBOL(netdev_upper_dev_link); 8051 8052 /** 8053 * netdev_master_upper_dev_link - Add a master link to the upper device 8054 * @dev: device 8055 * @upper_dev: new upper device 8056 * @upper_priv: upper device private 8057 * @upper_info: upper info to be passed down via notifier 8058 * @extack: netlink extended ack 8059 * 8060 * Adds a link to device which is upper to this one. In this case, only 8061 * one master upper device can be linked, although other non-master devices 8062 * might be linked as well. The caller must hold the RTNL lock. 8063 * On a failure a negative errno code is returned. On success the reference 8064 * counts are adjusted and the function returns zero. 8065 */ 8066 int netdev_master_upper_dev_link(struct net_device *dev, 8067 struct net_device *upper_dev, 8068 void *upper_priv, void *upper_info, 8069 struct netlink_ext_ack *extack) 8070 { 8071 struct netdev_nested_priv priv = { 8072 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 8073 .data = NULL, 8074 }; 8075 8076 return __netdev_upper_dev_link(dev, upper_dev, true, 8077 upper_priv, upper_info, &priv, extack); 8078 } 8079 EXPORT_SYMBOL(netdev_master_upper_dev_link); 8080 8081 static void __netdev_upper_dev_unlink(struct net_device *dev, 8082 struct net_device *upper_dev, 8083 struct netdev_nested_priv *priv) 8084 { 8085 struct netdev_notifier_changeupper_info changeupper_info = { 8086 .info = { 8087 .dev = dev, 8088 }, 8089 .upper_dev = upper_dev, 8090 .linking = false, 8091 }; 8092 8093 ASSERT_RTNL(); 8094 8095 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; 8096 8097 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 8098 &changeupper_info.info); 8099 8100 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 8101 8102 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 8103 &changeupper_info.info); 8104 8105 __netdev_update_upper_level(dev, NULL); 8106 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 8107 8108 __netdev_update_lower_level(upper_dev, priv); 8109 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 8110 priv); 8111 } 8112 8113 /** 8114 * netdev_upper_dev_unlink - Removes a link to upper device 8115 * @dev: device 8116 * @upper_dev: new upper device 8117 * 8118 * Removes a link to device which is upper to this one. The caller must hold 8119 * the RTNL lock. 8120 */ 8121 void netdev_upper_dev_unlink(struct net_device *dev, 8122 struct net_device *upper_dev) 8123 { 8124 struct netdev_nested_priv priv = { 8125 .flags = NESTED_SYNC_TODO, 8126 .data = NULL, 8127 }; 8128 8129 __netdev_upper_dev_unlink(dev, upper_dev, &priv); 8130 } 8131 EXPORT_SYMBOL(netdev_upper_dev_unlink); 8132 8133 static void __netdev_adjacent_dev_set(struct net_device *upper_dev, 8134 struct net_device *lower_dev, 8135 bool val) 8136 { 8137 struct netdev_adjacent *adj; 8138 8139 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower); 8140 if (adj) 8141 adj->ignore = val; 8142 8143 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper); 8144 if (adj) 8145 adj->ignore = val; 8146 } 8147 8148 static void netdev_adjacent_dev_disable(struct net_device *upper_dev, 8149 struct net_device *lower_dev) 8150 { 8151 __netdev_adjacent_dev_set(upper_dev, lower_dev, true); 8152 } 8153 8154 static void netdev_adjacent_dev_enable(struct net_device *upper_dev, 8155 struct net_device *lower_dev) 8156 { 8157 __netdev_adjacent_dev_set(upper_dev, lower_dev, false); 8158 } 8159 8160 int netdev_adjacent_change_prepare(struct net_device *old_dev, 8161 struct net_device *new_dev, 8162 struct net_device *dev, 8163 struct netlink_ext_ack *extack) 8164 { 8165 struct netdev_nested_priv priv = { 8166 .flags = 0, 8167 .data = NULL, 8168 }; 8169 int err; 8170 8171 if (!new_dev) 8172 return 0; 8173 8174 if (old_dev && new_dev != old_dev) 8175 netdev_adjacent_dev_disable(dev, old_dev); 8176 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv, 8177 extack); 8178 if (err) { 8179 if (old_dev && new_dev != old_dev) 8180 netdev_adjacent_dev_enable(dev, old_dev); 8181 return err; 8182 } 8183 8184 return 0; 8185 } 8186 EXPORT_SYMBOL(netdev_adjacent_change_prepare); 8187 8188 void netdev_adjacent_change_commit(struct net_device *old_dev, 8189 struct net_device *new_dev, 8190 struct net_device *dev) 8191 { 8192 struct netdev_nested_priv priv = { 8193 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 8194 .data = NULL, 8195 }; 8196 8197 if (!new_dev || !old_dev) 8198 return; 8199 8200 if (new_dev == old_dev) 8201 return; 8202 8203 netdev_adjacent_dev_enable(dev, old_dev); 8204 __netdev_upper_dev_unlink(old_dev, dev, &priv); 8205 } 8206 EXPORT_SYMBOL(netdev_adjacent_change_commit); 8207 8208 void netdev_adjacent_change_abort(struct net_device *old_dev, 8209 struct net_device *new_dev, 8210 struct net_device *dev) 8211 { 8212 struct netdev_nested_priv priv = { 8213 .flags = 0, 8214 .data = NULL, 8215 }; 8216 8217 if (!new_dev) 8218 return; 8219 8220 if (old_dev && new_dev != old_dev) 8221 netdev_adjacent_dev_enable(dev, old_dev); 8222 8223 __netdev_upper_dev_unlink(new_dev, dev, &priv); 8224 } 8225 EXPORT_SYMBOL(netdev_adjacent_change_abort); 8226 8227 /** 8228 * netdev_bonding_info_change - Dispatch event about slave change 8229 * @dev: device 8230 * @bonding_info: info to dispatch 8231 * 8232 * Send NETDEV_BONDING_INFO to netdev notifiers with info. 8233 * The caller must hold the RTNL lock. 8234 */ 8235 void netdev_bonding_info_change(struct net_device *dev, 8236 struct netdev_bonding_info *bonding_info) 8237 { 8238 struct netdev_notifier_bonding_info info = { 8239 .info.dev = dev, 8240 }; 8241 8242 memcpy(&info.bonding_info, bonding_info, 8243 sizeof(struct netdev_bonding_info)); 8244 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, 8245 &info.info); 8246 } 8247 EXPORT_SYMBOL(netdev_bonding_info_change); 8248 8249 /** 8250 * netdev_get_xmit_slave - Get the xmit slave of master device 8251 * @dev: device 8252 * @skb: The packet 8253 * @all_slaves: assume all the slaves are active 8254 * 8255 * The reference counters are not incremented so the caller must be 8256 * careful with locks. The caller must hold RCU lock. 8257 * %NULL is returned if no slave is found. 8258 */ 8259 8260 struct net_device *netdev_get_xmit_slave(struct net_device *dev, 8261 struct sk_buff *skb, 8262 bool all_slaves) 8263 { 8264 const struct net_device_ops *ops = dev->netdev_ops; 8265 8266 if (!ops->ndo_get_xmit_slave) 8267 return NULL; 8268 return ops->ndo_get_xmit_slave(dev, skb, all_slaves); 8269 } 8270 EXPORT_SYMBOL(netdev_get_xmit_slave); 8271 8272 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev, 8273 struct sock *sk) 8274 { 8275 const struct net_device_ops *ops = dev->netdev_ops; 8276 8277 if (!ops->ndo_sk_get_lower_dev) 8278 return NULL; 8279 return ops->ndo_sk_get_lower_dev(dev, sk); 8280 } 8281 8282 /** 8283 * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket 8284 * @dev: device 8285 * @sk: the socket 8286 * 8287 * %NULL is returned if no lower device is found. 8288 */ 8289 8290 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev, 8291 struct sock *sk) 8292 { 8293 struct net_device *lower; 8294 8295 lower = netdev_sk_get_lower_dev(dev, sk); 8296 while (lower) { 8297 dev = lower; 8298 lower = netdev_sk_get_lower_dev(dev, sk); 8299 } 8300 8301 return dev; 8302 } 8303 EXPORT_SYMBOL(netdev_sk_get_lowest_dev); 8304 8305 static void netdev_adjacent_add_links(struct net_device *dev) 8306 { 8307 struct netdev_adjacent *iter; 8308 8309 struct net *net = dev_net(dev); 8310 8311 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8312 if (!net_eq(net, dev_net(iter->dev))) 8313 continue; 8314 netdev_adjacent_sysfs_add(iter->dev, dev, 8315 &iter->dev->adj_list.lower); 8316 netdev_adjacent_sysfs_add(dev, iter->dev, 8317 &dev->adj_list.upper); 8318 } 8319 8320 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8321 if (!net_eq(net, dev_net(iter->dev))) 8322 continue; 8323 netdev_adjacent_sysfs_add(iter->dev, dev, 8324 &iter->dev->adj_list.upper); 8325 netdev_adjacent_sysfs_add(dev, iter->dev, 8326 &dev->adj_list.lower); 8327 } 8328 } 8329 8330 static void netdev_adjacent_del_links(struct net_device *dev) 8331 { 8332 struct netdev_adjacent *iter; 8333 8334 struct net *net = dev_net(dev); 8335 8336 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8337 if (!net_eq(net, dev_net(iter->dev))) 8338 continue; 8339 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8340 &iter->dev->adj_list.lower); 8341 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8342 &dev->adj_list.upper); 8343 } 8344 8345 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8346 if (!net_eq(net, dev_net(iter->dev))) 8347 continue; 8348 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8349 &iter->dev->adj_list.upper); 8350 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8351 &dev->adj_list.lower); 8352 } 8353 } 8354 8355 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) 8356 { 8357 struct netdev_adjacent *iter; 8358 8359 struct net *net = dev_net(dev); 8360 8361 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8362 if (!net_eq(net, dev_net(iter->dev))) 8363 continue; 8364 netdev_adjacent_sysfs_del(iter->dev, oldname, 8365 &iter->dev->adj_list.lower); 8366 netdev_adjacent_sysfs_add(iter->dev, dev, 8367 &iter->dev->adj_list.lower); 8368 } 8369 8370 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8371 if (!net_eq(net, dev_net(iter->dev))) 8372 continue; 8373 netdev_adjacent_sysfs_del(iter->dev, oldname, 8374 &iter->dev->adj_list.upper); 8375 netdev_adjacent_sysfs_add(iter->dev, dev, 8376 &iter->dev->adj_list.upper); 8377 } 8378 } 8379 8380 void *netdev_lower_dev_get_private(struct net_device *dev, 8381 struct net_device *lower_dev) 8382 { 8383 struct netdev_adjacent *lower; 8384 8385 if (!lower_dev) 8386 return NULL; 8387 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); 8388 if (!lower) 8389 return NULL; 8390 8391 return lower->private; 8392 } 8393 EXPORT_SYMBOL(netdev_lower_dev_get_private); 8394 8395 8396 /** 8397 * netdev_lower_state_changed - Dispatch event about lower device state change 8398 * @lower_dev: device 8399 * @lower_state_info: state to dispatch 8400 * 8401 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. 8402 * The caller must hold the RTNL lock. 8403 */ 8404 void netdev_lower_state_changed(struct net_device *lower_dev, 8405 void *lower_state_info) 8406 { 8407 struct netdev_notifier_changelowerstate_info changelowerstate_info = { 8408 .info.dev = lower_dev, 8409 }; 8410 8411 ASSERT_RTNL(); 8412 changelowerstate_info.lower_state_info = lower_state_info; 8413 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, 8414 &changelowerstate_info.info); 8415 } 8416 EXPORT_SYMBOL(netdev_lower_state_changed); 8417 8418 static void dev_change_rx_flags(struct net_device *dev, int flags) 8419 { 8420 const struct net_device_ops *ops = dev->netdev_ops; 8421 8422 if (ops->ndo_change_rx_flags) 8423 ops->ndo_change_rx_flags(dev, flags); 8424 } 8425 8426 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) 8427 { 8428 unsigned int old_flags = dev->flags; 8429 kuid_t uid; 8430 kgid_t gid; 8431 8432 ASSERT_RTNL(); 8433 8434 dev->flags |= IFF_PROMISC; 8435 dev->promiscuity += inc; 8436 if (dev->promiscuity == 0) { 8437 /* 8438 * Avoid overflow. 8439 * If inc causes overflow, untouch promisc and return error. 8440 */ 8441 if (inc < 0) 8442 dev->flags &= ~IFF_PROMISC; 8443 else { 8444 dev->promiscuity -= inc; 8445 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n", 8446 dev->name); 8447 return -EOVERFLOW; 8448 } 8449 } 8450 if (dev->flags != old_flags) { 8451 pr_info("device %s %s promiscuous mode\n", 8452 dev->name, 8453 dev->flags & IFF_PROMISC ? "entered" : "left"); 8454 if (audit_enabled) { 8455 current_uid_gid(&uid, &gid); 8456 audit_log(audit_context(), GFP_ATOMIC, 8457 AUDIT_ANOM_PROMISCUOUS, 8458 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", 8459 dev->name, (dev->flags & IFF_PROMISC), 8460 (old_flags & IFF_PROMISC), 8461 from_kuid(&init_user_ns, audit_get_loginuid(current)), 8462 from_kuid(&init_user_ns, uid), 8463 from_kgid(&init_user_ns, gid), 8464 audit_get_sessionid(current)); 8465 } 8466 8467 dev_change_rx_flags(dev, IFF_PROMISC); 8468 } 8469 if (notify) 8470 __dev_notify_flags(dev, old_flags, IFF_PROMISC); 8471 return 0; 8472 } 8473 8474 /** 8475 * dev_set_promiscuity - update promiscuity count on a device 8476 * @dev: device 8477 * @inc: modifier 8478 * 8479 * Add or remove promiscuity from a device. While the count in the device 8480 * remains above zero the interface remains promiscuous. Once it hits zero 8481 * the device reverts back to normal filtering operation. A negative inc 8482 * value is used to drop promiscuity on the device. 8483 * Return 0 if successful or a negative errno code on error. 8484 */ 8485 int dev_set_promiscuity(struct net_device *dev, int inc) 8486 { 8487 unsigned int old_flags = dev->flags; 8488 int err; 8489 8490 err = __dev_set_promiscuity(dev, inc, true); 8491 if (err < 0) 8492 return err; 8493 if (dev->flags != old_flags) 8494 dev_set_rx_mode(dev); 8495 return err; 8496 } 8497 EXPORT_SYMBOL(dev_set_promiscuity); 8498 8499 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) 8500 { 8501 unsigned int old_flags = dev->flags, old_gflags = dev->gflags; 8502 8503 ASSERT_RTNL(); 8504 8505 dev->flags |= IFF_ALLMULTI; 8506 dev->allmulti += inc; 8507 if (dev->allmulti == 0) { 8508 /* 8509 * Avoid overflow. 8510 * If inc causes overflow, untouch allmulti and return error. 8511 */ 8512 if (inc < 0) 8513 dev->flags &= ~IFF_ALLMULTI; 8514 else { 8515 dev->allmulti -= inc; 8516 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n", 8517 dev->name); 8518 return -EOVERFLOW; 8519 } 8520 } 8521 if (dev->flags ^ old_flags) { 8522 dev_change_rx_flags(dev, IFF_ALLMULTI); 8523 dev_set_rx_mode(dev); 8524 if (notify) 8525 __dev_notify_flags(dev, old_flags, 8526 dev->gflags ^ old_gflags); 8527 } 8528 return 0; 8529 } 8530 8531 /** 8532 * dev_set_allmulti - update allmulti count on a device 8533 * @dev: device 8534 * @inc: modifier 8535 * 8536 * Add or remove reception of all multicast frames to a device. While the 8537 * count in the device remains above zero the interface remains listening 8538 * to all interfaces. Once it hits zero the device reverts back to normal 8539 * filtering operation. A negative @inc value is used to drop the counter 8540 * when releasing a resource needing all multicasts. 8541 * Return 0 if successful or a negative errno code on error. 8542 */ 8543 8544 int dev_set_allmulti(struct net_device *dev, int inc) 8545 { 8546 return __dev_set_allmulti(dev, inc, true); 8547 } 8548 EXPORT_SYMBOL(dev_set_allmulti); 8549 8550 /* 8551 * Upload unicast and multicast address lists to device and 8552 * configure RX filtering. When the device doesn't support unicast 8553 * filtering it is put in promiscuous mode while unicast addresses 8554 * are present. 8555 */ 8556 void __dev_set_rx_mode(struct net_device *dev) 8557 { 8558 const struct net_device_ops *ops = dev->netdev_ops; 8559 8560 /* dev_open will call this function so the list will stay sane. */ 8561 if (!(dev->flags&IFF_UP)) 8562 return; 8563 8564 if (!netif_device_present(dev)) 8565 return; 8566 8567 if (!(dev->priv_flags & IFF_UNICAST_FLT)) { 8568 /* Unicast addresses changes may only happen under the rtnl, 8569 * therefore calling __dev_set_promiscuity here is safe. 8570 */ 8571 if (!netdev_uc_empty(dev) && !dev->uc_promisc) { 8572 __dev_set_promiscuity(dev, 1, false); 8573 dev->uc_promisc = true; 8574 } else if (netdev_uc_empty(dev) && dev->uc_promisc) { 8575 __dev_set_promiscuity(dev, -1, false); 8576 dev->uc_promisc = false; 8577 } 8578 } 8579 8580 if (ops->ndo_set_rx_mode) 8581 ops->ndo_set_rx_mode(dev); 8582 } 8583 8584 void dev_set_rx_mode(struct net_device *dev) 8585 { 8586 netif_addr_lock_bh(dev); 8587 __dev_set_rx_mode(dev); 8588 netif_addr_unlock_bh(dev); 8589 } 8590 8591 /** 8592 * dev_get_flags - get flags reported to userspace 8593 * @dev: device 8594 * 8595 * Get the combination of flag bits exported through APIs to userspace. 8596 */ 8597 unsigned int dev_get_flags(const struct net_device *dev) 8598 { 8599 unsigned int flags; 8600 8601 flags = (dev->flags & ~(IFF_PROMISC | 8602 IFF_ALLMULTI | 8603 IFF_RUNNING | 8604 IFF_LOWER_UP | 8605 IFF_DORMANT)) | 8606 (dev->gflags & (IFF_PROMISC | 8607 IFF_ALLMULTI)); 8608 8609 if (netif_running(dev)) { 8610 if (netif_oper_up(dev)) 8611 flags |= IFF_RUNNING; 8612 if (netif_carrier_ok(dev)) 8613 flags |= IFF_LOWER_UP; 8614 if (netif_dormant(dev)) 8615 flags |= IFF_DORMANT; 8616 } 8617 8618 return flags; 8619 } 8620 EXPORT_SYMBOL(dev_get_flags); 8621 8622 int __dev_change_flags(struct net_device *dev, unsigned int flags, 8623 struct netlink_ext_ack *extack) 8624 { 8625 unsigned int old_flags = dev->flags; 8626 int ret; 8627 8628 ASSERT_RTNL(); 8629 8630 /* 8631 * Set the flags on our device. 8632 */ 8633 8634 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | 8635 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | 8636 IFF_AUTOMEDIA)) | 8637 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | 8638 IFF_ALLMULTI)); 8639 8640 /* 8641 * Load in the correct multicast list now the flags have changed. 8642 */ 8643 8644 if ((old_flags ^ flags) & IFF_MULTICAST) 8645 dev_change_rx_flags(dev, IFF_MULTICAST); 8646 8647 dev_set_rx_mode(dev); 8648 8649 /* 8650 * Have we downed the interface. We handle IFF_UP ourselves 8651 * according to user attempts to set it, rather than blindly 8652 * setting it. 8653 */ 8654 8655 ret = 0; 8656 if ((old_flags ^ flags) & IFF_UP) { 8657 if (old_flags & IFF_UP) 8658 __dev_close(dev); 8659 else 8660 ret = __dev_open(dev, extack); 8661 } 8662 8663 if ((flags ^ dev->gflags) & IFF_PROMISC) { 8664 int inc = (flags & IFF_PROMISC) ? 1 : -1; 8665 unsigned int old_flags = dev->flags; 8666 8667 dev->gflags ^= IFF_PROMISC; 8668 8669 if (__dev_set_promiscuity(dev, inc, false) >= 0) 8670 if (dev->flags != old_flags) 8671 dev_set_rx_mode(dev); 8672 } 8673 8674 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI 8675 * is important. Some (broken) drivers set IFF_PROMISC, when 8676 * IFF_ALLMULTI is requested not asking us and not reporting. 8677 */ 8678 if ((flags ^ dev->gflags) & IFF_ALLMULTI) { 8679 int inc = (flags & IFF_ALLMULTI) ? 1 : -1; 8680 8681 dev->gflags ^= IFF_ALLMULTI; 8682 __dev_set_allmulti(dev, inc, false); 8683 } 8684 8685 return ret; 8686 } 8687 8688 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, 8689 unsigned int gchanges) 8690 { 8691 unsigned int changes = dev->flags ^ old_flags; 8692 8693 if (gchanges) 8694 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC); 8695 8696 if (changes & IFF_UP) { 8697 if (dev->flags & IFF_UP) 8698 call_netdevice_notifiers(NETDEV_UP, dev); 8699 else 8700 call_netdevice_notifiers(NETDEV_DOWN, dev); 8701 } 8702 8703 if (dev->flags & IFF_UP && 8704 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { 8705 struct netdev_notifier_change_info change_info = { 8706 .info = { 8707 .dev = dev, 8708 }, 8709 .flags_changed = changes, 8710 }; 8711 8712 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info); 8713 } 8714 } 8715 8716 /** 8717 * dev_change_flags - change device settings 8718 * @dev: device 8719 * @flags: device state flags 8720 * @extack: netlink extended ack 8721 * 8722 * Change settings on device based state flags. The flags are 8723 * in the userspace exported format. 8724 */ 8725 int dev_change_flags(struct net_device *dev, unsigned int flags, 8726 struct netlink_ext_ack *extack) 8727 { 8728 int ret; 8729 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; 8730 8731 ret = __dev_change_flags(dev, flags, extack); 8732 if (ret < 0) 8733 return ret; 8734 8735 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); 8736 __dev_notify_flags(dev, old_flags, changes); 8737 return ret; 8738 } 8739 EXPORT_SYMBOL(dev_change_flags); 8740 8741 int __dev_set_mtu(struct net_device *dev, int new_mtu) 8742 { 8743 const struct net_device_ops *ops = dev->netdev_ops; 8744 8745 if (ops->ndo_change_mtu) 8746 return ops->ndo_change_mtu(dev, new_mtu); 8747 8748 /* Pairs with all the lockless reads of dev->mtu in the stack */ 8749 WRITE_ONCE(dev->mtu, new_mtu); 8750 return 0; 8751 } 8752 EXPORT_SYMBOL(__dev_set_mtu); 8753 8754 int dev_validate_mtu(struct net_device *dev, int new_mtu, 8755 struct netlink_ext_ack *extack) 8756 { 8757 /* MTU must be positive, and in range */ 8758 if (new_mtu < 0 || new_mtu < dev->min_mtu) { 8759 NL_SET_ERR_MSG(extack, "mtu less than device minimum"); 8760 return -EINVAL; 8761 } 8762 8763 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { 8764 NL_SET_ERR_MSG(extack, "mtu greater than device maximum"); 8765 return -EINVAL; 8766 } 8767 return 0; 8768 } 8769 8770 /** 8771 * dev_set_mtu_ext - Change maximum transfer unit 8772 * @dev: device 8773 * @new_mtu: new transfer unit 8774 * @extack: netlink extended ack 8775 * 8776 * Change the maximum transfer size of the network device. 8777 */ 8778 int dev_set_mtu_ext(struct net_device *dev, int new_mtu, 8779 struct netlink_ext_ack *extack) 8780 { 8781 int err, orig_mtu; 8782 8783 if (new_mtu == dev->mtu) 8784 return 0; 8785 8786 err = dev_validate_mtu(dev, new_mtu, extack); 8787 if (err) 8788 return err; 8789 8790 if (!netif_device_present(dev)) 8791 return -ENODEV; 8792 8793 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); 8794 err = notifier_to_errno(err); 8795 if (err) 8796 return err; 8797 8798 orig_mtu = dev->mtu; 8799 err = __dev_set_mtu(dev, new_mtu); 8800 8801 if (!err) { 8802 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8803 orig_mtu); 8804 err = notifier_to_errno(err); 8805 if (err) { 8806 /* setting mtu back and notifying everyone again, 8807 * so that they have a chance to revert changes. 8808 */ 8809 __dev_set_mtu(dev, orig_mtu); 8810 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8811 new_mtu); 8812 } 8813 } 8814 return err; 8815 } 8816 8817 int dev_set_mtu(struct net_device *dev, int new_mtu) 8818 { 8819 struct netlink_ext_ack extack; 8820 int err; 8821 8822 memset(&extack, 0, sizeof(extack)); 8823 err = dev_set_mtu_ext(dev, new_mtu, &extack); 8824 if (err && extack._msg) 8825 net_err_ratelimited("%s: %s\n", dev->name, extack._msg); 8826 return err; 8827 } 8828 EXPORT_SYMBOL(dev_set_mtu); 8829 8830 /** 8831 * dev_change_tx_queue_len - Change TX queue length of a netdevice 8832 * @dev: device 8833 * @new_len: new tx queue length 8834 */ 8835 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len) 8836 { 8837 unsigned int orig_len = dev->tx_queue_len; 8838 int res; 8839 8840 if (new_len != (unsigned int)new_len) 8841 return -ERANGE; 8842 8843 if (new_len != orig_len) { 8844 dev->tx_queue_len = new_len; 8845 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev); 8846 res = notifier_to_errno(res); 8847 if (res) 8848 goto err_rollback; 8849 res = dev_qdisc_change_tx_queue_len(dev); 8850 if (res) 8851 goto err_rollback; 8852 } 8853 8854 return 0; 8855 8856 err_rollback: 8857 netdev_err(dev, "refused to change device tx_queue_len\n"); 8858 dev->tx_queue_len = orig_len; 8859 return res; 8860 } 8861 8862 /** 8863 * dev_set_group - Change group this device belongs to 8864 * @dev: device 8865 * @new_group: group this device should belong to 8866 */ 8867 void dev_set_group(struct net_device *dev, int new_group) 8868 { 8869 dev->group = new_group; 8870 } 8871 EXPORT_SYMBOL(dev_set_group); 8872 8873 /** 8874 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR. 8875 * @dev: device 8876 * @addr: new address 8877 * @extack: netlink extended ack 8878 */ 8879 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr, 8880 struct netlink_ext_ack *extack) 8881 { 8882 struct netdev_notifier_pre_changeaddr_info info = { 8883 .info.dev = dev, 8884 .info.extack = extack, 8885 .dev_addr = addr, 8886 }; 8887 int rc; 8888 8889 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info); 8890 return notifier_to_errno(rc); 8891 } 8892 EXPORT_SYMBOL(dev_pre_changeaddr_notify); 8893 8894 /** 8895 * dev_set_mac_address - Change Media Access Control Address 8896 * @dev: device 8897 * @sa: new address 8898 * @extack: netlink extended ack 8899 * 8900 * Change the hardware (MAC) address of the device 8901 */ 8902 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa, 8903 struct netlink_ext_ack *extack) 8904 { 8905 const struct net_device_ops *ops = dev->netdev_ops; 8906 int err; 8907 8908 if (!ops->ndo_set_mac_address) 8909 return -EOPNOTSUPP; 8910 if (sa->sa_family != dev->type) 8911 return -EINVAL; 8912 if (!netif_device_present(dev)) 8913 return -ENODEV; 8914 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack); 8915 if (err) 8916 return err; 8917 err = ops->ndo_set_mac_address(dev, sa); 8918 if (err) 8919 return err; 8920 dev->addr_assign_type = NET_ADDR_SET; 8921 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); 8922 add_device_randomness(dev->dev_addr, dev->addr_len); 8923 return 0; 8924 } 8925 EXPORT_SYMBOL(dev_set_mac_address); 8926 8927 static DECLARE_RWSEM(dev_addr_sem); 8928 8929 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa, 8930 struct netlink_ext_ack *extack) 8931 { 8932 int ret; 8933 8934 down_write(&dev_addr_sem); 8935 ret = dev_set_mac_address(dev, sa, extack); 8936 up_write(&dev_addr_sem); 8937 return ret; 8938 } 8939 EXPORT_SYMBOL(dev_set_mac_address_user); 8940 8941 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name) 8942 { 8943 size_t size = sizeof(sa->sa_data); 8944 struct net_device *dev; 8945 int ret = 0; 8946 8947 down_read(&dev_addr_sem); 8948 rcu_read_lock(); 8949 8950 dev = dev_get_by_name_rcu(net, dev_name); 8951 if (!dev) { 8952 ret = -ENODEV; 8953 goto unlock; 8954 } 8955 if (!dev->addr_len) 8956 memset(sa->sa_data, 0, size); 8957 else 8958 memcpy(sa->sa_data, dev->dev_addr, 8959 min_t(size_t, size, dev->addr_len)); 8960 sa->sa_family = dev->type; 8961 8962 unlock: 8963 rcu_read_unlock(); 8964 up_read(&dev_addr_sem); 8965 return ret; 8966 } 8967 EXPORT_SYMBOL(dev_get_mac_address); 8968 8969 /** 8970 * dev_change_carrier - Change device carrier 8971 * @dev: device 8972 * @new_carrier: new value 8973 * 8974 * Change device carrier 8975 */ 8976 int dev_change_carrier(struct net_device *dev, bool new_carrier) 8977 { 8978 const struct net_device_ops *ops = dev->netdev_ops; 8979 8980 if (!ops->ndo_change_carrier) 8981 return -EOPNOTSUPP; 8982 if (!netif_device_present(dev)) 8983 return -ENODEV; 8984 return ops->ndo_change_carrier(dev, new_carrier); 8985 } 8986 EXPORT_SYMBOL(dev_change_carrier); 8987 8988 /** 8989 * dev_get_phys_port_id - Get device physical port ID 8990 * @dev: device 8991 * @ppid: port ID 8992 * 8993 * Get device physical port ID 8994 */ 8995 int dev_get_phys_port_id(struct net_device *dev, 8996 struct netdev_phys_item_id *ppid) 8997 { 8998 const struct net_device_ops *ops = dev->netdev_ops; 8999 9000 if (!ops->ndo_get_phys_port_id) 9001 return -EOPNOTSUPP; 9002 return ops->ndo_get_phys_port_id(dev, ppid); 9003 } 9004 EXPORT_SYMBOL(dev_get_phys_port_id); 9005 9006 /** 9007 * dev_get_phys_port_name - Get device physical port name 9008 * @dev: device 9009 * @name: port name 9010 * @len: limit of bytes to copy to name 9011 * 9012 * Get device physical port name 9013 */ 9014 int dev_get_phys_port_name(struct net_device *dev, 9015 char *name, size_t len) 9016 { 9017 const struct net_device_ops *ops = dev->netdev_ops; 9018 int err; 9019 9020 if (ops->ndo_get_phys_port_name) { 9021 err = ops->ndo_get_phys_port_name(dev, name, len); 9022 if (err != -EOPNOTSUPP) 9023 return err; 9024 } 9025 return devlink_compat_phys_port_name_get(dev, name, len); 9026 } 9027 EXPORT_SYMBOL(dev_get_phys_port_name); 9028 9029 /** 9030 * dev_get_port_parent_id - Get the device's port parent identifier 9031 * @dev: network device 9032 * @ppid: pointer to a storage for the port's parent identifier 9033 * @recurse: allow/disallow recursion to lower devices 9034 * 9035 * Get the devices's port parent identifier 9036 */ 9037 int dev_get_port_parent_id(struct net_device *dev, 9038 struct netdev_phys_item_id *ppid, 9039 bool recurse) 9040 { 9041 const struct net_device_ops *ops = dev->netdev_ops; 9042 struct netdev_phys_item_id first = { }; 9043 struct net_device *lower_dev; 9044 struct list_head *iter; 9045 int err; 9046 9047 if (ops->ndo_get_port_parent_id) { 9048 err = ops->ndo_get_port_parent_id(dev, ppid); 9049 if (err != -EOPNOTSUPP) 9050 return err; 9051 } 9052 9053 err = devlink_compat_switch_id_get(dev, ppid); 9054 if (!err || err != -EOPNOTSUPP) 9055 return err; 9056 9057 if (!recurse) 9058 return -EOPNOTSUPP; 9059 9060 netdev_for_each_lower_dev(dev, lower_dev, iter) { 9061 err = dev_get_port_parent_id(lower_dev, ppid, recurse); 9062 if (err) 9063 break; 9064 if (!first.id_len) 9065 first = *ppid; 9066 else if (memcmp(&first, ppid, sizeof(*ppid))) 9067 return -EOPNOTSUPP; 9068 } 9069 9070 return err; 9071 } 9072 EXPORT_SYMBOL(dev_get_port_parent_id); 9073 9074 /** 9075 * netdev_port_same_parent_id - Indicate if two network devices have 9076 * the same port parent identifier 9077 * @a: first network device 9078 * @b: second network device 9079 */ 9080 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b) 9081 { 9082 struct netdev_phys_item_id a_id = { }; 9083 struct netdev_phys_item_id b_id = { }; 9084 9085 if (dev_get_port_parent_id(a, &a_id, true) || 9086 dev_get_port_parent_id(b, &b_id, true)) 9087 return false; 9088 9089 return netdev_phys_item_id_same(&a_id, &b_id); 9090 } 9091 EXPORT_SYMBOL(netdev_port_same_parent_id); 9092 9093 /** 9094 * dev_change_proto_down - update protocol port state information 9095 * @dev: device 9096 * @proto_down: new value 9097 * 9098 * This info can be used by switch drivers to set the phys state of the 9099 * port. 9100 */ 9101 int dev_change_proto_down(struct net_device *dev, bool proto_down) 9102 { 9103 const struct net_device_ops *ops = dev->netdev_ops; 9104 9105 if (!ops->ndo_change_proto_down) 9106 return -EOPNOTSUPP; 9107 if (!netif_device_present(dev)) 9108 return -ENODEV; 9109 return ops->ndo_change_proto_down(dev, proto_down); 9110 } 9111 EXPORT_SYMBOL(dev_change_proto_down); 9112 9113 /** 9114 * dev_change_proto_down_generic - generic implementation for 9115 * ndo_change_proto_down that sets carrier according to 9116 * proto_down. 9117 * 9118 * @dev: device 9119 * @proto_down: new value 9120 */ 9121 int dev_change_proto_down_generic(struct net_device *dev, bool proto_down) 9122 { 9123 if (proto_down) 9124 netif_carrier_off(dev); 9125 else 9126 netif_carrier_on(dev); 9127 dev->proto_down = proto_down; 9128 return 0; 9129 } 9130 EXPORT_SYMBOL(dev_change_proto_down_generic); 9131 9132 /** 9133 * dev_change_proto_down_reason - proto down reason 9134 * 9135 * @dev: device 9136 * @mask: proto down mask 9137 * @value: proto down value 9138 */ 9139 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask, 9140 u32 value) 9141 { 9142 int b; 9143 9144 if (!mask) { 9145 dev->proto_down_reason = value; 9146 } else { 9147 for_each_set_bit(b, &mask, 32) { 9148 if (value & (1 << b)) 9149 dev->proto_down_reason |= BIT(b); 9150 else 9151 dev->proto_down_reason &= ~BIT(b); 9152 } 9153 } 9154 } 9155 EXPORT_SYMBOL(dev_change_proto_down_reason); 9156 9157 struct bpf_xdp_link { 9158 struct bpf_link link; 9159 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */ 9160 int flags; 9161 }; 9162 9163 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags) 9164 { 9165 if (flags & XDP_FLAGS_HW_MODE) 9166 return XDP_MODE_HW; 9167 if (flags & XDP_FLAGS_DRV_MODE) 9168 return XDP_MODE_DRV; 9169 if (flags & XDP_FLAGS_SKB_MODE) 9170 return XDP_MODE_SKB; 9171 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB; 9172 } 9173 9174 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode) 9175 { 9176 switch (mode) { 9177 case XDP_MODE_SKB: 9178 return generic_xdp_install; 9179 case XDP_MODE_DRV: 9180 case XDP_MODE_HW: 9181 return dev->netdev_ops->ndo_bpf; 9182 default: 9183 return NULL; 9184 } 9185 } 9186 9187 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev, 9188 enum bpf_xdp_mode mode) 9189 { 9190 return dev->xdp_state[mode].link; 9191 } 9192 9193 static struct bpf_prog *dev_xdp_prog(struct net_device *dev, 9194 enum bpf_xdp_mode mode) 9195 { 9196 struct bpf_xdp_link *link = dev_xdp_link(dev, mode); 9197 9198 if (link) 9199 return link->link.prog; 9200 return dev->xdp_state[mode].prog; 9201 } 9202 9203 static u8 dev_xdp_prog_count(struct net_device *dev) 9204 { 9205 u8 count = 0; 9206 int i; 9207 9208 for (i = 0; i < __MAX_XDP_MODE; i++) 9209 if (dev->xdp_state[i].prog || dev->xdp_state[i].link) 9210 count++; 9211 return count; 9212 } 9213 9214 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode) 9215 { 9216 struct bpf_prog *prog = dev_xdp_prog(dev, mode); 9217 9218 return prog ? prog->aux->id : 0; 9219 } 9220 9221 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode, 9222 struct bpf_xdp_link *link) 9223 { 9224 dev->xdp_state[mode].link = link; 9225 dev->xdp_state[mode].prog = NULL; 9226 } 9227 9228 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode, 9229 struct bpf_prog *prog) 9230 { 9231 dev->xdp_state[mode].link = NULL; 9232 dev->xdp_state[mode].prog = prog; 9233 } 9234 9235 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode, 9236 bpf_op_t bpf_op, struct netlink_ext_ack *extack, 9237 u32 flags, struct bpf_prog *prog) 9238 { 9239 struct netdev_bpf xdp; 9240 int err; 9241 9242 memset(&xdp, 0, sizeof(xdp)); 9243 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG; 9244 xdp.extack = extack; 9245 xdp.flags = flags; 9246 xdp.prog = prog; 9247 9248 /* Drivers assume refcnt is already incremented (i.e, prog pointer is 9249 * "moved" into driver), so they don't increment it on their own, but 9250 * they do decrement refcnt when program is detached or replaced. 9251 * Given net_device also owns link/prog, we need to bump refcnt here 9252 * to prevent drivers from underflowing it. 9253 */ 9254 if (prog) 9255 bpf_prog_inc(prog); 9256 err = bpf_op(dev, &xdp); 9257 if (err) { 9258 if (prog) 9259 bpf_prog_put(prog); 9260 return err; 9261 } 9262 9263 if (mode != XDP_MODE_HW) 9264 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog); 9265 9266 return 0; 9267 } 9268 9269 static void dev_xdp_uninstall(struct net_device *dev) 9270 { 9271 struct bpf_xdp_link *link; 9272 struct bpf_prog *prog; 9273 enum bpf_xdp_mode mode; 9274 bpf_op_t bpf_op; 9275 9276 ASSERT_RTNL(); 9277 9278 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) { 9279 prog = dev_xdp_prog(dev, mode); 9280 if (!prog) 9281 continue; 9282 9283 bpf_op = dev_xdp_bpf_op(dev, mode); 9284 if (!bpf_op) 9285 continue; 9286 9287 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9288 9289 /* auto-detach link from net device */ 9290 link = dev_xdp_link(dev, mode); 9291 if (link) 9292 link->dev = NULL; 9293 else 9294 bpf_prog_put(prog); 9295 9296 dev_xdp_set_link(dev, mode, NULL); 9297 } 9298 } 9299 9300 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack, 9301 struct bpf_xdp_link *link, struct bpf_prog *new_prog, 9302 struct bpf_prog *old_prog, u32 flags) 9303 { 9304 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES); 9305 struct bpf_prog *cur_prog; 9306 enum bpf_xdp_mode mode; 9307 bpf_op_t bpf_op; 9308 int err; 9309 9310 ASSERT_RTNL(); 9311 9312 /* either link or prog attachment, never both */ 9313 if (link && (new_prog || old_prog)) 9314 return -EINVAL; 9315 /* link supports only XDP mode flags */ 9316 if (link && (flags & ~XDP_FLAGS_MODES)) { 9317 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment"); 9318 return -EINVAL; 9319 } 9320 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */ 9321 if (num_modes > 1) { 9322 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set"); 9323 return -EINVAL; 9324 } 9325 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */ 9326 if (!num_modes && dev_xdp_prog_count(dev) > 1) { 9327 NL_SET_ERR_MSG(extack, 9328 "More than one program loaded, unset mode is ambiguous"); 9329 return -EINVAL; 9330 } 9331 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */ 9332 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) { 9333 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified"); 9334 return -EINVAL; 9335 } 9336 9337 mode = dev_xdp_mode(dev, flags); 9338 /* can't replace attached link */ 9339 if (dev_xdp_link(dev, mode)) { 9340 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link"); 9341 return -EBUSY; 9342 } 9343 9344 cur_prog = dev_xdp_prog(dev, mode); 9345 /* can't replace attached prog with link */ 9346 if (link && cur_prog) { 9347 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link"); 9348 return -EBUSY; 9349 } 9350 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) { 9351 NL_SET_ERR_MSG(extack, "Active program does not match expected"); 9352 return -EEXIST; 9353 } 9354 9355 /* put effective new program into new_prog */ 9356 if (link) 9357 new_prog = link->link.prog; 9358 9359 if (new_prog) { 9360 bool offload = mode == XDP_MODE_HW; 9361 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB 9362 ? XDP_MODE_DRV : XDP_MODE_SKB; 9363 9364 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) { 9365 NL_SET_ERR_MSG(extack, "XDP program already attached"); 9366 return -EBUSY; 9367 } 9368 if (!offload && dev_xdp_prog(dev, other_mode)) { 9369 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time"); 9370 return -EEXIST; 9371 } 9372 if (!offload && bpf_prog_is_dev_bound(new_prog->aux)) { 9373 NL_SET_ERR_MSG(extack, "Using device-bound program without HW_MODE flag is not supported"); 9374 return -EINVAL; 9375 } 9376 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) { 9377 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device"); 9378 return -EINVAL; 9379 } 9380 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) { 9381 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device"); 9382 return -EINVAL; 9383 } 9384 } 9385 9386 /* don't call drivers if the effective program didn't change */ 9387 if (new_prog != cur_prog) { 9388 bpf_op = dev_xdp_bpf_op(dev, mode); 9389 if (!bpf_op) { 9390 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode"); 9391 return -EOPNOTSUPP; 9392 } 9393 9394 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog); 9395 if (err) 9396 return err; 9397 } 9398 9399 if (link) 9400 dev_xdp_set_link(dev, mode, link); 9401 else 9402 dev_xdp_set_prog(dev, mode, new_prog); 9403 if (cur_prog) 9404 bpf_prog_put(cur_prog); 9405 9406 return 0; 9407 } 9408 9409 static int dev_xdp_attach_link(struct net_device *dev, 9410 struct netlink_ext_ack *extack, 9411 struct bpf_xdp_link *link) 9412 { 9413 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags); 9414 } 9415 9416 static int dev_xdp_detach_link(struct net_device *dev, 9417 struct netlink_ext_ack *extack, 9418 struct bpf_xdp_link *link) 9419 { 9420 enum bpf_xdp_mode mode; 9421 bpf_op_t bpf_op; 9422 9423 ASSERT_RTNL(); 9424 9425 mode = dev_xdp_mode(dev, link->flags); 9426 if (dev_xdp_link(dev, mode) != link) 9427 return -EINVAL; 9428 9429 bpf_op = dev_xdp_bpf_op(dev, mode); 9430 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9431 dev_xdp_set_link(dev, mode, NULL); 9432 return 0; 9433 } 9434 9435 static void bpf_xdp_link_release(struct bpf_link *link) 9436 { 9437 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9438 9439 rtnl_lock(); 9440 9441 /* if racing with net_device's tear down, xdp_link->dev might be 9442 * already NULL, in which case link was already auto-detached 9443 */ 9444 if (xdp_link->dev) { 9445 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link)); 9446 xdp_link->dev = NULL; 9447 } 9448 9449 rtnl_unlock(); 9450 } 9451 9452 static int bpf_xdp_link_detach(struct bpf_link *link) 9453 { 9454 bpf_xdp_link_release(link); 9455 return 0; 9456 } 9457 9458 static void bpf_xdp_link_dealloc(struct bpf_link *link) 9459 { 9460 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9461 9462 kfree(xdp_link); 9463 } 9464 9465 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link, 9466 struct seq_file *seq) 9467 { 9468 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9469 u32 ifindex = 0; 9470 9471 rtnl_lock(); 9472 if (xdp_link->dev) 9473 ifindex = xdp_link->dev->ifindex; 9474 rtnl_unlock(); 9475 9476 seq_printf(seq, "ifindex:\t%u\n", ifindex); 9477 } 9478 9479 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link, 9480 struct bpf_link_info *info) 9481 { 9482 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9483 u32 ifindex = 0; 9484 9485 rtnl_lock(); 9486 if (xdp_link->dev) 9487 ifindex = xdp_link->dev->ifindex; 9488 rtnl_unlock(); 9489 9490 info->xdp.ifindex = ifindex; 9491 return 0; 9492 } 9493 9494 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog, 9495 struct bpf_prog *old_prog) 9496 { 9497 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9498 enum bpf_xdp_mode mode; 9499 bpf_op_t bpf_op; 9500 int err = 0; 9501 9502 rtnl_lock(); 9503 9504 /* link might have been auto-released already, so fail */ 9505 if (!xdp_link->dev) { 9506 err = -ENOLINK; 9507 goto out_unlock; 9508 } 9509 9510 if (old_prog && link->prog != old_prog) { 9511 err = -EPERM; 9512 goto out_unlock; 9513 } 9514 old_prog = link->prog; 9515 if (old_prog == new_prog) { 9516 /* no-op, don't disturb drivers */ 9517 bpf_prog_put(new_prog); 9518 goto out_unlock; 9519 } 9520 9521 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags); 9522 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode); 9523 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL, 9524 xdp_link->flags, new_prog); 9525 if (err) 9526 goto out_unlock; 9527 9528 old_prog = xchg(&link->prog, new_prog); 9529 bpf_prog_put(old_prog); 9530 9531 out_unlock: 9532 rtnl_unlock(); 9533 return err; 9534 } 9535 9536 static const struct bpf_link_ops bpf_xdp_link_lops = { 9537 .release = bpf_xdp_link_release, 9538 .dealloc = bpf_xdp_link_dealloc, 9539 .detach = bpf_xdp_link_detach, 9540 .show_fdinfo = bpf_xdp_link_show_fdinfo, 9541 .fill_link_info = bpf_xdp_link_fill_link_info, 9542 .update_prog = bpf_xdp_link_update, 9543 }; 9544 9545 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog) 9546 { 9547 struct net *net = current->nsproxy->net_ns; 9548 struct bpf_link_primer link_primer; 9549 struct bpf_xdp_link *link; 9550 struct net_device *dev; 9551 int err, fd; 9552 9553 dev = dev_get_by_index(net, attr->link_create.target_ifindex); 9554 if (!dev) 9555 return -EINVAL; 9556 9557 link = kzalloc(sizeof(*link), GFP_USER); 9558 if (!link) { 9559 err = -ENOMEM; 9560 goto out_put_dev; 9561 } 9562 9563 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog); 9564 link->dev = dev; 9565 link->flags = attr->link_create.flags; 9566 9567 err = bpf_link_prime(&link->link, &link_primer); 9568 if (err) { 9569 kfree(link); 9570 goto out_put_dev; 9571 } 9572 9573 rtnl_lock(); 9574 err = dev_xdp_attach_link(dev, NULL, link); 9575 rtnl_unlock(); 9576 9577 if (err) { 9578 bpf_link_cleanup(&link_primer); 9579 goto out_put_dev; 9580 } 9581 9582 fd = bpf_link_settle(&link_primer); 9583 /* link itself doesn't hold dev's refcnt to not complicate shutdown */ 9584 dev_put(dev); 9585 return fd; 9586 9587 out_put_dev: 9588 dev_put(dev); 9589 return err; 9590 } 9591 9592 /** 9593 * dev_change_xdp_fd - set or clear a bpf program for a device rx path 9594 * @dev: device 9595 * @extack: netlink extended ack 9596 * @fd: new program fd or negative value to clear 9597 * @expected_fd: old program fd that userspace expects to replace or clear 9598 * @flags: xdp-related flags 9599 * 9600 * Set or clear a bpf program for a device 9601 */ 9602 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, 9603 int fd, int expected_fd, u32 flags) 9604 { 9605 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags); 9606 struct bpf_prog *new_prog = NULL, *old_prog = NULL; 9607 int err; 9608 9609 ASSERT_RTNL(); 9610 9611 if (fd >= 0) { 9612 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP, 9613 mode != XDP_MODE_SKB); 9614 if (IS_ERR(new_prog)) 9615 return PTR_ERR(new_prog); 9616 } 9617 9618 if (expected_fd >= 0) { 9619 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP, 9620 mode != XDP_MODE_SKB); 9621 if (IS_ERR(old_prog)) { 9622 err = PTR_ERR(old_prog); 9623 old_prog = NULL; 9624 goto err_out; 9625 } 9626 } 9627 9628 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags); 9629 9630 err_out: 9631 if (err && new_prog) 9632 bpf_prog_put(new_prog); 9633 if (old_prog) 9634 bpf_prog_put(old_prog); 9635 return err; 9636 } 9637 9638 /** 9639 * dev_new_index - allocate an ifindex 9640 * @net: the applicable net namespace 9641 * 9642 * Returns a suitable unique value for a new device interface 9643 * number. The caller must hold the rtnl semaphore or the 9644 * dev_base_lock to be sure it remains unique. 9645 */ 9646 static int dev_new_index(struct net *net) 9647 { 9648 int ifindex = net->ifindex; 9649 9650 for (;;) { 9651 if (++ifindex <= 0) 9652 ifindex = 1; 9653 if (!__dev_get_by_index(net, ifindex)) 9654 return net->ifindex = ifindex; 9655 } 9656 } 9657 9658 /* Delayed registration/unregisteration */ 9659 static LIST_HEAD(net_todo_list); 9660 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); 9661 9662 static void net_set_todo(struct net_device *dev) 9663 { 9664 list_add_tail(&dev->todo_list, &net_todo_list); 9665 dev_net(dev)->dev_unreg_count++; 9666 } 9667 9668 static netdev_features_t netdev_sync_upper_features(struct net_device *lower, 9669 struct net_device *upper, netdev_features_t features) 9670 { 9671 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9672 netdev_features_t feature; 9673 int feature_bit; 9674 9675 for_each_netdev_feature(upper_disables, feature_bit) { 9676 feature = __NETIF_F_BIT(feature_bit); 9677 if (!(upper->wanted_features & feature) 9678 && (features & feature)) { 9679 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", 9680 &feature, upper->name); 9681 features &= ~feature; 9682 } 9683 } 9684 9685 return features; 9686 } 9687 9688 static void netdev_sync_lower_features(struct net_device *upper, 9689 struct net_device *lower, netdev_features_t features) 9690 { 9691 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9692 netdev_features_t feature; 9693 int feature_bit; 9694 9695 for_each_netdev_feature(upper_disables, feature_bit) { 9696 feature = __NETIF_F_BIT(feature_bit); 9697 if (!(features & feature) && (lower->features & feature)) { 9698 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", 9699 &feature, lower->name); 9700 lower->wanted_features &= ~feature; 9701 __netdev_update_features(lower); 9702 9703 if (unlikely(lower->features & feature)) 9704 netdev_WARN(upper, "failed to disable %pNF on %s!\n", 9705 &feature, lower->name); 9706 else 9707 netdev_features_change(lower); 9708 } 9709 } 9710 } 9711 9712 static netdev_features_t netdev_fix_features(struct net_device *dev, 9713 netdev_features_t features) 9714 { 9715 /* Fix illegal checksum combinations */ 9716 if ((features & NETIF_F_HW_CSUM) && 9717 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { 9718 netdev_warn(dev, "mixed HW and IP checksum settings.\n"); 9719 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); 9720 } 9721 9722 /* TSO requires that SG is present as well. */ 9723 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { 9724 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); 9725 features &= ~NETIF_F_ALL_TSO; 9726 } 9727 9728 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && 9729 !(features & NETIF_F_IP_CSUM)) { 9730 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); 9731 features &= ~NETIF_F_TSO; 9732 features &= ~NETIF_F_TSO_ECN; 9733 } 9734 9735 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && 9736 !(features & NETIF_F_IPV6_CSUM)) { 9737 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); 9738 features &= ~NETIF_F_TSO6; 9739 } 9740 9741 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ 9742 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) 9743 features &= ~NETIF_F_TSO_MANGLEID; 9744 9745 /* TSO ECN requires that TSO is present as well. */ 9746 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) 9747 features &= ~NETIF_F_TSO_ECN; 9748 9749 /* Software GSO depends on SG. */ 9750 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { 9751 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); 9752 features &= ~NETIF_F_GSO; 9753 } 9754 9755 /* GSO partial features require GSO partial be set */ 9756 if ((features & dev->gso_partial_features) && 9757 !(features & NETIF_F_GSO_PARTIAL)) { 9758 netdev_dbg(dev, 9759 "Dropping partially supported GSO features since no GSO partial.\n"); 9760 features &= ~dev->gso_partial_features; 9761 } 9762 9763 if (!(features & NETIF_F_RXCSUM)) { 9764 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet 9765 * successfully merged by hardware must also have the 9766 * checksum verified by hardware. If the user does not 9767 * want to enable RXCSUM, logically, we should disable GRO_HW. 9768 */ 9769 if (features & NETIF_F_GRO_HW) { 9770 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n"); 9771 features &= ~NETIF_F_GRO_HW; 9772 } 9773 } 9774 9775 /* LRO/HW-GRO features cannot be combined with RX-FCS */ 9776 if (features & NETIF_F_RXFCS) { 9777 if (features & NETIF_F_LRO) { 9778 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n"); 9779 features &= ~NETIF_F_LRO; 9780 } 9781 9782 if (features & NETIF_F_GRO_HW) { 9783 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n"); 9784 features &= ~NETIF_F_GRO_HW; 9785 } 9786 } 9787 9788 if (features & NETIF_F_HW_TLS_TX) { 9789 bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) == 9790 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM); 9791 bool hw_csum = features & NETIF_F_HW_CSUM; 9792 9793 if (!ip_csum && !hw_csum) { 9794 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n"); 9795 features &= ~NETIF_F_HW_TLS_TX; 9796 } 9797 } 9798 9799 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) { 9800 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n"); 9801 features &= ~NETIF_F_HW_TLS_RX; 9802 } 9803 9804 return features; 9805 } 9806 9807 int __netdev_update_features(struct net_device *dev) 9808 { 9809 struct net_device *upper, *lower; 9810 netdev_features_t features; 9811 struct list_head *iter; 9812 int err = -1; 9813 9814 ASSERT_RTNL(); 9815 9816 features = netdev_get_wanted_features(dev); 9817 9818 if (dev->netdev_ops->ndo_fix_features) 9819 features = dev->netdev_ops->ndo_fix_features(dev, features); 9820 9821 /* driver might be less strict about feature dependencies */ 9822 features = netdev_fix_features(dev, features); 9823 9824 /* some features can't be enabled if they're off on an upper device */ 9825 netdev_for_each_upper_dev_rcu(dev, upper, iter) 9826 features = netdev_sync_upper_features(dev, upper, features); 9827 9828 if (dev->features == features) 9829 goto sync_lower; 9830 9831 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", 9832 &dev->features, &features); 9833 9834 if (dev->netdev_ops->ndo_set_features) 9835 err = dev->netdev_ops->ndo_set_features(dev, features); 9836 else 9837 err = 0; 9838 9839 if (unlikely(err < 0)) { 9840 netdev_err(dev, 9841 "set_features() failed (%d); wanted %pNF, left %pNF\n", 9842 err, &features, &dev->features); 9843 /* return non-0 since some features might have changed and 9844 * it's better to fire a spurious notification than miss it 9845 */ 9846 return -1; 9847 } 9848 9849 sync_lower: 9850 /* some features must be disabled on lower devices when disabled 9851 * on an upper device (think: bonding master or bridge) 9852 */ 9853 netdev_for_each_lower_dev(dev, lower, iter) 9854 netdev_sync_lower_features(dev, lower, features); 9855 9856 if (!err) { 9857 netdev_features_t diff = features ^ dev->features; 9858 9859 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) { 9860 /* udp_tunnel_{get,drop}_rx_info both need 9861 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the 9862 * device, or they won't do anything. 9863 * Thus we need to update dev->features 9864 * *before* calling udp_tunnel_get_rx_info, 9865 * but *after* calling udp_tunnel_drop_rx_info. 9866 */ 9867 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) { 9868 dev->features = features; 9869 udp_tunnel_get_rx_info(dev); 9870 } else { 9871 udp_tunnel_drop_rx_info(dev); 9872 } 9873 } 9874 9875 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) { 9876 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) { 9877 dev->features = features; 9878 err |= vlan_get_rx_ctag_filter_info(dev); 9879 } else { 9880 vlan_drop_rx_ctag_filter_info(dev); 9881 } 9882 } 9883 9884 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) { 9885 if (features & NETIF_F_HW_VLAN_STAG_FILTER) { 9886 dev->features = features; 9887 err |= vlan_get_rx_stag_filter_info(dev); 9888 } else { 9889 vlan_drop_rx_stag_filter_info(dev); 9890 } 9891 } 9892 9893 dev->features = features; 9894 } 9895 9896 return err < 0 ? 0 : 1; 9897 } 9898 9899 /** 9900 * netdev_update_features - recalculate device features 9901 * @dev: the device to check 9902 * 9903 * Recalculate dev->features set and send notifications if it 9904 * has changed. Should be called after driver or hardware dependent 9905 * conditions might have changed that influence the features. 9906 */ 9907 void netdev_update_features(struct net_device *dev) 9908 { 9909 if (__netdev_update_features(dev)) 9910 netdev_features_change(dev); 9911 } 9912 EXPORT_SYMBOL(netdev_update_features); 9913 9914 /** 9915 * netdev_change_features - recalculate device features 9916 * @dev: the device to check 9917 * 9918 * Recalculate dev->features set and send notifications even 9919 * if they have not changed. Should be called instead of 9920 * netdev_update_features() if also dev->vlan_features might 9921 * have changed to allow the changes to be propagated to stacked 9922 * VLAN devices. 9923 */ 9924 void netdev_change_features(struct net_device *dev) 9925 { 9926 __netdev_update_features(dev); 9927 netdev_features_change(dev); 9928 } 9929 EXPORT_SYMBOL(netdev_change_features); 9930 9931 /** 9932 * netif_stacked_transfer_operstate - transfer operstate 9933 * @rootdev: the root or lower level device to transfer state from 9934 * @dev: the device to transfer operstate to 9935 * 9936 * Transfer operational state from root to device. This is normally 9937 * called when a stacking relationship exists between the root 9938 * device and the device(a leaf device). 9939 */ 9940 void netif_stacked_transfer_operstate(const struct net_device *rootdev, 9941 struct net_device *dev) 9942 { 9943 if (rootdev->operstate == IF_OPER_DORMANT) 9944 netif_dormant_on(dev); 9945 else 9946 netif_dormant_off(dev); 9947 9948 if (rootdev->operstate == IF_OPER_TESTING) 9949 netif_testing_on(dev); 9950 else 9951 netif_testing_off(dev); 9952 9953 if (netif_carrier_ok(rootdev)) 9954 netif_carrier_on(dev); 9955 else 9956 netif_carrier_off(dev); 9957 } 9958 EXPORT_SYMBOL(netif_stacked_transfer_operstate); 9959 9960 static int netif_alloc_rx_queues(struct net_device *dev) 9961 { 9962 unsigned int i, count = dev->num_rx_queues; 9963 struct netdev_rx_queue *rx; 9964 size_t sz = count * sizeof(*rx); 9965 int err = 0; 9966 9967 BUG_ON(count < 1); 9968 9969 rx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL); 9970 if (!rx) 9971 return -ENOMEM; 9972 9973 dev->_rx = rx; 9974 9975 for (i = 0; i < count; i++) { 9976 rx[i].dev = dev; 9977 9978 /* XDP RX-queue setup */ 9979 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0); 9980 if (err < 0) 9981 goto err_rxq_info; 9982 } 9983 return 0; 9984 9985 err_rxq_info: 9986 /* Rollback successful reg's and free other resources */ 9987 while (i--) 9988 xdp_rxq_info_unreg(&rx[i].xdp_rxq); 9989 kvfree(dev->_rx); 9990 dev->_rx = NULL; 9991 return err; 9992 } 9993 9994 static void netif_free_rx_queues(struct net_device *dev) 9995 { 9996 unsigned int i, count = dev->num_rx_queues; 9997 9998 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */ 9999 if (!dev->_rx) 10000 return; 10001 10002 for (i = 0; i < count; i++) 10003 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq); 10004 10005 kvfree(dev->_rx); 10006 } 10007 10008 static void netdev_init_one_queue(struct net_device *dev, 10009 struct netdev_queue *queue, void *_unused) 10010 { 10011 /* Initialize queue lock */ 10012 spin_lock_init(&queue->_xmit_lock); 10013 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); 10014 queue->xmit_lock_owner = -1; 10015 netdev_queue_numa_node_write(queue, NUMA_NO_NODE); 10016 queue->dev = dev; 10017 #ifdef CONFIG_BQL 10018 dql_init(&queue->dql, HZ); 10019 #endif 10020 } 10021 10022 static void netif_free_tx_queues(struct net_device *dev) 10023 { 10024 kvfree(dev->_tx); 10025 } 10026 10027 static int netif_alloc_netdev_queues(struct net_device *dev) 10028 { 10029 unsigned int count = dev->num_tx_queues; 10030 struct netdev_queue *tx; 10031 size_t sz = count * sizeof(*tx); 10032 10033 if (count < 1 || count > 0xffff) 10034 return -EINVAL; 10035 10036 tx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL); 10037 if (!tx) 10038 return -ENOMEM; 10039 10040 dev->_tx = tx; 10041 10042 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); 10043 spin_lock_init(&dev->tx_global_lock); 10044 10045 return 0; 10046 } 10047 10048 void netif_tx_stop_all_queues(struct net_device *dev) 10049 { 10050 unsigned int i; 10051 10052 for (i = 0; i < dev->num_tx_queues; i++) { 10053 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 10054 10055 netif_tx_stop_queue(txq); 10056 } 10057 } 10058 EXPORT_SYMBOL(netif_tx_stop_all_queues); 10059 10060 /** 10061 * register_netdevice - register a network device 10062 * @dev: device to register 10063 * 10064 * Take a completed network device structure and add it to the kernel 10065 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 10066 * chain. 0 is returned on success. A negative errno code is returned 10067 * on a failure to set up the device, or if the name is a duplicate. 10068 * 10069 * Callers must hold the rtnl semaphore. You may want 10070 * register_netdev() instead of this. 10071 * 10072 * BUGS: 10073 * The locking appears insufficient to guarantee two parallel registers 10074 * will not get the same name. 10075 */ 10076 10077 int register_netdevice(struct net_device *dev) 10078 { 10079 int ret; 10080 struct net *net = dev_net(dev); 10081 10082 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE < 10083 NETDEV_FEATURE_COUNT); 10084 BUG_ON(dev_boot_phase); 10085 ASSERT_RTNL(); 10086 10087 might_sleep(); 10088 10089 /* When net_device's are persistent, this will be fatal. */ 10090 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); 10091 BUG_ON(!net); 10092 10093 ret = ethtool_check_ops(dev->ethtool_ops); 10094 if (ret) 10095 return ret; 10096 10097 spin_lock_init(&dev->addr_list_lock); 10098 netdev_set_addr_lockdep_class(dev); 10099 10100 ret = dev_get_valid_name(net, dev, dev->name); 10101 if (ret < 0) 10102 goto out; 10103 10104 ret = -ENOMEM; 10105 dev->name_node = netdev_name_node_head_alloc(dev); 10106 if (!dev->name_node) 10107 goto out; 10108 10109 /* Init, if this function is available */ 10110 if (dev->netdev_ops->ndo_init) { 10111 ret = dev->netdev_ops->ndo_init(dev); 10112 if (ret) { 10113 if (ret > 0) 10114 ret = -EIO; 10115 goto err_free_name; 10116 } 10117 } 10118 10119 if (((dev->hw_features | dev->features) & 10120 NETIF_F_HW_VLAN_CTAG_FILTER) && 10121 (!dev->netdev_ops->ndo_vlan_rx_add_vid || 10122 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { 10123 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); 10124 ret = -EINVAL; 10125 goto err_uninit; 10126 } 10127 10128 ret = -EBUSY; 10129 if (!dev->ifindex) 10130 dev->ifindex = dev_new_index(net); 10131 else if (__dev_get_by_index(net, dev->ifindex)) 10132 goto err_uninit; 10133 10134 /* Transfer changeable features to wanted_features and enable 10135 * software offloads (GSO and GRO). 10136 */ 10137 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF); 10138 dev->features |= NETIF_F_SOFT_FEATURES; 10139 10140 if (dev->udp_tunnel_nic_info) { 10141 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10142 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10143 } 10144 10145 dev->wanted_features = dev->features & dev->hw_features; 10146 10147 if (!(dev->flags & IFF_LOOPBACK)) 10148 dev->hw_features |= NETIF_F_NOCACHE_COPY; 10149 10150 /* If IPv4 TCP segmentation offload is supported we should also 10151 * allow the device to enable segmenting the frame with the option 10152 * of ignoring a static IP ID value. This doesn't enable the 10153 * feature itself but allows the user to enable it later. 10154 */ 10155 if (dev->hw_features & NETIF_F_TSO) 10156 dev->hw_features |= NETIF_F_TSO_MANGLEID; 10157 if (dev->vlan_features & NETIF_F_TSO) 10158 dev->vlan_features |= NETIF_F_TSO_MANGLEID; 10159 if (dev->mpls_features & NETIF_F_TSO) 10160 dev->mpls_features |= NETIF_F_TSO_MANGLEID; 10161 if (dev->hw_enc_features & NETIF_F_TSO) 10162 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; 10163 10164 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. 10165 */ 10166 dev->vlan_features |= NETIF_F_HIGHDMA; 10167 10168 /* Make NETIF_F_SG inheritable to tunnel devices. 10169 */ 10170 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; 10171 10172 /* Make NETIF_F_SG inheritable to MPLS. 10173 */ 10174 dev->mpls_features |= NETIF_F_SG; 10175 10176 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); 10177 ret = notifier_to_errno(ret); 10178 if (ret) 10179 goto err_uninit; 10180 10181 ret = netdev_register_kobject(dev); 10182 if (ret) { 10183 dev->reg_state = NETREG_UNREGISTERED; 10184 goto err_uninit; 10185 } 10186 dev->reg_state = NETREG_REGISTERED; 10187 10188 __netdev_update_features(dev); 10189 10190 /* 10191 * Default initial state at registry is that the 10192 * device is present. 10193 */ 10194 10195 set_bit(__LINK_STATE_PRESENT, &dev->state); 10196 10197 linkwatch_init_dev(dev); 10198 10199 dev_init_scheduler(dev); 10200 dev_hold(dev); 10201 list_netdevice(dev); 10202 add_device_randomness(dev->dev_addr, dev->addr_len); 10203 10204 /* If the device has permanent device address, driver should 10205 * set dev_addr and also addr_assign_type should be set to 10206 * NET_ADDR_PERM (default value). 10207 */ 10208 if (dev->addr_assign_type == NET_ADDR_PERM) 10209 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); 10210 10211 /* Notify protocols, that a new device appeared. */ 10212 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); 10213 ret = notifier_to_errno(ret); 10214 if (ret) { 10215 /* Expect explicit free_netdev() on failure */ 10216 dev->needs_free_netdev = false; 10217 unregister_netdevice_queue(dev, NULL); 10218 goto out; 10219 } 10220 /* 10221 * Prevent userspace races by waiting until the network 10222 * device is fully setup before sending notifications. 10223 */ 10224 if (!dev->rtnl_link_ops || 10225 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 10226 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); 10227 10228 out: 10229 return ret; 10230 10231 err_uninit: 10232 if (dev->netdev_ops->ndo_uninit) 10233 dev->netdev_ops->ndo_uninit(dev); 10234 if (dev->priv_destructor) 10235 dev->priv_destructor(dev); 10236 err_free_name: 10237 netdev_name_node_free(dev->name_node); 10238 goto out; 10239 } 10240 EXPORT_SYMBOL(register_netdevice); 10241 10242 /** 10243 * init_dummy_netdev - init a dummy network device for NAPI 10244 * @dev: device to init 10245 * 10246 * This takes a network device structure and initialize the minimum 10247 * amount of fields so it can be used to schedule NAPI polls without 10248 * registering a full blown interface. This is to be used by drivers 10249 * that need to tie several hardware interfaces to a single NAPI 10250 * poll scheduler due to HW limitations. 10251 */ 10252 int init_dummy_netdev(struct net_device *dev) 10253 { 10254 /* Clear everything. Note we don't initialize spinlocks 10255 * are they aren't supposed to be taken by any of the 10256 * NAPI code and this dummy netdev is supposed to be 10257 * only ever used for NAPI polls 10258 */ 10259 memset(dev, 0, sizeof(struct net_device)); 10260 10261 /* make sure we BUG if trying to hit standard 10262 * register/unregister code path 10263 */ 10264 dev->reg_state = NETREG_DUMMY; 10265 10266 /* NAPI wants this */ 10267 INIT_LIST_HEAD(&dev->napi_list); 10268 10269 /* a dummy interface is started by default */ 10270 set_bit(__LINK_STATE_PRESENT, &dev->state); 10271 set_bit(__LINK_STATE_START, &dev->state); 10272 10273 /* napi_busy_loop stats accounting wants this */ 10274 dev_net_set(dev, &init_net); 10275 10276 /* Note : We dont allocate pcpu_refcnt for dummy devices, 10277 * because users of this 'device' dont need to change 10278 * its refcount. 10279 */ 10280 10281 return 0; 10282 } 10283 EXPORT_SYMBOL_GPL(init_dummy_netdev); 10284 10285 10286 /** 10287 * register_netdev - register a network device 10288 * @dev: device to register 10289 * 10290 * Take a completed network device structure and add it to the kernel 10291 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 10292 * chain. 0 is returned on success. A negative errno code is returned 10293 * on a failure to set up the device, or if the name is a duplicate. 10294 * 10295 * This is a wrapper around register_netdevice that takes the rtnl semaphore 10296 * and expands the device name if you passed a format string to 10297 * alloc_netdev. 10298 */ 10299 int register_netdev(struct net_device *dev) 10300 { 10301 int err; 10302 10303 if (rtnl_lock_killable()) 10304 return -EINTR; 10305 err = register_netdevice(dev); 10306 rtnl_unlock(); 10307 return err; 10308 } 10309 EXPORT_SYMBOL(register_netdev); 10310 10311 int netdev_refcnt_read(const struct net_device *dev) 10312 { 10313 #ifdef CONFIG_PCPU_DEV_REFCNT 10314 int i, refcnt = 0; 10315 10316 for_each_possible_cpu(i) 10317 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); 10318 return refcnt; 10319 #else 10320 return refcount_read(&dev->dev_refcnt); 10321 #endif 10322 } 10323 EXPORT_SYMBOL(netdev_refcnt_read); 10324 10325 #define WAIT_REFS_MIN_MSECS 1 10326 #define WAIT_REFS_MAX_MSECS 250 10327 /** 10328 * netdev_wait_allrefs - wait until all references are gone. 10329 * @dev: target net_device 10330 * 10331 * This is called when unregistering network devices. 10332 * 10333 * Any protocol or device that holds a reference should register 10334 * for netdevice notification, and cleanup and put back the 10335 * reference if they receive an UNREGISTER event. 10336 * We can get stuck here if buggy protocols don't correctly 10337 * call dev_put. 10338 */ 10339 static void netdev_wait_allrefs(struct net_device *dev) 10340 { 10341 unsigned long rebroadcast_time, warning_time; 10342 int wait = 0, refcnt; 10343 10344 linkwatch_forget_dev(dev); 10345 10346 rebroadcast_time = warning_time = jiffies; 10347 refcnt = netdev_refcnt_read(dev); 10348 10349 while (refcnt != 1) { 10350 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { 10351 rtnl_lock(); 10352 10353 /* Rebroadcast unregister notification */ 10354 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10355 10356 __rtnl_unlock(); 10357 rcu_barrier(); 10358 rtnl_lock(); 10359 10360 if (test_bit(__LINK_STATE_LINKWATCH_PENDING, 10361 &dev->state)) { 10362 /* We must not have linkwatch events 10363 * pending on unregister. If this 10364 * happens, we simply run the queue 10365 * unscheduled, resulting in a noop 10366 * for this device. 10367 */ 10368 linkwatch_run_queue(); 10369 } 10370 10371 __rtnl_unlock(); 10372 10373 rebroadcast_time = jiffies; 10374 } 10375 10376 if (!wait) { 10377 rcu_barrier(); 10378 wait = WAIT_REFS_MIN_MSECS; 10379 } else { 10380 msleep(wait); 10381 wait = min(wait << 1, WAIT_REFS_MAX_MSECS); 10382 } 10383 10384 refcnt = netdev_refcnt_read(dev); 10385 10386 if (refcnt != 1 && time_after(jiffies, warning_time + 10 * HZ)) { 10387 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", 10388 dev->name, refcnt); 10389 warning_time = jiffies; 10390 } 10391 } 10392 } 10393 10394 /* The sequence is: 10395 * 10396 * rtnl_lock(); 10397 * ... 10398 * register_netdevice(x1); 10399 * register_netdevice(x2); 10400 * ... 10401 * unregister_netdevice(y1); 10402 * unregister_netdevice(y2); 10403 * ... 10404 * rtnl_unlock(); 10405 * free_netdev(y1); 10406 * free_netdev(y2); 10407 * 10408 * We are invoked by rtnl_unlock(). 10409 * This allows us to deal with problems: 10410 * 1) We can delete sysfs objects which invoke hotplug 10411 * without deadlocking with linkwatch via keventd. 10412 * 2) Since we run with the RTNL semaphore not held, we can sleep 10413 * safely in order to wait for the netdev refcnt to drop to zero. 10414 * 10415 * We must not return until all unregister events added during 10416 * the interval the lock was held have been completed. 10417 */ 10418 void netdev_run_todo(void) 10419 { 10420 struct list_head list; 10421 #ifdef CONFIG_LOCKDEP 10422 struct list_head unlink_list; 10423 10424 list_replace_init(&net_unlink_list, &unlink_list); 10425 10426 while (!list_empty(&unlink_list)) { 10427 struct net_device *dev = list_first_entry(&unlink_list, 10428 struct net_device, 10429 unlink_list); 10430 list_del_init(&dev->unlink_list); 10431 dev->nested_level = dev->lower_level - 1; 10432 } 10433 #endif 10434 10435 /* Snapshot list, allow later requests */ 10436 list_replace_init(&net_todo_list, &list); 10437 10438 __rtnl_unlock(); 10439 10440 10441 /* Wait for rcu callbacks to finish before next phase */ 10442 if (!list_empty(&list)) 10443 rcu_barrier(); 10444 10445 while (!list_empty(&list)) { 10446 struct net_device *dev 10447 = list_first_entry(&list, struct net_device, todo_list); 10448 list_del(&dev->todo_list); 10449 10450 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { 10451 pr_err("network todo '%s' but state %d\n", 10452 dev->name, dev->reg_state); 10453 dump_stack(); 10454 continue; 10455 } 10456 10457 dev->reg_state = NETREG_UNREGISTERED; 10458 10459 netdev_wait_allrefs(dev); 10460 10461 /* paranoia */ 10462 BUG_ON(netdev_refcnt_read(dev) != 1); 10463 BUG_ON(!list_empty(&dev->ptype_all)); 10464 BUG_ON(!list_empty(&dev->ptype_specific)); 10465 WARN_ON(rcu_access_pointer(dev->ip_ptr)); 10466 WARN_ON(rcu_access_pointer(dev->ip6_ptr)); 10467 #if IS_ENABLED(CONFIG_DECNET) 10468 WARN_ON(dev->dn_ptr); 10469 #endif 10470 if (dev->priv_destructor) 10471 dev->priv_destructor(dev); 10472 if (dev->needs_free_netdev) 10473 free_netdev(dev); 10474 10475 /* Report a network device has been unregistered */ 10476 rtnl_lock(); 10477 dev_net(dev)->dev_unreg_count--; 10478 __rtnl_unlock(); 10479 wake_up(&netdev_unregistering_wq); 10480 10481 /* Free network device */ 10482 kobject_put(&dev->dev.kobj); 10483 } 10484 } 10485 10486 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has 10487 * all the same fields in the same order as net_device_stats, with only 10488 * the type differing, but rtnl_link_stats64 may have additional fields 10489 * at the end for newer counters. 10490 */ 10491 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, 10492 const struct net_device_stats *netdev_stats) 10493 { 10494 #if BITS_PER_LONG == 64 10495 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats)); 10496 memcpy(stats64, netdev_stats, sizeof(*netdev_stats)); 10497 /* zero out counters that only exist in rtnl_link_stats64 */ 10498 memset((char *)stats64 + sizeof(*netdev_stats), 0, 10499 sizeof(*stats64) - sizeof(*netdev_stats)); 10500 #else 10501 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long); 10502 const unsigned long *src = (const unsigned long *)netdev_stats; 10503 u64 *dst = (u64 *)stats64; 10504 10505 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); 10506 for (i = 0; i < n; i++) 10507 dst[i] = src[i]; 10508 /* zero out counters that only exist in rtnl_link_stats64 */ 10509 memset((char *)stats64 + n * sizeof(u64), 0, 10510 sizeof(*stats64) - n * sizeof(u64)); 10511 #endif 10512 } 10513 EXPORT_SYMBOL(netdev_stats_to_stats64); 10514 10515 /** 10516 * dev_get_stats - get network device statistics 10517 * @dev: device to get statistics from 10518 * @storage: place to store stats 10519 * 10520 * Get network statistics from device. Return @storage. 10521 * The device driver may provide its own method by setting 10522 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; 10523 * otherwise the internal statistics structure is used. 10524 */ 10525 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, 10526 struct rtnl_link_stats64 *storage) 10527 { 10528 const struct net_device_ops *ops = dev->netdev_ops; 10529 10530 if (ops->ndo_get_stats64) { 10531 memset(storage, 0, sizeof(*storage)); 10532 ops->ndo_get_stats64(dev, storage); 10533 } else if (ops->ndo_get_stats) { 10534 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); 10535 } else { 10536 netdev_stats_to_stats64(storage, &dev->stats); 10537 } 10538 storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped); 10539 storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped); 10540 storage->rx_nohandler += (unsigned long)atomic_long_read(&dev->rx_nohandler); 10541 return storage; 10542 } 10543 EXPORT_SYMBOL(dev_get_stats); 10544 10545 /** 10546 * dev_fetch_sw_netstats - get per-cpu network device statistics 10547 * @s: place to store stats 10548 * @netstats: per-cpu network stats to read from 10549 * 10550 * Read per-cpu network statistics and populate the related fields in @s. 10551 */ 10552 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s, 10553 const struct pcpu_sw_netstats __percpu *netstats) 10554 { 10555 int cpu; 10556 10557 for_each_possible_cpu(cpu) { 10558 const struct pcpu_sw_netstats *stats; 10559 struct pcpu_sw_netstats tmp; 10560 unsigned int start; 10561 10562 stats = per_cpu_ptr(netstats, cpu); 10563 do { 10564 start = u64_stats_fetch_begin_irq(&stats->syncp); 10565 tmp.rx_packets = stats->rx_packets; 10566 tmp.rx_bytes = stats->rx_bytes; 10567 tmp.tx_packets = stats->tx_packets; 10568 tmp.tx_bytes = stats->tx_bytes; 10569 } while (u64_stats_fetch_retry_irq(&stats->syncp, start)); 10570 10571 s->rx_packets += tmp.rx_packets; 10572 s->rx_bytes += tmp.rx_bytes; 10573 s->tx_packets += tmp.tx_packets; 10574 s->tx_bytes += tmp.tx_bytes; 10575 } 10576 } 10577 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats); 10578 10579 /** 10580 * dev_get_tstats64 - ndo_get_stats64 implementation 10581 * @dev: device to get statistics from 10582 * @s: place to store stats 10583 * 10584 * Populate @s from dev->stats and dev->tstats. Can be used as 10585 * ndo_get_stats64() callback. 10586 */ 10587 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s) 10588 { 10589 netdev_stats_to_stats64(s, &dev->stats); 10590 dev_fetch_sw_netstats(s, dev->tstats); 10591 } 10592 EXPORT_SYMBOL_GPL(dev_get_tstats64); 10593 10594 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) 10595 { 10596 struct netdev_queue *queue = dev_ingress_queue(dev); 10597 10598 #ifdef CONFIG_NET_CLS_ACT 10599 if (queue) 10600 return queue; 10601 queue = kzalloc(sizeof(*queue), GFP_KERNEL); 10602 if (!queue) 10603 return NULL; 10604 netdev_init_one_queue(dev, queue, NULL); 10605 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); 10606 queue->qdisc_sleeping = &noop_qdisc; 10607 rcu_assign_pointer(dev->ingress_queue, queue); 10608 #endif 10609 return queue; 10610 } 10611 10612 static const struct ethtool_ops default_ethtool_ops; 10613 10614 void netdev_set_default_ethtool_ops(struct net_device *dev, 10615 const struct ethtool_ops *ops) 10616 { 10617 if (dev->ethtool_ops == &default_ethtool_ops) 10618 dev->ethtool_ops = ops; 10619 } 10620 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); 10621 10622 void netdev_freemem(struct net_device *dev) 10623 { 10624 char *addr = (char *)dev - dev->padded; 10625 10626 kvfree(addr); 10627 } 10628 10629 /** 10630 * alloc_netdev_mqs - allocate network device 10631 * @sizeof_priv: size of private data to allocate space for 10632 * @name: device name format string 10633 * @name_assign_type: origin of device name 10634 * @setup: callback to initialize device 10635 * @txqs: the number of TX subqueues to allocate 10636 * @rxqs: the number of RX subqueues to allocate 10637 * 10638 * Allocates a struct net_device with private data area for driver use 10639 * and performs basic initialization. Also allocates subqueue structs 10640 * for each queue on the device. 10641 */ 10642 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, 10643 unsigned char name_assign_type, 10644 void (*setup)(struct net_device *), 10645 unsigned int txqs, unsigned int rxqs) 10646 { 10647 struct net_device *dev; 10648 unsigned int alloc_size; 10649 struct net_device *p; 10650 10651 BUG_ON(strlen(name) >= sizeof(dev->name)); 10652 10653 if (txqs < 1) { 10654 pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); 10655 return NULL; 10656 } 10657 10658 if (rxqs < 1) { 10659 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); 10660 return NULL; 10661 } 10662 10663 alloc_size = sizeof(struct net_device); 10664 if (sizeof_priv) { 10665 /* ensure 32-byte alignment of private area */ 10666 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); 10667 alloc_size += sizeof_priv; 10668 } 10669 /* ensure 32-byte alignment of whole construct */ 10670 alloc_size += NETDEV_ALIGN - 1; 10671 10672 p = kvzalloc(alloc_size, GFP_KERNEL | __GFP_RETRY_MAYFAIL); 10673 if (!p) 10674 return NULL; 10675 10676 dev = PTR_ALIGN(p, NETDEV_ALIGN); 10677 dev->padded = (char *)dev - (char *)p; 10678 10679 #ifdef CONFIG_PCPU_DEV_REFCNT 10680 dev->pcpu_refcnt = alloc_percpu(int); 10681 if (!dev->pcpu_refcnt) 10682 goto free_dev; 10683 dev_hold(dev); 10684 #else 10685 refcount_set(&dev->dev_refcnt, 1); 10686 #endif 10687 10688 if (dev_addr_init(dev)) 10689 goto free_pcpu; 10690 10691 dev_mc_init(dev); 10692 dev_uc_init(dev); 10693 10694 dev_net_set(dev, &init_net); 10695 10696 dev->gso_max_size = GSO_MAX_SIZE; 10697 dev->gso_max_segs = GSO_MAX_SEGS; 10698 dev->upper_level = 1; 10699 dev->lower_level = 1; 10700 #ifdef CONFIG_LOCKDEP 10701 dev->nested_level = 0; 10702 INIT_LIST_HEAD(&dev->unlink_list); 10703 #endif 10704 10705 INIT_LIST_HEAD(&dev->napi_list); 10706 INIT_LIST_HEAD(&dev->unreg_list); 10707 INIT_LIST_HEAD(&dev->close_list); 10708 INIT_LIST_HEAD(&dev->link_watch_list); 10709 INIT_LIST_HEAD(&dev->adj_list.upper); 10710 INIT_LIST_HEAD(&dev->adj_list.lower); 10711 INIT_LIST_HEAD(&dev->ptype_all); 10712 INIT_LIST_HEAD(&dev->ptype_specific); 10713 INIT_LIST_HEAD(&dev->net_notifier_list); 10714 #ifdef CONFIG_NET_SCHED 10715 hash_init(dev->qdisc_hash); 10716 #endif 10717 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; 10718 setup(dev); 10719 10720 if (!dev->tx_queue_len) { 10721 dev->priv_flags |= IFF_NO_QUEUE; 10722 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; 10723 } 10724 10725 dev->num_tx_queues = txqs; 10726 dev->real_num_tx_queues = txqs; 10727 if (netif_alloc_netdev_queues(dev)) 10728 goto free_all; 10729 10730 dev->num_rx_queues = rxqs; 10731 dev->real_num_rx_queues = rxqs; 10732 if (netif_alloc_rx_queues(dev)) 10733 goto free_all; 10734 10735 strcpy(dev->name, name); 10736 dev->name_assign_type = name_assign_type; 10737 dev->group = INIT_NETDEV_GROUP; 10738 if (!dev->ethtool_ops) 10739 dev->ethtool_ops = &default_ethtool_ops; 10740 10741 nf_hook_ingress_init(dev); 10742 10743 return dev; 10744 10745 free_all: 10746 free_netdev(dev); 10747 return NULL; 10748 10749 free_pcpu: 10750 #ifdef CONFIG_PCPU_DEV_REFCNT 10751 free_percpu(dev->pcpu_refcnt); 10752 free_dev: 10753 #endif 10754 netdev_freemem(dev); 10755 return NULL; 10756 } 10757 EXPORT_SYMBOL(alloc_netdev_mqs); 10758 10759 /** 10760 * free_netdev - free network device 10761 * @dev: device 10762 * 10763 * This function does the last stage of destroying an allocated device 10764 * interface. The reference to the device object is released. If this 10765 * is the last reference then it will be freed.Must be called in process 10766 * context. 10767 */ 10768 void free_netdev(struct net_device *dev) 10769 { 10770 struct napi_struct *p, *n; 10771 10772 might_sleep(); 10773 10774 /* When called immediately after register_netdevice() failed the unwind 10775 * handling may still be dismantling the device. Handle that case by 10776 * deferring the free. 10777 */ 10778 if (dev->reg_state == NETREG_UNREGISTERING) { 10779 ASSERT_RTNL(); 10780 dev->needs_free_netdev = true; 10781 return; 10782 } 10783 10784 netif_free_tx_queues(dev); 10785 netif_free_rx_queues(dev); 10786 10787 kfree(rcu_dereference_protected(dev->ingress_queue, 1)); 10788 10789 /* Flush device addresses */ 10790 dev_addr_flush(dev); 10791 10792 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) 10793 netif_napi_del(p); 10794 10795 #ifdef CONFIG_PCPU_DEV_REFCNT 10796 free_percpu(dev->pcpu_refcnt); 10797 dev->pcpu_refcnt = NULL; 10798 #endif 10799 free_percpu(dev->xdp_bulkq); 10800 dev->xdp_bulkq = NULL; 10801 10802 /* Compatibility with error handling in drivers */ 10803 if (dev->reg_state == NETREG_UNINITIALIZED) { 10804 netdev_freemem(dev); 10805 return; 10806 } 10807 10808 BUG_ON(dev->reg_state != NETREG_UNREGISTERED); 10809 dev->reg_state = NETREG_RELEASED; 10810 10811 /* will free via device release */ 10812 put_device(&dev->dev); 10813 } 10814 EXPORT_SYMBOL(free_netdev); 10815 10816 /** 10817 * synchronize_net - Synchronize with packet receive processing 10818 * 10819 * Wait for packets currently being received to be done. 10820 * Does not block later packets from starting. 10821 */ 10822 void synchronize_net(void) 10823 { 10824 might_sleep(); 10825 if (rtnl_is_locked()) 10826 synchronize_rcu_expedited(); 10827 else 10828 synchronize_rcu(); 10829 } 10830 EXPORT_SYMBOL(synchronize_net); 10831 10832 /** 10833 * unregister_netdevice_queue - remove device from the kernel 10834 * @dev: device 10835 * @head: list 10836 * 10837 * This function shuts down a device interface and removes it 10838 * from the kernel tables. 10839 * If head not NULL, device is queued to be unregistered later. 10840 * 10841 * Callers must hold the rtnl semaphore. You may want 10842 * unregister_netdev() instead of this. 10843 */ 10844 10845 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) 10846 { 10847 ASSERT_RTNL(); 10848 10849 if (head) { 10850 list_move_tail(&dev->unreg_list, head); 10851 } else { 10852 LIST_HEAD(single); 10853 10854 list_add(&dev->unreg_list, &single); 10855 unregister_netdevice_many(&single); 10856 } 10857 } 10858 EXPORT_SYMBOL(unregister_netdevice_queue); 10859 10860 /** 10861 * unregister_netdevice_many - unregister many devices 10862 * @head: list of devices 10863 * 10864 * Note: As most callers use a stack allocated list_head, 10865 * we force a list_del() to make sure stack wont be corrupted later. 10866 */ 10867 void unregister_netdevice_many(struct list_head *head) 10868 { 10869 struct net_device *dev, *tmp; 10870 LIST_HEAD(close_head); 10871 10872 BUG_ON(dev_boot_phase); 10873 ASSERT_RTNL(); 10874 10875 if (list_empty(head)) 10876 return; 10877 10878 list_for_each_entry_safe(dev, tmp, head, unreg_list) { 10879 /* Some devices call without registering 10880 * for initialization unwind. Remove those 10881 * devices and proceed with the remaining. 10882 */ 10883 if (dev->reg_state == NETREG_UNINITIALIZED) { 10884 pr_debug("unregister_netdevice: device %s/%p never was registered\n", 10885 dev->name, dev); 10886 10887 WARN_ON(1); 10888 list_del(&dev->unreg_list); 10889 continue; 10890 } 10891 dev->dismantle = true; 10892 BUG_ON(dev->reg_state != NETREG_REGISTERED); 10893 } 10894 10895 /* If device is running, close it first. */ 10896 list_for_each_entry(dev, head, unreg_list) 10897 list_add_tail(&dev->close_list, &close_head); 10898 dev_close_many(&close_head, true); 10899 10900 list_for_each_entry(dev, head, unreg_list) { 10901 /* And unlink it from device chain. */ 10902 unlist_netdevice(dev); 10903 10904 dev->reg_state = NETREG_UNREGISTERING; 10905 } 10906 flush_all_backlogs(); 10907 10908 synchronize_net(); 10909 10910 list_for_each_entry(dev, head, unreg_list) { 10911 struct sk_buff *skb = NULL; 10912 10913 /* Shutdown queueing discipline. */ 10914 dev_shutdown(dev); 10915 10916 dev_xdp_uninstall(dev); 10917 10918 /* Notify protocols, that we are about to destroy 10919 * this device. They should clean all the things. 10920 */ 10921 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10922 10923 if (!dev->rtnl_link_ops || 10924 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 10925 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0, 10926 GFP_KERNEL, NULL, 0); 10927 10928 /* 10929 * Flush the unicast and multicast chains 10930 */ 10931 dev_uc_flush(dev); 10932 dev_mc_flush(dev); 10933 10934 netdev_name_node_alt_flush(dev); 10935 netdev_name_node_free(dev->name_node); 10936 10937 if (dev->netdev_ops->ndo_uninit) 10938 dev->netdev_ops->ndo_uninit(dev); 10939 10940 if (skb) 10941 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL); 10942 10943 /* Notifier chain MUST detach us all upper devices. */ 10944 WARN_ON(netdev_has_any_upper_dev(dev)); 10945 WARN_ON(netdev_has_any_lower_dev(dev)); 10946 10947 /* Remove entries from kobject tree */ 10948 netdev_unregister_kobject(dev); 10949 #ifdef CONFIG_XPS 10950 /* Remove XPS queueing entries */ 10951 netif_reset_xps_queues_gt(dev, 0); 10952 #endif 10953 } 10954 10955 synchronize_net(); 10956 10957 list_for_each_entry(dev, head, unreg_list) { 10958 dev_put(dev); 10959 net_set_todo(dev); 10960 } 10961 10962 list_del(head); 10963 } 10964 EXPORT_SYMBOL(unregister_netdevice_many); 10965 10966 /** 10967 * unregister_netdev - remove device from the kernel 10968 * @dev: device 10969 * 10970 * This function shuts down a device interface and removes it 10971 * from the kernel tables. 10972 * 10973 * This is just a wrapper for unregister_netdevice that takes 10974 * the rtnl semaphore. In general you want to use this and not 10975 * unregister_netdevice. 10976 */ 10977 void unregister_netdev(struct net_device *dev) 10978 { 10979 rtnl_lock(); 10980 unregister_netdevice(dev); 10981 rtnl_unlock(); 10982 } 10983 EXPORT_SYMBOL(unregister_netdev); 10984 10985 /** 10986 * dev_change_net_namespace - move device to different nethost namespace 10987 * @dev: device 10988 * @net: network namespace 10989 * @pat: If not NULL name pattern to try if the current device name 10990 * is already taken in the destination network namespace. 10991 * 10992 * This function shuts down a device interface and moves it 10993 * to a new network namespace. On success 0 is returned, on 10994 * a failure a netagive errno code is returned. 10995 * 10996 * Callers must hold the rtnl semaphore. 10997 */ 10998 10999 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat) 11000 { 11001 struct net *net_old = dev_net(dev); 11002 int err, new_nsid, new_ifindex; 11003 11004 ASSERT_RTNL(); 11005 11006 /* Don't allow namespace local devices to be moved. */ 11007 err = -EINVAL; 11008 if (dev->features & NETIF_F_NETNS_LOCAL) 11009 goto out; 11010 11011 /* Ensure the device has been registrered */ 11012 if (dev->reg_state != NETREG_REGISTERED) 11013 goto out; 11014 11015 /* Get out if there is nothing todo */ 11016 err = 0; 11017 if (net_eq(net_old, net)) 11018 goto out; 11019 11020 /* Pick the destination device name, and ensure 11021 * we can use it in the destination network namespace. 11022 */ 11023 err = -EEXIST; 11024 if (__dev_get_by_name(net, dev->name)) { 11025 /* We get here if we can't use the current device name */ 11026 if (!pat) 11027 goto out; 11028 err = dev_get_valid_name(net, dev, pat); 11029 if (err < 0) 11030 goto out; 11031 } 11032 11033 /* 11034 * And now a mini version of register_netdevice unregister_netdevice. 11035 */ 11036 11037 /* If device is running close it first. */ 11038 dev_close(dev); 11039 11040 /* And unlink it from device chain */ 11041 unlist_netdevice(dev); 11042 11043 synchronize_net(); 11044 11045 /* Shutdown queueing discipline. */ 11046 dev_shutdown(dev); 11047 11048 /* Notify protocols, that we are about to destroy 11049 * this device. They should clean all the things. 11050 * 11051 * Note that dev->reg_state stays at NETREG_REGISTERED. 11052 * This is wanted because this way 8021q and macvlan know 11053 * the device is just moving and can keep their slaves up. 11054 */ 11055 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 11056 rcu_barrier(); 11057 11058 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL); 11059 /* If there is an ifindex conflict assign a new one */ 11060 if (__dev_get_by_index(net, dev->ifindex)) 11061 new_ifindex = dev_new_index(net); 11062 else 11063 new_ifindex = dev->ifindex; 11064 11065 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid, 11066 new_ifindex); 11067 11068 /* 11069 * Flush the unicast and multicast chains 11070 */ 11071 dev_uc_flush(dev); 11072 dev_mc_flush(dev); 11073 11074 /* Send a netdev-removed uevent to the old namespace */ 11075 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); 11076 netdev_adjacent_del_links(dev); 11077 11078 /* Move per-net netdevice notifiers that are following the netdevice */ 11079 move_netdevice_notifiers_dev_net(dev, net); 11080 11081 /* Actually switch the network namespace */ 11082 dev_net_set(dev, net); 11083 dev->ifindex = new_ifindex; 11084 11085 /* Send a netdev-add uevent to the new namespace */ 11086 kobject_uevent(&dev->dev.kobj, KOBJ_ADD); 11087 netdev_adjacent_add_links(dev); 11088 11089 /* Fixup kobjects */ 11090 err = device_rename(&dev->dev, dev->name); 11091 WARN_ON(err); 11092 11093 /* Adapt owner in case owning user namespace of target network 11094 * namespace is different from the original one. 11095 */ 11096 err = netdev_change_owner(dev, net_old, net); 11097 WARN_ON(err); 11098 11099 /* Add the device back in the hashes */ 11100 list_netdevice(dev); 11101 11102 /* Notify protocols, that a new device appeared. */ 11103 call_netdevice_notifiers(NETDEV_REGISTER, dev); 11104 11105 /* 11106 * Prevent userspace races by waiting until the network 11107 * device is fully setup before sending notifications. 11108 */ 11109 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); 11110 11111 synchronize_net(); 11112 err = 0; 11113 out: 11114 return err; 11115 } 11116 EXPORT_SYMBOL_GPL(dev_change_net_namespace); 11117 11118 static int dev_cpu_dead(unsigned int oldcpu) 11119 { 11120 struct sk_buff **list_skb; 11121 struct sk_buff *skb; 11122 unsigned int cpu; 11123 struct softnet_data *sd, *oldsd, *remsd = NULL; 11124 11125 local_irq_disable(); 11126 cpu = smp_processor_id(); 11127 sd = &per_cpu(softnet_data, cpu); 11128 oldsd = &per_cpu(softnet_data, oldcpu); 11129 11130 /* Find end of our completion_queue. */ 11131 list_skb = &sd->completion_queue; 11132 while (*list_skb) 11133 list_skb = &(*list_skb)->next; 11134 /* Append completion queue from offline CPU. */ 11135 *list_skb = oldsd->completion_queue; 11136 oldsd->completion_queue = NULL; 11137 11138 /* Append output queue from offline CPU. */ 11139 if (oldsd->output_queue) { 11140 *sd->output_queue_tailp = oldsd->output_queue; 11141 sd->output_queue_tailp = oldsd->output_queue_tailp; 11142 oldsd->output_queue = NULL; 11143 oldsd->output_queue_tailp = &oldsd->output_queue; 11144 } 11145 /* Append NAPI poll list from offline CPU, with one exception : 11146 * process_backlog() must be called by cpu owning percpu backlog. 11147 * We properly handle process_queue & input_pkt_queue later. 11148 */ 11149 while (!list_empty(&oldsd->poll_list)) { 11150 struct napi_struct *napi = list_first_entry(&oldsd->poll_list, 11151 struct napi_struct, 11152 poll_list); 11153 11154 list_del_init(&napi->poll_list); 11155 if (napi->poll == process_backlog) 11156 napi->state = 0; 11157 else 11158 ____napi_schedule(sd, napi); 11159 } 11160 11161 raise_softirq_irqoff(NET_TX_SOFTIRQ); 11162 local_irq_enable(); 11163 11164 #ifdef CONFIG_RPS 11165 remsd = oldsd->rps_ipi_list; 11166 oldsd->rps_ipi_list = NULL; 11167 #endif 11168 /* send out pending IPI's on offline CPU */ 11169 net_rps_send_ipi(remsd); 11170 11171 /* Process offline CPU's input_pkt_queue */ 11172 while ((skb = __skb_dequeue(&oldsd->process_queue))) { 11173 netif_rx_ni(skb); 11174 input_queue_head_incr(oldsd); 11175 } 11176 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { 11177 netif_rx_ni(skb); 11178 input_queue_head_incr(oldsd); 11179 } 11180 11181 return 0; 11182 } 11183 11184 /** 11185 * netdev_increment_features - increment feature set by one 11186 * @all: current feature set 11187 * @one: new feature set 11188 * @mask: mask feature set 11189 * 11190 * Computes a new feature set after adding a device with feature set 11191 * @one to the master device with current feature set @all. Will not 11192 * enable anything that is off in @mask. Returns the new feature set. 11193 */ 11194 netdev_features_t netdev_increment_features(netdev_features_t all, 11195 netdev_features_t one, netdev_features_t mask) 11196 { 11197 if (mask & NETIF_F_HW_CSUM) 11198 mask |= NETIF_F_CSUM_MASK; 11199 mask |= NETIF_F_VLAN_CHALLENGED; 11200 11201 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; 11202 all &= one | ~NETIF_F_ALL_FOR_ALL; 11203 11204 /* If one device supports hw checksumming, set for all. */ 11205 if (all & NETIF_F_HW_CSUM) 11206 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); 11207 11208 return all; 11209 } 11210 EXPORT_SYMBOL(netdev_increment_features); 11211 11212 static struct hlist_head * __net_init netdev_create_hash(void) 11213 { 11214 int i; 11215 struct hlist_head *hash; 11216 11217 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL); 11218 if (hash != NULL) 11219 for (i = 0; i < NETDEV_HASHENTRIES; i++) 11220 INIT_HLIST_HEAD(&hash[i]); 11221 11222 return hash; 11223 } 11224 11225 /* Initialize per network namespace state */ 11226 static int __net_init netdev_init(struct net *net) 11227 { 11228 BUILD_BUG_ON(GRO_HASH_BUCKETS > 11229 8 * sizeof_field(struct napi_struct, gro_bitmask)); 11230 11231 if (net != &init_net) 11232 INIT_LIST_HEAD(&net->dev_base_head); 11233 11234 net->dev_name_head = netdev_create_hash(); 11235 if (net->dev_name_head == NULL) 11236 goto err_name; 11237 11238 net->dev_index_head = netdev_create_hash(); 11239 if (net->dev_index_head == NULL) 11240 goto err_idx; 11241 11242 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain); 11243 11244 return 0; 11245 11246 err_idx: 11247 kfree(net->dev_name_head); 11248 err_name: 11249 return -ENOMEM; 11250 } 11251 11252 /** 11253 * netdev_drivername - network driver for the device 11254 * @dev: network device 11255 * 11256 * Determine network driver for device. 11257 */ 11258 const char *netdev_drivername(const struct net_device *dev) 11259 { 11260 const struct device_driver *driver; 11261 const struct device *parent; 11262 const char *empty = ""; 11263 11264 parent = dev->dev.parent; 11265 if (!parent) 11266 return empty; 11267 11268 driver = parent->driver; 11269 if (driver && driver->name) 11270 return driver->name; 11271 return empty; 11272 } 11273 11274 static void __netdev_printk(const char *level, const struct net_device *dev, 11275 struct va_format *vaf) 11276 { 11277 if (dev && dev->dev.parent) { 11278 dev_printk_emit(level[1] - '0', 11279 dev->dev.parent, 11280 "%s %s %s%s: %pV", 11281 dev_driver_string(dev->dev.parent), 11282 dev_name(dev->dev.parent), 11283 netdev_name(dev), netdev_reg_state(dev), 11284 vaf); 11285 } else if (dev) { 11286 printk("%s%s%s: %pV", 11287 level, netdev_name(dev), netdev_reg_state(dev), vaf); 11288 } else { 11289 printk("%s(NULL net_device): %pV", level, vaf); 11290 } 11291 } 11292 11293 void netdev_printk(const char *level, const struct net_device *dev, 11294 const char *format, ...) 11295 { 11296 struct va_format vaf; 11297 va_list args; 11298 11299 va_start(args, format); 11300 11301 vaf.fmt = format; 11302 vaf.va = &args; 11303 11304 __netdev_printk(level, dev, &vaf); 11305 11306 va_end(args); 11307 } 11308 EXPORT_SYMBOL(netdev_printk); 11309 11310 #define define_netdev_printk_level(func, level) \ 11311 void func(const struct net_device *dev, const char *fmt, ...) \ 11312 { \ 11313 struct va_format vaf; \ 11314 va_list args; \ 11315 \ 11316 va_start(args, fmt); \ 11317 \ 11318 vaf.fmt = fmt; \ 11319 vaf.va = &args; \ 11320 \ 11321 __netdev_printk(level, dev, &vaf); \ 11322 \ 11323 va_end(args); \ 11324 } \ 11325 EXPORT_SYMBOL(func); 11326 11327 define_netdev_printk_level(netdev_emerg, KERN_EMERG); 11328 define_netdev_printk_level(netdev_alert, KERN_ALERT); 11329 define_netdev_printk_level(netdev_crit, KERN_CRIT); 11330 define_netdev_printk_level(netdev_err, KERN_ERR); 11331 define_netdev_printk_level(netdev_warn, KERN_WARNING); 11332 define_netdev_printk_level(netdev_notice, KERN_NOTICE); 11333 define_netdev_printk_level(netdev_info, KERN_INFO); 11334 11335 static void __net_exit netdev_exit(struct net *net) 11336 { 11337 kfree(net->dev_name_head); 11338 kfree(net->dev_index_head); 11339 if (net != &init_net) 11340 WARN_ON_ONCE(!list_empty(&net->dev_base_head)); 11341 } 11342 11343 static struct pernet_operations __net_initdata netdev_net_ops = { 11344 .init = netdev_init, 11345 .exit = netdev_exit, 11346 }; 11347 11348 static void __net_exit default_device_exit(struct net *net) 11349 { 11350 struct net_device *dev, *aux; 11351 /* 11352 * Push all migratable network devices back to the 11353 * initial network namespace 11354 */ 11355 rtnl_lock(); 11356 for_each_netdev_safe(net, dev, aux) { 11357 int err; 11358 char fb_name[IFNAMSIZ]; 11359 11360 /* Ignore unmoveable devices (i.e. loopback) */ 11361 if (dev->features & NETIF_F_NETNS_LOCAL) 11362 continue; 11363 11364 /* Leave virtual devices for the generic cleanup */ 11365 if (dev->rtnl_link_ops) 11366 continue; 11367 11368 /* Push remaining network devices to init_net */ 11369 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); 11370 if (__dev_get_by_name(&init_net, fb_name)) 11371 snprintf(fb_name, IFNAMSIZ, "dev%%d"); 11372 err = dev_change_net_namespace(dev, &init_net, fb_name); 11373 if (err) { 11374 pr_emerg("%s: failed to move %s to init_net: %d\n", 11375 __func__, dev->name, err); 11376 BUG(); 11377 } 11378 } 11379 rtnl_unlock(); 11380 } 11381 11382 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list) 11383 { 11384 /* Return with the rtnl_lock held when there are no network 11385 * devices unregistering in any network namespace in net_list. 11386 */ 11387 struct net *net; 11388 bool unregistering; 11389 DEFINE_WAIT_FUNC(wait, woken_wake_function); 11390 11391 add_wait_queue(&netdev_unregistering_wq, &wait); 11392 for (;;) { 11393 unregistering = false; 11394 rtnl_lock(); 11395 list_for_each_entry(net, net_list, exit_list) { 11396 if (net->dev_unreg_count > 0) { 11397 unregistering = true; 11398 break; 11399 } 11400 } 11401 if (!unregistering) 11402 break; 11403 __rtnl_unlock(); 11404 11405 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); 11406 } 11407 remove_wait_queue(&netdev_unregistering_wq, &wait); 11408 } 11409 11410 static void __net_exit default_device_exit_batch(struct list_head *net_list) 11411 { 11412 /* At exit all network devices most be removed from a network 11413 * namespace. Do this in the reverse order of registration. 11414 * Do this across as many network namespaces as possible to 11415 * improve batching efficiency. 11416 */ 11417 struct net_device *dev; 11418 struct net *net; 11419 LIST_HEAD(dev_kill_list); 11420 11421 /* To prevent network device cleanup code from dereferencing 11422 * loopback devices or network devices that have been freed 11423 * wait here for all pending unregistrations to complete, 11424 * before unregistring the loopback device and allowing the 11425 * network namespace be freed. 11426 * 11427 * The netdev todo list containing all network devices 11428 * unregistrations that happen in default_device_exit_batch 11429 * will run in the rtnl_unlock() at the end of 11430 * default_device_exit_batch. 11431 */ 11432 rtnl_lock_unregistering(net_list); 11433 list_for_each_entry(net, net_list, exit_list) { 11434 for_each_netdev_reverse(net, dev) { 11435 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) 11436 dev->rtnl_link_ops->dellink(dev, &dev_kill_list); 11437 else 11438 unregister_netdevice_queue(dev, &dev_kill_list); 11439 } 11440 } 11441 unregister_netdevice_many(&dev_kill_list); 11442 rtnl_unlock(); 11443 } 11444 11445 static struct pernet_operations __net_initdata default_device_ops = { 11446 .exit = default_device_exit, 11447 .exit_batch = default_device_exit_batch, 11448 }; 11449 11450 /* 11451 * Initialize the DEV module. At boot time this walks the device list and 11452 * unhooks any devices that fail to initialise (normally hardware not 11453 * present) and leaves us with a valid list of present and active devices. 11454 * 11455 */ 11456 11457 /* 11458 * This is called single threaded during boot, so no need 11459 * to take the rtnl semaphore. 11460 */ 11461 static int __init net_dev_init(void) 11462 { 11463 int i, rc = -ENOMEM; 11464 11465 BUG_ON(!dev_boot_phase); 11466 11467 if (dev_proc_init()) 11468 goto out; 11469 11470 if (netdev_kobject_init()) 11471 goto out; 11472 11473 INIT_LIST_HEAD(&ptype_all); 11474 for (i = 0; i < PTYPE_HASH_SIZE; i++) 11475 INIT_LIST_HEAD(&ptype_base[i]); 11476 11477 INIT_LIST_HEAD(&offload_base); 11478 11479 if (register_pernet_subsys(&netdev_net_ops)) 11480 goto out; 11481 11482 /* 11483 * Initialise the packet receive queues. 11484 */ 11485 11486 for_each_possible_cpu(i) { 11487 struct work_struct *flush = per_cpu_ptr(&flush_works, i); 11488 struct softnet_data *sd = &per_cpu(softnet_data, i); 11489 11490 INIT_WORK(flush, flush_backlog); 11491 11492 skb_queue_head_init(&sd->input_pkt_queue); 11493 skb_queue_head_init(&sd->process_queue); 11494 #ifdef CONFIG_XFRM_OFFLOAD 11495 skb_queue_head_init(&sd->xfrm_backlog); 11496 #endif 11497 INIT_LIST_HEAD(&sd->poll_list); 11498 sd->output_queue_tailp = &sd->output_queue; 11499 #ifdef CONFIG_RPS 11500 INIT_CSD(&sd->csd, rps_trigger_softirq, sd); 11501 sd->cpu = i; 11502 #endif 11503 11504 init_gro_hash(&sd->backlog); 11505 sd->backlog.poll = process_backlog; 11506 sd->backlog.weight = weight_p; 11507 } 11508 11509 dev_boot_phase = 0; 11510 11511 /* The loopback device is special if any other network devices 11512 * is present in a network namespace the loopback device must 11513 * be present. Since we now dynamically allocate and free the 11514 * loopback device ensure this invariant is maintained by 11515 * keeping the loopback device as the first device on the 11516 * list of network devices. Ensuring the loopback devices 11517 * is the first device that appears and the last network device 11518 * that disappears. 11519 */ 11520 if (register_pernet_device(&loopback_net_ops)) 11521 goto out; 11522 11523 if (register_pernet_device(&default_device_ops)) 11524 goto out; 11525 11526 open_softirq(NET_TX_SOFTIRQ, net_tx_action); 11527 open_softirq(NET_RX_SOFTIRQ, net_rx_action); 11528 11529 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", 11530 NULL, dev_cpu_dead); 11531 WARN_ON(rc < 0); 11532 rc = 0; 11533 out: 11534 return rc; 11535 } 11536 11537 subsys_initcall(net_dev_init); 11538