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 unsigned int len; 2221 2222 if (!(dev->flags & IFF_UP)) 2223 return false; 2224 2225 len = dev->mtu + dev->hard_header_len + VLAN_HLEN; 2226 if (skb->len <= len) 2227 return true; 2228 2229 /* if TSO is enabled, we don't care about the length as the packet 2230 * could be forwarded without being segmented before 2231 */ 2232 if (skb_is_gso(skb)) 2233 return true; 2234 2235 return false; 2236 } 2237 EXPORT_SYMBOL_GPL(is_skb_forwardable); 2238 2239 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2240 { 2241 int ret = ____dev_forward_skb(dev, skb); 2242 2243 if (likely(!ret)) { 2244 skb->protocol = eth_type_trans(skb, dev); 2245 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); 2246 } 2247 2248 return ret; 2249 } 2250 EXPORT_SYMBOL_GPL(__dev_forward_skb); 2251 2252 /** 2253 * dev_forward_skb - loopback an skb to another netif 2254 * 2255 * @dev: destination network device 2256 * @skb: buffer to forward 2257 * 2258 * return values: 2259 * NET_RX_SUCCESS (no congestion) 2260 * NET_RX_DROP (packet was dropped, but freed) 2261 * 2262 * dev_forward_skb can be used for injecting an skb from the 2263 * start_xmit function of one device into the receive queue 2264 * of another device. 2265 * 2266 * The receiving device may be in another namespace, so 2267 * we have to clear all information in the skb that could 2268 * impact namespace isolation. 2269 */ 2270 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2271 { 2272 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); 2273 } 2274 EXPORT_SYMBOL_GPL(dev_forward_skb); 2275 2276 static inline int deliver_skb(struct sk_buff *skb, 2277 struct packet_type *pt_prev, 2278 struct net_device *orig_dev) 2279 { 2280 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 2281 return -ENOMEM; 2282 refcount_inc(&skb->users); 2283 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 2284 } 2285 2286 static inline void deliver_ptype_list_skb(struct sk_buff *skb, 2287 struct packet_type **pt, 2288 struct net_device *orig_dev, 2289 __be16 type, 2290 struct list_head *ptype_list) 2291 { 2292 struct packet_type *ptype, *pt_prev = *pt; 2293 2294 list_for_each_entry_rcu(ptype, ptype_list, list) { 2295 if (ptype->type != type) 2296 continue; 2297 if (pt_prev) 2298 deliver_skb(skb, pt_prev, orig_dev); 2299 pt_prev = ptype; 2300 } 2301 *pt = pt_prev; 2302 } 2303 2304 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) 2305 { 2306 if (!ptype->af_packet_priv || !skb->sk) 2307 return false; 2308 2309 if (ptype->id_match) 2310 return ptype->id_match(ptype, skb->sk); 2311 else if ((struct sock *)ptype->af_packet_priv == skb->sk) 2312 return true; 2313 2314 return false; 2315 } 2316 2317 /** 2318 * dev_nit_active - return true if any network interface taps are in use 2319 * 2320 * @dev: network device to check for the presence of taps 2321 */ 2322 bool dev_nit_active(struct net_device *dev) 2323 { 2324 return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all); 2325 } 2326 EXPORT_SYMBOL_GPL(dev_nit_active); 2327 2328 /* 2329 * Support routine. Sends outgoing frames to any network 2330 * taps currently in use. 2331 */ 2332 2333 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) 2334 { 2335 struct packet_type *ptype; 2336 struct sk_buff *skb2 = NULL; 2337 struct packet_type *pt_prev = NULL; 2338 struct list_head *ptype_list = &ptype_all; 2339 2340 rcu_read_lock(); 2341 again: 2342 list_for_each_entry_rcu(ptype, ptype_list, list) { 2343 if (ptype->ignore_outgoing) 2344 continue; 2345 2346 /* Never send packets back to the socket 2347 * they originated from - MvS (miquels@drinkel.ow.org) 2348 */ 2349 if (skb_loop_sk(ptype, skb)) 2350 continue; 2351 2352 if (pt_prev) { 2353 deliver_skb(skb2, pt_prev, skb->dev); 2354 pt_prev = ptype; 2355 continue; 2356 } 2357 2358 /* need to clone skb, done only once */ 2359 skb2 = skb_clone(skb, GFP_ATOMIC); 2360 if (!skb2) 2361 goto out_unlock; 2362 2363 net_timestamp_set(skb2); 2364 2365 /* skb->nh should be correctly 2366 * set by sender, so that the second statement is 2367 * just protection against buggy protocols. 2368 */ 2369 skb_reset_mac_header(skb2); 2370 2371 if (skb_network_header(skb2) < skb2->data || 2372 skb_network_header(skb2) > skb_tail_pointer(skb2)) { 2373 net_crit_ratelimited("protocol %04x is buggy, dev %s\n", 2374 ntohs(skb2->protocol), 2375 dev->name); 2376 skb_reset_network_header(skb2); 2377 } 2378 2379 skb2->transport_header = skb2->network_header; 2380 skb2->pkt_type = PACKET_OUTGOING; 2381 pt_prev = ptype; 2382 } 2383 2384 if (ptype_list == &ptype_all) { 2385 ptype_list = &dev->ptype_all; 2386 goto again; 2387 } 2388 out_unlock: 2389 if (pt_prev) { 2390 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC)) 2391 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); 2392 else 2393 kfree_skb(skb2); 2394 } 2395 rcu_read_unlock(); 2396 } 2397 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit); 2398 2399 /** 2400 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change 2401 * @dev: Network device 2402 * @txq: number of queues available 2403 * 2404 * If real_num_tx_queues is changed the tc mappings may no longer be 2405 * valid. To resolve this verify the tc mapping remains valid and if 2406 * not NULL the mapping. With no priorities mapping to this 2407 * offset/count pair it will no longer be used. In the worst case TC0 2408 * is invalid nothing can be done so disable priority mappings. If is 2409 * expected that drivers will fix this mapping if they can before 2410 * calling netif_set_real_num_tx_queues. 2411 */ 2412 static void netif_setup_tc(struct net_device *dev, unsigned int txq) 2413 { 2414 int i; 2415 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2416 2417 /* If TC0 is invalidated disable TC mapping */ 2418 if (tc->offset + tc->count > txq) { 2419 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); 2420 dev->num_tc = 0; 2421 return; 2422 } 2423 2424 /* Invalidated prio to tc mappings set to TC0 */ 2425 for (i = 1; i < TC_BITMASK + 1; i++) { 2426 int q = netdev_get_prio_tc_map(dev, i); 2427 2428 tc = &dev->tc_to_txq[q]; 2429 if (tc->offset + tc->count > txq) { 2430 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n", 2431 i, q); 2432 netdev_set_prio_tc_map(dev, i, 0); 2433 } 2434 } 2435 } 2436 2437 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq) 2438 { 2439 if (dev->num_tc) { 2440 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2441 int i; 2442 2443 /* walk through the TCs and see if it falls into any of them */ 2444 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) { 2445 if ((txq - tc->offset) < tc->count) 2446 return i; 2447 } 2448 2449 /* didn't find it, just return -1 to indicate no match */ 2450 return -1; 2451 } 2452 2453 return 0; 2454 } 2455 EXPORT_SYMBOL(netdev_txq_to_tc); 2456 2457 #ifdef CONFIG_XPS 2458 struct static_key xps_needed __read_mostly; 2459 EXPORT_SYMBOL(xps_needed); 2460 struct static_key xps_rxqs_needed __read_mostly; 2461 EXPORT_SYMBOL(xps_rxqs_needed); 2462 static DEFINE_MUTEX(xps_map_mutex); 2463 #define xmap_dereference(P) \ 2464 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) 2465 2466 static bool remove_xps_queue(struct xps_dev_maps *dev_maps, 2467 int tci, u16 index) 2468 { 2469 struct xps_map *map = NULL; 2470 int pos; 2471 2472 if (dev_maps) 2473 map = xmap_dereference(dev_maps->attr_map[tci]); 2474 if (!map) 2475 return false; 2476 2477 for (pos = map->len; pos--;) { 2478 if (map->queues[pos] != index) 2479 continue; 2480 2481 if (map->len > 1) { 2482 map->queues[pos] = map->queues[--map->len]; 2483 break; 2484 } 2485 2486 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2487 kfree_rcu(map, rcu); 2488 return false; 2489 } 2490 2491 return true; 2492 } 2493 2494 static bool remove_xps_queue_cpu(struct net_device *dev, 2495 struct xps_dev_maps *dev_maps, 2496 int cpu, u16 offset, u16 count) 2497 { 2498 int num_tc = dev->num_tc ? : 1; 2499 bool active = false; 2500 int tci; 2501 2502 for (tci = cpu * num_tc; num_tc--; tci++) { 2503 int i, j; 2504 2505 for (i = count, j = offset; i--; j++) { 2506 if (!remove_xps_queue(dev_maps, tci, j)) 2507 break; 2508 } 2509 2510 active |= i < 0; 2511 } 2512 2513 return active; 2514 } 2515 2516 static void reset_xps_maps(struct net_device *dev, 2517 struct xps_dev_maps *dev_maps, 2518 bool is_rxqs_map) 2519 { 2520 if (is_rxqs_map) { 2521 static_key_slow_dec_cpuslocked(&xps_rxqs_needed); 2522 RCU_INIT_POINTER(dev->xps_rxqs_map, NULL); 2523 } else { 2524 RCU_INIT_POINTER(dev->xps_cpus_map, NULL); 2525 } 2526 static_key_slow_dec_cpuslocked(&xps_needed); 2527 kfree_rcu(dev_maps, rcu); 2528 } 2529 2530 static void clean_xps_maps(struct net_device *dev, const unsigned long *mask, 2531 struct xps_dev_maps *dev_maps, unsigned int nr_ids, 2532 u16 offset, u16 count, bool is_rxqs_map) 2533 { 2534 bool active = false; 2535 int i, j; 2536 2537 for (j = -1; j = netif_attrmask_next(j, mask, nr_ids), 2538 j < nr_ids;) 2539 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, 2540 count); 2541 if (!active) 2542 reset_xps_maps(dev, dev_maps, is_rxqs_map); 2543 2544 if (!is_rxqs_map) { 2545 for (i = offset + (count - 1); count--; i--) { 2546 netdev_queue_numa_node_write( 2547 netdev_get_tx_queue(dev, i), 2548 NUMA_NO_NODE); 2549 } 2550 } 2551 } 2552 2553 static void netif_reset_xps_queues(struct net_device *dev, u16 offset, 2554 u16 count) 2555 { 2556 const unsigned long *possible_mask = NULL; 2557 struct xps_dev_maps *dev_maps; 2558 unsigned int nr_ids; 2559 2560 if (!static_key_false(&xps_needed)) 2561 return; 2562 2563 cpus_read_lock(); 2564 mutex_lock(&xps_map_mutex); 2565 2566 if (static_key_false(&xps_rxqs_needed)) { 2567 dev_maps = xmap_dereference(dev->xps_rxqs_map); 2568 if (dev_maps) { 2569 nr_ids = dev->num_rx_queues; 2570 clean_xps_maps(dev, possible_mask, dev_maps, nr_ids, 2571 offset, count, true); 2572 } 2573 } 2574 2575 dev_maps = xmap_dereference(dev->xps_cpus_map); 2576 if (!dev_maps) 2577 goto out_no_maps; 2578 2579 if (num_possible_cpus() > 1) 2580 possible_mask = cpumask_bits(cpu_possible_mask); 2581 nr_ids = nr_cpu_ids; 2582 clean_xps_maps(dev, possible_mask, dev_maps, nr_ids, offset, count, 2583 false); 2584 2585 out_no_maps: 2586 mutex_unlock(&xps_map_mutex); 2587 cpus_read_unlock(); 2588 } 2589 2590 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) 2591 { 2592 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index); 2593 } 2594 2595 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index, 2596 u16 index, bool is_rxqs_map) 2597 { 2598 struct xps_map *new_map; 2599 int alloc_len = XPS_MIN_MAP_ALLOC; 2600 int i, pos; 2601 2602 for (pos = 0; map && pos < map->len; pos++) { 2603 if (map->queues[pos] != index) 2604 continue; 2605 return map; 2606 } 2607 2608 /* Need to add tx-queue to this CPU's/rx-queue's existing map */ 2609 if (map) { 2610 if (pos < map->alloc_len) 2611 return map; 2612 2613 alloc_len = map->alloc_len * 2; 2614 } 2615 2616 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's 2617 * map 2618 */ 2619 if (is_rxqs_map) 2620 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL); 2621 else 2622 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, 2623 cpu_to_node(attr_index)); 2624 if (!new_map) 2625 return NULL; 2626 2627 for (i = 0; i < pos; i++) 2628 new_map->queues[i] = map->queues[i]; 2629 new_map->alloc_len = alloc_len; 2630 new_map->len = pos; 2631 2632 return new_map; 2633 } 2634 2635 /* Must be called under cpus_read_lock */ 2636 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask, 2637 u16 index, bool is_rxqs_map) 2638 { 2639 const unsigned long *online_mask = NULL, *possible_mask = NULL; 2640 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL; 2641 int i, j, tci, numa_node_id = -2; 2642 int maps_sz, num_tc = 1, tc = 0; 2643 struct xps_map *map, *new_map; 2644 bool active = false; 2645 unsigned int nr_ids; 2646 2647 if (dev->num_tc) { 2648 /* Do not allow XPS on subordinate device directly */ 2649 num_tc = dev->num_tc; 2650 if (num_tc < 0) 2651 return -EINVAL; 2652 2653 /* If queue belongs to subordinate dev use its map */ 2654 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev; 2655 2656 tc = netdev_txq_to_tc(dev, index); 2657 if (tc < 0) 2658 return -EINVAL; 2659 } 2660 2661 mutex_lock(&xps_map_mutex); 2662 if (is_rxqs_map) { 2663 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues); 2664 dev_maps = xmap_dereference(dev->xps_rxqs_map); 2665 nr_ids = dev->num_rx_queues; 2666 } else { 2667 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc); 2668 if (num_possible_cpus() > 1) { 2669 online_mask = cpumask_bits(cpu_online_mask); 2670 possible_mask = cpumask_bits(cpu_possible_mask); 2671 } 2672 dev_maps = xmap_dereference(dev->xps_cpus_map); 2673 nr_ids = nr_cpu_ids; 2674 } 2675 2676 if (maps_sz < L1_CACHE_BYTES) 2677 maps_sz = L1_CACHE_BYTES; 2678 2679 /* allocate memory for queue storage */ 2680 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids), 2681 j < nr_ids;) { 2682 if (!new_dev_maps) 2683 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); 2684 if (!new_dev_maps) { 2685 mutex_unlock(&xps_map_mutex); 2686 return -ENOMEM; 2687 } 2688 2689 tci = j * num_tc + tc; 2690 map = dev_maps ? xmap_dereference(dev_maps->attr_map[tci]) : 2691 NULL; 2692 2693 map = expand_xps_map(map, j, index, is_rxqs_map); 2694 if (!map) 2695 goto error; 2696 2697 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2698 } 2699 2700 if (!new_dev_maps) 2701 goto out_no_new_maps; 2702 2703 if (!dev_maps) { 2704 /* Increment static keys at most once per type */ 2705 static_key_slow_inc_cpuslocked(&xps_needed); 2706 if (is_rxqs_map) 2707 static_key_slow_inc_cpuslocked(&xps_rxqs_needed); 2708 } 2709 2710 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids), 2711 j < nr_ids;) { 2712 /* copy maps belonging to foreign traffic classes */ 2713 for (i = tc, tci = j * num_tc; dev_maps && i--; tci++) { 2714 /* fill in the new device map from the old device map */ 2715 map = xmap_dereference(dev_maps->attr_map[tci]); 2716 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2717 } 2718 2719 /* We need to explicitly update tci as prevous loop 2720 * could break out early if dev_maps is NULL. 2721 */ 2722 tci = j * num_tc + tc; 2723 2724 if (netif_attr_test_mask(j, mask, nr_ids) && 2725 netif_attr_test_online(j, online_mask, nr_ids)) { 2726 /* add tx-queue to CPU/rx-queue maps */ 2727 int pos = 0; 2728 2729 map = xmap_dereference(new_dev_maps->attr_map[tci]); 2730 while ((pos < map->len) && (map->queues[pos] != index)) 2731 pos++; 2732 2733 if (pos == map->len) 2734 map->queues[map->len++] = index; 2735 #ifdef CONFIG_NUMA 2736 if (!is_rxqs_map) { 2737 if (numa_node_id == -2) 2738 numa_node_id = cpu_to_node(j); 2739 else if (numa_node_id != cpu_to_node(j)) 2740 numa_node_id = -1; 2741 } 2742 #endif 2743 } else if (dev_maps) { 2744 /* fill in the new device map from the old device map */ 2745 map = xmap_dereference(dev_maps->attr_map[tci]); 2746 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2747 } 2748 2749 /* copy maps belonging to foreign traffic classes */ 2750 for (i = num_tc - tc, tci++; dev_maps && --i; tci++) { 2751 /* fill in the new device map from the old device map */ 2752 map = xmap_dereference(dev_maps->attr_map[tci]); 2753 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2754 } 2755 } 2756 2757 if (is_rxqs_map) 2758 rcu_assign_pointer(dev->xps_rxqs_map, new_dev_maps); 2759 else 2760 rcu_assign_pointer(dev->xps_cpus_map, new_dev_maps); 2761 2762 /* Cleanup old maps */ 2763 if (!dev_maps) 2764 goto out_no_old_maps; 2765 2766 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids), 2767 j < nr_ids;) { 2768 for (i = num_tc, tci = j * num_tc; i--; tci++) { 2769 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2770 map = xmap_dereference(dev_maps->attr_map[tci]); 2771 if (map && map != new_map) 2772 kfree_rcu(map, rcu); 2773 } 2774 } 2775 2776 kfree_rcu(dev_maps, rcu); 2777 2778 out_no_old_maps: 2779 dev_maps = new_dev_maps; 2780 active = true; 2781 2782 out_no_new_maps: 2783 if (!is_rxqs_map) { 2784 /* update Tx queue numa node */ 2785 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), 2786 (numa_node_id >= 0) ? 2787 numa_node_id : NUMA_NO_NODE); 2788 } 2789 2790 if (!dev_maps) 2791 goto out_no_maps; 2792 2793 /* removes tx-queue from unused CPUs/rx-queues */ 2794 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids), 2795 j < nr_ids;) { 2796 for (i = tc, tci = j * num_tc; i--; tci++) 2797 active |= remove_xps_queue(dev_maps, tci, index); 2798 if (!netif_attr_test_mask(j, mask, nr_ids) || 2799 !netif_attr_test_online(j, online_mask, nr_ids)) 2800 active |= remove_xps_queue(dev_maps, tci, index); 2801 for (i = num_tc - tc, tci++; --i; tci++) 2802 active |= remove_xps_queue(dev_maps, tci, index); 2803 } 2804 2805 /* free map if not active */ 2806 if (!active) 2807 reset_xps_maps(dev, dev_maps, is_rxqs_map); 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 = -1; j = netif_attrmask_next(j, possible_mask, nr_ids), 2816 j < nr_ids;) { 2817 for (i = num_tc, tci = j * num_tc; i--; tci++) { 2818 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2819 map = dev_maps ? 2820 xmap_dereference(dev_maps->attr_map[tci]) : 2821 NULL; 2822 if (new_map && new_map != map) 2823 kfree(new_map); 2824 } 2825 } 2826 2827 mutex_unlock(&xps_map_mutex); 2828 2829 kfree(new_dev_maps); 2830 return -ENOMEM; 2831 } 2832 EXPORT_SYMBOL_GPL(__netif_set_xps_queue); 2833 2834 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, 2835 u16 index) 2836 { 2837 int ret; 2838 2839 cpus_read_lock(); 2840 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, false); 2841 cpus_read_unlock(); 2842 2843 return ret; 2844 } 2845 EXPORT_SYMBOL(netif_set_xps_queue); 2846 2847 #endif 2848 static void netdev_unbind_all_sb_channels(struct net_device *dev) 2849 { 2850 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2851 2852 /* Unbind any subordinate channels */ 2853 while (txq-- != &dev->_tx[0]) { 2854 if (txq->sb_dev) 2855 netdev_unbind_sb_channel(dev, txq->sb_dev); 2856 } 2857 } 2858 2859 void netdev_reset_tc(struct net_device *dev) 2860 { 2861 #ifdef CONFIG_XPS 2862 netif_reset_xps_queues_gt(dev, 0); 2863 #endif 2864 netdev_unbind_all_sb_channels(dev); 2865 2866 /* Reset TC configuration of device */ 2867 dev->num_tc = 0; 2868 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq)); 2869 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map)); 2870 } 2871 EXPORT_SYMBOL(netdev_reset_tc); 2872 2873 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset) 2874 { 2875 if (tc >= dev->num_tc) 2876 return -EINVAL; 2877 2878 #ifdef CONFIG_XPS 2879 netif_reset_xps_queues(dev, offset, count); 2880 #endif 2881 dev->tc_to_txq[tc].count = count; 2882 dev->tc_to_txq[tc].offset = offset; 2883 return 0; 2884 } 2885 EXPORT_SYMBOL(netdev_set_tc_queue); 2886 2887 int netdev_set_num_tc(struct net_device *dev, u8 num_tc) 2888 { 2889 if (num_tc > TC_MAX_QUEUE) 2890 return -EINVAL; 2891 2892 #ifdef CONFIG_XPS 2893 netif_reset_xps_queues_gt(dev, 0); 2894 #endif 2895 netdev_unbind_all_sb_channels(dev); 2896 2897 dev->num_tc = num_tc; 2898 return 0; 2899 } 2900 EXPORT_SYMBOL(netdev_set_num_tc); 2901 2902 void netdev_unbind_sb_channel(struct net_device *dev, 2903 struct net_device *sb_dev) 2904 { 2905 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2906 2907 #ifdef CONFIG_XPS 2908 netif_reset_xps_queues_gt(sb_dev, 0); 2909 #endif 2910 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq)); 2911 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map)); 2912 2913 while (txq-- != &dev->_tx[0]) { 2914 if (txq->sb_dev == sb_dev) 2915 txq->sb_dev = NULL; 2916 } 2917 } 2918 EXPORT_SYMBOL(netdev_unbind_sb_channel); 2919 2920 int netdev_bind_sb_channel_queue(struct net_device *dev, 2921 struct net_device *sb_dev, 2922 u8 tc, u16 count, u16 offset) 2923 { 2924 /* Make certain the sb_dev and dev are already configured */ 2925 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc) 2926 return -EINVAL; 2927 2928 /* We cannot hand out queues we don't have */ 2929 if ((offset + count) > dev->real_num_tx_queues) 2930 return -EINVAL; 2931 2932 /* Record the mapping */ 2933 sb_dev->tc_to_txq[tc].count = count; 2934 sb_dev->tc_to_txq[tc].offset = offset; 2935 2936 /* Provide a way for Tx queue to find the tc_to_txq map or 2937 * XPS map for itself. 2938 */ 2939 while (count--) 2940 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev; 2941 2942 return 0; 2943 } 2944 EXPORT_SYMBOL(netdev_bind_sb_channel_queue); 2945 2946 int netdev_set_sb_channel(struct net_device *dev, u16 channel) 2947 { 2948 /* Do not use a multiqueue device to represent a subordinate channel */ 2949 if (netif_is_multiqueue(dev)) 2950 return -ENODEV; 2951 2952 /* We allow channels 1 - 32767 to be used for subordinate channels. 2953 * Channel 0 is meant to be "native" mode and used only to represent 2954 * the main root device. We allow writing 0 to reset the device back 2955 * to normal mode after being used as a subordinate channel. 2956 */ 2957 if (channel > S16_MAX) 2958 return -EINVAL; 2959 2960 dev->num_tc = -channel; 2961 2962 return 0; 2963 } 2964 EXPORT_SYMBOL(netdev_set_sb_channel); 2965 2966 /* 2967 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues 2968 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed. 2969 */ 2970 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) 2971 { 2972 bool disabling; 2973 int rc; 2974 2975 disabling = txq < dev->real_num_tx_queues; 2976 2977 if (txq < 1 || txq > dev->num_tx_queues) 2978 return -EINVAL; 2979 2980 if (dev->reg_state == NETREG_REGISTERED || 2981 dev->reg_state == NETREG_UNREGISTERING) { 2982 ASSERT_RTNL(); 2983 2984 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, 2985 txq); 2986 if (rc) 2987 return rc; 2988 2989 if (dev->num_tc) 2990 netif_setup_tc(dev, txq); 2991 2992 dev->real_num_tx_queues = txq; 2993 2994 if (disabling) { 2995 synchronize_net(); 2996 qdisc_reset_all_tx_gt(dev, txq); 2997 #ifdef CONFIG_XPS 2998 netif_reset_xps_queues_gt(dev, txq); 2999 #endif 3000 } 3001 } else { 3002 dev->real_num_tx_queues = txq; 3003 } 3004 3005 return 0; 3006 } 3007 EXPORT_SYMBOL(netif_set_real_num_tx_queues); 3008 3009 #ifdef CONFIG_SYSFS 3010 /** 3011 * netif_set_real_num_rx_queues - set actual number of RX queues used 3012 * @dev: Network device 3013 * @rxq: Actual number of RX queues 3014 * 3015 * This must be called either with the rtnl_lock held or before 3016 * registration of the net device. Returns 0 on success, or a 3017 * negative error code. If called before registration, it always 3018 * succeeds. 3019 */ 3020 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) 3021 { 3022 int rc; 3023 3024 if (rxq < 1 || rxq > dev->num_rx_queues) 3025 return -EINVAL; 3026 3027 if (dev->reg_state == NETREG_REGISTERED) { 3028 ASSERT_RTNL(); 3029 3030 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, 3031 rxq); 3032 if (rc) 3033 return rc; 3034 } 3035 3036 dev->real_num_rx_queues = rxq; 3037 return 0; 3038 } 3039 EXPORT_SYMBOL(netif_set_real_num_rx_queues); 3040 #endif 3041 3042 /** 3043 * netif_get_num_default_rss_queues - default number of RSS queues 3044 * 3045 * This routine should set an upper limit on the number of RSS queues 3046 * used by default by multiqueue devices. 3047 */ 3048 int netif_get_num_default_rss_queues(void) 3049 { 3050 return is_kdump_kernel() ? 3051 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus()); 3052 } 3053 EXPORT_SYMBOL(netif_get_num_default_rss_queues); 3054 3055 static void __netif_reschedule(struct Qdisc *q) 3056 { 3057 struct softnet_data *sd; 3058 unsigned long flags; 3059 3060 local_irq_save(flags); 3061 sd = this_cpu_ptr(&softnet_data); 3062 q->next_sched = NULL; 3063 *sd->output_queue_tailp = q; 3064 sd->output_queue_tailp = &q->next_sched; 3065 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3066 local_irq_restore(flags); 3067 } 3068 3069 void __netif_schedule(struct Qdisc *q) 3070 { 3071 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) 3072 __netif_reschedule(q); 3073 } 3074 EXPORT_SYMBOL(__netif_schedule); 3075 3076 struct dev_kfree_skb_cb { 3077 enum skb_free_reason reason; 3078 }; 3079 3080 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) 3081 { 3082 return (struct dev_kfree_skb_cb *)skb->cb; 3083 } 3084 3085 void netif_schedule_queue(struct netdev_queue *txq) 3086 { 3087 rcu_read_lock(); 3088 if (!netif_xmit_stopped(txq)) { 3089 struct Qdisc *q = rcu_dereference(txq->qdisc); 3090 3091 __netif_schedule(q); 3092 } 3093 rcu_read_unlock(); 3094 } 3095 EXPORT_SYMBOL(netif_schedule_queue); 3096 3097 void netif_tx_wake_queue(struct netdev_queue *dev_queue) 3098 { 3099 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { 3100 struct Qdisc *q; 3101 3102 rcu_read_lock(); 3103 q = rcu_dereference(dev_queue->qdisc); 3104 __netif_schedule(q); 3105 rcu_read_unlock(); 3106 } 3107 } 3108 EXPORT_SYMBOL(netif_tx_wake_queue); 3109 3110 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason) 3111 { 3112 unsigned long flags; 3113 3114 if (unlikely(!skb)) 3115 return; 3116 3117 if (likely(refcount_read(&skb->users) == 1)) { 3118 smp_rmb(); 3119 refcount_set(&skb->users, 0); 3120 } else if (likely(!refcount_dec_and_test(&skb->users))) { 3121 return; 3122 } 3123 get_kfree_skb_cb(skb)->reason = reason; 3124 local_irq_save(flags); 3125 skb->next = __this_cpu_read(softnet_data.completion_queue); 3126 __this_cpu_write(softnet_data.completion_queue, skb); 3127 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3128 local_irq_restore(flags); 3129 } 3130 EXPORT_SYMBOL(__dev_kfree_skb_irq); 3131 3132 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason) 3133 { 3134 if (in_irq() || irqs_disabled()) 3135 __dev_kfree_skb_irq(skb, reason); 3136 else 3137 dev_kfree_skb(skb); 3138 } 3139 EXPORT_SYMBOL(__dev_kfree_skb_any); 3140 3141 3142 /** 3143 * netif_device_detach - mark device as removed 3144 * @dev: network device 3145 * 3146 * Mark device as removed from system and therefore no longer available. 3147 */ 3148 void netif_device_detach(struct net_device *dev) 3149 { 3150 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && 3151 netif_running(dev)) { 3152 netif_tx_stop_all_queues(dev); 3153 } 3154 } 3155 EXPORT_SYMBOL(netif_device_detach); 3156 3157 /** 3158 * netif_device_attach - mark device as attached 3159 * @dev: network device 3160 * 3161 * Mark device as attached from system and restart if needed. 3162 */ 3163 void netif_device_attach(struct net_device *dev) 3164 { 3165 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && 3166 netif_running(dev)) { 3167 netif_tx_wake_all_queues(dev); 3168 __netdev_watchdog_up(dev); 3169 } 3170 } 3171 EXPORT_SYMBOL(netif_device_attach); 3172 3173 /* 3174 * Returns a Tx hash based on the given packet descriptor a Tx queues' number 3175 * to be used as a distribution range. 3176 */ 3177 static u16 skb_tx_hash(const struct net_device *dev, 3178 const struct net_device *sb_dev, 3179 struct sk_buff *skb) 3180 { 3181 u32 hash; 3182 u16 qoffset = 0; 3183 u16 qcount = dev->real_num_tx_queues; 3184 3185 if (dev->num_tc) { 3186 u8 tc = netdev_get_prio_tc_map(dev, skb->priority); 3187 3188 qoffset = sb_dev->tc_to_txq[tc].offset; 3189 qcount = sb_dev->tc_to_txq[tc].count; 3190 } 3191 3192 if (skb_rx_queue_recorded(skb)) { 3193 hash = skb_get_rx_queue(skb); 3194 if (hash >= qoffset) 3195 hash -= qoffset; 3196 while (unlikely(hash >= qcount)) 3197 hash -= qcount; 3198 return hash + qoffset; 3199 } 3200 3201 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; 3202 } 3203 3204 static void skb_warn_bad_offload(const struct sk_buff *skb) 3205 { 3206 static const netdev_features_t null_features; 3207 struct net_device *dev = skb->dev; 3208 const char *name = ""; 3209 3210 if (!net_ratelimit()) 3211 return; 3212 3213 if (dev) { 3214 if (dev->dev.parent) 3215 name = dev_driver_string(dev->dev.parent); 3216 else 3217 name = netdev_name(dev); 3218 } 3219 skb_dump(KERN_WARNING, skb, false); 3220 WARN(1, "%s: caps=(%pNF, %pNF)\n", 3221 name, dev ? &dev->features : &null_features, 3222 skb->sk ? &skb->sk->sk_route_caps : &null_features); 3223 } 3224 3225 /* 3226 * Invalidate hardware checksum when packet is to be mangled, and 3227 * complete checksum manually on outgoing path. 3228 */ 3229 int skb_checksum_help(struct sk_buff *skb) 3230 { 3231 __wsum csum; 3232 int ret = 0, offset; 3233 3234 if (skb->ip_summed == CHECKSUM_COMPLETE) 3235 goto out_set_summed; 3236 3237 if (unlikely(skb_is_gso(skb))) { 3238 skb_warn_bad_offload(skb); 3239 return -EINVAL; 3240 } 3241 3242 /* Before computing a checksum, we should make sure no frag could 3243 * be modified by an external entity : checksum could be wrong. 3244 */ 3245 if (skb_has_shared_frag(skb)) { 3246 ret = __skb_linearize(skb); 3247 if (ret) 3248 goto out; 3249 } 3250 3251 offset = skb_checksum_start_offset(skb); 3252 BUG_ON(offset >= skb_headlen(skb)); 3253 csum = skb_checksum(skb, offset, skb->len - offset, 0); 3254 3255 offset += skb->csum_offset; 3256 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); 3257 3258 ret = skb_ensure_writable(skb, offset + sizeof(__sum16)); 3259 if (ret) 3260 goto out; 3261 3262 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0; 3263 out_set_summed: 3264 skb->ip_summed = CHECKSUM_NONE; 3265 out: 3266 return ret; 3267 } 3268 EXPORT_SYMBOL(skb_checksum_help); 3269 3270 int skb_crc32c_csum_help(struct sk_buff *skb) 3271 { 3272 __le32 crc32c_csum; 3273 int ret = 0, offset, start; 3274 3275 if (skb->ip_summed != CHECKSUM_PARTIAL) 3276 goto out; 3277 3278 if (unlikely(skb_is_gso(skb))) 3279 goto out; 3280 3281 /* Before computing a checksum, we should make sure no frag could 3282 * be modified by an external entity : checksum could be wrong. 3283 */ 3284 if (unlikely(skb_has_shared_frag(skb))) { 3285 ret = __skb_linearize(skb); 3286 if (ret) 3287 goto out; 3288 } 3289 start = skb_checksum_start_offset(skb); 3290 offset = start + offsetof(struct sctphdr, checksum); 3291 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { 3292 ret = -EINVAL; 3293 goto out; 3294 } 3295 3296 ret = skb_ensure_writable(skb, offset + sizeof(__le32)); 3297 if (ret) 3298 goto out; 3299 3300 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start, 3301 skb->len - start, ~(__u32)0, 3302 crc32c_csum_stub)); 3303 *(__le32 *)(skb->data + offset) = crc32c_csum; 3304 skb->ip_summed = CHECKSUM_NONE; 3305 skb->csum_not_inet = 0; 3306 out: 3307 return ret; 3308 } 3309 3310 __be16 skb_network_protocol(struct sk_buff *skb, int *depth) 3311 { 3312 __be16 type = skb->protocol; 3313 3314 /* Tunnel gso handlers can set protocol to ethernet. */ 3315 if (type == htons(ETH_P_TEB)) { 3316 struct ethhdr *eth; 3317 3318 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) 3319 return 0; 3320 3321 eth = (struct ethhdr *)skb->data; 3322 type = eth->h_proto; 3323 } 3324 3325 return __vlan_get_protocol(skb, type, depth); 3326 } 3327 3328 /** 3329 * skb_mac_gso_segment - mac layer segmentation handler. 3330 * @skb: buffer to segment 3331 * @features: features for the output path (see dev->features) 3332 */ 3333 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb, 3334 netdev_features_t features) 3335 { 3336 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); 3337 struct packet_offload *ptype; 3338 int vlan_depth = skb->mac_len; 3339 __be16 type = skb_network_protocol(skb, &vlan_depth); 3340 3341 if (unlikely(!type)) 3342 return ERR_PTR(-EINVAL); 3343 3344 __skb_pull(skb, vlan_depth); 3345 3346 rcu_read_lock(); 3347 list_for_each_entry_rcu(ptype, &offload_base, list) { 3348 if (ptype->type == type && ptype->callbacks.gso_segment) { 3349 segs = ptype->callbacks.gso_segment(skb, features); 3350 break; 3351 } 3352 } 3353 rcu_read_unlock(); 3354 3355 __skb_push(skb, skb->data - skb_mac_header(skb)); 3356 3357 return segs; 3358 } 3359 EXPORT_SYMBOL(skb_mac_gso_segment); 3360 3361 3362 /* openvswitch calls this on rx path, so we need a different check. 3363 */ 3364 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path) 3365 { 3366 if (tx_path) 3367 return skb->ip_summed != CHECKSUM_PARTIAL && 3368 skb->ip_summed != CHECKSUM_UNNECESSARY; 3369 3370 return skb->ip_summed == CHECKSUM_NONE; 3371 } 3372 3373 /** 3374 * __skb_gso_segment - Perform segmentation on skb. 3375 * @skb: buffer to segment 3376 * @features: features for the output path (see dev->features) 3377 * @tx_path: whether it is called in TX path 3378 * 3379 * This function segments the given skb and returns a list of segments. 3380 * 3381 * It may return NULL if the skb requires no segmentation. This is 3382 * only possible when GSO is used for verifying header integrity. 3383 * 3384 * Segmentation preserves SKB_GSO_CB_OFFSET bytes of previous skb cb. 3385 */ 3386 struct sk_buff *__skb_gso_segment(struct sk_buff *skb, 3387 netdev_features_t features, bool tx_path) 3388 { 3389 struct sk_buff *segs; 3390 3391 if (unlikely(skb_needs_check(skb, tx_path))) { 3392 int err; 3393 3394 /* We're going to init ->check field in TCP or UDP header */ 3395 err = skb_cow_head(skb, 0); 3396 if (err < 0) 3397 return ERR_PTR(err); 3398 } 3399 3400 /* Only report GSO partial support if it will enable us to 3401 * support segmentation on this frame without needing additional 3402 * work. 3403 */ 3404 if (features & NETIF_F_GSO_PARTIAL) { 3405 netdev_features_t partial_features = NETIF_F_GSO_ROBUST; 3406 struct net_device *dev = skb->dev; 3407 3408 partial_features |= dev->features & dev->gso_partial_features; 3409 if (!skb_gso_ok(skb, features | partial_features)) 3410 features &= ~NETIF_F_GSO_PARTIAL; 3411 } 3412 3413 BUILD_BUG_ON(SKB_GSO_CB_OFFSET + 3414 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb)); 3415 3416 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb); 3417 SKB_GSO_CB(skb)->encap_level = 0; 3418 3419 skb_reset_mac_header(skb); 3420 skb_reset_mac_len(skb); 3421 3422 segs = skb_mac_gso_segment(skb, features); 3423 3424 if (segs != skb && unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs))) 3425 skb_warn_bad_offload(skb); 3426 3427 return segs; 3428 } 3429 EXPORT_SYMBOL(__skb_gso_segment); 3430 3431 /* Take action when hardware reception checksum errors are detected. */ 3432 #ifdef CONFIG_BUG 3433 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3434 { 3435 if (net_ratelimit()) { 3436 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>"); 3437 skb_dump(KERN_ERR, skb, true); 3438 dump_stack(); 3439 } 3440 } 3441 EXPORT_SYMBOL(netdev_rx_csum_fault); 3442 #endif 3443 3444 /* XXX: check that highmem exists at all on the given machine. */ 3445 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) 3446 { 3447 #ifdef CONFIG_HIGHMEM 3448 int i; 3449 3450 if (!(dev->features & NETIF_F_HIGHDMA)) { 3451 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 3452 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 3453 3454 if (PageHighMem(skb_frag_page(frag))) 3455 return 1; 3456 } 3457 } 3458 #endif 3459 return 0; 3460 } 3461 3462 /* If MPLS offload request, verify we are testing hardware MPLS features 3463 * instead of standard features for the netdev. 3464 */ 3465 #if IS_ENABLED(CONFIG_NET_MPLS_GSO) 3466 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3467 netdev_features_t features, 3468 __be16 type) 3469 { 3470 if (eth_p_mpls(type)) 3471 features &= skb->dev->mpls_features; 3472 3473 return features; 3474 } 3475 #else 3476 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3477 netdev_features_t features, 3478 __be16 type) 3479 { 3480 return features; 3481 } 3482 #endif 3483 3484 static netdev_features_t harmonize_features(struct sk_buff *skb, 3485 netdev_features_t features) 3486 { 3487 __be16 type; 3488 3489 type = skb_network_protocol(skb, NULL); 3490 features = net_mpls_features(skb, features, type); 3491 3492 if (skb->ip_summed != CHECKSUM_NONE && 3493 !can_checksum_protocol(features, type)) { 3494 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); 3495 } 3496 if (illegal_highdma(skb->dev, skb)) 3497 features &= ~NETIF_F_SG; 3498 3499 return features; 3500 } 3501 3502 netdev_features_t passthru_features_check(struct sk_buff *skb, 3503 struct net_device *dev, 3504 netdev_features_t features) 3505 { 3506 return features; 3507 } 3508 EXPORT_SYMBOL(passthru_features_check); 3509 3510 static netdev_features_t dflt_features_check(struct sk_buff *skb, 3511 struct net_device *dev, 3512 netdev_features_t features) 3513 { 3514 return vlan_features_check(skb, features); 3515 } 3516 3517 static netdev_features_t gso_features_check(const struct sk_buff *skb, 3518 struct net_device *dev, 3519 netdev_features_t features) 3520 { 3521 u16 gso_segs = skb_shinfo(skb)->gso_segs; 3522 3523 if (gso_segs > dev->gso_max_segs) 3524 return features & ~NETIF_F_GSO_MASK; 3525 3526 if (!skb_shinfo(skb)->gso_type) { 3527 skb_warn_bad_offload(skb); 3528 return features & ~NETIF_F_GSO_MASK; 3529 } 3530 3531 /* Support for GSO partial features requires software 3532 * intervention before we can actually process the packets 3533 * so we need to strip support for any partial features now 3534 * and we can pull them back in after we have partially 3535 * segmented the frame. 3536 */ 3537 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)) 3538 features &= ~dev->gso_partial_features; 3539 3540 /* Make sure to clear the IPv4 ID mangling feature if the 3541 * IPv4 header has the potential to be fragmented. 3542 */ 3543 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { 3544 struct iphdr *iph = skb->encapsulation ? 3545 inner_ip_hdr(skb) : ip_hdr(skb); 3546 3547 if (!(iph->frag_off & htons(IP_DF))) 3548 features &= ~NETIF_F_TSO_MANGLEID; 3549 } 3550 3551 return features; 3552 } 3553 3554 netdev_features_t netif_skb_features(struct sk_buff *skb) 3555 { 3556 struct net_device *dev = skb->dev; 3557 netdev_features_t features = dev->features; 3558 3559 if (skb_is_gso(skb)) 3560 features = gso_features_check(skb, dev, features); 3561 3562 /* If encapsulation offload request, verify we are testing 3563 * hardware encapsulation features instead of standard 3564 * features for the netdev 3565 */ 3566 if (skb->encapsulation) 3567 features &= dev->hw_enc_features; 3568 3569 if (skb_vlan_tagged(skb)) 3570 features = netdev_intersect_features(features, 3571 dev->vlan_features | 3572 NETIF_F_HW_VLAN_CTAG_TX | 3573 NETIF_F_HW_VLAN_STAG_TX); 3574 3575 if (dev->netdev_ops->ndo_features_check) 3576 features &= dev->netdev_ops->ndo_features_check(skb, dev, 3577 features); 3578 else 3579 features &= dflt_features_check(skb, dev, features); 3580 3581 return harmonize_features(skb, features); 3582 } 3583 EXPORT_SYMBOL(netif_skb_features); 3584 3585 static int xmit_one(struct sk_buff *skb, struct net_device *dev, 3586 struct netdev_queue *txq, bool more) 3587 { 3588 unsigned int len; 3589 int rc; 3590 3591 if (dev_nit_active(dev)) 3592 dev_queue_xmit_nit(skb, dev); 3593 3594 len = skb->len; 3595 PRANDOM_ADD_NOISE(skb, dev, txq, len + jiffies); 3596 trace_net_dev_start_xmit(skb, dev); 3597 rc = netdev_start_xmit(skb, dev, txq, more); 3598 trace_net_dev_xmit(skb, rc, dev, len); 3599 3600 return rc; 3601 } 3602 3603 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, 3604 struct netdev_queue *txq, int *ret) 3605 { 3606 struct sk_buff *skb = first; 3607 int rc = NETDEV_TX_OK; 3608 3609 while (skb) { 3610 struct sk_buff *next = skb->next; 3611 3612 skb_mark_not_on_list(skb); 3613 rc = xmit_one(skb, dev, txq, next != NULL); 3614 if (unlikely(!dev_xmit_complete(rc))) { 3615 skb->next = next; 3616 goto out; 3617 } 3618 3619 skb = next; 3620 if (netif_tx_queue_stopped(txq) && skb) { 3621 rc = NETDEV_TX_BUSY; 3622 break; 3623 } 3624 } 3625 3626 out: 3627 *ret = rc; 3628 return skb; 3629 } 3630 3631 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, 3632 netdev_features_t features) 3633 { 3634 if (skb_vlan_tag_present(skb) && 3635 !vlan_hw_offload_capable(features, skb->vlan_proto)) 3636 skb = __vlan_hwaccel_push_inside(skb); 3637 return skb; 3638 } 3639 3640 int skb_csum_hwoffload_help(struct sk_buff *skb, 3641 const netdev_features_t features) 3642 { 3643 if (unlikely(skb_csum_is_sctp(skb))) 3644 return !!(features & NETIF_F_SCTP_CRC) ? 0 : 3645 skb_crc32c_csum_help(skb); 3646 3647 if (features & NETIF_F_HW_CSUM) 3648 return 0; 3649 3650 if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) { 3651 switch (skb->csum_offset) { 3652 case offsetof(struct tcphdr, check): 3653 case offsetof(struct udphdr, check): 3654 return 0; 3655 } 3656 } 3657 3658 return skb_checksum_help(skb); 3659 } 3660 EXPORT_SYMBOL(skb_csum_hwoffload_help); 3661 3662 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again) 3663 { 3664 netdev_features_t features; 3665 3666 features = netif_skb_features(skb); 3667 skb = validate_xmit_vlan(skb, features); 3668 if (unlikely(!skb)) 3669 goto out_null; 3670 3671 skb = sk_validate_xmit_skb(skb, dev); 3672 if (unlikely(!skb)) 3673 goto out_null; 3674 3675 if (netif_needs_gso(skb, features)) { 3676 struct sk_buff *segs; 3677 3678 segs = skb_gso_segment(skb, features); 3679 if (IS_ERR(segs)) { 3680 goto out_kfree_skb; 3681 } else if (segs) { 3682 consume_skb(skb); 3683 skb = segs; 3684 } 3685 } else { 3686 if (skb_needs_linearize(skb, features) && 3687 __skb_linearize(skb)) 3688 goto out_kfree_skb; 3689 3690 /* If packet is not checksummed and device does not 3691 * support checksumming for this protocol, complete 3692 * checksumming here. 3693 */ 3694 if (skb->ip_summed == CHECKSUM_PARTIAL) { 3695 if (skb->encapsulation) 3696 skb_set_inner_transport_header(skb, 3697 skb_checksum_start_offset(skb)); 3698 else 3699 skb_set_transport_header(skb, 3700 skb_checksum_start_offset(skb)); 3701 if (skb_csum_hwoffload_help(skb, features)) 3702 goto out_kfree_skb; 3703 } 3704 } 3705 3706 skb = validate_xmit_xfrm(skb, features, again); 3707 3708 return skb; 3709 3710 out_kfree_skb: 3711 kfree_skb(skb); 3712 out_null: 3713 atomic_long_inc(&dev->tx_dropped); 3714 return NULL; 3715 } 3716 3717 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again) 3718 { 3719 struct sk_buff *next, *head = NULL, *tail; 3720 3721 for (; skb != NULL; skb = next) { 3722 next = skb->next; 3723 skb_mark_not_on_list(skb); 3724 3725 /* in case skb wont be segmented, point to itself */ 3726 skb->prev = skb; 3727 3728 skb = validate_xmit_skb(skb, dev, again); 3729 if (!skb) 3730 continue; 3731 3732 if (!head) 3733 head = skb; 3734 else 3735 tail->next = skb; 3736 /* If skb was segmented, skb->prev points to 3737 * the last segment. If not, it still contains skb. 3738 */ 3739 tail = skb->prev; 3740 } 3741 return head; 3742 } 3743 EXPORT_SYMBOL_GPL(validate_xmit_skb_list); 3744 3745 static void qdisc_pkt_len_init(struct sk_buff *skb) 3746 { 3747 const struct skb_shared_info *shinfo = skb_shinfo(skb); 3748 3749 qdisc_skb_cb(skb)->pkt_len = skb->len; 3750 3751 /* To get more precise estimation of bytes sent on wire, 3752 * we add to pkt_len the headers size of all segments 3753 */ 3754 if (shinfo->gso_size && skb_transport_header_was_set(skb)) { 3755 unsigned int hdr_len; 3756 u16 gso_segs = shinfo->gso_segs; 3757 3758 /* mac layer + network layer */ 3759 hdr_len = skb_transport_header(skb) - skb_mac_header(skb); 3760 3761 /* + transport layer */ 3762 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { 3763 const struct tcphdr *th; 3764 struct tcphdr _tcphdr; 3765 3766 th = skb_header_pointer(skb, skb_transport_offset(skb), 3767 sizeof(_tcphdr), &_tcphdr); 3768 if (likely(th)) 3769 hdr_len += __tcp_hdrlen(th); 3770 } else { 3771 struct udphdr _udphdr; 3772 3773 if (skb_header_pointer(skb, skb_transport_offset(skb), 3774 sizeof(_udphdr), &_udphdr)) 3775 hdr_len += sizeof(struct udphdr); 3776 } 3777 3778 if (shinfo->gso_type & SKB_GSO_DODGY) 3779 gso_segs = DIV_ROUND_UP(skb->len - hdr_len, 3780 shinfo->gso_size); 3781 3782 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; 3783 } 3784 } 3785 3786 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, 3787 struct net_device *dev, 3788 struct netdev_queue *txq) 3789 { 3790 spinlock_t *root_lock = qdisc_lock(q); 3791 struct sk_buff *to_free = NULL; 3792 bool contended; 3793 int rc; 3794 3795 qdisc_calculate_pkt_len(skb, q); 3796 3797 if (q->flags & TCQ_F_NOLOCK) { 3798 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK; 3799 qdisc_run(q); 3800 3801 if (unlikely(to_free)) 3802 kfree_skb_list(to_free); 3803 return rc; 3804 } 3805 3806 /* 3807 * Heuristic to force contended enqueues to serialize on a 3808 * separate lock before trying to get qdisc main lock. 3809 * This permits qdisc->running owner to get the lock more 3810 * often and dequeue packets faster. 3811 */ 3812 contended = qdisc_is_running(q); 3813 if (unlikely(contended)) 3814 spin_lock(&q->busylock); 3815 3816 spin_lock(root_lock); 3817 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { 3818 __qdisc_drop(skb, &to_free); 3819 rc = NET_XMIT_DROP; 3820 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && 3821 qdisc_run_begin(q)) { 3822 /* 3823 * This is a work-conserving queue; there are no old skbs 3824 * waiting to be sent out; and the qdisc is not running - 3825 * xmit the skb directly. 3826 */ 3827 3828 qdisc_bstats_update(q, skb); 3829 3830 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { 3831 if (unlikely(contended)) { 3832 spin_unlock(&q->busylock); 3833 contended = false; 3834 } 3835 __qdisc_run(q); 3836 } 3837 3838 qdisc_run_end(q); 3839 rc = NET_XMIT_SUCCESS; 3840 } else { 3841 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK; 3842 if (qdisc_run_begin(q)) { 3843 if (unlikely(contended)) { 3844 spin_unlock(&q->busylock); 3845 contended = false; 3846 } 3847 __qdisc_run(q); 3848 qdisc_run_end(q); 3849 } 3850 } 3851 spin_unlock(root_lock); 3852 if (unlikely(to_free)) 3853 kfree_skb_list(to_free); 3854 if (unlikely(contended)) 3855 spin_unlock(&q->busylock); 3856 return rc; 3857 } 3858 3859 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) 3860 static void skb_update_prio(struct sk_buff *skb) 3861 { 3862 const struct netprio_map *map; 3863 const struct sock *sk; 3864 unsigned int prioidx; 3865 3866 if (skb->priority) 3867 return; 3868 map = rcu_dereference_bh(skb->dev->priomap); 3869 if (!map) 3870 return; 3871 sk = skb_to_full_sk(skb); 3872 if (!sk) 3873 return; 3874 3875 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data); 3876 3877 if (prioidx < map->priomap_len) 3878 skb->priority = map->priomap[prioidx]; 3879 } 3880 #else 3881 #define skb_update_prio(skb) 3882 #endif 3883 3884 /** 3885 * dev_loopback_xmit - loop back @skb 3886 * @net: network namespace this loopback is happening in 3887 * @sk: sk needed to be a netfilter okfn 3888 * @skb: buffer to transmit 3889 */ 3890 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) 3891 { 3892 skb_reset_mac_header(skb); 3893 __skb_pull(skb, skb_network_offset(skb)); 3894 skb->pkt_type = PACKET_LOOPBACK; 3895 skb->ip_summed = CHECKSUM_UNNECESSARY; 3896 WARN_ON(!skb_dst(skb)); 3897 skb_dst_force(skb); 3898 netif_rx_ni(skb); 3899 return 0; 3900 } 3901 EXPORT_SYMBOL(dev_loopback_xmit); 3902 3903 #ifdef CONFIG_NET_EGRESS 3904 static struct sk_buff * 3905 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 3906 { 3907 struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress); 3908 struct tcf_result cl_res; 3909 3910 if (!miniq) 3911 return skb; 3912 3913 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */ 3914 qdisc_skb_cb(skb)->mru = 0; 3915 qdisc_skb_cb(skb)->post_ct = false; 3916 mini_qdisc_bstats_cpu_update(miniq, skb); 3917 3918 switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) { 3919 case TC_ACT_OK: 3920 case TC_ACT_RECLASSIFY: 3921 skb->tc_index = TC_H_MIN(cl_res.classid); 3922 break; 3923 case TC_ACT_SHOT: 3924 mini_qdisc_qstats_cpu_drop(miniq); 3925 *ret = NET_XMIT_DROP; 3926 kfree_skb(skb); 3927 return NULL; 3928 case TC_ACT_STOLEN: 3929 case TC_ACT_QUEUED: 3930 case TC_ACT_TRAP: 3931 *ret = NET_XMIT_SUCCESS; 3932 consume_skb(skb); 3933 return NULL; 3934 case TC_ACT_REDIRECT: 3935 /* No need to push/pop skb's mac_header here on egress! */ 3936 skb_do_redirect(skb); 3937 *ret = NET_XMIT_SUCCESS; 3938 return NULL; 3939 default: 3940 break; 3941 } 3942 3943 return skb; 3944 } 3945 #endif /* CONFIG_NET_EGRESS */ 3946 3947 #ifdef CONFIG_XPS 3948 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb, 3949 struct xps_dev_maps *dev_maps, unsigned int tci) 3950 { 3951 struct xps_map *map; 3952 int queue_index = -1; 3953 3954 if (dev->num_tc) { 3955 tci *= dev->num_tc; 3956 tci += netdev_get_prio_tc_map(dev, skb->priority); 3957 } 3958 3959 map = rcu_dereference(dev_maps->attr_map[tci]); 3960 if (map) { 3961 if (map->len == 1) 3962 queue_index = map->queues[0]; 3963 else 3964 queue_index = map->queues[reciprocal_scale( 3965 skb_get_hash(skb), map->len)]; 3966 if (unlikely(queue_index >= dev->real_num_tx_queues)) 3967 queue_index = -1; 3968 } 3969 return queue_index; 3970 } 3971 #endif 3972 3973 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev, 3974 struct sk_buff *skb) 3975 { 3976 #ifdef CONFIG_XPS 3977 struct xps_dev_maps *dev_maps; 3978 struct sock *sk = skb->sk; 3979 int queue_index = -1; 3980 3981 if (!static_key_false(&xps_needed)) 3982 return -1; 3983 3984 rcu_read_lock(); 3985 if (!static_key_false(&xps_rxqs_needed)) 3986 goto get_cpus_map; 3987 3988 dev_maps = rcu_dereference(sb_dev->xps_rxqs_map); 3989 if (dev_maps) { 3990 int tci = sk_rx_queue_get(sk); 3991 3992 if (tci >= 0 && tci < dev->num_rx_queues) 3993 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 3994 tci); 3995 } 3996 3997 get_cpus_map: 3998 if (queue_index < 0) { 3999 dev_maps = rcu_dereference(sb_dev->xps_cpus_map); 4000 if (dev_maps) { 4001 unsigned int tci = skb->sender_cpu - 1; 4002 4003 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4004 tci); 4005 } 4006 } 4007 rcu_read_unlock(); 4008 4009 return queue_index; 4010 #else 4011 return -1; 4012 #endif 4013 } 4014 4015 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb, 4016 struct net_device *sb_dev) 4017 { 4018 return 0; 4019 } 4020 EXPORT_SYMBOL(dev_pick_tx_zero); 4021 4022 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb, 4023 struct net_device *sb_dev) 4024 { 4025 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues; 4026 } 4027 EXPORT_SYMBOL(dev_pick_tx_cpu_id); 4028 4029 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb, 4030 struct net_device *sb_dev) 4031 { 4032 struct sock *sk = skb->sk; 4033 int queue_index = sk_tx_queue_get(sk); 4034 4035 sb_dev = sb_dev ? : dev; 4036 4037 if (queue_index < 0 || skb->ooo_okay || 4038 queue_index >= dev->real_num_tx_queues) { 4039 int new_index = get_xps_queue(dev, sb_dev, skb); 4040 4041 if (new_index < 0) 4042 new_index = skb_tx_hash(dev, sb_dev, skb); 4043 4044 if (queue_index != new_index && sk && 4045 sk_fullsock(sk) && 4046 rcu_access_pointer(sk->sk_dst_cache)) 4047 sk_tx_queue_set(sk, new_index); 4048 4049 queue_index = new_index; 4050 } 4051 4052 return queue_index; 4053 } 4054 EXPORT_SYMBOL(netdev_pick_tx); 4055 4056 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev, 4057 struct sk_buff *skb, 4058 struct net_device *sb_dev) 4059 { 4060 int queue_index = 0; 4061 4062 #ifdef CONFIG_XPS 4063 u32 sender_cpu = skb->sender_cpu - 1; 4064 4065 if (sender_cpu >= (u32)NR_CPUS) 4066 skb->sender_cpu = raw_smp_processor_id() + 1; 4067 #endif 4068 4069 if (dev->real_num_tx_queues != 1) { 4070 const struct net_device_ops *ops = dev->netdev_ops; 4071 4072 if (ops->ndo_select_queue) 4073 queue_index = ops->ndo_select_queue(dev, skb, sb_dev); 4074 else 4075 queue_index = netdev_pick_tx(dev, skb, sb_dev); 4076 4077 queue_index = netdev_cap_txqueue(dev, queue_index); 4078 } 4079 4080 skb_set_queue_mapping(skb, queue_index); 4081 return netdev_get_tx_queue(dev, queue_index); 4082 } 4083 4084 /** 4085 * __dev_queue_xmit - transmit a buffer 4086 * @skb: buffer to transmit 4087 * @sb_dev: suboordinate device used for L2 forwarding offload 4088 * 4089 * Queue a buffer for transmission to a network device. The caller must 4090 * have set the device and priority and built the buffer before calling 4091 * this function. The function can be called from an interrupt. 4092 * 4093 * A negative errno code is returned on a failure. A success does not 4094 * guarantee the frame will be transmitted as it may be dropped due 4095 * to congestion or traffic shaping. 4096 * 4097 * ----------------------------------------------------------------------------------- 4098 * I notice this method can also return errors from the queue disciplines, 4099 * including NET_XMIT_DROP, which is a positive value. So, errors can also 4100 * be positive. 4101 * 4102 * Regardless of the return value, the skb is consumed, so it is currently 4103 * difficult to retry a send to this method. (You can bump the ref count 4104 * before sending to hold a reference for retry if you are careful.) 4105 * 4106 * When calling this method, interrupts MUST be enabled. This is because 4107 * the BH enable code must have IRQs enabled so that it will not deadlock. 4108 * --BLG 4109 */ 4110 static int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev) 4111 { 4112 struct net_device *dev = skb->dev; 4113 struct netdev_queue *txq; 4114 struct Qdisc *q; 4115 int rc = -ENOMEM; 4116 bool again = false; 4117 4118 skb_reset_mac_header(skb); 4119 4120 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) 4121 __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED); 4122 4123 /* Disable soft irqs for various locks below. Also 4124 * stops preemption for RCU. 4125 */ 4126 rcu_read_lock_bh(); 4127 4128 skb_update_prio(skb); 4129 4130 qdisc_pkt_len_init(skb); 4131 #ifdef CONFIG_NET_CLS_ACT 4132 skb->tc_at_ingress = 0; 4133 # ifdef CONFIG_NET_EGRESS 4134 if (static_branch_unlikely(&egress_needed_key)) { 4135 skb = sch_handle_egress(skb, &rc, dev); 4136 if (!skb) 4137 goto out; 4138 } 4139 # endif 4140 #endif 4141 /* If device/qdisc don't need skb->dst, release it right now while 4142 * its hot in this cpu cache. 4143 */ 4144 if (dev->priv_flags & IFF_XMIT_DST_RELEASE) 4145 skb_dst_drop(skb); 4146 else 4147 skb_dst_force(skb); 4148 4149 txq = netdev_core_pick_tx(dev, skb, sb_dev); 4150 q = rcu_dereference_bh(txq->qdisc); 4151 4152 trace_net_dev_queue(skb); 4153 if (q->enqueue) { 4154 rc = __dev_xmit_skb(skb, q, dev, txq); 4155 goto out; 4156 } 4157 4158 /* The device has no queue. Common case for software devices: 4159 * loopback, all the sorts of tunnels... 4160 4161 * Really, it is unlikely that netif_tx_lock protection is necessary 4162 * here. (f.e. loopback and IP tunnels are clean ignoring statistics 4163 * counters.) 4164 * However, it is possible, that they rely on protection 4165 * made by us here. 4166 4167 * Check this and shot the lock. It is not prone from deadlocks. 4168 *Either shot noqueue qdisc, it is even simpler 8) 4169 */ 4170 if (dev->flags & IFF_UP) { 4171 int cpu = smp_processor_id(); /* ok because BHs are off */ 4172 4173 if (txq->xmit_lock_owner != cpu) { 4174 if (dev_xmit_recursion()) 4175 goto recursion_alert; 4176 4177 skb = validate_xmit_skb(skb, dev, &again); 4178 if (!skb) 4179 goto out; 4180 4181 PRANDOM_ADD_NOISE(skb, dev, txq, jiffies); 4182 HARD_TX_LOCK(dev, txq, cpu); 4183 4184 if (!netif_xmit_stopped(txq)) { 4185 dev_xmit_recursion_inc(); 4186 skb = dev_hard_start_xmit(skb, dev, txq, &rc); 4187 dev_xmit_recursion_dec(); 4188 if (dev_xmit_complete(rc)) { 4189 HARD_TX_UNLOCK(dev, txq); 4190 goto out; 4191 } 4192 } 4193 HARD_TX_UNLOCK(dev, txq); 4194 net_crit_ratelimited("Virtual device %s asks to queue packet!\n", 4195 dev->name); 4196 } else { 4197 /* Recursion is detected! It is possible, 4198 * unfortunately 4199 */ 4200 recursion_alert: 4201 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", 4202 dev->name); 4203 } 4204 } 4205 4206 rc = -ENETDOWN; 4207 rcu_read_unlock_bh(); 4208 4209 atomic_long_inc(&dev->tx_dropped); 4210 kfree_skb_list(skb); 4211 return rc; 4212 out: 4213 rcu_read_unlock_bh(); 4214 return rc; 4215 } 4216 4217 int dev_queue_xmit(struct sk_buff *skb) 4218 { 4219 return __dev_queue_xmit(skb, NULL); 4220 } 4221 EXPORT_SYMBOL(dev_queue_xmit); 4222 4223 int dev_queue_xmit_accel(struct sk_buff *skb, struct net_device *sb_dev) 4224 { 4225 return __dev_queue_xmit(skb, sb_dev); 4226 } 4227 EXPORT_SYMBOL(dev_queue_xmit_accel); 4228 4229 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id) 4230 { 4231 struct net_device *dev = skb->dev; 4232 struct sk_buff *orig_skb = skb; 4233 struct netdev_queue *txq; 4234 int ret = NETDEV_TX_BUSY; 4235 bool again = false; 4236 4237 if (unlikely(!netif_running(dev) || 4238 !netif_carrier_ok(dev))) 4239 goto drop; 4240 4241 skb = validate_xmit_skb_list(skb, dev, &again); 4242 if (skb != orig_skb) 4243 goto drop; 4244 4245 skb_set_queue_mapping(skb, queue_id); 4246 txq = skb_get_tx_queue(dev, skb); 4247 PRANDOM_ADD_NOISE(skb, dev, txq, jiffies); 4248 4249 local_bh_disable(); 4250 4251 dev_xmit_recursion_inc(); 4252 HARD_TX_LOCK(dev, txq, smp_processor_id()); 4253 if (!netif_xmit_frozen_or_drv_stopped(txq)) 4254 ret = netdev_start_xmit(skb, dev, txq, false); 4255 HARD_TX_UNLOCK(dev, txq); 4256 dev_xmit_recursion_dec(); 4257 4258 local_bh_enable(); 4259 return ret; 4260 drop: 4261 atomic_long_inc(&dev->tx_dropped); 4262 kfree_skb_list(skb); 4263 return NET_XMIT_DROP; 4264 } 4265 EXPORT_SYMBOL(__dev_direct_xmit); 4266 4267 /************************************************************************* 4268 * Receiver routines 4269 *************************************************************************/ 4270 4271 int netdev_max_backlog __read_mostly = 1000; 4272 EXPORT_SYMBOL(netdev_max_backlog); 4273 4274 int netdev_tstamp_prequeue __read_mostly = 1; 4275 int netdev_budget __read_mostly = 300; 4276 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */ 4277 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ; 4278 int weight_p __read_mostly = 64; /* old backlog weight */ 4279 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ 4280 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ 4281 int dev_rx_weight __read_mostly = 64; 4282 int dev_tx_weight __read_mostly = 64; 4283 /* Maximum number of GRO_NORMAL skbs to batch up for list-RX */ 4284 int gro_normal_batch __read_mostly = 8; 4285 4286 /* Called with irq disabled */ 4287 static inline void ____napi_schedule(struct softnet_data *sd, 4288 struct napi_struct *napi) 4289 { 4290 struct task_struct *thread; 4291 4292 if (test_bit(NAPI_STATE_THREADED, &napi->state)) { 4293 /* Paired with smp_mb__before_atomic() in 4294 * napi_enable()/dev_set_threaded(). 4295 * Use READ_ONCE() to guarantee a complete 4296 * read on napi->thread. Only call 4297 * wake_up_process() when it's not NULL. 4298 */ 4299 thread = READ_ONCE(napi->thread); 4300 if (thread) { 4301 wake_up_process(thread); 4302 return; 4303 } 4304 } 4305 4306 list_add_tail(&napi->poll_list, &sd->poll_list); 4307 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4308 } 4309 4310 #ifdef CONFIG_RPS 4311 4312 /* One global table that all flow-based protocols share. */ 4313 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; 4314 EXPORT_SYMBOL(rps_sock_flow_table); 4315 u32 rps_cpu_mask __read_mostly; 4316 EXPORT_SYMBOL(rps_cpu_mask); 4317 4318 struct static_key_false rps_needed __read_mostly; 4319 EXPORT_SYMBOL(rps_needed); 4320 struct static_key_false rfs_needed __read_mostly; 4321 EXPORT_SYMBOL(rfs_needed); 4322 4323 static struct rps_dev_flow * 4324 set_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4325 struct rps_dev_flow *rflow, u16 next_cpu) 4326 { 4327 if (next_cpu < nr_cpu_ids) { 4328 #ifdef CONFIG_RFS_ACCEL 4329 struct netdev_rx_queue *rxqueue; 4330 struct rps_dev_flow_table *flow_table; 4331 struct rps_dev_flow *old_rflow; 4332 u32 flow_id; 4333 u16 rxq_index; 4334 int rc; 4335 4336 /* Should we steer this flow to a different hardware queue? */ 4337 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || 4338 !(dev->features & NETIF_F_NTUPLE)) 4339 goto out; 4340 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); 4341 if (rxq_index == skb_get_rx_queue(skb)) 4342 goto out; 4343 4344 rxqueue = dev->_rx + rxq_index; 4345 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4346 if (!flow_table) 4347 goto out; 4348 flow_id = skb_get_hash(skb) & flow_table->mask; 4349 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, 4350 rxq_index, flow_id); 4351 if (rc < 0) 4352 goto out; 4353 old_rflow = rflow; 4354 rflow = &flow_table->flows[flow_id]; 4355 rflow->filter = rc; 4356 if (old_rflow->filter == rflow->filter) 4357 old_rflow->filter = RPS_NO_FILTER; 4358 out: 4359 #endif 4360 rflow->last_qtail = 4361 per_cpu(softnet_data, next_cpu).input_queue_head; 4362 } 4363 4364 rflow->cpu = next_cpu; 4365 return rflow; 4366 } 4367 4368 /* 4369 * get_rps_cpu is called from netif_receive_skb and returns the target 4370 * CPU from the RPS map of the receiving queue for a given skb. 4371 * rcu_read_lock must be held on entry. 4372 */ 4373 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4374 struct rps_dev_flow **rflowp) 4375 { 4376 const struct rps_sock_flow_table *sock_flow_table; 4377 struct netdev_rx_queue *rxqueue = dev->_rx; 4378 struct rps_dev_flow_table *flow_table; 4379 struct rps_map *map; 4380 int cpu = -1; 4381 u32 tcpu; 4382 u32 hash; 4383 4384 if (skb_rx_queue_recorded(skb)) { 4385 u16 index = skb_get_rx_queue(skb); 4386 4387 if (unlikely(index >= dev->real_num_rx_queues)) { 4388 WARN_ONCE(dev->real_num_rx_queues > 1, 4389 "%s received packet on queue %u, but number " 4390 "of RX queues is %u\n", 4391 dev->name, index, dev->real_num_rx_queues); 4392 goto done; 4393 } 4394 rxqueue += index; 4395 } 4396 4397 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ 4398 4399 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4400 map = rcu_dereference(rxqueue->rps_map); 4401 if (!flow_table && !map) 4402 goto done; 4403 4404 skb_reset_network_header(skb); 4405 hash = skb_get_hash(skb); 4406 if (!hash) 4407 goto done; 4408 4409 sock_flow_table = rcu_dereference(rps_sock_flow_table); 4410 if (flow_table && sock_flow_table) { 4411 struct rps_dev_flow *rflow; 4412 u32 next_cpu; 4413 u32 ident; 4414 4415 /* First check into global flow table if there is a match */ 4416 ident = sock_flow_table->ents[hash & sock_flow_table->mask]; 4417 if ((ident ^ hash) & ~rps_cpu_mask) 4418 goto try_rps; 4419 4420 next_cpu = ident & rps_cpu_mask; 4421 4422 /* OK, now we know there is a match, 4423 * we can look at the local (per receive queue) flow table 4424 */ 4425 rflow = &flow_table->flows[hash & flow_table->mask]; 4426 tcpu = rflow->cpu; 4427 4428 /* 4429 * If the desired CPU (where last recvmsg was done) is 4430 * different from current CPU (one in the rx-queue flow 4431 * table entry), switch if one of the following holds: 4432 * - Current CPU is unset (>= nr_cpu_ids). 4433 * - Current CPU is offline. 4434 * - The current CPU's queue tail has advanced beyond the 4435 * last packet that was enqueued using this table entry. 4436 * This guarantees that all previous packets for the flow 4437 * have been dequeued, thus preserving in order delivery. 4438 */ 4439 if (unlikely(tcpu != next_cpu) && 4440 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || 4441 ((int)(per_cpu(softnet_data, tcpu).input_queue_head - 4442 rflow->last_qtail)) >= 0)) { 4443 tcpu = next_cpu; 4444 rflow = set_rps_cpu(dev, skb, rflow, next_cpu); 4445 } 4446 4447 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { 4448 *rflowp = rflow; 4449 cpu = tcpu; 4450 goto done; 4451 } 4452 } 4453 4454 try_rps: 4455 4456 if (map) { 4457 tcpu = map->cpus[reciprocal_scale(hash, map->len)]; 4458 if (cpu_online(tcpu)) { 4459 cpu = tcpu; 4460 goto done; 4461 } 4462 } 4463 4464 done: 4465 return cpu; 4466 } 4467 4468 #ifdef CONFIG_RFS_ACCEL 4469 4470 /** 4471 * rps_may_expire_flow - check whether an RFS hardware filter may be removed 4472 * @dev: Device on which the filter was set 4473 * @rxq_index: RX queue index 4474 * @flow_id: Flow ID passed to ndo_rx_flow_steer() 4475 * @filter_id: Filter ID returned by ndo_rx_flow_steer() 4476 * 4477 * Drivers that implement ndo_rx_flow_steer() should periodically call 4478 * this function for each installed filter and remove the filters for 4479 * which it returns %true. 4480 */ 4481 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, 4482 u32 flow_id, u16 filter_id) 4483 { 4484 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; 4485 struct rps_dev_flow_table *flow_table; 4486 struct rps_dev_flow *rflow; 4487 bool expire = true; 4488 unsigned int cpu; 4489 4490 rcu_read_lock(); 4491 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4492 if (flow_table && flow_id <= flow_table->mask) { 4493 rflow = &flow_table->flows[flow_id]; 4494 cpu = READ_ONCE(rflow->cpu); 4495 if (rflow->filter == filter_id && cpu < nr_cpu_ids && 4496 ((int)(per_cpu(softnet_data, cpu).input_queue_head - 4497 rflow->last_qtail) < 4498 (int)(10 * flow_table->mask))) 4499 expire = false; 4500 } 4501 rcu_read_unlock(); 4502 return expire; 4503 } 4504 EXPORT_SYMBOL(rps_may_expire_flow); 4505 4506 #endif /* CONFIG_RFS_ACCEL */ 4507 4508 /* Called from hardirq (IPI) context */ 4509 static void rps_trigger_softirq(void *data) 4510 { 4511 struct softnet_data *sd = data; 4512 4513 ____napi_schedule(sd, &sd->backlog); 4514 sd->received_rps++; 4515 } 4516 4517 #endif /* CONFIG_RPS */ 4518 4519 /* 4520 * Check if this softnet_data structure is another cpu one 4521 * If yes, queue it to our IPI list and return 1 4522 * If no, return 0 4523 */ 4524 static int rps_ipi_queued(struct softnet_data *sd) 4525 { 4526 #ifdef CONFIG_RPS 4527 struct softnet_data *mysd = this_cpu_ptr(&softnet_data); 4528 4529 if (sd != mysd) { 4530 sd->rps_ipi_next = mysd->rps_ipi_list; 4531 mysd->rps_ipi_list = sd; 4532 4533 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4534 return 1; 4535 } 4536 #endif /* CONFIG_RPS */ 4537 return 0; 4538 } 4539 4540 #ifdef CONFIG_NET_FLOW_LIMIT 4541 int netdev_flow_limit_table_len __read_mostly = (1 << 12); 4542 #endif 4543 4544 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) 4545 { 4546 #ifdef CONFIG_NET_FLOW_LIMIT 4547 struct sd_flow_limit *fl; 4548 struct softnet_data *sd; 4549 unsigned int old_flow, new_flow; 4550 4551 if (qlen < (netdev_max_backlog >> 1)) 4552 return false; 4553 4554 sd = this_cpu_ptr(&softnet_data); 4555 4556 rcu_read_lock(); 4557 fl = rcu_dereference(sd->flow_limit); 4558 if (fl) { 4559 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); 4560 old_flow = fl->history[fl->history_head]; 4561 fl->history[fl->history_head] = new_flow; 4562 4563 fl->history_head++; 4564 fl->history_head &= FLOW_LIMIT_HISTORY - 1; 4565 4566 if (likely(fl->buckets[old_flow])) 4567 fl->buckets[old_flow]--; 4568 4569 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { 4570 fl->count++; 4571 rcu_read_unlock(); 4572 return true; 4573 } 4574 } 4575 rcu_read_unlock(); 4576 #endif 4577 return false; 4578 } 4579 4580 /* 4581 * enqueue_to_backlog is called to queue an skb to a per CPU backlog 4582 * queue (may be a remote CPU queue). 4583 */ 4584 static int enqueue_to_backlog(struct sk_buff *skb, int cpu, 4585 unsigned int *qtail) 4586 { 4587 struct softnet_data *sd; 4588 unsigned long flags; 4589 unsigned int qlen; 4590 4591 sd = &per_cpu(softnet_data, cpu); 4592 4593 local_irq_save(flags); 4594 4595 rps_lock(sd); 4596 if (!netif_running(skb->dev)) 4597 goto drop; 4598 qlen = skb_queue_len(&sd->input_pkt_queue); 4599 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) { 4600 if (qlen) { 4601 enqueue: 4602 __skb_queue_tail(&sd->input_pkt_queue, skb); 4603 input_queue_tail_incr_save(sd, qtail); 4604 rps_unlock(sd); 4605 local_irq_restore(flags); 4606 return NET_RX_SUCCESS; 4607 } 4608 4609 /* Schedule NAPI for backlog device 4610 * We can use non atomic operation since we own the queue lock 4611 */ 4612 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) { 4613 if (!rps_ipi_queued(sd)) 4614 ____napi_schedule(sd, &sd->backlog); 4615 } 4616 goto enqueue; 4617 } 4618 4619 drop: 4620 sd->dropped++; 4621 rps_unlock(sd); 4622 4623 local_irq_restore(flags); 4624 4625 atomic_long_inc(&skb->dev->rx_dropped); 4626 kfree_skb(skb); 4627 return NET_RX_DROP; 4628 } 4629 4630 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb) 4631 { 4632 struct net_device *dev = skb->dev; 4633 struct netdev_rx_queue *rxqueue; 4634 4635 rxqueue = dev->_rx; 4636 4637 if (skb_rx_queue_recorded(skb)) { 4638 u16 index = skb_get_rx_queue(skb); 4639 4640 if (unlikely(index >= dev->real_num_rx_queues)) { 4641 WARN_ONCE(dev->real_num_rx_queues > 1, 4642 "%s received packet on queue %u, but number " 4643 "of RX queues is %u\n", 4644 dev->name, index, dev->real_num_rx_queues); 4645 4646 return rxqueue; /* Return first rxqueue */ 4647 } 4648 rxqueue += index; 4649 } 4650 return rxqueue; 4651 } 4652 4653 static u32 netif_receive_generic_xdp(struct sk_buff *skb, 4654 struct xdp_buff *xdp, 4655 struct bpf_prog *xdp_prog) 4656 { 4657 void *orig_data, *orig_data_end, *hard_start; 4658 struct netdev_rx_queue *rxqueue; 4659 u32 metalen, act = XDP_DROP; 4660 u32 mac_len, frame_sz; 4661 __be16 orig_eth_type; 4662 struct ethhdr *eth; 4663 bool orig_bcast; 4664 int off; 4665 4666 /* Reinjected packets coming from act_mirred or similar should 4667 * not get XDP generic processing. 4668 */ 4669 if (skb_is_redirected(skb)) 4670 return XDP_PASS; 4671 4672 /* XDP packets must be linear and must have sufficient headroom 4673 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also 4674 * native XDP provides, thus we need to do it here as well. 4675 */ 4676 if (skb_cloned(skb) || skb_is_nonlinear(skb) || 4677 skb_headroom(skb) < XDP_PACKET_HEADROOM) { 4678 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb); 4679 int troom = skb->tail + skb->data_len - skb->end; 4680 4681 /* In case we have to go down the path and also linearize, 4682 * then lets do the pskb_expand_head() work just once here. 4683 */ 4684 if (pskb_expand_head(skb, 4685 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0, 4686 troom > 0 ? troom + 128 : 0, GFP_ATOMIC)) 4687 goto do_drop; 4688 if (skb_linearize(skb)) 4689 goto do_drop; 4690 } 4691 4692 /* The XDP program wants to see the packet starting at the MAC 4693 * header. 4694 */ 4695 mac_len = skb->data - skb_mac_header(skb); 4696 hard_start = skb->data - skb_headroom(skb); 4697 4698 /* SKB "head" area always have tailroom for skb_shared_info */ 4699 frame_sz = (void *)skb_end_pointer(skb) - hard_start; 4700 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 4701 4702 rxqueue = netif_get_rxqueue(skb); 4703 xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq); 4704 xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len, 4705 skb_headlen(skb) + mac_len, true); 4706 4707 orig_data_end = xdp->data_end; 4708 orig_data = xdp->data; 4709 eth = (struct ethhdr *)xdp->data; 4710 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest); 4711 orig_eth_type = eth->h_proto; 4712 4713 act = bpf_prog_run_xdp(xdp_prog, xdp); 4714 4715 /* check if bpf_xdp_adjust_head was used */ 4716 off = xdp->data - orig_data; 4717 if (off) { 4718 if (off > 0) 4719 __skb_pull(skb, off); 4720 else if (off < 0) 4721 __skb_push(skb, -off); 4722 4723 skb->mac_header += off; 4724 skb_reset_network_header(skb); 4725 } 4726 4727 /* check if bpf_xdp_adjust_tail was used */ 4728 off = xdp->data_end - orig_data_end; 4729 if (off != 0) { 4730 skb_set_tail_pointer(skb, xdp->data_end - xdp->data); 4731 skb->len += off; /* positive on grow, negative on shrink */ 4732 } 4733 4734 /* check if XDP changed eth hdr such SKB needs update */ 4735 eth = (struct ethhdr *)xdp->data; 4736 if ((orig_eth_type != eth->h_proto) || 4737 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) { 4738 __skb_push(skb, ETH_HLEN); 4739 skb->protocol = eth_type_trans(skb, skb->dev); 4740 } 4741 4742 switch (act) { 4743 case XDP_REDIRECT: 4744 case XDP_TX: 4745 __skb_push(skb, mac_len); 4746 break; 4747 case XDP_PASS: 4748 metalen = xdp->data - xdp->data_meta; 4749 if (metalen) 4750 skb_metadata_set(skb, metalen); 4751 break; 4752 default: 4753 bpf_warn_invalid_xdp_action(act); 4754 fallthrough; 4755 case XDP_ABORTED: 4756 trace_xdp_exception(skb->dev, xdp_prog, act); 4757 fallthrough; 4758 case XDP_DROP: 4759 do_drop: 4760 kfree_skb(skb); 4761 break; 4762 } 4763 4764 return act; 4765 } 4766 4767 /* When doing generic XDP we have to bypass the qdisc layer and the 4768 * network taps in order to match in-driver-XDP behavior. 4769 */ 4770 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog) 4771 { 4772 struct net_device *dev = skb->dev; 4773 struct netdev_queue *txq; 4774 bool free_skb = true; 4775 int cpu, rc; 4776 4777 txq = netdev_core_pick_tx(dev, skb, NULL); 4778 cpu = smp_processor_id(); 4779 HARD_TX_LOCK(dev, txq, cpu); 4780 if (!netif_xmit_stopped(txq)) { 4781 rc = netdev_start_xmit(skb, dev, txq, 0); 4782 if (dev_xmit_complete(rc)) 4783 free_skb = false; 4784 } 4785 HARD_TX_UNLOCK(dev, txq); 4786 if (free_skb) { 4787 trace_xdp_exception(dev, xdp_prog, XDP_TX); 4788 kfree_skb(skb); 4789 } 4790 } 4791 4792 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key); 4793 4794 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb) 4795 { 4796 if (xdp_prog) { 4797 struct xdp_buff xdp; 4798 u32 act; 4799 int err; 4800 4801 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog); 4802 if (act != XDP_PASS) { 4803 switch (act) { 4804 case XDP_REDIRECT: 4805 err = xdp_do_generic_redirect(skb->dev, skb, 4806 &xdp, xdp_prog); 4807 if (err) 4808 goto out_redir; 4809 break; 4810 case XDP_TX: 4811 generic_xdp_tx(skb, xdp_prog); 4812 break; 4813 } 4814 return XDP_DROP; 4815 } 4816 } 4817 return XDP_PASS; 4818 out_redir: 4819 kfree_skb(skb); 4820 return XDP_DROP; 4821 } 4822 EXPORT_SYMBOL_GPL(do_xdp_generic); 4823 4824 static int netif_rx_internal(struct sk_buff *skb) 4825 { 4826 int ret; 4827 4828 net_timestamp_check(netdev_tstamp_prequeue, skb); 4829 4830 trace_netif_rx(skb); 4831 4832 #ifdef CONFIG_RPS 4833 if (static_branch_unlikely(&rps_needed)) { 4834 struct rps_dev_flow voidflow, *rflow = &voidflow; 4835 int cpu; 4836 4837 preempt_disable(); 4838 rcu_read_lock(); 4839 4840 cpu = get_rps_cpu(skb->dev, skb, &rflow); 4841 if (cpu < 0) 4842 cpu = smp_processor_id(); 4843 4844 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 4845 4846 rcu_read_unlock(); 4847 preempt_enable(); 4848 } else 4849 #endif 4850 { 4851 unsigned int qtail; 4852 4853 ret = enqueue_to_backlog(skb, get_cpu(), &qtail); 4854 put_cpu(); 4855 } 4856 return ret; 4857 } 4858 4859 /** 4860 * netif_rx - post buffer to the network code 4861 * @skb: buffer to post 4862 * 4863 * This function receives a packet from a device driver and queues it for 4864 * the upper (protocol) levels to process. It always succeeds. The buffer 4865 * may be dropped during processing for congestion control or by the 4866 * protocol layers. 4867 * 4868 * return values: 4869 * NET_RX_SUCCESS (no congestion) 4870 * NET_RX_DROP (packet was dropped) 4871 * 4872 */ 4873 4874 int netif_rx(struct sk_buff *skb) 4875 { 4876 int ret; 4877 4878 trace_netif_rx_entry(skb); 4879 4880 ret = netif_rx_internal(skb); 4881 trace_netif_rx_exit(ret); 4882 4883 return ret; 4884 } 4885 EXPORT_SYMBOL(netif_rx); 4886 4887 int netif_rx_ni(struct sk_buff *skb) 4888 { 4889 int err; 4890 4891 trace_netif_rx_ni_entry(skb); 4892 4893 preempt_disable(); 4894 err = netif_rx_internal(skb); 4895 if (local_softirq_pending()) 4896 do_softirq(); 4897 preempt_enable(); 4898 trace_netif_rx_ni_exit(err); 4899 4900 return err; 4901 } 4902 EXPORT_SYMBOL(netif_rx_ni); 4903 4904 int netif_rx_any_context(struct sk_buff *skb) 4905 { 4906 /* 4907 * If invoked from contexts which do not invoke bottom half 4908 * processing either at return from interrupt or when softrqs are 4909 * reenabled, use netif_rx_ni() which invokes bottomhalf processing 4910 * directly. 4911 */ 4912 if (in_interrupt()) 4913 return netif_rx(skb); 4914 else 4915 return netif_rx_ni(skb); 4916 } 4917 EXPORT_SYMBOL(netif_rx_any_context); 4918 4919 static __latent_entropy void net_tx_action(struct softirq_action *h) 4920 { 4921 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 4922 4923 if (sd->completion_queue) { 4924 struct sk_buff *clist; 4925 4926 local_irq_disable(); 4927 clist = sd->completion_queue; 4928 sd->completion_queue = NULL; 4929 local_irq_enable(); 4930 4931 while (clist) { 4932 struct sk_buff *skb = clist; 4933 4934 clist = clist->next; 4935 4936 WARN_ON(refcount_read(&skb->users)); 4937 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED)) 4938 trace_consume_skb(skb); 4939 else 4940 trace_kfree_skb(skb, net_tx_action); 4941 4942 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) 4943 __kfree_skb(skb); 4944 else 4945 __kfree_skb_defer(skb); 4946 } 4947 4948 __kfree_skb_flush(); 4949 } 4950 4951 if (sd->output_queue) { 4952 struct Qdisc *head; 4953 4954 local_irq_disable(); 4955 head = sd->output_queue; 4956 sd->output_queue = NULL; 4957 sd->output_queue_tailp = &sd->output_queue; 4958 local_irq_enable(); 4959 4960 while (head) { 4961 struct Qdisc *q = head; 4962 spinlock_t *root_lock = NULL; 4963 4964 head = head->next_sched; 4965 4966 if (!(q->flags & TCQ_F_NOLOCK)) { 4967 root_lock = qdisc_lock(q); 4968 spin_lock(root_lock); 4969 } 4970 /* We need to make sure head->next_sched is read 4971 * before clearing __QDISC_STATE_SCHED 4972 */ 4973 smp_mb__before_atomic(); 4974 clear_bit(__QDISC_STATE_SCHED, &q->state); 4975 qdisc_run(q); 4976 if (root_lock) 4977 spin_unlock(root_lock); 4978 } 4979 } 4980 4981 xfrm_dev_backlog(sd); 4982 } 4983 4984 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) 4985 /* This hook is defined here for ATM LANE */ 4986 int (*br_fdb_test_addr_hook)(struct net_device *dev, 4987 unsigned char *addr) __read_mostly; 4988 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); 4989 #endif 4990 4991 static inline struct sk_buff * 4992 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 4993 struct net_device *orig_dev, bool *another) 4994 { 4995 #ifdef CONFIG_NET_CLS_ACT 4996 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress); 4997 struct tcf_result cl_res; 4998 4999 /* If there's at least one ingress present somewhere (so 5000 * we get here via enabled static key), remaining devices 5001 * that are not configured with an ingress qdisc will bail 5002 * out here. 5003 */ 5004 if (!miniq) 5005 return skb; 5006 5007 if (*pt_prev) { 5008 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5009 *pt_prev = NULL; 5010 } 5011 5012 qdisc_skb_cb(skb)->pkt_len = skb->len; 5013 qdisc_skb_cb(skb)->mru = 0; 5014 qdisc_skb_cb(skb)->post_ct = false; 5015 skb->tc_at_ingress = 1; 5016 mini_qdisc_bstats_cpu_update(miniq, skb); 5017 5018 switch (tcf_classify_ingress(skb, miniq->block, miniq->filter_list, 5019 &cl_res, false)) { 5020 case TC_ACT_OK: 5021 case TC_ACT_RECLASSIFY: 5022 skb->tc_index = TC_H_MIN(cl_res.classid); 5023 break; 5024 case TC_ACT_SHOT: 5025 mini_qdisc_qstats_cpu_drop(miniq); 5026 kfree_skb(skb); 5027 return NULL; 5028 case TC_ACT_STOLEN: 5029 case TC_ACT_QUEUED: 5030 case TC_ACT_TRAP: 5031 consume_skb(skb); 5032 return NULL; 5033 case TC_ACT_REDIRECT: 5034 /* skb_mac_header check was done by cls/act_bpf, so 5035 * we can safely push the L2 header back before 5036 * redirecting to another netdev 5037 */ 5038 __skb_push(skb, skb->mac_len); 5039 if (skb_do_redirect(skb) == -EAGAIN) { 5040 __skb_pull(skb, skb->mac_len); 5041 *another = true; 5042 break; 5043 } 5044 return NULL; 5045 case TC_ACT_CONSUMED: 5046 return NULL; 5047 default: 5048 break; 5049 } 5050 #endif /* CONFIG_NET_CLS_ACT */ 5051 return skb; 5052 } 5053 5054 /** 5055 * netdev_is_rx_handler_busy - check if receive handler is registered 5056 * @dev: device to check 5057 * 5058 * Check if a receive handler is already registered for a given device. 5059 * Return true if there one. 5060 * 5061 * The caller must hold the rtnl_mutex. 5062 */ 5063 bool netdev_is_rx_handler_busy(struct net_device *dev) 5064 { 5065 ASSERT_RTNL(); 5066 return dev && rtnl_dereference(dev->rx_handler); 5067 } 5068 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); 5069 5070 /** 5071 * netdev_rx_handler_register - register receive handler 5072 * @dev: device to register a handler for 5073 * @rx_handler: receive handler to register 5074 * @rx_handler_data: data pointer that is used by rx handler 5075 * 5076 * Register a receive handler for a device. This handler will then be 5077 * called from __netif_receive_skb. A negative errno code is returned 5078 * on a failure. 5079 * 5080 * The caller must hold the rtnl_mutex. 5081 * 5082 * For a general description of rx_handler, see enum rx_handler_result. 5083 */ 5084 int netdev_rx_handler_register(struct net_device *dev, 5085 rx_handler_func_t *rx_handler, 5086 void *rx_handler_data) 5087 { 5088 if (netdev_is_rx_handler_busy(dev)) 5089 return -EBUSY; 5090 5091 if (dev->priv_flags & IFF_NO_RX_HANDLER) 5092 return -EINVAL; 5093 5094 /* Note: rx_handler_data must be set before rx_handler */ 5095 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); 5096 rcu_assign_pointer(dev->rx_handler, rx_handler); 5097 5098 return 0; 5099 } 5100 EXPORT_SYMBOL_GPL(netdev_rx_handler_register); 5101 5102 /** 5103 * netdev_rx_handler_unregister - unregister receive handler 5104 * @dev: device to unregister a handler from 5105 * 5106 * Unregister a receive handler from a device. 5107 * 5108 * The caller must hold the rtnl_mutex. 5109 */ 5110 void netdev_rx_handler_unregister(struct net_device *dev) 5111 { 5112 5113 ASSERT_RTNL(); 5114 RCU_INIT_POINTER(dev->rx_handler, NULL); 5115 /* a reader seeing a non NULL rx_handler in a rcu_read_lock() 5116 * section has a guarantee to see a non NULL rx_handler_data 5117 * as well. 5118 */ 5119 synchronize_net(); 5120 RCU_INIT_POINTER(dev->rx_handler_data, NULL); 5121 } 5122 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); 5123 5124 /* 5125 * Limit the use of PFMEMALLOC reserves to those protocols that implement 5126 * the special handling of PFMEMALLOC skbs. 5127 */ 5128 static bool skb_pfmemalloc_protocol(struct sk_buff *skb) 5129 { 5130 switch (skb->protocol) { 5131 case htons(ETH_P_ARP): 5132 case htons(ETH_P_IP): 5133 case htons(ETH_P_IPV6): 5134 case htons(ETH_P_8021Q): 5135 case htons(ETH_P_8021AD): 5136 return true; 5137 default: 5138 return false; 5139 } 5140 } 5141 5142 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, 5143 int *ret, struct net_device *orig_dev) 5144 { 5145 if (nf_hook_ingress_active(skb)) { 5146 int ingress_retval; 5147 5148 if (*pt_prev) { 5149 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5150 *pt_prev = NULL; 5151 } 5152 5153 rcu_read_lock(); 5154 ingress_retval = nf_hook_ingress(skb); 5155 rcu_read_unlock(); 5156 return ingress_retval; 5157 } 5158 return 0; 5159 } 5160 5161 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc, 5162 struct packet_type **ppt_prev) 5163 { 5164 struct packet_type *ptype, *pt_prev; 5165 rx_handler_func_t *rx_handler; 5166 struct sk_buff *skb = *pskb; 5167 struct net_device *orig_dev; 5168 bool deliver_exact = false; 5169 int ret = NET_RX_DROP; 5170 __be16 type; 5171 5172 net_timestamp_check(!netdev_tstamp_prequeue, skb); 5173 5174 trace_netif_receive_skb(skb); 5175 5176 orig_dev = skb->dev; 5177 5178 skb_reset_network_header(skb); 5179 if (!skb_transport_header_was_set(skb)) 5180 skb_reset_transport_header(skb); 5181 skb_reset_mac_len(skb); 5182 5183 pt_prev = NULL; 5184 5185 another_round: 5186 skb->skb_iif = skb->dev->ifindex; 5187 5188 __this_cpu_inc(softnet_data.processed); 5189 5190 if (static_branch_unlikely(&generic_xdp_needed_key)) { 5191 int ret2; 5192 5193 preempt_disable(); 5194 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb); 5195 preempt_enable(); 5196 5197 if (ret2 != XDP_PASS) { 5198 ret = NET_RX_DROP; 5199 goto out; 5200 } 5201 skb_reset_mac_len(skb); 5202 } 5203 5204 if (eth_type_vlan(skb->protocol)) { 5205 skb = skb_vlan_untag(skb); 5206 if (unlikely(!skb)) 5207 goto out; 5208 } 5209 5210 if (skb_skip_tc_classify(skb)) 5211 goto skip_classify; 5212 5213 if (pfmemalloc) 5214 goto skip_taps; 5215 5216 list_for_each_entry_rcu(ptype, &ptype_all, list) { 5217 if (pt_prev) 5218 ret = deliver_skb(skb, pt_prev, orig_dev); 5219 pt_prev = ptype; 5220 } 5221 5222 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { 5223 if (pt_prev) 5224 ret = deliver_skb(skb, pt_prev, orig_dev); 5225 pt_prev = ptype; 5226 } 5227 5228 skip_taps: 5229 #ifdef CONFIG_NET_INGRESS 5230 if (static_branch_unlikely(&ingress_needed_key)) { 5231 bool another = false; 5232 5233 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev, 5234 &another); 5235 if (another) 5236 goto another_round; 5237 if (!skb) 5238 goto out; 5239 5240 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) 5241 goto out; 5242 } 5243 #endif 5244 skb_reset_redirect(skb); 5245 skip_classify: 5246 if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) 5247 goto drop; 5248 5249 if (skb_vlan_tag_present(skb)) { 5250 if (pt_prev) { 5251 ret = deliver_skb(skb, pt_prev, orig_dev); 5252 pt_prev = NULL; 5253 } 5254 if (vlan_do_receive(&skb)) 5255 goto another_round; 5256 else if (unlikely(!skb)) 5257 goto out; 5258 } 5259 5260 rx_handler = rcu_dereference(skb->dev->rx_handler); 5261 if (rx_handler) { 5262 if (pt_prev) { 5263 ret = deliver_skb(skb, pt_prev, orig_dev); 5264 pt_prev = NULL; 5265 } 5266 switch (rx_handler(&skb)) { 5267 case RX_HANDLER_CONSUMED: 5268 ret = NET_RX_SUCCESS; 5269 goto out; 5270 case RX_HANDLER_ANOTHER: 5271 goto another_round; 5272 case RX_HANDLER_EXACT: 5273 deliver_exact = true; 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 struct list_head *gro_list_prepare(struct napi_struct *napi, 5867 struct sk_buff *skb) 5868 { 5869 unsigned int maclen = skb->dev->hard_header_len; 5870 u32 hash = skb_get_hash_raw(skb); 5871 struct list_head *head; 5872 struct sk_buff *p; 5873 5874 head = &napi->gro_hash[hash & (GRO_HASH_BUCKETS - 1)].list; 5875 list_for_each_entry(p, head, list) { 5876 unsigned long diffs; 5877 5878 NAPI_GRO_CB(p)->flush = 0; 5879 5880 if (hash != skb_get_hash_raw(p)) { 5881 NAPI_GRO_CB(p)->same_flow = 0; 5882 continue; 5883 } 5884 5885 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev; 5886 diffs |= skb_vlan_tag_present(p) ^ skb_vlan_tag_present(skb); 5887 if (skb_vlan_tag_present(p)) 5888 diffs |= skb_vlan_tag_get(p) ^ skb_vlan_tag_get(skb); 5889 diffs |= skb_metadata_dst_cmp(p, skb); 5890 diffs |= skb_metadata_differs(p, skb); 5891 if (maclen == ETH_HLEN) 5892 diffs |= compare_ether_header(skb_mac_header(p), 5893 skb_mac_header(skb)); 5894 else if (!diffs) 5895 diffs = memcmp(skb_mac_header(p), 5896 skb_mac_header(skb), 5897 maclen); 5898 NAPI_GRO_CB(p)->same_flow = !diffs; 5899 } 5900 5901 return head; 5902 } 5903 5904 static void skb_gro_reset_offset(struct sk_buff *skb) 5905 { 5906 const struct skb_shared_info *pinfo = skb_shinfo(skb); 5907 const skb_frag_t *frag0 = &pinfo->frags[0]; 5908 5909 NAPI_GRO_CB(skb)->data_offset = 0; 5910 NAPI_GRO_CB(skb)->frag0 = NULL; 5911 NAPI_GRO_CB(skb)->frag0_len = 0; 5912 5913 if (!skb_headlen(skb) && pinfo->nr_frags && 5914 !PageHighMem(skb_frag_page(frag0))) { 5915 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0); 5916 NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int, 5917 skb_frag_size(frag0), 5918 skb->end - skb->tail); 5919 } 5920 } 5921 5922 static void gro_pull_from_frag0(struct sk_buff *skb, int grow) 5923 { 5924 struct skb_shared_info *pinfo = skb_shinfo(skb); 5925 5926 BUG_ON(skb->end - skb->tail < grow); 5927 5928 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); 5929 5930 skb->data_len -= grow; 5931 skb->tail += grow; 5932 5933 skb_frag_off_add(&pinfo->frags[0], grow); 5934 skb_frag_size_sub(&pinfo->frags[0], grow); 5935 5936 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) { 5937 skb_frag_unref(skb, 0); 5938 memmove(pinfo->frags, pinfo->frags + 1, 5939 --pinfo->nr_frags * sizeof(pinfo->frags[0])); 5940 } 5941 } 5942 5943 static void gro_flush_oldest(struct napi_struct *napi, struct list_head *head) 5944 { 5945 struct sk_buff *oldest; 5946 5947 oldest = list_last_entry(head, struct sk_buff, list); 5948 5949 /* We are called with head length >= MAX_GRO_SKBS, so this is 5950 * impossible. 5951 */ 5952 if (WARN_ON_ONCE(!oldest)) 5953 return; 5954 5955 /* Do not adjust napi->gro_hash[].count, caller is adding a new 5956 * SKB to the chain. 5957 */ 5958 skb_list_del_init(oldest); 5959 napi_gro_complete(napi, oldest); 5960 } 5961 5962 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) 5963 { 5964 u32 hash = skb_get_hash_raw(skb) & (GRO_HASH_BUCKETS - 1); 5965 struct list_head *head = &offload_base; 5966 struct packet_offload *ptype; 5967 __be16 type = skb->protocol; 5968 struct list_head *gro_head; 5969 struct sk_buff *pp = NULL; 5970 enum gro_result ret; 5971 int same_flow; 5972 int grow; 5973 5974 if (netif_elide_gro(skb->dev)) 5975 goto normal; 5976 5977 gro_head = gro_list_prepare(napi, skb); 5978 5979 rcu_read_lock(); 5980 list_for_each_entry_rcu(ptype, head, list) { 5981 if (ptype->type != type || !ptype->callbacks.gro_receive) 5982 continue; 5983 5984 skb_set_network_header(skb, skb_gro_offset(skb)); 5985 skb_reset_mac_len(skb); 5986 NAPI_GRO_CB(skb)->same_flow = 0; 5987 NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb); 5988 NAPI_GRO_CB(skb)->free = 0; 5989 NAPI_GRO_CB(skb)->encap_mark = 0; 5990 NAPI_GRO_CB(skb)->recursion_counter = 0; 5991 NAPI_GRO_CB(skb)->is_fou = 0; 5992 NAPI_GRO_CB(skb)->is_atomic = 1; 5993 NAPI_GRO_CB(skb)->gro_remcsum_start = 0; 5994 5995 /* Setup for GRO checksum validation */ 5996 switch (skb->ip_summed) { 5997 case CHECKSUM_COMPLETE: 5998 NAPI_GRO_CB(skb)->csum = skb->csum; 5999 NAPI_GRO_CB(skb)->csum_valid = 1; 6000 NAPI_GRO_CB(skb)->csum_cnt = 0; 6001 break; 6002 case CHECKSUM_UNNECESSARY: 6003 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1; 6004 NAPI_GRO_CB(skb)->csum_valid = 0; 6005 break; 6006 default: 6007 NAPI_GRO_CB(skb)->csum_cnt = 0; 6008 NAPI_GRO_CB(skb)->csum_valid = 0; 6009 } 6010 6011 pp = INDIRECT_CALL_INET(ptype->callbacks.gro_receive, 6012 ipv6_gro_receive, inet_gro_receive, 6013 gro_head, skb); 6014 break; 6015 } 6016 rcu_read_unlock(); 6017 6018 if (&ptype->list == head) 6019 goto normal; 6020 6021 if (PTR_ERR(pp) == -EINPROGRESS) { 6022 ret = GRO_CONSUMED; 6023 goto ok; 6024 } 6025 6026 same_flow = NAPI_GRO_CB(skb)->same_flow; 6027 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; 6028 6029 if (pp) { 6030 skb_list_del_init(pp); 6031 napi_gro_complete(napi, pp); 6032 napi->gro_hash[hash].count--; 6033 } 6034 6035 if (same_flow) 6036 goto ok; 6037 6038 if (NAPI_GRO_CB(skb)->flush) 6039 goto normal; 6040 6041 if (unlikely(napi->gro_hash[hash].count >= MAX_GRO_SKBS)) { 6042 gro_flush_oldest(napi, gro_head); 6043 } else { 6044 napi->gro_hash[hash].count++; 6045 } 6046 NAPI_GRO_CB(skb)->count = 1; 6047 NAPI_GRO_CB(skb)->age = jiffies; 6048 NAPI_GRO_CB(skb)->last = skb; 6049 skb_shinfo(skb)->gso_size = skb_gro_len(skb); 6050 list_add(&skb->list, gro_head); 6051 ret = GRO_HELD; 6052 6053 pull: 6054 grow = skb_gro_offset(skb) - skb_headlen(skb); 6055 if (grow > 0) 6056 gro_pull_from_frag0(skb, grow); 6057 ok: 6058 if (napi->gro_hash[hash].count) { 6059 if (!test_bit(hash, &napi->gro_bitmask)) 6060 __set_bit(hash, &napi->gro_bitmask); 6061 } else if (test_bit(hash, &napi->gro_bitmask)) { 6062 __clear_bit(hash, &napi->gro_bitmask); 6063 } 6064 6065 return ret; 6066 6067 normal: 6068 ret = GRO_NORMAL; 6069 goto pull; 6070 } 6071 6072 struct packet_offload *gro_find_receive_by_type(__be16 type) 6073 { 6074 struct list_head *offload_head = &offload_base; 6075 struct packet_offload *ptype; 6076 6077 list_for_each_entry_rcu(ptype, offload_head, list) { 6078 if (ptype->type != type || !ptype->callbacks.gro_receive) 6079 continue; 6080 return ptype; 6081 } 6082 return NULL; 6083 } 6084 EXPORT_SYMBOL(gro_find_receive_by_type); 6085 6086 struct packet_offload *gro_find_complete_by_type(__be16 type) 6087 { 6088 struct list_head *offload_head = &offload_base; 6089 struct packet_offload *ptype; 6090 6091 list_for_each_entry_rcu(ptype, offload_head, list) { 6092 if (ptype->type != type || !ptype->callbacks.gro_complete) 6093 continue; 6094 return ptype; 6095 } 6096 return NULL; 6097 } 6098 EXPORT_SYMBOL(gro_find_complete_by_type); 6099 6100 static void napi_skb_free_stolen_head(struct sk_buff *skb) 6101 { 6102 skb_dst_drop(skb); 6103 skb_ext_put(skb); 6104 kmem_cache_free(skbuff_head_cache, skb); 6105 } 6106 6107 static gro_result_t napi_skb_finish(struct napi_struct *napi, 6108 struct sk_buff *skb, 6109 gro_result_t ret) 6110 { 6111 switch (ret) { 6112 case GRO_NORMAL: 6113 gro_normal_one(napi, skb, 1); 6114 break; 6115 6116 case GRO_MERGED_FREE: 6117 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) 6118 napi_skb_free_stolen_head(skb); 6119 else 6120 __kfree_skb(skb); 6121 break; 6122 6123 case GRO_HELD: 6124 case GRO_MERGED: 6125 case GRO_CONSUMED: 6126 break; 6127 } 6128 6129 return ret; 6130 } 6131 6132 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) 6133 { 6134 gro_result_t ret; 6135 6136 skb_mark_napi_id(skb, napi); 6137 trace_napi_gro_receive_entry(skb); 6138 6139 skb_gro_reset_offset(skb); 6140 6141 ret = napi_skb_finish(napi, skb, dev_gro_receive(napi, skb)); 6142 trace_napi_gro_receive_exit(ret); 6143 6144 return ret; 6145 } 6146 EXPORT_SYMBOL(napi_gro_receive); 6147 6148 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) 6149 { 6150 if (unlikely(skb->pfmemalloc)) { 6151 consume_skb(skb); 6152 return; 6153 } 6154 __skb_pull(skb, skb_headlen(skb)); 6155 /* restore the reserve we had after netdev_alloc_skb_ip_align() */ 6156 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb)); 6157 __vlan_hwaccel_clear_tag(skb); 6158 skb->dev = napi->dev; 6159 skb->skb_iif = 0; 6160 6161 /* eth_type_trans() assumes pkt_type is PACKET_HOST */ 6162 skb->pkt_type = PACKET_HOST; 6163 6164 skb->encapsulation = 0; 6165 skb_shinfo(skb)->gso_type = 0; 6166 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); 6167 skb_ext_reset(skb); 6168 6169 napi->skb = skb; 6170 } 6171 6172 struct sk_buff *napi_get_frags(struct napi_struct *napi) 6173 { 6174 struct sk_buff *skb = napi->skb; 6175 6176 if (!skb) { 6177 skb = napi_alloc_skb(napi, GRO_MAX_HEAD); 6178 if (skb) { 6179 napi->skb = skb; 6180 skb_mark_napi_id(skb, napi); 6181 } 6182 } 6183 return skb; 6184 } 6185 EXPORT_SYMBOL(napi_get_frags); 6186 6187 static gro_result_t napi_frags_finish(struct napi_struct *napi, 6188 struct sk_buff *skb, 6189 gro_result_t ret) 6190 { 6191 switch (ret) { 6192 case GRO_NORMAL: 6193 case GRO_HELD: 6194 __skb_push(skb, ETH_HLEN); 6195 skb->protocol = eth_type_trans(skb, skb->dev); 6196 if (ret == GRO_NORMAL) 6197 gro_normal_one(napi, skb, 1); 6198 break; 6199 6200 case GRO_MERGED_FREE: 6201 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) 6202 napi_skb_free_stolen_head(skb); 6203 else 6204 napi_reuse_skb(napi, skb); 6205 break; 6206 6207 case GRO_MERGED: 6208 case GRO_CONSUMED: 6209 break; 6210 } 6211 6212 return ret; 6213 } 6214 6215 /* Upper GRO stack assumes network header starts at gro_offset=0 6216 * Drivers could call both napi_gro_frags() and napi_gro_receive() 6217 * We copy ethernet header into skb->data to have a common layout. 6218 */ 6219 static struct sk_buff *napi_frags_skb(struct napi_struct *napi) 6220 { 6221 struct sk_buff *skb = napi->skb; 6222 const struct ethhdr *eth; 6223 unsigned int hlen = sizeof(*eth); 6224 6225 napi->skb = NULL; 6226 6227 skb_reset_mac_header(skb); 6228 skb_gro_reset_offset(skb); 6229 6230 if (unlikely(skb_gro_header_hard(skb, hlen))) { 6231 eth = skb_gro_header_slow(skb, hlen, 0); 6232 if (unlikely(!eth)) { 6233 net_warn_ratelimited("%s: dropping impossible skb from %s\n", 6234 __func__, napi->dev->name); 6235 napi_reuse_skb(napi, skb); 6236 return NULL; 6237 } 6238 } else { 6239 eth = (const struct ethhdr *)skb->data; 6240 gro_pull_from_frag0(skb, hlen); 6241 NAPI_GRO_CB(skb)->frag0 += hlen; 6242 NAPI_GRO_CB(skb)->frag0_len -= hlen; 6243 } 6244 __skb_pull(skb, hlen); 6245 6246 /* 6247 * This works because the only protocols we care about don't require 6248 * special handling. 6249 * We'll fix it up properly in napi_frags_finish() 6250 */ 6251 skb->protocol = eth->h_proto; 6252 6253 return skb; 6254 } 6255 6256 gro_result_t napi_gro_frags(struct napi_struct *napi) 6257 { 6258 gro_result_t ret; 6259 struct sk_buff *skb = napi_frags_skb(napi); 6260 6261 trace_napi_gro_frags_entry(skb); 6262 6263 ret = napi_frags_finish(napi, skb, dev_gro_receive(napi, skb)); 6264 trace_napi_gro_frags_exit(ret); 6265 6266 return ret; 6267 } 6268 EXPORT_SYMBOL(napi_gro_frags); 6269 6270 /* Compute the checksum from gro_offset and return the folded value 6271 * after adding in any pseudo checksum. 6272 */ 6273 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb) 6274 { 6275 __wsum wsum; 6276 __sum16 sum; 6277 6278 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0); 6279 6280 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */ 6281 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum)); 6282 /* See comments in __skb_checksum_complete(). */ 6283 if (likely(!sum)) { 6284 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && 6285 !skb->csum_complete_sw) 6286 netdev_rx_csum_fault(skb->dev, skb); 6287 } 6288 6289 NAPI_GRO_CB(skb)->csum = wsum; 6290 NAPI_GRO_CB(skb)->csum_valid = 1; 6291 6292 return sum; 6293 } 6294 EXPORT_SYMBOL(__skb_gro_checksum_complete); 6295 6296 static void net_rps_send_ipi(struct softnet_data *remsd) 6297 { 6298 #ifdef CONFIG_RPS 6299 while (remsd) { 6300 struct softnet_data *next = remsd->rps_ipi_next; 6301 6302 if (cpu_online(remsd->cpu)) 6303 smp_call_function_single_async(remsd->cpu, &remsd->csd); 6304 remsd = next; 6305 } 6306 #endif 6307 } 6308 6309 /* 6310 * net_rps_action_and_irq_enable sends any pending IPI's for rps. 6311 * Note: called with local irq disabled, but exits with local irq enabled. 6312 */ 6313 static void net_rps_action_and_irq_enable(struct softnet_data *sd) 6314 { 6315 #ifdef CONFIG_RPS 6316 struct softnet_data *remsd = sd->rps_ipi_list; 6317 6318 if (remsd) { 6319 sd->rps_ipi_list = NULL; 6320 6321 local_irq_enable(); 6322 6323 /* Send pending IPI's to kick RPS processing on remote cpus. */ 6324 net_rps_send_ipi(remsd); 6325 } else 6326 #endif 6327 local_irq_enable(); 6328 } 6329 6330 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) 6331 { 6332 #ifdef CONFIG_RPS 6333 return sd->rps_ipi_list != NULL; 6334 #else 6335 return false; 6336 #endif 6337 } 6338 6339 static int process_backlog(struct napi_struct *napi, int quota) 6340 { 6341 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); 6342 bool again = true; 6343 int work = 0; 6344 6345 /* Check if we have pending ipi, its better to send them now, 6346 * not waiting net_rx_action() end. 6347 */ 6348 if (sd_has_rps_ipi_waiting(sd)) { 6349 local_irq_disable(); 6350 net_rps_action_and_irq_enable(sd); 6351 } 6352 6353 napi->weight = dev_rx_weight; 6354 while (again) { 6355 struct sk_buff *skb; 6356 6357 while ((skb = __skb_dequeue(&sd->process_queue))) { 6358 rcu_read_lock(); 6359 __netif_receive_skb(skb); 6360 rcu_read_unlock(); 6361 input_queue_head_incr(sd); 6362 if (++work >= quota) 6363 return work; 6364 6365 } 6366 6367 local_irq_disable(); 6368 rps_lock(sd); 6369 if (skb_queue_empty(&sd->input_pkt_queue)) { 6370 /* 6371 * Inline a custom version of __napi_complete(). 6372 * only current cpu owns and manipulates this napi, 6373 * and NAPI_STATE_SCHED is the only possible flag set 6374 * on backlog. 6375 * We can use a plain write instead of clear_bit(), 6376 * and we dont need an smp_mb() memory barrier. 6377 */ 6378 napi->state = 0; 6379 again = false; 6380 } else { 6381 skb_queue_splice_tail_init(&sd->input_pkt_queue, 6382 &sd->process_queue); 6383 } 6384 rps_unlock(sd); 6385 local_irq_enable(); 6386 } 6387 6388 return work; 6389 } 6390 6391 /** 6392 * __napi_schedule - schedule for receive 6393 * @n: entry to schedule 6394 * 6395 * The entry's receive function will be scheduled to run. 6396 * Consider using __napi_schedule_irqoff() if hard irqs are masked. 6397 */ 6398 void __napi_schedule(struct napi_struct *n) 6399 { 6400 unsigned long flags; 6401 6402 local_irq_save(flags); 6403 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6404 local_irq_restore(flags); 6405 } 6406 EXPORT_SYMBOL(__napi_schedule); 6407 6408 /** 6409 * napi_schedule_prep - check if napi can be scheduled 6410 * @n: napi context 6411 * 6412 * Test if NAPI routine is already running, and if not mark 6413 * it as running. This is used as a condition variable to 6414 * insure only one NAPI poll instance runs. We also make 6415 * sure there is no pending NAPI disable. 6416 */ 6417 bool napi_schedule_prep(struct napi_struct *n) 6418 { 6419 unsigned long val, new; 6420 6421 do { 6422 val = READ_ONCE(n->state); 6423 if (unlikely(val & NAPIF_STATE_DISABLE)) 6424 return false; 6425 new = val | NAPIF_STATE_SCHED; 6426 6427 /* Sets STATE_MISSED bit if STATE_SCHED was already set 6428 * This was suggested by Alexander Duyck, as compiler 6429 * emits better code than : 6430 * if (val & NAPIF_STATE_SCHED) 6431 * new |= NAPIF_STATE_MISSED; 6432 */ 6433 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * 6434 NAPIF_STATE_MISSED; 6435 } while (cmpxchg(&n->state, val, new) != val); 6436 6437 return !(val & NAPIF_STATE_SCHED); 6438 } 6439 EXPORT_SYMBOL(napi_schedule_prep); 6440 6441 /** 6442 * __napi_schedule_irqoff - schedule for receive 6443 * @n: entry to schedule 6444 * 6445 * Variant of __napi_schedule() assuming hard irqs are masked 6446 */ 6447 void __napi_schedule_irqoff(struct napi_struct *n) 6448 { 6449 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6450 } 6451 EXPORT_SYMBOL(__napi_schedule_irqoff); 6452 6453 bool napi_complete_done(struct napi_struct *n, int work_done) 6454 { 6455 unsigned long flags, val, new, timeout = 0; 6456 bool ret = true; 6457 6458 /* 6459 * 1) Don't let napi dequeue from the cpu poll list 6460 * just in case its running on a different cpu. 6461 * 2) If we are busy polling, do nothing here, we have 6462 * the guarantee we will be called later. 6463 */ 6464 if (unlikely(n->state & (NAPIF_STATE_NPSVC | 6465 NAPIF_STATE_IN_BUSY_POLL))) 6466 return false; 6467 6468 if (work_done) { 6469 if (n->gro_bitmask) 6470 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6471 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs); 6472 } 6473 if (n->defer_hard_irqs_count > 0) { 6474 n->defer_hard_irqs_count--; 6475 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6476 if (timeout) 6477 ret = false; 6478 } 6479 if (n->gro_bitmask) { 6480 /* When the NAPI instance uses a timeout and keeps postponing 6481 * it, we need to bound somehow the time packets are kept in 6482 * the GRO layer 6483 */ 6484 napi_gro_flush(n, !!timeout); 6485 } 6486 6487 gro_normal_list(n); 6488 6489 if (unlikely(!list_empty(&n->poll_list))) { 6490 /* If n->poll_list is not empty, we need to mask irqs */ 6491 local_irq_save(flags); 6492 list_del_init(&n->poll_list); 6493 local_irq_restore(flags); 6494 } 6495 6496 do { 6497 val = READ_ONCE(n->state); 6498 6499 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); 6500 6501 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED | 6502 NAPIF_STATE_PREFER_BUSY_POLL); 6503 6504 /* If STATE_MISSED was set, leave STATE_SCHED set, 6505 * because we will call napi->poll() one more time. 6506 * This C code was suggested by Alexander Duyck to help gcc. 6507 */ 6508 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * 6509 NAPIF_STATE_SCHED; 6510 } while (cmpxchg(&n->state, val, new) != val); 6511 6512 if (unlikely(val & NAPIF_STATE_MISSED)) { 6513 __napi_schedule(n); 6514 return false; 6515 } 6516 6517 if (timeout) 6518 hrtimer_start(&n->timer, ns_to_ktime(timeout), 6519 HRTIMER_MODE_REL_PINNED); 6520 return ret; 6521 } 6522 EXPORT_SYMBOL(napi_complete_done); 6523 6524 /* must be called under rcu_read_lock(), as we dont take a reference */ 6525 static struct napi_struct *napi_by_id(unsigned int napi_id) 6526 { 6527 unsigned int hash = napi_id % HASH_SIZE(napi_hash); 6528 struct napi_struct *napi; 6529 6530 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) 6531 if (napi->napi_id == napi_id) 6532 return napi; 6533 6534 return NULL; 6535 } 6536 6537 #if defined(CONFIG_NET_RX_BUSY_POLL) 6538 6539 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule) 6540 { 6541 if (!skip_schedule) { 6542 gro_normal_list(napi); 6543 __napi_schedule(napi); 6544 return; 6545 } 6546 6547 if (napi->gro_bitmask) { 6548 /* flush too old packets 6549 * If HZ < 1000, flush all packets. 6550 */ 6551 napi_gro_flush(napi, HZ >= 1000); 6552 } 6553 6554 gro_normal_list(napi); 6555 clear_bit(NAPI_STATE_SCHED, &napi->state); 6556 } 6557 6558 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll, 6559 u16 budget) 6560 { 6561 bool skip_schedule = false; 6562 unsigned long timeout; 6563 int rc; 6564 6565 /* Busy polling means there is a high chance device driver hard irq 6566 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was 6567 * set in napi_schedule_prep(). 6568 * Since we are about to call napi->poll() once more, we can safely 6569 * clear NAPI_STATE_MISSED. 6570 * 6571 * Note: x86 could use a single "lock and ..." instruction 6572 * to perform these two clear_bit() 6573 */ 6574 clear_bit(NAPI_STATE_MISSED, &napi->state); 6575 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); 6576 6577 local_bh_disable(); 6578 6579 if (prefer_busy_poll) { 6580 napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs); 6581 timeout = READ_ONCE(napi->dev->gro_flush_timeout); 6582 if (napi->defer_hard_irqs_count && timeout) { 6583 hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED); 6584 skip_schedule = true; 6585 } 6586 } 6587 6588 /* All we really want here is to re-enable device interrupts. 6589 * Ideally, a new ndo_busy_poll_stop() could avoid another round. 6590 */ 6591 rc = napi->poll(napi, budget); 6592 /* We can't gro_normal_list() here, because napi->poll() might have 6593 * rearmed the napi (napi_complete_done()) in which case it could 6594 * already be running on another CPU. 6595 */ 6596 trace_napi_poll(napi, rc, budget); 6597 netpoll_poll_unlock(have_poll_lock); 6598 if (rc == budget) 6599 __busy_poll_stop(napi, skip_schedule); 6600 local_bh_enable(); 6601 } 6602 6603 void napi_busy_loop(unsigned int napi_id, 6604 bool (*loop_end)(void *, unsigned long), 6605 void *loop_end_arg, bool prefer_busy_poll, u16 budget) 6606 { 6607 unsigned long start_time = loop_end ? busy_loop_current_time() : 0; 6608 int (*napi_poll)(struct napi_struct *napi, int budget); 6609 void *have_poll_lock = NULL; 6610 struct napi_struct *napi; 6611 6612 restart: 6613 napi_poll = NULL; 6614 6615 rcu_read_lock(); 6616 6617 napi = napi_by_id(napi_id); 6618 if (!napi) 6619 goto out; 6620 6621 preempt_disable(); 6622 for (;;) { 6623 int work = 0; 6624 6625 local_bh_disable(); 6626 if (!napi_poll) { 6627 unsigned long val = READ_ONCE(napi->state); 6628 6629 /* If multiple threads are competing for this napi, 6630 * we avoid dirtying napi->state as much as we can. 6631 */ 6632 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | 6633 NAPIF_STATE_IN_BUSY_POLL)) { 6634 if (prefer_busy_poll) 6635 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6636 goto count; 6637 } 6638 if (cmpxchg(&napi->state, val, 6639 val | NAPIF_STATE_IN_BUSY_POLL | 6640 NAPIF_STATE_SCHED) != val) { 6641 if (prefer_busy_poll) 6642 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6643 goto count; 6644 } 6645 have_poll_lock = netpoll_poll_lock(napi); 6646 napi_poll = napi->poll; 6647 } 6648 work = napi_poll(napi, budget); 6649 trace_napi_poll(napi, work, budget); 6650 gro_normal_list(napi); 6651 count: 6652 if (work > 0) 6653 __NET_ADD_STATS(dev_net(napi->dev), 6654 LINUX_MIB_BUSYPOLLRXPACKETS, work); 6655 local_bh_enable(); 6656 6657 if (!loop_end || loop_end(loop_end_arg, start_time)) 6658 break; 6659 6660 if (unlikely(need_resched())) { 6661 if (napi_poll) 6662 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6663 preempt_enable(); 6664 rcu_read_unlock(); 6665 cond_resched(); 6666 if (loop_end(loop_end_arg, start_time)) 6667 return; 6668 goto restart; 6669 } 6670 cpu_relax(); 6671 } 6672 if (napi_poll) 6673 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6674 preempt_enable(); 6675 out: 6676 rcu_read_unlock(); 6677 } 6678 EXPORT_SYMBOL(napi_busy_loop); 6679 6680 #endif /* CONFIG_NET_RX_BUSY_POLL */ 6681 6682 static void napi_hash_add(struct napi_struct *napi) 6683 { 6684 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state)) 6685 return; 6686 6687 spin_lock(&napi_hash_lock); 6688 6689 /* 0..NR_CPUS range is reserved for sender_cpu use */ 6690 do { 6691 if (unlikely(++napi_gen_id < MIN_NAPI_ID)) 6692 napi_gen_id = MIN_NAPI_ID; 6693 } while (napi_by_id(napi_gen_id)); 6694 napi->napi_id = napi_gen_id; 6695 6696 hlist_add_head_rcu(&napi->napi_hash_node, 6697 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); 6698 6699 spin_unlock(&napi_hash_lock); 6700 } 6701 6702 /* Warning : caller is responsible to make sure rcu grace period 6703 * is respected before freeing memory containing @napi 6704 */ 6705 static void napi_hash_del(struct napi_struct *napi) 6706 { 6707 spin_lock(&napi_hash_lock); 6708 6709 hlist_del_init_rcu(&napi->napi_hash_node); 6710 6711 spin_unlock(&napi_hash_lock); 6712 } 6713 6714 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) 6715 { 6716 struct napi_struct *napi; 6717 6718 napi = container_of(timer, struct napi_struct, timer); 6719 6720 /* Note : we use a relaxed variant of napi_schedule_prep() not setting 6721 * NAPI_STATE_MISSED, since we do not react to a device IRQ. 6722 */ 6723 if (!napi_disable_pending(napi) && 6724 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) { 6725 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6726 __napi_schedule_irqoff(napi); 6727 } 6728 6729 return HRTIMER_NORESTART; 6730 } 6731 6732 static void init_gro_hash(struct napi_struct *napi) 6733 { 6734 int i; 6735 6736 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6737 INIT_LIST_HEAD(&napi->gro_hash[i].list); 6738 napi->gro_hash[i].count = 0; 6739 } 6740 napi->gro_bitmask = 0; 6741 } 6742 6743 int dev_set_threaded(struct net_device *dev, bool threaded) 6744 { 6745 struct napi_struct *napi; 6746 int err = 0; 6747 6748 if (dev->threaded == threaded) 6749 return 0; 6750 6751 if (threaded) { 6752 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6753 if (!napi->thread) { 6754 err = napi_kthread_create(napi); 6755 if (err) { 6756 threaded = false; 6757 break; 6758 } 6759 } 6760 } 6761 } 6762 6763 dev->threaded = threaded; 6764 6765 /* Make sure kthread is created before THREADED bit 6766 * is set. 6767 */ 6768 smp_mb__before_atomic(); 6769 6770 /* Setting/unsetting threaded mode on a napi might not immediately 6771 * take effect, if the current napi instance is actively being 6772 * polled. In this case, the switch between threaded mode and 6773 * softirq mode will happen in the next round of napi_schedule(). 6774 * This should not cause hiccups/stalls to the live traffic. 6775 */ 6776 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6777 if (threaded) 6778 set_bit(NAPI_STATE_THREADED, &napi->state); 6779 else 6780 clear_bit(NAPI_STATE_THREADED, &napi->state); 6781 } 6782 6783 return err; 6784 } 6785 6786 void netif_napi_add(struct net_device *dev, struct napi_struct *napi, 6787 int (*poll)(struct napi_struct *, int), int weight) 6788 { 6789 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state))) 6790 return; 6791 6792 INIT_LIST_HEAD(&napi->poll_list); 6793 INIT_HLIST_NODE(&napi->napi_hash_node); 6794 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); 6795 napi->timer.function = napi_watchdog; 6796 init_gro_hash(napi); 6797 napi->skb = NULL; 6798 INIT_LIST_HEAD(&napi->rx_list); 6799 napi->rx_count = 0; 6800 napi->poll = poll; 6801 if (weight > NAPI_POLL_WEIGHT) 6802 netdev_err_once(dev, "%s() called with weight %d\n", __func__, 6803 weight); 6804 napi->weight = weight; 6805 napi->dev = dev; 6806 #ifdef CONFIG_NETPOLL 6807 napi->poll_owner = -1; 6808 #endif 6809 set_bit(NAPI_STATE_SCHED, &napi->state); 6810 set_bit(NAPI_STATE_NPSVC, &napi->state); 6811 list_add_rcu(&napi->dev_list, &dev->napi_list); 6812 napi_hash_add(napi); 6813 /* Create kthread for this napi if dev->threaded is set. 6814 * Clear dev->threaded if kthread creation failed so that 6815 * threaded mode will not be enabled in napi_enable(). 6816 */ 6817 if (dev->threaded && napi_kthread_create(napi)) 6818 dev->threaded = 0; 6819 } 6820 EXPORT_SYMBOL(netif_napi_add); 6821 6822 void napi_disable(struct napi_struct *n) 6823 { 6824 might_sleep(); 6825 set_bit(NAPI_STATE_DISABLE, &n->state); 6826 6827 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state)) 6828 msleep(1); 6829 while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state)) 6830 msleep(1); 6831 6832 hrtimer_cancel(&n->timer); 6833 6834 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &n->state); 6835 clear_bit(NAPI_STATE_DISABLE, &n->state); 6836 clear_bit(NAPI_STATE_THREADED, &n->state); 6837 } 6838 EXPORT_SYMBOL(napi_disable); 6839 6840 /** 6841 * napi_enable - enable NAPI scheduling 6842 * @n: NAPI context 6843 * 6844 * Resume NAPI from being scheduled on this context. 6845 * Must be paired with napi_disable. 6846 */ 6847 void napi_enable(struct napi_struct *n) 6848 { 6849 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state)); 6850 smp_mb__before_atomic(); 6851 clear_bit(NAPI_STATE_SCHED, &n->state); 6852 clear_bit(NAPI_STATE_NPSVC, &n->state); 6853 if (n->dev->threaded && n->thread) 6854 set_bit(NAPI_STATE_THREADED, &n->state); 6855 } 6856 EXPORT_SYMBOL(napi_enable); 6857 6858 static void flush_gro_hash(struct napi_struct *napi) 6859 { 6860 int i; 6861 6862 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6863 struct sk_buff *skb, *n; 6864 6865 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list) 6866 kfree_skb(skb); 6867 napi->gro_hash[i].count = 0; 6868 } 6869 } 6870 6871 /* Must be called in process context */ 6872 void __netif_napi_del(struct napi_struct *napi) 6873 { 6874 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state)) 6875 return; 6876 6877 napi_hash_del(napi); 6878 list_del_rcu(&napi->dev_list); 6879 napi_free_frags(napi); 6880 6881 flush_gro_hash(napi); 6882 napi->gro_bitmask = 0; 6883 6884 if (napi->thread) { 6885 kthread_stop(napi->thread); 6886 napi->thread = NULL; 6887 } 6888 } 6889 EXPORT_SYMBOL(__netif_napi_del); 6890 6891 static int __napi_poll(struct napi_struct *n, bool *repoll) 6892 { 6893 int work, weight; 6894 6895 weight = n->weight; 6896 6897 /* This NAPI_STATE_SCHED test is for avoiding a race 6898 * with netpoll's poll_napi(). Only the entity which 6899 * obtains the lock and sees NAPI_STATE_SCHED set will 6900 * actually make the ->poll() call. Therefore we avoid 6901 * accidentally calling ->poll() when NAPI is not scheduled. 6902 */ 6903 work = 0; 6904 if (test_bit(NAPI_STATE_SCHED, &n->state)) { 6905 work = n->poll(n, weight); 6906 trace_napi_poll(n, work, weight); 6907 } 6908 6909 if (unlikely(work > weight)) 6910 pr_err_once("NAPI poll function %pS returned %d, exceeding its budget of %d.\n", 6911 n->poll, work, weight); 6912 6913 if (likely(work < weight)) 6914 return work; 6915 6916 /* Drivers must not modify the NAPI state if they 6917 * consume the entire weight. In such cases this code 6918 * still "owns" the NAPI instance and therefore can 6919 * move the instance around on the list at-will. 6920 */ 6921 if (unlikely(napi_disable_pending(n))) { 6922 napi_complete(n); 6923 return work; 6924 } 6925 6926 /* The NAPI context has more processing work, but busy-polling 6927 * is preferred. Exit early. 6928 */ 6929 if (napi_prefer_busy_poll(n)) { 6930 if (napi_complete_done(n, work)) { 6931 /* If timeout is not set, we need to make sure 6932 * that the NAPI is re-scheduled. 6933 */ 6934 napi_schedule(n); 6935 } 6936 return work; 6937 } 6938 6939 if (n->gro_bitmask) { 6940 /* flush too old packets 6941 * If HZ < 1000, flush all packets. 6942 */ 6943 napi_gro_flush(n, HZ >= 1000); 6944 } 6945 6946 gro_normal_list(n); 6947 6948 /* Some drivers may have called napi_schedule 6949 * prior to exhausting their budget. 6950 */ 6951 if (unlikely(!list_empty(&n->poll_list))) { 6952 pr_warn_once("%s: Budget exhausted after napi rescheduled\n", 6953 n->dev ? n->dev->name : "backlog"); 6954 return work; 6955 } 6956 6957 *repoll = true; 6958 6959 return work; 6960 } 6961 6962 static int napi_poll(struct napi_struct *n, struct list_head *repoll) 6963 { 6964 bool do_repoll = false; 6965 void *have; 6966 int work; 6967 6968 list_del_init(&n->poll_list); 6969 6970 have = netpoll_poll_lock(n); 6971 6972 work = __napi_poll(n, &do_repoll); 6973 6974 if (do_repoll) 6975 list_add_tail(&n->poll_list, repoll); 6976 6977 netpoll_poll_unlock(have); 6978 6979 return work; 6980 } 6981 6982 static int napi_thread_wait(struct napi_struct *napi) 6983 { 6984 set_current_state(TASK_INTERRUPTIBLE); 6985 6986 while (!kthread_should_stop() && !napi_disable_pending(napi)) { 6987 if (test_bit(NAPI_STATE_SCHED, &napi->state)) { 6988 WARN_ON(!list_empty(&napi->poll_list)); 6989 __set_current_state(TASK_RUNNING); 6990 return 0; 6991 } 6992 6993 schedule(); 6994 set_current_state(TASK_INTERRUPTIBLE); 6995 } 6996 __set_current_state(TASK_RUNNING); 6997 return -1; 6998 } 6999 7000 static int napi_threaded_poll(void *data) 7001 { 7002 struct napi_struct *napi = data; 7003 void *have; 7004 7005 while (!napi_thread_wait(napi)) { 7006 for (;;) { 7007 bool repoll = false; 7008 7009 local_bh_disable(); 7010 7011 have = netpoll_poll_lock(napi); 7012 __napi_poll(napi, &repoll); 7013 netpoll_poll_unlock(have); 7014 7015 __kfree_skb_flush(); 7016 local_bh_enable(); 7017 7018 if (!repoll) 7019 break; 7020 7021 cond_resched(); 7022 } 7023 } 7024 return 0; 7025 } 7026 7027 static __latent_entropy void net_rx_action(struct softirq_action *h) 7028 { 7029 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 7030 unsigned long time_limit = jiffies + 7031 usecs_to_jiffies(netdev_budget_usecs); 7032 int budget = netdev_budget; 7033 LIST_HEAD(list); 7034 LIST_HEAD(repoll); 7035 7036 local_irq_disable(); 7037 list_splice_init(&sd->poll_list, &list); 7038 local_irq_enable(); 7039 7040 for (;;) { 7041 struct napi_struct *n; 7042 7043 if (list_empty(&list)) { 7044 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll)) 7045 goto out; 7046 break; 7047 } 7048 7049 n = list_first_entry(&list, struct napi_struct, poll_list); 7050 budget -= napi_poll(n, &repoll); 7051 7052 /* If softirq window is exhausted then punt. 7053 * Allow this to run for 2 jiffies since which will allow 7054 * an average latency of 1.5/HZ. 7055 */ 7056 if (unlikely(budget <= 0 || 7057 time_after_eq(jiffies, time_limit))) { 7058 sd->time_squeeze++; 7059 break; 7060 } 7061 } 7062 7063 local_irq_disable(); 7064 7065 list_splice_tail_init(&sd->poll_list, &list); 7066 list_splice_tail(&repoll, &list); 7067 list_splice(&list, &sd->poll_list); 7068 if (!list_empty(&sd->poll_list)) 7069 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 7070 7071 net_rps_action_and_irq_enable(sd); 7072 out: 7073 __kfree_skb_flush(); 7074 } 7075 7076 struct netdev_adjacent { 7077 struct net_device *dev; 7078 7079 /* upper master flag, there can only be one master device per list */ 7080 bool master; 7081 7082 /* lookup ignore flag */ 7083 bool ignore; 7084 7085 /* counter for the number of times this device was added to us */ 7086 u16 ref_nr; 7087 7088 /* private field for the users */ 7089 void *private; 7090 7091 struct list_head list; 7092 struct rcu_head rcu; 7093 }; 7094 7095 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, 7096 struct list_head *adj_list) 7097 { 7098 struct netdev_adjacent *adj; 7099 7100 list_for_each_entry(adj, adj_list, list) { 7101 if (adj->dev == adj_dev) 7102 return adj; 7103 } 7104 return NULL; 7105 } 7106 7107 static int ____netdev_has_upper_dev(struct net_device *upper_dev, 7108 struct netdev_nested_priv *priv) 7109 { 7110 struct net_device *dev = (struct net_device *)priv->data; 7111 7112 return upper_dev == dev; 7113 } 7114 7115 /** 7116 * netdev_has_upper_dev - Check if device is linked to an upper device 7117 * @dev: device 7118 * @upper_dev: upper device to check 7119 * 7120 * Find out if a device is linked to specified upper device and return true 7121 * in case it is. Note that this checks only immediate upper device, 7122 * not through a complete stack of devices. The caller must hold the RTNL lock. 7123 */ 7124 bool netdev_has_upper_dev(struct net_device *dev, 7125 struct net_device *upper_dev) 7126 { 7127 struct netdev_nested_priv priv = { 7128 .data = (void *)upper_dev, 7129 }; 7130 7131 ASSERT_RTNL(); 7132 7133 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 7134 &priv); 7135 } 7136 EXPORT_SYMBOL(netdev_has_upper_dev); 7137 7138 /** 7139 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device 7140 * @dev: device 7141 * @upper_dev: upper device to check 7142 * 7143 * Find out if a device is linked to specified upper device and return true 7144 * in case it is. Note that this checks the entire upper device chain. 7145 * The caller must hold rcu lock. 7146 */ 7147 7148 bool netdev_has_upper_dev_all_rcu(struct net_device *dev, 7149 struct net_device *upper_dev) 7150 { 7151 struct netdev_nested_priv priv = { 7152 .data = (void *)upper_dev, 7153 }; 7154 7155 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 7156 &priv); 7157 } 7158 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); 7159 7160 /** 7161 * netdev_has_any_upper_dev - Check if device is linked to some device 7162 * @dev: device 7163 * 7164 * Find out if a device is linked to an upper device and return true in case 7165 * it is. The caller must hold the RTNL lock. 7166 */ 7167 bool netdev_has_any_upper_dev(struct net_device *dev) 7168 { 7169 ASSERT_RTNL(); 7170 7171 return !list_empty(&dev->adj_list.upper); 7172 } 7173 EXPORT_SYMBOL(netdev_has_any_upper_dev); 7174 7175 /** 7176 * netdev_master_upper_dev_get - Get master upper device 7177 * @dev: device 7178 * 7179 * Find a master upper device and return pointer to it or NULL in case 7180 * it's not there. The caller must hold the RTNL lock. 7181 */ 7182 struct net_device *netdev_master_upper_dev_get(struct net_device *dev) 7183 { 7184 struct netdev_adjacent *upper; 7185 7186 ASSERT_RTNL(); 7187 7188 if (list_empty(&dev->adj_list.upper)) 7189 return NULL; 7190 7191 upper = list_first_entry(&dev->adj_list.upper, 7192 struct netdev_adjacent, list); 7193 if (likely(upper->master)) 7194 return upper->dev; 7195 return NULL; 7196 } 7197 EXPORT_SYMBOL(netdev_master_upper_dev_get); 7198 7199 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev) 7200 { 7201 struct netdev_adjacent *upper; 7202 7203 ASSERT_RTNL(); 7204 7205 if (list_empty(&dev->adj_list.upper)) 7206 return NULL; 7207 7208 upper = list_first_entry(&dev->adj_list.upper, 7209 struct netdev_adjacent, list); 7210 if (likely(upper->master) && !upper->ignore) 7211 return upper->dev; 7212 return NULL; 7213 } 7214 7215 /** 7216 * netdev_has_any_lower_dev - Check if device is linked to some device 7217 * @dev: device 7218 * 7219 * Find out if a device is linked to a lower device and return true in case 7220 * it is. The caller must hold the RTNL lock. 7221 */ 7222 static bool netdev_has_any_lower_dev(struct net_device *dev) 7223 { 7224 ASSERT_RTNL(); 7225 7226 return !list_empty(&dev->adj_list.lower); 7227 } 7228 7229 void *netdev_adjacent_get_private(struct list_head *adj_list) 7230 { 7231 struct netdev_adjacent *adj; 7232 7233 adj = list_entry(adj_list, struct netdev_adjacent, list); 7234 7235 return adj->private; 7236 } 7237 EXPORT_SYMBOL(netdev_adjacent_get_private); 7238 7239 /** 7240 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list 7241 * @dev: device 7242 * @iter: list_head ** of the current position 7243 * 7244 * Gets the next device from the dev's upper list, starting from iter 7245 * position. The caller must hold RCU read lock. 7246 */ 7247 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, 7248 struct list_head **iter) 7249 { 7250 struct netdev_adjacent *upper; 7251 7252 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 7253 7254 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7255 7256 if (&upper->list == &dev->adj_list.upper) 7257 return NULL; 7258 7259 *iter = &upper->list; 7260 7261 return upper->dev; 7262 } 7263 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); 7264 7265 static struct net_device *__netdev_next_upper_dev(struct net_device *dev, 7266 struct list_head **iter, 7267 bool *ignore) 7268 { 7269 struct netdev_adjacent *upper; 7270 7271 upper = list_entry((*iter)->next, struct netdev_adjacent, list); 7272 7273 if (&upper->list == &dev->adj_list.upper) 7274 return NULL; 7275 7276 *iter = &upper->list; 7277 *ignore = upper->ignore; 7278 7279 return upper->dev; 7280 } 7281 7282 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, 7283 struct list_head **iter) 7284 { 7285 struct netdev_adjacent *upper; 7286 7287 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 7288 7289 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7290 7291 if (&upper->list == &dev->adj_list.upper) 7292 return NULL; 7293 7294 *iter = &upper->list; 7295 7296 return upper->dev; 7297 } 7298 7299 static int __netdev_walk_all_upper_dev(struct net_device *dev, 7300 int (*fn)(struct net_device *dev, 7301 struct netdev_nested_priv *priv), 7302 struct netdev_nested_priv *priv) 7303 { 7304 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7305 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7306 int ret, cur = 0; 7307 bool ignore; 7308 7309 now = dev; 7310 iter = &dev->adj_list.upper; 7311 7312 while (1) { 7313 if (now != dev) { 7314 ret = fn(now, priv); 7315 if (ret) 7316 return ret; 7317 } 7318 7319 next = NULL; 7320 while (1) { 7321 udev = __netdev_next_upper_dev(now, &iter, &ignore); 7322 if (!udev) 7323 break; 7324 if (ignore) 7325 continue; 7326 7327 next = udev; 7328 niter = &udev->adj_list.upper; 7329 dev_stack[cur] = now; 7330 iter_stack[cur++] = iter; 7331 break; 7332 } 7333 7334 if (!next) { 7335 if (!cur) 7336 return 0; 7337 next = dev_stack[--cur]; 7338 niter = iter_stack[cur]; 7339 } 7340 7341 now = next; 7342 iter = niter; 7343 } 7344 7345 return 0; 7346 } 7347 7348 int netdev_walk_all_upper_dev_rcu(struct net_device *dev, 7349 int (*fn)(struct net_device *dev, 7350 struct netdev_nested_priv *priv), 7351 struct netdev_nested_priv *priv) 7352 { 7353 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7354 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7355 int ret, cur = 0; 7356 7357 now = dev; 7358 iter = &dev->adj_list.upper; 7359 7360 while (1) { 7361 if (now != dev) { 7362 ret = fn(now, priv); 7363 if (ret) 7364 return ret; 7365 } 7366 7367 next = NULL; 7368 while (1) { 7369 udev = netdev_next_upper_dev_rcu(now, &iter); 7370 if (!udev) 7371 break; 7372 7373 next = udev; 7374 niter = &udev->adj_list.upper; 7375 dev_stack[cur] = now; 7376 iter_stack[cur++] = iter; 7377 break; 7378 } 7379 7380 if (!next) { 7381 if (!cur) 7382 return 0; 7383 next = dev_stack[--cur]; 7384 niter = iter_stack[cur]; 7385 } 7386 7387 now = next; 7388 iter = niter; 7389 } 7390 7391 return 0; 7392 } 7393 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); 7394 7395 static bool __netdev_has_upper_dev(struct net_device *dev, 7396 struct net_device *upper_dev) 7397 { 7398 struct netdev_nested_priv priv = { 7399 .flags = 0, 7400 .data = (void *)upper_dev, 7401 }; 7402 7403 ASSERT_RTNL(); 7404 7405 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev, 7406 &priv); 7407 } 7408 7409 /** 7410 * netdev_lower_get_next_private - Get the next ->private from the 7411 * lower neighbour list 7412 * @dev: device 7413 * @iter: list_head ** of the current position 7414 * 7415 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7416 * list, starting from iter position. The caller must hold either hold the 7417 * RTNL lock or its own locking that guarantees that the neighbour lower 7418 * list will remain unchanged. 7419 */ 7420 void *netdev_lower_get_next_private(struct net_device *dev, 7421 struct list_head **iter) 7422 { 7423 struct netdev_adjacent *lower; 7424 7425 lower = list_entry(*iter, struct netdev_adjacent, list); 7426 7427 if (&lower->list == &dev->adj_list.lower) 7428 return NULL; 7429 7430 *iter = lower->list.next; 7431 7432 return lower->private; 7433 } 7434 EXPORT_SYMBOL(netdev_lower_get_next_private); 7435 7436 /** 7437 * netdev_lower_get_next_private_rcu - Get the next ->private from the 7438 * lower neighbour list, RCU 7439 * variant 7440 * @dev: device 7441 * @iter: list_head ** of the current position 7442 * 7443 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7444 * list, starting from iter position. The caller must hold RCU read lock. 7445 */ 7446 void *netdev_lower_get_next_private_rcu(struct net_device *dev, 7447 struct list_head **iter) 7448 { 7449 struct netdev_adjacent *lower; 7450 7451 WARN_ON_ONCE(!rcu_read_lock_held()); 7452 7453 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7454 7455 if (&lower->list == &dev->adj_list.lower) 7456 return NULL; 7457 7458 *iter = &lower->list; 7459 7460 return lower->private; 7461 } 7462 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); 7463 7464 /** 7465 * netdev_lower_get_next - Get the next device from the lower neighbour 7466 * list 7467 * @dev: device 7468 * @iter: list_head ** of the current position 7469 * 7470 * Gets the next netdev_adjacent from the dev's lower neighbour 7471 * list, starting from iter position. The caller must hold RTNL lock or 7472 * its own locking that guarantees that the neighbour lower 7473 * list will remain unchanged. 7474 */ 7475 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) 7476 { 7477 struct netdev_adjacent *lower; 7478 7479 lower = list_entry(*iter, struct netdev_adjacent, list); 7480 7481 if (&lower->list == &dev->adj_list.lower) 7482 return NULL; 7483 7484 *iter = lower->list.next; 7485 7486 return lower->dev; 7487 } 7488 EXPORT_SYMBOL(netdev_lower_get_next); 7489 7490 static struct net_device *netdev_next_lower_dev(struct net_device *dev, 7491 struct list_head **iter) 7492 { 7493 struct netdev_adjacent *lower; 7494 7495 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7496 7497 if (&lower->list == &dev->adj_list.lower) 7498 return NULL; 7499 7500 *iter = &lower->list; 7501 7502 return lower->dev; 7503 } 7504 7505 static struct net_device *__netdev_next_lower_dev(struct net_device *dev, 7506 struct list_head **iter, 7507 bool *ignore) 7508 { 7509 struct netdev_adjacent *lower; 7510 7511 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7512 7513 if (&lower->list == &dev->adj_list.lower) 7514 return NULL; 7515 7516 *iter = &lower->list; 7517 *ignore = lower->ignore; 7518 7519 return lower->dev; 7520 } 7521 7522 int netdev_walk_all_lower_dev(struct net_device *dev, 7523 int (*fn)(struct net_device *dev, 7524 struct netdev_nested_priv *priv), 7525 struct netdev_nested_priv *priv) 7526 { 7527 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7528 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7529 int ret, cur = 0; 7530 7531 now = dev; 7532 iter = &dev->adj_list.lower; 7533 7534 while (1) { 7535 if (now != dev) { 7536 ret = fn(now, priv); 7537 if (ret) 7538 return ret; 7539 } 7540 7541 next = NULL; 7542 while (1) { 7543 ldev = netdev_next_lower_dev(now, &iter); 7544 if (!ldev) 7545 break; 7546 7547 next = ldev; 7548 niter = &ldev->adj_list.lower; 7549 dev_stack[cur] = now; 7550 iter_stack[cur++] = iter; 7551 break; 7552 } 7553 7554 if (!next) { 7555 if (!cur) 7556 return 0; 7557 next = dev_stack[--cur]; 7558 niter = iter_stack[cur]; 7559 } 7560 7561 now = next; 7562 iter = niter; 7563 } 7564 7565 return 0; 7566 } 7567 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); 7568 7569 static int __netdev_walk_all_lower_dev(struct net_device *dev, 7570 int (*fn)(struct net_device *dev, 7571 struct netdev_nested_priv *priv), 7572 struct netdev_nested_priv *priv) 7573 { 7574 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7575 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7576 int ret, cur = 0; 7577 bool ignore; 7578 7579 now = dev; 7580 iter = &dev->adj_list.lower; 7581 7582 while (1) { 7583 if (now != dev) { 7584 ret = fn(now, priv); 7585 if (ret) 7586 return ret; 7587 } 7588 7589 next = NULL; 7590 while (1) { 7591 ldev = __netdev_next_lower_dev(now, &iter, &ignore); 7592 if (!ldev) 7593 break; 7594 if (ignore) 7595 continue; 7596 7597 next = ldev; 7598 niter = &ldev->adj_list.lower; 7599 dev_stack[cur] = now; 7600 iter_stack[cur++] = iter; 7601 break; 7602 } 7603 7604 if (!next) { 7605 if (!cur) 7606 return 0; 7607 next = dev_stack[--cur]; 7608 niter = iter_stack[cur]; 7609 } 7610 7611 now = next; 7612 iter = niter; 7613 } 7614 7615 return 0; 7616 } 7617 7618 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, 7619 struct list_head **iter) 7620 { 7621 struct netdev_adjacent *lower; 7622 7623 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7624 if (&lower->list == &dev->adj_list.lower) 7625 return NULL; 7626 7627 *iter = &lower->list; 7628 7629 return lower->dev; 7630 } 7631 EXPORT_SYMBOL(netdev_next_lower_dev_rcu); 7632 7633 static u8 __netdev_upper_depth(struct net_device *dev) 7634 { 7635 struct net_device *udev; 7636 struct list_head *iter; 7637 u8 max_depth = 0; 7638 bool ignore; 7639 7640 for (iter = &dev->adj_list.upper, 7641 udev = __netdev_next_upper_dev(dev, &iter, &ignore); 7642 udev; 7643 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) { 7644 if (ignore) 7645 continue; 7646 if (max_depth < udev->upper_level) 7647 max_depth = udev->upper_level; 7648 } 7649 7650 return max_depth; 7651 } 7652 7653 static u8 __netdev_lower_depth(struct net_device *dev) 7654 { 7655 struct net_device *ldev; 7656 struct list_head *iter; 7657 u8 max_depth = 0; 7658 bool ignore; 7659 7660 for (iter = &dev->adj_list.lower, 7661 ldev = __netdev_next_lower_dev(dev, &iter, &ignore); 7662 ldev; 7663 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) { 7664 if (ignore) 7665 continue; 7666 if (max_depth < ldev->lower_level) 7667 max_depth = ldev->lower_level; 7668 } 7669 7670 return max_depth; 7671 } 7672 7673 static int __netdev_update_upper_level(struct net_device *dev, 7674 struct netdev_nested_priv *__unused) 7675 { 7676 dev->upper_level = __netdev_upper_depth(dev) + 1; 7677 return 0; 7678 } 7679 7680 static int __netdev_update_lower_level(struct net_device *dev, 7681 struct netdev_nested_priv *priv) 7682 { 7683 dev->lower_level = __netdev_lower_depth(dev) + 1; 7684 7685 #ifdef CONFIG_LOCKDEP 7686 if (!priv) 7687 return 0; 7688 7689 if (priv->flags & NESTED_SYNC_IMM) 7690 dev->nested_level = dev->lower_level - 1; 7691 if (priv->flags & NESTED_SYNC_TODO) 7692 net_unlink_todo(dev); 7693 #endif 7694 return 0; 7695 } 7696 7697 int netdev_walk_all_lower_dev_rcu(struct net_device *dev, 7698 int (*fn)(struct net_device *dev, 7699 struct netdev_nested_priv *priv), 7700 struct netdev_nested_priv *priv) 7701 { 7702 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7703 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7704 int ret, cur = 0; 7705 7706 now = dev; 7707 iter = &dev->adj_list.lower; 7708 7709 while (1) { 7710 if (now != dev) { 7711 ret = fn(now, priv); 7712 if (ret) 7713 return ret; 7714 } 7715 7716 next = NULL; 7717 while (1) { 7718 ldev = netdev_next_lower_dev_rcu(now, &iter); 7719 if (!ldev) 7720 break; 7721 7722 next = ldev; 7723 niter = &ldev->adj_list.lower; 7724 dev_stack[cur] = now; 7725 iter_stack[cur++] = iter; 7726 break; 7727 } 7728 7729 if (!next) { 7730 if (!cur) 7731 return 0; 7732 next = dev_stack[--cur]; 7733 niter = iter_stack[cur]; 7734 } 7735 7736 now = next; 7737 iter = niter; 7738 } 7739 7740 return 0; 7741 } 7742 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); 7743 7744 /** 7745 * netdev_lower_get_first_private_rcu - Get the first ->private from the 7746 * lower neighbour list, RCU 7747 * variant 7748 * @dev: device 7749 * 7750 * Gets the first netdev_adjacent->private from the dev's lower neighbour 7751 * list. The caller must hold RCU read lock. 7752 */ 7753 void *netdev_lower_get_first_private_rcu(struct net_device *dev) 7754 { 7755 struct netdev_adjacent *lower; 7756 7757 lower = list_first_or_null_rcu(&dev->adj_list.lower, 7758 struct netdev_adjacent, list); 7759 if (lower) 7760 return lower->private; 7761 return NULL; 7762 } 7763 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); 7764 7765 /** 7766 * netdev_master_upper_dev_get_rcu - Get master upper device 7767 * @dev: device 7768 * 7769 * Find a master upper device and return pointer to it or NULL in case 7770 * it's not there. The caller must hold the RCU read lock. 7771 */ 7772 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) 7773 { 7774 struct netdev_adjacent *upper; 7775 7776 upper = list_first_or_null_rcu(&dev->adj_list.upper, 7777 struct netdev_adjacent, list); 7778 if (upper && likely(upper->master)) 7779 return upper->dev; 7780 return NULL; 7781 } 7782 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); 7783 7784 static int netdev_adjacent_sysfs_add(struct net_device *dev, 7785 struct net_device *adj_dev, 7786 struct list_head *dev_list) 7787 { 7788 char linkname[IFNAMSIZ+7]; 7789 7790 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7791 "upper_%s" : "lower_%s", adj_dev->name); 7792 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), 7793 linkname); 7794 } 7795 static void netdev_adjacent_sysfs_del(struct net_device *dev, 7796 char *name, 7797 struct list_head *dev_list) 7798 { 7799 char linkname[IFNAMSIZ+7]; 7800 7801 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7802 "upper_%s" : "lower_%s", name); 7803 sysfs_remove_link(&(dev->dev.kobj), linkname); 7804 } 7805 7806 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, 7807 struct net_device *adj_dev, 7808 struct list_head *dev_list) 7809 { 7810 return (dev_list == &dev->adj_list.upper || 7811 dev_list == &dev->adj_list.lower) && 7812 net_eq(dev_net(dev), dev_net(adj_dev)); 7813 } 7814 7815 static int __netdev_adjacent_dev_insert(struct net_device *dev, 7816 struct net_device *adj_dev, 7817 struct list_head *dev_list, 7818 void *private, bool master) 7819 { 7820 struct netdev_adjacent *adj; 7821 int ret; 7822 7823 adj = __netdev_find_adj(adj_dev, dev_list); 7824 7825 if (adj) { 7826 adj->ref_nr += 1; 7827 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", 7828 dev->name, adj_dev->name, adj->ref_nr); 7829 7830 return 0; 7831 } 7832 7833 adj = kmalloc(sizeof(*adj), GFP_KERNEL); 7834 if (!adj) 7835 return -ENOMEM; 7836 7837 adj->dev = adj_dev; 7838 adj->master = master; 7839 adj->ref_nr = 1; 7840 adj->private = private; 7841 adj->ignore = false; 7842 dev_hold(adj_dev); 7843 7844 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", 7845 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); 7846 7847 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { 7848 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); 7849 if (ret) 7850 goto free_adj; 7851 } 7852 7853 /* Ensure that master link is always the first item in list. */ 7854 if (master) { 7855 ret = sysfs_create_link(&(dev->dev.kobj), 7856 &(adj_dev->dev.kobj), "master"); 7857 if (ret) 7858 goto remove_symlinks; 7859 7860 list_add_rcu(&adj->list, dev_list); 7861 } else { 7862 list_add_tail_rcu(&adj->list, dev_list); 7863 } 7864 7865 return 0; 7866 7867 remove_symlinks: 7868 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7869 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7870 free_adj: 7871 kfree(adj); 7872 dev_put(adj_dev); 7873 7874 return ret; 7875 } 7876 7877 static void __netdev_adjacent_dev_remove(struct net_device *dev, 7878 struct net_device *adj_dev, 7879 u16 ref_nr, 7880 struct list_head *dev_list) 7881 { 7882 struct netdev_adjacent *adj; 7883 7884 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", 7885 dev->name, adj_dev->name, ref_nr); 7886 7887 adj = __netdev_find_adj(adj_dev, dev_list); 7888 7889 if (!adj) { 7890 pr_err("Adjacency does not exist for device %s from %s\n", 7891 dev->name, adj_dev->name); 7892 WARN_ON(1); 7893 return; 7894 } 7895 7896 if (adj->ref_nr > ref_nr) { 7897 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", 7898 dev->name, adj_dev->name, ref_nr, 7899 adj->ref_nr - ref_nr); 7900 adj->ref_nr -= ref_nr; 7901 return; 7902 } 7903 7904 if (adj->master) 7905 sysfs_remove_link(&(dev->dev.kobj), "master"); 7906 7907 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7908 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7909 7910 list_del_rcu(&adj->list); 7911 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", 7912 adj_dev->name, dev->name, adj_dev->name); 7913 dev_put(adj_dev); 7914 kfree_rcu(adj, rcu); 7915 } 7916 7917 static int __netdev_adjacent_dev_link_lists(struct net_device *dev, 7918 struct net_device *upper_dev, 7919 struct list_head *up_list, 7920 struct list_head *down_list, 7921 void *private, bool master) 7922 { 7923 int ret; 7924 7925 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, 7926 private, master); 7927 if (ret) 7928 return ret; 7929 7930 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, 7931 private, false); 7932 if (ret) { 7933 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); 7934 return ret; 7935 } 7936 7937 return 0; 7938 } 7939 7940 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, 7941 struct net_device *upper_dev, 7942 u16 ref_nr, 7943 struct list_head *up_list, 7944 struct list_head *down_list) 7945 { 7946 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); 7947 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); 7948 } 7949 7950 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, 7951 struct net_device *upper_dev, 7952 void *private, bool master) 7953 { 7954 return __netdev_adjacent_dev_link_lists(dev, upper_dev, 7955 &dev->adj_list.upper, 7956 &upper_dev->adj_list.lower, 7957 private, master); 7958 } 7959 7960 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, 7961 struct net_device *upper_dev) 7962 { 7963 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, 7964 &dev->adj_list.upper, 7965 &upper_dev->adj_list.lower); 7966 } 7967 7968 static int __netdev_upper_dev_link(struct net_device *dev, 7969 struct net_device *upper_dev, bool master, 7970 void *upper_priv, void *upper_info, 7971 struct netdev_nested_priv *priv, 7972 struct netlink_ext_ack *extack) 7973 { 7974 struct netdev_notifier_changeupper_info changeupper_info = { 7975 .info = { 7976 .dev = dev, 7977 .extack = extack, 7978 }, 7979 .upper_dev = upper_dev, 7980 .master = master, 7981 .linking = true, 7982 .upper_info = upper_info, 7983 }; 7984 struct net_device *master_dev; 7985 int ret = 0; 7986 7987 ASSERT_RTNL(); 7988 7989 if (dev == upper_dev) 7990 return -EBUSY; 7991 7992 /* To prevent loops, check if dev is not upper device to upper_dev. */ 7993 if (__netdev_has_upper_dev(upper_dev, dev)) 7994 return -EBUSY; 7995 7996 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV) 7997 return -EMLINK; 7998 7999 if (!master) { 8000 if (__netdev_has_upper_dev(dev, upper_dev)) 8001 return -EEXIST; 8002 } else { 8003 master_dev = __netdev_master_upper_dev_get(dev); 8004 if (master_dev) 8005 return master_dev == upper_dev ? -EEXIST : -EBUSY; 8006 } 8007 8008 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 8009 &changeupper_info.info); 8010 ret = notifier_to_errno(ret); 8011 if (ret) 8012 return ret; 8013 8014 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, 8015 master); 8016 if (ret) 8017 return ret; 8018 8019 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 8020 &changeupper_info.info); 8021 ret = notifier_to_errno(ret); 8022 if (ret) 8023 goto rollback; 8024 8025 __netdev_update_upper_level(dev, NULL); 8026 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 8027 8028 __netdev_update_lower_level(upper_dev, priv); 8029 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 8030 priv); 8031 8032 return 0; 8033 8034 rollback: 8035 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 8036 8037 return ret; 8038 } 8039 8040 /** 8041 * netdev_upper_dev_link - Add a link to the upper device 8042 * @dev: device 8043 * @upper_dev: new upper device 8044 * @extack: netlink extended ack 8045 * 8046 * Adds a link to device which is upper to this one. The caller must hold 8047 * the RTNL lock. On a failure a negative errno code is returned. 8048 * On success the reference counts are adjusted and the function 8049 * returns zero. 8050 */ 8051 int netdev_upper_dev_link(struct net_device *dev, 8052 struct net_device *upper_dev, 8053 struct netlink_ext_ack *extack) 8054 { 8055 struct netdev_nested_priv priv = { 8056 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 8057 .data = NULL, 8058 }; 8059 8060 return __netdev_upper_dev_link(dev, upper_dev, false, 8061 NULL, NULL, &priv, extack); 8062 } 8063 EXPORT_SYMBOL(netdev_upper_dev_link); 8064 8065 /** 8066 * netdev_master_upper_dev_link - Add a master link to the upper device 8067 * @dev: device 8068 * @upper_dev: new upper device 8069 * @upper_priv: upper device private 8070 * @upper_info: upper info to be passed down via notifier 8071 * @extack: netlink extended ack 8072 * 8073 * Adds a link to device which is upper to this one. In this case, only 8074 * one master upper device can be linked, although other non-master devices 8075 * might be linked as well. The caller must hold the RTNL lock. 8076 * On a failure a negative errno code is returned. On success the reference 8077 * counts are adjusted and the function returns zero. 8078 */ 8079 int netdev_master_upper_dev_link(struct net_device *dev, 8080 struct net_device *upper_dev, 8081 void *upper_priv, void *upper_info, 8082 struct netlink_ext_ack *extack) 8083 { 8084 struct netdev_nested_priv priv = { 8085 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 8086 .data = NULL, 8087 }; 8088 8089 return __netdev_upper_dev_link(dev, upper_dev, true, 8090 upper_priv, upper_info, &priv, extack); 8091 } 8092 EXPORT_SYMBOL(netdev_master_upper_dev_link); 8093 8094 static void __netdev_upper_dev_unlink(struct net_device *dev, 8095 struct net_device *upper_dev, 8096 struct netdev_nested_priv *priv) 8097 { 8098 struct netdev_notifier_changeupper_info changeupper_info = { 8099 .info = { 8100 .dev = dev, 8101 }, 8102 .upper_dev = upper_dev, 8103 .linking = false, 8104 }; 8105 8106 ASSERT_RTNL(); 8107 8108 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; 8109 8110 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 8111 &changeupper_info.info); 8112 8113 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 8114 8115 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 8116 &changeupper_info.info); 8117 8118 __netdev_update_upper_level(dev, NULL); 8119 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 8120 8121 __netdev_update_lower_level(upper_dev, priv); 8122 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 8123 priv); 8124 } 8125 8126 /** 8127 * netdev_upper_dev_unlink - Removes a link to upper device 8128 * @dev: device 8129 * @upper_dev: new upper device 8130 * 8131 * Removes a link to device which is upper to this one. The caller must hold 8132 * the RTNL lock. 8133 */ 8134 void netdev_upper_dev_unlink(struct net_device *dev, 8135 struct net_device *upper_dev) 8136 { 8137 struct netdev_nested_priv priv = { 8138 .flags = NESTED_SYNC_TODO, 8139 .data = NULL, 8140 }; 8141 8142 __netdev_upper_dev_unlink(dev, upper_dev, &priv); 8143 } 8144 EXPORT_SYMBOL(netdev_upper_dev_unlink); 8145 8146 static void __netdev_adjacent_dev_set(struct net_device *upper_dev, 8147 struct net_device *lower_dev, 8148 bool val) 8149 { 8150 struct netdev_adjacent *adj; 8151 8152 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower); 8153 if (adj) 8154 adj->ignore = val; 8155 8156 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper); 8157 if (adj) 8158 adj->ignore = val; 8159 } 8160 8161 static void netdev_adjacent_dev_disable(struct net_device *upper_dev, 8162 struct net_device *lower_dev) 8163 { 8164 __netdev_adjacent_dev_set(upper_dev, lower_dev, true); 8165 } 8166 8167 static void netdev_adjacent_dev_enable(struct net_device *upper_dev, 8168 struct net_device *lower_dev) 8169 { 8170 __netdev_adjacent_dev_set(upper_dev, lower_dev, false); 8171 } 8172 8173 int netdev_adjacent_change_prepare(struct net_device *old_dev, 8174 struct net_device *new_dev, 8175 struct net_device *dev, 8176 struct netlink_ext_ack *extack) 8177 { 8178 struct netdev_nested_priv priv = { 8179 .flags = 0, 8180 .data = NULL, 8181 }; 8182 int err; 8183 8184 if (!new_dev) 8185 return 0; 8186 8187 if (old_dev && new_dev != old_dev) 8188 netdev_adjacent_dev_disable(dev, old_dev); 8189 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv, 8190 extack); 8191 if (err) { 8192 if (old_dev && new_dev != old_dev) 8193 netdev_adjacent_dev_enable(dev, old_dev); 8194 return err; 8195 } 8196 8197 return 0; 8198 } 8199 EXPORT_SYMBOL(netdev_adjacent_change_prepare); 8200 8201 void netdev_adjacent_change_commit(struct net_device *old_dev, 8202 struct net_device *new_dev, 8203 struct net_device *dev) 8204 { 8205 struct netdev_nested_priv priv = { 8206 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 8207 .data = NULL, 8208 }; 8209 8210 if (!new_dev || !old_dev) 8211 return; 8212 8213 if (new_dev == old_dev) 8214 return; 8215 8216 netdev_adjacent_dev_enable(dev, old_dev); 8217 __netdev_upper_dev_unlink(old_dev, dev, &priv); 8218 } 8219 EXPORT_SYMBOL(netdev_adjacent_change_commit); 8220 8221 void netdev_adjacent_change_abort(struct net_device *old_dev, 8222 struct net_device *new_dev, 8223 struct net_device *dev) 8224 { 8225 struct netdev_nested_priv priv = { 8226 .flags = 0, 8227 .data = NULL, 8228 }; 8229 8230 if (!new_dev) 8231 return; 8232 8233 if (old_dev && new_dev != old_dev) 8234 netdev_adjacent_dev_enable(dev, old_dev); 8235 8236 __netdev_upper_dev_unlink(new_dev, dev, &priv); 8237 } 8238 EXPORT_SYMBOL(netdev_adjacent_change_abort); 8239 8240 /** 8241 * netdev_bonding_info_change - Dispatch event about slave change 8242 * @dev: device 8243 * @bonding_info: info to dispatch 8244 * 8245 * Send NETDEV_BONDING_INFO to netdev notifiers with info. 8246 * The caller must hold the RTNL lock. 8247 */ 8248 void netdev_bonding_info_change(struct net_device *dev, 8249 struct netdev_bonding_info *bonding_info) 8250 { 8251 struct netdev_notifier_bonding_info info = { 8252 .info.dev = dev, 8253 }; 8254 8255 memcpy(&info.bonding_info, bonding_info, 8256 sizeof(struct netdev_bonding_info)); 8257 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, 8258 &info.info); 8259 } 8260 EXPORT_SYMBOL(netdev_bonding_info_change); 8261 8262 /** 8263 * netdev_get_xmit_slave - Get the xmit slave of master device 8264 * @dev: device 8265 * @skb: The packet 8266 * @all_slaves: assume all the slaves are active 8267 * 8268 * The reference counters are not incremented so the caller must be 8269 * careful with locks. The caller must hold RCU lock. 8270 * %NULL is returned if no slave is found. 8271 */ 8272 8273 struct net_device *netdev_get_xmit_slave(struct net_device *dev, 8274 struct sk_buff *skb, 8275 bool all_slaves) 8276 { 8277 const struct net_device_ops *ops = dev->netdev_ops; 8278 8279 if (!ops->ndo_get_xmit_slave) 8280 return NULL; 8281 return ops->ndo_get_xmit_slave(dev, skb, all_slaves); 8282 } 8283 EXPORT_SYMBOL(netdev_get_xmit_slave); 8284 8285 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev, 8286 struct sock *sk) 8287 { 8288 const struct net_device_ops *ops = dev->netdev_ops; 8289 8290 if (!ops->ndo_sk_get_lower_dev) 8291 return NULL; 8292 return ops->ndo_sk_get_lower_dev(dev, sk); 8293 } 8294 8295 /** 8296 * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket 8297 * @dev: device 8298 * @sk: the socket 8299 * 8300 * %NULL is returned if no lower device is found. 8301 */ 8302 8303 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev, 8304 struct sock *sk) 8305 { 8306 struct net_device *lower; 8307 8308 lower = netdev_sk_get_lower_dev(dev, sk); 8309 while (lower) { 8310 dev = lower; 8311 lower = netdev_sk_get_lower_dev(dev, sk); 8312 } 8313 8314 return dev; 8315 } 8316 EXPORT_SYMBOL(netdev_sk_get_lowest_dev); 8317 8318 static void netdev_adjacent_add_links(struct net_device *dev) 8319 { 8320 struct netdev_adjacent *iter; 8321 8322 struct net *net = dev_net(dev); 8323 8324 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8325 if (!net_eq(net, dev_net(iter->dev))) 8326 continue; 8327 netdev_adjacent_sysfs_add(iter->dev, dev, 8328 &iter->dev->adj_list.lower); 8329 netdev_adjacent_sysfs_add(dev, iter->dev, 8330 &dev->adj_list.upper); 8331 } 8332 8333 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8334 if (!net_eq(net, dev_net(iter->dev))) 8335 continue; 8336 netdev_adjacent_sysfs_add(iter->dev, dev, 8337 &iter->dev->adj_list.upper); 8338 netdev_adjacent_sysfs_add(dev, iter->dev, 8339 &dev->adj_list.lower); 8340 } 8341 } 8342 8343 static void netdev_adjacent_del_links(struct net_device *dev) 8344 { 8345 struct netdev_adjacent *iter; 8346 8347 struct net *net = dev_net(dev); 8348 8349 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8350 if (!net_eq(net, dev_net(iter->dev))) 8351 continue; 8352 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8353 &iter->dev->adj_list.lower); 8354 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8355 &dev->adj_list.upper); 8356 } 8357 8358 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8359 if (!net_eq(net, dev_net(iter->dev))) 8360 continue; 8361 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8362 &iter->dev->adj_list.upper); 8363 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8364 &dev->adj_list.lower); 8365 } 8366 } 8367 8368 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) 8369 { 8370 struct netdev_adjacent *iter; 8371 8372 struct net *net = dev_net(dev); 8373 8374 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8375 if (!net_eq(net, dev_net(iter->dev))) 8376 continue; 8377 netdev_adjacent_sysfs_del(iter->dev, oldname, 8378 &iter->dev->adj_list.lower); 8379 netdev_adjacent_sysfs_add(iter->dev, dev, 8380 &iter->dev->adj_list.lower); 8381 } 8382 8383 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8384 if (!net_eq(net, dev_net(iter->dev))) 8385 continue; 8386 netdev_adjacent_sysfs_del(iter->dev, oldname, 8387 &iter->dev->adj_list.upper); 8388 netdev_adjacent_sysfs_add(iter->dev, dev, 8389 &iter->dev->adj_list.upper); 8390 } 8391 } 8392 8393 void *netdev_lower_dev_get_private(struct net_device *dev, 8394 struct net_device *lower_dev) 8395 { 8396 struct netdev_adjacent *lower; 8397 8398 if (!lower_dev) 8399 return NULL; 8400 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); 8401 if (!lower) 8402 return NULL; 8403 8404 return lower->private; 8405 } 8406 EXPORT_SYMBOL(netdev_lower_dev_get_private); 8407 8408 8409 /** 8410 * netdev_lower_state_changed - Dispatch event about lower device state change 8411 * @lower_dev: device 8412 * @lower_state_info: state to dispatch 8413 * 8414 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. 8415 * The caller must hold the RTNL lock. 8416 */ 8417 void netdev_lower_state_changed(struct net_device *lower_dev, 8418 void *lower_state_info) 8419 { 8420 struct netdev_notifier_changelowerstate_info changelowerstate_info = { 8421 .info.dev = lower_dev, 8422 }; 8423 8424 ASSERT_RTNL(); 8425 changelowerstate_info.lower_state_info = lower_state_info; 8426 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, 8427 &changelowerstate_info.info); 8428 } 8429 EXPORT_SYMBOL(netdev_lower_state_changed); 8430 8431 static void dev_change_rx_flags(struct net_device *dev, int flags) 8432 { 8433 const struct net_device_ops *ops = dev->netdev_ops; 8434 8435 if (ops->ndo_change_rx_flags) 8436 ops->ndo_change_rx_flags(dev, flags); 8437 } 8438 8439 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) 8440 { 8441 unsigned int old_flags = dev->flags; 8442 kuid_t uid; 8443 kgid_t gid; 8444 8445 ASSERT_RTNL(); 8446 8447 dev->flags |= IFF_PROMISC; 8448 dev->promiscuity += inc; 8449 if (dev->promiscuity == 0) { 8450 /* 8451 * Avoid overflow. 8452 * If inc causes overflow, untouch promisc and return error. 8453 */ 8454 if (inc < 0) 8455 dev->flags &= ~IFF_PROMISC; 8456 else { 8457 dev->promiscuity -= inc; 8458 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n", 8459 dev->name); 8460 return -EOVERFLOW; 8461 } 8462 } 8463 if (dev->flags != old_flags) { 8464 pr_info("device %s %s promiscuous mode\n", 8465 dev->name, 8466 dev->flags & IFF_PROMISC ? "entered" : "left"); 8467 if (audit_enabled) { 8468 current_uid_gid(&uid, &gid); 8469 audit_log(audit_context(), GFP_ATOMIC, 8470 AUDIT_ANOM_PROMISCUOUS, 8471 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", 8472 dev->name, (dev->flags & IFF_PROMISC), 8473 (old_flags & IFF_PROMISC), 8474 from_kuid(&init_user_ns, audit_get_loginuid(current)), 8475 from_kuid(&init_user_ns, uid), 8476 from_kgid(&init_user_ns, gid), 8477 audit_get_sessionid(current)); 8478 } 8479 8480 dev_change_rx_flags(dev, IFF_PROMISC); 8481 } 8482 if (notify) 8483 __dev_notify_flags(dev, old_flags, IFF_PROMISC); 8484 return 0; 8485 } 8486 8487 /** 8488 * dev_set_promiscuity - update promiscuity count on a device 8489 * @dev: device 8490 * @inc: modifier 8491 * 8492 * Add or remove promiscuity from a device. While the count in the device 8493 * remains above zero the interface remains promiscuous. Once it hits zero 8494 * the device reverts back to normal filtering operation. A negative inc 8495 * value is used to drop promiscuity on the device. 8496 * Return 0 if successful or a negative errno code on error. 8497 */ 8498 int dev_set_promiscuity(struct net_device *dev, int inc) 8499 { 8500 unsigned int old_flags = dev->flags; 8501 int err; 8502 8503 err = __dev_set_promiscuity(dev, inc, true); 8504 if (err < 0) 8505 return err; 8506 if (dev->flags != old_flags) 8507 dev_set_rx_mode(dev); 8508 return err; 8509 } 8510 EXPORT_SYMBOL(dev_set_promiscuity); 8511 8512 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) 8513 { 8514 unsigned int old_flags = dev->flags, old_gflags = dev->gflags; 8515 8516 ASSERT_RTNL(); 8517 8518 dev->flags |= IFF_ALLMULTI; 8519 dev->allmulti += inc; 8520 if (dev->allmulti == 0) { 8521 /* 8522 * Avoid overflow. 8523 * If inc causes overflow, untouch allmulti and return error. 8524 */ 8525 if (inc < 0) 8526 dev->flags &= ~IFF_ALLMULTI; 8527 else { 8528 dev->allmulti -= inc; 8529 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n", 8530 dev->name); 8531 return -EOVERFLOW; 8532 } 8533 } 8534 if (dev->flags ^ old_flags) { 8535 dev_change_rx_flags(dev, IFF_ALLMULTI); 8536 dev_set_rx_mode(dev); 8537 if (notify) 8538 __dev_notify_flags(dev, old_flags, 8539 dev->gflags ^ old_gflags); 8540 } 8541 return 0; 8542 } 8543 8544 /** 8545 * dev_set_allmulti - update allmulti count on a device 8546 * @dev: device 8547 * @inc: modifier 8548 * 8549 * Add or remove reception of all multicast frames to a device. While the 8550 * count in the device remains above zero the interface remains listening 8551 * to all interfaces. Once it hits zero the device reverts back to normal 8552 * filtering operation. A negative @inc value is used to drop the counter 8553 * when releasing a resource needing all multicasts. 8554 * Return 0 if successful or a negative errno code on error. 8555 */ 8556 8557 int dev_set_allmulti(struct net_device *dev, int inc) 8558 { 8559 return __dev_set_allmulti(dev, inc, true); 8560 } 8561 EXPORT_SYMBOL(dev_set_allmulti); 8562 8563 /* 8564 * Upload unicast and multicast address lists to device and 8565 * configure RX filtering. When the device doesn't support unicast 8566 * filtering it is put in promiscuous mode while unicast addresses 8567 * are present. 8568 */ 8569 void __dev_set_rx_mode(struct net_device *dev) 8570 { 8571 const struct net_device_ops *ops = dev->netdev_ops; 8572 8573 /* dev_open will call this function so the list will stay sane. */ 8574 if (!(dev->flags&IFF_UP)) 8575 return; 8576 8577 if (!netif_device_present(dev)) 8578 return; 8579 8580 if (!(dev->priv_flags & IFF_UNICAST_FLT)) { 8581 /* Unicast addresses changes may only happen under the rtnl, 8582 * therefore calling __dev_set_promiscuity here is safe. 8583 */ 8584 if (!netdev_uc_empty(dev) && !dev->uc_promisc) { 8585 __dev_set_promiscuity(dev, 1, false); 8586 dev->uc_promisc = true; 8587 } else if (netdev_uc_empty(dev) && dev->uc_promisc) { 8588 __dev_set_promiscuity(dev, -1, false); 8589 dev->uc_promisc = false; 8590 } 8591 } 8592 8593 if (ops->ndo_set_rx_mode) 8594 ops->ndo_set_rx_mode(dev); 8595 } 8596 8597 void dev_set_rx_mode(struct net_device *dev) 8598 { 8599 netif_addr_lock_bh(dev); 8600 __dev_set_rx_mode(dev); 8601 netif_addr_unlock_bh(dev); 8602 } 8603 8604 /** 8605 * dev_get_flags - get flags reported to userspace 8606 * @dev: device 8607 * 8608 * Get the combination of flag bits exported through APIs to userspace. 8609 */ 8610 unsigned int dev_get_flags(const struct net_device *dev) 8611 { 8612 unsigned int flags; 8613 8614 flags = (dev->flags & ~(IFF_PROMISC | 8615 IFF_ALLMULTI | 8616 IFF_RUNNING | 8617 IFF_LOWER_UP | 8618 IFF_DORMANT)) | 8619 (dev->gflags & (IFF_PROMISC | 8620 IFF_ALLMULTI)); 8621 8622 if (netif_running(dev)) { 8623 if (netif_oper_up(dev)) 8624 flags |= IFF_RUNNING; 8625 if (netif_carrier_ok(dev)) 8626 flags |= IFF_LOWER_UP; 8627 if (netif_dormant(dev)) 8628 flags |= IFF_DORMANT; 8629 } 8630 8631 return flags; 8632 } 8633 EXPORT_SYMBOL(dev_get_flags); 8634 8635 int __dev_change_flags(struct net_device *dev, unsigned int flags, 8636 struct netlink_ext_ack *extack) 8637 { 8638 unsigned int old_flags = dev->flags; 8639 int ret; 8640 8641 ASSERT_RTNL(); 8642 8643 /* 8644 * Set the flags on our device. 8645 */ 8646 8647 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | 8648 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | 8649 IFF_AUTOMEDIA)) | 8650 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | 8651 IFF_ALLMULTI)); 8652 8653 /* 8654 * Load in the correct multicast list now the flags have changed. 8655 */ 8656 8657 if ((old_flags ^ flags) & IFF_MULTICAST) 8658 dev_change_rx_flags(dev, IFF_MULTICAST); 8659 8660 dev_set_rx_mode(dev); 8661 8662 /* 8663 * Have we downed the interface. We handle IFF_UP ourselves 8664 * according to user attempts to set it, rather than blindly 8665 * setting it. 8666 */ 8667 8668 ret = 0; 8669 if ((old_flags ^ flags) & IFF_UP) { 8670 if (old_flags & IFF_UP) 8671 __dev_close(dev); 8672 else 8673 ret = __dev_open(dev, extack); 8674 } 8675 8676 if ((flags ^ dev->gflags) & IFF_PROMISC) { 8677 int inc = (flags & IFF_PROMISC) ? 1 : -1; 8678 unsigned int old_flags = dev->flags; 8679 8680 dev->gflags ^= IFF_PROMISC; 8681 8682 if (__dev_set_promiscuity(dev, inc, false) >= 0) 8683 if (dev->flags != old_flags) 8684 dev_set_rx_mode(dev); 8685 } 8686 8687 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI 8688 * is important. Some (broken) drivers set IFF_PROMISC, when 8689 * IFF_ALLMULTI is requested not asking us and not reporting. 8690 */ 8691 if ((flags ^ dev->gflags) & IFF_ALLMULTI) { 8692 int inc = (flags & IFF_ALLMULTI) ? 1 : -1; 8693 8694 dev->gflags ^= IFF_ALLMULTI; 8695 __dev_set_allmulti(dev, inc, false); 8696 } 8697 8698 return ret; 8699 } 8700 8701 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, 8702 unsigned int gchanges) 8703 { 8704 unsigned int changes = dev->flags ^ old_flags; 8705 8706 if (gchanges) 8707 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC); 8708 8709 if (changes & IFF_UP) { 8710 if (dev->flags & IFF_UP) 8711 call_netdevice_notifiers(NETDEV_UP, dev); 8712 else 8713 call_netdevice_notifiers(NETDEV_DOWN, dev); 8714 } 8715 8716 if (dev->flags & IFF_UP && 8717 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { 8718 struct netdev_notifier_change_info change_info = { 8719 .info = { 8720 .dev = dev, 8721 }, 8722 .flags_changed = changes, 8723 }; 8724 8725 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info); 8726 } 8727 } 8728 8729 /** 8730 * dev_change_flags - change device settings 8731 * @dev: device 8732 * @flags: device state flags 8733 * @extack: netlink extended ack 8734 * 8735 * Change settings on device based state flags. The flags are 8736 * in the userspace exported format. 8737 */ 8738 int dev_change_flags(struct net_device *dev, unsigned int flags, 8739 struct netlink_ext_ack *extack) 8740 { 8741 int ret; 8742 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; 8743 8744 ret = __dev_change_flags(dev, flags, extack); 8745 if (ret < 0) 8746 return ret; 8747 8748 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); 8749 __dev_notify_flags(dev, old_flags, changes); 8750 return ret; 8751 } 8752 EXPORT_SYMBOL(dev_change_flags); 8753 8754 int __dev_set_mtu(struct net_device *dev, int new_mtu) 8755 { 8756 const struct net_device_ops *ops = dev->netdev_ops; 8757 8758 if (ops->ndo_change_mtu) 8759 return ops->ndo_change_mtu(dev, new_mtu); 8760 8761 /* Pairs with all the lockless reads of dev->mtu in the stack */ 8762 WRITE_ONCE(dev->mtu, new_mtu); 8763 return 0; 8764 } 8765 EXPORT_SYMBOL(__dev_set_mtu); 8766 8767 int dev_validate_mtu(struct net_device *dev, int new_mtu, 8768 struct netlink_ext_ack *extack) 8769 { 8770 /* MTU must be positive, and in range */ 8771 if (new_mtu < 0 || new_mtu < dev->min_mtu) { 8772 NL_SET_ERR_MSG(extack, "mtu less than device minimum"); 8773 return -EINVAL; 8774 } 8775 8776 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { 8777 NL_SET_ERR_MSG(extack, "mtu greater than device maximum"); 8778 return -EINVAL; 8779 } 8780 return 0; 8781 } 8782 8783 /** 8784 * dev_set_mtu_ext - Change maximum transfer unit 8785 * @dev: device 8786 * @new_mtu: new transfer unit 8787 * @extack: netlink extended ack 8788 * 8789 * Change the maximum transfer size of the network device. 8790 */ 8791 int dev_set_mtu_ext(struct net_device *dev, int new_mtu, 8792 struct netlink_ext_ack *extack) 8793 { 8794 int err, orig_mtu; 8795 8796 if (new_mtu == dev->mtu) 8797 return 0; 8798 8799 err = dev_validate_mtu(dev, new_mtu, extack); 8800 if (err) 8801 return err; 8802 8803 if (!netif_device_present(dev)) 8804 return -ENODEV; 8805 8806 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); 8807 err = notifier_to_errno(err); 8808 if (err) 8809 return err; 8810 8811 orig_mtu = dev->mtu; 8812 err = __dev_set_mtu(dev, new_mtu); 8813 8814 if (!err) { 8815 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8816 orig_mtu); 8817 err = notifier_to_errno(err); 8818 if (err) { 8819 /* setting mtu back and notifying everyone again, 8820 * so that they have a chance to revert changes. 8821 */ 8822 __dev_set_mtu(dev, orig_mtu); 8823 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8824 new_mtu); 8825 } 8826 } 8827 return err; 8828 } 8829 8830 int dev_set_mtu(struct net_device *dev, int new_mtu) 8831 { 8832 struct netlink_ext_ack extack; 8833 int err; 8834 8835 memset(&extack, 0, sizeof(extack)); 8836 err = dev_set_mtu_ext(dev, new_mtu, &extack); 8837 if (err && extack._msg) 8838 net_err_ratelimited("%s: %s\n", dev->name, extack._msg); 8839 return err; 8840 } 8841 EXPORT_SYMBOL(dev_set_mtu); 8842 8843 /** 8844 * dev_change_tx_queue_len - Change TX queue length of a netdevice 8845 * @dev: device 8846 * @new_len: new tx queue length 8847 */ 8848 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len) 8849 { 8850 unsigned int orig_len = dev->tx_queue_len; 8851 int res; 8852 8853 if (new_len != (unsigned int)new_len) 8854 return -ERANGE; 8855 8856 if (new_len != orig_len) { 8857 dev->tx_queue_len = new_len; 8858 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev); 8859 res = notifier_to_errno(res); 8860 if (res) 8861 goto err_rollback; 8862 res = dev_qdisc_change_tx_queue_len(dev); 8863 if (res) 8864 goto err_rollback; 8865 } 8866 8867 return 0; 8868 8869 err_rollback: 8870 netdev_err(dev, "refused to change device tx_queue_len\n"); 8871 dev->tx_queue_len = orig_len; 8872 return res; 8873 } 8874 8875 /** 8876 * dev_set_group - Change group this device belongs to 8877 * @dev: device 8878 * @new_group: group this device should belong to 8879 */ 8880 void dev_set_group(struct net_device *dev, int new_group) 8881 { 8882 dev->group = new_group; 8883 } 8884 EXPORT_SYMBOL(dev_set_group); 8885 8886 /** 8887 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR. 8888 * @dev: device 8889 * @addr: new address 8890 * @extack: netlink extended ack 8891 */ 8892 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr, 8893 struct netlink_ext_ack *extack) 8894 { 8895 struct netdev_notifier_pre_changeaddr_info info = { 8896 .info.dev = dev, 8897 .info.extack = extack, 8898 .dev_addr = addr, 8899 }; 8900 int rc; 8901 8902 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info); 8903 return notifier_to_errno(rc); 8904 } 8905 EXPORT_SYMBOL(dev_pre_changeaddr_notify); 8906 8907 /** 8908 * dev_set_mac_address - Change Media Access Control Address 8909 * @dev: device 8910 * @sa: new address 8911 * @extack: netlink extended ack 8912 * 8913 * Change the hardware (MAC) address of the device 8914 */ 8915 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa, 8916 struct netlink_ext_ack *extack) 8917 { 8918 const struct net_device_ops *ops = dev->netdev_ops; 8919 int err; 8920 8921 if (!ops->ndo_set_mac_address) 8922 return -EOPNOTSUPP; 8923 if (sa->sa_family != dev->type) 8924 return -EINVAL; 8925 if (!netif_device_present(dev)) 8926 return -ENODEV; 8927 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack); 8928 if (err) 8929 return err; 8930 err = ops->ndo_set_mac_address(dev, sa); 8931 if (err) 8932 return err; 8933 dev->addr_assign_type = NET_ADDR_SET; 8934 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); 8935 add_device_randomness(dev->dev_addr, dev->addr_len); 8936 return 0; 8937 } 8938 EXPORT_SYMBOL(dev_set_mac_address); 8939 8940 static DECLARE_RWSEM(dev_addr_sem); 8941 8942 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa, 8943 struct netlink_ext_ack *extack) 8944 { 8945 int ret; 8946 8947 down_write(&dev_addr_sem); 8948 ret = dev_set_mac_address(dev, sa, extack); 8949 up_write(&dev_addr_sem); 8950 return ret; 8951 } 8952 EXPORT_SYMBOL(dev_set_mac_address_user); 8953 8954 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name) 8955 { 8956 size_t size = sizeof(sa->sa_data); 8957 struct net_device *dev; 8958 int ret = 0; 8959 8960 down_read(&dev_addr_sem); 8961 rcu_read_lock(); 8962 8963 dev = dev_get_by_name_rcu(net, dev_name); 8964 if (!dev) { 8965 ret = -ENODEV; 8966 goto unlock; 8967 } 8968 if (!dev->addr_len) 8969 memset(sa->sa_data, 0, size); 8970 else 8971 memcpy(sa->sa_data, dev->dev_addr, 8972 min_t(size_t, size, dev->addr_len)); 8973 sa->sa_family = dev->type; 8974 8975 unlock: 8976 rcu_read_unlock(); 8977 up_read(&dev_addr_sem); 8978 return ret; 8979 } 8980 EXPORT_SYMBOL(dev_get_mac_address); 8981 8982 /** 8983 * dev_change_carrier - Change device carrier 8984 * @dev: device 8985 * @new_carrier: new value 8986 * 8987 * Change device carrier 8988 */ 8989 int dev_change_carrier(struct net_device *dev, bool new_carrier) 8990 { 8991 const struct net_device_ops *ops = dev->netdev_ops; 8992 8993 if (!ops->ndo_change_carrier) 8994 return -EOPNOTSUPP; 8995 if (!netif_device_present(dev)) 8996 return -ENODEV; 8997 return ops->ndo_change_carrier(dev, new_carrier); 8998 } 8999 EXPORT_SYMBOL(dev_change_carrier); 9000 9001 /** 9002 * dev_get_phys_port_id - Get device physical port ID 9003 * @dev: device 9004 * @ppid: port ID 9005 * 9006 * Get device physical port ID 9007 */ 9008 int dev_get_phys_port_id(struct net_device *dev, 9009 struct netdev_phys_item_id *ppid) 9010 { 9011 const struct net_device_ops *ops = dev->netdev_ops; 9012 9013 if (!ops->ndo_get_phys_port_id) 9014 return -EOPNOTSUPP; 9015 return ops->ndo_get_phys_port_id(dev, ppid); 9016 } 9017 EXPORT_SYMBOL(dev_get_phys_port_id); 9018 9019 /** 9020 * dev_get_phys_port_name - Get device physical port name 9021 * @dev: device 9022 * @name: port name 9023 * @len: limit of bytes to copy to name 9024 * 9025 * Get device physical port name 9026 */ 9027 int dev_get_phys_port_name(struct net_device *dev, 9028 char *name, size_t len) 9029 { 9030 const struct net_device_ops *ops = dev->netdev_ops; 9031 int err; 9032 9033 if (ops->ndo_get_phys_port_name) { 9034 err = ops->ndo_get_phys_port_name(dev, name, len); 9035 if (err != -EOPNOTSUPP) 9036 return err; 9037 } 9038 return devlink_compat_phys_port_name_get(dev, name, len); 9039 } 9040 EXPORT_SYMBOL(dev_get_phys_port_name); 9041 9042 /** 9043 * dev_get_port_parent_id - Get the device's port parent identifier 9044 * @dev: network device 9045 * @ppid: pointer to a storage for the port's parent identifier 9046 * @recurse: allow/disallow recursion to lower devices 9047 * 9048 * Get the devices's port parent identifier 9049 */ 9050 int dev_get_port_parent_id(struct net_device *dev, 9051 struct netdev_phys_item_id *ppid, 9052 bool recurse) 9053 { 9054 const struct net_device_ops *ops = dev->netdev_ops; 9055 struct netdev_phys_item_id first = { }; 9056 struct net_device *lower_dev; 9057 struct list_head *iter; 9058 int err; 9059 9060 if (ops->ndo_get_port_parent_id) { 9061 err = ops->ndo_get_port_parent_id(dev, ppid); 9062 if (err != -EOPNOTSUPP) 9063 return err; 9064 } 9065 9066 err = devlink_compat_switch_id_get(dev, ppid); 9067 if (!err || err != -EOPNOTSUPP) 9068 return err; 9069 9070 if (!recurse) 9071 return -EOPNOTSUPP; 9072 9073 netdev_for_each_lower_dev(dev, lower_dev, iter) { 9074 err = dev_get_port_parent_id(lower_dev, ppid, recurse); 9075 if (err) 9076 break; 9077 if (!first.id_len) 9078 first = *ppid; 9079 else if (memcmp(&first, ppid, sizeof(*ppid))) 9080 return -EOPNOTSUPP; 9081 } 9082 9083 return err; 9084 } 9085 EXPORT_SYMBOL(dev_get_port_parent_id); 9086 9087 /** 9088 * netdev_port_same_parent_id - Indicate if two network devices have 9089 * the same port parent identifier 9090 * @a: first network device 9091 * @b: second network device 9092 */ 9093 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b) 9094 { 9095 struct netdev_phys_item_id a_id = { }; 9096 struct netdev_phys_item_id b_id = { }; 9097 9098 if (dev_get_port_parent_id(a, &a_id, true) || 9099 dev_get_port_parent_id(b, &b_id, true)) 9100 return false; 9101 9102 return netdev_phys_item_id_same(&a_id, &b_id); 9103 } 9104 EXPORT_SYMBOL(netdev_port_same_parent_id); 9105 9106 /** 9107 * dev_change_proto_down - update protocol port state information 9108 * @dev: device 9109 * @proto_down: new value 9110 * 9111 * This info can be used by switch drivers to set the phys state of the 9112 * port. 9113 */ 9114 int dev_change_proto_down(struct net_device *dev, bool proto_down) 9115 { 9116 const struct net_device_ops *ops = dev->netdev_ops; 9117 9118 if (!ops->ndo_change_proto_down) 9119 return -EOPNOTSUPP; 9120 if (!netif_device_present(dev)) 9121 return -ENODEV; 9122 return ops->ndo_change_proto_down(dev, proto_down); 9123 } 9124 EXPORT_SYMBOL(dev_change_proto_down); 9125 9126 /** 9127 * dev_change_proto_down_generic - generic implementation for 9128 * ndo_change_proto_down that sets carrier according to 9129 * proto_down. 9130 * 9131 * @dev: device 9132 * @proto_down: new value 9133 */ 9134 int dev_change_proto_down_generic(struct net_device *dev, bool proto_down) 9135 { 9136 if (proto_down) 9137 netif_carrier_off(dev); 9138 else 9139 netif_carrier_on(dev); 9140 dev->proto_down = proto_down; 9141 return 0; 9142 } 9143 EXPORT_SYMBOL(dev_change_proto_down_generic); 9144 9145 /** 9146 * dev_change_proto_down_reason - proto down reason 9147 * 9148 * @dev: device 9149 * @mask: proto down mask 9150 * @value: proto down value 9151 */ 9152 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask, 9153 u32 value) 9154 { 9155 int b; 9156 9157 if (!mask) { 9158 dev->proto_down_reason = value; 9159 } else { 9160 for_each_set_bit(b, &mask, 32) { 9161 if (value & (1 << b)) 9162 dev->proto_down_reason |= BIT(b); 9163 else 9164 dev->proto_down_reason &= ~BIT(b); 9165 } 9166 } 9167 } 9168 EXPORT_SYMBOL(dev_change_proto_down_reason); 9169 9170 struct bpf_xdp_link { 9171 struct bpf_link link; 9172 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */ 9173 int flags; 9174 }; 9175 9176 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags) 9177 { 9178 if (flags & XDP_FLAGS_HW_MODE) 9179 return XDP_MODE_HW; 9180 if (flags & XDP_FLAGS_DRV_MODE) 9181 return XDP_MODE_DRV; 9182 if (flags & XDP_FLAGS_SKB_MODE) 9183 return XDP_MODE_SKB; 9184 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB; 9185 } 9186 9187 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode) 9188 { 9189 switch (mode) { 9190 case XDP_MODE_SKB: 9191 return generic_xdp_install; 9192 case XDP_MODE_DRV: 9193 case XDP_MODE_HW: 9194 return dev->netdev_ops->ndo_bpf; 9195 default: 9196 return NULL; 9197 } 9198 } 9199 9200 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev, 9201 enum bpf_xdp_mode mode) 9202 { 9203 return dev->xdp_state[mode].link; 9204 } 9205 9206 static struct bpf_prog *dev_xdp_prog(struct net_device *dev, 9207 enum bpf_xdp_mode mode) 9208 { 9209 struct bpf_xdp_link *link = dev_xdp_link(dev, mode); 9210 9211 if (link) 9212 return link->link.prog; 9213 return dev->xdp_state[mode].prog; 9214 } 9215 9216 static u8 dev_xdp_prog_count(struct net_device *dev) 9217 { 9218 u8 count = 0; 9219 int i; 9220 9221 for (i = 0; i < __MAX_XDP_MODE; i++) 9222 if (dev->xdp_state[i].prog || dev->xdp_state[i].link) 9223 count++; 9224 return count; 9225 } 9226 9227 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode) 9228 { 9229 struct bpf_prog *prog = dev_xdp_prog(dev, mode); 9230 9231 return prog ? prog->aux->id : 0; 9232 } 9233 9234 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode, 9235 struct bpf_xdp_link *link) 9236 { 9237 dev->xdp_state[mode].link = link; 9238 dev->xdp_state[mode].prog = NULL; 9239 } 9240 9241 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode, 9242 struct bpf_prog *prog) 9243 { 9244 dev->xdp_state[mode].link = NULL; 9245 dev->xdp_state[mode].prog = prog; 9246 } 9247 9248 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode, 9249 bpf_op_t bpf_op, struct netlink_ext_ack *extack, 9250 u32 flags, struct bpf_prog *prog) 9251 { 9252 struct netdev_bpf xdp; 9253 int err; 9254 9255 memset(&xdp, 0, sizeof(xdp)); 9256 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG; 9257 xdp.extack = extack; 9258 xdp.flags = flags; 9259 xdp.prog = prog; 9260 9261 /* Drivers assume refcnt is already incremented (i.e, prog pointer is 9262 * "moved" into driver), so they don't increment it on their own, but 9263 * they do decrement refcnt when program is detached or replaced. 9264 * Given net_device also owns link/prog, we need to bump refcnt here 9265 * to prevent drivers from underflowing it. 9266 */ 9267 if (prog) 9268 bpf_prog_inc(prog); 9269 err = bpf_op(dev, &xdp); 9270 if (err) { 9271 if (prog) 9272 bpf_prog_put(prog); 9273 return err; 9274 } 9275 9276 if (mode != XDP_MODE_HW) 9277 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog); 9278 9279 return 0; 9280 } 9281 9282 static void dev_xdp_uninstall(struct net_device *dev) 9283 { 9284 struct bpf_xdp_link *link; 9285 struct bpf_prog *prog; 9286 enum bpf_xdp_mode mode; 9287 bpf_op_t bpf_op; 9288 9289 ASSERT_RTNL(); 9290 9291 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) { 9292 prog = dev_xdp_prog(dev, mode); 9293 if (!prog) 9294 continue; 9295 9296 bpf_op = dev_xdp_bpf_op(dev, mode); 9297 if (!bpf_op) 9298 continue; 9299 9300 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9301 9302 /* auto-detach link from net device */ 9303 link = dev_xdp_link(dev, mode); 9304 if (link) 9305 link->dev = NULL; 9306 else 9307 bpf_prog_put(prog); 9308 9309 dev_xdp_set_link(dev, mode, NULL); 9310 } 9311 } 9312 9313 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack, 9314 struct bpf_xdp_link *link, struct bpf_prog *new_prog, 9315 struct bpf_prog *old_prog, u32 flags) 9316 { 9317 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES); 9318 struct bpf_prog *cur_prog; 9319 enum bpf_xdp_mode mode; 9320 bpf_op_t bpf_op; 9321 int err; 9322 9323 ASSERT_RTNL(); 9324 9325 /* either link or prog attachment, never both */ 9326 if (link && (new_prog || old_prog)) 9327 return -EINVAL; 9328 /* link supports only XDP mode flags */ 9329 if (link && (flags & ~XDP_FLAGS_MODES)) { 9330 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment"); 9331 return -EINVAL; 9332 } 9333 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */ 9334 if (num_modes > 1) { 9335 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set"); 9336 return -EINVAL; 9337 } 9338 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */ 9339 if (!num_modes && dev_xdp_prog_count(dev) > 1) { 9340 NL_SET_ERR_MSG(extack, 9341 "More than one program loaded, unset mode is ambiguous"); 9342 return -EINVAL; 9343 } 9344 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */ 9345 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) { 9346 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified"); 9347 return -EINVAL; 9348 } 9349 9350 mode = dev_xdp_mode(dev, flags); 9351 /* can't replace attached link */ 9352 if (dev_xdp_link(dev, mode)) { 9353 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link"); 9354 return -EBUSY; 9355 } 9356 9357 cur_prog = dev_xdp_prog(dev, mode); 9358 /* can't replace attached prog with link */ 9359 if (link && cur_prog) { 9360 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link"); 9361 return -EBUSY; 9362 } 9363 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) { 9364 NL_SET_ERR_MSG(extack, "Active program does not match expected"); 9365 return -EEXIST; 9366 } 9367 9368 /* put effective new program into new_prog */ 9369 if (link) 9370 new_prog = link->link.prog; 9371 9372 if (new_prog) { 9373 bool offload = mode == XDP_MODE_HW; 9374 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB 9375 ? XDP_MODE_DRV : XDP_MODE_SKB; 9376 9377 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) { 9378 NL_SET_ERR_MSG(extack, "XDP program already attached"); 9379 return -EBUSY; 9380 } 9381 if (!offload && dev_xdp_prog(dev, other_mode)) { 9382 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time"); 9383 return -EEXIST; 9384 } 9385 if (!offload && bpf_prog_is_dev_bound(new_prog->aux)) { 9386 NL_SET_ERR_MSG(extack, "Using device-bound program without HW_MODE flag is not supported"); 9387 return -EINVAL; 9388 } 9389 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) { 9390 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device"); 9391 return -EINVAL; 9392 } 9393 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) { 9394 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device"); 9395 return -EINVAL; 9396 } 9397 } 9398 9399 /* don't call drivers if the effective program didn't change */ 9400 if (new_prog != cur_prog) { 9401 bpf_op = dev_xdp_bpf_op(dev, mode); 9402 if (!bpf_op) { 9403 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode"); 9404 return -EOPNOTSUPP; 9405 } 9406 9407 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog); 9408 if (err) 9409 return err; 9410 } 9411 9412 if (link) 9413 dev_xdp_set_link(dev, mode, link); 9414 else 9415 dev_xdp_set_prog(dev, mode, new_prog); 9416 if (cur_prog) 9417 bpf_prog_put(cur_prog); 9418 9419 return 0; 9420 } 9421 9422 static int dev_xdp_attach_link(struct net_device *dev, 9423 struct netlink_ext_ack *extack, 9424 struct bpf_xdp_link *link) 9425 { 9426 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags); 9427 } 9428 9429 static int dev_xdp_detach_link(struct net_device *dev, 9430 struct netlink_ext_ack *extack, 9431 struct bpf_xdp_link *link) 9432 { 9433 enum bpf_xdp_mode mode; 9434 bpf_op_t bpf_op; 9435 9436 ASSERT_RTNL(); 9437 9438 mode = dev_xdp_mode(dev, link->flags); 9439 if (dev_xdp_link(dev, mode) != link) 9440 return -EINVAL; 9441 9442 bpf_op = dev_xdp_bpf_op(dev, mode); 9443 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9444 dev_xdp_set_link(dev, mode, NULL); 9445 return 0; 9446 } 9447 9448 static void bpf_xdp_link_release(struct bpf_link *link) 9449 { 9450 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9451 9452 rtnl_lock(); 9453 9454 /* if racing with net_device's tear down, xdp_link->dev might be 9455 * already NULL, in which case link was already auto-detached 9456 */ 9457 if (xdp_link->dev) { 9458 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link)); 9459 xdp_link->dev = NULL; 9460 } 9461 9462 rtnl_unlock(); 9463 } 9464 9465 static int bpf_xdp_link_detach(struct bpf_link *link) 9466 { 9467 bpf_xdp_link_release(link); 9468 return 0; 9469 } 9470 9471 static void bpf_xdp_link_dealloc(struct bpf_link *link) 9472 { 9473 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9474 9475 kfree(xdp_link); 9476 } 9477 9478 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link, 9479 struct seq_file *seq) 9480 { 9481 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9482 u32 ifindex = 0; 9483 9484 rtnl_lock(); 9485 if (xdp_link->dev) 9486 ifindex = xdp_link->dev->ifindex; 9487 rtnl_unlock(); 9488 9489 seq_printf(seq, "ifindex:\t%u\n", ifindex); 9490 } 9491 9492 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link, 9493 struct bpf_link_info *info) 9494 { 9495 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9496 u32 ifindex = 0; 9497 9498 rtnl_lock(); 9499 if (xdp_link->dev) 9500 ifindex = xdp_link->dev->ifindex; 9501 rtnl_unlock(); 9502 9503 info->xdp.ifindex = ifindex; 9504 return 0; 9505 } 9506 9507 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog, 9508 struct bpf_prog *old_prog) 9509 { 9510 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9511 enum bpf_xdp_mode mode; 9512 bpf_op_t bpf_op; 9513 int err = 0; 9514 9515 rtnl_lock(); 9516 9517 /* link might have been auto-released already, so fail */ 9518 if (!xdp_link->dev) { 9519 err = -ENOLINK; 9520 goto out_unlock; 9521 } 9522 9523 if (old_prog && link->prog != old_prog) { 9524 err = -EPERM; 9525 goto out_unlock; 9526 } 9527 old_prog = link->prog; 9528 if (old_prog == new_prog) { 9529 /* no-op, don't disturb drivers */ 9530 bpf_prog_put(new_prog); 9531 goto out_unlock; 9532 } 9533 9534 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags); 9535 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode); 9536 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL, 9537 xdp_link->flags, new_prog); 9538 if (err) 9539 goto out_unlock; 9540 9541 old_prog = xchg(&link->prog, new_prog); 9542 bpf_prog_put(old_prog); 9543 9544 out_unlock: 9545 rtnl_unlock(); 9546 return err; 9547 } 9548 9549 static const struct bpf_link_ops bpf_xdp_link_lops = { 9550 .release = bpf_xdp_link_release, 9551 .dealloc = bpf_xdp_link_dealloc, 9552 .detach = bpf_xdp_link_detach, 9553 .show_fdinfo = bpf_xdp_link_show_fdinfo, 9554 .fill_link_info = bpf_xdp_link_fill_link_info, 9555 .update_prog = bpf_xdp_link_update, 9556 }; 9557 9558 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog) 9559 { 9560 struct net *net = current->nsproxy->net_ns; 9561 struct bpf_link_primer link_primer; 9562 struct bpf_xdp_link *link; 9563 struct net_device *dev; 9564 int err, fd; 9565 9566 dev = dev_get_by_index(net, attr->link_create.target_ifindex); 9567 if (!dev) 9568 return -EINVAL; 9569 9570 link = kzalloc(sizeof(*link), GFP_USER); 9571 if (!link) { 9572 err = -ENOMEM; 9573 goto out_put_dev; 9574 } 9575 9576 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog); 9577 link->dev = dev; 9578 link->flags = attr->link_create.flags; 9579 9580 err = bpf_link_prime(&link->link, &link_primer); 9581 if (err) { 9582 kfree(link); 9583 goto out_put_dev; 9584 } 9585 9586 rtnl_lock(); 9587 err = dev_xdp_attach_link(dev, NULL, link); 9588 rtnl_unlock(); 9589 9590 if (err) { 9591 bpf_link_cleanup(&link_primer); 9592 goto out_put_dev; 9593 } 9594 9595 fd = bpf_link_settle(&link_primer); 9596 /* link itself doesn't hold dev's refcnt to not complicate shutdown */ 9597 dev_put(dev); 9598 return fd; 9599 9600 out_put_dev: 9601 dev_put(dev); 9602 return err; 9603 } 9604 9605 /** 9606 * dev_change_xdp_fd - set or clear a bpf program for a device rx path 9607 * @dev: device 9608 * @extack: netlink extended ack 9609 * @fd: new program fd or negative value to clear 9610 * @expected_fd: old program fd that userspace expects to replace or clear 9611 * @flags: xdp-related flags 9612 * 9613 * Set or clear a bpf program for a device 9614 */ 9615 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, 9616 int fd, int expected_fd, u32 flags) 9617 { 9618 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags); 9619 struct bpf_prog *new_prog = NULL, *old_prog = NULL; 9620 int err; 9621 9622 ASSERT_RTNL(); 9623 9624 if (fd >= 0) { 9625 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP, 9626 mode != XDP_MODE_SKB); 9627 if (IS_ERR(new_prog)) 9628 return PTR_ERR(new_prog); 9629 } 9630 9631 if (expected_fd >= 0) { 9632 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP, 9633 mode != XDP_MODE_SKB); 9634 if (IS_ERR(old_prog)) { 9635 err = PTR_ERR(old_prog); 9636 old_prog = NULL; 9637 goto err_out; 9638 } 9639 } 9640 9641 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags); 9642 9643 err_out: 9644 if (err && new_prog) 9645 bpf_prog_put(new_prog); 9646 if (old_prog) 9647 bpf_prog_put(old_prog); 9648 return err; 9649 } 9650 9651 /** 9652 * dev_new_index - allocate an ifindex 9653 * @net: the applicable net namespace 9654 * 9655 * Returns a suitable unique value for a new device interface 9656 * number. The caller must hold the rtnl semaphore or the 9657 * dev_base_lock to be sure it remains unique. 9658 */ 9659 static int dev_new_index(struct net *net) 9660 { 9661 int ifindex = net->ifindex; 9662 9663 for (;;) { 9664 if (++ifindex <= 0) 9665 ifindex = 1; 9666 if (!__dev_get_by_index(net, ifindex)) 9667 return net->ifindex = ifindex; 9668 } 9669 } 9670 9671 /* Delayed registration/unregisteration */ 9672 static LIST_HEAD(net_todo_list); 9673 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); 9674 9675 static void net_set_todo(struct net_device *dev) 9676 { 9677 list_add_tail(&dev->todo_list, &net_todo_list); 9678 dev_net(dev)->dev_unreg_count++; 9679 } 9680 9681 static netdev_features_t netdev_sync_upper_features(struct net_device *lower, 9682 struct net_device *upper, netdev_features_t features) 9683 { 9684 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9685 netdev_features_t feature; 9686 int feature_bit; 9687 9688 for_each_netdev_feature(upper_disables, feature_bit) { 9689 feature = __NETIF_F_BIT(feature_bit); 9690 if (!(upper->wanted_features & feature) 9691 && (features & feature)) { 9692 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", 9693 &feature, upper->name); 9694 features &= ~feature; 9695 } 9696 } 9697 9698 return features; 9699 } 9700 9701 static void netdev_sync_lower_features(struct net_device *upper, 9702 struct net_device *lower, netdev_features_t features) 9703 { 9704 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9705 netdev_features_t feature; 9706 int feature_bit; 9707 9708 for_each_netdev_feature(upper_disables, feature_bit) { 9709 feature = __NETIF_F_BIT(feature_bit); 9710 if (!(features & feature) && (lower->features & feature)) { 9711 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", 9712 &feature, lower->name); 9713 lower->wanted_features &= ~feature; 9714 __netdev_update_features(lower); 9715 9716 if (unlikely(lower->features & feature)) 9717 netdev_WARN(upper, "failed to disable %pNF on %s!\n", 9718 &feature, lower->name); 9719 else 9720 netdev_features_change(lower); 9721 } 9722 } 9723 } 9724 9725 static netdev_features_t netdev_fix_features(struct net_device *dev, 9726 netdev_features_t features) 9727 { 9728 /* Fix illegal checksum combinations */ 9729 if ((features & NETIF_F_HW_CSUM) && 9730 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { 9731 netdev_warn(dev, "mixed HW and IP checksum settings.\n"); 9732 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); 9733 } 9734 9735 /* TSO requires that SG is present as well. */ 9736 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { 9737 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); 9738 features &= ~NETIF_F_ALL_TSO; 9739 } 9740 9741 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && 9742 !(features & NETIF_F_IP_CSUM)) { 9743 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); 9744 features &= ~NETIF_F_TSO; 9745 features &= ~NETIF_F_TSO_ECN; 9746 } 9747 9748 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && 9749 !(features & NETIF_F_IPV6_CSUM)) { 9750 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); 9751 features &= ~NETIF_F_TSO6; 9752 } 9753 9754 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ 9755 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) 9756 features &= ~NETIF_F_TSO_MANGLEID; 9757 9758 /* TSO ECN requires that TSO is present as well. */ 9759 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) 9760 features &= ~NETIF_F_TSO_ECN; 9761 9762 /* Software GSO depends on SG. */ 9763 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { 9764 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); 9765 features &= ~NETIF_F_GSO; 9766 } 9767 9768 /* GSO partial features require GSO partial be set */ 9769 if ((features & dev->gso_partial_features) && 9770 !(features & NETIF_F_GSO_PARTIAL)) { 9771 netdev_dbg(dev, 9772 "Dropping partially supported GSO features since no GSO partial.\n"); 9773 features &= ~dev->gso_partial_features; 9774 } 9775 9776 if (!(features & NETIF_F_RXCSUM)) { 9777 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet 9778 * successfully merged by hardware must also have the 9779 * checksum verified by hardware. If the user does not 9780 * want to enable RXCSUM, logically, we should disable GRO_HW. 9781 */ 9782 if (features & NETIF_F_GRO_HW) { 9783 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n"); 9784 features &= ~NETIF_F_GRO_HW; 9785 } 9786 } 9787 9788 /* LRO/HW-GRO features cannot be combined with RX-FCS */ 9789 if (features & NETIF_F_RXFCS) { 9790 if (features & NETIF_F_LRO) { 9791 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n"); 9792 features &= ~NETIF_F_LRO; 9793 } 9794 9795 if (features & NETIF_F_GRO_HW) { 9796 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n"); 9797 features &= ~NETIF_F_GRO_HW; 9798 } 9799 } 9800 9801 if (features & NETIF_F_HW_TLS_TX) { 9802 bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) == 9803 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM); 9804 bool hw_csum = features & NETIF_F_HW_CSUM; 9805 9806 if (!ip_csum && !hw_csum) { 9807 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n"); 9808 features &= ~NETIF_F_HW_TLS_TX; 9809 } 9810 } 9811 9812 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) { 9813 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n"); 9814 features &= ~NETIF_F_HW_TLS_RX; 9815 } 9816 9817 return features; 9818 } 9819 9820 int __netdev_update_features(struct net_device *dev) 9821 { 9822 struct net_device *upper, *lower; 9823 netdev_features_t features; 9824 struct list_head *iter; 9825 int err = -1; 9826 9827 ASSERT_RTNL(); 9828 9829 features = netdev_get_wanted_features(dev); 9830 9831 if (dev->netdev_ops->ndo_fix_features) 9832 features = dev->netdev_ops->ndo_fix_features(dev, features); 9833 9834 /* driver might be less strict about feature dependencies */ 9835 features = netdev_fix_features(dev, features); 9836 9837 /* some features can't be enabled if they're off on an upper device */ 9838 netdev_for_each_upper_dev_rcu(dev, upper, iter) 9839 features = netdev_sync_upper_features(dev, upper, features); 9840 9841 if (dev->features == features) 9842 goto sync_lower; 9843 9844 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", 9845 &dev->features, &features); 9846 9847 if (dev->netdev_ops->ndo_set_features) 9848 err = dev->netdev_ops->ndo_set_features(dev, features); 9849 else 9850 err = 0; 9851 9852 if (unlikely(err < 0)) { 9853 netdev_err(dev, 9854 "set_features() failed (%d); wanted %pNF, left %pNF\n", 9855 err, &features, &dev->features); 9856 /* return non-0 since some features might have changed and 9857 * it's better to fire a spurious notification than miss it 9858 */ 9859 return -1; 9860 } 9861 9862 sync_lower: 9863 /* some features must be disabled on lower devices when disabled 9864 * on an upper device (think: bonding master or bridge) 9865 */ 9866 netdev_for_each_lower_dev(dev, lower, iter) 9867 netdev_sync_lower_features(dev, lower, features); 9868 9869 if (!err) { 9870 netdev_features_t diff = features ^ dev->features; 9871 9872 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) { 9873 /* udp_tunnel_{get,drop}_rx_info both need 9874 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the 9875 * device, or they won't do anything. 9876 * Thus we need to update dev->features 9877 * *before* calling udp_tunnel_get_rx_info, 9878 * but *after* calling udp_tunnel_drop_rx_info. 9879 */ 9880 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) { 9881 dev->features = features; 9882 udp_tunnel_get_rx_info(dev); 9883 } else { 9884 udp_tunnel_drop_rx_info(dev); 9885 } 9886 } 9887 9888 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) { 9889 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) { 9890 dev->features = features; 9891 err |= vlan_get_rx_ctag_filter_info(dev); 9892 } else { 9893 vlan_drop_rx_ctag_filter_info(dev); 9894 } 9895 } 9896 9897 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) { 9898 if (features & NETIF_F_HW_VLAN_STAG_FILTER) { 9899 dev->features = features; 9900 err |= vlan_get_rx_stag_filter_info(dev); 9901 } else { 9902 vlan_drop_rx_stag_filter_info(dev); 9903 } 9904 } 9905 9906 dev->features = features; 9907 } 9908 9909 return err < 0 ? 0 : 1; 9910 } 9911 9912 /** 9913 * netdev_update_features - recalculate device features 9914 * @dev: the device to check 9915 * 9916 * Recalculate dev->features set and send notifications if it 9917 * has changed. Should be called after driver or hardware dependent 9918 * conditions might have changed that influence the features. 9919 */ 9920 void netdev_update_features(struct net_device *dev) 9921 { 9922 if (__netdev_update_features(dev)) 9923 netdev_features_change(dev); 9924 } 9925 EXPORT_SYMBOL(netdev_update_features); 9926 9927 /** 9928 * netdev_change_features - recalculate device features 9929 * @dev: the device to check 9930 * 9931 * Recalculate dev->features set and send notifications even 9932 * if they have not changed. Should be called instead of 9933 * netdev_update_features() if also dev->vlan_features might 9934 * have changed to allow the changes to be propagated to stacked 9935 * VLAN devices. 9936 */ 9937 void netdev_change_features(struct net_device *dev) 9938 { 9939 __netdev_update_features(dev); 9940 netdev_features_change(dev); 9941 } 9942 EXPORT_SYMBOL(netdev_change_features); 9943 9944 /** 9945 * netif_stacked_transfer_operstate - transfer operstate 9946 * @rootdev: the root or lower level device to transfer state from 9947 * @dev: the device to transfer operstate to 9948 * 9949 * Transfer operational state from root to device. This is normally 9950 * called when a stacking relationship exists between the root 9951 * device and the device(a leaf device). 9952 */ 9953 void netif_stacked_transfer_operstate(const struct net_device *rootdev, 9954 struct net_device *dev) 9955 { 9956 if (rootdev->operstate == IF_OPER_DORMANT) 9957 netif_dormant_on(dev); 9958 else 9959 netif_dormant_off(dev); 9960 9961 if (rootdev->operstate == IF_OPER_TESTING) 9962 netif_testing_on(dev); 9963 else 9964 netif_testing_off(dev); 9965 9966 if (netif_carrier_ok(rootdev)) 9967 netif_carrier_on(dev); 9968 else 9969 netif_carrier_off(dev); 9970 } 9971 EXPORT_SYMBOL(netif_stacked_transfer_operstate); 9972 9973 static int netif_alloc_rx_queues(struct net_device *dev) 9974 { 9975 unsigned int i, count = dev->num_rx_queues; 9976 struct netdev_rx_queue *rx; 9977 size_t sz = count * sizeof(*rx); 9978 int err = 0; 9979 9980 BUG_ON(count < 1); 9981 9982 rx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL); 9983 if (!rx) 9984 return -ENOMEM; 9985 9986 dev->_rx = rx; 9987 9988 for (i = 0; i < count; i++) { 9989 rx[i].dev = dev; 9990 9991 /* XDP RX-queue setup */ 9992 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0); 9993 if (err < 0) 9994 goto err_rxq_info; 9995 } 9996 return 0; 9997 9998 err_rxq_info: 9999 /* Rollback successful reg's and free other resources */ 10000 while (i--) 10001 xdp_rxq_info_unreg(&rx[i].xdp_rxq); 10002 kvfree(dev->_rx); 10003 dev->_rx = NULL; 10004 return err; 10005 } 10006 10007 static void netif_free_rx_queues(struct net_device *dev) 10008 { 10009 unsigned int i, count = dev->num_rx_queues; 10010 10011 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */ 10012 if (!dev->_rx) 10013 return; 10014 10015 for (i = 0; i < count; i++) 10016 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq); 10017 10018 kvfree(dev->_rx); 10019 } 10020 10021 static void netdev_init_one_queue(struct net_device *dev, 10022 struct netdev_queue *queue, void *_unused) 10023 { 10024 /* Initialize queue lock */ 10025 spin_lock_init(&queue->_xmit_lock); 10026 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); 10027 queue->xmit_lock_owner = -1; 10028 netdev_queue_numa_node_write(queue, NUMA_NO_NODE); 10029 queue->dev = dev; 10030 #ifdef CONFIG_BQL 10031 dql_init(&queue->dql, HZ); 10032 #endif 10033 } 10034 10035 static void netif_free_tx_queues(struct net_device *dev) 10036 { 10037 kvfree(dev->_tx); 10038 } 10039 10040 static int netif_alloc_netdev_queues(struct net_device *dev) 10041 { 10042 unsigned int count = dev->num_tx_queues; 10043 struct netdev_queue *tx; 10044 size_t sz = count * sizeof(*tx); 10045 10046 if (count < 1 || count > 0xffff) 10047 return -EINVAL; 10048 10049 tx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL); 10050 if (!tx) 10051 return -ENOMEM; 10052 10053 dev->_tx = tx; 10054 10055 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); 10056 spin_lock_init(&dev->tx_global_lock); 10057 10058 return 0; 10059 } 10060 10061 void netif_tx_stop_all_queues(struct net_device *dev) 10062 { 10063 unsigned int i; 10064 10065 for (i = 0; i < dev->num_tx_queues; i++) { 10066 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 10067 10068 netif_tx_stop_queue(txq); 10069 } 10070 } 10071 EXPORT_SYMBOL(netif_tx_stop_all_queues); 10072 10073 /** 10074 * register_netdevice - register a network device 10075 * @dev: device to register 10076 * 10077 * Take a completed network device structure and add it to the kernel 10078 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 10079 * chain. 0 is returned on success. A negative errno code is returned 10080 * on a failure to set up the device, or if the name is a duplicate. 10081 * 10082 * Callers must hold the rtnl semaphore. You may want 10083 * register_netdev() instead of this. 10084 * 10085 * BUGS: 10086 * The locking appears insufficient to guarantee two parallel registers 10087 * will not get the same name. 10088 */ 10089 10090 int register_netdevice(struct net_device *dev) 10091 { 10092 int ret; 10093 struct net *net = dev_net(dev); 10094 10095 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE < 10096 NETDEV_FEATURE_COUNT); 10097 BUG_ON(dev_boot_phase); 10098 ASSERT_RTNL(); 10099 10100 might_sleep(); 10101 10102 /* When net_device's are persistent, this will be fatal. */ 10103 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); 10104 BUG_ON(!net); 10105 10106 ret = ethtool_check_ops(dev->ethtool_ops); 10107 if (ret) 10108 return ret; 10109 10110 spin_lock_init(&dev->addr_list_lock); 10111 netdev_set_addr_lockdep_class(dev); 10112 10113 ret = dev_get_valid_name(net, dev, dev->name); 10114 if (ret < 0) 10115 goto out; 10116 10117 ret = -ENOMEM; 10118 dev->name_node = netdev_name_node_head_alloc(dev); 10119 if (!dev->name_node) 10120 goto out; 10121 10122 /* Init, if this function is available */ 10123 if (dev->netdev_ops->ndo_init) { 10124 ret = dev->netdev_ops->ndo_init(dev); 10125 if (ret) { 10126 if (ret > 0) 10127 ret = -EIO; 10128 goto err_free_name; 10129 } 10130 } 10131 10132 if (((dev->hw_features | dev->features) & 10133 NETIF_F_HW_VLAN_CTAG_FILTER) && 10134 (!dev->netdev_ops->ndo_vlan_rx_add_vid || 10135 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { 10136 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); 10137 ret = -EINVAL; 10138 goto err_uninit; 10139 } 10140 10141 ret = -EBUSY; 10142 if (!dev->ifindex) 10143 dev->ifindex = dev_new_index(net); 10144 else if (__dev_get_by_index(net, dev->ifindex)) 10145 goto err_uninit; 10146 10147 /* Transfer changeable features to wanted_features and enable 10148 * software offloads (GSO and GRO). 10149 */ 10150 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF); 10151 dev->features |= NETIF_F_SOFT_FEATURES; 10152 10153 if (dev->udp_tunnel_nic_info) { 10154 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10155 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10156 } 10157 10158 dev->wanted_features = dev->features & dev->hw_features; 10159 10160 if (!(dev->flags & IFF_LOOPBACK)) 10161 dev->hw_features |= NETIF_F_NOCACHE_COPY; 10162 10163 /* If IPv4 TCP segmentation offload is supported we should also 10164 * allow the device to enable segmenting the frame with the option 10165 * of ignoring a static IP ID value. This doesn't enable the 10166 * feature itself but allows the user to enable it later. 10167 */ 10168 if (dev->hw_features & NETIF_F_TSO) 10169 dev->hw_features |= NETIF_F_TSO_MANGLEID; 10170 if (dev->vlan_features & NETIF_F_TSO) 10171 dev->vlan_features |= NETIF_F_TSO_MANGLEID; 10172 if (dev->mpls_features & NETIF_F_TSO) 10173 dev->mpls_features |= NETIF_F_TSO_MANGLEID; 10174 if (dev->hw_enc_features & NETIF_F_TSO) 10175 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; 10176 10177 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. 10178 */ 10179 dev->vlan_features |= NETIF_F_HIGHDMA; 10180 10181 /* Make NETIF_F_SG inheritable to tunnel devices. 10182 */ 10183 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; 10184 10185 /* Make NETIF_F_SG inheritable to MPLS. 10186 */ 10187 dev->mpls_features |= NETIF_F_SG; 10188 10189 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); 10190 ret = notifier_to_errno(ret); 10191 if (ret) 10192 goto err_uninit; 10193 10194 ret = netdev_register_kobject(dev); 10195 if (ret) { 10196 dev->reg_state = NETREG_UNREGISTERED; 10197 goto err_uninit; 10198 } 10199 dev->reg_state = NETREG_REGISTERED; 10200 10201 __netdev_update_features(dev); 10202 10203 /* 10204 * Default initial state at registry is that the 10205 * device is present. 10206 */ 10207 10208 set_bit(__LINK_STATE_PRESENT, &dev->state); 10209 10210 linkwatch_init_dev(dev); 10211 10212 dev_init_scheduler(dev); 10213 dev_hold(dev); 10214 list_netdevice(dev); 10215 add_device_randomness(dev->dev_addr, dev->addr_len); 10216 10217 /* If the device has permanent device address, driver should 10218 * set dev_addr and also addr_assign_type should be set to 10219 * NET_ADDR_PERM (default value). 10220 */ 10221 if (dev->addr_assign_type == NET_ADDR_PERM) 10222 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); 10223 10224 /* Notify protocols, that a new device appeared. */ 10225 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); 10226 ret = notifier_to_errno(ret); 10227 if (ret) { 10228 /* Expect explicit free_netdev() on failure */ 10229 dev->needs_free_netdev = false; 10230 unregister_netdevice_queue(dev, NULL); 10231 goto out; 10232 } 10233 /* 10234 * Prevent userspace races by waiting until the network 10235 * device is fully setup before sending notifications. 10236 */ 10237 if (!dev->rtnl_link_ops || 10238 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 10239 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); 10240 10241 out: 10242 return ret; 10243 10244 err_uninit: 10245 if (dev->netdev_ops->ndo_uninit) 10246 dev->netdev_ops->ndo_uninit(dev); 10247 if (dev->priv_destructor) 10248 dev->priv_destructor(dev); 10249 err_free_name: 10250 netdev_name_node_free(dev->name_node); 10251 goto out; 10252 } 10253 EXPORT_SYMBOL(register_netdevice); 10254 10255 /** 10256 * init_dummy_netdev - init a dummy network device for NAPI 10257 * @dev: device to init 10258 * 10259 * This takes a network device structure and initialize the minimum 10260 * amount of fields so it can be used to schedule NAPI polls without 10261 * registering a full blown interface. This is to be used by drivers 10262 * that need to tie several hardware interfaces to a single NAPI 10263 * poll scheduler due to HW limitations. 10264 */ 10265 int init_dummy_netdev(struct net_device *dev) 10266 { 10267 /* Clear everything. Note we don't initialize spinlocks 10268 * are they aren't supposed to be taken by any of the 10269 * NAPI code and this dummy netdev is supposed to be 10270 * only ever used for NAPI polls 10271 */ 10272 memset(dev, 0, sizeof(struct net_device)); 10273 10274 /* make sure we BUG if trying to hit standard 10275 * register/unregister code path 10276 */ 10277 dev->reg_state = NETREG_DUMMY; 10278 10279 /* NAPI wants this */ 10280 INIT_LIST_HEAD(&dev->napi_list); 10281 10282 /* a dummy interface is started by default */ 10283 set_bit(__LINK_STATE_PRESENT, &dev->state); 10284 set_bit(__LINK_STATE_START, &dev->state); 10285 10286 /* napi_busy_loop stats accounting wants this */ 10287 dev_net_set(dev, &init_net); 10288 10289 /* Note : We dont allocate pcpu_refcnt for dummy devices, 10290 * because users of this 'device' dont need to change 10291 * its refcount. 10292 */ 10293 10294 return 0; 10295 } 10296 EXPORT_SYMBOL_GPL(init_dummy_netdev); 10297 10298 10299 /** 10300 * register_netdev - register a network device 10301 * @dev: device to register 10302 * 10303 * Take a completed network device structure and add it to the kernel 10304 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 10305 * chain. 0 is returned on success. A negative errno code is returned 10306 * on a failure to set up the device, or if the name is a duplicate. 10307 * 10308 * This is a wrapper around register_netdevice that takes the rtnl semaphore 10309 * and expands the device name if you passed a format string to 10310 * alloc_netdev. 10311 */ 10312 int register_netdev(struct net_device *dev) 10313 { 10314 int err; 10315 10316 if (rtnl_lock_killable()) 10317 return -EINTR; 10318 err = register_netdevice(dev); 10319 rtnl_unlock(); 10320 return err; 10321 } 10322 EXPORT_SYMBOL(register_netdev); 10323 10324 int netdev_refcnt_read(const struct net_device *dev) 10325 { 10326 int i, refcnt = 0; 10327 10328 for_each_possible_cpu(i) 10329 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); 10330 return refcnt; 10331 } 10332 EXPORT_SYMBOL(netdev_refcnt_read); 10333 10334 #define WAIT_REFS_MIN_MSECS 1 10335 #define WAIT_REFS_MAX_MSECS 250 10336 /** 10337 * netdev_wait_allrefs - wait until all references are gone. 10338 * @dev: target net_device 10339 * 10340 * This is called when unregistering network devices. 10341 * 10342 * Any protocol or device that holds a reference should register 10343 * for netdevice notification, and cleanup and put back the 10344 * reference if they receive an UNREGISTER event. 10345 * We can get stuck here if buggy protocols don't correctly 10346 * call dev_put. 10347 */ 10348 static void netdev_wait_allrefs(struct net_device *dev) 10349 { 10350 unsigned long rebroadcast_time, warning_time; 10351 int wait = 0, refcnt; 10352 10353 linkwatch_forget_dev(dev); 10354 10355 rebroadcast_time = warning_time = jiffies; 10356 refcnt = netdev_refcnt_read(dev); 10357 10358 while (refcnt != 0) { 10359 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { 10360 rtnl_lock(); 10361 10362 /* Rebroadcast unregister notification */ 10363 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10364 10365 __rtnl_unlock(); 10366 rcu_barrier(); 10367 rtnl_lock(); 10368 10369 if (test_bit(__LINK_STATE_LINKWATCH_PENDING, 10370 &dev->state)) { 10371 /* We must not have linkwatch events 10372 * pending on unregister. If this 10373 * happens, we simply run the queue 10374 * unscheduled, resulting in a noop 10375 * for this device. 10376 */ 10377 linkwatch_run_queue(); 10378 } 10379 10380 __rtnl_unlock(); 10381 10382 rebroadcast_time = jiffies; 10383 } 10384 10385 if (!wait) { 10386 rcu_barrier(); 10387 wait = WAIT_REFS_MIN_MSECS; 10388 } else { 10389 msleep(wait); 10390 wait = min(wait << 1, WAIT_REFS_MAX_MSECS); 10391 } 10392 10393 refcnt = netdev_refcnt_read(dev); 10394 10395 if (refcnt && time_after(jiffies, warning_time + 10 * HZ)) { 10396 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", 10397 dev->name, refcnt); 10398 warning_time = jiffies; 10399 } 10400 } 10401 } 10402 10403 /* The sequence is: 10404 * 10405 * rtnl_lock(); 10406 * ... 10407 * register_netdevice(x1); 10408 * register_netdevice(x2); 10409 * ... 10410 * unregister_netdevice(y1); 10411 * unregister_netdevice(y2); 10412 * ... 10413 * rtnl_unlock(); 10414 * free_netdev(y1); 10415 * free_netdev(y2); 10416 * 10417 * We are invoked by rtnl_unlock(). 10418 * This allows us to deal with problems: 10419 * 1) We can delete sysfs objects which invoke hotplug 10420 * without deadlocking with linkwatch via keventd. 10421 * 2) Since we run with the RTNL semaphore not held, we can sleep 10422 * safely in order to wait for the netdev refcnt to drop to zero. 10423 * 10424 * We must not return until all unregister events added during 10425 * the interval the lock was held have been completed. 10426 */ 10427 void netdev_run_todo(void) 10428 { 10429 struct list_head list; 10430 #ifdef CONFIG_LOCKDEP 10431 struct list_head unlink_list; 10432 10433 list_replace_init(&net_unlink_list, &unlink_list); 10434 10435 while (!list_empty(&unlink_list)) { 10436 struct net_device *dev = list_first_entry(&unlink_list, 10437 struct net_device, 10438 unlink_list); 10439 list_del_init(&dev->unlink_list); 10440 dev->nested_level = dev->lower_level - 1; 10441 } 10442 #endif 10443 10444 /* Snapshot list, allow later requests */ 10445 list_replace_init(&net_todo_list, &list); 10446 10447 __rtnl_unlock(); 10448 10449 10450 /* Wait for rcu callbacks to finish before next phase */ 10451 if (!list_empty(&list)) 10452 rcu_barrier(); 10453 10454 while (!list_empty(&list)) { 10455 struct net_device *dev 10456 = list_first_entry(&list, struct net_device, todo_list); 10457 list_del(&dev->todo_list); 10458 10459 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { 10460 pr_err("network todo '%s' but state %d\n", 10461 dev->name, dev->reg_state); 10462 dump_stack(); 10463 continue; 10464 } 10465 10466 dev->reg_state = NETREG_UNREGISTERED; 10467 10468 netdev_wait_allrefs(dev); 10469 10470 /* paranoia */ 10471 BUG_ON(netdev_refcnt_read(dev)); 10472 BUG_ON(!list_empty(&dev->ptype_all)); 10473 BUG_ON(!list_empty(&dev->ptype_specific)); 10474 WARN_ON(rcu_access_pointer(dev->ip_ptr)); 10475 WARN_ON(rcu_access_pointer(dev->ip6_ptr)); 10476 #if IS_ENABLED(CONFIG_DECNET) 10477 WARN_ON(dev->dn_ptr); 10478 #endif 10479 if (dev->priv_destructor) 10480 dev->priv_destructor(dev); 10481 if (dev->needs_free_netdev) 10482 free_netdev(dev); 10483 10484 /* Report a network device has been unregistered */ 10485 rtnl_lock(); 10486 dev_net(dev)->dev_unreg_count--; 10487 __rtnl_unlock(); 10488 wake_up(&netdev_unregistering_wq); 10489 10490 /* Free network device */ 10491 kobject_put(&dev->dev.kobj); 10492 } 10493 } 10494 10495 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has 10496 * all the same fields in the same order as net_device_stats, with only 10497 * the type differing, but rtnl_link_stats64 may have additional fields 10498 * at the end for newer counters. 10499 */ 10500 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, 10501 const struct net_device_stats *netdev_stats) 10502 { 10503 #if BITS_PER_LONG == 64 10504 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats)); 10505 memcpy(stats64, netdev_stats, sizeof(*netdev_stats)); 10506 /* zero out counters that only exist in rtnl_link_stats64 */ 10507 memset((char *)stats64 + sizeof(*netdev_stats), 0, 10508 sizeof(*stats64) - sizeof(*netdev_stats)); 10509 #else 10510 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long); 10511 const unsigned long *src = (const unsigned long *)netdev_stats; 10512 u64 *dst = (u64 *)stats64; 10513 10514 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); 10515 for (i = 0; i < n; i++) 10516 dst[i] = src[i]; 10517 /* zero out counters that only exist in rtnl_link_stats64 */ 10518 memset((char *)stats64 + n * sizeof(u64), 0, 10519 sizeof(*stats64) - n * sizeof(u64)); 10520 #endif 10521 } 10522 EXPORT_SYMBOL(netdev_stats_to_stats64); 10523 10524 /** 10525 * dev_get_stats - get network device statistics 10526 * @dev: device to get statistics from 10527 * @storage: place to store stats 10528 * 10529 * Get network statistics from device. Return @storage. 10530 * The device driver may provide its own method by setting 10531 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; 10532 * otherwise the internal statistics structure is used. 10533 */ 10534 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, 10535 struct rtnl_link_stats64 *storage) 10536 { 10537 const struct net_device_ops *ops = dev->netdev_ops; 10538 10539 if (ops->ndo_get_stats64) { 10540 memset(storage, 0, sizeof(*storage)); 10541 ops->ndo_get_stats64(dev, storage); 10542 } else if (ops->ndo_get_stats) { 10543 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); 10544 } else { 10545 netdev_stats_to_stats64(storage, &dev->stats); 10546 } 10547 storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped); 10548 storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped); 10549 storage->rx_nohandler += (unsigned long)atomic_long_read(&dev->rx_nohandler); 10550 return storage; 10551 } 10552 EXPORT_SYMBOL(dev_get_stats); 10553 10554 /** 10555 * dev_fetch_sw_netstats - get per-cpu network device statistics 10556 * @s: place to store stats 10557 * @netstats: per-cpu network stats to read from 10558 * 10559 * Read per-cpu network statistics and populate the related fields in @s. 10560 */ 10561 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s, 10562 const struct pcpu_sw_netstats __percpu *netstats) 10563 { 10564 int cpu; 10565 10566 for_each_possible_cpu(cpu) { 10567 const struct pcpu_sw_netstats *stats; 10568 struct pcpu_sw_netstats tmp; 10569 unsigned int start; 10570 10571 stats = per_cpu_ptr(netstats, cpu); 10572 do { 10573 start = u64_stats_fetch_begin_irq(&stats->syncp); 10574 tmp.rx_packets = stats->rx_packets; 10575 tmp.rx_bytes = stats->rx_bytes; 10576 tmp.tx_packets = stats->tx_packets; 10577 tmp.tx_bytes = stats->tx_bytes; 10578 } while (u64_stats_fetch_retry_irq(&stats->syncp, start)); 10579 10580 s->rx_packets += tmp.rx_packets; 10581 s->rx_bytes += tmp.rx_bytes; 10582 s->tx_packets += tmp.tx_packets; 10583 s->tx_bytes += tmp.tx_bytes; 10584 } 10585 } 10586 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats); 10587 10588 /** 10589 * dev_get_tstats64 - ndo_get_stats64 implementation 10590 * @dev: device to get statistics from 10591 * @s: place to store stats 10592 * 10593 * Populate @s from dev->stats and dev->tstats. Can be used as 10594 * ndo_get_stats64() callback. 10595 */ 10596 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s) 10597 { 10598 netdev_stats_to_stats64(s, &dev->stats); 10599 dev_fetch_sw_netstats(s, dev->tstats); 10600 } 10601 EXPORT_SYMBOL_GPL(dev_get_tstats64); 10602 10603 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) 10604 { 10605 struct netdev_queue *queue = dev_ingress_queue(dev); 10606 10607 #ifdef CONFIG_NET_CLS_ACT 10608 if (queue) 10609 return queue; 10610 queue = kzalloc(sizeof(*queue), GFP_KERNEL); 10611 if (!queue) 10612 return NULL; 10613 netdev_init_one_queue(dev, queue, NULL); 10614 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); 10615 queue->qdisc_sleeping = &noop_qdisc; 10616 rcu_assign_pointer(dev->ingress_queue, queue); 10617 #endif 10618 return queue; 10619 } 10620 10621 static const struct ethtool_ops default_ethtool_ops; 10622 10623 void netdev_set_default_ethtool_ops(struct net_device *dev, 10624 const struct ethtool_ops *ops) 10625 { 10626 if (dev->ethtool_ops == &default_ethtool_ops) 10627 dev->ethtool_ops = ops; 10628 } 10629 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); 10630 10631 void netdev_freemem(struct net_device *dev) 10632 { 10633 char *addr = (char *)dev - dev->padded; 10634 10635 kvfree(addr); 10636 } 10637 10638 /** 10639 * alloc_netdev_mqs - allocate network device 10640 * @sizeof_priv: size of private data to allocate space for 10641 * @name: device name format string 10642 * @name_assign_type: origin of device name 10643 * @setup: callback to initialize device 10644 * @txqs: the number of TX subqueues to allocate 10645 * @rxqs: the number of RX subqueues to allocate 10646 * 10647 * Allocates a struct net_device with private data area for driver use 10648 * and performs basic initialization. Also allocates subqueue structs 10649 * for each queue on the device. 10650 */ 10651 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, 10652 unsigned char name_assign_type, 10653 void (*setup)(struct net_device *), 10654 unsigned int txqs, unsigned int rxqs) 10655 { 10656 struct net_device *dev; 10657 unsigned int alloc_size; 10658 struct net_device *p; 10659 10660 BUG_ON(strlen(name) >= sizeof(dev->name)); 10661 10662 if (txqs < 1) { 10663 pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); 10664 return NULL; 10665 } 10666 10667 if (rxqs < 1) { 10668 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); 10669 return NULL; 10670 } 10671 10672 alloc_size = sizeof(struct net_device); 10673 if (sizeof_priv) { 10674 /* ensure 32-byte alignment of private area */ 10675 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); 10676 alloc_size += sizeof_priv; 10677 } 10678 /* ensure 32-byte alignment of whole construct */ 10679 alloc_size += NETDEV_ALIGN - 1; 10680 10681 p = kvzalloc(alloc_size, GFP_KERNEL | __GFP_RETRY_MAYFAIL); 10682 if (!p) 10683 return NULL; 10684 10685 dev = PTR_ALIGN(p, NETDEV_ALIGN); 10686 dev->padded = (char *)dev - (char *)p; 10687 10688 dev->pcpu_refcnt = alloc_percpu(int); 10689 if (!dev->pcpu_refcnt) 10690 goto free_dev; 10691 10692 if (dev_addr_init(dev)) 10693 goto free_pcpu; 10694 10695 dev_mc_init(dev); 10696 dev_uc_init(dev); 10697 10698 dev_net_set(dev, &init_net); 10699 10700 dev->gso_max_size = GSO_MAX_SIZE; 10701 dev->gso_max_segs = GSO_MAX_SEGS; 10702 dev->upper_level = 1; 10703 dev->lower_level = 1; 10704 #ifdef CONFIG_LOCKDEP 10705 dev->nested_level = 0; 10706 INIT_LIST_HEAD(&dev->unlink_list); 10707 #endif 10708 10709 INIT_LIST_HEAD(&dev->napi_list); 10710 INIT_LIST_HEAD(&dev->unreg_list); 10711 INIT_LIST_HEAD(&dev->close_list); 10712 INIT_LIST_HEAD(&dev->link_watch_list); 10713 INIT_LIST_HEAD(&dev->adj_list.upper); 10714 INIT_LIST_HEAD(&dev->adj_list.lower); 10715 INIT_LIST_HEAD(&dev->ptype_all); 10716 INIT_LIST_HEAD(&dev->ptype_specific); 10717 INIT_LIST_HEAD(&dev->net_notifier_list); 10718 #ifdef CONFIG_NET_SCHED 10719 hash_init(dev->qdisc_hash); 10720 #endif 10721 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; 10722 setup(dev); 10723 10724 if (!dev->tx_queue_len) { 10725 dev->priv_flags |= IFF_NO_QUEUE; 10726 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; 10727 } 10728 10729 dev->num_tx_queues = txqs; 10730 dev->real_num_tx_queues = txqs; 10731 if (netif_alloc_netdev_queues(dev)) 10732 goto free_all; 10733 10734 dev->num_rx_queues = rxqs; 10735 dev->real_num_rx_queues = rxqs; 10736 if (netif_alloc_rx_queues(dev)) 10737 goto free_all; 10738 10739 strcpy(dev->name, name); 10740 dev->name_assign_type = name_assign_type; 10741 dev->group = INIT_NETDEV_GROUP; 10742 if (!dev->ethtool_ops) 10743 dev->ethtool_ops = &default_ethtool_ops; 10744 10745 nf_hook_ingress_init(dev); 10746 10747 return dev; 10748 10749 free_all: 10750 free_netdev(dev); 10751 return NULL; 10752 10753 free_pcpu: 10754 free_percpu(dev->pcpu_refcnt); 10755 free_dev: 10756 netdev_freemem(dev); 10757 return NULL; 10758 } 10759 EXPORT_SYMBOL(alloc_netdev_mqs); 10760 10761 /** 10762 * free_netdev - free network device 10763 * @dev: device 10764 * 10765 * This function does the last stage of destroying an allocated device 10766 * interface. The reference to the device object is released. If this 10767 * is the last reference then it will be freed.Must be called in process 10768 * context. 10769 */ 10770 void free_netdev(struct net_device *dev) 10771 { 10772 struct napi_struct *p, *n; 10773 10774 might_sleep(); 10775 10776 /* When called immediately after register_netdevice() failed the unwind 10777 * handling may still be dismantling the device. Handle that case by 10778 * deferring the free. 10779 */ 10780 if (dev->reg_state == NETREG_UNREGISTERING) { 10781 ASSERT_RTNL(); 10782 dev->needs_free_netdev = true; 10783 return; 10784 } 10785 10786 netif_free_tx_queues(dev); 10787 netif_free_rx_queues(dev); 10788 10789 kfree(rcu_dereference_protected(dev->ingress_queue, 1)); 10790 10791 /* Flush device addresses */ 10792 dev_addr_flush(dev); 10793 10794 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) 10795 netif_napi_del(p); 10796 10797 free_percpu(dev->pcpu_refcnt); 10798 dev->pcpu_refcnt = NULL; 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