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 <trace/events/qdisc.h> 135 #include <linux/inetdevice.h> 136 #include <linux/cpu_rmap.h> 137 #include <linux/static_key.h> 138 #include <linux/hashtable.h> 139 #include <linux/vmalloc.h> 140 #include <linux/if_macvlan.h> 141 #include <linux/errqueue.h> 142 #include <linux/hrtimer.h> 143 #include <linux/netfilter_ingress.h> 144 #include <linux/crash_dump.h> 145 #include <linux/sctp.h> 146 #include <net/udp_tunnel.h> 147 #include <linux/net_namespace.h> 148 #include <linux/indirect_call_wrapper.h> 149 #include <net/devlink.h> 150 #include <linux/pm_runtime.h> 151 #include <linux/prandom.h> 152 #include <linux/once_lite.h> 153 154 #include "net-sysfs.h" 155 156 #define MAX_GRO_SKBS 8 157 158 /* This should be increased if a protocol with a bigger head is added. */ 159 #define GRO_MAX_HEAD (MAX_HEADER + 128) 160 161 static DEFINE_SPINLOCK(ptype_lock); 162 static DEFINE_SPINLOCK(offload_lock); 163 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; 164 struct list_head ptype_all __read_mostly; /* Taps */ 165 static struct list_head offload_base __read_mostly; 166 167 static int netif_rx_internal(struct sk_buff *skb); 168 static int call_netdevice_notifiers_info(unsigned long val, 169 struct netdev_notifier_info *info); 170 static int call_netdevice_notifiers_extack(unsigned long val, 171 struct net_device *dev, 172 struct netlink_ext_ack *extack); 173 static struct napi_struct *napi_by_id(unsigned int napi_id); 174 175 /* 176 * The @dev_base_head list is protected by @dev_base_lock and the rtnl 177 * semaphore. 178 * 179 * Pure readers hold dev_base_lock for reading, or rcu_read_lock() 180 * 181 * Writers must hold the rtnl semaphore while they loop through the 182 * dev_base_head list, and hold dev_base_lock for writing when they do the 183 * actual updates. This allows pure readers to access the list even 184 * while a writer is preparing to update it. 185 * 186 * To put it another way, dev_base_lock is held for writing only to 187 * protect against pure readers; the rtnl semaphore provides the 188 * protection against other writers. 189 * 190 * See, for example usages, register_netdevice() and 191 * unregister_netdevice(), which must be called with the rtnl 192 * semaphore held. 193 */ 194 DEFINE_RWLOCK(dev_base_lock); 195 EXPORT_SYMBOL(dev_base_lock); 196 197 static DEFINE_MUTEX(ifalias_mutex); 198 199 /* protects napi_hash addition/deletion and napi_gen_id */ 200 static DEFINE_SPINLOCK(napi_hash_lock); 201 202 static unsigned int napi_gen_id = NR_CPUS; 203 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8); 204 205 static DECLARE_RWSEM(devnet_rename_sem); 206 207 static inline void dev_base_seq_inc(struct net *net) 208 { 209 while (++net->dev_base_seq == 0) 210 ; 211 } 212 213 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) 214 { 215 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ)); 216 217 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; 218 } 219 220 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) 221 { 222 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; 223 } 224 225 static inline void rps_lock(struct softnet_data *sd) 226 { 227 #ifdef CONFIG_RPS 228 spin_lock(&sd->input_pkt_queue.lock); 229 #endif 230 } 231 232 static inline void rps_unlock(struct softnet_data *sd) 233 { 234 #ifdef CONFIG_RPS 235 spin_unlock(&sd->input_pkt_queue.lock); 236 #endif 237 } 238 239 static struct netdev_name_node *netdev_name_node_alloc(struct net_device *dev, 240 const char *name) 241 { 242 struct netdev_name_node *name_node; 243 244 name_node = kmalloc(sizeof(*name_node), GFP_KERNEL); 245 if (!name_node) 246 return NULL; 247 INIT_HLIST_NODE(&name_node->hlist); 248 name_node->dev = dev; 249 name_node->name = name; 250 return name_node; 251 } 252 253 static struct netdev_name_node * 254 netdev_name_node_head_alloc(struct net_device *dev) 255 { 256 struct netdev_name_node *name_node; 257 258 name_node = netdev_name_node_alloc(dev, dev->name); 259 if (!name_node) 260 return NULL; 261 INIT_LIST_HEAD(&name_node->list); 262 return name_node; 263 } 264 265 static void netdev_name_node_free(struct netdev_name_node *name_node) 266 { 267 kfree(name_node); 268 } 269 270 static void netdev_name_node_add(struct net *net, 271 struct netdev_name_node *name_node) 272 { 273 hlist_add_head_rcu(&name_node->hlist, 274 dev_name_hash(net, name_node->name)); 275 } 276 277 static void netdev_name_node_del(struct netdev_name_node *name_node) 278 { 279 hlist_del_rcu(&name_node->hlist); 280 } 281 282 static struct netdev_name_node *netdev_name_node_lookup(struct net *net, 283 const char *name) 284 { 285 struct hlist_head *head = dev_name_hash(net, name); 286 struct netdev_name_node *name_node; 287 288 hlist_for_each_entry(name_node, head, hlist) 289 if (!strcmp(name_node->name, name)) 290 return name_node; 291 return NULL; 292 } 293 294 static struct netdev_name_node *netdev_name_node_lookup_rcu(struct net *net, 295 const char *name) 296 { 297 struct hlist_head *head = dev_name_hash(net, name); 298 struct netdev_name_node *name_node; 299 300 hlist_for_each_entry_rcu(name_node, head, hlist) 301 if (!strcmp(name_node->name, name)) 302 return name_node; 303 return NULL; 304 } 305 306 int netdev_name_node_alt_create(struct net_device *dev, const char *name) 307 { 308 struct netdev_name_node *name_node; 309 struct net *net = dev_net(dev); 310 311 name_node = netdev_name_node_lookup(net, name); 312 if (name_node) 313 return -EEXIST; 314 name_node = netdev_name_node_alloc(dev, name); 315 if (!name_node) 316 return -ENOMEM; 317 netdev_name_node_add(net, name_node); 318 /* The node that holds dev->name acts as a head of per-device list. */ 319 list_add_tail(&name_node->list, &dev->name_node->list); 320 321 return 0; 322 } 323 EXPORT_SYMBOL(netdev_name_node_alt_create); 324 325 static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node) 326 { 327 list_del(&name_node->list); 328 netdev_name_node_del(name_node); 329 kfree(name_node->name); 330 netdev_name_node_free(name_node); 331 } 332 333 int netdev_name_node_alt_destroy(struct net_device *dev, const char *name) 334 { 335 struct netdev_name_node *name_node; 336 struct net *net = dev_net(dev); 337 338 name_node = netdev_name_node_lookup(net, name); 339 if (!name_node) 340 return -ENOENT; 341 /* lookup might have found our primary name or a name belonging 342 * to another device. 343 */ 344 if (name_node == dev->name_node || name_node->dev != dev) 345 return -EINVAL; 346 347 __netdev_name_node_alt_destroy(name_node); 348 349 return 0; 350 } 351 EXPORT_SYMBOL(netdev_name_node_alt_destroy); 352 353 static void netdev_name_node_alt_flush(struct net_device *dev) 354 { 355 struct netdev_name_node *name_node, *tmp; 356 357 list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list) 358 __netdev_name_node_alt_destroy(name_node); 359 } 360 361 /* Device list insertion */ 362 static void list_netdevice(struct net_device *dev) 363 { 364 struct net *net = dev_net(dev); 365 366 ASSERT_RTNL(); 367 368 write_lock_bh(&dev_base_lock); 369 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); 370 netdev_name_node_add(net, dev->name_node); 371 hlist_add_head_rcu(&dev->index_hlist, 372 dev_index_hash(net, dev->ifindex)); 373 write_unlock_bh(&dev_base_lock); 374 375 dev_base_seq_inc(net); 376 } 377 378 /* Device list removal 379 * caller must respect a RCU grace period before freeing/reusing dev 380 */ 381 static void unlist_netdevice(struct net_device *dev) 382 { 383 ASSERT_RTNL(); 384 385 /* Unlink dev from the device chain */ 386 write_lock_bh(&dev_base_lock); 387 list_del_rcu(&dev->dev_list); 388 netdev_name_node_del(dev->name_node); 389 hlist_del_rcu(&dev->index_hlist); 390 write_unlock_bh(&dev_base_lock); 391 392 dev_base_seq_inc(dev_net(dev)); 393 } 394 395 /* 396 * Our notifier list 397 */ 398 399 static RAW_NOTIFIER_HEAD(netdev_chain); 400 401 /* 402 * Device drivers call our routines to queue packets here. We empty the 403 * queue in the local softnet handler. 404 */ 405 406 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); 407 EXPORT_PER_CPU_SYMBOL(softnet_data); 408 409 #ifdef CONFIG_LOCKDEP 410 /* 411 * register_netdevice() inits txq->_xmit_lock and sets lockdep class 412 * according to dev->type 413 */ 414 static const unsigned short netdev_lock_type[] = { 415 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, 416 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, 417 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, 418 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, 419 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, 420 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, 421 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, 422 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, 423 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, 424 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, 425 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, 426 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, 427 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, 428 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, 429 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; 430 431 static const char *const netdev_lock_name[] = { 432 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", 433 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", 434 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", 435 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", 436 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", 437 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", 438 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", 439 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", 440 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", 441 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", 442 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", 443 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", 444 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", 445 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", 446 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; 447 448 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; 449 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; 450 451 static inline unsigned short netdev_lock_pos(unsigned short dev_type) 452 { 453 int i; 454 455 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) 456 if (netdev_lock_type[i] == dev_type) 457 return i; 458 /* the last key is used by default */ 459 return ARRAY_SIZE(netdev_lock_type) - 1; 460 } 461 462 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 463 unsigned short dev_type) 464 { 465 int i; 466 467 i = netdev_lock_pos(dev_type); 468 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], 469 netdev_lock_name[i]); 470 } 471 472 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 473 { 474 int i; 475 476 i = netdev_lock_pos(dev->type); 477 lockdep_set_class_and_name(&dev->addr_list_lock, 478 &netdev_addr_lock_key[i], 479 netdev_lock_name[i]); 480 } 481 #else 482 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 483 unsigned short dev_type) 484 { 485 } 486 487 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 488 { 489 } 490 #endif 491 492 /******************************************************************************* 493 * 494 * Protocol management and registration routines 495 * 496 *******************************************************************************/ 497 498 499 /* 500 * Add a protocol ID to the list. Now that the input handler is 501 * smarter we can dispense with all the messy stuff that used to be 502 * here. 503 * 504 * BEWARE!!! Protocol handlers, mangling input packets, 505 * MUST BE last in hash buckets and checking protocol handlers 506 * MUST start from promiscuous ptype_all chain in net_bh. 507 * It is true now, do not change it. 508 * Explanation follows: if protocol handler, mangling packet, will 509 * be the first on list, it is not able to sense, that packet 510 * is cloned and should be copied-on-write, so that it will 511 * change it and subsequent readers will get broken packet. 512 * --ANK (980803) 513 */ 514 515 static inline struct list_head *ptype_head(const struct packet_type *pt) 516 { 517 if (pt->type == htons(ETH_P_ALL)) 518 return pt->dev ? &pt->dev->ptype_all : &ptype_all; 519 else 520 return pt->dev ? &pt->dev->ptype_specific : 521 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; 522 } 523 524 /** 525 * dev_add_pack - add packet handler 526 * @pt: packet type declaration 527 * 528 * Add a protocol handler to the networking stack. The passed &packet_type 529 * is linked into kernel lists and may not be freed until it has been 530 * removed from the kernel lists. 531 * 532 * This call does not sleep therefore it can not 533 * guarantee all CPU's that are in middle of receiving packets 534 * will see the new packet type (until the next received packet). 535 */ 536 537 void dev_add_pack(struct packet_type *pt) 538 { 539 struct list_head *head = ptype_head(pt); 540 541 spin_lock(&ptype_lock); 542 list_add_rcu(&pt->list, head); 543 spin_unlock(&ptype_lock); 544 } 545 EXPORT_SYMBOL(dev_add_pack); 546 547 /** 548 * __dev_remove_pack - remove packet handler 549 * @pt: packet type declaration 550 * 551 * Remove a protocol handler that was previously added to the kernel 552 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 553 * from the kernel lists and can be freed or reused once this function 554 * returns. 555 * 556 * The packet type might still be in use by receivers 557 * and must not be freed until after all the CPU's have gone 558 * through a quiescent state. 559 */ 560 void __dev_remove_pack(struct packet_type *pt) 561 { 562 struct list_head *head = ptype_head(pt); 563 struct packet_type *pt1; 564 565 spin_lock(&ptype_lock); 566 567 list_for_each_entry(pt1, head, list) { 568 if (pt == pt1) { 569 list_del_rcu(&pt->list); 570 goto out; 571 } 572 } 573 574 pr_warn("dev_remove_pack: %p not found\n", pt); 575 out: 576 spin_unlock(&ptype_lock); 577 } 578 EXPORT_SYMBOL(__dev_remove_pack); 579 580 /** 581 * dev_remove_pack - remove packet handler 582 * @pt: packet type declaration 583 * 584 * Remove a protocol handler that was previously added to the kernel 585 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 586 * from the kernel lists and can be freed or reused once this function 587 * returns. 588 * 589 * This call sleeps to guarantee that no CPU is looking at the packet 590 * type after return. 591 */ 592 void dev_remove_pack(struct packet_type *pt) 593 { 594 __dev_remove_pack(pt); 595 596 synchronize_net(); 597 } 598 EXPORT_SYMBOL(dev_remove_pack); 599 600 601 /** 602 * dev_add_offload - register offload handlers 603 * @po: protocol offload declaration 604 * 605 * Add protocol offload handlers to the networking stack. The passed 606 * &proto_offload is linked into kernel lists and may not be freed until 607 * it has been removed from the kernel lists. 608 * 609 * This call does not sleep therefore it can not 610 * guarantee all CPU's that are in middle of receiving packets 611 * will see the new offload handlers (until the next received packet). 612 */ 613 void dev_add_offload(struct packet_offload *po) 614 { 615 struct packet_offload *elem; 616 617 spin_lock(&offload_lock); 618 list_for_each_entry(elem, &offload_base, list) { 619 if (po->priority < elem->priority) 620 break; 621 } 622 list_add_rcu(&po->list, elem->list.prev); 623 spin_unlock(&offload_lock); 624 } 625 EXPORT_SYMBOL(dev_add_offload); 626 627 /** 628 * __dev_remove_offload - remove offload handler 629 * @po: packet offload declaration 630 * 631 * Remove a protocol offload handler that was previously added to the 632 * kernel offload handlers by dev_add_offload(). The passed &offload_type 633 * is removed from the kernel lists and can be freed or reused once this 634 * function returns. 635 * 636 * The packet type might still be in use by receivers 637 * and must not be freed until after all the CPU's have gone 638 * through a quiescent state. 639 */ 640 static void __dev_remove_offload(struct packet_offload *po) 641 { 642 struct list_head *head = &offload_base; 643 struct packet_offload *po1; 644 645 spin_lock(&offload_lock); 646 647 list_for_each_entry(po1, head, list) { 648 if (po == po1) { 649 list_del_rcu(&po->list); 650 goto out; 651 } 652 } 653 654 pr_warn("dev_remove_offload: %p not found\n", po); 655 out: 656 spin_unlock(&offload_lock); 657 } 658 659 /** 660 * dev_remove_offload - remove packet offload handler 661 * @po: packet offload declaration 662 * 663 * Remove a packet offload handler that was previously added to the kernel 664 * offload handlers by dev_add_offload(). The passed &offload_type is 665 * removed from the kernel lists and can be freed or reused once this 666 * function returns. 667 * 668 * This call sleeps to guarantee that no CPU is looking at the packet 669 * type after return. 670 */ 671 void dev_remove_offload(struct packet_offload *po) 672 { 673 __dev_remove_offload(po); 674 675 synchronize_net(); 676 } 677 EXPORT_SYMBOL(dev_remove_offload); 678 679 /******************************************************************************* 680 * 681 * Device Interface Subroutines 682 * 683 *******************************************************************************/ 684 685 /** 686 * dev_get_iflink - get 'iflink' value of a interface 687 * @dev: targeted interface 688 * 689 * Indicates the ifindex the interface is linked to. 690 * Physical interfaces have the same 'ifindex' and 'iflink' values. 691 */ 692 693 int dev_get_iflink(const struct net_device *dev) 694 { 695 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink) 696 return dev->netdev_ops->ndo_get_iflink(dev); 697 698 return dev->ifindex; 699 } 700 EXPORT_SYMBOL(dev_get_iflink); 701 702 /** 703 * dev_fill_metadata_dst - Retrieve tunnel egress information. 704 * @dev: targeted interface 705 * @skb: The packet. 706 * 707 * For better visibility of tunnel traffic OVS needs to retrieve 708 * egress tunnel information for a packet. Following API allows 709 * user to get this info. 710 */ 711 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) 712 { 713 struct ip_tunnel_info *info; 714 715 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst) 716 return -EINVAL; 717 718 info = skb_tunnel_info_unclone(skb); 719 if (!info) 720 return -ENOMEM; 721 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX))) 722 return -EINVAL; 723 724 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb); 725 } 726 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst); 727 728 static struct net_device_path *dev_fwd_path(struct net_device_path_stack *stack) 729 { 730 int k = stack->num_paths++; 731 732 if (WARN_ON_ONCE(k >= NET_DEVICE_PATH_STACK_MAX)) 733 return NULL; 734 735 return &stack->path[k]; 736 } 737 738 int dev_fill_forward_path(const struct net_device *dev, const u8 *daddr, 739 struct net_device_path_stack *stack) 740 { 741 const struct net_device *last_dev; 742 struct net_device_path_ctx ctx = { 743 .dev = dev, 744 .daddr = daddr, 745 }; 746 struct net_device_path *path; 747 int ret = 0; 748 749 stack->num_paths = 0; 750 while (ctx.dev && ctx.dev->netdev_ops->ndo_fill_forward_path) { 751 last_dev = ctx.dev; 752 path = dev_fwd_path(stack); 753 if (!path) 754 return -1; 755 756 memset(path, 0, sizeof(struct net_device_path)); 757 ret = ctx.dev->netdev_ops->ndo_fill_forward_path(&ctx, path); 758 if (ret < 0) 759 return -1; 760 761 if (WARN_ON_ONCE(last_dev == ctx.dev)) 762 return -1; 763 } 764 path = dev_fwd_path(stack); 765 if (!path) 766 return -1; 767 path->type = DEV_PATH_ETHERNET; 768 path->dev = ctx.dev; 769 770 return ret; 771 } 772 EXPORT_SYMBOL_GPL(dev_fill_forward_path); 773 774 /** 775 * __dev_get_by_name - find a device by its name 776 * @net: the applicable net namespace 777 * @name: name to find 778 * 779 * Find an interface by name. Must be called under RTNL semaphore 780 * or @dev_base_lock. If the name is found a pointer to the device 781 * is returned. If the name is not found then %NULL is returned. The 782 * reference counters are not incremented so the caller must be 783 * careful with locks. 784 */ 785 786 struct net_device *__dev_get_by_name(struct net *net, const char *name) 787 { 788 struct netdev_name_node *node_name; 789 790 node_name = netdev_name_node_lookup(net, name); 791 return node_name ? node_name->dev : NULL; 792 } 793 EXPORT_SYMBOL(__dev_get_by_name); 794 795 /** 796 * dev_get_by_name_rcu - find a device by its name 797 * @net: the applicable net namespace 798 * @name: name to find 799 * 800 * Find an interface by name. 801 * If the name is found a pointer to the device is returned. 802 * If the name is not found then %NULL is returned. 803 * The reference counters are not incremented so the caller must be 804 * careful with locks. The caller must hold RCU lock. 805 */ 806 807 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) 808 { 809 struct netdev_name_node *node_name; 810 811 node_name = netdev_name_node_lookup_rcu(net, name); 812 return node_name ? node_name->dev : NULL; 813 } 814 EXPORT_SYMBOL(dev_get_by_name_rcu); 815 816 /** 817 * dev_get_by_name - find a device by its name 818 * @net: the applicable net namespace 819 * @name: name to find 820 * 821 * Find an interface by name. This can be called from any 822 * context and does its own locking. The returned handle has 823 * the usage count incremented and the caller must use dev_put() to 824 * release it when it is no longer needed. %NULL is returned if no 825 * matching device is found. 826 */ 827 828 struct net_device *dev_get_by_name(struct net *net, const char *name) 829 { 830 struct net_device *dev; 831 832 rcu_read_lock(); 833 dev = dev_get_by_name_rcu(net, name); 834 if (dev) 835 dev_hold(dev); 836 rcu_read_unlock(); 837 return dev; 838 } 839 EXPORT_SYMBOL(dev_get_by_name); 840 841 /** 842 * __dev_get_by_index - find a device by its ifindex 843 * @net: the applicable net namespace 844 * @ifindex: index of device 845 * 846 * Search for an interface by index. Returns %NULL if the device 847 * is not found or a pointer to the device. The device has not 848 * had its reference counter increased so the caller must be careful 849 * about locking. The caller must hold either the RTNL semaphore 850 * or @dev_base_lock. 851 */ 852 853 struct net_device *__dev_get_by_index(struct net *net, int ifindex) 854 { 855 struct net_device *dev; 856 struct hlist_head *head = dev_index_hash(net, ifindex); 857 858 hlist_for_each_entry(dev, head, index_hlist) 859 if (dev->ifindex == ifindex) 860 return dev; 861 862 return NULL; 863 } 864 EXPORT_SYMBOL(__dev_get_by_index); 865 866 /** 867 * dev_get_by_index_rcu - find a device by its ifindex 868 * @net: the applicable net namespace 869 * @ifindex: index of device 870 * 871 * Search for an interface by index. Returns %NULL if the device 872 * is not found or a pointer to the device. The device has not 873 * had its reference counter increased so the caller must be careful 874 * about locking. The caller must hold RCU lock. 875 */ 876 877 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) 878 { 879 struct net_device *dev; 880 struct hlist_head *head = dev_index_hash(net, ifindex); 881 882 hlist_for_each_entry_rcu(dev, head, index_hlist) 883 if (dev->ifindex == ifindex) 884 return dev; 885 886 return NULL; 887 } 888 EXPORT_SYMBOL(dev_get_by_index_rcu); 889 890 891 /** 892 * dev_get_by_index - find a device by its ifindex 893 * @net: the applicable net namespace 894 * @ifindex: index of device 895 * 896 * Search for an interface by index. Returns NULL if the device 897 * is not found or a pointer to the device. The device returned has 898 * had a reference added and the pointer is safe until the user calls 899 * dev_put to indicate they have finished with it. 900 */ 901 902 struct net_device *dev_get_by_index(struct net *net, int ifindex) 903 { 904 struct net_device *dev; 905 906 rcu_read_lock(); 907 dev = dev_get_by_index_rcu(net, ifindex); 908 if (dev) 909 dev_hold(dev); 910 rcu_read_unlock(); 911 return dev; 912 } 913 EXPORT_SYMBOL(dev_get_by_index); 914 915 /** 916 * dev_get_by_napi_id - find a device by napi_id 917 * @napi_id: ID of the NAPI struct 918 * 919 * Search for an interface by NAPI ID. Returns %NULL if the device 920 * is not found or a pointer to the device. The device has not had 921 * its reference counter increased so the caller must be careful 922 * about locking. The caller must hold RCU lock. 923 */ 924 925 struct net_device *dev_get_by_napi_id(unsigned int napi_id) 926 { 927 struct napi_struct *napi; 928 929 WARN_ON_ONCE(!rcu_read_lock_held()); 930 931 if (napi_id < MIN_NAPI_ID) 932 return NULL; 933 934 napi = napi_by_id(napi_id); 935 936 return napi ? napi->dev : NULL; 937 } 938 EXPORT_SYMBOL(dev_get_by_napi_id); 939 940 /** 941 * netdev_get_name - get a netdevice name, knowing its ifindex. 942 * @net: network namespace 943 * @name: a pointer to the buffer where the name will be stored. 944 * @ifindex: the ifindex of the interface to get the name from. 945 */ 946 int netdev_get_name(struct net *net, char *name, int ifindex) 947 { 948 struct net_device *dev; 949 int ret; 950 951 down_read(&devnet_rename_sem); 952 rcu_read_lock(); 953 954 dev = dev_get_by_index_rcu(net, ifindex); 955 if (!dev) { 956 ret = -ENODEV; 957 goto out; 958 } 959 960 strcpy(name, dev->name); 961 962 ret = 0; 963 out: 964 rcu_read_unlock(); 965 up_read(&devnet_rename_sem); 966 return ret; 967 } 968 969 /** 970 * dev_getbyhwaddr_rcu - find a device by its hardware address 971 * @net: the applicable net namespace 972 * @type: media type of device 973 * @ha: hardware address 974 * 975 * Search for an interface by MAC address. Returns NULL if the device 976 * is not found or a pointer to the device. 977 * The caller must hold RCU or RTNL. 978 * The returned device has not had its ref count increased 979 * and the caller must therefore be careful about locking 980 * 981 */ 982 983 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, 984 const char *ha) 985 { 986 struct net_device *dev; 987 988 for_each_netdev_rcu(net, dev) 989 if (dev->type == type && 990 !memcmp(dev->dev_addr, ha, dev->addr_len)) 991 return dev; 992 993 return NULL; 994 } 995 EXPORT_SYMBOL(dev_getbyhwaddr_rcu); 996 997 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) 998 { 999 struct net_device *dev, *ret = NULL; 1000 1001 rcu_read_lock(); 1002 for_each_netdev_rcu(net, dev) 1003 if (dev->type == type) { 1004 dev_hold(dev); 1005 ret = dev; 1006 break; 1007 } 1008 rcu_read_unlock(); 1009 return ret; 1010 } 1011 EXPORT_SYMBOL(dev_getfirstbyhwtype); 1012 1013 /** 1014 * __dev_get_by_flags - find any device with given flags 1015 * @net: the applicable net namespace 1016 * @if_flags: IFF_* values 1017 * @mask: bitmask of bits in if_flags to check 1018 * 1019 * Search for any interface with the given flags. Returns NULL if a device 1020 * is not found or a pointer to the device. Must be called inside 1021 * rtnl_lock(), and result refcount is unchanged. 1022 */ 1023 1024 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags, 1025 unsigned short mask) 1026 { 1027 struct net_device *dev, *ret; 1028 1029 ASSERT_RTNL(); 1030 1031 ret = NULL; 1032 for_each_netdev(net, dev) { 1033 if (((dev->flags ^ if_flags) & mask) == 0) { 1034 ret = dev; 1035 break; 1036 } 1037 } 1038 return ret; 1039 } 1040 EXPORT_SYMBOL(__dev_get_by_flags); 1041 1042 /** 1043 * dev_valid_name - check if name is okay for network device 1044 * @name: name string 1045 * 1046 * Network device names need to be valid file names to 1047 * allow sysfs to work. We also disallow any kind of 1048 * whitespace. 1049 */ 1050 bool dev_valid_name(const char *name) 1051 { 1052 if (*name == '\0') 1053 return false; 1054 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ) 1055 return false; 1056 if (!strcmp(name, ".") || !strcmp(name, "..")) 1057 return false; 1058 1059 while (*name) { 1060 if (*name == '/' || *name == ':' || isspace(*name)) 1061 return false; 1062 name++; 1063 } 1064 return true; 1065 } 1066 EXPORT_SYMBOL(dev_valid_name); 1067 1068 /** 1069 * __dev_alloc_name - allocate a name for a device 1070 * @net: network namespace to allocate the device name in 1071 * @name: name format string 1072 * @buf: scratch buffer and result name string 1073 * 1074 * Passed a format string - eg "lt%d" it will try and find a suitable 1075 * id. It scans list of devices to build up a free map, then chooses 1076 * the first empty slot. The caller must hold the dev_base or rtnl lock 1077 * while allocating the name and adding the device in order to avoid 1078 * duplicates. 1079 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1080 * Returns the number of the unit assigned or a negative errno code. 1081 */ 1082 1083 static int __dev_alloc_name(struct net *net, const char *name, char *buf) 1084 { 1085 int i = 0; 1086 const char *p; 1087 const int max_netdevices = 8*PAGE_SIZE; 1088 unsigned long *inuse; 1089 struct net_device *d; 1090 1091 if (!dev_valid_name(name)) 1092 return -EINVAL; 1093 1094 p = strchr(name, '%'); 1095 if (p) { 1096 /* 1097 * Verify the string as this thing may have come from 1098 * the user. There must be either one "%d" and no other "%" 1099 * characters. 1100 */ 1101 if (p[1] != 'd' || strchr(p + 2, '%')) 1102 return -EINVAL; 1103 1104 /* Use one page as a bit array of possible slots */ 1105 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); 1106 if (!inuse) 1107 return -ENOMEM; 1108 1109 for_each_netdev(net, d) { 1110 struct netdev_name_node *name_node; 1111 list_for_each_entry(name_node, &d->name_node->list, list) { 1112 if (!sscanf(name_node->name, name, &i)) 1113 continue; 1114 if (i < 0 || i >= max_netdevices) 1115 continue; 1116 1117 /* avoid cases where sscanf is not exact inverse of printf */ 1118 snprintf(buf, IFNAMSIZ, name, i); 1119 if (!strncmp(buf, name_node->name, IFNAMSIZ)) 1120 set_bit(i, inuse); 1121 } 1122 if (!sscanf(d->name, name, &i)) 1123 continue; 1124 if (i < 0 || i >= max_netdevices) 1125 continue; 1126 1127 /* avoid cases where sscanf is not exact inverse of printf */ 1128 snprintf(buf, IFNAMSIZ, name, i); 1129 if (!strncmp(buf, d->name, IFNAMSIZ)) 1130 set_bit(i, inuse); 1131 } 1132 1133 i = find_first_zero_bit(inuse, max_netdevices); 1134 free_page((unsigned long) inuse); 1135 } 1136 1137 snprintf(buf, IFNAMSIZ, name, i); 1138 if (!__dev_get_by_name(net, buf)) 1139 return i; 1140 1141 /* It is possible to run out of possible slots 1142 * when the name is long and there isn't enough space left 1143 * for the digits, or if all bits are used. 1144 */ 1145 return -ENFILE; 1146 } 1147 1148 static int dev_alloc_name_ns(struct net *net, 1149 struct net_device *dev, 1150 const char *name) 1151 { 1152 char buf[IFNAMSIZ]; 1153 int ret; 1154 1155 BUG_ON(!net); 1156 ret = __dev_alloc_name(net, name, buf); 1157 if (ret >= 0) 1158 strlcpy(dev->name, buf, IFNAMSIZ); 1159 return ret; 1160 } 1161 1162 /** 1163 * dev_alloc_name - allocate a name for a device 1164 * @dev: device 1165 * @name: name format string 1166 * 1167 * Passed a format string - eg "lt%d" it will try and find a suitable 1168 * id. It scans list of devices to build up a free map, then chooses 1169 * the first empty slot. The caller must hold the dev_base or rtnl lock 1170 * while allocating the name and adding the device in order to avoid 1171 * duplicates. 1172 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1173 * Returns the number of the unit assigned or a negative errno code. 1174 */ 1175 1176 int dev_alloc_name(struct net_device *dev, const char *name) 1177 { 1178 return dev_alloc_name_ns(dev_net(dev), dev, name); 1179 } 1180 EXPORT_SYMBOL(dev_alloc_name); 1181 1182 static int dev_get_valid_name(struct net *net, struct net_device *dev, 1183 const char *name) 1184 { 1185 BUG_ON(!net); 1186 1187 if (!dev_valid_name(name)) 1188 return -EINVAL; 1189 1190 if (strchr(name, '%')) 1191 return dev_alloc_name_ns(net, dev, name); 1192 else if (__dev_get_by_name(net, name)) 1193 return -EEXIST; 1194 else if (dev->name != name) 1195 strlcpy(dev->name, name, IFNAMSIZ); 1196 1197 return 0; 1198 } 1199 1200 /** 1201 * dev_change_name - change name of a device 1202 * @dev: device 1203 * @newname: name (or format string) must be at least IFNAMSIZ 1204 * 1205 * Change name of a device, can pass format strings "eth%d". 1206 * for wildcarding. 1207 */ 1208 int dev_change_name(struct net_device *dev, const char *newname) 1209 { 1210 unsigned char old_assign_type; 1211 char oldname[IFNAMSIZ]; 1212 int err = 0; 1213 int ret; 1214 struct net *net; 1215 1216 ASSERT_RTNL(); 1217 BUG_ON(!dev_net(dev)); 1218 1219 net = dev_net(dev); 1220 1221 /* Some auto-enslaved devices e.g. failover slaves are 1222 * special, as userspace might rename the device after 1223 * the interface had been brought up and running since 1224 * the point kernel initiated auto-enslavement. Allow 1225 * live name change even when these slave devices are 1226 * up and running. 1227 * 1228 * Typically, users of these auto-enslaving devices 1229 * don't actually care about slave name change, as 1230 * they are supposed to operate on master interface 1231 * directly. 1232 */ 1233 if (dev->flags & IFF_UP && 1234 likely(!(dev->priv_flags & IFF_LIVE_RENAME_OK))) 1235 return -EBUSY; 1236 1237 down_write(&devnet_rename_sem); 1238 1239 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { 1240 up_write(&devnet_rename_sem); 1241 return 0; 1242 } 1243 1244 memcpy(oldname, dev->name, IFNAMSIZ); 1245 1246 err = dev_get_valid_name(net, dev, newname); 1247 if (err < 0) { 1248 up_write(&devnet_rename_sem); 1249 return err; 1250 } 1251 1252 if (oldname[0] && !strchr(oldname, '%')) 1253 netdev_info(dev, "renamed from %s\n", oldname); 1254 1255 old_assign_type = dev->name_assign_type; 1256 dev->name_assign_type = NET_NAME_RENAMED; 1257 1258 rollback: 1259 ret = device_rename(&dev->dev, dev->name); 1260 if (ret) { 1261 memcpy(dev->name, oldname, IFNAMSIZ); 1262 dev->name_assign_type = old_assign_type; 1263 up_write(&devnet_rename_sem); 1264 return ret; 1265 } 1266 1267 up_write(&devnet_rename_sem); 1268 1269 netdev_adjacent_rename_links(dev, oldname); 1270 1271 write_lock_bh(&dev_base_lock); 1272 netdev_name_node_del(dev->name_node); 1273 write_unlock_bh(&dev_base_lock); 1274 1275 synchronize_rcu(); 1276 1277 write_lock_bh(&dev_base_lock); 1278 netdev_name_node_add(net, dev->name_node); 1279 write_unlock_bh(&dev_base_lock); 1280 1281 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); 1282 ret = notifier_to_errno(ret); 1283 1284 if (ret) { 1285 /* err >= 0 after dev_alloc_name() or stores the first errno */ 1286 if (err >= 0) { 1287 err = ret; 1288 down_write(&devnet_rename_sem); 1289 memcpy(dev->name, oldname, IFNAMSIZ); 1290 memcpy(oldname, newname, IFNAMSIZ); 1291 dev->name_assign_type = old_assign_type; 1292 old_assign_type = NET_NAME_RENAMED; 1293 goto rollback; 1294 } else { 1295 pr_err("%s: name change rollback failed: %d\n", 1296 dev->name, ret); 1297 } 1298 } 1299 1300 return err; 1301 } 1302 1303 /** 1304 * dev_set_alias - change ifalias of a device 1305 * @dev: device 1306 * @alias: name up to IFALIASZ 1307 * @len: limit of bytes to copy from info 1308 * 1309 * Set ifalias for a device, 1310 */ 1311 int dev_set_alias(struct net_device *dev, const char *alias, size_t len) 1312 { 1313 struct dev_ifalias *new_alias = NULL; 1314 1315 if (len >= IFALIASZ) 1316 return -EINVAL; 1317 1318 if (len) { 1319 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL); 1320 if (!new_alias) 1321 return -ENOMEM; 1322 1323 memcpy(new_alias->ifalias, alias, len); 1324 new_alias->ifalias[len] = 0; 1325 } 1326 1327 mutex_lock(&ifalias_mutex); 1328 new_alias = rcu_replace_pointer(dev->ifalias, new_alias, 1329 mutex_is_locked(&ifalias_mutex)); 1330 mutex_unlock(&ifalias_mutex); 1331 1332 if (new_alias) 1333 kfree_rcu(new_alias, rcuhead); 1334 1335 return len; 1336 } 1337 EXPORT_SYMBOL(dev_set_alias); 1338 1339 /** 1340 * dev_get_alias - get ifalias of a device 1341 * @dev: device 1342 * @name: buffer to store name of ifalias 1343 * @len: size of buffer 1344 * 1345 * get ifalias for a device. Caller must make sure dev cannot go 1346 * away, e.g. rcu read lock or own a reference count to device. 1347 */ 1348 int dev_get_alias(const struct net_device *dev, char *name, size_t len) 1349 { 1350 const struct dev_ifalias *alias; 1351 int ret = 0; 1352 1353 rcu_read_lock(); 1354 alias = rcu_dereference(dev->ifalias); 1355 if (alias) 1356 ret = snprintf(name, len, "%s", alias->ifalias); 1357 rcu_read_unlock(); 1358 1359 return ret; 1360 } 1361 1362 /** 1363 * netdev_features_change - device changes features 1364 * @dev: device to cause notification 1365 * 1366 * Called to indicate a device has changed features. 1367 */ 1368 void netdev_features_change(struct net_device *dev) 1369 { 1370 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); 1371 } 1372 EXPORT_SYMBOL(netdev_features_change); 1373 1374 /** 1375 * netdev_state_change - device changes state 1376 * @dev: device to cause notification 1377 * 1378 * Called to indicate a device has changed state. This function calls 1379 * the notifier chains for netdev_chain and sends a NEWLINK message 1380 * to the routing socket. 1381 */ 1382 void netdev_state_change(struct net_device *dev) 1383 { 1384 if (dev->flags & IFF_UP) { 1385 struct netdev_notifier_change_info change_info = { 1386 .info.dev = dev, 1387 }; 1388 1389 call_netdevice_notifiers_info(NETDEV_CHANGE, 1390 &change_info.info); 1391 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL); 1392 } 1393 } 1394 EXPORT_SYMBOL(netdev_state_change); 1395 1396 /** 1397 * __netdev_notify_peers - notify network peers about existence of @dev, 1398 * to be called when rtnl lock is already held. 1399 * @dev: network device 1400 * 1401 * Generate traffic such that interested network peers are aware of 1402 * @dev, such as by generating a gratuitous ARP. This may be used when 1403 * a device wants to inform the rest of the network about some sort of 1404 * reconfiguration such as a failover event or virtual machine 1405 * migration. 1406 */ 1407 void __netdev_notify_peers(struct net_device *dev) 1408 { 1409 ASSERT_RTNL(); 1410 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); 1411 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev); 1412 } 1413 EXPORT_SYMBOL(__netdev_notify_peers); 1414 1415 /** 1416 * netdev_notify_peers - notify network peers about existence of @dev 1417 * @dev: network device 1418 * 1419 * Generate traffic such that interested network peers are aware of 1420 * @dev, such as by generating a gratuitous ARP. This may be used when 1421 * a device wants to inform the rest of the network about some sort of 1422 * reconfiguration such as a failover event or virtual machine 1423 * migration. 1424 */ 1425 void netdev_notify_peers(struct net_device *dev) 1426 { 1427 rtnl_lock(); 1428 __netdev_notify_peers(dev); 1429 rtnl_unlock(); 1430 } 1431 EXPORT_SYMBOL(netdev_notify_peers); 1432 1433 static int napi_threaded_poll(void *data); 1434 1435 static int napi_kthread_create(struct napi_struct *n) 1436 { 1437 int err = 0; 1438 1439 /* Create and wake up the kthread once to put it in 1440 * TASK_INTERRUPTIBLE mode to avoid the blocked task 1441 * warning and work with loadavg. 1442 */ 1443 n->thread = kthread_run(napi_threaded_poll, n, "napi/%s-%d", 1444 n->dev->name, n->napi_id); 1445 if (IS_ERR(n->thread)) { 1446 err = PTR_ERR(n->thread); 1447 pr_err("kthread_run failed with err %d\n", err); 1448 n->thread = NULL; 1449 } 1450 1451 return err; 1452 } 1453 1454 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1455 { 1456 const struct net_device_ops *ops = dev->netdev_ops; 1457 int ret; 1458 1459 ASSERT_RTNL(); 1460 1461 if (!netif_device_present(dev)) { 1462 /* may be detached because parent is runtime-suspended */ 1463 if (dev->dev.parent) 1464 pm_runtime_resume(dev->dev.parent); 1465 if (!netif_device_present(dev)) 1466 return -ENODEV; 1467 } 1468 1469 /* Block netpoll from trying to do any rx path servicing. 1470 * If we don't do this there is a chance ndo_poll_controller 1471 * or ndo_poll may be running while we open the device 1472 */ 1473 netpoll_poll_disable(dev); 1474 1475 ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack); 1476 ret = notifier_to_errno(ret); 1477 if (ret) 1478 return ret; 1479 1480 set_bit(__LINK_STATE_START, &dev->state); 1481 1482 if (ops->ndo_validate_addr) 1483 ret = ops->ndo_validate_addr(dev); 1484 1485 if (!ret && ops->ndo_open) 1486 ret = ops->ndo_open(dev); 1487 1488 netpoll_poll_enable(dev); 1489 1490 if (ret) 1491 clear_bit(__LINK_STATE_START, &dev->state); 1492 else { 1493 dev->flags |= IFF_UP; 1494 dev_set_rx_mode(dev); 1495 dev_activate(dev); 1496 add_device_randomness(dev->dev_addr, dev->addr_len); 1497 } 1498 1499 return ret; 1500 } 1501 1502 /** 1503 * dev_open - prepare an interface for use. 1504 * @dev: device to open 1505 * @extack: netlink extended ack 1506 * 1507 * Takes a device from down to up state. The device's private open 1508 * function is invoked and then the multicast lists are loaded. Finally 1509 * the device is moved into the up state and a %NETDEV_UP message is 1510 * sent to the netdev notifier chain. 1511 * 1512 * Calling this function on an active interface is a nop. On a failure 1513 * a negative errno code is returned. 1514 */ 1515 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1516 { 1517 int ret; 1518 1519 if (dev->flags & IFF_UP) 1520 return 0; 1521 1522 ret = __dev_open(dev, extack); 1523 if (ret < 0) 1524 return ret; 1525 1526 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); 1527 call_netdevice_notifiers(NETDEV_UP, dev); 1528 1529 return ret; 1530 } 1531 EXPORT_SYMBOL(dev_open); 1532 1533 static void __dev_close_many(struct list_head *head) 1534 { 1535 struct net_device *dev; 1536 1537 ASSERT_RTNL(); 1538 might_sleep(); 1539 1540 list_for_each_entry(dev, head, close_list) { 1541 /* Temporarily disable netpoll until the interface is down */ 1542 netpoll_poll_disable(dev); 1543 1544 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); 1545 1546 clear_bit(__LINK_STATE_START, &dev->state); 1547 1548 /* Synchronize to scheduled poll. We cannot touch poll list, it 1549 * can be even on different cpu. So just clear netif_running(). 1550 * 1551 * dev->stop() will invoke napi_disable() on all of it's 1552 * napi_struct instances on this device. 1553 */ 1554 smp_mb__after_atomic(); /* Commit netif_running(). */ 1555 } 1556 1557 dev_deactivate_many(head); 1558 1559 list_for_each_entry(dev, head, close_list) { 1560 const struct net_device_ops *ops = dev->netdev_ops; 1561 1562 /* 1563 * Call the device specific close. This cannot fail. 1564 * Only if device is UP 1565 * 1566 * We allow it to be called even after a DETACH hot-plug 1567 * event. 1568 */ 1569 if (ops->ndo_stop) 1570 ops->ndo_stop(dev); 1571 1572 dev->flags &= ~IFF_UP; 1573 netpoll_poll_enable(dev); 1574 } 1575 } 1576 1577 static void __dev_close(struct net_device *dev) 1578 { 1579 LIST_HEAD(single); 1580 1581 list_add(&dev->close_list, &single); 1582 __dev_close_many(&single); 1583 list_del(&single); 1584 } 1585 1586 void dev_close_many(struct list_head *head, bool unlink) 1587 { 1588 struct net_device *dev, *tmp; 1589 1590 /* Remove the devices that don't need to be closed */ 1591 list_for_each_entry_safe(dev, tmp, head, close_list) 1592 if (!(dev->flags & IFF_UP)) 1593 list_del_init(&dev->close_list); 1594 1595 __dev_close_many(head); 1596 1597 list_for_each_entry_safe(dev, tmp, head, close_list) { 1598 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); 1599 call_netdevice_notifiers(NETDEV_DOWN, dev); 1600 if (unlink) 1601 list_del_init(&dev->close_list); 1602 } 1603 } 1604 EXPORT_SYMBOL(dev_close_many); 1605 1606 /** 1607 * dev_close - shutdown an interface. 1608 * @dev: device to shutdown 1609 * 1610 * This function moves an active device into down state. A 1611 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device 1612 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier 1613 * chain. 1614 */ 1615 void dev_close(struct net_device *dev) 1616 { 1617 if (dev->flags & IFF_UP) { 1618 LIST_HEAD(single); 1619 1620 list_add(&dev->close_list, &single); 1621 dev_close_many(&single, true); 1622 list_del(&single); 1623 } 1624 } 1625 EXPORT_SYMBOL(dev_close); 1626 1627 1628 /** 1629 * dev_disable_lro - disable Large Receive Offload on a device 1630 * @dev: device 1631 * 1632 * Disable Large Receive Offload (LRO) on a net device. Must be 1633 * called under RTNL. This is needed if received packets may be 1634 * forwarded to another interface. 1635 */ 1636 void dev_disable_lro(struct net_device *dev) 1637 { 1638 struct net_device *lower_dev; 1639 struct list_head *iter; 1640 1641 dev->wanted_features &= ~NETIF_F_LRO; 1642 netdev_update_features(dev); 1643 1644 if (unlikely(dev->features & NETIF_F_LRO)) 1645 netdev_WARN(dev, "failed to disable LRO!\n"); 1646 1647 netdev_for_each_lower_dev(dev, lower_dev, iter) 1648 dev_disable_lro(lower_dev); 1649 } 1650 EXPORT_SYMBOL(dev_disable_lro); 1651 1652 /** 1653 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device 1654 * @dev: device 1655 * 1656 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be 1657 * called under RTNL. This is needed if Generic XDP is installed on 1658 * the device. 1659 */ 1660 static void dev_disable_gro_hw(struct net_device *dev) 1661 { 1662 dev->wanted_features &= ~NETIF_F_GRO_HW; 1663 netdev_update_features(dev); 1664 1665 if (unlikely(dev->features & NETIF_F_GRO_HW)) 1666 netdev_WARN(dev, "failed to disable GRO_HW!\n"); 1667 } 1668 1669 const char *netdev_cmd_to_name(enum netdev_cmd cmd) 1670 { 1671 #define N(val) \ 1672 case NETDEV_##val: \ 1673 return "NETDEV_" __stringify(val); 1674 switch (cmd) { 1675 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER) 1676 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE) 1677 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE) 1678 N(POST_INIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) N(CHANGEUPPER) 1679 N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) N(BONDING_INFO) 1680 N(PRECHANGEUPPER) N(CHANGELOWERSTATE) N(UDP_TUNNEL_PUSH_INFO) 1681 N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN) 1682 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO) 1683 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO) 1684 N(PRE_CHANGEADDR) 1685 } 1686 #undef N 1687 return "UNKNOWN_NETDEV_EVENT"; 1688 } 1689 EXPORT_SYMBOL_GPL(netdev_cmd_to_name); 1690 1691 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, 1692 struct net_device *dev) 1693 { 1694 struct netdev_notifier_info info = { 1695 .dev = dev, 1696 }; 1697 1698 return nb->notifier_call(nb, val, &info); 1699 } 1700 1701 static int call_netdevice_register_notifiers(struct notifier_block *nb, 1702 struct net_device *dev) 1703 { 1704 int err; 1705 1706 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); 1707 err = notifier_to_errno(err); 1708 if (err) 1709 return err; 1710 1711 if (!(dev->flags & IFF_UP)) 1712 return 0; 1713 1714 call_netdevice_notifier(nb, NETDEV_UP, dev); 1715 return 0; 1716 } 1717 1718 static void call_netdevice_unregister_notifiers(struct notifier_block *nb, 1719 struct net_device *dev) 1720 { 1721 if (dev->flags & IFF_UP) { 1722 call_netdevice_notifier(nb, NETDEV_GOING_DOWN, 1723 dev); 1724 call_netdevice_notifier(nb, NETDEV_DOWN, dev); 1725 } 1726 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); 1727 } 1728 1729 static int call_netdevice_register_net_notifiers(struct notifier_block *nb, 1730 struct net *net) 1731 { 1732 struct net_device *dev; 1733 int err; 1734 1735 for_each_netdev(net, dev) { 1736 err = call_netdevice_register_notifiers(nb, dev); 1737 if (err) 1738 goto rollback; 1739 } 1740 return 0; 1741 1742 rollback: 1743 for_each_netdev_continue_reverse(net, dev) 1744 call_netdevice_unregister_notifiers(nb, dev); 1745 return err; 1746 } 1747 1748 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb, 1749 struct net *net) 1750 { 1751 struct net_device *dev; 1752 1753 for_each_netdev(net, dev) 1754 call_netdevice_unregister_notifiers(nb, dev); 1755 } 1756 1757 static int dev_boot_phase = 1; 1758 1759 /** 1760 * register_netdevice_notifier - register a network notifier block 1761 * @nb: notifier 1762 * 1763 * Register a notifier to be called when network device events occur. 1764 * The notifier passed is linked into the kernel structures and must 1765 * not be reused until it has been unregistered. A negative errno code 1766 * is returned on a failure. 1767 * 1768 * When registered all registration and up events are replayed 1769 * to the new notifier to allow device to have a race free 1770 * view of the network device list. 1771 */ 1772 1773 int register_netdevice_notifier(struct notifier_block *nb) 1774 { 1775 struct net *net; 1776 int err; 1777 1778 /* Close race with setup_net() and cleanup_net() */ 1779 down_write(&pernet_ops_rwsem); 1780 rtnl_lock(); 1781 err = raw_notifier_chain_register(&netdev_chain, nb); 1782 if (err) 1783 goto unlock; 1784 if (dev_boot_phase) 1785 goto unlock; 1786 for_each_net(net) { 1787 err = call_netdevice_register_net_notifiers(nb, net); 1788 if (err) 1789 goto rollback; 1790 } 1791 1792 unlock: 1793 rtnl_unlock(); 1794 up_write(&pernet_ops_rwsem); 1795 return err; 1796 1797 rollback: 1798 for_each_net_continue_reverse(net) 1799 call_netdevice_unregister_net_notifiers(nb, net); 1800 1801 raw_notifier_chain_unregister(&netdev_chain, nb); 1802 goto unlock; 1803 } 1804 EXPORT_SYMBOL(register_netdevice_notifier); 1805 1806 /** 1807 * unregister_netdevice_notifier - unregister a network notifier block 1808 * @nb: notifier 1809 * 1810 * Unregister a notifier previously registered by 1811 * register_netdevice_notifier(). The notifier is unlinked into the 1812 * kernel structures and may then be reused. A negative errno code 1813 * is returned on a failure. 1814 * 1815 * After unregistering unregister and down device events are synthesized 1816 * for all devices on the device list to the removed notifier to remove 1817 * the need for special case cleanup code. 1818 */ 1819 1820 int unregister_netdevice_notifier(struct notifier_block *nb) 1821 { 1822 struct net *net; 1823 int err; 1824 1825 /* Close race with setup_net() and cleanup_net() */ 1826 down_write(&pernet_ops_rwsem); 1827 rtnl_lock(); 1828 err = raw_notifier_chain_unregister(&netdev_chain, nb); 1829 if (err) 1830 goto unlock; 1831 1832 for_each_net(net) 1833 call_netdevice_unregister_net_notifiers(nb, net); 1834 1835 unlock: 1836 rtnl_unlock(); 1837 up_write(&pernet_ops_rwsem); 1838 return err; 1839 } 1840 EXPORT_SYMBOL(unregister_netdevice_notifier); 1841 1842 static int __register_netdevice_notifier_net(struct net *net, 1843 struct notifier_block *nb, 1844 bool ignore_call_fail) 1845 { 1846 int err; 1847 1848 err = raw_notifier_chain_register(&net->netdev_chain, nb); 1849 if (err) 1850 return err; 1851 if (dev_boot_phase) 1852 return 0; 1853 1854 err = call_netdevice_register_net_notifiers(nb, net); 1855 if (err && !ignore_call_fail) 1856 goto chain_unregister; 1857 1858 return 0; 1859 1860 chain_unregister: 1861 raw_notifier_chain_unregister(&net->netdev_chain, nb); 1862 return err; 1863 } 1864 1865 static int __unregister_netdevice_notifier_net(struct net *net, 1866 struct notifier_block *nb) 1867 { 1868 int err; 1869 1870 err = raw_notifier_chain_unregister(&net->netdev_chain, nb); 1871 if (err) 1872 return err; 1873 1874 call_netdevice_unregister_net_notifiers(nb, net); 1875 return 0; 1876 } 1877 1878 /** 1879 * register_netdevice_notifier_net - register a per-netns network notifier block 1880 * @net: network namespace 1881 * @nb: notifier 1882 * 1883 * Register a notifier to be called when network device events occur. 1884 * The notifier passed is linked into the kernel structures and must 1885 * not be reused until it has been unregistered. A negative errno code 1886 * is returned on a failure. 1887 * 1888 * When registered all registration and up events are replayed 1889 * to the new notifier to allow device to have a race free 1890 * view of the network device list. 1891 */ 1892 1893 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb) 1894 { 1895 int err; 1896 1897 rtnl_lock(); 1898 err = __register_netdevice_notifier_net(net, nb, false); 1899 rtnl_unlock(); 1900 return err; 1901 } 1902 EXPORT_SYMBOL(register_netdevice_notifier_net); 1903 1904 /** 1905 * unregister_netdevice_notifier_net - unregister a per-netns 1906 * network notifier block 1907 * @net: network namespace 1908 * @nb: notifier 1909 * 1910 * Unregister a notifier previously registered by 1911 * register_netdevice_notifier(). The notifier is unlinked into the 1912 * kernel structures and may then be reused. A negative errno code 1913 * is returned on a failure. 1914 * 1915 * After unregistering unregister and down device events are synthesized 1916 * for all devices on the device list to the removed notifier to remove 1917 * the need for special case cleanup code. 1918 */ 1919 1920 int unregister_netdevice_notifier_net(struct net *net, 1921 struct notifier_block *nb) 1922 { 1923 int err; 1924 1925 rtnl_lock(); 1926 err = __unregister_netdevice_notifier_net(net, nb); 1927 rtnl_unlock(); 1928 return err; 1929 } 1930 EXPORT_SYMBOL(unregister_netdevice_notifier_net); 1931 1932 int register_netdevice_notifier_dev_net(struct net_device *dev, 1933 struct notifier_block *nb, 1934 struct netdev_net_notifier *nn) 1935 { 1936 int err; 1937 1938 rtnl_lock(); 1939 err = __register_netdevice_notifier_net(dev_net(dev), nb, false); 1940 if (!err) { 1941 nn->nb = nb; 1942 list_add(&nn->list, &dev->net_notifier_list); 1943 } 1944 rtnl_unlock(); 1945 return err; 1946 } 1947 EXPORT_SYMBOL(register_netdevice_notifier_dev_net); 1948 1949 int unregister_netdevice_notifier_dev_net(struct net_device *dev, 1950 struct notifier_block *nb, 1951 struct netdev_net_notifier *nn) 1952 { 1953 int err; 1954 1955 rtnl_lock(); 1956 list_del(&nn->list); 1957 err = __unregister_netdevice_notifier_net(dev_net(dev), nb); 1958 rtnl_unlock(); 1959 return err; 1960 } 1961 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net); 1962 1963 static void move_netdevice_notifiers_dev_net(struct net_device *dev, 1964 struct net *net) 1965 { 1966 struct netdev_net_notifier *nn; 1967 1968 list_for_each_entry(nn, &dev->net_notifier_list, list) { 1969 __unregister_netdevice_notifier_net(dev_net(dev), nn->nb); 1970 __register_netdevice_notifier_net(net, nn->nb, true); 1971 } 1972 } 1973 1974 /** 1975 * call_netdevice_notifiers_info - call all network notifier blocks 1976 * @val: value passed unmodified to notifier function 1977 * @info: notifier information data 1978 * 1979 * Call all network notifier blocks. Parameters and return value 1980 * are as for raw_notifier_call_chain(). 1981 */ 1982 1983 static int call_netdevice_notifiers_info(unsigned long val, 1984 struct netdev_notifier_info *info) 1985 { 1986 struct net *net = dev_net(info->dev); 1987 int ret; 1988 1989 ASSERT_RTNL(); 1990 1991 /* Run per-netns notifier block chain first, then run the global one. 1992 * Hopefully, one day, the global one is going to be removed after 1993 * all notifier block registrators get converted to be per-netns. 1994 */ 1995 ret = raw_notifier_call_chain(&net->netdev_chain, val, info); 1996 if (ret & NOTIFY_STOP_MASK) 1997 return ret; 1998 return raw_notifier_call_chain(&netdev_chain, val, info); 1999 } 2000 2001 static int call_netdevice_notifiers_extack(unsigned long val, 2002 struct net_device *dev, 2003 struct netlink_ext_ack *extack) 2004 { 2005 struct netdev_notifier_info info = { 2006 .dev = dev, 2007 .extack = extack, 2008 }; 2009 2010 return call_netdevice_notifiers_info(val, &info); 2011 } 2012 2013 /** 2014 * call_netdevice_notifiers - call all network notifier blocks 2015 * @val: value passed unmodified to notifier function 2016 * @dev: net_device pointer passed unmodified to notifier function 2017 * 2018 * Call all network notifier blocks. Parameters and return value 2019 * are as for raw_notifier_call_chain(). 2020 */ 2021 2022 int call_netdevice_notifiers(unsigned long val, struct net_device *dev) 2023 { 2024 return call_netdevice_notifiers_extack(val, dev, NULL); 2025 } 2026 EXPORT_SYMBOL(call_netdevice_notifiers); 2027 2028 /** 2029 * call_netdevice_notifiers_mtu - call all network notifier blocks 2030 * @val: value passed unmodified to notifier function 2031 * @dev: net_device pointer passed unmodified to notifier function 2032 * @arg: additional u32 argument passed to the notifier function 2033 * 2034 * Call all network notifier blocks. Parameters and return value 2035 * are as for raw_notifier_call_chain(). 2036 */ 2037 static int call_netdevice_notifiers_mtu(unsigned long val, 2038 struct net_device *dev, u32 arg) 2039 { 2040 struct netdev_notifier_info_ext info = { 2041 .info.dev = dev, 2042 .ext.mtu = arg, 2043 }; 2044 2045 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0); 2046 2047 return call_netdevice_notifiers_info(val, &info.info); 2048 } 2049 2050 #ifdef CONFIG_NET_INGRESS 2051 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key); 2052 2053 void net_inc_ingress_queue(void) 2054 { 2055 static_branch_inc(&ingress_needed_key); 2056 } 2057 EXPORT_SYMBOL_GPL(net_inc_ingress_queue); 2058 2059 void net_dec_ingress_queue(void) 2060 { 2061 static_branch_dec(&ingress_needed_key); 2062 } 2063 EXPORT_SYMBOL_GPL(net_dec_ingress_queue); 2064 #endif 2065 2066 #ifdef CONFIG_NET_EGRESS 2067 static DEFINE_STATIC_KEY_FALSE(egress_needed_key); 2068 2069 void net_inc_egress_queue(void) 2070 { 2071 static_branch_inc(&egress_needed_key); 2072 } 2073 EXPORT_SYMBOL_GPL(net_inc_egress_queue); 2074 2075 void net_dec_egress_queue(void) 2076 { 2077 static_branch_dec(&egress_needed_key); 2078 } 2079 EXPORT_SYMBOL_GPL(net_dec_egress_queue); 2080 #endif 2081 2082 static DEFINE_STATIC_KEY_FALSE(netstamp_needed_key); 2083 #ifdef CONFIG_JUMP_LABEL 2084 static atomic_t netstamp_needed_deferred; 2085 static atomic_t netstamp_wanted; 2086 static void netstamp_clear(struct work_struct *work) 2087 { 2088 int deferred = atomic_xchg(&netstamp_needed_deferred, 0); 2089 int wanted; 2090 2091 wanted = atomic_add_return(deferred, &netstamp_wanted); 2092 if (wanted > 0) 2093 static_branch_enable(&netstamp_needed_key); 2094 else 2095 static_branch_disable(&netstamp_needed_key); 2096 } 2097 static DECLARE_WORK(netstamp_work, netstamp_clear); 2098 #endif 2099 2100 void net_enable_timestamp(void) 2101 { 2102 #ifdef CONFIG_JUMP_LABEL 2103 int wanted; 2104 2105 while (1) { 2106 wanted = atomic_read(&netstamp_wanted); 2107 if (wanted <= 0) 2108 break; 2109 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted) 2110 return; 2111 } 2112 atomic_inc(&netstamp_needed_deferred); 2113 schedule_work(&netstamp_work); 2114 #else 2115 static_branch_inc(&netstamp_needed_key); 2116 #endif 2117 } 2118 EXPORT_SYMBOL(net_enable_timestamp); 2119 2120 void net_disable_timestamp(void) 2121 { 2122 #ifdef CONFIG_JUMP_LABEL 2123 int wanted; 2124 2125 while (1) { 2126 wanted = atomic_read(&netstamp_wanted); 2127 if (wanted <= 1) 2128 break; 2129 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted) 2130 return; 2131 } 2132 atomic_dec(&netstamp_needed_deferred); 2133 schedule_work(&netstamp_work); 2134 #else 2135 static_branch_dec(&netstamp_needed_key); 2136 #endif 2137 } 2138 EXPORT_SYMBOL(net_disable_timestamp); 2139 2140 static inline void net_timestamp_set(struct sk_buff *skb) 2141 { 2142 skb->tstamp = 0; 2143 if (static_branch_unlikely(&netstamp_needed_key)) 2144 __net_timestamp(skb); 2145 } 2146 2147 #define net_timestamp_check(COND, SKB) \ 2148 if (static_branch_unlikely(&netstamp_needed_key)) { \ 2149 if ((COND) && !(SKB)->tstamp) \ 2150 __net_timestamp(SKB); \ 2151 } \ 2152 2153 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb) 2154 { 2155 return __is_skb_forwardable(dev, skb, true); 2156 } 2157 EXPORT_SYMBOL_GPL(is_skb_forwardable); 2158 2159 static int __dev_forward_skb2(struct net_device *dev, struct sk_buff *skb, 2160 bool check_mtu) 2161 { 2162 int ret = ____dev_forward_skb(dev, skb, check_mtu); 2163 2164 if (likely(!ret)) { 2165 skb->protocol = eth_type_trans(skb, dev); 2166 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); 2167 } 2168 2169 return ret; 2170 } 2171 2172 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2173 { 2174 return __dev_forward_skb2(dev, skb, true); 2175 } 2176 EXPORT_SYMBOL_GPL(__dev_forward_skb); 2177 2178 /** 2179 * dev_forward_skb - loopback an skb to another netif 2180 * 2181 * @dev: destination network device 2182 * @skb: buffer to forward 2183 * 2184 * return values: 2185 * NET_RX_SUCCESS (no congestion) 2186 * NET_RX_DROP (packet was dropped, but freed) 2187 * 2188 * dev_forward_skb can be used for injecting an skb from the 2189 * start_xmit function of one device into the receive queue 2190 * of another device. 2191 * 2192 * The receiving device may be in another namespace, so 2193 * we have to clear all information in the skb that could 2194 * impact namespace isolation. 2195 */ 2196 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2197 { 2198 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); 2199 } 2200 EXPORT_SYMBOL_GPL(dev_forward_skb); 2201 2202 int dev_forward_skb_nomtu(struct net_device *dev, struct sk_buff *skb) 2203 { 2204 return __dev_forward_skb2(dev, skb, false) ?: netif_rx_internal(skb); 2205 } 2206 2207 static inline int deliver_skb(struct sk_buff *skb, 2208 struct packet_type *pt_prev, 2209 struct net_device *orig_dev) 2210 { 2211 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 2212 return -ENOMEM; 2213 refcount_inc(&skb->users); 2214 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 2215 } 2216 2217 static inline void deliver_ptype_list_skb(struct sk_buff *skb, 2218 struct packet_type **pt, 2219 struct net_device *orig_dev, 2220 __be16 type, 2221 struct list_head *ptype_list) 2222 { 2223 struct packet_type *ptype, *pt_prev = *pt; 2224 2225 list_for_each_entry_rcu(ptype, ptype_list, list) { 2226 if (ptype->type != type) 2227 continue; 2228 if (pt_prev) 2229 deliver_skb(skb, pt_prev, orig_dev); 2230 pt_prev = ptype; 2231 } 2232 *pt = pt_prev; 2233 } 2234 2235 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) 2236 { 2237 if (!ptype->af_packet_priv || !skb->sk) 2238 return false; 2239 2240 if (ptype->id_match) 2241 return ptype->id_match(ptype, skb->sk); 2242 else if ((struct sock *)ptype->af_packet_priv == skb->sk) 2243 return true; 2244 2245 return false; 2246 } 2247 2248 /** 2249 * dev_nit_active - return true if any network interface taps are in use 2250 * 2251 * @dev: network device to check for the presence of taps 2252 */ 2253 bool dev_nit_active(struct net_device *dev) 2254 { 2255 return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all); 2256 } 2257 EXPORT_SYMBOL_GPL(dev_nit_active); 2258 2259 /* 2260 * Support routine. Sends outgoing frames to any network 2261 * taps currently in use. 2262 */ 2263 2264 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) 2265 { 2266 struct packet_type *ptype; 2267 struct sk_buff *skb2 = NULL; 2268 struct packet_type *pt_prev = NULL; 2269 struct list_head *ptype_list = &ptype_all; 2270 2271 rcu_read_lock(); 2272 again: 2273 list_for_each_entry_rcu(ptype, ptype_list, list) { 2274 if (ptype->ignore_outgoing) 2275 continue; 2276 2277 /* Never send packets back to the socket 2278 * they originated from - MvS (miquels@drinkel.ow.org) 2279 */ 2280 if (skb_loop_sk(ptype, skb)) 2281 continue; 2282 2283 if (pt_prev) { 2284 deliver_skb(skb2, pt_prev, skb->dev); 2285 pt_prev = ptype; 2286 continue; 2287 } 2288 2289 /* need to clone skb, done only once */ 2290 skb2 = skb_clone(skb, GFP_ATOMIC); 2291 if (!skb2) 2292 goto out_unlock; 2293 2294 net_timestamp_set(skb2); 2295 2296 /* skb->nh should be correctly 2297 * set by sender, so that the second statement is 2298 * just protection against buggy protocols. 2299 */ 2300 skb_reset_mac_header(skb2); 2301 2302 if (skb_network_header(skb2) < skb2->data || 2303 skb_network_header(skb2) > skb_tail_pointer(skb2)) { 2304 net_crit_ratelimited("protocol %04x is buggy, dev %s\n", 2305 ntohs(skb2->protocol), 2306 dev->name); 2307 skb_reset_network_header(skb2); 2308 } 2309 2310 skb2->transport_header = skb2->network_header; 2311 skb2->pkt_type = PACKET_OUTGOING; 2312 pt_prev = ptype; 2313 } 2314 2315 if (ptype_list == &ptype_all) { 2316 ptype_list = &dev->ptype_all; 2317 goto again; 2318 } 2319 out_unlock: 2320 if (pt_prev) { 2321 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC)) 2322 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); 2323 else 2324 kfree_skb(skb2); 2325 } 2326 rcu_read_unlock(); 2327 } 2328 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit); 2329 2330 /** 2331 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change 2332 * @dev: Network device 2333 * @txq: number of queues available 2334 * 2335 * If real_num_tx_queues is changed the tc mappings may no longer be 2336 * valid. To resolve this verify the tc mapping remains valid and if 2337 * not NULL the mapping. With no priorities mapping to this 2338 * offset/count pair it will no longer be used. In the worst case TC0 2339 * is invalid nothing can be done so disable priority mappings. If is 2340 * expected that drivers will fix this mapping if they can before 2341 * calling netif_set_real_num_tx_queues. 2342 */ 2343 static void netif_setup_tc(struct net_device *dev, unsigned int txq) 2344 { 2345 int i; 2346 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2347 2348 /* If TC0 is invalidated disable TC mapping */ 2349 if (tc->offset + tc->count > txq) { 2350 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); 2351 dev->num_tc = 0; 2352 return; 2353 } 2354 2355 /* Invalidated prio to tc mappings set to TC0 */ 2356 for (i = 1; i < TC_BITMASK + 1; i++) { 2357 int q = netdev_get_prio_tc_map(dev, i); 2358 2359 tc = &dev->tc_to_txq[q]; 2360 if (tc->offset + tc->count > txq) { 2361 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n", 2362 i, q); 2363 netdev_set_prio_tc_map(dev, i, 0); 2364 } 2365 } 2366 } 2367 2368 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq) 2369 { 2370 if (dev->num_tc) { 2371 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2372 int i; 2373 2374 /* walk through the TCs and see if it falls into any of them */ 2375 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) { 2376 if ((txq - tc->offset) < tc->count) 2377 return i; 2378 } 2379 2380 /* didn't find it, just return -1 to indicate no match */ 2381 return -1; 2382 } 2383 2384 return 0; 2385 } 2386 EXPORT_SYMBOL(netdev_txq_to_tc); 2387 2388 #ifdef CONFIG_XPS 2389 static struct static_key xps_needed __read_mostly; 2390 static struct static_key xps_rxqs_needed __read_mostly; 2391 static DEFINE_MUTEX(xps_map_mutex); 2392 #define xmap_dereference(P) \ 2393 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) 2394 2395 static bool remove_xps_queue(struct xps_dev_maps *dev_maps, 2396 struct xps_dev_maps *old_maps, int tci, u16 index) 2397 { 2398 struct xps_map *map = NULL; 2399 int pos; 2400 2401 if (dev_maps) 2402 map = xmap_dereference(dev_maps->attr_map[tci]); 2403 if (!map) 2404 return false; 2405 2406 for (pos = map->len; pos--;) { 2407 if (map->queues[pos] != index) 2408 continue; 2409 2410 if (map->len > 1) { 2411 map->queues[pos] = map->queues[--map->len]; 2412 break; 2413 } 2414 2415 if (old_maps) 2416 RCU_INIT_POINTER(old_maps->attr_map[tci], NULL); 2417 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2418 kfree_rcu(map, rcu); 2419 return false; 2420 } 2421 2422 return true; 2423 } 2424 2425 static bool remove_xps_queue_cpu(struct net_device *dev, 2426 struct xps_dev_maps *dev_maps, 2427 int cpu, u16 offset, u16 count) 2428 { 2429 int num_tc = dev_maps->num_tc; 2430 bool active = false; 2431 int tci; 2432 2433 for (tci = cpu * num_tc; num_tc--; tci++) { 2434 int i, j; 2435 2436 for (i = count, j = offset; i--; j++) { 2437 if (!remove_xps_queue(dev_maps, NULL, tci, j)) 2438 break; 2439 } 2440 2441 active |= i < 0; 2442 } 2443 2444 return active; 2445 } 2446 2447 static void reset_xps_maps(struct net_device *dev, 2448 struct xps_dev_maps *dev_maps, 2449 enum xps_map_type type) 2450 { 2451 static_key_slow_dec_cpuslocked(&xps_needed); 2452 if (type == XPS_RXQS) 2453 static_key_slow_dec_cpuslocked(&xps_rxqs_needed); 2454 2455 RCU_INIT_POINTER(dev->xps_maps[type], NULL); 2456 2457 kfree_rcu(dev_maps, rcu); 2458 } 2459 2460 static void clean_xps_maps(struct net_device *dev, enum xps_map_type type, 2461 u16 offset, u16 count) 2462 { 2463 struct xps_dev_maps *dev_maps; 2464 bool active = false; 2465 int i, j; 2466 2467 dev_maps = xmap_dereference(dev->xps_maps[type]); 2468 if (!dev_maps) 2469 return; 2470 2471 for (j = 0; j < dev_maps->nr_ids; j++) 2472 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, count); 2473 if (!active) 2474 reset_xps_maps(dev, dev_maps, type); 2475 2476 if (type == XPS_CPUS) { 2477 for (i = offset + (count - 1); count--; i--) 2478 netdev_queue_numa_node_write( 2479 netdev_get_tx_queue(dev, i), NUMA_NO_NODE); 2480 } 2481 } 2482 2483 static void netif_reset_xps_queues(struct net_device *dev, u16 offset, 2484 u16 count) 2485 { 2486 if (!static_key_false(&xps_needed)) 2487 return; 2488 2489 cpus_read_lock(); 2490 mutex_lock(&xps_map_mutex); 2491 2492 if (static_key_false(&xps_rxqs_needed)) 2493 clean_xps_maps(dev, XPS_RXQS, offset, count); 2494 2495 clean_xps_maps(dev, XPS_CPUS, offset, count); 2496 2497 mutex_unlock(&xps_map_mutex); 2498 cpus_read_unlock(); 2499 } 2500 2501 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) 2502 { 2503 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index); 2504 } 2505 2506 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index, 2507 u16 index, bool is_rxqs_map) 2508 { 2509 struct xps_map *new_map; 2510 int alloc_len = XPS_MIN_MAP_ALLOC; 2511 int i, pos; 2512 2513 for (pos = 0; map && pos < map->len; pos++) { 2514 if (map->queues[pos] != index) 2515 continue; 2516 return map; 2517 } 2518 2519 /* Need to add tx-queue to this CPU's/rx-queue's existing map */ 2520 if (map) { 2521 if (pos < map->alloc_len) 2522 return map; 2523 2524 alloc_len = map->alloc_len * 2; 2525 } 2526 2527 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's 2528 * map 2529 */ 2530 if (is_rxqs_map) 2531 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL); 2532 else 2533 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, 2534 cpu_to_node(attr_index)); 2535 if (!new_map) 2536 return NULL; 2537 2538 for (i = 0; i < pos; i++) 2539 new_map->queues[i] = map->queues[i]; 2540 new_map->alloc_len = alloc_len; 2541 new_map->len = pos; 2542 2543 return new_map; 2544 } 2545 2546 /* Copy xps maps at a given index */ 2547 static void xps_copy_dev_maps(struct xps_dev_maps *dev_maps, 2548 struct xps_dev_maps *new_dev_maps, int index, 2549 int tc, bool skip_tc) 2550 { 2551 int i, tci = index * dev_maps->num_tc; 2552 struct xps_map *map; 2553 2554 /* copy maps belonging to foreign traffic classes */ 2555 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2556 if (i == tc && skip_tc) 2557 continue; 2558 2559 /* fill in the new device map from the old device map */ 2560 map = xmap_dereference(dev_maps->attr_map[tci]); 2561 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2562 } 2563 } 2564 2565 /* Must be called under cpus_read_lock */ 2566 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask, 2567 u16 index, enum xps_map_type type) 2568 { 2569 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL, *old_dev_maps = NULL; 2570 const unsigned long *online_mask = NULL; 2571 bool active = false, copy = false; 2572 int i, j, tci, numa_node_id = -2; 2573 int maps_sz, num_tc = 1, tc = 0; 2574 struct xps_map *map, *new_map; 2575 unsigned int nr_ids; 2576 2577 if (dev->num_tc) { 2578 /* Do not allow XPS on subordinate device directly */ 2579 num_tc = dev->num_tc; 2580 if (num_tc < 0) 2581 return -EINVAL; 2582 2583 /* If queue belongs to subordinate dev use its map */ 2584 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev; 2585 2586 tc = netdev_txq_to_tc(dev, index); 2587 if (tc < 0) 2588 return -EINVAL; 2589 } 2590 2591 mutex_lock(&xps_map_mutex); 2592 2593 dev_maps = xmap_dereference(dev->xps_maps[type]); 2594 if (type == XPS_RXQS) { 2595 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues); 2596 nr_ids = dev->num_rx_queues; 2597 } else { 2598 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc); 2599 if (num_possible_cpus() > 1) 2600 online_mask = cpumask_bits(cpu_online_mask); 2601 nr_ids = nr_cpu_ids; 2602 } 2603 2604 if (maps_sz < L1_CACHE_BYTES) 2605 maps_sz = L1_CACHE_BYTES; 2606 2607 /* The old dev_maps could be larger or smaller than the one we're 2608 * setting up now, as dev->num_tc or nr_ids could have been updated in 2609 * between. We could try to be smart, but let's be safe instead and only 2610 * copy foreign traffic classes if the two map sizes match. 2611 */ 2612 if (dev_maps && 2613 dev_maps->num_tc == num_tc && dev_maps->nr_ids == nr_ids) 2614 copy = true; 2615 2616 /* allocate memory for queue storage */ 2617 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids), 2618 j < nr_ids;) { 2619 if (!new_dev_maps) { 2620 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); 2621 if (!new_dev_maps) { 2622 mutex_unlock(&xps_map_mutex); 2623 return -ENOMEM; 2624 } 2625 2626 new_dev_maps->nr_ids = nr_ids; 2627 new_dev_maps->num_tc = num_tc; 2628 } 2629 2630 tci = j * num_tc + tc; 2631 map = copy ? xmap_dereference(dev_maps->attr_map[tci]) : NULL; 2632 2633 map = expand_xps_map(map, j, index, type == XPS_RXQS); 2634 if (!map) 2635 goto error; 2636 2637 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2638 } 2639 2640 if (!new_dev_maps) 2641 goto out_no_new_maps; 2642 2643 if (!dev_maps) { 2644 /* Increment static keys at most once per type */ 2645 static_key_slow_inc_cpuslocked(&xps_needed); 2646 if (type == XPS_RXQS) 2647 static_key_slow_inc_cpuslocked(&xps_rxqs_needed); 2648 } 2649 2650 for (j = 0; j < nr_ids; j++) { 2651 bool skip_tc = false; 2652 2653 tci = j * num_tc + tc; 2654 if (netif_attr_test_mask(j, mask, nr_ids) && 2655 netif_attr_test_online(j, online_mask, nr_ids)) { 2656 /* add tx-queue to CPU/rx-queue maps */ 2657 int pos = 0; 2658 2659 skip_tc = true; 2660 2661 map = xmap_dereference(new_dev_maps->attr_map[tci]); 2662 while ((pos < map->len) && (map->queues[pos] != index)) 2663 pos++; 2664 2665 if (pos == map->len) 2666 map->queues[map->len++] = index; 2667 #ifdef CONFIG_NUMA 2668 if (type == XPS_CPUS) { 2669 if (numa_node_id == -2) 2670 numa_node_id = cpu_to_node(j); 2671 else if (numa_node_id != cpu_to_node(j)) 2672 numa_node_id = -1; 2673 } 2674 #endif 2675 } 2676 2677 if (copy) 2678 xps_copy_dev_maps(dev_maps, new_dev_maps, j, tc, 2679 skip_tc); 2680 } 2681 2682 rcu_assign_pointer(dev->xps_maps[type], new_dev_maps); 2683 2684 /* Cleanup old maps */ 2685 if (!dev_maps) 2686 goto out_no_old_maps; 2687 2688 for (j = 0; j < dev_maps->nr_ids; j++) { 2689 for (i = num_tc, tci = j * dev_maps->num_tc; i--; tci++) { 2690 map = xmap_dereference(dev_maps->attr_map[tci]); 2691 if (!map) 2692 continue; 2693 2694 if (copy) { 2695 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2696 if (map == new_map) 2697 continue; 2698 } 2699 2700 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2701 kfree_rcu(map, rcu); 2702 } 2703 } 2704 2705 old_dev_maps = dev_maps; 2706 2707 out_no_old_maps: 2708 dev_maps = new_dev_maps; 2709 active = true; 2710 2711 out_no_new_maps: 2712 if (type == XPS_CPUS) 2713 /* update Tx queue numa node */ 2714 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), 2715 (numa_node_id >= 0) ? 2716 numa_node_id : NUMA_NO_NODE); 2717 2718 if (!dev_maps) 2719 goto out_no_maps; 2720 2721 /* removes tx-queue from unused CPUs/rx-queues */ 2722 for (j = 0; j < dev_maps->nr_ids; j++) { 2723 tci = j * dev_maps->num_tc; 2724 2725 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2726 if (i == tc && 2727 netif_attr_test_mask(j, mask, dev_maps->nr_ids) && 2728 netif_attr_test_online(j, online_mask, dev_maps->nr_ids)) 2729 continue; 2730 2731 active |= remove_xps_queue(dev_maps, 2732 copy ? old_dev_maps : NULL, 2733 tci, index); 2734 } 2735 } 2736 2737 if (old_dev_maps) 2738 kfree_rcu(old_dev_maps, rcu); 2739 2740 /* free map if not active */ 2741 if (!active) 2742 reset_xps_maps(dev, dev_maps, type); 2743 2744 out_no_maps: 2745 mutex_unlock(&xps_map_mutex); 2746 2747 return 0; 2748 error: 2749 /* remove any maps that we added */ 2750 for (j = 0; j < nr_ids; j++) { 2751 for (i = num_tc, tci = j * num_tc; i--; tci++) { 2752 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2753 map = copy ? 2754 xmap_dereference(dev_maps->attr_map[tci]) : 2755 NULL; 2756 if (new_map && new_map != map) 2757 kfree(new_map); 2758 } 2759 } 2760 2761 mutex_unlock(&xps_map_mutex); 2762 2763 kfree(new_dev_maps); 2764 return -ENOMEM; 2765 } 2766 EXPORT_SYMBOL_GPL(__netif_set_xps_queue); 2767 2768 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, 2769 u16 index) 2770 { 2771 int ret; 2772 2773 cpus_read_lock(); 2774 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, XPS_CPUS); 2775 cpus_read_unlock(); 2776 2777 return ret; 2778 } 2779 EXPORT_SYMBOL(netif_set_xps_queue); 2780 2781 #endif 2782 static void netdev_unbind_all_sb_channels(struct net_device *dev) 2783 { 2784 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2785 2786 /* Unbind any subordinate channels */ 2787 while (txq-- != &dev->_tx[0]) { 2788 if (txq->sb_dev) 2789 netdev_unbind_sb_channel(dev, txq->sb_dev); 2790 } 2791 } 2792 2793 void netdev_reset_tc(struct net_device *dev) 2794 { 2795 #ifdef CONFIG_XPS 2796 netif_reset_xps_queues_gt(dev, 0); 2797 #endif 2798 netdev_unbind_all_sb_channels(dev); 2799 2800 /* Reset TC configuration of device */ 2801 dev->num_tc = 0; 2802 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq)); 2803 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map)); 2804 } 2805 EXPORT_SYMBOL(netdev_reset_tc); 2806 2807 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset) 2808 { 2809 if (tc >= dev->num_tc) 2810 return -EINVAL; 2811 2812 #ifdef CONFIG_XPS 2813 netif_reset_xps_queues(dev, offset, count); 2814 #endif 2815 dev->tc_to_txq[tc].count = count; 2816 dev->tc_to_txq[tc].offset = offset; 2817 return 0; 2818 } 2819 EXPORT_SYMBOL(netdev_set_tc_queue); 2820 2821 int netdev_set_num_tc(struct net_device *dev, u8 num_tc) 2822 { 2823 if (num_tc > TC_MAX_QUEUE) 2824 return -EINVAL; 2825 2826 #ifdef CONFIG_XPS 2827 netif_reset_xps_queues_gt(dev, 0); 2828 #endif 2829 netdev_unbind_all_sb_channels(dev); 2830 2831 dev->num_tc = num_tc; 2832 return 0; 2833 } 2834 EXPORT_SYMBOL(netdev_set_num_tc); 2835 2836 void netdev_unbind_sb_channel(struct net_device *dev, 2837 struct net_device *sb_dev) 2838 { 2839 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2840 2841 #ifdef CONFIG_XPS 2842 netif_reset_xps_queues_gt(sb_dev, 0); 2843 #endif 2844 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq)); 2845 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map)); 2846 2847 while (txq-- != &dev->_tx[0]) { 2848 if (txq->sb_dev == sb_dev) 2849 txq->sb_dev = NULL; 2850 } 2851 } 2852 EXPORT_SYMBOL(netdev_unbind_sb_channel); 2853 2854 int netdev_bind_sb_channel_queue(struct net_device *dev, 2855 struct net_device *sb_dev, 2856 u8 tc, u16 count, u16 offset) 2857 { 2858 /* Make certain the sb_dev and dev are already configured */ 2859 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc) 2860 return -EINVAL; 2861 2862 /* We cannot hand out queues we don't have */ 2863 if ((offset + count) > dev->real_num_tx_queues) 2864 return -EINVAL; 2865 2866 /* Record the mapping */ 2867 sb_dev->tc_to_txq[tc].count = count; 2868 sb_dev->tc_to_txq[tc].offset = offset; 2869 2870 /* Provide a way for Tx queue to find the tc_to_txq map or 2871 * XPS map for itself. 2872 */ 2873 while (count--) 2874 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev; 2875 2876 return 0; 2877 } 2878 EXPORT_SYMBOL(netdev_bind_sb_channel_queue); 2879 2880 int netdev_set_sb_channel(struct net_device *dev, u16 channel) 2881 { 2882 /* Do not use a multiqueue device to represent a subordinate channel */ 2883 if (netif_is_multiqueue(dev)) 2884 return -ENODEV; 2885 2886 /* We allow channels 1 - 32767 to be used for subordinate channels. 2887 * Channel 0 is meant to be "native" mode and used only to represent 2888 * the main root device. We allow writing 0 to reset the device back 2889 * to normal mode after being used as a subordinate channel. 2890 */ 2891 if (channel > S16_MAX) 2892 return -EINVAL; 2893 2894 dev->num_tc = -channel; 2895 2896 return 0; 2897 } 2898 EXPORT_SYMBOL(netdev_set_sb_channel); 2899 2900 /* 2901 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues 2902 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed. 2903 */ 2904 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) 2905 { 2906 bool disabling; 2907 int rc; 2908 2909 disabling = txq < dev->real_num_tx_queues; 2910 2911 if (txq < 1 || txq > dev->num_tx_queues) 2912 return -EINVAL; 2913 2914 if (dev->reg_state == NETREG_REGISTERED || 2915 dev->reg_state == NETREG_UNREGISTERING) { 2916 ASSERT_RTNL(); 2917 2918 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, 2919 txq); 2920 if (rc) 2921 return rc; 2922 2923 if (dev->num_tc) 2924 netif_setup_tc(dev, txq); 2925 2926 dev->real_num_tx_queues = txq; 2927 2928 if (disabling) { 2929 synchronize_net(); 2930 qdisc_reset_all_tx_gt(dev, txq); 2931 #ifdef CONFIG_XPS 2932 netif_reset_xps_queues_gt(dev, txq); 2933 #endif 2934 } 2935 } else { 2936 dev->real_num_tx_queues = txq; 2937 } 2938 2939 return 0; 2940 } 2941 EXPORT_SYMBOL(netif_set_real_num_tx_queues); 2942 2943 #ifdef CONFIG_SYSFS 2944 /** 2945 * netif_set_real_num_rx_queues - set actual number of RX queues used 2946 * @dev: Network device 2947 * @rxq: Actual number of RX queues 2948 * 2949 * This must be called either with the rtnl_lock held or before 2950 * registration of the net device. Returns 0 on success, or a 2951 * negative error code. If called before registration, it always 2952 * succeeds. 2953 */ 2954 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) 2955 { 2956 int rc; 2957 2958 if (rxq < 1 || rxq > dev->num_rx_queues) 2959 return -EINVAL; 2960 2961 if (dev->reg_state == NETREG_REGISTERED) { 2962 ASSERT_RTNL(); 2963 2964 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, 2965 rxq); 2966 if (rc) 2967 return rc; 2968 } 2969 2970 dev->real_num_rx_queues = rxq; 2971 return 0; 2972 } 2973 EXPORT_SYMBOL(netif_set_real_num_rx_queues); 2974 #endif 2975 2976 /** 2977 * netif_get_num_default_rss_queues - default number of RSS queues 2978 * 2979 * This routine should set an upper limit on the number of RSS queues 2980 * used by default by multiqueue devices. 2981 */ 2982 int netif_get_num_default_rss_queues(void) 2983 { 2984 return is_kdump_kernel() ? 2985 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus()); 2986 } 2987 EXPORT_SYMBOL(netif_get_num_default_rss_queues); 2988 2989 static void __netif_reschedule(struct Qdisc *q) 2990 { 2991 struct softnet_data *sd; 2992 unsigned long flags; 2993 2994 local_irq_save(flags); 2995 sd = this_cpu_ptr(&softnet_data); 2996 q->next_sched = NULL; 2997 *sd->output_queue_tailp = q; 2998 sd->output_queue_tailp = &q->next_sched; 2999 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3000 local_irq_restore(flags); 3001 } 3002 3003 void __netif_schedule(struct Qdisc *q) 3004 { 3005 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) 3006 __netif_reschedule(q); 3007 } 3008 EXPORT_SYMBOL(__netif_schedule); 3009 3010 struct dev_kfree_skb_cb { 3011 enum skb_free_reason reason; 3012 }; 3013 3014 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) 3015 { 3016 return (struct dev_kfree_skb_cb *)skb->cb; 3017 } 3018 3019 void netif_schedule_queue(struct netdev_queue *txq) 3020 { 3021 rcu_read_lock(); 3022 if (!netif_xmit_stopped(txq)) { 3023 struct Qdisc *q = rcu_dereference(txq->qdisc); 3024 3025 __netif_schedule(q); 3026 } 3027 rcu_read_unlock(); 3028 } 3029 EXPORT_SYMBOL(netif_schedule_queue); 3030 3031 void netif_tx_wake_queue(struct netdev_queue *dev_queue) 3032 { 3033 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { 3034 struct Qdisc *q; 3035 3036 rcu_read_lock(); 3037 q = rcu_dereference(dev_queue->qdisc); 3038 __netif_schedule(q); 3039 rcu_read_unlock(); 3040 } 3041 } 3042 EXPORT_SYMBOL(netif_tx_wake_queue); 3043 3044 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason) 3045 { 3046 unsigned long flags; 3047 3048 if (unlikely(!skb)) 3049 return; 3050 3051 if (likely(refcount_read(&skb->users) == 1)) { 3052 smp_rmb(); 3053 refcount_set(&skb->users, 0); 3054 } else if (likely(!refcount_dec_and_test(&skb->users))) { 3055 return; 3056 } 3057 get_kfree_skb_cb(skb)->reason = reason; 3058 local_irq_save(flags); 3059 skb->next = __this_cpu_read(softnet_data.completion_queue); 3060 __this_cpu_write(softnet_data.completion_queue, skb); 3061 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3062 local_irq_restore(flags); 3063 } 3064 EXPORT_SYMBOL(__dev_kfree_skb_irq); 3065 3066 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason) 3067 { 3068 if (in_irq() || irqs_disabled()) 3069 __dev_kfree_skb_irq(skb, reason); 3070 else 3071 dev_kfree_skb(skb); 3072 } 3073 EXPORT_SYMBOL(__dev_kfree_skb_any); 3074 3075 3076 /** 3077 * netif_device_detach - mark device as removed 3078 * @dev: network device 3079 * 3080 * Mark device as removed from system and therefore no longer available. 3081 */ 3082 void netif_device_detach(struct net_device *dev) 3083 { 3084 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && 3085 netif_running(dev)) { 3086 netif_tx_stop_all_queues(dev); 3087 } 3088 } 3089 EXPORT_SYMBOL(netif_device_detach); 3090 3091 /** 3092 * netif_device_attach - mark device as attached 3093 * @dev: network device 3094 * 3095 * Mark device as attached from system and restart if needed. 3096 */ 3097 void netif_device_attach(struct net_device *dev) 3098 { 3099 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && 3100 netif_running(dev)) { 3101 netif_tx_wake_all_queues(dev); 3102 __netdev_watchdog_up(dev); 3103 } 3104 } 3105 EXPORT_SYMBOL(netif_device_attach); 3106 3107 /* 3108 * Returns a Tx hash based on the given packet descriptor a Tx queues' number 3109 * to be used as a distribution range. 3110 */ 3111 static u16 skb_tx_hash(const struct net_device *dev, 3112 const struct net_device *sb_dev, 3113 struct sk_buff *skb) 3114 { 3115 u32 hash; 3116 u16 qoffset = 0; 3117 u16 qcount = dev->real_num_tx_queues; 3118 3119 if (dev->num_tc) { 3120 u8 tc = netdev_get_prio_tc_map(dev, skb->priority); 3121 3122 qoffset = sb_dev->tc_to_txq[tc].offset; 3123 qcount = sb_dev->tc_to_txq[tc].count; 3124 } 3125 3126 if (skb_rx_queue_recorded(skb)) { 3127 hash = skb_get_rx_queue(skb); 3128 if (hash >= qoffset) 3129 hash -= qoffset; 3130 while (unlikely(hash >= qcount)) 3131 hash -= qcount; 3132 return hash + qoffset; 3133 } 3134 3135 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; 3136 } 3137 3138 static void skb_warn_bad_offload(const struct sk_buff *skb) 3139 { 3140 static const netdev_features_t null_features; 3141 struct net_device *dev = skb->dev; 3142 const char *name = ""; 3143 3144 if (!net_ratelimit()) 3145 return; 3146 3147 if (dev) { 3148 if (dev->dev.parent) 3149 name = dev_driver_string(dev->dev.parent); 3150 else 3151 name = netdev_name(dev); 3152 } 3153 skb_dump(KERN_WARNING, skb, false); 3154 WARN(1, "%s: caps=(%pNF, %pNF)\n", 3155 name, dev ? &dev->features : &null_features, 3156 skb->sk ? &skb->sk->sk_route_caps : &null_features); 3157 } 3158 3159 /* 3160 * Invalidate hardware checksum when packet is to be mangled, and 3161 * complete checksum manually on outgoing path. 3162 */ 3163 int skb_checksum_help(struct sk_buff *skb) 3164 { 3165 __wsum csum; 3166 int ret = 0, offset; 3167 3168 if (skb->ip_summed == CHECKSUM_COMPLETE) 3169 goto out_set_summed; 3170 3171 if (unlikely(skb_is_gso(skb))) { 3172 skb_warn_bad_offload(skb); 3173 return -EINVAL; 3174 } 3175 3176 /* Before computing a checksum, we should make sure no frag could 3177 * be modified by an external entity : checksum could be wrong. 3178 */ 3179 if (skb_has_shared_frag(skb)) { 3180 ret = __skb_linearize(skb); 3181 if (ret) 3182 goto out; 3183 } 3184 3185 offset = skb_checksum_start_offset(skb); 3186 BUG_ON(offset >= skb_headlen(skb)); 3187 csum = skb_checksum(skb, offset, skb->len - offset, 0); 3188 3189 offset += skb->csum_offset; 3190 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); 3191 3192 ret = skb_ensure_writable(skb, offset + sizeof(__sum16)); 3193 if (ret) 3194 goto out; 3195 3196 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0; 3197 out_set_summed: 3198 skb->ip_summed = CHECKSUM_NONE; 3199 out: 3200 return ret; 3201 } 3202 EXPORT_SYMBOL(skb_checksum_help); 3203 3204 int skb_crc32c_csum_help(struct sk_buff *skb) 3205 { 3206 __le32 crc32c_csum; 3207 int ret = 0, offset, start; 3208 3209 if (skb->ip_summed != CHECKSUM_PARTIAL) 3210 goto out; 3211 3212 if (unlikely(skb_is_gso(skb))) 3213 goto out; 3214 3215 /* Before computing a checksum, we should make sure no frag could 3216 * be modified by an external entity : checksum could be wrong. 3217 */ 3218 if (unlikely(skb_has_shared_frag(skb))) { 3219 ret = __skb_linearize(skb); 3220 if (ret) 3221 goto out; 3222 } 3223 start = skb_checksum_start_offset(skb); 3224 offset = start + offsetof(struct sctphdr, checksum); 3225 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { 3226 ret = -EINVAL; 3227 goto out; 3228 } 3229 3230 ret = skb_ensure_writable(skb, offset + sizeof(__le32)); 3231 if (ret) 3232 goto out; 3233 3234 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start, 3235 skb->len - start, ~(__u32)0, 3236 crc32c_csum_stub)); 3237 *(__le32 *)(skb->data + offset) = crc32c_csum; 3238 skb->ip_summed = CHECKSUM_NONE; 3239 skb->csum_not_inet = 0; 3240 out: 3241 return ret; 3242 } 3243 3244 __be16 skb_network_protocol(struct sk_buff *skb, int *depth) 3245 { 3246 __be16 type = skb->protocol; 3247 3248 /* Tunnel gso handlers can set protocol to ethernet. */ 3249 if (type == htons(ETH_P_TEB)) { 3250 struct ethhdr *eth; 3251 3252 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) 3253 return 0; 3254 3255 eth = (struct ethhdr *)skb->data; 3256 type = eth->h_proto; 3257 } 3258 3259 return __vlan_get_protocol(skb, type, depth); 3260 } 3261 3262 /** 3263 * skb_mac_gso_segment - mac layer segmentation handler. 3264 * @skb: buffer to segment 3265 * @features: features for the output path (see dev->features) 3266 */ 3267 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb, 3268 netdev_features_t features) 3269 { 3270 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); 3271 struct packet_offload *ptype; 3272 int vlan_depth = skb->mac_len; 3273 __be16 type = skb_network_protocol(skb, &vlan_depth); 3274 3275 if (unlikely(!type)) 3276 return ERR_PTR(-EINVAL); 3277 3278 __skb_pull(skb, vlan_depth); 3279 3280 rcu_read_lock(); 3281 list_for_each_entry_rcu(ptype, &offload_base, list) { 3282 if (ptype->type == type && ptype->callbacks.gso_segment) { 3283 segs = ptype->callbacks.gso_segment(skb, features); 3284 break; 3285 } 3286 } 3287 rcu_read_unlock(); 3288 3289 __skb_push(skb, skb->data - skb_mac_header(skb)); 3290 3291 return segs; 3292 } 3293 EXPORT_SYMBOL(skb_mac_gso_segment); 3294 3295 3296 /* openvswitch calls this on rx path, so we need a different check. 3297 */ 3298 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path) 3299 { 3300 if (tx_path) 3301 return skb->ip_summed != CHECKSUM_PARTIAL && 3302 skb->ip_summed != CHECKSUM_UNNECESSARY; 3303 3304 return skb->ip_summed == CHECKSUM_NONE; 3305 } 3306 3307 /** 3308 * __skb_gso_segment - Perform segmentation on skb. 3309 * @skb: buffer to segment 3310 * @features: features for the output path (see dev->features) 3311 * @tx_path: whether it is called in TX path 3312 * 3313 * This function segments the given skb and returns a list of segments. 3314 * 3315 * It may return NULL if the skb requires no segmentation. This is 3316 * only possible when GSO is used for verifying header integrity. 3317 * 3318 * Segmentation preserves SKB_GSO_CB_OFFSET bytes of previous skb cb. 3319 */ 3320 struct sk_buff *__skb_gso_segment(struct sk_buff *skb, 3321 netdev_features_t features, bool tx_path) 3322 { 3323 struct sk_buff *segs; 3324 3325 if (unlikely(skb_needs_check(skb, tx_path))) { 3326 int err; 3327 3328 /* We're going to init ->check field in TCP or UDP header */ 3329 err = skb_cow_head(skb, 0); 3330 if (err < 0) 3331 return ERR_PTR(err); 3332 } 3333 3334 /* Only report GSO partial support if it will enable us to 3335 * support segmentation on this frame without needing additional 3336 * work. 3337 */ 3338 if (features & NETIF_F_GSO_PARTIAL) { 3339 netdev_features_t partial_features = NETIF_F_GSO_ROBUST; 3340 struct net_device *dev = skb->dev; 3341 3342 partial_features |= dev->features & dev->gso_partial_features; 3343 if (!skb_gso_ok(skb, features | partial_features)) 3344 features &= ~NETIF_F_GSO_PARTIAL; 3345 } 3346 3347 BUILD_BUG_ON(SKB_GSO_CB_OFFSET + 3348 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb)); 3349 3350 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb); 3351 SKB_GSO_CB(skb)->encap_level = 0; 3352 3353 skb_reset_mac_header(skb); 3354 skb_reset_mac_len(skb); 3355 3356 segs = skb_mac_gso_segment(skb, features); 3357 3358 if (segs != skb && unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs))) 3359 skb_warn_bad_offload(skb); 3360 3361 return segs; 3362 } 3363 EXPORT_SYMBOL(__skb_gso_segment); 3364 3365 /* Take action when hardware reception checksum errors are detected. */ 3366 #ifdef CONFIG_BUG 3367 static void do_netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3368 { 3369 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>"); 3370 skb_dump(KERN_ERR, skb, true); 3371 dump_stack(); 3372 } 3373 3374 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3375 { 3376 DO_ONCE_LITE(do_netdev_rx_csum_fault, dev, skb); 3377 } 3378 EXPORT_SYMBOL(netdev_rx_csum_fault); 3379 #endif 3380 3381 /* XXX: check that highmem exists at all on the given machine. */ 3382 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) 3383 { 3384 #ifdef CONFIG_HIGHMEM 3385 int i; 3386 3387 if (!(dev->features & NETIF_F_HIGHDMA)) { 3388 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 3389 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 3390 3391 if (PageHighMem(skb_frag_page(frag))) 3392 return 1; 3393 } 3394 } 3395 #endif 3396 return 0; 3397 } 3398 3399 /* If MPLS offload request, verify we are testing hardware MPLS features 3400 * instead of standard features for the netdev. 3401 */ 3402 #if IS_ENABLED(CONFIG_NET_MPLS_GSO) 3403 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3404 netdev_features_t features, 3405 __be16 type) 3406 { 3407 if (eth_p_mpls(type)) 3408 features &= skb->dev->mpls_features; 3409 3410 return features; 3411 } 3412 #else 3413 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3414 netdev_features_t features, 3415 __be16 type) 3416 { 3417 return features; 3418 } 3419 #endif 3420 3421 static netdev_features_t harmonize_features(struct sk_buff *skb, 3422 netdev_features_t features) 3423 { 3424 __be16 type; 3425 3426 type = skb_network_protocol(skb, NULL); 3427 features = net_mpls_features(skb, features, type); 3428 3429 if (skb->ip_summed != CHECKSUM_NONE && 3430 !can_checksum_protocol(features, type)) { 3431 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); 3432 } 3433 if (illegal_highdma(skb->dev, skb)) 3434 features &= ~NETIF_F_SG; 3435 3436 return features; 3437 } 3438 3439 netdev_features_t passthru_features_check(struct sk_buff *skb, 3440 struct net_device *dev, 3441 netdev_features_t features) 3442 { 3443 return features; 3444 } 3445 EXPORT_SYMBOL(passthru_features_check); 3446 3447 static netdev_features_t dflt_features_check(struct sk_buff *skb, 3448 struct net_device *dev, 3449 netdev_features_t features) 3450 { 3451 return vlan_features_check(skb, features); 3452 } 3453 3454 static netdev_features_t gso_features_check(const struct sk_buff *skb, 3455 struct net_device *dev, 3456 netdev_features_t features) 3457 { 3458 u16 gso_segs = skb_shinfo(skb)->gso_segs; 3459 3460 if (gso_segs > dev->gso_max_segs) 3461 return features & ~NETIF_F_GSO_MASK; 3462 3463 if (!skb_shinfo(skb)->gso_type) { 3464 skb_warn_bad_offload(skb); 3465 return features & ~NETIF_F_GSO_MASK; 3466 } 3467 3468 /* Support for GSO partial features requires software 3469 * intervention before we can actually process the packets 3470 * so we need to strip support for any partial features now 3471 * and we can pull them back in after we have partially 3472 * segmented the frame. 3473 */ 3474 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)) 3475 features &= ~dev->gso_partial_features; 3476 3477 /* Make sure to clear the IPv4 ID mangling feature if the 3478 * IPv4 header has the potential to be fragmented. 3479 */ 3480 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { 3481 struct iphdr *iph = skb->encapsulation ? 3482 inner_ip_hdr(skb) : ip_hdr(skb); 3483 3484 if (!(iph->frag_off & htons(IP_DF))) 3485 features &= ~NETIF_F_TSO_MANGLEID; 3486 } 3487 3488 return features; 3489 } 3490 3491 netdev_features_t netif_skb_features(struct sk_buff *skb) 3492 { 3493 struct net_device *dev = skb->dev; 3494 netdev_features_t features = dev->features; 3495 3496 if (skb_is_gso(skb)) 3497 features = gso_features_check(skb, dev, features); 3498 3499 /* If encapsulation offload request, verify we are testing 3500 * hardware encapsulation features instead of standard 3501 * features for the netdev 3502 */ 3503 if (skb->encapsulation) 3504 features &= dev->hw_enc_features; 3505 3506 if (skb_vlan_tagged(skb)) 3507 features = netdev_intersect_features(features, 3508 dev->vlan_features | 3509 NETIF_F_HW_VLAN_CTAG_TX | 3510 NETIF_F_HW_VLAN_STAG_TX); 3511 3512 if (dev->netdev_ops->ndo_features_check) 3513 features &= dev->netdev_ops->ndo_features_check(skb, dev, 3514 features); 3515 else 3516 features &= dflt_features_check(skb, dev, features); 3517 3518 return harmonize_features(skb, features); 3519 } 3520 EXPORT_SYMBOL(netif_skb_features); 3521 3522 static int xmit_one(struct sk_buff *skb, struct net_device *dev, 3523 struct netdev_queue *txq, bool more) 3524 { 3525 unsigned int len; 3526 int rc; 3527 3528 if (dev_nit_active(dev)) 3529 dev_queue_xmit_nit(skb, dev); 3530 3531 len = skb->len; 3532 PRANDOM_ADD_NOISE(skb, dev, txq, len + jiffies); 3533 trace_net_dev_start_xmit(skb, dev); 3534 rc = netdev_start_xmit(skb, dev, txq, more); 3535 trace_net_dev_xmit(skb, rc, dev, len); 3536 3537 return rc; 3538 } 3539 3540 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, 3541 struct netdev_queue *txq, int *ret) 3542 { 3543 struct sk_buff *skb = first; 3544 int rc = NETDEV_TX_OK; 3545 3546 while (skb) { 3547 struct sk_buff *next = skb->next; 3548 3549 skb_mark_not_on_list(skb); 3550 rc = xmit_one(skb, dev, txq, next != NULL); 3551 if (unlikely(!dev_xmit_complete(rc))) { 3552 skb->next = next; 3553 goto out; 3554 } 3555 3556 skb = next; 3557 if (netif_tx_queue_stopped(txq) && skb) { 3558 rc = NETDEV_TX_BUSY; 3559 break; 3560 } 3561 } 3562 3563 out: 3564 *ret = rc; 3565 return skb; 3566 } 3567 3568 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, 3569 netdev_features_t features) 3570 { 3571 if (skb_vlan_tag_present(skb) && 3572 !vlan_hw_offload_capable(features, skb->vlan_proto)) 3573 skb = __vlan_hwaccel_push_inside(skb); 3574 return skb; 3575 } 3576 3577 int skb_csum_hwoffload_help(struct sk_buff *skb, 3578 const netdev_features_t features) 3579 { 3580 if (unlikely(skb_csum_is_sctp(skb))) 3581 return !!(features & NETIF_F_SCTP_CRC) ? 0 : 3582 skb_crc32c_csum_help(skb); 3583 3584 if (features & NETIF_F_HW_CSUM) 3585 return 0; 3586 3587 if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) { 3588 switch (skb->csum_offset) { 3589 case offsetof(struct tcphdr, check): 3590 case offsetof(struct udphdr, check): 3591 return 0; 3592 } 3593 } 3594 3595 return skb_checksum_help(skb); 3596 } 3597 EXPORT_SYMBOL(skb_csum_hwoffload_help); 3598 3599 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again) 3600 { 3601 netdev_features_t features; 3602 3603 features = netif_skb_features(skb); 3604 skb = validate_xmit_vlan(skb, features); 3605 if (unlikely(!skb)) 3606 goto out_null; 3607 3608 skb = sk_validate_xmit_skb(skb, dev); 3609 if (unlikely(!skb)) 3610 goto out_null; 3611 3612 if (netif_needs_gso(skb, features)) { 3613 struct sk_buff *segs; 3614 3615 segs = skb_gso_segment(skb, features); 3616 if (IS_ERR(segs)) { 3617 goto out_kfree_skb; 3618 } else if (segs) { 3619 consume_skb(skb); 3620 skb = segs; 3621 } 3622 } else { 3623 if (skb_needs_linearize(skb, features) && 3624 __skb_linearize(skb)) 3625 goto out_kfree_skb; 3626 3627 /* If packet is not checksummed and device does not 3628 * support checksumming for this protocol, complete 3629 * checksumming here. 3630 */ 3631 if (skb->ip_summed == CHECKSUM_PARTIAL) { 3632 if (skb->encapsulation) 3633 skb_set_inner_transport_header(skb, 3634 skb_checksum_start_offset(skb)); 3635 else 3636 skb_set_transport_header(skb, 3637 skb_checksum_start_offset(skb)); 3638 if (skb_csum_hwoffload_help(skb, features)) 3639 goto out_kfree_skb; 3640 } 3641 } 3642 3643 skb = validate_xmit_xfrm(skb, features, again); 3644 3645 return skb; 3646 3647 out_kfree_skb: 3648 kfree_skb(skb); 3649 out_null: 3650 atomic_long_inc(&dev->tx_dropped); 3651 return NULL; 3652 } 3653 3654 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again) 3655 { 3656 struct sk_buff *next, *head = NULL, *tail; 3657 3658 for (; skb != NULL; skb = next) { 3659 next = skb->next; 3660 skb_mark_not_on_list(skb); 3661 3662 /* in case skb wont be segmented, point to itself */ 3663 skb->prev = skb; 3664 3665 skb = validate_xmit_skb(skb, dev, again); 3666 if (!skb) 3667 continue; 3668 3669 if (!head) 3670 head = skb; 3671 else 3672 tail->next = skb; 3673 /* If skb was segmented, skb->prev points to 3674 * the last segment. If not, it still contains skb. 3675 */ 3676 tail = skb->prev; 3677 } 3678 return head; 3679 } 3680 EXPORT_SYMBOL_GPL(validate_xmit_skb_list); 3681 3682 static void qdisc_pkt_len_init(struct sk_buff *skb) 3683 { 3684 const struct skb_shared_info *shinfo = skb_shinfo(skb); 3685 3686 qdisc_skb_cb(skb)->pkt_len = skb->len; 3687 3688 /* To get more precise estimation of bytes sent on wire, 3689 * we add to pkt_len the headers size of all segments 3690 */ 3691 if (shinfo->gso_size && skb_transport_header_was_set(skb)) { 3692 unsigned int hdr_len; 3693 u16 gso_segs = shinfo->gso_segs; 3694 3695 /* mac layer + network layer */ 3696 hdr_len = skb_transport_header(skb) - skb_mac_header(skb); 3697 3698 /* + transport layer */ 3699 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { 3700 const struct tcphdr *th; 3701 struct tcphdr _tcphdr; 3702 3703 th = skb_header_pointer(skb, skb_transport_offset(skb), 3704 sizeof(_tcphdr), &_tcphdr); 3705 if (likely(th)) 3706 hdr_len += __tcp_hdrlen(th); 3707 } else { 3708 struct udphdr _udphdr; 3709 3710 if (skb_header_pointer(skb, skb_transport_offset(skb), 3711 sizeof(_udphdr), &_udphdr)) 3712 hdr_len += sizeof(struct udphdr); 3713 } 3714 3715 if (shinfo->gso_type & SKB_GSO_DODGY) 3716 gso_segs = DIV_ROUND_UP(skb->len - hdr_len, 3717 shinfo->gso_size); 3718 3719 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; 3720 } 3721 } 3722 3723 static int dev_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *q, 3724 struct sk_buff **to_free, 3725 struct netdev_queue *txq) 3726 { 3727 int rc; 3728 3729 rc = q->enqueue(skb, q, to_free) & NET_XMIT_MASK; 3730 if (rc == NET_XMIT_SUCCESS) 3731 trace_qdisc_enqueue(q, txq, skb); 3732 return rc; 3733 } 3734 3735 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, 3736 struct net_device *dev, 3737 struct netdev_queue *txq) 3738 { 3739 spinlock_t *root_lock = qdisc_lock(q); 3740 struct sk_buff *to_free = NULL; 3741 bool contended; 3742 int rc; 3743 3744 qdisc_calculate_pkt_len(skb, q); 3745 3746 if (q->flags & TCQ_F_NOLOCK) { 3747 if (q->flags & TCQ_F_CAN_BYPASS && nolock_qdisc_is_empty(q) && 3748 qdisc_run_begin(q)) { 3749 /* Retest nolock_qdisc_is_empty() within the protection 3750 * of q->seqlock to protect from racing with requeuing. 3751 */ 3752 if (unlikely(!nolock_qdisc_is_empty(q))) { 3753 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3754 __qdisc_run(q); 3755 qdisc_run_end(q); 3756 3757 goto no_lock_out; 3758 } 3759 3760 qdisc_bstats_cpu_update(q, skb); 3761 if (sch_direct_xmit(skb, q, dev, txq, NULL, true) && 3762 !nolock_qdisc_is_empty(q)) 3763 __qdisc_run(q); 3764 3765 qdisc_run_end(q); 3766 return NET_XMIT_SUCCESS; 3767 } 3768 3769 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3770 qdisc_run(q); 3771 3772 no_lock_out: 3773 if (unlikely(to_free)) 3774 kfree_skb_list(to_free); 3775 return rc; 3776 } 3777 3778 /* 3779 * Heuristic to force contended enqueues to serialize on a 3780 * separate lock before trying to get qdisc main lock. 3781 * This permits qdisc->running owner to get the lock more 3782 * often and dequeue packets faster. 3783 */ 3784 contended = qdisc_is_running(q); 3785 if (unlikely(contended)) 3786 spin_lock(&q->busylock); 3787 3788 spin_lock(root_lock); 3789 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { 3790 __qdisc_drop(skb, &to_free); 3791 rc = NET_XMIT_DROP; 3792 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && 3793 qdisc_run_begin(q)) { 3794 /* 3795 * This is a work-conserving queue; there are no old skbs 3796 * waiting to be sent out; and the qdisc is not running - 3797 * xmit the skb directly. 3798 */ 3799 3800 qdisc_bstats_update(q, skb); 3801 3802 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { 3803 if (unlikely(contended)) { 3804 spin_unlock(&q->busylock); 3805 contended = false; 3806 } 3807 __qdisc_run(q); 3808 } 3809 3810 qdisc_run_end(q); 3811 rc = NET_XMIT_SUCCESS; 3812 } else { 3813 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3814 if (qdisc_run_begin(q)) { 3815 if (unlikely(contended)) { 3816 spin_unlock(&q->busylock); 3817 contended = false; 3818 } 3819 __qdisc_run(q); 3820 qdisc_run_end(q); 3821 } 3822 } 3823 spin_unlock(root_lock); 3824 if (unlikely(to_free)) 3825 kfree_skb_list(to_free); 3826 if (unlikely(contended)) 3827 spin_unlock(&q->busylock); 3828 return rc; 3829 } 3830 3831 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) 3832 static void skb_update_prio(struct sk_buff *skb) 3833 { 3834 const struct netprio_map *map; 3835 const struct sock *sk; 3836 unsigned int prioidx; 3837 3838 if (skb->priority) 3839 return; 3840 map = rcu_dereference_bh(skb->dev->priomap); 3841 if (!map) 3842 return; 3843 sk = skb_to_full_sk(skb); 3844 if (!sk) 3845 return; 3846 3847 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data); 3848 3849 if (prioidx < map->priomap_len) 3850 skb->priority = map->priomap[prioidx]; 3851 } 3852 #else 3853 #define skb_update_prio(skb) 3854 #endif 3855 3856 /** 3857 * dev_loopback_xmit - loop back @skb 3858 * @net: network namespace this loopback is happening in 3859 * @sk: sk needed to be a netfilter okfn 3860 * @skb: buffer to transmit 3861 */ 3862 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) 3863 { 3864 skb_reset_mac_header(skb); 3865 __skb_pull(skb, skb_network_offset(skb)); 3866 skb->pkt_type = PACKET_LOOPBACK; 3867 skb->ip_summed = CHECKSUM_UNNECESSARY; 3868 WARN_ON(!skb_dst(skb)); 3869 skb_dst_force(skb); 3870 netif_rx_ni(skb); 3871 return 0; 3872 } 3873 EXPORT_SYMBOL(dev_loopback_xmit); 3874 3875 #ifdef CONFIG_NET_EGRESS 3876 static struct sk_buff * 3877 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 3878 { 3879 struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress); 3880 struct tcf_result cl_res; 3881 3882 if (!miniq) 3883 return skb; 3884 3885 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */ 3886 qdisc_skb_cb(skb)->mru = 0; 3887 qdisc_skb_cb(skb)->post_ct = false; 3888 mini_qdisc_bstats_cpu_update(miniq, skb); 3889 3890 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) { 3891 case TC_ACT_OK: 3892 case TC_ACT_RECLASSIFY: 3893 skb->tc_index = TC_H_MIN(cl_res.classid); 3894 break; 3895 case TC_ACT_SHOT: 3896 mini_qdisc_qstats_cpu_drop(miniq); 3897 *ret = NET_XMIT_DROP; 3898 kfree_skb(skb); 3899 return NULL; 3900 case TC_ACT_STOLEN: 3901 case TC_ACT_QUEUED: 3902 case TC_ACT_TRAP: 3903 *ret = NET_XMIT_SUCCESS; 3904 consume_skb(skb); 3905 return NULL; 3906 case TC_ACT_REDIRECT: 3907 /* No need to push/pop skb's mac_header here on egress! */ 3908 skb_do_redirect(skb); 3909 *ret = NET_XMIT_SUCCESS; 3910 return NULL; 3911 default: 3912 break; 3913 } 3914 3915 return skb; 3916 } 3917 #endif /* CONFIG_NET_EGRESS */ 3918 3919 #ifdef CONFIG_XPS 3920 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb, 3921 struct xps_dev_maps *dev_maps, unsigned int tci) 3922 { 3923 int tc = netdev_get_prio_tc_map(dev, skb->priority); 3924 struct xps_map *map; 3925 int queue_index = -1; 3926 3927 if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids) 3928 return queue_index; 3929 3930 tci *= dev_maps->num_tc; 3931 tci += tc; 3932 3933 map = rcu_dereference(dev_maps->attr_map[tci]); 3934 if (map) { 3935 if (map->len == 1) 3936 queue_index = map->queues[0]; 3937 else 3938 queue_index = map->queues[reciprocal_scale( 3939 skb_get_hash(skb), map->len)]; 3940 if (unlikely(queue_index >= dev->real_num_tx_queues)) 3941 queue_index = -1; 3942 } 3943 return queue_index; 3944 } 3945 #endif 3946 3947 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev, 3948 struct sk_buff *skb) 3949 { 3950 #ifdef CONFIG_XPS 3951 struct xps_dev_maps *dev_maps; 3952 struct sock *sk = skb->sk; 3953 int queue_index = -1; 3954 3955 if (!static_key_false(&xps_needed)) 3956 return -1; 3957 3958 rcu_read_lock(); 3959 if (!static_key_false(&xps_rxqs_needed)) 3960 goto get_cpus_map; 3961 3962 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]); 3963 if (dev_maps) { 3964 int tci = sk_rx_queue_get(sk); 3965 3966 if (tci >= 0) 3967 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 3968 tci); 3969 } 3970 3971 get_cpus_map: 3972 if (queue_index < 0) { 3973 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]); 3974 if (dev_maps) { 3975 unsigned int tci = skb->sender_cpu - 1; 3976 3977 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 3978 tci); 3979 } 3980 } 3981 rcu_read_unlock(); 3982 3983 return queue_index; 3984 #else 3985 return -1; 3986 #endif 3987 } 3988 3989 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb, 3990 struct net_device *sb_dev) 3991 { 3992 return 0; 3993 } 3994 EXPORT_SYMBOL(dev_pick_tx_zero); 3995 3996 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb, 3997 struct net_device *sb_dev) 3998 { 3999 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues; 4000 } 4001 EXPORT_SYMBOL(dev_pick_tx_cpu_id); 4002 4003 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb, 4004 struct net_device *sb_dev) 4005 { 4006 struct sock *sk = skb->sk; 4007 int queue_index = sk_tx_queue_get(sk); 4008 4009 sb_dev = sb_dev ? : dev; 4010 4011 if (queue_index < 0 || skb->ooo_okay || 4012 queue_index >= dev->real_num_tx_queues) { 4013 int new_index = get_xps_queue(dev, sb_dev, skb); 4014 4015 if (new_index < 0) 4016 new_index = skb_tx_hash(dev, sb_dev, skb); 4017 4018 if (queue_index != new_index && sk && 4019 sk_fullsock(sk) && 4020 rcu_access_pointer(sk->sk_dst_cache)) 4021 sk_tx_queue_set(sk, new_index); 4022 4023 queue_index = new_index; 4024 } 4025 4026 return queue_index; 4027 } 4028 EXPORT_SYMBOL(netdev_pick_tx); 4029 4030 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev, 4031 struct sk_buff *skb, 4032 struct net_device *sb_dev) 4033 { 4034 int queue_index = 0; 4035 4036 #ifdef CONFIG_XPS 4037 u32 sender_cpu = skb->sender_cpu - 1; 4038 4039 if (sender_cpu >= (u32)NR_CPUS) 4040 skb->sender_cpu = raw_smp_processor_id() + 1; 4041 #endif 4042 4043 if (dev->real_num_tx_queues != 1) { 4044 const struct net_device_ops *ops = dev->netdev_ops; 4045 4046 if (ops->ndo_select_queue) 4047 queue_index = ops->ndo_select_queue(dev, skb, sb_dev); 4048 else 4049 queue_index = netdev_pick_tx(dev, skb, sb_dev); 4050 4051 queue_index = netdev_cap_txqueue(dev, queue_index); 4052 } 4053 4054 skb_set_queue_mapping(skb, queue_index); 4055 return netdev_get_tx_queue(dev, queue_index); 4056 } 4057 4058 /** 4059 * __dev_queue_xmit - transmit a buffer 4060 * @skb: buffer to transmit 4061 * @sb_dev: suboordinate device used for L2 forwarding offload 4062 * 4063 * Queue a buffer for transmission to a network device. The caller must 4064 * have set the device and priority and built the buffer before calling 4065 * this function. The function can be called from an interrupt. 4066 * 4067 * A negative errno code is returned on a failure. A success does not 4068 * guarantee the frame will be transmitted as it may be dropped due 4069 * to congestion or traffic shaping. 4070 * 4071 * ----------------------------------------------------------------------------------- 4072 * I notice this method can also return errors from the queue disciplines, 4073 * including NET_XMIT_DROP, which is a positive value. So, errors can also 4074 * be positive. 4075 * 4076 * Regardless of the return value, the skb is consumed, so it is currently 4077 * difficult to retry a send to this method. (You can bump the ref count 4078 * before sending to hold a reference for retry if you are careful.) 4079 * 4080 * When calling this method, interrupts MUST be enabled. This is because 4081 * the BH enable code must have IRQs enabled so that it will not deadlock. 4082 * --BLG 4083 */ 4084 static int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev) 4085 { 4086 struct net_device *dev = skb->dev; 4087 struct netdev_queue *txq; 4088 struct Qdisc *q; 4089 int rc = -ENOMEM; 4090 bool again = false; 4091 4092 skb_reset_mac_header(skb); 4093 4094 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) 4095 __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED); 4096 4097 /* Disable soft irqs for various locks below. Also 4098 * stops preemption for RCU. 4099 */ 4100 rcu_read_lock_bh(); 4101 4102 skb_update_prio(skb); 4103 4104 qdisc_pkt_len_init(skb); 4105 #ifdef CONFIG_NET_CLS_ACT 4106 skb->tc_at_ingress = 0; 4107 # ifdef CONFIG_NET_EGRESS 4108 if (static_branch_unlikely(&egress_needed_key)) { 4109 skb = sch_handle_egress(skb, &rc, dev); 4110 if (!skb) 4111 goto out; 4112 } 4113 # endif 4114 #endif 4115 /* If device/qdisc don't need skb->dst, release it right now while 4116 * its hot in this cpu cache. 4117 */ 4118 if (dev->priv_flags & IFF_XMIT_DST_RELEASE) 4119 skb_dst_drop(skb); 4120 else 4121 skb_dst_force(skb); 4122 4123 txq = netdev_core_pick_tx(dev, skb, sb_dev); 4124 q = rcu_dereference_bh(txq->qdisc); 4125 4126 trace_net_dev_queue(skb); 4127 if (q->enqueue) { 4128 rc = __dev_xmit_skb(skb, q, dev, txq); 4129 goto out; 4130 } 4131 4132 /* The device has no queue. Common case for software devices: 4133 * loopback, all the sorts of tunnels... 4134 4135 * Really, it is unlikely that netif_tx_lock protection is necessary 4136 * here. (f.e. loopback and IP tunnels are clean ignoring statistics 4137 * counters.) 4138 * However, it is possible, that they rely on protection 4139 * made by us here. 4140 4141 * Check this and shot the lock. It is not prone from deadlocks. 4142 *Either shot noqueue qdisc, it is even simpler 8) 4143 */ 4144 if (dev->flags & IFF_UP) { 4145 int cpu = smp_processor_id(); /* ok because BHs are off */ 4146 4147 if (txq->xmit_lock_owner != cpu) { 4148 if (dev_xmit_recursion()) 4149 goto recursion_alert; 4150 4151 skb = validate_xmit_skb(skb, dev, &again); 4152 if (!skb) 4153 goto out; 4154 4155 PRANDOM_ADD_NOISE(skb, dev, txq, jiffies); 4156 HARD_TX_LOCK(dev, txq, cpu); 4157 4158 if (!netif_xmit_stopped(txq)) { 4159 dev_xmit_recursion_inc(); 4160 skb = dev_hard_start_xmit(skb, dev, txq, &rc); 4161 dev_xmit_recursion_dec(); 4162 if (dev_xmit_complete(rc)) { 4163 HARD_TX_UNLOCK(dev, txq); 4164 goto out; 4165 } 4166 } 4167 HARD_TX_UNLOCK(dev, txq); 4168 net_crit_ratelimited("Virtual device %s asks to queue packet!\n", 4169 dev->name); 4170 } else { 4171 /* Recursion is detected! It is possible, 4172 * unfortunately 4173 */ 4174 recursion_alert: 4175 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", 4176 dev->name); 4177 } 4178 } 4179 4180 rc = -ENETDOWN; 4181 rcu_read_unlock_bh(); 4182 4183 atomic_long_inc(&dev->tx_dropped); 4184 kfree_skb_list(skb); 4185 return rc; 4186 out: 4187 rcu_read_unlock_bh(); 4188 return rc; 4189 } 4190 4191 int dev_queue_xmit(struct sk_buff *skb) 4192 { 4193 return __dev_queue_xmit(skb, NULL); 4194 } 4195 EXPORT_SYMBOL(dev_queue_xmit); 4196 4197 int dev_queue_xmit_accel(struct sk_buff *skb, struct net_device *sb_dev) 4198 { 4199 return __dev_queue_xmit(skb, sb_dev); 4200 } 4201 EXPORT_SYMBOL(dev_queue_xmit_accel); 4202 4203 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id) 4204 { 4205 struct net_device *dev = skb->dev; 4206 struct sk_buff *orig_skb = skb; 4207 struct netdev_queue *txq; 4208 int ret = NETDEV_TX_BUSY; 4209 bool again = false; 4210 4211 if (unlikely(!netif_running(dev) || 4212 !netif_carrier_ok(dev))) 4213 goto drop; 4214 4215 skb = validate_xmit_skb_list(skb, dev, &again); 4216 if (skb != orig_skb) 4217 goto drop; 4218 4219 skb_set_queue_mapping(skb, queue_id); 4220 txq = skb_get_tx_queue(dev, skb); 4221 PRANDOM_ADD_NOISE(skb, dev, txq, jiffies); 4222 4223 local_bh_disable(); 4224 4225 dev_xmit_recursion_inc(); 4226 HARD_TX_LOCK(dev, txq, smp_processor_id()); 4227 if (!netif_xmit_frozen_or_drv_stopped(txq)) 4228 ret = netdev_start_xmit(skb, dev, txq, false); 4229 HARD_TX_UNLOCK(dev, txq); 4230 dev_xmit_recursion_dec(); 4231 4232 local_bh_enable(); 4233 return ret; 4234 drop: 4235 atomic_long_inc(&dev->tx_dropped); 4236 kfree_skb_list(skb); 4237 return NET_XMIT_DROP; 4238 } 4239 EXPORT_SYMBOL(__dev_direct_xmit); 4240 4241 /************************************************************************* 4242 * Receiver routines 4243 *************************************************************************/ 4244 4245 int netdev_max_backlog __read_mostly = 1000; 4246 EXPORT_SYMBOL(netdev_max_backlog); 4247 4248 int netdev_tstamp_prequeue __read_mostly = 1; 4249 int netdev_budget __read_mostly = 300; 4250 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */ 4251 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ; 4252 int weight_p __read_mostly = 64; /* old backlog weight */ 4253 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ 4254 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ 4255 int dev_rx_weight __read_mostly = 64; 4256 int dev_tx_weight __read_mostly = 64; 4257 /* Maximum number of GRO_NORMAL skbs to batch up for list-RX */ 4258 int gro_normal_batch __read_mostly = 8; 4259 4260 /* Called with irq disabled */ 4261 static inline void ____napi_schedule(struct softnet_data *sd, 4262 struct napi_struct *napi) 4263 { 4264 struct task_struct *thread; 4265 4266 if (test_bit(NAPI_STATE_THREADED, &napi->state)) { 4267 /* Paired with smp_mb__before_atomic() in 4268 * napi_enable()/dev_set_threaded(). 4269 * Use READ_ONCE() to guarantee a complete 4270 * read on napi->thread. Only call 4271 * wake_up_process() when it's not NULL. 4272 */ 4273 thread = READ_ONCE(napi->thread); 4274 if (thread) { 4275 /* Avoid doing set_bit() if the thread is in 4276 * INTERRUPTIBLE state, cause napi_thread_wait() 4277 * makes sure to proceed with napi polling 4278 * if the thread is explicitly woken from here. 4279 */ 4280 if (READ_ONCE(thread->__state) != TASK_INTERRUPTIBLE) 4281 set_bit(NAPI_STATE_SCHED_THREADED, &napi->state); 4282 wake_up_process(thread); 4283 return; 4284 } 4285 } 4286 4287 list_add_tail(&napi->poll_list, &sd->poll_list); 4288 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4289 } 4290 4291 #ifdef CONFIG_RPS 4292 4293 /* One global table that all flow-based protocols share. */ 4294 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; 4295 EXPORT_SYMBOL(rps_sock_flow_table); 4296 u32 rps_cpu_mask __read_mostly; 4297 EXPORT_SYMBOL(rps_cpu_mask); 4298 4299 struct static_key_false rps_needed __read_mostly; 4300 EXPORT_SYMBOL(rps_needed); 4301 struct static_key_false rfs_needed __read_mostly; 4302 EXPORT_SYMBOL(rfs_needed); 4303 4304 static struct rps_dev_flow * 4305 set_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4306 struct rps_dev_flow *rflow, u16 next_cpu) 4307 { 4308 if (next_cpu < nr_cpu_ids) { 4309 #ifdef CONFIG_RFS_ACCEL 4310 struct netdev_rx_queue *rxqueue; 4311 struct rps_dev_flow_table *flow_table; 4312 struct rps_dev_flow *old_rflow; 4313 u32 flow_id; 4314 u16 rxq_index; 4315 int rc; 4316 4317 /* Should we steer this flow to a different hardware queue? */ 4318 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || 4319 !(dev->features & NETIF_F_NTUPLE)) 4320 goto out; 4321 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); 4322 if (rxq_index == skb_get_rx_queue(skb)) 4323 goto out; 4324 4325 rxqueue = dev->_rx + rxq_index; 4326 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4327 if (!flow_table) 4328 goto out; 4329 flow_id = skb_get_hash(skb) & flow_table->mask; 4330 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, 4331 rxq_index, flow_id); 4332 if (rc < 0) 4333 goto out; 4334 old_rflow = rflow; 4335 rflow = &flow_table->flows[flow_id]; 4336 rflow->filter = rc; 4337 if (old_rflow->filter == rflow->filter) 4338 old_rflow->filter = RPS_NO_FILTER; 4339 out: 4340 #endif 4341 rflow->last_qtail = 4342 per_cpu(softnet_data, next_cpu).input_queue_head; 4343 } 4344 4345 rflow->cpu = next_cpu; 4346 return rflow; 4347 } 4348 4349 /* 4350 * get_rps_cpu is called from netif_receive_skb and returns the target 4351 * CPU from the RPS map of the receiving queue for a given skb. 4352 * rcu_read_lock must be held on entry. 4353 */ 4354 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4355 struct rps_dev_flow **rflowp) 4356 { 4357 const struct rps_sock_flow_table *sock_flow_table; 4358 struct netdev_rx_queue *rxqueue = dev->_rx; 4359 struct rps_dev_flow_table *flow_table; 4360 struct rps_map *map; 4361 int cpu = -1; 4362 u32 tcpu; 4363 u32 hash; 4364 4365 if (skb_rx_queue_recorded(skb)) { 4366 u16 index = skb_get_rx_queue(skb); 4367 4368 if (unlikely(index >= dev->real_num_rx_queues)) { 4369 WARN_ONCE(dev->real_num_rx_queues > 1, 4370 "%s received packet on queue %u, but number " 4371 "of RX queues is %u\n", 4372 dev->name, index, dev->real_num_rx_queues); 4373 goto done; 4374 } 4375 rxqueue += index; 4376 } 4377 4378 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ 4379 4380 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4381 map = rcu_dereference(rxqueue->rps_map); 4382 if (!flow_table && !map) 4383 goto done; 4384 4385 skb_reset_network_header(skb); 4386 hash = skb_get_hash(skb); 4387 if (!hash) 4388 goto done; 4389 4390 sock_flow_table = rcu_dereference(rps_sock_flow_table); 4391 if (flow_table && sock_flow_table) { 4392 struct rps_dev_flow *rflow; 4393 u32 next_cpu; 4394 u32 ident; 4395 4396 /* First check into global flow table if there is a match */ 4397 ident = sock_flow_table->ents[hash & sock_flow_table->mask]; 4398 if ((ident ^ hash) & ~rps_cpu_mask) 4399 goto try_rps; 4400 4401 next_cpu = ident & rps_cpu_mask; 4402 4403 /* OK, now we know there is a match, 4404 * we can look at the local (per receive queue) flow table 4405 */ 4406 rflow = &flow_table->flows[hash & flow_table->mask]; 4407 tcpu = rflow->cpu; 4408 4409 /* 4410 * If the desired CPU (where last recvmsg was done) is 4411 * different from current CPU (one in the rx-queue flow 4412 * table entry), switch if one of the following holds: 4413 * - Current CPU is unset (>= nr_cpu_ids). 4414 * - Current CPU is offline. 4415 * - The current CPU's queue tail has advanced beyond the 4416 * last packet that was enqueued using this table entry. 4417 * This guarantees that all previous packets for the flow 4418 * have been dequeued, thus preserving in order delivery. 4419 */ 4420 if (unlikely(tcpu != next_cpu) && 4421 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || 4422 ((int)(per_cpu(softnet_data, tcpu).input_queue_head - 4423 rflow->last_qtail)) >= 0)) { 4424 tcpu = next_cpu; 4425 rflow = set_rps_cpu(dev, skb, rflow, next_cpu); 4426 } 4427 4428 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { 4429 *rflowp = rflow; 4430 cpu = tcpu; 4431 goto done; 4432 } 4433 } 4434 4435 try_rps: 4436 4437 if (map) { 4438 tcpu = map->cpus[reciprocal_scale(hash, map->len)]; 4439 if (cpu_online(tcpu)) { 4440 cpu = tcpu; 4441 goto done; 4442 } 4443 } 4444 4445 done: 4446 return cpu; 4447 } 4448 4449 #ifdef CONFIG_RFS_ACCEL 4450 4451 /** 4452 * rps_may_expire_flow - check whether an RFS hardware filter may be removed 4453 * @dev: Device on which the filter was set 4454 * @rxq_index: RX queue index 4455 * @flow_id: Flow ID passed to ndo_rx_flow_steer() 4456 * @filter_id: Filter ID returned by ndo_rx_flow_steer() 4457 * 4458 * Drivers that implement ndo_rx_flow_steer() should periodically call 4459 * this function for each installed filter and remove the filters for 4460 * which it returns %true. 4461 */ 4462 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, 4463 u32 flow_id, u16 filter_id) 4464 { 4465 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; 4466 struct rps_dev_flow_table *flow_table; 4467 struct rps_dev_flow *rflow; 4468 bool expire = true; 4469 unsigned int cpu; 4470 4471 rcu_read_lock(); 4472 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4473 if (flow_table && flow_id <= flow_table->mask) { 4474 rflow = &flow_table->flows[flow_id]; 4475 cpu = READ_ONCE(rflow->cpu); 4476 if (rflow->filter == filter_id && cpu < nr_cpu_ids && 4477 ((int)(per_cpu(softnet_data, cpu).input_queue_head - 4478 rflow->last_qtail) < 4479 (int)(10 * flow_table->mask))) 4480 expire = false; 4481 } 4482 rcu_read_unlock(); 4483 return expire; 4484 } 4485 EXPORT_SYMBOL(rps_may_expire_flow); 4486 4487 #endif /* CONFIG_RFS_ACCEL */ 4488 4489 /* Called from hardirq (IPI) context */ 4490 static void rps_trigger_softirq(void *data) 4491 { 4492 struct softnet_data *sd = data; 4493 4494 ____napi_schedule(sd, &sd->backlog); 4495 sd->received_rps++; 4496 } 4497 4498 #endif /* CONFIG_RPS */ 4499 4500 /* 4501 * Check if this softnet_data structure is another cpu one 4502 * If yes, queue it to our IPI list and return 1 4503 * If no, return 0 4504 */ 4505 static int rps_ipi_queued(struct softnet_data *sd) 4506 { 4507 #ifdef CONFIG_RPS 4508 struct softnet_data *mysd = this_cpu_ptr(&softnet_data); 4509 4510 if (sd != mysd) { 4511 sd->rps_ipi_next = mysd->rps_ipi_list; 4512 mysd->rps_ipi_list = sd; 4513 4514 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4515 return 1; 4516 } 4517 #endif /* CONFIG_RPS */ 4518 return 0; 4519 } 4520 4521 #ifdef CONFIG_NET_FLOW_LIMIT 4522 int netdev_flow_limit_table_len __read_mostly = (1 << 12); 4523 #endif 4524 4525 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) 4526 { 4527 #ifdef CONFIG_NET_FLOW_LIMIT 4528 struct sd_flow_limit *fl; 4529 struct softnet_data *sd; 4530 unsigned int old_flow, new_flow; 4531 4532 if (qlen < (netdev_max_backlog >> 1)) 4533 return false; 4534 4535 sd = this_cpu_ptr(&softnet_data); 4536 4537 rcu_read_lock(); 4538 fl = rcu_dereference(sd->flow_limit); 4539 if (fl) { 4540 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); 4541 old_flow = fl->history[fl->history_head]; 4542 fl->history[fl->history_head] = new_flow; 4543 4544 fl->history_head++; 4545 fl->history_head &= FLOW_LIMIT_HISTORY - 1; 4546 4547 if (likely(fl->buckets[old_flow])) 4548 fl->buckets[old_flow]--; 4549 4550 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { 4551 fl->count++; 4552 rcu_read_unlock(); 4553 return true; 4554 } 4555 } 4556 rcu_read_unlock(); 4557 #endif 4558 return false; 4559 } 4560 4561 /* 4562 * enqueue_to_backlog is called to queue an skb to a per CPU backlog 4563 * queue (may be a remote CPU queue). 4564 */ 4565 static int enqueue_to_backlog(struct sk_buff *skb, int cpu, 4566 unsigned int *qtail) 4567 { 4568 struct softnet_data *sd; 4569 unsigned long flags; 4570 unsigned int qlen; 4571 4572 sd = &per_cpu(softnet_data, cpu); 4573 4574 local_irq_save(flags); 4575 4576 rps_lock(sd); 4577 if (!netif_running(skb->dev)) 4578 goto drop; 4579 qlen = skb_queue_len(&sd->input_pkt_queue); 4580 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) { 4581 if (qlen) { 4582 enqueue: 4583 __skb_queue_tail(&sd->input_pkt_queue, skb); 4584 input_queue_tail_incr_save(sd, qtail); 4585 rps_unlock(sd); 4586 local_irq_restore(flags); 4587 return NET_RX_SUCCESS; 4588 } 4589 4590 /* Schedule NAPI for backlog device 4591 * We can use non atomic operation since we own the queue lock 4592 */ 4593 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) { 4594 if (!rps_ipi_queued(sd)) 4595 ____napi_schedule(sd, &sd->backlog); 4596 } 4597 goto enqueue; 4598 } 4599 4600 drop: 4601 sd->dropped++; 4602 rps_unlock(sd); 4603 4604 local_irq_restore(flags); 4605 4606 atomic_long_inc(&skb->dev->rx_dropped); 4607 kfree_skb(skb); 4608 return NET_RX_DROP; 4609 } 4610 4611 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb) 4612 { 4613 struct net_device *dev = skb->dev; 4614 struct netdev_rx_queue *rxqueue; 4615 4616 rxqueue = dev->_rx; 4617 4618 if (skb_rx_queue_recorded(skb)) { 4619 u16 index = skb_get_rx_queue(skb); 4620 4621 if (unlikely(index >= dev->real_num_rx_queues)) { 4622 WARN_ONCE(dev->real_num_rx_queues > 1, 4623 "%s received packet on queue %u, but number " 4624 "of RX queues is %u\n", 4625 dev->name, index, dev->real_num_rx_queues); 4626 4627 return rxqueue; /* Return first rxqueue */ 4628 } 4629 rxqueue += index; 4630 } 4631 return rxqueue; 4632 } 4633 4634 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp, 4635 struct bpf_prog *xdp_prog) 4636 { 4637 void *orig_data, *orig_data_end, *hard_start; 4638 struct netdev_rx_queue *rxqueue; 4639 bool orig_bcast, orig_host; 4640 u32 mac_len, frame_sz; 4641 __be16 orig_eth_type; 4642 struct ethhdr *eth; 4643 u32 metalen, act; 4644 int off; 4645 4646 /* The XDP program wants to see the packet starting at the MAC 4647 * header. 4648 */ 4649 mac_len = skb->data - skb_mac_header(skb); 4650 hard_start = skb->data - skb_headroom(skb); 4651 4652 /* SKB "head" area always have tailroom for skb_shared_info */ 4653 frame_sz = (void *)skb_end_pointer(skb) - hard_start; 4654 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 4655 4656 rxqueue = netif_get_rxqueue(skb); 4657 xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq); 4658 xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len, 4659 skb_headlen(skb) + mac_len, true); 4660 4661 orig_data_end = xdp->data_end; 4662 orig_data = xdp->data; 4663 eth = (struct ethhdr *)xdp->data; 4664 orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr); 4665 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest); 4666 orig_eth_type = eth->h_proto; 4667 4668 act = bpf_prog_run_xdp(xdp_prog, xdp); 4669 4670 /* check if bpf_xdp_adjust_head was used */ 4671 off = xdp->data - orig_data; 4672 if (off) { 4673 if (off > 0) 4674 __skb_pull(skb, off); 4675 else if (off < 0) 4676 __skb_push(skb, -off); 4677 4678 skb->mac_header += off; 4679 skb_reset_network_header(skb); 4680 } 4681 4682 /* check if bpf_xdp_adjust_tail was used */ 4683 off = xdp->data_end - orig_data_end; 4684 if (off != 0) { 4685 skb_set_tail_pointer(skb, xdp->data_end - xdp->data); 4686 skb->len += off; /* positive on grow, negative on shrink */ 4687 } 4688 4689 /* check if XDP changed eth hdr such SKB needs update */ 4690 eth = (struct ethhdr *)xdp->data; 4691 if ((orig_eth_type != eth->h_proto) || 4692 (orig_host != ether_addr_equal_64bits(eth->h_dest, 4693 skb->dev->dev_addr)) || 4694 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) { 4695 __skb_push(skb, ETH_HLEN); 4696 skb->pkt_type = PACKET_HOST; 4697 skb->protocol = eth_type_trans(skb, skb->dev); 4698 } 4699 4700 /* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull 4701 * before calling us again on redirect path. We do not call do_redirect 4702 * as we leave that up to the caller. 4703 * 4704 * Caller is responsible for managing lifetime of skb (i.e. calling 4705 * kfree_skb in response to actions it cannot handle/XDP_DROP). 4706 */ 4707 switch (act) { 4708 case XDP_REDIRECT: 4709 case XDP_TX: 4710 __skb_push(skb, mac_len); 4711 break; 4712 case XDP_PASS: 4713 metalen = xdp->data - xdp->data_meta; 4714 if (metalen) 4715 skb_metadata_set(skb, metalen); 4716 break; 4717 } 4718 4719 return act; 4720 } 4721 4722 static u32 netif_receive_generic_xdp(struct sk_buff *skb, 4723 struct xdp_buff *xdp, 4724 struct bpf_prog *xdp_prog) 4725 { 4726 u32 act = XDP_DROP; 4727 4728 /* Reinjected packets coming from act_mirred or similar should 4729 * not get XDP generic processing. 4730 */ 4731 if (skb_is_redirected(skb)) 4732 return XDP_PASS; 4733 4734 /* XDP packets must be linear and must have sufficient headroom 4735 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also 4736 * native XDP provides, thus we need to do it here as well. 4737 */ 4738 if (skb_cloned(skb) || skb_is_nonlinear(skb) || 4739 skb_headroom(skb) < XDP_PACKET_HEADROOM) { 4740 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb); 4741 int troom = skb->tail + skb->data_len - skb->end; 4742 4743 /* In case we have to go down the path and also linearize, 4744 * then lets do the pskb_expand_head() work just once here. 4745 */ 4746 if (pskb_expand_head(skb, 4747 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0, 4748 troom > 0 ? troom + 128 : 0, GFP_ATOMIC)) 4749 goto do_drop; 4750 if (skb_linearize(skb)) 4751 goto do_drop; 4752 } 4753 4754 act = bpf_prog_run_generic_xdp(skb, xdp, xdp_prog); 4755 switch (act) { 4756 case XDP_REDIRECT: 4757 case XDP_TX: 4758 case XDP_PASS: 4759 break; 4760 default: 4761 bpf_warn_invalid_xdp_action(act); 4762 fallthrough; 4763 case XDP_ABORTED: 4764 trace_xdp_exception(skb->dev, xdp_prog, act); 4765 fallthrough; 4766 case XDP_DROP: 4767 do_drop: 4768 kfree_skb(skb); 4769 break; 4770 } 4771 4772 return act; 4773 } 4774 4775 /* When doing generic XDP we have to bypass the qdisc layer and the 4776 * network taps in order to match in-driver-XDP behavior. 4777 */ 4778 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog) 4779 { 4780 struct net_device *dev = skb->dev; 4781 struct netdev_queue *txq; 4782 bool free_skb = true; 4783 int cpu, rc; 4784 4785 txq = netdev_core_pick_tx(dev, skb, NULL); 4786 cpu = smp_processor_id(); 4787 HARD_TX_LOCK(dev, txq, cpu); 4788 if (!netif_xmit_stopped(txq)) { 4789 rc = netdev_start_xmit(skb, dev, txq, 0); 4790 if (dev_xmit_complete(rc)) 4791 free_skb = false; 4792 } 4793 HARD_TX_UNLOCK(dev, txq); 4794 if (free_skb) { 4795 trace_xdp_exception(dev, xdp_prog, XDP_TX); 4796 kfree_skb(skb); 4797 } 4798 } 4799 4800 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key); 4801 4802 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb) 4803 { 4804 if (xdp_prog) { 4805 struct xdp_buff xdp; 4806 u32 act; 4807 int err; 4808 4809 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog); 4810 if (act != XDP_PASS) { 4811 switch (act) { 4812 case XDP_REDIRECT: 4813 err = xdp_do_generic_redirect(skb->dev, skb, 4814 &xdp, xdp_prog); 4815 if (err) 4816 goto out_redir; 4817 break; 4818 case XDP_TX: 4819 generic_xdp_tx(skb, xdp_prog); 4820 break; 4821 } 4822 return XDP_DROP; 4823 } 4824 } 4825 return XDP_PASS; 4826 out_redir: 4827 kfree_skb(skb); 4828 return XDP_DROP; 4829 } 4830 EXPORT_SYMBOL_GPL(do_xdp_generic); 4831 4832 static int netif_rx_internal(struct sk_buff *skb) 4833 { 4834 int ret; 4835 4836 net_timestamp_check(netdev_tstamp_prequeue, skb); 4837 4838 trace_netif_rx(skb); 4839 4840 #ifdef CONFIG_RPS 4841 if (static_branch_unlikely(&rps_needed)) { 4842 struct rps_dev_flow voidflow, *rflow = &voidflow; 4843 int cpu; 4844 4845 preempt_disable(); 4846 rcu_read_lock(); 4847 4848 cpu = get_rps_cpu(skb->dev, skb, &rflow); 4849 if (cpu < 0) 4850 cpu = smp_processor_id(); 4851 4852 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 4853 4854 rcu_read_unlock(); 4855 preempt_enable(); 4856 } else 4857 #endif 4858 { 4859 unsigned int qtail; 4860 4861 ret = enqueue_to_backlog(skb, get_cpu(), &qtail); 4862 put_cpu(); 4863 } 4864 return ret; 4865 } 4866 4867 /** 4868 * netif_rx - post buffer to the network code 4869 * @skb: buffer to post 4870 * 4871 * This function receives a packet from a device driver and queues it for 4872 * the upper (protocol) levels to process. It always succeeds. The buffer 4873 * may be dropped during processing for congestion control or by the 4874 * protocol layers. 4875 * 4876 * return values: 4877 * NET_RX_SUCCESS (no congestion) 4878 * NET_RX_DROP (packet was dropped) 4879 * 4880 */ 4881 4882 int netif_rx(struct sk_buff *skb) 4883 { 4884 int ret; 4885 4886 trace_netif_rx_entry(skb); 4887 4888 ret = netif_rx_internal(skb); 4889 trace_netif_rx_exit(ret); 4890 4891 return ret; 4892 } 4893 EXPORT_SYMBOL(netif_rx); 4894 4895 int netif_rx_ni(struct sk_buff *skb) 4896 { 4897 int err; 4898 4899 trace_netif_rx_ni_entry(skb); 4900 4901 preempt_disable(); 4902 err = netif_rx_internal(skb); 4903 if (local_softirq_pending()) 4904 do_softirq(); 4905 preempt_enable(); 4906 trace_netif_rx_ni_exit(err); 4907 4908 return err; 4909 } 4910 EXPORT_SYMBOL(netif_rx_ni); 4911 4912 int netif_rx_any_context(struct sk_buff *skb) 4913 { 4914 /* 4915 * If invoked from contexts which do not invoke bottom half 4916 * processing either at return from interrupt or when softrqs are 4917 * reenabled, use netif_rx_ni() which invokes bottomhalf processing 4918 * directly. 4919 */ 4920 if (in_interrupt()) 4921 return netif_rx(skb); 4922 else 4923 return netif_rx_ni(skb); 4924 } 4925 EXPORT_SYMBOL(netif_rx_any_context); 4926 4927 static __latent_entropy void net_tx_action(struct softirq_action *h) 4928 { 4929 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 4930 4931 if (sd->completion_queue) { 4932 struct sk_buff *clist; 4933 4934 local_irq_disable(); 4935 clist = sd->completion_queue; 4936 sd->completion_queue = NULL; 4937 local_irq_enable(); 4938 4939 while (clist) { 4940 struct sk_buff *skb = clist; 4941 4942 clist = clist->next; 4943 4944 WARN_ON(refcount_read(&skb->users)); 4945 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED)) 4946 trace_consume_skb(skb); 4947 else 4948 trace_kfree_skb(skb, net_tx_action); 4949 4950 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) 4951 __kfree_skb(skb); 4952 else 4953 __kfree_skb_defer(skb); 4954 } 4955 } 4956 4957 if (sd->output_queue) { 4958 struct Qdisc *head; 4959 4960 local_irq_disable(); 4961 head = sd->output_queue; 4962 sd->output_queue = NULL; 4963 sd->output_queue_tailp = &sd->output_queue; 4964 local_irq_enable(); 4965 4966 rcu_read_lock(); 4967 4968 while (head) { 4969 struct Qdisc *q = head; 4970 spinlock_t *root_lock = NULL; 4971 4972 head = head->next_sched; 4973 4974 /* We need to make sure head->next_sched is read 4975 * before clearing __QDISC_STATE_SCHED 4976 */ 4977 smp_mb__before_atomic(); 4978 4979 if (!(q->flags & TCQ_F_NOLOCK)) { 4980 root_lock = qdisc_lock(q); 4981 spin_lock(root_lock); 4982 } else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, 4983 &q->state))) { 4984 /* There is a synchronize_net() between 4985 * STATE_DEACTIVATED flag being set and 4986 * qdisc_reset()/some_qdisc_is_busy() in 4987 * dev_deactivate(), so we can safely bail out 4988 * early here to avoid data race between 4989 * qdisc_deactivate() and some_qdisc_is_busy() 4990 * for lockless qdisc. 4991 */ 4992 clear_bit(__QDISC_STATE_SCHED, &q->state); 4993 continue; 4994 } 4995 4996 clear_bit(__QDISC_STATE_SCHED, &q->state); 4997 qdisc_run(q); 4998 if (root_lock) 4999 spin_unlock(root_lock); 5000 } 5001 5002 rcu_read_unlock(); 5003 } 5004 5005 xfrm_dev_backlog(sd); 5006 } 5007 5008 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) 5009 /* This hook is defined here for ATM LANE */ 5010 int (*br_fdb_test_addr_hook)(struct net_device *dev, 5011 unsigned char *addr) __read_mostly; 5012 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); 5013 #endif 5014 5015 static inline struct sk_buff * 5016 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 5017 struct net_device *orig_dev, bool *another) 5018 { 5019 #ifdef CONFIG_NET_CLS_ACT 5020 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress); 5021 struct tcf_result cl_res; 5022 5023 /* If there's at least one ingress present somewhere (so 5024 * we get here via enabled static key), remaining devices 5025 * that are not configured with an ingress qdisc will bail 5026 * out here. 5027 */ 5028 if (!miniq) 5029 return skb; 5030 5031 if (*pt_prev) { 5032 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5033 *pt_prev = NULL; 5034 } 5035 5036 qdisc_skb_cb(skb)->pkt_len = skb->len; 5037 qdisc_skb_cb(skb)->mru = 0; 5038 qdisc_skb_cb(skb)->post_ct = false; 5039 skb->tc_at_ingress = 1; 5040 mini_qdisc_bstats_cpu_update(miniq, skb); 5041 5042 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) { 5043 case TC_ACT_OK: 5044 case TC_ACT_RECLASSIFY: 5045 skb->tc_index = TC_H_MIN(cl_res.classid); 5046 break; 5047 case TC_ACT_SHOT: 5048 mini_qdisc_qstats_cpu_drop(miniq); 5049 kfree_skb(skb); 5050 return NULL; 5051 case TC_ACT_STOLEN: 5052 case TC_ACT_QUEUED: 5053 case TC_ACT_TRAP: 5054 consume_skb(skb); 5055 return NULL; 5056 case TC_ACT_REDIRECT: 5057 /* skb_mac_header check was done by cls/act_bpf, so 5058 * we can safely push the L2 header back before 5059 * redirecting to another netdev 5060 */ 5061 __skb_push(skb, skb->mac_len); 5062 if (skb_do_redirect(skb) == -EAGAIN) { 5063 __skb_pull(skb, skb->mac_len); 5064 *another = true; 5065 break; 5066 } 5067 return NULL; 5068 case TC_ACT_CONSUMED: 5069 return NULL; 5070 default: 5071 break; 5072 } 5073 #endif /* CONFIG_NET_CLS_ACT */ 5074 return skb; 5075 } 5076 5077 /** 5078 * netdev_is_rx_handler_busy - check if receive handler is registered 5079 * @dev: device to check 5080 * 5081 * Check if a receive handler is already registered for a given device. 5082 * Return true if there one. 5083 * 5084 * The caller must hold the rtnl_mutex. 5085 */ 5086 bool netdev_is_rx_handler_busy(struct net_device *dev) 5087 { 5088 ASSERT_RTNL(); 5089 return dev && rtnl_dereference(dev->rx_handler); 5090 } 5091 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); 5092 5093 /** 5094 * netdev_rx_handler_register - register receive handler 5095 * @dev: device to register a handler for 5096 * @rx_handler: receive handler to register 5097 * @rx_handler_data: data pointer that is used by rx handler 5098 * 5099 * Register a receive handler for a device. This handler will then be 5100 * called from __netif_receive_skb. A negative errno code is returned 5101 * on a failure. 5102 * 5103 * The caller must hold the rtnl_mutex. 5104 * 5105 * For a general description of rx_handler, see enum rx_handler_result. 5106 */ 5107 int netdev_rx_handler_register(struct net_device *dev, 5108 rx_handler_func_t *rx_handler, 5109 void *rx_handler_data) 5110 { 5111 if (netdev_is_rx_handler_busy(dev)) 5112 return -EBUSY; 5113 5114 if (dev->priv_flags & IFF_NO_RX_HANDLER) 5115 return -EINVAL; 5116 5117 /* Note: rx_handler_data must be set before rx_handler */ 5118 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); 5119 rcu_assign_pointer(dev->rx_handler, rx_handler); 5120 5121 return 0; 5122 } 5123 EXPORT_SYMBOL_GPL(netdev_rx_handler_register); 5124 5125 /** 5126 * netdev_rx_handler_unregister - unregister receive handler 5127 * @dev: device to unregister a handler from 5128 * 5129 * Unregister a receive handler from a device. 5130 * 5131 * The caller must hold the rtnl_mutex. 5132 */ 5133 void netdev_rx_handler_unregister(struct net_device *dev) 5134 { 5135 5136 ASSERT_RTNL(); 5137 RCU_INIT_POINTER(dev->rx_handler, NULL); 5138 /* a reader seeing a non NULL rx_handler in a rcu_read_lock() 5139 * section has a guarantee to see a non NULL rx_handler_data 5140 * as well. 5141 */ 5142 synchronize_net(); 5143 RCU_INIT_POINTER(dev->rx_handler_data, NULL); 5144 } 5145 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); 5146 5147 /* 5148 * Limit the use of PFMEMALLOC reserves to those protocols that implement 5149 * the special handling of PFMEMALLOC skbs. 5150 */ 5151 static bool skb_pfmemalloc_protocol(struct sk_buff *skb) 5152 { 5153 switch (skb->protocol) { 5154 case htons(ETH_P_ARP): 5155 case htons(ETH_P_IP): 5156 case htons(ETH_P_IPV6): 5157 case htons(ETH_P_8021Q): 5158 case htons(ETH_P_8021AD): 5159 return true; 5160 default: 5161 return false; 5162 } 5163 } 5164 5165 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, 5166 int *ret, struct net_device *orig_dev) 5167 { 5168 if (nf_hook_ingress_active(skb)) { 5169 int ingress_retval; 5170 5171 if (*pt_prev) { 5172 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5173 *pt_prev = NULL; 5174 } 5175 5176 rcu_read_lock(); 5177 ingress_retval = nf_hook_ingress(skb); 5178 rcu_read_unlock(); 5179 return ingress_retval; 5180 } 5181 return 0; 5182 } 5183 5184 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc, 5185 struct packet_type **ppt_prev) 5186 { 5187 struct packet_type *ptype, *pt_prev; 5188 rx_handler_func_t *rx_handler; 5189 struct sk_buff *skb = *pskb; 5190 struct net_device *orig_dev; 5191 bool deliver_exact = false; 5192 int ret = NET_RX_DROP; 5193 __be16 type; 5194 5195 net_timestamp_check(!netdev_tstamp_prequeue, skb); 5196 5197 trace_netif_receive_skb(skb); 5198 5199 orig_dev = skb->dev; 5200 5201 skb_reset_network_header(skb); 5202 if (!skb_transport_header_was_set(skb)) 5203 skb_reset_transport_header(skb); 5204 skb_reset_mac_len(skb); 5205 5206 pt_prev = NULL; 5207 5208 another_round: 5209 skb->skb_iif = skb->dev->ifindex; 5210 5211 __this_cpu_inc(softnet_data.processed); 5212 5213 if (static_branch_unlikely(&generic_xdp_needed_key)) { 5214 int ret2; 5215 5216 migrate_disable(); 5217 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb); 5218 migrate_enable(); 5219 5220 if (ret2 != XDP_PASS) { 5221 ret = NET_RX_DROP; 5222 goto out; 5223 } 5224 } 5225 5226 if (eth_type_vlan(skb->protocol)) { 5227 skb = skb_vlan_untag(skb); 5228 if (unlikely(!skb)) 5229 goto out; 5230 } 5231 5232 if (skb_skip_tc_classify(skb)) 5233 goto skip_classify; 5234 5235 if (pfmemalloc) 5236 goto skip_taps; 5237 5238 list_for_each_entry_rcu(ptype, &ptype_all, list) { 5239 if (pt_prev) 5240 ret = deliver_skb(skb, pt_prev, orig_dev); 5241 pt_prev = ptype; 5242 } 5243 5244 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { 5245 if (pt_prev) 5246 ret = deliver_skb(skb, pt_prev, orig_dev); 5247 pt_prev = ptype; 5248 } 5249 5250 skip_taps: 5251 #ifdef CONFIG_NET_INGRESS 5252 if (static_branch_unlikely(&ingress_needed_key)) { 5253 bool another = false; 5254 5255 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev, 5256 &another); 5257 if (another) 5258 goto another_round; 5259 if (!skb) 5260 goto out; 5261 5262 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) 5263 goto out; 5264 } 5265 #endif 5266 skb_reset_redirect(skb); 5267 skip_classify: 5268 if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) 5269 goto drop; 5270 5271 if (skb_vlan_tag_present(skb)) { 5272 if (pt_prev) { 5273 ret = deliver_skb(skb, pt_prev, orig_dev); 5274 pt_prev = NULL; 5275 } 5276 if (vlan_do_receive(&skb)) 5277 goto another_round; 5278 else if (unlikely(!skb)) 5279 goto out; 5280 } 5281 5282 rx_handler = rcu_dereference(skb->dev->rx_handler); 5283 if (rx_handler) { 5284 if (pt_prev) { 5285 ret = deliver_skb(skb, pt_prev, orig_dev); 5286 pt_prev = NULL; 5287 } 5288 switch (rx_handler(&skb)) { 5289 case RX_HANDLER_CONSUMED: 5290 ret = NET_RX_SUCCESS; 5291 goto out; 5292 case RX_HANDLER_ANOTHER: 5293 goto another_round; 5294 case RX_HANDLER_EXACT: 5295 deliver_exact = true; 5296 break; 5297 case RX_HANDLER_PASS: 5298 break; 5299 default: 5300 BUG(); 5301 } 5302 } 5303 5304 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) { 5305 check_vlan_id: 5306 if (skb_vlan_tag_get_id(skb)) { 5307 /* Vlan id is non 0 and vlan_do_receive() above couldn't 5308 * find vlan device. 5309 */ 5310 skb->pkt_type = PACKET_OTHERHOST; 5311 } else if (eth_type_vlan(skb->protocol)) { 5312 /* Outer header is 802.1P with vlan 0, inner header is 5313 * 802.1Q or 802.1AD and vlan_do_receive() above could 5314 * not find vlan dev for vlan id 0. 5315 */ 5316 __vlan_hwaccel_clear_tag(skb); 5317 skb = skb_vlan_untag(skb); 5318 if (unlikely(!skb)) 5319 goto out; 5320 if (vlan_do_receive(&skb)) 5321 /* After stripping off 802.1P header with vlan 0 5322 * vlan dev is found for inner header. 5323 */ 5324 goto another_round; 5325 else if (unlikely(!skb)) 5326 goto out; 5327 else 5328 /* We have stripped outer 802.1P vlan 0 header. 5329 * But could not find vlan dev. 5330 * check again for vlan id to set OTHERHOST. 5331 */ 5332 goto check_vlan_id; 5333 } 5334 /* Note: we might in the future use prio bits 5335 * and set skb->priority like in vlan_do_receive() 5336 * For the time being, just ignore Priority Code Point 5337 */ 5338 __vlan_hwaccel_clear_tag(skb); 5339 } 5340 5341 type = skb->protocol; 5342 5343 /* deliver only exact match when indicated */ 5344 if (likely(!deliver_exact)) { 5345 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5346 &ptype_base[ntohs(type) & 5347 PTYPE_HASH_MASK]); 5348 } 5349 5350 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5351 &orig_dev->ptype_specific); 5352 5353 if (unlikely(skb->dev != orig_dev)) { 5354 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5355 &skb->dev->ptype_specific); 5356 } 5357 5358 if (pt_prev) { 5359 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 5360 goto drop; 5361 *ppt_prev = pt_prev; 5362 } else { 5363 drop: 5364 if (!deliver_exact) 5365 atomic_long_inc(&skb->dev->rx_dropped); 5366 else 5367 atomic_long_inc(&skb->dev->rx_nohandler); 5368 kfree_skb(skb); 5369 /* Jamal, now you will not able to escape explaining 5370 * me how you were going to use this. :-) 5371 */ 5372 ret = NET_RX_DROP; 5373 } 5374 5375 out: 5376 /* The invariant here is that if *ppt_prev is not NULL 5377 * then skb should also be non-NULL. 5378 * 5379 * Apparently *ppt_prev assignment above holds this invariant due to 5380 * skb dereferencing near it. 5381 */ 5382 *pskb = skb; 5383 return ret; 5384 } 5385 5386 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc) 5387 { 5388 struct net_device *orig_dev = skb->dev; 5389 struct packet_type *pt_prev = NULL; 5390 int ret; 5391 5392 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5393 if (pt_prev) 5394 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb, 5395 skb->dev, pt_prev, orig_dev); 5396 return ret; 5397 } 5398 5399 /** 5400 * netif_receive_skb_core - special purpose version of netif_receive_skb 5401 * @skb: buffer to process 5402 * 5403 * More direct receive version of netif_receive_skb(). It should 5404 * only be used by callers that have a need to skip RPS and Generic XDP. 5405 * Caller must also take care of handling if ``(page_is_)pfmemalloc``. 5406 * 5407 * This function may only be called from softirq context and interrupts 5408 * should be enabled. 5409 * 5410 * Return values (usually ignored): 5411 * NET_RX_SUCCESS: no congestion 5412 * NET_RX_DROP: packet was dropped 5413 */ 5414 int netif_receive_skb_core(struct sk_buff *skb) 5415 { 5416 int ret; 5417 5418 rcu_read_lock(); 5419 ret = __netif_receive_skb_one_core(skb, false); 5420 rcu_read_unlock(); 5421 5422 return ret; 5423 } 5424 EXPORT_SYMBOL(netif_receive_skb_core); 5425 5426 static inline void __netif_receive_skb_list_ptype(struct list_head *head, 5427 struct packet_type *pt_prev, 5428 struct net_device *orig_dev) 5429 { 5430 struct sk_buff *skb, *next; 5431 5432 if (!pt_prev) 5433 return; 5434 if (list_empty(head)) 5435 return; 5436 if (pt_prev->list_func != NULL) 5437 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv, 5438 ip_list_rcv, head, pt_prev, orig_dev); 5439 else 5440 list_for_each_entry_safe(skb, next, head, list) { 5441 skb_list_del_init(skb); 5442 pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 5443 } 5444 } 5445 5446 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc) 5447 { 5448 /* Fast-path assumptions: 5449 * - There is no RX handler. 5450 * - Only one packet_type matches. 5451 * If either of these fails, we will end up doing some per-packet 5452 * processing in-line, then handling the 'last ptype' for the whole 5453 * sublist. This can't cause out-of-order delivery to any single ptype, 5454 * because the 'last ptype' must be constant across the sublist, and all 5455 * other ptypes are handled per-packet. 5456 */ 5457 /* Current (common) ptype of sublist */ 5458 struct packet_type *pt_curr = NULL; 5459 /* Current (common) orig_dev of sublist */ 5460 struct net_device *od_curr = NULL; 5461 struct list_head sublist; 5462 struct sk_buff *skb, *next; 5463 5464 INIT_LIST_HEAD(&sublist); 5465 list_for_each_entry_safe(skb, next, head, list) { 5466 struct net_device *orig_dev = skb->dev; 5467 struct packet_type *pt_prev = NULL; 5468 5469 skb_list_del_init(skb); 5470 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5471 if (!pt_prev) 5472 continue; 5473 if (pt_curr != pt_prev || od_curr != orig_dev) { 5474 /* dispatch old sublist */ 5475 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5476 /* start new sublist */ 5477 INIT_LIST_HEAD(&sublist); 5478 pt_curr = pt_prev; 5479 od_curr = orig_dev; 5480 } 5481 list_add_tail(&skb->list, &sublist); 5482 } 5483 5484 /* dispatch final sublist */ 5485 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5486 } 5487 5488 static int __netif_receive_skb(struct sk_buff *skb) 5489 { 5490 int ret; 5491 5492 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { 5493 unsigned int noreclaim_flag; 5494 5495 /* 5496 * PFMEMALLOC skbs are special, they should 5497 * - be delivered to SOCK_MEMALLOC sockets only 5498 * - stay away from userspace 5499 * - have bounded memory usage 5500 * 5501 * Use PF_MEMALLOC as this saves us from propagating the allocation 5502 * context down to all allocation sites. 5503 */ 5504 noreclaim_flag = memalloc_noreclaim_save(); 5505 ret = __netif_receive_skb_one_core(skb, true); 5506 memalloc_noreclaim_restore(noreclaim_flag); 5507 } else 5508 ret = __netif_receive_skb_one_core(skb, false); 5509 5510 return ret; 5511 } 5512 5513 static void __netif_receive_skb_list(struct list_head *head) 5514 { 5515 unsigned long noreclaim_flag = 0; 5516 struct sk_buff *skb, *next; 5517 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */ 5518 5519 list_for_each_entry_safe(skb, next, head, list) { 5520 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) { 5521 struct list_head sublist; 5522 5523 /* Handle the previous sublist */ 5524 list_cut_before(&sublist, head, &skb->list); 5525 if (!list_empty(&sublist)) 5526 __netif_receive_skb_list_core(&sublist, pfmemalloc); 5527 pfmemalloc = !pfmemalloc; 5528 /* See comments in __netif_receive_skb */ 5529 if (pfmemalloc) 5530 noreclaim_flag = memalloc_noreclaim_save(); 5531 else 5532 memalloc_noreclaim_restore(noreclaim_flag); 5533 } 5534 } 5535 /* Handle the remaining sublist */ 5536 if (!list_empty(head)) 5537 __netif_receive_skb_list_core(head, pfmemalloc); 5538 /* Restore pflags */ 5539 if (pfmemalloc) 5540 memalloc_noreclaim_restore(noreclaim_flag); 5541 } 5542 5543 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp) 5544 { 5545 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog); 5546 struct bpf_prog *new = xdp->prog; 5547 int ret = 0; 5548 5549 switch (xdp->command) { 5550 case XDP_SETUP_PROG: 5551 rcu_assign_pointer(dev->xdp_prog, new); 5552 if (old) 5553 bpf_prog_put(old); 5554 5555 if (old && !new) { 5556 static_branch_dec(&generic_xdp_needed_key); 5557 } else if (new && !old) { 5558 static_branch_inc(&generic_xdp_needed_key); 5559 dev_disable_lro(dev); 5560 dev_disable_gro_hw(dev); 5561 } 5562 break; 5563 5564 default: 5565 ret = -EINVAL; 5566 break; 5567 } 5568 5569 return ret; 5570 } 5571 5572 static int netif_receive_skb_internal(struct sk_buff *skb) 5573 { 5574 int ret; 5575 5576 net_timestamp_check(netdev_tstamp_prequeue, skb); 5577 5578 if (skb_defer_rx_timestamp(skb)) 5579 return NET_RX_SUCCESS; 5580 5581 rcu_read_lock(); 5582 #ifdef CONFIG_RPS 5583 if (static_branch_unlikely(&rps_needed)) { 5584 struct rps_dev_flow voidflow, *rflow = &voidflow; 5585 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5586 5587 if (cpu >= 0) { 5588 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5589 rcu_read_unlock(); 5590 return ret; 5591 } 5592 } 5593 #endif 5594 ret = __netif_receive_skb(skb); 5595 rcu_read_unlock(); 5596 return ret; 5597 } 5598 5599 static void netif_receive_skb_list_internal(struct list_head *head) 5600 { 5601 struct sk_buff *skb, *next; 5602 struct list_head sublist; 5603 5604 INIT_LIST_HEAD(&sublist); 5605 list_for_each_entry_safe(skb, next, head, list) { 5606 net_timestamp_check(netdev_tstamp_prequeue, skb); 5607 skb_list_del_init(skb); 5608 if (!skb_defer_rx_timestamp(skb)) 5609 list_add_tail(&skb->list, &sublist); 5610 } 5611 list_splice_init(&sublist, head); 5612 5613 rcu_read_lock(); 5614 #ifdef CONFIG_RPS 5615 if (static_branch_unlikely(&rps_needed)) { 5616 list_for_each_entry_safe(skb, next, head, list) { 5617 struct rps_dev_flow voidflow, *rflow = &voidflow; 5618 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5619 5620 if (cpu >= 0) { 5621 /* Will be handled, remove from list */ 5622 skb_list_del_init(skb); 5623 enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5624 } 5625 } 5626 } 5627 #endif 5628 __netif_receive_skb_list(head); 5629 rcu_read_unlock(); 5630 } 5631 5632 /** 5633 * netif_receive_skb - process receive buffer from network 5634 * @skb: buffer to process 5635 * 5636 * netif_receive_skb() is the main receive data processing function. 5637 * It always succeeds. The buffer may be dropped during processing 5638 * for congestion control or by the protocol layers. 5639 * 5640 * This function may only be called from softirq context and interrupts 5641 * should be enabled. 5642 * 5643 * Return values (usually ignored): 5644 * NET_RX_SUCCESS: no congestion 5645 * NET_RX_DROP: packet was dropped 5646 */ 5647 int netif_receive_skb(struct sk_buff *skb) 5648 { 5649 int ret; 5650 5651 trace_netif_receive_skb_entry(skb); 5652 5653 ret = netif_receive_skb_internal(skb); 5654 trace_netif_receive_skb_exit(ret); 5655 5656 return ret; 5657 } 5658 EXPORT_SYMBOL(netif_receive_skb); 5659 5660 /** 5661 * netif_receive_skb_list - process many receive buffers from network 5662 * @head: list of skbs to process. 5663 * 5664 * Since return value of netif_receive_skb() is normally ignored, and 5665 * wouldn't be meaningful for a list, this function returns void. 5666 * 5667 * This function may only be called from softirq context and interrupts 5668 * should be enabled. 5669 */ 5670 void netif_receive_skb_list(struct list_head *head) 5671 { 5672 struct sk_buff *skb; 5673 5674 if (list_empty(head)) 5675 return; 5676 if (trace_netif_receive_skb_list_entry_enabled()) { 5677 list_for_each_entry(skb, head, list) 5678 trace_netif_receive_skb_list_entry(skb); 5679 } 5680 netif_receive_skb_list_internal(head); 5681 trace_netif_receive_skb_list_exit(0); 5682 } 5683 EXPORT_SYMBOL(netif_receive_skb_list); 5684 5685 static DEFINE_PER_CPU(struct work_struct, flush_works); 5686 5687 /* Network device is going away, flush any packets still pending */ 5688 static void flush_backlog(struct work_struct *work) 5689 { 5690 struct sk_buff *skb, *tmp; 5691 struct softnet_data *sd; 5692 5693 local_bh_disable(); 5694 sd = this_cpu_ptr(&softnet_data); 5695 5696 local_irq_disable(); 5697 rps_lock(sd); 5698 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { 5699 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5700 __skb_unlink(skb, &sd->input_pkt_queue); 5701 dev_kfree_skb_irq(skb); 5702 input_queue_head_incr(sd); 5703 } 5704 } 5705 rps_unlock(sd); 5706 local_irq_enable(); 5707 5708 skb_queue_walk_safe(&sd->process_queue, skb, tmp) { 5709 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5710 __skb_unlink(skb, &sd->process_queue); 5711 kfree_skb(skb); 5712 input_queue_head_incr(sd); 5713 } 5714 } 5715 local_bh_enable(); 5716 } 5717 5718 static bool flush_required(int cpu) 5719 { 5720 #if IS_ENABLED(CONFIG_RPS) 5721 struct softnet_data *sd = &per_cpu(softnet_data, cpu); 5722 bool do_flush; 5723 5724 local_irq_disable(); 5725 rps_lock(sd); 5726 5727 /* as insertion into process_queue happens with the rps lock held, 5728 * process_queue access may race only with dequeue 5729 */ 5730 do_flush = !skb_queue_empty(&sd->input_pkt_queue) || 5731 !skb_queue_empty_lockless(&sd->process_queue); 5732 rps_unlock(sd); 5733 local_irq_enable(); 5734 5735 return do_flush; 5736 #endif 5737 /* without RPS we can't safely check input_pkt_queue: during a 5738 * concurrent remote skb_queue_splice() we can detect as empty both 5739 * input_pkt_queue and process_queue even if the latter could end-up 5740 * containing a lot of packets. 5741 */ 5742 return true; 5743 } 5744 5745 static void flush_all_backlogs(void) 5746 { 5747 static cpumask_t flush_cpus; 5748 unsigned int cpu; 5749 5750 /* since we are under rtnl lock protection we can use static data 5751 * for the cpumask and avoid allocating on stack the possibly 5752 * large mask 5753 */ 5754 ASSERT_RTNL(); 5755 5756 get_online_cpus(); 5757 5758 cpumask_clear(&flush_cpus); 5759 for_each_online_cpu(cpu) { 5760 if (flush_required(cpu)) { 5761 queue_work_on(cpu, system_highpri_wq, 5762 per_cpu_ptr(&flush_works, cpu)); 5763 cpumask_set_cpu(cpu, &flush_cpus); 5764 } 5765 } 5766 5767 /* we can have in flight packet[s] on the cpus we are not flushing, 5768 * synchronize_net() in unregister_netdevice_many() will take care of 5769 * them 5770 */ 5771 for_each_cpu(cpu, &flush_cpus) 5772 flush_work(per_cpu_ptr(&flush_works, cpu)); 5773 5774 put_online_cpus(); 5775 } 5776 5777 /* Pass the currently batched GRO_NORMAL SKBs up to the stack. */ 5778 static void gro_normal_list(struct napi_struct *napi) 5779 { 5780 if (!napi->rx_count) 5781 return; 5782 netif_receive_skb_list_internal(&napi->rx_list); 5783 INIT_LIST_HEAD(&napi->rx_list); 5784 napi->rx_count = 0; 5785 } 5786 5787 /* Queue one GRO_NORMAL SKB up for list processing. If batch size exceeded, 5788 * pass the whole batch up to the stack. 5789 */ 5790 static void gro_normal_one(struct napi_struct *napi, struct sk_buff *skb, int segs) 5791 { 5792 list_add_tail(&skb->list, &napi->rx_list); 5793 napi->rx_count += segs; 5794 if (napi->rx_count >= gro_normal_batch) 5795 gro_normal_list(napi); 5796 } 5797 5798 static int napi_gro_complete(struct napi_struct *napi, struct sk_buff *skb) 5799 { 5800 struct packet_offload *ptype; 5801 __be16 type = skb->protocol; 5802 struct list_head *head = &offload_base; 5803 int err = -ENOENT; 5804 5805 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb)); 5806 5807 if (NAPI_GRO_CB(skb)->count == 1) { 5808 skb_shinfo(skb)->gso_size = 0; 5809 goto out; 5810 } 5811 5812 rcu_read_lock(); 5813 list_for_each_entry_rcu(ptype, head, list) { 5814 if (ptype->type != type || !ptype->callbacks.gro_complete) 5815 continue; 5816 5817 err = INDIRECT_CALL_INET(ptype->callbacks.gro_complete, 5818 ipv6_gro_complete, inet_gro_complete, 5819 skb, 0); 5820 break; 5821 } 5822 rcu_read_unlock(); 5823 5824 if (err) { 5825 WARN_ON(&ptype->list == head); 5826 kfree_skb(skb); 5827 return NET_RX_SUCCESS; 5828 } 5829 5830 out: 5831 gro_normal_one(napi, skb, NAPI_GRO_CB(skb)->count); 5832 return NET_RX_SUCCESS; 5833 } 5834 5835 static void __napi_gro_flush_chain(struct napi_struct *napi, u32 index, 5836 bool flush_old) 5837 { 5838 struct list_head *head = &napi->gro_hash[index].list; 5839 struct sk_buff *skb, *p; 5840 5841 list_for_each_entry_safe_reverse(skb, p, head, list) { 5842 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies) 5843 return; 5844 skb_list_del_init(skb); 5845 napi_gro_complete(napi, skb); 5846 napi->gro_hash[index].count--; 5847 } 5848 5849 if (!napi->gro_hash[index].count) 5850 __clear_bit(index, &napi->gro_bitmask); 5851 } 5852 5853 /* napi->gro_hash[].list contains packets ordered by age. 5854 * youngest packets at the head of it. 5855 * Complete skbs in reverse order to reduce latencies. 5856 */ 5857 void napi_gro_flush(struct napi_struct *napi, bool flush_old) 5858 { 5859 unsigned long bitmask = napi->gro_bitmask; 5860 unsigned int i, base = ~0U; 5861 5862 while ((i = ffs(bitmask)) != 0) { 5863 bitmask >>= i; 5864 base += i; 5865 __napi_gro_flush_chain(napi, base, flush_old); 5866 } 5867 } 5868 EXPORT_SYMBOL(napi_gro_flush); 5869 5870 static void gro_list_prepare(const struct list_head *head, 5871 const struct sk_buff *skb) 5872 { 5873 unsigned int maclen = skb->dev->hard_header_len; 5874 u32 hash = skb_get_hash_raw(skb); 5875 struct sk_buff *p; 5876 5877 list_for_each_entry(p, head, list) { 5878 unsigned long diffs; 5879 5880 NAPI_GRO_CB(p)->flush = 0; 5881 5882 if (hash != skb_get_hash_raw(p)) { 5883 NAPI_GRO_CB(p)->same_flow = 0; 5884 continue; 5885 } 5886 5887 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev; 5888 diffs |= skb_vlan_tag_present(p) ^ skb_vlan_tag_present(skb); 5889 if (skb_vlan_tag_present(p)) 5890 diffs |= skb_vlan_tag_get(p) ^ skb_vlan_tag_get(skb); 5891 diffs |= skb_metadata_differs(p, skb); 5892 if (maclen == ETH_HLEN) 5893 diffs |= compare_ether_header(skb_mac_header(p), 5894 skb_mac_header(skb)); 5895 else if (!diffs) 5896 diffs = memcmp(skb_mac_header(p), 5897 skb_mac_header(skb), 5898 maclen); 5899 5900 /* in most common scenarions 'slow_gro' is 0 5901 * otherwise we are already on some slower paths 5902 * either skip all the infrequent tests altogether or 5903 * avoid trying too hard to skip each of them individually 5904 */ 5905 if (!diffs && unlikely(skb->slow_gro | p->slow_gro)) { 5906 #if IS_ENABLED(CONFIG_SKB_EXTENSIONS) && IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 5907 struct tc_skb_ext *skb_ext; 5908 struct tc_skb_ext *p_ext; 5909 #endif 5910 5911 diffs |= p->sk != skb->sk; 5912 diffs |= skb_metadata_dst_cmp(p, skb); 5913 diffs |= skb_get_nfct(p) ^ skb_get_nfct(skb); 5914 5915 #if IS_ENABLED(CONFIG_SKB_EXTENSIONS) && IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 5916 skb_ext = skb_ext_find(skb, TC_SKB_EXT); 5917 p_ext = skb_ext_find(p, TC_SKB_EXT); 5918 5919 diffs |= (!!p_ext) ^ (!!skb_ext); 5920 if (!diffs && unlikely(skb_ext)) 5921 diffs |= p_ext->chain ^ skb_ext->chain; 5922 #endif 5923 } 5924 5925 NAPI_GRO_CB(p)->same_flow = !diffs; 5926 } 5927 } 5928 5929 static inline void skb_gro_reset_offset(struct sk_buff *skb, u32 nhoff) 5930 { 5931 const struct skb_shared_info *pinfo = skb_shinfo(skb); 5932 const skb_frag_t *frag0 = &pinfo->frags[0]; 5933 5934 NAPI_GRO_CB(skb)->data_offset = 0; 5935 NAPI_GRO_CB(skb)->frag0 = NULL; 5936 NAPI_GRO_CB(skb)->frag0_len = 0; 5937 5938 if (!skb_headlen(skb) && pinfo->nr_frags && 5939 !PageHighMem(skb_frag_page(frag0)) && 5940 (!NET_IP_ALIGN || !((skb_frag_off(frag0) + nhoff) & 3))) { 5941 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0); 5942 NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int, 5943 skb_frag_size(frag0), 5944 skb->end - skb->tail); 5945 } 5946 } 5947 5948 static void gro_pull_from_frag0(struct sk_buff *skb, int grow) 5949 { 5950 struct skb_shared_info *pinfo = skb_shinfo(skb); 5951 5952 BUG_ON(skb->end - skb->tail < grow); 5953 5954 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); 5955 5956 skb->data_len -= grow; 5957 skb->tail += grow; 5958 5959 skb_frag_off_add(&pinfo->frags[0], grow); 5960 skb_frag_size_sub(&pinfo->frags[0], grow); 5961 5962 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) { 5963 skb_frag_unref(skb, 0); 5964 memmove(pinfo->frags, pinfo->frags + 1, 5965 --pinfo->nr_frags * sizeof(pinfo->frags[0])); 5966 } 5967 } 5968 5969 static void gro_flush_oldest(struct napi_struct *napi, struct list_head *head) 5970 { 5971 struct sk_buff *oldest; 5972 5973 oldest = list_last_entry(head, struct sk_buff, list); 5974 5975 /* We are called with head length >= MAX_GRO_SKBS, so this is 5976 * impossible. 5977 */ 5978 if (WARN_ON_ONCE(!oldest)) 5979 return; 5980 5981 /* Do not adjust napi->gro_hash[].count, caller is adding a new 5982 * SKB to the chain. 5983 */ 5984 skb_list_del_init(oldest); 5985 napi_gro_complete(napi, oldest); 5986 } 5987 5988 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) 5989 { 5990 u32 bucket = skb_get_hash_raw(skb) & (GRO_HASH_BUCKETS - 1); 5991 struct gro_list *gro_list = &napi->gro_hash[bucket]; 5992 struct list_head *head = &offload_base; 5993 struct packet_offload *ptype; 5994 __be16 type = skb->protocol; 5995 struct sk_buff *pp = NULL; 5996 enum gro_result ret; 5997 int same_flow; 5998 int grow; 5999 6000 if (netif_elide_gro(skb->dev)) 6001 goto normal; 6002 6003 gro_list_prepare(&gro_list->list, skb); 6004 6005 rcu_read_lock(); 6006 list_for_each_entry_rcu(ptype, head, list) { 6007 if (ptype->type != type || !ptype->callbacks.gro_receive) 6008 continue; 6009 6010 skb_set_network_header(skb, skb_gro_offset(skb)); 6011 skb_reset_mac_len(skb); 6012 NAPI_GRO_CB(skb)->same_flow = 0; 6013 NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb); 6014 NAPI_GRO_CB(skb)->free = 0; 6015 NAPI_GRO_CB(skb)->encap_mark = 0; 6016 NAPI_GRO_CB(skb)->recursion_counter = 0; 6017 NAPI_GRO_CB(skb)->is_fou = 0; 6018 NAPI_GRO_CB(skb)->is_atomic = 1; 6019 NAPI_GRO_CB(skb)->gro_remcsum_start = 0; 6020 6021 /* Setup for GRO checksum validation */ 6022 switch (skb->ip_summed) { 6023 case CHECKSUM_COMPLETE: 6024 NAPI_GRO_CB(skb)->csum = skb->csum; 6025 NAPI_GRO_CB(skb)->csum_valid = 1; 6026 NAPI_GRO_CB(skb)->csum_cnt = 0; 6027 break; 6028 case CHECKSUM_UNNECESSARY: 6029 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1; 6030 NAPI_GRO_CB(skb)->csum_valid = 0; 6031 break; 6032 default: 6033 NAPI_GRO_CB(skb)->csum_cnt = 0; 6034 NAPI_GRO_CB(skb)->csum_valid = 0; 6035 } 6036 6037 pp = INDIRECT_CALL_INET(ptype->callbacks.gro_receive, 6038 ipv6_gro_receive, inet_gro_receive, 6039 &gro_list->list, skb); 6040 break; 6041 } 6042 rcu_read_unlock(); 6043 6044 if (&ptype->list == head) 6045 goto normal; 6046 6047 if (PTR_ERR(pp) == -EINPROGRESS) { 6048 ret = GRO_CONSUMED; 6049 goto ok; 6050 } 6051 6052 same_flow = NAPI_GRO_CB(skb)->same_flow; 6053 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; 6054 6055 if (pp) { 6056 skb_list_del_init(pp); 6057 napi_gro_complete(napi, pp); 6058 gro_list->count--; 6059 } 6060 6061 if (same_flow) 6062 goto ok; 6063 6064 if (NAPI_GRO_CB(skb)->flush) 6065 goto normal; 6066 6067 if (unlikely(gro_list->count >= MAX_GRO_SKBS)) 6068 gro_flush_oldest(napi, &gro_list->list); 6069 else 6070 gro_list->count++; 6071 6072 NAPI_GRO_CB(skb)->count = 1; 6073 NAPI_GRO_CB(skb)->age = jiffies; 6074 NAPI_GRO_CB(skb)->last = skb; 6075 skb_shinfo(skb)->gso_size = skb_gro_len(skb); 6076 list_add(&skb->list, &gro_list->list); 6077 ret = GRO_HELD; 6078 6079 pull: 6080 grow = skb_gro_offset(skb) - skb_headlen(skb); 6081 if (grow > 0) 6082 gro_pull_from_frag0(skb, grow); 6083 ok: 6084 if (gro_list->count) { 6085 if (!test_bit(bucket, &napi->gro_bitmask)) 6086 __set_bit(bucket, &napi->gro_bitmask); 6087 } else if (test_bit(bucket, &napi->gro_bitmask)) { 6088 __clear_bit(bucket, &napi->gro_bitmask); 6089 } 6090 6091 return ret; 6092 6093 normal: 6094 ret = GRO_NORMAL; 6095 goto pull; 6096 } 6097 6098 struct packet_offload *gro_find_receive_by_type(__be16 type) 6099 { 6100 struct list_head *offload_head = &offload_base; 6101 struct packet_offload *ptype; 6102 6103 list_for_each_entry_rcu(ptype, offload_head, list) { 6104 if (ptype->type != type || !ptype->callbacks.gro_receive) 6105 continue; 6106 return ptype; 6107 } 6108 return NULL; 6109 } 6110 EXPORT_SYMBOL(gro_find_receive_by_type); 6111 6112 struct packet_offload *gro_find_complete_by_type(__be16 type) 6113 { 6114 struct list_head *offload_head = &offload_base; 6115 struct packet_offload *ptype; 6116 6117 list_for_each_entry_rcu(ptype, offload_head, list) { 6118 if (ptype->type != type || !ptype->callbacks.gro_complete) 6119 continue; 6120 return ptype; 6121 } 6122 return NULL; 6123 } 6124 EXPORT_SYMBOL(gro_find_complete_by_type); 6125 6126 static gro_result_t napi_skb_finish(struct napi_struct *napi, 6127 struct sk_buff *skb, 6128 gro_result_t ret) 6129 { 6130 switch (ret) { 6131 case GRO_NORMAL: 6132 gro_normal_one(napi, skb, 1); 6133 break; 6134 6135 case GRO_MERGED_FREE: 6136 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) 6137 napi_skb_free_stolen_head(skb); 6138 else if (skb->fclone != SKB_FCLONE_UNAVAILABLE) 6139 __kfree_skb(skb); 6140 else 6141 __kfree_skb_defer(skb); 6142 break; 6143 6144 case GRO_HELD: 6145 case GRO_MERGED: 6146 case GRO_CONSUMED: 6147 break; 6148 } 6149 6150 return ret; 6151 } 6152 6153 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) 6154 { 6155 gro_result_t ret; 6156 6157 skb_mark_napi_id(skb, napi); 6158 trace_napi_gro_receive_entry(skb); 6159 6160 skb_gro_reset_offset(skb, 0); 6161 6162 ret = napi_skb_finish(napi, skb, dev_gro_receive(napi, skb)); 6163 trace_napi_gro_receive_exit(ret); 6164 6165 return ret; 6166 } 6167 EXPORT_SYMBOL(napi_gro_receive); 6168 6169 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) 6170 { 6171 if (unlikely(skb->pfmemalloc)) { 6172 consume_skb(skb); 6173 return; 6174 } 6175 __skb_pull(skb, skb_headlen(skb)); 6176 /* restore the reserve we had after netdev_alloc_skb_ip_align() */ 6177 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb)); 6178 __vlan_hwaccel_clear_tag(skb); 6179 skb->dev = napi->dev; 6180 skb->skb_iif = 0; 6181 6182 /* eth_type_trans() assumes pkt_type is PACKET_HOST */ 6183 skb->pkt_type = PACKET_HOST; 6184 6185 skb->encapsulation = 0; 6186 skb_shinfo(skb)->gso_type = 0; 6187 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); 6188 if (unlikely(skb->slow_gro)) { 6189 skb_orphan(skb); 6190 skb_ext_reset(skb); 6191 nf_reset_ct(skb); 6192 skb->slow_gro = 0; 6193 } 6194 6195 napi->skb = skb; 6196 } 6197 6198 struct sk_buff *napi_get_frags(struct napi_struct *napi) 6199 { 6200 struct sk_buff *skb = napi->skb; 6201 6202 if (!skb) { 6203 skb = napi_alloc_skb(napi, GRO_MAX_HEAD); 6204 if (skb) { 6205 napi->skb = skb; 6206 skb_mark_napi_id(skb, napi); 6207 } 6208 } 6209 return skb; 6210 } 6211 EXPORT_SYMBOL(napi_get_frags); 6212 6213 static gro_result_t napi_frags_finish(struct napi_struct *napi, 6214 struct sk_buff *skb, 6215 gro_result_t ret) 6216 { 6217 switch (ret) { 6218 case GRO_NORMAL: 6219 case GRO_HELD: 6220 __skb_push(skb, ETH_HLEN); 6221 skb->protocol = eth_type_trans(skb, skb->dev); 6222 if (ret == GRO_NORMAL) 6223 gro_normal_one(napi, skb, 1); 6224 break; 6225 6226 case GRO_MERGED_FREE: 6227 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) 6228 napi_skb_free_stolen_head(skb); 6229 else 6230 napi_reuse_skb(napi, skb); 6231 break; 6232 6233 case GRO_MERGED: 6234 case GRO_CONSUMED: 6235 break; 6236 } 6237 6238 return ret; 6239 } 6240 6241 /* Upper GRO stack assumes network header starts at gro_offset=0 6242 * Drivers could call both napi_gro_frags() and napi_gro_receive() 6243 * We copy ethernet header into skb->data to have a common layout. 6244 */ 6245 static struct sk_buff *napi_frags_skb(struct napi_struct *napi) 6246 { 6247 struct sk_buff *skb = napi->skb; 6248 const struct ethhdr *eth; 6249 unsigned int hlen = sizeof(*eth); 6250 6251 napi->skb = NULL; 6252 6253 skb_reset_mac_header(skb); 6254 skb_gro_reset_offset(skb, hlen); 6255 6256 if (unlikely(skb_gro_header_hard(skb, hlen))) { 6257 eth = skb_gro_header_slow(skb, hlen, 0); 6258 if (unlikely(!eth)) { 6259 net_warn_ratelimited("%s: dropping impossible skb from %s\n", 6260 __func__, napi->dev->name); 6261 napi_reuse_skb(napi, skb); 6262 return NULL; 6263 } 6264 } else { 6265 eth = (const struct ethhdr *)skb->data; 6266 gro_pull_from_frag0(skb, hlen); 6267 NAPI_GRO_CB(skb)->frag0 += hlen; 6268 NAPI_GRO_CB(skb)->frag0_len -= hlen; 6269 } 6270 __skb_pull(skb, hlen); 6271 6272 /* 6273 * This works because the only protocols we care about don't require 6274 * special handling. 6275 * We'll fix it up properly in napi_frags_finish() 6276 */ 6277 skb->protocol = eth->h_proto; 6278 6279 return skb; 6280 } 6281 6282 gro_result_t napi_gro_frags(struct napi_struct *napi) 6283 { 6284 gro_result_t ret; 6285 struct sk_buff *skb = napi_frags_skb(napi); 6286 6287 trace_napi_gro_frags_entry(skb); 6288 6289 ret = napi_frags_finish(napi, skb, dev_gro_receive(napi, skb)); 6290 trace_napi_gro_frags_exit(ret); 6291 6292 return ret; 6293 } 6294 EXPORT_SYMBOL(napi_gro_frags); 6295 6296 /* Compute the checksum from gro_offset and return the folded value 6297 * after adding in any pseudo checksum. 6298 */ 6299 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb) 6300 { 6301 __wsum wsum; 6302 __sum16 sum; 6303 6304 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0); 6305 6306 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */ 6307 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum)); 6308 /* See comments in __skb_checksum_complete(). */ 6309 if (likely(!sum)) { 6310 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && 6311 !skb->csum_complete_sw) 6312 netdev_rx_csum_fault(skb->dev, skb); 6313 } 6314 6315 NAPI_GRO_CB(skb)->csum = wsum; 6316 NAPI_GRO_CB(skb)->csum_valid = 1; 6317 6318 return sum; 6319 } 6320 EXPORT_SYMBOL(__skb_gro_checksum_complete); 6321 6322 static void net_rps_send_ipi(struct softnet_data *remsd) 6323 { 6324 #ifdef CONFIG_RPS 6325 while (remsd) { 6326 struct softnet_data *next = remsd->rps_ipi_next; 6327 6328 if (cpu_online(remsd->cpu)) 6329 smp_call_function_single_async(remsd->cpu, &remsd->csd); 6330 remsd = next; 6331 } 6332 #endif 6333 } 6334 6335 /* 6336 * net_rps_action_and_irq_enable sends any pending IPI's for rps. 6337 * Note: called with local irq disabled, but exits with local irq enabled. 6338 */ 6339 static void net_rps_action_and_irq_enable(struct softnet_data *sd) 6340 { 6341 #ifdef CONFIG_RPS 6342 struct softnet_data *remsd = sd->rps_ipi_list; 6343 6344 if (remsd) { 6345 sd->rps_ipi_list = NULL; 6346 6347 local_irq_enable(); 6348 6349 /* Send pending IPI's to kick RPS processing on remote cpus. */ 6350 net_rps_send_ipi(remsd); 6351 } else 6352 #endif 6353 local_irq_enable(); 6354 } 6355 6356 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) 6357 { 6358 #ifdef CONFIG_RPS 6359 return sd->rps_ipi_list != NULL; 6360 #else 6361 return false; 6362 #endif 6363 } 6364 6365 static int process_backlog(struct napi_struct *napi, int quota) 6366 { 6367 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); 6368 bool again = true; 6369 int work = 0; 6370 6371 /* Check if we have pending ipi, its better to send them now, 6372 * not waiting net_rx_action() end. 6373 */ 6374 if (sd_has_rps_ipi_waiting(sd)) { 6375 local_irq_disable(); 6376 net_rps_action_and_irq_enable(sd); 6377 } 6378 6379 napi->weight = dev_rx_weight; 6380 while (again) { 6381 struct sk_buff *skb; 6382 6383 while ((skb = __skb_dequeue(&sd->process_queue))) { 6384 rcu_read_lock(); 6385 __netif_receive_skb(skb); 6386 rcu_read_unlock(); 6387 input_queue_head_incr(sd); 6388 if (++work >= quota) 6389 return work; 6390 6391 } 6392 6393 local_irq_disable(); 6394 rps_lock(sd); 6395 if (skb_queue_empty(&sd->input_pkt_queue)) { 6396 /* 6397 * Inline a custom version of __napi_complete(). 6398 * only current cpu owns and manipulates this napi, 6399 * and NAPI_STATE_SCHED is the only possible flag set 6400 * on backlog. 6401 * We can use a plain write instead of clear_bit(), 6402 * and we dont need an smp_mb() memory barrier. 6403 */ 6404 napi->state = 0; 6405 again = false; 6406 } else { 6407 skb_queue_splice_tail_init(&sd->input_pkt_queue, 6408 &sd->process_queue); 6409 } 6410 rps_unlock(sd); 6411 local_irq_enable(); 6412 } 6413 6414 return work; 6415 } 6416 6417 /** 6418 * __napi_schedule - schedule for receive 6419 * @n: entry to schedule 6420 * 6421 * The entry's receive function will be scheduled to run. 6422 * Consider using __napi_schedule_irqoff() if hard irqs are masked. 6423 */ 6424 void __napi_schedule(struct napi_struct *n) 6425 { 6426 unsigned long flags; 6427 6428 local_irq_save(flags); 6429 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6430 local_irq_restore(flags); 6431 } 6432 EXPORT_SYMBOL(__napi_schedule); 6433 6434 /** 6435 * napi_schedule_prep - check if napi can be scheduled 6436 * @n: napi context 6437 * 6438 * Test if NAPI routine is already running, and if not mark 6439 * it as running. This is used as a condition variable to 6440 * insure only one NAPI poll instance runs. We also make 6441 * sure there is no pending NAPI disable. 6442 */ 6443 bool napi_schedule_prep(struct napi_struct *n) 6444 { 6445 unsigned long val, new; 6446 6447 do { 6448 val = READ_ONCE(n->state); 6449 if (unlikely(val & NAPIF_STATE_DISABLE)) 6450 return false; 6451 new = val | NAPIF_STATE_SCHED; 6452 6453 /* Sets STATE_MISSED bit if STATE_SCHED was already set 6454 * This was suggested by Alexander Duyck, as compiler 6455 * emits better code than : 6456 * if (val & NAPIF_STATE_SCHED) 6457 * new |= NAPIF_STATE_MISSED; 6458 */ 6459 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * 6460 NAPIF_STATE_MISSED; 6461 } while (cmpxchg(&n->state, val, new) != val); 6462 6463 return !(val & NAPIF_STATE_SCHED); 6464 } 6465 EXPORT_SYMBOL(napi_schedule_prep); 6466 6467 /** 6468 * __napi_schedule_irqoff - schedule for receive 6469 * @n: entry to schedule 6470 * 6471 * Variant of __napi_schedule() assuming hard irqs are masked. 6472 * 6473 * On PREEMPT_RT enabled kernels this maps to __napi_schedule() 6474 * because the interrupt disabled assumption might not be true 6475 * due to force-threaded interrupts and spinlock substitution. 6476 */ 6477 void __napi_schedule_irqoff(struct napi_struct *n) 6478 { 6479 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6480 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6481 else 6482 __napi_schedule(n); 6483 } 6484 EXPORT_SYMBOL(__napi_schedule_irqoff); 6485 6486 bool napi_complete_done(struct napi_struct *n, int work_done) 6487 { 6488 unsigned long flags, val, new, timeout = 0; 6489 bool ret = true; 6490 6491 /* 6492 * 1) Don't let napi dequeue from the cpu poll list 6493 * just in case its running on a different cpu. 6494 * 2) If we are busy polling, do nothing here, we have 6495 * the guarantee we will be called later. 6496 */ 6497 if (unlikely(n->state & (NAPIF_STATE_NPSVC | 6498 NAPIF_STATE_IN_BUSY_POLL))) 6499 return false; 6500 6501 if (work_done) { 6502 if (n->gro_bitmask) 6503 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6504 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs); 6505 } 6506 if (n->defer_hard_irqs_count > 0) { 6507 n->defer_hard_irqs_count--; 6508 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6509 if (timeout) 6510 ret = false; 6511 } 6512 if (n->gro_bitmask) { 6513 /* When the NAPI instance uses a timeout and keeps postponing 6514 * it, we need to bound somehow the time packets are kept in 6515 * the GRO layer 6516 */ 6517 napi_gro_flush(n, !!timeout); 6518 } 6519 6520 gro_normal_list(n); 6521 6522 if (unlikely(!list_empty(&n->poll_list))) { 6523 /* If n->poll_list is not empty, we need to mask irqs */ 6524 local_irq_save(flags); 6525 list_del_init(&n->poll_list); 6526 local_irq_restore(flags); 6527 } 6528 6529 do { 6530 val = READ_ONCE(n->state); 6531 6532 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); 6533 6534 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED | 6535 NAPIF_STATE_SCHED_THREADED | 6536 NAPIF_STATE_PREFER_BUSY_POLL); 6537 6538 /* If STATE_MISSED was set, leave STATE_SCHED set, 6539 * because we will call napi->poll() one more time. 6540 * This C code was suggested by Alexander Duyck to help gcc. 6541 */ 6542 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * 6543 NAPIF_STATE_SCHED; 6544 } while (cmpxchg(&n->state, val, new) != val); 6545 6546 if (unlikely(val & NAPIF_STATE_MISSED)) { 6547 __napi_schedule(n); 6548 return false; 6549 } 6550 6551 if (timeout) 6552 hrtimer_start(&n->timer, ns_to_ktime(timeout), 6553 HRTIMER_MODE_REL_PINNED); 6554 return ret; 6555 } 6556 EXPORT_SYMBOL(napi_complete_done); 6557 6558 /* must be called under rcu_read_lock(), as we dont take a reference */ 6559 static struct napi_struct *napi_by_id(unsigned int napi_id) 6560 { 6561 unsigned int hash = napi_id % HASH_SIZE(napi_hash); 6562 struct napi_struct *napi; 6563 6564 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) 6565 if (napi->napi_id == napi_id) 6566 return napi; 6567 6568 return NULL; 6569 } 6570 6571 #if defined(CONFIG_NET_RX_BUSY_POLL) 6572 6573 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule) 6574 { 6575 if (!skip_schedule) { 6576 gro_normal_list(napi); 6577 __napi_schedule(napi); 6578 return; 6579 } 6580 6581 if (napi->gro_bitmask) { 6582 /* flush too old packets 6583 * If HZ < 1000, flush all packets. 6584 */ 6585 napi_gro_flush(napi, HZ >= 1000); 6586 } 6587 6588 gro_normal_list(napi); 6589 clear_bit(NAPI_STATE_SCHED, &napi->state); 6590 } 6591 6592 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll, 6593 u16 budget) 6594 { 6595 bool skip_schedule = false; 6596 unsigned long timeout; 6597 int rc; 6598 6599 /* Busy polling means there is a high chance device driver hard irq 6600 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was 6601 * set in napi_schedule_prep(). 6602 * Since we are about to call napi->poll() once more, we can safely 6603 * clear NAPI_STATE_MISSED. 6604 * 6605 * Note: x86 could use a single "lock and ..." instruction 6606 * to perform these two clear_bit() 6607 */ 6608 clear_bit(NAPI_STATE_MISSED, &napi->state); 6609 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); 6610 6611 local_bh_disable(); 6612 6613 if (prefer_busy_poll) { 6614 napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs); 6615 timeout = READ_ONCE(napi->dev->gro_flush_timeout); 6616 if (napi->defer_hard_irqs_count && timeout) { 6617 hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED); 6618 skip_schedule = true; 6619 } 6620 } 6621 6622 /* All we really want here is to re-enable device interrupts. 6623 * Ideally, a new ndo_busy_poll_stop() could avoid another round. 6624 */ 6625 rc = napi->poll(napi, budget); 6626 /* We can't gro_normal_list() here, because napi->poll() might have 6627 * rearmed the napi (napi_complete_done()) in which case it could 6628 * already be running on another CPU. 6629 */ 6630 trace_napi_poll(napi, rc, budget); 6631 netpoll_poll_unlock(have_poll_lock); 6632 if (rc == budget) 6633 __busy_poll_stop(napi, skip_schedule); 6634 local_bh_enable(); 6635 } 6636 6637 void napi_busy_loop(unsigned int napi_id, 6638 bool (*loop_end)(void *, unsigned long), 6639 void *loop_end_arg, bool prefer_busy_poll, u16 budget) 6640 { 6641 unsigned long start_time = loop_end ? busy_loop_current_time() : 0; 6642 int (*napi_poll)(struct napi_struct *napi, int budget); 6643 void *have_poll_lock = NULL; 6644 struct napi_struct *napi; 6645 6646 restart: 6647 napi_poll = NULL; 6648 6649 rcu_read_lock(); 6650 6651 napi = napi_by_id(napi_id); 6652 if (!napi) 6653 goto out; 6654 6655 preempt_disable(); 6656 for (;;) { 6657 int work = 0; 6658 6659 local_bh_disable(); 6660 if (!napi_poll) { 6661 unsigned long val = READ_ONCE(napi->state); 6662 6663 /* If multiple threads are competing for this napi, 6664 * we avoid dirtying napi->state as much as we can. 6665 */ 6666 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | 6667 NAPIF_STATE_IN_BUSY_POLL)) { 6668 if (prefer_busy_poll) 6669 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6670 goto count; 6671 } 6672 if (cmpxchg(&napi->state, val, 6673 val | NAPIF_STATE_IN_BUSY_POLL | 6674 NAPIF_STATE_SCHED) != val) { 6675 if (prefer_busy_poll) 6676 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6677 goto count; 6678 } 6679 have_poll_lock = netpoll_poll_lock(napi); 6680 napi_poll = napi->poll; 6681 } 6682 work = napi_poll(napi, budget); 6683 trace_napi_poll(napi, work, budget); 6684 gro_normal_list(napi); 6685 count: 6686 if (work > 0) 6687 __NET_ADD_STATS(dev_net(napi->dev), 6688 LINUX_MIB_BUSYPOLLRXPACKETS, work); 6689 local_bh_enable(); 6690 6691 if (!loop_end || loop_end(loop_end_arg, start_time)) 6692 break; 6693 6694 if (unlikely(need_resched())) { 6695 if (napi_poll) 6696 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6697 preempt_enable(); 6698 rcu_read_unlock(); 6699 cond_resched(); 6700 if (loop_end(loop_end_arg, start_time)) 6701 return; 6702 goto restart; 6703 } 6704 cpu_relax(); 6705 } 6706 if (napi_poll) 6707 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6708 preempt_enable(); 6709 out: 6710 rcu_read_unlock(); 6711 } 6712 EXPORT_SYMBOL(napi_busy_loop); 6713 6714 #endif /* CONFIG_NET_RX_BUSY_POLL */ 6715 6716 static void napi_hash_add(struct napi_struct *napi) 6717 { 6718 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state)) 6719 return; 6720 6721 spin_lock(&napi_hash_lock); 6722 6723 /* 0..NR_CPUS range is reserved for sender_cpu use */ 6724 do { 6725 if (unlikely(++napi_gen_id < MIN_NAPI_ID)) 6726 napi_gen_id = MIN_NAPI_ID; 6727 } while (napi_by_id(napi_gen_id)); 6728 napi->napi_id = napi_gen_id; 6729 6730 hlist_add_head_rcu(&napi->napi_hash_node, 6731 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); 6732 6733 spin_unlock(&napi_hash_lock); 6734 } 6735 6736 /* Warning : caller is responsible to make sure rcu grace period 6737 * is respected before freeing memory containing @napi 6738 */ 6739 static void napi_hash_del(struct napi_struct *napi) 6740 { 6741 spin_lock(&napi_hash_lock); 6742 6743 hlist_del_init_rcu(&napi->napi_hash_node); 6744 6745 spin_unlock(&napi_hash_lock); 6746 } 6747 6748 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) 6749 { 6750 struct napi_struct *napi; 6751 6752 napi = container_of(timer, struct napi_struct, timer); 6753 6754 /* Note : we use a relaxed variant of napi_schedule_prep() not setting 6755 * NAPI_STATE_MISSED, since we do not react to a device IRQ. 6756 */ 6757 if (!napi_disable_pending(napi) && 6758 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) { 6759 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6760 __napi_schedule_irqoff(napi); 6761 } 6762 6763 return HRTIMER_NORESTART; 6764 } 6765 6766 static void init_gro_hash(struct napi_struct *napi) 6767 { 6768 int i; 6769 6770 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6771 INIT_LIST_HEAD(&napi->gro_hash[i].list); 6772 napi->gro_hash[i].count = 0; 6773 } 6774 napi->gro_bitmask = 0; 6775 } 6776 6777 int dev_set_threaded(struct net_device *dev, bool threaded) 6778 { 6779 struct napi_struct *napi; 6780 int err = 0; 6781 6782 if (dev->threaded == threaded) 6783 return 0; 6784 6785 if (threaded) { 6786 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6787 if (!napi->thread) { 6788 err = napi_kthread_create(napi); 6789 if (err) { 6790 threaded = false; 6791 break; 6792 } 6793 } 6794 } 6795 } 6796 6797 dev->threaded = threaded; 6798 6799 /* Make sure kthread is created before THREADED bit 6800 * is set. 6801 */ 6802 smp_mb__before_atomic(); 6803 6804 /* Setting/unsetting threaded mode on a napi might not immediately 6805 * take effect, if the current napi instance is actively being 6806 * polled. In this case, the switch between threaded mode and 6807 * softirq mode will happen in the next round of napi_schedule(). 6808 * This should not cause hiccups/stalls to the live traffic. 6809 */ 6810 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6811 if (threaded) 6812 set_bit(NAPI_STATE_THREADED, &napi->state); 6813 else 6814 clear_bit(NAPI_STATE_THREADED, &napi->state); 6815 } 6816 6817 return err; 6818 } 6819 EXPORT_SYMBOL(dev_set_threaded); 6820 6821 void netif_napi_add(struct net_device *dev, struct napi_struct *napi, 6822 int (*poll)(struct napi_struct *, int), int weight) 6823 { 6824 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state))) 6825 return; 6826 6827 INIT_LIST_HEAD(&napi->poll_list); 6828 INIT_HLIST_NODE(&napi->napi_hash_node); 6829 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); 6830 napi->timer.function = napi_watchdog; 6831 init_gro_hash(napi); 6832 napi->skb = NULL; 6833 INIT_LIST_HEAD(&napi->rx_list); 6834 napi->rx_count = 0; 6835 napi->poll = poll; 6836 if (weight > NAPI_POLL_WEIGHT) 6837 netdev_err_once(dev, "%s() called with weight %d\n", __func__, 6838 weight); 6839 napi->weight = weight; 6840 napi->dev = dev; 6841 #ifdef CONFIG_NETPOLL 6842 napi->poll_owner = -1; 6843 #endif 6844 set_bit(NAPI_STATE_SCHED, &napi->state); 6845 set_bit(NAPI_STATE_NPSVC, &napi->state); 6846 list_add_rcu(&napi->dev_list, &dev->napi_list); 6847 napi_hash_add(napi); 6848 /* Create kthread for this napi if dev->threaded is set. 6849 * Clear dev->threaded if kthread creation failed so that 6850 * threaded mode will not be enabled in napi_enable(). 6851 */ 6852 if (dev->threaded && napi_kthread_create(napi)) 6853 dev->threaded = 0; 6854 } 6855 EXPORT_SYMBOL(netif_napi_add); 6856 6857 void napi_disable(struct napi_struct *n) 6858 { 6859 might_sleep(); 6860 set_bit(NAPI_STATE_DISABLE, &n->state); 6861 6862 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state)) 6863 msleep(1); 6864 while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state)) 6865 msleep(1); 6866 6867 hrtimer_cancel(&n->timer); 6868 6869 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &n->state); 6870 clear_bit(NAPI_STATE_DISABLE, &n->state); 6871 clear_bit(NAPI_STATE_THREADED, &n->state); 6872 } 6873 EXPORT_SYMBOL(napi_disable); 6874 6875 /** 6876 * napi_enable - enable NAPI scheduling 6877 * @n: NAPI context 6878 * 6879 * Resume NAPI from being scheduled on this context. 6880 * Must be paired with napi_disable. 6881 */ 6882 void napi_enable(struct napi_struct *n) 6883 { 6884 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state)); 6885 smp_mb__before_atomic(); 6886 clear_bit(NAPI_STATE_SCHED, &n->state); 6887 clear_bit(NAPI_STATE_NPSVC, &n->state); 6888 if (n->dev->threaded && n->thread) 6889 set_bit(NAPI_STATE_THREADED, &n->state); 6890 } 6891 EXPORT_SYMBOL(napi_enable); 6892 6893 static void flush_gro_hash(struct napi_struct *napi) 6894 { 6895 int i; 6896 6897 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6898 struct sk_buff *skb, *n; 6899 6900 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list) 6901 kfree_skb(skb); 6902 napi->gro_hash[i].count = 0; 6903 } 6904 } 6905 6906 /* Must be called in process context */ 6907 void __netif_napi_del(struct napi_struct *napi) 6908 { 6909 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state)) 6910 return; 6911 6912 napi_hash_del(napi); 6913 list_del_rcu(&napi->dev_list); 6914 napi_free_frags(napi); 6915 6916 flush_gro_hash(napi); 6917 napi->gro_bitmask = 0; 6918 6919 if (napi->thread) { 6920 kthread_stop(napi->thread); 6921 napi->thread = NULL; 6922 } 6923 } 6924 EXPORT_SYMBOL(__netif_napi_del); 6925 6926 static int __napi_poll(struct napi_struct *n, bool *repoll) 6927 { 6928 int work, weight; 6929 6930 weight = n->weight; 6931 6932 /* This NAPI_STATE_SCHED test is for avoiding a race 6933 * with netpoll's poll_napi(). Only the entity which 6934 * obtains the lock and sees NAPI_STATE_SCHED set will 6935 * actually make the ->poll() call. Therefore we avoid 6936 * accidentally calling ->poll() when NAPI is not scheduled. 6937 */ 6938 work = 0; 6939 if (test_bit(NAPI_STATE_SCHED, &n->state)) { 6940 work = n->poll(n, weight); 6941 trace_napi_poll(n, work, weight); 6942 } 6943 6944 if (unlikely(work > weight)) 6945 pr_err_once("NAPI poll function %pS returned %d, exceeding its budget of %d.\n", 6946 n->poll, work, weight); 6947 6948 if (likely(work < weight)) 6949 return work; 6950 6951 /* Drivers must not modify the NAPI state if they 6952 * consume the entire weight. In such cases this code 6953 * still "owns" the NAPI instance and therefore can 6954 * move the instance around on the list at-will. 6955 */ 6956 if (unlikely(napi_disable_pending(n))) { 6957 napi_complete(n); 6958 return work; 6959 } 6960 6961 /* The NAPI context has more processing work, but busy-polling 6962 * is preferred. Exit early. 6963 */ 6964 if (napi_prefer_busy_poll(n)) { 6965 if (napi_complete_done(n, work)) { 6966 /* If timeout is not set, we need to make sure 6967 * that the NAPI is re-scheduled. 6968 */ 6969 napi_schedule(n); 6970 } 6971 return work; 6972 } 6973 6974 if (n->gro_bitmask) { 6975 /* flush too old packets 6976 * If HZ < 1000, flush all packets. 6977 */ 6978 napi_gro_flush(n, HZ >= 1000); 6979 } 6980 6981 gro_normal_list(n); 6982 6983 /* Some drivers may have called napi_schedule 6984 * prior to exhausting their budget. 6985 */ 6986 if (unlikely(!list_empty(&n->poll_list))) { 6987 pr_warn_once("%s: Budget exhausted after napi rescheduled\n", 6988 n->dev ? n->dev->name : "backlog"); 6989 return work; 6990 } 6991 6992 *repoll = true; 6993 6994 return work; 6995 } 6996 6997 static int napi_poll(struct napi_struct *n, struct list_head *repoll) 6998 { 6999 bool do_repoll = false; 7000 void *have; 7001 int work; 7002 7003 list_del_init(&n->poll_list); 7004 7005 have = netpoll_poll_lock(n); 7006 7007 work = __napi_poll(n, &do_repoll); 7008 7009 if (do_repoll) 7010 list_add_tail(&n->poll_list, repoll); 7011 7012 netpoll_poll_unlock(have); 7013 7014 return work; 7015 } 7016 7017 static int napi_thread_wait(struct napi_struct *napi) 7018 { 7019 bool woken = false; 7020 7021 set_current_state(TASK_INTERRUPTIBLE); 7022 7023 while (!kthread_should_stop()) { 7024 /* Testing SCHED_THREADED bit here to make sure the current 7025 * kthread owns this napi and could poll on this napi. 7026 * Testing SCHED bit is not enough because SCHED bit might be 7027 * set by some other busy poll thread or by napi_disable(). 7028 */ 7029 if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state) || woken) { 7030 WARN_ON(!list_empty(&napi->poll_list)); 7031 __set_current_state(TASK_RUNNING); 7032 return 0; 7033 } 7034 7035 schedule(); 7036 /* woken being true indicates this thread owns this napi. */ 7037 woken = true; 7038 set_current_state(TASK_INTERRUPTIBLE); 7039 } 7040 __set_current_state(TASK_RUNNING); 7041 7042 return -1; 7043 } 7044 7045 static int napi_threaded_poll(void *data) 7046 { 7047 struct napi_struct *napi = data; 7048 void *have; 7049 7050 while (!napi_thread_wait(napi)) { 7051 for (;;) { 7052 bool repoll = false; 7053 7054 local_bh_disable(); 7055 7056 have = netpoll_poll_lock(napi); 7057 __napi_poll(napi, &repoll); 7058 netpoll_poll_unlock(have); 7059 7060 local_bh_enable(); 7061 7062 if (!repoll) 7063 break; 7064 7065 cond_resched(); 7066 } 7067 } 7068 return 0; 7069 } 7070 7071 static __latent_entropy void net_rx_action(struct softirq_action *h) 7072 { 7073 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 7074 unsigned long time_limit = jiffies + 7075 usecs_to_jiffies(netdev_budget_usecs); 7076 int budget = netdev_budget; 7077 LIST_HEAD(list); 7078 LIST_HEAD(repoll); 7079 7080 local_irq_disable(); 7081 list_splice_init(&sd->poll_list, &list); 7082 local_irq_enable(); 7083 7084 for (;;) { 7085 struct napi_struct *n; 7086 7087 if (list_empty(&list)) { 7088 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll)) 7089 return; 7090 break; 7091 } 7092 7093 n = list_first_entry(&list, struct napi_struct, poll_list); 7094 budget -= napi_poll(n, &repoll); 7095 7096 /* If softirq window is exhausted then punt. 7097 * Allow this to run for 2 jiffies since which will allow 7098 * an average latency of 1.5/HZ. 7099 */ 7100 if (unlikely(budget <= 0 || 7101 time_after_eq(jiffies, time_limit))) { 7102 sd->time_squeeze++; 7103 break; 7104 } 7105 } 7106 7107 local_irq_disable(); 7108 7109 list_splice_tail_init(&sd->poll_list, &list); 7110 list_splice_tail(&repoll, &list); 7111 list_splice(&list, &sd->poll_list); 7112 if (!list_empty(&sd->poll_list)) 7113 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 7114 7115 net_rps_action_and_irq_enable(sd); 7116 } 7117 7118 struct netdev_adjacent { 7119 struct net_device *dev; 7120 7121 /* upper master flag, there can only be one master device per list */ 7122 bool master; 7123 7124 /* lookup ignore flag */ 7125 bool ignore; 7126 7127 /* counter for the number of times this device was added to us */ 7128 u16 ref_nr; 7129 7130 /* private field for the users */ 7131 void *private; 7132 7133 struct list_head list; 7134 struct rcu_head rcu; 7135 }; 7136 7137 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, 7138 struct list_head *adj_list) 7139 { 7140 struct netdev_adjacent *adj; 7141 7142 list_for_each_entry(adj, adj_list, list) { 7143 if (adj->dev == adj_dev) 7144 return adj; 7145 } 7146 return NULL; 7147 } 7148 7149 static int ____netdev_has_upper_dev(struct net_device *upper_dev, 7150 struct netdev_nested_priv *priv) 7151 { 7152 struct net_device *dev = (struct net_device *)priv->data; 7153 7154 return upper_dev == dev; 7155 } 7156 7157 /** 7158 * netdev_has_upper_dev - Check if device is linked to an upper device 7159 * @dev: device 7160 * @upper_dev: upper device to check 7161 * 7162 * Find out if a device is linked to specified upper device and return true 7163 * in case it is. Note that this checks only immediate upper device, 7164 * not through a complete stack of devices. The caller must hold the RTNL lock. 7165 */ 7166 bool netdev_has_upper_dev(struct net_device *dev, 7167 struct net_device *upper_dev) 7168 { 7169 struct netdev_nested_priv priv = { 7170 .data = (void *)upper_dev, 7171 }; 7172 7173 ASSERT_RTNL(); 7174 7175 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 7176 &priv); 7177 } 7178 EXPORT_SYMBOL(netdev_has_upper_dev); 7179 7180 /** 7181 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device 7182 * @dev: device 7183 * @upper_dev: upper device to check 7184 * 7185 * Find out if a device is linked to specified upper device and return true 7186 * in case it is. Note that this checks the entire upper device chain. 7187 * The caller must hold rcu lock. 7188 */ 7189 7190 bool netdev_has_upper_dev_all_rcu(struct net_device *dev, 7191 struct net_device *upper_dev) 7192 { 7193 struct netdev_nested_priv priv = { 7194 .data = (void *)upper_dev, 7195 }; 7196 7197 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 7198 &priv); 7199 } 7200 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); 7201 7202 /** 7203 * netdev_has_any_upper_dev - Check if device is linked to some device 7204 * @dev: device 7205 * 7206 * Find out if a device is linked to an upper device and return true in case 7207 * it is. The caller must hold the RTNL lock. 7208 */ 7209 bool netdev_has_any_upper_dev(struct net_device *dev) 7210 { 7211 ASSERT_RTNL(); 7212 7213 return !list_empty(&dev->adj_list.upper); 7214 } 7215 EXPORT_SYMBOL(netdev_has_any_upper_dev); 7216 7217 /** 7218 * netdev_master_upper_dev_get - Get master upper device 7219 * @dev: device 7220 * 7221 * Find a master upper device and return pointer to it or NULL in case 7222 * it's not there. The caller must hold the RTNL lock. 7223 */ 7224 struct net_device *netdev_master_upper_dev_get(struct net_device *dev) 7225 { 7226 struct netdev_adjacent *upper; 7227 7228 ASSERT_RTNL(); 7229 7230 if (list_empty(&dev->adj_list.upper)) 7231 return NULL; 7232 7233 upper = list_first_entry(&dev->adj_list.upper, 7234 struct netdev_adjacent, list); 7235 if (likely(upper->master)) 7236 return upper->dev; 7237 return NULL; 7238 } 7239 EXPORT_SYMBOL(netdev_master_upper_dev_get); 7240 7241 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev) 7242 { 7243 struct netdev_adjacent *upper; 7244 7245 ASSERT_RTNL(); 7246 7247 if (list_empty(&dev->adj_list.upper)) 7248 return NULL; 7249 7250 upper = list_first_entry(&dev->adj_list.upper, 7251 struct netdev_adjacent, list); 7252 if (likely(upper->master) && !upper->ignore) 7253 return upper->dev; 7254 return NULL; 7255 } 7256 7257 /** 7258 * netdev_has_any_lower_dev - Check if device is linked to some device 7259 * @dev: device 7260 * 7261 * Find out if a device is linked to a lower device and return true in case 7262 * it is. The caller must hold the RTNL lock. 7263 */ 7264 static bool netdev_has_any_lower_dev(struct net_device *dev) 7265 { 7266 ASSERT_RTNL(); 7267 7268 return !list_empty(&dev->adj_list.lower); 7269 } 7270 7271 void *netdev_adjacent_get_private(struct list_head *adj_list) 7272 { 7273 struct netdev_adjacent *adj; 7274 7275 adj = list_entry(adj_list, struct netdev_adjacent, list); 7276 7277 return adj->private; 7278 } 7279 EXPORT_SYMBOL(netdev_adjacent_get_private); 7280 7281 /** 7282 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list 7283 * @dev: device 7284 * @iter: list_head ** of the current position 7285 * 7286 * Gets the next device from the dev's upper list, starting from iter 7287 * position. The caller must hold RCU read lock. 7288 */ 7289 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, 7290 struct list_head **iter) 7291 { 7292 struct netdev_adjacent *upper; 7293 7294 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 7295 7296 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7297 7298 if (&upper->list == &dev->adj_list.upper) 7299 return NULL; 7300 7301 *iter = &upper->list; 7302 7303 return upper->dev; 7304 } 7305 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); 7306 7307 static struct net_device *__netdev_next_upper_dev(struct net_device *dev, 7308 struct list_head **iter, 7309 bool *ignore) 7310 { 7311 struct netdev_adjacent *upper; 7312 7313 upper = list_entry((*iter)->next, struct netdev_adjacent, list); 7314 7315 if (&upper->list == &dev->adj_list.upper) 7316 return NULL; 7317 7318 *iter = &upper->list; 7319 *ignore = upper->ignore; 7320 7321 return upper->dev; 7322 } 7323 7324 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, 7325 struct list_head **iter) 7326 { 7327 struct netdev_adjacent *upper; 7328 7329 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 7330 7331 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7332 7333 if (&upper->list == &dev->adj_list.upper) 7334 return NULL; 7335 7336 *iter = &upper->list; 7337 7338 return upper->dev; 7339 } 7340 7341 static int __netdev_walk_all_upper_dev(struct net_device *dev, 7342 int (*fn)(struct net_device *dev, 7343 struct netdev_nested_priv *priv), 7344 struct netdev_nested_priv *priv) 7345 { 7346 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7347 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7348 int ret, cur = 0; 7349 bool ignore; 7350 7351 now = dev; 7352 iter = &dev->adj_list.upper; 7353 7354 while (1) { 7355 if (now != dev) { 7356 ret = fn(now, priv); 7357 if (ret) 7358 return ret; 7359 } 7360 7361 next = NULL; 7362 while (1) { 7363 udev = __netdev_next_upper_dev(now, &iter, &ignore); 7364 if (!udev) 7365 break; 7366 if (ignore) 7367 continue; 7368 7369 next = udev; 7370 niter = &udev->adj_list.upper; 7371 dev_stack[cur] = now; 7372 iter_stack[cur++] = iter; 7373 break; 7374 } 7375 7376 if (!next) { 7377 if (!cur) 7378 return 0; 7379 next = dev_stack[--cur]; 7380 niter = iter_stack[cur]; 7381 } 7382 7383 now = next; 7384 iter = niter; 7385 } 7386 7387 return 0; 7388 } 7389 7390 int netdev_walk_all_upper_dev_rcu(struct net_device *dev, 7391 int (*fn)(struct net_device *dev, 7392 struct netdev_nested_priv *priv), 7393 struct netdev_nested_priv *priv) 7394 { 7395 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7396 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7397 int ret, cur = 0; 7398 7399 now = dev; 7400 iter = &dev->adj_list.upper; 7401 7402 while (1) { 7403 if (now != dev) { 7404 ret = fn(now, priv); 7405 if (ret) 7406 return ret; 7407 } 7408 7409 next = NULL; 7410 while (1) { 7411 udev = netdev_next_upper_dev_rcu(now, &iter); 7412 if (!udev) 7413 break; 7414 7415 next = udev; 7416 niter = &udev->adj_list.upper; 7417 dev_stack[cur] = now; 7418 iter_stack[cur++] = iter; 7419 break; 7420 } 7421 7422 if (!next) { 7423 if (!cur) 7424 return 0; 7425 next = dev_stack[--cur]; 7426 niter = iter_stack[cur]; 7427 } 7428 7429 now = next; 7430 iter = niter; 7431 } 7432 7433 return 0; 7434 } 7435 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); 7436 7437 static bool __netdev_has_upper_dev(struct net_device *dev, 7438 struct net_device *upper_dev) 7439 { 7440 struct netdev_nested_priv priv = { 7441 .flags = 0, 7442 .data = (void *)upper_dev, 7443 }; 7444 7445 ASSERT_RTNL(); 7446 7447 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev, 7448 &priv); 7449 } 7450 7451 /** 7452 * netdev_lower_get_next_private - Get the next ->private from the 7453 * lower neighbour list 7454 * @dev: device 7455 * @iter: list_head ** of the current position 7456 * 7457 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7458 * list, starting from iter position. The caller must hold either hold the 7459 * RTNL lock or its own locking that guarantees that the neighbour lower 7460 * list will remain unchanged. 7461 */ 7462 void *netdev_lower_get_next_private(struct net_device *dev, 7463 struct list_head **iter) 7464 { 7465 struct netdev_adjacent *lower; 7466 7467 lower = list_entry(*iter, struct netdev_adjacent, list); 7468 7469 if (&lower->list == &dev->adj_list.lower) 7470 return NULL; 7471 7472 *iter = lower->list.next; 7473 7474 return lower->private; 7475 } 7476 EXPORT_SYMBOL(netdev_lower_get_next_private); 7477 7478 /** 7479 * netdev_lower_get_next_private_rcu - Get the next ->private from the 7480 * lower neighbour list, RCU 7481 * variant 7482 * @dev: device 7483 * @iter: list_head ** of the current position 7484 * 7485 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7486 * list, starting from iter position. The caller must hold RCU read lock. 7487 */ 7488 void *netdev_lower_get_next_private_rcu(struct net_device *dev, 7489 struct list_head **iter) 7490 { 7491 struct netdev_adjacent *lower; 7492 7493 WARN_ON_ONCE(!rcu_read_lock_held()); 7494 7495 lower = list_entry_rcu((*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->private; 7503 } 7504 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); 7505 7506 /** 7507 * netdev_lower_get_next - Get the next device from the lower neighbour 7508 * list 7509 * @dev: device 7510 * @iter: list_head ** of the current position 7511 * 7512 * Gets the next netdev_adjacent from the dev's lower neighbour 7513 * list, starting from iter position. The caller must hold RTNL lock or 7514 * its own locking that guarantees that the neighbour lower 7515 * list will remain unchanged. 7516 */ 7517 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) 7518 { 7519 struct netdev_adjacent *lower; 7520 7521 lower = list_entry(*iter, struct netdev_adjacent, list); 7522 7523 if (&lower->list == &dev->adj_list.lower) 7524 return NULL; 7525 7526 *iter = lower->list.next; 7527 7528 return lower->dev; 7529 } 7530 EXPORT_SYMBOL(netdev_lower_get_next); 7531 7532 static struct net_device *netdev_next_lower_dev(struct net_device *dev, 7533 struct list_head **iter) 7534 { 7535 struct netdev_adjacent *lower; 7536 7537 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7538 7539 if (&lower->list == &dev->adj_list.lower) 7540 return NULL; 7541 7542 *iter = &lower->list; 7543 7544 return lower->dev; 7545 } 7546 7547 static struct net_device *__netdev_next_lower_dev(struct net_device *dev, 7548 struct list_head **iter, 7549 bool *ignore) 7550 { 7551 struct netdev_adjacent *lower; 7552 7553 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7554 7555 if (&lower->list == &dev->adj_list.lower) 7556 return NULL; 7557 7558 *iter = &lower->list; 7559 *ignore = lower->ignore; 7560 7561 return lower->dev; 7562 } 7563 7564 int netdev_walk_all_lower_dev(struct net_device *dev, 7565 int (*fn)(struct net_device *dev, 7566 struct netdev_nested_priv *priv), 7567 struct netdev_nested_priv *priv) 7568 { 7569 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7570 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7571 int ret, cur = 0; 7572 7573 now = dev; 7574 iter = &dev->adj_list.lower; 7575 7576 while (1) { 7577 if (now != dev) { 7578 ret = fn(now, priv); 7579 if (ret) 7580 return ret; 7581 } 7582 7583 next = NULL; 7584 while (1) { 7585 ldev = netdev_next_lower_dev(now, &iter); 7586 if (!ldev) 7587 break; 7588 7589 next = ldev; 7590 niter = &ldev->adj_list.lower; 7591 dev_stack[cur] = now; 7592 iter_stack[cur++] = iter; 7593 break; 7594 } 7595 7596 if (!next) { 7597 if (!cur) 7598 return 0; 7599 next = dev_stack[--cur]; 7600 niter = iter_stack[cur]; 7601 } 7602 7603 now = next; 7604 iter = niter; 7605 } 7606 7607 return 0; 7608 } 7609 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); 7610 7611 static int __netdev_walk_all_lower_dev(struct net_device *dev, 7612 int (*fn)(struct net_device *dev, 7613 struct netdev_nested_priv *priv), 7614 struct netdev_nested_priv *priv) 7615 { 7616 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7617 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7618 int ret, cur = 0; 7619 bool ignore; 7620 7621 now = dev; 7622 iter = &dev->adj_list.lower; 7623 7624 while (1) { 7625 if (now != dev) { 7626 ret = fn(now, priv); 7627 if (ret) 7628 return ret; 7629 } 7630 7631 next = NULL; 7632 while (1) { 7633 ldev = __netdev_next_lower_dev(now, &iter, &ignore); 7634 if (!ldev) 7635 break; 7636 if (ignore) 7637 continue; 7638 7639 next = ldev; 7640 niter = &ldev->adj_list.lower; 7641 dev_stack[cur] = now; 7642 iter_stack[cur++] = iter; 7643 break; 7644 } 7645 7646 if (!next) { 7647 if (!cur) 7648 return 0; 7649 next = dev_stack[--cur]; 7650 niter = iter_stack[cur]; 7651 } 7652 7653 now = next; 7654 iter = niter; 7655 } 7656 7657 return 0; 7658 } 7659 7660 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, 7661 struct list_head **iter) 7662 { 7663 struct netdev_adjacent *lower; 7664 7665 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7666 if (&lower->list == &dev->adj_list.lower) 7667 return NULL; 7668 7669 *iter = &lower->list; 7670 7671 return lower->dev; 7672 } 7673 EXPORT_SYMBOL(netdev_next_lower_dev_rcu); 7674 7675 static u8 __netdev_upper_depth(struct net_device *dev) 7676 { 7677 struct net_device *udev; 7678 struct list_head *iter; 7679 u8 max_depth = 0; 7680 bool ignore; 7681 7682 for (iter = &dev->adj_list.upper, 7683 udev = __netdev_next_upper_dev(dev, &iter, &ignore); 7684 udev; 7685 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) { 7686 if (ignore) 7687 continue; 7688 if (max_depth < udev->upper_level) 7689 max_depth = udev->upper_level; 7690 } 7691 7692 return max_depth; 7693 } 7694 7695 static u8 __netdev_lower_depth(struct net_device *dev) 7696 { 7697 struct net_device *ldev; 7698 struct list_head *iter; 7699 u8 max_depth = 0; 7700 bool ignore; 7701 7702 for (iter = &dev->adj_list.lower, 7703 ldev = __netdev_next_lower_dev(dev, &iter, &ignore); 7704 ldev; 7705 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) { 7706 if (ignore) 7707 continue; 7708 if (max_depth < ldev->lower_level) 7709 max_depth = ldev->lower_level; 7710 } 7711 7712 return max_depth; 7713 } 7714 7715 static int __netdev_update_upper_level(struct net_device *dev, 7716 struct netdev_nested_priv *__unused) 7717 { 7718 dev->upper_level = __netdev_upper_depth(dev) + 1; 7719 return 0; 7720 } 7721 7722 static int __netdev_update_lower_level(struct net_device *dev, 7723 struct netdev_nested_priv *priv) 7724 { 7725 dev->lower_level = __netdev_lower_depth(dev) + 1; 7726 7727 #ifdef CONFIG_LOCKDEP 7728 if (!priv) 7729 return 0; 7730 7731 if (priv->flags & NESTED_SYNC_IMM) 7732 dev->nested_level = dev->lower_level - 1; 7733 if (priv->flags & NESTED_SYNC_TODO) 7734 net_unlink_todo(dev); 7735 #endif 7736 return 0; 7737 } 7738 7739 int netdev_walk_all_lower_dev_rcu(struct net_device *dev, 7740 int (*fn)(struct net_device *dev, 7741 struct netdev_nested_priv *priv), 7742 struct netdev_nested_priv *priv) 7743 { 7744 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7745 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7746 int ret, cur = 0; 7747 7748 now = dev; 7749 iter = &dev->adj_list.lower; 7750 7751 while (1) { 7752 if (now != dev) { 7753 ret = fn(now, priv); 7754 if (ret) 7755 return ret; 7756 } 7757 7758 next = NULL; 7759 while (1) { 7760 ldev = netdev_next_lower_dev_rcu(now, &iter); 7761 if (!ldev) 7762 break; 7763 7764 next = ldev; 7765 niter = &ldev->adj_list.lower; 7766 dev_stack[cur] = now; 7767 iter_stack[cur++] = iter; 7768 break; 7769 } 7770 7771 if (!next) { 7772 if (!cur) 7773 return 0; 7774 next = dev_stack[--cur]; 7775 niter = iter_stack[cur]; 7776 } 7777 7778 now = next; 7779 iter = niter; 7780 } 7781 7782 return 0; 7783 } 7784 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); 7785 7786 /** 7787 * netdev_lower_get_first_private_rcu - Get the first ->private from the 7788 * lower neighbour list, RCU 7789 * variant 7790 * @dev: device 7791 * 7792 * Gets the first netdev_adjacent->private from the dev's lower neighbour 7793 * list. The caller must hold RCU read lock. 7794 */ 7795 void *netdev_lower_get_first_private_rcu(struct net_device *dev) 7796 { 7797 struct netdev_adjacent *lower; 7798 7799 lower = list_first_or_null_rcu(&dev->adj_list.lower, 7800 struct netdev_adjacent, list); 7801 if (lower) 7802 return lower->private; 7803 return NULL; 7804 } 7805 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); 7806 7807 /** 7808 * netdev_master_upper_dev_get_rcu - Get master upper device 7809 * @dev: device 7810 * 7811 * Find a master upper device and return pointer to it or NULL in case 7812 * it's not there. The caller must hold the RCU read lock. 7813 */ 7814 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) 7815 { 7816 struct netdev_adjacent *upper; 7817 7818 upper = list_first_or_null_rcu(&dev->adj_list.upper, 7819 struct netdev_adjacent, list); 7820 if (upper && likely(upper->master)) 7821 return upper->dev; 7822 return NULL; 7823 } 7824 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); 7825 7826 static int netdev_adjacent_sysfs_add(struct net_device *dev, 7827 struct net_device *adj_dev, 7828 struct list_head *dev_list) 7829 { 7830 char linkname[IFNAMSIZ+7]; 7831 7832 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7833 "upper_%s" : "lower_%s", adj_dev->name); 7834 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), 7835 linkname); 7836 } 7837 static void netdev_adjacent_sysfs_del(struct net_device *dev, 7838 char *name, 7839 struct list_head *dev_list) 7840 { 7841 char linkname[IFNAMSIZ+7]; 7842 7843 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7844 "upper_%s" : "lower_%s", name); 7845 sysfs_remove_link(&(dev->dev.kobj), linkname); 7846 } 7847 7848 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, 7849 struct net_device *adj_dev, 7850 struct list_head *dev_list) 7851 { 7852 return (dev_list == &dev->adj_list.upper || 7853 dev_list == &dev->adj_list.lower) && 7854 net_eq(dev_net(dev), dev_net(adj_dev)); 7855 } 7856 7857 static int __netdev_adjacent_dev_insert(struct net_device *dev, 7858 struct net_device *adj_dev, 7859 struct list_head *dev_list, 7860 void *private, bool master) 7861 { 7862 struct netdev_adjacent *adj; 7863 int ret; 7864 7865 adj = __netdev_find_adj(adj_dev, dev_list); 7866 7867 if (adj) { 7868 adj->ref_nr += 1; 7869 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", 7870 dev->name, adj_dev->name, adj->ref_nr); 7871 7872 return 0; 7873 } 7874 7875 adj = kmalloc(sizeof(*adj), GFP_KERNEL); 7876 if (!adj) 7877 return -ENOMEM; 7878 7879 adj->dev = adj_dev; 7880 adj->master = master; 7881 adj->ref_nr = 1; 7882 adj->private = private; 7883 adj->ignore = false; 7884 dev_hold(adj_dev); 7885 7886 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", 7887 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); 7888 7889 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { 7890 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); 7891 if (ret) 7892 goto free_adj; 7893 } 7894 7895 /* Ensure that master link is always the first item in list. */ 7896 if (master) { 7897 ret = sysfs_create_link(&(dev->dev.kobj), 7898 &(adj_dev->dev.kobj), "master"); 7899 if (ret) 7900 goto remove_symlinks; 7901 7902 list_add_rcu(&adj->list, dev_list); 7903 } else { 7904 list_add_tail_rcu(&adj->list, dev_list); 7905 } 7906 7907 return 0; 7908 7909 remove_symlinks: 7910 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7911 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7912 free_adj: 7913 kfree(adj); 7914 dev_put(adj_dev); 7915 7916 return ret; 7917 } 7918 7919 static void __netdev_adjacent_dev_remove(struct net_device *dev, 7920 struct net_device *adj_dev, 7921 u16 ref_nr, 7922 struct list_head *dev_list) 7923 { 7924 struct netdev_adjacent *adj; 7925 7926 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", 7927 dev->name, adj_dev->name, ref_nr); 7928 7929 adj = __netdev_find_adj(adj_dev, dev_list); 7930 7931 if (!adj) { 7932 pr_err("Adjacency does not exist for device %s from %s\n", 7933 dev->name, adj_dev->name); 7934 WARN_ON(1); 7935 return; 7936 } 7937 7938 if (adj->ref_nr > ref_nr) { 7939 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", 7940 dev->name, adj_dev->name, ref_nr, 7941 adj->ref_nr - ref_nr); 7942 adj->ref_nr -= ref_nr; 7943 return; 7944 } 7945 7946 if (adj->master) 7947 sysfs_remove_link(&(dev->dev.kobj), "master"); 7948 7949 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7950 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7951 7952 list_del_rcu(&adj->list); 7953 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", 7954 adj_dev->name, dev->name, adj_dev->name); 7955 dev_put(adj_dev); 7956 kfree_rcu(adj, rcu); 7957 } 7958 7959 static int __netdev_adjacent_dev_link_lists(struct net_device *dev, 7960 struct net_device *upper_dev, 7961 struct list_head *up_list, 7962 struct list_head *down_list, 7963 void *private, bool master) 7964 { 7965 int ret; 7966 7967 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, 7968 private, master); 7969 if (ret) 7970 return ret; 7971 7972 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, 7973 private, false); 7974 if (ret) { 7975 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); 7976 return ret; 7977 } 7978 7979 return 0; 7980 } 7981 7982 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, 7983 struct net_device *upper_dev, 7984 u16 ref_nr, 7985 struct list_head *up_list, 7986 struct list_head *down_list) 7987 { 7988 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); 7989 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); 7990 } 7991 7992 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, 7993 struct net_device *upper_dev, 7994 void *private, bool master) 7995 { 7996 return __netdev_adjacent_dev_link_lists(dev, upper_dev, 7997 &dev->adj_list.upper, 7998 &upper_dev->adj_list.lower, 7999 private, master); 8000 } 8001 8002 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, 8003 struct net_device *upper_dev) 8004 { 8005 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, 8006 &dev->adj_list.upper, 8007 &upper_dev->adj_list.lower); 8008 } 8009 8010 static int __netdev_upper_dev_link(struct net_device *dev, 8011 struct net_device *upper_dev, bool master, 8012 void *upper_priv, void *upper_info, 8013 struct netdev_nested_priv *priv, 8014 struct netlink_ext_ack *extack) 8015 { 8016 struct netdev_notifier_changeupper_info changeupper_info = { 8017 .info = { 8018 .dev = dev, 8019 .extack = extack, 8020 }, 8021 .upper_dev = upper_dev, 8022 .master = master, 8023 .linking = true, 8024 .upper_info = upper_info, 8025 }; 8026 struct net_device *master_dev; 8027 int ret = 0; 8028 8029 ASSERT_RTNL(); 8030 8031 if (dev == upper_dev) 8032 return -EBUSY; 8033 8034 /* To prevent loops, check if dev is not upper device to upper_dev. */ 8035 if (__netdev_has_upper_dev(upper_dev, dev)) 8036 return -EBUSY; 8037 8038 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV) 8039 return -EMLINK; 8040 8041 if (!master) { 8042 if (__netdev_has_upper_dev(dev, upper_dev)) 8043 return -EEXIST; 8044 } else { 8045 master_dev = __netdev_master_upper_dev_get(dev); 8046 if (master_dev) 8047 return master_dev == upper_dev ? -EEXIST : -EBUSY; 8048 } 8049 8050 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 8051 &changeupper_info.info); 8052 ret = notifier_to_errno(ret); 8053 if (ret) 8054 return ret; 8055 8056 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, 8057 master); 8058 if (ret) 8059 return ret; 8060 8061 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 8062 &changeupper_info.info); 8063 ret = notifier_to_errno(ret); 8064 if (ret) 8065 goto rollback; 8066 8067 __netdev_update_upper_level(dev, NULL); 8068 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 8069 8070 __netdev_update_lower_level(upper_dev, priv); 8071 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 8072 priv); 8073 8074 return 0; 8075 8076 rollback: 8077 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 8078 8079 return ret; 8080 } 8081 8082 /** 8083 * netdev_upper_dev_link - Add a link to the upper device 8084 * @dev: device 8085 * @upper_dev: new upper device 8086 * @extack: netlink extended ack 8087 * 8088 * Adds a link to device which is upper to this one. The caller must hold 8089 * the RTNL lock. On a failure a negative errno code is returned. 8090 * On success the reference counts are adjusted and the function 8091 * returns zero. 8092 */ 8093 int netdev_upper_dev_link(struct net_device *dev, 8094 struct net_device *upper_dev, 8095 struct netlink_ext_ack *extack) 8096 { 8097 struct netdev_nested_priv priv = { 8098 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 8099 .data = NULL, 8100 }; 8101 8102 return __netdev_upper_dev_link(dev, upper_dev, false, 8103 NULL, NULL, &priv, extack); 8104 } 8105 EXPORT_SYMBOL(netdev_upper_dev_link); 8106 8107 /** 8108 * netdev_master_upper_dev_link - Add a master link to the upper device 8109 * @dev: device 8110 * @upper_dev: new upper device 8111 * @upper_priv: upper device private 8112 * @upper_info: upper info to be passed down via notifier 8113 * @extack: netlink extended ack 8114 * 8115 * Adds a link to device which is upper to this one. In this case, only 8116 * one master upper device can be linked, although other non-master devices 8117 * might be linked as well. The caller must hold the RTNL lock. 8118 * On a failure a negative errno code is returned. On success the reference 8119 * counts are adjusted and the function returns zero. 8120 */ 8121 int netdev_master_upper_dev_link(struct net_device *dev, 8122 struct net_device *upper_dev, 8123 void *upper_priv, void *upper_info, 8124 struct netlink_ext_ack *extack) 8125 { 8126 struct netdev_nested_priv priv = { 8127 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 8128 .data = NULL, 8129 }; 8130 8131 return __netdev_upper_dev_link(dev, upper_dev, true, 8132 upper_priv, upper_info, &priv, extack); 8133 } 8134 EXPORT_SYMBOL(netdev_master_upper_dev_link); 8135 8136 static void __netdev_upper_dev_unlink(struct net_device *dev, 8137 struct net_device *upper_dev, 8138 struct netdev_nested_priv *priv) 8139 { 8140 struct netdev_notifier_changeupper_info changeupper_info = { 8141 .info = { 8142 .dev = dev, 8143 }, 8144 .upper_dev = upper_dev, 8145 .linking = false, 8146 }; 8147 8148 ASSERT_RTNL(); 8149 8150 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; 8151 8152 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 8153 &changeupper_info.info); 8154 8155 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 8156 8157 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 8158 &changeupper_info.info); 8159 8160 __netdev_update_upper_level(dev, NULL); 8161 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 8162 8163 __netdev_update_lower_level(upper_dev, priv); 8164 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 8165 priv); 8166 } 8167 8168 /** 8169 * netdev_upper_dev_unlink - Removes a link to upper device 8170 * @dev: device 8171 * @upper_dev: new upper device 8172 * 8173 * Removes a link to device which is upper to this one. The caller must hold 8174 * the RTNL lock. 8175 */ 8176 void netdev_upper_dev_unlink(struct net_device *dev, 8177 struct net_device *upper_dev) 8178 { 8179 struct netdev_nested_priv priv = { 8180 .flags = NESTED_SYNC_TODO, 8181 .data = NULL, 8182 }; 8183 8184 __netdev_upper_dev_unlink(dev, upper_dev, &priv); 8185 } 8186 EXPORT_SYMBOL(netdev_upper_dev_unlink); 8187 8188 static void __netdev_adjacent_dev_set(struct net_device *upper_dev, 8189 struct net_device *lower_dev, 8190 bool val) 8191 { 8192 struct netdev_adjacent *adj; 8193 8194 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower); 8195 if (adj) 8196 adj->ignore = val; 8197 8198 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper); 8199 if (adj) 8200 adj->ignore = val; 8201 } 8202 8203 static void netdev_adjacent_dev_disable(struct net_device *upper_dev, 8204 struct net_device *lower_dev) 8205 { 8206 __netdev_adjacent_dev_set(upper_dev, lower_dev, true); 8207 } 8208 8209 static void netdev_adjacent_dev_enable(struct net_device *upper_dev, 8210 struct net_device *lower_dev) 8211 { 8212 __netdev_adjacent_dev_set(upper_dev, lower_dev, false); 8213 } 8214 8215 int netdev_adjacent_change_prepare(struct net_device *old_dev, 8216 struct net_device *new_dev, 8217 struct net_device *dev, 8218 struct netlink_ext_ack *extack) 8219 { 8220 struct netdev_nested_priv priv = { 8221 .flags = 0, 8222 .data = NULL, 8223 }; 8224 int err; 8225 8226 if (!new_dev) 8227 return 0; 8228 8229 if (old_dev && new_dev != old_dev) 8230 netdev_adjacent_dev_disable(dev, old_dev); 8231 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv, 8232 extack); 8233 if (err) { 8234 if (old_dev && new_dev != old_dev) 8235 netdev_adjacent_dev_enable(dev, old_dev); 8236 return err; 8237 } 8238 8239 return 0; 8240 } 8241 EXPORT_SYMBOL(netdev_adjacent_change_prepare); 8242 8243 void netdev_adjacent_change_commit(struct net_device *old_dev, 8244 struct net_device *new_dev, 8245 struct net_device *dev) 8246 { 8247 struct netdev_nested_priv priv = { 8248 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 8249 .data = NULL, 8250 }; 8251 8252 if (!new_dev || !old_dev) 8253 return; 8254 8255 if (new_dev == old_dev) 8256 return; 8257 8258 netdev_adjacent_dev_enable(dev, old_dev); 8259 __netdev_upper_dev_unlink(old_dev, dev, &priv); 8260 } 8261 EXPORT_SYMBOL(netdev_adjacent_change_commit); 8262 8263 void netdev_adjacent_change_abort(struct net_device *old_dev, 8264 struct net_device *new_dev, 8265 struct net_device *dev) 8266 { 8267 struct netdev_nested_priv priv = { 8268 .flags = 0, 8269 .data = NULL, 8270 }; 8271 8272 if (!new_dev) 8273 return; 8274 8275 if (old_dev && new_dev != old_dev) 8276 netdev_adjacent_dev_enable(dev, old_dev); 8277 8278 __netdev_upper_dev_unlink(new_dev, dev, &priv); 8279 } 8280 EXPORT_SYMBOL(netdev_adjacent_change_abort); 8281 8282 /** 8283 * netdev_bonding_info_change - Dispatch event about slave change 8284 * @dev: device 8285 * @bonding_info: info to dispatch 8286 * 8287 * Send NETDEV_BONDING_INFO to netdev notifiers with info. 8288 * The caller must hold the RTNL lock. 8289 */ 8290 void netdev_bonding_info_change(struct net_device *dev, 8291 struct netdev_bonding_info *bonding_info) 8292 { 8293 struct netdev_notifier_bonding_info info = { 8294 .info.dev = dev, 8295 }; 8296 8297 memcpy(&info.bonding_info, bonding_info, 8298 sizeof(struct netdev_bonding_info)); 8299 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, 8300 &info.info); 8301 } 8302 EXPORT_SYMBOL(netdev_bonding_info_change); 8303 8304 /** 8305 * netdev_get_xmit_slave - Get the xmit slave of master device 8306 * @dev: device 8307 * @skb: The packet 8308 * @all_slaves: assume all the slaves are active 8309 * 8310 * The reference counters are not incremented so the caller must be 8311 * careful with locks. The caller must hold RCU lock. 8312 * %NULL is returned if no slave is found. 8313 */ 8314 8315 struct net_device *netdev_get_xmit_slave(struct net_device *dev, 8316 struct sk_buff *skb, 8317 bool all_slaves) 8318 { 8319 const struct net_device_ops *ops = dev->netdev_ops; 8320 8321 if (!ops->ndo_get_xmit_slave) 8322 return NULL; 8323 return ops->ndo_get_xmit_slave(dev, skb, all_slaves); 8324 } 8325 EXPORT_SYMBOL(netdev_get_xmit_slave); 8326 8327 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev, 8328 struct sock *sk) 8329 { 8330 const struct net_device_ops *ops = dev->netdev_ops; 8331 8332 if (!ops->ndo_sk_get_lower_dev) 8333 return NULL; 8334 return ops->ndo_sk_get_lower_dev(dev, sk); 8335 } 8336 8337 /** 8338 * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket 8339 * @dev: device 8340 * @sk: the socket 8341 * 8342 * %NULL is returned if no lower device is found. 8343 */ 8344 8345 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev, 8346 struct sock *sk) 8347 { 8348 struct net_device *lower; 8349 8350 lower = netdev_sk_get_lower_dev(dev, sk); 8351 while (lower) { 8352 dev = lower; 8353 lower = netdev_sk_get_lower_dev(dev, sk); 8354 } 8355 8356 return dev; 8357 } 8358 EXPORT_SYMBOL(netdev_sk_get_lowest_dev); 8359 8360 static void netdev_adjacent_add_links(struct net_device *dev) 8361 { 8362 struct netdev_adjacent *iter; 8363 8364 struct net *net = dev_net(dev); 8365 8366 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8367 if (!net_eq(net, dev_net(iter->dev))) 8368 continue; 8369 netdev_adjacent_sysfs_add(iter->dev, dev, 8370 &iter->dev->adj_list.lower); 8371 netdev_adjacent_sysfs_add(dev, iter->dev, 8372 &dev->adj_list.upper); 8373 } 8374 8375 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8376 if (!net_eq(net, dev_net(iter->dev))) 8377 continue; 8378 netdev_adjacent_sysfs_add(iter->dev, dev, 8379 &iter->dev->adj_list.upper); 8380 netdev_adjacent_sysfs_add(dev, iter->dev, 8381 &dev->adj_list.lower); 8382 } 8383 } 8384 8385 static void netdev_adjacent_del_links(struct net_device *dev) 8386 { 8387 struct netdev_adjacent *iter; 8388 8389 struct net *net = dev_net(dev); 8390 8391 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8392 if (!net_eq(net, dev_net(iter->dev))) 8393 continue; 8394 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8395 &iter->dev->adj_list.lower); 8396 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8397 &dev->adj_list.upper); 8398 } 8399 8400 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8401 if (!net_eq(net, dev_net(iter->dev))) 8402 continue; 8403 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8404 &iter->dev->adj_list.upper); 8405 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8406 &dev->adj_list.lower); 8407 } 8408 } 8409 8410 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) 8411 { 8412 struct netdev_adjacent *iter; 8413 8414 struct net *net = dev_net(dev); 8415 8416 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8417 if (!net_eq(net, dev_net(iter->dev))) 8418 continue; 8419 netdev_adjacent_sysfs_del(iter->dev, oldname, 8420 &iter->dev->adj_list.lower); 8421 netdev_adjacent_sysfs_add(iter->dev, dev, 8422 &iter->dev->adj_list.lower); 8423 } 8424 8425 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8426 if (!net_eq(net, dev_net(iter->dev))) 8427 continue; 8428 netdev_adjacent_sysfs_del(iter->dev, oldname, 8429 &iter->dev->adj_list.upper); 8430 netdev_adjacent_sysfs_add(iter->dev, dev, 8431 &iter->dev->adj_list.upper); 8432 } 8433 } 8434 8435 void *netdev_lower_dev_get_private(struct net_device *dev, 8436 struct net_device *lower_dev) 8437 { 8438 struct netdev_adjacent *lower; 8439 8440 if (!lower_dev) 8441 return NULL; 8442 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); 8443 if (!lower) 8444 return NULL; 8445 8446 return lower->private; 8447 } 8448 EXPORT_SYMBOL(netdev_lower_dev_get_private); 8449 8450 8451 /** 8452 * netdev_lower_state_changed - Dispatch event about lower device state change 8453 * @lower_dev: device 8454 * @lower_state_info: state to dispatch 8455 * 8456 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. 8457 * The caller must hold the RTNL lock. 8458 */ 8459 void netdev_lower_state_changed(struct net_device *lower_dev, 8460 void *lower_state_info) 8461 { 8462 struct netdev_notifier_changelowerstate_info changelowerstate_info = { 8463 .info.dev = lower_dev, 8464 }; 8465 8466 ASSERT_RTNL(); 8467 changelowerstate_info.lower_state_info = lower_state_info; 8468 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, 8469 &changelowerstate_info.info); 8470 } 8471 EXPORT_SYMBOL(netdev_lower_state_changed); 8472 8473 static void dev_change_rx_flags(struct net_device *dev, int flags) 8474 { 8475 const struct net_device_ops *ops = dev->netdev_ops; 8476 8477 if (ops->ndo_change_rx_flags) 8478 ops->ndo_change_rx_flags(dev, flags); 8479 } 8480 8481 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) 8482 { 8483 unsigned int old_flags = dev->flags; 8484 kuid_t uid; 8485 kgid_t gid; 8486 8487 ASSERT_RTNL(); 8488 8489 dev->flags |= IFF_PROMISC; 8490 dev->promiscuity += inc; 8491 if (dev->promiscuity == 0) { 8492 /* 8493 * Avoid overflow. 8494 * If inc causes overflow, untouch promisc and return error. 8495 */ 8496 if (inc < 0) 8497 dev->flags &= ~IFF_PROMISC; 8498 else { 8499 dev->promiscuity -= inc; 8500 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n", 8501 dev->name); 8502 return -EOVERFLOW; 8503 } 8504 } 8505 if (dev->flags != old_flags) { 8506 pr_info("device %s %s promiscuous mode\n", 8507 dev->name, 8508 dev->flags & IFF_PROMISC ? "entered" : "left"); 8509 if (audit_enabled) { 8510 current_uid_gid(&uid, &gid); 8511 audit_log(audit_context(), GFP_ATOMIC, 8512 AUDIT_ANOM_PROMISCUOUS, 8513 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", 8514 dev->name, (dev->flags & IFF_PROMISC), 8515 (old_flags & IFF_PROMISC), 8516 from_kuid(&init_user_ns, audit_get_loginuid(current)), 8517 from_kuid(&init_user_ns, uid), 8518 from_kgid(&init_user_ns, gid), 8519 audit_get_sessionid(current)); 8520 } 8521 8522 dev_change_rx_flags(dev, IFF_PROMISC); 8523 } 8524 if (notify) 8525 __dev_notify_flags(dev, old_flags, IFF_PROMISC); 8526 return 0; 8527 } 8528 8529 /** 8530 * dev_set_promiscuity - update promiscuity count on a device 8531 * @dev: device 8532 * @inc: modifier 8533 * 8534 * Add or remove promiscuity from a device. While the count in the device 8535 * remains above zero the interface remains promiscuous. Once it hits zero 8536 * the device reverts back to normal filtering operation. A negative inc 8537 * value is used to drop promiscuity on the device. 8538 * Return 0 if successful or a negative errno code on error. 8539 */ 8540 int dev_set_promiscuity(struct net_device *dev, int inc) 8541 { 8542 unsigned int old_flags = dev->flags; 8543 int err; 8544 8545 err = __dev_set_promiscuity(dev, inc, true); 8546 if (err < 0) 8547 return err; 8548 if (dev->flags != old_flags) 8549 dev_set_rx_mode(dev); 8550 return err; 8551 } 8552 EXPORT_SYMBOL(dev_set_promiscuity); 8553 8554 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) 8555 { 8556 unsigned int old_flags = dev->flags, old_gflags = dev->gflags; 8557 8558 ASSERT_RTNL(); 8559 8560 dev->flags |= IFF_ALLMULTI; 8561 dev->allmulti += inc; 8562 if (dev->allmulti == 0) { 8563 /* 8564 * Avoid overflow. 8565 * If inc causes overflow, untouch allmulti and return error. 8566 */ 8567 if (inc < 0) 8568 dev->flags &= ~IFF_ALLMULTI; 8569 else { 8570 dev->allmulti -= inc; 8571 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n", 8572 dev->name); 8573 return -EOVERFLOW; 8574 } 8575 } 8576 if (dev->flags ^ old_flags) { 8577 dev_change_rx_flags(dev, IFF_ALLMULTI); 8578 dev_set_rx_mode(dev); 8579 if (notify) 8580 __dev_notify_flags(dev, old_flags, 8581 dev->gflags ^ old_gflags); 8582 } 8583 return 0; 8584 } 8585 8586 /** 8587 * dev_set_allmulti - update allmulti count on a device 8588 * @dev: device 8589 * @inc: modifier 8590 * 8591 * Add or remove reception of all multicast frames to a device. While the 8592 * count in the device remains above zero the interface remains listening 8593 * to all interfaces. Once it hits zero the device reverts back to normal 8594 * filtering operation. A negative @inc value is used to drop the counter 8595 * when releasing a resource needing all multicasts. 8596 * Return 0 if successful or a negative errno code on error. 8597 */ 8598 8599 int dev_set_allmulti(struct net_device *dev, int inc) 8600 { 8601 return __dev_set_allmulti(dev, inc, true); 8602 } 8603 EXPORT_SYMBOL(dev_set_allmulti); 8604 8605 /* 8606 * Upload unicast and multicast address lists to device and 8607 * configure RX filtering. When the device doesn't support unicast 8608 * filtering it is put in promiscuous mode while unicast addresses 8609 * are present. 8610 */ 8611 void __dev_set_rx_mode(struct net_device *dev) 8612 { 8613 const struct net_device_ops *ops = dev->netdev_ops; 8614 8615 /* dev_open will call this function so the list will stay sane. */ 8616 if (!(dev->flags&IFF_UP)) 8617 return; 8618 8619 if (!netif_device_present(dev)) 8620 return; 8621 8622 if (!(dev->priv_flags & IFF_UNICAST_FLT)) { 8623 /* Unicast addresses changes may only happen under the rtnl, 8624 * therefore calling __dev_set_promiscuity here is safe. 8625 */ 8626 if (!netdev_uc_empty(dev) && !dev->uc_promisc) { 8627 __dev_set_promiscuity(dev, 1, false); 8628 dev->uc_promisc = true; 8629 } else if (netdev_uc_empty(dev) && dev->uc_promisc) { 8630 __dev_set_promiscuity(dev, -1, false); 8631 dev->uc_promisc = false; 8632 } 8633 } 8634 8635 if (ops->ndo_set_rx_mode) 8636 ops->ndo_set_rx_mode(dev); 8637 } 8638 8639 void dev_set_rx_mode(struct net_device *dev) 8640 { 8641 netif_addr_lock_bh(dev); 8642 __dev_set_rx_mode(dev); 8643 netif_addr_unlock_bh(dev); 8644 } 8645 8646 /** 8647 * dev_get_flags - get flags reported to userspace 8648 * @dev: device 8649 * 8650 * Get the combination of flag bits exported through APIs to userspace. 8651 */ 8652 unsigned int dev_get_flags(const struct net_device *dev) 8653 { 8654 unsigned int flags; 8655 8656 flags = (dev->flags & ~(IFF_PROMISC | 8657 IFF_ALLMULTI | 8658 IFF_RUNNING | 8659 IFF_LOWER_UP | 8660 IFF_DORMANT)) | 8661 (dev->gflags & (IFF_PROMISC | 8662 IFF_ALLMULTI)); 8663 8664 if (netif_running(dev)) { 8665 if (netif_oper_up(dev)) 8666 flags |= IFF_RUNNING; 8667 if (netif_carrier_ok(dev)) 8668 flags |= IFF_LOWER_UP; 8669 if (netif_dormant(dev)) 8670 flags |= IFF_DORMANT; 8671 } 8672 8673 return flags; 8674 } 8675 EXPORT_SYMBOL(dev_get_flags); 8676 8677 int __dev_change_flags(struct net_device *dev, unsigned int flags, 8678 struct netlink_ext_ack *extack) 8679 { 8680 unsigned int old_flags = dev->flags; 8681 int ret; 8682 8683 ASSERT_RTNL(); 8684 8685 /* 8686 * Set the flags on our device. 8687 */ 8688 8689 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | 8690 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | 8691 IFF_AUTOMEDIA)) | 8692 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | 8693 IFF_ALLMULTI)); 8694 8695 /* 8696 * Load in the correct multicast list now the flags have changed. 8697 */ 8698 8699 if ((old_flags ^ flags) & IFF_MULTICAST) 8700 dev_change_rx_flags(dev, IFF_MULTICAST); 8701 8702 dev_set_rx_mode(dev); 8703 8704 /* 8705 * Have we downed the interface. We handle IFF_UP ourselves 8706 * according to user attempts to set it, rather than blindly 8707 * setting it. 8708 */ 8709 8710 ret = 0; 8711 if ((old_flags ^ flags) & IFF_UP) { 8712 if (old_flags & IFF_UP) 8713 __dev_close(dev); 8714 else 8715 ret = __dev_open(dev, extack); 8716 } 8717 8718 if ((flags ^ dev->gflags) & IFF_PROMISC) { 8719 int inc = (flags & IFF_PROMISC) ? 1 : -1; 8720 unsigned int old_flags = dev->flags; 8721 8722 dev->gflags ^= IFF_PROMISC; 8723 8724 if (__dev_set_promiscuity(dev, inc, false) >= 0) 8725 if (dev->flags != old_flags) 8726 dev_set_rx_mode(dev); 8727 } 8728 8729 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI 8730 * is important. Some (broken) drivers set IFF_PROMISC, when 8731 * IFF_ALLMULTI is requested not asking us and not reporting. 8732 */ 8733 if ((flags ^ dev->gflags) & IFF_ALLMULTI) { 8734 int inc = (flags & IFF_ALLMULTI) ? 1 : -1; 8735 8736 dev->gflags ^= IFF_ALLMULTI; 8737 __dev_set_allmulti(dev, inc, false); 8738 } 8739 8740 return ret; 8741 } 8742 8743 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, 8744 unsigned int gchanges) 8745 { 8746 unsigned int changes = dev->flags ^ old_flags; 8747 8748 if (gchanges) 8749 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC); 8750 8751 if (changes & IFF_UP) { 8752 if (dev->flags & IFF_UP) 8753 call_netdevice_notifiers(NETDEV_UP, dev); 8754 else 8755 call_netdevice_notifiers(NETDEV_DOWN, dev); 8756 } 8757 8758 if (dev->flags & IFF_UP && 8759 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { 8760 struct netdev_notifier_change_info change_info = { 8761 .info = { 8762 .dev = dev, 8763 }, 8764 .flags_changed = changes, 8765 }; 8766 8767 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info); 8768 } 8769 } 8770 8771 /** 8772 * dev_change_flags - change device settings 8773 * @dev: device 8774 * @flags: device state flags 8775 * @extack: netlink extended ack 8776 * 8777 * Change settings on device based state flags. The flags are 8778 * in the userspace exported format. 8779 */ 8780 int dev_change_flags(struct net_device *dev, unsigned int flags, 8781 struct netlink_ext_ack *extack) 8782 { 8783 int ret; 8784 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; 8785 8786 ret = __dev_change_flags(dev, flags, extack); 8787 if (ret < 0) 8788 return ret; 8789 8790 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); 8791 __dev_notify_flags(dev, old_flags, changes); 8792 return ret; 8793 } 8794 EXPORT_SYMBOL(dev_change_flags); 8795 8796 int __dev_set_mtu(struct net_device *dev, int new_mtu) 8797 { 8798 const struct net_device_ops *ops = dev->netdev_ops; 8799 8800 if (ops->ndo_change_mtu) 8801 return ops->ndo_change_mtu(dev, new_mtu); 8802 8803 /* Pairs with all the lockless reads of dev->mtu in the stack */ 8804 WRITE_ONCE(dev->mtu, new_mtu); 8805 return 0; 8806 } 8807 EXPORT_SYMBOL(__dev_set_mtu); 8808 8809 int dev_validate_mtu(struct net_device *dev, int new_mtu, 8810 struct netlink_ext_ack *extack) 8811 { 8812 /* MTU must be positive, and in range */ 8813 if (new_mtu < 0 || new_mtu < dev->min_mtu) { 8814 NL_SET_ERR_MSG(extack, "mtu less than device minimum"); 8815 return -EINVAL; 8816 } 8817 8818 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { 8819 NL_SET_ERR_MSG(extack, "mtu greater than device maximum"); 8820 return -EINVAL; 8821 } 8822 return 0; 8823 } 8824 8825 /** 8826 * dev_set_mtu_ext - Change maximum transfer unit 8827 * @dev: device 8828 * @new_mtu: new transfer unit 8829 * @extack: netlink extended ack 8830 * 8831 * Change the maximum transfer size of the network device. 8832 */ 8833 int dev_set_mtu_ext(struct net_device *dev, int new_mtu, 8834 struct netlink_ext_ack *extack) 8835 { 8836 int err, orig_mtu; 8837 8838 if (new_mtu == dev->mtu) 8839 return 0; 8840 8841 err = dev_validate_mtu(dev, new_mtu, extack); 8842 if (err) 8843 return err; 8844 8845 if (!netif_device_present(dev)) 8846 return -ENODEV; 8847 8848 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); 8849 err = notifier_to_errno(err); 8850 if (err) 8851 return err; 8852 8853 orig_mtu = dev->mtu; 8854 err = __dev_set_mtu(dev, new_mtu); 8855 8856 if (!err) { 8857 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8858 orig_mtu); 8859 err = notifier_to_errno(err); 8860 if (err) { 8861 /* setting mtu back and notifying everyone again, 8862 * so that they have a chance to revert changes. 8863 */ 8864 __dev_set_mtu(dev, orig_mtu); 8865 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8866 new_mtu); 8867 } 8868 } 8869 return err; 8870 } 8871 8872 int dev_set_mtu(struct net_device *dev, int new_mtu) 8873 { 8874 struct netlink_ext_ack extack; 8875 int err; 8876 8877 memset(&extack, 0, sizeof(extack)); 8878 err = dev_set_mtu_ext(dev, new_mtu, &extack); 8879 if (err && extack._msg) 8880 net_err_ratelimited("%s: %s\n", dev->name, extack._msg); 8881 return err; 8882 } 8883 EXPORT_SYMBOL(dev_set_mtu); 8884 8885 /** 8886 * dev_change_tx_queue_len - Change TX queue length of a netdevice 8887 * @dev: device 8888 * @new_len: new tx queue length 8889 */ 8890 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len) 8891 { 8892 unsigned int orig_len = dev->tx_queue_len; 8893 int res; 8894 8895 if (new_len != (unsigned int)new_len) 8896 return -ERANGE; 8897 8898 if (new_len != orig_len) { 8899 dev->tx_queue_len = new_len; 8900 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev); 8901 res = notifier_to_errno(res); 8902 if (res) 8903 goto err_rollback; 8904 res = dev_qdisc_change_tx_queue_len(dev); 8905 if (res) 8906 goto err_rollback; 8907 } 8908 8909 return 0; 8910 8911 err_rollback: 8912 netdev_err(dev, "refused to change device tx_queue_len\n"); 8913 dev->tx_queue_len = orig_len; 8914 return res; 8915 } 8916 8917 /** 8918 * dev_set_group - Change group this device belongs to 8919 * @dev: device 8920 * @new_group: group this device should belong to 8921 */ 8922 void dev_set_group(struct net_device *dev, int new_group) 8923 { 8924 dev->group = new_group; 8925 } 8926 EXPORT_SYMBOL(dev_set_group); 8927 8928 /** 8929 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR. 8930 * @dev: device 8931 * @addr: new address 8932 * @extack: netlink extended ack 8933 */ 8934 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr, 8935 struct netlink_ext_ack *extack) 8936 { 8937 struct netdev_notifier_pre_changeaddr_info info = { 8938 .info.dev = dev, 8939 .info.extack = extack, 8940 .dev_addr = addr, 8941 }; 8942 int rc; 8943 8944 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info); 8945 return notifier_to_errno(rc); 8946 } 8947 EXPORT_SYMBOL(dev_pre_changeaddr_notify); 8948 8949 /** 8950 * dev_set_mac_address - Change Media Access Control Address 8951 * @dev: device 8952 * @sa: new address 8953 * @extack: netlink extended ack 8954 * 8955 * Change the hardware (MAC) address of the device 8956 */ 8957 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa, 8958 struct netlink_ext_ack *extack) 8959 { 8960 const struct net_device_ops *ops = dev->netdev_ops; 8961 int err; 8962 8963 if (!ops->ndo_set_mac_address) 8964 return -EOPNOTSUPP; 8965 if (sa->sa_family != dev->type) 8966 return -EINVAL; 8967 if (!netif_device_present(dev)) 8968 return -ENODEV; 8969 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack); 8970 if (err) 8971 return err; 8972 err = ops->ndo_set_mac_address(dev, sa); 8973 if (err) 8974 return err; 8975 dev->addr_assign_type = NET_ADDR_SET; 8976 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); 8977 add_device_randomness(dev->dev_addr, dev->addr_len); 8978 return 0; 8979 } 8980 EXPORT_SYMBOL(dev_set_mac_address); 8981 8982 static DECLARE_RWSEM(dev_addr_sem); 8983 8984 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa, 8985 struct netlink_ext_ack *extack) 8986 { 8987 int ret; 8988 8989 down_write(&dev_addr_sem); 8990 ret = dev_set_mac_address(dev, sa, extack); 8991 up_write(&dev_addr_sem); 8992 return ret; 8993 } 8994 EXPORT_SYMBOL(dev_set_mac_address_user); 8995 8996 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name) 8997 { 8998 size_t size = sizeof(sa->sa_data); 8999 struct net_device *dev; 9000 int ret = 0; 9001 9002 down_read(&dev_addr_sem); 9003 rcu_read_lock(); 9004 9005 dev = dev_get_by_name_rcu(net, dev_name); 9006 if (!dev) { 9007 ret = -ENODEV; 9008 goto unlock; 9009 } 9010 if (!dev->addr_len) 9011 memset(sa->sa_data, 0, size); 9012 else 9013 memcpy(sa->sa_data, dev->dev_addr, 9014 min_t(size_t, size, dev->addr_len)); 9015 sa->sa_family = dev->type; 9016 9017 unlock: 9018 rcu_read_unlock(); 9019 up_read(&dev_addr_sem); 9020 return ret; 9021 } 9022 EXPORT_SYMBOL(dev_get_mac_address); 9023 9024 /** 9025 * dev_change_carrier - Change device carrier 9026 * @dev: device 9027 * @new_carrier: new value 9028 * 9029 * Change device carrier 9030 */ 9031 int dev_change_carrier(struct net_device *dev, bool new_carrier) 9032 { 9033 const struct net_device_ops *ops = dev->netdev_ops; 9034 9035 if (!ops->ndo_change_carrier) 9036 return -EOPNOTSUPP; 9037 if (!netif_device_present(dev)) 9038 return -ENODEV; 9039 return ops->ndo_change_carrier(dev, new_carrier); 9040 } 9041 EXPORT_SYMBOL(dev_change_carrier); 9042 9043 /** 9044 * dev_get_phys_port_id - Get device physical port ID 9045 * @dev: device 9046 * @ppid: port ID 9047 * 9048 * Get device physical port ID 9049 */ 9050 int dev_get_phys_port_id(struct net_device *dev, 9051 struct netdev_phys_item_id *ppid) 9052 { 9053 const struct net_device_ops *ops = dev->netdev_ops; 9054 9055 if (!ops->ndo_get_phys_port_id) 9056 return -EOPNOTSUPP; 9057 return ops->ndo_get_phys_port_id(dev, ppid); 9058 } 9059 EXPORT_SYMBOL(dev_get_phys_port_id); 9060 9061 /** 9062 * dev_get_phys_port_name - Get device physical port name 9063 * @dev: device 9064 * @name: port name 9065 * @len: limit of bytes to copy to name 9066 * 9067 * Get device physical port name 9068 */ 9069 int dev_get_phys_port_name(struct net_device *dev, 9070 char *name, size_t len) 9071 { 9072 const struct net_device_ops *ops = dev->netdev_ops; 9073 int err; 9074 9075 if (ops->ndo_get_phys_port_name) { 9076 err = ops->ndo_get_phys_port_name(dev, name, len); 9077 if (err != -EOPNOTSUPP) 9078 return err; 9079 } 9080 return devlink_compat_phys_port_name_get(dev, name, len); 9081 } 9082 EXPORT_SYMBOL(dev_get_phys_port_name); 9083 9084 /** 9085 * dev_get_port_parent_id - Get the device's port parent identifier 9086 * @dev: network device 9087 * @ppid: pointer to a storage for the port's parent identifier 9088 * @recurse: allow/disallow recursion to lower devices 9089 * 9090 * Get the devices's port parent identifier 9091 */ 9092 int dev_get_port_parent_id(struct net_device *dev, 9093 struct netdev_phys_item_id *ppid, 9094 bool recurse) 9095 { 9096 const struct net_device_ops *ops = dev->netdev_ops; 9097 struct netdev_phys_item_id first = { }; 9098 struct net_device *lower_dev; 9099 struct list_head *iter; 9100 int err; 9101 9102 if (ops->ndo_get_port_parent_id) { 9103 err = ops->ndo_get_port_parent_id(dev, ppid); 9104 if (err != -EOPNOTSUPP) 9105 return err; 9106 } 9107 9108 err = devlink_compat_switch_id_get(dev, ppid); 9109 if (!err || err != -EOPNOTSUPP) 9110 return err; 9111 9112 if (!recurse) 9113 return -EOPNOTSUPP; 9114 9115 netdev_for_each_lower_dev(dev, lower_dev, iter) { 9116 err = dev_get_port_parent_id(lower_dev, ppid, recurse); 9117 if (err) 9118 break; 9119 if (!first.id_len) 9120 first = *ppid; 9121 else if (memcmp(&first, ppid, sizeof(*ppid))) 9122 return -EOPNOTSUPP; 9123 } 9124 9125 return err; 9126 } 9127 EXPORT_SYMBOL(dev_get_port_parent_id); 9128 9129 /** 9130 * netdev_port_same_parent_id - Indicate if two network devices have 9131 * the same port parent identifier 9132 * @a: first network device 9133 * @b: second network device 9134 */ 9135 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b) 9136 { 9137 struct netdev_phys_item_id a_id = { }; 9138 struct netdev_phys_item_id b_id = { }; 9139 9140 if (dev_get_port_parent_id(a, &a_id, true) || 9141 dev_get_port_parent_id(b, &b_id, true)) 9142 return false; 9143 9144 return netdev_phys_item_id_same(&a_id, &b_id); 9145 } 9146 EXPORT_SYMBOL(netdev_port_same_parent_id); 9147 9148 /** 9149 * dev_change_proto_down - update protocol port state information 9150 * @dev: device 9151 * @proto_down: new value 9152 * 9153 * This info can be used by switch drivers to set the phys state of the 9154 * port. 9155 */ 9156 int dev_change_proto_down(struct net_device *dev, bool proto_down) 9157 { 9158 const struct net_device_ops *ops = dev->netdev_ops; 9159 9160 if (!ops->ndo_change_proto_down) 9161 return -EOPNOTSUPP; 9162 if (!netif_device_present(dev)) 9163 return -ENODEV; 9164 return ops->ndo_change_proto_down(dev, proto_down); 9165 } 9166 EXPORT_SYMBOL(dev_change_proto_down); 9167 9168 /** 9169 * dev_change_proto_down_generic - generic implementation for 9170 * ndo_change_proto_down that sets carrier according to 9171 * proto_down. 9172 * 9173 * @dev: device 9174 * @proto_down: new value 9175 */ 9176 int dev_change_proto_down_generic(struct net_device *dev, bool proto_down) 9177 { 9178 if (proto_down) 9179 netif_carrier_off(dev); 9180 else 9181 netif_carrier_on(dev); 9182 dev->proto_down = proto_down; 9183 return 0; 9184 } 9185 EXPORT_SYMBOL(dev_change_proto_down_generic); 9186 9187 /** 9188 * dev_change_proto_down_reason - proto down reason 9189 * 9190 * @dev: device 9191 * @mask: proto down mask 9192 * @value: proto down value 9193 */ 9194 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask, 9195 u32 value) 9196 { 9197 int b; 9198 9199 if (!mask) { 9200 dev->proto_down_reason = value; 9201 } else { 9202 for_each_set_bit(b, &mask, 32) { 9203 if (value & (1 << b)) 9204 dev->proto_down_reason |= BIT(b); 9205 else 9206 dev->proto_down_reason &= ~BIT(b); 9207 } 9208 } 9209 } 9210 EXPORT_SYMBOL(dev_change_proto_down_reason); 9211 9212 struct bpf_xdp_link { 9213 struct bpf_link link; 9214 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */ 9215 int flags; 9216 }; 9217 9218 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags) 9219 { 9220 if (flags & XDP_FLAGS_HW_MODE) 9221 return XDP_MODE_HW; 9222 if (flags & XDP_FLAGS_DRV_MODE) 9223 return XDP_MODE_DRV; 9224 if (flags & XDP_FLAGS_SKB_MODE) 9225 return XDP_MODE_SKB; 9226 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB; 9227 } 9228 9229 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode) 9230 { 9231 switch (mode) { 9232 case XDP_MODE_SKB: 9233 return generic_xdp_install; 9234 case XDP_MODE_DRV: 9235 case XDP_MODE_HW: 9236 return dev->netdev_ops->ndo_bpf; 9237 default: 9238 return NULL; 9239 } 9240 } 9241 9242 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev, 9243 enum bpf_xdp_mode mode) 9244 { 9245 return dev->xdp_state[mode].link; 9246 } 9247 9248 static struct bpf_prog *dev_xdp_prog(struct net_device *dev, 9249 enum bpf_xdp_mode mode) 9250 { 9251 struct bpf_xdp_link *link = dev_xdp_link(dev, mode); 9252 9253 if (link) 9254 return link->link.prog; 9255 return dev->xdp_state[mode].prog; 9256 } 9257 9258 static u8 dev_xdp_prog_count(struct net_device *dev) 9259 { 9260 u8 count = 0; 9261 int i; 9262 9263 for (i = 0; i < __MAX_XDP_MODE; i++) 9264 if (dev->xdp_state[i].prog || dev->xdp_state[i].link) 9265 count++; 9266 return count; 9267 } 9268 9269 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode) 9270 { 9271 struct bpf_prog *prog = dev_xdp_prog(dev, mode); 9272 9273 return prog ? prog->aux->id : 0; 9274 } 9275 9276 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode, 9277 struct bpf_xdp_link *link) 9278 { 9279 dev->xdp_state[mode].link = link; 9280 dev->xdp_state[mode].prog = NULL; 9281 } 9282 9283 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode, 9284 struct bpf_prog *prog) 9285 { 9286 dev->xdp_state[mode].link = NULL; 9287 dev->xdp_state[mode].prog = prog; 9288 } 9289 9290 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode, 9291 bpf_op_t bpf_op, struct netlink_ext_ack *extack, 9292 u32 flags, struct bpf_prog *prog) 9293 { 9294 struct netdev_bpf xdp; 9295 int err; 9296 9297 memset(&xdp, 0, sizeof(xdp)); 9298 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG; 9299 xdp.extack = extack; 9300 xdp.flags = flags; 9301 xdp.prog = prog; 9302 9303 /* Drivers assume refcnt is already incremented (i.e, prog pointer is 9304 * "moved" into driver), so they don't increment it on their own, but 9305 * they do decrement refcnt when program is detached or replaced. 9306 * Given net_device also owns link/prog, we need to bump refcnt here 9307 * to prevent drivers from underflowing it. 9308 */ 9309 if (prog) 9310 bpf_prog_inc(prog); 9311 err = bpf_op(dev, &xdp); 9312 if (err) { 9313 if (prog) 9314 bpf_prog_put(prog); 9315 return err; 9316 } 9317 9318 if (mode != XDP_MODE_HW) 9319 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog); 9320 9321 return 0; 9322 } 9323 9324 static void dev_xdp_uninstall(struct net_device *dev) 9325 { 9326 struct bpf_xdp_link *link; 9327 struct bpf_prog *prog; 9328 enum bpf_xdp_mode mode; 9329 bpf_op_t bpf_op; 9330 9331 ASSERT_RTNL(); 9332 9333 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) { 9334 prog = dev_xdp_prog(dev, mode); 9335 if (!prog) 9336 continue; 9337 9338 bpf_op = dev_xdp_bpf_op(dev, mode); 9339 if (!bpf_op) 9340 continue; 9341 9342 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9343 9344 /* auto-detach link from net device */ 9345 link = dev_xdp_link(dev, mode); 9346 if (link) 9347 link->dev = NULL; 9348 else 9349 bpf_prog_put(prog); 9350 9351 dev_xdp_set_link(dev, mode, NULL); 9352 } 9353 } 9354 9355 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack, 9356 struct bpf_xdp_link *link, struct bpf_prog *new_prog, 9357 struct bpf_prog *old_prog, u32 flags) 9358 { 9359 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES); 9360 struct bpf_prog *cur_prog; 9361 enum bpf_xdp_mode mode; 9362 bpf_op_t bpf_op; 9363 int err; 9364 9365 ASSERT_RTNL(); 9366 9367 /* either link or prog attachment, never both */ 9368 if (link && (new_prog || old_prog)) 9369 return -EINVAL; 9370 /* link supports only XDP mode flags */ 9371 if (link && (flags & ~XDP_FLAGS_MODES)) { 9372 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment"); 9373 return -EINVAL; 9374 } 9375 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */ 9376 if (num_modes > 1) { 9377 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set"); 9378 return -EINVAL; 9379 } 9380 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */ 9381 if (!num_modes && dev_xdp_prog_count(dev) > 1) { 9382 NL_SET_ERR_MSG(extack, 9383 "More than one program loaded, unset mode is ambiguous"); 9384 return -EINVAL; 9385 } 9386 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */ 9387 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) { 9388 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified"); 9389 return -EINVAL; 9390 } 9391 9392 mode = dev_xdp_mode(dev, flags); 9393 /* can't replace attached link */ 9394 if (dev_xdp_link(dev, mode)) { 9395 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link"); 9396 return -EBUSY; 9397 } 9398 9399 cur_prog = dev_xdp_prog(dev, mode); 9400 /* can't replace attached prog with link */ 9401 if (link && cur_prog) { 9402 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link"); 9403 return -EBUSY; 9404 } 9405 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) { 9406 NL_SET_ERR_MSG(extack, "Active program does not match expected"); 9407 return -EEXIST; 9408 } 9409 9410 /* put effective new program into new_prog */ 9411 if (link) 9412 new_prog = link->link.prog; 9413 9414 if (new_prog) { 9415 bool offload = mode == XDP_MODE_HW; 9416 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB 9417 ? XDP_MODE_DRV : XDP_MODE_SKB; 9418 9419 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) { 9420 NL_SET_ERR_MSG(extack, "XDP program already attached"); 9421 return -EBUSY; 9422 } 9423 if (!offload && dev_xdp_prog(dev, other_mode)) { 9424 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time"); 9425 return -EEXIST; 9426 } 9427 if (!offload && bpf_prog_is_dev_bound(new_prog->aux)) { 9428 NL_SET_ERR_MSG(extack, "Using device-bound program without HW_MODE flag is not supported"); 9429 return -EINVAL; 9430 } 9431 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) { 9432 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device"); 9433 return -EINVAL; 9434 } 9435 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) { 9436 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device"); 9437 return -EINVAL; 9438 } 9439 } 9440 9441 /* don't call drivers if the effective program didn't change */ 9442 if (new_prog != cur_prog) { 9443 bpf_op = dev_xdp_bpf_op(dev, mode); 9444 if (!bpf_op) { 9445 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode"); 9446 return -EOPNOTSUPP; 9447 } 9448 9449 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog); 9450 if (err) 9451 return err; 9452 } 9453 9454 if (link) 9455 dev_xdp_set_link(dev, mode, link); 9456 else 9457 dev_xdp_set_prog(dev, mode, new_prog); 9458 if (cur_prog) 9459 bpf_prog_put(cur_prog); 9460 9461 return 0; 9462 } 9463 9464 static int dev_xdp_attach_link(struct net_device *dev, 9465 struct netlink_ext_ack *extack, 9466 struct bpf_xdp_link *link) 9467 { 9468 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags); 9469 } 9470 9471 static int dev_xdp_detach_link(struct net_device *dev, 9472 struct netlink_ext_ack *extack, 9473 struct bpf_xdp_link *link) 9474 { 9475 enum bpf_xdp_mode mode; 9476 bpf_op_t bpf_op; 9477 9478 ASSERT_RTNL(); 9479 9480 mode = dev_xdp_mode(dev, link->flags); 9481 if (dev_xdp_link(dev, mode) != link) 9482 return -EINVAL; 9483 9484 bpf_op = dev_xdp_bpf_op(dev, mode); 9485 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9486 dev_xdp_set_link(dev, mode, NULL); 9487 return 0; 9488 } 9489 9490 static void bpf_xdp_link_release(struct bpf_link *link) 9491 { 9492 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9493 9494 rtnl_lock(); 9495 9496 /* if racing with net_device's tear down, xdp_link->dev might be 9497 * already NULL, in which case link was already auto-detached 9498 */ 9499 if (xdp_link->dev) { 9500 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link)); 9501 xdp_link->dev = NULL; 9502 } 9503 9504 rtnl_unlock(); 9505 } 9506 9507 static int bpf_xdp_link_detach(struct bpf_link *link) 9508 { 9509 bpf_xdp_link_release(link); 9510 return 0; 9511 } 9512 9513 static void bpf_xdp_link_dealloc(struct bpf_link *link) 9514 { 9515 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9516 9517 kfree(xdp_link); 9518 } 9519 9520 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link, 9521 struct seq_file *seq) 9522 { 9523 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9524 u32 ifindex = 0; 9525 9526 rtnl_lock(); 9527 if (xdp_link->dev) 9528 ifindex = xdp_link->dev->ifindex; 9529 rtnl_unlock(); 9530 9531 seq_printf(seq, "ifindex:\t%u\n", ifindex); 9532 } 9533 9534 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link, 9535 struct bpf_link_info *info) 9536 { 9537 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9538 u32 ifindex = 0; 9539 9540 rtnl_lock(); 9541 if (xdp_link->dev) 9542 ifindex = xdp_link->dev->ifindex; 9543 rtnl_unlock(); 9544 9545 info->xdp.ifindex = ifindex; 9546 return 0; 9547 } 9548 9549 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog, 9550 struct bpf_prog *old_prog) 9551 { 9552 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9553 enum bpf_xdp_mode mode; 9554 bpf_op_t bpf_op; 9555 int err = 0; 9556 9557 rtnl_lock(); 9558 9559 /* link might have been auto-released already, so fail */ 9560 if (!xdp_link->dev) { 9561 err = -ENOLINK; 9562 goto out_unlock; 9563 } 9564 9565 if (old_prog && link->prog != old_prog) { 9566 err = -EPERM; 9567 goto out_unlock; 9568 } 9569 old_prog = link->prog; 9570 if (old_prog == new_prog) { 9571 /* no-op, don't disturb drivers */ 9572 bpf_prog_put(new_prog); 9573 goto out_unlock; 9574 } 9575 9576 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags); 9577 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode); 9578 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL, 9579 xdp_link->flags, new_prog); 9580 if (err) 9581 goto out_unlock; 9582 9583 old_prog = xchg(&link->prog, new_prog); 9584 bpf_prog_put(old_prog); 9585 9586 out_unlock: 9587 rtnl_unlock(); 9588 return err; 9589 } 9590 9591 static const struct bpf_link_ops bpf_xdp_link_lops = { 9592 .release = bpf_xdp_link_release, 9593 .dealloc = bpf_xdp_link_dealloc, 9594 .detach = bpf_xdp_link_detach, 9595 .show_fdinfo = bpf_xdp_link_show_fdinfo, 9596 .fill_link_info = bpf_xdp_link_fill_link_info, 9597 .update_prog = bpf_xdp_link_update, 9598 }; 9599 9600 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog) 9601 { 9602 struct net *net = current->nsproxy->net_ns; 9603 struct bpf_link_primer link_primer; 9604 struct bpf_xdp_link *link; 9605 struct net_device *dev; 9606 int err, fd; 9607 9608 rtnl_lock(); 9609 dev = dev_get_by_index(net, attr->link_create.target_ifindex); 9610 if (!dev) { 9611 rtnl_unlock(); 9612 return -EINVAL; 9613 } 9614 9615 link = kzalloc(sizeof(*link), GFP_USER); 9616 if (!link) { 9617 err = -ENOMEM; 9618 goto unlock; 9619 } 9620 9621 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog); 9622 link->dev = dev; 9623 link->flags = attr->link_create.flags; 9624 9625 err = bpf_link_prime(&link->link, &link_primer); 9626 if (err) { 9627 kfree(link); 9628 goto unlock; 9629 } 9630 9631 err = dev_xdp_attach_link(dev, NULL, link); 9632 rtnl_unlock(); 9633 9634 if (err) { 9635 link->dev = NULL; 9636 bpf_link_cleanup(&link_primer); 9637 goto out_put_dev; 9638 } 9639 9640 fd = bpf_link_settle(&link_primer); 9641 /* link itself doesn't hold dev's refcnt to not complicate shutdown */ 9642 dev_put(dev); 9643 return fd; 9644 9645 unlock: 9646 rtnl_unlock(); 9647 9648 out_put_dev: 9649 dev_put(dev); 9650 return err; 9651 } 9652 9653 /** 9654 * dev_change_xdp_fd - set or clear a bpf program for a device rx path 9655 * @dev: device 9656 * @extack: netlink extended ack 9657 * @fd: new program fd or negative value to clear 9658 * @expected_fd: old program fd that userspace expects to replace or clear 9659 * @flags: xdp-related flags 9660 * 9661 * Set or clear a bpf program for a device 9662 */ 9663 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, 9664 int fd, int expected_fd, u32 flags) 9665 { 9666 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags); 9667 struct bpf_prog *new_prog = NULL, *old_prog = NULL; 9668 int err; 9669 9670 ASSERT_RTNL(); 9671 9672 if (fd >= 0) { 9673 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP, 9674 mode != XDP_MODE_SKB); 9675 if (IS_ERR(new_prog)) 9676 return PTR_ERR(new_prog); 9677 } 9678 9679 if (expected_fd >= 0) { 9680 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP, 9681 mode != XDP_MODE_SKB); 9682 if (IS_ERR(old_prog)) { 9683 err = PTR_ERR(old_prog); 9684 old_prog = NULL; 9685 goto err_out; 9686 } 9687 } 9688 9689 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags); 9690 9691 err_out: 9692 if (err && new_prog) 9693 bpf_prog_put(new_prog); 9694 if (old_prog) 9695 bpf_prog_put(old_prog); 9696 return err; 9697 } 9698 9699 /** 9700 * dev_new_index - allocate an ifindex 9701 * @net: the applicable net namespace 9702 * 9703 * Returns a suitable unique value for a new device interface 9704 * number. The caller must hold the rtnl semaphore or the 9705 * dev_base_lock to be sure it remains unique. 9706 */ 9707 static int dev_new_index(struct net *net) 9708 { 9709 int ifindex = net->ifindex; 9710 9711 for (;;) { 9712 if (++ifindex <= 0) 9713 ifindex = 1; 9714 if (!__dev_get_by_index(net, ifindex)) 9715 return net->ifindex = ifindex; 9716 } 9717 } 9718 9719 /* Delayed registration/unregisteration */ 9720 static LIST_HEAD(net_todo_list); 9721 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); 9722 9723 static void net_set_todo(struct net_device *dev) 9724 { 9725 list_add_tail(&dev->todo_list, &net_todo_list); 9726 dev_net(dev)->dev_unreg_count++; 9727 } 9728 9729 static netdev_features_t netdev_sync_upper_features(struct net_device *lower, 9730 struct net_device *upper, netdev_features_t features) 9731 { 9732 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9733 netdev_features_t feature; 9734 int feature_bit; 9735 9736 for_each_netdev_feature(upper_disables, feature_bit) { 9737 feature = __NETIF_F_BIT(feature_bit); 9738 if (!(upper->wanted_features & feature) 9739 && (features & feature)) { 9740 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", 9741 &feature, upper->name); 9742 features &= ~feature; 9743 } 9744 } 9745 9746 return features; 9747 } 9748 9749 static void netdev_sync_lower_features(struct net_device *upper, 9750 struct net_device *lower, netdev_features_t features) 9751 { 9752 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9753 netdev_features_t feature; 9754 int feature_bit; 9755 9756 for_each_netdev_feature(upper_disables, feature_bit) { 9757 feature = __NETIF_F_BIT(feature_bit); 9758 if (!(features & feature) && (lower->features & feature)) { 9759 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", 9760 &feature, lower->name); 9761 lower->wanted_features &= ~feature; 9762 __netdev_update_features(lower); 9763 9764 if (unlikely(lower->features & feature)) 9765 netdev_WARN(upper, "failed to disable %pNF on %s!\n", 9766 &feature, lower->name); 9767 else 9768 netdev_features_change(lower); 9769 } 9770 } 9771 } 9772 9773 static netdev_features_t netdev_fix_features(struct net_device *dev, 9774 netdev_features_t features) 9775 { 9776 /* Fix illegal checksum combinations */ 9777 if ((features & NETIF_F_HW_CSUM) && 9778 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { 9779 netdev_warn(dev, "mixed HW and IP checksum settings.\n"); 9780 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); 9781 } 9782 9783 /* TSO requires that SG is present as well. */ 9784 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { 9785 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); 9786 features &= ~NETIF_F_ALL_TSO; 9787 } 9788 9789 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && 9790 !(features & NETIF_F_IP_CSUM)) { 9791 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); 9792 features &= ~NETIF_F_TSO; 9793 features &= ~NETIF_F_TSO_ECN; 9794 } 9795 9796 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && 9797 !(features & NETIF_F_IPV6_CSUM)) { 9798 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); 9799 features &= ~NETIF_F_TSO6; 9800 } 9801 9802 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ 9803 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) 9804 features &= ~NETIF_F_TSO_MANGLEID; 9805 9806 /* TSO ECN requires that TSO is present as well. */ 9807 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) 9808 features &= ~NETIF_F_TSO_ECN; 9809 9810 /* Software GSO depends on SG. */ 9811 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { 9812 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); 9813 features &= ~NETIF_F_GSO; 9814 } 9815 9816 /* GSO partial features require GSO partial be set */ 9817 if ((features & dev->gso_partial_features) && 9818 !(features & NETIF_F_GSO_PARTIAL)) { 9819 netdev_dbg(dev, 9820 "Dropping partially supported GSO features since no GSO partial.\n"); 9821 features &= ~dev->gso_partial_features; 9822 } 9823 9824 if (!(features & NETIF_F_RXCSUM)) { 9825 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet 9826 * successfully merged by hardware must also have the 9827 * checksum verified by hardware. If the user does not 9828 * want to enable RXCSUM, logically, we should disable GRO_HW. 9829 */ 9830 if (features & NETIF_F_GRO_HW) { 9831 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n"); 9832 features &= ~NETIF_F_GRO_HW; 9833 } 9834 } 9835 9836 /* LRO/HW-GRO features cannot be combined with RX-FCS */ 9837 if (features & NETIF_F_RXFCS) { 9838 if (features & NETIF_F_LRO) { 9839 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n"); 9840 features &= ~NETIF_F_LRO; 9841 } 9842 9843 if (features & NETIF_F_GRO_HW) { 9844 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n"); 9845 features &= ~NETIF_F_GRO_HW; 9846 } 9847 } 9848 9849 if (features & NETIF_F_HW_TLS_TX) { 9850 bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) == 9851 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM); 9852 bool hw_csum = features & NETIF_F_HW_CSUM; 9853 9854 if (!ip_csum && !hw_csum) { 9855 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n"); 9856 features &= ~NETIF_F_HW_TLS_TX; 9857 } 9858 } 9859 9860 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) { 9861 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n"); 9862 features &= ~NETIF_F_HW_TLS_RX; 9863 } 9864 9865 return features; 9866 } 9867 9868 int __netdev_update_features(struct net_device *dev) 9869 { 9870 struct net_device *upper, *lower; 9871 netdev_features_t features; 9872 struct list_head *iter; 9873 int err = -1; 9874 9875 ASSERT_RTNL(); 9876 9877 features = netdev_get_wanted_features(dev); 9878 9879 if (dev->netdev_ops->ndo_fix_features) 9880 features = dev->netdev_ops->ndo_fix_features(dev, features); 9881 9882 /* driver might be less strict about feature dependencies */ 9883 features = netdev_fix_features(dev, features); 9884 9885 /* some features can't be enabled if they're off on an upper device */ 9886 netdev_for_each_upper_dev_rcu(dev, upper, iter) 9887 features = netdev_sync_upper_features(dev, upper, features); 9888 9889 if (dev->features == features) 9890 goto sync_lower; 9891 9892 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", 9893 &dev->features, &features); 9894 9895 if (dev->netdev_ops->ndo_set_features) 9896 err = dev->netdev_ops->ndo_set_features(dev, features); 9897 else 9898 err = 0; 9899 9900 if (unlikely(err < 0)) { 9901 netdev_err(dev, 9902 "set_features() failed (%d); wanted %pNF, left %pNF\n", 9903 err, &features, &dev->features); 9904 /* return non-0 since some features might have changed and 9905 * it's better to fire a spurious notification than miss it 9906 */ 9907 return -1; 9908 } 9909 9910 sync_lower: 9911 /* some features must be disabled on lower devices when disabled 9912 * on an upper device (think: bonding master or bridge) 9913 */ 9914 netdev_for_each_lower_dev(dev, lower, iter) 9915 netdev_sync_lower_features(dev, lower, features); 9916 9917 if (!err) { 9918 netdev_features_t diff = features ^ dev->features; 9919 9920 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) { 9921 /* udp_tunnel_{get,drop}_rx_info both need 9922 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the 9923 * device, or they won't do anything. 9924 * Thus we need to update dev->features 9925 * *before* calling udp_tunnel_get_rx_info, 9926 * but *after* calling udp_tunnel_drop_rx_info. 9927 */ 9928 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) { 9929 dev->features = features; 9930 udp_tunnel_get_rx_info(dev); 9931 } else { 9932 udp_tunnel_drop_rx_info(dev); 9933 } 9934 } 9935 9936 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) { 9937 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) { 9938 dev->features = features; 9939 err |= vlan_get_rx_ctag_filter_info(dev); 9940 } else { 9941 vlan_drop_rx_ctag_filter_info(dev); 9942 } 9943 } 9944 9945 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) { 9946 if (features & NETIF_F_HW_VLAN_STAG_FILTER) { 9947 dev->features = features; 9948 err |= vlan_get_rx_stag_filter_info(dev); 9949 } else { 9950 vlan_drop_rx_stag_filter_info(dev); 9951 } 9952 } 9953 9954 dev->features = features; 9955 } 9956 9957 return err < 0 ? 0 : 1; 9958 } 9959 9960 /** 9961 * netdev_update_features - recalculate device features 9962 * @dev: the device to check 9963 * 9964 * Recalculate dev->features set and send notifications if it 9965 * has changed. Should be called after driver or hardware dependent 9966 * conditions might have changed that influence the features. 9967 */ 9968 void netdev_update_features(struct net_device *dev) 9969 { 9970 if (__netdev_update_features(dev)) 9971 netdev_features_change(dev); 9972 } 9973 EXPORT_SYMBOL(netdev_update_features); 9974 9975 /** 9976 * netdev_change_features - recalculate device features 9977 * @dev: the device to check 9978 * 9979 * Recalculate dev->features set and send notifications even 9980 * if they have not changed. Should be called instead of 9981 * netdev_update_features() if also dev->vlan_features might 9982 * have changed to allow the changes to be propagated to stacked 9983 * VLAN devices. 9984 */ 9985 void netdev_change_features(struct net_device *dev) 9986 { 9987 __netdev_update_features(dev); 9988 netdev_features_change(dev); 9989 } 9990 EXPORT_SYMBOL(netdev_change_features); 9991 9992 /** 9993 * netif_stacked_transfer_operstate - transfer operstate 9994 * @rootdev: the root or lower level device to transfer state from 9995 * @dev: the device to transfer operstate to 9996 * 9997 * Transfer operational state from root to device. This is normally 9998 * called when a stacking relationship exists between the root 9999 * device and the device(a leaf device). 10000 */ 10001 void netif_stacked_transfer_operstate(const struct net_device *rootdev, 10002 struct net_device *dev) 10003 { 10004 if (rootdev->operstate == IF_OPER_DORMANT) 10005 netif_dormant_on(dev); 10006 else 10007 netif_dormant_off(dev); 10008 10009 if (rootdev->operstate == IF_OPER_TESTING) 10010 netif_testing_on(dev); 10011 else 10012 netif_testing_off(dev); 10013 10014 if (netif_carrier_ok(rootdev)) 10015 netif_carrier_on(dev); 10016 else 10017 netif_carrier_off(dev); 10018 } 10019 EXPORT_SYMBOL(netif_stacked_transfer_operstate); 10020 10021 static int netif_alloc_rx_queues(struct net_device *dev) 10022 { 10023 unsigned int i, count = dev->num_rx_queues; 10024 struct netdev_rx_queue *rx; 10025 size_t sz = count * sizeof(*rx); 10026 int err = 0; 10027 10028 BUG_ON(count < 1); 10029 10030 rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10031 if (!rx) 10032 return -ENOMEM; 10033 10034 dev->_rx = rx; 10035 10036 for (i = 0; i < count; i++) { 10037 rx[i].dev = dev; 10038 10039 /* XDP RX-queue setup */ 10040 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0); 10041 if (err < 0) 10042 goto err_rxq_info; 10043 } 10044 return 0; 10045 10046 err_rxq_info: 10047 /* Rollback successful reg's and free other resources */ 10048 while (i--) 10049 xdp_rxq_info_unreg(&rx[i].xdp_rxq); 10050 kvfree(dev->_rx); 10051 dev->_rx = NULL; 10052 return err; 10053 } 10054 10055 static void netif_free_rx_queues(struct net_device *dev) 10056 { 10057 unsigned int i, count = dev->num_rx_queues; 10058 10059 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */ 10060 if (!dev->_rx) 10061 return; 10062 10063 for (i = 0; i < count; i++) 10064 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq); 10065 10066 kvfree(dev->_rx); 10067 } 10068 10069 static void netdev_init_one_queue(struct net_device *dev, 10070 struct netdev_queue *queue, void *_unused) 10071 { 10072 /* Initialize queue lock */ 10073 spin_lock_init(&queue->_xmit_lock); 10074 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); 10075 queue->xmit_lock_owner = -1; 10076 netdev_queue_numa_node_write(queue, NUMA_NO_NODE); 10077 queue->dev = dev; 10078 #ifdef CONFIG_BQL 10079 dql_init(&queue->dql, HZ); 10080 #endif 10081 } 10082 10083 static void netif_free_tx_queues(struct net_device *dev) 10084 { 10085 kvfree(dev->_tx); 10086 } 10087 10088 static int netif_alloc_netdev_queues(struct net_device *dev) 10089 { 10090 unsigned int count = dev->num_tx_queues; 10091 struct netdev_queue *tx; 10092 size_t sz = count * sizeof(*tx); 10093 10094 if (count < 1 || count > 0xffff) 10095 return -EINVAL; 10096 10097 tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10098 if (!tx) 10099 return -ENOMEM; 10100 10101 dev->_tx = tx; 10102 10103 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); 10104 spin_lock_init(&dev->tx_global_lock); 10105 10106 return 0; 10107 } 10108 10109 void netif_tx_stop_all_queues(struct net_device *dev) 10110 { 10111 unsigned int i; 10112 10113 for (i = 0; i < dev->num_tx_queues; i++) { 10114 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 10115 10116 netif_tx_stop_queue(txq); 10117 } 10118 } 10119 EXPORT_SYMBOL(netif_tx_stop_all_queues); 10120 10121 /** 10122 * register_netdevice - register a network device 10123 * @dev: device to register 10124 * 10125 * Take a completed network device structure and add it to the kernel 10126 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 10127 * chain. 0 is returned on success. A negative errno code is returned 10128 * on a failure to set up the device, or if the name is a duplicate. 10129 * 10130 * Callers must hold the rtnl semaphore. You may want 10131 * register_netdev() instead of this. 10132 * 10133 * BUGS: 10134 * The locking appears insufficient to guarantee two parallel registers 10135 * will not get the same name. 10136 */ 10137 10138 int register_netdevice(struct net_device *dev) 10139 { 10140 int ret; 10141 struct net *net = dev_net(dev); 10142 10143 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE < 10144 NETDEV_FEATURE_COUNT); 10145 BUG_ON(dev_boot_phase); 10146 ASSERT_RTNL(); 10147 10148 might_sleep(); 10149 10150 /* When net_device's are persistent, this will be fatal. */ 10151 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); 10152 BUG_ON(!net); 10153 10154 ret = ethtool_check_ops(dev->ethtool_ops); 10155 if (ret) 10156 return ret; 10157 10158 spin_lock_init(&dev->addr_list_lock); 10159 netdev_set_addr_lockdep_class(dev); 10160 10161 ret = dev_get_valid_name(net, dev, dev->name); 10162 if (ret < 0) 10163 goto out; 10164 10165 ret = -ENOMEM; 10166 dev->name_node = netdev_name_node_head_alloc(dev); 10167 if (!dev->name_node) 10168 goto out; 10169 10170 /* Init, if this function is available */ 10171 if (dev->netdev_ops->ndo_init) { 10172 ret = dev->netdev_ops->ndo_init(dev); 10173 if (ret) { 10174 if (ret > 0) 10175 ret = -EIO; 10176 goto err_free_name; 10177 } 10178 } 10179 10180 if (((dev->hw_features | dev->features) & 10181 NETIF_F_HW_VLAN_CTAG_FILTER) && 10182 (!dev->netdev_ops->ndo_vlan_rx_add_vid || 10183 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { 10184 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); 10185 ret = -EINVAL; 10186 goto err_uninit; 10187 } 10188 10189 ret = -EBUSY; 10190 if (!dev->ifindex) 10191 dev->ifindex = dev_new_index(net); 10192 else if (__dev_get_by_index(net, dev->ifindex)) 10193 goto err_uninit; 10194 10195 /* Transfer changeable features to wanted_features and enable 10196 * software offloads (GSO and GRO). 10197 */ 10198 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF); 10199 dev->features |= NETIF_F_SOFT_FEATURES; 10200 10201 if (dev->udp_tunnel_nic_info) { 10202 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10203 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10204 } 10205 10206 dev->wanted_features = dev->features & dev->hw_features; 10207 10208 if (!(dev->flags & IFF_LOOPBACK)) 10209 dev->hw_features |= NETIF_F_NOCACHE_COPY; 10210 10211 /* If IPv4 TCP segmentation offload is supported we should also 10212 * allow the device to enable segmenting the frame with the option 10213 * of ignoring a static IP ID value. This doesn't enable the 10214 * feature itself but allows the user to enable it later. 10215 */ 10216 if (dev->hw_features & NETIF_F_TSO) 10217 dev->hw_features |= NETIF_F_TSO_MANGLEID; 10218 if (dev->vlan_features & NETIF_F_TSO) 10219 dev->vlan_features |= NETIF_F_TSO_MANGLEID; 10220 if (dev->mpls_features & NETIF_F_TSO) 10221 dev->mpls_features |= NETIF_F_TSO_MANGLEID; 10222 if (dev->hw_enc_features & NETIF_F_TSO) 10223 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; 10224 10225 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. 10226 */ 10227 dev->vlan_features |= NETIF_F_HIGHDMA; 10228 10229 /* Make NETIF_F_SG inheritable to tunnel devices. 10230 */ 10231 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; 10232 10233 /* Make NETIF_F_SG inheritable to MPLS. 10234 */ 10235 dev->mpls_features |= NETIF_F_SG; 10236 10237 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); 10238 ret = notifier_to_errno(ret); 10239 if (ret) 10240 goto err_uninit; 10241 10242 ret = netdev_register_kobject(dev); 10243 if (ret) { 10244 dev->reg_state = NETREG_UNREGISTERED; 10245 goto err_uninit; 10246 } 10247 dev->reg_state = NETREG_REGISTERED; 10248 10249 __netdev_update_features(dev); 10250 10251 /* 10252 * Default initial state at registry is that the 10253 * device is present. 10254 */ 10255 10256 set_bit(__LINK_STATE_PRESENT, &dev->state); 10257 10258 linkwatch_init_dev(dev); 10259 10260 dev_init_scheduler(dev); 10261 dev_hold(dev); 10262 list_netdevice(dev); 10263 add_device_randomness(dev->dev_addr, dev->addr_len); 10264 10265 /* If the device has permanent device address, driver should 10266 * set dev_addr and also addr_assign_type should be set to 10267 * NET_ADDR_PERM (default value). 10268 */ 10269 if (dev->addr_assign_type == NET_ADDR_PERM) 10270 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); 10271 10272 /* Notify protocols, that a new device appeared. */ 10273 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); 10274 ret = notifier_to_errno(ret); 10275 if (ret) { 10276 /* Expect explicit free_netdev() on failure */ 10277 dev->needs_free_netdev = false; 10278 unregister_netdevice_queue(dev, NULL); 10279 goto out; 10280 } 10281 /* 10282 * Prevent userspace races by waiting until the network 10283 * device is fully setup before sending notifications. 10284 */ 10285 if (!dev->rtnl_link_ops || 10286 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 10287 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); 10288 10289 out: 10290 return ret; 10291 10292 err_uninit: 10293 if (dev->netdev_ops->ndo_uninit) 10294 dev->netdev_ops->ndo_uninit(dev); 10295 if (dev->priv_destructor) 10296 dev->priv_destructor(dev); 10297 err_free_name: 10298 netdev_name_node_free(dev->name_node); 10299 goto out; 10300 } 10301 EXPORT_SYMBOL(register_netdevice); 10302 10303 /** 10304 * init_dummy_netdev - init a dummy network device for NAPI 10305 * @dev: device to init 10306 * 10307 * This takes a network device structure and initialize the minimum 10308 * amount of fields so it can be used to schedule NAPI polls without 10309 * registering a full blown interface. This is to be used by drivers 10310 * that need to tie several hardware interfaces to a single NAPI 10311 * poll scheduler due to HW limitations. 10312 */ 10313 int init_dummy_netdev(struct net_device *dev) 10314 { 10315 /* Clear everything. Note we don't initialize spinlocks 10316 * are they aren't supposed to be taken by any of the 10317 * NAPI code and this dummy netdev is supposed to be 10318 * only ever used for NAPI polls 10319 */ 10320 memset(dev, 0, sizeof(struct net_device)); 10321 10322 /* make sure we BUG if trying to hit standard 10323 * register/unregister code path 10324 */ 10325 dev->reg_state = NETREG_DUMMY; 10326 10327 /* NAPI wants this */ 10328 INIT_LIST_HEAD(&dev->napi_list); 10329 10330 /* a dummy interface is started by default */ 10331 set_bit(__LINK_STATE_PRESENT, &dev->state); 10332 set_bit(__LINK_STATE_START, &dev->state); 10333 10334 /* napi_busy_loop stats accounting wants this */ 10335 dev_net_set(dev, &init_net); 10336 10337 /* Note : We dont allocate pcpu_refcnt for dummy devices, 10338 * because users of this 'device' dont need to change 10339 * its refcount. 10340 */ 10341 10342 return 0; 10343 } 10344 EXPORT_SYMBOL_GPL(init_dummy_netdev); 10345 10346 10347 /** 10348 * register_netdev - register a network device 10349 * @dev: device to register 10350 * 10351 * Take a completed network device structure and add it to the kernel 10352 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 10353 * chain. 0 is returned on success. A negative errno code is returned 10354 * on a failure to set up the device, or if the name is a duplicate. 10355 * 10356 * This is a wrapper around register_netdevice that takes the rtnl semaphore 10357 * and expands the device name if you passed a format string to 10358 * alloc_netdev. 10359 */ 10360 int register_netdev(struct net_device *dev) 10361 { 10362 int err; 10363 10364 if (rtnl_lock_killable()) 10365 return -EINTR; 10366 err = register_netdevice(dev); 10367 rtnl_unlock(); 10368 return err; 10369 } 10370 EXPORT_SYMBOL(register_netdev); 10371 10372 int netdev_refcnt_read(const struct net_device *dev) 10373 { 10374 #ifdef CONFIG_PCPU_DEV_REFCNT 10375 int i, refcnt = 0; 10376 10377 for_each_possible_cpu(i) 10378 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); 10379 return refcnt; 10380 #else 10381 return refcount_read(&dev->dev_refcnt); 10382 #endif 10383 } 10384 EXPORT_SYMBOL(netdev_refcnt_read); 10385 10386 int netdev_unregister_timeout_secs __read_mostly = 10; 10387 10388 #define WAIT_REFS_MIN_MSECS 1 10389 #define WAIT_REFS_MAX_MSECS 250 10390 /** 10391 * netdev_wait_allrefs - wait until all references are gone. 10392 * @dev: target net_device 10393 * 10394 * This is called when unregistering network devices. 10395 * 10396 * Any protocol or device that holds a reference should register 10397 * for netdevice notification, and cleanup and put back the 10398 * reference if they receive an UNREGISTER event. 10399 * We can get stuck here if buggy protocols don't correctly 10400 * call dev_put. 10401 */ 10402 static void netdev_wait_allrefs(struct net_device *dev) 10403 { 10404 unsigned long rebroadcast_time, warning_time; 10405 int wait = 0, refcnt; 10406 10407 linkwatch_forget_dev(dev); 10408 10409 rebroadcast_time = warning_time = jiffies; 10410 refcnt = netdev_refcnt_read(dev); 10411 10412 while (refcnt != 1) { 10413 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { 10414 rtnl_lock(); 10415 10416 /* Rebroadcast unregister notification */ 10417 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10418 10419 __rtnl_unlock(); 10420 rcu_barrier(); 10421 rtnl_lock(); 10422 10423 if (test_bit(__LINK_STATE_LINKWATCH_PENDING, 10424 &dev->state)) { 10425 /* We must not have linkwatch events 10426 * pending on unregister. If this 10427 * happens, we simply run the queue 10428 * unscheduled, resulting in a noop 10429 * for this device. 10430 */ 10431 linkwatch_run_queue(); 10432 } 10433 10434 __rtnl_unlock(); 10435 10436 rebroadcast_time = jiffies; 10437 } 10438 10439 if (!wait) { 10440 rcu_barrier(); 10441 wait = WAIT_REFS_MIN_MSECS; 10442 } else { 10443 msleep(wait); 10444 wait = min(wait << 1, WAIT_REFS_MAX_MSECS); 10445 } 10446 10447 refcnt = netdev_refcnt_read(dev); 10448 10449 if (refcnt != 1 && 10450 time_after(jiffies, warning_time + 10451 netdev_unregister_timeout_secs * HZ)) { 10452 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", 10453 dev->name, refcnt); 10454 warning_time = jiffies; 10455 } 10456 } 10457 } 10458 10459 /* The sequence is: 10460 * 10461 * rtnl_lock(); 10462 * ... 10463 * register_netdevice(x1); 10464 * register_netdevice(x2); 10465 * ... 10466 * unregister_netdevice(y1); 10467 * unregister_netdevice(y2); 10468 * ... 10469 * rtnl_unlock(); 10470 * free_netdev(y1); 10471 * free_netdev(y2); 10472 * 10473 * We are invoked by rtnl_unlock(). 10474 * This allows us to deal with problems: 10475 * 1) We can delete sysfs objects which invoke hotplug 10476 * without deadlocking with linkwatch via keventd. 10477 * 2) Since we run with the RTNL semaphore not held, we can sleep 10478 * safely in order to wait for the netdev refcnt to drop to zero. 10479 * 10480 * We must not return until all unregister events added during 10481 * the interval the lock was held have been completed. 10482 */ 10483 void netdev_run_todo(void) 10484 { 10485 struct list_head list; 10486 #ifdef CONFIG_LOCKDEP 10487 struct list_head unlink_list; 10488 10489 list_replace_init(&net_unlink_list, &unlink_list); 10490 10491 while (!list_empty(&unlink_list)) { 10492 struct net_device *dev = list_first_entry(&unlink_list, 10493 struct net_device, 10494 unlink_list); 10495 list_del_init(&dev->unlink_list); 10496 dev->nested_level = dev->lower_level - 1; 10497 } 10498 #endif 10499 10500 /* Snapshot list, allow later requests */ 10501 list_replace_init(&net_todo_list, &list); 10502 10503 __rtnl_unlock(); 10504 10505 10506 /* Wait for rcu callbacks to finish before next phase */ 10507 if (!list_empty(&list)) 10508 rcu_barrier(); 10509 10510 while (!list_empty(&list)) { 10511 struct net_device *dev 10512 = list_first_entry(&list, struct net_device, todo_list); 10513 list_del(&dev->todo_list); 10514 10515 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { 10516 pr_err("network todo '%s' but state %d\n", 10517 dev->name, dev->reg_state); 10518 dump_stack(); 10519 continue; 10520 } 10521 10522 dev->reg_state = NETREG_UNREGISTERED; 10523 10524 netdev_wait_allrefs(dev); 10525 10526 /* paranoia */ 10527 BUG_ON(netdev_refcnt_read(dev) != 1); 10528 BUG_ON(!list_empty(&dev->ptype_all)); 10529 BUG_ON(!list_empty(&dev->ptype_specific)); 10530 WARN_ON(rcu_access_pointer(dev->ip_ptr)); 10531 WARN_ON(rcu_access_pointer(dev->ip6_ptr)); 10532 #if IS_ENABLED(CONFIG_DECNET) 10533 WARN_ON(dev->dn_ptr); 10534 #endif 10535 if (dev->priv_destructor) 10536 dev->priv_destructor(dev); 10537 if (dev->needs_free_netdev) 10538 free_netdev(dev); 10539 10540 /* Report a network device has been unregistered */ 10541 rtnl_lock(); 10542 dev_net(dev)->dev_unreg_count--; 10543 __rtnl_unlock(); 10544 wake_up(&netdev_unregistering_wq); 10545 10546 /* Free network device */ 10547 kobject_put(&dev->dev.kobj); 10548 } 10549 } 10550 10551 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has 10552 * all the same fields in the same order as net_device_stats, with only 10553 * the type differing, but rtnl_link_stats64 may have additional fields 10554 * at the end for newer counters. 10555 */ 10556 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, 10557 const struct net_device_stats *netdev_stats) 10558 { 10559 #if BITS_PER_LONG == 64 10560 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats)); 10561 memcpy(stats64, netdev_stats, sizeof(*netdev_stats)); 10562 /* zero out counters that only exist in rtnl_link_stats64 */ 10563 memset((char *)stats64 + sizeof(*netdev_stats), 0, 10564 sizeof(*stats64) - sizeof(*netdev_stats)); 10565 #else 10566 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long); 10567 const unsigned long *src = (const unsigned long *)netdev_stats; 10568 u64 *dst = (u64 *)stats64; 10569 10570 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); 10571 for (i = 0; i < n; i++) 10572 dst[i] = src[i]; 10573 /* zero out counters that only exist in rtnl_link_stats64 */ 10574 memset((char *)stats64 + n * sizeof(u64), 0, 10575 sizeof(*stats64) - n * sizeof(u64)); 10576 #endif 10577 } 10578 EXPORT_SYMBOL(netdev_stats_to_stats64); 10579 10580 /** 10581 * dev_get_stats - get network device statistics 10582 * @dev: device to get statistics from 10583 * @storage: place to store stats 10584 * 10585 * Get network statistics from device. Return @storage. 10586 * The device driver may provide its own method by setting 10587 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; 10588 * otherwise the internal statistics structure is used. 10589 */ 10590 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, 10591 struct rtnl_link_stats64 *storage) 10592 { 10593 const struct net_device_ops *ops = dev->netdev_ops; 10594 10595 if (ops->ndo_get_stats64) { 10596 memset(storage, 0, sizeof(*storage)); 10597 ops->ndo_get_stats64(dev, storage); 10598 } else if (ops->ndo_get_stats) { 10599 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); 10600 } else { 10601 netdev_stats_to_stats64(storage, &dev->stats); 10602 } 10603 storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped); 10604 storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped); 10605 storage->rx_nohandler += (unsigned long)atomic_long_read(&dev->rx_nohandler); 10606 return storage; 10607 } 10608 EXPORT_SYMBOL(dev_get_stats); 10609 10610 /** 10611 * dev_fetch_sw_netstats - get per-cpu network device statistics 10612 * @s: place to store stats 10613 * @netstats: per-cpu network stats to read from 10614 * 10615 * Read per-cpu network statistics and populate the related fields in @s. 10616 */ 10617 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s, 10618 const struct pcpu_sw_netstats __percpu *netstats) 10619 { 10620 int cpu; 10621 10622 for_each_possible_cpu(cpu) { 10623 const struct pcpu_sw_netstats *stats; 10624 struct pcpu_sw_netstats tmp; 10625 unsigned int start; 10626 10627 stats = per_cpu_ptr(netstats, cpu); 10628 do { 10629 start = u64_stats_fetch_begin_irq(&stats->syncp); 10630 tmp.rx_packets = stats->rx_packets; 10631 tmp.rx_bytes = stats->rx_bytes; 10632 tmp.tx_packets = stats->tx_packets; 10633 tmp.tx_bytes = stats->tx_bytes; 10634 } while (u64_stats_fetch_retry_irq(&stats->syncp, start)); 10635 10636 s->rx_packets += tmp.rx_packets; 10637 s->rx_bytes += tmp.rx_bytes; 10638 s->tx_packets += tmp.tx_packets; 10639 s->tx_bytes += tmp.tx_bytes; 10640 } 10641 } 10642 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats); 10643 10644 /** 10645 * dev_get_tstats64 - ndo_get_stats64 implementation 10646 * @dev: device to get statistics from 10647 * @s: place to store stats 10648 * 10649 * Populate @s from dev->stats and dev->tstats. Can be used as 10650 * ndo_get_stats64() callback. 10651 */ 10652 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s) 10653 { 10654 netdev_stats_to_stats64(s, &dev->stats); 10655 dev_fetch_sw_netstats(s, dev->tstats); 10656 } 10657 EXPORT_SYMBOL_GPL(dev_get_tstats64); 10658 10659 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) 10660 { 10661 struct netdev_queue *queue = dev_ingress_queue(dev); 10662 10663 #ifdef CONFIG_NET_CLS_ACT 10664 if (queue) 10665 return queue; 10666 queue = kzalloc(sizeof(*queue), GFP_KERNEL); 10667 if (!queue) 10668 return NULL; 10669 netdev_init_one_queue(dev, queue, NULL); 10670 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); 10671 queue->qdisc_sleeping = &noop_qdisc; 10672 rcu_assign_pointer(dev->ingress_queue, queue); 10673 #endif 10674 return queue; 10675 } 10676 10677 static const struct ethtool_ops default_ethtool_ops; 10678 10679 void netdev_set_default_ethtool_ops(struct net_device *dev, 10680 const struct ethtool_ops *ops) 10681 { 10682 if (dev->ethtool_ops == &default_ethtool_ops) 10683 dev->ethtool_ops = ops; 10684 } 10685 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); 10686 10687 void netdev_freemem(struct net_device *dev) 10688 { 10689 char *addr = (char *)dev - dev->padded; 10690 10691 kvfree(addr); 10692 } 10693 10694 /** 10695 * alloc_netdev_mqs - allocate network device 10696 * @sizeof_priv: size of private data to allocate space for 10697 * @name: device name format string 10698 * @name_assign_type: origin of device name 10699 * @setup: callback to initialize device 10700 * @txqs: the number of TX subqueues to allocate 10701 * @rxqs: the number of RX subqueues to allocate 10702 * 10703 * Allocates a struct net_device with private data area for driver use 10704 * and performs basic initialization. Also allocates subqueue structs 10705 * for each queue on the device. 10706 */ 10707 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, 10708 unsigned char name_assign_type, 10709 void (*setup)(struct net_device *), 10710 unsigned int txqs, unsigned int rxqs) 10711 { 10712 struct net_device *dev; 10713 unsigned int alloc_size; 10714 struct net_device *p; 10715 10716 BUG_ON(strlen(name) >= sizeof(dev->name)); 10717 10718 if (txqs < 1) { 10719 pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); 10720 return NULL; 10721 } 10722 10723 if (rxqs < 1) { 10724 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); 10725 return NULL; 10726 } 10727 10728 alloc_size = sizeof(struct net_device); 10729 if (sizeof_priv) { 10730 /* ensure 32-byte alignment of private area */ 10731 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); 10732 alloc_size += sizeof_priv; 10733 } 10734 /* ensure 32-byte alignment of whole construct */ 10735 alloc_size += NETDEV_ALIGN - 1; 10736 10737 p = kvzalloc(alloc_size, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10738 if (!p) 10739 return NULL; 10740 10741 dev = PTR_ALIGN(p, NETDEV_ALIGN); 10742 dev->padded = (char *)dev - (char *)p; 10743 10744 #ifdef CONFIG_PCPU_DEV_REFCNT 10745 dev->pcpu_refcnt = alloc_percpu(int); 10746 if (!dev->pcpu_refcnt) 10747 goto free_dev; 10748 dev_hold(dev); 10749 #else 10750 refcount_set(&dev->dev_refcnt, 1); 10751 #endif 10752 10753 if (dev_addr_init(dev)) 10754 goto free_pcpu; 10755 10756 dev_mc_init(dev); 10757 dev_uc_init(dev); 10758 10759 dev_net_set(dev, &init_net); 10760 10761 dev->gso_max_size = GSO_MAX_SIZE; 10762 dev->gso_max_segs = GSO_MAX_SEGS; 10763 dev->upper_level = 1; 10764 dev->lower_level = 1; 10765 #ifdef CONFIG_LOCKDEP 10766 dev->nested_level = 0; 10767 INIT_LIST_HEAD(&dev->unlink_list); 10768 #endif 10769 10770 INIT_LIST_HEAD(&dev->napi_list); 10771 INIT_LIST_HEAD(&dev->unreg_list); 10772 INIT_LIST_HEAD(&dev->close_list); 10773 INIT_LIST_HEAD(&dev->link_watch_list); 10774 INIT_LIST_HEAD(&dev->adj_list.upper); 10775 INIT_LIST_HEAD(&dev->adj_list.lower); 10776 INIT_LIST_HEAD(&dev->ptype_all); 10777 INIT_LIST_HEAD(&dev->ptype_specific); 10778 INIT_LIST_HEAD(&dev->net_notifier_list); 10779 #ifdef CONFIG_NET_SCHED 10780 hash_init(dev->qdisc_hash); 10781 #endif 10782 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; 10783 setup(dev); 10784 10785 if (!dev->tx_queue_len) { 10786 dev->priv_flags |= IFF_NO_QUEUE; 10787 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; 10788 } 10789 10790 dev->num_tx_queues = txqs; 10791 dev->real_num_tx_queues = txqs; 10792 if (netif_alloc_netdev_queues(dev)) 10793 goto free_all; 10794 10795 dev->num_rx_queues = rxqs; 10796 dev->real_num_rx_queues = rxqs; 10797 if (netif_alloc_rx_queues(dev)) 10798 goto free_all; 10799 10800 strcpy(dev->name, name); 10801 dev->name_assign_type = name_assign_type; 10802 dev->group = INIT_NETDEV_GROUP; 10803 if (!dev->ethtool_ops) 10804 dev->ethtool_ops = &default_ethtool_ops; 10805 10806 nf_hook_ingress_init(dev); 10807 10808 return dev; 10809 10810 free_all: 10811 free_netdev(dev); 10812 return NULL; 10813 10814 free_pcpu: 10815 #ifdef CONFIG_PCPU_DEV_REFCNT 10816 free_percpu(dev->pcpu_refcnt); 10817 free_dev: 10818 #endif 10819 netdev_freemem(dev); 10820 return NULL; 10821 } 10822 EXPORT_SYMBOL(alloc_netdev_mqs); 10823 10824 /** 10825 * free_netdev - free network device 10826 * @dev: device 10827 * 10828 * This function does the last stage of destroying an allocated device 10829 * interface. The reference to the device object is released. If this 10830 * is the last reference then it will be freed.Must be called in process 10831 * context. 10832 */ 10833 void free_netdev(struct net_device *dev) 10834 { 10835 struct napi_struct *p, *n; 10836 10837 might_sleep(); 10838 10839 /* When called immediately after register_netdevice() failed the unwind 10840 * handling may still be dismantling the device. Handle that case by 10841 * deferring the free. 10842 */ 10843 if (dev->reg_state == NETREG_UNREGISTERING) { 10844 ASSERT_RTNL(); 10845 dev->needs_free_netdev = true; 10846 return; 10847 } 10848 10849 netif_free_tx_queues(dev); 10850 netif_free_rx_queues(dev); 10851 10852 kfree(rcu_dereference_protected(dev->ingress_queue, 1)); 10853 10854 /* Flush device addresses */ 10855 dev_addr_flush(dev); 10856 10857 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) 10858 netif_napi_del(p); 10859 10860 #ifdef CONFIG_PCPU_DEV_REFCNT 10861 free_percpu(dev->pcpu_refcnt); 10862 dev->pcpu_refcnt = NULL; 10863 #endif 10864 free_percpu(dev->xdp_bulkq); 10865 dev->xdp_bulkq = NULL; 10866 10867 /* Compatibility with error handling in drivers */ 10868 if (dev->reg_state == NETREG_UNINITIALIZED) { 10869 netdev_freemem(dev); 10870 return; 10871 } 10872 10873 BUG_ON(dev->reg_state != NETREG_UNREGISTERED); 10874 dev->reg_state = NETREG_RELEASED; 10875 10876 /* will free via device release */ 10877 put_device(&dev->dev); 10878 } 10879 EXPORT_SYMBOL(free_netdev); 10880 10881 /** 10882 * synchronize_net - Synchronize with packet receive processing 10883 * 10884 * Wait for packets currently being received to be done. 10885 * Does not block later packets from starting. 10886 */ 10887 void synchronize_net(void) 10888 { 10889 might_sleep(); 10890 if (rtnl_is_locked()) 10891 synchronize_rcu_expedited(); 10892 else 10893 synchronize_rcu(); 10894 } 10895 EXPORT_SYMBOL(synchronize_net); 10896 10897 /** 10898 * unregister_netdevice_queue - remove device from the kernel 10899 * @dev: device 10900 * @head: list 10901 * 10902 * This function shuts down a device interface and removes it 10903 * from the kernel tables. 10904 * If head not NULL, device is queued to be unregistered later. 10905 * 10906 * Callers must hold the rtnl semaphore. You may want 10907 * unregister_netdev() instead of this. 10908 */ 10909 10910 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) 10911 { 10912 ASSERT_RTNL(); 10913 10914 if (head) { 10915 list_move_tail(&dev->unreg_list, head); 10916 } else { 10917 LIST_HEAD(single); 10918 10919 list_add(&dev->unreg_list, &single); 10920 unregister_netdevice_many(&single); 10921 } 10922 } 10923 EXPORT_SYMBOL(unregister_netdevice_queue); 10924 10925 /** 10926 * unregister_netdevice_many - unregister many devices 10927 * @head: list of devices 10928 * 10929 * Note: As most callers use a stack allocated list_head, 10930 * we force a list_del() to make sure stack wont be corrupted later. 10931 */ 10932 void unregister_netdevice_many(struct list_head *head) 10933 { 10934 struct net_device *dev, *tmp; 10935 LIST_HEAD(close_head); 10936 10937 BUG_ON(dev_boot_phase); 10938 ASSERT_RTNL(); 10939 10940 if (list_empty(head)) 10941 return; 10942 10943 list_for_each_entry_safe(dev, tmp, head, unreg_list) { 10944 /* Some devices call without registering 10945 * for initialization unwind. Remove those 10946 * devices and proceed with the remaining. 10947 */ 10948 if (dev->reg_state == NETREG_UNINITIALIZED) { 10949 pr_debug("unregister_netdevice: device %s/%p never was registered\n", 10950 dev->name, dev); 10951 10952 WARN_ON(1); 10953 list_del(&dev->unreg_list); 10954 continue; 10955 } 10956 dev->dismantle = true; 10957 BUG_ON(dev->reg_state != NETREG_REGISTERED); 10958 } 10959 10960 /* If device is running, close it first. */ 10961 list_for_each_entry(dev, head, unreg_list) 10962 list_add_tail(&dev->close_list, &close_head); 10963 dev_close_many(&close_head, true); 10964 10965 list_for_each_entry(dev, head, unreg_list) { 10966 /* And unlink it from device chain. */ 10967 unlist_netdevice(dev); 10968 10969 dev->reg_state = NETREG_UNREGISTERING; 10970 } 10971 flush_all_backlogs(); 10972 10973 synchronize_net(); 10974 10975 list_for_each_entry(dev, head, unreg_list) { 10976 struct sk_buff *skb = NULL; 10977 10978 /* Shutdown queueing discipline. */ 10979 dev_shutdown(dev); 10980 10981 dev_xdp_uninstall(dev); 10982 10983 /* Notify protocols, that we are about to destroy 10984 * this device. They should clean all the things. 10985 */ 10986 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10987 10988 if (!dev->rtnl_link_ops || 10989 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 10990 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0, 10991 GFP_KERNEL, NULL, 0); 10992 10993 /* 10994 * Flush the unicast and multicast chains 10995 */ 10996 dev_uc_flush(dev); 10997 dev_mc_flush(dev); 10998 10999 netdev_name_node_alt_flush(dev); 11000 netdev_name_node_free(dev->name_node); 11001 11002 if (dev->netdev_ops->ndo_uninit) 11003 dev->netdev_ops->ndo_uninit(dev); 11004 11005 if (skb) 11006 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL); 11007 11008 /* Notifier chain MUST detach us all upper devices. */ 11009 WARN_ON(netdev_has_any_upper_dev(dev)); 11010 WARN_ON(netdev_has_any_lower_dev(dev)); 11011 11012 /* Remove entries from kobject tree */ 11013 netdev_unregister_kobject(dev); 11014 #ifdef CONFIG_XPS 11015 /* Remove XPS queueing entries */ 11016 netif_reset_xps_queues_gt(dev, 0); 11017 #endif 11018 } 11019 11020 synchronize_net(); 11021 11022 list_for_each_entry(dev, head, unreg_list) { 11023 dev_put(dev); 11024 net_set_todo(dev); 11025 } 11026 11027 list_del(head); 11028 } 11029 EXPORT_SYMBOL(unregister_netdevice_many); 11030 11031 /** 11032 * unregister_netdev - remove device from the kernel 11033 * @dev: device 11034 * 11035 * This function shuts down a device interface and removes it 11036 * from the kernel tables. 11037 * 11038 * This is just a wrapper for unregister_netdevice that takes 11039 * the rtnl semaphore. In general you want to use this and not 11040 * unregister_netdevice. 11041 */ 11042 void unregister_netdev(struct net_device *dev) 11043 { 11044 rtnl_lock(); 11045 unregister_netdevice(dev); 11046 rtnl_unlock(); 11047 } 11048 EXPORT_SYMBOL(unregister_netdev); 11049 11050 /** 11051 * __dev_change_net_namespace - move device to different nethost namespace 11052 * @dev: device 11053 * @net: network namespace 11054 * @pat: If not NULL name pattern to try if the current device name 11055 * is already taken in the destination network namespace. 11056 * @new_ifindex: If not zero, specifies device index in the target 11057 * namespace. 11058 * 11059 * This function shuts down a device interface and moves it 11060 * to a new network namespace. On success 0 is returned, on 11061 * a failure a netagive errno code is returned. 11062 * 11063 * Callers must hold the rtnl semaphore. 11064 */ 11065 11066 int __dev_change_net_namespace(struct net_device *dev, struct net *net, 11067 const char *pat, int new_ifindex) 11068 { 11069 struct net *net_old = dev_net(dev); 11070 int err, new_nsid; 11071 11072 ASSERT_RTNL(); 11073 11074 /* Don't allow namespace local devices to be moved. */ 11075 err = -EINVAL; 11076 if (dev->features & NETIF_F_NETNS_LOCAL) 11077 goto out; 11078 11079 /* Ensure the device has been registrered */ 11080 if (dev->reg_state != NETREG_REGISTERED) 11081 goto out; 11082 11083 /* Get out if there is nothing todo */ 11084 err = 0; 11085 if (net_eq(net_old, net)) 11086 goto out; 11087 11088 /* Pick the destination device name, and ensure 11089 * we can use it in the destination network namespace. 11090 */ 11091 err = -EEXIST; 11092 if (__dev_get_by_name(net, dev->name)) { 11093 /* We get here if we can't use the current device name */ 11094 if (!pat) 11095 goto out; 11096 err = dev_get_valid_name(net, dev, pat); 11097 if (err < 0) 11098 goto out; 11099 } 11100 11101 /* Check that new_ifindex isn't used yet. */ 11102 err = -EBUSY; 11103 if (new_ifindex && __dev_get_by_index(net, new_ifindex)) 11104 goto out; 11105 11106 /* 11107 * And now a mini version of register_netdevice unregister_netdevice. 11108 */ 11109 11110 /* If device is running close it first. */ 11111 dev_close(dev); 11112 11113 /* And unlink it from device chain */ 11114 unlist_netdevice(dev); 11115 11116 synchronize_net(); 11117 11118 /* Shutdown queueing discipline. */ 11119 dev_shutdown(dev); 11120 11121 /* Notify protocols, that we are about to destroy 11122 * this device. They should clean all the things. 11123 * 11124 * Note that dev->reg_state stays at NETREG_REGISTERED. 11125 * This is wanted because this way 8021q and macvlan know 11126 * the device is just moving and can keep their slaves up. 11127 */ 11128 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 11129 rcu_barrier(); 11130 11131 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL); 11132 /* If there is an ifindex conflict assign a new one */ 11133 if (!new_ifindex) { 11134 if (__dev_get_by_index(net, dev->ifindex)) 11135 new_ifindex = dev_new_index(net); 11136 else 11137 new_ifindex = dev->ifindex; 11138 } 11139 11140 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid, 11141 new_ifindex); 11142 11143 /* 11144 * Flush the unicast and multicast chains 11145 */ 11146 dev_uc_flush(dev); 11147 dev_mc_flush(dev); 11148 11149 /* Send a netdev-removed uevent to the old namespace */ 11150 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); 11151 netdev_adjacent_del_links(dev); 11152 11153 /* Move per-net netdevice notifiers that are following the netdevice */ 11154 move_netdevice_notifiers_dev_net(dev, net); 11155 11156 /* Actually switch the network namespace */ 11157 dev_net_set(dev, net); 11158 dev->ifindex = new_ifindex; 11159 11160 /* Send a netdev-add uevent to the new namespace */ 11161 kobject_uevent(&dev->dev.kobj, KOBJ_ADD); 11162 netdev_adjacent_add_links(dev); 11163 11164 /* Fixup kobjects */ 11165 err = device_rename(&dev->dev, dev->name); 11166 WARN_ON(err); 11167 11168 /* Adapt owner in case owning user namespace of target network 11169 * namespace is different from the original one. 11170 */ 11171 err = netdev_change_owner(dev, net_old, net); 11172 WARN_ON(err); 11173 11174 /* Add the device back in the hashes */ 11175 list_netdevice(dev); 11176 11177 /* Notify protocols, that a new device appeared. */ 11178 call_netdevice_notifiers(NETDEV_REGISTER, dev); 11179 11180 /* 11181 * Prevent userspace races by waiting until the network 11182 * device is fully setup before sending notifications. 11183 */ 11184 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); 11185 11186 synchronize_net(); 11187 err = 0; 11188 out: 11189 return err; 11190 } 11191 EXPORT_SYMBOL_GPL(__dev_change_net_namespace); 11192 11193 static int dev_cpu_dead(unsigned int oldcpu) 11194 { 11195 struct sk_buff **list_skb; 11196 struct sk_buff *skb; 11197 unsigned int cpu; 11198 struct softnet_data *sd, *oldsd, *remsd = NULL; 11199 11200 local_irq_disable(); 11201 cpu = smp_processor_id(); 11202 sd = &per_cpu(softnet_data, cpu); 11203 oldsd = &per_cpu(softnet_data, oldcpu); 11204 11205 /* Find end of our completion_queue. */ 11206 list_skb = &sd->completion_queue; 11207 while (*list_skb) 11208 list_skb = &(*list_skb)->next; 11209 /* Append completion queue from offline CPU. */ 11210 *list_skb = oldsd->completion_queue; 11211 oldsd->completion_queue = NULL; 11212 11213 /* Append output queue from offline CPU. */ 11214 if (oldsd->output_queue) { 11215 *sd->output_queue_tailp = oldsd->output_queue; 11216 sd->output_queue_tailp = oldsd->output_queue_tailp; 11217 oldsd->output_queue = NULL; 11218 oldsd->output_queue_tailp = &oldsd->output_queue; 11219 } 11220 /* Append NAPI poll list from offline CPU, with one exception : 11221 * process_backlog() must be called by cpu owning percpu backlog. 11222 * We properly handle process_queue & input_pkt_queue later. 11223 */ 11224 while (!list_empty(&oldsd->poll_list)) { 11225 struct napi_struct *napi = list_first_entry(&oldsd->poll_list, 11226 struct napi_struct, 11227 poll_list); 11228 11229 list_del_init(&napi->poll_list); 11230 if (napi->poll == process_backlog) 11231 napi->state = 0; 11232 else 11233 ____napi_schedule(sd, napi); 11234 } 11235 11236 raise_softirq_irqoff(NET_TX_SOFTIRQ); 11237 local_irq_enable(); 11238 11239 #ifdef CONFIG_RPS 11240 remsd = oldsd->rps_ipi_list; 11241 oldsd->rps_ipi_list = NULL; 11242 #endif 11243 /* send out pending IPI's on offline CPU */ 11244 net_rps_send_ipi(remsd); 11245 11246 /* Process offline CPU's input_pkt_queue */ 11247 while ((skb = __skb_dequeue(&oldsd->process_queue))) { 11248 netif_rx_ni(skb); 11249 input_queue_head_incr(oldsd); 11250 } 11251 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { 11252 netif_rx_ni(skb); 11253 input_queue_head_incr(oldsd); 11254 } 11255 11256 return 0; 11257 } 11258 11259 /** 11260 * netdev_increment_features - increment feature set by one 11261 * @all: current feature set 11262 * @one: new feature set 11263 * @mask: mask feature set 11264 * 11265 * Computes a new feature set after adding a device with feature set 11266 * @one to the master device with current feature set @all. Will not 11267 * enable anything that is off in @mask. Returns the new feature set. 11268 */ 11269 netdev_features_t netdev_increment_features(netdev_features_t all, 11270 netdev_features_t one, netdev_features_t mask) 11271 { 11272 if (mask & NETIF_F_HW_CSUM) 11273 mask |= NETIF_F_CSUM_MASK; 11274 mask |= NETIF_F_VLAN_CHALLENGED; 11275 11276 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; 11277 all &= one | ~NETIF_F_ALL_FOR_ALL; 11278 11279 /* If one device supports hw checksumming, set for all. */ 11280 if (all & NETIF_F_HW_CSUM) 11281 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); 11282 11283 return all; 11284 } 11285 EXPORT_SYMBOL(netdev_increment_features); 11286 11287 static struct hlist_head * __net_init netdev_create_hash(void) 11288 { 11289 int i; 11290 struct hlist_head *hash; 11291 11292 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL); 11293 if (hash != NULL) 11294 for (i = 0; i < NETDEV_HASHENTRIES; i++) 11295 INIT_HLIST_HEAD(&hash[i]); 11296 11297 return hash; 11298 } 11299 11300 /* Initialize per network namespace state */ 11301 static int __net_init netdev_init(struct net *net) 11302 { 11303 BUILD_BUG_ON(GRO_HASH_BUCKETS > 11304 8 * sizeof_field(struct napi_struct, gro_bitmask)); 11305 11306 if (net != &init_net) 11307 INIT_LIST_HEAD(&net->dev_base_head); 11308 11309 net->dev_name_head = netdev_create_hash(); 11310 if (net->dev_name_head == NULL) 11311 goto err_name; 11312 11313 net->dev_index_head = netdev_create_hash(); 11314 if (net->dev_index_head == NULL) 11315 goto err_idx; 11316 11317 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain); 11318 11319 return 0; 11320 11321 err_idx: 11322 kfree(net->dev_name_head); 11323 err_name: 11324 return -ENOMEM; 11325 } 11326 11327 /** 11328 * netdev_drivername - network driver for the device 11329 * @dev: network device 11330 * 11331 * Determine network driver for device. 11332 */ 11333 const char *netdev_drivername(const struct net_device *dev) 11334 { 11335 const struct device_driver *driver; 11336 const struct device *parent; 11337 const char *empty = ""; 11338 11339 parent = dev->dev.parent; 11340 if (!parent) 11341 return empty; 11342 11343 driver = parent->driver; 11344 if (driver && driver->name) 11345 return driver->name; 11346 return empty; 11347 } 11348 11349 static void __netdev_printk(const char *level, const struct net_device *dev, 11350 struct va_format *vaf) 11351 { 11352 if (dev && dev->dev.parent) { 11353 dev_printk_emit(level[1] - '0', 11354 dev->dev.parent, 11355 "%s %s %s%s: %pV", 11356 dev_driver_string(dev->dev.parent), 11357 dev_name(dev->dev.parent), 11358 netdev_name(dev), netdev_reg_state(dev), 11359 vaf); 11360 } else if (dev) { 11361 printk("%s%s%s: %pV", 11362 level, netdev_name(dev), netdev_reg_state(dev), vaf); 11363 } else { 11364 printk("%s(NULL net_device): %pV", level, vaf); 11365 } 11366 } 11367 11368 void netdev_printk(const char *level, const struct net_device *dev, 11369 const char *format, ...) 11370 { 11371 struct va_format vaf; 11372 va_list args; 11373 11374 va_start(args, format); 11375 11376 vaf.fmt = format; 11377 vaf.va = &args; 11378 11379 __netdev_printk(level, dev, &vaf); 11380 11381 va_end(args); 11382 } 11383 EXPORT_SYMBOL(netdev_printk); 11384 11385 #define define_netdev_printk_level(func, level) \ 11386 void func(const struct net_device *dev, const char *fmt, ...) \ 11387 { \ 11388 struct va_format vaf; \ 11389 va_list args; \ 11390 \ 11391 va_start(args, fmt); \ 11392 \ 11393 vaf.fmt = fmt; \ 11394 vaf.va = &args; \ 11395 \ 11396 __netdev_printk(level, dev, &vaf); \ 11397 \ 11398 va_end(args); \ 11399 } \ 11400 EXPORT_SYMBOL(func); 11401 11402 define_netdev_printk_level(netdev_emerg, KERN_EMERG); 11403 define_netdev_printk_level(netdev_alert, KERN_ALERT); 11404 define_netdev_printk_level(netdev_crit, KERN_CRIT); 11405 define_netdev_printk_level(netdev_err, KERN_ERR); 11406 define_netdev_printk_level(netdev_warn, KERN_WARNING); 11407 define_netdev_printk_level(netdev_notice, KERN_NOTICE); 11408 define_netdev_printk_level(netdev_info, KERN_INFO); 11409 11410 static void __net_exit netdev_exit(struct net *net) 11411 { 11412 kfree(net->dev_name_head); 11413 kfree(net->dev_index_head); 11414 if (net != &init_net) 11415 WARN_ON_ONCE(!list_empty(&net->dev_base_head)); 11416 } 11417 11418 static struct pernet_operations __net_initdata netdev_net_ops = { 11419 .init = netdev_init, 11420 .exit = netdev_exit, 11421 }; 11422 11423 static void __net_exit default_device_exit(struct net *net) 11424 { 11425 struct net_device *dev, *aux; 11426 /* 11427 * Push all migratable network devices back to the 11428 * initial network namespace 11429 */ 11430 rtnl_lock(); 11431 for_each_netdev_safe(net, dev, aux) { 11432 int err; 11433 char fb_name[IFNAMSIZ]; 11434 11435 /* Ignore unmoveable devices (i.e. loopback) */ 11436 if (dev->features & NETIF_F_NETNS_LOCAL) 11437 continue; 11438 11439 /* Leave virtual devices for the generic cleanup */ 11440 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund) 11441 continue; 11442 11443 /* Push remaining network devices to init_net */ 11444 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); 11445 if (__dev_get_by_name(&init_net, fb_name)) 11446 snprintf(fb_name, IFNAMSIZ, "dev%%d"); 11447 err = dev_change_net_namespace(dev, &init_net, fb_name); 11448 if (err) { 11449 pr_emerg("%s: failed to move %s to init_net: %d\n", 11450 __func__, dev->name, err); 11451 BUG(); 11452 } 11453 } 11454 rtnl_unlock(); 11455 } 11456 11457 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list) 11458 { 11459 /* Return with the rtnl_lock held when there are no network 11460 * devices unregistering in any network namespace in net_list. 11461 */ 11462 struct net *net; 11463 bool unregistering; 11464 DEFINE_WAIT_FUNC(wait, woken_wake_function); 11465 11466 add_wait_queue(&netdev_unregistering_wq, &wait); 11467 for (;;) { 11468 unregistering = false; 11469 rtnl_lock(); 11470 list_for_each_entry(net, net_list, exit_list) { 11471 if (net->dev_unreg_count > 0) { 11472 unregistering = true; 11473 break; 11474 } 11475 } 11476 if (!unregistering) 11477 break; 11478 __rtnl_unlock(); 11479 11480 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); 11481 } 11482 remove_wait_queue(&netdev_unregistering_wq, &wait); 11483 } 11484 11485 static void __net_exit default_device_exit_batch(struct list_head *net_list) 11486 { 11487 /* At exit all network devices most be removed from a network 11488 * namespace. Do this in the reverse order of registration. 11489 * Do this across as many network namespaces as possible to 11490 * improve batching efficiency. 11491 */ 11492 struct net_device *dev; 11493 struct net *net; 11494 LIST_HEAD(dev_kill_list); 11495 11496 /* To prevent network device cleanup code from dereferencing 11497 * loopback devices or network devices that have been freed 11498 * wait here for all pending unregistrations to complete, 11499 * before unregistring the loopback device and allowing the 11500 * network namespace be freed. 11501 * 11502 * The netdev todo list containing all network devices 11503 * unregistrations that happen in default_device_exit_batch 11504 * will run in the rtnl_unlock() at the end of 11505 * default_device_exit_batch. 11506 */ 11507 rtnl_lock_unregistering(net_list); 11508 list_for_each_entry(net, net_list, exit_list) { 11509 for_each_netdev_reverse(net, dev) { 11510 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) 11511 dev->rtnl_link_ops->dellink(dev, &dev_kill_list); 11512 else 11513 unregister_netdevice_queue(dev, &dev_kill_list); 11514 } 11515 } 11516 unregister_netdevice_many(&dev_kill_list); 11517 rtnl_unlock(); 11518 } 11519 11520 static struct pernet_operations __net_initdata default_device_ops = { 11521 .exit = default_device_exit, 11522 .exit_batch = default_device_exit_batch, 11523 }; 11524 11525 /* 11526 * Initialize the DEV module. At boot time this walks the device list and 11527 * unhooks any devices that fail to initialise (normally hardware not 11528 * present) and leaves us with a valid list of present and active devices. 11529 * 11530 */ 11531 11532 /* 11533 * This is called single threaded during boot, so no need 11534 * to take the rtnl semaphore. 11535 */ 11536 static int __init net_dev_init(void) 11537 { 11538 int i, rc = -ENOMEM; 11539 11540 BUG_ON(!dev_boot_phase); 11541 11542 if (dev_proc_init()) 11543 goto out; 11544 11545 if (netdev_kobject_init()) 11546 goto out; 11547 11548 INIT_LIST_HEAD(&ptype_all); 11549 for (i = 0; i < PTYPE_HASH_SIZE; i++) 11550 INIT_LIST_HEAD(&ptype_base[i]); 11551 11552 INIT_LIST_HEAD(&offload_base); 11553 11554 if (register_pernet_subsys(&netdev_net_ops)) 11555 goto out; 11556 11557 /* 11558 * Initialise the packet receive queues. 11559 */ 11560 11561 for_each_possible_cpu(i) { 11562 struct work_struct *flush = per_cpu_ptr(&flush_works, i); 11563 struct softnet_data *sd = &per_cpu(softnet_data, i); 11564 11565 INIT_WORK(flush, flush_backlog); 11566 11567 skb_queue_head_init(&sd->input_pkt_queue); 11568 skb_queue_head_init(&sd->process_queue); 11569 #ifdef CONFIG_XFRM_OFFLOAD 11570 skb_queue_head_init(&sd->xfrm_backlog); 11571 #endif 11572 INIT_LIST_HEAD(&sd->poll_list); 11573 sd->output_queue_tailp = &sd->output_queue; 11574 #ifdef CONFIG_RPS 11575 INIT_CSD(&sd->csd, rps_trigger_softirq, sd); 11576 sd->cpu = i; 11577 #endif 11578 11579 init_gro_hash(&sd->backlog); 11580 sd->backlog.poll = process_backlog; 11581 sd->backlog.weight = weight_p; 11582 } 11583 11584 dev_boot_phase = 0; 11585 11586 /* The loopback device is special if any other network devices 11587 * is present in a network namespace the loopback device must 11588 * be present. Since we now dynamically allocate and free the 11589 * loopback device ensure this invariant is maintained by 11590 * keeping the loopback device as the first device on the 11591 * list of network devices. Ensuring the loopback devices 11592 * is the first device that appears and the last network device 11593 * that disappears. 11594 */ 11595 if (register_pernet_device(&loopback_net_ops)) 11596 goto out; 11597 11598 if (register_pernet_device(&default_device_ops)) 11599 goto out; 11600 11601 open_softirq(NET_TX_SOFTIRQ, net_tx_action); 11602 open_softirq(NET_RX_SOFTIRQ, net_rx_action); 11603 11604 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", 11605 NULL, dev_cpu_dead); 11606 WARN_ON(rc < 0); 11607 rc = 0; 11608 out: 11609 return rc; 11610 } 11611 11612 subsys_initcall(net_dev_init); 11613