1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * NET3 Protocol independent device support routines. 4 * 5 * Derived from the non IP parts of dev.c 1.0.19 6 * Authors: Ross Biro 7 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 8 * Mark Evans, <evansmp@uhura.aston.ac.uk> 9 * 10 * Additional Authors: 11 * Florian la Roche <rzsfl@rz.uni-sb.de> 12 * Alan Cox <gw4pts@gw4pts.ampr.org> 13 * David Hinds <dahinds@users.sourceforge.net> 14 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru> 15 * Adam Sulmicki <adam@cfar.umd.edu> 16 * Pekka Riikonen <priikone@poesidon.pspt.fi> 17 * 18 * Changes: 19 * D.J. Barrow : Fixed bug where dev->refcnt gets set 20 * to 2 if register_netdev gets called 21 * before net_dev_init & also removed a 22 * few lines of code in the process. 23 * Alan Cox : device private ioctl copies fields back. 24 * Alan Cox : Transmit queue code does relevant 25 * stunts to keep the queue safe. 26 * Alan Cox : Fixed double lock. 27 * Alan Cox : Fixed promisc NULL pointer trap 28 * ???????? : Support the full private ioctl range 29 * Alan Cox : Moved ioctl permission check into 30 * drivers 31 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI 32 * Alan Cox : 100 backlog just doesn't cut it when 33 * you start doing multicast video 8) 34 * Alan Cox : Rewrote net_bh and list manager. 35 * Alan Cox : Fix ETH_P_ALL echoback lengths. 36 * Alan Cox : Took out transmit every packet pass 37 * Saved a few bytes in the ioctl handler 38 * Alan Cox : Network driver sets packet type before 39 * calling netif_rx. Saves a function 40 * call a packet. 41 * Alan Cox : Hashed net_bh() 42 * Richard Kooijman: Timestamp fixes. 43 * Alan Cox : Wrong field in SIOCGIFDSTADDR 44 * Alan Cox : Device lock protection. 45 * Alan Cox : Fixed nasty side effect of device close 46 * changes. 47 * Rudi Cilibrasi : Pass the right thing to 48 * set_mac_address() 49 * Dave Miller : 32bit quantity for the device lock to 50 * make it work out on a Sparc. 51 * Bjorn Ekwall : Added KERNELD hack. 52 * Alan Cox : Cleaned up the backlog initialise. 53 * Craig Metz : SIOCGIFCONF fix if space for under 54 * 1 device. 55 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there 56 * is no device open function. 57 * Andi Kleen : Fix error reporting for SIOCGIFCONF 58 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF 59 * Cyrus Durgin : Cleaned for KMOD 60 * Adam Sulmicki : Bug Fix : Network Device Unload 61 * A network device unload needs to purge 62 * the backlog queue. 63 * Paul Rusty Russell : SIOCSIFNAME 64 * Pekka Riikonen : Netdev boot-time settings code 65 * Andrew Morton : Make unregister_netdevice wait 66 * indefinitely on dev->refcnt 67 * J Hadi Salim : - Backlog queue sampling 68 * - netif_rx() feedback 69 */ 70 71 #include <linux/uaccess.h> 72 #include <linux/bitops.h> 73 #include <linux/capability.h> 74 #include <linux/cpu.h> 75 #include <linux/types.h> 76 #include <linux/kernel.h> 77 #include <linux/hash.h> 78 #include <linux/slab.h> 79 #include <linux/sched.h> 80 #include <linux/sched/mm.h> 81 #include <linux/mutex.h> 82 #include <linux/rwsem.h> 83 #include <linux/string.h> 84 #include <linux/mm.h> 85 #include <linux/socket.h> 86 #include <linux/sockios.h> 87 #include <linux/errno.h> 88 #include <linux/interrupt.h> 89 #include <linux/if_ether.h> 90 #include <linux/netdevice.h> 91 #include <linux/etherdevice.h> 92 #include <linux/ethtool.h> 93 #include <linux/skbuff.h> 94 #include <linux/kthread.h> 95 #include <linux/bpf.h> 96 #include <linux/bpf_trace.h> 97 #include <net/net_namespace.h> 98 #include <net/sock.h> 99 #include <net/busy_poll.h> 100 #include <linux/rtnetlink.h> 101 #include <linux/stat.h> 102 #include <net/dsa.h> 103 #include <net/dst.h> 104 #include <net/dst_metadata.h> 105 #include <net/gro.h> 106 #include <net/pkt_sched.h> 107 #include <net/pkt_cls.h> 108 #include <net/checksum.h> 109 #include <net/xfrm.h> 110 #include <linux/highmem.h> 111 #include <linux/init.h> 112 #include <linux/module.h> 113 #include <linux/netpoll.h> 114 #include <linux/rcupdate.h> 115 #include <linux/delay.h> 116 #include <net/iw_handler.h> 117 #include <asm/current.h> 118 #include <linux/audit.h> 119 #include <linux/dmaengine.h> 120 #include <linux/err.h> 121 #include <linux/ctype.h> 122 #include <linux/if_arp.h> 123 #include <linux/if_vlan.h> 124 #include <linux/ip.h> 125 #include <net/ip.h> 126 #include <net/mpls.h> 127 #include <linux/ipv6.h> 128 #include <linux/in.h> 129 #include <linux/jhash.h> 130 #include <linux/random.h> 131 #include <trace/events/napi.h> 132 #include <trace/events/net.h> 133 #include <trace/events/skb.h> 134 #include <linux/inetdevice.h> 135 #include <linux/cpu_rmap.h> 136 #include <linux/static_key.h> 137 #include <linux/hashtable.h> 138 #include <linux/vmalloc.h> 139 #include <linux/if_macvlan.h> 140 #include <linux/errqueue.h> 141 #include <linux/hrtimer.h> 142 #include <linux/netfilter_ingress.h> 143 #include <linux/crash_dump.h> 144 #include <linux/sctp.h> 145 #include <net/udp_tunnel.h> 146 #include <linux/net_namespace.h> 147 #include <linux/indirect_call_wrapper.h> 148 #include <net/devlink.h> 149 #include <linux/pm_runtime.h> 150 #include <linux/prandom.h> 151 152 #include "net-sysfs.h" 153 154 #define MAX_GRO_SKBS 8 155 156 /* This should be increased if a protocol with a bigger head is added. */ 157 #define GRO_MAX_HEAD (MAX_HEADER + 128) 158 159 static DEFINE_SPINLOCK(ptype_lock); 160 static DEFINE_SPINLOCK(offload_lock); 161 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; 162 struct list_head ptype_all __read_mostly; /* Taps */ 163 static struct list_head offload_base __read_mostly; 164 165 static int netif_rx_internal(struct sk_buff *skb); 166 static int call_netdevice_notifiers_info(unsigned long val, 167 struct netdev_notifier_info *info); 168 static int call_netdevice_notifiers_extack(unsigned long val, 169 struct net_device *dev, 170 struct netlink_ext_ack *extack); 171 static struct napi_struct *napi_by_id(unsigned int napi_id); 172 173 /* 174 * The @dev_base_head list is protected by @dev_base_lock and the rtnl 175 * semaphore. 176 * 177 * Pure readers hold dev_base_lock for reading, or rcu_read_lock() 178 * 179 * Writers must hold the rtnl semaphore while they loop through the 180 * dev_base_head list, and hold dev_base_lock for writing when they do the 181 * actual updates. This allows pure readers to access the list even 182 * while a writer is preparing to update it. 183 * 184 * To put it another way, dev_base_lock is held for writing only to 185 * protect against pure readers; the rtnl semaphore provides the 186 * protection against other writers. 187 * 188 * See, for example usages, register_netdevice() and 189 * unregister_netdevice(), which must be called with the rtnl 190 * semaphore held. 191 */ 192 DEFINE_RWLOCK(dev_base_lock); 193 EXPORT_SYMBOL(dev_base_lock); 194 195 static DEFINE_MUTEX(ifalias_mutex); 196 197 /* protects napi_hash addition/deletion and napi_gen_id */ 198 static DEFINE_SPINLOCK(napi_hash_lock); 199 200 static unsigned int napi_gen_id = NR_CPUS; 201 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8); 202 203 static DECLARE_RWSEM(devnet_rename_sem); 204 205 static inline void dev_base_seq_inc(struct net *net) 206 { 207 while (++net->dev_base_seq == 0) 208 ; 209 } 210 211 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) 212 { 213 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ)); 214 215 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; 216 } 217 218 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) 219 { 220 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; 221 } 222 223 static inline void rps_lock(struct softnet_data *sd) 224 { 225 #ifdef CONFIG_RPS 226 spin_lock(&sd->input_pkt_queue.lock); 227 #endif 228 } 229 230 static inline void rps_unlock(struct softnet_data *sd) 231 { 232 #ifdef CONFIG_RPS 233 spin_unlock(&sd->input_pkt_queue.lock); 234 #endif 235 } 236 237 static struct netdev_name_node *netdev_name_node_alloc(struct net_device *dev, 238 const char *name) 239 { 240 struct netdev_name_node *name_node; 241 242 name_node = kmalloc(sizeof(*name_node), GFP_KERNEL); 243 if (!name_node) 244 return NULL; 245 INIT_HLIST_NODE(&name_node->hlist); 246 name_node->dev = dev; 247 name_node->name = name; 248 return name_node; 249 } 250 251 static struct netdev_name_node * 252 netdev_name_node_head_alloc(struct net_device *dev) 253 { 254 struct netdev_name_node *name_node; 255 256 name_node = netdev_name_node_alloc(dev, dev->name); 257 if (!name_node) 258 return NULL; 259 INIT_LIST_HEAD(&name_node->list); 260 return name_node; 261 } 262 263 static void netdev_name_node_free(struct netdev_name_node *name_node) 264 { 265 kfree(name_node); 266 } 267 268 static void netdev_name_node_add(struct net *net, 269 struct netdev_name_node *name_node) 270 { 271 hlist_add_head_rcu(&name_node->hlist, 272 dev_name_hash(net, name_node->name)); 273 } 274 275 static void netdev_name_node_del(struct netdev_name_node *name_node) 276 { 277 hlist_del_rcu(&name_node->hlist); 278 } 279 280 static struct netdev_name_node *netdev_name_node_lookup(struct net *net, 281 const char *name) 282 { 283 struct hlist_head *head = dev_name_hash(net, name); 284 struct netdev_name_node *name_node; 285 286 hlist_for_each_entry(name_node, head, hlist) 287 if (!strcmp(name_node->name, name)) 288 return name_node; 289 return NULL; 290 } 291 292 static struct netdev_name_node *netdev_name_node_lookup_rcu(struct net *net, 293 const char *name) 294 { 295 struct hlist_head *head = dev_name_hash(net, name); 296 struct netdev_name_node *name_node; 297 298 hlist_for_each_entry_rcu(name_node, head, hlist) 299 if (!strcmp(name_node->name, name)) 300 return name_node; 301 return NULL; 302 } 303 304 int netdev_name_node_alt_create(struct net_device *dev, const char *name) 305 { 306 struct netdev_name_node *name_node; 307 struct net *net = dev_net(dev); 308 309 name_node = netdev_name_node_lookup(net, name); 310 if (name_node) 311 return -EEXIST; 312 name_node = netdev_name_node_alloc(dev, name); 313 if (!name_node) 314 return -ENOMEM; 315 netdev_name_node_add(net, name_node); 316 /* The node that holds dev->name acts as a head of per-device list. */ 317 list_add_tail(&name_node->list, &dev->name_node->list); 318 319 return 0; 320 } 321 EXPORT_SYMBOL(netdev_name_node_alt_create); 322 323 static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node) 324 { 325 list_del(&name_node->list); 326 netdev_name_node_del(name_node); 327 kfree(name_node->name); 328 netdev_name_node_free(name_node); 329 } 330 331 int netdev_name_node_alt_destroy(struct net_device *dev, const char *name) 332 { 333 struct netdev_name_node *name_node; 334 struct net *net = dev_net(dev); 335 336 name_node = netdev_name_node_lookup(net, name); 337 if (!name_node) 338 return -ENOENT; 339 /* lookup might have found our primary name or a name belonging 340 * to another device. 341 */ 342 if (name_node == dev->name_node || name_node->dev != dev) 343 return -EINVAL; 344 345 __netdev_name_node_alt_destroy(name_node); 346 347 return 0; 348 } 349 EXPORT_SYMBOL(netdev_name_node_alt_destroy); 350 351 static void netdev_name_node_alt_flush(struct net_device *dev) 352 { 353 struct netdev_name_node *name_node, *tmp; 354 355 list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list) 356 __netdev_name_node_alt_destroy(name_node); 357 } 358 359 /* Device list insertion */ 360 static void list_netdevice(struct net_device *dev) 361 { 362 struct net *net = dev_net(dev); 363 364 ASSERT_RTNL(); 365 366 write_lock_bh(&dev_base_lock); 367 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); 368 netdev_name_node_add(net, dev->name_node); 369 hlist_add_head_rcu(&dev->index_hlist, 370 dev_index_hash(net, dev->ifindex)); 371 write_unlock_bh(&dev_base_lock); 372 373 dev_base_seq_inc(net); 374 } 375 376 /* Device list removal 377 * caller must respect a RCU grace period before freeing/reusing dev 378 */ 379 static void unlist_netdevice(struct net_device *dev) 380 { 381 ASSERT_RTNL(); 382 383 /* Unlink dev from the device chain */ 384 write_lock_bh(&dev_base_lock); 385 list_del_rcu(&dev->dev_list); 386 netdev_name_node_del(dev->name_node); 387 hlist_del_rcu(&dev->index_hlist); 388 write_unlock_bh(&dev_base_lock); 389 390 dev_base_seq_inc(dev_net(dev)); 391 } 392 393 /* 394 * Our notifier list 395 */ 396 397 static RAW_NOTIFIER_HEAD(netdev_chain); 398 399 /* 400 * Device drivers call our routines to queue packets here. We empty the 401 * queue in the local softnet handler. 402 */ 403 404 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); 405 EXPORT_PER_CPU_SYMBOL(softnet_data); 406 407 #ifdef CONFIG_LOCKDEP 408 /* 409 * register_netdevice() inits txq->_xmit_lock and sets lockdep class 410 * according to dev->type 411 */ 412 static const unsigned short netdev_lock_type[] = { 413 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, 414 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, 415 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, 416 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, 417 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, 418 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, 419 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, 420 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, 421 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, 422 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, 423 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, 424 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, 425 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, 426 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, 427 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; 428 429 static const char *const netdev_lock_name[] = { 430 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", 431 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", 432 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", 433 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", 434 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", 435 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", 436 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", 437 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", 438 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", 439 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", 440 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", 441 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", 442 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", 443 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", 444 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; 445 446 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; 447 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; 448 449 static inline unsigned short netdev_lock_pos(unsigned short dev_type) 450 { 451 int i; 452 453 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) 454 if (netdev_lock_type[i] == dev_type) 455 return i; 456 /* the last key is used by default */ 457 return ARRAY_SIZE(netdev_lock_type) - 1; 458 } 459 460 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 461 unsigned short dev_type) 462 { 463 int i; 464 465 i = netdev_lock_pos(dev_type); 466 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], 467 netdev_lock_name[i]); 468 } 469 470 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 471 { 472 int i; 473 474 i = netdev_lock_pos(dev->type); 475 lockdep_set_class_and_name(&dev->addr_list_lock, 476 &netdev_addr_lock_key[i], 477 netdev_lock_name[i]); 478 } 479 #else 480 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 481 unsigned short dev_type) 482 { 483 } 484 485 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 486 { 487 } 488 #endif 489 490 /******************************************************************************* 491 * 492 * Protocol management and registration routines 493 * 494 *******************************************************************************/ 495 496 497 /* 498 * Add a protocol ID to the list. Now that the input handler is 499 * smarter we can dispense with all the messy stuff that used to be 500 * here. 501 * 502 * BEWARE!!! Protocol handlers, mangling input packets, 503 * MUST BE last in hash buckets and checking protocol handlers 504 * MUST start from promiscuous ptype_all chain in net_bh. 505 * It is true now, do not change it. 506 * Explanation follows: if protocol handler, mangling packet, will 507 * be the first on list, it is not able to sense, that packet 508 * is cloned and should be copied-on-write, so that it will 509 * change it and subsequent readers will get broken packet. 510 * --ANK (980803) 511 */ 512 513 static inline struct list_head *ptype_head(const struct packet_type *pt) 514 { 515 if (pt->type == htons(ETH_P_ALL)) 516 return pt->dev ? &pt->dev->ptype_all : &ptype_all; 517 else 518 return pt->dev ? &pt->dev->ptype_specific : 519 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; 520 } 521 522 /** 523 * dev_add_pack - add packet handler 524 * @pt: packet type declaration 525 * 526 * Add a protocol handler to the networking stack. The passed &packet_type 527 * is linked into kernel lists and may not be freed until it has been 528 * removed from the kernel lists. 529 * 530 * This call does not sleep therefore it can not 531 * guarantee all CPU's that are in middle of receiving packets 532 * will see the new packet type (until the next received packet). 533 */ 534 535 void dev_add_pack(struct packet_type *pt) 536 { 537 struct list_head *head = ptype_head(pt); 538 539 spin_lock(&ptype_lock); 540 list_add_rcu(&pt->list, head); 541 spin_unlock(&ptype_lock); 542 } 543 EXPORT_SYMBOL(dev_add_pack); 544 545 /** 546 * __dev_remove_pack - remove packet handler 547 * @pt: packet type declaration 548 * 549 * Remove a protocol handler that was previously added to the kernel 550 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 551 * from the kernel lists and can be freed or reused once this function 552 * returns. 553 * 554 * The packet type might still be in use by receivers 555 * and must not be freed until after all the CPU's have gone 556 * through a quiescent state. 557 */ 558 void __dev_remove_pack(struct packet_type *pt) 559 { 560 struct list_head *head = ptype_head(pt); 561 struct packet_type *pt1; 562 563 spin_lock(&ptype_lock); 564 565 list_for_each_entry(pt1, head, list) { 566 if (pt == pt1) { 567 list_del_rcu(&pt->list); 568 goto out; 569 } 570 } 571 572 pr_warn("dev_remove_pack: %p not found\n", pt); 573 out: 574 spin_unlock(&ptype_lock); 575 } 576 EXPORT_SYMBOL(__dev_remove_pack); 577 578 /** 579 * dev_remove_pack - remove packet handler 580 * @pt: packet type declaration 581 * 582 * Remove a protocol handler that was previously added to the kernel 583 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 584 * from the kernel lists and can be freed or reused once this function 585 * returns. 586 * 587 * This call sleeps to guarantee that no CPU is looking at the packet 588 * type after return. 589 */ 590 void dev_remove_pack(struct packet_type *pt) 591 { 592 __dev_remove_pack(pt); 593 594 synchronize_net(); 595 } 596 EXPORT_SYMBOL(dev_remove_pack); 597 598 599 /** 600 * dev_add_offload - register offload handlers 601 * @po: protocol offload declaration 602 * 603 * Add protocol offload handlers to the networking stack. The passed 604 * &proto_offload is linked into kernel lists and may not be freed until 605 * it has been removed from the kernel lists. 606 * 607 * This call does not sleep therefore it can not 608 * guarantee all CPU's that are in middle of receiving packets 609 * will see the new offload handlers (until the next received packet). 610 */ 611 void dev_add_offload(struct packet_offload *po) 612 { 613 struct packet_offload *elem; 614 615 spin_lock(&offload_lock); 616 list_for_each_entry(elem, &offload_base, list) { 617 if (po->priority < elem->priority) 618 break; 619 } 620 list_add_rcu(&po->list, elem->list.prev); 621 spin_unlock(&offload_lock); 622 } 623 EXPORT_SYMBOL(dev_add_offload); 624 625 /** 626 * __dev_remove_offload - remove offload handler 627 * @po: packet offload declaration 628 * 629 * Remove a protocol offload handler that was previously added to the 630 * kernel offload handlers by dev_add_offload(). The passed &offload_type 631 * is removed from the kernel lists and can be freed or reused once this 632 * function returns. 633 * 634 * The packet type might still be in use by receivers 635 * and must not be freed until after all the CPU's have gone 636 * through a quiescent state. 637 */ 638 static void __dev_remove_offload(struct packet_offload *po) 639 { 640 struct list_head *head = &offload_base; 641 struct packet_offload *po1; 642 643 spin_lock(&offload_lock); 644 645 list_for_each_entry(po1, head, list) { 646 if (po == po1) { 647 list_del_rcu(&po->list); 648 goto out; 649 } 650 } 651 652 pr_warn("dev_remove_offload: %p not found\n", po); 653 out: 654 spin_unlock(&offload_lock); 655 } 656 657 /** 658 * dev_remove_offload - remove packet offload handler 659 * @po: packet offload declaration 660 * 661 * Remove a packet offload handler that was previously added to the kernel 662 * offload handlers by dev_add_offload(). The passed &offload_type is 663 * removed from the kernel lists and can be freed or reused once this 664 * function returns. 665 * 666 * This call sleeps to guarantee that no CPU is looking at the packet 667 * type after return. 668 */ 669 void dev_remove_offload(struct packet_offload *po) 670 { 671 __dev_remove_offload(po); 672 673 synchronize_net(); 674 } 675 EXPORT_SYMBOL(dev_remove_offload); 676 677 /****************************************************************************** 678 * 679 * Device Boot-time Settings Routines 680 * 681 ******************************************************************************/ 682 683 /* Boot time configuration table */ 684 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX]; 685 686 /** 687 * netdev_boot_setup_add - add new setup entry 688 * @name: name of the device 689 * @map: configured settings for the device 690 * 691 * Adds new setup entry to the dev_boot_setup list. The function 692 * returns 0 on error and 1 on success. This is a generic routine to 693 * all netdevices. 694 */ 695 static int netdev_boot_setup_add(char *name, struct ifmap *map) 696 { 697 struct netdev_boot_setup *s; 698 int i; 699 700 s = dev_boot_setup; 701 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { 702 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') { 703 memset(s[i].name, 0, sizeof(s[i].name)); 704 strlcpy(s[i].name, name, IFNAMSIZ); 705 memcpy(&s[i].map, map, sizeof(s[i].map)); 706 break; 707 } 708 } 709 710 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1; 711 } 712 713 /** 714 * netdev_boot_setup_check - check boot time settings 715 * @dev: the netdevice 716 * 717 * Check boot time settings for the device. 718 * The found settings are set for the device to be used 719 * later in the device probing. 720 * Returns 0 if no settings found, 1 if they are. 721 */ 722 int netdev_boot_setup_check(struct net_device *dev) 723 { 724 struct netdev_boot_setup *s = dev_boot_setup; 725 int i; 726 727 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { 728 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' && 729 !strcmp(dev->name, s[i].name)) { 730 dev->irq = s[i].map.irq; 731 dev->base_addr = s[i].map.base_addr; 732 dev->mem_start = s[i].map.mem_start; 733 dev->mem_end = s[i].map.mem_end; 734 return 1; 735 } 736 } 737 return 0; 738 } 739 EXPORT_SYMBOL(netdev_boot_setup_check); 740 741 742 /** 743 * netdev_boot_base - get address from boot time settings 744 * @prefix: prefix for network device 745 * @unit: id for network device 746 * 747 * Check boot time settings for the base address of device. 748 * The found settings are set for the device to be used 749 * later in the device probing. 750 * Returns 0 if no settings found. 751 */ 752 unsigned long netdev_boot_base(const char *prefix, int unit) 753 { 754 const struct netdev_boot_setup *s = dev_boot_setup; 755 char name[IFNAMSIZ]; 756 int i; 757 758 sprintf(name, "%s%d", prefix, unit); 759 760 /* 761 * If device already registered then return base of 1 762 * to indicate not to probe for this interface 763 */ 764 if (__dev_get_by_name(&init_net, name)) 765 return 1; 766 767 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) 768 if (!strcmp(name, s[i].name)) 769 return s[i].map.base_addr; 770 return 0; 771 } 772 773 /* 774 * Saves at boot time configured settings for any netdevice. 775 */ 776 int __init netdev_boot_setup(char *str) 777 { 778 int ints[5]; 779 struct ifmap map; 780 781 str = get_options(str, ARRAY_SIZE(ints), ints); 782 if (!str || !*str) 783 return 0; 784 785 /* Save settings */ 786 memset(&map, 0, sizeof(map)); 787 if (ints[0] > 0) 788 map.irq = ints[1]; 789 if (ints[0] > 1) 790 map.base_addr = ints[2]; 791 if (ints[0] > 2) 792 map.mem_start = ints[3]; 793 if (ints[0] > 3) 794 map.mem_end = ints[4]; 795 796 /* Add new entry to the list */ 797 return netdev_boot_setup_add(str, &map); 798 } 799 800 __setup("netdev=", netdev_boot_setup); 801 802 /******************************************************************************* 803 * 804 * Device Interface Subroutines 805 * 806 *******************************************************************************/ 807 808 /** 809 * dev_get_iflink - get 'iflink' value of a interface 810 * @dev: targeted interface 811 * 812 * Indicates the ifindex the interface is linked to. 813 * Physical interfaces have the same 'ifindex' and 'iflink' values. 814 */ 815 816 int dev_get_iflink(const struct net_device *dev) 817 { 818 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink) 819 return dev->netdev_ops->ndo_get_iflink(dev); 820 821 return dev->ifindex; 822 } 823 EXPORT_SYMBOL(dev_get_iflink); 824 825 /** 826 * dev_fill_metadata_dst - Retrieve tunnel egress information. 827 * @dev: targeted interface 828 * @skb: The packet. 829 * 830 * For better visibility of tunnel traffic OVS needs to retrieve 831 * egress tunnel information for a packet. Following API allows 832 * user to get this info. 833 */ 834 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) 835 { 836 struct ip_tunnel_info *info; 837 838 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst) 839 return -EINVAL; 840 841 info = skb_tunnel_info_unclone(skb); 842 if (!info) 843 return -ENOMEM; 844 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX))) 845 return -EINVAL; 846 847 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb); 848 } 849 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst); 850 851 static struct net_device_path *dev_fwd_path(struct net_device_path_stack *stack) 852 { 853 int k = stack->num_paths++; 854 855 if (WARN_ON_ONCE(k >= NET_DEVICE_PATH_STACK_MAX)) 856 return NULL; 857 858 return &stack->path[k]; 859 } 860 861 int dev_fill_forward_path(const struct net_device *dev, const u8 *daddr, 862 struct net_device_path_stack *stack) 863 { 864 const struct net_device *last_dev; 865 struct net_device_path_ctx ctx = { 866 .dev = dev, 867 .daddr = daddr, 868 }; 869 struct net_device_path *path; 870 int ret = 0; 871 872 stack->num_paths = 0; 873 while (ctx.dev && ctx.dev->netdev_ops->ndo_fill_forward_path) { 874 last_dev = ctx.dev; 875 path = dev_fwd_path(stack); 876 if (!path) 877 return -1; 878 879 memset(path, 0, sizeof(struct net_device_path)); 880 ret = ctx.dev->netdev_ops->ndo_fill_forward_path(&ctx, path); 881 if (ret < 0) 882 return -1; 883 884 if (WARN_ON_ONCE(last_dev == ctx.dev)) 885 return -1; 886 } 887 path = dev_fwd_path(stack); 888 if (!path) 889 return -1; 890 path->type = DEV_PATH_ETHERNET; 891 path->dev = ctx.dev; 892 893 return ret; 894 } 895 EXPORT_SYMBOL_GPL(dev_fill_forward_path); 896 897 /** 898 * __dev_get_by_name - find a device by its name 899 * @net: the applicable net namespace 900 * @name: name to find 901 * 902 * Find an interface by name. Must be called under RTNL semaphore 903 * or @dev_base_lock. If the name is found a pointer to the device 904 * is returned. If the name is not found then %NULL is returned. The 905 * reference counters are not incremented so the caller must be 906 * careful with locks. 907 */ 908 909 struct net_device *__dev_get_by_name(struct net *net, const char *name) 910 { 911 struct netdev_name_node *node_name; 912 913 node_name = netdev_name_node_lookup(net, name); 914 return node_name ? node_name->dev : NULL; 915 } 916 EXPORT_SYMBOL(__dev_get_by_name); 917 918 /** 919 * dev_get_by_name_rcu - find a device by its name 920 * @net: the applicable net namespace 921 * @name: name to find 922 * 923 * Find an interface by name. 924 * If the name is found a pointer to the device is returned. 925 * If the name is not found then %NULL is returned. 926 * The reference counters are not incremented so the caller must be 927 * careful with locks. The caller must hold RCU lock. 928 */ 929 930 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) 931 { 932 struct netdev_name_node *node_name; 933 934 node_name = netdev_name_node_lookup_rcu(net, name); 935 return node_name ? node_name->dev : NULL; 936 } 937 EXPORT_SYMBOL(dev_get_by_name_rcu); 938 939 /** 940 * dev_get_by_name - find a device by its name 941 * @net: the applicable net namespace 942 * @name: name to find 943 * 944 * Find an interface by name. This can be called from any 945 * context and does its own locking. The returned handle has 946 * the usage count incremented and the caller must use dev_put() to 947 * release it when it is no longer needed. %NULL is returned if no 948 * matching device is found. 949 */ 950 951 struct net_device *dev_get_by_name(struct net *net, const char *name) 952 { 953 struct net_device *dev; 954 955 rcu_read_lock(); 956 dev = dev_get_by_name_rcu(net, name); 957 if (dev) 958 dev_hold(dev); 959 rcu_read_unlock(); 960 return dev; 961 } 962 EXPORT_SYMBOL(dev_get_by_name); 963 964 /** 965 * __dev_get_by_index - find a device by its ifindex 966 * @net: the applicable net namespace 967 * @ifindex: index of device 968 * 969 * Search for an interface by index. Returns %NULL if the device 970 * is not found or a pointer to the device. The device has not 971 * had its reference counter increased so the caller must be careful 972 * about locking. The caller must hold either the RTNL semaphore 973 * or @dev_base_lock. 974 */ 975 976 struct net_device *__dev_get_by_index(struct net *net, int ifindex) 977 { 978 struct net_device *dev; 979 struct hlist_head *head = dev_index_hash(net, ifindex); 980 981 hlist_for_each_entry(dev, head, index_hlist) 982 if (dev->ifindex == ifindex) 983 return dev; 984 985 return NULL; 986 } 987 EXPORT_SYMBOL(__dev_get_by_index); 988 989 /** 990 * dev_get_by_index_rcu - find a device by its ifindex 991 * @net: the applicable net namespace 992 * @ifindex: index of device 993 * 994 * Search for an interface by index. Returns %NULL if the device 995 * is not found or a pointer to the device. The device has not 996 * had its reference counter increased so the caller must be careful 997 * about locking. The caller must hold RCU lock. 998 */ 999 1000 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) 1001 { 1002 struct net_device *dev; 1003 struct hlist_head *head = dev_index_hash(net, ifindex); 1004 1005 hlist_for_each_entry_rcu(dev, head, index_hlist) 1006 if (dev->ifindex == ifindex) 1007 return dev; 1008 1009 return NULL; 1010 } 1011 EXPORT_SYMBOL(dev_get_by_index_rcu); 1012 1013 1014 /** 1015 * dev_get_by_index - find a device by its ifindex 1016 * @net: the applicable net namespace 1017 * @ifindex: index of device 1018 * 1019 * Search for an interface by index. Returns NULL if the device 1020 * is not found or a pointer to the device. The device returned has 1021 * had a reference added and the pointer is safe until the user calls 1022 * dev_put to indicate they have finished with it. 1023 */ 1024 1025 struct net_device *dev_get_by_index(struct net *net, int ifindex) 1026 { 1027 struct net_device *dev; 1028 1029 rcu_read_lock(); 1030 dev = dev_get_by_index_rcu(net, ifindex); 1031 if (dev) 1032 dev_hold(dev); 1033 rcu_read_unlock(); 1034 return dev; 1035 } 1036 EXPORT_SYMBOL(dev_get_by_index); 1037 1038 /** 1039 * dev_get_by_napi_id - find a device by napi_id 1040 * @napi_id: ID of the NAPI struct 1041 * 1042 * Search for an interface by NAPI ID. Returns %NULL if the device 1043 * is not found or a pointer to the device. The device has not had 1044 * its reference counter increased so the caller must be careful 1045 * about locking. The caller must hold RCU lock. 1046 */ 1047 1048 struct net_device *dev_get_by_napi_id(unsigned int napi_id) 1049 { 1050 struct napi_struct *napi; 1051 1052 WARN_ON_ONCE(!rcu_read_lock_held()); 1053 1054 if (napi_id < MIN_NAPI_ID) 1055 return NULL; 1056 1057 napi = napi_by_id(napi_id); 1058 1059 return napi ? napi->dev : NULL; 1060 } 1061 EXPORT_SYMBOL(dev_get_by_napi_id); 1062 1063 /** 1064 * netdev_get_name - get a netdevice name, knowing its ifindex. 1065 * @net: network namespace 1066 * @name: a pointer to the buffer where the name will be stored. 1067 * @ifindex: the ifindex of the interface to get the name from. 1068 */ 1069 int netdev_get_name(struct net *net, char *name, int ifindex) 1070 { 1071 struct net_device *dev; 1072 int ret; 1073 1074 down_read(&devnet_rename_sem); 1075 rcu_read_lock(); 1076 1077 dev = dev_get_by_index_rcu(net, ifindex); 1078 if (!dev) { 1079 ret = -ENODEV; 1080 goto out; 1081 } 1082 1083 strcpy(name, dev->name); 1084 1085 ret = 0; 1086 out: 1087 rcu_read_unlock(); 1088 up_read(&devnet_rename_sem); 1089 return ret; 1090 } 1091 1092 /** 1093 * dev_getbyhwaddr_rcu - find a device by its hardware address 1094 * @net: the applicable net namespace 1095 * @type: media type of device 1096 * @ha: hardware address 1097 * 1098 * Search for an interface by MAC address. Returns NULL if the device 1099 * is not found or a pointer to the device. 1100 * The caller must hold RCU or RTNL. 1101 * The returned device has not had its ref count increased 1102 * and the caller must therefore be careful about locking 1103 * 1104 */ 1105 1106 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, 1107 const char *ha) 1108 { 1109 struct net_device *dev; 1110 1111 for_each_netdev_rcu(net, dev) 1112 if (dev->type == type && 1113 !memcmp(dev->dev_addr, ha, dev->addr_len)) 1114 return dev; 1115 1116 return NULL; 1117 } 1118 EXPORT_SYMBOL(dev_getbyhwaddr_rcu); 1119 1120 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) 1121 { 1122 struct net_device *dev, *ret = NULL; 1123 1124 rcu_read_lock(); 1125 for_each_netdev_rcu(net, dev) 1126 if (dev->type == type) { 1127 dev_hold(dev); 1128 ret = dev; 1129 break; 1130 } 1131 rcu_read_unlock(); 1132 return ret; 1133 } 1134 EXPORT_SYMBOL(dev_getfirstbyhwtype); 1135 1136 /** 1137 * __dev_get_by_flags - find any device with given flags 1138 * @net: the applicable net namespace 1139 * @if_flags: IFF_* values 1140 * @mask: bitmask of bits in if_flags to check 1141 * 1142 * Search for any interface with the given flags. Returns NULL if a device 1143 * is not found or a pointer to the device. Must be called inside 1144 * rtnl_lock(), and result refcount is unchanged. 1145 */ 1146 1147 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags, 1148 unsigned short mask) 1149 { 1150 struct net_device *dev, *ret; 1151 1152 ASSERT_RTNL(); 1153 1154 ret = NULL; 1155 for_each_netdev(net, dev) { 1156 if (((dev->flags ^ if_flags) & mask) == 0) { 1157 ret = dev; 1158 break; 1159 } 1160 } 1161 return ret; 1162 } 1163 EXPORT_SYMBOL(__dev_get_by_flags); 1164 1165 /** 1166 * dev_valid_name - check if name is okay for network device 1167 * @name: name string 1168 * 1169 * Network device names need to be valid file names to 1170 * allow sysfs to work. We also disallow any kind of 1171 * whitespace. 1172 */ 1173 bool dev_valid_name(const char *name) 1174 { 1175 if (*name == '\0') 1176 return false; 1177 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ) 1178 return false; 1179 if (!strcmp(name, ".") || !strcmp(name, "..")) 1180 return false; 1181 1182 while (*name) { 1183 if (*name == '/' || *name == ':' || isspace(*name)) 1184 return false; 1185 name++; 1186 } 1187 return true; 1188 } 1189 EXPORT_SYMBOL(dev_valid_name); 1190 1191 /** 1192 * __dev_alloc_name - allocate a name for a device 1193 * @net: network namespace to allocate the device name in 1194 * @name: name format string 1195 * @buf: scratch buffer and result name string 1196 * 1197 * Passed a format string - eg "lt%d" it will try and find a suitable 1198 * id. It scans list of devices to build up a free map, then chooses 1199 * the first empty slot. The caller must hold the dev_base or rtnl lock 1200 * while allocating the name and adding the device in order to avoid 1201 * duplicates. 1202 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1203 * Returns the number of the unit assigned or a negative errno code. 1204 */ 1205 1206 static int __dev_alloc_name(struct net *net, const char *name, char *buf) 1207 { 1208 int i = 0; 1209 const char *p; 1210 const int max_netdevices = 8*PAGE_SIZE; 1211 unsigned long *inuse; 1212 struct net_device *d; 1213 1214 if (!dev_valid_name(name)) 1215 return -EINVAL; 1216 1217 p = strchr(name, '%'); 1218 if (p) { 1219 /* 1220 * Verify the string as this thing may have come from 1221 * the user. There must be either one "%d" and no other "%" 1222 * characters. 1223 */ 1224 if (p[1] != 'd' || strchr(p + 2, '%')) 1225 return -EINVAL; 1226 1227 /* Use one page as a bit array of possible slots */ 1228 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); 1229 if (!inuse) 1230 return -ENOMEM; 1231 1232 for_each_netdev(net, d) { 1233 struct netdev_name_node *name_node; 1234 list_for_each_entry(name_node, &d->name_node->list, list) { 1235 if (!sscanf(name_node->name, name, &i)) 1236 continue; 1237 if (i < 0 || i >= max_netdevices) 1238 continue; 1239 1240 /* avoid cases where sscanf is not exact inverse of printf */ 1241 snprintf(buf, IFNAMSIZ, name, i); 1242 if (!strncmp(buf, name_node->name, IFNAMSIZ)) 1243 set_bit(i, inuse); 1244 } 1245 if (!sscanf(d->name, name, &i)) 1246 continue; 1247 if (i < 0 || i >= max_netdevices) 1248 continue; 1249 1250 /* avoid cases where sscanf is not exact inverse of printf */ 1251 snprintf(buf, IFNAMSIZ, name, i); 1252 if (!strncmp(buf, d->name, IFNAMSIZ)) 1253 set_bit(i, inuse); 1254 } 1255 1256 i = find_first_zero_bit(inuse, max_netdevices); 1257 free_page((unsigned long) inuse); 1258 } 1259 1260 snprintf(buf, IFNAMSIZ, name, i); 1261 if (!__dev_get_by_name(net, buf)) 1262 return i; 1263 1264 /* It is possible to run out of possible slots 1265 * when the name is long and there isn't enough space left 1266 * for the digits, or if all bits are used. 1267 */ 1268 return -ENFILE; 1269 } 1270 1271 static int dev_alloc_name_ns(struct net *net, 1272 struct net_device *dev, 1273 const char *name) 1274 { 1275 char buf[IFNAMSIZ]; 1276 int ret; 1277 1278 BUG_ON(!net); 1279 ret = __dev_alloc_name(net, name, buf); 1280 if (ret >= 0) 1281 strlcpy(dev->name, buf, IFNAMSIZ); 1282 return ret; 1283 } 1284 1285 /** 1286 * dev_alloc_name - allocate a name for a device 1287 * @dev: device 1288 * @name: name format string 1289 * 1290 * Passed a format string - eg "lt%d" it will try and find a suitable 1291 * id. It scans list of devices to build up a free map, then chooses 1292 * the first empty slot. The caller must hold the dev_base or rtnl lock 1293 * while allocating the name and adding the device in order to avoid 1294 * duplicates. 1295 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1296 * Returns the number of the unit assigned or a negative errno code. 1297 */ 1298 1299 int dev_alloc_name(struct net_device *dev, const char *name) 1300 { 1301 return dev_alloc_name_ns(dev_net(dev), dev, name); 1302 } 1303 EXPORT_SYMBOL(dev_alloc_name); 1304 1305 static int dev_get_valid_name(struct net *net, struct net_device *dev, 1306 const char *name) 1307 { 1308 BUG_ON(!net); 1309 1310 if (!dev_valid_name(name)) 1311 return -EINVAL; 1312 1313 if (strchr(name, '%')) 1314 return dev_alloc_name_ns(net, dev, name); 1315 else if (__dev_get_by_name(net, name)) 1316 return -EEXIST; 1317 else if (dev->name != name) 1318 strlcpy(dev->name, name, IFNAMSIZ); 1319 1320 return 0; 1321 } 1322 1323 /** 1324 * dev_change_name - change name of a device 1325 * @dev: device 1326 * @newname: name (or format string) must be at least IFNAMSIZ 1327 * 1328 * Change name of a device, can pass format strings "eth%d". 1329 * for wildcarding. 1330 */ 1331 int dev_change_name(struct net_device *dev, const char *newname) 1332 { 1333 unsigned char old_assign_type; 1334 char oldname[IFNAMSIZ]; 1335 int err = 0; 1336 int ret; 1337 struct net *net; 1338 1339 ASSERT_RTNL(); 1340 BUG_ON(!dev_net(dev)); 1341 1342 net = dev_net(dev); 1343 1344 /* Some auto-enslaved devices e.g. failover slaves are 1345 * special, as userspace might rename the device after 1346 * the interface had been brought up and running since 1347 * the point kernel initiated auto-enslavement. Allow 1348 * live name change even when these slave devices are 1349 * up and running. 1350 * 1351 * Typically, users of these auto-enslaving devices 1352 * don't actually care about slave name change, as 1353 * they are supposed to operate on master interface 1354 * directly. 1355 */ 1356 if (dev->flags & IFF_UP && 1357 likely(!(dev->priv_flags & IFF_LIVE_RENAME_OK))) 1358 return -EBUSY; 1359 1360 down_write(&devnet_rename_sem); 1361 1362 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { 1363 up_write(&devnet_rename_sem); 1364 return 0; 1365 } 1366 1367 memcpy(oldname, dev->name, IFNAMSIZ); 1368 1369 err = dev_get_valid_name(net, dev, newname); 1370 if (err < 0) { 1371 up_write(&devnet_rename_sem); 1372 return err; 1373 } 1374 1375 if (oldname[0] && !strchr(oldname, '%')) 1376 netdev_info(dev, "renamed from %s\n", oldname); 1377 1378 old_assign_type = dev->name_assign_type; 1379 dev->name_assign_type = NET_NAME_RENAMED; 1380 1381 rollback: 1382 ret = device_rename(&dev->dev, dev->name); 1383 if (ret) { 1384 memcpy(dev->name, oldname, IFNAMSIZ); 1385 dev->name_assign_type = old_assign_type; 1386 up_write(&devnet_rename_sem); 1387 return ret; 1388 } 1389 1390 up_write(&devnet_rename_sem); 1391 1392 netdev_adjacent_rename_links(dev, oldname); 1393 1394 write_lock_bh(&dev_base_lock); 1395 netdev_name_node_del(dev->name_node); 1396 write_unlock_bh(&dev_base_lock); 1397 1398 synchronize_rcu(); 1399 1400 write_lock_bh(&dev_base_lock); 1401 netdev_name_node_add(net, dev->name_node); 1402 write_unlock_bh(&dev_base_lock); 1403 1404 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); 1405 ret = notifier_to_errno(ret); 1406 1407 if (ret) { 1408 /* err >= 0 after dev_alloc_name() or stores the first errno */ 1409 if (err >= 0) { 1410 err = ret; 1411 down_write(&devnet_rename_sem); 1412 memcpy(dev->name, oldname, IFNAMSIZ); 1413 memcpy(oldname, newname, IFNAMSIZ); 1414 dev->name_assign_type = old_assign_type; 1415 old_assign_type = NET_NAME_RENAMED; 1416 goto rollback; 1417 } else { 1418 pr_err("%s: name change rollback failed: %d\n", 1419 dev->name, ret); 1420 } 1421 } 1422 1423 return err; 1424 } 1425 1426 /** 1427 * dev_set_alias - change ifalias of a device 1428 * @dev: device 1429 * @alias: name up to IFALIASZ 1430 * @len: limit of bytes to copy from info 1431 * 1432 * Set ifalias for a device, 1433 */ 1434 int dev_set_alias(struct net_device *dev, const char *alias, size_t len) 1435 { 1436 struct dev_ifalias *new_alias = NULL; 1437 1438 if (len >= IFALIASZ) 1439 return -EINVAL; 1440 1441 if (len) { 1442 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL); 1443 if (!new_alias) 1444 return -ENOMEM; 1445 1446 memcpy(new_alias->ifalias, alias, len); 1447 new_alias->ifalias[len] = 0; 1448 } 1449 1450 mutex_lock(&ifalias_mutex); 1451 new_alias = rcu_replace_pointer(dev->ifalias, new_alias, 1452 mutex_is_locked(&ifalias_mutex)); 1453 mutex_unlock(&ifalias_mutex); 1454 1455 if (new_alias) 1456 kfree_rcu(new_alias, rcuhead); 1457 1458 return len; 1459 } 1460 EXPORT_SYMBOL(dev_set_alias); 1461 1462 /** 1463 * dev_get_alias - get ifalias of a device 1464 * @dev: device 1465 * @name: buffer to store name of ifalias 1466 * @len: size of buffer 1467 * 1468 * get ifalias for a device. Caller must make sure dev cannot go 1469 * away, e.g. rcu read lock or own a reference count to device. 1470 */ 1471 int dev_get_alias(const struct net_device *dev, char *name, size_t len) 1472 { 1473 const struct dev_ifalias *alias; 1474 int ret = 0; 1475 1476 rcu_read_lock(); 1477 alias = rcu_dereference(dev->ifalias); 1478 if (alias) 1479 ret = snprintf(name, len, "%s", alias->ifalias); 1480 rcu_read_unlock(); 1481 1482 return ret; 1483 } 1484 1485 /** 1486 * netdev_features_change - device changes features 1487 * @dev: device to cause notification 1488 * 1489 * Called to indicate a device has changed features. 1490 */ 1491 void netdev_features_change(struct net_device *dev) 1492 { 1493 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); 1494 } 1495 EXPORT_SYMBOL(netdev_features_change); 1496 1497 /** 1498 * netdev_state_change - device changes state 1499 * @dev: device to cause notification 1500 * 1501 * Called to indicate a device has changed state. This function calls 1502 * the notifier chains for netdev_chain and sends a NEWLINK message 1503 * to the routing socket. 1504 */ 1505 void netdev_state_change(struct net_device *dev) 1506 { 1507 if (dev->flags & IFF_UP) { 1508 struct netdev_notifier_change_info change_info = { 1509 .info.dev = dev, 1510 }; 1511 1512 call_netdevice_notifiers_info(NETDEV_CHANGE, 1513 &change_info.info); 1514 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL); 1515 } 1516 } 1517 EXPORT_SYMBOL(netdev_state_change); 1518 1519 /** 1520 * __netdev_notify_peers - notify network peers about existence of @dev, 1521 * to be called when rtnl lock is already held. 1522 * @dev: network device 1523 * 1524 * Generate traffic such that interested network peers are aware of 1525 * @dev, such as by generating a gratuitous ARP. This may be used when 1526 * a device wants to inform the rest of the network about some sort of 1527 * reconfiguration such as a failover event or virtual machine 1528 * migration. 1529 */ 1530 void __netdev_notify_peers(struct net_device *dev) 1531 { 1532 ASSERT_RTNL(); 1533 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); 1534 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev); 1535 } 1536 EXPORT_SYMBOL(__netdev_notify_peers); 1537 1538 /** 1539 * netdev_notify_peers - notify network peers about existence of @dev 1540 * @dev: network device 1541 * 1542 * Generate traffic such that interested network peers are aware of 1543 * @dev, such as by generating a gratuitous ARP. This may be used when 1544 * a device wants to inform the rest of the network about some sort of 1545 * reconfiguration such as a failover event or virtual machine 1546 * migration. 1547 */ 1548 void netdev_notify_peers(struct net_device *dev) 1549 { 1550 rtnl_lock(); 1551 __netdev_notify_peers(dev); 1552 rtnl_unlock(); 1553 } 1554 EXPORT_SYMBOL(netdev_notify_peers); 1555 1556 static int napi_threaded_poll(void *data); 1557 1558 static int napi_kthread_create(struct napi_struct *n) 1559 { 1560 int err = 0; 1561 1562 /* Create and wake up the kthread once to put it in 1563 * TASK_INTERRUPTIBLE mode to avoid the blocked task 1564 * warning and work with loadavg. 1565 */ 1566 n->thread = kthread_run(napi_threaded_poll, n, "napi/%s-%d", 1567 n->dev->name, n->napi_id); 1568 if (IS_ERR(n->thread)) { 1569 err = PTR_ERR(n->thread); 1570 pr_err("kthread_run failed with err %d\n", err); 1571 n->thread = NULL; 1572 } 1573 1574 return err; 1575 } 1576 1577 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1578 { 1579 const struct net_device_ops *ops = dev->netdev_ops; 1580 int ret; 1581 1582 ASSERT_RTNL(); 1583 1584 if (!netif_device_present(dev)) { 1585 /* may be detached because parent is runtime-suspended */ 1586 if (dev->dev.parent) 1587 pm_runtime_resume(dev->dev.parent); 1588 if (!netif_device_present(dev)) 1589 return -ENODEV; 1590 } 1591 1592 /* Block netpoll from trying to do any rx path servicing. 1593 * If we don't do this there is a chance ndo_poll_controller 1594 * or ndo_poll may be running while we open the device 1595 */ 1596 netpoll_poll_disable(dev); 1597 1598 ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack); 1599 ret = notifier_to_errno(ret); 1600 if (ret) 1601 return ret; 1602 1603 set_bit(__LINK_STATE_START, &dev->state); 1604 1605 if (ops->ndo_validate_addr) 1606 ret = ops->ndo_validate_addr(dev); 1607 1608 if (!ret && ops->ndo_open) 1609 ret = ops->ndo_open(dev); 1610 1611 netpoll_poll_enable(dev); 1612 1613 if (ret) 1614 clear_bit(__LINK_STATE_START, &dev->state); 1615 else { 1616 dev->flags |= IFF_UP; 1617 dev_set_rx_mode(dev); 1618 dev_activate(dev); 1619 add_device_randomness(dev->dev_addr, dev->addr_len); 1620 } 1621 1622 return ret; 1623 } 1624 1625 /** 1626 * dev_open - prepare an interface for use. 1627 * @dev: device to open 1628 * @extack: netlink extended ack 1629 * 1630 * Takes a device from down to up state. The device's private open 1631 * function is invoked and then the multicast lists are loaded. Finally 1632 * the device is moved into the up state and a %NETDEV_UP message is 1633 * sent to the netdev notifier chain. 1634 * 1635 * Calling this function on an active interface is a nop. On a failure 1636 * a negative errno code is returned. 1637 */ 1638 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1639 { 1640 int ret; 1641 1642 if (dev->flags & IFF_UP) 1643 return 0; 1644 1645 ret = __dev_open(dev, extack); 1646 if (ret < 0) 1647 return ret; 1648 1649 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); 1650 call_netdevice_notifiers(NETDEV_UP, dev); 1651 1652 return ret; 1653 } 1654 EXPORT_SYMBOL(dev_open); 1655 1656 static void __dev_close_many(struct list_head *head) 1657 { 1658 struct net_device *dev; 1659 1660 ASSERT_RTNL(); 1661 might_sleep(); 1662 1663 list_for_each_entry(dev, head, close_list) { 1664 /* Temporarily disable netpoll until the interface is down */ 1665 netpoll_poll_disable(dev); 1666 1667 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); 1668 1669 clear_bit(__LINK_STATE_START, &dev->state); 1670 1671 /* Synchronize to scheduled poll. We cannot touch poll list, it 1672 * can be even on different cpu. So just clear netif_running(). 1673 * 1674 * dev->stop() will invoke napi_disable() on all of it's 1675 * napi_struct instances on this device. 1676 */ 1677 smp_mb__after_atomic(); /* Commit netif_running(). */ 1678 } 1679 1680 dev_deactivate_many(head); 1681 1682 list_for_each_entry(dev, head, close_list) { 1683 const struct net_device_ops *ops = dev->netdev_ops; 1684 1685 /* 1686 * Call the device specific close. This cannot fail. 1687 * Only if device is UP 1688 * 1689 * We allow it to be called even after a DETACH hot-plug 1690 * event. 1691 */ 1692 if (ops->ndo_stop) 1693 ops->ndo_stop(dev); 1694 1695 dev->flags &= ~IFF_UP; 1696 netpoll_poll_enable(dev); 1697 } 1698 } 1699 1700 static void __dev_close(struct net_device *dev) 1701 { 1702 LIST_HEAD(single); 1703 1704 list_add(&dev->close_list, &single); 1705 __dev_close_many(&single); 1706 list_del(&single); 1707 } 1708 1709 void dev_close_many(struct list_head *head, bool unlink) 1710 { 1711 struct net_device *dev, *tmp; 1712 1713 /* Remove the devices that don't need to be closed */ 1714 list_for_each_entry_safe(dev, tmp, head, close_list) 1715 if (!(dev->flags & IFF_UP)) 1716 list_del_init(&dev->close_list); 1717 1718 __dev_close_many(head); 1719 1720 list_for_each_entry_safe(dev, tmp, head, close_list) { 1721 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); 1722 call_netdevice_notifiers(NETDEV_DOWN, dev); 1723 if (unlink) 1724 list_del_init(&dev->close_list); 1725 } 1726 } 1727 EXPORT_SYMBOL(dev_close_many); 1728 1729 /** 1730 * dev_close - shutdown an interface. 1731 * @dev: device to shutdown 1732 * 1733 * This function moves an active device into down state. A 1734 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device 1735 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier 1736 * chain. 1737 */ 1738 void dev_close(struct net_device *dev) 1739 { 1740 if (dev->flags & IFF_UP) { 1741 LIST_HEAD(single); 1742 1743 list_add(&dev->close_list, &single); 1744 dev_close_many(&single, true); 1745 list_del(&single); 1746 } 1747 } 1748 EXPORT_SYMBOL(dev_close); 1749 1750 1751 /** 1752 * dev_disable_lro - disable Large Receive Offload on a device 1753 * @dev: device 1754 * 1755 * Disable Large Receive Offload (LRO) on a net device. Must be 1756 * called under RTNL. This is needed if received packets may be 1757 * forwarded to another interface. 1758 */ 1759 void dev_disable_lro(struct net_device *dev) 1760 { 1761 struct net_device *lower_dev; 1762 struct list_head *iter; 1763 1764 dev->wanted_features &= ~NETIF_F_LRO; 1765 netdev_update_features(dev); 1766 1767 if (unlikely(dev->features & NETIF_F_LRO)) 1768 netdev_WARN(dev, "failed to disable LRO!\n"); 1769 1770 netdev_for_each_lower_dev(dev, lower_dev, iter) 1771 dev_disable_lro(lower_dev); 1772 } 1773 EXPORT_SYMBOL(dev_disable_lro); 1774 1775 /** 1776 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device 1777 * @dev: device 1778 * 1779 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be 1780 * called under RTNL. This is needed if Generic XDP is installed on 1781 * the device. 1782 */ 1783 static void dev_disable_gro_hw(struct net_device *dev) 1784 { 1785 dev->wanted_features &= ~NETIF_F_GRO_HW; 1786 netdev_update_features(dev); 1787 1788 if (unlikely(dev->features & NETIF_F_GRO_HW)) 1789 netdev_WARN(dev, "failed to disable GRO_HW!\n"); 1790 } 1791 1792 const char *netdev_cmd_to_name(enum netdev_cmd cmd) 1793 { 1794 #define N(val) \ 1795 case NETDEV_##val: \ 1796 return "NETDEV_" __stringify(val); 1797 switch (cmd) { 1798 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER) 1799 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE) 1800 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE) 1801 N(POST_INIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) N(CHANGEUPPER) 1802 N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) N(BONDING_INFO) 1803 N(PRECHANGEUPPER) N(CHANGELOWERSTATE) N(UDP_TUNNEL_PUSH_INFO) 1804 N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN) 1805 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO) 1806 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO) 1807 N(PRE_CHANGEADDR) 1808 } 1809 #undef N 1810 return "UNKNOWN_NETDEV_EVENT"; 1811 } 1812 EXPORT_SYMBOL_GPL(netdev_cmd_to_name); 1813 1814 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, 1815 struct net_device *dev) 1816 { 1817 struct netdev_notifier_info info = { 1818 .dev = dev, 1819 }; 1820 1821 return nb->notifier_call(nb, val, &info); 1822 } 1823 1824 static int call_netdevice_register_notifiers(struct notifier_block *nb, 1825 struct net_device *dev) 1826 { 1827 int err; 1828 1829 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); 1830 err = notifier_to_errno(err); 1831 if (err) 1832 return err; 1833 1834 if (!(dev->flags & IFF_UP)) 1835 return 0; 1836 1837 call_netdevice_notifier(nb, NETDEV_UP, dev); 1838 return 0; 1839 } 1840 1841 static void call_netdevice_unregister_notifiers(struct notifier_block *nb, 1842 struct net_device *dev) 1843 { 1844 if (dev->flags & IFF_UP) { 1845 call_netdevice_notifier(nb, NETDEV_GOING_DOWN, 1846 dev); 1847 call_netdevice_notifier(nb, NETDEV_DOWN, dev); 1848 } 1849 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); 1850 } 1851 1852 static int call_netdevice_register_net_notifiers(struct notifier_block *nb, 1853 struct net *net) 1854 { 1855 struct net_device *dev; 1856 int err; 1857 1858 for_each_netdev(net, dev) { 1859 err = call_netdevice_register_notifiers(nb, dev); 1860 if (err) 1861 goto rollback; 1862 } 1863 return 0; 1864 1865 rollback: 1866 for_each_netdev_continue_reverse(net, dev) 1867 call_netdevice_unregister_notifiers(nb, dev); 1868 return err; 1869 } 1870 1871 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb, 1872 struct net *net) 1873 { 1874 struct net_device *dev; 1875 1876 for_each_netdev(net, dev) 1877 call_netdevice_unregister_notifiers(nb, dev); 1878 } 1879 1880 static int dev_boot_phase = 1; 1881 1882 /** 1883 * register_netdevice_notifier - register a network notifier block 1884 * @nb: notifier 1885 * 1886 * Register a notifier to be called when network device events occur. 1887 * The notifier passed is linked into the kernel structures and must 1888 * not be reused until it has been unregistered. A negative errno code 1889 * is returned on a failure. 1890 * 1891 * When registered all registration and up events are replayed 1892 * to the new notifier to allow device to have a race free 1893 * view of the network device list. 1894 */ 1895 1896 int register_netdevice_notifier(struct notifier_block *nb) 1897 { 1898 struct net *net; 1899 int err; 1900 1901 /* Close race with setup_net() and cleanup_net() */ 1902 down_write(&pernet_ops_rwsem); 1903 rtnl_lock(); 1904 err = raw_notifier_chain_register(&netdev_chain, nb); 1905 if (err) 1906 goto unlock; 1907 if (dev_boot_phase) 1908 goto unlock; 1909 for_each_net(net) { 1910 err = call_netdevice_register_net_notifiers(nb, net); 1911 if (err) 1912 goto rollback; 1913 } 1914 1915 unlock: 1916 rtnl_unlock(); 1917 up_write(&pernet_ops_rwsem); 1918 return err; 1919 1920 rollback: 1921 for_each_net_continue_reverse(net) 1922 call_netdevice_unregister_net_notifiers(nb, net); 1923 1924 raw_notifier_chain_unregister(&netdev_chain, nb); 1925 goto unlock; 1926 } 1927 EXPORT_SYMBOL(register_netdevice_notifier); 1928 1929 /** 1930 * unregister_netdevice_notifier - unregister a network notifier block 1931 * @nb: notifier 1932 * 1933 * Unregister a notifier previously registered by 1934 * register_netdevice_notifier(). The notifier is unlinked into the 1935 * kernel structures and may then be reused. A negative errno code 1936 * is returned on a failure. 1937 * 1938 * After unregistering unregister and down device events are synthesized 1939 * for all devices on the device list to the removed notifier to remove 1940 * the need for special case cleanup code. 1941 */ 1942 1943 int unregister_netdevice_notifier(struct notifier_block *nb) 1944 { 1945 struct net *net; 1946 int err; 1947 1948 /* Close race with setup_net() and cleanup_net() */ 1949 down_write(&pernet_ops_rwsem); 1950 rtnl_lock(); 1951 err = raw_notifier_chain_unregister(&netdev_chain, nb); 1952 if (err) 1953 goto unlock; 1954 1955 for_each_net(net) 1956 call_netdevice_unregister_net_notifiers(nb, net); 1957 1958 unlock: 1959 rtnl_unlock(); 1960 up_write(&pernet_ops_rwsem); 1961 return err; 1962 } 1963 EXPORT_SYMBOL(unregister_netdevice_notifier); 1964 1965 static int __register_netdevice_notifier_net(struct net *net, 1966 struct notifier_block *nb, 1967 bool ignore_call_fail) 1968 { 1969 int err; 1970 1971 err = raw_notifier_chain_register(&net->netdev_chain, nb); 1972 if (err) 1973 return err; 1974 if (dev_boot_phase) 1975 return 0; 1976 1977 err = call_netdevice_register_net_notifiers(nb, net); 1978 if (err && !ignore_call_fail) 1979 goto chain_unregister; 1980 1981 return 0; 1982 1983 chain_unregister: 1984 raw_notifier_chain_unregister(&net->netdev_chain, nb); 1985 return err; 1986 } 1987 1988 static int __unregister_netdevice_notifier_net(struct net *net, 1989 struct notifier_block *nb) 1990 { 1991 int err; 1992 1993 err = raw_notifier_chain_unregister(&net->netdev_chain, nb); 1994 if (err) 1995 return err; 1996 1997 call_netdevice_unregister_net_notifiers(nb, net); 1998 return 0; 1999 } 2000 2001 /** 2002 * register_netdevice_notifier_net - register a per-netns network notifier block 2003 * @net: network namespace 2004 * @nb: notifier 2005 * 2006 * Register a notifier to be called when network device events occur. 2007 * The notifier passed is linked into the kernel structures and must 2008 * not be reused until it has been unregistered. A negative errno code 2009 * is returned on a failure. 2010 * 2011 * When registered all registration and up events are replayed 2012 * to the new notifier to allow device to have a race free 2013 * view of the network device list. 2014 */ 2015 2016 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb) 2017 { 2018 int err; 2019 2020 rtnl_lock(); 2021 err = __register_netdevice_notifier_net(net, nb, false); 2022 rtnl_unlock(); 2023 return err; 2024 } 2025 EXPORT_SYMBOL(register_netdevice_notifier_net); 2026 2027 /** 2028 * unregister_netdevice_notifier_net - unregister a per-netns 2029 * network notifier block 2030 * @net: network namespace 2031 * @nb: notifier 2032 * 2033 * Unregister a notifier previously registered by 2034 * register_netdevice_notifier(). The notifier is unlinked into the 2035 * kernel structures and may then be reused. A negative errno code 2036 * is returned on a failure. 2037 * 2038 * After unregistering unregister and down device events are synthesized 2039 * for all devices on the device list to the removed notifier to remove 2040 * the need for special case cleanup code. 2041 */ 2042 2043 int unregister_netdevice_notifier_net(struct net *net, 2044 struct notifier_block *nb) 2045 { 2046 int err; 2047 2048 rtnl_lock(); 2049 err = __unregister_netdevice_notifier_net(net, nb); 2050 rtnl_unlock(); 2051 return err; 2052 } 2053 EXPORT_SYMBOL(unregister_netdevice_notifier_net); 2054 2055 int register_netdevice_notifier_dev_net(struct net_device *dev, 2056 struct notifier_block *nb, 2057 struct netdev_net_notifier *nn) 2058 { 2059 int err; 2060 2061 rtnl_lock(); 2062 err = __register_netdevice_notifier_net(dev_net(dev), nb, false); 2063 if (!err) { 2064 nn->nb = nb; 2065 list_add(&nn->list, &dev->net_notifier_list); 2066 } 2067 rtnl_unlock(); 2068 return err; 2069 } 2070 EXPORT_SYMBOL(register_netdevice_notifier_dev_net); 2071 2072 int unregister_netdevice_notifier_dev_net(struct net_device *dev, 2073 struct notifier_block *nb, 2074 struct netdev_net_notifier *nn) 2075 { 2076 int err; 2077 2078 rtnl_lock(); 2079 list_del(&nn->list); 2080 err = __unregister_netdevice_notifier_net(dev_net(dev), nb); 2081 rtnl_unlock(); 2082 return err; 2083 } 2084 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net); 2085 2086 static void move_netdevice_notifiers_dev_net(struct net_device *dev, 2087 struct net *net) 2088 { 2089 struct netdev_net_notifier *nn; 2090 2091 list_for_each_entry(nn, &dev->net_notifier_list, list) { 2092 __unregister_netdevice_notifier_net(dev_net(dev), nn->nb); 2093 __register_netdevice_notifier_net(net, nn->nb, true); 2094 } 2095 } 2096 2097 /** 2098 * call_netdevice_notifiers_info - call all network notifier blocks 2099 * @val: value passed unmodified to notifier function 2100 * @info: notifier information data 2101 * 2102 * Call all network notifier blocks. Parameters and return value 2103 * are as for raw_notifier_call_chain(). 2104 */ 2105 2106 static int call_netdevice_notifiers_info(unsigned long val, 2107 struct netdev_notifier_info *info) 2108 { 2109 struct net *net = dev_net(info->dev); 2110 int ret; 2111 2112 ASSERT_RTNL(); 2113 2114 /* Run per-netns notifier block chain first, then run the global one. 2115 * Hopefully, one day, the global one is going to be removed after 2116 * all notifier block registrators get converted to be per-netns. 2117 */ 2118 ret = raw_notifier_call_chain(&net->netdev_chain, val, info); 2119 if (ret & NOTIFY_STOP_MASK) 2120 return ret; 2121 return raw_notifier_call_chain(&netdev_chain, val, info); 2122 } 2123 2124 static int call_netdevice_notifiers_extack(unsigned long val, 2125 struct net_device *dev, 2126 struct netlink_ext_ack *extack) 2127 { 2128 struct netdev_notifier_info info = { 2129 .dev = dev, 2130 .extack = extack, 2131 }; 2132 2133 return call_netdevice_notifiers_info(val, &info); 2134 } 2135 2136 /** 2137 * call_netdevice_notifiers - call all network notifier blocks 2138 * @val: value passed unmodified to notifier function 2139 * @dev: net_device pointer passed unmodified to notifier function 2140 * 2141 * Call all network notifier blocks. Parameters and return value 2142 * are as for raw_notifier_call_chain(). 2143 */ 2144 2145 int call_netdevice_notifiers(unsigned long val, struct net_device *dev) 2146 { 2147 return call_netdevice_notifiers_extack(val, dev, NULL); 2148 } 2149 EXPORT_SYMBOL(call_netdevice_notifiers); 2150 2151 /** 2152 * call_netdevice_notifiers_mtu - call all network notifier blocks 2153 * @val: value passed unmodified to notifier function 2154 * @dev: net_device pointer passed unmodified to notifier function 2155 * @arg: additional u32 argument passed to the notifier function 2156 * 2157 * Call all network notifier blocks. Parameters and return value 2158 * are as for raw_notifier_call_chain(). 2159 */ 2160 static int call_netdevice_notifiers_mtu(unsigned long val, 2161 struct net_device *dev, u32 arg) 2162 { 2163 struct netdev_notifier_info_ext info = { 2164 .info.dev = dev, 2165 .ext.mtu = arg, 2166 }; 2167 2168 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0); 2169 2170 return call_netdevice_notifiers_info(val, &info.info); 2171 } 2172 2173 #ifdef CONFIG_NET_INGRESS 2174 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key); 2175 2176 void net_inc_ingress_queue(void) 2177 { 2178 static_branch_inc(&ingress_needed_key); 2179 } 2180 EXPORT_SYMBOL_GPL(net_inc_ingress_queue); 2181 2182 void net_dec_ingress_queue(void) 2183 { 2184 static_branch_dec(&ingress_needed_key); 2185 } 2186 EXPORT_SYMBOL_GPL(net_dec_ingress_queue); 2187 #endif 2188 2189 #ifdef CONFIG_NET_EGRESS 2190 static DEFINE_STATIC_KEY_FALSE(egress_needed_key); 2191 2192 void net_inc_egress_queue(void) 2193 { 2194 static_branch_inc(&egress_needed_key); 2195 } 2196 EXPORT_SYMBOL_GPL(net_inc_egress_queue); 2197 2198 void net_dec_egress_queue(void) 2199 { 2200 static_branch_dec(&egress_needed_key); 2201 } 2202 EXPORT_SYMBOL_GPL(net_dec_egress_queue); 2203 #endif 2204 2205 static DEFINE_STATIC_KEY_FALSE(netstamp_needed_key); 2206 #ifdef CONFIG_JUMP_LABEL 2207 static atomic_t netstamp_needed_deferred; 2208 static atomic_t netstamp_wanted; 2209 static void netstamp_clear(struct work_struct *work) 2210 { 2211 int deferred = atomic_xchg(&netstamp_needed_deferred, 0); 2212 int wanted; 2213 2214 wanted = atomic_add_return(deferred, &netstamp_wanted); 2215 if (wanted > 0) 2216 static_branch_enable(&netstamp_needed_key); 2217 else 2218 static_branch_disable(&netstamp_needed_key); 2219 } 2220 static DECLARE_WORK(netstamp_work, netstamp_clear); 2221 #endif 2222 2223 void net_enable_timestamp(void) 2224 { 2225 #ifdef CONFIG_JUMP_LABEL 2226 int wanted; 2227 2228 while (1) { 2229 wanted = atomic_read(&netstamp_wanted); 2230 if (wanted <= 0) 2231 break; 2232 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted) 2233 return; 2234 } 2235 atomic_inc(&netstamp_needed_deferred); 2236 schedule_work(&netstamp_work); 2237 #else 2238 static_branch_inc(&netstamp_needed_key); 2239 #endif 2240 } 2241 EXPORT_SYMBOL(net_enable_timestamp); 2242 2243 void net_disable_timestamp(void) 2244 { 2245 #ifdef CONFIG_JUMP_LABEL 2246 int wanted; 2247 2248 while (1) { 2249 wanted = atomic_read(&netstamp_wanted); 2250 if (wanted <= 1) 2251 break; 2252 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted) 2253 return; 2254 } 2255 atomic_dec(&netstamp_needed_deferred); 2256 schedule_work(&netstamp_work); 2257 #else 2258 static_branch_dec(&netstamp_needed_key); 2259 #endif 2260 } 2261 EXPORT_SYMBOL(net_disable_timestamp); 2262 2263 static inline void net_timestamp_set(struct sk_buff *skb) 2264 { 2265 skb->tstamp = 0; 2266 if (static_branch_unlikely(&netstamp_needed_key)) 2267 __net_timestamp(skb); 2268 } 2269 2270 #define net_timestamp_check(COND, SKB) \ 2271 if (static_branch_unlikely(&netstamp_needed_key)) { \ 2272 if ((COND) && !(SKB)->tstamp) \ 2273 __net_timestamp(SKB); \ 2274 } \ 2275 2276 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb) 2277 { 2278 return __is_skb_forwardable(dev, skb, true); 2279 } 2280 EXPORT_SYMBOL_GPL(is_skb_forwardable); 2281 2282 static int __dev_forward_skb2(struct net_device *dev, struct sk_buff *skb, 2283 bool check_mtu) 2284 { 2285 int ret = ____dev_forward_skb(dev, skb, check_mtu); 2286 2287 if (likely(!ret)) { 2288 skb->protocol = eth_type_trans(skb, dev); 2289 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); 2290 } 2291 2292 return ret; 2293 } 2294 2295 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2296 { 2297 return __dev_forward_skb2(dev, skb, true); 2298 } 2299 EXPORT_SYMBOL_GPL(__dev_forward_skb); 2300 2301 /** 2302 * dev_forward_skb - loopback an skb to another netif 2303 * 2304 * @dev: destination network device 2305 * @skb: buffer to forward 2306 * 2307 * return values: 2308 * NET_RX_SUCCESS (no congestion) 2309 * NET_RX_DROP (packet was dropped, but freed) 2310 * 2311 * dev_forward_skb can be used for injecting an skb from the 2312 * start_xmit function of one device into the receive queue 2313 * of another device. 2314 * 2315 * The receiving device may be in another namespace, so 2316 * we have to clear all information in the skb that could 2317 * impact namespace isolation. 2318 */ 2319 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2320 { 2321 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); 2322 } 2323 EXPORT_SYMBOL_GPL(dev_forward_skb); 2324 2325 int dev_forward_skb_nomtu(struct net_device *dev, struct sk_buff *skb) 2326 { 2327 return __dev_forward_skb2(dev, skb, false) ?: netif_rx_internal(skb); 2328 } 2329 2330 static inline int deliver_skb(struct sk_buff *skb, 2331 struct packet_type *pt_prev, 2332 struct net_device *orig_dev) 2333 { 2334 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 2335 return -ENOMEM; 2336 refcount_inc(&skb->users); 2337 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 2338 } 2339 2340 static inline void deliver_ptype_list_skb(struct sk_buff *skb, 2341 struct packet_type **pt, 2342 struct net_device *orig_dev, 2343 __be16 type, 2344 struct list_head *ptype_list) 2345 { 2346 struct packet_type *ptype, *pt_prev = *pt; 2347 2348 list_for_each_entry_rcu(ptype, ptype_list, list) { 2349 if (ptype->type != type) 2350 continue; 2351 if (pt_prev) 2352 deliver_skb(skb, pt_prev, orig_dev); 2353 pt_prev = ptype; 2354 } 2355 *pt = pt_prev; 2356 } 2357 2358 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) 2359 { 2360 if (!ptype->af_packet_priv || !skb->sk) 2361 return false; 2362 2363 if (ptype->id_match) 2364 return ptype->id_match(ptype, skb->sk); 2365 else if ((struct sock *)ptype->af_packet_priv == skb->sk) 2366 return true; 2367 2368 return false; 2369 } 2370 2371 /** 2372 * dev_nit_active - return true if any network interface taps are in use 2373 * 2374 * @dev: network device to check for the presence of taps 2375 */ 2376 bool dev_nit_active(struct net_device *dev) 2377 { 2378 return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all); 2379 } 2380 EXPORT_SYMBOL_GPL(dev_nit_active); 2381 2382 /* 2383 * Support routine. Sends outgoing frames to any network 2384 * taps currently in use. 2385 */ 2386 2387 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) 2388 { 2389 struct packet_type *ptype; 2390 struct sk_buff *skb2 = NULL; 2391 struct packet_type *pt_prev = NULL; 2392 struct list_head *ptype_list = &ptype_all; 2393 2394 rcu_read_lock(); 2395 again: 2396 list_for_each_entry_rcu(ptype, ptype_list, list) { 2397 if (ptype->ignore_outgoing) 2398 continue; 2399 2400 /* Never send packets back to the socket 2401 * they originated from - MvS (miquels@drinkel.ow.org) 2402 */ 2403 if (skb_loop_sk(ptype, skb)) 2404 continue; 2405 2406 if (pt_prev) { 2407 deliver_skb(skb2, pt_prev, skb->dev); 2408 pt_prev = ptype; 2409 continue; 2410 } 2411 2412 /* need to clone skb, done only once */ 2413 skb2 = skb_clone(skb, GFP_ATOMIC); 2414 if (!skb2) 2415 goto out_unlock; 2416 2417 net_timestamp_set(skb2); 2418 2419 /* skb->nh should be correctly 2420 * set by sender, so that the second statement is 2421 * just protection against buggy protocols. 2422 */ 2423 skb_reset_mac_header(skb2); 2424 2425 if (skb_network_header(skb2) < skb2->data || 2426 skb_network_header(skb2) > skb_tail_pointer(skb2)) { 2427 net_crit_ratelimited("protocol %04x is buggy, dev %s\n", 2428 ntohs(skb2->protocol), 2429 dev->name); 2430 skb_reset_network_header(skb2); 2431 } 2432 2433 skb2->transport_header = skb2->network_header; 2434 skb2->pkt_type = PACKET_OUTGOING; 2435 pt_prev = ptype; 2436 } 2437 2438 if (ptype_list == &ptype_all) { 2439 ptype_list = &dev->ptype_all; 2440 goto again; 2441 } 2442 out_unlock: 2443 if (pt_prev) { 2444 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC)) 2445 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); 2446 else 2447 kfree_skb(skb2); 2448 } 2449 rcu_read_unlock(); 2450 } 2451 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit); 2452 2453 /** 2454 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change 2455 * @dev: Network device 2456 * @txq: number of queues available 2457 * 2458 * If real_num_tx_queues is changed the tc mappings may no longer be 2459 * valid. To resolve this verify the tc mapping remains valid and if 2460 * not NULL the mapping. With no priorities mapping to this 2461 * offset/count pair it will no longer be used. In the worst case TC0 2462 * is invalid nothing can be done so disable priority mappings. If is 2463 * expected that drivers will fix this mapping if they can before 2464 * calling netif_set_real_num_tx_queues. 2465 */ 2466 static void netif_setup_tc(struct net_device *dev, unsigned int txq) 2467 { 2468 int i; 2469 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2470 2471 /* If TC0 is invalidated disable TC mapping */ 2472 if (tc->offset + tc->count > txq) { 2473 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); 2474 dev->num_tc = 0; 2475 return; 2476 } 2477 2478 /* Invalidated prio to tc mappings set to TC0 */ 2479 for (i = 1; i < TC_BITMASK + 1; i++) { 2480 int q = netdev_get_prio_tc_map(dev, i); 2481 2482 tc = &dev->tc_to_txq[q]; 2483 if (tc->offset + tc->count > txq) { 2484 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n", 2485 i, q); 2486 netdev_set_prio_tc_map(dev, i, 0); 2487 } 2488 } 2489 } 2490 2491 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq) 2492 { 2493 if (dev->num_tc) { 2494 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2495 int i; 2496 2497 /* walk through the TCs and see if it falls into any of them */ 2498 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) { 2499 if ((txq - tc->offset) < tc->count) 2500 return i; 2501 } 2502 2503 /* didn't find it, just return -1 to indicate no match */ 2504 return -1; 2505 } 2506 2507 return 0; 2508 } 2509 EXPORT_SYMBOL(netdev_txq_to_tc); 2510 2511 #ifdef CONFIG_XPS 2512 static struct static_key xps_needed __read_mostly; 2513 static struct static_key xps_rxqs_needed __read_mostly; 2514 static DEFINE_MUTEX(xps_map_mutex); 2515 #define xmap_dereference(P) \ 2516 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) 2517 2518 static bool remove_xps_queue(struct xps_dev_maps *dev_maps, 2519 struct xps_dev_maps *old_maps, int tci, u16 index) 2520 { 2521 struct xps_map *map = NULL; 2522 int pos; 2523 2524 if (dev_maps) 2525 map = xmap_dereference(dev_maps->attr_map[tci]); 2526 if (!map) 2527 return false; 2528 2529 for (pos = map->len; pos--;) { 2530 if (map->queues[pos] != index) 2531 continue; 2532 2533 if (map->len > 1) { 2534 map->queues[pos] = map->queues[--map->len]; 2535 break; 2536 } 2537 2538 if (old_maps) 2539 RCU_INIT_POINTER(old_maps->attr_map[tci], NULL); 2540 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2541 kfree_rcu(map, rcu); 2542 return false; 2543 } 2544 2545 return true; 2546 } 2547 2548 static bool remove_xps_queue_cpu(struct net_device *dev, 2549 struct xps_dev_maps *dev_maps, 2550 int cpu, u16 offset, u16 count) 2551 { 2552 int num_tc = dev_maps->num_tc; 2553 bool active = false; 2554 int tci; 2555 2556 for (tci = cpu * num_tc; num_tc--; tci++) { 2557 int i, j; 2558 2559 for (i = count, j = offset; i--; j++) { 2560 if (!remove_xps_queue(dev_maps, NULL, tci, j)) 2561 break; 2562 } 2563 2564 active |= i < 0; 2565 } 2566 2567 return active; 2568 } 2569 2570 static void reset_xps_maps(struct net_device *dev, 2571 struct xps_dev_maps *dev_maps, 2572 enum xps_map_type type) 2573 { 2574 static_key_slow_dec_cpuslocked(&xps_needed); 2575 if (type == XPS_RXQS) 2576 static_key_slow_dec_cpuslocked(&xps_rxqs_needed); 2577 2578 RCU_INIT_POINTER(dev->xps_maps[type], NULL); 2579 2580 kfree_rcu(dev_maps, rcu); 2581 } 2582 2583 static void clean_xps_maps(struct net_device *dev, enum xps_map_type type, 2584 u16 offset, u16 count) 2585 { 2586 struct xps_dev_maps *dev_maps; 2587 bool active = false; 2588 int i, j; 2589 2590 dev_maps = xmap_dereference(dev->xps_maps[type]); 2591 if (!dev_maps) 2592 return; 2593 2594 for (j = 0; j < dev_maps->nr_ids; j++) 2595 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, count); 2596 if (!active) 2597 reset_xps_maps(dev, dev_maps, type); 2598 2599 if (type == XPS_CPUS) { 2600 for (i = offset + (count - 1); count--; i--) 2601 netdev_queue_numa_node_write( 2602 netdev_get_tx_queue(dev, i), NUMA_NO_NODE); 2603 } 2604 } 2605 2606 static void netif_reset_xps_queues(struct net_device *dev, u16 offset, 2607 u16 count) 2608 { 2609 if (!static_key_false(&xps_needed)) 2610 return; 2611 2612 cpus_read_lock(); 2613 mutex_lock(&xps_map_mutex); 2614 2615 if (static_key_false(&xps_rxqs_needed)) 2616 clean_xps_maps(dev, XPS_RXQS, offset, count); 2617 2618 clean_xps_maps(dev, XPS_CPUS, offset, count); 2619 2620 mutex_unlock(&xps_map_mutex); 2621 cpus_read_unlock(); 2622 } 2623 2624 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) 2625 { 2626 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index); 2627 } 2628 2629 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index, 2630 u16 index, bool is_rxqs_map) 2631 { 2632 struct xps_map *new_map; 2633 int alloc_len = XPS_MIN_MAP_ALLOC; 2634 int i, pos; 2635 2636 for (pos = 0; map && pos < map->len; pos++) { 2637 if (map->queues[pos] != index) 2638 continue; 2639 return map; 2640 } 2641 2642 /* Need to add tx-queue to this CPU's/rx-queue's existing map */ 2643 if (map) { 2644 if (pos < map->alloc_len) 2645 return map; 2646 2647 alloc_len = map->alloc_len * 2; 2648 } 2649 2650 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's 2651 * map 2652 */ 2653 if (is_rxqs_map) 2654 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL); 2655 else 2656 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, 2657 cpu_to_node(attr_index)); 2658 if (!new_map) 2659 return NULL; 2660 2661 for (i = 0; i < pos; i++) 2662 new_map->queues[i] = map->queues[i]; 2663 new_map->alloc_len = alloc_len; 2664 new_map->len = pos; 2665 2666 return new_map; 2667 } 2668 2669 /* Copy xps maps at a given index */ 2670 static void xps_copy_dev_maps(struct xps_dev_maps *dev_maps, 2671 struct xps_dev_maps *new_dev_maps, int index, 2672 int tc, bool skip_tc) 2673 { 2674 int i, tci = index * dev_maps->num_tc; 2675 struct xps_map *map; 2676 2677 /* copy maps belonging to foreign traffic classes */ 2678 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2679 if (i == tc && skip_tc) 2680 continue; 2681 2682 /* fill in the new device map from the old device map */ 2683 map = xmap_dereference(dev_maps->attr_map[tci]); 2684 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2685 } 2686 } 2687 2688 /* Must be called under cpus_read_lock */ 2689 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask, 2690 u16 index, enum xps_map_type type) 2691 { 2692 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL, *old_dev_maps = NULL; 2693 const unsigned long *online_mask = NULL; 2694 bool active = false, copy = false; 2695 int i, j, tci, numa_node_id = -2; 2696 int maps_sz, num_tc = 1, tc = 0; 2697 struct xps_map *map, *new_map; 2698 unsigned int nr_ids; 2699 2700 if (dev->num_tc) { 2701 /* Do not allow XPS on subordinate device directly */ 2702 num_tc = dev->num_tc; 2703 if (num_tc < 0) 2704 return -EINVAL; 2705 2706 /* If queue belongs to subordinate dev use its map */ 2707 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev; 2708 2709 tc = netdev_txq_to_tc(dev, index); 2710 if (tc < 0) 2711 return -EINVAL; 2712 } 2713 2714 mutex_lock(&xps_map_mutex); 2715 2716 dev_maps = xmap_dereference(dev->xps_maps[type]); 2717 if (type == XPS_RXQS) { 2718 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues); 2719 nr_ids = dev->num_rx_queues; 2720 } else { 2721 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc); 2722 if (num_possible_cpus() > 1) 2723 online_mask = cpumask_bits(cpu_online_mask); 2724 nr_ids = nr_cpu_ids; 2725 } 2726 2727 if (maps_sz < L1_CACHE_BYTES) 2728 maps_sz = L1_CACHE_BYTES; 2729 2730 /* The old dev_maps could be larger or smaller than the one we're 2731 * setting up now, as dev->num_tc or nr_ids could have been updated in 2732 * between. We could try to be smart, but let's be safe instead and only 2733 * copy foreign traffic classes if the two map sizes match. 2734 */ 2735 if (dev_maps && 2736 dev_maps->num_tc == num_tc && dev_maps->nr_ids == nr_ids) 2737 copy = true; 2738 2739 /* allocate memory for queue storage */ 2740 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids), 2741 j < nr_ids;) { 2742 if (!new_dev_maps) { 2743 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); 2744 if (!new_dev_maps) { 2745 mutex_unlock(&xps_map_mutex); 2746 return -ENOMEM; 2747 } 2748 2749 new_dev_maps->nr_ids = nr_ids; 2750 new_dev_maps->num_tc = num_tc; 2751 } 2752 2753 tci = j * num_tc + tc; 2754 map = copy ? xmap_dereference(dev_maps->attr_map[tci]) : NULL; 2755 2756 map = expand_xps_map(map, j, index, type == XPS_RXQS); 2757 if (!map) 2758 goto error; 2759 2760 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2761 } 2762 2763 if (!new_dev_maps) 2764 goto out_no_new_maps; 2765 2766 if (!dev_maps) { 2767 /* Increment static keys at most once per type */ 2768 static_key_slow_inc_cpuslocked(&xps_needed); 2769 if (type == XPS_RXQS) 2770 static_key_slow_inc_cpuslocked(&xps_rxqs_needed); 2771 } 2772 2773 for (j = 0; j < nr_ids; j++) { 2774 bool skip_tc = false; 2775 2776 tci = j * num_tc + tc; 2777 if (netif_attr_test_mask(j, mask, nr_ids) && 2778 netif_attr_test_online(j, online_mask, nr_ids)) { 2779 /* add tx-queue to CPU/rx-queue maps */ 2780 int pos = 0; 2781 2782 skip_tc = true; 2783 2784 map = xmap_dereference(new_dev_maps->attr_map[tci]); 2785 while ((pos < map->len) && (map->queues[pos] != index)) 2786 pos++; 2787 2788 if (pos == map->len) 2789 map->queues[map->len++] = index; 2790 #ifdef CONFIG_NUMA 2791 if (type == XPS_CPUS) { 2792 if (numa_node_id == -2) 2793 numa_node_id = cpu_to_node(j); 2794 else if (numa_node_id != cpu_to_node(j)) 2795 numa_node_id = -1; 2796 } 2797 #endif 2798 } 2799 2800 if (copy) 2801 xps_copy_dev_maps(dev_maps, new_dev_maps, j, tc, 2802 skip_tc); 2803 } 2804 2805 rcu_assign_pointer(dev->xps_maps[type], new_dev_maps); 2806 2807 /* Cleanup old maps */ 2808 if (!dev_maps) 2809 goto out_no_old_maps; 2810 2811 for (j = 0; j < dev_maps->nr_ids; j++) { 2812 for (i = num_tc, tci = j * dev_maps->num_tc; i--; tci++) { 2813 map = xmap_dereference(dev_maps->attr_map[tci]); 2814 if (!map) 2815 continue; 2816 2817 if (copy) { 2818 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2819 if (map == new_map) 2820 continue; 2821 } 2822 2823 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2824 kfree_rcu(map, rcu); 2825 } 2826 } 2827 2828 old_dev_maps = dev_maps; 2829 2830 out_no_old_maps: 2831 dev_maps = new_dev_maps; 2832 active = true; 2833 2834 out_no_new_maps: 2835 if (type == XPS_CPUS) 2836 /* update Tx queue numa node */ 2837 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), 2838 (numa_node_id >= 0) ? 2839 numa_node_id : NUMA_NO_NODE); 2840 2841 if (!dev_maps) 2842 goto out_no_maps; 2843 2844 /* removes tx-queue from unused CPUs/rx-queues */ 2845 for (j = 0; j < dev_maps->nr_ids; j++) { 2846 tci = j * dev_maps->num_tc; 2847 2848 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2849 if (i == tc && 2850 netif_attr_test_mask(j, mask, dev_maps->nr_ids) && 2851 netif_attr_test_online(j, online_mask, dev_maps->nr_ids)) 2852 continue; 2853 2854 active |= remove_xps_queue(dev_maps, 2855 copy ? old_dev_maps : NULL, 2856 tci, index); 2857 } 2858 } 2859 2860 if (old_dev_maps) 2861 kfree_rcu(old_dev_maps, rcu); 2862 2863 /* free map if not active */ 2864 if (!active) 2865 reset_xps_maps(dev, dev_maps, type); 2866 2867 out_no_maps: 2868 mutex_unlock(&xps_map_mutex); 2869 2870 return 0; 2871 error: 2872 /* remove any maps that we added */ 2873 for (j = 0; j < nr_ids; j++) { 2874 for (i = num_tc, tci = j * num_tc; i--; tci++) { 2875 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2876 map = copy ? 2877 xmap_dereference(dev_maps->attr_map[tci]) : 2878 NULL; 2879 if (new_map && new_map != map) 2880 kfree(new_map); 2881 } 2882 } 2883 2884 mutex_unlock(&xps_map_mutex); 2885 2886 kfree(new_dev_maps); 2887 return -ENOMEM; 2888 } 2889 EXPORT_SYMBOL_GPL(__netif_set_xps_queue); 2890 2891 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, 2892 u16 index) 2893 { 2894 int ret; 2895 2896 cpus_read_lock(); 2897 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, XPS_CPUS); 2898 cpus_read_unlock(); 2899 2900 return ret; 2901 } 2902 EXPORT_SYMBOL(netif_set_xps_queue); 2903 2904 #endif 2905 static void netdev_unbind_all_sb_channels(struct net_device *dev) 2906 { 2907 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2908 2909 /* Unbind any subordinate channels */ 2910 while (txq-- != &dev->_tx[0]) { 2911 if (txq->sb_dev) 2912 netdev_unbind_sb_channel(dev, txq->sb_dev); 2913 } 2914 } 2915 2916 void netdev_reset_tc(struct net_device *dev) 2917 { 2918 #ifdef CONFIG_XPS 2919 netif_reset_xps_queues_gt(dev, 0); 2920 #endif 2921 netdev_unbind_all_sb_channels(dev); 2922 2923 /* Reset TC configuration of device */ 2924 dev->num_tc = 0; 2925 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq)); 2926 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map)); 2927 } 2928 EXPORT_SYMBOL(netdev_reset_tc); 2929 2930 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset) 2931 { 2932 if (tc >= dev->num_tc) 2933 return -EINVAL; 2934 2935 #ifdef CONFIG_XPS 2936 netif_reset_xps_queues(dev, offset, count); 2937 #endif 2938 dev->tc_to_txq[tc].count = count; 2939 dev->tc_to_txq[tc].offset = offset; 2940 return 0; 2941 } 2942 EXPORT_SYMBOL(netdev_set_tc_queue); 2943 2944 int netdev_set_num_tc(struct net_device *dev, u8 num_tc) 2945 { 2946 if (num_tc > TC_MAX_QUEUE) 2947 return -EINVAL; 2948 2949 #ifdef CONFIG_XPS 2950 netif_reset_xps_queues_gt(dev, 0); 2951 #endif 2952 netdev_unbind_all_sb_channels(dev); 2953 2954 dev->num_tc = num_tc; 2955 return 0; 2956 } 2957 EXPORT_SYMBOL(netdev_set_num_tc); 2958 2959 void netdev_unbind_sb_channel(struct net_device *dev, 2960 struct net_device *sb_dev) 2961 { 2962 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2963 2964 #ifdef CONFIG_XPS 2965 netif_reset_xps_queues_gt(sb_dev, 0); 2966 #endif 2967 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq)); 2968 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map)); 2969 2970 while (txq-- != &dev->_tx[0]) { 2971 if (txq->sb_dev == sb_dev) 2972 txq->sb_dev = NULL; 2973 } 2974 } 2975 EXPORT_SYMBOL(netdev_unbind_sb_channel); 2976 2977 int netdev_bind_sb_channel_queue(struct net_device *dev, 2978 struct net_device *sb_dev, 2979 u8 tc, u16 count, u16 offset) 2980 { 2981 /* Make certain the sb_dev and dev are already configured */ 2982 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc) 2983 return -EINVAL; 2984 2985 /* We cannot hand out queues we don't have */ 2986 if ((offset + count) > dev->real_num_tx_queues) 2987 return -EINVAL; 2988 2989 /* Record the mapping */ 2990 sb_dev->tc_to_txq[tc].count = count; 2991 sb_dev->tc_to_txq[tc].offset = offset; 2992 2993 /* Provide a way for Tx queue to find the tc_to_txq map or 2994 * XPS map for itself. 2995 */ 2996 while (count--) 2997 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev; 2998 2999 return 0; 3000 } 3001 EXPORT_SYMBOL(netdev_bind_sb_channel_queue); 3002 3003 int netdev_set_sb_channel(struct net_device *dev, u16 channel) 3004 { 3005 /* Do not use a multiqueue device to represent a subordinate channel */ 3006 if (netif_is_multiqueue(dev)) 3007 return -ENODEV; 3008 3009 /* We allow channels 1 - 32767 to be used for subordinate channels. 3010 * Channel 0 is meant to be "native" mode and used only to represent 3011 * the main root device. We allow writing 0 to reset the device back 3012 * to normal mode after being used as a subordinate channel. 3013 */ 3014 if (channel > S16_MAX) 3015 return -EINVAL; 3016 3017 dev->num_tc = -channel; 3018 3019 return 0; 3020 } 3021 EXPORT_SYMBOL(netdev_set_sb_channel); 3022 3023 /* 3024 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues 3025 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed. 3026 */ 3027 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) 3028 { 3029 bool disabling; 3030 int rc; 3031 3032 disabling = txq < dev->real_num_tx_queues; 3033 3034 if (txq < 1 || txq > dev->num_tx_queues) 3035 return -EINVAL; 3036 3037 if (dev->reg_state == NETREG_REGISTERED || 3038 dev->reg_state == NETREG_UNREGISTERING) { 3039 ASSERT_RTNL(); 3040 3041 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, 3042 txq); 3043 if (rc) 3044 return rc; 3045 3046 if (dev->num_tc) 3047 netif_setup_tc(dev, txq); 3048 3049 dev->real_num_tx_queues = txq; 3050 3051 if (disabling) { 3052 synchronize_net(); 3053 qdisc_reset_all_tx_gt(dev, txq); 3054 #ifdef CONFIG_XPS 3055 netif_reset_xps_queues_gt(dev, txq); 3056 #endif 3057 } 3058 } else { 3059 dev->real_num_tx_queues = txq; 3060 } 3061 3062 return 0; 3063 } 3064 EXPORT_SYMBOL(netif_set_real_num_tx_queues); 3065 3066 #ifdef CONFIG_SYSFS 3067 /** 3068 * netif_set_real_num_rx_queues - set actual number of RX queues used 3069 * @dev: Network device 3070 * @rxq: Actual number of RX queues 3071 * 3072 * This must be called either with the rtnl_lock held or before 3073 * registration of the net device. Returns 0 on success, or a 3074 * negative error code. If called before registration, it always 3075 * succeeds. 3076 */ 3077 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) 3078 { 3079 int rc; 3080 3081 if (rxq < 1 || rxq > dev->num_rx_queues) 3082 return -EINVAL; 3083 3084 if (dev->reg_state == NETREG_REGISTERED) { 3085 ASSERT_RTNL(); 3086 3087 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, 3088 rxq); 3089 if (rc) 3090 return rc; 3091 } 3092 3093 dev->real_num_rx_queues = rxq; 3094 return 0; 3095 } 3096 EXPORT_SYMBOL(netif_set_real_num_rx_queues); 3097 #endif 3098 3099 /** 3100 * netif_get_num_default_rss_queues - default number of RSS queues 3101 * 3102 * This routine should set an upper limit on the number of RSS queues 3103 * used by default by multiqueue devices. 3104 */ 3105 int netif_get_num_default_rss_queues(void) 3106 { 3107 return is_kdump_kernel() ? 3108 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus()); 3109 } 3110 EXPORT_SYMBOL(netif_get_num_default_rss_queues); 3111 3112 static void __netif_reschedule(struct Qdisc *q) 3113 { 3114 struct softnet_data *sd; 3115 unsigned long flags; 3116 3117 local_irq_save(flags); 3118 sd = this_cpu_ptr(&softnet_data); 3119 q->next_sched = NULL; 3120 *sd->output_queue_tailp = q; 3121 sd->output_queue_tailp = &q->next_sched; 3122 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3123 local_irq_restore(flags); 3124 } 3125 3126 void __netif_schedule(struct Qdisc *q) 3127 { 3128 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) 3129 __netif_reschedule(q); 3130 } 3131 EXPORT_SYMBOL(__netif_schedule); 3132 3133 struct dev_kfree_skb_cb { 3134 enum skb_free_reason reason; 3135 }; 3136 3137 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) 3138 { 3139 return (struct dev_kfree_skb_cb *)skb->cb; 3140 } 3141 3142 void netif_schedule_queue(struct netdev_queue *txq) 3143 { 3144 rcu_read_lock(); 3145 if (!netif_xmit_stopped(txq)) { 3146 struct Qdisc *q = rcu_dereference(txq->qdisc); 3147 3148 __netif_schedule(q); 3149 } 3150 rcu_read_unlock(); 3151 } 3152 EXPORT_SYMBOL(netif_schedule_queue); 3153 3154 void netif_tx_wake_queue(struct netdev_queue *dev_queue) 3155 { 3156 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { 3157 struct Qdisc *q; 3158 3159 rcu_read_lock(); 3160 q = rcu_dereference(dev_queue->qdisc); 3161 __netif_schedule(q); 3162 rcu_read_unlock(); 3163 } 3164 } 3165 EXPORT_SYMBOL(netif_tx_wake_queue); 3166 3167 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason) 3168 { 3169 unsigned long flags; 3170 3171 if (unlikely(!skb)) 3172 return; 3173 3174 if (likely(refcount_read(&skb->users) == 1)) { 3175 smp_rmb(); 3176 refcount_set(&skb->users, 0); 3177 } else if (likely(!refcount_dec_and_test(&skb->users))) { 3178 return; 3179 } 3180 get_kfree_skb_cb(skb)->reason = reason; 3181 local_irq_save(flags); 3182 skb->next = __this_cpu_read(softnet_data.completion_queue); 3183 __this_cpu_write(softnet_data.completion_queue, skb); 3184 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3185 local_irq_restore(flags); 3186 } 3187 EXPORT_SYMBOL(__dev_kfree_skb_irq); 3188 3189 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason) 3190 { 3191 if (in_irq() || irqs_disabled()) 3192 __dev_kfree_skb_irq(skb, reason); 3193 else 3194 dev_kfree_skb(skb); 3195 } 3196 EXPORT_SYMBOL(__dev_kfree_skb_any); 3197 3198 3199 /** 3200 * netif_device_detach - mark device as removed 3201 * @dev: network device 3202 * 3203 * Mark device as removed from system and therefore no longer available. 3204 */ 3205 void netif_device_detach(struct net_device *dev) 3206 { 3207 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && 3208 netif_running(dev)) { 3209 netif_tx_stop_all_queues(dev); 3210 } 3211 } 3212 EXPORT_SYMBOL(netif_device_detach); 3213 3214 /** 3215 * netif_device_attach - mark device as attached 3216 * @dev: network device 3217 * 3218 * Mark device as attached from system and restart if needed. 3219 */ 3220 void netif_device_attach(struct net_device *dev) 3221 { 3222 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && 3223 netif_running(dev)) { 3224 netif_tx_wake_all_queues(dev); 3225 __netdev_watchdog_up(dev); 3226 } 3227 } 3228 EXPORT_SYMBOL(netif_device_attach); 3229 3230 /* 3231 * Returns a Tx hash based on the given packet descriptor a Tx queues' number 3232 * to be used as a distribution range. 3233 */ 3234 static u16 skb_tx_hash(const struct net_device *dev, 3235 const struct net_device *sb_dev, 3236 struct sk_buff *skb) 3237 { 3238 u32 hash; 3239 u16 qoffset = 0; 3240 u16 qcount = dev->real_num_tx_queues; 3241 3242 if (dev->num_tc) { 3243 u8 tc = netdev_get_prio_tc_map(dev, skb->priority); 3244 3245 qoffset = sb_dev->tc_to_txq[tc].offset; 3246 qcount = sb_dev->tc_to_txq[tc].count; 3247 } 3248 3249 if (skb_rx_queue_recorded(skb)) { 3250 hash = skb_get_rx_queue(skb); 3251 if (hash >= qoffset) 3252 hash -= qoffset; 3253 while (unlikely(hash >= qcount)) 3254 hash -= qcount; 3255 return hash + qoffset; 3256 } 3257 3258 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; 3259 } 3260 3261 static void skb_warn_bad_offload(const struct sk_buff *skb) 3262 { 3263 static const netdev_features_t null_features; 3264 struct net_device *dev = skb->dev; 3265 const char *name = ""; 3266 3267 if (!net_ratelimit()) 3268 return; 3269 3270 if (dev) { 3271 if (dev->dev.parent) 3272 name = dev_driver_string(dev->dev.parent); 3273 else 3274 name = netdev_name(dev); 3275 } 3276 skb_dump(KERN_WARNING, skb, false); 3277 WARN(1, "%s: caps=(%pNF, %pNF)\n", 3278 name, dev ? &dev->features : &null_features, 3279 skb->sk ? &skb->sk->sk_route_caps : &null_features); 3280 } 3281 3282 /* 3283 * Invalidate hardware checksum when packet is to be mangled, and 3284 * complete checksum manually on outgoing path. 3285 */ 3286 int skb_checksum_help(struct sk_buff *skb) 3287 { 3288 __wsum csum; 3289 int ret = 0, offset; 3290 3291 if (skb->ip_summed == CHECKSUM_COMPLETE) 3292 goto out_set_summed; 3293 3294 if (unlikely(skb_is_gso(skb))) { 3295 skb_warn_bad_offload(skb); 3296 return -EINVAL; 3297 } 3298 3299 /* Before computing a checksum, we should make sure no frag could 3300 * be modified by an external entity : checksum could be wrong. 3301 */ 3302 if (skb_has_shared_frag(skb)) { 3303 ret = __skb_linearize(skb); 3304 if (ret) 3305 goto out; 3306 } 3307 3308 offset = skb_checksum_start_offset(skb); 3309 BUG_ON(offset >= skb_headlen(skb)); 3310 csum = skb_checksum(skb, offset, skb->len - offset, 0); 3311 3312 offset += skb->csum_offset; 3313 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); 3314 3315 ret = skb_ensure_writable(skb, offset + sizeof(__sum16)); 3316 if (ret) 3317 goto out; 3318 3319 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0; 3320 out_set_summed: 3321 skb->ip_summed = CHECKSUM_NONE; 3322 out: 3323 return ret; 3324 } 3325 EXPORT_SYMBOL(skb_checksum_help); 3326 3327 int skb_crc32c_csum_help(struct sk_buff *skb) 3328 { 3329 __le32 crc32c_csum; 3330 int ret = 0, offset, start; 3331 3332 if (skb->ip_summed != CHECKSUM_PARTIAL) 3333 goto out; 3334 3335 if (unlikely(skb_is_gso(skb))) 3336 goto out; 3337 3338 /* Before computing a checksum, we should make sure no frag could 3339 * be modified by an external entity : checksum could be wrong. 3340 */ 3341 if (unlikely(skb_has_shared_frag(skb))) { 3342 ret = __skb_linearize(skb); 3343 if (ret) 3344 goto out; 3345 } 3346 start = skb_checksum_start_offset(skb); 3347 offset = start + offsetof(struct sctphdr, checksum); 3348 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { 3349 ret = -EINVAL; 3350 goto out; 3351 } 3352 3353 ret = skb_ensure_writable(skb, offset + sizeof(__le32)); 3354 if (ret) 3355 goto out; 3356 3357 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start, 3358 skb->len - start, ~(__u32)0, 3359 crc32c_csum_stub)); 3360 *(__le32 *)(skb->data + offset) = crc32c_csum; 3361 skb->ip_summed = CHECKSUM_NONE; 3362 skb->csum_not_inet = 0; 3363 out: 3364 return ret; 3365 } 3366 3367 __be16 skb_network_protocol(struct sk_buff *skb, int *depth) 3368 { 3369 __be16 type = skb->protocol; 3370 3371 /* Tunnel gso handlers can set protocol to ethernet. */ 3372 if (type == htons(ETH_P_TEB)) { 3373 struct ethhdr *eth; 3374 3375 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) 3376 return 0; 3377 3378 eth = (struct ethhdr *)skb->data; 3379 type = eth->h_proto; 3380 } 3381 3382 return __vlan_get_protocol(skb, type, depth); 3383 } 3384 3385 /** 3386 * skb_mac_gso_segment - mac layer segmentation handler. 3387 * @skb: buffer to segment 3388 * @features: features for the output path (see dev->features) 3389 */ 3390 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb, 3391 netdev_features_t features) 3392 { 3393 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); 3394 struct packet_offload *ptype; 3395 int vlan_depth = skb->mac_len; 3396 __be16 type = skb_network_protocol(skb, &vlan_depth); 3397 3398 if (unlikely(!type)) 3399 return ERR_PTR(-EINVAL); 3400 3401 __skb_pull(skb, vlan_depth); 3402 3403 rcu_read_lock(); 3404 list_for_each_entry_rcu(ptype, &offload_base, list) { 3405 if (ptype->type == type && ptype->callbacks.gso_segment) { 3406 segs = ptype->callbacks.gso_segment(skb, features); 3407 break; 3408 } 3409 } 3410 rcu_read_unlock(); 3411 3412 __skb_push(skb, skb->data - skb_mac_header(skb)); 3413 3414 return segs; 3415 } 3416 EXPORT_SYMBOL(skb_mac_gso_segment); 3417 3418 3419 /* openvswitch calls this on rx path, so we need a different check. 3420 */ 3421 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path) 3422 { 3423 if (tx_path) 3424 return skb->ip_summed != CHECKSUM_PARTIAL && 3425 skb->ip_summed != CHECKSUM_UNNECESSARY; 3426 3427 return skb->ip_summed == CHECKSUM_NONE; 3428 } 3429 3430 /** 3431 * __skb_gso_segment - Perform segmentation on skb. 3432 * @skb: buffer to segment 3433 * @features: features for the output path (see dev->features) 3434 * @tx_path: whether it is called in TX path 3435 * 3436 * This function segments the given skb and returns a list of segments. 3437 * 3438 * It may return NULL if the skb requires no segmentation. This is 3439 * only possible when GSO is used for verifying header integrity. 3440 * 3441 * Segmentation preserves SKB_GSO_CB_OFFSET bytes of previous skb cb. 3442 */ 3443 struct sk_buff *__skb_gso_segment(struct sk_buff *skb, 3444 netdev_features_t features, bool tx_path) 3445 { 3446 struct sk_buff *segs; 3447 3448 if (unlikely(skb_needs_check(skb, tx_path))) { 3449 int err; 3450 3451 /* We're going to init ->check field in TCP or UDP header */ 3452 err = skb_cow_head(skb, 0); 3453 if (err < 0) 3454 return ERR_PTR(err); 3455 } 3456 3457 /* Only report GSO partial support if it will enable us to 3458 * support segmentation on this frame without needing additional 3459 * work. 3460 */ 3461 if (features & NETIF_F_GSO_PARTIAL) { 3462 netdev_features_t partial_features = NETIF_F_GSO_ROBUST; 3463 struct net_device *dev = skb->dev; 3464 3465 partial_features |= dev->features & dev->gso_partial_features; 3466 if (!skb_gso_ok(skb, features | partial_features)) 3467 features &= ~NETIF_F_GSO_PARTIAL; 3468 } 3469 3470 BUILD_BUG_ON(SKB_GSO_CB_OFFSET + 3471 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb)); 3472 3473 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb); 3474 SKB_GSO_CB(skb)->encap_level = 0; 3475 3476 skb_reset_mac_header(skb); 3477 skb_reset_mac_len(skb); 3478 3479 segs = skb_mac_gso_segment(skb, features); 3480 3481 if (segs != skb && unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs))) 3482 skb_warn_bad_offload(skb); 3483 3484 return segs; 3485 } 3486 EXPORT_SYMBOL(__skb_gso_segment); 3487 3488 /* Take action when hardware reception checksum errors are detected. */ 3489 #ifdef CONFIG_BUG 3490 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3491 { 3492 if (net_ratelimit()) { 3493 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>"); 3494 skb_dump(KERN_ERR, skb, true); 3495 dump_stack(); 3496 } 3497 } 3498 EXPORT_SYMBOL(netdev_rx_csum_fault); 3499 #endif 3500 3501 /* XXX: check that highmem exists at all on the given machine. */ 3502 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) 3503 { 3504 #ifdef CONFIG_HIGHMEM 3505 int i; 3506 3507 if (!(dev->features & NETIF_F_HIGHDMA)) { 3508 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 3509 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 3510 3511 if (PageHighMem(skb_frag_page(frag))) 3512 return 1; 3513 } 3514 } 3515 #endif 3516 return 0; 3517 } 3518 3519 /* If MPLS offload request, verify we are testing hardware MPLS features 3520 * instead of standard features for the netdev. 3521 */ 3522 #if IS_ENABLED(CONFIG_NET_MPLS_GSO) 3523 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3524 netdev_features_t features, 3525 __be16 type) 3526 { 3527 if (eth_p_mpls(type)) 3528 features &= skb->dev->mpls_features; 3529 3530 return features; 3531 } 3532 #else 3533 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3534 netdev_features_t features, 3535 __be16 type) 3536 { 3537 return features; 3538 } 3539 #endif 3540 3541 static netdev_features_t harmonize_features(struct sk_buff *skb, 3542 netdev_features_t features) 3543 { 3544 __be16 type; 3545 3546 type = skb_network_protocol(skb, NULL); 3547 features = net_mpls_features(skb, features, type); 3548 3549 if (skb->ip_summed != CHECKSUM_NONE && 3550 !can_checksum_protocol(features, type)) { 3551 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); 3552 } 3553 if (illegal_highdma(skb->dev, skb)) 3554 features &= ~NETIF_F_SG; 3555 3556 return features; 3557 } 3558 3559 netdev_features_t passthru_features_check(struct sk_buff *skb, 3560 struct net_device *dev, 3561 netdev_features_t features) 3562 { 3563 return features; 3564 } 3565 EXPORT_SYMBOL(passthru_features_check); 3566 3567 static netdev_features_t dflt_features_check(struct sk_buff *skb, 3568 struct net_device *dev, 3569 netdev_features_t features) 3570 { 3571 return vlan_features_check(skb, features); 3572 } 3573 3574 static netdev_features_t gso_features_check(const struct sk_buff *skb, 3575 struct net_device *dev, 3576 netdev_features_t features) 3577 { 3578 u16 gso_segs = skb_shinfo(skb)->gso_segs; 3579 3580 if (gso_segs > dev->gso_max_segs) 3581 return features & ~NETIF_F_GSO_MASK; 3582 3583 if (!skb_shinfo(skb)->gso_type) { 3584 skb_warn_bad_offload(skb); 3585 return features & ~NETIF_F_GSO_MASK; 3586 } 3587 3588 /* Support for GSO partial features requires software 3589 * intervention before we can actually process the packets 3590 * so we need to strip support for any partial features now 3591 * and we can pull them back in after we have partially 3592 * segmented the frame. 3593 */ 3594 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)) 3595 features &= ~dev->gso_partial_features; 3596 3597 /* Make sure to clear the IPv4 ID mangling feature if the 3598 * IPv4 header has the potential to be fragmented. 3599 */ 3600 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { 3601 struct iphdr *iph = skb->encapsulation ? 3602 inner_ip_hdr(skb) : ip_hdr(skb); 3603 3604 if (!(iph->frag_off & htons(IP_DF))) 3605 features &= ~NETIF_F_TSO_MANGLEID; 3606 } 3607 3608 return features; 3609 } 3610 3611 netdev_features_t netif_skb_features(struct sk_buff *skb) 3612 { 3613 struct net_device *dev = skb->dev; 3614 netdev_features_t features = dev->features; 3615 3616 if (skb_is_gso(skb)) 3617 features = gso_features_check(skb, dev, features); 3618 3619 /* If encapsulation offload request, verify we are testing 3620 * hardware encapsulation features instead of standard 3621 * features for the netdev 3622 */ 3623 if (skb->encapsulation) 3624 features &= dev->hw_enc_features; 3625 3626 if (skb_vlan_tagged(skb)) 3627 features = netdev_intersect_features(features, 3628 dev->vlan_features | 3629 NETIF_F_HW_VLAN_CTAG_TX | 3630 NETIF_F_HW_VLAN_STAG_TX); 3631 3632 if (dev->netdev_ops->ndo_features_check) 3633 features &= dev->netdev_ops->ndo_features_check(skb, dev, 3634 features); 3635 else 3636 features &= dflt_features_check(skb, dev, features); 3637 3638 return harmonize_features(skb, features); 3639 } 3640 EXPORT_SYMBOL(netif_skb_features); 3641 3642 static int xmit_one(struct sk_buff *skb, struct net_device *dev, 3643 struct netdev_queue *txq, bool more) 3644 { 3645 unsigned int len; 3646 int rc; 3647 3648 if (dev_nit_active(dev)) 3649 dev_queue_xmit_nit(skb, dev); 3650 3651 len = skb->len; 3652 PRANDOM_ADD_NOISE(skb, dev, txq, len + jiffies); 3653 trace_net_dev_start_xmit(skb, dev); 3654 rc = netdev_start_xmit(skb, dev, txq, more); 3655 trace_net_dev_xmit(skb, rc, dev, len); 3656 3657 return rc; 3658 } 3659 3660 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, 3661 struct netdev_queue *txq, int *ret) 3662 { 3663 struct sk_buff *skb = first; 3664 int rc = NETDEV_TX_OK; 3665 3666 while (skb) { 3667 struct sk_buff *next = skb->next; 3668 3669 skb_mark_not_on_list(skb); 3670 rc = xmit_one(skb, dev, txq, next != NULL); 3671 if (unlikely(!dev_xmit_complete(rc))) { 3672 skb->next = next; 3673 goto out; 3674 } 3675 3676 skb = next; 3677 if (netif_tx_queue_stopped(txq) && skb) { 3678 rc = NETDEV_TX_BUSY; 3679 break; 3680 } 3681 } 3682 3683 out: 3684 *ret = rc; 3685 return skb; 3686 } 3687 3688 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, 3689 netdev_features_t features) 3690 { 3691 if (skb_vlan_tag_present(skb) && 3692 !vlan_hw_offload_capable(features, skb->vlan_proto)) 3693 skb = __vlan_hwaccel_push_inside(skb); 3694 return skb; 3695 } 3696 3697 int skb_csum_hwoffload_help(struct sk_buff *skb, 3698 const netdev_features_t features) 3699 { 3700 if (unlikely(skb_csum_is_sctp(skb))) 3701 return !!(features & NETIF_F_SCTP_CRC) ? 0 : 3702 skb_crc32c_csum_help(skb); 3703 3704 if (features & NETIF_F_HW_CSUM) 3705 return 0; 3706 3707 if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) { 3708 switch (skb->csum_offset) { 3709 case offsetof(struct tcphdr, check): 3710 case offsetof(struct udphdr, check): 3711 return 0; 3712 } 3713 } 3714 3715 return skb_checksum_help(skb); 3716 } 3717 EXPORT_SYMBOL(skb_csum_hwoffload_help); 3718 3719 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again) 3720 { 3721 netdev_features_t features; 3722 3723 features = netif_skb_features(skb); 3724 skb = validate_xmit_vlan(skb, features); 3725 if (unlikely(!skb)) 3726 goto out_null; 3727 3728 skb = sk_validate_xmit_skb(skb, dev); 3729 if (unlikely(!skb)) 3730 goto out_null; 3731 3732 if (netif_needs_gso(skb, features)) { 3733 struct sk_buff *segs; 3734 3735 segs = skb_gso_segment(skb, features); 3736 if (IS_ERR(segs)) { 3737 goto out_kfree_skb; 3738 } else if (segs) { 3739 consume_skb(skb); 3740 skb = segs; 3741 } 3742 } else { 3743 if (skb_needs_linearize(skb, features) && 3744 __skb_linearize(skb)) 3745 goto out_kfree_skb; 3746 3747 /* If packet is not checksummed and device does not 3748 * support checksumming for this protocol, complete 3749 * checksumming here. 3750 */ 3751 if (skb->ip_summed == CHECKSUM_PARTIAL) { 3752 if (skb->encapsulation) 3753 skb_set_inner_transport_header(skb, 3754 skb_checksum_start_offset(skb)); 3755 else 3756 skb_set_transport_header(skb, 3757 skb_checksum_start_offset(skb)); 3758 if (skb_csum_hwoffload_help(skb, features)) 3759 goto out_kfree_skb; 3760 } 3761 } 3762 3763 skb = validate_xmit_xfrm(skb, features, again); 3764 3765 return skb; 3766 3767 out_kfree_skb: 3768 kfree_skb(skb); 3769 out_null: 3770 atomic_long_inc(&dev->tx_dropped); 3771 return NULL; 3772 } 3773 3774 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again) 3775 { 3776 struct sk_buff *next, *head = NULL, *tail; 3777 3778 for (; skb != NULL; skb = next) { 3779 next = skb->next; 3780 skb_mark_not_on_list(skb); 3781 3782 /* in case skb wont be segmented, point to itself */ 3783 skb->prev = skb; 3784 3785 skb = validate_xmit_skb(skb, dev, again); 3786 if (!skb) 3787 continue; 3788 3789 if (!head) 3790 head = skb; 3791 else 3792 tail->next = skb; 3793 /* If skb was segmented, skb->prev points to 3794 * the last segment. If not, it still contains skb. 3795 */ 3796 tail = skb->prev; 3797 } 3798 return head; 3799 } 3800 EXPORT_SYMBOL_GPL(validate_xmit_skb_list); 3801 3802 static void qdisc_pkt_len_init(struct sk_buff *skb) 3803 { 3804 const struct skb_shared_info *shinfo = skb_shinfo(skb); 3805 3806 qdisc_skb_cb(skb)->pkt_len = skb->len; 3807 3808 /* To get more precise estimation of bytes sent on wire, 3809 * we add to pkt_len the headers size of all segments 3810 */ 3811 if (shinfo->gso_size && skb_transport_header_was_set(skb)) { 3812 unsigned int hdr_len; 3813 u16 gso_segs = shinfo->gso_segs; 3814 3815 /* mac layer + network layer */ 3816 hdr_len = skb_transport_header(skb) - skb_mac_header(skb); 3817 3818 /* + transport layer */ 3819 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { 3820 const struct tcphdr *th; 3821 struct tcphdr _tcphdr; 3822 3823 th = skb_header_pointer(skb, skb_transport_offset(skb), 3824 sizeof(_tcphdr), &_tcphdr); 3825 if (likely(th)) 3826 hdr_len += __tcp_hdrlen(th); 3827 } else { 3828 struct udphdr _udphdr; 3829 3830 if (skb_header_pointer(skb, skb_transport_offset(skb), 3831 sizeof(_udphdr), &_udphdr)) 3832 hdr_len += sizeof(struct udphdr); 3833 } 3834 3835 if (shinfo->gso_type & SKB_GSO_DODGY) 3836 gso_segs = DIV_ROUND_UP(skb->len - hdr_len, 3837 shinfo->gso_size); 3838 3839 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; 3840 } 3841 } 3842 3843 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, 3844 struct net_device *dev, 3845 struct netdev_queue *txq) 3846 { 3847 spinlock_t *root_lock = qdisc_lock(q); 3848 struct sk_buff *to_free = NULL; 3849 bool contended; 3850 int rc; 3851 3852 qdisc_calculate_pkt_len(skb, q); 3853 3854 if (q->flags & TCQ_F_NOLOCK) { 3855 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK; 3856 qdisc_run(q); 3857 3858 if (unlikely(to_free)) 3859 kfree_skb_list(to_free); 3860 return rc; 3861 } 3862 3863 /* 3864 * Heuristic to force contended enqueues to serialize on a 3865 * separate lock before trying to get qdisc main lock. 3866 * This permits qdisc->running owner to get the lock more 3867 * often and dequeue packets faster. 3868 */ 3869 contended = qdisc_is_running(q); 3870 if (unlikely(contended)) 3871 spin_lock(&q->busylock); 3872 3873 spin_lock(root_lock); 3874 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { 3875 __qdisc_drop(skb, &to_free); 3876 rc = NET_XMIT_DROP; 3877 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && 3878 qdisc_run_begin(q)) { 3879 /* 3880 * This is a work-conserving queue; there are no old skbs 3881 * waiting to be sent out; and the qdisc is not running - 3882 * xmit the skb directly. 3883 */ 3884 3885 qdisc_bstats_update(q, skb); 3886 3887 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { 3888 if (unlikely(contended)) { 3889 spin_unlock(&q->busylock); 3890 contended = false; 3891 } 3892 __qdisc_run(q); 3893 } 3894 3895 qdisc_run_end(q); 3896 rc = NET_XMIT_SUCCESS; 3897 } else { 3898 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK; 3899 if (qdisc_run_begin(q)) { 3900 if (unlikely(contended)) { 3901 spin_unlock(&q->busylock); 3902 contended = false; 3903 } 3904 __qdisc_run(q); 3905 qdisc_run_end(q); 3906 } 3907 } 3908 spin_unlock(root_lock); 3909 if (unlikely(to_free)) 3910 kfree_skb_list(to_free); 3911 if (unlikely(contended)) 3912 spin_unlock(&q->busylock); 3913 return rc; 3914 } 3915 3916 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) 3917 static void skb_update_prio(struct sk_buff *skb) 3918 { 3919 const struct netprio_map *map; 3920 const struct sock *sk; 3921 unsigned int prioidx; 3922 3923 if (skb->priority) 3924 return; 3925 map = rcu_dereference_bh(skb->dev->priomap); 3926 if (!map) 3927 return; 3928 sk = skb_to_full_sk(skb); 3929 if (!sk) 3930 return; 3931 3932 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data); 3933 3934 if (prioidx < map->priomap_len) 3935 skb->priority = map->priomap[prioidx]; 3936 } 3937 #else 3938 #define skb_update_prio(skb) 3939 #endif 3940 3941 /** 3942 * dev_loopback_xmit - loop back @skb 3943 * @net: network namespace this loopback is happening in 3944 * @sk: sk needed to be a netfilter okfn 3945 * @skb: buffer to transmit 3946 */ 3947 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) 3948 { 3949 skb_reset_mac_header(skb); 3950 __skb_pull(skb, skb_network_offset(skb)); 3951 skb->pkt_type = PACKET_LOOPBACK; 3952 skb->ip_summed = CHECKSUM_UNNECESSARY; 3953 WARN_ON(!skb_dst(skb)); 3954 skb_dst_force(skb); 3955 netif_rx_ni(skb); 3956 return 0; 3957 } 3958 EXPORT_SYMBOL(dev_loopback_xmit); 3959 3960 #ifdef CONFIG_NET_EGRESS 3961 static struct sk_buff * 3962 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 3963 { 3964 struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress); 3965 struct tcf_result cl_res; 3966 3967 if (!miniq) 3968 return skb; 3969 3970 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */ 3971 qdisc_skb_cb(skb)->mru = 0; 3972 qdisc_skb_cb(skb)->post_ct = false; 3973 mini_qdisc_bstats_cpu_update(miniq, skb); 3974 3975 switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) { 3976 case TC_ACT_OK: 3977 case TC_ACT_RECLASSIFY: 3978 skb->tc_index = TC_H_MIN(cl_res.classid); 3979 break; 3980 case TC_ACT_SHOT: 3981 mini_qdisc_qstats_cpu_drop(miniq); 3982 *ret = NET_XMIT_DROP; 3983 kfree_skb(skb); 3984 return NULL; 3985 case TC_ACT_STOLEN: 3986 case TC_ACT_QUEUED: 3987 case TC_ACT_TRAP: 3988 *ret = NET_XMIT_SUCCESS; 3989 consume_skb(skb); 3990 return NULL; 3991 case TC_ACT_REDIRECT: 3992 /* No need to push/pop skb's mac_header here on egress! */ 3993 skb_do_redirect(skb); 3994 *ret = NET_XMIT_SUCCESS; 3995 return NULL; 3996 default: 3997 break; 3998 } 3999 4000 return skb; 4001 } 4002 #endif /* CONFIG_NET_EGRESS */ 4003 4004 #ifdef CONFIG_XPS 4005 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb, 4006 struct xps_dev_maps *dev_maps, unsigned int tci) 4007 { 4008 int tc = netdev_get_prio_tc_map(dev, skb->priority); 4009 struct xps_map *map; 4010 int queue_index = -1; 4011 4012 if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids) 4013 return queue_index; 4014 4015 tci *= dev_maps->num_tc; 4016 tci += tc; 4017 4018 map = rcu_dereference(dev_maps->attr_map[tci]); 4019 if (map) { 4020 if (map->len == 1) 4021 queue_index = map->queues[0]; 4022 else 4023 queue_index = map->queues[reciprocal_scale( 4024 skb_get_hash(skb), map->len)]; 4025 if (unlikely(queue_index >= dev->real_num_tx_queues)) 4026 queue_index = -1; 4027 } 4028 return queue_index; 4029 } 4030 #endif 4031 4032 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev, 4033 struct sk_buff *skb) 4034 { 4035 #ifdef CONFIG_XPS 4036 struct xps_dev_maps *dev_maps; 4037 struct sock *sk = skb->sk; 4038 int queue_index = -1; 4039 4040 if (!static_key_false(&xps_needed)) 4041 return -1; 4042 4043 rcu_read_lock(); 4044 if (!static_key_false(&xps_rxqs_needed)) 4045 goto get_cpus_map; 4046 4047 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]); 4048 if (dev_maps) { 4049 int tci = sk_rx_queue_get(sk); 4050 4051 if (tci >= 0) 4052 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4053 tci); 4054 } 4055 4056 get_cpus_map: 4057 if (queue_index < 0) { 4058 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]); 4059 if (dev_maps) { 4060 unsigned int tci = skb->sender_cpu - 1; 4061 4062 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4063 tci); 4064 } 4065 } 4066 rcu_read_unlock(); 4067 4068 return queue_index; 4069 #else 4070 return -1; 4071 #endif 4072 } 4073 4074 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb, 4075 struct net_device *sb_dev) 4076 { 4077 return 0; 4078 } 4079 EXPORT_SYMBOL(dev_pick_tx_zero); 4080 4081 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb, 4082 struct net_device *sb_dev) 4083 { 4084 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues; 4085 } 4086 EXPORT_SYMBOL(dev_pick_tx_cpu_id); 4087 4088 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb, 4089 struct net_device *sb_dev) 4090 { 4091 struct sock *sk = skb->sk; 4092 int queue_index = sk_tx_queue_get(sk); 4093 4094 sb_dev = sb_dev ? : dev; 4095 4096 if (queue_index < 0 || skb->ooo_okay || 4097 queue_index >= dev->real_num_tx_queues) { 4098 int new_index = get_xps_queue(dev, sb_dev, skb); 4099 4100 if (new_index < 0) 4101 new_index = skb_tx_hash(dev, sb_dev, skb); 4102 4103 if (queue_index != new_index && sk && 4104 sk_fullsock(sk) && 4105 rcu_access_pointer(sk->sk_dst_cache)) 4106 sk_tx_queue_set(sk, new_index); 4107 4108 queue_index = new_index; 4109 } 4110 4111 return queue_index; 4112 } 4113 EXPORT_SYMBOL(netdev_pick_tx); 4114 4115 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev, 4116 struct sk_buff *skb, 4117 struct net_device *sb_dev) 4118 { 4119 int queue_index = 0; 4120 4121 #ifdef CONFIG_XPS 4122 u32 sender_cpu = skb->sender_cpu - 1; 4123 4124 if (sender_cpu >= (u32)NR_CPUS) 4125 skb->sender_cpu = raw_smp_processor_id() + 1; 4126 #endif 4127 4128 if (dev->real_num_tx_queues != 1) { 4129 const struct net_device_ops *ops = dev->netdev_ops; 4130 4131 if (ops->ndo_select_queue) 4132 queue_index = ops->ndo_select_queue(dev, skb, sb_dev); 4133 else 4134 queue_index = netdev_pick_tx(dev, skb, sb_dev); 4135 4136 queue_index = netdev_cap_txqueue(dev, queue_index); 4137 } 4138 4139 skb_set_queue_mapping(skb, queue_index); 4140 return netdev_get_tx_queue(dev, queue_index); 4141 } 4142 4143 /** 4144 * __dev_queue_xmit - transmit a buffer 4145 * @skb: buffer to transmit 4146 * @sb_dev: suboordinate device used for L2 forwarding offload 4147 * 4148 * Queue a buffer for transmission to a network device. The caller must 4149 * have set the device and priority and built the buffer before calling 4150 * this function. The function can be called from an interrupt. 4151 * 4152 * A negative errno code is returned on a failure. A success does not 4153 * guarantee the frame will be transmitted as it may be dropped due 4154 * to congestion or traffic shaping. 4155 * 4156 * ----------------------------------------------------------------------------------- 4157 * I notice this method can also return errors from the queue disciplines, 4158 * including NET_XMIT_DROP, which is a positive value. So, errors can also 4159 * be positive. 4160 * 4161 * Regardless of the return value, the skb is consumed, so it is currently 4162 * difficult to retry a send to this method. (You can bump the ref count 4163 * before sending to hold a reference for retry if you are careful.) 4164 * 4165 * When calling this method, interrupts MUST be enabled. This is because 4166 * the BH enable code must have IRQs enabled so that it will not deadlock. 4167 * --BLG 4168 */ 4169 static int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev) 4170 { 4171 struct net_device *dev = skb->dev; 4172 struct netdev_queue *txq; 4173 struct Qdisc *q; 4174 int rc = -ENOMEM; 4175 bool again = false; 4176 4177 skb_reset_mac_header(skb); 4178 4179 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) 4180 __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED); 4181 4182 /* Disable soft irqs for various locks below. Also 4183 * stops preemption for RCU. 4184 */ 4185 rcu_read_lock_bh(); 4186 4187 skb_update_prio(skb); 4188 4189 qdisc_pkt_len_init(skb); 4190 #ifdef CONFIG_NET_CLS_ACT 4191 skb->tc_at_ingress = 0; 4192 # ifdef CONFIG_NET_EGRESS 4193 if (static_branch_unlikely(&egress_needed_key)) { 4194 skb = sch_handle_egress(skb, &rc, dev); 4195 if (!skb) 4196 goto out; 4197 } 4198 # endif 4199 #endif 4200 /* If device/qdisc don't need skb->dst, release it right now while 4201 * its hot in this cpu cache. 4202 */ 4203 if (dev->priv_flags & IFF_XMIT_DST_RELEASE) 4204 skb_dst_drop(skb); 4205 else 4206 skb_dst_force(skb); 4207 4208 txq = netdev_core_pick_tx(dev, skb, sb_dev); 4209 q = rcu_dereference_bh(txq->qdisc); 4210 4211 trace_net_dev_queue(skb); 4212 if (q->enqueue) { 4213 rc = __dev_xmit_skb(skb, q, dev, txq); 4214 goto out; 4215 } 4216 4217 /* The device has no queue. Common case for software devices: 4218 * loopback, all the sorts of tunnels... 4219 4220 * Really, it is unlikely that netif_tx_lock protection is necessary 4221 * here. (f.e. loopback and IP tunnels are clean ignoring statistics 4222 * counters.) 4223 * However, it is possible, that they rely on protection 4224 * made by us here. 4225 4226 * Check this and shot the lock. It is not prone from deadlocks. 4227 *Either shot noqueue qdisc, it is even simpler 8) 4228 */ 4229 if (dev->flags & IFF_UP) { 4230 int cpu = smp_processor_id(); /* ok because BHs are off */ 4231 4232 if (txq->xmit_lock_owner != cpu) { 4233 if (dev_xmit_recursion()) 4234 goto recursion_alert; 4235 4236 skb = validate_xmit_skb(skb, dev, &again); 4237 if (!skb) 4238 goto out; 4239 4240 PRANDOM_ADD_NOISE(skb, dev, txq, jiffies); 4241 HARD_TX_LOCK(dev, txq, cpu); 4242 4243 if (!netif_xmit_stopped(txq)) { 4244 dev_xmit_recursion_inc(); 4245 skb = dev_hard_start_xmit(skb, dev, txq, &rc); 4246 dev_xmit_recursion_dec(); 4247 if (dev_xmit_complete(rc)) { 4248 HARD_TX_UNLOCK(dev, txq); 4249 goto out; 4250 } 4251 } 4252 HARD_TX_UNLOCK(dev, txq); 4253 net_crit_ratelimited("Virtual device %s asks to queue packet!\n", 4254 dev->name); 4255 } else { 4256 /* Recursion is detected! It is possible, 4257 * unfortunately 4258 */ 4259 recursion_alert: 4260 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", 4261 dev->name); 4262 } 4263 } 4264 4265 rc = -ENETDOWN; 4266 rcu_read_unlock_bh(); 4267 4268 atomic_long_inc(&dev->tx_dropped); 4269 kfree_skb_list(skb); 4270 return rc; 4271 out: 4272 rcu_read_unlock_bh(); 4273 return rc; 4274 } 4275 4276 int dev_queue_xmit(struct sk_buff *skb) 4277 { 4278 return __dev_queue_xmit(skb, NULL); 4279 } 4280 EXPORT_SYMBOL(dev_queue_xmit); 4281 4282 int dev_queue_xmit_accel(struct sk_buff *skb, struct net_device *sb_dev) 4283 { 4284 return __dev_queue_xmit(skb, sb_dev); 4285 } 4286 EXPORT_SYMBOL(dev_queue_xmit_accel); 4287 4288 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id) 4289 { 4290 struct net_device *dev = skb->dev; 4291 struct sk_buff *orig_skb = skb; 4292 struct netdev_queue *txq; 4293 int ret = NETDEV_TX_BUSY; 4294 bool again = false; 4295 4296 if (unlikely(!netif_running(dev) || 4297 !netif_carrier_ok(dev))) 4298 goto drop; 4299 4300 skb = validate_xmit_skb_list(skb, dev, &again); 4301 if (skb != orig_skb) 4302 goto drop; 4303 4304 skb_set_queue_mapping(skb, queue_id); 4305 txq = skb_get_tx_queue(dev, skb); 4306 PRANDOM_ADD_NOISE(skb, dev, txq, jiffies); 4307 4308 local_bh_disable(); 4309 4310 dev_xmit_recursion_inc(); 4311 HARD_TX_LOCK(dev, txq, smp_processor_id()); 4312 if (!netif_xmit_frozen_or_drv_stopped(txq)) 4313 ret = netdev_start_xmit(skb, dev, txq, false); 4314 HARD_TX_UNLOCK(dev, txq); 4315 dev_xmit_recursion_dec(); 4316 4317 local_bh_enable(); 4318 return ret; 4319 drop: 4320 atomic_long_inc(&dev->tx_dropped); 4321 kfree_skb_list(skb); 4322 return NET_XMIT_DROP; 4323 } 4324 EXPORT_SYMBOL(__dev_direct_xmit); 4325 4326 /************************************************************************* 4327 * Receiver routines 4328 *************************************************************************/ 4329 4330 int netdev_max_backlog __read_mostly = 1000; 4331 EXPORT_SYMBOL(netdev_max_backlog); 4332 4333 int netdev_tstamp_prequeue __read_mostly = 1; 4334 int netdev_budget __read_mostly = 300; 4335 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */ 4336 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ; 4337 int weight_p __read_mostly = 64; /* old backlog weight */ 4338 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ 4339 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ 4340 int dev_rx_weight __read_mostly = 64; 4341 int dev_tx_weight __read_mostly = 64; 4342 /* Maximum number of GRO_NORMAL skbs to batch up for list-RX */ 4343 int gro_normal_batch __read_mostly = 8; 4344 4345 /* Called with irq disabled */ 4346 static inline void ____napi_schedule(struct softnet_data *sd, 4347 struct napi_struct *napi) 4348 { 4349 struct task_struct *thread; 4350 4351 if (test_bit(NAPI_STATE_THREADED, &napi->state)) { 4352 /* Paired with smp_mb__before_atomic() in 4353 * napi_enable()/dev_set_threaded(). 4354 * Use READ_ONCE() to guarantee a complete 4355 * read on napi->thread. Only call 4356 * wake_up_process() when it's not NULL. 4357 */ 4358 thread = READ_ONCE(napi->thread); 4359 if (thread) { 4360 /* Avoid doing set_bit() if the thread is in 4361 * INTERRUPTIBLE state, cause napi_thread_wait() 4362 * makes sure to proceed with napi polling 4363 * if the thread is explicitly woken from here. 4364 */ 4365 if (READ_ONCE(thread->state) != TASK_INTERRUPTIBLE) 4366 set_bit(NAPI_STATE_SCHED_THREADED, &napi->state); 4367 wake_up_process(thread); 4368 return; 4369 } 4370 } 4371 4372 list_add_tail(&napi->poll_list, &sd->poll_list); 4373 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4374 } 4375 4376 #ifdef CONFIG_RPS 4377 4378 /* One global table that all flow-based protocols share. */ 4379 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; 4380 EXPORT_SYMBOL(rps_sock_flow_table); 4381 u32 rps_cpu_mask __read_mostly; 4382 EXPORT_SYMBOL(rps_cpu_mask); 4383 4384 struct static_key_false rps_needed __read_mostly; 4385 EXPORT_SYMBOL(rps_needed); 4386 struct static_key_false rfs_needed __read_mostly; 4387 EXPORT_SYMBOL(rfs_needed); 4388 4389 static struct rps_dev_flow * 4390 set_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4391 struct rps_dev_flow *rflow, u16 next_cpu) 4392 { 4393 if (next_cpu < nr_cpu_ids) { 4394 #ifdef CONFIG_RFS_ACCEL 4395 struct netdev_rx_queue *rxqueue; 4396 struct rps_dev_flow_table *flow_table; 4397 struct rps_dev_flow *old_rflow; 4398 u32 flow_id; 4399 u16 rxq_index; 4400 int rc; 4401 4402 /* Should we steer this flow to a different hardware queue? */ 4403 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || 4404 !(dev->features & NETIF_F_NTUPLE)) 4405 goto out; 4406 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); 4407 if (rxq_index == skb_get_rx_queue(skb)) 4408 goto out; 4409 4410 rxqueue = dev->_rx + rxq_index; 4411 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4412 if (!flow_table) 4413 goto out; 4414 flow_id = skb_get_hash(skb) & flow_table->mask; 4415 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, 4416 rxq_index, flow_id); 4417 if (rc < 0) 4418 goto out; 4419 old_rflow = rflow; 4420 rflow = &flow_table->flows[flow_id]; 4421 rflow->filter = rc; 4422 if (old_rflow->filter == rflow->filter) 4423 old_rflow->filter = RPS_NO_FILTER; 4424 out: 4425 #endif 4426 rflow->last_qtail = 4427 per_cpu(softnet_data, next_cpu).input_queue_head; 4428 } 4429 4430 rflow->cpu = next_cpu; 4431 return rflow; 4432 } 4433 4434 /* 4435 * get_rps_cpu is called from netif_receive_skb and returns the target 4436 * CPU from the RPS map of the receiving queue for a given skb. 4437 * rcu_read_lock must be held on entry. 4438 */ 4439 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4440 struct rps_dev_flow **rflowp) 4441 { 4442 const struct rps_sock_flow_table *sock_flow_table; 4443 struct netdev_rx_queue *rxqueue = dev->_rx; 4444 struct rps_dev_flow_table *flow_table; 4445 struct rps_map *map; 4446 int cpu = -1; 4447 u32 tcpu; 4448 u32 hash; 4449 4450 if (skb_rx_queue_recorded(skb)) { 4451 u16 index = skb_get_rx_queue(skb); 4452 4453 if (unlikely(index >= dev->real_num_rx_queues)) { 4454 WARN_ONCE(dev->real_num_rx_queues > 1, 4455 "%s received packet on queue %u, but number " 4456 "of RX queues is %u\n", 4457 dev->name, index, dev->real_num_rx_queues); 4458 goto done; 4459 } 4460 rxqueue += index; 4461 } 4462 4463 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ 4464 4465 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4466 map = rcu_dereference(rxqueue->rps_map); 4467 if (!flow_table && !map) 4468 goto done; 4469 4470 skb_reset_network_header(skb); 4471 hash = skb_get_hash(skb); 4472 if (!hash) 4473 goto done; 4474 4475 sock_flow_table = rcu_dereference(rps_sock_flow_table); 4476 if (flow_table && sock_flow_table) { 4477 struct rps_dev_flow *rflow; 4478 u32 next_cpu; 4479 u32 ident; 4480 4481 /* First check into global flow table if there is a match */ 4482 ident = sock_flow_table->ents[hash & sock_flow_table->mask]; 4483 if ((ident ^ hash) & ~rps_cpu_mask) 4484 goto try_rps; 4485 4486 next_cpu = ident & rps_cpu_mask; 4487 4488 /* OK, now we know there is a match, 4489 * we can look at the local (per receive queue) flow table 4490 */ 4491 rflow = &flow_table->flows[hash & flow_table->mask]; 4492 tcpu = rflow->cpu; 4493 4494 /* 4495 * If the desired CPU (where last recvmsg was done) is 4496 * different from current CPU (one in the rx-queue flow 4497 * table entry), switch if one of the following holds: 4498 * - Current CPU is unset (>= nr_cpu_ids). 4499 * - Current CPU is offline. 4500 * - The current CPU's queue tail has advanced beyond the 4501 * last packet that was enqueued using this table entry. 4502 * This guarantees that all previous packets for the flow 4503 * have been dequeued, thus preserving in order delivery. 4504 */ 4505 if (unlikely(tcpu != next_cpu) && 4506 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || 4507 ((int)(per_cpu(softnet_data, tcpu).input_queue_head - 4508 rflow->last_qtail)) >= 0)) { 4509 tcpu = next_cpu; 4510 rflow = set_rps_cpu(dev, skb, rflow, next_cpu); 4511 } 4512 4513 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { 4514 *rflowp = rflow; 4515 cpu = tcpu; 4516 goto done; 4517 } 4518 } 4519 4520 try_rps: 4521 4522 if (map) { 4523 tcpu = map->cpus[reciprocal_scale(hash, map->len)]; 4524 if (cpu_online(tcpu)) { 4525 cpu = tcpu; 4526 goto done; 4527 } 4528 } 4529 4530 done: 4531 return cpu; 4532 } 4533 4534 #ifdef CONFIG_RFS_ACCEL 4535 4536 /** 4537 * rps_may_expire_flow - check whether an RFS hardware filter may be removed 4538 * @dev: Device on which the filter was set 4539 * @rxq_index: RX queue index 4540 * @flow_id: Flow ID passed to ndo_rx_flow_steer() 4541 * @filter_id: Filter ID returned by ndo_rx_flow_steer() 4542 * 4543 * Drivers that implement ndo_rx_flow_steer() should periodically call 4544 * this function for each installed filter and remove the filters for 4545 * which it returns %true. 4546 */ 4547 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, 4548 u32 flow_id, u16 filter_id) 4549 { 4550 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; 4551 struct rps_dev_flow_table *flow_table; 4552 struct rps_dev_flow *rflow; 4553 bool expire = true; 4554 unsigned int cpu; 4555 4556 rcu_read_lock(); 4557 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4558 if (flow_table && flow_id <= flow_table->mask) { 4559 rflow = &flow_table->flows[flow_id]; 4560 cpu = READ_ONCE(rflow->cpu); 4561 if (rflow->filter == filter_id && cpu < nr_cpu_ids && 4562 ((int)(per_cpu(softnet_data, cpu).input_queue_head - 4563 rflow->last_qtail) < 4564 (int)(10 * flow_table->mask))) 4565 expire = false; 4566 } 4567 rcu_read_unlock(); 4568 return expire; 4569 } 4570 EXPORT_SYMBOL(rps_may_expire_flow); 4571 4572 #endif /* CONFIG_RFS_ACCEL */ 4573 4574 /* Called from hardirq (IPI) context */ 4575 static void rps_trigger_softirq(void *data) 4576 { 4577 struct softnet_data *sd = data; 4578 4579 ____napi_schedule(sd, &sd->backlog); 4580 sd->received_rps++; 4581 } 4582 4583 #endif /* CONFIG_RPS */ 4584 4585 /* 4586 * Check if this softnet_data structure is another cpu one 4587 * If yes, queue it to our IPI list and return 1 4588 * If no, return 0 4589 */ 4590 static int rps_ipi_queued(struct softnet_data *sd) 4591 { 4592 #ifdef CONFIG_RPS 4593 struct softnet_data *mysd = this_cpu_ptr(&softnet_data); 4594 4595 if (sd != mysd) { 4596 sd->rps_ipi_next = mysd->rps_ipi_list; 4597 mysd->rps_ipi_list = sd; 4598 4599 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4600 return 1; 4601 } 4602 #endif /* CONFIG_RPS */ 4603 return 0; 4604 } 4605 4606 #ifdef CONFIG_NET_FLOW_LIMIT 4607 int netdev_flow_limit_table_len __read_mostly = (1 << 12); 4608 #endif 4609 4610 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) 4611 { 4612 #ifdef CONFIG_NET_FLOW_LIMIT 4613 struct sd_flow_limit *fl; 4614 struct softnet_data *sd; 4615 unsigned int old_flow, new_flow; 4616 4617 if (qlen < (netdev_max_backlog >> 1)) 4618 return false; 4619 4620 sd = this_cpu_ptr(&softnet_data); 4621 4622 rcu_read_lock(); 4623 fl = rcu_dereference(sd->flow_limit); 4624 if (fl) { 4625 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); 4626 old_flow = fl->history[fl->history_head]; 4627 fl->history[fl->history_head] = new_flow; 4628 4629 fl->history_head++; 4630 fl->history_head &= FLOW_LIMIT_HISTORY - 1; 4631 4632 if (likely(fl->buckets[old_flow])) 4633 fl->buckets[old_flow]--; 4634 4635 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { 4636 fl->count++; 4637 rcu_read_unlock(); 4638 return true; 4639 } 4640 } 4641 rcu_read_unlock(); 4642 #endif 4643 return false; 4644 } 4645 4646 /* 4647 * enqueue_to_backlog is called to queue an skb to a per CPU backlog 4648 * queue (may be a remote CPU queue). 4649 */ 4650 static int enqueue_to_backlog(struct sk_buff *skb, int cpu, 4651 unsigned int *qtail) 4652 { 4653 struct softnet_data *sd; 4654 unsigned long flags; 4655 unsigned int qlen; 4656 4657 sd = &per_cpu(softnet_data, cpu); 4658 4659 local_irq_save(flags); 4660 4661 rps_lock(sd); 4662 if (!netif_running(skb->dev)) 4663 goto drop; 4664 qlen = skb_queue_len(&sd->input_pkt_queue); 4665 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) { 4666 if (qlen) { 4667 enqueue: 4668 __skb_queue_tail(&sd->input_pkt_queue, skb); 4669 input_queue_tail_incr_save(sd, qtail); 4670 rps_unlock(sd); 4671 local_irq_restore(flags); 4672 return NET_RX_SUCCESS; 4673 } 4674 4675 /* Schedule NAPI for backlog device 4676 * We can use non atomic operation since we own the queue lock 4677 */ 4678 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) { 4679 if (!rps_ipi_queued(sd)) 4680 ____napi_schedule(sd, &sd->backlog); 4681 } 4682 goto enqueue; 4683 } 4684 4685 drop: 4686 sd->dropped++; 4687 rps_unlock(sd); 4688 4689 local_irq_restore(flags); 4690 4691 atomic_long_inc(&skb->dev->rx_dropped); 4692 kfree_skb(skb); 4693 return NET_RX_DROP; 4694 } 4695 4696 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb) 4697 { 4698 struct net_device *dev = skb->dev; 4699 struct netdev_rx_queue *rxqueue; 4700 4701 rxqueue = dev->_rx; 4702 4703 if (skb_rx_queue_recorded(skb)) { 4704 u16 index = skb_get_rx_queue(skb); 4705 4706 if (unlikely(index >= dev->real_num_rx_queues)) { 4707 WARN_ONCE(dev->real_num_rx_queues > 1, 4708 "%s received packet on queue %u, but number " 4709 "of RX queues is %u\n", 4710 dev->name, index, dev->real_num_rx_queues); 4711 4712 return rxqueue; /* Return first rxqueue */ 4713 } 4714 rxqueue += index; 4715 } 4716 return rxqueue; 4717 } 4718 4719 static u32 netif_receive_generic_xdp(struct sk_buff *skb, 4720 struct xdp_buff *xdp, 4721 struct bpf_prog *xdp_prog) 4722 { 4723 void *orig_data, *orig_data_end, *hard_start; 4724 struct netdev_rx_queue *rxqueue; 4725 u32 metalen, act = XDP_DROP; 4726 bool orig_bcast, orig_host; 4727 u32 mac_len, frame_sz; 4728 __be16 orig_eth_type; 4729 struct ethhdr *eth; 4730 int off; 4731 4732 /* Reinjected packets coming from act_mirred or similar should 4733 * not get XDP generic processing. 4734 */ 4735 if (skb_is_redirected(skb)) 4736 return XDP_PASS; 4737 4738 /* XDP packets must be linear and must have sufficient headroom 4739 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also 4740 * native XDP provides, thus we need to do it here as well. 4741 */ 4742 if (skb_cloned(skb) || skb_is_nonlinear(skb) || 4743 skb_headroom(skb) < XDP_PACKET_HEADROOM) { 4744 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb); 4745 int troom = skb->tail + skb->data_len - skb->end; 4746 4747 /* In case we have to go down the path and also linearize, 4748 * then lets do the pskb_expand_head() work just once here. 4749 */ 4750 if (pskb_expand_head(skb, 4751 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0, 4752 troom > 0 ? troom + 128 : 0, GFP_ATOMIC)) 4753 goto do_drop; 4754 if (skb_linearize(skb)) 4755 goto do_drop; 4756 } 4757 4758 /* The XDP program wants to see the packet starting at the MAC 4759 * header. 4760 */ 4761 mac_len = skb->data - skb_mac_header(skb); 4762 hard_start = skb->data - skb_headroom(skb); 4763 4764 /* SKB "head" area always have tailroom for skb_shared_info */ 4765 frame_sz = (void *)skb_end_pointer(skb) - hard_start; 4766 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 4767 4768 rxqueue = netif_get_rxqueue(skb); 4769 xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq); 4770 xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len, 4771 skb_headlen(skb) + mac_len, true); 4772 4773 orig_data_end = xdp->data_end; 4774 orig_data = xdp->data; 4775 eth = (struct ethhdr *)xdp->data; 4776 orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr); 4777 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest); 4778 orig_eth_type = eth->h_proto; 4779 4780 act = bpf_prog_run_xdp(xdp_prog, xdp); 4781 4782 /* check if bpf_xdp_adjust_head was used */ 4783 off = xdp->data - orig_data; 4784 if (off) { 4785 if (off > 0) 4786 __skb_pull(skb, off); 4787 else if (off < 0) 4788 __skb_push(skb, -off); 4789 4790 skb->mac_header += off; 4791 skb_reset_network_header(skb); 4792 } 4793 4794 /* check if bpf_xdp_adjust_tail was used */ 4795 off = xdp->data_end - orig_data_end; 4796 if (off != 0) { 4797 skb_set_tail_pointer(skb, xdp->data_end - xdp->data); 4798 skb->len += off; /* positive on grow, negative on shrink */ 4799 } 4800 4801 /* check if XDP changed eth hdr such SKB needs update */ 4802 eth = (struct ethhdr *)xdp->data; 4803 if ((orig_eth_type != eth->h_proto) || 4804 (orig_host != ether_addr_equal_64bits(eth->h_dest, 4805 skb->dev->dev_addr)) || 4806 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) { 4807 __skb_push(skb, ETH_HLEN); 4808 skb->pkt_type = PACKET_HOST; 4809 skb->protocol = eth_type_trans(skb, skb->dev); 4810 } 4811 4812 switch (act) { 4813 case XDP_REDIRECT: 4814 case XDP_TX: 4815 __skb_push(skb, mac_len); 4816 break; 4817 case XDP_PASS: 4818 metalen = xdp->data - xdp->data_meta; 4819 if (metalen) 4820 skb_metadata_set(skb, metalen); 4821 break; 4822 default: 4823 bpf_warn_invalid_xdp_action(act); 4824 fallthrough; 4825 case XDP_ABORTED: 4826 trace_xdp_exception(skb->dev, xdp_prog, act); 4827 fallthrough; 4828 case XDP_DROP: 4829 do_drop: 4830 kfree_skb(skb); 4831 break; 4832 } 4833 4834 return act; 4835 } 4836 4837 /* When doing generic XDP we have to bypass the qdisc layer and the 4838 * network taps in order to match in-driver-XDP behavior. 4839 */ 4840 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog) 4841 { 4842 struct net_device *dev = skb->dev; 4843 struct netdev_queue *txq; 4844 bool free_skb = true; 4845 int cpu, rc; 4846 4847 txq = netdev_core_pick_tx(dev, skb, NULL); 4848 cpu = smp_processor_id(); 4849 HARD_TX_LOCK(dev, txq, cpu); 4850 if (!netif_xmit_stopped(txq)) { 4851 rc = netdev_start_xmit(skb, dev, txq, 0); 4852 if (dev_xmit_complete(rc)) 4853 free_skb = false; 4854 } 4855 HARD_TX_UNLOCK(dev, txq); 4856 if (free_skb) { 4857 trace_xdp_exception(dev, xdp_prog, XDP_TX); 4858 kfree_skb(skb); 4859 } 4860 } 4861 4862 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key); 4863 4864 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb) 4865 { 4866 if (xdp_prog) { 4867 struct xdp_buff xdp; 4868 u32 act; 4869 int err; 4870 4871 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog); 4872 if (act != XDP_PASS) { 4873 switch (act) { 4874 case XDP_REDIRECT: 4875 err = xdp_do_generic_redirect(skb->dev, skb, 4876 &xdp, xdp_prog); 4877 if (err) 4878 goto out_redir; 4879 break; 4880 case XDP_TX: 4881 generic_xdp_tx(skb, xdp_prog); 4882 break; 4883 } 4884 return XDP_DROP; 4885 } 4886 } 4887 return XDP_PASS; 4888 out_redir: 4889 kfree_skb(skb); 4890 return XDP_DROP; 4891 } 4892 EXPORT_SYMBOL_GPL(do_xdp_generic); 4893 4894 static int netif_rx_internal(struct sk_buff *skb) 4895 { 4896 int ret; 4897 4898 net_timestamp_check(netdev_tstamp_prequeue, skb); 4899 4900 trace_netif_rx(skb); 4901 4902 #ifdef CONFIG_RPS 4903 if (static_branch_unlikely(&rps_needed)) { 4904 struct rps_dev_flow voidflow, *rflow = &voidflow; 4905 int cpu; 4906 4907 preempt_disable(); 4908 rcu_read_lock(); 4909 4910 cpu = get_rps_cpu(skb->dev, skb, &rflow); 4911 if (cpu < 0) 4912 cpu = smp_processor_id(); 4913 4914 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 4915 4916 rcu_read_unlock(); 4917 preempt_enable(); 4918 } else 4919 #endif 4920 { 4921 unsigned int qtail; 4922 4923 ret = enqueue_to_backlog(skb, get_cpu(), &qtail); 4924 put_cpu(); 4925 } 4926 return ret; 4927 } 4928 4929 /** 4930 * netif_rx - post buffer to the network code 4931 * @skb: buffer to post 4932 * 4933 * This function receives a packet from a device driver and queues it for 4934 * the upper (protocol) levels to process. It always succeeds. The buffer 4935 * may be dropped during processing for congestion control or by the 4936 * protocol layers. 4937 * 4938 * return values: 4939 * NET_RX_SUCCESS (no congestion) 4940 * NET_RX_DROP (packet was dropped) 4941 * 4942 */ 4943 4944 int netif_rx(struct sk_buff *skb) 4945 { 4946 int ret; 4947 4948 trace_netif_rx_entry(skb); 4949 4950 ret = netif_rx_internal(skb); 4951 trace_netif_rx_exit(ret); 4952 4953 return ret; 4954 } 4955 EXPORT_SYMBOL(netif_rx); 4956 4957 int netif_rx_ni(struct sk_buff *skb) 4958 { 4959 int err; 4960 4961 trace_netif_rx_ni_entry(skb); 4962 4963 preempt_disable(); 4964 err = netif_rx_internal(skb); 4965 if (local_softirq_pending()) 4966 do_softirq(); 4967 preempt_enable(); 4968 trace_netif_rx_ni_exit(err); 4969 4970 return err; 4971 } 4972 EXPORT_SYMBOL(netif_rx_ni); 4973 4974 int netif_rx_any_context(struct sk_buff *skb) 4975 { 4976 /* 4977 * If invoked from contexts which do not invoke bottom half 4978 * processing either at return from interrupt or when softrqs are 4979 * reenabled, use netif_rx_ni() which invokes bottomhalf processing 4980 * directly. 4981 */ 4982 if (in_interrupt()) 4983 return netif_rx(skb); 4984 else 4985 return netif_rx_ni(skb); 4986 } 4987 EXPORT_SYMBOL(netif_rx_any_context); 4988 4989 static __latent_entropy void net_tx_action(struct softirq_action *h) 4990 { 4991 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 4992 4993 if (sd->completion_queue) { 4994 struct sk_buff *clist; 4995 4996 local_irq_disable(); 4997 clist = sd->completion_queue; 4998 sd->completion_queue = NULL; 4999 local_irq_enable(); 5000 5001 while (clist) { 5002 struct sk_buff *skb = clist; 5003 5004 clist = clist->next; 5005 5006 WARN_ON(refcount_read(&skb->users)); 5007 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED)) 5008 trace_consume_skb(skb); 5009 else 5010 trace_kfree_skb(skb, net_tx_action); 5011 5012 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) 5013 __kfree_skb(skb); 5014 else 5015 __kfree_skb_defer(skb); 5016 } 5017 } 5018 5019 if (sd->output_queue) { 5020 struct Qdisc *head; 5021 5022 local_irq_disable(); 5023 head = sd->output_queue; 5024 sd->output_queue = NULL; 5025 sd->output_queue_tailp = &sd->output_queue; 5026 local_irq_enable(); 5027 5028 while (head) { 5029 struct Qdisc *q = head; 5030 spinlock_t *root_lock = NULL; 5031 5032 head = head->next_sched; 5033 5034 if (!(q->flags & TCQ_F_NOLOCK)) { 5035 root_lock = qdisc_lock(q); 5036 spin_lock(root_lock); 5037 } 5038 /* We need to make sure head->next_sched is read 5039 * before clearing __QDISC_STATE_SCHED 5040 */ 5041 smp_mb__before_atomic(); 5042 clear_bit(__QDISC_STATE_SCHED, &q->state); 5043 qdisc_run(q); 5044 if (root_lock) 5045 spin_unlock(root_lock); 5046 } 5047 } 5048 5049 xfrm_dev_backlog(sd); 5050 } 5051 5052 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) 5053 /* This hook is defined here for ATM LANE */ 5054 int (*br_fdb_test_addr_hook)(struct net_device *dev, 5055 unsigned char *addr) __read_mostly; 5056 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); 5057 #endif 5058 5059 static inline struct sk_buff * 5060 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 5061 struct net_device *orig_dev, bool *another) 5062 { 5063 #ifdef CONFIG_NET_CLS_ACT 5064 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress); 5065 struct tcf_result cl_res; 5066 5067 /* If there's at least one ingress present somewhere (so 5068 * we get here via enabled static key), remaining devices 5069 * that are not configured with an ingress qdisc will bail 5070 * out here. 5071 */ 5072 if (!miniq) 5073 return skb; 5074 5075 if (*pt_prev) { 5076 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5077 *pt_prev = NULL; 5078 } 5079 5080 qdisc_skb_cb(skb)->pkt_len = skb->len; 5081 qdisc_skb_cb(skb)->mru = 0; 5082 qdisc_skb_cb(skb)->post_ct = false; 5083 skb->tc_at_ingress = 1; 5084 mini_qdisc_bstats_cpu_update(miniq, skb); 5085 5086 switch (tcf_classify_ingress(skb, miniq->block, miniq->filter_list, 5087 &cl_res, false)) { 5088 case TC_ACT_OK: 5089 case TC_ACT_RECLASSIFY: 5090 skb->tc_index = TC_H_MIN(cl_res.classid); 5091 break; 5092 case TC_ACT_SHOT: 5093 mini_qdisc_qstats_cpu_drop(miniq); 5094 kfree_skb(skb); 5095 return NULL; 5096 case TC_ACT_STOLEN: 5097 case TC_ACT_QUEUED: 5098 case TC_ACT_TRAP: 5099 consume_skb(skb); 5100 return NULL; 5101 case TC_ACT_REDIRECT: 5102 /* skb_mac_header check was done by cls/act_bpf, so 5103 * we can safely push the L2 header back before 5104 * redirecting to another netdev 5105 */ 5106 __skb_push(skb, skb->mac_len); 5107 if (skb_do_redirect(skb) == -EAGAIN) { 5108 __skb_pull(skb, skb->mac_len); 5109 *another = true; 5110 break; 5111 } 5112 return NULL; 5113 case TC_ACT_CONSUMED: 5114 return NULL; 5115 default: 5116 break; 5117 } 5118 #endif /* CONFIG_NET_CLS_ACT */ 5119 return skb; 5120 } 5121 5122 /** 5123 * netdev_is_rx_handler_busy - check if receive handler is registered 5124 * @dev: device to check 5125 * 5126 * Check if a receive handler is already registered for a given device. 5127 * Return true if there one. 5128 * 5129 * The caller must hold the rtnl_mutex. 5130 */ 5131 bool netdev_is_rx_handler_busy(struct net_device *dev) 5132 { 5133 ASSERT_RTNL(); 5134 return dev && rtnl_dereference(dev->rx_handler); 5135 } 5136 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); 5137 5138 /** 5139 * netdev_rx_handler_register - register receive handler 5140 * @dev: device to register a handler for 5141 * @rx_handler: receive handler to register 5142 * @rx_handler_data: data pointer that is used by rx handler 5143 * 5144 * Register a receive handler for a device. This handler will then be 5145 * called from __netif_receive_skb. A negative errno code is returned 5146 * on a failure. 5147 * 5148 * The caller must hold the rtnl_mutex. 5149 * 5150 * For a general description of rx_handler, see enum rx_handler_result. 5151 */ 5152 int netdev_rx_handler_register(struct net_device *dev, 5153 rx_handler_func_t *rx_handler, 5154 void *rx_handler_data) 5155 { 5156 if (netdev_is_rx_handler_busy(dev)) 5157 return -EBUSY; 5158 5159 if (dev->priv_flags & IFF_NO_RX_HANDLER) 5160 return -EINVAL; 5161 5162 /* Note: rx_handler_data must be set before rx_handler */ 5163 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); 5164 rcu_assign_pointer(dev->rx_handler, rx_handler); 5165 5166 return 0; 5167 } 5168 EXPORT_SYMBOL_GPL(netdev_rx_handler_register); 5169 5170 /** 5171 * netdev_rx_handler_unregister - unregister receive handler 5172 * @dev: device to unregister a handler from 5173 * 5174 * Unregister a receive handler from a device. 5175 * 5176 * The caller must hold the rtnl_mutex. 5177 */ 5178 void netdev_rx_handler_unregister(struct net_device *dev) 5179 { 5180 5181 ASSERT_RTNL(); 5182 RCU_INIT_POINTER(dev->rx_handler, NULL); 5183 /* a reader seeing a non NULL rx_handler in a rcu_read_lock() 5184 * section has a guarantee to see a non NULL rx_handler_data 5185 * as well. 5186 */ 5187 synchronize_net(); 5188 RCU_INIT_POINTER(dev->rx_handler_data, NULL); 5189 } 5190 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); 5191 5192 /* 5193 * Limit the use of PFMEMALLOC reserves to those protocols that implement 5194 * the special handling of PFMEMALLOC skbs. 5195 */ 5196 static bool skb_pfmemalloc_protocol(struct sk_buff *skb) 5197 { 5198 switch (skb->protocol) { 5199 case htons(ETH_P_ARP): 5200 case htons(ETH_P_IP): 5201 case htons(ETH_P_IPV6): 5202 case htons(ETH_P_8021Q): 5203 case htons(ETH_P_8021AD): 5204 return true; 5205 default: 5206 return false; 5207 } 5208 } 5209 5210 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, 5211 int *ret, struct net_device *orig_dev) 5212 { 5213 if (nf_hook_ingress_active(skb)) { 5214 int ingress_retval; 5215 5216 if (*pt_prev) { 5217 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5218 *pt_prev = NULL; 5219 } 5220 5221 rcu_read_lock(); 5222 ingress_retval = nf_hook_ingress(skb); 5223 rcu_read_unlock(); 5224 return ingress_retval; 5225 } 5226 return 0; 5227 } 5228 5229 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc, 5230 struct packet_type **ppt_prev) 5231 { 5232 struct packet_type *ptype, *pt_prev; 5233 rx_handler_func_t *rx_handler; 5234 struct sk_buff *skb = *pskb; 5235 struct net_device *orig_dev; 5236 bool deliver_exact = false; 5237 int ret = NET_RX_DROP; 5238 __be16 type; 5239 5240 net_timestamp_check(!netdev_tstamp_prequeue, skb); 5241 5242 trace_netif_receive_skb(skb); 5243 5244 orig_dev = skb->dev; 5245 5246 skb_reset_network_header(skb); 5247 if (!skb_transport_header_was_set(skb)) 5248 skb_reset_transport_header(skb); 5249 skb_reset_mac_len(skb); 5250 5251 pt_prev = NULL; 5252 5253 another_round: 5254 skb->skb_iif = skb->dev->ifindex; 5255 5256 __this_cpu_inc(softnet_data.processed); 5257 5258 if (static_branch_unlikely(&generic_xdp_needed_key)) { 5259 int ret2; 5260 5261 preempt_disable(); 5262 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb); 5263 preempt_enable(); 5264 5265 if (ret2 != XDP_PASS) { 5266 ret = NET_RX_DROP; 5267 goto out; 5268 } 5269 skb_reset_mac_len(skb); 5270 } 5271 5272 if (eth_type_vlan(skb->protocol)) { 5273 skb = skb_vlan_untag(skb); 5274 if (unlikely(!skb)) 5275 goto out; 5276 } 5277 5278 if (skb_skip_tc_classify(skb)) 5279 goto skip_classify; 5280 5281 if (pfmemalloc) 5282 goto skip_taps; 5283 5284 list_for_each_entry_rcu(ptype, &ptype_all, list) { 5285 if (pt_prev) 5286 ret = deliver_skb(skb, pt_prev, orig_dev); 5287 pt_prev = ptype; 5288 } 5289 5290 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { 5291 if (pt_prev) 5292 ret = deliver_skb(skb, pt_prev, orig_dev); 5293 pt_prev = ptype; 5294 } 5295 5296 skip_taps: 5297 #ifdef CONFIG_NET_INGRESS 5298 if (static_branch_unlikely(&ingress_needed_key)) { 5299 bool another = false; 5300 5301 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev, 5302 &another); 5303 if (another) 5304 goto another_round; 5305 if (!skb) 5306 goto out; 5307 5308 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) 5309 goto out; 5310 } 5311 #endif 5312 skb_reset_redirect(skb); 5313 skip_classify: 5314 if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) 5315 goto drop; 5316 5317 if (skb_vlan_tag_present(skb)) { 5318 if (pt_prev) { 5319 ret = deliver_skb(skb, pt_prev, orig_dev); 5320 pt_prev = NULL; 5321 } 5322 if (vlan_do_receive(&skb)) 5323 goto another_round; 5324 else if (unlikely(!skb)) 5325 goto out; 5326 } 5327 5328 rx_handler = rcu_dereference(skb->dev->rx_handler); 5329 if (rx_handler) { 5330 if (pt_prev) { 5331 ret = deliver_skb(skb, pt_prev, orig_dev); 5332 pt_prev = NULL; 5333 } 5334 switch (rx_handler(&skb)) { 5335 case RX_HANDLER_CONSUMED: 5336 ret = NET_RX_SUCCESS; 5337 goto out; 5338 case RX_HANDLER_ANOTHER: 5339 goto another_round; 5340 case RX_HANDLER_EXACT: 5341 deliver_exact = true; 5342 break; 5343 case RX_HANDLER_PASS: 5344 break; 5345 default: 5346 BUG(); 5347 } 5348 } 5349 5350 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) { 5351 check_vlan_id: 5352 if (skb_vlan_tag_get_id(skb)) { 5353 /* Vlan id is non 0 and vlan_do_receive() above couldn't 5354 * find vlan device. 5355 */ 5356 skb->pkt_type = PACKET_OTHERHOST; 5357 } else if (eth_type_vlan(skb->protocol)) { 5358 /* Outer header is 802.1P with vlan 0, inner header is 5359 * 802.1Q or 802.1AD and vlan_do_receive() above could 5360 * not find vlan dev for vlan id 0. 5361 */ 5362 __vlan_hwaccel_clear_tag(skb); 5363 skb = skb_vlan_untag(skb); 5364 if (unlikely(!skb)) 5365 goto out; 5366 if (vlan_do_receive(&skb)) 5367 /* After stripping off 802.1P header with vlan 0 5368 * vlan dev is found for inner header. 5369 */ 5370 goto another_round; 5371 else if (unlikely(!skb)) 5372 goto out; 5373 else 5374 /* We have stripped outer 802.1P vlan 0 header. 5375 * But could not find vlan dev. 5376 * check again for vlan id to set OTHERHOST. 5377 */ 5378 goto check_vlan_id; 5379 } 5380 /* Note: we might in the future use prio bits 5381 * and set skb->priority like in vlan_do_receive() 5382 * For the time being, just ignore Priority Code Point 5383 */ 5384 __vlan_hwaccel_clear_tag(skb); 5385 } 5386 5387 type = skb->protocol; 5388 5389 /* deliver only exact match when indicated */ 5390 if (likely(!deliver_exact)) { 5391 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5392 &ptype_base[ntohs(type) & 5393 PTYPE_HASH_MASK]); 5394 } 5395 5396 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5397 &orig_dev->ptype_specific); 5398 5399 if (unlikely(skb->dev != orig_dev)) { 5400 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5401 &skb->dev->ptype_specific); 5402 } 5403 5404 if (pt_prev) { 5405 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 5406 goto drop; 5407 *ppt_prev = pt_prev; 5408 } else { 5409 drop: 5410 if (!deliver_exact) 5411 atomic_long_inc(&skb->dev->rx_dropped); 5412 else 5413 atomic_long_inc(&skb->dev->rx_nohandler); 5414 kfree_skb(skb); 5415 /* Jamal, now you will not able to escape explaining 5416 * me how you were going to use this. :-) 5417 */ 5418 ret = NET_RX_DROP; 5419 } 5420 5421 out: 5422 /* The invariant here is that if *ppt_prev is not NULL 5423 * then skb should also be non-NULL. 5424 * 5425 * Apparently *ppt_prev assignment above holds this invariant due to 5426 * skb dereferencing near it. 5427 */ 5428 *pskb = skb; 5429 return ret; 5430 } 5431 5432 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc) 5433 { 5434 struct net_device *orig_dev = skb->dev; 5435 struct packet_type *pt_prev = NULL; 5436 int ret; 5437 5438 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5439 if (pt_prev) 5440 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb, 5441 skb->dev, pt_prev, orig_dev); 5442 return ret; 5443 } 5444 5445 /** 5446 * netif_receive_skb_core - special purpose version of netif_receive_skb 5447 * @skb: buffer to process 5448 * 5449 * More direct receive version of netif_receive_skb(). It should 5450 * only be used by callers that have a need to skip RPS and Generic XDP. 5451 * Caller must also take care of handling if ``(page_is_)pfmemalloc``. 5452 * 5453 * This function may only be called from softirq context and interrupts 5454 * should be enabled. 5455 * 5456 * Return values (usually ignored): 5457 * NET_RX_SUCCESS: no congestion 5458 * NET_RX_DROP: packet was dropped 5459 */ 5460 int netif_receive_skb_core(struct sk_buff *skb) 5461 { 5462 int ret; 5463 5464 rcu_read_lock(); 5465 ret = __netif_receive_skb_one_core(skb, false); 5466 rcu_read_unlock(); 5467 5468 return ret; 5469 } 5470 EXPORT_SYMBOL(netif_receive_skb_core); 5471 5472 static inline void __netif_receive_skb_list_ptype(struct list_head *head, 5473 struct packet_type *pt_prev, 5474 struct net_device *orig_dev) 5475 { 5476 struct sk_buff *skb, *next; 5477 5478 if (!pt_prev) 5479 return; 5480 if (list_empty(head)) 5481 return; 5482 if (pt_prev->list_func != NULL) 5483 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv, 5484 ip_list_rcv, head, pt_prev, orig_dev); 5485 else 5486 list_for_each_entry_safe(skb, next, head, list) { 5487 skb_list_del_init(skb); 5488 pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 5489 } 5490 } 5491 5492 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc) 5493 { 5494 /* Fast-path assumptions: 5495 * - There is no RX handler. 5496 * - Only one packet_type matches. 5497 * If either of these fails, we will end up doing some per-packet 5498 * processing in-line, then handling the 'last ptype' for the whole 5499 * sublist. This can't cause out-of-order delivery to any single ptype, 5500 * because the 'last ptype' must be constant across the sublist, and all 5501 * other ptypes are handled per-packet. 5502 */ 5503 /* Current (common) ptype of sublist */ 5504 struct packet_type *pt_curr = NULL; 5505 /* Current (common) orig_dev of sublist */ 5506 struct net_device *od_curr = NULL; 5507 struct list_head sublist; 5508 struct sk_buff *skb, *next; 5509 5510 INIT_LIST_HEAD(&sublist); 5511 list_for_each_entry_safe(skb, next, head, list) { 5512 struct net_device *orig_dev = skb->dev; 5513 struct packet_type *pt_prev = NULL; 5514 5515 skb_list_del_init(skb); 5516 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5517 if (!pt_prev) 5518 continue; 5519 if (pt_curr != pt_prev || od_curr != orig_dev) { 5520 /* dispatch old sublist */ 5521 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5522 /* start new sublist */ 5523 INIT_LIST_HEAD(&sublist); 5524 pt_curr = pt_prev; 5525 od_curr = orig_dev; 5526 } 5527 list_add_tail(&skb->list, &sublist); 5528 } 5529 5530 /* dispatch final sublist */ 5531 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5532 } 5533 5534 static int __netif_receive_skb(struct sk_buff *skb) 5535 { 5536 int ret; 5537 5538 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { 5539 unsigned int noreclaim_flag; 5540 5541 /* 5542 * PFMEMALLOC skbs are special, they should 5543 * - be delivered to SOCK_MEMALLOC sockets only 5544 * - stay away from userspace 5545 * - have bounded memory usage 5546 * 5547 * Use PF_MEMALLOC as this saves us from propagating the allocation 5548 * context down to all allocation sites. 5549 */ 5550 noreclaim_flag = memalloc_noreclaim_save(); 5551 ret = __netif_receive_skb_one_core(skb, true); 5552 memalloc_noreclaim_restore(noreclaim_flag); 5553 } else 5554 ret = __netif_receive_skb_one_core(skb, false); 5555 5556 return ret; 5557 } 5558 5559 static void __netif_receive_skb_list(struct list_head *head) 5560 { 5561 unsigned long noreclaim_flag = 0; 5562 struct sk_buff *skb, *next; 5563 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */ 5564 5565 list_for_each_entry_safe(skb, next, head, list) { 5566 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) { 5567 struct list_head sublist; 5568 5569 /* Handle the previous sublist */ 5570 list_cut_before(&sublist, head, &skb->list); 5571 if (!list_empty(&sublist)) 5572 __netif_receive_skb_list_core(&sublist, pfmemalloc); 5573 pfmemalloc = !pfmemalloc; 5574 /* See comments in __netif_receive_skb */ 5575 if (pfmemalloc) 5576 noreclaim_flag = memalloc_noreclaim_save(); 5577 else 5578 memalloc_noreclaim_restore(noreclaim_flag); 5579 } 5580 } 5581 /* Handle the remaining sublist */ 5582 if (!list_empty(head)) 5583 __netif_receive_skb_list_core(head, pfmemalloc); 5584 /* Restore pflags */ 5585 if (pfmemalloc) 5586 memalloc_noreclaim_restore(noreclaim_flag); 5587 } 5588 5589 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp) 5590 { 5591 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog); 5592 struct bpf_prog *new = xdp->prog; 5593 int ret = 0; 5594 5595 if (new) { 5596 u32 i; 5597 5598 mutex_lock(&new->aux->used_maps_mutex); 5599 5600 /* generic XDP does not work with DEVMAPs that can 5601 * have a bpf_prog installed on an entry 5602 */ 5603 for (i = 0; i < new->aux->used_map_cnt; i++) { 5604 if (dev_map_can_have_prog(new->aux->used_maps[i]) || 5605 cpu_map_prog_allowed(new->aux->used_maps[i])) { 5606 mutex_unlock(&new->aux->used_maps_mutex); 5607 return -EINVAL; 5608 } 5609 } 5610 5611 mutex_unlock(&new->aux->used_maps_mutex); 5612 } 5613 5614 switch (xdp->command) { 5615 case XDP_SETUP_PROG: 5616 rcu_assign_pointer(dev->xdp_prog, new); 5617 if (old) 5618 bpf_prog_put(old); 5619 5620 if (old && !new) { 5621 static_branch_dec(&generic_xdp_needed_key); 5622 } else if (new && !old) { 5623 static_branch_inc(&generic_xdp_needed_key); 5624 dev_disable_lro(dev); 5625 dev_disable_gro_hw(dev); 5626 } 5627 break; 5628 5629 default: 5630 ret = -EINVAL; 5631 break; 5632 } 5633 5634 return ret; 5635 } 5636 5637 static int netif_receive_skb_internal(struct sk_buff *skb) 5638 { 5639 int ret; 5640 5641 net_timestamp_check(netdev_tstamp_prequeue, skb); 5642 5643 if (skb_defer_rx_timestamp(skb)) 5644 return NET_RX_SUCCESS; 5645 5646 rcu_read_lock(); 5647 #ifdef CONFIG_RPS 5648 if (static_branch_unlikely(&rps_needed)) { 5649 struct rps_dev_flow voidflow, *rflow = &voidflow; 5650 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5651 5652 if (cpu >= 0) { 5653 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5654 rcu_read_unlock(); 5655 return ret; 5656 } 5657 } 5658 #endif 5659 ret = __netif_receive_skb(skb); 5660 rcu_read_unlock(); 5661 return ret; 5662 } 5663 5664 static void netif_receive_skb_list_internal(struct list_head *head) 5665 { 5666 struct sk_buff *skb, *next; 5667 struct list_head sublist; 5668 5669 INIT_LIST_HEAD(&sublist); 5670 list_for_each_entry_safe(skb, next, head, list) { 5671 net_timestamp_check(netdev_tstamp_prequeue, skb); 5672 skb_list_del_init(skb); 5673 if (!skb_defer_rx_timestamp(skb)) 5674 list_add_tail(&skb->list, &sublist); 5675 } 5676 list_splice_init(&sublist, head); 5677 5678 rcu_read_lock(); 5679 #ifdef CONFIG_RPS 5680 if (static_branch_unlikely(&rps_needed)) { 5681 list_for_each_entry_safe(skb, next, head, list) { 5682 struct rps_dev_flow voidflow, *rflow = &voidflow; 5683 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5684 5685 if (cpu >= 0) { 5686 /* Will be handled, remove from list */ 5687 skb_list_del_init(skb); 5688 enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5689 } 5690 } 5691 } 5692 #endif 5693 __netif_receive_skb_list(head); 5694 rcu_read_unlock(); 5695 } 5696 5697 /** 5698 * netif_receive_skb - process receive buffer from network 5699 * @skb: buffer to process 5700 * 5701 * netif_receive_skb() is the main receive data processing function. 5702 * It always succeeds. The buffer may be dropped during processing 5703 * for congestion control or by the protocol layers. 5704 * 5705 * This function may only be called from softirq context and interrupts 5706 * should be enabled. 5707 * 5708 * Return values (usually ignored): 5709 * NET_RX_SUCCESS: no congestion 5710 * NET_RX_DROP: packet was dropped 5711 */ 5712 int netif_receive_skb(struct sk_buff *skb) 5713 { 5714 int ret; 5715 5716 trace_netif_receive_skb_entry(skb); 5717 5718 ret = netif_receive_skb_internal(skb); 5719 trace_netif_receive_skb_exit(ret); 5720 5721 return ret; 5722 } 5723 EXPORT_SYMBOL(netif_receive_skb); 5724 5725 /** 5726 * netif_receive_skb_list - process many receive buffers from network 5727 * @head: list of skbs to process. 5728 * 5729 * Since return value of netif_receive_skb() is normally ignored, and 5730 * wouldn't be meaningful for a list, this function returns void. 5731 * 5732 * This function may only be called from softirq context and interrupts 5733 * should be enabled. 5734 */ 5735 void netif_receive_skb_list(struct list_head *head) 5736 { 5737 struct sk_buff *skb; 5738 5739 if (list_empty(head)) 5740 return; 5741 if (trace_netif_receive_skb_list_entry_enabled()) { 5742 list_for_each_entry(skb, head, list) 5743 trace_netif_receive_skb_list_entry(skb); 5744 } 5745 netif_receive_skb_list_internal(head); 5746 trace_netif_receive_skb_list_exit(0); 5747 } 5748 EXPORT_SYMBOL(netif_receive_skb_list); 5749 5750 static DEFINE_PER_CPU(struct work_struct, flush_works); 5751 5752 /* Network device is going away, flush any packets still pending */ 5753 static void flush_backlog(struct work_struct *work) 5754 { 5755 struct sk_buff *skb, *tmp; 5756 struct softnet_data *sd; 5757 5758 local_bh_disable(); 5759 sd = this_cpu_ptr(&softnet_data); 5760 5761 local_irq_disable(); 5762 rps_lock(sd); 5763 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { 5764 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5765 __skb_unlink(skb, &sd->input_pkt_queue); 5766 dev_kfree_skb_irq(skb); 5767 input_queue_head_incr(sd); 5768 } 5769 } 5770 rps_unlock(sd); 5771 local_irq_enable(); 5772 5773 skb_queue_walk_safe(&sd->process_queue, skb, tmp) { 5774 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5775 __skb_unlink(skb, &sd->process_queue); 5776 kfree_skb(skb); 5777 input_queue_head_incr(sd); 5778 } 5779 } 5780 local_bh_enable(); 5781 } 5782 5783 static bool flush_required(int cpu) 5784 { 5785 #if IS_ENABLED(CONFIG_RPS) 5786 struct softnet_data *sd = &per_cpu(softnet_data, cpu); 5787 bool do_flush; 5788 5789 local_irq_disable(); 5790 rps_lock(sd); 5791 5792 /* as insertion into process_queue happens with the rps lock held, 5793 * process_queue access may race only with dequeue 5794 */ 5795 do_flush = !skb_queue_empty(&sd->input_pkt_queue) || 5796 !skb_queue_empty_lockless(&sd->process_queue); 5797 rps_unlock(sd); 5798 local_irq_enable(); 5799 5800 return do_flush; 5801 #endif 5802 /* without RPS we can't safely check input_pkt_queue: during a 5803 * concurrent remote skb_queue_splice() we can detect as empty both 5804 * input_pkt_queue and process_queue even if the latter could end-up 5805 * containing a lot of packets. 5806 */ 5807 return true; 5808 } 5809 5810 static void flush_all_backlogs(void) 5811 { 5812 static cpumask_t flush_cpus; 5813 unsigned int cpu; 5814 5815 /* since we are under rtnl lock protection we can use static data 5816 * for the cpumask and avoid allocating on stack the possibly 5817 * large mask 5818 */ 5819 ASSERT_RTNL(); 5820 5821 get_online_cpus(); 5822 5823 cpumask_clear(&flush_cpus); 5824 for_each_online_cpu(cpu) { 5825 if (flush_required(cpu)) { 5826 queue_work_on(cpu, system_highpri_wq, 5827 per_cpu_ptr(&flush_works, cpu)); 5828 cpumask_set_cpu(cpu, &flush_cpus); 5829 } 5830 } 5831 5832 /* we can have in flight packet[s] on the cpus we are not flushing, 5833 * synchronize_net() in unregister_netdevice_many() will take care of 5834 * them 5835 */ 5836 for_each_cpu(cpu, &flush_cpus) 5837 flush_work(per_cpu_ptr(&flush_works, cpu)); 5838 5839 put_online_cpus(); 5840 } 5841 5842 /* Pass the currently batched GRO_NORMAL SKBs up to the stack. */ 5843 static void gro_normal_list(struct napi_struct *napi) 5844 { 5845 if (!napi->rx_count) 5846 return; 5847 netif_receive_skb_list_internal(&napi->rx_list); 5848 INIT_LIST_HEAD(&napi->rx_list); 5849 napi->rx_count = 0; 5850 } 5851 5852 /* Queue one GRO_NORMAL SKB up for list processing. If batch size exceeded, 5853 * pass the whole batch up to the stack. 5854 */ 5855 static void gro_normal_one(struct napi_struct *napi, struct sk_buff *skb, int segs) 5856 { 5857 list_add_tail(&skb->list, &napi->rx_list); 5858 napi->rx_count += segs; 5859 if (napi->rx_count >= gro_normal_batch) 5860 gro_normal_list(napi); 5861 } 5862 5863 static int napi_gro_complete(struct napi_struct *napi, struct sk_buff *skb) 5864 { 5865 struct packet_offload *ptype; 5866 __be16 type = skb->protocol; 5867 struct list_head *head = &offload_base; 5868 int err = -ENOENT; 5869 5870 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb)); 5871 5872 if (NAPI_GRO_CB(skb)->count == 1) { 5873 skb_shinfo(skb)->gso_size = 0; 5874 goto out; 5875 } 5876 5877 rcu_read_lock(); 5878 list_for_each_entry_rcu(ptype, head, list) { 5879 if (ptype->type != type || !ptype->callbacks.gro_complete) 5880 continue; 5881 5882 err = INDIRECT_CALL_INET(ptype->callbacks.gro_complete, 5883 ipv6_gro_complete, inet_gro_complete, 5884 skb, 0); 5885 break; 5886 } 5887 rcu_read_unlock(); 5888 5889 if (err) { 5890 WARN_ON(&ptype->list == head); 5891 kfree_skb(skb); 5892 return NET_RX_SUCCESS; 5893 } 5894 5895 out: 5896 gro_normal_one(napi, skb, NAPI_GRO_CB(skb)->count); 5897 return NET_RX_SUCCESS; 5898 } 5899 5900 static void __napi_gro_flush_chain(struct napi_struct *napi, u32 index, 5901 bool flush_old) 5902 { 5903 struct list_head *head = &napi->gro_hash[index].list; 5904 struct sk_buff *skb, *p; 5905 5906 list_for_each_entry_safe_reverse(skb, p, head, list) { 5907 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies) 5908 return; 5909 skb_list_del_init(skb); 5910 napi_gro_complete(napi, skb); 5911 napi->gro_hash[index].count--; 5912 } 5913 5914 if (!napi->gro_hash[index].count) 5915 __clear_bit(index, &napi->gro_bitmask); 5916 } 5917 5918 /* napi->gro_hash[].list contains packets ordered by age. 5919 * youngest packets at the head of it. 5920 * Complete skbs in reverse order to reduce latencies. 5921 */ 5922 void napi_gro_flush(struct napi_struct *napi, bool flush_old) 5923 { 5924 unsigned long bitmask = napi->gro_bitmask; 5925 unsigned int i, base = ~0U; 5926 5927 while ((i = ffs(bitmask)) != 0) { 5928 bitmask >>= i; 5929 base += i; 5930 __napi_gro_flush_chain(napi, base, flush_old); 5931 } 5932 } 5933 EXPORT_SYMBOL(napi_gro_flush); 5934 5935 static void gro_list_prepare(const struct list_head *head, 5936 const struct sk_buff *skb) 5937 { 5938 unsigned int maclen = skb->dev->hard_header_len; 5939 u32 hash = skb_get_hash_raw(skb); 5940 struct sk_buff *p; 5941 5942 list_for_each_entry(p, head, list) { 5943 unsigned long diffs; 5944 5945 NAPI_GRO_CB(p)->flush = 0; 5946 5947 if (hash != skb_get_hash_raw(p)) { 5948 NAPI_GRO_CB(p)->same_flow = 0; 5949 continue; 5950 } 5951 5952 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev; 5953 diffs |= skb_vlan_tag_present(p) ^ skb_vlan_tag_present(skb); 5954 if (skb_vlan_tag_present(p)) 5955 diffs |= skb_vlan_tag_get(p) ^ skb_vlan_tag_get(skb); 5956 diffs |= skb_metadata_dst_cmp(p, skb); 5957 diffs |= skb_metadata_differs(p, skb); 5958 if (maclen == ETH_HLEN) 5959 diffs |= compare_ether_header(skb_mac_header(p), 5960 skb_mac_header(skb)); 5961 else if (!diffs) 5962 diffs = memcmp(skb_mac_header(p), 5963 skb_mac_header(skb), 5964 maclen); 5965 NAPI_GRO_CB(p)->same_flow = !diffs; 5966 } 5967 } 5968 5969 static inline void skb_gro_reset_offset(struct sk_buff *skb, u32 nhoff) 5970 { 5971 const struct skb_shared_info *pinfo = skb_shinfo(skb); 5972 const skb_frag_t *frag0 = &pinfo->frags[0]; 5973 5974 NAPI_GRO_CB(skb)->data_offset = 0; 5975 NAPI_GRO_CB(skb)->frag0 = NULL; 5976 NAPI_GRO_CB(skb)->frag0_len = 0; 5977 5978 if (!skb_headlen(skb) && pinfo->nr_frags && 5979 !PageHighMem(skb_frag_page(frag0)) && 5980 (!NET_IP_ALIGN || !((skb_frag_off(frag0) + nhoff) & 3))) { 5981 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0); 5982 NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int, 5983 skb_frag_size(frag0), 5984 skb->end - skb->tail); 5985 } 5986 } 5987 5988 static void gro_pull_from_frag0(struct sk_buff *skb, int grow) 5989 { 5990 struct skb_shared_info *pinfo = skb_shinfo(skb); 5991 5992 BUG_ON(skb->end - skb->tail < grow); 5993 5994 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); 5995 5996 skb->data_len -= grow; 5997 skb->tail += grow; 5998 5999 skb_frag_off_add(&pinfo->frags[0], grow); 6000 skb_frag_size_sub(&pinfo->frags[0], grow); 6001 6002 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) { 6003 skb_frag_unref(skb, 0); 6004 memmove(pinfo->frags, pinfo->frags + 1, 6005 --pinfo->nr_frags * sizeof(pinfo->frags[0])); 6006 } 6007 } 6008 6009 static void gro_flush_oldest(struct napi_struct *napi, struct list_head *head) 6010 { 6011 struct sk_buff *oldest; 6012 6013 oldest = list_last_entry(head, struct sk_buff, list); 6014 6015 /* We are called with head length >= MAX_GRO_SKBS, so this is 6016 * impossible. 6017 */ 6018 if (WARN_ON_ONCE(!oldest)) 6019 return; 6020 6021 /* Do not adjust napi->gro_hash[].count, caller is adding a new 6022 * SKB to the chain. 6023 */ 6024 skb_list_del_init(oldest); 6025 napi_gro_complete(napi, oldest); 6026 } 6027 6028 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) 6029 { 6030 u32 bucket = skb_get_hash_raw(skb) & (GRO_HASH_BUCKETS - 1); 6031 struct gro_list *gro_list = &napi->gro_hash[bucket]; 6032 struct list_head *head = &offload_base; 6033 struct packet_offload *ptype; 6034 __be16 type = skb->protocol; 6035 struct sk_buff *pp = NULL; 6036 enum gro_result ret; 6037 int same_flow; 6038 int grow; 6039 6040 if (netif_elide_gro(skb->dev)) 6041 goto normal; 6042 6043 gro_list_prepare(&gro_list->list, skb); 6044 6045 rcu_read_lock(); 6046 list_for_each_entry_rcu(ptype, head, list) { 6047 if (ptype->type != type || !ptype->callbacks.gro_receive) 6048 continue; 6049 6050 skb_set_network_header(skb, skb_gro_offset(skb)); 6051 skb_reset_mac_len(skb); 6052 NAPI_GRO_CB(skb)->same_flow = 0; 6053 NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb); 6054 NAPI_GRO_CB(skb)->free = 0; 6055 NAPI_GRO_CB(skb)->encap_mark = 0; 6056 NAPI_GRO_CB(skb)->recursion_counter = 0; 6057 NAPI_GRO_CB(skb)->is_fou = 0; 6058 NAPI_GRO_CB(skb)->is_atomic = 1; 6059 NAPI_GRO_CB(skb)->gro_remcsum_start = 0; 6060 6061 /* Setup for GRO checksum validation */ 6062 switch (skb->ip_summed) { 6063 case CHECKSUM_COMPLETE: 6064 NAPI_GRO_CB(skb)->csum = skb->csum; 6065 NAPI_GRO_CB(skb)->csum_valid = 1; 6066 NAPI_GRO_CB(skb)->csum_cnt = 0; 6067 break; 6068 case CHECKSUM_UNNECESSARY: 6069 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1; 6070 NAPI_GRO_CB(skb)->csum_valid = 0; 6071 break; 6072 default: 6073 NAPI_GRO_CB(skb)->csum_cnt = 0; 6074 NAPI_GRO_CB(skb)->csum_valid = 0; 6075 } 6076 6077 pp = INDIRECT_CALL_INET(ptype->callbacks.gro_receive, 6078 ipv6_gro_receive, inet_gro_receive, 6079 &gro_list->list, skb); 6080 break; 6081 } 6082 rcu_read_unlock(); 6083 6084 if (&ptype->list == head) 6085 goto normal; 6086 6087 if (PTR_ERR(pp) == -EINPROGRESS) { 6088 ret = GRO_CONSUMED; 6089 goto ok; 6090 } 6091 6092 same_flow = NAPI_GRO_CB(skb)->same_flow; 6093 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; 6094 6095 if (pp) { 6096 skb_list_del_init(pp); 6097 napi_gro_complete(napi, pp); 6098 gro_list->count--; 6099 } 6100 6101 if (same_flow) 6102 goto ok; 6103 6104 if (NAPI_GRO_CB(skb)->flush) 6105 goto normal; 6106 6107 if (unlikely(gro_list->count >= MAX_GRO_SKBS)) 6108 gro_flush_oldest(napi, &gro_list->list); 6109 else 6110 gro_list->count++; 6111 6112 NAPI_GRO_CB(skb)->count = 1; 6113 NAPI_GRO_CB(skb)->age = jiffies; 6114 NAPI_GRO_CB(skb)->last = skb; 6115 skb_shinfo(skb)->gso_size = skb_gro_len(skb); 6116 list_add(&skb->list, &gro_list->list); 6117 ret = GRO_HELD; 6118 6119 pull: 6120 grow = skb_gro_offset(skb) - skb_headlen(skb); 6121 if (grow > 0) 6122 gro_pull_from_frag0(skb, grow); 6123 ok: 6124 if (gro_list->count) { 6125 if (!test_bit(bucket, &napi->gro_bitmask)) 6126 __set_bit(bucket, &napi->gro_bitmask); 6127 } else if (test_bit(bucket, &napi->gro_bitmask)) { 6128 __clear_bit(bucket, &napi->gro_bitmask); 6129 } 6130 6131 return ret; 6132 6133 normal: 6134 ret = GRO_NORMAL; 6135 goto pull; 6136 } 6137 6138 struct packet_offload *gro_find_receive_by_type(__be16 type) 6139 { 6140 struct list_head *offload_head = &offload_base; 6141 struct packet_offload *ptype; 6142 6143 list_for_each_entry_rcu(ptype, offload_head, list) { 6144 if (ptype->type != type || !ptype->callbacks.gro_receive) 6145 continue; 6146 return ptype; 6147 } 6148 return NULL; 6149 } 6150 EXPORT_SYMBOL(gro_find_receive_by_type); 6151 6152 struct packet_offload *gro_find_complete_by_type(__be16 type) 6153 { 6154 struct list_head *offload_head = &offload_base; 6155 struct packet_offload *ptype; 6156 6157 list_for_each_entry_rcu(ptype, offload_head, list) { 6158 if (ptype->type != type || !ptype->callbacks.gro_complete) 6159 continue; 6160 return ptype; 6161 } 6162 return NULL; 6163 } 6164 EXPORT_SYMBOL(gro_find_complete_by_type); 6165 6166 static gro_result_t napi_skb_finish(struct napi_struct *napi, 6167 struct sk_buff *skb, 6168 gro_result_t ret) 6169 { 6170 switch (ret) { 6171 case GRO_NORMAL: 6172 gro_normal_one(napi, skb, 1); 6173 break; 6174 6175 case GRO_MERGED_FREE: 6176 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) 6177 napi_skb_free_stolen_head(skb); 6178 else 6179 __kfree_skb_defer(skb); 6180 break; 6181 6182 case GRO_HELD: 6183 case GRO_MERGED: 6184 case GRO_CONSUMED: 6185 break; 6186 } 6187 6188 return ret; 6189 } 6190 6191 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) 6192 { 6193 gro_result_t ret; 6194 6195 skb_mark_napi_id(skb, napi); 6196 trace_napi_gro_receive_entry(skb); 6197 6198 skb_gro_reset_offset(skb, 0); 6199 6200 ret = napi_skb_finish(napi, skb, dev_gro_receive(napi, skb)); 6201 trace_napi_gro_receive_exit(ret); 6202 6203 return ret; 6204 } 6205 EXPORT_SYMBOL(napi_gro_receive); 6206 6207 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) 6208 { 6209 if (unlikely(skb->pfmemalloc)) { 6210 consume_skb(skb); 6211 return; 6212 } 6213 __skb_pull(skb, skb_headlen(skb)); 6214 /* restore the reserve we had after netdev_alloc_skb_ip_align() */ 6215 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb)); 6216 __vlan_hwaccel_clear_tag(skb); 6217 skb->dev = napi->dev; 6218 skb->skb_iif = 0; 6219 6220 /* eth_type_trans() assumes pkt_type is PACKET_HOST */ 6221 skb->pkt_type = PACKET_HOST; 6222 6223 skb->encapsulation = 0; 6224 skb_shinfo(skb)->gso_type = 0; 6225 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); 6226 skb_ext_reset(skb); 6227 6228 napi->skb = skb; 6229 } 6230 6231 struct sk_buff *napi_get_frags(struct napi_struct *napi) 6232 { 6233 struct sk_buff *skb = napi->skb; 6234 6235 if (!skb) { 6236 skb = napi_alloc_skb(napi, GRO_MAX_HEAD); 6237 if (skb) { 6238 napi->skb = skb; 6239 skb_mark_napi_id(skb, napi); 6240 } 6241 } 6242 return skb; 6243 } 6244 EXPORT_SYMBOL(napi_get_frags); 6245 6246 static gro_result_t napi_frags_finish(struct napi_struct *napi, 6247 struct sk_buff *skb, 6248 gro_result_t ret) 6249 { 6250 switch (ret) { 6251 case GRO_NORMAL: 6252 case GRO_HELD: 6253 __skb_push(skb, ETH_HLEN); 6254 skb->protocol = eth_type_trans(skb, skb->dev); 6255 if (ret == GRO_NORMAL) 6256 gro_normal_one(napi, skb, 1); 6257 break; 6258 6259 case GRO_MERGED_FREE: 6260 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) 6261 napi_skb_free_stolen_head(skb); 6262 else 6263 napi_reuse_skb(napi, skb); 6264 break; 6265 6266 case GRO_MERGED: 6267 case GRO_CONSUMED: 6268 break; 6269 } 6270 6271 return ret; 6272 } 6273 6274 /* Upper GRO stack assumes network header starts at gro_offset=0 6275 * Drivers could call both napi_gro_frags() and napi_gro_receive() 6276 * We copy ethernet header into skb->data to have a common layout. 6277 */ 6278 static struct sk_buff *napi_frags_skb(struct napi_struct *napi) 6279 { 6280 struct sk_buff *skb = napi->skb; 6281 const struct ethhdr *eth; 6282 unsigned int hlen = sizeof(*eth); 6283 6284 napi->skb = NULL; 6285 6286 skb_reset_mac_header(skb); 6287 skb_gro_reset_offset(skb, hlen); 6288 6289 if (unlikely(skb_gro_header_hard(skb, hlen))) { 6290 eth = skb_gro_header_slow(skb, hlen, 0); 6291 if (unlikely(!eth)) { 6292 net_warn_ratelimited("%s: dropping impossible skb from %s\n", 6293 __func__, napi->dev->name); 6294 napi_reuse_skb(napi, skb); 6295 return NULL; 6296 } 6297 } else { 6298 eth = (const struct ethhdr *)skb->data; 6299 gro_pull_from_frag0(skb, hlen); 6300 NAPI_GRO_CB(skb)->frag0 += hlen; 6301 NAPI_GRO_CB(skb)->frag0_len -= hlen; 6302 } 6303 __skb_pull(skb, hlen); 6304 6305 /* 6306 * This works because the only protocols we care about don't require 6307 * special handling. 6308 * We'll fix it up properly in napi_frags_finish() 6309 */ 6310 skb->protocol = eth->h_proto; 6311 6312 return skb; 6313 } 6314 6315 gro_result_t napi_gro_frags(struct napi_struct *napi) 6316 { 6317 gro_result_t ret; 6318 struct sk_buff *skb = napi_frags_skb(napi); 6319 6320 trace_napi_gro_frags_entry(skb); 6321 6322 ret = napi_frags_finish(napi, skb, dev_gro_receive(napi, skb)); 6323 trace_napi_gro_frags_exit(ret); 6324 6325 return ret; 6326 } 6327 EXPORT_SYMBOL(napi_gro_frags); 6328 6329 /* Compute the checksum from gro_offset and return the folded value 6330 * after adding in any pseudo checksum. 6331 */ 6332 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb) 6333 { 6334 __wsum wsum; 6335 __sum16 sum; 6336 6337 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0); 6338 6339 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */ 6340 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum)); 6341 /* See comments in __skb_checksum_complete(). */ 6342 if (likely(!sum)) { 6343 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && 6344 !skb->csum_complete_sw) 6345 netdev_rx_csum_fault(skb->dev, skb); 6346 } 6347 6348 NAPI_GRO_CB(skb)->csum = wsum; 6349 NAPI_GRO_CB(skb)->csum_valid = 1; 6350 6351 return sum; 6352 } 6353 EXPORT_SYMBOL(__skb_gro_checksum_complete); 6354 6355 static void net_rps_send_ipi(struct softnet_data *remsd) 6356 { 6357 #ifdef CONFIG_RPS 6358 while (remsd) { 6359 struct softnet_data *next = remsd->rps_ipi_next; 6360 6361 if (cpu_online(remsd->cpu)) 6362 smp_call_function_single_async(remsd->cpu, &remsd->csd); 6363 remsd = next; 6364 } 6365 #endif 6366 } 6367 6368 /* 6369 * net_rps_action_and_irq_enable sends any pending IPI's for rps. 6370 * Note: called with local irq disabled, but exits with local irq enabled. 6371 */ 6372 static void net_rps_action_and_irq_enable(struct softnet_data *sd) 6373 { 6374 #ifdef CONFIG_RPS 6375 struct softnet_data *remsd = sd->rps_ipi_list; 6376 6377 if (remsd) { 6378 sd->rps_ipi_list = NULL; 6379 6380 local_irq_enable(); 6381 6382 /* Send pending IPI's to kick RPS processing on remote cpus. */ 6383 net_rps_send_ipi(remsd); 6384 } else 6385 #endif 6386 local_irq_enable(); 6387 } 6388 6389 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) 6390 { 6391 #ifdef CONFIG_RPS 6392 return sd->rps_ipi_list != NULL; 6393 #else 6394 return false; 6395 #endif 6396 } 6397 6398 static int process_backlog(struct napi_struct *napi, int quota) 6399 { 6400 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); 6401 bool again = true; 6402 int work = 0; 6403 6404 /* Check if we have pending ipi, its better to send them now, 6405 * not waiting net_rx_action() end. 6406 */ 6407 if (sd_has_rps_ipi_waiting(sd)) { 6408 local_irq_disable(); 6409 net_rps_action_and_irq_enable(sd); 6410 } 6411 6412 napi->weight = dev_rx_weight; 6413 while (again) { 6414 struct sk_buff *skb; 6415 6416 while ((skb = __skb_dequeue(&sd->process_queue))) { 6417 rcu_read_lock(); 6418 __netif_receive_skb(skb); 6419 rcu_read_unlock(); 6420 input_queue_head_incr(sd); 6421 if (++work >= quota) 6422 return work; 6423 6424 } 6425 6426 local_irq_disable(); 6427 rps_lock(sd); 6428 if (skb_queue_empty(&sd->input_pkt_queue)) { 6429 /* 6430 * Inline a custom version of __napi_complete(). 6431 * only current cpu owns and manipulates this napi, 6432 * and NAPI_STATE_SCHED is the only possible flag set 6433 * on backlog. 6434 * We can use a plain write instead of clear_bit(), 6435 * and we dont need an smp_mb() memory barrier. 6436 */ 6437 napi->state = 0; 6438 again = false; 6439 } else { 6440 skb_queue_splice_tail_init(&sd->input_pkt_queue, 6441 &sd->process_queue); 6442 } 6443 rps_unlock(sd); 6444 local_irq_enable(); 6445 } 6446 6447 return work; 6448 } 6449 6450 /** 6451 * __napi_schedule - schedule for receive 6452 * @n: entry to schedule 6453 * 6454 * The entry's receive function will be scheduled to run. 6455 * Consider using __napi_schedule_irqoff() if hard irqs are masked. 6456 */ 6457 void __napi_schedule(struct napi_struct *n) 6458 { 6459 unsigned long flags; 6460 6461 local_irq_save(flags); 6462 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6463 local_irq_restore(flags); 6464 } 6465 EXPORT_SYMBOL(__napi_schedule); 6466 6467 /** 6468 * napi_schedule_prep - check if napi can be scheduled 6469 * @n: napi context 6470 * 6471 * Test if NAPI routine is already running, and if not mark 6472 * it as running. This is used as a condition variable to 6473 * insure only one NAPI poll instance runs. We also make 6474 * sure there is no pending NAPI disable. 6475 */ 6476 bool napi_schedule_prep(struct napi_struct *n) 6477 { 6478 unsigned long val, new; 6479 6480 do { 6481 val = READ_ONCE(n->state); 6482 if (unlikely(val & NAPIF_STATE_DISABLE)) 6483 return false; 6484 new = val | NAPIF_STATE_SCHED; 6485 6486 /* Sets STATE_MISSED bit if STATE_SCHED was already set 6487 * This was suggested by Alexander Duyck, as compiler 6488 * emits better code than : 6489 * if (val & NAPIF_STATE_SCHED) 6490 * new |= NAPIF_STATE_MISSED; 6491 */ 6492 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * 6493 NAPIF_STATE_MISSED; 6494 } while (cmpxchg(&n->state, val, new) != val); 6495 6496 return !(val & NAPIF_STATE_SCHED); 6497 } 6498 EXPORT_SYMBOL(napi_schedule_prep); 6499 6500 /** 6501 * __napi_schedule_irqoff - schedule for receive 6502 * @n: entry to schedule 6503 * 6504 * Variant of __napi_schedule() assuming hard irqs are masked 6505 */ 6506 void __napi_schedule_irqoff(struct napi_struct *n) 6507 { 6508 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6509 } 6510 EXPORT_SYMBOL(__napi_schedule_irqoff); 6511 6512 bool napi_complete_done(struct napi_struct *n, int work_done) 6513 { 6514 unsigned long flags, val, new, timeout = 0; 6515 bool ret = true; 6516 6517 /* 6518 * 1) Don't let napi dequeue from the cpu poll list 6519 * just in case its running on a different cpu. 6520 * 2) If we are busy polling, do nothing here, we have 6521 * the guarantee we will be called later. 6522 */ 6523 if (unlikely(n->state & (NAPIF_STATE_NPSVC | 6524 NAPIF_STATE_IN_BUSY_POLL))) 6525 return false; 6526 6527 if (work_done) { 6528 if (n->gro_bitmask) 6529 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6530 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs); 6531 } 6532 if (n->defer_hard_irqs_count > 0) { 6533 n->defer_hard_irqs_count--; 6534 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6535 if (timeout) 6536 ret = false; 6537 } 6538 if (n->gro_bitmask) { 6539 /* When the NAPI instance uses a timeout and keeps postponing 6540 * it, we need to bound somehow the time packets are kept in 6541 * the GRO layer 6542 */ 6543 napi_gro_flush(n, !!timeout); 6544 } 6545 6546 gro_normal_list(n); 6547 6548 if (unlikely(!list_empty(&n->poll_list))) { 6549 /* If n->poll_list is not empty, we need to mask irqs */ 6550 local_irq_save(flags); 6551 list_del_init(&n->poll_list); 6552 local_irq_restore(flags); 6553 } 6554 6555 do { 6556 val = READ_ONCE(n->state); 6557 6558 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); 6559 6560 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED | 6561 NAPIF_STATE_SCHED_THREADED | 6562 NAPIF_STATE_PREFER_BUSY_POLL); 6563 6564 /* If STATE_MISSED was set, leave STATE_SCHED set, 6565 * because we will call napi->poll() one more time. 6566 * This C code was suggested by Alexander Duyck to help gcc. 6567 */ 6568 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * 6569 NAPIF_STATE_SCHED; 6570 } while (cmpxchg(&n->state, val, new) != val); 6571 6572 if (unlikely(val & NAPIF_STATE_MISSED)) { 6573 __napi_schedule(n); 6574 return false; 6575 } 6576 6577 if (timeout) 6578 hrtimer_start(&n->timer, ns_to_ktime(timeout), 6579 HRTIMER_MODE_REL_PINNED); 6580 return ret; 6581 } 6582 EXPORT_SYMBOL(napi_complete_done); 6583 6584 /* must be called under rcu_read_lock(), as we dont take a reference */ 6585 static struct napi_struct *napi_by_id(unsigned int napi_id) 6586 { 6587 unsigned int hash = napi_id % HASH_SIZE(napi_hash); 6588 struct napi_struct *napi; 6589 6590 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) 6591 if (napi->napi_id == napi_id) 6592 return napi; 6593 6594 return NULL; 6595 } 6596 6597 #if defined(CONFIG_NET_RX_BUSY_POLL) 6598 6599 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule) 6600 { 6601 if (!skip_schedule) { 6602 gro_normal_list(napi); 6603 __napi_schedule(napi); 6604 return; 6605 } 6606 6607 if (napi->gro_bitmask) { 6608 /* flush too old packets 6609 * If HZ < 1000, flush all packets. 6610 */ 6611 napi_gro_flush(napi, HZ >= 1000); 6612 } 6613 6614 gro_normal_list(napi); 6615 clear_bit(NAPI_STATE_SCHED, &napi->state); 6616 } 6617 6618 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll, 6619 u16 budget) 6620 { 6621 bool skip_schedule = false; 6622 unsigned long timeout; 6623 int rc; 6624 6625 /* Busy polling means there is a high chance device driver hard irq 6626 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was 6627 * set in napi_schedule_prep(). 6628 * Since we are about to call napi->poll() once more, we can safely 6629 * clear NAPI_STATE_MISSED. 6630 * 6631 * Note: x86 could use a single "lock and ..." instruction 6632 * to perform these two clear_bit() 6633 */ 6634 clear_bit(NAPI_STATE_MISSED, &napi->state); 6635 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); 6636 6637 local_bh_disable(); 6638 6639 if (prefer_busy_poll) { 6640 napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs); 6641 timeout = READ_ONCE(napi->dev->gro_flush_timeout); 6642 if (napi->defer_hard_irqs_count && timeout) { 6643 hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED); 6644 skip_schedule = true; 6645 } 6646 } 6647 6648 /* All we really want here is to re-enable device interrupts. 6649 * Ideally, a new ndo_busy_poll_stop() could avoid another round. 6650 */ 6651 rc = napi->poll(napi, budget); 6652 /* We can't gro_normal_list() here, because napi->poll() might have 6653 * rearmed the napi (napi_complete_done()) in which case it could 6654 * already be running on another CPU. 6655 */ 6656 trace_napi_poll(napi, rc, budget); 6657 netpoll_poll_unlock(have_poll_lock); 6658 if (rc == budget) 6659 __busy_poll_stop(napi, skip_schedule); 6660 local_bh_enable(); 6661 } 6662 6663 void napi_busy_loop(unsigned int napi_id, 6664 bool (*loop_end)(void *, unsigned long), 6665 void *loop_end_arg, bool prefer_busy_poll, u16 budget) 6666 { 6667 unsigned long start_time = loop_end ? busy_loop_current_time() : 0; 6668 int (*napi_poll)(struct napi_struct *napi, int budget); 6669 void *have_poll_lock = NULL; 6670 struct napi_struct *napi; 6671 6672 restart: 6673 napi_poll = NULL; 6674 6675 rcu_read_lock(); 6676 6677 napi = napi_by_id(napi_id); 6678 if (!napi) 6679 goto out; 6680 6681 preempt_disable(); 6682 for (;;) { 6683 int work = 0; 6684 6685 local_bh_disable(); 6686 if (!napi_poll) { 6687 unsigned long val = READ_ONCE(napi->state); 6688 6689 /* If multiple threads are competing for this napi, 6690 * we avoid dirtying napi->state as much as we can. 6691 */ 6692 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | 6693 NAPIF_STATE_IN_BUSY_POLL)) { 6694 if (prefer_busy_poll) 6695 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6696 goto count; 6697 } 6698 if (cmpxchg(&napi->state, val, 6699 val | NAPIF_STATE_IN_BUSY_POLL | 6700 NAPIF_STATE_SCHED) != val) { 6701 if (prefer_busy_poll) 6702 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6703 goto count; 6704 } 6705 have_poll_lock = netpoll_poll_lock(napi); 6706 napi_poll = napi->poll; 6707 } 6708 work = napi_poll(napi, budget); 6709 trace_napi_poll(napi, work, budget); 6710 gro_normal_list(napi); 6711 count: 6712 if (work > 0) 6713 __NET_ADD_STATS(dev_net(napi->dev), 6714 LINUX_MIB_BUSYPOLLRXPACKETS, work); 6715 local_bh_enable(); 6716 6717 if (!loop_end || loop_end(loop_end_arg, start_time)) 6718 break; 6719 6720 if (unlikely(need_resched())) { 6721 if (napi_poll) 6722 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6723 preempt_enable(); 6724 rcu_read_unlock(); 6725 cond_resched(); 6726 if (loop_end(loop_end_arg, start_time)) 6727 return; 6728 goto restart; 6729 } 6730 cpu_relax(); 6731 } 6732 if (napi_poll) 6733 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6734 preempt_enable(); 6735 out: 6736 rcu_read_unlock(); 6737 } 6738 EXPORT_SYMBOL(napi_busy_loop); 6739 6740 #endif /* CONFIG_NET_RX_BUSY_POLL */ 6741 6742 static void napi_hash_add(struct napi_struct *napi) 6743 { 6744 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state)) 6745 return; 6746 6747 spin_lock(&napi_hash_lock); 6748 6749 /* 0..NR_CPUS range is reserved for sender_cpu use */ 6750 do { 6751 if (unlikely(++napi_gen_id < MIN_NAPI_ID)) 6752 napi_gen_id = MIN_NAPI_ID; 6753 } while (napi_by_id(napi_gen_id)); 6754 napi->napi_id = napi_gen_id; 6755 6756 hlist_add_head_rcu(&napi->napi_hash_node, 6757 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); 6758 6759 spin_unlock(&napi_hash_lock); 6760 } 6761 6762 /* Warning : caller is responsible to make sure rcu grace period 6763 * is respected before freeing memory containing @napi 6764 */ 6765 static void napi_hash_del(struct napi_struct *napi) 6766 { 6767 spin_lock(&napi_hash_lock); 6768 6769 hlist_del_init_rcu(&napi->napi_hash_node); 6770 6771 spin_unlock(&napi_hash_lock); 6772 } 6773 6774 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) 6775 { 6776 struct napi_struct *napi; 6777 6778 napi = container_of(timer, struct napi_struct, timer); 6779 6780 /* Note : we use a relaxed variant of napi_schedule_prep() not setting 6781 * NAPI_STATE_MISSED, since we do not react to a device IRQ. 6782 */ 6783 if (!napi_disable_pending(napi) && 6784 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) { 6785 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6786 __napi_schedule_irqoff(napi); 6787 } 6788 6789 return HRTIMER_NORESTART; 6790 } 6791 6792 static void init_gro_hash(struct napi_struct *napi) 6793 { 6794 int i; 6795 6796 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6797 INIT_LIST_HEAD(&napi->gro_hash[i].list); 6798 napi->gro_hash[i].count = 0; 6799 } 6800 napi->gro_bitmask = 0; 6801 } 6802 6803 int dev_set_threaded(struct net_device *dev, bool threaded) 6804 { 6805 struct napi_struct *napi; 6806 int err = 0; 6807 6808 if (dev->threaded == threaded) 6809 return 0; 6810 6811 if (threaded) { 6812 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6813 if (!napi->thread) { 6814 err = napi_kthread_create(napi); 6815 if (err) { 6816 threaded = false; 6817 break; 6818 } 6819 } 6820 } 6821 } 6822 6823 dev->threaded = threaded; 6824 6825 /* Make sure kthread is created before THREADED bit 6826 * is set. 6827 */ 6828 smp_mb__before_atomic(); 6829 6830 /* Setting/unsetting threaded mode on a napi might not immediately 6831 * take effect, if the current napi instance is actively being 6832 * polled. In this case, the switch between threaded mode and 6833 * softirq mode will happen in the next round of napi_schedule(). 6834 * This should not cause hiccups/stalls to the live traffic. 6835 */ 6836 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6837 if (threaded) 6838 set_bit(NAPI_STATE_THREADED, &napi->state); 6839 else 6840 clear_bit(NAPI_STATE_THREADED, &napi->state); 6841 } 6842 6843 return err; 6844 } 6845 EXPORT_SYMBOL(dev_set_threaded); 6846 6847 void netif_napi_add(struct net_device *dev, struct napi_struct *napi, 6848 int (*poll)(struct napi_struct *, int), int weight) 6849 { 6850 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state))) 6851 return; 6852 6853 INIT_LIST_HEAD(&napi->poll_list); 6854 INIT_HLIST_NODE(&napi->napi_hash_node); 6855 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); 6856 napi->timer.function = napi_watchdog; 6857 init_gro_hash(napi); 6858 napi->skb = NULL; 6859 INIT_LIST_HEAD(&napi->rx_list); 6860 napi->rx_count = 0; 6861 napi->poll = poll; 6862 if (weight > NAPI_POLL_WEIGHT) 6863 netdev_err_once(dev, "%s() called with weight %d\n", __func__, 6864 weight); 6865 napi->weight = weight; 6866 napi->dev = dev; 6867 #ifdef CONFIG_NETPOLL 6868 napi->poll_owner = -1; 6869 #endif 6870 set_bit(NAPI_STATE_SCHED, &napi->state); 6871 set_bit(NAPI_STATE_NPSVC, &napi->state); 6872 list_add_rcu(&napi->dev_list, &dev->napi_list); 6873 napi_hash_add(napi); 6874 /* Create kthread for this napi if dev->threaded is set. 6875 * Clear dev->threaded if kthread creation failed so that 6876 * threaded mode will not be enabled in napi_enable(). 6877 */ 6878 if (dev->threaded && napi_kthread_create(napi)) 6879 dev->threaded = 0; 6880 } 6881 EXPORT_SYMBOL(netif_napi_add); 6882 6883 void napi_disable(struct napi_struct *n) 6884 { 6885 might_sleep(); 6886 set_bit(NAPI_STATE_DISABLE, &n->state); 6887 6888 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state)) 6889 msleep(1); 6890 while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state)) 6891 msleep(1); 6892 6893 hrtimer_cancel(&n->timer); 6894 6895 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &n->state); 6896 clear_bit(NAPI_STATE_DISABLE, &n->state); 6897 clear_bit(NAPI_STATE_THREADED, &n->state); 6898 } 6899 EXPORT_SYMBOL(napi_disable); 6900 6901 /** 6902 * napi_enable - enable NAPI scheduling 6903 * @n: NAPI context 6904 * 6905 * Resume NAPI from being scheduled on this context. 6906 * Must be paired with napi_disable. 6907 */ 6908 void napi_enable(struct napi_struct *n) 6909 { 6910 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state)); 6911 smp_mb__before_atomic(); 6912 clear_bit(NAPI_STATE_SCHED, &n->state); 6913 clear_bit(NAPI_STATE_NPSVC, &n->state); 6914 if (n->dev->threaded && n->thread) 6915 set_bit(NAPI_STATE_THREADED, &n->state); 6916 } 6917 EXPORT_SYMBOL(napi_enable); 6918 6919 static void flush_gro_hash(struct napi_struct *napi) 6920 { 6921 int i; 6922 6923 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6924 struct sk_buff *skb, *n; 6925 6926 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list) 6927 kfree_skb(skb); 6928 napi->gro_hash[i].count = 0; 6929 } 6930 } 6931 6932 /* Must be called in process context */ 6933 void __netif_napi_del(struct napi_struct *napi) 6934 { 6935 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state)) 6936 return; 6937 6938 napi_hash_del(napi); 6939 list_del_rcu(&napi->dev_list); 6940 napi_free_frags(napi); 6941 6942 flush_gro_hash(napi); 6943 napi->gro_bitmask = 0; 6944 6945 if (napi->thread) { 6946 kthread_stop(napi->thread); 6947 napi->thread = NULL; 6948 } 6949 } 6950 EXPORT_SYMBOL(__netif_napi_del); 6951 6952 static int __napi_poll(struct napi_struct *n, bool *repoll) 6953 { 6954 int work, weight; 6955 6956 weight = n->weight; 6957 6958 /* This NAPI_STATE_SCHED test is for avoiding a race 6959 * with netpoll's poll_napi(). Only the entity which 6960 * obtains the lock and sees NAPI_STATE_SCHED set will 6961 * actually make the ->poll() call. Therefore we avoid 6962 * accidentally calling ->poll() when NAPI is not scheduled. 6963 */ 6964 work = 0; 6965 if (test_bit(NAPI_STATE_SCHED, &n->state)) { 6966 work = n->poll(n, weight); 6967 trace_napi_poll(n, work, weight); 6968 } 6969 6970 if (unlikely(work > weight)) 6971 pr_err_once("NAPI poll function %pS returned %d, exceeding its budget of %d.\n", 6972 n->poll, work, weight); 6973 6974 if (likely(work < weight)) 6975 return work; 6976 6977 /* Drivers must not modify the NAPI state if they 6978 * consume the entire weight. In such cases this code 6979 * still "owns" the NAPI instance and therefore can 6980 * move the instance around on the list at-will. 6981 */ 6982 if (unlikely(napi_disable_pending(n))) { 6983 napi_complete(n); 6984 return work; 6985 } 6986 6987 /* The NAPI context has more processing work, but busy-polling 6988 * is preferred. Exit early. 6989 */ 6990 if (napi_prefer_busy_poll(n)) { 6991 if (napi_complete_done(n, work)) { 6992 /* If timeout is not set, we need to make sure 6993 * that the NAPI is re-scheduled. 6994 */ 6995 napi_schedule(n); 6996 } 6997 return work; 6998 } 6999 7000 if (n->gro_bitmask) { 7001 /* flush too old packets 7002 * If HZ < 1000, flush all packets. 7003 */ 7004 napi_gro_flush(n, HZ >= 1000); 7005 } 7006 7007 gro_normal_list(n); 7008 7009 /* Some drivers may have called napi_schedule 7010 * prior to exhausting their budget. 7011 */ 7012 if (unlikely(!list_empty(&n->poll_list))) { 7013 pr_warn_once("%s: Budget exhausted after napi rescheduled\n", 7014 n->dev ? n->dev->name : "backlog"); 7015 return work; 7016 } 7017 7018 *repoll = true; 7019 7020 return work; 7021 } 7022 7023 static int napi_poll(struct napi_struct *n, struct list_head *repoll) 7024 { 7025 bool do_repoll = false; 7026 void *have; 7027 int work; 7028 7029 list_del_init(&n->poll_list); 7030 7031 have = netpoll_poll_lock(n); 7032 7033 work = __napi_poll(n, &do_repoll); 7034 7035 if (do_repoll) 7036 list_add_tail(&n->poll_list, repoll); 7037 7038 netpoll_poll_unlock(have); 7039 7040 return work; 7041 } 7042 7043 static int napi_thread_wait(struct napi_struct *napi) 7044 { 7045 bool woken = false; 7046 7047 set_current_state(TASK_INTERRUPTIBLE); 7048 7049 while (!kthread_should_stop()) { 7050 /* Testing SCHED_THREADED bit here to make sure the current 7051 * kthread owns this napi and could poll on this napi. 7052 * Testing SCHED bit is not enough because SCHED bit might be 7053 * set by some other busy poll thread or by napi_disable(). 7054 */ 7055 if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state) || woken) { 7056 WARN_ON(!list_empty(&napi->poll_list)); 7057 __set_current_state(TASK_RUNNING); 7058 return 0; 7059 } 7060 7061 schedule(); 7062 /* woken being true indicates this thread owns this napi. */ 7063 woken = true; 7064 set_current_state(TASK_INTERRUPTIBLE); 7065 } 7066 __set_current_state(TASK_RUNNING); 7067 7068 return -1; 7069 } 7070 7071 static int napi_threaded_poll(void *data) 7072 { 7073 struct napi_struct *napi = data; 7074 void *have; 7075 7076 while (!napi_thread_wait(napi)) { 7077 for (;;) { 7078 bool repoll = false; 7079 7080 local_bh_disable(); 7081 7082 have = netpoll_poll_lock(napi); 7083 __napi_poll(napi, &repoll); 7084 netpoll_poll_unlock(have); 7085 7086 local_bh_enable(); 7087 7088 if (!repoll) 7089 break; 7090 7091 cond_resched(); 7092 } 7093 } 7094 return 0; 7095 } 7096 7097 static __latent_entropy void net_rx_action(struct softirq_action *h) 7098 { 7099 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 7100 unsigned long time_limit = jiffies + 7101 usecs_to_jiffies(netdev_budget_usecs); 7102 int budget = netdev_budget; 7103 LIST_HEAD(list); 7104 LIST_HEAD(repoll); 7105 7106 local_irq_disable(); 7107 list_splice_init(&sd->poll_list, &list); 7108 local_irq_enable(); 7109 7110 for (;;) { 7111 struct napi_struct *n; 7112 7113 if (list_empty(&list)) { 7114 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll)) 7115 return; 7116 break; 7117 } 7118 7119 n = list_first_entry(&list, struct napi_struct, poll_list); 7120 budget -= napi_poll(n, &repoll); 7121 7122 /* If softirq window is exhausted then punt. 7123 * Allow this to run for 2 jiffies since which will allow 7124 * an average latency of 1.5/HZ. 7125 */ 7126 if (unlikely(budget <= 0 || 7127 time_after_eq(jiffies, time_limit))) { 7128 sd->time_squeeze++; 7129 break; 7130 } 7131 } 7132 7133 local_irq_disable(); 7134 7135 list_splice_tail_init(&sd->poll_list, &list); 7136 list_splice_tail(&repoll, &list); 7137 list_splice(&list, &sd->poll_list); 7138 if (!list_empty(&sd->poll_list)) 7139 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 7140 7141 net_rps_action_and_irq_enable(sd); 7142 } 7143 7144 struct netdev_adjacent { 7145 struct net_device *dev; 7146 7147 /* upper master flag, there can only be one master device per list */ 7148 bool master; 7149 7150 /* lookup ignore flag */ 7151 bool ignore; 7152 7153 /* counter for the number of times this device was added to us */ 7154 u16 ref_nr; 7155 7156 /* private field for the users */ 7157 void *private; 7158 7159 struct list_head list; 7160 struct rcu_head rcu; 7161 }; 7162 7163 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, 7164 struct list_head *adj_list) 7165 { 7166 struct netdev_adjacent *adj; 7167 7168 list_for_each_entry(adj, adj_list, list) { 7169 if (adj->dev == adj_dev) 7170 return adj; 7171 } 7172 return NULL; 7173 } 7174 7175 static int ____netdev_has_upper_dev(struct net_device *upper_dev, 7176 struct netdev_nested_priv *priv) 7177 { 7178 struct net_device *dev = (struct net_device *)priv->data; 7179 7180 return upper_dev == dev; 7181 } 7182 7183 /** 7184 * netdev_has_upper_dev - Check if device is linked to an upper device 7185 * @dev: device 7186 * @upper_dev: upper device to check 7187 * 7188 * Find out if a device is linked to specified upper device and return true 7189 * in case it is. Note that this checks only immediate upper device, 7190 * not through a complete stack of devices. The caller must hold the RTNL lock. 7191 */ 7192 bool netdev_has_upper_dev(struct net_device *dev, 7193 struct net_device *upper_dev) 7194 { 7195 struct netdev_nested_priv priv = { 7196 .data = (void *)upper_dev, 7197 }; 7198 7199 ASSERT_RTNL(); 7200 7201 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 7202 &priv); 7203 } 7204 EXPORT_SYMBOL(netdev_has_upper_dev); 7205 7206 /** 7207 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device 7208 * @dev: device 7209 * @upper_dev: upper device to check 7210 * 7211 * Find out if a device is linked to specified upper device and return true 7212 * in case it is. Note that this checks the entire upper device chain. 7213 * The caller must hold rcu lock. 7214 */ 7215 7216 bool netdev_has_upper_dev_all_rcu(struct net_device *dev, 7217 struct net_device *upper_dev) 7218 { 7219 struct netdev_nested_priv priv = { 7220 .data = (void *)upper_dev, 7221 }; 7222 7223 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 7224 &priv); 7225 } 7226 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); 7227 7228 /** 7229 * netdev_has_any_upper_dev - Check if device is linked to some device 7230 * @dev: device 7231 * 7232 * Find out if a device is linked to an upper device and return true in case 7233 * it is. The caller must hold the RTNL lock. 7234 */ 7235 bool netdev_has_any_upper_dev(struct net_device *dev) 7236 { 7237 ASSERT_RTNL(); 7238 7239 return !list_empty(&dev->adj_list.upper); 7240 } 7241 EXPORT_SYMBOL(netdev_has_any_upper_dev); 7242 7243 /** 7244 * netdev_master_upper_dev_get - Get master upper device 7245 * @dev: device 7246 * 7247 * Find a master upper device and return pointer to it or NULL in case 7248 * it's not there. The caller must hold the RTNL lock. 7249 */ 7250 struct net_device *netdev_master_upper_dev_get(struct net_device *dev) 7251 { 7252 struct netdev_adjacent *upper; 7253 7254 ASSERT_RTNL(); 7255 7256 if (list_empty(&dev->adj_list.upper)) 7257 return NULL; 7258 7259 upper = list_first_entry(&dev->adj_list.upper, 7260 struct netdev_adjacent, list); 7261 if (likely(upper->master)) 7262 return upper->dev; 7263 return NULL; 7264 } 7265 EXPORT_SYMBOL(netdev_master_upper_dev_get); 7266 7267 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev) 7268 { 7269 struct netdev_adjacent *upper; 7270 7271 ASSERT_RTNL(); 7272 7273 if (list_empty(&dev->adj_list.upper)) 7274 return NULL; 7275 7276 upper = list_first_entry(&dev->adj_list.upper, 7277 struct netdev_adjacent, list); 7278 if (likely(upper->master) && !upper->ignore) 7279 return upper->dev; 7280 return NULL; 7281 } 7282 7283 /** 7284 * netdev_has_any_lower_dev - Check if device is linked to some device 7285 * @dev: device 7286 * 7287 * Find out if a device is linked to a lower device and return true in case 7288 * it is. The caller must hold the RTNL lock. 7289 */ 7290 static bool netdev_has_any_lower_dev(struct net_device *dev) 7291 { 7292 ASSERT_RTNL(); 7293 7294 return !list_empty(&dev->adj_list.lower); 7295 } 7296 7297 void *netdev_adjacent_get_private(struct list_head *adj_list) 7298 { 7299 struct netdev_adjacent *adj; 7300 7301 adj = list_entry(adj_list, struct netdev_adjacent, list); 7302 7303 return adj->private; 7304 } 7305 EXPORT_SYMBOL(netdev_adjacent_get_private); 7306 7307 /** 7308 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list 7309 * @dev: device 7310 * @iter: list_head ** of the current position 7311 * 7312 * Gets the next device from the dev's upper list, starting from iter 7313 * position. The caller must hold RCU read lock. 7314 */ 7315 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, 7316 struct list_head **iter) 7317 { 7318 struct netdev_adjacent *upper; 7319 7320 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 7321 7322 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7323 7324 if (&upper->list == &dev->adj_list.upper) 7325 return NULL; 7326 7327 *iter = &upper->list; 7328 7329 return upper->dev; 7330 } 7331 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); 7332 7333 static struct net_device *__netdev_next_upper_dev(struct net_device *dev, 7334 struct list_head **iter, 7335 bool *ignore) 7336 { 7337 struct netdev_adjacent *upper; 7338 7339 upper = list_entry((*iter)->next, struct netdev_adjacent, list); 7340 7341 if (&upper->list == &dev->adj_list.upper) 7342 return NULL; 7343 7344 *iter = &upper->list; 7345 *ignore = upper->ignore; 7346 7347 return upper->dev; 7348 } 7349 7350 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, 7351 struct list_head **iter) 7352 { 7353 struct netdev_adjacent *upper; 7354 7355 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 7356 7357 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7358 7359 if (&upper->list == &dev->adj_list.upper) 7360 return NULL; 7361 7362 *iter = &upper->list; 7363 7364 return upper->dev; 7365 } 7366 7367 static int __netdev_walk_all_upper_dev(struct net_device *dev, 7368 int (*fn)(struct net_device *dev, 7369 struct netdev_nested_priv *priv), 7370 struct netdev_nested_priv *priv) 7371 { 7372 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7373 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7374 int ret, cur = 0; 7375 bool ignore; 7376 7377 now = dev; 7378 iter = &dev->adj_list.upper; 7379 7380 while (1) { 7381 if (now != dev) { 7382 ret = fn(now, priv); 7383 if (ret) 7384 return ret; 7385 } 7386 7387 next = NULL; 7388 while (1) { 7389 udev = __netdev_next_upper_dev(now, &iter, &ignore); 7390 if (!udev) 7391 break; 7392 if (ignore) 7393 continue; 7394 7395 next = udev; 7396 niter = &udev->adj_list.upper; 7397 dev_stack[cur] = now; 7398 iter_stack[cur++] = iter; 7399 break; 7400 } 7401 7402 if (!next) { 7403 if (!cur) 7404 return 0; 7405 next = dev_stack[--cur]; 7406 niter = iter_stack[cur]; 7407 } 7408 7409 now = next; 7410 iter = niter; 7411 } 7412 7413 return 0; 7414 } 7415 7416 int netdev_walk_all_upper_dev_rcu(struct net_device *dev, 7417 int (*fn)(struct net_device *dev, 7418 struct netdev_nested_priv *priv), 7419 struct netdev_nested_priv *priv) 7420 { 7421 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7422 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7423 int ret, cur = 0; 7424 7425 now = dev; 7426 iter = &dev->adj_list.upper; 7427 7428 while (1) { 7429 if (now != dev) { 7430 ret = fn(now, priv); 7431 if (ret) 7432 return ret; 7433 } 7434 7435 next = NULL; 7436 while (1) { 7437 udev = netdev_next_upper_dev_rcu(now, &iter); 7438 if (!udev) 7439 break; 7440 7441 next = udev; 7442 niter = &udev->adj_list.upper; 7443 dev_stack[cur] = now; 7444 iter_stack[cur++] = iter; 7445 break; 7446 } 7447 7448 if (!next) { 7449 if (!cur) 7450 return 0; 7451 next = dev_stack[--cur]; 7452 niter = iter_stack[cur]; 7453 } 7454 7455 now = next; 7456 iter = niter; 7457 } 7458 7459 return 0; 7460 } 7461 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); 7462 7463 static bool __netdev_has_upper_dev(struct net_device *dev, 7464 struct net_device *upper_dev) 7465 { 7466 struct netdev_nested_priv priv = { 7467 .flags = 0, 7468 .data = (void *)upper_dev, 7469 }; 7470 7471 ASSERT_RTNL(); 7472 7473 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev, 7474 &priv); 7475 } 7476 7477 /** 7478 * netdev_lower_get_next_private - Get the next ->private from the 7479 * lower neighbour list 7480 * @dev: device 7481 * @iter: list_head ** of the current position 7482 * 7483 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7484 * list, starting from iter position. The caller must hold either hold the 7485 * RTNL lock or its own locking that guarantees that the neighbour lower 7486 * list will remain unchanged. 7487 */ 7488 void *netdev_lower_get_next_private(struct net_device *dev, 7489 struct list_head **iter) 7490 { 7491 struct netdev_adjacent *lower; 7492 7493 lower = list_entry(*iter, struct netdev_adjacent, list); 7494 7495 if (&lower->list == &dev->adj_list.lower) 7496 return NULL; 7497 7498 *iter = lower->list.next; 7499 7500 return lower->private; 7501 } 7502 EXPORT_SYMBOL(netdev_lower_get_next_private); 7503 7504 /** 7505 * netdev_lower_get_next_private_rcu - Get the next ->private from the 7506 * lower neighbour list, RCU 7507 * variant 7508 * @dev: device 7509 * @iter: list_head ** of the current position 7510 * 7511 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7512 * list, starting from iter position. The caller must hold RCU read lock. 7513 */ 7514 void *netdev_lower_get_next_private_rcu(struct net_device *dev, 7515 struct list_head **iter) 7516 { 7517 struct netdev_adjacent *lower; 7518 7519 WARN_ON_ONCE(!rcu_read_lock_held()); 7520 7521 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7522 7523 if (&lower->list == &dev->adj_list.lower) 7524 return NULL; 7525 7526 *iter = &lower->list; 7527 7528 return lower->private; 7529 } 7530 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); 7531 7532 /** 7533 * netdev_lower_get_next - Get the next device from the lower neighbour 7534 * list 7535 * @dev: device 7536 * @iter: list_head ** of the current position 7537 * 7538 * Gets the next netdev_adjacent from the dev's lower neighbour 7539 * list, starting from iter position. The caller must hold RTNL lock or 7540 * its own locking that guarantees that the neighbour lower 7541 * list will remain unchanged. 7542 */ 7543 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) 7544 { 7545 struct netdev_adjacent *lower; 7546 7547 lower = list_entry(*iter, struct netdev_adjacent, list); 7548 7549 if (&lower->list == &dev->adj_list.lower) 7550 return NULL; 7551 7552 *iter = lower->list.next; 7553 7554 return lower->dev; 7555 } 7556 EXPORT_SYMBOL(netdev_lower_get_next); 7557 7558 static struct net_device *netdev_next_lower_dev(struct net_device *dev, 7559 struct list_head **iter) 7560 { 7561 struct netdev_adjacent *lower; 7562 7563 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7564 7565 if (&lower->list == &dev->adj_list.lower) 7566 return NULL; 7567 7568 *iter = &lower->list; 7569 7570 return lower->dev; 7571 } 7572 7573 static struct net_device *__netdev_next_lower_dev(struct net_device *dev, 7574 struct list_head **iter, 7575 bool *ignore) 7576 { 7577 struct netdev_adjacent *lower; 7578 7579 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7580 7581 if (&lower->list == &dev->adj_list.lower) 7582 return NULL; 7583 7584 *iter = &lower->list; 7585 *ignore = lower->ignore; 7586 7587 return lower->dev; 7588 } 7589 7590 int netdev_walk_all_lower_dev(struct net_device *dev, 7591 int (*fn)(struct net_device *dev, 7592 struct netdev_nested_priv *priv), 7593 struct netdev_nested_priv *priv) 7594 { 7595 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7596 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7597 int ret, cur = 0; 7598 7599 now = dev; 7600 iter = &dev->adj_list.lower; 7601 7602 while (1) { 7603 if (now != dev) { 7604 ret = fn(now, priv); 7605 if (ret) 7606 return ret; 7607 } 7608 7609 next = NULL; 7610 while (1) { 7611 ldev = netdev_next_lower_dev(now, &iter); 7612 if (!ldev) 7613 break; 7614 7615 next = ldev; 7616 niter = &ldev->adj_list.lower; 7617 dev_stack[cur] = now; 7618 iter_stack[cur++] = iter; 7619 break; 7620 } 7621 7622 if (!next) { 7623 if (!cur) 7624 return 0; 7625 next = dev_stack[--cur]; 7626 niter = iter_stack[cur]; 7627 } 7628 7629 now = next; 7630 iter = niter; 7631 } 7632 7633 return 0; 7634 } 7635 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); 7636 7637 static int __netdev_walk_all_lower_dev(struct net_device *dev, 7638 int (*fn)(struct net_device *dev, 7639 struct netdev_nested_priv *priv), 7640 struct netdev_nested_priv *priv) 7641 { 7642 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7643 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7644 int ret, cur = 0; 7645 bool ignore; 7646 7647 now = dev; 7648 iter = &dev->adj_list.lower; 7649 7650 while (1) { 7651 if (now != dev) { 7652 ret = fn(now, priv); 7653 if (ret) 7654 return ret; 7655 } 7656 7657 next = NULL; 7658 while (1) { 7659 ldev = __netdev_next_lower_dev(now, &iter, &ignore); 7660 if (!ldev) 7661 break; 7662 if (ignore) 7663 continue; 7664 7665 next = ldev; 7666 niter = &ldev->adj_list.lower; 7667 dev_stack[cur] = now; 7668 iter_stack[cur++] = iter; 7669 break; 7670 } 7671 7672 if (!next) { 7673 if (!cur) 7674 return 0; 7675 next = dev_stack[--cur]; 7676 niter = iter_stack[cur]; 7677 } 7678 7679 now = next; 7680 iter = niter; 7681 } 7682 7683 return 0; 7684 } 7685 7686 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, 7687 struct list_head **iter) 7688 { 7689 struct netdev_adjacent *lower; 7690 7691 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7692 if (&lower->list == &dev->adj_list.lower) 7693 return NULL; 7694 7695 *iter = &lower->list; 7696 7697 return lower->dev; 7698 } 7699 EXPORT_SYMBOL(netdev_next_lower_dev_rcu); 7700 7701 static u8 __netdev_upper_depth(struct net_device *dev) 7702 { 7703 struct net_device *udev; 7704 struct list_head *iter; 7705 u8 max_depth = 0; 7706 bool ignore; 7707 7708 for (iter = &dev->adj_list.upper, 7709 udev = __netdev_next_upper_dev(dev, &iter, &ignore); 7710 udev; 7711 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) { 7712 if (ignore) 7713 continue; 7714 if (max_depth < udev->upper_level) 7715 max_depth = udev->upper_level; 7716 } 7717 7718 return max_depth; 7719 } 7720 7721 static u8 __netdev_lower_depth(struct net_device *dev) 7722 { 7723 struct net_device *ldev; 7724 struct list_head *iter; 7725 u8 max_depth = 0; 7726 bool ignore; 7727 7728 for (iter = &dev->adj_list.lower, 7729 ldev = __netdev_next_lower_dev(dev, &iter, &ignore); 7730 ldev; 7731 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) { 7732 if (ignore) 7733 continue; 7734 if (max_depth < ldev->lower_level) 7735 max_depth = ldev->lower_level; 7736 } 7737 7738 return max_depth; 7739 } 7740 7741 static int __netdev_update_upper_level(struct net_device *dev, 7742 struct netdev_nested_priv *__unused) 7743 { 7744 dev->upper_level = __netdev_upper_depth(dev) + 1; 7745 return 0; 7746 } 7747 7748 static int __netdev_update_lower_level(struct net_device *dev, 7749 struct netdev_nested_priv *priv) 7750 { 7751 dev->lower_level = __netdev_lower_depth(dev) + 1; 7752 7753 #ifdef CONFIG_LOCKDEP 7754 if (!priv) 7755 return 0; 7756 7757 if (priv->flags & NESTED_SYNC_IMM) 7758 dev->nested_level = dev->lower_level - 1; 7759 if (priv->flags & NESTED_SYNC_TODO) 7760 net_unlink_todo(dev); 7761 #endif 7762 return 0; 7763 } 7764 7765 int netdev_walk_all_lower_dev_rcu(struct net_device *dev, 7766 int (*fn)(struct net_device *dev, 7767 struct netdev_nested_priv *priv), 7768 struct netdev_nested_priv *priv) 7769 { 7770 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7771 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7772 int ret, cur = 0; 7773 7774 now = dev; 7775 iter = &dev->adj_list.lower; 7776 7777 while (1) { 7778 if (now != dev) { 7779 ret = fn(now, priv); 7780 if (ret) 7781 return ret; 7782 } 7783 7784 next = NULL; 7785 while (1) { 7786 ldev = netdev_next_lower_dev_rcu(now, &iter); 7787 if (!ldev) 7788 break; 7789 7790 next = ldev; 7791 niter = &ldev->adj_list.lower; 7792 dev_stack[cur] = now; 7793 iter_stack[cur++] = iter; 7794 break; 7795 } 7796 7797 if (!next) { 7798 if (!cur) 7799 return 0; 7800 next = dev_stack[--cur]; 7801 niter = iter_stack[cur]; 7802 } 7803 7804 now = next; 7805 iter = niter; 7806 } 7807 7808 return 0; 7809 } 7810 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); 7811 7812 /** 7813 * netdev_lower_get_first_private_rcu - Get the first ->private from the 7814 * lower neighbour list, RCU 7815 * variant 7816 * @dev: device 7817 * 7818 * Gets the first netdev_adjacent->private from the dev's lower neighbour 7819 * list. The caller must hold RCU read lock. 7820 */ 7821 void *netdev_lower_get_first_private_rcu(struct net_device *dev) 7822 { 7823 struct netdev_adjacent *lower; 7824 7825 lower = list_first_or_null_rcu(&dev->adj_list.lower, 7826 struct netdev_adjacent, list); 7827 if (lower) 7828 return lower->private; 7829 return NULL; 7830 } 7831 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); 7832 7833 /** 7834 * netdev_master_upper_dev_get_rcu - Get master upper device 7835 * @dev: device 7836 * 7837 * Find a master upper device and return pointer to it or NULL in case 7838 * it's not there. The caller must hold the RCU read lock. 7839 */ 7840 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) 7841 { 7842 struct netdev_adjacent *upper; 7843 7844 upper = list_first_or_null_rcu(&dev->adj_list.upper, 7845 struct netdev_adjacent, list); 7846 if (upper && likely(upper->master)) 7847 return upper->dev; 7848 return NULL; 7849 } 7850 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); 7851 7852 static int netdev_adjacent_sysfs_add(struct net_device *dev, 7853 struct net_device *adj_dev, 7854 struct list_head *dev_list) 7855 { 7856 char linkname[IFNAMSIZ+7]; 7857 7858 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7859 "upper_%s" : "lower_%s", adj_dev->name); 7860 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), 7861 linkname); 7862 } 7863 static void netdev_adjacent_sysfs_del(struct net_device *dev, 7864 char *name, 7865 struct list_head *dev_list) 7866 { 7867 char linkname[IFNAMSIZ+7]; 7868 7869 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7870 "upper_%s" : "lower_%s", name); 7871 sysfs_remove_link(&(dev->dev.kobj), linkname); 7872 } 7873 7874 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, 7875 struct net_device *adj_dev, 7876 struct list_head *dev_list) 7877 { 7878 return (dev_list == &dev->adj_list.upper || 7879 dev_list == &dev->adj_list.lower) && 7880 net_eq(dev_net(dev), dev_net(adj_dev)); 7881 } 7882 7883 static int __netdev_adjacent_dev_insert(struct net_device *dev, 7884 struct net_device *adj_dev, 7885 struct list_head *dev_list, 7886 void *private, bool master) 7887 { 7888 struct netdev_adjacent *adj; 7889 int ret; 7890 7891 adj = __netdev_find_adj(adj_dev, dev_list); 7892 7893 if (adj) { 7894 adj->ref_nr += 1; 7895 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", 7896 dev->name, adj_dev->name, adj->ref_nr); 7897 7898 return 0; 7899 } 7900 7901 adj = kmalloc(sizeof(*adj), GFP_KERNEL); 7902 if (!adj) 7903 return -ENOMEM; 7904 7905 adj->dev = adj_dev; 7906 adj->master = master; 7907 adj->ref_nr = 1; 7908 adj->private = private; 7909 adj->ignore = false; 7910 dev_hold(adj_dev); 7911 7912 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", 7913 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); 7914 7915 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { 7916 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); 7917 if (ret) 7918 goto free_adj; 7919 } 7920 7921 /* Ensure that master link is always the first item in list. */ 7922 if (master) { 7923 ret = sysfs_create_link(&(dev->dev.kobj), 7924 &(adj_dev->dev.kobj), "master"); 7925 if (ret) 7926 goto remove_symlinks; 7927 7928 list_add_rcu(&adj->list, dev_list); 7929 } else { 7930 list_add_tail_rcu(&adj->list, dev_list); 7931 } 7932 7933 return 0; 7934 7935 remove_symlinks: 7936 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7937 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7938 free_adj: 7939 kfree(adj); 7940 dev_put(adj_dev); 7941 7942 return ret; 7943 } 7944 7945 static void __netdev_adjacent_dev_remove(struct net_device *dev, 7946 struct net_device *adj_dev, 7947 u16 ref_nr, 7948 struct list_head *dev_list) 7949 { 7950 struct netdev_adjacent *adj; 7951 7952 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", 7953 dev->name, adj_dev->name, ref_nr); 7954 7955 adj = __netdev_find_adj(adj_dev, dev_list); 7956 7957 if (!adj) { 7958 pr_err("Adjacency does not exist for device %s from %s\n", 7959 dev->name, adj_dev->name); 7960 WARN_ON(1); 7961 return; 7962 } 7963 7964 if (adj->ref_nr > ref_nr) { 7965 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", 7966 dev->name, adj_dev->name, ref_nr, 7967 adj->ref_nr - ref_nr); 7968 adj->ref_nr -= ref_nr; 7969 return; 7970 } 7971 7972 if (adj->master) 7973 sysfs_remove_link(&(dev->dev.kobj), "master"); 7974 7975 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7976 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7977 7978 list_del_rcu(&adj->list); 7979 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", 7980 adj_dev->name, dev->name, adj_dev->name); 7981 dev_put(adj_dev); 7982 kfree_rcu(adj, rcu); 7983 } 7984 7985 static int __netdev_adjacent_dev_link_lists(struct net_device *dev, 7986 struct net_device *upper_dev, 7987 struct list_head *up_list, 7988 struct list_head *down_list, 7989 void *private, bool master) 7990 { 7991 int ret; 7992 7993 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, 7994 private, master); 7995 if (ret) 7996 return ret; 7997 7998 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, 7999 private, false); 8000 if (ret) { 8001 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); 8002 return ret; 8003 } 8004 8005 return 0; 8006 } 8007 8008 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, 8009 struct net_device *upper_dev, 8010 u16 ref_nr, 8011 struct list_head *up_list, 8012 struct list_head *down_list) 8013 { 8014 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); 8015 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); 8016 } 8017 8018 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, 8019 struct net_device *upper_dev, 8020 void *private, bool master) 8021 { 8022 return __netdev_adjacent_dev_link_lists(dev, upper_dev, 8023 &dev->adj_list.upper, 8024 &upper_dev->adj_list.lower, 8025 private, master); 8026 } 8027 8028 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, 8029 struct net_device *upper_dev) 8030 { 8031 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, 8032 &dev->adj_list.upper, 8033 &upper_dev->adj_list.lower); 8034 } 8035 8036 static int __netdev_upper_dev_link(struct net_device *dev, 8037 struct net_device *upper_dev, bool master, 8038 void *upper_priv, void *upper_info, 8039 struct netdev_nested_priv *priv, 8040 struct netlink_ext_ack *extack) 8041 { 8042 struct netdev_notifier_changeupper_info changeupper_info = { 8043 .info = { 8044 .dev = dev, 8045 .extack = extack, 8046 }, 8047 .upper_dev = upper_dev, 8048 .master = master, 8049 .linking = true, 8050 .upper_info = upper_info, 8051 }; 8052 struct net_device *master_dev; 8053 int ret = 0; 8054 8055 ASSERT_RTNL(); 8056 8057 if (dev == upper_dev) 8058 return -EBUSY; 8059 8060 /* To prevent loops, check if dev is not upper device to upper_dev. */ 8061 if (__netdev_has_upper_dev(upper_dev, dev)) 8062 return -EBUSY; 8063 8064 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV) 8065 return -EMLINK; 8066 8067 if (!master) { 8068 if (__netdev_has_upper_dev(dev, upper_dev)) 8069 return -EEXIST; 8070 } else { 8071 master_dev = __netdev_master_upper_dev_get(dev); 8072 if (master_dev) 8073 return master_dev == upper_dev ? -EEXIST : -EBUSY; 8074 } 8075 8076 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 8077 &changeupper_info.info); 8078 ret = notifier_to_errno(ret); 8079 if (ret) 8080 return ret; 8081 8082 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, 8083 master); 8084 if (ret) 8085 return ret; 8086 8087 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 8088 &changeupper_info.info); 8089 ret = notifier_to_errno(ret); 8090 if (ret) 8091 goto rollback; 8092 8093 __netdev_update_upper_level(dev, NULL); 8094 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 8095 8096 __netdev_update_lower_level(upper_dev, priv); 8097 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 8098 priv); 8099 8100 return 0; 8101 8102 rollback: 8103 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 8104 8105 return ret; 8106 } 8107 8108 /** 8109 * netdev_upper_dev_link - Add a link to the upper device 8110 * @dev: device 8111 * @upper_dev: new upper device 8112 * @extack: netlink extended ack 8113 * 8114 * Adds a link to device which is upper to this one. The caller must hold 8115 * the RTNL lock. On a failure a negative errno code is returned. 8116 * On success the reference counts are adjusted and the function 8117 * returns zero. 8118 */ 8119 int netdev_upper_dev_link(struct net_device *dev, 8120 struct net_device *upper_dev, 8121 struct netlink_ext_ack *extack) 8122 { 8123 struct netdev_nested_priv priv = { 8124 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 8125 .data = NULL, 8126 }; 8127 8128 return __netdev_upper_dev_link(dev, upper_dev, false, 8129 NULL, NULL, &priv, extack); 8130 } 8131 EXPORT_SYMBOL(netdev_upper_dev_link); 8132 8133 /** 8134 * netdev_master_upper_dev_link - Add a master link to the upper device 8135 * @dev: device 8136 * @upper_dev: new upper device 8137 * @upper_priv: upper device private 8138 * @upper_info: upper info to be passed down via notifier 8139 * @extack: netlink extended ack 8140 * 8141 * Adds a link to device which is upper to this one. In this case, only 8142 * one master upper device can be linked, although other non-master devices 8143 * might be linked as well. The caller must hold the RTNL lock. 8144 * On a failure a negative errno code is returned. On success the reference 8145 * counts are adjusted and the function returns zero. 8146 */ 8147 int netdev_master_upper_dev_link(struct net_device *dev, 8148 struct net_device *upper_dev, 8149 void *upper_priv, void *upper_info, 8150 struct netlink_ext_ack *extack) 8151 { 8152 struct netdev_nested_priv priv = { 8153 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 8154 .data = NULL, 8155 }; 8156 8157 return __netdev_upper_dev_link(dev, upper_dev, true, 8158 upper_priv, upper_info, &priv, extack); 8159 } 8160 EXPORT_SYMBOL(netdev_master_upper_dev_link); 8161 8162 static void __netdev_upper_dev_unlink(struct net_device *dev, 8163 struct net_device *upper_dev, 8164 struct netdev_nested_priv *priv) 8165 { 8166 struct netdev_notifier_changeupper_info changeupper_info = { 8167 .info = { 8168 .dev = dev, 8169 }, 8170 .upper_dev = upper_dev, 8171 .linking = false, 8172 }; 8173 8174 ASSERT_RTNL(); 8175 8176 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; 8177 8178 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 8179 &changeupper_info.info); 8180 8181 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 8182 8183 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 8184 &changeupper_info.info); 8185 8186 __netdev_update_upper_level(dev, NULL); 8187 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 8188 8189 __netdev_update_lower_level(upper_dev, priv); 8190 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 8191 priv); 8192 } 8193 8194 /** 8195 * netdev_upper_dev_unlink - Removes a link to upper device 8196 * @dev: device 8197 * @upper_dev: new upper device 8198 * 8199 * Removes a link to device which is upper to this one. The caller must hold 8200 * the RTNL lock. 8201 */ 8202 void netdev_upper_dev_unlink(struct net_device *dev, 8203 struct net_device *upper_dev) 8204 { 8205 struct netdev_nested_priv priv = { 8206 .flags = NESTED_SYNC_TODO, 8207 .data = NULL, 8208 }; 8209 8210 __netdev_upper_dev_unlink(dev, upper_dev, &priv); 8211 } 8212 EXPORT_SYMBOL(netdev_upper_dev_unlink); 8213 8214 static void __netdev_adjacent_dev_set(struct net_device *upper_dev, 8215 struct net_device *lower_dev, 8216 bool val) 8217 { 8218 struct netdev_adjacent *adj; 8219 8220 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower); 8221 if (adj) 8222 adj->ignore = val; 8223 8224 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper); 8225 if (adj) 8226 adj->ignore = val; 8227 } 8228 8229 static void netdev_adjacent_dev_disable(struct net_device *upper_dev, 8230 struct net_device *lower_dev) 8231 { 8232 __netdev_adjacent_dev_set(upper_dev, lower_dev, true); 8233 } 8234 8235 static void netdev_adjacent_dev_enable(struct net_device *upper_dev, 8236 struct net_device *lower_dev) 8237 { 8238 __netdev_adjacent_dev_set(upper_dev, lower_dev, false); 8239 } 8240 8241 int netdev_adjacent_change_prepare(struct net_device *old_dev, 8242 struct net_device *new_dev, 8243 struct net_device *dev, 8244 struct netlink_ext_ack *extack) 8245 { 8246 struct netdev_nested_priv priv = { 8247 .flags = 0, 8248 .data = NULL, 8249 }; 8250 int err; 8251 8252 if (!new_dev) 8253 return 0; 8254 8255 if (old_dev && new_dev != old_dev) 8256 netdev_adjacent_dev_disable(dev, old_dev); 8257 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv, 8258 extack); 8259 if (err) { 8260 if (old_dev && new_dev != old_dev) 8261 netdev_adjacent_dev_enable(dev, old_dev); 8262 return err; 8263 } 8264 8265 return 0; 8266 } 8267 EXPORT_SYMBOL(netdev_adjacent_change_prepare); 8268 8269 void netdev_adjacent_change_commit(struct net_device *old_dev, 8270 struct net_device *new_dev, 8271 struct net_device *dev) 8272 { 8273 struct netdev_nested_priv priv = { 8274 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 8275 .data = NULL, 8276 }; 8277 8278 if (!new_dev || !old_dev) 8279 return; 8280 8281 if (new_dev == old_dev) 8282 return; 8283 8284 netdev_adjacent_dev_enable(dev, old_dev); 8285 __netdev_upper_dev_unlink(old_dev, dev, &priv); 8286 } 8287 EXPORT_SYMBOL(netdev_adjacent_change_commit); 8288 8289 void netdev_adjacent_change_abort(struct net_device *old_dev, 8290 struct net_device *new_dev, 8291 struct net_device *dev) 8292 { 8293 struct netdev_nested_priv priv = { 8294 .flags = 0, 8295 .data = NULL, 8296 }; 8297 8298 if (!new_dev) 8299 return; 8300 8301 if (old_dev && new_dev != old_dev) 8302 netdev_adjacent_dev_enable(dev, old_dev); 8303 8304 __netdev_upper_dev_unlink(new_dev, dev, &priv); 8305 } 8306 EXPORT_SYMBOL(netdev_adjacent_change_abort); 8307 8308 /** 8309 * netdev_bonding_info_change - Dispatch event about slave change 8310 * @dev: device 8311 * @bonding_info: info to dispatch 8312 * 8313 * Send NETDEV_BONDING_INFO to netdev notifiers with info. 8314 * The caller must hold the RTNL lock. 8315 */ 8316 void netdev_bonding_info_change(struct net_device *dev, 8317 struct netdev_bonding_info *bonding_info) 8318 { 8319 struct netdev_notifier_bonding_info info = { 8320 .info.dev = dev, 8321 }; 8322 8323 memcpy(&info.bonding_info, bonding_info, 8324 sizeof(struct netdev_bonding_info)); 8325 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, 8326 &info.info); 8327 } 8328 EXPORT_SYMBOL(netdev_bonding_info_change); 8329 8330 /** 8331 * netdev_get_xmit_slave - Get the xmit slave of master device 8332 * @dev: device 8333 * @skb: The packet 8334 * @all_slaves: assume all the slaves are active 8335 * 8336 * The reference counters are not incremented so the caller must be 8337 * careful with locks. The caller must hold RCU lock. 8338 * %NULL is returned if no slave is found. 8339 */ 8340 8341 struct net_device *netdev_get_xmit_slave(struct net_device *dev, 8342 struct sk_buff *skb, 8343 bool all_slaves) 8344 { 8345 const struct net_device_ops *ops = dev->netdev_ops; 8346 8347 if (!ops->ndo_get_xmit_slave) 8348 return NULL; 8349 return ops->ndo_get_xmit_slave(dev, skb, all_slaves); 8350 } 8351 EXPORT_SYMBOL(netdev_get_xmit_slave); 8352 8353 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev, 8354 struct sock *sk) 8355 { 8356 const struct net_device_ops *ops = dev->netdev_ops; 8357 8358 if (!ops->ndo_sk_get_lower_dev) 8359 return NULL; 8360 return ops->ndo_sk_get_lower_dev(dev, sk); 8361 } 8362 8363 /** 8364 * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket 8365 * @dev: device 8366 * @sk: the socket 8367 * 8368 * %NULL is returned if no lower device is found. 8369 */ 8370 8371 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev, 8372 struct sock *sk) 8373 { 8374 struct net_device *lower; 8375 8376 lower = netdev_sk_get_lower_dev(dev, sk); 8377 while (lower) { 8378 dev = lower; 8379 lower = netdev_sk_get_lower_dev(dev, sk); 8380 } 8381 8382 return dev; 8383 } 8384 EXPORT_SYMBOL(netdev_sk_get_lowest_dev); 8385 8386 static void netdev_adjacent_add_links(struct net_device *dev) 8387 { 8388 struct netdev_adjacent *iter; 8389 8390 struct net *net = dev_net(dev); 8391 8392 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8393 if (!net_eq(net, dev_net(iter->dev))) 8394 continue; 8395 netdev_adjacent_sysfs_add(iter->dev, dev, 8396 &iter->dev->adj_list.lower); 8397 netdev_adjacent_sysfs_add(dev, iter->dev, 8398 &dev->adj_list.upper); 8399 } 8400 8401 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8402 if (!net_eq(net, dev_net(iter->dev))) 8403 continue; 8404 netdev_adjacent_sysfs_add(iter->dev, dev, 8405 &iter->dev->adj_list.upper); 8406 netdev_adjacent_sysfs_add(dev, iter->dev, 8407 &dev->adj_list.lower); 8408 } 8409 } 8410 8411 static void netdev_adjacent_del_links(struct net_device *dev) 8412 { 8413 struct netdev_adjacent *iter; 8414 8415 struct net *net = dev_net(dev); 8416 8417 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8418 if (!net_eq(net, dev_net(iter->dev))) 8419 continue; 8420 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8421 &iter->dev->adj_list.lower); 8422 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8423 &dev->adj_list.upper); 8424 } 8425 8426 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8427 if (!net_eq(net, dev_net(iter->dev))) 8428 continue; 8429 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8430 &iter->dev->adj_list.upper); 8431 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8432 &dev->adj_list.lower); 8433 } 8434 } 8435 8436 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) 8437 { 8438 struct netdev_adjacent *iter; 8439 8440 struct net *net = dev_net(dev); 8441 8442 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8443 if (!net_eq(net, dev_net(iter->dev))) 8444 continue; 8445 netdev_adjacent_sysfs_del(iter->dev, oldname, 8446 &iter->dev->adj_list.lower); 8447 netdev_adjacent_sysfs_add(iter->dev, dev, 8448 &iter->dev->adj_list.lower); 8449 } 8450 8451 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8452 if (!net_eq(net, dev_net(iter->dev))) 8453 continue; 8454 netdev_adjacent_sysfs_del(iter->dev, oldname, 8455 &iter->dev->adj_list.upper); 8456 netdev_adjacent_sysfs_add(iter->dev, dev, 8457 &iter->dev->adj_list.upper); 8458 } 8459 } 8460 8461 void *netdev_lower_dev_get_private(struct net_device *dev, 8462 struct net_device *lower_dev) 8463 { 8464 struct netdev_adjacent *lower; 8465 8466 if (!lower_dev) 8467 return NULL; 8468 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); 8469 if (!lower) 8470 return NULL; 8471 8472 return lower->private; 8473 } 8474 EXPORT_SYMBOL(netdev_lower_dev_get_private); 8475 8476 8477 /** 8478 * netdev_lower_state_changed - Dispatch event about lower device state change 8479 * @lower_dev: device 8480 * @lower_state_info: state to dispatch 8481 * 8482 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. 8483 * The caller must hold the RTNL lock. 8484 */ 8485 void netdev_lower_state_changed(struct net_device *lower_dev, 8486 void *lower_state_info) 8487 { 8488 struct netdev_notifier_changelowerstate_info changelowerstate_info = { 8489 .info.dev = lower_dev, 8490 }; 8491 8492 ASSERT_RTNL(); 8493 changelowerstate_info.lower_state_info = lower_state_info; 8494 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, 8495 &changelowerstate_info.info); 8496 } 8497 EXPORT_SYMBOL(netdev_lower_state_changed); 8498 8499 static void dev_change_rx_flags(struct net_device *dev, int flags) 8500 { 8501 const struct net_device_ops *ops = dev->netdev_ops; 8502 8503 if (ops->ndo_change_rx_flags) 8504 ops->ndo_change_rx_flags(dev, flags); 8505 } 8506 8507 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) 8508 { 8509 unsigned int old_flags = dev->flags; 8510 kuid_t uid; 8511 kgid_t gid; 8512 8513 ASSERT_RTNL(); 8514 8515 dev->flags |= IFF_PROMISC; 8516 dev->promiscuity += inc; 8517 if (dev->promiscuity == 0) { 8518 /* 8519 * Avoid overflow. 8520 * If inc causes overflow, untouch promisc and return error. 8521 */ 8522 if (inc < 0) 8523 dev->flags &= ~IFF_PROMISC; 8524 else { 8525 dev->promiscuity -= inc; 8526 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n", 8527 dev->name); 8528 return -EOVERFLOW; 8529 } 8530 } 8531 if (dev->flags != old_flags) { 8532 pr_info("device %s %s promiscuous mode\n", 8533 dev->name, 8534 dev->flags & IFF_PROMISC ? "entered" : "left"); 8535 if (audit_enabled) { 8536 current_uid_gid(&uid, &gid); 8537 audit_log(audit_context(), GFP_ATOMIC, 8538 AUDIT_ANOM_PROMISCUOUS, 8539 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", 8540 dev->name, (dev->flags & IFF_PROMISC), 8541 (old_flags & IFF_PROMISC), 8542 from_kuid(&init_user_ns, audit_get_loginuid(current)), 8543 from_kuid(&init_user_ns, uid), 8544 from_kgid(&init_user_ns, gid), 8545 audit_get_sessionid(current)); 8546 } 8547 8548 dev_change_rx_flags(dev, IFF_PROMISC); 8549 } 8550 if (notify) 8551 __dev_notify_flags(dev, old_flags, IFF_PROMISC); 8552 return 0; 8553 } 8554 8555 /** 8556 * dev_set_promiscuity - update promiscuity count on a device 8557 * @dev: device 8558 * @inc: modifier 8559 * 8560 * Add or remove promiscuity from a device. While the count in the device 8561 * remains above zero the interface remains promiscuous. Once it hits zero 8562 * the device reverts back to normal filtering operation. A negative inc 8563 * value is used to drop promiscuity on the device. 8564 * Return 0 if successful or a negative errno code on error. 8565 */ 8566 int dev_set_promiscuity(struct net_device *dev, int inc) 8567 { 8568 unsigned int old_flags = dev->flags; 8569 int err; 8570 8571 err = __dev_set_promiscuity(dev, inc, true); 8572 if (err < 0) 8573 return err; 8574 if (dev->flags != old_flags) 8575 dev_set_rx_mode(dev); 8576 return err; 8577 } 8578 EXPORT_SYMBOL(dev_set_promiscuity); 8579 8580 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) 8581 { 8582 unsigned int old_flags = dev->flags, old_gflags = dev->gflags; 8583 8584 ASSERT_RTNL(); 8585 8586 dev->flags |= IFF_ALLMULTI; 8587 dev->allmulti += inc; 8588 if (dev->allmulti == 0) { 8589 /* 8590 * Avoid overflow. 8591 * If inc causes overflow, untouch allmulti and return error. 8592 */ 8593 if (inc < 0) 8594 dev->flags &= ~IFF_ALLMULTI; 8595 else { 8596 dev->allmulti -= inc; 8597 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n", 8598 dev->name); 8599 return -EOVERFLOW; 8600 } 8601 } 8602 if (dev->flags ^ old_flags) { 8603 dev_change_rx_flags(dev, IFF_ALLMULTI); 8604 dev_set_rx_mode(dev); 8605 if (notify) 8606 __dev_notify_flags(dev, old_flags, 8607 dev->gflags ^ old_gflags); 8608 } 8609 return 0; 8610 } 8611 8612 /** 8613 * dev_set_allmulti - update allmulti count on a device 8614 * @dev: device 8615 * @inc: modifier 8616 * 8617 * Add or remove reception of all multicast frames to a device. While the 8618 * count in the device remains above zero the interface remains listening 8619 * to all interfaces. Once it hits zero the device reverts back to normal 8620 * filtering operation. A negative @inc value is used to drop the counter 8621 * when releasing a resource needing all multicasts. 8622 * Return 0 if successful or a negative errno code on error. 8623 */ 8624 8625 int dev_set_allmulti(struct net_device *dev, int inc) 8626 { 8627 return __dev_set_allmulti(dev, inc, true); 8628 } 8629 EXPORT_SYMBOL(dev_set_allmulti); 8630 8631 /* 8632 * Upload unicast and multicast address lists to device and 8633 * configure RX filtering. When the device doesn't support unicast 8634 * filtering it is put in promiscuous mode while unicast addresses 8635 * are present. 8636 */ 8637 void __dev_set_rx_mode(struct net_device *dev) 8638 { 8639 const struct net_device_ops *ops = dev->netdev_ops; 8640 8641 /* dev_open will call this function so the list will stay sane. */ 8642 if (!(dev->flags&IFF_UP)) 8643 return; 8644 8645 if (!netif_device_present(dev)) 8646 return; 8647 8648 if (!(dev->priv_flags & IFF_UNICAST_FLT)) { 8649 /* Unicast addresses changes may only happen under the rtnl, 8650 * therefore calling __dev_set_promiscuity here is safe. 8651 */ 8652 if (!netdev_uc_empty(dev) && !dev->uc_promisc) { 8653 __dev_set_promiscuity(dev, 1, false); 8654 dev->uc_promisc = true; 8655 } else if (netdev_uc_empty(dev) && dev->uc_promisc) { 8656 __dev_set_promiscuity(dev, -1, false); 8657 dev->uc_promisc = false; 8658 } 8659 } 8660 8661 if (ops->ndo_set_rx_mode) 8662 ops->ndo_set_rx_mode(dev); 8663 } 8664 8665 void dev_set_rx_mode(struct net_device *dev) 8666 { 8667 netif_addr_lock_bh(dev); 8668 __dev_set_rx_mode(dev); 8669 netif_addr_unlock_bh(dev); 8670 } 8671 8672 /** 8673 * dev_get_flags - get flags reported to userspace 8674 * @dev: device 8675 * 8676 * Get the combination of flag bits exported through APIs to userspace. 8677 */ 8678 unsigned int dev_get_flags(const struct net_device *dev) 8679 { 8680 unsigned int flags; 8681 8682 flags = (dev->flags & ~(IFF_PROMISC | 8683 IFF_ALLMULTI | 8684 IFF_RUNNING | 8685 IFF_LOWER_UP | 8686 IFF_DORMANT)) | 8687 (dev->gflags & (IFF_PROMISC | 8688 IFF_ALLMULTI)); 8689 8690 if (netif_running(dev)) { 8691 if (netif_oper_up(dev)) 8692 flags |= IFF_RUNNING; 8693 if (netif_carrier_ok(dev)) 8694 flags |= IFF_LOWER_UP; 8695 if (netif_dormant(dev)) 8696 flags |= IFF_DORMANT; 8697 } 8698 8699 return flags; 8700 } 8701 EXPORT_SYMBOL(dev_get_flags); 8702 8703 int __dev_change_flags(struct net_device *dev, unsigned int flags, 8704 struct netlink_ext_ack *extack) 8705 { 8706 unsigned int old_flags = dev->flags; 8707 int ret; 8708 8709 ASSERT_RTNL(); 8710 8711 /* 8712 * Set the flags on our device. 8713 */ 8714 8715 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | 8716 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | 8717 IFF_AUTOMEDIA)) | 8718 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | 8719 IFF_ALLMULTI)); 8720 8721 /* 8722 * Load in the correct multicast list now the flags have changed. 8723 */ 8724 8725 if ((old_flags ^ flags) & IFF_MULTICAST) 8726 dev_change_rx_flags(dev, IFF_MULTICAST); 8727 8728 dev_set_rx_mode(dev); 8729 8730 /* 8731 * Have we downed the interface. We handle IFF_UP ourselves 8732 * according to user attempts to set it, rather than blindly 8733 * setting it. 8734 */ 8735 8736 ret = 0; 8737 if ((old_flags ^ flags) & IFF_UP) { 8738 if (old_flags & IFF_UP) 8739 __dev_close(dev); 8740 else 8741 ret = __dev_open(dev, extack); 8742 } 8743 8744 if ((flags ^ dev->gflags) & IFF_PROMISC) { 8745 int inc = (flags & IFF_PROMISC) ? 1 : -1; 8746 unsigned int old_flags = dev->flags; 8747 8748 dev->gflags ^= IFF_PROMISC; 8749 8750 if (__dev_set_promiscuity(dev, inc, false) >= 0) 8751 if (dev->flags != old_flags) 8752 dev_set_rx_mode(dev); 8753 } 8754 8755 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI 8756 * is important. Some (broken) drivers set IFF_PROMISC, when 8757 * IFF_ALLMULTI is requested not asking us and not reporting. 8758 */ 8759 if ((flags ^ dev->gflags) & IFF_ALLMULTI) { 8760 int inc = (flags & IFF_ALLMULTI) ? 1 : -1; 8761 8762 dev->gflags ^= IFF_ALLMULTI; 8763 __dev_set_allmulti(dev, inc, false); 8764 } 8765 8766 return ret; 8767 } 8768 8769 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, 8770 unsigned int gchanges) 8771 { 8772 unsigned int changes = dev->flags ^ old_flags; 8773 8774 if (gchanges) 8775 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC); 8776 8777 if (changes & IFF_UP) { 8778 if (dev->flags & IFF_UP) 8779 call_netdevice_notifiers(NETDEV_UP, dev); 8780 else 8781 call_netdevice_notifiers(NETDEV_DOWN, dev); 8782 } 8783 8784 if (dev->flags & IFF_UP && 8785 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { 8786 struct netdev_notifier_change_info change_info = { 8787 .info = { 8788 .dev = dev, 8789 }, 8790 .flags_changed = changes, 8791 }; 8792 8793 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info); 8794 } 8795 } 8796 8797 /** 8798 * dev_change_flags - change device settings 8799 * @dev: device 8800 * @flags: device state flags 8801 * @extack: netlink extended ack 8802 * 8803 * Change settings on device based state flags. The flags are 8804 * in the userspace exported format. 8805 */ 8806 int dev_change_flags(struct net_device *dev, unsigned int flags, 8807 struct netlink_ext_ack *extack) 8808 { 8809 int ret; 8810 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; 8811 8812 ret = __dev_change_flags(dev, flags, extack); 8813 if (ret < 0) 8814 return ret; 8815 8816 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); 8817 __dev_notify_flags(dev, old_flags, changes); 8818 return ret; 8819 } 8820 EXPORT_SYMBOL(dev_change_flags); 8821 8822 int __dev_set_mtu(struct net_device *dev, int new_mtu) 8823 { 8824 const struct net_device_ops *ops = dev->netdev_ops; 8825 8826 if (ops->ndo_change_mtu) 8827 return ops->ndo_change_mtu(dev, new_mtu); 8828 8829 /* Pairs with all the lockless reads of dev->mtu in the stack */ 8830 WRITE_ONCE(dev->mtu, new_mtu); 8831 return 0; 8832 } 8833 EXPORT_SYMBOL(__dev_set_mtu); 8834 8835 int dev_validate_mtu(struct net_device *dev, int new_mtu, 8836 struct netlink_ext_ack *extack) 8837 { 8838 /* MTU must be positive, and in range */ 8839 if (new_mtu < 0 || new_mtu < dev->min_mtu) { 8840 NL_SET_ERR_MSG(extack, "mtu less than device minimum"); 8841 return -EINVAL; 8842 } 8843 8844 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { 8845 NL_SET_ERR_MSG(extack, "mtu greater than device maximum"); 8846 return -EINVAL; 8847 } 8848 return 0; 8849 } 8850 8851 /** 8852 * dev_set_mtu_ext - Change maximum transfer unit 8853 * @dev: device 8854 * @new_mtu: new transfer unit 8855 * @extack: netlink extended ack 8856 * 8857 * Change the maximum transfer size of the network device. 8858 */ 8859 int dev_set_mtu_ext(struct net_device *dev, int new_mtu, 8860 struct netlink_ext_ack *extack) 8861 { 8862 int err, orig_mtu; 8863 8864 if (new_mtu == dev->mtu) 8865 return 0; 8866 8867 err = dev_validate_mtu(dev, new_mtu, extack); 8868 if (err) 8869 return err; 8870 8871 if (!netif_device_present(dev)) 8872 return -ENODEV; 8873 8874 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); 8875 err = notifier_to_errno(err); 8876 if (err) 8877 return err; 8878 8879 orig_mtu = dev->mtu; 8880 err = __dev_set_mtu(dev, new_mtu); 8881 8882 if (!err) { 8883 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8884 orig_mtu); 8885 err = notifier_to_errno(err); 8886 if (err) { 8887 /* setting mtu back and notifying everyone again, 8888 * so that they have a chance to revert changes. 8889 */ 8890 __dev_set_mtu(dev, orig_mtu); 8891 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8892 new_mtu); 8893 } 8894 } 8895 return err; 8896 } 8897 8898 int dev_set_mtu(struct net_device *dev, int new_mtu) 8899 { 8900 struct netlink_ext_ack extack; 8901 int err; 8902 8903 memset(&extack, 0, sizeof(extack)); 8904 err = dev_set_mtu_ext(dev, new_mtu, &extack); 8905 if (err && extack._msg) 8906 net_err_ratelimited("%s: %s\n", dev->name, extack._msg); 8907 return err; 8908 } 8909 EXPORT_SYMBOL(dev_set_mtu); 8910 8911 /** 8912 * dev_change_tx_queue_len - Change TX queue length of a netdevice 8913 * @dev: device 8914 * @new_len: new tx queue length 8915 */ 8916 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len) 8917 { 8918 unsigned int orig_len = dev->tx_queue_len; 8919 int res; 8920 8921 if (new_len != (unsigned int)new_len) 8922 return -ERANGE; 8923 8924 if (new_len != orig_len) { 8925 dev->tx_queue_len = new_len; 8926 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev); 8927 res = notifier_to_errno(res); 8928 if (res) 8929 goto err_rollback; 8930 res = dev_qdisc_change_tx_queue_len(dev); 8931 if (res) 8932 goto err_rollback; 8933 } 8934 8935 return 0; 8936 8937 err_rollback: 8938 netdev_err(dev, "refused to change device tx_queue_len\n"); 8939 dev->tx_queue_len = orig_len; 8940 return res; 8941 } 8942 8943 /** 8944 * dev_set_group - Change group this device belongs to 8945 * @dev: device 8946 * @new_group: group this device should belong to 8947 */ 8948 void dev_set_group(struct net_device *dev, int new_group) 8949 { 8950 dev->group = new_group; 8951 } 8952 EXPORT_SYMBOL(dev_set_group); 8953 8954 /** 8955 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR. 8956 * @dev: device 8957 * @addr: new address 8958 * @extack: netlink extended ack 8959 */ 8960 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr, 8961 struct netlink_ext_ack *extack) 8962 { 8963 struct netdev_notifier_pre_changeaddr_info info = { 8964 .info.dev = dev, 8965 .info.extack = extack, 8966 .dev_addr = addr, 8967 }; 8968 int rc; 8969 8970 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info); 8971 return notifier_to_errno(rc); 8972 } 8973 EXPORT_SYMBOL(dev_pre_changeaddr_notify); 8974 8975 /** 8976 * dev_set_mac_address - Change Media Access Control Address 8977 * @dev: device 8978 * @sa: new address 8979 * @extack: netlink extended ack 8980 * 8981 * Change the hardware (MAC) address of the device 8982 */ 8983 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa, 8984 struct netlink_ext_ack *extack) 8985 { 8986 const struct net_device_ops *ops = dev->netdev_ops; 8987 int err; 8988 8989 if (!ops->ndo_set_mac_address) 8990 return -EOPNOTSUPP; 8991 if (sa->sa_family != dev->type) 8992 return -EINVAL; 8993 if (!netif_device_present(dev)) 8994 return -ENODEV; 8995 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack); 8996 if (err) 8997 return err; 8998 err = ops->ndo_set_mac_address(dev, sa); 8999 if (err) 9000 return err; 9001 dev->addr_assign_type = NET_ADDR_SET; 9002 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); 9003 add_device_randomness(dev->dev_addr, dev->addr_len); 9004 return 0; 9005 } 9006 EXPORT_SYMBOL(dev_set_mac_address); 9007 9008 static DECLARE_RWSEM(dev_addr_sem); 9009 9010 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa, 9011 struct netlink_ext_ack *extack) 9012 { 9013 int ret; 9014 9015 down_write(&dev_addr_sem); 9016 ret = dev_set_mac_address(dev, sa, extack); 9017 up_write(&dev_addr_sem); 9018 return ret; 9019 } 9020 EXPORT_SYMBOL(dev_set_mac_address_user); 9021 9022 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name) 9023 { 9024 size_t size = sizeof(sa->sa_data); 9025 struct net_device *dev; 9026 int ret = 0; 9027 9028 down_read(&dev_addr_sem); 9029 rcu_read_lock(); 9030 9031 dev = dev_get_by_name_rcu(net, dev_name); 9032 if (!dev) { 9033 ret = -ENODEV; 9034 goto unlock; 9035 } 9036 if (!dev->addr_len) 9037 memset(sa->sa_data, 0, size); 9038 else 9039 memcpy(sa->sa_data, dev->dev_addr, 9040 min_t(size_t, size, dev->addr_len)); 9041 sa->sa_family = dev->type; 9042 9043 unlock: 9044 rcu_read_unlock(); 9045 up_read(&dev_addr_sem); 9046 return ret; 9047 } 9048 EXPORT_SYMBOL(dev_get_mac_address); 9049 9050 /** 9051 * dev_change_carrier - Change device carrier 9052 * @dev: device 9053 * @new_carrier: new value 9054 * 9055 * Change device carrier 9056 */ 9057 int dev_change_carrier(struct net_device *dev, bool new_carrier) 9058 { 9059 const struct net_device_ops *ops = dev->netdev_ops; 9060 9061 if (!ops->ndo_change_carrier) 9062 return -EOPNOTSUPP; 9063 if (!netif_device_present(dev)) 9064 return -ENODEV; 9065 return ops->ndo_change_carrier(dev, new_carrier); 9066 } 9067 EXPORT_SYMBOL(dev_change_carrier); 9068 9069 /** 9070 * dev_get_phys_port_id - Get device physical port ID 9071 * @dev: device 9072 * @ppid: port ID 9073 * 9074 * Get device physical port ID 9075 */ 9076 int dev_get_phys_port_id(struct net_device *dev, 9077 struct netdev_phys_item_id *ppid) 9078 { 9079 const struct net_device_ops *ops = dev->netdev_ops; 9080 9081 if (!ops->ndo_get_phys_port_id) 9082 return -EOPNOTSUPP; 9083 return ops->ndo_get_phys_port_id(dev, ppid); 9084 } 9085 EXPORT_SYMBOL(dev_get_phys_port_id); 9086 9087 /** 9088 * dev_get_phys_port_name - Get device physical port name 9089 * @dev: device 9090 * @name: port name 9091 * @len: limit of bytes to copy to name 9092 * 9093 * Get device physical port name 9094 */ 9095 int dev_get_phys_port_name(struct net_device *dev, 9096 char *name, size_t len) 9097 { 9098 const struct net_device_ops *ops = dev->netdev_ops; 9099 int err; 9100 9101 if (ops->ndo_get_phys_port_name) { 9102 err = ops->ndo_get_phys_port_name(dev, name, len); 9103 if (err != -EOPNOTSUPP) 9104 return err; 9105 } 9106 return devlink_compat_phys_port_name_get(dev, name, len); 9107 } 9108 EXPORT_SYMBOL(dev_get_phys_port_name); 9109 9110 /** 9111 * dev_get_port_parent_id - Get the device's port parent identifier 9112 * @dev: network device 9113 * @ppid: pointer to a storage for the port's parent identifier 9114 * @recurse: allow/disallow recursion to lower devices 9115 * 9116 * Get the devices's port parent identifier 9117 */ 9118 int dev_get_port_parent_id(struct net_device *dev, 9119 struct netdev_phys_item_id *ppid, 9120 bool recurse) 9121 { 9122 const struct net_device_ops *ops = dev->netdev_ops; 9123 struct netdev_phys_item_id first = { }; 9124 struct net_device *lower_dev; 9125 struct list_head *iter; 9126 int err; 9127 9128 if (ops->ndo_get_port_parent_id) { 9129 err = ops->ndo_get_port_parent_id(dev, ppid); 9130 if (err != -EOPNOTSUPP) 9131 return err; 9132 } 9133 9134 err = devlink_compat_switch_id_get(dev, ppid); 9135 if (!err || err != -EOPNOTSUPP) 9136 return err; 9137 9138 if (!recurse) 9139 return -EOPNOTSUPP; 9140 9141 netdev_for_each_lower_dev(dev, lower_dev, iter) { 9142 err = dev_get_port_parent_id(lower_dev, ppid, recurse); 9143 if (err) 9144 break; 9145 if (!first.id_len) 9146 first = *ppid; 9147 else if (memcmp(&first, ppid, sizeof(*ppid))) 9148 return -EOPNOTSUPP; 9149 } 9150 9151 return err; 9152 } 9153 EXPORT_SYMBOL(dev_get_port_parent_id); 9154 9155 /** 9156 * netdev_port_same_parent_id - Indicate if two network devices have 9157 * the same port parent identifier 9158 * @a: first network device 9159 * @b: second network device 9160 */ 9161 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b) 9162 { 9163 struct netdev_phys_item_id a_id = { }; 9164 struct netdev_phys_item_id b_id = { }; 9165 9166 if (dev_get_port_parent_id(a, &a_id, true) || 9167 dev_get_port_parent_id(b, &b_id, true)) 9168 return false; 9169 9170 return netdev_phys_item_id_same(&a_id, &b_id); 9171 } 9172 EXPORT_SYMBOL(netdev_port_same_parent_id); 9173 9174 /** 9175 * dev_change_proto_down - update protocol port state information 9176 * @dev: device 9177 * @proto_down: new value 9178 * 9179 * This info can be used by switch drivers to set the phys state of the 9180 * port. 9181 */ 9182 int dev_change_proto_down(struct net_device *dev, bool proto_down) 9183 { 9184 const struct net_device_ops *ops = dev->netdev_ops; 9185 9186 if (!ops->ndo_change_proto_down) 9187 return -EOPNOTSUPP; 9188 if (!netif_device_present(dev)) 9189 return -ENODEV; 9190 return ops->ndo_change_proto_down(dev, proto_down); 9191 } 9192 EXPORT_SYMBOL(dev_change_proto_down); 9193 9194 /** 9195 * dev_change_proto_down_generic - generic implementation for 9196 * ndo_change_proto_down that sets carrier according to 9197 * proto_down. 9198 * 9199 * @dev: device 9200 * @proto_down: new value 9201 */ 9202 int dev_change_proto_down_generic(struct net_device *dev, bool proto_down) 9203 { 9204 if (proto_down) 9205 netif_carrier_off(dev); 9206 else 9207 netif_carrier_on(dev); 9208 dev->proto_down = proto_down; 9209 return 0; 9210 } 9211 EXPORT_SYMBOL(dev_change_proto_down_generic); 9212 9213 /** 9214 * dev_change_proto_down_reason - proto down reason 9215 * 9216 * @dev: device 9217 * @mask: proto down mask 9218 * @value: proto down value 9219 */ 9220 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask, 9221 u32 value) 9222 { 9223 int b; 9224 9225 if (!mask) { 9226 dev->proto_down_reason = value; 9227 } else { 9228 for_each_set_bit(b, &mask, 32) { 9229 if (value & (1 << b)) 9230 dev->proto_down_reason |= BIT(b); 9231 else 9232 dev->proto_down_reason &= ~BIT(b); 9233 } 9234 } 9235 } 9236 EXPORT_SYMBOL(dev_change_proto_down_reason); 9237 9238 struct bpf_xdp_link { 9239 struct bpf_link link; 9240 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */ 9241 int flags; 9242 }; 9243 9244 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags) 9245 { 9246 if (flags & XDP_FLAGS_HW_MODE) 9247 return XDP_MODE_HW; 9248 if (flags & XDP_FLAGS_DRV_MODE) 9249 return XDP_MODE_DRV; 9250 if (flags & XDP_FLAGS_SKB_MODE) 9251 return XDP_MODE_SKB; 9252 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB; 9253 } 9254 9255 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode) 9256 { 9257 switch (mode) { 9258 case XDP_MODE_SKB: 9259 return generic_xdp_install; 9260 case XDP_MODE_DRV: 9261 case XDP_MODE_HW: 9262 return dev->netdev_ops->ndo_bpf; 9263 default: 9264 return NULL; 9265 } 9266 } 9267 9268 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev, 9269 enum bpf_xdp_mode mode) 9270 { 9271 return dev->xdp_state[mode].link; 9272 } 9273 9274 static struct bpf_prog *dev_xdp_prog(struct net_device *dev, 9275 enum bpf_xdp_mode mode) 9276 { 9277 struct bpf_xdp_link *link = dev_xdp_link(dev, mode); 9278 9279 if (link) 9280 return link->link.prog; 9281 return dev->xdp_state[mode].prog; 9282 } 9283 9284 static u8 dev_xdp_prog_count(struct net_device *dev) 9285 { 9286 u8 count = 0; 9287 int i; 9288 9289 for (i = 0; i < __MAX_XDP_MODE; i++) 9290 if (dev->xdp_state[i].prog || dev->xdp_state[i].link) 9291 count++; 9292 return count; 9293 } 9294 9295 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode) 9296 { 9297 struct bpf_prog *prog = dev_xdp_prog(dev, mode); 9298 9299 return prog ? prog->aux->id : 0; 9300 } 9301 9302 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode, 9303 struct bpf_xdp_link *link) 9304 { 9305 dev->xdp_state[mode].link = link; 9306 dev->xdp_state[mode].prog = NULL; 9307 } 9308 9309 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode, 9310 struct bpf_prog *prog) 9311 { 9312 dev->xdp_state[mode].link = NULL; 9313 dev->xdp_state[mode].prog = prog; 9314 } 9315 9316 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode, 9317 bpf_op_t bpf_op, struct netlink_ext_ack *extack, 9318 u32 flags, struct bpf_prog *prog) 9319 { 9320 struct netdev_bpf xdp; 9321 int err; 9322 9323 memset(&xdp, 0, sizeof(xdp)); 9324 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG; 9325 xdp.extack = extack; 9326 xdp.flags = flags; 9327 xdp.prog = prog; 9328 9329 /* Drivers assume refcnt is already incremented (i.e, prog pointer is 9330 * "moved" into driver), so they don't increment it on their own, but 9331 * they do decrement refcnt when program is detached or replaced. 9332 * Given net_device also owns link/prog, we need to bump refcnt here 9333 * to prevent drivers from underflowing it. 9334 */ 9335 if (prog) 9336 bpf_prog_inc(prog); 9337 err = bpf_op(dev, &xdp); 9338 if (err) { 9339 if (prog) 9340 bpf_prog_put(prog); 9341 return err; 9342 } 9343 9344 if (mode != XDP_MODE_HW) 9345 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog); 9346 9347 return 0; 9348 } 9349 9350 static void dev_xdp_uninstall(struct net_device *dev) 9351 { 9352 struct bpf_xdp_link *link; 9353 struct bpf_prog *prog; 9354 enum bpf_xdp_mode mode; 9355 bpf_op_t bpf_op; 9356 9357 ASSERT_RTNL(); 9358 9359 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) { 9360 prog = dev_xdp_prog(dev, mode); 9361 if (!prog) 9362 continue; 9363 9364 bpf_op = dev_xdp_bpf_op(dev, mode); 9365 if (!bpf_op) 9366 continue; 9367 9368 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9369 9370 /* auto-detach link from net device */ 9371 link = dev_xdp_link(dev, mode); 9372 if (link) 9373 link->dev = NULL; 9374 else 9375 bpf_prog_put(prog); 9376 9377 dev_xdp_set_link(dev, mode, NULL); 9378 } 9379 } 9380 9381 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack, 9382 struct bpf_xdp_link *link, struct bpf_prog *new_prog, 9383 struct bpf_prog *old_prog, u32 flags) 9384 { 9385 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES); 9386 struct bpf_prog *cur_prog; 9387 enum bpf_xdp_mode mode; 9388 bpf_op_t bpf_op; 9389 int err; 9390 9391 ASSERT_RTNL(); 9392 9393 /* either link or prog attachment, never both */ 9394 if (link && (new_prog || old_prog)) 9395 return -EINVAL; 9396 /* link supports only XDP mode flags */ 9397 if (link && (flags & ~XDP_FLAGS_MODES)) { 9398 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment"); 9399 return -EINVAL; 9400 } 9401 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */ 9402 if (num_modes > 1) { 9403 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set"); 9404 return -EINVAL; 9405 } 9406 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */ 9407 if (!num_modes && dev_xdp_prog_count(dev) > 1) { 9408 NL_SET_ERR_MSG(extack, 9409 "More than one program loaded, unset mode is ambiguous"); 9410 return -EINVAL; 9411 } 9412 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */ 9413 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) { 9414 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified"); 9415 return -EINVAL; 9416 } 9417 9418 mode = dev_xdp_mode(dev, flags); 9419 /* can't replace attached link */ 9420 if (dev_xdp_link(dev, mode)) { 9421 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link"); 9422 return -EBUSY; 9423 } 9424 9425 cur_prog = dev_xdp_prog(dev, mode); 9426 /* can't replace attached prog with link */ 9427 if (link && cur_prog) { 9428 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link"); 9429 return -EBUSY; 9430 } 9431 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) { 9432 NL_SET_ERR_MSG(extack, "Active program does not match expected"); 9433 return -EEXIST; 9434 } 9435 9436 /* put effective new program into new_prog */ 9437 if (link) 9438 new_prog = link->link.prog; 9439 9440 if (new_prog) { 9441 bool offload = mode == XDP_MODE_HW; 9442 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB 9443 ? XDP_MODE_DRV : XDP_MODE_SKB; 9444 9445 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) { 9446 NL_SET_ERR_MSG(extack, "XDP program already attached"); 9447 return -EBUSY; 9448 } 9449 if (!offload && dev_xdp_prog(dev, other_mode)) { 9450 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time"); 9451 return -EEXIST; 9452 } 9453 if (!offload && bpf_prog_is_dev_bound(new_prog->aux)) { 9454 NL_SET_ERR_MSG(extack, "Using device-bound program without HW_MODE flag is not supported"); 9455 return -EINVAL; 9456 } 9457 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) { 9458 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device"); 9459 return -EINVAL; 9460 } 9461 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) { 9462 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device"); 9463 return -EINVAL; 9464 } 9465 } 9466 9467 /* don't call drivers if the effective program didn't change */ 9468 if (new_prog != cur_prog) { 9469 bpf_op = dev_xdp_bpf_op(dev, mode); 9470 if (!bpf_op) { 9471 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode"); 9472 return -EOPNOTSUPP; 9473 } 9474 9475 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog); 9476 if (err) 9477 return err; 9478 } 9479 9480 if (link) 9481 dev_xdp_set_link(dev, mode, link); 9482 else 9483 dev_xdp_set_prog(dev, mode, new_prog); 9484 if (cur_prog) 9485 bpf_prog_put(cur_prog); 9486 9487 return 0; 9488 } 9489 9490 static int dev_xdp_attach_link(struct net_device *dev, 9491 struct netlink_ext_ack *extack, 9492 struct bpf_xdp_link *link) 9493 { 9494 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags); 9495 } 9496 9497 static int dev_xdp_detach_link(struct net_device *dev, 9498 struct netlink_ext_ack *extack, 9499 struct bpf_xdp_link *link) 9500 { 9501 enum bpf_xdp_mode mode; 9502 bpf_op_t bpf_op; 9503 9504 ASSERT_RTNL(); 9505 9506 mode = dev_xdp_mode(dev, link->flags); 9507 if (dev_xdp_link(dev, mode) != link) 9508 return -EINVAL; 9509 9510 bpf_op = dev_xdp_bpf_op(dev, mode); 9511 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9512 dev_xdp_set_link(dev, mode, NULL); 9513 return 0; 9514 } 9515 9516 static void bpf_xdp_link_release(struct bpf_link *link) 9517 { 9518 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9519 9520 rtnl_lock(); 9521 9522 /* if racing with net_device's tear down, xdp_link->dev might be 9523 * already NULL, in which case link was already auto-detached 9524 */ 9525 if (xdp_link->dev) { 9526 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link)); 9527 xdp_link->dev = NULL; 9528 } 9529 9530 rtnl_unlock(); 9531 } 9532 9533 static int bpf_xdp_link_detach(struct bpf_link *link) 9534 { 9535 bpf_xdp_link_release(link); 9536 return 0; 9537 } 9538 9539 static void bpf_xdp_link_dealloc(struct bpf_link *link) 9540 { 9541 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9542 9543 kfree(xdp_link); 9544 } 9545 9546 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link, 9547 struct seq_file *seq) 9548 { 9549 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9550 u32 ifindex = 0; 9551 9552 rtnl_lock(); 9553 if (xdp_link->dev) 9554 ifindex = xdp_link->dev->ifindex; 9555 rtnl_unlock(); 9556 9557 seq_printf(seq, "ifindex:\t%u\n", ifindex); 9558 } 9559 9560 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link, 9561 struct bpf_link_info *info) 9562 { 9563 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9564 u32 ifindex = 0; 9565 9566 rtnl_lock(); 9567 if (xdp_link->dev) 9568 ifindex = xdp_link->dev->ifindex; 9569 rtnl_unlock(); 9570 9571 info->xdp.ifindex = ifindex; 9572 return 0; 9573 } 9574 9575 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog, 9576 struct bpf_prog *old_prog) 9577 { 9578 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9579 enum bpf_xdp_mode mode; 9580 bpf_op_t bpf_op; 9581 int err = 0; 9582 9583 rtnl_lock(); 9584 9585 /* link might have been auto-released already, so fail */ 9586 if (!xdp_link->dev) { 9587 err = -ENOLINK; 9588 goto out_unlock; 9589 } 9590 9591 if (old_prog && link->prog != old_prog) { 9592 err = -EPERM; 9593 goto out_unlock; 9594 } 9595 old_prog = link->prog; 9596 if (old_prog == new_prog) { 9597 /* no-op, don't disturb drivers */ 9598 bpf_prog_put(new_prog); 9599 goto out_unlock; 9600 } 9601 9602 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags); 9603 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode); 9604 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL, 9605 xdp_link->flags, new_prog); 9606 if (err) 9607 goto out_unlock; 9608 9609 old_prog = xchg(&link->prog, new_prog); 9610 bpf_prog_put(old_prog); 9611 9612 out_unlock: 9613 rtnl_unlock(); 9614 return err; 9615 } 9616 9617 static const struct bpf_link_ops bpf_xdp_link_lops = { 9618 .release = bpf_xdp_link_release, 9619 .dealloc = bpf_xdp_link_dealloc, 9620 .detach = bpf_xdp_link_detach, 9621 .show_fdinfo = bpf_xdp_link_show_fdinfo, 9622 .fill_link_info = bpf_xdp_link_fill_link_info, 9623 .update_prog = bpf_xdp_link_update, 9624 }; 9625 9626 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog) 9627 { 9628 struct net *net = current->nsproxy->net_ns; 9629 struct bpf_link_primer link_primer; 9630 struct bpf_xdp_link *link; 9631 struct net_device *dev; 9632 int err, fd; 9633 9634 dev = dev_get_by_index(net, attr->link_create.target_ifindex); 9635 if (!dev) 9636 return -EINVAL; 9637 9638 link = kzalloc(sizeof(*link), GFP_USER); 9639 if (!link) { 9640 err = -ENOMEM; 9641 goto out_put_dev; 9642 } 9643 9644 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog); 9645 link->dev = dev; 9646 link->flags = attr->link_create.flags; 9647 9648 err = bpf_link_prime(&link->link, &link_primer); 9649 if (err) { 9650 kfree(link); 9651 goto out_put_dev; 9652 } 9653 9654 rtnl_lock(); 9655 err = dev_xdp_attach_link(dev, NULL, link); 9656 rtnl_unlock(); 9657 9658 if (err) { 9659 bpf_link_cleanup(&link_primer); 9660 goto out_put_dev; 9661 } 9662 9663 fd = bpf_link_settle(&link_primer); 9664 /* link itself doesn't hold dev's refcnt to not complicate shutdown */ 9665 dev_put(dev); 9666 return fd; 9667 9668 out_put_dev: 9669 dev_put(dev); 9670 return err; 9671 } 9672 9673 /** 9674 * dev_change_xdp_fd - set or clear a bpf program for a device rx path 9675 * @dev: device 9676 * @extack: netlink extended ack 9677 * @fd: new program fd or negative value to clear 9678 * @expected_fd: old program fd that userspace expects to replace or clear 9679 * @flags: xdp-related flags 9680 * 9681 * Set or clear a bpf program for a device 9682 */ 9683 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, 9684 int fd, int expected_fd, u32 flags) 9685 { 9686 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags); 9687 struct bpf_prog *new_prog = NULL, *old_prog = NULL; 9688 int err; 9689 9690 ASSERT_RTNL(); 9691 9692 if (fd >= 0) { 9693 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP, 9694 mode != XDP_MODE_SKB); 9695 if (IS_ERR(new_prog)) 9696 return PTR_ERR(new_prog); 9697 } 9698 9699 if (expected_fd >= 0) { 9700 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP, 9701 mode != XDP_MODE_SKB); 9702 if (IS_ERR(old_prog)) { 9703 err = PTR_ERR(old_prog); 9704 old_prog = NULL; 9705 goto err_out; 9706 } 9707 } 9708 9709 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags); 9710 9711 err_out: 9712 if (err && new_prog) 9713 bpf_prog_put(new_prog); 9714 if (old_prog) 9715 bpf_prog_put(old_prog); 9716 return err; 9717 } 9718 9719 /** 9720 * dev_new_index - allocate an ifindex 9721 * @net: the applicable net namespace 9722 * 9723 * Returns a suitable unique value for a new device interface 9724 * number. The caller must hold the rtnl semaphore or the 9725 * dev_base_lock to be sure it remains unique. 9726 */ 9727 static int dev_new_index(struct net *net) 9728 { 9729 int ifindex = net->ifindex; 9730 9731 for (;;) { 9732 if (++ifindex <= 0) 9733 ifindex = 1; 9734 if (!__dev_get_by_index(net, ifindex)) 9735 return net->ifindex = ifindex; 9736 } 9737 } 9738 9739 /* Delayed registration/unregisteration */ 9740 static LIST_HEAD(net_todo_list); 9741 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); 9742 9743 static void net_set_todo(struct net_device *dev) 9744 { 9745 list_add_tail(&dev->todo_list, &net_todo_list); 9746 dev_net(dev)->dev_unreg_count++; 9747 } 9748 9749 static netdev_features_t netdev_sync_upper_features(struct net_device *lower, 9750 struct net_device *upper, 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 (!(upper->wanted_features & feature) 9759 && (features & feature)) { 9760 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", 9761 &feature, upper->name); 9762 features &= ~feature; 9763 } 9764 } 9765 9766 return features; 9767 } 9768 9769 static void netdev_sync_lower_features(struct net_device *upper, 9770 struct net_device *lower, netdev_features_t features) 9771 { 9772 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9773 netdev_features_t feature; 9774 int feature_bit; 9775 9776 for_each_netdev_feature(upper_disables, feature_bit) { 9777 feature = __NETIF_F_BIT(feature_bit); 9778 if (!(features & feature) && (lower->features & feature)) { 9779 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", 9780 &feature, lower->name); 9781 lower->wanted_features &= ~feature; 9782 __netdev_update_features(lower); 9783 9784 if (unlikely(lower->features & feature)) 9785 netdev_WARN(upper, "failed to disable %pNF on %s!\n", 9786 &feature, lower->name); 9787 else 9788 netdev_features_change(lower); 9789 } 9790 } 9791 } 9792 9793 static netdev_features_t netdev_fix_features(struct net_device *dev, 9794 netdev_features_t features) 9795 { 9796 /* Fix illegal checksum combinations */ 9797 if ((features & NETIF_F_HW_CSUM) && 9798 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { 9799 netdev_warn(dev, "mixed HW and IP checksum settings.\n"); 9800 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); 9801 } 9802 9803 /* TSO requires that SG is present as well. */ 9804 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { 9805 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); 9806 features &= ~NETIF_F_ALL_TSO; 9807 } 9808 9809 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && 9810 !(features & NETIF_F_IP_CSUM)) { 9811 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); 9812 features &= ~NETIF_F_TSO; 9813 features &= ~NETIF_F_TSO_ECN; 9814 } 9815 9816 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && 9817 !(features & NETIF_F_IPV6_CSUM)) { 9818 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); 9819 features &= ~NETIF_F_TSO6; 9820 } 9821 9822 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ 9823 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) 9824 features &= ~NETIF_F_TSO_MANGLEID; 9825 9826 /* TSO ECN requires that TSO is present as well. */ 9827 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) 9828 features &= ~NETIF_F_TSO_ECN; 9829 9830 /* Software GSO depends on SG. */ 9831 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { 9832 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); 9833 features &= ~NETIF_F_GSO; 9834 } 9835 9836 /* GSO partial features require GSO partial be set */ 9837 if ((features & dev->gso_partial_features) && 9838 !(features & NETIF_F_GSO_PARTIAL)) { 9839 netdev_dbg(dev, 9840 "Dropping partially supported GSO features since no GSO partial.\n"); 9841 features &= ~dev->gso_partial_features; 9842 } 9843 9844 if (!(features & NETIF_F_RXCSUM)) { 9845 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet 9846 * successfully merged by hardware must also have the 9847 * checksum verified by hardware. If the user does not 9848 * want to enable RXCSUM, logically, we should disable GRO_HW. 9849 */ 9850 if (features & NETIF_F_GRO_HW) { 9851 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n"); 9852 features &= ~NETIF_F_GRO_HW; 9853 } 9854 } 9855 9856 /* LRO/HW-GRO features cannot be combined with RX-FCS */ 9857 if (features & NETIF_F_RXFCS) { 9858 if (features & NETIF_F_LRO) { 9859 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n"); 9860 features &= ~NETIF_F_LRO; 9861 } 9862 9863 if (features & NETIF_F_GRO_HW) { 9864 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n"); 9865 features &= ~NETIF_F_GRO_HW; 9866 } 9867 } 9868 9869 if (features & NETIF_F_HW_TLS_TX) { 9870 bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) == 9871 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM); 9872 bool hw_csum = features & NETIF_F_HW_CSUM; 9873 9874 if (!ip_csum && !hw_csum) { 9875 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n"); 9876 features &= ~NETIF_F_HW_TLS_TX; 9877 } 9878 } 9879 9880 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) { 9881 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n"); 9882 features &= ~NETIF_F_HW_TLS_RX; 9883 } 9884 9885 return features; 9886 } 9887 9888 int __netdev_update_features(struct net_device *dev) 9889 { 9890 struct net_device *upper, *lower; 9891 netdev_features_t features; 9892 struct list_head *iter; 9893 int err = -1; 9894 9895 ASSERT_RTNL(); 9896 9897 features = netdev_get_wanted_features(dev); 9898 9899 if (dev->netdev_ops->ndo_fix_features) 9900 features = dev->netdev_ops->ndo_fix_features(dev, features); 9901 9902 /* driver might be less strict about feature dependencies */ 9903 features = netdev_fix_features(dev, features); 9904 9905 /* some features can't be enabled if they're off on an upper device */ 9906 netdev_for_each_upper_dev_rcu(dev, upper, iter) 9907 features = netdev_sync_upper_features(dev, upper, features); 9908 9909 if (dev->features == features) 9910 goto sync_lower; 9911 9912 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", 9913 &dev->features, &features); 9914 9915 if (dev->netdev_ops->ndo_set_features) 9916 err = dev->netdev_ops->ndo_set_features(dev, features); 9917 else 9918 err = 0; 9919 9920 if (unlikely(err < 0)) { 9921 netdev_err(dev, 9922 "set_features() failed (%d); wanted %pNF, left %pNF\n", 9923 err, &features, &dev->features); 9924 /* return non-0 since some features might have changed and 9925 * it's better to fire a spurious notification than miss it 9926 */ 9927 return -1; 9928 } 9929 9930 sync_lower: 9931 /* some features must be disabled on lower devices when disabled 9932 * on an upper device (think: bonding master or bridge) 9933 */ 9934 netdev_for_each_lower_dev(dev, lower, iter) 9935 netdev_sync_lower_features(dev, lower, features); 9936 9937 if (!err) { 9938 netdev_features_t diff = features ^ dev->features; 9939 9940 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) { 9941 /* udp_tunnel_{get,drop}_rx_info both need 9942 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the 9943 * device, or they won't do anything. 9944 * Thus we need to update dev->features 9945 * *before* calling udp_tunnel_get_rx_info, 9946 * but *after* calling udp_tunnel_drop_rx_info. 9947 */ 9948 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) { 9949 dev->features = features; 9950 udp_tunnel_get_rx_info(dev); 9951 } else { 9952 udp_tunnel_drop_rx_info(dev); 9953 } 9954 } 9955 9956 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) { 9957 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) { 9958 dev->features = features; 9959 err |= vlan_get_rx_ctag_filter_info(dev); 9960 } else { 9961 vlan_drop_rx_ctag_filter_info(dev); 9962 } 9963 } 9964 9965 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) { 9966 if (features & NETIF_F_HW_VLAN_STAG_FILTER) { 9967 dev->features = features; 9968 err |= vlan_get_rx_stag_filter_info(dev); 9969 } else { 9970 vlan_drop_rx_stag_filter_info(dev); 9971 } 9972 } 9973 9974 dev->features = features; 9975 } 9976 9977 return err < 0 ? 0 : 1; 9978 } 9979 9980 /** 9981 * netdev_update_features - recalculate device features 9982 * @dev: the device to check 9983 * 9984 * Recalculate dev->features set and send notifications if it 9985 * has changed. Should be called after driver or hardware dependent 9986 * conditions might have changed that influence the features. 9987 */ 9988 void netdev_update_features(struct net_device *dev) 9989 { 9990 if (__netdev_update_features(dev)) 9991 netdev_features_change(dev); 9992 } 9993 EXPORT_SYMBOL(netdev_update_features); 9994 9995 /** 9996 * netdev_change_features - recalculate device features 9997 * @dev: the device to check 9998 * 9999 * Recalculate dev->features set and send notifications even 10000 * if they have not changed. Should be called instead of 10001 * netdev_update_features() if also dev->vlan_features might 10002 * have changed to allow the changes to be propagated to stacked 10003 * VLAN devices. 10004 */ 10005 void netdev_change_features(struct net_device *dev) 10006 { 10007 __netdev_update_features(dev); 10008 netdev_features_change(dev); 10009 } 10010 EXPORT_SYMBOL(netdev_change_features); 10011 10012 /** 10013 * netif_stacked_transfer_operstate - transfer operstate 10014 * @rootdev: the root or lower level device to transfer state from 10015 * @dev: the device to transfer operstate to 10016 * 10017 * Transfer operational state from root to device. This is normally 10018 * called when a stacking relationship exists between the root 10019 * device and the device(a leaf device). 10020 */ 10021 void netif_stacked_transfer_operstate(const struct net_device *rootdev, 10022 struct net_device *dev) 10023 { 10024 if (rootdev->operstate == IF_OPER_DORMANT) 10025 netif_dormant_on(dev); 10026 else 10027 netif_dormant_off(dev); 10028 10029 if (rootdev->operstate == IF_OPER_TESTING) 10030 netif_testing_on(dev); 10031 else 10032 netif_testing_off(dev); 10033 10034 if (netif_carrier_ok(rootdev)) 10035 netif_carrier_on(dev); 10036 else 10037 netif_carrier_off(dev); 10038 } 10039 EXPORT_SYMBOL(netif_stacked_transfer_operstate); 10040 10041 static int netif_alloc_rx_queues(struct net_device *dev) 10042 { 10043 unsigned int i, count = dev->num_rx_queues; 10044 struct netdev_rx_queue *rx; 10045 size_t sz = count * sizeof(*rx); 10046 int err = 0; 10047 10048 BUG_ON(count < 1); 10049 10050 rx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL); 10051 if (!rx) 10052 return -ENOMEM; 10053 10054 dev->_rx = rx; 10055 10056 for (i = 0; i < count; i++) { 10057 rx[i].dev = dev; 10058 10059 /* XDP RX-queue setup */ 10060 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0); 10061 if (err < 0) 10062 goto err_rxq_info; 10063 } 10064 return 0; 10065 10066 err_rxq_info: 10067 /* Rollback successful reg's and free other resources */ 10068 while (i--) 10069 xdp_rxq_info_unreg(&rx[i].xdp_rxq); 10070 kvfree(dev->_rx); 10071 dev->_rx = NULL; 10072 return err; 10073 } 10074 10075 static void netif_free_rx_queues(struct net_device *dev) 10076 { 10077 unsigned int i, count = dev->num_rx_queues; 10078 10079 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */ 10080 if (!dev->_rx) 10081 return; 10082 10083 for (i = 0; i < count; i++) 10084 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq); 10085 10086 kvfree(dev->_rx); 10087 } 10088 10089 static void netdev_init_one_queue(struct net_device *dev, 10090 struct netdev_queue *queue, void *_unused) 10091 { 10092 /* Initialize queue lock */ 10093 spin_lock_init(&queue->_xmit_lock); 10094 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); 10095 queue->xmit_lock_owner = -1; 10096 netdev_queue_numa_node_write(queue, NUMA_NO_NODE); 10097 queue->dev = dev; 10098 #ifdef CONFIG_BQL 10099 dql_init(&queue->dql, HZ); 10100 #endif 10101 } 10102 10103 static void netif_free_tx_queues(struct net_device *dev) 10104 { 10105 kvfree(dev->_tx); 10106 } 10107 10108 static int netif_alloc_netdev_queues(struct net_device *dev) 10109 { 10110 unsigned int count = dev->num_tx_queues; 10111 struct netdev_queue *tx; 10112 size_t sz = count * sizeof(*tx); 10113 10114 if (count < 1 || count > 0xffff) 10115 return -EINVAL; 10116 10117 tx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL); 10118 if (!tx) 10119 return -ENOMEM; 10120 10121 dev->_tx = tx; 10122 10123 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); 10124 spin_lock_init(&dev->tx_global_lock); 10125 10126 return 0; 10127 } 10128 10129 void netif_tx_stop_all_queues(struct net_device *dev) 10130 { 10131 unsigned int i; 10132 10133 for (i = 0; i < dev->num_tx_queues; i++) { 10134 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 10135 10136 netif_tx_stop_queue(txq); 10137 } 10138 } 10139 EXPORT_SYMBOL(netif_tx_stop_all_queues); 10140 10141 /** 10142 * register_netdevice - register a network device 10143 * @dev: device to register 10144 * 10145 * Take a completed network device structure and add it to the kernel 10146 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 10147 * chain. 0 is returned on success. A negative errno code is returned 10148 * on a failure to set up the device, or if the name is a duplicate. 10149 * 10150 * Callers must hold the rtnl semaphore. You may want 10151 * register_netdev() instead of this. 10152 * 10153 * BUGS: 10154 * The locking appears insufficient to guarantee two parallel registers 10155 * will not get the same name. 10156 */ 10157 10158 int register_netdevice(struct net_device *dev) 10159 { 10160 int ret; 10161 struct net *net = dev_net(dev); 10162 10163 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE < 10164 NETDEV_FEATURE_COUNT); 10165 BUG_ON(dev_boot_phase); 10166 ASSERT_RTNL(); 10167 10168 might_sleep(); 10169 10170 /* When net_device's are persistent, this will be fatal. */ 10171 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); 10172 BUG_ON(!net); 10173 10174 ret = ethtool_check_ops(dev->ethtool_ops); 10175 if (ret) 10176 return ret; 10177 10178 spin_lock_init(&dev->addr_list_lock); 10179 netdev_set_addr_lockdep_class(dev); 10180 10181 ret = dev_get_valid_name(net, dev, dev->name); 10182 if (ret < 0) 10183 goto out; 10184 10185 ret = -ENOMEM; 10186 dev->name_node = netdev_name_node_head_alloc(dev); 10187 if (!dev->name_node) 10188 goto out; 10189 10190 /* Init, if this function is available */ 10191 if (dev->netdev_ops->ndo_init) { 10192 ret = dev->netdev_ops->ndo_init(dev); 10193 if (ret) { 10194 if (ret > 0) 10195 ret = -EIO; 10196 goto err_free_name; 10197 } 10198 } 10199 10200 if (((dev->hw_features | dev->features) & 10201 NETIF_F_HW_VLAN_CTAG_FILTER) && 10202 (!dev->netdev_ops->ndo_vlan_rx_add_vid || 10203 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { 10204 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); 10205 ret = -EINVAL; 10206 goto err_uninit; 10207 } 10208 10209 ret = -EBUSY; 10210 if (!dev->ifindex) 10211 dev->ifindex = dev_new_index(net); 10212 else if (__dev_get_by_index(net, dev->ifindex)) 10213 goto err_uninit; 10214 10215 /* Transfer changeable features to wanted_features and enable 10216 * software offloads (GSO and GRO). 10217 */ 10218 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF); 10219 dev->features |= NETIF_F_SOFT_FEATURES; 10220 10221 if (dev->udp_tunnel_nic_info) { 10222 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10223 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10224 } 10225 10226 dev->wanted_features = dev->features & dev->hw_features; 10227 10228 if (!(dev->flags & IFF_LOOPBACK)) 10229 dev->hw_features |= NETIF_F_NOCACHE_COPY; 10230 10231 /* If IPv4 TCP segmentation offload is supported we should also 10232 * allow the device to enable segmenting the frame with the option 10233 * of ignoring a static IP ID value. This doesn't enable the 10234 * feature itself but allows the user to enable it later. 10235 */ 10236 if (dev->hw_features & NETIF_F_TSO) 10237 dev->hw_features |= NETIF_F_TSO_MANGLEID; 10238 if (dev->vlan_features & NETIF_F_TSO) 10239 dev->vlan_features |= NETIF_F_TSO_MANGLEID; 10240 if (dev->mpls_features & NETIF_F_TSO) 10241 dev->mpls_features |= NETIF_F_TSO_MANGLEID; 10242 if (dev->hw_enc_features & NETIF_F_TSO) 10243 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; 10244 10245 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. 10246 */ 10247 dev->vlan_features |= NETIF_F_HIGHDMA; 10248 10249 /* Make NETIF_F_SG inheritable to tunnel devices. 10250 */ 10251 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; 10252 10253 /* Make NETIF_F_SG inheritable to MPLS. 10254 */ 10255 dev->mpls_features |= NETIF_F_SG; 10256 10257 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); 10258 ret = notifier_to_errno(ret); 10259 if (ret) 10260 goto err_uninit; 10261 10262 ret = netdev_register_kobject(dev); 10263 if (ret) { 10264 dev->reg_state = NETREG_UNREGISTERED; 10265 goto err_uninit; 10266 } 10267 dev->reg_state = NETREG_REGISTERED; 10268 10269 __netdev_update_features(dev); 10270 10271 /* 10272 * Default initial state at registry is that the 10273 * device is present. 10274 */ 10275 10276 set_bit(__LINK_STATE_PRESENT, &dev->state); 10277 10278 linkwatch_init_dev(dev); 10279 10280 dev_init_scheduler(dev); 10281 dev_hold(dev); 10282 list_netdevice(dev); 10283 add_device_randomness(dev->dev_addr, dev->addr_len); 10284 10285 /* If the device has permanent device address, driver should 10286 * set dev_addr and also addr_assign_type should be set to 10287 * NET_ADDR_PERM (default value). 10288 */ 10289 if (dev->addr_assign_type == NET_ADDR_PERM) 10290 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); 10291 10292 /* Notify protocols, that a new device appeared. */ 10293 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); 10294 ret = notifier_to_errno(ret); 10295 if (ret) { 10296 /* Expect explicit free_netdev() on failure */ 10297 dev->needs_free_netdev = false; 10298 unregister_netdevice_queue(dev, NULL); 10299 goto out; 10300 } 10301 /* 10302 * Prevent userspace races by waiting until the network 10303 * device is fully setup before sending notifications. 10304 */ 10305 if (!dev->rtnl_link_ops || 10306 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 10307 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); 10308 10309 out: 10310 return ret; 10311 10312 err_uninit: 10313 if (dev->netdev_ops->ndo_uninit) 10314 dev->netdev_ops->ndo_uninit(dev); 10315 if (dev->priv_destructor) 10316 dev->priv_destructor(dev); 10317 err_free_name: 10318 netdev_name_node_free(dev->name_node); 10319 goto out; 10320 } 10321 EXPORT_SYMBOL(register_netdevice); 10322 10323 /** 10324 * init_dummy_netdev - init a dummy network device for NAPI 10325 * @dev: device to init 10326 * 10327 * This takes a network device structure and initialize the minimum 10328 * amount of fields so it can be used to schedule NAPI polls without 10329 * registering a full blown interface. This is to be used by drivers 10330 * that need to tie several hardware interfaces to a single NAPI 10331 * poll scheduler due to HW limitations. 10332 */ 10333 int init_dummy_netdev(struct net_device *dev) 10334 { 10335 /* Clear everything. Note we don't initialize spinlocks 10336 * are they aren't supposed to be taken by any of the 10337 * NAPI code and this dummy netdev is supposed to be 10338 * only ever used for NAPI polls 10339 */ 10340 memset(dev, 0, sizeof(struct net_device)); 10341 10342 /* make sure we BUG if trying to hit standard 10343 * register/unregister code path 10344 */ 10345 dev->reg_state = NETREG_DUMMY; 10346 10347 /* NAPI wants this */ 10348 INIT_LIST_HEAD(&dev->napi_list); 10349 10350 /* a dummy interface is started by default */ 10351 set_bit(__LINK_STATE_PRESENT, &dev->state); 10352 set_bit(__LINK_STATE_START, &dev->state); 10353 10354 /* napi_busy_loop stats accounting wants this */ 10355 dev_net_set(dev, &init_net); 10356 10357 /* Note : We dont allocate pcpu_refcnt for dummy devices, 10358 * because users of this 'device' dont need to change 10359 * its refcount. 10360 */ 10361 10362 return 0; 10363 } 10364 EXPORT_SYMBOL_GPL(init_dummy_netdev); 10365 10366 10367 /** 10368 * register_netdev - register a network device 10369 * @dev: device to register 10370 * 10371 * Take a completed network device structure and add it to the kernel 10372 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 10373 * chain. 0 is returned on success. A negative errno code is returned 10374 * on a failure to set up the device, or if the name is a duplicate. 10375 * 10376 * This is a wrapper around register_netdevice that takes the rtnl semaphore 10377 * and expands the device name if you passed a format string to 10378 * alloc_netdev. 10379 */ 10380 int register_netdev(struct net_device *dev) 10381 { 10382 int err; 10383 10384 if (rtnl_lock_killable()) 10385 return -EINTR; 10386 err = register_netdevice(dev); 10387 rtnl_unlock(); 10388 return err; 10389 } 10390 EXPORT_SYMBOL(register_netdev); 10391 10392 int netdev_refcnt_read(const struct net_device *dev) 10393 { 10394 #ifdef CONFIG_PCPU_DEV_REFCNT 10395 int i, refcnt = 0; 10396 10397 for_each_possible_cpu(i) 10398 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); 10399 return refcnt; 10400 #else 10401 return refcount_read(&dev->dev_refcnt); 10402 #endif 10403 } 10404 EXPORT_SYMBOL(netdev_refcnt_read); 10405 10406 int netdev_unregister_timeout_secs __read_mostly = 10; 10407 10408 #define WAIT_REFS_MIN_MSECS 1 10409 #define WAIT_REFS_MAX_MSECS 250 10410 /** 10411 * netdev_wait_allrefs - wait until all references are gone. 10412 * @dev: target net_device 10413 * 10414 * This is called when unregistering network devices. 10415 * 10416 * Any protocol or device that holds a reference should register 10417 * for netdevice notification, and cleanup and put back the 10418 * reference if they receive an UNREGISTER event. 10419 * We can get stuck here if buggy protocols don't correctly 10420 * call dev_put. 10421 */ 10422 static void netdev_wait_allrefs(struct net_device *dev) 10423 { 10424 unsigned long rebroadcast_time, warning_time; 10425 int wait = 0, refcnt; 10426 10427 linkwatch_forget_dev(dev); 10428 10429 rebroadcast_time = warning_time = jiffies; 10430 refcnt = netdev_refcnt_read(dev); 10431 10432 while (refcnt != 1) { 10433 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { 10434 rtnl_lock(); 10435 10436 /* Rebroadcast unregister notification */ 10437 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10438 10439 __rtnl_unlock(); 10440 rcu_barrier(); 10441 rtnl_lock(); 10442 10443 if (test_bit(__LINK_STATE_LINKWATCH_PENDING, 10444 &dev->state)) { 10445 /* We must not have linkwatch events 10446 * pending on unregister. If this 10447 * happens, we simply run the queue 10448 * unscheduled, resulting in a noop 10449 * for this device. 10450 */ 10451 linkwatch_run_queue(); 10452 } 10453 10454 __rtnl_unlock(); 10455 10456 rebroadcast_time = jiffies; 10457 } 10458 10459 if (!wait) { 10460 rcu_barrier(); 10461 wait = WAIT_REFS_MIN_MSECS; 10462 } else { 10463 msleep(wait); 10464 wait = min(wait << 1, WAIT_REFS_MAX_MSECS); 10465 } 10466 10467 refcnt = netdev_refcnt_read(dev); 10468 10469 if (refcnt != 1 && 10470 time_after(jiffies, warning_time + 10471 netdev_unregister_timeout_secs * HZ)) { 10472 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", 10473 dev->name, refcnt); 10474 warning_time = jiffies; 10475 } 10476 } 10477 } 10478 10479 /* The sequence is: 10480 * 10481 * rtnl_lock(); 10482 * ... 10483 * register_netdevice(x1); 10484 * register_netdevice(x2); 10485 * ... 10486 * unregister_netdevice(y1); 10487 * unregister_netdevice(y2); 10488 * ... 10489 * rtnl_unlock(); 10490 * free_netdev(y1); 10491 * free_netdev(y2); 10492 * 10493 * We are invoked by rtnl_unlock(). 10494 * This allows us to deal with problems: 10495 * 1) We can delete sysfs objects which invoke hotplug 10496 * without deadlocking with linkwatch via keventd. 10497 * 2) Since we run with the RTNL semaphore not held, we can sleep 10498 * safely in order to wait for the netdev refcnt to drop to zero. 10499 * 10500 * We must not return until all unregister events added during 10501 * the interval the lock was held have been completed. 10502 */ 10503 void netdev_run_todo(void) 10504 { 10505 struct list_head list; 10506 #ifdef CONFIG_LOCKDEP 10507 struct list_head unlink_list; 10508 10509 list_replace_init(&net_unlink_list, &unlink_list); 10510 10511 while (!list_empty(&unlink_list)) { 10512 struct net_device *dev = list_first_entry(&unlink_list, 10513 struct net_device, 10514 unlink_list); 10515 list_del_init(&dev->unlink_list); 10516 dev->nested_level = dev->lower_level - 1; 10517 } 10518 #endif 10519 10520 /* Snapshot list, allow later requests */ 10521 list_replace_init(&net_todo_list, &list); 10522 10523 __rtnl_unlock(); 10524 10525 10526 /* Wait for rcu callbacks to finish before next phase */ 10527 if (!list_empty(&list)) 10528 rcu_barrier(); 10529 10530 while (!list_empty(&list)) { 10531 struct net_device *dev 10532 = list_first_entry(&list, struct net_device, todo_list); 10533 list_del(&dev->todo_list); 10534 10535 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { 10536 pr_err("network todo '%s' but state %d\n", 10537 dev->name, dev->reg_state); 10538 dump_stack(); 10539 continue; 10540 } 10541 10542 dev->reg_state = NETREG_UNREGISTERED; 10543 10544 netdev_wait_allrefs(dev); 10545 10546 /* paranoia */ 10547 BUG_ON(netdev_refcnt_read(dev) != 1); 10548 BUG_ON(!list_empty(&dev->ptype_all)); 10549 BUG_ON(!list_empty(&dev->ptype_specific)); 10550 WARN_ON(rcu_access_pointer(dev->ip_ptr)); 10551 WARN_ON(rcu_access_pointer(dev->ip6_ptr)); 10552 #if IS_ENABLED(CONFIG_DECNET) 10553 WARN_ON(dev->dn_ptr); 10554 #endif 10555 if (dev->priv_destructor) 10556 dev->priv_destructor(dev); 10557 if (dev->needs_free_netdev) 10558 free_netdev(dev); 10559 10560 /* Report a network device has been unregistered */ 10561 rtnl_lock(); 10562 dev_net(dev)->dev_unreg_count--; 10563 __rtnl_unlock(); 10564 wake_up(&netdev_unregistering_wq); 10565 10566 /* Free network device */ 10567 kobject_put(&dev->dev.kobj); 10568 } 10569 } 10570 10571 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has 10572 * all the same fields in the same order as net_device_stats, with only 10573 * the type differing, but rtnl_link_stats64 may have additional fields 10574 * at the end for newer counters. 10575 */ 10576 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, 10577 const struct net_device_stats *netdev_stats) 10578 { 10579 #if BITS_PER_LONG == 64 10580 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats)); 10581 memcpy(stats64, netdev_stats, sizeof(*netdev_stats)); 10582 /* zero out counters that only exist in rtnl_link_stats64 */ 10583 memset((char *)stats64 + sizeof(*netdev_stats), 0, 10584 sizeof(*stats64) - sizeof(*netdev_stats)); 10585 #else 10586 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long); 10587 const unsigned long *src = (const unsigned long *)netdev_stats; 10588 u64 *dst = (u64 *)stats64; 10589 10590 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); 10591 for (i = 0; i < n; i++) 10592 dst[i] = src[i]; 10593 /* zero out counters that only exist in rtnl_link_stats64 */ 10594 memset((char *)stats64 + n * sizeof(u64), 0, 10595 sizeof(*stats64) - n * sizeof(u64)); 10596 #endif 10597 } 10598 EXPORT_SYMBOL(netdev_stats_to_stats64); 10599 10600 /** 10601 * dev_get_stats - get network device statistics 10602 * @dev: device to get statistics from 10603 * @storage: place to store stats 10604 * 10605 * Get network statistics from device. Return @storage. 10606 * The device driver may provide its own method by setting 10607 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; 10608 * otherwise the internal statistics structure is used. 10609 */ 10610 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, 10611 struct rtnl_link_stats64 *storage) 10612 { 10613 const struct net_device_ops *ops = dev->netdev_ops; 10614 10615 if (ops->ndo_get_stats64) { 10616 memset(storage, 0, sizeof(*storage)); 10617 ops->ndo_get_stats64(dev, storage); 10618 } else if (ops->ndo_get_stats) { 10619 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); 10620 } else { 10621 netdev_stats_to_stats64(storage, &dev->stats); 10622 } 10623 storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped); 10624 storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped); 10625 storage->rx_nohandler += (unsigned long)atomic_long_read(&dev->rx_nohandler); 10626 return storage; 10627 } 10628 EXPORT_SYMBOL(dev_get_stats); 10629 10630 /** 10631 * dev_fetch_sw_netstats - get per-cpu network device statistics 10632 * @s: place to store stats 10633 * @netstats: per-cpu network stats to read from 10634 * 10635 * Read per-cpu network statistics and populate the related fields in @s. 10636 */ 10637 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s, 10638 const struct pcpu_sw_netstats __percpu *netstats) 10639 { 10640 int cpu; 10641 10642 for_each_possible_cpu(cpu) { 10643 const struct pcpu_sw_netstats *stats; 10644 struct pcpu_sw_netstats tmp; 10645 unsigned int start; 10646 10647 stats = per_cpu_ptr(netstats, cpu); 10648 do { 10649 start = u64_stats_fetch_begin_irq(&stats->syncp); 10650 tmp.rx_packets = stats->rx_packets; 10651 tmp.rx_bytes = stats->rx_bytes; 10652 tmp.tx_packets = stats->tx_packets; 10653 tmp.tx_bytes = stats->tx_bytes; 10654 } while (u64_stats_fetch_retry_irq(&stats->syncp, start)); 10655 10656 s->rx_packets += tmp.rx_packets; 10657 s->rx_bytes += tmp.rx_bytes; 10658 s->tx_packets += tmp.tx_packets; 10659 s->tx_bytes += tmp.tx_bytes; 10660 } 10661 } 10662 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats); 10663 10664 /** 10665 * dev_get_tstats64 - ndo_get_stats64 implementation 10666 * @dev: device to get statistics from 10667 * @s: place to store stats 10668 * 10669 * Populate @s from dev->stats and dev->tstats. Can be used as 10670 * ndo_get_stats64() callback. 10671 */ 10672 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s) 10673 { 10674 netdev_stats_to_stats64(s, &dev->stats); 10675 dev_fetch_sw_netstats(s, dev->tstats); 10676 } 10677 EXPORT_SYMBOL_GPL(dev_get_tstats64); 10678 10679 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) 10680 { 10681 struct netdev_queue *queue = dev_ingress_queue(dev); 10682 10683 #ifdef CONFIG_NET_CLS_ACT 10684 if (queue) 10685 return queue; 10686 queue = kzalloc(sizeof(*queue), GFP_KERNEL); 10687 if (!queue) 10688 return NULL; 10689 netdev_init_one_queue(dev, queue, NULL); 10690 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); 10691 queue->qdisc_sleeping = &noop_qdisc; 10692 rcu_assign_pointer(dev->ingress_queue, queue); 10693 #endif 10694 return queue; 10695 } 10696 10697 static const struct ethtool_ops default_ethtool_ops; 10698 10699 void netdev_set_default_ethtool_ops(struct net_device *dev, 10700 const struct ethtool_ops *ops) 10701 { 10702 if (dev->ethtool_ops == &default_ethtool_ops) 10703 dev->ethtool_ops = ops; 10704 } 10705 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); 10706 10707 void netdev_freemem(struct net_device *dev) 10708 { 10709 char *addr = (char *)dev - dev->padded; 10710 10711 kvfree(addr); 10712 } 10713 10714 /** 10715 * alloc_netdev_mqs - allocate network device 10716 * @sizeof_priv: size of private data to allocate space for 10717 * @name: device name format string 10718 * @name_assign_type: origin of device name 10719 * @setup: callback to initialize device 10720 * @txqs: the number of TX subqueues to allocate 10721 * @rxqs: the number of RX subqueues to allocate 10722 * 10723 * Allocates a struct net_device with private data area for driver use 10724 * and performs basic initialization. Also allocates subqueue structs 10725 * for each queue on the device. 10726 */ 10727 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, 10728 unsigned char name_assign_type, 10729 void (*setup)(struct net_device *), 10730 unsigned int txqs, unsigned int rxqs) 10731 { 10732 struct net_device *dev; 10733 unsigned int alloc_size; 10734 struct net_device *p; 10735 10736 BUG_ON(strlen(name) >= sizeof(dev->name)); 10737 10738 if (txqs < 1) { 10739 pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); 10740 return NULL; 10741 } 10742 10743 if (rxqs < 1) { 10744 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); 10745 return NULL; 10746 } 10747 10748 alloc_size = sizeof(struct net_device); 10749 if (sizeof_priv) { 10750 /* ensure 32-byte alignment of private area */ 10751 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); 10752 alloc_size += sizeof_priv; 10753 } 10754 /* ensure 32-byte alignment of whole construct */ 10755 alloc_size += NETDEV_ALIGN - 1; 10756 10757 p = kvzalloc(alloc_size, GFP_KERNEL | __GFP_RETRY_MAYFAIL); 10758 if (!p) 10759 return NULL; 10760 10761 dev = PTR_ALIGN(p, NETDEV_ALIGN); 10762 dev->padded = (char *)dev - (char *)p; 10763 10764 #ifdef CONFIG_PCPU_DEV_REFCNT 10765 dev->pcpu_refcnt = alloc_percpu(int); 10766 if (!dev->pcpu_refcnt) 10767 goto free_dev; 10768 dev_hold(dev); 10769 #else 10770 refcount_set(&dev->dev_refcnt, 1); 10771 #endif 10772 10773 if (dev_addr_init(dev)) 10774 goto free_pcpu; 10775 10776 dev_mc_init(dev); 10777 dev_uc_init(dev); 10778 10779 dev_net_set(dev, &init_net); 10780 10781 dev->gso_max_size = GSO_MAX_SIZE; 10782 dev->gso_max_segs = GSO_MAX_SEGS; 10783 dev->upper_level = 1; 10784 dev->lower_level = 1; 10785 #ifdef CONFIG_LOCKDEP 10786 dev->nested_level = 0; 10787 INIT_LIST_HEAD(&dev->unlink_list); 10788 #endif 10789 10790 INIT_LIST_HEAD(&dev->napi_list); 10791 INIT_LIST_HEAD(&dev->unreg_list); 10792 INIT_LIST_HEAD(&dev->close_list); 10793 INIT_LIST_HEAD(&dev->link_watch_list); 10794 INIT_LIST_HEAD(&dev->adj_list.upper); 10795 INIT_LIST_HEAD(&dev->adj_list.lower); 10796 INIT_LIST_HEAD(&dev->ptype_all); 10797 INIT_LIST_HEAD(&dev->ptype_specific); 10798 INIT_LIST_HEAD(&dev->net_notifier_list); 10799 #ifdef CONFIG_NET_SCHED 10800 hash_init(dev->qdisc_hash); 10801 #endif 10802 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; 10803 setup(dev); 10804 10805 if (!dev->tx_queue_len) { 10806 dev->priv_flags |= IFF_NO_QUEUE; 10807 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; 10808 } 10809 10810 dev->num_tx_queues = txqs; 10811 dev->real_num_tx_queues = txqs; 10812 if (netif_alloc_netdev_queues(dev)) 10813 goto free_all; 10814 10815 dev->num_rx_queues = rxqs; 10816 dev->real_num_rx_queues = rxqs; 10817 if (netif_alloc_rx_queues(dev)) 10818 goto free_all; 10819 10820 strcpy(dev->name, name); 10821 dev->name_assign_type = name_assign_type; 10822 dev->group = INIT_NETDEV_GROUP; 10823 if (!dev->ethtool_ops) 10824 dev->ethtool_ops = &default_ethtool_ops; 10825 10826 nf_hook_ingress_init(dev); 10827 10828 return dev; 10829 10830 free_all: 10831 free_netdev(dev); 10832 return NULL; 10833 10834 free_pcpu: 10835 #ifdef CONFIG_PCPU_DEV_REFCNT 10836 free_percpu(dev->pcpu_refcnt); 10837 free_dev: 10838 #endif 10839 netdev_freemem(dev); 10840 return NULL; 10841 } 10842 EXPORT_SYMBOL(alloc_netdev_mqs); 10843 10844 /** 10845 * free_netdev - free network device 10846 * @dev: device 10847 * 10848 * This function does the last stage of destroying an allocated device 10849 * interface. The reference to the device object is released. If this 10850 * is the last reference then it will be freed.Must be called in process 10851 * context. 10852 */ 10853 void free_netdev(struct net_device *dev) 10854 { 10855 struct napi_struct *p, *n; 10856 10857 might_sleep(); 10858 10859 /* When called immediately after register_netdevice() failed the unwind 10860 * handling may still be dismantling the device. Handle that case by 10861 * deferring the free. 10862 */ 10863 if (dev->reg_state == NETREG_UNREGISTERING) { 10864 ASSERT_RTNL(); 10865 dev->needs_free_netdev = true; 10866 return; 10867 } 10868 10869 netif_free_tx_queues(dev); 10870 netif_free_rx_queues(dev); 10871 10872 kfree(rcu_dereference_protected(dev->ingress_queue, 1)); 10873 10874 /* Flush device addresses */ 10875 dev_addr_flush(dev); 10876 10877 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) 10878 netif_napi_del(p); 10879 10880 #ifdef CONFIG_PCPU_DEV_REFCNT 10881 free_percpu(dev->pcpu_refcnt); 10882 dev->pcpu_refcnt = NULL; 10883 #endif 10884 free_percpu(dev->xdp_bulkq); 10885 dev->xdp_bulkq = NULL; 10886 10887 /* Compatibility with error handling in drivers */ 10888 if (dev->reg_state == NETREG_UNINITIALIZED) { 10889 netdev_freemem(dev); 10890 return; 10891 } 10892 10893 BUG_ON(dev->reg_state != NETREG_UNREGISTERED); 10894 dev->reg_state = NETREG_RELEASED; 10895 10896 /* will free via device release */ 10897 put_device(&dev->dev); 10898 } 10899 EXPORT_SYMBOL(free_netdev); 10900 10901 /** 10902 * synchronize_net - Synchronize with packet receive processing 10903 * 10904 * Wait for packets currently being received to be done. 10905 * Does not block later packets from starting. 10906 */ 10907 void synchronize_net(void) 10908 { 10909 might_sleep(); 10910 if (rtnl_is_locked()) 10911 synchronize_rcu_expedited(); 10912 else 10913 synchronize_rcu(); 10914 } 10915 EXPORT_SYMBOL(synchronize_net); 10916 10917 /** 10918 * unregister_netdevice_queue - remove device from the kernel 10919 * @dev: device 10920 * @head: list 10921 * 10922 * This function shuts down a device interface and removes it 10923 * from the kernel tables. 10924 * If head not NULL, device is queued to be unregistered later. 10925 * 10926 * Callers must hold the rtnl semaphore. You may want 10927 * unregister_netdev() instead of this. 10928 */ 10929 10930 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) 10931 { 10932 ASSERT_RTNL(); 10933 10934 if (head) { 10935 list_move_tail(&dev->unreg_list, head); 10936 } else { 10937 LIST_HEAD(single); 10938 10939 list_add(&dev->unreg_list, &single); 10940 unregister_netdevice_many(&single); 10941 } 10942 } 10943 EXPORT_SYMBOL(unregister_netdevice_queue); 10944 10945 /** 10946 * unregister_netdevice_many - unregister many devices 10947 * @head: list of devices 10948 * 10949 * Note: As most callers use a stack allocated list_head, 10950 * we force a list_del() to make sure stack wont be corrupted later. 10951 */ 10952 void unregister_netdevice_many(struct list_head *head) 10953 { 10954 struct net_device *dev, *tmp; 10955 LIST_HEAD(close_head); 10956 10957 BUG_ON(dev_boot_phase); 10958 ASSERT_RTNL(); 10959 10960 if (list_empty(head)) 10961 return; 10962 10963 list_for_each_entry_safe(dev, tmp, head, unreg_list) { 10964 /* Some devices call without registering 10965 * for initialization unwind. Remove those 10966 * devices and proceed with the remaining. 10967 */ 10968 if (dev->reg_state == NETREG_UNINITIALIZED) { 10969 pr_debug("unregister_netdevice: device %s/%p never was registered\n", 10970 dev->name, dev); 10971 10972 WARN_ON(1); 10973 list_del(&dev->unreg_list); 10974 continue; 10975 } 10976 dev->dismantle = true; 10977 BUG_ON(dev->reg_state != NETREG_REGISTERED); 10978 } 10979 10980 /* If device is running, close it first. */ 10981 list_for_each_entry(dev, head, unreg_list) 10982 list_add_tail(&dev->close_list, &close_head); 10983 dev_close_many(&close_head, true); 10984 10985 list_for_each_entry(dev, head, unreg_list) { 10986 /* And unlink it from device chain. */ 10987 unlist_netdevice(dev); 10988 10989 dev->reg_state = NETREG_UNREGISTERING; 10990 } 10991 flush_all_backlogs(); 10992 10993 synchronize_net(); 10994 10995 list_for_each_entry(dev, head, unreg_list) { 10996 struct sk_buff *skb = NULL; 10997 10998 /* Shutdown queueing discipline. */ 10999 dev_shutdown(dev); 11000 11001 dev_xdp_uninstall(dev); 11002 11003 /* Notify protocols, that we are about to destroy 11004 * this device. They should clean all the things. 11005 */ 11006 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 11007 11008 if (!dev->rtnl_link_ops || 11009 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 11010 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0, 11011 GFP_KERNEL, NULL, 0); 11012 11013 /* 11014 * Flush the unicast and multicast chains 11015 */ 11016 dev_uc_flush(dev); 11017 dev_mc_flush(dev); 11018 11019 netdev_name_node_alt_flush(dev); 11020 netdev_name_node_free(dev->name_node); 11021 11022 if (dev->netdev_ops->ndo_uninit) 11023 dev->netdev_ops->ndo_uninit(dev); 11024 11025 if (skb) 11026 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL); 11027 11028 /* Notifier chain MUST detach us all upper devices. */ 11029 WARN_ON(netdev_has_any_upper_dev(dev)); 11030 WARN_ON(netdev_has_any_lower_dev(dev)); 11031 11032 /* Remove entries from kobject tree */ 11033 netdev_unregister_kobject(dev); 11034 #ifdef CONFIG_XPS 11035 /* Remove XPS queueing entries */ 11036 netif_reset_xps_queues_gt(dev, 0); 11037 #endif 11038 } 11039 11040 synchronize_net(); 11041 11042 list_for_each_entry(dev, head, unreg_list) { 11043 dev_put(dev); 11044 net_set_todo(dev); 11045 } 11046 11047 list_del(head); 11048 } 11049 EXPORT_SYMBOL(unregister_netdevice_many); 11050 11051 /** 11052 * unregister_netdev - remove device from the kernel 11053 * @dev: device 11054 * 11055 * This function shuts down a device interface and removes it 11056 * from the kernel tables. 11057 * 11058 * This is just a wrapper for unregister_netdevice that takes 11059 * the rtnl semaphore. In general you want to use this and not 11060 * unregister_netdevice. 11061 */ 11062 void unregister_netdev(struct net_device *dev) 11063 { 11064 rtnl_lock(); 11065 unregister_netdevice(dev); 11066 rtnl_unlock(); 11067 } 11068 EXPORT_SYMBOL(unregister_netdev); 11069 11070 /** 11071 * __dev_change_net_namespace - move device to different nethost namespace 11072 * @dev: device 11073 * @net: network namespace 11074 * @pat: If not NULL name pattern to try if the current device name 11075 * is already taken in the destination network namespace. 11076 * @new_ifindex: If not zero, specifies device index in the target 11077 * namespace. 11078 * 11079 * This function shuts down a device interface and moves it 11080 * to a new network namespace. On success 0 is returned, on 11081 * a failure a netagive errno code is returned. 11082 * 11083 * Callers must hold the rtnl semaphore. 11084 */ 11085 11086 int __dev_change_net_namespace(struct net_device *dev, struct net *net, 11087 const char *pat, int new_ifindex) 11088 { 11089 struct net *net_old = dev_net(dev); 11090 int err, new_nsid; 11091 11092 ASSERT_RTNL(); 11093 11094 /* Don't allow namespace local devices to be moved. */ 11095 err = -EINVAL; 11096 if (dev->features & NETIF_F_NETNS_LOCAL) 11097 goto out; 11098 11099 /* Ensure the device has been registrered */ 11100 if (dev->reg_state != NETREG_REGISTERED) 11101 goto out; 11102 11103 /* Get out if there is nothing todo */ 11104 err = 0; 11105 if (net_eq(net_old, net)) 11106 goto out; 11107 11108 /* Pick the destination device name, and ensure 11109 * we can use it in the destination network namespace. 11110 */ 11111 err = -EEXIST; 11112 if (__dev_get_by_name(net, dev->name)) { 11113 /* We get here if we can't use the current device name */ 11114 if (!pat) 11115 goto out; 11116 err = dev_get_valid_name(net, dev, pat); 11117 if (err < 0) 11118 goto out; 11119 } 11120 11121 /* Check that new_ifindex isn't used yet. */ 11122 err = -EBUSY; 11123 if (new_ifindex && __dev_get_by_index(net, new_ifindex)) 11124 goto out; 11125 11126 /* 11127 * And now a mini version of register_netdevice unregister_netdevice. 11128 */ 11129 11130 /* If device is running close it first. */ 11131 dev_close(dev); 11132 11133 /* And unlink it from device chain */ 11134 unlist_netdevice(dev); 11135 11136 synchronize_net(); 11137 11138 /* Shutdown queueing discipline. */ 11139 dev_shutdown(dev); 11140 11141 /* Notify protocols, that we are about to destroy 11142 * this device. They should clean all the things. 11143 * 11144 * Note that dev->reg_state stays at NETREG_REGISTERED. 11145 * This is wanted because this way 8021q and macvlan know 11146 * the device is just moving and can keep their slaves up. 11147 */ 11148 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 11149 rcu_barrier(); 11150 11151 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL); 11152 /* If there is an ifindex conflict assign a new one */ 11153 if (!new_ifindex) { 11154 if (__dev_get_by_index(net, dev->ifindex)) 11155 new_ifindex = dev_new_index(net); 11156 else 11157 new_ifindex = dev->ifindex; 11158 } 11159 11160 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid, 11161 new_ifindex); 11162 11163 /* 11164 * Flush the unicast and multicast chains 11165 */ 11166 dev_uc_flush(dev); 11167 dev_mc_flush(dev); 11168 11169 /* Send a netdev-removed uevent to the old namespace */ 11170 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); 11171 netdev_adjacent_del_links(dev); 11172 11173 /* Move per-net netdevice notifiers that are following the netdevice */ 11174 move_netdevice_notifiers_dev_net(dev, net); 11175 11176 /* Actually switch the network namespace */ 11177 dev_net_set(dev, net); 11178 dev->ifindex = new_ifindex; 11179 11180 /* Send a netdev-add uevent to the new namespace */ 11181 kobject_uevent(&dev->dev.kobj, KOBJ_ADD); 11182 netdev_adjacent_add_links(dev); 11183 11184 /* Fixup kobjects */ 11185 err = device_rename(&dev->dev, dev->name); 11186 WARN_ON(err); 11187 11188 /* Adapt owner in case owning user namespace of target network 11189 * namespace is different from the original one. 11190 */ 11191 err = netdev_change_owner(dev, net_old, net); 11192 WARN_ON(err); 11193 11194 /* Add the device back in the hashes */ 11195 list_netdevice(dev); 11196 11197 /* Notify protocols, that a new device appeared. */ 11198 call_netdevice_notifiers(NETDEV_REGISTER, dev); 11199 11200 /* 11201 * Prevent userspace races by waiting until the network 11202 * device is fully setup before sending notifications. 11203 */ 11204 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); 11205 11206 synchronize_net(); 11207 err = 0; 11208 out: 11209 return err; 11210 } 11211 EXPORT_SYMBOL_GPL(__dev_change_net_namespace); 11212 11213 static int dev_cpu_dead(unsigned int oldcpu) 11214 { 11215 struct sk_buff **list_skb; 11216 struct sk_buff *skb; 11217 unsigned int cpu; 11218 struct softnet_data *sd, *oldsd, *remsd = NULL; 11219 11220 local_irq_disable(); 11221 cpu = smp_processor_id(); 11222 sd = &per_cpu(softnet_data, cpu); 11223 oldsd = &per_cpu(softnet_data, oldcpu); 11224 11225 /* Find end of our completion_queue. */ 11226 list_skb = &sd->completion_queue; 11227 while (*list_skb) 11228 list_skb = &(*list_skb)->next; 11229 /* Append completion queue from offline CPU. */ 11230 *list_skb = oldsd->completion_queue; 11231 oldsd->completion_queue = NULL; 11232 11233 /* Append output queue from offline CPU. */ 11234 if (oldsd->output_queue) { 11235 *sd->output_queue_tailp = oldsd->output_queue; 11236 sd->output_queue_tailp = oldsd->output_queue_tailp; 11237 oldsd->output_queue = NULL; 11238 oldsd->output_queue_tailp = &oldsd->output_queue; 11239 } 11240 /* Append NAPI poll list from offline CPU, with one exception : 11241 * process_backlog() must be called by cpu owning percpu backlog. 11242 * We properly handle process_queue & input_pkt_queue later. 11243 */ 11244 while (!list_empty(&oldsd->poll_list)) { 11245 struct napi_struct *napi = list_first_entry(&oldsd->poll_list, 11246 struct napi_struct, 11247 poll_list); 11248 11249 list_del_init(&napi->poll_list); 11250 if (napi->poll == process_backlog) 11251 napi->state = 0; 11252 else 11253 ____napi_schedule(sd, napi); 11254 } 11255 11256 raise_softirq_irqoff(NET_TX_SOFTIRQ); 11257 local_irq_enable(); 11258 11259 #ifdef CONFIG_RPS 11260 remsd = oldsd->rps_ipi_list; 11261 oldsd->rps_ipi_list = NULL; 11262 #endif 11263 /* send out pending IPI's on offline CPU */ 11264 net_rps_send_ipi(remsd); 11265 11266 /* Process offline CPU's input_pkt_queue */ 11267 while ((skb = __skb_dequeue(&oldsd->process_queue))) { 11268 netif_rx_ni(skb); 11269 input_queue_head_incr(oldsd); 11270 } 11271 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { 11272 netif_rx_ni(skb); 11273 input_queue_head_incr(oldsd); 11274 } 11275 11276 return 0; 11277 } 11278 11279 /** 11280 * netdev_increment_features - increment feature set by one 11281 * @all: current feature set 11282 * @one: new feature set 11283 * @mask: mask feature set 11284 * 11285 * Computes a new feature set after adding a device with feature set 11286 * @one to the master device with current feature set @all. Will not 11287 * enable anything that is off in @mask. Returns the new feature set. 11288 */ 11289 netdev_features_t netdev_increment_features(netdev_features_t all, 11290 netdev_features_t one, netdev_features_t mask) 11291 { 11292 if (mask & NETIF_F_HW_CSUM) 11293 mask |= NETIF_F_CSUM_MASK; 11294 mask |= NETIF_F_VLAN_CHALLENGED; 11295 11296 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; 11297 all &= one | ~NETIF_F_ALL_FOR_ALL; 11298 11299 /* If one device supports hw checksumming, set for all. */ 11300 if (all & NETIF_F_HW_CSUM) 11301 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); 11302 11303 return all; 11304 } 11305 EXPORT_SYMBOL(netdev_increment_features); 11306 11307 static struct hlist_head * __net_init netdev_create_hash(void) 11308 { 11309 int i; 11310 struct hlist_head *hash; 11311 11312 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL); 11313 if (hash != NULL) 11314 for (i = 0; i < NETDEV_HASHENTRIES; i++) 11315 INIT_HLIST_HEAD(&hash[i]); 11316 11317 return hash; 11318 } 11319 11320 /* Initialize per network namespace state */ 11321 static int __net_init netdev_init(struct net *net) 11322 { 11323 BUILD_BUG_ON(GRO_HASH_BUCKETS > 11324 8 * sizeof_field(struct napi_struct, gro_bitmask)); 11325 11326 if (net != &init_net) 11327 INIT_LIST_HEAD(&net->dev_base_head); 11328 11329 net->dev_name_head = netdev_create_hash(); 11330 if (net->dev_name_head == NULL) 11331 goto err_name; 11332 11333 net->dev_index_head = netdev_create_hash(); 11334 if (net->dev_index_head == NULL) 11335 goto err_idx; 11336 11337 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain); 11338 11339 return 0; 11340 11341 err_idx: 11342 kfree(net->dev_name_head); 11343 err_name: 11344 return -ENOMEM; 11345 } 11346 11347 /** 11348 * netdev_drivername - network driver for the device 11349 * @dev: network device 11350 * 11351 * Determine network driver for device. 11352 */ 11353 const char *netdev_drivername(const struct net_device *dev) 11354 { 11355 const struct device_driver *driver; 11356 const struct device *parent; 11357 const char *empty = ""; 11358 11359 parent = dev->dev.parent; 11360 if (!parent) 11361 return empty; 11362 11363 driver = parent->driver; 11364 if (driver && driver->name) 11365 return driver->name; 11366 return empty; 11367 } 11368 11369 static void __netdev_printk(const char *level, const struct net_device *dev, 11370 struct va_format *vaf) 11371 { 11372 if (dev && dev->dev.parent) { 11373 dev_printk_emit(level[1] - '0', 11374 dev->dev.parent, 11375 "%s %s %s%s: %pV", 11376 dev_driver_string(dev->dev.parent), 11377 dev_name(dev->dev.parent), 11378 netdev_name(dev), netdev_reg_state(dev), 11379 vaf); 11380 } else if (dev) { 11381 printk("%s%s%s: %pV", 11382 level, netdev_name(dev), netdev_reg_state(dev), vaf); 11383 } else { 11384 printk("%s(NULL net_device): %pV", level, vaf); 11385 } 11386 } 11387 11388 void netdev_printk(const char *level, const struct net_device *dev, 11389 const char *format, ...) 11390 { 11391 struct va_format vaf; 11392 va_list args; 11393 11394 va_start(args, format); 11395 11396 vaf.fmt = format; 11397 vaf.va = &args; 11398 11399 __netdev_printk(level, dev, &vaf); 11400 11401 va_end(args); 11402 } 11403 EXPORT_SYMBOL(netdev_printk); 11404 11405 #define define_netdev_printk_level(func, level) \ 11406 void func(const struct net_device *dev, const char *fmt, ...) \ 11407 { \ 11408 struct va_format vaf; \ 11409 va_list args; \ 11410 \ 11411 va_start(args, fmt); \ 11412 \ 11413 vaf.fmt = fmt; \ 11414 vaf.va = &args; \ 11415 \ 11416 __netdev_printk(level, dev, &vaf); \ 11417 \ 11418 va_end(args); \ 11419 } \ 11420 EXPORT_SYMBOL(func); 11421 11422 define_netdev_printk_level(netdev_emerg, KERN_EMERG); 11423 define_netdev_printk_level(netdev_alert, KERN_ALERT); 11424 define_netdev_printk_level(netdev_crit, KERN_CRIT); 11425 define_netdev_printk_level(netdev_err, KERN_ERR); 11426 define_netdev_printk_level(netdev_warn, KERN_WARNING); 11427 define_netdev_printk_level(netdev_notice, KERN_NOTICE); 11428 define_netdev_printk_level(netdev_info, KERN_INFO); 11429 11430 static void __net_exit netdev_exit(struct net *net) 11431 { 11432 kfree(net->dev_name_head); 11433 kfree(net->dev_index_head); 11434 if (net != &init_net) 11435 WARN_ON_ONCE(!list_empty(&net->dev_base_head)); 11436 } 11437 11438 static struct pernet_operations __net_initdata netdev_net_ops = { 11439 .init = netdev_init, 11440 .exit = netdev_exit, 11441 }; 11442 11443 static void __net_exit default_device_exit(struct net *net) 11444 { 11445 struct net_device *dev, *aux; 11446 /* 11447 * Push all migratable network devices back to the 11448 * initial network namespace 11449 */ 11450 rtnl_lock(); 11451 for_each_netdev_safe(net, dev, aux) { 11452 int err; 11453 char fb_name[IFNAMSIZ]; 11454 11455 /* Ignore unmoveable devices (i.e. loopback) */ 11456 if (dev->features & NETIF_F_NETNS_LOCAL) 11457 continue; 11458 11459 /* Leave virtual devices for the generic cleanup */ 11460 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund) 11461 continue; 11462 11463 /* Push remaining network devices to init_net */ 11464 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); 11465 if (__dev_get_by_name(&init_net, fb_name)) 11466 snprintf(fb_name, IFNAMSIZ, "dev%%d"); 11467 err = dev_change_net_namespace(dev, &init_net, fb_name); 11468 if (err) { 11469 pr_emerg("%s: failed to move %s to init_net: %d\n", 11470 __func__, dev->name, err); 11471 BUG(); 11472 } 11473 } 11474 rtnl_unlock(); 11475 } 11476 11477 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list) 11478 { 11479 /* Return with the rtnl_lock held when there are no network 11480 * devices unregistering in any network namespace in net_list. 11481 */ 11482 struct net *net; 11483 bool unregistering; 11484 DEFINE_WAIT_FUNC(wait, woken_wake_function); 11485 11486 add_wait_queue(&netdev_unregistering_wq, &wait); 11487 for (;;) { 11488 unregistering = false; 11489 rtnl_lock(); 11490 list_for_each_entry(net, net_list, exit_list) { 11491 if (net->dev_unreg_count > 0) { 11492 unregistering = true; 11493 break; 11494 } 11495 } 11496 if (!unregistering) 11497 break; 11498 __rtnl_unlock(); 11499 11500 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); 11501 } 11502 remove_wait_queue(&netdev_unregistering_wq, &wait); 11503 } 11504 11505 static void __net_exit default_device_exit_batch(struct list_head *net_list) 11506 { 11507 /* At exit all network devices most be removed from a network 11508 * namespace. Do this in the reverse order of registration. 11509 * Do this across as many network namespaces as possible to 11510 * improve batching efficiency. 11511 */ 11512 struct net_device *dev; 11513 struct net *net; 11514 LIST_HEAD(dev_kill_list); 11515 11516 /* To prevent network device cleanup code from dereferencing 11517 * loopback devices or network devices that have been freed 11518 * wait here for all pending unregistrations to complete, 11519 * before unregistring the loopback device and allowing the 11520 * network namespace be freed. 11521 * 11522 * The netdev todo list containing all network devices 11523 * unregistrations that happen in default_device_exit_batch 11524 * will run in the rtnl_unlock() at the end of 11525 * default_device_exit_batch. 11526 */ 11527 rtnl_lock_unregistering(net_list); 11528 list_for_each_entry(net, net_list, exit_list) { 11529 for_each_netdev_reverse(net, dev) { 11530 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) 11531 dev->rtnl_link_ops->dellink(dev, &dev_kill_list); 11532 else 11533 unregister_netdevice_queue(dev, &dev_kill_list); 11534 } 11535 } 11536 unregister_netdevice_many(&dev_kill_list); 11537 rtnl_unlock(); 11538 } 11539 11540 static struct pernet_operations __net_initdata default_device_ops = { 11541 .exit = default_device_exit, 11542 .exit_batch = default_device_exit_batch, 11543 }; 11544 11545 /* 11546 * Initialize the DEV module. At boot time this walks the device list and 11547 * unhooks any devices that fail to initialise (normally hardware not 11548 * present) and leaves us with a valid list of present and active devices. 11549 * 11550 */ 11551 11552 /* 11553 * This is called single threaded during boot, so no need 11554 * to take the rtnl semaphore. 11555 */ 11556 static int __init net_dev_init(void) 11557 { 11558 int i, rc = -ENOMEM; 11559 11560 BUG_ON(!dev_boot_phase); 11561 11562 if (dev_proc_init()) 11563 goto out; 11564 11565 if (netdev_kobject_init()) 11566 goto out; 11567 11568 INIT_LIST_HEAD(&ptype_all); 11569 for (i = 0; i < PTYPE_HASH_SIZE; i++) 11570 INIT_LIST_HEAD(&ptype_base[i]); 11571 11572 INIT_LIST_HEAD(&offload_base); 11573 11574 if (register_pernet_subsys(&netdev_net_ops)) 11575 goto out; 11576 11577 /* 11578 * Initialise the packet receive queues. 11579 */ 11580 11581 for_each_possible_cpu(i) { 11582 struct work_struct *flush = per_cpu_ptr(&flush_works, i); 11583 struct softnet_data *sd = &per_cpu(softnet_data, i); 11584 11585 INIT_WORK(flush, flush_backlog); 11586 11587 skb_queue_head_init(&sd->input_pkt_queue); 11588 skb_queue_head_init(&sd->process_queue); 11589 #ifdef CONFIG_XFRM_OFFLOAD 11590 skb_queue_head_init(&sd->xfrm_backlog); 11591 #endif 11592 INIT_LIST_HEAD(&sd->poll_list); 11593 sd->output_queue_tailp = &sd->output_queue; 11594 #ifdef CONFIG_RPS 11595 INIT_CSD(&sd->csd, rps_trigger_softirq, sd); 11596 sd->cpu = i; 11597 #endif 11598 11599 init_gro_hash(&sd->backlog); 11600 sd->backlog.poll = process_backlog; 11601 sd->backlog.weight = weight_p; 11602 } 11603 11604 dev_boot_phase = 0; 11605 11606 /* The loopback device is special if any other network devices 11607 * is present in a network namespace the loopback device must 11608 * be present. Since we now dynamically allocate and free the 11609 * loopback device ensure this invariant is maintained by 11610 * keeping the loopback device as the first device on the 11611 * list of network devices. Ensuring the loopback devices 11612 * is the first device that appears and the last network device 11613 * that disappears. 11614 */ 11615 if (register_pernet_device(&loopback_net_ops)) 11616 goto out; 11617 11618 if (register_pernet_device(&default_device_ops)) 11619 goto out; 11620 11621 open_softirq(NET_TX_SOFTIRQ, net_tx_action); 11622 open_softirq(NET_RX_SOFTIRQ, net_rx_action); 11623 11624 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", 11625 NULL, dev_cpu_dead); 11626 WARN_ON(rc < 0); 11627 rc = 0; 11628 out: 11629 return rc; 11630 } 11631 11632 subsys_initcall(net_dev_init); 11633