1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * NET3 Protocol independent device support routines. 4 * 5 * Derived from the non IP parts of dev.c 1.0.19 6 * Authors: Ross Biro 7 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 8 * Mark Evans, <evansmp@uhura.aston.ac.uk> 9 * 10 * Additional Authors: 11 * Florian la Roche <rzsfl@rz.uni-sb.de> 12 * Alan Cox <gw4pts@gw4pts.ampr.org> 13 * David Hinds <dahinds@users.sourceforge.net> 14 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru> 15 * Adam Sulmicki <adam@cfar.umd.edu> 16 * Pekka Riikonen <priikone@poesidon.pspt.fi> 17 * 18 * Changes: 19 * D.J. Barrow : Fixed bug where dev->refcnt gets set 20 * to 2 if register_netdev gets called 21 * before net_dev_init & also removed a 22 * few lines of code in the process. 23 * Alan Cox : device private ioctl copies fields back. 24 * Alan Cox : Transmit queue code does relevant 25 * stunts to keep the queue safe. 26 * Alan Cox : Fixed double lock. 27 * Alan Cox : Fixed promisc NULL pointer trap 28 * ???????? : Support the full private ioctl range 29 * Alan Cox : Moved ioctl permission check into 30 * drivers 31 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI 32 * Alan Cox : 100 backlog just doesn't cut it when 33 * you start doing multicast video 8) 34 * Alan Cox : Rewrote net_bh and list manager. 35 * Alan Cox : Fix ETH_P_ALL echoback lengths. 36 * Alan Cox : Took out transmit every packet pass 37 * Saved a few bytes in the ioctl handler 38 * Alan Cox : Network driver sets packet type before 39 * calling netif_rx. Saves a function 40 * call a packet. 41 * Alan Cox : Hashed net_bh() 42 * Richard Kooijman: Timestamp fixes. 43 * Alan Cox : Wrong field in SIOCGIFDSTADDR 44 * Alan Cox : Device lock protection. 45 * Alan Cox : Fixed nasty side effect of device close 46 * changes. 47 * Rudi Cilibrasi : Pass the right thing to 48 * set_mac_address() 49 * Dave Miller : 32bit quantity for the device lock to 50 * make it work out on a Sparc. 51 * Bjorn Ekwall : Added KERNELD hack. 52 * Alan Cox : Cleaned up the backlog initialise. 53 * Craig Metz : SIOCGIFCONF fix if space for under 54 * 1 device. 55 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there 56 * is no device open function. 57 * Andi Kleen : Fix error reporting for SIOCGIFCONF 58 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF 59 * Cyrus Durgin : Cleaned for KMOD 60 * Adam Sulmicki : Bug Fix : Network Device Unload 61 * A network device unload needs to purge 62 * the backlog queue. 63 * Paul Rusty Russell : SIOCSIFNAME 64 * Pekka Riikonen : Netdev boot-time settings code 65 * Andrew Morton : Make unregister_netdevice wait 66 * indefinitely on dev->refcnt 67 * J Hadi Salim : - Backlog queue sampling 68 * - netif_rx() feedback 69 */ 70 71 #include <linux/uaccess.h> 72 #include <linux/bitops.h> 73 #include <linux/capability.h> 74 #include <linux/cpu.h> 75 #include <linux/types.h> 76 #include <linux/kernel.h> 77 #include <linux/hash.h> 78 #include <linux/slab.h> 79 #include <linux/sched.h> 80 #include <linux/sched/mm.h> 81 #include <linux/mutex.h> 82 #include <linux/rwsem.h> 83 #include <linux/string.h> 84 #include <linux/mm.h> 85 #include <linux/socket.h> 86 #include <linux/sockios.h> 87 #include <linux/errno.h> 88 #include <linux/interrupt.h> 89 #include <linux/if_ether.h> 90 #include <linux/netdevice.h> 91 #include <linux/etherdevice.h> 92 #include <linux/ethtool.h> 93 #include <linux/skbuff.h> 94 #include <linux/kthread.h> 95 #include <linux/bpf.h> 96 #include <linux/bpf_trace.h> 97 #include <net/net_namespace.h> 98 #include <net/sock.h> 99 #include <net/busy_poll.h> 100 #include <linux/rtnetlink.h> 101 #include <linux/stat.h> 102 #include <net/dsa.h> 103 #include <net/dst.h> 104 #include <net/dst_metadata.h> 105 #include <net/gro.h> 106 #include <net/pkt_sched.h> 107 #include <net/pkt_cls.h> 108 #include <net/checksum.h> 109 #include <net/xfrm.h> 110 #include <linux/highmem.h> 111 #include <linux/init.h> 112 #include <linux/module.h> 113 #include <linux/netpoll.h> 114 #include <linux/rcupdate.h> 115 #include <linux/delay.h> 116 #include <net/iw_handler.h> 117 #include <asm/current.h> 118 #include <linux/audit.h> 119 #include <linux/dmaengine.h> 120 #include <linux/err.h> 121 #include <linux/ctype.h> 122 #include <linux/if_arp.h> 123 #include <linux/if_vlan.h> 124 #include <linux/ip.h> 125 #include <net/ip.h> 126 #include <net/mpls.h> 127 #include <linux/ipv6.h> 128 #include <linux/in.h> 129 #include <linux/jhash.h> 130 #include <linux/random.h> 131 #include <trace/events/napi.h> 132 #include <trace/events/net.h> 133 #include <trace/events/skb.h> 134 #include <trace/events/qdisc.h> 135 #include <linux/inetdevice.h> 136 #include <linux/cpu_rmap.h> 137 #include <linux/static_key.h> 138 #include <linux/hashtable.h> 139 #include <linux/vmalloc.h> 140 #include <linux/if_macvlan.h> 141 #include <linux/errqueue.h> 142 #include <linux/hrtimer.h> 143 #include <linux/netfilter_ingress.h> 144 #include <linux/crash_dump.h> 145 #include <linux/sctp.h> 146 #include <net/udp_tunnel.h> 147 #include <linux/net_namespace.h> 148 #include <linux/indirect_call_wrapper.h> 149 #include <net/devlink.h> 150 #include <linux/pm_runtime.h> 151 #include <linux/prandom.h> 152 #include <linux/once_lite.h> 153 154 #include "net-sysfs.h" 155 156 #define MAX_GRO_SKBS 8 157 158 /* This should be increased if a protocol with a bigger head is added. */ 159 #define GRO_MAX_HEAD (MAX_HEADER + 128) 160 161 static DEFINE_SPINLOCK(ptype_lock); 162 static DEFINE_SPINLOCK(offload_lock); 163 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; 164 struct list_head ptype_all __read_mostly; /* Taps */ 165 static struct list_head offload_base __read_mostly; 166 167 static int netif_rx_internal(struct sk_buff *skb); 168 static int call_netdevice_notifiers_info(unsigned long val, 169 struct netdev_notifier_info *info); 170 static int call_netdevice_notifiers_extack(unsigned long val, 171 struct net_device *dev, 172 struct netlink_ext_ack *extack); 173 static struct napi_struct *napi_by_id(unsigned int napi_id); 174 175 /* 176 * The @dev_base_head list is protected by @dev_base_lock and the rtnl 177 * semaphore. 178 * 179 * Pure readers hold dev_base_lock for reading, or rcu_read_lock() 180 * 181 * Writers must hold the rtnl semaphore while they loop through the 182 * dev_base_head list, and hold dev_base_lock for writing when they do the 183 * actual updates. This allows pure readers to access the list even 184 * while a writer is preparing to update it. 185 * 186 * To put it another way, dev_base_lock is held for writing only to 187 * protect against pure readers; the rtnl semaphore provides the 188 * protection against other writers. 189 * 190 * See, for example usages, register_netdevice() and 191 * unregister_netdevice(), which must be called with the rtnl 192 * semaphore held. 193 */ 194 DEFINE_RWLOCK(dev_base_lock); 195 EXPORT_SYMBOL(dev_base_lock); 196 197 static DEFINE_MUTEX(ifalias_mutex); 198 199 /* protects napi_hash addition/deletion and napi_gen_id */ 200 static DEFINE_SPINLOCK(napi_hash_lock); 201 202 static unsigned int napi_gen_id = NR_CPUS; 203 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8); 204 205 static DECLARE_RWSEM(devnet_rename_sem); 206 207 static inline void dev_base_seq_inc(struct net *net) 208 { 209 while (++net->dev_base_seq == 0) 210 ; 211 } 212 213 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) 214 { 215 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ)); 216 217 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; 218 } 219 220 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) 221 { 222 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; 223 } 224 225 static inline void rps_lock(struct softnet_data *sd) 226 { 227 #ifdef CONFIG_RPS 228 spin_lock(&sd->input_pkt_queue.lock); 229 #endif 230 } 231 232 static inline void rps_unlock(struct softnet_data *sd) 233 { 234 #ifdef CONFIG_RPS 235 spin_unlock(&sd->input_pkt_queue.lock); 236 #endif 237 } 238 239 static struct netdev_name_node *netdev_name_node_alloc(struct net_device *dev, 240 const char *name) 241 { 242 struct netdev_name_node *name_node; 243 244 name_node = kmalloc(sizeof(*name_node), GFP_KERNEL); 245 if (!name_node) 246 return NULL; 247 INIT_HLIST_NODE(&name_node->hlist); 248 name_node->dev = dev; 249 name_node->name = name; 250 return name_node; 251 } 252 253 static struct netdev_name_node * 254 netdev_name_node_head_alloc(struct net_device *dev) 255 { 256 struct netdev_name_node *name_node; 257 258 name_node = netdev_name_node_alloc(dev, dev->name); 259 if (!name_node) 260 return NULL; 261 INIT_LIST_HEAD(&name_node->list); 262 return name_node; 263 } 264 265 static void netdev_name_node_free(struct netdev_name_node *name_node) 266 { 267 kfree(name_node); 268 } 269 270 static void netdev_name_node_add(struct net *net, 271 struct netdev_name_node *name_node) 272 { 273 hlist_add_head_rcu(&name_node->hlist, 274 dev_name_hash(net, name_node->name)); 275 } 276 277 static void netdev_name_node_del(struct netdev_name_node *name_node) 278 { 279 hlist_del_rcu(&name_node->hlist); 280 } 281 282 static struct netdev_name_node *netdev_name_node_lookup(struct net *net, 283 const char *name) 284 { 285 struct hlist_head *head = dev_name_hash(net, name); 286 struct netdev_name_node *name_node; 287 288 hlist_for_each_entry(name_node, head, hlist) 289 if (!strcmp(name_node->name, name)) 290 return name_node; 291 return NULL; 292 } 293 294 static struct netdev_name_node *netdev_name_node_lookup_rcu(struct net *net, 295 const char *name) 296 { 297 struct hlist_head *head = dev_name_hash(net, name); 298 struct netdev_name_node *name_node; 299 300 hlist_for_each_entry_rcu(name_node, head, hlist) 301 if (!strcmp(name_node->name, name)) 302 return name_node; 303 return NULL; 304 } 305 306 bool netdev_name_in_use(struct net *net, const char *name) 307 { 308 return netdev_name_node_lookup(net, name); 309 } 310 EXPORT_SYMBOL(netdev_name_in_use); 311 312 int netdev_name_node_alt_create(struct net_device *dev, const char *name) 313 { 314 struct netdev_name_node *name_node; 315 struct net *net = dev_net(dev); 316 317 name_node = netdev_name_node_lookup(net, name); 318 if (name_node) 319 return -EEXIST; 320 name_node = netdev_name_node_alloc(dev, name); 321 if (!name_node) 322 return -ENOMEM; 323 netdev_name_node_add(net, name_node); 324 /* The node that holds dev->name acts as a head of per-device list. */ 325 list_add_tail(&name_node->list, &dev->name_node->list); 326 327 return 0; 328 } 329 EXPORT_SYMBOL(netdev_name_node_alt_create); 330 331 static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node) 332 { 333 list_del(&name_node->list); 334 netdev_name_node_del(name_node); 335 kfree(name_node->name); 336 netdev_name_node_free(name_node); 337 } 338 339 int netdev_name_node_alt_destroy(struct net_device *dev, const char *name) 340 { 341 struct netdev_name_node *name_node; 342 struct net *net = dev_net(dev); 343 344 name_node = netdev_name_node_lookup(net, name); 345 if (!name_node) 346 return -ENOENT; 347 /* lookup might have found our primary name or a name belonging 348 * to another device. 349 */ 350 if (name_node == dev->name_node || name_node->dev != dev) 351 return -EINVAL; 352 353 __netdev_name_node_alt_destroy(name_node); 354 355 return 0; 356 } 357 EXPORT_SYMBOL(netdev_name_node_alt_destroy); 358 359 static void netdev_name_node_alt_flush(struct net_device *dev) 360 { 361 struct netdev_name_node *name_node, *tmp; 362 363 list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list) 364 __netdev_name_node_alt_destroy(name_node); 365 } 366 367 /* Device list insertion */ 368 static void list_netdevice(struct net_device *dev) 369 { 370 struct net *net = dev_net(dev); 371 372 ASSERT_RTNL(); 373 374 write_lock_bh(&dev_base_lock); 375 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); 376 netdev_name_node_add(net, dev->name_node); 377 hlist_add_head_rcu(&dev->index_hlist, 378 dev_index_hash(net, dev->ifindex)); 379 write_unlock_bh(&dev_base_lock); 380 381 dev_base_seq_inc(net); 382 } 383 384 /* Device list removal 385 * caller must respect a RCU grace period before freeing/reusing dev 386 */ 387 static void unlist_netdevice(struct net_device *dev) 388 { 389 ASSERT_RTNL(); 390 391 /* Unlink dev from the device chain */ 392 write_lock_bh(&dev_base_lock); 393 list_del_rcu(&dev->dev_list); 394 netdev_name_node_del(dev->name_node); 395 hlist_del_rcu(&dev->index_hlist); 396 write_unlock_bh(&dev_base_lock); 397 398 dev_base_seq_inc(dev_net(dev)); 399 } 400 401 /* 402 * Our notifier list 403 */ 404 405 static RAW_NOTIFIER_HEAD(netdev_chain); 406 407 /* 408 * Device drivers call our routines to queue packets here. We empty the 409 * queue in the local softnet handler. 410 */ 411 412 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); 413 EXPORT_PER_CPU_SYMBOL(softnet_data); 414 415 #ifdef CONFIG_LOCKDEP 416 /* 417 * register_netdevice() inits txq->_xmit_lock and sets lockdep class 418 * according to dev->type 419 */ 420 static const unsigned short netdev_lock_type[] = { 421 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, 422 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, 423 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, 424 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, 425 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, 426 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, 427 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, 428 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, 429 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, 430 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, 431 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, 432 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, 433 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, 434 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, 435 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; 436 437 static const char *const netdev_lock_name[] = { 438 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", 439 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", 440 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", 441 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", 442 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", 443 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", 444 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", 445 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", 446 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", 447 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", 448 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", 449 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", 450 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", 451 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", 452 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; 453 454 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; 455 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; 456 457 static inline unsigned short netdev_lock_pos(unsigned short dev_type) 458 { 459 int i; 460 461 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) 462 if (netdev_lock_type[i] == dev_type) 463 return i; 464 /* the last key is used by default */ 465 return ARRAY_SIZE(netdev_lock_type) - 1; 466 } 467 468 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 469 unsigned short dev_type) 470 { 471 int i; 472 473 i = netdev_lock_pos(dev_type); 474 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], 475 netdev_lock_name[i]); 476 } 477 478 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 479 { 480 int i; 481 482 i = netdev_lock_pos(dev->type); 483 lockdep_set_class_and_name(&dev->addr_list_lock, 484 &netdev_addr_lock_key[i], 485 netdev_lock_name[i]); 486 } 487 #else 488 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 489 unsigned short dev_type) 490 { 491 } 492 493 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 494 { 495 } 496 #endif 497 498 /******************************************************************************* 499 * 500 * Protocol management and registration routines 501 * 502 *******************************************************************************/ 503 504 505 /* 506 * Add a protocol ID to the list. Now that the input handler is 507 * smarter we can dispense with all the messy stuff that used to be 508 * here. 509 * 510 * BEWARE!!! Protocol handlers, mangling input packets, 511 * MUST BE last in hash buckets and checking protocol handlers 512 * MUST start from promiscuous ptype_all chain in net_bh. 513 * It is true now, do not change it. 514 * Explanation follows: if protocol handler, mangling packet, will 515 * be the first on list, it is not able to sense, that packet 516 * is cloned and should be copied-on-write, so that it will 517 * change it and subsequent readers will get broken packet. 518 * --ANK (980803) 519 */ 520 521 static inline struct list_head *ptype_head(const struct packet_type *pt) 522 { 523 if (pt->type == htons(ETH_P_ALL)) 524 return pt->dev ? &pt->dev->ptype_all : &ptype_all; 525 else 526 return pt->dev ? &pt->dev->ptype_specific : 527 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; 528 } 529 530 /** 531 * dev_add_pack - add packet handler 532 * @pt: packet type declaration 533 * 534 * Add a protocol handler to the networking stack. The passed &packet_type 535 * is linked into kernel lists and may not be freed until it has been 536 * removed from the kernel lists. 537 * 538 * This call does not sleep therefore it can not 539 * guarantee all CPU's that are in middle of receiving packets 540 * will see the new packet type (until the next received packet). 541 */ 542 543 void dev_add_pack(struct packet_type *pt) 544 { 545 struct list_head *head = ptype_head(pt); 546 547 spin_lock(&ptype_lock); 548 list_add_rcu(&pt->list, head); 549 spin_unlock(&ptype_lock); 550 } 551 EXPORT_SYMBOL(dev_add_pack); 552 553 /** 554 * __dev_remove_pack - remove packet handler 555 * @pt: packet type declaration 556 * 557 * Remove a protocol handler that was previously added to the kernel 558 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 559 * from the kernel lists and can be freed or reused once this function 560 * returns. 561 * 562 * The packet type might still be in use by receivers 563 * and must not be freed until after all the CPU's have gone 564 * through a quiescent state. 565 */ 566 void __dev_remove_pack(struct packet_type *pt) 567 { 568 struct list_head *head = ptype_head(pt); 569 struct packet_type *pt1; 570 571 spin_lock(&ptype_lock); 572 573 list_for_each_entry(pt1, head, list) { 574 if (pt == pt1) { 575 list_del_rcu(&pt->list); 576 goto out; 577 } 578 } 579 580 pr_warn("dev_remove_pack: %p not found\n", pt); 581 out: 582 spin_unlock(&ptype_lock); 583 } 584 EXPORT_SYMBOL(__dev_remove_pack); 585 586 /** 587 * dev_remove_pack - remove packet handler 588 * @pt: packet type declaration 589 * 590 * Remove a protocol handler that was previously added to the kernel 591 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 592 * from the kernel lists and can be freed or reused once this function 593 * returns. 594 * 595 * This call sleeps to guarantee that no CPU is looking at the packet 596 * type after return. 597 */ 598 void dev_remove_pack(struct packet_type *pt) 599 { 600 __dev_remove_pack(pt); 601 602 synchronize_net(); 603 } 604 EXPORT_SYMBOL(dev_remove_pack); 605 606 607 /** 608 * dev_add_offload - register offload handlers 609 * @po: protocol offload declaration 610 * 611 * Add protocol offload handlers to the networking stack. The passed 612 * &proto_offload is linked into kernel lists and may not be freed until 613 * it has been removed from the kernel lists. 614 * 615 * This call does not sleep therefore it can not 616 * guarantee all CPU's that are in middle of receiving packets 617 * will see the new offload handlers (until the next received packet). 618 */ 619 void dev_add_offload(struct packet_offload *po) 620 { 621 struct packet_offload *elem; 622 623 spin_lock(&offload_lock); 624 list_for_each_entry(elem, &offload_base, list) { 625 if (po->priority < elem->priority) 626 break; 627 } 628 list_add_rcu(&po->list, elem->list.prev); 629 spin_unlock(&offload_lock); 630 } 631 EXPORT_SYMBOL(dev_add_offload); 632 633 /** 634 * __dev_remove_offload - remove offload handler 635 * @po: packet offload declaration 636 * 637 * Remove a protocol offload handler that was previously added to the 638 * kernel offload handlers by dev_add_offload(). The passed &offload_type 639 * is removed from the kernel lists and can be freed or reused once this 640 * function returns. 641 * 642 * The packet type might still be in use by receivers 643 * and must not be freed until after all the CPU's have gone 644 * through a quiescent state. 645 */ 646 static void __dev_remove_offload(struct packet_offload *po) 647 { 648 struct list_head *head = &offload_base; 649 struct packet_offload *po1; 650 651 spin_lock(&offload_lock); 652 653 list_for_each_entry(po1, head, list) { 654 if (po == po1) { 655 list_del_rcu(&po->list); 656 goto out; 657 } 658 } 659 660 pr_warn("dev_remove_offload: %p not found\n", po); 661 out: 662 spin_unlock(&offload_lock); 663 } 664 665 /** 666 * dev_remove_offload - remove packet offload handler 667 * @po: packet offload declaration 668 * 669 * Remove a packet offload handler that was previously added to the kernel 670 * offload handlers by dev_add_offload(). The passed &offload_type is 671 * removed from the kernel lists and can be freed or reused once this 672 * function returns. 673 * 674 * This call sleeps to guarantee that no CPU is looking at the packet 675 * type after return. 676 */ 677 void dev_remove_offload(struct packet_offload *po) 678 { 679 __dev_remove_offload(po); 680 681 synchronize_net(); 682 } 683 EXPORT_SYMBOL(dev_remove_offload); 684 685 /******************************************************************************* 686 * 687 * Device Interface Subroutines 688 * 689 *******************************************************************************/ 690 691 /** 692 * dev_get_iflink - get 'iflink' value of a interface 693 * @dev: targeted interface 694 * 695 * Indicates the ifindex the interface is linked to. 696 * Physical interfaces have the same 'ifindex' and 'iflink' values. 697 */ 698 699 int dev_get_iflink(const struct net_device *dev) 700 { 701 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink) 702 return dev->netdev_ops->ndo_get_iflink(dev); 703 704 return dev->ifindex; 705 } 706 EXPORT_SYMBOL(dev_get_iflink); 707 708 /** 709 * dev_fill_metadata_dst - Retrieve tunnel egress information. 710 * @dev: targeted interface 711 * @skb: The packet. 712 * 713 * For better visibility of tunnel traffic OVS needs to retrieve 714 * egress tunnel information for a packet. Following API allows 715 * user to get this info. 716 */ 717 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) 718 { 719 struct ip_tunnel_info *info; 720 721 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst) 722 return -EINVAL; 723 724 info = skb_tunnel_info_unclone(skb); 725 if (!info) 726 return -ENOMEM; 727 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX))) 728 return -EINVAL; 729 730 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb); 731 } 732 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst); 733 734 static struct net_device_path *dev_fwd_path(struct net_device_path_stack *stack) 735 { 736 int k = stack->num_paths++; 737 738 if (WARN_ON_ONCE(k >= NET_DEVICE_PATH_STACK_MAX)) 739 return NULL; 740 741 return &stack->path[k]; 742 } 743 744 int dev_fill_forward_path(const struct net_device *dev, const u8 *daddr, 745 struct net_device_path_stack *stack) 746 { 747 const struct net_device *last_dev; 748 struct net_device_path_ctx ctx = { 749 .dev = dev, 750 .daddr = daddr, 751 }; 752 struct net_device_path *path; 753 int ret = 0; 754 755 stack->num_paths = 0; 756 while (ctx.dev && ctx.dev->netdev_ops->ndo_fill_forward_path) { 757 last_dev = ctx.dev; 758 path = dev_fwd_path(stack); 759 if (!path) 760 return -1; 761 762 memset(path, 0, sizeof(struct net_device_path)); 763 ret = ctx.dev->netdev_ops->ndo_fill_forward_path(&ctx, path); 764 if (ret < 0) 765 return -1; 766 767 if (WARN_ON_ONCE(last_dev == ctx.dev)) 768 return -1; 769 } 770 path = dev_fwd_path(stack); 771 if (!path) 772 return -1; 773 path->type = DEV_PATH_ETHERNET; 774 path->dev = ctx.dev; 775 776 return ret; 777 } 778 EXPORT_SYMBOL_GPL(dev_fill_forward_path); 779 780 /** 781 * __dev_get_by_name - find a device by its name 782 * @net: the applicable net namespace 783 * @name: name to find 784 * 785 * Find an interface by name. Must be called under RTNL semaphore 786 * or @dev_base_lock. If the name is found a pointer to the device 787 * is returned. If the name is not found then %NULL is returned. The 788 * reference counters are not incremented so the caller must be 789 * careful with locks. 790 */ 791 792 struct net_device *__dev_get_by_name(struct net *net, const char *name) 793 { 794 struct netdev_name_node *node_name; 795 796 node_name = netdev_name_node_lookup(net, name); 797 return node_name ? node_name->dev : NULL; 798 } 799 EXPORT_SYMBOL(__dev_get_by_name); 800 801 /** 802 * dev_get_by_name_rcu - find a device by its name 803 * @net: the applicable net namespace 804 * @name: name to find 805 * 806 * Find an interface by name. 807 * If the name is found a pointer to the device is returned. 808 * If the name is not found then %NULL is returned. 809 * The reference counters are not incremented so the caller must be 810 * careful with locks. The caller must hold RCU lock. 811 */ 812 813 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) 814 { 815 struct netdev_name_node *node_name; 816 817 node_name = netdev_name_node_lookup_rcu(net, name); 818 return node_name ? node_name->dev : NULL; 819 } 820 EXPORT_SYMBOL(dev_get_by_name_rcu); 821 822 /** 823 * dev_get_by_name - find a device by its name 824 * @net: the applicable net namespace 825 * @name: name to find 826 * 827 * Find an interface by name. This can be called from any 828 * context and does its own locking. The returned handle has 829 * the usage count incremented and the caller must use dev_put() to 830 * release it when it is no longer needed. %NULL is returned if no 831 * matching device is found. 832 */ 833 834 struct net_device *dev_get_by_name(struct net *net, const char *name) 835 { 836 struct net_device *dev; 837 838 rcu_read_lock(); 839 dev = dev_get_by_name_rcu(net, name); 840 dev_hold(dev); 841 rcu_read_unlock(); 842 return dev; 843 } 844 EXPORT_SYMBOL(dev_get_by_name); 845 846 /** 847 * __dev_get_by_index - find a device by its ifindex 848 * @net: the applicable net namespace 849 * @ifindex: index of device 850 * 851 * Search for an interface by index. Returns %NULL if the device 852 * is not found or a pointer to the device. The device has not 853 * had its reference counter increased so the caller must be careful 854 * about locking. The caller must hold either the RTNL semaphore 855 * or @dev_base_lock. 856 */ 857 858 struct net_device *__dev_get_by_index(struct net *net, int ifindex) 859 { 860 struct net_device *dev; 861 struct hlist_head *head = dev_index_hash(net, ifindex); 862 863 hlist_for_each_entry(dev, head, index_hlist) 864 if (dev->ifindex == ifindex) 865 return dev; 866 867 return NULL; 868 } 869 EXPORT_SYMBOL(__dev_get_by_index); 870 871 /** 872 * dev_get_by_index_rcu - find a device by its ifindex 873 * @net: the applicable net namespace 874 * @ifindex: index of device 875 * 876 * Search for an interface by index. Returns %NULL if the device 877 * is not found or a pointer to the device. The device has not 878 * had its reference counter increased so the caller must be careful 879 * about locking. The caller must hold RCU lock. 880 */ 881 882 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) 883 { 884 struct net_device *dev; 885 struct hlist_head *head = dev_index_hash(net, ifindex); 886 887 hlist_for_each_entry_rcu(dev, head, index_hlist) 888 if (dev->ifindex == ifindex) 889 return dev; 890 891 return NULL; 892 } 893 EXPORT_SYMBOL(dev_get_by_index_rcu); 894 895 896 /** 897 * dev_get_by_index - find a device by its ifindex 898 * @net: the applicable net namespace 899 * @ifindex: index of device 900 * 901 * Search for an interface by index. Returns NULL if the device 902 * is not found or a pointer to the device. The device returned has 903 * had a reference added and the pointer is safe until the user calls 904 * dev_put to indicate they have finished with it. 905 */ 906 907 struct net_device *dev_get_by_index(struct net *net, int ifindex) 908 { 909 struct net_device *dev; 910 911 rcu_read_lock(); 912 dev = dev_get_by_index_rcu(net, ifindex); 913 dev_hold(dev); 914 rcu_read_unlock(); 915 return dev; 916 } 917 EXPORT_SYMBOL(dev_get_by_index); 918 919 /** 920 * dev_get_by_napi_id - find a device by napi_id 921 * @napi_id: ID of the NAPI struct 922 * 923 * Search for an interface by NAPI ID. Returns %NULL if the device 924 * is not found or a pointer to the device. The device has not had 925 * its reference counter increased so the caller must be careful 926 * about locking. The caller must hold RCU lock. 927 */ 928 929 struct net_device *dev_get_by_napi_id(unsigned int napi_id) 930 { 931 struct napi_struct *napi; 932 933 WARN_ON_ONCE(!rcu_read_lock_held()); 934 935 if (napi_id < MIN_NAPI_ID) 936 return NULL; 937 938 napi = napi_by_id(napi_id); 939 940 return napi ? napi->dev : NULL; 941 } 942 EXPORT_SYMBOL(dev_get_by_napi_id); 943 944 /** 945 * netdev_get_name - get a netdevice name, knowing its ifindex. 946 * @net: network namespace 947 * @name: a pointer to the buffer where the name will be stored. 948 * @ifindex: the ifindex of the interface to get the name from. 949 */ 950 int netdev_get_name(struct net *net, char *name, int ifindex) 951 { 952 struct net_device *dev; 953 int ret; 954 955 down_read(&devnet_rename_sem); 956 rcu_read_lock(); 957 958 dev = dev_get_by_index_rcu(net, ifindex); 959 if (!dev) { 960 ret = -ENODEV; 961 goto out; 962 } 963 964 strcpy(name, dev->name); 965 966 ret = 0; 967 out: 968 rcu_read_unlock(); 969 up_read(&devnet_rename_sem); 970 return ret; 971 } 972 973 /** 974 * dev_getbyhwaddr_rcu - find a device by its hardware address 975 * @net: the applicable net namespace 976 * @type: media type of device 977 * @ha: hardware address 978 * 979 * Search for an interface by MAC address. Returns NULL if the device 980 * is not found or a pointer to the device. 981 * The caller must hold RCU or RTNL. 982 * The returned device has not had its ref count increased 983 * and the caller must therefore be careful about locking 984 * 985 */ 986 987 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, 988 const char *ha) 989 { 990 struct net_device *dev; 991 992 for_each_netdev_rcu(net, dev) 993 if (dev->type == type && 994 !memcmp(dev->dev_addr, ha, dev->addr_len)) 995 return dev; 996 997 return NULL; 998 } 999 EXPORT_SYMBOL(dev_getbyhwaddr_rcu); 1000 1001 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) 1002 { 1003 struct net_device *dev, *ret = NULL; 1004 1005 rcu_read_lock(); 1006 for_each_netdev_rcu(net, dev) 1007 if (dev->type == type) { 1008 dev_hold(dev); 1009 ret = dev; 1010 break; 1011 } 1012 rcu_read_unlock(); 1013 return ret; 1014 } 1015 EXPORT_SYMBOL(dev_getfirstbyhwtype); 1016 1017 /** 1018 * __dev_get_by_flags - find any device with given flags 1019 * @net: the applicable net namespace 1020 * @if_flags: IFF_* values 1021 * @mask: bitmask of bits in if_flags to check 1022 * 1023 * Search for any interface with the given flags. Returns NULL if a device 1024 * is not found or a pointer to the device. Must be called inside 1025 * rtnl_lock(), and result refcount is unchanged. 1026 */ 1027 1028 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags, 1029 unsigned short mask) 1030 { 1031 struct net_device *dev, *ret; 1032 1033 ASSERT_RTNL(); 1034 1035 ret = NULL; 1036 for_each_netdev(net, dev) { 1037 if (((dev->flags ^ if_flags) & mask) == 0) { 1038 ret = dev; 1039 break; 1040 } 1041 } 1042 return ret; 1043 } 1044 EXPORT_SYMBOL(__dev_get_by_flags); 1045 1046 /** 1047 * dev_valid_name - check if name is okay for network device 1048 * @name: name string 1049 * 1050 * Network device names need to be valid file names to 1051 * allow sysfs to work. We also disallow any kind of 1052 * whitespace. 1053 */ 1054 bool dev_valid_name(const char *name) 1055 { 1056 if (*name == '\0') 1057 return false; 1058 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ) 1059 return false; 1060 if (!strcmp(name, ".") || !strcmp(name, "..")) 1061 return false; 1062 1063 while (*name) { 1064 if (*name == '/' || *name == ':' || isspace(*name)) 1065 return false; 1066 name++; 1067 } 1068 return true; 1069 } 1070 EXPORT_SYMBOL(dev_valid_name); 1071 1072 /** 1073 * __dev_alloc_name - allocate a name for a device 1074 * @net: network namespace to allocate the device name in 1075 * @name: name format string 1076 * @buf: scratch buffer and result name string 1077 * 1078 * Passed a format string - eg "lt%d" it will try and find a suitable 1079 * id. It scans list of devices to build up a free map, then chooses 1080 * the first empty slot. The caller must hold the dev_base or rtnl lock 1081 * while allocating the name and adding the device in order to avoid 1082 * duplicates. 1083 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1084 * Returns the number of the unit assigned or a negative errno code. 1085 */ 1086 1087 static int __dev_alloc_name(struct net *net, const char *name, char *buf) 1088 { 1089 int i = 0; 1090 const char *p; 1091 const int max_netdevices = 8*PAGE_SIZE; 1092 unsigned long *inuse; 1093 struct net_device *d; 1094 1095 if (!dev_valid_name(name)) 1096 return -EINVAL; 1097 1098 p = strchr(name, '%'); 1099 if (p) { 1100 /* 1101 * Verify the string as this thing may have come from 1102 * the user. There must be either one "%d" and no other "%" 1103 * characters. 1104 */ 1105 if (p[1] != 'd' || strchr(p + 2, '%')) 1106 return -EINVAL; 1107 1108 /* Use one page as a bit array of possible slots */ 1109 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); 1110 if (!inuse) 1111 return -ENOMEM; 1112 1113 for_each_netdev(net, d) { 1114 struct netdev_name_node *name_node; 1115 list_for_each_entry(name_node, &d->name_node->list, list) { 1116 if (!sscanf(name_node->name, name, &i)) 1117 continue; 1118 if (i < 0 || i >= max_netdevices) 1119 continue; 1120 1121 /* avoid cases where sscanf is not exact inverse of printf */ 1122 snprintf(buf, IFNAMSIZ, name, i); 1123 if (!strncmp(buf, name_node->name, IFNAMSIZ)) 1124 set_bit(i, inuse); 1125 } 1126 if (!sscanf(d->name, name, &i)) 1127 continue; 1128 if (i < 0 || i >= max_netdevices) 1129 continue; 1130 1131 /* avoid cases where sscanf is not exact inverse of printf */ 1132 snprintf(buf, IFNAMSIZ, name, i); 1133 if (!strncmp(buf, d->name, IFNAMSIZ)) 1134 set_bit(i, inuse); 1135 } 1136 1137 i = find_first_zero_bit(inuse, max_netdevices); 1138 free_page((unsigned long) inuse); 1139 } 1140 1141 snprintf(buf, IFNAMSIZ, name, i); 1142 if (!netdev_name_in_use(net, buf)) 1143 return i; 1144 1145 /* It is possible to run out of possible slots 1146 * when the name is long and there isn't enough space left 1147 * for the digits, or if all bits are used. 1148 */ 1149 return -ENFILE; 1150 } 1151 1152 static int dev_alloc_name_ns(struct net *net, 1153 struct net_device *dev, 1154 const char *name) 1155 { 1156 char buf[IFNAMSIZ]; 1157 int ret; 1158 1159 BUG_ON(!net); 1160 ret = __dev_alloc_name(net, name, buf); 1161 if (ret >= 0) 1162 strlcpy(dev->name, buf, IFNAMSIZ); 1163 return ret; 1164 } 1165 1166 /** 1167 * dev_alloc_name - allocate a name for a device 1168 * @dev: device 1169 * @name: name format string 1170 * 1171 * Passed a format string - eg "lt%d" it will try and find a suitable 1172 * id. It scans list of devices to build up a free map, then chooses 1173 * the first empty slot. The caller must hold the dev_base or rtnl lock 1174 * while allocating the name and adding the device in order to avoid 1175 * duplicates. 1176 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1177 * Returns the number of the unit assigned or a negative errno code. 1178 */ 1179 1180 int dev_alloc_name(struct net_device *dev, const char *name) 1181 { 1182 return dev_alloc_name_ns(dev_net(dev), dev, name); 1183 } 1184 EXPORT_SYMBOL(dev_alloc_name); 1185 1186 static int dev_get_valid_name(struct net *net, struct net_device *dev, 1187 const char *name) 1188 { 1189 BUG_ON(!net); 1190 1191 if (!dev_valid_name(name)) 1192 return -EINVAL; 1193 1194 if (strchr(name, '%')) 1195 return dev_alloc_name_ns(net, dev, name); 1196 else if (netdev_name_in_use(net, name)) 1197 return -EEXIST; 1198 else if (dev->name != name) 1199 strlcpy(dev->name, name, IFNAMSIZ); 1200 1201 return 0; 1202 } 1203 1204 /** 1205 * dev_change_name - change name of a device 1206 * @dev: device 1207 * @newname: name (or format string) must be at least IFNAMSIZ 1208 * 1209 * Change name of a device, can pass format strings "eth%d". 1210 * for wildcarding. 1211 */ 1212 int dev_change_name(struct net_device *dev, const char *newname) 1213 { 1214 unsigned char old_assign_type; 1215 char oldname[IFNAMSIZ]; 1216 int err = 0; 1217 int ret; 1218 struct net *net; 1219 1220 ASSERT_RTNL(); 1221 BUG_ON(!dev_net(dev)); 1222 1223 net = dev_net(dev); 1224 1225 /* Some auto-enslaved devices e.g. failover slaves are 1226 * special, as userspace might rename the device after 1227 * the interface had been brought up and running since 1228 * the point kernel initiated auto-enslavement. Allow 1229 * live name change even when these slave devices are 1230 * up and running. 1231 * 1232 * Typically, users of these auto-enslaving devices 1233 * don't actually care about slave name change, as 1234 * they are supposed to operate on master interface 1235 * directly. 1236 */ 1237 if (dev->flags & IFF_UP && 1238 likely(!(dev->priv_flags & IFF_LIVE_RENAME_OK))) 1239 return -EBUSY; 1240 1241 down_write(&devnet_rename_sem); 1242 1243 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { 1244 up_write(&devnet_rename_sem); 1245 return 0; 1246 } 1247 1248 memcpy(oldname, dev->name, IFNAMSIZ); 1249 1250 err = dev_get_valid_name(net, dev, newname); 1251 if (err < 0) { 1252 up_write(&devnet_rename_sem); 1253 return err; 1254 } 1255 1256 if (oldname[0] && !strchr(oldname, '%')) 1257 netdev_info(dev, "renamed from %s\n", oldname); 1258 1259 old_assign_type = dev->name_assign_type; 1260 dev->name_assign_type = NET_NAME_RENAMED; 1261 1262 rollback: 1263 ret = device_rename(&dev->dev, dev->name); 1264 if (ret) { 1265 memcpy(dev->name, oldname, IFNAMSIZ); 1266 dev->name_assign_type = old_assign_type; 1267 up_write(&devnet_rename_sem); 1268 return ret; 1269 } 1270 1271 up_write(&devnet_rename_sem); 1272 1273 netdev_adjacent_rename_links(dev, oldname); 1274 1275 write_lock_bh(&dev_base_lock); 1276 netdev_name_node_del(dev->name_node); 1277 write_unlock_bh(&dev_base_lock); 1278 1279 synchronize_rcu(); 1280 1281 write_lock_bh(&dev_base_lock); 1282 netdev_name_node_add(net, dev->name_node); 1283 write_unlock_bh(&dev_base_lock); 1284 1285 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); 1286 ret = notifier_to_errno(ret); 1287 1288 if (ret) { 1289 /* err >= 0 after dev_alloc_name() or stores the first errno */ 1290 if (err >= 0) { 1291 err = ret; 1292 down_write(&devnet_rename_sem); 1293 memcpy(dev->name, oldname, IFNAMSIZ); 1294 memcpy(oldname, newname, IFNAMSIZ); 1295 dev->name_assign_type = old_assign_type; 1296 old_assign_type = NET_NAME_RENAMED; 1297 goto rollback; 1298 } else { 1299 pr_err("%s: name change rollback failed: %d\n", 1300 dev->name, ret); 1301 } 1302 } 1303 1304 return err; 1305 } 1306 1307 /** 1308 * dev_set_alias - change ifalias of a device 1309 * @dev: device 1310 * @alias: name up to IFALIASZ 1311 * @len: limit of bytes to copy from info 1312 * 1313 * Set ifalias for a device, 1314 */ 1315 int dev_set_alias(struct net_device *dev, const char *alias, size_t len) 1316 { 1317 struct dev_ifalias *new_alias = NULL; 1318 1319 if (len >= IFALIASZ) 1320 return -EINVAL; 1321 1322 if (len) { 1323 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL); 1324 if (!new_alias) 1325 return -ENOMEM; 1326 1327 memcpy(new_alias->ifalias, alias, len); 1328 new_alias->ifalias[len] = 0; 1329 } 1330 1331 mutex_lock(&ifalias_mutex); 1332 new_alias = rcu_replace_pointer(dev->ifalias, new_alias, 1333 mutex_is_locked(&ifalias_mutex)); 1334 mutex_unlock(&ifalias_mutex); 1335 1336 if (new_alias) 1337 kfree_rcu(new_alias, rcuhead); 1338 1339 return len; 1340 } 1341 EXPORT_SYMBOL(dev_set_alias); 1342 1343 /** 1344 * dev_get_alias - get ifalias of a device 1345 * @dev: device 1346 * @name: buffer to store name of ifalias 1347 * @len: size of buffer 1348 * 1349 * get ifalias for a device. Caller must make sure dev cannot go 1350 * away, e.g. rcu read lock or own a reference count to device. 1351 */ 1352 int dev_get_alias(const struct net_device *dev, char *name, size_t len) 1353 { 1354 const struct dev_ifalias *alias; 1355 int ret = 0; 1356 1357 rcu_read_lock(); 1358 alias = rcu_dereference(dev->ifalias); 1359 if (alias) 1360 ret = snprintf(name, len, "%s", alias->ifalias); 1361 rcu_read_unlock(); 1362 1363 return ret; 1364 } 1365 1366 /** 1367 * netdev_features_change - device changes features 1368 * @dev: device to cause notification 1369 * 1370 * Called to indicate a device has changed features. 1371 */ 1372 void netdev_features_change(struct net_device *dev) 1373 { 1374 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); 1375 } 1376 EXPORT_SYMBOL(netdev_features_change); 1377 1378 /** 1379 * netdev_state_change - device changes state 1380 * @dev: device to cause notification 1381 * 1382 * Called to indicate a device has changed state. This function calls 1383 * the notifier chains for netdev_chain and sends a NEWLINK message 1384 * to the routing socket. 1385 */ 1386 void netdev_state_change(struct net_device *dev) 1387 { 1388 if (dev->flags & IFF_UP) { 1389 struct netdev_notifier_change_info change_info = { 1390 .info.dev = dev, 1391 }; 1392 1393 call_netdevice_notifiers_info(NETDEV_CHANGE, 1394 &change_info.info); 1395 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL); 1396 } 1397 } 1398 EXPORT_SYMBOL(netdev_state_change); 1399 1400 /** 1401 * __netdev_notify_peers - notify network peers about existence of @dev, 1402 * to be called when rtnl lock is already held. 1403 * @dev: network device 1404 * 1405 * Generate traffic such that interested network peers are aware of 1406 * @dev, such as by generating a gratuitous ARP. This may be used when 1407 * a device wants to inform the rest of the network about some sort of 1408 * reconfiguration such as a failover event or virtual machine 1409 * migration. 1410 */ 1411 void __netdev_notify_peers(struct net_device *dev) 1412 { 1413 ASSERT_RTNL(); 1414 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); 1415 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev); 1416 } 1417 EXPORT_SYMBOL(__netdev_notify_peers); 1418 1419 /** 1420 * netdev_notify_peers - notify network peers about existence of @dev 1421 * @dev: network device 1422 * 1423 * Generate traffic such that interested network peers are aware of 1424 * @dev, such as by generating a gratuitous ARP. This may be used when 1425 * a device wants to inform the rest of the network about some sort of 1426 * reconfiguration such as a failover event or virtual machine 1427 * migration. 1428 */ 1429 void netdev_notify_peers(struct net_device *dev) 1430 { 1431 rtnl_lock(); 1432 __netdev_notify_peers(dev); 1433 rtnl_unlock(); 1434 } 1435 EXPORT_SYMBOL(netdev_notify_peers); 1436 1437 static int napi_threaded_poll(void *data); 1438 1439 static int napi_kthread_create(struct napi_struct *n) 1440 { 1441 int err = 0; 1442 1443 /* Create and wake up the kthread once to put it in 1444 * TASK_INTERRUPTIBLE mode to avoid the blocked task 1445 * warning and work with loadavg. 1446 */ 1447 n->thread = kthread_run(napi_threaded_poll, n, "napi/%s-%d", 1448 n->dev->name, n->napi_id); 1449 if (IS_ERR(n->thread)) { 1450 err = PTR_ERR(n->thread); 1451 pr_err("kthread_run failed with err %d\n", err); 1452 n->thread = NULL; 1453 } 1454 1455 return err; 1456 } 1457 1458 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1459 { 1460 const struct net_device_ops *ops = dev->netdev_ops; 1461 int ret; 1462 1463 ASSERT_RTNL(); 1464 1465 if (!netif_device_present(dev)) { 1466 /* may be detached because parent is runtime-suspended */ 1467 if (dev->dev.parent) 1468 pm_runtime_resume(dev->dev.parent); 1469 if (!netif_device_present(dev)) 1470 return -ENODEV; 1471 } 1472 1473 /* Block netpoll from trying to do any rx path servicing. 1474 * If we don't do this there is a chance ndo_poll_controller 1475 * or ndo_poll may be running while we open the device 1476 */ 1477 netpoll_poll_disable(dev); 1478 1479 ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack); 1480 ret = notifier_to_errno(ret); 1481 if (ret) 1482 return ret; 1483 1484 set_bit(__LINK_STATE_START, &dev->state); 1485 1486 if (ops->ndo_validate_addr) 1487 ret = ops->ndo_validate_addr(dev); 1488 1489 if (!ret && ops->ndo_open) 1490 ret = ops->ndo_open(dev); 1491 1492 netpoll_poll_enable(dev); 1493 1494 if (ret) 1495 clear_bit(__LINK_STATE_START, &dev->state); 1496 else { 1497 dev->flags |= IFF_UP; 1498 dev_set_rx_mode(dev); 1499 dev_activate(dev); 1500 add_device_randomness(dev->dev_addr, dev->addr_len); 1501 } 1502 1503 return ret; 1504 } 1505 1506 /** 1507 * dev_open - prepare an interface for use. 1508 * @dev: device to open 1509 * @extack: netlink extended ack 1510 * 1511 * Takes a device from down to up state. The device's private open 1512 * function is invoked and then the multicast lists are loaded. Finally 1513 * the device is moved into the up state and a %NETDEV_UP message is 1514 * sent to the netdev notifier chain. 1515 * 1516 * Calling this function on an active interface is a nop. On a failure 1517 * a negative errno code is returned. 1518 */ 1519 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1520 { 1521 int ret; 1522 1523 if (dev->flags & IFF_UP) 1524 return 0; 1525 1526 ret = __dev_open(dev, extack); 1527 if (ret < 0) 1528 return ret; 1529 1530 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); 1531 call_netdevice_notifiers(NETDEV_UP, dev); 1532 1533 return ret; 1534 } 1535 EXPORT_SYMBOL(dev_open); 1536 1537 static void __dev_close_many(struct list_head *head) 1538 { 1539 struct net_device *dev; 1540 1541 ASSERT_RTNL(); 1542 might_sleep(); 1543 1544 list_for_each_entry(dev, head, close_list) { 1545 /* Temporarily disable netpoll until the interface is down */ 1546 netpoll_poll_disable(dev); 1547 1548 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); 1549 1550 clear_bit(__LINK_STATE_START, &dev->state); 1551 1552 /* Synchronize to scheduled poll. We cannot touch poll list, it 1553 * can be even on different cpu. So just clear netif_running(). 1554 * 1555 * dev->stop() will invoke napi_disable() on all of it's 1556 * napi_struct instances on this device. 1557 */ 1558 smp_mb__after_atomic(); /* Commit netif_running(). */ 1559 } 1560 1561 dev_deactivate_many(head); 1562 1563 list_for_each_entry(dev, head, close_list) { 1564 const struct net_device_ops *ops = dev->netdev_ops; 1565 1566 /* 1567 * Call the device specific close. This cannot fail. 1568 * Only if device is UP 1569 * 1570 * We allow it to be called even after a DETACH hot-plug 1571 * event. 1572 */ 1573 if (ops->ndo_stop) 1574 ops->ndo_stop(dev); 1575 1576 dev->flags &= ~IFF_UP; 1577 netpoll_poll_enable(dev); 1578 } 1579 } 1580 1581 static void __dev_close(struct net_device *dev) 1582 { 1583 LIST_HEAD(single); 1584 1585 list_add(&dev->close_list, &single); 1586 __dev_close_many(&single); 1587 list_del(&single); 1588 } 1589 1590 void dev_close_many(struct list_head *head, bool unlink) 1591 { 1592 struct net_device *dev, *tmp; 1593 1594 /* Remove the devices that don't need to be closed */ 1595 list_for_each_entry_safe(dev, tmp, head, close_list) 1596 if (!(dev->flags & IFF_UP)) 1597 list_del_init(&dev->close_list); 1598 1599 __dev_close_many(head); 1600 1601 list_for_each_entry_safe(dev, tmp, head, close_list) { 1602 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); 1603 call_netdevice_notifiers(NETDEV_DOWN, dev); 1604 if (unlink) 1605 list_del_init(&dev->close_list); 1606 } 1607 } 1608 EXPORT_SYMBOL(dev_close_many); 1609 1610 /** 1611 * dev_close - shutdown an interface. 1612 * @dev: device to shutdown 1613 * 1614 * This function moves an active device into down state. A 1615 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device 1616 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier 1617 * chain. 1618 */ 1619 void dev_close(struct net_device *dev) 1620 { 1621 if (dev->flags & IFF_UP) { 1622 LIST_HEAD(single); 1623 1624 list_add(&dev->close_list, &single); 1625 dev_close_many(&single, true); 1626 list_del(&single); 1627 } 1628 } 1629 EXPORT_SYMBOL(dev_close); 1630 1631 1632 /** 1633 * dev_disable_lro - disable Large Receive Offload on a device 1634 * @dev: device 1635 * 1636 * Disable Large Receive Offload (LRO) on a net device. Must be 1637 * called under RTNL. This is needed if received packets may be 1638 * forwarded to another interface. 1639 */ 1640 void dev_disable_lro(struct net_device *dev) 1641 { 1642 struct net_device *lower_dev; 1643 struct list_head *iter; 1644 1645 dev->wanted_features &= ~NETIF_F_LRO; 1646 netdev_update_features(dev); 1647 1648 if (unlikely(dev->features & NETIF_F_LRO)) 1649 netdev_WARN(dev, "failed to disable LRO!\n"); 1650 1651 netdev_for_each_lower_dev(dev, lower_dev, iter) 1652 dev_disable_lro(lower_dev); 1653 } 1654 EXPORT_SYMBOL(dev_disable_lro); 1655 1656 /** 1657 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device 1658 * @dev: device 1659 * 1660 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be 1661 * called under RTNL. This is needed if Generic XDP is installed on 1662 * the device. 1663 */ 1664 static void dev_disable_gro_hw(struct net_device *dev) 1665 { 1666 dev->wanted_features &= ~NETIF_F_GRO_HW; 1667 netdev_update_features(dev); 1668 1669 if (unlikely(dev->features & NETIF_F_GRO_HW)) 1670 netdev_WARN(dev, "failed to disable GRO_HW!\n"); 1671 } 1672 1673 const char *netdev_cmd_to_name(enum netdev_cmd cmd) 1674 { 1675 #define N(val) \ 1676 case NETDEV_##val: \ 1677 return "NETDEV_" __stringify(val); 1678 switch (cmd) { 1679 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER) 1680 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE) 1681 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE) 1682 N(POST_INIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) N(CHANGEUPPER) 1683 N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) N(BONDING_INFO) 1684 N(PRECHANGEUPPER) N(CHANGELOWERSTATE) N(UDP_TUNNEL_PUSH_INFO) 1685 N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN) 1686 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO) 1687 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO) 1688 N(PRE_CHANGEADDR) 1689 } 1690 #undef N 1691 return "UNKNOWN_NETDEV_EVENT"; 1692 } 1693 EXPORT_SYMBOL_GPL(netdev_cmd_to_name); 1694 1695 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, 1696 struct net_device *dev) 1697 { 1698 struct netdev_notifier_info info = { 1699 .dev = dev, 1700 }; 1701 1702 return nb->notifier_call(nb, val, &info); 1703 } 1704 1705 static int call_netdevice_register_notifiers(struct notifier_block *nb, 1706 struct net_device *dev) 1707 { 1708 int err; 1709 1710 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); 1711 err = notifier_to_errno(err); 1712 if (err) 1713 return err; 1714 1715 if (!(dev->flags & IFF_UP)) 1716 return 0; 1717 1718 call_netdevice_notifier(nb, NETDEV_UP, dev); 1719 return 0; 1720 } 1721 1722 static void call_netdevice_unregister_notifiers(struct notifier_block *nb, 1723 struct net_device *dev) 1724 { 1725 if (dev->flags & IFF_UP) { 1726 call_netdevice_notifier(nb, NETDEV_GOING_DOWN, 1727 dev); 1728 call_netdevice_notifier(nb, NETDEV_DOWN, dev); 1729 } 1730 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); 1731 } 1732 1733 static int call_netdevice_register_net_notifiers(struct notifier_block *nb, 1734 struct net *net) 1735 { 1736 struct net_device *dev; 1737 int err; 1738 1739 for_each_netdev(net, dev) { 1740 err = call_netdevice_register_notifiers(nb, dev); 1741 if (err) 1742 goto rollback; 1743 } 1744 return 0; 1745 1746 rollback: 1747 for_each_netdev_continue_reverse(net, dev) 1748 call_netdevice_unregister_notifiers(nb, dev); 1749 return err; 1750 } 1751 1752 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb, 1753 struct net *net) 1754 { 1755 struct net_device *dev; 1756 1757 for_each_netdev(net, dev) 1758 call_netdevice_unregister_notifiers(nb, dev); 1759 } 1760 1761 static int dev_boot_phase = 1; 1762 1763 /** 1764 * register_netdevice_notifier - register a network notifier block 1765 * @nb: notifier 1766 * 1767 * Register a notifier to be called when network device events occur. 1768 * The notifier passed is linked into the kernel structures and must 1769 * not be reused until it has been unregistered. A negative errno code 1770 * is returned on a failure. 1771 * 1772 * When registered all registration and up events are replayed 1773 * to the new notifier to allow device to have a race free 1774 * view of the network device list. 1775 */ 1776 1777 int register_netdevice_notifier(struct notifier_block *nb) 1778 { 1779 struct net *net; 1780 int err; 1781 1782 /* Close race with setup_net() and cleanup_net() */ 1783 down_write(&pernet_ops_rwsem); 1784 rtnl_lock(); 1785 err = raw_notifier_chain_register(&netdev_chain, nb); 1786 if (err) 1787 goto unlock; 1788 if (dev_boot_phase) 1789 goto unlock; 1790 for_each_net(net) { 1791 err = call_netdevice_register_net_notifiers(nb, net); 1792 if (err) 1793 goto rollback; 1794 } 1795 1796 unlock: 1797 rtnl_unlock(); 1798 up_write(&pernet_ops_rwsem); 1799 return err; 1800 1801 rollback: 1802 for_each_net_continue_reverse(net) 1803 call_netdevice_unregister_net_notifiers(nb, net); 1804 1805 raw_notifier_chain_unregister(&netdev_chain, nb); 1806 goto unlock; 1807 } 1808 EXPORT_SYMBOL(register_netdevice_notifier); 1809 1810 /** 1811 * unregister_netdevice_notifier - unregister a network notifier block 1812 * @nb: notifier 1813 * 1814 * Unregister a notifier previously registered by 1815 * register_netdevice_notifier(). The notifier is unlinked into the 1816 * kernel structures and may then be reused. A negative errno code 1817 * is returned on a failure. 1818 * 1819 * After unregistering unregister and down device events are synthesized 1820 * for all devices on the device list to the removed notifier to remove 1821 * the need for special case cleanup code. 1822 */ 1823 1824 int unregister_netdevice_notifier(struct notifier_block *nb) 1825 { 1826 struct net *net; 1827 int err; 1828 1829 /* Close race with setup_net() and cleanup_net() */ 1830 down_write(&pernet_ops_rwsem); 1831 rtnl_lock(); 1832 err = raw_notifier_chain_unregister(&netdev_chain, nb); 1833 if (err) 1834 goto unlock; 1835 1836 for_each_net(net) 1837 call_netdevice_unregister_net_notifiers(nb, net); 1838 1839 unlock: 1840 rtnl_unlock(); 1841 up_write(&pernet_ops_rwsem); 1842 return err; 1843 } 1844 EXPORT_SYMBOL(unregister_netdevice_notifier); 1845 1846 static int __register_netdevice_notifier_net(struct net *net, 1847 struct notifier_block *nb, 1848 bool ignore_call_fail) 1849 { 1850 int err; 1851 1852 err = raw_notifier_chain_register(&net->netdev_chain, nb); 1853 if (err) 1854 return err; 1855 if (dev_boot_phase) 1856 return 0; 1857 1858 err = call_netdevice_register_net_notifiers(nb, net); 1859 if (err && !ignore_call_fail) 1860 goto chain_unregister; 1861 1862 return 0; 1863 1864 chain_unregister: 1865 raw_notifier_chain_unregister(&net->netdev_chain, nb); 1866 return err; 1867 } 1868 1869 static int __unregister_netdevice_notifier_net(struct net *net, 1870 struct notifier_block *nb) 1871 { 1872 int err; 1873 1874 err = raw_notifier_chain_unregister(&net->netdev_chain, nb); 1875 if (err) 1876 return err; 1877 1878 call_netdevice_unregister_net_notifiers(nb, net); 1879 return 0; 1880 } 1881 1882 /** 1883 * register_netdevice_notifier_net - register a per-netns network notifier block 1884 * @net: network namespace 1885 * @nb: notifier 1886 * 1887 * Register a notifier to be called when network device events occur. 1888 * The notifier passed is linked into the kernel structures and must 1889 * not be reused until it has been unregistered. A negative errno code 1890 * is returned on a failure. 1891 * 1892 * When registered all registration and up events are replayed 1893 * to the new notifier to allow device to have a race free 1894 * view of the network device list. 1895 */ 1896 1897 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb) 1898 { 1899 int err; 1900 1901 rtnl_lock(); 1902 err = __register_netdevice_notifier_net(net, nb, false); 1903 rtnl_unlock(); 1904 return err; 1905 } 1906 EXPORT_SYMBOL(register_netdevice_notifier_net); 1907 1908 /** 1909 * unregister_netdevice_notifier_net - unregister a per-netns 1910 * network notifier block 1911 * @net: network namespace 1912 * @nb: notifier 1913 * 1914 * Unregister a notifier previously registered by 1915 * register_netdevice_notifier(). The notifier is unlinked into the 1916 * kernel structures and may then be reused. A negative errno code 1917 * is returned on a failure. 1918 * 1919 * After unregistering unregister and down device events are synthesized 1920 * for all devices on the device list to the removed notifier to remove 1921 * the need for special case cleanup code. 1922 */ 1923 1924 int unregister_netdevice_notifier_net(struct net *net, 1925 struct notifier_block *nb) 1926 { 1927 int err; 1928 1929 rtnl_lock(); 1930 err = __unregister_netdevice_notifier_net(net, nb); 1931 rtnl_unlock(); 1932 return err; 1933 } 1934 EXPORT_SYMBOL(unregister_netdevice_notifier_net); 1935 1936 int register_netdevice_notifier_dev_net(struct net_device *dev, 1937 struct notifier_block *nb, 1938 struct netdev_net_notifier *nn) 1939 { 1940 int err; 1941 1942 rtnl_lock(); 1943 err = __register_netdevice_notifier_net(dev_net(dev), nb, false); 1944 if (!err) { 1945 nn->nb = nb; 1946 list_add(&nn->list, &dev->net_notifier_list); 1947 } 1948 rtnl_unlock(); 1949 return err; 1950 } 1951 EXPORT_SYMBOL(register_netdevice_notifier_dev_net); 1952 1953 int unregister_netdevice_notifier_dev_net(struct net_device *dev, 1954 struct notifier_block *nb, 1955 struct netdev_net_notifier *nn) 1956 { 1957 int err; 1958 1959 rtnl_lock(); 1960 list_del(&nn->list); 1961 err = __unregister_netdevice_notifier_net(dev_net(dev), nb); 1962 rtnl_unlock(); 1963 return err; 1964 } 1965 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net); 1966 1967 static void move_netdevice_notifiers_dev_net(struct net_device *dev, 1968 struct net *net) 1969 { 1970 struct netdev_net_notifier *nn; 1971 1972 list_for_each_entry(nn, &dev->net_notifier_list, list) { 1973 __unregister_netdevice_notifier_net(dev_net(dev), nn->nb); 1974 __register_netdevice_notifier_net(net, nn->nb, true); 1975 } 1976 } 1977 1978 /** 1979 * call_netdevice_notifiers_info - call all network notifier blocks 1980 * @val: value passed unmodified to notifier function 1981 * @info: notifier information data 1982 * 1983 * Call all network notifier blocks. Parameters and return value 1984 * are as for raw_notifier_call_chain(). 1985 */ 1986 1987 static int call_netdevice_notifiers_info(unsigned long val, 1988 struct netdev_notifier_info *info) 1989 { 1990 struct net *net = dev_net(info->dev); 1991 int ret; 1992 1993 ASSERT_RTNL(); 1994 1995 /* Run per-netns notifier block chain first, then run the global one. 1996 * Hopefully, one day, the global one is going to be removed after 1997 * all notifier block registrators get converted to be per-netns. 1998 */ 1999 ret = raw_notifier_call_chain(&net->netdev_chain, val, info); 2000 if (ret & NOTIFY_STOP_MASK) 2001 return ret; 2002 return raw_notifier_call_chain(&netdev_chain, val, info); 2003 } 2004 2005 static int call_netdevice_notifiers_extack(unsigned long val, 2006 struct net_device *dev, 2007 struct netlink_ext_ack *extack) 2008 { 2009 struct netdev_notifier_info info = { 2010 .dev = dev, 2011 .extack = extack, 2012 }; 2013 2014 return call_netdevice_notifiers_info(val, &info); 2015 } 2016 2017 /** 2018 * call_netdevice_notifiers - call all network notifier blocks 2019 * @val: value passed unmodified to notifier function 2020 * @dev: net_device pointer passed unmodified to notifier function 2021 * 2022 * Call all network notifier blocks. Parameters and return value 2023 * are as for raw_notifier_call_chain(). 2024 */ 2025 2026 int call_netdevice_notifiers(unsigned long val, struct net_device *dev) 2027 { 2028 return call_netdevice_notifiers_extack(val, dev, NULL); 2029 } 2030 EXPORT_SYMBOL(call_netdevice_notifiers); 2031 2032 /** 2033 * call_netdevice_notifiers_mtu - call all network notifier blocks 2034 * @val: value passed unmodified to notifier function 2035 * @dev: net_device pointer passed unmodified to notifier function 2036 * @arg: additional u32 argument passed to the notifier function 2037 * 2038 * Call all network notifier blocks. Parameters and return value 2039 * are as for raw_notifier_call_chain(). 2040 */ 2041 static int call_netdevice_notifiers_mtu(unsigned long val, 2042 struct net_device *dev, u32 arg) 2043 { 2044 struct netdev_notifier_info_ext info = { 2045 .info.dev = dev, 2046 .ext.mtu = arg, 2047 }; 2048 2049 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0); 2050 2051 return call_netdevice_notifiers_info(val, &info.info); 2052 } 2053 2054 #ifdef CONFIG_NET_INGRESS 2055 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key); 2056 2057 void net_inc_ingress_queue(void) 2058 { 2059 static_branch_inc(&ingress_needed_key); 2060 } 2061 EXPORT_SYMBOL_GPL(net_inc_ingress_queue); 2062 2063 void net_dec_ingress_queue(void) 2064 { 2065 static_branch_dec(&ingress_needed_key); 2066 } 2067 EXPORT_SYMBOL_GPL(net_dec_ingress_queue); 2068 #endif 2069 2070 #ifdef CONFIG_NET_EGRESS 2071 static DEFINE_STATIC_KEY_FALSE(egress_needed_key); 2072 2073 void net_inc_egress_queue(void) 2074 { 2075 static_branch_inc(&egress_needed_key); 2076 } 2077 EXPORT_SYMBOL_GPL(net_inc_egress_queue); 2078 2079 void net_dec_egress_queue(void) 2080 { 2081 static_branch_dec(&egress_needed_key); 2082 } 2083 EXPORT_SYMBOL_GPL(net_dec_egress_queue); 2084 #endif 2085 2086 static DEFINE_STATIC_KEY_FALSE(netstamp_needed_key); 2087 #ifdef CONFIG_JUMP_LABEL 2088 static atomic_t netstamp_needed_deferred; 2089 static atomic_t netstamp_wanted; 2090 static void netstamp_clear(struct work_struct *work) 2091 { 2092 int deferred = atomic_xchg(&netstamp_needed_deferred, 0); 2093 int wanted; 2094 2095 wanted = atomic_add_return(deferred, &netstamp_wanted); 2096 if (wanted > 0) 2097 static_branch_enable(&netstamp_needed_key); 2098 else 2099 static_branch_disable(&netstamp_needed_key); 2100 } 2101 static DECLARE_WORK(netstamp_work, netstamp_clear); 2102 #endif 2103 2104 void net_enable_timestamp(void) 2105 { 2106 #ifdef CONFIG_JUMP_LABEL 2107 int wanted; 2108 2109 while (1) { 2110 wanted = atomic_read(&netstamp_wanted); 2111 if (wanted <= 0) 2112 break; 2113 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted) 2114 return; 2115 } 2116 atomic_inc(&netstamp_needed_deferred); 2117 schedule_work(&netstamp_work); 2118 #else 2119 static_branch_inc(&netstamp_needed_key); 2120 #endif 2121 } 2122 EXPORT_SYMBOL(net_enable_timestamp); 2123 2124 void net_disable_timestamp(void) 2125 { 2126 #ifdef CONFIG_JUMP_LABEL 2127 int wanted; 2128 2129 while (1) { 2130 wanted = atomic_read(&netstamp_wanted); 2131 if (wanted <= 1) 2132 break; 2133 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted) 2134 return; 2135 } 2136 atomic_dec(&netstamp_needed_deferred); 2137 schedule_work(&netstamp_work); 2138 #else 2139 static_branch_dec(&netstamp_needed_key); 2140 #endif 2141 } 2142 EXPORT_SYMBOL(net_disable_timestamp); 2143 2144 static inline void net_timestamp_set(struct sk_buff *skb) 2145 { 2146 skb->tstamp = 0; 2147 if (static_branch_unlikely(&netstamp_needed_key)) 2148 __net_timestamp(skb); 2149 } 2150 2151 #define net_timestamp_check(COND, SKB) \ 2152 if (static_branch_unlikely(&netstamp_needed_key)) { \ 2153 if ((COND) && !(SKB)->tstamp) \ 2154 __net_timestamp(SKB); \ 2155 } \ 2156 2157 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb) 2158 { 2159 return __is_skb_forwardable(dev, skb, true); 2160 } 2161 EXPORT_SYMBOL_GPL(is_skb_forwardable); 2162 2163 static int __dev_forward_skb2(struct net_device *dev, struct sk_buff *skb, 2164 bool check_mtu) 2165 { 2166 int ret = ____dev_forward_skb(dev, skb, check_mtu); 2167 2168 if (likely(!ret)) { 2169 skb->protocol = eth_type_trans(skb, dev); 2170 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); 2171 } 2172 2173 return ret; 2174 } 2175 2176 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2177 { 2178 return __dev_forward_skb2(dev, skb, true); 2179 } 2180 EXPORT_SYMBOL_GPL(__dev_forward_skb); 2181 2182 /** 2183 * dev_forward_skb - loopback an skb to another netif 2184 * 2185 * @dev: destination network device 2186 * @skb: buffer to forward 2187 * 2188 * return values: 2189 * NET_RX_SUCCESS (no congestion) 2190 * NET_RX_DROP (packet was dropped, but freed) 2191 * 2192 * dev_forward_skb can be used for injecting an skb from the 2193 * start_xmit function of one device into the receive queue 2194 * of another device. 2195 * 2196 * The receiving device may be in another namespace, so 2197 * we have to clear all information in the skb that could 2198 * impact namespace isolation. 2199 */ 2200 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2201 { 2202 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); 2203 } 2204 EXPORT_SYMBOL_GPL(dev_forward_skb); 2205 2206 int dev_forward_skb_nomtu(struct net_device *dev, struct sk_buff *skb) 2207 { 2208 return __dev_forward_skb2(dev, skb, false) ?: netif_rx_internal(skb); 2209 } 2210 2211 static inline int deliver_skb(struct sk_buff *skb, 2212 struct packet_type *pt_prev, 2213 struct net_device *orig_dev) 2214 { 2215 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 2216 return -ENOMEM; 2217 refcount_inc(&skb->users); 2218 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 2219 } 2220 2221 static inline void deliver_ptype_list_skb(struct sk_buff *skb, 2222 struct packet_type **pt, 2223 struct net_device *orig_dev, 2224 __be16 type, 2225 struct list_head *ptype_list) 2226 { 2227 struct packet_type *ptype, *pt_prev = *pt; 2228 2229 list_for_each_entry_rcu(ptype, ptype_list, list) { 2230 if (ptype->type != type) 2231 continue; 2232 if (pt_prev) 2233 deliver_skb(skb, pt_prev, orig_dev); 2234 pt_prev = ptype; 2235 } 2236 *pt = pt_prev; 2237 } 2238 2239 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) 2240 { 2241 if (!ptype->af_packet_priv || !skb->sk) 2242 return false; 2243 2244 if (ptype->id_match) 2245 return ptype->id_match(ptype, skb->sk); 2246 else if ((struct sock *)ptype->af_packet_priv == skb->sk) 2247 return true; 2248 2249 return false; 2250 } 2251 2252 /** 2253 * dev_nit_active - return true if any network interface taps are in use 2254 * 2255 * @dev: network device to check for the presence of taps 2256 */ 2257 bool dev_nit_active(struct net_device *dev) 2258 { 2259 return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all); 2260 } 2261 EXPORT_SYMBOL_GPL(dev_nit_active); 2262 2263 /* 2264 * Support routine. Sends outgoing frames to any network 2265 * taps currently in use. 2266 */ 2267 2268 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) 2269 { 2270 struct packet_type *ptype; 2271 struct sk_buff *skb2 = NULL; 2272 struct packet_type *pt_prev = NULL; 2273 struct list_head *ptype_list = &ptype_all; 2274 2275 rcu_read_lock(); 2276 again: 2277 list_for_each_entry_rcu(ptype, ptype_list, list) { 2278 if (ptype->ignore_outgoing) 2279 continue; 2280 2281 /* Never send packets back to the socket 2282 * they originated from - MvS (miquels@drinkel.ow.org) 2283 */ 2284 if (skb_loop_sk(ptype, skb)) 2285 continue; 2286 2287 if (pt_prev) { 2288 deliver_skb(skb2, pt_prev, skb->dev); 2289 pt_prev = ptype; 2290 continue; 2291 } 2292 2293 /* need to clone skb, done only once */ 2294 skb2 = skb_clone(skb, GFP_ATOMIC); 2295 if (!skb2) 2296 goto out_unlock; 2297 2298 net_timestamp_set(skb2); 2299 2300 /* skb->nh should be correctly 2301 * set by sender, so that the second statement is 2302 * just protection against buggy protocols. 2303 */ 2304 skb_reset_mac_header(skb2); 2305 2306 if (skb_network_header(skb2) < skb2->data || 2307 skb_network_header(skb2) > skb_tail_pointer(skb2)) { 2308 net_crit_ratelimited("protocol %04x is buggy, dev %s\n", 2309 ntohs(skb2->protocol), 2310 dev->name); 2311 skb_reset_network_header(skb2); 2312 } 2313 2314 skb2->transport_header = skb2->network_header; 2315 skb2->pkt_type = PACKET_OUTGOING; 2316 pt_prev = ptype; 2317 } 2318 2319 if (ptype_list == &ptype_all) { 2320 ptype_list = &dev->ptype_all; 2321 goto again; 2322 } 2323 out_unlock: 2324 if (pt_prev) { 2325 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC)) 2326 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); 2327 else 2328 kfree_skb(skb2); 2329 } 2330 rcu_read_unlock(); 2331 } 2332 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit); 2333 2334 /** 2335 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change 2336 * @dev: Network device 2337 * @txq: number of queues available 2338 * 2339 * If real_num_tx_queues is changed the tc mappings may no longer be 2340 * valid. To resolve this verify the tc mapping remains valid and if 2341 * not NULL the mapping. With no priorities mapping to this 2342 * offset/count pair it will no longer be used. In the worst case TC0 2343 * is invalid nothing can be done so disable priority mappings. If is 2344 * expected that drivers will fix this mapping if they can before 2345 * calling netif_set_real_num_tx_queues. 2346 */ 2347 static void netif_setup_tc(struct net_device *dev, unsigned int txq) 2348 { 2349 int i; 2350 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2351 2352 /* If TC0 is invalidated disable TC mapping */ 2353 if (tc->offset + tc->count > txq) { 2354 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); 2355 dev->num_tc = 0; 2356 return; 2357 } 2358 2359 /* Invalidated prio to tc mappings set to TC0 */ 2360 for (i = 1; i < TC_BITMASK + 1; i++) { 2361 int q = netdev_get_prio_tc_map(dev, i); 2362 2363 tc = &dev->tc_to_txq[q]; 2364 if (tc->offset + tc->count > txq) { 2365 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n", 2366 i, q); 2367 netdev_set_prio_tc_map(dev, i, 0); 2368 } 2369 } 2370 } 2371 2372 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq) 2373 { 2374 if (dev->num_tc) { 2375 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2376 int i; 2377 2378 /* walk through the TCs and see if it falls into any of them */ 2379 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) { 2380 if ((txq - tc->offset) < tc->count) 2381 return i; 2382 } 2383 2384 /* didn't find it, just return -1 to indicate no match */ 2385 return -1; 2386 } 2387 2388 return 0; 2389 } 2390 EXPORT_SYMBOL(netdev_txq_to_tc); 2391 2392 #ifdef CONFIG_XPS 2393 static struct static_key xps_needed __read_mostly; 2394 static struct static_key xps_rxqs_needed __read_mostly; 2395 static DEFINE_MUTEX(xps_map_mutex); 2396 #define xmap_dereference(P) \ 2397 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) 2398 2399 static bool remove_xps_queue(struct xps_dev_maps *dev_maps, 2400 struct xps_dev_maps *old_maps, int tci, u16 index) 2401 { 2402 struct xps_map *map = NULL; 2403 int pos; 2404 2405 if (dev_maps) 2406 map = xmap_dereference(dev_maps->attr_map[tci]); 2407 if (!map) 2408 return false; 2409 2410 for (pos = map->len; pos--;) { 2411 if (map->queues[pos] != index) 2412 continue; 2413 2414 if (map->len > 1) { 2415 map->queues[pos] = map->queues[--map->len]; 2416 break; 2417 } 2418 2419 if (old_maps) 2420 RCU_INIT_POINTER(old_maps->attr_map[tci], NULL); 2421 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2422 kfree_rcu(map, rcu); 2423 return false; 2424 } 2425 2426 return true; 2427 } 2428 2429 static bool remove_xps_queue_cpu(struct net_device *dev, 2430 struct xps_dev_maps *dev_maps, 2431 int cpu, u16 offset, u16 count) 2432 { 2433 int num_tc = dev_maps->num_tc; 2434 bool active = false; 2435 int tci; 2436 2437 for (tci = cpu * num_tc; num_tc--; tci++) { 2438 int i, j; 2439 2440 for (i = count, j = offset; i--; j++) { 2441 if (!remove_xps_queue(dev_maps, NULL, tci, j)) 2442 break; 2443 } 2444 2445 active |= i < 0; 2446 } 2447 2448 return active; 2449 } 2450 2451 static void reset_xps_maps(struct net_device *dev, 2452 struct xps_dev_maps *dev_maps, 2453 enum xps_map_type type) 2454 { 2455 static_key_slow_dec_cpuslocked(&xps_needed); 2456 if (type == XPS_RXQS) 2457 static_key_slow_dec_cpuslocked(&xps_rxqs_needed); 2458 2459 RCU_INIT_POINTER(dev->xps_maps[type], NULL); 2460 2461 kfree_rcu(dev_maps, rcu); 2462 } 2463 2464 static void clean_xps_maps(struct net_device *dev, enum xps_map_type type, 2465 u16 offset, u16 count) 2466 { 2467 struct xps_dev_maps *dev_maps; 2468 bool active = false; 2469 int i, j; 2470 2471 dev_maps = xmap_dereference(dev->xps_maps[type]); 2472 if (!dev_maps) 2473 return; 2474 2475 for (j = 0; j < dev_maps->nr_ids; j++) 2476 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, count); 2477 if (!active) 2478 reset_xps_maps(dev, dev_maps, type); 2479 2480 if (type == XPS_CPUS) { 2481 for (i = offset + (count - 1); count--; i--) 2482 netdev_queue_numa_node_write( 2483 netdev_get_tx_queue(dev, i), NUMA_NO_NODE); 2484 } 2485 } 2486 2487 static void netif_reset_xps_queues(struct net_device *dev, u16 offset, 2488 u16 count) 2489 { 2490 if (!static_key_false(&xps_needed)) 2491 return; 2492 2493 cpus_read_lock(); 2494 mutex_lock(&xps_map_mutex); 2495 2496 if (static_key_false(&xps_rxqs_needed)) 2497 clean_xps_maps(dev, XPS_RXQS, offset, count); 2498 2499 clean_xps_maps(dev, XPS_CPUS, offset, count); 2500 2501 mutex_unlock(&xps_map_mutex); 2502 cpus_read_unlock(); 2503 } 2504 2505 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) 2506 { 2507 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index); 2508 } 2509 2510 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index, 2511 u16 index, bool is_rxqs_map) 2512 { 2513 struct xps_map *new_map; 2514 int alloc_len = XPS_MIN_MAP_ALLOC; 2515 int i, pos; 2516 2517 for (pos = 0; map && pos < map->len; pos++) { 2518 if (map->queues[pos] != index) 2519 continue; 2520 return map; 2521 } 2522 2523 /* Need to add tx-queue to this CPU's/rx-queue's existing map */ 2524 if (map) { 2525 if (pos < map->alloc_len) 2526 return map; 2527 2528 alloc_len = map->alloc_len * 2; 2529 } 2530 2531 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's 2532 * map 2533 */ 2534 if (is_rxqs_map) 2535 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL); 2536 else 2537 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, 2538 cpu_to_node(attr_index)); 2539 if (!new_map) 2540 return NULL; 2541 2542 for (i = 0; i < pos; i++) 2543 new_map->queues[i] = map->queues[i]; 2544 new_map->alloc_len = alloc_len; 2545 new_map->len = pos; 2546 2547 return new_map; 2548 } 2549 2550 /* Copy xps maps at a given index */ 2551 static void xps_copy_dev_maps(struct xps_dev_maps *dev_maps, 2552 struct xps_dev_maps *new_dev_maps, int index, 2553 int tc, bool skip_tc) 2554 { 2555 int i, tci = index * dev_maps->num_tc; 2556 struct xps_map *map; 2557 2558 /* copy maps belonging to foreign traffic classes */ 2559 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2560 if (i == tc && skip_tc) 2561 continue; 2562 2563 /* fill in the new device map from the old device map */ 2564 map = xmap_dereference(dev_maps->attr_map[tci]); 2565 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2566 } 2567 } 2568 2569 /* Must be called under cpus_read_lock */ 2570 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask, 2571 u16 index, enum xps_map_type type) 2572 { 2573 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL, *old_dev_maps = NULL; 2574 const unsigned long *online_mask = NULL; 2575 bool active = false, copy = false; 2576 int i, j, tci, numa_node_id = -2; 2577 int maps_sz, num_tc = 1, tc = 0; 2578 struct xps_map *map, *new_map; 2579 unsigned int nr_ids; 2580 2581 if (dev->num_tc) { 2582 /* Do not allow XPS on subordinate device directly */ 2583 num_tc = dev->num_tc; 2584 if (num_tc < 0) 2585 return -EINVAL; 2586 2587 /* If queue belongs to subordinate dev use its map */ 2588 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev; 2589 2590 tc = netdev_txq_to_tc(dev, index); 2591 if (tc < 0) 2592 return -EINVAL; 2593 } 2594 2595 mutex_lock(&xps_map_mutex); 2596 2597 dev_maps = xmap_dereference(dev->xps_maps[type]); 2598 if (type == XPS_RXQS) { 2599 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues); 2600 nr_ids = dev->num_rx_queues; 2601 } else { 2602 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc); 2603 if (num_possible_cpus() > 1) 2604 online_mask = cpumask_bits(cpu_online_mask); 2605 nr_ids = nr_cpu_ids; 2606 } 2607 2608 if (maps_sz < L1_CACHE_BYTES) 2609 maps_sz = L1_CACHE_BYTES; 2610 2611 /* The old dev_maps could be larger or smaller than the one we're 2612 * setting up now, as dev->num_tc or nr_ids could have been updated in 2613 * between. We could try to be smart, but let's be safe instead and only 2614 * copy foreign traffic classes if the two map sizes match. 2615 */ 2616 if (dev_maps && 2617 dev_maps->num_tc == num_tc && dev_maps->nr_ids == nr_ids) 2618 copy = true; 2619 2620 /* allocate memory for queue storage */ 2621 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids), 2622 j < nr_ids;) { 2623 if (!new_dev_maps) { 2624 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); 2625 if (!new_dev_maps) { 2626 mutex_unlock(&xps_map_mutex); 2627 return -ENOMEM; 2628 } 2629 2630 new_dev_maps->nr_ids = nr_ids; 2631 new_dev_maps->num_tc = num_tc; 2632 } 2633 2634 tci = j * num_tc + tc; 2635 map = copy ? xmap_dereference(dev_maps->attr_map[tci]) : NULL; 2636 2637 map = expand_xps_map(map, j, index, type == XPS_RXQS); 2638 if (!map) 2639 goto error; 2640 2641 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2642 } 2643 2644 if (!new_dev_maps) 2645 goto out_no_new_maps; 2646 2647 if (!dev_maps) { 2648 /* Increment static keys at most once per type */ 2649 static_key_slow_inc_cpuslocked(&xps_needed); 2650 if (type == XPS_RXQS) 2651 static_key_slow_inc_cpuslocked(&xps_rxqs_needed); 2652 } 2653 2654 for (j = 0; j < nr_ids; j++) { 2655 bool skip_tc = false; 2656 2657 tci = j * num_tc + tc; 2658 if (netif_attr_test_mask(j, mask, nr_ids) && 2659 netif_attr_test_online(j, online_mask, nr_ids)) { 2660 /* add tx-queue to CPU/rx-queue maps */ 2661 int pos = 0; 2662 2663 skip_tc = true; 2664 2665 map = xmap_dereference(new_dev_maps->attr_map[tci]); 2666 while ((pos < map->len) && (map->queues[pos] != index)) 2667 pos++; 2668 2669 if (pos == map->len) 2670 map->queues[map->len++] = index; 2671 #ifdef CONFIG_NUMA 2672 if (type == XPS_CPUS) { 2673 if (numa_node_id == -2) 2674 numa_node_id = cpu_to_node(j); 2675 else if (numa_node_id != cpu_to_node(j)) 2676 numa_node_id = -1; 2677 } 2678 #endif 2679 } 2680 2681 if (copy) 2682 xps_copy_dev_maps(dev_maps, new_dev_maps, j, tc, 2683 skip_tc); 2684 } 2685 2686 rcu_assign_pointer(dev->xps_maps[type], new_dev_maps); 2687 2688 /* Cleanup old maps */ 2689 if (!dev_maps) 2690 goto out_no_old_maps; 2691 2692 for (j = 0; j < dev_maps->nr_ids; j++) { 2693 for (i = num_tc, tci = j * dev_maps->num_tc; i--; tci++) { 2694 map = xmap_dereference(dev_maps->attr_map[tci]); 2695 if (!map) 2696 continue; 2697 2698 if (copy) { 2699 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2700 if (map == new_map) 2701 continue; 2702 } 2703 2704 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2705 kfree_rcu(map, rcu); 2706 } 2707 } 2708 2709 old_dev_maps = dev_maps; 2710 2711 out_no_old_maps: 2712 dev_maps = new_dev_maps; 2713 active = true; 2714 2715 out_no_new_maps: 2716 if (type == XPS_CPUS) 2717 /* update Tx queue numa node */ 2718 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), 2719 (numa_node_id >= 0) ? 2720 numa_node_id : NUMA_NO_NODE); 2721 2722 if (!dev_maps) 2723 goto out_no_maps; 2724 2725 /* removes tx-queue from unused CPUs/rx-queues */ 2726 for (j = 0; j < dev_maps->nr_ids; j++) { 2727 tci = j * dev_maps->num_tc; 2728 2729 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2730 if (i == tc && 2731 netif_attr_test_mask(j, mask, dev_maps->nr_ids) && 2732 netif_attr_test_online(j, online_mask, dev_maps->nr_ids)) 2733 continue; 2734 2735 active |= remove_xps_queue(dev_maps, 2736 copy ? old_dev_maps : NULL, 2737 tci, index); 2738 } 2739 } 2740 2741 if (old_dev_maps) 2742 kfree_rcu(old_dev_maps, rcu); 2743 2744 /* free map if not active */ 2745 if (!active) 2746 reset_xps_maps(dev, dev_maps, type); 2747 2748 out_no_maps: 2749 mutex_unlock(&xps_map_mutex); 2750 2751 return 0; 2752 error: 2753 /* remove any maps that we added */ 2754 for (j = 0; j < nr_ids; j++) { 2755 for (i = num_tc, tci = j * num_tc; i--; tci++) { 2756 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2757 map = copy ? 2758 xmap_dereference(dev_maps->attr_map[tci]) : 2759 NULL; 2760 if (new_map && new_map != map) 2761 kfree(new_map); 2762 } 2763 } 2764 2765 mutex_unlock(&xps_map_mutex); 2766 2767 kfree(new_dev_maps); 2768 return -ENOMEM; 2769 } 2770 EXPORT_SYMBOL_GPL(__netif_set_xps_queue); 2771 2772 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, 2773 u16 index) 2774 { 2775 int ret; 2776 2777 cpus_read_lock(); 2778 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, XPS_CPUS); 2779 cpus_read_unlock(); 2780 2781 return ret; 2782 } 2783 EXPORT_SYMBOL(netif_set_xps_queue); 2784 2785 #endif 2786 static void netdev_unbind_all_sb_channels(struct net_device *dev) 2787 { 2788 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2789 2790 /* Unbind any subordinate channels */ 2791 while (txq-- != &dev->_tx[0]) { 2792 if (txq->sb_dev) 2793 netdev_unbind_sb_channel(dev, txq->sb_dev); 2794 } 2795 } 2796 2797 void netdev_reset_tc(struct net_device *dev) 2798 { 2799 #ifdef CONFIG_XPS 2800 netif_reset_xps_queues_gt(dev, 0); 2801 #endif 2802 netdev_unbind_all_sb_channels(dev); 2803 2804 /* Reset TC configuration of device */ 2805 dev->num_tc = 0; 2806 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq)); 2807 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map)); 2808 } 2809 EXPORT_SYMBOL(netdev_reset_tc); 2810 2811 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset) 2812 { 2813 if (tc >= dev->num_tc) 2814 return -EINVAL; 2815 2816 #ifdef CONFIG_XPS 2817 netif_reset_xps_queues(dev, offset, count); 2818 #endif 2819 dev->tc_to_txq[tc].count = count; 2820 dev->tc_to_txq[tc].offset = offset; 2821 return 0; 2822 } 2823 EXPORT_SYMBOL(netdev_set_tc_queue); 2824 2825 int netdev_set_num_tc(struct net_device *dev, u8 num_tc) 2826 { 2827 if (num_tc > TC_MAX_QUEUE) 2828 return -EINVAL; 2829 2830 #ifdef CONFIG_XPS 2831 netif_reset_xps_queues_gt(dev, 0); 2832 #endif 2833 netdev_unbind_all_sb_channels(dev); 2834 2835 dev->num_tc = num_tc; 2836 return 0; 2837 } 2838 EXPORT_SYMBOL(netdev_set_num_tc); 2839 2840 void netdev_unbind_sb_channel(struct net_device *dev, 2841 struct net_device *sb_dev) 2842 { 2843 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2844 2845 #ifdef CONFIG_XPS 2846 netif_reset_xps_queues_gt(sb_dev, 0); 2847 #endif 2848 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq)); 2849 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map)); 2850 2851 while (txq-- != &dev->_tx[0]) { 2852 if (txq->sb_dev == sb_dev) 2853 txq->sb_dev = NULL; 2854 } 2855 } 2856 EXPORT_SYMBOL(netdev_unbind_sb_channel); 2857 2858 int netdev_bind_sb_channel_queue(struct net_device *dev, 2859 struct net_device *sb_dev, 2860 u8 tc, u16 count, u16 offset) 2861 { 2862 /* Make certain the sb_dev and dev are already configured */ 2863 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc) 2864 return -EINVAL; 2865 2866 /* We cannot hand out queues we don't have */ 2867 if ((offset + count) > dev->real_num_tx_queues) 2868 return -EINVAL; 2869 2870 /* Record the mapping */ 2871 sb_dev->tc_to_txq[tc].count = count; 2872 sb_dev->tc_to_txq[tc].offset = offset; 2873 2874 /* Provide a way for Tx queue to find the tc_to_txq map or 2875 * XPS map for itself. 2876 */ 2877 while (count--) 2878 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev; 2879 2880 return 0; 2881 } 2882 EXPORT_SYMBOL(netdev_bind_sb_channel_queue); 2883 2884 int netdev_set_sb_channel(struct net_device *dev, u16 channel) 2885 { 2886 /* Do not use a multiqueue device to represent a subordinate channel */ 2887 if (netif_is_multiqueue(dev)) 2888 return -ENODEV; 2889 2890 /* We allow channels 1 - 32767 to be used for subordinate channels. 2891 * Channel 0 is meant to be "native" mode and used only to represent 2892 * the main root device. We allow writing 0 to reset the device back 2893 * to normal mode after being used as a subordinate channel. 2894 */ 2895 if (channel > S16_MAX) 2896 return -EINVAL; 2897 2898 dev->num_tc = -channel; 2899 2900 return 0; 2901 } 2902 EXPORT_SYMBOL(netdev_set_sb_channel); 2903 2904 /* 2905 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues 2906 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed. 2907 */ 2908 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) 2909 { 2910 bool disabling; 2911 int rc; 2912 2913 disabling = txq < dev->real_num_tx_queues; 2914 2915 if (txq < 1 || txq > dev->num_tx_queues) 2916 return -EINVAL; 2917 2918 if (dev->reg_state == NETREG_REGISTERED || 2919 dev->reg_state == NETREG_UNREGISTERING) { 2920 ASSERT_RTNL(); 2921 2922 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, 2923 txq); 2924 if (rc) 2925 return rc; 2926 2927 if (dev->num_tc) 2928 netif_setup_tc(dev, txq); 2929 2930 dev_qdisc_change_real_num_tx(dev, txq); 2931 2932 dev->real_num_tx_queues = txq; 2933 2934 if (disabling) { 2935 synchronize_net(); 2936 qdisc_reset_all_tx_gt(dev, txq); 2937 #ifdef CONFIG_XPS 2938 netif_reset_xps_queues_gt(dev, txq); 2939 #endif 2940 } 2941 } else { 2942 dev->real_num_tx_queues = txq; 2943 } 2944 2945 return 0; 2946 } 2947 EXPORT_SYMBOL(netif_set_real_num_tx_queues); 2948 2949 #ifdef CONFIG_SYSFS 2950 /** 2951 * netif_set_real_num_rx_queues - set actual number of RX queues used 2952 * @dev: Network device 2953 * @rxq: Actual number of RX queues 2954 * 2955 * This must be called either with the rtnl_lock held or before 2956 * registration of the net device. Returns 0 on success, or a 2957 * negative error code. If called before registration, it always 2958 * succeeds. 2959 */ 2960 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) 2961 { 2962 int rc; 2963 2964 if (rxq < 1 || rxq > dev->num_rx_queues) 2965 return -EINVAL; 2966 2967 if (dev->reg_state == NETREG_REGISTERED) { 2968 ASSERT_RTNL(); 2969 2970 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, 2971 rxq); 2972 if (rc) 2973 return rc; 2974 } 2975 2976 dev->real_num_rx_queues = rxq; 2977 return 0; 2978 } 2979 EXPORT_SYMBOL(netif_set_real_num_rx_queues); 2980 #endif 2981 2982 /** 2983 * netif_set_real_num_queues - set actual number of RX and TX queues used 2984 * @dev: Network device 2985 * @txq: Actual number of TX queues 2986 * @rxq: Actual number of RX queues 2987 * 2988 * Set the real number of both TX and RX queues. 2989 * Does nothing if the number of queues is already correct. 2990 */ 2991 int netif_set_real_num_queues(struct net_device *dev, 2992 unsigned int txq, unsigned int rxq) 2993 { 2994 unsigned int old_rxq = dev->real_num_rx_queues; 2995 int err; 2996 2997 if (txq < 1 || txq > dev->num_tx_queues || 2998 rxq < 1 || rxq > dev->num_rx_queues) 2999 return -EINVAL; 3000 3001 /* Start from increases, so the error path only does decreases - 3002 * decreases can't fail. 3003 */ 3004 if (rxq > dev->real_num_rx_queues) { 3005 err = netif_set_real_num_rx_queues(dev, rxq); 3006 if (err) 3007 return err; 3008 } 3009 if (txq > dev->real_num_tx_queues) { 3010 err = netif_set_real_num_tx_queues(dev, txq); 3011 if (err) 3012 goto undo_rx; 3013 } 3014 if (rxq < dev->real_num_rx_queues) 3015 WARN_ON(netif_set_real_num_rx_queues(dev, rxq)); 3016 if (txq < dev->real_num_tx_queues) 3017 WARN_ON(netif_set_real_num_tx_queues(dev, txq)); 3018 3019 return 0; 3020 undo_rx: 3021 WARN_ON(netif_set_real_num_rx_queues(dev, old_rxq)); 3022 return err; 3023 } 3024 EXPORT_SYMBOL(netif_set_real_num_queues); 3025 3026 /** 3027 * netif_get_num_default_rss_queues - default number of RSS queues 3028 * 3029 * This routine should set an upper limit on the number of RSS queues 3030 * used by default by multiqueue devices. 3031 */ 3032 int netif_get_num_default_rss_queues(void) 3033 { 3034 return is_kdump_kernel() ? 3035 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus()); 3036 } 3037 EXPORT_SYMBOL(netif_get_num_default_rss_queues); 3038 3039 static void __netif_reschedule(struct Qdisc *q) 3040 { 3041 struct softnet_data *sd; 3042 unsigned long flags; 3043 3044 local_irq_save(flags); 3045 sd = this_cpu_ptr(&softnet_data); 3046 q->next_sched = NULL; 3047 *sd->output_queue_tailp = q; 3048 sd->output_queue_tailp = &q->next_sched; 3049 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3050 local_irq_restore(flags); 3051 } 3052 3053 void __netif_schedule(struct Qdisc *q) 3054 { 3055 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) 3056 __netif_reschedule(q); 3057 } 3058 EXPORT_SYMBOL(__netif_schedule); 3059 3060 struct dev_kfree_skb_cb { 3061 enum skb_free_reason reason; 3062 }; 3063 3064 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) 3065 { 3066 return (struct dev_kfree_skb_cb *)skb->cb; 3067 } 3068 3069 void netif_schedule_queue(struct netdev_queue *txq) 3070 { 3071 rcu_read_lock(); 3072 if (!netif_xmit_stopped(txq)) { 3073 struct Qdisc *q = rcu_dereference(txq->qdisc); 3074 3075 __netif_schedule(q); 3076 } 3077 rcu_read_unlock(); 3078 } 3079 EXPORT_SYMBOL(netif_schedule_queue); 3080 3081 void netif_tx_wake_queue(struct netdev_queue *dev_queue) 3082 { 3083 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { 3084 struct Qdisc *q; 3085 3086 rcu_read_lock(); 3087 q = rcu_dereference(dev_queue->qdisc); 3088 __netif_schedule(q); 3089 rcu_read_unlock(); 3090 } 3091 } 3092 EXPORT_SYMBOL(netif_tx_wake_queue); 3093 3094 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason) 3095 { 3096 unsigned long flags; 3097 3098 if (unlikely(!skb)) 3099 return; 3100 3101 if (likely(refcount_read(&skb->users) == 1)) { 3102 smp_rmb(); 3103 refcount_set(&skb->users, 0); 3104 } else if (likely(!refcount_dec_and_test(&skb->users))) { 3105 return; 3106 } 3107 get_kfree_skb_cb(skb)->reason = reason; 3108 local_irq_save(flags); 3109 skb->next = __this_cpu_read(softnet_data.completion_queue); 3110 __this_cpu_write(softnet_data.completion_queue, skb); 3111 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3112 local_irq_restore(flags); 3113 } 3114 EXPORT_SYMBOL(__dev_kfree_skb_irq); 3115 3116 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason) 3117 { 3118 if (in_hardirq() || irqs_disabled()) 3119 __dev_kfree_skb_irq(skb, reason); 3120 else 3121 dev_kfree_skb(skb); 3122 } 3123 EXPORT_SYMBOL(__dev_kfree_skb_any); 3124 3125 3126 /** 3127 * netif_device_detach - mark device as removed 3128 * @dev: network device 3129 * 3130 * Mark device as removed from system and therefore no longer available. 3131 */ 3132 void netif_device_detach(struct net_device *dev) 3133 { 3134 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && 3135 netif_running(dev)) { 3136 netif_tx_stop_all_queues(dev); 3137 } 3138 } 3139 EXPORT_SYMBOL(netif_device_detach); 3140 3141 /** 3142 * netif_device_attach - mark device as attached 3143 * @dev: network device 3144 * 3145 * Mark device as attached from system and restart if needed. 3146 */ 3147 void netif_device_attach(struct net_device *dev) 3148 { 3149 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && 3150 netif_running(dev)) { 3151 netif_tx_wake_all_queues(dev); 3152 __netdev_watchdog_up(dev); 3153 } 3154 } 3155 EXPORT_SYMBOL(netif_device_attach); 3156 3157 /* 3158 * Returns a Tx hash based on the given packet descriptor a Tx queues' number 3159 * to be used as a distribution range. 3160 */ 3161 static u16 skb_tx_hash(const struct net_device *dev, 3162 const struct net_device *sb_dev, 3163 struct sk_buff *skb) 3164 { 3165 u32 hash; 3166 u16 qoffset = 0; 3167 u16 qcount = dev->real_num_tx_queues; 3168 3169 if (dev->num_tc) { 3170 u8 tc = netdev_get_prio_tc_map(dev, skb->priority); 3171 3172 qoffset = sb_dev->tc_to_txq[tc].offset; 3173 qcount = sb_dev->tc_to_txq[tc].count; 3174 } 3175 3176 if (skb_rx_queue_recorded(skb)) { 3177 hash = skb_get_rx_queue(skb); 3178 if (hash >= qoffset) 3179 hash -= qoffset; 3180 while (unlikely(hash >= qcount)) 3181 hash -= qcount; 3182 return hash + qoffset; 3183 } 3184 3185 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; 3186 } 3187 3188 static void skb_warn_bad_offload(const struct sk_buff *skb) 3189 { 3190 static const netdev_features_t null_features; 3191 struct net_device *dev = skb->dev; 3192 const char *name = ""; 3193 3194 if (!net_ratelimit()) 3195 return; 3196 3197 if (dev) { 3198 if (dev->dev.parent) 3199 name = dev_driver_string(dev->dev.parent); 3200 else 3201 name = netdev_name(dev); 3202 } 3203 skb_dump(KERN_WARNING, skb, false); 3204 WARN(1, "%s: caps=(%pNF, %pNF)\n", 3205 name, dev ? &dev->features : &null_features, 3206 skb->sk ? &skb->sk->sk_route_caps : &null_features); 3207 } 3208 3209 /* 3210 * Invalidate hardware checksum when packet is to be mangled, and 3211 * complete checksum manually on outgoing path. 3212 */ 3213 int skb_checksum_help(struct sk_buff *skb) 3214 { 3215 __wsum csum; 3216 int ret = 0, offset; 3217 3218 if (skb->ip_summed == CHECKSUM_COMPLETE) 3219 goto out_set_summed; 3220 3221 if (unlikely(skb_is_gso(skb))) { 3222 skb_warn_bad_offload(skb); 3223 return -EINVAL; 3224 } 3225 3226 /* Before computing a checksum, we should make sure no frag could 3227 * be modified by an external entity : checksum could be wrong. 3228 */ 3229 if (skb_has_shared_frag(skb)) { 3230 ret = __skb_linearize(skb); 3231 if (ret) 3232 goto out; 3233 } 3234 3235 offset = skb_checksum_start_offset(skb); 3236 BUG_ON(offset >= skb_headlen(skb)); 3237 csum = skb_checksum(skb, offset, skb->len - offset, 0); 3238 3239 offset += skb->csum_offset; 3240 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); 3241 3242 ret = skb_ensure_writable(skb, offset + sizeof(__sum16)); 3243 if (ret) 3244 goto out; 3245 3246 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0; 3247 out_set_summed: 3248 skb->ip_summed = CHECKSUM_NONE; 3249 out: 3250 return ret; 3251 } 3252 EXPORT_SYMBOL(skb_checksum_help); 3253 3254 int skb_crc32c_csum_help(struct sk_buff *skb) 3255 { 3256 __le32 crc32c_csum; 3257 int ret = 0, offset, start; 3258 3259 if (skb->ip_summed != CHECKSUM_PARTIAL) 3260 goto out; 3261 3262 if (unlikely(skb_is_gso(skb))) 3263 goto out; 3264 3265 /* Before computing a checksum, we should make sure no frag could 3266 * be modified by an external entity : checksum could be wrong. 3267 */ 3268 if (unlikely(skb_has_shared_frag(skb))) { 3269 ret = __skb_linearize(skb); 3270 if (ret) 3271 goto out; 3272 } 3273 start = skb_checksum_start_offset(skb); 3274 offset = start + offsetof(struct sctphdr, checksum); 3275 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { 3276 ret = -EINVAL; 3277 goto out; 3278 } 3279 3280 ret = skb_ensure_writable(skb, offset + sizeof(__le32)); 3281 if (ret) 3282 goto out; 3283 3284 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start, 3285 skb->len - start, ~(__u32)0, 3286 crc32c_csum_stub)); 3287 *(__le32 *)(skb->data + offset) = crc32c_csum; 3288 skb->ip_summed = CHECKSUM_NONE; 3289 skb->csum_not_inet = 0; 3290 out: 3291 return ret; 3292 } 3293 3294 __be16 skb_network_protocol(struct sk_buff *skb, int *depth) 3295 { 3296 __be16 type = skb->protocol; 3297 3298 /* Tunnel gso handlers can set protocol to ethernet. */ 3299 if (type == htons(ETH_P_TEB)) { 3300 struct ethhdr *eth; 3301 3302 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) 3303 return 0; 3304 3305 eth = (struct ethhdr *)skb->data; 3306 type = eth->h_proto; 3307 } 3308 3309 return __vlan_get_protocol(skb, type, depth); 3310 } 3311 3312 /** 3313 * skb_mac_gso_segment - mac layer segmentation handler. 3314 * @skb: buffer to segment 3315 * @features: features for the output path (see dev->features) 3316 */ 3317 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb, 3318 netdev_features_t features) 3319 { 3320 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); 3321 struct packet_offload *ptype; 3322 int vlan_depth = skb->mac_len; 3323 __be16 type = skb_network_protocol(skb, &vlan_depth); 3324 3325 if (unlikely(!type)) 3326 return ERR_PTR(-EINVAL); 3327 3328 __skb_pull(skb, vlan_depth); 3329 3330 rcu_read_lock(); 3331 list_for_each_entry_rcu(ptype, &offload_base, list) { 3332 if (ptype->type == type && ptype->callbacks.gso_segment) { 3333 segs = ptype->callbacks.gso_segment(skb, features); 3334 break; 3335 } 3336 } 3337 rcu_read_unlock(); 3338 3339 __skb_push(skb, skb->data - skb_mac_header(skb)); 3340 3341 return segs; 3342 } 3343 EXPORT_SYMBOL(skb_mac_gso_segment); 3344 3345 3346 /* openvswitch calls this on rx path, so we need a different check. 3347 */ 3348 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path) 3349 { 3350 if (tx_path) 3351 return skb->ip_summed != CHECKSUM_PARTIAL && 3352 skb->ip_summed != CHECKSUM_UNNECESSARY; 3353 3354 return skb->ip_summed == CHECKSUM_NONE; 3355 } 3356 3357 /** 3358 * __skb_gso_segment - Perform segmentation on skb. 3359 * @skb: buffer to segment 3360 * @features: features for the output path (see dev->features) 3361 * @tx_path: whether it is called in TX path 3362 * 3363 * This function segments the given skb and returns a list of segments. 3364 * 3365 * It may return NULL if the skb requires no segmentation. This is 3366 * only possible when GSO is used for verifying header integrity. 3367 * 3368 * Segmentation preserves SKB_GSO_CB_OFFSET bytes of previous skb cb. 3369 */ 3370 struct sk_buff *__skb_gso_segment(struct sk_buff *skb, 3371 netdev_features_t features, bool tx_path) 3372 { 3373 struct sk_buff *segs; 3374 3375 if (unlikely(skb_needs_check(skb, tx_path))) { 3376 int err; 3377 3378 /* We're going to init ->check field in TCP or UDP header */ 3379 err = skb_cow_head(skb, 0); 3380 if (err < 0) 3381 return ERR_PTR(err); 3382 } 3383 3384 /* Only report GSO partial support if it will enable us to 3385 * support segmentation on this frame without needing additional 3386 * work. 3387 */ 3388 if (features & NETIF_F_GSO_PARTIAL) { 3389 netdev_features_t partial_features = NETIF_F_GSO_ROBUST; 3390 struct net_device *dev = skb->dev; 3391 3392 partial_features |= dev->features & dev->gso_partial_features; 3393 if (!skb_gso_ok(skb, features | partial_features)) 3394 features &= ~NETIF_F_GSO_PARTIAL; 3395 } 3396 3397 BUILD_BUG_ON(SKB_GSO_CB_OFFSET + 3398 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb)); 3399 3400 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb); 3401 SKB_GSO_CB(skb)->encap_level = 0; 3402 3403 skb_reset_mac_header(skb); 3404 skb_reset_mac_len(skb); 3405 3406 segs = skb_mac_gso_segment(skb, features); 3407 3408 if (segs != skb && unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs))) 3409 skb_warn_bad_offload(skb); 3410 3411 return segs; 3412 } 3413 EXPORT_SYMBOL(__skb_gso_segment); 3414 3415 /* Take action when hardware reception checksum errors are detected. */ 3416 #ifdef CONFIG_BUG 3417 static void do_netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3418 { 3419 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>"); 3420 skb_dump(KERN_ERR, skb, true); 3421 dump_stack(); 3422 } 3423 3424 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3425 { 3426 DO_ONCE_LITE(do_netdev_rx_csum_fault, dev, skb); 3427 } 3428 EXPORT_SYMBOL(netdev_rx_csum_fault); 3429 #endif 3430 3431 /* XXX: check that highmem exists at all on the given machine. */ 3432 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) 3433 { 3434 #ifdef CONFIG_HIGHMEM 3435 int i; 3436 3437 if (!(dev->features & NETIF_F_HIGHDMA)) { 3438 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 3439 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 3440 3441 if (PageHighMem(skb_frag_page(frag))) 3442 return 1; 3443 } 3444 } 3445 #endif 3446 return 0; 3447 } 3448 3449 /* If MPLS offload request, verify we are testing hardware MPLS features 3450 * instead of standard features for the netdev. 3451 */ 3452 #if IS_ENABLED(CONFIG_NET_MPLS_GSO) 3453 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3454 netdev_features_t features, 3455 __be16 type) 3456 { 3457 if (eth_p_mpls(type)) 3458 features &= skb->dev->mpls_features; 3459 3460 return features; 3461 } 3462 #else 3463 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3464 netdev_features_t features, 3465 __be16 type) 3466 { 3467 return features; 3468 } 3469 #endif 3470 3471 static netdev_features_t harmonize_features(struct sk_buff *skb, 3472 netdev_features_t features) 3473 { 3474 __be16 type; 3475 3476 type = skb_network_protocol(skb, NULL); 3477 features = net_mpls_features(skb, features, type); 3478 3479 if (skb->ip_summed != CHECKSUM_NONE && 3480 !can_checksum_protocol(features, type)) { 3481 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); 3482 } 3483 if (illegal_highdma(skb->dev, skb)) 3484 features &= ~NETIF_F_SG; 3485 3486 return features; 3487 } 3488 3489 netdev_features_t passthru_features_check(struct sk_buff *skb, 3490 struct net_device *dev, 3491 netdev_features_t features) 3492 { 3493 return features; 3494 } 3495 EXPORT_SYMBOL(passthru_features_check); 3496 3497 static netdev_features_t dflt_features_check(struct sk_buff *skb, 3498 struct net_device *dev, 3499 netdev_features_t features) 3500 { 3501 return vlan_features_check(skb, features); 3502 } 3503 3504 static netdev_features_t gso_features_check(const struct sk_buff *skb, 3505 struct net_device *dev, 3506 netdev_features_t features) 3507 { 3508 u16 gso_segs = skb_shinfo(skb)->gso_segs; 3509 3510 if (gso_segs > dev->gso_max_segs) 3511 return features & ~NETIF_F_GSO_MASK; 3512 3513 if (!skb_shinfo(skb)->gso_type) { 3514 skb_warn_bad_offload(skb); 3515 return features & ~NETIF_F_GSO_MASK; 3516 } 3517 3518 /* Support for GSO partial features requires software 3519 * intervention before we can actually process the packets 3520 * so we need to strip support for any partial features now 3521 * and we can pull them back in after we have partially 3522 * segmented the frame. 3523 */ 3524 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)) 3525 features &= ~dev->gso_partial_features; 3526 3527 /* Make sure to clear the IPv4 ID mangling feature if the 3528 * IPv4 header has the potential to be fragmented. 3529 */ 3530 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { 3531 struct iphdr *iph = skb->encapsulation ? 3532 inner_ip_hdr(skb) : ip_hdr(skb); 3533 3534 if (!(iph->frag_off & htons(IP_DF))) 3535 features &= ~NETIF_F_TSO_MANGLEID; 3536 } 3537 3538 return features; 3539 } 3540 3541 netdev_features_t netif_skb_features(struct sk_buff *skb) 3542 { 3543 struct net_device *dev = skb->dev; 3544 netdev_features_t features = dev->features; 3545 3546 if (skb_is_gso(skb)) 3547 features = gso_features_check(skb, dev, features); 3548 3549 /* If encapsulation offload request, verify we are testing 3550 * hardware encapsulation features instead of standard 3551 * features for the netdev 3552 */ 3553 if (skb->encapsulation) 3554 features &= dev->hw_enc_features; 3555 3556 if (skb_vlan_tagged(skb)) 3557 features = netdev_intersect_features(features, 3558 dev->vlan_features | 3559 NETIF_F_HW_VLAN_CTAG_TX | 3560 NETIF_F_HW_VLAN_STAG_TX); 3561 3562 if (dev->netdev_ops->ndo_features_check) 3563 features &= dev->netdev_ops->ndo_features_check(skb, dev, 3564 features); 3565 else 3566 features &= dflt_features_check(skb, dev, features); 3567 3568 return harmonize_features(skb, features); 3569 } 3570 EXPORT_SYMBOL(netif_skb_features); 3571 3572 static int xmit_one(struct sk_buff *skb, struct net_device *dev, 3573 struct netdev_queue *txq, bool more) 3574 { 3575 unsigned int len; 3576 int rc; 3577 3578 if (dev_nit_active(dev)) 3579 dev_queue_xmit_nit(skb, dev); 3580 3581 len = skb->len; 3582 PRANDOM_ADD_NOISE(skb, dev, txq, len + jiffies); 3583 trace_net_dev_start_xmit(skb, dev); 3584 rc = netdev_start_xmit(skb, dev, txq, more); 3585 trace_net_dev_xmit(skb, rc, dev, len); 3586 3587 return rc; 3588 } 3589 3590 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, 3591 struct netdev_queue *txq, int *ret) 3592 { 3593 struct sk_buff *skb = first; 3594 int rc = NETDEV_TX_OK; 3595 3596 while (skb) { 3597 struct sk_buff *next = skb->next; 3598 3599 skb_mark_not_on_list(skb); 3600 rc = xmit_one(skb, dev, txq, next != NULL); 3601 if (unlikely(!dev_xmit_complete(rc))) { 3602 skb->next = next; 3603 goto out; 3604 } 3605 3606 skb = next; 3607 if (netif_tx_queue_stopped(txq) && skb) { 3608 rc = NETDEV_TX_BUSY; 3609 break; 3610 } 3611 } 3612 3613 out: 3614 *ret = rc; 3615 return skb; 3616 } 3617 3618 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, 3619 netdev_features_t features) 3620 { 3621 if (skb_vlan_tag_present(skb) && 3622 !vlan_hw_offload_capable(features, skb->vlan_proto)) 3623 skb = __vlan_hwaccel_push_inside(skb); 3624 return skb; 3625 } 3626 3627 int skb_csum_hwoffload_help(struct sk_buff *skb, 3628 const netdev_features_t features) 3629 { 3630 if (unlikely(skb_csum_is_sctp(skb))) 3631 return !!(features & NETIF_F_SCTP_CRC) ? 0 : 3632 skb_crc32c_csum_help(skb); 3633 3634 if (features & NETIF_F_HW_CSUM) 3635 return 0; 3636 3637 if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) { 3638 switch (skb->csum_offset) { 3639 case offsetof(struct tcphdr, check): 3640 case offsetof(struct udphdr, check): 3641 return 0; 3642 } 3643 } 3644 3645 return skb_checksum_help(skb); 3646 } 3647 EXPORT_SYMBOL(skb_csum_hwoffload_help); 3648 3649 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again) 3650 { 3651 netdev_features_t features; 3652 3653 features = netif_skb_features(skb); 3654 skb = validate_xmit_vlan(skb, features); 3655 if (unlikely(!skb)) 3656 goto out_null; 3657 3658 skb = sk_validate_xmit_skb(skb, dev); 3659 if (unlikely(!skb)) 3660 goto out_null; 3661 3662 if (netif_needs_gso(skb, features)) { 3663 struct sk_buff *segs; 3664 3665 segs = skb_gso_segment(skb, features); 3666 if (IS_ERR(segs)) { 3667 goto out_kfree_skb; 3668 } else if (segs) { 3669 consume_skb(skb); 3670 skb = segs; 3671 } 3672 } else { 3673 if (skb_needs_linearize(skb, features) && 3674 __skb_linearize(skb)) 3675 goto out_kfree_skb; 3676 3677 /* If packet is not checksummed and device does not 3678 * support checksumming for this protocol, complete 3679 * checksumming here. 3680 */ 3681 if (skb->ip_summed == CHECKSUM_PARTIAL) { 3682 if (skb->encapsulation) 3683 skb_set_inner_transport_header(skb, 3684 skb_checksum_start_offset(skb)); 3685 else 3686 skb_set_transport_header(skb, 3687 skb_checksum_start_offset(skb)); 3688 if (skb_csum_hwoffload_help(skb, features)) 3689 goto out_kfree_skb; 3690 } 3691 } 3692 3693 skb = validate_xmit_xfrm(skb, features, again); 3694 3695 return skb; 3696 3697 out_kfree_skb: 3698 kfree_skb(skb); 3699 out_null: 3700 atomic_long_inc(&dev->tx_dropped); 3701 return NULL; 3702 } 3703 3704 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again) 3705 { 3706 struct sk_buff *next, *head = NULL, *tail; 3707 3708 for (; skb != NULL; skb = next) { 3709 next = skb->next; 3710 skb_mark_not_on_list(skb); 3711 3712 /* in case skb wont be segmented, point to itself */ 3713 skb->prev = skb; 3714 3715 skb = validate_xmit_skb(skb, dev, again); 3716 if (!skb) 3717 continue; 3718 3719 if (!head) 3720 head = skb; 3721 else 3722 tail->next = skb; 3723 /* If skb was segmented, skb->prev points to 3724 * the last segment. If not, it still contains skb. 3725 */ 3726 tail = skb->prev; 3727 } 3728 return head; 3729 } 3730 EXPORT_SYMBOL_GPL(validate_xmit_skb_list); 3731 3732 static void qdisc_pkt_len_init(struct sk_buff *skb) 3733 { 3734 const struct skb_shared_info *shinfo = skb_shinfo(skb); 3735 3736 qdisc_skb_cb(skb)->pkt_len = skb->len; 3737 3738 /* To get more precise estimation of bytes sent on wire, 3739 * we add to pkt_len the headers size of all segments 3740 */ 3741 if (shinfo->gso_size && skb_transport_header_was_set(skb)) { 3742 unsigned int hdr_len; 3743 u16 gso_segs = shinfo->gso_segs; 3744 3745 /* mac layer + network layer */ 3746 hdr_len = skb_transport_header(skb) - skb_mac_header(skb); 3747 3748 /* + transport layer */ 3749 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { 3750 const struct tcphdr *th; 3751 struct tcphdr _tcphdr; 3752 3753 th = skb_header_pointer(skb, skb_transport_offset(skb), 3754 sizeof(_tcphdr), &_tcphdr); 3755 if (likely(th)) 3756 hdr_len += __tcp_hdrlen(th); 3757 } else { 3758 struct udphdr _udphdr; 3759 3760 if (skb_header_pointer(skb, skb_transport_offset(skb), 3761 sizeof(_udphdr), &_udphdr)) 3762 hdr_len += sizeof(struct udphdr); 3763 } 3764 3765 if (shinfo->gso_type & SKB_GSO_DODGY) 3766 gso_segs = DIV_ROUND_UP(skb->len - hdr_len, 3767 shinfo->gso_size); 3768 3769 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; 3770 } 3771 } 3772 3773 static int dev_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *q, 3774 struct sk_buff **to_free, 3775 struct netdev_queue *txq) 3776 { 3777 int rc; 3778 3779 rc = q->enqueue(skb, q, to_free) & NET_XMIT_MASK; 3780 if (rc == NET_XMIT_SUCCESS) 3781 trace_qdisc_enqueue(q, txq, skb); 3782 return rc; 3783 } 3784 3785 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, 3786 struct net_device *dev, 3787 struct netdev_queue *txq) 3788 { 3789 spinlock_t *root_lock = qdisc_lock(q); 3790 struct sk_buff *to_free = NULL; 3791 bool contended; 3792 int rc; 3793 3794 qdisc_calculate_pkt_len(skb, q); 3795 3796 if (q->flags & TCQ_F_NOLOCK) { 3797 if (q->flags & TCQ_F_CAN_BYPASS && nolock_qdisc_is_empty(q) && 3798 qdisc_run_begin(q)) { 3799 /* Retest nolock_qdisc_is_empty() within the protection 3800 * of q->seqlock to protect from racing with requeuing. 3801 */ 3802 if (unlikely(!nolock_qdisc_is_empty(q))) { 3803 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3804 __qdisc_run(q); 3805 qdisc_run_end(q); 3806 3807 goto no_lock_out; 3808 } 3809 3810 qdisc_bstats_cpu_update(q, skb); 3811 if (sch_direct_xmit(skb, q, dev, txq, NULL, true) && 3812 !nolock_qdisc_is_empty(q)) 3813 __qdisc_run(q); 3814 3815 qdisc_run_end(q); 3816 return NET_XMIT_SUCCESS; 3817 } 3818 3819 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3820 qdisc_run(q); 3821 3822 no_lock_out: 3823 if (unlikely(to_free)) 3824 kfree_skb_list(to_free); 3825 return rc; 3826 } 3827 3828 /* 3829 * Heuristic to force contended enqueues to serialize on a 3830 * separate lock before trying to get qdisc main lock. 3831 * This permits qdisc->running owner to get the lock more 3832 * often and dequeue packets faster. 3833 */ 3834 contended = qdisc_is_running(q); 3835 if (unlikely(contended)) 3836 spin_lock(&q->busylock); 3837 3838 spin_lock(root_lock); 3839 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { 3840 __qdisc_drop(skb, &to_free); 3841 rc = NET_XMIT_DROP; 3842 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && 3843 qdisc_run_begin(q)) { 3844 /* 3845 * This is a work-conserving queue; there are no old skbs 3846 * waiting to be sent out; and the qdisc is not running - 3847 * xmit the skb directly. 3848 */ 3849 3850 qdisc_bstats_update(q, skb); 3851 3852 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { 3853 if (unlikely(contended)) { 3854 spin_unlock(&q->busylock); 3855 contended = false; 3856 } 3857 __qdisc_run(q); 3858 } 3859 3860 qdisc_run_end(q); 3861 rc = NET_XMIT_SUCCESS; 3862 } else { 3863 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3864 if (qdisc_run_begin(q)) { 3865 if (unlikely(contended)) { 3866 spin_unlock(&q->busylock); 3867 contended = false; 3868 } 3869 __qdisc_run(q); 3870 qdisc_run_end(q); 3871 } 3872 } 3873 spin_unlock(root_lock); 3874 if (unlikely(to_free)) 3875 kfree_skb_list(to_free); 3876 if (unlikely(contended)) 3877 spin_unlock(&q->busylock); 3878 return rc; 3879 } 3880 3881 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) 3882 static void skb_update_prio(struct sk_buff *skb) 3883 { 3884 const struct netprio_map *map; 3885 const struct sock *sk; 3886 unsigned int prioidx; 3887 3888 if (skb->priority) 3889 return; 3890 map = rcu_dereference_bh(skb->dev->priomap); 3891 if (!map) 3892 return; 3893 sk = skb_to_full_sk(skb); 3894 if (!sk) 3895 return; 3896 3897 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data); 3898 3899 if (prioidx < map->priomap_len) 3900 skb->priority = map->priomap[prioidx]; 3901 } 3902 #else 3903 #define skb_update_prio(skb) 3904 #endif 3905 3906 /** 3907 * dev_loopback_xmit - loop back @skb 3908 * @net: network namespace this loopback is happening in 3909 * @sk: sk needed to be a netfilter okfn 3910 * @skb: buffer to transmit 3911 */ 3912 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) 3913 { 3914 skb_reset_mac_header(skb); 3915 __skb_pull(skb, skb_network_offset(skb)); 3916 skb->pkt_type = PACKET_LOOPBACK; 3917 skb->ip_summed = CHECKSUM_UNNECESSARY; 3918 WARN_ON(!skb_dst(skb)); 3919 skb_dst_force(skb); 3920 netif_rx_ni(skb); 3921 return 0; 3922 } 3923 EXPORT_SYMBOL(dev_loopback_xmit); 3924 3925 #ifdef CONFIG_NET_EGRESS 3926 static struct sk_buff * 3927 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 3928 { 3929 struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress); 3930 struct tcf_result cl_res; 3931 3932 if (!miniq) 3933 return skb; 3934 3935 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */ 3936 qdisc_skb_cb(skb)->mru = 0; 3937 qdisc_skb_cb(skb)->post_ct = false; 3938 mini_qdisc_bstats_cpu_update(miniq, skb); 3939 3940 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) { 3941 case TC_ACT_OK: 3942 case TC_ACT_RECLASSIFY: 3943 skb->tc_index = TC_H_MIN(cl_res.classid); 3944 break; 3945 case TC_ACT_SHOT: 3946 mini_qdisc_qstats_cpu_drop(miniq); 3947 *ret = NET_XMIT_DROP; 3948 kfree_skb(skb); 3949 return NULL; 3950 case TC_ACT_STOLEN: 3951 case TC_ACT_QUEUED: 3952 case TC_ACT_TRAP: 3953 *ret = NET_XMIT_SUCCESS; 3954 consume_skb(skb); 3955 return NULL; 3956 case TC_ACT_REDIRECT: 3957 /* No need to push/pop skb's mac_header here on egress! */ 3958 skb_do_redirect(skb); 3959 *ret = NET_XMIT_SUCCESS; 3960 return NULL; 3961 default: 3962 break; 3963 } 3964 3965 return skb; 3966 } 3967 #endif /* CONFIG_NET_EGRESS */ 3968 3969 #ifdef CONFIG_XPS 3970 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb, 3971 struct xps_dev_maps *dev_maps, unsigned int tci) 3972 { 3973 int tc = netdev_get_prio_tc_map(dev, skb->priority); 3974 struct xps_map *map; 3975 int queue_index = -1; 3976 3977 if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids) 3978 return queue_index; 3979 3980 tci *= dev_maps->num_tc; 3981 tci += tc; 3982 3983 map = rcu_dereference(dev_maps->attr_map[tci]); 3984 if (map) { 3985 if (map->len == 1) 3986 queue_index = map->queues[0]; 3987 else 3988 queue_index = map->queues[reciprocal_scale( 3989 skb_get_hash(skb), map->len)]; 3990 if (unlikely(queue_index >= dev->real_num_tx_queues)) 3991 queue_index = -1; 3992 } 3993 return queue_index; 3994 } 3995 #endif 3996 3997 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev, 3998 struct sk_buff *skb) 3999 { 4000 #ifdef CONFIG_XPS 4001 struct xps_dev_maps *dev_maps; 4002 struct sock *sk = skb->sk; 4003 int queue_index = -1; 4004 4005 if (!static_key_false(&xps_needed)) 4006 return -1; 4007 4008 rcu_read_lock(); 4009 if (!static_key_false(&xps_rxqs_needed)) 4010 goto get_cpus_map; 4011 4012 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]); 4013 if (dev_maps) { 4014 int tci = sk_rx_queue_get(sk); 4015 4016 if (tci >= 0) 4017 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4018 tci); 4019 } 4020 4021 get_cpus_map: 4022 if (queue_index < 0) { 4023 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]); 4024 if (dev_maps) { 4025 unsigned int tci = skb->sender_cpu - 1; 4026 4027 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4028 tci); 4029 } 4030 } 4031 rcu_read_unlock(); 4032 4033 return queue_index; 4034 #else 4035 return -1; 4036 #endif 4037 } 4038 4039 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb, 4040 struct net_device *sb_dev) 4041 { 4042 return 0; 4043 } 4044 EXPORT_SYMBOL(dev_pick_tx_zero); 4045 4046 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb, 4047 struct net_device *sb_dev) 4048 { 4049 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues; 4050 } 4051 EXPORT_SYMBOL(dev_pick_tx_cpu_id); 4052 4053 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb, 4054 struct net_device *sb_dev) 4055 { 4056 struct sock *sk = skb->sk; 4057 int queue_index = sk_tx_queue_get(sk); 4058 4059 sb_dev = sb_dev ? : dev; 4060 4061 if (queue_index < 0 || skb->ooo_okay || 4062 queue_index >= dev->real_num_tx_queues) { 4063 int new_index = get_xps_queue(dev, sb_dev, skb); 4064 4065 if (new_index < 0) 4066 new_index = skb_tx_hash(dev, sb_dev, skb); 4067 4068 if (queue_index != new_index && sk && 4069 sk_fullsock(sk) && 4070 rcu_access_pointer(sk->sk_dst_cache)) 4071 sk_tx_queue_set(sk, new_index); 4072 4073 queue_index = new_index; 4074 } 4075 4076 return queue_index; 4077 } 4078 EXPORT_SYMBOL(netdev_pick_tx); 4079 4080 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev, 4081 struct sk_buff *skb, 4082 struct net_device *sb_dev) 4083 { 4084 int queue_index = 0; 4085 4086 #ifdef CONFIG_XPS 4087 u32 sender_cpu = skb->sender_cpu - 1; 4088 4089 if (sender_cpu >= (u32)NR_CPUS) 4090 skb->sender_cpu = raw_smp_processor_id() + 1; 4091 #endif 4092 4093 if (dev->real_num_tx_queues != 1) { 4094 const struct net_device_ops *ops = dev->netdev_ops; 4095 4096 if (ops->ndo_select_queue) 4097 queue_index = ops->ndo_select_queue(dev, skb, sb_dev); 4098 else 4099 queue_index = netdev_pick_tx(dev, skb, sb_dev); 4100 4101 queue_index = netdev_cap_txqueue(dev, queue_index); 4102 } 4103 4104 skb_set_queue_mapping(skb, queue_index); 4105 return netdev_get_tx_queue(dev, queue_index); 4106 } 4107 4108 /** 4109 * __dev_queue_xmit - transmit a buffer 4110 * @skb: buffer to transmit 4111 * @sb_dev: suboordinate device used for L2 forwarding offload 4112 * 4113 * Queue a buffer for transmission to a network device. The caller must 4114 * have set the device and priority and built the buffer before calling 4115 * this function. The function can be called from an interrupt. 4116 * 4117 * A negative errno code is returned on a failure. A success does not 4118 * guarantee the frame will be transmitted as it may be dropped due 4119 * to congestion or traffic shaping. 4120 * 4121 * ----------------------------------------------------------------------------------- 4122 * I notice this method can also return errors from the queue disciplines, 4123 * including NET_XMIT_DROP, which is a positive value. So, errors can also 4124 * be positive. 4125 * 4126 * Regardless of the return value, the skb is consumed, so it is currently 4127 * difficult to retry a send to this method. (You can bump the ref count 4128 * before sending to hold a reference for retry if you are careful.) 4129 * 4130 * When calling this method, interrupts MUST be enabled. This is because 4131 * the BH enable code must have IRQs enabled so that it will not deadlock. 4132 * --BLG 4133 */ 4134 static int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev) 4135 { 4136 struct net_device *dev = skb->dev; 4137 struct netdev_queue *txq; 4138 struct Qdisc *q; 4139 int rc = -ENOMEM; 4140 bool again = false; 4141 4142 skb_reset_mac_header(skb); 4143 4144 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) 4145 __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED); 4146 4147 /* Disable soft irqs for various locks below. Also 4148 * stops preemption for RCU. 4149 */ 4150 rcu_read_lock_bh(); 4151 4152 skb_update_prio(skb); 4153 4154 qdisc_pkt_len_init(skb); 4155 #ifdef CONFIG_NET_CLS_ACT 4156 skb->tc_at_ingress = 0; 4157 # ifdef CONFIG_NET_EGRESS 4158 if (static_branch_unlikely(&egress_needed_key)) { 4159 skb = sch_handle_egress(skb, &rc, dev); 4160 if (!skb) 4161 goto out; 4162 } 4163 # endif 4164 #endif 4165 /* If device/qdisc don't need skb->dst, release it right now while 4166 * its hot in this cpu cache. 4167 */ 4168 if (dev->priv_flags & IFF_XMIT_DST_RELEASE) 4169 skb_dst_drop(skb); 4170 else 4171 skb_dst_force(skb); 4172 4173 txq = netdev_core_pick_tx(dev, skb, sb_dev); 4174 q = rcu_dereference_bh(txq->qdisc); 4175 4176 trace_net_dev_queue(skb); 4177 if (q->enqueue) { 4178 rc = __dev_xmit_skb(skb, q, dev, txq); 4179 goto out; 4180 } 4181 4182 /* The device has no queue. Common case for software devices: 4183 * loopback, all the sorts of tunnels... 4184 4185 * Really, it is unlikely that netif_tx_lock protection is necessary 4186 * here. (f.e. loopback and IP tunnels are clean ignoring statistics 4187 * counters.) 4188 * However, it is possible, that they rely on protection 4189 * made by us here. 4190 4191 * Check this and shot the lock. It is not prone from deadlocks. 4192 *Either shot noqueue qdisc, it is even simpler 8) 4193 */ 4194 if (dev->flags & IFF_UP) { 4195 int cpu = smp_processor_id(); /* ok because BHs are off */ 4196 4197 if (txq->xmit_lock_owner != cpu) { 4198 if (dev_xmit_recursion()) 4199 goto recursion_alert; 4200 4201 skb = validate_xmit_skb(skb, dev, &again); 4202 if (!skb) 4203 goto out; 4204 4205 PRANDOM_ADD_NOISE(skb, dev, txq, jiffies); 4206 HARD_TX_LOCK(dev, txq, cpu); 4207 4208 if (!netif_xmit_stopped(txq)) { 4209 dev_xmit_recursion_inc(); 4210 skb = dev_hard_start_xmit(skb, dev, txq, &rc); 4211 dev_xmit_recursion_dec(); 4212 if (dev_xmit_complete(rc)) { 4213 HARD_TX_UNLOCK(dev, txq); 4214 goto out; 4215 } 4216 } 4217 HARD_TX_UNLOCK(dev, txq); 4218 net_crit_ratelimited("Virtual device %s asks to queue packet!\n", 4219 dev->name); 4220 } else { 4221 /* Recursion is detected! It is possible, 4222 * unfortunately 4223 */ 4224 recursion_alert: 4225 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", 4226 dev->name); 4227 } 4228 } 4229 4230 rc = -ENETDOWN; 4231 rcu_read_unlock_bh(); 4232 4233 atomic_long_inc(&dev->tx_dropped); 4234 kfree_skb_list(skb); 4235 return rc; 4236 out: 4237 rcu_read_unlock_bh(); 4238 return rc; 4239 } 4240 4241 int dev_queue_xmit(struct sk_buff *skb) 4242 { 4243 return __dev_queue_xmit(skb, NULL); 4244 } 4245 EXPORT_SYMBOL(dev_queue_xmit); 4246 4247 int dev_queue_xmit_accel(struct sk_buff *skb, struct net_device *sb_dev) 4248 { 4249 return __dev_queue_xmit(skb, sb_dev); 4250 } 4251 EXPORT_SYMBOL(dev_queue_xmit_accel); 4252 4253 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id) 4254 { 4255 struct net_device *dev = skb->dev; 4256 struct sk_buff *orig_skb = skb; 4257 struct netdev_queue *txq; 4258 int ret = NETDEV_TX_BUSY; 4259 bool again = false; 4260 4261 if (unlikely(!netif_running(dev) || 4262 !netif_carrier_ok(dev))) 4263 goto drop; 4264 4265 skb = validate_xmit_skb_list(skb, dev, &again); 4266 if (skb != orig_skb) 4267 goto drop; 4268 4269 skb_set_queue_mapping(skb, queue_id); 4270 txq = skb_get_tx_queue(dev, skb); 4271 PRANDOM_ADD_NOISE(skb, dev, txq, jiffies); 4272 4273 local_bh_disable(); 4274 4275 dev_xmit_recursion_inc(); 4276 HARD_TX_LOCK(dev, txq, smp_processor_id()); 4277 if (!netif_xmit_frozen_or_drv_stopped(txq)) 4278 ret = netdev_start_xmit(skb, dev, txq, false); 4279 HARD_TX_UNLOCK(dev, txq); 4280 dev_xmit_recursion_dec(); 4281 4282 local_bh_enable(); 4283 return ret; 4284 drop: 4285 atomic_long_inc(&dev->tx_dropped); 4286 kfree_skb_list(skb); 4287 return NET_XMIT_DROP; 4288 } 4289 EXPORT_SYMBOL(__dev_direct_xmit); 4290 4291 /************************************************************************* 4292 * Receiver routines 4293 *************************************************************************/ 4294 4295 int netdev_max_backlog __read_mostly = 1000; 4296 EXPORT_SYMBOL(netdev_max_backlog); 4297 4298 int netdev_tstamp_prequeue __read_mostly = 1; 4299 int netdev_budget __read_mostly = 300; 4300 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */ 4301 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ; 4302 int weight_p __read_mostly = 64; /* old backlog weight */ 4303 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ 4304 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ 4305 int dev_rx_weight __read_mostly = 64; 4306 int dev_tx_weight __read_mostly = 64; 4307 /* Maximum number of GRO_NORMAL skbs to batch up for list-RX */ 4308 int gro_normal_batch __read_mostly = 8; 4309 4310 /* Called with irq disabled */ 4311 static inline void ____napi_schedule(struct softnet_data *sd, 4312 struct napi_struct *napi) 4313 { 4314 struct task_struct *thread; 4315 4316 if (test_bit(NAPI_STATE_THREADED, &napi->state)) { 4317 /* Paired with smp_mb__before_atomic() in 4318 * napi_enable()/dev_set_threaded(). 4319 * Use READ_ONCE() to guarantee a complete 4320 * read on napi->thread. Only call 4321 * wake_up_process() when it's not NULL. 4322 */ 4323 thread = READ_ONCE(napi->thread); 4324 if (thread) { 4325 /* Avoid doing set_bit() if the thread is in 4326 * INTERRUPTIBLE state, cause napi_thread_wait() 4327 * makes sure to proceed with napi polling 4328 * if the thread is explicitly woken from here. 4329 */ 4330 if (READ_ONCE(thread->__state) != TASK_INTERRUPTIBLE) 4331 set_bit(NAPI_STATE_SCHED_THREADED, &napi->state); 4332 wake_up_process(thread); 4333 return; 4334 } 4335 } 4336 4337 list_add_tail(&napi->poll_list, &sd->poll_list); 4338 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4339 } 4340 4341 #ifdef CONFIG_RPS 4342 4343 /* One global table that all flow-based protocols share. */ 4344 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; 4345 EXPORT_SYMBOL(rps_sock_flow_table); 4346 u32 rps_cpu_mask __read_mostly; 4347 EXPORT_SYMBOL(rps_cpu_mask); 4348 4349 struct static_key_false rps_needed __read_mostly; 4350 EXPORT_SYMBOL(rps_needed); 4351 struct static_key_false rfs_needed __read_mostly; 4352 EXPORT_SYMBOL(rfs_needed); 4353 4354 static struct rps_dev_flow * 4355 set_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4356 struct rps_dev_flow *rflow, u16 next_cpu) 4357 { 4358 if (next_cpu < nr_cpu_ids) { 4359 #ifdef CONFIG_RFS_ACCEL 4360 struct netdev_rx_queue *rxqueue; 4361 struct rps_dev_flow_table *flow_table; 4362 struct rps_dev_flow *old_rflow; 4363 u32 flow_id; 4364 u16 rxq_index; 4365 int rc; 4366 4367 /* Should we steer this flow to a different hardware queue? */ 4368 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || 4369 !(dev->features & NETIF_F_NTUPLE)) 4370 goto out; 4371 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); 4372 if (rxq_index == skb_get_rx_queue(skb)) 4373 goto out; 4374 4375 rxqueue = dev->_rx + rxq_index; 4376 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4377 if (!flow_table) 4378 goto out; 4379 flow_id = skb_get_hash(skb) & flow_table->mask; 4380 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, 4381 rxq_index, flow_id); 4382 if (rc < 0) 4383 goto out; 4384 old_rflow = rflow; 4385 rflow = &flow_table->flows[flow_id]; 4386 rflow->filter = rc; 4387 if (old_rflow->filter == rflow->filter) 4388 old_rflow->filter = RPS_NO_FILTER; 4389 out: 4390 #endif 4391 rflow->last_qtail = 4392 per_cpu(softnet_data, next_cpu).input_queue_head; 4393 } 4394 4395 rflow->cpu = next_cpu; 4396 return rflow; 4397 } 4398 4399 /* 4400 * get_rps_cpu is called from netif_receive_skb and returns the target 4401 * CPU from the RPS map of the receiving queue for a given skb. 4402 * rcu_read_lock must be held on entry. 4403 */ 4404 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4405 struct rps_dev_flow **rflowp) 4406 { 4407 const struct rps_sock_flow_table *sock_flow_table; 4408 struct netdev_rx_queue *rxqueue = dev->_rx; 4409 struct rps_dev_flow_table *flow_table; 4410 struct rps_map *map; 4411 int cpu = -1; 4412 u32 tcpu; 4413 u32 hash; 4414 4415 if (skb_rx_queue_recorded(skb)) { 4416 u16 index = skb_get_rx_queue(skb); 4417 4418 if (unlikely(index >= dev->real_num_rx_queues)) { 4419 WARN_ONCE(dev->real_num_rx_queues > 1, 4420 "%s received packet on queue %u, but number " 4421 "of RX queues is %u\n", 4422 dev->name, index, dev->real_num_rx_queues); 4423 goto done; 4424 } 4425 rxqueue += index; 4426 } 4427 4428 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ 4429 4430 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4431 map = rcu_dereference(rxqueue->rps_map); 4432 if (!flow_table && !map) 4433 goto done; 4434 4435 skb_reset_network_header(skb); 4436 hash = skb_get_hash(skb); 4437 if (!hash) 4438 goto done; 4439 4440 sock_flow_table = rcu_dereference(rps_sock_flow_table); 4441 if (flow_table && sock_flow_table) { 4442 struct rps_dev_flow *rflow; 4443 u32 next_cpu; 4444 u32 ident; 4445 4446 /* First check into global flow table if there is a match */ 4447 ident = sock_flow_table->ents[hash & sock_flow_table->mask]; 4448 if ((ident ^ hash) & ~rps_cpu_mask) 4449 goto try_rps; 4450 4451 next_cpu = ident & rps_cpu_mask; 4452 4453 /* OK, now we know there is a match, 4454 * we can look at the local (per receive queue) flow table 4455 */ 4456 rflow = &flow_table->flows[hash & flow_table->mask]; 4457 tcpu = rflow->cpu; 4458 4459 /* 4460 * If the desired CPU (where last recvmsg was done) is 4461 * different from current CPU (one in the rx-queue flow 4462 * table entry), switch if one of the following holds: 4463 * - Current CPU is unset (>= nr_cpu_ids). 4464 * - Current CPU is offline. 4465 * - The current CPU's queue tail has advanced beyond the 4466 * last packet that was enqueued using this table entry. 4467 * This guarantees that all previous packets for the flow 4468 * have been dequeued, thus preserving in order delivery. 4469 */ 4470 if (unlikely(tcpu != next_cpu) && 4471 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || 4472 ((int)(per_cpu(softnet_data, tcpu).input_queue_head - 4473 rflow->last_qtail)) >= 0)) { 4474 tcpu = next_cpu; 4475 rflow = set_rps_cpu(dev, skb, rflow, next_cpu); 4476 } 4477 4478 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { 4479 *rflowp = rflow; 4480 cpu = tcpu; 4481 goto done; 4482 } 4483 } 4484 4485 try_rps: 4486 4487 if (map) { 4488 tcpu = map->cpus[reciprocal_scale(hash, map->len)]; 4489 if (cpu_online(tcpu)) { 4490 cpu = tcpu; 4491 goto done; 4492 } 4493 } 4494 4495 done: 4496 return cpu; 4497 } 4498 4499 #ifdef CONFIG_RFS_ACCEL 4500 4501 /** 4502 * rps_may_expire_flow - check whether an RFS hardware filter may be removed 4503 * @dev: Device on which the filter was set 4504 * @rxq_index: RX queue index 4505 * @flow_id: Flow ID passed to ndo_rx_flow_steer() 4506 * @filter_id: Filter ID returned by ndo_rx_flow_steer() 4507 * 4508 * Drivers that implement ndo_rx_flow_steer() should periodically call 4509 * this function for each installed filter and remove the filters for 4510 * which it returns %true. 4511 */ 4512 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, 4513 u32 flow_id, u16 filter_id) 4514 { 4515 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; 4516 struct rps_dev_flow_table *flow_table; 4517 struct rps_dev_flow *rflow; 4518 bool expire = true; 4519 unsigned int cpu; 4520 4521 rcu_read_lock(); 4522 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4523 if (flow_table && flow_id <= flow_table->mask) { 4524 rflow = &flow_table->flows[flow_id]; 4525 cpu = READ_ONCE(rflow->cpu); 4526 if (rflow->filter == filter_id && cpu < nr_cpu_ids && 4527 ((int)(per_cpu(softnet_data, cpu).input_queue_head - 4528 rflow->last_qtail) < 4529 (int)(10 * flow_table->mask))) 4530 expire = false; 4531 } 4532 rcu_read_unlock(); 4533 return expire; 4534 } 4535 EXPORT_SYMBOL(rps_may_expire_flow); 4536 4537 #endif /* CONFIG_RFS_ACCEL */ 4538 4539 /* Called from hardirq (IPI) context */ 4540 static void rps_trigger_softirq(void *data) 4541 { 4542 struct softnet_data *sd = data; 4543 4544 ____napi_schedule(sd, &sd->backlog); 4545 sd->received_rps++; 4546 } 4547 4548 #endif /* CONFIG_RPS */ 4549 4550 /* 4551 * Check if this softnet_data structure is another cpu one 4552 * If yes, queue it to our IPI list and return 1 4553 * If no, return 0 4554 */ 4555 static int rps_ipi_queued(struct softnet_data *sd) 4556 { 4557 #ifdef CONFIG_RPS 4558 struct softnet_data *mysd = this_cpu_ptr(&softnet_data); 4559 4560 if (sd != mysd) { 4561 sd->rps_ipi_next = mysd->rps_ipi_list; 4562 mysd->rps_ipi_list = sd; 4563 4564 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4565 return 1; 4566 } 4567 #endif /* CONFIG_RPS */ 4568 return 0; 4569 } 4570 4571 #ifdef CONFIG_NET_FLOW_LIMIT 4572 int netdev_flow_limit_table_len __read_mostly = (1 << 12); 4573 #endif 4574 4575 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) 4576 { 4577 #ifdef CONFIG_NET_FLOW_LIMIT 4578 struct sd_flow_limit *fl; 4579 struct softnet_data *sd; 4580 unsigned int old_flow, new_flow; 4581 4582 if (qlen < (netdev_max_backlog >> 1)) 4583 return false; 4584 4585 sd = this_cpu_ptr(&softnet_data); 4586 4587 rcu_read_lock(); 4588 fl = rcu_dereference(sd->flow_limit); 4589 if (fl) { 4590 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); 4591 old_flow = fl->history[fl->history_head]; 4592 fl->history[fl->history_head] = new_flow; 4593 4594 fl->history_head++; 4595 fl->history_head &= FLOW_LIMIT_HISTORY - 1; 4596 4597 if (likely(fl->buckets[old_flow])) 4598 fl->buckets[old_flow]--; 4599 4600 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { 4601 fl->count++; 4602 rcu_read_unlock(); 4603 return true; 4604 } 4605 } 4606 rcu_read_unlock(); 4607 #endif 4608 return false; 4609 } 4610 4611 /* 4612 * enqueue_to_backlog is called to queue an skb to a per CPU backlog 4613 * queue (may be a remote CPU queue). 4614 */ 4615 static int enqueue_to_backlog(struct sk_buff *skb, int cpu, 4616 unsigned int *qtail) 4617 { 4618 struct softnet_data *sd; 4619 unsigned long flags; 4620 unsigned int qlen; 4621 4622 sd = &per_cpu(softnet_data, cpu); 4623 4624 local_irq_save(flags); 4625 4626 rps_lock(sd); 4627 if (!netif_running(skb->dev)) 4628 goto drop; 4629 qlen = skb_queue_len(&sd->input_pkt_queue); 4630 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) { 4631 if (qlen) { 4632 enqueue: 4633 __skb_queue_tail(&sd->input_pkt_queue, skb); 4634 input_queue_tail_incr_save(sd, qtail); 4635 rps_unlock(sd); 4636 local_irq_restore(flags); 4637 return NET_RX_SUCCESS; 4638 } 4639 4640 /* Schedule NAPI for backlog device 4641 * We can use non atomic operation since we own the queue lock 4642 */ 4643 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) { 4644 if (!rps_ipi_queued(sd)) 4645 ____napi_schedule(sd, &sd->backlog); 4646 } 4647 goto enqueue; 4648 } 4649 4650 drop: 4651 sd->dropped++; 4652 rps_unlock(sd); 4653 4654 local_irq_restore(flags); 4655 4656 atomic_long_inc(&skb->dev->rx_dropped); 4657 kfree_skb(skb); 4658 return NET_RX_DROP; 4659 } 4660 4661 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb) 4662 { 4663 struct net_device *dev = skb->dev; 4664 struct netdev_rx_queue *rxqueue; 4665 4666 rxqueue = dev->_rx; 4667 4668 if (skb_rx_queue_recorded(skb)) { 4669 u16 index = skb_get_rx_queue(skb); 4670 4671 if (unlikely(index >= dev->real_num_rx_queues)) { 4672 WARN_ONCE(dev->real_num_rx_queues > 1, 4673 "%s received packet on queue %u, but number " 4674 "of RX queues is %u\n", 4675 dev->name, index, dev->real_num_rx_queues); 4676 4677 return rxqueue; /* Return first rxqueue */ 4678 } 4679 rxqueue += index; 4680 } 4681 return rxqueue; 4682 } 4683 4684 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp, 4685 struct bpf_prog *xdp_prog) 4686 { 4687 void *orig_data, *orig_data_end, *hard_start; 4688 struct netdev_rx_queue *rxqueue; 4689 bool orig_bcast, orig_host; 4690 u32 mac_len, frame_sz; 4691 __be16 orig_eth_type; 4692 struct ethhdr *eth; 4693 u32 metalen, act; 4694 int off; 4695 4696 /* The XDP program wants to see the packet starting at the MAC 4697 * header. 4698 */ 4699 mac_len = skb->data - skb_mac_header(skb); 4700 hard_start = skb->data - skb_headroom(skb); 4701 4702 /* SKB "head" area always have tailroom for skb_shared_info */ 4703 frame_sz = (void *)skb_end_pointer(skb) - hard_start; 4704 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 4705 4706 rxqueue = netif_get_rxqueue(skb); 4707 xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq); 4708 xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len, 4709 skb_headlen(skb) + mac_len, true); 4710 4711 orig_data_end = xdp->data_end; 4712 orig_data = xdp->data; 4713 eth = (struct ethhdr *)xdp->data; 4714 orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr); 4715 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest); 4716 orig_eth_type = eth->h_proto; 4717 4718 act = bpf_prog_run_xdp(xdp_prog, xdp); 4719 4720 /* check if bpf_xdp_adjust_head was used */ 4721 off = xdp->data - orig_data; 4722 if (off) { 4723 if (off > 0) 4724 __skb_pull(skb, off); 4725 else if (off < 0) 4726 __skb_push(skb, -off); 4727 4728 skb->mac_header += off; 4729 skb_reset_network_header(skb); 4730 } 4731 4732 /* check if bpf_xdp_adjust_tail was used */ 4733 off = xdp->data_end - orig_data_end; 4734 if (off != 0) { 4735 skb_set_tail_pointer(skb, xdp->data_end - xdp->data); 4736 skb->len += off; /* positive on grow, negative on shrink */ 4737 } 4738 4739 /* check if XDP changed eth hdr such SKB needs update */ 4740 eth = (struct ethhdr *)xdp->data; 4741 if ((orig_eth_type != eth->h_proto) || 4742 (orig_host != ether_addr_equal_64bits(eth->h_dest, 4743 skb->dev->dev_addr)) || 4744 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) { 4745 __skb_push(skb, ETH_HLEN); 4746 skb->pkt_type = PACKET_HOST; 4747 skb->protocol = eth_type_trans(skb, skb->dev); 4748 } 4749 4750 /* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull 4751 * before calling us again on redirect path. We do not call do_redirect 4752 * as we leave that up to the caller. 4753 * 4754 * Caller is responsible for managing lifetime of skb (i.e. calling 4755 * kfree_skb in response to actions it cannot handle/XDP_DROP). 4756 */ 4757 switch (act) { 4758 case XDP_REDIRECT: 4759 case XDP_TX: 4760 __skb_push(skb, mac_len); 4761 break; 4762 case XDP_PASS: 4763 metalen = xdp->data - xdp->data_meta; 4764 if (metalen) 4765 skb_metadata_set(skb, metalen); 4766 break; 4767 } 4768 4769 return act; 4770 } 4771 4772 static u32 netif_receive_generic_xdp(struct sk_buff *skb, 4773 struct xdp_buff *xdp, 4774 struct bpf_prog *xdp_prog) 4775 { 4776 u32 act = XDP_DROP; 4777 4778 /* Reinjected packets coming from act_mirred or similar should 4779 * not get XDP generic processing. 4780 */ 4781 if (skb_is_redirected(skb)) 4782 return XDP_PASS; 4783 4784 /* XDP packets must be linear and must have sufficient headroom 4785 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also 4786 * native XDP provides, thus we need to do it here as well. 4787 */ 4788 if (skb_cloned(skb) || skb_is_nonlinear(skb) || 4789 skb_headroom(skb) < XDP_PACKET_HEADROOM) { 4790 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb); 4791 int troom = skb->tail + skb->data_len - skb->end; 4792 4793 /* In case we have to go down the path and also linearize, 4794 * then lets do the pskb_expand_head() work just once here. 4795 */ 4796 if (pskb_expand_head(skb, 4797 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0, 4798 troom > 0 ? troom + 128 : 0, GFP_ATOMIC)) 4799 goto do_drop; 4800 if (skb_linearize(skb)) 4801 goto do_drop; 4802 } 4803 4804 act = bpf_prog_run_generic_xdp(skb, xdp, xdp_prog); 4805 switch (act) { 4806 case XDP_REDIRECT: 4807 case XDP_TX: 4808 case XDP_PASS: 4809 break; 4810 default: 4811 bpf_warn_invalid_xdp_action(act); 4812 fallthrough; 4813 case XDP_ABORTED: 4814 trace_xdp_exception(skb->dev, xdp_prog, act); 4815 fallthrough; 4816 case XDP_DROP: 4817 do_drop: 4818 kfree_skb(skb); 4819 break; 4820 } 4821 4822 return act; 4823 } 4824 4825 /* When doing generic XDP we have to bypass the qdisc layer and the 4826 * network taps in order to match in-driver-XDP behavior. 4827 */ 4828 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog) 4829 { 4830 struct net_device *dev = skb->dev; 4831 struct netdev_queue *txq; 4832 bool free_skb = true; 4833 int cpu, rc; 4834 4835 txq = netdev_core_pick_tx(dev, skb, NULL); 4836 cpu = smp_processor_id(); 4837 HARD_TX_LOCK(dev, txq, cpu); 4838 if (!netif_xmit_stopped(txq)) { 4839 rc = netdev_start_xmit(skb, dev, txq, 0); 4840 if (dev_xmit_complete(rc)) 4841 free_skb = false; 4842 } 4843 HARD_TX_UNLOCK(dev, txq); 4844 if (free_skb) { 4845 trace_xdp_exception(dev, xdp_prog, XDP_TX); 4846 kfree_skb(skb); 4847 } 4848 } 4849 4850 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key); 4851 4852 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb) 4853 { 4854 if (xdp_prog) { 4855 struct xdp_buff xdp; 4856 u32 act; 4857 int err; 4858 4859 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog); 4860 if (act != XDP_PASS) { 4861 switch (act) { 4862 case XDP_REDIRECT: 4863 err = xdp_do_generic_redirect(skb->dev, skb, 4864 &xdp, xdp_prog); 4865 if (err) 4866 goto out_redir; 4867 break; 4868 case XDP_TX: 4869 generic_xdp_tx(skb, xdp_prog); 4870 break; 4871 } 4872 return XDP_DROP; 4873 } 4874 } 4875 return XDP_PASS; 4876 out_redir: 4877 kfree_skb(skb); 4878 return XDP_DROP; 4879 } 4880 EXPORT_SYMBOL_GPL(do_xdp_generic); 4881 4882 static int netif_rx_internal(struct sk_buff *skb) 4883 { 4884 int ret; 4885 4886 net_timestamp_check(netdev_tstamp_prequeue, skb); 4887 4888 trace_netif_rx(skb); 4889 4890 #ifdef CONFIG_RPS 4891 if (static_branch_unlikely(&rps_needed)) { 4892 struct rps_dev_flow voidflow, *rflow = &voidflow; 4893 int cpu; 4894 4895 preempt_disable(); 4896 rcu_read_lock(); 4897 4898 cpu = get_rps_cpu(skb->dev, skb, &rflow); 4899 if (cpu < 0) 4900 cpu = smp_processor_id(); 4901 4902 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 4903 4904 rcu_read_unlock(); 4905 preempt_enable(); 4906 } else 4907 #endif 4908 { 4909 unsigned int qtail; 4910 4911 ret = enqueue_to_backlog(skb, get_cpu(), &qtail); 4912 put_cpu(); 4913 } 4914 return ret; 4915 } 4916 4917 /** 4918 * netif_rx - post buffer to the network code 4919 * @skb: buffer to post 4920 * 4921 * This function receives a packet from a device driver and queues it for 4922 * the upper (protocol) levels to process. It always succeeds. The buffer 4923 * may be dropped during processing for congestion control or by the 4924 * protocol layers. 4925 * 4926 * return values: 4927 * NET_RX_SUCCESS (no congestion) 4928 * NET_RX_DROP (packet was dropped) 4929 * 4930 */ 4931 4932 int netif_rx(struct sk_buff *skb) 4933 { 4934 int ret; 4935 4936 trace_netif_rx_entry(skb); 4937 4938 ret = netif_rx_internal(skb); 4939 trace_netif_rx_exit(ret); 4940 4941 return ret; 4942 } 4943 EXPORT_SYMBOL(netif_rx); 4944 4945 int netif_rx_ni(struct sk_buff *skb) 4946 { 4947 int err; 4948 4949 trace_netif_rx_ni_entry(skb); 4950 4951 preempt_disable(); 4952 err = netif_rx_internal(skb); 4953 if (local_softirq_pending()) 4954 do_softirq(); 4955 preempt_enable(); 4956 trace_netif_rx_ni_exit(err); 4957 4958 return err; 4959 } 4960 EXPORT_SYMBOL(netif_rx_ni); 4961 4962 int netif_rx_any_context(struct sk_buff *skb) 4963 { 4964 /* 4965 * If invoked from contexts which do not invoke bottom half 4966 * processing either at return from interrupt or when softrqs are 4967 * reenabled, use netif_rx_ni() which invokes bottomhalf processing 4968 * directly. 4969 */ 4970 if (in_interrupt()) 4971 return netif_rx(skb); 4972 else 4973 return netif_rx_ni(skb); 4974 } 4975 EXPORT_SYMBOL(netif_rx_any_context); 4976 4977 static __latent_entropy void net_tx_action(struct softirq_action *h) 4978 { 4979 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 4980 4981 if (sd->completion_queue) { 4982 struct sk_buff *clist; 4983 4984 local_irq_disable(); 4985 clist = sd->completion_queue; 4986 sd->completion_queue = NULL; 4987 local_irq_enable(); 4988 4989 while (clist) { 4990 struct sk_buff *skb = clist; 4991 4992 clist = clist->next; 4993 4994 WARN_ON(refcount_read(&skb->users)); 4995 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED)) 4996 trace_consume_skb(skb); 4997 else 4998 trace_kfree_skb(skb, net_tx_action); 4999 5000 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) 5001 __kfree_skb(skb); 5002 else 5003 __kfree_skb_defer(skb); 5004 } 5005 } 5006 5007 if (sd->output_queue) { 5008 struct Qdisc *head; 5009 5010 local_irq_disable(); 5011 head = sd->output_queue; 5012 sd->output_queue = NULL; 5013 sd->output_queue_tailp = &sd->output_queue; 5014 local_irq_enable(); 5015 5016 rcu_read_lock(); 5017 5018 while (head) { 5019 struct Qdisc *q = head; 5020 spinlock_t *root_lock = NULL; 5021 5022 head = head->next_sched; 5023 5024 /* We need to make sure head->next_sched is read 5025 * before clearing __QDISC_STATE_SCHED 5026 */ 5027 smp_mb__before_atomic(); 5028 5029 if (!(q->flags & TCQ_F_NOLOCK)) { 5030 root_lock = qdisc_lock(q); 5031 spin_lock(root_lock); 5032 } else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, 5033 &q->state))) { 5034 /* There is a synchronize_net() between 5035 * STATE_DEACTIVATED flag being set and 5036 * qdisc_reset()/some_qdisc_is_busy() in 5037 * dev_deactivate(), so we can safely bail out 5038 * early here to avoid data race between 5039 * qdisc_deactivate() and some_qdisc_is_busy() 5040 * for lockless qdisc. 5041 */ 5042 clear_bit(__QDISC_STATE_SCHED, &q->state); 5043 continue; 5044 } 5045 5046 clear_bit(__QDISC_STATE_SCHED, &q->state); 5047 qdisc_run(q); 5048 if (root_lock) 5049 spin_unlock(root_lock); 5050 } 5051 5052 rcu_read_unlock(); 5053 } 5054 5055 xfrm_dev_backlog(sd); 5056 } 5057 5058 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) 5059 /* This hook is defined here for ATM LANE */ 5060 int (*br_fdb_test_addr_hook)(struct net_device *dev, 5061 unsigned char *addr) __read_mostly; 5062 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); 5063 #endif 5064 5065 static inline struct sk_buff * 5066 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 5067 struct net_device *orig_dev, bool *another) 5068 { 5069 #ifdef CONFIG_NET_CLS_ACT 5070 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress); 5071 struct tcf_result cl_res; 5072 5073 /* If there's at least one ingress present somewhere (so 5074 * we get here via enabled static key), remaining devices 5075 * that are not configured with an ingress qdisc will bail 5076 * out here. 5077 */ 5078 if (!miniq) 5079 return skb; 5080 5081 if (*pt_prev) { 5082 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5083 *pt_prev = NULL; 5084 } 5085 5086 qdisc_skb_cb(skb)->pkt_len = skb->len; 5087 qdisc_skb_cb(skb)->mru = 0; 5088 qdisc_skb_cb(skb)->post_ct = false; 5089 skb->tc_at_ingress = 1; 5090 mini_qdisc_bstats_cpu_update(miniq, skb); 5091 5092 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) { 5093 case TC_ACT_OK: 5094 case TC_ACT_RECLASSIFY: 5095 skb->tc_index = TC_H_MIN(cl_res.classid); 5096 break; 5097 case TC_ACT_SHOT: 5098 mini_qdisc_qstats_cpu_drop(miniq); 5099 kfree_skb(skb); 5100 return NULL; 5101 case TC_ACT_STOLEN: 5102 case TC_ACT_QUEUED: 5103 case TC_ACT_TRAP: 5104 consume_skb(skb); 5105 return NULL; 5106 case TC_ACT_REDIRECT: 5107 /* skb_mac_header check was done by cls/act_bpf, so 5108 * we can safely push the L2 header back before 5109 * redirecting to another netdev 5110 */ 5111 __skb_push(skb, skb->mac_len); 5112 if (skb_do_redirect(skb) == -EAGAIN) { 5113 __skb_pull(skb, skb->mac_len); 5114 *another = true; 5115 break; 5116 } 5117 return NULL; 5118 case TC_ACT_CONSUMED: 5119 return NULL; 5120 default: 5121 break; 5122 } 5123 #endif /* CONFIG_NET_CLS_ACT */ 5124 return skb; 5125 } 5126 5127 /** 5128 * netdev_is_rx_handler_busy - check if receive handler is registered 5129 * @dev: device to check 5130 * 5131 * Check if a receive handler is already registered for a given device. 5132 * Return true if there one. 5133 * 5134 * The caller must hold the rtnl_mutex. 5135 */ 5136 bool netdev_is_rx_handler_busy(struct net_device *dev) 5137 { 5138 ASSERT_RTNL(); 5139 return dev && rtnl_dereference(dev->rx_handler); 5140 } 5141 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); 5142 5143 /** 5144 * netdev_rx_handler_register - register receive handler 5145 * @dev: device to register a handler for 5146 * @rx_handler: receive handler to register 5147 * @rx_handler_data: data pointer that is used by rx handler 5148 * 5149 * Register a receive handler for a device. This handler will then be 5150 * called from __netif_receive_skb. A negative errno code is returned 5151 * on a failure. 5152 * 5153 * The caller must hold the rtnl_mutex. 5154 * 5155 * For a general description of rx_handler, see enum rx_handler_result. 5156 */ 5157 int netdev_rx_handler_register(struct net_device *dev, 5158 rx_handler_func_t *rx_handler, 5159 void *rx_handler_data) 5160 { 5161 if (netdev_is_rx_handler_busy(dev)) 5162 return -EBUSY; 5163 5164 if (dev->priv_flags & IFF_NO_RX_HANDLER) 5165 return -EINVAL; 5166 5167 /* Note: rx_handler_data must be set before rx_handler */ 5168 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); 5169 rcu_assign_pointer(dev->rx_handler, rx_handler); 5170 5171 return 0; 5172 } 5173 EXPORT_SYMBOL_GPL(netdev_rx_handler_register); 5174 5175 /** 5176 * netdev_rx_handler_unregister - unregister receive handler 5177 * @dev: device to unregister a handler from 5178 * 5179 * Unregister a receive handler from a device. 5180 * 5181 * The caller must hold the rtnl_mutex. 5182 */ 5183 void netdev_rx_handler_unregister(struct net_device *dev) 5184 { 5185 5186 ASSERT_RTNL(); 5187 RCU_INIT_POINTER(dev->rx_handler, NULL); 5188 /* a reader seeing a non NULL rx_handler in a rcu_read_lock() 5189 * section has a guarantee to see a non NULL rx_handler_data 5190 * as well. 5191 */ 5192 synchronize_net(); 5193 RCU_INIT_POINTER(dev->rx_handler_data, NULL); 5194 } 5195 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); 5196 5197 /* 5198 * Limit the use of PFMEMALLOC reserves to those protocols that implement 5199 * the special handling of PFMEMALLOC skbs. 5200 */ 5201 static bool skb_pfmemalloc_protocol(struct sk_buff *skb) 5202 { 5203 switch (skb->protocol) { 5204 case htons(ETH_P_ARP): 5205 case htons(ETH_P_IP): 5206 case htons(ETH_P_IPV6): 5207 case htons(ETH_P_8021Q): 5208 case htons(ETH_P_8021AD): 5209 return true; 5210 default: 5211 return false; 5212 } 5213 } 5214 5215 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, 5216 int *ret, struct net_device *orig_dev) 5217 { 5218 if (nf_hook_ingress_active(skb)) { 5219 int ingress_retval; 5220 5221 if (*pt_prev) { 5222 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5223 *pt_prev = NULL; 5224 } 5225 5226 rcu_read_lock(); 5227 ingress_retval = nf_hook_ingress(skb); 5228 rcu_read_unlock(); 5229 return ingress_retval; 5230 } 5231 return 0; 5232 } 5233 5234 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc, 5235 struct packet_type **ppt_prev) 5236 { 5237 struct packet_type *ptype, *pt_prev; 5238 rx_handler_func_t *rx_handler; 5239 struct sk_buff *skb = *pskb; 5240 struct net_device *orig_dev; 5241 bool deliver_exact = false; 5242 int ret = NET_RX_DROP; 5243 __be16 type; 5244 5245 net_timestamp_check(!netdev_tstamp_prequeue, skb); 5246 5247 trace_netif_receive_skb(skb); 5248 5249 orig_dev = skb->dev; 5250 5251 skb_reset_network_header(skb); 5252 if (!skb_transport_header_was_set(skb)) 5253 skb_reset_transport_header(skb); 5254 skb_reset_mac_len(skb); 5255 5256 pt_prev = NULL; 5257 5258 another_round: 5259 skb->skb_iif = skb->dev->ifindex; 5260 5261 __this_cpu_inc(softnet_data.processed); 5262 5263 if (static_branch_unlikely(&generic_xdp_needed_key)) { 5264 int ret2; 5265 5266 migrate_disable(); 5267 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb); 5268 migrate_enable(); 5269 5270 if (ret2 != XDP_PASS) { 5271 ret = NET_RX_DROP; 5272 goto out; 5273 } 5274 } 5275 5276 if (eth_type_vlan(skb->protocol)) { 5277 skb = skb_vlan_untag(skb); 5278 if (unlikely(!skb)) 5279 goto out; 5280 } 5281 5282 if (skb_skip_tc_classify(skb)) 5283 goto skip_classify; 5284 5285 if (pfmemalloc) 5286 goto skip_taps; 5287 5288 list_for_each_entry_rcu(ptype, &ptype_all, list) { 5289 if (pt_prev) 5290 ret = deliver_skb(skb, pt_prev, orig_dev); 5291 pt_prev = ptype; 5292 } 5293 5294 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { 5295 if (pt_prev) 5296 ret = deliver_skb(skb, pt_prev, orig_dev); 5297 pt_prev = ptype; 5298 } 5299 5300 skip_taps: 5301 #ifdef CONFIG_NET_INGRESS 5302 if (static_branch_unlikely(&ingress_needed_key)) { 5303 bool another = false; 5304 5305 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev, 5306 &another); 5307 if (another) 5308 goto another_round; 5309 if (!skb) 5310 goto out; 5311 5312 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) 5313 goto out; 5314 } 5315 #endif 5316 skb_reset_redirect(skb); 5317 skip_classify: 5318 if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) 5319 goto drop; 5320 5321 if (skb_vlan_tag_present(skb)) { 5322 if (pt_prev) { 5323 ret = deliver_skb(skb, pt_prev, orig_dev); 5324 pt_prev = NULL; 5325 } 5326 if (vlan_do_receive(&skb)) 5327 goto another_round; 5328 else if (unlikely(!skb)) 5329 goto out; 5330 } 5331 5332 rx_handler = rcu_dereference(skb->dev->rx_handler); 5333 if (rx_handler) { 5334 if (pt_prev) { 5335 ret = deliver_skb(skb, pt_prev, orig_dev); 5336 pt_prev = NULL; 5337 } 5338 switch (rx_handler(&skb)) { 5339 case RX_HANDLER_CONSUMED: 5340 ret = NET_RX_SUCCESS; 5341 goto out; 5342 case RX_HANDLER_ANOTHER: 5343 goto another_round; 5344 case RX_HANDLER_EXACT: 5345 deliver_exact = true; 5346 break; 5347 case RX_HANDLER_PASS: 5348 break; 5349 default: 5350 BUG(); 5351 } 5352 } 5353 5354 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) { 5355 check_vlan_id: 5356 if (skb_vlan_tag_get_id(skb)) { 5357 /* Vlan id is non 0 and vlan_do_receive() above couldn't 5358 * find vlan device. 5359 */ 5360 skb->pkt_type = PACKET_OTHERHOST; 5361 } else if (eth_type_vlan(skb->protocol)) { 5362 /* Outer header is 802.1P with vlan 0, inner header is 5363 * 802.1Q or 802.1AD and vlan_do_receive() above could 5364 * not find vlan dev for vlan id 0. 5365 */ 5366 __vlan_hwaccel_clear_tag(skb); 5367 skb = skb_vlan_untag(skb); 5368 if (unlikely(!skb)) 5369 goto out; 5370 if (vlan_do_receive(&skb)) 5371 /* After stripping off 802.1P header with vlan 0 5372 * vlan dev is found for inner header. 5373 */ 5374 goto another_round; 5375 else if (unlikely(!skb)) 5376 goto out; 5377 else 5378 /* We have stripped outer 802.1P vlan 0 header. 5379 * But could not find vlan dev. 5380 * check again for vlan id to set OTHERHOST. 5381 */ 5382 goto check_vlan_id; 5383 } 5384 /* Note: we might in the future use prio bits 5385 * and set skb->priority like in vlan_do_receive() 5386 * For the time being, just ignore Priority Code Point 5387 */ 5388 __vlan_hwaccel_clear_tag(skb); 5389 } 5390 5391 type = skb->protocol; 5392 5393 /* deliver only exact match when indicated */ 5394 if (likely(!deliver_exact)) { 5395 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5396 &ptype_base[ntohs(type) & 5397 PTYPE_HASH_MASK]); 5398 } 5399 5400 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5401 &orig_dev->ptype_specific); 5402 5403 if (unlikely(skb->dev != orig_dev)) { 5404 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5405 &skb->dev->ptype_specific); 5406 } 5407 5408 if (pt_prev) { 5409 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 5410 goto drop; 5411 *ppt_prev = pt_prev; 5412 } else { 5413 drop: 5414 if (!deliver_exact) 5415 atomic_long_inc(&skb->dev->rx_dropped); 5416 else 5417 atomic_long_inc(&skb->dev->rx_nohandler); 5418 kfree_skb(skb); 5419 /* Jamal, now you will not able to escape explaining 5420 * me how you were going to use this. :-) 5421 */ 5422 ret = NET_RX_DROP; 5423 } 5424 5425 out: 5426 /* The invariant here is that if *ppt_prev is not NULL 5427 * then skb should also be non-NULL. 5428 * 5429 * Apparently *ppt_prev assignment above holds this invariant due to 5430 * skb dereferencing near it. 5431 */ 5432 *pskb = skb; 5433 return ret; 5434 } 5435 5436 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc) 5437 { 5438 struct net_device *orig_dev = skb->dev; 5439 struct packet_type *pt_prev = NULL; 5440 int ret; 5441 5442 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5443 if (pt_prev) 5444 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb, 5445 skb->dev, pt_prev, orig_dev); 5446 return ret; 5447 } 5448 5449 /** 5450 * netif_receive_skb_core - special purpose version of netif_receive_skb 5451 * @skb: buffer to process 5452 * 5453 * More direct receive version of netif_receive_skb(). It should 5454 * only be used by callers that have a need to skip RPS and Generic XDP. 5455 * Caller must also take care of handling if ``(page_is_)pfmemalloc``. 5456 * 5457 * This function may only be called from softirq context and interrupts 5458 * should be enabled. 5459 * 5460 * Return values (usually ignored): 5461 * NET_RX_SUCCESS: no congestion 5462 * NET_RX_DROP: packet was dropped 5463 */ 5464 int netif_receive_skb_core(struct sk_buff *skb) 5465 { 5466 int ret; 5467 5468 rcu_read_lock(); 5469 ret = __netif_receive_skb_one_core(skb, false); 5470 rcu_read_unlock(); 5471 5472 return ret; 5473 } 5474 EXPORT_SYMBOL(netif_receive_skb_core); 5475 5476 static inline void __netif_receive_skb_list_ptype(struct list_head *head, 5477 struct packet_type *pt_prev, 5478 struct net_device *orig_dev) 5479 { 5480 struct sk_buff *skb, *next; 5481 5482 if (!pt_prev) 5483 return; 5484 if (list_empty(head)) 5485 return; 5486 if (pt_prev->list_func != NULL) 5487 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv, 5488 ip_list_rcv, head, pt_prev, orig_dev); 5489 else 5490 list_for_each_entry_safe(skb, next, head, list) { 5491 skb_list_del_init(skb); 5492 pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 5493 } 5494 } 5495 5496 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc) 5497 { 5498 /* Fast-path assumptions: 5499 * - There is no RX handler. 5500 * - Only one packet_type matches. 5501 * If either of these fails, we will end up doing some per-packet 5502 * processing in-line, then handling the 'last ptype' for the whole 5503 * sublist. This can't cause out-of-order delivery to any single ptype, 5504 * because the 'last ptype' must be constant across the sublist, and all 5505 * other ptypes are handled per-packet. 5506 */ 5507 /* Current (common) ptype of sublist */ 5508 struct packet_type *pt_curr = NULL; 5509 /* Current (common) orig_dev of sublist */ 5510 struct net_device *od_curr = NULL; 5511 struct list_head sublist; 5512 struct sk_buff *skb, *next; 5513 5514 INIT_LIST_HEAD(&sublist); 5515 list_for_each_entry_safe(skb, next, head, list) { 5516 struct net_device *orig_dev = skb->dev; 5517 struct packet_type *pt_prev = NULL; 5518 5519 skb_list_del_init(skb); 5520 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5521 if (!pt_prev) 5522 continue; 5523 if (pt_curr != pt_prev || od_curr != orig_dev) { 5524 /* dispatch old sublist */ 5525 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5526 /* start new sublist */ 5527 INIT_LIST_HEAD(&sublist); 5528 pt_curr = pt_prev; 5529 od_curr = orig_dev; 5530 } 5531 list_add_tail(&skb->list, &sublist); 5532 } 5533 5534 /* dispatch final sublist */ 5535 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5536 } 5537 5538 static int __netif_receive_skb(struct sk_buff *skb) 5539 { 5540 int ret; 5541 5542 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { 5543 unsigned int noreclaim_flag; 5544 5545 /* 5546 * PFMEMALLOC skbs are special, they should 5547 * - be delivered to SOCK_MEMALLOC sockets only 5548 * - stay away from userspace 5549 * - have bounded memory usage 5550 * 5551 * Use PF_MEMALLOC as this saves us from propagating the allocation 5552 * context down to all allocation sites. 5553 */ 5554 noreclaim_flag = memalloc_noreclaim_save(); 5555 ret = __netif_receive_skb_one_core(skb, true); 5556 memalloc_noreclaim_restore(noreclaim_flag); 5557 } else 5558 ret = __netif_receive_skb_one_core(skb, false); 5559 5560 return ret; 5561 } 5562 5563 static void __netif_receive_skb_list(struct list_head *head) 5564 { 5565 unsigned long noreclaim_flag = 0; 5566 struct sk_buff *skb, *next; 5567 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */ 5568 5569 list_for_each_entry_safe(skb, next, head, list) { 5570 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) { 5571 struct list_head sublist; 5572 5573 /* Handle the previous sublist */ 5574 list_cut_before(&sublist, head, &skb->list); 5575 if (!list_empty(&sublist)) 5576 __netif_receive_skb_list_core(&sublist, pfmemalloc); 5577 pfmemalloc = !pfmemalloc; 5578 /* See comments in __netif_receive_skb */ 5579 if (pfmemalloc) 5580 noreclaim_flag = memalloc_noreclaim_save(); 5581 else 5582 memalloc_noreclaim_restore(noreclaim_flag); 5583 } 5584 } 5585 /* Handle the remaining sublist */ 5586 if (!list_empty(head)) 5587 __netif_receive_skb_list_core(head, pfmemalloc); 5588 /* Restore pflags */ 5589 if (pfmemalloc) 5590 memalloc_noreclaim_restore(noreclaim_flag); 5591 } 5592 5593 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp) 5594 { 5595 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog); 5596 struct bpf_prog *new = xdp->prog; 5597 int ret = 0; 5598 5599 switch (xdp->command) { 5600 case XDP_SETUP_PROG: 5601 rcu_assign_pointer(dev->xdp_prog, new); 5602 if (old) 5603 bpf_prog_put(old); 5604 5605 if (old && !new) { 5606 static_branch_dec(&generic_xdp_needed_key); 5607 } else if (new && !old) { 5608 static_branch_inc(&generic_xdp_needed_key); 5609 dev_disable_lro(dev); 5610 dev_disable_gro_hw(dev); 5611 } 5612 break; 5613 5614 default: 5615 ret = -EINVAL; 5616 break; 5617 } 5618 5619 return ret; 5620 } 5621 5622 static int netif_receive_skb_internal(struct sk_buff *skb) 5623 { 5624 int ret; 5625 5626 net_timestamp_check(netdev_tstamp_prequeue, skb); 5627 5628 if (skb_defer_rx_timestamp(skb)) 5629 return NET_RX_SUCCESS; 5630 5631 rcu_read_lock(); 5632 #ifdef CONFIG_RPS 5633 if (static_branch_unlikely(&rps_needed)) { 5634 struct rps_dev_flow voidflow, *rflow = &voidflow; 5635 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5636 5637 if (cpu >= 0) { 5638 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5639 rcu_read_unlock(); 5640 return ret; 5641 } 5642 } 5643 #endif 5644 ret = __netif_receive_skb(skb); 5645 rcu_read_unlock(); 5646 return ret; 5647 } 5648 5649 static void netif_receive_skb_list_internal(struct list_head *head) 5650 { 5651 struct sk_buff *skb, *next; 5652 struct list_head sublist; 5653 5654 INIT_LIST_HEAD(&sublist); 5655 list_for_each_entry_safe(skb, next, head, list) { 5656 net_timestamp_check(netdev_tstamp_prequeue, skb); 5657 skb_list_del_init(skb); 5658 if (!skb_defer_rx_timestamp(skb)) 5659 list_add_tail(&skb->list, &sublist); 5660 } 5661 list_splice_init(&sublist, head); 5662 5663 rcu_read_lock(); 5664 #ifdef CONFIG_RPS 5665 if (static_branch_unlikely(&rps_needed)) { 5666 list_for_each_entry_safe(skb, next, head, list) { 5667 struct rps_dev_flow voidflow, *rflow = &voidflow; 5668 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5669 5670 if (cpu >= 0) { 5671 /* Will be handled, remove from list */ 5672 skb_list_del_init(skb); 5673 enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5674 } 5675 } 5676 } 5677 #endif 5678 __netif_receive_skb_list(head); 5679 rcu_read_unlock(); 5680 } 5681 5682 /** 5683 * netif_receive_skb - process receive buffer from network 5684 * @skb: buffer to process 5685 * 5686 * netif_receive_skb() is the main receive data processing function. 5687 * It always succeeds. The buffer may be dropped during processing 5688 * for congestion control or by the protocol layers. 5689 * 5690 * This function may only be called from softirq context and interrupts 5691 * should be enabled. 5692 * 5693 * Return values (usually ignored): 5694 * NET_RX_SUCCESS: no congestion 5695 * NET_RX_DROP: packet was dropped 5696 */ 5697 int netif_receive_skb(struct sk_buff *skb) 5698 { 5699 int ret; 5700 5701 trace_netif_receive_skb_entry(skb); 5702 5703 ret = netif_receive_skb_internal(skb); 5704 trace_netif_receive_skb_exit(ret); 5705 5706 return ret; 5707 } 5708 EXPORT_SYMBOL(netif_receive_skb); 5709 5710 /** 5711 * netif_receive_skb_list - process many receive buffers from network 5712 * @head: list of skbs to process. 5713 * 5714 * Since return value of netif_receive_skb() is normally ignored, and 5715 * wouldn't be meaningful for a list, this function returns void. 5716 * 5717 * This function may only be called from softirq context and interrupts 5718 * should be enabled. 5719 */ 5720 void netif_receive_skb_list(struct list_head *head) 5721 { 5722 struct sk_buff *skb; 5723 5724 if (list_empty(head)) 5725 return; 5726 if (trace_netif_receive_skb_list_entry_enabled()) { 5727 list_for_each_entry(skb, head, list) 5728 trace_netif_receive_skb_list_entry(skb); 5729 } 5730 netif_receive_skb_list_internal(head); 5731 trace_netif_receive_skb_list_exit(0); 5732 } 5733 EXPORT_SYMBOL(netif_receive_skb_list); 5734 5735 static DEFINE_PER_CPU(struct work_struct, flush_works); 5736 5737 /* Network device is going away, flush any packets still pending */ 5738 static void flush_backlog(struct work_struct *work) 5739 { 5740 struct sk_buff *skb, *tmp; 5741 struct softnet_data *sd; 5742 5743 local_bh_disable(); 5744 sd = this_cpu_ptr(&softnet_data); 5745 5746 local_irq_disable(); 5747 rps_lock(sd); 5748 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { 5749 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5750 __skb_unlink(skb, &sd->input_pkt_queue); 5751 dev_kfree_skb_irq(skb); 5752 input_queue_head_incr(sd); 5753 } 5754 } 5755 rps_unlock(sd); 5756 local_irq_enable(); 5757 5758 skb_queue_walk_safe(&sd->process_queue, skb, tmp) { 5759 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5760 __skb_unlink(skb, &sd->process_queue); 5761 kfree_skb(skb); 5762 input_queue_head_incr(sd); 5763 } 5764 } 5765 local_bh_enable(); 5766 } 5767 5768 static bool flush_required(int cpu) 5769 { 5770 #if IS_ENABLED(CONFIG_RPS) 5771 struct softnet_data *sd = &per_cpu(softnet_data, cpu); 5772 bool do_flush; 5773 5774 local_irq_disable(); 5775 rps_lock(sd); 5776 5777 /* as insertion into process_queue happens with the rps lock held, 5778 * process_queue access may race only with dequeue 5779 */ 5780 do_flush = !skb_queue_empty(&sd->input_pkt_queue) || 5781 !skb_queue_empty_lockless(&sd->process_queue); 5782 rps_unlock(sd); 5783 local_irq_enable(); 5784 5785 return do_flush; 5786 #endif 5787 /* without RPS we can't safely check input_pkt_queue: during a 5788 * concurrent remote skb_queue_splice() we can detect as empty both 5789 * input_pkt_queue and process_queue even if the latter could end-up 5790 * containing a lot of packets. 5791 */ 5792 return true; 5793 } 5794 5795 static void flush_all_backlogs(void) 5796 { 5797 static cpumask_t flush_cpus; 5798 unsigned int cpu; 5799 5800 /* since we are under rtnl lock protection we can use static data 5801 * for the cpumask and avoid allocating on stack the possibly 5802 * large mask 5803 */ 5804 ASSERT_RTNL(); 5805 5806 cpus_read_lock(); 5807 5808 cpumask_clear(&flush_cpus); 5809 for_each_online_cpu(cpu) { 5810 if (flush_required(cpu)) { 5811 queue_work_on(cpu, system_highpri_wq, 5812 per_cpu_ptr(&flush_works, cpu)); 5813 cpumask_set_cpu(cpu, &flush_cpus); 5814 } 5815 } 5816 5817 /* we can have in flight packet[s] on the cpus we are not flushing, 5818 * synchronize_net() in unregister_netdevice_many() will take care of 5819 * them 5820 */ 5821 for_each_cpu(cpu, &flush_cpus) 5822 flush_work(per_cpu_ptr(&flush_works, cpu)); 5823 5824 cpus_read_unlock(); 5825 } 5826 5827 /* Pass the currently batched GRO_NORMAL SKBs up to the stack. */ 5828 static void gro_normal_list(struct napi_struct *napi) 5829 { 5830 if (!napi->rx_count) 5831 return; 5832 netif_receive_skb_list_internal(&napi->rx_list); 5833 INIT_LIST_HEAD(&napi->rx_list); 5834 napi->rx_count = 0; 5835 } 5836 5837 /* Queue one GRO_NORMAL SKB up for list processing. If batch size exceeded, 5838 * pass the whole batch up to the stack. 5839 */ 5840 static void gro_normal_one(struct napi_struct *napi, struct sk_buff *skb, int segs) 5841 { 5842 list_add_tail(&skb->list, &napi->rx_list); 5843 napi->rx_count += segs; 5844 if (napi->rx_count >= gro_normal_batch) 5845 gro_normal_list(napi); 5846 } 5847 5848 static void napi_gro_complete(struct napi_struct *napi, struct sk_buff *skb) 5849 { 5850 struct packet_offload *ptype; 5851 __be16 type = skb->protocol; 5852 struct list_head *head = &offload_base; 5853 int err = -ENOENT; 5854 5855 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb)); 5856 5857 if (NAPI_GRO_CB(skb)->count == 1) { 5858 skb_shinfo(skb)->gso_size = 0; 5859 goto out; 5860 } 5861 5862 rcu_read_lock(); 5863 list_for_each_entry_rcu(ptype, head, list) { 5864 if (ptype->type != type || !ptype->callbacks.gro_complete) 5865 continue; 5866 5867 err = INDIRECT_CALL_INET(ptype->callbacks.gro_complete, 5868 ipv6_gro_complete, inet_gro_complete, 5869 skb, 0); 5870 break; 5871 } 5872 rcu_read_unlock(); 5873 5874 if (err) { 5875 WARN_ON(&ptype->list == head); 5876 kfree_skb(skb); 5877 return; 5878 } 5879 5880 out: 5881 gro_normal_one(napi, skb, NAPI_GRO_CB(skb)->count); 5882 } 5883 5884 static void __napi_gro_flush_chain(struct napi_struct *napi, u32 index, 5885 bool flush_old) 5886 { 5887 struct list_head *head = &napi->gro_hash[index].list; 5888 struct sk_buff *skb, *p; 5889 5890 list_for_each_entry_safe_reverse(skb, p, head, list) { 5891 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies) 5892 return; 5893 skb_list_del_init(skb); 5894 napi_gro_complete(napi, skb); 5895 napi->gro_hash[index].count--; 5896 } 5897 5898 if (!napi->gro_hash[index].count) 5899 __clear_bit(index, &napi->gro_bitmask); 5900 } 5901 5902 /* napi->gro_hash[].list contains packets ordered by age. 5903 * youngest packets at the head of it. 5904 * Complete skbs in reverse order to reduce latencies. 5905 */ 5906 void napi_gro_flush(struct napi_struct *napi, bool flush_old) 5907 { 5908 unsigned long bitmask = napi->gro_bitmask; 5909 unsigned int i, base = ~0U; 5910 5911 while ((i = ffs(bitmask)) != 0) { 5912 bitmask >>= i; 5913 base += i; 5914 __napi_gro_flush_chain(napi, base, flush_old); 5915 } 5916 } 5917 EXPORT_SYMBOL(napi_gro_flush); 5918 5919 static void gro_list_prepare(const struct list_head *head, 5920 const struct sk_buff *skb) 5921 { 5922 unsigned int maclen = skb->dev->hard_header_len; 5923 u32 hash = skb_get_hash_raw(skb); 5924 struct sk_buff *p; 5925 5926 list_for_each_entry(p, head, list) { 5927 unsigned long diffs; 5928 5929 NAPI_GRO_CB(p)->flush = 0; 5930 5931 if (hash != skb_get_hash_raw(p)) { 5932 NAPI_GRO_CB(p)->same_flow = 0; 5933 continue; 5934 } 5935 5936 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev; 5937 diffs |= skb_vlan_tag_present(p) ^ skb_vlan_tag_present(skb); 5938 if (skb_vlan_tag_present(p)) 5939 diffs |= skb_vlan_tag_get(p) ^ skb_vlan_tag_get(skb); 5940 diffs |= skb_metadata_differs(p, skb); 5941 if (maclen == ETH_HLEN) 5942 diffs |= compare_ether_header(skb_mac_header(p), 5943 skb_mac_header(skb)); 5944 else if (!diffs) 5945 diffs = memcmp(skb_mac_header(p), 5946 skb_mac_header(skb), 5947 maclen); 5948 5949 /* in most common scenarions 'slow_gro' is 0 5950 * otherwise we are already on some slower paths 5951 * either skip all the infrequent tests altogether or 5952 * avoid trying too hard to skip each of them individually 5953 */ 5954 if (!diffs && unlikely(skb->slow_gro | p->slow_gro)) { 5955 #if IS_ENABLED(CONFIG_SKB_EXTENSIONS) && IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 5956 struct tc_skb_ext *skb_ext; 5957 struct tc_skb_ext *p_ext; 5958 #endif 5959 5960 diffs |= p->sk != skb->sk; 5961 diffs |= skb_metadata_dst_cmp(p, skb); 5962 diffs |= skb_get_nfct(p) ^ skb_get_nfct(skb); 5963 5964 #if IS_ENABLED(CONFIG_SKB_EXTENSIONS) && IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 5965 skb_ext = skb_ext_find(skb, TC_SKB_EXT); 5966 p_ext = skb_ext_find(p, TC_SKB_EXT); 5967 5968 diffs |= (!!p_ext) ^ (!!skb_ext); 5969 if (!diffs && unlikely(skb_ext)) 5970 diffs |= p_ext->chain ^ skb_ext->chain; 5971 #endif 5972 } 5973 5974 NAPI_GRO_CB(p)->same_flow = !diffs; 5975 } 5976 } 5977 5978 static inline void skb_gro_reset_offset(struct sk_buff *skb, u32 nhoff) 5979 { 5980 const struct skb_shared_info *pinfo = skb_shinfo(skb); 5981 const skb_frag_t *frag0 = &pinfo->frags[0]; 5982 5983 NAPI_GRO_CB(skb)->data_offset = 0; 5984 NAPI_GRO_CB(skb)->frag0 = NULL; 5985 NAPI_GRO_CB(skb)->frag0_len = 0; 5986 5987 if (!skb_headlen(skb) && pinfo->nr_frags && 5988 !PageHighMem(skb_frag_page(frag0)) && 5989 (!NET_IP_ALIGN || !((skb_frag_off(frag0) + nhoff) & 3))) { 5990 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0); 5991 NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int, 5992 skb_frag_size(frag0), 5993 skb->end - skb->tail); 5994 } 5995 } 5996 5997 static void gro_pull_from_frag0(struct sk_buff *skb, int grow) 5998 { 5999 struct skb_shared_info *pinfo = skb_shinfo(skb); 6000 6001 BUG_ON(skb->end - skb->tail < grow); 6002 6003 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); 6004 6005 skb->data_len -= grow; 6006 skb->tail += grow; 6007 6008 skb_frag_off_add(&pinfo->frags[0], grow); 6009 skb_frag_size_sub(&pinfo->frags[0], grow); 6010 6011 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) { 6012 skb_frag_unref(skb, 0); 6013 memmove(pinfo->frags, pinfo->frags + 1, 6014 --pinfo->nr_frags * sizeof(pinfo->frags[0])); 6015 } 6016 } 6017 6018 static void gro_flush_oldest(struct napi_struct *napi, struct list_head *head) 6019 { 6020 struct sk_buff *oldest; 6021 6022 oldest = list_last_entry(head, struct sk_buff, list); 6023 6024 /* We are called with head length >= MAX_GRO_SKBS, so this is 6025 * impossible. 6026 */ 6027 if (WARN_ON_ONCE(!oldest)) 6028 return; 6029 6030 /* Do not adjust napi->gro_hash[].count, caller is adding a new 6031 * SKB to the chain. 6032 */ 6033 skb_list_del_init(oldest); 6034 napi_gro_complete(napi, oldest); 6035 } 6036 6037 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) 6038 { 6039 u32 bucket = skb_get_hash_raw(skb) & (GRO_HASH_BUCKETS - 1); 6040 struct gro_list *gro_list = &napi->gro_hash[bucket]; 6041 struct list_head *head = &offload_base; 6042 struct packet_offload *ptype; 6043 __be16 type = skb->protocol; 6044 struct sk_buff *pp = NULL; 6045 enum gro_result ret; 6046 int same_flow; 6047 int grow; 6048 6049 if (netif_elide_gro(skb->dev)) 6050 goto normal; 6051 6052 gro_list_prepare(&gro_list->list, skb); 6053 6054 rcu_read_lock(); 6055 list_for_each_entry_rcu(ptype, head, list) { 6056 if (ptype->type != type || !ptype->callbacks.gro_receive) 6057 continue; 6058 6059 skb_set_network_header(skb, skb_gro_offset(skb)); 6060 skb_reset_mac_len(skb); 6061 NAPI_GRO_CB(skb)->same_flow = 0; 6062 NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb); 6063 NAPI_GRO_CB(skb)->free = 0; 6064 NAPI_GRO_CB(skb)->encap_mark = 0; 6065 NAPI_GRO_CB(skb)->recursion_counter = 0; 6066 NAPI_GRO_CB(skb)->is_fou = 0; 6067 NAPI_GRO_CB(skb)->is_atomic = 1; 6068 NAPI_GRO_CB(skb)->gro_remcsum_start = 0; 6069 6070 /* Setup for GRO checksum validation */ 6071 switch (skb->ip_summed) { 6072 case CHECKSUM_COMPLETE: 6073 NAPI_GRO_CB(skb)->csum = skb->csum; 6074 NAPI_GRO_CB(skb)->csum_valid = 1; 6075 NAPI_GRO_CB(skb)->csum_cnt = 0; 6076 break; 6077 case CHECKSUM_UNNECESSARY: 6078 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1; 6079 NAPI_GRO_CB(skb)->csum_valid = 0; 6080 break; 6081 default: 6082 NAPI_GRO_CB(skb)->csum_cnt = 0; 6083 NAPI_GRO_CB(skb)->csum_valid = 0; 6084 } 6085 6086 pp = INDIRECT_CALL_INET(ptype->callbacks.gro_receive, 6087 ipv6_gro_receive, inet_gro_receive, 6088 &gro_list->list, skb); 6089 break; 6090 } 6091 rcu_read_unlock(); 6092 6093 if (&ptype->list == head) 6094 goto normal; 6095 6096 if (PTR_ERR(pp) == -EINPROGRESS) { 6097 ret = GRO_CONSUMED; 6098 goto ok; 6099 } 6100 6101 same_flow = NAPI_GRO_CB(skb)->same_flow; 6102 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; 6103 6104 if (pp) { 6105 skb_list_del_init(pp); 6106 napi_gro_complete(napi, pp); 6107 gro_list->count--; 6108 } 6109 6110 if (same_flow) 6111 goto ok; 6112 6113 if (NAPI_GRO_CB(skb)->flush) 6114 goto normal; 6115 6116 if (unlikely(gro_list->count >= MAX_GRO_SKBS)) 6117 gro_flush_oldest(napi, &gro_list->list); 6118 else 6119 gro_list->count++; 6120 6121 NAPI_GRO_CB(skb)->count = 1; 6122 NAPI_GRO_CB(skb)->age = jiffies; 6123 NAPI_GRO_CB(skb)->last = skb; 6124 skb_shinfo(skb)->gso_size = skb_gro_len(skb); 6125 list_add(&skb->list, &gro_list->list); 6126 ret = GRO_HELD; 6127 6128 pull: 6129 grow = skb_gro_offset(skb) - skb_headlen(skb); 6130 if (grow > 0) 6131 gro_pull_from_frag0(skb, grow); 6132 ok: 6133 if (gro_list->count) { 6134 if (!test_bit(bucket, &napi->gro_bitmask)) 6135 __set_bit(bucket, &napi->gro_bitmask); 6136 } else if (test_bit(bucket, &napi->gro_bitmask)) { 6137 __clear_bit(bucket, &napi->gro_bitmask); 6138 } 6139 6140 return ret; 6141 6142 normal: 6143 ret = GRO_NORMAL; 6144 goto pull; 6145 } 6146 6147 struct packet_offload *gro_find_receive_by_type(__be16 type) 6148 { 6149 struct list_head *offload_head = &offload_base; 6150 struct packet_offload *ptype; 6151 6152 list_for_each_entry_rcu(ptype, offload_head, list) { 6153 if (ptype->type != type || !ptype->callbacks.gro_receive) 6154 continue; 6155 return ptype; 6156 } 6157 return NULL; 6158 } 6159 EXPORT_SYMBOL(gro_find_receive_by_type); 6160 6161 struct packet_offload *gro_find_complete_by_type(__be16 type) 6162 { 6163 struct list_head *offload_head = &offload_base; 6164 struct packet_offload *ptype; 6165 6166 list_for_each_entry_rcu(ptype, offload_head, list) { 6167 if (ptype->type != type || !ptype->callbacks.gro_complete) 6168 continue; 6169 return ptype; 6170 } 6171 return NULL; 6172 } 6173 EXPORT_SYMBOL(gro_find_complete_by_type); 6174 6175 static gro_result_t napi_skb_finish(struct napi_struct *napi, 6176 struct sk_buff *skb, 6177 gro_result_t ret) 6178 { 6179 switch (ret) { 6180 case GRO_NORMAL: 6181 gro_normal_one(napi, skb, 1); 6182 break; 6183 6184 case GRO_MERGED_FREE: 6185 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) 6186 napi_skb_free_stolen_head(skb); 6187 else if (skb->fclone != SKB_FCLONE_UNAVAILABLE) 6188 __kfree_skb(skb); 6189 else 6190 __kfree_skb_defer(skb); 6191 break; 6192 6193 case GRO_HELD: 6194 case GRO_MERGED: 6195 case GRO_CONSUMED: 6196 break; 6197 } 6198 6199 return ret; 6200 } 6201 6202 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) 6203 { 6204 gro_result_t ret; 6205 6206 skb_mark_napi_id(skb, napi); 6207 trace_napi_gro_receive_entry(skb); 6208 6209 skb_gro_reset_offset(skb, 0); 6210 6211 ret = napi_skb_finish(napi, skb, dev_gro_receive(napi, skb)); 6212 trace_napi_gro_receive_exit(ret); 6213 6214 return ret; 6215 } 6216 EXPORT_SYMBOL(napi_gro_receive); 6217 6218 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) 6219 { 6220 if (unlikely(skb->pfmemalloc)) { 6221 consume_skb(skb); 6222 return; 6223 } 6224 __skb_pull(skb, skb_headlen(skb)); 6225 /* restore the reserve we had after netdev_alloc_skb_ip_align() */ 6226 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb)); 6227 __vlan_hwaccel_clear_tag(skb); 6228 skb->dev = napi->dev; 6229 skb->skb_iif = 0; 6230 6231 /* eth_type_trans() assumes pkt_type is PACKET_HOST */ 6232 skb->pkt_type = PACKET_HOST; 6233 6234 skb->encapsulation = 0; 6235 skb_shinfo(skb)->gso_type = 0; 6236 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); 6237 if (unlikely(skb->slow_gro)) { 6238 skb_orphan(skb); 6239 skb_ext_reset(skb); 6240 nf_reset_ct(skb); 6241 skb->slow_gro = 0; 6242 } 6243 6244 napi->skb = skb; 6245 } 6246 6247 struct sk_buff *napi_get_frags(struct napi_struct *napi) 6248 { 6249 struct sk_buff *skb = napi->skb; 6250 6251 if (!skb) { 6252 skb = napi_alloc_skb(napi, GRO_MAX_HEAD); 6253 if (skb) { 6254 napi->skb = skb; 6255 skb_mark_napi_id(skb, napi); 6256 } 6257 } 6258 return skb; 6259 } 6260 EXPORT_SYMBOL(napi_get_frags); 6261 6262 static gro_result_t napi_frags_finish(struct napi_struct *napi, 6263 struct sk_buff *skb, 6264 gro_result_t ret) 6265 { 6266 switch (ret) { 6267 case GRO_NORMAL: 6268 case GRO_HELD: 6269 __skb_push(skb, ETH_HLEN); 6270 skb->protocol = eth_type_trans(skb, skb->dev); 6271 if (ret == GRO_NORMAL) 6272 gro_normal_one(napi, skb, 1); 6273 break; 6274 6275 case GRO_MERGED_FREE: 6276 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) 6277 napi_skb_free_stolen_head(skb); 6278 else 6279 napi_reuse_skb(napi, skb); 6280 break; 6281 6282 case GRO_MERGED: 6283 case GRO_CONSUMED: 6284 break; 6285 } 6286 6287 return ret; 6288 } 6289 6290 /* Upper GRO stack assumes network header starts at gro_offset=0 6291 * Drivers could call both napi_gro_frags() and napi_gro_receive() 6292 * We copy ethernet header into skb->data to have a common layout. 6293 */ 6294 static struct sk_buff *napi_frags_skb(struct napi_struct *napi) 6295 { 6296 struct sk_buff *skb = napi->skb; 6297 const struct ethhdr *eth; 6298 unsigned int hlen = sizeof(*eth); 6299 6300 napi->skb = NULL; 6301 6302 skb_reset_mac_header(skb); 6303 skb_gro_reset_offset(skb, hlen); 6304 6305 if (unlikely(skb_gro_header_hard(skb, hlen))) { 6306 eth = skb_gro_header_slow(skb, hlen, 0); 6307 if (unlikely(!eth)) { 6308 net_warn_ratelimited("%s: dropping impossible skb from %s\n", 6309 __func__, napi->dev->name); 6310 napi_reuse_skb(napi, skb); 6311 return NULL; 6312 } 6313 } else { 6314 eth = (const struct ethhdr *)skb->data; 6315 gro_pull_from_frag0(skb, hlen); 6316 NAPI_GRO_CB(skb)->frag0 += hlen; 6317 NAPI_GRO_CB(skb)->frag0_len -= hlen; 6318 } 6319 __skb_pull(skb, hlen); 6320 6321 /* 6322 * This works because the only protocols we care about don't require 6323 * special handling. 6324 * We'll fix it up properly in napi_frags_finish() 6325 */ 6326 skb->protocol = eth->h_proto; 6327 6328 return skb; 6329 } 6330 6331 gro_result_t napi_gro_frags(struct napi_struct *napi) 6332 { 6333 gro_result_t ret; 6334 struct sk_buff *skb = napi_frags_skb(napi); 6335 6336 trace_napi_gro_frags_entry(skb); 6337 6338 ret = napi_frags_finish(napi, skb, dev_gro_receive(napi, skb)); 6339 trace_napi_gro_frags_exit(ret); 6340 6341 return ret; 6342 } 6343 EXPORT_SYMBOL(napi_gro_frags); 6344 6345 /* Compute the checksum from gro_offset and return the folded value 6346 * after adding in any pseudo checksum. 6347 */ 6348 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb) 6349 { 6350 __wsum wsum; 6351 __sum16 sum; 6352 6353 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0); 6354 6355 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */ 6356 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum)); 6357 /* See comments in __skb_checksum_complete(). */ 6358 if (likely(!sum)) { 6359 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && 6360 !skb->csum_complete_sw) 6361 netdev_rx_csum_fault(skb->dev, skb); 6362 } 6363 6364 NAPI_GRO_CB(skb)->csum = wsum; 6365 NAPI_GRO_CB(skb)->csum_valid = 1; 6366 6367 return sum; 6368 } 6369 EXPORT_SYMBOL(__skb_gro_checksum_complete); 6370 6371 static void net_rps_send_ipi(struct softnet_data *remsd) 6372 { 6373 #ifdef CONFIG_RPS 6374 while (remsd) { 6375 struct softnet_data *next = remsd->rps_ipi_next; 6376 6377 if (cpu_online(remsd->cpu)) 6378 smp_call_function_single_async(remsd->cpu, &remsd->csd); 6379 remsd = next; 6380 } 6381 #endif 6382 } 6383 6384 /* 6385 * net_rps_action_and_irq_enable sends any pending IPI's for rps. 6386 * Note: called with local irq disabled, but exits with local irq enabled. 6387 */ 6388 static void net_rps_action_and_irq_enable(struct softnet_data *sd) 6389 { 6390 #ifdef CONFIG_RPS 6391 struct softnet_data *remsd = sd->rps_ipi_list; 6392 6393 if (remsd) { 6394 sd->rps_ipi_list = NULL; 6395 6396 local_irq_enable(); 6397 6398 /* Send pending IPI's to kick RPS processing on remote cpus. */ 6399 net_rps_send_ipi(remsd); 6400 } else 6401 #endif 6402 local_irq_enable(); 6403 } 6404 6405 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) 6406 { 6407 #ifdef CONFIG_RPS 6408 return sd->rps_ipi_list != NULL; 6409 #else 6410 return false; 6411 #endif 6412 } 6413 6414 static int process_backlog(struct napi_struct *napi, int quota) 6415 { 6416 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); 6417 bool again = true; 6418 int work = 0; 6419 6420 /* Check if we have pending ipi, its better to send them now, 6421 * not waiting net_rx_action() end. 6422 */ 6423 if (sd_has_rps_ipi_waiting(sd)) { 6424 local_irq_disable(); 6425 net_rps_action_and_irq_enable(sd); 6426 } 6427 6428 napi->weight = dev_rx_weight; 6429 while (again) { 6430 struct sk_buff *skb; 6431 6432 while ((skb = __skb_dequeue(&sd->process_queue))) { 6433 rcu_read_lock(); 6434 __netif_receive_skb(skb); 6435 rcu_read_unlock(); 6436 input_queue_head_incr(sd); 6437 if (++work >= quota) 6438 return work; 6439 6440 } 6441 6442 local_irq_disable(); 6443 rps_lock(sd); 6444 if (skb_queue_empty(&sd->input_pkt_queue)) { 6445 /* 6446 * Inline a custom version of __napi_complete(). 6447 * only current cpu owns and manipulates this napi, 6448 * and NAPI_STATE_SCHED is the only possible flag set 6449 * on backlog. 6450 * We can use a plain write instead of clear_bit(), 6451 * and we dont need an smp_mb() memory barrier. 6452 */ 6453 napi->state = 0; 6454 again = false; 6455 } else { 6456 skb_queue_splice_tail_init(&sd->input_pkt_queue, 6457 &sd->process_queue); 6458 } 6459 rps_unlock(sd); 6460 local_irq_enable(); 6461 } 6462 6463 return work; 6464 } 6465 6466 /** 6467 * __napi_schedule - schedule for receive 6468 * @n: entry to schedule 6469 * 6470 * The entry's receive function will be scheduled to run. 6471 * Consider using __napi_schedule_irqoff() if hard irqs are masked. 6472 */ 6473 void __napi_schedule(struct napi_struct *n) 6474 { 6475 unsigned long flags; 6476 6477 local_irq_save(flags); 6478 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6479 local_irq_restore(flags); 6480 } 6481 EXPORT_SYMBOL(__napi_schedule); 6482 6483 /** 6484 * napi_schedule_prep - check if napi can be scheduled 6485 * @n: napi context 6486 * 6487 * Test if NAPI routine is already running, and if not mark 6488 * it as running. This is used as a condition variable to 6489 * insure only one NAPI poll instance runs. We also make 6490 * sure there is no pending NAPI disable. 6491 */ 6492 bool napi_schedule_prep(struct napi_struct *n) 6493 { 6494 unsigned long val, new; 6495 6496 do { 6497 val = READ_ONCE(n->state); 6498 if (unlikely(val & NAPIF_STATE_DISABLE)) 6499 return false; 6500 new = val | NAPIF_STATE_SCHED; 6501 6502 /* Sets STATE_MISSED bit if STATE_SCHED was already set 6503 * This was suggested by Alexander Duyck, as compiler 6504 * emits better code than : 6505 * if (val & NAPIF_STATE_SCHED) 6506 * new |= NAPIF_STATE_MISSED; 6507 */ 6508 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * 6509 NAPIF_STATE_MISSED; 6510 } while (cmpxchg(&n->state, val, new) != val); 6511 6512 return !(val & NAPIF_STATE_SCHED); 6513 } 6514 EXPORT_SYMBOL(napi_schedule_prep); 6515 6516 /** 6517 * __napi_schedule_irqoff - schedule for receive 6518 * @n: entry to schedule 6519 * 6520 * Variant of __napi_schedule() assuming hard irqs are masked. 6521 * 6522 * On PREEMPT_RT enabled kernels this maps to __napi_schedule() 6523 * because the interrupt disabled assumption might not be true 6524 * due to force-threaded interrupts and spinlock substitution. 6525 */ 6526 void __napi_schedule_irqoff(struct napi_struct *n) 6527 { 6528 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6529 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6530 else 6531 __napi_schedule(n); 6532 } 6533 EXPORT_SYMBOL(__napi_schedule_irqoff); 6534 6535 bool napi_complete_done(struct napi_struct *n, int work_done) 6536 { 6537 unsigned long flags, val, new, timeout = 0; 6538 bool ret = true; 6539 6540 /* 6541 * 1) Don't let napi dequeue from the cpu poll list 6542 * just in case its running on a different cpu. 6543 * 2) If we are busy polling, do nothing here, we have 6544 * the guarantee we will be called later. 6545 */ 6546 if (unlikely(n->state & (NAPIF_STATE_NPSVC | 6547 NAPIF_STATE_IN_BUSY_POLL))) 6548 return false; 6549 6550 if (work_done) { 6551 if (n->gro_bitmask) 6552 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6553 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs); 6554 } 6555 if (n->defer_hard_irqs_count > 0) { 6556 n->defer_hard_irqs_count--; 6557 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6558 if (timeout) 6559 ret = false; 6560 } 6561 if (n->gro_bitmask) { 6562 /* When the NAPI instance uses a timeout and keeps postponing 6563 * it, we need to bound somehow the time packets are kept in 6564 * the GRO layer 6565 */ 6566 napi_gro_flush(n, !!timeout); 6567 } 6568 6569 gro_normal_list(n); 6570 6571 if (unlikely(!list_empty(&n->poll_list))) { 6572 /* If n->poll_list is not empty, we need to mask irqs */ 6573 local_irq_save(flags); 6574 list_del_init(&n->poll_list); 6575 local_irq_restore(flags); 6576 } 6577 6578 do { 6579 val = READ_ONCE(n->state); 6580 6581 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); 6582 6583 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED | 6584 NAPIF_STATE_SCHED_THREADED | 6585 NAPIF_STATE_PREFER_BUSY_POLL); 6586 6587 /* If STATE_MISSED was set, leave STATE_SCHED set, 6588 * because we will call napi->poll() one more time. 6589 * This C code was suggested by Alexander Duyck to help gcc. 6590 */ 6591 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * 6592 NAPIF_STATE_SCHED; 6593 } while (cmpxchg(&n->state, val, new) != val); 6594 6595 if (unlikely(val & NAPIF_STATE_MISSED)) { 6596 __napi_schedule(n); 6597 return false; 6598 } 6599 6600 if (timeout) 6601 hrtimer_start(&n->timer, ns_to_ktime(timeout), 6602 HRTIMER_MODE_REL_PINNED); 6603 return ret; 6604 } 6605 EXPORT_SYMBOL(napi_complete_done); 6606 6607 /* must be called under rcu_read_lock(), as we dont take a reference */ 6608 static struct napi_struct *napi_by_id(unsigned int napi_id) 6609 { 6610 unsigned int hash = napi_id % HASH_SIZE(napi_hash); 6611 struct napi_struct *napi; 6612 6613 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) 6614 if (napi->napi_id == napi_id) 6615 return napi; 6616 6617 return NULL; 6618 } 6619 6620 #if defined(CONFIG_NET_RX_BUSY_POLL) 6621 6622 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule) 6623 { 6624 if (!skip_schedule) { 6625 gro_normal_list(napi); 6626 __napi_schedule(napi); 6627 return; 6628 } 6629 6630 if (napi->gro_bitmask) { 6631 /* flush too old packets 6632 * If HZ < 1000, flush all packets. 6633 */ 6634 napi_gro_flush(napi, HZ >= 1000); 6635 } 6636 6637 gro_normal_list(napi); 6638 clear_bit(NAPI_STATE_SCHED, &napi->state); 6639 } 6640 6641 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll, 6642 u16 budget) 6643 { 6644 bool skip_schedule = false; 6645 unsigned long timeout; 6646 int rc; 6647 6648 /* Busy polling means there is a high chance device driver hard irq 6649 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was 6650 * set in napi_schedule_prep(). 6651 * Since we are about to call napi->poll() once more, we can safely 6652 * clear NAPI_STATE_MISSED. 6653 * 6654 * Note: x86 could use a single "lock and ..." instruction 6655 * to perform these two clear_bit() 6656 */ 6657 clear_bit(NAPI_STATE_MISSED, &napi->state); 6658 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); 6659 6660 local_bh_disable(); 6661 6662 if (prefer_busy_poll) { 6663 napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs); 6664 timeout = READ_ONCE(napi->dev->gro_flush_timeout); 6665 if (napi->defer_hard_irqs_count && timeout) { 6666 hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED); 6667 skip_schedule = true; 6668 } 6669 } 6670 6671 /* All we really want here is to re-enable device interrupts. 6672 * Ideally, a new ndo_busy_poll_stop() could avoid another round. 6673 */ 6674 rc = napi->poll(napi, budget); 6675 /* We can't gro_normal_list() here, because napi->poll() might have 6676 * rearmed the napi (napi_complete_done()) in which case it could 6677 * already be running on another CPU. 6678 */ 6679 trace_napi_poll(napi, rc, budget); 6680 netpoll_poll_unlock(have_poll_lock); 6681 if (rc == budget) 6682 __busy_poll_stop(napi, skip_schedule); 6683 local_bh_enable(); 6684 } 6685 6686 void napi_busy_loop(unsigned int napi_id, 6687 bool (*loop_end)(void *, unsigned long), 6688 void *loop_end_arg, bool prefer_busy_poll, u16 budget) 6689 { 6690 unsigned long start_time = loop_end ? busy_loop_current_time() : 0; 6691 int (*napi_poll)(struct napi_struct *napi, int budget); 6692 void *have_poll_lock = NULL; 6693 struct napi_struct *napi; 6694 6695 restart: 6696 napi_poll = NULL; 6697 6698 rcu_read_lock(); 6699 6700 napi = napi_by_id(napi_id); 6701 if (!napi) 6702 goto out; 6703 6704 preempt_disable(); 6705 for (;;) { 6706 int work = 0; 6707 6708 local_bh_disable(); 6709 if (!napi_poll) { 6710 unsigned long val = READ_ONCE(napi->state); 6711 6712 /* If multiple threads are competing for this napi, 6713 * we avoid dirtying napi->state as much as we can. 6714 */ 6715 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | 6716 NAPIF_STATE_IN_BUSY_POLL)) { 6717 if (prefer_busy_poll) 6718 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6719 goto count; 6720 } 6721 if (cmpxchg(&napi->state, val, 6722 val | NAPIF_STATE_IN_BUSY_POLL | 6723 NAPIF_STATE_SCHED) != val) { 6724 if (prefer_busy_poll) 6725 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6726 goto count; 6727 } 6728 have_poll_lock = netpoll_poll_lock(napi); 6729 napi_poll = napi->poll; 6730 } 6731 work = napi_poll(napi, budget); 6732 trace_napi_poll(napi, work, budget); 6733 gro_normal_list(napi); 6734 count: 6735 if (work > 0) 6736 __NET_ADD_STATS(dev_net(napi->dev), 6737 LINUX_MIB_BUSYPOLLRXPACKETS, work); 6738 local_bh_enable(); 6739 6740 if (!loop_end || loop_end(loop_end_arg, start_time)) 6741 break; 6742 6743 if (unlikely(need_resched())) { 6744 if (napi_poll) 6745 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6746 preempt_enable(); 6747 rcu_read_unlock(); 6748 cond_resched(); 6749 if (loop_end(loop_end_arg, start_time)) 6750 return; 6751 goto restart; 6752 } 6753 cpu_relax(); 6754 } 6755 if (napi_poll) 6756 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6757 preempt_enable(); 6758 out: 6759 rcu_read_unlock(); 6760 } 6761 EXPORT_SYMBOL(napi_busy_loop); 6762 6763 #endif /* CONFIG_NET_RX_BUSY_POLL */ 6764 6765 static void napi_hash_add(struct napi_struct *napi) 6766 { 6767 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state)) 6768 return; 6769 6770 spin_lock(&napi_hash_lock); 6771 6772 /* 0..NR_CPUS range is reserved for sender_cpu use */ 6773 do { 6774 if (unlikely(++napi_gen_id < MIN_NAPI_ID)) 6775 napi_gen_id = MIN_NAPI_ID; 6776 } while (napi_by_id(napi_gen_id)); 6777 napi->napi_id = napi_gen_id; 6778 6779 hlist_add_head_rcu(&napi->napi_hash_node, 6780 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); 6781 6782 spin_unlock(&napi_hash_lock); 6783 } 6784 6785 /* Warning : caller is responsible to make sure rcu grace period 6786 * is respected before freeing memory containing @napi 6787 */ 6788 static void napi_hash_del(struct napi_struct *napi) 6789 { 6790 spin_lock(&napi_hash_lock); 6791 6792 hlist_del_init_rcu(&napi->napi_hash_node); 6793 6794 spin_unlock(&napi_hash_lock); 6795 } 6796 6797 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) 6798 { 6799 struct napi_struct *napi; 6800 6801 napi = container_of(timer, struct napi_struct, timer); 6802 6803 /* Note : we use a relaxed variant of napi_schedule_prep() not setting 6804 * NAPI_STATE_MISSED, since we do not react to a device IRQ. 6805 */ 6806 if (!napi_disable_pending(napi) && 6807 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) { 6808 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6809 __napi_schedule_irqoff(napi); 6810 } 6811 6812 return HRTIMER_NORESTART; 6813 } 6814 6815 static void init_gro_hash(struct napi_struct *napi) 6816 { 6817 int i; 6818 6819 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6820 INIT_LIST_HEAD(&napi->gro_hash[i].list); 6821 napi->gro_hash[i].count = 0; 6822 } 6823 napi->gro_bitmask = 0; 6824 } 6825 6826 int dev_set_threaded(struct net_device *dev, bool threaded) 6827 { 6828 struct napi_struct *napi; 6829 int err = 0; 6830 6831 if (dev->threaded == threaded) 6832 return 0; 6833 6834 if (threaded) { 6835 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6836 if (!napi->thread) { 6837 err = napi_kthread_create(napi); 6838 if (err) { 6839 threaded = false; 6840 break; 6841 } 6842 } 6843 } 6844 } 6845 6846 dev->threaded = threaded; 6847 6848 /* Make sure kthread is created before THREADED bit 6849 * is set. 6850 */ 6851 smp_mb__before_atomic(); 6852 6853 /* Setting/unsetting threaded mode on a napi might not immediately 6854 * take effect, if the current napi instance is actively being 6855 * polled. In this case, the switch between threaded mode and 6856 * softirq mode will happen in the next round of napi_schedule(). 6857 * This should not cause hiccups/stalls to the live traffic. 6858 */ 6859 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6860 if (threaded) 6861 set_bit(NAPI_STATE_THREADED, &napi->state); 6862 else 6863 clear_bit(NAPI_STATE_THREADED, &napi->state); 6864 } 6865 6866 return err; 6867 } 6868 EXPORT_SYMBOL(dev_set_threaded); 6869 6870 void netif_napi_add(struct net_device *dev, struct napi_struct *napi, 6871 int (*poll)(struct napi_struct *, int), int weight) 6872 { 6873 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state))) 6874 return; 6875 6876 INIT_LIST_HEAD(&napi->poll_list); 6877 INIT_HLIST_NODE(&napi->napi_hash_node); 6878 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); 6879 napi->timer.function = napi_watchdog; 6880 init_gro_hash(napi); 6881 napi->skb = NULL; 6882 INIT_LIST_HEAD(&napi->rx_list); 6883 napi->rx_count = 0; 6884 napi->poll = poll; 6885 if (weight > NAPI_POLL_WEIGHT) 6886 netdev_err_once(dev, "%s() called with weight %d\n", __func__, 6887 weight); 6888 napi->weight = weight; 6889 napi->dev = dev; 6890 #ifdef CONFIG_NETPOLL 6891 napi->poll_owner = -1; 6892 #endif 6893 set_bit(NAPI_STATE_SCHED, &napi->state); 6894 set_bit(NAPI_STATE_NPSVC, &napi->state); 6895 list_add_rcu(&napi->dev_list, &dev->napi_list); 6896 napi_hash_add(napi); 6897 /* Create kthread for this napi if dev->threaded is set. 6898 * Clear dev->threaded if kthread creation failed so that 6899 * threaded mode will not be enabled in napi_enable(). 6900 */ 6901 if (dev->threaded && napi_kthread_create(napi)) 6902 dev->threaded = 0; 6903 } 6904 EXPORT_SYMBOL(netif_napi_add); 6905 6906 void napi_disable(struct napi_struct *n) 6907 { 6908 unsigned long val, new; 6909 6910 might_sleep(); 6911 set_bit(NAPI_STATE_DISABLE, &n->state); 6912 6913 do { 6914 val = READ_ONCE(n->state); 6915 if (val & (NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC)) { 6916 usleep_range(20, 200); 6917 continue; 6918 } 6919 6920 new = val | NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC; 6921 new &= ~(NAPIF_STATE_THREADED | NAPIF_STATE_PREFER_BUSY_POLL); 6922 } while (cmpxchg(&n->state, val, new) != val); 6923 6924 hrtimer_cancel(&n->timer); 6925 6926 clear_bit(NAPI_STATE_DISABLE, &n->state); 6927 } 6928 EXPORT_SYMBOL(napi_disable); 6929 6930 /** 6931 * napi_enable - enable NAPI scheduling 6932 * @n: NAPI context 6933 * 6934 * Resume NAPI from being scheduled on this context. 6935 * Must be paired with napi_disable. 6936 */ 6937 void napi_enable(struct napi_struct *n) 6938 { 6939 unsigned long val, new; 6940 6941 do { 6942 val = READ_ONCE(n->state); 6943 BUG_ON(!test_bit(NAPI_STATE_SCHED, &val)); 6944 6945 new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC); 6946 if (n->dev->threaded && n->thread) 6947 new |= NAPIF_STATE_THREADED; 6948 } while (cmpxchg(&n->state, val, new) != val); 6949 } 6950 EXPORT_SYMBOL(napi_enable); 6951 6952 static void flush_gro_hash(struct napi_struct *napi) 6953 { 6954 int i; 6955 6956 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6957 struct sk_buff *skb, *n; 6958 6959 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list) 6960 kfree_skb(skb); 6961 napi->gro_hash[i].count = 0; 6962 } 6963 } 6964 6965 /* Must be called in process context */ 6966 void __netif_napi_del(struct napi_struct *napi) 6967 { 6968 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state)) 6969 return; 6970 6971 napi_hash_del(napi); 6972 list_del_rcu(&napi->dev_list); 6973 napi_free_frags(napi); 6974 6975 flush_gro_hash(napi); 6976 napi->gro_bitmask = 0; 6977 6978 if (napi->thread) { 6979 kthread_stop(napi->thread); 6980 napi->thread = NULL; 6981 } 6982 } 6983 EXPORT_SYMBOL(__netif_napi_del); 6984 6985 static int __napi_poll(struct napi_struct *n, bool *repoll) 6986 { 6987 int work, weight; 6988 6989 weight = n->weight; 6990 6991 /* This NAPI_STATE_SCHED test is for avoiding a race 6992 * with netpoll's poll_napi(). Only the entity which 6993 * obtains the lock and sees NAPI_STATE_SCHED set will 6994 * actually make the ->poll() call. Therefore we avoid 6995 * accidentally calling ->poll() when NAPI is not scheduled. 6996 */ 6997 work = 0; 6998 if (test_bit(NAPI_STATE_SCHED, &n->state)) { 6999 work = n->poll(n, weight); 7000 trace_napi_poll(n, work, weight); 7001 } 7002 7003 if (unlikely(work > weight)) 7004 pr_err_once("NAPI poll function %pS returned %d, exceeding its budget of %d.\n", 7005 n->poll, work, weight); 7006 7007 if (likely(work < weight)) 7008 return work; 7009 7010 /* Drivers must not modify the NAPI state if they 7011 * consume the entire weight. In such cases this code 7012 * still "owns" the NAPI instance and therefore can 7013 * move the instance around on the list at-will. 7014 */ 7015 if (unlikely(napi_disable_pending(n))) { 7016 napi_complete(n); 7017 return work; 7018 } 7019 7020 /* The NAPI context has more processing work, but busy-polling 7021 * is preferred. Exit early. 7022 */ 7023 if (napi_prefer_busy_poll(n)) { 7024 if (napi_complete_done(n, work)) { 7025 /* If timeout is not set, we need to make sure 7026 * that the NAPI is re-scheduled. 7027 */ 7028 napi_schedule(n); 7029 } 7030 return work; 7031 } 7032 7033 if (n->gro_bitmask) { 7034 /* flush too old packets 7035 * If HZ < 1000, flush all packets. 7036 */ 7037 napi_gro_flush(n, HZ >= 1000); 7038 } 7039 7040 gro_normal_list(n); 7041 7042 /* Some drivers may have called napi_schedule 7043 * prior to exhausting their budget. 7044 */ 7045 if (unlikely(!list_empty(&n->poll_list))) { 7046 pr_warn_once("%s: Budget exhausted after napi rescheduled\n", 7047 n->dev ? n->dev->name : "backlog"); 7048 return work; 7049 } 7050 7051 *repoll = true; 7052 7053 return work; 7054 } 7055 7056 static int napi_poll(struct napi_struct *n, struct list_head *repoll) 7057 { 7058 bool do_repoll = false; 7059 void *have; 7060 int work; 7061 7062 list_del_init(&n->poll_list); 7063 7064 have = netpoll_poll_lock(n); 7065 7066 work = __napi_poll(n, &do_repoll); 7067 7068 if (do_repoll) 7069 list_add_tail(&n->poll_list, repoll); 7070 7071 netpoll_poll_unlock(have); 7072 7073 return work; 7074 } 7075 7076 static int napi_thread_wait(struct napi_struct *napi) 7077 { 7078 bool woken = false; 7079 7080 set_current_state(TASK_INTERRUPTIBLE); 7081 7082 while (!kthread_should_stop()) { 7083 /* Testing SCHED_THREADED bit here to make sure the current 7084 * kthread owns this napi and could poll on this napi. 7085 * Testing SCHED bit is not enough because SCHED bit might be 7086 * set by some other busy poll thread or by napi_disable(). 7087 */ 7088 if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state) || woken) { 7089 WARN_ON(!list_empty(&napi->poll_list)); 7090 __set_current_state(TASK_RUNNING); 7091 return 0; 7092 } 7093 7094 schedule(); 7095 /* woken being true indicates this thread owns this napi. */ 7096 woken = true; 7097 set_current_state(TASK_INTERRUPTIBLE); 7098 } 7099 __set_current_state(TASK_RUNNING); 7100 7101 return -1; 7102 } 7103 7104 static int napi_threaded_poll(void *data) 7105 { 7106 struct napi_struct *napi = data; 7107 void *have; 7108 7109 while (!napi_thread_wait(napi)) { 7110 for (;;) { 7111 bool repoll = false; 7112 7113 local_bh_disable(); 7114 7115 have = netpoll_poll_lock(napi); 7116 __napi_poll(napi, &repoll); 7117 netpoll_poll_unlock(have); 7118 7119 local_bh_enable(); 7120 7121 if (!repoll) 7122 break; 7123 7124 cond_resched(); 7125 } 7126 } 7127 return 0; 7128 } 7129 7130 static __latent_entropy void net_rx_action(struct softirq_action *h) 7131 { 7132 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 7133 unsigned long time_limit = jiffies + 7134 usecs_to_jiffies(netdev_budget_usecs); 7135 int budget = netdev_budget; 7136 LIST_HEAD(list); 7137 LIST_HEAD(repoll); 7138 7139 local_irq_disable(); 7140 list_splice_init(&sd->poll_list, &list); 7141 local_irq_enable(); 7142 7143 for (;;) { 7144 struct napi_struct *n; 7145 7146 if (list_empty(&list)) { 7147 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll)) 7148 return; 7149 break; 7150 } 7151 7152 n = list_first_entry(&list, struct napi_struct, poll_list); 7153 budget -= napi_poll(n, &repoll); 7154 7155 /* If softirq window is exhausted then punt. 7156 * Allow this to run for 2 jiffies since which will allow 7157 * an average latency of 1.5/HZ. 7158 */ 7159 if (unlikely(budget <= 0 || 7160 time_after_eq(jiffies, time_limit))) { 7161 sd->time_squeeze++; 7162 break; 7163 } 7164 } 7165 7166 local_irq_disable(); 7167 7168 list_splice_tail_init(&sd->poll_list, &list); 7169 list_splice_tail(&repoll, &list); 7170 list_splice(&list, &sd->poll_list); 7171 if (!list_empty(&sd->poll_list)) 7172 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 7173 7174 net_rps_action_and_irq_enable(sd); 7175 } 7176 7177 struct netdev_adjacent { 7178 struct net_device *dev; 7179 7180 /* upper master flag, there can only be one master device per list */ 7181 bool master; 7182 7183 /* lookup ignore flag */ 7184 bool ignore; 7185 7186 /* counter for the number of times this device was added to us */ 7187 u16 ref_nr; 7188 7189 /* private field for the users */ 7190 void *private; 7191 7192 struct list_head list; 7193 struct rcu_head rcu; 7194 }; 7195 7196 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, 7197 struct list_head *adj_list) 7198 { 7199 struct netdev_adjacent *adj; 7200 7201 list_for_each_entry(adj, adj_list, list) { 7202 if (adj->dev == adj_dev) 7203 return adj; 7204 } 7205 return NULL; 7206 } 7207 7208 static int ____netdev_has_upper_dev(struct net_device *upper_dev, 7209 struct netdev_nested_priv *priv) 7210 { 7211 struct net_device *dev = (struct net_device *)priv->data; 7212 7213 return upper_dev == dev; 7214 } 7215 7216 /** 7217 * netdev_has_upper_dev - Check if device is linked to an upper device 7218 * @dev: device 7219 * @upper_dev: upper device to check 7220 * 7221 * Find out if a device is linked to specified upper device and return true 7222 * in case it is. Note that this checks only immediate upper device, 7223 * not through a complete stack of devices. The caller must hold the RTNL lock. 7224 */ 7225 bool netdev_has_upper_dev(struct net_device *dev, 7226 struct net_device *upper_dev) 7227 { 7228 struct netdev_nested_priv priv = { 7229 .data = (void *)upper_dev, 7230 }; 7231 7232 ASSERT_RTNL(); 7233 7234 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 7235 &priv); 7236 } 7237 EXPORT_SYMBOL(netdev_has_upper_dev); 7238 7239 /** 7240 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device 7241 * @dev: device 7242 * @upper_dev: upper device to check 7243 * 7244 * Find out if a device is linked to specified upper device and return true 7245 * in case it is. Note that this checks the entire upper device chain. 7246 * The caller must hold rcu lock. 7247 */ 7248 7249 bool netdev_has_upper_dev_all_rcu(struct net_device *dev, 7250 struct net_device *upper_dev) 7251 { 7252 struct netdev_nested_priv priv = { 7253 .data = (void *)upper_dev, 7254 }; 7255 7256 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 7257 &priv); 7258 } 7259 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); 7260 7261 /** 7262 * netdev_has_any_upper_dev - Check if device is linked to some device 7263 * @dev: device 7264 * 7265 * Find out if a device is linked to an upper device and return true in case 7266 * it is. The caller must hold the RTNL lock. 7267 */ 7268 bool netdev_has_any_upper_dev(struct net_device *dev) 7269 { 7270 ASSERT_RTNL(); 7271 7272 return !list_empty(&dev->adj_list.upper); 7273 } 7274 EXPORT_SYMBOL(netdev_has_any_upper_dev); 7275 7276 /** 7277 * netdev_master_upper_dev_get - Get master upper device 7278 * @dev: device 7279 * 7280 * Find a master upper device and return pointer to it or NULL in case 7281 * it's not there. The caller must hold the RTNL lock. 7282 */ 7283 struct net_device *netdev_master_upper_dev_get(struct net_device *dev) 7284 { 7285 struct netdev_adjacent *upper; 7286 7287 ASSERT_RTNL(); 7288 7289 if (list_empty(&dev->adj_list.upper)) 7290 return NULL; 7291 7292 upper = list_first_entry(&dev->adj_list.upper, 7293 struct netdev_adjacent, list); 7294 if (likely(upper->master)) 7295 return upper->dev; 7296 return NULL; 7297 } 7298 EXPORT_SYMBOL(netdev_master_upper_dev_get); 7299 7300 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev) 7301 { 7302 struct netdev_adjacent *upper; 7303 7304 ASSERT_RTNL(); 7305 7306 if (list_empty(&dev->adj_list.upper)) 7307 return NULL; 7308 7309 upper = list_first_entry(&dev->adj_list.upper, 7310 struct netdev_adjacent, list); 7311 if (likely(upper->master) && !upper->ignore) 7312 return upper->dev; 7313 return NULL; 7314 } 7315 7316 /** 7317 * netdev_has_any_lower_dev - Check if device is linked to some device 7318 * @dev: device 7319 * 7320 * Find out if a device is linked to a lower device and return true in case 7321 * it is. The caller must hold the RTNL lock. 7322 */ 7323 static bool netdev_has_any_lower_dev(struct net_device *dev) 7324 { 7325 ASSERT_RTNL(); 7326 7327 return !list_empty(&dev->adj_list.lower); 7328 } 7329 7330 void *netdev_adjacent_get_private(struct list_head *adj_list) 7331 { 7332 struct netdev_adjacent *adj; 7333 7334 adj = list_entry(adj_list, struct netdev_adjacent, list); 7335 7336 return adj->private; 7337 } 7338 EXPORT_SYMBOL(netdev_adjacent_get_private); 7339 7340 /** 7341 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list 7342 * @dev: device 7343 * @iter: list_head ** of the current position 7344 * 7345 * Gets the next device from the dev's upper list, starting from iter 7346 * position. The caller must hold RCU read lock. 7347 */ 7348 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, 7349 struct list_head **iter) 7350 { 7351 struct netdev_adjacent *upper; 7352 7353 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 7354 7355 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7356 7357 if (&upper->list == &dev->adj_list.upper) 7358 return NULL; 7359 7360 *iter = &upper->list; 7361 7362 return upper->dev; 7363 } 7364 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); 7365 7366 static struct net_device *__netdev_next_upper_dev(struct net_device *dev, 7367 struct list_head **iter, 7368 bool *ignore) 7369 { 7370 struct netdev_adjacent *upper; 7371 7372 upper = list_entry((*iter)->next, struct netdev_adjacent, list); 7373 7374 if (&upper->list == &dev->adj_list.upper) 7375 return NULL; 7376 7377 *iter = &upper->list; 7378 *ignore = upper->ignore; 7379 7380 return upper->dev; 7381 } 7382 7383 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, 7384 struct list_head **iter) 7385 { 7386 struct netdev_adjacent *upper; 7387 7388 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 7389 7390 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7391 7392 if (&upper->list == &dev->adj_list.upper) 7393 return NULL; 7394 7395 *iter = &upper->list; 7396 7397 return upper->dev; 7398 } 7399 7400 static int __netdev_walk_all_upper_dev(struct net_device *dev, 7401 int (*fn)(struct net_device *dev, 7402 struct netdev_nested_priv *priv), 7403 struct netdev_nested_priv *priv) 7404 { 7405 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7406 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7407 int ret, cur = 0; 7408 bool ignore; 7409 7410 now = dev; 7411 iter = &dev->adj_list.upper; 7412 7413 while (1) { 7414 if (now != dev) { 7415 ret = fn(now, priv); 7416 if (ret) 7417 return ret; 7418 } 7419 7420 next = NULL; 7421 while (1) { 7422 udev = __netdev_next_upper_dev(now, &iter, &ignore); 7423 if (!udev) 7424 break; 7425 if (ignore) 7426 continue; 7427 7428 next = udev; 7429 niter = &udev->adj_list.upper; 7430 dev_stack[cur] = now; 7431 iter_stack[cur++] = iter; 7432 break; 7433 } 7434 7435 if (!next) { 7436 if (!cur) 7437 return 0; 7438 next = dev_stack[--cur]; 7439 niter = iter_stack[cur]; 7440 } 7441 7442 now = next; 7443 iter = niter; 7444 } 7445 7446 return 0; 7447 } 7448 7449 int netdev_walk_all_upper_dev_rcu(struct net_device *dev, 7450 int (*fn)(struct net_device *dev, 7451 struct netdev_nested_priv *priv), 7452 struct netdev_nested_priv *priv) 7453 { 7454 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7455 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7456 int ret, cur = 0; 7457 7458 now = dev; 7459 iter = &dev->adj_list.upper; 7460 7461 while (1) { 7462 if (now != dev) { 7463 ret = fn(now, priv); 7464 if (ret) 7465 return ret; 7466 } 7467 7468 next = NULL; 7469 while (1) { 7470 udev = netdev_next_upper_dev_rcu(now, &iter); 7471 if (!udev) 7472 break; 7473 7474 next = udev; 7475 niter = &udev->adj_list.upper; 7476 dev_stack[cur] = now; 7477 iter_stack[cur++] = iter; 7478 break; 7479 } 7480 7481 if (!next) { 7482 if (!cur) 7483 return 0; 7484 next = dev_stack[--cur]; 7485 niter = iter_stack[cur]; 7486 } 7487 7488 now = next; 7489 iter = niter; 7490 } 7491 7492 return 0; 7493 } 7494 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); 7495 7496 static bool __netdev_has_upper_dev(struct net_device *dev, 7497 struct net_device *upper_dev) 7498 { 7499 struct netdev_nested_priv priv = { 7500 .flags = 0, 7501 .data = (void *)upper_dev, 7502 }; 7503 7504 ASSERT_RTNL(); 7505 7506 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev, 7507 &priv); 7508 } 7509 7510 /** 7511 * netdev_lower_get_next_private - Get the next ->private from the 7512 * lower neighbour list 7513 * @dev: device 7514 * @iter: list_head ** of the current position 7515 * 7516 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7517 * list, starting from iter position. The caller must hold either hold the 7518 * RTNL lock or its own locking that guarantees that the neighbour lower 7519 * list will remain unchanged. 7520 */ 7521 void *netdev_lower_get_next_private(struct net_device *dev, 7522 struct list_head **iter) 7523 { 7524 struct netdev_adjacent *lower; 7525 7526 lower = list_entry(*iter, struct netdev_adjacent, list); 7527 7528 if (&lower->list == &dev->adj_list.lower) 7529 return NULL; 7530 7531 *iter = lower->list.next; 7532 7533 return lower->private; 7534 } 7535 EXPORT_SYMBOL(netdev_lower_get_next_private); 7536 7537 /** 7538 * netdev_lower_get_next_private_rcu - Get the next ->private from the 7539 * lower neighbour list, RCU 7540 * variant 7541 * @dev: device 7542 * @iter: list_head ** of the current position 7543 * 7544 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7545 * list, starting from iter position. The caller must hold RCU read lock. 7546 */ 7547 void *netdev_lower_get_next_private_rcu(struct net_device *dev, 7548 struct list_head **iter) 7549 { 7550 struct netdev_adjacent *lower; 7551 7552 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 7553 7554 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7555 7556 if (&lower->list == &dev->adj_list.lower) 7557 return NULL; 7558 7559 *iter = &lower->list; 7560 7561 return lower->private; 7562 } 7563 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); 7564 7565 /** 7566 * netdev_lower_get_next - Get the next device from the lower neighbour 7567 * list 7568 * @dev: device 7569 * @iter: list_head ** of the current position 7570 * 7571 * Gets the next netdev_adjacent from the dev's lower neighbour 7572 * list, starting from iter position. The caller must hold RTNL lock or 7573 * its own locking that guarantees that the neighbour lower 7574 * list will remain unchanged. 7575 */ 7576 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) 7577 { 7578 struct netdev_adjacent *lower; 7579 7580 lower = list_entry(*iter, struct netdev_adjacent, list); 7581 7582 if (&lower->list == &dev->adj_list.lower) 7583 return NULL; 7584 7585 *iter = lower->list.next; 7586 7587 return lower->dev; 7588 } 7589 EXPORT_SYMBOL(netdev_lower_get_next); 7590 7591 static struct net_device *netdev_next_lower_dev(struct net_device *dev, 7592 struct list_head **iter) 7593 { 7594 struct netdev_adjacent *lower; 7595 7596 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7597 7598 if (&lower->list == &dev->adj_list.lower) 7599 return NULL; 7600 7601 *iter = &lower->list; 7602 7603 return lower->dev; 7604 } 7605 7606 static struct net_device *__netdev_next_lower_dev(struct net_device *dev, 7607 struct list_head **iter, 7608 bool *ignore) 7609 { 7610 struct netdev_adjacent *lower; 7611 7612 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7613 7614 if (&lower->list == &dev->adj_list.lower) 7615 return NULL; 7616 7617 *iter = &lower->list; 7618 *ignore = lower->ignore; 7619 7620 return lower->dev; 7621 } 7622 7623 int netdev_walk_all_lower_dev(struct net_device *dev, 7624 int (*fn)(struct net_device *dev, 7625 struct netdev_nested_priv *priv), 7626 struct netdev_nested_priv *priv) 7627 { 7628 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7629 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7630 int ret, cur = 0; 7631 7632 now = dev; 7633 iter = &dev->adj_list.lower; 7634 7635 while (1) { 7636 if (now != dev) { 7637 ret = fn(now, priv); 7638 if (ret) 7639 return ret; 7640 } 7641 7642 next = NULL; 7643 while (1) { 7644 ldev = netdev_next_lower_dev(now, &iter); 7645 if (!ldev) 7646 break; 7647 7648 next = ldev; 7649 niter = &ldev->adj_list.lower; 7650 dev_stack[cur] = now; 7651 iter_stack[cur++] = iter; 7652 break; 7653 } 7654 7655 if (!next) { 7656 if (!cur) 7657 return 0; 7658 next = dev_stack[--cur]; 7659 niter = iter_stack[cur]; 7660 } 7661 7662 now = next; 7663 iter = niter; 7664 } 7665 7666 return 0; 7667 } 7668 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); 7669 7670 static int __netdev_walk_all_lower_dev(struct net_device *dev, 7671 int (*fn)(struct net_device *dev, 7672 struct netdev_nested_priv *priv), 7673 struct netdev_nested_priv *priv) 7674 { 7675 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7676 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7677 int ret, cur = 0; 7678 bool ignore; 7679 7680 now = dev; 7681 iter = &dev->adj_list.lower; 7682 7683 while (1) { 7684 if (now != dev) { 7685 ret = fn(now, priv); 7686 if (ret) 7687 return ret; 7688 } 7689 7690 next = NULL; 7691 while (1) { 7692 ldev = __netdev_next_lower_dev(now, &iter, &ignore); 7693 if (!ldev) 7694 break; 7695 if (ignore) 7696 continue; 7697 7698 next = ldev; 7699 niter = &ldev->adj_list.lower; 7700 dev_stack[cur] = now; 7701 iter_stack[cur++] = iter; 7702 break; 7703 } 7704 7705 if (!next) { 7706 if (!cur) 7707 return 0; 7708 next = dev_stack[--cur]; 7709 niter = iter_stack[cur]; 7710 } 7711 7712 now = next; 7713 iter = niter; 7714 } 7715 7716 return 0; 7717 } 7718 7719 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, 7720 struct list_head **iter) 7721 { 7722 struct netdev_adjacent *lower; 7723 7724 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7725 if (&lower->list == &dev->adj_list.lower) 7726 return NULL; 7727 7728 *iter = &lower->list; 7729 7730 return lower->dev; 7731 } 7732 EXPORT_SYMBOL(netdev_next_lower_dev_rcu); 7733 7734 static u8 __netdev_upper_depth(struct net_device *dev) 7735 { 7736 struct net_device *udev; 7737 struct list_head *iter; 7738 u8 max_depth = 0; 7739 bool ignore; 7740 7741 for (iter = &dev->adj_list.upper, 7742 udev = __netdev_next_upper_dev(dev, &iter, &ignore); 7743 udev; 7744 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) { 7745 if (ignore) 7746 continue; 7747 if (max_depth < udev->upper_level) 7748 max_depth = udev->upper_level; 7749 } 7750 7751 return max_depth; 7752 } 7753 7754 static u8 __netdev_lower_depth(struct net_device *dev) 7755 { 7756 struct net_device *ldev; 7757 struct list_head *iter; 7758 u8 max_depth = 0; 7759 bool ignore; 7760 7761 for (iter = &dev->adj_list.lower, 7762 ldev = __netdev_next_lower_dev(dev, &iter, &ignore); 7763 ldev; 7764 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) { 7765 if (ignore) 7766 continue; 7767 if (max_depth < ldev->lower_level) 7768 max_depth = ldev->lower_level; 7769 } 7770 7771 return max_depth; 7772 } 7773 7774 static int __netdev_update_upper_level(struct net_device *dev, 7775 struct netdev_nested_priv *__unused) 7776 { 7777 dev->upper_level = __netdev_upper_depth(dev) + 1; 7778 return 0; 7779 } 7780 7781 static int __netdev_update_lower_level(struct net_device *dev, 7782 struct netdev_nested_priv *priv) 7783 { 7784 dev->lower_level = __netdev_lower_depth(dev) + 1; 7785 7786 #ifdef CONFIG_LOCKDEP 7787 if (!priv) 7788 return 0; 7789 7790 if (priv->flags & NESTED_SYNC_IMM) 7791 dev->nested_level = dev->lower_level - 1; 7792 if (priv->flags & NESTED_SYNC_TODO) 7793 net_unlink_todo(dev); 7794 #endif 7795 return 0; 7796 } 7797 7798 int netdev_walk_all_lower_dev_rcu(struct net_device *dev, 7799 int (*fn)(struct net_device *dev, 7800 struct netdev_nested_priv *priv), 7801 struct netdev_nested_priv *priv) 7802 { 7803 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7804 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7805 int ret, cur = 0; 7806 7807 now = dev; 7808 iter = &dev->adj_list.lower; 7809 7810 while (1) { 7811 if (now != dev) { 7812 ret = fn(now, priv); 7813 if (ret) 7814 return ret; 7815 } 7816 7817 next = NULL; 7818 while (1) { 7819 ldev = netdev_next_lower_dev_rcu(now, &iter); 7820 if (!ldev) 7821 break; 7822 7823 next = ldev; 7824 niter = &ldev->adj_list.lower; 7825 dev_stack[cur] = now; 7826 iter_stack[cur++] = iter; 7827 break; 7828 } 7829 7830 if (!next) { 7831 if (!cur) 7832 return 0; 7833 next = dev_stack[--cur]; 7834 niter = iter_stack[cur]; 7835 } 7836 7837 now = next; 7838 iter = niter; 7839 } 7840 7841 return 0; 7842 } 7843 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); 7844 7845 /** 7846 * netdev_lower_get_first_private_rcu - Get the first ->private from the 7847 * lower neighbour list, RCU 7848 * variant 7849 * @dev: device 7850 * 7851 * Gets the first netdev_adjacent->private from the dev's lower neighbour 7852 * list. The caller must hold RCU read lock. 7853 */ 7854 void *netdev_lower_get_first_private_rcu(struct net_device *dev) 7855 { 7856 struct netdev_adjacent *lower; 7857 7858 lower = list_first_or_null_rcu(&dev->adj_list.lower, 7859 struct netdev_adjacent, list); 7860 if (lower) 7861 return lower->private; 7862 return NULL; 7863 } 7864 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); 7865 7866 /** 7867 * netdev_master_upper_dev_get_rcu - Get master upper device 7868 * @dev: device 7869 * 7870 * Find a master upper device and return pointer to it or NULL in case 7871 * it's not there. The caller must hold the RCU read lock. 7872 */ 7873 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) 7874 { 7875 struct netdev_adjacent *upper; 7876 7877 upper = list_first_or_null_rcu(&dev->adj_list.upper, 7878 struct netdev_adjacent, list); 7879 if (upper && likely(upper->master)) 7880 return upper->dev; 7881 return NULL; 7882 } 7883 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); 7884 7885 static int netdev_adjacent_sysfs_add(struct net_device *dev, 7886 struct net_device *adj_dev, 7887 struct list_head *dev_list) 7888 { 7889 char linkname[IFNAMSIZ+7]; 7890 7891 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7892 "upper_%s" : "lower_%s", adj_dev->name); 7893 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), 7894 linkname); 7895 } 7896 static void netdev_adjacent_sysfs_del(struct net_device *dev, 7897 char *name, 7898 struct list_head *dev_list) 7899 { 7900 char linkname[IFNAMSIZ+7]; 7901 7902 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7903 "upper_%s" : "lower_%s", name); 7904 sysfs_remove_link(&(dev->dev.kobj), linkname); 7905 } 7906 7907 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, 7908 struct net_device *adj_dev, 7909 struct list_head *dev_list) 7910 { 7911 return (dev_list == &dev->adj_list.upper || 7912 dev_list == &dev->adj_list.lower) && 7913 net_eq(dev_net(dev), dev_net(adj_dev)); 7914 } 7915 7916 static int __netdev_adjacent_dev_insert(struct net_device *dev, 7917 struct net_device *adj_dev, 7918 struct list_head *dev_list, 7919 void *private, bool master) 7920 { 7921 struct netdev_adjacent *adj; 7922 int ret; 7923 7924 adj = __netdev_find_adj(adj_dev, dev_list); 7925 7926 if (adj) { 7927 adj->ref_nr += 1; 7928 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", 7929 dev->name, adj_dev->name, adj->ref_nr); 7930 7931 return 0; 7932 } 7933 7934 adj = kmalloc(sizeof(*adj), GFP_KERNEL); 7935 if (!adj) 7936 return -ENOMEM; 7937 7938 adj->dev = adj_dev; 7939 adj->master = master; 7940 adj->ref_nr = 1; 7941 adj->private = private; 7942 adj->ignore = false; 7943 dev_hold(adj_dev); 7944 7945 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", 7946 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); 7947 7948 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { 7949 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); 7950 if (ret) 7951 goto free_adj; 7952 } 7953 7954 /* Ensure that master link is always the first item in list. */ 7955 if (master) { 7956 ret = sysfs_create_link(&(dev->dev.kobj), 7957 &(adj_dev->dev.kobj), "master"); 7958 if (ret) 7959 goto remove_symlinks; 7960 7961 list_add_rcu(&adj->list, dev_list); 7962 } else { 7963 list_add_tail_rcu(&adj->list, dev_list); 7964 } 7965 7966 return 0; 7967 7968 remove_symlinks: 7969 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7970 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7971 free_adj: 7972 kfree(adj); 7973 dev_put(adj_dev); 7974 7975 return ret; 7976 } 7977 7978 static void __netdev_adjacent_dev_remove(struct net_device *dev, 7979 struct net_device *adj_dev, 7980 u16 ref_nr, 7981 struct list_head *dev_list) 7982 { 7983 struct netdev_adjacent *adj; 7984 7985 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", 7986 dev->name, adj_dev->name, ref_nr); 7987 7988 adj = __netdev_find_adj(adj_dev, dev_list); 7989 7990 if (!adj) { 7991 pr_err("Adjacency does not exist for device %s from %s\n", 7992 dev->name, adj_dev->name); 7993 WARN_ON(1); 7994 return; 7995 } 7996 7997 if (adj->ref_nr > ref_nr) { 7998 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", 7999 dev->name, adj_dev->name, ref_nr, 8000 adj->ref_nr - ref_nr); 8001 adj->ref_nr -= ref_nr; 8002 return; 8003 } 8004 8005 if (adj->master) 8006 sysfs_remove_link(&(dev->dev.kobj), "master"); 8007 8008 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 8009 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 8010 8011 list_del_rcu(&adj->list); 8012 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", 8013 adj_dev->name, dev->name, adj_dev->name); 8014 dev_put(adj_dev); 8015 kfree_rcu(adj, rcu); 8016 } 8017 8018 static int __netdev_adjacent_dev_link_lists(struct net_device *dev, 8019 struct net_device *upper_dev, 8020 struct list_head *up_list, 8021 struct list_head *down_list, 8022 void *private, bool master) 8023 { 8024 int ret; 8025 8026 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, 8027 private, master); 8028 if (ret) 8029 return ret; 8030 8031 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, 8032 private, false); 8033 if (ret) { 8034 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); 8035 return ret; 8036 } 8037 8038 return 0; 8039 } 8040 8041 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, 8042 struct net_device *upper_dev, 8043 u16 ref_nr, 8044 struct list_head *up_list, 8045 struct list_head *down_list) 8046 { 8047 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); 8048 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); 8049 } 8050 8051 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, 8052 struct net_device *upper_dev, 8053 void *private, bool master) 8054 { 8055 return __netdev_adjacent_dev_link_lists(dev, upper_dev, 8056 &dev->adj_list.upper, 8057 &upper_dev->adj_list.lower, 8058 private, master); 8059 } 8060 8061 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, 8062 struct net_device *upper_dev) 8063 { 8064 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, 8065 &dev->adj_list.upper, 8066 &upper_dev->adj_list.lower); 8067 } 8068 8069 static int __netdev_upper_dev_link(struct net_device *dev, 8070 struct net_device *upper_dev, bool master, 8071 void *upper_priv, void *upper_info, 8072 struct netdev_nested_priv *priv, 8073 struct netlink_ext_ack *extack) 8074 { 8075 struct netdev_notifier_changeupper_info changeupper_info = { 8076 .info = { 8077 .dev = dev, 8078 .extack = extack, 8079 }, 8080 .upper_dev = upper_dev, 8081 .master = master, 8082 .linking = true, 8083 .upper_info = upper_info, 8084 }; 8085 struct net_device *master_dev; 8086 int ret = 0; 8087 8088 ASSERT_RTNL(); 8089 8090 if (dev == upper_dev) 8091 return -EBUSY; 8092 8093 /* To prevent loops, check if dev is not upper device to upper_dev. */ 8094 if (__netdev_has_upper_dev(upper_dev, dev)) 8095 return -EBUSY; 8096 8097 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV) 8098 return -EMLINK; 8099 8100 if (!master) { 8101 if (__netdev_has_upper_dev(dev, upper_dev)) 8102 return -EEXIST; 8103 } else { 8104 master_dev = __netdev_master_upper_dev_get(dev); 8105 if (master_dev) 8106 return master_dev == upper_dev ? -EEXIST : -EBUSY; 8107 } 8108 8109 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 8110 &changeupper_info.info); 8111 ret = notifier_to_errno(ret); 8112 if (ret) 8113 return ret; 8114 8115 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, 8116 master); 8117 if (ret) 8118 return ret; 8119 8120 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 8121 &changeupper_info.info); 8122 ret = notifier_to_errno(ret); 8123 if (ret) 8124 goto rollback; 8125 8126 __netdev_update_upper_level(dev, NULL); 8127 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 8128 8129 __netdev_update_lower_level(upper_dev, priv); 8130 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 8131 priv); 8132 8133 return 0; 8134 8135 rollback: 8136 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 8137 8138 return ret; 8139 } 8140 8141 /** 8142 * netdev_upper_dev_link - Add a link to the upper device 8143 * @dev: device 8144 * @upper_dev: new upper device 8145 * @extack: netlink extended ack 8146 * 8147 * Adds a link to device which is upper to this one. The caller must hold 8148 * the RTNL lock. On a failure a negative errno code is returned. 8149 * On success the reference counts are adjusted and the function 8150 * returns zero. 8151 */ 8152 int netdev_upper_dev_link(struct net_device *dev, 8153 struct net_device *upper_dev, 8154 struct netlink_ext_ack *extack) 8155 { 8156 struct netdev_nested_priv priv = { 8157 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 8158 .data = NULL, 8159 }; 8160 8161 return __netdev_upper_dev_link(dev, upper_dev, false, 8162 NULL, NULL, &priv, extack); 8163 } 8164 EXPORT_SYMBOL(netdev_upper_dev_link); 8165 8166 /** 8167 * netdev_master_upper_dev_link - Add a master link to the upper device 8168 * @dev: device 8169 * @upper_dev: new upper device 8170 * @upper_priv: upper device private 8171 * @upper_info: upper info to be passed down via notifier 8172 * @extack: netlink extended ack 8173 * 8174 * Adds a link to device which is upper to this one. In this case, only 8175 * one master upper device can be linked, although other non-master devices 8176 * might be linked as well. The caller must hold the RTNL lock. 8177 * On a failure a negative errno code is returned. On success the reference 8178 * counts are adjusted and the function returns zero. 8179 */ 8180 int netdev_master_upper_dev_link(struct net_device *dev, 8181 struct net_device *upper_dev, 8182 void *upper_priv, void *upper_info, 8183 struct netlink_ext_ack *extack) 8184 { 8185 struct netdev_nested_priv priv = { 8186 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 8187 .data = NULL, 8188 }; 8189 8190 return __netdev_upper_dev_link(dev, upper_dev, true, 8191 upper_priv, upper_info, &priv, extack); 8192 } 8193 EXPORT_SYMBOL(netdev_master_upper_dev_link); 8194 8195 static void __netdev_upper_dev_unlink(struct net_device *dev, 8196 struct net_device *upper_dev, 8197 struct netdev_nested_priv *priv) 8198 { 8199 struct netdev_notifier_changeupper_info changeupper_info = { 8200 .info = { 8201 .dev = dev, 8202 }, 8203 .upper_dev = upper_dev, 8204 .linking = false, 8205 }; 8206 8207 ASSERT_RTNL(); 8208 8209 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; 8210 8211 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 8212 &changeupper_info.info); 8213 8214 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 8215 8216 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 8217 &changeupper_info.info); 8218 8219 __netdev_update_upper_level(dev, NULL); 8220 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 8221 8222 __netdev_update_lower_level(upper_dev, priv); 8223 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 8224 priv); 8225 } 8226 8227 /** 8228 * netdev_upper_dev_unlink - Removes a link to upper device 8229 * @dev: device 8230 * @upper_dev: new upper device 8231 * 8232 * Removes a link to device which is upper to this one. The caller must hold 8233 * the RTNL lock. 8234 */ 8235 void netdev_upper_dev_unlink(struct net_device *dev, 8236 struct net_device *upper_dev) 8237 { 8238 struct netdev_nested_priv priv = { 8239 .flags = NESTED_SYNC_TODO, 8240 .data = NULL, 8241 }; 8242 8243 __netdev_upper_dev_unlink(dev, upper_dev, &priv); 8244 } 8245 EXPORT_SYMBOL(netdev_upper_dev_unlink); 8246 8247 static void __netdev_adjacent_dev_set(struct net_device *upper_dev, 8248 struct net_device *lower_dev, 8249 bool val) 8250 { 8251 struct netdev_adjacent *adj; 8252 8253 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower); 8254 if (adj) 8255 adj->ignore = val; 8256 8257 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper); 8258 if (adj) 8259 adj->ignore = val; 8260 } 8261 8262 static void netdev_adjacent_dev_disable(struct net_device *upper_dev, 8263 struct net_device *lower_dev) 8264 { 8265 __netdev_adjacent_dev_set(upper_dev, lower_dev, true); 8266 } 8267 8268 static void netdev_adjacent_dev_enable(struct net_device *upper_dev, 8269 struct net_device *lower_dev) 8270 { 8271 __netdev_adjacent_dev_set(upper_dev, lower_dev, false); 8272 } 8273 8274 int netdev_adjacent_change_prepare(struct net_device *old_dev, 8275 struct net_device *new_dev, 8276 struct net_device *dev, 8277 struct netlink_ext_ack *extack) 8278 { 8279 struct netdev_nested_priv priv = { 8280 .flags = 0, 8281 .data = NULL, 8282 }; 8283 int err; 8284 8285 if (!new_dev) 8286 return 0; 8287 8288 if (old_dev && new_dev != old_dev) 8289 netdev_adjacent_dev_disable(dev, old_dev); 8290 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv, 8291 extack); 8292 if (err) { 8293 if (old_dev && new_dev != old_dev) 8294 netdev_adjacent_dev_enable(dev, old_dev); 8295 return err; 8296 } 8297 8298 return 0; 8299 } 8300 EXPORT_SYMBOL(netdev_adjacent_change_prepare); 8301 8302 void netdev_adjacent_change_commit(struct net_device *old_dev, 8303 struct net_device *new_dev, 8304 struct net_device *dev) 8305 { 8306 struct netdev_nested_priv priv = { 8307 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 8308 .data = NULL, 8309 }; 8310 8311 if (!new_dev || !old_dev) 8312 return; 8313 8314 if (new_dev == old_dev) 8315 return; 8316 8317 netdev_adjacent_dev_enable(dev, old_dev); 8318 __netdev_upper_dev_unlink(old_dev, dev, &priv); 8319 } 8320 EXPORT_SYMBOL(netdev_adjacent_change_commit); 8321 8322 void netdev_adjacent_change_abort(struct net_device *old_dev, 8323 struct net_device *new_dev, 8324 struct net_device *dev) 8325 { 8326 struct netdev_nested_priv priv = { 8327 .flags = 0, 8328 .data = NULL, 8329 }; 8330 8331 if (!new_dev) 8332 return; 8333 8334 if (old_dev && new_dev != old_dev) 8335 netdev_adjacent_dev_enable(dev, old_dev); 8336 8337 __netdev_upper_dev_unlink(new_dev, dev, &priv); 8338 } 8339 EXPORT_SYMBOL(netdev_adjacent_change_abort); 8340 8341 /** 8342 * netdev_bonding_info_change - Dispatch event about slave change 8343 * @dev: device 8344 * @bonding_info: info to dispatch 8345 * 8346 * Send NETDEV_BONDING_INFO to netdev notifiers with info. 8347 * The caller must hold the RTNL lock. 8348 */ 8349 void netdev_bonding_info_change(struct net_device *dev, 8350 struct netdev_bonding_info *bonding_info) 8351 { 8352 struct netdev_notifier_bonding_info info = { 8353 .info.dev = dev, 8354 }; 8355 8356 memcpy(&info.bonding_info, bonding_info, 8357 sizeof(struct netdev_bonding_info)); 8358 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, 8359 &info.info); 8360 } 8361 EXPORT_SYMBOL(netdev_bonding_info_change); 8362 8363 /** 8364 * netdev_get_xmit_slave - Get the xmit slave of master device 8365 * @dev: device 8366 * @skb: The packet 8367 * @all_slaves: assume all the slaves are active 8368 * 8369 * The reference counters are not incremented so the caller must be 8370 * careful with locks. The caller must hold RCU lock. 8371 * %NULL is returned if no slave is found. 8372 */ 8373 8374 struct net_device *netdev_get_xmit_slave(struct net_device *dev, 8375 struct sk_buff *skb, 8376 bool all_slaves) 8377 { 8378 const struct net_device_ops *ops = dev->netdev_ops; 8379 8380 if (!ops->ndo_get_xmit_slave) 8381 return NULL; 8382 return ops->ndo_get_xmit_slave(dev, skb, all_slaves); 8383 } 8384 EXPORT_SYMBOL(netdev_get_xmit_slave); 8385 8386 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev, 8387 struct sock *sk) 8388 { 8389 const struct net_device_ops *ops = dev->netdev_ops; 8390 8391 if (!ops->ndo_sk_get_lower_dev) 8392 return NULL; 8393 return ops->ndo_sk_get_lower_dev(dev, sk); 8394 } 8395 8396 /** 8397 * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket 8398 * @dev: device 8399 * @sk: the socket 8400 * 8401 * %NULL is returned if no lower device is found. 8402 */ 8403 8404 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev, 8405 struct sock *sk) 8406 { 8407 struct net_device *lower; 8408 8409 lower = netdev_sk_get_lower_dev(dev, sk); 8410 while (lower) { 8411 dev = lower; 8412 lower = netdev_sk_get_lower_dev(dev, sk); 8413 } 8414 8415 return dev; 8416 } 8417 EXPORT_SYMBOL(netdev_sk_get_lowest_dev); 8418 8419 static void netdev_adjacent_add_links(struct net_device *dev) 8420 { 8421 struct netdev_adjacent *iter; 8422 8423 struct net *net = dev_net(dev); 8424 8425 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8426 if (!net_eq(net, dev_net(iter->dev))) 8427 continue; 8428 netdev_adjacent_sysfs_add(iter->dev, dev, 8429 &iter->dev->adj_list.lower); 8430 netdev_adjacent_sysfs_add(dev, iter->dev, 8431 &dev->adj_list.upper); 8432 } 8433 8434 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8435 if (!net_eq(net, dev_net(iter->dev))) 8436 continue; 8437 netdev_adjacent_sysfs_add(iter->dev, dev, 8438 &iter->dev->adj_list.upper); 8439 netdev_adjacent_sysfs_add(dev, iter->dev, 8440 &dev->adj_list.lower); 8441 } 8442 } 8443 8444 static void netdev_adjacent_del_links(struct net_device *dev) 8445 { 8446 struct netdev_adjacent *iter; 8447 8448 struct net *net = dev_net(dev); 8449 8450 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8451 if (!net_eq(net, dev_net(iter->dev))) 8452 continue; 8453 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8454 &iter->dev->adj_list.lower); 8455 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8456 &dev->adj_list.upper); 8457 } 8458 8459 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8460 if (!net_eq(net, dev_net(iter->dev))) 8461 continue; 8462 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8463 &iter->dev->adj_list.upper); 8464 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8465 &dev->adj_list.lower); 8466 } 8467 } 8468 8469 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) 8470 { 8471 struct netdev_adjacent *iter; 8472 8473 struct net *net = dev_net(dev); 8474 8475 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8476 if (!net_eq(net, dev_net(iter->dev))) 8477 continue; 8478 netdev_adjacent_sysfs_del(iter->dev, oldname, 8479 &iter->dev->adj_list.lower); 8480 netdev_adjacent_sysfs_add(iter->dev, dev, 8481 &iter->dev->adj_list.lower); 8482 } 8483 8484 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8485 if (!net_eq(net, dev_net(iter->dev))) 8486 continue; 8487 netdev_adjacent_sysfs_del(iter->dev, oldname, 8488 &iter->dev->adj_list.upper); 8489 netdev_adjacent_sysfs_add(iter->dev, dev, 8490 &iter->dev->adj_list.upper); 8491 } 8492 } 8493 8494 void *netdev_lower_dev_get_private(struct net_device *dev, 8495 struct net_device *lower_dev) 8496 { 8497 struct netdev_adjacent *lower; 8498 8499 if (!lower_dev) 8500 return NULL; 8501 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); 8502 if (!lower) 8503 return NULL; 8504 8505 return lower->private; 8506 } 8507 EXPORT_SYMBOL(netdev_lower_dev_get_private); 8508 8509 8510 /** 8511 * netdev_lower_state_changed - Dispatch event about lower device state change 8512 * @lower_dev: device 8513 * @lower_state_info: state to dispatch 8514 * 8515 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. 8516 * The caller must hold the RTNL lock. 8517 */ 8518 void netdev_lower_state_changed(struct net_device *lower_dev, 8519 void *lower_state_info) 8520 { 8521 struct netdev_notifier_changelowerstate_info changelowerstate_info = { 8522 .info.dev = lower_dev, 8523 }; 8524 8525 ASSERT_RTNL(); 8526 changelowerstate_info.lower_state_info = lower_state_info; 8527 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, 8528 &changelowerstate_info.info); 8529 } 8530 EXPORT_SYMBOL(netdev_lower_state_changed); 8531 8532 static void dev_change_rx_flags(struct net_device *dev, int flags) 8533 { 8534 const struct net_device_ops *ops = dev->netdev_ops; 8535 8536 if (ops->ndo_change_rx_flags) 8537 ops->ndo_change_rx_flags(dev, flags); 8538 } 8539 8540 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) 8541 { 8542 unsigned int old_flags = dev->flags; 8543 kuid_t uid; 8544 kgid_t gid; 8545 8546 ASSERT_RTNL(); 8547 8548 dev->flags |= IFF_PROMISC; 8549 dev->promiscuity += inc; 8550 if (dev->promiscuity == 0) { 8551 /* 8552 * Avoid overflow. 8553 * If inc causes overflow, untouch promisc and return error. 8554 */ 8555 if (inc < 0) 8556 dev->flags &= ~IFF_PROMISC; 8557 else { 8558 dev->promiscuity -= inc; 8559 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n", 8560 dev->name); 8561 return -EOVERFLOW; 8562 } 8563 } 8564 if (dev->flags != old_flags) { 8565 pr_info("device %s %s promiscuous mode\n", 8566 dev->name, 8567 dev->flags & IFF_PROMISC ? "entered" : "left"); 8568 if (audit_enabled) { 8569 current_uid_gid(&uid, &gid); 8570 audit_log(audit_context(), GFP_ATOMIC, 8571 AUDIT_ANOM_PROMISCUOUS, 8572 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", 8573 dev->name, (dev->flags & IFF_PROMISC), 8574 (old_flags & IFF_PROMISC), 8575 from_kuid(&init_user_ns, audit_get_loginuid(current)), 8576 from_kuid(&init_user_ns, uid), 8577 from_kgid(&init_user_ns, gid), 8578 audit_get_sessionid(current)); 8579 } 8580 8581 dev_change_rx_flags(dev, IFF_PROMISC); 8582 } 8583 if (notify) 8584 __dev_notify_flags(dev, old_flags, IFF_PROMISC); 8585 return 0; 8586 } 8587 8588 /** 8589 * dev_set_promiscuity - update promiscuity count on a device 8590 * @dev: device 8591 * @inc: modifier 8592 * 8593 * Add or remove promiscuity from a device. While the count in the device 8594 * remains above zero the interface remains promiscuous. Once it hits zero 8595 * the device reverts back to normal filtering operation. A negative inc 8596 * value is used to drop promiscuity on the device. 8597 * Return 0 if successful or a negative errno code on error. 8598 */ 8599 int dev_set_promiscuity(struct net_device *dev, int inc) 8600 { 8601 unsigned int old_flags = dev->flags; 8602 int err; 8603 8604 err = __dev_set_promiscuity(dev, inc, true); 8605 if (err < 0) 8606 return err; 8607 if (dev->flags != old_flags) 8608 dev_set_rx_mode(dev); 8609 return err; 8610 } 8611 EXPORT_SYMBOL(dev_set_promiscuity); 8612 8613 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) 8614 { 8615 unsigned int old_flags = dev->flags, old_gflags = dev->gflags; 8616 8617 ASSERT_RTNL(); 8618 8619 dev->flags |= IFF_ALLMULTI; 8620 dev->allmulti += inc; 8621 if (dev->allmulti == 0) { 8622 /* 8623 * Avoid overflow. 8624 * If inc causes overflow, untouch allmulti and return error. 8625 */ 8626 if (inc < 0) 8627 dev->flags &= ~IFF_ALLMULTI; 8628 else { 8629 dev->allmulti -= inc; 8630 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n", 8631 dev->name); 8632 return -EOVERFLOW; 8633 } 8634 } 8635 if (dev->flags ^ old_flags) { 8636 dev_change_rx_flags(dev, IFF_ALLMULTI); 8637 dev_set_rx_mode(dev); 8638 if (notify) 8639 __dev_notify_flags(dev, old_flags, 8640 dev->gflags ^ old_gflags); 8641 } 8642 return 0; 8643 } 8644 8645 /** 8646 * dev_set_allmulti - update allmulti count on a device 8647 * @dev: device 8648 * @inc: modifier 8649 * 8650 * Add or remove reception of all multicast frames to a device. While the 8651 * count in the device remains above zero the interface remains listening 8652 * to all interfaces. Once it hits zero the device reverts back to normal 8653 * filtering operation. A negative @inc value is used to drop the counter 8654 * when releasing a resource needing all multicasts. 8655 * Return 0 if successful or a negative errno code on error. 8656 */ 8657 8658 int dev_set_allmulti(struct net_device *dev, int inc) 8659 { 8660 return __dev_set_allmulti(dev, inc, true); 8661 } 8662 EXPORT_SYMBOL(dev_set_allmulti); 8663 8664 /* 8665 * Upload unicast and multicast address lists to device and 8666 * configure RX filtering. When the device doesn't support unicast 8667 * filtering it is put in promiscuous mode while unicast addresses 8668 * are present. 8669 */ 8670 void __dev_set_rx_mode(struct net_device *dev) 8671 { 8672 const struct net_device_ops *ops = dev->netdev_ops; 8673 8674 /* dev_open will call this function so the list will stay sane. */ 8675 if (!(dev->flags&IFF_UP)) 8676 return; 8677 8678 if (!netif_device_present(dev)) 8679 return; 8680 8681 if (!(dev->priv_flags & IFF_UNICAST_FLT)) { 8682 /* Unicast addresses changes may only happen under the rtnl, 8683 * therefore calling __dev_set_promiscuity here is safe. 8684 */ 8685 if (!netdev_uc_empty(dev) && !dev->uc_promisc) { 8686 __dev_set_promiscuity(dev, 1, false); 8687 dev->uc_promisc = true; 8688 } else if (netdev_uc_empty(dev) && dev->uc_promisc) { 8689 __dev_set_promiscuity(dev, -1, false); 8690 dev->uc_promisc = false; 8691 } 8692 } 8693 8694 if (ops->ndo_set_rx_mode) 8695 ops->ndo_set_rx_mode(dev); 8696 } 8697 8698 void dev_set_rx_mode(struct net_device *dev) 8699 { 8700 netif_addr_lock_bh(dev); 8701 __dev_set_rx_mode(dev); 8702 netif_addr_unlock_bh(dev); 8703 } 8704 8705 /** 8706 * dev_get_flags - get flags reported to userspace 8707 * @dev: device 8708 * 8709 * Get the combination of flag bits exported through APIs to userspace. 8710 */ 8711 unsigned int dev_get_flags(const struct net_device *dev) 8712 { 8713 unsigned int flags; 8714 8715 flags = (dev->flags & ~(IFF_PROMISC | 8716 IFF_ALLMULTI | 8717 IFF_RUNNING | 8718 IFF_LOWER_UP | 8719 IFF_DORMANT)) | 8720 (dev->gflags & (IFF_PROMISC | 8721 IFF_ALLMULTI)); 8722 8723 if (netif_running(dev)) { 8724 if (netif_oper_up(dev)) 8725 flags |= IFF_RUNNING; 8726 if (netif_carrier_ok(dev)) 8727 flags |= IFF_LOWER_UP; 8728 if (netif_dormant(dev)) 8729 flags |= IFF_DORMANT; 8730 } 8731 8732 return flags; 8733 } 8734 EXPORT_SYMBOL(dev_get_flags); 8735 8736 int __dev_change_flags(struct net_device *dev, unsigned int flags, 8737 struct netlink_ext_ack *extack) 8738 { 8739 unsigned int old_flags = dev->flags; 8740 int ret; 8741 8742 ASSERT_RTNL(); 8743 8744 /* 8745 * Set the flags on our device. 8746 */ 8747 8748 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | 8749 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | 8750 IFF_AUTOMEDIA)) | 8751 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | 8752 IFF_ALLMULTI)); 8753 8754 /* 8755 * Load in the correct multicast list now the flags have changed. 8756 */ 8757 8758 if ((old_flags ^ flags) & IFF_MULTICAST) 8759 dev_change_rx_flags(dev, IFF_MULTICAST); 8760 8761 dev_set_rx_mode(dev); 8762 8763 /* 8764 * Have we downed the interface. We handle IFF_UP ourselves 8765 * according to user attempts to set it, rather than blindly 8766 * setting it. 8767 */ 8768 8769 ret = 0; 8770 if ((old_flags ^ flags) & IFF_UP) { 8771 if (old_flags & IFF_UP) 8772 __dev_close(dev); 8773 else 8774 ret = __dev_open(dev, extack); 8775 } 8776 8777 if ((flags ^ dev->gflags) & IFF_PROMISC) { 8778 int inc = (flags & IFF_PROMISC) ? 1 : -1; 8779 unsigned int old_flags = dev->flags; 8780 8781 dev->gflags ^= IFF_PROMISC; 8782 8783 if (__dev_set_promiscuity(dev, inc, false) >= 0) 8784 if (dev->flags != old_flags) 8785 dev_set_rx_mode(dev); 8786 } 8787 8788 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI 8789 * is important. Some (broken) drivers set IFF_PROMISC, when 8790 * IFF_ALLMULTI is requested not asking us and not reporting. 8791 */ 8792 if ((flags ^ dev->gflags) & IFF_ALLMULTI) { 8793 int inc = (flags & IFF_ALLMULTI) ? 1 : -1; 8794 8795 dev->gflags ^= IFF_ALLMULTI; 8796 __dev_set_allmulti(dev, inc, false); 8797 } 8798 8799 return ret; 8800 } 8801 8802 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, 8803 unsigned int gchanges) 8804 { 8805 unsigned int changes = dev->flags ^ old_flags; 8806 8807 if (gchanges) 8808 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC); 8809 8810 if (changes & IFF_UP) { 8811 if (dev->flags & IFF_UP) 8812 call_netdevice_notifiers(NETDEV_UP, dev); 8813 else 8814 call_netdevice_notifiers(NETDEV_DOWN, dev); 8815 } 8816 8817 if (dev->flags & IFF_UP && 8818 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { 8819 struct netdev_notifier_change_info change_info = { 8820 .info = { 8821 .dev = dev, 8822 }, 8823 .flags_changed = changes, 8824 }; 8825 8826 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info); 8827 } 8828 } 8829 8830 /** 8831 * dev_change_flags - change device settings 8832 * @dev: device 8833 * @flags: device state flags 8834 * @extack: netlink extended ack 8835 * 8836 * Change settings on device based state flags. The flags are 8837 * in the userspace exported format. 8838 */ 8839 int dev_change_flags(struct net_device *dev, unsigned int flags, 8840 struct netlink_ext_ack *extack) 8841 { 8842 int ret; 8843 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; 8844 8845 ret = __dev_change_flags(dev, flags, extack); 8846 if (ret < 0) 8847 return ret; 8848 8849 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); 8850 __dev_notify_flags(dev, old_flags, changes); 8851 return ret; 8852 } 8853 EXPORT_SYMBOL(dev_change_flags); 8854 8855 int __dev_set_mtu(struct net_device *dev, int new_mtu) 8856 { 8857 const struct net_device_ops *ops = dev->netdev_ops; 8858 8859 if (ops->ndo_change_mtu) 8860 return ops->ndo_change_mtu(dev, new_mtu); 8861 8862 /* Pairs with all the lockless reads of dev->mtu in the stack */ 8863 WRITE_ONCE(dev->mtu, new_mtu); 8864 return 0; 8865 } 8866 EXPORT_SYMBOL(__dev_set_mtu); 8867 8868 int dev_validate_mtu(struct net_device *dev, int new_mtu, 8869 struct netlink_ext_ack *extack) 8870 { 8871 /* MTU must be positive, and in range */ 8872 if (new_mtu < 0 || new_mtu < dev->min_mtu) { 8873 NL_SET_ERR_MSG(extack, "mtu less than device minimum"); 8874 return -EINVAL; 8875 } 8876 8877 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { 8878 NL_SET_ERR_MSG(extack, "mtu greater than device maximum"); 8879 return -EINVAL; 8880 } 8881 return 0; 8882 } 8883 8884 /** 8885 * dev_set_mtu_ext - Change maximum transfer unit 8886 * @dev: device 8887 * @new_mtu: new transfer unit 8888 * @extack: netlink extended ack 8889 * 8890 * Change the maximum transfer size of the network device. 8891 */ 8892 int dev_set_mtu_ext(struct net_device *dev, int new_mtu, 8893 struct netlink_ext_ack *extack) 8894 { 8895 int err, orig_mtu; 8896 8897 if (new_mtu == dev->mtu) 8898 return 0; 8899 8900 err = dev_validate_mtu(dev, new_mtu, extack); 8901 if (err) 8902 return err; 8903 8904 if (!netif_device_present(dev)) 8905 return -ENODEV; 8906 8907 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); 8908 err = notifier_to_errno(err); 8909 if (err) 8910 return err; 8911 8912 orig_mtu = dev->mtu; 8913 err = __dev_set_mtu(dev, new_mtu); 8914 8915 if (!err) { 8916 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8917 orig_mtu); 8918 err = notifier_to_errno(err); 8919 if (err) { 8920 /* setting mtu back and notifying everyone again, 8921 * so that they have a chance to revert changes. 8922 */ 8923 __dev_set_mtu(dev, orig_mtu); 8924 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8925 new_mtu); 8926 } 8927 } 8928 return err; 8929 } 8930 8931 int dev_set_mtu(struct net_device *dev, int new_mtu) 8932 { 8933 struct netlink_ext_ack extack; 8934 int err; 8935 8936 memset(&extack, 0, sizeof(extack)); 8937 err = dev_set_mtu_ext(dev, new_mtu, &extack); 8938 if (err && extack._msg) 8939 net_err_ratelimited("%s: %s\n", dev->name, extack._msg); 8940 return err; 8941 } 8942 EXPORT_SYMBOL(dev_set_mtu); 8943 8944 /** 8945 * dev_change_tx_queue_len - Change TX queue length of a netdevice 8946 * @dev: device 8947 * @new_len: new tx queue length 8948 */ 8949 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len) 8950 { 8951 unsigned int orig_len = dev->tx_queue_len; 8952 int res; 8953 8954 if (new_len != (unsigned int)new_len) 8955 return -ERANGE; 8956 8957 if (new_len != orig_len) { 8958 dev->tx_queue_len = new_len; 8959 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev); 8960 res = notifier_to_errno(res); 8961 if (res) 8962 goto err_rollback; 8963 res = dev_qdisc_change_tx_queue_len(dev); 8964 if (res) 8965 goto err_rollback; 8966 } 8967 8968 return 0; 8969 8970 err_rollback: 8971 netdev_err(dev, "refused to change device tx_queue_len\n"); 8972 dev->tx_queue_len = orig_len; 8973 return res; 8974 } 8975 8976 /** 8977 * dev_set_group - Change group this device belongs to 8978 * @dev: device 8979 * @new_group: group this device should belong to 8980 */ 8981 void dev_set_group(struct net_device *dev, int new_group) 8982 { 8983 dev->group = new_group; 8984 } 8985 EXPORT_SYMBOL(dev_set_group); 8986 8987 /** 8988 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR. 8989 * @dev: device 8990 * @addr: new address 8991 * @extack: netlink extended ack 8992 */ 8993 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr, 8994 struct netlink_ext_ack *extack) 8995 { 8996 struct netdev_notifier_pre_changeaddr_info info = { 8997 .info.dev = dev, 8998 .info.extack = extack, 8999 .dev_addr = addr, 9000 }; 9001 int rc; 9002 9003 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info); 9004 return notifier_to_errno(rc); 9005 } 9006 EXPORT_SYMBOL(dev_pre_changeaddr_notify); 9007 9008 /** 9009 * dev_set_mac_address - Change Media Access Control Address 9010 * @dev: device 9011 * @sa: new address 9012 * @extack: netlink extended ack 9013 * 9014 * Change the hardware (MAC) address of the device 9015 */ 9016 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa, 9017 struct netlink_ext_ack *extack) 9018 { 9019 const struct net_device_ops *ops = dev->netdev_ops; 9020 int err; 9021 9022 if (!ops->ndo_set_mac_address) 9023 return -EOPNOTSUPP; 9024 if (sa->sa_family != dev->type) 9025 return -EINVAL; 9026 if (!netif_device_present(dev)) 9027 return -ENODEV; 9028 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack); 9029 if (err) 9030 return err; 9031 err = ops->ndo_set_mac_address(dev, sa); 9032 if (err) 9033 return err; 9034 dev->addr_assign_type = NET_ADDR_SET; 9035 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); 9036 add_device_randomness(dev->dev_addr, dev->addr_len); 9037 return 0; 9038 } 9039 EXPORT_SYMBOL(dev_set_mac_address); 9040 9041 static DECLARE_RWSEM(dev_addr_sem); 9042 9043 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa, 9044 struct netlink_ext_ack *extack) 9045 { 9046 int ret; 9047 9048 down_write(&dev_addr_sem); 9049 ret = dev_set_mac_address(dev, sa, extack); 9050 up_write(&dev_addr_sem); 9051 return ret; 9052 } 9053 EXPORT_SYMBOL(dev_set_mac_address_user); 9054 9055 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name) 9056 { 9057 size_t size = sizeof(sa->sa_data); 9058 struct net_device *dev; 9059 int ret = 0; 9060 9061 down_read(&dev_addr_sem); 9062 rcu_read_lock(); 9063 9064 dev = dev_get_by_name_rcu(net, dev_name); 9065 if (!dev) { 9066 ret = -ENODEV; 9067 goto unlock; 9068 } 9069 if (!dev->addr_len) 9070 memset(sa->sa_data, 0, size); 9071 else 9072 memcpy(sa->sa_data, dev->dev_addr, 9073 min_t(size_t, size, dev->addr_len)); 9074 sa->sa_family = dev->type; 9075 9076 unlock: 9077 rcu_read_unlock(); 9078 up_read(&dev_addr_sem); 9079 return ret; 9080 } 9081 EXPORT_SYMBOL(dev_get_mac_address); 9082 9083 /** 9084 * dev_change_carrier - Change device carrier 9085 * @dev: device 9086 * @new_carrier: new value 9087 * 9088 * Change device carrier 9089 */ 9090 int dev_change_carrier(struct net_device *dev, bool new_carrier) 9091 { 9092 const struct net_device_ops *ops = dev->netdev_ops; 9093 9094 if (!ops->ndo_change_carrier) 9095 return -EOPNOTSUPP; 9096 if (!netif_device_present(dev)) 9097 return -ENODEV; 9098 return ops->ndo_change_carrier(dev, new_carrier); 9099 } 9100 EXPORT_SYMBOL(dev_change_carrier); 9101 9102 /** 9103 * dev_get_phys_port_id - Get device physical port ID 9104 * @dev: device 9105 * @ppid: port ID 9106 * 9107 * Get device physical port ID 9108 */ 9109 int dev_get_phys_port_id(struct net_device *dev, 9110 struct netdev_phys_item_id *ppid) 9111 { 9112 const struct net_device_ops *ops = dev->netdev_ops; 9113 9114 if (!ops->ndo_get_phys_port_id) 9115 return -EOPNOTSUPP; 9116 return ops->ndo_get_phys_port_id(dev, ppid); 9117 } 9118 EXPORT_SYMBOL(dev_get_phys_port_id); 9119 9120 /** 9121 * dev_get_phys_port_name - Get device physical port name 9122 * @dev: device 9123 * @name: port name 9124 * @len: limit of bytes to copy to name 9125 * 9126 * Get device physical port name 9127 */ 9128 int dev_get_phys_port_name(struct net_device *dev, 9129 char *name, size_t len) 9130 { 9131 const struct net_device_ops *ops = dev->netdev_ops; 9132 int err; 9133 9134 if (ops->ndo_get_phys_port_name) { 9135 err = ops->ndo_get_phys_port_name(dev, name, len); 9136 if (err != -EOPNOTSUPP) 9137 return err; 9138 } 9139 return devlink_compat_phys_port_name_get(dev, name, len); 9140 } 9141 EXPORT_SYMBOL(dev_get_phys_port_name); 9142 9143 /** 9144 * dev_get_port_parent_id - Get the device's port parent identifier 9145 * @dev: network device 9146 * @ppid: pointer to a storage for the port's parent identifier 9147 * @recurse: allow/disallow recursion to lower devices 9148 * 9149 * Get the devices's port parent identifier 9150 */ 9151 int dev_get_port_parent_id(struct net_device *dev, 9152 struct netdev_phys_item_id *ppid, 9153 bool recurse) 9154 { 9155 const struct net_device_ops *ops = dev->netdev_ops; 9156 struct netdev_phys_item_id first = { }; 9157 struct net_device *lower_dev; 9158 struct list_head *iter; 9159 int err; 9160 9161 if (ops->ndo_get_port_parent_id) { 9162 err = ops->ndo_get_port_parent_id(dev, ppid); 9163 if (err != -EOPNOTSUPP) 9164 return err; 9165 } 9166 9167 err = devlink_compat_switch_id_get(dev, ppid); 9168 if (!recurse || err != -EOPNOTSUPP) 9169 return err; 9170 9171 netdev_for_each_lower_dev(dev, lower_dev, iter) { 9172 err = dev_get_port_parent_id(lower_dev, ppid, true); 9173 if (err) 9174 break; 9175 if (!first.id_len) 9176 first = *ppid; 9177 else if (memcmp(&first, ppid, sizeof(*ppid))) 9178 return -EOPNOTSUPP; 9179 } 9180 9181 return err; 9182 } 9183 EXPORT_SYMBOL(dev_get_port_parent_id); 9184 9185 /** 9186 * netdev_port_same_parent_id - Indicate if two network devices have 9187 * the same port parent identifier 9188 * @a: first network device 9189 * @b: second network device 9190 */ 9191 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b) 9192 { 9193 struct netdev_phys_item_id a_id = { }; 9194 struct netdev_phys_item_id b_id = { }; 9195 9196 if (dev_get_port_parent_id(a, &a_id, true) || 9197 dev_get_port_parent_id(b, &b_id, true)) 9198 return false; 9199 9200 return netdev_phys_item_id_same(&a_id, &b_id); 9201 } 9202 EXPORT_SYMBOL(netdev_port_same_parent_id); 9203 9204 /** 9205 * dev_change_proto_down - update protocol port state information 9206 * @dev: device 9207 * @proto_down: new value 9208 * 9209 * This info can be used by switch drivers to set the phys state of the 9210 * port. 9211 */ 9212 int dev_change_proto_down(struct net_device *dev, bool proto_down) 9213 { 9214 const struct net_device_ops *ops = dev->netdev_ops; 9215 9216 if (!ops->ndo_change_proto_down) 9217 return -EOPNOTSUPP; 9218 if (!netif_device_present(dev)) 9219 return -ENODEV; 9220 return ops->ndo_change_proto_down(dev, proto_down); 9221 } 9222 EXPORT_SYMBOL(dev_change_proto_down); 9223 9224 /** 9225 * dev_change_proto_down_generic - generic implementation for 9226 * ndo_change_proto_down that sets carrier according to 9227 * proto_down. 9228 * 9229 * @dev: device 9230 * @proto_down: new value 9231 */ 9232 int dev_change_proto_down_generic(struct net_device *dev, bool proto_down) 9233 { 9234 if (proto_down) 9235 netif_carrier_off(dev); 9236 else 9237 netif_carrier_on(dev); 9238 dev->proto_down = proto_down; 9239 return 0; 9240 } 9241 EXPORT_SYMBOL(dev_change_proto_down_generic); 9242 9243 /** 9244 * dev_change_proto_down_reason - proto down reason 9245 * 9246 * @dev: device 9247 * @mask: proto down mask 9248 * @value: proto down value 9249 */ 9250 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask, 9251 u32 value) 9252 { 9253 int b; 9254 9255 if (!mask) { 9256 dev->proto_down_reason = value; 9257 } else { 9258 for_each_set_bit(b, &mask, 32) { 9259 if (value & (1 << b)) 9260 dev->proto_down_reason |= BIT(b); 9261 else 9262 dev->proto_down_reason &= ~BIT(b); 9263 } 9264 } 9265 } 9266 EXPORT_SYMBOL(dev_change_proto_down_reason); 9267 9268 struct bpf_xdp_link { 9269 struct bpf_link link; 9270 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */ 9271 int flags; 9272 }; 9273 9274 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags) 9275 { 9276 if (flags & XDP_FLAGS_HW_MODE) 9277 return XDP_MODE_HW; 9278 if (flags & XDP_FLAGS_DRV_MODE) 9279 return XDP_MODE_DRV; 9280 if (flags & XDP_FLAGS_SKB_MODE) 9281 return XDP_MODE_SKB; 9282 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB; 9283 } 9284 9285 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode) 9286 { 9287 switch (mode) { 9288 case XDP_MODE_SKB: 9289 return generic_xdp_install; 9290 case XDP_MODE_DRV: 9291 case XDP_MODE_HW: 9292 return dev->netdev_ops->ndo_bpf; 9293 default: 9294 return NULL; 9295 } 9296 } 9297 9298 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev, 9299 enum bpf_xdp_mode mode) 9300 { 9301 return dev->xdp_state[mode].link; 9302 } 9303 9304 static struct bpf_prog *dev_xdp_prog(struct net_device *dev, 9305 enum bpf_xdp_mode mode) 9306 { 9307 struct bpf_xdp_link *link = dev_xdp_link(dev, mode); 9308 9309 if (link) 9310 return link->link.prog; 9311 return dev->xdp_state[mode].prog; 9312 } 9313 9314 u8 dev_xdp_prog_count(struct net_device *dev) 9315 { 9316 u8 count = 0; 9317 int i; 9318 9319 for (i = 0; i < __MAX_XDP_MODE; i++) 9320 if (dev->xdp_state[i].prog || dev->xdp_state[i].link) 9321 count++; 9322 return count; 9323 } 9324 EXPORT_SYMBOL_GPL(dev_xdp_prog_count); 9325 9326 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode) 9327 { 9328 struct bpf_prog *prog = dev_xdp_prog(dev, mode); 9329 9330 return prog ? prog->aux->id : 0; 9331 } 9332 9333 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode, 9334 struct bpf_xdp_link *link) 9335 { 9336 dev->xdp_state[mode].link = link; 9337 dev->xdp_state[mode].prog = NULL; 9338 } 9339 9340 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode, 9341 struct bpf_prog *prog) 9342 { 9343 dev->xdp_state[mode].link = NULL; 9344 dev->xdp_state[mode].prog = prog; 9345 } 9346 9347 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode, 9348 bpf_op_t bpf_op, struct netlink_ext_ack *extack, 9349 u32 flags, struct bpf_prog *prog) 9350 { 9351 struct netdev_bpf xdp; 9352 int err; 9353 9354 memset(&xdp, 0, sizeof(xdp)); 9355 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG; 9356 xdp.extack = extack; 9357 xdp.flags = flags; 9358 xdp.prog = prog; 9359 9360 /* Drivers assume refcnt is already incremented (i.e, prog pointer is 9361 * "moved" into driver), so they don't increment it on their own, but 9362 * they do decrement refcnt when program is detached or replaced. 9363 * Given net_device also owns link/prog, we need to bump refcnt here 9364 * to prevent drivers from underflowing it. 9365 */ 9366 if (prog) 9367 bpf_prog_inc(prog); 9368 err = bpf_op(dev, &xdp); 9369 if (err) { 9370 if (prog) 9371 bpf_prog_put(prog); 9372 return err; 9373 } 9374 9375 if (mode != XDP_MODE_HW) 9376 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog); 9377 9378 return 0; 9379 } 9380 9381 static void dev_xdp_uninstall(struct net_device *dev) 9382 { 9383 struct bpf_xdp_link *link; 9384 struct bpf_prog *prog; 9385 enum bpf_xdp_mode mode; 9386 bpf_op_t bpf_op; 9387 9388 ASSERT_RTNL(); 9389 9390 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) { 9391 prog = dev_xdp_prog(dev, mode); 9392 if (!prog) 9393 continue; 9394 9395 bpf_op = dev_xdp_bpf_op(dev, mode); 9396 if (!bpf_op) 9397 continue; 9398 9399 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9400 9401 /* auto-detach link from net device */ 9402 link = dev_xdp_link(dev, mode); 9403 if (link) 9404 link->dev = NULL; 9405 else 9406 bpf_prog_put(prog); 9407 9408 dev_xdp_set_link(dev, mode, NULL); 9409 } 9410 } 9411 9412 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack, 9413 struct bpf_xdp_link *link, struct bpf_prog *new_prog, 9414 struct bpf_prog *old_prog, u32 flags) 9415 { 9416 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES); 9417 struct bpf_prog *cur_prog; 9418 struct net_device *upper; 9419 struct list_head *iter; 9420 enum bpf_xdp_mode mode; 9421 bpf_op_t bpf_op; 9422 int err; 9423 9424 ASSERT_RTNL(); 9425 9426 /* either link or prog attachment, never both */ 9427 if (link && (new_prog || old_prog)) 9428 return -EINVAL; 9429 /* link supports only XDP mode flags */ 9430 if (link && (flags & ~XDP_FLAGS_MODES)) { 9431 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment"); 9432 return -EINVAL; 9433 } 9434 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */ 9435 if (num_modes > 1) { 9436 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set"); 9437 return -EINVAL; 9438 } 9439 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */ 9440 if (!num_modes && dev_xdp_prog_count(dev) > 1) { 9441 NL_SET_ERR_MSG(extack, 9442 "More than one program loaded, unset mode is ambiguous"); 9443 return -EINVAL; 9444 } 9445 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */ 9446 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) { 9447 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified"); 9448 return -EINVAL; 9449 } 9450 9451 mode = dev_xdp_mode(dev, flags); 9452 /* can't replace attached link */ 9453 if (dev_xdp_link(dev, mode)) { 9454 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link"); 9455 return -EBUSY; 9456 } 9457 9458 /* don't allow if an upper device already has a program */ 9459 netdev_for_each_upper_dev_rcu(dev, upper, iter) { 9460 if (dev_xdp_prog_count(upper) > 0) { 9461 NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program"); 9462 return -EEXIST; 9463 } 9464 } 9465 9466 cur_prog = dev_xdp_prog(dev, mode); 9467 /* can't replace attached prog with link */ 9468 if (link && cur_prog) { 9469 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link"); 9470 return -EBUSY; 9471 } 9472 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) { 9473 NL_SET_ERR_MSG(extack, "Active program does not match expected"); 9474 return -EEXIST; 9475 } 9476 9477 /* put effective new program into new_prog */ 9478 if (link) 9479 new_prog = link->link.prog; 9480 9481 if (new_prog) { 9482 bool offload = mode == XDP_MODE_HW; 9483 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB 9484 ? XDP_MODE_DRV : XDP_MODE_SKB; 9485 9486 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) { 9487 NL_SET_ERR_MSG(extack, "XDP program already attached"); 9488 return -EBUSY; 9489 } 9490 if (!offload && dev_xdp_prog(dev, other_mode)) { 9491 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time"); 9492 return -EEXIST; 9493 } 9494 if (!offload && bpf_prog_is_dev_bound(new_prog->aux)) { 9495 NL_SET_ERR_MSG(extack, "Using device-bound program without HW_MODE flag is not supported"); 9496 return -EINVAL; 9497 } 9498 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) { 9499 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device"); 9500 return -EINVAL; 9501 } 9502 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) { 9503 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device"); 9504 return -EINVAL; 9505 } 9506 } 9507 9508 /* don't call drivers if the effective program didn't change */ 9509 if (new_prog != cur_prog) { 9510 bpf_op = dev_xdp_bpf_op(dev, mode); 9511 if (!bpf_op) { 9512 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode"); 9513 return -EOPNOTSUPP; 9514 } 9515 9516 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog); 9517 if (err) 9518 return err; 9519 } 9520 9521 if (link) 9522 dev_xdp_set_link(dev, mode, link); 9523 else 9524 dev_xdp_set_prog(dev, mode, new_prog); 9525 if (cur_prog) 9526 bpf_prog_put(cur_prog); 9527 9528 return 0; 9529 } 9530 9531 static int dev_xdp_attach_link(struct net_device *dev, 9532 struct netlink_ext_ack *extack, 9533 struct bpf_xdp_link *link) 9534 { 9535 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags); 9536 } 9537 9538 static int dev_xdp_detach_link(struct net_device *dev, 9539 struct netlink_ext_ack *extack, 9540 struct bpf_xdp_link *link) 9541 { 9542 enum bpf_xdp_mode mode; 9543 bpf_op_t bpf_op; 9544 9545 ASSERT_RTNL(); 9546 9547 mode = dev_xdp_mode(dev, link->flags); 9548 if (dev_xdp_link(dev, mode) != link) 9549 return -EINVAL; 9550 9551 bpf_op = dev_xdp_bpf_op(dev, mode); 9552 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9553 dev_xdp_set_link(dev, mode, NULL); 9554 return 0; 9555 } 9556 9557 static void bpf_xdp_link_release(struct bpf_link *link) 9558 { 9559 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9560 9561 rtnl_lock(); 9562 9563 /* if racing with net_device's tear down, xdp_link->dev might be 9564 * already NULL, in which case link was already auto-detached 9565 */ 9566 if (xdp_link->dev) { 9567 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link)); 9568 xdp_link->dev = NULL; 9569 } 9570 9571 rtnl_unlock(); 9572 } 9573 9574 static int bpf_xdp_link_detach(struct bpf_link *link) 9575 { 9576 bpf_xdp_link_release(link); 9577 return 0; 9578 } 9579 9580 static void bpf_xdp_link_dealloc(struct bpf_link *link) 9581 { 9582 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9583 9584 kfree(xdp_link); 9585 } 9586 9587 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link, 9588 struct seq_file *seq) 9589 { 9590 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9591 u32 ifindex = 0; 9592 9593 rtnl_lock(); 9594 if (xdp_link->dev) 9595 ifindex = xdp_link->dev->ifindex; 9596 rtnl_unlock(); 9597 9598 seq_printf(seq, "ifindex:\t%u\n", ifindex); 9599 } 9600 9601 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link, 9602 struct bpf_link_info *info) 9603 { 9604 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9605 u32 ifindex = 0; 9606 9607 rtnl_lock(); 9608 if (xdp_link->dev) 9609 ifindex = xdp_link->dev->ifindex; 9610 rtnl_unlock(); 9611 9612 info->xdp.ifindex = ifindex; 9613 return 0; 9614 } 9615 9616 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog, 9617 struct bpf_prog *old_prog) 9618 { 9619 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9620 enum bpf_xdp_mode mode; 9621 bpf_op_t bpf_op; 9622 int err = 0; 9623 9624 rtnl_lock(); 9625 9626 /* link might have been auto-released already, so fail */ 9627 if (!xdp_link->dev) { 9628 err = -ENOLINK; 9629 goto out_unlock; 9630 } 9631 9632 if (old_prog && link->prog != old_prog) { 9633 err = -EPERM; 9634 goto out_unlock; 9635 } 9636 old_prog = link->prog; 9637 if (old_prog == new_prog) { 9638 /* no-op, don't disturb drivers */ 9639 bpf_prog_put(new_prog); 9640 goto out_unlock; 9641 } 9642 9643 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags); 9644 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode); 9645 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL, 9646 xdp_link->flags, new_prog); 9647 if (err) 9648 goto out_unlock; 9649 9650 old_prog = xchg(&link->prog, new_prog); 9651 bpf_prog_put(old_prog); 9652 9653 out_unlock: 9654 rtnl_unlock(); 9655 return err; 9656 } 9657 9658 static const struct bpf_link_ops bpf_xdp_link_lops = { 9659 .release = bpf_xdp_link_release, 9660 .dealloc = bpf_xdp_link_dealloc, 9661 .detach = bpf_xdp_link_detach, 9662 .show_fdinfo = bpf_xdp_link_show_fdinfo, 9663 .fill_link_info = bpf_xdp_link_fill_link_info, 9664 .update_prog = bpf_xdp_link_update, 9665 }; 9666 9667 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog) 9668 { 9669 struct net *net = current->nsproxy->net_ns; 9670 struct bpf_link_primer link_primer; 9671 struct bpf_xdp_link *link; 9672 struct net_device *dev; 9673 int err, fd; 9674 9675 rtnl_lock(); 9676 dev = dev_get_by_index(net, attr->link_create.target_ifindex); 9677 if (!dev) { 9678 rtnl_unlock(); 9679 return -EINVAL; 9680 } 9681 9682 link = kzalloc(sizeof(*link), GFP_USER); 9683 if (!link) { 9684 err = -ENOMEM; 9685 goto unlock; 9686 } 9687 9688 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog); 9689 link->dev = dev; 9690 link->flags = attr->link_create.flags; 9691 9692 err = bpf_link_prime(&link->link, &link_primer); 9693 if (err) { 9694 kfree(link); 9695 goto unlock; 9696 } 9697 9698 err = dev_xdp_attach_link(dev, NULL, link); 9699 rtnl_unlock(); 9700 9701 if (err) { 9702 link->dev = NULL; 9703 bpf_link_cleanup(&link_primer); 9704 goto out_put_dev; 9705 } 9706 9707 fd = bpf_link_settle(&link_primer); 9708 /* link itself doesn't hold dev's refcnt to not complicate shutdown */ 9709 dev_put(dev); 9710 return fd; 9711 9712 unlock: 9713 rtnl_unlock(); 9714 9715 out_put_dev: 9716 dev_put(dev); 9717 return err; 9718 } 9719 9720 /** 9721 * dev_change_xdp_fd - set or clear a bpf program for a device rx path 9722 * @dev: device 9723 * @extack: netlink extended ack 9724 * @fd: new program fd or negative value to clear 9725 * @expected_fd: old program fd that userspace expects to replace or clear 9726 * @flags: xdp-related flags 9727 * 9728 * Set or clear a bpf program for a device 9729 */ 9730 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, 9731 int fd, int expected_fd, u32 flags) 9732 { 9733 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags); 9734 struct bpf_prog *new_prog = NULL, *old_prog = NULL; 9735 int err; 9736 9737 ASSERT_RTNL(); 9738 9739 if (fd >= 0) { 9740 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP, 9741 mode != XDP_MODE_SKB); 9742 if (IS_ERR(new_prog)) 9743 return PTR_ERR(new_prog); 9744 } 9745 9746 if (expected_fd >= 0) { 9747 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP, 9748 mode != XDP_MODE_SKB); 9749 if (IS_ERR(old_prog)) { 9750 err = PTR_ERR(old_prog); 9751 old_prog = NULL; 9752 goto err_out; 9753 } 9754 } 9755 9756 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags); 9757 9758 err_out: 9759 if (err && new_prog) 9760 bpf_prog_put(new_prog); 9761 if (old_prog) 9762 bpf_prog_put(old_prog); 9763 return err; 9764 } 9765 9766 /** 9767 * dev_new_index - allocate an ifindex 9768 * @net: the applicable net namespace 9769 * 9770 * Returns a suitable unique value for a new device interface 9771 * number. The caller must hold the rtnl semaphore or the 9772 * dev_base_lock to be sure it remains unique. 9773 */ 9774 static int dev_new_index(struct net *net) 9775 { 9776 int ifindex = net->ifindex; 9777 9778 for (;;) { 9779 if (++ifindex <= 0) 9780 ifindex = 1; 9781 if (!__dev_get_by_index(net, ifindex)) 9782 return net->ifindex = ifindex; 9783 } 9784 } 9785 9786 /* Delayed registration/unregisteration */ 9787 static LIST_HEAD(net_todo_list); 9788 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); 9789 9790 static void net_set_todo(struct net_device *dev) 9791 { 9792 list_add_tail(&dev->todo_list, &net_todo_list); 9793 dev_net(dev)->dev_unreg_count++; 9794 } 9795 9796 static netdev_features_t netdev_sync_upper_features(struct net_device *lower, 9797 struct net_device *upper, netdev_features_t features) 9798 { 9799 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9800 netdev_features_t feature; 9801 int feature_bit; 9802 9803 for_each_netdev_feature(upper_disables, feature_bit) { 9804 feature = __NETIF_F_BIT(feature_bit); 9805 if (!(upper->wanted_features & feature) 9806 && (features & feature)) { 9807 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", 9808 &feature, upper->name); 9809 features &= ~feature; 9810 } 9811 } 9812 9813 return features; 9814 } 9815 9816 static void netdev_sync_lower_features(struct net_device *upper, 9817 struct net_device *lower, netdev_features_t features) 9818 { 9819 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9820 netdev_features_t feature; 9821 int feature_bit; 9822 9823 for_each_netdev_feature(upper_disables, feature_bit) { 9824 feature = __NETIF_F_BIT(feature_bit); 9825 if (!(features & feature) && (lower->features & feature)) { 9826 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", 9827 &feature, lower->name); 9828 lower->wanted_features &= ~feature; 9829 __netdev_update_features(lower); 9830 9831 if (unlikely(lower->features & feature)) 9832 netdev_WARN(upper, "failed to disable %pNF on %s!\n", 9833 &feature, lower->name); 9834 else 9835 netdev_features_change(lower); 9836 } 9837 } 9838 } 9839 9840 static netdev_features_t netdev_fix_features(struct net_device *dev, 9841 netdev_features_t features) 9842 { 9843 /* Fix illegal checksum combinations */ 9844 if ((features & NETIF_F_HW_CSUM) && 9845 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { 9846 netdev_warn(dev, "mixed HW and IP checksum settings.\n"); 9847 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); 9848 } 9849 9850 /* TSO requires that SG is present as well. */ 9851 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { 9852 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); 9853 features &= ~NETIF_F_ALL_TSO; 9854 } 9855 9856 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && 9857 !(features & NETIF_F_IP_CSUM)) { 9858 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); 9859 features &= ~NETIF_F_TSO; 9860 features &= ~NETIF_F_TSO_ECN; 9861 } 9862 9863 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && 9864 !(features & NETIF_F_IPV6_CSUM)) { 9865 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); 9866 features &= ~NETIF_F_TSO6; 9867 } 9868 9869 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ 9870 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) 9871 features &= ~NETIF_F_TSO_MANGLEID; 9872 9873 /* TSO ECN requires that TSO is present as well. */ 9874 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) 9875 features &= ~NETIF_F_TSO_ECN; 9876 9877 /* Software GSO depends on SG. */ 9878 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { 9879 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); 9880 features &= ~NETIF_F_GSO; 9881 } 9882 9883 /* GSO partial features require GSO partial be set */ 9884 if ((features & dev->gso_partial_features) && 9885 !(features & NETIF_F_GSO_PARTIAL)) { 9886 netdev_dbg(dev, 9887 "Dropping partially supported GSO features since no GSO partial.\n"); 9888 features &= ~dev->gso_partial_features; 9889 } 9890 9891 if (!(features & NETIF_F_RXCSUM)) { 9892 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet 9893 * successfully merged by hardware must also have the 9894 * checksum verified by hardware. If the user does not 9895 * want to enable RXCSUM, logically, we should disable GRO_HW. 9896 */ 9897 if (features & NETIF_F_GRO_HW) { 9898 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n"); 9899 features &= ~NETIF_F_GRO_HW; 9900 } 9901 } 9902 9903 /* LRO/HW-GRO features cannot be combined with RX-FCS */ 9904 if (features & NETIF_F_RXFCS) { 9905 if (features & NETIF_F_LRO) { 9906 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n"); 9907 features &= ~NETIF_F_LRO; 9908 } 9909 9910 if (features & NETIF_F_GRO_HW) { 9911 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n"); 9912 features &= ~NETIF_F_GRO_HW; 9913 } 9914 } 9915 9916 if (features & NETIF_F_HW_TLS_TX) { 9917 bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) == 9918 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM); 9919 bool hw_csum = features & NETIF_F_HW_CSUM; 9920 9921 if (!ip_csum && !hw_csum) { 9922 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n"); 9923 features &= ~NETIF_F_HW_TLS_TX; 9924 } 9925 } 9926 9927 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) { 9928 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n"); 9929 features &= ~NETIF_F_HW_TLS_RX; 9930 } 9931 9932 return features; 9933 } 9934 9935 int __netdev_update_features(struct net_device *dev) 9936 { 9937 struct net_device *upper, *lower; 9938 netdev_features_t features; 9939 struct list_head *iter; 9940 int err = -1; 9941 9942 ASSERT_RTNL(); 9943 9944 features = netdev_get_wanted_features(dev); 9945 9946 if (dev->netdev_ops->ndo_fix_features) 9947 features = dev->netdev_ops->ndo_fix_features(dev, features); 9948 9949 /* driver might be less strict about feature dependencies */ 9950 features = netdev_fix_features(dev, features); 9951 9952 /* some features can't be enabled if they're off on an upper device */ 9953 netdev_for_each_upper_dev_rcu(dev, upper, iter) 9954 features = netdev_sync_upper_features(dev, upper, features); 9955 9956 if (dev->features == features) 9957 goto sync_lower; 9958 9959 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", 9960 &dev->features, &features); 9961 9962 if (dev->netdev_ops->ndo_set_features) 9963 err = dev->netdev_ops->ndo_set_features(dev, features); 9964 else 9965 err = 0; 9966 9967 if (unlikely(err < 0)) { 9968 netdev_err(dev, 9969 "set_features() failed (%d); wanted %pNF, left %pNF\n", 9970 err, &features, &dev->features); 9971 /* return non-0 since some features might have changed and 9972 * it's better to fire a spurious notification than miss it 9973 */ 9974 return -1; 9975 } 9976 9977 sync_lower: 9978 /* some features must be disabled on lower devices when disabled 9979 * on an upper device (think: bonding master or bridge) 9980 */ 9981 netdev_for_each_lower_dev(dev, lower, iter) 9982 netdev_sync_lower_features(dev, lower, features); 9983 9984 if (!err) { 9985 netdev_features_t diff = features ^ dev->features; 9986 9987 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) { 9988 /* udp_tunnel_{get,drop}_rx_info both need 9989 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the 9990 * device, or they won't do anything. 9991 * Thus we need to update dev->features 9992 * *before* calling udp_tunnel_get_rx_info, 9993 * but *after* calling udp_tunnel_drop_rx_info. 9994 */ 9995 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) { 9996 dev->features = features; 9997 udp_tunnel_get_rx_info(dev); 9998 } else { 9999 udp_tunnel_drop_rx_info(dev); 10000 } 10001 } 10002 10003 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) { 10004 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) { 10005 dev->features = features; 10006 err |= vlan_get_rx_ctag_filter_info(dev); 10007 } else { 10008 vlan_drop_rx_ctag_filter_info(dev); 10009 } 10010 } 10011 10012 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) { 10013 if (features & NETIF_F_HW_VLAN_STAG_FILTER) { 10014 dev->features = features; 10015 err |= vlan_get_rx_stag_filter_info(dev); 10016 } else { 10017 vlan_drop_rx_stag_filter_info(dev); 10018 } 10019 } 10020 10021 dev->features = features; 10022 } 10023 10024 return err < 0 ? 0 : 1; 10025 } 10026 10027 /** 10028 * netdev_update_features - recalculate device features 10029 * @dev: the device to check 10030 * 10031 * Recalculate dev->features set and send notifications if it 10032 * has changed. Should be called after driver or hardware dependent 10033 * conditions might have changed that influence the features. 10034 */ 10035 void netdev_update_features(struct net_device *dev) 10036 { 10037 if (__netdev_update_features(dev)) 10038 netdev_features_change(dev); 10039 } 10040 EXPORT_SYMBOL(netdev_update_features); 10041 10042 /** 10043 * netdev_change_features - recalculate device features 10044 * @dev: the device to check 10045 * 10046 * Recalculate dev->features set and send notifications even 10047 * if they have not changed. Should be called instead of 10048 * netdev_update_features() if also dev->vlan_features might 10049 * have changed to allow the changes to be propagated to stacked 10050 * VLAN devices. 10051 */ 10052 void netdev_change_features(struct net_device *dev) 10053 { 10054 __netdev_update_features(dev); 10055 netdev_features_change(dev); 10056 } 10057 EXPORT_SYMBOL(netdev_change_features); 10058 10059 /** 10060 * netif_stacked_transfer_operstate - transfer operstate 10061 * @rootdev: the root or lower level device to transfer state from 10062 * @dev: the device to transfer operstate to 10063 * 10064 * Transfer operational state from root to device. This is normally 10065 * called when a stacking relationship exists between the root 10066 * device and the device(a leaf device). 10067 */ 10068 void netif_stacked_transfer_operstate(const struct net_device *rootdev, 10069 struct net_device *dev) 10070 { 10071 if (rootdev->operstate == IF_OPER_DORMANT) 10072 netif_dormant_on(dev); 10073 else 10074 netif_dormant_off(dev); 10075 10076 if (rootdev->operstate == IF_OPER_TESTING) 10077 netif_testing_on(dev); 10078 else 10079 netif_testing_off(dev); 10080 10081 if (netif_carrier_ok(rootdev)) 10082 netif_carrier_on(dev); 10083 else 10084 netif_carrier_off(dev); 10085 } 10086 EXPORT_SYMBOL(netif_stacked_transfer_operstate); 10087 10088 static int netif_alloc_rx_queues(struct net_device *dev) 10089 { 10090 unsigned int i, count = dev->num_rx_queues; 10091 struct netdev_rx_queue *rx; 10092 size_t sz = count * sizeof(*rx); 10093 int err = 0; 10094 10095 BUG_ON(count < 1); 10096 10097 rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10098 if (!rx) 10099 return -ENOMEM; 10100 10101 dev->_rx = rx; 10102 10103 for (i = 0; i < count; i++) { 10104 rx[i].dev = dev; 10105 10106 /* XDP RX-queue setup */ 10107 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0); 10108 if (err < 0) 10109 goto err_rxq_info; 10110 } 10111 return 0; 10112 10113 err_rxq_info: 10114 /* Rollback successful reg's and free other resources */ 10115 while (i--) 10116 xdp_rxq_info_unreg(&rx[i].xdp_rxq); 10117 kvfree(dev->_rx); 10118 dev->_rx = NULL; 10119 return err; 10120 } 10121 10122 static void netif_free_rx_queues(struct net_device *dev) 10123 { 10124 unsigned int i, count = dev->num_rx_queues; 10125 10126 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */ 10127 if (!dev->_rx) 10128 return; 10129 10130 for (i = 0; i < count; i++) 10131 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq); 10132 10133 kvfree(dev->_rx); 10134 } 10135 10136 static void netdev_init_one_queue(struct net_device *dev, 10137 struct netdev_queue *queue, void *_unused) 10138 { 10139 /* Initialize queue lock */ 10140 spin_lock_init(&queue->_xmit_lock); 10141 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); 10142 queue->xmit_lock_owner = -1; 10143 netdev_queue_numa_node_write(queue, NUMA_NO_NODE); 10144 queue->dev = dev; 10145 #ifdef CONFIG_BQL 10146 dql_init(&queue->dql, HZ); 10147 #endif 10148 } 10149 10150 static void netif_free_tx_queues(struct net_device *dev) 10151 { 10152 kvfree(dev->_tx); 10153 } 10154 10155 static int netif_alloc_netdev_queues(struct net_device *dev) 10156 { 10157 unsigned int count = dev->num_tx_queues; 10158 struct netdev_queue *tx; 10159 size_t sz = count * sizeof(*tx); 10160 10161 if (count < 1 || count > 0xffff) 10162 return -EINVAL; 10163 10164 tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10165 if (!tx) 10166 return -ENOMEM; 10167 10168 dev->_tx = tx; 10169 10170 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); 10171 spin_lock_init(&dev->tx_global_lock); 10172 10173 return 0; 10174 } 10175 10176 void netif_tx_stop_all_queues(struct net_device *dev) 10177 { 10178 unsigned int i; 10179 10180 for (i = 0; i < dev->num_tx_queues; i++) { 10181 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 10182 10183 netif_tx_stop_queue(txq); 10184 } 10185 } 10186 EXPORT_SYMBOL(netif_tx_stop_all_queues); 10187 10188 /** 10189 * register_netdevice - register a network device 10190 * @dev: device to register 10191 * 10192 * Take a completed network device structure and add it to the kernel 10193 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 10194 * chain. 0 is returned on success. A negative errno code is returned 10195 * on a failure to set up the device, or if the name is a duplicate. 10196 * 10197 * Callers must hold the rtnl semaphore. You may want 10198 * register_netdev() instead of this. 10199 * 10200 * BUGS: 10201 * The locking appears insufficient to guarantee two parallel registers 10202 * will not get the same name. 10203 */ 10204 10205 int register_netdevice(struct net_device *dev) 10206 { 10207 int ret; 10208 struct net *net = dev_net(dev); 10209 10210 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE < 10211 NETDEV_FEATURE_COUNT); 10212 BUG_ON(dev_boot_phase); 10213 ASSERT_RTNL(); 10214 10215 might_sleep(); 10216 10217 /* When net_device's are persistent, this will be fatal. */ 10218 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); 10219 BUG_ON(!net); 10220 10221 ret = ethtool_check_ops(dev->ethtool_ops); 10222 if (ret) 10223 return ret; 10224 10225 spin_lock_init(&dev->addr_list_lock); 10226 netdev_set_addr_lockdep_class(dev); 10227 10228 ret = dev_get_valid_name(net, dev, dev->name); 10229 if (ret < 0) 10230 goto out; 10231 10232 ret = -ENOMEM; 10233 dev->name_node = netdev_name_node_head_alloc(dev); 10234 if (!dev->name_node) 10235 goto out; 10236 10237 /* Init, if this function is available */ 10238 if (dev->netdev_ops->ndo_init) { 10239 ret = dev->netdev_ops->ndo_init(dev); 10240 if (ret) { 10241 if (ret > 0) 10242 ret = -EIO; 10243 goto err_free_name; 10244 } 10245 } 10246 10247 if (((dev->hw_features | dev->features) & 10248 NETIF_F_HW_VLAN_CTAG_FILTER) && 10249 (!dev->netdev_ops->ndo_vlan_rx_add_vid || 10250 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { 10251 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); 10252 ret = -EINVAL; 10253 goto err_uninit; 10254 } 10255 10256 ret = -EBUSY; 10257 if (!dev->ifindex) 10258 dev->ifindex = dev_new_index(net); 10259 else if (__dev_get_by_index(net, dev->ifindex)) 10260 goto err_uninit; 10261 10262 /* Transfer changeable features to wanted_features and enable 10263 * software offloads (GSO and GRO). 10264 */ 10265 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF); 10266 dev->features |= NETIF_F_SOFT_FEATURES; 10267 10268 if (dev->udp_tunnel_nic_info) { 10269 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10270 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10271 } 10272 10273 dev->wanted_features = dev->features & dev->hw_features; 10274 10275 if (!(dev->flags & IFF_LOOPBACK)) 10276 dev->hw_features |= NETIF_F_NOCACHE_COPY; 10277 10278 /* If IPv4 TCP segmentation offload is supported we should also 10279 * allow the device to enable segmenting the frame with the option 10280 * of ignoring a static IP ID value. This doesn't enable the 10281 * feature itself but allows the user to enable it later. 10282 */ 10283 if (dev->hw_features & NETIF_F_TSO) 10284 dev->hw_features |= NETIF_F_TSO_MANGLEID; 10285 if (dev->vlan_features & NETIF_F_TSO) 10286 dev->vlan_features |= NETIF_F_TSO_MANGLEID; 10287 if (dev->mpls_features & NETIF_F_TSO) 10288 dev->mpls_features |= NETIF_F_TSO_MANGLEID; 10289 if (dev->hw_enc_features & NETIF_F_TSO) 10290 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; 10291 10292 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. 10293 */ 10294 dev->vlan_features |= NETIF_F_HIGHDMA; 10295 10296 /* Make NETIF_F_SG inheritable to tunnel devices. 10297 */ 10298 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; 10299 10300 /* Make NETIF_F_SG inheritable to MPLS. 10301 */ 10302 dev->mpls_features |= NETIF_F_SG; 10303 10304 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); 10305 ret = notifier_to_errno(ret); 10306 if (ret) 10307 goto err_uninit; 10308 10309 ret = netdev_register_kobject(dev); 10310 if (ret) { 10311 dev->reg_state = NETREG_UNREGISTERED; 10312 goto err_uninit; 10313 } 10314 dev->reg_state = NETREG_REGISTERED; 10315 10316 __netdev_update_features(dev); 10317 10318 /* 10319 * Default initial state at registry is that the 10320 * device is present. 10321 */ 10322 10323 set_bit(__LINK_STATE_PRESENT, &dev->state); 10324 10325 linkwatch_init_dev(dev); 10326 10327 dev_init_scheduler(dev); 10328 dev_hold(dev); 10329 list_netdevice(dev); 10330 add_device_randomness(dev->dev_addr, dev->addr_len); 10331 10332 /* If the device has permanent device address, driver should 10333 * set dev_addr and also addr_assign_type should be set to 10334 * NET_ADDR_PERM (default value). 10335 */ 10336 if (dev->addr_assign_type == NET_ADDR_PERM) 10337 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); 10338 10339 /* Notify protocols, that a new device appeared. */ 10340 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); 10341 ret = notifier_to_errno(ret); 10342 if (ret) { 10343 /* Expect explicit free_netdev() on failure */ 10344 dev->needs_free_netdev = false; 10345 unregister_netdevice_queue(dev, NULL); 10346 goto out; 10347 } 10348 /* 10349 * Prevent userspace races by waiting until the network 10350 * device is fully setup before sending notifications. 10351 */ 10352 if (!dev->rtnl_link_ops || 10353 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 10354 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); 10355 10356 out: 10357 return ret; 10358 10359 err_uninit: 10360 if (dev->netdev_ops->ndo_uninit) 10361 dev->netdev_ops->ndo_uninit(dev); 10362 if (dev->priv_destructor) 10363 dev->priv_destructor(dev); 10364 err_free_name: 10365 netdev_name_node_free(dev->name_node); 10366 goto out; 10367 } 10368 EXPORT_SYMBOL(register_netdevice); 10369 10370 /** 10371 * init_dummy_netdev - init a dummy network device for NAPI 10372 * @dev: device to init 10373 * 10374 * This takes a network device structure and initialize the minimum 10375 * amount of fields so it can be used to schedule NAPI polls without 10376 * registering a full blown interface. This is to be used by drivers 10377 * that need to tie several hardware interfaces to a single NAPI 10378 * poll scheduler due to HW limitations. 10379 */ 10380 int init_dummy_netdev(struct net_device *dev) 10381 { 10382 /* Clear everything. Note we don't initialize spinlocks 10383 * are they aren't supposed to be taken by any of the 10384 * NAPI code and this dummy netdev is supposed to be 10385 * only ever used for NAPI polls 10386 */ 10387 memset(dev, 0, sizeof(struct net_device)); 10388 10389 /* make sure we BUG if trying to hit standard 10390 * register/unregister code path 10391 */ 10392 dev->reg_state = NETREG_DUMMY; 10393 10394 /* NAPI wants this */ 10395 INIT_LIST_HEAD(&dev->napi_list); 10396 10397 /* a dummy interface is started by default */ 10398 set_bit(__LINK_STATE_PRESENT, &dev->state); 10399 set_bit(__LINK_STATE_START, &dev->state); 10400 10401 /* napi_busy_loop stats accounting wants this */ 10402 dev_net_set(dev, &init_net); 10403 10404 /* Note : We dont allocate pcpu_refcnt for dummy devices, 10405 * because users of this 'device' dont need to change 10406 * its refcount. 10407 */ 10408 10409 return 0; 10410 } 10411 EXPORT_SYMBOL_GPL(init_dummy_netdev); 10412 10413 10414 /** 10415 * register_netdev - register a network device 10416 * @dev: device to register 10417 * 10418 * Take a completed network device structure and add it to the kernel 10419 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 10420 * chain. 0 is returned on success. A negative errno code is returned 10421 * on a failure to set up the device, or if the name is a duplicate. 10422 * 10423 * This is a wrapper around register_netdevice that takes the rtnl semaphore 10424 * and expands the device name if you passed a format string to 10425 * alloc_netdev. 10426 */ 10427 int register_netdev(struct net_device *dev) 10428 { 10429 int err; 10430 10431 if (rtnl_lock_killable()) 10432 return -EINTR; 10433 err = register_netdevice(dev); 10434 rtnl_unlock(); 10435 return err; 10436 } 10437 EXPORT_SYMBOL(register_netdev); 10438 10439 int netdev_refcnt_read(const struct net_device *dev) 10440 { 10441 #ifdef CONFIG_PCPU_DEV_REFCNT 10442 int i, refcnt = 0; 10443 10444 for_each_possible_cpu(i) 10445 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); 10446 return refcnt; 10447 #else 10448 return refcount_read(&dev->dev_refcnt); 10449 #endif 10450 } 10451 EXPORT_SYMBOL(netdev_refcnt_read); 10452 10453 int netdev_unregister_timeout_secs __read_mostly = 10; 10454 10455 #define WAIT_REFS_MIN_MSECS 1 10456 #define WAIT_REFS_MAX_MSECS 250 10457 /** 10458 * netdev_wait_allrefs - wait until all references are gone. 10459 * @dev: target net_device 10460 * 10461 * This is called when unregistering network devices. 10462 * 10463 * Any protocol or device that holds a reference should register 10464 * for netdevice notification, and cleanup and put back the 10465 * reference if they receive an UNREGISTER event. 10466 * We can get stuck here if buggy protocols don't correctly 10467 * call dev_put. 10468 */ 10469 static void netdev_wait_allrefs(struct net_device *dev) 10470 { 10471 unsigned long rebroadcast_time, warning_time; 10472 int wait = 0, refcnt; 10473 10474 linkwatch_forget_dev(dev); 10475 10476 rebroadcast_time = warning_time = jiffies; 10477 refcnt = netdev_refcnt_read(dev); 10478 10479 while (refcnt != 1) { 10480 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { 10481 rtnl_lock(); 10482 10483 /* Rebroadcast unregister notification */ 10484 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10485 10486 __rtnl_unlock(); 10487 rcu_barrier(); 10488 rtnl_lock(); 10489 10490 if (test_bit(__LINK_STATE_LINKWATCH_PENDING, 10491 &dev->state)) { 10492 /* We must not have linkwatch events 10493 * pending on unregister. If this 10494 * happens, we simply run the queue 10495 * unscheduled, resulting in a noop 10496 * for this device. 10497 */ 10498 linkwatch_run_queue(); 10499 } 10500 10501 __rtnl_unlock(); 10502 10503 rebroadcast_time = jiffies; 10504 } 10505 10506 if (!wait) { 10507 rcu_barrier(); 10508 wait = WAIT_REFS_MIN_MSECS; 10509 } else { 10510 msleep(wait); 10511 wait = min(wait << 1, WAIT_REFS_MAX_MSECS); 10512 } 10513 10514 refcnt = netdev_refcnt_read(dev); 10515 10516 if (refcnt != 1 && 10517 time_after(jiffies, warning_time + 10518 netdev_unregister_timeout_secs * HZ)) { 10519 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", 10520 dev->name, refcnt); 10521 warning_time = jiffies; 10522 } 10523 } 10524 } 10525 10526 /* The sequence is: 10527 * 10528 * rtnl_lock(); 10529 * ... 10530 * register_netdevice(x1); 10531 * register_netdevice(x2); 10532 * ... 10533 * unregister_netdevice(y1); 10534 * unregister_netdevice(y2); 10535 * ... 10536 * rtnl_unlock(); 10537 * free_netdev(y1); 10538 * free_netdev(y2); 10539 * 10540 * We are invoked by rtnl_unlock(). 10541 * This allows us to deal with problems: 10542 * 1) We can delete sysfs objects which invoke hotplug 10543 * without deadlocking with linkwatch via keventd. 10544 * 2) Since we run with the RTNL semaphore not held, we can sleep 10545 * safely in order to wait for the netdev refcnt to drop to zero. 10546 * 10547 * We must not return until all unregister events added during 10548 * the interval the lock was held have been completed. 10549 */ 10550 void netdev_run_todo(void) 10551 { 10552 struct list_head list; 10553 #ifdef CONFIG_LOCKDEP 10554 struct list_head unlink_list; 10555 10556 list_replace_init(&net_unlink_list, &unlink_list); 10557 10558 while (!list_empty(&unlink_list)) { 10559 struct net_device *dev = list_first_entry(&unlink_list, 10560 struct net_device, 10561 unlink_list); 10562 list_del_init(&dev->unlink_list); 10563 dev->nested_level = dev->lower_level - 1; 10564 } 10565 #endif 10566 10567 /* Snapshot list, allow later requests */ 10568 list_replace_init(&net_todo_list, &list); 10569 10570 __rtnl_unlock(); 10571 10572 10573 /* Wait for rcu callbacks to finish before next phase */ 10574 if (!list_empty(&list)) 10575 rcu_barrier(); 10576 10577 while (!list_empty(&list)) { 10578 struct net_device *dev 10579 = list_first_entry(&list, struct net_device, todo_list); 10580 list_del(&dev->todo_list); 10581 10582 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { 10583 pr_err("network todo '%s' but state %d\n", 10584 dev->name, dev->reg_state); 10585 dump_stack(); 10586 continue; 10587 } 10588 10589 dev->reg_state = NETREG_UNREGISTERED; 10590 10591 netdev_wait_allrefs(dev); 10592 10593 /* paranoia */ 10594 BUG_ON(netdev_refcnt_read(dev) != 1); 10595 BUG_ON(!list_empty(&dev->ptype_all)); 10596 BUG_ON(!list_empty(&dev->ptype_specific)); 10597 WARN_ON(rcu_access_pointer(dev->ip_ptr)); 10598 WARN_ON(rcu_access_pointer(dev->ip6_ptr)); 10599 #if IS_ENABLED(CONFIG_DECNET) 10600 WARN_ON(dev->dn_ptr); 10601 #endif 10602 if (dev->priv_destructor) 10603 dev->priv_destructor(dev); 10604 if (dev->needs_free_netdev) 10605 free_netdev(dev); 10606 10607 /* Report a network device has been unregistered */ 10608 rtnl_lock(); 10609 dev_net(dev)->dev_unreg_count--; 10610 __rtnl_unlock(); 10611 wake_up(&netdev_unregistering_wq); 10612 10613 /* Free network device */ 10614 kobject_put(&dev->dev.kobj); 10615 } 10616 } 10617 10618 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has 10619 * all the same fields in the same order as net_device_stats, with only 10620 * the type differing, but rtnl_link_stats64 may have additional fields 10621 * at the end for newer counters. 10622 */ 10623 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, 10624 const struct net_device_stats *netdev_stats) 10625 { 10626 #if BITS_PER_LONG == 64 10627 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats)); 10628 memcpy(stats64, netdev_stats, sizeof(*netdev_stats)); 10629 /* zero out counters that only exist in rtnl_link_stats64 */ 10630 memset((char *)stats64 + sizeof(*netdev_stats), 0, 10631 sizeof(*stats64) - sizeof(*netdev_stats)); 10632 #else 10633 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long); 10634 const unsigned long *src = (const unsigned long *)netdev_stats; 10635 u64 *dst = (u64 *)stats64; 10636 10637 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); 10638 for (i = 0; i < n; i++) 10639 dst[i] = src[i]; 10640 /* zero out counters that only exist in rtnl_link_stats64 */ 10641 memset((char *)stats64 + n * sizeof(u64), 0, 10642 sizeof(*stats64) - n * sizeof(u64)); 10643 #endif 10644 } 10645 EXPORT_SYMBOL(netdev_stats_to_stats64); 10646 10647 /** 10648 * dev_get_stats - get network device statistics 10649 * @dev: device to get statistics from 10650 * @storage: place to store stats 10651 * 10652 * Get network statistics from device. Return @storage. 10653 * The device driver may provide its own method by setting 10654 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; 10655 * otherwise the internal statistics structure is used. 10656 */ 10657 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, 10658 struct rtnl_link_stats64 *storage) 10659 { 10660 const struct net_device_ops *ops = dev->netdev_ops; 10661 10662 if (ops->ndo_get_stats64) { 10663 memset(storage, 0, sizeof(*storage)); 10664 ops->ndo_get_stats64(dev, storage); 10665 } else if (ops->ndo_get_stats) { 10666 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); 10667 } else { 10668 netdev_stats_to_stats64(storage, &dev->stats); 10669 } 10670 storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped); 10671 storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped); 10672 storage->rx_nohandler += (unsigned long)atomic_long_read(&dev->rx_nohandler); 10673 return storage; 10674 } 10675 EXPORT_SYMBOL(dev_get_stats); 10676 10677 /** 10678 * dev_fetch_sw_netstats - get per-cpu network device statistics 10679 * @s: place to store stats 10680 * @netstats: per-cpu network stats to read from 10681 * 10682 * Read per-cpu network statistics and populate the related fields in @s. 10683 */ 10684 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s, 10685 const struct pcpu_sw_netstats __percpu *netstats) 10686 { 10687 int cpu; 10688 10689 for_each_possible_cpu(cpu) { 10690 const struct pcpu_sw_netstats *stats; 10691 struct pcpu_sw_netstats tmp; 10692 unsigned int start; 10693 10694 stats = per_cpu_ptr(netstats, cpu); 10695 do { 10696 start = u64_stats_fetch_begin_irq(&stats->syncp); 10697 tmp.rx_packets = stats->rx_packets; 10698 tmp.rx_bytes = stats->rx_bytes; 10699 tmp.tx_packets = stats->tx_packets; 10700 tmp.tx_bytes = stats->tx_bytes; 10701 } while (u64_stats_fetch_retry_irq(&stats->syncp, start)); 10702 10703 s->rx_packets += tmp.rx_packets; 10704 s->rx_bytes += tmp.rx_bytes; 10705 s->tx_packets += tmp.tx_packets; 10706 s->tx_bytes += tmp.tx_bytes; 10707 } 10708 } 10709 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats); 10710 10711 /** 10712 * dev_get_tstats64 - ndo_get_stats64 implementation 10713 * @dev: device to get statistics from 10714 * @s: place to store stats 10715 * 10716 * Populate @s from dev->stats and dev->tstats. Can be used as 10717 * ndo_get_stats64() callback. 10718 */ 10719 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s) 10720 { 10721 netdev_stats_to_stats64(s, &dev->stats); 10722 dev_fetch_sw_netstats(s, dev->tstats); 10723 } 10724 EXPORT_SYMBOL_GPL(dev_get_tstats64); 10725 10726 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) 10727 { 10728 struct netdev_queue *queue = dev_ingress_queue(dev); 10729 10730 #ifdef CONFIG_NET_CLS_ACT 10731 if (queue) 10732 return queue; 10733 queue = kzalloc(sizeof(*queue), GFP_KERNEL); 10734 if (!queue) 10735 return NULL; 10736 netdev_init_one_queue(dev, queue, NULL); 10737 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); 10738 queue->qdisc_sleeping = &noop_qdisc; 10739 rcu_assign_pointer(dev->ingress_queue, queue); 10740 #endif 10741 return queue; 10742 } 10743 10744 static const struct ethtool_ops default_ethtool_ops; 10745 10746 void netdev_set_default_ethtool_ops(struct net_device *dev, 10747 const struct ethtool_ops *ops) 10748 { 10749 if (dev->ethtool_ops == &default_ethtool_ops) 10750 dev->ethtool_ops = ops; 10751 } 10752 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); 10753 10754 void netdev_freemem(struct net_device *dev) 10755 { 10756 char *addr = (char *)dev - dev->padded; 10757 10758 kvfree(addr); 10759 } 10760 10761 /** 10762 * alloc_netdev_mqs - allocate network device 10763 * @sizeof_priv: size of private data to allocate space for 10764 * @name: device name format string 10765 * @name_assign_type: origin of device name 10766 * @setup: callback to initialize device 10767 * @txqs: the number of TX subqueues to allocate 10768 * @rxqs: the number of RX subqueues to allocate 10769 * 10770 * Allocates a struct net_device with private data area for driver use 10771 * and performs basic initialization. Also allocates subqueue structs 10772 * for each queue on the device. 10773 */ 10774 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, 10775 unsigned char name_assign_type, 10776 void (*setup)(struct net_device *), 10777 unsigned int txqs, unsigned int rxqs) 10778 { 10779 struct net_device *dev; 10780 unsigned int alloc_size; 10781 struct net_device *p; 10782 10783 BUG_ON(strlen(name) >= sizeof(dev->name)); 10784 10785 if (txqs < 1) { 10786 pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); 10787 return NULL; 10788 } 10789 10790 if (rxqs < 1) { 10791 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); 10792 return NULL; 10793 } 10794 10795 alloc_size = sizeof(struct net_device); 10796 if (sizeof_priv) { 10797 /* ensure 32-byte alignment of private area */ 10798 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); 10799 alloc_size += sizeof_priv; 10800 } 10801 /* ensure 32-byte alignment of whole construct */ 10802 alloc_size += NETDEV_ALIGN - 1; 10803 10804 p = kvzalloc(alloc_size, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10805 if (!p) 10806 return NULL; 10807 10808 dev = PTR_ALIGN(p, NETDEV_ALIGN); 10809 dev->padded = (char *)dev - (char *)p; 10810 10811 #ifdef CONFIG_PCPU_DEV_REFCNT 10812 dev->pcpu_refcnt = alloc_percpu(int); 10813 if (!dev->pcpu_refcnt) 10814 goto free_dev; 10815 dev_hold(dev); 10816 #else 10817 refcount_set(&dev->dev_refcnt, 1); 10818 #endif 10819 10820 if (dev_addr_init(dev)) 10821 goto free_pcpu; 10822 10823 dev_mc_init(dev); 10824 dev_uc_init(dev); 10825 10826 dev_net_set(dev, &init_net); 10827 10828 dev->gso_max_size = GSO_MAX_SIZE; 10829 dev->gso_max_segs = GSO_MAX_SEGS; 10830 dev->upper_level = 1; 10831 dev->lower_level = 1; 10832 #ifdef CONFIG_LOCKDEP 10833 dev->nested_level = 0; 10834 INIT_LIST_HEAD(&dev->unlink_list); 10835 #endif 10836 10837 INIT_LIST_HEAD(&dev->napi_list); 10838 INIT_LIST_HEAD(&dev->unreg_list); 10839 INIT_LIST_HEAD(&dev->close_list); 10840 INIT_LIST_HEAD(&dev->link_watch_list); 10841 INIT_LIST_HEAD(&dev->adj_list.upper); 10842 INIT_LIST_HEAD(&dev->adj_list.lower); 10843 INIT_LIST_HEAD(&dev->ptype_all); 10844 INIT_LIST_HEAD(&dev->ptype_specific); 10845 INIT_LIST_HEAD(&dev->net_notifier_list); 10846 #ifdef CONFIG_NET_SCHED 10847 hash_init(dev->qdisc_hash); 10848 #endif 10849 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; 10850 setup(dev); 10851 10852 if (!dev->tx_queue_len) { 10853 dev->priv_flags |= IFF_NO_QUEUE; 10854 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; 10855 } 10856 10857 dev->num_tx_queues = txqs; 10858 dev->real_num_tx_queues = txqs; 10859 if (netif_alloc_netdev_queues(dev)) 10860 goto free_all; 10861 10862 dev->num_rx_queues = rxqs; 10863 dev->real_num_rx_queues = rxqs; 10864 if (netif_alloc_rx_queues(dev)) 10865 goto free_all; 10866 10867 strcpy(dev->name, name); 10868 dev->name_assign_type = name_assign_type; 10869 dev->group = INIT_NETDEV_GROUP; 10870 if (!dev->ethtool_ops) 10871 dev->ethtool_ops = &default_ethtool_ops; 10872 10873 nf_hook_ingress_init(dev); 10874 10875 return dev; 10876 10877 free_all: 10878 free_netdev(dev); 10879 return NULL; 10880 10881 free_pcpu: 10882 #ifdef CONFIG_PCPU_DEV_REFCNT 10883 free_percpu(dev->pcpu_refcnt); 10884 free_dev: 10885 #endif 10886 netdev_freemem(dev); 10887 return NULL; 10888 } 10889 EXPORT_SYMBOL(alloc_netdev_mqs); 10890 10891 /** 10892 * free_netdev - free network device 10893 * @dev: device 10894 * 10895 * This function does the last stage of destroying an allocated device 10896 * interface. The reference to the device object is released. If this 10897 * is the last reference then it will be freed.Must be called in process 10898 * context. 10899 */ 10900 void free_netdev(struct net_device *dev) 10901 { 10902 struct napi_struct *p, *n; 10903 10904 might_sleep(); 10905 10906 /* When called immediately after register_netdevice() failed the unwind 10907 * handling may still be dismantling the device. Handle that case by 10908 * deferring the free. 10909 */ 10910 if (dev->reg_state == NETREG_UNREGISTERING) { 10911 ASSERT_RTNL(); 10912 dev->needs_free_netdev = true; 10913 return; 10914 } 10915 10916 netif_free_tx_queues(dev); 10917 netif_free_rx_queues(dev); 10918 10919 kfree(rcu_dereference_protected(dev->ingress_queue, 1)); 10920 10921 /* Flush device addresses */ 10922 dev_addr_flush(dev); 10923 10924 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) 10925 netif_napi_del(p); 10926 10927 #ifdef CONFIG_PCPU_DEV_REFCNT 10928 free_percpu(dev->pcpu_refcnt); 10929 dev->pcpu_refcnt = NULL; 10930 #endif 10931 free_percpu(dev->xdp_bulkq); 10932 dev->xdp_bulkq = NULL; 10933 10934 /* Compatibility with error handling in drivers */ 10935 if (dev->reg_state == NETREG_UNINITIALIZED) { 10936 netdev_freemem(dev); 10937 return; 10938 } 10939 10940 BUG_ON(dev->reg_state != NETREG_UNREGISTERED); 10941 dev->reg_state = NETREG_RELEASED; 10942 10943 /* will free via device release */ 10944 put_device(&dev->dev); 10945 } 10946 EXPORT_SYMBOL(free_netdev); 10947 10948 /** 10949 * synchronize_net - Synchronize with packet receive processing 10950 * 10951 * Wait for packets currently being received to be done. 10952 * Does not block later packets from starting. 10953 */ 10954 void synchronize_net(void) 10955 { 10956 might_sleep(); 10957 if (rtnl_is_locked()) 10958 synchronize_rcu_expedited(); 10959 else 10960 synchronize_rcu(); 10961 } 10962 EXPORT_SYMBOL(synchronize_net); 10963 10964 /** 10965 * unregister_netdevice_queue - remove device from the kernel 10966 * @dev: device 10967 * @head: list 10968 * 10969 * This function shuts down a device interface and removes it 10970 * from the kernel tables. 10971 * If head not NULL, device is queued to be unregistered later. 10972 * 10973 * Callers must hold the rtnl semaphore. You may want 10974 * unregister_netdev() instead of this. 10975 */ 10976 10977 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) 10978 { 10979 ASSERT_RTNL(); 10980 10981 if (head) { 10982 list_move_tail(&dev->unreg_list, head); 10983 } else { 10984 LIST_HEAD(single); 10985 10986 list_add(&dev->unreg_list, &single); 10987 unregister_netdevice_many(&single); 10988 } 10989 } 10990 EXPORT_SYMBOL(unregister_netdevice_queue); 10991 10992 /** 10993 * unregister_netdevice_many - unregister many devices 10994 * @head: list of devices 10995 * 10996 * Note: As most callers use a stack allocated list_head, 10997 * we force a list_del() to make sure stack wont be corrupted later. 10998 */ 10999 void unregister_netdevice_many(struct list_head *head) 11000 { 11001 struct net_device *dev, *tmp; 11002 LIST_HEAD(close_head); 11003 11004 BUG_ON(dev_boot_phase); 11005 ASSERT_RTNL(); 11006 11007 if (list_empty(head)) 11008 return; 11009 11010 list_for_each_entry_safe(dev, tmp, head, unreg_list) { 11011 /* Some devices call without registering 11012 * for initialization unwind. Remove those 11013 * devices and proceed with the remaining. 11014 */ 11015 if (dev->reg_state == NETREG_UNINITIALIZED) { 11016 pr_debug("unregister_netdevice: device %s/%p never was registered\n", 11017 dev->name, dev); 11018 11019 WARN_ON(1); 11020 list_del(&dev->unreg_list); 11021 continue; 11022 } 11023 dev->dismantle = true; 11024 BUG_ON(dev->reg_state != NETREG_REGISTERED); 11025 } 11026 11027 /* If device is running, close it first. */ 11028 list_for_each_entry(dev, head, unreg_list) 11029 list_add_tail(&dev->close_list, &close_head); 11030 dev_close_many(&close_head, true); 11031 11032 list_for_each_entry(dev, head, unreg_list) { 11033 /* And unlink it from device chain. */ 11034 unlist_netdevice(dev); 11035 11036 dev->reg_state = NETREG_UNREGISTERING; 11037 } 11038 flush_all_backlogs(); 11039 11040 synchronize_net(); 11041 11042 list_for_each_entry(dev, head, unreg_list) { 11043 struct sk_buff *skb = NULL; 11044 11045 /* Shutdown queueing discipline. */ 11046 dev_shutdown(dev); 11047 11048 dev_xdp_uninstall(dev); 11049 11050 /* Notify protocols, that we are about to destroy 11051 * this device. They should clean all the things. 11052 */ 11053 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 11054 11055 if (!dev->rtnl_link_ops || 11056 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 11057 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0, 11058 GFP_KERNEL, NULL, 0); 11059 11060 /* 11061 * Flush the unicast and multicast chains 11062 */ 11063 dev_uc_flush(dev); 11064 dev_mc_flush(dev); 11065 11066 netdev_name_node_alt_flush(dev); 11067 netdev_name_node_free(dev->name_node); 11068 11069 if (dev->netdev_ops->ndo_uninit) 11070 dev->netdev_ops->ndo_uninit(dev); 11071 11072 if (skb) 11073 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL); 11074 11075 /* Notifier chain MUST detach us all upper devices. */ 11076 WARN_ON(netdev_has_any_upper_dev(dev)); 11077 WARN_ON(netdev_has_any_lower_dev(dev)); 11078 11079 /* Remove entries from kobject tree */ 11080 netdev_unregister_kobject(dev); 11081 #ifdef CONFIG_XPS 11082 /* Remove XPS queueing entries */ 11083 netif_reset_xps_queues_gt(dev, 0); 11084 #endif 11085 } 11086 11087 synchronize_net(); 11088 11089 list_for_each_entry(dev, head, unreg_list) { 11090 dev_put(dev); 11091 net_set_todo(dev); 11092 } 11093 11094 list_del(head); 11095 } 11096 EXPORT_SYMBOL(unregister_netdevice_many); 11097 11098 /** 11099 * unregister_netdev - remove device from the kernel 11100 * @dev: device 11101 * 11102 * This function shuts down a device interface and removes it 11103 * from the kernel tables. 11104 * 11105 * This is just a wrapper for unregister_netdevice that takes 11106 * the rtnl semaphore. In general you want to use this and not 11107 * unregister_netdevice. 11108 */ 11109 void unregister_netdev(struct net_device *dev) 11110 { 11111 rtnl_lock(); 11112 unregister_netdevice(dev); 11113 rtnl_unlock(); 11114 } 11115 EXPORT_SYMBOL(unregister_netdev); 11116 11117 /** 11118 * __dev_change_net_namespace - move device to different nethost namespace 11119 * @dev: device 11120 * @net: network namespace 11121 * @pat: If not NULL name pattern to try if the current device name 11122 * is already taken in the destination network namespace. 11123 * @new_ifindex: If not zero, specifies device index in the target 11124 * namespace. 11125 * 11126 * This function shuts down a device interface and moves it 11127 * to a new network namespace. On success 0 is returned, on 11128 * a failure a netagive errno code is returned. 11129 * 11130 * Callers must hold the rtnl semaphore. 11131 */ 11132 11133 int __dev_change_net_namespace(struct net_device *dev, struct net *net, 11134 const char *pat, int new_ifindex) 11135 { 11136 struct net *net_old = dev_net(dev); 11137 int err, new_nsid; 11138 11139 ASSERT_RTNL(); 11140 11141 /* Don't allow namespace local devices to be moved. */ 11142 err = -EINVAL; 11143 if (dev->features & NETIF_F_NETNS_LOCAL) 11144 goto out; 11145 11146 /* Ensure the device has been registrered */ 11147 if (dev->reg_state != NETREG_REGISTERED) 11148 goto out; 11149 11150 /* Get out if there is nothing todo */ 11151 err = 0; 11152 if (net_eq(net_old, net)) 11153 goto out; 11154 11155 /* Pick the destination device name, and ensure 11156 * we can use it in the destination network namespace. 11157 */ 11158 err = -EEXIST; 11159 if (netdev_name_in_use(net, dev->name)) { 11160 /* We get here if we can't use the current device name */ 11161 if (!pat) 11162 goto out; 11163 err = dev_get_valid_name(net, dev, pat); 11164 if (err < 0) 11165 goto out; 11166 } 11167 11168 /* Check that new_ifindex isn't used yet. */ 11169 err = -EBUSY; 11170 if (new_ifindex && __dev_get_by_index(net, new_ifindex)) 11171 goto out; 11172 11173 /* 11174 * And now a mini version of register_netdevice unregister_netdevice. 11175 */ 11176 11177 /* If device is running close it first. */ 11178 dev_close(dev); 11179 11180 /* And unlink it from device chain */ 11181 unlist_netdevice(dev); 11182 11183 synchronize_net(); 11184 11185 /* Shutdown queueing discipline. */ 11186 dev_shutdown(dev); 11187 11188 /* Notify protocols, that we are about to destroy 11189 * this device. They should clean all the things. 11190 * 11191 * Note that dev->reg_state stays at NETREG_REGISTERED. 11192 * This is wanted because this way 8021q and macvlan know 11193 * the device is just moving and can keep their slaves up. 11194 */ 11195 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 11196 rcu_barrier(); 11197 11198 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL); 11199 /* If there is an ifindex conflict assign a new one */ 11200 if (!new_ifindex) { 11201 if (__dev_get_by_index(net, dev->ifindex)) 11202 new_ifindex = dev_new_index(net); 11203 else 11204 new_ifindex = dev->ifindex; 11205 } 11206 11207 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid, 11208 new_ifindex); 11209 11210 /* 11211 * Flush the unicast and multicast chains 11212 */ 11213 dev_uc_flush(dev); 11214 dev_mc_flush(dev); 11215 11216 /* Send a netdev-removed uevent to the old namespace */ 11217 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); 11218 netdev_adjacent_del_links(dev); 11219 11220 /* Move per-net netdevice notifiers that are following the netdevice */ 11221 move_netdevice_notifiers_dev_net(dev, net); 11222 11223 /* Actually switch the network namespace */ 11224 dev_net_set(dev, net); 11225 dev->ifindex = new_ifindex; 11226 11227 /* Send a netdev-add uevent to the new namespace */ 11228 kobject_uevent(&dev->dev.kobj, KOBJ_ADD); 11229 netdev_adjacent_add_links(dev); 11230 11231 /* Fixup kobjects */ 11232 err = device_rename(&dev->dev, dev->name); 11233 WARN_ON(err); 11234 11235 /* Adapt owner in case owning user namespace of target network 11236 * namespace is different from the original one. 11237 */ 11238 err = netdev_change_owner(dev, net_old, net); 11239 WARN_ON(err); 11240 11241 /* Add the device back in the hashes */ 11242 list_netdevice(dev); 11243 11244 /* Notify protocols, that a new device appeared. */ 11245 call_netdevice_notifiers(NETDEV_REGISTER, dev); 11246 11247 /* 11248 * Prevent userspace races by waiting until the network 11249 * device is fully setup before sending notifications. 11250 */ 11251 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); 11252 11253 synchronize_net(); 11254 err = 0; 11255 out: 11256 return err; 11257 } 11258 EXPORT_SYMBOL_GPL(__dev_change_net_namespace); 11259 11260 static int dev_cpu_dead(unsigned int oldcpu) 11261 { 11262 struct sk_buff **list_skb; 11263 struct sk_buff *skb; 11264 unsigned int cpu; 11265 struct softnet_data *sd, *oldsd, *remsd = NULL; 11266 11267 local_irq_disable(); 11268 cpu = smp_processor_id(); 11269 sd = &per_cpu(softnet_data, cpu); 11270 oldsd = &per_cpu(softnet_data, oldcpu); 11271 11272 /* Find end of our completion_queue. */ 11273 list_skb = &sd->completion_queue; 11274 while (*list_skb) 11275 list_skb = &(*list_skb)->next; 11276 /* Append completion queue from offline CPU. */ 11277 *list_skb = oldsd->completion_queue; 11278 oldsd->completion_queue = NULL; 11279 11280 /* Append output queue from offline CPU. */ 11281 if (oldsd->output_queue) { 11282 *sd->output_queue_tailp = oldsd->output_queue; 11283 sd->output_queue_tailp = oldsd->output_queue_tailp; 11284 oldsd->output_queue = NULL; 11285 oldsd->output_queue_tailp = &oldsd->output_queue; 11286 } 11287 /* Append NAPI poll list from offline CPU, with one exception : 11288 * process_backlog() must be called by cpu owning percpu backlog. 11289 * We properly handle process_queue & input_pkt_queue later. 11290 */ 11291 while (!list_empty(&oldsd->poll_list)) { 11292 struct napi_struct *napi = list_first_entry(&oldsd->poll_list, 11293 struct napi_struct, 11294 poll_list); 11295 11296 list_del_init(&napi->poll_list); 11297 if (napi->poll == process_backlog) 11298 napi->state = 0; 11299 else 11300 ____napi_schedule(sd, napi); 11301 } 11302 11303 raise_softirq_irqoff(NET_TX_SOFTIRQ); 11304 local_irq_enable(); 11305 11306 #ifdef CONFIG_RPS 11307 remsd = oldsd->rps_ipi_list; 11308 oldsd->rps_ipi_list = NULL; 11309 #endif 11310 /* send out pending IPI's on offline CPU */ 11311 net_rps_send_ipi(remsd); 11312 11313 /* Process offline CPU's input_pkt_queue */ 11314 while ((skb = __skb_dequeue(&oldsd->process_queue))) { 11315 netif_rx_ni(skb); 11316 input_queue_head_incr(oldsd); 11317 } 11318 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { 11319 netif_rx_ni(skb); 11320 input_queue_head_incr(oldsd); 11321 } 11322 11323 return 0; 11324 } 11325 11326 /** 11327 * netdev_increment_features - increment feature set by one 11328 * @all: current feature set 11329 * @one: new feature set 11330 * @mask: mask feature set 11331 * 11332 * Computes a new feature set after adding a device with feature set 11333 * @one to the master device with current feature set @all. Will not 11334 * enable anything that is off in @mask. Returns the new feature set. 11335 */ 11336 netdev_features_t netdev_increment_features(netdev_features_t all, 11337 netdev_features_t one, netdev_features_t mask) 11338 { 11339 if (mask & NETIF_F_HW_CSUM) 11340 mask |= NETIF_F_CSUM_MASK; 11341 mask |= NETIF_F_VLAN_CHALLENGED; 11342 11343 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; 11344 all &= one | ~NETIF_F_ALL_FOR_ALL; 11345 11346 /* If one device supports hw checksumming, set for all. */ 11347 if (all & NETIF_F_HW_CSUM) 11348 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); 11349 11350 return all; 11351 } 11352 EXPORT_SYMBOL(netdev_increment_features); 11353 11354 static struct hlist_head * __net_init netdev_create_hash(void) 11355 { 11356 int i; 11357 struct hlist_head *hash; 11358 11359 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL); 11360 if (hash != NULL) 11361 for (i = 0; i < NETDEV_HASHENTRIES; i++) 11362 INIT_HLIST_HEAD(&hash[i]); 11363 11364 return hash; 11365 } 11366 11367 /* Initialize per network namespace state */ 11368 static int __net_init netdev_init(struct net *net) 11369 { 11370 BUILD_BUG_ON(GRO_HASH_BUCKETS > 11371 8 * sizeof_field(struct napi_struct, gro_bitmask)); 11372 11373 if (net != &init_net) 11374 INIT_LIST_HEAD(&net->dev_base_head); 11375 11376 net->dev_name_head = netdev_create_hash(); 11377 if (net->dev_name_head == NULL) 11378 goto err_name; 11379 11380 net->dev_index_head = netdev_create_hash(); 11381 if (net->dev_index_head == NULL) 11382 goto err_idx; 11383 11384 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain); 11385 11386 return 0; 11387 11388 err_idx: 11389 kfree(net->dev_name_head); 11390 err_name: 11391 return -ENOMEM; 11392 } 11393 11394 /** 11395 * netdev_drivername - network driver for the device 11396 * @dev: network device 11397 * 11398 * Determine network driver for device. 11399 */ 11400 const char *netdev_drivername(const struct net_device *dev) 11401 { 11402 const struct device_driver *driver; 11403 const struct device *parent; 11404 const char *empty = ""; 11405 11406 parent = dev->dev.parent; 11407 if (!parent) 11408 return empty; 11409 11410 driver = parent->driver; 11411 if (driver && driver->name) 11412 return driver->name; 11413 return empty; 11414 } 11415 11416 static void __netdev_printk(const char *level, const struct net_device *dev, 11417 struct va_format *vaf) 11418 { 11419 if (dev && dev->dev.parent) { 11420 dev_printk_emit(level[1] - '0', 11421 dev->dev.parent, 11422 "%s %s %s%s: %pV", 11423 dev_driver_string(dev->dev.parent), 11424 dev_name(dev->dev.parent), 11425 netdev_name(dev), netdev_reg_state(dev), 11426 vaf); 11427 } else if (dev) { 11428 printk("%s%s%s: %pV", 11429 level, netdev_name(dev), netdev_reg_state(dev), vaf); 11430 } else { 11431 printk("%s(NULL net_device): %pV", level, vaf); 11432 } 11433 } 11434 11435 void netdev_printk(const char *level, const struct net_device *dev, 11436 const char *format, ...) 11437 { 11438 struct va_format vaf; 11439 va_list args; 11440 11441 va_start(args, format); 11442 11443 vaf.fmt = format; 11444 vaf.va = &args; 11445 11446 __netdev_printk(level, dev, &vaf); 11447 11448 va_end(args); 11449 } 11450 EXPORT_SYMBOL(netdev_printk); 11451 11452 #define define_netdev_printk_level(func, level) \ 11453 void func(const struct net_device *dev, const char *fmt, ...) \ 11454 { \ 11455 struct va_format vaf; \ 11456 va_list args; \ 11457 \ 11458 va_start(args, fmt); \ 11459 \ 11460 vaf.fmt = fmt; \ 11461 vaf.va = &args; \ 11462 \ 11463 __netdev_printk(level, dev, &vaf); \ 11464 \ 11465 va_end(args); \ 11466 } \ 11467 EXPORT_SYMBOL(func); 11468 11469 define_netdev_printk_level(netdev_emerg, KERN_EMERG); 11470 define_netdev_printk_level(netdev_alert, KERN_ALERT); 11471 define_netdev_printk_level(netdev_crit, KERN_CRIT); 11472 define_netdev_printk_level(netdev_err, KERN_ERR); 11473 define_netdev_printk_level(netdev_warn, KERN_WARNING); 11474 define_netdev_printk_level(netdev_notice, KERN_NOTICE); 11475 define_netdev_printk_level(netdev_info, KERN_INFO); 11476 11477 static void __net_exit netdev_exit(struct net *net) 11478 { 11479 kfree(net->dev_name_head); 11480 kfree(net->dev_index_head); 11481 if (net != &init_net) 11482 WARN_ON_ONCE(!list_empty(&net->dev_base_head)); 11483 } 11484 11485 static struct pernet_operations __net_initdata netdev_net_ops = { 11486 .init = netdev_init, 11487 .exit = netdev_exit, 11488 }; 11489 11490 static void __net_exit default_device_exit(struct net *net) 11491 { 11492 struct net_device *dev, *aux; 11493 /* 11494 * Push all migratable network devices back to the 11495 * initial network namespace 11496 */ 11497 rtnl_lock(); 11498 for_each_netdev_safe(net, dev, aux) { 11499 int err; 11500 char fb_name[IFNAMSIZ]; 11501 11502 /* Ignore unmoveable devices (i.e. loopback) */ 11503 if (dev->features & NETIF_F_NETNS_LOCAL) 11504 continue; 11505 11506 /* Leave virtual devices for the generic cleanup */ 11507 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund) 11508 continue; 11509 11510 /* Push remaining network devices to init_net */ 11511 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); 11512 if (netdev_name_in_use(&init_net, fb_name)) 11513 snprintf(fb_name, IFNAMSIZ, "dev%%d"); 11514 err = dev_change_net_namespace(dev, &init_net, fb_name); 11515 if (err) { 11516 pr_emerg("%s: failed to move %s to init_net: %d\n", 11517 __func__, dev->name, err); 11518 BUG(); 11519 } 11520 } 11521 rtnl_unlock(); 11522 } 11523 11524 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list) 11525 { 11526 /* Return with the rtnl_lock held when there are no network 11527 * devices unregistering in any network namespace in net_list. 11528 */ 11529 struct net *net; 11530 bool unregistering; 11531 DEFINE_WAIT_FUNC(wait, woken_wake_function); 11532 11533 add_wait_queue(&netdev_unregistering_wq, &wait); 11534 for (;;) { 11535 unregistering = false; 11536 rtnl_lock(); 11537 list_for_each_entry(net, net_list, exit_list) { 11538 if (net->dev_unreg_count > 0) { 11539 unregistering = true; 11540 break; 11541 } 11542 } 11543 if (!unregistering) 11544 break; 11545 __rtnl_unlock(); 11546 11547 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); 11548 } 11549 remove_wait_queue(&netdev_unregistering_wq, &wait); 11550 } 11551 11552 static void __net_exit default_device_exit_batch(struct list_head *net_list) 11553 { 11554 /* At exit all network devices most be removed from a network 11555 * namespace. Do this in the reverse order of registration. 11556 * Do this across as many network namespaces as possible to 11557 * improve batching efficiency. 11558 */ 11559 struct net_device *dev; 11560 struct net *net; 11561 LIST_HEAD(dev_kill_list); 11562 11563 /* To prevent network device cleanup code from dereferencing 11564 * loopback devices or network devices that have been freed 11565 * wait here for all pending unregistrations to complete, 11566 * before unregistring the loopback device and allowing the 11567 * network namespace be freed. 11568 * 11569 * The netdev todo list containing all network devices 11570 * unregistrations that happen in default_device_exit_batch 11571 * will run in the rtnl_unlock() at the end of 11572 * default_device_exit_batch. 11573 */ 11574 rtnl_lock_unregistering(net_list); 11575 list_for_each_entry(net, net_list, exit_list) { 11576 for_each_netdev_reverse(net, dev) { 11577 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) 11578 dev->rtnl_link_ops->dellink(dev, &dev_kill_list); 11579 else 11580 unregister_netdevice_queue(dev, &dev_kill_list); 11581 } 11582 } 11583 unregister_netdevice_many(&dev_kill_list); 11584 rtnl_unlock(); 11585 } 11586 11587 static struct pernet_operations __net_initdata default_device_ops = { 11588 .exit = default_device_exit, 11589 .exit_batch = default_device_exit_batch, 11590 }; 11591 11592 /* 11593 * Initialize the DEV module. At boot time this walks the device list and 11594 * unhooks any devices that fail to initialise (normally hardware not 11595 * present) and leaves us with a valid list of present and active devices. 11596 * 11597 */ 11598 11599 /* 11600 * This is called single threaded during boot, so no need 11601 * to take the rtnl semaphore. 11602 */ 11603 static int __init net_dev_init(void) 11604 { 11605 int i, rc = -ENOMEM; 11606 11607 BUG_ON(!dev_boot_phase); 11608 11609 if (dev_proc_init()) 11610 goto out; 11611 11612 if (netdev_kobject_init()) 11613 goto out; 11614 11615 INIT_LIST_HEAD(&ptype_all); 11616 for (i = 0; i < PTYPE_HASH_SIZE; i++) 11617 INIT_LIST_HEAD(&ptype_base[i]); 11618 11619 INIT_LIST_HEAD(&offload_base); 11620 11621 if (register_pernet_subsys(&netdev_net_ops)) 11622 goto out; 11623 11624 /* 11625 * Initialise the packet receive queues. 11626 */ 11627 11628 for_each_possible_cpu(i) { 11629 struct work_struct *flush = per_cpu_ptr(&flush_works, i); 11630 struct softnet_data *sd = &per_cpu(softnet_data, i); 11631 11632 INIT_WORK(flush, flush_backlog); 11633 11634 skb_queue_head_init(&sd->input_pkt_queue); 11635 skb_queue_head_init(&sd->process_queue); 11636 #ifdef CONFIG_XFRM_OFFLOAD 11637 skb_queue_head_init(&sd->xfrm_backlog); 11638 #endif 11639 INIT_LIST_HEAD(&sd->poll_list); 11640 sd->output_queue_tailp = &sd->output_queue; 11641 #ifdef CONFIG_RPS 11642 INIT_CSD(&sd->csd, rps_trigger_softirq, sd); 11643 sd->cpu = i; 11644 #endif 11645 11646 init_gro_hash(&sd->backlog); 11647 sd->backlog.poll = process_backlog; 11648 sd->backlog.weight = weight_p; 11649 } 11650 11651 dev_boot_phase = 0; 11652 11653 /* The loopback device is special if any other network devices 11654 * is present in a network namespace the loopback device must 11655 * be present. Since we now dynamically allocate and free the 11656 * loopback device ensure this invariant is maintained by 11657 * keeping the loopback device as the first device on the 11658 * list of network devices. Ensuring the loopback devices 11659 * is the first device that appears and the last network device 11660 * that disappears. 11661 */ 11662 if (register_pernet_device(&loopback_net_ops)) 11663 goto out; 11664 11665 if (register_pernet_device(&default_device_ops)) 11666 goto out; 11667 11668 open_softirq(NET_TX_SOFTIRQ, net_tx_action); 11669 open_softirq(NET_RX_SOFTIRQ, net_rx_action); 11670 11671 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", 11672 NULL, dev_cpu_dead); 11673 WARN_ON(rc < 0); 11674 rc = 0; 11675 out: 11676 return rc; 11677 } 11678 11679 subsys_initcall(net_dev_init); 11680