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_netdev.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 netdev_err(dev, "name change rollback failed: %d\n", 1300 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 netdev_warn(dev, "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 netdev_warn(dev, "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 if (unlikely(!qcount)) { 3175 net_warn_ratelimited("%s: invalid qcount, qoffset %u for tc %u\n", 3176 sb_dev->name, qoffset, tc); 3177 qoffset = 0; 3178 qcount = dev->real_num_tx_queues; 3179 } 3180 } 3181 3182 if (skb_rx_queue_recorded(skb)) { 3183 hash = skb_get_rx_queue(skb); 3184 if (hash >= qoffset) 3185 hash -= qoffset; 3186 while (unlikely(hash >= qcount)) 3187 hash -= qcount; 3188 return hash + qoffset; 3189 } 3190 3191 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; 3192 } 3193 3194 static void skb_warn_bad_offload(const struct sk_buff *skb) 3195 { 3196 static const netdev_features_t null_features; 3197 struct net_device *dev = skb->dev; 3198 const char *name = ""; 3199 3200 if (!net_ratelimit()) 3201 return; 3202 3203 if (dev) { 3204 if (dev->dev.parent) 3205 name = dev_driver_string(dev->dev.parent); 3206 else 3207 name = netdev_name(dev); 3208 } 3209 skb_dump(KERN_WARNING, skb, false); 3210 WARN(1, "%s: caps=(%pNF, %pNF)\n", 3211 name, dev ? &dev->features : &null_features, 3212 skb->sk ? &skb->sk->sk_route_caps : &null_features); 3213 } 3214 3215 /* 3216 * Invalidate hardware checksum when packet is to be mangled, and 3217 * complete checksum manually on outgoing path. 3218 */ 3219 int skb_checksum_help(struct sk_buff *skb) 3220 { 3221 __wsum csum; 3222 int ret = 0, offset; 3223 3224 if (skb->ip_summed == CHECKSUM_COMPLETE) 3225 goto out_set_summed; 3226 3227 if (unlikely(skb_is_gso(skb))) { 3228 skb_warn_bad_offload(skb); 3229 return -EINVAL; 3230 } 3231 3232 /* Before computing a checksum, we should make sure no frag could 3233 * be modified by an external entity : checksum could be wrong. 3234 */ 3235 if (skb_has_shared_frag(skb)) { 3236 ret = __skb_linearize(skb); 3237 if (ret) 3238 goto out; 3239 } 3240 3241 offset = skb_checksum_start_offset(skb); 3242 BUG_ON(offset >= skb_headlen(skb)); 3243 csum = skb_checksum(skb, offset, skb->len - offset, 0); 3244 3245 offset += skb->csum_offset; 3246 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); 3247 3248 ret = skb_ensure_writable(skb, offset + sizeof(__sum16)); 3249 if (ret) 3250 goto out; 3251 3252 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0; 3253 out_set_summed: 3254 skb->ip_summed = CHECKSUM_NONE; 3255 out: 3256 return ret; 3257 } 3258 EXPORT_SYMBOL(skb_checksum_help); 3259 3260 int skb_crc32c_csum_help(struct sk_buff *skb) 3261 { 3262 __le32 crc32c_csum; 3263 int ret = 0, offset, start; 3264 3265 if (skb->ip_summed != CHECKSUM_PARTIAL) 3266 goto out; 3267 3268 if (unlikely(skb_is_gso(skb))) 3269 goto out; 3270 3271 /* Before computing a checksum, we should make sure no frag could 3272 * be modified by an external entity : checksum could be wrong. 3273 */ 3274 if (unlikely(skb_has_shared_frag(skb))) { 3275 ret = __skb_linearize(skb); 3276 if (ret) 3277 goto out; 3278 } 3279 start = skb_checksum_start_offset(skb); 3280 offset = start + offsetof(struct sctphdr, checksum); 3281 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { 3282 ret = -EINVAL; 3283 goto out; 3284 } 3285 3286 ret = skb_ensure_writable(skb, offset + sizeof(__le32)); 3287 if (ret) 3288 goto out; 3289 3290 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start, 3291 skb->len - start, ~(__u32)0, 3292 crc32c_csum_stub)); 3293 *(__le32 *)(skb->data + offset) = crc32c_csum; 3294 skb->ip_summed = CHECKSUM_NONE; 3295 skb->csum_not_inet = 0; 3296 out: 3297 return ret; 3298 } 3299 3300 __be16 skb_network_protocol(struct sk_buff *skb, int *depth) 3301 { 3302 __be16 type = skb->protocol; 3303 3304 /* Tunnel gso handlers can set protocol to ethernet. */ 3305 if (type == htons(ETH_P_TEB)) { 3306 struct ethhdr *eth; 3307 3308 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) 3309 return 0; 3310 3311 eth = (struct ethhdr *)skb->data; 3312 type = eth->h_proto; 3313 } 3314 3315 return __vlan_get_protocol(skb, type, depth); 3316 } 3317 3318 /** 3319 * skb_mac_gso_segment - mac layer segmentation handler. 3320 * @skb: buffer to segment 3321 * @features: features for the output path (see dev->features) 3322 */ 3323 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb, 3324 netdev_features_t features) 3325 { 3326 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); 3327 struct packet_offload *ptype; 3328 int vlan_depth = skb->mac_len; 3329 __be16 type = skb_network_protocol(skb, &vlan_depth); 3330 3331 if (unlikely(!type)) 3332 return ERR_PTR(-EINVAL); 3333 3334 __skb_pull(skb, vlan_depth); 3335 3336 rcu_read_lock(); 3337 list_for_each_entry_rcu(ptype, &offload_base, list) { 3338 if (ptype->type == type && ptype->callbacks.gso_segment) { 3339 segs = ptype->callbacks.gso_segment(skb, features); 3340 break; 3341 } 3342 } 3343 rcu_read_unlock(); 3344 3345 __skb_push(skb, skb->data - skb_mac_header(skb)); 3346 3347 return segs; 3348 } 3349 EXPORT_SYMBOL(skb_mac_gso_segment); 3350 3351 3352 /* openvswitch calls this on rx path, so we need a different check. 3353 */ 3354 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path) 3355 { 3356 if (tx_path) 3357 return skb->ip_summed != CHECKSUM_PARTIAL && 3358 skb->ip_summed != CHECKSUM_UNNECESSARY; 3359 3360 return skb->ip_summed == CHECKSUM_NONE; 3361 } 3362 3363 /** 3364 * __skb_gso_segment - Perform segmentation on skb. 3365 * @skb: buffer to segment 3366 * @features: features for the output path (see dev->features) 3367 * @tx_path: whether it is called in TX path 3368 * 3369 * This function segments the given skb and returns a list of segments. 3370 * 3371 * It may return NULL if the skb requires no segmentation. This is 3372 * only possible when GSO is used for verifying header integrity. 3373 * 3374 * Segmentation preserves SKB_GSO_CB_OFFSET bytes of previous skb cb. 3375 */ 3376 struct sk_buff *__skb_gso_segment(struct sk_buff *skb, 3377 netdev_features_t features, bool tx_path) 3378 { 3379 struct sk_buff *segs; 3380 3381 if (unlikely(skb_needs_check(skb, tx_path))) { 3382 int err; 3383 3384 /* We're going to init ->check field in TCP or UDP header */ 3385 err = skb_cow_head(skb, 0); 3386 if (err < 0) 3387 return ERR_PTR(err); 3388 } 3389 3390 /* Only report GSO partial support if it will enable us to 3391 * support segmentation on this frame without needing additional 3392 * work. 3393 */ 3394 if (features & NETIF_F_GSO_PARTIAL) { 3395 netdev_features_t partial_features = NETIF_F_GSO_ROBUST; 3396 struct net_device *dev = skb->dev; 3397 3398 partial_features |= dev->features & dev->gso_partial_features; 3399 if (!skb_gso_ok(skb, features | partial_features)) 3400 features &= ~NETIF_F_GSO_PARTIAL; 3401 } 3402 3403 BUILD_BUG_ON(SKB_GSO_CB_OFFSET + 3404 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb)); 3405 3406 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb); 3407 SKB_GSO_CB(skb)->encap_level = 0; 3408 3409 skb_reset_mac_header(skb); 3410 skb_reset_mac_len(skb); 3411 3412 segs = skb_mac_gso_segment(skb, features); 3413 3414 if (segs != skb && unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs))) 3415 skb_warn_bad_offload(skb); 3416 3417 return segs; 3418 } 3419 EXPORT_SYMBOL(__skb_gso_segment); 3420 3421 /* Take action when hardware reception checksum errors are detected. */ 3422 #ifdef CONFIG_BUG 3423 static void do_netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3424 { 3425 netdev_err(dev, "hw csum failure\n"); 3426 skb_dump(KERN_ERR, skb, true); 3427 dump_stack(); 3428 } 3429 3430 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3431 { 3432 DO_ONCE_LITE(do_netdev_rx_csum_fault, dev, skb); 3433 } 3434 EXPORT_SYMBOL(netdev_rx_csum_fault); 3435 #endif 3436 3437 /* XXX: check that highmem exists at all on the given machine. */ 3438 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) 3439 { 3440 #ifdef CONFIG_HIGHMEM 3441 int i; 3442 3443 if (!(dev->features & NETIF_F_HIGHDMA)) { 3444 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 3445 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 3446 3447 if (PageHighMem(skb_frag_page(frag))) 3448 return 1; 3449 } 3450 } 3451 #endif 3452 return 0; 3453 } 3454 3455 /* If MPLS offload request, verify we are testing hardware MPLS features 3456 * instead of standard features for the netdev. 3457 */ 3458 #if IS_ENABLED(CONFIG_NET_MPLS_GSO) 3459 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3460 netdev_features_t features, 3461 __be16 type) 3462 { 3463 if (eth_p_mpls(type)) 3464 features &= skb->dev->mpls_features; 3465 3466 return features; 3467 } 3468 #else 3469 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3470 netdev_features_t features, 3471 __be16 type) 3472 { 3473 return features; 3474 } 3475 #endif 3476 3477 static netdev_features_t harmonize_features(struct sk_buff *skb, 3478 netdev_features_t features) 3479 { 3480 __be16 type; 3481 3482 type = skb_network_protocol(skb, NULL); 3483 features = net_mpls_features(skb, features, type); 3484 3485 if (skb->ip_summed != CHECKSUM_NONE && 3486 !can_checksum_protocol(features, type)) { 3487 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); 3488 } 3489 if (illegal_highdma(skb->dev, skb)) 3490 features &= ~NETIF_F_SG; 3491 3492 return features; 3493 } 3494 3495 netdev_features_t passthru_features_check(struct sk_buff *skb, 3496 struct net_device *dev, 3497 netdev_features_t features) 3498 { 3499 return features; 3500 } 3501 EXPORT_SYMBOL(passthru_features_check); 3502 3503 static netdev_features_t dflt_features_check(struct sk_buff *skb, 3504 struct net_device *dev, 3505 netdev_features_t features) 3506 { 3507 return vlan_features_check(skb, features); 3508 } 3509 3510 static netdev_features_t gso_features_check(const struct sk_buff *skb, 3511 struct net_device *dev, 3512 netdev_features_t features) 3513 { 3514 u16 gso_segs = skb_shinfo(skb)->gso_segs; 3515 3516 if (gso_segs > dev->gso_max_segs) 3517 return features & ~NETIF_F_GSO_MASK; 3518 3519 if (!skb_shinfo(skb)->gso_type) { 3520 skb_warn_bad_offload(skb); 3521 return features & ~NETIF_F_GSO_MASK; 3522 } 3523 3524 /* Support for GSO partial features requires software 3525 * intervention before we can actually process the packets 3526 * so we need to strip support for any partial features now 3527 * and we can pull them back in after we have partially 3528 * segmented the frame. 3529 */ 3530 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)) 3531 features &= ~dev->gso_partial_features; 3532 3533 /* Make sure to clear the IPv4 ID mangling feature if the 3534 * IPv4 header has the potential to be fragmented. 3535 */ 3536 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { 3537 struct iphdr *iph = skb->encapsulation ? 3538 inner_ip_hdr(skb) : ip_hdr(skb); 3539 3540 if (!(iph->frag_off & htons(IP_DF))) 3541 features &= ~NETIF_F_TSO_MANGLEID; 3542 } 3543 3544 return features; 3545 } 3546 3547 netdev_features_t netif_skb_features(struct sk_buff *skb) 3548 { 3549 struct net_device *dev = skb->dev; 3550 netdev_features_t features = dev->features; 3551 3552 if (skb_is_gso(skb)) 3553 features = gso_features_check(skb, dev, features); 3554 3555 /* If encapsulation offload request, verify we are testing 3556 * hardware encapsulation features instead of standard 3557 * features for the netdev 3558 */ 3559 if (skb->encapsulation) 3560 features &= dev->hw_enc_features; 3561 3562 if (skb_vlan_tagged(skb)) 3563 features = netdev_intersect_features(features, 3564 dev->vlan_features | 3565 NETIF_F_HW_VLAN_CTAG_TX | 3566 NETIF_F_HW_VLAN_STAG_TX); 3567 3568 if (dev->netdev_ops->ndo_features_check) 3569 features &= dev->netdev_ops->ndo_features_check(skb, dev, 3570 features); 3571 else 3572 features &= dflt_features_check(skb, dev, features); 3573 3574 return harmonize_features(skb, features); 3575 } 3576 EXPORT_SYMBOL(netif_skb_features); 3577 3578 static int xmit_one(struct sk_buff *skb, struct net_device *dev, 3579 struct netdev_queue *txq, bool more) 3580 { 3581 unsigned int len; 3582 int rc; 3583 3584 if (dev_nit_active(dev)) 3585 dev_queue_xmit_nit(skb, dev); 3586 3587 len = skb->len; 3588 PRANDOM_ADD_NOISE(skb, dev, txq, len + jiffies); 3589 trace_net_dev_start_xmit(skb, dev); 3590 rc = netdev_start_xmit(skb, dev, txq, more); 3591 trace_net_dev_xmit(skb, rc, dev, len); 3592 3593 return rc; 3594 } 3595 3596 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, 3597 struct netdev_queue *txq, int *ret) 3598 { 3599 struct sk_buff *skb = first; 3600 int rc = NETDEV_TX_OK; 3601 3602 while (skb) { 3603 struct sk_buff *next = skb->next; 3604 3605 skb_mark_not_on_list(skb); 3606 rc = xmit_one(skb, dev, txq, next != NULL); 3607 if (unlikely(!dev_xmit_complete(rc))) { 3608 skb->next = next; 3609 goto out; 3610 } 3611 3612 skb = next; 3613 if (netif_tx_queue_stopped(txq) && skb) { 3614 rc = NETDEV_TX_BUSY; 3615 break; 3616 } 3617 } 3618 3619 out: 3620 *ret = rc; 3621 return skb; 3622 } 3623 3624 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, 3625 netdev_features_t features) 3626 { 3627 if (skb_vlan_tag_present(skb) && 3628 !vlan_hw_offload_capable(features, skb->vlan_proto)) 3629 skb = __vlan_hwaccel_push_inside(skb); 3630 return skb; 3631 } 3632 3633 int skb_csum_hwoffload_help(struct sk_buff *skb, 3634 const netdev_features_t features) 3635 { 3636 if (unlikely(skb_csum_is_sctp(skb))) 3637 return !!(features & NETIF_F_SCTP_CRC) ? 0 : 3638 skb_crc32c_csum_help(skb); 3639 3640 if (features & NETIF_F_HW_CSUM) 3641 return 0; 3642 3643 if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) { 3644 switch (skb->csum_offset) { 3645 case offsetof(struct tcphdr, check): 3646 case offsetof(struct udphdr, check): 3647 return 0; 3648 } 3649 } 3650 3651 return skb_checksum_help(skb); 3652 } 3653 EXPORT_SYMBOL(skb_csum_hwoffload_help); 3654 3655 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again) 3656 { 3657 netdev_features_t features; 3658 3659 features = netif_skb_features(skb); 3660 skb = validate_xmit_vlan(skb, features); 3661 if (unlikely(!skb)) 3662 goto out_null; 3663 3664 skb = sk_validate_xmit_skb(skb, dev); 3665 if (unlikely(!skb)) 3666 goto out_null; 3667 3668 if (netif_needs_gso(skb, features)) { 3669 struct sk_buff *segs; 3670 3671 segs = skb_gso_segment(skb, features); 3672 if (IS_ERR(segs)) { 3673 goto out_kfree_skb; 3674 } else if (segs) { 3675 consume_skb(skb); 3676 skb = segs; 3677 } 3678 } else { 3679 if (skb_needs_linearize(skb, features) && 3680 __skb_linearize(skb)) 3681 goto out_kfree_skb; 3682 3683 /* If packet is not checksummed and device does not 3684 * support checksumming for this protocol, complete 3685 * checksumming here. 3686 */ 3687 if (skb->ip_summed == CHECKSUM_PARTIAL) { 3688 if (skb->encapsulation) 3689 skb_set_inner_transport_header(skb, 3690 skb_checksum_start_offset(skb)); 3691 else 3692 skb_set_transport_header(skb, 3693 skb_checksum_start_offset(skb)); 3694 if (skb_csum_hwoffload_help(skb, features)) 3695 goto out_kfree_skb; 3696 } 3697 } 3698 3699 skb = validate_xmit_xfrm(skb, features, again); 3700 3701 return skb; 3702 3703 out_kfree_skb: 3704 kfree_skb(skb); 3705 out_null: 3706 atomic_long_inc(&dev->tx_dropped); 3707 return NULL; 3708 } 3709 3710 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again) 3711 { 3712 struct sk_buff *next, *head = NULL, *tail; 3713 3714 for (; skb != NULL; skb = next) { 3715 next = skb->next; 3716 skb_mark_not_on_list(skb); 3717 3718 /* in case skb wont be segmented, point to itself */ 3719 skb->prev = skb; 3720 3721 skb = validate_xmit_skb(skb, dev, again); 3722 if (!skb) 3723 continue; 3724 3725 if (!head) 3726 head = skb; 3727 else 3728 tail->next = skb; 3729 /* If skb was segmented, skb->prev points to 3730 * the last segment. If not, it still contains skb. 3731 */ 3732 tail = skb->prev; 3733 } 3734 return head; 3735 } 3736 EXPORT_SYMBOL_GPL(validate_xmit_skb_list); 3737 3738 static void qdisc_pkt_len_init(struct sk_buff *skb) 3739 { 3740 const struct skb_shared_info *shinfo = skb_shinfo(skb); 3741 3742 qdisc_skb_cb(skb)->pkt_len = skb->len; 3743 3744 /* To get more precise estimation of bytes sent on wire, 3745 * we add to pkt_len the headers size of all segments 3746 */ 3747 if (shinfo->gso_size && skb_transport_header_was_set(skb)) { 3748 unsigned int hdr_len; 3749 u16 gso_segs = shinfo->gso_segs; 3750 3751 /* mac layer + network layer */ 3752 hdr_len = skb_transport_header(skb) - skb_mac_header(skb); 3753 3754 /* + transport layer */ 3755 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { 3756 const struct tcphdr *th; 3757 struct tcphdr _tcphdr; 3758 3759 th = skb_header_pointer(skb, skb_transport_offset(skb), 3760 sizeof(_tcphdr), &_tcphdr); 3761 if (likely(th)) 3762 hdr_len += __tcp_hdrlen(th); 3763 } else { 3764 struct udphdr _udphdr; 3765 3766 if (skb_header_pointer(skb, skb_transport_offset(skb), 3767 sizeof(_udphdr), &_udphdr)) 3768 hdr_len += sizeof(struct udphdr); 3769 } 3770 3771 if (shinfo->gso_type & SKB_GSO_DODGY) 3772 gso_segs = DIV_ROUND_UP(skb->len - hdr_len, 3773 shinfo->gso_size); 3774 3775 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; 3776 } 3777 } 3778 3779 static int dev_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *q, 3780 struct sk_buff **to_free, 3781 struct netdev_queue *txq) 3782 { 3783 int rc; 3784 3785 rc = q->enqueue(skb, q, to_free) & NET_XMIT_MASK; 3786 if (rc == NET_XMIT_SUCCESS) 3787 trace_qdisc_enqueue(q, txq, skb); 3788 return rc; 3789 } 3790 3791 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, 3792 struct net_device *dev, 3793 struct netdev_queue *txq) 3794 { 3795 spinlock_t *root_lock = qdisc_lock(q); 3796 struct sk_buff *to_free = NULL; 3797 bool contended; 3798 int rc; 3799 3800 qdisc_calculate_pkt_len(skb, q); 3801 3802 if (q->flags & TCQ_F_NOLOCK) { 3803 if (q->flags & TCQ_F_CAN_BYPASS && nolock_qdisc_is_empty(q) && 3804 qdisc_run_begin(q)) { 3805 /* Retest nolock_qdisc_is_empty() within the protection 3806 * of q->seqlock to protect from racing with requeuing. 3807 */ 3808 if (unlikely(!nolock_qdisc_is_empty(q))) { 3809 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3810 __qdisc_run(q); 3811 qdisc_run_end(q); 3812 3813 goto no_lock_out; 3814 } 3815 3816 qdisc_bstats_cpu_update(q, skb); 3817 if (sch_direct_xmit(skb, q, dev, txq, NULL, true) && 3818 !nolock_qdisc_is_empty(q)) 3819 __qdisc_run(q); 3820 3821 qdisc_run_end(q); 3822 return NET_XMIT_SUCCESS; 3823 } 3824 3825 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3826 qdisc_run(q); 3827 3828 no_lock_out: 3829 if (unlikely(to_free)) 3830 kfree_skb_list(to_free); 3831 return rc; 3832 } 3833 3834 /* 3835 * Heuristic to force contended enqueues to serialize on a 3836 * separate lock before trying to get qdisc main lock. 3837 * This permits qdisc->running owner to get the lock more 3838 * often and dequeue packets faster. 3839 */ 3840 contended = qdisc_is_running(q); 3841 if (unlikely(contended)) 3842 spin_lock(&q->busylock); 3843 3844 spin_lock(root_lock); 3845 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { 3846 __qdisc_drop(skb, &to_free); 3847 rc = NET_XMIT_DROP; 3848 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && 3849 qdisc_run_begin(q)) { 3850 /* 3851 * This is a work-conserving queue; there are no old skbs 3852 * waiting to be sent out; and the qdisc is not running - 3853 * xmit the skb directly. 3854 */ 3855 3856 qdisc_bstats_update(q, skb); 3857 3858 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { 3859 if (unlikely(contended)) { 3860 spin_unlock(&q->busylock); 3861 contended = false; 3862 } 3863 __qdisc_run(q); 3864 } 3865 3866 qdisc_run_end(q); 3867 rc = NET_XMIT_SUCCESS; 3868 } else { 3869 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3870 if (qdisc_run_begin(q)) { 3871 if (unlikely(contended)) { 3872 spin_unlock(&q->busylock); 3873 contended = false; 3874 } 3875 __qdisc_run(q); 3876 qdisc_run_end(q); 3877 } 3878 } 3879 spin_unlock(root_lock); 3880 if (unlikely(to_free)) 3881 kfree_skb_list(to_free); 3882 if (unlikely(contended)) 3883 spin_unlock(&q->busylock); 3884 return rc; 3885 } 3886 3887 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) 3888 static void skb_update_prio(struct sk_buff *skb) 3889 { 3890 const struct netprio_map *map; 3891 const struct sock *sk; 3892 unsigned int prioidx; 3893 3894 if (skb->priority) 3895 return; 3896 map = rcu_dereference_bh(skb->dev->priomap); 3897 if (!map) 3898 return; 3899 sk = skb_to_full_sk(skb); 3900 if (!sk) 3901 return; 3902 3903 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data); 3904 3905 if (prioidx < map->priomap_len) 3906 skb->priority = map->priomap[prioidx]; 3907 } 3908 #else 3909 #define skb_update_prio(skb) 3910 #endif 3911 3912 /** 3913 * dev_loopback_xmit - loop back @skb 3914 * @net: network namespace this loopback is happening in 3915 * @sk: sk needed to be a netfilter okfn 3916 * @skb: buffer to transmit 3917 */ 3918 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) 3919 { 3920 skb_reset_mac_header(skb); 3921 __skb_pull(skb, skb_network_offset(skb)); 3922 skb->pkt_type = PACKET_LOOPBACK; 3923 if (skb->ip_summed == CHECKSUM_NONE) 3924 skb->ip_summed = CHECKSUM_UNNECESSARY; 3925 WARN_ON(!skb_dst(skb)); 3926 skb_dst_force(skb); 3927 netif_rx_ni(skb); 3928 return 0; 3929 } 3930 EXPORT_SYMBOL(dev_loopback_xmit); 3931 3932 #ifdef CONFIG_NET_EGRESS 3933 static struct sk_buff * 3934 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 3935 { 3936 #ifdef CONFIG_NET_CLS_ACT 3937 struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress); 3938 struct tcf_result cl_res; 3939 3940 if (!miniq) 3941 return skb; 3942 3943 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */ 3944 qdisc_skb_cb(skb)->mru = 0; 3945 qdisc_skb_cb(skb)->post_ct = false; 3946 mini_qdisc_bstats_cpu_update(miniq, skb); 3947 3948 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) { 3949 case TC_ACT_OK: 3950 case TC_ACT_RECLASSIFY: 3951 skb->tc_index = TC_H_MIN(cl_res.classid); 3952 break; 3953 case TC_ACT_SHOT: 3954 mini_qdisc_qstats_cpu_drop(miniq); 3955 *ret = NET_XMIT_DROP; 3956 kfree_skb(skb); 3957 return NULL; 3958 case TC_ACT_STOLEN: 3959 case TC_ACT_QUEUED: 3960 case TC_ACT_TRAP: 3961 *ret = NET_XMIT_SUCCESS; 3962 consume_skb(skb); 3963 return NULL; 3964 case TC_ACT_REDIRECT: 3965 /* No need to push/pop skb's mac_header here on egress! */ 3966 skb_do_redirect(skb); 3967 *ret = NET_XMIT_SUCCESS; 3968 return NULL; 3969 default: 3970 break; 3971 } 3972 #endif /* CONFIG_NET_CLS_ACT */ 3973 3974 return skb; 3975 } 3976 #endif /* CONFIG_NET_EGRESS */ 3977 3978 #ifdef CONFIG_XPS 3979 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb, 3980 struct xps_dev_maps *dev_maps, unsigned int tci) 3981 { 3982 int tc = netdev_get_prio_tc_map(dev, skb->priority); 3983 struct xps_map *map; 3984 int queue_index = -1; 3985 3986 if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids) 3987 return queue_index; 3988 3989 tci *= dev_maps->num_tc; 3990 tci += tc; 3991 3992 map = rcu_dereference(dev_maps->attr_map[tci]); 3993 if (map) { 3994 if (map->len == 1) 3995 queue_index = map->queues[0]; 3996 else 3997 queue_index = map->queues[reciprocal_scale( 3998 skb_get_hash(skb), map->len)]; 3999 if (unlikely(queue_index >= dev->real_num_tx_queues)) 4000 queue_index = -1; 4001 } 4002 return queue_index; 4003 } 4004 #endif 4005 4006 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev, 4007 struct sk_buff *skb) 4008 { 4009 #ifdef CONFIG_XPS 4010 struct xps_dev_maps *dev_maps; 4011 struct sock *sk = skb->sk; 4012 int queue_index = -1; 4013 4014 if (!static_key_false(&xps_needed)) 4015 return -1; 4016 4017 rcu_read_lock(); 4018 if (!static_key_false(&xps_rxqs_needed)) 4019 goto get_cpus_map; 4020 4021 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]); 4022 if (dev_maps) { 4023 int tci = sk_rx_queue_get(sk); 4024 4025 if (tci >= 0) 4026 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4027 tci); 4028 } 4029 4030 get_cpus_map: 4031 if (queue_index < 0) { 4032 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]); 4033 if (dev_maps) { 4034 unsigned int tci = skb->sender_cpu - 1; 4035 4036 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4037 tci); 4038 } 4039 } 4040 rcu_read_unlock(); 4041 4042 return queue_index; 4043 #else 4044 return -1; 4045 #endif 4046 } 4047 4048 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb, 4049 struct net_device *sb_dev) 4050 { 4051 return 0; 4052 } 4053 EXPORT_SYMBOL(dev_pick_tx_zero); 4054 4055 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb, 4056 struct net_device *sb_dev) 4057 { 4058 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues; 4059 } 4060 EXPORT_SYMBOL(dev_pick_tx_cpu_id); 4061 4062 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb, 4063 struct net_device *sb_dev) 4064 { 4065 struct sock *sk = skb->sk; 4066 int queue_index = sk_tx_queue_get(sk); 4067 4068 sb_dev = sb_dev ? : dev; 4069 4070 if (queue_index < 0 || skb->ooo_okay || 4071 queue_index >= dev->real_num_tx_queues) { 4072 int new_index = get_xps_queue(dev, sb_dev, skb); 4073 4074 if (new_index < 0) 4075 new_index = skb_tx_hash(dev, sb_dev, skb); 4076 4077 if (queue_index != new_index && sk && 4078 sk_fullsock(sk) && 4079 rcu_access_pointer(sk->sk_dst_cache)) 4080 sk_tx_queue_set(sk, new_index); 4081 4082 queue_index = new_index; 4083 } 4084 4085 return queue_index; 4086 } 4087 EXPORT_SYMBOL(netdev_pick_tx); 4088 4089 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev, 4090 struct sk_buff *skb, 4091 struct net_device *sb_dev) 4092 { 4093 int queue_index = 0; 4094 4095 #ifdef CONFIG_XPS 4096 u32 sender_cpu = skb->sender_cpu - 1; 4097 4098 if (sender_cpu >= (u32)NR_CPUS) 4099 skb->sender_cpu = raw_smp_processor_id() + 1; 4100 #endif 4101 4102 if (dev->real_num_tx_queues != 1) { 4103 const struct net_device_ops *ops = dev->netdev_ops; 4104 4105 if (ops->ndo_select_queue) 4106 queue_index = ops->ndo_select_queue(dev, skb, sb_dev); 4107 else 4108 queue_index = netdev_pick_tx(dev, skb, sb_dev); 4109 4110 queue_index = netdev_cap_txqueue(dev, queue_index); 4111 } 4112 4113 skb_set_queue_mapping(skb, queue_index); 4114 return netdev_get_tx_queue(dev, queue_index); 4115 } 4116 4117 /** 4118 * __dev_queue_xmit - transmit a buffer 4119 * @skb: buffer to transmit 4120 * @sb_dev: suboordinate device used for L2 forwarding offload 4121 * 4122 * Queue a buffer for transmission to a network device. The caller must 4123 * have set the device and priority and built the buffer before calling 4124 * this function. The function can be called from an interrupt. 4125 * 4126 * A negative errno code is returned on a failure. A success does not 4127 * guarantee the frame will be transmitted as it may be dropped due 4128 * to congestion or traffic shaping. 4129 * 4130 * ----------------------------------------------------------------------------------- 4131 * I notice this method can also return errors from the queue disciplines, 4132 * including NET_XMIT_DROP, which is a positive value. So, errors can also 4133 * be positive. 4134 * 4135 * Regardless of the return value, the skb is consumed, so it is currently 4136 * difficult to retry a send to this method. (You can bump the ref count 4137 * before sending to hold a reference for retry if you are careful.) 4138 * 4139 * When calling this method, interrupts MUST be enabled. This is because 4140 * the BH enable code must have IRQs enabled so that it will not deadlock. 4141 * --BLG 4142 */ 4143 static int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev) 4144 { 4145 struct net_device *dev = skb->dev; 4146 struct netdev_queue *txq; 4147 struct Qdisc *q; 4148 int rc = -ENOMEM; 4149 bool again = false; 4150 4151 skb_reset_mac_header(skb); 4152 4153 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) 4154 __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED); 4155 4156 /* Disable soft irqs for various locks below. Also 4157 * stops preemption for RCU. 4158 */ 4159 rcu_read_lock_bh(); 4160 4161 skb_update_prio(skb); 4162 4163 qdisc_pkt_len_init(skb); 4164 #ifdef CONFIG_NET_CLS_ACT 4165 skb->tc_at_ingress = 0; 4166 #endif 4167 #ifdef CONFIG_NET_EGRESS 4168 if (static_branch_unlikely(&egress_needed_key)) { 4169 if (nf_hook_egress_active()) { 4170 skb = nf_hook_egress(skb, &rc, dev); 4171 if (!skb) 4172 goto out; 4173 } 4174 nf_skip_egress(skb, true); 4175 skb = sch_handle_egress(skb, &rc, dev); 4176 if (!skb) 4177 goto out; 4178 nf_skip_egress(skb, false); 4179 } 4180 #endif 4181 /* If device/qdisc don't need skb->dst, release it right now while 4182 * its hot in this cpu cache. 4183 */ 4184 if (dev->priv_flags & IFF_XMIT_DST_RELEASE) 4185 skb_dst_drop(skb); 4186 else 4187 skb_dst_force(skb); 4188 4189 txq = netdev_core_pick_tx(dev, skb, sb_dev); 4190 q = rcu_dereference_bh(txq->qdisc); 4191 4192 trace_net_dev_queue(skb); 4193 if (q->enqueue) { 4194 rc = __dev_xmit_skb(skb, q, dev, txq); 4195 goto out; 4196 } 4197 4198 /* The device has no queue. Common case for software devices: 4199 * loopback, all the sorts of tunnels... 4200 4201 * Really, it is unlikely that netif_tx_lock protection is necessary 4202 * here. (f.e. loopback and IP tunnels are clean ignoring statistics 4203 * counters.) 4204 * However, it is possible, that they rely on protection 4205 * made by us here. 4206 4207 * Check this and shot the lock. It is not prone from deadlocks. 4208 *Either shot noqueue qdisc, it is even simpler 8) 4209 */ 4210 if (dev->flags & IFF_UP) { 4211 int cpu = smp_processor_id(); /* ok because BHs are off */ 4212 4213 /* Other cpus might concurrently change txq->xmit_lock_owner 4214 * to -1 or to their cpu id, but not to our id. 4215 */ 4216 if (READ_ONCE(txq->xmit_lock_owner) != cpu) { 4217 if (dev_xmit_recursion()) 4218 goto recursion_alert; 4219 4220 skb = validate_xmit_skb(skb, dev, &again); 4221 if (!skb) 4222 goto out; 4223 4224 PRANDOM_ADD_NOISE(skb, dev, txq, jiffies); 4225 HARD_TX_LOCK(dev, txq, cpu); 4226 4227 if (!netif_xmit_stopped(txq)) { 4228 dev_xmit_recursion_inc(); 4229 skb = dev_hard_start_xmit(skb, dev, txq, &rc); 4230 dev_xmit_recursion_dec(); 4231 if (dev_xmit_complete(rc)) { 4232 HARD_TX_UNLOCK(dev, txq); 4233 goto out; 4234 } 4235 } 4236 HARD_TX_UNLOCK(dev, txq); 4237 net_crit_ratelimited("Virtual device %s asks to queue packet!\n", 4238 dev->name); 4239 } else { 4240 /* Recursion is detected! It is possible, 4241 * unfortunately 4242 */ 4243 recursion_alert: 4244 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", 4245 dev->name); 4246 } 4247 } 4248 4249 rc = -ENETDOWN; 4250 rcu_read_unlock_bh(); 4251 4252 atomic_long_inc(&dev->tx_dropped); 4253 kfree_skb_list(skb); 4254 return rc; 4255 out: 4256 rcu_read_unlock_bh(); 4257 return rc; 4258 } 4259 4260 int dev_queue_xmit(struct sk_buff *skb) 4261 { 4262 return __dev_queue_xmit(skb, NULL); 4263 } 4264 EXPORT_SYMBOL(dev_queue_xmit); 4265 4266 int dev_queue_xmit_accel(struct sk_buff *skb, struct net_device *sb_dev) 4267 { 4268 return __dev_queue_xmit(skb, sb_dev); 4269 } 4270 EXPORT_SYMBOL(dev_queue_xmit_accel); 4271 4272 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id) 4273 { 4274 struct net_device *dev = skb->dev; 4275 struct sk_buff *orig_skb = skb; 4276 struct netdev_queue *txq; 4277 int ret = NETDEV_TX_BUSY; 4278 bool again = false; 4279 4280 if (unlikely(!netif_running(dev) || 4281 !netif_carrier_ok(dev))) 4282 goto drop; 4283 4284 skb = validate_xmit_skb_list(skb, dev, &again); 4285 if (skb != orig_skb) 4286 goto drop; 4287 4288 skb_set_queue_mapping(skb, queue_id); 4289 txq = skb_get_tx_queue(dev, skb); 4290 PRANDOM_ADD_NOISE(skb, dev, txq, jiffies); 4291 4292 local_bh_disable(); 4293 4294 dev_xmit_recursion_inc(); 4295 HARD_TX_LOCK(dev, txq, smp_processor_id()); 4296 if (!netif_xmit_frozen_or_drv_stopped(txq)) 4297 ret = netdev_start_xmit(skb, dev, txq, false); 4298 HARD_TX_UNLOCK(dev, txq); 4299 dev_xmit_recursion_dec(); 4300 4301 local_bh_enable(); 4302 return ret; 4303 drop: 4304 atomic_long_inc(&dev->tx_dropped); 4305 kfree_skb_list(skb); 4306 return NET_XMIT_DROP; 4307 } 4308 EXPORT_SYMBOL(__dev_direct_xmit); 4309 4310 /************************************************************************* 4311 * Receiver routines 4312 *************************************************************************/ 4313 4314 int netdev_max_backlog __read_mostly = 1000; 4315 EXPORT_SYMBOL(netdev_max_backlog); 4316 4317 int netdev_tstamp_prequeue __read_mostly = 1; 4318 int netdev_budget __read_mostly = 300; 4319 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */ 4320 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ; 4321 int weight_p __read_mostly = 64; /* old backlog weight */ 4322 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ 4323 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ 4324 int dev_rx_weight __read_mostly = 64; 4325 int dev_tx_weight __read_mostly = 64; 4326 /* Maximum number of GRO_NORMAL skbs to batch up for list-RX */ 4327 int gro_normal_batch __read_mostly = 8; 4328 4329 /* Called with irq disabled */ 4330 static inline void ____napi_schedule(struct softnet_data *sd, 4331 struct napi_struct *napi) 4332 { 4333 struct task_struct *thread; 4334 4335 if (test_bit(NAPI_STATE_THREADED, &napi->state)) { 4336 /* Paired with smp_mb__before_atomic() in 4337 * napi_enable()/dev_set_threaded(). 4338 * Use READ_ONCE() to guarantee a complete 4339 * read on napi->thread. Only call 4340 * wake_up_process() when it's not NULL. 4341 */ 4342 thread = READ_ONCE(napi->thread); 4343 if (thread) { 4344 /* Avoid doing set_bit() if the thread is in 4345 * INTERRUPTIBLE state, cause napi_thread_wait() 4346 * makes sure to proceed with napi polling 4347 * if the thread is explicitly woken from here. 4348 */ 4349 if (READ_ONCE(thread->__state) != TASK_INTERRUPTIBLE) 4350 set_bit(NAPI_STATE_SCHED_THREADED, &napi->state); 4351 wake_up_process(thread); 4352 return; 4353 } 4354 } 4355 4356 list_add_tail(&napi->poll_list, &sd->poll_list); 4357 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4358 } 4359 4360 #ifdef CONFIG_RPS 4361 4362 /* One global table that all flow-based protocols share. */ 4363 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; 4364 EXPORT_SYMBOL(rps_sock_flow_table); 4365 u32 rps_cpu_mask __read_mostly; 4366 EXPORT_SYMBOL(rps_cpu_mask); 4367 4368 struct static_key_false rps_needed __read_mostly; 4369 EXPORT_SYMBOL(rps_needed); 4370 struct static_key_false rfs_needed __read_mostly; 4371 EXPORT_SYMBOL(rfs_needed); 4372 4373 static struct rps_dev_flow * 4374 set_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4375 struct rps_dev_flow *rflow, u16 next_cpu) 4376 { 4377 if (next_cpu < nr_cpu_ids) { 4378 #ifdef CONFIG_RFS_ACCEL 4379 struct netdev_rx_queue *rxqueue; 4380 struct rps_dev_flow_table *flow_table; 4381 struct rps_dev_flow *old_rflow; 4382 u32 flow_id; 4383 u16 rxq_index; 4384 int rc; 4385 4386 /* Should we steer this flow to a different hardware queue? */ 4387 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || 4388 !(dev->features & NETIF_F_NTUPLE)) 4389 goto out; 4390 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); 4391 if (rxq_index == skb_get_rx_queue(skb)) 4392 goto out; 4393 4394 rxqueue = dev->_rx + rxq_index; 4395 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4396 if (!flow_table) 4397 goto out; 4398 flow_id = skb_get_hash(skb) & flow_table->mask; 4399 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, 4400 rxq_index, flow_id); 4401 if (rc < 0) 4402 goto out; 4403 old_rflow = rflow; 4404 rflow = &flow_table->flows[flow_id]; 4405 rflow->filter = rc; 4406 if (old_rflow->filter == rflow->filter) 4407 old_rflow->filter = RPS_NO_FILTER; 4408 out: 4409 #endif 4410 rflow->last_qtail = 4411 per_cpu(softnet_data, next_cpu).input_queue_head; 4412 } 4413 4414 rflow->cpu = next_cpu; 4415 return rflow; 4416 } 4417 4418 /* 4419 * get_rps_cpu is called from netif_receive_skb and returns the target 4420 * CPU from the RPS map of the receiving queue for a given skb. 4421 * rcu_read_lock must be held on entry. 4422 */ 4423 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4424 struct rps_dev_flow **rflowp) 4425 { 4426 const struct rps_sock_flow_table *sock_flow_table; 4427 struct netdev_rx_queue *rxqueue = dev->_rx; 4428 struct rps_dev_flow_table *flow_table; 4429 struct rps_map *map; 4430 int cpu = -1; 4431 u32 tcpu; 4432 u32 hash; 4433 4434 if (skb_rx_queue_recorded(skb)) { 4435 u16 index = skb_get_rx_queue(skb); 4436 4437 if (unlikely(index >= dev->real_num_rx_queues)) { 4438 WARN_ONCE(dev->real_num_rx_queues > 1, 4439 "%s received packet on queue %u, but number " 4440 "of RX queues is %u\n", 4441 dev->name, index, dev->real_num_rx_queues); 4442 goto done; 4443 } 4444 rxqueue += index; 4445 } 4446 4447 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ 4448 4449 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4450 map = rcu_dereference(rxqueue->rps_map); 4451 if (!flow_table && !map) 4452 goto done; 4453 4454 skb_reset_network_header(skb); 4455 hash = skb_get_hash(skb); 4456 if (!hash) 4457 goto done; 4458 4459 sock_flow_table = rcu_dereference(rps_sock_flow_table); 4460 if (flow_table && sock_flow_table) { 4461 struct rps_dev_flow *rflow; 4462 u32 next_cpu; 4463 u32 ident; 4464 4465 /* First check into global flow table if there is a match */ 4466 ident = sock_flow_table->ents[hash & sock_flow_table->mask]; 4467 if ((ident ^ hash) & ~rps_cpu_mask) 4468 goto try_rps; 4469 4470 next_cpu = ident & rps_cpu_mask; 4471 4472 /* OK, now we know there is a match, 4473 * we can look at the local (per receive queue) flow table 4474 */ 4475 rflow = &flow_table->flows[hash & flow_table->mask]; 4476 tcpu = rflow->cpu; 4477 4478 /* 4479 * If the desired CPU (where last recvmsg was done) is 4480 * different from current CPU (one in the rx-queue flow 4481 * table entry), switch if one of the following holds: 4482 * - Current CPU is unset (>= nr_cpu_ids). 4483 * - Current CPU is offline. 4484 * - The current CPU's queue tail has advanced beyond the 4485 * last packet that was enqueued using this table entry. 4486 * This guarantees that all previous packets for the flow 4487 * have been dequeued, thus preserving in order delivery. 4488 */ 4489 if (unlikely(tcpu != next_cpu) && 4490 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || 4491 ((int)(per_cpu(softnet_data, tcpu).input_queue_head - 4492 rflow->last_qtail)) >= 0)) { 4493 tcpu = next_cpu; 4494 rflow = set_rps_cpu(dev, skb, rflow, next_cpu); 4495 } 4496 4497 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { 4498 *rflowp = rflow; 4499 cpu = tcpu; 4500 goto done; 4501 } 4502 } 4503 4504 try_rps: 4505 4506 if (map) { 4507 tcpu = map->cpus[reciprocal_scale(hash, map->len)]; 4508 if (cpu_online(tcpu)) { 4509 cpu = tcpu; 4510 goto done; 4511 } 4512 } 4513 4514 done: 4515 return cpu; 4516 } 4517 4518 #ifdef CONFIG_RFS_ACCEL 4519 4520 /** 4521 * rps_may_expire_flow - check whether an RFS hardware filter may be removed 4522 * @dev: Device on which the filter was set 4523 * @rxq_index: RX queue index 4524 * @flow_id: Flow ID passed to ndo_rx_flow_steer() 4525 * @filter_id: Filter ID returned by ndo_rx_flow_steer() 4526 * 4527 * Drivers that implement ndo_rx_flow_steer() should periodically call 4528 * this function for each installed filter and remove the filters for 4529 * which it returns %true. 4530 */ 4531 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, 4532 u32 flow_id, u16 filter_id) 4533 { 4534 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; 4535 struct rps_dev_flow_table *flow_table; 4536 struct rps_dev_flow *rflow; 4537 bool expire = true; 4538 unsigned int cpu; 4539 4540 rcu_read_lock(); 4541 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4542 if (flow_table && flow_id <= flow_table->mask) { 4543 rflow = &flow_table->flows[flow_id]; 4544 cpu = READ_ONCE(rflow->cpu); 4545 if (rflow->filter == filter_id && cpu < nr_cpu_ids && 4546 ((int)(per_cpu(softnet_data, cpu).input_queue_head - 4547 rflow->last_qtail) < 4548 (int)(10 * flow_table->mask))) 4549 expire = false; 4550 } 4551 rcu_read_unlock(); 4552 return expire; 4553 } 4554 EXPORT_SYMBOL(rps_may_expire_flow); 4555 4556 #endif /* CONFIG_RFS_ACCEL */ 4557 4558 /* Called from hardirq (IPI) context */ 4559 static void rps_trigger_softirq(void *data) 4560 { 4561 struct softnet_data *sd = data; 4562 4563 ____napi_schedule(sd, &sd->backlog); 4564 sd->received_rps++; 4565 } 4566 4567 #endif /* CONFIG_RPS */ 4568 4569 /* 4570 * Check if this softnet_data structure is another cpu one 4571 * If yes, queue it to our IPI list and return 1 4572 * If no, return 0 4573 */ 4574 static int rps_ipi_queued(struct softnet_data *sd) 4575 { 4576 #ifdef CONFIG_RPS 4577 struct softnet_data *mysd = this_cpu_ptr(&softnet_data); 4578 4579 if (sd != mysd) { 4580 sd->rps_ipi_next = mysd->rps_ipi_list; 4581 mysd->rps_ipi_list = sd; 4582 4583 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4584 return 1; 4585 } 4586 #endif /* CONFIG_RPS */ 4587 return 0; 4588 } 4589 4590 #ifdef CONFIG_NET_FLOW_LIMIT 4591 int netdev_flow_limit_table_len __read_mostly = (1 << 12); 4592 #endif 4593 4594 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) 4595 { 4596 #ifdef CONFIG_NET_FLOW_LIMIT 4597 struct sd_flow_limit *fl; 4598 struct softnet_data *sd; 4599 unsigned int old_flow, new_flow; 4600 4601 if (qlen < (netdev_max_backlog >> 1)) 4602 return false; 4603 4604 sd = this_cpu_ptr(&softnet_data); 4605 4606 rcu_read_lock(); 4607 fl = rcu_dereference(sd->flow_limit); 4608 if (fl) { 4609 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); 4610 old_flow = fl->history[fl->history_head]; 4611 fl->history[fl->history_head] = new_flow; 4612 4613 fl->history_head++; 4614 fl->history_head &= FLOW_LIMIT_HISTORY - 1; 4615 4616 if (likely(fl->buckets[old_flow])) 4617 fl->buckets[old_flow]--; 4618 4619 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { 4620 fl->count++; 4621 rcu_read_unlock(); 4622 return true; 4623 } 4624 } 4625 rcu_read_unlock(); 4626 #endif 4627 return false; 4628 } 4629 4630 /* 4631 * enqueue_to_backlog is called to queue an skb to a per CPU backlog 4632 * queue (may be a remote CPU queue). 4633 */ 4634 static int enqueue_to_backlog(struct sk_buff *skb, int cpu, 4635 unsigned int *qtail) 4636 { 4637 struct softnet_data *sd; 4638 unsigned long flags; 4639 unsigned int qlen; 4640 4641 sd = &per_cpu(softnet_data, cpu); 4642 4643 local_irq_save(flags); 4644 4645 rps_lock(sd); 4646 if (!netif_running(skb->dev)) 4647 goto drop; 4648 qlen = skb_queue_len(&sd->input_pkt_queue); 4649 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) { 4650 if (qlen) { 4651 enqueue: 4652 __skb_queue_tail(&sd->input_pkt_queue, skb); 4653 input_queue_tail_incr_save(sd, qtail); 4654 rps_unlock(sd); 4655 local_irq_restore(flags); 4656 return NET_RX_SUCCESS; 4657 } 4658 4659 /* Schedule NAPI for backlog device 4660 * We can use non atomic operation since we own the queue lock 4661 */ 4662 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) { 4663 if (!rps_ipi_queued(sd)) 4664 ____napi_schedule(sd, &sd->backlog); 4665 } 4666 goto enqueue; 4667 } 4668 4669 drop: 4670 sd->dropped++; 4671 rps_unlock(sd); 4672 4673 local_irq_restore(flags); 4674 4675 atomic_long_inc(&skb->dev->rx_dropped); 4676 kfree_skb(skb); 4677 return NET_RX_DROP; 4678 } 4679 4680 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb) 4681 { 4682 struct net_device *dev = skb->dev; 4683 struct netdev_rx_queue *rxqueue; 4684 4685 rxqueue = dev->_rx; 4686 4687 if (skb_rx_queue_recorded(skb)) { 4688 u16 index = skb_get_rx_queue(skb); 4689 4690 if (unlikely(index >= dev->real_num_rx_queues)) { 4691 WARN_ONCE(dev->real_num_rx_queues > 1, 4692 "%s received packet on queue %u, but number " 4693 "of RX queues is %u\n", 4694 dev->name, index, dev->real_num_rx_queues); 4695 4696 return rxqueue; /* Return first rxqueue */ 4697 } 4698 rxqueue += index; 4699 } 4700 return rxqueue; 4701 } 4702 4703 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp, 4704 struct bpf_prog *xdp_prog) 4705 { 4706 void *orig_data, *orig_data_end, *hard_start; 4707 struct netdev_rx_queue *rxqueue; 4708 bool orig_bcast, orig_host; 4709 u32 mac_len, frame_sz; 4710 __be16 orig_eth_type; 4711 struct ethhdr *eth; 4712 u32 metalen, act; 4713 int off; 4714 4715 /* The XDP program wants to see the packet starting at the MAC 4716 * header. 4717 */ 4718 mac_len = skb->data - skb_mac_header(skb); 4719 hard_start = skb->data - skb_headroom(skb); 4720 4721 /* SKB "head" area always have tailroom for skb_shared_info */ 4722 frame_sz = (void *)skb_end_pointer(skb) - hard_start; 4723 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 4724 4725 rxqueue = netif_get_rxqueue(skb); 4726 xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq); 4727 xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len, 4728 skb_headlen(skb) + mac_len, true); 4729 4730 orig_data_end = xdp->data_end; 4731 orig_data = xdp->data; 4732 eth = (struct ethhdr *)xdp->data; 4733 orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr); 4734 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest); 4735 orig_eth_type = eth->h_proto; 4736 4737 act = bpf_prog_run_xdp(xdp_prog, xdp); 4738 4739 /* check if bpf_xdp_adjust_head was used */ 4740 off = xdp->data - orig_data; 4741 if (off) { 4742 if (off > 0) 4743 __skb_pull(skb, off); 4744 else if (off < 0) 4745 __skb_push(skb, -off); 4746 4747 skb->mac_header += off; 4748 skb_reset_network_header(skb); 4749 } 4750 4751 /* check if bpf_xdp_adjust_tail was used */ 4752 off = xdp->data_end - orig_data_end; 4753 if (off != 0) { 4754 skb_set_tail_pointer(skb, xdp->data_end - xdp->data); 4755 skb->len += off; /* positive on grow, negative on shrink */ 4756 } 4757 4758 /* check if XDP changed eth hdr such SKB needs update */ 4759 eth = (struct ethhdr *)xdp->data; 4760 if ((orig_eth_type != eth->h_proto) || 4761 (orig_host != ether_addr_equal_64bits(eth->h_dest, 4762 skb->dev->dev_addr)) || 4763 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) { 4764 __skb_push(skb, ETH_HLEN); 4765 skb->pkt_type = PACKET_HOST; 4766 skb->protocol = eth_type_trans(skb, skb->dev); 4767 } 4768 4769 /* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull 4770 * before calling us again on redirect path. We do not call do_redirect 4771 * as we leave that up to the caller. 4772 * 4773 * Caller is responsible for managing lifetime of skb (i.e. calling 4774 * kfree_skb in response to actions it cannot handle/XDP_DROP). 4775 */ 4776 switch (act) { 4777 case XDP_REDIRECT: 4778 case XDP_TX: 4779 __skb_push(skb, mac_len); 4780 break; 4781 case XDP_PASS: 4782 metalen = xdp->data - xdp->data_meta; 4783 if (metalen) 4784 skb_metadata_set(skb, metalen); 4785 break; 4786 } 4787 4788 return act; 4789 } 4790 4791 static u32 netif_receive_generic_xdp(struct sk_buff *skb, 4792 struct xdp_buff *xdp, 4793 struct bpf_prog *xdp_prog) 4794 { 4795 u32 act = XDP_DROP; 4796 4797 /* Reinjected packets coming from act_mirred or similar should 4798 * not get XDP generic processing. 4799 */ 4800 if (skb_is_redirected(skb)) 4801 return XDP_PASS; 4802 4803 /* XDP packets must be linear and must have sufficient headroom 4804 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also 4805 * native XDP provides, thus we need to do it here as well. 4806 */ 4807 if (skb_cloned(skb) || skb_is_nonlinear(skb) || 4808 skb_headroom(skb) < XDP_PACKET_HEADROOM) { 4809 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb); 4810 int troom = skb->tail + skb->data_len - skb->end; 4811 4812 /* In case we have to go down the path and also linearize, 4813 * then lets do the pskb_expand_head() work just once here. 4814 */ 4815 if (pskb_expand_head(skb, 4816 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0, 4817 troom > 0 ? troom + 128 : 0, GFP_ATOMIC)) 4818 goto do_drop; 4819 if (skb_linearize(skb)) 4820 goto do_drop; 4821 } 4822 4823 act = bpf_prog_run_generic_xdp(skb, xdp, xdp_prog); 4824 switch (act) { 4825 case XDP_REDIRECT: 4826 case XDP_TX: 4827 case XDP_PASS: 4828 break; 4829 default: 4830 bpf_warn_invalid_xdp_action(act); 4831 fallthrough; 4832 case XDP_ABORTED: 4833 trace_xdp_exception(skb->dev, xdp_prog, act); 4834 fallthrough; 4835 case XDP_DROP: 4836 do_drop: 4837 kfree_skb(skb); 4838 break; 4839 } 4840 4841 return act; 4842 } 4843 4844 /* When doing generic XDP we have to bypass the qdisc layer and the 4845 * network taps in order to match in-driver-XDP behavior. 4846 */ 4847 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog) 4848 { 4849 struct net_device *dev = skb->dev; 4850 struct netdev_queue *txq; 4851 bool free_skb = true; 4852 int cpu, rc; 4853 4854 txq = netdev_core_pick_tx(dev, skb, NULL); 4855 cpu = smp_processor_id(); 4856 HARD_TX_LOCK(dev, txq, cpu); 4857 if (!netif_xmit_stopped(txq)) { 4858 rc = netdev_start_xmit(skb, dev, txq, 0); 4859 if (dev_xmit_complete(rc)) 4860 free_skb = false; 4861 } 4862 HARD_TX_UNLOCK(dev, txq); 4863 if (free_skb) { 4864 trace_xdp_exception(dev, xdp_prog, XDP_TX); 4865 kfree_skb(skb); 4866 } 4867 } 4868 4869 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key); 4870 4871 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb) 4872 { 4873 if (xdp_prog) { 4874 struct xdp_buff xdp; 4875 u32 act; 4876 int err; 4877 4878 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog); 4879 if (act != XDP_PASS) { 4880 switch (act) { 4881 case XDP_REDIRECT: 4882 err = xdp_do_generic_redirect(skb->dev, skb, 4883 &xdp, xdp_prog); 4884 if (err) 4885 goto out_redir; 4886 break; 4887 case XDP_TX: 4888 generic_xdp_tx(skb, xdp_prog); 4889 break; 4890 } 4891 return XDP_DROP; 4892 } 4893 } 4894 return XDP_PASS; 4895 out_redir: 4896 kfree_skb(skb); 4897 return XDP_DROP; 4898 } 4899 EXPORT_SYMBOL_GPL(do_xdp_generic); 4900 4901 static int netif_rx_internal(struct sk_buff *skb) 4902 { 4903 int ret; 4904 4905 net_timestamp_check(netdev_tstamp_prequeue, skb); 4906 4907 trace_netif_rx(skb); 4908 4909 #ifdef CONFIG_RPS 4910 if (static_branch_unlikely(&rps_needed)) { 4911 struct rps_dev_flow voidflow, *rflow = &voidflow; 4912 int cpu; 4913 4914 preempt_disable(); 4915 rcu_read_lock(); 4916 4917 cpu = get_rps_cpu(skb->dev, skb, &rflow); 4918 if (cpu < 0) 4919 cpu = smp_processor_id(); 4920 4921 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 4922 4923 rcu_read_unlock(); 4924 preempt_enable(); 4925 } else 4926 #endif 4927 { 4928 unsigned int qtail; 4929 4930 ret = enqueue_to_backlog(skb, get_cpu(), &qtail); 4931 put_cpu(); 4932 } 4933 return ret; 4934 } 4935 4936 /** 4937 * netif_rx - post buffer to the network code 4938 * @skb: buffer to post 4939 * 4940 * This function receives a packet from a device driver and queues it for 4941 * the upper (protocol) levels to process. It always succeeds. The buffer 4942 * may be dropped during processing for congestion control or by the 4943 * protocol layers. 4944 * 4945 * return values: 4946 * NET_RX_SUCCESS (no congestion) 4947 * NET_RX_DROP (packet was dropped) 4948 * 4949 */ 4950 4951 int netif_rx(struct sk_buff *skb) 4952 { 4953 int ret; 4954 4955 trace_netif_rx_entry(skb); 4956 4957 ret = netif_rx_internal(skb); 4958 trace_netif_rx_exit(ret); 4959 4960 return ret; 4961 } 4962 EXPORT_SYMBOL(netif_rx); 4963 4964 int netif_rx_ni(struct sk_buff *skb) 4965 { 4966 int err; 4967 4968 trace_netif_rx_ni_entry(skb); 4969 4970 preempt_disable(); 4971 err = netif_rx_internal(skb); 4972 if (local_softirq_pending()) 4973 do_softirq(); 4974 preempt_enable(); 4975 trace_netif_rx_ni_exit(err); 4976 4977 return err; 4978 } 4979 EXPORT_SYMBOL(netif_rx_ni); 4980 4981 int netif_rx_any_context(struct sk_buff *skb) 4982 { 4983 /* 4984 * If invoked from contexts which do not invoke bottom half 4985 * processing either at return from interrupt or when softrqs are 4986 * reenabled, use netif_rx_ni() which invokes bottomhalf processing 4987 * directly. 4988 */ 4989 if (in_interrupt()) 4990 return netif_rx(skb); 4991 else 4992 return netif_rx_ni(skb); 4993 } 4994 EXPORT_SYMBOL(netif_rx_any_context); 4995 4996 static __latent_entropy void net_tx_action(struct softirq_action *h) 4997 { 4998 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 4999 5000 if (sd->completion_queue) { 5001 struct sk_buff *clist; 5002 5003 local_irq_disable(); 5004 clist = sd->completion_queue; 5005 sd->completion_queue = NULL; 5006 local_irq_enable(); 5007 5008 while (clist) { 5009 struct sk_buff *skb = clist; 5010 5011 clist = clist->next; 5012 5013 WARN_ON(refcount_read(&skb->users)); 5014 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED)) 5015 trace_consume_skb(skb); 5016 else 5017 trace_kfree_skb(skb, net_tx_action); 5018 5019 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) 5020 __kfree_skb(skb); 5021 else 5022 __kfree_skb_defer(skb); 5023 } 5024 } 5025 5026 if (sd->output_queue) { 5027 struct Qdisc *head; 5028 5029 local_irq_disable(); 5030 head = sd->output_queue; 5031 sd->output_queue = NULL; 5032 sd->output_queue_tailp = &sd->output_queue; 5033 local_irq_enable(); 5034 5035 rcu_read_lock(); 5036 5037 while (head) { 5038 struct Qdisc *q = head; 5039 spinlock_t *root_lock = NULL; 5040 5041 head = head->next_sched; 5042 5043 /* We need to make sure head->next_sched is read 5044 * before clearing __QDISC_STATE_SCHED 5045 */ 5046 smp_mb__before_atomic(); 5047 5048 if (!(q->flags & TCQ_F_NOLOCK)) { 5049 root_lock = qdisc_lock(q); 5050 spin_lock(root_lock); 5051 } else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, 5052 &q->state))) { 5053 /* There is a synchronize_net() between 5054 * STATE_DEACTIVATED flag being set and 5055 * qdisc_reset()/some_qdisc_is_busy() in 5056 * dev_deactivate(), so we can safely bail out 5057 * early here to avoid data race between 5058 * qdisc_deactivate() and some_qdisc_is_busy() 5059 * for lockless qdisc. 5060 */ 5061 clear_bit(__QDISC_STATE_SCHED, &q->state); 5062 continue; 5063 } 5064 5065 clear_bit(__QDISC_STATE_SCHED, &q->state); 5066 qdisc_run(q); 5067 if (root_lock) 5068 spin_unlock(root_lock); 5069 } 5070 5071 rcu_read_unlock(); 5072 } 5073 5074 xfrm_dev_backlog(sd); 5075 } 5076 5077 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) 5078 /* This hook is defined here for ATM LANE */ 5079 int (*br_fdb_test_addr_hook)(struct net_device *dev, 5080 unsigned char *addr) __read_mostly; 5081 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); 5082 #endif 5083 5084 static inline struct sk_buff * 5085 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 5086 struct net_device *orig_dev, bool *another) 5087 { 5088 #ifdef CONFIG_NET_CLS_ACT 5089 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress); 5090 struct tcf_result cl_res; 5091 5092 /* If there's at least one ingress present somewhere (so 5093 * we get here via enabled static key), remaining devices 5094 * that are not configured with an ingress qdisc will bail 5095 * out here. 5096 */ 5097 if (!miniq) 5098 return skb; 5099 5100 if (*pt_prev) { 5101 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5102 *pt_prev = NULL; 5103 } 5104 5105 qdisc_skb_cb(skb)->pkt_len = skb->len; 5106 qdisc_skb_cb(skb)->mru = 0; 5107 qdisc_skb_cb(skb)->post_ct = false; 5108 skb->tc_at_ingress = 1; 5109 mini_qdisc_bstats_cpu_update(miniq, skb); 5110 5111 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) { 5112 case TC_ACT_OK: 5113 case TC_ACT_RECLASSIFY: 5114 skb->tc_index = TC_H_MIN(cl_res.classid); 5115 break; 5116 case TC_ACT_SHOT: 5117 mini_qdisc_qstats_cpu_drop(miniq); 5118 kfree_skb(skb); 5119 return NULL; 5120 case TC_ACT_STOLEN: 5121 case TC_ACT_QUEUED: 5122 case TC_ACT_TRAP: 5123 consume_skb(skb); 5124 return NULL; 5125 case TC_ACT_REDIRECT: 5126 /* skb_mac_header check was done by cls/act_bpf, so 5127 * we can safely push the L2 header back before 5128 * redirecting to another netdev 5129 */ 5130 __skb_push(skb, skb->mac_len); 5131 if (skb_do_redirect(skb) == -EAGAIN) { 5132 __skb_pull(skb, skb->mac_len); 5133 *another = true; 5134 break; 5135 } 5136 return NULL; 5137 case TC_ACT_CONSUMED: 5138 return NULL; 5139 default: 5140 break; 5141 } 5142 #endif /* CONFIG_NET_CLS_ACT */ 5143 return skb; 5144 } 5145 5146 /** 5147 * netdev_is_rx_handler_busy - check if receive handler is registered 5148 * @dev: device to check 5149 * 5150 * Check if a receive handler is already registered for a given device. 5151 * Return true if there one. 5152 * 5153 * The caller must hold the rtnl_mutex. 5154 */ 5155 bool netdev_is_rx_handler_busy(struct net_device *dev) 5156 { 5157 ASSERT_RTNL(); 5158 return dev && rtnl_dereference(dev->rx_handler); 5159 } 5160 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); 5161 5162 /** 5163 * netdev_rx_handler_register - register receive handler 5164 * @dev: device to register a handler for 5165 * @rx_handler: receive handler to register 5166 * @rx_handler_data: data pointer that is used by rx handler 5167 * 5168 * Register a receive handler for a device. This handler will then be 5169 * called from __netif_receive_skb. A negative errno code is returned 5170 * on a failure. 5171 * 5172 * The caller must hold the rtnl_mutex. 5173 * 5174 * For a general description of rx_handler, see enum rx_handler_result. 5175 */ 5176 int netdev_rx_handler_register(struct net_device *dev, 5177 rx_handler_func_t *rx_handler, 5178 void *rx_handler_data) 5179 { 5180 if (netdev_is_rx_handler_busy(dev)) 5181 return -EBUSY; 5182 5183 if (dev->priv_flags & IFF_NO_RX_HANDLER) 5184 return -EINVAL; 5185 5186 /* Note: rx_handler_data must be set before rx_handler */ 5187 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); 5188 rcu_assign_pointer(dev->rx_handler, rx_handler); 5189 5190 return 0; 5191 } 5192 EXPORT_SYMBOL_GPL(netdev_rx_handler_register); 5193 5194 /** 5195 * netdev_rx_handler_unregister - unregister receive handler 5196 * @dev: device to unregister a handler from 5197 * 5198 * Unregister a receive handler from a device. 5199 * 5200 * The caller must hold the rtnl_mutex. 5201 */ 5202 void netdev_rx_handler_unregister(struct net_device *dev) 5203 { 5204 5205 ASSERT_RTNL(); 5206 RCU_INIT_POINTER(dev->rx_handler, NULL); 5207 /* a reader seeing a non NULL rx_handler in a rcu_read_lock() 5208 * section has a guarantee to see a non NULL rx_handler_data 5209 * as well. 5210 */ 5211 synchronize_net(); 5212 RCU_INIT_POINTER(dev->rx_handler_data, NULL); 5213 } 5214 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); 5215 5216 /* 5217 * Limit the use of PFMEMALLOC reserves to those protocols that implement 5218 * the special handling of PFMEMALLOC skbs. 5219 */ 5220 static bool skb_pfmemalloc_protocol(struct sk_buff *skb) 5221 { 5222 switch (skb->protocol) { 5223 case htons(ETH_P_ARP): 5224 case htons(ETH_P_IP): 5225 case htons(ETH_P_IPV6): 5226 case htons(ETH_P_8021Q): 5227 case htons(ETH_P_8021AD): 5228 return true; 5229 default: 5230 return false; 5231 } 5232 } 5233 5234 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, 5235 int *ret, struct net_device *orig_dev) 5236 { 5237 if (nf_hook_ingress_active(skb)) { 5238 int ingress_retval; 5239 5240 if (*pt_prev) { 5241 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5242 *pt_prev = NULL; 5243 } 5244 5245 rcu_read_lock(); 5246 ingress_retval = nf_hook_ingress(skb); 5247 rcu_read_unlock(); 5248 return ingress_retval; 5249 } 5250 return 0; 5251 } 5252 5253 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc, 5254 struct packet_type **ppt_prev) 5255 { 5256 struct packet_type *ptype, *pt_prev; 5257 rx_handler_func_t *rx_handler; 5258 struct sk_buff *skb = *pskb; 5259 struct net_device *orig_dev; 5260 bool deliver_exact = false; 5261 int ret = NET_RX_DROP; 5262 __be16 type; 5263 5264 net_timestamp_check(!netdev_tstamp_prequeue, skb); 5265 5266 trace_netif_receive_skb(skb); 5267 5268 orig_dev = skb->dev; 5269 5270 skb_reset_network_header(skb); 5271 if (!skb_transport_header_was_set(skb)) 5272 skb_reset_transport_header(skb); 5273 skb_reset_mac_len(skb); 5274 5275 pt_prev = NULL; 5276 5277 another_round: 5278 skb->skb_iif = skb->dev->ifindex; 5279 5280 __this_cpu_inc(softnet_data.processed); 5281 5282 if (static_branch_unlikely(&generic_xdp_needed_key)) { 5283 int ret2; 5284 5285 migrate_disable(); 5286 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb); 5287 migrate_enable(); 5288 5289 if (ret2 != XDP_PASS) { 5290 ret = NET_RX_DROP; 5291 goto out; 5292 } 5293 } 5294 5295 if (eth_type_vlan(skb->protocol)) { 5296 skb = skb_vlan_untag(skb); 5297 if (unlikely(!skb)) 5298 goto out; 5299 } 5300 5301 if (skb_skip_tc_classify(skb)) 5302 goto skip_classify; 5303 5304 if (pfmemalloc) 5305 goto skip_taps; 5306 5307 list_for_each_entry_rcu(ptype, &ptype_all, list) { 5308 if (pt_prev) 5309 ret = deliver_skb(skb, pt_prev, orig_dev); 5310 pt_prev = ptype; 5311 } 5312 5313 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { 5314 if (pt_prev) 5315 ret = deliver_skb(skb, pt_prev, orig_dev); 5316 pt_prev = ptype; 5317 } 5318 5319 skip_taps: 5320 #ifdef CONFIG_NET_INGRESS 5321 if (static_branch_unlikely(&ingress_needed_key)) { 5322 bool another = false; 5323 5324 nf_skip_egress(skb, true); 5325 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev, 5326 &another); 5327 if (another) 5328 goto another_round; 5329 if (!skb) 5330 goto out; 5331 5332 nf_skip_egress(skb, false); 5333 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) 5334 goto out; 5335 } 5336 #endif 5337 skb_reset_redirect(skb); 5338 skip_classify: 5339 if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) 5340 goto drop; 5341 5342 if (skb_vlan_tag_present(skb)) { 5343 if (pt_prev) { 5344 ret = deliver_skb(skb, pt_prev, orig_dev); 5345 pt_prev = NULL; 5346 } 5347 if (vlan_do_receive(&skb)) 5348 goto another_round; 5349 else if (unlikely(!skb)) 5350 goto out; 5351 } 5352 5353 rx_handler = rcu_dereference(skb->dev->rx_handler); 5354 if (rx_handler) { 5355 if (pt_prev) { 5356 ret = deliver_skb(skb, pt_prev, orig_dev); 5357 pt_prev = NULL; 5358 } 5359 switch (rx_handler(&skb)) { 5360 case RX_HANDLER_CONSUMED: 5361 ret = NET_RX_SUCCESS; 5362 goto out; 5363 case RX_HANDLER_ANOTHER: 5364 goto another_round; 5365 case RX_HANDLER_EXACT: 5366 deliver_exact = true; 5367 break; 5368 case RX_HANDLER_PASS: 5369 break; 5370 default: 5371 BUG(); 5372 } 5373 } 5374 5375 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) { 5376 check_vlan_id: 5377 if (skb_vlan_tag_get_id(skb)) { 5378 /* Vlan id is non 0 and vlan_do_receive() above couldn't 5379 * find vlan device. 5380 */ 5381 skb->pkt_type = PACKET_OTHERHOST; 5382 } else if (eth_type_vlan(skb->protocol)) { 5383 /* Outer header is 802.1P with vlan 0, inner header is 5384 * 802.1Q or 802.1AD and vlan_do_receive() above could 5385 * not find vlan dev for vlan id 0. 5386 */ 5387 __vlan_hwaccel_clear_tag(skb); 5388 skb = skb_vlan_untag(skb); 5389 if (unlikely(!skb)) 5390 goto out; 5391 if (vlan_do_receive(&skb)) 5392 /* After stripping off 802.1P header with vlan 0 5393 * vlan dev is found for inner header. 5394 */ 5395 goto another_round; 5396 else if (unlikely(!skb)) 5397 goto out; 5398 else 5399 /* We have stripped outer 802.1P vlan 0 header. 5400 * But could not find vlan dev. 5401 * check again for vlan id to set OTHERHOST. 5402 */ 5403 goto check_vlan_id; 5404 } 5405 /* Note: we might in the future use prio bits 5406 * and set skb->priority like in vlan_do_receive() 5407 * For the time being, just ignore Priority Code Point 5408 */ 5409 __vlan_hwaccel_clear_tag(skb); 5410 } 5411 5412 type = skb->protocol; 5413 5414 /* deliver only exact match when indicated */ 5415 if (likely(!deliver_exact)) { 5416 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5417 &ptype_base[ntohs(type) & 5418 PTYPE_HASH_MASK]); 5419 } 5420 5421 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5422 &orig_dev->ptype_specific); 5423 5424 if (unlikely(skb->dev != orig_dev)) { 5425 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5426 &skb->dev->ptype_specific); 5427 } 5428 5429 if (pt_prev) { 5430 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 5431 goto drop; 5432 *ppt_prev = pt_prev; 5433 } else { 5434 drop: 5435 if (!deliver_exact) 5436 atomic_long_inc(&skb->dev->rx_dropped); 5437 else 5438 atomic_long_inc(&skb->dev->rx_nohandler); 5439 kfree_skb(skb); 5440 /* Jamal, now you will not able to escape explaining 5441 * me how you were going to use this. :-) 5442 */ 5443 ret = NET_RX_DROP; 5444 } 5445 5446 out: 5447 /* The invariant here is that if *ppt_prev is not NULL 5448 * then skb should also be non-NULL. 5449 * 5450 * Apparently *ppt_prev assignment above holds this invariant due to 5451 * skb dereferencing near it. 5452 */ 5453 *pskb = skb; 5454 return ret; 5455 } 5456 5457 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc) 5458 { 5459 struct net_device *orig_dev = skb->dev; 5460 struct packet_type *pt_prev = NULL; 5461 int ret; 5462 5463 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5464 if (pt_prev) 5465 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb, 5466 skb->dev, pt_prev, orig_dev); 5467 return ret; 5468 } 5469 5470 /** 5471 * netif_receive_skb_core - special purpose version of netif_receive_skb 5472 * @skb: buffer to process 5473 * 5474 * More direct receive version of netif_receive_skb(). It should 5475 * only be used by callers that have a need to skip RPS and Generic XDP. 5476 * Caller must also take care of handling if ``(page_is_)pfmemalloc``. 5477 * 5478 * This function may only be called from softirq context and interrupts 5479 * should be enabled. 5480 * 5481 * Return values (usually ignored): 5482 * NET_RX_SUCCESS: no congestion 5483 * NET_RX_DROP: packet was dropped 5484 */ 5485 int netif_receive_skb_core(struct sk_buff *skb) 5486 { 5487 int ret; 5488 5489 rcu_read_lock(); 5490 ret = __netif_receive_skb_one_core(skb, false); 5491 rcu_read_unlock(); 5492 5493 return ret; 5494 } 5495 EXPORT_SYMBOL(netif_receive_skb_core); 5496 5497 static inline void __netif_receive_skb_list_ptype(struct list_head *head, 5498 struct packet_type *pt_prev, 5499 struct net_device *orig_dev) 5500 { 5501 struct sk_buff *skb, *next; 5502 5503 if (!pt_prev) 5504 return; 5505 if (list_empty(head)) 5506 return; 5507 if (pt_prev->list_func != NULL) 5508 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv, 5509 ip_list_rcv, head, pt_prev, orig_dev); 5510 else 5511 list_for_each_entry_safe(skb, next, head, list) { 5512 skb_list_del_init(skb); 5513 pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 5514 } 5515 } 5516 5517 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc) 5518 { 5519 /* Fast-path assumptions: 5520 * - There is no RX handler. 5521 * - Only one packet_type matches. 5522 * If either of these fails, we will end up doing some per-packet 5523 * processing in-line, then handling the 'last ptype' for the whole 5524 * sublist. This can't cause out-of-order delivery to any single ptype, 5525 * because the 'last ptype' must be constant across the sublist, and all 5526 * other ptypes are handled per-packet. 5527 */ 5528 /* Current (common) ptype of sublist */ 5529 struct packet_type *pt_curr = NULL; 5530 /* Current (common) orig_dev of sublist */ 5531 struct net_device *od_curr = NULL; 5532 struct list_head sublist; 5533 struct sk_buff *skb, *next; 5534 5535 INIT_LIST_HEAD(&sublist); 5536 list_for_each_entry_safe(skb, next, head, list) { 5537 struct net_device *orig_dev = skb->dev; 5538 struct packet_type *pt_prev = NULL; 5539 5540 skb_list_del_init(skb); 5541 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5542 if (!pt_prev) 5543 continue; 5544 if (pt_curr != pt_prev || od_curr != orig_dev) { 5545 /* dispatch old sublist */ 5546 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5547 /* start new sublist */ 5548 INIT_LIST_HEAD(&sublist); 5549 pt_curr = pt_prev; 5550 od_curr = orig_dev; 5551 } 5552 list_add_tail(&skb->list, &sublist); 5553 } 5554 5555 /* dispatch final sublist */ 5556 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5557 } 5558 5559 static int __netif_receive_skb(struct sk_buff *skb) 5560 { 5561 int ret; 5562 5563 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { 5564 unsigned int noreclaim_flag; 5565 5566 /* 5567 * PFMEMALLOC skbs are special, they should 5568 * - be delivered to SOCK_MEMALLOC sockets only 5569 * - stay away from userspace 5570 * - have bounded memory usage 5571 * 5572 * Use PF_MEMALLOC as this saves us from propagating the allocation 5573 * context down to all allocation sites. 5574 */ 5575 noreclaim_flag = memalloc_noreclaim_save(); 5576 ret = __netif_receive_skb_one_core(skb, true); 5577 memalloc_noreclaim_restore(noreclaim_flag); 5578 } else 5579 ret = __netif_receive_skb_one_core(skb, false); 5580 5581 return ret; 5582 } 5583 5584 static void __netif_receive_skb_list(struct list_head *head) 5585 { 5586 unsigned long noreclaim_flag = 0; 5587 struct sk_buff *skb, *next; 5588 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */ 5589 5590 list_for_each_entry_safe(skb, next, head, list) { 5591 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) { 5592 struct list_head sublist; 5593 5594 /* Handle the previous sublist */ 5595 list_cut_before(&sublist, head, &skb->list); 5596 if (!list_empty(&sublist)) 5597 __netif_receive_skb_list_core(&sublist, pfmemalloc); 5598 pfmemalloc = !pfmemalloc; 5599 /* See comments in __netif_receive_skb */ 5600 if (pfmemalloc) 5601 noreclaim_flag = memalloc_noreclaim_save(); 5602 else 5603 memalloc_noreclaim_restore(noreclaim_flag); 5604 } 5605 } 5606 /* Handle the remaining sublist */ 5607 if (!list_empty(head)) 5608 __netif_receive_skb_list_core(head, pfmemalloc); 5609 /* Restore pflags */ 5610 if (pfmemalloc) 5611 memalloc_noreclaim_restore(noreclaim_flag); 5612 } 5613 5614 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp) 5615 { 5616 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog); 5617 struct bpf_prog *new = xdp->prog; 5618 int ret = 0; 5619 5620 switch (xdp->command) { 5621 case XDP_SETUP_PROG: 5622 rcu_assign_pointer(dev->xdp_prog, new); 5623 if (old) 5624 bpf_prog_put(old); 5625 5626 if (old && !new) { 5627 static_branch_dec(&generic_xdp_needed_key); 5628 } else if (new && !old) { 5629 static_branch_inc(&generic_xdp_needed_key); 5630 dev_disable_lro(dev); 5631 dev_disable_gro_hw(dev); 5632 } 5633 break; 5634 5635 default: 5636 ret = -EINVAL; 5637 break; 5638 } 5639 5640 return ret; 5641 } 5642 5643 static int netif_receive_skb_internal(struct sk_buff *skb) 5644 { 5645 int ret; 5646 5647 net_timestamp_check(netdev_tstamp_prequeue, skb); 5648 5649 if (skb_defer_rx_timestamp(skb)) 5650 return NET_RX_SUCCESS; 5651 5652 rcu_read_lock(); 5653 #ifdef CONFIG_RPS 5654 if (static_branch_unlikely(&rps_needed)) { 5655 struct rps_dev_flow voidflow, *rflow = &voidflow; 5656 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5657 5658 if (cpu >= 0) { 5659 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5660 rcu_read_unlock(); 5661 return ret; 5662 } 5663 } 5664 #endif 5665 ret = __netif_receive_skb(skb); 5666 rcu_read_unlock(); 5667 return ret; 5668 } 5669 5670 static void netif_receive_skb_list_internal(struct list_head *head) 5671 { 5672 struct sk_buff *skb, *next; 5673 struct list_head sublist; 5674 5675 INIT_LIST_HEAD(&sublist); 5676 list_for_each_entry_safe(skb, next, head, list) { 5677 net_timestamp_check(netdev_tstamp_prequeue, skb); 5678 skb_list_del_init(skb); 5679 if (!skb_defer_rx_timestamp(skb)) 5680 list_add_tail(&skb->list, &sublist); 5681 } 5682 list_splice_init(&sublist, head); 5683 5684 rcu_read_lock(); 5685 #ifdef CONFIG_RPS 5686 if (static_branch_unlikely(&rps_needed)) { 5687 list_for_each_entry_safe(skb, next, head, list) { 5688 struct rps_dev_flow voidflow, *rflow = &voidflow; 5689 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5690 5691 if (cpu >= 0) { 5692 /* Will be handled, remove from list */ 5693 skb_list_del_init(skb); 5694 enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5695 } 5696 } 5697 } 5698 #endif 5699 __netif_receive_skb_list(head); 5700 rcu_read_unlock(); 5701 } 5702 5703 /** 5704 * netif_receive_skb - process receive buffer from network 5705 * @skb: buffer to process 5706 * 5707 * netif_receive_skb() is the main receive data processing function. 5708 * It always succeeds. The buffer may be dropped during processing 5709 * for congestion control or by the protocol layers. 5710 * 5711 * This function may only be called from softirq context and interrupts 5712 * should be enabled. 5713 * 5714 * Return values (usually ignored): 5715 * NET_RX_SUCCESS: no congestion 5716 * NET_RX_DROP: packet was dropped 5717 */ 5718 int netif_receive_skb(struct sk_buff *skb) 5719 { 5720 int ret; 5721 5722 trace_netif_receive_skb_entry(skb); 5723 5724 ret = netif_receive_skb_internal(skb); 5725 trace_netif_receive_skb_exit(ret); 5726 5727 return ret; 5728 } 5729 EXPORT_SYMBOL(netif_receive_skb); 5730 5731 /** 5732 * netif_receive_skb_list - process many receive buffers from network 5733 * @head: list of skbs to process. 5734 * 5735 * Since return value of netif_receive_skb() is normally ignored, and 5736 * wouldn't be meaningful for a list, this function returns void. 5737 * 5738 * This function may only be called from softirq context and interrupts 5739 * should be enabled. 5740 */ 5741 void netif_receive_skb_list(struct list_head *head) 5742 { 5743 struct sk_buff *skb; 5744 5745 if (list_empty(head)) 5746 return; 5747 if (trace_netif_receive_skb_list_entry_enabled()) { 5748 list_for_each_entry(skb, head, list) 5749 trace_netif_receive_skb_list_entry(skb); 5750 } 5751 netif_receive_skb_list_internal(head); 5752 trace_netif_receive_skb_list_exit(0); 5753 } 5754 EXPORT_SYMBOL(netif_receive_skb_list); 5755 5756 static DEFINE_PER_CPU(struct work_struct, flush_works); 5757 5758 /* Network device is going away, flush any packets still pending */ 5759 static void flush_backlog(struct work_struct *work) 5760 { 5761 struct sk_buff *skb, *tmp; 5762 struct softnet_data *sd; 5763 5764 local_bh_disable(); 5765 sd = this_cpu_ptr(&softnet_data); 5766 5767 local_irq_disable(); 5768 rps_lock(sd); 5769 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { 5770 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5771 __skb_unlink(skb, &sd->input_pkt_queue); 5772 dev_kfree_skb_irq(skb); 5773 input_queue_head_incr(sd); 5774 } 5775 } 5776 rps_unlock(sd); 5777 local_irq_enable(); 5778 5779 skb_queue_walk_safe(&sd->process_queue, skb, tmp) { 5780 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5781 __skb_unlink(skb, &sd->process_queue); 5782 kfree_skb(skb); 5783 input_queue_head_incr(sd); 5784 } 5785 } 5786 local_bh_enable(); 5787 } 5788 5789 static bool flush_required(int cpu) 5790 { 5791 #if IS_ENABLED(CONFIG_RPS) 5792 struct softnet_data *sd = &per_cpu(softnet_data, cpu); 5793 bool do_flush; 5794 5795 local_irq_disable(); 5796 rps_lock(sd); 5797 5798 /* as insertion into process_queue happens with the rps lock held, 5799 * process_queue access may race only with dequeue 5800 */ 5801 do_flush = !skb_queue_empty(&sd->input_pkt_queue) || 5802 !skb_queue_empty_lockless(&sd->process_queue); 5803 rps_unlock(sd); 5804 local_irq_enable(); 5805 5806 return do_flush; 5807 #endif 5808 /* without RPS we can't safely check input_pkt_queue: during a 5809 * concurrent remote skb_queue_splice() we can detect as empty both 5810 * input_pkt_queue and process_queue even if the latter could end-up 5811 * containing a lot of packets. 5812 */ 5813 return true; 5814 } 5815 5816 static void flush_all_backlogs(void) 5817 { 5818 static cpumask_t flush_cpus; 5819 unsigned int cpu; 5820 5821 /* since we are under rtnl lock protection we can use static data 5822 * for the cpumask and avoid allocating on stack the possibly 5823 * large mask 5824 */ 5825 ASSERT_RTNL(); 5826 5827 cpus_read_lock(); 5828 5829 cpumask_clear(&flush_cpus); 5830 for_each_online_cpu(cpu) { 5831 if (flush_required(cpu)) { 5832 queue_work_on(cpu, system_highpri_wq, 5833 per_cpu_ptr(&flush_works, cpu)); 5834 cpumask_set_cpu(cpu, &flush_cpus); 5835 } 5836 } 5837 5838 /* we can have in flight packet[s] on the cpus we are not flushing, 5839 * synchronize_net() in unregister_netdevice_many() will take care of 5840 * them 5841 */ 5842 for_each_cpu(cpu, &flush_cpus) 5843 flush_work(per_cpu_ptr(&flush_works, cpu)); 5844 5845 cpus_read_unlock(); 5846 } 5847 5848 /* Pass the currently batched GRO_NORMAL SKBs up to the stack. */ 5849 static void gro_normal_list(struct napi_struct *napi) 5850 { 5851 if (!napi->rx_count) 5852 return; 5853 netif_receive_skb_list_internal(&napi->rx_list); 5854 INIT_LIST_HEAD(&napi->rx_list); 5855 napi->rx_count = 0; 5856 } 5857 5858 /* Queue one GRO_NORMAL SKB up for list processing. If batch size exceeded, 5859 * pass the whole batch up to the stack. 5860 */ 5861 static void gro_normal_one(struct napi_struct *napi, struct sk_buff *skb, int segs) 5862 { 5863 list_add_tail(&skb->list, &napi->rx_list); 5864 napi->rx_count += segs; 5865 if (napi->rx_count >= gro_normal_batch) 5866 gro_normal_list(napi); 5867 } 5868 5869 static void napi_gro_complete(struct napi_struct *napi, struct sk_buff *skb) 5870 { 5871 struct packet_offload *ptype; 5872 __be16 type = skb->protocol; 5873 struct list_head *head = &offload_base; 5874 int err = -ENOENT; 5875 5876 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb)); 5877 5878 if (NAPI_GRO_CB(skb)->count == 1) { 5879 skb_shinfo(skb)->gso_size = 0; 5880 goto out; 5881 } 5882 5883 rcu_read_lock(); 5884 list_for_each_entry_rcu(ptype, head, list) { 5885 if (ptype->type != type || !ptype->callbacks.gro_complete) 5886 continue; 5887 5888 err = INDIRECT_CALL_INET(ptype->callbacks.gro_complete, 5889 ipv6_gro_complete, inet_gro_complete, 5890 skb, 0); 5891 break; 5892 } 5893 rcu_read_unlock(); 5894 5895 if (err) { 5896 WARN_ON(&ptype->list == head); 5897 kfree_skb(skb); 5898 return; 5899 } 5900 5901 out: 5902 gro_normal_one(napi, skb, NAPI_GRO_CB(skb)->count); 5903 } 5904 5905 static void __napi_gro_flush_chain(struct napi_struct *napi, u32 index, 5906 bool flush_old) 5907 { 5908 struct list_head *head = &napi->gro_hash[index].list; 5909 struct sk_buff *skb, *p; 5910 5911 list_for_each_entry_safe_reverse(skb, p, head, list) { 5912 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies) 5913 return; 5914 skb_list_del_init(skb); 5915 napi_gro_complete(napi, skb); 5916 napi->gro_hash[index].count--; 5917 } 5918 5919 if (!napi->gro_hash[index].count) 5920 __clear_bit(index, &napi->gro_bitmask); 5921 } 5922 5923 /* napi->gro_hash[].list contains packets ordered by age. 5924 * youngest packets at the head of it. 5925 * Complete skbs in reverse order to reduce latencies. 5926 */ 5927 void napi_gro_flush(struct napi_struct *napi, bool flush_old) 5928 { 5929 unsigned long bitmask = napi->gro_bitmask; 5930 unsigned int i, base = ~0U; 5931 5932 while ((i = ffs(bitmask)) != 0) { 5933 bitmask >>= i; 5934 base += i; 5935 __napi_gro_flush_chain(napi, base, flush_old); 5936 } 5937 } 5938 EXPORT_SYMBOL(napi_gro_flush); 5939 5940 static void gro_list_prepare(const struct list_head *head, 5941 const struct sk_buff *skb) 5942 { 5943 unsigned int maclen = skb->dev->hard_header_len; 5944 u32 hash = skb_get_hash_raw(skb); 5945 struct sk_buff *p; 5946 5947 list_for_each_entry(p, head, list) { 5948 unsigned long diffs; 5949 5950 NAPI_GRO_CB(p)->flush = 0; 5951 5952 if (hash != skb_get_hash_raw(p)) { 5953 NAPI_GRO_CB(p)->same_flow = 0; 5954 continue; 5955 } 5956 5957 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev; 5958 diffs |= skb_vlan_tag_present(p) ^ skb_vlan_tag_present(skb); 5959 if (skb_vlan_tag_present(p)) 5960 diffs |= skb_vlan_tag_get(p) ^ skb_vlan_tag_get(skb); 5961 diffs |= skb_metadata_differs(p, skb); 5962 if (maclen == ETH_HLEN) 5963 diffs |= compare_ether_header(skb_mac_header(p), 5964 skb_mac_header(skb)); 5965 else if (!diffs) 5966 diffs = memcmp(skb_mac_header(p), 5967 skb_mac_header(skb), 5968 maclen); 5969 5970 /* in most common scenarions 'slow_gro' is 0 5971 * otherwise we are already on some slower paths 5972 * either skip all the infrequent tests altogether or 5973 * avoid trying too hard to skip each of them individually 5974 */ 5975 if (!diffs && unlikely(skb->slow_gro | p->slow_gro)) { 5976 #if IS_ENABLED(CONFIG_SKB_EXTENSIONS) && IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 5977 struct tc_skb_ext *skb_ext; 5978 struct tc_skb_ext *p_ext; 5979 #endif 5980 5981 diffs |= p->sk != skb->sk; 5982 diffs |= skb_metadata_dst_cmp(p, skb); 5983 diffs |= skb_get_nfct(p) ^ skb_get_nfct(skb); 5984 5985 #if IS_ENABLED(CONFIG_SKB_EXTENSIONS) && IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 5986 skb_ext = skb_ext_find(skb, TC_SKB_EXT); 5987 p_ext = skb_ext_find(p, TC_SKB_EXT); 5988 5989 diffs |= (!!p_ext) ^ (!!skb_ext); 5990 if (!diffs && unlikely(skb_ext)) 5991 diffs |= p_ext->chain ^ skb_ext->chain; 5992 #endif 5993 } 5994 5995 NAPI_GRO_CB(p)->same_flow = !diffs; 5996 } 5997 } 5998 5999 static inline void skb_gro_reset_offset(struct sk_buff *skb, u32 nhoff) 6000 { 6001 const struct skb_shared_info *pinfo = skb_shinfo(skb); 6002 const skb_frag_t *frag0 = &pinfo->frags[0]; 6003 6004 NAPI_GRO_CB(skb)->data_offset = 0; 6005 NAPI_GRO_CB(skb)->frag0 = NULL; 6006 NAPI_GRO_CB(skb)->frag0_len = 0; 6007 6008 if (!skb_headlen(skb) && pinfo->nr_frags && 6009 !PageHighMem(skb_frag_page(frag0)) && 6010 (!NET_IP_ALIGN || !((skb_frag_off(frag0) + nhoff) & 3))) { 6011 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0); 6012 NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int, 6013 skb_frag_size(frag0), 6014 skb->end - skb->tail); 6015 } 6016 } 6017 6018 static void gro_pull_from_frag0(struct sk_buff *skb, int grow) 6019 { 6020 struct skb_shared_info *pinfo = skb_shinfo(skb); 6021 6022 BUG_ON(skb->end - skb->tail < grow); 6023 6024 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); 6025 6026 skb->data_len -= grow; 6027 skb->tail += grow; 6028 6029 skb_frag_off_add(&pinfo->frags[0], grow); 6030 skb_frag_size_sub(&pinfo->frags[0], grow); 6031 6032 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) { 6033 skb_frag_unref(skb, 0); 6034 memmove(pinfo->frags, pinfo->frags + 1, 6035 --pinfo->nr_frags * sizeof(pinfo->frags[0])); 6036 } 6037 } 6038 6039 static void gro_flush_oldest(struct napi_struct *napi, struct list_head *head) 6040 { 6041 struct sk_buff *oldest; 6042 6043 oldest = list_last_entry(head, struct sk_buff, list); 6044 6045 /* We are called with head length >= MAX_GRO_SKBS, so this is 6046 * impossible. 6047 */ 6048 if (WARN_ON_ONCE(!oldest)) 6049 return; 6050 6051 /* Do not adjust napi->gro_hash[].count, caller is adding a new 6052 * SKB to the chain. 6053 */ 6054 skb_list_del_init(oldest); 6055 napi_gro_complete(napi, oldest); 6056 } 6057 6058 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) 6059 { 6060 u32 bucket = skb_get_hash_raw(skb) & (GRO_HASH_BUCKETS - 1); 6061 struct gro_list *gro_list = &napi->gro_hash[bucket]; 6062 struct list_head *head = &offload_base; 6063 struct packet_offload *ptype; 6064 __be16 type = skb->protocol; 6065 struct sk_buff *pp = NULL; 6066 enum gro_result ret; 6067 int same_flow; 6068 int grow; 6069 6070 if (netif_elide_gro(skb->dev)) 6071 goto normal; 6072 6073 gro_list_prepare(&gro_list->list, skb); 6074 6075 rcu_read_lock(); 6076 list_for_each_entry_rcu(ptype, head, list) { 6077 if (ptype->type != type || !ptype->callbacks.gro_receive) 6078 continue; 6079 6080 skb_set_network_header(skb, skb_gro_offset(skb)); 6081 skb_reset_mac_len(skb); 6082 NAPI_GRO_CB(skb)->same_flow = 0; 6083 NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb); 6084 NAPI_GRO_CB(skb)->free = 0; 6085 NAPI_GRO_CB(skb)->encap_mark = 0; 6086 NAPI_GRO_CB(skb)->recursion_counter = 0; 6087 NAPI_GRO_CB(skb)->is_fou = 0; 6088 NAPI_GRO_CB(skb)->is_atomic = 1; 6089 NAPI_GRO_CB(skb)->gro_remcsum_start = 0; 6090 6091 /* Setup for GRO checksum validation */ 6092 switch (skb->ip_summed) { 6093 case CHECKSUM_COMPLETE: 6094 NAPI_GRO_CB(skb)->csum = skb->csum; 6095 NAPI_GRO_CB(skb)->csum_valid = 1; 6096 NAPI_GRO_CB(skb)->csum_cnt = 0; 6097 break; 6098 case CHECKSUM_UNNECESSARY: 6099 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1; 6100 NAPI_GRO_CB(skb)->csum_valid = 0; 6101 break; 6102 default: 6103 NAPI_GRO_CB(skb)->csum_cnt = 0; 6104 NAPI_GRO_CB(skb)->csum_valid = 0; 6105 } 6106 6107 pp = INDIRECT_CALL_INET(ptype->callbacks.gro_receive, 6108 ipv6_gro_receive, inet_gro_receive, 6109 &gro_list->list, skb); 6110 break; 6111 } 6112 rcu_read_unlock(); 6113 6114 if (&ptype->list == head) 6115 goto normal; 6116 6117 if (PTR_ERR(pp) == -EINPROGRESS) { 6118 ret = GRO_CONSUMED; 6119 goto ok; 6120 } 6121 6122 same_flow = NAPI_GRO_CB(skb)->same_flow; 6123 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; 6124 6125 if (pp) { 6126 skb_list_del_init(pp); 6127 napi_gro_complete(napi, pp); 6128 gro_list->count--; 6129 } 6130 6131 if (same_flow) 6132 goto ok; 6133 6134 if (NAPI_GRO_CB(skb)->flush) 6135 goto normal; 6136 6137 if (unlikely(gro_list->count >= MAX_GRO_SKBS)) 6138 gro_flush_oldest(napi, &gro_list->list); 6139 else 6140 gro_list->count++; 6141 6142 NAPI_GRO_CB(skb)->count = 1; 6143 NAPI_GRO_CB(skb)->age = jiffies; 6144 NAPI_GRO_CB(skb)->last = skb; 6145 skb_shinfo(skb)->gso_size = skb_gro_len(skb); 6146 list_add(&skb->list, &gro_list->list); 6147 ret = GRO_HELD; 6148 6149 pull: 6150 grow = skb_gro_offset(skb) - skb_headlen(skb); 6151 if (grow > 0) 6152 gro_pull_from_frag0(skb, grow); 6153 ok: 6154 if (gro_list->count) { 6155 if (!test_bit(bucket, &napi->gro_bitmask)) 6156 __set_bit(bucket, &napi->gro_bitmask); 6157 } else if (test_bit(bucket, &napi->gro_bitmask)) { 6158 __clear_bit(bucket, &napi->gro_bitmask); 6159 } 6160 6161 return ret; 6162 6163 normal: 6164 ret = GRO_NORMAL; 6165 goto pull; 6166 } 6167 6168 struct packet_offload *gro_find_receive_by_type(__be16 type) 6169 { 6170 struct list_head *offload_head = &offload_base; 6171 struct packet_offload *ptype; 6172 6173 list_for_each_entry_rcu(ptype, offload_head, list) { 6174 if (ptype->type != type || !ptype->callbacks.gro_receive) 6175 continue; 6176 return ptype; 6177 } 6178 return NULL; 6179 } 6180 EXPORT_SYMBOL(gro_find_receive_by_type); 6181 6182 struct packet_offload *gro_find_complete_by_type(__be16 type) 6183 { 6184 struct list_head *offload_head = &offload_base; 6185 struct packet_offload *ptype; 6186 6187 list_for_each_entry_rcu(ptype, offload_head, list) { 6188 if (ptype->type != type || !ptype->callbacks.gro_complete) 6189 continue; 6190 return ptype; 6191 } 6192 return NULL; 6193 } 6194 EXPORT_SYMBOL(gro_find_complete_by_type); 6195 6196 static gro_result_t napi_skb_finish(struct napi_struct *napi, 6197 struct sk_buff *skb, 6198 gro_result_t ret) 6199 { 6200 switch (ret) { 6201 case GRO_NORMAL: 6202 gro_normal_one(napi, skb, 1); 6203 break; 6204 6205 case GRO_MERGED_FREE: 6206 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) 6207 napi_skb_free_stolen_head(skb); 6208 else if (skb->fclone != SKB_FCLONE_UNAVAILABLE) 6209 __kfree_skb(skb); 6210 else 6211 __kfree_skb_defer(skb); 6212 break; 6213 6214 case GRO_HELD: 6215 case GRO_MERGED: 6216 case GRO_CONSUMED: 6217 break; 6218 } 6219 6220 return ret; 6221 } 6222 6223 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) 6224 { 6225 gro_result_t ret; 6226 6227 skb_mark_napi_id(skb, napi); 6228 trace_napi_gro_receive_entry(skb); 6229 6230 skb_gro_reset_offset(skb, 0); 6231 6232 ret = napi_skb_finish(napi, skb, dev_gro_receive(napi, skb)); 6233 trace_napi_gro_receive_exit(ret); 6234 6235 return ret; 6236 } 6237 EXPORT_SYMBOL(napi_gro_receive); 6238 6239 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) 6240 { 6241 if (unlikely(skb->pfmemalloc)) { 6242 consume_skb(skb); 6243 return; 6244 } 6245 __skb_pull(skb, skb_headlen(skb)); 6246 /* restore the reserve we had after netdev_alloc_skb_ip_align() */ 6247 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb)); 6248 __vlan_hwaccel_clear_tag(skb); 6249 skb->dev = napi->dev; 6250 skb->skb_iif = 0; 6251 6252 /* eth_type_trans() assumes pkt_type is PACKET_HOST */ 6253 skb->pkt_type = PACKET_HOST; 6254 6255 skb->encapsulation = 0; 6256 skb_shinfo(skb)->gso_type = 0; 6257 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); 6258 if (unlikely(skb->slow_gro)) { 6259 skb_orphan(skb); 6260 skb_ext_reset(skb); 6261 nf_reset_ct(skb); 6262 skb->slow_gro = 0; 6263 } 6264 6265 napi->skb = skb; 6266 } 6267 6268 struct sk_buff *napi_get_frags(struct napi_struct *napi) 6269 { 6270 struct sk_buff *skb = napi->skb; 6271 6272 if (!skb) { 6273 skb = napi_alloc_skb(napi, GRO_MAX_HEAD); 6274 if (skb) { 6275 napi->skb = skb; 6276 skb_mark_napi_id(skb, napi); 6277 } 6278 } 6279 return skb; 6280 } 6281 EXPORT_SYMBOL(napi_get_frags); 6282 6283 static gro_result_t napi_frags_finish(struct napi_struct *napi, 6284 struct sk_buff *skb, 6285 gro_result_t ret) 6286 { 6287 switch (ret) { 6288 case GRO_NORMAL: 6289 case GRO_HELD: 6290 __skb_push(skb, ETH_HLEN); 6291 skb->protocol = eth_type_trans(skb, skb->dev); 6292 if (ret == GRO_NORMAL) 6293 gro_normal_one(napi, skb, 1); 6294 break; 6295 6296 case GRO_MERGED_FREE: 6297 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) 6298 napi_skb_free_stolen_head(skb); 6299 else 6300 napi_reuse_skb(napi, skb); 6301 break; 6302 6303 case GRO_MERGED: 6304 case GRO_CONSUMED: 6305 break; 6306 } 6307 6308 return ret; 6309 } 6310 6311 /* Upper GRO stack assumes network header starts at gro_offset=0 6312 * Drivers could call both napi_gro_frags() and napi_gro_receive() 6313 * We copy ethernet header into skb->data to have a common layout. 6314 */ 6315 static struct sk_buff *napi_frags_skb(struct napi_struct *napi) 6316 { 6317 struct sk_buff *skb = napi->skb; 6318 const struct ethhdr *eth; 6319 unsigned int hlen = sizeof(*eth); 6320 6321 napi->skb = NULL; 6322 6323 skb_reset_mac_header(skb); 6324 skb_gro_reset_offset(skb, hlen); 6325 6326 if (unlikely(skb_gro_header_hard(skb, hlen))) { 6327 eth = skb_gro_header_slow(skb, hlen, 0); 6328 if (unlikely(!eth)) { 6329 net_warn_ratelimited("%s: dropping impossible skb from %s\n", 6330 __func__, napi->dev->name); 6331 napi_reuse_skb(napi, skb); 6332 return NULL; 6333 } 6334 } else { 6335 eth = (const struct ethhdr *)skb->data; 6336 gro_pull_from_frag0(skb, hlen); 6337 NAPI_GRO_CB(skb)->frag0 += hlen; 6338 NAPI_GRO_CB(skb)->frag0_len -= hlen; 6339 } 6340 __skb_pull(skb, hlen); 6341 6342 /* 6343 * This works because the only protocols we care about don't require 6344 * special handling. 6345 * We'll fix it up properly in napi_frags_finish() 6346 */ 6347 skb->protocol = eth->h_proto; 6348 6349 return skb; 6350 } 6351 6352 gro_result_t napi_gro_frags(struct napi_struct *napi) 6353 { 6354 gro_result_t ret; 6355 struct sk_buff *skb = napi_frags_skb(napi); 6356 6357 trace_napi_gro_frags_entry(skb); 6358 6359 ret = napi_frags_finish(napi, skb, dev_gro_receive(napi, skb)); 6360 trace_napi_gro_frags_exit(ret); 6361 6362 return ret; 6363 } 6364 EXPORT_SYMBOL(napi_gro_frags); 6365 6366 /* Compute the checksum from gro_offset and return the folded value 6367 * after adding in any pseudo checksum. 6368 */ 6369 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb) 6370 { 6371 __wsum wsum; 6372 __sum16 sum; 6373 6374 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0); 6375 6376 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */ 6377 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum)); 6378 /* See comments in __skb_checksum_complete(). */ 6379 if (likely(!sum)) { 6380 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && 6381 !skb->csum_complete_sw) 6382 netdev_rx_csum_fault(skb->dev, skb); 6383 } 6384 6385 NAPI_GRO_CB(skb)->csum = wsum; 6386 NAPI_GRO_CB(skb)->csum_valid = 1; 6387 6388 return sum; 6389 } 6390 EXPORT_SYMBOL(__skb_gro_checksum_complete); 6391 6392 static void net_rps_send_ipi(struct softnet_data *remsd) 6393 { 6394 #ifdef CONFIG_RPS 6395 while (remsd) { 6396 struct softnet_data *next = remsd->rps_ipi_next; 6397 6398 if (cpu_online(remsd->cpu)) 6399 smp_call_function_single_async(remsd->cpu, &remsd->csd); 6400 remsd = next; 6401 } 6402 #endif 6403 } 6404 6405 /* 6406 * net_rps_action_and_irq_enable sends any pending IPI's for rps. 6407 * Note: called with local irq disabled, but exits with local irq enabled. 6408 */ 6409 static void net_rps_action_and_irq_enable(struct softnet_data *sd) 6410 { 6411 #ifdef CONFIG_RPS 6412 struct softnet_data *remsd = sd->rps_ipi_list; 6413 6414 if (remsd) { 6415 sd->rps_ipi_list = NULL; 6416 6417 local_irq_enable(); 6418 6419 /* Send pending IPI's to kick RPS processing on remote cpus. */ 6420 net_rps_send_ipi(remsd); 6421 } else 6422 #endif 6423 local_irq_enable(); 6424 } 6425 6426 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) 6427 { 6428 #ifdef CONFIG_RPS 6429 return sd->rps_ipi_list != NULL; 6430 #else 6431 return false; 6432 #endif 6433 } 6434 6435 static int process_backlog(struct napi_struct *napi, int quota) 6436 { 6437 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); 6438 bool again = true; 6439 int work = 0; 6440 6441 /* Check if we have pending ipi, its better to send them now, 6442 * not waiting net_rx_action() end. 6443 */ 6444 if (sd_has_rps_ipi_waiting(sd)) { 6445 local_irq_disable(); 6446 net_rps_action_and_irq_enable(sd); 6447 } 6448 6449 napi->weight = dev_rx_weight; 6450 while (again) { 6451 struct sk_buff *skb; 6452 6453 while ((skb = __skb_dequeue(&sd->process_queue))) { 6454 rcu_read_lock(); 6455 __netif_receive_skb(skb); 6456 rcu_read_unlock(); 6457 input_queue_head_incr(sd); 6458 if (++work >= quota) 6459 return work; 6460 6461 } 6462 6463 local_irq_disable(); 6464 rps_lock(sd); 6465 if (skb_queue_empty(&sd->input_pkt_queue)) { 6466 /* 6467 * Inline a custom version of __napi_complete(). 6468 * only current cpu owns and manipulates this napi, 6469 * and NAPI_STATE_SCHED is the only possible flag set 6470 * on backlog. 6471 * We can use a plain write instead of clear_bit(), 6472 * and we dont need an smp_mb() memory barrier. 6473 */ 6474 napi->state = 0; 6475 again = false; 6476 } else { 6477 skb_queue_splice_tail_init(&sd->input_pkt_queue, 6478 &sd->process_queue); 6479 } 6480 rps_unlock(sd); 6481 local_irq_enable(); 6482 } 6483 6484 return work; 6485 } 6486 6487 /** 6488 * __napi_schedule - schedule for receive 6489 * @n: entry to schedule 6490 * 6491 * The entry's receive function will be scheduled to run. 6492 * Consider using __napi_schedule_irqoff() if hard irqs are masked. 6493 */ 6494 void __napi_schedule(struct napi_struct *n) 6495 { 6496 unsigned long flags; 6497 6498 local_irq_save(flags); 6499 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6500 local_irq_restore(flags); 6501 } 6502 EXPORT_SYMBOL(__napi_schedule); 6503 6504 /** 6505 * napi_schedule_prep - check if napi can be scheduled 6506 * @n: napi context 6507 * 6508 * Test if NAPI routine is already running, and if not mark 6509 * it as running. This is used as a condition variable to 6510 * insure only one NAPI poll instance runs. We also make 6511 * sure there is no pending NAPI disable. 6512 */ 6513 bool napi_schedule_prep(struct napi_struct *n) 6514 { 6515 unsigned long val, new; 6516 6517 do { 6518 val = READ_ONCE(n->state); 6519 if (unlikely(val & NAPIF_STATE_DISABLE)) 6520 return false; 6521 new = val | NAPIF_STATE_SCHED; 6522 6523 /* Sets STATE_MISSED bit if STATE_SCHED was already set 6524 * This was suggested by Alexander Duyck, as compiler 6525 * emits better code than : 6526 * if (val & NAPIF_STATE_SCHED) 6527 * new |= NAPIF_STATE_MISSED; 6528 */ 6529 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * 6530 NAPIF_STATE_MISSED; 6531 } while (cmpxchg(&n->state, val, new) != val); 6532 6533 return !(val & NAPIF_STATE_SCHED); 6534 } 6535 EXPORT_SYMBOL(napi_schedule_prep); 6536 6537 /** 6538 * __napi_schedule_irqoff - schedule for receive 6539 * @n: entry to schedule 6540 * 6541 * Variant of __napi_schedule() assuming hard irqs are masked. 6542 * 6543 * On PREEMPT_RT enabled kernels this maps to __napi_schedule() 6544 * because the interrupt disabled assumption might not be true 6545 * due to force-threaded interrupts and spinlock substitution. 6546 */ 6547 void __napi_schedule_irqoff(struct napi_struct *n) 6548 { 6549 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6550 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6551 else 6552 __napi_schedule(n); 6553 } 6554 EXPORT_SYMBOL(__napi_schedule_irqoff); 6555 6556 bool napi_complete_done(struct napi_struct *n, int work_done) 6557 { 6558 unsigned long flags, val, new, timeout = 0; 6559 bool ret = true; 6560 6561 /* 6562 * 1) Don't let napi dequeue from the cpu poll list 6563 * just in case its running on a different cpu. 6564 * 2) If we are busy polling, do nothing here, we have 6565 * the guarantee we will be called later. 6566 */ 6567 if (unlikely(n->state & (NAPIF_STATE_NPSVC | 6568 NAPIF_STATE_IN_BUSY_POLL))) 6569 return false; 6570 6571 if (work_done) { 6572 if (n->gro_bitmask) 6573 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6574 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs); 6575 } 6576 if (n->defer_hard_irqs_count > 0) { 6577 n->defer_hard_irqs_count--; 6578 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6579 if (timeout) 6580 ret = false; 6581 } 6582 if (n->gro_bitmask) { 6583 /* When the NAPI instance uses a timeout and keeps postponing 6584 * it, we need to bound somehow the time packets are kept in 6585 * the GRO layer 6586 */ 6587 napi_gro_flush(n, !!timeout); 6588 } 6589 6590 gro_normal_list(n); 6591 6592 if (unlikely(!list_empty(&n->poll_list))) { 6593 /* If n->poll_list is not empty, we need to mask irqs */ 6594 local_irq_save(flags); 6595 list_del_init(&n->poll_list); 6596 local_irq_restore(flags); 6597 } 6598 6599 do { 6600 val = READ_ONCE(n->state); 6601 6602 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); 6603 6604 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED | 6605 NAPIF_STATE_SCHED_THREADED | 6606 NAPIF_STATE_PREFER_BUSY_POLL); 6607 6608 /* If STATE_MISSED was set, leave STATE_SCHED set, 6609 * because we will call napi->poll() one more time. 6610 * This C code was suggested by Alexander Duyck to help gcc. 6611 */ 6612 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * 6613 NAPIF_STATE_SCHED; 6614 } while (cmpxchg(&n->state, val, new) != val); 6615 6616 if (unlikely(val & NAPIF_STATE_MISSED)) { 6617 __napi_schedule(n); 6618 return false; 6619 } 6620 6621 if (timeout) 6622 hrtimer_start(&n->timer, ns_to_ktime(timeout), 6623 HRTIMER_MODE_REL_PINNED); 6624 return ret; 6625 } 6626 EXPORT_SYMBOL(napi_complete_done); 6627 6628 /* must be called under rcu_read_lock(), as we dont take a reference */ 6629 static struct napi_struct *napi_by_id(unsigned int napi_id) 6630 { 6631 unsigned int hash = napi_id % HASH_SIZE(napi_hash); 6632 struct napi_struct *napi; 6633 6634 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) 6635 if (napi->napi_id == napi_id) 6636 return napi; 6637 6638 return NULL; 6639 } 6640 6641 #if defined(CONFIG_NET_RX_BUSY_POLL) 6642 6643 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule) 6644 { 6645 if (!skip_schedule) { 6646 gro_normal_list(napi); 6647 __napi_schedule(napi); 6648 return; 6649 } 6650 6651 if (napi->gro_bitmask) { 6652 /* flush too old packets 6653 * If HZ < 1000, flush all packets. 6654 */ 6655 napi_gro_flush(napi, HZ >= 1000); 6656 } 6657 6658 gro_normal_list(napi); 6659 clear_bit(NAPI_STATE_SCHED, &napi->state); 6660 } 6661 6662 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll, 6663 u16 budget) 6664 { 6665 bool skip_schedule = false; 6666 unsigned long timeout; 6667 int rc; 6668 6669 /* Busy polling means there is a high chance device driver hard irq 6670 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was 6671 * set in napi_schedule_prep(). 6672 * Since we are about to call napi->poll() once more, we can safely 6673 * clear NAPI_STATE_MISSED. 6674 * 6675 * Note: x86 could use a single "lock and ..." instruction 6676 * to perform these two clear_bit() 6677 */ 6678 clear_bit(NAPI_STATE_MISSED, &napi->state); 6679 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); 6680 6681 local_bh_disable(); 6682 6683 if (prefer_busy_poll) { 6684 napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs); 6685 timeout = READ_ONCE(napi->dev->gro_flush_timeout); 6686 if (napi->defer_hard_irqs_count && timeout) { 6687 hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED); 6688 skip_schedule = true; 6689 } 6690 } 6691 6692 /* All we really want here is to re-enable device interrupts. 6693 * Ideally, a new ndo_busy_poll_stop() could avoid another round. 6694 */ 6695 rc = napi->poll(napi, budget); 6696 /* We can't gro_normal_list() here, because napi->poll() might have 6697 * rearmed the napi (napi_complete_done()) in which case it could 6698 * already be running on another CPU. 6699 */ 6700 trace_napi_poll(napi, rc, budget); 6701 netpoll_poll_unlock(have_poll_lock); 6702 if (rc == budget) 6703 __busy_poll_stop(napi, skip_schedule); 6704 local_bh_enable(); 6705 } 6706 6707 void napi_busy_loop(unsigned int napi_id, 6708 bool (*loop_end)(void *, unsigned long), 6709 void *loop_end_arg, bool prefer_busy_poll, u16 budget) 6710 { 6711 unsigned long start_time = loop_end ? busy_loop_current_time() : 0; 6712 int (*napi_poll)(struct napi_struct *napi, int budget); 6713 void *have_poll_lock = NULL; 6714 struct napi_struct *napi; 6715 6716 restart: 6717 napi_poll = NULL; 6718 6719 rcu_read_lock(); 6720 6721 napi = napi_by_id(napi_id); 6722 if (!napi) 6723 goto out; 6724 6725 preempt_disable(); 6726 for (;;) { 6727 int work = 0; 6728 6729 local_bh_disable(); 6730 if (!napi_poll) { 6731 unsigned long val = READ_ONCE(napi->state); 6732 6733 /* If multiple threads are competing for this napi, 6734 * we avoid dirtying napi->state as much as we can. 6735 */ 6736 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | 6737 NAPIF_STATE_IN_BUSY_POLL)) { 6738 if (prefer_busy_poll) 6739 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6740 goto count; 6741 } 6742 if (cmpxchg(&napi->state, val, 6743 val | NAPIF_STATE_IN_BUSY_POLL | 6744 NAPIF_STATE_SCHED) != val) { 6745 if (prefer_busy_poll) 6746 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6747 goto count; 6748 } 6749 have_poll_lock = netpoll_poll_lock(napi); 6750 napi_poll = napi->poll; 6751 } 6752 work = napi_poll(napi, budget); 6753 trace_napi_poll(napi, work, budget); 6754 gro_normal_list(napi); 6755 count: 6756 if (work > 0) 6757 __NET_ADD_STATS(dev_net(napi->dev), 6758 LINUX_MIB_BUSYPOLLRXPACKETS, work); 6759 local_bh_enable(); 6760 6761 if (!loop_end || loop_end(loop_end_arg, start_time)) 6762 break; 6763 6764 if (unlikely(need_resched())) { 6765 if (napi_poll) 6766 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6767 preempt_enable(); 6768 rcu_read_unlock(); 6769 cond_resched(); 6770 if (loop_end(loop_end_arg, start_time)) 6771 return; 6772 goto restart; 6773 } 6774 cpu_relax(); 6775 } 6776 if (napi_poll) 6777 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6778 preempt_enable(); 6779 out: 6780 rcu_read_unlock(); 6781 } 6782 EXPORT_SYMBOL(napi_busy_loop); 6783 6784 #endif /* CONFIG_NET_RX_BUSY_POLL */ 6785 6786 static void napi_hash_add(struct napi_struct *napi) 6787 { 6788 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state)) 6789 return; 6790 6791 spin_lock(&napi_hash_lock); 6792 6793 /* 0..NR_CPUS range is reserved for sender_cpu use */ 6794 do { 6795 if (unlikely(++napi_gen_id < MIN_NAPI_ID)) 6796 napi_gen_id = MIN_NAPI_ID; 6797 } while (napi_by_id(napi_gen_id)); 6798 napi->napi_id = napi_gen_id; 6799 6800 hlist_add_head_rcu(&napi->napi_hash_node, 6801 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); 6802 6803 spin_unlock(&napi_hash_lock); 6804 } 6805 6806 /* Warning : caller is responsible to make sure rcu grace period 6807 * is respected before freeing memory containing @napi 6808 */ 6809 static void napi_hash_del(struct napi_struct *napi) 6810 { 6811 spin_lock(&napi_hash_lock); 6812 6813 hlist_del_init_rcu(&napi->napi_hash_node); 6814 6815 spin_unlock(&napi_hash_lock); 6816 } 6817 6818 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) 6819 { 6820 struct napi_struct *napi; 6821 6822 napi = container_of(timer, struct napi_struct, timer); 6823 6824 /* Note : we use a relaxed variant of napi_schedule_prep() not setting 6825 * NAPI_STATE_MISSED, since we do not react to a device IRQ. 6826 */ 6827 if (!napi_disable_pending(napi) && 6828 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) { 6829 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6830 __napi_schedule_irqoff(napi); 6831 } 6832 6833 return HRTIMER_NORESTART; 6834 } 6835 6836 static void init_gro_hash(struct napi_struct *napi) 6837 { 6838 int i; 6839 6840 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6841 INIT_LIST_HEAD(&napi->gro_hash[i].list); 6842 napi->gro_hash[i].count = 0; 6843 } 6844 napi->gro_bitmask = 0; 6845 } 6846 6847 int dev_set_threaded(struct net_device *dev, bool threaded) 6848 { 6849 struct napi_struct *napi; 6850 int err = 0; 6851 6852 if (dev->threaded == threaded) 6853 return 0; 6854 6855 if (threaded) { 6856 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6857 if (!napi->thread) { 6858 err = napi_kthread_create(napi); 6859 if (err) { 6860 threaded = false; 6861 break; 6862 } 6863 } 6864 } 6865 } 6866 6867 dev->threaded = threaded; 6868 6869 /* Make sure kthread is created before THREADED bit 6870 * is set. 6871 */ 6872 smp_mb__before_atomic(); 6873 6874 /* Setting/unsetting threaded mode on a napi might not immediately 6875 * take effect, if the current napi instance is actively being 6876 * polled. In this case, the switch between threaded mode and 6877 * softirq mode will happen in the next round of napi_schedule(). 6878 * This should not cause hiccups/stalls to the live traffic. 6879 */ 6880 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6881 if (threaded) 6882 set_bit(NAPI_STATE_THREADED, &napi->state); 6883 else 6884 clear_bit(NAPI_STATE_THREADED, &napi->state); 6885 } 6886 6887 return err; 6888 } 6889 EXPORT_SYMBOL(dev_set_threaded); 6890 6891 void netif_napi_add(struct net_device *dev, struct napi_struct *napi, 6892 int (*poll)(struct napi_struct *, int), int weight) 6893 { 6894 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state))) 6895 return; 6896 6897 INIT_LIST_HEAD(&napi->poll_list); 6898 INIT_HLIST_NODE(&napi->napi_hash_node); 6899 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); 6900 napi->timer.function = napi_watchdog; 6901 init_gro_hash(napi); 6902 napi->skb = NULL; 6903 INIT_LIST_HEAD(&napi->rx_list); 6904 napi->rx_count = 0; 6905 napi->poll = poll; 6906 if (weight > NAPI_POLL_WEIGHT) 6907 netdev_err_once(dev, "%s() called with weight %d\n", __func__, 6908 weight); 6909 napi->weight = weight; 6910 napi->dev = dev; 6911 #ifdef CONFIG_NETPOLL 6912 napi->poll_owner = -1; 6913 #endif 6914 set_bit(NAPI_STATE_SCHED, &napi->state); 6915 set_bit(NAPI_STATE_NPSVC, &napi->state); 6916 list_add_rcu(&napi->dev_list, &dev->napi_list); 6917 napi_hash_add(napi); 6918 /* Create kthread for this napi if dev->threaded is set. 6919 * Clear dev->threaded if kthread creation failed so that 6920 * threaded mode will not be enabled in napi_enable(). 6921 */ 6922 if (dev->threaded && napi_kthread_create(napi)) 6923 dev->threaded = 0; 6924 } 6925 EXPORT_SYMBOL(netif_napi_add); 6926 6927 void napi_disable(struct napi_struct *n) 6928 { 6929 unsigned long val, new; 6930 6931 might_sleep(); 6932 set_bit(NAPI_STATE_DISABLE, &n->state); 6933 6934 for ( ; ; ) { 6935 val = READ_ONCE(n->state); 6936 if (val & (NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC)) { 6937 usleep_range(20, 200); 6938 continue; 6939 } 6940 6941 new = val | NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC; 6942 new &= ~(NAPIF_STATE_THREADED | NAPIF_STATE_PREFER_BUSY_POLL); 6943 6944 if (cmpxchg(&n->state, val, new) == val) 6945 break; 6946 } 6947 6948 hrtimer_cancel(&n->timer); 6949 6950 clear_bit(NAPI_STATE_DISABLE, &n->state); 6951 } 6952 EXPORT_SYMBOL(napi_disable); 6953 6954 /** 6955 * napi_enable - enable NAPI scheduling 6956 * @n: NAPI context 6957 * 6958 * Resume NAPI from being scheduled on this context. 6959 * Must be paired with napi_disable. 6960 */ 6961 void napi_enable(struct napi_struct *n) 6962 { 6963 unsigned long val, new; 6964 6965 do { 6966 val = READ_ONCE(n->state); 6967 BUG_ON(!test_bit(NAPI_STATE_SCHED, &val)); 6968 6969 new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC); 6970 if (n->dev->threaded && n->thread) 6971 new |= NAPIF_STATE_THREADED; 6972 } while (cmpxchg(&n->state, val, new) != val); 6973 } 6974 EXPORT_SYMBOL(napi_enable); 6975 6976 static void flush_gro_hash(struct napi_struct *napi) 6977 { 6978 int i; 6979 6980 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6981 struct sk_buff *skb, *n; 6982 6983 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list) 6984 kfree_skb(skb); 6985 napi->gro_hash[i].count = 0; 6986 } 6987 } 6988 6989 /* Must be called in process context */ 6990 void __netif_napi_del(struct napi_struct *napi) 6991 { 6992 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state)) 6993 return; 6994 6995 napi_hash_del(napi); 6996 list_del_rcu(&napi->dev_list); 6997 napi_free_frags(napi); 6998 6999 flush_gro_hash(napi); 7000 napi->gro_bitmask = 0; 7001 7002 if (napi->thread) { 7003 kthread_stop(napi->thread); 7004 napi->thread = NULL; 7005 } 7006 } 7007 EXPORT_SYMBOL(__netif_napi_del); 7008 7009 static int __napi_poll(struct napi_struct *n, bool *repoll) 7010 { 7011 int work, weight; 7012 7013 weight = n->weight; 7014 7015 /* This NAPI_STATE_SCHED test is for avoiding a race 7016 * with netpoll's poll_napi(). Only the entity which 7017 * obtains the lock and sees NAPI_STATE_SCHED set will 7018 * actually make the ->poll() call. Therefore we avoid 7019 * accidentally calling ->poll() when NAPI is not scheduled. 7020 */ 7021 work = 0; 7022 if (test_bit(NAPI_STATE_SCHED, &n->state)) { 7023 work = n->poll(n, weight); 7024 trace_napi_poll(n, work, weight); 7025 } 7026 7027 if (unlikely(work > weight)) 7028 netdev_err_once(n->dev, "NAPI poll function %pS returned %d, exceeding its budget of %d.\n", 7029 n->poll, work, weight); 7030 7031 if (likely(work < weight)) 7032 return work; 7033 7034 /* Drivers must not modify the NAPI state if they 7035 * consume the entire weight. In such cases this code 7036 * still "owns" the NAPI instance and therefore can 7037 * move the instance around on the list at-will. 7038 */ 7039 if (unlikely(napi_disable_pending(n))) { 7040 napi_complete(n); 7041 return work; 7042 } 7043 7044 /* The NAPI context has more processing work, but busy-polling 7045 * is preferred. Exit early. 7046 */ 7047 if (napi_prefer_busy_poll(n)) { 7048 if (napi_complete_done(n, work)) { 7049 /* If timeout is not set, we need to make sure 7050 * that the NAPI is re-scheduled. 7051 */ 7052 napi_schedule(n); 7053 } 7054 return work; 7055 } 7056 7057 if (n->gro_bitmask) { 7058 /* flush too old packets 7059 * If HZ < 1000, flush all packets. 7060 */ 7061 napi_gro_flush(n, HZ >= 1000); 7062 } 7063 7064 gro_normal_list(n); 7065 7066 /* Some drivers may have called napi_schedule 7067 * prior to exhausting their budget. 7068 */ 7069 if (unlikely(!list_empty(&n->poll_list))) { 7070 pr_warn_once("%s: Budget exhausted after napi rescheduled\n", 7071 n->dev ? n->dev->name : "backlog"); 7072 return work; 7073 } 7074 7075 *repoll = true; 7076 7077 return work; 7078 } 7079 7080 static int napi_poll(struct napi_struct *n, struct list_head *repoll) 7081 { 7082 bool do_repoll = false; 7083 void *have; 7084 int work; 7085 7086 list_del_init(&n->poll_list); 7087 7088 have = netpoll_poll_lock(n); 7089 7090 work = __napi_poll(n, &do_repoll); 7091 7092 if (do_repoll) 7093 list_add_tail(&n->poll_list, repoll); 7094 7095 netpoll_poll_unlock(have); 7096 7097 return work; 7098 } 7099 7100 static int napi_thread_wait(struct napi_struct *napi) 7101 { 7102 bool woken = false; 7103 7104 set_current_state(TASK_INTERRUPTIBLE); 7105 7106 while (!kthread_should_stop()) { 7107 /* Testing SCHED_THREADED bit here to make sure the current 7108 * kthread owns this napi and could poll on this napi. 7109 * Testing SCHED bit is not enough because SCHED bit might be 7110 * set by some other busy poll thread or by napi_disable(). 7111 */ 7112 if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state) || woken) { 7113 WARN_ON(!list_empty(&napi->poll_list)); 7114 __set_current_state(TASK_RUNNING); 7115 return 0; 7116 } 7117 7118 schedule(); 7119 /* woken being true indicates this thread owns this napi. */ 7120 woken = true; 7121 set_current_state(TASK_INTERRUPTIBLE); 7122 } 7123 __set_current_state(TASK_RUNNING); 7124 7125 return -1; 7126 } 7127 7128 static int napi_threaded_poll(void *data) 7129 { 7130 struct napi_struct *napi = data; 7131 void *have; 7132 7133 while (!napi_thread_wait(napi)) { 7134 for (;;) { 7135 bool repoll = false; 7136 7137 local_bh_disable(); 7138 7139 have = netpoll_poll_lock(napi); 7140 __napi_poll(napi, &repoll); 7141 netpoll_poll_unlock(have); 7142 7143 local_bh_enable(); 7144 7145 if (!repoll) 7146 break; 7147 7148 cond_resched(); 7149 } 7150 } 7151 return 0; 7152 } 7153 7154 static __latent_entropy void net_rx_action(struct softirq_action *h) 7155 { 7156 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 7157 unsigned long time_limit = jiffies + 7158 usecs_to_jiffies(netdev_budget_usecs); 7159 int budget = netdev_budget; 7160 LIST_HEAD(list); 7161 LIST_HEAD(repoll); 7162 7163 local_irq_disable(); 7164 list_splice_init(&sd->poll_list, &list); 7165 local_irq_enable(); 7166 7167 for (;;) { 7168 struct napi_struct *n; 7169 7170 if (list_empty(&list)) { 7171 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll)) 7172 return; 7173 break; 7174 } 7175 7176 n = list_first_entry(&list, struct napi_struct, poll_list); 7177 budget -= napi_poll(n, &repoll); 7178 7179 /* If softirq window is exhausted then punt. 7180 * Allow this to run for 2 jiffies since which will allow 7181 * an average latency of 1.5/HZ. 7182 */ 7183 if (unlikely(budget <= 0 || 7184 time_after_eq(jiffies, time_limit))) { 7185 sd->time_squeeze++; 7186 break; 7187 } 7188 } 7189 7190 local_irq_disable(); 7191 7192 list_splice_tail_init(&sd->poll_list, &list); 7193 list_splice_tail(&repoll, &list); 7194 list_splice(&list, &sd->poll_list); 7195 if (!list_empty(&sd->poll_list)) 7196 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 7197 7198 net_rps_action_and_irq_enable(sd); 7199 } 7200 7201 struct netdev_adjacent { 7202 struct net_device *dev; 7203 7204 /* upper master flag, there can only be one master device per list */ 7205 bool master; 7206 7207 /* lookup ignore flag */ 7208 bool ignore; 7209 7210 /* counter for the number of times this device was added to us */ 7211 u16 ref_nr; 7212 7213 /* private field for the users */ 7214 void *private; 7215 7216 struct list_head list; 7217 struct rcu_head rcu; 7218 }; 7219 7220 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, 7221 struct list_head *adj_list) 7222 { 7223 struct netdev_adjacent *adj; 7224 7225 list_for_each_entry(adj, adj_list, list) { 7226 if (adj->dev == adj_dev) 7227 return adj; 7228 } 7229 return NULL; 7230 } 7231 7232 static int ____netdev_has_upper_dev(struct net_device *upper_dev, 7233 struct netdev_nested_priv *priv) 7234 { 7235 struct net_device *dev = (struct net_device *)priv->data; 7236 7237 return upper_dev == dev; 7238 } 7239 7240 /** 7241 * netdev_has_upper_dev - Check if device is linked to an upper device 7242 * @dev: device 7243 * @upper_dev: upper device to check 7244 * 7245 * Find out if a device is linked to specified upper device and return true 7246 * in case it is. Note that this checks only immediate upper device, 7247 * not through a complete stack of devices. The caller must hold the RTNL lock. 7248 */ 7249 bool netdev_has_upper_dev(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 ASSERT_RTNL(); 7257 7258 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 7259 &priv); 7260 } 7261 EXPORT_SYMBOL(netdev_has_upper_dev); 7262 7263 /** 7264 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device 7265 * @dev: device 7266 * @upper_dev: upper device to check 7267 * 7268 * Find out if a device is linked to specified upper device and return true 7269 * in case it is. Note that this checks the entire upper device chain. 7270 * The caller must hold rcu lock. 7271 */ 7272 7273 bool netdev_has_upper_dev_all_rcu(struct net_device *dev, 7274 struct net_device *upper_dev) 7275 { 7276 struct netdev_nested_priv priv = { 7277 .data = (void *)upper_dev, 7278 }; 7279 7280 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 7281 &priv); 7282 } 7283 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); 7284 7285 /** 7286 * netdev_has_any_upper_dev - Check if device is linked to some device 7287 * @dev: device 7288 * 7289 * Find out if a device is linked to an upper device and return true in case 7290 * it is. The caller must hold the RTNL lock. 7291 */ 7292 bool netdev_has_any_upper_dev(struct net_device *dev) 7293 { 7294 ASSERT_RTNL(); 7295 7296 return !list_empty(&dev->adj_list.upper); 7297 } 7298 EXPORT_SYMBOL(netdev_has_any_upper_dev); 7299 7300 /** 7301 * netdev_master_upper_dev_get - Get master upper device 7302 * @dev: device 7303 * 7304 * Find a master upper device and return pointer to it or NULL in case 7305 * it's not there. The caller must hold the RTNL lock. 7306 */ 7307 struct net_device *netdev_master_upper_dev_get(struct net_device *dev) 7308 { 7309 struct netdev_adjacent *upper; 7310 7311 ASSERT_RTNL(); 7312 7313 if (list_empty(&dev->adj_list.upper)) 7314 return NULL; 7315 7316 upper = list_first_entry(&dev->adj_list.upper, 7317 struct netdev_adjacent, list); 7318 if (likely(upper->master)) 7319 return upper->dev; 7320 return NULL; 7321 } 7322 EXPORT_SYMBOL(netdev_master_upper_dev_get); 7323 7324 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev) 7325 { 7326 struct netdev_adjacent *upper; 7327 7328 ASSERT_RTNL(); 7329 7330 if (list_empty(&dev->adj_list.upper)) 7331 return NULL; 7332 7333 upper = list_first_entry(&dev->adj_list.upper, 7334 struct netdev_adjacent, list); 7335 if (likely(upper->master) && !upper->ignore) 7336 return upper->dev; 7337 return NULL; 7338 } 7339 7340 /** 7341 * netdev_has_any_lower_dev - Check if device is linked to some device 7342 * @dev: device 7343 * 7344 * Find out if a device is linked to a lower device and return true in case 7345 * it is. The caller must hold the RTNL lock. 7346 */ 7347 static bool netdev_has_any_lower_dev(struct net_device *dev) 7348 { 7349 ASSERT_RTNL(); 7350 7351 return !list_empty(&dev->adj_list.lower); 7352 } 7353 7354 void *netdev_adjacent_get_private(struct list_head *adj_list) 7355 { 7356 struct netdev_adjacent *adj; 7357 7358 adj = list_entry(adj_list, struct netdev_adjacent, list); 7359 7360 return adj->private; 7361 } 7362 EXPORT_SYMBOL(netdev_adjacent_get_private); 7363 7364 /** 7365 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list 7366 * @dev: device 7367 * @iter: list_head ** of the current position 7368 * 7369 * Gets the next device from the dev's upper list, starting from iter 7370 * position. The caller must hold RCU read lock. 7371 */ 7372 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, 7373 struct list_head **iter) 7374 { 7375 struct netdev_adjacent *upper; 7376 7377 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 7378 7379 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7380 7381 if (&upper->list == &dev->adj_list.upper) 7382 return NULL; 7383 7384 *iter = &upper->list; 7385 7386 return upper->dev; 7387 } 7388 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); 7389 7390 static struct net_device *__netdev_next_upper_dev(struct net_device *dev, 7391 struct list_head **iter, 7392 bool *ignore) 7393 { 7394 struct netdev_adjacent *upper; 7395 7396 upper = list_entry((*iter)->next, struct netdev_adjacent, list); 7397 7398 if (&upper->list == &dev->adj_list.upper) 7399 return NULL; 7400 7401 *iter = &upper->list; 7402 *ignore = upper->ignore; 7403 7404 return upper->dev; 7405 } 7406 7407 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, 7408 struct list_head **iter) 7409 { 7410 struct netdev_adjacent *upper; 7411 7412 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 7413 7414 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7415 7416 if (&upper->list == &dev->adj_list.upper) 7417 return NULL; 7418 7419 *iter = &upper->list; 7420 7421 return upper->dev; 7422 } 7423 7424 static int __netdev_walk_all_upper_dev(struct net_device *dev, 7425 int (*fn)(struct net_device *dev, 7426 struct netdev_nested_priv *priv), 7427 struct netdev_nested_priv *priv) 7428 { 7429 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7430 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7431 int ret, cur = 0; 7432 bool ignore; 7433 7434 now = dev; 7435 iter = &dev->adj_list.upper; 7436 7437 while (1) { 7438 if (now != dev) { 7439 ret = fn(now, priv); 7440 if (ret) 7441 return ret; 7442 } 7443 7444 next = NULL; 7445 while (1) { 7446 udev = __netdev_next_upper_dev(now, &iter, &ignore); 7447 if (!udev) 7448 break; 7449 if (ignore) 7450 continue; 7451 7452 next = udev; 7453 niter = &udev->adj_list.upper; 7454 dev_stack[cur] = now; 7455 iter_stack[cur++] = iter; 7456 break; 7457 } 7458 7459 if (!next) { 7460 if (!cur) 7461 return 0; 7462 next = dev_stack[--cur]; 7463 niter = iter_stack[cur]; 7464 } 7465 7466 now = next; 7467 iter = niter; 7468 } 7469 7470 return 0; 7471 } 7472 7473 int netdev_walk_all_upper_dev_rcu(struct net_device *dev, 7474 int (*fn)(struct net_device *dev, 7475 struct netdev_nested_priv *priv), 7476 struct netdev_nested_priv *priv) 7477 { 7478 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7479 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7480 int ret, cur = 0; 7481 7482 now = dev; 7483 iter = &dev->adj_list.upper; 7484 7485 while (1) { 7486 if (now != dev) { 7487 ret = fn(now, priv); 7488 if (ret) 7489 return ret; 7490 } 7491 7492 next = NULL; 7493 while (1) { 7494 udev = netdev_next_upper_dev_rcu(now, &iter); 7495 if (!udev) 7496 break; 7497 7498 next = udev; 7499 niter = &udev->adj_list.upper; 7500 dev_stack[cur] = now; 7501 iter_stack[cur++] = iter; 7502 break; 7503 } 7504 7505 if (!next) { 7506 if (!cur) 7507 return 0; 7508 next = dev_stack[--cur]; 7509 niter = iter_stack[cur]; 7510 } 7511 7512 now = next; 7513 iter = niter; 7514 } 7515 7516 return 0; 7517 } 7518 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); 7519 7520 static bool __netdev_has_upper_dev(struct net_device *dev, 7521 struct net_device *upper_dev) 7522 { 7523 struct netdev_nested_priv priv = { 7524 .flags = 0, 7525 .data = (void *)upper_dev, 7526 }; 7527 7528 ASSERT_RTNL(); 7529 7530 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev, 7531 &priv); 7532 } 7533 7534 /** 7535 * netdev_lower_get_next_private - Get the next ->private from the 7536 * lower neighbour list 7537 * @dev: device 7538 * @iter: list_head ** of the current position 7539 * 7540 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7541 * list, starting from iter position. The caller must hold either hold the 7542 * RTNL lock or its own locking that guarantees that the neighbour lower 7543 * list will remain unchanged. 7544 */ 7545 void *netdev_lower_get_next_private(struct net_device *dev, 7546 struct list_head **iter) 7547 { 7548 struct netdev_adjacent *lower; 7549 7550 lower = list_entry(*iter, struct netdev_adjacent, list); 7551 7552 if (&lower->list == &dev->adj_list.lower) 7553 return NULL; 7554 7555 *iter = lower->list.next; 7556 7557 return lower->private; 7558 } 7559 EXPORT_SYMBOL(netdev_lower_get_next_private); 7560 7561 /** 7562 * netdev_lower_get_next_private_rcu - Get the next ->private from the 7563 * lower neighbour list, RCU 7564 * variant 7565 * @dev: device 7566 * @iter: list_head ** of the current position 7567 * 7568 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7569 * list, starting from iter position. The caller must hold RCU read lock. 7570 */ 7571 void *netdev_lower_get_next_private_rcu(struct net_device *dev, 7572 struct list_head **iter) 7573 { 7574 struct netdev_adjacent *lower; 7575 7576 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 7577 7578 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7579 7580 if (&lower->list == &dev->adj_list.lower) 7581 return NULL; 7582 7583 *iter = &lower->list; 7584 7585 return lower->private; 7586 } 7587 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); 7588 7589 /** 7590 * netdev_lower_get_next - Get the next device from the lower neighbour 7591 * list 7592 * @dev: device 7593 * @iter: list_head ** of the current position 7594 * 7595 * Gets the next netdev_adjacent from the dev's lower neighbour 7596 * list, starting from iter position. The caller must hold RTNL lock or 7597 * its own locking that guarantees that the neighbour lower 7598 * list will remain unchanged. 7599 */ 7600 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) 7601 { 7602 struct netdev_adjacent *lower; 7603 7604 lower = list_entry(*iter, struct netdev_adjacent, list); 7605 7606 if (&lower->list == &dev->adj_list.lower) 7607 return NULL; 7608 7609 *iter = lower->list.next; 7610 7611 return lower->dev; 7612 } 7613 EXPORT_SYMBOL(netdev_lower_get_next); 7614 7615 static struct net_device *netdev_next_lower_dev(struct net_device *dev, 7616 struct list_head **iter) 7617 { 7618 struct netdev_adjacent *lower; 7619 7620 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7621 7622 if (&lower->list == &dev->adj_list.lower) 7623 return NULL; 7624 7625 *iter = &lower->list; 7626 7627 return lower->dev; 7628 } 7629 7630 static struct net_device *__netdev_next_lower_dev(struct net_device *dev, 7631 struct list_head **iter, 7632 bool *ignore) 7633 { 7634 struct netdev_adjacent *lower; 7635 7636 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7637 7638 if (&lower->list == &dev->adj_list.lower) 7639 return NULL; 7640 7641 *iter = &lower->list; 7642 *ignore = lower->ignore; 7643 7644 return lower->dev; 7645 } 7646 7647 int netdev_walk_all_lower_dev(struct net_device *dev, 7648 int (*fn)(struct net_device *dev, 7649 struct netdev_nested_priv *priv), 7650 struct netdev_nested_priv *priv) 7651 { 7652 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7653 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7654 int ret, cur = 0; 7655 7656 now = dev; 7657 iter = &dev->adj_list.lower; 7658 7659 while (1) { 7660 if (now != dev) { 7661 ret = fn(now, priv); 7662 if (ret) 7663 return ret; 7664 } 7665 7666 next = NULL; 7667 while (1) { 7668 ldev = netdev_next_lower_dev(now, &iter); 7669 if (!ldev) 7670 break; 7671 7672 next = ldev; 7673 niter = &ldev->adj_list.lower; 7674 dev_stack[cur] = now; 7675 iter_stack[cur++] = iter; 7676 break; 7677 } 7678 7679 if (!next) { 7680 if (!cur) 7681 return 0; 7682 next = dev_stack[--cur]; 7683 niter = iter_stack[cur]; 7684 } 7685 7686 now = next; 7687 iter = niter; 7688 } 7689 7690 return 0; 7691 } 7692 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); 7693 7694 static int __netdev_walk_all_lower_dev(struct net_device *dev, 7695 int (*fn)(struct net_device *dev, 7696 struct netdev_nested_priv *priv), 7697 struct netdev_nested_priv *priv) 7698 { 7699 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7700 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7701 int ret, cur = 0; 7702 bool ignore; 7703 7704 now = dev; 7705 iter = &dev->adj_list.lower; 7706 7707 while (1) { 7708 if (now != dev) { 7709 ret = fn(now, priv); 7710 if (ret) 7711 return ret; 7712 } 7713 7714 next = NULL; 7715 while (1) { 7716 ldev = __netdev_next_lower_dev(now, &iter, &ignore); 7717 if (!ldev) 7718 break; 7719 if (ignore) 7720 continue; 7721 7722 next = ldev; 7723 niter = &ldev->adj_list.lower; 7724 dev_stack[cur] = now; 7725 iter_stack[cur++] = iter; 7726 break; 7727 } 7728 7729 if (!next) { 7730 if (!cur) 7731 return 0; 7732 next = dev_stack[--cur]; 7733 niter = iter_stack[cur]; 7734 } 7735 7736 now = next; 7737 iter = niter; 7738 } 7739 7740 return 0; 7741 } 7742 7743 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, 7744 struct list_head **iter) 7745 { 7746 struct netdev_adjacent *lower; 7747 7748 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7749 if (&lower->list == &dev->adj_list.lower) 7750 return NULL; 7751 7752 *iter = &lower->list; 7753 7754 return lower->dev; 7755 } 7756 EXPORT_SYMBOL(netdev_next_lower_dev_rcu); 7757 7758 static u8 __netdev_upper_depth(struct net_device *dev) 7759 { 7760 struct net_device *udev; 7761 struct list_head *iter; 7762 u8 max_depth = 0; 7763 bool ignore; 7764 7765 for (iter = &dev->adj_list.upper, 7766 udev = __netdev_next_upper_dev(dev, &iter, &ignore); 7767 udev; 7768 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) { 7769 if (ignore) 7770 continue; 7771 if (max_depth < udev->upper_level) 7772 max_depth = udev->upper_level; 7773 } 7774 7775 return max_depth; 7776 } 7777 7778 static u8 __netdev_lower_depth(struct net_device *dev) 7779 { 7780 struct net_device *ldev; 7781 struct list_head *iter; 7782 u8 max_depth = 0; 7783 bool ignore; 7784 7785 for (iter = &dev->adj_list.lower, 7786 ldev = __netdev_next_lower_dev(dev, &iter, &ignore); 7787 ldev; 7788 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) { 7789 if (ignore) 7790 continue; 7791 if (max_depth < ldev->lower_level) 7792 max_depth = ldev->lower_level; 7793 } 7794 7795 return max_depth; 7796 } 7797 7798 static int __netdev_update_upper_level(struct net_device *dev, 7799 struct netdev_nested_priv *__unused) 7800 { 7801 dev->upper_level = __netdev_upper_depth(dev) + 1; 7802 return 0; 7803 } 7804 7805 static int __netdev_update_lower_level(struct net_device *dev, 7806 struct netdev_nested_priv *priv) 7807 { 7808 dev->lower_level = __netdev_lower_depth(dev) + 1; 7809 7810 #ifdef CONFIG_LOCKDEP 7811 if (!priv) 7812 return 0; 7813 7814 if (priv->flags & NESTED_SYNC_IMM) 7815 dev->nested_level = dev->lower_level - 1; 7816 if (priv->flags & NESTED_SYNC_TODO) 7817 net_unlink_todo(dev); 7818 #endif 7819 return 0; 7820 } 7821 7822 int netdev_walk_all_lower_dev_rcu(struct net_device *dev, 7823 int (*fn)(struct net_device *dev, 7824 struct netdev_nested_priv *priv), 7825 struct netdev_nested_priv *priv) 7826 { 7827 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7828 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7829 int ret, cur = 0; 7830 7831 now = dev; 7832 iter = &dev->adj_list.lower; 7833 7834 while (1) { 7835 if (now != dev) { 7836 ret = fn(now, priv); 7837 if (ret) 7838 return ret; 7839 } 7840 7841 next = NULL; 7842 while (1) { 7843 ldev = netdev_next_lower_dev_rcu(now, &iter); 7844 if (!ldev) 7845 break; 7846 7847 next = ldev; 7848 niter = &ldev->adj_list.lower; 7849 dev_stack[cur] = now; 7850 iter_stack[cur++] = iter; 7851 break; 7852 } 7853 7854 if (!next) { 7855 if (!cur) 7856 return 0; 7857 next = dev_stack[--cur]; 7858 niter = iter_stack[cur]; 7859 } 7860 7861 now = next; 7862 iter = niter; 7863 } 7864 7865 return 0; 7866 } 7867 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); 7868 7869 /** 7870 * netdev_lower_get_first_private_rcu - Get the first ->private from the 7871 * lower neighbour list, RCU 7872 * variant 7873 * @dev: device 7874 * 7875 * Gets the first netdev_adjacent->private from the dev's lower neighbour 7876 * list. The caller must hold RCU read lock. 7877 */ 7878 void *netdev_lower_get_first_private_rcu(struct net_device *dev) 7879 { 7880 struct netdev_adjacent *lower; 7881 7882 lower = list_first_or_null_rcu(&dev->adj_list.lower, 7883 struct netdev_adjacent, list); 7884 if (lower) 7885 return lower->private; 7886 return NULL; 7887 } 7888 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); 7889 7890 /** 7891 * netdev_master_upper_dev_get_rcu - Get master upper device 7892 * @dev: device 7893 * 7894 * Find a master upper device and return pointer to it or NULL in case 7895 * it's not there. The caller must hold the RCU read lock. 7896 */ 7897 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) 7898 { 7899 struct netdev_adjacent *upper; 7900 7901 upper = list_first_or_null_rcu(&dev->adj_list.upper, 7902 struct netdev_adjacent, list); 7903 if (upper && likely(upper->master)) 7904 return upper->dev; 7905 return NULL; 7906 } 7907 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); 7908 7909 static int netdev_adjacent_sysfs_add(struct net_device *dev, 7910 struct net_device *adj_dev, 7911 struct list_head *dev_list) 7912 { 7913 char linkname[IFNAMSIZ+7]; 7914 7915 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7916 "upper_%s" : "lower_%s", adj_dev->name); 7917 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), 7918 linkname); 7919 } 7920 static void netdev_adjacent_sysfs_del(struct net_device *dev, 7921 char *name, 7922 struct list_head *dev_list) 7923 { 7924 char linkname[IFNAMSIZ+7]; 7925 7926 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7927 "upper_%s" : "lower_%s", name); 7928 sysfs_remove_link(&(dev->dev.kobj), linkname); 7929 } 7930 7931 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, 7932 struct net_device *adj_dev, 7933 struct list_head *dev_list) 7934 { 7935 return (dev_list == &dev->adj_list.upper || 7936 dev_list == &dev->adj_list.lower) && 7937 net_eq(dev_net(dev), dev_net(adj_dev)); 7938 } 7939 7940 static int __netdev_adjacent_dev_insert(struct net_device *dev, 7941 struct net_device *adj_dev, 7942 struct list_head *dev_list, 7943 void *private, bool master) 7944 { 7945 struct netdev_adjacent *adj; 7946 int ret; 7947 7948 adj = __netdev_find_adj(adj_dev, dev_list); 7949 7950 if (adj) { 7951 adj->ref_nr += 1; 7952 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", 7953 dev->name, adj_dev->name, adj->ref_nr); 7954 7955 return 0; 7956 } 7957 7958 adj = kmalloc(sizeof(*adj), GFP_KERNEL); 7959 if (!adj) 7960 return -ENOMEM; 7961 7962 adj->dev = adj_dev; 7963 adj->master = master; 7964 adj->ref_nr = 1; 7965 adj->private = private; 7966 adj->ignore = false; 7967 dev_hold(adj_dev); 7968 7969 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", 7970 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); 7971 7972 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { 7973 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); 7974 if (ret) 7975 goto free_adj; 7976 } 7977 7978 /* Ensure that master link is always the first item in list. */ 7979 if (master) { 7980 ret = sysfs_create_link(&(dev->dev.kobj), 7981 &(adj_dev->dev.kobj), "master"); 7982 if (ret) 7983 goto remove_symlinks; 7984 7985 list_add_rcu(&adj->list, dev_list); 7986 } else { 7987 list_add_tail_rcu(&adj->list, dev_list); 7988 } 7989 7990 return 0; 7991 7992 remove_symlinks: 7993 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7994 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7995 free_adj: 7996 kfree(adj); 7997 dev_put(adj_dev); 7998 7999 return ret; 8000 } 8001 8002 static void __netdev_adjacent_dev_remove(struct net_device *dev, 8003 struct net_device *adj_dev, 8004 u16 ref_nr, 8005 struct list_head *dev_list) 8006 { 8007 struct netdev_adjacent *adj; 8008 8009 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", 8010 dev->name, adj_dev->name, ref_nr); 8011 8012 adj = __netdev_find_adj(adj_dev, dev_list); 8013 8014 if (!adj) { 8015 pr_err("Adjacency does not exist for device %s from %s\n", 8016 dev->name, adj_dev->name); 8017 WARN_ON(1); 8018 return; 8019 } 8020 8021 if (adj->ref_nr > ref_nr) { 8022 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", 8023 dev->name, adj_dev->name, ref_nr, 8024 adj->ref_nr - ref_nr); 8025 adj->ref_nr -= ref_nr; 8026 return; 8027 } 8028 8029 if (adj->master) 8030 sysfs_remove_link(&(dev->dev.kobj), "master"); 8031 8032 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 8033 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 8034 8035 list_del_rcu(&adj->list); 8036 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", 8037 adj_dev->name, dev->name, adj_dev->name); 8038 dev_put(adj_dev); 8039 kfree_rcu(adj, rcu); 8040 } 8041 8042 static int __netdev_adjacent_dev_link_lists(struct net_device *dev, 8043 struct net_device *upper_dev, 8044 struct list_head *up_list, 8045 struct list_head *down_list, 8046 void *private, bool master) 8047 { 8048 int ret; 8049 8050 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, 8051 private, master); 8052 if (ret) 8053 return ret; 8054 8055 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, 8056 private, false); 8057 if (ret) { 8058 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); 8059 return ret; 8060 } 8061 8062 return 0; 8063 } 8064 8065 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, 8066 struct net_device *upper_dev, 8067 u16 ref_nr, 8068 struct list_head *up_list, 8069 struct list_head *down_list) 8070 { 8071 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); 8072 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); 8073 } 8074 8075 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, 8076 struct net_device *upper_dev, 8077 void *private, bool master) 8078 { 8079 return __netdev_adjacent_dev_link_lists(dev, upper_dev, 8080 &dev->adj_list.upper, 8081 &upper_dev->adj_list.lower, 8082 private, master); 8083 } 8084 8085 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, 8086 struct net_device *upper_dev) 8087 { 8088 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, 8089 &dev->adj_list.upper, 8090 &upper_dev->adj_list.lower); 8091 } 8092 8093 static int __netdev_upper_dev_link(struct net_device *dev, 8094 struct net_device *upper_dev, bool master, 8095 void *upper_priv, void *upper_info, 8096 struct netdev_nested_priv *priv, 8097 struct netlink_ext_ack *extack) 8098 { 8099 struct netdev_notifier_changeupper_info changeupper_info = { 8100 .info = { 8101 .dev = dev, 8102 .extack = extack, 8103 }, 8104 .upper_dev = upper_dev, 8105 .master = master, 8106 .linking = true, 8107 .upper_info = upper_info, 8108 }; 8109 struct net_device *master_dev; 8110 int ret = 0; 8111 8112 ASSERT_RTNL(); 8113 8114 if (dev == upper_dev) 8115 return -EBUSY; 8116 8117 /* To prevent loops, check if dev is not upper device to upper_dev. */ 8118 if (__netdev_has_upper_dev(upper_dev, dev)) 8119 return -EBUSY; 8120 8121 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV) 8122 return -EMLINK; 8123 8124 if (!master) { 8125 if (__netdev_has_upper_dev(dev, upper_dev)) 8126 return -EEXIST; 8127 } else { 8128 master_dev = __netdev_master_upper_dev_get(dev); 8129 if (master_dev) 8130 return master_dev == upper_dev ? -EEXIST : -EBUSY; 8131 } 8132 8133 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 8134 &changeupper_info.info); 8135 ret = notifier_to_errno(ret); 8136 if (ret) 8137 return ret; 8138 8139 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, 8140 master); 8141 if (ret) 8142 return ret; 8143 8144 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 8145 &changeupper_info.info); 8146 ret = notifier_to_errno(ret); 8147 if (ret) 8148 goto rollback; 8149 8150 __netdev_update_upper_level(dev, NULL); 8151 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 8152 8153 __netdev_update_lower_level(upper_dev, priv); 8154 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 8155 priv); 8156 8157 return 0; 8158 8159 rollback: 8160 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 8161 8162 return ret; 8163 } 8164 8165 /** 8166 * netdev_upper_dev_link - Add a link to the upper device 8167 * @dev: device 8168 * @upper_dev: new upper device 8169 * @extack: netlink extended ack 8170 * 8171 * Adds a link to device which is upper to this one. The caller must hold 8172 * the RTNL lock. On a failure a negative errno code is returned. 8173 * On success the reference counts are adjusted and the function 8174 * returns zero. 8175 */ 8176 int netdev_upper_dev_link(struct net_device *dev, 8177 struct net_device *upper_dev, 8178 struct netlink_ext_ack *extack) 8179 { 8180 struct netdev_nested_priv priv = { 8181 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 8182 .data = NULL, 8183 }; 8184 8185 return __netdev_upper_dev_link(dev, upper_dev, false, 8186 NULL, NULL, &priv, extack); 8187 } 8188 EXPORT_SYMBOL(netdev_upper_dev_link); 8189 8190 /** 8191 * netdev_master_upper_dev_link - Add a master link to the upper device 8192 * @dev: device 8193 * @upper_dev: new upper device 8194 * @upper_priv: upper device private 8195 * @upper_info: upper info to be passed down via notifier 8196 * @extack: netlink extended ack 8197 * 8198 * Adds a link to device which is upper to this one. In this case, only 8199 * one master upper device can be linked, although other non-master devices 8200 * might be linked as well. The caller must hold the RTNL lock. 8201 * On a failure a negative errno code is returned. On success the reference 8202 * counts are adjusted and the function returns zero. 8203 */ 8204 int netdev_master_upper_dev_link(struct net_device *dev, 8205 struct net_device *upper_dev, 8206 void *upper_priv, void *upper_info, 8207 struct netlink_ext_ack *extack) 8208 { 8209 struct netdev_nested_priv priv = { 8210 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 8211 .data = NULL, 8212 }; 8213 8214 return __netdev_upper_dev_link(dev, upper_dev, true, 8215 upper_priv, upper_info, &priv, extack); 8216 } 8217 EXPORT_SYMBOL(netdev_master_upper_dev_link); 8218 8219 static void __netdev_upper_dev_unlink(struct net_device *dev, 8220 struct net_device *upper_dev, 8221 struct netdev_nested_priv *priv) 8222 { 8223 struct netdev_notifier_changeupper_info changeupper_info = { 8224 .info = { 8225 .dev = dev, 8226 }, 8227 .upper_dev = upper_dev, 8228 .linking = false, 8229 }; 8230 8231 ASSERT_RTNL(); 8232 8233 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; 8234 8235 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 8236 &changeupper_info.info); 8237 8238 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 8239 8240 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 8241 &changeupper_info.info); 8242 8243 __netdev_update_upper_level(dev, NULL); 8244 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 8245 8246 __netdev_update_lower_level(upper_dev, priv); 8247 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 8248 priv); 8249 } 8250 8251 /** 8252 * netdev_upper_dev_unlink - Removes a link to upper device 8253 * @dev: device 8254 * @upper_dev: new upper device 8255 * 8256 * Removes a link to device which is upper to this one. The caller must hold 8257 * the RTNL lock. 8258 */ 8259 void netdev_upper_dev_unlink(struct net_device *dev, 8260 struct net_device *upper_dev) 8261 { 8262 struct netdev_nested_priv priv = { 8263 .flags = NESTED_SYNC_TODO, 8264 .data = NULL, 8265 }; 8266 8267 __netdev_upper_dev_unlink(dev, upper_dev, &priv); 8268 } 8269 EXPORT_SYMBOL(netdev_upper_dev_unlink); 8270 8271 static void __netdev_adjacent_dev_set(struct net_device *upper_dev, 8272 struct net_device *lower_dev, 8273 bool val) 8274 { 8275 struct netdev_adjacent *adj; 8276 8277 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower); 8278 if (adj) 8279 adj->ignore = val; 8280 8281 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper); 8282 if (adj) 8283 adj->ignore = val; 8284 } 8285 8286 static void netdev_adjacent_dev_disable(struct net_device *upper_dev, 8287 struct net_device *lower_dev) 8288 { 8289 __netdev_adjacent_dev_set(upper_dev, lower_dev, true); 8290 } 8291 8292 static void netdev_adjacent_dev_enable(struct net_device *upper_dev, 8293 struct net_device *lower_dev) 8294 { 8295 __netdev_adjacent_dev_set(upper_dev, lower_dev, false); 8296 } 8297 8298 int netdev_adjacent_change_prepare(struct net_device *old_dev, 8299 struct net_device *new_dev, 8300 struct net_device *dev, 8301 struct netlink_ext_ack *extack) 8302 { 8303 struct netdev_nested_priv priv = { 8304 .flags = 0, 8305 .data = NULL, 8306 }; 8307 int err; 8308 8309 if (!new_dev) 8310 return 0; 8311 8312 if (old_dev && new_dev != old_dev) 8313 netdev_adjacent_dev_disable(dev, old_dev); 8314 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv, 8315 extack); 8316 if (err) { 8317 if (old_dev && new_dev != old_dev) 8318 netdev_adjacent_dev_enable(dev, old_dev); 8319 return err; 8320 } 8321 8322 return 0; 8323 } 8324 EXPORT_SYMBOL(netdev_adjacent_change_prepare); 8325 8326 void netdev_adjacent_change_commit(struct net_device *old_dev, 8327 struct net_device *new_dev, 8328 struct net_device *dev) 8329 { 8330 struct netdev_nested_priv priv = { 8331 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 8332 .data = NULL, 8333 }; 8334 8335 if (!new_dev || !old_dev) 8336 return; 8337 8338 if (new_dev == old_dev) 8339 return; 8340 8341 netdev_adjacent_dev_enable(dev, old_dev); 8342 __netdev_upper_dev_unlink(old_dev, dev, &priv); 8343 } 8344 EXPORT_SYMBOL(netdev_adjacent_change_commit); 8345 8346 void netdev_adjacent_change_abort(struct net_device *old_dev, 8347 struct net_device *new_dev, 8348 struct net_device *dev) 8349 { 8350 struct netdev_nested_priv priv = { 8351 .flags = 0, 8352 .data = NULL, 8353 }; 8354 8355 if (!new_dev) 8356 return; 8357 8358 if (old_dev && new_dev != old_dev) 8359 netdev_adjacent_dev_enable(dev, old_dev); 8360 8361 __netdev_upper_dev_unlink(new_dev, dev, &priv); 8362 } 8363 EXPORT_SYMBOL(netdev_adjacent_change_abort); 8364 8365 /** 8366 * netdev_bonding_info_change - Dispatch event about slave change 8367 * @dev: device 8368 * @bonding_info: info to dispatch 8369 * 8370 * Send NETDEV_BONDING_INFO to netdev notifiers with info. 8371 * The caller must hold the RTNL lock. 8372 */ 8373 void netdev_bonding_info_change(struct net_device *dev, 8374 struct netdev_bonding_info *bonding_info) 8375 { 8376 struct netdev_notifier_bonding_info info = { 8377 .info.dev = dev, 8378 }; 8379 8380 memcpy(&info.bonding_info, bonding_info, 8381 sizeof(struct netdev_bonding_info)); 8382 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, 8383 &info.info); 8384 } 8385 EXPORT_SYMBOL(netdev_bonding_info_change); 8386 8387 /** 8388 * netdev_get_xmit_slave - Get the xmit slave of master device 8389 * @dev: device 8390 * @skb: The packet 8391 * @all_slaves: assume all the slaves are active 8392 * 8393 * The reference counters are not incremented so the caller must be 8394 * careful with locks. The caller must hold RCU lock. 8395 * %NULL is returned if no slave is found. 8396 */ 8397 8398 struct net_device *netdev_get_xmit_slave(struct net_device *dev, 8399 struct sk_buff *skb, 8400 bool all_slaves) 8401 { 8402 const struct net_device_ops *ops = dev->netdev_ops; 8403 8404 if (!ops->ndo_get_xmit_slave) 8405 return NULL; 8406 return ops->ndo_get_xmit_slave(dev, skb, all_slaves); 8407 } 8408 EXPORT_SYMBOL(netdev_get_xmit_slave); 8409 8410 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev, 8411 struct sock *sk) 8412 { 8413 const struct net_device_ops *ops = dev->netdev_ops; 8414 8415 if (!ops->ndo_sk_get_lower_dev) 8416 return NULL; 8417 return ops->ndo_sk_get_lower_dev(dev, sk); 8418 } 8419 8420 /** 8421 * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket 8422 * @dev: device 8423 * @sk: the socket 8424 * 8425 * %NULL is returned if no lower device is found. 8426 */ 8427 8428 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev, 8429 struct sock *sk) 8430 { 8431 struct net_device *lower; 8432 8433 lower = netdev_sk_get_lower_dev(dev, sk); 8434 while (lower) { 8435 dev = lower; 8436 lower = netdev_sk_get_lower_dev(dev, sk); 8437 } 8438 8439 return dev; 8440 } 8441 EXPORT_SYMBOL(netdev_sk_get_lowest_dev); 8442 8443 static void netdev_adjacent_add_links(struct net_device *dev) 8444 { 8445 struct netdev_adjacent *iter; 8446 8447 struct net *net = dev_net(dev); 8448 8449 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8450 if (!net_eq(net, dev_net(iter->dev))) 8451 continue; 8452 netdev_adjacent_sysfs_add(iter->dev, dev, 8453 &iter->dev->adj_list.lower); 8454 netdev_adjacent_sysfs_add(dev, iter->dev, 8455 &dev->adj_list.upper); 8456 } 8457 8458 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8459 if (!net_eq(net, dev_net(iter->dev))) 8460 continue; 8461 netdev_adjacent_sysfs_add(iter->dev, dev, 8462 &iter->dev->adj_list.upper); 8463 netdev_adjacent_sysfs_add(dev, iter->dev, 8464 &dev->adj_list.lower); 8465 } 8466 } 8467 8468 static void netdev_adjacent_del_links(struct net_device *dev) 8469 { 8470 struct netdev_adjacent *iter; 8471 8472 struct net *net = dev_net(dev); 8473 8474 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8475 if (!net_eq(net, dev_net(iter->dev))) 8476 continue; 8477 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8478 &iter->dev->adj_list.lower); 8479 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8480 &dev->adj_list.upper); 8481 } 8482 8483 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8484 if (!net_eq(net, dev_net(iter->dev))) 8485 continue; 8486 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8487 &iter->dev->adj_list.upper); 8488 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8489 &dev->adj_list.lower); 8490 } 8491 } 8492 8493 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) 8494 { 8495 struct netdev_adjacent *iter; 8496 8497 struct net *net = dev_net(dev); 8498 8499 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8500 if (!net_eq(net, dev_net(iter->dev))) 8501 continue; 8502 netdev_adjacent_sysfs_del(iter->dev, oldname, 8503 &iter->dev->adj_list.lower); 8504 netdev_adjacent_sysfs_add(iter->dev, dev, 8505 &iter->dev->adj_list.lower); 8506 } 8507 8508 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8509 if (!net_eq(net, dev_net(iter->dev))) 8510 continue; 8511 netdev_adjacent_sysfs_del(iter->dev, oldname, 8512 &iter->dev->adj_list.upper); 8513 netdev_adjacent_sysfs_add(iter->dev, dev, 8514 &iter->dev->adj_list.upper); 8515 } 8516 } 8517 8518 void *netdev_lower_dev_get_private(struct net_device *dev, 8519 struct net_device *lower_dev) 8520 { 8521 struct netdev_adjacent *lower; 8522 8523 if (!lower_dev) 8524 return NULL; 8525 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); 8526 if (!lower) 8527 return NULL; 8528 8529 return lower->private; 8530 } 8531 EXPORT_SYMBOL(netdev_lower_dev_get_private); 8532 8533 8534 /** 8535 * netdev_lower_state_changed - Dispatch event about lower device state change 8536 * @lower_dev: device 8537 * @lower_state_info: state to dispatch 8538 * 8539 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. 8540 * The caller must hold the RTNL lock. 8541 */ 8542 void netdev_lower_state_changed(struct net_device *lower_dev, 8543 void *lower_state_info) 8544 { 8545 struct netdev_notifier_changelowerstate_info changelowerstate_info = { 8546 .info.dev = lower_dev, 8547 }; 8548 8549 ASSERT_RTNL(); 8550 changelowerstate_info.lower_state_info = lower_state_info; 8551 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, 8552 &changelowerstate_info.info); 8553 } 8554 EXPORT_SYMBOL(netdev_lower_state_changed); 8555 8556 static void dev_change_rx_flags(struct net_device *dev, int flags) 8557 { 8558 const struct net_device_ops *ops = dev->netdev_ops; 8559 8560 if (ops->ndo_change_rx_flags) 8561 ops->ndo_change_rx_flags(dev, flags); 8562 } 8563 8564 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) 8565 { 8566 unsigned int old_flags = dev->flags; 8567 kuid_t uid; 8568 kgid_t gid; 8569 8570 ASSERT_RTNL(); 8571 8572 dev->flags |= IFF_PROMISC; 8573 dev->promiscuity += inc; 8574 if (dev->promiscuity == 0) { 8575 /* 8576 * Avoid overflow. 8577 * If inc causes overflow, untouch promisc and return error. 8578 */ 8579 if (inc < 0) 8580 dev->flags &= ~IFF_PROMISC; 8581 else { 8582 dev->promiscuity -= inc; 8583 netdev_warn(dev, "promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n"); 8584 return -EOVERFLOW; 8585 } 8586 } 8587 if (dev->flags != old_flags) { 8588 pr_info("device %s %s promiscuous mode\n", 8589 dev->name, 8590 dev->flags & IFF_PROMISC ? "entered" : "left"); 8591 if (audit_enabled) { 8592 current_uid_gid(&uid, &gid); 8593 audit_log(audit_context(), GFP_ATOMIC, 8594 AUDIT_ANOM_PROMISCUOUS, 8595 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", 8596 dev->name, (dev->flags & IFF_PROMISC), 8597 (old_flags & IFF_PROMISC), 8598 from_kuid(&init_user_ns, audit_get_loginuid(current)), 8599 from_kuid(&init_user_ns, uid), 8600 from_kgid(&init_user_ns, gid), 8601 audit_get_sessionid(current)); 8602 } 8603 8604 dev_change_rx_flags(dev, IFF_PROMISC); 8605 } 8606 if (notify) 8607 __dev_notify_flags(dev, old_flags, IFF_PROMISC); 8608 return 0; 8609 } 8610 8611 /** 8612 * dev_set_promiscuity - update promiscuity count on a device 8613 * @dev: device 8614 * @inc: modifier 8615 * 8616 * Add or remove promiscuity from a device. While the count in the device 8617 * remains above zero the interface remains promiscuous. Once it hits zero 8618 * the device reverts back to normal filtering operation. A negative inc 8619 * value is used to drop promiscuity on the device. 8620 * Return 0 if successful or a negative errno code on error. 8621 */ 8622 int dev_set_promiscuity(struct net_device *dev, int inc) 8623 { 8624 unsigned int old_flags = dev->flags; 8625 int err; 8626 8627 err = __dev_set_promiscuity(dev, inc, true); 8628 if (err < 0) 8629 return err; 8630 if (dev->flags != old_flags) 8631 dev_set_rx_mode(dev); 8632 return err; 8633 } 8634 EXPORT_SYMBOL(dev_set_promiscuity); 8635 8636 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) 8637 { 8638 unsigned int old_flags = dev->flags, old_gflags = dev->gflags; 8639 8640 ASSERT_RTNL(); 8641 8642 dev->flags |= IFF_ALLMULTI; 8643 dev->allmulti += inc; 8644 if (dev->allmulti == 0) { 8645 /* 8646 * Avoid overflow. 8647 * If inc causes overflow, untouch allmulti and return error. 8648 */ 8649 if (inc < 0) 8650 dev->flags &= ~IFF_ALLMULTI; 8651 else { 8652 dev->allmulti -= inc; 8653 netdev_warn(dev, "allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n"); 8654 return -EOVERFLOW; 8655 } 8656 } 8657 if (dev->flags ^ old_flags) { 8658 dev_change_rx_flags(dev, IFF_ALLMULTI); 8659 dev_set_rx_mode(dev); 8660 if (notify) 8661 __dev_notify_flags(dev, old_flags, 8662 dev->gflags ^ old_gflags); 8663 } 8664 return 0; 8665 } 8666 8667 /** 8668 * dev_set_allmulti - update allmulti count on a device 8669 * @dev: device 8670 * @inc: modifier 8671 * 8672 * Add or remove reception of all multicast frames to a device. While the 8673 * count in the device remains above zero the interface remains listening 8674 * to all interfaces. Once it hits zero the device reverts back to normal 8675 * filtering operation. A negative @inc value is used to drop the counter 8676 * when releasing a resource needing all multicasts. 8677 * Return 0 if successful or a negative errno code on error. 8678 */ 8679 8680 int dev_set_allmulti(struct net_device *dev, int inc) 8681 { 8682 return __dev_set_allmulti(dev, inc, true); 8683 } 8684 EXPORT_SYMBOL(dev_set_allmulti); 8685 8686 /* 8687 * Upload unicast and multicast address lists to device and 8688 * configure RX filtering. When the device doesn't support unicast 8689 * filtering it is put in promiscuous mode while unicast addresses 8690 * are present. 8691 */ 8692 void __dev_set_rx_mode(struct net_device *dev) 8693 { 8694 const struct net_device_ops *ops = dev->netdev_ops; 8695 8696 /* dev_open will call this function so the list will stay sane. */ 8697 if (!(dev->flags&IFF_UP)) 8698 return; 8699 8700 if (!netif_device_present(dev)) 8701 return; 8702 8703 if (!(dev->priv_flags & IFF_UNICAST_FLT)) { 8704 /* Unicast addresses changes may only happen under the rtnl, 8705 * therefore calling __dev_set_promiscuity here is safe. 8706 */ 8707 if (!netdev_uc_empty(dev) && !dev->uc_promisc) { 8708 __dev_set_promiscuity(dev, 1, false); 8709 dev->uc_promisc = true; 8710 } else if (netdev_uc_empty(dev) && dev->uc_promisc) { 8711 __dev_set_promiscuity(dev, -1, false); 8712 dev->uc_promisc = false; 8713 } 8714 } 8715 8716 if (ops->ndo_set_rx_mode) 8717 ops->ndo_set_rx_mode(dev); 8718 } 8719 8720 void dev_set_rx_mode(struct net_device *dev) 8721 { 8722 netif_addr_lock_bh(dev); 8723 __dev_set_rx_mode(dev); 8724 netif_addr_unlock_bh(dev); 8725 } 8726 8727 /** 8728 * dev_get_flags - get flags reported to userspace 8729 * @dev: device 8730 * 8731 * Get the combination of flag bits exported through APIs to userspace. 8732 */ 8733 unsigned int dev_get_flags(const struct net_device *dev) 8734 { 8735 unsigned int flags; 8736 8737 flags = (dev->flags & ~(IFF_PROMISC | 8738 IFF_ALLMULTI | 8739 IFF_RUNNING | 8740 IFF_LOWER_UP | 8741 IFF_DORMANT)) | 8742 (dev->gflags & (IFF_PROMISC | 8743 IFF_ALLMULTI)); 8744 8745 if (netif_running(dev)) { 8746 if (netif_oper_up(dev)) 8747 flags |= IFF_RUNNING; 8748 if (netif_carrier_ok(dev)) 8749 flags |= IFF_LOWER_UP; 8750 if (netif_dormant(dev)) 8751 flags |= IFF_DORMANT; 8752 } 8753 8754 return flags; 8755 } 8756 EXPORT_SYMBOL(dev_get_flags); 8757 8758 int __dev_change_flags(struct net_device *dev, unsigned int flags, 8759 struct netlink_ext_ack *extack) 8760 { 8761 unsigned int old_flags = dev->flags; 8762 int ret; 8763 8764 ASSERT_RTNL(); 8765 8766 /* 8767 * Set the flags on our device. 8768 */ 8769 8770 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | 8771 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | 8772 IFF_AUTOMEDIA)) | 8773 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | 8774 IFF_ALLMULTI)); 8775 8776 /* 8777 * Load in the correct multicast list now the flags have changed. 8778 */ 8779 8780 if ((old_flags ^ flags) & IFF_MULTICAST) 8781 dev_change_rx_flags(dev, IFF_MULTICAST); 8782 8783 dev_set_rx_mode(dev); 8784 8785 /* 8786 * Have we downed the interface. We handle IFF_UP ourselves 8787 * according to user attempts to set it, rather than blindly 8788 * setting it. 8789 */ 8790 8791 ret = 0; 8792 if ((old_flags ^ flags) & IFF_UP) { 8793 if (old_flags & IFF_UP) 8794 __dev_close(dev); 8795 else 8796 ret = __dev_open(dev, extack); 8797 } 8798 8799 if ((flags ^ dev->gflags) & IFF_PROMISC) { 8800 int inc = (flags & IFF_PROMISC) ? 1 : -1; 8801 unsigned int old_flags = dev->flags; 8802 8803 dev->gflags ^= IFF_PROMISC; 8804 8805 if (__dev_set_promiscuity(dev, inc, false) >= 0) 8806 if (dev->flags != old_flags) 8807 dev_set_rx_mode(dev); 8808 } 8809 8810 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI 8811 * is important. Some (broken) drivers set IFF_PROMISC, when 8812 * IFF_ALLMULTI is requested not asking us and not reporting. 8813 */ 8814 if ((flags ^ dev->gflags) & IFF_ALLMULTI) { 8815 int inc = (flags & IFF_ALLMULTI) ? 1 : -1; 8816 8817 dev->gflags ^= IFF_ALLMULTI; 8818 __dev_set_allmulti(dev, inc, false); 8819 } 8820 8821 return ret; 8822 } 8823 8824 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, 8825 unsigned int gchanges) 8826 { 8827 unsigned int changes = dev->flags ^ old_flags; 8828 8829 if (gchanges) 8830 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC); 8831 8832 if (changes & IFF_UP) { 8833 if (dev->flags & IFF_UP) 8834 call_netdevice_notifiers(NETDEV_UP, dev); 8835 else 8836 call_netdevice_notifiers(NETDEV_DOWN, dev); 8837 } 8838 8839 if (dev->flags & IFF_UP && 8840 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { 8841 struct netdev_notifier_change_info change_info = { 8842 .info = { 8843 .dev = dev, 8844 }, 8845 .flags_changed = changes, 8846 }; 8847 8848 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info); 8849 } 8850 } 8851 8852 /** 8853 * dev_change_flags - change device settings 8854 * @dev: device 8855 * @flags: device state flags 8856 * @extack: netlink extended ack 8857 * 8858 * Change settings on device based state flags. The flags are 8859 * in the userspace exported format. 8860 */ 8861 int dev_change_flags(struct net_device *dev, unsigned int flags, 8862 struct netlink_ext_ack *extack) 8863 { 8864 int ret; 8865 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; 8866 8867 ret = __dev_change_flags(dev, flags, extack); 8868 if (ret < 0) 8869 return ret; 8870 8871 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); 8872 __dev_notify_flags(dev, old_flags, changes); 8873 return ret; 8874 } 8875 EXPORT_SYMBOL(dev_change_flags); 8876 8877 int __dev_set_mtu(struct net_device *dev, int new_mtu) 8878 { 8879 const struct net_device_ops *ops = dev->netdev_ops; 8880 8881 if (ops->ndo_change_mtu) 8882 return ops->ndo_change_mtu(dev, new_mtu); 8883 8884 /* Pairs with all the lockless reads of dev->mtu in the stack */ 8885 WRITE_ONCE(dev->mtu, new_mtu); 8886 return 0; 8887 } 8888 EXPORT_SYMBOL(__dev_set_mtu); 8889 8890 int dev_validate_mtu(struct net_device *dev, int new_mtu, 8891 struct netlink_ext_ack *extack) 8892 { 8893 /* MTU must be positive, and in range */ 8894 if (new_mtu < 0 || new_mtu < dev->min_mtu) { 8895 NL_SET_ERR_MSG(extack, "mtu less than device minimum"); 8896 return -EINVAL; 8897 } 8898 8899 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { 8900 NL_SET_ERR_MSG(extack, "mtu greater than device maximum"); 8901 return -EINVAL; 8902 } 8903 return 0; 8904 } 8905 8906 /** 8907 * dev_set_mtu_ext - Change maximum transfer unit 8908 * @dev: device 8909 * @new_mtu: new transfer unit 8910 * @extack: netlink extended ack 8911 * 8912 * Change the maximum transfer size of the network device. 8913 */ 8914 int dev_set_mtu_ext(struct net_device *dev, int new_mtu, 8915 struct netlink_ext_ack *extack) 8916 { 8917 int err, orig_mtu; 8918 8919 if (new_mtu == dev->mtu) 8920 return 0; 8921 8922 err = dev_validate_mtu(dev, new_mtu, extack); 8923 if (err) 8924 return err; 8925 8926 if (!netif_device_present(dev)) 8927 return -ENODEV; 8928 8929 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); 8930 err = notifier_to_errno(err); 8931 if (err) 8932 return err; 8933 8934 orig_mtu = dev->mtu; 8935 err = __dev_set_mtu(dev, new_mtu); 8936 8937 if (!err) { 8938 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8939 orig_mtu); 8940 err = notifier_to_errno(err); 8941 if (err) { 8942 /* setting mtu back and notifying everyone again, 8943 * so that they have a chance to revert changes. 8944 */ 8945 __dev_set_mtu(dev, orig_mtu); 8946 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8947 new_mtu); 8948 } 8949 } 8950 return err; 8951 } 8952 8953 int dev_set_mtu(struct net_device *dev, int new_mtu) 8954 { 8955 struct netlink_ext_ack extack; 8956 int err; 8957 8958 memset(&extack, 0, sizeof(extack)); 8959 err = dev_set_mtu_ext(dev, new_mtu, &extack); 8960 if (err && extack._msg) 8961 net_err_ratelimited("%s: %s\n", dev->name, extack._msg); 8962 return err; 8963 } 8964 EXPORT_SYMBOL(dev_set_mtu); 8965 8966 /** 8967 * dev_change_tx_queue_len - Change TX queue length of a netdevice 8968 * @dev: device 8969 * @new_len: new tx queue length 8970 */ 8971 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len) 8972 { 8973 unsigned int orig_len = dev->tx_queue_len; 8974 int res; 8975 8976 if (new_len != (unsigned int)new_len) 8977 return -ERANGE; 8978 8979 if (new_len != orig_len) { 8980 dev->tx_queue_len = new_len; 8981 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev); 8982 res = notifier_to_errno(res); 8983 if (res) 8984 goto err_rollback; 8985 res = dev_qdisc_change_tx_queue_len(dev); 8986 if (res) 8987 goto err_rollback; 8988 } 8989 8990 return 0; 8991 8992 err_rollback: 8993 netdev_err(dev, "refused to change device tx_queue_len\n"); 8994 dev->tx_queue_len = orig_len; 8995 return res; 8996 } 8997 8998 /** 8999 * dev_set_group - Change group this device belongs to 9000 * @dev: device 9001 * @new_group: group this device should belong to 9002 */ 9003 void dev_set_group(struct net_device *dev, int new_group) 9004 { 9005 dev->group = new_group; 9006 } 9007 EXPORT_SYMBOL(dev_set_group); 9008 9009 /** 9010 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR. 9011 * @dev: device 9012 * @addr: new address 9013 * @extack: netlink extended ack 9014 */ 9015 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr, 9016 struct netlink_ext_ack *extack) 9017 { 9018 struct netdev_notifier_pre_changeaddr_info info = { 9019 .info.dev = dev, 9020 .info.extack = extack, 9021 .dev_addr = addr, 9022 }; 9023 int rc; 9024 9025 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info); 9026 return notifier_to_errno(rc); 9027 } 9028 EXPORT_SYMBOL(dev_pre_changeaddr_notify); 9029 9030 /** 9031 * dev_set_mac_address - Change Media Access Control Address 9032 * @dev: device 9033 * @sa: new address 9034 * @extack: netlink extended ack 9035 * 9036 * Change the hardware (MAC) address of the device 9037 */ 9038 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa, 9039 struct netlink_ext_ack *extack) 9040 { 9041 const struct net_device_ops *ops = dev->netdev_ops; 9042 int err; 9043 9044 if (!ops->ndo_set_mac_address) 9045 return -EOPNOTSUPP; 9046 if (sa->sa_family != dev->type) 9047 return -EINVAL; 9048 if (!netif_device_present(dev)) 9049 return -ENODEV; 9050 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack); 9051 if (err) 9052 return err; 9053 err = ops->ndo_set_mac_address(dev, sa); 9054 if (err) 9055 return err; 9056 dev->addr_assign_type = NET_ADDR_SET; 9057 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); 9058 add_device_randomness(dev->dev_addr, dev->addr_len); 9059 return 0; 9060 } 9061 EXPORT_SYMBOL(dev_set_mac_address); 9062 9063 static DECLARE_RWSEM(dev_addr_sem); 9064 9065 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa, 9066 struct netlink_ext_ack *extack) 9067 { 9068 int ret; 9069 9070 down_write(&dev_addr_sem); 9071 ret = dev_set_mac_address(dev, sa, extack); 9072 up_write(&dev_addr_sem); 9073 return ret; 9074 } 9075 EXPORT_SYMBOL(dev_set_mac_address_user); 9076 9077 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name) 9078 { 9079 size_t size = sizeof(sa->sa_data); 9080 struct net_device *dev; 9081 int ret = 0; 9082 9083 down_read(&dev_addr_sem); 9084 rcu_read_lock(); 9085 9086 dev = dev_get_by_name_rcu(net, dev_name); 9087 if (!dev) { 9088 ret = -ENODEV; 9089 goto unlock; 9090 } 9091 if (!dev->addr_len) 9092 memset(sa->sa_data, 0, size); 9093 else 9094 memcpy(sa->sa_data, dev->dev_addr, 9095 min_t(size_t, size, dev->addr_len)); 9096 sa->sa_family = dev->type; 9097 9098 unlock: 9099 rcu_read_unlock(); 9100 up_read(&dev_addr_sem); 9101 return ret; 9102 } 9103 EXPORT_SYMBOL(dev_get_mac_address); 9104 9105 /** 9106 * dev_change_carrier - Change device carrier 9107 * @dev: device 9108 * @new_carrier: new value 9109 * 9110 * Change device carrier 9111 */ 9112 int dev_change_carrier(struct net_device *dev, bool new_carrier) 9113 { 9114 const struct net_device_ops *ops = dev->netdev_ops; 9115 9116 if (!ops->ndo_change_carrier) 9117 return -EOPNOTSUPP; 9118 if (!netif_device_present(dev)) 9119 return -ENODEV; 9120 return ops->ndo_change_carrier(dev, new_carrier); 9121 } 9122 EXPORT_SYMBOL(dev_change_carrier); 9123 9124 /** 9125 * dev_get_phys_port_id - Get device physical port ID 9126 * @dev: device 9127 * @ppid: port ID 9128 * 9129 * Get device physical port ID 9130 */ 9131 int dev_get_phys_port_id(struct net_device *dev, 9132 struct netdev_phys_item_id *ppid) 9133 { 9134 const struct net_device_ops *ops = dev->netdev_ops; 9135 9136 if (!ops->ndo_get_phys_port_id) 9137 return -EOPNOTSUPP; 9138 return ops->ndo_get_phys_port_id(dev, ppid); 9139 } 9140 EXPORT_SYMBOL(dev_get_phys_port_id); 9141 9142 /** 9143 * dev_get_phys_port_name - Get device physical port name 9144 * @dev: device 9145 * @name: port name 9146 * @len: limit of bytes to copy to name 9147 * 9148 * Get device physical port name 9149 */ 9150 int dev_get_phys_port_name(struct net_device *dev, 9151 char *name, size_t len) 9152 { 9153 const struct net_device_ops *ops = dev->netdev_ops; 9154 int err; 9155 9156 if (ops->ndo_get_phys_port_name) { 9157 err = ops->ndo_get_phys_port_name(dev, name, len); 9158 if (err != -EOPNOTSUPP) 9159 return err; 9160 } 9161 return devlink_compat_phys_port_name_get(dev, name, len); 9162 } 9163 EXPORT_SYMBOL(dev_get_phys_port_name); 9164 9165 /** 9166 * dev_get_port_parent_id - Get the device's port parent identifier 9167 * @dev: network device 9168 * @ppid: pointer to a storage for the port's parent identifier 9169 * @recurse: allow/disallow recursion to lower devices 9170 * 9171 * Get the devices's port parent identifier 9172 */ 9173 int dev_get_port_parent_id(struct net_device *dev, 9174 struct netdev_phys_item_id *ppid, 9175 bool recurse) 9176 { 9177 const struct net_device_ops *ops = dev->netdev_ops; 9178 struct netdev_phys_item_id first = { }; 9179 struct net_device *lower_dev; 9180 struct list_head *iter; 9181 int err; 9182 9183 if (ops->ndo_get_port_parent_id) { 9184 err = ops->ndo_get_port_parent_id(dev, ppid); 9185 if (err != -EOPNOTSUPP) 9186 return err; 9187 } 9188 9189 err = devlink_compat_switch_id_get(dev, ppid); 9190 if (!recurse || err != -EOPNOTSUPP) 9191 return err; 9192 9193 netdev_for_each_lower_dev(dev, lower_dev, iter) { 9194 err = dev_get_port_parent_id(lower_dev, ppid, true); 9195 if (err) 9196 break; 9197 if (!first.id_len) 9198 first = *ppid; 9199 else if (memcmp(&first, ppid, sizeof(*ppid))) 9200 return -EOPNOTSUPP; 9201 } 9202 9203 return err; 9204 } 9205 EXPORT_SYMBOL(dev_get_port_parent_id); 9206 9207 /** 9208 * netdev_port_same_parent_id - Indicate if two network devices have 9209 * the same port parent identifier 9210 * @a: first network device 9211 * @b: second network device 9212 */ 9213 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b) 9214 { 9215 struct netdev_phys_item_id a_id = { }; 9216 struct netdev_phys_item_id b_id = { }; 9217 9218 if (dev_get_port_parent_id(a, &a_id, true) || 9219 dev_get_port_parent_id(b, &b_id, true)) 9220 return false; 9221 9222 return netdev_phys_item_id_same(&a_id, &b_id); 9223 } 9224 EXPORT_SYMBOL(netdev_port_same_parent_id); 9225 9226 /** 9227 * dev_change_proto_down - update protocol port state information 9228 * @dev: device 9229 * @proto_down: new value 9230 * 9231 * This info can be used by switch drivers to set the phys state of the 9232 * port. 9233 */ 9234 int dev_change_proto_down(struct net_device *dev, bool proto_down) 9235 { 9236 const struct net_device_ops *ops = dev->netdev_ops; 9237 9238 if (!ops->ndo_change_proto_down) 9239 return -EOPNOTSUPP; 9240 if (!netif_device_present(dev)) 9241 return -ENODEV; 9242 return ops->ndo_change_proto_down(dev, proto_down); 9243 } 9244 EXPORT_SYMBOL(dev_change_proto_down); 9245 9246 /** 9247 * dev_change_proto_down_generic - generic implementation for 9248 * ndo_change_proto_down that sets carrier according to 9249 * proto_down. 9250 * 9251 * @dev: device 9252 * @proto_down: new value 9253 */ 9254 int dev_change_proto_down_generic(struct net_device *dev, bool proto_down) 9255 { 9256 if (proto_down) 9257 netif_carrier_off(dev); 9258 else 9259 netif_carrier_on(dev); 9260 dev->proto_down = proto_down; 9261 return 0; 9262 } 9263 EXPORT_SYMBOL(dev_change_proto_down_generic); 9264 9265 /** 9266 * dev_change_proto_down_reason - proto down reason 9267 * 9268 * @dev: device 9269 * @mask: proto down mask 9270 * @value: proto down value 9271 */ 9272 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask, 9273 u32 value) 9274 { 9275 int b; 9276 9277 if (!mask) { 9278 dev->proto_down_reason = value; 9279 } else { 9280 for_each_set_bit(b, &mask, 32) { 9281 if (value & (1 << b)) 9282 dev->proto_down_reason |= BIT(b); 9283 else 9284 dev->proto_down_reason &= ~BIT(b); 9285 } 9286 } 9287 } 9288 EXPORT_SYMBOL(dev_change_proto_down_reason); 9289 9290 struct bpf_xdp_link { 9291 struct bpf_link link; 9292 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */ 9293 int flags; 9294 }; 9295 9296 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags) 9297 { 9298 if (flags & XDP_FLAGS_HW_MODE) 9299 return XDP_MODE_HW; 9300 if (flags & XDP_FLAGS_DRV_MODE) 9301 return XDP_MODE_DRV; 9302 if (flags & XDP_FLAGS_SKB_MODE) 9303 return XDP_MODE_SKB; 9304 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB; 9305 } 9306 9307 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode) 9308 { 9309 switch (mode) { 9310 case XDP_MODE_SKB: 9311 return generic_xdp_install; 9312 case XDP_MODE_DRV: 9313 case XDP_MODE_HW: 9314 return dev->netdev_ops->ndo_bpf; 9315 default: 9316 return NULL; 9317 } 9318 } 9319 9320 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev, 9321 enum bpf_xdp_mode mode) 9322 { 9323 return dev->xdp_state[mode].link; 9324 } 9325 9326 static struct bpf_prog *dev_xdp_prog(struct net_device *dev, 9327 enum bpf_xdp_mode mode) 9328 { 9329 struct bpf_xdp_link *link = dev_xdp_link(dev, mode); 9330 9331 if (link) 9332 return link->link.prog; 9333 return dev->xdp_state[mode].prog; 9334 } 9335 9336 u8 dev_xdp_prog_count(struct net_device *dev) 9337 { 9338 u8 count = 0; 9339 int i; 9340 9341 for (i = 0; i < __MAX_XDP_MODE; i++) 9342 if (dev->xdp_state[i].prog || dev->xdp_state[i].link) 9343 count++; 9344 return count; 9345 } 9346 EXPORT_SYMBOL_GPL(dev_xdp_prog_count); 9347 9348 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode) 9349 { 9350 struct bpf_prog *prog = dev_xdp_prog(dev, mode); 9351 9352 return prog ? prog->aux->id : 0; 9353 } 9354 9355 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode, 9356 struct bpf_xdp_link *link) 9357 { 9358 dev->xdp_state[mode].link = link; 9359 dev->xdp_state[mode].prog = NULL; 9360 } 9361 9362 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode, 9363 struct bpf_prog *prog) 9364 { 9365 dev->xdp_state[mode].link = NULL; 9366 dev->xdp_state[mode].prog = prog; 9367 } 9368 9369 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode, 9370 bpf_op_t bpf_op, struct netlink_ext_ack *extack, 9371 u32 flags, struct bpf_prog *prog) 9372 { 9373 struct netdev_bpf xdp; 9374 int err; 9375 9376 memset(&xdp, 0, sizeof(xdp)); 9377 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG; 9378 xdp.extack = extack; 9379 xdp.flags = flags; 9380 xdp.prog = prog; 9381 9382 /* Drivers assume refcnt is already incremented (i.e, prog pointer is 9383 * "moved" into driver), so they don't increment it on their own, but 9384 * they do decrement refcnt when program is detached or replaced. 9385 * Given net_device also owns link/prog, we need to bump refcnt here 9386 * to prevent drivers from underflowing it. 9387 */ 9388 if (prog) 9389 bpf_prog_inc(prog); 9390 err = bpf_op(dev, &xdp); 9391 if (err) { 9392 if (prog) 9393 bpf_prog_put(prog); 9394 return err; 9395 } 9396 9397 if (mode != XDP_MODE_HW) 9398 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog); 9399 9400 return 0; 9401 } 9402 9403 static void dev_xdp_uninstall(struct net_device *dev) 9404 { 9405 struct bpf_xdp_link *link; 9406 struct bpf_prog *prog; 9407 enum bpf_xdp_mode mode; 9408 bpf_op_t bpf_op; 9409 9410 ASSERT_RTNL(); 9411 9412 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) { 9413 prog = dev_xdp_prog(dev, mode); 9414 if (!prog) 9415 continue; 9416 9417 bpf_op = dev_xdp_bpf_op(dev, mode); 9418 if (!bpf_op) 9419 continue; 9420 9421 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9422 9423 /* auto-detach link from net device */ 9424 link = dev_xdp_link(dev, mode); 9425 if (link) 9426 link->dev = NULL; 9427 else 9428 bpf_prog_put(prog); 9429 9430 dev_xdp_set_link(dev, mode, NULL); 9431 } 9432 } 9433 9434 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack, 9435 struct bpf_xdp_link *link, struct bpf_prog *new_prog, 9436 struct bpf_prog *old_prog, u32 flags) 9437 { 9438 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES); 9439 struct bpf_prog *cur_prog; 9440 struct net_device *upper; 9441 struct list_head *iter; 9442 enum bpf_xdp_mode mode; 9443 bpf_op_t bpf_op; 9444 int err; 9445 9446 ASSERT_RTNL(); 9447 9448 /* either link or prog attachment, never both */ 9449 if (link && (new_prog || old_prog)) 9450 return -EINVAL; 9451 /* link supports only XDP mode flags */ 9452 if (link && (flags & ~XDP_FLAGS_MODES)) { 9453 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment"); 9454 return -EINVAL; 9455 } 9456 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */ 9457 if (num_modes > 1) { 9458 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set"); 9459 return -EINVAL; 9460 } 9461 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */ 9462 if (!num_modes && dev_xdp_prog_count(dev) > 1) { 9463 NL_SET_ERR_MSG(extack, 9464 "More than one program loaded, unset mode is ambiguous"); 9465 return -EINVAL; 9466 } 9467 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */ 9468 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) { 9469 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified"); 9470 return -EINVAL; 9471 } 9472 9473 mode = dev_xdp_mode(dev, flags); 9474 /* can't replace attached link */ 9475 if (dev_xdp_link(dev, mode)) { 9476 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link"); 9477 return -EBUSY; 9478 } 9479 9480 /* don't allow if an upper device already has a program */ 9481 netdev_for_each_upper_dev_rcu(dev, upper, iter) { 9482 if (dev_xdp_prog_count(upper) > 0) { 9483 NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program"); 9484 return -EEXIST; 9485 } 9486 } 9487 9488 cur_prog = dev_xdp_prog(dev, mode); 9489 /* can't replace attached prog with link */ 9490 if (link && cur_prog) { 9491 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link"); 9492 return -EBUSY; 9493 } 9494 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) { 9495 NL_SET_ERR_MSG(extack, "Active program does not match expected"); 9496 return -EEXIST; 9497 } 9498 9499 /* put effective new program into new_prog */ 9500 if (link) 9501 new_prog = link->link.prog; 9502 9503 if (new_prog) { 9504 bool offload = mode == XDP_MODE_HW; 9505 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB 9506 ? XDP_MODE_DRV : XDP_MODE_SKB; 9507 9508 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) { 9509 NL_SET_ERR_MSG(extack, "XDP program already attached"); 9510 return -EBUSY; 9511 } 9512 if (!offload && dev_xdp_prog(dev, other_mode)) { 9513 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time"); 9514 return -EEXIST; 9515 } 9516 if (!offload && bpf_prog_is_dev_bound(new_prog->aux)) { 9517 NL_SET_ERR_MSG(extack, "Using device-bound program without HW_MODE flag is not supported"); 9518 return -EINVAL; 9519 } 9520 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) { 9521 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device"); 9522 return -EINVAL; 9523 } 9524 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) { 9525 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device"); 9526 return -EINVAL; 9527 } 9528 } 9529 9530 /* don't call drivers if the effective program didn't change */ 9531 if (new_prog != cur_prog) { 9532 bpf_op = dev_xdp_bpf_op(dev, mode); 9533 if (!bpf_op) { 9534 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode"); 9535 return -EOPNOTSUPP; 9536 } 9537 9538 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog); 9539 if (err) 9540 return err; 9541 } 9542 9543 if (link) 9544 dev_xdp_set_link(dev, mode, link); 9545 else 9546 dev_xdp_set_prog(dev, mode, new_prog); 9547 if (cur_prog) 9548 bpf_prog_put(cur_prog); 9549 9550 return 0; 9551 } 9552 9553 static int dev_xdp_attach_link(struct net_device *dev, 9554 struct netlink_ext_ack *extack, 9555 struct bpf_xdp_link *link) 9556 { 9557 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags); 9558 } 9559 9560 static int dev_xdp_detach_link(struct net_device *dev, 9561 struct netlink_ext_ack *extack, 9562 struct bpf_xdp_link *link) 9563 { 9564 enum bpf_xdp_mode mode; 9565 bpf_op_t bpf_op; 9566 9567 ASSERT_RTNL(); 9568 9569 mode = dev_xdp_mode(dev, link->flags); 9570 if (dev_xdp_link(dev, mode) != link) 9571 return -EINVAL; 9572 9573 bpf_op = dev_xdp_bpf_op(dev, mode); 9574 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9575 dev_xdp_set_link(dev, mode, NULL); 9576 return 0; 9577 } 9578 9579 static void bpf_xdp_link_release(struct bpf_link *link) 9580 { 9581 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9582 9583 rtnl_lock(); 9584 9585 /* if racing with net_device's tear down, xdp_link->dev might be 9586 * already NULL, in which case link was already auto-detached 9587 */ 9588 if (xdp_link->dev) { 9589 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link)); 9590 xdp_link->dev = NULL; 9591 } 9592 9593 rtnl_unlock(); 9594 } 9595 9596 static int bpf_xdp_link_detach(struct bpf_link *link) 9597 { 9598 bpf_xdp_link_release(link); 9599 return 0; 9600 } 9601 9602 static void bpf_xdp_link_dealloc(struct bpf_link *link) 9603 { 9604 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9605 9606 kfree(xdp_link); 9607 } 9608 9609 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link, 9610 struct seq_file *seq) 9611 { 9612 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9613 u32 ifindex = 0; 9614 9615 rtnl_lock(); 9616 if (xdp_link->dev) 9617 ifindex = xdp_link->dev->ifindex; 9618 rtnl_unlock(); 9619 9620 seq_printf(seq, "ifindex:\t%u\n", ifindex); 9621 } 9622 9623 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link, 9624 struct bpf_link_info *info) 9625 { 9626 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9627 u32 ifindex = 0; 9628 9629 rtnl_lock(); 9630 if (xdp_link->dev) 9631 ifindex = xdp_link->dev->ifindex; 9632 rtnl_unlock(); 9633 9634 info->xdp.ifindex = ifindex; 9635 return 0; 9636 } 9637 9638 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog, 9639 struct bpf_prog *old_prog) 9640 { 9641 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9642 enum bpf_xdp_mode mode; 9643 bpf_op_t bpf_op; 9644 int err = 0; 9645 9646 rtnl_lock(); 9647 9648 /* link might have been auto-released already, so fail */ 9649 if (!xdp_link->dev) { 9650 err = -ENOLINK; 9651 goto out_unlock; 9652 } 9653 9654 if (old_prog && link->prog != old_prog) { 9655 err = -EPERM; 9656 goto out_unlock; 9657 } 9658 old_prog = link->prog; 9659 if (old_prog == new_prog) { 9660 /* no-op, don't disturb drivers */ 9661 bpf_prog_put(new_prog); 9662 goto out_unlock; 9663 } 9664 9665 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags); 9666 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode); 9667 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL, 9668 xdp_link->flags, new_prog); 9669 if (err) 9670 goto out_unlock; 9671 9672 old_prog = xchg(&link->prog, new_prog); 9673 bpf_prog_put(old_prog); 9674 9675 out_unlock: 9676 rtnl_unlock(); 9677 return err; 9678 } 9679 9680 static const struct bpf_link_ops bpf_xdp_link_lops = { 9681 .release = bpf_xdp_link_release, 9682 .dealloc = bpf_xdp_link_dealloc, 9683 .detach = bpf_xdp_link_detach, 9684 .show_fdinfo = bpf_xdp_link_show_fdinfo, 9685 .fill_link_info = bpf_xdp_link_fill_link_info, 9686 .update_prog = bpf_xdp_link_update, 9687 }; 9688 9689 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog) 9690 { 9691 struct net *net = current->nsproxy->net_ns; 9692 struct bpf_link_primer link_primer; 9693 struct bpf_xdp_link *link; 9694 struct net_device *dev; 9695 int err, fd; 9696 9697 rtnl_lock(); 9698 dev = dev_get_by_index(net, attr->link_create.target_ifindex); 9699 if (!dev) { 9700 rtnl_unlock(); 9701 return -EINVAL; 9702 } 9703 9704 link = kzalloc(sizeof(*link), GFP_USER); 9705 if (!link) { 9706 err = -ENOMEM; 9707 goto unlock; 9708 } 9709 9710 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog); 9711 link->dev = dev; 9712 link->flags = attr->link_create.flags; 9713 9714 err = bpf_link_prime(&link->link, &link_primer); 9715 if (err) { 9716 kfree(link); 9717 goto unlock; 9718 } 9719 9720 err = dev_xdp_attach_link(dev, NULL, link); 9721 rtnl_unlock(); 9722 9723 if (err) { 9724 link->dev = NULL; 9725 bpf_link_cleanup(&link_primer); 9726 goto out_put_dev; 9727 } 9728 9729 fd = bpf_link_settle(&link_primer); 9730 /* link itself doesn't hold dev's refcnt to not complicate shutdown */ 9731 dev_put(dev); 9732 return fd; 9733 9734 unlock: 9735 rtnl_unlock(); 9736 9737 out_put_dev: 9738 dev_put(dev); 9739 return err; 9740 } 9741 9742 /** 9743 * dev_change_xdp_fd - set or clear a bpf program for a device rx path 9744 * @dev: device 9745 * @extack: netlink extended ack 9746 * @fd: new program fd or negative value to clear 9747 * @expected_fd: old program fd that userspace expects to replace or clear 9748 * @flags: xdp-related flags 9749 * 9750 * Set or clear a bpf program for a device 9751 */ 9752 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, 9753 int fd, int expected_fd, u32 flags) 9754 { 9755 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags); 9756 struct bpf_prog *new_prog = NULL, *old_prog = NULL; 9757 int err; 9758 9759 ASSERT_RTNL(); 9760 9761 if (fd >= 0) { 9762 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP, 9763 mode != XDP_MODE_SKB); 9764 if (IS_ERR(new_prog)) 9765 return PTR_ERR(new_prog); 9766 } 9767 9768 if (expected_fd >= 0) { 9769 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP, 9770 mode != XDP_MODE_SKB); 9771 if (IS_ERR(old_prog)) { 9772 err = PTR_ERR(old_prog); 9773 old_prog = NULL; 9774 goto err_out; 9775 } 9776 } 9777 9778 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags); 9779 9780 err_out: 9781 if (err && new_prog) 9782 bpf_prog_put(new_prog); 9783 if (old_prog) 9784 bpf_prog_put(old_prog); 9785 return err; 9786 } 9787 9788 /** 9789 * dev_new_index - allocate an ifindex 9790 * @net: the applicable net namespace 9791 * 9792 * Returns a suitable unique value for a new device interface 9793 * number. The caller must hold the rtnl semaphore or the 9794 * dev_base_lock to be sure it remains unique. 9795 */ 9796 static int dev_new_index(struct net *net) 9797 { 9798 int ifindex = net->ifindex; 9799 9800 for (;;) { 9801 if (++ifindex <= 0) 9802 ifindex = 1; 9803 if (!__dev_get_by_index(net, ifindex)) 9804 return net->ifindex = ifindex; 9805 } 9806 } 9807 9808 /* Delayed registration/unregisteration */ 9809 static LIST_HEAD(net_todo_list); 9810 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); 9811 9812 static void net_set_todo(struct net_device *dev) 9813 { 9814 list_add_tail(&dev->todo_list, &net_todo_list); 9815 dev_net(dev)->dev_unreg_count++; 9816 } 9817 9818 static netdev_features_t netdev_sync_upper_features(struct net_device *lower, 9819 struct net_device *upper, netdev_features_t features) 9820 { 9821 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9822 netdev_features_t feature; 9823 int feature_bit; 9824 9825 for_each_netdev_feature(upper_disables, feature_bit) { 9826 feature = __NETIF_F_BIT(feature_bit); 9827 if (!(upper->wanted_features & feature) 9828 && (features & feature)) { 9829 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", 9830 &feature, upper->name); 9831 features &= ~feature; 9832 } 9833 } 9834 9835 return features; 9836 } 9837 9838 static void netdev_sync_lower_features(struct net_device *upper, 9839 struct net_device *lower, netdev_features_t features) 9840 { 9841 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9842 netdev_features_t feature; 9843 int feature_bit; 9844 9845 for_each_netdev_feature(upper_disables, feature_bit) { 9846 feature = __NETIF_F_BIT(feature_bit); 9847 if (!(features & feature) && (lower->features & feature)) { 9848 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", 9849 &feature, lower->name); 9850 lower->wanted_features &= ~feature; 9851 __netdev_update_features(lower); 9852 9853 if (unlikely(lower->features & feature)) 9854 netdev_WARN(upper, "failed to disable %pNF on %s!\n", 9855 &feature, lower->name); 9856 else 9857 netdev_features_change(lower); 9858 } 9859 } 9860 } 9861 9862 static netdev_features_t netdev_fix_features(struct net_device *dev, 9863 netdev_features_t features) 9864 { 9865 /* Fix illegal checksum combinations */ 9866 if ((features & NETIF_F_HW_CSUM) && 9867 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { 9868 netdev_warn(dev, "mixed HW and IP checksum settings.\n"); 9869 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); 9870 } 9871 9872 /* TSO requires that SG is present as well. */ 9873 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { 9874 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); 9875 features &= ~NETIF_F_ALL_TSO; 9876 } 9877 9878 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && 9879 !(features & NETIF_F_IP_CSUM)) { 9880 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); 9881 features &= ~NETIF_F_TSO; 9882 features &= ~NETIF_F_TSO_ECN; 9883 } 9884 9885 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && 9886 !(features & NETIF_F_IPV6_CSUM)) { 9887 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); 9888 features &= ~NETIF_F_TSO6; 9889 } 9890 9891 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ 9892 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) 9893 features &= ~NETIF_F_TSO_MANGLEID; 9894 9895 /* TSO ECN requires that TSO is present as well. */ 9896 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) 9897 features &= ~NETIF_F_TSO_ECN; 9898 9899 /* Software GSO depends on SG. */ 9900 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { 9901 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); 9902 features &= ~NETIF_F_GSO; 9903 } 9904 9905 /* GSO partial features require GSO partial be set */ 9906 if ((features & dev->gso_partial_features) && 9907 !(features & NETIF_F_GSO_PARTIAL)) { 9908 netdev_dbg(dev, 9909 "Dropping partially supported GSO features since no GSO partial.\n"); 9910 features &= ~dev->gso_partial_features; 9911 } 9912 9913 if (!(features & NETIF_F_RXCSUM)) { 9914 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet 9915 * successfully merged by hardware must also have the 9916 * checksum verified by hardware. If the user does not 9917 * want to enable RXCSUM, logically, we should disable GRO_HW. 9918 */ 9919 if (features & NETIF_F_GRO_HW) { 9920 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n"); 9921 features &= ~NETIF_F_GRO_HW; 9922 } 9923 } 9924 9925 /* LRO/HW-GRO features cannot be combined with RX-FCS */ 9926 if (features & NETIF_F_RXFCS) { 9927 if (features & NETIF_F_LRO) { 9928 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n"); 9929 features &= ~NETIF_F_LRO; 9930 } 9931 9932 if (features & NETIF_F_GRO_HW) { 9933 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n"); 9934 features &= ~NETIF_F_GRO_HW; 9935 } 9936 } 9937 9938 if ((features & NETIF_F_GRO_HW) && (features & NETIF_F_LRO)) { 9939 netdev_dbg(dev, "Dropping LRO feature since HW-GRO is requested.\n"); 9940 features &= ~NETIF_F_LRO; 9941 } 9942 9943 if (features & NETIF_F_HW_TLS_TX) { 9944 bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) == 9945 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM); 9946 bool hw_csum = features & NETIF_F_HW_CSUM; 9947 9948 if (!ip_csum && !hw_csum) { 9949 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n"); 9950 features &= ~NETIF_F_HW_TLS_TX; 9951 } 9952 } 9953 9954 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) { 9955 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n"); 9956 features &= ~NETIF_F_HW_TLS_RX; 9957 } 9958 9959 return features; 9960 } 9961 9962 int __netdev_update_features(struct net_device *dev) 9963 { 9964 struct net_device *upper, *lower; 9965 netdev_features_t features; 9966 struct list_head *iter; 9967 int err = -1; 9968 9969 ASSERT_RTNL(); 9970 9971 features = netdev_get_wanted_features(dev); 9972 9973 if (dev->netdev_ops->ndo_fix_features) 9974 features = dev->netdev_ops->ndo_fix_features(dev, features); 9975 9976 /* driver might be less strict about feature dependencies */ 9977 features = netdev_fix_features(dev, features); 9978 9979 /* some features can't be enabled if they're off on an upper device */ 9980 netdev_for_each_upper_dev_rcu(dev, upper, iter) 9981 features = netdev_sync_upper_features(dev, upper, features); 9982 9983 if (dev->features == features) 9984 goto sync_lower; 9985 9986 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", 9987 &dev->features, &features); 9988 9989 if (dev->netdev_ops->ndo_set_features) 9990 err = dev->netdev_ops->ndo_set_features(dev, features); 9991 else 9992 err = 0; 9993 9994 if (unlikely(err < 0)) { 9995 netdev_err(dev, 9996 "set_features() failed (%d); wanted %pNF, left %pNF\n", 9997 err, &features, &dev->features); 9998 /* return non-0 since some features might have changed and 9999 * it's better to fire a spurious notification than miss it 10000 */ 10001 return -1; 10002 } 10003 10004 sync_lower: 10005 /* some features must be disabled on lower devices when disabled 10006 * on an upper device (think: bonding master or bridge) 10007 */ 10008 netdev_for_each_lower_dev(dev, lower, iter) 10009 netdev_sync_lower_features(dev, lower, features); 10010 10011 if (!err) { 10012 netdev_features_t diff = features ^ dev->features; 10013 10014 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) { 10015 /* udp_tunnel_{get,drop}_rx_info both need 10016 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the 10017 * device, or they won't do anything. 10018 * Thus we need to update dev->features 10019 * *before* calling udp_tunnel_get_rx_info, 10020 * but *after* calling udp_tunnel_drop_rx_info. 10021 */ 10022 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) { 10023 dev->features = features; 10024 udp_tunnel_get_rx_info(dev); 10025 } else { 10026 udp_tunnel_drop_rx_info(dev); 10027 } 10028 } 10029 10030 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) { 10031 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) { 10032 dev->features = features; 10033 err |= vlan_get_rx_ctag_filter_info(dev); 10034 } else { 10035 vlan_drop_rx_ctag_filter_info(dev); 10036 } 10037 } 10038 10039 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) { 10040 if (features & NETIF_F_HW_VLAN_STAG_FILTER) { 10041 dev->features = features; 10042 err |= vlan_get_rx_stag_filter_info(dev); 10043 } else { 10044 vlan_drop_rx_stag_filter_info(dev); 10045 } 10046 } 10047 10048 dev->features = features; 10049 } 10050 10051 return err < 0 ? 0 : 1; 10052 } 10053 10054 /** 10055 * netdev_update_features - recalculate device features 10056 * @dev: the device to check 10057 * 10058 * Recalculate dev->features set and send notifications if it 10059 * has changed. Should be called after driver or hardware dependent 10060 * conditions might have changed that influence the features. 10061 */ 10062 void netdev_update_features(struct net_device *dev) 10063 { 10064 if (__netdev_update_features(dev)) 10065 netdev_features_change(dev); 10066 } 10067 EXPORT_SYMBOL(netdev_update_features); 10068 10069 /** 10070 * netdev_change_features - recalculate device features 10071 * @dev: the device to check 10072 * 10073 * Recalculate dev->features set and send notifications even 10074 * if they have not changed. Should be called instead of 10075 * netdev_update_features() if also dev->vlan_features might 10076 * have changed to allow the changes to be propagated to stacked 10077 * VLAN devices. 10078 */ 10079 void netdev_change_features(struct net_device *dev) 10080 { 10081 __netdev_update_features(dev); 10082 netdev_features_change(dev); 10083 } 10084 EXPORT_SYMBOL(netdev_change_features); 10085 10086 /** 10087 * netif_stacked_transfer_operstate - transfer operstate 10088 * @rootdev: the root or lower level device to transfer state from 10089 * @dev: the device to transfer operstate to 10090 * 10091 * Transfer operational state from root to device. This is normally 10092 * called when a stacking relationship exists between the root 10093 * device and the device(a leaf device). 10094 */ 10095 void netif_stacked_transfer_operstate(const struct net_device *rootdev, 10096 struct net_device *dev) 10097 { 10098 if (rootdev->operstate == IF_OPER_DORMANT) 10099 netif_dormant_on(dev); 10100 else 10101 netif_dormant_off(dev); 10102 10103 if (rootdev->operstate == IF_OPER_TESTING) 10104 netif_testing_on(dev); 10105 else 10106 netif_testing_off(dev); 10107 10108 if (netif_carrier_ok(rootdev)) 10109 netif_carrier_on(dev); 10110 else 10111 netif_carrier_off(dev); 10112 } 10113 EXPORT_SYMBOL(netif_stacked_transfer_operstate); 10114 10115 static int netif_alloc_rx_queues(struct net_device *dev) 10116 { 10117 unsigned int i, count = dev->num_rx_queues; 10118 struct netdev_rx_queue *rx; 10119 size_t sz = count * sizeof(*rx); 10120 int err = 0; 10121 10122 BUG_ON(count < 1); 10123 10124 rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10125 if (!rx) 10126 return -ENOMEM; 10127 10128 dev->_rx = rx; 10129 10130 for (i = 0; i < count; i++) { 10131 rx[i].dev = dev; 10132 10133 /* XDP RX-queue setup */ 10134 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0); 10135 if (err < 0) 10136 goto err_rxq_info; 10137 } 10138 return 0; 10139 10140 err_rxq_info: 10141 /* Rollback successful reg's and free other resources */ 10142 while (i--) 10143 xdp_rxq_info_unreg(&rx[i].xdp_rxq); 10144 kvfree(dev->_rx); 10145 dev->_rx = NULL; 10146 return err; 10147 } 10148 10149 static void netif_free_rx_queues(struct net_device *dev) 10150 { 10151 unsigned int i, count = dev->num_rx_queues; 10152 10153 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */ 10154 if (!dev->_rx) 10155 return; 10156 10157 for (i = 0; i < count; i++) 10158 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq); 10159 10160 kvfree(dev->_rx); 10161 } 10162 10163 static void netdev_init_one_queue(struct net_device *dev, 10164 struct netdev_queue *queue, void *_unused) 10165 { 10166 /* Initialize queue lock */ 10167 spin_lock_init(&queue->_xmit_lock); 10168 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); 10169 queue->xmit_lock_owner = -1; 10170 netdev_queue_numa_node_write(queue, NUMA_NO_NODE); 10171 queue->dev = dev; 10172 #ifdef CONFIG_BQL 10173 dql_init(&queue->dql, HZ); 10174 #endif 10175 } 10176 10177 static void netif_free_tx_queues(struct net_device *dev) 10178 { 10179 kvfree(dev->_tx); 10180 } 10181 10182 static int netif_alloc_netdev_queues(struct net_device *dev) 10183 { 10184 unsigned int count = dev->num_tx_queues; 10185 struct netdev_queue *tx; 10186 size_t sz = count * sizeof(*tx); 10187 10188 if (count < 1 || count > 0xffff) 10189 return -EINVAL; 10190 10191 tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10192 if (!tx) 10193 return -ENOMEM; 10194 10195 dev->_tx = tx; 10196 10197 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); 10198 spin_lock_init(&dev->tx_global_lock); 10199 10200 return 0; 10201 } 10202 10203 void netif_tx_stop_all_queues(struct net_device *dev) 10204 { 10205 unsigned int i; 10206 10207 for (i = 0; i < dev->num_tx_queues; i++) { 10208 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 10209 10210 netif_tx_stop_queue(txq); 10211 } 10212 } 10213 EXPORT_SYMBOL(netif_tx_stop_all_queues); 10214 10215 /** 10216 * register_netdevice - register a network device 10217 * @dev: device to register 10218 * 10219 * Take a completed network device structure and add it to the kernel 10220 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 10221 * chain. 0 is returned on success. A negative errno code is returned 10222 * on a failure to set up the device, or if the name is a duplicate. 10223 * 10224 * Callers must hold the rtnl semaphore. You may want 10225 * register_netdev() instead of this. 10226 * 10227 * BUGS: 10228 * The locking appears insufficient to guarantee two parallel registers 10229 * will not get the same name. 10230 */ 10231 10232 int register_netdevice(struct net_device *dev) 10233 { 10234 int ret; 10235 struct net *net = dev_net(dev); 10236 10237 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE < 10238 NETDEV_FEATURE_COUNT); 10239 BUG_ON(dev_boot_phase); 10240 ASSERT_RTNL(); 10241 10242 might_sleep(); 10243 10244 /* When net_device's are persistent, this will be fatal. */ 10245 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); 10246 BUG_ON(!net); 10247 10248 ret = ethtool_check_ops(dev->ethtool_ops); 10249 if (ret) 10250 return ret; 10251 10252 spin_lock_init(&dev->addr_list_lock); 10253 netdev_set_addr_lockdep_class(dev); 10254 10255 ret = dev_get_valid_name(net, dev, dev->name); 10256 if (ret < 0) 10257 goto out; 10258 10259 ret = -ENOMEM; 10260 dev->name_node = netdev_name_node_head_alloc(dev); 10261 if (!dev->name_node) 10262 goto out; 10263 10264 /* Init, if this function is available */ 10265 if (dev->netdev_ops->ndo_init) { 10266 ret = dev->netdev_ops->ndo_init(dev); 10267 if (ret) { 10268 if (ret > 0) 10269 ret = -EIO; 10270 goto err_free_name; 10271 } 10272 } 10273 10274 if (((dev->hw_features | dev->features) & 10275 NETIF_F_HW_VLAN_CTAG_FILTER) && 10276 (!dev->netdev_ops->ndo_vlan_rx_add_vid || 10277 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { 10278 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); 10279 ret = -EINVAL; 10280 goto err_uninit; 10281 } 10282 10283 ret = -EBUSY; 10284 if (!dev->ifindex) 10285 dev->ifindex = dev_new_index(net); 10286 else if (__dev_get_by_index(net, dev->ifindex)) 10287 goto err_uninit; 10288 10289 /* Transfer changeable features to wanted_features and enable 10290 * software offloads (GSO and GRO). 10291 */ 10292 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF); 10293 dev->features |= NETIF_F_SOFT_FEATURES; 10294 10295 if (dev->udp_tunnel_nic_info) { 10296 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10297 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10298 } 10299 10300 dev->wanted_features = dev->features & dev->hw_features; 10301 10302 if (!(dev->flags & IFF_LOOPBACK)) 10303 dev->hw_features |= NETIF_F_NOCACHE_COPY; 10304 10305 /* If IPv4 TCP segmentation offload is supported we should also 10306 * allow the device to enable segmenting the frame with the option 10307 * of ignoring a static IP ID value. This doesn't enable the 10308 * feature itself but allows the user to enable it later. 10309 */ 10310 if (dev->hw_features & NETIF_F_TSO) 10311 dev->hw_features |= NETIF_F_TSO_MANGLEID; 10312 if (dev->vlan_features & NETIF_F_TSO) 10313 dev->vlan_features |= NETIF_F_TSO_MANGLEID; 10314 if (dev->mpls_features & NETIF_F_TSO) 10315 dev->mpls_features |= NETIF_F_TSO_MANGLEID; 10316 if (dev->hw_enc_features & NETIF_F_TSO) 10317 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; 10318 10319 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. 10320 */ 10321 dev->vlan_features |= NETIF_F_HIGHDMA; 10322 10323 /* Make NETIF_F_SG inheritable to tunnel devices. 10324 */ 10325 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; 10326 10327 /* Make NETIF_F_SG inheritable to MPLS. 10328 */ 10329 dev->mpls_features |= NETIF_F_SG; 10330 10331 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); 10332 ret = notifier_to_errno(ret); 10333 if (ret) 10334 goto err_uninit; 10335 10336 ret = netdev_register_kobject(dev); 10337 if (ret) { 10338 dev->reg_state = NETREG_UNREGISTERED; 10339 goto err_uninit; 10340 } 10341 dev->reg_state = NETREG_REGISTERED; 10342 10343 __netdev_update_features(dev); 10344 10345 /* 10346 * Default initial state at registry is that the 10347 * device is present. 10348 */ 10349 10350 set_bit(__LINK_STATE_PRESENT, &dev->state); 10351 10352 linkwatch_init_dev(dev); 10353 10354 dev_init_scheduler(dev); 10355 dev_hold(dev); 10356 list_netdevice(dev); 10357 add_device_randomness(dev->dev_addr, dev->addr_len); 10358 10359 /* If the device has permanent device address, driver should 10360 * set dev_addr and also addr_assign_type should be set to 10361 * NET_ADDR_PERM (default value). 10362 */ 10363 if (dev->addr_assign_type == NET_ADDR_PERM) 10364 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); 10365 10366 /* Notify protocols, that a new device appeared. */ 10367 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); 10368 ret = notifier_to_errno(ret); 10369 if (ret) { 10370 /* Expect explicit free_netdev() on failure */ 10371 dev->needs_free_netdev = false; 10372 unregister_netdevice_queue(dev, NULL); 10373 goto out; 10374 } 10375 /* 10376 * Prevent userspace races by waiting until the network 10377 * device is fully setup before sending notifications. 10378 */ 10379 if (!dev->rtnl_link_ops || 10380 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 10381 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); 10382 10383 out: 10384 return ret; 10385 10386 err_uninit: 10387 if (dev->netdev_ops->ndo_uninit) 10388 dev->netdev_ops->ndo_uninit(dev); 10389 if (dev->priv_destructor) 10390 dev->priv_destructor(dev); 10391 err_free_name: 10392 netdev_name_node_free(dev->name_node); 10393 goto out; 10394 } 10395 EXPORT_SYMBOL(register_netdevice); 10396 10397 /** 10398 * init_dummy_netdev - init a dummy network device for NAPI 10399 * @dev: device to init 10400 * 10401 * This takes a network device structure and initialize the minimum 10402 * amount of fields so it can be used to schedule NAPI polls without 10403 * registering a full blown interface. This is to be used by drivers 10404 * that need to tie several hardware interfaces to a single NAPI 10405 * poll scheduler due to HW limitations. 10406 */ 10407 int init_dummy_netdev(struct net_device *dev) 10408 { 10409 /* Clear everything. Note we don't initialize spinlocks 10410 * are they aren't supposed to be taken by any of the 10411 * NAPI code and this dummy netdev is supposed to be 10412 * only ever used for NAPI polls 10413 */ 10414 memset(dev, 0, sizeof(struct net_device)); 10415 10416 /* make sure we BUG if trying to hit standard 10417 * register/unregister code path 10418 */ 10419 dev->reg_state = NETREG_DUMMY; 10420 10421 /* NAPI wants this */ 10422 INIT_LIST_HEAD(&dev->napi_list); 10423 10424 /* a dummy interface is started by default */ 10425 set_bit(__LINK_STATE_PRESENT, &dev->state); 10426 set_bit(__LINK_STATE_START, &dev->state); 10427 10428 /* napi_busy_loop stats accounting wants this */ 10429 dev_net_set(dev, &init_net); 10430 10431 /* Note : We dont allocate pcpu_refcnt for dummy devices, 10432 * because users of this 'device' dont need to change 10433 * its refcount. 10434 */ 10435 10436 return 0; 10437 } 10438 EXPORT_SYMBOL_GPL(init_dummy_netdev); 10439 10440 10441 /** 10442 * register_netdev - register a network device 10443 * @dev: device to register 10444 * 10445 * Take a completed network device structure and add it to the kernel 10446 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 10447 * chain. 0 is returned on success. A negative errno code is returned 10448 * on a failure to set up the device, or if the name is a duplicate. 10449 * 10450 * This is a wrapper around register_netdevice that takes the rtnl semaphore 10451 * and expands the device name if you passed a format string to 10452 * alloc_netdev. 10453 */ 10454 int register_netdev(struct net_device *dev) 10455 { 10456 int err; 10457 10458 if (rtnl_lock_killable()) 10459 return -EINTR; 10460 err = register_netdevice(dev); 10461 rtnl_unlock(); 10462 return err; 10463 } 10464 EXPORT_SYMBOL(register_netdev); 10465 10466 int netdev_refcnt_read(const struct net_device *dev) 10467 { 10468 #ifdef CONFIG_PCPU_DEV_REFCNT 10469 int i, refcnt = 0; 10470 10471 for_each_possible_cpu(i) 10472 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); 10473 return refcnt; 10474 #else 10475 return refcount_read(&dev->dev_refcnt); 10476 #endif 10477 } 10478 EXPORT_SYMBOL(netdev_refcnt_read); 10479 10480 int netdev_unregister_timeout_secs __read_mostly = 10; 10481 10482 #define WAIT_REFS_MIN_MSECS 1 10483 #define WAIT_REFS_MAX_MSECS 250 10484 /** 10485 * netdev_wait_allrefs - wait until all references are gone. 10486 * @dev: target net_device 10487 * 10488 * This is called when unregistering network devices. 10489 * 10490 * Any protocol or device that holds a reference should register 10491 * for netdevice notification, and cleanup and put back the 10492 * reference if they receive an UNREGISTER event. 10493 * We can get stuck here if buggy protocols don't correctly 10494 * call dev_put. 10495 */ 10496 static void netdev_wait_allrefs(struct net_device *dev) 10497 { 10498 unsigned long rebroadcast_time, warning_time; 10499 int wait = 0, refcnt; 10500 10501 linkwatch_forget_dev(dev); 10502 10503 rebroadcast_time = warning_time = jiffies; 10504 refcnt = netdev_refcnt_read(dev); 10505 10506 while (refcnt != 1) { 10507 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { 10508 rtnl_lock(); 10509 10510 /* Rebroadcast unregister notification */ 10511 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10512 10513 __rtnl_unlock(); 10514 rcu_barrier(); 10515 rtnl_lock(); 10516 10517 if (test_bit(__LINK_STATE_LINKWATCH_PENDING, 10518 &dev->state)) { 10519 /* We must not have linkwatch events 10520 * pending on unregister. If this 10521 * happens, we simply run the queue 10522 * unscheduled, resulting in a noop 10523 * for this device. 10524 */ 10525 linkwatch_run_queue(); 10526 } 10527 10528 __rtnl_unlock(); 10529 10530 rebroadcast_time = jiffies; 10531 } 10532 10533 if (!wait) { 10534 rcu_barrier(); 10535 wait = WAIT_REFS_MIN_MSECS; 10536 } else { 10537 msleep(wait); 10538 wait = min(wait << 1, WAIT_REFS_MAX_MSECS); 10539 } 10540 10541 refcnt = netdev_refcnt_read(dev); 10542 10543 if (refcnt != 1 && 10544 time_after(jiffies, warning_time + 10545 netdev_unregister_timeout_secs * HZ)) { 10546 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", 10547 dev->name, refcnt); 10548 warning_time = jiffies; 10549 } 10550 } 10551 } 10552 10553 /* The sequence is: 10554 * 10555 * rtnl_lock(); 10556 * ... 10557 * register_netdevice(x1); 10558 * register_netdevice(x2); 10559 * ... 10560 * unregister_netdevice(y1); 10561 * unregister_netdevice(y2); 10562 * ... 10563 * rtnl_unlock(); 10564 * free_netdev(y1); 10565 * free_netdev(y2); 10566 * 10567 * We are invoked by rtnl_unlock(). 10568 * This allows us to deal with problems: 10569 * 1) We can delete sysfs objects which invoke hotplug 10570 * without deadlocking with linkwatch via keventd. 10571 * 2) Since we run with the RTNL semaphore not held, we can sleep 10572 * safely in order to wait for the netdev refcnt to drop to zero. 10573 * 10574 * We must not return until all unregister events added during 10575 * the interval the lock was held have been completed. 10576 */ 10577 void netdev_run_todo(void) 10578 { 10579 struct list_head list; 10580 #ifdef CONFIG_LOCKDEP 10581 struct list_head unlink_list; 10582 10583 list_replace_init(&net_unlink_list, &unlink_list); 10584 10585 while (!list_empty(&unlink_list)) { 10586 struct net_device *dev = list_first_entry(&unlink_list, 10587 struct net_device, 10588 unlink_list); 10589 list_del_init(&dev->unlink_list); 10590 dev->nested_level = dev->lower_level - 1; 10591 } 10592 #endif 10593 10594 /* Snapshot list, allow later requests */ 10595 list_replace_init(&net_todo_list, &list); 10596 10597 __rtnl_unlock(); 10598 10599 10600 /* Wait for rcu callbacks to finish before next phase */ 10601 if (!list_empty(&list)) 10602 rcu_barrier(); 10603 10604 while (!list_empty(&list)) { 10605 struct net_device *dev 10606 = list_first_entry(&list, struct net_device, todo_list); 10607 list_del(&dev->todo_list); 10608 10609 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { 10610 pr_err("network todo '%s' but state %d\n", 10611 dev->name, dev->reg_state); 10612 dump_stack(); 10613 continue; 10614 } 10615 10616 dev->reg_state = NETREG_UNREGISTERED; 10617 10618 netdev_wait_allrefs(dev); 10619 10620 /* paranoia */ 10621 BUG_ON(netdev_refcnt_read(dev) != 1); 10622 BUG_ON(!list_empty(&dev->ptype_all)); 10623 BUG_ON(!list_empty(&dev->ptype_specific)); 10624 WARN_ON(rcu_access_pointer(dev->ip_ptr)); 10625 WARN_ON(rcu_access_pointer(dev->ip6_ptr)); 10626 #if IS_ENABLED(CONFIG_DECNET) 10627 WARN_ON(dev->dn_ptr); 10628 #endif 10629 if (dev->priv_destructor) 10630 dev->priv_destructor(dev); 10631 if (dev->needs_free_netdev) 10632 free_netdev(dev); 10633 10634 /* Report a network device has been unregistered */ 10635 rtnl_lock(); 10636 dev_net(dev)->dev_unreg_count--; 10637 __rtnl_unlock(); 10638 wake_up(&netdev_unregistering_wq); 10639 10640 /* Free network device */ 10641 kobject_put(&dev->dev.kobj); 10642 } 10643 } 10644 10645 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has 10646 * all the same fields in the same order as net_device_stats, with only 10647 * the type differing, but rtnl_link_stats64 may have additional fields 10648 * at the end for newer counters. 10649 */ 10650 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, 10651 const struct net_device_stats *netdev_stats) 10652 { 10653 #if BITS_PER_LONG == 64 10654 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats)); 10655 memcpy(stats64, netdev_stats, sizeof(*netdev_stats)); 10656 /* zero out counters that only exist in rtnl_link_stats64 */ 10657 memset((char *)stats64 + sizeof(*netdev_stats), 0, 10658 sizeof(*stats64) - sizeof(*netdev_stats)); 10659 #else 10660 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long); 10661 const unsigned long *src = (const unsigned long *)netdev_stats; 10662 u64 *dst = (u64 *)stats64; 10663 10664 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); 10665 for (i = 0; i < n; i++) 10666 dst[i] = src[i]; 10667 /* zero out counters that only exist in rtnl_link_stats64 */ 10668 memset((char *)stats64 + n * sizeof(u64), 0, 10669 sizeof(*stats64) - n * sizeof(u64)); 10670 #endif 10671 } 10672 EXPORT_SYMBOL(netdev_stats_to_stats64); 10673 10674 /** 10675 * dev_get_stats - get network device statistics 10676 * @dev: device to get statistics from 10677 * @storage: place to store stats 10678 * 10679 * Get network statistics from device. Return @storage. 10680 * The device driver may provide its own method by setting 10681 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; 10682 * otherwise the internal statistics structure is used. 10683 */ 10684 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, 10685 struct rtnl_link_stats64 *storage) 10686 { 10687 const struct net_device_ops *ops = dev->netdev_ops; 10688 10689 if (ops->ndo_get_stats64) { 10690 memset(storage, 0, sizeof(*storage)); 10691 ops->ndo_get_stats64(dev, storage); 10692 } else if (ops->ndo_get_stats) { 10693 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); 10694 } else { 10695 netdev_stats_to_stats64(storage, &dev->stats); 10696 } 10697 storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped); 10698 storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped); 10699 storage->rx_nohandler += (unsigned long)atomic_long_read(&dev->rx_nohandler); 10700 return storage; 10701 } 10702 EXPORT_SYMBOL(dev_get_stats); 10703 10704 /** 10705 * dev_fetch_sw_netstats - get per-cpu network device statistics 10706 * @s: place to store stats 10707 * @netstats: per-cpu network stats to read from 10708 * 10709 * Read per-cpu network statistics and populate the related fields in @s. 10710 */ 10711 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s, 10712 const struct pcpu_sw_netstats __percpu *netstats) 10713 { 10714 int cpu; 10715 10716 for_each_possible_cpu(cpu) { 10717 const struct pcpu_sw_netstats *stats; 10718 struct pcpu_sw_netstats tmp; 10719 unsigned int start; 10720 10721 stats = per_cpu_ptr(netstats, cpu); 10722 do { 10723 start = u64_stats_fetch_begin_irq(&stats->syncp); 10724 tmp.rx_packets = stats->rx_packets; 10725 tmp.rx_bytes = stats->rx_bytes; 10726 tmp.tx_packets = stats->tx_packets; 10727 tmp.tx_bytes = stats->tx_bytes; 10728 } while (u64_stats_fetch_retry_irq(&stats->syncp, start)); 10729 10730 s->rx_packets += tmp.rx_packets; 10731 s->rx_bytes += tmp.rx_bytes; 10732 s->tx_packets += tmp.tx_packets; 10733 s->tx_bytes += tmp.tx_bytes; 10734 } 10735 } 10736 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats); 10737 10738 /** 10739 * dev_get_tstats64 - ndo_get_stats64 implementation 10740 * @dev: device to get statistics from 10741 * @s: place to store stats 10742 * 10743 * Populate @s from dev->stats and dev->tstats. Can be used as 10744 * ndo_get_stats64() callback. 10745 */ 10746 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s) 10747 { 10748 netdev_stats_to_stats64(s, &dev->stats); 10749 dev_fetch_sw_netstats(s, dev->tstats); 10750 } 10751 EXPORT_SYMBOL_GPL(dev_get_tstats64); 10752 10753 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) 10754 { 10755 struct netdev_queue *queue = dev_ingress_queue(dev); 10756 10757 #ifdef CONFIG_NET_CLS_ACT 10758 if (queue) 10759 return queue; 10760 queue = kzalloc(sizeof(*queue), GFP_KERNEL); 10761 if (!queue) 10762 return NULL; 10763 netdev_init_one_queue(dev, queue, NULL); 10764 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); 10765 queue->qdisc_sleeping = &noop_qdisc; 10766 rcu_assign_pointer(dev->ingress_queue, queue); 10767 #endif 10768 return queue; 10769 } 10770 10771 static const struct ethtool_ops default_ethtool_ops; 10772 10773 void netdev_set_default_ethtool_ops(struct net_device *dev, 10774 const struct ethtool_ops *ops) 10775 { 10776 if (dev->ethtool_ops == &default_ethtool_ops) 10777 dev->ethtool_ops = ops; 10778 } 10779 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); 10780 10781 void netdev_freemem(struct net_device *dev) 10782 { 10783 char *addr = (char *)dev - dev->padded; 10784 10785 kvfree(addr); 10786 } 10787 10788 /** 10789 * alloc_netdev_mqs - allocate network device 10790 * @sizeof_priv: size of private data to allocate space for 10791 * @name: device name format string 10792 * @name_assign_type: origin of device name 10793 * @setup: callback to initialize device 10794 * @txqs: the number of TX subqueues to allocate 10795 * @rxqs: the number of RX subqueues to allocate 10796 * 10797 * Allocates a struct net_device with private data area for driver use 10798 * and performs basic initialization. Also allocates subqueue structs 10799 * for each queue on the device. 10800 */ 10801 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, 10802 unsigned char name_assign_type, 10803 void (*setup)(struct net_device *), 10804 unsigned int txqs, unsigned int rxqs) 10805 { 10806 struct net_device *dev; 10807 unsigned int alloc_size; 10808 struct net_device *p; 10809 10810 BUG_ON(strlen(name) >= sizeof(dev->name)); 10811 10812 if (txqs < 1) { 10813 pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); 10814 return NULL; 10815 } 10816 10817 if (rxqs < 1) { 10818 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); 10819 return NULL; 10820 } 10821 10822 alloc_size = sizeof(struct net_device); 10823 if (sizeof_priv) { 10824 /* ensure 32-byte alignment of private area */ 10825 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); 10826 alloc_size += sizeof_priv; 10827 } 10828 /* ensure 32-byte alignment of whole construct */ 10829 alloc_size += NETDEV_ALIGN - 1; 10830 10831 p = kvzalloc(alloc_size, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10832 if (!p) 10833 return NULL; 10834 10835 dev = PTR_ALIGN(p, NETDEV_ALIGN); 10836 dev->padded = (char *)dev - (char *)p; 10837 10838 #ifdef CONFIG_PCPU_DEV_REFCNT 10839 dev->pcpu_refcnt = alloc_percpu(int); 10840 if (!dev->pcpu_refcnt) 10841 goto free_dev; 10842 dev_hold(dev); 10843 #else 10844 refcount_set(&dev->dev_refcnt, 1); 10845 #endif 10846 10847 if (dev_addr_init(dev)) 10848 goto free_pcpu; 10849 10850 dev_mc_init(dev); 10851 dev_uc_init(dev); 10852 10853 dev_net_set(dev, &init_net); 10854 10855 dev->gso_max_size = GSO_MAX_SIZE; 10856 dev->gso_max_segs = GSO_MAX_SEGS; 10857 dev->upper_level = 1; 10858 dev->lower_level = 1; 10859 #ifdef CONFIG_LOCKDEP 10860 dev->nested_level = 0; 10861 INIT_LIST_HEAD(&dev->unlink_list); 10862 #endif 10863 10864 INIT_LIST_HEAD(&dev->napi_list); 10865 INIT_LIST_HEAD(&dev->unreg_list); 10866 INIT_LIST_HEAD(&dev->close_list); 10867 INIT_LIST_HEAD(&dev->link_watch_list); 10868 INIT_LIST_HEAD(&dev->adj_list.upper); 10869 INIT_LIST_HEAD(&dev->adj_list.lower); 10870 INIT_LIST_HEAD(&dev->ptype_all); 10871 INIT_LIST_HEAD(&dev->ptype_specific); 10872 INIT_LIST_HEAD(&dev->net_notifier_list); 10873 #ifdef CONFIG_NET_SCHED 10874 hash_init(dev->qdisc_hash); 10875 #endif 10876 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; 10877 setup(dev); 10878 10879 if (!dev->tx_queue_len) { 10880 dev->priv_flags |= IFF_NO_QUEUE; 10881 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; 10882 } 10883 10884 dev->num_tx_queues = txqs; 10885 dev->real_num_tx_queues = txqs; 10886 if (netif_alloc_netdev_queues(dev)) 10887 goto free_all; 10888 10889 dev->num_rx_queues = rxqs; 10890 dev->real_num_rx_queues = rxqs; 10891 if (netif_alloc_rx_queues(dev)) 10892 goto free_all; 10893 10894 strcpy(dev->name, name); 10895 dev->name_assign_type = name_assign_type; 10896 dev->group = INIT_NETDEV_GROUP; 10897 if (!dev->ethtool_ops) 10898 dev->ethtool_ops = &default_ethtool_ops; 10899 10900 nf_hook_netdev_init(dev); 10901 10902 return dev; 10903 10904 free_all: 10905 free_netdev(dev); 10906 return NULL; 10907 10908 free_pcpu: 10909 #ifdef CONFIG_PCPU_DEV_REFCNT 10910 free_percpu(dev->pcpu_refcnt); 10911 free_dev: 10912 #endif 10913 netdev_freemem(dev); 10914 return NULL; 10915 } 10916 EXPORT_SYMBOL(alloc_netdev_mqs); 10917 10918 /** 10919 * free_netdev - free network device 10920 * @dev: device 10921 * 10922 * This function does the last stage of destroying an allocated device 10923 * interface. The reference to the device object is released. If this 10924 * is the last reference then it will be freed.Must be called in process 10925 * context. 10926 */ 10927 void free_netdev(struct net_device *dev) 10928 { 10929 struct napi_struct *p, *n; 10930 10931 might_sleep(); 10932 10933 /* When called immediately after register_netdevice() failed the unwind 10934 * handling may still be dismantling the device. Handle that case by 10935 * deferring the free. 10936 */ 10937 if (dev->reg_state == NETREG_UNREGISTERING) { 10938 ASSERT_RTNL(); 10939 dev->needs_free_netdev = true; 10940 return; 10941 } 10942 10943 netif_free_tx_queues(dev); 10944 netif_free_rx_queues(dev); 10945 10946 kfree(rcu_dereference_protected(dev->ingress_queue, 1)); 10947 10948 /* Flush device addresses */ 10949 dev_addr_flush(dev); 10950 10951 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) 10952 netif_napi_del(p); 10953 10954 #ifdef CONFIG_PCPU_DEV_REFCNT 10955 free_percpu(dev->pcpu_refcnt); 10956 dev->pcpu_refcnt = NULL; 10957 #endif 10958 free_percpu(dev->xdp_bulkq); 10959 dev->xdp_bulkq = NULL; 10960 10961 /* Compatibility with error handling in drivers */ 10962 if (dev->reg_state == NETREG_UNINITIALIZED) { 10963 netdev_freemem(dev); 10964 return; 10965 } 10966 10967 BUG_ON(dev->reg_state != NETREG_UNREGISTERED); 10968 dev->reg_state = NETREG_RELEASED; 10969 10970 /* will free via device release */ 10971 put_device(&dev->dev); 10972 } 10973 EXPORT_SYMBOL(free_netdev); 10974 10975 /** 10976 * synchronize_net - Synchronize with packet receive processing 10977 * 10978 * Wait for packets currently being received to be done. 10979 * Does not block later packets from starting. 10980 */ 10981 void synchronize_net(void) 10982 { 10983 might_sleep(); 10984 if (rtnl_is_locked()) 10985 synchronize_rcu_expedited(); 10986 else 10987 synchronize_rcu(); 10988 } 10989 EXPORT_SYMBOL(synchronize_net); 10990 10991 /** 10992 * unregister_netdevice_queue - remove device from the kernel 10993 * @dev: device 10994 * @head: list 10995 * 10996 * This function shuts down a device interface and removes it 10997 * from the kernel tables. 10998 * If head not NULL, device is queued to be unregistered later. 10999 * 11000 * Callers must hold the rtnl semaphore. You may want 11001 * unregister_netdev() instead of this. 11002 */ 11003 11004 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) 11005 { 11006 ASSERT_RTNL(); 11007 11008 if (head) { 11009 list_move_tail(&dev->unreg_list, head); 11010 } else { 11011 LIST_HEAD(single); 11012 11013 list_add(&dev->unreg_list, &single); 11014 unregister_netdevice_many(&single); 11015 } 11016 } 11017 EXPORT_SYMBOL(unregister_netdevice_queue); 11018 11019 /** 11020 * unregister_netdevice_many - unregister many devices 11021 * @head: list of devices 11022 * 11023 * Note: As most callers use a stack allocated list_head, 11024 * we force a list_del() to make sure stack wont be corrupted later. 11025 */ 11026 void unregister_netdevice_many(struct list_head *head) 11027 { 11028 struct net_device *dev, *tmp; 11029 LIST_HEAD(close_head); 11030 11031 BUG_ON(dev_boot_phase); 11032 ASSERT_RTNL(); 11033 11034 if (list_empty(head)) 11035 return; 11036 11037 list_for_each_entry_safe(dev, tmp, head, unreg_list) { 11038 /* Some devices call without registering 11039 * for initialization unwind. Remove those 11040 * devices and proceed with the remaining. 11041 */ 11042 if (dev->reg_state == NETREG_UNINITIALIZED) { 11043 pr_debug("unregister_netdevice: device %s/%p never was registered\n", 11044 dev->name, dev); 11045 11046 WARN_ON(1); 11047 list_del(&dev->unreg_list); 11048 continue; 11049 } 11050 dev->dismantle = true; 11051 BUG_ON(dev->reg_state != NETREG_REGISTERED); 11052 } 11053 11054 /* If device is running, close it first. */ 11055 list_for_each_entry(dev, head, unreg_list) 11056 list_add_tail(&dev->close_list, &close_head); 11057 dev_close_many(&close_head, true); 11058 11059 list_for_each_entry(dev, head, unreg_list) { 11060 /* And unlink it from device chain. */ 11061 unlist_netdevice(dev); 11062 11063 dev->reg_state = NETREG_UNREGISTERING; 11064 } 11065 flush_all_backlogs(); 11066 11067 synchronize_net(); 11068 11069 list_for_each_entry(dev, head, unreg_list) { 11070 struct sk_buff *skb = NULL; 11071 11072 /* Shutdown queueing discipline. */ 11073 dev_shutdown(dev); 11074 11075 dev_xdp_uninstall(dev); 11076 11077 /* Notify protocols, that we are about to destroy 11078 * this device. They should clean all the things. 11079 */ 11080 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 11081 11082 if (!dev->rtnl_link_ops || 11083 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 11084 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0, 11085 GFP_KERNEL, NULL, 0); 11086 11087 /* 11088 * Flush the unicast and multicast chains 11089 */ 11090 dev_uc_flush(dev); 11091 dev_mc_flush(dev); 11092 11093 netdev_name_node_alt_flush(dev); 11094 netdev_name_node_free(dev->name_node); 11095 11096 if (dev->netdev_ops->ndo_uninit) 11097 dev->netdev_ops->ndo_uninit(dev); 11098 11099 if (skb) 11100 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL); 11101 11102 /* Notifier chain MUST detach us all upper devices. */ 11103 WARN_ON(netdev_has_any_upper_dev(dev)); 11104 WARN_ON(netdev_has_any_lower_dev(dev)); 11105 11106 /* Remove entries from kobject tree */ 11107 netdev_unregister_kobject(dev); 11108 #ifdef CONFIG_XPS 11109 /* Remove XPS queueing entries */ 11110 netif_reset_xps_queues_gt(dev, 0); 11111 #endif 11112 } 11113 11114 synchronize_net(); 11115 11116 list_for_each_entry(dev, head, unreg_list) { 11117 dev_put(dev); 11118 net_set_todo(dev); 11119 } 11120 11121 list_del(head); 11122 } 11123 EXPORT_SYMBOL(unregister_netdevice_many); 11124 11125 /** 11126 * unregister_netdev - remove device from the kernel 11127 * @dev: device 11128 * 11129 * This function shuts down a device interface and removes it 11130 * from the kernel tables. 11131 * 11132 * This is just a wrapper for unregister_netdevice that takes 11133 * the rtnl semaphore. In general you want to use this and not 11134 * unregister_netdevice. 11135 */ 11136 void unregister_netdev(struct net_device *dev) 11137 { 11138 rtnl_lock(); 11139 unregister_netdevice(dev); 11140 rtnl_unlock(); 11141 } 11142 EXPORT_SYMBOL(unregister_netdev); 11143 11144 /** 11145 * __dev_change_net_namespace - move device to different nethost namespace 11146 * @dev: device 11147 * @net: network namespace 11148 * @pat: If not NULL name pattern to try if the current device name 11149 * is already taken in the destination network namespace. 11150 * @new_ifindex: If not zero, specifies device index in the target 11151 * namespace. 11152 * 11153 * This function shuts down a device interface and moves it 11154 * to a new network namespace. On success 0 is returned, on 11155 * a failure a netagive errno code is returned. 11156 * 11157 * Callers must hold the rtnl semaphore. 11158 */ 11159 11160 int __dev_change_net_namespace(struct net_device *dev, struct net *net, 11161 const char *pat, int new_ifindex) 11162 { 11163 struct net *net_old = dev_net(dev); 11164 int err, new_nsid; 11165 11166 ASSERT_RTNL(); 11167 11168 /* Don't allow namespace local devices to be moved. */ 11169 err = -EINVAL; 11170 if (dev->features & NETIF_F_NETNS_LOCAL) 11171 goto out; 11172 11173 /* Ensure the device has been registrered */ 11174 if (dev->reg_state != NETREG_REGISTERED) 11175 goto out; 11176 11177 /* Get out if there is nothing todo */ 11178 err = 0; 11179 if (net_eq(net_old, net)) 11180 goto out; 11181 11182 /* Pick the destination device name, and ensure 11183 * we can use it in the destination network namespace. 11184 */ 11185 err = -EEXIST; 11186 if (netdev_name_in_use(net, dev->name)) { 11187 /* We get here if we can't use the current device name */ 11188 if (!pat) 11189 goto out; 11190 err = dev_get_valid_name(net, dev, pat); 11191 if (err < 0) 11192 goto out; 11193 } 11194 11195 /* Check that new_ifindex isn't used yet. */ 11196 err = -EBUSY; 11197 if (new_ifindex && __dev_get_by_index(net, new_ifindex)) 11198 goto out; 11199 11200 /* 11201 * And now a mini version of register_netdevice unregister_netdevice. 11202 */ 11203 11204 /* If device is running close it first. */ 11205 dev_close(dev); 11206 11207 /* And unlink it from device chain */ 11208 unlist_netdevice(dev); 11209 11210 synchronize_net(); 11211 11212 /* Shutdown queueing discipline. */ 11213 dev_shutdown(dev); 11214 11215 /* Notify protocols, that we are about to destroy 11216 * this device. They should clean all the things. 11217 * 11218 * Note that dev->reg_state stays at NETREG_REGISTERED. 11219 * This is wanted because this way 8021q and macvlan know 11220 * the device is just moving and can keep their slaves up. 11221 */ 11222 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 11223 rcu_barrier(); 11224 11225 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL); 11226 /* If there is an ifindex conflict assign a new one */ 11227 if (!new_ifindex) { 11228 if (__dev_get_by_index(net, dev->ifindex)) 11229 new_ifindex = dev_new_index(net); 11230 else 11231 new_ifindex = dev->ifindex; 11232 } 11233 11234 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid, 11235 new_ifindex); 11236 11237 /* 11238 * Flush the unicast and multicast chains 11239 */ 11240 dev_uc_flush(dev); 11241 dev_mc_flush(dev); 11242 11243 /* Send a netdev-removed uevent to the old namespace */ 11244 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); 11245 netdev_adjacent_del_links(dev); 11246 11247 /* Move per-net netdevice notifiers that are following the netdevice */ 11248 move_netdevice_notifiers_dev_net(dev, net); 11249 11250 /* Actually switch the network namespace */ 11251 dev_net_set(dev, net); 11252 dev->ifindex = new_ifindex; 11253 11254 /* Send a netdev-add uevent to the new namespace */ 11255 kobject_uevent(&dev->dev.kobj, KOBJ_ADD); 11256 netdev_adjacent_add_links(dev); 11257 11258 /* Fixup kobjects */ 11259 err = device_rename(&dev->dev, dev->name); 11260 WARN_ON(err); 11261 11262 /* Adapt owner in case owning user namespace of target network 11263 * namespace is different from the original one. 11264 */ 11265 err = netdev_change_owner(dev, net_old, net); 11266 WARN_ON(err); 11267 11268 /* Add the device back in the hashes */ 11269 list_netdevice(dev); 11270 11271 /* Notify protocols, that a new device appeared. */ 11272 call_netdevice_notifiers(NETDEV_REGISTER, dev); 11273 11274 /* 11275 * Prevent userspace races by waiting until the network 11276 * device is fully setup before sending notifications. 11277 */ 11278 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); 11279 11280 synchronize_net(); 11281 err = 0; 11282 out: 11283 return err; 11284 } 11285 EXPORT_SYMBOL_GPL(__dev_change_net_namespace); 11286 11287 static int dev_cpu_dead(unsigned int oldcpu) 11288 { 11289 struct sk_buff **list_skb; 11290 struct sk_buff *skb; 11291 unsigned int cpu; 11292 struct softnet_data *sd, *oldsd, *remsd = NULL; 11293 11294 local_irq_disable(); 11295 cpu = smp_processor_id(); 11296 sd = &per_cpu(softnet_data, cpu); 11297 oldsd = &per_cpu(softnet_data, oldcpu); 11298 11299 /* Find end of our completion_queue. */ 11300 list_skb = &sd->completion_queue; 11301 while (*list_skb) 11302 list_skb = &(*list_skb)->next; 11303 /* Append completion queue from offline CPU. */ 11304 *list_skb = oldsd->completion_queue; 11305 oldsd->completion_queue = NULL; 11306 11307 /* Append output queue from offline CPU. */ 11308 if (oldsd->output_queue) { 11309 *sd->output_queue_tailp = oldsd->output_queue; 11310 sd->output_queue_tailp = oldsd->output_queue_tailp; 11311 oldsd->output_queue = NULL; 11312 oldsd->output_queue_tailp = &oldsd->output_queue; 11313 } 11314 /* Append NAPI poll list from offline CPU, with one exception : 11315 * process_backlog() must be called by cpu owning percpu backlog. 11316 * We properly handle process_queue & input_pkt_queue later. 11317 */ 11318 while (!list_empty(&oldsd->poll_list)) { 11319 struct napi_struct *napi = list_first_entry(&oldsd->poll_list, 11320 struct napi_struct, 11321 poll_list); 11322 11323 list_del_init(&napi->poll_list); 11324 if (napi->poll == process_backlog) 11325 napi->state = 0; 11326 else 11327 ____napi_schedule(sd, napi); 11328 } 11329 11330 raise_softirq_irqoff(NET_TX_SOFTIRQ); 11331 local_irq_enable(); 11332 11333 #ifdef CONFIG_RPS 11334 remsd = oldsd->rps_ipi_list; 11335 oldsd->rps_ipi_list = NULL; 11336 #endif 11337 /* send out pending IPI's on offline CPU */ 11338 net_rps_send_ipi(remsd); 11339 11340 /* Process offline CPU's input_pkt_queue */ 11341 while ((skb = __skb_dequeue(&oldsd->process_queue))) { 11342 netif_rx_ni(skb); 11343 input_queue_head_incr(oldsd); 11344 } 11345 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { 11346 netif_rx_ni(skb); 11347 input_queue_head_incr(oldsd); 11348 } 11349 11350 return 0; 11351 } 11352 11353 /** 11354 * netdev_increment_features - increment feature set by one 11355 * @all: current feature set 11356 * @one: new feature set 11357 * @mask: mask feature set 11358 * 11359 * Computes a new feature set after adding a device with feature set 11360 * @one to the master device with current feature set @all. Will not 11361 * enable anything that is off in @mask. Returns the new feature set. 11362 */ 11363 netdev_features_t netdev_increment_features(netdev_features_t all, 11364 netdev_features_t one, netdev_features_t mask) 11365 { 11366 if (mask & NETIF_F_HW_CSUM) 11367 mask |= NETIF_F_CSUM_MASK; 11368 mask |= NETIF_F_VLAN_CHALLENGED; 11369 11370 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; 11371 all &= one | ~NETIF_F_ALL_FOR_ALL; 11372 11373 /* If one device supports hw checksumming, set for all. */ 11374 if (all & NETIF_F_HW_CSUM) 11375 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); 11376 11377 return all; 11378 } 11379 EXPORT_SYMBOL(netdev_increment_features); 11380 11381 static struct hlist_head * __net_init netdev_create_hash(void) 11382 { 11383 int i; 11384 struct hlist_head *hash; 11385 11386 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL); 11387 if (hash != NULL) 11388 for (i = 0; i < NETDEV_HASHENTRIES; i++) 11389 INIT_HLIST_HEAD(&hash[i]); 11390 11391 return hash; 11392 } 11393 11394 /* Initialize per network namespace state */ 11395 static int __net_init netdev_init(struct net *net) 11396 { 11397 BUILD_BUG_ON(GRO_HASH_BUCKETS > 11398 8 * sizeof_field(struct napi_struct, gro_bitmask)); 11399 11400 if (net != &init_net) 11401 INIT_LIST_HEAD(&net->dev_base_head); 11402 11403 net->dev_name_head = netdev_create_hash(); 11404 if (net->dev_name_head == NULL) 11405 goto err_name; 11406 11407 net->dev_index_head = netdev_create_hash(); 11408 if (net->dev_index_head == NULL) 11409 goto err_idx; 11410 11411 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain); 11412 11413 return 0; 11414 11415 err_idx: 11416 kfree(net->dev_name_head); 11417 err_name: 11418 return -ENOMEM; 11419 } 11420 11421 /** 11422 * netdev_drivername - network driver for the device 11423 * @dev: network device 11424 * 11425 * Determine network driver for device. 11426 */ 11427 const char *netdev_drivername(const struct net_device *dev) 11428 { 11429 const struct device_driver *driver; 11430 const struct device *parent; 11431 const char *empty = ""; 11432 11433 parent = dev->dev.parent; 11434 if (!parent) 11435 return empty; 11436 11437 driver = parent->driver; 11438 if (driver && driver->name) 11439 return driver->name; 11440 return empty; 11441 } 11442 11443 static void __netdev_printk(const char *level, const struct net_device *dev, 11444 struct va_format *vaf) 11445 { 11446 if (dev && dev->dev.parent) { 11447 dev_printk_emit(level[1] - '0', 11448 dev->dev.parent, 11449 "%s %s %s%s: %pV", 11450 dev_driver_string(dev->dev.parent), 11451 dev_name(dev->dev.parent), 11452 netdev_name(dev), netdev_reg_state(dev), 11453 vaf); 11454 } else if (dev) { 11455 printk("%s%s%s: %pV", 11456 level, netdev_name(dev), netdev_reg_state(dev), vaf); 11457 } else { 11458 printk("%s(NULL net_device): %pV", level, vaf); 11459 } 11460 } 11461 11462 void netdev_printk(const char *level, const struct net_device *dev, 11463 const char *format, ...) 11464 { 11465 struct va_format vaf; 11466 va_list args; 11467 11468 va_start(args, format); 11469 11470 vaf.fmt = format; 11471 vaf.va = &args; 11472 11473 __netdev_printk(level, dev, &vaf); 11474 11475 va_end(args); 11476 } 11477 EXPORT_SYMBOL(netdev_printk); 11478 11479 #define define_netdev_printk_level(func, level) \ 11480 void func(const struct net_device *dev, const char *fmt, ...) \ 11481 { \ 11482 struct va_format vaf; \ 11483 va_list args; \ 11484 \ 11485 va_start(args, fmt); \ 11486 \ 11487 vaf.fmt = fmt; \ 11488 vaf.va = &args; \ 11489 \ 11490 __netdev_printk(level, dev, &vaf); \ 11491 \ 11492 va_end(args); \ 11493 } \ 11494 EXPORT_SYMBOL(func); 11495 11496 define_netdev_printk_level(netdev_emerg, KERN_EMERG); 11497 define_netdev_printk_level(netdev_alert, KERN_ALERT); 11498 define_netdev_printk_level(netdev_crit, KERN_CRIT); 11499 define_netdev_printk_level(netdev_err, KERN_ERR); 11500 define_netdev_printk_level(netdev_warn, KERN_WARNING); 11501 define_netdev_printk_level(netdev_notice, KERN_NOTICE); 11502 define_netdev_printk_level(netdev_info, KERN_INFO); 11503 11504 static void __net_exit netdev_exit(struct net *net) 11505 { 11506 kfree(net->dev_name_head); 11507 kfree(net->dev_index_head); 11508 if (net != &init_net) 11509 WARN_ON_ONCE(!list_empty(&net->dev_base_head)); 11510 } 11511 11512 static struct pernet_operations __net_initdata netdev_net_ops = { 11513 .init = netdev_init, 11514 .exit = netdev_exit, 11515 }; 11516 11517 static void __net_exit default_device_exit(struct net *net) 11518 { 11519 struct net_device *dev, *aux; 11520 /* 11521 * Push all migratable network devices back to the 11522 * initial network namespace 11523 */ 11524 rtnl_lock(); 11525 for_each_netdev_safe(net, dev, aux) { 11526 int err; 11527 char fb_name[IFNAMSIZ]; 11528 11529 /* Ignore unmoveable devices (i.e. loopback) */ 11530 if (dev->features & NETIF_F_NETNS_LOCAL) 11531 continue; 11532 11533 /* Leave virtual devices for the generic cleanup */ 11534 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund) 11535 continue; 11536 11537 /* Push remaining network devices to init_net */ 11538 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); 11539 if (netdev_name_in_use(&init_net, fb_name)) 11540 snprintf(fb_name, IFNAMSIZ, "dev%%d"); 11541 err = dev_change_net_namespace(dev, &init_net, fb_name); 11542 if (err) { 11543 pr_emerg("%s: failed to move %s to init_net: %d\n", 11544 __func__, dev->name, err); 11545 BUG(); 11546 } 11547 } 11548 rtnl_unlock(); 11549 } 11550 11551 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list) 11552 { 11553 /* Return with the rtnl_lock held when there are no network 11554 * devices unregistering in any network namespace in net_list. 11555 */ 11556 struct net *net; 11557 bool unregistering; 11558 DEFINE_WAIT_FUNC(wait, woken_wake_function); 11559 11560 add_wait_queue(&netdev_unregistering_wq, &wait); 11561 for (;;) { 11562 unregistering = false; 11563 rtnl_lock(); 11564 list_for_each_entry(net, net_list, exit_list) { 11565 if (net->dev_unreg_count > 0) { 11566 unregistering = true; 11567 break; 11568 } 11569 } 11570 if (!unregistering) 11571 break; 11572 __rtnl_unlock(); 11573 11574 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); 11575 } 11576 remove_wait_queue(&netdev_unregistering_wq, &wait); 11577 } 11578 11579 static void __net_exit default_device_exit_batch(struct list_head *net_list) 11580 { 11581 /* At exit all network devices most be removed from a network 11582 * namespace. Do this in the reverse order of registration. 11583 * Do this across as many network namespaces as possible to 11584 * improve batching efficiency. 11585 */ 11586 struct net_device *dev; 11587 struct net *net; 11588 LIST_HEAD(dev_kill_list); 11589 11590 /* To prevent network device cleanup code from dereferencing 11591 * loopback devices or network devices that have been freed 11592 * wait here for all pending unregistrations to complete, 11593 * before unregistring the loopback device and allowing the 11594 * network namespace be freed. 11595 * 11596 * The netdev todo list containing all network devices 11597 * unregistrations that happen in default_device_exit_batch 11598 * will run in the rtnl_unlock() at the end of 11599 * default_device_exit_batch. 11600 */ 11601 rtnl_lock_unregistering(net_list); 11602 list_for_each_entry(net, net_list, exit_list) { 11603 for_each_netdev_reverse(net, dev) { 11604 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) 11605 dev->rtnl_link_ops->dellink(dev, &dev_kill_list); 11606 else 11607 unregister_netdevice_queue(dev, &dev_kill_list); 11608 } 11609 } 11610 unregister_netdevice_many(&dev_kill_list); 11611 rtnl_unlock(); 11612 } 11613 11614 static struct pernet_operations __net_initdata default_device_ops = { 11615 .exit = default_device_exit, 11616 .exit_batch = default_device_exit_batch, 11617 }; 11618 11619 /* 11620 * Initialize the DEV module. At boot time this walks the device list and 11621 * unhooks any devices that fail to initialise (normally hardware not 11622 * present) and leaves us with a valid list of present and active devices. 11623 * 11624 */ 11625 11626 /* 11627 * This is called single threaded during boot, so no need 11628 * to take the rtnl semaphore. 11629 */ 11630 static int __init net_dev_init(void) 11631 { 11632 int i, rc = -ENOMEM; 11633 11634 BUG_ON(!dev_boot_phase); 11635 11636 if (dev_proc_init()) 11637 goto out; 11638 11639 if (netdev_kobject_init()) 11640 goto out; 11641 11642 INIT_LIST_HEAD(&ptype_all); 11643 for (i = 0; i < PTYPE_HASH_SIZE; i++) 11644 INIT_LIST_HEAD(&ptype_base[i]); 11645 11646 INIT_LIST_HEAD(&offload_base); 11647 11648 if (register_pernet_subsys(&netdev_net_ops)) 11649 goto out; 11650 11651 /* 11652 * Initialise the packet receive queues. 11653 */ 11654 11655 for_each_possible_cpu(i) { 11656 struct work_struct *flush = per_cpu_ptr(&flush_works, i); 11657 struct softnet_data *sd = &per_cpu(softnet_data, i); 11658 11659 INIT_WORK(flush, flush_backlog); 11660 11661 skb_queue_head_init(&sd->input_pkt_queue); 11662 skb_queue_head_init(&sd->process_queue); 11663 #ifdef CONFIG_XFRM_OFFLOAD 11664 skb_queue_head_init(&sd->xfrm_backlog); 11665 #endif 11666 INIT_LIST_HEAD(&sd->poll_list); 11667 sd->output_queue_tailp = &sd->output_queue; 11668 #ifdef CONFIG_RPS 11669 INIT_CSD(&sd->csd, rps_trigger_softirq, sd); 11670 sd->cpu = i; 11671 #endif 11672 11673 init_gro_hash(&sd->backlog); 11674 sd->backlog.poll = process_backlog; 11675 sd->backlog.weight = weight_p; 11676 } 11677 11678 dev_boot_phase = 0; 11679 11680 /* The loopback device is special if any other network devices 11681 * is present in a network namespace the loopback device must 11682 * be present. Since we now dynamically allocate and free the 11683 * loopback device ensure this invariant is maintained by 11684 * keeping the loopback device as the first device on the 11685 * list of network devices. Ensuring the loopback devices 11686 * is the first device that appears and the last network device 11687 * that disappears. 11688 */ 11689 if (register_pernet_device(&loopback_net_ops)) 11690 goto out; 11691 11692 if (register_pernet_device(&default_device_ops)) 11693 goto out; 11694 11695 open_softirq(NET_TX_SOFTIRQ, net_tx_action); 11696 open_softirq(NET_RX_SOFTIRQ, net_rx_action); 11697 11698 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", 11699 NULL, dev_cpu_dead); 11700 WARN_ON(rc < 0); 11701 rc = 0; 11702 out: 11703 return rc; 11704 } 11705 11706 subsys_initcall(net_dev_init); 11707