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 if (txq->xmit_lock_owner != cpu) { 4214 if (dev_xmit_recursion()) 4215 goto recursion_alert; 4216 4217 skb = validate_xmit_skb(skb, dev, &again); 4218 if (!skb) 4219 goto out; 4220 4221 PRANDOM_ADD_NOISE(skb, dev, txq, jiffies); 4222 HARD_TX_LOCK(dev, txq, cpu); 4223 4224 if (!netif_xmit_stopped(txq)) { 4225 dev_xmit_recursion_inc(); 4226 skb = dev_hard_start_xmit(skb, dev, txq, &rc); 4227 dev_xmit_recursion_dec(); 4228 if (dev_xmit_complete(rc)) { 4229 HARD_TX_UNLOCK(dev, txq); 4230 goto out; 4231 } 4232 } 4233 HARD_TX_UNLOCK(dev, txq); 4234 net_crit_ratelimited("Virtual device %s asks to queue packet!\n", 4235 dev->name); 4236 } else { 4237 /* Recursion is detected! It is possible, 4238 * unfortunately 4239 */ 4240 recursion_alert: 4241 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", 4242 dev->name); 4243 } 4244 } 4245 4246 rc = -ENETDOWN; 4247 rcu_read_unlock_bh(); 4248 4249 atomic_long_inc(&dev->tx_dropped); 4250 kfree_skb_list(skb); 4251 return rc; 4252 out: 4253 rcu_read_unlock_bh(); 4254 return rc; 4255 } 4256 4257 int dev_queue_xmit(struct sk_buff *skb) 4258 { 4259 return __dev_queue_xmit(skb, NULL); 4260 } 4261 EXPORT_SYMBOL(dev_queue_xmit); 4262 4263 int dev_queue_xmit_accel(struct sk_buff *skb, struct net_device *sb_dev) 4264 { 4265 return __dev_queue_xmit(skb, sb_dev); 4266 } 4267 EXPORT_SYMBOL(dev_queue_xmit_accel); 4268 4269 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id) 4270 { 4271 struct net_device *dev = skb->dev; 4272 struct sk_buff *orig_skb = skb; 4273 struct netdev_queue *txq; 4274 int ret = NETDEV_TX_BUSY; 4275 bool again = false; 4276 4277 if (unlikely(!netif_running(dev) || 4278 !netif_carrier_ok(dev))) 4279 goto drop; 4280 4281 skb = validate_xmit_skb_list(skb, dev, &again); 4282 if (skb != orig_skb) 4283 goto drop; 4284 4285 skb_set_queue_mapping(skb, queue_id); 4286 txq = skb_get_tx_queue(dev, skb); 4287 PRANDOM_ADD_NOISE(skb, dev, txq, jiffies); 4288 4289 local_bh_disable(); 4290 4291 dev_xmit_recursion_inc(); 4292 HARD_TX_LOCK(dev, txq, smp_processor_id()); 4293 if (!netif_xmit_frozen_or_drv_stopped(txq)) 4294 ret = netdev_start_xmit(skb, dev, txq, false); 4295 HARD_TX_UNLOCK(dev, txq); 4296 dev_xmit_recursion_dec(); 4297 4298 local_bh_enable(); 4299 return ret; 4300 drop: 4301 atomic_long_inc(&dev->tx_dropped); 4302 kfree_skb_list(skb); 4303 return NET_XMIT_DROP; 4304 } 4305 EXPORT_SYMBOL(__dev_direct_xmit); 4306 4307 /************************************************************************* 4308 * Receiver routines 4309 *************************************************************************/ 4310 4311 int netdev_max_backlog __read_mostly = 1000; 4312 EXPORT_SYMBOL(netdev_max_backlog); 4313 4314 int netdev_tstamp_prequeue __read_mostly = 1; 4315 int netdev_budget __read_mostly = 300; 4316 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */ 4317 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ; 4318 int weight_p __read_mostly = 64; /* old backlog weight */ 4319 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ 4320 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ 4321 int dev_rx_weight __read_mostly = 64; 4322 int dev_tx_weight __read_mostly = 64; 4323 /* Maximum number of GRO_NORMAL skbs to batch up for list-RX */ 4324 int gro_normal_batch __read_mostly = 8; 4325 4326 /* Called with irq disabled */ 4327 static inline void ____napi_schedule(struct softnet_data *sd, 4328 struct napi_struct *napi) 4329 { 4330 struct task_struct *thread; 4331 4332 if (test_bit(NAPI_STATE_THREADED, &napi->state)) { 4333 /* Paired with smp_mb__before_atomic() in 4334 * napi_enable()/dev_set_threaded(). 4335 * Use READ_ONCE() to guarantee a complete 4336 * read on napi->thread. Only call 4337 * wake_up_process() when it's not NULL. 4338 */ 4339 thread = READ_ONCE(napi->thread); 4340 if (thread) { 4341 /* Avoid doing set_bit() if the thread is in 4342 * INTERRUPTIBLE state, cause napi_thread_wait() 4343 * makes sure to proceed with napi polling 4344 * if the thread is explicitly woken from here. 4345 */ 4346 if (READ_ONCE(thread->__state) != TASK_INTERRUPTIBLE) 4347 set_bit(NAPI_STATE_SCHED_THREADED, &napi->state); 4348 wake_up_process(thread); 4349 return; 4350 } 4351 } 4352 4353 list_add_tail(&napi->poll_list, &sd->poll_list); 4354 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4355 } 4356 4357 #ifdef CONFIG_RPS 4358 4359 /* One global table that all flow-based protocols share. */ 4360 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; 4361 EXPORT_SYMBOL(rps_sock_flow_table); 4362 u32 rps_cpu_mask __read_mostly; 4363 EXPORT_SYMBOL(rps_cpu_mask); 4364 4365 struct static_key_false rps_needed __read_mostly; 4366 EXPORT_SYMBOL(rps_needed); 4367 struct static_key_false rfs_needed __read_mostly; 4368 EXPORT_SYMBOL(rfs_needed); 4369 4370 static struct rps_dev_flow * 4371 set_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4372 struct rps_dev_flow *rflow, u16 next_cpu) 4373 { 4374 if (next_cpu < nr_cpu_ids) { 4375 #ifdef CONFIG_RFS_ACCEL 4376 struct netdev_rx_queue *rxqueue; 4377 struct rps_dev_flow_table *flow_table; 4378 struct rps_dev_flow *old_rflow; 4379 u32 flow_id; 4380 u16 rxq_index; 4381 int rc; 4382 4383 /* Should we steer this flow to a different hardware queue? */ 4384 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || 4385 !(dev->features & NETIF_F_NTUPLE)) 4386 goto out; 4387 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); 4388 if (rxq_index == skb_get_rx_queue(skb)) 4389 goto out; 4390 4391 rxqueue = dev->_rx + rxq_index; 4392 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4393 if (!flow_table) 4394 goto out; 4395 flow_id = skb_get_hash(skb) & flow_table->mask; 4396 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, 4397 rxq_index, flow_id); 4398 if (rc < 0) 4399 goto out; 4400 old_rflow = rflow; 4401 rflow = &flow_table->flows[flow_id]; 4402 rflow->filter = rc; 4403 if (old_rflow->filter == rflow->filter) 4404 old_rflow->filter = RPS_NO_FILTER; 4405 out: 4406 #endif 4407 rflow->last_qtail = 4408 per_cpu(softnet_data, next_cpu).input_queue_head; 4409 } 4410 4411 rflow->cpu = next_cpu; 4412 return rflow; 4413 } 4414 4415 /* 4416 * get_rps_cpu is called from netif_receive_skb and returns the target 4417 * CPU from the RPS map of the receiving queue for a given skb. 4418 * rcu_read_lock must be held on entry. 4419 */ 4420 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4421 struct rps_dev_flow **rflowp) 4422 { 4423 const struct rps_sock_flow_table *sock_flow_table; 4424 struct netdev_rx_queue *rxqueue = dev->_rx; 4425 struct rps_dev_flow_table *flow_table; 4426 struct rps_map *map; 4427 int cpu = -1; 4428 u32 tcpu; 4429 u32 hash; 4430 4431 if (skb_rx_queue_recorded(skb)) { 4432 u16 index = skb_get_rx_queue(skb); 4433 4434 if (unlikely(index >= dev->real_num_rx_queues)) { 4435 WARN_ONCE(dev->real_num_rx_queues > 1, 4436 "%s received packet on queue %u, but number " 4437 "of RX queues is %u\n", 4438 dev->name, index, dev->real_num_rx_queues); 4439 goto done; 4440 } 4441 rxqueue += index; 4442 } 4443 4444 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ 4445 4446 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4447 map = rcu_dereference(rxqueue->rps_map); 4448 if (!flow_table && !map) 4449 goto done; 4450 4451 skb_reset_network_header(skb); 4452 hash = skb_get_hash(skb); 4453 if (!hash) 4454 goto done; 4455 4456 sock_flow_table = rcu_dereference(rps_sock_flow_table); 4457 if (flow_table && sock_flow_table) { 4458 struct rps_dev_flow *rflow; 4459 u32 next_cpu; 4460 u32 ident; 4461 4462 /* First check into global flow table if there is a match */ 4463 ident = sock_flow_table->ents[hash & sock_flow_table->mask]; 4464 if ((ident ^ hash) & ~rps_cpu_mask) 4465 goto try_rps; 4466 4467 next_cpu = ident & rps_cpu_mask; 4468 4469 /* OK, now we know there is a match, 4470 * we can look at the local (per receive queue) flow table 4471 */ 4472 rflow = &flow_table->flows[hash & flow_table->mask]; 4473 tcpu = rflow->cpu; 4474 4475 /* 4476 * If the desired CPU (where last recvmsg was done) is 4477 * different from current CPU (one in the rx-queue flow 4478 * table entry), switch if one of the following holds: 4479 * - Current CPU is unset (>= nr_cpu_ids). 4480 * - Current CPU is offline. 4481 * - The current CPU's queue tail has advanced beyond the 4482 * last packet that was enqueued using this table entry. 4483 * This guarantees that all previous packets for the flow 4484 * have been dequeued, thus preserving in order delivery. 4485 */ 4486 if (unlikely(tcpu != next_cpu) && 4487 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || 4488 ((int)(per_cpu(softnet_data, tcpu).input_queue_head - 4489 rflow->last_qtail)) >= 0)) { 4490 tcpu = next_cpu; 4491 rflow = set_rps_cpu(dev, skb, rflow, next_cpu); 4492 } 4493 4494 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { 4495 *rflowp = rflow; 4496 cpu = tcpu; 4497 goto done; 4498 } 4499 } 4500 4501 try_rps: 4502 4503 if (map) { 4504 tcpu = map->cpus[reciprocal_scale(hash, map->len)]; 4505 if (cpu_online(tcpu)) { 4506 cpu = tcpu; 4507 goto done; 4508 } 4509 } 4510 4511 done: 4512 return cpu; 4513 } 4514 4515 #ifdef CONFIG_RFS_ACCEL 4516 4517 /** 4518 * rps_may_expire_flow - check whether an RFS hardware filter may be removed 4519 * @dev: Device on which the filter was set 4520 * @rxq_index: RX queue index 4521 * @flow_id: Flow ID passed to ndo_rx_flow_steer() 4522 * @filter_id: Filter ID returned by ndo_rx_flow_steer() 4523 * 4524 * Drivers that implement ndo_rx_flow_steer() should periodically call 4525 * this function for each installed filter and remove the filters for 4526 * which it returns %true. 4527 */ 4528 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, 4529 u32 flow_id, u16 filter_id) 4530 { 4531 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; 4532 struct rps_dev_flow_table *flow_table; 4533 struct rps_dev_flow *rflow; 4534 bool expire = true; 4535 unsigned int cpu; 4536 4537 rcu_read_lock(); 4538 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4539 if (flow_table && flow_id <= flow_table->mask) { 4540 rflow = &flow_table->flows[flow_id]; 4541 cpu = READ_ONCE(rflow->cpu); 4542 if (rflow->filter == filter_id && cpu < nr_cpu_ids && 4543 ((int)(per_cpu(softnet_data, cpu).input_queue_head - 4544 rflow->last_qtail) < 4545 (int)(10 * flow_table->mask))) 4546 expire = false; 4547 } 4548 rcu_read_unlock(); 4549 return expire; 4550 } 4551 EXPORT_SYMBOL(rps_may_expire_flow); 4552 4553 #endif /* CONFIG_RFS_ACCEL */ 4554 4555 /* Called from hardirq (IPI) context */ 4556 static void rps_trigger_softirq(void *data) 4557 { 4558 struct softnet_data *sd = data; 4559 4560 ____napi_schedule(sd, &sd->backlog); 4561 sd->received_rps++; 4562 } 4563 4564 #endif /* CONFIG_RPS */ 4565 4566 /* 4567 * Check if this softnet_data structure is another cpu one 4568 * If yes, queue it to our IPI list and return 1 4569 * If no, return 0 4570 */ 4571 static int rps_ipi_queued(struct softnet_data *sd) 4572 { 4573 #ifdef CONFIG_RPS 4574 struct softnet_data *mysd = this_cpu_ptr(&softnet_data); 4575 4576 if (sd != mysd) { 4577 sd->rps_ipi_next = mysd->rps_ipi_list; 4578 mysd->rps_ipi_list = sd; 4579 4580 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4581 return 1; 4582 } 4583 #endif /* CONFIG_RPS */ 4584 return 0; 4585 } 4586 4587 #ifdef CONFIG_NET_FLOW_LIMIT 4588 int netdev_flow_limit_table_len __read_mostly = (1 << 12); 4589 #endif 4590 4591 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) 4592 { 4593 #ifdef CONFIG_NET_FLOW_LIMIT 4594 struct sd_flow_limit *fl; 4595 struct softnet_data *sd; 4596 unsigned int old_flow, new_flow; 4597 4598 if (qlen < (netdev_max_backlog >> 1)) 4599 return false; 4600 4601 sd = this_cpu_ptr(&softnet_data); 4602 4603 rcu_read_lock(); 4604 fl = rcu_dereference(sd->flow_limit); 4605 if (fl) { 4606 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); 4607 old_flow = fl->history[fl->history_head]; 4608 fl->history[fl->history_head] = new_flow; 4609 4610 fl->history_head++; 4611 fl->history_head &= FLOW_LIMIT_HISTORY - 1; 4612 4613 if (likely(fl->buckets[old_flow])) 4614 fl->buckets[old_flow]--; 4615 4616 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { 4617 fl->count++; 4618 rcu_read_unlock(); 4619 return true; 4620 } 4621 } 4622 rcu_read_unlock(); 4623 #endif 4624 return false; 4625 } 4626 4627 /* 4628 * enqueue_to_backlog is called to queue an skb to a per CPU backlog 4629 * queue (may be a remote CPU queue). 4630 */ 4631 static int enqueue_to_backlog(struct sk_buff *skb, int cpu, 4632 unsigned int *qtail) 4633 { 4634 struct softnet_data *sd; 4635 unsigned long flags; 4636 unsigned int qlen; 4637 4638 sd = &per_cpu(softnet_data, cpu); 4639 4640 local_irq_save(flags); 4641 4642 rps_lock(sd); 4643 if (!netif_running(skb->dev)) 4644 goto drop; 4645 qlen = skb_queue_len(&sd->input_pkt_queue); 4646 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) { 4647 if (qlen) { 4648 enqueue: 4649 __skb_queue_tail(&sd->input_pkt_queue, skb); 4650 input_queue_tail_incr_save(sd, qtail); 4651 rps_unlock(sd); 4652 local_irq_restore(flags); 4653 return NET_RX_SUCCESS; 4654 } 4655 4656 /* Schedule NAPI for backlog device 4657 * We can use non atomic operation since we own the queue lock 4658 */ 4659 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) { 4660 if (!rps_ipi_queued(sd)) 4661 ____napi_schedule(sd, &sd->backlog); 4662 } 4663 goto enqueue; 4664 } 4665 4666 drop: 4667 sd->dropped++; 4668 rps_unlock(sd); 4669 4670 local_irq_restore(flags); 4671 4672 atomic_long_inc(&skb->dev->rx_dropped); 4673 kfree_skb(skb); 4674 return NET_RX_DROP; 4675 } 4676 4677 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb) 4678 { 4679 struct net_device *dev = skb->dev; 4680 struct netdev_rx_queue *rxqueue; 4681 4682 rxqueue = dev->_rx; 4683 4684 if (skb_rx_queue_recorded(skb)) { 4685 u16 index = skb_get_rx_queue(skb); 4686 4687 if (unlikely(index >= dev->real_num_rx_queues)) { 4688 WARN_ONCE(dev->real_num_rx_queues > 1, 4689 "%s received packet on queue %u, but number " 4690 "of RX queues is %u\n", 4691 dev->name, index, dev->real_num_rx_queues); 4692 4693 return rxqueue; /* Return first rxqueue */ 4694 } 4695 rxqueue += index; 4696 } 4697 return rxqueue; 4698 } 4699 4700 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp, 4701 struct bpf_prog *xdp_prog) 4702 { 4703 void *orig_data, *orig_data_end, *hard_start; 4704 struct netdev_rx_queue *rxqueue; 4705 bool orig_bcast, orig_host; 4706 u32 mac_len, frame_sz; 4707 __be16 orig_eth_type; 4708 struct ethhdr *eth; 4709 u32 metalen, act; 4710 int off; 4711 4712 /* The XDP program wants to see the packet starting at the MAC 4713 * header. 4714 */ 4715 mac_len = skb->data - skb_mac_header(skb); 4716 hard_start = skb->data - skb_headroom(skb); 4717 4718 /* SKB "head" area always have tailroom for skb_shared_info */ 4719 frame_sz = (void *)skb_end_pointer(skb) - hard_start; 4720 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 4721 4722 rxqueue = netif_get_rxqueue(skb); 4723 xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq); 4724 xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len, 4725 skb_headlen(skb) + mac_len, true); 4726 4727 orig_data_end = xdp->data_end; 4728 orig_data = xdp->data; 4729 eth = (struct ethhdr *)xdp->data; 4730 orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr); 4731 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest); 4732 orig_eth_type = eth->h_proto; 4733 4734 act = bpf_prog_run_xdp(xdp_prog, xdp); 4735 4736 /* check if bpf_xdp_adjust_head was used */ 4737 off = xdp->data - orig_data; 4738 if (off) { 4739 if (off > 0) 4740 __skb_pull(skb, off); 4741 else if (off < 0) 4742 __skb_push(skb, -off); 4743 4744 skb->mac_header += off; 4745 skb_reset_network_header(skb); 4746 } 4747 4748 /* check if bpf_xdp_adjust_tail was used */ 4749 off = xdp->data_end - orig_data_end; 4750 if (off != 0) { 4751 skb_set_tail_pointer(skb, xdp->data_end - xdp->data); 4752 skb->len += off; /* positive on grow, negative on shrink */ 4753 } 4754 4755 /* check if XDP changed eth hdr such SKB needs update */ 4756 eth = (struct ethhdr *)xdp->data; 4757 if ((orig_eth_type != eth->h_proto) || 4758 (orig_host != ether_addr_equal_64bits(eth->h_dest, 4759 skb->dev->dev_addr)) || 4760 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) { 4761 __skb_push(skb, ETH_HLEN); 4762 skb->pkt_type = PACKET_HOST; 4763 skb->protocol = eth_type_trans(skb, skb->dev); 4764 } 4765 4766 /* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull 4767 * before calling us again on redirect path. We do not call do_redirect 4768 * as we leave that up to the caller. 4769 * 4770 * Caller is responsible for managing lifetime of skb (i.e. calling 4771 * kfree_skb in response to actions it cannot handle/XDP_DROP). 4772 */ 4773 switch (act) { 4774 case XDP_REDIRECT: 4775 case XDP_TX: 4776 __skb_push(skb, mac_len); 4777 break; 4778 case XDP_PASS: 4779 metalen = xdp->data - xdp->data_meta; 4780 if (metalen) 4781 skb_metadata_set(skb, metalen); 4782 break; 4783 } 4784 4785 return act; 4786 } 4787 4788 static u32 netif_receive_generic_xdp(struct sk_buff *skb, 4789 struct xdp_buff *xdp, 4790 struct bpf_prog *xdp_prog) 4791 { 4792 u32 act = XDP_DROP; 4793 4794 /* Reinjected packets coming from act_mirred or similar should 4795 * not get XDP generic processing. 4796 */ 4797 if (skb_is_redirected(skb)) 4798 return XDP_PASS; 4799 4800 /* XDP packets must be linear and must have sufficient headroom 4801 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also 4802 * native XDP provides, thus we need to do it here as well. 4803 */ 4804 if (skb_cloned(skb) || skb_is_nonlinear(skb) || 4805 skb_headroom(skb) < XDP_PACKET_HEADROOM) { 4806 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb); 4807 int troom = skb->tail + skb->data_len - skb->end; 4808 4809 /* In case we have to go down the path and also linearize, 4810 * then lets do the pskb_expand_head() work just once here. 4811 */ 4812 if (pskb_expand_head(skb, 4813 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0, 4814 troom > 0 ? troom + 128 : 0, GFP_ATOMIC)) 4815 goto do_drop; 4816 if (skb_linearize(skb)) 4817 goto do_drop; 4818 } 4819 4820 act = bpf_prog_run_generic_xdp(skb, xdp, xdp_prog); 4821 switch (act) { 4822 case XDP_REDIRECT: 4823 case XDP_TX: 4824 case XDP_PASS: 4825 break; 4826 default: 4827 bpf_warn_invalid_xdp_action(act); 4828 fallthrough; 4829 case XDP_ABORTED: 4830 trace_xdp_exception(skb->dev, xdp_prog, act); 4831 fallthrough; 4832 case XDP_DROP: 4833 do_drop: 4834 kfree_skb(skb); 4835 break; 4836 } 4837 4838 return act; 4839 } 4840 4841 /* When doing generic XDP we have to bypass the qdisc layer and the 4842 * network taps in order to match in-driver-XDP behavior. 4843 */ 4844 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog) 4845 { 4846 struct net_device *dev = skb->dev; 4847 struct netdev_queue *txq; 4848 bool free_skb = true; 4849 int cpu, rc; 4850 4851 txq = netdev_core_pick_tx(dev, skb, NULL); 4852 cpu = smp_processor_id(); 4853 HARD_TX_LOCK(dev, txq, cpu); 4854 if (!netif_xmit_stopped(txq)) { 4855 rc = netdev_start_xmit(skb, dev, txq, 0); 4856 if (dev_xmit_complete(rc)) 4857 free_skb = false; 4858 } 4859 HARD_TX_UNLOCK(dev, txq); 4860 if (free_skb) { 4861 trace_xdp_exception(dev, xdp_prog, XDP_TX); 4862 kfree_skb(skb); 4863 } 4864 } 4865 4866 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key); 4867 4868 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb) 4869 { 4870 if (xdp_prog) { 4871 struct xdp_buff xdp; 4872 u32 act; 4873 int err; 4874 4875 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog); 4876 if (act != XDP_PASS) { 4877 switch (act) { 4878 case XDP_REDIRECT: 4879 err = xdp_do_generic_redirect(skb->dev, skb, 4880 &xdp, xdp_prog); 4881 if (err) 4882 goto out_redir; 4883 break; 4884 case XDP_TX: 4885 generic_xdp_tx(skb, xdp_prog); 4886 break; 4887 } 4888 return XDP_DROP; 4889 } 4890 } 4891 return XDP_PASS; 4892 out_redir: 4893 kfree_skb(skb); 4894 return XDP_DROP; 4895 } 4896 EXPORT_SYMBOL_GPL(do_xdp_generic); 4897 4898 static int netif_rx_internal(struct sk_buff *skb) 4899 { 4900 int ret; 4901 4902 net_timestamp_check(netdev_tstamp_prequeue, skb); 4903 4904 trace_netif_rx(skb); 4905 4906 #ifdef CONFIG_RPS 4907 if (static_branch_unlikely(&rps_needed)) { 4908 struct rps_dev_flow voidflow, *rflow = &voidflow; 4909 int cpu; 4910 4911 preempt_disable(); 4912 rcu_read_lock(); 4913 4914 cpu = get_rps_cpu(skb->dev, skb, &rflow); 4915 if (cpu < 0) 4916 cpu = smp_processor_id(); 4917 4918 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 4919 4920 rcu_read_unlock(); 4921 preempt_enable(); 4922 } else 4923 #endif 4924 { 4925 unsigned int qtail; 4926 4927 ret = enqueue_to_backlog(skb, get_cpu(), &qtail); 4928 put_cpu(); 4929 } 4930 return ret; 4931 } 4932 4933 /** 4934 * netif_rx - post buffer to the network code 4935 * @skb: buffer to post 4936 * 4937 * This function receives a packet from a device driver and queues it for 4938 * the upper (protocol) levels to process. It always succeeds. The buffer 4939 * may be dropped during processing for congestion control or by the 4940 * protocol layers. 4941 * 4942 * return values: 4943 * NET_RX_SUCCESS (no congestion) 4944 * NET_RX_DROP (packet was dropped) 4945 * 4946 */ 4947 4948 int netif_rx(struct sk_buff *skb) 4949 { 4950 int ret; 4951 4952 trace_netif_rx_entry(skb); 4953 4954 ret = netif_rx_internal(skb); 4955 trace_netif_rx_exit(ret); 4956 4957 return ret; 4958 } 4959 EXPORT_SYMBOL(netif_rx); 4960 4961 int netif_rx_ni(struct sk_buff *skb) 4962 { 4963 int err; 4964 4965 trace_netif_rx_ni_entry(skb); 4966 4967 preempt_disable(); 4968 err = netif_rx_internal(skb); 4969 if (local_softirq_pending()) 4970 do_softirq(); 4971 preempt_enable(); 4972 trace_netif_rx_ni_exit(err); 4973 4974 return err; 4975 } 4976 EXPORT_SYMBOL(netif_rx_ni); 4977 4978 int netif_rx_any_context(struct sk_buff *skb) 4979 { 4980 /* 4981 * If invoked from contexts which do not invoke bottom half 4982 * processing either at return from interrupt or when softrqs are 4983 * reenabled, use netif_rx_ni() which invokes bottomhalf processing 4984 * directly. 4985 */ 4986 if (in_interrupt()) 4987 return netif_rx(skb); 4988 else 4989 return netif_rx_ni(skb); 4990 } 4991 EXPORT_SYMBOL(netif_rx_any_context); 4992 4993 static __latent_entropy void net_tx_action(struct softirq_action *h) 4994 { 4995 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 4996 4997 if (sd->completion_queue) { 4998 struct sk_buff *clist; 4999 5000 local_irq_disable(); 5001 clist = sd->completion_queue; 5002 sd->completion_queue = NULL; 5003 local_irq_enable(); 5004 5005 while (clist) { 5006 struct sk_buff *skb = clist; 5007 5008 clist = clist->next; 5009 5010 WARN_ON(refcount_read(&skb->users)); 5011 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED)) 5012 trace_consume_skb(skb); 5013 else 5014 trace_kfree_skb(skb, net_tx_action); 5015 5016 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) 5017 __kfree_skb(skb); 5018 else 5019 __kfree_skb_defer(skb); 5020 } 5021 } 5022 5023 if (sd->output_queue) { 5024 struct Qdisc *head; 5025 5026 local_irq_disable(); 5027 head = sd->output_queue; 5028 sd->output_queue = NULL; 5029 sd->output_queue_tailp = &sd->output_queue; 5030 local_irq_enable(); 5031 5032 rcu_read_lock(); 5033 5034 while (head) { 5035 struct Qdisc *q = head; 5036 spinlock_t *root_lock = NULL; 5037 5038 head = head->next_sched; 5039 5040 /* We need to make sure head->next_sched is read 5041 * before clearing __QDISC_STATE_SCHED 5042 */ 5043 smp_mb__before_atomic(); 5044 5045 if (!(q->flags & TCQ_F_NOLOCK)) { 5046 root_lock = qdisc_lock(q); 5047 spin_lock(root_lock); 5048 } else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, 5049 &q->state))) { 5050 /* There is a synchronize_net() between 5051 * STATE_DEACTIVATED flag being set and 5052 * qdisc_reset()/some_qdisc_is_busy() in 5053 * dev_deactivate(), so we can safely bail out 5054 * early here to avoid data race between 5055 * qdisc_deactivate() and some_qdisc_is_busy() 5056 * for lockless qdisc. 5057 */ 5058 clear_bit(__QDISC_STATE_SCHED, &q->state); 5059 continue; 5060 } 5061 5062 clear_bit(__QDISC_STATE_SCHED, &q->state); 5063 qdisc_run(q); 5064 if (root_lock) 5065 spin_unlock(root_lock); 5066 } 5067 5068 rcu_read_unlock(); 5069 } 5070 5071 xfrm_dev_backlog(sd); 5072 } 5073 5074 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) 5075 /* This hook is defined here for ATM LANE */ 5076 int (*br_fdb_test_addr_hook)(struct net_device *dev, 5077 unsigned char *addr) __read_mostly; 5078 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); 5079 #endif 5080 5081 static inline struct sk_buff * 5082 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 5083 struct net_device *orig_dev, bool *another) 5084 { 5085 #ifdef CONFIG_NET_CLS_ACT 5086 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress); 5087 struct tcf_result cl_res; 5088 5089 /* If there's at least one ingress present somewhere (so 5090 * we get here via enabled static key), remaining devices 5091 * that are not configured with an ingress qdisc will bail 5092 * out here. 5093 */ 5094 if (!miniq) 5095 return skb; 5096 5097 if (*pt_prev) { 5098 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5099 *pt_prev = NULL; 5100 } 5101 5102 qdisc_skb_cb(skb)->pkt_len = skb->len; 5103 qdisc_skb_cb(skb)->mru = 0; 5104 qdisc_skb_cb(skb)->post_ct = false; 5105 skb->tc_at_ingress = 1; 5106 mini_qdisc_bstats_cpu_update(miniq, skb); 5107 5108 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) { 5109 case TC_ACT_OK: 5110 case TC_ACT_RECLASSIFY: 5111 skb->tc_index = TC_H_MIN(cl_res.classid); 5112 break; 5113 case TC_ACT_SHOT: 5114 mini_qdisc_qstats_cpu_drop(miniq); 5115 kfree_skb(skb); 5116 return NULL; 5117 case TC_ACT_STOLEN: 5118 case TC_ACT_QUEUED: 5119 case TC_ACT_TRAP: 5120 consume_skb(skb); 5121 return NULL; 5122 case TC_ACT_REDIRECT: 5123 /* skb_mac_header check was done by cls/act_bpf, so 5124 * we can safely push the L2 header back before 5125 * redirecting to another netdev 5126 */ 5127 __skb_push(skb, skb->mac_len); 5128 if (skb_do_redirect(skb) == -EAGAIN) { 5129 __skb_pull(skb, skb->mac_len); 5130 *another = true; 5131 break; 5132 } 5133 return NULL; 5134 case TC_ACT_CONSUMED: 5135 return NULL; 5136 default: 5137 break; 5138 } 5139 #endif /* CONFIG_NET_CLS_ACT */ 5140 return skb; 5141 } 5142 5143 /** 5144 * netdev_is_rx_handler_busy - check if receive handler is registered 5145 * @dev: device to check 5146 * 5147 * Check if a receive handler is already registered for a given device. 5148 * Return true if there one. 5149 * 5150 * The caller must hold the rtnl_mutex. 5151 */ 5152 bool netdev_is_rx_handler_busy(struct net_device *dev) 5153 { 5154 ASSERT_RTNL(); 5155 return dev && rtnl_dereference(dev->rx_handler); 5156 } 5157 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); 5158 5159 /** 5160 * netdev_rx_handler_register - register receive handler 5161 * @dev: device to register a handler for 5162 * @rx_handler: receive handler to register 5163 * @rx_handler_data: data pointer that is used by rx handler 5164 * 5165 * Register a receive handler for a device. This handler will then be 5166 * called from __netif_receive_skb. A negative errno code is returned 5167 * on a failure. 5168 * 5169 * The caller must hold the rtnl_mutex. 5170 * 5171 * For a general description of rx_handler, see enum rx_handler_result. 5172 */ 5173 int netdev_rx_handler_register(struct net_device *dev, 5174 rx_handler_func_t *rx_handler, 5175 void *rx_handler_data) 5176 { 5177 if (netdev_is_rx_handler_busy(dev)) 5178 return -EBUSY; 5179 5180 if (dev->priv_flags & IFF_NO_RX_HANDLER) 5181 return -EINVAL; 5182 5183 /* Note: rx_handler_data must be set before rx_handler */ 5184 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); 5185 rcu_assign_pointer(dev->rx_handler, rx_handler); 5186 5187 return 0; 5188 } 5189 EXPORT_SYMBOL_GPL(netdev_rx_handler_register); 5190 5191 /** 5192 * netdev_rx_handler_unregister - unregister receive handler 5193 * @dev: device to unregister a handler from 5194 * 5195 * Unregister a receive handler from a device. 5196 * 5197 * The caller must hold the rtnl_mutex. 5198 */ 5199 void netdev_rx_handler_unregister(struct net_device *dev) 5200 { 5201 5202 ASSERT_RTNL(); 5203 RCU_INIT_POINTER(dev->rx_handler, NULL); 5204 /* a reader seeing a non NULL rx_handler in a rcu_read_lock() 5205 * section has a guarantee to see a non NULL rx_handler_data 5206 * as well. 5207 */ 5208 synchronize_net(); 5209 RCU_INIT_POINTER(dev->rx_handler_data, NULL); 5210 } 5211 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); 5212 5213 /* 5214 * Limit the use of PFMEMALLOC reserves to those protocols that implement 5215 * the special handling of PFMEMALLOC skbs. 5216 */ 5217 static bool skb_pfmemalloc_protocol(struct sk_buff *skb) 5218 { 5219 switch (skb->protocol) { 5220 case htons(ETH_P_ARP): 5221 case htons(ETH_P_IP): 5222 case htons(ETH_P_IPV6): 5223 case htons(ETH_P_8021Q): 5224 case htons(ETH_P_8021AD): 5225 return true; 5226 default: 5227 return false; 5228 } 5229 } 5230 5231 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, 5232 int *ret, struct net_device *orig_dev) 5233 { 5234 if (nf_hook_ingress_active(skb)) { 5235 int ingress_retval; 5236 5237 if (*pt_prev) { 5238 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5239 *pt_prev = NULL; 5240 } 5241 5242 rcu_read_lock(); 5243 ingress_retval = nf_hook_ingress(skb); 5244 rcu_read_unlock(); 5245 return ingress_retval; 5246 } 5247 return 0; 5248 } 5249 5250 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc, 5251 struct packet_type **ppt_prev) 5252 { 5253 struct packet_type *ptype, *pt_prev; 5254 rx_handler_func_t *rx_handler; 5255 struct sk_buff *skb = *pskb; 5256 struct net_device *orig_dev; 5257 bool deliver_exact = false; 5258 int ret = NET_RX_DROP; 5259 __be16 type; 5260 5261 net_timestamp_check(!netdev_tstamp_prequeue, skb); 5262 5263 trace_netif_receive_skb(skb); 5264 5265 orig_dev = skb->dev; 5266 5267 skb_reset_network_header(skb); 5268 if (!skb_transport_header_was_set(skb)) 5269 skb_reset_transport_header(skb); 5270 skb_reset_mac_len(skb); 5271 5272 pt_prev = NULL; 5273 5274 another_round: 5275 skb->skb_iif = skb->dev->ifindex; 5276 5277 __this_cpu_inc(softnet_data.processed); 5278 5279 if (static_branch_unlikely(&generic_xdp_needed_key)) { 5280 int ret2; 5281 5282 migrate_disable(); 5283 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb); 5284 migrate_enable(); 5285 5286 if (ret2 != XDP_PASS) { 5287 ret = NET_RX_DROP; 5288 goto out; 5289 } 5290 } 5291 5292 if (eth_type_vlan(skb->protocol)) { 5293 skb = skb_vlan_untag(skb); 5294 if (unlikely(!skb)) 5295 goto out; 5296 } 5297 5298 if (skb_skip_tc_classify(skb)) 5299 goto skip_classify; 5300 5301 if (pfmemalloc) 5302 goto skip_taps; 5303 5304 list_for_each_entry_rcu(ptype, &ptype_all, list) { 5305 if (pt_prev) 5306 ret = deliver_skb(skb, pt_prev, orig_dev); 5307 pt_prev = ptype; 5308 } 5309 5310 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { 5311 if (pt_prev) 5312 ret = deliver_skb(skb, pt_prev, orig_dev); 5313 pt_prev = ptype; 5314 } 5315 5316 skip_taps: 5317 #ifdef CONFIG_NET_INGRESS 5318 if (static_branch_unlikely(&ingress_needed_key)) { 5319 bool another = false; 5320 5321 nf_skip_egress(skb, true); 5322 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev, 5323 &another); 5324 if (another) 5325 goto another_round; 5326 if (!skb) 5327 goto out; 5328 5329 nf_skip_egress(skb, false); 5330 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) 5331 goto out; 5332 } 5333 #endif 5334 skb_reset_redirect(skb); 5335 skip_classify: 5336 if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) 5337 goto drop; 5338 5339 if (skb_vlan_tag_present(skb)) { 5340 if (pt_prev) { 5341 ret = deliver_skb(skb, pt_prev, orig_dev); 5342 pt_prev = NULL; 5343 } 5344 if (vlan_do_receive(&skb)) 5345 goto another_round; 5346 else if (unlikely(!skb)) 5347 goto out; 5348 } 5349 5350 rx_handler = rcu_dereference(skb->dev->rx_handler); 5351 if (rx_handler) { 5352 if (pt_prev) { 5353 ret = deliver_skb(skb, pt_prev, orig_dev); 5354 pt_prev = NULL; 5355 } 5356 switch (rx_handler(&skb)) { 5357 case RX_HANDLER_CONSUMED: 5358 ret = NET_RX_SUCCESS; 5359 goto out; 5360 case RX_HANDLER_ANOTHER: 5361 goto another_round; 5362 case RX_HANDLER_EXACT: 5363 deliver_exact = true; 5364 break; 5365 case RX_HANDLER_PASS: 5366 break; 5367 default: 5368 BUG(); 5369 } 5370 } 5371 5372 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) { 5373 check_vlan_id: 5374 if (skb_vlan_tag_get_id(skb)) { 5375 /* Vlan id is non 0 and vlan_do_receive() above couldn't 5376 * find vlan device. 5377 */ 5378 skb->pkt_type = PACKET_OTHERHOST; 5379 } else if (eth_type_vlan(skb->protocol)) { 5380 /* Outer header is 802.1P with vlan 0, inner header is 5381 * 802.1Q or 802.1AD and vlan_do_receive() above could 5382 * not find vlan dev for vlan id 0. 5383 */ 5384 __vlan_hwaccel_clear_tag(skb); 5385 skb = skb_vlan_untag(skb); 5386 if (unlikely(!skb)) 5387 goto out; 5388 if (vlan_do_receive(&skb)) 5389 /* After stripping off 802.1P header with vlan 0 5390 * vlan dev is found for inner header. 5391 */ 5392 goto another_round; 5393 else if (unlikely(!skb)) 5394 goto out; 5395 else 5396 /* We have stripped outer 802.1P vlan 0 header. 5397 * But could not find vlan dev. 5398 * check again for vlan id to set OTHERHOST. 5399 */ 5400 goto check_vlan_id; 5401 } 5402 /* Note: we might in the future use prio bits 5403 * and set skb->priority like in vlan_do_receive() 5404 * For the time being, just ignore Priority Code Point 5405 */ 5406 __vlan_hwaccel_clear_tag(skb); 5407 } 5408 5409 type = skb->protocol; 5410 5411 /* deliver only exact match when indicated */ 5412 if (likely(!deliver_exact)) { 5413 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5414 &ptype_base[ntohs(type) & 5415 PTYPE_HASH_MASK]); 5416 } 5417 5418 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5419 &orig_dev->ptype_specific); 5420 5421 if (unlikely(skb->dev != orig_dev)) { 5422 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5423 &skb->dev->ptype_specific); 5424 } 5425 5426 if (pt_prev) { 5427 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 5428 goto drop; 5429 *ppt_prev = pt_prev; 5430 } else { 5431 drop: 5432 if (!deliver_exact) 5433 atomic_long_inc(&skb->dev->rx_dropped); 5434 else 5435 atomic_long_inc(&skb->dev->rx_nohandler); 5436 kfree_skb(skb); 5437 /* Jamal, now you will not able to escape explaining 5438 * me how you were going to use this. :-) 5439 */ 5440 ret = NET_RX_DROP; 5441 } 5442 5443 out: 5444 /* The invariant here is that if *ppt_prev is not NULL 5445 * then skb should also be non-NULL. 5446 * 5447 * Apparently *ppt_prev assignment above holds this invariant due to 5448 * skb dereferencing near it. 5449 */ 5450 *pskb = skb; 5451 return ret; 5452 } 5453 5454 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc) 5455 { 5456 struct net_device *orig_dev = skb->dev; 5457 struct packet_type *pt_prev = NULL; 5458 int ret; 5459 5460 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5461 if (pt_prev) 5462 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb, 5463 skb->dev, pt_prev, orig_dev); 5464 return ret; 5465 } 5466 5467 /** 5468 * netif_receive_skb_core - special purpose version of netif_receive_skb 5469 * @skb: buffer to process 5470 * 5471 * More direct receive version of netif_receive_skb(). It should 5472 * only be used by callers that have a need to skip RPS and Generic XDP. 5473 * Caller must also take care of handling if ``(page_is_)pfmemalloc``. 5474 * 5475 * This function may only be called from softirq context and interrupts 5476 * should be enabled. 5477 * 5478 * Return values (usually ignored): 5479 * NET_RX_SUCCESS: no congestion 5480 * NET_RX_DROP: packet was dropped 5481 */ 5482 int netif_receive_skb_core(struct sk_buff *skb) 5483 { 5484 int ret; 5485 5486 rcu_read_lock(); 5487 ret = __netif_receive_skb_one_core(skb, false); 5488 rcu_read_unlock(); 5489 5490 return ret; 5491 } 5492 EXPORT_SYMBOL(netif_receive_skb_core); 5493 5494 static inline void __netif_receive_skb_list_ptype(struct list_head *head, 5495 struct packet_type *pt_prev, 5496 struct net_device *orig_dev) 5497 { 5498 struct sk_buff *skb, *next; 5499 5500 if (!pt_prev) 5501 return; 5502 if (list_empty(head)) 5503 return; 5504 if (pt_prev->list_func != NULL) 5505 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv, 5506 ip_list_rcv, head, pt_prev, orig_dev); 5507 else 5508 list_for_each_entry_safe(skb, next, head, list) { 5509 skb_list_del_init(skb); 5510 pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 5511 } 5512 } 5513 5514 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc) 5515 { 5516 /* Fast-path assumptions: 5517 * - There is no RX handler. 5518 * - Only one packet_type matches. 5519 * If either of these fails, we will end up doing some per-packet 5520 * processing in-line, then handling the 'last ptype' for the whole 5521 * sublist. This can't cause out-of-order delivery to any single ptype, 5522 * because the 'last ptype' must be constant across the sublist, and all 5523 * other ptypes are handled per-packet. 5524 */ 5525 /* Current (common) ptype of sublist */ 5526 struct packet_type *pt_curr = NULL; 5527 /* Current (common) orig_dev of sublist */ 5528 struct net_device *od_curr = NULL; 5529 struct list_head sublist; 5530 struct sk_buff *skb, *next; 5531 5532 INIT_LIST_HEAD(&sublist); 5533 list_for_each_entry_safe(skb, next, head, list) { 5534 struct net_device *orig_dev = skb->dev; 5535 struct packet_type *pt_prev = NULL; 5536 5537 skb_list_del_init(skb); 5538 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5539 if (!pt_prev) 5540 continue; 5541 if (pt_curr != pt_prev || od_curr != orig_dev) { 5542 /* dispatch old sublist */ 5543 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5544 /* start new sublist */ 5545 INIT_LIST_HEAD(&sublist); 5546 pt_curr = pt_prev; 5547 od_curr = orig_dev; 5548 } 5549 list_add_tail(&skb->list, &sublist); 5550 } 5551 5552 /* dispatch final sublist */ 5553 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5554 } 5555 5556 static int __netif_receive_skb(struct sk_buff *skb) 5557 { 5558 int ret; 5559 5560 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { 5561 unsigned int noreclaim_flag; 5562 5563 /* 5564 * PFMEMALLOC skbs are special, they should 5565 * - be delivered to SOCK_MEMALLOC sockets only 5566 * - stay away from userspace 5567 * - have bounded memory usage 5568 * 5569 * Use PF_MEMALLOC as this saves us from propagating the allocation 5570 * context down to all allocation sites. 5571 */ 5572 noreclaim_flag = memalloc_noreclaim_save(); 5573 ret = __netif_receive_skb_one_core(skb, true); 5574 memalloc_noreclaim_restore(noreclaim_flag); 5575 } else 5576 ret = __netif_receive_skb_one_core(skb, false); 5577 5578 return ret; 5579 } 5580 5581 static void __netif_receive_skb_list(struct list_head *head) 5582 { 5583 unsigned long noreclaim_flag = 0; 5584 struct sk_buff *skb, *next; 5585 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */ 5586 5587 list_for_each_entry_safe(skb, next, head, list) { 5588 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) { 5589 struct list_head sublist; 5590 5591 /* Handle the previous sublist */ 5592 list_cut_before(&sublist, head, &skb->list); 5593 if (!list_empty(&sublist)) 5594 __netif_receive_skb_list_core(&sublist, pfmemalloc); 5595 pfmemalloc = !pfmemalloc; 5596 /* See comments in __netif_receive_skb */ 5597 if (pfmemalloc) 5598 noreclaim_flag = memalloc_noreclaim_save(); 5599 else 5600 memalloc_noreclaim_restore(noreclaim_flag); 5601 } 5602 } 5603 /* Handle the remaining sublist */ 5604 if (!list_empty(head)) 5605 __netif_receive_skb_list_core(head, pfmemalloc); 5606 /* Restore pflags */ 5607 if (pfmemalloc) 5608 memalloc_noreclaim_restore(noreclaim_flag); 5609 } 5610 5611 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp) 5612 { 5613 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog); 5614 struct bpf_prog *new = xdp->prog; 5615 int ret = 0; 5616 5617 switch (xdp->command) { 5618 case XDP_SETUP_PROG: 5619 rcu_assign_pointer(dev->xdp_prog, new); 5620 if (old) 5621 bpf_prog_put(old); 5622 5623 if (old && !new) { 5624 static_branch_dec(&generic_xdp_needed_key); 5625 } else if (new && !old) { 5626 static_branch_inc(&generic_xdp_needed_key); 5627 dev_disable_lro(dev); 5628 dev_disable_gro_hw(dev); 5629 } 5630 break; 5631 5632 default: 5633 ret = -EINVAL; 5634 break; 5635 } 5636 5637 return ret; 5638 } 5639 5640 static int netif_receive_skb_internal(struct sk_buff *skb) 5641 { 5642 int ret; 5643 5644 net_timestamp_check(netdev_tstamp_prequeue, skb); 5645 5646 if (skb_defer_rx_timestamp(skb)) 5647 return NET_RX_SUCCESS; 5648 5649 rcu_read_lock(); 5650 #ifdef CONFIG_RPS 5651 if (static_branch_unlikely(&rps_needed)) { 5652 struct rps_dev_flow voidflow, *rflow = &voidflow; 5653 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5654 5655 if (cpu >= 0) { 5656 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5657 rcu_read_unlock(); 5658 return ret; 5659 } 5660 } 5661 #endif 5662 ret = __netif_receive_skb(skb); 5663 rcu_read_unlock(); 5664 return ret; 5665 } 5666 5667 static void netif_receive_skb_list_internal(struct list_head *head) 5668 { 5669 struct sk_buff *skb, *next; 5670 struct list_head sublist; 5671 5672 INIT_LIST_HEAD(&sublist); 5673 list_for_each_entry_safe(skb, next, head, list) { 5674 net_timestamp_check(netdev_tstamp_prequeue, skb); 5675 skb_list_del_init(skb); 5676 if (!skb_defer_rx_timestamp(skb)) 5677 list_add_tail(&skb->list, &sublist); 5678 } 5679 list_splice_init(&sublist, head); 5680 5681 rcu_read_lock(); 5682 #ifdef CONFIG_RPS 5683 if (static_branch_unlikely(&rps_needed)) { 5684 list_for_each_entry_safe(skb, next, head, list) { 5685 struct rps_dev_flow voidflow, *rflow = &voidflow; 5686 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5687 5688 if (cpu >= 0) { 5689 /* Will be handled, remove from list */ 5690 skb_list_del_init(skb); 5691 enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5692 } 5693 } 5694 } 5695 #endif 5696 __netif_receive_skb_list(head); 5697 rcu_read_unlock(); 5698 } 5699 5700 /** 5701 * netif_receive_skb - process receive buffer from network 5702 * @skb: buffer to process 5703 * 5704 * netif_receive_skb() is the main receive data processing function. 5705 * It always succeeds. The buffer may be dropped during processing 5706 * for congestion control or by the protocol layers. 5707 * 5708 * This function may only be called from softirq context and interrupts 5709 * should be enabled. 5710 * 5711 * Return values (usually ignored): 5712 * NET_RX_SUCCESS: no congestion 5713 * NET_RX_DROP: packet was dropped 5714 */ 5715 int netif_receive_skb(struct sk_buff *skb) 5716 { 5717 int ret; 5718 5719 trace_netif_receive_skb_entry(skb); 5720 5721 ret = netif_receive_skb_internal(skb); 5722 trace_netif_receive_skb_exit(ret); 5723 5724 return ret; 5725 } 5726 EXPORT_SYMBOL(netif_receive_skb); 5727 5728 /** 5729 * netif_receive_skb_list - process many receive buffers from network 5730 * @head: list of skbs to process. 5731 * 5732 * Since return value of netif_receive_skb() is normally ignored, and 5733 * wouldn't be meaningful for a list, this function returns void. 5734 * 5735 * This function may only be called from softirq context and interrupts 5736 * should be enabled. 5737 */ 5738 void netif_receive_skb_list(struct list_head *head) 5739 { 5740 struct sk_buff *skb; 5741 5742 if (list_empty(head)) 5743 return; 5744 if (trace_netif_receive_skb_list_entry_enabled()) { 5745 list_for_each_entry(skb, head, list) 5746 trace_netif_receive_skb_list_entry(skb); 5747 } 5748 netif_receive_skb_list_internal(head); 5749 trace_netif_receive_skb_list_exit(0); 5750 } 5751 EXPORT_SYMBOL(netif_receive_skb_list); 5752 5753 static DEFINE_PER_CPU(struct work_struct, flush_works); 5754 5755 /* Network device is going away, flush any packets still pending */ 5756 static void flush_backlog(struct work_struct *work) 5757 { 5758 struct sk_buff *skb, *tmp; 5759 struct softnet_data *sd; 5760 5761 local_bh_disable(); 5762 sd = this_cpu_ptr(&softnet_data); 5763 5764 local_irq_disable(); 5765 rps_lock(sd); 5766 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { 5767 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5768 __skb_unlink(skb, &sd->input_pkt_queue); 5769 dev_kfree_skb_irq(skb); 5770 input_queue_head_incr(sd); 5771 } 5772 } 5773 rps_unlock(sd); 5774 local_irq_enable(); 5775 5776 skb_queue_walk_safe(&sd->process_queue, skb, tmp) { 5777 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5778 __skb_unlink(skb, &sd->process_queue); 5779 kfree_skb(skb); 5780 input_queue_head_incr(sd); 5781 } 5782 } 5783 local_bh_enable(); 5784 } 5785 5786 static bool flush_required(int cpu) 5787 { 5788 #if IS_ENABLED(CONFIG_RPS) 5789 struct softnet_data *sd = &per_cpu(softnet_data, cpu); 5790 bool do_flush; 5791 5792 local_irq_disable(); 5793 rps_lock(sd); 5794 5795 /* as insertion into process_queue happens with the rps lock held, 5796 * process_queue access may race only with dequeue 5797 */ 5798 do_flush = !skb_queue_empty(&sd->input_pkt_queue) || 5799 !skb_queue_empty_lockless(&sd->process_queue); 5800 rps_unlock(sd); 5801 local_irq_enable(); 5802 5803 return do_flush; 5804 #endif 5805 /* without RPS we can't safely check input_pkt_queue: during a 5806 * concurrent remote skb_queue_splice() we can detect as empty both 5807 * input_pkt_queue and process_queue even if the latter could end-up 5808 * containing a lot of packets. 5809 */ 5810 return true; 5811 } 5812 5813 static void flush_all_backlogs(void) 5814 { 5815 static cpumask_t flush_cpus; 5816 unsigned int cpu; 5817 5818 /* since we are under rtnl lock protection we can use static data 5819 * for the cpumask and avoid allocating on stack the possibly 5820 * large mask 5821 */ 5822 ASSERT_RTNL(); 5823 5824 cpus_read_lock(); 5825 5826 cpumask_clear(&flush_cpus); 5827 for_each_online_cpu(cpu) { 5828 if (flush_required(cpu)) { 5829 queue_work_on(cpu, system_highpri_wq, 5830 per_cpu_ptr(&flush_works, cpu)); 5831 cpumask_set_cpu(cpu, &flush_cpus); 5832 } 5833 } 5834 5835 /* we can have in flight packet[s] on the cpus we are not flushing, 5836 * synchronize_net() in unregister_netdevice_many() will take care of 5837 * them 5838 */ 5839 for_each_cpu(cpu, &flush_cpus) 5840 flush_work(per_cpu_ptr(&flush_works, cpu)); 5841 5842 cpus_read_unlock(); 5843 } 5844 5845 /* Pass the currently batched GRO_NORMAL SKBs up to the stack. */ 5846 static void gro_normal_list(struct napi_struct *napi) 5847 { 5848 if (!napi->rx_count) 5849 return; 5850 netif_receive_skb_list_internal(&napi->rx_list); 5851 INIT_LIST_HEAD(&napi->rx_list); 5852 napi->rx_count = 0; 5853 } 5854 5855 /* Queue one GRO_NORMAL SKB up for list processing. If batch size exceeded, 5856 * pass the whole batch up to the stack. 5857 */ 5858 static void gro_normal_one(struct napi_struct *napi, struct sk_buff *skb, int segs) 5859 { 5860 list_add_tail(&skb->list, &napi->rx_list); 5861 napi->rx_count += segs; 5862 if (napi->rx_count >= gro_normal_batch) 5863 gro_normal_list(napi); 5864 } 5865 5866 static void napi_gro_complete(struct napi_struct *napi, struct sk_buff *skb) 5867 { 5868 struct packet_offload *ptype; 5869 __be16 type = skb->protocol; 5870 struct list_head *head = &offload_base; 5871 int err = -ENOENT; 5872 5873 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb)); 5874 5875 if (NAPI_GRO_CB(skb)->count == 1) { 5876 skb_shinfo(skb)->gso_size = 0; 5877 goto out; 5878 } 5879 5880 rcu_read_lock(); 5881 list_for_each_entry_rcu(ptype, head, list) { 5882 if (ptype->type != type || !ptype->callbacks.gro_complete) 5883 continue; 5884 5885 err = INDIRECT_CALL_INET(ptype->callbacks.gro_complete, 5886 ipv6_gro_complete, inet_gro_complete, 5887 skb, 0); 5888 break; 5889 } 5890 rcu_read_unlock(); 5891 5892 if (err) { 5893 WARN_ON(&ptype->list == head); 5894 kfree_skb(skb); 5895 return; 5896 } 5897 5898 out: 5899 gro_normal_one(napi, skb, NAPI_GRO_CB(skb)->count); 5900 } 5901 5902 static void __napi_gro_flush_chain(struct napi_struct *napi, u32 index, 5903 bool flush_old) 5904 { 5905 struct list_head *head = &napi->gro_hash[index].list; 5906 struct sk_buff *skb, *p; 5907 5908 list_for_each_entry_safe_reverse(skb, p, head, list) { 5909 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies) 5910 return; 5911 skb_list_del_init(skb); 5912 napi_gro_complete(napi, skb); 5913 napi->gro_hash[index].count--; 5914 } 5915 5916 if (!napi->gro_hash[index].count) 5917 __clear_bit(index, &napi->gro_bitmask); 5918 } 5919 5920 /* napi->gro_hash[].list contains packets ordered by age. 5921 * youngest packets at the head of it. 5922 * Complete skbs in reverse order to reduce latencies. 5923 */ 5924 void napi_gro_flush(struct napi_struct *napi, bool flush_old) 5925 { 5926 unsigned long bitmask = napi->gro_bitmask; 5927 unsigned int i, base = ~0U; 5928 5929 while ((i = ffs(bitmask)) != 0) { 5930 bitmask >>= i; 5931 base += i; 5932 __napi_gro_flush_chain(napi, base, flush_old); 5933 } 5934 } 5935 EXPORT_SYMBOL(napi_gro_flush); 5936 5937 static void gro_list_prepare(const struct list_head *head, 5938 const struct sk_buff *skb) 5939 { 5940 unsigned int maclen = skb->dev->hard_header_len; 5941 u32 hash = skb_get_hash_raw(skb); 5942 struct sk_buff *p; 5943 5944 list_for_each_entry(p, head, list) { 5945 unsigned long diffs; 5946 5947 NAPI_GRO_CB(p)->flush = 0; 5948 5949 if (hash != skb_get_hash_raw(p)) { 5950 NAPI_GRO_CB(p)->same_flow = 0; 5951 continue; 5952 } 5953 5954 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev; 5955 diffs |= skb_vlan_tag_present(p) ^ skb_vlan_tag_present(skb); 5956 if (skb_vlan_tag_present(p)) 5957 diffs |= skb_vlan_tag_get(p) ^ skb_vlan_tag_get(skb); 5958 diffs |= skb_metadata_differs(p, skb); 5959 if (maclen == ETH_HLEN) 5960 diffs |= compare_ether_header(skb_mac_header(p), 5961 skb_mac_header(skb)); 5962 else if (!diffs) 5963 diffs = memcmp(skb_mac_header(p), 5964 skb_mac_header(skb), 5965 maclen); 5966 5967 /* in most common scenarions 'slow_gro' is 0 5968 * otherwise we are already on some slower paths 5969 * either skip all the infrequent tests altogether or 5970 * avoid trying too hard to skip each of them individually 5971 */ 5972 if (!diffs && unlikely(skb->slow_gro | p->slow_gro)) { 5973 #if IS_ENABLED(CONFIG_SKB_EXTENSIONS) && IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 5974 struct tc_skb_ext *skb_ext; 5975 struct tc_skb_ext *p_ext; 5976 #endif 5977 5978 diffs |= p->sk != skb->sk; 5979 diffs |= skb_metadata_dst_cmp(p, skb); 5980 diffs |= skb_get_nfct(p) ^ skb_get_nfct(skb); 5981 5982 #if IS_ENABLED(CONFIG_SKB_EXTENSIONS) && IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 5983 skb_ext = skb_ext_find(skb, TC_SKB_EXT); 5984 p_ext = skb_ext_find(p, TC_SKB_EXT); 5985 5986 diffs |= (!!p_ext) ^ (!!skb_ext); 5987 if (!diffs && unlikely(skb_ext)) 5988 diffs |= p_ext->chain ^ skb_ext->chain; 5989 #endif 5990 } 5991 5992 NAPI_GRO_CB(p)->same_flow = !diffs; 5993 } 5994 } 5995 5996 static inline void skb_gro_reset_offset(struct sk_buff *skb, u32 nhoff) 5997 { 5998 const struct skb_shared_info *pinfo = skb_shinfo(skb); 5999 const skb_frag_t *frag0 = &pinfo->frags[0]; 6000 6001 NAPI_GRO_CB(skb)->data_offset = 0; 6002 NAPI_GRO_CB(skb)->frag0 = NULL; 6003 NAPI_GRO_CB(skb)->frag0_len = 0; 6004 6005 if (!skb_headlen(skb) && pinfo->nr_frags && 6006 !PageHighMem(skb_frag_page(frag0)) && 6007 (!NET_IP_ALIGN || !((skb_frag_off(frag0) + nhoff) & 3))) { 6008 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0); 6009 NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int, 6010 skb_frag_size(frag0), 6011 skb->end - skb->tail); 6012 } 6013 } 6014 6015 static void gro_pull_from_frag0(struct sk_buff *skb, int grow) 6016 { 6017 struct skb_shared_info *pinfo = skb_shinfo(skb); 6018 6019 BUG_ON(skb->end - skb->tail < grow); 6020 6021 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); 6022 6023 skb->data_len -= grow; 6024 skb->tail += grow; 6025 6026 skb_frag_off_add(&pinfo->frags[0], grow); 6027 skb_frag_size_sub(&pinfo->frags[0], grow); 6028 6029 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) { 6030 skb_frag_unref(skb, 0); 6031 memmove(pinfo->frags, pinfo->frags + 1, 6032 --pinfo->nr_frags * sizeof(pinfo->frags[0])); 6033 } 6034 } 6035 6036 static void gro_flush_oldest(struct napi_struct *napi, struct list_head *head) 6037 { 6038 struct sk_buff *oldest; 6039 6040 oldest = list_last_entry(head, struct sk_buff, list); 6041 6042 /* We are called with head length >= MAX_GRO_SKBS, so this is 6043 * impossible. 6044 */ 6045 if (WARN_ON_ONCE(!oldest)) 6046 return; 6047 6048 /* Do not adjust napi->gro_hash[].count, caller is adding a new 6049 * SKB to the chain. 6050 */ 6051 skb_list_del_init(oldest); 6052 napi_gro_complete(napi, oldest); 6053 } 6054 6055 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) 6056 { 6057 u32 bucket = skb_get_hash_raw(skb) & (GRO_HASH_BUCKETS - 1); 6058 struct gro_list *gro_list = &napi->gro_hash[bucket]; 6059 struct list_head *head = &offload_base; 6060 struct packet_offload *ptype; 6061 __be16 type = skb->protocol; 6062 struct sk_buff *pp = NULL; 6063 enum gro_result ret; 6064 int same_flow; 6065 int grow; 6066 6067 if (netif_elide_gro(skb->dev)) 6068 goto normal; 6069 6070 gro_list_prepare(&gro_list->list, skb); 6071 6072 rcu_read_lock(); 6073 list_for_each_entry_rcu(ptype, head, list) { 6074 if (ptype->type != type || !ptype->callbacks.gro_receive) 6075 continue; 6076 6077 skb_set_network_header(skb, skb_gro_offset(skb)); 6078 skb_reset_mac_len(skb); 6079 NAPI_GRO_CB(skb)->same_flow = 0; 6080 NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb); 6081 NAPI_GRO_CB(skb)->free = 0; 6082 NAPI_GRO_CB(skb)->encap_mark = 0; 6083 NAPI_GRO_CB(skb)->recursion_counter = 0; 6084 NAPI_GRO_CB(skb)->is_fou = 0; 6085 NAPI_GRO_CB(skb)->is_atomic = 1; 6086 NAPI_GRO_CB(skb)->gro_remcsum_start = 0; 6087 6088 /* Setup for GRO checksum validation */ 6089 switch (skb->ip_summed) { 6090 case CHECKSUM_COMPLETE: 6091 NAPI_GRO_CB(skb)->csum = skb->csum; 6092 NAPI_GRO_CB(skb)->csum_valid = 1; 6093 NAPI_GRO_CB(skb)->csum_cnt = 0; 6094 break; 6095 case CHECKSUM_UNNECESSARY: 6096 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1; 6097 NAPI_GRO_CB(skb)->csum_valid = 0; 6098 break; 6099 default: 6100 NAPI_GRO_CB(skb)->csum_cnt = 0; 6101 NAPI_GRO_CB(skb)->csum_valid = 0; 6102 } 6103 6104 pp = INDIRECT_CALL_INET(ptype->callbacks.gro_receive, 6105 ipv6_gro_receive, inet_gro_receive, 6106 &gro_list->list, skb); 6107 break; 6108 } 6109 rcu_read_unlock(); 6110 6111 if (&ptype->list == head) 6112 goto normal; 6113 6114 if (PTR_ERR(pp) == -EINPROGRESS) { 6115 ret = GRO_CONSUMED; 6116 goto ok; 6117 } 6118 6119 same_flow = NAPI_GRO_CB(skb)->same_flow; 6120 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; 6121 6122 if (pp) { 6123 skb_list_del_init(pp); 6124 napi_gro_complete(napi, pp); 6125 gro_list->count--; 6126 } 6127 6128 if (same_flow) 6129 goto ok; 6130 6131 if (NAPI_GRO_CB(skb)->flush) 6132 goto normal; 6133 6134 if (unlikely(gro_list->count >= MAX_GRO_SKBS)) 6135 gro_flush_oldest(napi, &gro_list->list); 6136 else 6137 gro_list->count++; 6138 6139 NAPI_GRO_CB(skb)->count = 1; 6140 NAPI_GRO_CB(skb)->age = jiffies; 6141 NAPI_GRO_CB(skb)->last = skb; 6142 skb_shinfo(skb)->gso_size = skb_gro_len(skb); 6143 list_add(&skb->list, &gro_list->list); 6144 ret = GRO_HELD; 6145 6146 pull: 6147 grow = skb_gro_offset(skb) - skb_headlen(skb); 6148 if (grow > 0) 6149 gro_pull_from_frag0(skb, grow); 6150 ok: 6151 if (gro_list->count) { 6152 if (!test_bit(bucket, &napi->gro_bitmask)) 6153 __set_bit(bucket, &napi->gro_bitmask); 6154 } else if (test_bit(bucket, &napi->gro_bitmask)) { 6155 __clear_bit(bucket, &napi->gro_bitmask); 6156 } 6157 6158 return ret; 6159 6160 normal: 6161 ret = GRO_NORMAL; 6162 goto pull; 6163 } 6164 6165 struct packet_offload *gro_find_receive_by_type(__be16 type) 6166 { 6167 struct list_head *offload_head = &offload_base; 6168 struct packet_offload *ptype; 6169 6170 list_for_each_entry_rcu(ptype, offload_head, list) { 6171 if (ptype->type != type || !ptype->callbacks.gro_receive) 6172 continue; 6173 return ptype; 6174 } 6175 return NULL; 6176 } 6177 EXPORT_SYMBOL(gro_find_receive_by_type); 6178 6179 struct packet_offload *gro_find_complete_by_type(__be16 type) 6180 { 6181 struct list_head *offload_head = &offload_base; 6182 struct packet_offload *ptype; 6183 6184 list_for_each_entry_rcu(ptype, offload_head, list) { 6185 if (ptype->type != type || !ptype->callbacks.gro_complete) 6186 continue; 6187 return ptype; 6188 } 6189 return NULL; 6190 } 6191 EXPORT_SYMBOL(gro_find_complete_by_type); 6192 6193 static gro_result_t napi_skb_finish(struct napi_struct *napi, 6194 struct sk_buff *skb, 6195 gro_result_t ret) 6196 { 6197 switch (ret) { 6198 case GRO_NORMAL: 6199 gro_normal_one(napi, skb, 1); 6200 break; 6201 6202 case GRO_MERGED_FREE: 6203 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) 6204 napi_skb_free_stolen_head(skb); 6205 else if (skb->fclone != SKB_FCLONE_UNAVAILABLE) 6206 __kfree_skb(skb); 6207 else 6208 __kfree_skb_defer(skb); 6209 break; 6210 6211 case GRO_HELD: 6212 case GRO_MERGED: 6213 case GRO_CONSUMED: 6214 break; 6215 } 6216 6217 return ret; 6218 } 6219 6220 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) 6221 { 6222 gro_result_t ret; 6223 6224 skb_mark_napi_id(skb, napi); 6225 trace_napi_gro_receive_entry(skb); 6226 6227 skb_gro_reset_offset(skb, 0); 6228 6229 ret = napi_skb_finish(napi, skb, dev_gro_receive(napi, skb)); 6230 trace_napi_gro_receive_exit(ret); 6231 6232 return ret; 6233 } 6234 EXPORT_SYMBOL(napi_gro_receive); 6235 6236 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) 6237 { 6238 if (unlikely(skb->pfmemalloc)) { 6239 consume_skb(skb); 6240 return; 6241 } 6242 __skb_pull(skb, skb_headlen(skb)); 6243 /* restore the reserve we had after netdev_alloc_skb_ip_align() */ 6244 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb)); 6245 __vlan_hwaccel_clear_tag(skb); 6246 skb->dev = napi->dev; 6247 skb->skb_iif = 0; 6248 6249 /* eth_type_trans() assumes pkt_type is PACKET_HOST */ 6250 skb->pkt_type = PACKET_HOST; 6251 6252 skb->encapsulation = 0; 6253 skb_shinfo(skb)->gso_type = 0; 6254 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); 6255 if (unlikely(skb->slow_gro)) { 6256 skb_orphan(skb); 6257 skb_ext_reset(skb); 6258 nf_reset_ct(skb); 6259 skb->slow_gro = 0; 6260 } 6261 6262 napi->skb = skb; 6263 } 6264 6265 struct sk_buff *napi_get_frags(struct napi_struct *napi) 6266 { 6267 struct sk_buff *skb = napi->skb; 6268 6269 if (!skb) { 6270 skb = napi_alloc_skb(napi, GRO_MAX_HEAD); 6271 if (skb) { 6272 napi->skb = skb; 6273 skb_mark_napi_id(skb, napi); 6274 } 6275 } 6276 return skb; 6277 } 6278 EXPORT_SYMBOL(napi_get_frags); 6279 6280 static gro_result_t napi_frags_finish(struct napi_struct *napi, 6281 struct sk_buff *skb, 6282 gro_result_t ret) 6283 { 6284 switch (ret) { 6285 case GRO_NORMAL: 6286 case GRO_HELD: 6287 __skb_push(skb, ETH_HLEN); 6288 skb->protocol = eth_type_trans(skb, skb->dev); 6289 if (ret == GRO_NORMAL) 6290 gro_normal_one(napi, skb, 1); 6291 break; 6292 6293 case GRO_MERGED_FREE: 6294 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) 6295 napi_skb_free_stolen_head(skb); 6296 else 6297 napi_reuse_skb(napi, skb); 6298 break; 6299 6300 case GRO_MERGED: 6301 case GRO_CONSUMED: 6302 break; 6303 } 6304 6305 return ret; 6306 } 6307 6308 /* Upper GRO stack assumes network header starts at gro_offset=0 6309 * Drivers could call both napi_gro_frags() and napi_gro_receive() 6310 * We copy ethernet header into skb->data to have a common layout. 6311 */ 6312 static struct sk_buff *napi_frags_skb(struct napi_struct *napi) 6313 { 6314 struct sk_buff *skb = napi->skb; 6315 const struct ethhdr *eth; 6316 unsigned int hlen = sizeof(*eth); 6317 6318 napi->skb = NULL; 6319 6320 skb_reset_mac_header(skb); 6321 skb_gro_reset_offset(skb, hlen); 6322 6323 if (unlikely(skb_gro_header_hard(skb, hlen))) { 6324 eth = skb_gro_header_slow(skb, hlen, 0); 6325 if (unlikely(!eth)) { 6326 net_warn_ratelimited("%s: dropping impossible skb from %s\n", 6327 __func__, napi->dev->name); 6328 napi_reuse_skb(napi, skb); 6329 return NULL; 6330 } 6331 } else { 6332 eth = (const struct ethhdr *)skb->data; 6333 gro_pull_from_frag0(skb, hlen); 6334 NAPI_GRO_CB(skb)->frag0 += hlen; 6335 NAPI_GRO_CB(skb)->frag0_len -= hlen; 6336 } 6337 __skb_pull(skb, hlen); 6338 6339 /* 6340 * This works because the only protocols we care about don't require 6341 * special handling. 6342 * We'll fix it up properly in napi_frags_finish() 6343 */ 6344 skb->protocol = eth->h_proto; 6345 6346 return skb; 6347 } 6348 6349 gro_result_t napi_gro_frags(struct napi_struct *napi) 6350 { 6351 gro_result_t ret; 6352 struct sk_buff *skb = napi_frags_skb(napi); 6353 6354 trace_napi_gro_frags_entry(skb); 6355 6356 ret = napi_frags_finish(napi, skb, dev_gro_receive(napi, skb)); 6357 trace_napi_gro_frags_exit(ret); 6358 6359 return ret; 6360 } 6361 EXPORT_SYMBOL(napi_gro_frags); 6362 6363 /* Compute the checksum from gro_offset and return the folded value 6364 * after adding in any pseudo checksum. 6365 */ 6366 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb) 6367 { 6368 __wsum wsum; 6369 __sum16 sum; 6370 6371 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0); 6372 6373 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */ 6374 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum)); 6375 /* See comments in __skb_checksum_complete(). */ 6376 if (likely(!sum)) { 6377 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && 6378 !skb->csum_complete_sw) 6379 netdev_rx_csum_fault(skb->dev, skb); 6380 } 6381 6382 NAPI_GRO_CB(skb)->csum = wsum; 6383 NAPI_GRO_CB(skb)->csum_valid = 1; 6384 6385 return sum; 6386 } 6387 EXPORT_SYMBOL(__skb_gro_checksum_complete); 6388 6389 static void net_rps_send_ipi(struct softnet_data *remsd) 6390 { 6391 #ifdef CONFIG_RPS 6392 while (remsd) { 6393 struct softnet_data *next = remsd->rps_ipi_next; 6394 6395 if (cpu_online(remsd->cpu)) 6396 smp_call_function_single_async(remsd->cpu, &remsd->csd); 6397 remsd = next; 6398 } 6399 #endif 6400 } 6401 6402 /* 6403 * net_rps_action_and_irq_enable sends any pending IPI's for rps. 6404 * Note: called with local irq disabled, but exits with local irq enabled. 6405 */ 6406 static void net_rps_action_and_irq_enable(struct softnet_data *sd) 6407 { 6408 #ifdef CONFIG_RPS 6409 struct softnet_data *remsd = sd->rps_ipi_list; 6410 6411 if (remsd) { 6412 sd->rps_ipi_list = NULL; 6413 6414 local_irq_enable(); 6415 6416 /* Send pending IPI's to kick RPS processing on remote cpus. */ 6417 net_rps_send_ipi(remsd); 6418 } else 6419 #endif 6420 local_irq_enable(); 6421 } 6422 6423 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) 6424 { 6425 #ifdef CONFIG_RPS 6426 return sd->rps_ipi_list != NULL; 6427 #else 6428 return false; 6429 #endif 6430 } 6431 6432 static int process_backlog(struct napi_struct *napi, int quota) 6433 { 6434 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); 6435 bool again = true; 6436 int work = 0; 6437 6438 /* Check if we have pending ipi, its better to send them now, 6439 * not waiting net_rx_action() end. 6440 */ 6441 if (sd_has_rps_ipi_waiting(sd)) { 6442 local_irq_disable(); 6443 net_rps_action_and_irq_enable(sd); 6444 } 6445 6446 napi->weight = dev_rx_weight; 6447 while (again) { 6448 struct sk_buff *skb; 6449 6450 while ((skb = __skb_dequeue(&sd->process_queue))) { 6451 rcu_read_lock(); 6452 __netif_receive_skb(skb); 6453 rcu_read_unlock(); 6454 input_queue_head_incr(sd); 6455 if (++work >= quota) 6456 return work; 6457 6458 } 6459 6460 local_irq_disable(); 6461 rps_lock(sd); 6462 if (skb_queue_empty(&sd->input_pkt_queue)) { 6463 /* 6464 * Inline a custom version of __napi_complete(). 6465 * only current cpu owns and manipulates this napi, 6466 * and NAPI_STATE_SCHED is the only possible flag set 6467 * on backlog. 6468 * We can use a plain write instead of clear_bit(), 6469 * and we dont need an smp_mb() memory barrier. 6470 */ 6471 napi->state = 0; 6472 again = false; 6473 } else { 6474 skb_queue_splice_tail_init(&sd->input_pkt_queue, 6475 &sd->process_queue); 6476 } 6477 rps_unlock(sd); 6478 local_irq_enable(); 6479 } 6480 6481 return work; 6482 } 6483 6484 /** 6485 * __napi_schedule - schedule for receive 6486 * @n: entry to schedule 6487 * 6488 * The entry's receive function will be scheduled to run. 6489 * Consider using __napi_schedule_irqoff() if hard irqs are masked. 6490 */ 6491 void __napi_schedule(struct napi_struct *n) 6492 { 6493 unsigned long flags; 6494 6495 local_irq_save(flags); 6496 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6497 local_irq_restore(flags); 6498 } 6499 EXPORT_SYMBOL(__napi_schedule); 6500 6501 /** 6502 * napi_schedule_prep - check if napi can be scheduled 6503 * @n: napi context 6504 * 6505 * Test if NAPI routine is already running, and if not mark 6506 * it as running. This is used as a condition variable to 6507 * insure only one NAPI poll instance runs. We also make 6508 * sure there is no pending NAPI disable. 6509 */ 6510 bool napi_schedule_prep(struct napi_struct *n) 6511 { 6512 unsigned long val, new; 6513 6514 do { 6515 val = READ_ONCE(n->state); 6516 if (unlikely(val & NAPIF_STATE_DISABLE)) 6517 return false; 6518 new = val | NAPIF_STATE_SCHED; 6519 6520 /* Sets STATE_MISSED bit if STATE_SCHED was already set 6521 * This was suggested by Alexander Duyck, as compiler 6522 * emits better code than : 6523 * if (val & NAPIF_STATE_SCHED) 6524 * new |= NAPIF_STATE_MISSED; 6525 */ 6526 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * 6527 NAPIF_STATE_MISSED; 6528 } while (cmpxchg(&n->state, val, new) != val); 6529 6530 return !(val & NAPIF_STATE_SCHED); 6531 } 6532 EXPORT_SYMBOL(napi_schedule_prep); 6533 6534 /** 6535 * __napi_schedule_irqoff - schedule for receive 6536 * @n: entry to schedule 6537 * 6538 * Variant of __napi_schedule() assuming hard irqs are masked. 6539 * 6540 * On PREEMPT_RT enabled kernels this maps to __napi_schedule() 6541 * because the interrupt disabled assumption might not be true 6542 * due to force-threaded interrupts and spinlock substitution. 6543 */ 6544 void __napi_schedule_irqoff(struct napi_struct *n) 6545 { 6546 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6547 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6548 else 6549 __napi_schedule(n); 6550 } 6551 EXPORT_SYMBOL(__napi_schedule_irqoff); 6552 6553 bool napi_complete_done(struct napi_struct *n, int work_done) 6554 { 6555 unsigned long flags, val, new, timeout = 0; 6556 bool ret = true; 6557 6558 /* 6559 * 1) Don't let napi dequeue from the cpu poll list 6560 * just in case its running on a different cpu. 6561 * 2) If we are busy polling, do nothing here, we have 6562 * the guarantee we will be called later. 6563 */ 6564 if (unlikely(n->state & (NAPIF_STATE_NPSVC | 6565 NAPIF_STATE_IN_BUSY_POLL))) 6566 return false; 6567 6568 if (work_done) { 6569 if (n->gro_bitmask) 6570 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6571 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs); 6572 } 6573 if (n->defer_hard_irqs_count > 0) { 6574 n->defer_hard_irqs_count--; 6575 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6576 if (timeout) 6577 ret = false; 6578 } 6579 if (n->gro_bitmask) { 6580 /* When the NAPI instance uses a timeout and keeps postponing 6581 * it, we need to bound somehow the time packets are kept in 6582 * the GRO layer 6583 */ 6584 napi_gro_flush(n, !!timeout); 6585 } 6586 6587 gro_normal_list(n); 6588 6589 if (unlikely(!list_empty(&n->poll_list))) { 6590 /* If n->poll_list is not empty, we need to mask irqs */ 6591 local_irq_save(flags); 6592 list_del_init(&n->poll_list); 6593 local_irq_restore(flags); 6594 } 6595 6596 do { 6597 val = READ_ONCE(n->state); 6598 6599 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); 6600 6601 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED | 6602 NAPIF_STATE_SCHED_THREADED | 6603 NAPIF_STATE_PREFER_BUSY_POLL); 6604 6605 /* If STATE_MISSED was set, leave STATE_SCHED set, 6606 * because we will call napi->poll() one more time. 6607 * This C code was suggested by Alexander Duyck to help gcc. 6608 */ 6609 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * 6610 NAPIF_STATE_SCHED; 6611 } while (cmpxchg(&n->state, val, new) != val); 6612 6613 if (unlikely(val & NAPIF_STATE_MISSED)) { 6614 __napi_schedule(n); 6615 return false; 6616 } 6617 6618 if (timeout) 6619 hrtimer_start(&n->timer, ns_to_ktime(timeout), 6620 HRTIMER_MODE_REL_PINNED); 6621 return ret; 6622 } 6623 EXPORT_SYMBOL(napi_complete_done); 6624 6625 /* must be called under rcu_read_lock(), as we dont take a reference */ 6626 static struct napi_struct *napi_by_id(unsigned int napi_id) 6627 { 6628 unsigned int hash = napi_id % HASH_SIZE(napi_hash); 6629 struct napi_struct *napi; 6630 6631 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) 6632 if (napi->napi_id == napi_id) 6633 return napi; 6634 6635 return NULL; 6636 } 6637 6638 #if defined(CONFIG_NET_RX_BUSY_POLL) 6639 6640 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule) 6641 { 6642 if (!skip_schedule) { 6643 gro_normal_list(napi); 6644 __napi_schedule(napi); 6645 return; 6646 } 6647 6648 if (napi->gro_bitmask) { 6649 /* flush too old packets 6650 * If HZ < 1000, flush all packets. 6651 */ 6652 napi_gro_flush(napi, HZ >= 1000); 6653 } 6654 6655 gro_normal_list(napi); 6656 clear_bit(NAPI_STATE_SCHED, &napi->state); 6657 } 6658 6659 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll, 6660 u16 budget) 6661 { 6662 bool skip_schedule = false; 6663 unsigned long timeout; 6664 int rc; 6665 6666 /* Busy polling means there is a high chance device driver hard irq 6667 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was 6668 * set in napi_schedule_prep(). 6669 * Since we are about to call napi->poll() once more, we can safely 6670 * clear NAPI_STATE_MISSED. 6671 * 6672 * Note: x86 could use a single "lock and ..." instruction 6673 * to perform these two clear_bit() 6674 */ 6675 clear_bit(NAPI_STATE_MISSED, &napi->state); 6676 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); 6677 6678 local_bh_disable(); 6679 6680 if (prefer_busy_poll) { 6681 napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs); 6682 timeout = READ_ONCE(napi->dev->gro_flush_timeout); 6683 if (napi->defer_hard_irqs_count && timeout) { 6684 hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED); 6685 skip_schedule = true; 6686 } 6687 } 6688 6689 /* All we really want here is to re-enable device interrupts. 6690 * Ideally, a new ndo_busy_poll_stop() could avoid another round. 6691 */ 6692 rc = napi->poll(napi, budget); 6693 /* We can't gro_normal_list() here, because napi->poll() might have 6694 * rearmed the napi (napi_complete_done()) in which case it could 6695 * already be running on another CPU. 6696 */ 6697 trace_napi_poll(napi, rc, budget); 6698 netpoll_poll_unlock(have_poll_lock); 6699 if (rc == budget) 6700 __busy_poll_stop(napi, skip_schedule); 6701 local_bh_enable(); 6702 } 6703 6704 void napi_busy_loop(unsigned int napi_id, 6705 bool (*loop_end)(void *, unsigned long), 6706 void *loop_end_arg, bool prefer_busy_poll, u16 budget) 6707 { 6708 unsigned long start_time = loop_end ? busy_loop_current_time() : 0; 6709 int (*napi_poll)(struct napi_struct *napi, int budget); 6710 void *have_poll_lock = NULL; 6711 struct napi_struct *napi; 6712 6713 restart: 6714 napi_poll = NULL; 6715 6716 rcu_read_lock(); 6717 6718 napi = napi_by_id(napi_id); 6719 if (!napi) 6720 goto out; 6721 6722 preempt_disable(); 6723 for (;;) { 6724 int work = 0; 6725 6726 local_bh_disable(); 6727 if (!napi_poll) { 6728 unsigned long val = READ_ONCE(napi->state); 6729 6730 /* If multiple threads are competing for this napi, 6731 * we avoid dirtying napi->state as much as we can. 6732 */ 6733 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | 6734 NAPIF_STATE_IN_BUSY_POLL)) { 6735 if (prefer_busy_poll) 6736 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6737 goto count; 6738 } 6739 if (cmpxchg(&napi->state, val, 6740 val | NAPIF_STATE_IN_BUSY_POLL | 6741 NAPIF_STATE_SCHED) != val) { 6742 if (prefer_busy_poll) 6743 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6744 goto count; 6745 } 6746 have_poll_lock = netpoll_poll_lock(napi); 6747 napi_poll = napi->poll; 6748 } 6749 work = napi_poll(napi, budget); 6750 trace_napi_poll(napi, work, budget); 6751 gro_normal_list(napi); 6752 count: 6753 if (work > 0) 6754 __NET_ADD_STATS(dev_net(napi->dev), 6755 LINUX_MIB_BUSYPOLLRXPACKETS, work); 6756 local_bh_enable(); 6757 6758 if (!loop_end || loop_end(loop_end_arg, start_time)) 6759 break; 6760 6761 if (unlikely(need_resched())) { 6762 if (napi_poll) 6763 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6764 preempt_enable(); 6765 rcu_read_unlock(); 6766 cond_resched(); 6767 if (loop_end(loop_end_arg, start_time)) 6768 return; 6769 goto restart; 6770 } 6771 cpu_relax(); 6772 } 6773 if (napi_poll) 6774 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6775 preempt_enable(); 6776 out: 6777 rcu_read_unlock(); 6778 } 6779 EXPORT_SYMBOL(napi_busy_loop); 6780 6781 #endif /* CONFIG_NET_RX_BUSY_POLL */ 6782 6783 static void napi_hash_add(struct napi_struct *napi) 6784 { 6785 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state)) 6786 return; 6787 6788 spin_lock(&napi_hash_lock); 6789 6790 /* 0..NR_CPUS range is reserved for sender_cpu use */ 6791 do { 6792 if (unlikely(++napi_gen_id < MIN_NAPI_ID)) 6793 napi_gen_id = MIN_NAPI_ID; 6794 } while (napi_by_id(napi_gen_id)); 6795 napi->napi_id = napi_gen_id; 6796 6797 hlist_add_head_rcu(&napi->napi_hash_node, 6798 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); 6799 6800 spin_unlock(&napi_hash_lock); 6801 } 6802 6803 /* Warning : caller is responsible to make sure rcu grace period 6804 * is respected before freeing memory containing @napi 6805 */ 6806 static void napi_hash_del(struct napi_struct *napi) 6807 { 6808 spin_lock(&napi_hash_lock); 6809 6810 hlist_del_init_rcu(&napi->napi_hash_node); 6811 6812 spin_unlock(&napi_hash_lock); 6813 } 6814 6815 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) 6816 { 6817 struct napi_struct *napi; 6818 6819 napi = container_of(timer, struct napi_struct, timer); 6820 6821 /* Note : we use a relaxed variant of napi_schedule_prep() not setting 6822 * NAPI_STATE_MISSED, since we do not react to a device IRQ. 6823 */ 6824 if (!napi_disable_pending(napi) && 6825 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) { 6826 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6827 __napi_schedule_irqoff(napi); 6828 } 6829 6830 return HRTIMER_NORESTART; 6831 } 6832 6833 static void init_gro_hash(struct napi_struct *napi) 6834 { 6835 int i; 6836 6837 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6838 INIT_LIST_HEAD(&napi->gro_hash[i].list); 6839 napi->gro_hash[i].count = 0; 6840 } 6841 napi->gro_bitmask = 0; 6842 } 6843 6844 int dev_set_threaded(struct net_device *dev, bool threaded) 6845 { 6846 struct napi_struct *napi; 6847 int err = 0; 6848 6849 if (dev->threaded == threaded) 6850 return 0; 6851 6852 if (threaded) { 6853 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6854 if (!napi->thread) { 6855 err = napi_kthread_create(napi); 6856 if (err) { 6857 threaded = false; 6858 break; 6859 } 6860 } 6861 } 6862 } 6863 6864 dev->threaded = threaded; 6865 6866 /* Make sure kthread is created before THREADED bit 6867 * is set. 6868 */ 6869 smp_mb__before_atomic(); 6870 6871 /* Setting/unsetting threaded mode on a napi might not immediately 6872 * take effect, if the current napi instance is actively being 6873 * polled. In this case, the switch between threaded mode and 6874 * softirq mode will happen in the next round of napi_schedule(). 6875 * This should not cause hiccups/stalls to the live traffic. 6876 */ 6877 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6878 if (threaded) 6879 set_bit(NAPI_STATE_THREADED, &napi->state); 6880 else 6881 clear_bit(NAPI_STATE_THREADED, &napi->state); 6882 } 6883 6884 return err; 6885 } 6886 EXPORT_SYMBOL(dev_set_threaded); 6887 6888 void netif_napi_add(struct net_device *dev, struct napi_struct *napi, 6889 int (*poll)(struct napi_struct *, int), int weight) 6890 { 6891 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state))) 6892 return; 6893 6894 INIT_LIST_HEAD(&napi->poll_list); 6895 INIT_HLIST_NODE(&napi->napi_hash_node); 6896 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); 6897 napi->timer.function = napi_watchdog; 6898 init_gro_hash(napi); 6899 napi->skb = NULL; 6900 INIT_LIST_HEAD(&napi->rx_list); 6901 napi->rx_count = 0; 6902 napi->poll = poll; 6903 if (weight > NAPI_POLL_WEIGHT) 6904 netdev_err_once(dev, "%s() called with weight %d\n", __func__, 6905 weight); 6906 napi->weight = weight; 6907 napi->dev = dev; 6908 #ifdef CONFIG_NETPOLL 6909 napi->poll_owner = -1; 6910 #endif 6911 set_bit(NAPI_STATE_SCHED, &napi->state); 6912 set_bit(NAPI_STATE_NPSVC, &napi->state); 6913 list_add_rcu(&napi->dev_list, &dev->napi_list); 6914 napi_hash_add(napi); 6915 /* Create kthread for this napi if dev->threaded is set. 6916 * Clear dev->threaded if kthread creation failed so that 6917 * threaded mode will not be enabled in napi_enable(). 6918 */ 6919 if (dev->threaded && napi_kthread_create(napi)) 6920 dev->threaded = 0; 6921 } 6922 EXPORT_SYMBOL(netif_napi_add); 6923 6924 void napi_disable(struct napi_struct *n) 6925 { 6926 unsigned long val, new; 6927 6928 might_sleep(); 6929 set_bit(NAPI_STATE_DISABLE, &n->state); 6930 6931 for ( ; ; ) { 6932 val = READ_ONCE(n->state); 6933 if (val & (NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC)) { 6934 usleep_range(20, 200); 6935 continue; 6936 } 6937 6938 new = val | NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC; 6939 new &= ~(NAPIF_STATE_THREADED | NAPIF_STATE_PREFER_BUSY_POLL); 6940 6941 if (cmpxchg(&n->state, val, new) == val) 6942 break; 6943 } 6944 6945 hrtimer_cancel(&n->timer); 6946 6947 clear_bit(NAPI_STATE_DISABLE, &n->state); 6948 } 6949 EXPORT_SYMBOL(napi_disable); 6950 6951 /** 6952 * napi_enable - enable NAPI scheduling 6953 * @n: NAPI context 6954 * 6955 * Resume NAPI from being scheduled on this context. 6956 * Must be paired with napi_disable. 6957 */ 6958 void napi_enable(struct napi_struct *n) 6959 { 6960 unsigned long val, new; 6961 6962 do { 6963 val = READ_ONCE(n->state); 6964 BUG_ON(!test_bit(NAPI_STATE_SCHED, &val)); 6965 6966 new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC); 6967 if (n->dev->threaded && n->thread) 6968 new |= NAPIF_STATE_THREADED; 6969 } while (cmpxchg(&n->state, val, new) != val); 6970 } 6971 EXPORT_SYMBOL(napi_enable); 6972 6973 static void flush_gro_hash(struct napi_struct *napi) 6974 { 6975 int i; 6976 6977 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6978 struct sk_buff *skb, *n; 6979 6980 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list) 6981 kfree_skb(skb); 6982 napi->gro_hash[i].count = 0; 6983 } 6984 } 6985 6986 /* Must be called in process context */ 6987 void __netif_napi_del(struct napi_struct *napi) 6988 { 6989 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state)) 6990 return; 6991 6992 napi_hash_del(napi); 6993 list_del_rcu(&napi->dev_list); 6994 napi_free_frags(napi); 6995 6996 flush_gro_hash(napi); 6997 napi->gro_bitmask = 0; 6998 6999 if (napi->thread) { 7000 kthread_stop(napi->thread); 7001 napi->thread = NULL; 7002 } 7003 } 7004 EXPORT_SYMBOL(__netif_napi_del); 7005 7006 static int __napi_poll(struct napi_struct *n, bool *repoll) 7007 { 7008 int work, weight; 7009 7010 weight = n->weight; 7011 7012 /* This NAPI_STATE_SCHED test is for avoiding a race 7013 * with netpoll's poll_napi(). Only the entity which 7014 * obtains the lock and sees NAPI_STATE_SCHED set will 7015 * actually make the ->poll() call. Therefore we avoid 7016 * accidentally calling ->poll() when NAPI is not scheduled. 7017 */ 7018 work = 0; 7019 if (test_bit(NAPI_STATE_SCHED, &n->state)) { 7020 work = n->poll(n, weight); 7021 trace_napi_poll(n, work, weight); 7022 } 7023 7024 if (unlikely(work > weight)) 7025 netdev_err_once(n->dev, "NAPI poll function %pS returned %d, exceeding its budget of %d.\n", 7026 n->poll, work, weight); 7027 7028 if (likely(work < weight)) 7029 return work; 7030 7031 /* Drivers must not modify the NAPI state if they 7032 * consume the entire weight. In such cases this code 7033 * still "owns" the NAPI instance and therefore can 7034 * move the instance around on the list at-will. 7035 */ 7036 if (unlikely(napi_disable_pending(n))) { 7037 napi_complete(n); 7038 return work; 7039 } 7040 7041 /* The NAPI context has more processing work, but busy-polling 7042 * is preferred. Exit early. 7043 */ 7044 if (napi_prefer_busy_poll(n)) { 7045 if (napi_complete_done(n, work)) { 7046 /* If timeout is not set, we need to make sure 7047 * that the NAPI is re-scheduled. 7048 */ 7049 napi_schedule(n); 7050 } 7051 return work; 7052 } 7053 7054 if (n->gro_bitmask) { 7055 /* flush too old packets 7056 * If HZ < 1000, flush all packets. 7057 */ 7058 napi_gro_flush(n, HZ >= 1000); 7059 } 7060 7061 gro_normal_list(n); 7062 7063 /* Some drivers may have called napi_schedule 7064 * prior to exhausting their budget. 7065 */ 7066 if (unlikely(!list_empty(&n->poll_list))) { 7067 pr_warn_once("%s: Budget exhausted after napi rescheduled\n", 7068 n->dev ? n->dev->name : "backlog"); 7069 return work; 7070 } 7071 7072 *repoll = true; 7073 7074 return work; 7075 } 7076 7077 static int napi_poll(struct napi_struct *n, struct list_head *repoll) 7078 { 7079 bool do_repoll = false; 7080 void *have; 7081 int work; 7082 7083 list_del_init(&n->poll_list); 7084 7085 have = netpoll_poll_lock(n); 7086 7087 work = __napi_poll(n, &do_repoll); 7088 7089 if (do_repoll) 7090 list_add_tail(&n->poll_list, repoll); 7091 7092 netpoll_poll_unlock(have); 7093 7094 return work; 7095 } 7096 7097 static int napi_thread_wait(struct napi_struct *napi) 7098 { 7099 bool woken = false; 7100 7101 set_current_state(TASK_INTERRUPTIBLE); 7102 7103 while (!kthread_should_stop()) { 7104 /* Testing SCHED_THREADED bit here to make sure the current 7105 * kthread owns this napi and could poll on this napi. 7106 * Testing SCHED bit is not enough because SCHED bit might be 7107 * set by some other busy poll thread or by napi_disable(). 7108 */ 7109 if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state) || woken) { 7110 WARN_ON(!list_empty(&napi->poll_list)); 7111 __set_current_state(TASK_RUNNING); 7112 return 0; 7113 } 7114 7115 schedule(); 7116 /* woken being true indicates this thread owns this napi. */ 7117 woken = true; 7118 set_current_state(TASK_INTERRUPTIBLE); 7119 } 7120 __set_current_state(TASK_RUNNING); 7121 7122 return -1; 7123 } 7124 7125 static int napi_threaded_poll(void *data) 7126 { 7127 struct napi_struct *napi = data; 7128 void *have; 7129 7130 while (!napi_thread_wait(napi)) { 7131 for (;;) { 7132 bool repoll = false; 7133 7134 local_bh_disable(); 7135 7136 have = netpoll_poll_lock(napi); 7137 __napi_poll(napi, &repoll); 7138 netpoll_poll_unlock(have); 7139 7140 local_bh_enable(); 7141 7142 if (!repoll) 7143 break; 7144 7145 cond_resched(); 7146 } 7147 } 7148 return 0; 7149 } 7150 7151 static __latent_entropy void net_rx_action(struct softirq_action *h) 7152 { 7153 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 7154 unsigned long time_limit = jiffies + 7155 usecs_to_jiffies(netdev_budget_usecs); 7156 int budget = netdev_budget; 7157 LIST_HEAD(list); 7158 LIST_HEAD(repoll); 7159 7160 local_irq_disable(); 7161 list_splice_init(&sd->poll_list, &list); 7162 local_irq_enable(); 7163 7164 for (;;) { 7165 struct napi_struct *n; 7166 7167 if (list_empty(&list)) { 7168 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll)) 7169 return; 7170 break; 7171 } 7172 7173 n = list_first_entry(&list, struct napi_struct, poll_list); 7174 budget -= napi_poll(n, &repoll); 7175 7176 /* If softirq window is exhausted then punt. 7177 * Allow this to run for 2 jiffies since which will allow 7178 * an average latency of 1.5/HZ. 7179 */ 7180 if (unlikely(budget <= 0 || 7181 time_after_eq(jiffies, time_limit))) { 7182 sd->time_squeeze++; 7183 break; 7184 } 7185 } 7186 7187 local_irq_disable(); 7188 7189 list_splice_tail_init(&sd->poll_list, &list); 7190 list_splice_tail(&repoll, &list); 7191 list_splice(&list, &sd->poll_list); 7192 if (!list_empty(&sd->poll_list)) 7193 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 7194 7195 net_rps_action_and_irq_enable(sd); 7196 } 7197 7198 struct netdev_adjacent { 7199 struct net_device *dev; 7200 7201 /* upper master flag, there can only be one master device per list */ 7202 bool master; 7203 7204 /* lookup ignore flag */ 7205 bool ignore; 7206 7207 /* counter for the number of times this device was added to us */ 7208 u16 ref_nr; 7209 7210 /* private field for the users */ 7211 void *private; 7212 7213 struct list_head list; 7214 struct rcu_head rcu; 7215 }; 7216 7217 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, 7218 struct list_head *adj_list) 7219 { 7220 struct netdev_adjacent *adj; 7221 7222 list_for_each_entry(adj, adj_list, list) { 7223 if (adj->dev == adj_dev) 7224 return adj; 7225 } 7226 return NULL; 7227 } 7228 7229 static int ____netdev_has_upper_dev(struct net_device *upper_dev, 7230 struct netdev_nested_priv *priv) 7231 { 7232 struct net_device *dev = (struct net_device *)priv->data; 7233 7234 return upper_dev == dev; 7235 } 7236 7237 /** 7238 * netdev_has_upper_dev - Check if device is linked to an upper device 7239 * @dev: device 7240 * @upper_dev: upper device to check 7241 * 7242 * Find out if a device is linked to specified upper device and return true 7243 * in case it is. Note that this checks only immediate upper device, 7244 * not through a complete stack of devices. The caller must hold the RTNL lock. 7245 */ 7246 bool netdev_has_upper_dev(struct net_device *dev, 7247 struct net_device *upper_dev) 7248 { 7249 struct netdev_nested_priv priv = { 7250 .data = (void *)upper_dev, 7251 }; 7252 7253 ASSERT_RTNL(); 7254 7255 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 7256 &priv); 7257 } 7258 EXPORT_SYMBOL(netdev_has_upper_dev); 7259 7260 /** 7261 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device 7262 * @dev: device 7263 * @upper_dev: upper device to check 7264 * 7265 * Find out if a device is linked to specified upper device and return true 7266 * in case it is. Note that this checks the entire upper device chain. 7267 * The caller must hold rcu lock. 7268 */ 7269 7270 bool netdev_has_upper_dev_all_rcu(struct net_device *dev, 7271 struct net_device *upper_dev) 7272 { 7273 struct netdev_nested_priv priv = { 7274 .data = (void *)upper_dev, 7275 }; 7276 7277 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 7278 &priv); 7279 } 7280 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); 7281 7282 /** 7283 * netdev_has_any_upper_dev - Check if device is linked to some device 7284 * @dev: device 7285 * 7286 * Find out if a device is linked to an upper device and return true in case 7287 * it is. The caller must hold the RTNL lock. 7288 */ 7289 bool netdev_has_any_upper_dev(struct net_device *dev) 7290 { 7291 ASSERT_RTNL(); 7292 7293 return !list_empty(&dev->adj_list.upper); 7294 } 7295 EXPORT_SYMBOL(netdev_has_any_upper_dev); 7296 7297 /** 7298 * netdev_master_upper_dev_get - Get master upper device 7299 * @dev: device 7300 * 7301 * Find a master upper device and return pointer to it or NULL in case 7302 * it's not there. The caller must hold the RTNL lock. 7303 */ 7304 struct net_device *netdev_master_upper_dev_get(struct net_device *dev) 7305 { 7306 struct netdev_adjacent *upper; 7307 7308 ASSERT_RTNL(); 7309 7310 if (list_empty(&dev->adj_list.upper)) 7311 return NULL; 7312 7313 upper = list_first_entry(&dev->adj_list.upper, 7314 struct netdev_adjacent, list); 7315 if (likely(upper->master)) 7316 return upper->dev; 7317 return NULL; 7318 } 7319 EXPORT_SYMBOL(netdev_master_upper_dev_get); 7320 7321 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev) 7322 { 7323 struct netdev_adjacent *upper; 7324 7325 ASSERT_RTNL(); 7326 7327 if (list_empty(&dev->adj_list.upper)) 7328 return NULL; 7329 7330 upper = list_first_entry(&dev->adj_list.upper, 7331 struct netdev_adjacent, list); 7332 if (likely(upper->master) && !upper->ignore) 7333 return upper->dev; 7334 return NULL; 7335 } 7336 7337 /** 7338 * netdev_has_any_lower_dev - Check if device is linked to some device 7339 * @dev: device 7340 * 7341 * Find out if a device is linked to a lower device and return true in case 7342 * it is. The caller must hold the RTNL lock. 7343 */ 7344 static bool netdev_has_any_lower_dev(struct net_device *dev) 7345 { 7346 ASSERT_RTNL(); 7347 7348 return !list_empty(&dev->adj_list.lower); 7349 } 7350 7351 void *netdev_adjacent_get_private(struct list_head *adj_list) 7352 { 7353 struct netdev_adjacent *adj; 7354 7355 adj = list_entry(adj_list, struct netdev_adjacent, list); 7356 7357 return adj->private; 7358 } 7359 EXPORT_SYMBOL(netdev_adjacent_get_private); 7360 7361 /** 7362 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list 7363 * @dev: device 7364 * @iter: list_head ** of the current position 7365 * 7366 * Gets the next device from the dev's upper list, starting from iter 7367 * position. The caller must hold RCU read lock. 7368 */ 7369 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, 7370 struct list_head **iter) 7371 { 7372 struct netdev_adjacent *upper; 7373 7374 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 7375 7376 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7377 7378 if (&upper->list == &dev->adj_list.upper) 7379 return NULL; 7380 7381 *iter = &upper->list; 7382 7383 return upper->dev; 7384 } 7385 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); 7386 7387 static struct net_device *__netdev_next_upper_dev(struct net_device *dev, 7388 struct list_head **iter, 7389 bool *ignore) 7390 { 7391 struct netdev_adjacent *upper; 7392 7393 upper = list_entry((*iter)->next, struct netdev_adjacent, list); 7394 7395 if (&upper->list == &dev->adj_list.upper) 7396 return NULL; 7397 7398 *iter = &upper->list; 7399 *ignore = upper->ignore; 7400 7401 return upper->dev; 7402 } 7403 7404 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, 7405 struct list_head **iter) 7406 { 7407 struct netdev_adjacent *upper; 7408 7409 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 7410 7411 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7412 7413 if (&upper->list == &dev->adj_list.upper) 7414 return NULL; 7415 7416 *iter = &upper->list; 7417 7418 return upper->dev; 7419 } 7420 7421 static int __netdev_walk_all_upper_dev(struct net_device *dev, 7422 int (*fn)(struct net_device *dev, 7423 struct netdev_nested_priv *priv), 7424 struct netdev_nested_priv *priv) 7425 { 7426 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7427 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7428 int ret, cur = 0; 7429 bool ignore; 7430 7431 now = dev; 7432 iter = &dev->adj_list.upper; 7433 7434 while (1) { 7435 if (now != dev) { 7436 ret = fn(now, priv); 7437 if (ret) 7438 return ret; 7439 } 7440 7441 next = NULL; 7442 while (1) { 7443 udev = __netdev_next_upper_dev(now, &iter, &ignore); 7444 if (!udev) 7445 break; 7446 if (ignore) 7447 continue; 7448 7449 next = udev; 7450 niter = &udev->adj_list.upper; 7451 dev_stack[cur] = now; 7452 iter_stack[cur++] = iter; 7453 break; 7454 } 7455 7456 if (!next) { 7457 if (!cur) 7458 return 0; 7459 next = dev_stack[--cur]; 7460 niter = iter_stack[cur]; 7461 } 7462 7463 now = next; 7464 iter = niter; 7465 } 7466 7467 return 0; 7468 } 7469 7470 int netdev_walk_all_upper_dev_rcu(struct net_device *dev, 7471 int (*fn)(struct net_device *dev, 7472 struct netdev_nested_priv *priv), 7473 struct netdev_nested_priv *priv) 7474 { 7475 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7476 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7477 int ret, cur = 0; 7478 7479 now = dev; 7480 iter = &dev->adj_list.upper; 7481 7482 while (1) { 7483 if (now != dev) { 7484 ret = fn(now, priv); 7485 if (ret) 7486 return ret; 7487 } 7488 7489 next = NULL; 7490 while (1) { 7491 udev = netdev_next_upper_dev_rcu(now, &iter); 7492 if (!udev) 7493 break; 7494 7495 next = udev; 7496 niter = &udev->adj_list.upper; 7497 dev_stack[cur] = now; 7498 iter_stack[cur++] = iter; 7499 break; 7500 } 7501 7502 if (!next) { 7503 if (!cur) 7504 return 0; 7505 next = dev_stack[--cur]; 7506 niter = iter_stack[cur]; 7507 } 7508 7509 now = next; 7510 iter = niter; 7511 } 7512 7513 return 0; 7514 } 7515 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); 7516 7517 static bool __netdev_has_upper_dev(struct net_device *dev, 7518 struct net_device *upper_dev) 7519 { 7520 struct netdev_nested_priv priv = { 7521 .flags = 0, 7522 .data = (void *)upper_dev, 7523 }; 7524 7525 ASSERT_RTNL(); 7526 7527 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev, 7528 &priv); 7529 } 7530 7531 /** 7532 * netdev_lower_get_next_private - Get the next ->private from the 7533 * lower neighbour list 7534 * @dev: device 7535 * @iter: list_head ** of the current position 7536 * 7537 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7538 * list, starting from iter position. The caller must hold either hold the 7539 * RTNL lock or its own locking that guarantees that the neighbour lower 7540 * list will remain unchanged. 7541 */ 7542 void *netdev_lower_get_next_private(struct net_device *dev, 7543 struct list_head **iter) 7544 { 7545 struct netdev_adjacent *lower; 7546 7547 lower = list_entry(*iter, struct netdev_adjacent, list); 7548 7549 if (&lower->list == &dev->adj_list.lower) 7550 return NULL; 7551 7552 *iter = lower->list.next; 7553 7554 return lower->private; 7555 } 7556 EXPORT_SYMBOL(netdev_lower_get_next_private); 7557 7558 /** 7559 * netdev_lower_get_next_private_rcu - Get the next ->private from the 7560 * lower neighbour list, RCU 7561 * variant 7562 * @dev: device 7563 * @iter: list_head ** of the current position 7564 * 7565 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7566 * list, starting from iter position. The caller must hold RCU read lock. 7567 */ 7568 void *netdev_lower_get_next_private_rcu(struct net_device *dev, 7569 struct list_head **iter) 7570 { 7571 struct netdev_adjacent *lower; 7572 7573 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 7574 7575 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7576 7577 if (&lower->list == &dev->adj_list.lower) 7578 return NULL; 7579 7580 *iter = &lower->list; 7581 7582 return lower->private; 7583 } 7584 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); 7585 7586 /** 7587 * netdev_lower_get_next - Get the next device from the lower neighbour 7588 * list 7589 * @dev: device 7590 * @iter: list_head ** of the current position 7591 * 7592 * Gets the next netdev_adjacent from the dev's lower neighbour 7593 * list, starting from iter position. The caller must hold RTNL lock or 7594 * its own locking that guarantees that the neighbour lower 7595 * list will remain unchanged. 7596 */ 7597 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) 7598 { 7599 struct netdev_adjacent *lower; 7600 7601 lower = list_entry(*iter, struct netdev_adjacent, list); 7602 7603 if (&lower->list == &dev->adj_list.lower) 7604 return NULL; 7605 7606 *iter = lower->list.next; 7607 7608 return lower->dev; 7609 } 7610 EXPORT_SYMBOL(netdev_lower_get_next); 7611 7612 static struct net_device *netdev_next_lower_dev(struct net_device *dev, 7613 struct list_head **iter) 7614 { 7615 struct netdev_adjacent *lower; 7616 7617 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7618 7619 if (&lower->list == &dev->adj_list.lower) 7620 return NULL; 7621 7622 *iter = &lower->list; 7623 7624 return lower->dev; 7625 } 7626 7627 static struct net_device *__netdev_next_lower_dev(struct net_device *dev, 7628 struct list_head **iter, 7629 bool *ignore) 7630 { 7631 struct netdev_adjacent *lower; 7632 7633 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7634 7635 if (&lower->list == &dev->adj_list.lower) 7636 return NULL; 7637 7638 *iter = &lower->list; 7639 *ignore = lower->ignore; 7640 7641 return lower->dev; 7642 } 7643 7644 int netdev_walk_all_lower_dev(struct net_device *dev, 7645 int (*fn)(struct net_device *dev, 7646 struct netdev_nested_priv *priv), 7647 struct netdev_nested_priv *priv) 7648 { 7649 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7650 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7651 int ret, cur = 0; 7652 7653 now = dev; 7654 iter = &dev->adj_list.lower; 7655 7656 while (1) { 7657 if (now != dev) { 7658 ret = fn(now, priv); 7659 if (ret) 7660 return ret; 7661 } 7662 7663 next = NULL; 7664 while (1) { 7665 ldev = netdev_next_lower_dev(now, &iter); 7666 if (!ldev) 7667 break; 7668 7669 next = ldev; 7670 niter = &ldev->adj_list.lower; 7671 dev_stack[cur] = now; 7672 iter_stack[cur++] = iter; 7673 break; 7674 } 7675 7676 if (!next) { 7677 if (!cur) 7678 return 0; 7679 next = dev_stack[--cur]; 7680 niter = iter_stack[cur]; 7681 } 7682 7683 now = next; 7684 iter = niter; 7685 } 7686 7687 return 0; 7688 } 7689 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); 7690 7691 static int __netdev_walk_all_lower_dev(struct net_device *dev, 7692 int (*fn)(struct net_device *dev, 7693 struct netdev_nested_priv *priv), 7694 struct netdev_nested_priv *priv) 7695 { 7696 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7697 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7698 int ret, cur = 0; 7699 bool ignore; 7700 7701 now = dev; 7702 iter = &dev->adj_list.lower; 7703 7704 while (1) { 7705 if (now != dev) { 7706 ret = fn(now, priv); 7707 if (ret) 7708 return ret; 7709 } 7710 7711 next = NULL; 7712 while (1) { 7713 ldev = __netdev_next_lower_dev(now, &iter, &ignore); 7714 if (!ldev) 7715 break; 7716 if (ignore) 7717 continue; 7718 7719 next = ldev; 7720 niter = &ldev->adj_list.lower; 7721 dev_stack[cur] = now; 7722 iter_stack[cur++] = iter; 7723 break; 7724 } 7725 7726 if (!next) { 7727 if (!cur) 7728 return 0; 7729 next = dev_stack[--cur]; 7730 niter = iter_stack[cur]; 7731 } 7732 7733 now = next; 7734 iter = niter; 7735 } 7736 7737 return 0; 7738 } 7739 7740 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, 7741 struct list_head **iter) 7742 { 7743 struct netdev_adjacent *lower; 7744 7745 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7746 if (&lower->list == &dev->adj_list.lower) 7747 return NULL; 7748 7749 *iter = &lower->list; 7750 7751 return lower->dev; 7752 } 7753 EXPORT_SYMBOL(netdev_next_lower_dev_rcu); 7754 7755 static u8 __netdev_upper_depth(struct net_device *dev) 7756 { 7757 struct net_device *udev; 7758 struct list_head *iter; 7759 u8 max_depth = 0; 7760 bool ignore; 7761 7762 for (iter = &dev->adj_list.upper, 7763 udev = __netdev_next_upper_dev(dev, &iter, &ignore); 7764 udev; 7765 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) { 7766 if (ignore) 7767 continue; 7768 if (max_depth < udev->upper_level) 7769 max_depth = udev->upper_level; 7770 } 7771 7772 return max_depth; 7773 } 7774 7775 static u8 __netdev_lower_depth(struct net_device *dev) 7776 { 7777 struct net_device *ldev; 7778 struct list_head *iter; 7779 u8 max_depth = 0; 7780 bool ignore; 7781 7782 for (iter = &dev->adj_list.lower, 7783 ldev = __netdev_next_lower_dev(dev, &iter, &ignore); 7784 ldev; 7785 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) { 7786 if (ignore) 7787 continue; 7788 if (max_depth < ldev->lower_level) 7789 max_depth = ldev->lower_level; 7790 } 7791 7792 return max_depth; 7793 } 7794 7795 static int __netdev_update_upper_level(struct net_device *dev, 7796 struct netdev_nested_priv *__unused) 7797 { 7798 dev->upper_level = __netdev_upper_depth(dev) + 1; 7799 return 0; 7800 } 7801 7802 static int __netdev_update_lower_level(struct net_device *dev, 7803 struct netdev_nested_priv *priv) 7804 { 7805 dev->lower_level = __netdev_lower_depth(dev) + 1; 7806 7807 #ifdef CONFIG_LOCKDEP 7808 if (!priv) 7809 return 0; 7810 7811 if (priv->flags & NESTED_SYNC_IMM) 7812 dev->nested_level = dev->lower_level - 1; 7813 if (priv->flags & NESTED_SYNC_TODO) 7814 net_unlink_todo(dev); 7815 #endif 7816 return 0; 7817 } 7818 7819 int netdev_walk_all_lower_dev_rcu(struct net_device *dev, 7820 int (*fn)(struct net_device *dev, 7821 struct netdev_nested_priv *priv), 7822 struct netdev_nested_priv *priv) 7823 { 7824 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7825 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7826 int ret, cur = 0; 7827 7828 now = dev; 7829 iter = &dev->adj_list.lower; 7830 7831 while (1) { 7832 if (now != dev) { 7833 ret = fn(now, priv); 7834 if (ret) 7835 return ret; 7836 } 7837 7838 next = NULL; 7839 while (1) { 7840 ldev = netdev_next_lower_dev_rcu(now, &iter); 7841 if (!ldev) 7842 break; 7843 7844 next = ldev; 7845 niter = &ldev->adj_list.lower; 7846 dev_stack[cur] = now; 7847 iter_stack[cur++] = iter; 7848 break; 7849 } 7850 7851 if (!next) { 7852 if (!cur) 7853 return 0; 7854 next = dev_stack[--cur]; 7855 niter = iter_stack[cur]; 7856 } 7857 7858 now = next; 7859 iter = niter; 7860 } 7861 7862 return 0; 7863 } 7864 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); 7865 7866 /** 7867 * netdev_lower_get_first_private_rcu - Get the first ->private from the 7868 * lower neighbour list, RCU 7869 * variant 7870 * @dev: device 7871 * 7872 * Gets the first netdev_adjacent->private from the dev's lower neighbour 7873 * list. The caller must hold RCU read lock. 7874 */ 7875 void *netdev_lower_get_first_private_rcu(struct net_device *dev) 7876 { 7877 struct netdev_adjacent *lower; 7878 7879 lower = list_first_or_null_rcu(&dev->adj_list.lower, 7880 struct netdev_adjacent, list); 7881 if (lower) 7882 return lower->private; 7883 return NULL; 7884 } 7885 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); 7886 7887 /** 7888 * netdev_master_upper_dev_get_rcu - Get master upper device 7889 * @dev: device 7890 * 7891 * Find a master upper device and return pointer to it or NULL in case 7892 * it's not there. The caller must hold the RCU read lock. 7893 */ 7894 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) 7895 { 7896 struct netdev_adjacent *upper; 7897 7898 upper = list_first_or_null_rcu(&dev->adj_list.upper, 7899 struct netdev_adjacent, list); 7900 if (upper && likely(upper->master)) 7901 return upper->dev; 7902 return NULL; 7903 } 7904 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); 7905 7906 static int netdev_adjacent_sysfs_add(struct net_device *dev, 7907 struct net_device *adj_dev, 7908 struct list_head *dev_list) 7909 { 7910 char linkname[IFNAMSIZ+7]; 7911 7912 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7913 "upper_%s" : "lower_%s", adj_dev->name); 7914 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), 7915 linkname); 7916 } 7917 static void netdev_adjacent_sysfs_del(struct net_device *dev, 7918 char *name, 7919 struct list_head *dev_list) 7920 { 7921 char linkname[IFNAMSIZ+7]; 7922 7923 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7924 "upper_%s" : "lower_%s", name); 7925 sysfs_remove_link(&(dev->dev.kobj), linkname); 7926 } 7927 7928 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, 7929 struct net_device *adj_dev, 7930 struct list_head *dev_list) 7931 { 7932 return (dev_list == &dev->adj_list.upper || 7933 dev_list == &dev->adj_list.lower) && 7934 net_eq(dev_net(dev), dev_net(adj_dev)); 7935 } 7936 7937 static int __netdev_adjacent_dev_insert(struct net_device *dev, 7938 struct net_device *adj_dev, 7939 struct list_head *dev_list, 7940 void *private, bool master) 7941 { 7942 struct netdev_adjacent *adj; 7943 int ret; 7944 7945 adj = __netdev_find_adj(adj_dev, dev_list); 7946 7947 if (adj) { 7948 adj->ref_nr += 1; 7949 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", 7950 dev->name, adj_dev->name, adj->ref_nr); 7951 7952 return 0; 7953 } 7954 7955 adj = kmalloc(sizeof(*adj), GFP_KERNEL); 7956 if (!adj) 7957 return -ENOMEM; 7958 7959 adj->dev = adj_dev; 7960 adj->master = master; 7961 adj->ref_nr = 1; 7962 adj->private = private; 7963 adj->ignore = false; 7964 dev_hold(adj_dev); 7965 7966 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", 7967 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); 7968 7969 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { 7970 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); 7971 if (ret) 7972 goto free_adj; 7973 } 7974 7975 /* Ensure that master link is always the first item in list. */ 7976 if (master) { 7977 ret = sysfs_create_link(&(dev->dev.kobj), 7978 &(adj_dev->dev.kobj), "master"); 7979 if (ret) 7980 goto remove_symlinks; 7981 7982 list_add_rcu(&adj->list, dev_list); 7983 } else { 7984 list_add_tail_rcu(&adj->list, dev_list); 7985 } 7986 7987 return 0; 7988 7989 remove_symlinks: 7990 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7991 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7992 free_adj: 7993 kfree(adj); 7994 dev_put(adj_dev); 7995 7996 return ret; 7997 } 7998 7999 static void __netdev_adjacent_dev_remove(struct net_device *dev, 8000 struct net_device *adj_dev, 8001 u16 ref_nr, 8002 struct list_head *dev_list) 8003 { 8004 struct netdev_adjacent *adj; 8005 8006 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", 8007 dev->name, adj_dev->name, ref_nr); 8008 8009 adj = __netdev_find_adj(adj_dev, dev_list); 8010 8011 if (!adj) { 8012 pr_err("Adjacency does not exist for device %s from %s\n", 8013 dev->name, adj_dev->name); 8014 WARN_ON(1); 8015 return; 8016 } 8017 8018 if (adj->ref_nr > ref_nr) { 8019 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", 8020 dev->name, adj_dev->name, ref_nr, 8021 adj->ref_nr - ref_nr); 8022 adj->ref_nr -= ref_nr; 8023 return; 8024 } 8025 8026 if (adj->master) 8027 sysfs_remove_link(&(dev->dev.kobj), "master"); 8028 8029 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 8030 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 8031 8032 list_del_rcu(&adj->list); 8033 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", 8034 adj_dev->name, dev->name, adj_dev->name); 8035 dev_put(adj_dev); 8036 kfree_rcu(adj, rcu); 8037 } 8038 8039 static int __netdev_adjacent_dev_link_lists(struct net_device *dev, 8040 struct net_device *upper_dev, 8041 struct list_head *up_list, 8042 struct list_head *down_list, 8043 void *private, bool master) 8044 { 8045 int ret; 8046 8047 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, 8048 private, master); 8049 if (ret) 8050 return ret; 8051 8052 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, 8053 private, false); 8054 if (ret) { 8055 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); 8056 return ret; 8057 } 8058 8059 return 0; 8060 } 8061 8062 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, 8063 struct net_device *upper_dev, 8064 u16 ref_nr, 8065 struct list_head *up_list, 8066 struct list_head *down_list) 8067 { 8068 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); 8069 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); 8070 } 8071 8072 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, 8073 struct net_device *upper_dev, 8074 void *private, bool master) 8075 { 8076 return __netdev_adjacent_dev_link_lists(dev, upper_dev, 8077 &dev->adj_list.upper, 8078 &upper_dev->adj_list.lower, 8079 private, master); 8080 } 8081 8082 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, 8083 struct net_device *upper_dev) 8084 { 8085 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, 8086 &dev->adj_list.upper, 8087 &upper_dev->adj_list.lower); 8088 } 8089 8090 static int __netdev_upper_dev_link(struct net_device *dev, 8091 struct net_device *upper_dev, bool master, 8092 void *upper_priv, void *upper_info, 8093 struct netdev_nested_priv *priv, 8094 struct netlink_ext_ack *extack) 8095 { 8096 struct netdev_notifier_changeupper_info changeupper_info = { 8097 .info = { 8098 .dev = dev, 8099 .extack = extack, 8100 }, 8101 .upper_dev = upper_dev, 8102 .master = master, 8103 .linking = true, 8104 .upper_info = upper_info, 8105 }; 8106 struct net_device *master_dev; 8107 int ret = 0; 8108 8109 ASSERT_RTNL(); 8110 8111 if (dev == upper_dev) 8112 return -EBUSY; 8113 8114 /* To prevent loops, check if dev is not upper device to upper_dev. */ 8115 if (__netdev_has_upper_dev(upper_dev, dev)) 8116 return -EBUSY; 8117 8118 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV) 8119 return -EMLINK; 8120 8121 if (!master) { 8122 if (__netdev_has_upper_dev(dev, upper_dev)) 8123 return -EEXIST; 8124 } else { 8125 master_dev = __netdev_master_upper_dev_get(dev); 8126 if (master_dev) 8127 return master_dev == upper_dev ? -EEXIST : -EBUSY; 8128 } 8129 8130 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 8131 &changeupper_info.info); 8132 ret = notifier_to_errno(ret); 8133 if (ret) 8134 return ret; 8135 8136 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, 8137 master); 8138 if (ret) 8139 return ret; 8140 8141 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 8142 &changeupper_info.info); 8143 ret = notifier_to_errno(ret); 8144 if (ret) 8145 goto rollback; 8146 8147 __netdev_update_upper_level(dev, NULL); 8148 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 8149 8150 __netdev_update_lower_level(upper_dev, priv); 8151 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 8152 priv); 8153 8154 return 0; 8155 8156 rollback: 8157 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 8158 8159 return ret; 8160 } 8161 8162 /** 8163 * netdev_upper_dev_link - Add a link to the upper device 8164 * @dev: device 8165 * @upper_dev: new upper device 8166 * @extack: netlink extended ack 8167 * 8168 * Adds a link to device which is upper to this one. The caller must hold 8169 * the RTNL lock. On a failure a negative errno code is returned. 8170 * On success the reference counts are adjusted and the function 8171 * returns zero. 8172 */ 8173 int netdev_upper_dev_link(struct net_device *dev, 8174 struct net_device *upper_dev, 8175 struct netlink_ext_ack *extack) 8176 { 8177 struct netdev_nested_priv priv = { 8178 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 8179 .data = NULL, 8180 }; 8181 8182 return __netdev_upper_dev_link(dev, upper_dev, false, 8183 NULL, NULL, &priv, extack); 8184 } 8185 EXPORT_SYMBOL(netdev_upper_dev_link); 8186 8187 /** 8188 * netdev_master_upper_dev_link - Add a master link to the upper device 8189 * @dev: device 8190 * @upper_dev: new upper device 8191 * @upper_priv: upper device private 8192 * @upper_info: upper info to be passed down via notifier 8193 * @extack: netlink extended ack 8194 * 8195 * Adds a link to device which is upper to this one. In this case, only 8196 * one master upper device can be linked, although other non-master devices 8197 * might be linked as well. The caller must hold the RTNL lock. 8198 * On a failure a negative errno code is returned. On success the reference 8199 * counts are adjusted and the function returns zero. 8200 */ 8201 int netdev_master_upper_dev_link(struct net_device *dev, 8202 struct net_device *upper_dev, 8203 void *upper_priv, void *upper_info, 8204 struct netlink_ext_ack *extack) 8205 { 8206 struct netdev_nested_priv priv = { 8207 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 8208 .data = NULL, 8209 }; 8210 8211 return __netdev_upper_dev_link(dev, upper_dev, true, 8212 upper_priv, upper_info, &priv, extack); 8213 } 8214 EXPORT_SYMBOL(netdev_master_upper_dev_link); 8215 8216 static void __netdev_upper_dev_unlink(struct net_device *dev, 8217 struct net_device *upper_dev, 8218 struct netdev_nested_priv *priv) 8219 { 8220 struct netdev_notifier_changeupper_info changeupper_info = { 8221 .info = { 8222 .dev = dev, 8223 }, 8224 .upper_dev = upper_dev, 8225 .linking = false, 8226 }; 8227 8228 ASSERT_RTNL(); 8229 8230 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; 8231 8232 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 8233 &changeupper_info.info); 8234 8235 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 8236 8237 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 8238 &changeupper_info.info); 8239 8240 __netdev_update_upper_level(dev, NULL); 8241 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 8242 8243 __netdev_update_lower_level(upper_dev, priv); 8244 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 8245 priv); 8246 } 8247 8248 /** 8249 * netdev_upper_dev_unlink - Removes a link to upper device 8250 * @dev: device 8251 * @upper_dev: new upper device 8252 * 8253 * Removes a link to device which is upper to this one. The caller must hold 8254 * the RTNL lock. 8255 */ 8256 void netdev_upper_dev_unlink(struct net_device *dev, 8257 struct net_device *upper_dev) 8258 { 8259 struct netdev_nested_priv priv = { 8260 .flags = NESTED_SYNC_TODO, 8261 .data = NULL, 8262 }; 8263 8264 __netdev_upper_dev_unlink(dev, upper_dev, &priv); 8265 } 8266 EXPORT_SYMBOL(netdev_upper_dev_unlink); 8267 8268 static void __netdev_adjacent_dev_set(struct net_device *upper_dev, 8269 struct net_device *lower_dev, 8270 bool val) 8271 { 8272 struct netdev_adjacent *adj; 8273 8274 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower); 8275 if (adj) 8276 adj->ignore = val; 8277 8278 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper); 8279 if (adj) 8280 adj->ignore = val; 8281 } 8282 8283 static void netdev_adjacent_dev_disable(struct net_device *upper_dev, 8284 struct net_device *lower_dev) 8285 { 8286 __netdev_adjacent_dev_set(upper_dev, lower_dev, true); 8287 } 8288 8289 static void netdev_adjacent_dev_enable(struct net_device *upper_dev, 8290 struct net_device *lower_dev) 8291 { 8292 __netdev_adjacent_dev_set(upper_dev, lower_dev, false); 8293 } 8294 8295 int netdev_adjacent_change_prepare(struct net_device *old_dev, 8296 struct net_device *new_dev, 8297 struct net_device *dev, 8298 struct netlink_ext_ack *extack) 8299 { 8300 struct netdev_nested_priv priv = { 8301 .flags = 0, 8302 .data = NULL, 8303 }; 8304 int err; 8305 8306 if (!new_dev) 8307 return 0; 8308 8309 if (old_dev && new_dev != old_dev) 8310 netdev_adjacent_dev_disable(dev, old_dev); 8311 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv, 8312 extack); 8313 if (err) { 8314 if (old_dev && new_dev != old_dev) 8315 netdev_adjacent_dev_enable(dev, old_dev); 8316 return err; 8317 } 8318 8319 return 0; 8320 } 8321 EXPORT_SYMBOL(netdev_adjacent_change_prepare); 8322 8323 void netdev_adjacent_change_commit(struct net_device *old_dev, 8324 struct net_device *new_dev, 8325 struct net_device *dev) 8326 { 8327 struct netdev_nested_priv priv = { 8328 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 8329 .data = NULL, 8330 }; 8331 8332 if (!new_dev || !old_dev) 8333 return; 8334 8335 if (new_dev == old_dev) 8336 return; 8337 8338 netdev_adjacent_dev_enable(dev, old_dev); 8339 __netdev_upper_dev_unlink(old_dev, dev, &priv); 8340 } 8341 EXPORT_SYMBOL(netdev_adjacent_change_commit); 8342 8343 void netdev_adjacent_change_abort(struct net_device *old_dev, 8344 struct net_device *new_dev, 8345 struct net_device *dev) 8346 { 8347 struct netdev_nested_priv priv = { 8348 .flags = 0, 8349 .data = NULL, 8350 }; 8351 8352 if (!new_dev) 8353 return; 8354 8355 if (old_dev && new_dev != old_dev) 8356 netdev_adjacent_dev_enable(dev, old_dev); 8357 8358 __netdev_upper_dev_unlink(new_dev, dev, &priv); 8359 } 8360 EXPORT_SYMBOL(netdev_adjacent_change_abort); 8361 8362 /** 8363 * netdev_bonding_info_change - Dispatch event about slave change 8364 * @dev: device 8365 * @bonding_info: info to dispatch 8366 * 8367 * Send NETDEV_BONDING_INFO to netdev notifiers with info. 8368 * The caller must hold the RTNL lock. 8369 */ 8370 void netdev_bonding_info_change(struct net_device *dev, 8371 struct netdev_bonding_info *bonding_info) 8372 { 8373 struct netdev_notifier_bonding_info info = { 8374 .info.dev = dev, 8375 }; 8376 8377 memcpy(&info.bonding_info, bonding_info, 8378 sizeof(struct netdev_bonding_info)); 8379 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, 8380 &info.info); 8381 } 8382 EXPORT_SYMBOL(netdev_bonding_info_change); 8383 8384 /** 8385 * netdev_get_xmit_slave - Get the xmit slave of master device 8386 * @dev: device 8387 * @skb: The packet 8388 * @all_slaves: assume all the slaves are active 8389 * 8390 * The reference counters are not incremented so the caller must be 8391 * careful with locks. The caller must hold RCU lock. 8392 * %NULL is returned if no slave is found. 8393 */ 8394 8395 struct net_device *netdev_get_xmit_slave(struct net_device *dev, 8396 struct sk_buff *skb, 8397 bool all_slaves) 8398 { 8399 const struct net_device_ops *ops = dev->netdev_ops; 8400 8401 if (!ops->ndo_get_xmit_slave) 8402 return NULL; 8403 return ops->ndo_get_xmit_slave(dev, skb, all_slaves); 8404 } 8405 EXPORT_SYMBOL(netdev_get_xmit_slave); 8406 8407 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev, 8408 struct sock *sk) 8409 { 8410 const struct net_device_ops *ops = dev->netdev_ops; 8411 8412 if (!ops->ndo_sk_get_lower_dev) 8413 return NULL; 8414 return ops->ndo_sk_get_lower_dev(dev, sk); 8415 } 8416 8417 /** 8418 * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket 8419 * @dev: device 8420 * @sk: the socket 8421 * 8422 * %NULL is returned if no lower device is found. 8423 */ 8424 8425 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev, 8426 struct sock *sk) 8427 { 8428 struct net_device *lower; 8429 8430 lower = netdev_sk_get_lower_dev(dev, sk); 8431 while (lower) { 8432 dev = lower; 8433 lower = netdev_sk_get_lower_dev(dev, sk); 8434 } 8435 8436 return dev; 8437 } 8438 EXPORT_SYMBOL(netdev_sk_get_lowest_dev); 8439 8440 static void netdev_adjacent_add_links(struct net_device *dev) 8441 { 8442 struct netdev_adjacent *iter; 8443 8444 struct net *net = dev_net(dev); 8445 8446 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8447 if (!net_eq(net, dev_net(iter->dev))) 8448 continue; 8449 netdev_adjacent_sysfs_add(iter->dev, dev, 8450 &iter->dev->adj_list.lower); 8451 netdev_adjacent_sysfs_add(dev, iter->dev, 8452 &dev->adj_list.upper); 8453 } 8454 8455 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8456 if (!net_eq(net, dev_net(iter->dev))) 8457 continue; 8458 netdev_adjacent_sysfs_add(iter->dev, dev, 8459 &iter->dev->adj_list.upper); 8460 netdev_adjacent_sysfs_add(dev, iter->dev, 8461 &dev->adj_list.lower); 8462 } 8463 } 8464 8465 static void netdev_adjacent_del_links(struct net_device *dev) 8466 { 8467 struct netdev_adjacent *iter; 8468 8469 struct net *net = dev_net(dev); 8470 8471 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8472 if (!net_eq(net, dev_net(iter->dev))) 8473 continue; 8474 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8475 &iter->dev->adj_list.lower); 8476 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8477 &dev->adj_list.upper); 8478 } 8479 8480 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8481 if (!net_eq(net, dev_net(iter->dev))) 8482 continue; 8483 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8484 &iter->dev->adj_list.upper); 8485 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8486 &dev->adj_list.lower); 8487 } 8488 } 8489 8490 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) 8491 { 8492 struct netdev_adjacent *iter; 8493 8494 struct net *net = dev_net(dev); 8495 8496 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8497 if (!net_eq(net, dev_net(iter->dev))) 8498 continue; 8499 netdev_adjacent_sysfs_del(iter->dev, oldname, 8500 &iter->dev->adj_list.lower); 8501 netdev_adjacent_sysfs_add(iter->dev, dev, 8502 &iter->dev->adj_list.lower); 8503 } 8504 8505 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8506 if (!net_eq(net, dev_net(iter->dev))) 8507 continue; 8508 netdev_adjacent_sysfs_del(iter->dev, oldname, 8509 &iter->dev->adj_list.upper); 8510 netdev_adjacent_sysfs_add(iter->dev, dev, 8511 &iter->dev->adj_list.upper); 8512 } 8513 } 8514 8515 void *netdev_lower_dev_get_private(struct net_device *dev, 8516 struct net_device *lower_dev) 8517 { 8518 struct netdev_adjacent *lower; 8519 8520 if (!lower_dev) 8521 return NULL; 8522 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); 8523 if (!lower) 8524 return NULL; 8525 8526 return lower->private; 8527 } 8528 EXPORT_SYMBOL(netdev_lower_dev_get_private); 8529 8530 8531 /** 8532 * netdev_lower_state_changed - Dispatch event about lower device state change 8533 * @lower_dev: device 8534 * @lower_state_info: state to dispatch 8535 * 8536 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. 8537 * The caller must hold the RTNL lock. 8538 */ 8539 void netdev_lower_state_changed(struct net_device *lower_dev, 8540 void *lower_state_info) 8541 { 8542 struct netdev_notifier_changelowerstate_info changelowerstate_info = { 8543 .info.dev = lower_dev, 8544 }; 8545 8546 ASSERT_RTNL(); 8547 changelowerstate_info.lower_state_info = lower_state_info; 8548 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, 8549 &changelowerstate_info.info); 8550 } 8551 EXPORT_SYMBOL(netdev_lower_state_changed); 8552 8553 static void dev_change_rx_flags(struct net_device *dev, int flags) 8554 { 8555 const struct net_device_ops *ops = dev->netdev_ops; 8556 8557 if (ops->ndo_change_rx_flags) 8558 ops->ndo_change_rx_flags(dev, flags); 8559 } 8560 8561 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) 8562 { 8563 unsigned int old_flags = dev->flags; 8564 kuid_t uid; 8565 kgid_t gid; 8566 8567 ASSERT_RTNL(); 8568 8569 dev->flags |= IFF_PROMISC; 8570 dev->promiscuity += inc; 8571 if (dev->promiscuity == 0) { 8572 /* 8573 * Avoid overflow. 8574 * If inc causes overflow, untouch promisc and return error. 8575 */ 8576 if (inc < 0) 8577 dev->flags &= ~IFF_PROMISC; 8578 else { 8579 dev->promiscuity -= inc; 8580 netdev_warn(dev, "promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n"); 8581 return -EOVERFLOW; 8582 } 8583 } 8584 if (dev->flags != old_flags) { 8585 pr_info("device %s %s promiscuous mode\n", 8586 dev->name, 8587 dev->flags & IFF_PROMISC ? "entered" : "left"); 8588 if (audit_enabled) { 8589 current_uid_gid(&uid, &gid); 8590 audit_log(audit_context(), GFP_ATOMIC, 8591 AUDIT_ANOM_PROMISCUOUS, 8592 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", 8593 dev->name, (dev->flags & IFF_PROMISC), 8594 (old_flags & IFF_PROMISC), 8595 from_kuid(&init_user_ns, audit_get_loginuid(current)), 8596 from_kuid(&init_user_ns, uid), 8597 from_kgid(&init_user_ns, gid), 8598 audit_get_sessionid(current)); 8599 } 8600 8601 dev_change_rx_flags(dev, IFF_PROMISC); 8602 } 8603 if (notify) 8604 __dev_notify_flags(dev, old_flags, IFF_PROMISC); 8605 return 0; 8606 } 8607 8608 /** 8609 * dev_set_promiscuity - update promiscuity count on a device 8610 * @dev: device 8611 * @inc: modifier 8612 * 8613 * Add or remove promiscuity from a device. While the count in the device 8614 * remains above zero the interface remains promiscuous. Once it hits zero 8615 * the device reverts back to normal filtering operation. A negative inc 8616 * value is used to drop promiscuity on the device. 8617 * Return 0 if successful or a negative errno code on error. 8618 */ 8619 int dev_set_promiscuity(struct net_device *dev, int inc) 8620 { 8621 unsigned int old_flags = dev->flags; 8622 int err; 8623 8624 err = __dev_set_promiscuity(dev, inc, true); 8625 if (err < 0) 8626 return err; 8627 if (dev->flags != old_flags) 8628 dev_set_rx_mode(dev); 8629 return err; 8630 } 8631 EXPORT_SYMBOL(dev_set_promiscuity); 8632 8633 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) 8634 { 8635 unsigned int old_flags = dev->flags, old_gflags = dev->gflags; 8636 8637 ASSERT_RTNL(); 8638 8639 dev->flags |= IFF_ALLMULTI; 8640 dev->allmulti += inc; 8641 if (dev->allmulti == 0) { 8642 /* 8643 * Avoid overflow. 8644 * If inc causes overflow, untouch allmulti and return error. 8645 */ 8646 if (inc < 0) 8647 dev->flags &= ~IFF_ALLMULTI; 8648 else { 8649 dev->allmulti -= inc; 8650 netdev_warn(dev, "allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n"); 8651 return -EOVERFLOW; 8652 } 8653 } 8654 if (dev->flags ^ old_flags) { 8655 dev_change_rx_flags(dev, IFF_ALLMULTI); 8656 dev_set_rx_mode(dev); 8657 if (notify) 8658 __dev_notify_flags(dev, old_flags, 8659 dev->gflags ^ old_gflags); 8660 } 8661 return 0; 8662 } 8663 8664 /** 8665 * dev_set_allmulti - update allmulti count on a device 8666 * @dev: device 8667 * @inc: modifier 8668 * 8669 * Add or remove reception of all multicast frames to a device. While the 8670 * count in the device remains above zero the interface remains listening 8671 * to all interfaces. Once it hits zero the device reverts back to normal 8672 * filtering operation. A negative @inc value is used to drop the counter 8673 * when releasing a resource needing all multicasts. 8674 * Return 0 if successful or a negative errno code on error. 8675 */ 8676 8677 int dev_set_allmulti(struct net_device *dev, int inc) 8678 { 8679 return __dev_set_allmulti(dev, inc, true); 8680 } 8681 EXPORT_SYMBOL(dev_set_allmulti); 8682 8683 /* 8684 * Upload unicast and multicast address lists to device and 8685 * configure RX filtering. When the device doesn't support unicast 8686 * filtering it is put in promiscuous mode while unicast addresses 8687 * are present. 8688 */ 8689 void __dev_set_rx_mode(struct net_device *dev) 8690 { 8691 const struct net_device_ops *ops = dev->netdev_ops; 8692 8693 /* dev_open will call this function so the list will stay sane. */ 8694 if (!(dev->flags&IFF_UP)) 8695 return; 8696 8697 if (!netif_device_present(dev)) 8698 return; 8699 8700 if (!(dev->priv_flags & IFF_UNICAST_FLT)) { 8701 /* Unicast addresses changes may only happen under the rtnl, 8702 * therefore calling __dev_set_promiscuity here is safe. 8703 */ 8704 if (!netdev_uc_empty(dev) && !dev->uc_promisc) { 8705 __dev_set_promiscuity(dev, 1, false); 8706 dev->uc_promisc = true; 8707 } else if (netdev_uc_empty(dev) && dev->uc_promisc) { 8708 __dev_set_promiscuity(dev, -1, false); 8709 dev->uc_promisc = false; 8710 } 8711 } 8712 8713 if (ops->ndo_set_rx_mode) 8714 ops->ndo_set_rx_mode(dev); 8715 } 8716 8717 void dev_set_rx_mode(struct net_device *dev) 8718 { 8719 netif_addr_lock_bh(dev); 8720 __dev_set_rx_mode(dev); 8721 netif_addr_unlock_bh(dev); 8722 } 8723 8724 /** 8725 * dev_get_flags - get flags reported to userspace 8726 * @dev: device 8727 * 8728 * Get the combination of flag bits exported through APIs to userspace. 8729 */ 8730 unsigned int dev_get_flags(const struct net_device *dev) 8731 { 8732 unsigned int flags; 8733 8734 flags = (dev->flags & ~(IFF_PROMISC | 8735 IFF_ALLMULTI | 8736 IFF_RUNNING | 8737 IFF_LOWER_UP | 8738 IFF_DORMANT)) | 8739 (dev->gflags & (IFF_PROMISC | 8740 IFF_ALLMULTI)); 8741 8742 if (netif_running(dev)) { 8743 if (netif_oper_up(dev)) 8744 flags |= IFF_RUNNING; 8745 if (netif_carrier_ok(dev)) 8746 flags |= IFF_LOWER_UP; 8747 if (netif_dormant(dev)) 8748 flags |= IFF_DORMANT; 8749 } 8750 8751 return flags; 8752 } 8753 EXPORT_SYMBOL(dev_get_flags); 8754 8755 int __dev_change_flags(struct net_device *dev, unsigned int flags, 8756 struct netlink_ext_ack *extack) 8757 { 8758 unsigned int old_flags = dev->flags; 8759 int ret; 8760 8761 ASSERT_RTNL(); 8762 8763 /* 8764 * Set the flags on our device. 8765 */ 8766 8767 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | 8768 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | 8769 IFF_AUTOMEDIA)) | 8770 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | 8771 IFF_ALLMULTI)); 8772 8773 /* 8774 * Load in the correct multicast list now the flags have changed. 8775 */ 8776 8777 if ((old_flags ^ flags) & IFF_MULTICAST) 8778 dev_change_rx_flags(dev, IFF_MULTICAST); 8779 8780 dev_set_rx_mode(dev); 8781 8782 /* 8783 * Have we downed the interface. We handle IFF_UP ourselves 8784 * according to user attempts to set it, rather than blindly 8785 * setting it. 8786 */ 8787 8788 ret = 0; 8789 if ((old_flags ^ flags) & IFF_UP) { 8790 if (old_flags & IFF_UP) 8791 __dev_close(dev); 8792 else 8793 ret = __dev_open(dev, extack); 8794 } 8795 8796 if ((flags ^ dev->gflags) & IFF_PROMISC) { 8797 int inc = (flags & IFF_PROMISC) ? 1 : -1; 8798 unsigned int old_flags = dev->flags; 8799 8800 dev->gflags ^= IFF_PROMISC; 8801 8802 if (__dev_set_promiscuity(dev, inc, false) >= 0) 8803 if (dev->flags != old_flags) 8804 dev_set_rx_mode(dev); 8805 } 8806 8807 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI 8808 * is important. Some (broken) drivers set IFF_PROMISC, when 8809 * IFF_ALLMULTI is requested not asking us and not reporting. 8810 */ 8811 if ((flags ^ dev->gflags) & IFF_ALLMULTI) { 8812 int inc = (flags & IFF_ALLMULTI) ? 1 : -1; 8813 8814 dev->gflags ^= IFF_ALLMULTI; 8815 __dev_set_allmulti(dev, inc, false); 8816 } 8817 8818 return ret; 8819 } 8820 8821 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, 8822 unsigned int gchanges) 8823 { 8824 unsigned int changes = dev->flags ^ old_flags; 8825 8826 if (gchanges) 8827 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC); 8828 8829 if (changes & IFF_UP) { 8830 if (dev->flags & IFF_UP) 8831 call_netdevice_notifiers(NETDEV_UP, dev); 8832 else 8833 call_netdevice_notifiers(NETDEV_DOWN, dev); 8834 } 8835 8836 if (dev->flags & IFF_UP && 8837 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { 8838 struct netdev_notifier_change_info change_info = { 8839 .info = { 8840 .dev = dev, 8841 }, 8842 .flags_changed = changes, 8843 }; 8844 8845 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info); 8846 } 8847 } 8848 8849 /** 8850 * dev_change_flags - change device settings 8851 * @dev: device 8852 * @flags: device state flags 8853 * @extack: netlink extended ack 8854 * 8855 * Change settings on device based state flags. The flags are 8856 * in the userspace exported format. 8857 */ 8858 int dev_change_flags(struct net_device *dev, unsigned int flags, 8859 struct netlink_ext_ack *extack) 8860 { 8861 int ret; 8862 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; 8863 8864 ret = __dev_change_flags(dev, flags, extack); 8865 if (ret < 0) 8866 return ret; 8867 8868 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); 8869 __dev_notify_flags(dev, old_flags, changes); 8870 return ret; 8871 } 8872 EXPORT_SYMBOL(dev_change_flags); 8873 8874 int __dev_set_mtu(struct net_device *dev, int new_mtu) 8875 { 8876 const struct net_device_ops *ops = dev->netdev_ops; 8877 8878 if (ops->ndo_change_mtu) 8879 return ops->ndo_change_mtu(dev, new_mtu); 8880 8881 /* Pairs with all the lockless reads of dev->mtu in the stack */ 8882 WRITE_ONCE(dev->mtu, new_mtu); 8883 return 0; 8884 } 8885 EXPORT_SYMBOL(__dev_set_mtu); 8886 8887 int dev_validate_mtu(struct net_device *dev, int new_mtu, 8888 struct netlink_ext_ack *extack) 8889 { 8890 /* MTU must be positive, and in range */ 8891 if (new_mtu < 0 || new_mtu < dev->min_mtu) { 8892 NL_SET_ERR_MSG(extack, "mtu less than device minimum"); 8893 return -EINVAL; 8894 } 8895 8896 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { 8897 NL_SET_ERR_MSG(extack, "mtu greater than device maximum"); 8898 return -EINVAL; 8899 } 8900 return 0; 8901 } 8902 8903 /** 8904 * dev_set_mtu_ext - Change maximum transfer unit 8905 * @dev: device 8906 * @new_mtu: new transfer unit 8907 * @extack: netlink extended ack 8908 * 8909 * Change the maximum transfer size of the network device. 8910 */ 8911 int dev_set_mtu_ext(struct net_device *dev, int new_mtu, 8912 struct netlink_ext_ack *extack) 8913 { 8914 int err, orig_mtu; 8915 8916 if (new_mtu == dev->mtu) 8917 return 0; 8918 8919 err = dev_validate_mtu(dev, new_mtu, extack); 8920 if (err) 8921 return err; 8922 8923 if (!netif_device_present(dev)) 8924 return -ENODEV; 8925 8926 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); 8927 err = notifier_to_errno(err); 8928 if (err) 8929 return err; 8930 8931 orig_mtu = dev->mtu; 8932 err = __dev_set_mtu(dev, new_mtu); 8933 8934 if (!err) { 8935 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8936 orig_mtu); 8937 err = notifier_to_errno(err); 8938 if (err) { 8939 /* setting mtu back and notifying everyone again, 8940 * so that they have a chance to revert changes. 8941 */ 8942 __dev_set_mtu(dev, orig_mtu); 8943 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8944 new_mtu); 8945 } 8946 } 8947 return err; 8948 } 8949 8950 int dev_set_mtu(struct net_device *dev, int new_mtu) 8951 { 8952 struct netlink_ext_ack extack; 8953 int err; 8954 8955 memset(&extack, 0, sizeof(extack)); 8956 err = dev_set_mtu_ext(dev, new_mtu, &extack); 8957 if (err && extack._msg) 8958 net_err_ratelimited("%s: %s\n", dev->name, extack._msg); 8959 return err; 8960 } 8961 EXPORT_SYMBOL(dev_set_mtu); 8962 8963 /** 8964 * dev_change_tx_queue_len - Change TX queue length of a netdevice 8965 * @dev: device 8966 * @new_len: new tx queue length 8967 */ 8968 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len) 8969 { 8970 unsigned int orig_len = dev->tx_queue_len; 8971 int res; 8972 8973 if (new_len != (unsigned int)new_len) 8974 return -ERANGE; 8975 8976 if (new_len != orig_len) { 8977 dev->tx_queue_len = new_len; 8978 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev); 8979 res = notifier_to_errno(res); 8980 if (res) 8981 goto err_rollback; 8982 res = dev_qdisc_change_tx_queue_len(dev); 8983 if (res) 8984 goto err_rollback; 8985 } 8986 8987 return 0; 8988 8989 err_rollback: 8990 netdev_err(dev, "refused to change device tx_queue_len\n"); 8991 dev->tx_queue_len = orig_len; 8992 return res; 8993 } 8994 8995 /** 8996 * dev_set_group - Change group this device belongs to 8997 * @dev: device 8998 * @new_group: group this device should belong to 8999 */ 9000 void dev_set_group(struct net_device *dev, int new_group) 9001 { 9002 dev->group = new_group; 9003 } 9004 EXPORT_SYMBOL(dev_set_group); 9005 9006 /** 9007 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR. 9008 * @dev: device 9009 * @addr: new address 9010 * @extack: netlink extended ack 9011 */ 9012 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr, 9013 struct netlink_ext_ack *extack) 9014 { 9015 struct netdev_notifier_pre_changeaddr_info info = { 9016 .info.dev = dev, 9017 .info.extack = extack, 9018 .dev_addr = addr, 9019 }; 9020 int rc; 9021 9022 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info); 9023 return notifier_to_errno(rc); 9024 } 9025 EXPORT_SYMBOL(dev_pre_changeaddr_notify); 9026 9027 /** 9028 * dev_set_mac_address - Change Media Access Control Address 9029 * @dev: device 9030 * @sa: new address 9031 * @extack: netlink extended ack 9032 * 9033 * Change the hardware (MAC) address of the device 9034 */ 9035 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa, 9036 struct netlink_ext_ack *extack) 9037 { 9038 const struct net_device_ops *ops = dev->netdev_ops; 9039 int err; 9040 9041 if (!ops->ndo_set_mac_address) 9042 return -EOPNOTSUPP; 9043 if (sa->sa_family != dev->type) 9044 return -EINVAL; 9045 if (!netif_device_present(dev)) 9046 return -ENODEV; 9047 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack); 9048 if (err) 9049 return err; 9050 err = ops->ndo_set_mac_address(dev, sa); 9051 if (err) 9052 return err; 9053 dev->addr_assign_type = NET_ADDR_SET; 9054 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); 9055 add_device_randomness(dev->dev_addr, dev->addr_len); 9056 return 0; 9057 } 9058 EXPORT_SYMBOL(dev_set_mac_address); 9059 9060 static DECLARE_RWSEM(dev_addr_sem); 9061 9062 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa, 9063 struct netlink_ext_ack *extack) 9064 { 9065 int ret; 9066 9067 down_write(&dev_addr_sem); 9068 ret = dev_set_mac_address(dev, sa, extack); 9069 up_write(&dev_addr_sem); 9070 return ret; 9071 } 9072 EXPORT_SYMBOL(dev_set_mac_address_user); 9073 9074 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name) 9075 { 9076 size_t size = sizeof(sa->sa_data); 9077 struct net_device *dev; 9078 int ret = 0; 9079 9080 down_read(&dev_addr_sem); 9081 rcu_read_lock(); 9082 9083 dev = dev_get_by_name_rcu(net, dev_name); 9084 if (!dev) { 9085 ret = -ENODEV; 9086 goto unlock; 9087 } 9088 if (!dev->addr_len) 9089 memset(sa->sa_data, 0, size); 9090 else 9091 memcpy(sa->sa_data, dev->dev_addr, 9092 min_t(size_t, size, dev->addr_len)); 9093 sa->sa_family = dev->type; 9094 9095 unlock: 9096 rcu_read_unlock(); 9097 up_read(&dev_addr_sem); 9098 return ret; 9099 } 9100 EXPORT_SYMBOL(dev_get_mac_address); 9101 9102 /** 9103 * dev_change_carrier - Change device carrier 9104 * @dev: device 9105 * @new_carrier: new value 9106 * 9107 * Change device carrier 9108 */ 9109 int dev_change_carrier(struct net_device *dev, bool new_carrier) 9110 { 9111 const struct net_device_ops *ops = dev->netdev_ops; 9112 9113 if (!ops->ndo_change_carrier) 9114 return -EOPNOTSUPP; 9115 if (!netif_device_present(dev)) 9116 return -ENODEV; 9117 return ops->ndo_change_carrier(dev, new_carrier); 9118 } 9119 EXPORT_SYMBOL(dev_change_carrier); 9120 9121 /** 9122 * dev_get_phys_port_id - Get device physical port ID 9123 * @dev: device 9124 * @ppid: port ID 9125 * 9126 * Get device physical port ID 9127 */ 9128 int dev_get_phys_port_id(struct net_device *dev, 9129 struct netdev_phys_item_id *ppid) 9130 { 9131 const struct net_device_ops *ops = dev->netdev_ops; 9132 9133 if (!ops->ndo_get_phys_port_id) 9134 return -EOPNOTSUPP; 9135 return ops->ndo_get_phys_port_id(dev, ppid); 9136 } 9137 EXPORT_SYMBOL(dev_get_phys_port_id); 9138 9139 /** 9140 * dev_get_phys_port_name - Get device physical port name 9141 * @dev: device 9142 * @name: port name 9143 * @len: limit of bytes to copy to name 9144 * 9145 * Get device physical port name 9146 */ 9147 int dev_get_phys_port_name(struct net_device *dev, 9148 char *name, size_t len) 9149 { 9150 const struct net_device_ops *ops = dev->netdev_ops; 9151 int err; 9152 9153 if (ops->ndo_get_phys_port_name) { 9154 err = ops->ndo_get_phys_port_name(dev, name, len); 9155 if (err != -EOPNOTSUPP) 9156 return err; 9157 } 9158 return devlink_compat_phys_port_name_get(dev, name, len); 9159 } 9160 EXPORT_SYMBOL(dev_get_phys_port_name); 9161 9162 /** 9163 * dev_get_port_parent_id - Get the device's port parent identifier 9164 * @dev: network device 9165 * @ppid: pointer to a storage for the port's parent identifier 9166 * @recurse: allow/disallow recursion to lower devices 9167 * 9168 * Get the devices's port parent identifier 9169 */ 9170 int dev_get_port_parent_id(struct net_device *dev, 9171 struct netdev_phys_item_id *ppid, 9172 bool recurse) 9173 { 9174 const struct net_device_ops *ops = dev->netdev_ops; 9175 struct netdev_phys_item_id first = { }; 9176 struct net_device *lower_dev; 9177 struct list_head *iter; 9178 int err; 9179 9180 if (ops->ndo_get_port_parent_id) { 9181 err = ops->ndo_get_port_parent_id(dev, ppid); 9182 if (err != -EOPNOTSUPP) 9183 return err; 9184 } 9185 9186 err = devlink_compat_switch_id_get(dev, ppid); 9187 if (!recurse || err != -EOPNOTSUPP) 9188 return err; 9189 9190 netdev_for_each_lower_dev(dev, lower_dev, iter) { 9191 err = dev_get_port_parent_id(lower_dev, ppid, true); 9192 if (err) 9193 break; 9194 if (!first.id_len) 9195 first = *ppid; 9196 else if (memcmp(&first, ppid, sizeof(*ppid))) 9197 return -EOPNOTSUPP; 9198 } 9199 9200 return err; 9201 } 9202 EXPORT_SYMBOL(dev_get_port_parent_id); 9203 9204 /** 9205 * netdev_port_same_parent_id - Indicate if two network devices have 9206 * the same port parent identifier 9207 * @a: first network device 9208 * @b: second network device 9209 */ 9210 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b) 9211 { 9212 struct netdev_phys_item_id a_id = { }; 9213 struct netdev_phys_item_id b_id = { }; 9214 9215 if (dev_get_port_parent_id(a, &a_id, true) || 9216 dev_get_port_parent_id(b, &b_id, true)) 9217 return false; 9218 9219 return netdev_phys_item_id_same(&a_id, &b_id); 9220 } 9221 EXPORT_SYMBOL(netdev_port_same_parent_id); 9222 9223 /** 9224 * dev_change_proto_down - update protocol port state information 9225 * @dev: device 9226 * @proto_down: new value 9227 * 9228 * This info can be used by switch drivers to set the phys state of the 9229 * port. 9230 */ 9231 int dev_change_proto_down(struct net_device *dev, bool proto_down) 9232 { 9233 const struct net_device_ops *ops = dev->netdev_ops; 9234 9235 if (!ops->ndo_change_proto_down) 9236 return -EOPNOTSUPP; 9237 if (!netif_device_present(dev)) 9238 return -ENODEV; 9239 return ops->ndo_change_proto_down(dev, proto_down); 9240 } 9241 EXPORT_SYMBOL(dev_change_proto_down); 9242 9243 /** 9244 * dev_change_proto_down_generic - generic implementation for 9245 * ndo_change_proto_down that sets carrier according to 9246 * proto_down. 9247 * 9248 * @dev: device 9249 * @proto_down: new value 9250 */ 9251 int dev_change_proto_down_generic(struct net_device *dev, bool proto_down) 9252 { 9253 if (proto_down) 9254 netif_carrier_off(dev); 9255 else 9256 netif_carrier_on(dev); 9257 dev->proto_down = proto_down; 9258 return 0; 9259 } 9260 EXPORT_SYMBOL(dev_change_proto_down_generic); 9261 9262 /** 9263 * dev_change_proto_down_reason - proto down reason 9264 * 9265 * @dev: device 9266 * @mask: proto down mask 9267 * @value: proto down value 9268 */ 9269 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask, 9270 u32 value) 9271 { 9272 int b; 9273 9274 if (!mask) { 9275 dev->proto_down_reason = value; 9276 } else { 9277 for_each_set_bit(b, &mask, 32) { 9278 if (value & (1 << b)) 9279 dev->proto_down_reason |= BIT(b); 9280 else 9281 dev->proto_down_reason &= ~BIT(b); 9282 } 9283 } 9284 } 9285 EXPORT_SYMBOL(dev_change_proto_down_reason); 9286 9287 struct bpf_xdp_link { 9288 struct bpf_link link; 9289 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */ 9290 int flags; 9291 }; 9292 9293 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags) 9294 { 9295 if (flags & XDP_FLAGS_HW_MODE) 9296 return XDP_MODE_HW; 9297 if (flags & XDP_FLAGS_DRV_MODE) 9298 return XDP_MODE_DRV; 9299 if (flags & XDP_FLAGS_SKB_MODE) 9300 return XDP_MODE_SKB; 9301 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB; 9302 } 9303 9304 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode) 9305 { 9306 switch (mode) { 9307 case XDP_MODE_SKB: 9308 return generic_xdp_install; 9309 case XDP_MODE_DRV: 9310 case XDP_MODE_HW: 9311 return dev->netdev_ops->ndo_bpf; 9312 default: 9313 return NULL; 9314 } 9315 } 9316 9317 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev, 9318 enum bpf_xdp_mode mode) 9319 { 9320 return dev->xdp_state[mode].link; 9321 } 9322 9323 static struct bpf_prog *dev_xdp_prog(struct net_device *dev, 9324 enum bpf_xdp_mode mode) 9325 { 9326 struct bpf_xdp_link *link = dev_xdp_link(dev, mode); 9327 9328 if (link) 9329 return link->link.prog; 9330 return dev->xdp_state[mode].prog; 9331 } 9332 9333 u8 dev_xdp_prog_count(struct net_device *dev) 9334 { 9335 u8 count = 0; 9336 int i; 9337 9338 for (i = 0; i < __MAX_XDP_MODE; i++) 9339 if (dev->xdp_state[i].prog || dev->xdp_state[i].link) 9340 count++; 9341 return count; 9342 } 9343 EXPORT_SYMBOL_GPL(dev_xdp_prog_count); 9344 9345 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode) 9346 { 9347 struct bpf_prog *prog = dev_xdp_prog(dev, mode); 9348 9349 return prog ? prog->aux->id : 0; 9350 } 9351 9352 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode, 9353 struct bpf_xdp_link *link) 9354 { 9355 dev->xdp_state[mode].link = link; 9356 dev->xdp_state[mode].prog = NULL; 9357 } 9358 9359 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode, 9360 struct bpf_prog *prog) 9361 { 9362 dev->xdp_state[mode].link = NULL; 9363 dev->xdp_state[mode].prog = prog; 9364 } 9365 9366 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode, 9367 bpf_op_t bpf_op, struct netlink_ext_ack *extack, 9368 u32 flags, struct bpf_prog *prog) 9369 { 9370 struct netdev_bpf xdp; 9371 int err; 9372 9373 memset(&xdp, 0, sizeof(xdp)); 9374 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG; 9375 xdp.extack = extack; 9376 xdp.flags = flags; 9377 xdp.prog = prog; 9378 9379 /* Drivers assume refcnt is already incremented (i.e, prog pointer is 9380 * "moved" into driver), so they don't increment it on their own, but 9381 * they do decrement refcnt when program is detached or replaced. 9382 * Given net_device also owns link/prog, we need to bump refcnt here 9383 * to prevent drivers from underflowing it. 9384 */ 9385 if (prog) 9386 bpf_prog_inc(prog); 9387 err = bpf_op(dev, &xdp); 9388 if (err) { 9389 if (prog) 9390 bpf_prog_put(prog); 9391 return err; 9392 } 9393 9394 if (mode != XDP_MODE_HW) 9395 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog); 9396 9397 return 0; 9398 } 9399 9400 static void dev_xdp_uninstall(struct net_device *dev) 9401 { 9402 struct bpf_xdp_link *link; 9403 struct bpf_prog *prog; 9404 enum bpf_xdp_mode mode; 9405 bpf_op_t bpf_op; 9406 9407 ASSERT_RTNL(); 9408 9409 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) { 9410 prog = dev_xdp_prog(dev, mode); 9411 if (!prog) 9412 continue; 9413 9414 bpf_op = dev_xdp_bpf_op(dev, mode); 9415 if (!bpf_op) 9416 continue; 9417 9418 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9419 9420 /* auto-detach link from net device */ 9421 link = dev_xdp_link(dev, mode); 9422 if (link) 9423 link->dev = NULL; 9424 else 9425 bpf_prog_put(prog); 9426 9427 dev_xdp_set_link(dev, mode, NULL); 9428 } 9429 } 9430 9431 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack, 9432 struct bpf_xdp_link *link, struct bpf_prog *new_prog, 9433 struct bpf_prog *old_prog, u32 flags) 9434 { 9435 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES); 9436 struct bpf_prog *cur_prog; 9437 struct net_device *upper; 9438 struct list_head *iter; 9439 enum bpf_xdp_mode mode; 9440 bpf_op_t bpf_op; 9441 int err; 9442 9443 ASSERT_RTNL(); 9444 9445 /* either link or prog attachment, never both */ 9446 if (link && (new_prog || old_prog)) 9447 return -EINVAL; 9448 /* link supports only XDP mode flags */ 9449 if (link && (flags & ~XDP_FLAGS_MODES)) { 9450 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment"); 9451 return -EINVAL; 9452 } 9453 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */ 9454 if (num_modes > 1) { 9455 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set"); 9456 return -EINVAL; 9457 } 9458 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */ 9459 if (!num_modes && dev_xdp_prog_count(dev) > 1) { 9460 NL_SET_ERR_MSG(extack, 9461 "More than one program loaded, unset mode is ambiguous"); 9462 return -EINVAL; 9463 } 9464 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */ 9465 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) { 9466 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified"); 9467 return -EINVAL; 9468 } 9469 9470 mode = dev_xdp_mode(dev, flags); 9471 /* can't replace attached link */ 9472 if (dev_xdp_link(dev, mode)) { 9473 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link"); 9474 return -EBUSY; 9475 } 9476 9477 /* don't allow if an upper device already has a program */ 9478 netdev_for_each_upper_dev_rcu(dev, upper, iter) { 9479 if (dev_xdp_prog_count(upper) > 0) { 9480 NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program"); 9481 return -EEXIST; 9482 } 9483 } 9484 9485 cur_prog = dev_xdp_prog(dev, mode); 9486 /* can't replace attached prog with link */ 9487 if (link && cur_prog) { 9488 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link"); 9489 return -EBUSY; 9490 } 9491 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) { 9492 NL_SET_ERR_MSG(extack, "Active program does not match expected"); 9493 return -EEXIST; 9494 } 9495 9496 /* put effective new program into new_prog */ 9497 if (link) 9498 new_prog = link->link.prog; 9499 9500 if (new_prog) { 9501 bool offload = mode == XDP_MODE_HW; 9502 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB 9503 ? XDP_MODE_DRV : XDP_MODE_SKB; 9504 9505 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) { 9506 NL_SET_ERR_MSG(extack, "XDP program already attached"); 9507 return -EBUSY; 9508 } 9509 if (!offload && dev_xdp_prog(dev, other_mode)) { 9510 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time"); 9511 return -EEXIST; 9512 } 9513 if (!offload && bpf_prog_is_dev_bound(new_prog->aux)) { 9514 NL_SET_ERR_MSG(extack, "Using device-bound program without HW_MODE flag is not supported"); 9515 return -EINVAL; 9516 } 9517 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) { 9518 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device"); 9519 return -EINVAL; 9520 } 9521 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) { 9522 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device"); 9523 return -EINVAL; 9524 } 9525 } 9526 9527 /* don't call drivers if the effective program didn't change */ 9528 if (new_prog != cur_prog) { 9529 bpf_op = dev_xdp_bpf_op(dev, mode); 9530 if (!bpf_op) { 9531 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode"); 9532 return -EOPNOTSUPP; 9533 } 9534 9535 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog); 9536 if (err) 9537 return err; 9538 } 9539 9540 if (link) 9541 dev_xdp_set_link(dev, mode, link); 9542 else 9543 dev_xdp_set_prog(dev, mode, new_prog); 9544 if (cur_prog) 9545 bpf_prog_put(cur_prog); 9546 9547 return 0; 9548 } 9549 9550 static int dev_xdp_attach_link(struct net_device *dev, 9551 struct netlink_ext_ack *extack, 9552 struct bpf_xdp_link *link) 9553 { 9554 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags); 9555 } 9556 9557 static int dev_xdp_detach_link(struct net_device *dev, 9558 struct netlink_ext_ack *extack, 9559 struct bpf_xdp_link *link) 9560 { 9561 enum bpf_xdp_mode mode; 9562 bpf_op_t bpf_op; 9563 9564 ASSERT_RTNL(); 9565 9566 mode = dev_xdp_mode(dev, link->flags); 9567 if (dev_xdp_link(dev, mode) != link) 9568 return -EINVAL; 9569 9570 bpf_op = dev_xdp_bpf_op(dev, mode); 9571 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9572 dev_xdp_set_link(dev, mode, NULL); 9573 return 0; 9574 } 9575 9576 static void bpf_xdp_link_release(struct bpf_link *link) 9577 { 9578 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9579 9580 rtnl_lock(); 9581 9582 /* if racing with net_device's tear down, xdp_link->dev might be 9583 * already NULL, in which case link was already auto-detached 9584 */ 9585 if (xdp_link->dev) { 9586 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link)); 9587 xdp_link->dev = NULL; 9588 } 9589 9590 rtnl_unlock(); 9591 } 9592 9593 static int bpf_xdp_link_detach(struct bpf_link *link) 9594 { 9595 bpf_xdp_link_release(link); 9596 return 0; 9597 } 9598 9599 static void bpf_xdp_link_dealloc(struct bpf_link *link) 9600 { 9601 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9602 9603 kfree(xdp_link); 9604 } 9605 9606 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link, 9607 struct seq_file *seq) 9608 { 9609 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9610 u32 ifindex = 0; 9611 9612 rtnl_lock(); 9613 if (xdp_link->dev) 9614 ifindex = xdp_link->dev->ifindex; 9615 rtnl_unlock(); 9616 9617 seq_printf(seq, "ifindex:\t%u\n", ifindex); 9618 } 9619 9620 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link, 9621 struct bpf_link_info *info) 9622 { 9623 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9624 u32 ifindex = 0; 9625 9626 rtnl_lock(); 9627 if (xdp_link->dev) 9628 ifindex = xdp_link->dev->ifindex; 9629 rtnl_unlock(); 9630 9631 info->xdp.ifindex = ifindex; 9632 return 0; 9633 } 9634 9635 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog, 9636 struct bpf_prog *old_prog) 9637 { 9638 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9639 enum bpf_xdp_mode mode; 9640 bpf_op_t bpf_op; 9641 int err = 0; 9642 9643 rtnl_lock(); 9644 9645 /* link might have been auto-released already, so fail */ 9646 if (!xdp_link->dev) { 9647 err = -ENOLINK; 9648 goto out_unlock; 9649 } 9650 9651 if (old_prog && link->prog != old_prog) { 9652 err = -EPERM; 9653 goto out_unlock; 9654 } 9655 old_prog = link->prog; 9656 if (old_prog == new_prog) { 9657 /* no-op, don't disturb drivers */ 9658 bpf_prog_put(new_prog); 9659 goto out_unlock; 9660 } 9661 9662 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags); 9663 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode); 9664 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL, 9665 xdp_link->flags, new_prog); 9666 if (err) 9667 goto out_unlock; 9668 9669 old_prog = xchg(&link->prog, new_prog); 9670 bpf_prog_put(old_prog); 9671 9672 out_unlock: 9673 rtnl_unlock(); 9674 return err; 9675 } 9676 9677 static const struct bpf_link_ops bpf_xdp_link_lops = { 9678 .release = bpf_xdp_link_release, 9679 .dealloc = bpf_xdp_link_dealloc, 9680 .detach = bpf_xdp_link_detach, 9681 .show_fdinfo = bpf_xdp_link_show_fdinfo, 9682 .fill_link_info = bpf_xdp_link_fill_link_info, 9683 .update_prog = bpf_xdp_link_update, 9684 }; 9685 9686 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog) 9687 { 9688 struct net *net = current->nsproxy->net_ns; 9689 struct bpf_link_primer link_primer; 9690 struct bpf_xdp_link *link; 9691 struct net_device *dev; 9692 int err, fd; 9693 9694 rtnl_lock(); 9695 dev = dev_get_by_index(net, attr->link_create.target_ifindex); 9696 if (!dev) { 9697 rtnl_unlock(); 9698 return -EINVAL; 9699 } 9700 9701 link = kzalloc(sizeof(*link), GFP_USER); 9702 if (!link) { 9703 err = -ENOMEM; 9704 goto unlock; 9705 } 9706 9707 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog); 9708 link->dev = dev; 9709 link->flags = attr->link_create.flags; 9710 9711 err = bpf_link_prime(&link->link, &link_primer); 9712 if (err) { 9713 kfree(link); 9714 goto unlock; 9715 } 9716 9717 err = dev_xdp_attach_link(dev, NULL, link); 9718 rtnl_unlock(); 9719 9720 if (err) { 9721 link->dev = NULL; 9722 bpf_link_cleanup(&link_primer); 9723 goto out_put_dev; 9724 } 9725 9726 fd = bpf_link_settle(&link_primer); 9727 /* link itself doesn't hold dev's refcnt to not complicate shutdown */ 9728 dev_put(dev); 9729 return fd; 9730 9731 unlock: 9732 rtnl_unlock(); 9733 9734 out_put_dev: 9735 dev_put(dev); 9736 return err; 9737 } 9738 9739 /** 9740 * dev_change_xdp_fd - set or clear a bpf program for a device rx path 9741 * @dev: device 9742 * @extack: netlink extended ack 9743 * @fd: new program fd or negative value to clear 9744 * @expected_fd: old program fd that userspace expects to replace or clear 9745 * @flags: xdp-related flags 9746 * 9747 * Set or clear a bpf program for a device 9748 */ 9749 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, 9750 int fd, int expected_fd, u32 flags) 9751 { 9752 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags); 9753 struct bpf_prog *new_prog = NULL, *old_prog = NULL; 9754 int err; 9755 9756 ASSERT_RTNL(); 9757 9758 if (fd >= 0) { 9759 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP, 9760 mode != XDP_MODE_SKB); 9761 if (IS_ERR(new_prog)) 9762 return PTR_ERR(new_prog); 9763 } 9764 9765 if (expected_fd >= 0) { 9766 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP, 9767 mode != XDP_MODE_SKB); 9768 if (IS_ERR(old_prog)) { 9769 err = PTR_ERR(old_prog); 9770 old_prog = NULL; 9771 goto err_out; 9772 } 9773 } 9774 9775 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags); 9776 9777 err_out: 9778 if (err && new_prog) 9779 bpf_prog_put(new_prog); 9780 if (old_prog) 9781 bpf_prog_put(old_prog); 9782 return err; 9783 } 9784 9785 /** 9786 * dev_new_index - allocate an ifindex 9787 * @net: the applicable net namespace 9788 * 9789 * Returns a suitable unique value for a new device interface 9790 * number. The caller must hold the rtnl semaphore or the 9791 * dev_base_lock to be sure it remains unique. 9792 */ 9793 static int dev_new_index(struct net *net) 9794 { 9795 int ifindex = net->ifindex; 9796 9797 for (;;) { 9798 if (++ifindex <= 0) 9799 ifindex = 1; 9800 if (!__dev_get_by_index(net, ifindex)) 9801 return net->ifindex = ifindex; 9802 } 9803 } 9804 9805 /* Delayed registration/unregisteration */ 9806 static LIST_HEAD(net_todo_list); 9807 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); 9808 9809 static void net_set_todo(struct net_device *dev) 9810 { 9811 list_add_tail(&dev->todo_list, &net_todo_list); 9812 dev_net(dev)->dev_unreg_count++; 9813 } 9814 9815 static netdev_features_t netdev_sync_upper_features(struct net_device *lower, 9816 struct net_device *upper, netdev_features_t features) 9817 { 9818 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9819 netdev_features_t feature; 9820 int feature_bit; 9821 9822 for_each_netdev_feature(upper_disables, feature_bit) { 9823 feature = __NETIF_F_BIT(feature_bit); 9824 if (!(upper->wanted_features & feature) 9825 && (features & feature)) { 9826 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", 9827 &feature, upper->name); 9828 features &= ~feature; 9829 } 9830 } 9831 9832 return features; 9833 } 9834 9835 static void netdev_sync_lower_features(struct net_device *upper, 9836 struct net_device *lower, netdev_features_t features) 9837 { 9838 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9839 netdev_features_t feature; 9840 int feature_bit; 9841 9842 for_each_netdev_feature(upper_disables, feature_bit) { 9843 feature = __NETIF_F_BIT(feature_bit); 9844 if (!(features & feature) && (lower->features & feature)) { 9845 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", 9846 &feature, lower->name); 9847 lower->wanted_features &= ~feature; 9848 __netdev_update_features(lower); 9849 9850 if (unlikely(lower->features & feature)) 9851 netdev_WARN(upper, "failed to disable %pNF on %s!\n", 9852 &feature, lower->name); 9853 else 9854 netdev_features_change(lower); 9855 } 9856 } 9857 } 9858 9859 static netdev_features_t netdev_fix_features(struct net_device *dev, 9860 netdev_features_t features) 9861 { 9862 /* Fix illegal checksum combinations */ 9863 if ((features & NETIF_F_HW_CSUM) && 9864 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { 9865 netdev_warn(dev, "mixed HW and IP checksum settings.\n"); 9866 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); 9867 } 9868 9869 /* TSO requires that SG is present as well. */ 9870 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { 9871 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); 9872 features &= ~NETIF_F_ALL_TSO; 9873 } 9874 9875 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && 9876 !(features & NETIF_F_IP_CSUM)) { 9877 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); 9878 features &= ~NETIF_F_TSO; 9879 features &= ~NETIF_F_TSO_ECN; 9880 } 9881 9882 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && 9883 !(features & NETIF_F_IPV6_CSUM)) { 9884 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); 9885 features &= ~NETIF_F_TSO6; 9886 } 9887 9888 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ 9889 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) 9890 features &= ~NETIF_F_TSO_MANGLEID; 9891 9892 /* TSO ECN requires that TSO is present as well. */ 9893 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) 9894 features &= ~NETIF_F_TSO_ECN; 9895 9896 /* Software GSO depends on SG. */ 9897 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { 9898 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); 9899 features &= ~NETIF_F_GSO; 9900 } 9901 9902 /* GSO partial features require GSO partial be set */ 9903 if ((features & dev->gso_partial_features) && 9904 !(features & NETIF_F_GSO_PARTIAL)) { 9905 netdev_dbg(dev, 9906 "Dropping partially supported GSO features since no GSO partial.\n"); 9907 features &= ~dev->gso_partial_features; 9908 } 9909 9910 if (!(features & NETIF_F_RXCSUM)) { 9911 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet 9912 * successfully merged by hardware must also have the 9913 * checksum verified by hardware. If the user does not 9914 * want to enable RXCSUM, logically, we should disable GRO_HW. 9915 */ 9916 if (features & NETIF_F_GRO_HW) { 9917 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n"); 9918 features &= ~NETIF_F_GRO_HW; 9919 } 9920 } 9921 9922 /* LRO/HW-GRO features cannot be combined with RX-FCS */ 9923 if (features & NETIF_F_RXFCS) { 9924 if (features & NETIF_F_LRO) { 9925 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n"); 9926 features &= ~NETIF_F_LRO; 9927 } 9928 9929 if (features & NETIF_F_GRO_HW) { 9930 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n"); 9931 features &= ~NETIF_F_GRO_HW; 9932 } 9933 } 9934 9935 if ((features & NETIF_F_GRO_HW) && (features & NETIF_F_LRO)) { 9936 netdev_dbg(dev, "Dropping LRO feature since HW-GRO is requested.\n"); 9937 features &= ~NETIF_F_LRO; 9938 } 9939 9940 if (features & NETIF_F_HW_TLS_TX) { 9941 bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) == 9942 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM); 9943 bool hw_csum = features & NETIF_F_HW_CSUM; 9944 9945 if (!ip_csum && !hw_csum) { 9946 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n"); 9947 features &= ~NETIF_F_HW_TLS_TX; 9948 } 9949 } 9950 9951 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) { 9952 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n"); 9953 features &= ~NETIF_F_HW_TLS_RX; 9954 } 9955 9956 return features; 9957 } 9958 9959 int __netdev_update_features(struct net_device *dev) 9960 { 9961 struct net_device *upper, *lower; 9962 netdev_features_t features; 9963 struct list_head *iter; 9964 int err = -1; 9965 9966 ASSERT_RTNL(); 9967 9968 features = netdev_get_wanted_features(dev); 9969 9970 if (dev->netdev_ops->ndo_fix_features) 9971 features = dev->netdev_ops->ndo_fix_features(dev, features); 9972 9973 /* driver might be less strict about feature dependencies */ 9974 features = netdev_fix_features(dev, features); 9975 9976 /* some features can't be enabled if they're off on an upper device */ 9977 netdev_for_each_upper_dev_rcu(dev, upper, iter) 9978 features = netdev_sync_upper_features(dev, upper, features); 9979 9980 if (dev->features == features) 9981 goto sync_lower; 9982 9983 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", 9984 &dev->features, &features); 9985 9986 if (dev->netdev_ops->ndo_set_features) 9987 err = dev->netdev_ops->ndo_set_features(dev, features); 9988 else 9989 err = 0; 9990 9991 if (unlikely(err < 0)) { 9992 netdev_err(dev, 9993 "set_features() failed (%d); wanted %pNF, left %pNF\n", 9994 err, &features, &dev->features); 9995 /* return non-0 since some features might have changed and 9996 * it's better to fire a spurious notification than miss it 9997 */ 9998 return -1; 9999 } 10000 10001 sync_lower: 10002 /* some features must be disabled on lower devices when disabled 10003 * on an upper device (think: bonding master or bridge) 10004 */ 10005 netdev_for_each_lower_dev(dev, lower, iter) 10006 netdev_sync_lower_features(dev, lower, features); 10007 10008 if (!err) { 10009 netdev_features_t diff = features ^ dev->features; 10010 10011 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) { 10012 /* udp_tunnel_{get,drop}_rx_info both need 10013 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the 10014 * device, or they won't do anything. 10015 * Thus we need to update dev->features 10016 * *before* calling udp_tunnel_get_rx_info, 10017 * but *after* calling udp_tunnel_drop_rx_info. 10018 */ 10019 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) { 10020 dev->features = features; 10021 udp_tunnel_get_rx_info(dev); 10022 } else { 10023 udp_tunnel_drop_rx_info(dev); 10024 } 10025 } 10026 10027 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) { 10028 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) { 10029 dev->features = features; 10030 err |= vlan_get_rx_ctag_filter_info(dev); 10031 } else { 10032 vlan_drop_rx_ctag_filter_info(dev); 10033 } 10034 } 10035 10036 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) { 10037 if (features & NETIF_F_HW_VLAN_STAG_FILTER) { 10038 dev->features = features; 10039 err |= vlan_get_rx_stag_filter_info(dev); 10040 } else { 10041 vlan_drop_rx_stag_filter_info(dev); 10042 } 10043 } 10044 10045 dev->features = features; 10046 } 10047 10048 return err < 0 ? 0 : 1; 10049 } 10050 10051 /** 10052 * netdev_update_features - recalculate device features 10053 * @dev: the device to check 10054 * 10055 * Recalculate dev->features set and send notifications if it 10056 * has changed. Should be called after driver or hardware dependent 10057 * conditions might have changed that influence the features. 10058 */ 10059 void netdev_update_features(struct net_device *dev) 10060 { 10061 if (__netdev_update_features(dev)) 10062 netdev_features_change(dev); 10063 } 10064 EXPORT_SYMBOL(netdev_update_features); 10065 10066 /** 10067 * netdev_change_features - recalculate device features 10068 * @dev: the device to check 10069 * 10070 * Recalculate dev->features set and send notifications even 10071 * if they have not changed. Should be called instead of 10072 * netdev_update_features() if also dev->vlan_features might 10073 * have changed to allow the changes to be propagated to stacked 10074 * VLAN devices. 10075 */ 10076 void netdev_change_features(struct net_device *dev) 10077 { 10078 __netdev_update_features(dev); 10079 netdev_features_change(dev); 10080 } 10081 EXPORT_SYMBOL(netdev_change_features); 10082 10083 /** 10084 * netif_stacked_transfer_operstate - transfer operstate 10085 * @rootdev: the root or lower level device to transfer state from 10086 * @dev: the device to transfer operstate to 10087 * 10088 * Transfer operational state from root to device. This is normally 10089 * called when a stacking relationship exists between the root 10090 * device and the device(a leaf device). 10091 */ 10092 void netif_stacked_transfer_operstate(const struct net_device *rootdev, 10093 struct net_device *dev) 10094 { 10095 if (rootdev->operstate == IF_OPER_DORMANT) 10096 netif_dormant_on(dev); 10097 else 10098 netif_dormant_off(dev); 10099 10100 if (rootdev->operstate == IF_OPER_TESTING) 10101 netif_testing_on(dev); 10102 else 10103 netif_testing_off(dev); 10104 10105 if (netif_carrier_ok(rootdev)) 10106 netif_carrier_on(dev); 10107 else 10108 netif_carrier_off(dev); 10109 } 10110 EXPORT_SYMBOL(netif_stacked_transfer_operstate); 10111 10112 static int netif_alloc_rx_queues(struct net_device *dev) 10113 { 10114 unsigned int i, count = dev->num_rx_queues; 10115 struct netdev_rx_queue *rx; 10116 size_t sz = count * sizeof(*rx); 10117 int err = 0; 10118 10119 BUG_ON(count < 1); 10120 10121 rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10122 if (!rx) 10123 return -ENOMEM; 10124 10125 dev->_rx = rx; 10126 10127 for (i = 0; i < count; i++) { 10128 rx[i].dev = dev; 10129 10130 /* XDP RX-queue setup */ 10131 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0); 10132 if (err < 0) 10133 goto err_rxq_info; 10134 } 10135 return 0; 10136 10137 err_rxq_info: 10138 /* Rollback successful reg's and free other resources */ 10139 while (i--) 10140 xdp_rxq_info_unreg(&rx[i].xdp_rxq); 10141 kvfree(dev->_rx); 10142 dev->_rx = NULL; 10143 return err; 10144 } 10145 10146 static void netif_free_rx_queues(struct net_device *dev) 10147 { 10148 unsigned int i, count = dev->num_rx_queues; 10149 10150 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */ 10151 if (!dev->_rx) 10152 return; 10153 10154 for (i = 0; i < count; i++) 10155 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq); 10156 10157 kvfree(dev->_rx); 10158 } 10159 10160 static void netdev_init_one_queue(struct net_device *dev, 10161 struct netdev_queue *queue, void *_unused) 10162 { 10163 /* Initialize queue lock */ 10164 spin_lock_init(&queue->_xmit_lock); 10165 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); 10166 queue->xmit_lock_owner = -1; 10167 netdev_queue_numa_node_write(queue, NUMA_NO_NODE); 10168 queue->dev = dev; 10169 #ifdef CONFIG_BQL 10170 dql_init(&queue->dql, HZ); 10171 #endif 10172 } 10173 10174 static void netif_free_tx_queues(struct net_device *dev) 10175 { 10176 kvfree(dev->_tx); 10177 } 10178 10179 static int netif_alloc_netdev_queues(struct net_device *dev) 10180 { 10181 unsigned int count = dev->num_tx_queues; 10182 struct netdev_queue *tx; 10183 size_t sz = count * sizeof(*tx); 10184 10185 if (count < 1 || count > 0xffff) 10186 return -EINVAL; 10187 10188 tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10189 if (!tx) 10190 return -ENOMEM; 10191 10192 dev->_tx = tx; 10193 10194 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); 10195 spin_lock_init(&dev->tx_global_lock); 10196 10197 return 0; 10198 } 10199 10200 void netif_tx_stop_all_queues(struct net_device *dev) 10201 { 10202 unsigned int i; 10203 10204 for (i = 0; i < dev->num_tx_queues; i++) { 10205 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 10206 10207 netif_tx_stop_queue(txq); 10208 } 10209 } 10210 EXPORT_SYMBOL(netif_tx_stop_all_queues); 10211 10212 /** 10213 * register_netdevice - register a network device 10214 * @dev: device to register 10215 * 10216 * Take a completed network device structure and add it to the kernel 10217 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 10218 * chain. 0 is returned on success. A negative errno code is returned 10219 * on a failure to set up the device, or if the name is a duplicate. 10220 * 10221 * Callers must hold the rtnl semaphore. You may want 10222 * register_netdev() instead of this. 10223 * 10224 * BUGS: 10225 * The locking appears insufficient to guarantee two parallel registers 10226 * will not get the same name. 10227 */ 10228 10229 int register_netdevice(struct net_device *dev) 10230 { 10231 int ret; 10232 struct net *net = dev_net(dev); 10233 10234 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE < 10235 NETDEV_FEATURE_COUNT); 10236 BUG_ON(dev_boot_phase); 10237 ASSERT_RTNL(); 10238 10239 might_sleep(); 10240 10241 /* When net_device's are persistent, this will be fatal. */ 10242 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); 10243 BUG_ON(!net); 10244 10245 ret = ethtool_check_ops(dev->ethtool_ops); 10246 if (ret) 10247 return ret; 10248 10249 spin_lock_init(&dev->addr_list_lock); 10250 netdev_set_addr_lockdep_class(dev); 10251 10252 ret = dev_get_valid_name(net, dev, dev->name); 10253 if (ret < 0) 10254 goto out; 10255 10256 ret = -ENOMEM; 10257 dev->name_node = netdev_name_node_head_alloc(dev); 10258 if (!dev->name_node) 10259 goto out; 10260 10261 /* Init, if this function is available */ 10262 if (dev->netdev_ops->ndo_init) { 10263 ret = dev->netdev_ops->ndo_init(dev); 10264 if (ret) { 10265 if (ret > 0) 10266 ret = -EIO; 10267 goto err_free_name; 10268 } 10269 } 10270 10271 if (((dev->hw_features | dev->features) & 10272 NETIF_F_HW_VLAN_CTAG_FILTER) && 10273 (!dev->netdev_ops->ndo_vlan_rx_add_vid || 10274 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { 10275 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); 10276 ret = -EINVAL; 10277 goto err_uninit; 10278 } 10279 10280 ret = -EBUSY; 10281 if (!dev->ifindex) 10282 dev->ifindex = dev_new_index(net); 10283 else if (__dev_get_by_index(net, dev->ifindex)) 10284 goto err_uninit; 10285 10286 /* Transfer changeable features to wanted_features and enable 10287 * software offloads (GSO and GRO). 10288 */ 10289 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF); 10290 dev->features |= NETIF_F_SOFT_FEATURES; 10291 10292 if (dev->udp_tunnel_nic_info) { 10293 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10294 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10295 } 10296 10297 dev->wanted_features = dev->features & dev->hw_features; 10298 10299 if (!(dev->flags & IFF_LOOPBACK)) 10300 dev->hw_features |= NETIF_F_NOCACHE_COPY; 10301 10302 /* If IPv4 TCP segmentation offload is supported we should also 10303 * allow the device to enable segmenting the frame with the option 10304 * of ignoring a static IP ID value. This doesn't enable the 10305 * feature itself but allows the user to enable it later. 10306 */ 10307 if (dev->hw_features & NETIF_F_TSO) 10308 dev->hw_features |= NETIF_F_TSO_MANGLEID; 10309 if (dev->vlan_features & NETIF_F_TSO) 10310 dev->vlan_features |= NETIF_F_TSO_MANGLEID; 10311 if (dev->mpls_features & NETIF_F_TSO) 10312 dev->mpls_features |= NETIF_F_TSO_MANGLEID; 10313 if (dev->hw_enc_features & NETIF_F_TSO) 10314 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; 10315 10316 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. 10317 */ 10318 dev->vlan_features |= NETIF_F_HIGHDMA; 10319 10320 /* Make NETIF_F_SG inheritable to tunnel devices. 10321 */ 10322 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; 10323 10324 /* Make NETIF_F_SG inheritable to MPLS. 10325 */ 10326 dev->mpls_features |= NETIF_F_SG; 10327 10328 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); 10329 ret = notifier_to_errno(ret); 10330 if (ret) 10331 goto err_uninit; 10332 10333 ret = netdev_register_kobject(dev); 10334 if (ret) { 10335 dev->reg_state = NETREG_UNREGISTERED; 10336 goto err_uninit; 10337 } 10338 dev->reg_state = NETREG_REGISTERED; 10339 10340 __netdev_update_features(dev); 10341 10342 /* 10343 * Default initial state at registry is that the 10344 * device is present. 10345 */ 10346 10347 set_bit(__LINK_STATE_PRESENT, &dev->state); 10348 10349 linkwatch_init_dev(dev); 10350 10351 dev_init_scheduler(dev); 10352 dev_hold(dev); 10353 list_netdevice(dev); 10354 add_device_randomness(dev->dev_addr, dev->addr_len); 10355 10356 /* If the device has permanent device address, driver should 10357 * set dev_addr and also addr_assign_type should be set to 10358 * NET_ADDR_PERM (default value). 10359 */ 10360 if (dev->addr_assign_type == NET_ADDR_PERM) 10361 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); 10362 10363 /* Notify protocols, that a new device appeared. */ 10364 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); 10365 ret = notifier_to_errno(ret); 10366 if (ret) { 10367 /* Expect explicit free_netdev() on failure */ 10368 dev->needs_free_netdev = false; 10369 unregister_netdevice_queue(dev, NULL); 10370 goto out; 10371 } 10372 /* 10373 * Prevent userspace races by waiting until the network 10374 * device is fully setup before sending notifications. 10375 */ 10376 if (!dev->rtnl_link_ops || 10377 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 10378 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); 10379 10380 out: 10381 return ret; 10382 10383 err_uninit: 10384 if (dev->netdev_ops->ndo_uninit) 10385 dev->netdev_ops->ndo_uninit(dev); 10386 if (dev->priv_destructor) 10387 dev->priv_destructor(dev); 10388 err_free_name: 10389 netdev_name_node_free(dev->name_node); 10390 goto out; 10391 } 10392 EXPORT_SYMBOL(register_netdevice); 10393 10394 /** 10395 * init_dummy_netdev - init a dummy network device for NAPI 10396 * @dev: device to init 10397 * 10398 * This takes a network device structure and initialize the minimum 10399 * amount of fields so it can be used to schedule NAPI polls without 10400 * registering a full blown interface. This is to be used by drivers 10401 * that need to tie several hardware interfaces to a single NAPI 10402 * poll scheduler due to HW limitations. 10403 */ 10404 int init_dummy_netdev(struct net_device *dev) 10405 { 10406 /* Clear everything. Note we don't initialize spinlocks 10407 * are they aren't supposed to be taken by any of the 10408 * NAPI code and this dummy netdev is supposed to be 10409 * only ever used for NAPI polls 10410 */ 10411 memset(dev, 0, sizeof(struct net_device)); 10412 10413 /* make sure we BUG if trying to hit standard 10414 * register/unregister code path 10415 */ 10416 dev->reg_state = NETREG_DUMMY; 10417 10418 /* NAPI wants this */ 10419 INIT_LIST_HEAD(&dev->napi_list); 10420 10421 /* a dummy interface is started by default */ 10422 set_bit(__LINK_STATE_PRESENT, &dev->state); 10423 set_bit(__LINK_STATE_START, &dev->state); 10424 10425 /* napi_busy_loop stats accounting wants this */ 10426 dev_net_set(dev, &init_net); 10427 10428 /* Note : We dont allocate pcpu_refcnt for dummy devices, 10429 * because users of this 'device' dont need to change 10430 * its refcount. 10431 */ 10432 10433 return 0; 10434 } 10435 EXPORT_SYMBOL_GPL(init_dummy_netdev); 10436 10437 10438 /** 10439 * register_netdev - register a network device 10440 * @dev: device to register 10441 * 10442 * Take a completed network device structure and add it to the kernel 10443 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 10444 * chain. 0 is returned on success. A negative errno code is returned 10445 * on a failure to set up the device, or if the name is a duplicate. 10446 * 10447 * This is a wrapper around register_netdevice that takes the rtnl semaphore 10448 * and expands the device name if you passed a format string to 10449 * alloc_netdev. 10450 */ 10451 int register_netdev(struct net_device *dev) 10452 { 10453 int err; 10454 10455 if (rtnl_lock_killable()) 10456 return -EINTR; 10457 err = register_netdevice(dev); 10458 rtnl_unlock(); 10459 return err; 10460 } 10461 EXPORT_SYMBOL(register_netdev); 10462 10463 int netdev_refcnt_read(const struct net_device *dev) 10464 { 10465 #ifdef CONFIG_PCPU_DEV_REFCNT 10466 int i, refcnt = 0; 10467 10468 for_each_possible_cpu(i) 10469 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); 10470 return refcnt; 10471 #else 10472 return refcount_read(&dev->dev_refcnt); 10473 #endif 10474 } 10475 EXPORT_SYMBOL(netdev_refcnt_read); 10476 10477 int netdev_unregister_timeout_secs __read_mostly = 10; 10478 10479 #define WAIT_REFS_MIN_MSECS 1 10480 #define WAIT_REFS_MAX_MSECS 250 10481 /** 10482 * netdev_wait_allrefs - wait until all references are gone. 10483 * @dev: target net_device 10484 * 10485 * This is called when unregistering network devices. 10486 * 10487 * Any protocol or device that holds a reference should register 10488 * for netdevice notification, and cleanup and put back the 10489 * reference if they receive an UNREGISTER event. 10490 * We can get stuck here if buggy protocols don't correctly 10491 * call dev_put. 10492 */ 10493 static void netdev_wait_allrefs(struct net_device *dev) 10494 { 10495 unsigned long rebroadcast_time, warning_time; 10496 int wait = 0, refcnt; 10497 10498 linkwatch_forget_dev(dev); 10499 10500 rebroadcast_time = warning_time = jiffies; 10501 refcnt = netdev_refcnt_read(dev); 10502 10503 while (refcnt != 1) { 10504 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { 10505 rtnl_lock(); 10506 10507 /* Rebroadcast unregister notification */ 10508 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10509 10510 __rtnl_unlock(); 10511 rcu_barrier(); 10512 rtnl_lock(); 10513 10514 if (test_bit(__LINK_STATE_LINKWATCH_PENDING, 10515 &dev->state)) { 10516 /* We must not have linkwatch events 10517 * pending on unregister. If this 10518 * happens, we simply run the queue 10519 * unscheduled, resulting in a noop 10520 * for this device. 10521 */ 10522 linkwatch_run_queue(); 10523 } 10524 10525 __rtnl_unlock(); 10526 10527 rebroadcast_time = jiffies; 10528 } 10529 10530 if (!wait) { 10531 rcu_barrier(); 10532 wait = WAIT_REFS_MIN_MSECS; 10533 } else { 10534 msleep(wait); 10535 wait = min(wait << 1, WAIT_REFS_MAX_MSECS); 10536 } 10537 10538 refcnt = netdev_refcnt_read(dev); 10539 10540 if (refcnt != 1 && 10541 time_after(jiffies, warning_time + 10542 netdev_unregister_timeout_secs * HZ)) { 10543 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", 10544 dev->name, refcnt); 10545 warning_time = jiffies; 10546 } 10547 } 10548 } 10549 10550 /* The sequence is: 10551 * 10552 * rtnl_lock(); 10553 * ... 10554 * register_netdevice(x1); 10555 * register_netdevice(x2); 10556 * ... 10557 * unregister_netdevice(y1); 10558 * unregister_netdevice(y2); 10559 * ... 10560 * rtnl_unlock(); 10561 * free_netdev(y1); 10562 * free_netdev(y2); 10563 * 10564 * We are invoked by rtnl_unlock(). 10565 * This allows us to deal with problems: 10566 * 1) We can delete sysfs objects which invoke hotplug 10567 * without deadlocking with linkwatch via keventd. 10568 * 2) Since we run with the RTNL semaphore not held, we can sleep 10569 * safely in order to wait for the netdev refcnt to drop to zero. 10570 * 10571 * We must not return until all unregister events added during 10572 * the interval the lock was held have been completed. 10573 */ 10574 void netdev_run_todo(void) 10575 { 10576 struct list_head list; 10577 #ifdef CONFIG_LOCKDEP 10578 struct list_head unlink_list; 10579 10580 list_replace_init(&net_unlink_list, &unlink_list); 10581 10582 while (!list_empty(&unlink_list)) { 10583 struct net_device *dev = list_first_entry(&unlink_list, 10584 struct net_device, 10585 unlink_list); 10586 list_del_init(&dev->unlink_list); 10587 dev->nested_level = dev->lower_level - 1; 10588 } 10589 #endif 10590 10591 /* Snapshot list, allow later requests */ 10592 list_replace_init(&net_todo_list, &list); 10593 10594 __rtnl_unlock(); 10595 10596 10597 /* Wait for rcu callbacks to finish before next phase */ 10598 if (!list_empty(&list)) 10599 rcu_barrier(); 10600 10601 while (!list_empty(&list)) { 10602 struct net_device *dev 10603 = list_first_entry(&list, struct net_device, todo_list); 10604 list_del(&dev->todo_list); 10605 10606 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { 10607 pr_err("network todo '%s' but state %d\n", 10608 dev->name, dev->reg_state); 10609 dump_stack(); 10610 continue; 10611 } 10612 10613 dev->reg_state = NETREG_UNREGISTERED; 10614 10615 netdev_wait_allrefs(dev); 10616 10617 /* paranoia */ 10618 BUG_ON(netdev_refcnt_read(dev) != 1); 10619 BUG_ON(!list_empty(&dev->ptype_all)); 10620 BUG_ON(!list_empty(&dev->ptype_specific)); 10621 WARN_ON(rcu_access_pointer(dev->ip_ptr)); 10622 WARN_ON(rcu_access_pointer(dev->ip6_ptr)); 10623 #if IS_ENABLED(CONFIG_DECNET) 10624 WARN_ON(dev->dn_ptr); 10625 #endif 10626 if (dev->priv_destructor) 10627 dev->priv_destructor(dev); 10628 if (dev->needs_free_netdev) 10629 free_netdev(dev); 10630 10631 /* Report a network device has been unregistered */ 10632 rtnl_lock(); 10633 dev_net(dev)->dev_unreg_count--; 10634 __rtnl_unlock(); 10635 wake_up(&netdev_unregistering_wq); 10636 10637 /* Free network device */ 10638 kobject_put(&dev->dev.kobj); 10639 } 10640 } 10641 10642 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has 10643 * all the same fields in the same order as net_device_stats, with only 10644 * the type differing, but rtnl_link_stats64 may have additional fields 10645 * at the end for newer counters. 10646 */ 10647 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, 10648 const struct net_device_stats *netdev_stats) 10649 { 10650 #if BITS_PER_LONG == 64 10651 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats)); 10652 memcpy(stats64, netdev_stats, sizeof(*netdev_stats)); 10653 /* zero out counters that only exist in rtnl_link_stats64 */ 10654 memset((char *)stats64 + sizeof(*netdev_stats), 0, 10655 sizeof(*stats64) - sizeof(*netdev_stats)); 10656 #else 10657 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long); 10658 const unsigned long *src = (const unsigned long *)netdev_stats; 10659 u64 *dst = (u64 *)stats64; 10660 10661 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); 10662 for (i = 0; i < n; i++) 10663 dst[i] = src[i]; 10664 /* zero out counters that only exist in rtnl_link_stats64 */ 10665 memset((char *)stats64 + n * sizeof(u64), 0, 10666 sizeof(*stats64) - n * sizeof(u64)); 10667 #endif 10668 } 10669 EXPORT_SYMBOL(netdev_stats_to_stats64); 10670 10671 /** 10672 * dev_get_stats - get network device statistics 10673 * @dev: device to get statistics from 10674 * @storage: place to store stats 10675 * 10676 * Get network statistics from device. Return @storage. 10677 * The device driver may provide its own method by setting 10678 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; 10679 * otherwise the internal statistics structure is used. 10680 */ 10681 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, 10682 struct rtnl_link_stats64 *storage) 10683 { 10684 const struct net_device_ops *ops = dev->netdev_ops; 10685 10686 if (ops->ndo_get_stats64) { 10687 memset(storage, 0, sizeof(*storage)); 10688 ops->ndo_get_stats64(dev, storage); 10689 } else if (ops->ndo_get_stats) { 10690 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); 10691 } else { 10692 netdev_stats_to_stats64(storage, &dev->stats); 10693 } 10694 storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped); 10695 storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped); 10696 storage->rx_nohandler += (unsigned long)atomic_long_read(&dev->rx_nohandler); 10697 return storage; 10698 } 10699 EXPORT_SYMBOL(dev_get_stats); 10700 10701 /** 10702 * dev_fetch_sw_netstats - get per-cpu network device statistics 10703 * @s: place to store stats 10704 * @netstats: per-cpu network stats to read from 10705 * 10706 * Read per-cpu network statistics and populate the related fields in @s. 10707 */ 10708 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s, 10709 const struct pcpu_sw_netstats __percpu *netstats) 10710 { 10711 int cpu; 10712 10713 for_each_possible_cpu(cpu) { 10714 const struct pcpu_sw_netstats *stats; 10715 struct pcpu_sw_netstats tmp; 10716 unsigned int start; 10717 10718 stats = per_cpu_ptr(netstats, cpu); 10719 do { 10720 start = u64_stats_fetch_begin_irq(&stats->syncp); 10721 tmp.rx_packets = stats->rx_packets; 10722 tmp.rx_bytes = stats->rx_bytes; 10723 tmp.tx_packets = stats->tx_packets; 10724 tmp.tx_bytes = stats->tx_bytes; 10725 } while (u64_stats_fetch_retry_irq(&stats->syncp, start)); 10726 10727 s->rx_packets += tmp.rx_packets; 10728 s->rx_bytes += tmp.rx_bytes; 10729 s->tx_packets += tmp.tx_packets; 10730 s->tx_bytes += tmp.tx_bytes; 10731 } 10732 } 10733 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats); 10734 10735 /** 10736 * dev_get_tstats64 - ndo_get_stats64 implementation 10737 * @dev: device to get statistics from 10738 * @s: place to store stats 10739 * 10740 * Populate @s from dev->stats and dev->tstats. Can be used as 10741 * ndo_get_stats64() callback. 10742 */ 10743 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s) 10744 { 10745 netdev_stats_to_stats64(s, &dev->stats); 10746 dev_fetch_sw_netstats(s, dev->tstats); 10747 } 10748 EXPORT_SYMBOL_GPL(dev_get_tstats64); 10749 10750 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) 10751 { 10752 struct netdev_queue *queue = dev_ingress_queue(dev); 10753 10754 #ifdef CONFIG_NET_CLS_ACT 10755 if (queue) 10756 return queue; 10757 queue = kzalloc(sizeof(*queue), GFP_KERNEL); 10758 if (!queue) 10759 return NULL; 10760 netdev_init_one_queue(dev, queue, NULL); 10761 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); 10762 queue->qdisc_sleeping = &noop_qdisc; 10763 rcu_assign_pointer(dev->ingress_queue, queue); 10764 #endif 10765 return queue; 10766 } 10767 10768 static const struct ethtool_ops default_ethtool_ops; 10769 10770 void netdev_set_default_ethtool_ops(struct net_device *dev, 10771 const struct ethtool_ops *ops) 10772 { 10773 if (dev->ethtool_ops == &default_ethtool_ops) 10774 dev->ethtool_ops = ops; 10775 } 10776 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); 10777 10778 void netdev_freemem(struct net_device *dev) 10779 { 10780 char *addr = (char *)dev - dev->padded; 10781 10782 kvfree(addr); 10783 } 10784 10785 /** 10786 * alloc_netdev_mqs - allocate network device 10787 * @sizeof_priv: size of private data to allocate space for 10788 * @name: device name format string 10789 * @name_assign_type: origin of device name 10790 * @setup: callback to initialize device 10791 * @txqs: the number of TX subqueues to allocate 10792 * @rxqs: the number of RX subqueues to allocate 10793 * 10794 * Allocates a struct net_device with private data area for driver use 10795 * and performs basic initialization. Also allocates subqueue structs 10796 * for each queue on the device. 10797 */ 10798 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, 10799 unsigned char name_assign_type, 10800 void (*setup)(struct net_device *), 10801 unsigned int txqs, unsigned int rxqs) 10802 { 10803 struct net_device *dev; 10804 unsigned int alloc_size; 10805 struct net_device *p; 10806 10807 BUG_ON(strlen(name) >= sizeof(dev->name)); 10808 10809 if (txqs < 1) { 10810 pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); 10811 return NULL; 10812 } 10813 10814 if (rxqs < 1) { 10815 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); 10816 return NULL; 10817 } 10818 10819 alloc_size = sizeof(struct net_device); 10820 if (sizeof_priv) { 10821 /* ensure 32-byte alignment of private area */ 10822 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); 10823 alloc_size += sizeof_priv; 10824 } 10825 /* ensure 32-byte alignment of whole construct */ 10826 alloc_size += NETDEV_ALIGN - 1; 10827 10828 p = kvzalloc(alloc_size, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10829 if (!p) 10830 return NULL; 10831 10832 dev = PTR_ALIGN(p, NETDEV_ALIGN); 10833 dev->padded = (char *)dev - (char *)p; 10834 10835 #ifdef CONFIG_PCPU_DEV_REFCNT 10836 dev->pcpu_refcnt = alloc_percpu(int); 10837 if (!dev->pcpu_refcnt) 10838 goto free_dev; 10839 dev_hold(dev); 10840 #else 10841 refcount_set(&dev->dev_refcnt, 1); 10842 #endif 10843 10844 if (dev_addr_init(dev)) 10845 goto free_pcpu; 10846 10847 dev_mc_init(dev); 10848 dev_uc_init(dev); 10849 10850 dev_net_set(dev, &init_net); 10851 10852 dev->gso_max_size = GSO_MAX_SIZE; 10853 dev->gso_max_segs = GSO_MAX_SEGS; 10854 dev->upper_level = 1; 10855 dev->lower_level = 1; 10856 #ifdef CONFIG_LOCKDEP 10857 dev->nested_level = 0; 10858 INIT_LIST_HEAD(&dev->unlink_list); 10859 #endif 10860 10861 INIT_LIST_HEAD(&dev->napi_list); 10862 INIT_LIST_HEAD(&dev->unreg_list); 10863 INIT_LIST_HEAD(&dev->close_list); 10864 INIT_LIST_HEAD(&dev->link_watch_list); 10865 INIT_LIST_HEAD(&dev->adj_list.upper); 10866 INIT_LIST_HEAD(&dev->adj_list.lower); 10867 INIT_LIST_HEAD(&dev->ptype_all); 10868 INIT_LIST_HEAD(&dev->ptype_specific); 10869 INIT_LIST_HEAD(&dev->net_notifier_list); 10870 #ifdef CONFIG_NET_SCHED 10871 hash_init(dev->qdisc_hash); 10872 #endif 10873 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; 10874 setup(dev); 10875 10876 if (!dev->tx_queue_len) { 10877 dev->priv_flags |= IFF_NO_QUEUE; 10878 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; 10879 } 10880 10881 dev->num_tx_queues = txqs; 10882 dev->real_num_tx_queues = txqs; 10883 if (netif_alloc_netdev_queues(dev)) 10884 goto free_all; 10885 10886 dev->num_rx_queues = rxqs; 10887 dev->real_num_rx_queues = rxqs; 10888 if (netif_alloc_rx_queues(dev)) 10889 goto free_all; 10890 10891 strcpy(dev->name, name); 10892 dev->name_assign_type = name_assign_type; 10893 dev->group = INIT_NETDEV_GROUP; 10894 if (!dev->ethtool_ops) 10895 dev->ethtool_ops = &default_ethtool_ops; 10896 10897 nf_hook_netdev_init(dev); 10898 10899 return dev; 10900 10901 free_all: 10902 free_netdev(dev); 10903 return NULL; 10904 10905 free_pcpu: 10906 #ifdef CONFIG_PCPU_DEV_REFCNT 10907 free_percpu(dev->pcpu_refcnt); 10908 free_dev: 10909 #endif 10910 netdev_freemem(dev); 10911 return NULL; 10912 } 10913 EXPORT_SYMBOL(alloc_netdev_mqs); 10914 10915 /** 10916 * free_netdev - free network device 10917 * @dev: device 10918 * 10919 * This function does the last stage of destroying an allocated device 10920 * interface. The reference to the device object is released. If this 10921 * is the last reference then it will be freed.Must be called in process 10922 * context. 10923 */ 10924 void free_netdev(struct net_device *dev) 10925 { 10926 struct napi_struct *p, *n; 10927 10928 might_sleep(); 10929 10930 /* When called immediately after register_netdevice() failed the unwind 10931 * handling may still be dismantling the device. Handle that case by 10932 * deferring the free. 10933 */ 10934 if (dev->reg_state == NETREG_UNREGISTERING) { 10935 ASSERT_RTNL(); 10936 dev->needs_free_netdev = true; 10937 return; 10938 } 10939 10940 netif_free_tx_queues(dev); 10941 netif_free_rx_queues(dev); 10942 10943 kfree(rcu_dereference_protected(dev->ingress_queue, 1)); 10944 10945 /* Flush device addresses */ 10946 dev_addr_flush(dev); 10947 10948 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) 10949 netif_napi_del(p); 10950 10951 #ifdef CONFIG_PCPU_DEV_REFCNT 10952 free_percpu(dev->pcpu_refcnt); 10953 dev->pcpu_refcnt = NULL; 10954 #endif 10955 free_percpu(dev->xdp_bulkq); 10956 dev->xdp_bulkq = NULL; 10957 10958 /* Compatibility with error handling in drivers */ 10959 if (dev->reg_state == NETREG_UNINITIALIZED) { 10960 netdev_freemem(dev); 10961 return; 10962 } 10963 10964 BUG_ON(dev->reg_state != NETREG_UNREGISTERED); 10965 dev->reg_state = NETREG_RELEASED; 10966 10967 /* will free via device release */ 10968 put_device(&dev->dev); 10969 } 10970 EXPORT_SYMBOL(free_netdev); 10971 10972 /** 10973 * synchronize_net - Synchronize with packet receive processing 10974 * 10975 * Wait for packets currently being received to be done. 10976 * Does not block later packets from starting. 10977 */ 10978 void synchronize_net(void) 10979 { 10980 might_sleep(); 10981 if (rtnl_is_locked()) 10982 synchronize_rcu_expedited(); 10983 else 10984 synchronize_rcu(); 10985 } 10986 EXPORT_SYMBOL(synchronize_net); 10987 10988 /** 10989 * unregister_netdevice_queue - remove device from the kernel 10990 * @dev: device 10991 * @head: list 10992 * 10993 * This function shuts down a device interface and removes it 10994 * from the kernel tables. 10995 * If head not NULL, device is queued to be unregistered later. 10996 * 10997 * Callers must hold the rtnl semaphore. You may want 10998 * unregister_netdev() instead of this. 10999 */ 11000 11001 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) 11002 { 11003 ASSERT_RTNL(); 11004 11005 if (head) { 11006 list_move_tail(&dev->unreg_list, head); 11007 } else { 11008 LIST_HEAD(single); 11009 11010 list_add(&dev->unreg_list, &single); 11011 unregister_netdevice_many(&single); 11012 } 11013 } 11014 EXPORT_SYMBOL(unregister_netdevice_queue); 11015 11016 /** 11017 * unregister_netdevice_many - unregister many devices 11018 * @head: list of devices 11019 * 11020 * Note: As most callers use a stack allocated list_head, 11021 * we force a list_del() to make sure stack wont be corrupted later. 11022 */ 11023 void unregister_netdevice_many(struct list_head *head) 11024 { 11025 struct net_device *dev, *tmp; 11026 LIST_HEAD(close_head); 11027 11028 BUG_ON(dev_boot_phase); 11029 ASSERT_RTNL(); 11030 11031 if (list_empty(head)) 11032 return; 11033 11034 list_for_each_entry_safe(dev, tmp, head, unreg_list) { 11035 /* Some devices call without registering 11036 * for initialization unwind. Remove those 11037 * devices and proceed with the remaining. 11038 */ 11039 if (dev->reg_state == NETREG_UNINITIALIZED) { 11040 pr_debug("unregister_netdevice: device %s/%p never was registered\n", 11041 dev->name, dev); 11042 11043 WARN_ON(1); 11044 list_del(&dev->unreg_list); 11045 continue; 11046 } 11047 dev->dismantle = true; 11048 BUG_ON(dev->reg_state != NETREG_REGISTERED); 11049 } 11050 11051 /* If device is running, close it first. */ 11052 list_for_each_entry(dev, head, unreg_list) 11053 list_add_tail(&dev->close_list, &close_head); 11054 dev_close_many(&close_head, true); 11055 11056 list_for_each_entry(dev, head, unreg_list) { 11057 /* And unlink it from device chain. */ 11058 unlist_netdevice(dev); 11059 11060 dev->reg_state = NETREG_UNREGISTERING; 11061 } 11062 flush_all_backlogs(); 11063 11064 synchronize_net(); 11065 11066 list_for_each_entry(dev, head, unreg_list) { 11067 struct sk_buff *skb = NULL; 11068 11069 /* Shutdown queueing discipline. */ 11070 dev_shutdown(dev); 11071 11072 dev_xdp_uninstall(dev); 11073 11074 /* Notify protocols, that we are about to destroy 11075 * this device. They should clean all the things. 11076 */ 11077 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 11078 11079 if (!dev->rtnl_link_ops || 11080 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 11081 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0, 11082 GFP_KERNEL, NULL, 0); 11083 11084 /* 11085 * Flush the unicast and multicast chains 11086 */ 11087 dev_uc_flush(dev); 11088 dev_mc_flush(dev); 11089 11090 netdev_name_node_alt_flush(dev); 11091 netdev_name_node_free(dev->name_node); 11092 11093 if (dev->netdev_ops->ndo_uninit) 11094 dev->netdev_ops->ndo_uninit(dev); 11095 11096 if (skb) 11097 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL); 11098 11099 /* Notifier chain MUST detach us all upper devices. */ 11100 WARN_ON(netdev_has_any_upper_dev(dev)); 11101 WARN_ON(netdev_has_any_lower_dev(dev)); 11102 11103 /* Remove entries from kobject tree */ 11104 netdev_unregister_kobject(dev); 11105 #ifdef CONFIG_XPS 11106 /* Remove XPS queueing entries */ 11107 netif_reset_xps_queues_gt(dev, 0); 11108 #endif 11109 } 11110 11111 synchronize_net(); 11112 11113 list_for_each_entry(dev, head, unreg_list) { 11114 dev_put(dev); 11115 net_set_todo(dev); 11116 } 11117 11118 list_del(head); 11119 } 11120 EXPORT_SYMBOL(unregister_netdevice_many); 11121 11122 /** 11123 * unregister_netdev - remove device from the kernel 11124 * @dev: device 11125 * 11126 * This function shuts down a device interface and removes it 11127 * from the kernel tables. 11128 * 11129 * This is just a wrapper for unregister_netdevice that takes 11130 * the rtnl semaphore. In general you want to use this and not 11131 * unregister_netdevice. 11132 */ 11133 void unregister_netdev(struct net_device *dev) 11134 { 11135 rtnl_lock(); 11136 unregister_netdevice(dev); 11137 rtnl_unlock(); 11138 } 11139 EXPORT_SYMBOL(unregister_netdev); 11140 11141 /** 11142 * __dev_change_net_namespace - move device to different nethost namespace 11143 * @dev: device 11144 * @net: network namespace 11145 * @pat: If not NULL name pattern to try if the current device name 11146 * is already taken in the destination network namespace. 11147 * @new_ifindex: If not zero, specifies device index in the target 11148 * namespace. 11149 * 11150 * This function shuts down a device interface and moves it 11151 * to a new network namespace. On success 0 is returned, on 11152 * a failure a netagive errno code is returned. 11153 * 11154 * Callers must hold the rtnl semaphore. 11155 */ 11156 11157 int __dev_change_net_namespace(struct net_device *dev, struct net *net, 11158 const char *pat, int new_ifindex) 11159 { 11160 struct net *net_old = dev_net(dev); 11161 int err, new_nsid; 11162 11163 ASSERT_RTNL(); 11164 11165 /* Don't allow namespace local devices to be moved. */ 11166 err = -EINVAL; 11167 if (dev->features & NETIF_F_NETNS_LOCAL) 11168 goto out; 11169 11170 /* Ensure the device has been registrered */ 11171 if (dev->reg_state != NETREG_REGISTERED) 11172 goto out; 11173 11174 /* Get out if there is nothing todo */ 11175 err = 0; 11176 if (net_eq(net_old, net)) 11177 goto out; 11178 11179 /* Pick the destination device name, and ensure 11180 * we can use it in the destination network namespace. 11181 */ 11182 err = -EEXIST; 11183 if (netdev_name_in_use(net, dev->name)) { 11184 /* We get here if we can't use the current device name */ 11185 if (!pat) 11186 goto out; 11187 err = dev_get_valid_name(net, dev, pat); 11188 if (err < 0) 11189 goto out; 11190 } 11191 11192 /* Check that new_ifindex isn't used yet. */ 11193 err = -EBUSY; 11194 if (new_ifindex && __dev_get_by_index(net, new_ifindex)) 11195 goto out; 11196 11197 /* 11198 * And now a mini version of register_netdevice unregister_netdevice. 11199 */ 11200 11201 /* If device is running close it first. */ 11202 dev_close(dev); 11203 11204 /* And unlink it from device chain */ 11205 unlist_netdevice(dev); 11206 11207 synchronize_net(); 11208 11209 /* Shutdown queueing discipline. */ 11210 dev_shutdown(dev); 11211 11212 /* Notify protocols, that we are about to destroy 11213 * this device. They should clean all the things. 11214 * 11215 * Note that dev->reg_state stays at NETREG_REGISTERED. 11216 * This is wanted because this way 8021q and macvlan know 11217 * the device is just moving and can keep their slaves up. 11218 */ 11219 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 11220 rcu_barrier(); 11221 11222 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL); 11223 /* If there is an ifindex conflict assign a new one */ 11224 if (!new_ifindex) { 11225 if (__dev_get_by_index(net, dev->ifindex)) 11226 new_ifindex = dev_new_index(net); 11227 else 11228 new_ifindex = dev->ifindex; 11229 } 11230 11231 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid, 11232 new_ifindex); 11233 11234 /* 11235 * Flush the unicast and multicast chains 11236 */ 11237 dev_uc_flush(dev); 11238 dev_mc_flush(dev); 11239 11240 /* Send a netdev-removed uevent to the old namespace */ 11241 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); 11242 netdev_adjacent_del_links(dev); 11243 11244 /* Move per-net netdevice notifiers that are following the netdevice */ 11245 move_netdevice_notifiers_dev_net(dev, net); 11246 11247 /* Actually switch the network namespace */ 11248 dev_net_set(dev, net); 11249 dev->ifindex = new_ifindex; 11250 11251 /* Send a netdev-add uevent to the new namespace */ 11252 kobject_uevent(&dev->dev.kobj, KOBJ_ADD); 11253 netdev_adjacent_add_links(dev); 11254 11255 /* Fixup kobjects */ 11256 err = device_rename(&dev->dev, dev->name); 11257 WARN_ON(err); 11258 11259 /* Adapt owner in case owning user namespace of target network 11260 * namespace is different from the original one. 11261 */ 11262 err = netdev_change_owner(dev, net_old, net); 11263 WARN_ON(err); 11264 11265 /* Add the device back in the hashes */ 11266 list_netdevice(dev); 11267 11268 /* Notify protocols, that a new device appeared. */ 11269 call_netdevice_notifiers(NETDEV_REGISTER, dev); 11270 11271 /* 11272 * Prevent userspace races by waiting until the network 11273 * device is fully setup before sending notifications. 11274 */ 11275 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); 11276 11277 synchronize_net(); 11278 err = 0; 11279 out: 11280 return err; 11281 } 11282 EXPORT_SYMBOL_GPL(__dev_change_net_namespace); 11283 11284 static int dev_cpu_dead(unsigned int oldcpu) 11285 { 11286 struct sk_buff **list_skb; 11287 struct sk_buff *skb; 11288 unsigned int cpu; 11289 struct softnet_data *sd, *oldsd, *remsd = NULL; 11290 11291 local_irq_disable(); 11292 cpu = smp_processor_id(); 11293 sd = &per_cpu(softnet_data, cpu); 11294 oldsd = &per_cpu(softnet_data, oldcpu); 11295 11296 /* Find end of our completion_queue. */ 11297 list_skb = &sd->completion_queue; 11298 while (*list_skb) 11299 list_skb = &(*list_skb)->next; 11300 /* Append completion queue from offline CPU. */ 11301 *list_skb = oldsd->completion_queue; 11302 oldsd->completion_queue = NULL; 11303 11304 /* Append output queue from offline CPU. */ 11305 if (oldsd->output_queue) { 11306 *sd->output_queue_tailp = oldsd->output_queue; 11307 sd->output_queue_tailp = oldsd->output_queue_tailp; 11308 oldsd->output_queue = NULL; 11309 oldsd->output_queue_tailp = &oldsd->output_queue; 11310 } 11311 /* Append NAPI poll list from offline CPU, with one exception : 11312 * process_backlog() must be called by cpu owning percpu backlog. 11313 * We properly handle process_queue & input_pkt_queue later. 11314 */ 11315 while (!list_empty(&oldsd->poll_list)) { 11316 struct napi_struct *napi = list_first_entry(&oldsd->poll_list, 11317 struct napi_struct, 11318 poll_list); 11319 11320 list_del_init(&napi->poll_list); 11321 if (napi->poll == process_backlog) 11322 napi->state = 0; 11323 else 11324 ____napi_schedule(sd, napi); 11325 } 11326 11327 raise_softirq_irqoff(NET_TX_SOFTIRQ); 11328 local_irq_enable(); 11329 11330 #ifdef CONFIG_RPS 11331 remsd = oldsd->rps_ipi_list; 11332 oldsd->rps_ipi_list = NULL; 11333 #endif 11334 /* send out pending IPI's on offline CPU */ 11335 net_rps_send_ipi(remsd); 11336 11337 /* Process offline CPU's input_pkt_queue */ 11338 while ((skb = __skb_dequeue(&oldsd->process_queue))) { 11339 netif_rx_ni(skb); 11340 input_queue_head_incr(oldsd); 11341 } 11342 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { 11343 netif_rx_ni(skb); 11344 input_queue_head_incr(oldsd); 11345 } 11346 11347 return 0; 11348 } 11349 11350 /** 11351 * netdev_increment_features - increment feature set by one 11352 * @all: current feature set 11353 * @one: new feature set 11354 * @mask: mask feature set 11355 * 11356 * Computes a new feature set after adding a device with feature set 11357 * @one to the master device with current feature set @all. Will not 11358 * enable anything that is off in @mask. Returns the new feature set. 11359 */ 11360 netdev_features_t netdev_increment_features(netdev_features_t all, 11361 netdev_features_t one, netdev_features_t mask) 11362 { 11363 if (mask & NETIF_F_HW_CSUM) 11364 mask |= NETIF_F_CSUM_MASK; 11365 mask |= NETIF_F_VLAN_CHALLENGED; 11366 11367 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; 11368 all &= one | ~NETIF_F_ALL_FOR_ALL; 11369 11370 /* If one device supports hw checksumming, set for all. */ 11371 if (all & NETIF_F_HW_CSUM) 11372 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); 11373 11374 return all; 11375 } 11376 EXPORT_SYMBOL(netdev_increment_features); 11377 11378 static struct hlist_head * __net_init netdev_create_hash(void) 11379 { 11380 int i; 11381 struct hlist_head *hash; 11382 11383 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL); 11384 if (hash != NULL) 11385 for (i = 0; i < NETDEV_HASHENTRIES; i++) 11386 INIT_HLIST_HEAD(&hash[i]); 11387 11388 return hash; 11389 } 11390 11391 /* Initialize per network namespace state */ 11392 static int __net_init netdev_init(struct net *net) 11393 { 11394 BUILD_BUG_ON(GRO_HASH_BUCKETS > 11395 8 * sizeof_field(struct napi_struct, gro_bitmask)); 11396 11397 if (net != &init_net) 11398 INIT_LIST_HEAD(&net->dev_base_head); 11399 11400 net->dev_name_head = netdev_create_hash(); 11401 if (net->dev_name_head == NULL) 11402 goto err_name; 11403 11404 net->dev_index_head = netdev_create_hash(); 11405 if (net->dev_index_head == NULL) 11406 goto err_idx; 11407 11408 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain); 11409 11410 return 0; 11411 11412 err_idx: 11413 kfree(net->dev_name_head); 11414 err_name: 11415 return -ENOMEM; 11416 } 11417 11418 /** 11419 * netdev_drivername - network driver for the device 11420 * @dev: network device 11421 * 11422 * Determine network driver for device. 11423 */ 11424 const char *netdev_drivername(const struct net_device *dev) 11425 { 11426 const struct device_driver *driver; 11427 const struct device *parent; 11428 const char *empty = ""; 11429 11430 parent = dev->dev.parent; 11431 if (!parent) 11432 return empty; 11433 11434 driver = parent->driver; 11435 if (driver && driver->name) 11436 return driver->name; 11437 return empty; 11438 } 11439 11440 static void __netdev_printk(const char *level, const struct net_device *dev, 11441 struct va_format *vaf) 11442 { 11443 if (dev && dev->dev.parent) { 11444 dev_printk_emit(level[1] - '0', 11445 dev->dev.parent, 11446 "%s %s %s%s: %pV", 11447 dev_driver_string(dev->dev.parent), 11448 dev_name(dev->dev.parent), 11449 netdev_name(dev), netdev_reg_state(dev), 11450 vaf); 11451 } else if (dev) { 11452 printk("%s%s%s: %pV", 11453 level, netdev_name(dev), netdev_reg_state(dev), vaf); 11454 } else { 11455 printk("%s(NULL net_device): %pV", level, vaf); 11456 } 11457 } 11458 11459 void netdev_printk(const char *level, const struct net_device *dev, 11460 const char *format, ...) 11461 { 11462 struct va_format vaf; 11463 va_list args; 11464 11465 va_start(args, format); 11466 11467 vaf.fmt = format; 11468 vaf.va = &args; 11469 11470 __netdev_printk(level, dev, &vaf); 11471 11472 va_end(args); 11473 } 11474 EXPORT_SYMBOL(netdev_printk); 11475 11476 #define define_netdev_printk_level(func, level) \ 11477 void func(const struct net_device *dev, const char *fmt, ...) \ 11478 { \ 11479 struct va_format vaf; \ 11480 va_list args; \ 11481 \ 11482 va_start(args, fmt); \ 11483 \ 11484 vaf.fmt = fmt; \ 11485 vaf.va = &args; \ 11486 \ 11487 __netdev_printk(level, dev, &vaf); \ 11488 \ 11489 va_end(args); \ 11490 } \ 11491 EXPORT_SYMBOL(func); 11492 11493 define_netdev_printk_level(netdev_emerg, KERN_EMERG); 11494 define_netdev_printk_level(netdev_alert, KERN_ALERT); 11495 define_netdev_printk_level(netdev_crit, KERN_CRIT); 11496 define_netdev_printk_level(netdev_err, KERN_ERR); 11497 define_netdev_printk_level(netdev_warn, KERN_WARNING); 11498 define_netdev_printk_level(netdev_notice, KERN_NOTICE); 11499 define_netdev_printk_level(netdev_info, KERN_INFO); 11500 11501 static void __net_exit netdev_exit(struct net *net) 11502 { 11503 kfree(net->dev_name_head); 11504 kfree(net->dev_index_head); 11505 if (net != &init_net) 11506 WARN_ON_ONCE(!list_empty(&net->dev_base_head)); 11507 } 11508 11509 static struct pernet_operations __net_initdata netdev_net_ops = { 11510 .init = netdev_init, 11511 .exit = netdev_exit, 11512 }; 11513 11514 static void __net_exit default_device_exit(struct net *net) 11515 { 11516 struct net_device *dev, *aux; 11517 /* 11518 * Push all migratable network devices back to the 11519 * initial network namespace 11520 */ 11521 rtnl_lock(); 11522 for_each_netdev_safe(net, dev, aux) { 11523 int err; 11524 char fb_name[IFNAMSIZ]; 11525 11526 /* Ignore unmoveable devices (i.e. loopback) */ 11527 if (dev->features & NETIF_F_NETNS_LOCAL) 11528 continue; 11529 11530 /* Leave virtual devices for the generic cleanup */ 11531 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund) 11532 continue; 11533 11534 /* Push remaining network devices to init_net */ 11535 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); 11536 if (netdev_name_in_use(&init_net, fb_name)) 11537 snprintf(fb_name, IFNAMSIZ, "dev%%d"); 11538 err = dev_change_net_namespace(dev, &init_net, fb_name); 11539 if (err) { 11540 pr_emerg("%s: failed to move %s to init_net: %d\n", 11541 __func__, dev->name, err); 11542 BUG(); 11543 } 11544 } 11545 rtnl_unlock(); 11546 } 11547 11548 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list) 11549 { 11550 /* Return with the rtnl_lock held when there are no network 11551 * devices unregistering in any network namespace in net_list. 11552 */ 11553 struct net *net; 11554 bool unregistering; 11555 DEFINE_WAIT_FUNC(wait, woken_wake_function); 11556 11557 add_wait_queue(&netdev_unregistering_wq, &wait); 11558 for (;;) { 11559 unregistering = false; 11560 rtnl_lock(); 11561 list_for_each_entry(net, net_list, exit_list) { 11562 if (net->dev_unreg_count > 0) { 11563 unregistering = true; 11564 break; 11565 } 11566 } 11567 if (!unregistering) 11568 break; 11569 __rtnl_unlock(); 11570 11571 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); 11572 } 11573 remove_wait_queue(&netdev_unregistering_wq, &wait); 11574 } 11575 11576 static void __net_exit default_device_exit_batch(struct list_head *net_list) 11577 { 11578 /* At exit all network devices most be removed from a network 11579 * namespace. Do this in the reverse order of registration. 11580 * Do this across as many network namespaces as possible to 11581 * improve batching efficiency. 11582 */ 11583 struct net_device *dev; 11584 struct net *net; 11585 LIST_HEAD(dev_kill_list); 11586 11587 /* To prevent network device cleanup code from dereferencing 11588 * loopback devices or network devices that have been freed 11589 * wait here for all pending unregistrations to complete, 11590 * before unregistring the loopback device and allowing the 11591 * network namespace be freed. 11592 * 11593 * The netdev todo list containing all network devices 11594 * unregistrations that happen in default_device_exit_batch 11595 * will run in the rtnl_unlock() at the end of 11596 * default_device_exit_batch. 11597 */ 11598 rtnl_lock_unregistering(net_list); 11599 list_for_each_entry(net, net_list, exit_list) { 11600 for_each_netdev_reverse(net, dev) { 11601 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) 11602 dev->rtnl_link_ops->dellink(dev, &dev_kill_list); 11603 else 11604 unregister_netdevice_queue(dev, &dev_kill_list); 11605 } 11606 } 11607 unregister_netdevice_many(&dev_kill_list); 11608 rtnl_unlock(); 11609 } 11610 11611 static struct pernet_operations __net_initdata default_device_ops = { 11612 .exit = default_device_exit, 11613 .exit_batch = default_device_exit_batch, 11614 }; 11615 11616 /* 11617 * Initialize the DEV module. At boot time this walks the device list and 11618 * unhooks any devices that fail to initialise (normally hardware not 11619 * present) and leaves us with a valid list of present and active devices. 11620 * 11621 */ 11622 11623 /* 11624 * This is called single threaded during boot, so no need 11625 * to take the rtnl semaphore. 11626 */ 11627 static int __init net_dev_init(void) 11628 { 11629 int i, rc = -ENOMEM; 11630 11631 BUG_ON(!dev_boot_phase); 11632 11633 if (dev_proc_init()) 11634 goto out; 11635 11636 if (netdev_kobject_init()) 11637 goto out; 11638 11639 INIT_LIST_HEAD(&ptype_all); 11640 for (i = 0; i < PTYPE_HASH_SIZE; i++) 11641 INIT_LIST_HEAD(&ptype_base[i]); 11642 11643 INIT_LIST_HEAD(&offload_base); 11644 11645 if (register_pernet_subsys(&netdev_net_ops)) 11646 goto out; 11647 11648 /* 11649 * Initialise the packet receive queues. 11650 */ 11651 11652 for_each_possible_cpu(i) { 11653 struct work_struct *flush = per_cpu_ptr(&flush_works, i); 11654 struct softnet_data *sd = &per_cpu(softnet_data, i); 11655 11656 INIT_WORK(flush, flush_backlog); 11657 11658 skb_queue_head_init(&sd->input_pkt_queue); 11659 skb_queue_head_init(&sd->process_queue); 11660 #ifdef CONFIG_XFRM_OFFLOAD 11661 skb_queue_head_init(&sd->xfrm_backlog); 11662 #endif 11663 INIT_LIST_HEAD(&sd->poll_list); 11664 sd->output_queue_tailp = &sd->output_queue; 11665 #ifdef CONFIG_RPS 11666 INIT_CSD(&sd->csd, rps_trigger_softirq, sd); 11667 sd->cpu = i; 11668 #endif 11669 11670 init_gro_hash(&sd->backlog); 11671 sd->backlog.poll = process_backlog; 11672 sd->backlog.weight = weight_p; 11673 } 11674 11675 dev_boot_phase = 0; 11676 11677 /* The loopback device is special if any other network devices 11678 * is present in a network namespace the loopback device must 11679 * be present. Since we now dynamically allocate and free the 11680 * loopback device ensure this invariant is maintained by 11681 * keeping the loopback device as the first device on the 11682 * list of network devices. Ensuring the loopback devices 11683 * is the first device that appears and the last network device 11684 * that disappears. 11685 */ 11686 if (register_pernet_device(&loopback_net_ops)) 11687 goto out; 11688 11689 if (register_pernet_device(&default_device_ops)) 11690 goto out; 11691 11692 open_softirq(NET_TX_SOFTIRQ, net_tx_action); 11693 open_softirq(NET_RX_SOFTIRQ, net_rx_action); 11694 11695 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", 11696 NULL, dev_cpu_dead); 11697 WARN_ON(rc < 0); 11698 rc = 0; 11699 out: 11700 return rc; 11701 } 11702 11703 subsys_initcall(net_dev_init); 11704