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