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_bh(&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_bh(&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_bh(&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_bh(&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_bh(&dev_base_lock); 1192 netdev_name_node_del(dev->name_node); 1193 write_unlock_bh(&dev_base_lock); 1194 1195 synchronize_rcu(); 1196 1197 write_lock_bh(&dev_base_lock); 1198 netdev_name_node_add(net, dev->name_node); 1199 write_unlock_bh(&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 if (txq->xmit_lock_owner != cpu) { 4097 if (dev_xmit_recursion()) 4098 goto recursion_alert; 4099 4100 skb = validate_xmit_skb(skb, dev, &again); 4101 if (!skb) 4102 goto out; 4103 4104 PRANDOM_ADD_NOISE(skb, dev, txq, jiffies); 4105 HARD_TX_LOCK(dev, txq, cpu); 4106 4107 if (!netif_xmit_stopped(txq)) { 4108 dev_xmit_recursion_inc(); 4109 skb = dev_hard_start_xmit(skb, dev, txq, &rc); 4110 dev_xmit_recursion_dec(); 4111 if (dev_xmit_complete(rc)) { 4112 HARD_TX_UNLOCK(dev, txq); 4113 goto out; 4114 } 4115 } 4116 HARD_TX_UNLOCK(dev, txq); 4117 net_crit_ratelimited("Virtual device %s asks to queue packet!\n", 4118 dev->name); 4119 } else { 4120 /* Recursion is detected! It is possible, 4121 * unfortunately 4122 */ 4123 recursion_alert: 4124 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", 4125 dev->name); 4126 } 4127 } 4128 4129 rc = -ENETDOWN; 4130 rcu_read_unlock_bh(); 4131 4132 atomic_long_inc(&dev->tx_dropped); 4133 kfree_skb_list(skb); 4134 return rc; 4135 out: 4136 rcu_read_unlock_bh(); 4137 return rc; 4138 } 4139 4140 int dev_queue_xmit(struct sk_buff *skb) 4141 { 4142 return __dev_queue_xmit(skb, NULL); 4143 } 4144 EXPORT_SYMBOL(dev_queue_xmit); 4145 4146 int dev_queue_xmit_accel(struct sk_buff *skb, struct net_device *sb_dev) 4147 { 4148 return __dev_queue_xmit(skb, sb_dev); 4149 } 4150 EXPORT_SYMBOL(dev_queue_xmit_accel); 4151 4152 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id) 4153 { 4154 struct net_device *dev = skb->dev; 4155 struct sk_buff *orig_skb = skb; 4156 struct netdev_queue *txq; 4157 int ret = NETDEV_TX_BUSY; 4158 bool again = false; 4159 4160 if (unlikely(!netif_running(dev) || 4161 !netif_carrier_ok(dev))) 4162 goto drop; 4163 4164 skb = validate_xmit_skb_list(skb, dev, &again); 4165 if (skb != orig_skb) 4166 goto drop; 4167 4168 skb_set_queue_mapping(skb, queue_id); 4169 txq = skb_get_tx_queue(dev, skb); 4170 PRANDOM_ADD_NOISE(skb, dev, txq, jiffies); 4171 4172 local_bh_disable(); 4173 4174 dev_xmit_recursion_inc(); 4175 HARD_TX_LOCK(dev, txq, smp_processor_id()); 4176 if (!netif_xmit_frozen_or_drv_stopped(txq)) 4177 ret = netdev_start_xmit(skb, dev, txq, false); 4178 HARD_TX_UNLOCK(dev, txq); 4179 dev_xmit_recursion_dec(); 4180 4181 local_bh_enable(); 4182 return ret; 4183 drop: 4184 atomic_long_inc(&dev->tx_dropped); 4185 kfree_skb_list(skb); 4186 return NET_XMIT_DROP; 4187 } 4188 EXPORT_SYMBOL(__dev_direct_xmit); 4189 4190 /************************************************************************* 4191 * Receiver routines 4192 *************************************************************************/ 4193 4194 int netdev_max_backlog __read_mostly = 1000; 4195 EXPORT_SYMBOL(netdev_max_backlog); 4196 4197 int netdev_tstamp_prequeue __read_mostly = 1; 4198 int netdev_budget __read_mostly = 300; 4199 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */ 4200 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ; 4201 int weight_p __read_mostly = 64; /* old backlog weight */ 4202 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ 4203 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ 4204 int dev_rx_weight __read_mostly = 64; 4205 int dev_tx_weight __read_mostly = 64; 4206 4207 /* Called with irq disabled */ 4208 static inline void ____napi_schedule(struct softnet_data *sd, 4209 struct napi_struct *napi) 4210 { 4211 struct task_struct *thread; 4212 4213 if (test_bit(NAPI_STATE_THREADED, &napi->state)) { 4214 /* Paired with smp_mb__before_atomic() in 4215 * napi_enable()/dev_set_threaded(). 4216 * Use READ_ONCE() to guarantee a complete 4217 * read on napi->thread. Only call 4218 * wake_up_process() when it's not NULL. 4219 */ 4220 thread = READ_ONCE(napi->thread); 4221 if (thread) { 4222 /* Avoid doing set_bit() if the thread is in 4223 * INTERRUPTIBLE state, cause napi_thread_wait() 4224 * makes sure to proceed with napi polling 4225 * if the thread is explicitly woken from here. 4226 */ 4227 if (READ_ONCE(thread->__state) != TASK_INTERRUPTIBLE) 4228 set_bit(NAPI_STATE_SCHED_THREADED, &napi->state); 4229 wake_up_process(thread); 4230 return; 4231 } 4232 } 4233 4234 list_add_tail(&napi->poll_list, &sd->poll_list); 4235 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4236 } 4237 4238 #ifdef CONFIG_RPS 4239 4240 /* One global table that all flow-based protocols share. */ 4241 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; 4242 EXPORT_SYMBOL(rps_sock_flow_table); 4243 u32 rps_cpu_mask __read_mostly; 4244 EXPORT_SYMBOL(rps_cpu_mask); 4245 4246 struct static_key_false rps_needed __read_mostly; 4247 EXPORT_SYMBOL(rps_needed); 4248 struct static_key_false rfs_needed __read_mostly; 4249 EXPORT_SYMBOL(rfs_needed); 4250 4251 static struct rps_dev_flow * 4252 set_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4253 struct rps_dev_flow *rflow, u16 next_cpu) 4254 { 4255 if (next_cpu < nr_cpu_ids) { 4256 #ifdef CONFIG_RFS_ACCEL 4257 struct netdev_rx_queue *rxqueue; 4258 struct rps_dev_flow_table *flow_table; 4259 struct rps_dev_flow *old_rflow; 4260 u32 flow_id; 4261 u16 rxq_index; 4262 int rc; 4263 4264 /* Should we steer this flow to a different hardware queue? */ 4265 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || 4266 !(dev->features & NETIF_F_NTUPLE)) 4267 goto out; 4268 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); 4269 if (rxq_index == skb_get_rx_queue(skb)) 4270 goto out; 4271 4272 rxqueue = dev->_rx + rxq_index; 4273 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4274 if (!flow_table) 4275 goto out; 4276 flow_id = skb_get_hash(skb) & flow_table->mask; 4277 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, 4278 rxq_index, flow_id); 4279 if (rc < 0) 4280 goto out; 4281 old_rflow = rflow; 4282 rflow = &flow_table->flows[flow_id]; 4283 rflow->filter = rc; 4284 if (old_rflow->filter == rflow->filter) 4285 old_rflow->filter = RPS_NO_FILTER; 4286 out: 4287 #endif 4288 rflow->last_qtail = 4289 per_cpu(softnet_data, next_cpu).input_queue_head; 4290 } 4291 4292 rflow->cpu = next_cpu; 4293 return rflow; 4294 } 4295 4296 /* 4297 * get_rps_cpu is called from netif_receive_skb and returns the target 4298 * CPU from the RPS map of the receiving queue for a given skb. 4299 * rcu_read_lock must be held on entry. 4300 */ 4301 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4302 struct rps_dev_flow **rflowp) 4303 { 4304 const struct rps_sock_flow_table *sock_flow_table; 4305 struct netdev_rx_queue *rxqueue = dev->_rx; 4306 struct rps_dev_flow_table *flow_table; 4307 struct rps_map *map; 4308 int cpu = -1; 4309 u32 tcpu; 4310 u32 hash; 4311 4312 if (skb_rx_queue_recorded(skb)) { 4313 u16 index = skb_get_rx_queue(skb); 4314 4315 if (unlikely(index >= dev->real_num_rx_queues)) { 4316 WARN_ONCE(dev->real_num_rx_queues > 1, 4317 "%s received packet on queue %u, but number " 4318 "of RX queues is %u\n", 4319 dev->name, index, dev->real_num_rx_queues); 4320 goto done; 4321 } 4322 rxqueue += index; 4323 } 4324 4325 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ 4326 4327 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4328 map = rcu_dereference(rxqueue->rps_map); 4329 if (!flow_table && !map) 4330 goto done; 4331 4332 skb_reset_network_header(skb); 4333 hash = skb_get_hash(skb); 4334 if (!hash) 4335 goto done; 4336 4337 sock_flow_table = rcu_dereference(rps_sock_flow_table); 4338 if (flow_table && sock_flow_table) { 4339 struct rps_dev_flow *rflow; 4340 u32 next_cpu; 4341 u32 ident; 4342 4343 /* First check into global flow table if there is a match */ 4344 ident = sock_flow_table->ents[hash & sock_flow_table->mask]; 4345 if ((ident ^ hash) & ~rps_cpu_mask) 4346 goto try_rps; 4347 4348 next_cpu = ident & rps_cpu_mask; 4349 4350 /* OK, now we know there is a match, 4351 * we can look at the local (per receive queue) flow table 4352 */ 4353 rflow = &flow_table->flows[hash & flow_table->mask]; 4354 tcpu = rflow->cpu; 4355 4356 /* 4357 * If the desired CPU (where last recvmsg was done) is 4358 * different from current CPU (one in the rx-queue flow 4359 * table entry), switch if one of the following holds: 4360 * - Current CPU is unset (>= nr_cpu_ids). 4361 * - Current CPU is offline. 4362 * - The current CPU's queue tail has advanced beyond the 4363 * last packet that was enqueued using this table entry. 4364 * This guarantees that all previous packets for the flow 4365 * have been dequeued, thus preserving in order delivery. 4366 */ 4367 if (unlikely(tcpu != next_cpu) && 4368 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || 4369 ((int)(per_cpu(softnet_data, tcpu).input_queue_head - 4370 rflow->last_qtail)) >= 0)) { 4371 tcpu = next_cpu; 4372 rflow = set_rps_cpu(dev, skb, rflow, next_cpu); 4373 } 4374 4375 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { 4376 *rflowp = rflow; 4377 cpu = tcpu; 4378 goto done; 4379 } 4380 } 4381 4382 try_rps: 4383 4384 if (map) { 4385 tcpu = map->cpus[reciprocal_scale(hash, map->len)]; 4386 if (cpu_online(tcpu)) { 4387 cpu = tcpu; 4388 goto done; 4389 } 4390 } 4391 4392 done: 4393 return cpu; 4394 } 4395 4396 #ifdef CONFIG_RFS_ACCEL 4397 4398 /** 4399 * rps_may_expire_flow - check whether an RFS hardware filter may be removed 4400 * @dev: Device on which the filter was set 4401 * @rxq_index: RX queue index 4402 * @flow_id: Flow ID passed to ndo_rx_flow_steer() 4403 * @filter_id: Filter ID returned by ndo_rx_flow_steer() 4404 * 4405 * Drivers that implement ndo_rx_flow_steer() should periodically call 4406 * this function for each installed filter and remove the filters for 4407 * which it returns %true. 4408 */ 4409 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, 4410 u32 flow_id, u16 filter_id) 4411 { 4412 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; 4413 struct rps_dev_flow_table *flow_table; 4414 struct rps_dev_flow *rflow; 4415 bool expire = true; 4416 unsigned int cpu; 4417 4418 rcu_read_lock(); 4419 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4420 if (flow_table && flow_id <= flow_table->mask) { 4421 rflow = &flow_table->flows[flow_id]; 4422 cpu = READ_ONCE(rflow->cpu); 4423 if (rflow->filter == filter_id && cpu < nr_cpu_ids && 4424 ((int)(per_cpu(softnet_data, cpu).input_queue_head - 4425 rflow->last_qtail) < 4426 (int)(10 * flow_table->mask))) 4427 expire = false; 4428 } 4429 rcu_read_unlock(); 4430 return expire; 4431 } 4432 EXPORT_SYMBOL(rps_may_expire_flow); 4433 4434 #endif /* CONFIG_RFS_ACCEL */ 4435 4436 /* Called from hardirq (IPI) context */ 4437 static void rps_trigger_softirq(void *data) 4438 { 4439 struct softnet_data *sd = data; 4440 4441 ____napi_schedule(sd, &sd->backlog); 4442 sd->received_rps++; 4443 } 4444 4445 #endif /* CONFIG_RPS */ 4446 4447 /* 4448 * Check if this softnet_data structure is another cpu one 4449 * If yes, queue it to our IPI list and return 1 4450 * If no, return 0 4451 */ 4452 static int rps_ipi_queued(struct softnet_data *sd) 4453 { 4454 #ifdef CONFIG_RPS 4455 struct softnet_data *mysd = this_cpu_ptr(&softnet_data); 4456 4457 if (sd != mysd) { 4458 sd->rps_ipi_next = mysd->rps_ipi_list; 4459 mysd->rps_ipi_list = sd; 4460 4461 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4462 return 1; 4463 } 4464 #endif /* CONFIG_RPS */ 4465 return 0; 4466 } 4467 4468 #ifdef CONFIG_NET_FLOW_LIMIT 4469 int netdev_flow_limit_table_len __read_mostly = (1 << 12); 4470 #endif 4471 4472 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) 4473 { 4474 #ifdef CONFIG_NET_FLOW_LIMIT 4475 struct sd_flow_limit *fl; 4476 struct softnet_data *sd; 4477 unsigned int old_flow, new_flow; 4478 4479 if (qlen < (netdev_max_backlog >> 1)) 4480 return false; 4481 4482 sd = this_cpu_ptr(&softnet_data); 4483 4484 rcu_read_lock(); 4485 fl = rcu_dereference(sd->flow_limit); 4486 if (fl) { 4487 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); 4488 old_flow = fl->history[fl->history_head]; 4489 fl->history[fl->history_head] = new_flow; 4490 4491 fl->history_head++; 4492 fl->history_head &= FLOW_LIMIT_HISTORY - 1; 4493 4494 if (likely(fl->buckets[old_flow])) 4495 fl->buckets[old_flow]--; 4496 4497 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { 4498 fl->count++; 4499 rcu_read_unlock(); 4500 return true; 4501 } 4502 } 4503 rcu_read_unlock(); 4504 #endif 4505 return false; 4506 } 4507 4508 /* 4509 * enqueue_to_backlog is called to queue an skb to a per CPU backlog 4510 * queue (may be a remote CPU queue). 4511 */ 4512 static int enqueue_to_backlog(struct sk_buff *skb, int cpu, 4513 unsigned int *qtail) 4514 { 4515 struct softnet_data *sd; 4516 unsigned long flags; 4517 unsigned int qlen; 4518 4519 sd = &per_cpu(softnet_data, cpu); 4520 4521 local_irq_save(flags); 4522 4523 rps_lock(sd); 4524 if (!netif_running(skb->dev)) 4525 goto drop; 4526 qlen = skb_queue_len(&sd->input_pkt_queue); 4527 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) { 4528 if (qlen) { 4529 enqueue: 4530 __skb_queue_tail(&sd->input_pkt_queue, skb); 4531 input_queue_tail_incr_save(sd, qtail); 4532 rps_unlock(sd); 4533 local_irq_restore(flags); 4534 return NET_RX_SUCCESS; 4535 } 4536 4537 /* Schedule NAPI for backlog device 4538 * We can use non atomic operation since we own the queue lock 4539 */ 4540 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) { 4541 if (!rps_ipi_queued(sd)) 4542 ____napi_schedule(sd, &sd->backlog); 4543 } 4544 goto enqueue; 4545 } 4546 4547 drop: 4548 sd->dropped++; 4549 rps_unlock(sd); 4550 4551 local_irq_restore(flags); 4552 4553 atomic_long_inc(&skb->dev->rx_dropped); 4554 kfree_skb(skb); 4555 return NET_RX_DROP; 4556 } 4557 4558 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb) 4559 { 4560 struct net_device *dev = skb->dev; 4561 struct netdev_rx_queue *rxqueue; 4562 4563 rxqueue = dev->_rx; 4564 4565 if (skb_rx_queue_recorded(skb)) { 4566 u16 index = skb_get_rx_queue(skb); 4567 4568 if (unlikely(index >= dev->real_num_rx_queues)) { 4569 WARN_ONCE(dev->real_num_rx_queues > 1, 4570 "%s received packet on queue %u, but number " 4571 "of RX queues is %u\n", 4572 dev->name, index, dev->real_num_rx_queues); 4573 4574 return rxqueue; /* Return first rxqueue */ 4575 } 4576 rxqueue += index; 4577 } 4578 return rxqueue; 4579 } 4580 4581 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp, 4582 struct bpf_prog *xdp_prog) 4583 { 4584 void *orig_data, *orig_data_end, *hard_start; 4585 struct netdev_rx_queue *rxqueue; 4586 bool orig_bcast, orig_host; 4587 u32 mac_len, frame_sz; 4588 __be16 orig_eth_type; 4589 struct ethhdr *eth; 4590 u32 metalen, act; 4591 int off; 4592 4593 /* The XDP program wants to see the packet starting at the MAC 4594 * header. 4595 */ 4596 mac_len = skb->data - skb_mac_header(skb); 4597 hard_start = skb->data - skb_headroom(skb); 4598 4599 /* SKB "head" area always have tailroom for skb_shared_info */ 4600 frame_sz = (void *)skb_end_pointer(skb) - hard_start; 4601 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 4602 4603 rxqueue = netif_get_rxqueue(skb); 4604 xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq); 4605 xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len, 4606 skb_headlen(skb) + mac_len, true); 4607 4608 orig_data_end = xdp->data_end; 4609 orig_data = xdp->data; 4610 eth = (struct ethhdr *)xdp->data; 4611 orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr); 4612 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest); 4613 orig_eth_type = eth->h_proto; 4614 4615 act = bpf_prog_run_xdp(xdp_prog, xdp); 4616 4617 /* check if bpf_xdp_adjust_head was used */ 4618 off = xdp->data - orig_data; 4619 if (off) { 4620 if (off > 0) 4621 __skb_pull(skb, off); 4622 else if (off < 0) 4623 __skb_push(skb, -off); 4624 4625 skb->mac_header += off; 4626 skb_reset_network_header(skb); 4627 } 4628 4629 /* check if bpf_xdp_adjust_tail was used */ 4630 off = xdp->data_end - orig_data_end; 4631 if (off != 0) { 4632 skb_set_tail_pointer(skb, xdp->data_end - xdp->data); 4633 skb->len += off; /* positive on grow, negative on shrink */ 4634 } 4635 4636 /* check if XDP changed eth hdr such SKB needs update */ 4637 eth = (struct ethhdr *)xdp->data; 4638 if ((orig_eth_type != eth->h_proto) || 4639 (orig_host != ether_addr_equal_64bits(eth->h_dest, 4640 skb->dev->dev_addr)) || 4641 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) { 4642 __skb_push(skb, ETH_HLEN); 4643 skb->pkt_type = PACKET_HOST; 4644 skb->protocol = eth_type_trans(skb, skb->dev); 4645 } 4646 4647 /* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull 4648 * before calling us again on redirect path. We do not call do_redirect 4649 * as we leave that up to the caller. 4650 * 4651 * Caller is responsible for managing lifetime of skb (i.e. calling 4652 * kfree_skb in response to actions it cannot handle/XDP_DROP). 4653 */ 4654 switch (act) { 4655 case XDP_REDIRECT: 4656 case XDP_TX: 4657 __skb_push(skb, mac_len); 4658 break; 4659 case XDP_PASS: 4660 metalen = xdp->data - xdp->data_meta; 4661 if (metalen) 4662 skb_metadata_set(skb, metalen); 4663 break; 4664 } 4665 4666 return act; 4667 } 4668 4669 static u32 netif_receive_generic_xdp(struct sk_buff *skb, 4670 struct xdp_buff *xdp, 4671 struct bpf_prog *xdp_prog) 4672 { 4673 u32 act = XDP_DROP; 4674 4675 /* Reinjected packets coming from act_mirred or similar should 4676 * not get XDP generic processing. 4677 */ 4678 if (skb_is_redirected(skb)) 4679 return XDP_PASS; 4680 4681 /* XDP packets must be linear and must have sufficient headroom 4682 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also 4683 * native XDP provides, thus we need to do it here as well. 4684 */ 4685 if (skb_cloned(skb) || skb_is_nonlinear(skb) || 4686 skb_headroom(skb) < XDP_PACKET_HEADROOM) { 4687 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb); 4688 int troom = skb->tail + skb->data_len - skb->end; 4689 4690 /* In case we have to go down the path and also linearize, 4691 * then lets do the pskb_expand_head() work just once here. 4692 */ 4693 if (pskb_expand_head(skb, 4694 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0, 4695 troom > 0 ? troom + 128 : 0, GFP_ATOMIC)) 4696 goto do_drop; 4697 if (skb_linearize(skb)) 4698 goto do_drop; 4699 } 4700 4701 act = bpf_prog_run_generic_xdp(skb, xdp, xdp_prog); 4702 switch (act) { 4703 case XDP_REDIRECT: 4704 case XDP_TX: 4705 case XDP_PASS: 4706 break; 4707 default: 4708 bpf_warn_invalid_xdp_action(act); 4709 fallthrough; 4710 case XDP_ABORTED: 4711 trace_xdp_exception(skb->dev, xdp_prog, act); 4712 fallthrough; 4713 case XDP_DROP: 4714 do_drop: 4715 kfree_skb(skb); 4716 break; 4717 } 4718 4719 return act; 4720 } 4721 4722 /* When doing generic XDP we have to bypass the qdisc layer and the 4723 * network taps in order to match in-driver-XDP behavior. 4724 */ 4725 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog) 4726 { 4727 struct net_device *dev = skb->dev; 4728 struct netdev_queue *txq; 4729 bool free_skb = true; 4730 int cpu, rc; 4731 4732 txq = netdev_core_pick_tx(dev, skb, NULL); 4733 cpu = smp_processor_id(); 4734 HARD_TX_LOCK(dev, txq, cpu); 4735 if (!netif_xmit_stopped(txq)) { 4736 rc = netdev_start_xmit(skb, dev, txq, 0); 4737 if (dev_xmit_complete(rc)) 4738 free_skb = false; 4739 } 4740 HARD_TX_UNLOCK(dev, txq); 4741 if (free_skb) { 4742 trace_xdp_exception(dev, xdp_prog, XDP_TX); 4743 kfree_skb(skb); 4744 } 4745 } 4746 4747 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key); 4748 4749 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb) 4750 { 4751 if (xdp_prog) { 4752 struct xdp_buff xdp; 4753 u32 act; 4754 int err; 4755 4756 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog); 4757 if (act != XDP_PASS) { 4758 switch (act) { 4759 case XDP_REDIRECT: 4760 err = xdp_do_generic_redirect(skb->dev, skb, 4761 &xdp, xdp_prog); 4762 if (err) 4763 goto out_redir; 4764 break; 4765 case XDP_TX: 4766 generic_xdp_tx(skb, xdp_prog); 4767 break; 4768 } 4769 return XDP_DROP; 4770 } 4771 } 4772 return XDP_PASS; 4773 out_redir: 4774 kfree_skb(skb); 4775 return XDP_DROP; 4776 } 4777 EXPORT_SYMBOL_GPL(do_xdp_generic); 4778 4779 static int netif_rx_internal(struct sk_buff *skb) 4780 { 4781 int ret; 4782 4783 net_timestamp_check(netdev_tstamp_prequeue, skb); 4784 4785 trace_netif_rx(skb); 4786 4787 #ifdef CONFIG_RPS 4788 if (static_branch_unlikely(&rps_needed)) { 4789 struct rps_dev_flow voidflow, *rflow = &voidflow; 4790 int cpu; 4791 4792 preempt_disable(); 4793 rcu_read_lock(); 4794 4795 cpu = get_rps_cpu(skb->dev, skb, &rflow); 4796 if (cpu < 0) 4797 cpu = smp_processor_id(); 4798 4799 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 4800 4801 rcu_read_unlock(); 4802 preempt_enable(); 4803 } else 4804 #endif 4805 { 4806 unsigned int qtail; 4807 4808 ret = enqueue_to_backlog(skb, get_cpu(), &qtail); 4809 put_cpu(); 4810 } 4811 return ret; 4812 } 4813 4814 /** 4815 * netif_rx - post buffer to the network code 4816 * @skb: buffer to post 4817 * 4818 * This function receives a packet from a device driver and queues it for 4819 * the upper (protocol) levels to process. It always succeeds. The buffer 4820 * may be dropped during processing for congestion control or by the 4821 * protocol layers. 4822 * 4823 * return values: 4824 * NET_RX_SUCCESS (no congestion) 4825 * NET_RX_DROP (packet was dropped) 4826 * 4827 */ 4828 4829 int netif_rx(struct sk_buff *skb) 4830 { 4831 int ret; 4832 4833 trace_netif_rx_entry(skb); 4834 4835 ret = netif_rx_internal(skb); 4836 trace_netif_rx_exit(ret); 4837 4838 return ret; 4839 } 4840 EXPORT_SYMBOL(netif_rx); 4841 4842 int netif_rx_ni(struct sk_buff *skb) 4843 { 4844 int err; 4845 4846 trace_netif_rx_ni_entry(skb); 4847 4848 preempt_disable(); 4849 err = netif_rx_internal(skb); 4850 if (local_softirq_pending()) 4851 do_softirq(); 4852 preempt_enable(); 4853 trace_netif_rx_ni_exit(err); 4854 4855 return err; 4856 } 4857 EXPORT_SYMBOL(netif_rx_ni); 4858 4859 int netif_rx_any_context(struct sk_buff *skb) 4860 { 4861 /* 4862 * If invoked from contexts which do not invoke bottom half 4863 * processing either at return from interrupt or when softrqs are 4864 * reenabled, use netif_rx_ni() which invokes bottomhalf processing 4865 * directly. 4866 */ 4867 if (in_interrupt()) 4868 return netif_rx(skb); 4869 else 4870 return netif_rx_ni(skb); 4871 } 4872 EXPORT_SYMBOL(netif_rx_any_context); 4873 4874 static __latent_entropy void net_tx_action(struct softirq_action *h) 4875 { 4876 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 4877 4878 if (sd->completion_queue) { 4879 struct sk_buff *clist; 4880 4881 local_irq_disable(); 4882 clist = sd->completion_queue; 4883 sd->completion_queue = NULL; 4884 local_irq_enable(); 4885 4886 while (clist) { 4887 struct sk_buff *skb = clist; 4888 4889 clist = clist->next; 4890 4891 WARN_ON(refcount_read(&skb->users)); 4892 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED)) 4893 trace_consume_skb(skb); 4894 else 4895 trace_kfree_skb(skb, net_tx_action); 4896 4897 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) 4898 __kfree_skb(skb); 4899 else 4900 __kfree_skb_defer(skb); 4901 } 4902 } 4903 4904 if (sd->output_queue) { 4905 struct Qdisc *head; 4906 4907 local_irq_disable(); 4908 head = sd->output_queue; 4909 sd->output_queue = NULL; 4910 sd->output_queue_tailp = &sd->output_queue; 4911 local_irq_enable(); 4912 4913 rcu_read_lock(); 4914 4915 while (head) { 4916 struct Qdisc *q = head; 4917 spinlock_t *root_lock = NULL; 4918 4919 head = head->next_sched; 4920 4921 /* We need to make sure head->next_sched is read 4922 * before clearing __QDISC_STATE_SCHED 4923 */ 4924 smp_mb__before_atomic(); 4925 4926 if (!(q->flags & TCQ_F_NOLOCK)) { 4927 root_lock = qdisc_lock(q); 4928 spin_lock(root_lock); 4929 } else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, 4930 &q->state))) { 4931 /* There is a synchronize_net() between 4932 * STATE_DEACTIVATED flag being set and 4933 * qdisc_reset()/some_qdisc_is_busy() in 4934 * dev_deactivate(), so we can safely bail out 4935 * early here to avoid data race between 4936 * qdisc_deactivate() and some_qdisc_is_busy() 4937 * for lockless qdisc. 4938 */ 4939 clear_bit(__QDISC_STATE_SCHED, &q->state); 4940 continue; 4941 } 4942 4943 clear_bit(__QDISC_STATE_SCHED, &q->state); 4944 qdisc_run(q); 4945 if (root_lock) 4946 spin_unlock(root_lock); 4947 } 4948 4949 rcu_read_unlock(); 4950 } 4951 4952 xfrm_dev_backlog(sd); 4953 } 4954 4955 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) 4956 /* This hook is defined here for ATM LANE */ 4957 int (*br_fdb_test_addr_hook)(struct net_device *dev, 4958 unsigned char *addr) __read_mostly; 4959 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); 4960 #endif 4961 4962 static inline struct sk_buff * 4963 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 4964 struct net_device *orig_dev, bool *another) 4965 { 4966 #ifdef CONFIG_NET_CLS_ACT 4967 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress); 4968 struct tcf_result cl_res; 4969 4970 /* If there's at least one ingress present somewhere (so 4971 * we get here via enabled static key), remaining devices 4972 * that are not configured with an ingress qdisc will bail 4973 * out here. 4974 */ 4975 if (!miniq) 4976 return skb; 4977 4978 if (*pt_prev) { 4979 *ret = deliver_skb(skb, *pt_prev, orig_dev); 4980 *pt_prev = NULL; 4981 } 4982 4983 qdisc_skb_cb(skb)->pkt_len = skb->len; 4984 qdisc_skb_cb(skb)->mru = 0; 4985 qdisc_skb_cb(skb)->post_ct = false; 4986 skb->tc_at_ingress = 1; 4987 mini_qdisc_bstats_cpu_update(miniq, skb); 4988 4989 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) { 4990 case TC_ACT_OK: 4991 case TC_ACT_RECLASSIFY: 4992 skb->tc_index = TC_H_MIN(cl_res.classid); 4993 break; 4994 case TC_ACT_SHOT: 4995 mini_qdisc_qstats_cpu_drop(miniq); 4996 kfree_skb(skb); 4997 return NULL; 4998 case TC_ACT_STOLEN: 4999 case TC_ACT_QUEUED: 5000 case TC_ACT_TRAP: 5001 consume_skb(skb); 5002 return NULL; 5003 case TC_ACT_REDIRECT: 5004 /* skb_mac_header check was done by cls/act_bpf, so 5005 * we can safely push the L2 header back before 5006 * redirecting to another netdev 5007 */ 5008 __skb_push(skb, skb->mac_len); 5009 if (skb_do_redirect(skb) == -EAGAIN) { 5010 __skb_pull(skb, skb->mac_len); 5011 *another = true; 5012 break; 5013 } 5014 return NULL; 5015 case TC_ACT_CONSUMED: 5016 return NULL; 5017 default: 5018 break; 5019 } 5020 #endif /* CONFIG_NET_CLS_ACT */ 5021 return skb; 5022 } 5023 5024 /** 5025 * netdev_is_rx_handler_busy - check if receive handler is registered 5026 * @dev: device to check 5027 * 5028 * Check if a receive handler is already registered for a given device. 5029 * Return true if there one. 5030 * 5031 * The caller must hold the rtnl_mutex. 5032 */ 5033 bool netdev_is_rx_handler_busy(struct net_device *dev) 5034 { 5035 ASSERT_RTNL(); 5036 return dev && rtnl_dereference(dev->rx_handler); 5037 } 5038 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); 5039 5040 /** 5041 * netdev_rx_handler_register - register receive handler 5042 * @dev: device to register a handler for 5043 * @rx_handler: receive handler to register 5044 * @rx_handler_data: data pointer that is used by rx handler 5045 * 5046 * Register a receive handler for a device. This handler will then be 5047 * called from __netif_receive_skb. A negative errno code is returned 5048 * on a failure. 5049 * 5050 * The caller must hold the rtnl_mutex. 5051 * 5052 * For a general description of rx_handler, see enum rx_handler_result. 5053 */ 5054 int netdev_rx_handler_register(struct net_device *dev, 5055 rx_handler_func_t *rx_handler, 5056 void *rx_handler_data) 5057 { 5058 if (netdev_is_rx_handler_busy(dev)) 5059 return -EBUSY; 5060 5061 if (dev->priv_flags & IFF_NO_RX_HANDLER) 5062 return -EINVAL; 5063 5064 /* Note: rx_handler_data must be set before rx_handler */ 5065 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); 5066 rcu_assign_pointer(dev->rx_handler, rx_handler); 5067 5068 return 0; 5069 } 5070 EXPORT_SYMBOL_GPL(netdev_rx_handler_register); 5071 5072 /** 5073 * netdev_rx_handler_unregister - unregister receive handler 5074 * @dev: device to unregister a handler from 5075 * 5076 * Unregister a receive handler from a device. 5077 * 5078 * The caller must hold the rtnl_mutex. 5079 */ 5080 void netdev_rx_handler_unregister(struct net_device *dev) 5081 { 5082 5083 ASSERT_RTNL(); 5084 RCU_INIT_POINTER(dev->rx_handler, NULL); 5085 /* a reader seeing a non NULL rx_handler in a rcu_read_lock() 5086 * section has a guarantee to see a non NULL rx_handler_data 5087 * as well. 5088 */ 5089 synchronize_net(); 5090 RCU_INIT_POINTER(dev->rx_handler_data, NULL); 5091 } 5092 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); 5093 5094 /* 5095 * Limit the use of PFMEMALLOC reserves to those protocols that implement 5096 * the special handling of PFMEMALLOC skbs. 5097 */ 5098 static bool skb_pfmemalloc_protocol(struct sk_buff *skb) 5099 { 5100 switch (skb->protocol) { 5101 case htons(ETH_P_ARP): 5102 case htons(ETH_P_IP): 5103 case htons(ETH_P_IPV6): 5104 case htons(ETH_P_8021Q): 5105 case htons(ETH_P_8021AD): 5106 return true; 5107 default: 5108 return false; 5109 } 5110 } 5111 5112 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, 5113 int *ret, struct net_device *orig_dev) 5114 { 5115 if (nf_hook_ingress_active(skb)) { 5116 int ingress_retval; 5117 5118 if (*pt_prev) { 5119 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5120 *pt_prev = NULL; 5121 } 5122 5123 rcu_read_lock(); 5124 ingress_retval = nf_hook_ingress(skb); 5125 rcu_read_unlock(); 5126 return ingress_retval; 5127 } 5128 return 0; 5129 } 5130 5131 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc, 5132 struct packet_type **ppt_prev) 5133 { 5134 struct packet_type *ptype, *pt_prev; 5135 rx_handler_func_t *rx_handler; 5136 struct sk_buff *skb = *pskb; 5137 struct net_device *orig_dev; 5138 bool deliver_exact = false; 5139 int ret = NET_RX_DROP; 5140 __be16 type; 5141 5142 net_timestamp_check(!netdev_tstamp_prequeue, skb); 5143 5144 trace_netif_receive_skb(skb); 5145 5146 orig_dev = skb->dev; 5147 5148 skb_reset_network_header(skb); 5149 if (!skb_transport_header_was_set(skb)) 5150 skb_reset_transport_header(skb); 5151 skb_reset_mac_len(skb); 5152 5153 pt_prev = NULL; 5154 5155 another_round: 5156 skb->skb_iif = skb->dev->ifindex; 5157 5158 __this_cpu_inc(softnet_data.processed); 5159 5160 if (static_branch_unlikely(&generic_xdp_needed_key)) { 5161 int ret2; 5162 5163 migrate_disable(); 5164 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb); 5165 migrate_enable(); 5166 5167 if (ret2 != XDP_PASS) { 5168 ret = NET_RX_DROP; 5169 goto out; 5170 } 5171 } 5172 5173 if (eth_type_vlan(skb->protocol)) { 5174 skb = skb_vlan_untag(skb); 5175 if (unlikely(!skb)) 5176 goto out; 5177 } 5178 5179 if (skb_skip_tc_classify(skb)) 5180 goto skip_classify; 5181 5182 if (pfmemalloc) 5183 goto skip_taps; 5184 5185 list_for_each_entry_rcu(ptype, &ptype_all, list) { 5186 if (pt_prev) 5187 ret = deliver_skb(skb, pt_prev, orig_dev); 5188 pt_prev = ptype; 5189 } 5190 5191 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { 5192 if (pt_prev) 5193 ret = deliver_skb(skb, pt_prev, orig_dev); 5194 pt_prev = ptype; 5195 } 5196 5197 skip_taps: 5198 #ifdef CONFIG_NET_INGRESS 5199 if (static_branch_unlikely(&ingress_needed_key)) { 5200 bool another = false; 5201 5202 nf_skip_egress(skb, true); 5203 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev, 5204 &another); 5205 if (another) 5206 goto another_round; 5207 if (!skb) 5208 goto out; 5209 5210 nf_skip_egress(skb, false); 5211 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) 5212 goto out; 5213 } 5214 #endif 5215 skb_reset_redirect(skb); 5216 skip_classify: 5217 if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) 5218 goto drop; 5219 5220 if (skb_vlan_tag_present(skb)) { 5221 if (pt_prev) { 5222 ret = deliver_skb(skb, pt_prev, orig_dev); 5223 pt_prev = NULL; 5224 } 5225 if (vlan_do_receive(&skb)) 5226 goto another_round; 5227 else if (unlikely(!skb)) 5228 goto out; 5229 } 5230 5231 rx_handler = rcu_dereference(skb->dev->rx_handler); 5232 if (rx_handler) { 5233 if (pt_prev) { 5234 ret = deliver_skb(skb, pt_prev, orig_dev); 5235 pt_prev = NULL; 5236 } 5237 switch (rx_handler(&skb)) { 5238 case RX_HANDLER_CONSUMED: 5239 ret = NET_RX_SUCCESS; 5240 goto out; 5241 case RX_HANDLER_ANOTHER: 5242 goto another_round; 5243 case RX_HANDLER_EXACT: 5244 deliver_exact = true; 5245 break; 5246 case RX_HANDLER_PASS: 5247 break; 5248 default: 5249 BUG(); 5250 } 5251 } 5252 5253 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) { 5254 check_vlan_id: 5255 if (skb_vlan_tag_get_id(skb)) { 5256 /* Vlan id is non 0 and vlan_do_receive() above couldn't 5257 * find vlan device. 5258 */ 5259 skb->pkt_type = PACKET_OTHERHOST; 5260 } else if (eth_type_vlan(skb->protocol)) { 5261 /* Outer header is 802.1P with vlan 0, inner header is 5262 * 802.1Q or 802.1AD and vlan_do_receive() above could 5263 * not find vlan dev for vlan id 0. 5264 */ 5265 __vlan_hwaccel_clear_tag(skb); 5266 skb = skb_vlan_untag(skb); 5267 if (unlikely(!skb)) 5268 goto out; 5269 if (vlan_do_receive(&skb)) 5270 /* After stripping off 802.1P header with vlan 0 5271 * vlan dev is found for inner header. 5272 */ 5273 goto another_round; 5274 else if (unlikely(!skb)) 5275 goto out; 5276 else 5277 /* We have stripped outer 802.1P vlan 0 header. 5278 * But could not find vlan dev. 5279 * check again for vlan id to set OTHERHOST. 5280 */ 5281 goto check_vlan_id; 5282 } 5283 /* Note: we might in the future use prio bits 5284 * and set skb->priority like in vlan_do_receive() 5285 * For the time being, just ignore Priority Code Point 5286 */ 5287 __vlan_hwaccel_clear_tag(skb); 5288 } 5289 5290 type = skb->protocol; 5291 5292 /* deliver only exact match when indicated */ 5293 if (likely(!deliver_exact)) { 5294 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5295 &ptype_base[ntohs(type) & 5296 PTYPE_HASH_MASK]); 5297 } 5298 5299 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5300 &orig_dev->ptype_specific); 5301 5302 if (unlikely(skb->dev != orig_dev)) { 5303 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5304 &skb->dev->ptype_specific); 5305 } 5306 5307 if (pt_prev) { 5308 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 5309 goto drop; 5310 *ppt_prev = pt_prev; 5311 } else { 5312 drop: 5313 if (!deliver_exact) 5314 atomic_long_inc(&skb->dev->rx_dropped); 5315 else 5316 atomic_long_inc(&skb->dev->rx_nohandler); 5317 kfree_skb(skb); 5318 /* Jamal, now you will not able to escape explaining 5319 * me how you were going to use this. :-) 5320 */ 5321 ret = NET_RX_DROP; 5322 } 5323 5324 out: 5325 /* The invariant here is that if *ppt_prev is not NULL 5326 * then skb should also be non-NULL. 5327 * 5328 * Apparently *ppt_prev assignment above holds this invariant due to 5329 * skb dereferencing near it. 5330 */ 5331 *pskb = skb; 5332 return ret; 5333 } 5334 5335 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc) 5336 { 5337 struct net_device *orig_dev = skb->dev; 5338 struct packet_type *pt_prev = NULL; 5339 int ret; 5340 5341 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5342 if (pt_prev) 5343 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb, 5344 skb->dev, pt_prev, orig_dev); 5345 return ret; 5346 } 5347 5348 /** 5349 * netif_receive_skb_core - special purpose version of netif_receive_skb 5350 * @skb: buffer to process 5351 * 5352 * More direct receive version of netif_receive_skb(). It should 5353 * only be used by callers that have a need to skip RPS and Generic XDP. 5354 * Caller must also take care of handling if ``(page_is_)pfmemalloc``. 5355 * 5356 * This function may only be called from softirq context and interrupts 5357 * should be enabled. 5358 * 5359 * Return values (usually ignored): 5360 * NET_RX_SUCCESS: no congestion 5361 * NET_RX_DROP: packet was dropped 5362 */ 5363 int netif_receive_skb_core(struct sk_buff *skb) 5364 { 5365 int ret; 5366 5367 rcu_read_lock(); 5368 ret = __netif_receive_skb_one_core(skb, false); 5369 rcu_read_unlock(); 5370 5371 return ret; 5372 } 5373 EXPORT_SYMBOL(netif_receive_skb_core); 5374 5375 static inline void __netif_receive_skb_list_ptype(struct list_head *head, 5376 struct packet_type *pt_prev, 5377 struct net_device *orig_dev) 5378 { 5379 struct sk_buff *skb, *next; 5380 5381 if (!pt_prev) 5382 return; 5383 if (list_empty(head)) 5384 return; 5385 if (pt_prev->list_func != NULL) 5386 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv, 5387 ip_list_rcv, head, pt_prev, orig_dev); 5388 else 5389 list_for_each_entry_safe(skb, next, head, list) { 5390 skb_list_del_init(skb); 5391 pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 5392 } 5393 } 5394 5395 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc) 5396 { 5397 /* Fast-path assumptions: 5398 * - There is no RX handler. 5399 * - Only one packet_type matches. 5400 * If either of these fails, we will end up doing some per-packet 5401 * processing in-line, then handling the 'last ptype' for the whole 5402 * sublist. This can't cause out-of-order delivery to any single ptype, 5403 * because the 'last ptype' must be constant across the sublist, and all 5404 * other ptypes are handled per-packet. 5405 */ 5406 /* Current (common) ptype of sublist */ 5407 struct packet_type *pt_curr = NULL; 5408 /* Current (common) orig_dev of sublist */ 5409 struct net_device *od_curr = NULL; 5410 struct list_head sublist; 5411 struct sk_buff *skb, *next; 5412 5413 INIT_LIST_HEAD(&sublist); 5414 list_for_each_entry_safe(skb, next, head, list) { 5415 struct net_device *orig_dev = skb->dev; 5416 struct packet_type *pt_prev = NULL; 5417 5418 skb_list_del_init(skb); 5419 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5420 if (!pt_prev) 5421 continue; 5422 if (pt_curr != pt_prev || od_curr != orig_dev) { 5423 /* dispatch old sublist */ 5424 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5425 /* start new sublist */ 5426 INIT_LIST_HEAD(&sublist); 5427 pt_curr = pt_prev; 5428 od_curr = orig_dev; 5429 } 5430 list_add_tail(&skb->list, &sublist); 5431 } 5432 5433 /* dispatch final sublist */ 5434 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5435 } 5436 5437 static int __netif_receive_skb(struct sk_buff *skb) 5438 { 5439 int ret; 5440 5441 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { 5442 unsigned int noreclaim_flag; 5443 5444 /* 5445 * PFMEMALLOC skbs are special, they should 5446 * - be delivered to SOCK_MEMALLOC sockets only 5447 * - stay away from userspace 5448 * - have bounded memory usage 5449 * 5450 * Use PF_MEMALLOC as this saves us from propagating the allocation 5451 * context down to all allocation sites. 5452 */ 5453 noreclaim_flag = memalloc_noreclaim_save(); 5454 ret = __netif_receive_skb_one_core(skb, true); 5455 memalloc_noreclaim_restore(noreclaim_flag); 5456 } else 5457 ret = __netif_receive_skb_one_core(skb, false); 5458 5459 return ret; 5460 } 5461 5462 static void __netif_receive_skb_list(struct list_head *head) 5463 { 5464 unsigned long noreclaim_flag = 0; 5465 struct sk_buff *skb, *next; 5466 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */ 5467 5468 list_for_each_entry_safe(skb, next, head, list) { 5469 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) { 5470 struct list_head sublist; 5471 5472 /* Handle the previous sublist */ 5473 list_cut_before(&sublist, head, &skb->list); 5474 if (!list_empty(&sublist)) 5475 __netif_receive_skb_list_core(&sublist, pfmemalloc); 5476 pfmemalloc = !pfmemalloc; 5477 /* See comments in __netif_receive_skb */ 5478 if (pfmemalloc) 5479 noreclaim_flag = memalloc_noreclaim_save(); 5480 else 5481 memalloc_noreclaim_restore(noreclaim_flag); 5482 } 5483 } 5484 /* Handle the remaining sublist */ 5485 if (!list_empty(head)) 5486 __netif_receive_skb_list_core(head, pfmemalloc); 5487 /* Restore pflags */ 5488 if (pfmemalloc) 5489 memalloc_noreclaim_restore(noreclaim_flag); 5490 } 5491 5492 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp) 5493 { 5494 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog); 5495 struct bpf_prog *new = xdp->prog; 5496 int ret = 0; 5497 5498 switch (xdp->command) { 5499 case XDP_SETUP_PROG: 5500 rcu_assign_pointer(dev->xdp_prog, new); 5501 if (old) 5502 bpf_prog_put(old); 5503 5504 if (old && !new) { 5505 static_branch_dec(&generic_xdp_needed_key); 5506 } else if (new && !old) { 5507 static_branch_inc(&generic_xdp_needed_key); 5508 dev_disable_lro(dev); 5509 dev_disable_gro_hw(dev); 5510 } 5511 break; 5512 5513 default: 5514 ret = -EINVAL; 5515 break; 5516 } 5517 5518 return ret; 5519 } 5520 5521 static int netif_receive_skb_internal(struct sk_buff *skb) 5522 { 5523 int ret; 5524 5525 net_timestamp_check(netdev_tstamp_prequeue, skb); 5526 5527 if (skb_defer_rx_timestamp(skb)) 5528 return NET_RX_SUCCESS; 5529 5530 rcu_read_lock(); 5531 #ifdef CONFIG_RPS 5532 if (static_branch_unlikely(&rps_needed)) { 5533 struct rps_dev_flow voidflow, *rflow = &voidflow; 5534 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5535 5536 if (cpu >= 0) { 5537 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5538 rcu_read_unlock(); 5539 return ret; 5540 } 5541 } 5542 #endif 5543 ret = __netif_receive_skb(skb); 5544 rcu_read_unlock(); 5545 return ret; 5546 } 5547 5548 void netif_receive_skb_list_internal(struct list_head *head) 5549 { 5550 struct sk_buff *skb, *next; 5551 struct list_head sublist; 5552 5553 INIT_LIST_HEAD(&sublist); 5554 list_for_each_entry_safe(skb, next, head, list) { 5555 net_timestamp_check(netdev_tstamp_prequeue, skb); 5556 skb_list_del_init(skb); 5557 if (!skb_defer_rx_timestamp(skb)) 5558 list_add_tail(&skb->list, &sublist); 5559 } 5560 list_splice_init(&sublist, head); 5561 5562 rcu_read_lock(); 5563 #ifdef CONFIG_RPS 5564 if (static_branch_unlikely(&rps_needed)) { 5565 list_for_each_entry_safe(skb, next, head, list) { 5566 struct rps_dev_flow voidflow, *rflow = &voidflow; 5567 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5568 5569 if (cpu >= 0) { 5570 /* Will be handled, remove from list */ 5571 skb_list_del_init(skb); 5572 enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5573 } 5574 } 5575 } 5576 #endif 5577 __netif_receive_skb_list(head); 5578 rcu_read_unlock(); 5579 } 5580 5581 /** 5582 * netif_receive_skb - process receive buffer from network 5583 * @skb: buffer to process 5584 * 5585 * netif_receive_skb() is the main receive data processing function. 5586 * It always succeeds. The buffer may be dropped during processing 5587 * for congestion control or by the protocol layers. 5588 * 5589 * This function may only be called from softirq context and interrupts 5590 * should be enabled. 5591 * 5592 * Return values (usually ignored): 5593 * NET_RX_SUCCESS: no congestion 5594 * NET_RX_DROP: packet was dropped 5595 */ 5596 int netif_receive_skb(struct sk_buff *skb) 5597 { 5598 int ret; 5599 5600 trace_netif_receive_skb_entry(skb); 5601 5602 ret = netif_receive_skb_internal(skb); 5603 trace_netif_receive_skb_exit(ret); 5604 5605 return ret; 5606 } 5607 EXPORT_SYMBOL(netif_receive_skb); 5608 5609 /** 5610 * netif_receive_skb_list - process many receive buffers from network 5611 * @head: list of skbs to process. 5612 * 5613 * Since return value of netif_receive_skb() is normally ignored, and 5614 * wouldn't be meaningful for a list, this function returns void. 5615 * 5616 * This function may only be called from softirq context and interrupts 5617 * should be enabled. 5618 */ 5619 void netif_receive_skb_list(struct list_head *head) 5620 { 5621 struct sk_buff *skb; 5622 5623 if (list_empty(head)) 5624 return; 5625 if (trace_netif_receive_skb_list_entry_enabled()) { 5626 list_for_each_entry(skb, head, list) 5627 trace_netif_receive_skb_list_entry(skb); 5628 } 5629 netif_receive_skb_list_internal(head); 5630 trace_netif_receive_skb_list_exit(0); 5631 } 5632 EXPORT_SYMBOL(netif_receive_skb_list); 5633 5634 static DEFINE_PER_CPU(struct work_struct, flush_works); 5635 5636 /* Network device is going away, flush any packets still pending */ 5637 static void flush_backlog(struct work_struct *work) 5638 { 5639 struct sk_buff *skb, *tmp; 5640 struct softnet_data *sd; 5641 5642 local_bh_disable(); 5643 sd = this_cpu_ptr(&softnet_data); 5644 5645 local_irq_disable(); 5646 rps_lock(sd); 5647 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { 5648 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5649 __skb_unlink(skb, &sd->input_pkt_queue); 5650 dev_kfree_skb_irq(skb); 5651 input_queue_head_incr(sd); 5652 } 5653 } 5654 rps_unlock(sd); 5655 local_irq_enable(); 5656 5657 skb_queue_walk_safe(&sd->process_queue, skb, tmp) { 5658 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5659 __skb_unlink(skb, &sd->process_queue); 5660 kfree_skb(skb); 5661 input_queue_head_incr(sd); 5662 } 5663 } 5664 local_bh_enable(); 5665 } 5666 5667 static bool flush_required(int cpu) 5668 { 5669 #if IS_ENABLED(CONFIG_RPS) 5670 struct softnet_data *sd = &per_cpu(softnet_data, cpu); 5671 bool do_flush; 5672 5673 local_irq_disable(); 5674 rps_lock(sd); 5675 5676 /* as insertion into process_queue happens with the rps lock held, 5677 * process_queue access may race only with dequeue 5678 */ 5679 do_flush = !skb_queue_empty(&sd->input_pkt_queue) || 5680 !skb_queue_empty_lockless(&sd->process_queue); 5681 rps_unlock(sd); 5682 local_irq_enable(); 5683 5684 return do_flush; 5685 #endif 5686 /* without RPS we can't safely check input_pkt_queue: during a 5687 * concurrent remote skb_queue_splice() we can detect as empty both 5688 * input_pkt_queue and process_queue even if the latter could end-up 5689 * containing a lot of packets. 5690 */ 5691 return true; 5692 } 5693 5694 static void flush_all_backlogs(void) 5695 { 5696 static cpumask_t flush_cpus; 5697 unsigned int cpu; 5698 5699 /* since we are under rtnl lock protection we can use static data 5700 * for the cpumask and avoid allocating on stack the possibly 5701 * large mask 5702 */ 5703 ASSERT_RTNL(); 5704 5705 cpus_read_lock(); 5706 5707 cpumask_clear(&flush_cpus); 5708 for_each_online_cpu(cpu) { 5709 if (flush_required(cpu)) { 5710 queue_work_on(cpu, system_highpri_wq, 5711 per_cpu_ptr(&flush_works, cpu)); 5712 cpumask_set_cpu(cpu, &flush_cpus); 5713 } 5714 } 5715 5716 /* we can have in flight packet[s] on the cpus we are not flushing, 5717 * synchronize_net() in unregister_netdevice_many() will take care of 5718 * them 5719 */ 5720 for_each_cpu(cpu, &flush_cpus) 5721 flush_work(per_cpu_ptr(&flush_works, cpu)); 5722 5723 cpus_read_unlock(); 5724 } 5725 5726 static void net_rps_send_ipi(struct softnet_data *remsd) 5727 { 5728 #ifdef CONFIG_RPS 5729 while (remsd) { 5730 struct softnet_data *next = remsd->rps_ipi_next; 5731 5732 if (cpu_online(remsd->cpu)) 5733 smp_call_function_single_async(remsd->cpu, &remsd->csd); 5734 remsd = next; 5735 } 5736 #endif 5737 } 5738 5739 /* 5740 * net_rps_action_and_irq_enable sends any pending IPI's for rps. 5741 * Note: called with local irq disabled, but exits with local irq enabled. 5742 */ 5743 static void net_rps_action_and_irq_enable(struct softnet_data *sd) 5744 { 5745 #ifdef CONFIG_RPS 5746 struct softnet_data *remsd = sd->rps_ipi_list; 5747 5748 if (remsd) { 5749 sd->rps_ipi_list = NULL; 5750 5751 local_irq_enable(); 5752 5753 /* Send pending IPI's to kick RPS processing on remote cpus. */ 5754 net_rps_send_ipi(remsd); 5755 } else 5756 #endif 5757 local_irq_enable(); 5758 } 5759 5760 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) 5761 { 5762 #ifdef CONFIG_RPS 5763 return sd->rps_ipi_list != NULL; 5764 #else 5765 return false; 5766 #endif 5767 } 5768 5769 static int process_backlog(struct napi_struct *napi, int quota) 5770 { 5771 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); 5772 bool again = true; 5773 int work = 0; 5774 5775 /* Check if we have pending ipi, its better to send them now, 5776 * not waiting net_rx_action() end. 5777 */ 5778 if (sd_has_rps_ipi_waiting(sd)) { 5779 local_irq_disable(); 5780 net_rps_action_and_irq_enable(sd); 5781 } 5782 5783 napi->weight = dev_rx_weight; 5784 while (again) { 5785 struct sk_buff *skb; 5786 5787 while ((skb = __skb_dequeue(&sd->process_queue))) { 5788 rcu_read_lock(); 5789 __netif_receive_skb(skb); 5790 rcu_read_unlock(); 5791 input_queue_head_incr(sd); 5792 if (++work >= quota) 5793 return work; 5794 5795 } 5796 5797 local_irq_disable(); 5798 rps_lock(sd); 5799 if (skb_queue_empty(&sd->input_pkt_queue)) { 5800 /* 5801 * Inline a custom version of __napi_complete(). 5802 * only current cpu owns and manipulates this napi, 5803 * and NAPI_STATE_SCHED is the only possible flag set 5804 * on backlog. 5805 * We can use a plain write instead of clear_bit(), 5806 * and we dont need an smp_mb() memory barrier. 5807 */ 5808 napi->state = 0; 5809 again = false; 5810 } else { 5811 skb_queue_splice_tail_init(&sd->input_pkt_queue, 5812 &sd->process_queue); 5813 } 5814 rps_unlock(sd); 5815 local_irq_enable(); 5816 } 5817 5818 return work; 5819 } 5820 5821 /** 5822 * __napi_schedule - schedule for receive 5823 * @n: entry to schedule 5824 * 5825 * The entry's receive function will be scheduled to run. 5826 * Consider using __napi_schedule_irqoff() if hard irqs are masked. 5827 */ 5828 void __napi_schedule(struct napi_struct *n) 5829 { 5830 unsigned long flags; 5831 5832 local_irq_save(flags); 5833 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 5834 local_irq_restore(flags); 5835 } 5836 EXPORT_SYMBOL(__napi_schedule); 5837 5838 /** 5839 * napi_schedule_prep - check if napi can be scheduled 5840 * @n: napi context 5841 * 5842 * Test if NAPI routine is already running, and if not mark 5843 * it as running. This is used as a condition variable to 5844 * insure only one NAPI poll instance runs. We also make 5845 * sure there is no pending NAPI disable. 5846 */ 5847 bool napi_schedule_prep(struct napi_struct *n) 5848 { 5849 unsigned long val, new; 5850 5851 do { 5852 val = READ_ONCE(n->state); 5853 if (unlikely(val & NAPIF_STATE_DISABLE)) 5854 return false; 5855 new = val | NAPIF_STATE_SCHED; 5856 5857 /* Sets STATE_MISSED bit if STATE_SCHED was already set 5858 * This was suggested by Alexander Duyck, as compiler 5859 * emits better code than : 5860 * if (val & NAPIF_STATE_SCHED) 5861 * new |= NAPIF_STATE_MISSED; 5862 */ 5863 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * 5864 NAPIF_STATE_MISSED; 5865 } while (cmpxchg(&n->state, val, new) != val); 5866 5867 return !(val & NAPIF_STATE_SCHED); 5868 } 5869 EXPORT_SYMBOL(napi_schedule_prep); 5870 5871 /** 5872 * __napi_schedule_irqoff - schedule for receive 5873 * @n: entry to schedule 5874 * 5875 * Variant of __napi_schedule() assuming hard irqs are masked. 5876 * 5877 * On PREEMPT_RT enabled kernels this maps to __napi_schedule() 5878 * because the interrupt disabled assumption might not be true 5879 * due to force-threaded interrupts and spinlock substitution. 5880 */ 5881 void __napi_schedule_irqoff(struct napi_struct *n) 5882 { 5883 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 5884 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 5885 else 5886 __napi_schedule(n); 5887 } 5888 EXPORT_SYMBOL(__napi_schedule_irqoff); 5889 5890 bool napi_complete_done(struct napi_struct *n, int work_done) 5891 { 5892 unsigned long flags, val, new, timeout = 0; 5893 bool ret = true; 5894 5895 /* 5896 * 1) Don't let napi dequeue from the cpu poll list 5897 * just in case its running on a different cpu. 5898 * 2) If we are busy polling, do nothing here, we have 5899 * the guarantee we will be called later. 5900 */ 5901 if (unlikely(n->state & (NAPIF_STATE_NPSVC | 5902 NAPIF_STATE_IN_BUSY_POLL))) 5903 return false; 5904 5905 if (work_done) { 5906 if (n->gro_bitmask) 5907 timeout = READ_ONCE(n->dev->gro_flush_timeout); 5908 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs); 5909 } 5910 if (n->defer_hard_irqs_count > 0) { 5911 n->defer_hard_irqs_count--; 5912 timeout = READ_ONCE(n->dev->gro_flush_timeout); 5913 if (timeout) 5914 ret = false; 5915 } 5916 if (n->gro_bitmask) { 5917 /* When the NAPI instance uses a timeout and keeps postponing 5918 * it, we need to bound somehow the time packets are kept in 5919 * the GRO layer 5920 */ 5921 napi_gro_flush(n, !!timeout); 5922 } 5923 5924 gro_normal_list(n); 5925 5926 if (unlikely(!list_empty(&n->poll_list))) { 5927 /* If n->poll_list is not empty, we need to mask irqs */ 5928 local_irq_save(flags); 5929 list_del_init(&n->poll_list); 5930 local_irq_restore(flags); 5931 } 5932 5933 do { 5934 val = READ_ONCE(n->state); 5935 5936 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); 5937 5938 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED | 5939 NAPIF_STATE_SCHED_THREADED | 5940 NAPIF_STATE_PREFER_BUSY_POLL); 5941 5942 /* If STATE_MISSED was set, leave STATE_SCHED set, 5943 * because we will call napi->poll() one more time. 5944 * This C code was suggested by Alexander Duyck to help gcc. 5945 */ 5946 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * 5947 NAPIF_STATE_SCHED; 5948 } while (cmpxchg(&n->state, val, new) != val); 5949 5950 if (unlikely(val & NAPIF_STATE_MISSED)) { 5951 __napi_schedule(n); 5952 return false; 5953 } 5954 5955 if (timeout) 5956 hrtimer_start(&n->timer, ns_to_ktime(timeout), 5957 HRTIMER_MODE_REL_PINNED); 5958 return ret; 5959 } 5960 EXPORT_SYMBOL(napi_complete_done); 5961 5962 /* must be called under rcu_read_lock(), as we dont take a reference */ 5963 static struct napi_struct *napi_by_id(unsigned int napi_id) 5964 { 5965 unsigned int hash = napi_id % HASH_SIZE(napi_hash); 5966 struct napi_struct *napi; 5967 5968 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) 5969 if (napi->napi_id == napi_id) 5970 return napi; 5971 5972 return NULL; 5973 } 5974 5975 #if defined(CONFIG_NET_RX_BUSY_POLL) 5976 5977 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule) 5978 { 5979 if (!skip_schedule) { 5980 gro_normal_list(napi); 5981 __napi_schedule(napi); 5982 return; 5983 } 5984 5985 if (napi->gro_bitmask) { 5986 /* flush too old packets 5987 * If HZ < 1000, flush all packets. 5988 */ 5989 napi_gro_flush(napi, HZ >= 1000); 5990 } 5991 5992 gro_normal_list(napi); 5993 clear_bit(NAPI_STATE_SCHED, &napi->state); 5994 } 5995 5996 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll, 5997 u16 budget) 5998 { 5999 bool skip_schedule = false; 6000 unsigned long timeout; 6001 int rc; 6002 6003 /* Busy polling means there is a high chance device driver hard irq 6004 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was 6005 * set in napi_schedule_prep(). 6006 * Since we are about to call napi->poll() once more, we can safely 6007 * clear NAPI_STATE_MISSED. 6008 * 6009 * Note: x86 could use a single "lock and ..." instruction 6010 * to perform these two clear_bit() 6011 */ 6012 clear_bit(NAPI_STATE_MISSED, &napi->state); 6013 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); 6014 6015 local_bh_disable(); 6016 6017 if (prefer_busy_poll) { 6018 napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs); 6019 timeout = READ_ONCE(napi->dev->gro_flush_timeout); 6020 if (napi->defer_hard_irqs_count && timeout) { 6021 hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED); 6022 skip_schedule = true; 6023 } 6024 } 6025 6026 /* All we really want here is to re-enable device interrupts. 6027 * Ideally, a new ndo_busy_poll_stop() could avoid another round. 6028 */ 6029 rc = napi->poll(napi, budget); 6030 /* We can't gro_normal_list() here, because napi->poll() might have 6031 * rearmed the napi (napi_complete_done()) in which case it could 6032 * already be running on another CPU. 6033 */ 6034 trace_napi_poll(napi, rc, budget); 6035 netpoll_poll_unlock(have_poll_lock); 6036 if (rc == budget) 6037 __busy_poll_stop(napi, skip_schedule); 6038 local_bh_enable(); 6039 } 6040 6041 void napi_busy_loop(unsigned int napi_id, 6042 bool (*loop_end)(void *, unsigned long), 6043 void *loop_end_arg, bool prefer_busy_poll, u16 budget) 6044 { 6045 unsigned long start_time = loop_end ? busy_loop_current_time() : 0; 6046 int (*napi_poll)(struct napi_struct *napi, int budget); 6047 void *have_poll_lock = NULL; 6048 struct napi_struct *napi; 6049 6050 restart: 6051 napi_poll = NULL; 6052 6053 rcu_read_lock(); 6054 6055 napi = napi_by_id(napi_id); 6056 if (!napi) 6057 goto out; 6058 6059 preempt_disable(); 6060 for (;;) { 6061 int work = 0; 6062 6063 local_bh_disable(); 6064 if (!napi_poll) { 6065 unsigned long val = READ_ONCE(napi->state); 6066 6067 /* If multiple threads are competing for this napi, 6068 * we avoid dirtying napi->state as much as we can. 6069 */ 6070 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | 6071 NAPIF_STATE_IN_BUSY_POLL)) { 6072 if (prefer_busy_poll) 6073 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6074 goto count; 6075 } 6076 if (cmpxchg(&napi->state, val, 6077 val | NAPIF_STATE_IN_BUSY_POLL | 6078 NAPIF_STATE_SCHED) != val) { 6079 if (prefer_busy_poll) 6080 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6081 goto count; 6082 } 6083 have_poll_lock = netpoll_poll_lock(napi); 6084 napi_poll = napi->poll; 6085 } 6086 work = napi_poll(napi, budget); 6087 trace_napi_poll(napi, work, budget); 6088 gro_normal_list(napi); 6089 count: 6090 if (work > 0) 6091 __NET_ADD_STATS(dev_net(napi->dev), 6092 LINUX_MIB_BUSYPOLLRXPACKETS, work); 6093 local_bh_enable(); 6094 6095 if (!loop_end || loop_end(loop_end_arg, start_time)) 6096 break; 6097 6098 if (unlikely(need_resched())) { 6099 if (napi_poll) 6100 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6101 preempt_enable(); 6102 rcu_read_unlock(); 6103 cond_resched(); 6104 if (loop_end(loop_end_arg, start_time)) 6105 return; 6106 goto restart; 6107 } 6108 cpu_relax(); 6109 } 6110 if (napi_poll) 6111 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6112 preempt_enable(); 6113 out: 6114 rcu_read_unlock(); 6115 } 6116 EXPORT_SYMBOL(napi_busy_loop); 6117 6118 #endif /* CONFIG_NET_RX_BUSY_POLL */ 6119 6120 static void napi_hash_add(struct napi_struct *napi) 6121 { 6122 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state)) 6123 return; 6124 6125 spin_lock(&napi_hash_lock); 6126 6127 /* 0..NR_CPUS range is reserved for sender_cpu use */ 6128 do { 6129 if (unlikely(++napi_gen_id < MIN_NAPI_ID)) 6130 napi_gen_id = MIN_NAPI_ID; 6131 } while (napi_by_id(napi_gen_id)); 6132 napi->napi_id = napi_gen_id; 6133 6134 hlist_add_head_rcu(&napi->napi_hash_node, 6135 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); 6136 6137 spin_unlock(&napi_hash_lock); 6138 } 6139 6140 /* Warning : caller is responsible to make sure rcu grace period 6141 * is respected before freeing memory containing @napi 6142 */ 6143 static void napi_hash_del(struct napi_struct *napi) 6144 { 6145 spin_lock(&napi_hash_lock); 6146 6147 hlist_del_init_rcu(&napi->napi_hash_node); 6148 6149 spin_unlock(&napi_hash_lock); 6150 } 6151 6152 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) 6153 { 6154 struct napi_struct *napi; 6155 6156 napi = container_of(timer, struct napi_struct, timer); 6157 6158 /* Note : we use a relaxed variant of napi_schedule_prep() not setting 6159 * NAPI_STATE_MISSED, since we do not react to a device IRQ. 6160 */ 6161 if (!napi_disable_pending(napi) && 6162 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) { 6163 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6164 __napi_schedule_irqoff(napi); 6165 } 6166 6167 return HRTIMER_NORESTART; 6168 } 6169 6170 static void init_gro_hash(struct napi_struct *napi) 6171 { 6172 int i; 6173 6174 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6175 INIT_LIST_HEAD(&napi->gro_hash[i].list); 6176 napi->gro_hash[i].count = 0; 6177 } 6178 napi->gro_bitmask = 0; 6179 } 6180 6181 int dev_set_threaded(struct net_device *dev, bool threaded) 6182 { 6183 struct napi_struct *napi; 6184 int err = 0; 6185 6186 if (dev->threaded == threaded) 6187 return 0; 6188 6189 if (threaded) { 6190 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6191 if (!napi->thread) { 6192 err = napi_kthread_create(napi); 6193 if (err) { 6194 threaded = false; 6195 break; 6196 } 6197 } 6198 } 6199 } 6200 6201 dev->threaded = threaded; 6202 6203 /* Make sure kthread is created before THREADED bit 6204 * is set. 6205 */ 6206 smp_mb__before_atomic(); 6207 6208 /* Setting/unsetting threaded mode on a napi might not immediately 6209 * take effect, if the current napi instance is actively being 6210 * polled. In this case, the switch between threaded mode and 6211 * softirq mode will happen in the next round of napi_schedule(). 6212 * This should not cause hiccups/stalls to the live traffic. 6213 */ 6214 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6215 if (threaded) 6216 set_bit(NAPI_STATE_THREADED, &napi->state); 6217 else 6218 clear_bit(NAPI_STATE_THREADED, &napi->state); 6219 } 6220 6221 return err; 6222 } 6223 EXPORT_SYMBOL(dev_set_threaded); 6224 6225 void netif_napi_add(struct net_device *dev, struct napi_struct *napi, 6226 int (*poll)(struct napi_struct *, int), int weight) 6227 { 6228 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state))) 6229 return; 6230 6231 INIT_LIST_HEAD(&napi->poll_list); 6232 INIT_HLIST_NODE(&napi->napi_hash_node); 6233 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); 6234 napi->timer.function = napi_watchdog; 6235 init_gro_hash(napi); 6236 napi->skb = NULL; 6237 INIT_LIST_HEAD(&napi->rx_list); 6238 napi->rx_count = 0; 6239 napi->poll = poll; 6240 if (weight > NAPI_POLL_WEIGHT) 6241 netdev_err_once(dev, "%s() called with weight %d\n", __func__, 6242 weight); 6243 napi->weight = weight; 6244 napi->dev = dev; 6245 #ifdef CONFIG_NETPOLL 6246 napi->poll_owner = -1; 6247 #endif 6248 set_bit(NAPI_STATE_SCHED, &napi->state); 6249 set_bit(NAPI_STATE_NPSVC, &napi->state); 6250 list_add_rcu(&napi->dev_list, &dev->napi_list); 6251 napi_hash_add(napi); 6252 /* Create kthread for this napi if dev->threaded is set. 6253 * Clear dev->threaded if kthread creation failed so that 6254 * threaded mode will not be enabled in napi_enable(). 6255 */ 6256 if (dev->threaded && napi_kthread_create(napi)) 6257 dev->threaded = 0; 6258 } 6259 EXPORT_SYMBOL(netif_napi_add); 6260 6261 void napi_disable(struct napi_struct *n) 6262 { 6263 unsigned long val, new; 6264 6265 might_sleep(); 6266 set_bit(NAPI_STATE_DISABLE, &n->state); 6267 6268 for ( ; ; ) { 6269 val = READ_ONCE(n->state); 6270 if (val & (NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC)) { 6271 usleep_range(20, 200); 6272 continue; 6273 } 6274 6275 new = val | NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC; 6276 new &= ~(NAPIF_STATE_THREADED | NAPIF_STATE_PREFER_BUSY_POLL); 6277 6278 if (cmpxchg(&n->state, val, new) == val) 6279 break; 6280 } 6281 6282 hrtimer_cancel(&n->timer); 6283 6284 clear_bit(NAPI_STATE_DISABLE, &n->state); 6285 } 6286 EXPORT_SYMBOL(napi_disable); 6287 6288 /** 6289 * napi_enable - enable NAPI scheduling 6290 * @n: NAPI context 6291 * 6292 * Resume NAPI from being scheduled on this context. 6293 * Must be paired with napi_disable. 6294 */ 6295 void napi_enable(struct napi_struct *n) 6296 { 6297 unsigned long val, new; 6298 6299 do { 6300 val = READ_ONCE(n->state); 6301 BUG_ON(!test_bit(NAPI_STATE_SCHED, &val)); 6302 6303 new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC); 6304 if (n->dev->threaded && n->thread) 6305 new |= NAPIF_STATE_THREADED; 6306 } while (cmpxchg(&n->state, val, new) != val); 6307 } 6308 EXPORT_SYMBOL(napi_enable); 6309 6310 static void flush_gro_hash(struct napi_struct *napi) 6311 { 6312 int i; 6313 6314 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6315 struct sk_buff *skb, *n; 6316 6317 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list) 6318 kfree_skb(skb); 6319 napi->gro_hash[i].count = 0; 6320 } 6321 } 6322 6323 /* Must be called in process context */ 6324 void __netif_napi_del(struct napi_struct *napi) 6325 { 6326 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state)) 6327 return; 6328 6329 napi_hash_del(napi); 6330 list_del_rcu(&napi->dev_list); 6331 napi_free_frags(napi); 6332 6333 flush_gro_hash(napi); 6334 napi->gro_bitmask = 0; 6335 6336 if (napi->thread) { 6337 kthread_stop(napi->thread); 6338 napi->thread = NULL; 6339 } 6340 } 6341 EXPORT_SYMBOL(__netif_napi_del); 6342 6343 static int __napi_poll(struct napi_struct *n, bool *repoll) 6344 { 6345 int work, weight; 6346 6347 weight = n->weight; 6348 6349 /* This NAPI_STATE_SCHED test is for avoiding a race 6350 * with netpoll's poll_napi(). Only the entity which 6351 * obtains the lock and sees NAPI_STATE_SCHED set will 6352 * actually make the ->poll() call. Therefore we avoid 6353 * accidentally calling ->poll() when NAPI is not scheduled. 6354 */ 6355 work = 0; 6356 if (test_bit(NAPI_STATE_SCHED, &n->state)) { 6357 work = n->poll(n, weight); 6358 trace_napi_poll(n, work, weight); 6359 } 6360 6361 if (unlikely(work > weight)) 6362 netdev_err_once(n->dev, "NAPI poll function %pS returned %d, exceeding its budget of %d.\n", 6363 n->poll, work, weight); 6364 6365 if (likely(work < weight)) 6366 return work; 6367 6368 /* Drivers must not modify the NAPI state if they 6369 * consume the entire weight. In such cases this code 6370 * still "owns" the NAPI instance and therefore can 6371 * move the instance around on the list at-will. 6372 */ 6373 if (unlikely(napi_disable_pending(n))) { 6374 napi_complete(n); 6375 return work; 6376 } 6377 6378 /* The NAPI context has more processing work, but busy-polling 6379 * is preferred. Exit early. 6380 */ 6381 if (napi_prefer_busy_poll(n)) { 6382 if (napi_complete_done(n, work)) { 6383 /* If timeout is not set, we need to make sure 6384 * that the NAPI is re-scheduled. 6385 */ 6386 napi_schedule(n); 6387 } 6388 return work; 6389 } 6390 6391 if (n->gro_bitmask) { 6392 /* flush too old packets 6393 * If HZ < 1000, flush all packets. 6394 */ 6395 napi_gro_flush(n, HZ >= 1000); 6396 } 6397 6398 gro_normal_list(n); 6399 6400 /* Some drivers may have called napi_schedule 6401 * prior to exhausting their budget. 6402 */ 6403 if (unlikely(!list_empty(&n->poll_list))) { 6404 pr_warn_once("%s: Budget exhausted after napi rescheduled\n", 6405 n->dev ? n->dev->name : "backlog"); 6406 return work; 6407 } 6408 6409 *repoll = true; 6410 6411 return work; 6412 } 6413 6414 static int napi_poll(struct napi_struct *n, struct list_head *repoll) 6415 { 6416 bool do_repoll = false; 6417 void *have; 6418 int work; 6419 6420 list_del_init(&n->poll_list); 6421 6422 have = netpoll_poll_lock(n); 6423 6424 work = __napi_poll(n, &do_repoll); 6425 6426 if (do_repoll) 6427 list_add_tail(&n->poll_list, repoll); 6428 6429 netpoll_poll_unlock(have); 6430 6431 return work; 6432 } 6433 6434 static int napi_thread_wait(struct napi_struct *napi) 6435 { 6436 bool woken = false; 6437 6438 set_current_state(TASK_INTERRUPTIBLE); 6439 6440 while (!kthread_should_stop()) { 6441 /* Testing SCHED_THREADED bit here to make sure the current 6442 * kthread owns this napi and could poll on this napi. 6443 * Testing SCHED bit is not enough because SCHED bit might be 6444 * set by some other busy poll thread or by napi_disable(). 6445 */ 6446 if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state) || woken) { 6447 WARN_ON(!list_empty(&napi->poll_list)); 6448 __set_current_state(TASK_RUNNING); 6449 return 0; 6450 } 6451 6452 schedule(); 6453 /* woken being true indicates this thread owns this napi. */ 6454 woken = true; 6455 set_current_state(TASK_INTERRUPTIBLE); 6456 } 6457 __set_current_state(TASK_RUNNING); 6458 6459 return -1; 6460 } 6461 6462 static int napi_threaded_poll(void *data) 6463 { 6464 struct napi_struct *napi = data; 6465 void *have; 6466 6467 while (!napi_thread_wait(napi)) { 6468 for (;;) { 6469 bool repoll = false; 6470 6471 local_bh_disable(); 6472 6473 have = netpoll_poll_lock(napi); 6474 __napi_poll(napi, &repoll); 6475 netpoll_poll_unlock(have); 6476 6477 local_bh_enable(); 6478 6479 if (!repoll) 6480 break; 6481 6482 cond_resched(); 6483 } 6484 } 6485 return 0; 6486 } 6487 6488 static __latent_entropy void net_rx_action(struct softirq_action *h) 6489 { 6490 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 6491 unsigned long time_limit = jiffies + 6492 usecs_to_jiffies(netdev_budget_usecs); 6493 int budget = netdev_budget; 6494 LIST_HEAD(list); 6495 LIST_HEAD(repoll); 6496 6497 local_irq_disable(); 6498 list_splice_init(&sd->poll_list, &list); 6499 local_irq_enable(); 6500 6501 for (;;) { 6502 struct napi_struct *n; 6503 6504 if (list_empty(&list)) { 6505 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll)) 6506 return; 6507 break; 6508 } 6509 6510 n = list_first_entry(&list, struct napi_struct, poll_list); 6511 budget -= napi_poll(n, &repoll); 6512 6513 /* If softirq window is exhausted then punt. 6514 * Allow this to run for 2 jiffies since which will allow 6515 * an average latency of 1.5/HZ. 6516 */ 6517 if (unlikely(budget <= 0 || 6518 time_after_eq(jiffies, time_limit))) { 6519 sd->time_squeeze++; 6520 break; 6521 } 6522 } 6523 6524 local_irq_disable(); 6525 6526 list_splice_tail_init(&sd->poll_list, &list); 6527 list_splice_tail(&repoll, &list); 6528 list_splice(&list, &sd->poll_list); 6529 if (!list_empty(&sd->poll_list)) 6530 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 6531 6532 net_rps_action_and_irq_enable(sd); 6533 } 6534 6535 struct netdev_adjacent { 6536 struct net_device *dev; 6537 6538 /* upper master flag, there can only be one master device per list */ 6539 bool master; 6540 6541 /* lookup ignore flag */ 6542 bool ignore; 6543 6544 /* counter for the number of times this device was added to us */ 6545 u16 ref_nr; 6546 6547 /* private field for the users */ 6548 void *private; 6549 6550 struct list_head list; 6551 struct rcu_head rcu; 6552 }; 6553 6554 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, 6555 struct list_head *adj_list) 6556 { 6557 struct netdev_adjacent *adj; 6558 6559 list_for_each_entry(adj, adj_list, list) { 6560 if (adj->dev == adj_dev) 6561 return adj; 6562 } 6563 return NULL; 6564 } 6565 6566 static int ____netdev_has_upper_dev(struct net_device *upper_dev, 6567 struct netdev_nested_priv *priv) 6568 { 6569 struct net_device *dev = (struct net_device *)priv->data; 6570 6571 return upper_dev == dev; 6572 } 6573 6574 /** 6575 * netdev_has_upper_dev - Check if device is linked to an upper device 6576 * @dev: device 6577 * @upper_dev: upper device to check 6578 * 6579 * Find out if a device is linked to specified upper device and return true 6580 * in case it is. Note that this checks only immediate upper device, 6581 * not through a complete stack of devices. The caller must hold the RTNL lock. 6582 */ 6583 bool netdev_has_upper_dev(struct net_device *dev, 6584 struct net_device *upper_dev) 6585 { 6586 struct netdev_nested_priv priv = { 6587 .data = (void *)upper_dev, 6588 }; 6589 6590 ASSERT_RTNL(); 6591 6592 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 6593 &priv); 6594 } 6595 EXPORT_SYMBOL(netdev_has_upper_dev); 6596 6597 /** 6598 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device 6599 * @dev: device 6600 * @upper_dev: upper device to check 6601 * 6602 * Find out if a device is linked to specified upper device and return true 6603 * in case it is. Note that this checks the entire upper device chain. 6604 * The caller must hold rcu lock. 6605 */ 6606 6607 bool netdev_has_upper_dev_all_rcu(struct net_device *dev, 6608 struct net_device *upper_dev) 6609 { 6610 struct netdev_nested_priv priv = { 6611 .data = (void *)upper_dev, 6612 }; 6613 6614 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 6615 &priv); 6616 } 6617 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); 6618 6619 /** 6620 * netdev_has_any_upper_dev - Check if device is linked to some device 6621 * @dev: device 6622 * 6623 * Find out if a device is linked to an upper device and return true in case 6624 * it is. The caller must hold the RTNL lock. 6625 */ 6626 bool netdev_has_any_upper_dev(struct net_device *dev) 6627 { 6628 ASSERT_RTNL(); 6629 6630 return !list_empty(&dev->adj_list.upper); 6631 } 6632 EXPORT_SYMBOL(netdev_has_any_upper_dev); 6633 6634 /** 6635 * netdev_master_upper_dev_get - Get master upper device 6636 * @dev: device 6637 * 6638 * Find a master upper device and return pointer to it or NULL in case 6639 * it's not there. The caller must hold the RTNL lock. 6640 */ 6641 struct net_device *netdev_master_upper_dev_get(struct net_device *dev) 6642 { 6643 struct netdev_adjacent *upper; 6644 6645 ASSERT_RTNL(); 6646 6647 if (list_empty(&dev->adj_list.upper)) 6648 return NULL; 6649 6650 upper = list_first_entry(&dev->adj_list.upper, 6651 struct netdev_adjacent, list); 6652 if (likely(upper->master)) 6653 return upper->dev; 6654 return NULL; 6655 } 6656 EXPORT_SYMBOL(netdev_master_upper_dev_get); 6657 6658 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev) 6659 { 6660 struct netdev_adjacent *upper; 6661 6662 ASSERT_RTNL(); 6663 6664 if (list_empty(&dev->adj_list.upper)) 6665 return NULL; 6666 6667 upper = list_first_entry(&dev->adj_list.upper, 6668 struct netdev_adjacent, list); 6669 if (likely(upper->master) && !upper->ignore) 6670 return upper->dev; 6671 return NULL; 6672 } 6673 6674 /** 6675 * netdev_has_any_lower_dev - Check if device is linked to some device 6676 * @dev: device 6677 * 6678 * Find out if a device is linked to a lower device and return true in case 6679 * it is. The caller must hold the RTNL lock. 6680 */ 6681 static bool netdev_has_any_lower_dev(struct net_device *dev) 6682 { 6683 ASSERT_RTNL(); 6684 6685 return !list_empty(&dev->adj_list.lower); 6686 } 6687 6688 void *netdev_adjacent_get_private(struct list_head *adj_list) 6689 { 6690 struct netdev_adjacent *adj; 6691 6692 adj = list_entry(adj_list, struct netdev_adjacent, list); 6693 6694 return adj->private; 6695 } 6696 EXPORT_SYMBOL(netdev_adjacent_get_private); 6697 6698 /** 6699 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list 6700 * @dev: device 6701 * @iter: list_head ** of the current position 6702 * 6703 * Gets the next device from the dev's upper list, starting from iter 6704 * position. The caller must hold RCU read lock. 6705 */ 6706 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, 6707 struct list_head **iter) 6708 { 6709 struct netdev_adjacent *upper; 6710 6711 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 6712 6713 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 6714 6715 if (&upper->list == &dev->adj_list.upper) 6716 return NULL; 6717 6718 *iter = &upper->list; 6719 6720 return upper->dev; 6721 } 6722 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); 6723 6724 static struct net_device *__netdev_next_upper_dev(struct net_device *dev, 6725 struct list_head **iter, 6726 bool *ignore) 6727 { 6728 struct netdev_adjacent *upper; 6729 6730 upper = list_entry((*iter)->next, struct netdev_adjacent, list); 6731 6732 if (&upper->list == &dev->adj_list.upper) 6733 return NULL; 6734 6735 *iter = &upper->list; 6736 *ignore = upper->ignore; 6737 6738 return upper->dev; 6739 } 6740 6741 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, 6742 struct list_head **iter) 6743 { 6744 struct netdev_adjacent *upper; 6745 6746 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 6747 6748 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 6749 6750 if (&upper->list == &dev->adj_list.upper) 6751 return NULL; 6752 6753 *iter = &upper->list; 6754 6755 return upper->dev; 6756 } 6757 6758 static int __netdev_walk_all_upper_dev(struct net_device *dev, 6759 int (*fn)(struct net_device *dev, 6760 struct netdev_nested_priv *priv), 6761 struct netdev_nested_priv *priv) 6762 { 6763 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 6764 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 6765 int ret, cur = 0; 6766 bool ignore; 6767 6768 now = dev; 6769 iter = &dev->adj_list.upper; 6770 6771 while (1) { 6772 if (now != dev) { 6773 ret = fn(now, priv); 6774 if (ret) 6775 return ret; 6776 } 6777 6778 next = NULL; 6779 while (1) { 6780 udev = __netdev_next_upper_dev(now, &iter, &ignore); 6781 if (!udev) 6782 break; 6783 if (ignore) 6784 continue; 6785 6786 next = udev; 6787 niter = &udev->adj_list.upper; 6788 dev_stack[cur] = now; 6789 iter_stack[cur++] = iter; 6790 break; 6791 } 6792 6793 if (!next) { 6794 if (!cur) 6795 return 0; 6796 next = dev_stack[--cur]; 6797 niter = iter_stack[cur]; 6798 } 6799 6800 now = next; 6801 iter = niter; 6802 } 6803 6804 return 0; 6805 } 6806 6807 int netdev_walk_all_upper_dev_rcu(struct net_device *dev, 6808 int (*fn)(struct net_device *dev, 6809 struct netdev_nested_priv *priv), 6810 struct netdev_nested_priv *priv) 6811 { 6812 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 6813 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 6814 int ret, cur = 0; 6815 6816 now = dev; 6817 iter = &dev->adj_list.upper; 6818 6819 while (1) { 6820 if (now != dev) { 6821 ret = fn(now, priv); 6822 if (ret) 6823 return ret; 6824 } 6825 6826 next = NULL; 6827 while (1) { 6828 udev = netdev_next_upper_dev_rcu(now, &iter); 6829 if (!udev) 6830 break; 6831 6832 next = udev; 6833 niter = &udev->adj_list.upper; 6834 dev_stack[cur] = now; 6835 iter_stack[cur++] = iter; 6836 break; 6837 } 6838 6839 if (!next) { 6840 if (!cur) 6841 return 0; 6842 next = dev_stack[--cur]; 6843 niter = iter_stack[cur]; 6844 } 6845 6846 now = next; 6847 iter = niter; 6848 } 6849 6850 return 0; 6851 } 6852 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); 6853 6854 static bool __netdev_has_upper_dev(struct net_device *dev, 6855 struct net_device *upper_dev) 6856 { 6857 struct netdev_nested_priv priv = { 6858 .flags = 0, 6859 .data = (void *)upper_dev, 6860 }; 6861 6862 ASSERT_RTNL(); 6863 6864 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev, 6865 &priv); 6866 } 6867 6868 /** 6869 * netdev_lower_get_next_private - Get the next ->private from the 6870 * lower neighbour list 6871 * @dev: device 6872 * @iter: list_head ** of the current position 6873 * 6874 * Gets the next netdev_adjacent->private from the dev's lower neighbour 6875 * list, starting from iter position. The caller must hold either hold the 6876 * RTNL lock or its own locking that guarantees that the neighbour lower 6877 * list will remain unchanged. 6878 */ 6879 void *netdev_lower_get_next_private(struct net_device *dev, 6880 struct list_head **iter) 6881 { 6882 struct netdev_adjacent *lower; 6883 6884 lower = list_entry(*iter, struct netdev_adjacent, list); 6885 6886 if (&lower->list == &dev->adj_list.lower) 6887 return NULL; 6888 6889 *iter = lower->list.next; 6890 6891 return lower->private; 6892 } 6893 EXPORT_SYMBOL(netdev_lower_get_next_private); 6894 6895 /** 6896 * netdev_lower_get_next_private_rcu - Get the next ->private from the 6897 * lower neighbour list, RCU 6898 * variant 6899 * @dev: device 6900 * @iter: list_head ** of the current position 6901 * 6902 * Gets the next netdev_adjacent->private from the dev's lower neighbour 6903 * list, starting from iter position. The caller must hold RCU read lock. 6904 */ 6905 void *netdev_lower_get_next_private_rcu(struct net_device *dev, 6906 struct list_head **iter) 6907 { 6908 struct netdev_adjacent *lower; 6909 6910 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 6911 6912 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 6913 6914 if (&lower->list == &dev->adj_list.lower) 6915 return NULL; 6916 6917 *iter = &lower->list; 6918 6919 return lower->private; 6920 } 6921 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); 6922 6923 /** 6924 * netdev_lower_get_next - Get the next device from the lower neighbour 6925 * list 6926 * @dev: device 6927 * @iter: list_head ** of the current position 6928 * 6929 * Gets the next netdev_adjacent from the dev's lower neighbour 6930 * list, starting from iter position. The caller must hold RTNL lock or 6931 * its own locking that guarantees that the neighbour lower 6932 * list will remain unchanged. 6933 */ 6934 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) 6935 { 6936 struct netdev_adjacent *lower; 6937 6938 lower = list_entry(*iter, struct netdev_adjacent, list); 6939 6940 if (&lower->list == &dev->adj_list.lower) 6941 return NULL; 6942 6943 *iter = lower->list.next; 6944 6945 return lower->dev; 6946 } 6947 EXPORT_SYMBOL(netdev_lower_get_next); 6948 6949 static struct net_device *netdev_next_lower_dev(struct net_device *dev, 6950 struct list_head **iter) 6951 { 6952 struct netdev_adjacent *lower; 6953 6954 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 6955 6956 if (&lower->list == &dev->adj_list.lower) 6957 return NULL; 6958 6959 *iter = &lower->list; 6960 6961 return lower->dev; 6962 } 6963 6964 static struct net_device *__netdev_next_lower_dev(struct net_device *dev, 6965 struct list_head **iter, 6966 bool *ignore) 6967 { 6968 struct netdev_adjacent *lower; 6969 6970 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 6971 6972 if (&lower->list == &dev->adj_list.lower) 6973 return NULL; 6974 6975 *iter = &lower->list; 6976 *ignore = lower->ignore; 6977 6978 return lower->dev; 6979 } 6980 6981 int netdev_walk_all_lower_dev(struct net_device *dev, 6982 int (*fn)(struct net_device *dev, 6983 struct netdev_nested_priv *priv), 6984 struct netdev_nested_priv *priv) 6985 { 6986 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 6987 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 6988 int ret, cur = 0; 6989 6990 now = dev; 6991 iter = &dev->adj_list.lower; 6992 6993 while (1) { 6994 if (now != dev) { 6995 ret = fn(now, priv); 6996 if (ret) 6997 return ret; 6998 } 6999 7000 next = NULL; 7001 while (1) { 7002 ldev = netdev_next_lower_dev(now, &iter); 7003 if (!ldev) 7004 break; 7005 7006 next = ldev; 7007 niter = &ldev->adj_list.lower; 7008 dev_stack[cur] = now; 7009 iter_stack[cur++] = iter; 7010 break; 7011 } 7012 7013 if (!next) { 7014 if (!cur) 7015 return 0; 7016 next = dev_stack[--cur]; 7017 niter = iter_stack[cur]; 7018 } 7019 7020 now = next; 7021 iter = niter; 7022 } 7023 7024 return 0; 7025 } 7026 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); 7027 7028 static int __netdev_walk_all_lower_dev(struct net_device *dev, 7029 int (*fn)(struct net_device *dev, 7030 struct netdev_nested_priv *priv), 7031 struct netdev_nested_priv *priv) 7032 { 7033 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7034 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7035 int ret, cur = 0; 7036 bool ignore; 7037 7038 now = dev; 7039 iter = &dev->adj_list.lower; 7040 7041 while (1) { 7042 if (now != dev) { 7043 ret = fn(now, priv); 7044 if (ret) 7045 return ret; 7046 } 7047 7048 next = NULL; 7049 while (1) { 7050 ldev = __netdev_next_lower_dev(now, &iter, &ignore); 7051 if (!ldev) 7052 break; 7053 if (ignore) 7054 continue; 7055 7056 next = ldev; 7057 niter = &ldev->adj_list.lower; 7058 dev_stack[cur] = now; 7059 iter_stack[cur++] = iter; 7060 break; 7061 } 7062 7063 if (!next) { 7064 if (!cur) 7065 return 0; 7066 next = dev_stack[--cur]; 7067 niter = iter_stack[cur]; 7068 } 7069 7070 now = next; 7071 iter = niter; 7072 } 7073 7074 return 0; 7075 } 7076 7077 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, 7078 struct list_head **iter) 7079 { 7080 struct netdev_adjacent *lower; 7081 7082 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7083 if (&lower->list == &dev->adj_list.lower) 7084 return NULL; 7085 7086 *iter = &lower->list; 7087 7088 return lower->dev; 7089 } 7090 EXPORT_SYMBOL(netdev_next_lower_dev_rcu); 7091 7092 static u8 __netdev_upper_depth(struct net_device *dev) 7093 { 7094 struct net_device *udev; 7095 struct list_head *iter; 7096 u8 max_depth = 0; 7097 bool ignore; 7098 7099 for (iter = &dev->adj_list.upper, 7100 udev = __netdev_next_upper_dev(dev, &iter, &ignore); 7101 udev; 7102 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) { 7103 if (ignore) 7104 continue; 7105 if (max_depth < udev->upper_level) 7106 max_depth = udev->upper_level; 7107 } 7108 7109 return max_depth; 7110 } 7111 7112 static u8 __netdev_lower_depth(struct net_device *dev) 7113 { 7114 struct net_device *ldev; 7115 struct list_head *iter; 7116 u8 max_depth = 0; 7117 bool ignore; 7118 7119 for (iter = &dev->adj_list.lower, 7120 ldev = __netdev_next_lower_dev(dev, &iter, &ignore); 7121 ldev; 7122 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) { 7123 if (ignore) 7124 continue; 7125 if (max_depth < ldev->lower_level) 7126 max_depth = ldev->lower_level; 7127 } 7128 7129 return max_depth; 7130 } 7131 7132 static int __netdev_update_upper_level(struct net_device *dev, 7133 struct netdev_nested_priv *__unused) 7134 { 7135 dev->upper_level = __netdev_upper_depth(dev) + 1; 7136 return 0; 7137 } 7138 7139 static int __netdev_update_lower_level(struct net_device *dev, 7140 struct netdev_nested_priv *priv) 7141 { 7142 dev->lower_level = __netdev_lower_depth(dev) + 1; 7143 7144 #ifdef CONFIG_LOCKDEP 7145 if (!priv) 7146 return 0; 7147 7148 if (priv->flags & NESTED_SYNC_IMM) 7149 dev->nested_level = dev->lower_level - 1; 7150 if (priv->flags & NESTED_SYNC_TODO) 7151 net_unlink_todo(dev); 7152 #endif 7153 return 0; 7154 } 7155 7156 int netdev_walk_all_lower_dev_rcu(struct net_device *dev, 7157 int (*fn)(struct net_device *dev, 7158 struct netdev_nested_priv *priv), 7159 struct netdev_nested_priv *priv) 7160 { 7161 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7162 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7163 int ret, cur = 0; 7164 7165 now = dev; 7166 iter = &dev->adj_list.lower; 7167 7168 while (1) { 7169 if (now != dev) { 7170 ret = fn(now, priv); 7171 if (ret) 7172 return ret; 7173 } 7174 7175 next = NULL; 7176 while (1) { 7177 ldev = netdev_next_lower_dev_rcu(now, &iter); 7178 if (!ldev) 7179 break; 7180 7181 next = ldev; 7182 niter = &ldev->adj_list.lower; 7183 dev_stack[cur] = now; 7184 iter_stack[cur++] = iter; 7185 break; 7186 } 7187 7188 if (!next) { 7189 if (!cur) 7190 return 0; 7191 next = dev_stack[--cur]; 7192 niter = iter_stack[cur]; 7193 } 7194 7195 now = next; 7196 iter = niter; 7197 } 7198 7199 return 0; 7200 } 7201 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); 7202 7203 /** 7204 * netdev_lower_get_first_private_rcu - Get the first ->private from the 7205 * lower neighbour list, RCU 7206 * variant 7207 * @dev: device 7208 * 7209 * Gets the first netdev_adjacent->private from the dev's lower neighbour 7210 * list. The caller must hold RCU read lock. 7211 */ 7212 void *netdev_lower_get_first_private_rcu(struct net_device *dev) 7213 { 7214 struct netdev_adjacent *lower; 7215 7216 lower = list_first_or_null_rcu(&dev->adj_list.lower, 7217 struct netdev_adjacent, list); 7218 if (lower) 7219 return lower->private; 7220 return NULL; 7221 } 7222 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); 7223 7224 /** 7225 * netdev_master_upper_dev_get_rcu - Get master upper device 7226 * @dev: device 7227 * 7228 * Find a master upper device and return pointer to it or NULL in case 7229 * it's not there. The caller must hold the RCU read lock. 7230 */ 7231 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) 7232 { 7233 struct netdev_adjacent *upper; 7234 7235 upper = list_first_or_null_rcu(&dev->adj_list.upper, 7236 struct netdev_adjacent, list); 7237 if (upper && likely(upper->master)) 7238 return upper->dev; 7239 return NULL; 7240 } 7241 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); 7242 7243 static int netdev_adjacent_sysfs_add(struct net_device *dev, 7244 struct net_device *adj_dev, 7245 struct list_head *dev_list) 7246 { 7247 char linkname[IFNAMSIZ+7]; 7248 7249 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7250 "upper_%s" : "lower_%s", adj_dev->name); 7251 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), 7252 linkname); 7253 } 7254 static void netdev_adjacent_sysfs_del(struct net_device *dev, 7255 char *name, 7256 struct list_head *dev_list) 7257 { 7258 char linkname[IFNAMSIZ+7]; 7259 7260 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7261 "upper_%s" : "lower_%s", name); 7262 sysfs_remove_link(&(dev->dev.kobj), linkname); 7263 } 7264 7265 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, 7266 struct net_device *adj_dev, 7267 struct list_head *dev_list) 7268 { 7269 return (dev_list == &dev->adj_list.upper || 7270 dev_list == &dev->adj_list.lower) && 7271 net_eq(dev_net(dev), dev_net(adj_dev)); 7272 } 7273 7274 static int __netdev_adjacent_dev_insert(struct net_device *dev, 7275 struct net_device *adj_dev, 7276 struct list_head *dev_list, 7277 void *private, bool master) 7278 { 7279 struct netdev_adjacent *adj; 7280 int ret; 7281 7282 adj = __netdev_find_adj(adj_dev, dev_list); 7283 7284 if (adj) { 7285 adj->ref_nr += 1; 7286 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", 7287 dev->name, adj_dev->name, adj->ref_nr); 7288 7289 return 0; 7290 } 7291 7292 adj = kmalloc(sizeof(*adj), GFP_KERNEL); 7293 if (!adj) 7294 return -ENOMEM; 7295 7296 adj->dev = adj_dev; 7297 adj->master = master; 7298 adj->ref_nr = 1; 7299 adj->private = private; 7300 adj->ignore = false; 7301 dev_hold(adj_dev); 7302 7303 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", 7304 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); 7305 7306 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { 7307 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); 7308 if (ret) 7309 goto free_adj; 7310 } 7311 7312 /* Ensure that master link is always the first item in list. */ 7313 if (master) { 7314 ret = sysfs_create_link(&(dev->dev.kobj), 7315 &(adj_dev->dev.kobj), "master"); 7316 if (ret) 7317 goto remove_symlinks; 7318 7319 list_add_rcu(&adj->list, dev_list); 7320 } else { 7321 list_add_tail_rcu(&adj->list, dev_list); 7322 } 7323 7324 return 0; 7325 7326 remove_symlinks: 7327 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7328 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7329 free_adj: 7330 kfree(adj); 7331 dev_put(adj_dev); 7332 7333 return ret; 7334 } 7335 7336 static void __netdev_adjacent_dev_remove(struct net_device *dev, 7337 struct net_device *adj_dev, 7338 u16 ref_nr, 7339 struct list_head *dev_list) 7340 { 7341 struct netdev_adjacent *adj; 7342 7343 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", 7344 dev->name, adj_dev->name, ref_nr); 7345 7346 adj = __netdev_find_adj(adj_dev, dev_list); 7347 7348 if (!adj) { 7349 pr_err("Adjacency does not exist for device %s from %s\n", 7350 dev->name, adj_dev->name); 7351 WARN_ON(1); 7352 return; 7353 } 7354 7355 if (adj->ref_nr > ref_nr) { 7356 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", 7357 dev->name, adj_dev->name, ref_nr, 7358 adj->ref_nr - ref_nr); 7359 adj->ref_nr -= ref_nr; 7360 return; 7361 } 7362 7363 if (adj->master) 7364 sysfs_remove_link(&(dev->dev.kobj), "master"); 7365 7366 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7367 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7368 7369 list_del_rcu(&adj->list); 7370 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", 7371 adj_dev->name, dev->name, adj_dev->name); 7372 dev_put(adj_dev); 7373 kfree_rcu(adj, rcu); 7374 } 7375 7376 static int __netdev_adjacent_dev_link_lists(struct net_device *dev, 7377 struct net_device *upper_dev, 7378 struct list_head *up_list, 7379 struct list_head *down_list, 7380 void *private, bool master) 7381 { 7382 int ret; 7383 7384 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, 7385 private, master); 7386 if (ret) 7387 return ret; 7388 7389 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, 7390 private, false); 7391 if (ret) { 7392 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); 7393 return ret; 7394 } 7395 7396 return 0; 7397 } 7398 7399 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, 7400 struct net_device *upper_dev, 7401 u16 ref_nr, 7402 struct list_head *up_list, 7403 struct list_head *down_list) 7404 { 7405 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); 7406 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); 7407 } 7408 7409 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, 7410 struct net_device *upper_dev, 7411 void *private, bool master) 7412 { 7413 return __netdev_adjacent_dev_link_lists(dev, upper_dev, 7414 &dev->adj_list.upper, 7415 &upper_dev->adj_list.lower, 7416 private, master); 7417 } 7418 7419 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, 7420 struct net_device *upper_dev) 7421 { 7422 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, 7423 &dev->adj_list.upper, 7424 &upper_dev->adj_list.lower); 7425 } 7426 7427 static int __netdev_upper_dev_link(struct net_device *dev, 7428 struct net_device *upper_dev, bool master, 7429 void *upper_priv, void *upper_info, 7430 struct netdev_nested_priv *priv, 7431 struct netlink_ext_ack *extack) 7432 { 7433 struct netdev_notifier_changeupper_info changeupper_info = { 7434 .info = { 7435 .dev = dev, 7436 .extack = extack, 7437 }, 7438 .upper_dev = upper_dev, 7439 .master = master, 7440 .linking = true, 7441 .upper_info = upper_info, 7442 }; 7443 struct net_device *master_dev; 7444 int ret = 0; 7445 7446 ASSERT_RTNL(); 7447 7448 if (dev == upper_dev) 7449 return -EBUSY; 7450 7451 /* To prevent loops, check if dev is not upper device to upper_dev. */ 7452 if (__netdev_has_upper_dev(upper_dev, dev)) 7453 return -EBUSY; 7454 7455 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV) 7456 return -EMLINK; 7457 7458 if (!master) { 7459 if (__netdev_has_upper_dev(dev, upper_dev)) 7460 return -EEXIST; 7461 } else { 7462 master_dev = __netdev_master_upper_dev_get(dev); 7463 if (master_dev) 7464 return master_dev == upper_dev ? -EEXIST : -EBUSY; 7465 } 7466 7467 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7468 &changeupper_info.info); 7469 ret = notifier_to_errno(ret); 7470 if (ret) 7471 return ret; 7472 7473 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, 7474 master); 7475 if (ret) 7476 return ret; 7477 7478 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 7479 &changeupper_info.info); 7480 ret = notifier_to_errno(ret); 7481 if (ret) 7482 goto rollback; 7483 7484 __netdev_update_upper_level(dev, NULL); 7485 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 7486 7487 __netdev_update_lower_level(upper_dev, priv); 7488 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 7489 priv); 7490 7491 return 0; 7492 7493 rollback: 7494 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 7495 7496 return ret; 7497 } 7498 7499 /** 7500 * netdev_upper_dev_link - Add a link to the upper device 7501 * @dev: device 7502 * @upper_dev: new upper device 7503 * @extack: netlink extended ack 7504 * 7505 * Adds a link to device which is upper to this one. The caller must hold 7506 * the RTNL lock. On a failure a negative errno code is returned. 7507 * On success the reference counts are adjusted and the function 7508 * returns zero. 7509 */ 7510 int netdev_upper_dev_link(struct net_device *dev, 7511 struct net_device *upper_dev, 7512 struct netlink_ext_ack *extack) 7513 { 7514 struct netdev_nested_priv priv = { 7515 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7516 .data = NULL, 7517 }; 7518 7519 return __netdev_upper_dev_link(dev, upper_dev, false, 7520 NULL, NULL, &priv, extack); 7521 } 7522 EXPORT_SYMBOL(netdev_upper_dev_link); 7523 7524 /** 7525 * netdev_master_upper_dev_link - Add a master link to the upper device 7526 * @dev: device 7527 * @upper_dev: new upper device 7528 * @upper_priv: upper device private 7529 * @upper_info: upper info to be passed down via notifier 7530 * @extack: netlink extended ack 7531 * 7532 * Adds a link to device which is upper to this one. In this case, only 7533 * one master upper device can be linked, although other non-master devices 7534 * might be linked as well. The caller must hold the RTNL lock. 7535 * On a failure a negative errno code is returned. On success the reference 7536 * counts are adjusted and the function returns zero. 7537 */ 7538 int netdev_master_upper_dev_link(struct net_device *dev, 7539 struct net_device *upper_dev, 7540 void *upper_priv, void *upper_info, 7541 struct netlink_ext_ack *extack) 7542 { 7543 struct netdev_nested_priv priv = { 7544 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7545 .data = NULL, 7546 }; 7547 7548 return __netdev_upper_dev_link(dev, upper_dev, true, 7549 upper_priv, upper_info, &priv, extack); 7550 } 7551 EXPORT_SYMBOL(netdev_master_upper_dev_link); 7552 7553 static void __netdev_upper_dev_unlink(struct net_device *dev, 7554 struct net_device *upper_dev, 7555 struct netdev_nested_priv *priv) 7556 { 7557 struct netdev_notifier_changeupper_info changeupper_info = { 7558 .info = { 7559 .dev = dev, 7560 }, 7561 .upper_dev = upper_dev, 7562 .linking = false, 7563 }; 7564 7565 ASSERT_RTNL(); 7566 7567 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; 7568 7569 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7570 &changeupper_info.info); 7571 7572 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 7573 7574 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 7575 &changeupper_info.info); 7576 7577 __netdev_update_upper_level(dev, NULL); 7578 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 7579 7580 __netdev_update_lower_level(upper_dev, priv); 7581 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 7582 priv); 7583 } 7584 7585 /** 7586 * netdev_upper_dev_unlink - Removes a link to upper device 7587 * @dev: device 7588 * @upper_dev: new upper device 7589 * 7590 * Removes a link to device which is upper to this one. The caller must hold 7591 * the RTNL lock. 7592 */ 7593 void netdev_upper_dev_unlink(struct net_device *dev, 7594 struct net_device *upper_dev) 7595 { 7596 struct netdev_nested_priv priv = { 7597 .flags = NESTED_SYNC_TODO, 7598 .data = NULL, 7599 }; 7600 7601 __netdev_upper_dev_unlink(dev, upper_dev, &priv); 7602 } 7603 EXPORT_SYMBOL(netdev_upper_dev_unlink); 7604 7605 static void __netdev_adjacent_dev_set(struct net_device *upper_dev, 7606 struct net_device *lower_dev, 7607 bool val) 7608 { 7609 struct netdev_adjacent *adj; 7610 7611 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower); 7612 if (adj) 7613 adj->ignore = val; 7614 7615 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper); 7616 if (adj) 7617 adj->ignore = val; 7618 } 7619 7620 static void netdev_adjacent_dev_disable(struct net_device *upper_dev, 7621 struct net_device *lower_dev) 7622 { 7623 __netdev_adjacent_dev_set(upper_dev, lower_dev, true); 7624 } 7625 7626 static void netdev_adjacent_dev_enable(struct net_device *upper_dev, 7627 struct net_device *lower_dev) 7628 { 7629 __netdev_adjacent_dev_set(upper_dev, lower_dev, false); 7630 } 7631 7632 int netdev_adjacent_change_prepare(struct net_device *old_dev, 7633 struct net_device *new_dev, 7634 struct net_device *dev, 7635 struct netlink_ext_ack *extack) 7636 { 7637 struct netdev_nested_priv priv = { 7638 .flags = 0, 7639 .data = NULL, 7640 }; 7641 int err; 7642 7643 if (!new_dev) 7644 return 0; 7645 7646 if (old_dev && new_dev != old_dev) 7647 netdev_adjacent_dev_disable(dev, old_dev); 7648 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv, 7649 extack); 7650 if (err) { 7651 if (old_dev && new_dev != old_dev) 7652 netdev_adjacent_dev_enable(dev, old_dev); 7653 return err; 7654 } 7655 7656 return 0; 7657 } 7658 EXPORT_SYMBOL(netdev_adjacent_change_prepare); 7659 7660 void netdev_adjacent_change_commit(struct net_device *old_dev, 7661 struct net_device *new_dev, 7662 struct net_device *dev) 7663 { 7664 struct netdev_nested_priv priv = { 7665 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7666 .data = NULL, 7667 }; 7668 7669 if (!new_dev || !old_dev) 7670 return; 7671 7672 if (new_dev == old_dev) 7673 return; 7674 7675 netdev_adjacent_dev_enable(dev, old_dev); 7676 __netdev_upper_dev_unlink(old_dev, dev, &priv); 7677 } 7678 EXPORT_SYMBOL(netdev_adjacent_change_commit); 7679 7680 void netdev_adjacent_change_abort(struct net_device *old_dev, 7681 struct net_device *new_dev, 7682 struct net_device *dev) 7683 { 7684 struct netdev_nested_priv priv = { 7685 .flags = 0, 7686 .data = NULL, 7687 }; 7688 7689 if (!new_dev) 7690 return; 7691 7692 if (old_dev && new_dev != old_dev) 7693 netdev_adjacent_dev_enable(dev, old_dev); 7694 7695 __netdev_upper_dev_unlink(new_dev, dev, &priv); 7696 } 7697 EXPORT_SYMBOL(netdev_adjacent_change_abort); 7698 7699 /** 7700 * netdev_bonding_info_change - Dispatch event about slave change 7701 * @dev: device 7702 * @bonding_info: info to dispatch 7703 * 7704 * Send NETDEV_BONDING_INFO to netdev notifiers with info. 7705 * The caller must hold the RTNL lock. 7706 */ 7707 void netdev_bonding_info_change(struct net_device *dev, 7708 struct netdev_bonding_info *bonding_info) 7709 { 7710 struct netdev_notifier_bonding_info info = { 7711 .info.dev = dev, 7712 }; 7713 7714 memcpy(&info.bonding_info, bonding_info, 7715 sizeof(struct netdev_bonding_info)); 7716 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, 7717 &info.info); 7718 } 7719 EXPORT_SYMBOL(netdev_bonding_info_change); 7720 7721 /** 7722 * netdev_get_xmit_slave - Get the xmit slave of master device 7723 * @dev: device 7724 * @skb: The packet 7725 * @all_slaves: assume all the slaves are active 7726 * 7727 * The reference counters are not incremented so the caller must be 7728 * careful with locks. The caller must hold RCU lock. 7729 * %NULL is returned if no slave is found. 7730 */ 7731 7732 struct net_device *netdev_get_xmit_slave(struct net_device *dev, 7733 struct sk_buff *skb, 7734 bool all_slaves) 7735 { 7736 const struct net_device_ops *ops = dev->netdev_ops; 7737 7738 if (!ops->ndo_get_xmit_slave) 7739 return NULL; 7740 return ops->ndo_get_xmit_slave(dev, skb, all_slaves); 7741 } 7742 EXPORT_SYMBOL(netdev_get_xmit_slave); 7743 7744 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev, 7745 struct sock *sk) 7746 { 7747 const struct net_device_ops *ops = dev->netdev_ops; 7748 7749 if (!ops->ndo_sk_get_lower_dev) 7750 return NULL; 7751 return ops->ndo_sk_get_lower_dev(dev, sk); 7752 } 7753 7754 /** 7755 * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket 7756 * @dev: device 7757 * @sk: the socket 7758 * 7759 * %NULL is returned if no lower device is found. 7760 */ 7761 7762 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev, 7763 struct sock *sk) 7764 { 7765 struct net_device *lower; 7766 7767 lower = netdev_sk_get_lower_dev(dev, sk); 7768 while (lower) { 7769 dev = lower; 7770 lower = netdev_sk_get_lower_dev(dev, sk); 7771 } 7772 7773 return dev; 7774 } 7775 EXPORT_SYMBOL(netdev_sk_get_lowest_dev); 7776 7777 static void netdev_adjacent_add_links(struct net_device *dev) 7778 { 7779 struct netdev_adjacent *iter; 7780 7781 struct net *net = dev_net(dev); 7782 7783 list_for_each_entry(iter, &dev->adj_list.upper, list) { 7784 if (!net_eq(net, dev_net(iter->dev))) 7785 continue; 7786 netdev_adjacent_sysfs_add(iter->dev, dev, 7787 &iter->dev->adj_list.lower); 7788 netdev_adjacent_sysfs_add(dev, iter->dev, 7789 &dev->adj_list.upper); 7790 } 7791 7792 list_for_each_entry(iter, &dev->adj_list.lower, list) { 7793 if (!net_eq(net, dev_net(iter->dev))) 7794 continue; 7795 netdev_adjacent_sysfs_add(iter->dev, dev, 7796 &iter->dev->adj_list.upper); 7797 netdev_adjacent_sysfs_add(dev, iter->dev, 7798 &dev->adj_list.lower); 7799 } 7800 } 7801 7802 static void netdev_adjacent_del_links(struct net_device *dev) 7803 { 7804 struct netdev_adjacent *iter; 7805 7806 struct net *net = dev_net(dev); 7807 7808 list_for_each_entry(iter, &dev->adj_list.upper, list) { 7809 if (!net_eq(net, dev_net(iter->dev))) 7810 continue; 7811 netdev_adjacent_sysfs_del(iter->dev, dev->name, 7812 &iter->dev->adj_list.lower); 7813 netdev_adjacent_sysfs_del(dev, iter->dev->name, 7814 &dev->adj_list.upper); 7815 } 7816 7817 list_for_each_entry(iter, &dev->adj_list.lower, list) { 7818 if (!net_eq(net, dev_net(iter->dev))) 7819 continue; 7820 netdev_adjacent_sysfs_del(iter->dev, dev->name, 7821 &iter->dev->adj_list.upper); 7822 netdev_adjacent_sysfs_del(dev, iter->dev->name, 7823 &dev->adj_list.lower); 7824 } 7825 } 7826 7827 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) 7828 { 7829 struct netdev_adjacent *iter; 7830 7831 struct net *net = dev_net(dev); 7832 7833 list_for_each_entry(iter, &dev->adj_list.upper, list) { 7834 if (!net_eq(net, dev_net(iter->dev))) 7835 continue; 7836 netdev_adjacent_sysfs_del(iter->dev, oldname, 7837 &iter->dev->adj_list.lower); 7838 netdev_adjacent_sysfs_add(iter->dev, dev, 7839 &iter->dev->adj_list.lower); 7840 } 7841 7842 list_for_each_entry(iter, &dev->adj_list.lower, list) { 7843 if (!net_eq(net, dev_net(iter->dev))) 7844 continue; 7845 netdev_adjacent_sysfs_del(iter->dev, oldname, 7846 &iter->dev->adj_list.upper); 7847 netdev_adjacent_sysfs_add(iter->dev, dev, 7848 &iter->dev->adj_list.upper); 7849 } 7850 } 7851 7852 void *netdev_lower_dev_get_private(struct net_device *dev, 7853 struct net_device *lower_dev) 7854 { 7855 struct netdev_adjacent *lower; 7856 7857 if (!lower_dev) 7858 return NULL; 7859 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); 7860 if (!lower) 7861 return NULL; 7862 7863 return lower->private; 7864 } 7865 EXPORT_SYMBOL(netdev_lower_dev_get_private); 7866 7867 7868 /** 7869 * netdev_lower_state_changed - Dispatch event about lower device state change 7870 * @lower_dev: device 7871 * @lower_state_info: state to dispatch 7872 * 7873 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. 7874 * The caller must hold the RTNL lock. 7875 */ 7876 void netdev_lower_state_changed(struct net_device *lower_dev, 7877 void *lower_state_info) 7878 { 7879 struct netdev_notifier_changelowerstate_info changelowerstate_info = { 7880 .info.dev = lower_dev, 7881 }; 7882 7883 ASSERT_RTNL(); 7884 changelowerstate_info.lower_state_info = lower_state_info; 7885 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, 7886 &changelowerstate_info.info); 7887 } 7888 EXPORT_SYMBOL(netdev_lower_state_changed); 7889 7890 static void dev_change_rx_flags(struct net_device *dev, int flags) 7891 { 7892 const struct net_device_ops *ops = dev->netdev_ops; 7893 7894 if (ops->ndo_change_rx_flags) 7895 ops->ndo_change_rx_flags(dev, flags); 7896 } 7897 7898 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) 7899 { 7900 unsigned int old_flags = dev->flags; 7901 kuid_t uid; 7902 kgid_t gid; 7903 7904 ASSERT_RTNL(); 7905 7906 dev->flags |= IFF_PROMISC; 7907 dev->promiscuity += inc; 7908 if (dev->promiscuity == 0) { 7909 /* 7910 * Avoid overflow. 7911 * If inc causes overflow, untouch promisc and return error. 7912 */ 7913 if (inc < 0) 7914 dev->flags &= ~IFF_PROMISC; 7915 else { 7916 dev->promiscuity -= inc; 7917 netdev_warn(dev, "promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n"); 7918 return -EOVERFLOW; 7919 } 7920 } 7921 if (dev->flags != old_flags) { 7922 pr_info("device %s %s promiscuous mode\n", 7923 dev->name, 7924 dev->flags & IFF_PROMISC ? "entered" : "left"); 7925 if (audit_enabled) { 7926 current_uid_gid(&uid, &gid); 7927 audit_log(audit_context(), GFP_ATOMIC, 7928 AUDIT_ANOM_PROMISCUOUS, 7929 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", 7930 dev->name, (dev->flags & IFF_PROMISC), 7931 (old_flags & IFF_PROMISC), 7932 from_kuid(&init_user_ns, audit_get_loginuid(current)), 7933 from_kuid(&init_user_ns, uid), 7934 from_kgid(&init_user_ns, gid), 7935 audit_get_sessionid(current)); 7936 } 7937 7938 dev_change_rx_flags(dev, IFF_PROMISC); 7939 } 7940 if (notify) 7941 __dev_notify_flags(dev, old_flags, IFF_PROMISC); 7942 return 0; 7943 } 7944 7945 /** 7946 * dev_set_promiscuity - update promiscuity count on a device 7947 * @dev: device 7948 * @inc: modifier 7949 * 7950 * Add or remove promiscuity from a device. While the count in the device 7951 * remains above zero the interface remains promiscuous. Once it hits zero 7952 * the device reverts back to normal filtering operation. A negative inc 7953 * value is used to drop promiscuity on the device. 7954 * Return 0 if successful or a negative errno code on error. 7955 */ 7956 int dev_set_promiscuity(struct net_device *dev, int inc) 7957 { 7958 unsigned int old_flags = dev->flags; 7959 int err; 7960 7961 err = __dev_set_promiscuity(dev, inc, true); 7962 if (err < 0) 7963 return err; 7964 if (dev->flags != old_flags) 7965 dev_set_rx_mode(dev); 7966 return err; 7967 } 7968 EXPORT_SYMBOL(dev_set_promiscuity); 7969 7970 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) 7971 { 7972 unsigned int old_flags = dev->flags, old_gflags = dev->gflags; 7973 7974 ASSERT_RTNL(); 7975 7976 dev->flags |= IFF_ALLMULTI; 7977 dev->allmulti += inc; 7978 if (dev->allmulti == 0) { 7979 /* 7980 * Avoid overflow. 7981 * If inc causes overflow, untouch allmulti and return error. 7982 */ 7983 if (inc < 0) 7984 dev->flags &= ~IFF_ALLMULTI; 7985 else { 7986 dev->allmulti -= inc; 7987 netdev_warn(dev, "allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n"); 7988 return -EOVERFLOW; 7989 } 7990 } 7991 if (dev->flags ^ old_flags) { 7992 dev_change_rx_flags(dev, IFF_ALLMULTI); 7993 dev_set_rx_mode(dev); 7994 if (notify) 7995 __dev_notify_flags(dev, old_flags, 7996 dev->gflags ^ old_gflags); 7997 } 7998 return 0; 7999 } 8000 8001 /** 8002 * dev_set_allmulti - update allmulti count on a device 8003 * @dev: device 8004 * @inc: modifier 8005 * 8006 * Add or remove reception of all multicast frames to a device. While the 8007 * count in the device remains above zero the interface remains listening 8008 * to all interfaces. Once it hits zero the device reverts back to normal 8009 * filtering operation. A negative @inc value is used to drop the counter 8010 * when releasing a resource needing all multicasts. 8011 * Return 0 if successful or a negative errno code on error. 8012 */ 8013 8014 int dev_set_allmulti(struct net_device *dev, int inc) 8015 { 8016 return __dev_set_allmulti(dev, inc, true); 8017 } 8018 EXPORT_SYMBOL(dev_set_allmulti); 8019 8020 /* 8021 * Upload unicast and multicast address lists to device and 8022 * configure RX filtering. When the device doesn't support unicast 8023 * filtering it is put in promiscuous mode while unicast addresses 8024 * are present. 8025 */ 8026 void __dev_set_rx_mode(struct net_device *dev) 8027 { 8028 const struct net_device_ops *ops = dev->netdev_ops; 8029 8030 /* dev_open will call this function so the list will stay sane. */ 8031 if (!(dev->flags&IFF_UP)) 8032 return; 8033 8034 if (!netif_device_present(dev)) 8035 return; 8036 8037 if (!(dev->priv_flags & IFF_UNICAST_FLT)) { 8038 /* Unicast addresses changes may only happen under the rtnl, 8039 * therefore calling __dev_set_promiscuity here is safe. 8040 */ 8041 if (!netdev_uc_empty(dev) && !dev->uc_promisc) { 8042 __dev_set_promiscuity(dev, 1, false); 8043 dev->uc_promisc = true; 8044 } else if (netdev_uc_empty(dev) && dev->uc_promisc) { 8045 __dev_set_promiscuity(dev, -1, false); 8046 dev->uc_promisc = false; 8047 } 8048 } 8049 8050 if (ops->ndo_set_rx_mode) 8051 ops->ndo_set_rx_mode(dev); 8052 } 8053 8054 void dev_set_rx_mode(struct net_device *dev) 8055 { 8056 netif_addr_lock_bh(dev); 8057 __dev_set_rx_mode(dev); 8058 netif_addr_unlock_bh(dev); 8059 } 8060 8061 /** 8062 * dev_get_flags - get flags reported to userspace 8063 * @dev: device 8064 * 8065 * Get the combination of flag bits exported through APIs to userspace. 8066 */ 8067 unsigned int dev_get_flags(const struct net_device *dev) 8068 { 8069 unsigned int flags; 8070 8071 flags = (dev->flags & ~(IFF_PROMISC | 8072 IFF_ALLMULTI | 8073 IFF_RUNNING | 8074 IFF_LOWER_UP | 8075 IFF_DORMANT)) | 8076 (dev->gflags & (IFF_PROMISC | 8077 IFF_ALLMULTI)); 8078 8079 if (netif_running(dev)) { 8080 if (netif_oper_up(dev)) 8081 flags |= IFF_RUNNING; 8082 if (netif_carrier_ok(dev)) 8083 flags |= IFF_LOWER_UP; 8084 if (netif_dormant(dev)) 8085 flags |= IFF_DORMANT; 8086 } 8087 8088 return flags; 8089 } 8090 EXPORT_SYMBOL(dev_get_flags); 8091 8092 int __dev_change_flags(struct net_device *dev, unsigned int flags, 8093 struct netlink_ext_ack *extack) 8094 { 8095 unsigned int old_flags = dev->flags; 8096 int ret; 8097 8098 ASSERT_RTNL(); 8099 8100 /* 8101 * Set the flags on our device. 8102 */ 8103 8104 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | 8105 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | 8106 IFF_AUTOMEDIA)) | 8107 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | 8108 IFF_ALLMULTI)); 8109 8110 /* 8111 * Load in the correct multicast list now the flags have changed. 8112 */ 8113 8114 if ((old_flags ^ flags) & IFF_MULTICAST) 8115 dev_change_rx_flags(dev, IFF_MULTICAST); 8116 8117 dev_set_rx_mode(dev); 8118 8119 /* 8120 * Have we downed the interface. We handle IFF_UP ourselves 8121 * according to user attempts to set it, rather than blindly 8122 * setting it. 8123 */ 8124 8125 ret = 0; 8126 if ((old_flags ^ flags) & IFF_UP) { 8127 if (old_flags & IFF_UP) 8128 __dev_close(dev); 8129 else 8130 ret = __dev_open(dev, extack); 8131 } 8132 8133 if ((flags ^ dev->gflags) & IFF_PROMISC) { 8134 int inc = (flags & IFF_PROMISC) ? 1 : -1; 8135 unsigned int old_flags = dev->flags; 8136 8137 dev->gflags ^= IFF_PROMISC; 8138 8139 if (__dev_set_promiscuity(dev, inc, false) >= 0) 8140 if (dev->flags != old_flags) 8141 dev_set_rx_mode(dev); 8142 } 8143 8144 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI 8145 * is important. Some (broken) drivers set IFF_PROMISC, when 8146 * IFF_ALLMULTI is requested not asking us and not reporting. 8147 */ 8148 if ((flags ^ dev->gflags) & IFF_ALLMULTI) { 8149 int inc = (flags & IFF_ALLMULTI) ? 1 : -1; 8150 8151 dev->gflags ^= IFF_ALLMULTI; 8152 __dev_set_allmulti(dev, inc, false); 8153 } 8154 8155 return ret; 8156 } 8157 8158 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, 8159 unsigned int gchanges) 8160 { 8161 unsigned int changes = dev->flags ^ old_flags; 8162 8163 if (gchanges) 8164 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC); 8165 8166 if (changes & IFF_UP) { 8167 if (dev->flags & IFF_UP) 8168 call_netdevice_notifiers(NETDEV_UP, dev); 8169 else 8170 call_netdevice_notifiers(NETDEV_DOWN, dev); 8171 } 8172 8173 if (dev->flags & IFF_UP && 8174 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { 8175 struct netdev_notifier_change_info change_info = { 8176 .info = { 8177 .dev = dev, 8178 }, 8179 .flags_changed = changes, 8180 }; 8181 8182 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info); 8183 } 8184 } 8185 8186 /** 8187 * dev_change_flags - change device settings 8188 * @dev: device 8189 * @flags: device state flags 8190 * @extack: netlink extended ack 8191 * 8192 * Change settings on device based state flags. The flags are 8193 * in the userspace exported format. 8194 */ 8195 int dev_change_flags(struct net_device *dev, unsigned int flags, 8196 struct netlink_ext_ack *extack) 8197 { 8198 int ret; 8199 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; 8200 8201 ret = __dev_change_flags(dev, flags, extack); 8202 if (ret < 0) 8203 return ret; 8204 8205 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); 8206 __dev_notify_flags(dev, old_flags, changes); 8207 return ret; 8208 } 8209 EXPORT_SYMBOL(dev_change_flags); 8210 8211 int __dev_set_mtu(struct net_device *dev, int new_mtu) 8212 { 8213 const struct net_device_ops *ops = dev->netdev_ops; 8214 8215 if (ops->ndo_change_mtu) 8216 return ops->ndo_change_mtu(dev, new_mtu); 8217 8218 /* Pairs with all the lockless reads of dev->mtu in the stack */ 8219 WRITE_ONCE(dev->mtu, new_mtu); 8220 return 0; 8221 } 8222 EXPORT_SYMBOL(__dev_set_mtu); 8223 8224 int dev_validate_mtu(struct net_device *dev, int new_mtu, 8225 struct netlink_ext_ack *extack) 8226 { 8227 /* MTU must be positive, and in range */ 8228 if (new_mtu < 0 || new_mtu < dev->min_mtu) { 8229 NL_SET_ERR_MSG(extack, "mtu less than device minimum"); 8230 return -EINVAL; 8231 } 8232 8233 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { 8234 NL_SET_ERR_MSG(extack, "mtu greater than device maximum"); 8235 return -EINVAL; 8236 } 8237 return 0; 8238 } 8239 8240 /** 8241 * dev_set_mtu_ext - Change maximum transfer unit 8242 * @dev: device 8243 * @new_mtu: new transfer unit 8244 * @extack: netlink extended ack 8245 * 8246 * Change the maximum transfer size of the network device. 8247 */ 8248 int dev_set_mtu_ext(struct net_device *dev, int new_mtu, 8249 struct netlink_ext_ack *extack) 8250 { 8251 int err, orig_mtu; 8252 8253 if (new_mtu == dev->mtu) 8254 return 0; 8255 8256 err = dev_validate_mtu(dev, new_mtu, extack); 8257 if (err) 8258 return err; 8259 8260 if (!netif_device_present(dev)) 8261 return -ENODEV; 8262 8263 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); 8264 err = notifier_to_errno(err); 8265 if (err) 8266 return err; 8267 8268 orig_mtu = dev->mtu; 8269 err = __dev_set_mtu(dev, new_mtu); 8270 8271 if (!err) { 8272 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8273 orig_mtu); 8274 err = notifier_to_errno(err); 8275 if (err) { 8276 /* setting mtu back and notifying everyone again, 8277 * so that they have a chance to revert changes. 8278 */ 8279 __dev_set_mtu(dev, orig_mtu); 8280 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8281 new_mtu); 8282 } 8283 } 8284 return err; 8285 } 8286 8287 int dev_set_mtu(struct net_device *dev, int new_mtu) 8288 { 8289 struct netlink_ext_ack extack; 8290 int err; 8291 8292 memset(&extack, 0, sizeof(extack)); 8293 err = dev_set_mtu_ext(dev, new_mtu, &extack); 8294 if (err && extack._msg) 8295 net_err_ratelimited("%s: %s\n", dev->name, extack._msg); 8296 return err; 8297 } 8298 EXPORT_SYMBOL(dev_set_mtu); 8299 8300 /** 8301 * dev_change_tx_queue_len - Change TX queue length of a netdevice 8302 * @dev: device 8303 * @new_len: new tx queue length 8304 */ 8305 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len) 8306 { 8307 unsigned int orig_len = dev->tx_queue_len; 8308 int res; 8309 8310 if (new_len != (unsigned int)new_len) 8311 return -ERANGE; 8312 8313 if (new_len != orig_len) { 8314 dev->tx_queue_len = new_len; 8315 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev); 8316 res = notifier_to_errno(res); 8317 if (res) 8318 goto err_rollback; 8319 res = dev_qdisc_change_tx_queue_len(dev); 8320 if (res) 8321 goto err_rollback; 8322 } 8323 8324 return 0; 8325 8326 err_rollback: 8327 netdev_err(dev, "refused to change device tx_queue_len\n"); 8328 dev->tx_queue_len = orig_len; 8329 return res; 8330 } 8331 8332 /** 8333 * dev_set_group - Change group this device belongs to 8334 * @dev: device 8335 * @new_group: group this device should belong to 8336 */ 8337 void dev_set_group(struct net_device *dev, int new_group) 8338 { 8339 dev->group = new_group; 8340 } 8341 EXPORT_SYMBOL(dev_set_group); 8342 8343 /** 8344 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR. 8345 * @dev: device 8346 * @addr: new address 8347 * @extack: netlink extended ack 8348 */ 8349 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr, 8350 struct netlink_ext_ack *extack) 8351 { 8352 struct netdev_notifier_pre_changeaddr_info info = { 8353 .info.dev = dev, 8354 .info.extack = extack, 8355 .dev_addr = addr, 8356 }; 8357 int rc; 8358 8359 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info); 8360 return notifier_to_errno(rc); 8361 } 8362 EXPORT_SYMBOL(dev_pre_changeaddr_notify); 8363 8364 /** 8365 * dev_set_mac_address - Change Media Access Control Address 8366 * @dev: device 8367 * @sa: new address 8368 * @extack: netlink extended ack 8369 * 8370 * Change the hardware (MAC) address of the device 8371 */ 8372 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa, 8373 struct netlink_ext_ack *extack) 8374 { 8375 const struct net_device_ops *ops = dev->netdev_ops; 8376 int err; 8377 8378 if (!ops->ndo_set_mac_address) 8379 return -EOPNOTSUPP; 8380 if (sa->sa_family != dev->type) 8381 return -EINVAL; 8382 if (!netif_device_present(dev)) 8383 return -ENODEV; 8384 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack); 8385 if (err) 8386 return err; 8387 err = ops->ndo_set_mac_address(dev, sa); 8388 if (err) 8389 return err; 8390 dev->addr_assign_type = NET_ADDR_SET; 8391 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); 8392 add_device_randomness(dev->dev_addr, dev->addr_len); 8393 return 0; 8394 } 8395 EXPORT_SYMBOL(dev_set_mac_address); 8396 8397 static DECLARE_RWSEM(dev_addr_sem); 8398 8399 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa, 8400 struct netlink_ext_ack *extack) 8401 { 8402 int ret; 8403 8404 down_write(&dev_addr_sem); 8405 ret = dev_set_mac_address(dev, sa, extack); 8406 up_write(&dev_addr_sem); 8407 return ret; 8408 } 8409 EXPORT_SYMBOL(dev_set_mac_address_user); 8410 8411 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name) 8412 { 8413 size_t size = sizeof(sa->sa_data); 8414 struct net_device *dev; 8415 int ret = 0; 8416 8417 down_read(&dev_addr_sem); 8418 rcu_read_lock(); 8419 8420 dev = dev_get_by_name_rcu(net, dev_name); 8421 if (!dev) { 8422 ret = -ENODEV; 8423 goto unlock; 8424 } 8425 if (!dev->addr_len) 8426 memset(sa->sa_data, 0, size); 8427 else 8428 memcpy(sa->sa_data, dev->dev_addr, 8429 min_t(size_t, size, dev->addr_len)); 8430 sa->sa_family = dev->type; 8431 8432 unlock: 8433 rcu_read_unlock(); 8434 up_read(&dev_addr_sem); 8435 return ret; 8436 } 8437 EXPORT_SYMBOL(dev_get_mac_address); 8438 8439 /** 8440 * dev_change_carrier - Change device carrier 8441 * @dev: device 8442 * @new_carrier: new value 8443 * 8444 * Change device carrier 8445 */ 8446 int dev_change_carrier(struct net_device *dev, bool new_carrier) 8447 { 8448 const struct net_device_ops *ops = dev->netdev_ops; 8449 8450 if (!ops->ndo_change_carrier) 8451 return -EOPNOTSUPP; 8452 if (!netif_device_present(dev)) 8453 return -ENODEV; 8454 return ops->ndo_change_carrier(dev, new_carrier); 8455 } 8456 EXPORT_SYMBOL(dev_change_carrier); 8457 8458 /** 8459 * dev_get_phys_port_id - Get device physical port ID 8460 * @dev: device 8461 * @ppid: port ID 8462 * 8463 * Get device physical port ID 8464 */ 8465 int dev_get_phys_port_id(struct net_device *dev, 8466 struct netdev_phys_item_id *ppid) 8467 { 8468 const struct net_device_ops *ops = dev->netdev_ops; 8469 8470 if (!ops->ndo_get_phys_port_id) 8471 return -EOPNOTSUPP; 8472 return ops->ndo_get_phys_port_id(dev, ppid); 8473 } 8474 EXPORT_SYMBOL(dev_get_phys_port_id); 8475 8476 /** 8477 * dev_get_phys_port_name - Get device physical port name 8478 * @dev: device 8479 * @name: port name 8480 * @len: limit of bytes to copy to name 8481 * 8482 * Get device physical port name 8483 */ 8484 int dev_get_phys_port_name(struct net_device *dev, 8485 char *name, size_t len) 8486 { 8487 const struct net_device_ops *ops = dev->netdev_ops; 8488 int err; 8489 8490 if (ops->ndo_get_phys_port_name) { 8491 err = ops->ndo_get_phys_port_name(dev, name, len); 8492 if (err != -EOPNOTSUPP) 8493 return err; 8494 } 8495 return devlink_compat_phys_port_name_get(dev, name, len); 8496 } 8497 EXPORT_SYMBOL(dev_get_phys_port_name); 8498 8499 /** 8500 * dev_get_port_parent_id - Get the device's port parent identifier 8501 * @dev: network device 8502 * @ppid: pointer to a storage for the port's parent identifier 8503 * @recurse: allow/disallow recursion to lower devices 8504 * 8505 * Get the devices's port parent identifier 8506 */ 8507 int dev_get_port_parent_id(struct net_device *dev, 8508 struct netdev_phys_item_id *ppid, 8509 bool recurse) 8510 { 8511 const struct net_device_ops *ops = dev->netdev_ops; 8512 struct netdev_phys_item_id first = { }; 8513 struct net_device *lower_dev; 8514 struct list_head *iter; 8515 int err; 8516 8517 if (ops->ndo_get_port_parent_id) { 8518 err = ops->ndo_get_port_parent_id(dev, ppid); 8519 if (err != -EOPNOTSUPP) 8520 return err; 8521 } 8522 8523 err = devlink_compat_switch_id_get(dev, ppid); 8524 if (!recurse || err != -EOPNOTSUPP) 8525 return err; 8526 8527 netdev_for_each_lower_dev(dev, lower_dev, iter) { 8528 err = dev_get_port_parent_id(lower_dev, ppid, true); 8529 if (err) 8530 break; 8531 if (!first.id_len) 8532 first = *ppid; 8533 else if (memcmp(&first, ppid, sizeof(*ppid))) 8534 return -EOPNOTSUPP; 8535 } 8536 8537 return err; 8538 } 8539 EXPORT_SYMBOL(dev_get_port_parent_id); 8540 8541 /** 8542 * netdev_port_same_parent_id - Indicate if two network devices have 8543 * the same port parent identifier 8544 * @a: first network device 8545 * @b: second network device 8546 */ 8547 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b) 8548 { 8549 struct netdev_phys_item_id a_id = { }; 8550 struct netdev_phys_item_id b_id = { }; 8551 8552 if (dev_get_port_parent_id(a, &a_id, true) || 8553 dev_get_port_parent_id(b, &b_id, true)) 8554 return false; 8555 8556 return netdev_phys_item_id_same(&a_id, &b_id); 8557 } 8558 EXPORT_SYMBOL(netdev_port_same_parent_id); 8559 8560 /** 8561 * dev_change_proto_down - set carrier according to proto_down. 8562 * 8563 * @dev: device 8564 * @proto_down: new value 8565 */ 8566 int dev_change_proto_down(struct net_device *dev, bool proto_down) 8567 { 8568 if (!(dev->priv_flags & IFF_CHANGE_PROTO_DOWN)) 8569 return -EOPNOTSUPP; 8570 if (!netif_device_present(dev)) 8571 return -ENODEV; 8572 if (proto_down) 8573 netif_carrier_off(dev); 8574 else 8575 netif_carrier_on(dev); 8576 dev->proto_down = proto_down; 8577 return 0; 8578 } 8579 EXPORT_SYMBOL(dev_change_proto_down); 8580 8581 /** 8582 * dev_change_proto_down_reason - proto down reason 8583 * 8584 * @dev: device 8585 * @mask: proto down mask 8586 * @value: proto down value 8587 */ 8588 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask, 8589 u32 value) 8590 { 8591 int b; 8592 8593 if (!mask) { 8594 dev->proto_down_reason = value; 8595 } else { 8596 for_each_set_bit(b, &mask, 32) { 8597 if (value & (1 << b)) 8598 dev->proto_down_reason |= BIT(b); 8599 else 8600 dev->proto_down_reason &= ~BIT(b); 8601 } 8602 } 8603 } 8604 EXPORT_SYMBOL(dev_change_proto_down_reason); 8605 8606 struct bpf_xdp_link { 8607 struct bpf_link link; 8608 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */ 8609 int flags; 8610 }; 8611 8612 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags) 8613 { 8614 if (flags & XDP_FLAGS_HW_MODE) 8615 return XDP_MODE_HW; 8616 if (flags & XDP_FLAGS_DRV_MODE) 8617 return XDP_MODE_DRV; 8618 if (flags & XDP_FLAGS_SKB_MODE) 8619 return XDP_MODE_SKB; 8620 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB; 8621 } 8622 8623 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode) 8624 { 8625 switch (mode) { 8626 case XDP_MODE_SKB: 8627 return generic_xdp_install; 8628 case XDP_MODE_DRV: 8629 case XDP_MODE_HW: 8630 return dev->netdev_ops->ndo_bpf; 8631 default: 8632 return NULL; 8633 } 8634 } 8635 8636 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev, 8637 enum bpf_xdp_mode mode) 8638 { 8639 return dev->xdp_state[mode].link; 8640 } 8641 8642 static struct bpf_prog *dev_xdp_prog(struct net_device *dev, 8643 enum bpf_xdp_mode mode) 8644 { 8645 struct bpf_xdp_link *link = dev_xdp_link(dev, mode); 8646 8647 if (link) 8648 return link->link.prog; 8649 return dev->xdp_state[mode].prog; 8650 } 8651 8652 u8 dev_xdp_prog_count(struct net_device *dev) 8653 { 8654 u8 count = 0; 8655 int i; 8656 8657 for (i = 0; i < __MAX_XDP_MODE; i++) 8658 if (dev->xdp_state[i].prog || dev->xdp_state[i].link) 8659 count++; 8660 return count; 8661 } 8662 EXPORT_SYMBOL_GPL(dev_xdp_prog_count); 8663 8664 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode) 8665 { 8666 struct bpf_prog *prog = dev_xdp_prog(dev, mode); 8667 8668 return prog ? prog->aux->id : 0; 8669 } 8670 8671 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode, 8672 struct bpf_xdp_link *link) 8673 { 8674 dev->xdp_state[mode].link = link; 8675 dev->xdp_state[mode].prog = NULL; 8676 } 8677 8678 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode, 8679 struct bpf_prog *prog) 8680 { 8681 dev->xdp_state[mode].link = NULL; 8682 dev->xdp_state[mode].prog = prog; 8683 } 8684 8685 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode, 8686 bpf_op_t bpf_op, struct netlink_ext_ack *extack, 8687 u32 flags, struct bpf_prog *prog) 8688 { 8689 struct netdev_bpf xdp; 8690 int err; 8691 8692 memset(&xdp, 0, sizeof(xdp)); 8693 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG; 8694 xdp.extack = extack; 8695 xdp.flags = flags; 8696 xdp.prog = prog; 8697 8698 /* Drivers assume refcnt is already incremented (i.e, prog pointer is 8699 * "moved" into driver), so they don't increment it on their own, but 8700 * they do decrement refcnt when program is detached or replaced. 8701 * Given net_device also owns link/prog, we need to bump refcnt here 8702 * to prevent drivers from underflowing it. 8703 */ 8704 if (prog) 8705 bpf_prog_inc(prog); 8706 err = bpf_op(dev, &xdp); 8707 if (err) { 8708 if (prog) 8709 bpf_prog_put(prog); 8710 return err; 8711 } 8712 8713 if (mode != XDP_MODE_HW) 8714 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog); 8715 8716 return 0; 8717 } 8718 8719 static void dev_xdp_uninstall(struct net_device *dev) 8720 { 8721 struct bpf_xdp_link *link; 8722 struct bpf_prog *prog; 8723 enum bpf_xdp_mode mode; 8724 bpf_op_t bpf_op; 8725 8726 ASSERT_RTNL(); 8727 8728 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) { 8729 prog = dev_xdp_prog(dev, mode); 8730 if (!prog) 8731 continue; 8732 8733 bpf_op = dev_xdp_bpf_op(dev, mode); 8734 if (!bpf_op) 8735 continue; 8736 8737 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 8738 8739 /* auto-detach link from net device */ 8740 link = dev_xdp_link(dev, mode); 8741 if (link) 8742 link->dev = NULL; 8743 else 8744 bpf_prog_put(prog); 8745 8746 dev_xdp_set_link(dev, mode, NULL); 8747 } 8748 } 8749 8750 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack, 8751 struct bpf_xdp_link *link, struct bpf_prog *new_prog, 8752 struct bpf_prog *old_prog, u32 flags) 8753 { 8754 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES); 8755 struct bpf_prog *cur_prog; 8756 struct net_device *upper; 8757 struct list_head *iter; 8758 enum bpf_xdp_mode mode; 8759 bpf_op_t bpf_op; 8760 int err; 8761 8762 ASSERT_RTNL(); 8763 8764 /* either link or prog attachment, never both */ 8765 if (link && (new_prog || old_prog)) 8766 return -EINVAL; 8767 /* link supports only XDP mode flags */ 8768 if (link && (flags & ~XDP_FLAGS_MODES)) { 8769 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment"); 8770 return -EINVAL; 8771 } 8772 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */ 8773 if (num_modes > 1) { 8774 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set"); 8775 return -EINVAL; 8776 } 8777 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */ 8778 if (!num_modes && dev_xdp_prog_count(dev) > 1) { 8779 NL_SET_ERR_MSG(extack, 8780 "More than one program loaded, unset mode is ambiguous"); 8781 return -EINVAL; 8782 } 8783 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */ 8784 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) { 8785 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified"); 8786 return -EINVAL; 8787 } 8788 8789 mode = dev_xdp_mode(dev, flags); 8790 /* can't replace attached link */ 8791 if (dev_xdp_link(dev, mode)) { 8792 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link"); 8793 return -EBUSY; 8794 } 8795 8796 /* don't allow if an upper device already has a program */ 8797 netdev_for_each_upper_dev_rcu(dev, upper, iter) { 8798 if (dev_xdp_prog_count(upper) > 0) { 8799 NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program"); 8800 return -EEXIST; 8801 } 8802 } 8803 8804 cur_prog = dev_xdp_prog(dev, mode); 8805 /* can't replace attached prog with link */ 8806 if (link && cur_prog) { 8807 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link"); 8808 return -EBUSY; 8809 } 8810 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) { 8811 NL_SET_ERR_MSG(extack, "Active program does not match expected"); 8812 return -EEXIST; 8813 } 8814 8815 /* put effective new program into new_prog */ 8816 if (link) 8817 new_prog = link->link.prog; 8818 8819 if (new_prog) { 8820 bool offload = mode == XDP_MODE_HW; 8821 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB 8822 ? XDP_MODE_DRV : XDP_MODE_SKB; 8823 8824 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) { 8825 NL_SET_ERR_MSG(extack, "XDP program already attached"); 8826 return -EBUSY; 8827 } 8828 if (!offload && dev_xdp_prog(dev, other_mode)) { 8829 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time"); 8830 return -EEXIST; 8831 } 8832 if (!offload && bpf_prog_is_dev_bound(new_prog->aux)) { 8833 NL_SET_ERR_MSG(extack, "Using device-bound program without HW_MODE flag is not supported"); 8834 return -EINVAL; 8835 } 8836 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) { 8837 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device"); 8838 return -EINVAL; 8839 } 8840 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) { 8841 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device"); 8842 return -EINVAL; 8843 } 8844 } 8845 8846 /* don't call drivers if the effective program didn't change */ 8847 if (new_prog != cur_prog) { 8848 bpf_op = dev_xdp_bpf_op(dev, mode); 8849 if (!bpf_op) { 8850 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode"); 8851 return -EOPNOTSUPP; 8852 } 8853 8854 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog); 8855 if (err) 8856 return err; 8857 } 8858 8859 if (link) 8860 dev_xdp_set_link(dev, mode, link); 8861 else 8862 dev_xdp_set_prog(dev, mode, new_prog); 8863 if (cur_prog) 8864 bpf_prog_put(cur_prog); 8865 8866 return 0; 8867 } 8868 8869 static int dev_xdp_attach_link(struct net_device *dev, 8870 struct netlink_ext_ack *extack, 8871 struct bpf_xdp_link *link) 8872 { 8873 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags); 8874 } 8875 8876 static int dev_xdp_detach_link(struct net_device *dev, 8877 struct netlink_ext_ack *extack, 8878 struct bpf_xdp_link *link) 8879 { 8880 enum bpf_xdp_mode mode; 8881 bpf_op_t bpf_op; 8882 8883 ASSERT_RTNL(); 8884 8885 mode = dev_xdp_mode(dev, link->flags); 8886 if (dev_xdp_link(dev, mode) != link) 8887 return -EINVAL; 8888 8889 bpf_op = dev_xdp_bpf_op(dev, mode); 8890 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 8891 dev_xdp_set_link(dev, mode, NULL); 8892 return 0; 8893 } 8894 8895 static void bpf_xdp_link_release(struct bpf_link *link) 8896 { 8897 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 8898 8899 rtnl_lock(); 8900 8901 /* if racing with net_device's tear down, xdp_link->dev might be 8902 * already NULL, in which case link was already auto-detached 8903 */ 8904 if (xdp_link->dev) { 8905 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link)); 8906 xdp_link->dev = NULL; 8907 } 8908 8909 rtnl_unlock(); 8910 } 8911 8912 static int bpf_xdp_link_detach(struct bpf_link *link) 8913 { 8914 bpf_xdp_link_release(link); 8915 return 0; 8916 } 8917 8918 static void bpf_xdp_link_dealloc(struct bpf_link *link) 8919 { 8920 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 8921 8922 kfree(xdp_link); 8923 } 8924 8925 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link, 8926 struct seq_file *seq) 8927 { 8928 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 8929 u32 ifindex = 0; 8930 8931 rtnl_lock(); 8932 if (xdp_link->dev) 8933 ifindex = xdp_link->dev->ifindex; 8934 rtnl_unlock(); 8935 8936 seq_printf(seq, "ifindex:\t%u\n", ifindex); 8937 } 8938 8939 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link, 8940 struct bpf_link_info *info) 8941 { 8942 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 8943 u32 ifindex = 0; 8944 8945 rtnl_lock(); 8946 if (xdp_link->dev) 8947 ifindex = xdp_link->dev->ifindex; 8948 rtnl_unlock(); 8949 8950 info->xdp.ifindex = ifindex; 8951 return 0; 8952 } 8953 8954 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog, 8955 struct bpf_prog *old_prog) 8956 { 8957 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 8958 enum bpf_xdp_mode mode; 8959 bpf_op_t bpf_op; 8960 int err = 0; 8961 8962 rtnl_lock(); 8963 8964 /* link might have been auto-released already, so fail */ 8965 if (!xdp_link->dev) { 8966 err = -ENOLINK; 8967 goto out_unlock; 8968 } 8969 8970 if (old_prog && link->prog != old_prog) { 8971 err = -EPERM; 8972 goto out_unlock; 8973 } 8974 old_prog = link->prog; 8975 if (old_prog == new_prog) { 8976 /* no-op, don't disturb drivers */ 8977 bpf_prog_put(new_prog); 8978 goto out_unlock; 8979 } 8980 8981 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags); 8982 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode); 8983 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL, 8984 xdp_link->flags, new_prog); 8985 if (err) 8986 goto out_unlock; 8987 8988 old_prog = xchg(&link->prog, new_prog); 8989 bpf_prog_put(old_prog); 8990 8991 out_unlock: 8992 rtnl_unlock(); 8993 return err; 8994 } 8995 8996 static const struct bpf_link_ops bpf_xdp_link_lops = { 8997 .release = bpf_xdp_link_release, 8998 .dealloc = bpf_xdp_link_dealloc, 8999 .detach = bpf_xdp_link_detach, 9000 .show_fdinfo = bpf_xdp_link_show_fdinfo, 9001 .fill_link_info = bpf_xdp_link_fill_link_info, 9002 .update_prog = bpf_xdp_link_update, 9003 }; 9004 9005 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog) 9006 { 9007 struct net *net = current->nsproxy->net_ns; 9008 struct bpf_link_primer link_primer; 9009 struct bpf_xdp_link *link; 9010 struct net_device *dev; 9011 int err, fd; 9012 9013 rtnl_lock(); 9014 dev = dev_get_by_index(net, attr->link_create.target_ifindex); 9015 if (!dev) { 9016 rtnl_unlock(); 9017 return -EINVAL; 9018 } 9019 9020 link = kzalloc(sizeof(*link), GFP_USER); 9021 if (!link) { 9022 err = -ENOMEM; 9023 goto unlock; 9024 } 9025 9026 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog); 9027 link->dev = dev; 9028 link->flags = attr->link_create.flags; 9029 9030 err = bpf_link_prime(&link->link, &link_primer); 9031 if (err) { 9032 kfree(link); 9033 goto unlock; 9034 } 9035 9036 err = dev_xdp_attach_link(dev, NULL, link); 9037 rtnl_unlock(); 9038 9039 if (err) { 9040 link->dev = NULL; 9041 bpf_link_cleanup(&link_primer); 9042 goto out_put_dev; 9043 } 9044 9045 fd = bpf_link_settle(&link_primer); 9046 /* link itself doesn't hold dev's refcnt to not complicate shutdown */ 9047 dev_put(dev); 9048 return fd; 9049 9050 unlock: 9051 rtnl_unlock(); 9052 9053 out_put_dev: 9054 dev_put(dev); 9055 return err; 9056 } 9057 9058 /** 9059 * dev_change_xdp_fd - set or clear a bpf program for a device rx path 9060 * @dev: device 9061 * @extack: netlink extended ack 9062 * @fd: new program fd or negative value to clear 9063 * @expected_fd: old program fd that userspace expects to replace or clear 9064 * @flags: xdp-related flags 9065 * 9066 * Set or clear a bpf program for a device 9067 */ 9068 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, 9069 int fd, int expected_fd, u32 flags) 9070 { 9071 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags); 9072 struct bpf_prog *new_prog = NULL, *old_prog = NULL; 9073 int err; 9074 9075 ASSERT_RTNL(); 9076 9077 if (fd >= 0) { 9078 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP, 9079 mode != XDP_MODE_SKB); 9080 if (IS_ERR(new_prog)) 9081 return PTR_ERR(new_prog); 9082 } 9083 9084 if (expected_fd >= 0) { 9085 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP, 9086 mode != XDP_MODE_SKB); 9087 if (IS_ERR(old_prog)) { 9088 err = PTR_ERR(old_prog); 9089 old_prog = NULL; 9090 goto err_out; 9091 } 9092 } 9093 9094 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags); 9095 9096 err_out: 9097 if (err && new_prog) 9098 bpf_prog_put(new_prog); 9099 if (old_prog) 9100 bpf_prog_put(old_prog); 9101 return err; 9102 } 9103 9104 /** 9105 * dev_new_index - allocate an ifindex 9106 * @net: the applicable net namespace 9107 * 9108 * Returns a suitable unique value for a new device interface 9109 * number. The caller must hold the rtnl semaphore or the 9110 * dev_base_lock to be sure it remains unique. 9111 */ 9112 static int dev_new_index(struct net *net) 9113 { 9114 int ifindex = net->ifindex; 9115 9116 for (;;) { 9117 if (++ifindex <= 0) 9118 ifindex = 1; 9119 if (!__dev_get_by_index(net, ifindex)) 9120 return net->ifindex = ifindex; 9121 } 9122 } 9123 9124 /* Delayed registration/unregisteration */ 9125 static LIST_HEAD(net_todo_list); 9126 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); 9127 9128 static void net_set_todo(struct net_device *dev) 9129 { 9130 list_add_tail(&dev->todo_list, &net_todo_list); 9131 dev_net(dev)->dev_unreg_count++; 9132 } 9133 9134 static netdev_features_t netdev_sync_upper_features(struct net_device *lower, 9135 struct net_device *upper, netdev_features_t features) 9136 { 9137 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9138 netdev_features_t feature; 9139 int feature_bit; 9140 9141 for_each_netdev_feature(upper_disables, feature_bit) { 9142 feature = __NETIF_F_BIT(feature_bit); 9143 if (!(upper->wanted_features & feature) 9144 && (features & feature)) { 9145 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", 9146 &feature, upper->name); 9147 features &= ~feature; 9148 } 9149 } 9150 9151 return features; 9152 } 9153 9154 static void netdev_sync_lower_features(struct net_device *upper, 9155 struct net_device *lower, netdev_features_t features) 9156 { 9157 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9158 netdev_features_t feature; 9159 int feature_bit; 9160 9161 for_each_netdev_feature(upper_disables, feature_bit) { 9162 feature = __NETIF_F_BIT(feature_bit); 9163 if (!(features & feature) && (lower->features & feature)) { 9164 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", 9165 &feature, lower->name); 9166 lower->wanted_features &= ~feature; 9167 __netdev_update_features(lower); 9168 9169 if (unlikely(lower->features & feature)) 9170 netdev_WARN(upper, "failed to disable %pNF on %s!\n", 9171 &feature, lower->name); 9172 else 9173 netdev_features_change(lower); 9174 } 9175 } 9176 } 9177 9178 static netdev_features_t netdev_fix_features(struct net_device *dev, 9179 netdev_features_t features) 9180 { 9181 /* Fix illegal checksum combinations */ 9182 if ((features & NETIF_F_HW_CSUM) && 9183 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { 9184 netdev_warn(dev, "mixed HW and IP checksum settings.\n"); 9185 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); 9186 } 9187 9188 /* TSO requires that SG is present as well. */ 9189 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { 9190 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); 9191 features &= ~NETIF_F_ALL_TSO; 9192 } 9193 9194 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && 9195 !(features & NETIF_F_IP_CSUM)) { 9196 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); 9197 features &= ~NETIF_F_TSO; 9198 features &= ~NETIF_F_TSO_ECN; 9199 } 9200 9201 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && 9202 !(features & NETIF_F_IPV6_CSUM)) { 9203 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); 9204 features &= ~NETIF_F_TSO6; 9205 } 9206 9207 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ 9208 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) 9209 features &= ~NETIF_F_TSO_MANGLEID; 9210 9211 /* TSO ECN requires that TSO is present as well. */ 9212 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) 9213 features &= ~NETIF_F_TSO_ECN; 9214 9215 /* Software GSO depends on SG. */ 9216 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { 9217 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); 9218 features &= ~NETIF_F_GSO; 9219 } 9220 9221 /* GSO partial features require GSO partial be set */ 9222 if ((features & dev->gso_partial_features) && 9223 !(features & NETIF_F_GSO_PARTIAL)) { 9224 netdev_dbg(dev, 9225 "Dropping partially supported GSO features since no GSO partial.\n"); 9226 features &= ~dev->gso_partial_features; 9227 } 9228 9229 if (!(features & NETIF_F_RXCSUM)) { 9230 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet 9231 * successfully merged by hardware must also have the 9232 * checksum verified by hardware. If the user does not 9233 * want to enable RXCSUM, logically, we should disable GRO_HW. 9234 */ 9235 if (features & NETIF_F_GRO_HW) { 9236 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n"); 9237 features &= ~NETIF_F_GRO_HW; 9238 } 9239 } 9240 9241 /* LRO/HW-GRO features cannot be combined with RX-FCS */ 9242 if (features & NETIF_F_RXFCS) { 9243 if (features & NETIF_F_LRO) { 9244 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n"); 9245 features &= ~NETIF_F_LRO; 9246 } 9247 9248 if (features & NETIF_F_GRO_HW) { 9249 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n"); 9250 features &= ~NETIF_F_GRO_HW; 9251 } 9252 } 9253 9254 if ((features & NETIF_F_GRO_HW) && (features & NETIF_F_LRO)) { 9255 netdev_dbg(dev, "Dropping LRO feature since HW-GRO is requested.\n"); 9256 features &= ~NETIF_F_LRO; 9257 } 9258 9259 if (features & NETIF_F_HW_TLS_TX) { 9260 bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) == 9261 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM); 9262 bool hw_csum = features & NETIF_F_HW_CSUM; 9263 9264 if (!ip_csum && !hw_csum) { 9265 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n"); 9266 features &= ~NETIF_F_HW_TLS_TX; 9267 } 9268 } 9269 9270 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) { 9271 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n"); 9272 features &= ~NETIF_F_HW_TLS_RX; 9273 } 9274 9275 return features; 9276 } 9277 9278 int __netdev_update_features(struct net_device *dev) 9279 { 9280 struct net_device *upper, *lower; 9281 netdev_features_t features; 9282 struct list_head *iter; 9283 int err = -1; 9284 9285 ASSERT_RTNL(); 9286 9287 features = netdev_get_wanted_features(dev); 9288 9289 if (dev->netdev_ops->ndo_fix_features) 9290 features = dev->netdev_ops->ndo_fix_features(dev, features); 9291 9292 /* driver might be less strict about feature dependencies */ 9293 features = netdev_fix_features(dev, features); 9294 9295 /* some features can't be enabled if they're off on an upper device */ 9296 netdev_for_each_upper_dev_rcu(dev, upper, iter) 9297 features = netdev_sync_upper_features(dev, upper, features); 9298 9299 if (dev->features == features) 9300 goto sync_lower; 9301 9302 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", 9303 &dev->features, &features); 9304 9305 if (dev->netdev_ops->ndo_set_features) 9306 err = dev->netdev_ops->ndo_set_features(dev, features); 9307 else 9308 err = 0; 9309 9310 if (unlikely(err < 0)) { 9311 netdev_err(dev, 9312 "set_features() failed (%d); wanted %pNF, left %pNF\n", 9313 err, &features, &dev->features); 9314 /* return non-0 since some features might have changed and 9315 * it's better to fire a spurious notification than miss it 9316 */ 9317 return -1; 9318 } 9319 9320 sync_lower: 9321 /* some features must be disabled on lower devices when disabled 9322 * on an upper device (think: bonding master or bridge) 9323 */ 9324 netdev_for_each_lower_dev(dev, lower, iter) 9325 netdev_sync_lower_features(dev, lower, features); 9326 9327 if (!err) { 9328 netdev_features_t diff = features ^ dev->features; 9329 9330 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) { 9331 /* udp_tunnel_{get,drop}_rx_info both need 9332 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the 9333 * device, or they won't do anything. 9334 * Thus we need to update dev->features 9335 * *before* calling udp_tunnel_get_rx_info, 9336 * but *after* calling udp_tunnel_drop_rx_info. 9337 */ 9338 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) { 9339 dev->features = features; 9340 udp_tunnel_get_rx_info(dev); 9341 } else { 9342 udp_tunnel_drop_rx_info(dev); 9343 } 9344 } 9345 9346 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) { 9347 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) { 9348 dev->features = features; 9349 err |= vlan_get_rx_ctag_filter_info(dev); 9350 } else { 9351 vlan_drop_rx_ctag_filter_info(dev); 9352 } 9353 } 9354 9355 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) { 9356 if (features & NETIF_F_HW_VLAN_STAG_FILTER) { 9357 dev->features = features; 9358 err |= vlan_get_rx_stag_filter_info(dev); 9359 } else { 9360 vlan_drop_rx_stag_filter_info(dev); 9361 } 9362 } 9363 9364 dev->features = features; 9365 } 9366 9367 return err < 0 ? 0 : 1; 9368 } 9369 9370 /** 9371 * netdev_update_features - recalculate device features 9372 * @dev: the device to check 9373 * 9374 * Recalculate dev->features set and send notifications if it 9375 * has changed. Should be called after driver or hardware dependent 9376 * conditions might have changed that influence the features. 9377 */ 9378 void netdev_update_features(struct net_device *dev) 9379 { 9380 if (__netdev_update_features(dev)) 9381 netdev_features_change(dev); 9382 } 9383 EXPORT_SYMBOL(netdev_update_features); 9384 9385 /** 9386 * netdev_change_features - recalculate device features 9387 * @dev: the device to check 9388 * 9389 * Recalculate dev->features set and send notifications even 9390 * if they have not changed. Should be called instead of 9391 * netdev_update_features() if also dev->vlan_features might 9392 * have changed to allow the changes to be propagated to stacked 9393 * VLAN devices. 9394 */ 9395 void netdev_change_features(struct net_device *dev) 9396 { 9397 __netdev_update_features(dev); 9398 netdev_features_change(dev); 9399 } 9400 EXPORT_SYMBOL(netdev_change_features); 9401 9402 /** 9403 * netif_stacked_transfer_operstate - transfer operstate 9404 * @rootdev: the root or lower level device to transfer state from 9405 * @dev: the device to transfer operstate to 9406 * 9407 * Transfer operational state from root to device. This is normally 9408 * called when a stacking relationship exists between the root 9409 * device and the device(a leaf device). 9410 */ 9411 void netif_stacked_transfer_operstate(const struct net_device *rootdev, 9412 struct net_device *dev) 9413 { 9414 if (rootdev->operstate == IF_OPER_DORMANT) 9415 netif_dormant_on(dev); 9416 else 9417 netif_dormant_off(dev); 9418 9419 if (rootdev->operstate == IF_OPER_TESTING) 9420 netif_testing_on(dev); 9421 else 9422 netif_testing_off(dev); 9423 9424 if (netif_carrier_ok(rootdev)) 9425 netif_carrier_on(dev); 9426 else 9427 netif_carrier_off(dev); 9428 } 9429 EXPORT_SYMBOL(netif_stacked_transfer_operstate); 9430 9431 static int netif_alloc_rx_queues(struct net_device *dev) 9432 { 9433 unsigned int i, count = dev->num_rx_queues; 9434 struct netdev_rx_queue *rx; 9435 size_t sz = count * sizeof(*rx); 9436 int err = 0; 9437 9438 BUG_ON(count < 1); 9439 9440 rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 9441 if (!rx) 9442 return -ENOMEM; 9443 9444 dev->_rx = rx; 9445 9446 for (i = 0; i < count; i++) { 9447 rx[i].dev = dev; 9448 9449 /* XDP RX-queue setup */ 9450 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0); 9451 if (err < 0) 9452 goto err_rxq_info; 9453 } 9454 return 0; 9455 9456 err_rxq_info: 9457 /* Rollback successful reg's and free other resources */ 9458 while (i--) 9459 xdp_rxq_info_unreg(&rx[i].xdp_rxq); 9460 kvfree(dev->_rx); 9461 dev->_rx = NULL; 9462 return err; 9463 } 9464 9465 static void netif_free_rx_queues(struct net_device *dev) 9466 { 9467 unsigned int i, count = dev->num_rx_queues; 9468 9469 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */ 9470 if (!dev->_rx) 9471 return; 9472 9473 for (i = 0; i < count; i++) 9474 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq); 9475 9476 kvfree(dev->_rx); 9477 } 9478 9479 static void netdev_init_one_queue(struct net_device *dev, 9480 struct netdev_queue *queue, void *_unused) 9481 { 9482 /* Initialize queue lock */ 9483 spin_lock_init(&queue->_xmit_lock); 9484 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); 9485 queue->xmit_lock_owner = -1; 9486 netdev_queue_numa_node_write(queue, NUMA_NO_NODE); 9487 queue->dev = dev; 9488 #ifdef CONFIG_BQL 9489 dql_init(&queue->dql, HZ); 9490 #endif 9491 } 9492 9493 static void netif_free_tx_queues(struct net_device *dev) 9494 { 9495 kvfree(dev->_tx); 9496 } 9497 9498 static int netif_alloc_netdev_queues(struct net_device *dev) 9499 { 9500 unsigned int count = dev->num_tx_queues; 9501 struct netdev_queue *tx; 9502 size_t sz = count * sizeof(*tx); 9503 9504 if (count < 1 || count > 0xffff) 9505 return -EINVAL; 9506 9507 tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 9508 if (!tx) 9509 return -ENOMEM; 9510 9511 dev->_tx = tx; 9512 9513 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); 9514 spin_lock_init(&dev->tx_global_lock); 9515 9516 return 0; 9517 } 9518 9519 void netif_tx_stop_all_queues(struct net_device *dev) 9520 { 9521 unsigned int i; 9522 9523 for (i = 0; i < dev->num_tx_queues; i++) { 9524 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 9525 9526 netif_tx_stop_queue(txq); 9527 } 9528 } 9529 EXPORT_SYMBOL(netif_tx_stop_all_queues); 9530 9531 /** 9532 * register_netdevice - register a network device 9533 * @dev: device to register 9534 * 9535 * Take a completed network device structure and add it to the kernel 9536 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 9537 * chain. 0 is returned on success. A negative errno code is returned 9538 * on a failure to set up the device, or if the name is a duplicate. 9539 * 9540 * Callers must hold the rtnl semaphore. You may want 9541 * register_netdev() instead of this. 9542 * 9543 * BUGS: 9544 * The locking appears insufficient to guarantee two parallel registers 9545 * will not get the same name. 9546 */ 9547 9548 int register_netdevice(struct net_device *dev) 9549 { 9550 int ret; 9551 struct net *net = dev_net(dev); 9552 9553 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE < 9554 NETDEV_FEATURE_COUNT); 9555 BUG_ON(dev_boot_phase); 9556 ASSERT_RTNL(); 9557 9558 might_sleep(); 9559 9560 /* When net_device's are persistent, this will be fatal. */ 9561 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); 9562 BUG_ON(!net); 9563 9564 ret = ethtool_check_ops(dev->ethtool_ops); 9565 if (ret) 9566 return ret; 9567 9568 spin_lock_init(&dev->addr_list_lock); 9569 netdev_set_addr_lockdep_class(dev); 9570 9571 ret = dev_get_valid_name(net, dev, dev->name); 9572 if (ret < 0) 9573 goto out; 9574 9575 ret = -ENOMEM; 9576 dev->name_node = netdev_name_node_head_alloc(dev); 9577 if (!dev->name_node) 9578 goto out; 9579 9580 /* Init, if this function is available */ 9581 if (dev->netdev_ops->ndo_init) { 9582 ret = dev->netdev_ops->ndo_init(dev); 9583 if (ret) { 9584 if (ret > 0) 9585 ret = -EIO; 9586 goto err_free_name; 9587 } 9588 } 9589 9590 if (((dev->hw_features | dev->features) & 9591 NETIF_F_HW_VLAN_CTAG_FILTER) && 9592 (!dev->netdev_ops->ndo_vlan_rx_add_vid || 9593 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { 9594 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); 9595 ret = -EINVAL; 9596 goto err_uninit; 9597 } 9598 9599 ret = -EBUSY; 9600 if (!dev->ifindex) 9601 dev->ifindex = dev_new_index(net); 9602 else if (__dev_get_by_index(net, dev->ifindex)) 9603 goto err_uninit; 9604 9605 /* Transfer changeable features to wanted_features and enable 9606 * software offloads (GSO and GRO). 9607 */ 9608 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF); 9609 dev->features |= NETIF_F_SOFT_FEATURES; 9610 9611 if (dev->udp_tunnel_nic_info) { 9612 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT; 9613 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT; 9614 } 9615 9616 dev->wanted_features = dev->features & dev->hw_features; 9617 9618 if (!(dev->flags & IFF_LOOPBACK)) 9619 dev->hw_features |= NETIF_F_NOCACHE_COPY; 9620 9621 /* If IPv4 TCP segmentation offload is supported we should also 9622 * allow the device to enable segmenting the frame with the option 9623 * of ignoring a static IP ID value. This doesn't enable the 9624 * feature itself but allows the user to enable it later. 9625 */ 9626 if (dev->hw_features & NETIF_F_TSO) 9627 dev->hw_features |= NETIF_F_TSO_MANGLEID; 9628 if (dev->vlan_features & NETIF_F_TSO) 9629 dev->vlan_features |= NETIF_F_TSO_MANGLEID; 9630 if (dev->mpls_features & NETIF_F_TSO) 9631 dev->mpls_features |= NETIF_F_TSO_MANGLEID; 9632 if (dev->hw_enc_features & NETIF_F_TSO) 9633 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; 9634 9635 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. 9636 */ 9637 dev->vlan_features |= NETIF_F_HIGHDMA; 9638 9639 /* Make NETIF_F_SG inheritable to tunnel devices. 9640 */ 9641 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; 9642 9643 /* Make NETIF_F_SG inheritable to MPLS. 9644 */ 9645 dev->mpls_features |= NETIF_F_SG; 9646 9647 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); 9648 ret = notifier_to_errno(ret); 9649 if (ret) 9650 goto err_uninit; 9651 9652 ret = netdev_register_kobject(dev); 9653 if (ret) { 9654 dev->reg_state = NETREG_UNREGISTERED; 9655 goto err_uninit; 9656 } 9657 dev->reg_state = NETREG_REGISTERED; 9658 9659 __netdev_update_features(dev); 9660 9661 /* 9662 * Default initial state at registry is that the 9663 * device is present. 9664 */ 9665 9666 set_bit(__LINK_STATE_PRESENT, &dev->state); 9667 9668 linkwatch_init_dev(dev); 9669 9670 dev_init_scheduler(dev); 9671 dev_hold(dev); 9672 list_netdevice(dev); 9673 add_device_randomness(dev->dev_addr, dev->addr_len); 9674 9675 /* If the device has permanent device address, driver should 9676 * set dev_addr and also addr_assign_type should be set to 9677 * NET_ADDR_PERM (default value). 9678 */ 9679 if (dev->addr_assign_type == NET_ADDR_PERM) 9680 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); 9681 9682 /* Notify protocols, that a new device appeared. */ 9683 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); 9684 ret = notifier_to_errno(ret); 9685 if (ret) { 9686 /* Expect explicit free_netdev() on failure */ 9687 dev->needs_free_netdev = false; 9688 unregister_netdevice_queue(dev, NULL); 9689 goto out; 9690 } 9691 /* 9692 * Prevent userspace races by waiting until the network 9693 * device is fully setup before sending notifications. 9694 */ 9695 if (!dev->rtnl_link_ops || 9696 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 9697 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); 9698 9699 out: 9700 return ret; 9701 9702 err_uninit: 9703 if (dev->netdev_ops->ndo_uninit) 9704 dev->netdev_ops->ndo_uninit(dev); 9705 if (dev->priv_destructor) 9706 dev->priv_destructor(dev); 9707 err_free_name: 9708 netdev_name_node_free(dev->name_node); 9709 goto out; 9710 } 9711 EXPORT_SYMBOL(register_netdevice); 9712 9713 /** 9714 * init_dummy_netdev - init a dummy network device for NAPI 9715 * @dev: device to init 9716 * 9717 * This takes a network device structure and initialize the minimum 9718 * amount of fields so it can be used to schedule NAPI polls without 9719 * registering a full blown interface. This is to be used by drivers 9720 * that need to tie several hardware interfaces to a single NAPI 9721 * poll scheduler due to HW limitations. 9722 */ 9723 int init_dummy_netdev(struct net_device *dev) 9724 { 9725 /* Clear everything. Note we don't initialize spinlocks 9726 * are they aren't supposed to be taken by any of the 9727 * NAPI code and this dummy netdev is supposed to be 9728 * only ever used for NAPI polls 9729 */ 9730 memset(dev, 0, sizeof(struct net_device)); 9731 9732 /* make sure we BUG if trying to hit standard 9733 * register/unregister code path 9734 */ 9735 dev->reg_state = NETREG_DUMMY; 9736 9737 /* NAPI wants this */ 9738 INIT_LIST_HEAD(&dev->napi_list); 9739 9740 /* a dummy interface is started by default */ 9741 set_bit(__LINK_STATE_PRESENT, &dev->state); 9742 set_bit(__LINK_STATE_START, &dev->state); 9743 9744 /* napi_busy_loop stats accounting wants this */ 9745 dev_net_set(dev, &init_net); 9746 9747 /* Note : We dont allocate pcpu_refcnt for dummy devices, 9748 * because users of this 'device' dont need to change 9749 * its refcount. 9750 */ 9751 9752 return 0; 9753 } 9754 EXPORT_SYMBOL_GPL(init_dummy_netdev); 9755 9756 9757 /** 9758 * register_netdev - register a network device 9759 * @dev: device to register 9760 * 9761 * Take a completed network device structure and add it to the kernel 9762 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 9763 * chain. 0 is returned on success. A negative errno code is returned 9764 * on a failure to set up the device, or if the name is a duplicate. 9765 * 9766 * This is a wrapper around register_netdevice that takes the rtnl semaphore 9767 * and expands the device name if you passed a format string to 9768 * alloc_netdev. 9769 */ 9770 int register_netdev(struct net_device *dev) 9771 { 9772 int err; 9773 9774 if (rtnl_lock_killable()) 9775 return -EINTR; 9776 err = register_netdevice(dev); 9777 rtnl_unlock(); 9778 return err; 9779 } 9780 EXPORT_SYMBOL(register_netdev); 9781 9782 int netdev_refcnt_read(const struct net_device *dev) 9783 { 9784 #ifdef CONFIG_PCPU_DEV_REFCNT 9785 int i, refcnt = 0; 9786 9787 for_each_possible_cpu(i) 9788 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); 9789 return refcnt; 9790 #else 9791 return refcount_read(&dev->dev_refcnt); 9792 #endif 9793 } 9794 EXPORT_SYMBOL(netdev_refcnt_read); 9795 9796 int netdev_unregister_timeout_secs __read_mostly = 10; 9797 9798 #define WAIT_REFS_MIN_MSECS 1 9799 #define WAIT_REFS_MAX_MSECS 250 9800 /** 9801 * netdev_wait_allrefs - wait until all references are gone. 9802 * @dev: target net_device 9803 * 9804 * This is called when unregistering network devices. 9805 * 9806 * Any protocol or device that holds a reference should register 9807 * for netdevice notification, and cleanup and put back the 9808 * reference if they receive an UNREGISTER event. 9809 * We can get stuck here if buggy protocols don't correctly 9810 * call dev_put. 9811 */ 9812 static void netdev_wait_allrefs(struct net_device *dev) 9813 { 9814 unsigned long rebroadcast_time, warning_time; 9815 int wait = 0, refcnt; 9816 9817 linkwatch_forget_dev(dev); 9818 9819 rebroadcast_time = warning_time = jiffies; 9820 refcnt = netdev_refcnt_read(dev); 9821 9822 while (refcnt != 1) { 9823 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { 9824 rtnl_lock(); 9825 9826 /* Rebroadcast unregister notification */ 9827 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 9828 9829 __rtnl_unlock(); 9830 rcu_barrier(); 9831 rtnl_lock(); 9832 9833 if (test_bit(__LINK_STATE_LINKWATCH_PENDING, 9834 &dev->state)) { 9835 /* We must not have linkwatch events 9836 * pending on unregister. If this 9837 * happens, we simply run the queue 9838 * unscheduled, resulting in a noop 9839 * for this device. 9840 */ 9841 linkwatch_run_queue(); 9842 } 9843 9844 __rtnl_unlock(); 9845 9846 rebroadcast_time = jiffies; 9847 } 9848 9849 if (!wait) { 9850 rcu_barrier(); 9851 wait = WAIT_REFS_MIN_MSECS; 9852 } else { 9853 msleep(wait); 9854 wait = min(wait << 1, WAIT_REFS_MAX_MSECS); 9855 } 9856 9857 refcnt = netdev_refcnt_read(dev); 9858 9859 if (refcnt != 1 && 9860 time_after(jiffies, warning_time + 9861 netdev_unregister_timeout_secs * HZ)) { 9862 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", 9863 dev->name, refcnt); 9864 warning_time = jiffies; 9865 } 9866 } 9867 } 9868 9869 /* The sequence is: 9870 * 9871 * rtnl_lock(); 9872 * ... 9873 * register_netdevice(x1); 9874 * register_netdevice(x2); 9875 * ... 9876 * unregister_netdevice(y1); 9877 * unregister_netdevice(y2); 9878 * ... 9879 * rtnl_unlock(); 9880 * free_netdev(y1); 9881 * free_netdev(y2); 9882 * 9883 * We are invoked by rtnl_unlock(). 9884 * This allows us to deal with problems: 9885 * 1) We can delete sysfs objects which invoke hotplug 9886 * without deadlocking with linkwatch via keventd. 9887 * 2) Since we run with the RTNL semaphore not held, we can sleep 9888 * safely in order to wait for the netdev refcnt to drop to zero. 9889 * 9890 * We must not return until all unregister events added during 9891 * the interval the lock was held have been completed. 9892 */ 9893 void netdev_run_todo(void) 9894 { 9895 struct list_head list; 9896 #ifdef CONFIG_LOCKDEP 9897 struct list_head unlink_list; 9898 9899 list_replace_init(&net_unlink_list, &unlink_list); 9900 9901 while (!list_empty(&unlink_list)) { 9902 struct net_device *dev = list_first_entry(&unlink_list, 9903 struct net_device, 9904 unlink_list); 9905 list_del_init(&dev->unlink_list); 9906 dev->nested_level = dev->lower_level - 1; 9907 } 9908 #endif 9909 9910 /* Snapshot list, allow later requests */ 9911 list_replace_init(&net_todo_list, &list); 9912 9913 __rtnl_unlock(); 9914 9915 9916 /* Wait for rcu callbacks to finish before next phase */ 9917 if (!list_empty(&list)) 9918 rcu_barrier(); 9919 9920 while (!list_empty(&list)) { 9921 struct net_device *dev 9922 = list_first_entry(&list, struct net_device, todo_list); 9923 list_del(&dev->todo_list); 9924 9925 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { 9926 pr_err("network todo '%s' but state %d\n", 9927 dev->name, dev->reg_state); 9928 dump_stack(); 9929 continue; 9930 } 9931 9932 dev->reg_state = NETREG_UNREGISTERED; 9933 9934 netdev_wait_allrefs(dev); 9935 9936 /* paranoia */ 9937 BUG_ON(netdev_refcnt_read(dev) != 1); 9938 BUG_ON(!list_empty(&dev->ptype_all)); 9939 BUG_ON(!list_empty(&dev->ptype_specific)); 9940 WARN_ON(rcu_access_pointer(dev->ip_ptr)); 9941 WARN_ON(rcu_access_pointer(dev->ip6_ptr)); 9942 #if IS_ENABLED(CONFIG_DECNET) 9943 WARN_ON(dev->dn_ptr); 9944 #endif 9945 if (dev->priv_destructor) 9946 dev->priv_destructor(dev); 9947 if (dev->needs_free_netdev) 9948 free_netdev(dev); 9949 9950 /* Report a network device has been unregistered */ 9951 rtnl_lock(); 9952 dev_net(dev)->dev_unreg_count--; 9953 __rtnl_unlock(); 9954 wake_up(&netdev_unregistering_wq); 9955 9956 /* Free network device */ 9957 kobject_put(&dev->dev.kobj); 9958 } 9959 } 9960 9961 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has 9962 * all the same fields in the same order as net_device_stats, with only 9963 * the type differing, but rtnl_link_stats64 may have additional fields 9964 * at the end for newer counters. 9965 */ 9966 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, 9967 const struct net_device_stats *netdev_stats) 9968 { 9969 #if BITS_PER_LONG == 64 9970 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats)); 9971 memcpy(stats64, netdev_stats, sizeof(*netdev_stats)); 9972 /* zero out counters that only exist in rtnl_link_stats64 */ 9973 memset((char *)stats64 + sizeof(*netdev_stats), 0, 9974 sizeof(*stats64) - sizeof(*netdev_stats)); 9975 #else 9976 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long); 9977 const unsigned long *src = (const unsigned long *)netdev_stats; 9978 u64 *dst = (u64 *)stats64; 9979 9980 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); 9981 for (i = 0; i < n; i++) 9982 dst[i] = src[i]; 9983 /* zero out counters that only exist in rtnl_link_stats64 */ 9984 memset((char *)stats64 + n * sizeof(u64), 0, 9985 sizeof(*stats64) - n * sizeof(u64)); 9986 #endif 9987 } 9988 EXPORT_SYMBOL(netdev_stats_to_stats64); 9989 9990 /** 9991 * dev_get_stats - get network device statistics 9992 * @dev: device to get statistics from 9993 * @storage: place to store stats 9994 * 9995 * Get network statistics from device. Return @storage. 9996 * The device driver may provide its own method by setting 9997 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; 9998 * otherwise the internal statistics structure is used. 9999 */ 10000 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, 10001 struct rtnl_link_stats64 *storage) 10002 { 10003 const struct net_device_ops *ops = dev->netdev_ops; 10004 10005 if (ops->ndo_get_stats64) { 10006 memset(storage, 0, sizeof(*storage)); 10007 ops->ndo_get_stats64(dev, storage); 10008 } else if (ops->ndo_get_stats) { 10009 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); 10010 } else { 10011 netdev_stats_to_stats64(storage, &dev->stats); 10012 } 10013 storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped); 10014 storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped); 10015 storage->rx_nohandler += (unsigned long)atomic_long_read(&dev->rx_nohandler); 10016 return storage; 10017 } 10018 EXPORT_SYMBOL(dev_get_stats); 10019 10020 /** 10021 * dev_fetch_sw_netstats - get per-cpu network device statistics 10022 * @s: place to store stats 10023 * @netstats: per-cpu network stats to read from 10024 * 10025 * Read per-cpu network statistics and populate the related fields in @s. 10026 */ 10027 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s, 10028 const struct pcpu_sw_netstats __percpu *netstats) 10029 { 10030 int cpu; 10031 10032 for_each_possible_cpu(cpu) { 10033 const struct pcpu_sw_netstats *stats; 10034 struct pcpu_sw_netstats tmp; 10035 unsigned int start; 10036 10037 stats = per_cpu_ptr(netstats, cpu); 10038 do { 10039 start = u64_stats_fetch_begin_irq(&stats->syncp); 10040 tmp.rx_packets = stats->rx_packets; 10041 tmp.rx_bytes = stats->rx_bytes; 10042 tmp.tx_packets = stats->tx_packets; 10043 tmp.tx_bytes = stats->tx_bytes; 10044 } while (u64_stats_fetch_retry_irq(&stats->syncp, start)); 10045 10046 s->rx_packets += tmp.rx_packets; 10047 s->rx_bytes += tmp.rx_bytes; 10048 s->tx_packets += tmp.tx_packets; 10049 s->tx_bytes += tmp.tx_bytes; 10050 } 10051 } 10052 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats); 10053 10054 /** 10055 * dev_get_tstats64 - ndo_get_stats64 implementation 10056 * @dev: device to get statistics from 10057 * @s: place to store stats 10058 * 10059 * Populate @s from dev->stats and dev->tstats. Can be used as 10060 * ndo_get_stats64() callback. 10061 */ 10062 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s) 10063 { 10064 netdev_stats_to_stats64(s, &dev->stats); 10065 dev_fetch_sw_netstats(s, dev->tstats); 10066 } 10067 EXPORT_SYMBOL_GPL(dev_get_tstats64); 10068 10069 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) 10070 { 10071 struct netdev_queue *queue = dev_ingress_queue(dev); 10072 10073 #ifdef CONFIG_NET_CLS_ACT 10074 if (queue) 10075 return queue; 10076 queue = kzalloc(sizeof(*queue), GFP_KERNEL); 10077 if (!queue) 10078 return NULL; 10079 netdev_init_one_queue(dev, queue, NULL); 10080 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); 10081 queue->qdisc_sleeping = &noop_qdisc; 10082 rcu_assign_pointer(dev->ingress_queue, queue); 10083 #endif 10084 return queue; 10085 } 10086 10087 static const struct ethtool_ops default_ethtool_ops; 10088 10089 void netdev_set_default_ethtool_ops(struct net_device *dev, 10090 const struct ethtool_ops *ops) 10091 { 10092 if (dev->ethtool_ops == &default_ethtool_ops) 10093 dev->ethtool_ops = ops; 10094 } 10095 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); 10096 10097 void netdev_freemem(struct net_device *dev) 10098 { 10099 char *addr = (char *)dev - dev->padded; 10100 10101 kvfree(addr); 10102 } 10103 10104 /** 10105 * alloc_netdev_mqs - allocate network device 10106 * @sizeof_priv: size of private data to allocate space for 10107 * @name: device name format string 10108 * @name_assign_type: origin of device name 10109 * @setup: callback to initialize device 10110 * @txqs: the number of TX subqueues to allocate 10111 * @rxqs: the number of RX subqueues to allocate 10112 * 10113 * Allocates a struct net_device with private data area for driver use 10114 * and performs basic initialization. Also allocates subqueue structs 10115 * for each queue on the device. 10116 */ 10117 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, 10118 unsigned char name_assign_type, 10119 void (*setup)(struct net_device *), 10120 unsigned int txqs, unsigned int rxqs) 10121 { 10122 struct net_device *dev; 10123 unsigned int alloc_size; 10124 struct net_device *p; 10125 10126 BUG_ON(strlen(name) >= sizeof(dev->name)); 10127 10128 if (txqs < 1) { 10129 pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); 10130 return NULL; 10131 } 10132 10133 if (rxqs < 1) { 10134 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); 10135 return NULL; 10136 } 10137 10138 alloc_size = sizeof(struct net_device); 10139 if (sizeof_priv) { 10140 /* ensure 32-byte alignment of private area */ 10141 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); 10142 alloc_size += sizeof_priv; 10143 } 10144 /* ensure 32-byte alignment of whole construct */ 10145 alloc_size += NETDEV_ALIGN - 1; 10146 10147 p = kvzalloc(alloc_size, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10148 if (!p) 10149 return NULL; 10150 10151 dev = PTR_ALIGN(p, NETDEV_ALIGN); 10152 dev->padded = (char *)dev - (char *)p; 10153 10154 #ifdef CONFIG_PCPU_DEV_REFCNT 10155 dev->pcpu_refcnt = alloc_percpu(int); 10156 if (!dev->pcpu_refcnt) 10157 goto free_dev; 10158 dev_hold(dev); 10159 #else 10160 refcount_set(&dev->dev_refcnt, 1); 10161 #endif 10162 10163 if (dev_addr_init(dev)) 10164 goto free_pcpu; 10165 10166 dev_mc_init(dev); 10167 dev_uc_init(dev); 10168 10169 dev_net_set(dev, &init_net); 10170 10171 dev->gso_max_size = GSO_MAX_SIZE; 10172 dev->gso_max_segs = GSO_MAX_SEGS; 10173 dev->upper_level = 1; 10174 dev->lower_level = 1; 10175 #ifdef CONFIG_LOCKDEP 10176 dev->nested_level = 0; 10177 INIT_LIST_HEAD(&dev->unlink_list); 10178 #endif 10179 10180 INIT_LIST_HEAD(&dev->napi_list); 10181 INIT_LIST_HEAD(&dev->unreg_list); 10182 INIT_LIST_HEAD(&dev->close_list); 10183 INIT_LIST_HEAD(&dev->link_watch_list); 10184 INIT_LIST_HEAD(&dev->adj_list.upper); 10185 INIT_LIST_HEAD(&dev->adj_list.lower); 10186 INIT_LIST_HEAD(&dev->ptype_all); 10187 INIT_LIST_HEAD(&dev->ptype_specific); 10188 INIT_LIST_HEAD(&dev->net_notifier_list); 10189 #ifdef CONFIG_NET_SCHED 10190 hash_init(dev->qdisc_hash); 10191 #endif 10192 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; 10193 setup(dev); 10194 10195 if (!dev->tx_queue_len) { 10196 dev->priv_flags |= IFF_NO_QUEUE; 10197 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; 10198 } 10199 10200 dev->num_tx_queues = txqs; 10201 dev->real_num_tx_queues = txqs; 10202 if (netif_alloc_netdev_queues(dev)) 10203 goto free_all; 10204 10205 dev->num_rx_queues = rxqs; 10206 dev->real_num_rx_queues = rxqs; 10207 if (netif_alloc_rx_queues(dev)) 10208 goto free_all; 10209 10210 strcpy(dev->name, name); 10211 dev->name_assign_type = name_assign_type; 10212 dev->group = INIT_NETDEV_GROUP; 10213 if (!dev->ethtool_ops) 10214 dev->ethtool_ops = &default_ethtool_ops; 10215 10216 nf_hook_netdev_init(dev); 10217 10218 return dev; 10219 10220 free_all: 10221 free_netdev(dev); 10222 return NULL; 10223 10224 free_pcpu: 10225 #ifdef CONFIG_PCPU_DEV_REFCNT 10226 free_percpu(dev->pcpu_refcnt); 10227 free_dev: 10228 #endif 10229 netdev_freemem(dev); 10230 return NULL; 10231 } 10232 EXPORT_SYMBOL(alloc_netdev_mqs); 10233 10234 /** 10235 * free_netdev - free network device 10236 * @dev: device 10237 * 10238 * This function does the last stage of destroying an allocated device 10239 * interface. The reference to the device object is released. If this 10240 * is the last reference then it will be freed.Must be called in process 10241 * context. 10242 */ 10243 void free_netdev(struct net_device *dev) 10244 { 10245 struct napi_struct *p, *n; 10246 10247 might_sleep(); 10248 10249 /* When called immediately after register_netdevice() failed the unwind 10250 * handling may still be dismantling the device. Handle that case by 10251 * deferring the free. 10252 */ 10253 if (dev->reg_state == NETREG_UNREGISTERING) { 10254 ASSERT_RTNL(); 10255 dev->needs_free_netdev = true; 10256 return; 10257 } 10258 10259 netif_free_tx_queues(dev); 10260 netif_free_rx_queues(dev); 10261 10262 kfree(rcu_dereference_protected(dev->ingress_queue, 1)); 10263 10264 /* Flush device addresses */ 10265 dev_addr_flush(dev); 10266 10267 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) 10268 netif_napi_del(p); 10269 10270 #ifdef CONFIG_PCPU_DEV_REFCNT 10271 free_percpu(dev->pcpu_refcnt); 10272 dev->pcpu_refcnt = NULL; 10273 #endif 10274 free_percpu(dev->xdp_bulkq); 10275 dev->xdp_bulkq = NULL; 10276 10277 /* Compatibility with error handling in drivers */ 10278 if (dev->reg_state == NETREG_UNINITIALIZED) { 10279 netdev_freemem(dev); 10280 return; 10281 } 10282 10283 BUG_ON(dev->reg_state != NETREG_UNREGISTERED); 10284 dev->reg_state = NETREG_RELEASED; 10285 10286 /* will free via device release */ 10287 put_device(&dev->dev); 10288 } 10289 EXPORT_SYMBOL(free_netdev); 10290 10291 /** 10292 * synchronize_net - Synchronize with packet receive processing 10293 * 10294 * Wait for packets currently being received to be done. 10295 * Does not block later packets from starting. 10296 */ 10297 void synchronize_net(void) 10298 { 10299 might_sleep(); 10300 if (rtnl_is_locked()) 10301 synchronize_rcu_expedited(); 10302 else 10303 synchronize_rcu(); 10304 } 10305 EXPORT_SYMBOL(synchronize_net); 10306 10307 /** 10308 * unregister_netdevice_queue - remove device from the kernel 10309 * @dev: device 10310 * @head: list 10311 * 10312 * This function shuts down a device interface and removes it 10313 * from the kernel tables. 10314 * If head not NULL, device is queued to be unregistered later. 10315 * 10316 * Callers must hold the rtnl semaphore. You may want 10317 * unregister_netdev() instead of this. 10318 */ 10319 10320 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) 10321 { 10322 ASSERT_RTNL(); 10323 10324 if (head) { 10325 list_move_tail(&dev->unreg_list, head); 10326 } else { 10327 LIST_HEAD(single); 10328 10329 list_add(&dev->unreg_list, &single); 10330 unregister_netdevice_many(&single); 10331 } 10332 } 10333 EXPORT_SYMBOL(unregister_netdevice_queue); 10334 10335 /** 10336 * unregister_netdevice_many - unregister many devices 10337 * @head: list of devices 10338 * 10339 * Note: As most callers use a stack allocated list_head, 10340 * we force a list_del() to make sure stack wont be corrupted later. 10341 */ 10342 void unregister_netdevice_many(struct list_head *head) 10343 { 10344 struct net_device *dev, *tmp; 10345 LIST_HEAD(close_head); 10346 10347 BUG_ON(dev_boot_phase); 10348 ASSERT_RTNL(); 10349 10350 if (list_empty(head)) 10351 return; 10352 10353 list_for_each_entry_safe(dev, tmp, head, unreg_list) { 10354 /* Some devices call without registering 10355 * for initialization unwind. Remove those 10356 * devices and proceed with the remaining. 10357 */ 10358 if (dev->reg_state == NETREG_UNINITIALIZED) { 10359 pr_debug("unregister_netdevice: device %s/%p never was registered\n", 10360 dev->name, dev); 10361 10362 WARN_ON(1); 10363 list_del(&dev->unreg_list); 10364 continue; 10365 } 10366 dev->dismantle = true; 10367 BUG_ON(dev->reg_state != NETREG_REGISTERED); 10368 } 10369 10370 /* If device is running, close it first. */ 10371 list_for_each_entry(dev, head, unreg_list) 10372 list_add_tail(&dev->close_list, &close_head); 10373 dev_close_many(&close_head, true); 10374 10375 list_for_each_entry(dev, head, unreg_list) { 10376 /* And unlink it from device chain. */ 10377 unlist_netdevice(dev); 10378 10379 dev->reg_state = NETREG_UNREGISTERING; 10380 } 10381 flush_all_backlogs(); 10382 10383 synchronize_net(); 10384 10385 list_for_each_entry(dev, head, unreg_list) { 10386 struct sk_buff *skb = NULL; 10387 10388 /* Shutdown queueing discipline. */ 10389 dev_shutdown(dev); 10390 10391 dev_xdp_uninstall(dev); 10392 10393 /* Notify protocols, that we are about to destroy 10394 * this device. They should clean all the things. 10395 */ 10396 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10397 10398 if (!dev->rtnl_link_ops || 10399 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 10400 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0, 10401 GFP_KERNEL, NULL, 0); 10402 10403 /* 10404 * Flush the unicast and multicast chains 10405 */ 10406 dev_uc_flush(dev); 10407 dev_mc_flush(dev); 10408 10409 netdev_name_node_alt_flush(dev); 10410 netdev_name_node_free(dev->name_node); 10411 10412 if (dev->netdev_ops->ndo_uninit) 10413 dev->netdev_ops->ndo_uninit(dev); 10414 10415 if (skb) 10416 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL); 10417 10418 /* Notifier chain MUST detach us all upper devices. */ 10419 WARN_ON(netdev_has_any_upper_dev(dev)); 10420 WARN_ON(netdev_has_any_lower_dev(dev)); 10421 10422 /* Remove entries from kobject tree */ 10423 netdev_unregister_kobject(dev); 10424 #ifdef CONFIG_XPS 10425 /* Remove XPS queueing entries */ 10426 netif_reset_xps_queues_gt(dev, 0); 10427 #endif 10428 } 10429 10430 synchronize_net(); 10431 10432 list_for_each_entry(dev, head, unreg_list) { 10433 dev_put(dev); 10434 net_set_todo(dev); 10435 } 10436 10437 list_del(head); 10438 } 10439 EXPORT_SYMBOL(unregister_netdevice_many); 10440 10441 /** 10442 * unregister_netdev - remove device from the kernel 10443 * @dev: device 10444 * 10445 * This function shuts down a device interface and removes it 10446 * from the kernel tables. 10447 * 10448 * This is just a wrapper for unregister_netdevice that takes 10449 * the rtnl semaphore. In general you want to use this and not 10450 * unregister_netdevice. 10451 */ 10452 void unregister_netdev(struct net_device *dev) 10453 { 10454 rtnl_lock(); 10455 unregister_netdevice(dev); 10456 rtnl_unlock(); 10457 } 10458 EXPORT_SYMBOL(unregister_netdev); 10459 10460 /** 10461 * __dev_change_net_namespace - move device to different nethost namespace 10462 * @dev: device 10463 * @net: network namespace 10464 * @pat: If not NULL name pattern to try if the current device name 10465 * is already taken in the destination network namespace. 10466 * @new_ifindex: If not zero, specifies device index in the target 10467 * namespace. 10468 * 10469 * This function shuts down a device interface and moves it 10470 * to a new network namespace. On success 0 is returned, on 10471 * a failure a netagive errno code is returned. 10472 * 10473 * Callers must hold the rtnl semaphore. 10474 */ 10475 10476 int __dev_change_net_namespace(struct net_device *dev, struct net *net, 10477 const char *pat, int new_ifindex) 10478 { 10479 struct net *net_old = dev_net(dev); 10480 int err, new_nsid; 10481 10482 ASSERT_RTNL(); 10483 10484 /* Don't allow namespace local devices to be moved. */ 10485 err = -EINVAL; 10486 if (dev->features & NETIF_F_NETNS_LOCAL) 10487 goto out; 10488 10489 /* Ensure the device has been registrered */ 10490 if (dev->reg_state != NETREG_REGISTERED) 10491 goto out; 10492 10493 /* Get out if there is nothing todo */ 10494 err = 0; 10495 if (net_eq(net_old, net)) 10496 goto out; 10497 10498 /* Pick the destination device name, and ensure 10499 * we can use it in the destination network namespace. 10500 */ 10501 err = -EEXIST; 10502 if (netdev_name_in_use(net, dev->name)) { 10503 /* We get here if we can't use the current device name */ 10504 if (!pat) 10505 goto out; 10506 err = dev_get_valid_name(net, dev, pat); 10507 if (err < 0) 10508 goto out; 10509 } 10510 10511 /* Check that new_ifindex isn't used yet. */ 10512 err = -EBUSY; 10513 if (new_ifindex && __dev_get_by_index(net, new_ifindex)) 10514 goto out; 10515 10516 /* 10517 * And now a mini version of register_netdevice unregister_netdevice. 10518 */ 10519 10520 /* If device is running close it first. */ 10521 dev_close(dev); 10522 10523 /* And unlink it from device chain */ 10524 unlist_netdevice(dev); 10525 10526 synchronize_net(); 10527 10528 /* Shutdown queueing discipline. */ 10529 dev_shutdown(dev); 10530 10531 /* Notify protocols, that we are about to destroy 10532 * this device. They should clean all the things. 10533 * 10534 * Note that dev->reg_state stays at NETREG_REGISTERED. 10535 * This is wanted because this way 8021q and macvlan know 10536 * the device is just moving and can keep their slaves up. 10537 */ 10538 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10539 rcu_barrier(); 10540 10541 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL); 10542 /* If there is an ifindex conflict assign a new one */ 10543 if (!new_ifindex) { 10544 if (__dev_get_by_index(net, dev->ifindex)) 10545 new_ifindex = dev_new_index(net); 10546 else 10547 new_ifindex = dev->ifindex; 10548 } 10549 10550 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid, 10551 new_ifindex); 10552 10553 /* 10554 * Flush the unicast and multicast chains 10555 */ 10556 dev_uc_flush(dev); 10557 dev_mc_flush(dev); 10558 10559 /* Send a netdev-removed uevent to the old namespace */ 10560 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); 10561 netdev_adjacent_del_links(dev); 10562 10563 /* Move per-net netdevice notifiers that are following the netdevice */ 10564 move_netdevice_notifiers_dev_net(dev, net); 10565 10566 /* Actually switch the network namespace */ 10567 dev_net_set(dev, net); 10568 dev->ifindex = new_ifindex; 10569 10570 /* Send a netdev-add uevent to the new namespace */ 10571 kobject_uevent(&dev->dev.kobj, KOBJ_ADD); 10572 netdev_adjacent_add_links(dev); 10573 10574 /* Fixup kobjects */ 10575 err = device_rename(&dev->dev, dev->name); 10576 WARN_ON(err); 10577 10578 /* Adapt owner in case owning user namespace of target network 10579 * namespace is different from the original one. 10580 */ 10581 err = netdev_change_owner(dev, net_old, net); 10582 WARN_ON(err); 10583 10584 /* Add the device back in the hashes */ 10585 list_netdevice(dev); 10586 10587 /* Notify protocols, that a new device appeared. */ 10588 call_netdevice_notifiers(NETDEV_REGISTER, dev); 10589 10590 /* 10591 * Prevent userspace races by waiting until the network 10592 * device is fully setup before sending notifications. 10593 */ 10594 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); 10595 10596 synchronize_net(); 10597 err = 0; 10598 out: 10599 return err; 10600 } 10601 EXPORT_SYMBOL_GPL(__dev_change_net_namespace); 10602 10603 static int dev_cpu_dead(unsigned int oldcpu) 10604 { 10605 struct sk_buff **list_skb; 10606 struct sk_buff *skb; 10607 unsigned int cpu; 10608 struct softnet_data *sd, *oldsd, *remsd = NULL; 10609 10610 local_irq_disable(); 10611 cpu = smp_processor_id(); 10612 sd = &per_cpu(softnet_data, cpu); 10613 oldsd = &per_cpu(softnet_data, oldcpu); 10614 10615 /* Find end of our completion_queue. */ 10616 list_skb = &sd->completion_queue; 10617 while (*list_skb) 10618 list_skb = &(*list_skb)->next; 10619 /* Append completion queue from offline CPU. */ 10620 *list_skb = oldsd->completion_queue; 10621 oldsd->completion_queue = NULL; 10622 10623 /* Append output queue from offline CPU. */ 10624 if (oldsd->output_queue) { 10625 *sd->output_queue_tailp = oldsd->output_queue; 10626 sd->output_queue_tailp = oldsd->output_queue_tailp; 10627 oldsd->output_queue = NULL; 10628 oldsd->output_queue_tailp = &oldsd->output_queue; 10629 } 10630 /* Append NAPI poll list from offline CPU, with one exception : 10631 * process_backlog() must be called by cpu owning percpu backlog. 10632 * We properly handle process_queue & input_pkt_queue later. 10633 */ 10634 while (!list_empty(&oldsd->poll_list)) { 10635 struct napi_struct *napi = list_first_entry(&oldsd->poll_list, 10636 struct napi_struct, 10637 poll_list); 10638 10639 list_del_init(&napi->poll_list); 10640 if (napi->poll == process_backlog) 10641 napi->state = 0; 10642 else 10643 ____napi_schedule(sd, napi); 10644 } 10645 10646 raise_softirq_irqoff(NET_TX_SOFTIRQ); 10647 local_irq_enable(); 10648 10649 #ifdef CONFIG_RPS 10650 remsd = oldsd->rps_ipi_list; 10651 oldsd->rps_ipi_list = NULL; 10652 #endif 10653 /* send out pending IPI's on offline CPU */ 10654 net_rps_send_ipi(remsd); 10655 10656 /* Process offline CPU's input_pkt_queue */ 10657 while ((skb = __skb_dequeue(&oldsd->process_queue))) { 10658 netif_rx_ni(skb); 10659 input_queue_head_incr(oldsd); 10660 } 10661 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { 10662 netif_rx_ni(skb); 10663 input_queue_head_incr(oldsd); 10664 } 10665 10666 return 0; 10667 } 10668 10669 /** 10670 * netdev_increment_features - increment feature set by one 10671 * @all: current feature set 10672 * @one: new feature set 10673 * @mask: mask feature set 10674 * 10675 * Computes a new feature set after adding a device with feature set 10676 * @one to the master device with current feature set @all. Will not 10677 * enable anything that is off in @mask. Returns the new feature set. 10678 */ 10679 netdev_features_t netdev_increment_features(netdev_features_t all, 10680 netdev_features_t one, netdev_features_t mask) 10681 { 10682 if (mask & NETIF_F_HW_CSUM) 10683 mask |= NETIF_F_CSUM_MASK; 10684 mask |= NETIF_F_VLAN_CHALLENGED; 10685 10686 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; 10687 all &= one | ~NETIF_F_ALL_FOR_ALL; 10688 10689 /* If one device supports hw checksumming, set for all. */ 10690 if (all & NETIF_F_HW_CSUM) 10691 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); 10692 10693 return all; 10694 } 10695 EXPORT_SYMBOL(netdev_increment_features); 10696 10697 static struct hlist_head * __net_init netdev_create_hash(void) 10698 { 10699 int i; 10700 struct hlist_head *hash; 10701 10702 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL); 10703 if (hash != NULL) 10704 for (i = 0; i < NETDEV_HASHENTRIES; i++) 10705 INIT_HLIST_HEAD(&hash[i]); 10706 10707 return hash; 10708 } 10709 10710 /* Initialize per network namespace state */ 10711 static int __net_init netdev_init(struct net *net) 10712 { 10713 BUILD_BUG_ON(GRO_HASH_BUCKETS > 10714 8 * sizeof_field(struct napi_struct, gro_bitmask)); 10715 10716 if (net != &init_net) 10717 INIT_LIST_HEAD(&net->dev_base_head); 10718 10719 net->dev_name_head = netdev_create_hash(); 10720 if (net->dev_name_head == NULL) 10721 goto err_name; 10722 10723 net->dev_index_head = netdev_create_hash(); 10724 if (net->dev_index_head == NULL) 10725 goto err_idx; 10726 10727 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain); 10728 10729 return 0; 10730 10731 err_idx: 10732 kfree(net->dev_name_head); 10733 err_name: 10734 return -ENOMEM; 10735 } 10736 10737 /** 10738 * netdev_drivername - network driver for the device 10739 * @dev: network device 10740 * 10741 * Determine network driver for device. 10742 */ 10743 const char *netdev_drivername(const struct net_device *dev) 10744 { 10745 const struct device_driver *driver; 10746 const struct device *parent; 10747 const char *empty = ""; 10748 10749 parent = dev->dev.parent; 10750 if (!parent) 10751 return empty; 10752 10753 driver = parent->driver; 10754 if (driver && driver->name) 10755 return driver->name; 10756 return empty; 10757 } 10758 10759 static void __netdev_printk(const char *level, const struct net_device *dev, 10760 struct va_format *vaf) 10761 { 10762 if (dev && dev->dev.parent) { 10763 dev_printk_emit(level[1] - '0', 10764 dev->dev.parent, 10765 "%s %s %s%s: %pV", 10766 dev_driver_string(dev->dev.parent), 10767 dev_name(dev->dev.parent), 10768 netdev_name(dev), netdev_reg_state(dev), 10769 vaf); 10770 } else if (dev) { 10771 printk("%s%s%s: %pV", 10772 level, netdev_name(dev), netdev_reg_state(dev), vaf); 10773 } else { 10774 printk("%s(NULL net_device): %pV", level, vaf); 10775 } 10776 } 10777 10778 void netdev_printk(const char *level, const struct net_device *dev, 10779 const char *format, ...) 10780 { 10781 struct va_format vaf; 10782 va_list args; 10783 10784 va_start(args, format); 10785 10786 vaf.fmt = format; 10787 vaf.va = &args; 10788 10789 __netdev_printk(level, dev, &vaf); 10790 10791 va_end(args); 10792 } 10793 EXPORT_SYMBOL(netdev_printk); 10794 10795 #define define_netdev_printk_level(func, level) \ 10796 void func(const struct net_device *dev, const char *fmt, ...) \ 10797 { \ 10798 struct va_format vaf; \ 10799 va_list args; \ 10800 \ 10801 va_start(args, fmt); \ 10802 \ 10803 vaf.fmt = fmt; \ 10804 vaf.va = &args; \ 10805 \ 10806 __netdev_printk(level, dev, &vaf); \ 10807 \ 10808 va_end(args); \ 10809 } \ 10810 EXPORT_SYMBOL(func); 10811 10812 define_netdev_printk_level(netdev_emerg, KERN_EMERG); 10813 define_netdev_printk_level(netdev_alert, KERN_ALERT); 10814 define_netdev_printk_level(netdev_crit, KERN_CRIT); 10815 define_netdev_printk_level(netdev_err, KERN_ERR); 10816 define_netdev_printk_level(netdev_warn, KERN_WARNING); 10817 define_netdev_printk_level(netdev_notice, KERN_NOTICE); 10818 define_netdev_printk_level(netdev_info, KERN_INFO); 10819 10820 static void __net_exit netdev_exit(struct net *net) 10821 { 10822 kfree(net->dev_name_head); 10823 kfree(net->dev_index_head); 10824 if (net != &init_net) 10825 WARN_ON_ONCE(!list_empty(&net->dev_base_head)); 10826 } 10827 10828 static struct pernet_operations __net_initdata netdev_net_ops = { 10829 .init = netdev_init, 10830 .exit = netdev_exit, 10831 }; 10832 10833 static void __net_exit default_device_exit(struct net *net) 10834 { 10835 struct net_device *dev, *aux; 10836 /* 10837 * Push all migratable network devices back to the 10838 * initial network namespace 10839 */ 10840 rtnl_lock(); 10841 for_each_netdev_safe(net, dev, aux) { 10842 int err; 10843 char fb_name[IFNAMSIZ]; 10844 10845 /* Ignore unmoveable devices (i.e. loopback) */ 10846 if (dev->features & NETIF_F_NETNS_LOCAL) 10847 continue; 10848 10849 /* Leave virtual devices for the generic cleanup */ 10850 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund) 10851 continue; 10852 10853 /* Push remaining network devices to init_net */ 10854 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); 10855 if (netdev_name_in_use(&init_net, fb_name)) 10856 snprintf(fb_name, IFNAMSIZ, "dev%%d"); 10857 err = dev_change_net_namespace(dev, &init_net, fb_name); 10858 if (err) { 10859 pr_emerg("%s: failed to move %s to init_net: %d\n", 10860 __func__, dev->name, err); 10861 BUG(); 10862 } 10863 } 10864 rtnl_unlock(); 10865 } 10866 10867 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list) 10868 { 10869 /* Return with the rtnl_lock held when there are no network 10870 * devices unregistering in any network namespace in net_list. 10871 */ 10872 struct net *net; 10873 bool unregistering; 10874 DEFINE_WAIT_FUNC(wait, woken_wake_function); 10875 10876 add_wait_queue(&netdev_unregistering_wq, &wait); 10877 for (;;) { 10878 unregistering = false; 10879 rtnl_lock(); 10880 list_for_each_entry(net, net_list, exit_list) { 10881 if (net->dev_unreg_count > 0) { 10882 unregistering = true; 10883 break; 10884 } 10885 } 10886 if (!unregistering) 10887 break; 10888 __rtnl_unlock(); 10889 10890 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); 10891 } 10892 remove_wait_queue(&netdev_unregistering_wq, &wait); 10893 } 10894 10895 static void __net_exit default_device_exit_batch(struct list_head *net_list) 10896 { 10897 /* At exit all network devices most be removed from a network 10898 * namespace. Do this in the reverse order of registration. 10899 * Do this across as many network namespaces as possible to 10900 * improve batching efficiency. 10901 */ 10902 struct net_device *dev; 10903 struct net *net; 10904 LIST_HEAD(dev_kill_list); 10905 10906 /* To prevent network device cleanup code from dereferencing 10907 * loopback devices or network devices that have been freed 10908 * wait here for all pending unregistrations to complete, 10909 * before unregistring the loopback device and allowing the 10910 * network namespace be freed. 10911 * 10912 * The netdev todo list containing all network devices 10913 * unregistrations that happen in default_device_exit_batch 10914 * will run in the rtnl_unlock() at the end of 10915 * default_device_exit_batch. 10916 */ 10917 rtnl_lock_unregistering(net_list); 10918 list_for_each_entry(net, net_list, exit_list) { 10919 for_each_netdev_reverse(net, dev) { 10920 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) 10921 dev->rtnl_link_ops->dellink(dev, &dev_kill_list); 10922 else 10923 unregister_netdevice_queue(dev, &dev_kill_list); 10924 } 10925 } 10926 unregister_netdevice_many(&dev_kill_list); 10927 rtnl_unlock(); 10928 } 10929 10930 static struct pernet_operations __net_initdata default_device_ops = { 10931 .exit = default_device_exit, 10932 .exit_batch = default_device_exit_batch, 10933 }; 10934 10935 /* 10936 * Initialize the DEV module. At boot time this walks the device list and 10937 * unhooks any devices that fail to initialise (normally hardware not 10938 * present) and leaves us with a valid list of present and active devices. 10939 * 10940 */ 10941 10942 /* 10943 * This is called single threaded during boot, so no need 10944 * to take the rtnl semaphore. 10945 */ 10946 static int __init net_dev_init(void) 10947 { 10948 int i, rc = -ENOMEM; 10949 10950 BUG_ON(!dev_boot_phase); 10951 10952 if (dev_proc_init()) 10953 goto out; 10954 10955 if (netdev_kobject_init()) 10956 goto out; 10957 10958 INIT_LIST_HEAD(&ptype_all); 10959 for (i = 0; i < PTYPE_HASH_SIZE; i++) 10960 INIT_LIST_HEAD(&ptype_base[i]); 10961 10962 if (register_pernet_subsys(&netdev_net_ops)) 10963 goto out; 10964 10965 /* 10966 * Initialise the packet receive queues. 10967 */ 10968 10969 for_each_possible_cpu(i) { 10970 struct work_struct *flush = per_cpu_ptr(&flush_works, i); 10971 struct softnet_data *sd = &per_cpu(softnet_data, i); 10972 10973 INIT_WORK(flush, flush_backlog); 10974 10975 skb_queue_head_init(&sd->input_pkt_queue); 10976 skb_queue_head_init(&sd->process_queue); 10977 #ifdef CONFIG_XFRM_OFFLOAD 10978 skb_queue_head_init(&sd->xfrm_backlog); 10979 #endif 10980 INIT_LIST_HEAD(&sd->poll_list); 10981 sd->output_queue_tailp = &sd->output_queue; 10982 #ifdef CONFIG_RPS 10983 INIT_CSD(&sd->csd, rps_trigger_softirq, sd); 10984 sd->cpu = i; 10985 #endif 10986 10987 init_gro_hash(&sd->backlog); 10988 sd->backlog.poll = process_backlog; 10989 sd->backlog.weight = weight_p; 10990 } 10991 10992 dev_boot_phase = 0; 10993 10994 /* The loopback device is special if any other network devices 10995 * is present in a network namespace the loopback device must 10996 * be present. Since we now dynamically allocate and free the 10997 * loopback device ensure this invariant is maintained by 10998 * keeping the loopback device as the first device on the 10999 * list of network devices. Ensuring the loopback devices 11000 * is the first device that appears and the last network device 11001 * that disappears. 11002 */ 11003 if (register_pernet_device(&loopback_net_ops)) 11004 goto out; 11005 11006 if (register_pernet_device(&default_device_ops)) 11007 goto out; 11008 11009 open_softirq(NET_TX_SOFTIRQ, net_tx_action); 11010 open_softirq(NET_RX_SOFTIRQ, net_rx_action); 11011 11012 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", 11013 NULL, dev_cpu_dead); 11014 WARN_ON(rc < 0); 11015 rc = 0; 11016 out: 11017 return rc; 11018 } 11019 11020 subsys_initcall(net_dev_init); 11021