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