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