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_ingress.h> 144 #include <linux/crash_dump.h> 145 #include <linux/sctp.h> 146 #include <net/udp_tunnel.h> 147 #include <linux/net_namespace.h> 148 #include <linux/indirect_call_wrapper.h> 149 #include <net/devlink.h> 150 #include <linux/pm_runtime.h> 151 #include <linux/prandom.h> 152 #include <linux/once_lite.h> 153 154 #include "net-sysfs.h" 155 156 #define MAX_GRO_SKBS 8 157 158 /* This should be increased if a protocol with a bigger head is added. */ 159 #define GRO_MAX_HEAD (MAX_HEADER + 128) 160 161 static DEFINE_SPINLOCK(ptype_lock); 162 static DEFINE_SPINLOCK(offload_lock); 163 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; 164 struct list_head ptype_all __read_mostly; /* Taps */ 165 static struct list_head offload_base __read_mostly; 166 167 static int netif_rx_internal(struct sk_buff *skb); 168 static int call_netdevice_notifiers_info(unsigned long val, 169 struct netdev_notifier_info *info); 170 static int call_netdevice_notifiers_extack(unsigned long val, 171 struct net_device *dev, 172 struct netlink_ext_ack *extack); 173 static struct napi_struct *napi_by_id(unsigned int napi_id); 174 175 /* 176 * The @dev_base_head list is protected by @dev_base_lock and the rtnl 177 * semaphore. 178 * 179 * Pure readers hold dev_base_lock for reading, or rcu_read_lock() 180 * 181 * Writers must hold the rtnl semaphore while they loop through the 182 * dev_base_head list, and hold dev_base_lock for writing when they do the 183 * actual updates. This allows pure readers to access the list even 184 * while a writer is preparing to update it. 185 * 186 * To put it another way, dev_base_lock is held for writing only to 187 * protect against pure readers; the rtnl semaphore provides the 188 * protection against other writers. 189 * 190 * See, for example usages, register_netdevice() and 191 * unregister_netdevice(), which must be called with the rtnl 192 * semaphore held. 193 */ 194 DEFINE_RWLOCK(dev_base_lock); 195 EXPORT_SYMBOL(dev_base_lock); 196 197 static DEFINE_MUTEX(ifalias_mutex); 198 199 /* protects napi_hash addition/deletion and napi_gen_id */ 200 static DEFINE_SPINLOCK(napi_hash_lock); 201 202 static unsigned int napi_gen_id = NR_CPUS; 203 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8); 204 205 static DECLARE_RWSEM(devnet_rename_sem); 206 207 static inline void dev_base_seq_inc(struct net *net) 208 { 209 while (++net->dev_base_seq == 0) 210 ; 211 } 212 213 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) 214 { 215 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ)); 216 217 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; 218 } 219 220 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) 221 { 222 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; 223 } 224 225 static inline void rps_lock(struct softnet_data *sd) 226 { 227 #ifdef CONFIG_RPS 228 spin_lock(&sd->input_pkt_queue.lock); 229 #endif 230 } 231 232 static inline void rps_unlock(struct softnet_data *sd) 233 { 234 #ifdef CONFIG_RPS 235 spin_unlock(&sd->input_pkt_queue.lock); 236 #endif 237 } 238 239 static struct netdev_name_node *netdev_name_node_alloc(struct net_device *dev, 240 const char *name) 241 { 242 struct netdev_name_node *name_node; 243 244 name_node = kmalloc(sizeof(*name_node), GFP_KERNEL); 245 if (!name_node) 246 return NULL; 247 INIT_HLIST_NODE(&name_node->hlist); 248 name_node->dev = dev; 249 name_node->name = name; 250 return name_node; 251 } 252 253 static struct netdev_name_node * 254 netdev_name_node_head_alloc(struct net_device *dev) 255 { 256 struct netdev_name_node *name_node; 257 258 name_node = netdev_name_node_alloc(dev, dev->name); 259 if (!name_node) 260 return NULL; 261 INIT_LIST_HEAD(&name_node->list); 262 return name_node; 263 } 264 265 static void netdev_name_node_free(struct netdev_name_node *name_node) 266 { 267 kfree(name_node); 268 } 269 270 static void netdev_name_node_add(struct net *net, 271 struct netdev_name_node *name_node) 272 { 273 hlist_add_head_rcu(&name_node->hlist, 274 dev_name_hash(net, name_node->name)); 275 } 276 277 static void netdev_name_node_del(struct netdev_name_node *name_node) 278 { 279 hlist_del_rcu(&name_node->hlist); 280 } 281 282 static struct netdev_name_node *netdev_name_node_lookup(struct net *net, 283 const char *name) 284 { 285 struct hlist_head *head = dev_name_hash(net, name); 286 struct netdev_name_node *name_node; 287 288 hlist_for_each_entry(name_node, head, hlist) 289 if (!strcmp(name_node->name, name)) 290 return name_node; 291 return NULL; 292 } 293 294 static struct netdev_name_node *netdev_name_node_lookup_rcu(struct net *net, 295 const char *name) 296 { 297 struct hlist_head *head = dev_name_hash(net, name); 298 struct netdev_name_node *name_node; 299 300 hlist_for_each_entry_rcu(name_node, head, hlist) 301 if (!strcmp(name_node->name, name)) 302 return name_node; 303 return NULL; 304 } 305 306 int netdev_name_node_alt_create(struct net_device *dev, const char *name) 307 { 308 struct netdev_name_node *name_node; 309 struct net *net = dev_net(dev); 310 311 name_node = netdev_name_node_lookup(net, name); 312 if (name_node) 313 return -EEXIST; 314 name_node = netdev_name_node_alloc(dev, name); 315 if (!name_node) 316 return -ENOMEM; 317 netdev_name_node_add(net, name_node); 318 /* The node that holds dev->name acts as a head of per-device list. */ 319 list_add_tail(&name_node->list, &dev->name_node->list); 320 321 return 0; 322 } 323 EXPORT_SYMBOL(netdev_name_node_alt_create); 324 325 static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node) 326 { 327 list_del(&name_node->list); 328 netdev_name_node_del(name_node); 329 kfree(name_node->name); 330 netdev_name_node_free(name_node); 331 } 332 333 int netdev_name_node_alt_destroy(struct net_device *dev, const char *name) 334 { 335 struct netdev_name_node *name_node; 336 struct net *net = dev_net(dev); 337 338 name_node = netdev_name_node_lookup(net, name); 339 if (!name_node) 340 return -ENOENT; 341 /* lookup might have found our primary name or a name belonging 342 * to another device. 343 */ 344 if (name_node == dev->name_node || name_node->dev != dev) 345 return -EINVAL; 346 347 __netdev_name_node_alt_destroy(name_node); 348 349 return 0; 350 } 351 EXPORT_SYMBOL(netdev_name_node_alt_destroy); 352 353 static void netdev_name_node_alt_flush(struct net_device *dev) 354 { 355 struct netdev_name_node *name_node, *tmp; 356 357 list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list) 358 __netdev_name_node_alt_destroy(name_node); 359 } 360 361 /* Device list insertion */ 362 static void list_netdevice(struct net_device *dev) 363 { 364 struct net *net = dev_net(dev); 365 366 ASSERT_RTNL(); 367 368 write_lock_bh(&dev_base_lock); 369 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); 370 netdev_name_node_add(net, dev->name_node); 371 hlist_add_head_rcu(&dev->index_hlist, 372 dev_index_hash(net, dev->ifindex)); 373 write_unlock_bh(&dev_base_lock); 374 375 dev_base_seq_inc(net); 376 } 377 378 /* Device list removal 379 * caller must respect a RCU grace period before freeing/reusing dev 380 */ 381 static void unlist_netdevice(struct net_device *dev) 382 { 383 ASSERT_RTNL(); 384 385 /* Unlink dev from the device chain */ 386 write_lock_bh(&dev_base_lock); 387 list_del_rcu(&dev->dev_list); 388 netdev_name_node_del(dev->name_node); 389 hlist_del_rcu(&dev->index_hlist); 390 write_unlock_bh(&dev_base_lock); 391 392 dev_base_seq_inc(dev_net(dev)); 393 } 394 395 /* 396 * Our notifier list 397 */ 398 399 static RAW_NOTIFIER_HEAD(netdev_chain); 400 401 /* 402 * Device drivers call our routines to queue packets here. We empty the 403 * queue in the local softnet handler. 404 */ 405 406 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); 407 EXPORT_PER_CPU_SYMBOL(softnet_data); 408 409 #ifdef CONFIG_LOCKDEP 410 /* 411 * register_netdevice() inits txq->_xmit_lock and sets lockdep class 412 * according to dev->type 413 */ 414 static const unsigned short netdev_lock_type[] = { 415 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, 416 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, 417 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, 418 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, 419 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, 420 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, 421 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, 422 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, 423 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, 424 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, 425 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, 426 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, 427 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, 428 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, 429 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; 430 431 static const char *const netdev_lock_name[] = { 432 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", 433 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", 434 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", 435 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", 436 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", 437 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", 438 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", 439 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", 440 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", 441 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", 442 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", 443 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", 444 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", 445 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", 446 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; 447 448 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; 449 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; 450 451 static inline unsigned short netdev_lock_pos(unsigned short dev_type) 452 { 453 int i; 454 455 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) 456 if (netdev_lock_type[i] == dev_type) 457 return i; 458 /* the last key is used by default */ 459 return ARRAY_SIZE(netdev_lock_type) - 1; 460 } 461 462 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 463 unsigned short dev_type) 464 { 465 int i; 466 467 i = netdev_lock_pos(dev_type); 468 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], 469 netdev_lock_name[i]); 470 } 471 472 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 473 { 474 int i; 475 476 i = netdev_lock_pos(dev->type); 477 lockdep_set_class_and_name(&dev->addr_list_lock, 478 &netdev_addr_lock_key[i], 479 netdev_lock_name[i]); 480 } 481 #else 482 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 483 unsigned short dev_type) 484 { 485 } 486 487 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 488 { 489 } 490 #endif 491 492 /******************************************************************************* 493 * 494 * Protocol management and registration routines 495 * 496 *******************************************************************************/ 497 498 499 /* 500 * Add a protocol ID to the list. Now that the input handler is 501 * smarter we can dispense with all the messy stuff that used to be 502 * here. 503 * 504 * BEWARE!!! Protocol handlers, mangling input packets, 505 * MUST BE last in hash buckets and checking protocol handlers 506 * MUST start from promiscuous ptype_all chain in net_bh. 507 * It is true now, do not change it. 508 * Explanation follows: if protocol handler, mangling packet, will 509 * be the first on list, it is not able to sense, that packet 510 * is cloned and should be copied-on-write, so that it will 511 * change it and subsequent readers will get broken packet. 512 * --ANK (980803) 513 */ 514 515 static inline struct list_head *ptype_head(const struct packet_type *pt) 516 { 517 if (pt->type == htons(ETH_P_ALL)) 518 return pt->dev ? &pt->dev->ptype_all : &ptype_all; 519 else 520 return pt->dev ? &pt->dev->ptype_specific : 521 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; 522 } 523 524 /** 525 * dev_add_pack - add packet handler 526 * @pt: packet type declaration 527 * 528 * Add a protocol handler to the networking stack. The passed &packet_type 529 * is linked into kernel lists and may not be freed until it has been 530 * removed from the kernel lists. 531 * 532 * This call does not sleep therefore it can not 533 * guarantee all CPU's that are in middle of receiving packets 534 * will see the new packet type (until the next received packet). 535 */ 536 537 void dev_add_pack(struct packet_type *pt) 538 { 539 struct list_head *head = ptype_head(pt); 540 541 spin_lock(&ptype_lock); 542 list_add_rcu(&pt->list, head); 543 spin_unlock(&ptype_lock); 544 } 545 EXPORT_SYMBOL(dev_add_pack); 546 547 /** 548 * __dev_remove_pack - remove packet handler 549 * @pt: packet type declaration 550 * 551 * Remove a protocol handler that was previously added to the kernel 552 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 553 * from the kernel lists and can be freed or reused once this function 554 * returns. 555 * 556 * The packet type might still be in use by receivers 557 * and must not be freed until after all the CPU's have gone 558 * through a quiescent state. 559 */ 560 void __dev_remove_pack(struct packet_type *pt) 561 { 562 struct list_head *head = ptype_head(pt); 563 struct packet_type *pt1; 564 565 spin_lock(&ptype_lock); 566 567 list_for_each_entry(pt1, head, list) { 568 if (pt == pt1) { 569 list_del_rcu(&pt->list); 570 goto out; 571 } 572 } 573 574 pr_warn("dev_remove_pack: %p not found\n", pt); 575 out: 576 spin_unlock(&ptype_lock); 577 } 578 EXPORT_SYMBOL(__dev_remove_pack); 579 580 /** 581 * dev_remove_pack - remove packet handler 582 * @pt: packet type declaration 583 * 584 * Remove a protocol handler that was previously added to the kernel 585 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 586 * from the kernel lists and can be freed or reused once this function 587 * returns. 588 * 589 * This call sleeps to guarantee that no CPU is looking at the packet 590 * type after return. 591 */ 592 void dev_remove_pack(struct packet_type *pt) 593 { 594 __dev_remove_pack(pt); 595 596 synchronize_net(); 597 } 598 EXPORT_SYMBOL(dev_remove_pack); 599 600 601 /** 602 * dev_add_offload - register offload handlers 603 * @po: protocol offload declaration 604 * 605 * Add protocol offload handlers to the networking stack. The passed 606 * &proto_offload is linked into kernel lists and may not be freed until 607 * it has been removed from the kernel lists. 608 * 609 * This call does not sleep therefore it can not 610 * guarantee all CPU's that are in middle of receiving packets 611 * will see the new offload handlers (until the next received packet). 612 */ 613 void dev_add_offload(struct packet_offload *po) 614 { 615 struct packet_offload *elem; 616 617 spin_lock(&offload_lock); 618 list_for_each_entry(elem, &offload_base, list) { 619 if (po->priority < elem->priority) 620 break; 621 } 622 list_add_rcu(&po->list, elem->list.prev); 623 spin_unlock(&offload_lock); 624 } 625 EXPORT_SYMBOL(dev_add_offload); 626 627 /** 628 * __dev_remove_offload - remove offload handler 629 * @po: packet offload declaration 630 * 631 * Remove a protocol offload handler that was previously added to the 632 * kernel offload handlers by dev_add_offload(). The passed &offload_type 633 * is removed from the kernel lists and can be freed or reused once this 634 * function returns. 635 * 636 * The packet type might still be in use by receivers 637 * and must not be freed until after all the CPU's have gone 638 * through a quiescent state. 639 */ 640 static void __dev_remove_offload(struct packet_offload *po) 641 { 642 struct list_head *head = &offload_base; 643 struct packet_offload *po1; 644 645 spin_lock(&offload_lock); 646 647 list_for_each_entry(po1, head, list) { 648 if (po == po1) { 649 list_del_rcu(&po->list); 650 goto out; 651 } 652 } 653 654 pr_warn("dev_remove_offload: %p not found\n", po); 655 out: 656 spin_unlock(&offload_lock); 657 } 658 659 /** 660 * dev_remove_offload - remove packet offload handler 661 * @po: packet offload declaration 662 * 663 * Remove a packet offload handler that was previously added to the kernel 664 * offload handlers by dev_add_offload(). The passed &offload_type is 665 * removed from the kernel lists and can be freed or reused once this 666 * function returns. 667 * 668 * This call sleeps to guarantee that no CPU is looking at the packet 669 * type after return. 670 */ 671 void dev_remove_offload(struct packet_offload *po) 672 { 673 __dev_remove_offload(po); 674 675 synchronize_net(); 676 } 677 EXPORT_SYMBOL(dev_remove_offload); 678 679 /******************************************************************************* 680 * 681 * Device Interface Subroutines 682 * 683 *******************************************************************************/ 684 685 /** 686 * dev_get_iflink - get 'iflink' value of a interface 687 * @dev: targeted interface 688 * 689 * Indicates the ifindex the interface is linked to. 690 * Physical interfaces have the same 'ifindex' and 'iflink' values. 691 */ 692 693 int dev_get_iflink(const struct net_device *dev) 694 { 695 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink) 696 return dev->netdev_ops->ndo_get_iflink(dev); 697 698 return dev->ifindex; 699 } 700 EXPORT_SYMBOL(dev_get_iflink); 701 702 /** 703 * dev_fill_metadata_dst - Retrieve tunnel egress information. 704 * @dev: targeted interface 705 * @skb: The packet. 706 * 707 * For better visibility of tunnel traffic OVS needs to retrieve 708 * egress tunnel information for a packet. Following API allows 709 * user to get this info. 710 */ 711 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) 712 { 713 struct ip_tunnel_info *info; 714 715 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst) 716 return -EINVAL; 717 718 info = skb_tunnel_info_unclone(skb); 719 if (!info) 720 return -ENOMEM; 721 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX))) 722 return -EINVAL; 723 724 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb); 725 } 726 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst); 727 728 static struct net_device_path *dev_fwd_path(struct net_device_path_stack *stack) 729 { 730 int k = stack->num_paths++; 731 732 if (WARN_ON_ONCE(k >= NET_DEVICE_PATH_STACK_MAX)) 733 return NULL; 734 735 return &stack->path[k]; 736 } 737 738 int dev_fill_forward_path(const struct net_device *dev, const u8 *daddr, 739 struct net_device_path_stack *stack) 740 { 741 const struct net_device *last_dev; 742 struct net_device_path_ctx ctx = { 743 .dev = dev, 744 .daddr = daddr, 745 }; 746 struct net_device_path *path; 747 int ret = 0; 748 749 stack->num_paths = 0; 750 while (ctx.dev && ctx.dev->netdev_ops->ndo_fill_forward_path) { 751 last_dev = ctx.dev; 752 path = dev_fwd_path(stack); 753 if (!path) 754 return -1; 755 756 memset(path, 0, sizeof(struct net_device_path)); 757 ret = ctx.dev->netdev_ops->ndo_fill_forward_path(&ctx, path); 758 if (ret < 0) 759 return -1; 760 761 if (WARN_ON_ONCE(last_dev == ctx.dev)) 762 return -1; 763 } 764 path = dev_fwd_path(stack); 765 if (!path) 766 return -1; 767 path->type = DEV_PATH_ETHERNET; 768 path->dev = ctx.dev; 769 770 return ret; 771 } 772 EXPORT_SYMBOL_GPL(dev_fill_forward_path); 773 774 /** 775 * __dev_get_by_name - find a device by its name 776 * @net: the applicable net namespace 777 * @name: name to find 778 * 779 * Find an interface by name. Must be called under RTNL semaphore 780 * or @dev_base_lock. If the name is found a pointer to the device 781 * is returned. If the name is not found then %NULL is returned. The 782 * reference counters are not incremented so the caller must be 783 * careful with locks. 784 */ 785 786 struct net_device *__dev_get_by_name(struct net *net, const char *name) 787 { 788 struct netdev_name_node *node_name; 789 790 node_name = netdev_name_node_lookup(net, name); 791 return node_name ? node_name->dev : NULL; 792 } 793 EXPORT_SYMBOL(__dev_get_by_name); 794 795 /** 796 * dev_get_by_name_rcu - find a device by its name 797 * @net: the applicable net namespace 798 * @name: name to find 799 * 800 * Find an interface by name. 801 * If the name is found a pointer to the device is returned. 802 * If the name is not found then %NULL is returned. 803 * The reference counters are not incremented so the caller must be 804 * careful with locks. The caller must hold RCU lock. 805 */ 806 807 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) 808 { 809 struct netdev_name_node *node_name; 810 811 node_name = netdev_name_node_lookup_rcu(net, name); 812 return node_name ? node_name->dev : NULL; 813 } 814 EXPORT_SYMBOL(dev_get_by_name_rcu); 815 816 /** 817 * dev_get_by_name - find a device by its name 818 * @net: the applicable net namespace 819 * @name: name to find 820 * 821 * Find an interface by name. This can be called from any 822 * context and does its own locking. The returned handle has 823 * the usage count incremented and the caller must use dev_put() to 824 * release it when it is no longer needed. %NULL is returned if no 825 * matching device is found. 826 */ 827 828 struct net_device *dev_get_by_name(struct net *net, const char *name) 829 { 830 struct net_device *dev; 831 832 rcu_read_lock(); 833 dev = dev_get_by_name_rcu(net, name); 834 dev_hold(dev); 835 rcu_read_unlock(); 836 return dev; 837 } 838 EXPORT_SYMBOL(dev_get_by_name); 839 840 /** 841 * __dev_get_by_index - find a device by its ifindex 842 * @net: the applicable net namespace 843 * @ifindex: index of device 844 * 845 * Search for an interface by index. Returns %NULL if the device 846 * is not found or a pointer to the device. The device has not 847 * had its reference counter increased so the caller must be careful 848 * about locking. The caller must hold either the RTNL semaphore 849 * or @dev_base_lock. 850 */ 851 852 struct net_device *__dev_get_by_index(struct net *net, int ifindex) 853 { 854 struct net_device *dev; 855 struct hlist_head *head = dev_index_hash(net, ifindex); 856 857 hlist_for_each_entry(dev, head, index_hlist) 858 if (dev->ifindex == ifindex) 859 return dev; 860 861 return NULL; 862 } 863 EXPORT_SYMBOL(__dev_get_by_index); 864 865 /** 866 * dev_get_by_index_rcu - find a device by its ifindex 867 * @net: the applicable net namespace 868 * @ifindex: index of device 869 * 870 * Search for an interface by index. Returns %NULL if the device 871 * is not found or a pointer to the device. The device has not 872 * had its reference counter increased so the caller must be careful 873 * about locking. The caller must hold RCU lock. 874 */ 875 876 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) 877 { 878 struct net_device *dev; 879 struct hlist_head *head = dev_index_hash(net, ifindex); 880 881 hlist_for_each_entry_rcu(dev, head, index_hlist) 882 if (dev->ifindex == ifindex) 883 return dev; 884 885 return NULL; 886 } 887 EXPORT_SYMBOL(dev_get_by_index_rcu); 888 889 890 /** 891 * dev_get_by_index - find a device by its ifindex 892 * @net: the applicable net namespace 893 * @ifindex: index of device 894 * 895 * Search for an interface by index. Returns NULL if the device 896 * is not found or a pointer to the device. The device returned has 897 * had a reference added and the pointer is safe until the user calls 898 * dev_put to indicate they have finished with it. 899 */ 900 901 struct net_device *dev_get_by_index(struct net *net, int ifindex) 902 { 903 struct net_device *dev; 904 905 rcu_read_lock(); 906 dev = dev_get_by_index_rcu(net, ifindex); 907 dev_hold(dev); 908 rcu_read_unlock(); 909 return dev; 910 } 911 EXPORT_SYMBOL(dev_get_by_index); 912 913 /** 914 * dev_get_by_napi_id - find a device by napi_id 915 * @napi_id: ID of the NAPI struct 916 * 917 * Search for an interface by NAPI ID. Returns %NULL if the device 918 * is not found or a pointer to the device. The device has not had 919 * its reference counter increased so the caller must be careful 920 * about locking. The caller must hold RCU lock. 921 */ 922 923 struct net_device *dev_get_by_napi_id(unsigned int napi_id) 924 { 925 struct napi_struct *napi; 926 927 WARN_ON_ONCE(!rcu_read_lock_held()); 928 929 if (napi_id < MIN_NAPI_ID) 930 return NULL; 931 932 napi = napi_by_id(napi_id); 933 934 return napi ? napi->dev : NULL; 935 } 936 EXPORT_SYMBOL(dev_get_by_napi_id); 937 938 /** 939 * netdev_get_name - get a netdevice name, knowing its ifindex. 940 * @net: network namespace 941 * @name: a pointer to the buffer where the name will be stored. 942 * @ifindex: the ifindex of the interface to get the name from. 943 */ 944 int netdev_get_name(struct net *net, char *name, int ifindex) 945 { 946 struct net_device *dev; 947 int ret; 948 949 down_read(&devnet_rename_sem); 950 rcu_read_lock(); 951 952 dev = dev_get_by_index_rcu(net, ifindex); 953 if (!dev) { 954 ret = -ENODEV; 955 goto out; 956 } 957 958 strcpy(name, dev->name); 959 960 ret = 0; 961 out: 962 rcu_read_unlock(); 963 up_read(&devnet_rename_sem); 964 return ret; 965 } 966 967 /** 968 * dev_getbyhwaddr_rcu - find a device by its hardware address 969 * @net: the applicable net namespace 970 * @type: media type of device 971 * @ha: hardware address 972 * 973 * Search for an interface by MAC address. Returns NULL if the device 974 * is not found or a pointer to the device. 975 * The caller must hold RCU or RTNL. 976 * The returned device has not had its ref count increased 977 * and the caller must therefore be careful about locking 978 * 979 */ 980 981 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, 982 const char *ha) 983 { 984 struct net_device *dev; 985 986 for_each_netdev_rcu(net, dev) 987 if (dev->type == type && 988 !memcmp(dev->dev_addr, ha, dev->addr_len)) 989 return dev; 990 991 return NULL; 992 } 993 EXPORT_SYMBOL(dev_getbyhwaddr_rcu); 994 995 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) 996 { 997 struct net_device *dev, *ret = NULL; 998 999 rcu_read_lock(); 1000 for_each_netdev_rcu(net, dev) 1001 if (dev->type == type) { 1002 dev_hold(dev); 1003 ret = dev; 1004 break; 1005 } 1006 rcu_read_unlock(); 1007 return ret; 1008 } 1009 EXPORT_SYMBOL(dev_getfirstbyhwtype); 1010 1011 /** 1012 * __dev_get_by_flags - find any device with given flags 1013 * @net: the applicable net namespace 1014 * @if_flags: IFF_* values 1015 * @mask: bitmask of bits in if_flags to check 1016 * 1017 * Search for any interface with the given flags. Returns NULL if a device 1018 * is not found or a pointer to the device. Must be called inside 1019 * rtnl_lock(), and result refcount is unchanged. 1020 */ 1021 1022 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags, 1023 unsigned short mask) 1024 { 1025 struct net_device *dev, *ret; 1026 1027 ASSERT_RTNL(); 1028 1029 ret = NULL; 1030 for_each_netdev(net, dev) { 1031 if (((dev->flags ^ if_flags) & mask) == 0) { 1032 ret = dev; 1033 break; 1034 } 1035 } 1036 return ret; 1037 } 1038 EXPORT_SYMBOL(__dev_get_by_flags); 1039 1040 /** 1041 * dev_valid_name - check if name is okay for network device 1042 * @name: name string 1043 * 1044 * Network device names need to be valid file names to 1045 * allow sysfs to work. We also disallow any kind of 1046 * whitespace. 1047 */ 1048 bool dev_valid_name(const char *name) 1049 { 1050 if (*name == '\0') 1051 return false; 1052 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ) 1053 return false; 1054 if (!strcmp(name, ".") || !strcmp(name, "..")) 1055 return false; 1056 1057 while (*name) { 1058 if (*name == '/' || *name == ':' || isspace(*name)) 1059 return false; 1060 name++; 1061 } 1062 return true; 1063 } 1064 EXPORT_SYMBOL(dev_valid_name); 1065 1066 /** 1067 * __dev_alloc_name - allocate a name for a device 1068 * @net: network namespace to allocate the device name in 1069 * @name: name format string 1070 * @buf: scratch buffer and result name string 1071 * 1072 * Passed a format string - eg "lt%d" it will try and find a suitable 1073 * id. It scans list of devices to build up a free map, then chooses 1074 * the first empty slot. The caller must hold the dev_base or rtnl lock 1075 * while allocating the name and adding the device in order to avoid 1076 * duplicates. 1077 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1078 * Returns the number of the unit assigned or a negative errno code. 1079 */ 1080 1081 static int __dev_alloc_name(struct net *net, const char *name, char *buf) 1082 { 1083 int i = 0; 1084 const char *p; 1085 const int max_netdevices = 8*PAGE_SIZE; 1086 unsigned long *inuse; 1087 struct net_device *d; 1088 1089 if (!dev_valid_name(name)) 1090 return -EINVAL; 1091 1092 p = strchr(name, '%'); 1093 if (p) { 1094 /* 1095 * Verify the string as this thing may have come from 1096 * the user. There must be either one "%d" and no other "%" 1097 * characters. 1098 */ 1099 if (p[1] != 'd' || strchr(p + 2, '%')) 1100 return -EINVAL; 1101 1102 /* Use one page as a bit array of possible slots */ 1103 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); 1104 if (!inuse) 1105 return -ENOMEM; 1106 1107 for_each_netdev(net, d) { 1108 struct netdev_name_node *name_node; 1109 list_for_each_entry(name_node, &d->name_node->list, list) { 1110 if (!sscanf(name_node->name, name, &i)) 1111 continue; 1112 if (i < 0 || i >= max_netdevices) 1113 continue; 1114 1115 /* avoid cases where sscanf is not exact inverse of printf */ 1116 snprintf(buf, IFNAMSIZ, name, i); 1117 if (!strncmp(buf, name_node->name, IFNAMSIZ)) 1118 set_bit(i, inuse); 1119 } 1120 if (!sscanf(d->name, name, &i)) 1121 continue; 1122 if (i < 0 || i >= max_netdevices) 1123 continue; 1124 1125 /* avoid cases where sscanf is not exact inverse of printf */ 1126 snprintf(buf, IFNAMSIZ, name, i); 1127 if (!strncmp(buf, d->name, IFNAMSIZ)) 1128 set_bit(i, inuse); 1129 } 1130 1131 i = find_first_zero_bit(inuse, max_netdevices); 1132 free_page((unsigned long) inuse); 1133 } 1134 1135 snprintf(buf, IFNAMSIZ, name, i); 1136 if (!__dev_get_by_name(net, buf)) 1137 return i; 1138 1139 /* It is possible to run out of possible slots 1140 * when the name is long and there isn't enough space left 1141 * for the digits, or if all bits are used. 1142 */ 1143 return -ENFILE; 1144 } 1145 1146 static int dev_alloc_name_ns(struct net *net, 1147 struct net_device *dev, 1148 const char *name) 1149 { 1150 char buf[IFNAMSIZ]; 1151 int ret; 1152 1153 BUG_ON(!net); 1154 ret = __dev_alloc_name(net, name, buf); 1155 if (ret >= 0) 1156 strlcpy(dev->name, buf, IFNAMSIZ); 1157 return ret; 1158 } 1159 1160 /** 1161 * dev_alloc_name - allocate a name for a device 1162 * @dev: device 1163 * @name: name format string 1164 * 1165 * Passed a format string - eg "lt%d" it will try and find a suitable 1166 * id. It scans list of devices to build up a free map, then chooses 1167 * the first empty slot. The caller must hold the dev_base or rtnl lock 1168 * while allocating the name and adding the device in order to avoid 1169 * duplicates. 1170 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1171 * Returns the number of the unit assigned or a negative errno code. 1172 */ 1173 1174 int dev_alloc_name(struct net_device *dev, const char *name) 1175 { 1176 return dev_alloc_name_ns(dev_net(dev), dev, name); 1177 } 1178 EXPORT_SYMBOL(dev_alloc_name); 1179 1180 static int dev_get_valid_name(struct net *net, struct net_device *dev, 1181 const char *name) 1182 { 1183 BUG_ON(!net); 1184 1185 if (!dev_valid_name(name)) 1186 return -EINVAL; 1187 1188 if (strchr(name, '%')) 1189 return dev_alloc_name_ns(net, dev, name); 1190 else if (__dev_get_by_name(net, name)) 1191 return -EEXIST; 1192 else if (dev->name != name) 1193 strlcpy(dev->name, name, IFNAMSIZ); 1194 1195 return 0; 1196 } 1197 1198 /** 1199 * dev_change_name - change name of a device 1200 * @dev: device 1201 * @newname: name (or format string) must be at least IFNAMSIZ 1202 * 1203 * Change name of a device, can pass format strings "eth%d". 1204 * for wildcarding. 1205 */ 1206 int dev_change_name(struct net_device *dev, const char *newname) 1207 { 1208 unsigned char old_assign_type; 1209 char oldname[IFNAMSIZ]; 1210 int err = 0; 1211 int ret; 1212 struct net *net; 1213 1214 ASSERT_RTNL(); 1215 BUG_ON(!dev_net(dev)); 1216 1217 net = dev_net(dev); 1218 1219 /* Some auto-enslaved devices e.g. failover slaves are 1220 * special, as userspace might rename the device after 1221 * the interface had been brought up and running since 1222 * the point kernel initiated auto-enslavement. Allow 1223 * live name change even when these slave devices are 1224 * up and running. 1225 * 1226 * Typically, users of these auto-enslaving devices 1227 * don't actually care about slave name change, as 1228 * they are supposed to operate on master interface 1229 * directly. 1230 */ 1231 if (dev->flags & IFF_UP && 1232 likely(!(dev->priv_flags & IFF_LIVE_RENAME_OK))) 1233 return -EBUSY; 1234 1235 down_write(&devnet_rename_sem); 1236 1237 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { 1238 up_write(&devnet_rename_sem); 1239 return 0; 1240 } 1241 1242 memcpy(oldname, dev->name, IFNAMSIZ); 1243 1244 err = dev_get_valid_name(net, dev, newname); 1245 if (err < 0) { 1246 up_write(&devnet_rename_sem); 1247 return err; 1248 } 1249 1250 if (oldname[0] && !strchr(oldname, '%')) 1251 netdev_info(dev, "renamed from %s\n", oldname); 1252 1253 old_assign_type = dev->name_assign_type; 1254 dev->name_assign_type = NET_NAME_RENAMED; 1255 1256 rollback: 1257 ret = device_rename(&dev->dev, dev->name); 1258 if (ret) { 1259 memcpy(dev->name, oldname, IFNAMSIZ); 1260 dev->name_assign_type = old_assign_type; 1261 up_write(&devnet_rename_sem); 1262 return ret; 1263 } 1264 1265 up_write(&devnet_rename_sem); 1266 1267 netdev_adjacent_rename_links(dev, oldname); 1268 1269 write_lock_bh(&dev_base_lock); 1270 netdev_name_node_del(dev->name_node); 1271 write_unlock_bh(&dev_base_lock); 1272 1273 synchronize_rcu(); 1274 1275 write_lock_bh(&dev_base_lock); 1276 netdev_name_node_add(net, dev->name_node); 1277 write_unlock_bh(&dev_base_lock); 1278 1279 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); 1280 ret = notifier_to_errno(ret); 1281 1282 if (ret) { 1283 /* err >= 0 after dev_alloc_name() or stores the first errno */ 1284 if (err >= 0) { 1285 err = ret; 1286 down_write(&devnet_rename_sem); 1287 memcpy(dev->name, oldname, IFNAMSIZ); 1288 memcpy(oldname, newname, IFNAMSIZ); 1289 dev->name_assign_type = old_assign_type; 1290 old_assign_type = NET_NAME_RENAMED; 1291 goto rollback; 1292 } else { 1293 pr_err("%s: name change rollback failed: %d\n", 1294 dev->name, ret); 1295 } 1296 } 1297 1298 return err; 1299 } 1300 1301 /** 1302 * dev_set_alias - change ifalias of a device 1303 * @dev: device 1304 * @alias: name up to IFALIASZ 1305 * @len: limit of bytes to copy from info 1306 * 1307 * Set ifalias for a device, 1308 */ 1309 int dev_set_alias(struct net_device *dev, const char *alias, size_t len) 1310 { 1311 struct dev_ifalias *new_alias = NULL; 1312 1313 if (len >= IFALIASZ) 1314 return -EINVAL; 1315 1316 if (len) { 1317 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL); 1318 if (!new_alias) 1319 return -ENOMEM; 1320 1321 memcpy(new_alias->ifalias, alias, len); 1322 new_alias->ifalias[len] = 0; 1323 } 1324 1325 mutex_lock(&ifalias_mutex); 1326 new_alias = rcu_replace_pointer(dev->ifalias, new_alias, 1327 mutex_is_locked(&ifalias_mutex)); 1328 mutex_unlock(&ifalias_mutex); 1329 1330 if (new_alias) 1331 kfree_rcu(new_alias, rcuhead); 1332 1333 return len; 1334 } 1335 EXPORT_SYMBOL(dev_set_alias); 1336 1337 /** 1338 * dev_get_alias - get ifalias of a device 1339 * @dev: device 1340 * @name: buffer to store name of ifalias 1341 * @len: size of buffer 1342 * 1343 * get ifalias for a device. Caller must make sure dev cannot go 1344 * away, e.g. rcu read lock or own a reference count to device. 1345 */ 1346 int dev_get_alias(const struct net_device *dev, char *name, size_t len) 1347 { 1348 const struct dev_ifalias *alias; 1349 int ret = 0; 1350 1351 rcu_read_lock(); 1352 alias = rcu_dereference(dev->ifalias); 1353 if (alias) 1354 ret = snprintf(name, len, "%s", alias->ifalias); 1355 rcu_read_unlock(); 1356 1357 return ret; 1358 } 1359 1360 /** 1361 * netdev_features_change - device changes features 1362 * @dev: device to cause notification 1363 * 1364 * Called to indicate a device has changed features. 1365 */ 1366 void netdev_features_change(struct net_device *dev) 1367 { 1368 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); 1369 } 1370 EXPORT_SYMBOL(netdev_features_change); 1371 1372 /** 1373 * netdev_state_change - device changes state 1374 * @dev: device to cause notification 1375 * 1376 * Called to indicate a device has changed state. This function calls 1377 * the notifier chains for netdev_chain and sends a NEWLINK message 1378 * to the routing socket. 1379 */ 1380 void netdev_state_change(struct net_device *dev) 1381 { 1382 if (dev->flags & IFF_UP) { 1383 struct netdev_notifier_change_info change_info = { 1384 .info.dev = dev, 1385 }; 1386 1387 call_netdevice_notifiers_info(NETDEV_CHANGE, 1388 &change_info.info); 1389 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL); 1390 } 1391 } 1392 EXPORT_SYMBOL(netdev_state_change); 1393 1394 /** 1395 * __netdev_notify_peers - notify network peers about existence of @dev, 1396 * to be called when rtnl lock is already held. 1397 * @dev: network device 1398 * 1399 * Generate traffic such that interested network peers are aware of 1400 * @dev, such as by generating a gratuitous ARP. This may be used when 1401 * a device wants to inform the rest of the network about some sort of 1402 * reconfiguration such as a failover event or virtual machine 1403 * migration. 1404 */ 1405 void __netdev_notify_peers(struct net_device *dev) 1406 { 1407 ASSERT_RTNL(); 1408 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); 1409 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev); 1410 } 1411 EXPORT_SYMBOL(__netdev_notify_peers); 1412 1413 /** 1414 * netdev_notify_peers - notify network peers about existence of @dev 1415 * @dev: network device 1416 * 1417 * Generate traffic such that interested network peers are aware of 1418 * @dev, such as by generating a gratuitous ARP. This may be used when 1419 * a device wants to inform the rest of the network about some sort of 1420 * reconfiguration such as a failover event or virtual machine 1421 * migration. 1422 */ 1423 void netdev_notify_peers(struct net_device *dev) 1424 { 1425 rtnl_lock(); 1426 __netdev_notify_peers(dev); 1427 rtnl_unlock(); 1428 } 1429 EXPORT_SYMBOL(netdev_notify_peers); 1430 1431 static int napi_threaded_poll(void *data); 1432 1433 static int napi_kthread_create(struct napi_struct *n) 1434 { 1435 int err = 0; 1436 1437 /* Create and wake up the kthread once to put it in 1438 * TASK_INTERRUPTIBLE mode to avoid the blocked task 1439 * warning and work with loadavg. 1440 */ 1441 n->thread = kthread_run(napi_threaded_poll, n, "napi/%s-%d", 1442 n->dev->name, n->napi_id); 1443 if (IS_ERR(n->thread)) { 1444 err = PTR_ERR(n->thread); 1445 pr_err("kthread_run failed with err %d\n", err); 1446 n->thread = NULL; 1447 } 1448 1449 return err; 1450 } 1451 1452 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1453 { 1454 const struct net_device_ops *ops = dev->netdev_ops; 1455 int ret; 1456 1457 ASSERT_RTNL(); 1458 1459 if (!netif_device_present(dev)) { 1460 /* may be detached because parent is runtime-suspended */ 1461 if (dev->dev.parent) 1462 pm_runtime_resume(dev->dev.parent); 1463 if (!netif_device_present(dev)) 1464 return -ENODEV; 1465 } 1466 1467 /* Block netpoll from trying to do any rx path servicing. 1468 * If we don't do this there is a chance ndo_poll_controller 1469 * or ndo_poll may be running while we open the device 1470 */ 1471 netpoll_poll_disable(dev); 1472 1473 ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack); 1474 ret = notifier_to_errno(ret); 1475 if (ret) 1476 return ret; 1477 1478 set_bit(__LINK_STATE_START, &dev->state); 1479 1480 if (ops->ndo_validate_addr) 1481 ret = ops->ndo_validate_addr(dev); 1482 1483 if (!ret && ops->ndo_open) 1484 ret = ops->ndo_open(dev); 1485 1486 netpoll_poll_enable(dev); 1487 1488 if (ret) 1489 clear_bit(__LINK_STATE_START, &dev->state); 1490 else { 1491 dev->flags |= IFF_UP; 1492 dev_set_rx_mode(dev); 1493 dev_activate(dev); 1494 add_device_randomness(dev->dev_addr, dev->addr_len); 1495 } 1496 1497 return ret; 1498 } 1499 1500 /** 1501 * dev_open - prepare an interface for use. 1502 * @dev: device to open 1503 * @extack: netlink extended ack 1504 * 1505 * Takes a device from down to up state. The device's private open 1506 * function is invoked and then the multicast lists are loaded. Finally 1507 * the device is moved into the up state and a %NETDEV_UP message is 1508 * sent to the netdev notifier chain. 1509 * 1510 * Calling this function on an active interface is a nop. On a failure 1511 * a negative errno code is returned. 1512 */ 1513 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1514 { 1515 int ret; 1516 1517 if (dev->flags & IFF_UP) 1518 return 0; 1519 1520 ret = __dev_open(dev, extack); 1521 if (ret < 0) 1522 return ret; 1523 1524 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); 1525 call_netdevice_notifiers(NETDEV_UP, dev); 1526 1527 return ret; 1528 } 1529 EXPORT_SYMBOL(dev_open); 1530 1531 static void __dev_close_many(struct list_head *head) 1532 { 1533 struct net_device *dev; 1534 1535 ASSERT_RTNL(); 1536 might_sleep(); 1537 1538 list_for_each_entry(dev, head, close_list) { 1539 /* Temporarily disable netpoll until the interface is down */ 1540 netpoll_poll_disable(dev); 1541 1542 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); 1543 1544 clear_bit(__LINK_STATE_START, &dev->state); 1545 1546 /* Synchronize to scheduled poll. We cannot touch poll list, it 1547 * can be even on different cpu. So just clear netif_running(). 1548 * 1549 * dev->stop() will invoke napi_disable() on all of it's 1550 * napi_struct instances on this device. 1551 */ 1552 smp_mb__after_atomic(); /* Commit netif_running(). */ 1553 } 1554 1555 dev_deactivate_many(head); 1556 1557 list_for_each_entry(dev, head, close_list) { 1558 const struct net_device_ops *ops = dev->netdev_ops; 1559 1560 /* 1561 * Call the device specific close. This cannot fail. 1562 * Only if device is UP 1563 * 1564 * We allow it to be called even after a DETACH hot-plug 1565 * event. 1566 */ 1567 if (ops->ndo_stop) 1568 ops->ndo_stop(dev); 1569 1570 dev->flags &= ~IFF_UP; 1571 netpoll_poll_enable(dev); 1572 } 1573 } 1574 1575 static void __dev_close(struct net_device *dev) 1576 { 1577 LIST_HEAD(single); 1578 1579 list_add(&dev->close_list, &single); 1580 __dev_close_many(&single); 1581 list_del(&single); 1582 } 1583 1584 void dev_close_many(struct list_head *head, bool unlink) 1585 { 1586 struct net_device *dev, *tmp; 1587 1588 /* Remove the devices that don't need to be closed */ 1589 list_for_each_entry_safe(dev, tmp, head, close_list) 1590 if (!(dev->flags & IFF_UP)) 1591 list_del_init(&dev->close_list); 1592 1593 __dev_close_many(head); 1594 1595 list_for_each_entry_safe(dev, tmp, head, close_list) { 1596 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); 1597 call_netdevice_notifiers(NETDEV_DOWN, dev); 1598 if (unlink) 1599 list_del_init(&dev->close_list); 1600 } 1601 } 1602 EXPORT_SYMBOL(dev_close_many); 1603 1604 /** 1605 * dev_close - shutdown an interface. 1606 * @dev: device to shutdown 1607 * 1608 * This function moves an active device into down state. A 1609 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device 1610 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier 1611 * chain. 1612 */ 1613 void dev_close(struct net_device *dev) 1614 { 1615 if (dev->flags & IFF_UP) { 1616 LIST_HEAD(single); 1617 1618 list_add(&dev->close_list, &single); 1619 dev_close_many(&single, true); 1620 list_del(&single); 1621 } 1622 } 1623 EXPORT_SYMBOL(dev_close); 1624 1625 1626 /** 1627 * dev_disable_lro - disable Large Receive Offload on a device 1628 * @dev: device 1629 * 1630 * Disable Large Receive Offload (LRO) on a net device. Must be 1631 * called under RTNL. This is needed if received packets may be 1632 * forwarded to another interface. 1633 */ 1634 void dev_disable_lro(struct net_device *dev) 1635 { 1636 struct net_device *lower_dev; 1637 struct list_head *iter; 1638 1639 dev->wanted_features &= ~NETIF_F_LRO; 1640 netdev_update_features(dev); 1641 1642 if (unlikely(dev->features & NETIF_F_LRO)) 1643 netdev_WARN(dev, "failed to disable LRO!\n"); 1644 1645 netdev_for_each_lower_dev(dev, lower_dev, iter) 1646 dev_disable_lro(lower_dev); 1647 } 1648 EXPORT_SYMBOL(dev_disable_lro); 1649 1650 /** 1651 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device 1652 * @dev: device 1653 * 1654 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be 1655 * called under RTNL. This is needed if Generic XDP is installed on 1656 * the device. 1657 */ 1658 static void dev_disable_gro_hw(struct net_device *dev) 1659 { 1660 dev->wanted_features &= ~NETIF_F_GRO_HW; 1661 netdev_update_features(dev); 1662 1663 if (unlikely(dev->features & NETIF_F_GRO_HW)) 1664 netdev_WARN(dev, "failed to disable GRO_HW!\n"); 1665 } 1666 1667 const char *netdev_cmd_to_name(enum netdev_cmd cmd) 1668 { 1669 #define N(val) \ 1670 case NETDEV_##val: \ 1671 return "NETDEV_" __stringify(val); 1672 switch (cmd) { 1673 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER) 1674 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE) 1675 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE) 1676 N(POST_INIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) N(CHANGEUPPER) 1677 N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) N(BONDING_INFO) 1678 N(PRECHANGEUPPER) N(CHANGELOWERSTATE) N(UDP_TUNNEL_PUSH_INFO) 1679 N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN) 1680 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO) 1681 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO) 1682 N(PRE_CHANGEADDR) 1683 } 1684 #undef N 1685 return "UNKNOWN_NETDEV_EVENT"; 1686 } 1687 EXPORT_SYMBOL_GPL(netdev_cmd_to_name); 1688 1689 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, 1690 struct net_device *dev) 1691 { 1692 struct netdev_notifier_info info = { 1693 .dev = dev, 1694 }; 1695 1696 return nb->notifier_call(nb, val, &info); 1697 } 1698 1699 static int call_netdevice_register_notifiers(struct notifier_block *nb, 1700 struct net_device *dev) 1701 { 1702 int err; 1703 1704 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); 1705 err = notifier_to_errno(err); 1706 if (err) 1707 return err; 1708 1709 if (!(dev->flags & IFF_UP)) 1710 return 0; 1711 1712 call_netdevice_notifier(nb, NETDEV_UP, dev); 1713 return 0; 1714 } 1715 1716 static void call_netdevice_unregister_notifiers(struct notifier_block *nb, 1717 struct net_device *dev) 1718 { 1719 if (dev->flags & IFF_UP) { 1720 call_netdevice_notifier(nb, NETDEV_GOING_DOWN, 1721 dev); 1722 call_netdevice_notifier(nb, NETDEV_DOWN, dev); 1723 } 1724 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); 1725 } 1726 1727 static int call_netdevice_register_net_notifiers(struct notifier_block *nb, 1728 struct net *net) 1729 { 1730 struct net_device *dev; 1731 int err; 1732 1733 for_each_netdev(net, dev) { 1734 err = call_netdevice_register_notifiers(nb, dev); 1735 if (err) 1736 goto rollback; 1737 } 1738 return 0; 1739 1740 rollback: 1741 for_each_netdev_continue_reverse(net, dev) 1742 call_netdevice_unregister_notifiers(nb, dev); 1743 return err; 1744 } 1745 1746 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb, 1747 struct net *net) 1748 { 1749 struct net_device *dev; 1750 1751 for_each_netdev(net, dev) 1752 call_netdevice_unregister_notifiers(nb, dev); 1753 } 1754 1755 static int dev_boot_phase = 1; 1756 1757 /** 1758 * register_netdevice_notifier - register a network notifier block 1759 * @nb: notifier 1760 * 1761 * Register a notifier to be called when network device events occur. 1762 * The notifier passed is linked into the kernel structures and must 1763 * not be reused until it has been unregistered. A negative errno code 1764 * is returned on a failure. 1765 * 1766 * When registered all registration and up events are replayed 1767 * to the new notifier to allow device to have a race free 1768 * view of the network device list. 1769 */ 1770 1771 int register_netdevice_notifier(struct notifier_block *nb) 1772 { 1773 struct net *net; 1774 int err; 1775 1776 /* Close race with setup_net() and cleanup_net() */ 1777 down_write(&pernet_ops_rwsem); 1778 rtnl_lock(); 1779 err = raw_notifier_chain_register(&netdev_chain, nb); 1780 if (err) 1781 goto unlock; 1782 if (dev_boot_phase) 1783 goto unlock; 1784 for_each_net(net) { 1785 err = call_netdevice_register_net_notifiers(nb, net); 1786 if (err) 1787 goto rollback; 1788 } 1789 1790 unlock: 1791 rtnl_unlock(); 1792 up_write(&pernet_ops_rwsem); 1793 return err; 1794 1795 rollback: 1796 for_each_net_continue_reverse(net) 1797 call_netdevice_unregister_net_notifiers(nb, net); 1798 1799 raw_notifier_chain_unregister(&netdev_chain, nb); 1800 goto unlock; 1801 } 1802 EXPORT_SYMBOL(register_netdevice_notifier); 1803 1804 /** 1805 * unregister_netdevice_notifier - unregister a network notifier block 1806 * @nb: notifier 1807 * 1808 * Unregister a notifier previously registered by 1809 * register_netdevice_notifier(). The notifier is unlinked into the 1810 * kernel structures and may then be reused. A negative errno code 1811 * is returned on a failure. 1812 * 1813 * After unregistering unregister and down device events are synthesized 1814 * for all devices on the device list to the removed notifier to remove 1815 * the need for special case cleanup code. 1816 */ 1817 1818 int unregister_netdevice_notifier(struct notifier_block *nb) 1819 { 1820 struct net *net; 1821 int err; 1822 1823 /* Close race with setup_net() and cleanup_net() */ 1824 down_write(&pernet_ops_rwsem); 1825 rtnl_lock(); 1826 err = raw_notifier_chain_unregister(&netdev_chain, nb); 1827 if (err) 1828 goto unlock; 1829 1830 for_each_net(net) 1831 call_netdevice_unregister_net_notifiers(nb, net); 1832 1833 unlock: 1834 rtnl_unlock(); 1835 up_write(&pernet_ops_rwsem); 1836 return err; 1837 } 1838 EXPORT_SYMBOL(unregister_netdevice_notifier); 1839 1840 static int __register_netdevice_notifier_net(struct net *net, 1841 struct notifier_block *nb, 1842 bool ignore_call_fail) 1843 { 1844 int err; 1845 1846 err = raw_notifier_chain_register(&net->netdev_chain, nb); 1847 if (err) 1848 return err; 1849 if (dev_boot_phase) 1850 return 0; 1851 1852 err = call_netdevice_register_net_notifiers(nb, net); 1853 if (err && !ignore_call_fail) 1854 goto chain_unregister; 1855 1856 return 0; 1857 1858 chain_unregister: 1859 raw_notifier_chain_unregister(&net->netdev_chain, nb); 1860 return err; 1861 } 1862 1863 static int __unregister_netdevice_notifier_net(struct net *net, 1864 struct notifier_block *nb) 1865 { 1866 int err; 1867 1868 err = raw_notifier_chain_unregister(&net->netdev_chain, nb); 1869 if (err) 1870 return err; 1871 1872 call_netdevice_unregister_net_notifiers(nb, net); 1873 return 0; 1874 } 1875 1876 /** 1877 * register_netdevice_notifier_net - register a per-netns network notifier block 1878 * @net: network namespace 1879 * @nb: notifier 1880 * 1881 * Register a notifier to be called when network device events occur. 1882 * The notifier passed is linked into the kernel structures and must 1883 * not be reused until it has been unregistered. A negative errno code 1884 * is returned on a failure. 1885 * 1886 * When registered all registration and up events are replayed 1887 * to the new notifier to allow device to have a race free 1888 * view of the network device list. 1889 */ 1890 1891 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb) 1892 { 1893 int err; 1894 1895 rtnl_lock(); 1896 err = __register_netdevice_notifier_net(net, nb, false); 1897 rtnl_unlock(); 1898 return err; 1899 } 1900 EXPORT_SYMBOL(register_netdevice_notifier_net); 1901 1902 /** 1903 * unregister_netdevice_notifier_net - unregister a per-netns 1904 * network notifier block 1905 * @net: network namespace 1906 * @nb: notifier 1907 * 1908 * Unregister a notifier previously registered by 1909 * register_netdevice_notifier(). The notifier is unlinked into the 1910 * kernel structures and may then be reused. A negative errno code 1911 * is returned on a failure. 1912 * 1913 * After unregistering unregister and down device events are synthesized 1914 * for all devices on the device list to the removed notifier to remove 1915 * the need for special case cleanup code. 1916 */ 1917 1918 int unregister_netdevice_notifier_net(struct net *net, 1919 struct notifier_block *nb) 1920 { 1921 int err; 1922 1923 rtnl_lock(); 1924 err = __unregister_netdevice_notifier_net(net, nb); 1925 rtnl_unlock(); 1926 return err; 1927 } 1928 EXPORT_SYMBOL(unregister_netdevice_notifier_net); 1929 1930 int register_netdevice_notifier_dev_net(struct net_device *dev, 1931 struct notifier_block *nb, 1932 struct netdev_net_notifier *nn) 1933 { 1934 int err; 1935 1936 rtnl_lock(); 1937 err = __register_netdevice_notifier_net(dev_net(dev), nb, false); 1938 if (!err) { 1939 nn->nb = nb; 1940 list_add(&nn->list, &dev->net_notifier_list); 1941 } 1942 rtnl_unlock(); 1943 return err; 1944 } 1945 EXPORT_SYMBOL(register_netdevice_notifier_dev_net); 1946 1947 int unregister_netdevice_notifier_dev_net(struct net_device *dev, 1948 struct notifier_block *nb, 1949 struct netdev_net_notifier *nn) 1950 { 1951 int err; 1952 1953 rtnl_lock(); 1954 list_del(&nn->list); 1955 err = __unregister_netdevice_notifier_net(dev_net(dev), nb); 1956 rtnl_unlock(); 1957 return err; 1958 } 1959 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net); 1960 1961 static void move_netdevice_notifiers_dev_net(struct net_device *dev, 1962 struct net *net) 1963 { 1964 struct netdev_net_notifier *nn; 1965 1966 list_for_each_entry(nn, &dev->net_notifier_list, list) { 1967 __unregister_netdevice_notifier_net(dev_net(dev), nn->nb); 1968 __register_netdevice_notifier_net(net, nn->nb, true); 1969 } 1970 } 1971 1972 /** 1973 * call_netdevice_notifiers_info - call all network notifier blocks 1974 * @val: value passed unmodified to notifier function 1975 * @info: notifier information data 1976 * 1977 * Call all network notifier blocks. Parameters and return value 1978 * are as for raw_notifier_call_chain(). 1979 */ 1980 1981 static int call_netdevice_notifiers_info(unsigned long val, 1982 struct netdev_notifier_info *info) 1983 { 1984 struct net *net = dev_net(info->dev); 1985 int ret; 1986 1987 ASSERT_RTNL(); 1988 1989 /* Run per-netns notifier block chain first, then run the global one. 1990 * Hopefully, one day, the global one is going to be removed after 1991 * all notifier block registrators get converted to be per-netns. 1992 */ 1993 ret = raw_notifier_call_chain(&net->netdev_chain, val, info); 1994 if (ret & NOTIFY_STOP_MASK) 1995 return ret; 1996 return raw_notifier_call_chain(&netdev_chain, val, info); 1997 } 1998 1999 static int call_netdevice_notifiers_extack(unsigned long val, 2000 struct net_device *dev, 2001 struct netlink_ext_ack *extack) 2002 { 2003 struct netdev_notifier_info info = { 2004 .dev = dev, 2005 .extack = extack, 2006 }; 2007 2008 return call_netdevice_notifiers_info(val, &info); 2009 } 2010 2011 /** 2012 * call_netdevice_notifiers - call all network notifier blocks 2013 * @val: value passed unmodified to notifier function 2014 * @dev: net_device pointer passed unmodified to notifier function 2015 * 2016 * Call all network notifier blocks. Parameters and return value 2017 * are as for raw_notifier_call_chain(). 2018 */ 2019 2020 int call_netdevice_notifiers(unsigned long val, struct net_device *dev) 2021 { 2022 return call_netdevice_notifiers_extack(val, dev, NULL); 2023 } 2024 EXPORT_SYMBOL(call_netdevice_notifiers); 2025 2026 /** 2027 * call_netdevice_notifiers_mtu - call all network notifier blocks 2028 * @val: value passed unmodified to notifier function 2029 * @dev: net_device pointer passed unmodified to notifier function 2030 * @arg: additional u32 argument passed to the notifier function 2031 * 2032 * Call all network notifier blocks. Parameters and return value 2033 * are as for raw_notifier_call_chain(). 2034 */ 2035 static int call_netdevice_notifiers_mtu(unsigned long val, 2036 struct net_device *dev, u32 arg) 2037 { 2038 struct netdev_notifier_info_ext info = { 2039 .info.dev = dev, 2040 .ext.mtu = arg, 2041 }; 2042 2043 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0); 2044 2045 return call_netdevice_notifiers_info(val, &info.info); 2046 } 2047 2048 #ifdef CONFIG_NET_INGRESS 2049 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key); 2050 2051 void net_inc_ingress_queue(void) 2052 { 2053 static_branch_inc(&ingress_needed_key); 2054 } 2055 EXPORT_SYMBOL_GPL(net_inc_ingress_queue); 2056 2057 void net_dec_ingress_queue(void) 2058 { 2059 static_branch_dec(&ingress_needed_key); 2060 } 2061 EXPORT_SYMBOL_GPL(net_dec_ingress_queue); 2062 #endif 2063 2064 #ifdef CONFIG_NET_EGRESS 2065 static DEFINE_STATIC_KEY_FALSE(egress_needed_key); 2066 2067 void net_inc_egress_queue(void) 2068 { 2069 static_branch_inc(&egress_needed_key); 2070 } 2071 EXPORT_SYMBOL_GPL(net_inc_egress_queue); 2072 2073 void net_dec_egress_queue(void) 2074 { 2075 static_branch_dec(&egress_needed_key); 2076 } 2077 EXPORT_SYMBOL_GPL(net_dec_egress_queue); 2078 #endif 2079 2080 static DEFINE_STATIC_KEY_FALSE(netstamp_needed_key); 2081 #ifdef CONFIG_JUMP_LABEL 2082 static atomic_t netstamp_needed_deferred; 2083 static atomic_t netstamp_wanted; 2084 static void netstamp_clear(struct work_struct *work) 2085 { 2086 int deferred = atomic_xchg(&netstamp_needed_deferred, 0); 2087 int wanted; 2088 2089 wanted = atomic_add_return(deferred, &netstamp_wanted); 2090 if (wanted > 0) 2091 static_branch_enable(&netstamp_needed_key); 2092 else 2093 static_branch_disable(&netstamp_needed_key); 2094 } 2095 static DECLARE_WORK(netstamp_work, netstamp_clear); 2096 #endif 2097 2098 void net_enable_timestamp(void) 2099 { 2100 #ifdef CONFIG_JUMP_LABEL 2101 int wanted; 2102 2103 while (1) { 2104 wanted = atomic_read(&netstamp_wanted); 2105 if (wanted <= 0) 2106 break; 2107 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted) 2108 return; 2109 } 2110 atomic_inc(&netstamp_needed_deferred); 2111 schedule_work(&netstamp_work); 2112 #else 2113 static_branch_inc(&netstamp_needed_key); 2114 #endif 2115 } 2116 EXPORT_SYMBOL(net_enable_timestamp); 2117 2118 void net_disable_timestamp(void) 2119 { 2120 #ifdef CONFIG_JUMP_LABEL 2121 int wanted; 2122 2123 while (1) { 2124 wanted = atomic_read(&netstamp_wanted); 2125 if (wanted <= 1) 2126 break; 2127 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted) 2128 return; 2129 } 2130 atomic_dec(&netstamp_needed_deferred); 2131 schedule_work(&netstamp_work); 2132 #else 2133 static_branch_dec(&netstamp_needed_key); 2134 #endif 2135 } 2136 EXPORT_SYMBOL(net_disable_timestamp); 2137 2138 static inline void net_timestamp_set(struct sk_buff *skb) 2139 { 2140 skb->tstamp = 0; 2141 if (static_branch_unlikely(&netstamp_needed_key)) 2142 __net_timestamp(skb); 2143 } 2144 2145 #define net_timestamp_check(COND, SKB) \ 2146 if (static_branch_unlikely(&netstamp_needed_key)) { \ 2147 if ((COND) && !(SKB)->tstamp) \ 2148 __net_timestamp(SKB); \ 2149 } \ 2150 2151 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb) 2152 { 2153 return __is_skb_forwardable(dev, skb, true); 2154 } 2155 EXPORT_SYMBOL_GPL(is_skb_forwardable); 2156 2157 static int __dev_forward_skb2(struct net_device *dev, struct sk_buff *skb, 2158 bool check_mtu) 2159 { 2160 int ret = ____dev_forward_skb(dev, skb, check_mtu); 2161 2162 if (likely(!ret)) { 2163 skb->protocol = eth_type_trans(skb, dev); 2164 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); 2165 } 2166 2167 return ret; 2168 } 2169 2170 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2171 { 2172 return __dev_forward_skb2(dev, skb, true); 2173 } 2174 EXPORT_SYMBOL_GPL(__dev_forward_skb); 2175 2176 /** 2177 * dev_forward_skb - loopback an skb to another netif 2178 * 2179 * @dev: destination network device 2180 * @skb: buffer to forward 2181 * 2182 * return values: 2183 * NET_RX_SUCCESS (no congestion) 2184 * NET_RX_DROP (packet was dropped, but freed) 2185 * 2186 * dev_forward_skb can be used for injecting an skb from the 2187 * start_xmit function of one device into the receive queue 2188 * of another device. 2189 * 2190 * The receiving device may be in another namespace, so 2191 * we have to clear all information in the skb that could 2192 * impact namespace isolation. 2193 */ 2194 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2195 { 2196 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); 2197 } 2198 EXPORT_SYMBOL_GPL(dev_forward_skb); 2199 2200 int dev_forward_skb_nomtu(struct net_device *dev, struct sk_buff *skb) 2201 { 2202 return __dev_forward_skb2(dev, skb, false) ?: netif_rx_internal(skb); 2203 } 2204 2205 static inline int deliver_skb(struct sk_buff *skb, 2206 struct packet_type *pt_prev, 2207 struct net_device *orig_dev) 2208 { 2209 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 2210 return -ENOMEM; 2211 refcount_inc(&skb->users); 2212 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 2213 } 2214 2215 static inline void deliver_ptype_list_skb(struct sk_buff *skb, 2216 struct packet_type **pt, 2217 struct net_device *orig_dev, 2218 __be16 type, 2219 struct list_head *ptype_list) 2220 { 2221 struct packet_type *ptype, *pt_prev = *pt; 2222 2223 list_for_each_entry_rcu(ptype, ptype_list, list) { 2224 if (ptype->type != type) 2225 continue; 2226 if (pt_prev) 2227 deliver_skb(skb, pt_prev, orig_dev); 2228 pt_prev = ptype; 2229 } 2230 *pt = pt_prev; 2231 } 2232 2233 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) 2234 { 2235 if (!ptype->af_packet_priv || !skb->sk) 2236 return false; 2237 2238 if (ptype->id_match) 2239 return ptype->id_match(ptype, skb->sk); 2240 else if ((struct sock *)ptype->af_packet_priv == skb->sk) 2241 return true; 2242 2243 return false; 2244 } 2245 2246 /** 2247 * dev_nit_active - return true if any network interface taps are in use 2248 * 2249 * @dev: network device to check for the presence of taps 2250 */ 2251 bool dev_nit_active(struct net_device *dev) 2252 { 2253 return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all); 2254 } 2255 EXPORT_SYMBOL_GPL(dev_nit_active); 2256 2257 /* 2258 * Support routine. Sends outgoing frames to any network 2259 * taps currently in use. 2260 */ 2261 2262 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) 2263 { 2264 struct packet_type *ptype; 2265 struct sk_buff *skb2 = NULL; 2266 struct packet_type *pt_prev = NULL; 2267 struct list_head *ptype_list = &ptype_all; 2268 2269 rcu_read_lock(); 2270 again: 2271 list_for_each_entry_rcu(ptype, ptype_list, list) { 2272 if (ptype->ignore_outgoing) 2273 continue; 2274 2275 /* Never send packets back to the socket 2276 * they originated from - MvS (miquels@drinkel.ow.org) 2277 */ 2278 if (skb_loop_sk(ptype, skb)) 2279 continue; 2280 2281 if (pt_prev) { 2282 deliver_skb(skb2, pt_prev, skb->dev); 2283 pt_prev = ptype; 2284 continue; 2285 } 2286 2287 /* need to clone skb, done only once */ 2288 skb2 = skb_clone(skb, GFP_ATOMIC); 2289 if (!skb2) 2290 goto out_unlock; 2291 2292 net_timestamp_set(skb2); 2293 2294 /* skb->nh should be correctly 2295 * set by sender, so that the second statement is 2296 * just protection against buggy protocols. 2297 */ 2298 skb_reset_mac_header(skb2); 2299 2300 if (skb_network_header(skb2) < skb2->data || 2301 skb_network_header(skb2) > skb_tail_pointer(skb2)) { 2302 net_crit_ratelimited("protocol %04x is buggy, dev %s\n", 2303 ntohs(skb2->protocol), 2304 dev->name); 2305 skb_reset_network_header(skb2); 2306 } 2307 2308 skb2->transport_header = skb2->network_header; 2309 skb2->pkt_type = PACKET_OUTGOING; 2310 pt_prev = ptype; 2311 } 2312 2313 if (ptype_list == &ptype_all) { 2314 ptype_list = &dev->ptype_all; 2315 goto again; 2316 } 2317 out_unlock: 2318 if (pt_prev) { 2319 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC)) 2320 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); 2321 else 2322 kfree_skb(skb2); 2323 } 2324 rcu_read_unlock(); 2325 } 2326 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit); 2327 2328 /** 2329 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change 2330 * @dev: Network device 2331 * @txq: number of queues available 2332 * 2333 * If real_num_tx_queues is changed the tc mappings may no longer be 2334 * valid. To resolve this verify the tc mapping remains valid and if 2335 * not NULL the mapping. With no priorities mapping to this 2336 * offset/count pair it will no longer be used. In the worst case TC0 2337 * is invalid nothing can be done so disable priority mappings. If is 2338 * expected that drivers will fix this mapping if they can before 2339 * calling netif_set_real_num_tx_queues. 2340 */ 2341 static void netif_setup_tc(struct net_device *dev, unsigned int txq) 2342 { 2343 int i; 2344 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2345 2346 /* If TC0 is invalidated disable TC mapping */ 2347 if (tc->offset + tc->count > txq) { 2348 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); 2349 dev->num_tc = 0; 2350 return; 2351 } 2352 2353 /* Invalidated prio to tc mappings set to TC0 */ 2354 for (i = 1; i < TC_BITMASK + 1; i++) { 2355 int q = netdev_get_prio_tc_map(dev, i); 2356 2357 tc = &dev->tc_to_txq[q]; 2358 if (tc->offset + tc->count > txq) { 2359 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n", 2360 i, q); 2361 netdev_set_prio_tc_map(dev, i, 0); 2362 } 2363 } 2364 } 2365 2366 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq) 2367 { 2368 if (dev->num_tc) { 2369 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2370 int i; 2371 2372 /* walk through the TCs and see if it falls into any of them */ 2373 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) { 2374 if ((txq - tc->offset) < tc->count) 2375 return i; 2376 } 2377 2378 /* didn't find it, just return -1 to indicate no match */ 2379 return -1; 2380 } 2381 2382 return 0; 2383 } 2384 EXPORT_SYMBOL(netdev_txq_to_tc); 2385 2386 #ifdef CONFIG_XPS 2387 static struct static_key xps_needed __read_mostly; 2388 static struct static_key xps_rxqs_needed __read_mostly; 2389 static DEFINE_MUTEX(xps_map_mutex); 2390 #define xmap_dereference(P) \ 2391 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) 2392 2393 static bool remove_xps_queue(struct xps_dev_maps *dev_maps, 2394 struct xps_dev_maps *old_maps, int tci, u16 index) 2395 { 2396 struct xps_map *map = NULL; 2397 int pos; 2398 2399 if (dev_maps) 2400 map = xmap_dereference(dev_maps->attr_map[tci]); 2401 if (!map) 2402 return false; 2403 2404 for (pos = map->len; pos--;) { 2405 if (map->queues[pos] != index) 2406 continue; 2407 2408 if (map->len > 1) { 2409 map->queues[pos] = map->queues[--map->len]; 2410 break; 2411 } 2412 2413 if (old_maps) 2414 RCU_INIT_POINTER(old_maps->attr_map[tci], NULL); 2415 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2416 kfree_rcu(map, rcu); 2417 return false; 2418 } 2419 2420 return true; 2421 } 2422 2423 static bool remove_xps_queue_cpu(struct net_device *dev, 2424 struct xps_dev_maps *dev_maps, 2425 int cpu, u16 offset, u16 count) 2426 { 2427 int num_tc = dev_maps->num_tc; 2428 bool active = false; 2429 int tci; 2430 2431 for (tci = cpu * num_tc; num_tc--; tci++) { 2432 int i, j; 2433 2434 for (i = count, j = offset; i--; j++) { 2435 if (!remove_xps_queue(dev_maps, NULL, tci, j)) 2436 break; 2437 } 2438 2439 active |= i < 0; 2440 } 2441 2442 return active; 2443 } 2444 2445 static void reset_xps_maps(struct net_device *dev, 2446 struct xps_dev_maps *dev_maps, 2447 enum xps_map_type type) 2448 { 2449 static_key_slow_dec_cpuslocked(&xps_needed); 2450 if (type == XPS_RXQS) 2451 static_key_slow_dec_cpuslocked(&xps_rxqs_needed); 2452 2453 RCU_INIT_POINTER(dev->xps_maps[type], NULL); 2454 2455 kfree_rcu(dev_maps, rcu); 2456 } 2457 2458 static void clean_xps_maps(struct net_device *dev, enum xps_map_type type, 2459 u16 offset, u16 count) 2460 { 2461 struct xps_dev_maps *dev_maps; 2462 bool active = false; 2463 int i, j; 2464 2465 dev_maps = xmap_dereference(dev->xps_maps[type]); 2466 if (!dev_maps) 2467 return; 2468 2469 for (j = 0; j < dev_maps->nr_ids; j++) 2470 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, count); 2471 if (!active) 2472 reset_xps_maps(dev, dev_maps, type); 2473 2474 if (type == XPS_CPUS) { 2475 for (i = offset + (count - 1); count--; i--) 2476 netdev_queue_numa_node_write( 2477 netdev_get_tx_queue(dev, i), NUMA_NO_NODE); 2478 } 2479 } 2480 2481 static void netif_reset_xps_queues(struct net_device *dev, u16 offset, 2482 u16 count) 2483 { 2484 if (!static_key_false(&xps_needed)) 2485 return; 2486 2487 cpus_read_lock(); 2488 mutex_lock(&xps_map_mutex); 2489 2490 if (static_key_false(&xps_rxqs_needed)) 2491 clean_xps_maps(dev, XPS_RXQS, offset, count); 2492 2493 clean_xps_maps(dev, XPS_CPUS, offset, count); 2494 2495 mutex_unlock(&xps_map_mutex); 2496 cpus_read_unlock(); 2497 } 2498 2499 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) 2500 { 2501 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index); 2502 } 2503 2504 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index, 2505 u16 index, bool is_rxqs_map) 2506 { 2507 struct xps_map *new_map; 2508 int alloc_len = XPS_MIN_MAP_ALLOC; 2509 int i, pos; 2510 2511 for (pos = 0; map && pos < map->len; pos++) { 2512 if (map->queues[pos] != index) 2513 continue; 2514 return map; 2515 } 2516 2517 /* Need to add tx-queue to this CPU's/rx-queue's existing map */ 2518 if (map) { 2519 if (pos < map->alloc_len) 2520 return map; 2521 2522 alloc_len = map->alloc_len * 2; 2523 } 2524 2525 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's 2526 * map 2527 */ 2528 if (is_rxqs_map) 2529 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL); 2530 else 2531 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, 2532 cpu_to_node(attr_index)); 2533 if (!new_map) 2534 return NULL; 2535 2536 for (i = 0; i < pos; i++) 2537 new_map->queues[i] = map->queues[i]; 2538 new_map->alloc_len = alloc_len; 2539 new_map->len = pos; 2540 2541 return new_map; 2542 } 2543 2544 /* Copy xps maps at a given index */ 2545 static void xps_copy_dev_maps(struct xps_dev_maps *dev_maps, 2546 struct xps_dev_maps *new_dev_maps, int index, 2547 int tc, bool skip_tc) 2548 { 2549 int i, tci = index * dev_maps->num_tc; 2550 struct xps_map *map; 2551 2552 /* copy maps belonging to foreign traffic classes */ 2553 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2554 if (i == tc && skip_tc) 2555 continue; 2556 2557 /* fill in the new device map from the old device map */ 2558 map = xmap_dereference(dev_maps->attr_map[tci]); 2559 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2560 } 2561 } 2562 2563 /* Must be called under cpus_read_lock */ 2564 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask, 2565 u16 index, enum xps_map_type type) 2566 { 2567 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL, *old_dev_maps = NULL; 2568 const unsigned long *online_mask = NULL; 2569 bool active = false, copy = false; 2570 int i, j, tci, numa_node_id = -2; 2571 int maps_sz, num_tc = 1, tc = 0; 2572 struct xps_map *map, *new_map; 2573 unsigned int nr_ids; 2574 2575 if (dev->num_tc) { 2576 /* Do not allow XPS on subordinate device directly */ 2577 num_tc = dev->num_tc; 2578 if (num_tc < 0) 2579 return -EINVAL; 2580 2581 /* If queue belongs to subordinate dev use its map */ 2582 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev; 2583 2584 tc = netdev_txq_to_tc(dev, index); 2585 if (tc < 0) 2586 return -EINVAL; 2587 } 2588 2589 mutex_lock(&xps_map_mutex); 2590 2591 dev_maps = xmap_dereference(dev->xps_maps[type]); 2592 if (type == XPS_RXQS) { 2593 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues); 2594 nr_ids = dev->num_rx_queues; 2595 } else { 2596 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc); 2597 if (num_possible_cpus() > 1) 2598 online_mask = cpumask_bits(cpu_online_mask); 2599 nr_ids = nr_cpu_ids; 2600 } 2601 2602 if (maps_sz < L1_CACHE_BYTES) 2603 maps_sz = L1_CACHE_BYTES; 2604 2605 /* The old dev_maps could be larger or smaller than the one we're 2606 * setting up now, as dev->num_tc or nr_ids could have been updated in 2607 * between. We could try to be smart, but let's be safe instead and only 2608 * copy foreign traffic classes if the two map sizes match. 2609 */ 2610 if (dev_maps && 2611 dev_maps->num_tc == num_tc && dev_maps->nr_ids == nr_ids) 2612 copy = true; 2613 2614 /* allocate memory for queue storage */ 2615 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids), 2616 j < nr_ids;) { 2617 if (!new_dev_maps) { 2618 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); 2619 if (!new_dev_maps) { 2620 mutex_unlock(&xps_map_mutex); 2621 return -ENOMEM; 2622 } 2623 2624 new_dev_maps->nr_ids = nr_ids; 2625 new_dev_maps->num_tc = num_tc; 2626 } 2627 2628 tci = j * num_tc + tc; 2629 map = copy ? xmap_dereference(dev_maps->attr_map[tci]) : NULL; 2630 2631 map = expand_xps_map(map, j, index, type == XPS_RXQS); 2632 if (!map) 2633 goto error; 2634 2635 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2636 } 2637 2638 if (!new_dev_maps) 2639 goto out_no_new_maps; 2640 2641 if (!dev_maps) { 2642 /* Increment static keys at most once per type */ 2643 static_key_slow_inc_cpuslocked(&xps_needed); 2644 if (type == XPS_RXQS) 2645 static_key_slow_inc_cpuslocked(&xps_rxqs_needed); 2646 } 2647 2648 for (j = 0; j < nr_ids; j++) { 2649 bool skip_tc = false; 2650 2651 tci = j * num_tc + tc; 2652 if (netif_attr_test_mask(j, mask, nr_ids) && 2653 netif_attr_test_online(j, online_mask, nr_ids)) { 2654 /* add tx-queue to CPU/rx-queue maps */ 2655 int pos = 0; 2656 2657 skip_tc = true; 2658 2659 map = xmap_dereference(new_dev_maps->attr_map[tci]); 2660 while ((pos < map->len) && (map->queues[pos] != index)) 2661 pos++; 2662 2663 if (pos == map->len) 2664 map->queues[map->len++] = index; 2665 #ifdef CONFIG_NUMA 2666 if (type == XPS_CPUS) { 2667 if (numa_node_id == -2) 2668 numa_node_id = cpu_to_node(j); 2669 else if (numa_node_id != cpu_to_node(j)) 2670 numa_node_id = -1; 2671 } 2672 #endif 2673 } 2674 2675 if (copy) 2676 xps_copy_dev_maps(dev_maps, new_dev_maps, j, tc, 2677 skip_tc); 2678 } 2679 2680 rcu_assign_pointer(dev->xps_maps[type], new_dev_maps); 2681 2682 /* Cleanup old maps */ 2683 if (!dev_maps) 2684 goto out_no_old_maps; 2685 2686 for (j = 0; j < dev_maps->nr_ids; j++) { 2687 for (i = num_tc, tci = j * dev_maps->num_tc; i--; tci++) { 2688 map = xmap_dereference(dev_maps->attr_map[tci]); 2689 if (!map) 2690 continue; 2691 2692 if (copy) { 2693 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2694 if (map == new_map) 2695 continue; 2696 } 2697 2698 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2699 kfree_rcu(map, rcu); 2700 } 2701 } 2702 2703 old_dev_maps = dev_maps; 2704 2705 out_no_old_maps: 2706 dev_maps = new_dev_maps; 2707 active = true; 2708 2709 out_no_new_maps: 2710 if (type == XPS_CPUS) 2711 /* update Tx queue numa node */ 2712 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), 2713 (numa_node_id >= 0) ? 2714 numa_node_id : NUMA_NO_NODE); 2715 2716 if (!dev_maps) 2717 goto out_no_maps; 2718 2719 /* removes tx-queue from unused CPUs/rx-queues */ 2720 for (j = 0; j < dev_maps->nr_ids; j++) { 2721 tci = j * dev_maps->num_tc; 2722 2723 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2724 if (i == tc && 2725 netif_attr_test_mask(j, mask, dev_maps->nr_ids) && 2726 netif_attr_test_online(j, online_mask, dev_maps->nr_ids)) 2727 continue; 2728 2729 active |= remove_xps_queue(dev_maps, 2730 copy ? old_dev_maps : NULL, 2731 tci, index); 2732 } 2733 } 2734 2735 if (old_dev_maps) 2736 kfree_rcu(old_dev_maps, rcu); 2737 2738 /* free map if not active */ 2739 if (!active) 2740 reset_xps_maps(dev, dev_maps, type); 2741 2742 out_no_maps: 2743 mutex_unlock(&xps_map_mutex); 2744 2745 return 0; 2746 error: 2747 /* remove any maps that we added */ 2748 for (j = 0; j < nr_ids; j++) { 2749 for (i = num_tc, tci = j * num_tc; i--; tci++) { 2750 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2751 map = copy ? 2752 xmap_dereference(dev_maps->attr_map[tci]) : 2753 NULL; 2754 if (new_map && new_map != map) 2755 kfree(new_map); 2756 } 2757 } 2758 2759 mutex_unlock(&xps_map_mutex); 2760 2761 kfree(new_dev_maps); 2762 return -ENOMEM; 2763 } 2764 EXPORT_SYMBOL_GPL(__netif_set_xps_queue); 2765 2766 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, 2767 u16 index) 2768 { 2769 int ret; 2770 2771 cpus_read_lock(); 2772 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, XPS_CPUS); 2773 cpus_read_unlock(); 2774 2775 return ret; 2776 } 2777 EXPORT_SYMBOL(netif_set_xps_queue); 2778 2779 #endif 2780 static void netdev_unbind_all_sb_channels(struct net_device *dev) 2781 { 2782 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2783 2784 /* Unbind any subordinate channels */ 2785 while (txq-- != &dev->_tx[0]) { 2786 if (txq->sb_dev) 2787 netdev_unbind_sb_channel(dev, txq->sb_dev); 2788 } 2789 } 2790 2791 void netdev_reset_tc(struct net_device *dev) 2792 { 2793 #ifdef CONFIG_XPS 2794 netif_reset_xps_queues_gt(dev, 0); 2795 #endif 2796 netdev_unbind_all_sb_channels(dev); 2797 2798 /* Reset TC configuration of device */ 2799 dev->num_tc = 0; 2800 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq)); 2801 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map)); 2802 } 2803 EXPORT_SYMBOL(netdev_reset_tc); 2804 2805 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset) 2806 { 2807 if (tc >= dev->num_tc) 2808 return -EINVAL; 2809 2810 #ifdef CONFIG_XPS 2811 netif_reset_xps_queues(dev, offset, count); 2812 #endif 2813 dev->tc_to_txq[tc].count = count; 2814 dev->tc_to_txq[tc].offset = offset; 2815 return 0; 2816 } 2817 EXPORT_SYMBOL(netdev_set_tc_queue); 2818 2819 int netdev_set_num_tc(struct net_device *dev, u8 num_tc) 2820 { 2821 if (num_tc > TC_MAX_QUEUE) 2822 return -EINVAL; 2823 2824 #ifdef CONFIG_XPS 2825 netif_reset_xps_queues_gt(dev, 0); 2826 #endif 2827 netdev_unbind_all_sb_channels(dev); 2828 2829 dev->num_tc = num_tc; 2830 return 0; 2831 } 2832 EXPORT_SYMBOL(netdev_set_num_tc); 2833 2834 void netdev_unbind_sb_channel(struct net_device *dev, 2835 struct net_device *sb_dev) 2836 { 2837 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2838 2839 #ifdef CONFIG_XPS 2840 netif_reset_xps_queues_gt(sb_dev, 0); 2841 #endif 2842 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq)); 2843 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map)); 2844 2845 while (txq-- != &dev->_tx[0]) { 2846 if (txq->sb_dev == sb_dev) 2847 txq->sb_dev = NULL; 2848 } 2849 } 2850 EXPORT_SYMBOL(netdev_unbind_sb_channel); 2851 2852 int netdev_bind_sb_channel_queue(struct net_device *dev, 2853 struct net_device *sb_dev, 2854 u8 tc, u16 count, u16 offset) 2855 { 2856 /* Make certain the sb_dev and dev are already configured */ 2857 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc) 2858 return -EINVAL; 2859 2860 /* We cannot hand out queues we don't have */ 2861 if ((offset + count) > dev->real_num_tx_queues) 2862 return -EINVAL; 2863 2864 /* Record the mapping */ 2865 sb_dev->tc_to_txq[tc].count = count; 2866 sb_dev->tc_to_txq[tc].offset = offset; 2867 2868 /* Provide a way for Tx queue to find the tc_to_txq map or 2869 * XPS map for itself. 2870 */ 2871 while (count--) 2872 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev; 2873 2874 return 0; 2875 } 2876 EXPORT_SYMBOL(netdev_bind_sb_channel_queue); 2877 2878 int netdev_set_sb_channel(struct net_device *dev, u16 channel) 2879 { 2880 /* Do not use a multiqueue device to represent a subordinate channel */ 2881 if (netif_is_multiqueue(dev)) 2882 return -ENODEV; 2883 2884 /* We allow channels 1 - 32767 to be used for subordinate channels. 2885 * Channel 0 is meant to be "native" mode and used only to represent 2886 * the main root device. We allow writing 0 to reset the device back 2887 * to normal mode after being used as a subordinate channel. 2888 */ 2889 if (channel > S16_MAX) 2890 return -EINVAL; 2891 2892 dev->num_tc = -channel; 2893 2894 return 0; 2895 } 2896 EXPORT_SYMBOL(netdev_set_sb_channel); 2897 2898 /* 2899 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues 2900 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed. 2901 */ 2902 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) 2903 { 2904 bool disabling; 2905 int rc; 2906 2907 disabling = txq < dev->real_num_tx_queues; 2908 2909 if (txq < 1 || txq > dev->num_tx_queues) 2910 return -EINVAL; 2911 2912 if (dev->reg_state == NETREG_REGISTERED || 2913 dev->reg_state == NETREG_UNREGISTERING) { 2914 ASSERT_RTNL(); 2915 2916 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, 2917 txq); 2918 if (rc) 2919 return rc; 2920 2921 if (dev->num_tc) 2922 netif_setup_tc(dev, txq); 2923 2924 dev_qdisc_change_real_num_tx(dev, txq); 2925 2926 dev->real_num_tx_queues = txq; 2927 2928 if (disabling) { 2929 synchronize_net(); 2930 qdisc_reset_all_tx_gt(dev, txq); 2931 #ifdef CONFIG_XPS 2932 netif_reset_xps_queues_gt(dev, txq); 2933 #endif 2934 } 2935 } else { 2936 dev->real_num_tx_queues = txq; 2937 } 2938 2939 return 0; 2940 } 2941 EXPORT_SYMBOL(netif_set_real_num_tx_queues); 2942 2943 #ifdef CONFIG_SYSFS 2944 /** 2945 * netif_set_real_num_rx_queues - set actual number of RX queues used 2946 * @dev: Network device 2947 * @rxq: Actual number of RX queues 2948 * 2949 * This must be called either with the rtnl_lock held or before 2950 * registration of the net device. Returns 0 on success, or a 2951 * negative error code. If called before registration, it always 2952 * succeeds. 2953 */ 2954 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) 2955 { 2956 int rc; 2957 2958 if (rxq < 1 || rxq > dev->num_rx_queues) 2959 return -EINVAL; 2960 2961 if (dev->reg_state == NETREG_REGISTERED) { 2962 ASSERT_RTNL(); 2963 2964 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, 2965 rxq); 2966 if (rc) 2967 return rc; 2968 } 2969 2970 dev->real_num_rx_queues = rxq; 2971 return 0; 2972 } 2973 EXPORT_SYMBOL(netif_set_real_num_rx_queues); 2974 #endif 2975 2976 /** 2977 * netif_set_real_num_queues - set actual number of RX and TX queues used 2978 * @dev: Network device 2979 * @txq: Actual number of TX queues 2980 * @rxq: Actual number of RX queues 2981 * 2982 * Set the real number of both TX and RX queues. 2983 * Does nothing if the number of queues is already correct. 2984 */ 2985 int netif_set_real_num_queues(struct net_device *dev, 2986 unsigned int txq, unsigned int rxq) 2987 { 2988 unsigned int old_rxq = dev->real_num_rx_queues; 2989 int err; 2990 2991 if (txq < 1 || txq > dev->num_tx_queues || 2992 rxq < 1 || rxq > dev->num_rx_queues) 2993 return -EINVAL; 2994 2995 /* Start from increases, so the error path only does decreases - 2996 * decreases can't fail. 2997 */ 2998 if (rxq > dev->real_num_rx_queues) { 2999 err = netif_set_real_num_rx_queues(dev, rxq); 3000 if (err) 3001 return err; 3002 } 3003 if (txq > dev->real_num_tx_queues) { 3004 err = netif_set_real_num_tx_queues(dev, txq); 3005 if (err) 3006 goto undo_rx; 3007 } 3008 if (rxq < dev->real_num_rx_queues) 3009 WARN_ON(netif_set_real_num_rx_queues(dev, rxq)); 3010 if (txq < dev->real_num_tx_queues) 3011 WARN_ON(netif_set_real_num_tx_queues(dev, txq)); 3012 3013 return 0; 3014 undo_rx: 3015 WARN_ON(netif_set_real_num_rx_queues(dev, old_rxq)); 3016 return err; 3017 } 3018 EXPORT_SYMBOL(netif_set_real_num_queues); 3019 3020 /** 3021 * netif_get_num_default_rss_queues - default number of RSS queues 3022 * 3023 * This routine should set an upper limit on the number of RSS queues 3024 * used by default by multiqueue devices. 3025 */ 3026 int netif_get_num_default_rss_queues(void) 3027 { 3028 return is_kdump_kernel() ? 3029 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus()); 3030 } 3031 EXPORT_SYMBOL(netif_get_num_default_rss_queues); 3032 3033 static void __netif_reschedule(struct Qdisc *q) 3034 { 3035 struct softnet_data *sd; 3036 unsigned long flags; 3037 3038 local_irq_save(flags); 3039 sd = this_cpu_ptr(&softnet_data); 3040 q->next_sched = NULL; 3041 *sd->output_queue_tailp = q; 3042 sd->output_queue_tailp = &q->next_sched; 3043 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3044 local_irq_restore(flags); 3045 } 3046 3047 void __netif_schedule(struct Qdisc *q) 3048 { 3049 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) 3050 __netif_reschedule(q); 3051 } 3052 EXPORT_SYMBOL(__netif_schedule); 3053 3054 struct dev_kfree_skb_cb { 3055 enum skb_free_reason reason; 3056 }; 3057 3058 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) 3059 { 3060 return (struct dev_kfree_skb_cb *)skb->cb; 3061 } 3062 3063 void netif_schedule_queue(struct netdev_queue *txq) 3064 { 3065 rcu_read_lock(); 3066 if (!netif_xmit_stopped(txq)) { 3067 struct Qdisc *q = rcu_dereference(txq->qdisc); 3068 3069 __netif_schedule(q); 3070 } 3071 rcu_read_unlock(); 3072 } 3073 EXPORT_SYMBOL(netif_schedule_queue); 3074 3075 void netif_tx_wake_queue(struct netdev_queue *dev_queue) 3076 { 3077 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { 3078 struct Qdisc *q; 3079 3080 rcu_read_lock(); 3081 q = rcu_dereference(dev_queue->qdisc); 3082 __netif_schedule(q); 3083 rcu_read_unlock(); 3084 } 3085 } 3086 EXPORT_SYMBOL(netif_tx_wake_queue); 3087 3088 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason) 3089 { 3090 unsigned long flags; 3091 3092 if (unlikely(!skb)) 3093 return; 3094 3095 if (likely(refcount_read(&skb->users) == 1)) { 3096 smp_rmb(); 3097 refcount_set(&skb->users, 0); 3098 } else if (likely(!refcount_dec_and_test(&skb->users))) { 3099 return; 3100 } 3101 get_kfree_skb_cb(skb)->reason = reason; 3102 local_irq_save(flags); 3103 skb->next = __this_cpu_read(softnet_data.completion_queue); 3104 __this_cpu_write(softnet_data.completion_queue, skb); 3105 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3106 local_irq_restore(flags); 3107 } 3108 EXPORT_SYMBOL(__dev_kfree_skb_irq); 3109 3110 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason) 3111 { 3112 if (in_hardirq() || irqs_disabled()) 3113 __dev_kfree_skb_irq(skb, reason); 3114 else 3115 dev_kfree_skb(skb); 3116 } 3117 EXPORT_SYMBOL(__dev_kfree_skb_any); 3118 3119 3120 /** 3121 * netif_device_detach - mark device as removed 3122 * @dev: network device 3123 * 3124 * Mark device as removed from system and therefore no longer available. 3125 */ 3126 void netif_device_detach(struct net_device *dev) 3127 { 3128 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && 3129 netif_running(dev)) { 3130 netif_tx_stop_all_queues(dev); 3131 } 3132 } 3133 EXPORT_SYMBOL(netif_device_detach); 3134 3135 /** 3136 * netif_device_attach - mark device as attached 3137 * @dev: network device 3138 * 3139 * Mark device as attached from system and restart if needed. 3140 */ 3141 void netif_device_attach(struct net_device *dev) 3142 { 3143 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && 3144 netif_running(dev)) { 3145 netif_tx_wake_all_queues(dev); 3146 __netdev_watchdog_up(dev); 3147 } 3148 } 3149 EXPORT_SYMBOL(netif_device_attach); 3150 3151 /* 3152 * Returns a Tx hash based on the given packet descriptor a Tx queues' number 3153 * to be used as a distribution range. 3154 */ 3155 static u16 skb_tx_hash(const struct net_device *dev, 3156 const struct net_device *sb_dev, 3157 struct sk_buff *skb) 3158 { 3159 u32 hash; 3160 u16 qoffset = 0; 3161 u16 qcount = dev->real_num_tx_queues; 3162 3163 if (dev->num_tc) { 3164 u8 tc = netdev_get_prio_tc_map(dev, skb->priority); 3165 3166 qoffset = sb_dev->tc_to_txq[tc].offset; 3167 qcount = sb_dev->tc_to_txq[tc].count; 3168 } 3169 3170 if (skb_rx_queue_recorded(skb)) { 3171 hash = skb_get_rx_queue(skb); 3172 if (hash >= qoffset) 3173 hash -= qoffset; 3174 while (unlikely(hash >= qcount)) 3175 hash -= qcount; 3176 return hash + qoffset; 3177 } 3178 3179 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; 3180 } 3181 3182 static void skb_warn_bad_offload(const struct sk_buff *skb) 3183 { 3184 static const netdev_features_t null_features; 3185 struct net_device *dev = skb->dev; 3186 const char *name = ""; 3187 3188 if (!net_ratelimit()) 3189 return; 3190 3191 if (dev) { 3192 if (dev->dev.parent) 3193 name = dev_driver_string(dev->dev.parent); 3194 else 3195 name = netdev_name(dev); 3196 } 3197 skb_dump(KERN_WARNING, skb, false); 3198 WARN(1, "%s: caps=(%pNF, %pNF)\n", 3199 name, dev ? &dev->features : &null_features, 3200 skb->sk ? &skb->sk->sk_route_caps : &null_features); 3201 } 3202 3203 /* 3204 * Invalidate hardware checksum when packet is to be mangled, and 3205 * complete checksum manually on outgoing path. 3206 */ 3207 int skb_checksum_help(struct sk_buff *skb) 3208 { 3209 __wsum csum; 3210 int ret = 0, offset; 3211 3212 if (skb->ip_summed == CHECKSUM_COMPLETE) 3213 goto out_set_summed; 3214 3215 if (unlikely(skb_is_gso(skb))) { 3216 skb_warn_bad_offload(skb); 3217 return -EINVAL; 3218 } 3219 3220 /* Before computing a checksum, we should make sure no frag could 3221 * be modified by an external entity : checksum could be wrong. 3222 */ 3223 if (skb_has_shared_frag(skb)) { 3224 ret = __skb_linearize(skb); 3225 if (ret) 3226 goto out; 3227 } 3228 3229 offset = skb_checksum_start_offset(skb); 3230 BUG_ON(offset >= skb_headlen(skb)); 3231 csum = skb_checksum(skb, offset, skb->len - offset, 0); 3232 3233 offset += skb->csum_offset; 3234 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); 3235 3236 ret = skb_ensure_writable(skb, offset + sizeof(__sum16)); 3237 if (ret) 3238 goto out; 3239 3240 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0; 3241 out_set_summed: 3242 skb->ip_summed = CHECKSUM_NONE; 3243 out: 3244 return ret; 3245 } 3246 EXPORT_SYMBOL(skb_checksum_help); 3247 3248 int skb_crc32c_csum_help(struct sk_buff *skb) 3249 { 3250 __le32 crc32c_csum; 3251 int ret = 0, offset, start; 3252 3253 if (skb->ip_summed != CHECKSUM_PARTIAL) 3254 goto out; 3255 3256 if (unlikely(skb_is_gso(skb))) 3257 goto out; 3258 3259 /* Before computing a checksum, we should make sure no frag could 3260 * be modified by an external entity : checksum could be wrong. 3261 */ 3262 if (unlikely(skb_has_shared_frag(skb))) { 3263 ret = __skb_linearize(skb); 3264 if (ret) 3265 goto out; 3266 } 3267 start = skb_checksum_start_offset(skb); 3268 offset = start + offsetof(struct sctphdr, checksum); 3269 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { 3270 ret = -EINVAL; 3271 goto out; 3272 } 3273 3274 ret = skb_ensure_writable(skb, offset + sizeof(__le32)); 3275 if (ret) 3276 goto out; 3277 3278 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start, 3279 skb->len - start, ~(__u32)0, 3280 crc32c_csum_stub)); 3281 *(__le32 *)(skb->data + offset) = crc32c_csum; 3282 skb->ip_summed = CHECKSUM_NONE; 3283 skb->csum_not_inet = 0; 3284 out: 3285 return ret; 3286 } 3287 3288 __be16 skb_network_protocol(struct sk_buff *skb, int *depth) 3289 { 3290 __be16 type = skb->protocol; 3291 3292 /* Tunnel gso handlers can set protocol to ethernet. */ 3293 if (type == htons(ETH_P_TEB)) { 3294 struct ethhdr *eth; 3295 3296 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) 3297 return 0; 3298 3299 eth = (struct ethhdr *)skb->data; 3300 type = eth->h_proto; 3301 } 3302 3303 return __vlan_get_protocol(skb, type, depth); 3304 } 3305 3306 /** 3307 * skb_mac_gso_segment - mac layer segmentation handler. 3308 * @skb: buffer to segment 3309 * @features: features for the output path (see dev->features) 3310 */ 3311 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb, 3312 netdev_features_t features) 3313 { 3314 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); 3315 struct packet_offload *ptype; 3316 int vlan_depth = skb->mac_len; 3317 __be16 type = skb_network_protocol(skb, &vlan_depth); 3318 3319 if (unlikely(!type)) 3320 return ERR_PTR(-EINVAL); 3321 3322 __skb_pull(skb, vlan_depth); 3323 3324 rcu_read_lock(); 3325 list_for_each_entry_rcu(ptype, &offload_base, list) { 3326 if (ptype->type == type && ptype->callbacks.gso_segment) { 3327 segs = ptype->callbacks.gso_segment(skb, features); 3328 break; 3329 } 3330 } 3331 rcu_read_unlock(); 3332 3333 __skb_push(skb, skb->data - skb_mac_header(skb)); 3334 3335 return segs; 3336 } 3337 EXPORT_SYMBOL(skb_mac_gso_segment); 3338 3339 3340 /* openvswitch calls this on rx path, so we need a different check. 3341 */ 3342 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path) 3343 { 3344 if (tx_path) 3345 return skb->ip_summed != CHECKSUM_PARTIAL && 3346 skb->ip_summed != CHECKSUM_UNNECESSARY; 3347 3348 return skb->ip_summed == CHECKSUM_NONE; 3349 } 3350 3351 /** 3352 * __skb_gso_segment - Perform segmentation on skb. 3353 * @skb: buffer to segment 3354 * @features: features for the output path (see dev->features) 3355 * @tx_path: whether it is called in TX path 3356 * 3357 * This function segments the given skb and returns a list of segments. 3358 * 3359 * It may return NULL if the skb requires no segmentation. This is 3360 * only possible when GSO is used for verifying header integrity. 3361 * 3362 * Segmentation preserves SKB_GSO_CB_OFFSET bytes of previous skb cb. 3363 */ 3364 struct sk_buff *__skb_gso_segment(struct sk_buff *skb, 3365 netdev_features_t features, bool tx_path) 3366 { 3367 struct sk_buff *segs; 3368 3369 if (unlikely(skb_needs_check(skb, tx_path))) { 3370 int err; 3371 3372 /* We're going to init ->check field in TCP or UDP header */ 3373 err = skb_cow_head(skb, 0); 3374 if (err < 0) 3375 return ERR_PTR(err); 3376 } 3377 3378 /* Only report GSO partial support if it will enable us to 3379 * support segmentation on this frame without needing additional 3380 * work. 3381 */ 3382 if (features & NETIF_F_GSO_PARTIAL) { 3383 netdev_features_t partial_features = NETIF_F_GSO_ROBUST; 3384 struct net_device *dev = skb->dev; 3385 3386 partial_features |= dev->features & dev->gso_partial_features; 3387 if (!skb_gso_ok(skb, features | partial_features)) 3388 features &= ~NETIF_F_GSO_PARTIAL; 3389 } 3390 3391 BUILD_BUG_ON(SKB_GSO_CB_OFFSET + 3392 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb)); 3393 3394 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb); 3395 SKB_GSO_CB(skb)->encap_level = 0; 3396 3397 skb_reset_mac_header(skb); 3398 skb_reset_mac_len(skb); 3399 3400 segs = skb_mac_gso_segment(skb, features); 3401 3402 if (segs != skb && unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs))) 3403 skb_warn_bad_offload(skb); 3404 3405 return segs; 3406 } 3407 EXPORT_SYMBOL(__skb_gso_segment); 3408 3409 /* Take action when hardware reception checksum errors are detected. */ 3410 #ifdef CONFIG_BUG 3411 static void do_netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3412 { 3413 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>"); 3414 skb_dump(KERN_ERR, skb, true); 3415 dump_stack(); 3416 } 3417 3418 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3419 { 3420 DO_ONCE_LITE(do_netdev_rx_csum_fault, dev, skb); 3421 } 3422 EXPORT_SYMBOL(netdev_rx_csum_fault); 3423 #endif 3424 3425 /* XXX: check that highmem exists at all on the given machine. */ 3426 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) 3427 { 3428 #ifdef CONFIG_HIGHMEM 3429 int i; 3430 3431 if (!(dev->features & NETIF_F_HIGHDMA)) { 3432 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 3433 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 3434 3435 if (PageHighMem(skb_frag_page(frag))) 3436 return 1; 3437 } 3438 } 3439 #endif 3440 return 0; 3441 } 3442 3443 /* If MPLS offload request, verify we are testing hardware MPLS features 3444 * instead of standard features for the netdev. 3445 */ 3446 #if IS_ENABLED(CONFIG_NET_MPLS_GSO) 3447 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3448 netdev_features_t features, 3449 __be16 type) 3450 { 3451 if (eth_p_mpls(type)) 3452 features &= skb->dev->mpls_features; 3453 3454 return features; 3455 } 3456 #else 3457 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3458 netdev_features_t features, 3459 __be16 type) 3460 { 3461 return features; 3462 } 3463 #endif 3464 3465 static netdev_features_t harmonize_features(struct sk_buff *skb, 3466 netdev_features_t features) 3467 { 3468 __be16 type; 3469 3470 type = skb_network_protocol(skb, NULL); 3471 features = net_mpls_features(skb, features, type); 3472 3473 if (skb->ip_summed != CHECKSUM_NONE && 3474 !can_checksum_protocol(features, type)) { 3475 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); 3476 } 3477 if (illegal_highdma(skb->dev, skb)) 3478 features &= ~NETIF_F_SG; 3479 3480 return features; 3481 } 3482 3483 netdev_features_t passthru_features_check(struct sk_buff *skb, 3484 struct net_device *dev, 3485 netdev_features_t features) 3486 { 3487 return features; 3488 } 3489 EXPORT_SYMBOL(passthru_features_check); 3490 3491 static netdev_features_t dflt_features_check(struct sk_buff *skb, 3492 struct net_device *dev, 3493 netdev_features_t features) 3494 { 3495 return vlan_features_check(skb, features); 3496 } 3497 3498 static netdev_features_t gso_features_check(const struct sk_buff *skb, 3499 struct net_device *dev, 3500 netdev_features_t features) 3501 { 3502 u16 gso_segs = skb_shinfo(skb)->gso_segs; 3503 3504 if (gso_segs > dev->gso_max_segs) 3505 return features & ~NETIF_F_GSO_MASK; 3506 3507 if (!skb_shinfo(skb)->gso_type) { 3508 skb_warn_bad_offload(skb); 3509 return features & ~NETIF_F_GSO_MASK; 3510 } 3511 3512 /* Support for GSO partial features requires software 3513 * intervention before we can actually process the packets 3514 * so we need to strip support for any partial features now 3515 * and we can pull them back in after we have partially 3516 * segmented the frame. 3517 */ 3518 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)) 3519 features &= ~dev->gso_partial_features; 3520 3521 /* Make sure to clear the IPv4 ID mangling feature if the 3522 * IPv4 header has the potential to be fragmented. 3523 */ 3524 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { 3525 struct iphdr *iph = skb->encapsulation ? 3526 inner_ip_hdr(skb) : ip_hdr(skb); 3527 3528 if (!(iph->frag_off & htons(IP_DF))) 3529 features &= ~NETIF_F_TSO_MANGLEID; 3530 } 3531 3532 return features; 3533 } 3534 3535 netdev_features_t netif_skb_features(struct sk_buff *skb) 3536 { 3537 struct net_device *dev = skb->dev; 3538 netdev_features_t features = dev->features; 3539 3540 if (skb_is_gso(skb)) 3541 features = gso_features_check(skb, dev, features); 3542 3543 /* If encapsulation offload request, verify we are testing 3544 * hardware encapsulation features instead of standard 3545 * features for the netdev 3546 */ 3547 if (skb->encapsulation) 3548 features &= dev->hw_enc_features; 3549 3550 if (skb_vlan_tagged(skb)) 3551 features = netdev_intersect_features(features, 3552 dev->vlan_features | 3553 NETIF_F_HW_VLAN_CTAG_TX | 3554 NETIF_F_HW_VLAN_STAG_TX); 3555 3556 if (dev->netdev_ops->ndo_features_check) 3557 features &= dev->netdev_ops->ndo_features_check(skb, dev, 3558 features); 3559 else 3560 features &= dflt_features_check(skb, dev, features); 3561 3562 return harmonize_features(skb, features); 3563 } 3564 EXPORT_SYMBOL(netif_skb_features); 3565 3566 static int xmit_one(struct sk_buff *skb, struct net_device *dev, 3567 struct netdev_queue *txq, bool more) 3568 { 3569 unsigned int len; 3570 int rc; 3571 3572 if (dev_nit_active(dev)) 3573 dev_queue_xmit_nit(skb, dev); 3574 3575 len = skb->len; 3576 PRANDOM_ADD_NOISE(skb, dev, txq, len + jiffies); 3577 trace_net_dev_start_xmit(skb, dev); 3578 rc = netdev_start_xmit(skb, dev, txq, more); 3579 trace_net_dev_xmit(skb, rc, dev, len); 3580 3581 return rc; 3582 } 3583 3584 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, 3585 struct netdev_queue *txq, int *ret) 3586 { 3587 struct sk_buff *skb = first; 3588 int rc = NETDEV_TX_OK; 3589 3590 while (skb) { 3591 struct sk_buff *next = skb->next; 3592 3593 skb_mark_not_on_list(skb); 3594 rc = xmit_one(skb, dev, txq, next != NULL); 3595 if (unlikely(!dev_xmit_complete(rc))) { 3596 skb->next = next; 3597 goto out; 3598 } 3599 3600 skb = next; 3601 if (netif_tx_queue_stopped(txq) && skb) { 3602 rc = NETDEV_TX_BUSY; 3603 break; 3604 } 3605 } 3606 3607 out: 3608 *ret = rc; 3609 return skb; 3610 } 3611 3612 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, 3613 netdev_features_t features) 3614 { 3615 if (skb_vlan_tag_present(skb) && 3616 !vlan_hw_offload_capable(features, skb->vlan_proto)) 3617 skb = __vlan_hwaccel_push_inside(skb); 3618 return skb; 3619 } 3620 3621 int skb_csum_hwoffload_help(struct sk_buff *skb, 3622 const netdev_features_t features) 3623 { 3624 if (unlikely(skb_csum_is_sctp(skb))) 3625 return !!(features & NETIF_F_SCTP_CRC) ? 0 : 3626 skb_crc32c_csum_help(skb); 3627 3628 if (features & NETIF_F_HW_CSUM) 3629 return 0; 3630 3631 if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) { 3632 switch (skb->csum_offset) { 3633 case offsetof(struct tcphdr, check): 3634 case offsetof(struct udphdr, check): 3635 return 0; 3636 } 3637 } 3638 3639 return skb_checksum_help(skb); 3640 } 3641 EXPORT_SYMBOL(skb_csum_hwoffload_help); 3642 3643 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again) 3644 { 3645 netdev_features_t features; 3646 3647 features = netif_skb_features(skb); 3648 skb = validate_xmit_vlan(skb, features); 3649 if (unlikely(!skb)) 3650 goto out_null; 3651 3652 skb = sk_validate_xmit_skb(skb, dev); 3653 if (unlikely(!skb)) 3654 goto out_null; 3655 3656 if (netif_needs_gso(skb, features)) { 3657 struct sk_buff *segs; 3658 3659 segs = skb_gso_segment(skb, features); 3660 if (IS_ERR(segs)) { 3661 goto out_kfree_skb; 3662 } else if (segs) { 3663 consume_skb(skb); 3664 skb = segs; 3665 } 3666 } else { 3667 if (skb_needs_linearize(skb, features) && 3668 __skb_linearize(skb)) 3669 goto out_kfree_skb; 3670 3671 /* If packet is not checksummed and device does not 3672 * support checksumming for this protocol, complete 3673 * checksumming here. 3674 */ 3675 if (skb->ip_summed == CHECKSUM_PARTIAL) { 3676 if (skb->encapsulation) 3677 skb_set_inner_transport_header(skb, 3678 skb_checksum_start_offset(skb)); 3679 else 3680 skb_set_transport_header(skb, 3681 skb_checksum_start_offset(skb)); 3682 if (skb_csum_hwoffload_help(skb, features)) 3683 goto out_kfree_skb; 3684 } 3685 } 3686 3687 skb = validate_xmit_xfrm(skb, features, again); 3688 3689 return skb; 3690 3691 out_kfree_skb: 3692 kfree_skb(skb); 3693 out_null: 3694 atomic_long_inc(&dev->tx_dropped); 3695 return NULL; 3696 } 3697 3698 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again) 3699 { 3700 struct sk_buff *next, *head = NULL, *tail; 3701 3702 for (; skb != NULL; skb = next) { 3703 next = skb->next; 3704 skb_mark_not_on_list(skb); 3705 3706 /* in case skb wont be segmented, point to itself */ 3707 skb->prev = skb; 3708 3709 skb = validate_xmit_skb(skb, dev, again); 3710 if (!skb) 3711 continue; 3712 3713 if (!head) 3714 head = skb; 3715 else 3716 tail->next = skb; 3717 /* If skb was segmented, skb->prev points to 3718 * the last segment. If not, it still contains skb. 3719 */ 3720 tail = skb->prev; 3721 } 3722 return head; 3723 } 3724 EXPORT_SYMBOL_GPL(validate_xmit_skb_list); 3725 3726 static void qdisc_pkt_len_init(struct sk_buff *skb) 3727 { 3728 const struct skb_shared_info *shinfo = skb_shinfo(skb); 3729 3730 qdisc_skb_cb(skb)->pkt_len = skb->len; 3731 3732 /* To get more precise estimation of bytes sent on wire, 3733 * we add to pkt_len the headers size of all segments 3734 */ 3735 if (shinfo->gso_size && skb_transport_header_was_set(skb)) { 3736 unsigned int hdr_len; 3737 u16 gso_segs = shinfo->gso_segs; 3738 3739 /* mac layer + network layer */ 3740 hdr_len = skb_transport_header(skb) - skb_mac_header(skb); 3741 3742 /* + transport layer */ 3743 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { 3744 const struct tcphdr *th; 3745 struct tcphdr _tcphdr; 3746 3747 th = skb_header_pointer(skb, skb_transport_offset(skb), 3748 sizeof(_tcphdr), &_tcphdr); 3749 if (likely(th)) 3750 hdr_len += __tcp_hdrlen(th); 3751 } else { 3752 struct udphdr _udphdr; 3753 3754 if (skb_header_pointer(skb, skb_transport_offset(skb), 3755 sizeof(_udphdr), &_udphdr)) 3756 hdr_len += sizeof(struct udphdr); 3757 } 3758 3759 if (shinfo->gso_type & SKB_GSO_DODGY) 3760 gso_segs = DIV_ROUND_UP(skb->len - hdr_len, 3761 shinfo->gso_size); 3762 3763 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; 3764 } 3765 } 3766 3767 static int dev_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *q, 3768 struct sk_buff **to_free, 3769 struct netdev_queue *txq) 3770 { 3771 int rc; 3772 3773 rc = q->enqueue(skb, q, to_free) & NET_XMIT_MASK; 3774 if (rc == NET_XMIT_SUCCESS) 3775 trace_qdisc_enqueue(q, txq, skb); 3776 return rc; 3777 } 3778 3779 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, 3780 struct net_device *dev, 3781 struct netdev_queue *txq) 3782 { 3783 spinlock_t *root_lock = qdisc_lock(q); 3784 struct sk_buff *to_free = NULL; 3785 bool contended; 3786 int rc; 3787 3788 qdisc_calculate_pkt_len(skb, q); 3789 3790 if (q->flags & TCQ_F_NOLOCK) { 3791 if (q->flags & TCQ_F_CAN_BYPASS && nolock_qdisc_is_empty(q) && 3792 qdisc_run_begin(q)) { 3793 /* Retest nolock_qdisc_is_empty() within the protection 3794 * of q->seqlock to protect from racing with requeuing. 3795 */ 3796 if (unlikely(!nolock_qdisc_is_empty(q))) { 3797 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3798 __qdisc_run(q); 3799 qdisc_run_end(q); 3800 3801 goto no_lock_out; 3802 } 3803 3804 qdisc_bstats_cpu_update(q, skb); 3805 if (sch_direct_xmit(skb, q, dev, txq, NULL, true) && 3806 !nolock_qdisc_is_empty(q)) 3807 __qdisc_run(q); 3808 3809 qdisc_run_end(q); 3810 return NET_XMIT_SUCCESS; 3811 } 3812 3813 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3814 qdisc_run(q); 3815 3816 no_lock_out: 3817 if (unlikely(to_free)) 3818 kfree_skb_list(to_free); 3819 return rc; 3820 } 3821 3822 /* 3823 * Heuristic to force contended enqueues to serialize on a 3824 * separate lock before trying to get qdisc main lock. 3825 * This permits qdisc->running owner to get the lock more 3826 * often and dequeue packets faster. 3827 */ 3828 contended = qdisc_is_running(q); 3829 if (unlikely(contended)) 3830 spin_lock(&q->busylock); 3831 3832 spin_lock(root_lock); 3833 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { 3834 __qdisc_drop(skb, &to_free); 3835 rc = NET_XMIT_DROP; 3836 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && 3837 qdisc_run_begin(q)) { 3838 /* 3839 * This is a work-conserving queue; there are no old skbs 3840 * waiting to be sent out; and the qdisc is not running - 3841 * xmit the skb directly. 3842 */ 3843 3844 qdisc_bstats_update(q, skb); 3845 3846 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { 3847 if (unlikely(contended)) { 3848 spin_unlock(&q->busylock); 3849 contended = false; 3850 } 3851 __qdisc_run(q); 3852 } 3853 3854 qdisc_run_end(q); 3855 rc = NET_XMIT_SUCCESS; 3856 } else { 3857 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3858 if (qdisc_run_begin(q)) { 3859 if (unlikely(contended)) { 3860 spin_unlock(&q->busylock); 3861 contended = false; 3862 } 3863 __qdisc_run(q); 3864 qdisc_run_end(q); 3865 } 3866 } 3867 spin_unlock(root_lock); 3868 if (unlikely(to_free)) 3869 kfree_skb_list(to_free); 3870 if (unlikely(contended)) 3871 spin_unlock(&q->busylock); 3872 return rc; 3873 } 3874 3875 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) 3876 static void skb_update_prio(struct sk_buff *skb) 3877 { 3878 const struct netprio_map *map; 3879 const struct sock *sk; 3880 unsigned int prioidx; 3881 3882 if (skb->priority) 3883 return; 3884 map = rcu_dereference_bh(skb->dev->priomap); 3885 if (!map) 3886 return; 3887 sk = skb_to_full_sk(skb); 3888 if (!sk) 3889 return; 3890 3891 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data); 3892 3893 if (prioidx < map->priomap_len) 3894 skb->priority = map->priomap[prioidx]; 3895 } 3896 #else 3897 #define skb_update_prio(skb) 3898 #endif 3899 3900 /** 3901 * dev_loopback_xmit - loop back @skb 3902 * @net: network namespace this loopback is happening in 3903 * @sk: sk needed to be a netfilter okfn 3904 * @skb: buffer to transmit 3905 */ 3906 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) 3907 { 3908 skb_reset_mac_header(skb); 3909 __skb_pull(skb, skb_network_offset(skb)); 3910 skb->pkt_type = PACKET_LOOPBACK; 3911 skb->ip_summed = CHECKSUM_UNNECESSARY; 3912 WARN_ON(!skb_dst(skb)); 3913 skb_dst_force(skb); 3914 netif_rx_ni(skb); 3915 return 0; 3916 } 3917 EXPORT_SYMBOL(dev_loopback_xmit); 3918 3919 #ifdef CONFIG_NET_EGRESS 3920 static struct sk_buff * 3921 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 3922 { 3923 struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress); 3924 struct tcf_result cl_res; 3925 3926 if (!miniq) 3927 return skb; 3928 3929 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */ 3930 qdisc_skb_cb(skb)->mru = 0; 3931 qdisc_skb_cb(skb)->post_ct = false; 3932 mini_qdisc_bstats_cpu_update(miniq, skb); 3933 3934 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) { 3935 case TC_ACT_OK: 3936 case TC_ACT_RECLASSIFY: 3937 skb->tc_index = TC_H_MIN(cl_res.classid); 3938 break; 3939 case TC_ACT_SHOT: 3940 mini_qdisc_qstats_cpu_drop(miniq); 3941 *ret = NET_XMIT_DROP; 3942 kfree_skb(skb); 3943 return NULL; 3944 case TC_ACT_STOLEN: 3945 case TC_ACT_QUEUED: 3946 case TC_ACT_TRAP: 3947 *ret = NET_XMIT_SUCCESS; 3948 consume_skb(skb); 3949 return NULL; 3950 case TC_ACT_REDIRECT: 3951 /* No need to push/pop skb's mac_header here on egress! */ 3952 skb_do_redirect(skb); 3953 *ret = NET_XMIT_SUCCESS; 3954 return NULL; 3955 default: 3956 break; 3957 } 3958 3959 return skb; 3960 } 3961 #endif /* CONFIG_NET_EGRESS */ 3962 3963 #ifdef CONFIG_XPS 3964 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb, 3965 struct xps_dev_maps *dev_maps, unsigned int tci) 3966 { 3967 int tc = netdev_get_prio_tc_map(dev, skb->priority); 3968 struct xps_map *map; 3969 int queue_index = -1; 3970 3971 if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids) 3972 return queue_index; 3973 3974 tci *= dev_maps->num_tc; 3975 tci += tc; 3976 3977 map = rcu_dereference(dev_maps->attr_map[tci]); 3978 if (map) { 3979 if (map->len == 1) 3980 queue_index = map->queues[0]; 3981 else 3982 queue_index = map->queues[reciprocal_scale( 3983 skb_get_hash(skb), map->len)]; 3984 if (unlikely(queue_index >= dev->real_num_tx_queues)) 3985 queue_index = -1; 3986 } 3987 return queue_index; 3988 } 3989 #endif 3990 3991 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev, 3992 struct sk_buff *skb) 3993 { 3994 #ifdef CONFIG_XPS 3995 struct xps_dev_maps *dev_maps; 3996 struct sock *sk = skb->sk; 3997 int queue_index = -1; 3998 3999 if (!static_key_false(&xps_needed)) 4000 return -1; 4001 4002 rcu_read_lock(); 4003 if (!static_key_false(&xps_rxqs_needed)) 4004 goto get_cpus_map; 4005 4006 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]); 4007 if (dev_maps) { 4008 int tci = sk_rx_queue_get(sk); 4009 4010 if (tci >= 0) 4011 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4012 tci); 4013 } 4014 4015 get_cpus_map: 4016 if (queue_index < 0) { 4017 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]); 4018 if (dev_maps) { 4019 unsigned int tci = skb->sender_cpu - 1; 4020 4021 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4022 tci); 4023 } 4024 } 4025 rcu_read_unlock(); 4026 4027 return queue_index; 4028 #else 4029 return -1; 4030 #endif 4031 } 4032 4033 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb, 4034 struct net_device *sb_dev) 4035 { 4036 return 0; 4037 } 4038 EXPORT_SYMBOL(dev_pick_tx_zero); 4039 4040 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb, 4041 struct net_device *sb_dev) 4042 { 4043 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues; 4044 } 4045 EXPORT_SYMBOL(dev_pick_tx_cpu_id); 4046 4047 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb, 4048 struct net_device *sb_dev) 4049 { 4050 struct sock *sk = skb->sk; 4051 int queue_index = sk_tx_queue_get(sk); 4052 4053 sb_dev = sb_dev ? : dev; 4054 4055 if (queue_index < 0 || skb->ooo_okay || 4056 queue_index >= dev->real_num_tx_queues) { 4057 int new_index = get_xps_queue(dev, sb_dev, skb); 4058 4059 if (new_index < 0) 4060 new_index = skb_tx_hash(dev, sb_dev, skb); 4061 4062 if (queue_index != new_index && sk && 4063 sk_fullsock(sk) && 4064 rcu_access_pointer(sk->sk_dst_cache)) 4065 sk_tx_queue_set(sk, new_index); 4066 4067 queue_index = new_index; 4068 } 4069 4070 return queue_index; 4071 } 4072 EXPORT_SYMBOL(netdev_pick_tx); 4073 4074 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev, 4075 struct sk_buff *skb, 4076 struct net_device *sb_dev) 4077 { 4078 int queue_index = 0; 4079 4080 #ifdef CONFIG_XPS 4081 u32 sender_cpu = skb->sender_cpu - 1; 4082 4083 if (sender_cpu >= (u32)NR_CPUS) 4084 skb->sender_cpu = raw_smp_processor_id() + 1; 4085 #endif 4086 4087 if (dev->real_num_tx_queues != 1) { 4088 const struct net_device_ops *ops = dev->netdev_ops; 4089 4090 if (ops->ndo_select_queue) 4091 queue_index = ops->ndo_select_queue(dev, skb, sb_dev); 4092 else 4093 queue_index = netdev_pick_tx(dev, skb, sb_dev); 4094 4095 queue_index = netdev_cap_txqueue(dev, queue_index); 4096 } 4097 4098 skb_set_queue_mapping(skb, queue_index); 4099 return netdev_get_tx_queue(dev, queue_index); 4100 } 4101 4102 /** 4103 * __dev_queue_xmit - transmit a buffer 4104 * @skb: buffer to transmit 4105 * @sb_dev: suboordinate device used for L2 forwarding offload 4106 * 4107 * Queue a buffer for transmission to a network device. The caller must 4108 * have set the device and priority and built the buffer before calling 4109 * this function. The function can be called from an interrupt. 4110 * 4111 * A negative errno code is returned on a failure. A success does not 4112 * guarantee the frame will be transmitted as it may be dropped due 4113 * to congestion or traffic shaping. 4114 * 4115 * ----------------------------------------------------------------------------------- 4116 * I notice this method can also return errors from the queue disciplines, 4117 * including NET_XMIT_DROP, which is a positive value. So, errors can also 4118 * be positive. 4119 * 4120 * Regardless of the return value, the skb is consumed, so it is currently 4121 * difficult to retry a send to this method. (You can bump the ref count 4122 * before sending to hold a reference for retry if you are careful.) 4123 * 4124 * When calling this method, interrupts MUST be enabled. This is because 4125 * the BH enable code must have IRQs enabled so that it will not deadlock. 4126 * --BLG 4127 */ 4128 static int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev) 4129 { 4130 struct net_device *dev = skb->dev; 4131 struct netdev_queue *txq; 4132 struct Qdisc *q; 4133 int rc = -ENOMEM; 4134 bool again = false; 4135 4136 skb_reset_mac_header(skb); 4137 4138 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) 4139 __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED); 4140 4141 /* Disable soft irqs for various locks below. Also 4142 * stops preemption for RCU. 4143 */ 4144 rcu_read_lock_bh(); 4145 4146 skb_update_prio(skb); 4147 4148 qdisc_pkt_len_init(skb); 4149 #ifdef CONFIG_NET_CLS_ACT 4150 skb->tc_at_ingress = 0; 4151 # ifdef CONFIG_NET_EGRESS 4152 if (static_branch_unlikely(&egress_needed_key)) { 4153 skb = sch_handle_egress(skb, &rc, dev); 4154 if (!skb) 4155 goto out; 4156 } 4157 # endif 4158 #endif 4159 /* If device/qdisc don't need skb->dst, release it right now while 4160 * its hot in this cpu cache. 4161 */ 4162 if (dev->priv_flags & IFF_XMIT_DST_RELEASE) 4163 skb_dst_drop(skb); 4164 else 4165 skb_dst_force(skb); 4166 4167 txq = netdev_core_pick_tx(dev, skb, sb_dev); 4168 q = rcu_dereference_bh(txq->qdisc); 4169 4170 trace_net_dev_queue(skb); 4171 if (q->enqueue) { 4172 rc = __dev_xmit_skb(skb, q, dev, txq); 4173 goto out; 4174 } 4175 4176 /* The device has no queue. Common case for software devices: 4177 * loopback, all the sorts of tunnels... 4178 4179 * Really, it is unlikely that netif_tx_lock protection is necessary 4180 * here. (f.e. loopback and IP tunnels are clean ignoring statistics 4181 * counters.) 4182 * However, it is possible, that they rely on protection 4183 * made by us here. 4184 4185 * Check this and shot the lock. It is not prone from deadlocks. 4186 *Either shot noqueue qdisc, it is even simpler 8) 4187 */ 4188 if (dev->flags & IFF_UP) { 4189 int cpu = smp_processor_id(); /* ok because BHs are off */ 4190 4191 if (txq->xmit_lock_owner != cpu) { 4192 if (dev_xmit_recursion()) 4193 goto recursion_alert; 4194 4195 skb = validate_xmit_skb(skb, dev, &again); 4196 if (!skb) 4197 goto out; 4198 4199 PRANDOM_ADD_NOISE(skb, dev, txq, jiffies); 4200 HARD_TX_LOCK(dev, txq, cpu); 4201 4202 if (!netif_xmit_stopped(txq)) { 4203 dev_xmit_recursion_inc(); 4204 skb = dev_hard_start_xmit(skb, dev, txq, &rc); 4205 dev_xmit_recursion_dec(); 4206 if (dev_xmit_complete(rc)) { 4207 HARD_TX_UNLOCK(dev, txq); 4208 goto out; 4209 } 4210 } 4211 HARD_TX_UNLOCK(dev, txq); 4212 net_crit_ratelimited("Virtual device %s asks to queue packet!\n", 4213 dev->name); 4214 } else { 4215 /* Recursion is detected! It is possible, 4216 * unfortunately 4217 */ 4218 recursion_alert: 4219 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", 4220 dev->name); 4221 } 4222 } 4223 4224 rc = -ENETDOWN; 4225 rcu_read_unlock_bh(); 4226 4227 atomic_long_inc(&dev->tx_dropped); 4228 kfree_skb_list(skb); 4229 return rc; 4230 out: 4231 rcu_read_unlock_bh(); 4232 return rc; 4233 } 4234 4235 int dev_queue_xmit(struct sk_buff *skb) 4236 { 4237 return __dev_queue_xmit(skb, NULL); 4238 } 4239 EXPORT_SYMBOL(dev_queue_xmit); 4240 4241 int dev_queue_xmit_accel(struct sk_buff *skb, struct net_device *sb_dev) 4242 { 4243 return __dev_queue_xmit(skb, sb_dev); 4244 } 4245 EXPORT_SYMBOL(dev_queue_xmit_accel); 4246 4247 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id) 4248 { 4249 struct net_device *dev = skb->dev; 4250 struct sk_buff *orig_skb = skb; 4251 struct netdev_queue *txq; 4252 int ret = NETDEV_TX_BUSY; 4253 bool again = false; 4254 4255 if (unlikely(!netif_running(dev) || 4256 !netif_carrier_ok(dev))) 4257 goto drop; 4258 4259 skb = validate_xmit_skb_list(skb, dev, &again); 4260 if (skb != orig_skb) 4261 goto drop; 4262 4263 skb_set_queue_mapping(skb, queue_id); 4264 txq = skb_get_tx_queue(dev, skb); 4265 PRANDOM_ADD_NOISE(skb, dev, txq, jiffies); 4266 4267 local_bh_disable(); 4268 4269 dev_xmit_recursion_inc(); 4270 HARD_TX_LOCK(dev, txq, smp_processor_id()); 4271 if (!netif_xmit_frozen_or_drv_stopped(txq)) 4272 ret = netdev_start_xmit(skb, dev, txq, false); 4273 HARD_TX_UNLOCK(dev, txq); 4274 dev_xmit_recursion_dec(); 4275 4276 local_bh_enable(); 4277 return ret; 4278 drop: 4279 atomic_long_inc(&dev->tx_dropped); 4280 kfree_skb_list(skb); 4281 return NET_XMIT_DROP; 4282 } 4283 EXPORT_SYMBOL(__dev_direct_xmit); 4284 4285 /************************************************************************* 4286 * Receiver routines 4287 *************************************************************************/ 4288 4289 int netdev_max_backlog __read_mostly = 1000; 4290 EXPORT_SYMBOL(netdev_max_backlog); 4291 4292 int netdev_tstamp_prequeue __read_mostly = 1; 4293 int netdev_budget __read_mostly = 300; 4294 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */ 4295 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ; 4296 int weight_p __read_mostly = 64; /* old backlog weight */ 4297 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ 4298 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ 4299 int dev_rx_weight __read_mostly = 64; 4300 int dev_tx_weight __read_mostly = 64; 4301 /* Maximum number of GRO_NORMAL skbs to batch up for list-RX */ 4302 int gro_normal_batch __read_mostly = 8; 4303 4304 /* Called with irq disabled */ 4305 static inline void ____napi_schedule(struct softnet_data *sd, 4306 struct napi_struct *napi) 4307 { 4308 struct task_struct *thread; 4309 4310 if (test_bit(NAPI_STATE_THREADED, &napi->state)) { 4311 /* Paired with smp_mb__before_atomic() in 4312 * napi_enable()/dev_set_threaded(). 4313 * Use READ_ONCE() to guarantee a complete 4314 * read on napi->thread. Only call 4315 * wake_up_process() when it's not NULL. 4316 */ 4317 thread = READ_ONCE(napi->thread); 4318 if (thread) { 4319 /* Avoid doing set_bit() if the thread is in 4320 * INTERRUPTIBLE state, cause napi_thread_wait() 4321 * makes sure to proceed with napi polling 4322 * if the thread is explicitly woken from here. 4323 */ 4324 if (READ_ONCE(thread->__state) != TASK_INTERRUPTIBLE) 4325 set_bit(NAPI_STATE_SCHED_THREADED, &napi->state); 4326 wake_up_process(thread); 4327 return; 4328 } 4329 } 4330 4331 list_add_tail(&napi->poll_list, &sd->poll_list); 4332 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4333 } 4334 4335 #ifdef CONFIG_RPS 4336 4337 /* One global table that all flow-based protocols share. */ 4338 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; 4339 EXPORT_SYMBOL(rps_sock_flow_table); 4340 u32 rps_cpu_mask __read_mostly; 4341 EXPORT_SYMBOL(rps_cpu_mask); 4342 4343 struct static_key_false rps_needed __read_mostly; 4344 EXPORT_SYMBOL(rps_needed); 4345 struct static_key_false rfs_needed __read_mostly; 4346 EXPORT_SYMBOL(rfs_needed); 4347 4348 static struct rps_dev_flow * 4349 set_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4350 struct rps_dev_flow *rflow, u16 next_cpu) 4351 { 4352 if (next_cpu < nr_cpu_ids) { 4353 #ifdef CONFIG_RFS_ACCEL 4354 struct netdev_rx_queue *rxqueue; 4355 struct rps_dev_flow_table *flow_table; 4356 struct rps_dev_flow *old_rflow; 4357 u32 flow_id; 4358 u16 rxq_index; 4359 int rc; 4360 4361 /* Should we steer this flow to a different hardware queue? */ 4362 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || 4363 !(dev->features & NETIF_F_NTUPLE)) 4364 goto out; 4365 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); 4366 if (rxq_index == skb_get_rx_queue(skb)) 4367 goto out; 4368 4369 rxqueue = dev->_rx + rxq_index; 4370 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4371 if (!flow_table) 4372 goto out; 4373 flow_id = skb_get_hash(skb) & flow_table->mask; 4374 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, 4375 rxq_index, flow_id); 4376 if (rc < 0) 4377 goto out; 4378 old_rflow = rflow; 4379 rflow = &flow_table->flows[flow_id]; 4380 rflow->filter = rc; 4381 if (old_rflow->filter == rflow->filter) 4382 old_rflow->filter = RPS_NO_FILTER; 4383 out: 4384 #endif 4385 rflow->last_qtail = 4386 per_cpu(softnet_data, next_cpu).input_queue_head; 4387 } 4388 4389 rflow->cpu = next_cpu; 4390 return rflow; 4391 } 4392 4393 /* 4394 * get_rps_cpu is called from netif_receive_skb and returns the target 4395 * CPU from the RPS map of the receiving queue for a given skb. 4396 * rcu_read_lock must be held on entry. 4397 */ 4398 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4399 struct rps_dev_flow **rflowp) 4400 { 4401 const struct rps_sock_flow_table *sock_flow_table; 4402 struct netdev_rx_queue *rxqueue = dev->_rx; 4403 struct rps_dev_flow_table *flow_table; 4404 struct rps_map *map; 4405 int cpu = -1; 4406 u32 tcpu; 4407 u32 hash; 4408 4409 if (skb_rx_queue_recorded(skb)) { 4410 u16 index = skb_get_rx_queue(skb); 4411 4412 if (unlikely(index >= dev->real_num_rx_queues)) { 4413 WARN_ONCE(dev->real_num_rx_queues > 1, 4414 "%s received packet on queue %u, but number " 4415 "of RX queues is %u\n", 4416 dev->name, index, dev->real_num_rx_queues); 4417 goto done; 4418 } 4419 rxqueue += index; 4420 } 4421 4422 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ 4423 4424 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4425 map = rcu_dereference(rxqueue->rps_map); 4426 if (!flow_table && !map) 4427 goto done; 4428 4429 skb_reset_network_header(skb); 4430 hash = skb_get_hash(skb); 4431 if (!hash) 4432 goto done; 4433 4434 sock_flow_table = rcu_dereference(rps_sock_flow_table); 4435 if (flow_table && sock_flow_table) { 4436 struct rps_dev_flow *rflow; 4437 u32 next_cpu; 4438 u32 ident; 4439 4440 /* First check into global flow table if there is a match */ 4441 ident = sock_flow_table->ents[hash & sock_flow_table->mask]; 4442 if ((ident ^ hash) & ~rps_cpu_mask) 4443 goto try_rps; 4444 4445 next_cpu = ident & rps_cpu_mask; 4446 4447 /* OK, now we know there is a match, 4448 * we can look at the local (per receive queue) flow table 4449 */ 4450 rflow = &flow_table->flows[hash & flow_table->mask]; 4451 tcpu = rflow->cpu; 4452 4453 /* 4454 * If the desired CPU (where last recvmsg was done) is 4455 * different from current CPU (one in the rx-queue flow 4456 * table entry), switch if one of the following holds: 4457 * - Current CPU is unset (>= nr_cpu_ids). 4458 * - Current CPU is offline. 4459 * - The current CPU's queue tail has advanced beyond the 4460 * last packet that was enqueued using this table entry. 4461 * This guarantees that all previous packets for the flow 4462 * have been dequeued, thus preserving in order delivery. 4463 */ 4464 if (unlikely(tcpu != next_cpu) && 4465 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || 4466 ((int)(per_cpu(softnet_data, tcpu).input_queue_head - 4467 rflow->last_qtail)) >= 0)) { 4468 tcpu = next_cpu; 4469 rflow = set_rps_cpu(dev, skb, rflow, next_cpu); 4470 } 4471 4472 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { 4473 *rflowp = rflow; 4474 cpu = tcpu; 4475 goto done; 4476 } 4477 } 4478 4479 try_rps: 4480 4481 if (map) { 4482 tcpu = map->cpus[reciprocal_scale(hash, map->len)]; 4483 if (cpu_online(tcpu)) { 4484 cpu = tcpu; 4485 goto done; 4486 } 4487 } 4488 4489 done: 4490 return cpu; 4491 } 4492 4493 #ifdef CONFIG_RFS_ACCEL 4494 4495 /** 4496 * rps_may_expire_flow - check whether an RFS hardware filter may be removed 4497 * @dev: Device on which the filter was set 4498 * @rxq_index: RX queue index 4499 * @flow_id: Flow ID passed to ndo_rx_flow_steer() 4500 * @filter_id: Filter ID returned by ndo_rx_flow_steer() 4501 * 4502 * Drivers that implement ndo_rx_flow_steer() should periodically call 4503 * this function for each installed filter and remove the filters for 4504 * which it returns %true. 4505 */ 4506 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, 4507 u32 flow_id, u16 filter_id) 4508 { 4509 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; 4510 struct rps_dev_flow_table *flow_table; 4511 struct rps_dev_flow *rflow; 4512 bool expire = true; 4513 unsigned int cpu; 4514 4515 rcu_read_lock(); 4516 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4517 if (flow_table && flow_id <= flow_table->mask) { 4518 rflow = &flow_table->flows[flow_id]; 4519 cpu = READ_ONCE(rflow->cpu); 4520 if (rflow->filter == filter_id && cpu < nr_cpu_ids && 4521 ((int)(per_cpu(softnet_data, cpu).input_queue_head - 4522 rflow->last_qtail) < 4523 (int)(10 * flow_table->mask))) 4524 expire = false; 4525 } 4526 rcu_read_unlock(); 4527 return expire; 4528 } 4529 EXPORT_SYMBOL(rps_may_expire_flow); 4530 4531 #endif /* CONFIG_RFS_ACCEL */ 4532 4533 /* Called from hardirq (IPI) context */ 4534 static void rps_trigger_softirq(void *data) 4535 { 4536 struct softnet_data *sd = data; 4537 4538 ____napi_schedule(sd, &sd->backlog); 4539 sd->received_rps++; 4540 } 4541 4542 #endif /* CONFIG_RPS */ 4543 4544 /* 4545 * Check if this softnet_data structure is another cpu one 4546 * If yes, queue it to our IPI list and return 1 4547 * If no, return 0 4548 */ 4549 static int rps_ipi_queued(struct softnet_data *sd) 4550 { 4551 #ifdef CONFIG_RPS 4552 struct softnet_data *mysd = this_cpu_ptr(&softnet_data); 4553 4554 if (sd != mysd) { 4555 sd->rps_ipi_next = mysd->rps_ipi_list; 4556 mysd->rps_ipi_list = sd; 4557 4558 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4559 return 1; 4560 } 4561 #endif /* CONFIG_RPS */ 4562 return 0; 4563 } 4564 4565 #ifdef CONFIG_NET_FLOW_LIMIT 4566 int netdev_flow_limit_table_len __read_mostly = (1 << 12); 4567 #endif 4568 4569 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) 4570 { 4571 #ifdef CONFIG_NET_FLOW_LIMIT 4572 struct sd_flow_limit *fl; 4573 struct softnet_data *sd; 4574 unsigned int old_flow, new_flow; 4575 4576 if (qlen < (netdev_max_backlog >> 1)) 4577 return false; 4578 4579 sd = this_cpu_ptr(&softnet_data); 4580 4581 rcu_read_lock(); 4582 fl = rcu_dereference(sd->flow_limit); 4583 if (fl) { 4584 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); 4585 old_flow = fl->history[fl->history_head]; 4586 fl->history[fl->history_head] = new_flow; 4587 4588 fl->history_head++; 4589 fl->history_head &= FLOW_LIMIT_HISTORY - 1; 4590 4591 if (likely(fl->buckets[old_flow])) 4592 fl->buckets[old_flow]--; 4593 4594 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { 4595 fl->count++; 4596 rcu_read_unlock(); 4597 return true; 4598 } 4599 } 4600 rcu_read_unlock(); 4601 #endif 4602 return false; 4603 } 4604 4605 /* 4606 * enqueue_to_backlog is called to queue an skb to a per CPU backlog 4607 * queue (may be a remote CPU queue). 4608 */ 4609 static int enqueue_to_backlog(struct sk_buff *skb, int cpu, 4610 unsigned int *qtail) 4611 { 4612 struct softnet_data *sd; 4613 unsigned long flags; 4614 unsigned int qlen; 4615 4616 sd = &per_cpu(softnet_data, cpu); 4617 4618 local_irq_save(flags); 4619 4620 rps_lock(sd); 4621 if (!netif_running(skb->dev)) 4622 goto drop; 4623 qlen = skb_queue_len(&sd->input_pkt_queue); 4624 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) { 4625 if (qlen) { 4626 enqueue: 4627 __skb_queue_tail(&sd->input_pkt_queue, skb); 4628 input_queue_tail_incr_save(sd, qtail); 4629 rps_unlock(sd); 4630 local_irq_restore(flags); 4631 return NET_RX_SUCCESS; 4632 } 4633 4634 /* Schedule NAPI for backlog device 4635 * We can use non atomic operation since we own the queue lock 4636 */ 4637 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) { 4638 if (!rps_ipi_queued(sd)) 4639 ____napi_schedule(sd, &sd->backlog); 4640 } 4641 goto enqueue; 4642 } 4643 4644 drop: 4645 sd->dropped++; 4646 rps_unlock(sd); 4647 4648 local_irq_restore(flags); 4649 4650 atomic_long_inc(&skb->dev->rx_dropped); 4651 kfree_skb(skb); 4652 return NET_RX_DROP; 4653 } 4654 4655 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb) 4656 { 4657 struct net_device *dev = skb->dev; 4658 struct netdev_rx_queue *rxqueue; 4659 4660 rxqueue = dev->_rx; 4661 4662 if (skb_rx_queue_recorded(skb)) { 4663 u16 index = skb_get_rx_queue(skb); 4664 4665 if (unlikely(index >= dev->real_num_rx_queues)) { 4666 WARN_ONCE(dev->real_num_rx_queues > 1, 4667 "%s received packet on queue %u, but number " 4668 "of RX queues is %u\n", 4669 dev->name, index, dev->real_num_rx_queues); 4670 4671 return rxqueue; /* Return first rxqueue */ 4672 } 4673 rxqueue += index; 4674 } 4675 return rxqueue; 4676 } 4677 4678 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp, 4679 struct bpf_prog *xdp_prog) 4680 { 4681 void *orig_data, *orig_data_end, *hard_start; 4682 struct netdev_rx_queue *rxqueue; 4683 bool orig_bcast, orig_host; 4684 u32 mac_len, frame_sz; 4685 __be16 orig_eth_type; 4686 struct ethhdr *eth; 4687 u32 metalen, act; 4688 int off; 4689 4690 /* The XDP program wants to see the packet starting at the MAC 4691 * header. 4692 */ 4693 mac_len = skb->data - skb_mac_header(skb); 4694 hard_start = skb->data - skb_headroom(skb); 4695 4696 /* SKB "head" area always have tailroom for skb_shared_info */ 4697 frame_sz = (void *)skb_end_pointer(skb) - hard_start; 4698 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 4699 4700 rxqueue = netif_get_rxqueue(skb); 4701 xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq); 4702 xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len, 4703 skb_headlen(skb) + mac_len, true); 4704 4705 orig_data_end = xdp->data_end; 4706 orig_data = xdp->data; 4707 eth = (struct ethhdr *)xdp->data; 4708 orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr); 4709 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest); 4710 orig_eth_type = eth->h_proto; 4711 4712 act = bpf_prog_run_xdp(xdp_prog, xdp); 4713 4714 /* check if bpf_xdp_adjust_head was used */ 4715 off = xdp->data - orig_data; 4716 if (off) { 4717 if (off > 0) 4718 __skb_pull(skb, off); 4719 else if (off < 0) 4720 __skb_push(skb, -off); 4721 4722 skb->mac_header += off; 4723 skb_reset_network_header(skb); 4724 } 4725 4726 /* check if bpf_xdp_adjust_tail was used */ 4727 off = xdp->data_end - orig_data_end; 4728 if (off != 0) { 4729 skb_set_tail_pointer(skb, xdp->data_end - xdp->data); 4730 skb->len += off; /* positive on grow, negative on shrink */ 4731 } 4732 4733 /* check if XDP changed eth hdr such SKB needs update */ 4734 eth = (struct ethhdr *)xdp->data; 4735 if ((orig_eth_type != eth->h_proto) || 4736 (orig_host != ether_addr_equal_64bits(eth->h_dest, 4737 skb->dev->dev_addr)) || 4738 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) { 4739 __skb_push(skb, ETH_HLEN); 4740 skb->pkt_type = PACKET_HOST; 4741 skb->protocol = eth_type_trans(skb, skb->dev); 4742 } 4743 4744 /* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull 4745 * before calling us again on redirect path. We do not call do_redirect 4746 * as we leave that up to the caller. 4747 * 4748 * Caller is responsible for managing lifetime of skb (i.e. calling 4749 * kfree_skb in response to actions it cannot handle/XDP_DROP). 4750 */ 4751 switch (act) { 4752 case XDP_REDIRECT: 4753 case XDP_TX: 4754 __skb_push(skb, mac_len); 4755 break; 4756 case XDP_PASS: 4757 metalen = xdp->data - xdp->data_meta; 4758 if (metalen) 4759 skb_metadata_set(skb, metalen); 4760 break; 4761 } 4762 4763 return act; 4764 } 4765 4766 static u32 netif_receive_generic_xdp(struct sk_buff *skb, 4767 struct xdp_buff *xdp, 4768 struct bpf_prog *xdp_prog) 4769 { 4770 u32 act = XDP_DROP; 4771 4772 /* Reinjected packets coming from act_mirred or similar should 4773 * not get XDP generic processing. 4774 */ 4775 if (skb_is_redirected(skb)) 4776 return XDP_PASS; 4777 4778 /* XDP packets must be linear and must have sufficient headroom 4779 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also 4780 * native XDP provides, thus we need to do it here as well. 4781 */ 4782 if (skb_cloned(skb) || skb_is_nonlinear(skb) || 4783 skb_headroom(skb) < XDP_PACKET_HEADROOM) { 4784 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb); 4785 int troom = skb->tail + skb->data_len - skb->end; 4786 4787 /* In case we have to go down the path and also linearize, 4788 * then lets do the pskb_expand_head() work just once here. 4789 */ 4790 if (pskb_expand_head(skb, 4791 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0, 4792 troom > 0 ? troom + 128 : 0, GFP_ATOMIC)) 4793 goto do_drop; 4794 if (skb_linearize(skb)) 4795 goto do_drop; 4796 } 4797 4798 act = bpf_prog_run_generic_xdp(skb, xdp, xdp_prog); 4799 switch (act) { 4800 case XDP_REDIRECT: 4801 case XDP_TX: 4802 case XDP_PASS: 4803 break; 4804 default: 4805 bpf_warn_invalid_xdp_action(act); 4806 fallthrough; 4807 case XDP_ABORTED: 4808 trace_xdp_exception(skb->dev, xdp_prog, act); 4809 fallthrough; 4810 case XDP_DROP: 4811 do_drop: 4812 kfree_skb(skb); 4813 break; 4814 } 4815 4816 return act; 4817 } 4818 4819 /* When doing generic XDP we have to bypass the qdisc layer and the 4820 * network taps in order to match in-driver-XDP behavior. 4821 */ 4822 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog) 4823 { 4824 struct net_device *dev = skb->dev; 4825 struct netdev_queue *txq; 4826 bool free_skb = true; 4827 int cpu, rc; 4828 4829 txq = netdev_core_pick_tx(dev, skb, NULL); 4830 cpu = smp_processor_id(); 4831 HARD_TX_LOCK(dev, txq, cpu); 4832 if (!netif_xmit_stopped(txq)) { 4833 rc = netdev_start_xmit(skb, dev, txq, 0); 4834 if (dev_xmit_complete(rc)) 4835 free_skb = false; 4836 } 4837 HARD_TX_UNLOCK(dev, txq); 4838 if (free_skb) { 4839 trace_xdp_exception(dev, xdp_prog, XDP_TX); 4840 kfree_skb(skb); 4841 } 4842 } 4843 4844 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key); 4845 4846 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb) 4847 { 4848 if (xdp_prog) { 4849 struct xdp_buff xdp; 4850 u32 act; 4851 int err; 4852 4853 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog); 4854 if (act != XDP_PASS) { 4855 switch (act) { 4856 case XDP_REDIRECT: 4857 err = xdp_do_generic_redirect(skb->dev, skb, 4858 &xdp, xdp_prog); 4859 if (err) 4860 goto out_redir; 4861 break; 4862 case XDP_TX: 4863 generic_xdp_tx(skb, xdp_prog); 4864 break; 4865 } 4866 return XDP_DROP; 4867 } 4868 } 4869 return XDP_PASS; 4870 out_redir: 4871 kfree_skb(skb); 4872 return XDP_DROP; 4873 } 4874 EXPORT_SYMBOL_GPL(do_xdp_generic); 4875 4876 static int netif_rx_internal(struct sk_buff *skb) 4877 { 4878 int ret; 4879 4880 net_timestamp_check(netdev_tstamp_prequeue, skb); 4881 4882 trace_netif_rx(skb); 4883 4884 #ifdef CONFIG_RPS 4885 if (static_branch_unlikely(&rps_needed)) { 4886 struct rps_dev_flow voidflow, *rflow = &voidflow; 4887 int cpu; 4888 4889 preempt_disable(); 4890 rcu_read_lock(); 4891 4892 cpu = get_rps_cpu(skb->dev, skb, &rflow); 4893 if (cpu < 0) 4894 cpu = smp_processor_id(); 4895 4896 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 4897 4898 rcu_read_unlock(); 4899 preempt_enable(); 4900 } else 4901 #endif 4902 { 4903 unsigned int qtail; 4904 4905 ret = enqueue_to_backlog(skb, get_cpu(), &qtail); 4906 put_cpu(); 4907 } 4908 return ret; 4909 } 4910 4911 /** 4912 * netif_rx - post buffer to the network code 4913 * @skb: buffer to post 4914 * 4915 * This function receives a packet from a device driver and queues it for 4916 * the upper (protocol) levels to process. It always succeeds. The buffer 4917 * may be dropped during processing for congestion control or by the 4918 * protocol layers. 4919 * 4920 * return values: 4921 * NET_RX_SUCCESS (no congestion) 4922 * NET_RX_DROP (packet was dropped) 4923 * 4924 */ 4925 4926 int netif_rx(struct sk_buff *skb) 4927 { 4928 int ret; 4929 4930 trace_netif_rx_entry(skb); 4931 4932 ret = netif_rx_internal(skb); 4933 trace_netif_rx_exit(ret); 4934 4935 return ret; 4936 } 4937 EXPORT_SYMBOL(netif_rx); 4938 4939 int netif_rx_ni(struct sk_buff *skb) 4940 { 4941 int err; 4942 4943 trace_netif_rx_ni_entry(skb); 4944 4945 preempt_disable(); 4946 err = netif_rx_internal(skb); 4947 if (local_softirq_pending()) 4948 do_softirq(); 4949 preempt_enable(); 4950 trace_netif_rx_ni_exit(err); 4951 4952 return err; 4953 } 4954 EXPORT_SYMBOL(netif_rx_ni); 4955 4956 int netif_rx_any_context(struct sk_buff *skb) 4957 { 4958 /* 4959 * If invoked from contexts which do not invoke bottom half 4960 * processing either at return from interrupt or when softrqs are 4961 * reenabled, use netif_rx_ni() which invokes bottomhalf processing 4962 * directly. 4963 */ 4964 if (in_interrupt()) 4965 return netif_rx(skb); 4966 else 4967 return netif_rx_ni(skb); 4968 } 4969 EXPORT_SYMBOL(netif_rx_any_context); 4970 4971 static __latent_entropy void net_tx_action(struct softirq_action *h) 4972 { 4973 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 4974 4975 if (sd->completion_queue) { 4976 struct sk_buff *clist; 4977 4978 local_irq_disable(); 4979 clist = sd->completion_queue; 4980 sd->completion_queue = NULL; 4981 local_irq_enable(); 4982 4983 while (clist) { 4984 struct sk_buff *skb = clist; 4985 4986 clist = clist->next; 4987 4988 WARN_ON(refcount_read(&skb->users)); 4989 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED)) 4990 trace_consume_skb(skb); 4991 else 4992 trace_kfree_skb(skb, net_tx_action); 4993 4994 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) 4995 __kfree_skb(skb); 4996 else 4997 __kfree_skb_defer(skb); 4998 } 4999 } 5000 5001 if (sd->output_queue) { 5002 struct Qdisc *head; 5003 5004 local_irq_disable(); 5005 head = sd->output_queue; 5006 sd->output_queue = NULL; 5007 sd->output_queue_tailp = &sd->output_queue; 5008 local_irq_enable(); 5009 5010 rcu_read_lock(); 5011 5012 while (head) { 5013 struct Qdisc *q = head; 5014 spinlock_t *root_lock = NULL; 5015 5016 head = head->next_sched; 5017 5018 /* We need to make sure head->next_sched is read 5019 * before clearing __QDISC_STATE_SCHED 5020 */ 5021 smp_mb__before_atomic(); 5022 5023 if (!(q->flags & TCQ_F_NOLOCK)) { 5024 root_lock = qdisc_lock(q); 5025 spin_lock(root_lock); 5026 } else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, 5027 &q->state))) { 5028 /* There is a synchronize_net() between 5029 * STATE_DEACTIVATED flag being set and 5030 * qdisc_reset()/some_qdisc_is_busy() in 5031 * dev_deactivate(), so we can safely bail out 5032 * early here to avoid data race between 5033 * qdisc_deactivate() and some_qdisc_is_busy() 5034 * for lockless qdisc. 5035 */ 5036 clear_bit(__QDISC_STATE_SCHED, &q->state); 5037 continue; 5038 } 5039 5040 clear_bit(__QDISC_STATE_SCHED, &q->state); 5041 qdisc_run(q); 5042 if (root_lock) 5043 spin_unlock(root_lock); 5044 } 5045 5046 rcu_read_unlock(); 5047 } 5048 5049 xfrm_dev_backlog(sd); 5050 } 5051 5052 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) 5053 /* This hook is defined here for ATM LANE */ 5054 int (*br_fdb_test_addr_hook)(struct net_device *dev, 5055 unsigned char *addr) __read_mostly; 5056 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); 5057 #endif 5058 5059 static inline struct sk_buff * 5060 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 5061 struct net_device *orig_dev, bool *another) 5062 { 5063 #ifdef CONFIG_NET_CLS_ACT 5064 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress); 5065 struct tcf_result cl_res; 5066 5067 /* If there's at least one ingress present somewhere (so 5068 * we get here via enabled static key), remaining devices 5069 * that are not configured with an ingress qdisc will bail 5070 * out here. 5071 */ 5072 if (!miniq) 5073 return skb; 5074 5075 if (*pt_prev) { 5076 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5077 *pt_prev = NULL; 5078 } 5079 5080 qdisc_skb_cb(skb)->pkt_len = skb->len; 5081 qdisc_skb_cb(skb)->mru = 0; 5082 qdisc_skb_cb(skb)->post_ct = false; 5083 skb->tc_at_ingress = 1; 5084 mini_qdisc_bstats_cpu_update(miniq, skb); 5085 5086 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) { 5087 case TC_ACT_OK: 5088 case TC_ACT_RECLASSIFY: 5089 skb->tc_index = TC_H_MIN(cl_res.classid); 5090 break; 5091 case TC_ACT_SHOT: 5092 mini_qdisc_qstats_cpu_drop(miniq); 5093 kfree_skb(skb); 5094 return NULL; 5095 case TC_ACT_STOLEN: 5096 case TC_ACT_QUEUED: 5097 case TC_ACT_TRAP: 5098 consume_skb(skb); 5099 return NULL; 5100 case TC_ACT_REDIRECT: 5101 /* skb_mac_header check was done by cls/act_bpf, so 5102 * we can safely push the L2 header back before 5103 * redirecting to another netdev 5104 */ 5105 __skb_push(skb, skb->mac_len); 5106 if (skb_do_redirect(skb) == -EAGAIN) { 5107 __skb_pull(skb, skb->mac_len); 5108 *another = true; 5109 break; 5110 } 5111 return NULL; 5112 case TC_ACT_CONSUMED: 5113 return NULL; 5114 default: 5115 break; 5116 } 5117 #endif /* CONFIG_NET_CLS_ACT */ 5118 return skb; 5119 } 5120 5121 /** 5122 * netdev_is_rx_handler_busy - check if receive handler is registered 5123 * @dev: device to check 5124 * 5125 * Check if a receive handler is already registered for a given device. 5126 * Return true if there one. 5127 * 5128 * The caller must hold the rtnl_mutex. 5129 */ 5130 bool netdev_is_rx_handler_busy(struct net_device *dev) 5131 { 5132 ASSERT_RTNL(); 5133 return dev && rtnl_dereference(dev->rx_handler); 5134 } 5135 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); 5136 5137 /** 5138 * netdev_rx_handler_register - register receive handler 5139 * @dev: device to register a handler for 5140 * @rx_handler: receive handler to register 5141 * @rx_handler_data: data pointer that is used by rx handler 5142 * 5143 * Register a receive handler for a device. This handler will then be 5144 * called from __netif_receive_skb. A negative errno code is returned 5145 * on a failure. 5146 * 5147 * The caller must hold the rtnl_mutex. 5148 * 5149 * For a general description of rx_handler, see enum rx_handler_result. 5150 */ 5151 int netdev_rx_handler_register(struct net_device *dev, 5152 rx_handler_func_t *rx_handler, 5153 void *rx_handler_data) 5154 { 5155 if (netdev_is_rx_handler_busy(dev)) 5156 return -EBUSY; 5157 5158 if (dev->priv_flags & IFF_NO_RX_HANDLER) 5159 return -EINVAL; 5160 5161 /* Note: rx_handler_data must be set before rx_handler */ 5162 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); 5163 rcu_assign_pointer(dev->rx_handler, rx_handler); 5164 5165 return 0; 5166 } 5167 EXPORT_SYMBOL_GPL(netdev_rx_handler_register); 5168 5169 /** 5170 * netdev_rx_handler_unregister - unregister receive handler 5171 * @dev: device to unregister a handler from 5172 * 5173 * Unregister a receive handler from a device. 5174 * 5175 * The caller must hold the rtnl_mutex. 5176 */ 5177 void netdev_rx_handler_unregister(struct net_device *dev) 5178 { 5179 5180 ASSERT_RTNL(); 5181 RCU_INIT_POINTER(dev->rx_handler, NULL); 5182 /* a reader seeing a non NULL rx_handler in a rcu_read_lock() 5183 * section has a guarantee to see a non NULL rx_handler_data 5184 * as well. 5185 */ 5186 synchronize_net(); 5187 RCU_INIT_POINTER(dev->rx_handler_data, NULL); 5188 } 5189 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); 5190 5191 /* 5192 * Limit the use of PFMEMALLOC reserves to those protocols that implement 5193 * the special handling of PFMEMALLOC skbs. 5194 */ 5195 static bool skb_pfmemalloc_protocol(struct sk_buff *skb) 5196 { 5197 switch (skb->protocol) { 5198 case htons(ETH_P_ARP): 5199 case htons(ETH_P_IP): 5200 case htons(ETH_P_IPV6): 5201 case htons(ETH_P_8021Q): 5202 case htons(ETH_P_8021AD): 5203 return true; 5204 default: 5205 return false; 5206 } 5207 } 5208 5209 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, 5210 int *ret, struct net_device *orig_dev) 5211 { 5212 if (nf_hook_ingress_active(skb)) { 5213 int ingress_retval; 5214 5215 if (*pt_prev) { 5216 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5217 *pt_prev = NULL; 5218 } 5219 5220 rcu_read_lock(); 5221 ingress_retval = nf_hook_ingress(skb); 5222 rcu_read_unlock(); 5223 return ingress_retval; 5224 } 5225 return 0; 5226 } 5227 5228 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc, 5229 struct packet_type **ppt_prev) 5230 { 5231 struct packet_type *ptype, *pt_prev; 5232 rx_handler_func_t *rx_handler; 5233 struct sk_buff *skb = *pskb; 5234 struct net_device *orig_dev; 5235 bool deliver_exact = false; 5236 int ret = NET_RX_DROP; 5237 __be16 type; 5238 5239 net_timestamp_check(!netdev_tstamp_prequeue, skb); 5240 5241 trace_netif_receive_skb(skb); 5242 5243 orig_dev = skb->dev; 5244 5245 skb_reset_network_header(skb); 5246 if (!skb_transport_header_was_set(skb)) 5247 skb_reset_transport_header(skb); 5248 skb_reset_mac_len(skb); 5249 5250 pt_prev = NULL; 5251 5252 another_round: 5253 skb->skb_iif = skb->dev->ifindex; 5254 5255 __this_cpu_inc(softnet_data.processed); 5256 5257 if (static_branch_unlikely(&generic_xdp_needed_key)) { 5258 int ret2; 5259 5260 migrate_disable(); 5261 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb); 5262 migrate_enable(); 5263 5264 if (ret2 != XDP_PASS) { 5265 ret = NET_RX_DROP; 5266 goto out; 5267 } 5268 } 5269 5270 if (eth_type_vlan(skb->protocol)) { 5271 skb = skb_vlan_untag(skb); 5272 if (unlikely(!skb)) 5273 goto out; 5274 } 5275 5276 if (skb_skip_tc_classify(skb)) 5277 goto skip_classify; 5278 5279 if (pfmemalloc) 5280 goto skip_taps; 5281 5282 list_for_each_entry_rcu(ptype, &ptype_all, list) { 5283 if (pt_prev) 5284 ret = deliver_skb(skb, pt_prev, orig_dev); 5285 pt_prev = ptype; 5286 } 5287 5288 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { 5289 if (pt_prev) 5290 ret = deliver_skb(skb, pt_prev, orig_dev); 5291 pt_prev = ptype; 5292 } 5293 5294 skip_taps: 5295 #ifdef CONFIG_NET_INGRESS 5296 if (static_branch_unlikely(&ingress_needed_key)) { 5297 bool another = false; 5298 5299 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev, 5300 &another); 5301 if (another) 5302 goto another_round; 5303 if (!skb) 5304 goto out; 5305 5306 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) 5307 goto out; 5308 } 5309 #endif 5310 skb_reset_redirect(skb); 5311 skip_classify: 5312 if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) 5313 goto drop; 5314 5315 if (skb_vlan_tag_present(skb)) { 5316 if (pt_prev) { 5317 ret = deliver_skb(skb, pt_prev, orig_dev); 5318 pt_prev = NULL; 5319 } 5320 if (vlan_do_receive(&skb)) 5321 goto another_round; 5322 else if (unlikely(!skb)) 5323 goto out; 5324 } 5325 5326 rx_handler = rcu_dereference(skb->dev->rx_handler); 5327 if (rx_handler) { 5328 if (pt_prev) { 5329 ret = deliver_skb(skb, pt_prev, orig_dev); 5330 pt_prev = NULL; 5331 } 5332 switch (rx_handler(&skb)) { 5333 case RX_HANDLER_CONSUMED: 5334 ret = NET_RX_SUCCESS; 5335 goto out; 5336 case RX_HANDLER_ANOTHER: 5337 goto another_round; 5338 case RX_HANDLER_EXACT: 5339 deliver_exact = true; 5340 break; 5341 case RX_HANDLER_PASS: 5342 break; 5343 default: 5344 BUG(); 5345 } 5346 } 5347 5348 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) { 5349 check_vlan_id: 5350 if (skb_vlan_tag_get_id(skb)) { 5351 /* Vlan id is non 0 and vlan_do_receive() above couldn't 5352 * find vlan device. 5353 */ 5354 skb->pkt_type = PACKET_OTHERHOST; 5355 } else if (eth_type_vlan(skb->protocol)) { 5356 /* Outer header is 802.1P with vlan 0, inner header is 5357 * 802.1Q or 802.1AD and vlan_do_receive() above could 5358 * not find vlan dev for vlan id 0. 5359 */ 5360 __vlan_hwaccel_clear_tag(skb); 5361 skb = skb_vlan_untag(skb); 5362 if (unlikely(!skb)) 5363 goto out; 5364 if (vlan_do_receive(&skb)) 5365 /* After stripping off 802.1P header with vlan 0 5366 * vlan dev is found for inner header. 5367 */ 5368 goto another_round; 5369 else if (unlikely(!skb)) 5370 goto out; 5371 else 5372 /* We have stripped outer 802.1P vlan 0 header. 5373 * But could not find vlan dev. 5374 * check again for vlan id to set OTHERHOST. 5375 */ 5376 goto check_vlan_id; 5377 } 5378 /* Note: we might in the future use prio bits 5379 * and set skb->priority like in vlan_do_receive() 5380 * For the time being, just ignore Priority Code Point 5381 */ 5382 __vlan_hwaccel_clear_tag(skb); 5383 } 5384 5385 type = skb->protocol; 5386 5387 /* deliver only exact match when indicated */ 5388 if (likely(!deliver_exact)) { 5389 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5390 &ptype_base[ntohs(type) & 5391 PTYPE_HASH_MASK]); 5392 } 5393 5394 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5395 &orig_dev->ptype_specific); 5396 5397 if (unlikely(skb->dev != orig_dev)) { 5398 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5399 &skb->dev->ptype_specific); 5400 } 5401 5402 if (pt_prev) { 5403 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 5404 goto drop; 5405 *ppt_prev = pt_prev; 5406 } else { 5407 drop: 5408 if (!deliver_exact) 5409 atomic_long_inc(&skb->dev->rx_dropped); 5410 else 5411 atomic_long_inc(&skb->dev->rx_nohandler); 5412 kfree_skb(skb); 5413 /* Jamal, now you will not able to escape explaining 5414 * me how you were going to use this. :-) 5415 */ 5416 ret = NET_RX_DROP; 5417 } 5418 5419 out: 5420 /* The invariant here is that if *ppt_prev is not NULL 5421 * then skb should also be non-NULL. 5422 * 5423 * Apparently *ppt_prev assignment above holds this invariant due to 5424 * skb dereferencing near it. 5425 */ 5426 *pskb = skb; 5427 return ret; 5428 } 5429 5430 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc) 5431 { 5432 struct net_device *orig_dev = skb->dev; 5433 struct packet_type *pt_prev = NULL; 5434 int ret; 5435 5436 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5437 if (pt_prev) 5438 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb, 5439 skb->dev, pt_prev, orig_dev); 5440 return ret; 5441 } 5442 5443 /** 5444 * netif_receive_skb_core - special purpose version of netif_receive_skb 5445 * @skb: buffer to process 5446 * 5447 * More direct receive version of netif_receive_skb(). It should 5448 * only be used by callers that have a need to skip RPS and Generic XDP. 5449 * Caller must also take care of handling if ``(page_is_)pfmemalloc``. 5450 * 5451 * This function may only be called from softirq context and interrupts 5452 * should be enabled. 5453 * 5454 * Return values (usually ignored): 5455 * NET_RX_SUCCESS: no congestion 5456 * NET_RX_DROP: packet was dropped 5457 */ 5458 int netif_receive_skb_core(struct sk_buff *skb) 5459 { 5460 int ret; 5461 5462 rcu_read_lock(); 5463 ret = __netif_receive_skb_one_core(skb, false); 5464 rcu_read_unlock(); 5465 5466 return ret; 5467 } 5468 EXPORT_SYMBOL(netif_receive_skb_core); 5469 5470 static inline void __netif_receive_skb_list_ptype(struct list_head *head, 5471 struct packet_type *pt_prev, 5472 struct net_device *orig_dev) 5473 { 5474 struct sk_buff *skb, *next; 5475 5476 if (!pt_prev) 5477 return; 5478 if (list_empty(head)) 5479 return; 5480 if (pt_prev->list_func != NULL) 5481 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv, 5482 ip_list_rcv, head, pt_prev, orig_dev); 5483 else 5484 list_for_each_entry_safe(skb, next, head, list) { 5485 skb_list_del_init(skb); 5486 pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 5487 } 5488 } 5489 5490 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc) 5491 { 5492 /* Fast-path assumptions: 5493 * - There is no RX handler. 5494 * - Only one packet_type matches. 5495 * If either of these fails, we will end up doing some per-packet 5496 * processing in-line, then handling the 'last ptype' for the whole 5497 * sublist. This can't cause out-of-order delivery to any single ptype, 5498 * because the 'last ptype' must be constant across the sublist, and all 5499 * other ptypes are handled per-packet. 5500 */ 5501 /* Current (common) ptype of sublist */ 5502 struct packet_type *pt_curr = NULL; 5503 /* Current (common) orig_dev of sublist */ 5504 struct net_device *od_curr = NULL; 5505 struct list_head sublist; 5506 struct sk_buff *skb, *next; 5507 5508 INIT_LIST_HEAD(&sublist); 5509 list_for_each_entry_safe(skb, next, head, list) { 5510 struct net_device *orig_dev = skb->dev; 5511 struct packet_type *pt_prev = NULL; 5512 5513 skb_list_del_init(skb); 5514 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5515 if (!pt_prev) 5516 continue; 5517 if (pt_curr != pt_prev || od_curr != orig_dev) { 5518 /* dispatch old sublist */ 5519 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5520 /* start new sublist */ 5521 INIT_LIST_HEAD(&sublist); 5522 pt_curr = pt_prev; 5523 od_curr = orig_dev; 5524 } 5525 list_add_tail(&skb->list, &sublist); 5526 } 5527 5528 /* dispatch final sublist */ 5529 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5530 } 5531 5532 static int __netif_receive_skb(struct sk_buff *skb) 5533 { 5534 int ret; 5535 5536 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { 5537 unsigned int noreclaim_flag; 5538 5539 /* 5540 * PFMEMALLOC skbs are special, they should 5541 * - be delivered to SOCK_MEMALLOC sockets only 5542 * - stay away from userspace 5543 * - have bounded memory usage 5544 * 5545 * Use PF_MEMALLOC as this saves us from propagating the allocation 5546 * context down to all allocation sites. 5547 */ 5548 noreclaim_flag = memalloc_noreclaim_save(); 5549 ret = __netif_receive_skb_one_core(skb, true); 5550 memalloc_noreclaim_restore(noreclaim_flag); 5551 } else 5552 ret = __netif_receive_skb_one_core(skb, false); 5553 5554 return ret; 5555 } 5556 5557 static void __netif_receive_skb_list(struct list_head *head) 5558 { 5559 unsigned long noreclaim_flag = 0; 5560 struct sk_buff *skb, *next; 5561 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */ 5562 5563 list_for_each_entry_safe(skb, next, head, list) { 5564 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) { 5565 struct list_head sublist; 5566 5567 /* Handle the previous sublist */ 5568 list_cut_before(&sublist, head, &skb->list); 5569 if (!list_empty(&sublist)) 5570 __netif_receive_skb_list_core(&sublist, pfmemalloc); 5571 pfmemalloc = !pfmemalloc; 5572 /* See comments in __netif_receive_skb */ 5573 if (pfmemalloc) 5574 noreclaim_flag = memalloc_noreclaim_save(); 5575 else 5576 memalloc_noreclaim_restore(noreclaim_flag); 5577 } 5578 } 5579 /* Handle the remaining sublist */ 5580 if (!list_empty(head)) 5581 __netif_receive_skb_list_core(head, pfmemalloc); 5582 /* Restore pflags */ 5583 if (pfmemalloc) 5584 memalloc_noreclaim_restore(noreclaim_flag); 5585 } 5586 5587 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp) 5588 { 5589 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog); 5590 struct bpf_prog *new = xdp->prog; 5591 int ret = 0; 5592 5593 switch (xdp->command) { 5594 case XDP_SETUP_PROG: 5595 rcu_assign_pointer(dev->xdp_prog, new); 5596 if (old) 5597 bpf_prog_put(old); 5598 5599 if (old && !new) { 5600 static_branch_dec(&generic_xdp_needed_key); 5601 } else if (new && !old) { 5602 static_branch_inc(&generic_xdp_needed_key); 5603 dev_disable_lro(dev); 5604 dev_disable_gro_hw(dev); 5605 } 5606 break; 5607 5608 default: 5609 ret = -EINVAL; 5610 break; 5611 } 5612 5613 return ret; 5614 } 5615 5616 static int netif_receive_skb_internal(struct sk_buff *skb) 5617 { 5618 int ret; 5619 5620 net_timestamp_check(netdev_tstamp_prequeue, skb); 5621 5622 if (skb_defer_rx_timestamp(skb)) 5623 return NET_RX_SUCCESS; 5624 5625 rcu_read_lock(); 5626 #ifdef CONFIG_RPS 5627 if (static_branch_unlikely(&rps_needed)) { 5628 struct rps_dev_flow voidflow, *rflow = &voidflow; 5629 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5630 5631 if (cpu >= 0) { 5632 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5633 rcu_read_unlock(); 5634 return ret; 5635 } 5636 } 5637 #endif 5638 ret = __netif_receive_skb(skb); 5639 rcu_read_unlock(); 5640 return ret; 5641 } 5642 5643 static void netif_receive_skb_list_internal(struct list_head *head) 5644 { 5645 struct sk_buff *skb, *next; 5646 struct list_head sublist; 5647 5648 INIT_LIST_HEAD(&sublist); 5649 list_for_each_entry_safe(skb, next, head, list) { 5650 net_timestamp_check(netdev_tstamp_prequeue, skb); 5651 skb_list_del_init(skb); 5652 if (!skb_defer_rx_timestamp(skb)) 5653 list_add_tail(&skb->list, &sublist); 5654 } 5655 list_splice_init(&sublist, head); 5656 5657 rcu_read_lock(); 5658 #ifdef CONFIG_RPS 5659 if (static_branch_unlikely(&rps_needed)) { 5660 list_for_each_entry_safe(skb, next, head, list) { 5661 struct rps_dev_flow voidflow, *rflow = &voidflow; 5662 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5663 5664 if (cpu >= 0) { 5665 /* Will be handled, remove from list */ 5666 skb_list_del_init(skb); 5667 enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5668 } 5669 } 5670 } 5671 #endif 5672 __netif_receive_skb_list(head); 5673 rcu_read_unlock(); 5674 } 5675 5676 /** 5677 * netif_receive_skb - process receive buffer from network 5678 * @skb: buffer to process 5679 * 5680 * netif_receive_skb() is the main receive data processing function. 5681 * It always succeeds. The buffer may be dropped during processing 5682 * for congestion control or by the protocol layers. 5683 * 5684 * This function may only be called from softirq context and interrupts 5685 * should be enabled. 5686 * 5687 * Return values (usually ignored): 5688 * NET_RX_SUCCESS: no congestion 5689 * NET_RX_DROP: packet was dropped 5690 */ 5691 int netif_receive_skb(struct sk_buff *skb) 5692 { 5693 int ret; 5694 5695 trace_netif_receive_skb_entry(skb); 5696 5697 ret = netif_receive_skb_internal(skb); 5698 trace_netif_receive_skb_exit(ret); 5699 5700 return ret; 5701 } 5702 EXPORT_SYMBOL(netif_receive_skb); 5703 5704 /** 5705 * netif_receive_skb_list - process many receive buffers from network 5706 * @head: list of skbs to process. 5707 * 5708 * Since return value of netif_receive_skb() is normally ignored, and 5709 * wouldn't be meaningful for a list, this function returns void. 5710 * 5711 * This function may only be called from softirq context and interrupts 5712 * should be enabled. 5713 */ 5714 void netif_receive_skb_list(struct list_head *head) 5715 { 5716 struct sk_buff *skb; 5717 5718 if (list_empty(head)) 5719 return; 5720 if (trace_netif_receive_skb_list_entry_enabled()) { 5721 list_for_each_entry(skb, head, list) 5722 trace_netif_receive_skb_list_entry(skb); 5723 } 5724 netif_receive_skb_list_internal(head); 5725 trace_netif_receive_skb_list_exit(0); 5726 } 5727 EXPORT_SYMBOL(netif_receive_skb_list); 5728 5729 static DEFINE_PER_CPU(struct work_struct, flush_works); 5730 5731 /* Network device is going away, flush any packets still pending */ 5732 static void flush_backlog(struct work_struct *work) 5733 { 5734 struct sk_buff *skb, *tmp; 5735 struct softnet_data *sd; 5736 5737 local_bh_disable(); 5738 sd = this_cpu_ptr(&softnet_data); 5739 5740 local_irq_disable(); 5741 rps_lock(sd); 5742 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { 5743 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5744 __skb_unlink(skb, &sd->input_pkt_queue); 5745 dev_kfree_skb_irq(skb); 5746 input_queue_head_incr(sd); 5747 } 5748 } 5749 rps_unlock(sd); 5750 local_irq_enable(); 5751 5752 skb_queue_walk_safe(&sd->process_queue, skb, tmp) { 5753 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5754 __skb_unlink(skb, &sd->process_queue); 5755 kfree_skb(skb); 5756 input_queue_head_incr(sd); 5757 } 5758 } 5759 local_bh_enable(); 5760 } 5761 5762 static bool flush_required(int cpu) 5763 { 5764 #if IS_ENABLED(CONFIG_RPS) 5765 struct softnet_data *sd = &per_cpu(softnet_data, cpu); 5766 bool do_flush; 5767 5768 local_irq_disable(); 5769 rps_lock(sd); 5770 5771 /* as insertion into process_queue happens with the rps lock held, 5772 * process_queue access may race only with dequeue 5773 */ 5774 do_flush = !skb_queue_empty(&sd->input_pkt_queue) || 5775 !skb_queue_empty_lockless(&sd->process_queue); 5776 rps_unlock(sd); 5777 local_irq_enable(); 5778 5779 return do_flush; 5780 #endif 5781 /* without RPS we can't safely check input_pkt_queue: during a 5782 * concurrent remote skb_queue_splice() we can detect as empty both 5783 * input_pkt_queue and process_queue even if the latter could end-up 5784 * containing a lot of packets. 5785 */ 5786 return true; 5787 } 5788 5789 static void flush_all_backlogs(void) 5790 { 5791 static cpumask_t flush_cpus; 5792 unsigned int cpu; 5793 5794 /* since we are under rtnl lock protection we can use static data 5795 * for the cpumask and avoid allocating on stack the possibly 5796 * large mask 5797 */ 5798 ASSERT_RTNL(); 5799 5800 cpus_read_lock(); 5801 5802 cpumask_clear(&flush_cpus); 5803 for_each_online_cpu(cpu) { 5804 if (flush_required(cpu)) { 5805 queue_work_on(cpu, system_highpri_wq, 5806 per_cpu_ptr(&flush_works, cpu)); 5807 cpumask_set_cpu(cpu, &flush_cpus); 5808 } 5809 } 5810 5811 /* we can have in flight packet[s] on the cpus we are not flushing, 5812 * synchronize_net() in unregister_netdevice_many() will take care of 5813 * them 5814 */ 5815 for_each_cpu(cpu, &flush_cpus) 5816 flush_work(per_cpu_ptr(&flush_works, cpu)); 5817 5818 cpus_read_unlock(); 5819 } 5820 5821 /* Pass the currently batched GRO_NORMAL SKBs up to the stack. */ 5822 static void gro_normal_list(struct napi_struct *napi) 5823 { 5824 if (!napi->rx_count) 5825 return; 5826 netif_receive_skb_list_internal(&napi->rx_list); 5827 INIT_LIST_HEAD(&napi->rx_list); 5828 napi->rx_count = 0; 5829 } 5830 5831 /* Queue one GRO_NORMAL SKB up for list processing. If batch size exceeded, 5832 * pass the whole batch up to the stack. 5833 */ 5834 static void gro_normal_one(struct napi_struct *napi, struct sk_buff *skb, int segs) 5835 { 5836 list_add_tail(&skb->list, &napi->rx_list); 5837 napi->rx_count += segs; 5838 if (napi->rx_count >= gro_normal_batch) 5839 gro_normal_list(napi); 5840 } 5841 5842 static void napi_gro_complete(struct napi_struct *napi, struct sk_buff *skb) 5843 { 5844 struct packet_offload *ptype; 5845 __be16 type = skb->protocol; 5846 struct list_head *head = &offload_base; 5847 int err = -ENOENT; 5848 5849 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb)); 5850 5851 if (NAPI_GRO_CB(skb)->count == 1) { 5852 skb_shinfo(skb)->gso_size = 0; 5853 goto out; 5854 } 5855 5856 rcu_read_lock(); 5857 list_for_each_entry_rcu(ptype, head, list) { 5858 if (ptype->type != type || !ptype->callbacks.gro_complete) 5859 continue; 5860 5861 err = INDIRECT_CALL_INET(ptype->callbacks.gro_complete, 5862 ipv6_gro_complete, inet_gro_complete, 5863 skb, 0); 5864 break; 5865 } 5866 rcu_read_unlock(); 5867 5868 if (err) { 5869 WARN_ON(&ptype->list == head); 5870 kfree_skb(skb); 5871 return; 5872 } 5873 5874 out: 5875 gro_normal_one(napi, skb, NAPI_GRO_CB(skb)->count); 5876 } 5877 5878 static void __napi_gro_flush_chain(struct napi_struct *napi, u32 index, 5879 bool flush_old) 5880 { 5881 struct list_head *head = &napi->gro_hash[index].list; 5882 struct sk_buff *skb, *p; 5883 5884 list_for_each_entry_safe_reverse(skb, p, head, list) { 5885 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies) 5886 return; 5887 skb_list_del_init(skb); 5888 napi_gro_complete(napi, skb); 5889 napi->gro_hash[index].count--; 5890 } 5891 5892 if (!napi->gro_hash[index].count) 5893 __clear_bit(index, &napi->gro_bitmask); 5894 } 5895 5896 /* napi->gro_hash[].list contains packets ordered by age. 5897 * youngest packets at the head of it. 5898 * Complete skbs in reverse order to reduce latencies. 5899 */ 5900 void napi_gro_flush(struct napi_struct *napi, bool flush_old) 5901 { 5902 unsigned long bitmask = napi->gro_bitmask; 5903 unsigned int i, base = ~0U; 5904 5905 while ((i = ffs(bitmask)) != 0) { 5906 bitmask >>= i; 5907 base += i; 5908 __napi_gro_flush_chain(napi, base, flush_old); 5909 } 5910 } 5911 EXPORT_SYMBOL(napi_gro_flush); 5912 5913 static void gro_list_prepare(const struct list_head *head, 5914 const struct sk_buff *skb) 5915 { 5916 unsigned int maclen = skb->dev->hard_header_len; 5917 u32 hash = skb_get_hash_raw(skb); 5918 struct sk_buff *p; 5919 5920 list_for_each_entry(p, head, list) { 5921 unsigned long diffs; 5922 5923 NAPI_GRO_CB(p)->flush = 0; 5924 5925 if (hash != skb_get_hash_raw(p)) { 5926 NAPI_GRO_CB(p)->same_flow = 0; 5927 continue; 5928 } 5929 5930 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev; 5931 diffs |= skb_vlan_tag_present(p) ^ skb_vlan_tag_present(skb); 5932 if (skb_vlan_tag_present(p)) 5933 diffs |= skb_vlan_tag_get(p) ^ skb_vlan_tag_get(skb); 5934 diffs |= skb_metadata_differs(p, skb); 5935 if (maclen == ETH_HLEN) 5936 diffs |= compare_ether_header(skb_mac_header(p), 5937 skb_mac_header(skb)); 5938 else if (!diffs) 5939 diffs = memcmp(skb_mac_header(p), 5940 skb_mac_header(skb), 5941 maclen); 5942 5943 /* in most common scenarions 'slow_gro' is 0 5944 * otherwise we are already on some slower paths 5945 * either skip all the infrequent tests altogether or 5946 * avoid trying too hard to skip each of them individually 5947 */ 5948 if (!diffs && unlikely(skb->slow_gro | p->slow_gro)) { 5949 #if IS_ENABLED(CONFIG_SKB_EXTENSIONS) && IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 5950 struct tc_skb_ext *skb_ext; 5951 struct tc_skb_ext *p_ext; 5952 #endif 5953 5954 diffs |= p->sk != skb->sk; 5955 diffs |= skb_metadata_dst_cmp(p, skb); 5956 diffs |= skb_get_nfct(p) ^ skb_get_nfct(skb); 5957 5958 #if IS_ENABLED(CONFIG_SKB_EXTENSIONS) && IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 5959 skb_ext = skb_ext_find(skb, TC_SKB_EXT); 5960 p_ext = skb_ext_find(p, TC_SKB_EXT); 5961 5962 diffs |= (!!p_ext) ^ (!!skb_ext); 5963 if (!diffs && unlikely(skb_ext)) 5964 diffs |= p_ext->chain ^ skb_ext->chain; 5965 #endif 5966 } 5967 5968 NAPI_GRO_CB(p)->same_flow = !diffs; 5969 } 5970 } 5971 5972 static inline void skb_gro_reset_offset(struct sk_buff *skb, u32 nhoff) 5973 { 5974 const struct skb_shared_info *pinfo = skb_shinfo(skb); 5975 const skb_frag_t *frag0 = &pinfo->frags[0]; 5976 5977 NAPI_GRO_CB(skb)->data_offset = 0; 5978 NAPI_GRO_CB(skb)->frag0 = NULL; 5979 NAPI_GRO_CB(skb)->frag0_len = 0; 5980 5981 if (!skb_headlen(skb) && pinfo->nr_frags && 5982 !PageHighMem(skb_frag_page(frag0)) && 5983 (!NET_IP_ALIGN || !((skb_frag_off(frag0) + nhoff) & 3))) { 5984 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0); 5985 NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int, 5986 skb_frag_size(frag0), 5987 skb->end - skb->tail); 5988 } 5989 } 5990 5991 static void gro_pull_from_frag0(struct sk_buff *skb, int grow) 5992 { 5993 struct skb_shared_info *pinfo = skb_shinfo(skb); 5994 5995 BUG_ON(skb->end - skb->tail < grow); 5996 5997 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); 5998 5999 skb->data_len -= grow; 6000 skb->tail += grow; 6001 6002 skb_frag_off_add(&pinfo->frags[0], grow); 6003 skb_frag_size_sub(&pinfo->frags[0], grow); 6004 6005 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) { 6006 skb_frag_unref(skb, 0); 6007 memmove(pinfo->frags, pinfo->frags + 1, 6008 --pinfo->nr_frags * sizeof(pinfo->frags[0])); 6009 } 6010 } 6011 6012 static void gro_flush_oldest(struct napi_struct *napi, struct list_head *head) 6013 { 6014 struct sk_buff *oldest; 6015 6016 oldest = list_last_entry(head, struct sk_buff, list); 6017 6018 /* We are called with head length >= MAX_GRO_SKBS, so this is 6019 * impossible. 6020 */ 6021 if (WARN_ON_ONCE(!oldest)) 6022 return; 6023 6024 /* Do not adjust napi->gro_hash[].count, caller is adding a new 6025 * SKB to the chain. 6026 */ 6027 skb_list_del_init(oldest); 6028 napi_gro_complete(napi, oldest); 6029 } 6030 6031 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) 6032 { 6033 u32 bucket = skb_get_hash_raw(skb) & (GRO_HASH_BUCKETS - 1); 6034 struct gro_list *gro_list = &napi->gro_hash[bucket]; 6035 struct list_head *head = &offload_base; 6036 struct packet_offload *ptype; 6037 __be16 type = skb->protocol; 6038 struct sk_buff *pp = NULL; 6039 enum gro_result ret; 6040 int same_flow; 6041 int grow; 6042 6043 if (netif_elide_gro(skb->dev)) 6044 goto normal; 6045 6046 gro_list_prepare(&gro_list->list, skb); 6047 6048 rcu_read_lock(); 6049 list_for_each_entry_rcu(ptype, head, list) { 6050 if (ptype->type != type || !ptype->callbacks.gro_receive) 6051 continue; 6052 6053 skb_set_network_header(skb, skb_gro_offset(skb)); 6054 skb_reset_mac_len(skb); 6055 NAPI_GRO_CB(skb)->same_flow = 0; 6056 NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb); 6057 NAPI_GRO_CB(skb)->free = 0; 6058 NAPI_GRO_CB(skb)->encap_mark = 0; 6059 NAPI_GRO_CB(skb)->recursion_counter = 0; 6060 NAPI_GRO_CB(skb)->is_fou = 0; 6061 NAPI_GRO_CB(skb)->is_atomic = 1; 6062 NAPI_GRO_CB(skb)->gro_remcsum_start = 0; 6063 6064 /* Setup for GRO checksum validation */ 6065 switch (skb->ip_summed) { 6066 case CHECKSUM_COMPLETE: 6067 NAPI_GRO_CB(skb)->csum = skb->csum; 6068 NAPI_GRO_CB(skb)->csum_valid = 1; 6069 NAPI_GRO_CB(skb)->csum_cnt = 0; 6070 break; 6071 case CHECKSUM_UNNECESSARY: 6072 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1; 6073 NAPI_GRO_CB(skb)->csum_valid = 0; 6074 break; 6075 default: 6076 NAPI_GRO_CB(skb)->csum_cnt = 0; 6077 NAPI_GRO_CB(skb)->csum_valid = 0; 6078 } 6079 6080 pp = INDIRECT_CALL_INET(ptype->callbacks.gro_receive, 6081 ipv6_gro_receive, inet_gro_receive, 6082 &gro_list->list, skb); 6083 break; 6084 } 6085 rcu_read_unlock(); 6086 6087 if (&ptype->list == head) 6088 goto normal; 6089 6090 if (PTR_ERR(pp) == -EINPROGRESS) { 6091 ret = GRO_CONSUMED; 6092 goto ok; 6093 } 6094 6095 same_flow = NAPI_GRO_CB(skb)->same_flow; 6096 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; 6097 6098 if (pp) { 6099 skb_list_del_init(pp); 6100 napi_gro_complete(napi, pp); 6101 gro_list->count--; 6102 } 6103 6104 if (same_flow) 6105 goto ok; 6106 6107 if (NAPI_GRO_CB(skb)->flush) 6108 goto normal; 6109 6110 if (unlikely(gro_list->count >= MAX_GRO_SKBS)) 6111 gro_flush_oldest(napi, &gro_list->list); 6112 else 6113 gro_list->count++; 6114 6115 NAPI_GRO_CB(skb)->count = 1; 6116 NAPI_GRO_CB(skb)->age = jiffies; 6117 NAPI_GRO_CB(skb)->last = skb; 6118 skb_shinfo(skb)->gso_size = skb_gro_len(skb); 6119 list_add(&skb->list, &gro_list->list); 6120 ret = GRO_HELD; 6121 6122 pull: 6123 grow = skb_gro_offset(skb) - skb_headlen(skb); 6124 if (grow > 0) 6125 gro_pull_from_frag0(skb, grow); 6126 ok: 6127 if (gro_list->count) { 6128 if (!test_bit(bucket, &napi->gro_bitmask)) 6129 __set_bit(bucket, &napi->gro_bitmask); 6130 } else if (test_bit(bucket, &napi->gro_bitmask)) { 6131 __clear_bit(bucket, &napi->gro_bitmask); 6132 } 6133 6134 return ret; 6135 6136 normal: 6137 ret = GRO_NORMAL; 6138 goto pull; 6139 } 6140 6141 struct packet_offload *gro_find_receive_by_type(__be16 type) 6142 { 6143 struct list_head *offload_head = &offload_base; 6144 struct packet_offload *ptype; 6145 6146 list_for_each_entry_rcu(ptype, offload_head, list) { 6147 if (ptype->type != type || !ptype->callbacks.gro_receive) 6148 continue; 6149 return ptype; 6150 } 6151 return NULL; 6152 } 6153 EXPORT_SYMBOL(gro_find_receive_by_type); 6154 6155 struct packet_offload *gro_find_complete_by_type(__be16 type) 6156 { 6157 struct list_head *offload_head = &offload_base; 6158 struct packet_offload *ptype; 6159 6160 list_for_each_entry_rcu(ptype, offload_head, list) { 6161 if (ptype->type != type || !ptype->callbacks.gro_complete) 6162 continue; 6163 return ptype; 6164 } 6165 return NULL; 6166 } 6167 EXPORT_SYMBOL(gro_find_complete_by_type); 6168 6169 static gro_result_t napi_skb_finish(struct napi_struct *napi, 6170 struct sk_buff *skb, 6171 gro_result_t ret) 6172 { 6173 switch (ret) { 6174 case GRO_NORMAL: 6175 gro_normal_one(napi, skb, 1); 6176 break; 6177 6178 case GRO_MERGED_FREE: 6179 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) 6180 napi_skb_free_stolen_head(skb); 6181 else if (skb->fclone != SKB_FCLONE_UNAVAILABLE) 6182 __kfree_skb(skb); 6183 else 6184 __kfree_skb_defer(skb); 6185 break; 6186 6187 case GRO_HELD: 6188 case GRO_MERGED: 6189 case GRO_CONSUMED: 6190 break; 6191 } 6192 6193 return ret; 6194 } 6195 6196 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) 6197 { 6198 gro_result_t ret; 6199 6200 skb_mark_napi_id(skb, napi); 6201 trace_napi_gro_receive_entry(skb); 6202 6203 skb_gro_reset_offset(skb, 0); 6204 6205 ret = napi_skb_finish(napi, skb, dev_gro_receive(napi, skb)); 6206 trace_napi_gro_receive_exit(ret); 6207 6208 return ret; 6209 } 6210 EXPORT_SYMBOL(napi_gro_receive); 6211 6212 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) 6213 { 6214 if (unlikely(skb->pfmemalloc)) { 6215 consume_skb(skb); 6216 return; 6217 } 6218 __skb_pull(skb, skb_headlen(skb)); 6219 /* restore the reserve we had after netdev_alloc_skb_ip_align() */ 6220 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb)); 6221 __vlan_hwaccel_clear_tag(skb); 6222 skb->dev = napi->dev; 6223 skb->skb_iif = 0; 6224 6225 /* eth_type_trans() assumes pkt_type is PACKET_HOST */ 6226 skb->pkt_type = PACKET_HOST; 6227 6228 skb->encapsulation = 0; 6229 skb_shinfo(skb)->gso_type = 0; 6230 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); 6231 if (unlikely(skb->slow_gro)) { 6232 skb_orphan(skb); 6233 skb_ext_reset(skb); 6234 nf_reset_ct(skb); 6235 skb->slow_gro = 0; 6236 } 6237 6238 napi->skb = skb; 6239 } 6240 6241 struct sk_buff *napi_get_frags(struct napi_struct *napi) 6242 { 6243 struct sk_buff *skb = napi->skb; 6244 6245 if (!skb) { 6246 skb = napi_alloc_skb(napi, GRO_MAX_HEAD); 6247 if (skb) { 6248 napi->skb = skb; 6249 skb_mark_napi_id(skb, napi); 6250 } 6251 } 6252 return skb; 6253 } 6254 EXPORT_SYMBOL(napi_get_frags); 6255 6256 static gro_result_t napi_frags_finish(struct napi_struct *napi, 6257 struct sk_buff *skb, 6258 gro_result_t ret) 6259 { 6260 switch (ret) { 6261 case GRO_NORMAL: 6262 case GRO_HELD: 6263 __skb_push(skb, ETH_HLEN); 6264 skb->protocol = eth_type_trans(skb, skb->dev); 6265 if (ret == GRO_NORMAL) 6266 gro_normal_one(napi, skb, 1); 6267 break; 6268 6269 case GRO_MERGED_FREE: 6270 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) 6271 napi_skb_free_stolen_head(skb); 6272 else 6273 napi_reuse_skb(napi, skb); 6274 break; 6275 6276 case GRO_MERGED: 6277 case GRO_CONSUMED: 6278 break; 6279 } 6280 6281 return ret; 6282 } 6283 6284 /* Upper GRO stack assumes network header starts at gro_offset=0 6285 * Drivers could call both napi_gro_frags() and napi_gro_receive() 6286 * We copy ethernet header into skb->data to have a common layout. 6287 */ 6288 static struct sk_buff *napi_frags_skb(struct napi_struct *napi) 6289 { 6290 struct sk_buff *skb = napi->skb; 6291 const struct ethhdr *eth; 6292 unsigned int hlen = sizeof(*eth); 6293 6294 napi->skb = NULL; 6295 6296 skb_reset_mac_header(skb); 6297 skb_gro_reset_offset(skb, hlen); 6298 6299 if (unlikely(skb_gro_header_hard(skb, hlen))) { 6300 eth = skb_gro_header_slow(skb, hlen, 0); 6301 if (unlikely(!eth)) { 6302 net_warn_ratelimited("%s: dropping impossible skb from %s\n", 6303 __func__, napi->dev->name); 6304 napi_reuse_skb(napi, skb); 6305 return NULL; 6306 } 6307 } else { 6308 eth = (const struct ethhdr *)skb->data; 6309 gro_pull_from_frag0(skb, hlen); 6310 NAPI_GRO_CB(skb)->frag0 += hlen; 6311 NAPI_GRO_CB(skb)->frag0_len -= hlen; 6312 } 6313 __skb_pull(skb, hlen); 6314 6315 /* 6316 * This works because the only protocols we care about don't require 6317 * special handling. 6318 * We'll fix it up properly in napi_frags_finish() 6319 */ 6320 skb->protocol = eth->h_proto; 6321 6322 return skb; 6323 } 6324 6325 gro_result_t napi_gro_frags(struct napi_struct *napi) 6326 { 6327 gro_result_t ret; 6328 struct sk_buff *skb = napi_frags_skb(napi); 6329 6330 trace_napi_gro_frags_entry(skb); 6331 6332 ret = napi_frags_finish(napi, skb, dev_gro_receive(napi, skb)); 6333 trace_napi_gro_frags_exit(ret); 6334 6335 return ret; 6336 } 6337 EXPORT_SYMBOL(napi_gro_frags); 6338 6339 /* Compute the checksum from gro_offset and return the folded value 6340 * after adding in any pseudo checksum. 6341 */ 6342 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb) 6343 { 6344 __wsum wsum; 6345 __sum16 sum; 6346 6347 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0); 6348 6349 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */ 6350 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum)); 6351 /* See comments in __skb_checksum_complete(). */ 6352 if (likely(!sum)) { 6353 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && 6354 !skb->csum_complete_sw) 6355 netdev_rx_csum_fault(skb->dev, skb); 6356 } 6357 6358 NAPI_GRO_CB(skb)->csum = wsum; 6359 NAPI_GRO_CB(skb)->csum_valid = 1; 6360 6361 return sum; 6362 } 6363 EXPORT_SYMBOL(__skb_gro_checksum_complete); 6364 6365 static void net_rps_send_ipi(struct softnet_data *remsd) 6366 { 6367 #ifdef CONFIG_RPS 6368 while (remsd) { 6369 struct softnet_data *next = remsd->rps_ipi_next; 6370 6371 if (cpu_online(remsd->cpu)) 6372 smp_call_function_single_async(remsd->cpu, &remsd->csd); 6373 remsd = next; 6374 } 6375 #endif 6376 } 6377 6378 /* 6379 * net_rps_action_and_irq_enable sends any pending IPI's for rps. 6380 * Note: called with local irq disabled, but exits with local irq enabled. 6381 */ 6382 static void net_rps_action_and_irq_enable(struct softnet_data *sd) 6383 { 6384 #ifdef CONFIG_RPS 6385 struct softnet_data *remsd = sd->rps_ipi_list; 6386 6387 if (remsd) { 6388 sd->rps_ipi_list = NULL; 6389 6390 local_irq_enable(); 6391 6392 /* Send pending IPI's to kick RPS processing on remote cpus. */ 6393 net_rps_send_ipi(remsd); 6394 } else 6395 #endif 6396 local_irq_enable(); 6397 } 6398 6399 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) 6400 { 6401 #ifdef CONFIG_RPS 6402 return sd->rps_ipi_list != NULL; 6403 #else 6404 return false; 6405 #endif 6406 } 6407 6408 static int process_backlog(struct napi_struct *napi, int quota) 6409 { 6410 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); 6411 bool again = true; 6412 int work = 0; 6413 6414 /* Check if we have pending ipi, its better to send them now, 6415 * not waiting net_rx_action() end. 6416 */ 6417 if (sd_has_rps_ipi_waiting(sd)) { 6418 local_irq_disable(); 6419 net_rps_action_and_irq_enable(sd); 6420 } 6421 6422 napi->weight = dev_rx_weight; 6423 while (again) { 6424 struct sk_buff *skb; 6425 6426 while ((skb = __skb_dequeue(&sd->process_queue))) { 6427 rcu_read_lock(); 6428 __netif_receive_skb(skb); 6429 rcu_read_unlock(); 6430 input_queue_head_incr(sd); 6431 if (++work >= quota) 6432 return work; 6433 6434 } 6435 6436 local_irq_disable(); 6437 rps_lock(sd); 6438 if (skb_queue_empty(&sd->input_pkt_queue)) { 6439 /* 6440 * Inline a custom version of __napi_complete(). 6441 * only current cpu owns and manipulates this napi, 6442 * and NAPI_STATE_SCHED is the only possible flag set 6443 * on backlog. 6444 * We can use a plain write instead of clear_bit(), 6445 * and we dont need an smp_mb() memory barrier. 6446 */ 6447 napi->state = 0; 6448 again = false; 6449 } else { 6450 skb_queue_splice_tail_init(&sd->input_pkt_queue, 6451 &sd->process_queue); 6452 } 6453 rps_unlock(sd); 6454 local_irq_enable(); 6455 } 6456 6457 return work; 6458 } 6459 6460 /** 6461 * __napi_schedule - schedule for receive 6462 * @n: entry to schedule 6463 * 6464 * The entry's receive function will be scheduled to run. 6465 * Consider using __napi_schedule_irqoff() if hard irqs are masked. 6466 */ 6467 void __napi_schedule(struct napi_struct *n) 6468 { 6469 unsigned long flags; 6470 6471 local_irq_save(flags); 6472 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6473 local_irq_restore(flags); 6474 } 6475 EXPORT_SYMBOL(__napi_schedule); 6476 6477 /** 6478 * napi_schedule_prep - check if napi can be scheduled 6479 * @n: napi context 6480 * 6481 * Test if NAPI routine is already running, and if not mark 6482 * it as running. This is used as a condition variable to 6483 * insure only one NAPI poll instance runs. We also make 6484 * sure there is no pending NAPI disable. 6485 */ 6486 bool napi_schedule_prep(struct napi_struct *n) 6487 { 6488 unsigned long val, new; 6489 6490 do { 6491 val = READ_ONCE(n->state); 6492 if (unlikely(val & NAPIF_STATE_DISABLE)) 6493 return false; 6494 new = val | NAPIF_STATE_SCHED; 6495 6496 /* Sets STATE_MISSED bit if STATE_SCHED was already set 6497 * This was suggested by Alexander Duyck, as compiler 6498 * emits better code than : 6499 * if (val & NAPIF_STATE_SCHED) 6500 * new |= NAPIF_STATE_MISSED; 6501 */ 6502 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * 6503 NAPIF_STATE_MISSED; 6504 } while (cmpxchg(&n->state, val, new) != val); 6505 6506 return !(val & NAPIF_STATE_SCHED); 6507 } 6508 EXPORT_SYMBOL(napi_schedule_prep); 6509 6510 /** 6511 * __napi_schedule_irqoff - schedule for receive 6512 * @n: entry to schedule 6513 * 6514 * Variant of __napi_schedule() assuming hard irqs are masked. 6515 * 6516 * On PREEMPT_RT enabled kernels this maps to __napi_schedule() 6517 * because the interrupt disabled assumption might not be true 6518 * due to force-threaded interrupts and spinlock substitution. 6519 */ 6520 void __napi_schedule_irqoff(struct napi_struct *n) 6521 { 6522 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6523 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6524 else 6525 __napi_schedule(n); 6526 } 6527 EXPORT_SYMBOL(__napi_schedule_irqoff); 6528 6529 bool napi_complete_done(struct napi_struct *n, int work_done) 6530 { 6531 unsigned long flags, val, new, timeout = 0; 6532 bool ret = true; 6533 6534 /* 6535 * 1) Don't let napi dequeue from the cpu poll list 6536 * just in case its running on a different cpu. 6537 * 2) If we are busy polling, do nothing here, we have 6538 * the guarantee we will be called later. 6539 */ 6540 if (unlikely(n->state & (NAPIF_STATE_NPSVC | 6541 NAPIF_STATE_IN_BUSY_POLL))) 6542 return false; 6543 6544 if (work_done) { 6545 if (n->gro_bitmask) 6546 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6547 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs); 6548 } 6549 if (n->defer_hard_irqs_count > 0) { 6550 n->defer_hard_irqs_count--; 6551 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6552 if (timeout) 6553 ret = false; 6554 } 6555 if (n->gro_bitmask) { 6556 /* When the NAPI instance uses a timeout and keeps postponing 6557 * it, we need to bound somehow the time packets are kept in 6558 * the GRO layer 6559 */ 6560 napi_gro_flush(n, !!timeout); 6561 } 6562 6563 gro_normal_list(n); 6564 6565 if (unlikely(!list_empty(&n->poll_list))) { 6566 /* If n->poll_list is not empty, we need to mask irqs */ 6567 local_irq_save(flags); 6568 list_del_init(&n->poll_list); 6569 local_irq_restore(flags); 6570 } 6571 6572 do { 6573 val = READ_ONCE(n->state); 6574 6575 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); 6576 6577 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED | 6578 NAPIF_STATE_SCHED_THREADED | 6579 NAPIF_STATE_PREFER_BUSY_POLL); 6580 6581 /* If STATE_MISSED was set, leave STATE_SCHED set, 6582 * because we will call napi->poll() one more time. 6583 * This C code was suggested by Alexander Duyck to help gcc. 6584 */ 6585 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * 6586 NAPIF_STATE_SCHED; 6587 } while (cmpxchg(&n->state, val, new) != val); 6588 6589 if (unlikely(val & NAPIF_STATE_MISSED)) { 6590 __napi_schedule(n); 6591 return false; 6592 } 6593 6594 if (timeout) 6595 hrtimer_start(&n->timer, ns_to_ktime(timeout), 6596 HRTIMER_MODE_REL_PINNED); 6597 return ret; 6598 } 6599 EXPORT_SYMBOL(napi_complete_done); 6600 6601 /* must be called under rcu_read_lock(), as we dont take a reference */ 6602 static struct napi_struct *napi_by_id(unsigned int napi_id) 6603 { 6604 unsigned int hash = napi_id % HASH_SIZE(napi_hash); 6605 struct napi_struct *napi; 6606 6607 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) 6608 if (napi->napi_id == napi_id) 6609 return napi; 6610 6611 return NULL; 6612 } 6613 6614 #if defined(CONFIG_NET_RX_BUSY_POLL) 6615 6616 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule) 6617 { 6618 if (!skip_schedule) { 6619 gro_normal_list(napi); 6620 __napi_schedule(napi); 6621 return; 6622 } 6623 6624 if (napi->gro_bitmask) { 6625 /* flush too old packets 6626 * If HZ < 1000, flush all packets. 6627 */ 6628 napi_gro_flush(napi, HZ >= 1000); 6629 } 6630 6631 gro_normal_list(napi); 6632 clear_bit(NAPI_STATE_SCHED, &napi->state); 6633 } 6634 6635 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll, 6636 u16 budget) 6637 { 6638 bool skip_schedule = false; 6639 unsigned long timeout; 6640 int rc; 6641 6642 /* Busy polling means there is a high chance device driver hard irq 6643 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was 6644 * set in napi_schedule_prep(). 6645 * Since we are about to call napi->poll() once more, we can safely 6646 * clear NAPI_STATE_MISSED. 6647 * 6648 * Note: x86 could use a single "lock and ..." instruction 6649 * to perform these two clear_bit() 6650 */ 6651 clear_bit(NAPI_STATE_MISSED, &napi->state); 6652 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); 6653 6654 local_bh_disable(); 6655 6656 if (prefer_busy_poll) { 6657 napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs); 6658 timeout = READ_ONCE(napi->dev->gro_flush_timeout); 6659 if (napi->defer_hard_irqs_count && timeout) { 6660 hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED); 6661 skip_schedule = true; 6662 } 6663 } 6664 6665 /* All we really want here is to re-enable device interrupts. 6666 * Ideally, a new ndo_busy_poll_stop() could avoid another round. 6667 */ 6668 rc = napi->poll(napi, budget); 6669 /* We can't gro_normal_list() here, because napi->poll() might have 6670 * rearmed the napi (napi_complete_done()) in which case it could 6671 * already be running on another CPU. 6672 */ 6673 trace_napi_poll(napi, rc, budget); 6674 netpoll_poll_unlock(have_poll_lock); 6675 if (rc == budget) 6676 __busy_poll_stop(napi, skip_schedule); 6677 local_bh_enable(); 6678 } 6679 6680 void napi_busy_loop(unsigned int napi_id, 6681 bool (*loop_end)(void *, unsigned long), 6682 void *loop_end_arg, bool prefer_busy_poll, u16 budget) 6683 { 6684 unsigned long start_time = loop_end ? busy_loop_current_time() : 0; 6685 int (*napi_poll)(struct napi_struct *napi, int budget); 6686 void *have_poll_lock = NULL; 6687 struct napi_struct *napi; 6688 6689 restart: 6690 napi_poll = NULL; 6691 6692 rcu_read_lock(); 6693 6694 napi = napi_by_id(napi_id); 6695 if (!napi) 6696 goto out; 6697 6698 preempt_disable(); 6699 for (;;) { 6700 int work = 0; 6701 6702 local_bh_disable(); 6703 if (!napi_poll) { 6704 unsigned long val = READ_ONCE(napi->state); 6705 6706 /* If multiple threads are competing for this napi, 6707 * we avoid dirtying napi->state as much as we can. 6708 */ 6709 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | 6710 NAPIF_STATE_IN_BUSY_POLL)) { 6711 if (prefer_busy_poll) 6712 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6713 goto count; 6714 } 6715 if (cmpxchg(&napi->state, val, 6716 val | NAPIF_STATE_IN_BUSY_POLL | 6717 NAPIF_STATE_SCHED) != val) { 6718 if (prefer_busy_poll) 6719 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6720 goto count; 6721 } 6722 have_poll_lock = netpoll_poll_lock(napi); 6723 napi_poll = napi->poll; 6724 } 6725 work = napi_poll(napi, budget); 6726 trace_napi_poll(napi, work, budget); 6727 gro_normal_list(napi); 6728 count: 6729 if (work > 0) 6730 __NET_ADD_STATS(dev_net(napi->dev), 6731 LINUX_MIB_BUSYPOLLRXPACKETS, work); 6732 local_bh_enable(); 6733 6734 if (!loop_end || loop_end(loop_end_arg, start_time)) 6735 break; 6736 6737 if (unlikely(need_resched())) { 6738 if (napi_poll) 6739 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6740 preempt_enable(); 6741 rcu_read_unlock(); 6742 cond_resched(); 6743 if (loop_end(loop_end_arg, start_time)) 6744 return; 6745 goto restart; 6746 } 6747 cpu_relax(); 6748 } 6749 if (napi_poll) 6750 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6751 preempt_enable(); 6752 out: 6753 rcu_read_unlock(); 6754 } 6755 EXPORT_SYMBOL(napi_busy_loop); 6756 6757 #endif /* CONFIG_NET_RX_BUSY_POLL */ 6758 6759 static void napi_hash_add(struct napi_struct *napi) 6760 { 6761 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state)) 6762 return; 6763 6764 spin_lock(&napi_hash_lock); 6765 6766 /* 0..NR_CPUS range is reserved for sender_cpu use */ 6767 do { 6768 if (unlikely(++napi_gen_id < MIN_NAPI_ID)) 6769 napi_gen_id = MIN_NAPI_ID; 6770 } while (napi_by_id(napi_gen_id)); 6771 napi->napi_id = napi_gen_id; 6772 6773 hlist_add_head_rcu(&napi->napi_hash_node, 6774 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); 6775 6776 spin_unlock(&napi_hash_lock); 6777 } 6778 6779 /* Warning : caller is responsible to make sure rcu grace period 6780 * is respected before freeing memory containing @napi 6781 */ 6782 static void napi_hash_del(struct napi_struct *napi) 6783 { 6784 spin_lock(&napi_hash_lock); 6785 6786 hlist_del_init_rcu(&napi->napi_hash_node); 6787 6788 spin_unlock(&napi_hash_lock); 6789 } 6790 6791 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) 6792 { 6793 struct napi_struct *napi; 6794 6795 napi = container_of(timer, struct napi_struct, timer); 6796 6797 /* Note : we use a relaxed variant of napi_schedule_prep() not setting 6798 * NAPI_STATE_MISSED, since we do not react to a device IRQ. 6799 */ 6800 if (!napi_disable_pending(napi) && 6801 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) { 6802 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6803 __napi_schedule_irqoff(napi); 6804 } 6805 6806 return HRTIMER_NORESTART; 6807 } 6808 6809 static void init_gro_hash(struct napi_struct *napi) 6810 { 6811 int i; 6812 6813 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6814 INIT_LIST_HEAD(&napi->gro_hash[i].list); 6815 napi->gro_hash[i].count = 0; 6816 } 6817 napi->gro_bitmask = 0; 6818 } 6819 6820 int dev_set_threaded(struct net_device *dev, bool threaded) 6821 { 6822 struct napi_struct *napi; 6823 int err = 0; 6824 6825 if (dev->threaded == threaded) 6826 return 0; 6827 6828 if (threaded) { 6829 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6830 if (!napi->thread) { 6831 err = napi_kthread_create(napi); 6832 if (err) { 6833 threaded = false; 6834 break; 6835 } 6836 } 6837 } 6838 } 6839 6840 dev->threaded = threaded; 6841 6842 /* Make sure kthread is created before THREADED bit 6843 * is set. 6844 */ 6845 smp_mb__before_atomic(); 6846 6847 /* Setting/unsetting threaded mode on a napi might not immediately 6848 * take effect, if the current napi instance is actively being 6849 * polled. In this case, the switch between threaded mode and 6850 * softirq mode will happen in the next round of napi_schedule(). 6851 * This should not cause hiccups/stalls to the live traffic. 6852 */ 6853 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6854 if (threaded) 6855 set_bit(NAPI_STATE_THREADED, &napi->state); 6856 else 6857 clear_bit(NAPI_STATE_THREADED, &napi->state); 6858 } 6859 6860 return err; 6861 } 6862 EXPORT_SYMBOL(dev_set_threaded); 6863 6864 void netif_napi_add(struct net_device *dev, struct napi_struct *napi, 6865 int (*poll)(struct napi_struct *, int), int weight) 6866 { 6867 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state))) 6868 return; 6869 6870 INIT_LIST_HEAD(&napi->poll_list); 6871 INIT_HLIST_NODE(&napi->napi_hash_node); 6872 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); 6873 napi->timer.function = napi_watchdog; 6874 init_gro_hash(napi); 6875 napi->skb = NULL; 6876 INIT_LIST_HEAD(&napi->rx_list); 6877 napi->rx_count = 0; 6878 napi->poll = poll; 6879 if (weight > NAPI_POLL_WEIGHT) 6880 netdev_err_once(dev, "%s() called with weight %d\n", __func__, 6881 weight); 6882 napi->weight = weight; 6883 napi->dev = dev; 6884 #ifdef CONFIG_NETPOLL 6885 napi->poll_owner = -1; 6886 #endif 6887 set_bit(NAPI_STATE_SCHED, &napi->state); 6888 set_bit(NAPI_STATE_NPSVC, &napi->state); 6889 list_add_rcu(&napi->dev_list, &dev->napi_list); 6890 napi_hash_add(napi); 6891 /* Create kthread for this napi if dev->threaded is set. 6892 * Clear dev->threaded if kthread creation failed so that 6893 * threaded mode will not be enabled in napi_enable(). 6894 */ 6895 if (dev->threaded && napi_kthread_create(napi)) 6896 dev->threaded = 0; 6897 } 6898 EXPORT_SYMBOL(netif_napi_add); 6899 6900 void napi_disable(struct napi_struct *n) 6901 { 6902 unsigned long val, new; 6903 6904 might_sleep(); 6905 set_bit(NAPI_STATE_DISABLE, &n->state); 6906 6907 do { 6908 val = READ_ONCE(n->state); 6909 if (val & (NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC)) { 6910 usleep_range(20, 200); 6911 continue; 6912 } 6913 6914 new = val | NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC; 6915 new &= ~(NAPIF_STATE_THREADED | NAPIF_STATE_PREFER_BUSY_POLL); 6916 } while (cmpxchg(&n->state, val, new) != val); 6917 6918 hrtimer_cancel(&n->timer); 6919 6920 clear_bit(NAPI_STATE_DISABLE, &n->state); 6921 } 6922 EXPORT_SYMBOL(napi_disable); 6923 6924 /** 6925 * napi_enable - enable NAPI scheduling 6926 * @n: NAPI context 6927 * 6928 * Resume NAPI from being scheduled on this context. 6929 * Must be paired with napi_disable. 6930 */ 6931 void napi_enable(struct napi_struct *n) 6932 { 6933 unsigned long val, new; 6934 6935 do { 6936 val = READ_ONCE(n->state); 6937 BUG_ON(!test_bit(NAPI_STATE_SCHED, &val)); 6938 6939 new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC); 6940 if (n->dev->threaded && n->thread) 6941 new |= NAPIF_STATE_THREADED; 6942 } while (cmpxchg(&n->state, val, new) != val); 6943 } 6944 EXPORT_SYMBOL(napi_enable); 6945 6946 static void flush_gro_hash(struct napi_struct *napi) 6947 { 6948 int i; 6949 6950 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6951 struct sk_buff *skb, *n; 6952 6953 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list) 6954 kfree_skb(skb); 6955 napi->gro_hash[i].count = 0; 6956 } 6957 } 6958 6959 /* Must be called in process context */ 6960 void __netif_napi_del(struct napi_struct *napi) 6961 { 6962 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state)) 6963 return; 6964 6965 napi_hash_del(napi); 6966 list_del_rcu(&napi->dev_list); 6967 napi_free_frags(napi); 6968 6969 flush_gro_hash(napi); 6970 napi->gro_bitmask = 0; 6971 6972 if (napi->thread) { 6973 kthread_stop(napi->thread); 6974 napi->thread = NULL; 6975 } 6976 } 6977 EXPORT_SYMBOL(__netif_napi_del); 6978 6979 static int __napi_poll(struct napi_struct *n, bool *repoll) 6980 { 6981 int work, weight; 6982 6983 weight = n->weight; 6984 6985 /* This NAPI_STATE_SCHED test is for avoiding a race 6986 * with netpoll's poll_napi(). Only the entity which 6987 * obtains the lock and sees NAPI_STATE_SCHED set will 6988 * actually make the ->poll() call. Therefore we avoid 6989 * accidentally calling ->poll() when NAPI is not scheduled. 6990 */ 6991 work = 0; 6992 if (test_bit(NAPI_STATE_SCHED, &n->state)) { 6993 work = n->poll(n, weight); 6994 trace_napi_poll(n, work, weight); 6995 } 6996 6997 if (unlikely(work > weight)) 6998 pr_err_once("NAPI poll function %pS returned %d, exceeding its budget of %d.\n", 6999 n->poll, work, weight); 7000 7001 if (likely(work < weight)) 7002 return work; 7003 7004 /* Drivers must not modify the NAPI state if they 7005 * consume the entire weight. In such cases this code 7006 * still "owns" the NAPI instance and therefore can 7007 * move the instance around on the list at-will. 7008 */ 7009 if (unlikely(napi_disable_pending(n))) { 7010 napi_complete(n); 7011 return work; 7012 } 7013 7014 /* The NAPI context has more processing work, but busy-polling 7015 * is preferred. Exit early. 7016 */ 7017 if (napi_prefer_busy_poll(n)) { 7018 if (napi_complete_done(n, work)) { 7019 /* If timeout is not set, we need to make sure 7020 * that the NAPI is re-scheduled. 7021 */ 7022 napi_schedule(n); 7023 } 7024 return work; 7025 } 7026 7027 if (n->gro_bitmask) { 7028 /* flush too old packets 7029 * If HZ < 1000, flush all packets. 7030 */ 7031 napi_gro_flush(n, HZ >= 1000); 7032 } 7033 7034 gro_normal_list(n); 7035 7036 /* Some drivers may have called napi_schedule 7037 * prior to exhausting their budget. 7038 */ 7039 if (unlikely(!list_empty(&n->poll_list))) { 7040 pr_warn_once("%s: Budget exhausted after napi rescheduled\n", 7041 n->dev ? n->dev->name : "backlog"); 7042 return work; 7043 } 7044 7045 *repoll = true; 7046 7047 return work; 7048 } 7049 7050 static int napi_poll(struct napi_struct *n, struct list_head *repoll) 7051 { 7052 bool do_repoll = false; 7053 void *have; 7054 int work; 7055 7056 list_del_init(&n->poll_list); 7057 7058 have = netpoll_poll_lock(n); 7059 7060 work = __napi_poll(n, &do_repoll); 7061 7062 if (do_repoll) 7063 list_add_tail(&n->poll_list, repoll); 7064 7065 netpoll_poll_unlock(have); 7066 7067 return work; 7068 } 7069 7070 static int napi_thread_wait(struct napi_struct *napi) 7071 { 7072 bool woken = false; 7073 7074 set_current_state(TASK_INTERRUPTIBLE); 7075 7076 while (!kthread_should_stop()) { 7077 /* Testing SCHED_THREADED bit here to make sure the current 7078 * kthread owns this napi and could poll on this napi. 7079 * Testing SCHED bit is not enough because SCHED bit might be 7080 * set by some other busy poll thread or by napi_disable(). 7081 */ 7082 if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state) || woken) { 7083 WARN_ON(!list_empty(&napi->poll_list)); 7084 __set_current_state(TASK_RUNNING); 7085 return 0; 7086 } 7087 7088 schedule(); 7089 /* woken being true indicates this thread owns this napi. */ 7090 woken = true; 7091 set_current_state(TASK_INTERRUPTIBLE); 7092 } 7093 __set_current_state(TASK_RUNNING); 7094 7095 return -1; 7096 } 7097 7098 static int napi_threaded_poll(void *data) 7099 { 7100 struct napi_struct *napi = data; 7101 void *have; 7102 7103 while (!napi_thread_wait(napi)) { 7104 for (;;) { 7105 bool repoll = false; 7106 7107 local_bh_disable(); 7108 7109 have = netpoll_poll_lock(napi); 7110 __napi_poll(napi, &repoll); 7111 netpoll_poll_unlock(have); 7112 7113 local_bh_enable(); 7114 7115 if (!repoll) 7116 break; 7117 7118 cond_resched(); 7119 } 7120 } 7121 return 0; 7122 } 7123 7124 static __latent_entropy void net_rx_action(struct softirq_action *h) 7125 { 7126 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 7127 unsigned long time_limit = jiffies + 7128 usecs_to_jiffies(netdev_budget_usecs); 7129 int budget = netdev_budget; 7130 LIST_HEAD(list); 7131 LIST_HEAD(repoll); 7132 7133 local_irq_disable(); 7134 list_splice_init(&sd->poll_list, &list); 7135 local_irq_enable(); 7136 7137 for (;;) { 7138 struct napi_struct *n; 7139 7140 if (list_empty(&list)) { 7141 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll)) 7142 return; 7143 break; 7144 } 7145 7146 n = list_first_entry(&list, struct napi_struct, poll_list); 7147 budget -= napi_poll(n, &repoll); 7148 7149 /* If softirq window is exhausted then punt. 7150 * Allow this to run for 2 jiffies since which will allow 7151 * an average latency of 1.5/HZ. 7152 */ 7153 if (unlikely(budget <= 0 || 7154 time_after_eq(jiffies, time_limit))) { 7155 sd->time_squeeze++; 7156 break; 7157 } 7158 } 7159 7160 local_irq_disable(); 7161 7162 list_splice_tail_init(&sd->poll_list, &list); 7163 list_splice_tail(&repoll, &list); 7164 list_splice(&list, &sd->poll_list); 7165 if (!list_empty(&sd->poll_list)) 7166 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 7167 7168 net_rps_action_and_irq_enable(sd); 7169 } 7170 7171 struct netdev_adjacent { 7172 struct net_device *dev; 7173 7174 /* upper master flag, there can only be one master device per list */ 7175 bool master; 7176 7177 /* lookup ignore flag */ 7178 bool ignore; 7179 7180 /* counter for the number of times this device was added to us */ 7181 u16 ref_nr; 7182 7183 /* private field for the users */ 7184 void *private; 7185 7186 struct list_head list; 7187 struct rcu_head rcu; 7188 }; 7189 7190 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, 7191 struct list_head *adj_list) 7192 { 7193 struct netdev_adjacent *adj; 7194 7195 list_for_each_entry(adj, adj_list, list) { 7196 if (adj->dev == adj_dev) 7197 return adj; 7198 } 7199 return NULL; 7200 } 7201 7202 static int ____netdev_has_upper_dev(struct net_device *upper_dev, 7203 struct netdev_nested_priv *priv) 7204 { 7205 struct net_device *dev = (struct net_device *)priv->data; 7206 7207 return upper_dev == dev; 7208 } 7209 7210 /** 7211 * netdev_has_upper_dev - Check if device is linked to an upper device 7212 * @dev: device 7213 * @upper_dev: upper device to check 7214 * 7215 * Find out if a device is linked to specified upper device and return true 7216 * in case it is. Note that this checks only immediate upper device, 7217 * not through a complete stack of devices. The caller must hold the RTNL lock. 7218 */ 7219 bool netdev_has_upper_dev(struct net_device *dev, 7220 struct net_device *upper_dev) 7221 { 7222 struct netdev_nested_priv priv = { 7223 .data = (void *)upper_dev, 7224 }; 7225 7226 ASSERT_RTNL(); 7227 7228 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 7229 &priv); 7230 } 7231 EXPORT_SYMBOL(netdev_has_upper_dev); 7232 7233 /** 7234 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device 7235 * @dev: device 7236 * @upper_dev: upper device to check 7237 * 7238 * Find out if a device is linked to specified upper device and return true 7239 * in case it is. Note that this checks the entire upper device chain. 7240 * The caller must hold rcu lock. 7241 */ 7242 7243 bool netdev_has_upper_dev_all_rcu(struct net_device *dev, 7244 struct net_device *upper_dev) 7245 { 7246 struct netdev_nested_priv priv = { 7247 .data = (void *)upper_dev, 7248 }; 7249 7250 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 7251 &priv); 7252 } 7253 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); 7254 7255 /** 7256 * netdev_has_any_upper_dev - Check if device is linked to some device 7257 * @dev: device 7258 * 7259 * Find out if a device is linked to an upper device and return true in case 7260 * it is. The caller must hold the RTNL lock. 7261 */ 7262 bool netdev_has_any_upper_dev(struct net_device *dev) 7263 { 7264 ASSERT_RTNL(); 7265 7266 return !list_empty(&dev->adj_list.upper); 7267 } 7268 EXPORT_SYMBOL(netdev_has_any_upper_dev); 7269 7270 /** 7271 * netdev_master_upper_dev_get - Get master upper device 7272 * @dev: device 7273 * 7274 * Find a master upper device and return pointer to it or NULL in case 7275 * it's not there. The caller must hold the RTNL lock. 7276 */ 7277 struct net_device *netdev_master_upper_dev_get(struct net_device *dev) 7278 { 7279 struct netdev_adjacent *upper; 7280 7281 ASSERT_RTNL(); 7282 7283 if (list_empty(&dev->adj_list.upper)) 7284 return NULL; 7285 7286 upper = list_first_entry(&dev->adj_list.upper, 7287 struct netdev_adjacent, list); 7288 if (likely(upper->master)) 7289 return upper->dev; 7290 return NULL; 7291 } 7292 EXPORT_SYMBOL(netdev_master_upper_dev_get); 7293 7294 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev) 7295 { 7296 struct netdev_adjacent *upper; 7297 7298 ASSERT_RTNL(); 7299 7300 if (list_empty(&dev->adj_list.upper)) 7301 return NULL; 7302 7303 upper = list_first_entry(&dev->adj_list.upper, 7304 struct netdev_adjacent, list); 7305 if (likely(upper->master) && !upper->ignore) 7306 return upper->dev; 7307 return NULL; 7308 } 7309 7310 /** 7311 * netdev_has_any_lower_dev - Check if device is linked to some device 7312 * @dev: device 7313 * 7314 * Find out if a device is linked to a lower device and return true in case 7315 * it is. The caller must hold the RTNL lock. 7316 */ 7317 static bool netdev_has_any_lower_dev(struct net_device *dev) 7318 { 7319 ASSERT_RTNL(); 7320 7321 return !list_empty(&dev->adj_list.lower); 7322 } 7323 7324 void *netdev_adjacent_get_private(struct list_head *adj_list) 7325 { 7326 struct netdev_adjacent *adj; 7327 7328 adj = list_entry(adj_list, struct netdev_adjacent, list); 7329 7330 return adj->private; 7331 } 7332 EXPORT_SYMBOL(netdev_adjacent_get_private); 7333 7334 /** 7335 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list 7336 * @dev: device 7337 * @iter: list_head ** of the current position 7338 * 7339 * Gets the next device from the dev's upper list, starting from iter 7340 * position. The caller must hold RCU read lock. 7341 */ 7342 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, 7343 struct list_head **iter) 7344 { 7345 struct netdev_adjacent *upper; 7346 7347 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 7348 7349 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7350 7351 if (&upper->list == &dev->adj_list.upper) 7352 return NULL; 7353 7354 *iter = &upper->list; 7355 7356 return upper->dev; 7357 } 7358 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); 7359 7360 static struct net_device *__netdev_next_upper_dev(struct net_device *dev, 7361 struct list_head **iter, 7362 bool *ignore) 7363 { 7364 struct netdev_adjacent *upper; 7365 7366 upper = list_entry((*iter)->next, struct netdev_adjacent, list); 7367 7368 if (&upper->list == &dev->adj_list.upper) 7369 return NULL; 7370 7371 *iter = &upper->list; 7372 *ignore = upper->ignore; 7373 7374 return upper->dev; 7375 } 7376 7377 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, 7378 struct list_head **iter) 7379 { 7380 struct netdev_adjacent *upper; 7381 7382 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 7383 7384 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7385 7386 if (&upper->list == &dev->adj_list.upper) 7387 return NULL; 7388 7389 *iter = &upper->list; 7390 7391 return upper->dev; 7392 } 7393 7394 static int __netdev_walk_all_upper_dev(struct net_device *dev, 7395 int (*fn)(struct net_device *dev, 7396 struct netdev_nested_priv *priv), 7397 struct netdev_nested_priv *priv) 7398 { 7399 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7400 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7401 int ret, cur = 0; 7402 bool ignore; 7403 7404 now = dev; 7405 iter = &dev->adj_list.upper; 7406 7407 while (1) { 7408 if (now != dev) { 7409 ret = fn(now, priv); 7410 if (ret) 7411 return ret; 7412 } 7413 7414 next = NULL; 7415 while (1) { 7416 udev = __netdev_next_upper_dev(now, &iter, &ignore); 7417 if (!udev) 7418 break; 7419 if (ignore) 7420 continue; 7421 7422 next = udev; 7423 niter = &udev->adj_list.upper; 7424 dev_stack[cur] = now; 7425 iter_stack[cur++] = iter; 7426 break; 7427 } 7428 7429 if (!next) { 7430 if (!cur) 7431 return 0; 7432 next = dev_stack[--cur]; 7433 niter = iter_stack[cur]; 7434 } 7435 7436 now = next; 7437 iter = niter; 7438 } 7439 7440 return 0; 7441 } 7442 7443 int netdev_walk_all_upper_dev_rcu(struct net_device *dev, 7444 int (*fn)(struct net_device *dev, 7445 struct netdev_nested_priv *priv), 7446 struct netdev_nested_priv *priv) 7447 { 7448 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7449 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7450 int ret, cur = 0; 7451 7452 now = dev; 7453 iter = &dev->adj_list.upper; 7454 7455 while (1) { 7456 if (now != dev) { 7457 ret = fn(now, priv); 7458 if (ret) 7459 return ret; 7460 } 7461 7462 next = NULL; 7463 while (1) { 7464 udev = netdev_next_upper_dev_rcu(now, &iter); 7465 if (!udev) 7466 break; 7467 7468 next = udev; 7469 niter = &udev->adj_list.upper; 7470 dev_stack[cur] = now; 7471 iter_stack[cur++] = iter; 7472 break; 7473 } 7474 7475 if (!next) { 7476 if (!cur) 7477 return 0; 7478 next = dev_stack[--cur]; 7479 niter = iter_stack[cur]; 7480 } 7481 7482 now = next; 7483 iter = niter; 7484 } 7485 7486 return 0; 7487 } 7488 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); 7489 7490 static bool __netdev_has_upper_dev(struct net_device *dev, 7491 struct net_device *upper_dev) 7492 { 7493 struct netdev_nested_priv priv = { 7494 .flags = 0, 7495 .data = (void *)upper_dev, 7496 }; 7497 7498 ASSERT_RTNL(); 7499 7500 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev, 7501 &priv); 7502 } 7503 7504 /** 7505 * netdev_lower_get_next_private - Get the next ->private from the 7506 * lower neighbour list 7507 * @dev: device 7508 * @iter: list_head ** of the current position 7509 * 7510 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7511 * list, starting from iter position. The caller must hold either hold the 7512 * RTNL lock or its own locking that guarantees that the neighbour lower 7513 * list will remain unchanged. 7514 */ 7515 void *netdev_lower_get_next_private(struct net_device *dev, 7516 struct list_head **iter) 7517 { 7518 struct netdev_adjacent *lower; 7519 7520 lower = list_entry(*iter, struct netdev_adjacent, list); 7521 7522 if (&lower->list == &dev->adj_list.lower) 7523 return NULL; 7524 7525 *iter = lower->list.next; 7526 7527 return lower->private; 7528 } 7529 EXPORT_SYMBOL(netdev_lower_get_next_private); 7530 7531 /** 7532 * netdev_lower_get_next_private_rcu - Get the next ->private from the 7533 * lower neighbour list, RCU 7534 * variant 7535 * @dev: device 7536 * @iter: list_head ** of the current position 7537 * 7538 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7539 * list, starting from iter position. The caller must hold RCU read lock. 7540 */ 7541 void *netdev_lower_get_next_private_rcu(struct net_device *dev, 7542 struct list_head **iter) 7543 { 7544 struct netdev_adjacent *lower; 7545 7546 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 7547 7548 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7549 7550 if (&lower->list == &dev->adj_list.lower) 7551 return NULL; 7552 7553 *iter = &lower->list; 7554 7555 return lower->private; 7556 } 7557 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); 7558 7559 /** 7560 * netdev_lower_get_next - Get the next device from the lower neighbour 7561 * list 7562 * @dev: device 7563 * @iter: list_head ** of the current position 7564 * 7565 * Gets the next netdev_adjacent from the dev's lower neighbour 7566 * list, starting from iter position. The caller must hold RTNL lock or 7567 * its own locking that guarantees that the neighbour lower 7568 * list will remain unchanged. 7569 */ 7570 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) 7571 { 7572 struct netdev_adjacent *lower; 7573 7574 lower = list_entry(*iter, struct netdev_adjacent, list); 7575 7576 if (&lower->list == &dev->adj_list.lower) 7577 return NULL; 7578 7579 *iter = lower->list.next; 7580 7581 return lower->dev; 7582 } 7583 EXPORT_SYMBOL(netdev_lower_get_next); 7584 7585 static struct net_device *netdev_next_lower_dev(struct net_device *dev, 7586 struct list_head **iter) 7587 { 7588 struct netdev_adjacent *lower; 7589 7590 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7591 7592 if (&lower->list == &dev->adj_list.lower) 7593 return NULL; 7594 7595 *iter = &lower->list; 7596 7597 return lower->dev; 7598 } 7599 7600 static struct net_device *__netdev_next_lower_dev(struct net_device *dev, 7601 struct list_head **iter, 7602 bool *ignore) 7603 { 7604 struct netdev_adjacent *lower; 7605 7606 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7607 7608 if (&lower->list == &dev->adj_list.lower) 7609 return NULL; 7610 7611 *iter = &lower->list; 7612 *ignore = lower->ignore; 7613 7614 return lower->dev; 7615 } 7616 7617 int netdev_walk_all_lower_dev(struct net_device *dev, 7618 int (*fn)(struct net_device *dev, 7619 struct netdev_nested_priv *priv), 7620 struct netdev_nested_priv *priv) 7621 { 7622 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7623 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7624 int ret, cur = 0; 7625 7626 now = dev; 7627 iter = &dev->adj_list.lower; 7628 7629 while (1) { 7630 if (now != dev) { 7631 ret = fn(now, priv); 7632 if (ret) 7633 return ret; 7634 } 7635 7636 next = NULL; 7637 while (1) { 7638 ldev = netdev_next_lower_dev(now, &iter); 7639 if (!ldev) 7640 break; 7641 7642 next = ldev; 7643 niter = &ldev->adj_list.lower; 7644 dev_stack[cur] = now; 7645 iter_stack[cur++] = iter; 7646 break; 7647 } 7648 7649 if (!next) { 7650 if (!cur) 7651 return 0; 7652 next = dev_stack[--cur]; 7653 niter = iter_stack[cur]; 7654 } 7655 7656 now = next; 7657 iter = niter; 7658 } 7659 7660 return 0; 7661 } 7662 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); 7663 7664 static int __netdev_walk_all_lower_dev(struct net_device *dev, 7665 int (*fn)(struct net_device *dev, 7666 struct netdev_nested_priv *priv), 7667 struct netdev_nested_priv *priv) 7668 { 7669 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7670 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7671 int ret, cur = 0; 7672 bool ignore; 7673 7674 now = dev; 7675 iter = &dev->adj_list.lower; 7676 7677 while (1) { 7678 if (now != dev) { 7679 ret = fn(now, priv); 7680 if (ret) 7681 return ret; 7682 } 7683 7684 next = NULL; 7685 while (1) { 7686 ldev = __netdev_next_lower_dev(now, &iter, &ignore); 7687 if (!ldev) 7688 break; 7689 if (ignore) 7690 continue; 7691 7692 next = ldev; 7693 niter = &ldev->adj_list.lower; 7694 dev_stack[cur] = now; 7695 iter_stack[cur++] = iter; 7696 break; 7697 } 7698 7699 if (!next) { 7700 if (!cur) 7701 return 0; 7702 next = dev_stack[--cur]; 7703 niter = iter_stack[cur]; 7704 } 7705 7706 now = next; 7707 iter = niter; 7708 } 7709 7710 return 0; 7711 } 7712 7713 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, 7714 struct list_head **iter) 7715 { 7716 struct netdev_adjacent *lower; 7717 7718 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7719 if (&lower->list == &dev->adj_list.lower) 7720 return NULL; 7721 7722 *iter = &lower->list; 7723 7724 return lower->dev; 7725 } 7726 EXPORT_SYMBOL(netdev_next_lower_dev_rcu); 7727 7728 static u8 __netdev_upper_depth(struct net_device *dev) 7729 { 7730 struct net_device *udev; 7731 struct list_head *iter; 7732 u8 max_depth = 0; 7733 bool ignore; 7734 7735 for (iter = &dev->adj_list.upper, 7736 udev = __netdev_next_upper_dev(dev, &iter, &ignore); 7737 udev; 7738 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) { 7739 if (ignore) 7740 continue; 7741 if (max_depth < udev->upper_level) 7742 max_depth = udev->upper_level; 7743 } 7744 7745 return max_depth; 7746 } 7747 7748 static u8 __netdev_lower_depth(struct net_device *dev) 7749 { 7750 struct net_device *ldev; 7751 struct list_head *iter; 7752 u8 max_depth = 0; 7753 bool ignore; 7754 7755 for (iter = &dev->adj_list.lower, 7756 ldev = __netdev_next_lower_dev(dev, &iter, &ignore); 7757 ldev; 7758 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) { 7759 if (ignore) 7760 continue; 7761 if (max_depth < ldev->lower_level) 7762 max_depth = ldev->lower_level; 7763 } 7764 7765 return max_depth; 7766 } 7767 7768 static int __netdev_update_upper_level(struct net_device *dev, 7769 struct netdev_nested_priv *__unused) 7770 { 7771 dev->upper_level = __netdev_upper_depth(dev) + 1; 7772 return 0; 7773 } 7774 7775 static int __netdev_update_lower_level(struct net_device *dev, 7776 struct netdev_nested_priv *priv) 7777 { 7778 dev->lower_level = __netdev_lower_depth(dev) + 1; 7779 7780 #ifdef CONFIG_LOCKDEP 7781 if (!priv) 7782 return 0; 7783 7784 if (priv->flags & NESTED_SYNC_IMM) 7785 dev->nested_level = dev->lower_level - 1; 7786 if (priv->flags & NESTED_SYNC_TODO) 7787 net_unlink_todo(dev); 7788 #endif 7789 return 0; 7790 } 7791 7792 int netdev_walk_all_lower_dev_rcu(struct net_device *dev, 7793 int (*fn)(struct net_device *dev, 7794 struct netdev_nested_priv *priv), 7795 struct netdev_nested_priv *priv) 7796 { 7797 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7798 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7799 int ret, cur = 0; 7800 7801 now = dev; 7802 iter = &dev->adj_list.lower; 7803 7804 while (1) { 7805 if (now != dev) { 7806 ret = fn(now, priv); 7807 if (ret) 7808 return ret; 7809 } 7810 7811 next = NULL; 7812 while (1) { 7813 ldev = netdev_next_lower_dev_rcu(now, &iter); 7814 if (!ldev) 7815 break; 7816 7817 next = ldev; 7818 niter = &ldev->adj_list.lower; 7819 dev_stack[cur] = now; 7820 iter_stack[cur++] = iter; 7821 break; 7822 } 7823 7824 if (!next) { 7825 if (!cur) 7826 return 0; 7827 next = dev_stack[--cur]; 7828 niter = iter_stack[cur]; 7829 } 7830 7831 now = next; 7832 iter = niter; 7833 } 7834 7835 return 0; 7836 } 7837 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); 7838 7839 /** 7840 * netdev_lower_get_first_private_rcu - Get the first ->private from the 7841 * lower neighbour list, RCU 7842 * variant 7843 * @dev: device 7844 * 7845 * Gets the first netdev_adjacent->private from the dev's lower neighbour 7846 * list. The caller must hold RCU read lock. 7847 */ 7848 void *netdev_lower_get_first_private_rcu(struct net_device *dev) 7849 { 7850 struct netdev_adjacent *lower; 7851 7852 lower = list_first_or_null_rcu(&dev->adj_list.lower, 7853 struct netdev_adjacent, list); 7854 if (lower) 7855 return lower->private; 7856 return NULL; 7857 } 7858 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); 7859 7860 /** 7861 * netdev_master_upper_dev_get_rcu - Get master upper device 7862 * @dev: device 7863 * 7864 * Find a master upper device and return pointer to it or NULL in case 7865 * it's not there. The caller must hold the RCU read lock. 7866 */ 7867 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) 7868 { 7869 struct netdev_adjacent *upper; 7870 7871 upper = list_first_or_null_rcu(&dev->adj_list.upper, 7872 struct netdev_adjacent, list); 7873 if (upper && likely(upper->master)) 7874 return upper->dev; 7875 return NULL; 7876 } 7877 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); 7878 7879 static int netdev_adjacent_sysfs_add(struct net_device *dev, 7880 struct net_device *adj_dev, 7881 struct list_head *dev_list) 7882 { 7883 char linkname[IFNAMSIZ+7]; 7884 7885 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7886 "upper_%s" : "lower_%s", adj_dev->name); 7887 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), 7888 linkname); 7889 } 7890 static void netdev_adjacent_sysfs_del(struct net_device *dev, 7891 char *name, 7892 struct list_head *dev_list) 7893 { 7894 char linkname[IFNAMSIZ+7]; 7895 7896 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7897 "upper_%s" : "lower_%s", name); 7898 sysfs_remove_link(&(dev->dev.kobj), linkname); 7899 } 7900 7901 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, 7902 struct net_device *adj_dev, 7903 struct list_head *dev_list) 7904 { 7905 return (dev_list == &dev->adj_list.upper || 7906 dev_list == &dev->adj_list.lower) && 7907 net_eq(dev_net(dev), dev_net(adj_dev)); 7908 } 7909 7910 static int __netdev_adjacent_dev_insert(struct net_device *dev, 7911 struct net_device *adj_dev, 7912 struct list_head *dev_list, 7913 void *private, bool master) 7914 { 7915 struct netdev_adjacent *adj; 7916 int ret; 7917 7918 adj = __netdev_find_adj(adj_dev, dev_list); 7919 7920 if (adj) { 7921 adj->ref_nr += 1; 7922 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", 7923 dev->name, adj_dev->name, adj->ref_nr); 7924 7925 return 0; 7926 } 7927 7928 adj = kmalloc(sizeof(*adj), GFP_KERNEL); 7929 if (!adj) 7930 return -ENOMEM; 7931 7932 adj->dev = adj_dev; 7933 adj->master = master; 7934 adj->ref_nr = 1; 7935 adj->private = private; 7936 adj->ignore = false; 7937 dev_hold(adj_dev); 7938 7939 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", 7940 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); 7941 7942 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { 7943 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); 7944 if (ret) 7945 goto free_adj; 7946 } 7947 7948 /* Ensure that master link is always the first item in list. */ 7949 if (master) { 7950 ret = sysfs_create_link(&(dev->dev.kobj), 7951 &(adj_dev->dev.kobj), "master"); 7952 if (ret) 7953 goto remove_symlinks; 7954 7955 list_add_rcu(&adj->list, dev_list); 7956 } else { 7957 list_add_tail_rcu(&adj->list, dev_list); 7958 } 7959 7960 return 0; 7961 7962 remove_symlinks: 7963 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7964 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7965 free_adj: 7966 kfree(adj); 7967 dev_put(adj_dev); 7968 7969 return ret; 7970 } 7971 7972 static void __netdev_adjacent_dev_remove(struct net_device *dev, 7973 struct net_device *adj_dev, 7974 u16 ref_nr, 7975 struct list_head *dev_list) 7976 { 7977 struct netdev_adjacent *adj; 7978 7979 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", 7980 dev->name, adj_dev->name, ref_nr); 7981 7982 adj = __netdev_find_adj(adj_dev, dev_list); 7983 7984 if (!adj) { 7985 pr_err("Adjacency does not exist for device %s from %s\n", 7986 dev->name, adj_dev->name); 7987 WARN_ON(1); 7988 return; 7989 } 7990 7991 if (adj->ref_nr > ref_nr) { 7992 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", 7993 dev->name, adj_dev->name, ref_nr, 7994 adj->ref_nr - ref_nr); 7995 adj->ref_nr -= ref_nr; 7996 return; 7997 } 7998 7999 if (adj->master) 8000 sysfs_remove_link(&(dev->dev.kobj), "master"); 8001 8002 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 8003 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 8004 8005 list_del_rcu(&adj->list); 8006 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", 8007 adj_dev->name, dev->name, adj_dev->name); 8008 dev_put(adj_dev); 8009 kfree_rcu(adj, rcu); 8010 } 8011 8012 static int __netdev_adjacent_dev_link_lists(struct net_device *dev, 8013 struct net_device *upper_dev, 8014 struct list_head *up_list, 8015 struct list_head *down_list, 8016 void *private, bool master) 8017 { 8018 int ret; 8019 8020 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, 8021 private, master); 8022 if (ret) 8023 return ret; 8024 8025 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, 8026 private, false); 8027 if (ret) { 8028 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); 8029 return ret; 8030 } 8031 8032 return 0; 8033 } 8034 8035 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, 8036 struct net_device *upper_dev, 8037 u16 ref_nr, 8038 struct list_head *up_list, 8039 struct list_head *down_list) 8040 { 8041 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); 8042 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); 8043 } 8044 8045 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, 8046 struct net_device *upper_dev, 8047 void *private, bool master) 8048 { 8049 return __netdev_adjacent_dev_link_lists(dev, upper_dev, 8050 &dev->adj_list.upper, 8051 &upper_dev->adj_list.lower, 8052 private, master); 8053 } 8054 8055 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, 8056 struct net_device *upper_dev) 8057 { 8058 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, 8059 &dev->adj_list.upper, 8060 &upper_dev->adj_list.lower); 8061 } 8062 8063 static int __netdev_upper_dev_link(struct net_device *dev, 8064 struct net_device *upper_dev, bool master, 8065 void *upper_priv, void *upper_info, 8066 struct netdev_nested_priv *priv, 8067 struct netlink_ext_ack *extack) 8068 { 8069 struct netdev_notifier_changeupper_info changeupper_info = { 8070 .info = { 8071 .dev = dev, 8072 .extack = extack, 8073 }, 8074 .upper_dev = upper_dev, 8075 .master = master, 8076 .linking = true, 8077 .upper_info = upper_info, 8078 }; 8079 struct net_device *master_dev; 8080 int ret = 0; 8081 8082 ASSERT_RTNL(); 8083 8084 if (dev == upper_dev) 8085 return -EBUSY; 8086 8087 /* To prevent loops, check if dev is not upper device to upper_dev. */ 8088 if (__netdev_has_upper_dev(upper_dev, dev)) 8089 return -EBUSY; 8090 8091 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV) 8092 return -EMLINK; 8093 8094 if (!master) { 8095 if (__netdev_has_upper_dev(dev, upper_dev)) 8096 return -EEXIST; 8097 } else { 8098 master_dev = __netdev_master_upper_dev_get(dev); 8099 if (master_dev) 8100 return master_dev == upper_dev ? -EEXIST : -EBUSY; 8101 } 8102 8103 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 8104 &changeupper_info.info); 8105 ret = notifier_to_errno(ret); 8106 if (ret) 8107 return ret; 8108 8109 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, 8110 master); 8111 if (ret) 8112 return ret; 8113 8114 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 8115 &changeupper_info.info); 8116 ret = notifier_to_errno(ret); 8117 if (ret) 8118 goto rollback; 8119 8120 __netdev_update_upper_level(dev, NULL); 8121 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 8122 8123 __netdev_update_lower_level(upper_dev, priv); 8124 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 8125 priv); 8126 8127 return 0; 8128 8129 rollback: 8130 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 8131 8132 return ret; 8133 } 8134 8135 /** 8136 * netdev_upper_dev_link - Add a link to the upper device 8137 * @dev: device 8138 * @upper_dev: new upper device 8139 * @extack: netlink extended ack 8140 * 8141 * Adds a link to device which is upper to this one. The caller must hold 8142 * the RTNL lock. On a failure a negative errno code is returned. 8143 * On success the reference counts are adjusted and the function 8144 * returns zero. 8145 */ 8146 int netdev_upper_dev_link(struct net_device *dev, 8147 struct net_device *upper_dev, 8148 struct netlink_ext_ack *extack) 8149 { 8150 struct netdev_nested_priv priv = { 8151 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 8152 .data = NULL, 8153 }; 8154 8155 return __netdev_upper_dev_link(dev, upper_dev, false, 8156 NULL, NULL, &priv, extack); 8157 } 8158 EXPORT_SYMBOL(netdev_upper_dev_link); 8159 8160 /** 8161 * netdev_master_upper_dev_link - Add a master link to the upper device 8162 * @dev: device 8163 * @upper_dev: new upper device 8164 * @upper_priv: upper device private 8165 * @upper_info: upper info to be passed down via notifier 8166 * @extack: netlink extended ack 8167 * 8168 * Adds a link to device which is upper to this one. In this case, only 8169 * one master upper device can be linked, although other non-master devices 8170 * might be linked as well. The caller must hold the RTNL lock. 8171 * On a failure a negative errno code is returned. On success the reference 8172 * counts are adjusted and the function returns zero. 8173 */ 8174 int netdev_master_upper_dev_link(struct net_device *dev, 8175 struct net_device *upper_dev, 8176 void *upper_priv, void *upper_info, 8177 struct netlink_ext_ack *extack) 8178 { 8179 struct netdev_nested_priv priv = { 8180 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 8181 .data = NULL, 8182 }; 8183 8184 return __netdev_upper_dev_link(dev, upper_dev, true, 8185 upper_priv, upper_info, &priv, extack); 8186 } 8187 EXPORT_SYMBOL(netdev_master_upper_dev_link); 8188 8189 static void __netdev_upper_dev_unlink(struct net_device *dev, 8190 struct net_device *upper_dev, 8191 struct netdev_nested_priv *priv) 8192 { 8193 struct netdev_notifier_changeupper_info changeupper_info = { 8194 .info = { 8195 .dev = dev, 8196 }, 8197 .upper_dev = upper_dev, 8198 .linking = false, 8199 }; 8200 8201 ASSERT_RTNL(); 8202 8203 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; 8204 8205 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 8206 &changeupper_info.info); 8207 8208 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 8209 8210 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 8211 &changeupper_info.info); 8212 8213 __netdev_update_upper_level(dev, NULL); 8214 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 8215 8216 __netdev_update_lower_level(upper_dev, priv); 8217 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 8218 priv); 8219 } 8220 8221 /** 8222 * netdev_upper_dev_unlink - Removes a link to upper device 8223 * @dev: device 8224 * @upper_dev: new upper device 8225 * 8226 * Removes a link to device which is upper to this one. The caller must hold 8227 * the RTNL lock. 8228 */ 8229 void netdev_upper_dev_unlink(struct net_device *dev, 8230 struct net_device *upper_dev) 8231 { 8232 struct netdev_nested_priv priv = { 8233 .flags = NESTED_SYNC_TODO, 8234 .data = NULL, 8235 }; 8236 8237 __netdev_upper_dev_unlink(dev, upper_dev, &priv); 8238 } 8239 EXPORT_SYMBOL(netdev_upper_dev_unlink); 8240 8241 static void __netdev_adjacent_dev_set(struct net_device *upper_dev, 8242 struct net_device *lower_dev, 8243 bool val) 8244 { 8245 struct netdev_adjacent *adj; 8246 8247 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower); 8248 if (adj) 8249 adj->ignore = val; 8250 8251 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper); 8252 if (adj) 8253 adj->ignore = val; 8254 } 8255 8256 static void netdev_adjacent_dev_disable(struct net_device *upper_dev, 8257 struct net_device *lower_dev) 8258 { 8259 __netdev_adjacent_dev_set(upper_dev, lower_dev, true); 8260 } 8261 8262 static void netdev_adjacent_dev_enable(struct net_device *upper_dev, 8263 struct net_device *lower_dev) 8264 { 8265 __netdev_adjacent_dev_set(upper_dev, lower_dev, false); 8266 } 8267 8268 int netdev_adjacent_change_prepare(struct net_device *old_dev, 8269 struct net_device *new_dev, 8270 struct net_device *dev, 8271 struct netlink_ext_ack *extack) 8272 { 8273 struct netdev_nested_priv priv = { 8274 .flags = 0, 8275 .data = NULL, 8276 }; 8277 int err; 8278 8279 if (!new_dev) 8280 return 0; 8281 8282 if (old_dev && new_dev != old_dev) 8283 netdev_adjacent_dev_disable(dev, old_dev); 8284 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv, 8285 extack); 8286 if (err) { 8287 if (old_dev && new_dev != old_dev) 8288 netdev_adjacent_dev_enable(dev, old_dev); 8289 return err; 8290 } 8291 8292 return 0; 8293 } 8294 EXPORT_SYMBOL(netdev_adjacent_change_prepare); 8295 8296 void netdev_adjacent_change_commit(struct net_device *old_dev, 8297 struct net_device *new_dev, 8298 struct net_device *dev) 8299 { 8300 struct netdev_nested_priv priv = { 8301 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 8302 .data = NULL, 8303 }; 8304 8305 if (!new_dev || !old_dev) 8306 return; 8307 8308 if (new_dev == old_dev) 8309 return; 8310 8311 netdev_adjacent_dev_enable(dev, old_dev); 8312 __netdev_upper_dev_unlink(old_dev, dev, &priv); 8313 } 8314 EXPORT_SYMBOL(netdev_adjacent_change_commit); 8315 8316 void netdev_adjacent_change_abort(struct net_device *old_dev, 8317 struct net_device *new_dev, 8318 struct net_device *dev) 8319 { 8320 struct netdev_nested_priv priv = { 8321 .flags = 0, 8322 .data = NULL, 8323 }; 8324 8325 if (!new_dev) 8326 return; 8327 8328 if (old_dev && new_dev != old_dev) 8329 netdev_adjacent_dev_enable(dev, old_dev); 8330 8331 __netdev_upper_dev_unlink(new_dev, dev, &priv); 8332 } 8333 EXPORT_SYMBOL(netdev_adjacent_change_abort); 8334 8335 /** 8336 * netdev_bonding_info_change - Dispatch event about slave change 8337 * @dev: device 8338 * @bonding_info: info to dispatch 8339 * 8340 * Send NETDEV_BONDING_INFO to netdev notifiers with info. 8341 * The caller must hold the RTNL lock. 8342 */ 8343 void netdev_bonding_info_change(struct net_device *dev, 8344 struct netdev_bonding_info *bonding_info) 8345 { 8346 struct netdev_notifier_bonding_info info = { 8347 .info.dev = dev, 8348 }; 8349 8350 memcpy(&info.bonding_info, bonding_info, 8351 sizeof(struct netdev_bonding_info)); 8352 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, 8353 &info.info); 8354 } 8355 EXPORT_SYMBOL(netdev_bonding_info_change); 8356 8357 /** 8358 * netdev_get_xmit_slave - Get the xmit slave of master device 8359 * @dev: device 8360 * @skb: The packet 8361 * @all_slaves: assume all the slaves are active 8362 * 8363 * The reference counters are not incremented so the caller must be 8364 * careful with locks. The caller must hold RCU lock. 8365 * %NULL is returned if no slave is found. 8366 */ 8367 8368 struct net_device *netdev_get_xmit_slave(struct net_device *dev, 8369 struct sk_buff *skb, 8370 bool all_slaves) 8371 { 8372 const struct net_device_ops *ops = dev->netdev_ops; 8373 8374 if (!ops->ndo_get_xmit_slave) 8375 return NULL; 8376 return ops->ndo_get_xmit_slave(dev, skb, all_slaves); 8377 } 8378 EXPORT_SYMBOL(netdev_get_xmit_slave); 8379 8380 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev, 8381 struct sock *sk) 8382 { 8383 const struct net_device_ops *ops = dev->netdev_ops; 8384 8385 if (!ops->ndo_sk_get_lower_dev) 8386 return NULL; 8387 return ops->ndo_sk_get_lower_dev(dev, sk); 8388 } 8389 8390 /** 8391 * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket 8392 * @dev: device 8393 * @sk: the socket 8394 * 8395 * %NULL is returned if no lower device is found. 8396 */ 8397 8398 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev, 8399 struct sock *sk) 8400 { 8401 struct net_device *lower; 8402 8403 lower = netdev_sk_get_lower_dev(dev, sk); 8404 while (lower) { 8405 dev = lower; 8406 lower = netdev_sk_get_lower_dev(dev, sk); 8407 } 8408 8409 return dev; 8410 } 8411 EXPORT_SYMBOL(netdev_sk_get_lowest_dev); 8412 8413 static void netdev_adjacent_add_links(struct net_device *dev) 8414 { 8415 struct netdev_adjacent *iter; 8416 8417 struct net *net = dev_net(dev); 8418 8419 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8420 if (!net_eq(net, dev_net(iter->dev))) 8421 continue; 8422 netdev_adjacent_sysfs_add(iter->dev, dev, 8423 &iter->dev->adj_list.lower); 8424 netdev_adjacent_sysfs_add(dev, iter->dev, 8425 &dev->adj_list.upper); 8426 } 8427 8428 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8429 if (!net_eq(net, dev_net(iter->dev))) 8430 continue; 8431 netdev_adjacent_sysfs_add(iter->dev, dev, 8432 &iter->dev->adj_list.upper); 8433 netdev_adjacent_sysfs_add(dev, iter->dev, 8434 &dev->adj_list.lower); 8435 } 8436 } 8437 8438 static void netdev_adjacent_del_links(struct net_device *dev) 8439 { 8440 struct netdev_adjacent *iter; 8441 8442 struct net *net = dev_net(dev); 8443 8444 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8445 if (!net_eq(net, dev_net(iter->dev))) 8446 continue; 8447 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8448 &iter->dev->adj_list.lower); 8449 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8450 &dev->adj_list.upper); 8451 } 8452 8453 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8454 if (!net_eq(net, dev_net(iter->dev))) 8455 continue; 8456 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8457 &iter->dev->adj_list.upper); 8458 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8459 &dev->adj_list.lower); 8460 } 8461 } 8462 8463 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) 8464 { 8465 struct netdev_adjacent *iter; 8466 8467 struct net *net = dev_net(dev); 8468 8469 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8470 if (!net_eq(net, dev_net(iter->dev))) 8471 continue; 8472 netdev_adjacent_sysfs_del(iter->dev, oldname, 8473 &iter->dev->adj_list.lower); 8474 netdev_adjacent_sysfs_add(iter->dev, dev, 8475 &iter->dev->adj_list.lower); 8476 } 8477 8478 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8479 if (!net_eq(net, dev_net(iter->dev))) 8480 continue; 8481 netdev_adjacent_sysfs_del(iter->dev, oldname, 8482 &iter->dev->adj_list.upper); 8483 netdev_adjacent_sysfs_add(iter->dev, dev, 8484 &iter->dev->adj_list.upper); 8485 } 8486 } 8487 8488 void *netdev_lower_dev_get_private(struct net_device *dev, 8489 struct net_device *lower_dev) 8490 { 8491 struct netdev_adjacent *lower; 8492 8493 if (!lower_dev) 8494 return NULL; 8495 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); 8496 if (!lower) 8497 return NULL; 8498 8499 return lower->private; 8500 } 8501 EXPORT_SYMBOL(netdev_lower_dev_get_private); 8502 8503 8504 /** 8505 * netdev_lower_state_changed - Dispatch event about lower device state change 8506 * @lower_dev: device 8507 * @lower_state_info: state to dispatch 8508 * 8509 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. 8510 * The caller must hold the RTNL lock. 8511 */ 8512 void netdev_lower_state_changed(struct net_device *lower_dev, 8513 void *lower_state_info) 8514 { 8515 struct netdev_notifier_changelowerstate_info changelowerstate_info = { 8516 .info.dev = lower_dev, 8517 }; 8518 8519 ASSERT_RTNL(); 8520 changelowerstate_info.lower_state_info = lower_state_info; 8521 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, 8522 &changelowerstate_info.info); 8523 } 8524 EXPORT_SYMBOL(netdev_lower_state_changed); 8525 8526 static void dev_change_rx_flags(struct net_device *dev, int flags) 8527 { 8528 const struct net_device_ops *ops = dev->netdev_ops; 8529 8530 if (ops->ndo_change_rx_flags) 8531 ops->ndo_change_rx_flags(dev, flags); 8532 } 8533 8534 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) 8535 { 8536 unsigned int old_flags = dev->flags; 8537 kuid_t uid; 8538 kgid_t gid; 8539 8540 ASSERT_RTNL(); 8541 8542 dev->flags |= IFF_PROMISC; 8543 dev->promiscuity += inc; 8544 if (dev->promiscuity == 0) { 8545 /* 8546 * Avoid overflow. 8547 * If inc causes overflow, untouch promisc and return error. 8548 */ 8549 if (inc < 0) 8550 dev->flags &= ~IFF_PROMISC; 8551 else { 8552 dev->promiscuity -= inc; 8553 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n", 8554 dev->name); 8555 return -EOVERFLOW; 8556 } 8557 } 8558 if (dev->flags != old_flags) { 8559 pr_info("device %s %s promiscuous mode\n", 8560 dev->name, 8561 dev->flags & IFF_PROMISC ? "entered" : "left"); 8562 if (audit_enabled) { 8563 current_uid_gid(&uid, &gid); 8564 audit_log(audit_context(), GFP_ATOMIC, 8565 AUDIT_ANOM_PROMISCUOUS, 8566 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", 8567 dev->name, (dev->flags & IFF_PROMISC), 8568 (old_flags & IFF_PROMISC), 8569 from_kuid(&init_user_ns, audit_get_loginuid(current)), 8570 from_kuid(&init_user_ns, uid), 8571 from_kgid(&init_user_ns, gid), 8572 audit_get_sessionid(current)); 8573 } 8574 8575 dev_change_rx_flags(dev, IFF_PROMISC); 8576 } 8577 if (notify) 8578 __dev_notify_flags(dev, old_flags, IFF_PROMISC); 8579 return 0; 8580 } 8581 8582 /** 8583 * dev_set_promiscuity - update promiscuity count on a device 8584 * @dev: device 8585 * @inc: modifier 8586 * 8587 * Add or remove promiscuity from a device. While the count in the device 8588 * remains above zero the interface remains promiscuous. Once it hits zero 8589 * the device reverts back to normal filtering operation. A negative inc 8590 * value is used to drop promiscuity on the device. 8591 * Return 0 if successful or a negative errno code on error. 8592 */ 8593 int dev_set_promiscuity(struct net_device *dev, int inc) 8594 { 8595 unsigned int old_flags = dev->flags; 8596 int err; 8597 8598 err = __dev_set_promiscuity(dev, inc, true); 8599 if (err < 0) 8600 return err; 8601 if (dev->flags != old_flags) 8602 dev_set_rx_mode(dev); 8603 return err; 8604 } 8605 EXPORT_SYMBOL(dev_set_promiscuity); 8606 8607 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) 8608 { 8609 unsigned int old_flags = dev->flags, old_gflags = dev->gflags; 8610 8611 ASSERT_RTNL(); 8612 8613 dev->flags |= IFF_ALLMULTI; 8614 dev->allmulti += inc; 8615 if (dev->allmulti == 0) { 8616 /* 8617 * Avoid overflow. 8618 * If inc causes overflow, untouch allmulti and return error. 8619 */ 8620 if (inc < 0) 8621 dev->flags &= ~IFF_ALLMULTI; 8622 else { 8623 dev->allmulti -= inc; 8624 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n", 8625 dev->name); 8626 return -EOVERFLOW; 8627 } 8628 } 8629 if (dev->flags ^ old_flags) { 8630 dev_change_rx_flags(dev, IFF_ALLMULTI); 8631 dev_set_rx_mode(dev); 8632 if (notify) 8633 __dev_notify_flags(dev, old_flags, 8634 dev->gflags ^ old_gflags); 8635 } 8636 return 0; 8637 } 8638 8639 /** 8640 * dev_set_allmulti - update allmulti count on a device 8641 * @dev: device 8642 * @inc: modifier 8643 * 8644 * Add or remove reception of all multicast frames to a device. While the 8645 * count in the device remains above zero the interface remains listening 8646 * to all interfaces. Once it hits zero the device reverts back to normal 8647 * filtering operation. A negative @inc value is used to drop the counter 8648 * when releasing a resource needing all multicasts. 8649 * Return 0 if successful or a negative errno code on error. 8650 */ 8651 8652 int dev_set_allmulti(struct net_device *dev, int inc) 8653 { 8654 return __dev_set_allmulti(dev, inc, true); 8655 } 8656 EXPORT_SYMBOL(dev_set_allmulti); 8657 8658 /* 8659 * Upload unicast and multicast address lists to device and 8660 * configure RX filtering. When the device doesn't support unicast 8661 * filtering it is put in promiscuous mode while unicast addresses 8662 * are present. 8663 */ 8664 void __dev_set_rx_mode(struct net_device *dev) 8665 { 8666 const struct net_device_ops *ops = dev->netdev_ops; 8667 8668 /* dev_open will call this function so the list will stay sane. */ 8669 if (!(dev->flags&IFF_UP)) 8670 return; 8671 8672 if (!netif_device_present(dev)) 8673 return; 8674 8675 if (!(dev->priv_flags & IFF_UNICAST_FLT)) { 8676 /* Unicast addresses changes may only happen under the rtnl, 8677 * therefore calling __dev_set_promiscuity here is safe. 8678 */ 8679 if (!netdev_uc_empty(dev) && !dev->uc_promisc) { 8680 __dev_set_promiscuity(dev, 1, false); 8681 dev->uc_promisc = true; 8682 } else if (netdev_uc_empty(dev) && dev->uc_promisc) { 8683 __dev_set_promiscuity(dev, -1, false); 8684 dev->uc_promisc = false; 8685 } 8686 } 8687 8688 if (ops->ndo_set_rx_mode) 8689 ops->ndo_set_rx_mode(dev); 8690 } 8691 8692 void dev_set_rx_mode(struct net_device *dev) 8693 { 8694 netif_addr_lock_bh(dev); 8695 __dev_set_rx_mode(dev); 8696 netif_addr_unlock_bh(dev); 8697 } 8698 8699 /** 8700 * dev_get_flags - get flags reported to userspace 8701 * @dev: device 8702 * 8703 * Get the combination of flag bits exported through APIs to userspace. 8704 */ 8705 unsigned int dev_get_flags(const struct net_device *dev) 8706 { 8707 unsigned int flags; 8708 8709 flags = (dev->flags & ~(IFF_PROMISC | 8710 IFF_ALLMULTI | 8711 IFF_RUNNING | 8712 IFF_LOWER_UP | 8713 IFF_DORMANT)) | 8714 (dev->gflags & (IFF_PROMISC | 8715 IFF_ALLMULTI)); 8716 8717 if (netif_running(dev)) { 8718 if (netif_oper_up(dev)) 8719 flags |= IFF_RUNNING; 8720 if (netif_carrier_ok(dev)) 8721 flags |= IFF_LOWER_UP; 8722 if (netif_dormant(dev)) 8723 flags |= IFF_DORMANT; 8724 } 8725 8726 return flags; 8727 } 8728 EXPORT_SYMBOL(dev_get_flags); 8729 8730 int __dev_change_flags(struct net_device *dev, unsigned int flags, 8731 struct netlink_ext_ack *extack) 8732 { 8733 unsigned int old_flags = dev->flags; 8734 int ret; 8735 8736 ASSERT_RTNL(); 8737 8738 /* 8739 * Set the flags on our device. 8740 */ 8741 8742 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | 8743 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | 8744 IFF_AUTOMEDIA)) | 8745 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | 8746 IFF_ALLMULTI)); 8747 8748 /* 8749 * Load in the correct multicast list now the flags have changed. 8750 */ 8751 8752 if ((old_flags ^ flags) & IFF_MULTICAST) 8753 dev_change_rx_flags(dev, IFF_MULTICAST); 8754 8755 dev_set_rx_mode(dev); 8756 8757 /* 8758 * Have we downed the interface. We handle IFF_UP ourselves 8759 * according to user attempts to set it, rather than blindly 8760 * setting it. 8761 */ 8762 8763 ret = 0; 8764 if ((old_flags ^ flags) & IFF_UP) { 8765 if (old_flags & IFF_UP) 8766 __dev_close(dev); 8767 else 8768 ret = __dev_open(dev, extack); 8769 } 8770 8771 if ((flags ^ dev->gflags) & IFF_PROMISC) { 8772 int inc = (flags & IFF_PROMISC) ? 1 : -1; 8773 unsigned int old_flags = dev->flags; 8774 8775 dev->gflags ^= IFF_PROMISC; 8776 8777 if (__dev_set_promiscuity(dev, inc, false) >= 0) 8778 if (dev->flags != old_flags) 8779 dev_set_rx_mode(dev); 8780 } 8781 8782 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI 8783 * is important. Some (broken) drivers set IFF_PROMISC, when 8784 * IFF_ALLMULTI is requested not asking us and not reporting. 8785 */ 8786 if ((flags ^ dev->gflags) & IFF_ALLMULTI) { 8787 int inc = (flags & IFF_ALLMULTI) ? 1 : -1; 8788 8789 dev->gflags ^= IFF_ALLMULTI; 8790 __dev_set_allmulti(dev, inc, false); 8791 } 8792 8793 return ret; 8794 } 8795 8796 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, 8797 unsigned int gchanges) 8798 { 8799 unsigned int changes = dev->flags ^ old_flags; 8800 8801 if (gchanges) 8802 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC); 8803 8804 if (changes & IFF_UP) { 8805 if (dev->flags & IFF_UP) 8806 call_netdevice_notifiers(NETDEV_UP, dev); 8807 else 8808 call_netdevice_notifiers(NETDEV_DOWN, dev); 8809 } 8810 8811 if (dev->flags & IFF_UP && 8812 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { 8813 struct netdev_notifier_change_info change_info = { 8814 .info = { 8815 .dev = dev, 8816 }, 8817 .flags_changed = changes, 8818 }; 8819 8820 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info); 8821 } 8822 } 8823 8824 /** 8825 * dev_change_flags - change device settings 8826 * @dev: device 8827 * @flags: device state flags 8828 * @extack: netlink extended ack 8829 * 8830 * Change settings on device based state flags. The flags are 8831 * in the userspace exported format. 8832 */ 8833 int dev_change_flags(struct net_device *dev, unsigned int flags, 8834 struct netlink_ext_ack *extack) 8835 { 8836 int ret; 8837 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; 8838 8839 ret = __dev_change_flags(dev, flags, extack); 8840 if (ret < 0) 8841 return ret; 8842 8843 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); 8844 __dev_notify_flags(dev, old_flags, changes); 8845 return ret; 8846 } 8847 EXPORT_SYMBOL(dev_change_flags); 8848 8849 int __dev_set_mtu(struct net_device *dev, int new_mtu) 8850 { 8851 const struct net_device_ops *ops = dev->netdev_ops; 8852 8853 if (ops->ndo_change_mtu) 8854 return ops->ndo_change_mtu(dev, new_mtu); 8855 8856 /* Pairs with all the lockless reads of dev->mtu in the stack */ 8857 WRITE_ONCE(dev->mtu, new_mtu); 8858 return 0; 8859 } 8860 EXPORT_SYMBOL(__dev_set_mtu); 8861 8862 int dev_validate_mtu(struct net_device *dev, int new_mtu, 8863 struct netlink_ext_ack *extack) 8864 { 8865 /* MTU must be positive, and in range */ 8866 if (new_mtu < 0 || new_mtu < dev->min_mtu) { 8867 NL_SET_ERR_MSG(extack, "mtu less than device minimum"); 8868 return -EINVAL; 8869 } 8870 8871 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { 8872 NL_SET_ERR_MSG(extack, "mtu greater than device maximum"); 8873 return -EINVAL; 8874 } 8875 return 0; 8876 } 8877 8878 /** 8879 * dev_set_mtu_ext - Change maximum transfer unit 8880 * @dev: device 8881 * @new_mtu: new transfer unit 8882 * @extack: netlink extended ack 8883 * 8884 * Change the maximum transfer size of the network device. 8885 */ 8886 int dev_set_mtu_ext(struct net_device *dev, int new_mtu, 8887 struct netlink_ext_ack *extack) 8888 { 8889 int err, orig_mtu; 8890 8891 if (new_mtu == dev->mtu) 8892 return 0; 8893 8894 err = dev_validate_mtu(dev, new_mtu, extack); 8895 if (err) 8896 return err; 8897 8898 if (!netif_device_present(dev)) 8899 return -ENODEV; 8900 8901 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); 8902 err = notifier_to_errno(err); 8903 if (err) 8904 return err; 8905 8906 orig_mtu = dev->mtu; 8907 err = __dev_set_mtu(dev, new_mtu); 8908 8909 if (!err) { 8910 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8911 orig_mtu); 8912 err = notifier_to_errno(err); 8913 if (err) { 8914 /* setting mtu back and notifying everyone again, 8915 * so that they have a chance to revert changes. 8916 */ 8917 __dev_set_mtu(dev, orig_mtu); 8918 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8919 new_mtu); 8920 } 8921 } 8922 return err; 8923 } 8924 8925 int dev_set_mtu(struct net_device *dev, int new_mtu) 8926 { 8927 struct netlink_ext_ack extack; 8928 int err; 8929 8930 memset(&extack, 0, sizeof(extack)); 8931 err = dev_set_mtu_ext(dev, new_mtu, &extack); 8932 if (err && extack._msg) 8933 net_err_ratelimited("%s: %s\n", dev->name, extack._msg); 8934 return err; 8935 } 8936 EXPORT_SYMBOL(dev_set_mtu); 8937 8938 /** 8939 * dev_change_tx_queue_len - Change TX queue length of a netdevice 8940 * @dev: device 8941 * @new_len: new tx queue length 8942 */ 8943 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len) 8944 { 8945 unsigned int orig_len = dev->tx_queue_len; 8946 int res; 8947 8948 if (new_len != (unsigned int)new_len) 8949 return -ERANGE; 8950 8951 if (new_len != orig_len) { 8952 dev->tx_queue_len = new_len; 8953 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev); 8954 res = notifier_to_errno(res); 8955 if (res) 8956 goto err_rollback; 8957 res = dev_qdisc_change_tx_queue_len(dev); 8958 if (res) 8959 goto err_rollback; 8960 } 8961 8962 return 0; 8963 8964 err_rollback: 8965 netdev_err(dev, "refused to change device tx_queue_len\n"); 8966 dev->tx_queue_len = orig_len; 8967 return res; 8968 } 8969 8970 /** 8971 * dev_set_group - Change group this device belongs to 8972 * @dev: device 8973 * @new_group: group this device should belong to 8974 */ 8975 void dev_set_group(struct net_device *dev, int new_group) 8976 { 8977 dev->group = new_group; 8978 } 8979 EXPORT_SYMBOL(dev_set_group); 8980 8981 /** 8982 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR. 8983 * @dev: device 8984 * @addr: new address 8985 * @extack: netlink extended ack 8986 */ 8987 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr, 8988 struct netlink_ext_ack *extack) 8989 { 8990 struct netdev_notifier_pre_changeaddr_info info = { 8991 .info.dev = dev, 8992 .info.extack = extack, 8993 .dev_addr = addr, 8994 }; 8995 int rc; 8996 8997 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info); 8998 return notifier_to_errno(rc); 8999 } 9000 EXPORT_SYMBOL(dev_pre_changeaddr_notify); 9001 9002 /** 9003 * dev_set_mac_address - Change Media Access Control Address 9004 * @dev: device 9005 * @sa: new address 9006 * @extack: netlink extended ack 9007 * 9008 * Change the hardware (MAC) address of the device 9009 */ 9010 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa, 9011 struct netlink_ext_ack *extack) 9012 { 9013 const struct net_device_ops *ops = dev->netdev_ops; 9014 int err; 9015 9016 if (!ops->ndo_set_mac_address) 9017 return -EOPNOTSUPP; 9018 if (sa->sa_family != dev->type) 9019 return -EINVAL; 9020 if (!netif_device_present(dev)) 9021 return -ENODEV; 9022 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack); 9023 if (err) 9024 return err; 9025 err = ops->ndo_set_mac_address(dev, sa); 9026 if (err) 9027 return err; 9028 dev->addr_assign_type = NET_ADDR_SET; 9029 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); 9030 add_device_randomness(dev->dev_addr, dev->addr_len); 9031 return 0; 9032 } 9033 EXPORT_SYMBOL(dev_set_mac_address); 9034 9035 static DECLARE_RWSEM(dev_addr_sem); 9036 9037 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa, 9038 struct netlink_ext_ack *extack) 9039 { 9040 int ret; 9041 9042 down_write(&dev_addr_sem); 9043 ret = dev_set_mac_address(dev, sa, extack); 9044 up_write(&dev_addr_sem); 9045 return ret; 9046 } 9047 EXPORT_SYMBOL(dev_set_mac_address_user); 9048 9049 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name) 9050 { 9051 size_t size = sizeof(sa->sa_data); 9052 struct net_device *dev; 9053 int ret = 0; 9054 9055 down_read(&dev_addr_sem); 9056 rcu_read_lock(); 9057 9058 dev = dev_get_by_name_rcu(net, dev_name); 9059 if (!dev) { 9060 ret = -ENODEV; 9061 goto unlock; 9062 } 9063 if (!dev->addr_len) 9064 memset(sa->sa_data, 0, size); 9065 else 9066 memcpy(sa->sa_data, dev->dev_addr, 9067 min_t(size_t, size, dev->addr_len)); 9068 sa->sa_family = dev->type; 9069 9070 unlock: 9071 rcu_read_unlock(); 9072 up_read(&dev_addr_sem); 9073 return ret; 9074 } 9075 EXPORT_SYMBOL(dev_get_mac_address); 9076 9077 /** 9078 * dev_change_carrier - Change device carrier 9079 * @dev: device 9080 * @new_carrier: new value 9081 * 9082 * Change device carrier 9083 */ 9084 int dev_change_carrier(struct net_device *dev, bool new_carrier) 9085 { 9086 const struct net_device_ops *ops = dev->netdev_ops; 9087 9088 if (!ops->ndo_change_carrier) 9089 return -EOPNOTSUPP; 9090 if (!netif_device_present(dev)) 9091 return -ENODEV; 9092 return ops->ndo_change_carrier(dev, new_carrier); 9093 } 9094 EXPORT_SYMBOL(dev_change_carrier); 9095 9096 /** 9097 * dev_get_phys_port_id - Get device physical port ID 9098 * @dev: device 9099 * @ppid: port ID 9100 * 9101 * Get device physical port ID 9102 */ 9103 int dev_get_phys_port_id(struct net_device *dev, 9104 struct netdev_phys_item_id *ppid) 9105 { 9106 const struct net_device_ops *ops = dev->netdev_ops; 9107 9108 if (!ops->ndo_get_phys_port_id) 9109 return -EOPNOTSUPP; 9110 return ops->ndo_get_phys_port_id(dev, ppid); 9111 } 9112 EXPORT_SYMBOL(dev_get_phys_port_id); 9113 9114 /** 9115 * dev_get_phys_port_name - Get device physical port name 9116 * @dev: device 9117 * @name: port name 9118 * @len: limit of bytes to copy to name 9119 * 9120 * Get device physical port name 9121 */ 9122 int dev_get_phys_port_name(struct net_device *dev, 9123 char *name, size_t len) 9124 { 9125 const struct net_device_ops *ops = dev->netdev_ops; 9126 int err; 9127 9128 if (ops->ndo_get_phys_port_name) { 9129 err = ops->ndo_get_phys_port_name(dev, name, len); 9130 if (err != -EOPNOTSUPP) 9131 return err; 9132 } 9133 return devlink_compat_phys_port_name_get(dev, name, len); 9134 } 9135 EXPORT_SYMBOL(dev_get_phys_port_name); 9136 9137 /** 9138 * dev_get_port_parent_id - Get the device's port parent identifier 9139 * @dev: network device 9140 * @ppid: pointer to a storage for the port's parent identifier 9141 * @recurse: allow/disallow recursion to lower devices 9142 * 9143 * Get the devices's port parent identifier 9144 */ 9145 int dev_get_port_parent_id(struct net_device *dev, 9146 struct netdev_phys_item_id *ppid, 9147 bool recurse) 9148 { 9149 const struct net_device_ops *ops = dev->netdev_ops; 9150 struct netdev_phys_item_id first = { }; 9151 struct net_device *lower_dev; 9152 struct list_head *iter; 9153 int err; 9154 9155 if (ops->ndo_get_port_parent_id) { 9156 err = ops->ndo_get_port_parent_id(dev, ppid); 9157 if (err != -EOPNOTSUPP) 9158 return err; 9159 } 9160 9161 err = devlink_compat_switch_id_get(dev, ppid); 9162 if (!err || err != -EOPNOTSUPP) 9163 return err; 9164 9165 if (!recurse) 9166 return -EOPNOTSUPP; 9167 9168 netdev_for_each_lower_dev(dev, lower_dev, iter) { 9169 err = dev_get_port_parent_id(lower_dev, ppid, recurse); 9170 if (err) 9171 break; 9172 if (!first.id_len) 9173 first = *ppid; 9174 else if (memcmp(&first, ppid, sizeof(*ppid))) 9175 return -EOPNOTSUPP; 9176 } 9177 9178 return err; 9179 } 9180 EXPORT_SYMBOL(dev_get_port_parent_id); 9181 9182 /** 9183 * netdev_port_same_parent_id - Indicate if two network devices have 9184 * the same port parent identifier 9185 * @a: first network device 9186 * @b: second network device 9187 */ 9188 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b) 9189 { 9190 struct netdev_phys_item_id a_id = { }; 9191 struct netdev_phys_item_id b_id = { }; 9192 9193 if (dev_get_port_parent_id(a, &a_id, true) || 9194 dev_get_port_parent_id(b, &b_id, true)) 9195 return false; 9196 9197 return netdev_phys_item_id_same(&a_id, &b_id); 9198 } 9199 EXPORT_SYMBOL(netdev_port_same_parent_id); 9200 9201 /** 9202 * dev_change_proto_down - update protocol port state information 9203 * @dev: device 9204 * @proto_down: new value 9205 * 9206 * This info can be used by switch drivers to set the phys state of the 9207 * port. 9208 */ 9209 int dev_change_proto_down(struct net_device *dev, bool proto_down) 9210 { 9211 const struct net_device_ops *ops = dev->netdev_ops; 9212 9213 if (!ops->ndo_change_proto_down) 9214 return -EOPNOTSUPP; 9215 if (!netif_device_present(dev)) 9216 return -ENODEV; 9217 return ops->ndo_change_proto_down(dev, proto_down); 9218 } 9219 EXPORT_SYMBOL(dev_change_proto_down); 9220 9221 /** 9222 * dev_change_proto_down_generic - generic implementation for 9223 * ndo_change_proto_down that sets carrier according to 9224 * proto_down. 9225 * 9226 * @dev: device 9227 * @proto_down: new value 9228 */ 9229 int dev_change_proto_down_generic(struct net_device *dev, bool proto_down) 9230 { 9231 if (proto_down) 9232 netif_carrier_off(dev); 9233 else 9234 netif_carrier_on(dev); 9235 dev->proto_down = proto_down; 9236 return 0; 9237 } 9238 EXPORT_SYMBOL(dev_change_proto_down_generic); 9239 9240 /** 9241 * dev_change_proto_down_reason - proto down reason 9242 * 9243 * @dev: device 9244 * @mask: proto down mask 9245 * @value: proto down value 9246 */ 9247 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask, 9248 u32 value) 9249 { 9250 int b; 9251 9252 if (!mask) { 9253 dev->proto_down_reason = value; 9254 } else { 9255 for_each_set_bit(b, &mask, 32) { 9256 if (value & (1 << b)) 9257 dev->proto_down_reason |= BIT(b); 9258 else 9259 dev->proto_down_reason &= ~BIT(b); 9260 } 9261 } 9262 } 9263 EXPORT_SYMBOL(dev_change_proto_down_reason); 9264 9265 struct bpf_xdp_link { 9266 struct bpf_link link; 9267 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */ 9268 int flags; 9269 }; 9270 9271 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags) 9272 { 9273 if (flags & XDP_FLAGS_HW_MODE) 9274 return XDP_MODE_HW; 9275 if (flags & XDP_FLAGS_DRV_MODE) 9276 return XDP_MODE_DRV; 9277 if (flags & XDP_FLAGS_SKB_MODE) 9278 return XDP_MODE_SKB; 9279 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB; 9280 } 9281 9282 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode) 9283 { 9284 switch (mode) { 9285 case XDP_MODE_SKB: 9286 return generic_xdp_install; 9287 case XDP_MODE_DRV: 9288 case XDP_MODE_HW: 9289 return dev->netdev_ops->ndo_bpf; 9290 default: 9291 return NULL; 9292 } 9293 } 9294 9295 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev, 9296 enum bpf_xdp_mode mode) 9297 { 9298 return dev->xdp_state[mode].link; 9299 } 9300 9301 static struct bpf_prog *dev_xdp_prog(struct net_device *dev, 9302 enum bpf_xdp_mode mode) 9303 { 9304 struct bpf_xdp_link *link = dev_xdp_link(dev, mode); 9305 9306 if (link) 9307 return link->link.prog; 9308 return dev->xdp_state[mode].prog; 9309 } 9310 9311 u8 dev_xdp_prog_count(struct net_device *dev) 9312 { 9313 u8 count = 0; 9314 int i; 9315 9316 for (i = 0; i < __MAX_XDP_MODE; i++) 9317 if (dev->xdp_state[i].prog || dev->xdp_state[i].link) 9318 count++; 9319 return count; 9320 } 9321 EXPORT_SYMBOL_GPL(dev_xdp_prog_count); 9322 9323 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode) 9324 { 9325 struct bpf_prog *prog = dev_xdp_prog(dev, mode); 9326 9327 return prog ? prog->aux->id : 0; 9328 } 9329 9330 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode, 9331 struct bpf_xdp_link *link) 9332 { 9333 dev->xdp_state[mode].link = link; 9334 dev->xdp_state[mode].prog = NULL; 9335 } 9336 9337 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode, 9338 struct bpf_prog *prog) 9339 { 9340 dev->xdp_state[mode].link = NULL; 9341 dev->xdp_state[mode].prog = prog; 9342 } 9343 9344 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode, 9345 bpf_op_t bpf_op, struct netlink_ext_ack *extack, 9346 u32 flags, struct bpf_prog *prog) 9347 { 9348 struct netdev_bpf xdp; 9349 int err; 9350 9351 memset(&xdp, 0, sizeof(xdp)); 9352 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG; 9353 xdp.extack = extack; 9354 xdp.flags = flags; 9355 xdp.prog = prog; 9356 9357 /* Drivers assume refcnt is already incremented (i.e, prog pointer is 9358 * "moved" into driver), so they don't increment it on their own, but 9359 * they do decrement refcnt when program is detached or replaced. 9360 * Given net_device also owns link/prog, we need to bump refcnt here 9361 * to prevent drivers from underflowing it. 9362 */ 9363 if (prog) 9364 bpf_prog_inc(prog); 9365 err = bpf_op(dev, &xdp); 9366 if (err) { 9367 if (prog) 9368 bpf_prog_put(prog); 9369 return err; 9370 } 9371 9372 if (mode != XDP_MODE_HW) 9373 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog); 9374 9375 return 0; 9376 } 9377 9378 static void dev_xdp_uninstall(struct net_device *dev) 9379 { 9380 struct bpf_xdp_link *link; 9381 struct bpf_prog *prog; 9382 enum bpf_xdp_mode mode; 9383 bpf_op_t bpf_op; 9384 9385 ASSERT_RTNL(); 9386 9387 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) { 9388 prog = dev_xdp_prog(dev, mode); 9389 if (!prog) 9390 continue; 9391 9392 bpf_op = dev_xdp_bpf_op(dev, mode); 9393 if (!bpf_op) 9394 continue; 9395 9396 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9397 9398 /* auto-detach link from net device */ 9399 link = dev_xdp_link(dev, mode); 9400 if (link) 9401 link->dev = NULL; 9402 else 9403 bpf_prog_put(prog); 9404 9405 dev_xdp_set_link(dev, mode, NULL); 9406 } 9407 } 9408 9409 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack, 9410 struct bpf_xdp_link *link, struct bpf_prog *new_prog, 9411 struct bpf_prog *old_prog, u32 flags) 9412 { 9413 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES); 9414 struct bpf_prog *cur_prog; 9415 struct net_device *upper; 9416 struct list_head *iter; 9417 enum bpf_xdp_mode mode; 9418 bpf_op_t bpf_op; 9419 int err; 9420 9421 ASSERT_RTNL(); 9422 9423 /* either link or prog attachment, never both */ 9424 if (link && (new_prog || old_prog)) 9425 return -EINVAL; 9426 /* link supports only XDP mode flags */ 9427 if (link && (flags & ~XDP_FLAGS_MODES)) { 9428 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment"); 9429 return -EINVAL; 9430 } 9431 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */ 9432 if (num_modes > 1) { 9433 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set"); 9434 return -EINVAL; 9435 } 9436 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */ 9437 if (!num_modes && dev_xdp_prog_count(dev) > 1) { 9438 NL_SET_ERR_MSG(extack, 9439 "More than one program loaded, unset mode is ambiguous"); 9440 return -EINVAL; 9441 } 9442 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */ 9443 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) { 9444 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified"); 9445 return -EINVAL; 9446 } 9447 9448 mode = dev_xdp_mode(dev, flags); 9449 /* can't replace attached link */ 9450 if (dev_xdp_link(dev, mode)) { 9451 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link"); 9452 return -EBUSY; 9453 } 9454 9455 /* don't allow if an upper device already has a program */ 9456 netdev_for_each_upper_dev_rcu(dev, upper, iter) { 9457 if (dev_xdp_prog_count(upper) > 0) { 9458 NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program"); 9459 return -EEXIST; 9460 } 9461 } 9462 9463 cur_prog = dev_xdp_prog(dev, mode); 9464 /* can't replace attached prog with link */ 9465 if (link && cur_prog) { 9466 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link"); 9467 return -EBUSY; 9468 } 9469 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) { 9470 NL_SET_ERR_MSG(extack, "Active program does not match expected"); 9471 return -EEXIST; 9472 } 9473 9474 /* put effective new program into new_prog */ 9475 if (link) 9476 new_prog = link->link.prog; 9477 9478 if (new_prog) { 9479 bool offload = mode == XDP_MODE_HW; 9480 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB 9481 ? XDP_MODE_DRV : XDP_MODE_SKB; 9482 9483 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) { 9484 NL_SET_ERR_MSG(extack, "XDP program already attached"); 9485 return -EBUSY; 9486 } 9487 if (!offload && dev_xdp_prog(dev, other_mode)) { 9488 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time"); 9489 return -EEXIST; 9490 } 9491 if (!offload && bpf_prog_is_dev_bound(new_prog->aux)) { 9492 NL_SET_ERR_MSG(extack, "Using device-bound program without HW_MODE flag is not supported"); 9493 return -EINVAL; 9494 } 9495 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) { 9496 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device"); 9497 return -EINVAL; 9498 } 9499 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) { 9500 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device"); 9501 return -EINVAL; 9502 } 9503 } 9504 9505 /* don't call drivers if the effective program didn't change */ 9506 if (new_prog != cur_prog) { 9507 bpf_op = dev_xdp_bpf_op(dev, mode); 9508 if (!bpf_op) { 9509 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode"); 9510 return -EOPNOTSUPP; 9511 } 9512 9513 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog); 9514 if (err) 9515 return err; 9516 } 9517 9518 if (link) 9519 dev_xdp_set_link(dev, mode, link); 9520 else 9521 dev_xdp_set_prog(dev, mode, new_prog); 9522 if (cur_prog) 9523 bpf_prog_put(cur_prog); 9524 9525 return 0; 9526 } 9527 9528 static int dev_xdp_attach_link(struct net_device *dev, 9529 struct netlink_ext_ack *extack, 9530 struct bpf_xdp_link *link) 9531 { 9532 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags); 9533 } 9534 9535 static int dev_xdp_detach_link(struct net_device *dev, 9536 struct netlink_ext_ack *extack, 9537 struct bpf_xdp_link *link) 9538 { 9539 enum bpf_xdp_mode mode; 9540 bpf_op_t bpf_op; 9541 9542 ASSERT_RTNL(); 9543 9544 mode = dev_xdp_mode(dev, link->flags); 9545 if (dev_xdp_link(dev, mode) != link) 9546 return -EINVAL; 9547 9548 bpf_op = dev_xdp_bpf_op(dev, mode); 9549 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9550 dev_xdp_set_link(dev, mode, NULL); 9551 return 0; 9552 } 9553 9554 static void bpf_xdp_link_release(struct bpf_link *link) 9555 { 9556 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9557 9558 rtnl_lock(); 9559 9560 /* if racing with net_device's tear down, xdp_link->dev might be 9561 * already NULL, in which case link was already auto-detached 9562 */ 9563 if (xdp_link->dev) { 9564 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link)); 9565 xdp_link->dev = NULL; 9566 } 9567 9568 rtnl_unlock(); 9569 } 9570 9571 static int bpf_xdp_link_detach(struct bpf_link *link) 9572 { 9573 bpf_xdp_link_release(link); 9574 return 0; 9575 } 9576 9577 static void bpf_xdp_link_dealloc(struct bpf_link *link) 9578 { 9579 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9580 9581 kfree(xdp_link); 9582 } 9583 9584 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link, 9585 struct seq_file *seq) 9586 { 9587 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9588 u32 ifindex = 0; 9589 9590 rtnl_lock(); 9591 if (xdp_link->dev) 9592 ifindex = xdp_link->dev->ifindex; 9593 rtnl_unlock(); 9594 9595 seq_printf(seq, "ifindex:\t%u\n", ifindex); 9596 } 9597 9598 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link, 9599 struct bpf_link_info *info) 9600 { 9601 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9602 u32 ifindex = 0; 9603 9604 rtnl_lock(); 9605 if (xdp_link->dev) 9606 ifindex = xdp_link->dev->ifindex; 9607 rtnl_unlock(); 9608 9609 info->xdp.ifindex = ifindex; 9610 return 0; 9611 } 9612 9613 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog, 9614 struct bpf_prog *old_prog) 9615 { 9616 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9617 enum bpf_xdp_mode mode; 9618 bpf_op_t bpf_op; 9619 int err = 0; 9620 9621 rtnl_lock(); 9622 9623 /* link might have been auto-released already, so fail */ 9624 if (!xdp_link->dev) { 9625 err = -ENOLINK; 9626 goto out_unlock; 9627 } 9628 9629 if (old_prog && link->prog != old_prog) { 9630 err = -EPERM; 9631 goto out_unlock; 9632 } 9633 old_prog = link->prog; 9634 if (old_prog == new_prog) { 9635 /* no-op, don't disturb drivers */ 9636 bpf_prog_put(new_prog); 9637 goto out_unlock; 9638 } 9639 9640 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags); 9641 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode); 9642 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL, 9643 xdp_link->flags, new_prog); 9644 if (err) 9645 goto out_unlock; 9646 9647 old_prog = xchg(&link->prog, new_prog); 9648 bpf_prog_put(old_prog); 9649 9650 out_unlock: 9651 rtnl_unlock(); 9652 return err; 9653 } 9654 9655 static const struct bpf_link_ops bpf_xdp_link_lops = { 9656 .release = bpf_xdp_link_release, 9657 .dealloc = bpf_xdp_link_dealloc, 9658 .detach = bpf_xdp_link_detach, 9659 .show_fdinfo = bpf_xdp_link_show_fdinfo, 9660 .fill_link_info = bpf_xdp_link_fill_link_info, 9661 .update_prog = bpf_xdp_link_update, 9662 }; 9663 9664 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog) 9665 { 9666 struct net *net = current->nsproxy->net_ns; 9667 struct bpf_link_primer link_primer; 9668 struct bpf_xdp_link *link; 9669 struct net_device *dev; 9670 int err, fd; 9671 9672 rtnl_lock(); 9673 dev = dev_get_by_index(net, attr->link_create.target_ifindex); 9674 if (!dev) { 9675 rtnl_unlock(); 9676 return -EINVAL; 9677 } 9678 9679 link = kzalloc(sizeof(*link), GFP_USER); 9680 if (!link) { 9681 err = -ENOMEM; 9682 goto unlock; 9683 } 9684 9685 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog); 9686 link->dev = dev; 9687 link->flags = attr->link_create.flags; 9688 9689 err = bpf_link_prime(&link->link, &link_primer); 9690 if (err) { 9691 kfree(link); 9692 goto unlock; 9693 } 9694 9695 err = dev_xdp_attach_link(dev, NULL, link); 9696 rtnl_unlock(); 9697 9698 if (err) { 9699 link->dev = NULL; 9700 bpf_link_cleanup(&link_primer); 9701 goto out_put_dev; 9702 } 9703 9704 fd = bpf_link_settle(&link_primer); 9705 /* link itself doesn't hold dev's refcnt to not complicate shutdown */ 9706 dev_put(dev); 9707 return fd; 9708 9709 unlock: 9710 rtnl_unlock(); 9711 9712 out_put_dev: 9713 dev_put(dev); 9714 return err; 9715 } 9716 9717 /** 9718 * dev_change_xdp_fd - set or clear a bpf program for a device rx path 9719 * @dev: device 9720 * @extack: netlink extended ack 9721 * @fd: new program fd or negative value to clear 9722 * @expected_fd: old program fd that userspace expects to replace or clear 9723 * @flags: xdp-related flags 9724 * 9725 * Set or clear a bpf program for a device 9726 */ 9727 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, 9728 int fd, int expected_fd, u32 flags) 9729 { 9730 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags); 9731 struct bpf_prog *new_prog = NULL, *old_prog = NULL; 9732 int err; 9733 9734 ASSERT_RTNL(); 9735 9736 if (fd >= 0) { 9737 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP, 9738 mode != XDP_MODE_SKB); 9739 if (IS_ERR(new_prog)) 9740 return PTR_ERR(new_prog); 9741 } 9742 9743 if (expected_fd >= 0) { 9744 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP, 9745 mode != XDP_MODE_SKB); 9746 if (IS_ERR(old_prog)) { 9747 err = PTR_ERR(old_prog); 9748 old_prog = NULL; 9749 goto err_out; 9750 } 9751 } 9752 9753 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags); 9754 9755 err_out: 9756 if (err && new_prog) 9757 bpf_prog_put(new_prog); 9758 if (old_prog) 9759 bpf_prog_put(old_prog); 9760 return err; 9761 } 9762 9763 /** 9764 * dev_new_index - allocate an ifindex 9765 * @net: the applicable net namespace 9766 * 9767 * Returns a suitable unique value for a new device interface 9768 * number. The caller must hold the rtnl semaphore or the 9769 * dev_base_lock to be sure it remains unique. 9770 */ 9771 static int dev_new_index(struct net *net) 9772 { 9773 int ifindex = net->ifindex; 9774 9775 for (;;) { 9776 if (++ifindex <= 0) 9777 ifindex = 1; 9778 if (!__dev_get_by_index(net, ifindex)) 9779 return net->ifindex = ifindex; 9780 } 9781 } 9782 9783 /* Delayed registration/unregisteration */ 9784 static LIST_HEAD(net_todo_list); 9785 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); 9786 9787 static void net_set_todo(struct net_device *dev) 9788 { 9789 list_add_tail(&dev->todo_list, &net_todo_list); 9790 dev_net(dev)->dev_unreg_count++; 9791 } 9792 9793 static netdev_features_t netdev_sync_upper_features(struct net_device *lower, 9794 struct net_device *upper, netdev_features_t features) 9795 { 9796 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9797 netdev_features_t feature; 9798 int feature_bit; 9799 9800 for_each_netdev_feature(upper_disables, feature_bit) { 9801 feature = __NETIF_F_BIT(feature_bit); 9802 if (!(upper->wanted_features & feature) 9803 && (features & feature)) { 9804 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", 9805 &feature, upper->name); 9806 features &= ~feature; 9807 } 9808 } 9809 9810 return features; 9811 } 9812 9813 static void netdev_sync_lower_features(struct net_device *upper, 9814 struct net_device *lower, netdev_features_t features) 9815 { 9816 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9817 netdev_features_t feature; 9818 int feature_bit; 9819 9820 for_each_netdev_feature(upper_disables, feature_bit) { 9821 feature = __NETIF_F_BIT(feature_bit); 9822 if (!(features & feature) && (lower->features & feature)) { 9823 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", 9824 &feature, lower->name); 9825 lower->wanted_features &= ~feature; 9826 __netdev_update_features(lower); 9827 9828 if (unlikely(lower->features & feature)) 9829 netdev_WARN(upper, "failed to disable %pNF on %s!\n", 9830 &feature, lower->name); 9831 else 9832 netdev_features_change(lower); 9833 } 9834 } 9835 } 9836 9837 static netdev_features_t netdev_fix_features(struct net_device *dev, 9838 netdev_features_t features) 9839 { 9840 /* Fix illegal checksum combinations */ 9841 if ((features & NETIF_F_HW_CSUM) && 9842 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { 9843 netdev_warn(dev, "mixed HW and IP checksum settings.\n"); 9844 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); 9845 } 9846 9847 /* TSO requires that SG is present as well. */ 9848 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { 9849 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); 9850 features &= ~NETIF_F_ALL_TSO; 9851 } 9852 9853 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && 9854 !(features & NETIF_F_IP_CSUM)) { 9855 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); 9856 features &= ~NETIF_F_TSO; 9857 features &= ~NETIF_F_TSO_ECN; 9858 } 9859 9860 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && 9861 !(features & NETIF_F_IPV6_CSUM)) { 9862 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); 9863 features &= ~NETIF_F_TSO6; 9864 } 9865 9866 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ 9867 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) 9868 features &= ~NETIF_F_TSO_MANGLEID; 9869 9870 /* TSO ECN requires that TSO is present as well. */ 9871 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) 9872 features &= ~NETIF_F_TSO_ECN; 9873 9874 /* Software GSO depends on SG. */ 9875 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { 9876 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); 9877 features &= ~NETIF_F_GSO; 9878 } 9879 9880 /* GSO partial features require GSO partial be set */ 9881 if ((features & dev->gso_partial_features) && 9882 !(features & NETIF_F_GSO_PARTIAL)) { 9883 netdev_dbg(dev, 9884 "Dropping partially supported GSO features since no GSO partial.\n"); 9885 features &= ~dev->gso_partial_features; 9886 } 9887 9888 if (!(features & NETIF_F_RXCSUM)) { 9889 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet 9890 * successfully merged by hardware must also have the 9891 * checksum verified by hardware. If the user does not 9892 * want to enable RXCSUM, logically, we should disable GRO_HW. 9893 */ 9894 if (features & NETIF_F_GRO_HW) { 9895 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n"); 9896 features &= ~NETIF_F_GRO_HW; 9897 } 9898 } 9899 9900 /* LRO/HW-GRO features cannot be combined with RX-FCS */ 9901 if (features & NETIF_F_RXFCS) { 9902 if (features & NETIF_F_LRO) { 9903 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n"); 9904 features &= ~NETIF_F_LRO; 9905 } 9906 9907 if (features & NETIF_F_GRO_HW) { 9908 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n"); 9909 features &= ~NETIF_F_GRO_HW; 9910 } 9911 } 9912 9913 if (features & NETIF_F_HW_TLS_TX) { 9914 bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) == 9915 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM); 9916 bool hw_csum = features & NETIF_F_HW_CSUM; 9917 9918 if (!ip_csum && !hw_csum) { 9919 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n"); 9920 features &= ~NETIF_F_HW_TLS_TX; 9921 } 9922 } 9923 9924 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) { 9925 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n"); 9926 features &= ~NETIF_F_HW_TLS_RX; 9927 } 9928 9929 return features; 9930 } 9931 9932 int __netdev_update_features(struct net_device *dev) 9933 { 9934 struct net_device *upper, *lower; 9935 netdev_features_t features; 9936 struct list_head *iter; 9937 int err = -1; 9938 9939 ASSERT_RTNL(); 9940 9941 features = netdev_get_wanted_features(dev); 9942 9943 if (dev->netdev_ops->ndo_fix_features) 9944 features = dev->netdev_ops->ndo_fix_features(dev, features); 9945 9946 /* driver might be less strict about feature dependencies */ 9947 features = netdev_fix_features(dev, features); 9948 9949 /* some features can't be enabled if they're off on an upper device */ 9950 netdev_for_each_upper_dev_rcu(dev, upper, iter) 9951 features = netdev_sync_upper_features(dev, upper, features); 9952 9953 if (dev->features == features) 9954 goto sync_lower; 9955 9956 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", 9957 &dev->features, &features); 9958 9959 if (dev->netdev_ops->ndo_set_features) 9960 err = dev->netdev_ops->ndo_set_features(dev, features); 9961 else 9962 err = 0; 9963 9964 if (unlikely(err < 0)) { 9965 netdev_err(dev, 9966 "set_features() failed (%d); wanted %pNF, left %pNF\n", 9967 err, &features, &dev->features); 9968 /* return non-0 since some features might have changed and 9969 * it's better to fire a spurious notification than miss it 9970 */ 9971 return -1; 9972 } 9973 9974 sync_lower: 9975 /* some features must be disabled on lower devices when disabled 9976 * on an upper device (think: bonding master or bridge) 9977 */ 9978 netdev_for_each_lower_dev(dev, lower, iter) 9979 netdev_sync_lower_features(dev, lower, features); 9980 9981 if (!err) { 9982 netdev_features_t diff = features ^ dev->features; 9983 9984 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) { 9985 /* udp_tunnel_{get,drop}_rx_info both need 9986 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the 9987 * device, or they won't do anything. 9988 * Thus we need to update dev->features 9989 * *before* calling udp_tunnel_get_rx_info, 9990 * but *after* calling udp_tunnel_drop_rx_info. 9991 */ 9992 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) { 9993 dev->features = features; 9994 udp_tunnel_get_rx_info(dev); 9995 } else { 9996 udp_tunnel_drop_rx_info(dev); 9997 } 9998 } 9999 10000 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) { 10001 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) { 10002 dev->features = features; 10003 err |= vlan_get_rx_ctag_filter_info(dev); 10004 } else { 10005 vlan_drop_rx_ctag_filter_info(dev); 10006 } 10007 } 10008 10009 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) { 10010 if (features & NETIF_F_HW_VLAN_STAG_FILTER) { 10011 dev->features = features; 10012 err |= vlan_get_rx_stag_filter_info(dev); 10013 } else { 10014 vlan_drop_rx_stag_filter_info(dev); 10015 } 10016 } 10017 10018 dev->features = features; 10019 } 10020 10021 return err < 0 ? 0 : 1; 10022 } 10023 10024 /** 10025 * netdev_update_features - recalculate device features 10026 * @dev: the device to check 10027 * 10028 * Recalculate dev->features set and send notifications if it 10029 * has changed. Should be called after driver or hardware dependent 10030 * conditions might have changed that influence the features. 10031 */ 10032 void netdev_update_features(struct net_device *dev) 10033 { 10034 if (__netdev_update_features(dev)) 10035 netdev_features_change(dev); 10036 } 10037 EXPORT_SYMBOL(netdev_update_features); 10038 10039 /** 10040 * netdev_change_features - recalculate device features 10041 * @dev: the device to check 10042 * 10043 * Recalculate dev->features set and send notifications even 10044 * if they have not changed. Should be called instead of 10045 * netdev_update_features() if also dev->vlan_features might 10046 * have changed to allow the changes to be propagated to stacked 10047 * VLAN devices. 10048 */ 10049 void netdev_change_features(struct net_device *dev) 10050 { 10051 __netdev_update_features(dev); 10052 netdev_features_change(dev); 10053 } 10054 EXPORT_SYMBOL(netdev_change_features); 10055 10056 /** 10057 * netif_stacked_transfer_operstate - transfer operstate 10058 * @rootdev: the root or lower level device to transfer state from 10059 * @dev: the device to transfer operstate to 10060 * 10061 * Transfer operational state from root to device. This is normally 10062 * called when a stacking relationship exists between the root 10063 * device and the device(a leaf device). 10064 */ 10065 void netif_stacked_transfer_operstate(const struct net_device *rootdev, 10066 struct net_device *dev) 10067 { 10068 if (rootdev->operstate == IF_OPER_DORMANT) 10069 netif_dormant_on(dev); 10070 else 10071 netif_dormant_off(dev); 10072 10073 if (rootdev->operstate == IF_OPER_TESTING) 10074 netif_testing_on(dev); 10075 else 10076 netif_testing_off(dev); 10077 10078 if (netif_carrier_ok(rootdev)) 10079 netif_carrier_on(dev); 10080 else 10081 netif_carrier_off(dev); 10082 } 10083 EXPORT_SYMBOL(netif_stacked_transfer_operstate); 10084 10085 static int netif_alloc_rx_queues(struct net_device *dev) 10086 { 10087 unsigned int i, count = dev->num_rx_queues; 10088 struct netdev_rx_queue *rx; 10089 size_t sz = count * sizeof(*rx); 10090 int err = 0; 10091 10092 BUG_ON(count < 1); 10093 10094 rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10095 if (!rx) 10096 return -ENOMEM; 10097 10098 dev->_rx = rx; 10099 10100 for (i = 0; i < count; i++) { 10101 rx[i].dev = dev; 10102 10103 /* XDP RX-queue setup */ 10104 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0); 10105 if (err < 0) 10106 goto err_rxq_info; 10107 } 10108 return 0; 10109 10110 err_rxq_info: 10111 /* Rollback successful reg's and free other resources */ 10112 while (i--) 10113 xdp_rxq_info_unreg(&rx[i].xdp_rxq); 10114 kvfree(dev->_rx); 10115 dev->_rx = NULL; 10116 return err; 10117 } 10118 10119 static void netif_free_rx_queues(struct net_device *dev) 10120 { 10121 unsigned int i, count = dev->num_rx_queues; 10122 10123 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */ 10124 if (!dev->_rx) 10125 return; 10126 10127 for (i = 0; i < count; i++) 10128 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq); 10129 10130 kvfree(dev->_rx); 10131 } 10132 10133 static void netdev_init_one_queue(struct net_device *dev, 10134 struct netdev_queue *queue, void *_unused) 10135 { 10136 /* Initialize queue lock */ 10137 spin_lock_init(&queue->_xmit_lock); 10138 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); 10139 queue->xmit_lock_owner = -1; 10140 netdev_queue_numa_node_write(queue, NUMA_NO_NODE); 10141 queue->dev = dev; 10142 #ifdef CONFIG_BQL 10143 dql_init(&queue->dql, HZ); 10144 #endif 10145 } 10146 10147 static void netif_free_tx_queues(struct net_device *dev) 10148 { 10149 kvfree(dev->_tx); 10150 } 10151 10152 static int netif_alloc_netdev_queues(struct net_device *dev) 10153 { 10154 unsigned int count = dev->num_tx_queues; 10155 struct netdev_queue *tx; 10156 size_t sz = count * sizeof(*tx); 10157 10158 if (count < 1 || count > 0xffff) 10159 return -EINVAL; 10160 10161 tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10162 if (!tx) 10163 return -ENOMEM; 10164 10165 dev->_tx = tx; 10166 10167 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); 10168 spin_lock_init(&dev->tx_global_lock); 10169 10170 return 0; 10171 } 10172 10173 void netif_tx_stop_all_queues(struct net_device *dev) 10174 { 10175 unsigned int i; 10176 10177 for (i = 0; i < dev->num_tx_queues; i++) { 10178 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 10179 10180 netif_tx_stop_queue(txq); 10181 } 10182 } 10183 EXPORT_SYMBOL(netif_tx_stop_all_queues); 10184 10185 /** 10186 * register_netdevice - register a network device 10187 * @dev: device to register 10188 * 10189 * Take a completed network device structure and add it to the kernel 10190 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 10191 * chain. 0 is returned on success. A negative errno code is returned 10192 * on a failure to set up the device, or if the name is a duplicate. 10193 * 10194 * Callers must hold the rtnl semaphore. You may want 10195 * register_netdev() instead of this. 10196 * 10197 * BUGS: 10198 * The locking appears insufficient to guarantee two parallel registers 10199 * will not get the same name. 10200 */ 10201 10202 int register_netdevice(struct net_device *dev) 10203 { 10204 int ret; 10205 struct net *net = dev_net(dev); 10206 10207 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE < 10208 NETDEV_FEATURE_COUNT); 10209 BUG_ON(dev_boot_phase); 10210 ASSERT_RTNL(); 10211 10212 might_sleep(); 10213 10214 /* When net_device's are persistent, this will be fatal. */ 10215 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); 10216 BUG_ON(!net); 10217 10218 ret = ethtool_check_ops(dev->ethtool_ops); 10219 if (ret) 10220 return ret; 10221 10222 spin_lock_init(&dev->addr_list_lock); 10223 netdev_set_addr_lockdep_class(dev); 10224 10225 ret = dev_get_valid_name(net, dev, dev->name); 10226 if (ret < 0) 10227 goto out; 10228 10229 ret = -ENOMEM; 10230 dev->name_node = netdev_name_node_head_alloc(dev); 10231 if (!dev->name_node) 10232 goto out; 10233 10234 /* Init, if this function is available */ 10235 if (dev->netdev_ops->ndo_init) { 10236 ret = dev->netdev_ops->ndo_init(dev); 10237 if (ret) { 10238 if (ret > 0) 10239 ret = -EIO; 10240 goto err_free_name; 10241 } 10242 } 10243 10244 if (((dev->hw_features | dev->features) & 10245 NETIF_F_HW_VLAN_CTAG_FILTER) && 10246 (!dev->netdev_ops->ndo_vlan_rx_add_vid || 10247 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { 10248 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); 10249 ret = -EINVAL; 10250 goto err_uninit; 10251 } 10252 10253 ret = -EBUSY; 10254 if (!dev->ifindex) 10255 dev->ifindex = dev_new_index(net); 10256 else if (__dev_get_by_index(net, dev->ifindex)) 10257 goto err_uninit; 10258 10259 /* Transfer changeable features to wanted_features and enable 10260 * software offloads (GSO and GRO). 10261 */ 10262 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF); 10263 dev->features |= NETIF_F_SOFT_FEATURES; 10264 10265 if (dev->udp_tunnel_nic_info) { 10266 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10267 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10268 } 10269 10270 dev->wanted_features = dev->features & dev->hw_features; 10271 10272 if (!(dev->flags & IFF_LOOPBACK)) 10273 dev->hw_features |= NETIF_F_NOCACHE_COPY; 10274 10275 /* If IPv4 TCP segmentation offload is supported we should also 10276 * allow the device to enable segmenting the frame with the option 10277 * of ignoring a static IP ID value. This doesn't enable the 10278 * feature itself but allows the user to enable it later. 10279 */ 10280 if (dev->hw_features & NETIF_F_TSO) 10281 dev->hw_features |= NETIF_F_TSO_MANGLEID; 10282 if (dev->vlan_features & NETIF_F_TSO) 10283 dev->vlan_features |= NETIF_F_TSO_MANGLEID; 10284 if (dev->mpls_features & NETIF_F_TSO) 10285 dev->mpls_features |= NETIF_F_TSO_MANGLEID; 10286 if (dev->hw_enc_features & NETIF_F_TSO) 10287 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; 10288 10289 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. 10290 */ 10291 dev->vlan_features |= NETIF_F_HIGHDMA; 10292 10293 /* Make NETIF_F_SG inheritable to tunnel devices. 10294 */ 10295 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; 10296 10297 /* Make NETIF_F_SG inheritable to MPLS. 10298 */ 10299 dev->mpls_features |= NETIF_F_SG; 10300 10301 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); 10302 ret = notifier_to_errno(ret); 10303 if (ret) 10304 goto err_uninit; 10305 10306 ret = netdev_register_kobject(dev); 10307 if (ret) { 10308 dev->reg_state = NETREG_UNREGISTERED; 10309 goto err_uninit; 10310 } 10311 dev->reg_state = NETREG_REGISTERED; 10312 10313 __netdev_update_features(dev); 10314 10315 /* 10316 * Default initial state at registry is that the 10317 * device is present. 10318 */ 10319 10320 set_bit(__LINK_STATE_PRESENT, &dev->state); 10321 10322 linkwatch_init_dev(dev); 10323 10324 dev_init_scheduler(dev); 10325 dev_hold(dev); 10326 list_netdevice(dev); 10327 add_device_randomness(dev->dev_addr, dev->addr_len); 10328 10329 /* If the device has permanent device address, driver should 10330 * set dev_addr and also addr_assign_type should be set to 10331 * NET_ADDR_PERM (default value). 10332 */ 10333 if (dev->addr_assign_type == NET_ADDR_PERM) 10334 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); 10335 10336 /* Notify protocols, that a new device appeared. */ 10337 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); 10338 ret = notifier_to_errno(ret); 10339 if (ret) { 10340 /* Expect explicit free_netdev() on failure */ 10341 dev->needs_free_netdev = false; 10342 unregister_netdevice_queue(dev, NULL); 10343 goto out; 10344 } 10345 /* 10346 * Prevent userspace races by waiting until the network 10347 * device is fully setup before sending notifications. 10348 */ 10349 if (!dev->rtnl_link_ops || 10350 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 10351 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); 10352 10353 out: 10354 return ret; 10355 10356 err_uninit: 10357 if (dev->netdev_ops->ndo_uninit) 10358 dev->netdev_ops->ndo_uninit(dev); 10359 if (dev->priv_destructor) 10360 dev->priv_destructor(dev); 10361 err_free_name: 10362 netdev_name_node_free(dev->name_node); 10363 goto out; 10364 } 10365 EXPORT_SYMBOL(register_netdevice); 10366 10367 /** 10368 * init_dummy_netdev - init a dummy network device for NAPI 10369 * @dev: device to init 10370 * 10371 * This takes a network device structure and initialize the minimum 10372 * amount of fields so it can be used to schedule NAPI polls without 10373 * registering a full blown interface. This is to be used by drivers 10374 * that need to tie several hardware interfaces to a single NAPI 10375 * poll scheduler due to HW limitations. 10376 */ 10377 int init_dummy_netdev(struct net_device *dev) 10378 { 10379 /* Clear everything. Note we don't initialize spinlocks 10380 * are they aren't supposed to be taken by any of the 10381 * NAPI code and this dummy netdev is supposed to be 10382 * only ever used for NAPI polls 10383 */ 10384 memset(dev, 0, sizeof(struct net_device)); 10385 10386 /* make sure we BUG if trying to hit standard 10387 * register/unregister code path 10388 */ 10389 dev->reg_state = NETREG_DUMMY; 10390 10391 /* NAPI wants this */ 10392 INIT_LIST_HEAD(&dev->napi_list); 10393 10394 /* a dummy interface is started by default */ 10395 set_bit(__LINK_STATE_PRESENT, &dev->state); 10396 set_bit(__LINK_STATE_START, &dev->state); 10397 10398 /* napi_busy_loop stats accounting wants this */ 10399 dev_net_set(dev, &init_net); 10400 10401 /* Note : We dont allocate pcpu_refcnt for dummy devices, 10402 * because users of this 'device' dont need to change 10403 * its refcount. 10404 */ 10405 10406 return 0; 10407 } 10408 EXPORT_SYMBOL_GPL(init_dummy_netdev); 10409 10410 10411 /** 10412 * register_netdev - register a network device 10413 * @dev: device to register 10414 * 10415 * Take a completed network device structure and add it to the kernel 10416 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 10417 * chain. 0 is returned on success. A negative errno code is returned 10418 * on a failure to set up the device, or if the name is a duplicate. 10419 * 10420 * This is a wrapper around register_netdevice that takes the rtnl semaphore 10421 * and expands the device name if you passed a format string to 10422 * alloc_netdev. 10423 */ 10424 int register_netdev(struct net_device *dev) 10425 { 10426 int err; 10427 10428 if (rtnl_lock_killable()) 10429 return -EINTR; 10430 err = register_netdevice(dev); 10431 rtnl_unlock(); 10432 return err; 10433 } 10434 EXPORT_SYMBOL(register_netdev); 10435 10436 int netdev_refcnt_read(const struct net_device *dev) 10437 { 10438 #ifdef CONFIG_PCPU_DEV_REFCNT 10439 int i, refcnt = 0; 10440 10441 for_each_possible_cpu(i) 10442 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); 10443 return refcnt; 10444 #else 10445 return refcount_read(&dev->dev_refcnt); 10446 #endif 10447 } 10448 EXPORT_SYMBOL(netdev_refcnt_read); 10449 10450 int netdev_unregister_timeout_secs __read_mostly = 10; 10451 10452 #define WAIT_REFS_MIN_MSECS 1 10453 #define WAIT_REFS_MAX_MSECS 250 10454 /** 10455 * netdev_wait_allrefs - wait until all references are gone. 10456 * @dev: target net_device 10457 * 10458 * This is called when unregistering network devices. 10459 * 10460 * Any protocol or device that holds a reference should register 10461 * for netdevice notification, and cleanup and put back the 10462 * reference if they receive an UNREGISTER event. 10463 * We can get stuck here if buggy protocols don't correctly 10464 * call dev_put. 10465 */ 10466 static void netdev_wait_allrefs(struct net_device *dev) 10467 { 10468 unsigned long rebroadcast_time, warning_time; 10469 int wait = 0, refcnt; 10470 10471 linkwatch_forget_dev(dev); 10472 10473 rebroadcast_time = warning_time = jiffies; 10474 refcnt = netdev_refcnt_read(dev); 10475 10476 while (refcnt != 1) { 10477 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { 10478 rtnl_lock(); 10479 10480 /* Rebroadcast unregister notification */ 10481 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10482 10483 __rtnl_unlock(); 10484 rcu_barrier(); 10485 rtnl_lock(); 10486 10487 if (test_bit(__LINK_STATE_LINKWATCH_PENDING, 10488 &dev->state)) { 10489 /* We must not have linkwatch events 10490 * pending on unregister. If this 10491 * happens, we simply run the queue 10492 * unscheduled, resulting in a noop 10493 * for this device. 10494 */ 10495 linkwatch_run_queue(); 10496 } 10497 10498 __rtnl_unlock(); 10499 10500 rebroadcast_time = jiffies; 10501 } 10502 10503 if (!wait) { 10504 rcu_barrier(); 10505 wait = WAIT_REFS_MIN_MSECS; 10506 } else { 10507 msleep(wait); 10508 wait = min(wait << 1, WAIT_REFS_MAX_MSECS); 10509 } 10510 10511 refcnt = netdev_refcnt_read(dev); 10512 10513 if (refcnt != 1 && 10514 time_after(jiffies, warning_time + 10515 netdev_unregister_timeout_secs * HZ)) { 10516 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", 10517 dev->name, refcnt); 10518 warning_time = jiffies; 10519 } 10520 } 10521 } 10522 10523 /* The sequence is: 10524 * 10525 * rtnl_lock(); 10526 * ... 10527 * register_netdevice(x1); 10528 * register_netdevice(x2); 10529 * ... 10530 * unregister_netdevice(y1); 10531 * unregister_netdevice(y2); 10532 * ... 10533 * rtnl_unlock(); 10534 * free_netdev(y1); 10535 * free_netdev(y2); 10536 * 10537 * We are invoked by rtnl_unlock(). 10538 * This allows us to deal with problems: 10539 * 1) We can delete sysfs objects which invoke hotplug 10540 * without deadlocking with linkwatch via keventd. 10541 * 2) Since we run with the RTNL semaphore not held, we can sleep 10542 * safely in order to wait for the netdev refcnt to drop to zero. 10543 * 10544 * We must not return until all unregister events added during 10545 * the interval the lock was held have been completed. 10546 */ 10547 void netdev_run_todo(void) 10548 { 10549 struct list_head list; 10550 #ifdef CONFIG_LOCKDEP 10551 struct list_head unlink_list; 10552 10553 list_replace_init(&net_unlink_list, &unlink_list); 10554 10555 while (!list_empty(&unlink_list)) { 10556 struct net_device *dev = list_first_entry(&unlink_list, 10557 struct net_device, 10558 unlink_list); 10559 list_del_init(&dev->unlink_list); 10560 dev->nested_level = dev->lower_level - 1; 10561 } 10562 #endif 10563 10564 /* Snapshot list, allow later requests */ 10565 list_replace_init(&net_todo_list, &list); 10566 10567 __rtnl_unlock(); 10568 10569 10570 /* Wait for rcu callbacks to finish before next phase */ 10571 if (!list_empty(&list)) 10572 rcu_barrier(); 10573 10574 while (!list_empty(&list)) { 10575 struct net_device *dev 10576 = list_first_entry(&list, struct net_device, todo_list); 10577 list_del(&dev->todo_list); 10578 10579 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { 10580 pr_err("network todo '%s' but state %d\n", 10581 dev->name, dev->reg_state); 10582 dump_stack(); 10583 continue; 10584 } 10585 10586 dev->reg_state = NETREG_UNREGISTERED; 10587 10588 netdev_wait_allrefs(dev); 10589 10590 /* paranoia */ 10591 BUG_ON(netdev_refcnt_read(dev) != 1); 10592 BUG_ON(!list_empty(&dev->ptype_all)); 10593 BUG_ON(!list_empty(&dev->ptype_specific)); 10594 WARN_ON(rcu_access_pointer(dev->ip_ptr)); 10595 WARN_ON(rcu_access_pointer(dev->ip6_ptr)); 10596 #if IS_ENABLED(CONFIG_DECNET) 10597 WARN_ON(dev->dn_ptr); 10598 #endif 10599 if (dev->priv_destructor) 10600 dev->priv_destructor(dev); 10601 if (dev->needs_free_netdev) 10602 free_netdev(dev); 10603 10604 /* Report a network device has been unregistered */ 10605 rtnl_lock(); 10606 dev_net(dev)->dev_unreg_count--; 10607 __rtnl_unlock(); 10608 wake_up(&netdev_unregistering_wq); 10609 10610 /* Free network device */ 10611 kobject_put(&dev->dev.kobj); 10612 } 10613 } 10614 10615 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has 10616 * all the same fields in the same order as net_device_stats, with only 10617 * the type differing, but rtnl_link_stats64 may have additional fields 10618 * at the end for newer counters. 10619 */ 10620 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, 10621 const struct net_device_stats *netdev_stats) 10622 { 10623 #if BITS_PER_LONG == 64 10624 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats)); 10625 memcpy(stats64, netdev_stats, sizeof(*netdev_stats)); 10626 /* zero out counters that only exist in rtnl_link_stats64 */ 10627 memset((char *)stats64 + sizeof(*netdev_stats), 0, 10628 sizeof(*stats64) - sizeof(*netdev_stats)); 10629 #else 10630 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long); 10631 const unsigned long *src = (const unsigned long *)netdev_stats; 10632 u64 *dst = (u64 *)stats64; 10633 10634 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); 10635 for (i = 0; i < n; i++) 10636 dst[i] = src[i]; 10637 /* zero out counters that only exist in rtnl_link_stats64 */ 10638 memset((char *)stats64 + n * sizeof(u64), 0, 10639 sizeof(*stats64) - n * sizeof(u64)); 10640 #endif 10641 } 10642 EXPORT_SYMBOL(netdev_stats_to_stats64); 10643 10644 /** 10645 * dev_get_stats - get network device statistics 10646 * @dev: device to get statistics from 10647 * @storage: place to store stats 10648 * 10649 * Get network statistics from device. Return @storage. 10650 * The device driver may provide its own method by setting 10651 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; 10652 * otherwise the internal statistics structure is used. 10653 */ 10654 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, 10655 struct rtnl_link_stats64 *storage) 10656 { 10657 const struct net_device_ops *ops = dev->netdev_ops; 10658 10659 if (ops->ndo_get_stats64) { 10660 memset(storage, 0, sizeof(*storage)); 10661 ops->ndo_get_stats64(dev, storage); 10662 } else if (ops->ndo_get_stats) { 10663 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); 10664 } else { 10665 netdev_stats_to_stats64(storage, &dev->stats); 10666 } 10667 storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped); 10668 storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped); 10669 storage->rx_nohandler += (unsigned long)atomic_long_read(&dev->rx_nohandler); 10670 return storage; 10671 } 10672 EXPORT_SYMBOL(dev_get_stats); 10673 10674 /** 10675 * dev_fetch_sw_netstats - get per-cpu network device statistics 10676 * @s: place to store stats 10677 * @netstats: per-cpu network stats to read from 10678 * 10679 * Read per-cpu network statistics and populate the related fields in @s. 10680 */ 10681 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s, 10682 const struct pcpu_sw_netstats __percpu *netstats) 10683 { 10684 int cpu; 10685 10686 for_each_possible_cpu(cpu) { 10687 const struct pcpu_sw_netstats *stats; 10688 struct pcpu_sw_netstats tmp; 10689 unsigned int start; 10690 10691 stats = per_cpu_ptr(netstats, cpu); 10692 do { 10693 start = u64_stats_fetch_begin_irq(&stats->syncp); 10694 tmp.rx_packets = stats->rx_packets; 10695 tmp.rx_bytes = stats->rx_bytes; 10696 tmp.tx_packets = stats->tx_packets; 10697 tmp.tx_bytes = stats->tx_bytes; 10698 } while (u64_stats_fetch_retry_irq(&stats->syncp, start)); 10699 10700 s->rx_packets += tmp.rx_packets; 10701 s->rx_bytes += tmp.rx_bytes; 10702 s->tx_packets += tmp.tx_packets; 10703 s->tx_bytes += tmp.tx_bytes; 10704 } 10705 } 10706 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats); 10707 10708 /** 10709 * dev_get_tstats64 - ndo_get_stats64 implementation 10710 * @dev: device to get statistics from 10711 * @s: place to store stats 10712 * 10713 * Populate @s from dev->stats and dev->tstats. Can be used as 10714 * ndo_get_stats64() callback. 10715 */ 10716 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s) 10717 { 10718 netdev_stats_to_stats64(s, &dev->stats); 10719 dev_fetch_sw_netstats(s, dev->tstats); 10720 } 10721 EXPORT_SYMBOL_GPL(dev_get_tstats64); 10722 10723 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) 10724 { 10725 struct netdev_queue *queue = dev_ingress_queue(dev); 10726 10727 #ifdef CONFIG_NET_CLS_ACT 10728 if (queue) 10729 return queue; 10730 queue = kzalloc(sizeof(*queue), GFP_KERNEL); 10731 if (!queue) 10732 return NULL; 10733 netdev_init_one_queue(dev, queue, NULL); 10734 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); 10735 queue->qdisc_sleeping = &noop_qdisc; 10736 rcu_assign_pointer(dev->ingress_queue, queue); 10737 #endif 10738 return queue; 10739 } 10740 10741 static const struct ethtool_ops default_ethtool_ops; 10742 10743 void netdev_set_default_ethtool_ops(struct net_device *dev, 10744 const struct ethtool_ops *ops) 10745 { 10746 if (dev->ethtool_ops == &default_ethtool_ops) 10747 dev->ethtool_ops = ops; 10748 } 10749 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); 10750 10751 void netdev_freemem(struct net_device *dev) 10752 { 10753 char *addr = (char *)dev - dev->padded; 10754 10755 kvfree(addr); 10756 } 10757 10758 /** 10759 * alloc_netdev_mqs - allocate network device 10760 * @sizeof_priv: size of private data to allocate space for 10761 * @name: device name format string 10762 * @name_assign_type: origin of device name 10763 * @setup: callback to initialize device 10764 * @txqs: the number of TX subqueues to allocate 10765 * @rxqs: the number of RX subqueues to allocate 10766 * 10767 * Allocates a struct net_device with private data area for driver use 10768 * and performs basic initialization. Also allocates subqueue structs 10769 * for each queue on the device. 10770 */ 10771 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, 10772 unsigned char name_assign_type, 10773 void (*setup)(struct net_device *), 10774 unsigned int txqs, unsigned int rxqs) 10775 { 10776 struct net_device *dev; 10777 unsigned int alloc_size; 10778 struct net_device *p; 10779 10780 BUG_ON(strlen(name) >= sizeof(dev->name)); 10781 10782 if (txqs < 1) { 10783 pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); 10784 return NULL; 10785 } 10786 10787 if (rxqs < 1) { 10788 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); 10789 return NULL; 10790 } 10791 10792 alloc_size = sizeof(struct net_device); 10793 if (sizeof_priv) { 10794 /* ensure 32-byte alignment of private area */ 10795 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); 10796 alloc_size += sizeof_priv; 10797 } 10798 /* ensure 32-byte alignment of whole construct */ 10799 alloc_size += NETDEV_ALIGN - 1; 10800 10801 p = kvzalloc(alloc_size, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10802 if (!p) 10803 return NULL; 10804 10805 dev = PTR_ALIGN(p, NETDEV_ALIGN); 10806 dev->padded = (char *)dev - (char *)p; 10807 10808 #ifdef CONFIG_PCPU_DEV_REFCNT 10809 dev->pcpu_refcnt = alloc_percpu(int); 10810 if (!dev->pcpu_refcnt) 10811 goto free_dev; 10812 dev_hold(dev); 10813 #else 10814 refcount_set(&dev->dev_refcnt, 1); 10815 #endif 10816 10817 if (dev_addr_init(dev)) 10818 goto free_pcpu; 10819 10820 dev_mc_init(dev); 10821 dev_uc_init(dev); 10822 10823 dev_net_set(dev, &init_net); 10824 10825 dev->gso_max_size = GSO_MAX_SIZE; 10826 dev->gso_max_segs = GSO_MAX_SEGS; 10827 dev->upper_level = 1; 10828 dev->lower_level = 1; 10829 #ifdef CONFIG_LOCKDEP 10830 dev->nested_level = 0; 10831 INIT_LIST_HEAD(&dev->unlink_list); 10832 #endif 10833 10834 INIT_LIST_HEAD(&dev->napi_list); 10835 INIT_LIST_HEAD(&dev->unreg_list); 10836 INIT_LIST_HEAD(&dev->close_list); 10837 INIT_LIST_HEAD(&dev->link_watch_list); 10838 INIT_LIST_HEAD(&dev->adj_list.upper); 10839 INIT_LIST_HEAD(&dev->adj_list.lower); 10840 INIT_LIST_HEAD(&dev->ptype_all); 10841 INIT_LIST_HEAD(&dev->ptype_specific); 10842 INIT_LIST_HEAD(&dev->net_notifier_list); 10843 #ifdef CONFIG_NET_SCHED 10844 hash_init(dev->qdisc_hash); 10845 #endif 10846 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; 10847 setup(dev); 10848 10849 if (!dev->tx_queue_len) { 10850 dev->priv_flags |= IFF_NO_QUEUE; 10851 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; 10852 } 10853 10854 dev->num_tx_queues = txqs; 10855 dev->real_num_tx_queues = txqs; 10856 if (netif_alloc_netdev_queues(dev)) 10857 goto free_all; 10858 10859 dev->num_rx_queues = rxqs; 10860 dev->real_num_rx_queues = rxqs; 10861 if (netif_alloc_rx_queues(dev)) 10862 goto free_all; 10863 10864 strcpy(dev->name, name); 10865 dev->name_assign_type = name_assign_type; 10866 dev->group = INIT_NETDEV_GROUP; 10867 if (!dev->ethtool_ops) 10868 dev->ethtool_ops = &default_ethtool_ops; 10869 10870 nf_hook_ingress_init(dev); 10871 10872 return dev; 10873 10874 free_all: 10875 free_netdev(dev); 10876 return NULL; 10877 10878 free_pcpu: 10879 #ifdef CONFIG_PCPU_DEV_REFCNT 10880 free_percpu(dev->pcpu_refcnt); 10881 free_dev: 10882 #endif 10883 netdev_freemem(dev); 10884 return NULL; 10885 } 10886 EXPORT_SYMBOL(alloc_netdev_mqs); 10887 10888 /** 10889 * free_netdev - free network device 10890 * @dev: device 10891 * 10892 * This function does the last stage of destroying an allocated device 10893 * interface. The reference to the device object is released. If this 10894 * is the last reference then it will be freed.Must be called in process 10895 * context. 10896 */ 10897 void free_netdev(struct net_device *dev) 10898 { 10899 struct napi_struct *p, *n; 10900 10901 might_sleep(); 10902 10903 /* When called immediately after register_netdevice() failed the unwind 10904 * handling may still be dismantling the device. Handle that case by 10905 * deferring the free. 10906 */ 10907 if (dev->reg_state == NETREG_UNREGISTERING) { 10908 ASSERT_RTNL(); 10909 dev->needs_free_netdev = true; 10910 return; 10911 } 10912 10913 netif_free_tx_queues(dev); 10914 netif_free_rx_queues(dev); 10915 10916 kfree(rcu_dereference_protected(dev->ingress_queue, 1)); 10917 10918 /* Flush device addresses */ 10919 dev_addr_flush(dev); 10920 10921 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) 10922 netif_napi_del(p); 10923 10924 #ifdef CONFIG_PCPU_DEV_REFCNT 10925 free_percpu(dev->pcpu_refcnt); 10926 dev->pcpu_refcnt = NULL; 10927 #endif 10928 free_percpu(dev->xdp_bulkq); 10929 dev->xdp_bulkq = NULL; 10930 10931 /* Compatibility with error handling in drivers */ 10932 if (dev->reg_state == NETREG_UNINITIALIZED) { 10933 netdev_freemem(dev); 10934 return; 10935 } 10936 10937 BUG_ON(dev->reg_state != NETREG_UNREGISTERED); 10938 dev->reg_state = NETREG_RELEASED; 10939 10940 /* will free via device release */ 10941 put_device(&dev->dev); 10942 } 10943 EXPORT_SYMBOL(free_netdev); 10944 10945 /** 10946 * synchronize_net - Synchronize with packet receive processing 10947 * 10948 * Wait for packets currently being received to be done. 10949 * Does not block later packets from starting. 10950 */ 10951 void synchronize_net(void) 10952 { 10953 might_sleep(); 10954 if (rtnl_is_locked()) 10955 synchronize_rcu_expedited(); 10956 else 10957 synchronize_rcu(); 10958 } 10959 EXPORT_SYMBOL(synchronize_net); 10960 10961 /** 10962 * unregister_netdevice_queue - remove device from the kernel 10963 * @dev: device 10964 * @head: list 10965 * 10966 * This function shuts down a device interface and removes it 10967 * from the kernel tables. 10968 * If head not NULL, device is queued to be unregistered later. 10969 * 10970 * Callers must hold the rtnl semaphore. You may want 10971 * unregister_netdev() instead of this. 10972 */ 10973 10974 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) 10975 { 10976 ASSERT_RTNL(); 10977 10978 if (head) { 10979 list_move_tail(&dev->unreg_list, head); 10980 } else { 10981 LIST_HEAD(single); 10982 10983 list_add(&dev->unreg_list, &single); 10984 unregister_netdevice_many(&single); 10985 } 10986 } 10987 EXPORT_SYMBOL(unregister_netdevice_queue); 10988 10989 /** 10990 * unregister_netdevice_many - unregister many devices 10991 * @head: list of devices 10992 * 10993 * Note: As most callers use a stack allocated list_head, 10994 * we force a list_del() to make sure stack wont be corrupted later. 10995 */ 10996 void unregister_netdevice_many(struct list_head *head) 10997 { 10998 struct net_device *dev, *tmp; 10999 LIST_HEAD(close_head); 11000 11001 BUG_ON(dev_boot_phase); 11002 ASSERT_RTNL(); 11003 11004 if (list_empty(head)) 11005 return; 11006 11007 list_for_each_entry_safe(dev, tmp, head, unreg_list) { 11008 /* Some devices call without registering 11009 * for initialization unwind. Remove those 11010 * devices and proceed with the remaining. 11011 */ 11012 if (dev->reg_state == NETREG_UNINITIALIZED) { 11013 pr_debug("unregister_netdevice: device %s/%p never was registered\n", 11014 dev->name, dev); 11015 11016 WARN_ON(1); 11017 list_del(&dev->unreg_list); 11018 continue; 11019 } 11020 dev->dismantle = true; 11021 BUG_ON(dev->reg_state != NETREG_REGISTERED); 11022 } 11023 11024 /* If device is running, close it first. */ 11025 list_for_each_entry(dev, head, unreg_list) 11026 list_add_tail(&dev->close_list, &close_head); 11027 dev_close_many(&close_head, true); 11028 11029 list_for_each_entry(dev, head, unreg_list) { 11030 /* And unlink it from device chain. */ 11031 unlist_netdevice(dev); 11032 11033 dev->reg_state = NETREG_UNREGISTERING; 11034 } 11035 flush_all_backlogs(); 11036 11037 synchronize_net(); 11038 11039 list_for_each_entry(dev, head, unreg_list) { 11040 struct sk_buff *skb = NULL; 11041 11042 /* Shutdown queueing discipline. */ 11043 dev_shutdown(dev); 11044 11045 dev_xdp_uninstall(dev); 11046 11047 /* Notify protocols, that we are about to destroy 11048 * this device. They should clean all the things. 11049 */ 11050 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 11051 11052 if (!dev->rtnl_link_ops || 11053 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 11054 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0, 11055 GFP_KERNEL, NULL, 0); 11056 11057 /* 11058 * Flush the unicast and multicast chains 11059 */ 11060 dev_uc_flush(dev); 11061 dev_mc_flush(dev); 11062 11063 netdev_name_node_alt_flush(dev); 11064 netdev_name_node_free(dev->name_node); 11065 11066 if (dev->netdev_ops->ndo_uninit) 11067 dev->netdev_ops->ndo_uninit(dev); 11068 11069 if (skb) 11070 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL); 11071 11072 /* Notifier chain MUST detach us all upper devices. */ 11073 WARN_ON(netdev_has_any_upper_dev(dev)); 11074 WARN_ON(netdev_has_any_lower_dev(dev)); 11075 11076 /* Remove entries from kobject tree */ 11077 netdev_unregister_kobject(dev); 11078 #ifdef CONFIG_XPS 11079 /* Remove XPS queueing entries */ 11080 netif_reset_xps_queues_gt(dev, 0); 11081 #endif 11082 } 11083 11084 synchronize_net(); 11085 11086 list_for_each_entry(dev, head, unreg_list) { 11087 dev_put(dev); 11088 net_set_todo(dev); 11089 } 11090 11091 list_del(head); 11092 } 11093 EXPORT_SYMBOL(unregister_netdevice_many); 11094 11095 /** 11096 * unregister_netdev - remove device from the kernel 11097 * @dev: device 11098 * 11099 * This function shuts down a device interface and removes it 11100 * from the kernel tables. 11101 * 11102 * This is just a wrapper for unregister_netdevice that takes 11103 * the rtnl semaphore. In general you want to use this and not 11104 * unregister_netdevice. 11105 */ 11106 void unregister_netdev(struct net_device *dev) 11107 { 11108 rtnl_lock(); 11109 unregister_netdevice(dev); 11110 rtnl_unlock(); 11111 } 11112 EXPORT_SYMBOL(unregister_netdev); 11113 11114 /** 11115 * __dev_change_net_namespace - move device to different nethost namespace 11116 * @dev: device 11117 * @net: network namespace 11118 * @pat: If not NULL name pattern to try if the current device name 11119 * is already taken in the destination network namespace. 11120 * @new_ifindex: If not zero, specifies device index in the target 11121 * namespace. 11122 * 11123 * This function shuts down a device interface and moves it 11124 * to a new network namespace. On success 0 is returned, on 11125 * a failure a netagive errno code is returned. 11126 * 11127 * Callers must hold the rtnl semaphore. 11128 */ 11129 11130 int __dev_change_net_namespace(struct net_device *dev, struct net *net, 11131 const char *pat, int new_ifindex) 11132 { 11133 struct net *net_old = dev_net(dev); 11134 int err, new_nsid; 11135 11136 ASSERT_RTNL(); 11137 11138 /* Don't allow namespace local devices to be moved. */ 11139 err = -EINVAL; 11140 if (dev->features & NETIF_F_NETNS_LOCAL) 11141 goto out; 11142 11143 /* Ensure the device has been registrered */ 11144 if (dev->reg_state != NETREG_REGISTERED) 11145 goto out; 11146 11147 /* Get out if there is nothing todo */ 11148 err = 0; 11149 if (net_eq(net_old, net)) 11150 goto out; 11151 11152 /* Pick the destination device name, and ensure 11153 * we can use it in the destination network namespace. 11154 */ 11155 err = -EEXIST; 11156 if (__dev_get_by_name(net, dev->name)) { 11157 /* We get here if we can't use the current device name */ 11158 if (!pat) 11159 goto out; 11160 err = dev_get_valid_name(net, dev, pat); 11161 if (err < 0) 11162 goto out; 11163 } 11164 11165 /* Check that new_ifindex isn't used yet. */ 11166 err = -EBUSY; 11167 if (new_ifindex && __dev_get_by_index(net, new_ifindex)) 11168 goto out; 11169 11170 /* 11171 * And now a mini version of register_netdevice unregister_netdevice. 11172 */ 11173 11174 /* If device is running close it first. */ 11175 dev_close(dev); 11176 11177 /* And unlink it from device chain */ 11178 unlist_netdevice(dev); 11179 11180 synchronize_net(); 11181 11182 /* Shutdown queueing discipline. */ 11183 dev_shutdown(dev); 11184 11185 /* Notify protocols, that we are about to destroy 11186 * this device. They should clean all the things. 11187 * 11188 * Note that dev->reg_state stays at NETREG_REGISTERED. 11189 * This is wanted because this way 8021q and macvlan know 11190 * the device is just moving and can keep their slaves up. 11191 */ 11192 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 11193 rcu_barrier(); 11194 11195 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL); 11196 /* If there is an ifindex conflict assign a new one */ 11197 if (!new_ifindex) { 11198 if (__dev_get_by_index(net, dev->ifindex)) 11199 new_ifindex = dev_new_index(net); 11200 else 11201 new_ifindex = dev->ifindex; 11202 } 11203 11204 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid, 11205 new_ifindex); 11206 11207 /* 11208 * Flush the unicast and multicast chains 11209 */ 11210 dev_uc_flush(dev); 11211 dev_mc_flush(dev); 11212 11213 /* Send a netdev-removed uevent to the old namespace */ 11214 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); 11215 netdev_adjacent_del_links(dev); 11216 11217 /* Move per-net netdevice notifiers that are following the netdevice */ 11218 move_netdevice_notifiers_dev_net(dev, net); 11219 11220 /* Actually switch the network namespace */ 11221 dev_net_set(dev, net); 11222 dev->ifindex = new_ifindex; 11223 11224 /* Send a netdev-add uevent to the new namespace */ 11225 kobject_uevent(&dev->dev.kobj, KOBJ_ADD); 11226 netdev_adjacent_add_links(dev); 11227 11228 /* Fixup kobjects */ 11229 err = device_rename(&dev->dev, dev->name); 11230 WARN_ON(err); 11231 11232 /* Adapt owner in case owning user namespace of target network 11233 * namespace is different from the original one. 11234 */ 11235 err = netdev_change_owner(dev, net_old, net); 11236 WARN_ON(err); 11237 11238 /* Add the device back in the hashes */ 11239 list_netdevice(dev); 11240 11241 /* Notify protocols, that a new device appeared. */ 11242 call_netdevice_notifiers(NETDEV_REGISTER, dev); 11243 11244 /* 11245 * Prevent userspace races by waiting until the network 11246 * device is fully setup before sending notifications. 11247 */ 11248 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); 11249 11250 synchronize_net(); 11251 err = 0; 11252 out: 11253 return err; 11254 } 11255 EXPORT_SYMBOL_GPL(__dev_change_net_namespace); 11256 11257 static int dev_cpu_dead(unsigned int oldcpu) 11258 { 11259 struct sk_buff **list_skb; 11260 struct sk_buff *skb; 11261 unsigned int cpu; 11262 struct softnet_data *sd, *oldsd, *remsd = NULL; 11263 11264 local_irq_disable(); 11265 cpu = smp_processor_id(); 11266 sd = &per_cpu(softnet_data, cpu); 11267 oldsd = &per_cpu(softnet_data, oldcpu); 11268 11269 /* Find end of our completion_queue. */ 11270 list_skb = &sd->completion_queue; 11271 while (*list_skb) 11272 list_skb = &(*list_skb)->next; 11273 /* Append completion queue from offline CPU. */ 11274 *list_skb = oldsd->completion_queue; 11275 oldsd->completion_queue = NULL; 11276 11277 /* Append output queue from offline CPU. */ 11278 if (oldsd->output_queue) { 11279 *sd->output_queue_tailp = oldsd->output_queue; 11280 sd->output_queue_tailp = oldsd->output_queue_tailp; 11281 oldsd->output_queue = NULL; 11282 oldsd->output_queue_tailp = &oldsd->output_queue; 11283 } 11284 /* Append NAPI poll list from offline CPU, with one exception : 11285 * process_backlog() must be called by cpu owning percpu backlog. 11286 * We properly handle process_queue & input_pkt_queue later. 11287 */ 11288 while (!list_empty(&oldsd->poll_list)) { 11289 struct napi_struct *napi = list_first_entry(&oldsd->poll_list, 11290 struct napi_struct, 11291 poll_list); 11292 11293 list_del_init(&napi->poll_list); 11294 if (napi->poll == process_backlog) 11295 napi->state = 0; 11296 else 11297 ____napi_schedule(sd, napi); 11298 } 11299 11300 raise_softirq_irqoff(NET_TX_SOFTIRQ); 11301 local_irq_enable(); 11302 11303 #ifdef CONFIG_RPS 11304 remsd = oldsd->rps_ipi_list; 11305 oldsd->rps_ipi_list = NULL; 11306 #endif 11307 /* send out pending IPI's on offline CPU */ 11308 net_rps_send_ipi(remsd); 11309 11310 /* Process offline CPU's input_pkt_queue */ 11311 while ((skb = __skb_dequeue(&oldsd->process_queue))) { 11312 netif_rx_ni(skb); 11313 input_queue_head_incr(oldsd); 11314 } 11315 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { 11316 netif_rx_ni(skb); 11317 input_queue_head_incr(oldsd); 11318 } 11319 11320 return 0; 11321 } 11322 11323 /** 11324 * netdev_increment_features - increment feature set by one 11325 * @all: current feature set 11326 * @one: new feature set 11327 * @mask: mask feature set 11328 * 11329 * Computes a new feature set after adding a device with feature set 11330 * @one to the master device with current feature set @all. Will not 11331 * enable anything that is off in @mask. Returns the new feature set. 11332 */ 11333 netdev_features_t netdev_increment_features(netdev_features_t all, 11334 netdev_features_t one, netdev_features_t mask) 11335 { 11336 if (mask & NETIF_F_HW_CSUM) 11337 mask |= NETIF_F_CSUM_MASK; 11338 mask |= NETIF_F_VLAN_CHALLENGED; 11339 11340 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; 11341 all &= one | ~NETIF_F_ALL_FOR_ALL; 11342 11343 /* If one device supports hw checksumming, set for all. */ 11344 if (all & NETIF_F_HW_CSUM) 11345 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); 11346 11347 return all; 11348 } 11349 EXPORT_SYMBOL(netdev_increment_features); 11350 11351 static struct hlist_head * __net_init netdev_create_hash(void) 11352 { 11353 int i; 11354 struct hlist_head *hash; 11355 11356 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL); 11357 if (hash != NULL) 11358 for (i = 0; i < NETDEV_HASHENTRIES; i++) 11359 INIT_HLIST_HEAD(&hash[i]); 11360 11361 return hash; 11362 } 11363 11364 /* Initialize per network namespace state */ 11365 static int __net_init netdev_init(struct net *net) 11366 { 11367 BUILD_BUG_ON(GRO_HASH_BUCKETS > 11368 8 * sizeof_field(struct napi_struct, gro_bitmask)); 11369 11370 if (net != &init_net) 11371 INIT_LIST_HEAD(&net->dev_base_head); 11372 11373 net->dev_name_head = netdev_create_hash(); 11374 if (net->dev_name_head == NULL) 11375 goto err_name; 11376 11377 net->dev_index_head = netdev_create_hash(); 11378 if (net->dev_index_head == NULL) 11379 goto err_idx; 11380 11381 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain); 11382 11383 return 0; 11384 11385 err_idx: 11386 kfree(net->dev_name_head); 11387 err_name: 11388 return -ENOMEM; 11389 } 11390 11391 /** 11392 * netdev_drivername - network driver for the device 11393 * @dev: network device 11394 * 11395 * Determine network driver for device. 11396 */ 11397 const char *netdev_drivername(const struct net_device *dev) 11398 { 11399 const struct device_driver *driver; 11400 const struct device *parent; 11401 const char *empty = ""; 11402 11403 parent = dev->dev.parent; 11404 if (!parent) 11405 return empty; 11406 11407 driver = parent->driver; 11408 if (driver && driver->name) 11409 return driver->name; 11410 return empty; 11411 } 11412 11413 static void __netdev_printk(const char *level, const struct net_device *dev, 11414 struct va_format *vaf) 11415 { 11416 if (dev && dev->dev.parent) { 11417 dev_printk_emit(level[1] - '0', 11418 dev->dev.parent, 11419 "%s %s %s%s: %pV", 11420 dev_driver_string(dev->dev.parent), 11421 dev_name(dev->dev.parent), 11422 netdev_name(dev), netdev_reg_state(dev), 11423 vaf); 11424 } else if (dev) { 11425 printk("%s%s%s: %pV", 11426 level, netdev_name(dev), netdev_reg_state(dev), vaf); 11427 } else { 11428 printk("%s(NULL net_device): %pV", level, vaf); 11429 } 11430 } 11431 11432 void netdev_printk(const char *level, const struct net_device *dev, 11433 const char *format, ...) 11434 { 11435 struct va_format vaf; 11436 va_list args; 11437 11438 va_start(args, format); 11439 11440 vaf.fmt = format; 11441 vaf.va = &args; 11442 11443 __netdev_printk(level, dev, &vaf); 11444 11445 va_end(args); 11446 } 11447 EXPORT_SYMBOL(netdev_printk); 11448 11449 #define define_netdev_printk_level(func, level) \ 11450 void func(const struct net_device *dev, const char *fmt, ...) \ 11451 { \ 11452 struct va_format vaf; \ 11453 va_list args; \ 11454 \ 11455 va_start(args, fmt); \ 11456 \ 11457 vaf.fmt = fmt; \ 11458 vaf.va = &args; \ 11459 \ 11460 __netdev_printk(level, dev, &vaf); \ 11461 \ 11462 va_end(args); \ 11463 } \ 11464 EXPORT_SYMBOL(func); 11465 11466 define_netdev_printk_level(netdev_emerg, KERN_EMERG); 11467 define_netdev_printk_level(netdev_alert, KERN_ALERT); 11468 define_netdev_printk_level(netdev_crit, KERN_CRIT); 11469 define_netdev_printk_level(netdev_err, KERN_ERR); 11470 define_netdev_printk_level(netdev_warn, KERN_WARNING); 11471 define_netdev_printk_level(netdev_notice, KERN_NOTICE); 11472 define_netdev_printk_level(netdev_info, KERN_INFO); 11473 11474 static void __net_exit netdev_exit(struct net *net) 11475 { 11476 kfree(net->dev_name_head); 11477 kfree(net->dev_index_head); 11478 if (net != &init_net) 11479 WARN_ON_ONCE(!list_empty(&net->dev_base_head)); 11480 } 11481 11482 static struct pernet_operations __net_initdata netdev_net_ops = { 11483 .init = netdev_init, 11484 .exit = netdev_exit, 11485 }; 11486 11487 static void __net_exit default_device_exit(struct net *net) 11488 { 11489 struct net_device *dev, *aux; 11490 /* 11491 * Push all migratable network devices back to the 11492 * initial network namespace 11493 */ 11494 rtnl_lock(); 11495 for_each_netdev_safe(net, dev, aux) { 11496 int err; 11497 char fb_name[IFNAMSIZ]; 11498 11499 /* Ignore unmoveable devices (i.e. loopback) */ 11500 if (dev->features & NETIF_F_NETNS_LOCAL) 11501 continue; 11502 11503 /* Leave virtual devices for the generic cleanup */ 11504 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund) 11505 continue; 11506 11507 /* Push remaining network devices to init_net */ 11508 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); 11509 if (__dev_get_by_name(&init_net, fb_name)) 11510 snprintf(fb_name, IFNAMSIZ, "dev%%d"); 11511 err = dev_change_net_namespace(dev, &init_net, fb_name); 11512 if (err) { 11513 pr_emerg("%s: failed to move %s to init_net: %d\n", 11514 __func__, dev->name, err); 11515 BUG(); 11516 } 11517 } 11518 rtnl_unlock(); 11519 } 11520 11521 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list) 11522 { 11523 /* Return with the rtnl_lock held when there are no network 11524 * devices unregistering in any network namespace in net_list. 11525 */ 11526 struct net *net; 11527 bool unregistering; 11528 DEFINE_WAIT_FUNC(wait, woken_wake_function); 11529 11530 add_wait_queue(&netdev_unregistering_wq, &wait); 11531 for (;;) { 11532 unregistering = false; 11533 rtnl_lock(); 11534 list_for_each_entry(net, net_list, exit_list) { 11535 if (net->dev_unreg_count > 0) { 11536 unregistering = true; 11537 break; 11538 } 11539 } 11540 if (!unregistering) 11541 break; 11542 __rtnl_unlock(); 11543 11544 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); 11545 } 11546 remove_wait_queue(&netdev_unregistering_wq, &wait); 11547 } 11548 11549 static void __net_exit default_device_exit_batch(struct list_head *net_list) 11550 { 11551 /* At exit all network devices most be removed from a network 11552 * namespace. Do this in the reverse order of registration. 11553 * Do this across as many network namespaces as possible to 11554 * improve batching efficiency. 11555 */ 11556 struct net_device *dev; 11557 struct net *net; 11558 LIST_HEAD(dev_kill_list); 11559 11560 /* To prevent network device cleanup code from dereferencing 11561 * loopback devices or network devices that have been freed 11562 * wait here for all pending unregistrations to complete, 11563 * before unregistring the loopback device and allowing the 11564 * network namespace be freed. 11565 * 11566 * The netdev todo list containing all network devices 11567 * unregistrations that happen in default_device_exit_batch 11568 * will run in the rtnl_unlock() at the end of 11569 * default_device_exit_batch. 11570 */ 11571 rtnl_lock_unregistering(net_list); 11572 list_for_each_entry(net, net_list, exit_list) { 11573 for_each_netdev_reverse(net, dev) { 11574 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) 11575 dev->rtnl_link_ops->dellink(dev, &dev_kill_list); 11576 else 11577 unregister_netdevice_queue(dev, &dev_kill_list); 11578 } 11579 } 11580 unregister_netdevice_many(&dev_kill_list); 11581 rtnl_unlock(); 11582 } 11583 11584 static struct pernet_operations __net_initdata default_device_ops = { 11585 .exit = default_device_exit, 11586 .exit_batch = default_device_exit_batch, 11587 }; 11588 11589 /* 11590 * Initialize the DEV module. At boot time this walks the device list and 11591 * unhooks any devices that fail to initialise (normally hardware not 11592 * present) and leaves us with a valid list of present and active devices. 11593 * 11594 */ 11595 11596 /* 11597 * This is called single threaded during boot, so no need 11598 * to take the rtnl semaphore. 11599 */ 11600 static int __init net_dev_init(void) 11601 { 11602 int i, rc = -ENOMEM; 11603 11604 BUG_ON(!dev_boot_phase); 11605 11606 if (dev_proc_init()) 11607 goto out; 11608 11609 if (netdev_kobject_init()) 11610 goto out; 11611 11612 INIT_LIST_HEAD(&ptype_all); 11613 for (i = 0; i < PTYPE_HASH_SIZE; i++) 11614 INIT_LIST_HEAD(&ptype_base[i]); 11615 11616 INIT_LIST_HEAD(&offload_base); 11617 11618 if (register_pernet_subsys(&netdev_net_ops)) 11619 goto out; 11620 11621 /* 11622 * Initialise the packet receive queues. 11623 */ 11624 11625 for_each_possible_cpu(i) { 11626 struct work_struct *flush = per_cpu_ptr(&flush_works, i); 11627 struct softnet_data *sd = &per_cpu(softnet_data, i); 11628 11629 INIT_WORK(flush, flush_backlog); 11630 11631 skb_queue_head_init(&sd->input_pkt_queue); 11632 skb_queue_head_init(&sd->process_queue); 11633 #ifdef CONFIG_XFRM_OFFLOAD 11634 skb_queue_head_init(&sd->xfrm_backlog); 11635 #endif 11636 INIT_LIST_HEAD(&sd->poll_list); 11637 sd->output_queue_tailp = &sd->output_queue; 11638 #ifdef CONFIG_RPS 11639 INIT_CSD(&sd->csd, rps_trigger_softirq, sd); 11640 sd->cpu = i; 11641 #endif 11642 11643 init_gro_hash(&sd->backlog); 11644 sd->backlog.poll = process_backlog; 11645 sd->backlog.weight = weight_p; 11646 } 11647 11648 dev_boot_phase = 0; 11649 11650 /* The loopback device is special if any other network devices 11651 * is present in a network namespace the loopback device must 11652 * be present. Since we now dynamically allocate and free the 11653 * loopback device ensure this invariant is maintained by 11654 * keeping the loopback device as the first device on the 11655 * list of network devices. Ensuring the loopback devices 11656 * is the first device that appears and the last network device 11657 * that disappears. 11658 */ 11659 if (register_pernet_device(&loopback_net_ops)) 11660 goto out; 11661 11662 if (register_pernet_device(&default_device_ops)) 11663 goto out; 11664 11665 open_softirq(NET_TX_SOFTIRQ, net_tx_action); 11666 open_softirq(NET_RX_SOFTIRQ, net_rx_action); 11667 11668 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", 11669 NULL, dev_cpu_dead); 11670 WARN_ON(rc < 0); 11671 rc = 0; 11672 out: 11673 return rc; 11674 } 11675 11676 subsys_initcall(net_dev_init); 11677