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, <[email protected]> 8 * Mark Evans, <[email protected]> 9 * 10 * Additional Authors: 11 * Florian la Roche <[email protected]> 12 * Alan Cox <[email protected]> 13 * David Hinds <[email protected]> 14 * Alexey Kuznetsov <[email protected]> 15 * Adam Sulmicki <[email protected]> 16 * Pekka Riikonen <[email protected]> 17 * 18 * Changes: 19 * D.J. Barrow : Fixed bug where dev->refcnt gets set 20 * to 2 if register_netdev gets called 21 * before net_dev_init & also removed a 22 * few lines of code in the process. 23 * Alan Cox : device private ioctl copies fields back. 24 * Alan Cox : Transmit queue code does relevant 25 * stunts to keep the queue safe. 26 * Alan Cox : Fixed double lock. 27 * Alan Cox : Fixed promisc NULL pointer trap 28 * ???????? : Support the full private ioctl range 29 * Alan Cox : Moved ioctl permission check into 30 * drivers 31 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI 32 * Alan Cox : 100 backlog just doesn't cut it when 33 * you start doing multicast video 8) 34 * Alan Cox : Rewrote net_bh and list manager. 35 * Alan Cox : Fix ETH_P_ALL echoback lengths. 36 * Alan Cox : Took out transmit every packet pass 37 * Saved a few bytes in the ioctl handler 38 * Alan Cox : Network driver sets packet type before 39 * calling netif_rx. Saves a function 40 * call a packet. 41 * Alan Cox : Hashed net_bh() 42 * Richard Kooijman: Timestamp fixes. 43 * Alan Cox : Wrong field in SIOCGIFDSTADDR 44 * Alan Cox : Device lock protection. 45 * Alan Cox : Fixed nasty side effect of device close 46 * changes. 47 * Rudi Cilibrasi : Pass the right thing to 48 * set_mac_address() 49 * Dave Miller : 32bit quantity for the device lock to 50 * make it work out on a Sparc. 51 * Bjorn Ekwall : Added KERNELD hack. 52 * Alan Cox : Cleaned up the backlog initialise. 53 * Craig Metz : SIOCGIFCONF fix if space for under 54 * 1 device. 55 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there 56 * is no device open function. 57 * Andi Kleen : Fix error reporting for SIOCGIFCONF 58 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF 59 * Cyrus Durgin : Cleaned for KMOD 60 * Adam Sulmicki : Bug Fix : Network Device Unload 61 * A network device unload needs to purge 62 * the backlog queue. 63 * Paul Rusty Russell : SIOCSIFNAME 64 * Pekka Riikonen : Netdev boot-time settings code 65 * Andrew Morton : Make unregister_netdevice wait 66 * indefinitely on dev->refcnt 67 * J Hadi Salim : - Backlog queue sampling 68 * - netif_rx() feedback 69 */ 70 71 #include <linux/uaccess.h> 72 #include <linux/bitops.h> 73 #include <linux/capability.h> 74 #include <linux/cpu.h> 75 #include <linux/types.h> 76 #include <linux/kernel.h> 77 #include <linux/hash.h> 78 #include <linux/slab.h> 79 #include <linux/sched.h> 80 #include <linux/sched/mm.h> 81 #include <linux/mutex.h> 82 #include <linux/rwsem.h> 83 #include <linux/string.h> 84 #include <linux/mm.h> 85 #include <linux/socket.h> 86 #include <linux/sockios.h> 87 #include <linux/errno.h> 88 #include <linux/interrupt.h> 89 #include <linux/if_ether.h> 90 #include <linux/netdevice.h> 91 #include <linux/etherdevice.h> 92 #include <linux/ethtool.h> 93 #include <linux/skbuff.h> 94 #include <linux/kthread.h> 95 #include <linux/bpf.h> 96 #include <linux/bpf_trace.h> 97 #include <net/net_namespace.h> 98 #include <net/sock.h> 99 #include <net/busy_poll.h> 100 #include <linux/rtnetlink.h> 101 #include <linux/stat.h> 102 #include <net/dsa.h> 103 #include <net/dst.h> 104 #include <net/dst_metadata.h> 105 #include <net/gro.h> 106 #include <net/pkt_sched.h> 107 #include <net/pkt_cls.h> 108 #include <net/checksum.h> 109 #include <net/xfrm.h> 110 #include <linux/highmem.h> 111 #include <linux/init.h> 112 #include <linux/module.h> 113 #include <linux/netpoll.h> 114 #include <linux/rcupdate.h> 115 #include <linux/delay.h> 116 #include <net/iw_handler.h> 117 #include <asm/current.h> 118 #include <linux/audit.h> 119 #include <linux/dmaengine.h> 120 #include <linux/err.h> 121 #include <linux/ctype.h> 122 #include <linux/if_arp.h> 123 #include <linux/if_vlan.h> 124 #include <linux/ip.h> 125 #include <net/ip.h> 126 #include <net/mpls.h> 127 #include <linux/ipv6.h> 128 #include <linux/in.h> 129 #include <linux/jhash.h> 130 #include <linux/random.h> 131 #include <trace/events/napi.h> 132 #include <trace/events/net.h> 133 #include <trace/events/skb.h> 134 #include <trace/events/qdisc.h> 135 #include <linux/inetdevice.h> 136 #include <linux/cpu_rmap.h> 137 #include <linux/static_key.h> 138 #include <linux/hashtable.h> 139 #include <linux/vmalloc.h> 140 #include <linux/if_macvlan.h> 141 #include <linux/errqueue.h> 142 #include <linux/hrtimer.h> 143 #include <linux/netfilter_netdev.h> 144 #include <linux/crash_dump.h> 145 #include <linux/sctp.h> 146 #include <net/udp_tunnel.h> 147 #include <linux/net_namespace.h> 148 #include <linux/indirect_call_wrapper.h> 149 #include <net/devlink.h> 150 #include <linux/pm_runtime.h> 151 #include <linux/prandom.h> 152 #include <linux/once_lite.h> 153 154 #include "net-sysfs.h" 155 156 #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 bool netdev_name_in_use(struct net *net, const char *name) 307 { 308 return netdev_name_node_lookup(net, name); 309 } 310 EXPORT_SYMBOL(netdev_name_in_use); 311 312 int netdev_name_node_alt_create(struct net_device *dev, const char *name) 313 { 314 struct netdev_name_node *name_node; 315 struct net *net = dev_net(dev); 316 317 name_node = netdev_name_node_lookup(net, name); 318 if (name_node) 319 return -EEXIST; 320 name_node = netdev_name_node_alloc(dev, name); 321 if (!name_node) 322 return -ENOMEM; 323 netdev_name_node_add(net, name_node); 324 /* The node that holds dev->name acts as a head of per-device list. */ 325 list_add_tail(&name_node->list, &dev->name_node->list); 326 327 return 0; 328 } 329 EXPORT_SYMBOL(netdev_name_node_alt_create); 330 331 static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node) 332 { 333 list_del(&name_node->list); 334 netdev_name_node_del(name_node); 335 kfree(name_node->name); 336 netdev_name_node_free(name_node); 337 } 338 339 int netdev_name_node_alt_destroy(struct net_device *dev, const char *name) 340 { 341 struct netdev_name_node *name_node; 342 struct net *net = dev_net(dev); 343 344 name_node = netdev_name_node_lookup(net, name); 345 if (!name_node) 346 return -ENOENT; 347 /* lookup might have found our primary name or a name belonging 348 * to another device. 349 */ 350 if (name_node == dev->name_node || name_node->dev != dev) 351 return -EINVAL; 352 353 __netdev_name_node_alt_destroy(name_node); 354 355 return 0; 356 } 357 EXPORT_SYMBOL(netdev_name_node_alt_destroy); 358 359 static void netdev_name_node_alt_flush(struct net_device *dev) 360 { 361 struct netdev_name_node *name_node, *tmp; 362 363 list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list) 364 __netdev_name_node_alt_destroy(name_node); 365 } 366 367 /* Device list insertion */ 368 static void list_netdevice(struct net_device *dev) 369 { 370 struct net *net = dev_net(dev); 371 372 ASSERT_RTNL(); 373 374 write_lock_bh(&dev_base_lock); 375 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); 376 netdev_name_node_add(net, dev->name_node); 377 hlist_add_head_rcu(&dev->index_hlist, 378 dev_index_hash(net, dev->ifindex)); 379 write_unlock_bh(&dev_base_lock); 380 381 dev_base_seq_inc(net); 382 } 383 384 /* Device list removal 385 * caller must respect a RCU grace period before freeing/reusing dev 386 */ 387 static void unlist_netdevice(struct net_device *dev) 388 { 389 ASSERT_RTNL(); 390 391 /* Unlink dev from the device chain */ 392 write_lock_bh(&dev_base_lock); 393 list_del_rcu(&dev->dev_list); 394 netdev_name_node_del(dev->name_node); 395 hlist_del_rcu(&dev->index_hlist); 396 write_unlock_bh(&dev_base_lock); 397 398 dev_base_seq_inc(dev_net(dev)); 399 } 400 401 /* 402 * Our notifier list 403 */ 404 405 static RAW_NOTIFIER_HEAD(netdev_chain); 406 407 /* 408 * Device drivers call our routines to queue packets here. We empty the 409 * queue in the local softnet handler. 410 */ 411 412 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); 413 EXPORT_PER_CPU_SYMBOL(softnet_data); 414 415 #ifdef CONFIG_LOCKDEP 416 /* 417 * register_netdevice() inits txq->_xmit_lock and sets lockdep class 418 * according to dev->type 419 */ 420 static const unsigned short netdev_lock_type[] = { 421 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, 422 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, 423 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, 424 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, 425 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, 426 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, 427 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, 428 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, 429 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, 430 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, 431 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, 432 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, 433 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, 434 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, 435 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; 436 437 static const char *const netdev_lock_name[] = { 438 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", 439 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", 440 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", 441 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", 442 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", 443 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", 444 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", 445 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", 446 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", 447 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", 448 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", 449 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", 450 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", 451 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", 452 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; 453 454 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; 455 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; 456 457 static inline unsigned short netdev_lock_pos(unsigned short dev_type) 458 { 459 int i; 460 461 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) 462 if (netdev_lock_type[i] == dev_type) 463 return i; 464 /* the last key is used by default */ 465 return ARRAY_SIZE(netdev_lock_type) - 1; 466 } 467 468 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 469 unsigned short dev_type) 470 { 471 int i; 472 473 i = netdev_lock_pos(dev_type); 474 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], 475 netdev_lock_name[i]); 476 } 477 478 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 479 { 480 int i; 481 482 i = netdev_lock_pos(dev->type); 483 lockdep_set_class_and_name(&dev->addr_list_lock, 484 &netdev_addr_lock_key[i], 485 netdev_lock_name[i]); 486 } 487 #else 488 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 489 unsigned short dev_type) 490 { 491 } 492 493 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 494 { 495 } 496 #endif 497 498 /******************************************************************************* 499 * 500 * Protocol management and registration routines 501 * 502 *******************************************************************************/ 503 504 505 /* 506 * Add a protocol ID to the list. Now that the input handler is 507 * smarter we can dispense with all the messy stuff that used to be 508 * here. 509 * 510 * BEWARE!!! Protocol handlers, mangling input packets, 511 * MUST BE last in hash buckets and checking protocol handlers 512 * MUST start from promiscuous ptype_all chain in net_bh. 513 * It is true now, do not change it. 514 * Explanation follows: if protocol handler, mangling packet, will 515 * be the first on list, it is not able to sense, that packet 516 * is cloned and should be copied-on-write, so that it will 517 * change it and subsequent readers will get broken packet. 518 * --ANK (980803) 519 */ 520 521 static inline struct list_head *ptype_head(const struct packet_type *pt) 522 { 523 if (pt->type == htons(ETH_P_ALL)) 524 return pt->dev ? &pt->dev->ptype_all : &ptype_all; 525 else 526 return pt->dev ? &pt->dev->ptype_specific : 527 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; 528 } 529 530 /** 531 * dev_add_pack - add packet handler 532 * @pt: packet type declaration 533 * 534 * Add a protocol handler to the networking stack. The passed &packet_type 535 * is linked into kernel lists and may not be freed until it has been 536 * removed from the kernel lists. 537 * 538 * This call does not sleep therefore it can not 539 * guarantee all CPU's that are in middle of receiving packets 540 * will see the new packet type (until the next received packet). 541 */ 542 543 void dev_add_pack(struct packet_type *pt) 544 { 545 struct list_head *head = ptype_head(pt); 546 547 spin_lock(&ptype_lock); 548 list_add_rcu(&pt->list, head); 549 spin_unlock(&ptype_lock); 550 } 551 EXPORT_SYMBOL(dev_add_pack); 552 553 /** 554 * __dev_remove_pack - remove packet handler 555 * @pt: packet type declaration 556 * 557 * Remove a protocol handler that was previously added to the kernel 558 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 559 * from the kernel lists and can be freed or reused once this function 560 * returns. 561 * 562 * The packet type might still be in use by receivers 563 * and must not be freed until after all the CPU's have gone 564 * through a quiescent state. 565 */ 566 void __dev_remove_pack(struct packet_type *pt) 567 { 568 struct list_head *head = ptype_head(pt); 569 struct packet_type *pt1; 570 571 spin_lock(&ptype_lock); 572 573 list_for_each_entry(pt1, head, list) { 574 if (pt == pt1) { 575 list_del_rcu(&pt->list); 576 goto out; 577 } 578 } 579 580 pr_warn("dev_remove_pack: %p not found\n", pt); 581 out: 582 spin_unlock(&ptype_lock); 583 } 584 EXPORT_SYMBOL(__dev_remove_pack); 585 586 /** 587 * dev_remove_pack - remove packet handler 588 * @pt: packet type declaration 589 * 590 * Remove a protocol handler that was previously added to the kernel 591 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 592 * from the kernel lists and can be freed or reused once this function 593 * returns. 594 * 595 * This call sleeps to guarantee that no CPU is looking at the packet 596 * type after return. 597 */ 598 void dev_remove_pack(struct packet_type *pt) 599 { 600 __dev_remove_pack(pt); 601 602 synchronize_net(); 603 } 604 EXPORT_SYMBOL(dev_remove_pack); 605 606 607 /** 608 * dev_add_offload - register offload handlers 609 * @po: protocol offload declaration 610 * 611 * Add protocol offload handlers to the networking stack. The passed 612 * &proto_offload is linked into kernel lists and may not be freed until 613 * it has been removed from the kernel lists. 614 * 615 * This call does not sleep therefore it can not 616 * guarantee all CPU's that are in middle of receiving packets 617 * will see the new offload handlers (until the next received packet). 618 */ 619 void dev_add_offload(struct packet_offload *po) 620 { 621 struct packet_offload *elem; 622 623 spin_lock(&offload_lock); 624 list_for_each_entry(elem, &offload_base, list) { 625 if (po->priority < elem->priority) 626 break; 627 } 628 list_add_rcu(&po->list, elem->list.prev); 629 spin_unlock(&offload_lock); 630 } 631 EXPORT_SYMBOL(dev_add_offload); 632 633 /** 634 * __dev_remove_offload - remove offload handler 635 * @po: packet offload declaration 636 * 637 * Remove a protocol offload handler that was previously added to the 638 * kernel offload handlers by dev_add_offload(). The passed &offload_type 639 * is removed from the kernel lists and can be freed or reused once this 640 * function returns. 641 * 642 * The packet type might still be in use by receivers 643 * and must not be freed until after all the CPU's have gone 644 * through a quiescent state. 645 */ 646 static void __dev_remove_offload(struct packet_offload *po) 647 { 648 struct list_head *head = &offload_base; 649 struct packet_offload *po1; 650 651 spin_lock(&offload_lock); 652 653 list_for_each_entry(po1, head, list) { 654 if (po == po1) { 655 list_del_rcu(&po->list); 656 goto out; 657 } 658 } 659 660 pr_warn("dev_remove_offload: %p not found\n", po); 661 out: 662 spin_unlock(&offload_lock); 663 } 664 665 /** 666 * dev_remove_offload - remove packet offload handler 667 * @po: packet offload declaration 668 * 669 * Remove a packet offload handler that was previously added to the kernel 670 * offload handlers by dev_add_offload(). The passed &offload_type is 671 * removed from the kernel lists and can be freed or reused once this 672 * function returns. 673 * 674 * This call sleeps to guarantee that no CPU is looking at the packet 675 * type after return. 676 */ 677 void dev_remove_offload(struct packet_offload *po) 678 { 679 __dev_remove_offload(po); 680 681 synchronize_net(); 682 } 683 EXPORT_SYMBOL(dev_remove_offload); 684 685 /******************************************************************************* 686 * 687 * Device Interface Subroutines 688 * 689 *******************************************************************************/ 690 691 /** 692 * dev_get_iflink - get 'iflink' value of a interface 693 * @dev: targeted interface 694 * 695 * Indicates the ifindex the interface is linked to. 696 * Physical interfaces have the same 'ifindex' and 'iflink' values. 697 */ 698 699 int dev_get_iflink(const struct net_device *dev) 700 { 701 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink) 702 return dev->netdev_ops->ndo_get_iflink(dev); 703 704 return dev->ifindex; 705 } 706 EXPORT_SYMBOL(dev_get_iflink); 707 708 /** 709 * dev_fill_metadata_dst - Retrieve tunnel egress information. 710 * @dev: targeted interface 711 * @skb: The packet. 712 * 713 * For better visibility of tunnel traffic OVS needs to retrieve 714 * egress tunnel information for a packet. Following API allows 715 * user to get this info. 716 */ 717 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) 718 { 719 struct ip_tunnel_info *info; 720 721 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst) 722 return -EINVAL; 723 724 info = skb_tunnel_info_unclone(skb); 725 if (!info) 726 return -ENOMEM; 727 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX))) 728 return -EINVAL; 729 730 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb); 731 } 732 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst); 733 734 static struct net_device_path *dev_fwd_path(struct net_device_path_stack *stack) 735 { 736 int k = stack->num_paths++; 737 738 if (WARN_ON_ONCE(k >= NET_DEVICE_PATH_STACK_MAX)) 739 return NULL; 740 741 return &stack->path[k]; 742 } 743 744 int dev_fill_forward_path(const struct net_device *dev, const u8 *daddr, 745 struct net_device_path_stack *stack) 746 { 747 const struct net_device *last_dev; 748 struct net_device_path_ctx ctx = { 749 .dev = dev, 750 .daddr = daddr, 751 }; 752 struct net_device_path *path; 753 int ret = 0; 754 755 stack->num_paths = 0; 756 while (ctx.dev && ctx.dev->netdev_ops->ndo_fill_forward_path) { 757 last_dev = ctx.dev; 758 path = dev_fwd_path(stack); 759 if (!path) 760 return -1; 761 762 memset(path, 0, sizeof(struct net_device_path)); 763 ret = ctx.dev->netdev_ops->ndo_fill_forward_path(&ctx, path); 764 if (ret < 0) 765 return -1; 766 767 if (WARN_ON_ONCE(last_dev == ctx.dev)) 768 return -1; 769 } 770 path = dev_fwd_path(stack); 771 if (!path) 772 return -1; 773 path->type = DEV_PATH_ETHERNET; 774 path->dev = ctx.dev; 775 776 return ret; 777 } 778 EXPORT_SYMBOL_GPL(dev_fill_forward_path); 779 780 /** 781 * __dev_get_by_name - find a device by its name 782 * @net: the applicable net namespace 783 * @name: name to find 784 * 785 * Find an interface by name. Must be called under RTNL semaphore 786 * or @dev_base_lock. If the name is found a pointer to the device 787 * is returned. If the name is not found then %NULL is returned. The 788 * reference counters are not incremented so the caller must be 789 * careful with locks. 790 */ 791 792 struct net_device *__dev_get_by_name(struct net *net, const char *name) 793 { 794 struct netdev_name_node *node_name; 795 796 node_name = netdev_name_node_lookup(net, name); 797 return node_name ? node_name->dev : NULL; 798 } 799 EXPORT_SYMBOL(__dev_get_by_name); 800 801 /** 802 * dev_get_by_name_rcu - find a device by its name 803 * @net: the applicable net namespace 804 * @name: name to find 805 * 806 * Find an interface by name. 807 * If the name is found a pointer to the device is returned. 808 * If the name is not found then %NULL is returned. 809 * The reference counters are not incremented so the caller must be 810 * careful with locks. The caller must hold RCU lock. 811 */ 812 813 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) 814 { 815 struct netdev_name_node *node_name; 816 817 node_name = netdev_name_node_lookup_rcu(net, name); 818 return node_name ? node_name->dev : NULL; 819 } 820 EXPORT_SYMBOL(dev_get_by_name_rcu); 821 822 /** 823 * dev_get_by_name - find a device by its name 824 * @net: the applicable net namespace 825 * @name: name to find 826 * 827 * Find an interface by name. This can be called from any 828 * context and does its own locking. The returned handle has 829 * the usage count incremented and the caller must use dev_put() to 830 * release it when it is no longer needed. %NULL is returned if no 831 * matching device is found. 832 */ 833 834 struct net_device *dev_get_by_name(struct net *net, const char *name) 835 { 836 struct net_device *dev; 837 838 rcu_read_lock(); 839 dev = dev_get_by_name_rcu(net, name); 840 dev_hold(dev); 841 rcu_read_unlock(); 842 return dev; 843 } 844 EXPORT_SYMBOL(dev_get_by_name); 845 846 /** 847 * __dev_get_by_index - find a device by its ifindex 848 * @net: the applicable net namespace 849 * @ifindex: index of device 850 * 851 * Search for an interface by index. Returns %NULL if the device 852 * is not found or a pointer to the device. The device has not 853 * had its reference counter increased so the caller must be careful 854 * about locking. The caller must hold either the RTNL semaphore 855 * or @dev_base_lock. 856 */ 857 858 struct net_device *__dev_get_by_index(struct net *net, int ifindex) 859 { 860 struct net_device *dev; 861 struct hlist_head *head = dev_index_hash(net, ifindex); 862 863 hlist_for_each_entry(dev, head, index_hlist) 864 if (dev->ifindex == ifindex) 865 return dev; 866 867 return NULL; 868 } 869 EXPORT_SYMBOL(__dev_get_by_index); 870 871 /** 872 * dev_get_by_index_rcu - find a device by its ifindex 873 * @net: the applicable net namespace 874 * @ifindex: index of device 875 * 876 * Search for an interface by index. Returns %NULL if the device 877 * is not found or a pointer to the device. The device has not 878 * had its reference counter increased so the caller must be careful 879 * about locking. The caller must hold RCU lock. 880 */ 881 882 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) 883 { 884 struct net_device *dev; 885 struct hlist_head *head = dev_index_hash(net, ifindex); 886 887 hlist_for_each_entry_rcu(dev, head, index_hlist) 888 if (dev->ifindex == ifindex) 889 return dev; 890 891 return NULL; 892 } 893 EXPORT_SYMBOL(dev_get_by_index_rcu); 894 895 896 /** 897 * dev_get_by_index - find a device by its ifindex 898 * @net: the applicable net namespace 899 * @ifindex: index of device 900 * 901 * Search for an interface by index. Returns NULL if the device 902 * is not found or a pointer to the device. The device returned has 903 * had a reference added and the pointer is safe until the user calls 904 * dev_put to indicate they have finished with it. 905 */ 906 907 struct net_device *dev_get_by_index(struct net *net, int ifindex) 908 { 909 struct net_device *dev; 910 911 rcu_read_lock(); 912 dev = dev_get_by_index_rcu(net, ifindex); 913 dev_hold(dev); 914 rcu_read_unlock(); 915 return dev; 916 } 917 EXPORT_SYMBOL(dev_get_by_index); 918 919 /** 920 * dev_get_by_napi_id - find a device by napi_id 921 * @napi_id: ID of the NAPI struct 922 * 923 * Search for an interface by NAPI ID. Returns %NULL if the device 924 * is not found or a pointer to the device. The device has not had 925 * its reference counter increased so the caller must be careful 926 * about locking. The caller must hold RCU lock. 927 */ 928 929 struct net_device *dev_get_by_napi_id(unsigned int napi_id) 930 { 931 struct napi_struct *napi; 932 933 WARN_ON_ONCE(!rcu_read_lock_held()); 934 935 if (napi_id < MIN_NAPI_ID) 936 return NULL; 937 938 napi = napi_by_id(napi_id); 939 940 return napi ? napi->dev : NULL; 941 } 942 EXPORT_SYMBOL(dev_get_by_napi_id); 943 944 /** 945 * netdev_get_name - get a netdevice name, knowing its ifindex. 946 * @net: network namespace 947 * @name: a pointer to the buffer where the name will be stored. 948 * @ifindex: the ifindex of the interface to get the name from. 949 */ 950 int netdev_get_name(struct net *net, char *name, int ifindex) 951 { 952 struct net_device *dev; 953 int ret; 954 955 down_read(&devnet_rename_sem); 956 rcu_read_lock(); 957 958 dev = dev_get_by_index_rcu(net, ifindex); 959 if (!dev) { 960 ret = -ENODEV; 961 goto out; 962 } 963 964 strcpy(name, dev->name); 965 966 ret = 0; 967 out: 968 rcu_read_unlock(); 969 up_read(&devnet_rename_sem); 970 return ret; 971 } 972 973 /** 974 * dev_getbyhwaddr_rcu - find a device by its hardware address 975 * @net: the applicable net namespace 976 * @type: media type of device 977 * @ha: hardware address 978 * 979 * Search for an interface by MAC address. Returns NULL if the device 980 * is not found or a pointer to the device. 981 * The caller must hold RCU or RTNL. 982 * The returned device has not had its ref count increased 983 * and the caller must therefore be careful about locking 984 * 985 */ 986 987 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, 988 const char *ha) 989 { 990 struct net_device *dev; 991 992 for_each_netdev_rcu(net, dev) 993 if (dev->type == type && 994 !memcmp(dev->dev_addr, ha, dev->addr_len)) 995 return dev; 996 997 return NULL; 998 } 999 EXPORT_SYMBOL(dev_getbyhwaddr_rcu); 1000 1001 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) 1002 { 1003 struct net_device *dev, *ret = NULL; 1004 1005 rcu_read_lock(); 1006 for_each_netdev_rcu(net, dev) 1007 if (dev->type == type) { 1008 dev_hold(dev); 1009 ret = dev; 1010 break; 1011 } 1012 rcu_read_unlock(); 1013 return ret; 1014 } 1015 EXPORT_SYMBOL(dev_getfirstbyhwtype); 1016 1017 /** 1018 * __dev_get_by_flags - find any device with given flags 1019 * @net: the applicable net namespace 1020 * @if_flags: IFF_* values 1021 * @mask: bitmask of bits in if_flags to check 1022 * 1023 * Search for any interface with the given flags. Returns NULL if a device 1024 * is not found or a pointer to the device. Must be called inside 1025 * rtnl_lock(), and result refcount is unchanged. 1026 */ 1027 1028 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags, 1029 unsigned short mask) 1030 { 1031 struct net_device *dev, *ret; 1032 1033 ASSERT_RTNL(); 1034 1035 ret = NULL; 1036 for_each_netdev(net, dev) { 1037 if (((dev->flags ^ if_flags) & mask) == 0) { 1038 ret = dev; 1039 break; 1040 } 1041 } 1042 return ret; 1043 } 1044 EXPORT_SYMBOL(__dev_get_by_flags); 1045 1046 /** 1047 * dev_valid_name - check if name is okay for network device 1048 * @name: name string 1049 * 1050 * Network device names need to be valid file names to 1051 * allow sysfs to work. We also disallow any kind of 1052 * whitespace. 1053 */ 1054 bool dev_valid_name(const char *name) 1055 { 1056 if (*name == '\0') 1057 return false; 1058 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ) 1059 return false; 1060 if (!strcmp(name, ".") || !strcmp(name, "..")) 1061 return false; 1062 1063 while (*name) { 1064 if (*name == '/' || *name == ':' || isspace(*name)) 1065 return false; 1066 name++; 1067 } 1068 return true; 1069 } 1070 EXPORT_SYMBOL(dev_valid_name); 1071 1072 /** 1073 * __dev_alloc_name - allocate a name for a device 1074 * @net: network namespace to allocate the device name in 1075 * @name: name format string 1076 * @buf: scratch buffer and result name string 1077 * 1078 * Passed a format string - eg "lt%d" it will try and find a suitable 1079 * id. It scans list of devices to build up a free map, then chooses 1080 * the first empty slot. The caller must hold the dev_base or rtnl lock 1081 * while allocating the name and adding the device in order to avoid 1082 * duplicates. 1083 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1084 * Returns the number of the unit assigned or a negative errno code. 1085 */ 1086 1087 static int __dev_alloc_name(struct net *net, const char *name, char *buf) 1088 { 1089 int i = 0; 1090 const char *p; 1091 const int max_netdevices = 8*PAGE_SIZE; 1092 unsigned long *inuse; 1093 struct net_device *d; 1094 1095 if (!dev_valid_name(name)) 1096 return -EINVAL; 1097 1098 p = strchr(name, '%'); 1099 if (p) { 1100 /* 1101 * Verify the string as this thing may have come from 1102 * the user. There must be either one "%d" and no other "%" 1103 * characters. 1104 */ 1105 if (p[1] != 'd' || strchr(p + 2, '%')) 1106 return -EINVAL; 1107 1108 /* Use one page as a bit array of possible slots */ 1109 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); 1110 if (!inuse) 1111 return -ENOMEM; 1112 1113 for_each_netdev(net, d) { 1114 struct netdev_name_node *name_node; 1115 list_for_each_entry(name_node, &d->name_node->list, list) { 1116 if (!sscanf(name_node->name, name, &i)) 1117 continue; 1118 if (i < 0 || i >= max_netdevices) 1119 continue; 1120 1121 /* avoid cases where sscanf is not exact inverse of printf */ 1122 snprintf(buf, IFNAMSIZ, name, i); 1123 if (!strncmp(buf, name_node->name, IFNAMSIZ)) 1124 set_bit(i, inuse); 1125 } 1126 if (!sscanf(d->name, name, &i)) 1127 continue; 1128 if (i < 0 || i >= max_netdevices) 1129 continue; 1130 1131 /* avoid cases where sscanf is not exact inverse of printf */ 1132 snprintf(buf, IFNAMSIZ, name, i); 1133 if (!strncmp(buf, d->name, IFNAMSIZ)) 1134 set_bit(i, inuse); 1135 } 1136 1137 i = find_first_zero_bit(inuse, max_netdevices); 1138 free_page((unsigned long) inuse); 1139 } 1140 1141 snprintf(buf, IFNAMSIZ, name, i); 1142 if (!netdev_name_in_use(net, buf)) 1143 return i; 1144 1145 /* It is possible to run out of possible slots 1146 * when the name is long and there isn't enough space left 1147 * for the digits, or if all bits are used. 1148 */ 1149 return -ENFILE; 1150 } 1151 1152 static int dev_alloc_name_ns(struct net *net, 1153 struct net_device *dev, 1154 const char *name) 1155 { 1156 char buf[IFNAMSIZ]; 1157 int ret; 1158 1159 BUG_ON(!net); 1160 ret = __dev_alloc_name(net, name, buf); 1161 if (ret >= 0) 1162 strlcpy(dev->name, buf, IFNAMSIZ); 1163 return ret; 1164 } 1165 1166 /** 1167 * dev_alloc_name - allocate a name for a device 1168 * @dev: device 1169 * @name: name format string 1170 * 1171 * Passed a format string - eg "lt%d" it will try and find a suitable 1172 * id. It scans list of devices to build up a free map, then chooses 1173 * the first empty slot. The caller must hold the dev_base or rtnl lock 1174 * while allocating the name and adding the device in order to avoid 1175 * duplicates. 1176 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1177 * Returns the number of the unit assigned or a negative errno code. 1178 */ 1179 1180 int dev_alloc_name(struct net_device *dev, const char *name) 1181 { 1182 return dev_alloc_name_ns(dev_net(dev), dev, name); 1183 } 1184 EXPORT_SYMBOL(dev_alloc_name); 1185 1186 static int dev_get_valid_name(struct net *net, struct net_device *dev, 1187 const char *name) 1188 { 1189 BUG_ON(!net); 1190 1191 if (!dev_valid_name(name)) 1192 return -EINVAL; 1193 1194 if (strchr(name, '%')) 1195 return dev_alloc_name_ns(net, dev, name); 1196 else if (netdev_name_in_use(net, name)) 1197 return -EEXIST; 1198 else if (dev->name != name) 1199 strlcpy(dev->name, name, IFNAMSIZ); 1200 1201 return 0; 1202 } 1203 1204 /** 1205 * dev_change_name - change name of a device 1206 * @dev: device 1207 * @newname: name (or format string) must be at least IFNAMSIZ 1208 * 1209 * Change name of a device, can pass format strings "eth%d". 1210 * for wildcarding. 1211 */ 1212 int dev_change_name(struct net_device *dev, const char *newname) 1213 { 1214 unsigned char old_assign_type; 1215 char oldname[IFNAMSIZ]; 1216 int err = 0; 1217 int ret; 1218 struct net *net; 1219 1220 ASSERT_RTNL(); 1221 BUG_ON(!dev_net(dev)); 1222 1223 net = dev_net(dev); 1224 1225 /* Some auto-enslaved devices e.g. failover slaves are 1226 * special, as userspace might rename the device after 1227 * the interface had been brought up and running since 1228 * the point kernel initiated auto-enslavement. Allow 1229 * live name change even when these slave devices are 1230 * up and running. 1231 * 1232 * Typically, users of these auto-enslaving devices 1233 * don't actually care about slave name change, as 1234 * they are supposed to operate on master interface 1235 * directly. 1236 */ 1237 if (dev->flags & IFF_UP && 1238 likely(!(dev->priv_flags & IFF_LIVE_RENAME_OK))) 1239 return -EBUSY; 1240 1241 down_write(&devnet_rename_sem); 1242 1243 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { 1244 up_write(&devnet_rename_sem); 1245 return 0; 1246 } 1247 1248 memcpy(oldname, dev->name, IFNAMSIZ); 1249 1250 err = dev_get_valid_name(net, dev, newname); 1251 if (err < 0) { 1252 up_write(&devnet_rename_sem); 1253 return err; 1254 } 1255 1256 if (oldname[0] && !strchr(oldname, '%')) 1257 netdev_info(dev, "renamed from %s\n", oldname); 1258 1259 old_assign_type = dev->name_assign_type; 1260 dev->name_assign_type = NET_NAME_RENAMED; 1261 1262 rollback: 1263 ret = device_rename(&dev->dev, dev->name); 1264 if (ret) { 1265 memcpy(dev->name, oldname, IFNAMSIZ); 1266 dev->name_assign_type = old_assign_type; 1267 up_write(&devnet_rename_sem); 1268 return ret; 1269 } 1270 1271 up_write(&devnet_rename_sem); 1272 1273 netdev_adjacent_rename_links(dev, oldname); 1274 1275 write_lock_bh(&dev_base_lock); 1276 netdev_name_node_del(dev->name_node); 1277 write_unlock_bh(&dev_base_lock); 1278 1279 synchronize_rcu(); 1280 1281 write_lock_bh(&dev_base_lock); 1282 netdev_name_node_add(net, dev->name_node); 1283 write_unlock_bh(&dev_base_lock); 1284 1285 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); 1286 ret = notifier_to_errno(ret); 1287 1288 if (ret) { 1289 /* err >= 0 after dev_alloc_name() or stores the first errno */ 1290 if (err >= 0) { 1291 err = ret; 1292 down_write(&devnet_rename_sem); 1293 memcpy(dev->name, oldname, IFNAMSIZ); 1294 memcpy(oldname, newname, IFNAMSIZ); 1295 dev->name_assign_type = old_assign_type; 1296 old_assign_type = NET_NAME_RENAMED; 1297 goto rollback; 1298 } else { 1299 netdev_err(dev, "name change rollback failed: %d\n", 1300 ret); 1301 } 1302 } 1303 1304 return err; 1305 } 1306 1307 /** 1308 * dev_set_alias - change ifalias of a device 1309 * @dev: device 1310 * @alias: name up to IFALIASZ 1311 * @len: limit of bytes to copy from info 1312 * 1313 * Set ifalias for a device, 1314 */ 1315 int dev_set_alias(struct net_device *dev, const char *alias, size_t len) 1316 { 1317 struct dev_ifalias *new_alias = NULL; 1318 1319 if (len >= IFALIASZ) 1320 return -EINVAL; 1321 1322 if (len) { 1323 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL); 1324 if (!new_alias) 1325 return -ENOMEM; 1326 1327 memcpy(new_alias->ifalias, alias, len); 1328 new_alias->ifalias[len] = 0; 1329 } 1330 1331 mutex_lock(&ifalias_mutex); 1332 new_alias = rcu_replace_pointer(dev->ifalias, new_alias, 1333 mutex_is_locked(&ifalias_mutex)); 1334 mutex_unlock(&ifalias_mutex); 1335 1336 if (new_alias) 1337 kfree_rcu(new_alias, rcuhead); 1338 1339 return len; 1340 } 1341 EXPORT_SYMBOL(dev_set_alias); 1342 1343 /** 1344 * dev_get_alias - get ifalias of a device 1345 * @dev: device 1346 * @name: buffer to store name of ifalias 1347 * @len: size of buffer 1348 * 1349 * get ifalias for a device. Caller must make sure dev cannot go 1350 * away, e.g. rcu read lock or own a reference count to device. 1351 */ 1352 int dev_get_alias(const struct net_device *dev, char *name, size_t len) 1353 { 1354 const struct dev_ifalias *alias; 1355 int ret = 0; 1356 1357 rcu_read_lock(); 1358 alias = rcu_dereference(dev->ifalias); 1359 if (alias) 1360 ret = snprintf(name, len, "%s", alias->ifalias); 1361 rcu_read_unlock(); 1362 1363 return ret; 1364 } 1365 1366 /** 1367 * netdev_features_change - device changes features 1368 * @dev: device to cause notification 1369 * 1370 * Called to indicate a device has changed features. 1371 */ 1372 void netdev_features_change(struct net_device *dev) 1373 { 1374 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); 1375 } 1376 EXPORT_SYMBOL(netdev_features_change); 1377 1378 /** 1379 * netdev_state_change - device changes state 1380 * @dev: device to cause notification 1381 * 1382 * Called to indicate a device has changed state. This function calls 1383 * the notifier chains for netdev_chain and sends a NEWLINK message 1384 * to the routing socket. 1385 */ 1386 void netdev_state_change(struct net_device *dev) 1387 { 1388 if (dev->flags & IFF_UP) { 1389 struct netdev_notifier_change_info change_info = { 1390 .info.dev = dev, 1391 }; 1392 1393 call_netdevice_notifiers_info(NETDEV_CHANGE, 1394 &change_info.info); 1395 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL); 1396 } 1397 } 1398 EXPORT_SYMBOL(netdev_state_change); 1399 1400 /** 1401 * __netdev_notify_peers - notify network peers about existence of @dev, 1402 * to be called when rtnl lock is already held. 1403 * @dev: network device 1404 * 1405 * Generate traffic such that interested network peers are aware of 1406 * @dev, such as by generating a gratuitous ARP. This may be used when 1407 * a device wants to inform the rest of the network about some sort of 1408 * reconfiguration such as a failover event or virtual machine 1409 * migration. 1410 */ 1411 void __netdev_notify_peers(struct net_device *dev) 1412 { 1413 ASSERT_RTNL(); 1414 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); 1415 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev); 1416 } 1417 EXPORT_SYMBOL(__netdev_notify_peers); 1418 1419 /** 1420 * netdev_notify_peers - notify network peers about existence of @dev 1421 * @dev: network device 1422 * 1423 * Generate traffic such that interested network peers are aware of 1424 * @dev, such as by generating a gratuitous ARP. This may be used when 1425 * a device wants to inform the rest of the network about some sort of 1426 * reconfiguration such as a failover event or virtual machine 1427 * migration. 1428 */ 1429 void netdev_notify_peers(struct net_device *dev) 1430 { 1431 rtnl_lock(); 1432 __netdev_notify_peers(dev); 1433 rtnl_unlock(); 1434 } 1435 EXPORT_SYMBOL(netdev_notify_peers); 1436 1437 static int napi_threaded_poll(void *data); 1438 1439 static int napi_kthread_create(struct napi_struct *n) 1440 { 1441 int err = 0; 1442 1443 /* Create and wake up the kthread once to put it in 1444 * TASK_INTERRUPTIBLE mode to avoid the blocked task 1445 * warning and work with loadavg. 1446 */ 1447 n->thread = kthread_run(napi_threaded_poll, n, "napi/%s-%d", 1448 n->dev->name, n->napi_id); 1449 if (IS_ERR(n->thread)) { 1450 err = PTR_ERR(n->thread); 1451 pr_err("kthread_run failed with err %d\n", err); 1452 n->thread = NULL; 1453 } 1454 1455 return err; 1456 } 1457 1458 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1459 { 1460 const struct net_device_ops *ops = dev->netdev_ops; 1461 int ret; 1462 1463 ASSERT_RTNL(); 1464 1465 if (!netif_device_present(dev)) { 1466 /* may be detached because parent is runtime-suspended */ 1467 if (dev->dev.parent) 1468 pm_runtime_resume(dev->dev.parent); 1469 if (!netif_device_present(dev)) 1470 return -ENODEV; 1471 } 1472 1473 /* Block netpoll from trying to do any rx path servicing. 1474 * If we don't do this there is a chance ndo_poll_controller 1475 * or ndo_poll may be running while we open the device 1476 */ 1477 netpoll_poll_disable(dev); 1478 1479 ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack); 1480 ret = notifier_to_errno(ret); 1481 if (ret) 1482 return ret; 1483 1484 set_bit(__LINK_STATE_START, &dev->state); 1485 1486 if (ops->ndo_validate_addr) 1487 ret = ops->ndo_validate_addr(dev); 1488 1489 if (!ret && ops->ndo_open) 1490 ret = ops->ndo_open(dev); 1491 1492 netpoll_poll_enable(dev); 1493 1494 if (ret) 1495 clear_bit(__LINK_STATE_START, &dev->state); 1496 else { 1497 dev->flags |= IFF_UP; 1498 dev_set_rx_mode(dev); 1499 dev_activate(dev); 1500 add_device_randomness(dev->dev_addr, dev->addr_len); 1501 } 1502 1503 return ret; 1504 } 1505 1506 /** 1507 * dev_open - prepare an interface for use. 1508 * @dev: device to open 1509 * @extack: netlink extended ack 1510 * 1511 * Takes a device from down to up state. The device's private open 1512 * function is invoked and then the multicast lists are loaded. Finally 1513 * the device is moved into the up state and a %NETDEV_UP message is 1514 * sent to the netdev notifier chain. 1515 * 1516 * Calling this function on an active interface is a nop. On a failure 1517 * a negative errno code is returned. 1518 */ 1519 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1520 { 1521 int ret; 1522 1523 if (dev->flags & IFF_UP) 1524 return 0; 1525 1526 ret = __dev_open(dev, extack); 1527 if (ret < 0) 1528 return ret; 1529 1530 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); 1531 call_netdevice_notifiers(NETDEV_UP, dev); 1532 1533 return ret; 1534 } 1535 EXPORT_SYMBOL(dev_open); 1536 1537 static void __dev_close_many(struct list_head *head) 1538 { 1539 struct net_device *dev; 1540 1541 ASSERT_RTNL(); 1542 might_sleep(); 1543 1544 list_for_each_entry(dev, head, close_list) { 1545 /* Temporarily disable netpoll until the interface is down */ 1546 netpoll_poll_disable(dev); 1547 1548 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); 1549 1550 clear_bit(__LINK_STATE_START, &dev->state); 1551 1552 /* Synchronize to scheduled poll. We cannot touch poll list, it 1553 * can be even on different cpu. So just clear netif_running(). 1554 * 1555 * dev->stop() will invoke napi_disable() on all of it's 1556 * napi_struct instances on this device. 1557 */ 1558 smp_mb__after_atomic(); /* Commit netif_running(). */ 1559 } 1560 1561 dev_deactivate_many(head); 1562 1563 list_for_each_entry(dev, head, close_list) { 1564 const struct net_device_ops *ops = dev->netdev_ops; 1565 1566 /* 1567 * Call the device specific close. This cannot fail. 1568 * Only if device is UP 1569 * 1570 * We allow it to be called even after a DETACH hot-plug 1571 * event. 1572 */ 1573 if (ops->ndo_stop) 1574 ops->ndo_stop(dev); 1575 1576 dev->flags &= ~IFF_UP; 1577 netpoll_poll_enable(dev); 1578 } 1579 } 1580 1581 static void __dev_close(struct net_device *dev) 1582 { 1583 LIST_HEAD(single); 1584 1585 list_add(&dev->close_list, &single); 1586 __dev_close_many(&single); 1587 list_del(&single); 1588 } 1589 1590 void dev_close_many(struct list_head *head, bool unlink) 1591 { 1592 struct net_device *dev, *tmp; 1593 1594 /* Remove the devices that don't need to be closed */ 1595 list_for_each_entry_safe(dev, tmp, head, close_list) 1596 if (!(dev->flags & IFF_UP)) 1597 list_del_init(&dev->close_list); 1598 1599 __dev_close_many(head); 1600 1601 list_for_each_entry_safe(dev, tmp, head, close_list) { 1602 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); 1603 call_netdevice_notifiers(NETDEV_DOWN, dev); 1604 if (unlink) 1605 list_del_init(&dev->close_list); 1606 } 1607 } 1608 EXPORT_SYMBOL(dev_close_many); 1609 1610 /** 1611 * dev_close - shutdown an interface. 1612 * @dev: device to shutdown 1613 * 1614 * This function moves an active device into down state. A 1615 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device 1616 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier 1617 * chain. 1618 */ 1619 void dev_close(struct net_device *dev) 1620 { 1621 if (dev->flags & IFF_UP) { 1622 LIST_HEAD(single); 1623 1624 list_add(&dev->close_list, &single); 1625 dev_close_many(&single, true); 1626 list_del(&single); 1627 } 1628 } 1629 EXPORT_SYMBOL(dev_close); 1630 1631 1632 /** 1633 * dev_disable_lro - disable Large Receive Offload on a device 1634 * @dev: device 1635 * 1636 * Disable Large Receive Offload (LRO) on a net device. Must be 1637 * called under RTNL. This is needed if received packets may be 1638 * forwarded to another interface. 1639 */ 1640 void dev_disable_lro(struct net_device *dev) 1641 { 1642 struct net_device *lower_dev; 1643 struct list_head *iter; 1644 1645 dev->wanted_features &= ~NETIF_F_LRO; 1646 netdev_update_features(dev); 1647 1648 if (unlikely(dev->features & NETIF_F_LRO)) 1649 netdev_WARN(dev, "failed to disable LRO!\n"); 1650 1651 netdev_for_each_lower_dev(dev, lower_dev, iter) 1652 dev_disable_lro(lower_dev); 1653 } 1654 EXPORT_SYMBOL(dev_disable_lro); 1655 1656 /** 1657 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device 1658 * @dev: device 1659 * 1660 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be 1661 * called under RTNL. This is needed if Generic XDP is installed on 1662 * the device. 1663 */ 1664 static void dev_disable_gro_hw(struct net_device *dev) 1665 { 1666 dev->wanted_features &= ~NETIF_F_GRO_HW; 1667 netdev_update_features(dev); 1668 1669 if (unlikely(dev->features & NETIF_F_GRO_HW)) 1670 netdev_WARN(dev, "failed to disable GRO_HW!\n"); 1671 } 1672 1673 const char *netdev_cmd_to_name(enum netdev_cmd cmd) 1674 { 1675 #define N(val) \ 1676 case NETDEV_##val: \ 1677 return "NETDEV_" __stringify(val); 1678 switch (cmd) { 1679 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER) 1680 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE) 1681 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE) 1682 N(POST_INIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) N(CHANGEUPPER) 1683 N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) N(BONDING_INFO) 1684 N(PRECHANGEUPPER) N(CHANGELOWERSTATE) N(UDP_TUNNEL_PUSH_INFO) 1685 N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN) 1686 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO) 1687 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO) 1688 N(PRE_CHANGEADDR) 1689 } 1690 #undef N 1691 return "UNKNOWN_NETDEV_EVENT"; 1692 } 1693 EXPORT_SYMBOL_GPL(netdev_cmd_to_name); 1694 1695 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, 1696 struct net_device *dev) 1697 { 1698 struct netdev_notifier_info info = { 1699 .dev = dev, 1700 }; 1701 1702 return nb->notifier_call(nb, val, &info); 1703 } 1704 1705 static int call_netdevice_register_notifiers(struct notifier_block *nb, 1706 struct net_device *dev) 1707 { 1708 int err; 1709 1710 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); 1711 err = notifier_to_errno(err); 1712 if (err) 1713 return err; 1714 1715 if (!(dev->flags & IFF_UP)) 1716 return 0; 1717 1718 call_netdevice_notifier(nb, NETDEV_UP, dev); 1719 return 0; 1720 } 1721 1722 static void call_netdevice_unregister_notifiers(struct notifier_block *nb, 1723 struct net_device *dev) 1724 { 1725 if (dev->flags & IFF_UP) { 1726 call_netdevice_notifier(nb, NETDEV_GOING_DOWN, 1727 dev); 1728 call_netdevice_notifier(nb, NETDEV_DOWN, dev); 1729 } 1730 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); 1731 } 1732 1733 static int call_netdevice_register_net_notifiers(struct notifier_block *nb, 1734 struct net *net) 1735 { 1736 struct net_device *dev; 1737 int err; 1738 1739 for_each_netdev(net, dev) { 1740 err = call_netdevice_register_notifiers(nb, dev); 1741 if (err) 1742 goto rollback; 1743 } 1744 return 0; 1745 1746 rollback: 1747 for_each_netdev_continue_reverse(net, dev) 1748 call_netdevice_unregister_notifiers(nb, dev); 1749 return err; 1750 } 1751 1752 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb, 1753 struct net *net) 1754 { 1755 struct net_device *dev; 1756 1757 for_each_netdev(net, dev) 1758 call_netdevice_unregister_notifiers(nb, dev); 1759 } 1760 1761 static int dev_boot_phase = 1; 1762 1763 /** 1764 * register_netdevice_notifier - register a network notifier block 1765 * @nb: notifier 1766 * 1767 * Register a notifier to be called when network device events occur. 1768 * The notifier passed is linked into the kernel structures and must 1769 * not be reused until it has been unregistered. A negative errno code 1770 * is returned on a failure. 1771 * 1772 * When registered all registration and up events are replayed 1773 * to the new notifier to allow device to have a race free 1774 * view of the network device list. 1775 */ 1776 1777 int register_netdevice_notifier(struct notifier_block *nb) 1778 { 1779 struct net *net; 1780 int err; 1781 1782 /* Close race with setup_net() and cleanup_net() */ 1783 down_write(&pernet_ops_rwsem); 1784 rtnl_lock(); 1785 err = raw_notifier_chain_register(&netdev_chain, nb); 1786 if (err) 1787 goto unlock; 1788 if (dev_boot_phase) 1789 goto unlock; 1790 for_each_net(net) { 1791 err = call_netdevice_register_net_notifiers(nb, net); 1792 if (err) 1793 goto rollback; 1794 } 1795 1796 unlock: 1797 rtnl_unlock(); 1798 up_write(&pernet_ops_rwsem); 1799 return err; 1800 1801 rollback: 1802 for_each_net_continue_reverse(net) 1803 call_netdevice_unregister_net_notifiers(nb, net); 1804 1805 raw_notifier_chain_unregister(&netdev_chain, nb); 1806 goto unlock; 1807 } 1808 EXPORT_SYMBOL(register_netdevice_notifier); 1809 1810 /** 1811 * unregister_netdevice_notifier - unregister a network notifier block 1812 * @nb: notifier 1813 * 1814 * Unregister a notifier previously registered by 1815 * register_netdevice_notifier(). The notifier is unlinked into the 1816 * kernel structures and may then be reused. A negative errno code 1817 * is returned on a failure. 1818 * 1819 * After unregistering unregister and down device events are synthesized 1820 * for all devices on the device list to the removed notifier to remove 1821 * the need for special case cleanup code. 1822 */ 1823 1824 int unregister_netdevice_notifier(struct notifier_block *nb) 1825 { 1826 struct net *net; 1827 int err; 1828 1829 /* Close race with setup_net() and cleanup_net() */ 1830 down_write(&pernet_ops_rwsem); 1831 rtnl_lock(); 1832 err = raw_notifier_chain_unregister(&netdev_chain, nb); 1833 if (err) 1834 goto unlock; 1835 1836 for_each_net(net) 1837 call_netdevice_unregister_net_notifiers(nb, net); 1838 1839 unlock: 1840 rtnl_unlock(); 1841 up_write(&pernet_ops_rwsem); 1842 return err; 1843 } 1844 EXPORT_SYMBOL(unregister_netdevice_notifier); 1845 1846 static int __register_netdevice_notifier_net(struct net *net, 1847 struct notifier_block *nb, 1848 bool ignore_call_fail) 1849 { 1850 int err; 1851 1852 err = raw_notifier_chain_register(&net->netdev_chain, nb); 1853 if (err) 1854 return err; 1855 if (dev_boot_phase) 1856 return 0; 1857 1858 err = call_netdevice_register_net_notifiers(nb, net); 1859 if (err && !ignore_call_fail) 1860 goto chain_unregister; 1861 1862 return 0; 1863 1864 chain_unregister: 1865 raw_notifier_chain_unregister(&net->netdev_chain, nb); 1866 return err; 1867 } 1868 1869 static int __unregister_netdevice_notifier_net(struct net *net, 1870 struct notifier_block *nb) 1871 { 1872 int err; 1873 1874 err = raw_notifier_chain_unregister(&net->netdev_chain, nb); 1875 if (err) 1876 return err; 1877 1878 call_netdevice_unregister_net_notifiers(nb, net); 1879 return 0; 1880 } 1881 1882 /** 1883 * register_netdevice_notifier_net - register a per-netns network notifier block 1884 * @net: network namespace 1885 * @nb: notifier 1886 * 1887 * Register a notifier to be called when network device events occur. 1888 * The notifier passed is linked into the kernel structures and must 1889 * not be reused until it has been unregistered. A negative errno code 1890 * is returned on a failure. 1891 * 1892 * When registered all registration and up events are replayed 1893 * to the new notifier to allow device to have a race free 1894 * view of the network device list. 1895 */ 1896 1897 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb) 1898 { 1899 int err; 1900 1901 rtnl_lock(); 1902 err = __register_netdevice_notifier_net(net, nb, false); 1903 rtnl_unlock(); 1904 return err; 1905 } 1906 EXPORT_SYMBOL(register_netdevice_notifier_net); 1907 1908 /** 1909 * unregister_netdevice_notifier_net - unregister a per-netns 1910 * network notifier block 1911 * @net: network namespace 1912 * @nb: notifier 1913 * 1914 * Unregister a notifier previously registered by 1915 * register_netdevice_notifier(). The notifier is unlinked into the 1916 * kernel structures and may then be reused. A negative errno code 1917 * is returned on a failure. 1918 * 1919 * After unregistering unregister and down device events are synthesized 1920 * for all devices on the device list to the removed notifier to remove 1921 * the need for special case cleanup code. 1922 */ 1923 1924 int unregister_netdevice_notifier_net(struct net *net, 1925 struct notifier_block *nb) 1926 { 1927 int err; 1928 1929 rtnl_lock(); 1930 err = __unregister_netdevice_notifier_net(net, nb); 1931 rtnl_unlock(); 1932 return err; 1933 } 1934 EXPORT_SYMBOL(unregister_netdevice_notifier_net); 1935 1936 int register_netdevice_notifier_dev_net(struct net_device *dev, 1937 struct notifier_block *nb, 1938 struct netdev_net_notifier *nn) 1939 { 1940 int err; 1941 1942 rtnl_lock(); 1943 err = __register_netdevice_notifier_net(dev_net(dev), nb, false); 1944 if (!err) { 1945 nn->nb = nb; 1946 list_add(&nn->list, &dev->net_notifier_list); 1947 } 1948 rtnl_unlock(); 1949 return err; 1950 } 1951 EXPORT_SYMBOL(register_netdevice_notifier_dev_net); 1952 1953 int unregister_netdevice_notifier_dev_net(struct net_device *dev, 1954 struct notifier_block *nb, 1955 struct netdev_net_notifier *nn) 1956 { 1957 int err; 1958 1959 rtnl_lock(); 1960 list_del(&nn->list); 1961 err = __unregister_netdevice_notifier_net(dev_net(dev), nb); 1962 rtnl_unlock(); 1963 return err; 1964 } 1965 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net); 1966 1967 static void move_netdevice_notifiers_dev_net(struct net_device *dev, 1968 struct net *net) 1969 { 1970 struct netdev_net_notifier *nn; 1971 1972 list_for_each_entry(nn, &dev->net_notifier_list, list) { 1973 __unregister_netdevice_notifier_net(dev_net(dev), nn->nb); 1974 __register_netdevice_notifier_net(net, nn->nb, true); 1975 } 1976 } 1977 1978 /** 1979 * call_netdevice_notifiers_info - call all network notifier blocks 1980 * @val: value passed unmodified to notifier function 1981 * @info: notifier information data 1982 * 1983 * Call all network notifier blocks. Parameters and return value 1984 * are as for raw_notifier_call_chain(). 1985 */ 1986 1987 static int call_netdevice_notifiers_info(unsigned long val, 1988 struct netdev_notifier_info *info) 1989 { 1990 struct net *net = dev_net(info->dev); 1991 int ret; 1992 1993 ASSERT_RTNL(); 1994 1995 /* Run per-netns notifier block chain first, then run the global one. 1996 * Hopefully, one day, the global one is going to be removed after 1997 * all notifier block registrators get converted to be per-netns. 1998 */ 1999 ret = raw_notifier_call_chain(&net->netdev_chain, val, info); 2000 if (ret & NOTIFY_STOP_MASK) 2001 return ret; 2002 return raw_notifier_call_chain(&netdev_chain, val, info); 2003 } 2004 2005 static int call_netdevice_notifiers_extack(unsigned long val, 2006 struct net_device *dev, 2007 struct netlink_ext_ack *extack) 2008 { 2009 struct netdev_notifier_info info = { 2010 .dev = dev, 2011 .extack = extack, 2012 }; 2013 2014 return call_netdevice_notifiers_info(val, &info); 2015 } 2016 2017 /** 2018 * call_netdevice_notifiers - call all network notifier blocks 2019 * @val: value passed unmodified to notifier function 2020 * @dev: net_device pointer passed unmodified to notifier function 2021 * 2022 * Call all network notifier blocks. Parameters and return value 2023 * are as for raw_notifier_call_chain(). 2024 */ 2025 2026 int call_netdevice_notifiers(unsigned long val, struct net_device *dev) 2027 { 2028 return call_netdevice_notifiers_extack(val, dev, NULL); 2029 } 2030 EXPORT_SYMBOL(call_netdevice_notifiers); 2031 2032 /** 2033 * call_netdevice_notifiers_mtu - call all network notifier blocks 2034 * @val: value passed unmodified to notifier function 2035 * @dev: net_device pointer passed unmodified to notifier function 2036 * @arg: additional u32 argument passed to the notifier function 2037 * 2038 * Call all network notifier blocks. Parameters and return value 2039 * are as for raw_notifier_call_chain(). 2040 */ 2041 static int call_netdevice_notifiers_mtu(unsigned long val, 2042 struct net_device *dev, u32 arg) 2043 { 2044 struct netdev_notifier_info_ext info = { 2045 .info.dev = dev, 2046 .ext.mtu = arg, 2047 }; 2048 2049 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0); 2050 2051 return call_netdevice_notifiers_info(val, &info.info); 2052 } 2053 2054 #ifdef CONFIG_NET_INGRESS 2055 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key); 2056 2057 void net_inc_ingress_queue(void) 2058 { 2059 static_branch_inc(&ingress_needed_key); 2060 } 2061 EXPORT_SYMBOL_GPL(net_inc_ingress_queue); 2062 2063 void net_dec_ingress_queue(void) 2064 { 2065 static_branch_dec(&ingress_needed_key); 2066 } 2067 EXPORT_SYMBOL_GPL(net_dec_ingress_queue); 2068 #endif 2069 2070 #ifdef CONFIG_NET_EGRESS 2071 static DEFINE_STATIC_KEY_FALSE(egress_needed_key); 2072 2073 void net_inc_egress_queue(void) 2074 { 2075 static_branch_inc(&egress_needed_key); 2076 } 2077 EXPORT_SYMBOL_GPL(net_inc_egress_queue); 2078 2079 void net_dec_egress_queue(void) 2080 { 2081 static_branch_dec(&egress_needed_key); 2082 } 2083 EXPORT_SYMBOL_GPL(net_dec_egress_queue); 2084 #endif 2085 2086 static DEFINE_STATIC_KEY_FALSE(netstamp_needed_key); 2087 #ifdef CONFIG_JUMP_LABEL 2088 static atomic_t netstamp_needed_deferred; 2089 static atomic_t netstamp_wanted; 2090 static void netstamp_clear(struct work_struct *work) 2091 { 2092 int deferred = atomic_xchg(&netstamp_needed_deferred, 0); 2093 int wanted; 2094 2095 wanted = atomic_add_return(deferred, &netstamp_wanted); 2096 if (wanted > 0) 2097 static_branch_enable(&netstamp_needed_key); 2098 else 2099 static_branch_disable(&netstamp_needed_key); 2100 } 2101 static DECLARE_WORK(netstamp_work, netstamp_clear); 2102 #endif 2103 2104 void net_enable_timestamp(void) 2105 { 2106 #ifdef CONFIG_JUMP_LABEL 2107 int wanted; 2108 2109 while (1) { 2110 wanted = atomic_read(&netstamp_wanted); 2111 if (wanted <= 0) 2112 break; 2113 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted) 2114 return; 2115 } 2116 atomic_inc(&netstamp_needed_deferred); 2117 schedule_work(&netstamp_work); 2118 #else 2119 static_branch_inc(&netstamp_needed_key); 2120 #endif 2121 } 2122 EXPORT_SYMBOL(net_enable_timestamp); 2123 2124 void net_disable_timestamp(void) 2125 { 2126 #ifdef CONFIG_JUMP_LABEL 2127 int wanted; 2128 2129 while (1) { 2130 wanted = atomic_read(&netstamp_wanted); 2131 if (wanted <= 1) 2132 break; 2133 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted) 2134 return; 2135 } 2136 atomic_dec(&netstamp_needed_deferred); 2137 schedule_work(&netstamp_work); 2138 #else 2139 static_branch_dec(&netstamp_needed_key); 2140 #endif 2141 } 2142 EXPORT_SYMBOL(net_disable_timestamp); 2143 2144 static inline void net_timestamp_set(struct sk_buff *skb) 2145 { 2146 skb->tstamp = 0; 2147 if (static_branch_unlikely(&netstamp_needed_key)) 2148 __net_timestamp(skb); 2149 } 2150 2151 #define net_timestamp_check(COND, SKB) \ 2152 if (static_branch_unlikely(&netstamp_needed_key)) { \ 2153 if ((COND) && !(SKB)->tstamp) \ 2154 __net_timestamp(SKB); \ 2155 } \ 2156 2157 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb) 2158 { 2159 return __is_skb_forwardable(dev, skb, true); 2160 } 2161 EXPORT_SYMBOL_GPL(is_skb_forwardable); 2162 2163 static int __dev_forward_skb2(struct net_device *dev, struct sk_buff *skb, 2164 bool check_mtu) 2165 { 2166 int ret = ____dev_forward_skb(dev, skb, check_mtu); 2167 2168 if (likely(!ret)) { 2169 skb->protocol = eth_type_trans(skb, dev); 2170 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); 2171 } 2172 2173 return ret; 2174 } 2175 2176 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2177 { 2178 return __dev_forward_skb2(dev, skb, true); 2179 } 2180 EXPORT_SYMBOL_GPL(__dev_forward_skb); 2181 2182 /** 2183 * dev_forward_skb - loopback an skb to another netif 2184 * 2185 * @dev: destination network device 2186 * @skb: buffer to forward 2187 * 2188 * return values: 2189 * NET_RX_SUCCESS (no congestion) 2190 * NET_RX_DROP (packet was dropped, but freed) 2191 * 2192 * dev_forward_skb can be used for injecting an skb from the 2193 * start_xmit function of one device into the receive queue 2194 * of another device. 2195 * 2196 * The receiving device may be in another namespace, so 2197 * we have to clear all information in the skb that could 2198 * impact namespace isolation. 2199 */ 2200 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2201 { 2202 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); 2203 } 2204 EXPORT_SYMBOL_GPL(dev_forward_skb); 2205 2206 int dev_forward_skb_nomtu(struct net_device *dev, struct sk_buff *skb) 2207 { 2208 return __dev_forward_skb2(dev, skb, false) ?: netif_rx_internal(skb); 2209 } 2210 2211 static inline int deliver_skb(struct sk_buff *skb, 2212 struct packet_type *pt_prev, 2213 struct net_device *orig_dev) 2214 { 2215 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 2216 return -ENOMEM; 2217 refcount_inc(&skb->users); 2218 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 2219 } 2220 2221 static inline void deliver_ptype_list_skb(struct sk_buff *skb, 2222 struct packet_type **pt, 2223 struct net_device *orig_dev, 2224 __be16 type, 2225 struct list_head *ptype_list) 2226 { 2227 struct packet_type *ptype, *pt_prev = *pt; 2228 2229 list_for_each_entry_rcu(ptype, ptype_list, list) { 2230 if (ptype->type != type) 2231 continue; 2232 if (pt_prev) 2233 deliver_skb(skb, pt_prev, orig_dev); 2234 pt_prev = ptype; 2235 } 2236 *pt = pt_prev; 2237 } 2238 2239 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) 2240 { 2241 if (!ptype->af_packet_priv || !skb->sk) 2242 return false; 2243 2244 if (ptype->id_match) 2245 return ptype->id_match(ptype, skb->sk); 2246 else if ((struct sock *)ptype->af_packet_priv == skb->sk) 2247 return true; 2248 2249 return false; 2250 } 2251 2252 /** 2253 * dev_nit_active - return true if any network interface taps are in use 2254 * 2255 * @dev: network device to check for the presence of taps 2256 */ 2257 bool dev_nit_active(struct net_device *dev) 2258 { 2259 return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all); 2260 } 2261 EXPORT_SYMBOL_GPL(dev_nit_active); 2262 2263 /* 2264 * Support routine. Sends outgoing frames to any network 2265 * taps currently in use. 2266 */ 2267 2268 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) 2269 { 2270 struct packet_type *ptype; 2271 struct sk_buff *skb2 = NULL; 2272 struct packet_type *pt_prev = NULL; 2273 struct list_head *ptype_list = &ptype_all; 2274 2275 rcu_read_lock(); 2276 again: 2277 list_for_each_entry_rcu(ptype, ptype_list, list) { 2278 if (ptype->ignore_outgoing) 2279 continue; 2280 2281 /* Never send packets back to the socket 2282 * they originated from - MvS ([email protected]) 2283 */ 2284 if (skb_loop_sk(ptype, skb)) 2285 continue; 2286 2287 if (pt_prev) { 2288 deliver_skb(skb2, pt_prev, skb->dev); 2289 pt_prev = ptype; 2290 continue; 2291 } 2292 2293 /* need to clone skb, done only once */ 2294 skb2 = skb_clone(skb, GFP_ATOMIC); 2295 if (!skb2) 2296 goto out_unlock; 2297 2298 net_timestamp_set(skb2); 2299 2300 /* skb->nh should be correctly 2301 * set by sender, so that the second statement is 2302 * just protection against buggy protocols. 2303 */ 2304 skb_reset_mac_header(skb2); 2305 2306 if (skb_network_header(skb2) < skb2->data || 2307 skb_network_header(skb2) > skb_tail_pointer(skb2)) { 2308 net_crit_ratelimited("protocol %04x is buggy, dev %s\n", 2309 ntohs(skb2->protocol), 2310 dev->name); 2311 skb_reset_network_header(skb2); 2312 } 2313 2314 skb2->transport_header = skb2->network_header; 2315 skb2->pkt_type = PACKET_OUTGOING; 2316 pt_prev = ptype; 2317 } 2318 2319 if (ptype_list == &ptype_all) { 2320 ptype_list = &dev->ptype_all; 2321 goto again; 2322 } 2323 out_unlock: 2324 if (pt_prev) { 2325 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC)) 2326 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); 2327 else 2328 kfree_skb(skb2); 2329 } 2330 rcu_read_unlock(); 2331 } 2332 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit); 2333 2334 /** 2335 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change 2336 * @dev: Network device 2337 * @txq: number of queues available 2338 * 2339 * If real_num_tx_queues is changed the tc mappings may no longer be 2340 * valid. To resolve this verify the tc mapping remains valid and if 2341 * not NULL the mapping. With no priorities mapping to this 2342 * offset/count pair it will no longer be used. In the worst case TC0 2343 * is invalid nothing can be done so disable priority mappings. If is 2344 * expected that drivers will fix this mapping if they can before 2345 * calling netif_set_real_num_tx_queues. 2346 */ 2347 static void netif_setup_tc(struct net_device *dev, unsigned int txq) 2348 { 2349 int i; 2350 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2351 2352 /* If TC0 is invalidated disable TC mapping */ 2353 if (tc->offset + tc->count > txq) { 2354 netdev_warn(dev, "Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); 2355 dev->num_tc = 0; 2356 return; 2357 } 2358 2359 /* Invalidated prio to tc mappings set to TC0 */ 2360 for (i = 1; i < TC_BITMASK + 1; i++) { 2361 int q = netdev_get_prio_tc_map(dev, i); 2362 2363 tc = &dev->tc_to_txq[q]; 2364 if (tc->offset + tc->count > txq) { 2365 netdev_warn(dev, "Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n", 2366 i, q); 2367 netdev_set_prio_tc_map(dev, i, 0); 2368 } 2369 } 2370 } 2371 2372 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq) 2373 { 2374 if (dev->num_tc) { 2375 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2376 int i; 2377 2378 /* walk through the TCs and see if it falls into any of them */ 2379 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) { 2380 if ((txq - tc->offset) < tc->count) 2381 return i; 2382 } 2383 2384 /* didn't find it, just return -1 to indicate no match */ 2385 return -1; 2386 } 2387 2388 return 0; 2389 } 2390 EXPORT_SYMBOL(netdev_txq_to_tc); 2391 2392 #ifdef CONFIG_XPS 2393 static struct static_key xps_needed __read_mostly; 2394 static struct static_key xps_rxqs_needed __read_mostly; 2395 static DEFINE_MUTEX(xps_map_mutex); 2396 #define xmap_dereference(P) \ 2397 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) 2398 2399 static bool remove_xps_queue(struct xps_dev_maps *dev_maps, 2400 struct xps_dev_maps *old_maps, int tci, u16 index) 2401 { 2402 struct xps_map *map = NULL; 2403 int pos; 2404 2405 if (dev_maps) 2406 map = xmap_dereference(dev_maps->attr_map[tci]); 2407 if (!map) 2408 return false; 2409 2410 for (pos = map->len; pos--;) { 2411 if (map->queues[pos] != index) 2412 continue; 2413 2414 if (map->len > 1) { 2415 map->queues[pos] = map->queues[--map->len]; 2416 break; 2417 } 2418 2419 if (old_maps) 2420 RCU_INIT_POINTER(old_maps->attr_map[tci], NULL); 2421 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2422 kfree_rcu(map, rcu); 2423 return false; 2424 } 2425 2426 return true; 2427 } 2428 2429 static bool remove_xps_queue_cpu(struct net_device *dev, 2430 struct xps_dev_maps *dev_maps, 2431 int cpu, u16 offset, u16 count) 2432 { 2433 int num_tc = dev_maps->num_tc; 2434 bool active = false; 2435 int tci; 2436 2437 for (tci = cpu * num_tc; num_tc--; tci++) { 2438 int i, j; 2439 2440 for (i = count, j = offset; i--; j++) { 2441 if (!remove_xps_queue(dev_maps, NULL, tci, j)) 2442 break; 2443 } 2444 2445 active |= i < 0; 2446 } 2447 2448 return active; 2449 } 2450 2451 static void reset_xps_maps(struct net_device *dev, 2452 struct xps_dev_maps *dev_maps, 2453 enum xps_map_type type) 2454 { 2455 static_key_slow_dec_cpuslocked(&xps_needed); 2456 if (type == XPS_RXQS) 2457 static_key_slow_dec_cpuslocked(&xps_rxqs_needed); 2458 2459 RCU_INIT_POINTER(dev->xps_maps[type], NULL); 2460 2461 kfree_rcu(dev_maps, rcu); 2462 } 2463 2464 static void clean_xps_maps(struct net_device *dev, enum xps_map_type type, 2465 u16 offset, u16 count) 2466 { 2467 struct xps_dev_maps *dev_maps; 2468 bool active = false; 2469 int i, j; 2470 2471 dev_maps = xmap_dereference(dev->xps_maps[type]); 2472 if (!dev_maps) 2473 return; 2474 2475 for (j = 0; j < dev_maps->nr_ids; j++) 2476 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, count); 2477 if (!active) 2478 reset_xps_maps(dev, dev_maps, type); 2479 2480 if (type == XPS_CPUS) { 2481 for (i = offset + (count - 1); count--; i--) 2482 netdev_queue_numa_node_write( 2483 netdev_get_tx_queue(dev, i), NUMA_NO_NODE); 2484 } 2485 } 2486 2487 static void netif_reset_xps_queues(struct net_device *dev, u16 offset, 2488 u16 count) 2489 { 2490 if (!static_key_false(&xps_needed)) 2491 return; 2492 2493 cpus_read_lock(); 2494 mutex_lock(&xps_map_mutex); 2495 2496 if (static_key_false(&xps_rxqs_needed)) 2497 clean_xps_maps(dev, XPS_RXQS, offset, count); 2498 2499 clean_xps_maps(dev, XPS_CPUS, offset, count); 2500 2501 mutex_unlock(&xps_map_mutex); 2502 cpus_read_unlock(); 2503 } 2504 2505 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) 2506 { 2507 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index); 2508 } 2509 2510 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index, 2511 u16 index, bool is_rxqs_map) 2512 { 2513 struct xps_map *new_map; 2514 int alloc_len = XPS_MIN_MAP_ALLOC; 2515 int i, pos; 2516 2517 for (pos = 0; map && pos < map->len; pos++) { 2518 if (map->queues[pos] != index) 2519 continue; 2520 return map; 2521 } 2522 2523 /* Need to add tx-queue to this CPU's/rx-queue's existing map */ 2524 if (map) { 2525 if (pos < map->alloc_len) 2526 return map; 2527 2528 alloc_len = map->alloc_len * 2; 2529 } 2530 2531 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's 2532 * map 2533 */ 2534 if (is_rxqs_map) 2535 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL); 2536 else 2537 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, 2538 cpu_to_node(attr_index)); 2539 if (!new_map) 2540 return NULL; 2541 2542 for (i = 0; i < pos; i++) 2543 new_map->queues[i] = map->queues[i]; 2544 new_map->alloc_len = alloc_len; 2545 new_map->len = pos; 2546 2547 return new_map; 2548 } 2549 2550 /* Copy xps maps at a given index */ 2551 static void xps_copy_dev_maps(struct xps_dev_maps *dev_maps, 2552 struct xps_dev_maps *new_dev_maps, int index, 2553 int tc, bool skip_tc) 2554 { 2555 int i, tci = index * dev_maps->num_tc; 2556 struct xps_map *map; 2557 2558 /* copy maps belonging to foreign traffic classes */ 2559 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2560 if (i == tc && skip_tc) 2561 continue; 2562 2563 /* fill in the new device map from the old device map */ 2564 map = xmap_dereference(dev_maps->attr_map[tci]); 2565 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2566 } 2567 } 2568 2569 /* Must be called under cpus_read_lock */ 2570 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask, 2571 u16 index, enum xps_map_type type) 2572 { 2573 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL, *old_dev_maps = NULL; 2574 const unsigned long *online_mask = NULL; 2575 bool active = false, copy = false; 2576 int i, j, tci, numa_node_id = -2; 2577 int maps_sz, num_tc = 1, tc = 0; 2578 struct xps_map *map, *new_map; 2579 unsigned int nr_ids; 2580 2581 if (dev->num_tc) { 2582 /* Do not allow XPS on subordinate device directly */ 2583 num_tc = dev->num_tc; 2584 if (num_tc < 0) 2585 return -EINVAL; 2586 2587 /* If queue belongs to subordinate dev use its map */ 2588 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev; 2589 2590 tc = netdev_txq_to_tc(dev, index); 2591 if (tc < 0) 2592 return -EINVAL; 2593 } 2594 2595 mutex_lock(&xps_map_mutex); 2596 2597 dev_maps = xmap_dereference(dev->xps_maps[type]); 2598 if (type == XPS_RXQS) { 2599 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues); 2600 nr_ids = dev->num_rx_queues; 2601 } else { 2602 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc); 2603 if (num_possible_cpus() > 1) 2604 online_mask = cpumask_bits(cpu_online_mask); 2605 nr_ids = nr_cpu_ids; 2606 } 2607 2608 if (maps_sz < L1_CACHE_BYTES) 2609 maps_sz = L1_CACHE_BYTES; 2610 2611 /* The old dev_maps could be larger or smaller than the one we're 2612 * setting up now, as dev->num_tc or nr_ids could have been updated in 2613 * between. We could try to be smart, but let's be safe instead and only 2614 * copy foreign traffic classes if the two map sizes match. 2615 */ 2616 if (dev_maps && 2617 dev_maps->num_tc == num_tc && dev_maps->nr_ids == nr_ids) 2618 copy = true; 2619 2620 /* allocate memory for queue storage */ 2621 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids), 2622 j < nr_ids;) { 2623 if (!new_dev_maps) { 2624 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); 2625 if (!new_dev_maps) { 2626 mutex_unlock(&xps_map_mutex); 2627 return -ENOMEM; 2628 } 2629 2630 new_dev_maps->nr_ids = nr_ids; 2631 new_dev_maps->num_tc = num_tc; 2632 } 2633 2634 tci = j * num_tc + tc; 2635 map = copy ? xmap_dereference(dev_maps->attr_map[tci]) : NULL; 2636 2637 map = expand_xps_map(map, j, index, type == XPS_RXQS); 2638 if (!map) 2639 goto error; 2640 2641 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2642 } 2643 2644 if (!new_dev_maps) 2645 goto out_no_new_maps; 2646 2647 if (!dev_maps) { 2648 /* Increment static keys at most once per type */ 2649 static_key_slow_inc_cpuslocked(&xps_needed); 2650 if (type == XPS_RXQS) 2651 static_key_slow_inc_cpuslocked(&xps_rxqs_needed); 2652 } 2653 2654 for (j = 0; j < nr_ids; j++) { 2655 bool skip_tc = false; 2656 2657 tci = j * num_tc + tc; 2658 if (netif_attr_test_mask(j, mask, nr_ids) && 2659 netif_attr_test_online(j, online_mask, nr_ids)) { 2660 /* add tx-queue to CPU/rx-queue maps */ 2661 int pos = 0; 2662 2663 skip_tc = true; 2664 2665 map = xmap_dereference(new_dev_maps->attr_map[tci]); 2666 while ((pos < map->len) && (map->queues[pos] != index)) 2667 pos++; 2668 2669 if (pos == map->len) 2670 map->queues[map->len++] = index; 2671 #ifdef CONFIG_NUMA 2672 if (type == XPS_CPUS) { 2673 if (numa_node_id == -2) 2674 numa_node_id = cpu_to_node(j); 2675 else if (numa_node_id != cpu_to_node(j)) 2676 numa_node_id = -1; 2677 } 2678 #endif 2679 } 2680 2681 if (copy) 2682 xps_copy_dev_maps(dev_maps, new_dev_maps, j, tc, 2683 skip_tc); 2684 } 2685 2686 rcu_assign_pointer(dev->xps_maps[type], new_dev_maps); 2687 2688 /* Cleanup old maps */ 2689 if (!dev_maps) 2690 goto out_no_old_maps; 2691 2692 for (j = 0; j < dev_maps->nr_ids; j++) { 2693 for (i = num_tc, tci = j * dev_maps->num_tc; i--; tci++) { 2694 map = xmap_dereference(dev_maps->attr_map[tci]); 2695 if (!map) 2696 continue; 2697 2698 if (copy) { 2699 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2700 if (map == new_map) 2701 continue; 2702 } 2703 2704 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2705 kfree_rcu(map, rcu); 2706 } 2707 } 2708 2709 old_dev_maps = dev_maps; 2710 2711 out_no_old_maps: 2712 dev_maps = new_dev_maps; 2713 active = true; 2714 2715 out_no_new_maps: 2716 if (type == XPS_CPUS) 2717 /* update Tx queue numa node */ 2718 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), 2719 (numa_node_id >= 0) ? 2720 numa_node_id : NUMA_NO_NODE); 2721 2722 if (!dev_maps) 2723 goto out_no_maps; 2724 2725 /* removes tx-queue from unused CPUs/rx-queues */ 2726 for (j = 0; j < dev_maps->nr_ids; j++) { 2727 tci = j * dev_maps->num_tc; 2728 2729 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2730 if (i == tc && 2731 netif_attr_test_mask(j, mask, dev_maps->nr_ids) && 2732 netif_attr_test_online(j, online_mask, dev_maps->nr_ids)) 2733 continue; 2734 2735 active |= remove_xps_queue(dev_maps, 2736 copy ? old_dev_maps : NULL, 2737 tci, index); 2738 } 2739 } 2740 2741 if (old_dev_maps) 2742 kfree_rcu(old_dev_maps, rcu); 2743 2744 /* free map if not active */ 2745 if (!active) 2746 reset_xps_maps(dev, dev_maps, type); 2747 2748 out_no_maps: 2749 mutex_unlock(&xps_map_mutex); 2750 2751 return 0; 2752 error: 2753 /* remove any maps that we added */ 2754 for (j = 0; j < nr_ids; j++) { 2755 for (i = num_tc, tci = j * num_tc; i--; tci++) { 2756 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2757 map = copy ? 2758 xmap_dereference(dev_maps->attr_map[tci]) : 2759 NULL; 2760 if (new_map && new_map != map) 2761 kfree(new_map); 2762 } 2763 } 2764 2765 mutex_unlock(&xps_map_mutex); 2766 2767 kfree(new_dev_maps); 2768 return -ENOMEM; 2769 } 2770 EXPORT_SYMBOL_GPL(__netif_set_xps_queue); 2771 2772 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, 2773 u16 index) 2774 { 2775 int ret; 2776 2777 cpus_read_lock(); 2778 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, XPS_CPUS); 2779 cpus_read_unlock(); 2780 2781 return ret; 2782 } 2783 EXPORT_SYMBOL(netif_set_xps_queue); 2784 2785 #endif 2786 static void netdev_unbind_all_sb_channels(struct net_device *dev) 2787 { 2788 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2789 2790 /* Unbind any subordinate channels */ 2791 while (txq-- != &dev->_tx[0]) { 2792 if (txq->sb_dev) 2793 netdev_unbind_sb_channel(dev, txq->sb_dev); 2794 } 2795 } 2796 2797 void netdev_reset_tc(struct net_device *dev) 2798 { 2799 #ifdef CONFIG_XPS 2800 netif_reset_xps_queues_gt(dev, 0); 2801 #endif 2802 netdev_unbind_all_sb_channels(dev); 2803 2804 /* Reset TC configuration of device */ 2805 dev->num_tc = 0; 2806 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq)); 2807 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map)); 2808 } 2809 EXPORT_SYMBOL(netdev_reset_tc); 2810 2811 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset) 2812 { 2813 if (tc >= dev->num_tc) 2814 return -EINVAL; 2815 2816 #ifdef CONFIG_XPS 2817 netif_reset_xps_queues(dev, offset, count); 2818 #endif 2819 dev->tc_to_txq[tc].count = count; 2820 dev->tc_to_txq[tc].offset = offset; 2821 return 0; 2822 } 2823 EXPORT_SYMBOL(netdev_set_tc_queue); 2824 2825 int netdev_set_num_tc(struct net_device *dev, u8 num_tc) 2826 { 2827 if (num_tc > TC_MAX_QUEUE) 2828 return -EINVAL; 2829 2830 #ifdef CONFIG_XPS 2831 netif_reset_xps_queues_gt(dev, 0); 2832 #endif 2833 netdev_unbind_all_sb_channels(dev); 2834 2835 dev->num_tc = num_tc; 2836 return 0; 2837 } 2838 EXPORT_SYMBOL(netdev_set_num_tc); 2839 2840 void netdev_unbind_sb_channel(struct net_device *dev, 2841 struct net_device *sb_dev) 2842 { 2843 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2844 2845 #ifdef CONFIG_XPS 2846 netif_reset_xps_queues_gt(sb_dev, 0); 2847 #endif 2848 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq)); 2849 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map)); 2850 2851 while (txq-- != &dev->_tx[0]) { 2852 if (txq->sb_dev == sb_dev) 2853 txq->sb_dev = NULL; 2854 } 2855 } 2856 EXPORT_SYMBOL(netdev_unbind_sb_channel); 2857 2858 int netdev_bind_sb_channel_queue(struct net_device *dev, 2859 struct net_device *sb_dev, 2860 u8 tc, u16 count, u16 offset) 2861 { 2862 /* Make certain the sb_dev and dev are already configured */ 2863 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc) 2864 return -EINVAL; 2865 2866 /* We cannot hand out queues we don't have */ 2867 if ((offset + count) > dev->real_num_tx_queues) 2868 return -EINVAL; 2869 2870 /* Record the mapping */ 2871 sb_dev->tc_to_txq[tc].count = count; 2872 sb_dev->tc_to_txq[tc].offset = offset; 2873 2874 /* Provide a way for Tx queue to find the tc_to_txq map or 2875 * XPS map for itself. 2876 */ 2877 while (count--) 2878 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev; 2879 2880 return 0; 2881 } 2882 EXPORT_SYMBOL(netdev_bind_sb_channel_queue); 2883 2884 int netdev_set_sb_channel(struct net_device *dev, u16 channel) 2885 { 2886 /* Do not use a multiqueue device to represent a subordinate channel */ 2887 if (netif_is_multiqueue(dev)) 2888 return -ENODEV; 2889 2890 /* We allow channels 1 - 32767 to be used for subordinate channels. 2891 * Channel 0 is meant to be "native" mode and used only to represent 2892 * the main root device. We allow writing 0 to reset the device back 2893 * to normal mode after being used as a subordinate channel. 2894 */ 2895 if (channel > S16_MAX) 2896 return -EINVAL; 2897 2898 dev->num_tc = -channel; 2899 2900 return 0; 2901 } 2902 EXPORT_SYMBOL(netdev_set_sb_channel); 2903 2904 /* 2905 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues 2906 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed. 2907 */ 2908 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) 2909 { 2910 bool disabling; 2911 int rc; 2912 2913 disabling = txq < dev->real_num_tx_queues; 2914 2915 if (txq < 1 || txq > dev->num_tx_queues) 2916 return -EINVAL; 2917 2918 if (dev->reg_state == NETREG_REGISTERED || 2919 dev->reg_state == NETREG_UNREGISTERING) { 2920 ASSERT_RTNL(); 2921 2922 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, 2923 txq); 2924 if (rc) 2925 return rc; 2926 2927 if (dev->num_tc) 2928 netif_setup_tc(dev, txq); 2929 2930 dev_qdisc_change_real_num_tx(dev, txq); 2931 2932 dev->real_num_tx_queues = txq; 2933 2934 if (disabling) { 2935 synchronize_net(); 2936 qdisc_reset_all_tx_gt(dev, txq); 2937 #ifdef CONFIG_XPS 2938 netif_reset_xps_queues_gt(dev, txq); 2939 #endif 2940 } 2941 } else { 2942 dev->real_num_tx_queues = txq; 2943 } 2944 2945 return 0; 2946 } 2947 EXPORT_SYMBOL(netif_set_real_num_tx_queues); 2948 2949 #ifdef CONFIG_SYSFS 2950 /** 2951 * netif_set_real_num_rx_queues - set actual number of RX queues used 2952 * @dev: Network device 2953 * @rxq: Actual number of RX queues 2954 * 2955 * This must be called either with the rtnl_lock held or before 2956 * registration of the net device. Returns 0 on success, or a 2957 * negative error code. If called before registration, it always 2958 * succeeds. 2959 */ 2960 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) 2961 { 2962 int rc; 2963 2964 if (rxq < 1 || rxq > dev->num_rx_queues) 2965 return -EINVAL; 2966 2967 if (dev->reg_state == NETREG_REGISTERED) { 2968 ASSERT_RTNL(); 2969 2970 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, 2971 rxq); 2972 if (rc) 2973 return rc; 2974 } 2975 2976 dev->real_num_rx_queues = rxq; 2977 return 0; 2978 } 2979 EXPORT_SYMBOL(netif_set_real_num_rx_queues); 2980 #endif 2981 2982 /** 2983 * netif_set_real_num_queues - set actual number of RX and TX queues used 2984 * @dev: Network device 2985 * @txq: Actual number of TX queues 2986 * @rxq: Actual number of RX queues 2987 * 2988 * Set the real number of both TX and RX queues. 2989 * Does nothing if the number of queues is already correct. 2990 */ 2991 int netif_set_real_num_queues(struct net_device *dev, 2992 unsigned int txq, unsigned int rxq) 2993 { 2994 unsigned int old_rxq = dev->real_num_rx_queues; 2995 int err; 2996 2997 if (txq < 1 || txq > dev->num_tx_queues || 2998 rxq < 1 || rxq > dev->num_rx_queues) 2999 return -EINVAL; 3000 3001 /* Start from increases, so the error path only does decreases - 3002 * decreases can't fail. 3003 */ 3004 if (rxq > dev->real_num_rx_queues) { 3005 err = netif_set_real_num_rx_queues(dev, rxq); 3006 if (err) 3007 return err; 3008 } 3009 if (txq > dev->real_num_tx_queues) { 3010 err = netif_set_real_num_tx_queues(dev, txq); 3011 if (err) 3012 goto undo_rx; 3013 } 3014 if (rxq < dev->real_num_rx_queues) 3015 WARN_ON(netif_set_real_num_rx_queues(dev, rxq)); 3016 if (txq < dev->real_num_tx_queues) 3017 WARN_ON(netif_set_real_num_tx_queues(dev, txq)); 3018 3019 return 0; 3020 undo_rx: 3021 WARN_ON(netif_set_real_num_rx_queues(dev, old_rxq)); 3022 return err; 3023 } 3024 EXPORT_SYMBOL(netif_set_real_num_queues); 3025 3026 /** 3027 * netif_get_num_default_rss_queues - default number of RSS queues 3028 * 3029 * This routine should set an upper limit on the number of RSS queues 3030 * used by default by multiqueue devices. 3031 */ 3032 int netif_get_num_default_rss_queues(void) 3033 { 3034 return is_kdump_kernel() ? 3035 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus()); 3036 } 3037 EXPORT_SYMBOL(netif_get_num_default_rss_queues); 3038 3039 static void __netif_reschedule(struct Qdisc *q) 3040 { 3041 struct softnet_data *sd; 3042 unsigned long flags; 3043 3044 local_irq_save(flags); 3045 sd = this_cpu_ptr(&softnet_data); 3046 q->next_sched = NULL; 3047 *sd->output_queue_tailp = q; 3048 sd->output_queue_tailp = &q->next_sched; 3049 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3050 local_irq_restore(flags); 3051 } 3052 3053 void __netif_schedule(struct Qdisc *q) 3054 { 3055 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) 3056 __netif_reschedule(q); 3057 } 3058 EXPORT_SYMBOL(__netif_schedule); 3059 3060 struct dev_kfree_skb_cb { 3061 enum skb_free_reason reason; 3062 }; 3063 3064 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) 3065 { 3066 return (struct dev_kfree_skb_cb *)skb->cb; 3067 } 3068 3069 void netif_schedule_queue(struct netdev_queue *txq) 3070 { 3071 rcu_read_lock(); 3072 if (!netif_xmit_stopped(txq)) { 3073 struct Qdisc *q = rcu_dereference(txq->qdisc); 3074 3075 __netif_schedule(q); 3076 } 3077 rcu_read_unlock(); 3078 } 3079 EXPORT_SYMBOL(netif_schedule_queue); 3080 3081 void netif_tx_wake_queue(struct netdev_queue *dev_queue) 3082 { 3083 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { 3084 struct Qdisc *q; 3085 3086 rcu_read_lock(); 3087 q = rcu_dereference(dev_queue->qdisc); 3088 __netif_schedule(q); 3089 rcu_read_unlock(); 3090 } 3091 } 3092 EXPORT_SYMBOL(netif_tx_wake_queue); 3093 3094 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason) 3095 { 3096 unsigned long flags; 3097 3098 if (unlikely(!skb)) 3099 return; 3100 3101 if (likely(refcount_read(&skb->users) == 1)) { 3102 smp_rmb(); 3103 refcount_set(&skb->users, 0); 3104 } else if (likely(!refcount_dec_and_test(&skb->users))) { 3105 return; 3106 } 3107 get_kfree_skb_cb(skb)->reason = reason; 3108 local_irq_save(flags); 3109 skb->next = __this_cpu_read(softnet_data.completion_queue); 3110 __this_cpu_write(softnet_data.completion_queue, skb); 3111 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3112 local_irq_restore(flags); 3113 } 3114 EXPORT_SYMBOL(__dev_kfree_skb_irq); 3115 3116 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason) 3117 { 3118 if (in_hardirq() || irqs_disabled()) 3119 __dev_kfree_skb_irq(skb, reason); 3120 else 3121 dev_kfree_skb(skb); 3122 } 3123 EXPORT_SYMBOL(__dev_kfree_skb_any); 3124 3125 3126 /** 3127 * netif_device_detach - mark device as removed 3128 * @dev: network device 3129 * 3130 * Mark device as removed from system and therefore no longer available. 3131 */ 3132 void netif_device_detach(struct net_device *dev) 3133 { 3134 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && 3135 netif_running(dev)) { 3136 netif_tx_stop_all_queues(dev); 3137 } 3138 } 3139 EXPORT_SYMBOL(netif_device_detach); 3140 3141 /** 3142 * netif_device_attach - mark device as attached 3143 * @dev: network device 3144 * 3145 * Mark device as attached from system and restart if needed. 3146 */ 3147 void netif_device_attach(struct net_device *dev) 3148 { 3149 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && 3150 netif_running(dev)) { 3151 netif_tx_wake_all_queues(dev); 3152 __netdev_watchdog_up(dev); 3153 } 3154 } 3155 EXPORT_SYMBOL(netif_device_attach); 3156 3157 /* 3158 * Returns a Tx hash based on the given packet descriptor a Tx queues' number 3159 * to be used as a distribution range. 3160 */ 3161 static u16 skb_tx_hash(const struct net_device *dev, 3162 const struct net_device *sb_dev, 3163 struct sk_buff *skb) 3164 { 3165 u32 hash; 3166 u16 qoffset = 0; 3167 u16 qcount = dev->real_num_tx_queues; 3168 3169 if (dev->num_tc) { 3170 u8 tc = netdev_get_prio_tc_map(dev, skb->priority); 3171 3172 qoffset = sb_dev->tc_to_txq[tc].offset; 3173 qcount = sb_dev->tc_to_txq[tc].count; 3174 } 3175 3176 if (skb_rx_queue_recorded(skb)) { 3177 hash = skb_get_rx_queue(skb); 3178 if (hash >= qoffset) 3179 hash -= qoffset; 3180 while (unlikely(hash >= qcount)) 3181 hash -= qcount; 3182 return hash + qoffset; 3183 } 3184 3185 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; 3186 } 3187 3188 static void skb_warn_bad_offload(const struct sk_buff *skb) 3189 { 3190 static const netdev_features_t null_features; 3191 struct net_device *dev = skb->dev; 3192 const char *name = ""; 3193 3194 if (!net_ratelimit()) 3195 return; 3196 3197 if (dev) { 3198 if (dev->dev.parent) 3199 name = dev_driver_string(dev->dev.parent); 3200 else 3201 name = netdev_name(dev); 3202 } 3203 skb_dump(KERN_WARNING, skb, false); 3204 WARN(1, "%s: caps=(%pNF, %pNF)\n", 3205 name, dev ? &dev->features : &null_features, 3206 skb->sk ? &skb->sk->sk_route_caps : &null_features); 3207 } 3208 3209 /* 3210 * Invalidate hardware checksum when packet is to be mangled, and 3211 * complete checksum manually on outgoing path. 3212 */ 3213 int skb_checksum_help(struct sk_buff *skb) 3214 { 3215 __wsum csum; 3216 int ret = 0, offset; 3217 3218 if (skb->ip_summed == CHECKSUM_COMPLETE) 3219 goto out_set_summed; 3220 3221 if (unlikely(skb_is_gso(skb))) { 3222 skb_warn_bad_offload(skb); 3223 return -EINVAL; 3224 } 3225 3226 /* Before computing a checksum, we should make sure no frag could 3227 * be modified by an external entity : checksum could be wrong. 3228 */ 3229 if (skb_has_shared_frag(skb)) { 3230 ret = __skb_linearize(skb); 3231 if (ret) 3232 goto out; 3233 } 3234 3235 offset = skb_checksum_start_offset(skb); 3236 BUG_ON(offset >= skb_headlen(skb)); 3237 csum = skb_checksum(skb, offset, skb->len - offset, 0); 3238 3239 offset += skb->csum_offset; 3240 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); 3241 3242 ret = skb_ensure_writable(skb, offset + sizeof(__sum16)); 3243 if (ret) 3244 goto out; 3245 3246 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0; 3247 out_set_summed: 3248 skb->ip_summed = CHECKSUM_NONE; 3249 out: 3250 return ret; 3251 } 3252 EXPORT_SYMBOL(skb_checksum_help); 3253 3254 int skb_crc32c_csum_help(struct sk_buff *skb) 3255 { 3256 __le32 crc32c_csum; 3257 int ret = 0, offset, start; 3258 3259 if (skb->ip_summed != CHECKSUM_PARTIAL) 3260 goto out; 3261 3262 if (unlikely(skb_is_gso(skb))) 3263 goto out; 3264 3265 /* Before computing a checksum, we should make sure no frag could 3266 * be modified by an external entity : checksum could be wrong. 3267 */ 3268 if (unlikely(skb_has_shared_frag(skb))) { 3269 ret = __skb_linearize(skb); 3270 if (ret) 3271 goto out; 3272 } 3273 start = skb_checksum_start_offset(skb); 3274 offset = start + offsetof(struct sctphdr, checksum); 3275 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { 3276 ret = -EINVAL; 3277 goto out; 3278 } 3279 3280 ret = skb_ensure_writable(skb, offset + sizeof(__le32)); 3281 if (ret) 3282 goto out; 3283 3284 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start, 3285 skb->len - start, ~(__u32)0, 3286 crc32c_csum_stub)); 3287 *(__le32 *)(skb->data + offset) = crc32c_csum; 3288 skb->ip_summed = CHECKSUM_NONE; 3289 skb->csum_not_inet = 0; 3290 out: 3291 return ret; 3292 } 3293 3294 __be16 skb_network_protocol(struct sk_buff *skb, int *depth) 3295 { 3296 __be16 type = skb->protocol; 3297 3298 /* Tunnel gso handlers can set protocol to ethernet. */ 3299 if (type == htons(ETH_P_TEB)) { 3300 struct ethhdr *eth; 3301 3302 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) 3303 return 0; 3304 3305 eth = (struct ethhdr *)skb->data; 3306 type = eth->h_proto; 3307 } 3308 3309 return __vlan_get_protocol(skb, type, depth); 3310 } 3311 3312 /** 3313 * skb_mac_gso_segment - mac layer segmentation handler. 3314 * @skb: buffer to segment 3315 * @features: features for the output path (see dev->features) 3316 */ 3317 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb, 3318 netdev_features_t features) 3319 { 3320 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); 3321 struct packet_offload *ptype; 3322 int vlan_depth = skb->mac_len; 3323 __be16 type = skb_network_protocol(skb, &vlan_depth); 3324 3325 if (unlikely(!type)) 3326 return ERR_PTR(-EINVAL); 3327 3328 __skb_pull(skb, vlan_depth); 3329 3330 rcu_read_lock(); 3331 list_for_each_entry_rcu(ptype, &offload_base, list) { 3332 if (ptype->type == type && ptype->callbacks.gso_segment) { 3333 segs = ptype->callbacks.gso_segment(skb, features); 3334 break; 3335 } 3336 } 3337 rcu_read_unlock(); 3338 3339 __skb_push(skb, skb->data - skb_mac_header(skb)); 3340 3341 return segs; 3342 } 3343 EXPORT_SYMBOL(skb_mac_gso_segment); 3344 3345 3346 /* openvswitch calls this on rx path, so we need a different check. 3347 */ 3348 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path) 3349 { 3350 if (tx_path) 3351 return skb->ip_summed != CHECKSUM_PARTIAL && 3352 skb->ip_summed != CHECKSUM_UNNECESSARY; 3353 3354 return skb->ip_summed == CHECKSUM_NONE; 3355 } 3356 3357 /** 3358 * __skb_gso_segment - Perform segmentation on skb. 3359 * @skb: buffer to segment 3360 * @features: features for the output path (see dev->features) 3361 * @tx_path: whether it is called in TX path 3362 * 3363 * This function segments the given skb and returns a list of segments. 3364 * 3365 * It may return NULL if the skb requires no segmentation. This is 3366 * only possible when GSO is used for verifying header integrity. 3367 * 3368 * Segmentation preserves SKB_GSO_CB_OFFSET bytes of previous skb cb. 3369 */ 3370 struct sk_buff *__skb_gso_segment(struct sk_buff *skb, 3371 netdev_features_t features, bool tx_path) 3372 { 3373 struct sk_buff *segs; 3374 3375 if (unlikely(skb_needs_check(skb, tx_path))) { 3376 int err; 3377 3378 /* We're going to init ->check field in TCP or UDP header */ 3379 err = skb_cow_head(skb, 0); 3380 if (err < 0) 3381 return ERR_PTR(err); 3382 } 3383 3384 /* Only report GSO partial support if it will enable us to 3385 * support segmentation on this frame without needing additional 3386 * work. 3387 */ 3388 if (features & NETIF_F_GSO_PARTIAL) { 3389 netdev_features_t partial_features = NETIF_F_GSO_ROBUST; 3390 struct net_device *dev = skb->dev; 3391 3392 partial_features |= dev->features & dev->gso_partial_features; 3393 if (!skb_gso_ok(skb, features | partial_features)) 3394 features &= ~NETIF_F_GSO_PARTIAL; 3395 } 3396 3397 BUILD_BUG_ON(SKB_GSO_CB_OFFSET + 3398 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb)); 3399 3400 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb); 3401 SKB_GSO_CB(skb)->encap_level = 0; 3402 3403 skb_reset_mac_header(skb); 3404 skb_reset_mac_len(skb); 3405 3406 segs = skb_mac_gso_segment(skb, features); 3407 3408 if (segs != skb && unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs))) 3409 skb_warn_bad_offload(skb); 3410 3411 return segs; 3412 } 3413 EXPORT_SYMBOL(__skb_gso_segment); 3414 3415 /* Take action when hardware reception checksum errors are detected. */ 3416 #ifdef CONFIG_BUG 3417 static void do_netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3418 { 3419 netdev_err(dev, "hw csum failure\n"); 3420 skb_dump(KERN_ERR, skb, true); 3421 dump_stack(); 3422 } 3423 3424 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3425 { 3426 DO_ONCE_LITE(do_netdev_rx_csum_fault, dev, skb); 3427 } 3428 EXPORT_SYMBOL(netdev_rx_csum_fault); 3429 #endif 3430 3431 /* XXX: check that highmem exists at all on the given machine. */ 3432 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) 3433 { 3434 #ifdef CONFIG_HIGHMEM 3435 int i; 3436 3437 if (!(dev->features & NETIF_F_HIGHDMA)) { 3438 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 3439 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 3440 3441 if (PageHighMem(skb_frag_page(frag))) 3442 return 1; 3443 } 3444 } 3445 #endif 3446 return 0; 3447 } 3448 3449 /* If MPLS offload request, verify we are testing hardware MPLS features 3450 * instead of standard features for the netdev. 3451 */ 3452 #if IS_ENABLED(CONFIG_NET_MPLS_GSO) 3453 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3454 netdev_features_t features, 3455 __be16 type) 3456 { 3457 if (eth_p_mpls(type)) 3458 features &= skb->dev->mpls_features; 3459 3460 return features; 3461 } 3462 #else 3463 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3464 netdev_features_t features, 3465 __be16 type) 3466 { 3467 return features; 3468 } 3469 #endif 3470 3471 static netdev_features_t harmonize_features(struct sk_buff *skb, 3472 netdev_features_t features) 3473 { 3474 __be16 type; 3475 3476 type = skb_network_protocol(skb, NULL); 3477 features = net_mpls_features(skb, features, type); 3478 3479 if (skb->ip_summed != CHECKSUM_NONE && 3480 !can_checksum_protocol(features, type)) { 3481 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); 3482 } 3483 if (illegal_highdma(skb->dev, skb)) 3484 features &= ~NETIF_F_SG; 3485 3486 return features; 3487 } 3488 3489 netdev_features_t passthru_features_check(struct sk_buff *skb, 3490 struct net_device *dev, 3491 netdev_features_t features) 3492 { 3493 return features; 3494 } 3495 EXPORT_SYMBOL(passthru_features_check); 3496 3497 static netdev_features_t dflt_features_check(struct sk_buff *skb, 3498 struct net_device *dev, 3499 netdev_features_t features) 3500 { 3501 return vlan_features_check(skb, features); 3502 } 3503 3504 static netdev_features_t gso_features_check(const struct sk_buff *skb, 3505 struct net_device *dev, 3506 netdev_features_t features) 3507 { 3508 u16 gso_segs = skb_shinfo(skb)->gso_segs; 3509 3510 if (gso_segs > dev->gso_max_segs) 3511 return features & ~NETIF_F_GSO_MASK; 3512 3513 if (!skb_shinfo(skb)->gso_type) { 3514 skb_warn_bad_offload(skb); 3515 return features & ~NETIF_F_GSO_MASK; 3516 } 3517 3518 /* Support for GSO partial features requires software 3519 * intervention before we can actually process the packets 3520 * so we need to strip support for any partial features now 3521 * and we can pull them back in after we have partially 3522 * segmented the frame. 3523 */ 3524 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)) 3525 features &= ~dev->gso_partial_features; 3526 3527 /* Make sure to clear the IPv4 ID mangling feature if the 3528 * IPv4 header has the potential to be fragmented. 3529 */ 3530 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { 3531 struct iphdr *iph = skb->encapsulation ? 3532 inner_ip_hdr(skb) : ip_hdr(skb); 3533 3534 if (!(iph->frag_off & htons(IP_DF))) 3535 features &= ~NETIF_F_TSO_MANGLEID; 3536 } 3537 3538 return features; 3539 } 3540 3541 netdev_features_t netif_skb_features(struct sk_buff *skb) 3542 { 3543 struct net_device *dev = skb->dev; 3544 netdev_features_t features = dev->features; 3545 3546 if (skb_is_gso(skb)) 3547 features = gso_features_check(skb, dev, features); 3548 3549 /* If encapsulation offload request, verify we are testing 3550 * hardware encapsulation features instead of standard 3551 * features for the netdev 3552 */ 3553 if (skb->encapsulation) 3554 features &= dev->hw_enc_features; 3555 3556 if (skb_vlan_tagged(skb)) 3557 features = netdev_intersect_features(features, 3558 dev->vlan_features | 3559 NETIF_F_HW_VLAN_CTAG_TX | 3560 NETIF_F_HW_VLAN_STAG_TX); 3561 3562 if (dev->netdev_ops->ndo_features_check) 3563 features &= dev->netdev_ops->ndo_features_check(skb, dev, 3564 features); 3565 else 3566 features &= dflt_features_check(skb, dev, features); 3567 3568 return harmonize_features(skb, features); 3569 } 3570 EXPORT_SYMBOL(netif_skb_features); 3571 3572 static int xmit_one(struct sk_buff *skb, struct net_device *dev, 3573 struct netdev_queue *txq, bool more) 3574 { 3575 unsigned int len; 3576 int rc; 3577 3578 if (dev_nit_active(dev)) 3579 dev_queue_xmit_nit(skb, dev); 3580 3581 len = skb->len; 3582 PRANDOM_ADD_NOISE(skb, dev, txq, len + jiffies); 3583 trace_net_dev_start_xmit(skb, dev); 3584 rc = netdev_start_xmit(skb, dev, txq, more); 3585 trace_net_dev_xmit(skb, rc, dev, len); 3586 3587 return rc; 3588 } 3589 3590 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, 3591 struct netdev_queue *txq, int *ret) 3592 { 3593 struct sk_buff *skb = first; 3594 int rc = NETDEV_TX_OK; 3595 3596 while (skb) { 3597 struct sk_buff *next = skb->next; 3598 3599 skb_mark_not_on_list(skb); 3600 rc = xmit_one(skb, dev, txq, next != NULL); 3601 if (unlikely(!dev_xmit_complete(rc))) { 3602 skb->next = next; 3603 goto out; 3604 } 3605 3606 skb = next; 3607 if (netif_tx_queue_stopped(txq) && skb) { 3608 rc = NETDEV_TX_BUSY; 3609 break; 3610 } 3611 } 3612 3613 out: 3614 *ret = rc; 3615 return skb; 3616 } 3617 3618 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, 3619 netdev_features_t features) 3620 { 3621 if (skb_vlan_tag_present(skb) && 3622 !vlan_hw_offload_capable(features, skb->vlan_proto)) 3623 skb = __vlan_hwaccel_push_inside(skb); 3624 return skb; 3625 } 3626 3627 int skb_csum_hwoffload_help(struct sk_buff *skb, 3628 const netdev_features_t features) 3629 { 3630 if (unlikely(skb_csum_is_sctp(skb))) 3631 return !!(features & NETIF_F_SCTP_CRC) ? 0 : 3632 skb_crc32c_csum_help(skb); 3633 3634 if (features & NETIF_F_HW_CSUM) 3635 return 0; 3636 3637 if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) { 3638 switch (skb->csum_offset) { 3639 case offsetof(struct tcphdr, check): 3640 case offsetof(struct udphdr, check): 3641 return 0; 3642 } 3643 } 3644 3645 return skb_checksum_help(skb); 3646 } 3647 EXPORT_SYMBOL(skb_csum_hwoffload_help); 3648 3649 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again) 3650 { 3651 netdev_features_t features; 3652 3653 features = netif_skb_features(skb); 3654 skb = validate_xmit_vlan(skb, features); 3655 if (unlikely(!skb)) 3656 goto out_null; 3657 3658 skb = sk_validate_xmit_skb(skb, dev); 3659 if (unlikely(!skb)) 3660 goto out_null; 3661 3662 if (netif_needs_gso(skb, features)) { 3663 struct sk_buff *segs; 3664 3665 segs = skb_gso_segment(skb, features); 3666 if (IS_ERR(segs)) { 3667 goto out_kfree_skb; 3668 } else if (segs) { 3669 consume_skb(skb); 3670 skb = segs; 3671 } 3672 } else { 3673 if (skb_needs_linearize(skb, features) && 3674 __skb_linearize(skb)) 3675 goto out_kfree_skb; 3676 3677 /* If packet is not checksummed and device does not 3678 * support checksumming for this protocol, complete 3679 * checksumming here. 3680 */ 3681 if (skb->ip_summed == CHECKSUM_PARTIAL) { 3682 if (skb->encapsulation) 3683 skb_set_inner_transport_header(skb, 3684 skb_checksum_start_offset(skb)); 3685 else 3686 skb_set_transport_header(skb, 3687 skb_checksum_start_offset(skb)); 3688 if (skb_csum_hwoffload_help(skb, features)) 3689 goto out_kfree_skb; 3690 } 3691 } 3692 3693 skb = validate_xmit_xfrm(skb, features, again); 3694 3695 return skb; 3696 3697 out_kfree_skb: 3698 kfree_skb(skb); 3699 out_null: 3700 atomic_long_inc(&dev->tx_dropped); 3701 return NULL; 3702 } 3703 3704 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again) 3705 { 3706 struct sk_buff *next, *head = NULL, *tail; 3707 3708 for (; skb != NULL; skb = next) { 3709 next = skb->next; 3710 skb_mark_not_on_list(skb); 3711 3712 /* in case skb wont be segmented, point to itself */ 3713 skb->prev = skb; 3714 3715 skb = validate_xmit_skb(skb, dev, again); 3716 if (!skb) 3717 continue; 3718 3719 if (!head) 3720 head = skb; 3721 else 3722 tail->next = skb; 3723 /* If skb was segmented, skb->prev points to 3724 * the last segment. If not, it still contains skb. 3725 */ 3726 tail = skb->prev; 3727 } 3728 return head; 3729 } 3730 EXPORT_SYMBOL_GPL(validate_xmit_skb_list); 3731 3732 static void qdisc_pkt_len_init(struct sk_buff *skb) 3733 { 3734 const struct skb_shared_info *shinfo = skb_shinfo(skb); 3735 3736 qdisc_skb_cb(skb)->pkt_len = skb->len; 3737 3738 /* To get more precise estimation of bytes sent on wire, 3739 * we add to pkt_len the headers size of all segments 3740 */ 3741 if (shinfo->gso_size && skb_transport_header_was_set(skb)) { 3742 unsigned int hdr_len; 3743 u16 gso_segs = shinfo->gso_segs; 3744 3745 /* mac layer + network layer */ 3746 hdr_len = skb_transport_header(skb) - skb_mac_header(skb); 3747 3748 /* + transport layer */ 3749 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { 3750 const struct tcphdr *th; 3751 struct tcphdr _tcphdr; 3752 3753 th = skb_header_pointer(skb, skb_transport_offset(skb), 3754 sizeof(_tcphdr), &_tcphdr); 3755 if (likely(th)) 3756 hdr_len += __tcp_hdrlen(th); 3757 } else { 3758 struct udphdr _udphdr; 3759 3760 if (skb_header_pointer(skb, skb_transport_offset(skb), 3761 sizeof(_udphdr), &_udphdr)) 3762 hdr_len += sizeof(struct udphdr); 3763 } 3764 3765 if (shinfo->gso_type & SKB_GSO_DODGY) 3766 gso_segs = DIV_ROUND_UP(skb->len - hdr_len, 3767 shinfo->gso_size); 3768 3769 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; 3770 } 3771 } 3772 3773 static int dev_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *q, 3774 struct sk_buff **to_free, 3775 struct netdev_queue *txq) 3776 { 3777 int rc; 3778 3779 rc = q->enqueue(skb, q, to_free) & NET_XMIT_MASK; 3780 if (rc == NET_XMIT_SUCCESS) 3781 trace_qdisc_enqueue(q, txq, skb); 3782 return rc; 3783 } 3784 3785 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, 3786 struct net_device *dev, 3787 struct netdev_queue *txq) 3788 { 3789 spinlock_t *root_lock = qdisc_lock(q); 3790 struct sk_buff *to_free = NULL; 3791 bool contended; 3792 int rc; 3793 3794 qdisc_calculate_pkt_len(skb, q); 3795 3796 if (q->flags & TCQ_F_NOLOCK) { 3797 if (q->flags & TCQ_F_CAN_BYPASS && nolock_qdisc_is_empty(q) && 3798 qdisc_run_begin(q)) { 3799 /* Retest nolock_qdisc_is_empty() within the protection 3800 * of q->seqlock to protect from racing with requeuing. 3801 */ 3802 if (unlikely(!nolock_qdisc_is_empty(q))) { 3803 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3804 __qdisc_run(q); 3805 qdisc_run_end(q); 3806 3807 goto no_lock_out; 3808 } 3809 3810 qdisc_bstats_cpu_update(q, skb); 3811 if (sch_direct_xmit(skb, q, dev, txq, NULL, true) && 3812 !nolock_qdisc_is_empty(q)) 3813 __qdisc_run(q); 3814 3815 qdisc_run_end(q); 3816 return NET_XMIT_SUCCESS; 3817 } 3818 3819 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3820 qdisc_run(q); 3821 3822 no_lock_out: 3823 if (unlikely(to_free)) 3824 kfree_skb_list(to_free); 3825 return rc; 3826 } 3827 3828 /* 3829 * Heuristic to force contended enqueues to serialize on a 3830 * separate lock before trying to get qdisc main lock. 3831 * This permits qdisc->running owner to get the lock more 3832 * often and dequeue packets faster. 3833 */ 3834 contended = qdisc_is_running(q); 3835 if (unlikely(contended)) 3836 spin_lock(&q->busylock); 3837 3838 spin_lock(root_lock); 3839 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { 3840 __qdisc_drop(skb, &to_free); 3841 rc = NET_XMIT_DROP; 3842 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && 3843 qdisc_run_begin(q)) { 3844 /* 3845 * This is a work-conserving queue; there are no old skbs 3846 * waiting to be sent out; and the qdisc is not running - 3847 * xmit the skb directly. 3848 */ 3849 3850 qdisc_bstats_update(q, skb); 3851 3852 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { 3853 if (unlikely(contended)) { 3854 spin_unlock(&q->busylock); 3855 contended = false; 3856 } 3857 __qdisc_run(q); 3858 } 3859 3860 qdisc_run_end(q); 3861 rc = NET_XMIT_SUCCESS; 3862 } else { 3863 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3864 if (qdisc_run_begin(q)) { 3865 if (unlikely(contended)) { 3866 spin_unlock(&q->busylock); 3867 contended = false; 3868 } 3869 __qdisc_run(q); 3870 qdisc_run_end(q); 3871 } 3872 } 3873 spin_unlock(root_lock); 3874 if (unlikely(to_free)) 3875 kfree_skb_list(to_free); 3876 if (unlikely(contended)) 3877 spin_unlock(&q->busylock); 3878 return rc; 3879 } 3880 3881 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) 3882 static void skb_update_prio(struct sk_buff *skb) 3883 { 3884 const struct netprio_map *map; 3885 const struct sock *sk; 3886 unsigned int prioidx; 3887 3888 if (skb->priority) 3889 return; 3890 map = rcu_dereference_bh(skb->dev->priomap); 3891 if (!map) 3892 return; 3893 sk = skb_to_full_sk(skb); 3894 if (!sk) 3895 return; 3896 3897 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data); 3898 3899 if (prioidx < map->priomap_len) 3900 skb->priority = map->priomap[prioidx]; 3901 } 3902 #else 3903 #define skb_update_prio(skb) 3904 #endif 3905 3906 /** 3907 * dev_loopback_xmit - loop back @skb 3908 * @net: network namespace this loopback is happening in 3909 * @sk: sk needed to be a netfilter okfn 3910 * @skb: buffer to transmit 3911 */ 3912 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) 3913 { 3914 skb_reset_mac_header(skb); 3915 __skb_pull(skb, skb_network_offset(skb)); 3916 skb->pkt_type = PACKET_LOOPBACK; 3917 skb->ip_summed = CHECKSUM_UNNECESSARY; 3918 WARN_ON(!skb_dst(skb)); 3919 skb_dst_force(skb); 3920 netif_rx_ni(skb); 3921 return 0; 3922 } 3923 EXPORT_SYMBOL(dev_loopback_xmit); 3924 3925 #ifdef CONFIG_NET_EGRESS 3926 static struct sk_buff * 3927 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 3928 { 3929 #ifdef CONFIG_NET_CLS_ACT 3930 struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress); 3931 struct tcf_result cl_res; 3932 3933 if (!miniq) 3934 return skb; 3935 3936 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */ 3937 qdisc_skb_cb(skb)->mru = 0; 3938 qdisc_skb_cb(skb)->post_ct = false; 3939 mini_qdisc_bstats_cpu_update(miniq, skb); 3940 3941 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) { 3942 case TC_ACT_OK: 3943 case TC_ACT_RECLASSIFY: 3944 skb->tc_index = TC_H_MIN(cl_res.classid); 3945 break; 3946 case TC_ACT_SHOT: 3947 mini_qdisc_qstats_cpu_drop(miniq); 3948 *ret = NET_XMIT_DROP; 3949 kfree_skb(skb); 3950 return NULL; 3951 case TC_ACT_STOLEN: 3952 case TC_ACT_QUEUED: 3953 case TC_ACT_TRAP: 3954 *ret = NET_XMIT_SUCCESS; 3955 consume_skb(skb); 3956 return NULL; 3957 case TC_ACT_REDIRECT: 3958 /* No need to push/pop skb's mac_header here on egress! */ 3959 skb_do_redirect(skb); 3960 *ret = NET_XMIT_SUCCESS; 3961 return NULL; 3962 default: 3963 break; 3964 } 3965 #endif /* CONFIG_NET_CLS_ACT */ 3966 3967 return skb; 3968 } 3969 #endif /* CONFIG_NET_EGRESS */ 3970 3971 #ifdef CONFIG_XPS 3972 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb, 3973 struct xps_dev_maps *dev_maps, unsigned int tci) 3974 { 3975 int tc = netdev_get_prio_tc_map(dev, skb->priority); 3976 struct xps_map *map; 3977 int queue_index = -1; 3978 3979 if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids) 3980 return queue_index; 3981 3982 tci *= dev_maps->num_tc; 3983 tci += tc; 3984 3985 map = rcu_dereference(dev_maps->attr_map[tci]); 3986 if (map) { 3987 if (map->len == 1) 3988 queue_index = map->queues[0]; 3989 else 3990 queue_index = map->queues[reciprocal_scale( 3991 skb_get_hash(skb), map->len)]; 3992 if (unlikely(queue_index >= dev->real_num_tx_queues)) 3993 queue_index = -1; 3994 } 3995 return queue_index; 3996 } 3997 #endif 3998 3999 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev, 4000 struct sk_buff *skb) 4001 { 4002 #ifdef CONFIG_XPS 4003 struct xps_dev_maps *dev_maps; 4004 struct sock *sk = skb->sk; 4005 int queue_index = -1; 4006 4007 if (!static_key_false(&xps_needed)) 4008 return -1; 4009 4010 rcu_read_lock(); 4011 if (!static_key_false(&xps_rxqs_needed)) 4012 goto get_cpus_map; 4013 4014 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]); 4015 if (dev_maps) { 4016 int tci = sk_rx_queue_get(sk); 4017 4018 if (tci >= 0) 4019 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4020 tci); 4021 } 4022 4023 get_cpus_map: 4024 if (queue_index < 0) { 4025 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]); 4026 if (dev_maps) { 4027 unsigned int tci = skb->sender_cpu - 1; 4028 4029 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4030 tci); 4031 } 4032 } 4033 rcu_read_unlock(); 4034 4035 return queue_index; 4036 #else 4037 return -1; 4038 #endif 4039 } 4040 4041 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb, 4042 struct net_device *sb_dev) 4043 { 4044 return 0; 4045 } 4046 EXPORT_SYMBOL(dev_pick_tx_zero); 4047 4048 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb, 4049 struct net_device *sb_dev) 4050 { 4051 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues; 4052 } 4053 EXPORT_SYMBOL(dev_pick_tx_cpu_id); 4054 4055 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb, 4056 struct net_device *sb_dev) 4057 { 4058 struct sock *sk = skb->sk; 4059 int queue_index = sk_tx_queue_get(sk); 4060 4061 sb_dev = sb_dev ? : dev; 4062 4063 if (queue_index < 0 || skb->ooo_okay || 4064 queue_index >= dev->real_num_tx_queues) { 4065 int new_index = get_xps_queue(dev, sb_dev, skb); 4066 4067 if (new_index < 0) 4068 new_index = skb_tx_hash(dev, sb_dev, skb); 4069 4070 if (queue_index != new_index && sk && 4071 sk_fullsock(sk) && 4072 rcu_access_pointer(sk->sk_dst_cache)) 4073 sk_tx_queue_set(sk, new_index); 4074 4075 queue_index = new_index; 4076 } 4077 4078 return queue_index; 4079 } 4080 EXPORT_SYMBOL(netdev_pick_tx); 4081 4082 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev, 4083 struct sk_buff *skb, 4084 struct net_device *sb_dev) 4085 { 4086 int queue_index = 0; 4087 4088 #ifdef CONFIG_XPS 4089 u32 sender_cpu = skb->sender_cpu - 1; 4090 4091 if (sender_cpu >= (u32)NR_CPUS) 4092 skb->sender_cpu = raw_smp_processor_id() + 1; 4093 #endif 4094 4095 if (dev->real_num_tx_queues != 1) { 4096 const struct net_device_ops *ops = dev->netdev_ops; 4097 4098 if (ops->ndo_select_queue) 4099 queue_index = ops->ndo_select_queue(dev, skb, sb_dev); 4100 else 4101 queue_index = netdev_pick_tx(dev, skb, sb_dev); 4102 4103 queue_index = netdev_cap_txqueue(dev, queue_index); 4104 } 4105 4106 skb_set_queue_mapping(skb, queue_index); 4107 return netdev_get_tx_queue(dev, queue_index); 4108 } 4109 4110 /** 4111 * __dev_queue_xmit - transmit a buffer 4112 * @skb: buffer to transmit 4113 * @sb_dev: suboordinate device used for L2 forwarding offload 4114 * 4115 * Queue a buffer for transmission to a network device. The caller must 4116 * have set the device and priority and built the buffer before calling 4117 * this function. The function can be called from an interrupt. 4118 * 4119 * A negative errno code is returned on a failure. A success does not 4120 * guarantee the frame will be transmitted as it may be dropped due 4121 * to congestion or traffic shaping. 4122 * 4123 * ----------------------------------------------------------------------------------- 4124 * I notice this method can also return errors from the queue disciplines, 4125 * including NET_XMIT_DROP, which is a positive value. So, errors can also 4126 * be positive. 4127 * 4128 * Regardless of the return value, the skb is consumed, so it is currently 4129 * difficult to retry a send to this method. (You can bump the ref count 4130 * before sending to hold a reference for retry if you are careful.) 4131 * 4132 * When calling this method, interrupts MUST be enabled. This is because 4133 * the BH enable code must have IRQs enabled so that it will not deadlock. 4134 * --BLG 4135 */ 4136 static int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev) 4137 { 4138 struct net_device *dev = skb->dev; 4139 struct netdev_queue *txq; 4140 struct Qdisc *q; 4141 int rc = -ENOMEM; 4142 bool again = false; 4143 4144 skb_reset_mac_header(skb); 4145 4146 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) 4147 __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED); 4148 4149 /* Disable soft irqs for various locks below. Also 4150 * stops preemption for RCU. 4151 */ 4152 rcu_read_lock_bh(); 4153 4154 skb_update_prio(skb); 4155 4156 qdisc_pkt_len_init(skb); 4157 #ifdef CONFIG_NET_CLS_ACT 4158 skb->tc_at_ingress = 0; 4159 #endif 4160 #ifdef CONFIG_NET_EGRESS 4161 if (static_branch_unlikely(&egress_needed_key)) { 4162 if (nf_hook_egress_active()) { 4163 skb = nf_hook_egress(skb, &rc, dev); 4164 if (!skb) 4165 goto out; 4166 } 4167 nf_skip_egress(skb, true); 4168 skb = sch_handle_egress(skb, &rc, dev); 4169 if (!skb) 4170 goto out; 4171 nf_skip_egress(skb, false); 4172 } 4173 #endif 4174 /* If device/qdisc don't need skb->dst, release it right now while 4175 * its hot in this cpu cache. 4176 */ 4177 if (dev->priv_flags & IFF_XMIT_DST_RELEASE) 4178 skb_dst_drop(skb); 4179 else 4180 skb_dst_force(skb); 4181 4182 txq = netdev_core_pick_tx(dev, skb, sb_dev); 4183 q = rcu_dereference_bh(txq->qdisc); 4184 4185 trace_net_dev_queue(skb); 4186 if (q->enqueue) { 4187 rc = __dev_xmit_skb(skb, q, dev, txq); 4188 goto out; 4189 } 4190 4191 /* The device has no queue. Common case for software devices: 4192 * loopback, all the sorts of tunnels... 4193 4194 * Really, it is unlikely that netif_tx_lock protection is necessary 4195 * here. (f.e. loopback and IP tunnels are clean ignoring statistics 4196 * counters.) 4197 * However, it is possible, that they rely on protection 4198 * made by us here. 4199 4200 * Check this and shot the lock. It is not prone from deadlocks. 4201 *Either shot noqueue qdisc, it is even simpler 8) 4202 */ 4203 if (dev->flags & IFF_UP) { 4204 int cpu = smp_processor_id(); /* ok because BHs are off */ 4205 4206 if (txq->xmit_lock_owner != cpu) { 4207 if (dev_xmit_recursion()) 4208 goto recursion_alert; 4209 4210 skb = validate_xmit_skb(skb, dev, &again); 4211 if (!skb) 4212 goto out; 4213 4214 PRANDOM_ADD_NOISE(skb, dev, txq, jiffies); 4215 HARD_TX_LOCK(dev, txq, cpu); 4216 4217 if (!netif_xmit_stopped(txq)) { 4218 dev_xmit_recursion_inc(); 4219 skb = dev_hard_start_xmit(skb, dev, txq, &rc); 4220 dev_xmit_recursion_dec(); 4221 if (dev_xmit_complete(rc)) { 4222 HARD_TX_UNLOCK(dev, txq); 4223 goto out; 4224 } 4225 } 4226 HARD_TX_UNLOCK(dev, txq); 4227 net_crit_ratelimited("Virtual device %s asks to queue packet!\n", 4228 dev->name); 4229 } else { 4230 /* Recursion is detected! It is possible, 4231 * unfortunately 4232 */ 4233 recursion_alert: 4234 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", 4235 dev->name); 4236 } 4237 } 4238 4239 rc = -ENETDOWN; 4240 rcu_read_unlock_bh(); 4241 4242 atomic_long_inc(&dev->tx_dropped); 4243 kfree_skb_list(skb); 4244 return rc; 4245 out: 4246 rcu_read_unlock_bh(); 4247 return rc; 4248 } 4249 4250 int dev_queue_xmit(struct sk_buff *skb) 4251 { 4252 return __dev_queue_xmit(skb, NULL); 4253 } 4254 EXPORT_SYMBOL(dev_queue_xmit); 4255 4256 int dev_queue_xmit_accel(struct sk_buff *skb, struct net_device *sb_dev) 4257 { 4258 return __dev_queue_xmit(skb, sb_dev); 4259 } 4260 EXPORT_SYMBOL(dev_queue_xmit_accel); 4261 4262 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id) 4263 { 4264 struct net_device *dev = skb->dev; 4265 struct sk_buff *orig_skb = skb; 4266 struct netdev_queue *txq; 4267 int ret = NETDEV_TX_BUSY; 4268 bool again = false; 4269 4270 if (unlikely(!netif_running(dev) || 4271 !netif_carrier_ok(dev))) 4272 goto drop; 4273 4274 skb = validate_xmit_skb_list(skb, dev, &again); 4275 if (skb != orig_skb) 4276 goto drop; 4277 4278 skb_set_queue_mapping(skb, queue_id); 4279 txq = skb_get_tx_queue(dev, skb); 4280 PRANDOM_ADD_NOISE(skb, dev, txq, jiffies); 4281 4282 local_bh_disable(); 4283 4284 dev_xmit_recursion_inc(); 4285 HARD_TX_LOCK(dev, txq, smp_processor_id()); 4286 if (!netif_xmit_frozen_or_drv_stopped(txq)) 4287 ret = netdev_start_xmit(skb, dev, txq, false); 4288 HARD_TX_UNLOCK(dev, txq); 4289 dev_xmit_recursion_dec(); 4290 4291 local_bh_enable(); 4292 return ret; 4293 drop: 4294 atomic_long_inc(&dev->tx_dropped); 4295 kfree_skb_list(skb); 4296 return NET_XMIT_DROP; 4297 } 4298 EXPORT_SYMBOL(__dev_direct_xmit); 4299 4300 /************************************************************************* 4301 * Receiver routines 4302 *************************************************************************/ 4303 4304 int netdev_max_backlog __read_mostly = 1000; 4305 EXPORT_SYMBOL(netdev_max_backlog); 4306 4307 int netdev_tstamp_prequeue __read_mostly = 1; 4308 int netdev_budget __read_mostly = 300; 4309 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */ 4310 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ; 4311 int weight_p __read_mostly = 64; /* old backlog weight */ 4312 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ 4313 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ 4314 int dev_rx_weight __read_mostly = 64; 4315 int dev_tx_weight __read_mostly = 64; 4316 /* Maximum number of GRO_NORMAL skbs to batch up for list-RX */ 4317 int gro_normal_batch __read_mostly = 8; 4318 4319 /* Called with irq disabled */ 4320 static inline void ____napi_schedule(struct softnet_data *sd, 4321 struct napi_struct *napi) 4322 { 4323 struct task_struct *thread; 4324 4325 if (test_bit(NAPI_STATE_THREADED, &napi->state)) { 4326 /* Paired with smp_mb__before_atomic() in 4327 * napi_enable()/dev_set_threaded(). 4328 * Use READ_ONCE() to guarantee a complete 4329 * read on napi->thread. Only call 4330 * wake_up_process() when it's not NULL. 4331 */ 4332 thread = READ_ONCE(napi->thread); 4333 if (thread) { 4334 /* Avoid doing set_bit() if the thread is in 4335 * INTERRUPTIBLE state, cause napi_thread_wait() 4336 * makes sure to proceed with napi polling 4337 * if the thread is explicitly woken from here. 4338 */ 4339 if (READ_ONCE(thread->__state) != TASK_INTERRUPTIBLE) 4340 set_bit(NAPI_STATE_SCHED_THREADED, &napi->state); 4341 wake_up_process(thread); 4342 return; 4343 } 4344 } 4345 4346 list_add_tail(&napi->poll_list, &sd->poll_list); 4347 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4348 } 4349 4350 #ifdef CONFIG_RPS 4351 4352 /* One global table that all flow-based protocols share. */ 4353 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; 4354 EXPORT_SYMBOL(rps_sock_flow_table); 4355 u32 rps_cpu_mask __read_mostly; 4356 EXPORT_SYMBOL(rps_cpu_mask); 4357 4358 struct static_key_false rps_needed __read_mostly; 4359 EXPORT_SYMBOL(rps_needed); 4360 struct static_key_false rfs_needed __read_mostly; 4361 EXPORT_SYMBOL(rfs_needed); 4362 4363 static struct rps_dev_flow * 4364 set_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4365 struct rps_dev_flow *rflow, u16 next_cpu) 4366 { 4367 if (next_cpu < nr_cpu_ids) { 4368 #ifdef CONFIG_RFS_ACCEL 4369 struct netdev_rx_queue *rxqueue; 4370 struct rps_dev_flow_table *flow_table; 4371 struct rps_dev_flow *old_rflow; 4372 u32 flow_id; 4373 u16 rxq_index; 4374 int rc; 4375 4376 /* Should we steer this flow to a different hardware queue? */ 4377 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || 4378 !(dev->features & NETIF_F_NTUPLE)) 4379 goto out; 4380 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); 4381 if (rxq_index == skb_get_rx_queue(skb)) 4382 goto out; 4383 4384 rxqueue = dev->_rx + rxq_index; 4385 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4386 if (!flow_table) 4387 goto out; 4388 flow_id = skb_get_hash(skb) & flow_table->mask; 4389 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, 4390 rxq_index, flow_id); 4391 if (rc < 0) 4392 goto out; 4393 old_rflow = rflow; 4394 rflow = &flow_table->flows[flow_id]; 4395 rflow->filter = rc; 4396 if (old_rflow->filter == rflow->filter) 4397 old_rflow->filter = RPS_NO_FILTER; 4398 out: 4399 #endif 4400 rflow->last_qtail = 4401 per_cpu(softnet_data, next_cpu).input_queue_head; 4402 } 4403 4404 rflow->cpu = next_cpu; 4405 return rflow; 4406 } 4407 4408 /* 4409 * get_rps_cpu is called from netif_receive_skb and returns the target 4410 * CPU from the RPS map of the receiving queue for a given skb. 4411 * rcu_read_lock must be held on entry. 4412 */ 4413 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4414 struct rps_dev_flow **rflowp) 4415 { 4416 const struct rps_sock_flow_table *sock_flow_table; 4417 struct netdev_rx_queue *rxqueue = dev->_rx; 4418 struct rps_dev_flow_table *flow_table; 4419 struct rps_map *map; 4420 int cpu = -1; 4421 u32 tcpu; 4422 u32 hash; 4423 4424 if (skb_rx_queue_recorded(skb)) { 4425 u16 index = skb_get_rx_queue(skb); 4426 4427 if (unlikely(index >= dev->real_num_rx_queues)) { 4428 WARN_ONCE(dev->real_num_rx_queues > 1, 4429 "%s received packet on queue %u, but number " 4430 "of RX queues is %u\n", 4431 dev->name, index, dev->real_num_rx_queues); 4432 goto done; 4433 } 4434 rxqueue += index; 4435 } 4436 4437 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ 4438 4439 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4440 map = rcu_dereference(rxqueue->rps_map); 4441 if (!flow_table && !map) 4442 goto done; 4443 4444 skb_reset_network_header(skb); 4445 hash = skb_get_hash(skb); 4446 if (!hash) 4447 goto done; 4448 4449 sock_flow_table = rcu_dereference(rps_sock_flow_table); 4450 if (flow_table && sock_flow_table) { 4451 struct rps_dev_flow *rflow; 4452 u32 next_cpu; 4453 u32 ident; 4454 4455 /* First check into global flow table if there is a match */ 4456 ident = sock_flow_table->ents[hash & sock_flow_table->mask]; 4457 if ((ident ^ hash) & ~rps_cpu_mask) 4458 goto try_rps; 4459 4460 next_cpu = ident & rps_cpu_mask; 4461 4462 /* OK, now we know there is a match, 4463 * we can look at the local (per receive queue) flow table 4464 */ 4465 rflow = &flow_table->flows[hash & flow_table->mask]; 4466 tcpu = rflow->cpu; 4467 4468 /* 4469 * If the desired CPU (where last recvmsg was done) is 4470 * different from current CPU (one in the rx-queue flow 4471 * table entry), switch if one of the following holds: 4472 * - Current CPU is unset (>= nr_cpu_ids). 4473 * - Current CPU is offline. 4474 * - The current CPU's queue tail has advanced beyond the 4475 * last packet that was enqueued using this table entry. 4476 * This guarantees that all previous packets for the flow 4477 * have been dequeued, thus preserving in order delivery. 4478 */ 4479 if (unlikely(tcpu != next_cpu) && 4480 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || 4481 ((int)(per_cpu(softnet_data, tcpu).input_queue_head - 4482 rflow->last_qtail)) >= 0)) { 4483 tcpu = next_cpu; 4484 rflow = set_rps_cpu(dev, skb, rflow, next_cpu); 4485 } 4486 4487 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { 4488 *rflowp = rflow; 4489 cpu = tcpu; 4490 goto done; 4491 } 4492 } 4493 4494 try_rps: 4495 4496 if (map) { 4497 tcpu = map->cpus[reciprocal_scale(hash, map->len)]; 4498 if (cpu_online(tcpu)) { 4499 cpu = tcpu; 4500 goto done; 4501 } 4502 } 4503 4504 done: 4505 return cpu; 4506 } 4507 4508 #ifdef CONFIG_RFS_ACCEL 4509 4510 /** 4511 * rps_may_expire_flow - check whether an RFS hardware filter may be removed 4512 * @dev: Device on which the filter was set 4513 * @rxq_index: RX queue index 4514 * @flow_id: Flow ID passed to ndo_rx_flow_steer() 4515 * @filter_id: Filter ID returned by ndo_rx_flow_steer() 4516 * 4517 * Drivers that implement ndo_rx_flow_steer() should periodically call 4518 * this function for each installed filter and remove the filters for 4519 * which it returns %true. 4520 */ 4521 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, 4522 u32 flow_id, u16 filter_id) 4523 { 4524 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; 4525 struct rps_dev_flow_table *flow_table; 4526 struct rps_dev_flow *rflow; 4527 bool expire = true; 4528 unsigned int cpu; 4529 4530 rcu_read_lock(); 4531 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4532 if (flow_table && flow_id <= flow_table->mask) { 4533 rflow = &flow_table->flows[flow_id]; 4534 cpu = READ_ONCE(rflow->cpu); 4535 if (rflow->filter == filter_id && cpu < nr_cpu_ids && 4536 ((int)(per_cpu(softnet_data, cpu).input_queue_head - 4537 rflow->last_qtail) < 4538 (int)(10 * flow_table->mask))) 4539 expire = false; 4540 } 4541 rcu_read_unlock(); 4542 return expire; 4543 } 4544 EXPORT_SYMBOL(rps_may_expire_flow); 4545 4546 #endif /* CONFIG_RFS_ACCEL */ 4547 4548 /* Called from hardirq (IPI) context */ 4549 static void rps_trigger_softirq(void *data) 4550 { 4551 struct softnet_data *sd = data; 4552 4553 ____napi_schedule(sd, &sd->backlog); 4554 sd->received_rps++; 4555 } 4556 4557 #endif /* CONFIG_RPS */ 4558 4559 /* 4560 * Check if this softnet_data structure is another cpu one 4561 * If yes, queue it to our IPI list and return 1 4562 * If no, return 0 4563 */ 4564 static int rps_ipi_queued(struct softnet_data *sd) 4565 { 4566 #ifdef CONFIG_RPS 4567 struct softnet_data *mysd = this_cpu_ptr(&softnet_data); 4568 4569 if (sd != mysd) { 4570 sd->rps_ipi_next = mysd->rps_ipi_list; 4571 mysd->rps_ipi_list = sd; 4572 4573 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4574 return 1; 4575 } 4576 #endif /* CONFIG_RPS */ 4577 return 0; 4578 } 4579 4580 #ifdef CONFIG_NET_FLOW_LIMIT 4581 int netdev_flow_limit_table_len __read_mostly = (1 << 12); 4582 #endif 4583 4584 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) 4585 { 4586 #ifdef CONFIG_NET_FLOW_LIMIT 4587 struct sd_flow_limit *fl; 4588 struct softnet_data *sd; 4589 unsigned int old_flow, new_flow; 4590 4591 if (qlen < (netdev_max_backlog >> 1)) 4592 return false; 4593 4594 sd = this_cpu_ptr(&softnet_data); 4595 4596 rcu_read_lock(); 4597 fl = rcu_dereference(sd->flow_limit); 4598 if (fl) { 4599 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); 4600 old_flow = fl->history[fl->history_head]; 4601 fl->history[fl->history_head] = new_flow; 4602 4603 fl->history_head++; 4604 fl->history_head &= FLOW_LIMIT_HISTORY - 1; 4605 4606 if (likely(fl->buckets[old_flow])) 4607 fl->buckets[old_flow]--; 4608 4609 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { 4610 fl->count++; 4611 rcu_read_unlock(); 4612 return true; 4613 } 4614 } 4615 rcu_read_unlock(); 4616 #endif 4617 return false; 4618 } 4619 4620 /* 4621 * enqueue_to_backlog is called to queue an skb to a per CPU backlog 4622 * queue (may be a remote CPU queue). 4623 */ 4624 static int enqueue_to_backlog(struct sk_buff *skb, int cpu, 4625 unsigned int *qtail) 4626 { 4627 struct softnet_data *sd; 4628 unsigned long flags; 4629 unsigned int qlen; 4630 4631 sd = &per_cpu(softnet_data, cpu); 4632 4633 local_irq_save(flags); 4634 4635 rps_lock(sd); 4636 if (!netif_running(skb->dev)) 4637 goto drop; 4638 qlen = skb_queue_len(&sd->input_pkt_queue); 4639 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) { 4640 if (qlen) { 4641 enqueue: 4642 __skb_queue_tail(&sd->input_pkt_queue, skb); 4643 input_queue_tail_incr_save(sd, qtail); 4644 rps_unlock(sd); 4645 local_irq_restore(flags); 4646 return NET_RX_SUCCESS; 4647 } 4648 4649 /* Schedule NAPI for backlog device 4650 * We can use non atomic operation since we own the queue lock 4651 */ 4652 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) { 4653 if (!rps_ipi_queued(sd)) 4654 ____napi_schedule(sd, &sd->backlog); 4655 } 4656 goto enqueue; 4657 } 4658 4659 drop: 4660 sd->dropped++; 4661 rps_unlock(sd); 4662 4663 local_irq_restore(flags); 4664 4665 atomic_long_inc(&skb->dev->rx_dropped); 4666 kfree_skb(skb); 4667 return NET_RX_DROP; 4668 } 4669 4670 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb) 4671 { 4672 struct net_device *dev = skb->dev; 4673 struct netdev_rx_queue *rxqueue; 4674 4675 rxqueue = dev->_rx; 4676 4677 if (skb_rx_queue_recorded(skb)) { 4678 u16 index = skb_get_rx_queue(skb); 4679 4680 if (unlikely(index >= dev->real_num_rx_queues)) { 4681 WARN_ONCE(dev->real_num_rx_queues > 1, 4682 "%s received packet on queue %u, but number " 4683 "of RX queues is %u\n", 4684 dev->name, index, dev->real_num_rx_queues); 4685 4686 return rxqueue; /* Return first rxqueue */ 4687 } 4688 rxqueue += index; 4689 } 4690 return rxqueue; 4691 } 4692 4693 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp, 4694 struct bpf_prog *xdp_prog) 4695 { 4696 void *orig_data, *orig_data_end, *hard_start; 4697 struct netdev_rx_queue *rxqueue; 4698 bool orig_bcast, orig_host; 4699 u32 mac_len, frame_sz; 4700 __be16 orig_eth_type; 4701 struct ethhdr *eth; 4702 u32 metalen, act; 4703 int off; 4704 4705 /* The XDP program wants to see the packet starting at the MAC 4706 * header. 4707 */ 4708 mac_len = skb->data - skb_mac_header(skb); 4709 hard_start = skb->data - skb_headroom(skb); 4710 4711 /* SKB "head" area always have tailroom for skb_shared_info */ 4712 frame_sz = (void *)skb_end_pointer(skb) - hard_start; 4713 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 4714 4715 rxqueue = netif_get_rxqueue(skb); 4716 xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq); 4717 xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len, 4718 skb_headlen(skb) + mac_len, true); 4719 4720 orig_data_end = xdp->data_end; 4721 orig_data = xdp->data; 4722 eth = (struct ethhdr *)xdp->data; 4723 orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr); 4724 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest); 4725 orig_eth_type = eth->h_proto; 4726 4727 act = bpf_prog_run_xdp(xdp_prog, xdp); 4728 4729 /* check if bpf_xdp_adjust_head was used */ 4730 off = xdp->data - orig_data; 4731 if (off) { 4732 if (off > 0) 4733 __skb_pull(skb, off); 4734 else if (off < 0) 4735 __skb_push(skb, -off); 4736 4737 skb->mac_header += off; 4738 skb_reset_network_header(skb); 4739 } 4740 4741 /* check if bpf_xdp_adjust_tail was used */ 4742 off = xdp->data_end - orig_data_end; 4743 if (off != 0) { 4744 skb_set_tail_pointer(skb, xdp->data_end - xdp->data); 4745 skb->len += off; /* positive on grow, negative on shrink */ 4746 } 4747 4748 /* check if XDP changed eth hdr such SKB needs update */ 4749 eth = (struct ethhdr *)xdp->data; 4750 if ((orig_eth_type != eth->h_proto) || 4751 (orig_host != ether_addr_equal_64bits(eth->h_dest, 4752 skb->dev->dev_addr)) || 4753 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) { 4754 __skb_push(skb, ETH_HLEN); 4755 skb->pkt_type = PACKET_HOST; 4756 skb->protocol = eth_type_trans(skb, skb->dev); 4757 } 4758 4759 /* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull 4760 * before calling us again on redirect path. We do not call do_redirect 4761 * as we leave that up to the caller. 4762 * 4763 * Caller is responsible for managing lifetime of skb (i.e. calling 4764 * kfree_skb in response to actions it cannot handle/XDP_DROP). 4765 */ 4766 switch (act) { 4767 case XDP_REDIRECT: 4768 case XDP_TX: 4769 __skb_push(skb, mac_len); 4770 break; 4771 case XDP_PASS: 4772 metalen = xdp->data - xdp->data_meta; 4773 if (metalen) 4774 skb_metadata_set(skb, metalen); 4775 break; 4776 } 4777 4778 return act; 4779 } 4780 4781 static u32 netif_receive_generic_xdp(struct sk_buff *skb, 4782 struct xdp_buff *xdp, 4783 struct bpf_prog *xdp_prog) 4784 { 4785 u32 act = XDP_DROP; 4786 4787 /* Reinjected packets coming from act_mirred or similar should 4788 * not get XDP generic processing. 4789 */ 4790 if (skb_is_redirected(skb)) 4791 return XDP_PASS; 4792 4793 /* XDP packets must be linear and must have sufficient headroom 4794 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also 4795 * native XDP provides, thus we need to do it here as well. 4796 */ 4797 if (skb_cloned(skb) || skb_is_nonlinear(skb) || 4798 skb_headroom(skb) < XDP_PACKET_HEADROOM) { 4799 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb); 4800 int troom = skb->tail + skb->data_len - skb->end; 4801 4802 /* In case we have to go down the path and also linearize, 4803 * then lets do the pskb_expand_head() work just once here. 4804 */ 4805 if (pskb_expand_head(skb, 4806 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0, 4807 troom > 0 ? troom + 128 : 0, GFP_ATOMIC)) 4808 goto do_drop; 4809 if (skb_linearize(skb)) 4810 goto do_drop; 4811 } 4812 4813 act = bpf_prog_run_generic_xdp(skb, xdp, xdp_prog); 4814 switch (act) { 4815 case XDP_REDIRECT: 4816 case XDP_TX: 4817 case XDP_PASS: 4818 break; 4819 default: 4820 bpf_warn_invalid_xdp_action(act); 4821 fallthrough; 4822 case XDP_ABORTED: 4823 trace_xdp_exception(skb->dev, xdp_prog, act); 4824 fallthrough; 4825 case XDP_DROP: 4826 do_drop: 4827 kfree_skb(skb); 4828 break; 4829 } 4830 4831 return act; 4832 } 4833 4834 /* When doing generic XDP we have to bypass the qdisc layer and the 4835 * network taps in order to match in-driver-XDP behavior. 4836 */ 4837 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog) 4838 { 4839 struct net_device *dev = skb->dev; 4840 struct netdev_queue *txq; 4841 bool free_skb = true; 4842 int cpu, rc; 4843 4844 txq = netdev_core_pick_tx(dev, skb, NULL); 4845 cpu = smp_processor_id(); 4846 HARD_TX_LOCK(dev, txq, cpu); 4847 if (!netif_xmit_stopped(txq)) { 4848 rc = netdev_start_xmit(skb, dev, txq, 0); 4849 if (dev_xmit_complete(rc)) 4850 free_skb = false; 4851 } 4852 HARD_TX_UNLOCK(dev, txq); 4853 if (free_skb) { 4854 trace_xdp_exception(dev, xdp_prog, XDP_TX); 4855 kfree_skb(skb); 4856 } 4857 } 4858 4859 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key); 4860 4861 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb) 4862 { 4863 if (xdp_prog) { 4864 struct xdp_buff xdp; 4865 u32 act; 4866 int err; 4867 4868 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog); 4869 if (act != XDP_PASS) { 4870 switch (act) { 4871 case XDP_REDIRECT: 4872 err = xdp_do_generic_redirect(skb->dev, skb, 4873 &xdp, xdp_prog); 4874 if (err) 4875 goto out_redir; 4876 break; 4877 case XDP_TX: 4878 generic_xdp_tx(skb, xdp_prog); 4879 break; 4880 } 4881 return XDP_DROP; 4882 } 4883 } 4884 return XDP_PASS; 4885 out_redir: 4886 kfree_skb(skb); 4887 return XDP_DROP; 4888 } 4889 EXPORT_SYMBOL_GPL(do_xdp_generic); 4890 4891 static int netif_rx_internal(struct sk_buff *skb) 4892 { 4893 int ret; 4894 4895 net_timestamp_check(netdev_tstamp_prequeue, skb); 4896 4897 trace_netif_rx(skb); 4898 4899 #ifdef CONFIG_RPS 4900 if (static_branch_unlikely(&rps_needed)) { 4901 struct rps_dev_flow voidflow, *rflow = &voidflow; 4902 int cpu; 4903 4904 preempt_disable(); 4905 rcu_read_lock(); 4906 4907 cpu = get_rps_cpu(skb->dev, skb, &rflow); 4908 if (cpu < 0) 4909 cpu = smp_processor_id(); 4910 4911 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 4912 4913 rcu_read_unlock(); 4914 preempt_enable(); 4915 } else 4916 #endif 4917 { 4918 unsigned int qtail; 4919 4920 ret = enqueue_to_backlog(skb, get_cpu(), &qtail); 4921 put_cpu(); 4922 } 4923 return ret; 4924 } 4925 4926 /** 4927 * netif_rx - post buffer to the network code 4928 * @skb: buffer to post 4929 * 4930 * This function receives a packet from a device driver and queues it for 4931 * the upper (protocol) levels to process. It always succeeds. The buffer 4932 * may be dropped during processing for congestion control or by the 4933 * protocol layers. 4934 * 4935 * return values: 4936 * NET_RX_SUCCESS (no congestion) 4937 * NET_RX_DROP (packet was dropped) 4938 * 4939 */ 4940 4941 int netif_rx(struct sk_buff *skb) 4942 { 4943 int ret; 4944 4945 trace_netif_rx_entry(skb); 4946 4947 ret = netif_rx_internal(skb); 4948 trace_netif_rx_exit(ret); 4949 4950 return ret; 4951 } 4952 EXPORT_SYMBOL(netif_rx); 4953 4954 int netif_rx_ni(struct sk_buff *skb) 4955 { 4956 int err; 4957 4958 trace_netif_rx_ni_entry(skb); 4959 4960 preempt_disable(); 4961 err = netif_rx_internal(skb); 4962 if (local_softirq_pending()) 4963 do_softirq(); 4964 preempt_enable(); 4965 trace_netif_rx_ni_exit(err); 4966 4967 return err; 4968 } 4969 EXPORT_SYMBOL(netif_rx_ni); 4970 4971 int netif_rx_any_context(struct sk_buff *skb) 4972 { 4973 /* 4974 * If invoked from contexts which do not invoke bottom half 4975 * processing either at return from interrupt or when softrqs are 4976 * reenabled, use netif_rx_ni() which invokes bottomhalf processing 4977 * directly. 4978 */ 4979 if (in_interrupt()) 4980 return netif_rx(skb); 4981 else 4982 return netif_rx_ni(skb); 4983 } 4984 EXPORT_SYMBOL(netif_rx_any_context); 4985 4986 static __latent_entropy void net_tx_action(struct softirq_action *h) 4987 { 4988 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 4989 4990 if (sd->completion_queue) { 4991 struct sk_buff *clist; 4992 4993 local_irq_disable(); 4994 clist = sd->completion_queue; 4995 sd->completion_queue = NULL; 4996 local_irq_enable(); 4997 4998 while (clist) { 4999 struct sk_buff *skb = clist; 5000 5001 clist = clist->next; 5002 5003 WARN_ON(refcount_read(&skb->users)); 5004 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED)) 5005 trace_consume_skb(skb); 5006 else 5007 trace_kfree_skb(skb, net_tx_action); 5008 5009 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) 5010 __kfree_skb(skb); 5011 else 5012 __kfree_skb_defer(skb); 5013 } 5014 } 5015 5016 if (sd->output_queue) { 5017 struct Qdisc *head; 5018 5019 local_irq_disable(); 5020 head = sd->output_queue; 5021 sd->output_queue = NULL; 5022 sd->output_queue_tailp = &sd->output_queue; 5023 local_irq_enable(); 5024 5025 rcu_read_lock(); 5026 5027 while (head) { 5028 struct Qdisc *q = head; 5029 spinlock_t *root_lock = NULL; 5030 5031 head = head->next_sched; 5032 5033 /* We need to make sure head->next_sched is read 5034 * before clearing __QDISC_STATE_SCHED 5035 */ 5036 smp_mb__before_atomic(); 5037 5038 if (!(q->flags & TCQ_F_NOLOCK)) { 5039 root_lock = qdisc_lock(q); 5040 spin_lock(root_lock); 5041 } else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, 5042 &q->state))) { 5043 /* There is a synchronize_net() between 5044 * STATE_DEACTIVATED flag being set and 5045 * qdisc_reset()/some_qdisc_is_busy() in 5046 * dev_deactivate(), so we can safely bail out 5047 * early here to avoid data race between 5048 * qdisc_deactivate() and some_qdisc_is_busy() 5049 * for lockless qdisc. 5050 */ 5051 clear_bit(__QDISC_STATE_SCHED, &q->state); 5052 continue; 5053 } 5054 5055 clear_bit(__QDISC_STATE_SCHED, &q->state); 5056 qdisc_run(q); 5057 if (root_lock) 5058 spin_unlock(root_lock); 5059 } 5060 5061 rcu_read_unlock(); 5062 } 5063 5064 xfrm_dev_backlog(sd); 5065 } 5066 5067 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) 5068 /* This hook is defined here for ATM LANE */ 5069 int (*br_fdb_test_addr_hook)(struct net_device *dev, 5070 unsigned char *addr) __read_mostly; 5071 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); 5072 #endif 5073 5074 static inline struct sk_buff * 5075 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 5076 struct net_device *orig_dev, bool *another) 5077 { 5078 #ifdef CONFIG_NET_CLS_ACT 5079 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress); 5080 struct tcf_result cl_res; 5081 5082 /* If there's at least one ingress present somewhere (so 5083 * we get here via enabled static key), remaining devices 5084 * that are not configured with an ingress qdisc will bail 5085 * out here. 5086 */ 5087 if (!miniq) 5088 return skb; 5089 5090 if (*pt_prev) { 5091 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5092 *pt_prev = NULL; 5093 } 5094 5095 qdisc_skb_cb(skb)->pkt_len = skb->len; 5096 qdisc_skb_cb(skb)->mru = 0; 5097 qdisc_skb_cb(skb)->post_ct = false; 5098 skb->tc_at_ingress = 1; 5099 mini_qdisc_bstats_cpu_update(miniq, skb); 5100 5101 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) { 5102 case TC_ACT_OK: 5103 case TC_ACT_RECLASSIFY: 5104 skb->tc_index = TC_H_MIN(cl_res.classid); 5105 break; 5106 case TC_ACT_SHOT: 5107 mini_qdisc_qstats_cpu_drop(miniq); 5108 kfree_skb(skb); 5109 return NULL; 5110 case TC_ACT_STOLEN: 5111 case TC_ACT_QUEUED: 5112 case TC_ACT_TRAP: 5113 consume_skb(skb); 5114 return NULL; 5115 case TC_ACT_REDIRECT: 5116 /* skb_mac_header check was done by cls/act_bpf, so 5117 * we can safely push the L2 header back before 5118 * redirecting to another netdev 5119 */ 5120 __skb_push(skb, skb->mac_len); 5121 if (skb_do_redirect(skb) == -EAGAIN) { 5122 __skb_pull(skb, skb->mac_len); 5123 *another = true; 5124 break; 5125 } 5126 return NULL; 5127 case TC_ACT_CONSUMED: 5128 return NULL; 5129 default: 5130 break; 5131 } 5132 #endif /* CONFIG_NET_CLS_ACT */ 5133 return skb; 5134 } 5135 5136 /** 5137 * netdev_is_rx_handler_busy - check if receive handler is registered 5138 * @dev: device to check 5139 * 5140 * Check if a receive handler is already registered for a given device. 5141 * Return true if there one. 5142 * 5143 * The caller must hold the rtnl_mutex. 5144 */ 5145 bool netdev_is_rx_handler_busy(struct net_device *dev) 5146 { 5147 ASSERT_RTNL(); 5148 return dev && rtnl_dereference(dev->rx_handler); 5149 } 5150 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); 5151 5152 /** 5153 * netdev_rx_handler_register - register receive handler 5154 * @dev: device to register a handler for 5155 * @rx_handler: receive handler to register 5156 * @rx_handler_data: data pointer that is used by rx handler 5157 * 5158 * Register a receive handler for a device. This handler will then be 5159 * called from __netif_receive_skb. A negative errno code is returned 5160 * on a failure. 5161 * 5162 * The caller must hold the rtnl_mutex. 5163 * 5164 * For a general description of rx_handler, see enum rx_handler_result. 5165 */ 5166 int netdev_rx_handler_register(struct net_device *dev, 5167 rx_handler_func_t *rx_handler, 5168 void *rx_handler_data) 5169 { 5170 if (netdev_is_rx_handler_busy(dev)) 5171 return -EBUSY; 5172 5173 if (dev->priv_flags & IFF_NO_RX_HANDLER) 5174 return -EINVAL; 5175 5176 /* Note: rx_handler_data must be set before rx_handler */ 5177 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); 5178 rcu_assign_pointer(dev->rx_handler, rx_handler); 5179 5180 return 0; 5181 } 5182 EXPORT_SYMBOL_GPL(netdev_rx_handler_register); 5183 5184 /** 5185 * netdev_rx_handler_unregister - unregister receive handler 5186 * @dev: device to unregister a handler from 5187 * 5188 * Unregister a receive handler from a device. 5189 * 5190 * The caller must hold the rtnl_mutex. 5191 */ 5192 void netdev_rx_handler_unregister(struct net_device *dev) 5193 { 5194 5195 ASSERT_RTNL(); 5196 RCU_INIT_POINTER(dev->rx_handler, NULL); 5197 /* a reader seeing a non NULL rx_handler in a rcu_read_lock() 5198 * section has a guarantee to see a non NULL rx_handler_data 5199 * as well. 5200 */ 5201 synchronize_net(); 5202 RCU_INIT_POINTER(dev->rx_handler_data, NULL); 5203 } 5204 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); 5205 5206 /* 5207 * Limit the use of PFMEMALLOC reserves to those protocols that implement 5208 * the special handling of PFMEMALLOC skbs. 5209 */ 5210 static bool skb_pfmemalloc_protocol(struct sk_buff *skb) 5211 { 5212 switch (skb->protocol) { 5213 case htons(ETH_P_ARP): 5214 case htons(ETH_P_IP): 5215 case htons(ETH_P_IPV6): 5216 case htons(ETH_P_8021Q): 5217 case htons(ETH_P_8021AD): 5218 return true; 5219 default: 5220 return false; 5221 } 5222 } 5223 5224 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, 5225 int *ret, struct net_device *orig_dev) 5226 { 5227 if (nf_hook_ingress_active(skb)) { 5228 int ingress_retval; 5229 5230 if (*pt_prev) { 5231 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5232 *pt_prev = NULL; 5233 } 5234 5235 rcu_read_lock(); 5236 ingress_retval = nf_hook_ingress(skb); 5237 rcu_read_unlock(); 5238 return ingress_retval; 5239 } 5240 return 0; 5241 } 5242 5243 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc, 5244 struct packet_type **ppt_prev) 5245 { 5246 struct packet_type *ptype, *pt_prev; 5247 rx_handler_func_t *rx_handler; 5248 struct sk_buff *skb = *pskb; 5249 struct net_device *orig_dev; 5250 bool deliver_exact = false; 5251 int ret = NET_RX_DROP; 5252 __be16 type; 5253 5254 net_timestamp_check(!netdev_tstamp_prequeue, skb); 5255 5256 trace_netif_receive_skb(skb); 5257 5258 orig_dev = skb->dev; 5259 5260 skb_reset_network_header(skb); 5261 if (!skb_transport_header_was_set(skb)) 5262 skb_reset_transport_header(skb); 5263 skb_reset_mac_len(skb); 5264 5265 pt_prev = NULL; 5266 5267 another_round: 5268 skb->skb_iif = skb->dev->ifindex; 5269 5270 __this_cpu_inc(softnet_data.processed); 5271 5272 if (static_branch_unlikely(&generic_xdp_needed_key)) { 5273 int ret2; 5274 5275 migrate_disable(); 5276 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb); 5277 migrate_enable(); 5278 5279 if (ret2 != XDP_PASS) { 5280 ret = NET_RX_DROP; 5281 goto out; 5282 } 5283 } 5284 5285 if (eth_type_vlan(skb->protocol)) { 5286 skb = skb_vlan_untag(skb); 5287 if (unlikely(!skb)) 5288 goto out; 5289 } 5290 5291 if (skb_skip_tc_classify(skb)) 5292 goto skip_classify; 5293 5294 if (pfmemalloc) 5295 goto skip_taps; 5296 5297 list_for_each_entry_rcu(ptype, &ptype_all, list) { 5298 if (pt_prev) 5299 ret = deliver_skb(skb, pt_prev, orig_dev); 5300 pt_prev = ptype; 5301 } 5302 5303 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { 5304 if (pt_prev) 5305 ret = deliver_skb(skb, pt_prev, orig_dev); 5306 pt_prev = ptype; 5307 } 5308 5309 skip_taps: 5310 #ifdef CONFIG_NET_INGRESS 5311 if (static_branch_unlikely(&ingress_needed_key)) { 5312 bool another = false; 5313 5314 nf_skip_egress(skb, true); 5315 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev, 5316 &another); 5317 if (another) 5318 goto another_round; 5319 if (!skb) 5320 goto out; 5321 5322 nf_skip_egress(skb, false); 5323 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) 5324 goto out; 5325 } 5326 #endif 5327 skb_reset_redirect(skb); 5328 skip_classify: 5329 if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) 5330 goto drop; 5331 5332 if (skb_vlan_tag_present(skb)) { 5333 if (pt_prev) { 5334 ret = deliver_skb(skb, pt_prev, orig_dev); 5335 pt_prev = NULL; 5336 } 5337 if (vlan_do_receive(&skb)) 5338 goto another_round; 5339 else if (unlikely(!skb)) 5340 goto out; 5341 } 5342 5343 rx_handler = rcu_dereference(skb->dev->rx_handler); 5344 if (rx_handler) { 5345 if (pt_prev) { 5346 ret = deliver_skb(skb, pt_prev, orig_dev); 5347 pt_prev = NULL; 5348 } 5349 switch (rx_handler(&skb)) { 5350 case RX_HANDLER_CONSUMED: 5351 ret = NET_RX_SUCCESS; 5352 goto out; 5353 case RX_HANDLER_ANOTHER: 5354 goto another_round; 5355 case RX_HANDLER_EXACT: 5356 deliver_exact = true; 5357 break; 5358 case RX_HANDLER_PASS: 5359 break; 5360 default: 5361 BUG(); 5362 } 5363 } 5364 5365 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) { 5366 check_vlan_id: 5367 if (skb_vlan_tag_get_id(skb)) { 5368 /* Vlan id is non 0 and vlan_do_receive() above couldn't 5369 * find vlan device. 5370 */ 5371 skb->pkt_type = PACKET_OTHERHOST; 5372 } else if (eth_type_vlan(skb->protocol)) { 5373 /* Outer header is 802.1P with vlan 0, inner header is 5374 * 802.1Q or 802.1AD and vlan_do_receive() above could 5375 * not find vlan dev for vlan id 0. 5376 */ 5377 __vlan_hwaccel_clear_tag(skb); 5378 skb = skb_vlan_untag(skb); 5379 if (unlikely(!skb)) 5380 goto out; 5381 if (vlan_do_receive(&skb)) 5382 /* After stripping off 802.1P header with vlan 0 5383 * vlan dev is found for inner header. 5384 */ 5385 goto another_round; 5386 else if (unlikely(!skb)) 5387 goto out; 5388 else 5389 /* We have stripped outer 802.1P vlan 0 header. 5390 * But could not find vlan dev. 5391 * check again for vlan id to set OTHERHOST. 5392 */ 5393 goto check_vlan_id; 5394 } 5395 /* Note: we might in the future use prio bits 5396 * and set skb->priority like in vlan_do_receive() 5397 * For the time being, just ignore Priority Code Point 5398 */ 5399 __vlan_hwaccel_clear_tag(skb); 5400 } 5401 5402 type = skb->protocol; 5403 5404 /* deliver only exact match when indicated */ 5405 if (likely(!deliver_exact)) { 5406 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5407 &ptype_base[ntohs(type) & 5408 PTYPE_HASH_MASK]); 5409 } 5410 5411 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5412 &orig_dev->ptype_specific); 5413 5414 if (unlikely(skb->dev != orig_dev)) { 5415 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5416 &skb->dev->ptype_specific); 5417 } 5418 5419 if (pt_prev) { 5420 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 5421 goto drop; 5422 *ppt_prev = pt_prev; 5423 } else { 5424 drop: 5425 if (!deliver_exact) 5426 atomic_long_inc(&skb->dev->rx_dropped); 5427 else 5428 atomic_long_inc(&skb->dev->rx_nohandler); 5429 kfree_skb(skb); 5430 /* Jamal, now you will not able to escape explaining 5431 * me how you were going to use this. :-) 5432 */ 5433 ret = NET_RX_DROP; 5434 } 5435 5436 out: 5437 /* The invariant here is that if *ppt_prev is not NULL 5438 * then skb should also be non-NULL. 5439 * 5440 * Apparently *ppt_prev assignment above holds this invariant due to 5441 * skb dereferencing near it. 5442 */ 5443 *pskb = skb; 5444 return ret; 5445 } 5446 5447 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc) 5448 { 5449 struct net_device *orig_dev = skb->dev; 5450 struct packet_type *pt_prev = NULL; 5451 int ret; 5452 5453 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5454 if (pt_prev) 5455 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb, 5456 skb->dev, pt_prev, orig_dev); 5457 return ret; 5458 } 5459 5460 /** 5461 * netif_receive_skb_core - special purpose version of netif_receive_skb 5462 * @skb: buffer to process 5463 * 5464 * More direct receive version of netif_receive_skb(). It should 5465 * only be used by callers that have a need to skip RPS and Generic XDP. 5466 * Caller must also take care of handling if ``(page_is_)pfmemalloc``. 5467 * 5468 * This function may only be called from softirq context and interrupts 5469 * should be enabled. 5470 * 5471 * Return values (usually ignored): 5472 * NET_RX_SUCCESS: no congestion 5473 * NET_RX_DROP: packet was dropped 5474 */ 5475 int netif_receive_skb_core(struct sk_buff *skb) 5476 { 5477 int ret; 5478 5479 rcu_read_lock(); 5480 ret = __netif_receive_skb_one_core(skb, false); 5481 rcu_read_unlock(); 5482 5483 return ret; 5484 } 5485 EXPORT_SYMBOL(netif_receive_skb_core); 5486 5487 static inline void __netif_receive_skb_list_ptype(struct list_head *head, 5488 struct packet_type *pt_prev, 5489 struct net_device *orig_dev) 5490 { 5491 struct sk_buff *skb, *next; 5492 5493 if (!pt_prev) 5494 return; 5495 if (list_empty(head)) 5496 return; 5497 if (pt_prev->list_func != NULL) 5498 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv, 5499 ip_list_rcv, head, pt_prev, orig_dev); 5500 else 5501 list_for_each_entry_safe(skb, next, head, list) { 5502 skb_list_del_init(skb); 5503 pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 5504 } 5505 } 5506 5507 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc) 5508 { 5509 /* Fast-path assumptions: 5510 * - There is no RX handler. 5511 * - Only one packet_type matches. 5512 * If either of these fails, we will end up doing some per-packet 5513 * processing in-line, then handling the 'last ptype' for the whole 5514 * sublist. This can't cause out-of-order delivery to any single ptype, 5515 * because the 'last ptype' must be constant across the sublist, and all 5516 * other ptypes are handled per-packet. 5517 */ 5518 /* Current (common) ptype of sublist */ 5519 struct packet_type *pt_curr = NULL; 5520 /* Current (common) orig_dev of sublist */ 5521 struct net_device *od_curr = NULL; 5522 struct list_head sublist; 5523 struct sk_buff *skb, *next; 5524 5525 INIT_LIST_HEAD(&sublist); 5526 list_for_each_entry_safe(skb, next, head, list) { 5527 struct net_device *orig_dev = skb->dev; 5528 struct packet_type *pt_prev = NULL; 5529 5530 skb_list_del_init(skb); 5531 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5532 if (!pt_prev) 5533 continue; 5534 if (pt_curr != pt_prev || od_curr != orig_dev) { 5535 /* dispatch old sublist */ 5536 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5537 /* start new sublist */ 5538 INIT_LIST_HEAD(&sublist); 5539 pt_curr = pt_prev; 5540 od_curr = orig_dev; 5541 } 5542 list_add_tail(&skb->list, &sublist); 5543 } 5544 5545 /* dispatch final sublist */ 5546 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5547 } 5548 5549 static int __netif_receive_skb(struct sk_buff *skb) 5550 { 5551 int ret; 5552 5553 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { 5554 unsigned int noreclaim_flag; 5555 5556 /* 5557 * PFMEMALLOC skbs are special, they should 5558 * - be delivered to SOCK_MEMALLOC sockets only 5559 * - stay away from userspace 5560 * - have bounded memory usage 5561 * 5562 * Use PF_MEMALLOC as this saves us from propagating the allocation 5563 * context down to all allocation sites. 5564 */ 5565 noreclaim_flag = memalloc_noreclaim_save(); 5566 ret = __netif_receive_skb_one_core(skb, true); 5567 memalloc_noreclaim_restore(noreclaim_flag); 5568 } else 5569 ret = __netif_receive_skb_one_core(skb, false); 5570 5571 return ret; 5572 } 5573 5574 static void __netif_receive_skb_list(struct list_head *head) 5575 { 5576 unsigned long noreclaim_flag = 0; 5577 struct sk_buff *skb, *next; 5578 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */ 5579 5580 list_for_each_entry_safe(skb, next, head, list) { 5581 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) { 5582 struct list_head sublist; 5583 5584 /* Handle the previous sublist */ 5585 list_cut_before(&sublist, head, &skb->list); 5586 if (!list_empty(&sublist)) 5587 __netif_receive_skb_list_core(&sublist, pfmemalloc); 5588 pfmemalloc = !pfmemalloc; 5589 /* See comments in __netif_receive_skb */ 5590 if (pfmemalloc) 5591 noreclaim_flag = memalloc_noreclaim_save(); 5592 else 5593 memalloc_noreclaim_restore(noreclaim_flag); 5594 } 5595 } 5596 /* Handle the remaining sublist */ 5597 if (!list_empty(head)) 5598 __netif_receive_skb_list_core(head, pfmemalloc); 5599 /* Restore pflags */ 5600 if (pfmemalloc) 5601 memalloc_noreclaim_restore(noreclaim_flag); 5602 } 5603 5604 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp) 5605 { 5606 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog); 5607 struct bpf_prog *new = xdp->prog; 5608 int ret = 0; 5609 5610 switch (xdp->command) { 5611 case XDP_SETUP_PROG: 5612 rcu_assign_pointer(dev->xdp_prog, new); 5613 if (old) 5614 bpf_prog_put(old); 5615 5616 if (old && !new) { 5617 static_branch_dec(&generic_xdp_needed_key); 5618 } else if (new && !old) { 5619 static_branch_inc(&generic_xdp_needed_key); 5620 dev_disable_lro(dev); 5621 dev_disable_gro_hw(dev); 5622 } 5623 break; 5624 5625 default: 5626 ret = -EINVAL; 5627 break; 5628 } 5629 5630 return ret; 5631 } 5632 5633 static int netif_receive_skb_internal(struct sk_buff *skb) 5634 { 5635 int ret; 5636 5637 net_timestamp_check(netdev_tstamp_prequeue, skb); 5638 5639 if (skb_defer_rx_timestamp(skb)) 5640 return NET_RX_SUCCESS; 5641 5642 rcu_read_lock(); 5643 #ifdef CONFIG_RPS 5644 if (static_branch_unlikely(&rps_needed)) { 5645 struct rps_dev_flow voidflow, *rflow = &voidflow; 5646 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5647 5648 if (cpu >= 0) { 5649 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5650 rcu_read_unlock(); 5651 return ret; 5652 } 5653 } 5654 #endif 5655 ret = __netif_receive_skb(skb); 5656 rcu_read_unlock(); 5657 return ret; 5658 } 5659 5660 static void netif_receive_skb_list_internal(struct list_head *head) 5661 { 5662 struct sk_buff *skb, *next; 5663 struct list_head sublist; 5664 5665 INIT_LIST_HEAD(&sublist); 5666 list_for_each_entry_safe(skb, next, head, list) { 5667 net_timestamp_check(netdev_tstamp_prequeue, skb); 5668 skb_list_del_init(skb); 5669 if (!skb_defer_rx_timestamp(skb)) 5670 list_add_tail(&skb->list, &sublist); 5671 } 5672 list_splice_init(&sublist, head); 5673 5674 rcu_read_lock(); 5675 #ifdef CONFIG_RPS 5676 if (static_branch_unlikely(&rps_needed)) { 5677 list_for_each_entry_safe(skb, next, head, list) { 5678 struct rps_dev_flow voidflow, *rflow = &voidflow; 5679 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5680 5681 if (cpu >= 0) { 5682 /* Will be handled, remove from list */ 5683 skb_list_del_init(skb); 5684 enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5685 } 5686 } 5687 } 5688 #endif 5689 __netif_receive_skb_list(head); 5690 rcu_read_unlock(); 5691 } 5692 5693 /** 5694 * netif_receive_skb - process receive buffer from network 5695 * @skb: buffer to process 5696 * 5697 * netif_receive_skb() is the main receive data processing function. 5698 * It always succeeds. The buffer may be dropped during processing 5699 * for congestion control or by the protocol layers. 5700 * 5701 * This function may only be called from softirq context and interrupts 5702 * should be enabled. 5703 * 5704 * Return values (usually ignored): 5705 * NET_RX_SUCCESS: no congestion 5706 * NET_RX_DROP: packet was dropped 5707 */ 5708 int netif_receive_skb(struct sk_buff *skb) 5709 { 5710 int ret; 5711 5712 trace_netif_receive_skb_entry(skb); 5713 5714 ret = netif_receive_skb_internal(skb); 5715 trace_netif_receive_skb_exit(ret); 5716 5717 return ret; 5718 } 5719 EXPORT_SYMBOL(netif_receive_skb); 5720 5721 /** 5722 * netif_receive_skb_list - process many receive buffers from network 5723 * @head: list of skbs to process. 5724 * 5725 * Since return value of netif_receive_skb() is normally ignored, and 5726 * wouldn't be meaningful for a list, this function returns void. 5727 * 5728 * This function may only be called from softirq context and interrupts 5729 * should be enabled. 5730 */ 5731 void netif_receive_skb_list(struct list_head *head) 5732 { 5733 struct sk_buff *skb; 5734 5735 if (list_empty(head)) 5736 return; 5737 if (trace_netif_receive_skb_list_entry_enabled()) { 5738 list_for_each_entry(skb, head, list) 5739 trace_netif_receive_skb_list_entry(skb); 5740 } 5741 netif_receive_skb_list_internal(head); 5742 trace_netif_receive_skb_list_exit(0); 5743 } 5744 EXPORT_SYMBOL(netif_receive_skb_list); 5745 5746 static DEFINE_PER_CPU(struct work_struct, flush_works); 5747 5748 /* Network device is going away, flush any packets still pending */ 5749 static void flush_backlog(struct work_struct *work) 5750 { 5751 struct sk_buff *skb, *tmp; 5752 struct softnet_data *sd; 5753 5754 local_bh_disable(); 5755 sd = this_cpu_ptr(&softnet_data); 5756 5757 local_irq_disable(); 5758 rps_lock(sd); 5759 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { 5760 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5761 __skb_unlink(skb, &sd->input_pkt_queue); 5762 dev_kfree_skb_irq(skb); 5763 input_queue_head_incr(sd); 5764 } 5765 } 5766 rps_unlock(sd); 5767 local_irq_enable(); 5768 5769 skb_queue_walk_safe(&sd->process_queue, skb, tmp) { 5770 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5771 __skb_unlink(skb, &sd->process_queue); 5772 kfree_skb(skb); 5773 input_queue_head_incr(sd); 5774 } 5775 } 5776 local_bh_enable(); 5777 } 5778 5779 static bool flush_required(int cpu) 5780 { 5781 #if IS_ENABLED(CONFIG_RPS) 5782 struct softnet_data *sd = &per_cpu(softnet_data, cpu); 5783 bool do_flush; 5784 5785 local_irq_disable(); 5786 rps_lock(sd); 5787 5788 /* as insertion into process_queue happens with the rps lock held, 5789 * process_queue access may race only with dequeue 5790 */ 5791 do_flush = !skb_queue_empty(&sd->input_pkt_queue) || 5792 !skb_queue_empty_lockless(&sd->process_queue); 5793 rps_unlock(sd); 5794 local_irq_enable(); 5795 5796 return do_flush; 5797 #endif 5798 /* without RPS we can't safely check input_pkt_queue: during a 5799 * concurrent remote skb_queue_splice() we can detect as empty both 5800 * input_pkt_queue and process_queue even if the latter could end-up 5801 * containing a lot of packets. 5802 */ 5803 return true; 5804 } 5805 5806 static void flush_all_backlogs(void) 5807 { 5808 static cpumask_t flush_cpus; 5809 unsigned int cpu; 5810 5811 /* since we are under rtnl lock protection we can use static data 5812 * for the cpumask and avoid allocating on stack the possibly 5813 * large mask 5814 */ 5815 ASSERT_RTNL(); 5816 5817 cpus_read_lock(); 5818 5819 cpumask_clear(&flush_cpus); 5820 for_each_online_cpu(cpu) { 5821 if (flush_required(cpu)) { 5822 queue_work_on(cpu, system_highpri_wq, 5823 per_cpu_ptr(&flush_works, cpu)); 5824 cpumask_set_cpu(cpu, &flush_cpus); 5825 } 5826 } 5827 5828 /* we can have in flight packet[s] on the cpus we are not flushing, 5829 * synchronize_net() in unregister_netdevice_many() will take care of 5830 * them 5831 */ 5832 for_each_cpu(cpu, &flush_cpus) 5833 flush_work(per_cpu_ptr(&flush_works, cpu)); 5834 5835 cpus_read_unlock(); 5836 } 5837 5838 /* Pass the currently batched GRO_NORMAL SKBs up to the stack. */ 5839 static void gro_normal_list(struct napi_struct *napi) 5840 { 5841 if (!napi->rx_count) 5842 return; 5843 netif_receive_skb_list_internal(&napi->rx_list); 5844 INIT_LIST_HEAD(&napi->rx_list); 5845 napi->rx_count = 0; 5846 } 5847 5848 /* Queue one GRO_NORMAL SKB up for list processing. If batch size exceeded, 5849 * pass the whole batch up to the stack. 5850 */ 5851 static void gro_normal_one(struct napi_struct *napi, struct sk_buff *skb, int segs) 5852 { 5853 list_add_tail(&skb->list, &napi->rx_list); 5854 napi->rx_count += segs; 5855 if (napi->rx_count >= gro_normal_batch) 5856 gro_normal_list(napi); 5857 } 5858 5859 static void napi_gro_complete(struct napi_struct *napi, struct sk_buff *skb) 5860 { 5861 struct packet_offload *ptype; 5862 __be16 type = skb->protocol; 5863 struct list_head *head = &offload_base; 5864 int err = -ENOENT; 5865 5866 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb)); 5867 5868 if (NAPI_GRO_CB(skb)->count == 1) { 5869 skb_shinfo(skb)->gso_size = 0; 5870 goto out; 5871 } 5872 5873 rcu_read_lock(); 5874 list_for_each_entry_rcu(ptype, head, list) { 5875 if (ptype->type != type || !ptype->callbacks.gro_complete) 5876 continue; 5877 5878 err = INDIRECT_CALL_INET(ptype->callbacks.gro_complete, 5879 ipv6_gro_complete, inet_gro_complete, 5880 skb, 0); 5881 break; 5882 } 5883 rcu_read_unlock(); 5884 5885 if (err) { 5886 WARN_ON(&ptype->list == head); 5887 kfree_skb(skb); 5888 return; 5889 } 5890 5891 out: 5892 gro_normal_one(napi, skb, NAPI_GRO_CB(skb)->count); 5893 } 5894 5895 static void __napi_gro_flush_chain(struct napi_struct *napi, u32 index, 5896 bool flush_old) 5897 { 5898 struct list_head *head = &napi->gro_hash[index].list; 5899 struct sk_buff *skb, *p; 5900 5901 list_for_each_entry_safe_reverse(skb, p, head, list) { 5902 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies) 5903 return; 5904 skb_list_del_init(skb); 5905 napi_gro_complete(napi, skb); 5906 napi->gro_hash[index].count--; 5907 } 5908 5909 if (!napi->gro_hash[index].count) 5910 __clear_bit(index, &napi->gro_bitmask); 5911 } 5912 5913 /* napi->gro_hash[].list contains packets ordered by age. 5914 * youngest packets at the head of it. 5915 * Complete skbs in reverse order to reduce latencies. 5916 */ 5917 void napi_gro_flush(struct napi_struct *napi, bool flush_old) 5918 { 5919 unsigned long bitmask = napi->gro_bitmask; 5920 unsigned int i, base = ~0U; 5921 5922 while ((i = ffs(bitmask)) != 0) { 5923 bitmask >>= i; 5924 base += i; 5925 __napi_gro_flush_chain(napi, base, flush_old); 5926 } 5927 } 5928 EXPORT_SYMBOL(napi_gro_flush); 5929 5930 static void gro_list_prepare(const struct list_head *head, 5931 const struct sk_buff *skb) 5932 { 5933 unsigned int maclen = skb->dev->hard_header_len; 5934 u32 hash = skb_get_hash_raw(skb); 5935 struct sk_buff *p; 5936 5937 list_for_each_entry(p, head, list) { 5938 unsigned long diffs; 5939 5940 NAPI_GRO_CB(p)->flush = 0; 5941 5942 if (hash != skb_get_hash_raw(p)) { 5943 NAPI_GRO_CB(p)->same_flow = 0; 5944 continue; 5945 } 5946 5947 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev; 5948 diffs |= skb_vlan_tag_present(p) ^ skb_vlan_tag_present(skb); 5949 if (skb_vlan_tag_present(p)) 5950 diffs |= skb_vlan_tag_get(p) ^ skb_vlan_tag_get(skb); 5951 diffs |= skb_metadata_differs(p, skb); 5952 if (maclen == ETH_HLEN) 5953 diffs |= compare_ether_header(skb_mac_header(p), 5954 skb_mac_header(skb)); 5955 else if (!diffs) 5956 diffs = memcmp(skb_mac_header(p), 5957 skb_mac_header(skb), 5958 maclen); 5959 5960 /* in most common scenarions 'slow_gro' is 0 5961 * otherwise we are already on some slower paths 5962 * either skip all the infrequent tests altogether or 5963 * avoid trying too hard to skip each of them individually 5964 */ 5965 if (!diffs && unlikely(skb->slow_gro | p->slow_gro)) { 5966 #if IS_ENABLED(CONFIG_SKB_EXTENSIONS) && IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 5967 struct tc_skb_ext *skb_ext; 5968 struct tc_skb_ext *p_ext; 5969 #endif 5970 5971 diffs |= p->sk != skb->sk; 5972 diffs |= skb_metadata_dst_cmp(p, skb); 5973 diffs |= skb_get_nfct(p) ^ skb_get_nfct(skb); 5974 5975 #if IS_ENABLED(CONFIG_SKB_EXTENSIONS) && IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 5976 skb_ext = skb_ext_find(skb, TC_SKB_EXT); 5977 p_ext = skb_ext_find(p, TC_SKB_EXT); 5978 5979 diffs |= (!!p_ext) ^ (!!skb_ext); 5980 if (!diffs && unlikely(skb_ext)) 5981 diffs |= p_ext->chain ^ skb_ext->chain; 5982 #endif 5983 } 5984 5985 NAPI_GRO_CB(p)->same_flow = !diffs; 5986 } 5987 } 5988 5989 static inline void skb_gro_reset_offset(struct sk_buff *skb, u32 nhoff) 5990 { 5991 const struct skb_shared_info *pinfo = skb_shinfo(skb); 5992 const skb_frag_t *frag0 = &pinfo->frags[0]; 5993 5994 NAPI_GRO_CB(skb)->data_offset = 0; 5995 NAPI_GRO_CB(skb)->frag0 = NULL; 5996 NAPI_GRO_CB(skb)->frag0_len = 0; 5997 5998 if (!skb_headlen(skb) && pinfo->nr_frags && 5999 !PageHighMem(skb_frag_page(frag0)) && 6000 (!NET_IP_ALIGN || !((skb_frag_off(frag0) + nhoff) & 3))) { 6001 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0); 6002 NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int, 6003 skb_frag_size(frag0), 6004 skb->end - skb->tail); 6005 } 6006 } 6007 6008 static void gro_pull_from_frag0(struct sk_buff *skb, int grow) 6009 { 6010 struct skb_shared_info *pinfo = skb_shinfo(skb); 6011 6012 BUG_ON(skb->end - skb->tail < grow); 6013 6014 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); 6015 6016 skb->data_len -= grow; 6017 skb->tail += grow; 6018 6019 skb_frag_off_add(&pinfo->frags[0], grow); 6020 skb_frag_size_sub(&pinfo->frags[0], grow); 6021 6022 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) { 6023 skb_frag_unref(skb, 0); 6024 memmove(pinfo->frags, pinfo->frags + 1, 6025 --pinfo->nr_frags * sizeof(pinfo->frags[0])); 6026 } 6027 } 6028 6029 static void gro_flush_oldest(struct napi_struct *napi, struct list_head *head) 6030 { 6031 struct sk_buff *oldest; 6032 6033 oldest = list_last_entry(head, struct sk_buff, list); 6034 6035 /* We are called with head length >= MAX_GRO_SKBS, so this is 6036 * impossible. 6037 */ 6038 if (WARN_ON_ONCE(!oldest)) 6039 return; 6040 6041 /* Do not adjust napi->gro_hash[].count, caller is adding a new 6042 * SKB to the chain. 6043 */ 6044 skb_list_del_init(oldest); 6045 napi_gro_complete(napi, oldest); 6046 } 6047 6048 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) 6049 { 6050 u32 bucket = skb_get_hash_raw(skb) & (GRO_HASH_BUCKETS - 1); 6051 struct gro_list *gro_list = &napi->gro_hash[bucket]; 6052 struct list_head *head = &offload_base; 6053 struct packet_offload *ptype; 6054 __be16 type = skb->protocol; 6055 struct sk_buff *pp = NULL; 6056 enum gro_result ret; 6057 int same_flow; 6058 int grow; 6059 6060 if (netif_elide_gro(skb->dev)) 6061 goto normal; 6062 6063 gro_list_prepare(&gro_list->list, skb); 6064 6065 rcu_read_lock(); 6066 list_for_each_entry_rcu(ptype, head, list) { 6067 if (ptype->type != type || !ptype->callbacks.gro_receive) 6068 continue; 6069 6070 skb_set_network_header(skb, skb_gro_offset(skb)); 6071 skb_reset_mac_len(skb); 6072 NAPI_GRO_CB(skb)->same_flow = 0; 6073 NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb); 6074 NAPI_GRO_CB(skb)->free = 0; 6075 NAPI_GRO_CB(skb)->encap_mark = 0; 6076 NAPI_GRO_CB(skb)->recursion_counter = 0; 6077 NAPI_GRO_CB(skb)->is_fou = 0; 6078 NAPI_GRO_CB(skb)->is_atomic = 1; 6079 NAPI_GRO_CB(skb)->gro_remcsum_start = 0; 6080 6081 /* Setup for GRO checksum validation */ 6082 switch (skb->ip_summed) { 6083 case CHECKSUM_COMPLETE: 6084 NAPI_GRO_CB(skb)->csum = skb->csum; 6085 NAPI_GRO_CB(skb)->csum_valid = 1; 6086 NAPI_GRO_CB(skb)->csum_cnt = 0; 6087 break; 6088 case CHECKSUM_UNNECESSARY: 6089 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1; 6090 NAPI_GRO_CB(skb)->csum_valid = 0; 6091 break; 6092 default: 6093 NAPI_GRO_CB(skb)->csum_cnt = 0; 6094 NAPI_GRO_CB(skb)->csum_valid = 0; 6095 } 6096 6097 pp = INDIRECT_CALL_INET(ptype->callbacks.gro_receive, 6098 ipv6_gro_receive, inet_gro_receive, 6099 &gro_list->list, skb); 6100 break; 6101 } 6102 rcu_read_unlock(); 6103 6104 if (&ptype->list == head) 6105 goto normal; 6106 6107 if (PTR_ERR(pp) == -EINPROGRESS) { 6108 ret = GRO_CONSUMED; 6109 goto ok; 6110 } 6111 6112 same_flow = NAPI_GRO_CB(skb)->same_flow; 6113 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; 6114 6115 if (pp) { 6116 skb_list_del_init(pp); 6117 napi_gro_complete(napi, pp); 6118 gro_list->count--; 6119 } 6120 6121 if (same_flow) 6122 goto ok; 6123 6124 if (NAPI_GRO_CB(skb)->flush) 6125 goto normal; 6126 6127 if (unlikely(gro_list->count >= MAX_GRO_SKBS)) 6128 gro_flush_oldest(napi, &gro_list->list); 6129 else 6130 gro_list->count++; 6131 6132 NAPI_GRO_CB(skb)->count = 1; 6133 NAPI_GRO_CB(skb)->age = jiffies; 6134 NAPI_GRO_CB(skb)->last = skb; 6135 skb_shinfo(skb)->gso_size = skb_gro_len(skb); 6136 list_add(&skb->list, &gro_list->list); 6137 ret = GRO_HELD; 6138 6139 pull: 6140 grow = skb_gro_offset(skb) - skb_headlen(skb); 6141 if (grow > 0) 6142 gro_pull_from_frag0(skb, grow); 6143 ok: 6144 if (gro_list->count) { 6145 if (!test_bit(bucket, &napi->gro_bitmask)) 6146 __set_bit(bucket, &napi->gro_bitmask); 6147 } else if (test_bit(bucket, &napi->gro_bitmask)) { 6148 __clear_bit(bucket, &napi->gro_bitmask); 6149 } 6150 6151 return ret; 6152 6153 normal: 6154 ret = GRO_NORMAL; 6155 goto pull; 6156 } 6157 6158 struct packet_offload *gro_find_receive_by_type(__be16 type) 6159 { 6160 struct list_head *offload_head = &offload_base; 6161 struct packet_offload *ptype; 6162 6163 list_for_each_entry_rcu(ptype, offload_head, list) { 6164 if (ptype->type != type || !ptype->callbacks.gro_receive) 6165 continue; 6166 return ptype; 6167 } 6168 return NULL; 6169 } 6170 EXPORT_SYMBOL(gro_find_receive_by_type); 6171 6172 struct packet_offload *gro_find_complete_by_type(__be16 type) 6173 { 6174 struct list_head *offload_head = &offload_base; 6175 struct packet_offload *ptype; 6176 6177 list_for_each_entry_rcu(ptype, offload_head, list) { 6178 if (ptype->type != type || !ptype->callbacks.gro_complete) 6179 continue; 6180 return ptype; 6181 } 6182 return NULL; 6183 } 6184 EXPORT_SYMBOL(gro_find_complete_by_type); 6185 6186 static gro_result_t napi_skb_finish(struct napi_struct *napi, 6187 struct sk_buff *skb, 6188 gro_result_t ret) 6189 { 6190 switch (ret) { 6191 case GRO_NORMAL: 6192 gro_normal_one(napi, skb, 1); 6193 break; 6194 6195 case GRO_MERGED_FREE: 6196 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) 6197 napi_skb_free_stolen_head(skb); 6198 else if (skb->fclone != SKB_FCLONE_UNAVAILABLE) 6199 __kfree_skb(skb); 6200 else 6201 __kfree_skb_defer(skb); 6202 break; 6203 6204 case GRO_HELD: 6205 case GRO_MERGED: 6206 case GRO_CONSUMED: 6207 break; 6208 } 6209 6210 return ret; 6211 } 6212 6213 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) 6214 { 6215 gro_result_t ret; 6216 6217 skb_mark_napi_id(skb, napi); 6218 trace_napi_gro_receive_entry(skb); 6219 6220 skb_gro_reset_offset(skb, 0); 6221 6222 ret = napi_skb_finish(napi, skb, dev_gro_receive(napi, skb)); 6223 trace_napi_gro_receive_exit(ret); 6224 6225 return ret; 6226 } 6227 EXPORT_SYMBOL(napi_gro_receive); 6228 6229 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) 6230 { 6231 if (unlikely(skb->pfmemalloc)) { 6232 consume_skb(skb); 6233 return; 6234 } 6235 __skb_pull(skb, skb_headlen(skb)); 6236 /* restore the reserve we had after netdev_alloc_skb_ip_align() */ 6237 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb)); 6238 __vlan_hwaccel_clear_tag(skb); 6239 skb->dev = napi->dev; 6240 skb->skb_iif = 0; 6241 6242 /* eth_type_trans() assumes pkt_type is PACKET_HOST */ 6243 skb->pkt_type = PACKET_HOST; 6244 6245 skb->encapsulation = 0; 6246 skb_shinfo(skb)->gso_type = 0; 6247 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); 6248 if (unlikely(skb->slow_gro)) { 6249 skb_orphan(skb); 6250 skb_ext_reset(skb); 6251 nf_reset_ct(skb); 6252 skb->slow_gro = 0; 6253 } 6254 6255 napi->skb = skb; 6256 } 6257 6258 struct sk_buff *napi_get_frags(struct napi_struct *napi) 6259 { 6260 struct sk_buff *skb = napi->skb; 6261 6262 if (!skb) { 6263 skb = napi_alloc_skb(napi, GRO_MAX_HEAD); 6264 if (skb) { 6265 napi->skb = skb; 6266 skb_mark_napi_id(skb, napi); 6267 } 6268 } 6269 return skb; 6270 } 6271 EXPORT_SYMBOL(napi_get_frags); 6272 6273 static gro_result_t napi_frags_finish(struct napi_struct *napi, 6274 struct sk_buff *skb, 6275 gro_result_t ret) 6276 { 6277 switch (ret) { 6278 case GRO_NORMAL: 6279 case GRO_HELD: 6280 __skb_push(skb, ETH_HLEN); 6281 skb->protocol = eth_type_trans(skb, skb->dev); 6282 if (ret == GRO_NORMAL) 6283 gro_normal_one(napi, skb, 1); 6284 break; 6285 6286 case GRO_MERGED_FREE: 6287 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) 6288 napi_skb_free_stolen_head(skb); 6289 else 6290 napi_reuse_skb(napi, skb); 6291 break; 6292 6293 case GRO_MERGED: 6294 case GRO_CONSUMED: 6295 break; 6296 } 6297 6298 return ret; 6299 } 6300 6301 /* Upper GRO stack assumes network header starts at gro_offset=0 6302 * Drivers could call both napi_gro_frags() and napi_gro_receive() 6303 * We copy ethernet header into skb->data to have a common layout. 6304 */ 6305 static struct sk_buff *napi_frags_skb(struct napi_struct *napi) 6306 { 6307 struct sk_buff *skb = napi->skb; 6308 const struct ethhdr *eth; 6309 unsigned int hlen = sizeof(*eth); 6310 6311 napi->skb = NULL; 6312 6313 skb_reset_mac_header(skb); 6314 skb_gro_reset_offset(skb, hlen); 6315 6316 if (unlikely(skb_gro_header_hard(skb, hlen))) { 6317 eth = skb_gro_header_slow(skb, hlen, 0); 6318 if (unlikely(!eth)) { 6319 net_warn_ratelimited("%s: dropping impossible skb from %s\n", 6320 __func__, napi->dev->name); 6321 napi_reuse_skb(napi, skb); 6322 return NULL; 6323 } 6324 } else { 6325 eth = (const struct ethhdr *)skb->data; 6326 gro_pull_from_frag0(skb, hlen); 6327 NAPI_GRO_CB(skb)->frag0 += hlen; 6328 NAPI_GRO_CB(skb)->frag0_len -= hlen; 6329 } 6330 __skb_pull(skb, hlen); 6331 6332 /* 6333 * This works because the only protocols we care about don't require 6334 * special handling. 6335 * We'll fix it up properly in napi_frags_finish() 6336 */ 6337 skb->protocol = eth->h_proto; 6338 6339 return skb; 6340 } 6341 6342 gro_result_t napi_gro_frags(struct napi_struct *napi) 6343 { 6344 gro_result_t ret; 6345 struct sk_buff *skb = napi_frags_skb(napi); 6346 6347 trace_napi_gro_frags_entry(skb); 6348 6349 ret = napi_frags_finish(napi, skb, dev_gro_receive(napi, skb)); 6350 trace_napi_gro_frags_exit(ret); 6351 6352 return ret; 6353 } 6354 EXPORT_SYMBOL(napi_gro_frags); 6355 6356 /* Compute the checksum from gro_offset and return the folded value 6357 * after adding in any pseudo checksum. 6358 */ 6359 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb) 6360 { 6361 __wsum wsum; 6362 __sum16 sum; 6363 6364 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0); 6365 6366 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */ 6367 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum)); 6368 /* See comments in __skb_checksum_complete(). */ 6369 if (likely(!sum)) { 6370 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && 6371 !skb->csum_complete_sw) 6372 netdev_rx_csum_fault(skb->dev, skb); 6373 } 6374 6375 NAPI_GRO_CB(skb)->csum = wsum; 6376 NAPI_GRO_CB(skb)->csum_valid = 1; 6377 6378 return sum; 6379 } 6380 EXPORT_SYMBOL(__skb_gro_checksum_complete); 6381 6382 static void net_rps_send_ipi(struct softnet_data *remsd) 6383 { 6384 #ifdef CONFIG_RPS 6385 while (remsd) { 6386 struct softnet_data *next = remsd->rps_ipi_next; 6387 6388 if (cpu_online(remsd->cpu)) 6389 smp_call_function_single_async(remsd->cpu, &remsd->csd); 6390 remsd = next; 6391 } 6392 #endif 6393 } 6394 6395 /* 6396 * net_rps_action_and_irq_enable sends any pending IPI's for rps. 6397 * Note: called with local irq disabled, but exits with local irq enabled. 6398 */ 6399 static void net_rps_action_and_irq_enable(struct softnet_data *sd) 6400 { 6401 #ifdef CONFIG_RPS 6402 struct softnet_data *remsd = sd->rps_ipi_list; 6403 6404 if (remsd) { 6405 sd->rps_ipi_list = NULL; 6406 6407 local_irq_enable(); 6408 6409 /* Send pending IPI's to kick RPS processing on remote cpus. */ 6410 net_rps_send_ipi(remsd); 6411 } else 6412 #endif 6413 local_irq_enable(); 6414 } 6415 6416 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) 6417 { 6418 #ifdef CONFIG_RPS 6419 return sd->rps_ipi_list != NULL; 6420 #else 6421 return false; 6422 #endif 6423 } 6424 6425 static int process_backlog(struct napi_struct *napi, int quota) 6426 { 6427 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); 6428 bool again = true; 6429 int work = 0; 6430 6431 /* Check if we have pending ipi, its better to send them now, 6432 * not waiting net_rx_action() end. 6433 */ 6434 if (sd_has_rps_ipi_waiting(sd)) { 6435 local_irq_disable(); 6436 net_rps_action_and_irq_enable(sd); 6437 } 6438 6439 napi->weight = dev_rx_weight; 6440 while (again) { 6441 struct sk_buff *skb; 6442 6443 while ((skb = __skb_dequeue(&sd->process_queue))) { 6444 rcu_read_lock(); 6445 __netif_receive_skb(skb); 6446 rcu_read_unlock(); 6447 input_queue_head_incr(sd); 6448 if (++work >= quota) 6449 return work; 6450 6451 } 6452 6453 local_irq_disable(); 6454 rps_lock(sd); 6455 if (skb_queue_empty(&sd->input_pkt_queue)) { 6456 /* 6457 * Inline a custom version of __napi_complete(). 6458 * only current cpu owns and manipulates this napi, 6459 * and NAPI_STATE_SCHED is the only possible flag set 6460 * on backlog. 6461 * We can use a plain write instead of clear_bit(), 6462 * and we dont need an smp_mb() memory barrier. 6463 */ 6464 napi->state = 0; 6465 again = false; 6466 } else { 6467 skb_queue_splice_tail_init(&sd->input_pkt_queue, 6468 &sd->process_queue); 6469 } 6470 rps_unlock(sd); 6471 local_irq_enable(); 6472 } 6473 6474 return work; 6475 } 6476 6477 /** 6478 * __napi_schedule - schedule for receive 6479 * @n: entry to schedule 6480 * 6481 * The entry's receive function will be scheduled to run. 6482 * Consider using __napi_schedule_irqoff() if hard irqs are masked. 6483 */ 6484 void __napi_schedule(struct napi_struct *n) 6485 { 6486 unsigned long flags; 6487 6488 local_irq_save(flags); 6489 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6490 local_irq_restore(flags); 6491 } 6492 EXPORT_SYMBOL(__napi_schedule); 6493 6494 /** 6495 * napi_schedule_prep - check if napi can be scheduled 6496 * @n: napi context 6497 * 6498 * Test if NAPI routine is already running, and if not mark 6499 * it as running. This is used as a condition variable to 6500 * insure only one NAPI poll instance runs. We also make 6501 * sure there is no pending NAPI disable. 6502 */ 6503 bool napi_schedule_prep(struct napi_struct *n) 6504 { 6505 unsigned long val, new; 6506 6507 do { 6508 val = READ_ONCE(n->state); 6509 if (unlikely(val & NAPIF_STATE_DISABLE)) 6510 return false; 6511 new = val | NAPIF_STATE_SCHED; 6512 6513 /* Sets STATE_MISSED bit if STATE_SCHED was already set 6514 * This was suggested by Alexander Duyck, as compiler 6515 * emits better code than : 6516 * if (val & NAPIF_STATE_SCHED) 6517 * new |= NAPIF_STATE_MISSED; 6518 */ 6519 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * 6520 NAPIF_STATE_MISSED; 6521 } while (cmpxchg(&n->state, val, new) != val); 6522 6523 return !(val & NAPIF_STATE_SCHED); 6524 } 6525 EXPORT_SYMBOL(napi_schedule_prep); 6526 6527 /** 6528 * __napi_schedule_irqoff - schedule for receive 6529 * @n: entry to schedule 6530 * 6531 * Variant of __napi_schedule() assuming hard irqs are masked. 6532 * 6533 * On PREEMPT_RT enabled kernels this maps to __napi_schedule() 6534 * because the interrupt disabled assumption might not be true 6535 * due to force-threaded interrupts and spinlock substitution. 6536 */ 6537 void __napi_schedule_irqoff(struct napi_struct *n) 6538 { 6539 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6540 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6541 else 6542 __napi_schedule(n); 6543 } 6544 EXPORT_SYMBOL(__napi_schedule_irqoff); 6545 6546 bool napi_complete_done(struct napi_struct *n, int work_done) 6547 { 6548 unsigned long flags, val, new, timeout = 0; 6549 bool ret = true; 6550 6551 /* 6552 * 1) Don't let napi dequeue from the cpu poll list 6553 * just in case its running on a different cpu. 6554 * 2) If we are busy polling, do nothing here, we have 6555 * the guarantee we will be called later. 6556 */ 6557 if (unlikely(n->state & (NAPIF_STATE_NPSVC | 6558 NAPIF_STATE_IN_BUSY_POLL))) 6559 return false; 6560 6561 if (work_done) { 6562 if (n->gro_bitmask) 6563 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6564 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs); 6565 } 6566 if (n->defer_hard_irqs_count > 0) { 6567 n->defer_hard_irqs_count--; 6568 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6569 if (timeout) 6570 ret = false; 6571 } 6572 if (n->gro_bitmask) { 6573 /* When the NAPI instance uses a timeout and keeps postponing 6574 * it, we need to bound somehow the time packets are kept in 6575 * the GRO layer 6576 */ 6577 napi_gro_flush(n, !!timeout); 6578 } 6579 6580 gro_normal_list(n); 6581 6582 if (unlikely(!list_empty(&n->poll_list))) { 6583 /* If n->poll_list is not empty, we need to mask irqs */ 6584 local_irq_save(flags); 6585 list_del_init(&n->poll_list); 6586 local_irq_restore(flags); 6587 } 6588 6589 do { 6590 val = READ_ONCE(n->state); 6591 6592 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); 6593 6594 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED | 6595 NAPIF_STATE_SCHED_THREADED | 6596 NAPIF_STATE_PREFER_BUSY_POLL); 6597 6598 /* If STATE_MISSED was set, leave STATE_SCHED set, 6599 * because we will call napi->poll() one more time. 6600 * This C code was suggested by Alexander Duyck to help gcc. 6601 */ 6602 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * 6603 NAPIF_STATE_SCHED; 6604 } while (cmpxchg(&n->state, val, new) != val); 6605 6606 if (unlikely(val & NAPIF_STATE_MISSED)) { 6607 __napi_schedule(n); 6608 return false; 6609 } 6610 6611 if (timeout) 6612 hrtimer_start(&n->timer, ns_to_ktime(timeout), 6613 HRTIMER_MODE_REL_PINNED); 6614 return ret; 6615 } 6616 EXPORT_SYMBOL(napi_complete_done); 6617 6618 /* must be called under rcu_read_lock(), as we dont take a reference */ 6619 static struct napi_struct *napi_by_id(unsigned int napi_id) 6620 { 6621 unsigned int hash = napi_id % HASH_SIZE(napi_hash); 6622 struct napi_struct *napi; 6623 6624 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) 6625 if (napi->napi_id == napi_id) 6626 return napi; 6627 6628 return NULL; 6629 } 6630 6631 #if defined(CONFIG_NET_RX_BUSY_POLL) 6632 6633 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule) 6634 { 6635 if (!skip_schedule) { 6636 gro_normal_list(napi); 6637 __napi_schedule(napi); 6638 return; 6639 } 6640 6641 if (napi->gro_bitmask) { 6642 /* flush too old packets 6643 * If HZ < 1000, flush all packets. 6644 */ 6645 napi_gro_flush(napi, HZ >= 1000); 6646 } 6647 6648 gro_normal_list(napi); 6649 clear_bit(NAPI_STATE_SCHED, &napi->state); 6650 } 6651 6652 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll, 6653 u16 budget) 6654 { 6655 bool skip_schedule = false; 6656 unsigned long timeout; 6657 int rc; 6658 6659 /* Busy polling means there is a high chance device driver hard irq 6660 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was 6661 * set in napi_schedule_prep(). 6662 * Since we are about to call napi->poll() once more, we can safely 6663 * clear NAPI_STATE_MISSED. 6664 * 6665 * Note: x86 could use a single "lock and ..." instruction 6666 * to perform these two clear_bit() 6667 */ 6668 clear_bit(NAPI_STATE_MISSED, &napi->state); 6669 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); 6670 6671 local_bh_disable(); 6672 6673 if (prefer_busy_poll) { 6674 napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs); 6675 timeout = READ_ONCE(napi->dev->gro_flush_timeout); 6676 if (napi->defer_hard_irqs_count && timeout) { 6677 hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED); 6678 skip_schedule = true; 6679 } 6680 } 6681 6682 /* All we really want here is to re-enable device interrupts. 6683 * Ideally, a new ndo_busy_poll_stop() could avoid another round. 6684 */ 6685 rc = napi->poll(napi, budget); 6686 /* We can't gro_normal_list() here, because napi->poll() might have 6687 * rearmed the napi (napi_complete_done()) in which case it could 6688 * already be running on another CPU. 6689 */ 6690 trace_napi_poll(napi, rc, budget); 6691 netpoll_poll_unlock(have_poll_lock); 6692 if (rc == budget) 6693 __busy_poll_stop(napi, skip_schedule); 6694 local_bh_enable(); 6695 } 6696 6697 void napi_busy_loop(unsigned int napi_id, 6698 bool (*loop_end)(void *, unsigned long), 6699 void *loop_end_arg, bool prefer_busy_poll, u16 budget) 6700 { 6701 unsigned long start_time = loop_end ? busy_loop_current_time() : 0; 6702 int (*napi_poll)(struct napi_struct *napi, int budget); 6703 void *have_poll_lock = NULL; 6704 struct napi_struct *napi; 6705 6706 restart: 6707 napi_poll = NULL; 6708 6709 rcu_read_lock(); 6710 6711 napi = napi_by_id(napi_id); 6712 if (!napi) 6713 goto out; 6714 6715 preempt_disable(); 6716 for (;;) { 6717 int work = 0; 6718 6719 local_bh_disable(); 6720 if (!napi_poll) { 6721 unsigned long val = READ_ONCE(napi->state); 6722 6723 /* If multiple threads are competing for this napi, 6724 * we avoid dirtying napi->state as much as we can. 6725 */ 6726 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | 6727 NAPIF_STATE_IN_BUSY_POLL)) { 6728 if (prefer_busy_poll) 6729 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6730 goto count; 6731 } 6732 if (cmpxchg(&napi->state, val, 6733 val | NAPIF_STATE_IN_BUSY_POLL | 6734 NAPIF_STATE_SCHED) != val) { 6735 if (prefer_busy_poll) 6736 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6737 goto count; 6738 } 6739 have_poll_lock = netpoll_poll_lock(napi); 6740 napi_poll = napi->poll; 6741 } 6742 work = napi_poll(napi, budget); 6743 trace_napi_poll(napi, work, budget); 6744 gro_normal_list(napi); 6745 count: 6746 if (work > 0) 6747 __NET_ADD_STATS(dev_net(napi->dev), 6748 LINUX_MIB_BUSYPOLLRXPACKETS, work); 6749 local_bh_enable(); 6750 6751 if (!loop_end || loop_end(loop_end_arg, start_time)) 6752 break; 6753 6754 if (unlikely(need_resched())) { 6755 if (napi_poll) 6756 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6757 preempt_enable(); 6758 rcu_read_unlock(); 6759 cond_resched(); 6760 if (loop_end(loop_end_arg, start_time)) 6761 return; 6762 goto restart; 6763 } 6764 cpu_relax(); 6765 } 6766 if (napi_poll) 6767 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6768 preempt_enable(); 6769 out: 6770 rcu_read_unlock(); 6771 } 6772 EXPORT_SYMBOL(napi_busy_loop); 6773 6774 #endif /* CONFIG_NET_RX_BUSY_POLL */ 6775 6776 static void napi_hash_add(struct napi_struct *napi) 6777 { 6778 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state)) 6779 return; 6780 6781 spin_lock(&napi_hash_lock); 6782 6783 /* 0..NR_CPUS range is reserved for sender_cpu use */ 6784 do { 6785 if (unlikely(++napi_gen_id < MIN_NAPI_ID)) 6786 napi_gen_id = MIN_NAPI_ID; 6787 } while (napi_by_id(napi_gen_id)); 6788 napi->napi_id = napi_gen_id; 6789 6790 hlist_add_head_rcu(&napi->napi_hash_node, 6791 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); 6792 6793 spin_unlock(&napi_hash_lock); 6794 } 6795 6796 /* Warning : caller is responsible to make sure rcu grace period 6797 * is respected before freeing memory containing @napi 6798 */ 6799 static void napi_hash_del(struct napi_struct *napi) 6800 { 6801 spin_lock(&napi_hash_lock); 6802 6803 hlist_del_init_rcu(&napi->napi_hash_node); 6804 6805 spin_unlock(&napi_hash_lock); 6806 } 6807 6808 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) 6809 { 6810 struct napi_struct *napi; 6811 6812 napi = container_of(timer, struct napi_struct, timer); 6813 6814 /* Note : we use a relaxed variant of napi_schedule_prep() not setting 6815 * NAPI_STATE_MISSED, since we do not react to a device IRQ. 6816 */ 6817 if (!napi_disable_pending(napi) && 6818 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) { 6819 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6820 __napi_schedule_irqoff(napi); 6821 } 6822 6823 return HRTIMER_NORESTART; 6824 } 6825 6826 static void init_gro_hash(struct napi_struct *napi) 6827 { 6828 int i; 6829 6830 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6831 INIT_LIST_HEAD(&napi->gro_hash[i].list); 6832 napi->gro_hash[i].count = 0; 6833 } 6834 napi->gro_bitmask = 0; 6835 } 6836 6837 int dev_set_threaded(struct net_device *dev, bool threaded) 6838 { 6839 struct napi_struct *napi; 6840 int err = 0; 6841 6842 if (dev->threaded == threaded) 6843 return 0; 6844 6845 if (threaded) { 6846 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6847 if (!napi->thread) { 6848 err = napi_kthread_create(napi); 6849 if (err) { 6850 threaded = false; 6851 break; 6852 } 6853 } 6854 } 6855 } 6856 6857 dev->threaded = threaded; 6858 6859 /* Make sure kthread is created before THREADED bit 6860 * is set. 6861 */ 6862 smp_mb__before_atomic(); 6863 6864 /* Setting/unsetting threaded mode on a napi might not immediately 6865 * take effect, if the current napi instance is actively being 6866 * polled. In this case, the switch between threaded mode and 6867 * softirq mode will happen in the next round of napi_schedule(). 6868 * This should not cause hiccups/stalls to the live traffic. 6869 */ 6870 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6871 if (threaded) 6872 set_bit(NAPI_STATE_THREADED, &napi->state); 6873 else 6874 clear_bit(NAPI_STATE_THREADED, &napi->state); 6875 } 6876 6877 return err; 6878 } 6879 EXPORT_SYMBOL(dev_set_threaded); 6880 6881 void netif_napi_add(struct net_device *dev, struct napi_struct *napi, 6882 int (*poll)(struct napi_struct *, int), int weight) 6883 { 6884 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state))) 6885 return; 6886 6887 INIT_LIST_HEAD(&napi->poll_list); 6888 INIT_HLIST_NODE(&napi->napi_hash_node); 6889 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); 6890 napi->timer.function = napi_watchdog; 6891 init_gro_hash(napi); 6892 napi->skb = NULL; 6893 INIT_LIST_HEAD(&napi->rx_list); 6894 napi->rx_count = 0; 6895 napi->poll = poll; 6896 if (weight > NAPI_POLL_WEIGHT) 6897 netdev_err_once(dev, "%s() called with weight %d\n", __func__, 6898 weight); 6899 napi->weight = weight; 6900 napi->dev = dev; 6901 #ifdef CONFIG_NETPOLL 6902 napi->poll_owner = -1; 6903 #endif 6904 set_bit(NAPI_STATE_SCHED, &napi->state); 6905 set_bit(NAPI_STATE_NPSVC, &napi->state); 6906 list_add_rcu(&napi->dev_list, &dev->napi_list); 6907 napi_hash_add(napi); 6908 /* Create kthread for this napi if dev->threaded is set. 6909 * Clear dev->threaded if kthread creation failed so that 6910 * threaded mode will not be enabled in napi_enable(). 6911 */ 6912 if (dev->threaded && napi_kthread_create(napi)) 6913 dev->threaded = 0; 6914 } 6915 EXPORT_SYMBOL(netif_napi_add); 6916 6917 void napi_disable(struct napi_struct *n) 6918 { 6919 unsigned long val, new; 6920 6921 might_sleep(); 6922 set_bit(NAPI_STATE_DISABLE, &n->state); 6923 6924 do { 6925 val = READ_ONCE(n->state); 6926 if (val & (NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC)) { 6927 usleep_range(20, 200); 6928 continue; 6929 } 6930 6931 new = val | NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC; 6932 new &= ~(NAPIF_STATE_THREADED | NAPIF_STATE_PREFER_BUSY_POLL); 6933 } while (cmpxchg(&n->state, val, new) != val); 6934 6935 hrtimer_cancel(&n->timer); 6936 6937 clear_bit(NAPI_STATE_DISABLE, &n->state); 6938 } 6939 EXPORT_SYMBOL(napi_disable); 6940 6941 /** 6942 * napi_enable - enable NAPI scheduling 6943 * @n: NAPI context 6944 * 6945 * Resume NAPI from being scheduled on this context. 6946 * Must be paired with napi_disable. 6947 */ 6948 void napi_enable(struct napi_struct *n) 6949 { 6950 unsigned long val, new; 6951 6952 do { 6953 val = READ_ONCE(n->state); 6954 BUG_ON(!test_bit(NAPI_STATE_SCHED, &val)); 6955 6956 new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC); 6957 if (n->dev->threaded && n->thread) 6958 new |= NAPIF_STATE_THREADED; 6959 } while (cmpxchg(&n->state, val, new) != val); 6960 } 6961 EXPORT_SYMBOL(napi_enable); 6962 6963 static void flush_gro_hash(struct napi_struct *napi) 6964 { 6965 int i; 6966 6967 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6968 struct sk_buff *skb, *n; 6969 6970 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list) 6971 kfree_skb(skb); 6972 napi->gro_hash[i].count = 0; 6973 } 6974 } 6975 6976 /* Must be called in process context */ 6977 void __netif_napi_del(struct napi_struct *napi) 6978 { 6979 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state)) 6980 return; 6981 6982 napi_hash_del(napi); 6983 list_del_rcu(&napi->dev_list); 6984 napi_free_frags(napi); 6985 6986 flush_gro_hash(napi); 6987 napi->gro_bitmask = 0; 6988 6989 if (napi->thread) { 6990 kthread_stop(napi->thread); 6991 napi->thread = NULL; 6992 } 6993 } 6994 EXPORT_SYMBOL(__netif_napi_del); 6995 6996 static int __napi_poll(struct napi_struct *n, bool *repoll) 6997 { 6998 int work, weight; 6999 7000 weight = n->weight; 7001 7002 /* This NAPI_STATE_SCHED test is for avoiding a race 7003 * with netpoll's poll_napi(). Only the entity which 7004 * obtains the lock and sees NAPI_STATE_SCHED set will 7005 * actually make the ->poll() call. Therefore we avoid 7006 * accidentally calling ->poll() when NAPI is not scheduled. 7007 */ 7008 work = 0; 7009 if (test_bit(NAPI_STATE_SCHED, &n->state)) { 7010 work = n->poll(n, weight); 7011 trace_napi_poll(n, work, weight); 7012 } 7013 7014 if (unlikely(work > weight)) 7015 netdev_err_once(n->dev, "NAPI poll function %pS returned %d, exceeding its budget of %d.\n", 7016 n->poll, work, weight); 7017 7018 if (likely(work < weight)) 7019 return work; 7020 7021 /* Drivers must not modify the NAPI state if they 7022 * consume the entire weight. In such cases this code 7023 * still "owns" the NAPI instance and therefore can 7024 * move the instance around on the list at-will. 7025 */ 7026 if (unlikely(napi_disable_pending(n))) { 7027 napi_complete(n); 7028 return work; 7029 } 7030 7031 /* The NAPI context has more processing work, but busy-polling 7032 * is preferred. Exit early. 7033 */ 7034 if (napi_prefer_busy_poll(n)) { 7035 if (napi_complete_done(n, work)) { 7036 /* If timeout is not set, we need to make sure 7037 * that the NAPI is re-scheduled. 7038 */ 7039 napi_schedule(n); 7040 } 7041 return work; 7042 } 7043 7044 if (n->gro_bitmask) { 7045 /* flush too old packets 7046 * If HZ < 1000, flush all packets. 7047 */ 7048 napi_gro_flush(n, HZ >= 1000); 7049 } 7050 7051 gro_normal_list(n); 7052 7053 /* Some drivers may have called napi_schedule 7054 * prior to exhausting their budget. 7055 */ 7056 if (unlikely(!list_empty(&n->poll_list))) { 7057 pr_warn_once("%s: Budget exhausted after napi rescheduled\n", 7058 n->dev ? n->dev->name : "backlog"); 7059 return work; 7060 } 7061 7062 *repoll = true; 7063 7064 return work; 7065 } 7066 7067 static int napi_poll(struct napi_struct *n, struct list_head *repoll) 7068 { 7069 bool do_repoll = false; 7070 void *have; 7071 int work; 7072 7073 list_del_init(&n->poll_list); 7074 7075 have = netpoll_poll_lock(n); 7076 7077 work = __napi_poll(n, &do_repoll); 7078 7079 if (do_repoll) 7080 list_add_tail(&n->poll_list, repoll); 7081 7082 netpoll_poll_unlock(have); 7083 7084 return work; 7085 } 7086 7087 static int napi_thread_wait(struct napi_struct *napi) 7088 { 7089 bool woken = false; 7090 7091 set_current_state(TASK_INTERRUPTIBLE); 7092 7093 while (!kthread_should_stop()) { 7094 /* Testing SCHED_THREADED bit here to make sure the current 7095 * kthread owns this napi and could poll on this napi. 7096 * Testing SCHED bit is not enough because SCHED bit might be 7097 * set by some other busy poll thread or by napi_disable(). 7098 */ 7099 if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state) || woken) { 7100 WARN_ON(!list_empty(&napi->poll_list)); 7101 __set_current_state(TASK_RUNNING); 7102 return 0; 7103 } 7104 7105 schedule(); 7106 /* woken being true indicates this thread owns this napi. */ 7107 woken = true; 7108 set_current_state(TASK_INTERRUPTIBLE); 7109 } 7110 __set_current_state(TASK_RUNNING); 7111 7112 return -1; 7113 } 7114 7115 static int napi_threaded_poll(void *data) 7116 { 7117 struct napi_struct *napi = data; 7118 void *have; 7119 7120 while (!napi_thread_wait(napi)) { 7121 for (;;) { 7122 bool repoll = false; 7123 7124 local_bh_disable(); 7125 7126 have = netpoll_poll_lock(napi); 7127 __napi_poll(napi, &repoll); 7128 netpoll_poll_unlock(have); 7129 7130 local_bh_enable(); 7131 7132 if (!repoll) 7133 break; 7134 7135 cond_resched(); 7136 } 7137 } 7138 return 0; 7139 } 7140 7141 static __latent_entropy void net_rx_action(struct softirq_action *h) 7142 { 7143 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 7144 unsigned long time_limit = jiffies + 7145 usecs_to_jiffies(netdev_budget_usecs); 7146 int budget = netdev_budget; 7147 LIST_HEAD(list); 7148 LIST_HEAD(repoll); 7149 7150 local_irq_disable(); 7151 list_splice_init(&sd->poll_list, &list); 7152 local_irq_enable(); 7153 7154 for (;;) { 7155 struct napi_struct *n; 7156 7157 if (list_empty(&list)) { 7158 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll)) 7159 return; 7160 break; 7161 } 7162 7163 n = list_first_entry(&list, struct napi_struct, poll_list); 7164 budget -= napi_poll(n, &repoll); 7165 7166 /* If softirq window is exhausted then punt. 7167 * Allow this to run for 2 jiffies since which will allow 7168 * an average latency of 1.5/HZ. 7169 */ 7170 if (unlikely(budget <= 0 || 7171 time_after_eq(jiffies, time_limit))) { 7172 sd->time_squeeze++; 7173 break; 7174 } 7175 } 7176 7177 local_irq_disable(); 7178 7179 list_splice_tail_init(&sd->poll_list, &list); 7180 list_splice_tail(&repoll, &list); 7181 list_splice(&list, &sd->poll_list); 7182 if (!list_empty(&sd->poll_list)) 7183 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 7184 7185 net_rps_action_and_irq_enable(sd); 7186 } 7187 7188 struct netdev_adjacent { 7189 struct net_device *dev; 7190 7191 /* upper master flag, there can only be one master device per list */ 7192 bool master; 7193 7194 /* lookup ignore flag */ 7195 bool ignore; 7196 7197 /* counter for the number of times this device was added to us */ 7198 u16 ref_nr; 7199 7200 /* private field for the users */ 7201 void *private; 7202 7203 struct list_head list; 7204 struct rcu_head rcu; 7205 }; 7206 7207 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, 7208 struct list_head *adj_list) 7209 { 7210 struct netdev_adjacent *adj; 7211 7212 list_for_each_entry(adj, adj_list, list) { 7213 if (adj->dev == adj_dev) 7214 return adj; 7215 } 7216 return NULL; 7217 } 7218 7219 static int ____netdev_has_upper_dev(struct net_device *upper_dev, 7220 struct netdev_nested_priv *priv) 7221 { 7222 struct net_device *dev = (struct net_device *)priv->data; 7223 7224 return upper_dev == dev; 7225 } 7226 7227 /** 7228 * netdev_has_upper_dev - Check if device is linked to an upper device 7229 * @dev: device 7230 * @upper_dev: upper device to check 7231 * 7232 * Find out if a device is linked to specified upper device and return true 7233 * in case it is. Note that this checks only immediate upper device, 7234 * not through a complete stack of devices. The caller must hold the RTNL lock. 7235 */ 7236 bool netdev_has_upper_dev(struct net_device *dev, 7237 struct net_device *upper_dev) 7238 { 7239 struct netdev_nested_priv priv = { 7240 .data = (void *)upper_dev, 7241 }; 7242 7243 ASSERT_RTNL(); 7244 7245 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 7246 &priv); 7247 } 7248 EXPORT_SYMBOL(netdev_has_upper_dev); 7249 7250 /** 7251 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device 7252 * @dev: device 7253 * @upper_dev: upper device to check 7254 * 7255 * Find out if a device is linked to specified upper device and return true 7256 * in case it is. Note that this checks the entire upper device chain. 7257 * The caller must hold rcu lock. 7258 */ 7259 7260 bool netdev_has_upper_dev_all_rcu(struct net_device *dev, 7261 struct net_device *upper_dev) 7262 { 7263 struct netdev_nested_priv priv = { 7264 .data = (void *)upper_dev, 7265 }; 7266 7267 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 7268 &priv); 7269 } 7270 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); 7271 7272 /** 7273 * netdev_has_any_upper_dev - Check if device is linked to some device 7274 * @dev: device 7275 * 7276 * Find out if a device is linked to an upper device and return true in case 7277 * it is. The caller must hold the RTNL lock. 7278 */ 7279 bool netdev_has_any_upper_dev(struct net_device *dev) 7280 { 7281 ASSERT_RTNL(); 7282 7283 return !list_empty(&dev->adj_list.upper); 7284 } 7285 EXPORT_SYMBOL(netdev_has_any_upper_dev); 7286 7287 /** 7288 * netdev_master_upper_dev_get - Get master upper device 7289 * @dev: device 7290 * 7291 * Find a master upper device and return pointer to it or NULL in case 7292 * it's not there. The caller must hold the RTNL lock. 7293 */ 7294 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)) 7306 return upper->dev; 7307 return NULL; 7308 } 7309 EXPORT_SYMBOL(netdev_master_upper_dev_get); 7310 7311 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev) 7312 { 7313 struct netdev_adjacent *upper; 7314 7315 ASSERT_RTNL(); 7316 7317 if (list_empty(&dev->adj_list.upper)) 7318 return NULL; 7319 7320 upper = list_first_entry(&dev->adj_list.upper, 7321 struct netdev_adjacent, list); 7322 if (likely(upper->master) && !upper->ignore) 7323 return upper->dev; 7324 return NULL; 7325 } 7326 7327 /** 7328 * netdev_has_any_lower_dev - Check if device is linked to some device 7329 * @dev: device 7330 * 7331 * Find out if a device is linked to a lower device and return true in case 7332 * it is. The caller must hold the RTNL lock. 7333 */ 7334 static bool netdev_has_any_lower_dev(struct net_device *dev) 7335 { 7336 ASSERT_RTNL(); 7337 7338 return !list_empty(&dev->adj_list.lower); 7339 } 7340 7341 void *netdev_adjacent_get_private(struct list_head *adj_list) 7342 { 7343 struct netdev_adjacent *adj; 7344 7345 adj = list_entry(adj_list, struct netdev_adjacent, list); 7346 7347 return adj->private; 7348 } 7349 EXPORT_SYMBOL(netdev_adjacent_get_private); 7350 7351 /** 7352 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list 7353 * @dev: device 7354 * @iter: list_head ** of the current position 7355 * 7356 * Gets the next device from the dev's upper list, starting from iter 7357 * position. The caller must hold RCU read lock. 7358 */ 7359 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, 7360 struct list_head **iter) 7361 { 7362 struct netdev_adjacent *upper; 7363 7364 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 7365 7366 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7367 7368 if (&upper->list == &dev->adj_list.upper) 7369 return NULL; 7370 7371 *iter = &upper->list; 7372 7373 return upper->dev; 7374 } 7375 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); 7376 7377 static struct net_device *__netdev_next_upper_dev(struct net_device *dev, 7378 struct list_head **iter, 7379 bool *ignore) 7380 { 7381 struct netdev_adjacent *upper; 7382 7383 upper = list_entry((*iter)->next, struct netdev_adjacent, list); 7384 7385 if (&upper->list == &dev->adj_list.upper) 7386 return NULL; 7387 7388 *iter = &upper->list; 7389 *ignore = upper->ignore; 7390 7391 return upper->dev; 7392 } 7393 7394 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, 7395 struct list_head **iter) 7396 { 7397 struct netdev_adjacent *upper; 7398 7399 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 7400 7401 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7402 7403 if (&upper->list == &dev->adj_list.upper) 7404 return NULL; 7405 7406 *iter = &upper->list; 7407 7408 return upper->dev; 7409 } 7410 7411 static int __netdev_walk_all_upper_dev(struct net_device *dev, 7412 int (*fn)(struct net_device *dev, 7413 struct netdev_nested_priv *priv), 7414 struct netdev_nested_priv *priv) 7415 { 7416 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7417 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7418 int ret, cur = 0; 7419 bool ignore; 7420 7421 now = dev; 7422 iter = &dev->adj_list.upper; 7423 7424 while (1) { 7425 if (now != dev) { 7426 ret = fn(now, priv); 7427 if (ret) 7428 return ret; 7429 } 7430 7431 next = NULL; 7432 while (1) { 7433 udev = __netdev_next_upper_dev(now, &iter, &ignore); 7434 if (!udev) 7435 break; 7436 if (ignore) 7437 continue; 7438 7439 next = udev; 7440 niter = &udev->adj_list.upper; 7441 dev_stack[cur] = now; 7442 iter_stack[cur++] = iter; 7443 break; 7444 } 7445 7446 if (!next) { 7447 if (!cur) 7448 return 0; 7449 next = dev_stack[--cur]; 7450 niter = iter_stack[cur]; 7451 } 7452 7453 now = next; 7454 iter = niter; 7455 } 7456 7457 return 0; 7458 } 7459 7460 int netdev_walk_all_upper_dev_rcu(struct net_device *dev, 7461 int (*fn)(struct net_device *dev, 7462 struct netdev_nested_priv *priv), 7463 struct netdev_nested_priv *priv) 7464 { 7465 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7466 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7467 int ret, cur = 0; 7468 7469 now = dev; 7470 iter = &dev->adj_list.upper; 7471 7472 while (1) { 7473 if (now != dev) { 7474 ret = fn(now, priv); 7475 if (ret) 7476 return ret; 7477 } 7478 7479 next = NULL; 7480 while (1) { 7481 udev = netdev_next_upper_dev_rcu(now, &iter); 7482 if (!udev) 7483 break; 7484 7485 next = udev; 7486 niter = &udev->adj_list.upper; 7487 dev_stack[cur] = now; 7488 iter_stack[cur++] = iter; 7489 break; 7490 } 7491 7492 if (!next) { 7493 if (!cur) 7494 return 0; 7495 next = dev_stack[--cur]; 7496 niter = iter_stack[cur]; 7497 } 7498 7499 now = next; 7500 iter = niter; 7501 } 7502 7503 return 0; 7504 } 7505 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); 7506 7507 static bool __netdev_has_upper_dev(struct net_device *dev, 7508 struct net_device *upper_dev) 7509 { 7510 struct netdev_nested_priv priv = { 7511 .flags = 0, 7512 .data = (void *)upper_dev, 7513 }; 7514 7515 ASSERT_RTNL(); 7516 7517 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev, 7518 &priv); 7519 } 7520 7521 /** 7522 * netdev_lower_get_next_private - Get the next ->private from the 7523 * lower neighbour list 7524 * @dev: device 7525 * @iter: list_head ** of the current position 7526 * 7527 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7528 * list, starting from iter position. The caller must hold either hold the 7529 * RTNL lock or its own locking that guarantees that the neighbour lower 7530 * list will remain unchanged. 7531 */ 7532 void *netdev_lower_get_next_private(struct net_device *dev, 7533 struct list_head **iter) 7534 { 7535 struct netdev_adjacent *lower; 7536 7537 lower = list_entry(*iter, struct netdev_adjacent, list); 7538 7539 if (&lower->list == &dev->adj_list.lower) 7540 return NULL; 7541 7542 *iter = lower->list.next; 7543 7544 return lower->private; 7545 } 7546 EXPORT_SYMBOL(netdev_lower_get_next_private); 7547 7548 /** 7549 * netdev_lower_get_next_private_rcu - Get the next ->private from the 7550 * lower neighbour list, RCU 7551 * variant 7552 * @dev: device 7553 * @iter: list_head ** of the current position 7554 * 7555 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7556 * list, starting from iter position. The caller must hold RCU read lock. 7557 */ 7558 void *netdev_lower_get_next_private_rcu(struct net_device *dev, 7559 struct list_head **iter) 7560 { 7561 struct netdev_adjacent *lower; 7562 7563 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 7564 7565 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7566 7567 if (&lower->list == &dev->adj_list.lower) 7568 return NULL; 7569 7570 *iter = &lower->list; 7571 7572 return lower->private; 7573 } 7574 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); 7575 7576 /** 7577 * netdev_lower_get_next - Get the next device from the lower neighbour 7578 * list 7579 * @dev: device 7580 * @iter: list_head ** of the current position 7581 * 7582 * Gets the next netdev_adjacent from the dev's lower neighbour 7583 * list, starting from iter position. The caller must hold RTNL lock or 7584 * its own locking that guarantees that the neighbour lower 7585 * list will remain unchanged. 7586 */ 7587 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) 7588 { 7589 struct netdev_adjacent *lower; 7590 7591 lower = list_entry(*iter, struct netdev_adjacent, list); 7592 7593 if (&lower->list == &dev->adj_list.lower) 7594 return NULL; 7595 7596 *iter = lower->list.next; 7597 7598 return lower->dev; 7599 } 7600 EXPORT_SYMBOL(netdev_lower_get_next); 7601 7602 static struct net_device *netdev_next_lower_dev(struct net_device *dev, 7603 struct list_head **iter) 7604 { 7605 struct netdev_adjacent *lower; 7606 7607 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7608 7609 if (&lower->list == &dev->adj_list.lower) 7610 return NULL; 7611 7612 *iter = &lower->list; 7613 7614 return lower->dev; 7615 } 7616 7617 static struct net_device *__netdev_next_lower_dev(struct net_device *dev, 7618 struct list_head **iter, 7619 bool *ignore) 7620 { 7621 struct netdev_adjacent *lower; 7622 7623 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7624 7625 if (&lower->list == &dev->adj_list.lower) 7626 return NULL; 7627 7628 *iter = &lower->list; 7629 *ignore = lower->ignore; 7630 7631 return lower->dev; 7632 } 7633 7634 int netdev_walk_all_lower_dev(struct net_device *dev, 7635 int (*fn)(struct net_device *dev, 7636 struct netdev_nested_priv *priv), 7637 struct netdev_nested_priv *priv) 7638 { 7639 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7640 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7641 int ret, cur = 0; 7642 7643 now = dev; 7644 iter = &dev->adj_list.lower; 7645 7646 while (1) { 7647 if (now != dev) { 7648 ret = fn(now, priv); 7649 if (ret) 7650 return ret; 7651 } 7652 7653 next = NULL; 7654 while (1) { 7655 ldev = netdev_next_lower_dev(now, &iter); 7656 if (!ldev) 7657 break; 7658 7659 next = ldev; 7660 niter = &ldev->adj_list.lower; 7661 dev_stack[cur] = now; 7662 iter_stack[cur++] = iter; 7663 break; 7664 } 7665 7666 if (!next) { 7667 if (!cur) 7668 return 0; 7669 next = dev_stack[--cur]; 7670 niter = iter_stack[cur]; 7671 } 7672 7673 now = next; 7674 iter = niter; 7675 } 7676 7677 return 0; 7678 } 7679 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); 7680 7681 static int __netdev_walk_all_lower_dev(struct net_device *dev, 7682 int (*fn)(struct net_device *dev, 7683 struct netdev_nested_priv *priv), 7684 struct netdev_nested_priv *priv) 7685 { 7686 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7687 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7688 int ret, cur = 0; 7689 bool ignore; 7690 7691 now = dev; 7692 iter = &dev->adj_list.lower; 7693 7694 while (1) { 7695 if (now != dev) { 7696 ret = fn(now, priv); 7697 if (ret) 7698 return ret; 7699 } 7700 7701 next = NULL; 7702 while (1) { 7703 ldev = __netdev_next_lower_dev(now, &iter, &ignore); 7704 if (!ldev) 7705 break; 7706 if (ignore) 7707 continue; 7708 7709 next = ldev; 7710 niter = &ldev->adj_list.lower; 7711 dev_stack[cur] = now; 7712 iter_stack[cur++] = iter; 7713 break; 7714 } 7715 7716 if (!next) { 7717 if (!cur) 7718 return 0; 7719 next = dev_stack[--cur]; 7720 niter = iter_stack[cur]; 7721 } 7722 7723 now = next; 7724 iter = niter; 7725 } 7726 7727 return 0; 7728 } 7729 7730 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, 7731 struct list_head **iter) 7732 { 7733 struct netdev_adjacent *lower; 7734 7735 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7736 if (&lower->list == &dev->adj_list.lower) 7737 return NULL; 7738 7739 *iter = &lower->list; 7740 7741 return lower->dev; 7742 } 7743 EXPORT_SYMBOL(netdev_next_lower_dev_rcu); 7744 7745 static u8 __netdev_upper_depth(struct net_device *dev) 7746 { 7747 struct net_device *udev; 7748 struct list_head *iter; 7749 u8 max_depth = 0; 7750 bool ignore; 7751 7752 for (iter = &dev->adj_list.upper, 7753 udev = __netdev_next_upper_dev(dev, &iter, &ignore); 7754 udev; 7755 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) { 7756 if (ignore) 7757 continue; 7758 if (max_depth < udev->upper_level) 7759 max_depth = udev->upper_level; 7760 } 7761 7762 return max_depth; 7763 } 7764 7765 static u8 __netdev_lower_depth(struct net_device *dev) 7766 { 7767 struct net_device *ldev; 7768 struct list_head *iter; 7769 u8 max_depth = 0; 7770 bool ignore; 7771 7772 for (iter = &dev->adj_list.lower, 7773 ldev = __netdev_next_lower_dev(dev, &iter, &ignore); 7774 ldev; 7775 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) { 7776 if (ignore) 7777 continue; 7778 if (max_depth < ldev->lower_level) 7779 max_depth = ldev->lower_level; 7780 } 7781 7782 return max_depth; 7783 } 7784 7785 static int __netdev_update_upper_level(struct net_device *dev, 7786 struct netdev_nested_priv *__unused) 7787 { 7788 dev->upper_level = __netdev_upper_depth(dev) + 1; 7789 return 0; 7790 } 7791 7792 static int __netdev_update_lower_level(struct net_device *dev, 7793 struct netdev_nested_priv *priv) 7794 { 7795 dev->lower_level = __netdev_lower_depth(dev) + 1; 7796 7797 #ifdef CONFIG_LOCKDEP 7798 if (!priv) 7799 return 0; 7800 7801 if (priv->flags & NESTED_SYNC_IMM) 7802 dev->nested_level = dev->lower_level - 1; 7803 if (priv->flags & NESTED_SYNC_TODO) 7804 net_unlink_todo(dev); 7805 #endif 7806 return 0; 7807 } 7808 7809 int netdev_walk_all_lower_dev_rcu(struct net_device *dev, 7810 int (*fn)(struct net_device *dev, 7811 struct netdev_nested_priv *priv), 7812 struct netdev_nested_priv *priv) 7813 { 7814 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7815 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7816 int ret, cur = 0; 7817 7818 now = dev; 7819 iter = &dev->adj_list.lower; 7820 7821 while (1) { 7822 if (now != dev) { 7823 ret = fn(now, priv); 7824 if (ret) 7825 return ret; 7826 } 7827 7828 next = NULL; 7829 while (1) { 7830 ldev = netdev_next_lower_dev_rcu(now, &iter); 7831 if (!ldev) 7832 break; 7833 7834 next = ldev; 7835 niter = &ldev->adj_list.lower; 7836 dev_stack[cur] = now; 7837 iter_stack[cur++] = iter; 7838 break; 7839 } 7840 7841 if (!next) { 7842 if (!cur) 7843 return 0; 7844 next = dev_stack[--cur]; 7845 niter = iter_stack[cur]; 7846 } 7847 7848 now = next; 7849 iter = niter; 7850 } 7851 7852 return 0; 7853 } 7854 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); 7855 7856 /** 7857 * netdev_lower_get_first_private_rcu - Get the first ->private from the 7858 * lower neighbour list, RCU 7859 * variant 7860 * @dev: device 7861 * 7862 * Gets the first netdev_adjacent->private from the dev's lower neighbour 7863 * list. The caller must hold RCU read lock. 7864 */ 7865 void *netdev_lower_get_first_private_rcu(struct net_device *dev) 7866 { 7867 struct netdev_adjacent *lower; 7868 7869 lower = list_first_or_null_rcu(&dev->adj_list.lower, 7870 struct netdev_adjacent, list); 7871 if (lower) 7872 return lower->private; 7873 return NULL; 7874 } 7875 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); 7876 7877 /** 7878 * netdev_master_upper_dev_get_rcu - Get master upper device 7879 * @dev: device 7880 * 7881 * Find a master upper device and return pointer to it or NULL in case 7882 * it's not there. The caller must hold the RCU read lock. 7883 */ 7884 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) 7885 { 7886 struct netdev_adjacent *upper; 7887 7888 upper = list_first_or_null_rcu(&dev->adj_list.upper, 7889 struct netdev_adjacent, list); 7890 if (upper && likely(upper->master)) 7891 return upper->dev; 7892 return NULL; 7893 } 7894 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); 7895 7896 static int netdev_adjacent_sysfs_add(struct net_device *dev, 7897 struct net_device *adj_dev, 7898 struct list_head *dev_list) 7899 { 7900 char linkname[IFNAMSIZ+7]; 7901 7902 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7903 "upper_%s" : "lower_%s", adj_dev->name); 7904 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), 7905 linkname); 7906 } 7907 static void netdev_adjacent_sysfs_del(struct net_device *dev, 7908 char *name, 7909 struct list_head *dev_list) 7910 { 7911 char linkname[IFNAMSIZ+7]; 7912 7913 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7914 "upper_%s" : "lower_%s", name); 7915 sysfs_remove_link(&(dev->dev.kobj), linkname); 7916 } 7917 7918 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, 7919 struct net_device *adj_dev, 7920 struct list_head *dev_list) 7921 { 7922 return (dev_list == &dev->adj_list.upper || 7923 dev_list == &dev->adj_list.lower) && 7924 net_eq(dev_net(dev), dev_net(adj_dev)); 7925 } 7926 7927 static int __netdev_adjacent_dev_insert(struct net_device *dev, 7928 struct net_device *adj_dev, 7929 struct list_head *dev_list, 7930 void *private, bool master) 7931 { 7932 struct netdev_adjacent *adj; 7933 int ret; 7934 7935 adj = __netdev_find_adj(adj_dev, dev_list); 7936 7937 if (adj) { 7938 adj->ref_nr += 1; 7939 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", 7940 dev->name, adj_dev->name, adj->ref_nr); 7941 7942 return 0; 7943 } 7944 7945 adj = kmalloc(sizeof(*adj), GFP_KERNEL); 7946 if (!adj) 7947 return -ENOMEM; 7948 7949 adj->dev = adj_dev; 7950 adj->master = master; 7951 adj->ref_nr = 1; 7952 adj->private = private; 7953 adj->ignore = false; 7954 dev_hold(adj_dev); 7955 7956 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", 7957 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); 7958 7959 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { 7960 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); 7961 if (ret) 7962 goto free_adj; 7963 } 7964 7965 /* Ensure that master link is always the first item in list. */ 7966 if (master) { 7967 ret = sysfs_create_link(&(dev->dev.kobj), 7968 &(adj_dev->dev.kobj), "master"); 7969 if (ret) 7970 goto remove_symlinks; 7971 7972 list_add_rcu(&adj->list, dev_list); 7973 } else { 7974 list_add_tail_rcu(&adj->list, dev_list); 7975 } 7976 7977 return 0; 7978 7979 remove_symlinks: 7980 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7981 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7982 free_adj: 7983 kfree(adj); 7984 dev_put(adj_dev); 7985 7986 return ret; 7987 } 7988 7989 static void __netdev_adjacent_dev_remove(struct net_device *dev, 7990 struct net_device *adj_dev, 7991 u16 ref_nr, 7992 struct list_head *dev_list) 7993 { 7994 struct netdev_adjacent *adj; 7995 7996 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", 7997 dev->name, adj_dev->name, ref_nr); 7998 7999 adj = __netdev_find_adj(adj_dev, dev_list); 8000 8001 if (!adj) { 8002 pr_err("Adjacency does not exist for device %s from %s\n", 8003 dev->name, adj_dev->name); 8004 WARN_ON(1); 8005 return; 8006 } 8007 8008 if (adj->ref_nr > ref_nr) { 8009 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", 8010 dev->name, adj_dev->name, ref_nr, 8011 adj->ref_nr - ref_nr); 8012 adj->ref_nr -= ref_nr; 8013 return; 8014 } 8015 8016 if (adj->master) 8017 sysfs_remove_link(&(dev->dev.kobj), "master"); 8018 8019 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 8020 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 8021 8022 list_del_rcu(&adj->list); 8023 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", 8024 adj_dev->name, dev->name, adj_dev->name); 8025 dev_put(adj_dev); 8026 kfree_rcu(adj, rcu); 8027 } 8028 8029 static int __netdev_adjacent_dev_link_lists(struct net_device *dev, 8030 struct net_device *upper_dev, 8031 struct list_head *up_list, 8032 struct list_head *down_list, 8033 void *private, bool master) 8034 { 8035 int ret; 8036 8037 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, 8038 private, master); 8039 if (ret) 8040 return ret; 8041 8042 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, 8043 private, false); 8044 if (ret) { 8045 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); 8046 return ret; 8047 } 8048 8049 return 0; 8050 } 8051 8052 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, 8053 struct net_device *upper_dev, 8054 u16 ref_nr, 8055 struct list_head *up_list, 8056 struct list_head *down_list) 8057 { 8058 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); 8059 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); 8060 } 8061 8062 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, 8063 struct net_device *upper_dev, 8064 void *private, bool master) 8065 { 8066 return __netdev_adjacent_dev_link_lists(dev, upper_dev, 8067 &dev->adj_list.upper, 8068 &upper_dev->adj_list.lower, 8069 private, master); 8070 } 8071 8072 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, 8073 struct net_device *upper_dev) 8074 { 8075 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, 8076 &dev->adj_list.upper, 8077 &upper_dev->adj_list.lower); 8078 } 8079 8080 static int __netdev_upper_dev_link(struct net_device *dev, 8081 struct net_device *upper_dev, bool master, 8082 void *upper_priv, void *upper_info, 8083 struct netdev_nested_priv *priv, 8084 struct netlink_ext_ack *extack) 8085 { 8086 struct netdev_notifier_changeupper_info changeupper_info = { 8087 .info = { 8088 .dev = dev, 8089 .extack = extack, 8090 }, 8091 .upper_dev = upper_dev, 8092 .master = master, 8093 .linking = true, 8094 .upper_info = upper_info, 8095 }; 8096 struct net_device *master_dev; 8097 int ret = 0; 8098 8099 ASSERT_RTNL(); 8100 8101 if (dev == upper_dev) 8102 return -EBUSY; 8103 8104 /* To prevent loops, check if dev is not upper device to upper_dev. */ 8105 if (__netdev_has_upper_dev(upper_dev, dev)) 8106 return -EBUSY; 8107 8108 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV) 8109 return -EMLINK; 8110 8111 if (!master) { 8112 if (__netdev_has_upper_dev(dev, upper_dev)) 8113 return -EEXIST; 8114 } else { 8115 master_dev = __netdev_master_upper_dev_get(dev); 8116 if (master_dev) 8117 return master_dev == upper_dev ? -EEXIST : -EBUSY; 8118 } 8119 8120 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 8121 &changeupper_info.info); 8122 ret = notifier_to_errno(ret); 8123 if (ret) 8124 return ret; 8125 8126 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, 8127 master); 8128 if (ret) 8129 return ret; 8130 8131 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 8132 &changeupper_info.info); 8133 ret = notifier_to_errno(ret); 8134 if (ret) 8135 goto rollback; 8136 8137 __netdev_update_upper_level(dev, NULL); 8138 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 8139 8140 __netdev_update_lower_level(upper_dev, priv); 8141 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 8142 priv); 8143 8144 return 0; 8145 8146 rollback: 8147 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 8148 8149 return ret; 8150 } 8151 8152 /** 8153 * netdev_upper_dev_link - Add a link to the upper device 8154 * @dev: device 8155 * @upper_dev: new upper device 8156 * @extack: netlink extended ack 8157 * 8158 * Adds a link to device which is upper to this one. The caller must hold 8159 * the RTNL lock. On a failure a negative errno code is returned. 8160 * On success the reference counts are adjusted and the function 8161 * returns zero. 8162 */ 8163 int netdev_upper_dev_link(struct net_device *dev, 8164 struct net_device *upper_dev, 8165 struct netlink_ext_ack *extack) 8166 { 8167 struct netdev_nested_priv priv = { 8168 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 8169 .data = NULL, 8170 }; 8171 8172 return __netdev_upper_dev_link(dev, upper_dev, false, 8173 NULL, NULL, &priv, extack); 8174 } 8175 EXPORT_SYMBOL(netdev_upper_dev_link); 8176 8177 /** 8178 * netdev_master_upper_dev_link - Add a master link to the upper device 8179 * @dev: device 8180 * @upper_dev: new upper device 8181 * @upper_priv: upper device private 8182 * @upper_info: upper info to be passed down via notifier 8183 * @extack: netlink extended ack 8184 * 8185 * Adds a link to device which is upper to this one. In this case, only 8186 * one master upper device can be linked, although other non-master devices 8187 * might be linked as well. The caller must hold the RTNL lock. 8188 * On a failure a negative errno code is returned. On success the reference 8189 * counts are adjusted and the function returns zero. 8190 */ 8191 int netdev_master_upper_dev_link(struct net_device *dev, 8192 struct net_device *upper_dev, 8193 void *upper_priv, void *upper_info, 8194 struct netlink_ext_ack *extack) 8195 { 8196 struct netdev_nested_priv priv = { 8197 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 8198 .data = NULL, 8199 }; 8200 8201 return __netdev_upper_dev_link(dev, upper_dev, true, 8202 upper_priv, upper_info, &priv, extack); 8203 } 8204 EXPORT_SYMBOL(netdev_master_upper_dev_link); 8205 8206 static void __netdev_upper_dev_unlink(struct net_device *dev, 8207 struct net_device *upper_dev, 8208 struct netdev_nested_priv *priv) 8209 { 8210 struct netdev_notifier_changeupper_info changeupper_info = { 8211 .info = { 8212 .dev = dev, 8213 }, 8214 .upper_dev = upper_dev, 8215 .linking = false, 8216 }; 8217 8218 ASSERT_RTNL(); 8219 8220 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; 8221 8222 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 8223 &changeupper_info.info); 8224 8225 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 8226 8227 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 8228 &changeupper_info.info); 8229 8230 __netdev_update_upper_level(dev, NULL); 8231 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 8232 8233 __netdev_update_lower_level(upper_dev, priv); 8234 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 8235 priv); 8236 } 8237 8238 /** 8239 * netdev_upper_dev_unlink - Removes a link to upper device 8240 * @dev: device 8241 * @upper_dev: new upper device 8242 * 8243 * Removes a link to device which is upper to this one. The caller must hold 8244 * the RTNL lock. 8245 */ 8246 void netdev_upper_dev_unlink(struct net_device *dev, 8247 struct net_device *upper_dev) 8248 { 8249 struct netdev_nested_priv priv = { 8250 .flags = NESTED_SYNC_TODO, 8251 .data = NULL, 8252 }; 8253 8254 __netdev_upper_dev_unlink(dev, upper_dev, &priv); 8255 } 8256 EXPORT_SYMBOL(netdev_upper_dev_unlink); 8257 8258 static void __netdev_adjacent_dev_set(struct net_device *upper_dev, 8259 struct net_device *lower_dev, 8260 bool val) 8261 { 8262 struct netdev_adjacent *adj; 8263 8264 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower); 8265 if (adj) 8266 adj->ignore = val; 8267 8268 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper); 8269 if (adj) 8270 adj->ignore = val; 8271 } 8272 8273 static void netdev_adjacent_dev_disable(struct net_device *upper_dev, 8274 struct net_device *lower_dev) 8275 { 8276 __netdev_adjacent_dev_set(upper_dev, lower_dev, true); 8277 } 8278 8279 static void netdev_adjacent_dev_enable(struct net_device *upper_dev, 8280 struct net_device *lower_dev) 8281 { 8282 __netdev_adjacent_dev_set(upper_dev, lower_dev, false); 8283 } 8284 8285 int netdev_adjacent_change_prepare(struct net_device *old_dev, 8286 struct net_device *new_dev, 8287 struct net_device *dev, 8288 struct netlink_ext_ack *extack) 8289 { 8290 struct netdev_nested_priv priv = { 8291 .flags = 0, 8292 .data = NULL, 8293 }; 8294 int err; 8295 8296 if (!new_dev) 8297 return 0; 8298 8299 if (old_dev && new_dev != old_dev) 8300 netdev_adjacent_dev_disable(dev, old_dev); 8301 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv, 8302 extack); 8303 if (err) { 8304 if (old_dev && new_dev != old_dev) 8305 netdev_adjacent_dev_enable(dev, old_dev); 8306 return err; 8307 } 8308 8309 return 0; 8310 } 8311 EXPORT_SYMBOL(netdev_adjacent_change_prepare); 8312 8313 void netdev_adjacent_change_commit(struct net_device *old_dev, 8314 struct net_device *new_dev, 8315 struct net_device *dev) 8316 { 8317 struct netdev_nested_priv priv = { 8318 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 8319 .data = NULL, 8320 }; 8321 8322 if (!new_dev || !old_dev) 8323 return; 8324 8325 if (new_dev == old_dev) 8326 return; 8327 8328 netdev_adjacent_dev_enable(dev, old_dev); 8329 __netdev_upper_dev_unlink(old_dev, dev, &priv); 8330 } 8331 EXPORT_SYMBOL(netdev_adjacent_change_commit); 8332 8333 void netdev_adjacent_change_abort(struct net_device *old_dev, 8334 struct net_device *new_dev, 8335 struct net_device *dev) 8336 { 8337 struct netdev_nested_priv priv = { 8338 .flags = 0, 8339 .data = NULL, 8340 }; 8341 8342 if (!new_dev) 8343 return; 8344 8345 if (old_dev && new_dev != old_dev) 8346 netdev_adjacent_dev_enable(dev, old_dev); 8347 8348 __netdev_upper_dev_unlink(new_dev, dev, &priv); 8349 } 8350 EXPORT_SYMBOL(netdev_adjacent_change_abort); 8351 8352 /** 8353 * netdev_bonding_info_change - Dispatch event about slave change 8354 * @dev: device 8355 * @bonding_info: info to dispatch 8356 * 8357 * Send NETDEV_BONDING_INFO to netdev notifiers with info. 8358 * The caller must hold the RTNL lock. 8359 */ 8360 void netdev_bonding_info_change(struct net_device *dev, 8361 struct netdev_bonding_info *bonding_info) 8362 { 8363 struct netdev_notifier_bonding_info info = { 8364 .info.dev = dev, 8365 }; 8366 8367 memcpy(&info.bonding_info, bonding_info, 8368 sizeof(struct netdev_bonding_info)); 8369 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, 8370 &info.info); 8371 } 8372 EXPORT_SYMBOL(netdev_bonding_info_change); 8373 8374 /** 8375 * netdev_get_xmit_slave - Get the xmit slave of master device 8376 * @dev: device 8377 * @skb: The packet 8378 * @all_slaves: assume all the slaves are active 8379 * 8380 * The reference counters are not incremented so the caller must be 8381 * careful with locks. The caller must hold RCU lock. 8382 * %NULL is returned if no slave is found. 8383 */ 8384 8385 struct net_device *netdev_get_xmit_slave(struct net_device *dev, 8386 struct sk_buff *skb, 8387 bool all_slaves) 8388 { 8389 const struct net_device_ops *ops = dev->netdev_ops; 8390 8391 if (!ops->ndo_get_xmit_slave) 8392 return NULL; 8393 return ops->ndo_get_xmit_slave(dev, skb, all_slaves); 8394 } 8395 EXPORT_SYMBOL(netdev_get_xmit_slave); 8396 8397 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev, 8398 struct sock *sk) 8399 { 8400 const struct net_device_ops *ops = dev->netdev_ops; 8401 8402 if (!ops->ndo_sk_get_lower_dev) 8403 return NULL; 8404 return ops->ndo_sk_get_lower_dev(dev, sk); 8405 } 8406 8407 /** 8408 * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket 8409 * @dev: device 8410 * @sk: the socket 8411 * 8412 * %NULL is returned if no lower device is found. 8413 */ 8414 8415 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev, 8416 struct sock *sk) 8417 { 8418 struct net_device *lower; 8419 8420 lower = netdev_sk_get_lower_dev(dev, sk); 8421 while (lower) { 8422 dev = lower; 8423 lower = netdev_sk_get_lower_dev(dev, sk); 8424 } 8425 8426 return dev; 8427 } 8428 EXPORT_SYMBOL(netdev_sk_get_lowest_dev); 8429 8430 static void netdev_adjacent_add_links(struct net_device *dev) 8431 { 8432 struct netdev_adjacent *iter; 8433 8434 struct net *net = dev_net(dev); 8435 8436 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8437 if (!net_eq(net, dev_net(iter->dev))) 8438 continue; 8439 netdev_adjacent_sysfs_add(iter->dev, dev, 8440 &iter->dev->adj_list.lower); 8441 netdev_adjacent_sysfs_add(dev, iter->dev, 8442 &dev->adj_list.upper); 8443 } 8444 8445 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8446 if (!net_eq(net, dev_net(iter->dev))) 8447 continue; 8448 netdev_adjacent_sysfs_add(iter->dev, dev, 8449 &iter->dev->adj_list.upper); 8450 netdev_adjacent_sysfs_add(dev, iter->dev, 8451 &dev->adj_list.lower); 8452 } 8453 } 8454 8455 static void netdev_adjacent_del_links(struct net_device *dev) 8456 { 8457 struct netdev_adjacent *iter; 8458 8459 struct net *net = dev_net(dev); 8460 8461 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8462 if (!net_eq(net, dev_net(iter->dev))) 8463 continue; 8464 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8465 &iter->dev->adj_list.lower); 8466 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8467 &dev->adj_list.upper); 8468 } 8469 8470 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8471 if (!net_eq(net, dev_net(iter->dev))) 8472 continue; 8473 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8474 &iter->dev->adj_list.upper); 8475 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8476 &dev->adj_list.lower); 8477 } 8478 } 8479 8480 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) 8481 { 8482 struct netdev_adjacent *iter; 8483 8484 struct net *net = dev_net(dev); 8485 8486 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8487 if (!net_eq(net, dev_net(iter->dev))) 8488 continue; 8489 netdev_adjacent_sysfs_del(iter->dev, oldname, 8490 &iter->dev->adj_list.lower); 8491 netdev_adjacent_sysfs_add(iter->dev, dev, 8492 &iter->dev->adj_list.lower); 8493 } 8494 8495 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8496 if (!net_eq(net, dev_net(iter->dev))) 8497 continue; 8498 netdev_adjacent_sysfs_del(iter->dev, oldname, 8499 &iter->dev->adj_list.upper); 8500 netdev_adjacent_sysfs_add(iter->dev, dev, 8501 &iter->dev->adj_list.upper); 8502 } 8503 } 8504 8505 void *netdev_lower_dev_get_private(struct net_device *dev, 8506 struct net_device *lower_dev) 8507 { 8508 struct netdev_adjacent *lower; 8509 8510 if (!lower_dev) 8511 return NULL; 8512 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); 8513 if (!lower) 8514 return NULL; 8515 8516 return lower->private; 8517 } 8518 EXPORT_SYMBOL(netdev_lower_dev_get_private); 8519 8520 8521 /** 8522 * netdev_lower_state_changed - Dispatch event about lower device state change 8523 * @lower_dev: device 8524 * @lower_state_info: state to dispatch 8525 * 8526 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. 8527 * The caller must hold the RTNL lock. 8528 */ 8529 void netdev_lower_state_changed(struct net_device *lower_dev, 8530 void *lower_state_info) 8531 { 8532 struct netdev_notifier_changelowerstate_info changelowerstate_info = { 8533 .info.dev = lower_dev, 8534 }; 8535 8536 ASSERT_RTNL(); 8537 changelowerstate_info.lower_state_info = lower_state_info; 8538 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, 8539 &changelowerstate_info.info); 8540 } 8541 EXPORT_SYMBOL(netdev_lower_state_changed); 8542 8543 static void dev_change_rx_flags(struct net_device *dev, int flags) 8544 { 8545 const struct net_device_ops *ops = dev->netdev_ops; 8546 8547 if (ops->ndo_change_rx_flags) 8548 ops->ndo_change_rx_flags(dev, flags); 8549 } 8550 8551 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) 8552 { 8553 unsigned int old_flags = dev->flags; 8554 kuid_t uid; 8555 kgid_t gid; 8556 8557 ASSERT_RTNL(); 8558 8559 dev->flags |= IFF_PROMISC; 8560 dev->promiscuity += inc; 8561 if (dev->promiscuity == 0) { 8562 /* 8563 * Avoid overflow. 8564 * If inc causes overflow, untouch promisc and return error. 8565 */ 8566 if (inc < 0) 8567 dev->flags &= ~IFF_PROMISC; 8568 else { 8569 dev->promiscuity -= inc; 8570 netdev_warn(dev, "promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n"); 8571 return -EOVERFLOW; 8572 } 8573 } 8574 if (dev->flags != old_flags) { 8575 pr_info("device %s %s promiscuous mode\n", 8576 dev->name, 8577 dev->flags & IFF_PROMISC ? "entered" : "left"); 8578 if (audit_enabled) { 8579 current_uid_gid(&uid, &gid); 8580 audit_log(audit_context(), GFP_ATOMIC, 8581 AUDIT_ANOM_PROMISCUOUS, 8582 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", 8583 dev->name, (dev->flags & IFF_PROMISC), 8584 (old_flags & IFF_PROMISC), 8585 from_kuid(&init_user_ns, audit_get_loginuid(current)), 8586 from_kuid(&init_user_ns, uid), 8587 from_kgid(&init_user_ns, gid), 8588 audit_get_sessionid(current)); 8589 } 8590 8591 dev_change_rx_flags(dev, IFF_PROMISC); 8592 } 8593 if (notify) 8594 __dev_notify_flags(dev, old_flags, IFF_PROMISC); 8595 return 0; 8596 } 8597 8598 /** 8599 * dev_set_promiscuity - update promiscuity count on a device 8600 * @dev: device 8601 * @inc: modifier 8602 * 8603 * Add or remove promiscuity from a device. While the count in the device 8604 * remains above zero the interface remains promiscuous. Once it hits zero 8605 * the device reverts back to normal filtering operation. A negative inc 8606 * value is used to drop promiscuity on the device. 8607 * Return 0 if successful or a negative errno code on error. 8608 */ 8609 int dev_set_promiscuity(struct net_device *dev, int inc) 8610 { 8611 unsigned int old_flags = dev->flags; 8612 int err; 8613 8614 err = __dev_set_promiscuity(dev, inc, true); 8615 if (err < 0) 8616 return err; 8617 if (dev->flags != old_flags) 8618 dev_set_rx_mode(dev); 8619 return err; 8620 } 8621 EXPORT_SYMBOL(dev_set_promiscuity); 8622 8623 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) 8624 { 8625 unsigned int old_flags = dev->flags, old_gflags = dev->gflags; 8626 8627 ASSERT_RTNL(); 8628 8629 dev->flags |= IFF_ALLMULTI; 8630 dev->allmulti += inc; 8631 if (dev->allmulti == 0) { 8632 /* 8633 * Avoid overflow. 8634 * If inc causes overflow, untouch allmulti and return error. 8635 */ 8636 if (inc < 0) 8637 dev->flags &= ~IFF_ALLMULTI; 8638 else { 8639 dev->allmulti -= inc; 8640 netdev_warn(dev, "allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n"); 8641 return -EOVERFLOW; 8642 } 8643 } 8644 if (dev->flags ^ old_flags) { 8645 dev_change_rx_flags(dev, IFF_ALLMULTI); 8646 dev_set_rx_mode(dev); 8647 if (notify) 8648 __dev_notify_flags(dev, old_flags, 8649 dev->gflags ^ old_gflags); 8650 } 8651 return 0; 8652 } 8653 8654 /** 8655 * dev_set_allmulti - update allmulti count on a device 8656 * @dev: device 8657 * @inc: modifier 8658 * 8659 * Add or remove reception of all multicast frames to a device. While the 8660 * count in the device remains above zero the interface remains listening 8661 * to all interfaces. Once it hits zero the device reverts back to normal 8662 * filtering operation. A negative @inc value is used to drop the counter 8663 * when releasing a resource needing all multicasts. 8664 * Return 0 if successful or a negative errno code on error. 8665 */ 8666 8667 int dev_set_allmulti(struct net_device *dev, int inc) 8668 { 8669 return __dev_set_allmulti(dev, inc, true); 8670 } 8671 EXPORT_SYMBOL(dev_set_allmulti); 8672 8673 /* 8674 * Upload unicast and multicast address lists to device and 8675 * configure RX filtering. When the device doesn't support unicast 8676 * filtering it is put in promiscuous mode while unicast addresses 8677 * are present. 8678 */ 8679 void __dev_set_rx_mode(struct net_device *dev) 8680 { 8681 const struct net_device_ops *ops = dev->netdev_ops; 8682 8683 /* dev_open will call this function so the list will stay sane. */ 8684 if (!(dev->flags&IFF_UP)) 8685 return; 8686 8687 if (!netif_device_present(dev)) 8688 return; 8689 8690 if (!(dev->priv_flags & IFF_UNICAST_FLT)) { 8691 /* Unicast addresses changes may only happen under the rtnl, 8692 * therefore calling __dev_set_promiscuity here is safe. 8693 */ 8694 if (!netdev_uc_empty(dev) && !dev->uc_promisc) { 8695 __dev_set_promiscuity(dev, 1, false); 8696 dev->uc_promisc = true; 8697 } else if (netdev_uc_empty(dev) && dev->uc_promisc) { 8698 __dev_set_promiscuity(dev, -1, false); 8699 dev->uc_promisc = false; 8700 } 8701 } 8702 8703 if (ops->ndo_set_rx_mode) 8704 ops->ndo_set_rx_mode(dev); 8705 } 8706 8707 void dev_set_rx_mode(struct net_device *dev) 8708 { 8709 netif_addr_lock_bh(dev); 8710 __dev_set_rx_mode(dev); 8711 netif_addr_unlock_bh(dev); 8712 } 8713 8714 /** 8715 * dev_get_flags - get flags reported to userspace 8716 * @dev: device 8717 * 8718 * Get the combination of flag bits exported through APIs to userspace. 8719 */ 8720 unsigned int dev_get_flags(const struct net_device *dev) 8721 { 8722 unsigned int flags; 8723 8724 flags = (dev->flags & ~(IFF_PROMISC | 8725 IFF_ALLMULTI | 8726 IFF_RUNNING | 8727 IFF_LOWER_UP | 8728 IFF_DORMANT)) | 8729 (dev->gflags & (IFF_PROMISC | 8730 IFF_ALLMULTI)); 8731 8732 if (netif_running(dev)) { 8733 if (netif_oper_up(dev)) 8734 flags |= IFF_RUNNING; 8735 if (netif_carrier_ok(dev)) 8736 flags |= IFF_LOWER_UP; 8737 if (netif_dormant(dev)) 8738 flags |= IFF_DORMANT; 8739 } 8740 8741 return flags; 8742 } 8743 EXPORT_SYMBOL(dev_get_flags); 8744 8745 int __dev_change_flags(struct net_device *dev, unsigned int flags, 8746 struct netlink_ext_ack *extack) 8747 { 8748 unsigned int old_flags = dev->flags; 8749 int ret; 8750 8751 ASSERT_RTNL(); 8752 8753 /* 8754 * Set the flags on our device. 8755 */ 8756 8757 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | 8758 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | 8759 IFF_AUTOMEDIA)) | 8760 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | 8761 IFF_ALLMULTI)); 8762 8763 /* 8764 * Load in the correct multicast list now the flags have changed. 8765 */ 8766 8767 if ((old_flags ^ flags) & IFF_MULTICAST) 8768 dev_change_rx_flags(dev, IFF_MULTICAST); 8769 8770 dev_set_rx_mode(dev); 8771 8772 /* 8773 * Have we downed the interface. We handle IFF_UP ourselves 8774 * according to user attempts to set it, rather than blindly 8775 * setting it. 8776 */ 8777 8778 ret = 0; 8779 if ((old_flags ^ flags) & IFF_UP) { 8780 if (old_flags & IFF_UP) 8781 __dev_close(dev); 8782 else 8783 ret = __dev_open(dev, extack); 8784 } 8785 8786 if ((flags ^ dev->gflags) & IFF_PROMISC) { 8787 int inc = (flags & IFF_PROMISC) ? 1 : -1; 8788 unsigned int old_flags = dev->flags; 8789 8790 dev->gflags ^= IFF_PROMISC; 8791 8792 if (__dev_set_promiscuity(dev, inc, false) >= 0) 8793 if (dev->flags != old_flags) 8794 dev_set_rx_mode(dev); 8795 } 8796 8797 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI 8798 * is important. Some (broken) drivers set IFF_PROMISC, when 8799 * IFF_ALLMULTI is requested not asking us and not reporting. 8800 */ 8801 if ((flags ^ dev->gflags) & IFF_ALLMULTI) { 8802 int inc = (flags & IFF_ALLMULTI) ? 1 : -1; 8803 8804 dev->gflags ^= IFF_ALLMULTI; 8805 __dev_set_allmulti(dev, inc, false); 8806 } 8807 8808 return ret; 8809 } 8810 8811 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, 8812 unsigned int gchanges) 8813 { 8814 unsigned int changes = dev->flags ^ old_flags; 8815 8816 if (gchanges) 8817 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC); 8818 8819 if (changes & IFF_UP) { 8820 if (dev->flags & IFF_UP) 8821 call_netdevice_notifiers(NETDEV_UP, dev); 8822 else 8823 call_netdevice_notifiers(NETDEV_DOWN, dev); 8824 } 8825 8826 if (dev->flags & IFF_UP && 8827 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { 8828 struct netdev_notifier_change_info change_info = { 8829 .info = { 8830 .dev = dev, 8831 }, 8832 .flags_changed = changes, 8833 }; 8834 8835 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info); 8836 } 8837 } 8838 8839 /** 8840 * dev_change_flags - change device settings 8841 * @dev: device 8842 * @flags: device state flags 8843 * @extack: netlink extended ack 8844 * 8845 * Change settings on device based state flags. The flags are 8846 * in the userspace exported format. 8847 */ 8848 int dev_change_flags(struct net_device *dev, unsigned int flags, 8849 struct netlink_ext_ack *extack) 8850 { 8851 int ret; 8852 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; 8853 8854 ret = __dev_change_flags(dev, flags, extack); 8855 if (ret < 0) 8856 return ret; 8857 8858 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); 8859 __dev_notify_flags(dev, old_flags, changes); 8860 return ret; 8861 } 8862 EXPORT_SYMBOL(dev_change_flags); 8863 8864 int __dev_set_mtu(struct net_device *dev, int new_mtu) 8865 { 8866 const struct net_device_ops *ops = dev->netdev_ops; 8867 8868 if (ops->ndo_change_mtu) 8869 return ops->ndo_change_mtu(dev, new_mtu); 8870 8871 /* Pairs with all the lockless reads of dev->mtu in the stack */ 8872 WRITE_ONCE(dev->mtu, new_mtu); 8873 return 0; 8874 } 8875 EXPORT_SYMBOL(__dev_set_mtu); 8876 8877 int dev_validate_mtu(struct net_device *dev, int new_mtu, 8878 struct netlink_ext_ack *extack) 8879 { 8880 /* MTU must be positive, and in range */ 8881 if (new_mtu < 0 || new_mtu < dev->min_mtu) { 8882 NL_SET_ERR_MSG(extack, "mtu less than device minimum"); 8883 return -EINVAL; 8884 } 8885 8886 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { 8887 NL_SET_ERR_MSG(extack, "mtu greater than device maximum"); 8888 return -EINVAL; 8889 } 8890 return 0; 8891 } 8892 8893 /** 8894 * dev_set_mtu_ext - Change maximum transfer unit 8895 * @dev: device 8896 * @new_mtu: new transfer unit 8897 * @extack: netlink extended ack 8898 * 8899 * Change the maximum transfer size of the network device. 8900 */ 8901 int dev_set_mtu_ext(struct net_device *dev, int new_mtu, 8902 struct netlink_ext_ack *extack) 8903 { 8904 int err, orig_mtu; 8905 8906 if (new_mtu == dev->mtu) 8907 return 0; 8908 8909 err = dev_validate_mtu(dev, new_mtu, extack); 8910 if (err) 8911 return err; 8912 8913 if (!netif_device_present(dev)) 8914 return -ENODEV; 8915 8916 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); 8917 err = notifier_to_errno(err); 8918 if (err) 8919 return err; 8920 8921 orig_mtu = dev->mtu; 8922 err = __dev_set_mtu(dev, new_mtu); 8923 8924 if (!err) { 8925 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8926 orig_mtu); 8927 err = notifier_to_errno(err); 8928 if (err) { 8929 /* setting mtu back and notifying everyone again, 8930 * so that they have a chance to revert changes. 8931 */ 8932 __dev_set_mtu(dev, orig_mtu); 8933 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8934 new_mtu); 8935 } 8936 } 8937 return err; 8938 } 8939 8940 int dev_set_mtu(struct net_device *dev, int new_mtu) 8941 { 8942 struct netlink_ext_ack extack; 8943 int err; 8944 8945 memset(&extack, 0, sizeof(extack)); 8946 err = dev_set_mtu_ext(dev, new_mtu, &extack); 8947 if (err && extack._msg) 8948 net_err_ratelimited("%s: %s\n", dev->name, extack._msg); 8949 return err; 8950 } 8951 EXPORT_SYMBOL(dev_set_mtu); 8952 8953 /** 8954 * dev_change_tx_queue_len - Change TX queue length of a netdevice 8955 * @dev: device 8956 * @new_len: new tx queue length 8957 */ 8958 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len) 8959 { 8960 unsigned int orig_len = dev->tx_queue_len; 8961 int res; 8962 8963 if (new_len != (unsigned int)new_len) 8964 return -ERANGE; 8965 8966 if (new_len != orig_len) { 8967 dev->tx_queue_len = new_len; 8968 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev); 8969 res = notifier_to_errno(res); 8970 if (res) 8971 goto err_rollback; 8972 res = dev_qdisc_change_tx_queue_len(dev); 8973 if (res) 8974 goto err_rollback; 8975 } 8976 8977 return 0; 8978 8979 err_rollback: 8980 netdev_err(dev, "refused to change device tx_queue_len\n"); 8981 dev->tx_queue_len = orig_len; 8982 return res; 8983 } 8984 8985 /** 8986 * dev_set_group - Change group this device belongs to 8987 * @dev: device 8988 * @new_group: group this device should belong to 8989 */ 8990 void dev_set_group(struct net_device *dev, int new_group) 8991 { 8992 dev->group = new_group; 8993 } 8994 EXPORT_SYMBOL(dev_set_group); 8995 8996 /** 8997 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR. 8998 * @dev: device 8999 * @addr: new address 9000 * @extack: netlink extended ack 9001 */ 9002 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr, 9003 struct netlink_ext_ack *extack) 9004 { 9005 struct netdev_notifier_pre_changeaddr_info info = { 9006 .info.dev = dev, 9007 .info.extack = extack, 9008 .dev_addr = addr, 9009 }; 9010 int rc; 9011 9012 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info); 9013 return notifier_to_errno(rc); 9014 } 9015 EXPORT_SYMBOL(dev_pre_changeaddr_notify); 9016 9017 /** 9018 * dev_set_mac_address - Change Media Access Control Address 9019 * @dev: device 9020 * @sa: new address 9021 * @extack: netlink extended ack 9022 * 9023 * Change the hardware (MAC) address of the device 9024 */ 9025 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa, 9026 struct netlink_ext_ack *extack) 9027 { 9028 const struct net_device_ops *ops = dev->netdev_ops; 9029 int err; 9030 9031 if (!ops->ndo_set_mac_address) 9032 return -EOPNOTSUPP; 9033 if (sa->sa_family != dev->type) 9034 return -EINVAL; 9035 if (!netif_device_present(dev)) 9036 return -ENODEV; 9037 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack); 9038 if (err) 9039 return err; 9040 err = ops->ndo_set_mac_address(dev, sa); 9041 if (err) 9042 return err; 9043 dev->addr_assign_type = NET_ADDR_SET; 9044 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); 9045 add_device_randomness(dev->dev_addr, dev->addr_len); 9046 return 0; 9047 } 9048 EXPORT_SYMBOL(dev_set_mac_address); 9049 9050 static DECLARE_RWSEM(dev_addr_sem); 9051 9052 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa, 9053 struct netlink_ext_ack *extack) 9054 { 9055 int ret; 9056 9057 down_write(&dev_addr_sem); 9058 ret = dev_set_mac_address(dev, sa, extack); 9059 up_write(&dev_addr_sem); 9060 return ret; 9061 } 9062 EXPORT_SYMBOL(dev_set_mac_address_user); 9063 9064 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name) 9065 { 9066 size_t size = sizeof(sa->sa_data); 9067 struct net_device *dev; 9068 int ret = 0; 9069 9070 down_read(&dev_addr_sem); 9071 rcu_read_lock(); 9072 9073 dev = dev_get_by_name_rcu(net, dev_name); 9074 if (!dev) { 9075 ret = -ENODEV; 9076 goto unlock; 9077 } 9078 if (!dev->addr_len) 9079 memset(sa->sa_data, 0, size); 9080 else 9081 memcpy(sa->sa_data, dev->dev_addr, 9082 min_t(size_t, size, dev->addr_len)); 9083 sa->sa_family = dev->type; 9084 9085 unlock: 9086 rcu_read_unlock(); 9087 up_read(&dev_addr_sem); 9088 return ret; 9089 } 9090 EXPORT_SYMBOL(dev_get_mac_address); 9091 9092 /** 9093 * dev_change_carrier - Change device carrier 9094 * @dev: device 9095 * @new_carrier: new value 9096 * 9097 * Change device carrier 9098 */ 9099 int dev_change_carrier(struct net_device *dev, bool new_carrier) 9100 { 9101 const struct net_device_ops *ops = dev->netdev_ops; 9102 9103 if (!ops->ndo_change_carrier) 9104 return -EOPNOTSUPP; 9105 if (!netif_device_present(dev)) 9106 return -ENODEV; 9107 return ops->ndo_change_carrier(dev, new_carrier); 9108 } 9109 EXPORT_SYMBOL(dev_change_carrier); 9110 9111 /** 9112 * dev_get_phys_port_id - Get device physical port ID 9113 * @dev: device 9114 * @ppid: port ID 9115 * 9116 * Get device physical port ID 9117 */ 9118 int dev_get_phys_port_id(struct net_device *dev, 9119 struct netdev_phys_item_id *ppid) 9120 { 9121 const struct net_device_ops *ops = dev->netdev_ops; 9122 9123 if (!ops->ndo_get_phys_port_id) 9124 return -EOPNOTSUPP; 9125 return ops->ndo_get_phys_port_id(dev, ppid); 9126 } 9127 EXPORT_SYMBOL(dev_get_phys_port_id); 9128 9129 /** 9130 * dev_get_phys_port_name - Get device physical port name 9131 * @dev: device 9132 * @name: port name 9133 * @len: limit of bytes to copy to name 9134 * 9135 * Get device physical port name 9136 */ 9137 int dev_get_phys_port_name(struct net_device *dev, 9138 char *name, size_t len) 9139 { 9140 const struct net_device_ops *ops = dev->netdev_ops; 9141 int err; 9142 9143 if (ops->ndo_get_phys_port_name) { 9144 err = ops->ndo_get_phys_port_name(dev, name, len); 9145 if (err != -EOPNOTSUPP) 9146 return err; 9147 } 9148 return devlink_compat_phys_port_name_get(dev, name, len); 9149 } 9150 EXPORT_SYMBOL(dev_get_phys_port_name); 9151 9152 /** 9153 * dev_get_port_parent_id - Get the device's port parent identifier 9154 * @dev: network device 9155 * @ppid: pointer to a storage for the port's parent identifier 9156 * @recurse: allow/disallow recursion to lower devices 9157 * 9158 * Get the devices's port parent identifier 9159 */ 9160 int dev_get_port_parent_id(struct net_device *dev, 9161 struct netdev_phys_item_id *ppid, 9162 bool recurse) 9163 { 9164 const struct net_device_ops *ops = dev->netdev_ops; 9165 struct netdev_phys_item_id first = { }; 9166 struct net_device *lower_dev; 9167 struct list_head *iter; 9168 int err; 9169 9170 if (ops->ndo_get_port_parent_id) { 9171 err = ops->ndo_get_port_parent_id(dev, ppid); 9172 if (err != -EOPNOTSUPP) 9173 return err; 9174 } 9175 9176 err = devlink_compat_switch_id_get(dev, ppid); 9177 if (!recurse || err != -EOPNOTSUPP) 9178 return err; 9179 9180 netdev_for_each_lower_dev(dev, lower_dev, iter) { 9181 err = dev_get_port_parent_id(lower_dev, ppid, true); 9182 if (err) 9183 break; 9184 if (!first.id_len) 9185 first = *ppid; 9186 else if (memcmp(&first, ppid, sizeof(*ppid))) 9187 return -EOPNOTSUPP; 9188 } 9189 9190 return err; 9191 } 9192 EXPORT_SYMBOL(dev_get_port_parent_id); 9193 9194 /** 9195 * netdev_port_same_parent_id - Indicate if two network devices have 9196 * the same port parent identifier 9197 * @a: first network device 9198 * @b: second network device 9199 */ 9200 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b) 9201 { 9202 struct netdev_phys_item_id a_id = { }; 9203 struct netdev_phys_item_id b_id = { }; 9204 9205 if (dev_get_port_parent_id(a, &a_id, true) || 9206 dev_get_port_parent_id(b, &b_id, true)) 9207 return false; 9208 9209 return netdev_phys_item_id_same(&a_id, &b_id); 9210 } 9211 EXPORT_SYMBOL(netdev_port_same_parent_id); 9212 9213 /** 9214 * dev_change_proto_down - update protocol port state information 9215 * @dev: device 9216 * @proto_down: new value 9217 * 9218 * This info can be used by switch drivers to set the phys state of the 9219 * port. 9220 */ 9221 int dev_change_proto_down(struct net_device *dev, bool proto_down) 9222 { 9223 const struct net_device_ops *ops = dev->netdev_ops; 9224 9225 if (!ops->ndo_change_proto_down) 9226 return -EOPNOTSUPP; 9227 if (!netif_device_present(dev)) 9228 return -ENODEV; 9229 return ops->ndo_change_proto_down(dev, proto_down); 9230 } 9231 EXPORT_SYMBOL(dev_change_proto_down); 9232 9233 /** 9234 * dev_change_proto_down_generic - generic implementation for 9235 * ndo_change_proto_down that sets carrier according to 9236 * proto_down. 9237 * 9238 * @dev: device 9239 * @proto_down: new value 9240 */ 9241 int dev_change_proto_down_generic(struct net_device *dev, bool proto_down) 9242 { 9243 if (proto_down) 9244 netif_carrier_off(dev); 9245 else 9246 netif_carrier_on(dev); 9247 dev->proto_down = proto_down; 9248 return 0; 9249 } 9250 EXPORT_SYMBOL(dev_change_proto_down_generic); 9251 9252 /** 9253 * dev_change_proto_down_reason - proto down reason 9254 * 9255 * @dev: device 9256 * @mask: proto down mask 9257 * @value: proto down value 9258 */ 9259 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask, 9260 u32 value) 9261 { 9262 int b; 9263 9264 if (!mask) { 9265 dev->proto_down_reason = value; 9266 } else { 9267 for_each_set_bit(b, &mask, 32) { 9268 if (value & (1 << b)) 9269 dev->proto_down_reason |= BIT(b); 9270 else 9271 dev->proto_down_reason &= ~BIT(b); 9272 } 9273 } 9274 } 9275 EXPORT_SYMBOL(dev_change_proto_down_reason); 9276 9277 struct bpf_xdp_link { 9278 struct bpf_link link; 9279 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */ 9280 int flags; 9281 }; 9282 9283 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags) 9284 { 9285 if (flags & XDP_FLAGS_HW_MODE) 9286 return XDP_MODE_HW; 9287 if (flags & XDP_FLAGS_DRV_MODE) 9288 return XDP_MODE_DRV; 9289 if (flags & XDP_FLAGS_SKB_MODE) 9290 return XDP_MODE_SKB; 9291 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB; 9292 } 9293 9294 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode) 9295 { 9296 switch (mode) { 9297 case XDP_MODE_SKB: 9298 return generic_xdp_install; 9299 case XDP_MODE_DRV: 9300 case XDP_MODE_HW: 9301 return dev->netdev_ops->ndo_bpf; 9302 default: 9303 return NULL; 9304 } 9305 } 9306 9307 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev, 9308 enum bpf_xdp_mode mode) 9309 { 9310 return dev->xdp_state[mode].link; 9311 } 9312 9313 static struct bpf_prog *dev_xdp_prog(struct net_device *dev, 9314 enum bpf_xdp_mode mode) 9315 { 9316 struct bpf_xdp_link *link = dev_xdp_link(dev, mode); 9317 9318 if (link) 9319 return link->link.prog; 9320 return dev->xdp_state[mode].prog; 9321 } 9322 9323 u8 dev_xdp_prog_count(struct net_device *dev) 9324 { 9325 u8 count = 0; 9326 int i; 9327 9328 for (i = 0; i < __MAX_XDP_MODE; i++) 9329 if (dev->xdp_state[i].prog || dev->xdp_state[i].link) 9330 count++; 9331 return count; 9332 } 9333 EXPORT_SYMBOL_GPL(dev_xdp_prog_count); 9334 9335 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode) 9336 { 9337 struct bpf_prog *prog = dev_xdp_prog(dev, mode); 9338 9339 return prog ? prog->aux->id : 0; 9340 } 9341 9342 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode, 9343 struct bpf_xdp_link *link) 9344 { 9345 dev->xdp_state[mode].link = link; 9346 dev->xdp_state[mode].prog = NULL; 9347 } 9348 9349 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode, 9350 struct bpf_prog *prog) 9351 { 9352 dev->xdp_state[mode].link = NULL; 9353 dev->xdp_state[mode].prog = prog; 9354 } 9355 9356 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode, 9357 bpf_op_t bpf_op, struct netlink_ext_ack *extack, 9358 u32 flags, struct bpf_prog *prog) 9359 { 9360 struct netdev_bpf xdp; 9361 int err; 9362 9363 memset(&xdp, 0, sizeof(xdp)); 9364 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG; 9365 xdp.extack = extack; 9366 xdp.flags = flags; 9367 xdp.prog = prog; 9368 9369 /* Drivers assume refcnt is already incremented (i.e, prog pointer is 9370 * "moved" into driver), so they don't increment it on their own, but 9371 * they do decrement refcnt when program is detached or replaced. 9372 * Given net_device also owns link/prog, we need to bump refcnt here 9373 * to prevent drivers from underflowing it. 9374 */ 9375 if (prog) 9376 bpf_prog_inc(prog); 9377 err = bpf_op(dev, &xdp); 9378 if (err) { 9379 if (prog) 9380 bpf_prog_put(prog); 9381 return err; 9382 } 9383 9384 if (mode != XDP_MODE_HW) 9385 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog); 9386 9387 return 0; 9388 } 9389 9390 static void dev_xdp_uninstall(struct net_device *dev) 9391 { 9392 struct bpf_xdp_link *link; 9393 struct bpf_prog *prog; 9394 enum bpf_xdp_mode mode; 9395 bpf_op_t bpf_op; 9396 9397 ASSERT_RTNL(); 9398 9399 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) { 9400 prog = dev_xdp_prog(dev, mode); 9401 if (!prog) 9402 continue; 9403 9404 bpf_op = dev_xdp_bpf_op(dev, mode); 9405 if (!bpf_op) 9406 continue; 9407 9408 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9409 9410 /* auto-detach link from net device */ 9411 link = dev_xdp_link(dev, mode); 9412 if (link) 9413 link->dev = NULL; 9414 else 9415 bpf_prog_put(prog); 9416 9417 dev_xdp_set_link(dev, mode, NULL); 9418 } 9419 } 9420 9421 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack, 9422 struct bpf_xdp_link *link, struct bpf_prog *new_prog, 9423 struct bpf_prog *old_prog, u32 flags) 9424 { 9425 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES); 9426 struct bpf_prog *cur_prog; 9427 struct net_device *upper; 9428 struct list_head *iter; 9429 enum bpf_xdp_mode mode; 9430 bpf_op_t bpf_op; 9431 int err; 9432 9433 ASSERT_RTNL(); 9434 9435 /* either link or prog attachment, never both */ 9436 if (link && (new_prog || old_prog)) 9437 return -EINVAL; 9438 /* link supports only XDP mode flags */ 9439 if (link && (flags & ~XDP_FLAGS_MODES)) { 9440 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment"); 9441 return -EINVAL; 9442 } 9443 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */ 9444 if (num_modes > 1) { 9445 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set"); 9446 return -EINVAL; 9447 } 9448 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */ 9449 if (!num_modes && dev_xdp_prog_count(dev) > 1) { 9450 NL_SET_ERR_MSG(extack, 9451 "More than one program loaded, unset mode is ambiguous"); 9452 return -EINVAL; 9453 } 9454 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */ 9455 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) { 9456 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified"); 9457 return -EINVAL; 9458 } 9459 9460 mode = dev_xdp_mode(dev, flags); 9461 /* can't replace attached link */ 9462 if (dev_xdp_link(dev, mode)) { 9463 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link"); 9464 return -EBUSY; 9465 } 9466 9467 /* don't allow if an upper device already has a program */ 9468 netdev_for_each_upper_dev_rcu(dev, upper, iter) { 9469 if (dev_xdp_prog_count(upper) > 0) { 9470 NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program"); 9471 return -EEXIST; 9472 } 9473 } 9474 9475 cur_prog = dev_xdp_prog(dev, mode); 9476 /* can't replace attached prog with link */ 9477 if (link && cur_prog) { 9478 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link"); 9479 return -EBUSY; 9480 } 9481 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) { 9482 NL_SET_ERR_MSG(extack, "Active program does not match expected"); 9483 return -EEXIST; 9484 } 9485 9486 /* put effective new program into new_prog */ 9487 if (link) 9488 new_prog = link->link.prog; 9489 9490 if (new_prog) { 9491 bool offload = mode == XDP_MODE_HW; 9492 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB 9493 ? XDP_MODE_DRV : XDP_MODE_SKB; 9494 9495 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) { 9496 NL_SET_ERR_MSG(extack, "XDP program already attached"); 9497 return -EBUSY; 9498 } 9499 if (!offload && dev_xdp_prog(dev, other_mode)) { 9500 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time"); 9501 return -EEXIST; 9502 } 9503 if (!offload && bpf_prog_is_dev_bound(new_prog->aux)) { 9504 NL_SET_ERR_MSG(extack, "Using device-bound program without HW_MODE flag is not supported"); 9505 return -EINVAL; 9506 } 9507 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) { 9508 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device"); 9509 return -EINVAL; 9510 } 9511 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) { 9512 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device"); 9513 return -EINVAL; 9514 } 9515 } 9516 9517 /* don't call drivers if the effective program didn't change */ 9518 if (new_prog != cur_prog) { 9519 bpf_op = dev_xdp_bpf_op(dev, mode); 9520 if (!bpf_op) { 9521 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode"); 9522 return -EOPNOTSUPP; 9523 } 9524 9525 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog); 9526 if (err) 9527 return err; 9528 } 9529 9530 if (link) 9531 dev_xdp_set_link(dev, mode, link); 9532 else 9533 dev_xdp_set_prog(dev, mode, new_prog); 9534 if (cur_prog) 9535 bpf_prog_put(cur_prog); 9536 9537 return 0; 9538 } 9539 9540 static int dev_xdp_attach_link(struct net_device *dev, 9541 struct netlink_ext_ack *extack, 9542 struct bpf_xdp_link *link) 9543 { 9544 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags); 9545 } 9546 9547 static int dev_xdp_detach_link(struct net_device *dev, 9548 struct netlink_ext_ack *extack, 9549 struct bpf_xdp_link *link) 9550 { 9551 enum bpf_xdp_mode mode; 9552 bpf_op_t bpf_op; 9553 9554 ASSERT_RTNL(); 9555 9556 mode = dev_xdp_mode(dev, link->flags); 9557 if (dev_xdp_link(dev, mode) != link) 9558 return -EINVAL; 9559 9560 bpf_op = dev_xdp_bpf_op(dev, mode); 9561 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9562 dev_xdp_set_link(dev, mode, NULL); 9563 return 0; 9564 } 9565 9566 static void bpf_xdp_link_release(struct bpf_link *link) 9567 { 9568 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9569 9570 rtnl_lock(); 9571 9572 /* if racing with net_device's tear down, xdp_link->dev might be 9573 * already NULL, in which case link was already auto-detached 9574 */ 9575 if (xdp_link->dev) { 9576 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link)); 9577 xdp_link->dev = NULL; 9578 } 9579 9580 rtnl_unlock(); 9581 } 9582 9583 static int bpf_xdp_link_detach(struct bpf_link *link) 9584 { 9585 bpf_xdp_link_release(link); 9586 return 0; 9587 } 9588 9589 static void bpf_xdp_link_dealloc(struct bpf_link *link) 9590 { 9591 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9592 9593 kfree(xdp_link); 9594 } 9595 9596 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link, 9597 struct seq_file *seq) 9598 { 9599 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9600 u32 ifindex = 0; 9601 9602 rtnl_lock(); 9603 if (xdp_link->dev) 9604 ifindex = xdp_link->dev->ifindex; 9605 rtnl_unlock(); 9606 9607 seq_printf(seq, "ifindex:\t%u\n", ifindex); 9608 } 9609 9610 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link, 9611 struct bpf_link_info *info) 9612 { 9613 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9614 u32 ifindex = 0; 9615 9616 rtnl_lock(); 9617 if (xdp_link->dev) 9618 ifindex = xdp_link->dev->ifindex; 9619 rtnl_unlock(); 9620 9621 info->xdp.ifindex = ifindex; 9622 return 0; 9623 } 9624 9625 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog, 9626 struct bpf_prog *old_prog) 9627 { 9628 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9629 enum bpf_xdp_mode mode; 9630 bpf_op_t bpf_op; 9631 int err = 0; 9632 9633 rtnl_lock(); 9634 9635 /* link might have been auto-released already, so fail */ 9636 if (!xdp_link->dev) { 9637 err = -ENOLINK; 9638 goto out_unlock; 9639 } 9640 9641 if (old_prog && link->prog != old_prog) { 9642 err = -EPERM; 9643 goto out_unlock; 9644 } 9645 old_prog = link->prog; 9646 if (old_prog == new_prog) { 9647 /* no-op, don't disturb drivers */ 9648 bpf_prog_put(new_prog); 9649 goto out_unlock; 9650 } 9651 9652 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags); 9653 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode); 9654 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL, 9655 xdp_link->flags, new_prog); 9656 if (err) 9657 goto out_unlock; 9658 9659 old_prog = xchg(&link->prog, new_prog); 9660 bpf_prog_put(old_prog); 9661 9662 out_unlock: 9663 rtnl_unlock(); 9664 return err; 9665 } 9666 9667 static const struct bpf_link_ops bpf_xdp_link_lops = { 9668 .release = bpf_xdp_link_release, 9669 .dealloc = bpf_xdp_link_dealloc, 9670 .detach = bpf_xdp_link_detach, 9671 .show_fdinfo = bpf_xdp_link_show_fdinfo, 9672 .fill_link_info = bpf_xdp_link_fill_link_info, 9673 .update_prog = bpf_xdp_link_update, 9674 }; 9675 9676 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog) 9677 { 9678 struct net *net = current->nsproxy->net_ns; 9679 struct bpf_link_primer link_primer; 9680 struct bpf_xdp_link *link; 9681 struct net_device *dev; 9682 int err, fd; 9683 9684 rtnl_lock(); 9685 dev = dev_get_by_index(net, attr->link_create.target_ifindex); 9686 if (!dev) { 9687 rtnl_unlock(); 9688 return -EINVAL; 9689 } 9690 9691 link = kzalloc(sizeof(*link), GFP_USER); 9692 if (!link) { 9693 err = -ENOMEM; 9694 goto unlock; 9695 } 9696 9697 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog); 9698 link->dev = dev; 9699 link->flags = attr->link_create.flags; 9700 9701 err = bpf_link_prime(&link->link, &link_primer); 9702 if (err) { 9703 kfree(link); 9704 goto unlock; 9705 } 9706 9707 err = dev_xdp_attach_link(dev, NULL, link); 9708 rtnl_unlock(); 9709 9710 if (err) { 9711 link->dev = NULL; 9712 bpf_link_cleanup(&link_primer); 9713 goto out_put_dev; 9714 } 9715 9716 fd = bpf_link_settle(&link_primer); 9717 /* link itself doesn't hold dev's refcnt to not complicate shutdown */ 9718 dev_put(dev); 9719 return fd; 9720 9721 unlock: 9722 rtnl_unlock(); 9723 9724 out_put_dev: 9725 dev_put(dev); 9726 return err; 9727 } 9728 9729 /** 9730 * dev_change_xdp_fd - set or clear a bpf program for a device rx path 9731 * @dev: device 9732 * @extack: netlink extended ack 9733 * @fd: new program fd or negative value to clear 9734 * @expected_fd: old program fd that userspace expects to replace or clear 9735 * @flags: xdp-related flags 9736 * 9737 * Set or clear a bpf program for a device 9738 */ 9739 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, 9740 int fd, int expected_fd, u32 flags) 9741 { 9742 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags); 9743 struct bpf_prog *new_prog = NULL, *old_prog = NULL; 9744 int err; 9745 9746 ASSERT_RTNL(); 9747 9748 if (fd >= 0) { 9749 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP, 9750 mode != XDP_MODE_SKB); 9751 if (IS_ERR(new_prog)) 9752 return PTR_ERR(new_prog); 9753 } 9754 9755 if (expected_fd >= 0) { 9756 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP, 9757 mode != XDP_MODE_SKB); 9758 if (IS_ERR(old_prog)) { 9759 err = PTR_ERR(old_prog); 9760 old_prog = NULL; 9761 goto err_out; 9762 } 9763 } 9764 9765 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags); 9766 9767 err_out: 9768 if (err && new_prog) 9769 bpf_prog_put(new_prog); 9770 if (old_prog) 9771 bpf_prog_put(old_prog); 9772 return err; 9773 } 9774 9775 /** 9776 * dev_new_index - allocate an ifindex 9777 * @net: the applicable net namespace 9778 * 9779 * Returns a suitable unique value for a new device interface 9780 * number. The caller must hold the rtnl semaphore or the 9781 * dev_base_lock to be sure it remains unique. 9782 */ 9783 static int dev_new_index(struct net *net) 9784 { 9785 int ifindex = net->ifindex; 9786 9787 for (;;) { 9788 if (++ifindex <= 0) 9789 ifindex = 1; 9790 if (!__dev_get_by_index(net, ifindex)) 9791 return net->ifindex = ifindex; 9792 } 9793 } 9794 9795 /* Delayed registration/unregisteration */ 9796 static LIST_HEAD(net_todo_list); 9797 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); 9798 9799 static void net_set_todo(struct net_device *dev) 9800 { 9801 list_add_tail(&dev->todo_list, &net_todo_list); 9802 dev_net(dev)->dev_unreg_count++; 9803 } 9804 9805 static netdev_features_t netdev_sync_upper_features(struct net_device *lower, 9806 struct net_device *upper, netdev_features_t features) 9807 { 9808 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9809 netdev_features_t feature; 9810 int feature_bit; 9811 9812 for_each_netdev_feature(upper_disables, feature_bit) { 9813 feature = __NETIF_F_BIT(feature_bit); 9814 if (!(upper->wanted_features & feature) 9815 && (features & feature)) { 9816 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", 9817 &feature, upper->name); 9818 features &= ~feature; 9819 } 9820 } 9821 9822 return features; 9823 } 9824 9825 static void netdev_sync_lower_features(struct net_device *upper, 9826 struct net_device *lower, netdev_features_t features) 9827 { 9828 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9829 netdev_features_t feature; 9830 int feature_bit; 9831 9832 for_each_netdev_feature(upper_disables, feature_bit) { 9833 feature = __NETIF_F_BIT(feature_bit); 9834 if (!(features & feature) && (lower->features & feature)) { 9835 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", 9836 &feature, lower->name); 9837 lower->wanted_features &= ~feature; 9838 __netdev_update_features(lower); 9839 9840 if (unlikely(lower->features & feature)) 9841 netdev_WARN(upper, "failed to disable %pNF on %s!\n", 9842 &feature, lower->name); 9843 else 9844 netdev_features_change(lower); 9845 } 9846 } 9847 } 9848 9849 static netdev_features_t netdev_fix_features(struct net_device *dev, 9850 netdev_features_t features) 9851 { 9852 /* Fix illegal checksum combinations */ 9853 if ((features & NETIF_F_HW_CSUM) && 9854 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { 9855 netdev_warn(dev, "mixed HW and IP checksum settings.\n"); 9856 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); 9857 } 9858 9859 /* TSO requires that SG is present as well. */ 9860 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { 9861 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); 9862 features &= ~NETIF_F_ALL_TSO; 9863 } 9864 9865 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && 9866 !(features & NETIF_F_IP_CSUM)) { 9867 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); 9868 features &= ~NETIF_F_TSO; 9869 features &= ~NETIF_F_TSO_ECN; 9870 } 9871 9872 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && 9873 !(features & NETIF_F_IPV6_CSUM)) { 9874 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); 9875 features &= ~NETIF_F_TSO6; 9876 } 9877 9878 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ 9879 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) 9880 features &= ~NETIF_F_TSO_MANGLEID; 9881 9882 /* TSO ECN requires that TSO is present as well. */ 9883 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) 9884 features &= ~NETIF_F_TSO_ECN; 9885 9886 /* Software GSO depends on SG. */ 9887 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { 9888 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); 9889 features &= ~NETIF_F_GSO; 9890 } 9891 9892 /* GSO partial features require GSO partial be set */ 9893 if ((features & dev->gso_partial_features) && 9894 !(features & NETIF_F_GSO_PARTIAL)) { 9895 netdev_dbg(dev, 9896 "Dropping partially supported GSO features since no GSO partial.\n"); 9897 features &= ~dev->gso_partial_features; 9898 } 9899 9900 if (!(features & NETIF_F_RXCSUM)) { 9901 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet 9902 * successfully merged by hardware must also have the 9903 * checksum verified by hardware. If the user does not 9904 * want to enable RXCSUM, logically, we should disable GRO_HW. 9905 */ 9906 if (features & NETIF_F_GRO_HW) { 9907 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n"); 9908 features &= ~NETIF_F_GRO_HW; 9909 } 9910 } 9911 9912 /* LRO/HW-GRO features cannot be combined with RX-FCS */ 9913 if (features & NETIF_F_RXFCS) { 9914 if (features & NETIF_F_LRO) { 9915 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n"); 9916 features &= ~NETIF_F_LRO; 9917 } 9918 9919 if (features & NETIF_F_GRO_HW) { 9920 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n"); 9921 features &= ~NETIF_F_GRO_HW; 9922 } 9923 } 9924 9925 if ((features & NETIF_F_GRO_HW) && (features & NETIF_F_LRO)) { 9926 netdev_dbg(dev, "Dropping LRO feature since HW-GRO is requested.\n"); 9927 features &= ~NETIF_F_LRO; 9928 } 9929 9930 if (features & NETIF_F_HW_TLS_TX) { 9931 bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) == 9932 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM); 9933 bool hw_csum = features & NETIF_F_HW_CSUM; 9934 9935 if (!ip_csum && !hw_csum) { 9936 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n"); 9937 features &= ~NETIF_F_HW_TLS_TX; 9938 } 9939 } 9940 9941 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) { 9942 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n"); 9943 features &= ~NETIF_F_HW_TLS_RX; 9944 } 9945 9946 return features; 9947 } 9948 9949 int __netdev_update_features(struct net_device *dev) 9950 { 9951 struct net_device *upper, *lower; 9952 netdev_features_t features; 9953 struct list_head *iter; 9954 int err = -1; 9955 9956 ASSERT_RTNL(); 9957 9958 features = netdev_get_wanted_features(dev); 9959 9960 if (dev->netdev_ops->ndo_fix_features) 9961 features = dev->netdev_ops->ndo_fix_features(dev, features); 9962 9963 /* driver might be less strict about feature dependencies */ 9964 features = netdev_fix_features(dev, features); 9965 9966 /* some features can't be enabled if they're off on an upper device */ 9967 netdev_for_each_upper_dev_rcu(dev, upper, iter) 9968 features = netdev_sync_upper_features(dev, upper, features); 9969 9970 if (dev->features == features) 9971 goto sync_lower; 9972 9973 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", 9974 &dev->features, &features); 9975 9976 if (dev->netdev_ops->ndo_set_features) 9977 err = dev->netdev_ops->ndo_set_features(dev, features); 9978 else 9979 err = 0; 9980 9981 if (unlikely(err < 0)) { 9982 netdev_err(dev, 9983 "set_features() failed (%d); wanted %pNF, left %pNF\n", 9984 err, &features, &dev->features); 9985 /* return non-0 since some features might have changed and 9986 * it's better to fire a spurious notification than miss it 9987 */ 9988 return -1; 9989 } 9990 9991 sync_lower: 9992 /* some features must be disabled on lower devices when disabled 9993 * on an upper device (think: bonding master or bridge) 9994 */ 9995 netdev_for_each_lower_dev(dev, lower, iter) 9996 netdev_sync_lower_features(dev, lower, features); 9997 9998 if (!err) { 9999 netdev_features_t diff = features ^ dev->features; 10000 10001 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) { 10002 /* udp_tunnel_{get,drop}_rx_info both need 10003 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the 10004 * device, or they won't do anything. 10005 * Thus we need to update dev->features 10006 * *before* calling udp_tunnel_get_rx_info, 10007 * but *after* calling udp_tunnel_drop_rx_info. 10008 */ 10009 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) { 10010 dev->features = features; 10011 udp_tunnel_get_rx_info(dev); 10012 } else { 10013 udp_tunnel_drop_rx_info(dev); 10014 } 10015 } 10016 10017 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) { 10018 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) { 10019 dev->features = features; 10020 err |= vlan_get_rx_ctag_filter_info(dev); 10021 } else { 10022 vlan_drop_rx_ctag_filter_info(dev); 10023 } 10024 } 10025 10026 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) { 10027 if (features & NETIF_F_HW_VLAN_STAG_FILTER) { 10028 dev->features = features; 10029 err |= vlan_get_rx_stag_filter_info(dev); 10030 } else { 10031 vlan_drop_rx_stag_filter_info(dev); 10032 } 10033 } 10034 10035 dev->features = features; 10036 } 10037 10038 return err < 0 ? 0 : 1; 10039 } 10040 10041 /** 10042 * netdev_update_features - recalculate device features 10043 * @dev: the device to check 10044 * 10045 * Recalculate dev->features set and send notifications if it 10046 * has changed. Should be called after driver or hardware dependent 10047 * conditions might have changed that influence the features. 10048 */ 10049 void netdev_update_features(struct net_device *dev) 10050 { 10051 if (__netdev_update_features(dev)) 10052 netdev_features_change(dev); 10053 } 10054 EXPORT_SYMBOL(netdev_update_features); 10055 10056 /** 10057 * netdev_change_features - recalculate device features 10058 * @dev: the device to check 10059 * 10060 * Recalculate dev->features set and send notifications even 10061 * if they have not changed. Should be called instead of 10062 * netdev_update_features() if also dev->vlan_features might 10063 * have changed to allow the changes to be propagated to stacked 10064 * VLAN devices. 10065 */ 10066 void netdev_change_features(struct net_device *dev) 10067 { 10068 __netdev_update_features(dev); 10069 netdev_features_change(dev); 10070 } 10071 EXPORT_SYMBOL(netdev_change_features); 10072 10073 /** 10074 * netif_stacked_transfer_operstate - transfer operstate 10075 * @rootdev: the root or lower level device to transfer state from 10076 * @dev: the device to transfer operstate to 10077 * 10078 * Transfer operational state from root to device. This is normally 10079 * called when a stacking relationship exists between the root 10080 * device and the device(a leaf device). 10081 */ 10082 void netif_stacked_transfer_operstate(const struct net_device *rootdev, 10083 struct net_device *dev) 10084 { 10085 if (rootdev->operstate == IF_OPER_DORMANT) 10086 netif_dormant_on(dev); 10087 else 10088 netif_dormant_off(dev); 10089 10090 if (rootdev->operstate == IF_OPER_TESTING) 10091 netif_testing_on(dev); 10092 else 10093 netif_testing_off(dev); 10094 10095 if (netif_carrier_ok(rootdev)) 10096 netif_carrier_on(dev); 10097 else 10098 netif_carrier_off(dev); 10099 } 10100 EXPORT_SYMBOL(netif_stacked_transfer_operstate); 10101 10102 static int netif_alloc_rx_queues(struct net_device *dev) 10103 { 10104 unsigned int i, count = dev->num_rx_queues; 10105 struct netdev_rx_queue *rx; 10106 size_t sz = count * sizeof(*rx); 10107 int err = 0; 10108 10109 BUG_ON(count < 1); 10110 10111 rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10112 if (!rx) 10113 return -ENOMEM; 10114 10115 dev->_rx = rx; 10116 10117 for (i = 0; i < count; i++) { 10118 rx[i].dev = dev; 10119 10120 /* XDP RX-queue setup */ 10121 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0); 10122 if (err < 0) 10123 goto err_rxq_info; 10124 } 10125 return 0; 10126 10127 err_rxq_info: 10128 /* Rollback successful reg's and free other resources */ 10129 while (i--) 10130 xdp_rxq_info_unreg(&rx[i].xdp_rxq); 10131 kvfree(dev->_rx); 10132 dev->_rx = NULL; 10133 return err; 10134 } 10135 10136 static void netif_free_rx_queues(struct net_device *dev) 10137 { 10138 unsigned int i, count = dev->num_rx_queues; 10139 10140 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */ 10141 if (!dev->_rx) 10142 return; 10143 10144 for (i = 0; i < count; i++) 10145 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq); 10146 10147 kvfree(dev->_rx); 10148 } 10149 10150 static void netdev_init_one_queue(struct net_device *dev, 10151 struct netdev_queue *queue, void *_unused) 10152 { 10153 /* Initialize queue lock */ 10154 spin_lock_init(&queue->_xmit_lock); 10155 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); 10156 queue->xmit_lock_owner = -1; 10157 netdev_queue_numa_node_write(queue, NUMA_NO_NODE); 10158 queue->dev = dev; 10159 #ifdef CONFIG_BQL 10160 dql_init(&queue->dql, HZ); 10161 #endif 10162 } 10163 10164 static void netif_free_tx_queues(struct net_device *dev) 10165 { 10166 kvfree(dev->_tx); 10167 } 10168 10169 static int netif_alloc_netdev_queues(struct net_device *dev) 10170 { 10171 unsigned int count = dev->num_tx_queues; 10172 struct netdev_queue *tx; 10173 size_t sz = count * sizeof(*tx); 10174 10175 if (count < 1 || count > 0xffff) 10176 return -EINVAL; 10177 10178 tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10179 if (!tx) 10180 return -ENOMEM; 10181 10182 dev->_tx = tx; 10183 10184 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); 10185 spin_lock_init(&dev->tx_global_lock); 10186 10187 return 0; 10188 } 10189 10190 void netif_tx_stop_all_queues(struct net_device *dev) 10191 { 10192 unsigned int i; 10193 10194 for (i = 0; i < dev->num_tx_queues; i++) { 10195 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 10196 10197 netif_tx_stop_queue(txq); 10198 } 10199 } 10200 EXPORT_SYMBOL(netif_tx_stop_all_queues); 10201 10202 /** 10203 * register_netdevice - register a network device 10204 * @dev: device to register 10205 * 10206 * Take a completed network device structure and add it to the kernel 10207 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 10208 * chain. 0 is returned on success. A negative errno code is returned 10209 * on a failure to set up the device, or if the name is a duplicate. 10210 * 10211 * Callers must hold the rtnl semaphore. You may want 10212 * register_netdev() instead of this. 10213 * 10214 * BUGS: 10215 * The locking appears insufficient to guarantee two parallel registers 10216 * will not get the same name. 10217 */ 10218 10219 int register_netdevice(struct net_device *dev) 10220 { 10221 int ret; 10222 struct net *net = dev_net(dev); 10223 10224 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE < 10225 NETDEV_FEATURE_COUNT); 10226 BUG_ON(dev_boot_phase); 10227 ASSERT_RTNL(); 10228 10229 might_sleep(); 10230 10231 /* When net_device's are persistent, this will be fatal. */ 10232 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); 10233 BUG_ON(!net); 10234 10235 ret = ethtool_check_ops(dev->ethtool_ops); 10236 if (ret) 10237 return ret; 10238 10239 spin_lock_init(&dev->addr_list_lock); 10240 netdev_set_addr_lockdep_class(dev); 10241 10242 ret = dev_get_valid_name(net, dev, dev->name); 10243 if (ret < 0) 10244 goto out; 10245 10246 ret = -ENOMEM; 10247 dev->name_node = netdev_name_node_head_alloc(dev); 10248 if (!dev->name_node) 10249 goto out; 10250 10251 /* Init, if this function is available */ 10252 if (dev->netdev_ops->ndo_init) { 10253 ret = dev->netdev_ops->ndo_init(dev); 10254 if (ret) { 10255 if (ret > 0) 10256 ret = -EIO; 10257 goto err_free_name; 10258 } 10259 } 10260 10261 if (((dev->hw_features | dev->features) & 10262 NETIF_F_HW_VLAN_CTAG_FILTER) && 10263 (!dev->netdev_ops->ndo_vlan_rx_add_vid || 10264 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { 10265 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); 10266 ret = -EINVAL; 10267 goto err_uninit; 10268 } 10269 10270 ret = -EBUSY; 10271 if (!dev->ifindex) 10272 dev->ifindex = dev_new_index(net); 10273 else if (__dev_get_by_index(net, dev->ifindex)) 10274 goto err_uninit; 10275 10276 /* Transfer changeable features to wanted_features and enable 10277 * software offloads (GSO and GRO). 10278 */ 10279 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF); 10280 dev->features |= NETIF_F_SOFT_FEATURES; 10281 10282 if (dev->udp_tunnel_nic_info) { 10283 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10284 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10285 } 10286 10287 dev->wanted_features = dev->features & dev->hw_features; 10288 10289 if (!(dev->flags & IFF_LOOPBACK)) 10290 dev->hw_features |= NETIF_F_NOCACHE_COPY; 10291 10292 /* If IPv4 TCP segmentation offload is supported we should also 10293 * allow the device to enable segmenting the frame with the option 10294 * of ignoring a static IP ID value. This doesn't enable the 10295 * feature itself but allows the user to enable it later. 10296 */ 10297 if (dev->hw_features & NETIF_F_TSO) 10298 dev->hw_features |= NETIF_F_TSO_MANGLEID; 10299 if (dev->vlan_features & NETIF_F_TSO) 10300 dev->vlan_features |= NETIF_F_TSO_MANGLEID; 10301 if (dev->mpls_features & NETIF_F_TSO) 10302 dev->mpls_features |= NETIF_F_TSO_MANGLEID; 10303 if (dev->hw_enc_features & NETIF_F_TSO) 10304 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; 10305 10306 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. 10307 */ 10308 dev->vlan_features |= NETIF_F_HIGHDMA; 10309 10310 /* Make NETIF_F_SG inheritable to tunnel devices. 10311 */ 10312 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; 10313 10314 /* Make NETIF_F_SG inheritable to MPLS. 10315 */ 10316 dev->mpls_features |= NETIF_F_SG; 10317 10318 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); 10319 ret = notifier_to_errno(ret); 10320 if (ret) 10321 goto err_uninit; 10322 10323 ret = netdev_register_kobject(dev); 10324 if (ret) { 10325 dev->reg_state = NETREG_UNREGISTERED; 10326 goto err_uninit; 10327 } 10328 dev->reg_state = NETREG_REGISTERED; 10329 10330 __netdev_update_features(dev); 10331 10332 /* 10333 * Default initial state at registry is that the 10334 * device is present. 10335 */ 10336 10337 set_bit(__LINK_STATE_PRESENT, &dev->state); 10338 10339 linkwatch_init_dev(dev); 10340 10341 dev_init_scheduler(dev); 10342 dev_hold(dev); 10343 list_netdevice(dev); 10344 add_device_randomness(dev->dev_addr, dev->addr_len); 10345 10346 /* If the device has permanent device address, driver should 10347 * set dev_addr and also addr_assign_type should be set to 10348 * NET_ADDR_PERM (default value). 10349 */ 10350 if (dev->addr_assign_type == NET_ADDR_PERM) 10351 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); 10352 10353 /* Notify protocols, that a new device appeared. */ 10354 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); 10355 ret = notifier_to_errno(ret); 10356 if (ret) { 10357 /* Expect explicit free_netdev() on failure */ 10358 dev->needs_free_netdev = false; 10359 unregister_netdevice_queue(dev, NULL); 10360 goto out; 10361 } 10362 /* 10363 * Prevent userspace races by waiting until the network 10364 * device is fully setup before sending notifications. 10365 */ 10366 if (!dev->rtnl_link_ops || 10367 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 10368 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); 10369 10370 out: 10371 return ret; 10372 10373 err_uninit: 10374 if (dev->netdev_ops->ndo_uninit) 10375 dev->netdev_ops->ndo_uninit(dev); 10376 if (dev->priv_destructor) 10377 dev->priv_destructor(dev); 10378 err_free_name: 10379 netdev_name_node_free(dev->name_node); 10380 goto out; 10381 } 10382 EXPORT_SYMBOL(register_netdevice); 10383 10384 /** 10385 * init_dummy_netdev - init a dummy network device for NAPI 10386 * @dev: device to init 10387 * 10388 * This takes a network device structure and initialize the minimum 10389 * amount of fields so it can be used to schedule NAPI polls without 10390 * registering a full blown interface. This is to be used by drivers 10391 * that need to tie several hardware interfaces to a single NAPI 10392 * poll scheduler due to HW limitations. 10393 */ 10394 int init_dummy_netdev(struct net_device *dev) 10395 { 10396 /* Clear everything. Note we don't initialize spinlocks 10397 * are they aren't supposed to be taken by any of the 10398 * NAPI code and this dummy netdev is supposed to be 10399 * only ever used for NAPI polls 10400 */ 10401 memset(dev, 0, sizeof(struct net_device)); 10402 10403 /* make sure we BUG if trying to hit standard 10404 * register/unregister code path 10405 */ 10406 dev->reg_state = NETREG_DUMMY; 10407 10408 /* NAPI wants this */ 10409 INIT_LIST_HEAD(&dev->napi_list); 10410 10411 /* a dummy interface is started by default */ 10412 set_bit(__LINK_STATE_PRESENT, &dev->state); 10413 set_bit(__LINK_STATE_START, &dev->state); 10414 10415 /* napi_busy_loop stats accounting wants this */ 10416 dev_net_set(dev, &init_net); 10417 10418 /* Note : We dont allocate pcpu_refcnt for dummy devices, 10419 * because users of this 'device' dont need to change 10420 * its refcount. 10421 */ 10422 10423 return 0; 10424 } 10425 EXPORT_SYMBOL_GPL(init_dummy_netdev); 10426 10427 10428 /** 10429 * register_netdev - register a network device 10430 * @dev: device to register 10431 * 10432 * Take a completed network device structure and add it to the kernel 10433 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 10434 * chain. 0 is returned on success. A negative errno code is returned 10435 * on a failure to set up the device, or if the name is a duplicate. 10436 * 10437 * This is a wrapper around register_netdevice that takes the rtnl semaphore 10438 * and expands the device name if you passed a format string to 10439 * alloc_netdev. 10440 */ 10441 int register_netdev(struct net_device *dev) 10442 { 10443 int err; 10444 10445 if (rtnl_lock_killable()) 10446 return -EINTR; 10447 err = register_netdevice(dev); 10448 rtnl_unlock(); 10449 return err; 10450 } 10451 EXPORT_SYMBOL(register_netdev); 10452 10453 int netdev_refcnt_read(const struct net_device *dev) 10454 { 10455 #ifdef CONFIG_PCPU_DEV_REFCNT 10456 int i, refcnt = 0; 10457 10458 for_each_possible_cpu(i) 10459 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); 10460 return refcnt; 10461 #else 10462 return refcount_read(&dev->dev_refcnt); 10463 #endif 10464 } 10465 EXPORT_SYMBOL(netdev_refcnt_read); 10466 10467 int netdev_unregister_timeout_secs __read_mostly = 10; 10468 10469 #define WAIT_REFS_MIN_MSECS 1 10470 #define WAIT_REFS_MAX_MSECS 250 10471 /** 10472 * netdev_wait_allrefs - wait until all references are gone. 10473 * @dev: target net_device 10474 * 10475 * This is called when unregistering network devices. 10476 * 10477 * Any protocol or device that holds a reference should register 10478 * for netdevice notification, and cleanup and put back the 10479 * reference if they receive an UNREGISTER event. 10480 * We can get stuck here if buggy protocols don't correctly 10481 * call dev_put. 10482 */ 10483 static void netdev_wait_allrefs(struct net_device *dev) 10484 { 10485 unsigned long rebroadcast_time, warning_time; 10486 int wait = 0, refcnt; 10487 10488 linkwatch_forget_dev(dev); 10489 10490 rebroadcast_time = warning_time = jiffies; 10491 refcnt = netdev_refcnt_read(dev); 10492 10493 while (refcnt != 1) { 10494 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { 10495 rtnl_lock(); 10496 10497 /* Rebroadcast unregister notification */ 10498 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10499 10500 __rtnl_unlock(); 10501 rcu_barrier(); 10502 rtnl_lock(); 10503 10504 if (test_bit(__LINK_STATE_LINKWATCH_PENDING, 10505 &dev->state)) { 10506 /* We must not have linkwatch events 10507 * pending on unregister. If this 10508 * happens, we simply run the queue 10509 * unscheduled, resulting in a noop 10510 * for this device. 10511 */ 10512 linkwatch_run_queue(); 10513 } 10514 10515 __rtnl_unlock(); 10516 10517 rebroadcast_time = jiffies; 10518 } 10519 10520 if (!wait) { 10521 rcu_barrier(); 10522 wait = WAIT_REFS_MIN_MSECS; 10523 } else { 10524 msleep(wait); 10525 wait = min(wait << 1, WAIT_REFS_MAX_MSECS); 10526 } 10527 10528 refcnt = netdev_refcnt_read(dev); 10529 10530 if (refcnt != 1 && 10531 time_after(jiffies, warning_time + 10532 netdev_unregister_timeout_secs * HZ)) { 10533 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", 10534 dev->name, refcnt); 10535 warning_time = jiffies; 10536 } 10537 } 10538 } 10539 10540 /* The sequence is: 10541 * 10542 * rtnl_lock(); 10543 * ... 10544 * register_netdevice(x1); 10545 * register_netdevice(x2); 10546 * ... 10547 * unregister_netdevice(y1); 10548 * unregister_netdevice(y2); 10549 * ... 10550 * rtnl_unlock(); 10551 * free_netdev(y1); 10552 * free_netdev(y2); 10553 * 10554 * We are invoked by rtnl_unlock(). 10555 * This allows us to deal with problems: 10556 * 1) We can delete sysfs objects which invoke hotplug 10557 * without deadlocking with linkwatch via keventd. 10558 * 2) Since we run with the RTNL semaphore not held, we can sleep 10559 * safely in order to wait for the netdev refcnt to drop to zero. 10560 * 10561 * We must not return until all unregister events added during 10562 * the interval the lock was held have been completed. 10563 */ 10564 void netdev_run_todo(void) 10565 { 10566 struct list_head list; 10567 #ifdef CONFIG_LOCKDEP 10568 struct list_head unlink_list; 10569 10570 list_replace_init(&net_unlink_list, &unlink_list); 10571 10572 while (!list_empty(&unlink_list)) { 10573 struct net_device *dev = list_first_entry(&unlink_list, 10574 struct net_device, 10575 unlink_list); 10576 list_del_init(&dev->unlink_list); 10577 dev->nested_level = dev->lower_level - 1; 10578 } 10579 #endif 10580 10581 /* Snapshot list, allow later requests */ 10582 list_replace_init(&net_todo_list, &list); 10583 10584 __rtnl_unlock(); 10585 10586 10587 /* Wait for rcu callbacks to finish before next phase */ 10588 if (!list_empty(&list)) 10589 rcu_barrier(); 10590 10591 while (!list_empty(&list)) { 10592 struct net_device *dev 10593 = list_first_entry(&list, struct net_device, todo_list); 10594 list_del(&dev->todo_list); 10595 10596 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { 10597 pr_err("network todo '%s' but state %d\n", 10598 dev->name, dev->reg_state); 10599 dump_stack(); 10600 continue; 10601 } 10602 10603 dev->reg_state = NETREG_UNREGISTERED; 10604 10605 netdev_wait_allrefs(dev); 10606 10607 /* paranoia */ 10608 BUG_ON(netdev_refcnt_read(dev) != 1); 10609 BUG_ON(!list_empty(&dev->ptype_all)); 10610 BUG_ON(!list_empty(&dev->ptype_specific)); 10611 WARN_ON(rcu_access_pointer(dev->ip_ptr)); 10612 WARN_ON(rcu_access_pointer(dev->ip6_ptr)); 10613 #if IS_ENABLED(CONFIG_DECNET) 10614 WARN_ON(dev->dn_ptr); 10615 #endif 10616 if (dev->priv_destructor) 10617 dev->priv_destructor(dev); 10618 if (dev->needs_free_netdev) 10619 free_netdev(dev); 10620 10621 /* Report a network device has been unregistered */ 10622 rtnl_lock(); 10623 dev_net(dev)->dev_unreg_count--; 10624 __rtnl_unlock(); 10625 wake_up(&netdev_unregistering_wq); 10626 10627 /* Free network device */ 10628 kobject_put(&dev->dev.kobj); 10629 } 10630 } 10631 10632 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has 10633 * all the same fields in the same order as net_device_stats, with only 10634 * the type differing, but rtnl_link_stats64 may have additional fields 10635 * at the end for newer counters. 10636 */ 10637 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, 10638 const struct net_device_stats *netdev_stats) 10639 { 10640 #if BITS_PER_LONG == 64 10641 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats)); 10642 memcpy(stats64, netdev_stats, sizeof(*netdev_stats)); 10643 /* zero out counters that only exist in rtnl_link_stats64 */ 10644 memset((char *)stats64 + sizeof(*netdev_stats), 0, 10645 sizeof(*stats64) - sizeof(*netdev_stats)); 10646 #else 10647 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long); 10648 const unsigned long *src = (const unsigned long *)netdev_stats; 10649 u64 *dst = (u64 *)stats64; 10650 10651 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); 10652 for (i = 0; i < n; i++) 10653 dst[i] = src[i]; 10654 /* zero out counters that only exist in rtnl_link_stats64 */ 10655 memset((char *)stats64 + n * sizeof(u64), 0, 10656 sizeof(*stats64) - n * sizeof(u64)); 10657 #endif 10658 } 10659 EXPORT_SYMBOL(netdev_stats_to_stats64); 10660 10661 /** 10662 * dev_get_stats - get network device statistics 10663 * @dev: device to get statistics from 10664 * @storage: place to store stats 10665 * 10666 * Get network statistics from device. Return @storage. 10667 * The device driver may provide its own method by setting 10668 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; 10669 * otherwise the internal statistics structure is used. 10670 */ 10671 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, 10672 struct rtnl_link_stats64 *storage) 10673 { 10674 const struct net_device_ops *ops = dev->netdev_ops; 10675 10676 if (ops->ndo_get_stats64) { 10677 memset(storage, 0, sizeof(*storage)); 10678 ops->ndo_get_stats64(dev, storage); 10679 } else if (ops->ndo_get_stats) { 10680 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); 10681 } else { 10682 netdev_stats_to_stats64(storage, &dev->stats); 10683 } 10684 storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped); 10685 storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped); 10686 storage->rx_nohandler += (unsigned long)atomic_long_read(&dev->rx_nohandler); 10687 return storage; 10688 } 10689 EXPORT_SYMBOL(dev_get_stats); 10690 10691 /** 10692 * dev_fetch_sw_netstats - get per-cpu network device statistics 10693 * @s: place to store stats 10694 * @netstats: per-cpu network stats to read from 10695 * 10696 * Read per-cpu network statistics and populate the related fields in @s. 10697 */ 10698 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s, 10699 const struct pcpu_sw_netstats __percpu *netstats) 10700 { 10701 int cpu; 10702 10703 for_each_possible_cpu(cpu) { 10704 const struct pcpu_sw_netstats *stats; 10705 struct pcpu_sw_netstats tmp; 10706 unsigned int start; 10707 10708 stats = per_cpu_ptr(netstats, cpu); 10709 do { 10710 start = u64_stats_fetch_begin_irq(&stats->syncp); 10711 tmp.rx_packets = stats->rx_packets; 10712 tmp.rx_bytes = stats->rx_bytes; 10713 tmp.tx_packets = stats->tx_packets; 10714 tmp.tx_bytes = stats->tx_bytes; 10715 } while (u64_stats_fetch_retry_irq(&stats->syncp, start)); 10716 10717 s->rx_packets += tmp.rx_packets; 10718 s->rx_bytes += tmp.rx_bytes; 10719 s->tx_packets += tmp.tx_packets; 10720 s->tx_bytes += tmp.tx_bytes; 10721 } 10722 } 10723 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats); 10724 10725 /** 10726 * dev_get_tstats64 - ndo_get_stats64 implementation 10727 * @dev: device to get statistics from 10728 * @s: place to store stats 10729 * 10730 * Populate @s from dev->stats and dev->tstats. Can be used as 10731 * ndo_get_stats64() callback. 10732 */ 10733 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s) 10734 { 10735 netdev_stats_to_stats64(s, &dev->stats); 10736 dev_fetch_sw_netstats(s, dev->tstats); 10737 } 10738 EXPORT_SYMBOL_GPL(dev_get_tstats64); 10739 10740 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) 10741 { 10742 struct netdev_queue *queue = dev_ingress_queue(dev); 10743 10744 #ifdef CONFIG_NET_CLS_ACT 10745 if (queue) 10746 return queue; 10747 queue = kzalloc(sizeof(*queue), GFP_KERNEL); 10748 if (!queue) 10749 return NULL; 10750 netdev_init_one_queue(dev, queue, NULL); 10751 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); 10752 queue->qdisc_sleeping = &noop_qdisc; 10753 rcu_assign_pointer(dev->ingress_queue, queue); 10754 #endif 10755 return queue; 10756 } 10757 10758 static const struct ethtool_ops default_ethtool_ops; 10759 10760 void netdev_set_default_ethtool_ops(struct net_device *dev, 10761 const struct ethtool_ops *ops) 10762 { 10763 if (dev->ethtool_ops == &default_ethtool_ops) 10764 dev->ethtool_ops = ops; 10765 } 10766 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); 10767 10768 void netdev_freemem(struct net_device *dev) 10769 { 10770 char *addr = (char *)dev - dev->padded; 10771 10772 kvfree(addr); 10773 } 10774 10775 /** 10776 * alloc_netdev_mqs - allocate network device 10777 * @sizeof_priv: size of private data to allocate space for 10778 * @name: device name format string 10779 * @name_assign_type: origin of device name 10780 * @setup: callback to initialize device 10781 * @txqs: the number of TX subqueues to allocate 10782 * @rxqs: the number of RX subqueues to allocate 10783 * 10784 * Allocates a struct net_device with private data area for driver use 10785 * and performs basic initialization. Also allocates subqueue structs 10786 * for each queue on the device. 10787 */ 10788 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, 10789 unsigned char name_assign_type, 10790 void (*setup)(struct net_device *), 10791 unsigned int txqs, unsigned int rxqs) 10792 { 10793 struct net_device *dev; 10794 unsigned int alloc_size; 10795 struct net_device *p; 10796 10797 BUG_ON(strlen(name) >= sizeof(dev->name)); 10798 10799 if (txqs < 1) { 10800 pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); 10801 return NULL; 10802 } 10803 10804 if (rxqs < 1) { 10805 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); 10806 return NULL; 10807 } 10808 10809 alloc_size = sizeof(struct net_device); 10810 if (sizeof_priv) { 10811 /* ensure 32-byte alignment of private area */ 10812 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); 10813 alloc_size += sizeof_priv; 10814 } 10815 /* ensure 32-byte alignment of whole construct */ 10816 alloc_size += NETDEV_ALIGN - 1; 10817 10818 p = kvzalloc(alloc_size, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10819 if (!p) 10820 return NULL; 10821 10822 dev = PTR_ALIGN(p, NETDEV_ALIGN); 10823 dev->padded = (char *)dev - (char *)p; 10824 10825 #ifdef CONFIG_PCPU_DEV_REFCNT 10826 dev->pcpu_refcnt = alloc_percpu(int); 10827 if (!dev->pcpu_refcnt) 10828 goto free_dev; 10829 dev_hold(dev); 10830 #else 10831 refcount_set(&dev->dev_refcnt, 1); 10832 #endif 10833 10834 if (dev_addr_init(dev)) 10835 goto free_pcpu; 10836 10837 dev_mc_init(dev); 10838 dev_uc_init(dev); 10839 10840 dev_net_set(dev, &init_net); 10841 10842 dev->gso_max_size = GSO_MAX_SIZE; 10843 dev->gso_max_segs = GSO_MAX_SEGS; 10844 dev->upper_level = 1; 10845 dev->lower_level = 1; 10846 #ifdef CONFIG_LOCKDEP 10847 dev->nested_level = 0; 10848 INIT_LIST_HEAD(&dev->unlink_list); 10849 #endif 10850 10851 INIT_LIST_HEAD(&dev->napi_list); 10852 INIT_LIST_HEAD(&dev->unreg_list); 10853 INIT_LIST_HEAD(&dev->close_list); 10854 INIT_LIST_HEAD(&dev->link_watch_list); 10855 INIT_LIST_HEAD(&dev->adj_list.upper); 10856 INIT_LIST_HEAD(&dev->adj_list.lower); 10857 INIT_LIST_HEAD(&dev->ptype_all); 10858 INIT_LIST_HEAD(&dev->ptype_specific); 10859 INIT_LIST_HEAD(&dev->net_notifier_list); 10860 #ifdef CONFIG_NET_SCHED 10861 hash_init(dev->qdisc_hash); 10862 #endif 10863 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; 10864 setup(dev); 10865 10866 if (!dev->tx_queue_len) { 10867 dev->priv_flags |= IFF_NO_QUEUE; 10868 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; 10869 } 10870 10871 dev->num_tx_queues = txqs; 10872 dev->real_num_tx_queues = txqs; 10873 if (netif_alloc_netdev_queues(dev)) 10874 goto free_all; 10875 10876 dev->num_rx_queues = rxqs; 10877 dev->real_num_rx_queues = rxqs; 10878 if (netif_alloc_rx_queues(dev)) 10879 goto free_all; 10880 10881 strcpy(dev->name, name); 10882 dev->name_assign_type = name_assign_type; 10883 dev->group = INIT_NETDEV_GROUP; 10884 if (!dev->ethtool_ops) 10885 dev->ethtool_ops = &default_ethtool_ops; 10886 10887 nf_hook_netdev_init(dev); 10888 10889 return dev; 10890 10891 free_all: 10892 free_netdev(dev); 10893 return NULL; 10894 10895 free_pcpu: 10896 #ifdef CONFIG_PCPU_DEV_REFCNT 10897 free_percpu(dev->pcpu_refcnt); 10898 free_dev: 10899 #endif 10900 netdev_freemem(dev); 10901 return NULL; 10902 } 10903 EXPORT_SYMBOL(alloc_netdev_mqs); 10904 10905 /** 10906 * free_netdev - free network device 10907 * @dev: device 10908 * 10909 * This function does the last stage of destroying an allocated device 10910 * interface. The reference to the device object is released. If this 10911 * is the last reference then it will be freed.Must be called in process 10912 * context. 10913 */ 10914 void free_netdev(struct net_device *dev) 10915 { 10916 struct napi_struct *p, *n; 10917 10918 might_sleep(); 10919 10920 /* When called immediately after register_netdevice() failed the unwind 10921 * handling may still be dismantling the device. Handle that case by 10922 * deferring the free. 10923 */ 10924 if (dev->reg_state == NETREG_UNREGISTERING) { 10925 ASSERT_RTNL(); 10926 dev->needs_free_netdev = true; 10927 return; 10928 } 10929 10930 netif_free_tx_queues(dev); 10931 netif_free_rx_queues(dev); 10932 10933 kfree(rcu_dereference_protected(dev->ingress_queue, 1)); 10934 10935 /* Flush device addresses */ 10936 dev_addr_flush(dev); 10937 10938 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) 10939 netif_napi_del(p); 10940 10941 #ifdef CONFIG_PCPU_DEV_REFCNT 10942 free_percpu(dev->pcpu_refcnt); 10943 dev->pcpu_refcnt = NULL; 10944 #endif 10945 free_percpu(dev->xdp_bulkq); 10946 dev->xdp_bulkq = NULL; 10947 10948 /* Compatibility with error handling in drivers */ 10949 if (dev->reg_state == NETREG_UNINITIALIZED) { 10950 netdev_freemem(dev); 10951 return; 10952 } 10953 10954 BUG_ON(dev->reg_state != NETREG_UNREGISTERED); 10955 dev->reg_state = NETREG_RELEASED; 10956 10957 /* will free via device release */ 10958 put_device(&dev->dev); 10959 } 10960 EXPORT_SYMBOL(free_netdev); 10961 10962 /** 10963 * synchronize_net - Synchronize with packet receive processing 10964 * 10965 * Wait for packets currently being received to be done. 10966 * Does not block later packets from starting. 10967 */ 10968 void synchronize_net(void) 10969 { 10970 might_sleep(); 10971 if (rtnl_is_locked()) 10972 synchronize_rcu_expedited(); 10973 else 10974 synchronize_rcu(); 10975 } 10976 EXPORT_SYMBOL(synchronize_net); 10977 10978 /** 10979 * unregister_netdevice_queue - remove device from the kernel 10980 * @dev: device 10981 * @head: list 10982 * 10983 * This function shuts down a device interface and removes it 10984 * from the kernel tables. 10985 * If head not NULL, device is queued to be unregistered later. 10986 * 10987 * Callers must hold the rtnl semaphore. You may want 10988 * unregister_netdev() instead of this. 10989 */ 10990 10991 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) 10992 { 10993 ASSERT_RTNL(); 10994 10995 if (head) { 10996 list_move_tail(&dev->unreg_list, head); 10997 } else { 10998 LIST_HEAD(single); 10999 11000 list_add(&dev->unreg_list, &single); 11001 unregister_netdevice_many(&single); 11002 } 11003 } 11004 EXPORT_SYMBOL(unregister_netdevice_queue); 11005 11006 /** 11007 * unregister_netdevice_many - unregister many devices 11008 * @head: list of devices 11009 * 11010 * Note: As most callers use a stack allocated list_head, 11011 * we force a list_del() to make sure stack wont be corrupted later. 11012 */ 11013 void unregister_netdevice_many(struct list_head *head) 11014 { 11015 struct net_device *dev, *tmp; 11016 LIST_HEAD(close_head); 11017 11018 BUG_ON(dev_boot_phase); 11019 ASSERT_RTNL(); 11020 11021 if (list_empty(head)) 11022 return; 11023 11024 list_for_each_entry_safe(dev, tmp, head, unreg_list) { 11025 /* Some devices call without registering 11026 * for initialization unwind. Remove those 11027 * devices and proceed with the remaining. 11028 */ 11029 if (dev->reg_state == NETREG_UNINITIALIZED) { 11030 pr_debug("unregister_netdevice: device %s/%p never was registered\n", 11031 dev->name, dev); 11032 11033 WARN_ON(1); 11034 list_del(&dev->unreg_list); 11035 continue; 11036 } 11037 dev->dismantle = true; 11038 BUG_ON(dev->reg_state != NETREG_REGISTERED); 11039 } 11040 11041 /* If device is running, close it first. */ 11042 list_for_each_entry(dev, head, unreg_list) 11043 list_add_tail(&dev->close_list, &close_head); 11044 dev_close_many(&close_head, true); 11045 11046 list_for_each_entry(dev, head, unreg_list) { 11047 /* And unlink it from device chain. */ 11048 unlist_netdevice(dev); 11049 11050 dev->reg_state = NETREG_UNREGISTERING; 11051 } 11052 flush_all_backlogs(); 11053 11054 synchronize_net(); 11055 11056 list_for_each_entry(dev, head, unreg_list) { 11057 struct sk_buff *skb = NULL; 11058 11059 /* Shutdown queueing discipline. */ 11060 dev_shutdown(dev); 11061 11062 dev_xdp_uninstall(dev); 11063 11064 /* Notify protocols, that we are about to destroy 11065 * this device. They should clean all the things. 11066 */ 11067 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 11068 11069 if (!dev->rtnl_link_ops || 11070 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 11071 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0, 11072 GFP_KERNEL, NULL, 0); 11073 11074 /* 11075 * Flush the unicast and multicast chains 11076 */ 11077 dev_uc_flush(dev); 11078 dev_mc_flush(dev); 11079 11080 netdev_name_node_alt_flush(dev); 11081 netdev_name_node_free(dev->name_node); 11082 11083 if (dev->netdev_ops->ndo_uninit) 11084 dev->netdev_ops->ndo_uninit(dev); 11085 11086 if (skb) 11087 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL); 11088 11089 /* Notifier chain MUST detach us all upper devices. */ 11090 WARN_ON(netdev_has_any_upper_dev(dev)); 11091 WARN_ON(netdev_has_any_lower_dev(dev)); 11092 11093 /* Remove entries from kobject tree */ 11094 netdev_unregister_kobject(dev); 11095 #ifdef CONFIG_XPS 11096 /* Remove XPS queueing entries */ 11097 netif_reset_xps_queues_gt(dev, 0); 11098 #endif 11099 } 11100 11101 synchronize_net(); 11102 11103 list_for_each_entry(dev, head, unreg_list) { 11104 dev_put(dev); 11105 net_set_todo(dev); 11106 } 11107 11108 list_del(head); 11109 } 11110 EXPORT_SYMBOL(unregister_netdevice_many); 11111 11112 /** 11113 * unregister_netdev - remove device from the kernel 11114 * @dev: device 11115 * 11116 * This function shuts down a device interface and removes it 11117 * from the kernel tables. 11118 * 11119 * This is just a wrapper for unregister_netdevice that takes 11120 * the rtnl semaphore. In general you want to use this and not 11121 * unregister_netdevice. 11122 */ 11123 void unregister_netdev(struct net_device *dev) 11124 { 11125 rtnl_lock(); 11126 unregister_netdevice(dev); 11127 rtnl_unlock(); 11128 } 11129 EXPORT_SYMBOL(unregister_netdev); 11130 11131 /** 11132 * __dev_change_net_namespace - move device to different nethost namespace 11133 * @dev: device 11134 * @net: network namespace 11135 * @pat: If not NULL name pattern to try if the current device name 11136 * is already taken in the destination network namespace. 11137 * @new_ifindex: If not zero, specifies device index in the target 11138 * namespace. 11139 * 11140 * This function shuts down a device interface and moves it 11141 * to a new network namespace. On success 0 is returned, on 11142 * a failure a netagive errno code is returned. 11143 * 11144 * Callers must hold the rtnl semaphore. 11145 */ 11146 11147 int __dev_change_net_namespace(struct net_device *dev, struct net *net, 11148 const char *pat, int new_ifindex) 11149 { 11150 struct net *net_old = dev_net(dev); 11151 int err, new_nsid; 11152 11153 ASSERT_RTNL(); 11154 11155 /* Don't allow namespace local devices to be moved. */ 11156 err = -EINVAL; 11157 if (dev->features & NETIF_F_NETNS_LOCAL) 11158 goto out; 11159 11160 /* Ensure the device has been registrered */ 11161 if (dev->reg_state != NETREG_REGISTERED) 11162 goto out; 11163 11164 /* Get out if there is nothing todo */ 11165 err = 0; 11166 if (net_eq(net_old, net)) 11167 goto out; 11168 11169 /* Pick the destination device name, and ensure 11170 * we can use it in the destination network namespace. 11171 */ 11172 err = -EEXIST; 11173 if (netdev_name_in_use(net, dev->name)) { 11174 /* We get here if we can't use the current device name */ 11175 if (!pat) 11176 goto out; 11177 err = dev_get_valid_name(net, dev, pat); 11178 if (err < 0) 11179 goto out; 11180 } 11181 11182 /* Check that new_ifindex isn't used yet. */ 11183 err = -EBUSY; 11184 if (new_ifindex && __dev_get_by_index(net, new_ifindex)) 11185 goto out; 11186 11187 /* 11188 * And now a mini version of register_netdevice unregister_netdevice. 11189 */ 11190 11191 /* If device is running close it first. */ 11192 dev_close(dev); 11193 11194 /* And unlink it from device chain */ 11195 unlist_netdevice(dev); 11196 11197 synchronize_net(); 11198 11199 /* Shutdown queueing discipline. */ 11200 dev_shutdown(dev); 11201 11202 /* Notify protocols, that we are about to destroy 11203 * this device. They should clean all the things. 11204 * 11205 * Note that dev->reg_state stays at NETREG_REGISTERED. 11206 * This is wanted because this way 8021q and macvlan know 11207 * the device is just moving and can keep their slaves up. 11208 */ 11209 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 11210 rcu_barrier(); 11211 11212 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL); 11213 /* If there is an ifindex conflict assign a new one */ 11214 if (!new_ifindex) { 11215 if (__dev_get_by_index(net, dev->ifindex)) 11216 new_ifindex = dev_new_index(net); 11217 else 11218 new_ifindex = dev->ifindex; 11219 } 11220 11221 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid, 11222 new_ifindex); 11223 11224 /* 11225 * Flush the unicast and multicast chains 11226 */ 11227 dev_uc_flush(dev); 11228 dev_mc_flush(dev); 11229 11230 /* Send a netdev-removed uevent to the old namespace */ 11231 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); 11232 netdev_adjacent_del_links(dev); 11233 11234 /* Move per-net netdevice notifiers that are following the netdevice */ 11235 move_netdevice_notifiers_dev_net(dev, net); 11236 11237 /* Actually switch the network namespace */ 11238 dev_net_set(dev, net); 11239 dev->ifindex = new_ifindex; 11240 11241 /* Send a netdev-add uevent to the new namespace */ 11242 kobject_uevent(&dev->dev.kobj, KOBJ_ADD); 11243 netdev_adjacent_add_links(dev); 11244 11245 /* Fixup kobjects */ 11246 err = device_rename(&dev->dev, dev->name); 11247 WARN_ON(err); 11248 11249 /* Adapt owner in case owning user namespace of target network 11250 * namespace is different from the original one. 11251 */ 11252 err = netdev_change_owner(dev, net_old, net); 11253 WARN_ON(err); 11254 11255 /* Add the device back in the hashes */ 11256 list_netdevice(dev); 11257 11258 /* Notify protocols, that a new device appeared. */ 11259 call_netdevice_notifiers(NETDEV_REGISTER, dev); 11260 11261 /* 11262 * Prevent userspace races by waiting until the network 11263 * device is fully setup before sending notifications. 11264 */ 11265 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); 11266 11267 synchronize_net(); 11268 err = 0; 11269 out: 11270 return err; 11271 } 11272 EXPORT_SYMBOL_GPL(__dev_change_net_namespace); 11273 11274 static int dev_cpu_dead(unsigned int oldcpu) 11275 { 11276 struct sk_buff **list_skb; 11277 struct sk_buff *skb; 11278 unsigned int cpu; 11279 struct softnet_data *sd, *oldsd, *remsd = NULL; 11280 11281 local_irq_disable(); 11282 cpu = smp_processor_id(); 11283 sd = &per_cpu(softnet_data, cpu); 11284 oldsd = &per_cpu(softnet_data, oldcpu); 11285 11286 /* Find end of our completion_queue. */ 11287 list_skb = &sd->completion_queue; 11288 while (*list_skb) 11289 list_skb = &(*list_skb)->next; 11290 /* Append completion queue from offline CPU. */ 11291 *list_skb = oldsd->completion_queue; 11292 oldsd->completion_queue = NULL; 11293 11294 /* Append output queue from offline CPU. */ 11295 if (oldsd->output_queue) { 11296 *sd->output_queue_tailp = oldsd->output_queue; 11297 sd->output_queue_tailp = oldsd->output_queue_tailp; 11298 oldsd->output_queue = NULL; 11299 oldsd->output_queue_tailp = &oldsd->output_queue; 11300 } 11301 /* Append NAPI poll list from offline CPU, with one exception : 11302 * process_backlog() must be called by cpu owning percpu backlog. 11303 * We properly handle process_queue & input_pkt_queue later. 11304 */ 11305 while (!list_empty(&oldsd->poll_list)) { 11306 struct napi_struct *napi = list_first_entry(&oldsd->poll_list, 11307 struct napi_struct, 11308 poll_list); 11309 11310 list_del_init(&napi->poll_list); 11311 if (napi->poll == process_backlog) 11312 napi->state = 0; 11313 else 11314 ____napi_schedule(sd, napi); 11315 } 11316 11317 raise_softirq_irqoff(NET_TX_SOFTIRQ); 11318 local_irq_enable(); 11319 11320 #ifdef CONFIG_RPS 11321 remsd = oldsd->rps_ipi_list; 11322 oldsd->rps_ipi_list = NULL; 11323 #endif 11324 /* send out pending IPI's on offline CPU */ 11325 net_rps_send_ipi(remsd); 11326 11327 /* Process offline CPU's input_pkt_queue */ 11328 while ((skb = __skb_dequeue(&oldsd->process_queue))) { 11329 netif_rx_ni(skb); 11330 input_queue_head_incr(oldsd); 11331 } 11332 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { 11333 netif_rx_ni(skb); 11334 input_queue_head_incr(oldsd); 11335 } 11336 11337 return 0; 11338 } 11339 11340 /** 11341 * netdev_increment_features - increment feature set by one 11342 * @all: current feature set 11343 * @one: new feature set 11344 * @mask: mask feature set 11345 * 11346 * Computes a new feature set after adding a device with feature set 11347 * @one to the master device with current feature set @all. Will not 11348 * enable anything that is off in @mask. Returns the new feature set. 11349 */ 11350 netdev_features_t netdev_increment_features(netdev_features_t all, 11351 netdev_features_t one, netdev_features_t mask) 11352 { 11353 if (mask & NETIF_F_HW_CSUM) 11354 mask |= NETIF_F_CSUM_MASK; 11355 mask |= NETIF_F_VLAN_CHALLENGED; 11356 11357 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; 11358 all &= one | ~NETIF_F_ALL_FOR_ALL; 11359 11360 /* If one device supports hw checksumming, set for all. */ 11361 if (all & NETIF_F_HW_CSUM) 11362 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); 11363 11364 return all; 11365 } 11366 EXPORT_SYMBOL(netdev_increment_features); 11367 11368 static struct hlist_head * __net_init netdev_create_hash(void) 11369 { 11370 int i; 11371 struct hlist_head *hash; 11372 11373 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL); 11374 if (hash != NULL) 11375 for (i = 0; i < NETDEV_HASHENTRIES; i++) 11376 INIT_HLIST_HEAD(&hash[i]); 11377 11378 return hash; 11379 } 11380 11381 /* Initialize per network namespace state */ 11382 static int __net_init netdev_init(struct net *net) 11383 { 11384 BUILD_BUG_ON(GRO_HASH_BUCKETS > 11385 8 * sizeof_field(struct napi_struct, gro_bitmask)); 11386 11387 if (net != &init_net) 11388 INIT_LIST_HEAD(&net->dev_base_head); 11389 11390 net->dev_name_head = netdev_create_hash(); 11391 if (net->dev_name_head == NULL) 11392 goto err_name; 11393 11394 net->dev_index_head = netdev_create_hash(); 11395 if (net->dev_index_head == NULL) 11396 goto err_idx; 11397 11398 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain); 11399 11400 return 0; 11401 11402 err_idx: 11403 kfree(net->dev_name_head); 11404 err_name: 11405 return -ENOMEM; 11406 } 11407 11408 /** 11409 * netdev_drivername - network driver for the device 11410 * @dev: network device 11411 * 11412 * Determine network driver for device. 11413 */ 11414 const char *netdev_drivername(const struct net_device *dev) 11415 { 11416 const struct device_driver *driver; 11417 const struct device *parent; 11418 const char *empty = ""; 11419 11420 parent = dev->dev.parent; 11421 if (!parent) 11422 return empty; 11423 11424 driver = parent->driver; 11425 if (driver && driver->name) 11426 return driver->name; 11427 return empty; 11428 } 11429 11430 static void __netdev_printk(const char *level, const struct net_device *dev, 11431 struct va_format *vaf) 11432 { 11433 if (dev && dev->dev.parent) { 11434 dev_printk_emit(level[1] - '0', 11435 dev->dev.parent, 11436 "%s %s %s%s: %pV", 11437 dev_driver_string(dev->dev.parent), 11438 dev_name(dev->dev.parent), 11439 netdev_name(dev), netdev_reg_state(dev), 11440 vaf); 11441 } else if (dev) { 11442 printk("%s%s%s: %pV", 11443 level, netdev_name(dev), netdev_reg_state(dev), vaf); 11444 } else { 11445 printk("%s(NULL net_device): %pV", level, vaf); 11446 } 11447 } 11448 11449 void netdev_printk(const char *level, const struct net_device *dev, 11450 const char *format, ...) 11451 { 11452 struct va_format vaf; 11453 va_list args; 11454 11455 va_start(args, format); 11456 11457 vaf.fmt = format; 11458 vaf.va = &args; 11459 11460 __netdev_printk(level, dev, &vaf); 11461 11462 va_end(args); 11463 } 11464 EXPORT_SYMBOL(netdev_printk); 11465 11466 #define define_netdev_printk_level(func, level) \ 11467 void func(const struct net_device *dev, const char *fmt, ...) \ 11468 { \ 11469 struct va_format vaf; \ 11470 va_list args; \ 11471 \ 11472 va_start(args, fmt); \ 11473 \ 11474 vaf.fmt = fmt; \ 11475 vaf.va = &args; \ 11476 \ 11477 __netdev_printk(level, dev, &vaf); \ 11478 \ 11479 va_end(args); \ 11480 } \ 11481 EXPORT_SYMBOL(func); 11482 11483 define_netdev_printk_level(netdev_emerg, KERN_EMERG); 11484 define_netdev_printk_level(netdev_alert, KERN_ALERT); 11485 define_netdev_printk_level(netdev_crit, KERN_CRIT); 11486 define_netdev_printk_level(netdev_err, KERN_ERR); 11487 define_netdev_printk_level(netdev_warn, KERN_WARNING); 11488 define_netdev_printk_level(netdev_notice, KERN_NOTICE); 11489 define_netdev_printk_level(netdev_info, KERN_INFO); 11490 11491 static void __net_exit netdev_exit(struct net *net) 11492 { 11493 kfree(net->dev_name_head); 11494 kfree(net->dev_index_head); 11495 if (net != &init_net) 11496 WARN_ON_ONCE(!list_empty(&net->dev_base_head)); 11497 } 11498 11499 static struct pernet_operations __net_initdata netdev_net_ops = { 11500 .init = netdev_init, 11501 .exit = netdev_exit, 11502 }; 11503 11504 static void __net_exit default_device_exit(struct net *net) 11505 { 11506 struct net_device *dev, *aux; 11507 /* 11508 * Push all migratable network devices back to the 11509 * initial network namespace 11510 */ 11511 rtnl_lock(); 11512 for_each_netdev_safe(net, dev, aux) { 11513 int err; 11514 char fb_name[IFNAMSIZ]; 11515 11516 /* Ignore unmoveable devices (i.e. loopback) */ 11517 if (dev->features & NETIF_F_NETNS_LOCAL) 11518 continue; 11519 11520 /* Leave virtual devices for the generic cleanup */ 11521 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund) 11522 continue; 11523 11524 /* Push remaining network devices to init_net */ 11525 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); 11526 if (netdev_name_in_use(&init_net, fb_name)) 11527 snprintf(fb_name, IFNAMSIZ, "dev%%d"); 11528 err = dev_change_net_namespace(dev, &init_net, fb_name); 11529 if (err) { 11530 pr_emerg("%s: failed to move %s to init_net: %d\n", 11531 __func__, dev->name, err); 11532 BUG(); 11533 } 11534 } 11535 rtnl_unlock(); 11536 } 11537 11538 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list) 11539 { 11540 /* Return with the rtnl_lock held when there are no network 11541 * devices unregistering in any network namespace in net_list. 11542 */ 11543 struct net *net; 11544 bool unregistering; 11545 DEFINE_WAIT_FUNC(wait, woken_wake_function); 11546 11547 add_wait_queue(&netdev_unregistering_wq, &wait); 11548 for (;;) { 11549 unregistering = false; 11550 rtnl_lock(); 11551 list_for_each_entry(net, net_list, exit_list) { 11552 if (net->dev_unreg_count > 0) { 11553 unregistering = true; 11554 break; 11555 } 11556 } 11557 if (!unregistering) 11558 break; 11559 __rtnl_unlock(); 11560 11561 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); 11562 } 11563 remove_wait_queue(&netdev_unregistering_wq, &wait); 11564 } 11565 11566 static void __net_exit default_device_exit_batch(struct list_head *net_list) 11567 { 11568 /* At exit all network devices most be removed from a network 11569 * namespace. Do this in the reverse order of registration. 11570 * Do this across as many network namespaces as possible to 11571 * improve batching efficiency. 11572 */ 11573 struct net_device *dev; 11574 struct net *net; 11575 LIST_HEAD(dev_kill_list); 11576 11577 /* To prevent network device cleanup code from dereferencing 11578 * loopback devices or network devices that have been freed 11579 * wait here for all pending unregistrations to complete, 11580 * before unregistring the loopback device and allowing the 11581 * network namespace be freed. 11582 * 11583 * The netdev todo list containing all network devices 11584 * unregistrations that happen in default_device_exit_batch 11585 * will run in the rtnl_unlock() at the end of 11586 * default_device_exit_batch. 11587 */ 11588 rtnl_lock_unregistering(net_list); 11589 list_for_each_entry(net, net_list, exit_list) { 11590 for_each_netdev_reverse(net, dev) { 11591 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) 11592 dev->rtnl_link_ops->dellink(dev, &dev_kill_list); 11593 else 11594 unregister_netdevice_queue(dev, &dev_kill_list); 11595 } 11596 } 11597 unregister_netdevice_many(&dev_kill_list); 11598 rtnl_unlock(); 11599 } 11600 11601 static struct pernet_operations __net_initdata default_device_ops = { 11602 .exit = default_device_exit, 11603 .exit_batch = default_device_exit_batch, 11604 }; 11605 11606 /* 11607 * Initialize the DEV module. At boot time this walks the device list and 11608 * unhooks any devices that fail to initialise (normally hardware not 11609 * present) and leaves us with a valid list of present and active devices. 11610 * 11611 */ 11612 11613 /* 11614 * This is called single threaded during boot, so no need 11615 * to take the rtnl semaphore. 11616 */ 11617 static int __init net_dev_init(void) 11618 { 11619 int i, rc = -ENOMEM; 11620 11621 BUG_ON(!dev_boot_phase); 11622 11623 if (dev_proc_init()) 11624 goto out; 11625 11626 if (netdev_kobject_init()) 11627 goto out; 11628 11629 INIT_LIST_HEAD(&ptype_all); 11630 for (i = 0; i < PTYPE_HASH_SIZE; i++) 11631 INIT_LIST_HEAD(&ptype_base[i]); 11632 11633 INIT_LIST_HEAD(&offload_base); 11634 11635 if (register_pernet_subsys(&netdev_net_ops)) 11636 goto out; 11637 11638 /* 11639 * Initialise the packet receive queues. 11640 */ 11641 11642 for_each_possible_cpu(i) { 11643 struct work_struct *flush = per_cpu_ptr(&flush_works, i); 11644 struct softnet_data *sd = &per_cpu(softnet_data, i); 11645 11646 INIT_WORK(flush, flush_backlog); 11647 11648 skb_queue_head_init(&sd->input_pkt_queue); 11649 skb_queue_head_init(&sd->process_queue); 11650 #ifdef CONFIG_XFRM_OFFLOAD 11651 skb_queue_head_init(&sd->xfrm_backlog); 11652 #endif 11653 INIT_LIST_HEAD(&sd->poll_list); 11654 sd->output_queue_tailp = &sd->output_queue; 11655 #ifdef CONFIG_RPS 11656 INIT_CSD(&sd->csd, rps_trigger_softirq, sd); 11657 sd->cpu = i; 11658 #endif 11659 11660 init_gro_hash(&sd->backlog); 11661 sd->backlog.poll = process_backlog; 11662 sd->backlog.weight = weight_p; 11663 } 11664 11665 dev_boot_phase = 0; 11666 11667 /* The loopback device is special if any other network devices 11668 * is present in a network namespace the loopback device must 11669 * be present. Since we now dynamically allocate and free the 11670 * loopback device ensure this invariant is maintained by 11671 * keeping the loopback device as the first device on the 11672 * list of network devices. Ensuring the loopback devices 11673 * is the first device that appears and the last network device 11674 * that disappears. 11675 */ 11676 if (register_pernet_device(&loopback_net_ops)) 11677 goto out; 11678 11679 if (register_pernet_device(&default_device_ops)) 11680 goto out; 11681 11682 open_softirq(NET_TX_SOFTIRQ, net_tx_action); 11683 open_softirq(NET_RX_SOFTIRQ, net_rx_action); 11684 11685 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", 11686 NULL, dev_cpu_dead); 11687 WARN_ON(rc < 0); 11688 rc = 0; 11689 out: 11690 return rc; 11691 } 11692 11693 subsys_initcall(net_dev_init); 11694