1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * NET3 Protocol independent device support routines. 4 * 5 * Derived from the non IP parts of dev.c 1.0.19 6 * Authors: Ross Biro 7 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 8 * Mark Evans, <evansmp@uhura.aston.ac.uk> 9 * 10 * Additional Authors: 11 * Florian la Roche <rzsfl@rz.uni-sb.de> 12 * Alan Cox <gw4pts@gw4pts.ampr.org> 13 * David Hinds <dahinds@users.sourceforge.net> 14 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru> 15 * Adam Sulmicki <adam@cfar.umd.edu> 16 * Pekka Riikonen <priikone@poesidon.pspt.fi> 17 * 18 * Changes: 19 * D.J. Barrow : Fixed bug where dev->refcnt gets set 20 * to 2 if register_netdev gets called 21 * before net_dev_init & also removed a 22 * few lines of code in the process. 23 * Alan Cox : device private ioctl copies fields back. 24 * Alan Cox : Transmit queue code does relevant 25 * stunts to keep the queue safe. 26 * Alan Cox : Fixed double lock. 27 * Alan Cox : Fixed promisc NULL pointer trap 28 * ???????? : Support the full private ioctl range 29 * Alan Cox : Moved ioctl permission check into 30 * drivers 31 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI 32 * Alan Cox : 100 backlog just doesn't cut it when 33 * you start doing multicast video 8) 34 * Alan Cox : Rewrote net_bh and list manager. 35 * Alan Cox : Fix ETH_P_ALL echoback lengths. 36 * Alan Cox : Took out transmit every packet pass 37 * Saved a few bytes in the ioctl handler 38 * Alan Cox : Network driver sets packet type before 39 * calling netif_rx. Saves a function 40 * call a packet. 41 * Alan Cox : Hashed net_bh() 42 * Richard Kooijman: Timestamp fixes. 43 * Alan Cox : Wrong field in SIOCGIFDSTADDR 44 * Alan Cox : Device lock protection. 45 * Alan Cox : Fixed nasty side effect of device close 46 * changes. 47 * Rudi Cilibrasi : Pass the right thing to 48 * set_mac_address() 49 * Dave Miller : 32bit quantity for the device lock to 50 * make it work out on a Sparc. 51 * Bjorn Ekwall : Added KERNELD hack. 52 * Alan Cox : Cleaned up the backlog initialise. 53 * Craig Metz : SIOCGIFCONF fix if space for under 54 * 1 device. 55 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there 56 * is no device open function. 57 * Andi Kleen : Fix error reporting for SIOCGIFCONF 58 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF 59 * Cyrus Durgin : Cleaned for KMOD 60 * Adam Sulmicki : Bug Fix : Network Device Unload 61 * A network device unload needs to purge 62 * the backlog queue. 63 * Paul Rusty Russell : SIOCSIFNAME 64 * Pekka Riikonen : Netdev boot-time settings code 65 * Andrew Morton : Make unregister_netdevice wait 66 * indefinitely on dev->refcnt 67 * J Hadi Salim : - Backlog queue sampling 68 * - netif_rx() feedback 69 */ 70 71 #include <linux/uaccess.h> 72 #include <linux/bitops.h> 73 #include <linux/capability.h> 74 #include <linux/cpu.h> 75 #include <linux/types.h> 76 #include <linux/kernel.h> 77 #include <linux/hash.h> 78 #include <linux/slab.h> 79 #include <linux/sched.h> 80 #include <linux/sched/mm.h> 81 #include <linux/mutex.h> 82 #include <linux/rwsem.h> 83 #include <linux/string.h> 84 #include <linux/mm.h> 85 #include <linux/socket.h> 86 #include <linux/sockios.h> 87 #include <linux/errno.h> 88 #include <linux/interrupt.h> 89 #include <linux/if_ether.h> 90 #include <linux/netdevice.h> 91 #include <linux/etherdevice.h> 92 #include <linux/ethtool.h> 93 #include <linux/skbuff.h> 94 #include <linux/bpf.h> 95 #include <linux/bpf_trace.h> 96 #include <net/net_namespace.h> 97 #include <net/sock.h> 98 #include <net/busy_poll.h> 99 #include <linux/rtnetlink.h> 100 #include <linux/stat.h> 101 #include <net/dsa.h> 102 #include <net/dst.h> 103 #include <net/dst_metadata.h> 104 #include <net/pkt_sched.h> 105 #include <net/pkt_cls.h> 106 #include <net/checksum.h> 107 #include <net/xfrm.h> 108 #include <linux/highmem.h> 109 #include <linux/init.h> 110 #include <linux/module.h> 111 #include <linux/netpoll.h> 112 #include <linux/rcupdate.h> 113 #include <linux/delay.h> 114 #include <net/iw_handler.h> 115 #include <asm/current.h> 116 #include <linux/audit.h> 117 #include <linux/dmaengine.h> 118 #include <linux/err.h> 119 #include <linux/ctype.h> 120 #include <linux/if_arp.h> 121 #include <linux/if_vlan.h> 122 #include <linux/ip.h> 123 #include <net/ip.h> 124 #include <net/mpls.h> 125 #include <linux/ipv6.h> 126 #include <linux/in.h> 127 #include <linux/jhash.h> 128 #include <linux/random.h> 129 #include <trace/events/napi.h> 130 #include <trace/events/net.h> 131 #include <trace/events/skb.h> 132 #include <linux/inetdevice.h> 133 #include <linux/cpu_rmap.h> 134 #include <linux/static_key.h> 135 #include <linux/hashtable.h> 136 #include <linux/vmalloc.h> 137 #include <linux/if_macvlan.h> 138 #include <linux/errqueue.h> 139 #include <linux/hrtimer.h> 140 #include <linux/netfilter_ingress.h> 141 #include <linux/crash_dump.h> 142 #include <linux/sctp.h> 143 #include <net/udp_tunnel.h> 144 #include <linux/net_namespace.h> 145 #include <linux/indirect_call_wrapper.h> 146 #include <net/devlink.h> 147 #include <linux/pm_runtime.h> 148 #include <linux/prandom.h> 149 150 #include "net-sysfs.h" 151 152 #define MAX_GRO_SKBS 8 153 154 /* This should be increased if a protocol with a bigger head is added. */ 155 #define GRO_MAX_HEAD (MAX_HEADER + 128) 156 157 static DEFINE_SPINLOCK(ptype_lock); 158 static DEFINE_SPINLOCK(offload_lock); 159 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; 160 struct list_head ptype_all __read_mostly; /* Taps */ 161 static struct list_head offload_base __read_mostly; 162 163 static int netif_rx_internal(struct sk_buff *skb); 164 static int call_netdevice_notifiers_info(unsigned long val, 165 struct netdev_notifier_info *info); 166 static int call_netdevice_notifiers_extack(unsigned long val, 167 struct net_device *dev, 168 struct netlink_ext_ack *extack); 169 static struct napi_struct *napi_by_id(unsigned int napi_id); 170 171 /* 172 * The @dev_base_head list is protected by @dev_base_lock and the rtnl 173 * semaphore. 174 * 175 * Pure readers hold dev_base_lock for reading, or rcu_read_lock() 176 * 177 * Writers must hold the rtnl semaphore while they loop through the 178 * dev_base_head list, and hold dev_base_lock for writing when they do the 179 * actual updates. This allows pure readers to access the list even 180 * while a writer is preparing to update it. 181 * 182 * To put it another way, dev_base_lock is held for writing only to 183 * protect against pure readers; the rtnl semaphore provides the 184 * protection against other writers. 185 * 186 * See, for example usages, register_netdevice() and 187 * unregister_netdevice(), which must be called with the rtnl 188 * semaphore held. 189 */ 190 DEFINE_RWLOCK(dev_base_lock); 191 EXPORT_SYMBOL(dev_base_lock); 192 193 static DEFINE_MUTEX(ifalias_mutex); 194 195 /* protects napi_hash addition/deletion and napi_gen_id */ 196 static DEFINE_SPINLOCK(napi_hash_lock); 197 198 static unsigned int napi_gen_id = NR_CPUS; 199 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8); 200 201 static DECLARE_RWSEM(devnet_rename_sem); 202 203 static inline void dev_base_seq_inc(struct net *net) 204 { 205 while (++net->dev_base_seq == 0) 206 ; 207 } 208 209 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) 210 { 211 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ)); 212 213 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; 214 } 215 216 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) 217 { 218 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; 219 } 220 221 static inline void rps_lock(struct softnet_data *sd) 222 { 223 #ifdef CONFIG_RPS 224 spin_lock(&sd->input_pkt_queue.lock); 225 #endif 226 } 227 228 static inline void rps_unlock(struct softnet_data *sd) 229 { 230 #ifdef CONFIG_RPS 231 spin_unlock(&sd->input_pkt_queue.lock); 232 #endif 233 } 234 235 static struct netdev_name_node *netdev_name_node_alloc(struct net_device *dev, 236 const char *name) 237 { 238 struct netdev_name_node *name_node; 239 240 name_node = kmalloc(sizeof(*name_node), GFP_KERNEL); 241 if (!name_node) 242 return NULL; 243 INIT_HLIST_NODE(&name_node->hlist); 244 name_node->dev = dev; 245 name_node->name = name; 246 return name_node; 247 } 248 249 static struct netdev_name_node * 250 netdev_name_node_head_alloc(struct net_device *dev) 251 { 252 struct netdev_name_node *name_node; 253 254 name_node = netdev_name_node_alloc(dev, dev->name); 255 if (!name_node) 256 return NULL; 257 INIT_LIST_HEAD(&name_node->list); 258 return name_node; 259 } 260 261 static void netdev_name_node_free(struct netdev_name_node *name_node) 262 { 263 kfree(name_node); 264 } 265 266 static void netdev_name_node_add(struct net *net, 267 struct netdev_name_node *name_node) 268 { 269 hlist_add_head_rcu(&name_node->hlist, 270 dev_name_hash(net, name_node->name)); 271 } 272 273 static void netdev_name_node_del(struct netdev_name_node *name_node) 274 { 275 hlist_del_rcu(&name_node->hlist); 276 } 277 278 static struct netdev_name_node *netdev_name_node_lookup(struct net *net, 279 const char *name) 280 { 281 struct hlist_head *head = dev_name_hash(net, name); 282 struct netdev_name_node *name_node; 283 284 hlist_for_each_entry(name_node, head, hlist) 285 if (!strcmp(name_node->name, name)) 286 return name_node; 287 return NULL; 288 } 289 290 static struct netdev_name_node *netdev_name_node_lookup_rcu(struct net *net, 291 const char *name) 292 { 293 struct hlist_head *head = dev_name_hash(net, name); 294 struct netdev_name_node *name_node; 295 296 hlist_for_each_entry_rcu(name_node, head, hlist) 297 if (!strcmp(name_node->name, name)) 298 return name_node; 299 return NULL; 300 } 301 302 int netdev_name_node_alt_create(struct net_device *dev, const char *name) 303 { 304 struct netdev_name_node *name_node; 305 struct net *net = dev_net(dev); 306 307 name_node = netdev_name_node_lookup(net, name); 308 if (name_node) 309 return -EEXIST; 310 name_node = netdev_name_node_alloc(dev, name); 311 if (!name_node) 312 return -ENOMEM; 313 netdev_name_node_add(net, name_node); 314 /* The node that holds dev->name acts as a head of per-device list. */ 315 list_add_tail(&name_node->list, &dev->name_node->list); 316 317 return 0; 318 } 319 EXPORT_SYMBOL(netdev_name_node_alt_create); 320 321 static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node) 322 { 323 list_del(&name_node->list); 324 netdev_name_node_del(name_node); 325 kfree(name_node->name); 326 netdev_name_node_free(name_node); 327 } 328 329 int netdev_name_node_alt_destroy(struct net_device *dev, const char *name) 330 { 331 struct netdev_name_node *name_node; 332 struct net *net = dev_net(dev); 333 334 name_node = netdev_name_node_lookup(net, name); 335 if (!name_node) 336 return -ENOENT; 337 /* lookup might have found our primary name or a name belonging 338 * to another device. 339 */ 340 if (name_node == dev->name_node || name_node->dev != dev) 341 return -EINVAL; 342 343 __netdev_name_node_alt_destroy(name_node); 344 345 return 0; 346 } 347 EXPORT_SYMBOL(netdev_name_node_alt_destroy); 348 349 static void netdev_name_node_alt_flush(struct net_device *dev) 350 { 351 struct netdev_name_node *name_node, *tmp; 352 353 list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list) 354 __netdev_name_node_alt_destroy(name_node); 355 } 356 357 /* Device list insertion */ 358 static void list_netdevice(struct net_device *dev) 359 { 360 struct net *net = dev_net(dev); 361 362 ASSERT_RTNL(); 363 364 write_lock_bh(&dev_base_lock); 365 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); 366 netdev_name_node_add(net, dev->name_node); 367 hlist_add_head_rcu(&dev->index_hlist, 368 dev_index_hash(net, dev->ifindex)); 369 write_unlock_bh(&dev_base_lock); 370 371 dev_base_seq_inc(net); 372 } 373 374 /* Device list removal 375 * caller must respect a RCU grace period before freeing/reusing dev 376 */ 377 static void unlist_netdevice(struct net_device *dev) 378 { 379 ASSERT_RTNL(); 380 381 /* Unlink dev from the device chain */ 382 write_lock_bh(&dev_base_lock); 383 list_del_rcu(&dev->dev_list); 384 netdev_name_node_del(dev->name_node); 385 hlist_del_rcu(&dev->index_hlist); 386 write_unlock_bh(&dev_base_lock); 387 388 dev_base_seq_inc(dev_net(dev)); 389 } 390 391 /* 392 * Our notifier list 393 */ 394 395 static RAW_NOTIFIER_HEAD(netdev_chain); 396 397 /* 398 * Device drivers call our routines to queue packets here. We empty the 399 * queue in the local softnet handler. 400 */ 401 402 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); 403 EXPORT_PER_CPU_SYMBOL(softnet_data); 404 405 #ifdef CONFIG_LOCKDEP 406 /* 407 * register_netdevice() inits txq->_xmit_lock and sets lockdep class 408 * according to dev->type 409 */ 410 static const unsigned short netdev_lock_type[] = { 411 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, 412 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, 413 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, 414 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, 415 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, 416 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, 417 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, 418 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, 419 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, 420 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, 421 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, 422 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, 423 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, 424 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, 425 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; 426 427 static const char *const netdev_lock_name[] = { 428 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", 429 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", 430 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", 431 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", 432 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", 433 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", 434 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", 435 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", 436 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", 437 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", 438 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", 439 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", 440 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", 441 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", 442 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; 443 444 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; 445 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; 446 447 static inline unsigned short netdev_lock_pos(unsigned short dev_type) 448 { 449 int i; 450 451 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) 452 if (netdev_lock_type[i] == dev_type) 453 return i; 454 /* the last key is used by default */ 455 return ARRAY_SIZE(netdev_lock_type) - 1; 456 } 457 458 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 459 unsigned short dev_type) 460 { 461 int i; 462 463 i = netdev_lock_pos(dev_type); 464 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], 465 netdev_lock_name[i]); 466 } 467 468 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 469 { 470 int i; 471 472 i = netdev_lock_pos(dev->type); 473 lockdep_set_class_and_name(&dev->addr_list_lock, 474 &netdev_addr_lock_key[i], 475 netdev_lock_name[i]); 476 } 477 #else 478 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 479 unsigned short dev_type) 480 { 481 } 482 483 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 484 { 485 } 486 #endif 487 488 /******************************************************************************* 489 * 490 * Protocol management and registration routines 491 * 492 *******************************************************************************/ 493 494 495 /* 496 * Add a protocol ID to the list. Now that the input handler is 497 * smarter we can dispense with all the messy stuff that used to be 498 * here. 499 * 500 * BEWARE!!! Protocol handlers, mangling input packets, 501 * MUST BE last in hash buckets and checking protocol handlers 502 * MUST start from promiscuous ptype_all chain in net_bh. 503 * It is true now, do not change it. 504 * Explanation follows: if protocol handler, mangling packet, will 505 * be the first on list, it is not able to sense, that packet 506 * is cloned and should be copied-on-write, so that it will 507 * change it and subsequent readers will get broken packet. 508 * --ANK (980803) 509 */ 510 511 static inline struct list_head *ptype_head(const struct packet_type *pt) 512 { 513 if (pt->type == htons(ETH_P_ALL)) 514 return pt->dev ? &pt->dev->ptype_all : &ptype_all; 515 else 516 return pt->dev ? &pt->dev->ptype_specific : 517 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; 518 } 519 520 /** 521 * dev_add_pack - add packet handler 522 * @pt: packet type declaration 523 * 524 * Add a protocol handler to the networking stack. The passed &packet_type 525 * is linked into kernel lists and may not be freed until it has been 526 * removed from the kernel lists. 527 * 528 * This call does not sleep therefore it can not 529 * guarantee all CPU's that are in middle of receiving packets 530 * will see the new packet type (until the next received packet). 531 */ 532 533 void dev_add_pack(struct packet_type *pt) 534 { 535 struct list_head *head = ptype_head(pt); 536 537 spin_lock(&ptype_lock); 538 list_add_rcu(&pt->list, head); 539 spin_unlock(&ptype_lock); 540 } 541 EXPORT_SYMBOL(dev_add_pack); 542 543 /** 544 * __dev_remove_pack - remove packet handler 545 * @pt: packet type declaration 546 * 547 * Remove a protocol handler that was previously added to the kernel 548 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 549 * from the kernel lists and can be freed or reused once this function 550 * returns. 551 * 552 * The packet type might still be in use by receivers 553 * and must not be freed until after all the CPU's have gone 554 * through a quiescent state. 555 */ 556 void __dev_remove_pack(struct packet_type *pt) 557 { 558 struct list_head *head = ptype_head(pt); 559 struct packet_type *pt1; 560 561 spin_lock(&ptype_lock); 562 563 list_for_each_entry(pt1, head, list) { 564 if (pt == pt1) { 565 list_del_rcu(&pt->list); 566 goto out; 567 } 568 } 569 570 pr_warn("dev_remove_pack: %p not found\n", pt); 571 out: 572 spin_unlock(&ptype_lock); 573 } 574 EXPORT_SYMBOL(__dev_remove_pack); 575 576 /** 577 * dev_remove_pack - remove packet handler 578 * @pt: packet type declaration 579 * 580 * Remove a protocol handler that was previously added to the kernel 581 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 582 * from the kernel lists and can be freed or reused once this function 583 * returns. 584 * 585 * This call sleeps to guarantee that no CPU is looking at the packet 586 * type after return. 587 */ 588 void dev_remove_pack(struct packet_type *pt) 589 { 590 __dev_remove_pack(pt); 591 592 synchronize_net(); 593 } 594 EXPORT_SYMBOL(dev_remove_pack); 595 596 597 /** 598 * dev_add_offload - register offload handlers 599 * @po: protocol offload declaration 600 * 601 * Add protocol offload handlers to the networking stack. The passed 602 * &proto_offload is linked into kernel lists and may not be freed until 603 * it has been removed from the kernel lists. 604 * 605 * This call does not sleep therefore it can not 606 * guarantee all CPU's that are in middle of receiving packets 607 * will see the new offload handlers (until the next received packet). 608 */ 609 void dev_add_offload(struct packet_offload *po) 610 { 611 struct packet_offload *elem; 612 613 spin_lock(&offload_lock); 614 list_for_each_entry(elem, &offload_base, list) { 615 if (po->priority < elem->priority) 616 break; 617 } 618 list_add_rcu(&po->list, elem->list.prev); 619 spin_unlock(&offload_lock); 620 } 621 EXPORT_SYMBOL(dev_add_offload); 622 623 /** 624 * __dev_remove_offload - remove offload handler 625 * @po: packet offload declaration 626 * 627 * Remove a protocol offload handler that was previously added to the 628 * kernel offload handlers by dev_add_offload(). The passed &offload_type 629 * is removed from the kernel lists and can be freed or reused once this 630 * function returns. 631 * 632 * The packet type might still be in use by receivers 633 * and must not be freed until after all the CPU's have gone 634 * through a quiescent state. 635 */ 636 static void __dev_remove_offload(struct packet_offload *po) 637 { 638 struct list_head *head = &offload_base; 639 struct packet_offload *po1; 640 641 spin_lock(&offload_lock); 642 643 list_for_each_entry(po1, head, list) { 644 if (po == po1) { 645 list_del_rcu(&po->list); 646 goto out; 647 } 648 } 649 650 pr_warn("dev_remove_offload: %p not found\n", po); 651 out: 652 spin_unlock(&offload_lock); 653 } 654 655 /** 656 * dev_remove_offload - remove packet offload handler 657 * @po: packet offload declaration 658 * 659 * Remove a packet offload handler that was previously added to the kernel 660 * offload handlers by dev_add_offload(). The passed &offload_type is 661 * removed from the kernel lists and can be freed or reused once this 662 * function returns. 663 * 664 * This call sleeps to guarantee that no CPU is looking at the packet 665 * type after return. 666 */ 667 void dev_remove_offload(struct packet_offload *po) 668 { 669 __dev_remove_offload(po); 670 671 synchronize_net(); 672 } 673 EXPORT_SYMBOL(dev_remove_offload); 674 675 /****************************************************************************** 676 * 677 * Device Boot-time Settings Routines 678 * 679 ******************************************************************************/ 680 681 /* Boot time configuration table */ 682 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX]; 683 684 /** 685 * netdev_boot_setup_add - add new setup entry 686 * @name: name of the device 687 * @map: configured settings for the device 688 * 689 * Adds new setup entry to the dev_boot_setup list. The function 690 * returns 0 on error and 1 on success. This is a generic routine to 691 * all netdevices. 692 */ 693 static int netdev_boot_setup_add(char *name, struct ifmap *map) 694 { 695 struct netdev_boot_setup *s; 696 int i; 697 698 s = dev_boot_setup; 699 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { 700 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') { 701 memset(s[i].name, 0, sizeof(s[i].name)); 702 strlcpy(s[i].name, name, IFNAMSIZ); 703 memcpy(&s[i].map, map, sizeof(s[i].map)); 704 break; 705 } 706 } 707 708 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1; 709 } 710 711 /** 712 * netdev_boot_setup_check - check boot time settings 713 * @dev: the netdevice 714 * 715 * Check boot time settings for the device. 716 * The found settings are set for the device to be used 717 * later in the device probing. 718 * Returns 0 if no settings found, 1 if they are. 719 */ 720 int netdev_boot_setup_check(struct net_device *dev) 721 { 722 struct netdev_boot_setup *s = dev_boot_setup; 723 int i; 724 725 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { 726 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' && 727 !strcmp(dev->name, s[i].name)) { 728 dev->irq = s[i].map.irq; 729 dev->base_addr = s[i].map.base_addr; 730 dev->mem_start = s[i].map.mem_start; 731 dev->mem_end = s[i].map.mem_end; 732 return 1; 733 } 734 } 735 return 0; 736 } 737 EXPORT_SYMBOL(netdev_boot_setup_check); 738 739 740 /** 741 * netdev_boot_base - get address from boot time settings 742 * @prefix: prefix for network device 743 * @unit: id for network device 744 * 745 * Check boot time settings for the base address of device. 746 * The found settings are set for the device to be used 747 * later in the device probing. 748 * Returns 0 if no settings found. 749 */ 750 unsigned long netdev_boot_base(const char *prefix, int unit) 751 { 752 const struct netdev_boot_setup *s = dev_boot_setup; 753 char name[IFNAMSIZ]; 754 int i; 755 756 sprintf(name, "%s%d", prefix, unit); 757 758 /* 759 * If device already registered then return base of 1 760 * to indicate not to probe for this interface 761 */ 762 if (__dev_get_by_name(&init_net, name)) 763 return 1; 764 765 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) 766 if (!strcmp(name, s[i].name)) 767 return s[i].map.base_addr; 768 return 0; 769 } 770 771 /* 772 * Saves at boot time configured settings for any netdevice. 773 */ 774 int __init netdev_boot_setup(char *str) 775 { 776 int ints[5]; 777 struct ifmap map; 778 779 str = get_options(str, ARRAY_SIZE(ints), ints); 780 if (!str || !*str) 781 return 0; 782 783 /* Save settings */ 784 memset(&map, 0, sizeof(map)); 785 if (ints[0] > 0) 786 map.irq = ints[1]; 787 if (ints[0] > 1) 788 map.base_addr = ints[2]; 789 if (ints[0] > 2) 790 map.mem_start = ints[3]; 791 if (ints[0] > 3) 792 map.mem_end = ints[4]; 793 794 /* Add new entry to the list */ 795 return netdev_boot_setup_add(str, &map); 796 } 797 798 __setup("netdev=", netdev_boot_setup); 799 800 /******************************************************************************* 801 * 802 * Device Interface Subroutines 803 * 804 *******************************************************************************/ 805 806 /** 807 * dev_get_iflink - get 'iflink' value of a interface 808 * @dev: targeted interface 809 * 810 * Indicates the ifindex the interface is linked to. 811 * Physical interfaces have the same 'ifindex' and 'iflink' values. 812 */ 813 814 int dev_get_iflink(const struct net_device *dev) 815 { 816 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink) 817 return dev->netdev_ops->ndo_get_iflink(dev); 818 819 return dev->ifindex; 820 } 821 EXPORT_SYMBOL(dev_get_iflink); 822 823 /** 824 * dev_fill_metadata_dst - Retrieve tunnel egress information. 825 * @dev: targeted interface 826 * @skb: The packet. 827 * 828 * For better visibility of tunnel traffic OVS needs to retrieve 829 * egress tunnel information for a packet. Following API allows 830 * user to get this info. 831 */ 832 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) 833 { 834 struct ip_tunnel_info *info; 835 836 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst) 837 return -EINVAL; 838 839 info = skb_tunnel_info_unclone(skb); 840 if (!info) 841 return -ENOMEM; 842 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX))) 843 return -EINVAL; 844 845 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb); 846 } 847 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst); 848 849 /** 850 * __dev_get_by_name - find a device by its name 851 * @net: the applicable net namespace 852 * @name: name to find 853 * 854 * Find an interface by name. Must be called under RTNL semaphore 855 * or @dev_base_lock. If the name is found a pointer to the device 856 * is returned. If the name is not found then %NULL is returned. The 857 * reference counters are not incremented so the caller must be 858 * careful with locks. 859 */ 860 861 struct net_device *__dev_get_by_name(struct net *net, const char *name) 862 { 863 struct netdev_name_node *node_name; 864 865 node_name = netdev_name_node_lookup(net, name); 866 return node_name ? node_name->dev : NULL; 867 } 868 EXPORT_SYMBOL(__dev_get_by_name); 869 870 /** 871 * dev_get_by_name_rcu - find a device by its name 872 * @net: the applicable net namespace 873 * @name: name to find 874 * 875 * Find an interface by name. 876 * If the name is found a pointer to the device is returned. 877 * If the name is not found then %NULL is returned. 878 * The reference counters are not incremented so the caller must be 879 * careful with locks. The caller must hold RCU lock. 880 */ 881 882 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) 883 { 884 struct netdev_name_node *node_name; 885 886 node_name = netdev_name_node_lookup_rcu(net, name); 887 return node_name ? node_name->dev : NULL; 888 } 889 EXPORT_SYMBOL(dev_get_by_name_rcu); 890 891 /** 892 * dev_get_by_name - find a device by its name 893 * @net: the applicable net namespace 894 * @name: name to find 895 * 896 * Find an interface by name. This can be called from any 897 * context and does its own locking. The returned handle has 898 * the usage count incremented and the caller must use dev_put() to 899 * release it when it is no longer needed. %NULL is returned if no 900 * matching device is found. 901 */ 902 903 struct net_device *dev_get_by_name(struct net *net, const char *name) 904 { 905 struct net_device *dev; 906 907 rcu_read_lock(); 908 dev = dev_get_by_name_rcu(net, name); 909 if (dev) 910 dev_hold(dev); 911 rcu_read_unlock(); 912 return dev; 913 } 914 EXPORT_SYMBOL(dev_get_by_name); 915 916 /** 917 * __dev_get_by_index - find a device by its ifindex 918 * @net: the applicable net namespace 919 * @ifindex: index of device 920 * 921 * Search for an interface by index. Returns %NULL if the device 922 * is not found or a pointer to the device. The device has not 923 * had its reference counter increased so the caller must be careful 924 * about locking. The caller must hold either the RTNL semaphore 925 * or @dev_base_lock. 926 */ 927 928 struct net_device *__dev_get_by_index(struct net *net, int ifindex) 929 { 930 struct net_device *dev; 931 struct hlist_head *head = dev_index_hash(net, ifindex); 932 933 hlist_for_each_entry(dev, head, index_hlist) 934 if (dev->ifindex == ifindex) 935 return dev; 936 937 return NULL; 938 } 939 EXPORT_SYMBOL(__dev_get_by_index); 940 941 /** 942 * dev_get_by_index_rcu - find a device by its ifindex 943 * @net: the applicable net namespace 944 * @ifindex: index of device 945 * 946 * Search for an interface by index. Returns %NULL if the device 947 * is not found or a pointer to the device. The device has not 948 * had its reference counter increased so the caller must be careful 949 * about locking. The caller must hold RCU lock. 950 */ 951 952 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) 953 { 954 struct net_device *dev; 955 struct hlist_head *head = dev_index_hash(net, ifindex); 956 957 hlist_for_each_entry_rcu(dev, head, index_hlist) 958 if (dev->ifindex == ifindex) 959 return dev; 960 961 return NULL; 962 } 963 EXPORT_SYMBOL(dev_get_by_index_rcu); 964 965 966 /** 967 * dev_get_by_index - find a device by its ifindex 968 * @net: the applicable net namespace 969 * @ifindex: index of device 970 * 971 * Search for an interface by index. Returns NULL if the device 972 * is not found or a pointer to the device. The device returned has 973 * had a reference added and the pointer is safe until the user calls 974 * dev_put to indicate they have finished with it. 975 */ 976 977 struct net_device *dev_get_by_index(struct net *net, int ifindex) 978 { 979 struct net_device *dev; 980 981 rcu_read_lock(); 982 dev = dev_get_by_index_rcu(net, ifindex); 983 if (dev) 984 dev_hold(dev); 985 rcu_read_unlock(); 986 return dev; 987 } 988 EXPORT_SYMBOL(dev_get_by_index); 989 990 /** 991 * dev_get_by_napi_id - find a device by napi_id 992 * @napi_id: ID of the NAPI struct 993 * 994 * Search for an interface by NAPI ID. Returns %NULL if the device 995 * is not found or a pointer to the device. The device has not had 996 * its reference counter increased so the caller must be careful 997 * about locking. The caller must hold RCU lock. 998 */ 999 1000 struct net_device *dev_get_by_napi_id(unsigned int napi_id) 1001 { 1002 struct napi_struct *napi; 1003 1004 WARN_ON_ONCE(!rcu_read_lock_held()); 1005 1006 if (napi_id < MIN_NAPI_ID) 1007 return NULL; 1008 1009 napi = napi_by_id(napi_id); 1010 1011 return napi ? napi->dev : NULL; 1012 } 1013 EXPORT_SYMBOL(dev_get_by_napi_id); 1014 1015 /** 1016 * netdev_get_name - get a netdevice name, knowing its ifindex. 1017 * @net: network namespace 1018 * @name: a pointer to the buffer where the name will be stored. 1019 * @ifindex: the ifindex of the interface to get the name from. 1020 */ 1021 int netdev_get_name(struct net *net, char *name, int ifindex) 1022 { 1023 struct net_device *dev; 1024 int ret; 1025 1026 down_read(&devnet_rename_sem); 1027 rcu_read_lock(); 1028 1029 dev = dev_get_by_index_rcu(net, ifindex); 1030 if (!dev) { 1031 ret = -ENODEV; 1032 goto out; 1033 } 1034 1035 strcpy(name, dev->name); 1036 1037 ret = 0; 1038 out: 1039 rcu_read_unlock(); 1040 up_read(&devnet_rename_sem); 1041 return ret; 1042 } 1043 1044 /** 1045 * dev_getbyhwaddr_rcu - find a device by its hardware address 1046 * @net: the applicable net namespace 1047 * @type: media type of device 1048 * @ha: hardware address 1049 * 1050 * Search for an interface by MAC address. Returns NULL if the device 1051 * is not found or a pointer to the device. 1052 * The caller must hold RCU or RTNL. 1053 * The returned device has not had its ref count increased 1054 * and the caller must therefore be careful about locking 1055 * 1056 */ 1057 1058 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, 1059 const char *ha) 1060 { 1061 struct net_device *dev; 1062 1063 for_each_netdev_rcu(net, dev) 1064 if (dev->type == type && 1065 !memcmp(dev->dev_addr, ha, dev->addr_len)) 1066 return dev; 1067 1068 return NULL; 1069 } 1070 EXPORT_SYMBOL(dev_getbyhwaddr_rcu); 1071 1072 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) 1073 { 1074 struct net_device *dev, *ret = NULL; 1075 1076 rcu_read_lock(); 1077 for_each_netdev_rcu(net, dev) 1078 if (dev->type == type) { 1079 dev_hold(dev); 1080 ret = dev; 1081 break; 1082 } 1083 rcu_read_unlock(); 1084 return ret; 1085 } 1086 EXPORT_SYMBOL(dev_getfirstbyhwtype); 1087 1088 /** 1089 * __dev_get_by_flags - find any device with given flags 1090 * @net: the applicable net namespace 1091 * @if_flags: IFF_* values 1092 * @mask: bitmask of bits in if_flags to check 1093 * 1094 * Search for any interface with the given flags. Returns NULL if a device 1095 * is not found or a pointer to the device. Must be called inside 1096 * rtnl_lock(), and result refcount is unchanged. 1097 */ 1098 1099 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags, 1100 unsigned short mask) 1101 { 1102 struct net_device *dev, *ret; 1103 1104 ASSERT_RTNL(); 1105 1106 ret = NULL; 1107 for_each_netdev(net, dev) { 1108 if (((dev->flags ^ if_flags) & mask) == 0) { 1109 ret = dev; 1110 break; 1111 } 1112 } 1113 return ret; 1114 } 1115 EXPORT_SYMBOL(__dev_get_by_flags); 1116 1117 /** 1118 * dev_valid_name - check if name is okay for network device 1119 * @name: name string 1120 * 1121 * Network device names need to be valid file names to 1122 * allow sysfs to work. We also disallow any kind of 1123 * whitespace. 1124 */ 1125 bool dev_valid_name(const char *name) 1126 { 1127 if (*name == '\0') 1128 return false; 1129 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ) 1130 return false; 1131 if (!strcmp(name, ".") || !strcmp(name, "..")) 1132 return false; 1133 1134 while (*name) { 1135 if (*name == '/' || *name == ':' || isspace(*name)) 1136 return false; 1137 name++; 1138 } 1139 return true; 1140 } 1141 EXPORT_SYMBOL(dev_valid_name); 1142 1143 /** 1144 * __dev_alloc_name - allocate a name for a device 1145 * @net: network namespace to allocate the device name in 1146 * @name: name format string 1147 * @buf: scratch buffer and result name string 1148 * 1149 * Passed a format string - eg "lt%d" it will try and find a suitable 1150 * id. It scans list of devices to build up a free map, then chooses 1151 * the first empty slot. The caller must hold the dev_base or rtnl lock 1152 * while allocating the name and adding the device in order to avoid 1153 * duplicates. 1154 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1155 * Returns the number of the unit assigned or a negative errno code. 1156 */ 1157 1158 static int __dev_alloc_name(struct net *net, const char *name, char *buf) 1159 { 1160 int i = 0; 1161 const char *p; 1162 const int max_netdevices = 8*PAGE_SIZE; 1163 unsigned long *inuse; 1164 struct net_device *d; 1165 1166 if (!dev_valid_name(name)) 1167 return -EINVAL; 1168 1169 p = strchr(name, '%'); 1170 if (p) { 1171 /* 1172 * Verify the string as this thing may have come from 1173 * the user. There must be either one "%d" and no other "%" 1174 * characters. 1175 */ 1176 if (p[1] != 'd' || strchr(p + 2, '%')) 1177 return -EINVAL; 1178 1179 /* Use one page as a bit array of possible slots */ 1180 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); 1181 if (!inuse) 1182 return -ENOMEM; 1183 1184 for_each_netdev(net, d) { 1185 if (!sscanf(d->name, name, &i)) 1186 continue; 1187 if (i < 0 || i >= max_netdevices) 1188 continue; 1189 1190 /* avoid cases where sscanf is not exact inverse of printf */ 1191 snprintf(buf, IFNAMSIZ, name, i); 1192 if (!strncmp(buf, d->name, IFNAMSIZ)) 1193 set_bit(i, inuse); 1194 } 1195 1196 i = find_first_zero_bit(inuse, max_netdevices); 1197 free_page((unsigned long) inuse); 1198 } 1199 1200 snprintf(buf, IFNAMSIZ, name, i); 1201 if (!__dev_get_by_name(net, buf)) 1202 return i; 1203 1204 /* It is possible to run out of possible slots 1205 * when the name is long and there isn't enough space left 1206 * for the digits, or if all bits are used. 1207 */ 1208 return -ENFILE; 1209 } 1210 1211 static int dev_alloc_name_ns(struct net *net, 1212 struct net_device *dev, 1213 const char *name) 1214 { 1215 char buf[IFNAMSIZ]; 1216 int ret; 1217 1218 BUG_ON(!net); 1219 ret = __dev_alloc_name(net, name, buf); 1220 if (ret >= 0) 1221 strlcpy(dev->name, buf, IFNAMSIZ); 1222 return ret; 1223 } 1224 1225 /** 1226 * dev_alloc_name - allocate a name for a device 1227 * @dev: device 1228 * @name: name format string 1229 * 1230 * Passed a format string - eg "lt%d" it will try and find a suitable 1231 * id. It scans list of devices to build up a free map, then chooses 1232 * the first empty slot. The caller must hold the dev_base or rtnl lock 1233 * while allocating the name and adding the device in order to avoid 1234 * duplicates. 1235 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1236 * Returns the number of the unit assigned or a negative errno code. 1237 */ 1238 1239 int dev_alloc_name(struct net_device *dev, const char *name) 1240 { 1241 return dev_alloc_name_ns(dev_net(dev), dev, name); 1242 } 1243 EXPORT_SYMBOL(dev_alloc_name); 1244 1245 static int dev_get_valid_name(struct net *net, struct net_device *dev, 1246 const char *name) 1247 { 1248 BUG_ON(!net); 1249 1250 if (!dev_valid_name(name)) 1251 return -EINVAL; 1252 1253 if (strchr(name, '%')) 1254 return dev_alloc_name_ns(net, dev, name); 1255 else if (__dev_get_by_name(net, name)) 1256 return -EEXIST; 1257 else if (dev->name != name) 1258 strlcpy(dev->name, name, IFNAMSIZ); 1259 1260 return 0; 1261 } 1262 1263 /** 1264 * dev_change_name - change name of a device 1265 * @dev: device 1266 * @newname: name (or format string) must be at least IFNAMSIZ 1267 * 1268 * Change name of a device, can pass format strings "eth%d". 1269 * for wildcarding. 1270 */ 1271 int dev_change_name(struct net_device *dev, const char *newname) 1272 { 1273 unsigned char old_assign_type; 1274 char oldname[IFNAMSIZ]; 1275 int err = 0; 1276 int ret; 1277 struct net *net; 1278 1279 ASSERT_RTNL(); 1280 BUG_ON(!dev_net(dev)); 1281 1282 net = dev_net(dev); 1283 1284 /* Some auto-enslaved devices e.g. failover slaves are 1285 * special, as userspace might rename the device after 1286 * the interface had been brought up and running since 1287 * the point kernel initiated auto-enslavement. Allow 1288 * live name change even when these slave devices are 1289 * up and running. 1290 * 1291 * Typically, users of these auto-enslaving devices 1292 * don't actually care about slave name change, as 1293 * they are supposed to operate on master interface 1294 * directly. 1295 */ 1296 if (dev->flags & IFF_UP && 1297 likely(!(dev->priv_flags & IFF_LIVE_RENAME_OK))) 1298 return -EBUSY; 1299 1300 down_write(&devnet_rename_sem); 1301 1302 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { 1303 up_write(&devnet_rename_sem); 1304 return 0; 1305 } 1306 1307 memcpy(oldname, dev->name, IFNAMSIZ); 1308 1309 err = dev_get_valid_name(net, dev, newname); 1310 if (err < 0) { 1311 up_write(&devnet_rename_sem); 1312 return err; 1313 } 1314 1315 if (oldname[0] && !strchr(oldname, '%')) 1316 netdev_info(dev, "renamed from %s\n", oldname); 1317 1318 old_assign_type = dev->name_assign_type; 1319 dev->name_assign_type = NET_NAME_RENAMED; 1320 1321 rollback: 1322 ret = device_rename(&dev->dev, dev->name); 1323 if (ret) { 1324 memcpy(dev->name, oldname, IFNAMSIZ); 1325 dev->name_assign_type = old_assign_type; 1326 up_write(&devnet_rename_sem); 1327 return ret; 1328 } 1329 1330 up_write(&devnet_rename_sem); 1331 1332 netdev_adjacent_rename_links(dev, oldname); 1333 1334 write_lock_bh(&dev_base_lock); 1335 netdev_name_node_del(dev->name_node); 1336 write_unlock_bh(&dev_base_lock); 1337 1338 synchronize_rcu(); 1339 1340 write_lock_bh(&dev_base_lock); 1341 netdev_name_node_add(net, dev->name_node); 1342 write_unlock_bh(&dev_base_lock); 1343 1344 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); 1345 ret = notifier_to_errno(ret); 1346 1347 if (ret) { 1348 /* err >= 0 after dev_alloc_name() or stores the first errno */ 1349 if (err >= 0) { 1350 err = ret; 1351 down_write(&devnet_rename_sem); 1352 memcpy(dev->name, oldname, IFNAMSIZ); 1353 memcpy(oldname, newname, IFNAMSIZ); 1354 dev->name_assign_type = old_assign_type; 1355 old_assign_type = NET_NAME_RENAMED; 1356 goto rollback; 1357 } else { 1358 pr_err("%s: name change rollback failed: %d\n", 1359 dev->name, ret); 1360 } 1361 } 1362 1363 return err; 1364 } 1365 1366 /** 1367 * dev_set_alias - change ifalias of a device 1368 * @dev: device 1369 * @alias: name up to IFALIASZ 1370 * @len: limit of bytes to copy from info 1371 * 1372 * Set ifalias for a device, 1373 */ 1374 int dev_set_alias(struct net_device *dev, const char *alias, size_t len) 1375 { 1376 struct dev_ifalias *new_alias = NULL; 1377 1378 if (len >= IFALIASZ) 1379 return -EINVAL; 1380 1381 if (len) { 1382 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL); 1383 if (!new_alias) 1384 return -ENOMEM; 1385 1386 memcpy(new_alias->ifalias, alias, len); 1387 new_alias->ifalias[len] = 0; 1388 } 1389 1390 mutex_lock(&ifalias_mutex); 1391 new_alias = rcu_replace_pointer(dev->ifalias, new_alias, 1392 mutex_is_locked(&ifalias_mutex)); 1393 mutex_unlock(&ifalias_mutex); 1394 1395 if (new_alias) 1396 kfree_rcu(new_alias, rcuhead); 1397 1398 return len; 1399 } 1400 EXPORT_SYMBOL(dev_set_alias); 1401 1402 /** 1403 * dev_get_alias - get ifalias of a device 1404 * @dev: device 1405 * @name: buffer to store name of ifalias 1406 * @len: size of buffer 1407 * 1408 * get ifalias for a device. Caller must make sure dev cannot go 1409 * away, e.g. rcu read lock or own a reference count to device. 1410 */ 1411 int dev_get_alias(const struct net_device *dev, char *name, size_t len) 1412 { 1413 const struct dev_ifalias *alias; 1414 int ret = 0; 1415 1416 rcu_read_lock(); 1417 alias = rcu_dereference(dev->ifalias); 1418 if (alias) 1419 ret = snprintf(name, len, "%s", alias->ifalias); 1420 rcu_read_unlock(); 1421 1422 return ret; 1423 } 1424 1425 /** 1426 * netdev_features_change - device changes features 1427 * @dev: device to cause notification 1428 * 1429 * Called to indicate a device has changed features. 1430 */ 1431 void netdev_features_change(struct net_device *dev) 1432 { 1433 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); 1434 } 1435 EXPORT_SYMBOL(netdev_features_change); 1436 1437 /** 1438 * netdev_state_change - device changes state 1439 * @dev: device to cause notification 1440 * 1441 * Called to indicate a device has changed state. This function calls 1442 * the notifier chains for netdev_chain and sends a NEWLINK message 1443 * to the routing socket. 1444 */ 1445 void netdev_state_change(struct net_device *dev) 1446 { 1447 if (dev->flags & IFF_UP) { 1448 struct netdev_notifier_change_info change_info = { 1449 .info.dev = dev, 1450 }; 1451 1452 call_netdevice_notifiers_info(NETDEV_CHANGE, 1453 &change_info.info); 1454 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL); 1455 } 1456 } 1457 EXPORT_SYMBOL(netdev_state_change); 1458 1459 /** 1460 * __netdev_notify_peers - notify network peers about existence of @dev, 1461 * to be called when rtnl lock is already held. 1462 * @dev: network device 1463 * 1464 * Generate traffic such that interested network peers are aware of 1465 * @dev, such as by generating a gratuitous ARP. This may be used when 1466 * a device wants to inform the rest of the network about some sort of 1467 * reconfiguration such as a failover event or virtual machine 1468 * migration. 1469 */ 1470 void __netdev_notify_peers(struct net_device *dev) 1471 { 1472 ASSERT_RTNL(); 1473 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); 1474 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev); 1475 } 1476 EXPORT_SYMBOL(__netdev_notify_peers); 1477 1478 /** 1479 * netdev_notify_peers - notify network peers about existence of @dev 1480 * @dev: network device 1481 * 1482 * Generate traffic such that interested network peers are aware of 1483 * @dev, such as by generating a gratuitous ARP. This may be used when 1484 * a device wants to inform the rest of the network about some sort of 1485 * reconfiguration such as a failover event or virtual machine 1486 * migration. 1487 */ 1488 void netdev_notify_peers(struct net_device *dev) 1489 { 1490 rtnl_lock(); 1491 __netdev_notify_peers(dev); 1492 rtnl_unlock(); 1493 } 1494 EXPORT_SYMBOL(netdev_notify_peers); 1495 1496 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1497 { 1498 const struct net_device_ops *ops = dev->netdev_ops; 1499 int ret; 1500 1501 ASSERT_RTNL(); 1502 1503 if (!netif_device_present(dev)) { 1504 /* may be detached because parent is runtime-suspended */ 1505 if (dev->dev.parent) 1506 pm_runtime_resume(dev->dev.parent); 1507 if (!netif_device_present(dev)) 1508 return -ENODEV; 1509 } 1510 1511 /* Block netpoll from trying to do any rx path servicing. 1512 * If we don't do this there is a chance ndo_poll_controller 1513 * or ndo_poll may be running while we open the device 1514 */ 1515 netpoll_poll_disable(dev); 1516 1517 ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack); 1518 ret = notifier_to_errno(ret); 1519 if (ret) 1520 return ret; 1521 1522 set_bit(__LINK_STATE_START, &dev->state); 1523 1524 if (ops->ndo_validate_addr) 1525 ret = ops->ndo_validate_addr(dev); 1526 1527 if (!ret && ops->ndo_open) 1528 ret = ops->ndo_open(dev); 1529 1530 netpoll_poll_enable(dev); 1531 1532 if (ret) 1533 clear_bit(__LINK_STATE_START, &dev->state); 1534 else { 1535 dev->flags |= IFF_UP; 1536 dev_set_rx_mode(dev); 1537 dev_activate(dev); 1538 add_device_randomness(dev->dev_addr, dev->addr_len); 1539 } 1540 1541 return ret; 1542 } 1543 1544 /** 1545 * dev_open - prepare an interface for use. 1546 * @dev: device to open 1547 * @extack: netlink extended ack 1548 * 1549 * Takes a device from down to up state. The device's private open 1550 * function is invoked and then the multicast lists are loaded. Finally 1551 * the device is moved into the up state and a %NETDEV_UP message is 1552 * sent to the netdev notifier chain. 1553 * 1554 * Calling this function on an active interface is a nop. On a failure 1555 * a negative errno code is returned. 1556 */ 1557 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1558 { 1559 int ret; 1560 1561 if (dev->flags & IFF_UP) 1562 return 0; 1563 1564 ret = __dev_open(dev, extack); 1565 if (ret < 0) 1566 return ret; 1567 1568 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); 1569 call_netdevice_notifiers(NETDEV_UP, dev); 1570 1571 return ret; 1572 } 1573 EXPORT_SYMBOL(dev_open); 1574 1575 static void __dev_close_many(struct list_head *head) 1576 { 1577 struct net_device *dev; 1578 1579 ASSERT_RTNL(); 1580 might_sleep(); 1581 1582 list_for_each_entry(dev, head, close_list) { 1583 /* Temporarily disable netpoll until the interface is down */ 1584 netpoll_poll_disable(dev); 1585 1586 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); 1587 1588 clear_bit(__LINK_STATE_START, &dev->state); 1589 1590 /* Synchronize to scheduled poll. We cannot touch poll list, it 1591 * can be even on different cpu. So just clear netif_running(). 1592 * 1593 * dev->stop() will invoke napi_disable() on all of it's 1594 * napi_struct instances on this device. 1595 */ 1596 smp_mb__after_atomic(); /* Commit netif_running(). */ 1597 } 1598 1599 dev_deactivate_many(head); 1600 1601 list_for_each_entry(dev, head, close_list) { 1602 const struct net_device_ops *ops = dev->netdev_ops; 1603 1604 /* 1605 * Call the device specific close. This cannot fail. 1606 * Only if device is UP 1607 * 1608 * We allow it to be called even after a DETACH hot-plug 1609 * event. 1610 */ 1611 if (ops->ndo_stop) 1612 ops->ndo_stop(dev); 1613 1614 dev->flags &= ~IFF_UP; 1615 netpoll_poll_enable(dev); 1616 } 1617 } 1618 1619 static void __dev_close(struct net_device *dev) 1620 { 1621 LIST_HEAD(single); 1622 1623 list_add(&dev->close_list, &single); 1624 __dev_close_many(&single); 1625 list_del(&single); 1626 } 1627 1628 void dev_close_many(struct list_head *head, bool unlink) 1629 { 1630 struct net_device *dev, *tmp; 1631 1632 /* Remove the devices that don't need to be closed */ 1633 list_for_each_entry_safe(dev, tmp, head, close_list) 1634 if (!(dev->flags & IFF_UP)) 1635 list_del_init(&dev->close_list); 1636 1637 __dev_close_many(head); 1638 1639 list_for_each_entry_safe(dev, tmp, head, close_list) { 1640 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); 1641 call_netdevice_notifiers(NETDEV_DOWN, dev); 1642 if (unlink) 1643 list_del_init(&dev->close_list); 1644 } 1645 } 1646 EXPORT_SYMBOL(dev_close_many); 1647 1648 /** 1649 * dev_close - shutdown an interface. 1650 * @dev: device to shutdown 1651 * 1652 * This function moves an active device into down state. A 1653 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device 1654 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier 1655 * chain. 1656 */ 1657 void dev_close(struct net_device *dev) 1658 { 1659 if (dev->flags & IFF_UP) { 1660 LIST_HEAD(single); 1661 1662 list_add(&dev->close_list, &single); 1663 dev_close_many(&single, true); 1664 list_del(&single); 1665 } 1666 } 1667 EXPORT_SYMBOL(dev_close); 1668 1669 1670 /** 1671 * dev_disable_lro - disable Large Receive Offload on a device 1672 * @dev: device 1673 * 1674 * Disable Large Receive Offload (LRO) on a net device. Must be 1675 * called under RTNL. This is needed if received packets may be 1676 * forwarded to another interface. 1677 */ 1678 void dev_disable_lro(struct net_device *dev) 1679 { 1680 struct net_device *lower_dev; 1681 struct list_head *iter; 1682 1683 dev->wanted_features &= ~NETIF_F_LRO; 1684 netdev_update_features(dev); 1685 1686 if (unlikely(dev->features & NETIF_F_LRO)) 1687 netdev_WARN(dev, "failed to disable LRO!\n"); 1688 1689 netdev_for_each_lower_dev(dev, lower_dev, iter) 1690 dev_disable_lro(lower_dev); 1691 } 1692 EXPORT_SYMBOL(dev_disable_lro); 1693 1694 /** 1695 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device 1696 * @dev: device 1697 * 1698 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be 1699 * called under RTNL. This is needed if Generic XDP is installed on 1700 * the device. 1701 */ 1702 static void dev_disable_gro_hw(struct net_device *dev) 1703 { 1704 dev->wanted_features &= ~NETIF_F_GRO_HW; 1705 netdev_update_features(dev); 1706 1707 if (unlikely(dev->features & NETIF_F_GRO_HW)) 1708 netdev_WARN(dev, "failed to disable GRO_HW!\n"); 1709 } 1710 1711 const char *netdev_cmd_to_name(enum netdev_cmd cmd) 1712 { 1713 #define N(val) \ 1714 case NETDEV_##val: \ 1715 return "NETDEV_" __stringify(val); 1716 switch (cmd) { 1717 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER) 1718 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE) 1719 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE) 1720 N(POST_INIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) N(CHANGEUPPER) 1721 N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) N(BONDING_INFO) 1722 N(PRECHANGEUPPER) N(CHANGELOWERSTATE) N(UDP_TUNNEL_PUSH_INFO) 1723 N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN) 1724 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO) 1725 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO) 1726 N(PRE_CHANGEADDR) 1727 } 1728 #undef N 1729 return "UNKNOWN_NETDEV_EVENT"; 1730 } 1731 EXPORT_SYMBOL_GPL(netdev_cmd_to_name); 1732 1733 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, 1734 struct net_device *dev) 1735 { 1736 struct netdev_notifier_info info = { 1737 .dev = dev, 1738 }; 1739 1740 return nb->notifier_call(nb, val, &info); 1741 } 1742 1743 static int call_netdevice_register_notifiers(struct notifier_block *nb, 1744 struct net_device *dev) 1745 { 1746 int err; 1747 1748 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); 1749 err = notifier_to_errno(err); 1750 if (err) 1751 return err; 1752 1753 if (!(dev->flags & IFF_UP)) 1754 return 0; 1755 1756 call_netdevice_notifier(nb, NETDEV_UP, dev); 1757 return 0; 1758 } 1759 1760 static void call_netdevice_unregister_notifiers(struct notifier_block *nb, 1761 struct net_device *dev) 1762 { 1763 if (dev->flags & IFF_UP) { 1764 call_netdevice_notifier(nb, NETDEV_GOING_DOWN, 1765 dev); 1766 call_netdevice_notifier(nb, NETDEV_DOWN, dev); 1767 } 1768 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); 1769 } 1770 1771 static int call_netdevice_register_net_notifiers(struct notifier_block *nb, 1772 struct net *net) 1773 { 1774 struct net_device *dev; 1775 int err; 1776 1777 for_each_netdev(net, dev) { 1778 err = call_netdevice_register_notifiers(nb, dev); 1779 if (err) 1780 goto rollback; 1781 } 1782 return 0; 1783 1784 rollback: 1785 for_each_netdev_continue_reverse(net, dev) 1786 call_netdevice_unregister_notifiers(nb, dev); 1787 return err; 1788 } 1789 1790 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb, 1791 struct net *net) 1792 { 1793 struct net_device *dev; 1794 1795 for_each_netdev(net, dev) 1796 call_netdevice_unregister_notifiers(nb, dev); 1797 } 1798 1799 static int dev_boot_phase = 1; 1800 1801 /** 1802 * register_netdevice_notifier - register a network notifier block 1803 * @nb: notifier 1804 * 1805 * Register a notifier to be called when network device events occur. 1806 * The notifier passed is linked into the kernel structures and must 1807 * not be reused until it has been unregistered. A negative errno code 1808 * is returned on a failure. 1809 * 1810 * When registered all registration and up events are replayed 1811 * to the new notifier to allow device to have a race free 1812 * view of the network device list. 1813 */ 1814 1815 int register_netdevice_notifier(struct notifier_block *nb) 1816 { 1817 struct net *net; 1818 int err; 1819 1820 /* Close race with setup_net() and cleanup_net() */ 1821 down_write(&pernet_ops_rwsem); 1822 rtnl_lock(); 1823 err = raw_notifier_chain_register(&netdev_chain, nb); 1824 if (err) 1825 goto unlock; 1826 if (dev_boot_phase) 1827 goto unlock; 1828 for_each_net(net) { 1829 err = call_netdevice_register_net_notifiers(nb, net); 1830 if (err) 1831 goto rollback; 1832 } 1833 1834 unlock: 1835 rtnl_unlock(); 1836 up_write(&pernet_ops_rwsem); 1837 return err; 1838 1839 rollback: 1840 for_each_net_continue_reverse(net) 1841 call_netdevice_unregister_net_notifiers(nb, net); 1842 1843 raw_notifier_chain_unregister(&netdev_chain, nb); 1844 goto unlock; 1845 } 1846 EXPORT_SYMBOL(register_netdevice_notifier); 1847 1848 /** 1849 * unregister_netdevice_notifier - unregister a network notifier block 1850 * @nb: notifier 1851 * 1852 * Unregister a notifier previously registered by 1853 * register_netdevice_notifier(). The notifier is unlinked into the 1854 * kernel structures and may then be reused. A negative errno code 1855 * is returned on a failure. 1856 * 1857 * After unregistering unregister and down device events are synthesized 1858 * for all devices on the device list to the removed notifier to remove 1859 * the need for special case cleanup code. 1860 */ 1861 1862 int unregister_netdevice_notifier(struct notifier_block *nb) 1863 { 1864 struct net *net; 1865 int err; 1866 1867 /* Close race with setup_net() and cleanup_net() */ 1868 down_write(&pernet_ops_rwsem); 1869 rtnl_lock(); 1870 err = raw_notifier_chain_unregister(&netdev_chain, nb); 1871 if (err) 1872 goto unlock; 1873 1874 for_each_net(net) 1875 call_netdevice_unregister_net_notifiers(nb, net); 1876 1877 unlock: 1878 rtnl_unlock(); 1879 up_write(&pernet_ops_rwsem); 1880 return err; 1881 } 1882 EXPORT_SYMBOL(unregister_netdevice_notifier); 1883 1884 static int __register_netdevice_notifier_net(struct net *net, 1885 struct notifier_block *nb, 1886 bool ignore_call_fail) 1887 { 1888 int err; 1889 1890 err = raw_notifier_chain_register(&net->netdev_chain, nb); 1891 if (err) 1892 return err; 1893 if (dev_boot_phase) 1894 return 0; 1895 1896 err = call_netdevice_register_net_notifiers(nb, net); 1897 if (err && !ignore_call_fail) 1898 goto chain_unregister; 1899 1900 return 0; 1901 1902 chain_unregister: 1903 raw_notifier_chain_unregister(&net->netdev_chain, nb); 1904 return err; 1905 } 1906 1907 static int __unregister_netdevice_notifier_net(struct net *net, 1908 struct notifier_block *nb) 1909 { 1910 int err; 1911 1912 err = raw_notifier_chain_unregister(&net->netdev_chain, nb); 1913 if (err) 1914 return err; 1915 1916 call_netdevice_unregister_net_notifiers(nb, net); 1917 return 0; 1918 } 1919 1920 /** 1921 * register_netdevice_notifier_net - register a per-netns network notifier block 1922 * @net: network namespace 1923 * @nb: notifier 1924 * 1925 * Register a notifier to be called when network device events occur. 1926 * The notifier passed is linked into the kernel structures and must 1927 * not be reused until it has been unregistered. A negative errno code 1928 * is returned on a failure. 1929 * 1930 * When registered all registration and up events are replayed 1931 * to the new notifier to allow device to have a race free 1932 * view of the network device list. 1933 */ 1934 1935 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb) 1936 { 1937 int err; 1938 1939 rtnl_lock(); 1940 err = __register_netdevice_notifier_net(net, nb, false); 1941 rtnl_unlock(); 1942 return err; 1943 } 1944 EXPORT_SYMBOL(register_netdevice_notifier_net); 1945 1946 /** 1947 * unregister_netdevice_notifier_net - unregister a per-netns 1948 * network notifier block 1949 * @net: network namespace 1950 * @nb: notifier 1951 * 1952 * Unregister a notifier previously registered by 1953 * register_netdevice_notifier(). The notifier is unlinked into the 1954 * kernel structures and may then be reused. A negative errno code 1955 * is returned on a failure. 1956 * 1957 * After unregistering unregister and down device events are synthesized 1958 * for all devices on the device list to the removed notifier to remove 1959 * the need for special case cleanup code. 1960 */ 1961 1962 int unregister_netdevice_notifier_net(struct net *net, 1963 struct notifier_block *nb) 1964 { 1965 int err; 1966 1967 rtnl_lock(); 1968 err = __unregister_netdevice_notifier_net(net, nb); 1969 rtnl_unlock(); 1970 return err; 1971 } 1972 EXPORT_SYMBOL(unregister_netdevice_notifier_net); 1973 1974 int register_netdevice_notifier_dev_net(struct net_device *dev, 1975 struct notifier_block *nb, 1976 struct netdev_net_notifier *nn) 1977 { 1978 int err; 1979 1980 rtnl_lock(); 1981 err = __register_netdevice_notifier_net(dev_net(dev), nb, false); 1982 if (!err) { 1983 nn->nb = nb; 1984 list_add(&nn->list, &dev->net_notifier_list); 1985 } 1986 rtnl_unlock(); 1987 return err; 1988 } 1989 EXPORT_SYMBOL(register_netdevice_notifier_dev_net); 1990 1991 int unregister_netdevice_notifier_dev_net(struct net_device *dev, 1992 struct notifier_block *nb, 1993 struct netdev_net_notifier *nn) 1994 { 1995 int err; 1996 1997 rtnl_lock(); 1998 list_del(&nn->list); 1999 err = __unregister_netdevice_notifier_net(dev_net(dev), nb); 2000 rtnl_unlock(); 2001 return err; 2002 } 2003 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net); 2004 2005 static void move_netdevice_notifiers_dev_net(struct net_device *dev, 2006 struct net *net) 2007 { 2008 struct netdev_net_notifier *nn; 2009 2010 list_for_each_entry(nn, &dev->net_notifier_list, list) { 2011 __unregister_netdevice_notifier_net(dev_net(dev), nn->nb); 2012 __register_netdevice_notifier_net(net, nn->nb, true); 2013 } 2014 } 2015 2016 /** 2017 * call_netdevice_notifiers_info - call all network notifier blocks 2018 * @val: value passed unmodified to notifier function 2019 * @info: notifier information data 2020 * 2021 * Call all network notifier blocks. Parameters and return value 2022 * are as for raw_notifier_call_chain(). 2023 */ 2024 2025 static int call_netdevice_notifiers_info(unsigned long val, 2026 struct netdev_notifier_info *info) 2027 { 2028 struct net *net = dev_net(info->dev); 2029 int ret; 2030 2031 ASSERT_RTNL(); 2032 2033 /* Run per-netns notifier block chain first, then run the global one. 2034 * Hopefully, one day, the global one is going to be removed after 2035 * all notifier block registrators get converted to be per-netns. 2036 */ 2037 ret = raw_notifier_call_chain(&net->netdev_chain, val, info); 2038 if (ret & NOTIFY_STOP_MASK) 2039 return ret; 2040 return raw_notifier_call_chain(&netdev_chain, val, info); 2041 } 2042 2043 static int call_netdevice_notifiers_extack(unsigned long val, 2044 struct net_device *dev, 2045 struct netlink_ext_ack *extack) 2046 { 2047 struct netdev_notifier_info info = { 2048 .dev = dev, 2049 .extack = extack, 2050 }; 2051 2052 return call_netdevice_notifiers_info(val, &info); 2053 } 2054 2055 /** 2056 * call_netdevice_notifiers - call all network notifier blocks 2057 * @val: value passed unmodified to notifier function 2058 * @dev: net_device pointer passed unmodified to notifier function 2059 * 2060 * Call all network notifier blocks. Parameters and return value 2061 * are as for raw_notifier_call_chain(). 2062 */ 2063 2064 int call_netdevice_notifiers(unsigned long val, struct net_device *dev) 2065 { 2066 return call_netdevice_notifiers_extack(val, dev, NULL); 2067 } 2068 EXPORT_SYMBOL(call_netdevice_notifiers); 2069 2070 /** 2071 * call_netdevice_notifiers_mtu - call all network notifier blocks 2072 * @val: value passed unmodified to notifier function 2073 * @dev: net_device pointer passed unmodified to notifier function 2074 * @arg: additional u32 argument passed to the notifier function 2075 * 2076 * Call all network notifier blocks. Parameters and return value 2077 * are as for raw_notifier_call_chain(). 2078 */ 2079 static int call_netdevice_notifiers_mtu(unsigned long val, 2080 struct net_device *dev, u32 arg) 2081 { 2082 struct netdev_notifier_info_ext info = { 2083 .info.dev = dev, 2084 .ext.mtu = arg, 2085 }; 2086 2087 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0); 2088 2089 return call_netdevice_notifiers_info(val, &info.info); 2090 } 2091 2092 #ifdef CONFIG_NET_INGRESS 2093 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key); 2094 2095 void net_inc_ingress_queue(void) 2096 { 2097 static_branch_inc(&ingress_needed_key); 2098 } 2099 EXPORT_SYMBOL_GPL(net_inc_ingress_queue); 2100 2101 void net_dec_ingress_queue(void) 2102 { 2103 static_branch_dec(&ingress_needed_key); 2104 } 2105 EXPORT_SYMBOL_GPL(net_dec_ingress_queue); 2106 #endif 2107 2108 #ifdef CONFIG_NET_EGRESS 2109 static DEFINE_STATIC_KEY_FALSE(egress_needed_key); 2110 2111 void net_inc_egress_queue(void) 2112 { 2113 static_branch_inc(&egress_needed_key); 2114 } 2115 EXPORT_SYMBOL_GPL(net_inc_egress_queue); 2116 2117 void net_dec_egress_queue(void) 2118 { 2119 static_branch_dec(&egress_needed_key); 2120 } 2121 EXPORT_SYMBOL_GPL(net_dec_egress_queue); 2122 #endif 2123 2124 static DEFINE_STATIC_KEY_FALSE(netstamp_needed_key); 2125 #ifdef CONFIG_JUMP_LABEL 2126 static atomic_t netstamp_needed_deferred; 2127 static atomic_t netstamp_wanted; 2128 static void netstamp_clear(struct work_struct *work) 2129 { 2130 int deferred = atomic_xchg(&netstamp_needed_deferred, 0); 2131 int wanted; 2132 2133 wanted = atomic_add_return(deferred, &netstamp_wanted); 2134 if (wanted > 0) 2135 static_branch_enable(&netstamp_needed_key); 2136 else 2137 static_branch_disable(&netstamp_needed_key); 2138 } 2139 static DECLARE_WORK(netstamp_work, netstamp_clear); 2140 #endif 2141 2142 void net_enable_timestamp(void) 2143 { 2144 #ifdef CONFIG_JUMP_LABEL 2145 int wanted; 2146 2147 while (1) { 2148 wanted = atomic_read(&netstamp_wanted); 2149 if (wanted <= 0) 2150 break; 2151 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted) 2152 return; 2153 } 2154 atomic_inc(&netstamp_needed_deferred); 2155 schedule_work(&netstamp_work); 2156 #else 2157 static_branch_inc(&netstamp_needed_key); 2158 #endif 2159 } 2160 EXPORT_SYMBOL(net_enable_timestamp); 2161 2162 void net_disable_timestamp(void) 2163 { 2164 #ifdef CONFIG_JUMP_LABEL 2165 int wanted; 2166 2167 while (1) { 2168 wanted = atomic_read(&netstamp_wanted); 2169 if (wanted <= 1) 2170 break; 2171 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted) 2172 return; 2173 } 2174 atomic_dec(&netstamp_needed_deferred); 2175 schedule_work(&netstamp_work); 2176 #else 2177 static_branch_dec(&netstamp_needed_key); 2178 #endif 2179 } 2180 EXPORT_SYMBOL(net_disable_timestamp); 2181 2182 static inline void net_timestamp_set(struct sk_buff *skb) 2183 { 2184 skb->tstamp = 0; 2185 if (static_branch_unlikely(&netstamp_needed_key)) 2186 __net_timestamp(skb); 2187 } 2188 2189 #define net_timestamp_check(COND, SKB) \ 2190 if (static_branch_unlikely(&netstamp_needed_key)) { \ 2191 if ((COND) && !(SKB)->tstamp) \ 2192 __net_timestamp(SKB); \ 2193 } \ 2194 2195 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb) 2196 { 2197 unsigned int len; 2198 2199 if (!(dev->flags & IFF_UP)) 2200 return false; 2201 2202 len = dev->mtu + dev->hard_header_len + VLAN_HLEN; 2203 if (skb->len <= len) 2204 return true; 2205 2206 /* if TSO is enabled, we don't care about the length as the packet 2207 * could be forwarded without being segmented before 2208 */ 2209 if (skb_is_gso(skb)) 2210 return true; 2211 2212 return false; 2213 } 2214 EXPORT_SYMBOL_GPL(is_skb_forwardable); 2215 2216 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2217 { 2218 int ret = ____dev_forward_skb(dev, skb); 2219 2220 if (likely(!ret)) { 2221 skb->protocol = eth_type_trans(skb, dev); 2222 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); 2223 } 2224 2225 return ret; 2226 } 2227 EXPORT_SYMBOL_GPL(__dev_forward_skb); 2228 2229 /** 2230 * dev_forward_skb - loopback an skb to another netif 2231 * 2232 * @dev: destination network device 2233 * @skb: buffer to forward 2234 * 2235 * return values: 2236 * NET_RX_SUCCESS (no congestion) 2237 * NET_RX_DROP (packet was dropped, but freed) 2238 * 2239 * dev_forward_skb can be used for injecting an skb from the 2240 * start_xmit function of one device into the receive queue 2241 * of another device. 2242 * 2243 * The receiving device may be in another namespace, so 2244 * we have to clear all information in the skb that could 2245 * impact namespace isolation. 2246 */ 2247 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2248 { 2249 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); 2250 } 2251 EXPORT_SYMBOL_GPL(dev_forward_skb); 2252 2253 static inline int deliver_skb(struct sk_buff *skb, 2254 struct packet_type *pt_prev, 2255 struct net_device *orig_dev) 2256 { 2257 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 2258 return -ENOMEM; 2259 refcount_inc(&skb->users); 2260 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 2261 } 2262 2263 static inline void deliver_ptype_list_skb(struct sk_buff *skb, 2264 struct packet_type **pt, 2265 struct net_device *orig_dev, 2266 __be16 type, 2267 struct list_head *ptype_list) 2268 { 2269 struct packet_type *ptype, *pt_prev = *pt; 2270 2271 list_for_each_entry_rcu(ptype, ptype_list, list) { 2272 if (ptype->type != type) 2273 continue; 2274 if (pt_prev) 2275 deliver_skb(skb, pt_prev, orig_dev); 2276 pt_prev = ptype; 2277 } 2278 *pt = pt_prev; 2279 } 2280 2281 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) 2282 { 2283 if (!ptype->af_packet_priv || !skb->sk) 2284 return false; 2285 2286 if (ptype->id_match) 2287 return ptype->id_match(ptype, skb->sk); 2288 else if ((struct sock *)ptype->af_packet_priv == skb->sk) 2289 return true; 2290 2291 return false; 2292 } 2293 2294 /** 2295 * dev_nit_active - return true if any network interface taps are in use 2296 * 2297 * @dev: network device to check for the presence of taps 2298 */ 2299 bool dev_nit_active(struct net_device *dev) 2300 { 2301 return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all); 2302 } 2303 EXPORT_SYMBOL_GPL(dev_nit_active); 2304 2305 /* 2306 * Support routine. Sends outgoing frames to any network 2307 * taps currently in use. 2308 */ 2309 2310 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) 2311 { 2312 struct packet_type *ptype; 2313 struct sk_buff *skb2 = NULL; 2314 struct packet_type *pt_prev = NULL; 2315 struct list_head *ptype_list = &ptype_all; 2316 2317 rcu_read_lock(); 2318 again: 2319 list_for_each_entry_rcu(ptype, ptype_list, list) { 2320 if (ptype->ignore_outgoing) 2321 continue; 2322 2323 /* Never send packets back to the socket 2324 * they originated from - MvS (miquels@drinkel.ow.org) 2325 */ 2326 if (skb_loop_sk(ptype, skb)) 2327 continue; 2328 2329 if (pt_prev) { 2330 deliver_skb(skb2, pt_prev, skb->dev); 2331 pt_prev = ptype; 2332 continue; 2333 } 2334 2335 /* need to clone skb, done only once */ 2336 skb2 = skb_clone(skb, GFP_ATOMIC); 2337 if (!skb2) 2338 goto out_unlock; 2339 2340 net_timestamp_set(skb2); 2341 2342 /* skb->nh should be correctly 2343 * set by sender, so that the second statement is 2344 * just protection against buggy protocols. 2345 */ 2346 skb_reset_mac_header(skb2); 2347 2348 if (skb_network_header(skb2) < skb2->data || 2349 skb_network_header(skb2) > skb_tail_pointer(skb2)) { 2350 net_crit_ratelimited("protocol %04x is buggy, dev %s\n", 2351 ntohs(skb2->protocol), 2352 dev->name); 2353 skb_reset_network_header(skb2); 2354 } 2355 2356 skb2->transport_header = skb2->network_header; 2357 skb2->pkt_type = PACKET_OUTGOING; 2358 pt_prev = ptype; 2359 } 2360 2361 if (ptype_list == &ptype_all) { 2362 ptype_list = &dev->ptype_all; 2363 goto again; 2364 } 2365 out_unlock: 2366 if (pt_prev) { 2367 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC)) 2368 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); 2369 else 2370 kfree_skb(skb2); 2371 } 2372 rcu_read_unlock(); 2373 } 2374 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit); 2375 2376 /** 2377 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change 2378 * @dev: Network device 2379 * @txq: number of queues available 2380 * 2381 * If real_num_tx_queues is changed the tc mappings may no longer be 2382 * valid. To resolve this verify the tc mapping remains valid and if 2383 * not NULL the mapping. With no priorities mapping to this 2384 * offset/count pair it will no longer be used. In the worst case TC0 2385 * is invalid nothing can be done so disable priority mappings. If is 2386 * expected that drivers will fix this mapping if they can before 2387 * calling netif_set_real_num_tx_queues. 2388 */ 2389 static void netif_setup_tc(struct net_device *dev, unsigned int txq) 2390 { 2391 int i; 2392 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2393 2394 /* If TC0 is invalidated disable TC mapping */ 2395 if (tc->offset + tc->count > txq) { 2396 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); 2397 dev->num_tc = 0; 2398 return; 2399 } 2400 2401 /* Invalidated prio to tc mappings set to TC0 */ 2402 for (i = 1; i < TC_BITMASK + 1; i++) { 2403 int q = netdev_get_prio_tc_map(dev, i); 2404 2405 tc = &dev->tc_to_txq[q]; 2406 if (tc->offset + tc->count > txq) { 2407 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n", 2408 i, q); 2409 netdev_set_prio_tc_map(dev, i, 0); 2410 } 2411 } 2412 } 2413 2414 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq) 2415 { 2416 if (dev->num_tc) { 2417 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2418 int i; 2419 2420 /* walk through the TCs and see if it falls into any of them */ 2421 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) { 2422 if ((txq - tc->offset) < tc->count) 2423 return i; 2424 } 2425 2426 /* didn't find it, just return -1 to indicate no match */ 2427 return -1; 2428 } 2429 2430 return 0; 2431 } 2432 EXPORT_SYMBOL(netdev_txq_to_tc); 2433 2434 #ifdef CONFIG_XPS 2435 struct static_key xps_needed __read_mostly; 2436 EXPORT_SYMBOL(xps_needed); 2437 struct static_key xps_rxqs_needed __read_mostly; 2438 EXPORT_SYMBOL(xps_rxqs_needed); 2439 static DEFINE_MUTEX(xps_map_mutex); 2440 #define xmap_dereference(P) \ 2441 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) 2442 2443 static bool remove_xps_queue(struct xps_dev_maps *dev_maps, 2444 int tci, u16 index) 2445 { 2446 struct xps_map *map = NULL; 2447 int pos; 2448 2449 if (dev_maps) 2450 map = xmap_dereference(dev_maps->attr_map[tci]); 2451 if (!map) 2452 return false; 2453 2454 for (pos = map->len; pos--;) { 2455 if (map->queues[pos] != index) 2456 continue; 2457 2458 if (map->len > 1) { 2459 map->queues[pos] = map->queues[--map->len]; 2460 break; 2461 } 2462 2463 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2464 kfree_rcu(map, rcu); 2465 return false; 2466 } 2467 2468 return true; 2469 } 2470 2471 static bool remove_xps_queue_cpu(struct net_device *dev, 2472 struct xps_dev_maps *dev_maps, 2473 int cpu, u16 offset, u16 count) 2474 { 2475 int num_tc = dev->num_tc ? : 1; 2476 bool active = false; 2477 int tci; 2478 2479 for (tci = cpu * num_tc; num_tc--; tci++) { 2480 int i, j; 2481 2482 for (i = count, j = offset; i--; j++) { 2483 if (!remove_xps_queue(dev_maps, tci, j)) 2484 break; 2485 } 2486 2487 active |= i < 0; 2488 } 2489 2490 return active; 2491 } 2492 2493 static void reset_xps_maps(struct net_device *dev, 2494 struct xps_dev_maps *dev_maps, 2495 bool is_rxqs_map) 2496 { 2497 if (is_rxqs_map) { 2498 static_key_slow_dec_cpuslocked(&xps_rxqs_needed); 2499 RCU_INIT_POINTER(dev->xps_rxqs_map, NULL); 2500 } else { 2501 RCU_INIT_POINTER(dev->xps_cpus_map, NULL); 2502 } 2503 static_key_slow_dec_cpuslocked(&xps_needed); 2504 kfree_rcu(dev_maps, rcu); 2505 } 2506 2507 static void clean_xps_maps(struct net_device *dev, const unsigned long *mask, 2508 struct xps_dev_maps *dev_maps, unsigned int nr_ids, 2509 u16 offset, u16 count, bool is_rxqs_map) 2510 { 2511 bool active = false; 2512 int i, j; 2513 2514 for (j = -1; j = netif_attrmask_next(j, mask, nr_ids), 2515 j < nr_ids;) 2516 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, 2517 count); 2518 if (!active) 2519 reset_xps_maps(dev, dev_maps, is_rxqs_map); 2520 2521 if (!is_rxqs_map) { 2522 for (i = offset + (count - 1); count--; i--) { 2523 netdev_queue_numa_node_write( 2524 netdev_get_tx_queue(dev, i), 2525 NUMA_NO_NODE); 2526 } 2527 } 2528 } 2529 2530 static void netif_reset_xps_queues(struct net_device *dev, u16 offset, 2531 u16 count) 2532 { 2533 const unsigned long *possible_mask = NULL; 2534 struct xps_dev_maps *dev_maps; 2535 unsigned int nr_ids; 2536 2537 if (!static_key_false(&xps_needed)) 2538 return; 2539 2540 cpus_read_lock(); 2541 mutex_lock(&xps_map_mutex); 2542 2543 if (static_key_false(&xps_rxqs_needed)) { 2544 dev_maps = xmap_dereference(dev->xps_rxqs_map); 2545 if (dev_maps) { 2546 nr_ids = dev->num_rx_queues; 2547 clean_xps_maps(dev, possible_mask, dev_maps, nr_ids, 2548 offset, count, true); 2549 } 2550 } 2551 2552 dev_maps = xmap_dereference(dev->xps_cpus_map); 2553 if (!dev_maps) 2554 goto out_no_maps; 2555 2556 if (num_possible_cpus() > 1) 2557 possible_mask = cpumask_bits(cpu_possible_mask); 2558 nr_ids = nr_cpu_ids; 2559 clean_xps_maps(dev, possible_mask, dev_maps, nr_ids, offset, count, 2560 false); 2561 2562 out_no_maps: 2563 mutex_unlock(&xps_map_mutex); 2564 cpus_read_unlock(); 2565 } 2566 2567 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) 2568 { 2569 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index); 2570 } 2571 2572 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index, 2573 u16 index, bool is_rxqs_map) 2574 { 2575 struct xps_map *new_map; 2576 int alloc_len = XPS_MIN_MAP_ALLOC; 2577 int i, pos; 2578 2579 for (pos = 0; map && pos < map->len; pos++) { 2580 if (map->queues[pos] != index) 2581 continue; 2582 return map; 2583 } 2584 2585 /* Need to add tx-queue to this CPU's/rx-queue's existing map */ 2586 if (map) { 2587 if (pos < map->alloc_len) 2588 return map; 2589 2590 alloc_len = map->alloc_len * 2; 2591 } 2592 2593 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's 2594 * map 2595 */ 2596 if (is_rxqs_map) 2597 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL); 2598 else 2599 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, 2600 cpu_to_node(attr_index)); 2601 if (!new_map) 2602 return NULL; 2603 2604 for (i = 0; i < pos; i++) 2605 new_map->queues[i] = map->queues[i]; 2606 new_map->alloc_len = alloc_len; 2607 new_map->len = pos; 2608 2609 return new_map; 2610 } 2611 2612 /* Must be called under cpus_read_lock */ 2613 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask, 2614 u16 index, bool is_rxqs_map) 2615 { 2616 const unsigned long *online_mask = NULL, *possible_mask = NULL; 2617 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL; 2618 int i, j, tci, numa_node_id = -2; 2619 int maps_sz, num_tc = 1, tc = 0; 2620 struct xps_map *map, *new_map; 2621 bool active = false; 2622 unsigned int nr_ids; 2623 2624 if (dev->num_tc) { 2625 /* Do not allow XPS on subordinate device directly */ 2626 num_tc = dev->num_tc; 2627 if (num_tc < 0) 2628 return -EINVAL; 2629 2630 /* If queue belongs to subordinate dev use its map */ 2631 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev; 2632 2633 tc = netdev_txq_to_tc(dev, index); 2634 if (tc < 0) 2635 return -EINVAL; 2636 } 2637 2638 mutex_lock(&xps_map_mutex); 2639 if (is_rxqs_map) { 2640 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues); 2641 dev_maps = xmap_dereference(dev->xps_rxqs_map); 2642 nr_ids = dev->num_rx_queues; 2643 } else { 2644 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc); 2645 if (num_possible_cpus() > 1) { 2646 online_mask = cpumask_bits(cpu_online_mask); 2647 possible_mask = cpumask_bits(cpu_possible_mask); 2648 } 2649 dev_maps = xmap_dereference(dev->xps_cpus_map); 2650 nr_ids = nr_cpu_ids; 2651 } 2652 2653 if (maps_sz < L1_CACHE_BYTES) 2654 maps_sz = L1_CACHE_BYTES; 2655 2656 /* allocate memory for queue storage */ 2657 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids), 2658 j < nr_ids;) { 2659 if (!new_dev_maps) 2660 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); 2661 if (!new_dev_maps) { 2662 mutex_unlock(&xps_map_mutex); 2663 return -ENOMEM; 2664 } 2665 2666 tci = j * num_tc + tc; 2667 map = dev_maps ? xmap_dereference(dev_maps->attr_map[tci]) : 2668 NULL; 2669 2670 map = expand_xps_map(map, j, index, is_rxqs_map); 2671 if (!map) 2672 goto error; 2673 2674 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2675 } 2676 2677 if (!new_dev_maps) 2678 goto out_no_new_maps; 2679 2680 if (!dev_maps) { 2681 /* Increment static keys at most once per type */ 2682 static_key_slow_inc_cpuslocked(&xps_needed); 2683 if (is_rxqs_map) 2684 static_key_slow_inc_cpuslocked(&xps_rxqs_needed); 2685 } 2686 2687 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids), 2688 j < nr_ids;) { 2689 /* copy maps belonging to foreign traffic classes */ 2690 for (i = tc, tci = j * num_tc; dev_maps && i--; tci++) { 2691 /* fill in the new device map from the old device map */ 2692 map = xmap_dereference(dev_maps->attr_map[tci]); 2693 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2694 } 2695 2696 /* We need to explicitly update tci as prevous loop 2697 * could break out early if dev_maps is NULL. 2698 */ 2699 tci = j * num_tc + tc; 2700 2701 if (netif_attr_test_mask(j, mask, nr_ids) && 2702 netif_attr_test_online(j, online_mask, nr_ids)) { 2703 /* add tx-queue to CPU/rx-queue maps */ 2704 int pos = 0; 2705 2706 map = xmap_dereference(new_dev_maps->attr_map[tci]); 2707 while ((pos < map->len) && (map->queues[pos] != index)) 2708 pos++; 2709 2710 if (pos == map->len) 2711 map->queues[map->len++] = index; 2712 #ifdef CONFIG_NUMA 2713 if (!is_rxqs_map) { 2714 if (numa_node_id == -2) 2715 numa_node_id = cpu_to_node(j); 2716 else if (numa_node_id != cpu_to_node(j)) 2717 numa_node_id = -1; 2718 } 2719 #endif 2720 } else if (dev_maps) { 2721 /* fill in the new device map from the old device map */ 2722 map = xmap_dereference(dev_maps->attr_map[tci]); 2723 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2724 } 2725 2726 /* copy maps belonging to foreign traffic classes */ 2727 for (i = num_tc - tc, tci++; dev_maps && --i; tci++) { 2728 /* fill in the new device map from the old device map */ 2729 map = xmap_dereference(dev_maps->attr_map[tci]); 2730 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2731 } 2732 } 2733 2734 if (is_rxqs_map) 2735 rcu_assign_pointer(dev->xps_rxqs_map, new_dev_maps); 2736 else 2737 rcu_assign_pointer(dev->xps_cpus_map, new_dev_maps); 2738 2739 /* Cleanup old maps */ 2740 if (!dev_maps) 2741 goto out_no_old_maps; 2742 2743 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids), 2744 j < nr_ids;) { 2745 for (i = num_tc, tci = j * num_tc; i--; tci++) { 2746 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2747 map = xmap_dereference(dev_maps->attr_map[tci]); 2748 if (map && map != new_map) 2749 kfree_rcu(map, rcu); 2750 } 2751 } 2752 2753 kfree_rcu(dev_maps, rcu); 2754 2755 out_no_old_maps: 2756 dev_maps = new_dev_maps; 2757 active = true; 2758 2759 out_no_new_maps: 2760 if (!is_rxqs_map) { 2761 /* update Tx queue numa node */ 2762 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), 2763 (numa_node_id >= 0) ? 2764 numa_node_id : NUMA_NO_NODE); 2765 } 2766 2767 if (!dev_maps) 2768 goto out_no_maps; 2769 2770 /* removes tx-queue from unused CPUs/rx-queues */ 2771 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids), 2772 j < nr_ids;) { 2773 for (i = tc, tci = j * num_tc; i--; tci++) 2774 active |= remove_xps_queue(dev_maps, tci, index); 2775 if (!netif_attr_test_mask(j, mask, nr_ids) || 2776 !netif_attr_test_online(j, online_mask, nr_ids)) 2777 active |= remove_xps_queue(dev_maps, tci, index); 2778 for (i = num_tc - tc, tci++; --i; tci++) 2779 active |= remove_xps_queue(dev_maps, tci, index); 2780 } 2781 2782 /* free map if not active */ 2783 if (!active) 2784 reset_xps_maps(dev, dev_maps, is_rxqs_map); 2785 2786 out_no_maps: 2787 mutex_unlock(&xps_map_mutex); 2788 2789 return 0; 2790 error: 2791 /* remove any maps that we added */ 2792 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids), 2793 j < nr_ids;) { 2794 for (i = num_tc, tci = j * num_tc; i--; tci++) { 2795 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2796 map = dev_maps ? 2797 xmap_dereference(dev_maps->attr_map[tci]) : 2798 NULL; 2799 if (new_map && new_map != map) 2800 kfree(new_map); 2801 } 2802 } 2803 2804 mutex_unlock(&xps_map_mutex); 2805 2806 kfree(new_dev_maps); 2807 return -ENOMEM; 2808 } 2809 EXPORT_SYMBOL_GPL(__netif_set_xps_queue); 2810 2811 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, 2812 u16 index) 2813 { 2814 int ret; 2815 2816 cpus_read_lock(); 2817 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, false); 2818 cpus_read_unlock(); 2819 2820 return ret; 2821 } 2822 EXPORT_SYMBOL(netif_set_xps_queue); 2823 2824 #endif 2825 static void netdev_unbind_all_sb_channels(struct net_device *dev) 2826 { 2827 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2828 2829 /* Unbind any subordinate channels */ 2830 while (txq-- != &dev->_tx[0]) { 2831 if (txq->sb_dev) 2832 netdev_unbind_sb_channel(dev, txq->sb_dev); 2833 } 2834 } 2835 2836 void netdev_reset_tc(struct net_device *dev) 2837 { 2838 #ifdef CONFIG_XPS 2839 netif_reset_xps_queues_gt(dev, 0); 2840 #endif 2841 netdev_unbind_all_sb_channels(dev); 2842 2843 /* Reset TC configuration of device */ 2844 dev->num_tc = 0; 2845 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq)); 2846 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map)); 2847 } 2848 EXPORT_SYMBOL(netdev_reset_tc); 2849 2850 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset) 2851 { 2852 if (tc >= dev->num_tc) 2853 return -EINVAL; 2854 2855 #ifdef CONFIG_XPS 2856 netif_reset_xps_queues(dev, offset, count); 2857 #endif 2858 dev->tc_to_txq[tc].count = count; 2859 dev->tc_to_txq[tc].offset = offset; 2860 return 0; 2861 } 2862 EXPORT_SYMBOL(netdev_set_tc_queue); 2863 2864 int netdev_set_num_tc(struct net_device *dev, u8 num_tc) 2865 { 2866 if (num_tc > TC_MAX_QUEUE) 2867 return -EINVAL; 2868 2869 #ifdef CONFIG_XPS 2870 netif_reset_xps_queues_gt(dev, 0); 2871 #endif 2872 netdev_unbind_all_sb_channels(dev); 2873 2874 dev->num_tc = num_tc; 2875 return 0; 2876 } 2877 EXPORT_SYMBOL(netdev_set_num_tc); 2878 2879 void netdev_unbind_sb_channel(struct net_device *dev, 2880 struct net_device *sb_dev) 2881 { 2882 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2883 2884 #ifdef CONFIG_XPS 2885 netif_reset_xps_queues_gt(sb_dev, 0); 2886 #endif 2887 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq)); 2888 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map)); 2889 2890 while (txq-- != &dev->_tx[0]) { 2891 if (txq->sb_dev == sb_dev) 2892 txq->sb_dev = NULL; 2893 } 2894 } 2895 EXPORT_SYMBOL(netdev_unbind_sb_channel); 2896 2897 int netdev_bind_sb_channel_queue(struct net_device *dev, 2898 struct net_device *sb_dev, 2899 u8 tc, u16 count, u16 offset) 2900 { 2901 /* Make certain the sb_dev and dev are already configured */ 2902 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc) 2903 return -EINVAL; 2904 2905 /* We cannot hand out queues we don't have */ 2906 if ((offset + count) > dev->real_num_tx_queues) 2907 return -EINVAL; 2908 2909 /* Record the mapping */ 2910 sb_dev->tc_to_txq[tc].count = count; 2911 sb_dev->tc_to_txq[tc].offset = offset; 2912 2913 /* Provide a way for Tx queue to find the tc_to_txq map or 2914 * XPS map for itself. 2915 */ 2916 while (count--) 2917 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev; 2918 2919 return 0; 2920 } 2921 EXPORT_SYMBOL(netdev_bind_sb_channel_queue); 2922 2923 int netdev_set_sb_channel(struct net_device *dev, u16 channel) 2924 { 2925 /* Do not use a multiqueue device to represent a subordinate channel */ 2926 if (netif_is_multiqueue(dev)) 2927 return -ENODEV; 2928 2929 /* We allow channels 1 - 32767 to be used for subordinate channels. 2930 * Channel 0 is meant to be "native" mode and used only to represent 2931 * the main root device. We allow writing 0 to reset the device back 2932 * to normal mode after being used as a subordinate channel. 2933 */ 2934 if (channel > S16_MAX) 2935 return -EINVAL; 2936 2937 dev->num_tc = -channel; 2938 2939 return 0; 2940 } 2941 EXPORT_SYMBOL(netdev_set_sb_channel); 2942 2943 /* 2944 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues 2945 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed. 2946 */ 2947 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) 2948 { 2949 bool disabling; 2950 int rc; 2951 2952 disabling = txq < dev->real_num_tx_queues; 2953 2954 if (txq < 1 || txq > dev->num_tx_queues) 2955 return -EINVAL; 2956 2957 if (dev->reg_state == NETREG_REGISTERED || 2958 dev->reg_state == NETREG_UNREGISTERING) { 2959 ASSERT_RTNL(); 2960 2961 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, 2962 txq); 2963 if (rc) 2964 return rc; 2965 2966 if (dev->num_tc) 2967 netif_setup_tc(dev, txq); 2968 2969 dev->real_num_tx_queues = txq; 2970 2971 if (disabling) { 2972 synchronize_net(); 2973 qdisc_reset_all_tx_gt(dev, txq); 2974 #ifdef CONFIG_XPS 2975 netif_reset_xps_queues_gt(dev, txq); 2976 #endif 2977 } 2978 } else { 2979 dev->real_num_tx_queues = txq; 2980 } 2981 2982 return 0; 2983 } 2984 EXPORT_SYMBOL(netif_set_real_num_tx_queues); 2985 2986 #ifdef CONFIG_SYSFS 2987 /** 2988 * netif_set_real_num_rx_queues - set actual number of RX queues used 2989 * @dev: Network device 2990 * @rxq: Actual number of RX queues 2991 * 2992 * This must be called either with the rtnl_lock held or before 2993 * registration of the net device. Returns 0 on success, or a 2994 * negative error code. If called before registration, it always 2995 * succeeds. 2996 */ 2997 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) 2998 { 2999 int rc; 3000 3001 if (rxq < 1 || rxq > dev->num_rx_queues) 3002 return -EINVAL; 3003 3004 if (dev->reg_state == NETREG_REGISTERED) { 3005 ASSERT_RTNL(); 3006 3007 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, 3008 rxq); 3009 if (rc) 3010 return rc; 3011 } 3012 3013 dev->real_num_rx_queues = rxq; 3014 return 0; 3015 } 3016 EXPORT_SYMBOL(netif_set_real_num_rx_queues); 3017 #endif 3018 3019 /** 3020 * netif_get_num_default_rss_queues - default number of RSS queues 3021 * 3022 * This routine should set an upper limit on the number of RSS queues 3023 * used by default by multiqueue devices. 3024 */ 3025 int netif_get_num_default_rss_queues(void) 3026 { 3027 return is_kdump_kernel() ? 3028 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus()); 3029 } 3030 EXPORT_SYMBOL(netif_get_num_default_rss_queues); 3031 3032 static void __netif_reschedule(struct Qdisc *q) 3033 { 3034 struct softnet_data *sd; 3035 unsigned long flags; 3036 3037 local_irq_save(flags); 3038 sd = this_cpu_ptr(&softnet_data); 3039 q->next_sched = NULL; 3040 *sd->output_queue_tailp = q; 3041 sd->output_queue_tailp = &q->next_sched; 3042 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3043 local_irq_restore(flags); 3044 } 3045 3046 void __netif_schedule(struct Qdisc *q) 3047 { 3048 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) 3049 __netif_reschedule(q); 3050 } 3051 EXPORT_SYMBOL(__netif_schedule); 3052 3053 struct dev_kfree_skb_cb { 3054 enum skb_free_reason reason; 3055 }; 3056 3057 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) 3058 { 3059 return (struct dev_kfree_skb_cb *)skb->cb; 3060 } 3061 3062 void netif_schedule_queue(struct netdev_queue *txq) 3063 { 3064 rcu_read_lock(); 3065 if (!netif_xmit_stopped(txq)) { 3066 struct Qdisc *q = rcu_dereference(txq->qdisc); 3067 3068 __netif_schedule(q); 3069 } 3070 rcu_read_unlock(); 3071 } 3072 EXPORT_SYMBOL(netif_schedule_queue); 3073 3074 void netif_tx_wake_queue(struct netdev_queue *dev_queue) 3075 { 3076 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { 3077 struct Qdisc *q; 3078 3079 rcu_read_lock(); 3080 q = rcu_dereference(dev_queue->qdisc); 3081 __netif_schedule(q); 3082 rcu_read_unlock(); 3083 } 3084 } 3085 EXPORT_SYMBOL(netif_tx_wake_queue); 3086 3087 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason) 3088 { 3089 unsigned long flags; 3090 3091 if (unlikely(!skb)) 3092 return; 3093 3094 if (likely(refcount_read(&skb->users) == 1)) { 3095 smp_rmb(); 3096 refcount_set(&skb->users, 0); 3097 } else if (likely(!refcount_dec_and_test(&skb->users))) { 3098 return; 3099 } 3100 get_kfree_skb_cb(skb)->reason = reason; 3101 local_irq_save(flags); 3102 skb->next = __this_cpu_read(softnet_data.completion_queue); 3103 __this_cpu_write(softnet_data.completion_queue, skb); 3104 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3105 local_irq_restore(flags); 3106 } 3107 EXPORT_SYMBOL(__dev_kfree_skb_irq); 3108 3109 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason) 3110 { 3111 if (in_irq() || irqs_disabled()) 3112 __dev_kfree_skb_irq(skb, reason); 3113 else 3114 dev_kfree_skb(skb); 3115 } 3116 EXPORT_SYMBOL(__dev_kfree_skb_any); 3117 3118 3119 /** 3120 * netif_device_detach - mark device as removed 3121 * @dev: network device 3122 * 3123 * Mark device as removed from system and therefore no longer available. 3124 */ 3125 void netif_device_detach(struct net_device *dev) 3126 { 3127 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && 3128 netif_running(dev)) { 3129 netif_tx_stop_all_queues(dev); 3130 } 3131 } 3132 EXPORT_SYMBOL(netif_device_detach); 3133 3134 /** 3135 * netif_device_attach - mark device as attached 3136 * @dev: network device 3137 * 3138 * Mark device as attached from system and restart if needed. 3139 */ 3140 void netif_device_attach(struct net_device *dev) 3141 { 3142 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && 3143 netif_running(dev)) { 3144 netif_tx_wake_all_queues(dev); 3145 __netdev_watchdog_up(dev); 3146 } 3147 } 3148 EXPORT_SYMBOL(netif_device_attach); 3149 3150 /* 3151 * Returns a Tx hash based on the given packet descriptor a Tx queues' number 3152 * to be used as a distribution range. 3153 */ 3154 static u16 skb_tx_hash(const struct net_device *dev, 3155 const struct net_device *sb_dev, 3156 struct sk_buff *skb) 3157 { 3158 u32 hash; 3159 u16 qoffset = 0; 3160 u16 qcount = dev->real_num_tx_queues; 3161 3162 if (dev->num_tc) { 3163 u8 tc = netdev_get_prio_tc_map(dev, skb->priority); 3164 3165 qoffset = sb_dev->tc_to_txq[tc].offset; 3166 qcount = sb_dev->tc_to_txq[tc].count; 3167 } 3168 3169 if (skb_rx_queue_recorded(skb)) { 3170 hash = skb_get_rx_queue(skb); 3171 if (hash >= qoffset) 3172 hash -= qoffset; 3173 while (unlikely(hash >= qcount)) 3174 hash -= qcount; 3175 return hash + qoffset; 3176 } 3177 3178 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; 3179 } 3180 3181 static void skb_warn_bad_offload(const struct sk_buff *skb) 3182 { 3183 static const netdev_features_t null_features; 3184 struct net_device *dev = skb->dev; 3185 const char *name = ""; 3186 3187 if (!net_ratelimit()) 3188 return; 3189 3190 if (dev) { 3191 if (dev->dev.parent) 3192 name = dev_driver_string(dev->dev.parent); 3193 else 3194 name = netdev_name(dev); 3195 } 3196 skb_dump(KERN_WARNING, skb, false); 3197 WARN(1, "%s: caps=(%pNF, %pNF)\n", 3198 name, dev ? &dev->features : &null_features, 3199 skb->sk ? &skb->sk->sk_route_caps : &null_features); 3200 } 3201 3202 /* 3203 * Invalidate hardware checksum when packet is to be mangled, and 3204 * complete checksum manually on outgoing path. 3205 */ 3206 int skb_checksum_help(struct sk_buff *skb) 3207 { 3208 __wsum csum; 3209 int ret = 0, offset; 3210 3211 if (skb->ip_summed == CHECKSUM_COMPLETE) 3212 goto out_set_summed; 3213 3214 if (unlikely(skb_is_gso(skb))) { 3215 skb_warn_bad_offload(skb); 3216 return -EINVAL; 3217 } 3218 3219 /* Before computing a checksum, we should make sure no frag could 3220 * be modified by an external entity : checksum could be wrong. 3221 */ 3222 if (skb_has_shared_frag(skb)) { 3223 ret = __skb_linearize(skb); 3224 if (ret) 3225 goto out; 3226 } 3227 3228 offset = skb_checksum_start_offset(skb); 3229 BUG_ON(offset >= skb_headlen(skb)); 3230 csum = skb_checksum(skb, offset, skb->len - offset, 0); 3231 3232 offset += skb->csum_offset; 3233 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); 3234 3235 ret = skb_ensure_writable(skb, offset + sizeof(__sum16)); 3236 if (ret) 3237 goto out; 3238 3239 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0; 3240 out_set_summed: 3241 skb->ip_summed = CHECKSUM_NONE; 3242 out: 3243 return ret; 3244 } 3245 EXPORT_SYMBOL(skb_checksum_help); 3246 3247 int skb_crc32c_csum_help(struct sk_buff *skb) 3248 { 3249 __le32 crc32c_csum; 3250 int ret = 0, offset, start; 3251 3252 if (skb->ip_summed != CHECKSUM_PARTIAL) 3253 goto out; 3254 3255 if (unlikely(skb_is_gso(skb))) 3256 goto out; 3257 3258 /* Before computing a checksum, we should make sure no frag could 3259 * be modified by an external entity : checksum could be wrong. 3260 */ 3261 if (unlikely(skb_has_shared_frag(skb))) { 3262 ret = __skb_linearize(skb); 3263 if (ret) 3264 goto out; 3265 } 3266 start = skb_checksum_start_offset(skb); 3267 offset = start + offsetof(struct sctphdr, checksum); 3268 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { 3269 ret = -EINVAL; 3270 goto out; 3271 } 3272 3273 ret = skb_ensure_writable(skb, offset + sizeof(__le32)); 3274 if (ret) 3275 goto out; 3276 3277 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start, 3278 skb->len - start, ~(__u32)0, 3279 crc32c_csum_stub)); 3280 *(__le32 *)(skb->data + offset) = crc32c_csum; 3281 skb->ip_summed = CHECKSUM_NONE; 3282 skb->csum_not_inet = 0; 3283 out: 3284 return ret; 3285 } 3286 3287 __be16 skb_network_protocol(struct sk_buff *skb, int *depth) 3288 { 3289 __be16 type = skb->protocol; 3290 3291 /* Tunnel gso handlers can set protocol to ethernet. */ 3292 if (type == htons(ETH_P_TEB)) { 3293 struct ethhdr *eth; 3294 3295 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) 3296 return 0; 3297 3298 eth = (struct ethhdr *)skb->data; 3299 type = eth->h_proto; 3300 } 3301 3302 return __vlan_get_protocol(skb, type, depth); 3303 } 3304 3305 /** 3306 * skb_mac_gso_segment - mac layer segmentation handler. 3307 * @skb: buffer to segment 3308 * @features: features for the output path (see dev->features) 3309 */ 3310 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb, 3311 netdev_features_t features) 3312 { 3313 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); 3314 struct packet_offload *ptype; 3315 int vlan_depth = skb->mac_len; 3316 __be16 type = skb_network_protocol(skb, &vlan_depth); 3317 3318 if (unlikely(!type)) 3319 return ERR_PTR(-EINVAL); 3320 3321 __skb_pull(skb, vlan_depth); 3322 3323 rcu_read_lock(); 3324 list_for_each_entry_rcu(ptype, &offload_base, list) { 3325 if (ptype->type == type && ptype->callbacks.gso_segment) { 3326 segs = ptype->callbacks.gso_segment(skb, features); 3327 break; 3328 } 3329 } 3330 rcu_read_unlock(); 3331 3332 __skb_push(skb, skb->data - skb_mac_header(skb)); 3333 3334 return segs; 3335 } 3336 EXPORT_SYMBOL(skb_mac_gso_segment); 3337 3338 3339 /* openvswitch calls this on rx path, so we need a different check. 3340 */ 3341 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path) 3342 { 3343 if (tx_path) 3344 return skb->ip_summed != CHECKSUM_PARTIAL && 3345 skb->ip_summed != CHECKSUM_UNNECESSARY; 3346 3347 return skb->ip_summed == CHECKSUM_NONE; 3348 } 3349 3350 /** 3351 * __skb_gso_segment - Perform segmentation on skb. 3352 * @skb: buffer to segment 3353 * @features: features for the output path (see dev->features) 3354 * @tx_path: whether it is called in TX path 3355 * 3356 * This function segments the given skb and returns a list of segments. 3357 * 3358 * It may return NULL if the skb requires no segmentation. This is 3359 * only possible when GSO is used for verifying header integrity. 3360 * 3361 * Segmentation preserves SKB_GSO_CB_OFFSET bytes of previous skb cb. 3362 */ 3363 struct sk_buff *__skb_gso_segment(struct sk_buff *skb, 3364 netdev_features_t features, bool tx_path) 3365 { 3366 struct sk_buff *segs; 3367 3368 if (unlikely(skb_needs_check(skb, tx_path))) { 3369 int err; 3370 3371 /* We're going to init ->check field in TCP or UDP header */ 3372 err = skb_cow_head(skb, 0); 3373 if (err < 0) 3374 return ERR_PTR(err); 3375 } 3376 3377 /* Only report GSO partial support if it will enable us to 3378 * support segmentation on this frame without needing additional 3379 * work. 3380 */ 3381 if (features & NETIF_F_GSO_PARTIAL) { 3382 netdev_features_t partial_features = NETIF_F_GSO_ROBUST; 3383 struct net_device *dev = skb->dev; 3384 3385 partial_features |= dev->features & dev->gso_partial_features; 3386 if (!skb_gso_ok(skb, features | partial_features)) 3387 features &= ~NETIF_F_GSO_PARTIAL; 3388 } 3389 3390 BUILD_BUG_ON(SKB_GSO_CB_OFFSET + 3391 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb)); 3392 3393 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb); 3394 SKB_GSO_CB(skb)->encap_level = 0; 3395 3396 skb_reset_mac_header(skb); 3397 skb_reset_mac_len(skb); 3398 3399 segs = skb_mac_gso_segment(skb, features); 3400 3401 if (segs != skb && unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs))) 3402 skb_warn_bad_offload(skb); 3403 3404 return segs; 3405 } 3406 EXPORT_SYMBOL(__skb_gso_segment); 3407 3408 /* Take action when hardware reception checksum errors are detected. */ 3409 #ifdef CONFIG_BUG 3410 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3411 { 3412 if (net_ratelimit()) { 3413 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>"); 3414 skb_dump(KERN_ERR, skb, true); 3415 dump_stack(); 3416 } 3417 } 3418 EXPORT_SYMBOL(netdev_rx_csum_fault); 3419 #endif 3420 3421 /* XXX: check that highmem exists at all on the given machine. */ 3422 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) 3423 { 3424 #ifdef CONFIG_HIGHMEM 3425 int i; 3426 3427 if (!(dev->features & NETIF_F_HIGHDMA)) { 3428 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 3429 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 3430 3431 if (PageHighMem(skb_frag_page(frag))) 3432 return 1; 3433 } 3434 } 3435 #endif 3436 return 0; 3437 } 3438 3439 /* If MPLS offload request, verify we are testing hardware MPLS features 3440 * instead of standard features for the netdev. 3441 */ 3442 #if IS_ENABLED(CONFIG_NET_MPLS_GSO) 3443 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3444 netdev_features_t features, 3445 __be16 type) 3446 { 3447 if (eth_p_mpls(type)) 3448 features &= skb->dev->mpls_features; 3449 3450 return features; 3451 } 3452 #else 3453 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3454 netdev_features_t features, 3455 __be16 type) 3456 { 3457 return features; 3458 } 3459 #endif 3460 3461 static netdev_features_t harmonize_features(struct sk_buff *skb, 3462 netdev_features_t features) 3463 { 3464 __be16 type; 3465 3466 type = skb_network_protocol(skb, NULL); 3467 features = net_mpls_features(skb, features, type); 3468 3469 if (skb->ip_summed != CHECKSUM_NONE && 3470 !can_checksum_protocol(features, type)) { 3471 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); 3472 } 3473 if (illegal_highdma(skb->dev, skb)) 3474 features &= ~NETIF_F_SG; 3475 3476 return features; 3477 } 3478 3479 netdev_features_t passthru_features_check(struct sk_buff *skb, 3480 struct net_device *dev, 3481 netdev_features_t features) 3482 { 3483 return features; 3484 } 3485 EXPORT_SYMBOL(passthru_features_check); 3486 3487 static netdev_features_t dflt_features_check(struct sk_buff *skb, 3488 struct net_device *dev, 3489 netdev_features_t features) 3490 { 3491 return vlan_features_check(skb, features); 3492 } 3493 3494 static netdev_features_t gso_features_check(const struct sk_buff *skb, 3495 struct net_device *dev, 3496 netdev_features_t features) 3497 { 3498 u16 gso_segs = skb_shinfo(skb)->gso_segs; 3499 3500 if (gso_segs > dev->gso_max_segs) 3501 return features & ~NETIF_F_GSO_MASK; 3502 3503 if (!skb_shinfo(skb)->gso_type) { 3504 skb_warn_bad_offload(skb); 3505 return features & ~NETIF_F_GSO_MASK; 3506 } 3507 3508 /* Support for GSO partial features requires software 3509 * intervention before we can actually process the packets 3510 * so we need to strip support for any partial features now 3511 * and we can pull them back in after we have partially 3512 * segmented the frame. 3513 */ 3514 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)) 3515 features &= ~dev->gso_partial_features; 3516 3517 /* Make sure to clear the IPv4 ID mangling feature if the 3518 * IPv4 header has the potential to be fragmented. 3519 */ 3520 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { 3521 struct iphdr *iph = skb->encapsulation ? 3522 inner_ip_hdr(skb) : ip_hdr(skb); 3523 3524 if (!(iph->frag_off & htons(IP_DF))) 3525 features &= ~NETIF_F_TSO_MANGLEID; 3526 } 3527 3528 return features; 3529 } 3530 3531 netdev_features_t netif_skb_features(struct sk_buff *skb) 3532 { 3533 struct net_device *dev = skb->dev; 3534 netdev_features_t features = dev->features; 3535 3536 if (skb_is_gso(skb)) 3537 features = gso_features_check(skb, dev, features); 3538 3539 /* If encapsulation offload request, verify we are testing 3540 * hardware encapsulation features instead of standard 3541 * features for the netdev 3542 */ 3543 if (skb->encapsulation) 3544 features &= dev->hw_enc_features; 3545 3546 if (skb_vlan_tagged(skb)) 3547 features = netdev_intersect_features(features, 3548 dev->vlan_features | 3549 NETIF_F_HW_VLAN_CTAG_TX | 3550 NETIF_F_HW_VLAN_STAG_TX); 3551 3552 if (dev->netdev_ops->ndo_features_check) 3553 features &= dev->netdev_ops->ndo_features_check(skb, dev, 3554 features); 3555 else 3556 features &= dflt_features_check(skb, dev, features); 3557 3558 return harmonize_features(skb, features); 3559 } 3560 EXPORT_SYMBOL(netif_skb_features); 3561 3562 static int xmit_one(struct sk_buff *skb, struct net_device *dev, 3563 struct netdev_queue *txq, bool more) 3564 { 3565 unsigned int len; 3566 int rc; 3567 3568 if (dev_nit_active(dev)) 3569 dev_queue_xmit_nit(skb, dev); 3570 3571 len = skb->len; 3572 PRANDOM_ADD_NOISE(skb, dev, txq, len + jiffies); 3573 trace_net_dev_start_xmit(skb, dev); 3574 rc = netdev_start_xmit(skb, dev, txq, more); 3575 trace_net_dev_xmit(skb, rc, dev, len); 3576 3577 return rc; 3578 } 3579 3580 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, 3581 struct netdev_queue *txq, int *ret) 3582 { 3583 struct sk_buff *skb = first; 3584 int rc = NETDEV_TX_OK; 3585 3586 while (skb) { 3587 struct sk_buff *next = skb->next; 3588 3589 skb_mark_not_on_list(skb); 3590 rc = xmit_one(skb, dev, txq, next != NULL); 3591 if (unlikely(!dev_xmit_complete(rc))) { 3592 skb->next = next; 3593 goto out; 3594 } 3595 3596 skb = next; 3597 if (netif_tx_queue_stopped(txq) && skb) { 3598 rc = NETDEV_TX_BUSY; 3599 break; 3600 } 3601 } 3602 3603 out: 3604 *ret = rc; 3605 return skb; 3606 } 3607 3608 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, 3609 netdev_features_t features) 3610 { 3611 if (skb_vlan_tag_present(skb) && 3612 !vlan_hw_offload_capable(features, skb->vlan_proto)) 3613 skb = __vlan_hwaccel_push_inside(skb); 3614 return skb; 3615 } 3616 3617 int skb_csum_hwoffload_help(struct sk_buff *skb, 3618 const netdev_features_t features) 3619 { 3620 if (unlikely(skb->csum_not_inet)) 3621 return !!(features & NETIF_F_SCTP_CRC) ? 0 : 3622 skb_crc32c_csum_help(skb); 3623 3624 return !!(features & NETIF_F_CSUM_MASK) ? 0 : skb_checksum_help(skb); 3625 } 3626 EXPORT_SYMBOL(skb_csum_hwoffload_help); 3627 3628 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again) 3629 { 3630 netdev_features_t features; 3631 3632 features = netif_skb_features(skb); 3633 skb = validate_xmit_vlan(skb, features); 3634 if (unlikely(!skb)) 3635 goto out_null; 3636 3637 skb = sk_validate_xmit_skb(skb, dev); 3638 if (unlikely(!skb)) 3639 goto out_null; 3640 3641 if (netif_needs_gso(skb, features)) { 3642 struct sk_buff *segs; 3643 3644 segs = skb_gso_segment(skb, features); 3645 if (IS_ERR(segs)) { 3646 goto out_kfree_skb; 3647 } else if (segs) { 3648 consume_skb(skb); 3649 skb = segs; 3650 } 3651 } else { 3652 if (skb_needs_linearize(skb, features) && 3653 __skb_linearize(skb)) 3654 goto out_kfree_skb; 3655 3656 /* If packet is not checksummed and device does not 3657 * support checksumming for this protocol, complete 3658 * checksumming here. 3659 */ 3660 if (skb->ip_summed == CHECKSUM_PARTIAL) { 3661 if (skb->encapsulation) 3662 skb_set_inner_transport_header(skb, 3663 skb_checksum_start_offset(skb)); 3664 else 3665 skb_set_transport_header(skb, 3666 skb_checksum_start_offset(skb)); 3667 if (skb_csum_hwoffload_help(skb, features)) 3668 goto out_kfree_skb; 3669 } 3670 } 3671 3672 skb = validate_xmit_xfrm(skb, features, again); 3673 3674 return skb; 3675 3676 out_kfree_skb: 3677 kfree_skb(skb); 3678 out_null: 3679 atomic_long_inc(&dev->tx_dropped); 3680 return NULL; 3681 } 3682 3683 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again) 3684 { 3685 struct sk_buff *next, *head = NULL, *tail; 3686 3687 for (; skb != NULL; skb = next) { 3688 next = skb->next; 3689 skb_mark_not_on_list(skb); 3690 3691 /* in case skb wont be segmented, point to itself */ 3692 skb->prev = skb; 3693 3694 skb = validate_xmit_skb(skb, dev, again); 3695 if (!skb) 3696 continue; 3697 3698 if (!head) 3699 head = skb; 3700 else 3701 tail->next = skb; 3702 /* If skb was segmented, skb->prev points to 3703 * the last segment. If not, it still contains skb. 3704 */ 3705 tail = skb->prev; 3706 } 3707 return head; 3708 } 3709 EXPORT_SYMBOL_GPL(validate_xmit_skb_list); 3710 3711 static void qdisc_pkt_len_init(struct sk_buff *skb) 3712 { 3713 const struct skb_shared_info *shinfo = skb_shinfo(skb); 3714 3715 qdisc_skb_cb(skb)->pkt_len = skb->len; 3716 3717 /* To get more precise estimation of bytes sent on wire, 3718 * we add to pkt_len the headers size of all segments 3719 */ 3720 if (shinfo->gso_size && skb_transport_header_was_set(skb)) { 3721 unsigned int hdr_len; 3722 u16 gso_segs = shinfo->gso_segs; 3723 3724 /* mac layer + network layer */ 3725 hdr_len = skb_transport_header(skb) - skb_mac_header(skb); 3726 3727 /* + transport layer */ 3728 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { 3729 const struct tcphdr *th; 3730 struct tcphdr _tcphdr; 3731 3732 th = skb_header_pointer(skb, skb_transport_offset(skb), 3733 sizeof(_tcphdr), &_tcphdr); 3734 if (likely(th)) 3735 hdr_len += __tcp_hdrlen(th); 3736 } else { 3737 struct udphdr _udphdr; 3738 3739 if (skb_header_pointer(skb, skb_transport_offset(skb), 3740 sizeof(_udphdr), &_udphdr)) 3741 hdr_len += sizeof(struct udphdr); 3742 } 3743 3744 if (shinfo->gso_type & SKB_GSO_DODGY) 3745 gso_segs = DIV_ROUND_UP(skb->len - hdr_len, 3746 shinfo->gso_size); 3747 3748 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; 3749 } 3750 } 3751 3752 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, 3753 struct net_device *dev, 3754 struct netdev_queue *txq) 3755 { 3756 spinlock_t *root_lock = qdisc_lock(q); 3757 struct sk_buff *to_free = NULL; 3758 bool contended; 3759 int rc; 3760 3761 qdisc_calculate_pkt_len(skb, q); 3762 3763 if (q->flags & TCQ_F_NOLOCK) { 3764 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK; 3765 qdisc_run(q); 3766 3767 if (unlikely(to_free)) 3768 kfree_skb_list(to_free); 3769 return rc; 3770 } 3771 3772 /* 3773 * Heuristic to force contended enqueues to serialize on a 3774 * separate lock before trying to get qdisc main lock. 3775 * This permits qdisc->running owner to get the lock more 3776 * often and dequeue packets faster. 3777 */ 3778 contended = qdisc_is_running(q); 3779 if (unlikely(contended)) 3780 spin_lock(&q->busylock); 3781 3782 spin_lock(root_lock); 3783 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { 3784 __qdisc_drop(skb, &to_free); 3785 rc = NET_XMIT_DROP; 3786 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && 3787 qdisc_run_begin(q)) { 3788 /* 3789 * This is a work-conserving queue; there are no old skbs 3790 * waiting to be sent out; and the qdisc is not running - 3791 * xmit the skb directly. 3792 */ 3793 3794 qdisc_bstats_update(q, skb); 3795 3796 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { 3797 if (unlikely(contended)) { 3798 spin_unlock(&q->busylock); 3799 contended = false; 3800 } 3801 __qdisc_run(q); 3802 } 3803 3804 qdisc_run_end(q); 3805 rc = NET_XMIT_SUCCESS; 3806 } else { 3807 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK; 3808 if (qdisc_run_begin(q)) { 3809 if (unlikely(contended)) { 3810 spin_unlock(&q->busylock); 3811 contended = false; 3812 } 3813 __qdisc_run(q); 3814 qdisc_run_end(q); 3815 } 3816 } 3817 spin_unlock(root_lock); 3818 if (unlikely(to_free)) 3819 kfree_skb_list(to_free); 3820 if (unlikely(contended)) 3821 spin_unlock(&q->busylock); 3822 return rc; 3823 } 3824 3825 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) 3826 static void skb_update_prio(struct sk_buff *skb) 3827 { 3828 const struct netprio_map *map; 3829 const struct sock *sk; 3830 unsigned int prioidx; 3831 3832 if (skb->priority) 3833 return; 3834 map = rcu_dereference_bh(skb->dev->priomap); 3835 if (!map) 3836 return; 3837 sk = skb_to_full_sk(skb); 3838 if (!sk) 3839 return; 3840 3841 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data); 3842 3843 if (prioidx < map->priomap_len) 3844 skb->priority = map->priomap[prioidx]; 3845 } 3846 #else 3847 #define skb_update_prio(skb) 3848 #endif 3849 3850 /** 3851 * dev_loopback_xmit - loop back @skb 3852 * @net: network namespace this loopback is happening in 3853 * @sk: sk needed to be a netfilter okfn 3854 * @skb: buffer to transmit 3855 */ 3856 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) 3857 { 3858 skb_reset_mac_header(skb); 3859 __skb_pull(skb, skb_network_offset(skb)); 3860 skb->pkt_type = PACKET_LOOPBACK; 3861 skb->ip_summed = CHECKSUM_UNNECESSARY; 3862 WARN_ON(!skb_dst(skb)); 3863 skb_dst_force(skb); 3864 netif_rx_ni(skb); 3865 return 0; 3866 } 3867 EXPORT_SYMBOL(dev_loopback_xmit); 3868 3869 #ifdef CONFIG_NET_EGRESS 3870 static struct sk_buff * 3871 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 3872 { 3873 struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress); 3874 struct tcf_result cl_res; 3875 3876 if (!miniq) 3877 return skb; 3878 3879 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */ 3880 qdisc_skb_cb(skb)->mru = 0; 3881 mini_qdisc_bstats_cpu_update(miniq, skb); 3882 3883 switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) { 3884 case TC_ACT_OK: 3885 case TC_ACT_RECLASSIFY: 3886 skb->tc_index = TC_H_MIN(cl_res.classid); 3887 break; 3888 case TC_ACT_SHOT: 3889 mini_qdisc_qstats_cpu_drop(miniq); 3890 *ret = NET_XMIT_DROP; 3891 kfree_skb(skb); 3892 return NULL; 3893 case TC_ACT_STOLEN: 3894 case TC_ACT_QUEUED: 3895 case TC_ACT_TRAP: 3896 *ret = NET_XMIT_SUCCESS; 3897 consume_skb(skb); 3898 return NULL; 3899 case TC_ACT_REDIRECT: 3900 /* No need to push/pop skb's mac_header here on egress! */ 3901 skb_do_redirect(skb); 3902 *ret = NET_XMIT_SUCCESS; 3903 return NULL; 3904 default: 3905 break; 3906 } 3907 3908 return skb; 3909 } 3910 #endif /* CONFIG_NET_EGRESS */ 3911 3912 #ifdef CONFIG_XPS 3913 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb, 3914 struct xps_dev_maps *dev_maps, unsigned int tci) 3915 { 3916 struct xps_map *map; 3917 int queue_index = -1; 3918 3919 if (dev->num_tc) { 3920 tci *= dev->num_tc; 3921 tci += netdev_get_prio_tc_map(dev, skb->priority); 3922 } 3923 3924 map = rcu_dereference(dev_maps->attr_map[tci]); 3925 if (map) { 3926 if (map->len == 1) 3927 queue_index = map->queues[0]; 3928 else 3929 queue_index = map->queues[reciprocal_scale( 3930 skb_get_hash(skb), map->len)]; 3931 if (unlikely(queue_index >= dev->real_num_tx_queues)) 3932 queue_index = -1; 3933 } 3934 return queue_index; 3935 } 3936 #endif 3937 3938 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev, 3939 struct sk_buff *skb) 3940 { 3941 #ifdef CONFIG_XPS 3942 struct xps_dev_maps *dev_maps; 3943 struct sock *sk = skb->sk; 3944 int queue_index = -1; 3945 3946 if (!static_key_false(&xps_needed)) 3947 return -1; 3948 3949 rcu_read_lock(); 3950 if (!static_key_false(&xps_rxqs_needed)) 3951 goto get_cpus_map; 3952 3953 dev_maps = rcu_dereference(sb_dev->xps_rxqs_map); 3954 if (dev_maps) { 3955 int tci = sk_rx_queue_get(sk); 3956 3957 if (tci >= 0 && tci < dev->num_rx_queues) 3958 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 3959 tci); 3960 } 3961 3962 get_cpus_map: 3963 if (queue_index < 0) { 3964 dev_maps = rcu_dereference(sb_dev->xps_cpus_map); 3965 if (dev_maps) { 3966 unsigned int tci = skb->sender_cpu - 1; 3967 3968 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 3969 tci); 3970 } 3971 } 3972 rcu_read_unlock(); 3973 3974 return queue_index; 3975 #else 3976 return -1; 3977 #endif 3978 } 3979 3980 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb, 3981 struct net_device *sb_dev) 3982 { 3983 return 0; 3984 } 3985 EXPORT_SYMBOL(dev_pick_tx_zero); 3986 3987 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb, 3988 struct net_device *sb_dev) 3989 { 3990 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues; 3991 } 3992 EXPORT_SYMBOL(dev_pick_tx_cpu_id); 3993 3994 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb, 3995 struct net_device *sb_dev) 3996 { 3997 struct sock *sk = skb->sk; 3998 int queue_index = sk_tx_queue_get(sk); 3999 4000 sb_dev = sb_dev ? : dev; 4001 4002 if (queue_index < 0 || skb->ooo_okay || 4003 queue_index >= dev->real_num_tx_queues) { 4004 int new_index = get_xps_queue(dev, sb_dev, skb); 4005 4006 if (new_index < 0) 4007 new_index = skb_tx_hash(dev, sb_dev, skb); 4008 4009 if (queue_index != new_index && sk && 4010 sk_fullsock(sk) && 4011 rcu_access_pointer(sk->sk_dst_cache)) 4012 sk_tx_queue_set(sk, new_index); 4013 4014 queue_index = new_index; 4015 } 4016 4017 return queue_index; 4018 } 4019 EXPORT_SYMBOL(netdev_pick_tx); 4020 4021 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev, 4022 struct sk_buff *skb, 4023 struct net_device *sb_dev) 4024 { 4025 int queue_index = 0; 4026 4027 #ifdef CONFIG_XPS 4028 u32 sender_cpu = skb->sender_cpu - 1; 4029 4030 if (sender_cpu >= (u32)NR_CPUS) 4031 skb->sender_cpu = raw_smp_processor_id() + 1; 4032 #endif 4033 4034 if (dev->real_num_tx_queues != 1) { 4035 const struct net_device_ops *ops = dev->netdev_ops; 4036 4037 if (ops->ndo_select_queue) 4038 queue_index = ops->ndo_select_queue(dev, skb, sb_dev); 4039 else 4040 queue_index = netdev_pick_tx(dev, skb, sb_dev); 4041 4042 queue_index = netdev_cap_txqueue(dev, queue_index); 4043 } 4044 4045 skb_set_queue_mapping(skb, queue_index); 4046 return netdev_get_tx_queue(dev, queue_index); 4047 } 4048 4049 /** 4050 * __dev_queue_xmit - transmit a buffer 4051 * @skb: buffer to transmit 4052 * @sb_dev: suboordinate device used for L2 forwarding offload 4053 * 4054 * Queue a buffer for transmission to a network device. The caller must 4055 * have set the device and priority and built the buffer before calling 4056 * this function. The function can be called from an interrupt. 4057 * 4058 * A negative errno code is returned on a failure. A success does not 4059 * guarantee the frame will be transmitted as it may be dropped due 4060 * to congestion or traffic shaping. 4061 * 4062 * ----------------------------------------------------------------------------------- 4063 * I notice this method can also return errors from the queue disciplines, 4064 * including NET_XMIT_DROP, which is a positive value. So, errors can also 4065 * be positive. 4066 * 4067 * Regardless of the return value, the skb is consumed, so it is currently 4068 * difficult to retry a send to this method. (You can bump the ref count 4069 * before sending to hold a reference for retry if you are careful.) 4070 * 4071 * When calling this method, interrupts MUST be enabled. This is because 4072 * the BH enable code must have IRQs enabled so that it will not deadlock. 4073 * --BLG 4074 */ 4075 static int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev) 4076 { 4077 struct net_device *dev = skb->dev; 4078 struct netdev_queue *txq; 4079 struct Qdisc *q; 4080 int rc = -ENOMEM; 4081 bool again = false; 4082 4083 skb_reset_mac_header(skb); 4084 4085 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) 4086 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED); 4087 4088 /* Disable soft irqs for various locks below. Also 4089 * stops preemption for RCU. 4090 */ 4091 rcu_read_lock_bh(); 4092 4093 skb_update_prio(skb); 4094 4095 qdisc_pkt_len_init(skb); 4096 #ifdef CONFIG_NET_CLS_ACT 4097 skb->tc_at_ingress = 0; 4098 # ifdef CONFIG_NET_EGRESS 4099 if (static_branch_unlikely(&egress_needed_key)) { 4100 skb = sch_handle_egress(skb, &rc, dev); 4101 if (!skb) 4102 goto out; 4103 } 4104 # endif 4105 #endif 4106 /* If device/qdisc don't need skb->dst, release it right now while 4107 * its hot in this cpu cache. 4108 */ 4109 if (dev->priv_flags & IFF_XMIT_DST_RELEASE) 4110 skb_dst_drop(skb); 4111 else 4112 skb_dst_force(skb); 4113 4114 txq = netdev_core_pick_tx(dev, skb, sb_dev); 4115 q = rcu_dereference_bh(txq->qdisc); 4116 4117 trace_net_dev_queue(skb); 4118 if (q->enqueue) { 4119 rc = __dev_xmit_skb(skb, q, dev, txq); 4120 goto out; 4121 } 4122 4123 /* The device has no queue. Common case for software devices: 4124 * loopback, all the sorts of tunnels... 4125 4126 * Really, it is unlikely that netif_tx_lock protection is necessary 4127 * here. (f.e. loopback and IP tunnels are clean ignoring statistics 4128 * counters.) 4129 * However, it is possible, that they rely on protection 4130 * made by us here. 4131 4132 * Check this and shot the lock. It is not prone from deadlocks. 4133 *Either shot noqueue qdisc, it is even simpler 8) 4134 */ 4135 if (dev->flags & IFF_UP) { 4136 int cpu = smp_processor_id(); /* ok because BHs are off */ 4137 4138 if (txq->xmit_lock_owner != cpu) { 4139 if (dev_xmit_recursion()) 4140 goto recursion_alert; 4141 4142 skb = validate_xmit_skb(skb, dev, &again); 4143 if (!skb) 4144 goto out; 4145 4146 PRANDOM_ADD_NOISE(skb, dev, txq, jiffies); 4147 HARD_TX_LOCK(dev, txq, cpu); 4148 4149 if (!netif_xmit_stopped(txq)) { 4150 dev_xmit_recursion_inc(); 4151 skb = dev_hard_start_xmit(skb, dev, txq, &rc); 4152 dev_xmit_recursion_dec(); 4153 if (dev_xmit_complete(rc)) { 4154 HARD_TX_UNLOCK(dev, txq); 4155 goto out; 4156 } 4157 } 4158 HARD_TX_UNLOCK(dev, txq); 4159 net_crit_ratelimited("Virtual device %s asks to queue packet!\n", 4160 dev->name); 4161 } else { 4162 /* Recursion is detected! It is possible, 4163 * unfortunately 4164 */ 4165 recursion_alert: 4166 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", 4167 dev->name); 4168 } 4169 } 4170 4171 rc = -ENETDOWN; 4172 rcu_read_unlock_bh(); 4173 4174 atomic_long_inc(&dev->tx_dropped); 4175 kfree_skb_list(skb); 4176 return rc; 4177 out: 4178 rcu_read_unlock_bh(); 4179 return rc; 4180 } 4181 4182 int dev_queue_xmit(struct sk_buff *skb) 4183 { 4184 return __dev_queue_xmit(skb, NULL); 4185 } 4186 EXPORT_SYMBOL(dev_queue_xmit); 4187 4188 int dev_queue_xmit_accel(struct sk_buff *skb, struct net_device *sb_dev) 4189 { 4190 return __dev_queue_xmit(skb, sb_dev); 4191 } 4192 EXPORT_SYMBOL(dev_queue_xmit_accel); 4193 4194 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id) 4195 { 4196 struct net_device *dev = skb->dev; 4197 struct sk_buff *orig_skb = skb; 4198 struct netdev_queue *txq; 4199 int ret = NETDEV_TX_BUSY; 4200 bool again = false; 4201 4202 if (unlikely(!netif_running(dev) || 4203 !netif_carrier_ok(dev))) 4204 goto drop; 4205 4206 skb = validate_xmit_skb_list(skb, dev, &again); 4207 if (skb != orig_skb) 4208 goto drop; 4209 4210 skb_set_queue_mapping(skb, queue_id); 4211 txq = skb_get_tx_queue(dev, skb); 4212 PRANDOM_ADD_NOISE(skb, dev, txq, jiffies); 4213 4214 local_bh_disable(); 4215 4216 dev_xmit_recursion_inc(); 4217 HARD_TX_LOCK(dev, txq, smp_processor_id()); 4218 if (!netif_xmit_frozen_or_drv_stopped(txq)) 4219 ret = netdev_start_xmit(skb, dev, txq, false); 4220 HARD_TX_UNLOCK(dev, txq); 4221 dev_xmit_recursion_dec(); 4222 4223 local_bh_enable(); 4224 return ret; 4225 drop: 4226 atomic_long_inc(&dev->tx_dropped); 4227 kfree_skb_list(skb); 4228 return NET_XMIT_DROP; 4229 } 4230 EXPORT_SYMBOL(__dev_direct_xmit); 4231 4232 /************************************************************************* 4233 * Receiver routines 4234 *************************************************************************/ 4235 4236 int netdev_max_backlog __read_mostly = 1000; 4237 EXPORT_SYMBOL(netdev_max_backlog); 4238 4239 int netdev_tstamp_prequeue __read_mostly = 1; 4240 int netdev_budget __read_mostly = 300; 4241 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */ 4242 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ; 4243 int weight_p __read_mostly = 64; /* old backlog weight */ 4244 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ 4245 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ 4246 int dev_rx_weight __read_mostly = 64; 4247 int dev_tx_weight __read_mostly = 64; 4248 /* Maximum number of GRO_NORMAL skbs to batch up for list-RX */ 4249 int gro_normal_batch __read_mostly = 8; 4250 4251 /* Called with irq disabled */ 4252 static inline void ____napi_schedule(struct softnet_data *sd, 4253 struct napi_struct *napi) 4254 { 4255 list_add_tail(&napi->poll_list, &sd->poll_list); 4256 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4257 } 4258 4259 #ifdef CONFIG_RPS 4260 4261 /* One global table that all flow-based protocols share. */ 4262 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; 4263 EXPORT_SYMBOL(rps_sock_flow_table); 4264 u32 rps_cpu_mask __read_mostly; 4265 EXPORT_SYMBOL(rps_cpu_mask); 4266 4267 struct static_key_false rps_needed __read_mostly; 4268 EXPORT_SYMBOL(rps_needed); 4269 struct static_key_false rfs_needed __read_mostly; 4270 EXPORT_SYMBOL(rfs_needed); 4271 4272 static struct rps_dev_flow * 4273 set_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4274 struct rps_dev_flow *rflow, u16 next_cpu) 4275 { 4276 if (next_cpu < nr_cpu_ids) { 4277 #ifdef CONFIG_RFS_ACCEL 4278 struct netdev_rx_queue *rxqueue; 4279 struct rps_dev_flow_table *flow_table; 4280 struct rps_dev_flow *old_rflow; 4281 u32 flow_id; 4282 u16 rxq_index; 4283 int rc; 4284 4285 /* Should we steer this flow to a different hardware queue? */ 4286 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || 4287 !(dev->features & NETIF_F_NTUPLE)) 4288 goto out; 4289 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); 4290 if (rxq_index == skb_get_rx_queue(skb)) 4291 goto out; 4292 4293 rxqueue = dev->_rx + rxq_index; 4294 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4295 if (!flow_table) 4296 goto out; 4297 flow_id = skb_get_hash(skb) & flow_table->mask; 4298 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, 4299 rxq_index, flow_id); 4300 if (rc < 0) 4301 goto out; 4302 old_rflow = rflow; 4303 rflow = &flow_table->flows[flow_id]; 4304 rflow->filter = rc; 4305 if (old_rflow->filter == rflow->filter) 4306 old_rflow->filter = RPS_NO_FILTER; 4307 out: 4308 #endif 4309 rflow->last_qtail = 4310 per_cpu(softnet_data, next_cpu).input_queue_head; 4311 } 4312 4313 rflow->cpu = next_cpu; 4314 return rflow; 4315 } 4316 4317 /* 4318 * get_rps_cpu is called from netif_receive_skb and returns the target 4319 * CPU from the RPS map of the receiving queue for a given skb. 4320 * rcu_read_lock must be held on entry. 4321 */ 4322 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4323 struct rps_dev_flow **rflowp) 4324 { 4325 const struct rps_sock_flow_table *sock_flow_table; 4326 struct netdev_rx_queue *rxqueue = dev->_rx; 4327 struct rps_dev_flow_table *flow_table; 4328 struct rps_map *map; 4329 int cpu = -1; 4330 u32 tcpu; 4331 u32 hash; 4332 4333 if (skb_rx_queue_recorded(skb)) { 4334 u16 index = skb_get_rx_queue(skb); 4335 4336 if (unlikely(index >= dev->real_num_rx_queues)) { 4337 WARN_ONCE(dev->real_num_rx_queues > 1, 4338 "%s received packet on queue %u, but number " 4339 "of RX queues is %u\n", 4340 dev->name, index, dev->real_num_rx_queues); 4341 goto done; 4342 } 4343 rxqueue += index; 4344 } 4345 4346 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ 4347 4348 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4349 map = rcu_dereference(rxqueue->rps_map); 4350 if (!flow_table && !map) 4351 goto done; 4352 4353 skb_reset_network_header(skb); 4354 hash = skb_get_hash(skb); 4355 if (!hash) 4356 goto done; 4357 4358 sock_flow_table = rcu_dereference(rps_sock_flow_table); 4359 if (flow_table && sock_flow_table) { 4360 struct rps_dev_flow *rflow; 4361 u32 next_cpu; 4362 u32 ident; 4363 4364 /* First check into global flow table if there is a match */ 4365 ident = sock_flow_table->ents[hash & sock_flow_table->mask]; 4366 if ((ident ^ hash) & ~rps_cpu_mask) 4367 goto try_rps; 4368 4369 next_cpu = ident & rps_cpu_mask; 4370 4371 /* OK, now we know there is a match, 4372 * we can look at the local (per receive queue) flow table 4373 */ 4374 rflow = &flow_table->flows[hash & flow_table->mask]; 4375 tcpu = rflow->cpu; 4376 4377 /* 4378 * If the desired CPU (where last recvmsg was done) is 4379 * different from current CPU (one in the rx-queue flow 4380 * table entry), switch if one of the following holds: 4381 * - Current CPU is unset (>= nr_cpu_ids). 4382 * - Current CPU is offline. 4383 * - The current CPU's queue tail has advanced beyond the 4384 * last packet that was enqueued using this table entry. 4385 * This guarantees that all previous packets for the flow 4386 * have been dequeued, thus preserving in order delivery. 4387 */ 4388 if (unlikely(tcpu != next_cpu) && 4389 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || 4390 ((int)(per_cpu(softnet_data, tcpu).input_queue_head - 4391 rflow->last_qtail)) >= 0)) { 4392 tcpu = next_cpu; 4393 rflow = set_rps_cpu(dev, skb, rflow, next_cpu); 4394 } 4395 4396 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { 4397 *rflowp = rflow; 4398 cpu = tcpu; 4399 goto done; 4400 } 4401 } 4402 4403 try_rps: 4404 4405 if (map) { 4406 tcpu = map->cpus[reciprocal_scale(hash, map->len)]; 4407 if (cpu_online(tcpu)) { 4408 cpu = tcpu; 4409 goto done; 4410 } 4411 } 4412 4413 done: 4414 return cpu; 4415 } 4416 4417 #ifdef CONFIG_RFS_ACCEL 4418 4419 /** 4420 * rps_may_expire_flow - check whether an RFS hardware filter may be removed 4421 * @dev: Device on which the filter was set 4422 * @rxq_index: RX queue index 4423 * @flow_id: Flow ID passed to ndo_rx_flow_steer() 4424 * @filter_id: Filter ID returned by ndo_rx_flow_steer() 4425 * 4426 * Drivers that implement ndo_rx_flow_steer() should periodically call 4427 * this function for each installed filter and remove the filters for 4428 * which it returns %true. 4429 */ 4430 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, 4431 u32 flow_id, u16 filter_id) 4432 { 4433 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; 4434 struct rps_dev_flow_table *flow_table; 4435 struct rps_dev_flow *rflow; 4436 bool expire = true; 4437 unsigned int cpu; 4438 4439 rcu_read_lock(); 4440 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4441 if (flow_table && flow_id <= flow_table->mask) { 4442 rflow = &flow_table->flows[flow_id]; 4443 cpu = READ_ONCE(rflow->cpu); 4444 if (rflow->filter == filter_id && cpu < nr_cpu_ids && 4445 ((int)(per_cpu(softnet_data, cpu).input_queue_head - 4446 rflow->last_qtail) < 4447 (int)(10 * flow_table->mask))) 4448 expire = false; 4449 } 4450 rcu_read_unlock(); 4451 return expire; 4452 } 4453 EXPORT_SYMBOL(rps_may_expire_flow); 4454 4455 #endif /* CONFIG_RFS_ACCEL */ 4456 4457 /* Called from hardirq (IPI) context */ 4458 static void rps_trigger_softirq(void *data) 4459 { 4460 struct softnet_data *sd = data; 4461 4462 ____napi_schedule(sd, &sd->backlog); 4463 sd->received_rps++; 4464 } 4465 4466 #endif /* CONFIG_RPS */ 4467 4468 /* 4469 * Check if this softnet_data structure is another cpu one 4470 * If yes, queue it to our IPI list and return 1 4471 * If no, return 0 4472 */ 4473 static int rps_ipi_queued(struct softnet_data *sd) 4474 { 4475 #ifdef CONFIG_RPS 4476 struct softnet_data *mysd = this_cpu_ptr(&softnet_data); 4477 4478 if (sd != mysd) { 4479 sd->rps_ipi_next = mysd->rps_ipi_list; 4480 mysd->rps_ipi_list = sd; 4481 4482 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4483 return 1; 4484 } 4485 #endif /* CONFIG_RPS */ 4486 return 0; 4487 } 4488 4489 #ifdef CONFIG_NET_FLOW_LIMIT 4490 int netdev_flow_limit_table_len __read_mostly = (1 << 12); 4491 #endif 4492 4493 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) 4494 { 4495 #ifdef CONFIG_NET_FLOW_LIMIT 4496 struct sd_flow_limit *fl; 4497 struct softnet_data *sd; 4498 unsigned int old_flow, new_flow; 4499 4500 if (qlen < (netdev_max_backlog >> 1)) 4501 return false; 4502 4503 sd = this_cpu_ptr(&softnet_data); 4504 4505 rcu_read_lock(); 4506 fl = rcu_dereference(sd->flow_limit); 4507 if (fl) { 4508 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); 4509 old_flow = fl->history[fl->history_head]; 4510 fl->history[fl->history_head] = new_flow; 4511 4512 fl->history_head++; 4513 fl->history_head &= FLOW_LIMIT_HISTORY - 1; 4514 4515 if (likely(fl->buckets[old_flow])) 4516 fl->buckets[old_flow]--; 4517 4518 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { 4519 fl->count++; 4520 rcu_read_unlock(); 4521 return true; 4522 } 4523 } 4524 rcu_read_unlock(); 4525 #endif 4526 return false; 4527 } 4528 4529 /* 4530 * enqueue_to_backlog is called to queue an skb to a per CPU backlog 4531 * queue (may be a remote CPU queue). 4532 */ 4533 static int enqueue_to_backlog(struct sk_buff *skb, int cpu, 4534 unsigned int *qtail) 4535 { 4536 struct softnet_data *sd; 4537 unsigned long flags; 4538 unsigned int qlen; 4539 4540 sd = &per_cpu(softnet_data, cpu); 4541 4542 local_irq_save(flags); 4543 4544 rps_lock(sd); 4545 if (!netif_running(skb->dev)) 4546 goto drop; 4547 qlen = skb_queue_len(&sd->input_pkt_queue); 4548 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) { 4549 if (qlen) { 4550 enqueue: 4551 __skb_queue_tail(&sd->input_pkt_queue, skb); 4552 input_queue_tail_incr_save(sd, qtail); 4553 rps_unlock(sd); 4554 local_irq_restore(flags); 4555 return NET_RX_SUCCESS; 4556 } 4557 4558 /* Schedule NAPI for backlog device 4559 * We can use non atomic operation since we own the queue lock 4560 */ 4561 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) { 4562 if (!rps_ipi_queued(sd)) 4563 ____napi_schedule(sd, &sd->backlog); 4564 } 4565 goto enqueue; 4566 } 4567 4568 drop: 4569 sd->dropped++; 4570 rps_unlock(sd); 4571 4572 local_irq_restore(flags); 4573 4574 atomic_long_inc(&skb->dev->rx_dropped); 4575 kfree_skb(skb); 4576 return NET_RX_DROP; 4577 } 4578 4579 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb) 4580 { 4581 struct net_device *dev = skb->dev; 4582 struct netdev_rx_queue *rxqueue; 4583 4584 rxqueue = dev->_rx; 4585 4586 if (skb_rx_queue_recorded(skb)) { 4587 u16 index = skb_get_rx_queue(skb); 4588 4589 if (unlikely(index >= dev->real_num_rx_queues)) { 4590 WARN_ONCE(dev->real_num_rx_queues > 1, 4591 "%s received packet on queue %u, but number " 4592 "of RX queues is %u\n", 4593 dev->name, index, dev->real_num_rx_queues); 4594 4595 return rxqueue; /* Return first rxqueue */ 4596 } 4597 rxqueue += index; 4598 } 4599 return rxqueue; 4600 } 4601 4602 static u32 netif_receive_generic_xdp(struct sk_buff *skb, 4603 struct xdp_buff *xdp, 4604 struct bpf_prog *xdp_prog) 4605 { 4606 struct netdev_rx_queue *rxqueue; 4607 void *orig_data, *orig_data_end; 4608 u32 metalen, act = XDP_DROP; 4609 __be16 orig_eth_type; 4610 struct ethhdr *eth; 4611 bool orig_bcast; 4612 int hlen, off; 4613 u32 mac_len; 4614 4615 /* Reinjected packets coming from act_mirred or similar should 4616 * not get XDP generic processing. 4617 */ 4618 if (skb_is_redirected(skb)) 4619 return XDP_PASS; 4620 4621 /* XDP packets must be linear and must have sufficient headroom 4622 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also 4623 * native XDP provides, thus we need to do it here as well. 4624 */ 4625 if (skb_cloned(skb) || skb_is_nonlinear(skb) || 4626 skb_headroom(skb) < XDP_PACKET_HEADROOM) { 4627 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb); 4628 int troom = skb->tail + skb->data_len - skb->end; 4629 4630 /* In case we have to go down the path and also linearize, 4631 * then lets do the pskb_expand_head() work just once here. 4632 */ 4633 if (pskb_expand_head(skb, 4634 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0, 4635 troom > 0 ? troom + 128 : 0, GFP_ATOMIC)) 4636 goto do_drop; 4637 if (skb_linearize(skb)) 4638 goto do_drop; 4639 } 4640 4641 /* The XDP program wants to see the packet starting at the MAC 4642 * header. 4643 */ 4644 mac_len = skb->data - skb_mac_header(skb); 4645 hlen = skb_headlen(skb) + mac_len; 4646 xdp->data = skb->data - mac_len; 4647 xdp->data_meta = xdp->data; 4648 xdp->data_end = xdp->data + hlen; 4649 xdp->data_hard_start = skb->data - skb_headroom(skb); 4650 4651 /* SKB "head" area always have tailroom for skb_shared_info */ 4652 xdp->frame_sz = (void *)skb_end_pointer(skb) - xdp->data_hard_start; 4653 xdp->frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 4654 4655 orig_data_end = xdp->data_end; 4656 orig_data = xdp->data; 4657 eth = (struct ethhdr *)xdp->data; 4658 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest); 4659 orig_eth_type = eth->h_proto; 4660 4661 rxqueue = netif_get_rxqueue(skb); 4662 xdp->rxq = &rxqueue->xdp_rxq; 4663 4664 act = bpf_prog_run_xdp(xdp_prog, xdp); 4665 4666 /* check if bpf_xdp_adjust_head was used */ 4667 off = xdp->data - orig_data; 4668 if (off) { 4669 if (off > 0) 4670 __skb_pull(skb, off); 4671 else if (off < 0) 4672 __skb_push(skb, -off); 4673 4674 skb->mac_header += off; 4675 skb_reset_network_header(skb); 4676 } 4677 4678 /* check if bpf_xdp_adjust_tail was used */ 4679 off = xdp->data_end - orig_data_end; 4680 if (off != 0) { 4681 skb_set_tail_pointer(skb, xdp->data_end - xdp->data); 4682 skb->len += off; /* positive on grow, negative on shrink */ 4683 } 4684 4685 /* check if XDP changed eth hdr such SKB needs update */ 4686 eth = (struct ethhdr *)xdp->data; 4687 if ((orig_eth_type != eth->h_proto) || 4688 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) { 4689 __skb_push(skb, ETH_HLEN); 4690 skb->protocol = eth_type_trans(skb, skb->dev); 4691 } 4692 4693 switch (act) { 4694 case XDP_REDIRECT: 4695 case XDP_TX: 4696 __skb_push(skb, mac_len); 4697 break; 4698 case XDP_PASS: 4699 metalen = xdp->data - xdp->data_meta; 4700 if (metalen) 4701 skb_metadata_set(skb, metalen); 4702 break; 4703 default: 4704 bpf_warn_invalid_xdp_action(act); 4705 fallthrough; 4706 case XDP_ABORTED: 4707 trace_xdp_exception(skb->dev, xdp_prog, act); 4708 fallthrough; 4709 case XDP_DROP: 4710 do_drop: 4711 kfree_skb(skb); 4712 break; 4713 } 4714 4715 return act; 4716 } 4717 4718 /* When doing generic XDP we have to bypass the qdisc layer and the 4719 * network taps in order to match in-driver-XDP behavior. 4720 */ 4721 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog) 4722 { 4723 struct net_device *dev = skb->dev; 4724 struct netdev_queue *txq; 4725 bool free_skb = true; 4726 int cpu, rc; 4727 4728 txq = netdev_core_pick_tx(dev, skb, NULL); 4729 cpu = smp_processor_id(); 4730 HARD_TX_LOCK(dev, txq, cpu); 4731 if (!netif_xmit_stopped(txq)) { 4732 rc = netdev_start_xmit(skb, dev, txq, 0); 4733 if (dev_xmit_complete(rc)) 4734 free_skb = false; 4735 } 4736 HARD_TX_UNLOCK(dev, txq); 4737 if (free_skb) { 4738 trace_xdp_exception(dev, xdp_prog, XDP_TX); 4739 kfree_skb(skb); 4740 } 4741 } 4742 4743 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key); 4744 4745 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb) 4746 { 4747 if (xdp_prog) { 4748 struct xdp_buff xdp; 4749 u32 act; 4750 int err; 4751 4752 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog); 4753 if (act != XDP_PASS) { 4754 switch (act) { 4755 case XDP_REDIRECT: 4756 err = xdp_do_generic_redirect(skb->dev, skb, 4757 &xdp, xdp_prog); 4758 if (err) 4759 goto out_redir; 4760 break; 4761 case XDP_TX: 4762 generic_xdp_tx(skb, xdp_prog); 4763 break; 4764 } 4765 return XDP_DROP; 4766 } 4767 } 4768 return XDP_PASS; 4769 out_redir: 4770 kfree_skb(skb); 4771 return XDP_DROP; 4772 } 4773 EXPORT_SYMBOL_GPL(do_xdp_generic); 4774 4775 static int netif_rx_internal(struct sk_buff *skb) 4776 { 4777 int ret; 4778 4779 net_timestamp_check(netdev_tstamp_prequeue, skb); 4780 4781 trace_netif_rx(skb); 4782 4783 #ifdef CONFIG_RPS 4784 if (static_branch_unlikely(&rps_needed)) { 4785 struct rps_dev_flow voidflow, *rflow = &voidflow; 4786 int cpu; 4787 4788 preempt_disable(); 4789 rcu_read_lock(); 4790 4791 cpu = get_rps_cpu(skb->dev, skb, &rflow); 4792 if (cpu < 0) 4793 cpu = smp_processor_id(); 4794 4795 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 4796 4797 rcu_read_unlock(); 4798 preempt_enable(); 4799 } else 4800 #endif 4801 { 4802 unsigned int qtail; 4803 4804 ret = enqueue_to_backlog(skb, get_cpu(), &qtail); 4805 put_cpu(); 4806 } 4807 return ret; 4808 } 4809 4810 /** 4811 * netif_rx - post buffer to the network code 4812 * @skb: buffer to post 4813 * 4814 * This function receives a packet from a device driver and queues it for 4815 * the upper (protocol) levels to process. It always succeeds. The buffer 4816 * may be dropped during processing for congestion control or by the 4817 * protocol layers. 4818 * 4819 * return values: 4820 * NET_RX_SUCCESS (no congestion) 4821 * NET_RX_DROP (packet was dropped) 4822 * 4823 */ 4824 4825 int netif_rx(struct sk_buff *skb) 4826 { 4827 int ret; 4828 4829 trace_netif_rx_entry(skb); 4830 4831 ret = netif_rx_internal(skb); 4832 trace_netif_rx_exit(ret); 4833 4834 return ret; 4835 } 4836 EXPORT_SYMBOL(netif_rx); 4837 4838 int netif_rx_ni(struct sk_buff *skb) 4839 { 4840 int err; 4841 4842 trace_netif_rx_ni_entry(skb); 4843 4844 preempt_disable(); 4845 err = netif_rx_internal(skb); 4846 if (local_softirq_pending()) 4847 do_softirq(); 4848 preempt_enable(); 4849 trace_netif_rx_ni_exit(err); 4850 4851 return err; 4852 } 4853 EXPORT_SYMBOL(netif_rx_ni); 4854 4855 int netif_rx_any_context(struct sk_buff *skb) 4856 { 4857 /* 4858 * If invoked from contexts which do not invoke bottom half 4859 * processing either at return from interrupt or when softrqs are 4860 * reenabled, use netif_rx_ni() which invokes bottomhalf processing 4861 * directly. 4862 */ 4863 if (in_interrupt()) 4864 return netif_rx(skb); 4865 else 4866 return netif_rx_ni(skb); 4867 } 4868 EXPORT_SYMBOL(netif_rx_any_context); 4869 4870 static __latent_entropy void net_tx_action(struct softirq_action *h) 4871 { 4872 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 4873 4874 if (sd->completion_queue) { 4875 struct sk_buff *clist; 4876 4877 local_irq_disable(); 4878 clist = sd->completion_queue; 4879 sd->completion_queue = NULL; 4880 local_irq_enable(); 4881 4882 while (clist) { 4883 struct sk_buff *skb = clist; 4884 4885 clist = clist->next; 4886 4887 WARN_ON(refcount_read(&skb->users)); 4888 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED)) 4889 trace_consume_skb(skb); 4890 else 4891 trace_kfree_skb(skb, net_tx_action); 4892 4893 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) 4894 __kfree_skb(skb); 4895 else 4896 __kfree_skb_defer(skb); 4897 } 4898 4899 __kfree_skb_flush(); 4900 } 4901 4902 if (sd->output_queue) { 4903 struct Qdisc *head; 4904 4905 local_irq_disable(); 4906 head = sd->output_queue; 4907 sd->output_queue = NULL; 4908 sd->output_queue_tailp = &sd->output_queue; 4909 local_irq_enable(); 4910 4911 while (head) { 4912 struct Qdisc *q = head; 4913 spinlock_t *root_lock = NULL; 4914 4915 head = head->next_sched; 4916 4917 if (!(q->flags & TCQ_F_NOLOCK)) { 4918 root_lock = qdisc_lock(q); 4919 spin_lock(root_lock); 4920 } 4921 /* We need to make sure head->next_sched is read 4922 * before clearing __QDISC_STATE_SCHED 4923 */ 4924 smp_mb__before_atomic(); 4925 clear_bit(__QDISC_STATE_SCHED, &q->state); 4926 qdisc_run(q); 4927 if (root_lock) 4928 spin_unlock(root_lock); 4929 } 4930 } 4931 4932 xfrm_dev_backlog(sd); 4933 } 4934 4935 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) 4936 /* This hook is defined here for ATM LANE */ 4937 int (*br_fdb_test_addr_hook)(struct net_device *dev, 4938 unsigned char *addr) __read_mostly; 4939 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); 4940 #endif 4941 4942 static inline struct sk_buff * 4943 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 4944 struct net_device *orig_dev, bool *another) 4945 { 4946 #ifdef CONFIG_NET_CLS_ACT 4947 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress); 4948 struct tcf_result cl_res; 4949 4950 /* If there's at least one ingress present somewhere (so 4951 * we get here via enabled static key), remaining devices 4952 * that are not configured with an ingress qdisc will bail 4953 * out here. 4954 */ 4955 if (!miniq) 4956 return skb; 4957 4958 if (*pt_prev) { 4959 *ret = deliver_skb(skb, *pt_prev, orig_dev); 4960 *pt_prev = NULL; 4961 } 4962 4963 qdisc_skb_cb(skb)->pkt_len = skb->len; 4964 qdisc_skb_cb(skb)->mru = 0; 4965 skb->tc_at_ingress = 1; 4966 mini_qdisc_bstats_cpu_update(miniq, skb); 4967 4968 switch (tcf_classify_ingress(skb, miniq->block, miniq->filter_list, 4969 &cl_res, false)) { 4970 case TC_ACT_OK: 4971 case TC_ACT_RECLASSIFY: 4972 skb->tc_index = TC_H_MIN(cl_res.classid); 4973 break; 4974 case TC_ACT_SHOT: 4975 mini_qdisc_qstats_cpu_drop(miniq); 4976 kfree_skb(skb); 4977 return NULL; 4978 case TC_ACT_STOLEN: 4979 case TC_ACT_QUEUED: 4980 case TC_ACT_TRAP: 4981 consume_skb(skb); 4982 return NULL; 4983 case TC_ACT_REDIRECT: 4984 /* skb_mac_header check was done by cls/act_bpf, so 4985 * we can safely push the L2 header back before 4986 * redirecting to another netdev 4987 */ 4988 __skb_push(skb, skb->mac_len); 4989 if (skb_do_redirect(skb) == -EAGAIN) { 4990 __skb_pull(skb, skb->mac_len); 4991 *another = true; 4992 break; 4993 } 4994 return NULL; 4995 case TC_ACT_CONSUMED: 4996 return NULL; 4997 default: 4998 break; 4999 } 5000 #endif /* CONFIG_NET_CLS_ACT */ 5001 return skb; 5002 } 5003 5004 /** 5005 * netdev_is_rx_handler_busy - check if receive handler is registered 5006 * @dev: device to check 5007 * 5008 * Check if a receive handler is already registered for a given device. 5009 * Return true if there one. 5010 * 5011 * The caller must hold the rtnl_mutex. 5012 */ 5013 bool netdev_is_rx_handler_busy(struct net_device *dev) 5014 { 5015 ASSERT_RTNL(); 5016 return dev && rtnl_dereference(dev->rx_handler); 5017 } 5018 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); 5019 5020 /** 5021 * netdev_rx_handler_register - register receive handler 5022 * @dev: device to register a handler for 5023 * @rx_handler: receive handler to register 5024 * @rx_handler_data: data pointer that is used by rx handler 5025 * 5026 * Register a receive handler for a device. This handler will then be 5027 * called from __netif_receive_skb. A negative errno code is returned 5028 * on a failure. 5029 * 5030 * The caller must hold the rtnl_mutex. 5031 * 5032 * For a general description of rx_handler, see enum rx_handler_result. 5033 */ 5034 int netdev_rx_handler_register(struct net_device *dev, 5035 rx_handler_func_t *rx_handler, 5036 void *rx_handler_data) 5037 { 5038 if (netdev_is_rx_handler_busy(dev)) 5039 return -EBUSY; 5040 5041 if (dev->priv_flags & IFF_NO_RX_HANDLER) 5042 return -EINVAL; 5043 5044 /* Note: rx_handler_data must be set before rx_handler */ 5045 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); 5046 rcu_assign_pointer(dev->rx_handler, rx_handler); 5047 5048 return 0; 5049 } 5050 EXPORT_SYMBOL_GPL(netdev_rx_handler_register); 5051 5052 /** 5053 * netdev_rx_handler_unregister - unregister receive handler 5054 * @dev: device to unregister a handler from 5055 * 5056 * Unregister a receive handler from a device. 5057 * 5058 * The caller must hold the rtnl_mutex. 5059 */ 5060 void netdev_rx_handler_unregister(struct net_device *dev) 5061 { 5062 5063 ASSERT_RTNL(); 5064 RCU_INIT_POINTER(dev->rx_handler, NULL); 5065 /* a reader seeing a non NULL rx_handler in a rcu_read_lock() 5066 * section has a guarantee to see a non NULL rx_handler_data 5067 * as well. 5068 */ 5069 synchronize_net(); 5070 RCU_INIT_POINTER(dev->rx_handler_data, NULL); 5071 } 5072 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); 5073 5074 /* 5075 * Limit the use of PFMEMALLOC reserves to those protocols that implement 5076 * the special handling of PFMEMALLOC skbs. 5077 */ 5078 static bool skb_pfmemalloc_protocol(struct sk_buff *skb) 5079 { 5080 switch (skb->protocol) { 5081 case htons(ETH_P_ARP): 5082 case htons(ETH_P_IP): 5083 case htons(ETH_P_IPV6): 5084 case htons(ETH_P_8021Q): 5085 case htons(ETH_P_8021AD): 5086 return true; 5087 default: 5088 return false; 5089 } 5090 } 5091 5092 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, 5093 int *ret, struct net_device *orig_dev) 5094 { 5095 if (nf_hook_ingress_active(skb)) { 5096 int ingress_retval; 5097 5098 if (*pt_prev) { 5099 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5100 *pt_prev = NULL; 5101 } 5102 5103 rcu_read_lock(); 5104 ingress_retval = nf_hook_ingress(skb); 5105 rcu_read_unlock(); 5106 return ingress_retval; 5107 } 5108 return 0; 5109 } 5110 5111 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc, 5112 struct packet_type **ppt_prev) 5113 { 5114 struct packet_type *ptype, *pt_prev; 5115 rx_handler_func_t *rx_handler; 5116 struct sk_buff *skb = *pskb; 5117 struct net_device *orig_dev; 5118 bool deliver_exact = false; 5119 int ret = NET_RX_DROP; 5120 __be16 type; 5121 5122 net_timestamp_check(!netdev_tstamp_prequeue, skb); 5123 5124 trace_netif_receive_skb(skb); 5125 5126 orig_dev = skb->dev; 5127 5128 skb_reset_network_header(skb); 5129 if (!skb_transport_header_was_set(skb)) 5130 skb_reset_transport_header(skb); 5131 skb_reset_mac_len(skb); 5132 5133 pt_prev = NULL; 5134 5135 another_round: 5136 skb->skb_iif = skb->dev->ifindex; 5137 5138 __this_cpu_inc(softnet_data.processed); 5139 5140 if (static_branch_unlikely(&generic_xdp_needed_key)) { 5141 int ret2; 5142 5143 preempt_disable(); 5144 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb); 5145 preempt_enable(); 5146 5147 if (ret2 != XDP_PASS) { 5148 ret = NET_RX_DROP; 5149 goto out; 5150 } 5151 skb_reset_mac_len(skb); 5152 } 5153 5154 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) || 5155 skb->protocol == cpu_to_be16(ETH_P_8021AD)) { 5156 skb = skb_vlan_untag(skb); 5157 if (unlikely(!skb)) 5158 goto out; 5159 } 5160 5161 if (skb_skip_tc_classify(skb)) 5162 goto skip_classify; 5163 5164 if (pfmemalloc) 5165 goto skip_taps; 5166 5167 list_for_each_entry_rcu(ptype, &ptype_all, list) { 5168 if (pt_prev) 5169 ret = deliver_skb(skb, pt_prev, orig_dev); 5170 pt_prev = ptype; 5171 } 5172 5173 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { 5174 if (pt_prev) 5175 ret = deliver_skb(skb, pt_prev, orig_dev); 5176 pt_prev = ptype; 5177 } 5178 5179 skip_taps: 5180 #ifdef CONFIG_NET_INGRESS 5181 if (static_branch_unlikely(&ingress_needed_key)) { 5182 bool another = false; 5183 5184 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev, 5185 &another); 5186 if (another) 5187 goto another_round; 5188 if (!skb) 5189 goto out; 5190 5191 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) 5192 goto out; 5193 } 5194 #endif 5195 skb_reset_redirect(skb); 5196 skip_classify: 5197 if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) 5198 goto drop; 5199 5200 if (skb_vlan_tag_present(skb)) { 5201 if (pt_prev) { 5202 ret = deliver_skb(skb, pt_prev, orig_dev); 5203 pt_prev = NULL; 5204 } 5205 if (vlan_do_receive(&skb)) 5206 goto another_round; 5207 else if (unlikely(!skb)) 5208 goto out; 5209 } 5210 5211 rx_handler = rcu_dereference(skb->dev->rx_handler); 5212 if (rx_handler) { 5213 if (pt_prev) { 5214 ret = deliver_skb(skb, pt_prev, orig_dev); 5215 pt_prev = NULL; 5216 } 5217 switch (rx_handler(&skb)) { 5218 case RX_HANDLER_CONSUMED: 5219 ret = NET_RX_SUCCESS; 5220 goto out; 5221 case RX_HANDLER_ANOTHER: 5222 goto another_round; 5223 case RX_HANDLER_EXACT: 5224 deliver_exact = true; 5225 case RX_HANDLER_PASS: 5226 break; 5227 default: 5228 BUG(); 5229 } 5230 } 5231 5232 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) { 5233 check_vlan_id: 5234 if (skb_vlan_tag_get_id(skb)) { 5235 /* Vlan id is non 0 and vlan_do_receive() above couldn't 5236 * find vlan device. 5237 */ 5238 skb->pkt_type = PACKET_OTHERHOST; 5239 } else if (skb->protocol == cpu_to_be16(ETH_P_8021Q) || 5240 skb->protocol == cpu_to_be16(ETH_P_8021AD)) { 5241 /* Outer header is 802.1P with vlan 0, inner header is 5242 * 802.1Q or 802.1AD and vlan_do_receive() above could 5243 * not find vlan dev for vlan id 0. 5244 */ 5245 __vlan_hwaccel_clear_tag(skb); 5246 skb = skb_vlan_untag(skb); 5247 if (unlikely(!skb)) 5248 goto out; 5249 if (vlan_do_receive(&skb)) 5250 /* After stripping off 802.1P header with vlan 0 5251 * vlan dev is found for inner header. 5252 */ 5253 goto another_round; 5254 else if (unlikely(!skb)) 5255 goto out; 5256 else 5257 /* We have stripped outer 802.1P vlan 0 header. 5258 * But could not find vlan dev. 5259 * check again for vlan id to set OTHERHOST. 5260 */ 5261 goto check_vlan_id; 5262 } 5263 /* Note: we might in the future use prio bits 5264 * and set skb->priority like in vlan_do_receive() 5265 * For the time being, just ignore Priority Code Point 5266 */ 5267 __vlan_hwaccel_clear_tag(skb); 5268 } 5269 5270 type = skb->protocol; 5271 5272 /* deliver only exact match when indicated */ 5273 if (likely(!deliver_exact)) { 5274 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5275 &ptype_base[ntohs(type) & 5276 PTYPE_HASH_MASK]); 5277 } 5278 5279 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5280 &orig_dev->ptype_specific); 5281 5282 if (unlikely(skb->dev != orig_dev)) { 5283 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5284 &skb->dev->ptype_specific); 5285 } 5286 5287 if (pt_prev) { 5288 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 5289 goto drop; 5290 *ppt_prev = pt_prev; 5291 } else { 5292 drop: 5293 if (!deliver_exact) 5294 atomic_long_inc(&skb->dev->rx_dropped); 5295 else 5296 atomic_long_inc(&skb->dev->rx_nohandler); 5297 kfree_skb(skb); 5298 /* Jamal, now you will not able to escape explaining 5299 * me how you were going to use this. :-) 5300 */ 5301 ret = NET_RX_DROP; 5302 } 5303 5304 out: 5305 /* The invariant here is that if *ppt_prev is not NULL 5306 * then skb should also be non-NULL. 5307 * 5308 * Apparently *ppt_prev assignment above holds this invariant due to 5309 * skb dereferencing near it. 5310 */ 5311 *pskb = skb; 5312 return ret; 5313 } 5314 5315 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc) 5316 { 5317 struct net_device *orig_dev = skb->dev; 5318 struct packet_type *pt_prev = NULL; 5319 int ret; 5320 5321 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5322 if (pt_prev) 5323 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb, 5324 skb->dev, pt_prev, orig_dev); 5325 return ret; 5326 } 5327 5328 /** 5329 * netif_receive_skb_core - special purpose version of netif_receive_skb 5330 * @skb: buffer to process 5331 * 5332 * More direct receive version of netif_receive_skb(). It should 5333 * only be used by callers that have a need to skip RPS and Generic XDP. 5334 * Caller must also take care of handling if ``(page_is_)pfmemalloc``. 5335 * 5336 * This function may only be called from softirq context and interrupts 5337 * should be enabled. 5338 * 5339 * Return values (usually ignored): 5340 * NET_RX_SUCCESS: no congestion 5341 * NET_RX_DROP: packet was dropped 5342 */ 5343 int netif_receive_skb_core(struct sk_buff *skb) 5344 { 5345 int ret; 5346 5347 rcu_read_lock(); 5348 ret = __netif_receive_skb_one_core(skb, false); 5349 rcu_read_unlock(); 5350 5351 return ret; 5352 } 5353 EXPORT_SYMBOL(netif_receive_skb_core); 5354 5355 static inline void __netif_receive_skb_list_ptype(struct list_head *head, 5356 struct packet_type *pt_prev, 5357 struct net_device *orig_dev) 5358 { 5359 struct sk_buff *skb, *next; 5360 5361 if (!pt_prev) 5362 return; 5363 if (list_empty(head)) 5364 return; 5365 if (pt_prev->list_func != NULL) 5366 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv, 5367 ip_list_rcv, head, pt_prev, orig_dev); 5368 else 5369 list_for_each_entry_safe(skb, next, head, list) { 5370 skb_list_del_init(skb); 5371 pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 5372 } 5373 } 5374 5375 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc) 5376 { 5377 /* Fast-path assumptions: 5378 * - There is no RX handler. 5379 * - Only one packet_type matches. 5380 * If either of these fails, we will end up doing some per-packet 5381 * processing in-line, then handling the 'last ptype' for the whole 5382 * sublist. This can't cause out-of-order delivery to any single ptype, 5383 * because the 'last ptype' must be constant across the sublist, and all 5384 * other ptypes are handled per-packet. 5385 */ 5386 /* Current (common) ptype of sublist */ 5387 struct packet_type *pt_curr = NULL; 5388 /* Current (common) orig_dev of sublist */ 5389 struct net_device *od_curr = NULL; 5390 struct list_head sublist; 5391 struct sk_buff *skb, *next; 5392 5393 INIT_LIST_HEAD(&sublist); 5394 list_for_each_entry_safe(skb, next, head, list) { 5395 struct net_device *orig_dev = skb->dev; 5396 struct packet_type *pt_prev = NULL; 5397 5398 skb_list_del_init(skb); 5399 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5400 if (!pt_prev) 5401 continue; 5402 if (pt_curr != pt_prev || od_curr != orig_dev) { 5403 /* dispatch old sublist */ 5404 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5405 /* start new sublist */ 5406 INIT_LIST_HEAD(&sublist); 5407 pt_curr = pt_prev; 5408 od_curr = orig_dev; 5409 } 5410 list_add_tail(&skb->list, &sublist); 5411 } 5412 5413 /* dispatch final sublist */ 5414 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5415 } 5416 5417 static int __netif_receive_skb(struct sk_buff *skb) 5418 { 5419 int ret; 5420 5421 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { 5422 unsigned int noreclaim_flag; 5423 5424 /* 5425 * PFMEMALLOC skbs are special, they should 5426 * - be delivered to SOCK_MEMALLOC sockets only 5427 * - stay away from userspace 5428 * - have bounded memory usage 5429 * 5430 * Use PF_MEMALLOC as this saves us from propagating the allocation 5431 * context down to all allocation sites. 5432 */ 5433 noreclaim_flag = memalloc_noreclaim_save(); 5434 ret = __netif_receive_skb_one_core(skb, true); 5435 memalloc_noreclaim_restore(noreclaim_flag); 5436 } else 5437 ret = __netif_receive_skb_one_core(skb, false); 5438 5439 return ret; 5440 } 5441 5442 static void __netif_receive_skb_list(struct list_head *head) 5443 { 5444 unsigned long noreclaim_flag = 0; 5445 struct sk_buff *skb, *next; 5446 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */ 5447 5448 list_for_each_entry_safe(skb, next, head, list) { 5449 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) { 5450 struct list_head sublist; 5451 5452 /* Handle the previous sublist */ 5453 list_cut_before(&sublist, head, &skb->list); 5454 if (!list_empty(&sublist)) 5455 __netif_receive_skb_list_core(&sublist, pfmemalloc); 5456 pfmemalloc = !pfmemalloc; 5457 /* See comments in __netif_receive_skb */ 5458 if (pfmemalloc) 5459 noreclaim_flag = memalloc_noreclaim_save(); 5460 else 5461 memalloc_noreclaim_restore(noreclaim_flag); 5462 } 5463 } 5464 /* Handle the remaining sublist */ 5465 if (!list_empty(head)) 5466 __netif_receive_skb_list_core(head, pfmemalloc); 5467 /* Restore pflags */ 5468 if (pfmemalloc) 5469 memalloc_noreclaim_restore(noreclaim_flag); 5470 } 5471 5472 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp) 5473 { 5474 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog); 5475 struct bpf_prog *new = xdp->prog; 5476 int ret = 0; 5477 5478 if (new) { 5479 u32 i; 5480 5481 mutex_lock(&new->aux->used_maps_mutex); 5482 5483 /* generic XDP does not work with DEVMAPs that can 5484 * have a bpf_prog installed on an entry 5485 */ 5486 for (i = 0; i < new->aux->used_map_cnt; i++) { 5487 if (dev_map_can_have_prog(new->aux->used_maps[i]) || 5488 cpu_map_prog_allowed(new->aux->used_maps[i])) { 5489 mutex_unlock(&new->aux->used_maps_mutex); 5490 return -EINVAL; 5491 } 5492 } 5493 5494 mutex_unlock(&new->aux->used_maps_mutex); 5495 } 5496 5497 switch (xdp->command) { 5498 case XDP_SETUP_PROG: 5499 rcu_assign_pointer(dev->xdp_prog, new); 5500 if (old) 5501 bpf_prog_put(old); 5502 5503 if (old && !new) { 5504 static_branch_dec(&generic_xdp_needed_key); 5505 } else if (new && !old) { 5506 static_branch_inc(&generic_xdp_needed_key); 5507 dev_disable_lro(dev); 5508 dev_disable_gro_hw(dev); 5509 } 5510 break; 5511 5512 default: 5513 ret = -EINVAL; 5514 break; 5515 } 5516 5517 return ret; 5518 } 5519 5520 static int netif_receive_skb_internal(struct sk_buff *skb) 5521 { 5522 int ret; 5523 5524 net_timestamp_check(netdev_tstamp_prequeue, skb); 5525 5526 if (skb_defer_rx_timestamp(skb)) 5527 return NET_RX_SUCCESS; 5528 5529 rcu_read_lock(); 5530 #ifdef CONFIG_RPS 5531 if (static_branch_unlikely(&rps_needed)) { 5532 struct rps_dev_flow voidflow, *rflow = &voidflow; 5533 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5534 5535 if (cpu >= 0) { 5536 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5537 rcu_read_unlock(); 5538 return ret; 5539 } 5540 } 5541 #endif 5542 ret = __netif_receive_skb(skb); 5543 rcu_read_unlock(); 5544 return ret; 5545 } 5546 5547 static void netif_receive_skb_list_internal(struct list_head *head) 5548 { 5549 struct sk_buff *skb, *next; 5550 struct list_head sublist; 5551 5552 INIT_LIST_HEAD(&sublist); 5553 list_for_each_entry_safe(skb, next, head, list) { 5554 net_timestamp_check(netdev_tstamp_prequeue, skb); 5555 skb_list_del_init(skb); 5556 if (!skb_defer_rx_timestamp(skb)) 5557 list_add_tail(&skb->list, &sublist); 5558 } 5559 list_splice_init(&sublist, head); 5560 5561 rcu_read_lock(); 5562 #ifdef CONFIG_RPS 5563 if (static_branch_unlikely(&rps_needed)) { 5564 list_for_each_entry_safe(skb, next, head, list) { 5565 struct rps_dev_flow voidflow, *rflow = &voidflow; 5566 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5567 5568 if (cpu >= 0) { 5569 /* Will be handled, remove from list */ 5570 skb_list_del_init(skb); 5571 enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5572 } 5573 } 5574 } 5575 #endif 5576 __netif_receive_skb_list(head); 5577 rcu_read_unlock(); 5578 } 5579 5580 /** 5581 * netif_receive_skb - process receive buffer from network 5582 * @skb: buffer to process 5583 * 5584 * netif_receive_skb() is the main receive data processing function. 5585 * It always succeeds. The buffer may be dropped during processing 5586 * for congestion control or by the protocol layers. 5587 * 5588 * This function may only be called from softirq context and interrupts 5589 * should be enabled. 5590 * 5591 * Return values (usually ignored): 5592 * NET_RX_SUCCESS: no congestion 5593 * NET_RX_DROP: packet was dropped 5594 */ 5595 int netif_receive_skb(struct sk_buff *skb) 5596 { 5597 int ret; 5598 5599 trace_netif_receive_skb_entry(skb); 5600 5601 ret = netif_receive_skb_internal(skb); 5602 trace_netif_receive_skb_exit(ret); 5603 5604 return ret; 5605 } 5606 EXPORT_SYMBOL(netif_receive_skb); 5607 5608 /** 5609 * netif_receive_skb_list - process many receive buffers from network 5610 * @head: list of skbs to process. 5611 * 5612 * Since return value of netif_receive_skb() is normally ignored, and 5613 * wouldn't be meaningful for a list, this function returns void. 5614 * 5615 * This function may only be called from softirq context and interrupts 5616 * should be enabled. 5617 */ 5618 void netif_receive_skb_list(struct list_head *head) 5619 { 5620 struct sk_buff *skb; 5621 5622 if (list_empty(head)) 5623 return; 5624 if (trace_netif_receive_skb_list_entry_enabled()) { 5625 list_for_each_entry(skb, head, list) 5626 trace_netif_receive_skb_list_entry(skb); 5627 } 5628 netif_receive_skb_list_internal(head); 5629 trace_netif_receive_skb_list_exit(0); 5630 } 5631 EXPORT_SYMBOL(netif_receive_skb_list); 5632 5633 static DEFINE_PER_CPU(struct work_struct, flush_works); 5634 5635 /* Network device is going away, flush any packets still pending */ 5636 static void flush_backlog(struct work_struct *work) 5637 { 5638 struct sk_buff *skb, *tmp; 5639 struct softnet_data *sd; 5640 5641 local_bh_disable(); 5642 sd = this_cpu_ptr(&softnet_data); 5643 5644 local_irq_disable(); 5645 rps_lock(sd); 5646 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { 5647 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5648 __skb_unlink(skb, &sd->input_pkt_queue); 5649 dev_kfree_skb_irq(skb); 5650 input_queue_head_incr(sd); 5651 } 5652 } 5653 rps_unlock(sd); 5654 local_irq_enable(); 5655 5656 skb_queue_walk_safe(&sd->process_queue, skb, tmp) { 5657 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5658 __skb_unlink(skb, &sd->process_queue); 5659 kfree_skb(skb); 5660 input_queue_head_incr(sd); 5661 } 5662 } 5663 local_bh_enable(); 5664 } 5665 5666 static bool flush_required(int cpu) 5667 { 5668 #if IS_ENABLED(CONFIG_RPS) 5669 struct softnet_data *sd = &per_cpu(softnet_data, cpu); 5670 bool do_flush; 5671 5672 local_irq_disable(); 5673 rps_lock(sd); 5674 5675 /* as insertion into process_queue happens with the rps lock held, 5676 * process_queue access may race only with dequeue 5677 */ 5678 do_flush = !skb_queue_empty(&sd->input_pkt_queue) || 5679 !skb_queue_empty_lockless(&sd->process_queue); 5680 rps_unlock(sd); 5681 local_irq_enable(); 5682 5683 return do_flush; 5684 #endif 5685 /* without RPS we can't safely check input_pkt_queue: during a 5686 * concurrent remote skb_queue_splice() we can detect as empty both 5687 * input_pkt_queue and process_queue even if the latter could end-up 5688 * containing a lot of packets. 5689 */ 5690 return true; 5691 } 5692 5693 static void flush_all_backlogs(void) 5694 { 5695 static cpumask_t flush_cpus; 5696 unsigned int cpu; 5697 5698 /* since we are under rtnl lock protection we can use static data 5699 * for the cpumask and avoid allocating on stack the possibly 5700 * large mask 5701 */ 5702 ASSERT_RTNL(); 5703 5704 get_online_cpus(); 5705 5706 cpumask_clear(&flush_cpus); 5707 for_each_online_cpu(cpu) { 5708 if (flush_required(cpu)) { 5709 queue_work_on(cpu, system_highpri_wq, 5710 per_cpu_ptr(&flush_works, cpu)); 5711 cpumask_set_cpu(cpu, &flush_cpus); 5712 } 5713 } 5714 5715 /* we can have in flight packet[s] on the cpus we are not flushing, 5716 * synchronize_net() in rollback_registered_many() will take care of 5717 * them 5718 */ 5719 for_each_cpu(cpu, &flush_cpus) 5720 flush_work(per_cpu_ptr(&flush_works, cpu)); 5721 5722 put_online_cpus(); 5723 } 5724 5725 /* Pass the currently batched GRO_NORMAL SKBs up to the stack. */ 5726 static void gro_normal_list(struct napi_struct *napi) 5727 { 5728 if (!napi->rx_count) 5729 return; 5730 netif_receive_skb_list_internal(&napi->rx_list); 5731 INIT_LIST_HEAD(&napi->rx_list); 5732 napi->rx_count = 0; 5733 } 5734 5735 /* Queue one GRO_NORMAL SKB up for list processing. If batch size exceeded, 5736 * pass the whole batch up to the stack. 5737 */ 5738 static void gro_normal_one(struct napi_struct *napi, struct sk_buff *skb, int segs) 5739 { 5740 list_add_tail(&skb->list, &napi->rx_list); 5741 napi->rx_count += segs; 5742 if (napi->rx_count >= gro_normal_batch) 5743 gro_normal_list(napi); 5744 } 5745 5746 INDIRECT_CALLABLE_DECLARE(int inet_gro_complete(struct sk_buff *, int)); 5747 INDIRECT_CALLABLE_DECLARE(int ipv6_gro_complete(struct sk_buff *, int)); 5748 static int napi_gro_complete(struct napi_struct *napi, struct sk_buff *skb) 5749 { 5750 struct packet_offload *ptype; 5751 __be16 type = skb->protocol; 5752 struct list_head *head = &offload_base; 5753 int err = -ENOENT; 5754 5755 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb)); 5756 5757 if (NAPI_GRO_CB(skb)->count == 1) { 5758 skb_shinfo(skb)->gso_size = 0; 5759 goto out; 5760 } 5761 5762 rcu_read_lock(); 5763 list_for_each_entry_rcu(ptype, head, list) { 5764 if (ptype->type != type || !ptype->callbacks.gro_complete) 5765 continue; 5766 5767 err = INDIRECT_CALL_INET(ptype->callbacks.gro_complete, 5768 ipv6_gro_complete, inet_gro_complete, 5769 skb, 0); 5770 break; 5771 } 5772 rcu_read_unlock(); 5773 5774 if (err) { 5775 WARN_ON(&ptype->list == head); 5776 kfree_skb(skb); 5777 return NET_RX_SUCCESS; 5778 } 5779 5780 out: 5781 gro_normal_one(napi, skb, NAPI_GRO_CB(skb)->count); 5782 return NET_RX_SUCCESS; 5783 } 5784 5785 static void __napi_gro_flush_chain(struct napi_struct *napi, u32 index, 5786 bool flush_old) 5787 { 5788 struct list_head *head = &napi->gro_hash[index].list; 5789 struct sk_buff *skb, *p; 5790 5791 list_for_each_entry_safe_reverse(skb, p, head, list) { 5792 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies) 5793 return; 5794 skb_list_del_init(skb); 5795 napi_gro_complete(napi, skb); 5796 napi->gro_hash[index].count--; 5797 } 5798 5799 if (!napi->gro_hash[index].count) 5800 __clear_bit(index, &napi->gro_bitmask); 5801 } 5802 5803 /* napi->gro_hash[].list contains packets ordered by age. 5804 * youngest packets at the head of it. 5805 * Complete skbs in reverse order to reduce latencies. 5806 */ 5807 void napi_gro_flush(struct napi_struct *napi, bool flush_old) 5808 { 5809 unsigned long bitmask = napi->gro_bitmask; 5810 unsigned int i, base = ~0U; 5811 5812 while ((i = ffs(bitmask)) != 0) { 5813 bitmask >>= i; 5814 base += i; 5815 __napi_gro_flush_chain(napi, base, flush_old); 5816 } 5817 } 5818 EXPORT_SYMBOL(napi_gro_flush); 5819 5820 static struct list_head *gro_list_prepare(struct napi_struct *napi, 5821 struct sk_buff *skb) 5822 { 5823 unsigned int maclen = skb->dev->hard_header_len; 5824 u32 hash = skb_get_hash_raw(skb); 5825 struct list_head *head; 5826 struct sk_buff *p; 5827 5828 head = &napi->gro_hash[hash & (GRO_HASH_BUCKETS - 1)].list; 5829 list_for_each_entry(p, head, list) { 5830 unsigned long diffs; 5831 5832 NAPI_GRO_CB(p)->flush = 0; 5833 5834 if (hash != skb_get_hash_raw(p)) { 5835 NAPI_GRO_CB(p)->same_flow = 0; 5836 continue; 5837 } 5838 5839 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev; 5840 diffs |= skb_vlan_tag_present(p) ^ skb_vlan_tag_present(skb); 5841 if (skb_vlan_tag_present(p)) 5842 diffs |= skb_vlan_tag_get(p) ^ skb_vlan_tag_get(skb); 5843 diffs |= skb_metadata_dst_cmp(p, skb); 5844 diffs |= skb_metadata_differs(p, skb); 5845 if (maclen == ETH_HLEN) 5846 diffs |= compare_ether_header(skb_mac_header(p), 5847 skb_mac_header(skb)); 5848 else if (!diffs) 5849 diffs = memcmp(skb_mac_header(p), 5850 skb_mac_header(skb), 5851 maclen); 5852 NAPI_GRO_CB(p)->same_flow = !diffs; 5853 } 5854 5855 return head; 5856 } 5857 5858 static void skb_gro_reset_offset(struct sk_buff *skb) 5859 { 5860 const struct skb_shared_info *pinfo = skb_shinfo(skb); 5861 const skb_frag_t *frag0 = &pinfo->frags[0]; 5862 5863 NAPI_GRO_CB(skb)->data_offset = 0; 5864 NAPI_GRO_CB(skb)->frag0 = NULL; 5865 NAPI_GRO_CB(skb)->frag0_len = 0; 5866 5867 if (!skb_headlen(skb) && pinfo->nr_frags && 5868 !PageHighMem(skb_frag_page(frag0))) { 5869 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0); 5870 NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int, 5871 skb_frag_size(frag0), 5872 skb->end - skb->tail); 5873 } 5874 } 5875 5876 static void gro_pull_from_frag0(struct sk_buff *skb, int grow) 5877 { 5878 struct skb_shared_info *pinfo = skb_shinfo(skb); 5879 5880 BUG_ON(skb->end - skb->tail < grow); 5881 5882 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); 5883 5884 skb->data_len -= grow; 5885 skb->tail += grow; 5886 5887 skb_frag_off_add(&pinfo->frags[0], grow); 5888 skb_frag_size_sub(&pinfo->frags[0], grow); 5889 5890 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) { 5891 skb_frag_unref(skb, 0); 5892 memmove(pinfo->frags, pinfo->frags + 1, 5893 --pinfo->nr_frags * sizeof(pinfo->frags[0])); 5894 } 5895 } 5896 5897 static void gro_flush_oldest(struct napi_struct *napi, struct list_head *head) 5898 { 5899 struct sk_buff *oldest; 5900 5901 oldest = list_last_entry(head, struct sk_buff, list); 5902 5903 /* We are called with head length >= MAX_GRO_SKBS, so this is 5904 * impossible. 5905 */ 5906 if (WARN_ON_ONCE(!oldest)) 5907 return; 5908 5909 /* Do not adjust napi->gro_hash[].count, caller is adding a new 5910 * SKB to the chain. 5911 */ 5912 skb_list_del_init(oldest); 5913 napi_gro_complete(napi, oldest); 5914 } 5915 5916 INDIRECT_CALLABLE_DECLARE(struct sk_buff *inet_gro_receive(struct list_head *, 5917 struct sk_buff *)); 5918 INDIRECT_CALLABLE_DECLARE(struct sk_buff *ipv6_gro_receive(struct list_head *, 5919 struct sk_buff *)); 5920 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) 5921 { 5922 u32 hash = skb_get_hash_raw(skb) & (GRO_HASH_BUCKETS - 1); 5923 struct list_head *head = &offload_base; 5924 struct packet_offload *ptype; 5925 __be16 type = skb->protocol; 5926 struct list_head *gro_head; 5927 struct sk_buff *pp = NULL; 5928 enum gro_result ret; 5929 int same_flow; 5930 int grow; 5931 5932 if (netif_elide_gro(skb->dev)) 5933 goto normal; 5934 5935 gro_head = gro_list_prepare(napi, skb); 5936 5937 rcu_read_lock(); 5938 list_for_each_entry_rcu(ptype, head, list) { 5939 if (ptype->type != type || !ptype->callbacks.gro_receive) 5940 continue; 5941 5942 skb_set_network_header(skb, skb_gro_offset(skb)); 5943 skb_reset_mac_len(skb); 5944 NAPI_GRO_CB(skb)->same_flow = 0; 5945 NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb); 5946 NAPI_GRO_CB(skb)->free = 0; 5947 NAPI_GRO_CB(skb)->encap_mark = 0; 5948 NAPI_GRO_CB(skb)->recursion_counter = 0; 5949 NAPI_GRO_CB(skb)->is_fou = 0; 5950 NAPI_GRO_CB(skb)->is_atomic = 1; 5951 NAPI_GRO_CB(skb)->gro_remcsum_start = 0; 5952 5953 /* Setup for GRO checksum validation */ 5954 switch (skb->ip_summed) { 5955 case CHECKSUM_COMPLETE: 5956 NAPI_GRO_CB(skb)->csum = skb->csum; 5957 NAPI_GRO_CB(skb)->csum_valid = 1; 5958 NAPI_GRO_CB(skb)->csum_cnt = 0; 5959 break; 5960 case CHECKSUM_UNNECESSARY: 5961 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1; 5962 NAPI_GRO_CB(skb)->csum_valid = 0; 5963 break; 5964 default: 5965 NAPI_GRO_CB(skb)->csum_cnt = 0; 5966 NAPI_GRO_CB(skb)->csum_valid = 0; 5967 } 5968 5969 pp = INDIRECT_CALL_INET(ptype->callbacks.gro_receive, 5970 ipv6_gro_receive, inet_gro_receive, 5971 gro_head, skb); 5972 break; 5973 } 5974 rcu_read_unlock(); 5975 5976 if (&ptype->list == head) 5977 goto normal; 5978 5979 if (PTR_ERR(pp) == -EINPROGRESS) { 5980 ret = GRO_CONSUMED; 5981 goto ok; 5982 } 5983 5984 same_flow = NAPI_GRO_CB(skb)->same_flow; 5985 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; 5986 5987 if (pp) { 5988 skb_list_del_init(pp); 5989 napi_gro_complete(napi, pp); 5990 napi->gro_hash[hash].count--; 5991 } 5992 5993 if (same_flow) 5994 goto ok; 5995 5996 if (NAPI_GRO_CB(skb)->flush) 5997 goto normal; 5998 5999 if (unlikely(napi->gro_hash[hash].count >= MAX_GRO_SKBS)) { 6000 gro_flush_oldest(napi, gro_head); 6001 } else { 6002 napi->gro_hash[hash].count++; 6003 } 6004 NAPI_GRO_CB(skb)->count = 1; 6005 NAPI_GRO_CB(skb)->age = jiffies; 6006 NAPI_GRO_CB(skb)->last = skb; 6007 skb_shinfo(skb)->gso_size = skb_gro_len(skb); 6008 list_add(&skb->list, gro_head); 6009 ret = GRO_HELD; 6010 6011 pull: 6012 grow = skb_gro_offset(skb) - skb_headlen(skb); 6013 if (grow > 0) 6014 gro_pull_from_frag0(skb, grow); 6015 ok: 6016 if (napi->gro_hash[hash].count) { 6017 if (!test_bit(hash, &napi->gro_bitmask)) 6018 __set_bit(hash, &napi->gro_bitmask); 6019 } else if (test_bit(hash, &napi->gro_bitmask)) { 6020 __clear_bit(hash, &napi->gro_bitmask); 6021 } 6022 6023 return ret; 6024 6025 normal: 6026 ret = GRO_NORMAL; 6027 goto pull; 6028 } 6029 6030 struct packet_offload *gro_find_receive_by_type(__be16 type) 6031 { 6032 struct list_head *offload_head = &offload_base; 6033 struct packet_offload *ptype; 6034 6035 list_for_each_entry_rcu(ptype, offload_head, list) { 6036 if (ptype->type != type || !ptype->callbacks.gro_receive) 6037 continue; 6038 return ptype; 6039 } 6040 return NULL; 6041 } 6042 EXPORT_SYMBOL(gro_find_receive_by_type); 6043 6044 struct packet_offload *gro_find_complete_by_type(__be16 type) 6045 { 6046 struct list_head *offload_head = &offload_base; 6047 struct packet_offload *ptype; 6048 6049 list_for_each_entry_rcu(ptype, offload_head, list) { 6050 if (ptype->type != type || !ptype->callbacks.gro_complete) 6051 continue; 6052 return ptype; 6053 } 6054 return NULL; 6055 } 6056 EXPORT_SYMBOL(gro_find_complete_by_type); 6057 6058 static void napi_skb_free_stolen_head(struct sk_buff *skb) 6059 { 6060 skb_dst_drop(skb); 6061 skb_ext_put(skb); 6062 kmem_cache_free(skbuff_head_cache, skb); 6063 } 6064 6065 static gro_result_t napi_skb_finish(struct napi_struct *napi, 6066 struct sk_buff *skb, 6067 gro_result_t ret) 6068 { 6069 switch (ret) { 6070 case GRO_NORMAL: 6071 gro_normal_one(napi, skb, 1); 6072 break; 6073 6074 case GRO_DROP: 6075 kfree_skb(skb); 6076 break; 6077 6078 case GRO_MERGED_FREE: 6079 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) 6080 napi_skb_free_stolen_head(skb); 6081 else 6082 __kfree_skb(skb); 6083 break; 6084 6085 case GRO_HELD: 6086 case GRO_MERGED: 6087 case GRO_CONSUMED: 6088 break; 6089 } 6090 6091 return ret; 6092 } 6093 6094 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) 6095 { 6096 gro_result_t ret; 6097 6098 skb_mark_napi_id(skb, napi); 6099 trace_napi_gro_receive_entry(skb); 6100 6101 skb_gro_reset_offset(skb); 6102 6103 ret = napi_skb_finish(napi, skb, dev_gro_receive(napi, skb)); 6104 trace_napi_gro_receive_exit(ret); 6105 6106 return ret; 6107 } 6108 EXPORT_SYMBOL(napi_gro_receive); 6109 6110 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) 6111 { 6112 if (unlikely(skb->pfmemalloc)) { 6113 consume_skb(skb); 6114 return; 6115 } 6116 __skb_pull(skb, skb_headlen(skb)); 6117 /* restore the reserve we had after netdev_alloc_skb_ip_align() */ 6118 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb)); 6119 __vlan_hwaccel_clear_tag(skb); 6120 skb->dev = napi->dev; 6121 skb->skb_iif = 0; 6122 6123 /* eth_type_trans() assumes pkt_type is PACKET_HOST */ 6124 skb->pkt_type = PACKET_HOST; 6125 6126 skb->encapsulation = 0; 6127 skb_shinfo(skb)->gso_type = 0; 6128 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); 6129 skb_ext_reset(skb); 6130 6131 napi->skb = skb; 6132 } 6133 6134 struct sk_buff *napi_get_frags(struct napi_struct *napi) 6135 { 6136 struct sk_buff *skb = napi->skb; 6137 6138 if (!skb) { 6139 skb = napi_alloc_skb(napi, GRO_MAX_HEAD); 6140 if (skb) { 6141 napi->skb = skb; 6142 skb_mark_napi_id(skb, napi); 6143 } 6144 } 6145 return skb; 6146 } 6147 EXPORT_SYMBOL(napi_get_frags); 6148 6149 static gro_result_t napi_frags_finish(struct napi_struct *napi, 6150 struct sk_buff *skb, 6151 gro_result_t ret) 6152 { 6153 switch (ret) { 6154 case GRO_NORMAL: 6155 case GRO_HELD: 6156 __skb_push(skb, ETH_HLEN); 6157 skb->protocol = eth_type_trans(skb, skb->dev); 6158 if (ret == GRO_NORMAL) 6159 gro_normal_one(napi, skb, 1); 6160 break; 6161 6162 case GRO_DROP: 6163 napi_reuse_skb(napi, skb); 6164 break; 6165 6166 case GRO_MERGED_FREE: 6167 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) 6168 napi_skb_free_stolen_head(skb); 6169 else 6170 napi_reuse_skb(napi, skb); 6171 break; 6172 6173 case GRO_MERGED: 6174 case GRO_CONSUMED: 6175 break; 6176 } 6177 6178 return ret; 6179 } 6180 6181 /* Upper GRO stack assumes network header starts at gro_offset=0 6182 * Drivers could call both napi_gro_frags() and napi_gro_receive() 6183 * We copy ethernet header into skb->data to have a common layout. 6184 */ 6185 static struct sk_buff *napi_frags_skb(struct napi_struct *napi) 6186 { 6187 struct sk_buff *skb = napi->skb; 6188 const struct ethhdr *eth; 6189 unsigned int hlen = sizeof(*eth); 6190 6191 napi->skb = NULL; 6192 6193 skb_reset_mac_header(skb); 6194 skb_gro_reset_offset(skb); 6195 6196 if (unlikely(skb_gro_header_hard(skb, hlen))) { 6197 eth = skb_gro_header_slow(skb, hlen, 0); 6198 if (unlikely(!eth)) { 6199 net_warn_ratelimited("%s: dropping impossible skb from %s\n", 6200 __func__, napi->dev->name); 6201 napi_reuse_skb(napi, skb); 6202 return NULL; 6203 } 6204 } else { 6205 eth = (const struct ethhdr *)skb->data; 6206 gro_pull_from_frag0(skb, hlen); 6207 NAPI_GRO_CB(skb)->frag0 += hlen; 6208 NAPI_GRO_CB(skb)->frag0_len -= hlen; 6209 } 6210 __skb_pull(skb, hlen); 6211 6212 /* 6213 * This works because the only protocols we care about don't require 6214 * special handling. 6215 * We'll fix it up properly in napi_frags_finish() 6216 */ 6217 skb->protocol = eth->h_proto; 6218 6219 return skb; 6220 } 6221 6222 gro_result_t napi_gro_frags(struct napi_struct *napi) 6223 { 6224 gro_result_t ret; 6225 struct sk_buff *skb = napi_frags_skb(napi); 6226 6227 if (!skb) 6228 return GRO_DROP; 6229 6230 trace_napi_gro_frags_entry(skb); 6231 6232 ret = napi_frags_finish(napi, skb, dev_gro_receive(napi, skb)); 6233 trace_napi_gro_frags_exit(ret); 6234 6235 return ret; 6236 } 6237 EXPORT_SYMBOL(napi_gro_frags); 6238 6239 /* Compute the checksum from gro_offset and return the folded value 6240 * after adding in any pseudo checksum. 6241 */ 6242 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb) 6243 { 6244 __wsum wsum; 6245 __sum16 sum; 6246 6247 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0); 6248 6249 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */ 6250 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum)); 6251 /* See comments in __skb_checksum_complete(). */ 6252 if (likely(!sum)) { 6253 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && 6254 !skb->csum_complete_sw) 6255 netdev_rx_csum_fault(skb->dev, skb); 6256 } 6257 6258 NAPI_GRO_CB(skb)->csum = wsum; 6259 NAPI_GRO_CB(skb)->csum_valid = 1; 6260 6261 return sum; 6262 } 6263 EXPORT_SYMBOL(__skb_gro_checksum_complete); 6264 6265 static void net_rps_send_ipi(struct softnet_data *remsd) 6266 { 6267 #ifdef CONFIG_RPS 6268 while (remsd) { 6269 struct softnet_data *next = remsd->rps_ipi_next; 6270 6271 if (cpu_online(remsd->cpu)) 6272 smp_call_function_single_async(remsd->cpu, &remsd->csd); 6273 remsd = next; 6274 } 6275 #endif 6276 } 6277 6278 /* 6279 * net_rps_action_and_irq_enable sends any pending IPI's for rps. 6280 * Note: called with local irq disabled, but exits with local irq enabled. 6281 */ 6282 static void net_rps_action_and_irq_enable(struct softnet_data *sd) 6283 { 6284 #ifdef CONFIG_RPS 6285 struct softnet_data *remsd = sd->rps_ipi_list; 6286 6287 if (remsd) { 6288 sd->rps_ipi_list = NULL; 6289 6290 local_irq_enable(); 6291 6292 /* Send pending IPI's to kick RPS processing on remote cpus. */ 6293 net_rps_send_ipi(remsd); 6294 } else 6295 #endif 6296 local_irq_enable(); 6297 } 6298 6299 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) 6300 { 6301 #ifdef CONFIG_RPS 6302 return sd->rps_ipi_list != NULL; 6303 #else 6304 return false; 6305 #endif 6306 } 6307 6308 static int process_backlog(struct napi_struct *napi, int quota) 6309 { 6310 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); 6311 bool again = true; 6312 int work = 0; 6313 6314 /* Check if we have pending ipi, its better to send them now, 6315 * not waiting net_rx_action() end. 6316 */ 6317 if (sd_has_rps_ipi_waiting(sd)) { 6318 local_irq_disable(); 6319 net_rps_action_and_irq_enable(sd); 6320 } 6321 6322 napi->weight = dev_rx_weight; 6323 while (again) { 6324 struct sk_buff *skb; 6325 6326 while ((skb = __skb_dequeue(&sd->process_queue))) { 6327 rcu_read_lock(); 6328 __netif_receive_skb(skb); 6329 rcu_read_unlock(); 6330 input_queue_head_incr(sd); 6331 if (++work >= quota) 6332 return work; 6333 6334 } 6335 6336 local_irq_disable(); 6337 rps_lock(sd); 6338 if (skb_queue_empty(&sd->input_pkt_queue)) { 6339 /* 6340 * Inline a custom version of __napi_complete(). 6341 * only current cpu owns and manipulates this napi, 6342 * and NAPI_STATE_SCHED is the only possible flag set 6343 * on backlog. 6344 * We can use a plain write instead of clear_bit(), 6345 * and we dont need an smp_mb() memory barrier. 6346 */ 6347 napi->state = 0; 6348 again = false; 6349 } else { 6350 skb_queue_splice_tail_init(&sd->input_pkt_queue, 6351 &sd->process_queue); 6352 } 6353 rps_unlock(sd); 6354 local_irq_enable(); 6355 } 6356 6357 return work; 6358 } 6359 6360 /** 6361 * __napi_schedule - schedule for receive 6362 * @n: entry to schedule 6363 * 6364 * The entry's receive function will be scheduled to run. 6365 * Consider using __napi_schedule_irqoff() if hard irqs are masked. 6366 */ 6367 void __napi_schedule(struct napi_struct *n) 6368 { 6369 unsigned long flags; 6370 6371 local_irq_save(flags); 6372 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6373 local_irq_restore(flags); 6374 } 6375 EXPORT_SYMBOL(__napi_schedule); 6376 6377 /** 6378 * napi_schedule_prep - check if napi can be scheduled 6379 * @n: napi context 6380 * 6381 * Test if NAPI routine is already running, and if not mark 6382 * it as running. This is used as a condition variable to 6383 * insure only one NAPI poll instance runs. We also make 6384 * sure there is no pending NAPI disable. 6385 */ 6386 bool napi_schedule_prep(struct napi_struct *n) 6387 { 6388 unsigned long val, new; 6389 6390 do { 6391 val = READ_ONCE(n->state); 6392 if (unlikely(val & NAPIF_STATE_DISABLE)) 6393 return false; 6394 new = val | NAPIF_STATE_SCHED; 6395 6396 /* Sets STATE_MISSED bit if STATE_SCHED was already set 6397 * This was suggested by Alexander Duyck, as compiler 6398 * emits better code than : 6399 * if (val & NAPIF_STATE_SCHED) 6400 * new |= NAPIF_STATE_MISSED; 6401 */ 6402 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * 6403 NAPIF_STATE_MISSED; 6404 } while (cmpxchg(&n->state, val, new) != val); 6405 6406 return !(val & NAPIF_STATE_SCHED); 6407 } 6408 EXPORT_SYMBOL(napi_schedule_prep); 6409 6410 /** 6411 * __napi_schedule_irqoff - schedule for receive 6412 * @n: entry to schedule 6413 * 6414 * Variant of __napi_schedule() assuming hard irqs are masked 6415 */ 6416 void __napi_schedule_irqoff(struct napi_struct *n) 6417 { 6418 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6419 } 6420 EXPORT_SYMBOL(__napi_schedule_irqoff); 6421 6422 bool napi_complete_done(struct napi_struct *n, int work_done) 6423 { 6424 unsigned long flags, val, new, timeout = 0; 6425 bool ret = true; 6426 6427 /* 6428 * 1) Don't let napi dequeue from the cpu poll list 6429 * just in case its running on a different cpu. 6430 * 2) If we are busy polling, do nothing here, we have 6431 * the guarantee we will be called later. 6432 */ 6433 if (unlikely(n->state & (NAPIF_STATE_NPSVC | 6434 NAPIF_STATE_IN_BUSY_POLL))) 6435 return false; 6436 6437 if (work_done) { 6438 if (n->gro_bitmask) 6439 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6440 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs); 6441 } 6442 if (n->defer_hard_irqs_count > 0) { 6443 n->defer_hard_irqs_count--; 6444 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6445 if (timeout) 6446 ret = false; 6447 } 6448 if (n->gro_bitmask) { 6449 /* When the NAPI instance uses a timeout and keeps postponing 6450 * it, we need to bound somehow the time packets are kept in 6451 * the GRO layer 6452 */ 6453 napi_gro_flush(n, !!timeout); 6454 } 6455 6456 gro_normal_list(n); 6457 6458 if (unlikely(!list_empty(&n->poll_list))) { 6459 /* If n->poll_list is not empty, we need to mask irqs */ 6460 local_irq_save(flags); 6461 list_del_init(&n->poll_list); 6462 local_irq_restore(flags); 6463 } 6464 6465 do { 6466 val = READ_ONCE(n->state); 6467 6468 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); 6469 6470 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED | 6471 NAPIF_STATE_PREFER_BUSY_POLL); 6472 6473 /* If STATE_MISSED was set, leave STATE_SCHED set, 6474 * because we will call napi->poll() one more time. 6475 * This C code was suggested by Alexander Duyck to help gcc. 6476 */ 6477 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * 6478 NAPIF_STATE_SCHED; 6479 } while (cmpxchg(&n->state, val, new) != val); 6480 6481 if (unlikely(val & NAPIF_STATE_MISSED)) { 6482 __napi_schedule(n); 6483 return false; 6484 } 6485 6486 if (timeout) 6487 hrtimer_start(&n->timer, ns_to_ktime(timeout), 6488 HRTIMER_MODE_REL_PINNED); 6489 return ret; 6490 } 6491 EXPORT_SYMBOL(napi_complete_done); 6492 6493 /* must be called under rcu_read_lock(), as we dont take a reference */ 6494 static struct napi_struct *napi_by_id(unsigned int napi_id) 6495 { 6496 unsigned int hash = napi_id % HASH_SIZE(napi_hash); 6497 struct napi_struct *napi; 6498 6499 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) 6500 if (napi->napi_id == napi_id) 6501 return napi; 6502 6503 return NULL; 6504 } 6505 6506 #if defined(CONFIG_NET_RX_BUSY_POLL) 6507 6508 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule) 6509 { 6510 if (!skip_schedule) { 6511 gro_normal_list(napi); 6512 __napi_schedule(napi); 6513 return; 6514 } 6515 6516 if (napi->gro_bitmask) { 6517 /* flush too old packets 6518 * If HZ < 1000, flush all packets. 6519 */ 6520 napi_gro_flush(napi, HZ >= 1000); 6521 } 6522 6523 gro_normal_list(napi); 6524 clear_bit(NAPI_STATE_SCHED, &napi->state); 6525 } 6526 6527 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll, 6528 u16 budget) 6529 { 6530 bool skip_schedule = false; 6531 unsigned long timeout; 6532 int rc; 6533 6534 /* Busy polling means there is a high chance device driver hard irq 6535 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was 6536 * set in napi_schedule_prep(). 6537 * Since we are about to call napi->poll() once more, we can safely 6538 * clear NAPI_STATE_MISSED. 6539 * 6540 * Note: x86 could use a single "lock and ..." instruction 6541 * to perform these two clear_bit() 6542 */ 6543 clear_bit(NAPI_STATE_MISSED, &napi->state); 6544 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); 6545 6546 local_bh_disable(); 6547 6548 if (prefer_busy_poll) { 6549 napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs); 6550 timeout = READ_ONCE(napi->dev->gro_flush_timeout); 6551 if (napi->defer_hard_irqs_count && timeout) { 6552 hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED); 6553 skip_schedule = true; 6554 } 6555 } 6556 6557 /* All we really want here is to re-enable device interrupts. 6558 * Ideally, a new ndo_busy_poll_stop() could avoid another round. 6559 */ 6560 rc = napi->poll(napi, budget); 6561 /* We can't gro_normal_list() here, because napi->poll() might have 6562 * rearmed the napi (napi_complete_done()) in which case it could 6563 * already be running on another CPU. 6564 */ 6565 trace_napi_poll(napi, rc, budget); 6566 netpoll_poll_unlock(have_poll_lock); 6567 if (rc == budget) 6568 __busy_poll_stop(napi, skip_schedule); 6569 local_bh_enable(); 6570 } 6571 6572 void napi_busy_loop(unsigned int napi_id, 6573 bool (*loop_end)(void *, unsigned long), 6574 void *loop_end_arg, bool prefer_busy_poll, u16 budget) 6575 { 6576 unsigned long start_time = loop_end ? busy_loop_current_time() : 0; 6577 int (*napi_poll)(struct napi_struct *napi, int budget); 6578 void *have_poll_lock = NULL; 6579 struct napi_struct *napi; 6580 6581 restart: 6582 napi_poll = NULL; 6583 6584 rcu_read_lock(); 6585 6586 napi = napi_by_id(napi_id); 6587 if (!napi) 6588 goto out; 6589 6590 preempt_disable(); 6591 for (;;) { 6592 int work = 0; 6593 6594 local_bh_disable(); 6595 if (!napi_poll) { 6596 unsigned long val = READ_ONCE(napi->state); 6597 6598 /* If multiple threads are competing for this napi, 6599 * we avoid dirtying napi->state as much as we can. 6600 */ 6601 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | 6602 NAPIF_STATE_IN_BUSY_POLL)) { 6603 if (prefer_busy_poll) 6604 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6605 goto count; 6606 } 6607 if (cmpxchg(&napi->state, val, 6608 val | NAPIF_STATE_IN_BUSY_POLL | 6609 NAPIF_STATE_SCHED) != val) { 6610 if (prefer_busy_poll) 6611 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6612 goto count; 6613 } 6614 have_poll_lock = netpoll_poll_lock(napi); 6615 napi_poll = napi->poll; 6616 } 6617 work = napi_poll(napi, budget); 6618 trace_napi_poll(napi, work, budget); 6619 gro_normal_list(napi); 6620 count: 6621 if (work > 0) 6622 __NET_ADD_STATS(dev_net(napi->dev), 6623 LINUX_MIB_BUSYPOLLRXPACKETS, work); 6624 local_bh_enable(); 6625 6626 if (!loop_end || loop_end(loop_end_arg, start_time)) 6627 break; 6628 6629 if (unlikely(need_resched())) { 6630 if (napi_poll) 6631 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6632 preempt_enable(); 6633 rcu_read_unlock(); 6634 cond_resched(); 6635 if (loop_end(loop_end_arg, start_time)) 6636 return; 6637 goto restart; 6638 } 6639 cpu_relax(); 6640 } 6641 if (napi_poll) 6642 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6643 preempt_enable(); 6644 out: 6645 rcu_read_unlock(); 6646 } 6647 EXPORT_SYMBOL(napi_busy_loop); 6648 6649 #endif /* CONFIG_NET_RX_BUSY_POLL */ 6650 6651 static void napi_hash_add(struct napi_struct *napi) 6652 { 6653 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state)) 6654 return; 6655 6656 spin_lock(&napi_hash_lock); 6657 6658 /* 0..NR_CPUS range is reserved for sender_cpu use */ 6659 do { 6660 if (unlikely(++napi_gen_id < MIN_NAPI_ID)) 6661 napi_gen_id = MIN_NAPI_ID; 6662 } while (napi_by_id(napi_gen_id)); 6663 napi->napi_id = napi_gen_id; 6664 6665 hlist_add_head_rcu(&napi->napi_hash_node, 6666 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); 6667 6668 spin_unlock(&napi_hash_lock); 6669 } 6670 6671 /* Warning : caller is responsible to make sure rcu grace period 6672 * is respected before freeing memory containing @napi 6673 */ 6674 static void napi_hash_del(struct napi_struct *napi) 6675 { 6676 spin_lock(&napi_hash_lock); 6677 6678 hlist_del_init_rcu(&napi->napi_hash_node); 6679 6680 spin_unlock(&napi_hash_lock); 6681 } 6682 6683 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) 6684 { 6685 struct napi_struct *napi; 6686 6687 napi = container_of(timer, struct napi_struct, timer); 6688 6689 /* Note : we use a relaxed variant of napi_schedule_prep() not setting 6690 * NAPI_STATE_MISSED, since we do not react to a device IRQ. 6691 */ 6692 if (!napi_disable_pending(napi) && 6693 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) { 6694 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6695 __napi_schedule_irqoff(napi); 6696 } 6697 6698 return HRTIMER_NORESTART; 6699 } 6700 6701 static void init_gro_hash(struct napi_struct *napi) 6702 { 6703 int i; 6704 6705 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6706 INIT_LIST_HEAD(&napi->gro_hash[i].list); 6707 napi->gro_hash[i].count = 0; 6708 } 6709 napi->gro_bitmask = 0; 6710 } 6711 6712 void netif_napi_add(struct net_device *dev, struct napi_struct *napi, 6713 int (*poll)(struct napi_struct *, int), int weight) 6714 { 6715 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state))) 6716 return; 6717 6718 INIT_LIST_HEAD(&napi->poll_list); 6719 INIT_HLIST_NODE(&napi->napi_hash_node); 6720 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); 6721 napi->timer.function = napi_watchdog; 6722 init_gro_hash(napi); 6723 napi->skb = NULL; 6724 INIT_LIST_HEAD(&napi->rx_list); 6725 napi->rx_count = 0; 6726 napi->poll = poll; 6727 if (weight > NAPI_POLL_WEIGHT) 6728 netdev_err_once(dev, "%s() called with weight %d\n", __func__, 6729 weight); 6730 napi->weight = weight; 6731 napi->dev = dev; 6732 #ifdef CONFIG_NETPOLL 6733 napi->poll_owner = -1; 6734 #endif 6735 set_bit(NAPI_STATE_SCHED, &napi->state); 6736 set_bit(NAPI_STATE_NPSVC, &napi->state); 6737 list_add_rcu(&napi->dev_list, &dev->napi_list); 6738 napi_hash_add(napi); 6739 } 6740 EXPORT_SYMBOL(netif_napi_add); 6741 6742 void napi_disable(struct napi_struct *n) 6743 { 6744 might_sleep(); 6745 set_bit(NAPI_STATE_DISABLE, &n->state); 6746 6747 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state)) 6748 msleep(1); 6749 while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state)) 6750 msleep(1); 6751 6752 hrtimer_cancel(&n->timer); 6753 6754 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &n->state); 6755 clear_bit(NAPI_STATE_DISABLE, &n->state); 6756 } 6757 EXPORT_SYMBOL(napi_disable); 6758 6759 static void flush_gro_hash(struct napi_struct *napi) 6760 { 6761 int i; 6762 6763 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6764 struct sk_buff *skb, *n; 6765 6766 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list) 6767 kfree_skb(skb); 6768 napi->gro_hash[i].count = 0; 6769 } 6770 } 6771 6772 /* Must be called in process context */ 6773 void __netif_napi_del(struct napi_struct *napi) 6774 { 6775 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state)) 6776 return; 6777 6778 napi_hash_del(napi); 6779 list_del_rcu(&napi->dev_list); 6780 napi_free_frags(napi); 6781 6782 flush_gro_hash(napi); 6783 napi->gro_bitmask = 0; 6784 } 6785 EXPORT_SYMBOL(__netif_napi_del); 6786 6787 static int napi_poll(struct napi_struct *n, struct list_head *repoll) 6788 { 6789 void *have; 6790 int work, weight; 6791 6792 list_del_init(&n->poll_list); 6793 6794 have = netpoll_poll_lock(n); 6795 6796 weight = n->weight; 6797 6798 /* This NAPI_STATE_SCHED test is for avoiding a race 6799 * with netpoll's poll_napi(). Only the entity which 6800 * obtains the lock and sees NAPI_STATE_SCHED set will 6801 * actually make the ->poll() call. Therefore we avoid 6802 * accidentally calling ->poll() when NAPI is not scheduled. 6803 */ 6804 work = 0; 6805 if (test_bit(NAPI_STATE_SCHED, &n->state)) { 6806 work = n->poll(n, weight); 6807 trace_napi_poll(n, work, weight); 6808 } 6809 6810 if (unlikely(work > weight)) 6811 pr_err_once("NAPI poll function %pS returned %d, exceeding its budget of %d.\n", 6812 n->poll, work, weight); 6813 6814 if (likely(work < weight)) 6815 goto out_unlock; 6816 6817 /* Drivers must not modify the NAPI state if they 6818 * consume the entire weight. In such cases this code 6819 * still "owns" the NAPI instance and therefore can 6820 * move the instance around on the list at-will. 6821 */ 6822 if (unlikely(napi_disable_pending(n))) { 6823 napi_complete(n); 6824 goto out_unlock; 6825 } 6826 6827 /* The NAPI context has more processing work, but busy-polling 6828 * is preferred. Exit early. 6829 */ 6830 if (napi_prefer_busy_poll(n)) { 6831 if (napi_complete_done(n, work)) { 6832 /* If timeout is not set, we need to make sure 6833 * that the NAPI is re-scheduled. 6834 */ 6835 napi_schedule(n); 6836 } 6837 goto out_unlock; 6838 } 6839 6840 if (n->gro_bitmask) { 6841 /* flush too old packets 6842 * If HZ < 1000, flush all packets. 6843 */ 6844 napi_gro_flush(n, HZ >= 1000); 6845 } 6846 6847 gro_normal_list(n); 6848 6849 /* Some drivers may have called napi_schedule 6850 * prior to exhausting their budget. 6851 */ 6852 if (unlikely(!list_empty(&n->poll_list))) { 6853 pr_warn_once("%s: Budget exhausted after napi rescheduled\n", 6854 n->dev ? n->dev->name : "backlog"); 6855 goto out_unlock; 6856 } 6857 6858 list_add_tail(&n->poll_list, repoll); 6859 6860 out_unlock: 6861 netpoll_poll_unlock(have); 6862 6863 return work; 6864 } 6865 6866 static __latent_entropy void net_rx_action(struct softirq_action *h) 6867 { 6868 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 6869 unsigned long time_limit = jiffies + 6870 usecs_to_jiffies(netdev_budget_usecs); 6871 int budget = netdev_budget; 6872 LIST_HEAD(list); 6873 LIST_HEAD(repoll); 6874 6875 local_irq_disable(); 6876 list_splice_init(&sd->poll_list, &list); 6877 local_irq_enable(); 6878 6879 for (;;) { 6880 struct napi_struct *n; 6881 6882 if (list_empty(&list)) { 6883 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll)) 6884 goto out; 6885 break; 6886 } 6887 6888 n = list_first_entry(&list, struct napi_struct, poll_list); 6889 budget -= napi_poll(n, &repoll); 6890 6891 /* If softirq window is exhausted then punt. 6892 * Allow this to run for 2 jiffies since which will allow 6893 * an average latency of 1.5/HZ. 6894 */ 6895 if (unlikely(budget <= 0 || 6896 time_after_eq(jiffies, time_limit))) { 6897 sd->time_squeeze++; 6898 break; 6899 } 6900 } 6901 6902 local_irq_disable(); 6903 6904 list_splice_tail_init(&sd->poll_list, &list); 6905 list_splice_tail(&repoll, &list); 6906 list_splice(&list, &sd->poll_list); 6907 if (!list_empty(&sd->poll_list)) 6908 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 6909 6910 net_rps_action_and_irq_enable(sd); 6911 out: 6912 __kfree_skb_flush(); 6913 } 6914 6915 struct netdev_adjacent { 6916 struct net_device *dev; 6917 6918 /* upper master flag, there can only be one master device per list */ 6919 bool master; 6920 6921 /* lookup ignore flag */ 6922 bool ignore; 6923 6924 /* counter for the number of times this device was added to us */ 6925 u16 ref_nr; 6926 6927 /* private field for the users */ 6928 void *private; 6929 6930 struct list_head list; 6931 struct rcu_head rcu; 6932 }; 6933 6934 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, 6935 struct list_head *adj_list) 6936 { 6937 struct netdev_adjacent *adj; 6938 6939 list_for_each_entry(adj, adj_list, list) { 6940 if (adj->dev == adj_dev) 6941 return adj; 6942 } 6943 return NULL; 6944 } 6945 6946 static int ____netdev_has_upper_dev(struct net_device *upper_dev, 6947 struct netdev_nested_priv *priv) 6948 { 6949 struct net_device *dev = (struct net_device *)priv->data; 6950 6951 return upper_dev == dev; 6952 } 6953 6954 /** 6955 * netdev_has_upper_dev - Check if device is linked to an upper device 6956 * @dev: device 6957 * @upper_dev: upper device to check 6958 * 6959 * Find out if a device is linked to specified upper device and return true 6960 * in case it is. Note that this checks only immediate upper device, 6961 * not through a complete stack of devices. The caller must hold the RTNL lock. 6962 */ 6963 bool netdev_has_upper_dev(struct net_device *dev, 6964 struct net_device *upper_dev) 6965 { 6966 struct netdev_nested_priv priv = { 6967 .data = (void *)upper_dev, 6968 }; 6969 6970 ASSERT_RTNL(); 6971 6972 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 6973 &priv); 6974 } 6975 EXPORT_SYMBOL(netdev_has_upper_dev); 6976 6977 /** 6978 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device 6979 * @dev: device 6980 * @upper_dev: upper device to check 6981 * 6982 * Find out if a device is linked to specified upper device and return true 6983 * in case it is. Note that this checks the entire upper device chain. 6984 * The caller must hold rcu lock. 6985 */ 6986 6987 bool netdev_has_upper_dev_all_rcu(struct net_device *dev, 6988 struct net_device *upper_dev) 6989 { 6990 struct netdev_nested_priv priv = { 6991 .data = (void *)upper_dev, 6992 }; 6993 6994 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 6995 &priv); 6996 } 6997 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); 6998 6999 /** 7000 * netdev_has_any_upper_dev - Check if device is linked to some device 7001 * @dev: device 7002 * 7003 * Find out if a device is linked to an upper device and return true in case 7004 * it is. The caller must hold the RTNL lock. 7005 */ 7006 bool netdev_has_any_upper_dev(struct net_device *dev) 7007 { 7008 ASSERT_RTNL(); 7009 7010 return !list_empty(&dev->adj_list.upper); 7011 } 7012 EXPORT_SYMBOL(netdev_has_any_upper_dev); 7013 7014 /** 7015 * netdev_master_upper_dev_get - Get master upper device 7016 * @dev: device 7017 * 7018 * Find a master upper device and return pointer to it or NULL in case 7019 * it's not there. The caller must hold the RTNL lock. 7020 */ 7021 struct net_device *netdev_master_upper_dev_get(struct net_device *dev) 7022 { 7023 struct netdev_adjacent *upper; 7024 7025 ASSERT_RTNL(); 7026 7027 if (list_empty(&dev->adj_list.upper)) 7028 return NULL; 7029 7030 upper = list_first_entry(&dev->adj_list.upper, 7031 struct netdev_adjacent, list); 7032 if (likely(upper->master)) 7033 return upper->dev; 7034 return NULL; 7035 } 7036 EXPORT_SYMBOL(netdev_master_upper_dev_get); 7037 7038 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev) 7039 { 7040 struct netdev_adjacent *upper; 7041 7042 ASSERT_RTNL(); 7043 7044 if (list_empty(&dev->adj_list.upper)) 7045 return NULL; 7046 7047 upper = list_first_entry(&dev->adj_list.upper, 7048 struct netdev_adjacent, list); 7049 if (likely(upper->master) && !upper->ignore) 7050 return upper->dev; 7051 return NULL; 7052 } 7053 7054 /** 7055 * netdev_has_any_lower_dev - Check if device is linked to some device 7056 * @dev: device 7057 * 7058 * Find out if a device is linked to a lower device and return true in case 7059 * it is. The caller must hold the RTNL lock. 7060 */ 7061 static bool netdev_has_any_lower_dev(struct net_device *dev) 7062 { 7063 ASSERT_RTNL(); 7064 7065 return !list_empty(&dev->adj_list.lower); 7066 } 7067 7068 void *netdev_adjacent_get_private(struct list_head *adj_list) 7069 { 7070 struct netdev_adjacent *adj; 7071 7072 adj = list_entry(adj_list, struct netdev_adjacent, list); 7073 7074 return adj->private; 7075 } 7076 EXPORT_SYMBOL(netdev_adjacent_get_private); 7077 7078 /** 7079 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list 7080 * @dev: device 7081 * @iter: list_head ** of the current position 7082 * 7083 * Gets the next device from the dev's upper list, starting from iter 7084 * position. The caller must hold RCU read lock. 7085 */ 7086 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, 7087 struct list_head **iter) 7088 { 7089 struct netdev_adjacent *upper; 7090 7091 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 7092 7093 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7094 7095 if (&upper->list == &dev->adj_list.upper) 7096 return NULL; 7097 7098 *iter = &upper->list; 7099 7100 return upper->dev; 7101 } 7102 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); 7103 7104 static struct net_device *__netdev_next_upper_dev(struct net_device *dev, 7105 struct list_head **iter, 7106 bool *ignore) 7107 { 7108 struct netdev_adjacent *upper; 7109 7110 upper = list_entry((*iter)->next, struct netdev_adjacent, list); 7111 7112 if (&upper->list == &dev->adj_list.upper) 7113 return NULL; 7114 7115 *iter = &upper->list; 7116 *ignore = upper->ignore; 7117 7118 return upper->dev; 7119 } 7120 7121 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, 7122 struct list_head **iter) 7123 { 7124 struct netdev_adjacent *upper; 7125 7126 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 7127 7128 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7129 7130 if (&upper->list == &dev->adj_list.upper) 7131 return NULL; 7132 7133 *iter = &upper->list; 7134 7135 return upper->dev; 7136 } 7137 7138 static int __netdev_walk_all_upper_dev(struct net_device *dev, 7139 int (*fn)(struct net_device *dev, 7140 struct netdev_nested_priv *priv), 7141 struct netdev_nested_priv *priv) 7142 { 7143 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7144 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7145 int ret, cur = 0; 7146 bool ignore; 7147 7148 now = dev; 7149 iter = &dev->adj_list.upper; 7150 7151 while (1) { 7152 if (now != dev) { 7153 ret = fn(now, priv); 7154 if (ret) 7155 return ret; 7156 } 7157 7158 next = NULL; 7159 while (1) { 7160 udev = __netdev_next_upper_dev(now, &iter, &ignore); 7161 if (!udev) 7162 break; 7163 if (ignore) 7164 continue; 7165 7166 next = udev; 7167 niter = &udev->adj_list.upper; 7168 dev_stack[cur] = now; 7169 iter_stack[cur++] = iter; 7170 break; 7171 } 7172 7173 if (!next) { 7174 if (!cur) 7175 return 0; 7176 next = dev_stack[--cur]; 7177 niter = iter_stack[cur]; 7178 } 7179 7180 now = next; 7181 iter = niter; 7182 } 7183 7184 return 0; 7185 } 7186 7187 int netdev_walk_all_upper_dev_rcu(struct net_device *dev, 7188 int (*fn)(struct net_device *dev, 7189 struct netdev_nested_priv *priv), 7190 struct netdev_nested_priv *priv) 7191 { 7192 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7193 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7194 int ret, cur = 0; 7195 7196 now = dev; 7197 iter = &dev->adj_list.upper; 7198 7199 while (1) { 7200 if (now != dev) { 7201 ret = fn(now, priv); 7202 if (ret) 7203 return ret; 7204 } 7205 7206 next = NULL; 7207 while (1) { 7208 udev = netdev_next_upper_dev_rcu(now, &iter); 7209 if (!udev) 7210 break; 7211 7212 next = udev; 7213 niter = &udev->adj_list.upper; 7214 dev_stack[cur] = now; 7215 iter_stack[cur++] = iter; 7216 break; 7217 } 7218 7219 if (!next) { 7220 if (!cur) 7221 return 0; 7222 next = dev_stack[--cur]; 7223 niter = iter_stack[cur]; 7224 } 7225 7226 now = next; 7227 iter = niter; 7228 } 7229 7230 return 0; 7231 } 7232 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); 7233 7234 static bool __netdev_has_upper_dev(struct net_device *dev, 7235 struct net_device *upper_dev) 7236 { 7237 struct netdev_nested_priv priv = { 7238 .flags = 0, 7239 .data = (void *)upper_dev, 7240 }; 7241 7242 ASSERT_RTNL(); 7243 7244 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev, 7245 &priv); 7246 } 7247 7248 /** 7249 * netdev_lower_get_next_private - Get the next ->private from the 7250 * lower neighbour list 7251 * @dev: device 7252 * @iter: list_head ** of the current position 7253 * 7254 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7255 * list, starting from iter position. The caller must hold either hold the 7256 * RTNL lock or its own locking that guarantees that the neighbour lower 7257 * list will remain unchanged. 7258 */ 7259 void *netdev_lower_get_next_private(struct net_device *dev, 7260 struct list_head **iter) 7261 { 7262 struct netdev_adjacent *lower; 7263 7264 lower = list_entry(*iter, struct netdev_adjacent, list); 7265 7266 if (&lower->list == &dev->adj_list.lower) 7267 return NULL; 7268 7269 *iter = lower->list.next; 7270 7271 return lower->private; 7272 } 7273 EXPORT_SYMBOL(netdev_lower_get_next_private); 7274 7275 /** 7276 * netdev_lower_get_next_private_rcu - Get the next ->private from the 7277 * lower neighbour list, RCU 7278 * variant 7279 * @dev: device 7280 * @iter: list_head ** of the current position 7281 * 7282 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7283 * list, starting from iter position. The caller must hold RCU read lock. 7284 */ 7285 void *netdev_lower_get_next_private_rcu(struct net_device *dev, 7286 struct list_head **iter) 7287 { 7288 struct netdev_adjacent *lower; 7289 7290 WARN_ON_ONCE(!rcu_read_lock_held()); 7291 7292 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7293 7294 if (&lower->list == &dev->adj_list.lower) 7295 return NULL; 7296 7297 *iter = &lower->list; 7298 7299 return lower->private; 7300 } 7301 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); 7302 7303 /** 7304 * netdev_lower_get_next - Get the next device from the lower neighbour 7305 * list 7306 * @dev: device 7307 * @iter: list_head ** of the current position 7308 * 7309 * Gets the next netdev_adjacent from the dev's lower neighbour 7310 * list, starting from iter position. The caller must hold RTNL lock or 7311 * its own locking that guarantees that the neighbour lower 7312 * list will remain unchanged. 7313 */ 7314 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) 7315 { 7316 struct netdev_adjacent *lower; 7317 7318 lower = list_entry(*iter, struct netdev_adjacent, list); 7319 7320 if (&lower->list == &dev->adj_list.lower) 7321 return NULL; 7322 7323 *iter = lower->list.next; 7324 7325 return lower->dev; 7326 } 7327 EXPORT_SYMBOL(netdev_lower_get_next); 7328 7329 static struct net_device *netdev_next_lower_dev(struct net_device *dev, 7330 struct list_head **iter) 7331 { 7332 struct netdev_adjacent *lower; 7333 7334 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7335 7336 if (&lower->list == &dev->adj_list.lower) 7337 return NULL; 7338 7339 *iter = &lower->list; 7340 7341 return lower->dev; 7342 } 7343 7344 static struct net_device *__netdev_next_lower_dev(struct net_device *dev, 7345 struct list_head **iter, 7346 bool *ignore) 7347 { 7348 struct netdev_adjacent *lower; 7349 7350 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7351 7352 if (&lower->list == &dev->adj_list.lower) 7353 return NULL; 7354 7355 *iter = &lower->list; 7356 *ignore = lower->ignore; 7357 7358 return lower->dev; 7359 } 7360 7361 int netdev_walk_all_lower_dev(struct net_device *dev, 7362 int (*fn)(struct net_device *dev, 7363 struct netdev_nested_priv *priv), 7364 struct netdev_nested_priv *priv) 7365 { 7366 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7367 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7368 int ret, cur = 0; 7369 7370 now = dev; 7371 iter = &dev->adj_list.lower; 7372 7373 while (1) { 7374 if (now != dev) { 7375 ret = fn(now, priv); 7376 if (ret) 7377 return ret; 7378 } 7379 7380 next = NULL; 7381 while (1) { 7382 ldev = netdev_next_lower_dev(now, &iter); 7383 if (!ldev) 7384 break; 7385 7386 next = ldev; 7387 niter = &ldev->adj_list.lower; 7388 dev_stack[cur] = now; 7389 iter_stack[cur++] = iter; 7390 break; 7391 } 7392 7393 if (!next) { 7394 if (!cur) 7395 return 0; 7396 next = dev_stack[--cur]; 7397 niter = iter_stack[cur]; 7398 } 7399 7400 now = next; 7401 iter = niter; 7402 } 7403 7404 return 0; 7405 } 7406 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); 7407 7408 static int __netdev_walk_all_lower_dev(struct net_device *dev, 7409 int (*fn)(struct net_device *dev, 7410 struct netdev_nested_priv *priv), 7411 struct netdev_nested_priv *priv) 7412 { 7413 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7414 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7415 int ret, cur = 0; 7416 bool ignore; 7417 7418 now = dev; 7419 iter = &dev->adj_list.lower; 7420 7421 while (1) { 7422 if (now != dev) { 7423 ret = fn(now, priv); 7424 if (ret) 7425 return ret; 7426 } 7427 7428 next = NULL; 7429 while (1) { 7430 ldev = __netdev_next_lower_dev(now, &iter, &ignore); 7431 if (!ldev) 7432 break; 7433 if (ignore) 7434 continue; 7435 7436 next = ldev; 7437 niter = &ldev->adj_list.lower; 7438 dev_stack[cur] = now; 7439 iter_stack[cur++] = iter; 7440 break; 7441 } 7442 7443 if (!next) { 7444 if (!cur) 7445 return 0; 7446 next = dev_stack[--cur]; 7447 niter = iter_stack[cur]; 7448 } 7449 7450 now = next; 7451 iter = niter; 7452 } 7453 7454 return 0; 7455 } 7456 7457 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, 7458 struct list_head **iter) 7459 { 7460 struct netdev_adjacent *lower; 7461 7462 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7463 if (&lower->list == &dev->adj_list.lower) 7464 return NULL; 7465 7466 *iter = &lower->list; 7467 7468 return lower->dev; 7469 } 7470 EXPORT_SYMBOL(netdev_next_lower_dev_rcu); 7471 7472 static u8 __netdev_upper_depth(struct net_device *dev) 7473 { 7474 struct net_device *udev; 7475 struct list_head *iter; 7476 u8 max_depth = 0; 7477 bool ignore; 7478 7479 for (iter = &dev->adj_list.upper, 7480 udev = __netdev_next_upper_dev(dev, &iter, &ignore); 7481 udev; 7482 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) { 7483 if (ignore) 7484 continue; 7485 if (max_depth < udev->upper_level) 7486 max_depth = udev->upper_level; 7487 } 7488 7489 return max_depth; 7490 } 7491 7492 static u8 __netdev_lower_depth(struct net_device *dev) 7493 { 7494 struct net_device *ldev; 7495 struct list_head *iter; 7496 u8 max_depth = 0; 7497 bool ignore; 7498 7499 for (iter = &dev->adj_list.lower, 7500 ldev = __netdev_next_lower_dev(dev, &iter, &ignore); 7501 ldev; 7502 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) { 7503 if (ignore) 7504 continue; 7505 if (max_depth < ldev->lower_level) 7506 max_depth = ldev->lower_level; 7507 } 7508 7509 return max_depth; 7510 } 7511 7512 static int __netdev_update_upper_level(struct net_device *dev, 7513 struct netdev_nested_priv *__unused) 7514 { 7515 dev->upper_level = __netdev_upper_depth(dev) + 1; 7516 return 0; 7517 } 7518 7519 static int __netdev_update_lower_level(struct net_device *dev, 7520 struct netdev_nested_priv *priv) 7521 { 7522 dev->lower_level = __netdev_lower_depth(dev) + 1; 7523 7524 #ifdef CONFIG_LOCKDEP 7525 if (!priv) 7526 return 0; 7527 7528 if (priv->flags & NESTED_SYNC_IMM) 7529 dev->nested_level = dev->lower_level - 1; 7530 if (priv->flags & NESTED_SYNC_TODO) 7531 net_unlink_todo(dev); 7532 #endif 7533 return 0; 7534 } 7535 7536 int netdev_walk_all_lower_dev_rcu(struct net_device *dev, 7537 int (*fn)(struct net_device *dev, 7538 struct netdev_nested_priv *priv), 7539 struct netdev_nested_priv *priv) 7540 { 7541 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7542 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7543 int ret, cur = 0; 7544 7545 now = dev; 7546 iter = &dev->adj_list.lower; 7547 7548 while (1) { 7549 if (now != dev) { 7550 ret = fn(now, priv); 7551 if (ret) 7552 return ret; 7553 } 7554 7555 next = NULL; 7556 while (1) { 7557 ldev = netdev_next_lower_dev_rcu(now, &iter); 7558 if (!ldev) 7559 break; 7560 7561 next = ldev; 7562 niter = &ldev->adj_list.lower; 7563 dev_stack[cur] = now; 7564 iter_stack[cur++] = iter; 7565 break; 7566 } 7567 7568 if (!next) { 7569 if (!cur) 7570 return 0; 7571 next = dev_stack[--cur]; 7572 niter = iter_stack[cur]; 7573 } 7574 7575 now = next; 7576 iter = niter; 7577 } 7578 7579 return 0; 7580 } 7581 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); 7582 7583 /** 7584 * netdev_lower_get_first_private_rcu - Get the first ->private from the 7585 * lower neighbour list, RCU 7586 * variant 7587 * @dev: device 7588 * 7589 * Gets the first netdev_adjacent->private from the dev's lower neighbour 7590 * list. The caller must hold RCU read lock. 7591 */ 7592 void *netdev_lower_get_first_private_rcu(struct net_device *dev) 7593 { 7594 struct netdev_adjacent *lower; 7595 7596 lower = list_first_or_null_rcu(&dev->adj_list.lower, 7597 struct netdev_adjacent, list); 7598 if (lower) 7599 return lower->private; 7600 return NULL; 7601 } 7602 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); 7603 7604 /** 7605 * netdev_master_upper_dev_get_rcu - Get master upper device 7606 * @dev: device 7607 * 7608 * Find a master upper device and return pointer to it or NULL in case 7609 * it's not there. The caller must hold the RCU read lock. 7610 */ 7611 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) 7612 { 7613 struct netdev_adjacent *upper; 7614 7615 upper = list_first_or_null_rcu(&dev->adj_list.upper, 7616 struct netdev_adjacent, list); 7617 if (upper && likely(upper->master)) 7618 return upper->dev; 7619 return NULL; 7620 } 7621 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); 7622 7623 static int netdev_adjacent_sysfs_add(struct net_device *dev, 7624 struct net_device *adj_dev, 7625 struct list_head *dev_list) 7626 { 7627 char linkname[IFNAMSIZ+7]; 7628 7629 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7630 "upper_%s" : "lower_%s", adj_dev->name); 7631 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), 7632 linkname); 7633 } 7634 static void netdev_adjacent_sysfs_del(struct net_device *dev, 7635 char *name, 7636 struct list_head *dev_list) 7637 { 7638 char linkname[IFNAMSIZ+7]; 7639 7640 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7641 "upper_%s" : "lower_%s", name); 7642 sysfs_remove_link(&(dev->dev.kobj), linkname); 7643 } 7644 7645 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, 7646 struct net_device *adj_dev, 7647 struct list_head *dev_list) 7648 { 7649 return (dev_list == &dev->adj_list.upper || 7650 dev_list == &dev->adj_list.lower) && 7651 net_eq(dev_net(dev), dev_net(adj_dev)); 7652 } 7653 7654 static int __netdev_adjacent_dev_insert(struct net_device *dev, 7655 struct net_device *adj_dev, 7656 struct list_head *dev_list, 7657 void *private, bool master) 7658 { 7659 struct netdev_adjacent *adj; 7660 int ret; 7661 7662 adj = __netdev_find_adj(adj_dev, dev_list); 7663 7664 if (adj) { 7665 adj->ref_nr += 1; 7666 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", 7667 dev->name, adj_dev->name, adj->ref_nr); 7668 7669 return 0; 7670 } 7671 7672 adj = kmalloc(sizeof(*adj), GFP_KERNEL); 7673 if (!adj) 7674 return -ENOMEM; 7675 7676 adj->dev = adj_dev; 7677 adj->master = master; 7678 adj->ref_nr = 1; 7679 adj->private = private; 7680 adj->ignore = false; 7681 dev_hold(adj_dev); 7682 7683 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", 7684 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); 7685 7686 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { 7687 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); 7688 if (ret) 7689 goto free_adj; 7690 } 7691 7692 /* Ensure that master link is always the first item in list. */ 7693 if (master) { 7694 ret = sysfs_create_link(&(dev->dev.kobj), 7695 &(adj_dev->dev.kobj), "master"); 7696 if (ret) 7697 goto remove_symlinks; 7698 7699 list_add_rcu(&adj->list, dev_list); 7700 } else { 7701 list_add_tail_rcu(&adj->list, dev_list); 7702 } 7703 7704 return 0; 7705 7706 remove_symlinks: 7707 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7708 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7709 free_adj: 7710 kfree(adj); 7711 dev_put(adj_dev); 7712 7713 return ret; 7714 } 7715 7716 static void __netdev_adjacent_dev_remove(struct net_device *dev, 7717 struct net_device *adj_dev, 7718 u16 ref_nr, 7719 struct list_head *dev_list) 7720 { 7721 struct netdev_adjacent *adj; 7722 7723 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", 7724 dev->name, adj_dev->name, ref_nr); 7725 7726 adj = __netdev_find_adj(adj_dev, dev_list); 7727 7728 if (!adj) { 7729 pr_err("Adjacency does not exist for device %s from %s\n", 7730 dev->name, adj_dev->name); 7731 WARN_ON(1); 7732 return; 7733 } 7734 7735 if (adj->ref_nr > ref_nr) { 7736 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", 7737 dev->name, adj_dev->name, ref_nr, 7738 adj->ref_nr - ref_nr); 7739 adj->ref_nr -= ref_nr; 7740 return; 7741 } 7742 7743 if (adj->master) 7744 sysfs_remove_link(&(dev->dev.kobj), "master"); 7745 7746 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7747 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7748 7749 list_del_rcu(&adj->list); 7750 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", 7751 adj_dev->name, dev->name, adj_dev->name); 7752 dev_put(adj_dev); 7753 kfree_rcu(adj, rcu); 7754 } 7755 7756 static int __netdev_adjacent_dev_link_lists(struct net_device *dev, 7757 struct net_device *upper_dev, 7758 struct list_head *up_list, 7759 struct list_head *down_list, 7760 void *private, bool master) 7761 { 7762 int ret; 7763 7764 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, 7765 private, master); 7766 if (ret) 7767 return ret; 7768 7769 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, 7770 private, false); 7771 if (ret) { 7772 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); 7773 return ret; 7774 } 7775 7776 return 0; 7777 } 7778 7779 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, 7780 struct net_device *upper_dev, 7781 u16 ref_nr, 7782 struct list_head *up_list, 7783 struct list_head *down_list) 7784 { 7785 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); 7786 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); 7787 } 7788 7789 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, 7790 struct net_device *upper_dev, 7791 void *private, bool master) 7792 { 7793 return __netdev_adjacent_dev_link_lists(dev, upper_dev, 7794 &dev->adj_list.upper, 7795 &upper_dev->adj_list.lower, 7796 private, master); 7797 } 7798 7799 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, 7800 struct net_device *upper_dev) 7801 { 7802 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, 7803 &dev->adj_list.upper, 7804 &upper_dev->adj_list.lower); 7805 } 7806 7807 static int __netdev_upper_dev_link(struct net_device *dev, 7808 struct net_device *upper_dev, bool master, 7809 void *upper_priv, void *upper_info, 7810 struct netdev_nested_priv *priv, 7811 struct netlink_ext_ack *extack) 7812 { 7813 struct netdev_notifier_changeupper_info changeupper_info = { 7814 .info = { 7815 .dev = dev, 7816 .extack = extack, 7817 }, 7818 .upper_dev = upper_dev, 7819 .master = master, 7820 .linking = true, 7821 .upper_info = upper_info, 7822 }; 7823 struct net_device *master_dev; 7824 int ret = 0; 7825 7826 ASSERT_RTNL(); 7827 7828 if (dev == upper_dev) 7829 return -EBUSY; 7830 7831 /* To prevent loops, check if dev is not upper device to upper_dev. */ 7832 if (__netdev_has_upper_dev(upper_dev, dev)) 7833 return -EBUSY; 7834 7835 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV) 7836 return -EMLINK; 7837 7838 if (!master) { 7839 if (__netdev_has_upper_dev(dev, upper_dev)) 7840 return -EEXIST; 7841 } else { 7842 master_dev = __netdev_master_upper_dev_get(dev); 7843 if (master_dev) 7844 return master_dev == upper_dev ? -EEXIST : -EBUSY; 7845 } 7846 7847 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7848 &changeupper_info.info); 7849 ret = notifier_to_errno(ret); 7850 if (ret) 7851 return ret; 7852 7853 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, 7854 master); 7855 if (ret) 7856 return ret; 7857 7858 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 7859 &changeupper_info.info); 7860 ret = notifier_to_errno(ret); 7861 if (ret) 7862 goto rollback; 7863 7864 __netdev_update_upper_level(dev, NULL); 7865 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 7866 7867 __netdev_update_lower_level(upper_dev, priv); 7868 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 7869 priv); 7870 7871 return 0; 7872 7873 rollback: 7874 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 7875 7876 return ret; 7877 } 7878 7879 /** 7880 * netdev_upper_dev_link - Add a link to the upper device 7881 * @dev: device 7882 * @upper_dev: new upper device 7883 * @extack: netlink extended ack 7884 * 7885 * Adds a link to device which is upper to this one. The caller must hold 7886 * the RTNL lock. On a failure a negative errno code is returned. 7887 * On success the reference counts are adjusted and the function 7888 * returns zero. 7889 */ 7890 int netdev_upper_dev_link(struct net_device *dev, 7891 struct net_device *upper_dev, 7892 struct netlink_ext_ack *extack) 7893 { 7894 struct netdev_nested_priv priv = { 7895 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7896 .data = NULL, 7897 }; 7898 7899 return __netdev_upper_dev_link(dev, upper_dev, false, 7900 NULL, NULL, &priv, extack); 7901 } 7902 EXPORT_SYMBOL(netdev_upper_dev_link); 7903 7904 /** 7905 * netdev_master_upper_dev_link - Add a master link to the upper device 7906 * @dev: device 7907 * @upper_dev: new upper device 7908 * @upper_priv: upper device private 7909 * @upper_info: upper info to be passed down via notifier 7910 * @extack: netlink extended ack 7911 * 7912 * Adds a link to device which is upper to this one. In this case, only 7913 * one master upper device can be linked, although other non-master devices 7914 * might be linked as well. The caller must hold the RTNL lock. 7915 * On a failure a negative errno code is returned. On success the reference 7916 * counts are adjusted and the function returns zero. 7917 */ 7918 int netdev_master_upper_dev_link(struct net_device *dev, 7919 struct net_device *upper_dev, 7920 void *upper_priv, void *upper_info, 7921 struct netlink_ext_ack *extack) 7922 { 7923 struct netdev_nested_priv priv = { 7924 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7925 .data = NULL, 7926 }; 7927 7928 return __netdev_upper_dev_link(dev, upper_dev, true, 7929 upper_priv, upper_info, &priv, extack); 7930 } 7931 EXPORT_SYMBOL(netdev_master_upper_dev_link); 7932 7933 static void __netdev_upper_dev_unlink(struct net_device *dev, 7934 struct net_device *upper_dev, 7935 struct netdev_nested_priv *priv) 7936 { 7937 struct netdev_notifier_changeupper_info changeupper_info = { 7938 .info = { 7939 .dev = dev, 7940 }, 7941 .upper_dev = upper_dev, 7942 .linking = false, 7943 }; 7944 7945 ASSERT_RTNL(); 7946 7947 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; 7948 7949 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7950 &changeupper_info.info); 7951 7952 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 7953 7954 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 7955 &changeupper_info.info); 7956 7957 __netdev_update_upper_level(dev, NULL); 7958 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 7959 7960 __netdev_update_lower_level(upper_dev, priv); 7961 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 7962 priv); 7963 } 7964 7965 /** 7966 * netdev_upper_dev_unlink - Removes a link to upper device 7967 * @dev: device 7968 * @upper_dev: new upper device 7969 * 7970 * Removes a link to device which is upper to this one. The caller must hold 7971 * the RTNL lock. 7972 */ 7973 void netdev_upper_dev_unlink(struct net_device *dev, 7974 struct net_device *upper_dev) 7975 { 7976 struct netdev_nested_priv priv = { 7977 .flags = NESTED_SYNC_TODO, 7978 .data = NULL, 7979 }; 7980 7981 __netdev_upper_dev_unlink(dev, upper_dev, &priv); 7982 } 7983 EXPORT_SYMBOL(netdev_upper_dev_unlink); 7984 7985 static void __netdev_adjacent_dev_set(struct net_device *upper_dev, 7986 struct net_device *lower_dev, 7987 bool val) 7988 { 7989 struct netdev_adjacent *adj; 7990 7991 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower); 7992 if (adj) 7993 adj->ignore = val; 7994 7995 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper); 7996 if (adj) 7997 adj->ignore = val; 7998 } 7999 8000 static void netdev_adjacent_dev_disable(struct net_device *upper_dev, 8001 struct net_device *lower_dev) 8002 { 8003 __netdev_adjacent_dev_set(upper_dev, lower_dev, true); 8004 } 8005 8006 static void netdev_adjacent_dev_enable(struct net_device *upper_dev, 8007 struct net_device *lower_dev) 8008 { 8009 __netdev_adjacent_dev_set(upper_dev, lower_dev, false); 8010 } 8011 8012 int netdev_adjacent_change_prepare(struct net_device *old_dev, 8013 struct net_device *new_dev, 8014 struct net_device *dev, 8015 struct netlink_ext_ack *extack) 8016 { 8017 struct netdev_nested_priv priv = { 8018 .flags = 0, 8019 .data = NULL, 8020 }; 8021 int err; 8022 8023 if (!new_dev) 8024 return 0; 8025 8026 if (old_dev && new_dev != old_dev) 8027 netdev_adjacent_dev_disable(dev, old_dev); 8028 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv, 8029 extack); 8030 if (err) { 8031 if (old_dev && new_dev != old_dev) 8032 netdev_adjacent_dev_enable(dev, old_dev); 8033 return err; 8034 } 8035 8036 return 0; 8037 } 8038 EXPORT_SYMBOL(netdev_adjacent_change_prepare); 8039 8040 void netdev_adjacent_change_commit(struct net_device *old_dev, 8041 struct net_device *new_dev, 8042 struct net_device *dev) 8043 { 8044 struct netdev_nested_priv priv = { 8045 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 8046 .data = NULL, 8047 }; 8048 8049 if (!new_dev || !old_dev) 8050 return; 8051 8052 if (new_dev == old_dev) 8053 return; 8054 8055 netdev_adjacent_dev_enable(dev, old_dev); 8056 __netdev_upper_dev_unlink(old_dev, dev, &priv); 8057 } 8058 EXPORT_SYMBOL(netdev_adjacent_change_commit); 8059 8060 void netdev_adjacent_change_abort(struct net_device *old_dev, 8061 struct net_device *new_dev, 8062 struct net_device *dev) 8063 { 8064 struct netdev_nested_priv priv = { 8065 .flags = 0, 8066 .data = NULL, 8067 }; 8068 8069 if (!new_dev) 8070 return; 8071 8072 if (old_dev && new_dev != old_dev) 8073 netdev_adjacent_dev_enable(dev, old_dev); 8074 8075 __netdev_upper_dev_unlink(new_dev, dev, &priv); 8076 } 8077 EXPORT_SYMBOL(netdev_adjacent_change_abort); 8078 8079 /** 8080 * netdev_bonding_info_change - Dispatch event about slave change 8081 * @dev: device 8082 * @bonding_info: info to dispatch 8083 * 8084 * Send NETDEV_BONDING_INFO to netdev notifiers with info. 8085 * The caller must hold the RTNL lock. 8086 */ 8087 void netdev_bonding_info_change(struct net_device *dev, 8088 struct netdev_bonding_info *bonding_info) 8089 { 8090 struct netdev_notifier_bonding_info info = { 8091 .info.dev = dev, 8092 }; 8093 8094 memcpy(&info.bonding_info, bonding_info, 8095 sizeof(struct netdev_bonding_info)); 8096 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, 8097 &info.info); 8098 } 8099 EXPORT_SYMBOL(netdev_bonding_info_change); 8100 8101 /** 8102 * netdev_get_xmit_slave - Get the xmit slave of master device 8103 * @dev: device 8104 * @skb: The packet 8105 * @all_slaves: assume all the slaves are active 8106 * 8107 * The reference counters are not incremented so the caller must be 8108 * careful with locks. The caller must hold RCU lock. 8109 * %NULL is returned if no slave is found. 8110 */ 8111 8112 struct net_device *netdev_get_xmit_slave(struct net_device *dev, 8113 struct sk_buff *skb, 8114 bool all_slaves) 8115 { 8116 const struct net_device_ops *ops = dev->netdev_ops; 8117 8118 if (!ops->ndo_get_xmit_slave) 8119 return NULL; 8120 return ops->ndo_get_xmit_slave(dev, skb, all_slaves); 8121 } 8122 EXPORT_SYMBOL(netdev_get_xmit_slave); 8123 8124 static void netdev_adjacent_add_links(struct net_device *dev) 8125 { 8126 struct netdev_adjacent *iter; 8127 8128 struct net *net = dev_net(dev); 8129 8130 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8131 if (!net_eq(net, dev_net(iter->dev))) 8132 continue; 8133 netdev_adjacent_sysfs_add(iter->dev, dev, 8134 &iter->dev->adj_list.lower); 8135 netdev_adjacent_sysfs_add(dev, iter->dev, 8136 &dev->adj_list.upper); 8137 } 8138 8139 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8140 if (!net_eq(net, dev_net(iter->dev))) 8141 continue; 8142 netdev_adjacent_sysfs_add(iter->dev, dev, 8143 &iter->dev->adj_list.upper); 8144 netdev_adjacent_sysfs_add(dev, iter->dev, 8145 &dev->adj_list.lower); 8146 } 8147 } 8148 8149 static void netdev_adjacent_del_links(struct net_device *dev) 8150 { 8151 struct netdev_adjacent *iter; 8152 8153 struct net *net = dev_net(dev); 8154 8155 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8156 if (!net_eq(net, dev_net(iter->dev))) 8157 continue; 8158 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8159 &iter->dev->adj_list.lower); 8160 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8161 &dev->adj_list.upper); 8162 } 8163 8164 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8165 if (!net_eq(net, dev_net(iter->dev))) 8166 continue; 8167 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8168 &iter->dev->adj_list.upper); 8169 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8170 &dev->adj_list.lower); 8171 } 8172 } 8173 8174 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) 8175 { 8176 struct netdev_adjacent *iter; 8177 8178 struct net *net = dev_net(dev); 8179 8180 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8181 if (!net_eq(net, dev_net(iter->dev))) 8182 continue; 8183 netdev_adjacent_sysfs_del(iter->dev, oldname, 8184 &iter->dev->adj_list.lower); 8185 netdev_adjacent_sysfs_add(iter->dev, dev, 8186 &iter->dev->adj_list.lower); 8187 } 8188 8189 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8190 if (!net_eq(net, dev_net(iter->dev))) 8191 continue; 8192 netdev_adjacent_sysfs_del(iter->dev, oldname, 8193 &iter->dev->adj_list.upper); 8194 netdev_adjacent_sysfs_add(iter->dev, dev, 8195 &iter->dev->adj_list.upper); 8196 } 8197 } 8198 8199 void *netdev_lower_dev_get_private(struct net_device *dev, 8200 struct net_device *lower_dev) 8201 { 8202 struct netdev_adjacent *lower; 8203 8204 if (!lower_dev) 8205 return NULL; 8206 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); 8207 if (!lower) 8208 return NULL; 8209 8210 return lower->private; 8211 } 8212 EXPORT_SYMBOL(netdev_lower_dev_get_private); 8213 8214 8215 /** 8216 * netdev_lower_state_changed - Dispatch event about lower device state change 8217 * @lower_dev: device 8218 * @lower_state_info: state to dispatch 8219 * 8220 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. 8221 * The caller must hold the RTNL lock. 8222 */ 8223 void netdev_lower_state_changed(struct net_device *lower_dev, 8224 void *lower_state_info) 8225 { 8226 struct netdev_notifier_changelowerstate_info changelowerstate_info = { 8227 .info.dev = lower_dev, 8228 }; 8229 8230 ASSERT_RTNL(); 8231 changelowerstate_info.lower_state_info = lower_state_info; 8232 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, 8233 &changelowerstate_info.info); 8234 } 8235 EXPORT_SYMBOL(netdev_lower_state_changed); 8236 8237 static void dev_change_rx_flags(struct net_device *dev, int flags) 8238 { 8239 const struct net_device_ops *ops = dev->netdev_ops; 8240 8241 if (ops->ndo_change_rx_flags) 8242 ops->ndo_change_rx_flags(dev, flags); 8243 } 8244 8245 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) 8246 { 8247 unsigned int old_flags = dev->flags; 8248 kuid_t uid; 8249 kgid_t gid; 8250 8251 ASSERT_RTNL(); 8252 8253 dev->flags |= IFF_PROMISC; 8254 dev->promiscuity += inc; 8255 if (dev->promiscuity == 0) { 8256 /* 8257 * Avoid overflow. 8258 * If inc causes overflow, untouch promisc and return error. 8259 */ 8260 if (inc < 0) 8261 dev->flags &= ~IFF_PROMISC; 8262 else { 8263 dev->promiscuity -= inc; 8264 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n", 8265 dev->name); 8266 return -EOVERFLOW; 8267 } 8268 } 8269 if (dev->flags != old_flags) { 8270 pr_info("device %s %s promiscuous mode\n", 8271 dev->name, 8272 dev->flags & IFF_PROMISC ? "entered" : "left"); 8273 if (audit_enabled) { 8274 current_uid_gid(&uid, &gid); 8275 audit_log(audit_context(), GFP_ATOMIC, 8276 AUDIT_ANOM_PROMISCUOUS, 8277 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", 8278 dev->name, (dev->flags & IFF_PROMISC), 8279 (old_flags & IFF_PROMISC), 8280 from_kuid(&init_user_ns, audit_get_loginuid(current)), 8281 from_kuid(&init_user_ns, uid), 8282 from_kgid(&init_user_ns, gid), 8283 audit_get_sessionid(current)); 8284 } 8285 8286 dev_change_rx_flags(dev, IFF_PROMISC); 8287 } 8288 if (notify) 8289 __dev_notify_flags(dev, old_flags, IFF_PROMISC); 8290 return 0; 8291 } 8292 8293 /** 8294 * dev_set_promiscuity - update promiscuity count on a device 8295 * @dev: device 8296 * @inc: modifier 8297 * 8298 * Add or remove promiscuity from a device. While the count in the device 8299 * remains above zero the interface remains promiscuous. Once it hits zero 8300 * the device reverts back to normal filtering operation. A negative inc 8301 * value is used to drop promiscuity on the device. 8302 * Return 0 if successful or a negative errno code on error. 8303 */ 8304 int dev_set_promiscuity(struct net_device *dev, int inc) 8305 { 8306 unsigned int old_flags = dev->flags; 8307 int err; 8308 8309 err = __dev_set_promiscuity(dev, inc, true); 8310 if (err < 0) 8311 return err; 8312 if (dev->flags != old_flags) 8313 dev_set_rx_mode(dev); 8314 return err; 8315 } 8316 EXPORT_SYMBOL(dev_set_promiscuity); 8317 8318 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) 8319 { 8320 unsigned int old_flags = dev->flags, old_gflags = dev->gflags; 8321 8322 ASSERT_RTNL(); 8323 8324 dev->flags |= IFF_ALLMULTI; 8325 dev->allmulti += inc; 8326 if (dev->allmulti == 0) { 8327 /* 8328 * Avoid overflow. 8329 * If inc causes overflow, untouch allmulti and return error. 8330 */ 8331 if (inc < 0) 8332 dev->flags &= ~IFF_ALLMULTI; 8333 else { 8334 dev->allmulti -= inc; 8335 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n", 8336 dev->name); 8337 return -EOVERFLOW; 8338 } 8339 } 8340 if (dev->flags ^ old_flags) { 8341 dev_change_rx_flags(dev, IFF_ALLMULTI); 8342 dev_set_rx_mode(dev); 8343 if (notify) 8344 __dev_notify_flags(dev, old_flags, 8345 dev->gflags ^ old_gflags); 8346 } 8347 return 0; 8348 } 8349 8350 /** 8351 * dev_set_allmulti - update allmulti count on a device 8352 * @dev: device 8353 * @inc: modifier 8354 * 8355 * Add or remove reception of all multicast frames to a device. While the 8356 * count in the device remains above zero the interface remains listening 8357 * to all interfaces. Once it hits zero the device reverts back to normal 8358 * filtering operation. A negative @inc value is used to drop the counter 8359 * when releasing a resource needing all multicasts. 8360 * Return 0 if successful or a negative errno code on error. 8361 */ 8362 8363 int dev_set_allmulti(struct net_device *dev, int inc) 8364 { 8365 return __dev_set_allmulti(dev, inc, true); 8366 } 8367 EXPORT_SYMBOL(dev_set_allmulti); 8368 8369 /* 8370 * Upload unicast and multicast address lists to device and 8371 * configure RX filtering. When the device doesn't support unicast 8372 * filtering it is put in promiscuous mode while unicast addresses 8373 * are present. 8374 */ 8375 void __dev_set_rx_mode(struct net_device *dev) 8376 { 8377 const struct net_device_ops *ops = dev->netdev_ops; 8378 8379 /* dev_open will call this function so the list will stay sane. */ 8380 if (!(dev->flags&IFF_UP)) 8381 return; 8382 8383 if (!netif_device_present(dev)) 8384 return; 8385 8386 if (!(dev->priv_flags & IFF_UNICAST_FLT)) { 8387 /* Unicast addresses changes may only happen under the rtnl, 8388 * therefore calling __dev_set_promiscuity here is safe. 8389 */ 8390 if (!netdev_uc_empty(dev) && !dev->uc_promisc) { 8391 __dev_set_promiscuity(dev, 1, false); 8392 dev->uc_promisc = true; 8393 } else if (netdev_uc_empty(dev) && dev->uc_promisc) { 8394 __dev_set_promiscuity(dev, -1, false); 8395 dev->uc_promisc = false; 8396 } 8397 } 8398 8399 if (ops->ndo_set_rx_mode) 8400 ops->ndo_set_rx_mode(dev); 8401 } 8402 8403 void dev_set_rx_mode(struct net_device *dev) 8404 { 8405 netif_addr_lock_bh(dev); 8406 __dev_set_rx_mode(dev); 8407 netif_addr_unlock_bh(dev); 8408 } 8409 8410 /** 8411 * dev_get_flags - get flags reported to userspace 8412 * @dev: device 8413 * 8414 * Get the combination of flag bits exported through APIs to userspace. 8415 */ 8416 unsigned int dev_get_flags(const struct net_device *dev) 8417 { 8418 unsigned int flags; 8419 8420 flags = (dev->flags & ~(IFF_PROMISC | 8421 IFF_ALLMULTI | 8422 IFF_RUNNING | 8423 IFF_LOWER_UP | 8424 IFF_DORMANT)) | 8425 (dev->gflags & (IFF_PROMISC | 8426 IFF_ALLMULTI)); 8427 8428 if (netif_running(dev)) { 8429 if (netif_oper_up(dev)) 8430 flags |= IFF_RUNNING; 8431 if (netif_carrier_ok(dev)) 8432 flags |= IFF_LOWER_UP; 8433 if (netif_dormant(dev)) 8434 flags |= IFF_DORMANT; 8435 } 8436 8437 return flags; 8438 } 8439 EXPORT_SYMBOL(dev_get_flags); 8440 8441 int __dev_change_flags(struct net_device *dev, unsigned int flags, 8442 struct netlink_ext_ack *extack) 8443 { 8444 unsigned int old_flags = dev->flags; 8445 int ret; 8446 8447 ASSERT_RTNL(); 8448 8449 /* 8450 * Set the flags on our device. 8451 */ 8452 8453 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | 8454 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | 8455 IFF_AUTOMEDIA)) | 8456 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | 8457 IFF_ALLMULTI)); 8458 8459 /* 8460 * Load in the correct multicast list now the flags have changed. 8461 */ 8462 8463 if ((old_flags ^ flags) & IFF_MULTICAST) 8464 dev_change_rx_flags(dev, IFF_MULTICAST); 8465 8466 dev_set_rx_mode(dev); 8467 8468 /* 8469 * Have we downed the interface. We handle IFF_UP ourselves 8470 * according to user attempts to set it, rather than blindly 8471 * setting it. 8472 */ 8473 8474 ret = 0; 8475 if ((old_flags ^ flags) & IFF_UP) { 8476 if (old_flags & IFF_UP) 8477 __dev_close(dev); 8478 else 8479 ret = __dev_open(dev, extack); 8480 } 8481 8482 if ((flags ^ dev->gflags) & IFF_PROMISC) { 8483 int inc = (flags & IFF_PROMISC) ? 1 : -1; 8484 unsigned int old_flags = dev->flags; 8485 8486 dev->gflags ^= IFF_PROMISC; 8487 8488 if (__dev_set_promiscuity(dev, inc, false) >= 0) 8489 if (dev->flags != old_flags) 8490 dev_set_rx_mode(dev); 8491 } 8492 8493 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI 8494 * is important. Some (broken) drivers set IFF_PROMISC, when 8495 * IFF_ALLMULTI is requested not asking us and not reporting. 8496 */ 8497 if ((flags ^ dev->gflags) & IFF_ALLMULTI) { 8498 int inc = (flags & IFF_ALLMULTI) ? 1 : -1; 8499 8500 dev->gflags ^= IFF_ALLMULTI; 8501 __dev_set_allmulti(dev, inc, false); 8502 } 8503 8504 return ret; 8505 } 8506 8507 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, 8508 unsigned int gchanges) 8509 { 8510 unsigned int changes = dev->flags ^ old_flags; 8511 8512 if (gchanges) 8513 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC); 8514 8515 if (changes & IFF_UP) { 8516 if (dev->flags & IFF_UP) 8517 call_netdevice_notifiers(NETDEV_UP, dev); 8518 else 8519 call_netdevice_notifiers(NETDEV_DOWN, dev); 8520 } 8521 8522 if (dev->flags & IFF_UP && 8523 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { 8524 struct netdev_notifier_change_info change_info = { 8525 .info = { 8526 .dev = dev, 8527 }, 8528 .flags_changed = changes, 8529 }; 8530 8531 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info); 8532 } 8533 } 8534 8535 /** 8536 * dev_change_flags - change device settings 8537 * @dev: device 8538 * @flags: device state flags 8539 * @extack: netlink extended ack 8540 * 8541 * Change settings on device based state flags. The flags are 8542 * in the userspace exported format. 8543 */ 8544 int dev_change_flags(struct net_device *dev, unsigned int flags, 8545 struct netlink_ext_ack *extack) 8546 { 8547 int ret; 8548 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; 8549 8550 ret = __dev_change_flags(dev, flags, extack); 8551 if (ret < 0) 8552 return ret; 8553 8554 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); 8555 __dev_notify_flags(dev, old_flags, changes); 8556 return ret; 8557 } 8558 EXPORT_SYMBOL(dev_change_flags); 8559 8560 int __dev_set_mtu(struct net_device *dev, int new_mtu) 8561 { 8562 const struct net_device_ops *ops = dev->netdev_ops; 8563 8564 if (ops->ndo_change_mtu) 8565 return ops->ndo_change_mtu(dev, new_mtu); 8566 8567 /* Pairs with all the lockless reads of dev->mtu in the stack */ 8568 WRITE_ONCE(dev->mtu, new_mtu); 8569 return 0; 8570 } 8571 EXPORT_SYMBOL(__dev_set_mtu); 8572 8573 int dev_validate_mtu(struct net_device *dev, int new_mtu, 8574 struct netlink_ext_ack *extack) 8575 { 8576 /* MTU must be positive, and in range */ 8577 if (new_mtu < 0 || new_mtu < dev->min_mtu) { 8578 NL_SET_ERR_MSG(extack, "mtu less than device minimum"); 8579 return -EINVAL; 8580 } 8581 8582 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { 8583 NL_SET_ERR_MSG(extack, "mtu greater than device maximum"); 8584 return -EINVAL; 8585 } 8586 return 0; 8587 } 8588 8589 /** 8590 * dev_set_mtu_ext - Change maximum transfer unit 8591 * @dev: device 8592 * @new_mtu: new transfer unit 8593 * @extack: netlink extended ack 8594 * 8595 * Change the maximum transfer size of the network device. 8596 */ 8597 int dev_set_mtu_ext(struct net_device *dev, int new_mtu, 8598 struct netlink_ext_ack *extack) 8599 { 8600 int err, orig_mtu; 8601 8602 if (new_mtu == dev->mtu) 8603 return 0; 8604 8605 err = dev_validate_mtu(dev, new_mtu, extack); 8606 if (err) 8607 return err; 8608 8609 if (!netif_device_present(dev)) 8610 return -ENODEV; 8611 8612 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); 8613 err = notifier_to_errno(err); 8614 if (err) 8615 return err; 8616 8617 orig_mtu = dev->mtu; 8618 err = __dev_set_mtu(dev, new_mtu); 8619 8620 if (!err) { 8621 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8622 orig_mtu); 8623 err = notifier_to_errno(err); 8624 if (err) { 8625 /* setting mtu back and notifying everyone again, 8626 * so that they have a chance to revert changes. 8627 */ 8628 __dev_set_mtu(dev, orig_mtu); 8629 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8630 new_mtu); 8631 } 8632 } 8633 return err; 8634 } 8635 8636 int dev_set_mtu(struct net_device *dev, int new_mtu) 8637 { 8638 struct netlink_ext_ack extack; 8639 int err; 8640 8641 memset(&extack, 0, sizeof(extack)); 8642 err = dev_set_mtu_ext(dev, new_mtu, &extack); 8643 if (err && extack._msg) 8644 net_err_ratelimited("%s: %s\n", dev->name, extack._msg); 8645 return err; 8646 } 8647 EXPORT_SYMBOL(dev_set_mtu); 8648 8649 /** 8650 * dev_change_tx_queue_len - Change TX queue length of a netdevice 8651 * @dev: device 8652 * @new_len: new tx queue length 8653 */ 8654 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len) 8655 { 8656 unsigned int orig_len = dev->tx_queue_len; 8657 int res; 8658 8659 if (new_len != (unsigned int)new_len) 8660 return -ERANGE; 8661 8662 if (new_len != orig_len) { 8663 dev->tx_queue_len = new_len; 8664 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev); 8665 res = notifier_to_errno(res); 8666 if (res) 8667 goto err_rollback; 8668 res = dev_qdisc_change_tx_queue_len(dev); 8669 if (res) 8670 goto err_rollback; 8671 } 8672 8673 return 0; 8674 8675 err_rollback: 8676 netdev_err(dev, "refused to change device tx_queue_len\n"); 8677 dev->tx_queue_len = orig_len; 8678 return res; 8679 } 8680 8681 /** 8682 * dev_set_group - Change group this device belongs to 8683 * @dev: device 8684 * @new_group: group this device should belong to 8685 */ 8686 void dev_set_group(struct net_device *dev, int new_group) 8687 { 8688 dev->group = new_group; 8689 } 8690 EXPORT_SYMBOL(dev_set_group); 8691 8692 /** 8693 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR. 8694 * @dev: device 8695 * @addr: new address 8696 * @extack: netlink extended ack 8697 */ 8698 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr, 8699 struct netlink_ext_ack *extack) 8700 { 8701 struct netdev_notifier_pre_changeaddr_info info = { 8702 .info.dev = dev, 8703 .info.extack = extack, 8704 .dev_addr = addr, 8705 }; 8706 int rc; 8707 8708 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info); 8709 return notifier_to_errno(rc); 8710 } 8711 EXPORT_SYMBOL(dev_pre_changeaddr_notify); 8712 8713 /** 8714 * dev_set_mac_address - Change Media Access Control Address 8715 * @dev: device 8716 * @sa: new address 8717 * @extack: netlink extended ack 8718 * 8719 * Change the hardware (MAC) address of the device 8720 */ 8721 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa, 8722 struct netlink_ext_ack *extack) 8723 { 8724 const struct net_device_ops *ops = dev->netdev_ops; 8725 int err; 8726 8727 if (!ops->ndo_set_mac_address) 8728 return -EOPNOTSUPP; 8729 if (sa->sa_family != dev->type) 8730 return -EINVAL; 8731 if (!netif_device_present(dev)) 8732 return -ENODEV; 8733 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack); 8734 if (err) 8735 return err; 8736 err = ops->ndo_set_mac_address(dev, sa); 8737 if (err) 8738 return err; 8739 dev->addr_assign_type = NET_ADDR_SET; 8740 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); 8741 add_device_randomness(dev->dev_addr, dev->addr_len); 8742 return 0; 8743 } 8744 EXPORT_SYMBOL(dev_set_mac_address); 8745 8746 /** 8747 * dev_change_carrier - Change device carrier 8748 * @dev: device 8749 * @new_carrier: new value 8750 * 8751 * Change device carrier 8752 */ 8753 int dev_change_carrier(struct net_device *dev, bool new_carrier) 8754 { 8755 const struct net_device_ops *ops = dev->netdev_ops; 8756 8757 if (!ops->ndo_change_carrier) 8758 return -EOPNOTSUPP; 8759 if (!netif_device_present(dev)) 8760 return -ENODEV; 8761 return ops->ndo_change_carrier(dev, new_carrier); 8762 } 8763 EXPORT_SYMBOL(dev_change_carrier); 8764 8765 /** 8766 * dev_get_phys_port_id - Get device physical port ID 8767 * @dev: device 8768 * @ppid: port ID 8769 * 8770 * Get device physical port ID 8771 */ 8772 int dev_get_phys_port_id(struct net_device *dev, 8773 struct netdev_phys_item_id *ppid) 8774 { 8775 const struct net_device_ops *ops = dev->netdev_ops; 8776 8777 if (!ops->ndo_get_phys_port_id) 8778 return -EOPNOTSUPP; 8779 return ops->ndo_get_phys_port_id(dev, ppid); 8780 } 8781 EXPORT_SYMBOL(dev_get_phys_port_id); 8782 8783 /** 8784 * dev_get_phys_port_name - Get device physical port name 8785 * @dev: device 8786 * @name: port name 8787 * @len: limit of bytes to copy to name 8788 * 8789 * Get device physical port name 8790 */ 8791 int dev_get_phys_port_name(struct net_device *dev, 8792 char *name, size_t len) 8793 { 8794 const struct net_device_ops *ops = dev->netdev_ops; 8795 int err; 8796 8797 if (ops->ndo_get_phys_port_name) { 8798 err = ops->ndo_get_phys_port_name(dev, name, len); 8799 if (err != -EOPNOTSUPP) 8800 return err; 8801 } 8802 return devlink_compat_phys_port_name_get(dev, name, len); 8803 } 8804 EXPORT_SYMBOL(dev_get_phys_port_name); 8805 8806 /** 8807 * dev_get_port_parent_id - Get the device's port parent identifier 8808 * @dev: network device 8809 * @ppid: pointer to a storage for the port's parent identifier 8810 * @recurse: allow/disallow recursion to lower devices 8811 * 8812 * Get the devices's port parent identifier 8813 */ 8814 int dev_get_port_parent_id(struct net_device *dev, 8815 struct netdev_phys_item_id *ppid, 8816 bool recurse) 8817 { 8818 const struct net_device_ops *ops = dev->netdev_ops; 8819 struct netdev_phys_item_id first = { }; 8820 struct net_device *lower_dev; 8821 struct list_head *iter; 8822 int err; 8823 8824 if (ops->ndo_get_port_parent_id) { 8825 err = ops->ndo_get_port_parent_id(dev, ppid); 8826 if (err != -EOPNOTSUPP) 8827 return err; 8828 } 8829 8830 err = devlink_compat_switch_id_get(dev, ppid); 8831 if (!err || err != -EOPNOTSUPP) 8832 return err; 8833 8834 if (!recurse) 8835 return -EOPNOTSUPP; 8836 8837 netdev_for_each_lower_dev(dev, lower_dev, iter) { 8838 err = dev_get_port_parent_id(lower_dev, ppid, recurse); 8839 if (err) 8840 break; 8841 if (!first.id_len) 8842 first = *ppid; 8843 else if (memcmp(&first, ppid, sizeof(*ppid))) 8844 return -EOPNOTSUPP; 8845 } 8846 8847 return err; 8848 } 8849 EXPORT_SYMBOL(dev_get_port_parent_id); 8850 8851 /** 8852 * netdev_port_same_parent_id - Indicate if two network devices have 8853 * the same port parent identifier 8854 * @a: first network device 8855 * @b: second network device 8856 */ 8857 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b) 8858 { 8859 struct netdev_phys_item_id a_id = { }; 8860 struct netdev_phys_item_id b_id = { }; 8861 8862 if (dev_get_port_parent_id(a, &a_id, true) || 8863 dev_get_port_parent_id(b, &b_id, true)) 8864 return false; 8865 8866 return netdev_phys_item_id_same(&a_id, &b_id); 8867 } 8868 EXPORT_SYMBOL(netdev_port_same_parent_id); 8869 8870 /** 8871 * dev_change_proto_down - update protocol port state information 8872 * @dev: device 8873 * @proto_down: new value 8874 * 8875 * This info can be used by switch drivers to set the phys state of the 8876 * port. 8877 */ 8878 int dev_change_proto_down(struct net_device *dev, bool proto_down) 8879 { 8880 const struct net_device_ops *ops = dev->netdev_ops; 8881 8882 if (!ops->ndo_change_proto_down) 8883 return -EOPNOTSUPP; 8884 if (!netif_device_present(dev)) 8885 return -ENODEV; 8886 return ops->ndo_change_proto_down(dev, proto_down); 8887 } 8888 EXPORT_SYMBOL(dev_change_proto_down); 8889 8890 /** 8891 * dev_change_proto_down_generic - generic implementation for 8892 * ndo_change_proto_down that sets carrier according to 8893 * proto_down. 8894 * 8895 * @dev: device 8896 * @proto_down: new value 8897 */ 8898 int dev_change_proto_down_generic(struct net_device *dev, bool proto_down) 8899 { 8900 if (proto_down) 8901 netif_carrier_off(dev); 8902 else 8903 netif_carrier_on(dev); 8904 dev->proto_down = proto_down; 8905 return 0; 8906 } 8907 EXPORT_SYMBOL(dev_change_proto_down_generic); 8908 8909 /** 8910 * dev_change_proto_down_reason - proto down reason 8911 * 8912 * @dev: device 8913 * @mask: proto down mask 8914 * @value: proto down value 8915 */ 8916 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask, 8917 u32 value) 8918 { 8919 int b; 8920 8921 if (!mask) { 8922 dev->proto_down_reason = value; 8923 } else { 8924 for_each_set_bit(b, &mask, 32) { 8925 if (value & (1 << b)) 8926 dev->proto_down_reason |= BIT(b); 8927 else 8928 dev->proto_down_reason &= ~BIT(b); 8929 } 8930 } 8931 } 8932 EXPORT_SYMBOL(dev_change_proto_down_reason); 8933 8934 struct bpf_xdp_link { 8935 struct bpf_link link; 8936 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */ 8937 int flags; 8938 }; 8939 8940 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags) 8941 { 8942 if (flags & XDP_FLAGS_HW_MODE) 8943 return XDP_MODE_HW; 8944 if (flags & XDP_FLAGS_DRV_MODE) 8945 return XDP_MODE_DRV; 8946 if (flags & XDP_FLAGS_SKB_MODE) 8947 return XDP_MODE_SKB; 8948 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB; 8949 } 8950 8951 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode) 8952 { 8953 switch (mode) { 8954 case XDP_MODE_SKB: 8955 return generic_xdp_install; 8956 case XDP_MODE_DRV: 8957 case XDP_MODE_HW: 8958 return dev->netdev_ops->ndo_bpf; 8959 default: 8960 return NULL; 8961 } 8962 } 8963 8964 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev, 8965 enum bpf_xdp_mode mode) 8966 { 8967 return dev->xdp_state[mode].link; 8968 } 8969 8970 static struct bpf_prog *dev_xdp_prog(struct net_device *dev, 8971 enum bpf_xdp_mode mode) 8972 { 8973 struct bpf_xdp_link *link = dev_xdp_link(dev, mode); 8974 8975 if (link) 8976 return link->link.prog; 8977 return dev->xdp_state[mode].prog; 8978 } 8979 8980 static u8 dev_xdp_prog_count(struct net_device *dev) 8981 { 8982 u8 count = 0; 8983 int i; 8984 8985 for (i = 0; i < __MAX_XDP_MODE; i++) 8986 if (dev->xdp_state[i].prog || dev->xdp_state[i].link) 8987 count++; 8988 return count; 8989 } 8990 8991 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode) 8992 { 8993 struct bpf_prog *prog = dev_xdp_prog(dev, mode); 8994 8995 return prog ? prog->aux->id : 0; 8996 } 8997 8998 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode, 8999 struct bpf_xdp_link *link) 9000 { 9001 dev->xdp_state[mode].link = link; 9002 dev->xdp_state[mode].prog = NULL; 9003 } 9004 9005 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode, 9006 struct bpf_prog *prog) 9007 { 9008 dev->xdp_state[mode].link = NULL; 9009 dev->xdp_state[mode].prog = prog; 9010 } 9011 9012 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode, 9013 bpf_op_t bpf_op, struct netlink_ext_ack *extack, 9014 u32 flags, struct bpf_prog *prog) 9015 { 9016 struct netdev_bpf xdp; 9017 int err; 9018 9019 memset(&xdp, 0, sizeof(xdp)); 9020 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG; 9021 xdp.extack = extack; 9022 xdp.flags = flags; 9023 xdp.prog = prog; 9024 9025 /* Drivers assume refcnt is already incremented (i.e, prog pointer is 9026 * "moved" into driver), so they don't increment it on their own, but 9027 * they do decrement refcnt when program is detached or replaced. 9028 * Given net_device also owns link/prog, we need to bump refcnt here 9029 * to prevent drivers from underflowing it. 9030 */ 9031 if (prog) 9032 bpf_prog_inc(prog); 9033 err = bpf_op(dev, &xdp); 9034 if (err) { 9035 if (prog) 9036 bpf_prog_put(prog); 9037 return err; 9038 } 9039 9040 if (mode != XDP_MODE_HW) 9041 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog); 9042 9043 return 0; 9044 } 9045 9046 static void dev_xdp_uninstall(struct net_device *dev) 9047 { 9048 struct bpf_xdp_link *link; 9049 struct bpf_prog *prog; 9050 enum bpf_xdp_mode mode; 9051 bpf_op_t bpf_op; 9052 9053 ASSERT_RTNL(); 9054 9055 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) { 9056 prog = dev_xdp_prog(dev, mode); 9057 if (!prog) 9058 continue; 9059 9060 bpf_op = dev_xdp_bpf_op(dev, mode); 9061 if (!bpf_op) 9062 continue; 9063 9064 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9065 9066 /* auto-detach link from net device */ 9067 link = dev_xdp_link(dev, mode); 9068 if (link) 9069 link->dev = NULL; 9070 else 9071 bpf_prog_put(prog); 9072 9073 dev_xdp_set_link(dev, mode, NULL); 9074 } 9075 } 9076 9077 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack, 9078 struct bpf_xdp_link *link, struct bpf_prog *new_prog, 9079 struct bpf_prog *old_prog, u32 flags) 9080 { 9081 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES); 9082 struct bpf_prog *cur_prog; 9083 enum bpf_xdp_mode mode; 9084 bpf_op_t bpf_op; 9085 int err; 9086 9087 ASSERT_RTNL(); 9088 9089 /* either link or prog attachment, never both */ 9090 if (link && (new_prog || old_prog)) 9091 return -EINVAL; 9092 /* link supports only XDP mode flags */ 9093 if (link && (flags & ~XDP_FLAGS_MODES)) { 9094 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment"); 9095 return -EINVAL; 9096 } 9097 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */ 9098 if (num_modes > 1) { 9099 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set"); 9100 return -EINVAL; 9101 } 9102 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */ 9103 if (!num_modes && dev_xdp_prog_count(dev) > 1) { 9104 NL_SET_ERR_MSG(extack, 9105 "More than one program loaded, unset mode is ambiguous"); 9106 return -EINVAL; 9107 } 9108 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */ 9109 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) { 9110 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified"); 9111 return -EINVAL; 9112 } 9113 9114 mode = dev_xdp_mode(dev, flags); 9115 /* can't replace attached link */ 9116 if (dev_xdp_link(dev, mode)) { 9117 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link"); 9118 return -EBUSY; 9119 } 9120 9121 cur_prog = dev_xdp_prog(dev, mode); 9122 /* can't replace attached prog with link */ 9123 if (link && cur_prog) { 9124 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link"); 9125 return -EBUSY; 9126 } 9127 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) { 9128 NL_SET_ERR_MSG(extack, "Active program does not match expected"); 9129 return -EEXIST; 9130 } 9131 9132 /* put effective new program into new_prog */ 9133 if (link) 9134 new_prog = link->link.prog; 9135 9136 if (new_prog) { 9137 bool offload = mode == XDP_MODE_HW; 9138 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB 9139 ? XDP_MODE_DRV : XDP_MODE_SKB; 9140 9141 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) { 9142 NL_SET_ERR_MSG(extack, "XDP program already attached"); 9143 return -EBUSY; 9144 } 9145 if (!offload && dev_xdp_prog(dev, other_mode)) { 9146 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time"); 9147 return -EEXIST; 9148 } 9149 if (!offload && bpf_prog_is_dev_bound(new_prog->aux)) { 9150 NL_SET_ERR_MSG(extack, "Using device-bound program without HW_MODE flag is not supported"); 9151 return -EINVAL; 9152 } 9153 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) { 9154 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device"); 9155 return -EINVAL; 9156 } 9157 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) { 9158 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device"); 9159 return -EINVAL; 9160 } 9161 } 9162 9163 /* don't call drivers if the effective program didn't change */ 9164 if (new_prog != cur_prog) { 9165 bpf_op = dev_xdp_bpf_op(dev, mode); 9166 if (!bpf_op) { 9167 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode"); 9168 return -EOPNOTSUPP; 9169 } 9170 9171 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog); 9172 if (err) 9173 return err; 9174 } 9175 9176 if (link) 9177 dev_xdp_set_link(dev, mode, link); 9178 else 9179 dev_xdp_set_prog(dev, mode, new_prog); 9180 if (cur_prog) 9181 bpf_prog_put(cur_prog); 9182 9183 return 0; 9184 } 9185 9186 static int dev_xdp_attach_link(struct net_device *dev, 9187 struct netlink_ext_ack *extack, 9188 struct bpf_xdp_link *link) 9189 { 9190 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags); 9191 } 9192 9193 static int dev_xdp_detach_link(struct net_device *dev, 9194 struct netlink_ext_ack *extack, 9195 struct bpf_xdp_link *link) 9196 { 9197 enum bpf_xdp_mode mode; 9198 bpf_op_t bpf_op; 9199 9200 ASSERT_RTNL(); 9201 9202 mode = dev_xdp_mode(dev, link->flags); 9203 if (dev_xdp_link(dev, mode) != link) 9204 return -EINVAL; 9205 9206 bpf_op = dev_xdp_bpf_op(dev, mode); 9207 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9208 dev_xdp_set_link(dev, mode, NULL); 9209 return 0; 9210 } 9211 9212 static void bpf_xdp_link_release(struct bpf_link *link) 9213 { 9214 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9215 9216 rtnl_lock(); 9217 9218 /* if racing with net_device's tear down, xdp_link->dev might be 9219 * already NULL, in which case link was already auto-detached 9220 */ 9221 if (xdp_link->dev) { 9222 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link)); 9223 xdp_link->dev = NULL; 9224 } 9225 9226 rtnl_unlock(); 9227 } 9228 9229 static int bpf_xdp_link_detach(struct bpf_link *link) 9230 { 9231 bpf_xdp_link_release(link); 9232 return 0; 9233 } 9234 9235 static void bpf_xdp_link_dealloc(struct bpf_link *link) 9236 { 9237 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9238 9239 kfree(xdp_link); 9240 } 9241 9242 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link, 9243 struct seq_file *seq) 9244 { 9245 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9246 u32 ifindex = 0; 9247 9248 rtnl_lock(); 9249 if (xdp_link->dev) 9250 ifindex = xdp_link->dev->ifindex; 9251 rtnl_unlock(); 9252 9253 seq_printf(seq, "ifindex:\t%u\n", ifindex); 9254 } 9255 9256 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link, 9257 struct bpf_link_info *info) 9258 { 9259 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9260 u32 ifindex = 0; 9261 9262 rtnl_lock(); 9263 if (xdp_link->dev) 9264 ifindex = xdp_link->dev->ifindex; 9265 rtnl_unlock(); 9266 9267 info->xdp.ifindex = ifindex; 9268 return 0; 9269 } 9270 9271 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog, 9272 struct bpf_prog *old_prog) 9273 { 9274 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9275 enum bpf_xdp_mode mode; 9276 bpf_op_t bpf_op; 9277 int err = 0; 9278 9279 rtnl_lock(); 9280 9281 /* link might have been auto-released already, so fail */ 9282 if (!xdp_link->dev) { 9283 err = -ENOLINK; 9284 goto out_unlock; 9285 } 9286 9287 if (old_prog && link->prog != old_prog) { 9288 err = -EPERM; 9289 goto out_unlock; 9290 } 9291 old_prog = link->prog; 9292 if (old_prog == new_prog) { 9293 /* no-op, don't disturb drivers */ 9294 bpf_prog_put(new_prog); 9295 goto out_unlock; 9296 } 9297 9298 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags); 9299 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode); 9300 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL, 9301 xdp_link->flags, new_prog); 9302 if (err) 9303 goto out_unlock; 9304 9305 old_prog = xchg(&link->prog, new_prog); 9306 bpf_prog_put(old_prog); 9307 9308 out_unlock: 9309 rtnl_unlock(); 9310 return err; 9311 } 9312 9313 static const struct bpf_link_ops bpf_xdp_link_lops = { 9314 .release = bpf_xdp_link_release, 9315 .dealloc = bpf_xdp_link_dealloc, 9316 .detach = bpf_xdp_link_detach, 9317 .show_fdinfo = bpf_xdp_link_show_fdinfo, 9318 .fill_link_info = bpf_xdp_link_fill_link_info, 9319 .update_prog = bpf_xdp_link_update, 9320 }; 9321 9322 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog) 9323 { 9324 struct net *net = current->nsproxy->net_ns; 9325 struct bpf_link_primer link_primer; 9326 struct bpf_xdp_link *link; 9327 struct net_device *dev; 9328 int err, fd; 9329 9330 dev = dev_get_by_index(net, attr->link_create.target_ifindex); 9331 if (!dev) 9332 return -EINVAL; 9333 9334 link = kzalloc(sizeof(*link), GFP_USER); 9335 if (!link) { 9336 err = -ENOMEM; 9337 goto out_put_dev; 9338 } 9339 9340 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog); 9341 link->dev = dev; 9342 link->flags = attr->link_create.flags; 9343 9344 err = bpf_link_prime(&link->link, &link_primer); 9345 if (err) { 9346 kfree(link); 9347 goto out_put_dev; 9348 } 9349 9350 rtnl_lock(); 9351 err = dev_xdp_attach_link(dev, NULL, link); 9352 rtnl_unlock(); 9353 9354 if (err) { 9355 bpf_link_cleanup(&link_primer); 9356 goto out_put_dev; 9357 } 9358 9359 fd = bpf_link_settle(&link_primer); 9360 /* link itself doesn't hold dev's refcnt to not complicate shutdown */ 9361 dev_put(dev); 9362 return fd; 9363 9364 out_put_dev: 9365 dev_put(dev); 9366 return err; 9367 } 9368 9369 /** 9370 * dev_change_xdp_fd - set or clear a bpf program for a device rx path 9371 * @dev: device 9372 * @extack: netlink extended ack 9373 * @fd: new program fd or negative value to clear 9374 * @expected_fd: old program fd that userspace expects to replace or clear 9375 * @flags: xdp-related flags 9376 * 9377 * Set or clear a bpf program for a device 9378 */ 9379 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, 9380 int fd, int expected_fd, u32 flags) 9381 { 9382 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags); 9383 struct bpf_prog *new_prog = NULL, *old_prog = NULL; 9384 int err; 9385 9386 ASSERT_RTNL(); 9387 9388 if (fd >= 0) { 9389 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP, 9390 mode != XDP_MODE_SKB); 9391 if (IS_ERR(new_prog)) 9392 return PTR_ERR(new_prog); 9393 } 9394 9395 if (expected_fd >= 0) { 9396 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP, 9397 mode != XDP_MODE_SKB); 9398 if (IS_ERR(old_prog)) { 9399 err = PTR_ERR(old_prog); 9400 old_prog = NULL; 9401 goto err_out; 9402 } 9403 } 9404 9405 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags); 9406 9407 err_out: 9408 if (err && new_prog) 9409 bpf_prog_put(new_prog); 9410 if (old_prog) 9411 bpf_prog_put(old_prog); 9412 return err; 9413 } 9414 9415 /** 9416 * dev_new_index - allocate an ifindex 9417 * @net: the applicable net namespace 9418 * 9419 * Returns a suitable unique value for a new device interface 9420 * number. The caller must hold the rtnl semaphore or the 9421 * dev_base_lock to be sure it remains unique. 9422 */ 9423 static int dev_new_index(struct net *net) 9424 { 9425 int ifindex = net->ifindex; 9426 9427 for (;;) { 9428 if (++ifindex <= 0) 9429 ifindex = 1; 9430 if (!__dev_get_by_index(net, ifindex)) 9431 return net->ifindex = ifindex; 9432 } 9433 } 9434 9435 /* Delayed registration/unregisteration */ 9436 static LIST_HEAD(net_todo_list); 9437 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); 9438 9439 static void net_set_todo(struct net_device *dev) 9440 { 9441 list_add_tail(&dev->todo_list, &net_todo_list); 9442 dev_net(dev)->dev_unreg_count++; 9443 } 9444 9445 static void rollback_registered_many(struct list_head *head) 9446 { 9447 struct net_device *dev, *tmp; 9448 LIST_HEAD(close_head); 9449 9450 BUG_ON(dev_boot_phase); 9451 ASSERT_RTNL(); 9452 9453 list_for_each_entry_safe(dev, tmp, head, unreg_list) { 9454 /* Some devices call without registering 9455 * for initialization unwind. Remove those 9456 * devices and proceed with the remaining. 9457 */ 9458 if (dev->reg_state == NETREG_UNINITIALIZED) { 9459 pr_debug("unregister_netdevice: device %s/%p never was registered\n", 9460 dev->name, dev); 9461 9462 WARN_ON(1); 9463 list_del(&dev->unreg_list); 9464 continue; 9465 } 9466 dev->dismantle = true; 9467 BUG_ON(dev->reg_state != NETREG_REGISTERED); 9468 } 9469 9470 /* If device is running, close it first. */ 9471 list_for_each_entry(dev, head, unreg_list) 9472 list_add_tail(&dev->close_list, &close_head); 9473 dev_close_many(&close_head, true); 9474 9475 list_for_each_entry(dev, head, unreg_list) { 9476 /* And unlink it from device chain. */ 9477 unlist_netdevice(dev); 9478 9479 dev->reg_state = NETREG_UNREGISTERING; 9480 } 9481 flush_all_backlogs(); 9482 9483 synchronize_net(); 9484 9485 list_for_each_entry(dev, head, unreg_list) { 9486 struct sk_buff *skb = NULL; 9487 9488 /* Shutdown queueing discipline. */ 9489 dev_shutdown(dev); 9490 9491 dev_xdp_uninstall(dev); 9492 9493 /* Notify protocols, that we are about to destroy 9494 * this device. They should clean all the things. 9495 */ 9496 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 9497 9498 if (!dev->rtnl_link_ops || 9499 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 9500 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0, 9501 GFP_KERNEL, NULL, 0); 9502 9503 /* 9504 * Flush the unicast and multicast chains 9505 */ 9506 dev_uc_flush(dev); 9507 dev_mc_flush(dev); 9508 9509 netdev_name_node_alt_flush(dev); 9510 netdev_name_node_free(dev->name_node); 9511 9512 if (dev->netdev_ops->ndo_uninit) 9513 dev->netdev_ops->ndo_uninit(dev); 9514 9515 if (skb) 9516 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL); 9517 9518 /* Notifier chain MUST detach us all upper devices. */ 9519 WARN_ON(netdev_has_any_upper_dev(dev)); 9520 WARN_ON(netdev_has_any_lower_dev(dev)); 9521 9522 /* Remove entries from kobject tree */ 9523 netdev_unregister_kobject(dev); 9524 #ifdef CONFIG_XPS 9525 /* Remove XPS queueing entries */ 9526 netif_reset_xps_queues_gt(dev, 0); 9527 #endif 9528 } 9529 9530 synchronize_net(); 9531 9532 list_for_each_entry(dev, head, unreg_list) 9533 dev_put(dev); 9534 } 9535 9536 static void rollback_registered(struct net_device *dev) 9537 { 9538 LIST_HEAD(single); 9539 9540 list_add(&dev->unreg_list, &single); 9541 rollback_registered_many(&single); 9542 list_del(&single); 9543 } 9544 9545 static netdev_features_t netdev_sync_upper_features(struct net_device *lower, 9546 struct net_device *upper, netdev_features_t features) 9547 { 9548 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9549 netdev_features_t feature; 9550 int feature_bit; 9551 9552 for_each_netdev_feature(upper_disables, feature_bit) { 9553 feature = __NETIF_F_BIT(feature_bit); 9554 if (!(upper->wanted_features & feature) 9555 && (features & feature)) { 9556 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", 9557 &feature, upper->name); 9558 features &= ~feature; 9559 } 9560 } 9561 9562 return features; 9563 } 9564 9565 static void netdev_sync_lower_features(struct net_device *upper, 9566 struct net_device *lower, netdev_features_t features) 9567 { 9568 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9569 netdev_features_t feature; 9570 int feature_bit; 9571 9572 for_each_netdev_feature(upper_disables, feature_bit) { 9573 feature = __NETIF_F_BIT(feature_bit); 9574 if (!(features & feature) && (lower->features & feature)) { 9575 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", 9576 &feature, lower->name); 9577 lower->wanted_features &= ~feature; 9578 __netdev_update_features(lower); 9579 9580 if (unlikely(lower->features & feature)) 9581 netdev_WARN(upper, "failed to disable %pNF on %s!\n", 9582 &feature, lower->name); 9583 else 9584 netdev_features_change(lower); 9585 } 9586 } 9587 } 9588 9589 static netdev_features_t netdev_fix_features(struct net_device *dev, 9590 netdev_features_t features) 9591 { 9592 /* Fix illegal checksum combinations */ 9593 if ((features & NETIF_F_HW_CSUM) && 9594 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { 9595 netdev_warn(dev, "mixed HW and IP checksum settings.\n"); 9596 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); 9597 } 9598 9599 /* TSO requires that SG is present as well. */ 9600 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { 9601 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); 9602 features &= ~NETIF_F_ALL_TSO; 9603 } 9604 9605 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && 9606 !(features & NETIF_F_IP_CSUM)) { 9607 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); 9608 features &= ~NETIF_F_TSO; 9609 features &= ~NETIF_F_TSO_ECN; 9610 } 9611 9612 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && 9613 !(features & NETIF_F_IPV6_CSUM)) { 9614 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); 9615 features &= ~NETIF_F_TSO6; 9616 } 9617 9618 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ 9619 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) 9620 features &= ~NETIF_F_TSO_MANGLEID; 9621 9622 /* TSO ECN requires that TSO is present as well. */ 9623 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) 9624 features &= ~NETIF_F_TSO_ECN; 9625 9626 /* Software GSO depends on SG. */ 9627 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { 9628 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); 9629 features &= ~NETIF_F_GSO; 9630 } 9631 9632 /* GSO partial features require GSO partial be set */ 9633 if ((features & dev->gso_partial_features) && 9634 !(features & NETIF_F_GSO_PARTIAL)) { 9635 netdev_dbg(dev, 9636 "Dropping partially supported GSO features since no GSO partial.\n"); 9637 features &= ~dev->gso_partial_features; 9638 } 9639 9640 if (!(features & NETIF_F_RXCSUM)) { 9641 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet 9642 * successfully merged by hardware must also have the 9643 * checksum verified by hardware. If the user does not 9644 * want to enable RXCSUM, logically, we should disable GRO_HW. 9645 */ 9646 if (features & NETIF_F_GRO_HW) { 9647 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n"); 9648 features &= ~NETIF_F_GRO_HW; 9649 } 9650 } 9651 9652 /* LRO/HW-GRO features cannot be combined with RX-FCS */ 9653 if (features & NETIF_F_RXFCS) { 9654 if (features & NETIF_F_LRO) { 9655 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n"); 9656 features &= ~NETIF_F_LRO; 9657 } 9658 9659 if (features & NETIF_F_GRO_HW) { 9660 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n"); 9661 features &= ~NETIF_F_GRO_HW; 9662 } 9663 } 9664 9665 if (features & NETIF_F_HW_TLS_TX) { 9666 bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) == 9667 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM); 9668 bool hw_csum = features & NETIF_F_HW_CSUM; 9669 9670 if (!ip_csum && !hw_csum) { 9671 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n"); 9672 features &= ~NETIF_F_HW_TLS_TX; 9673 } 9674 } 9675 9676 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) { 9677 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n"); 9678 features &= ~NETIF_F_HW_TLS_RX; 9679 } 9680 9681 return features; 9682 } 9683 9684 int __netdev_update_features(struct net_device *dev) 9685 { 9686 struct net_device *upper, *lower; 9687 netdev_features_t features; 9688 struct list_head *iter; 9689 int err = -1; 9690 9691 ASSERT_RTNL(); 9692 9693 features = netdev_get_wanted_features(dev); 9694 9695 if (dev->netdev_ops->ndo_fix_features) 9696 features = dev->netdev_ops->ndo_fix_features(dev, features); 9697 9698 /* driver might be less strict about feature dependencies */ 9699 features = netdev_fix_features(dev, features); 9700 9701 /* some features can't be enabled if they're off on an upper device */ 9702 netdev_for_each_upper_dev_rcu(dev, upper, iter) 9703 features = netdev_sync_upper_features(dev, upper, features); 9704 9705 if (dev->features == features) 9706 goto sync_lower; 9707 9708 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", 9709 &dev->features, &features); 9710 9711 if (dev->netdev_ops->ndo_set_features) 9712 err = dev->netdev_ops->ndo_set_features(dev, features); 9713 else 9714 err = 0; 9715 9716 if (unlikely(err < 0)) { 9717 netdev_err(dev, 9718 "set_features() failed (%d); wanted %pNF, left %pNF\n", 9719 err, &features, &dev->features); 9720 /* return non-0 since some features might have changed and 9721 * it's better to fire a spurious notification than miss it 9722 */ 9723 return -1; 9724 } 9725 9726 sync_lower: 9727 /* some features must be disabled on lower devices when disabled 9728 * on an upper device (think: bonding master or bridge) 9729 */ 9730 netdev_for_each_lower_dev(dev, lower, iter) 9731 netdev_sync_lower_features(dev, lower, features); 9732 9733 if (!err) { 9734 netdev_features_t diff = features ^ dev->features; 9735 9736 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) { 9737 /* udp_tunnel_{get,drop}_rx_info both need 9738 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the 9739 * device, or they won't do anything. 9740 * Thus we need to update dev->features 9741 * *before* calling udp_tunnel_get_rx_info, 9742 * but *after* calling udp_tunnel_drop_rx_info. 9743 */ 9744 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) { 9745 dev->features = features; 9746 udp_tunnel_get_rx_info(dev); 9747 } else { 9748 udp_tunnel_drop_rx_info(dev); 9749 } 9750 } 9751 9752 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) { 9753 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) { 9754 dev->features = features; 9755 err |= vlan_get_rx_ctag_filter_info(dev); 9756 } else { 9757 vlan_drop_rx_ctag_filter_info(dev); 9758 } 9759 } 9760 9761 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) { 9762 if (features & NETIF_F_HW_VLAN_STAG_FILTER) { 9763 dev->features = features; 9764 err |= vlan_get_rx_stag_filter_info(dev); 9765 } else { 9766 vlan_drop_rx_stag_filter_info(dev); 9767 } 9768 } 9769 9770 dev->features = features; 9771 } 9772 9773 return err < 0 ? 0 : 1; 9774 } 9775 9776 /** 9777 * netdev_update_features - recalculate device features 9778 * @dev: the device to check 9779 * 9780 * Recalculate dev->features set and send notifications if it 9781 * has changed. Should be called after driver or hardware dependent 9782 * conditions might have changed that influence the features. 9783 */ 9784 void netdev_update_features(struct net_device *dev) 9785 { 9786 if (__netdev_update_features(dev)) 9787 netdev_features_change(dev); 9788 } 9789 EXPORT_SYMBOL(netdev_update_features); 9790 9791 /** 9792 * netdev_change_features - recalculate device features 9793 * @dev: the device to check 9794 * 9795 * Recalculate dev->features set and send notifications even 9796 * if they have not changed. Should be called instead of 9797 * netdev_update_features() if also dev->vlan_features might 9798 * have changed to allow the changes to be propagated to stacked 9799 * VLAN devices. 9800 */ 9801 void netdev_change_features(struct net_device *dev) 9802 { 9803 __netdev_update_features(dev); 9804 netdev_features_change(dev); 9805 } 9806 EXPORT_SYMBOL(netdev_change_features); 9807 9808 /** 9809 * netif_stacked_transfer_operstate - transfer operstate 9810 * @rootdev: the root or lower level device to transfer state from 9811 * @dev: the device to transfer operstate to 9812 * 9813 * Transfer operational state from root to device. This is normally 9814 * called when a stacking relationship exists between the root 9815 * device and the device(a leaf device). 9816 */ 9817 void netif_stacked_transfer_operstate(const struct net_device *rootdev, 9818 struct net_device *dev) 9819 { 9820 if (rootdev->operstate == IF_OPER_DORMANT) 9821 netif_dormant_on(dev); 9822 else 9823 netif_dormant_off(dev); 9824 9825 if (rootdev->operstate == IF_OPER_TESTING) 9826 netif_testing_on(dev); 9827 else 9828 netif_testing_off(dev); 9829 9830 if (netif_carrier_ok(rootdev)) 9831 netif_carrier_on(dev); 9832 else 9833 netif_carrier_off(dev); 9834 } 9835 EXPORT_SYMBOL(netif_stacked_transfer_operstate); 9836 9837 static int netif_alloc_rx_queues(struct net_device *dev) 9838 { 9839 unsigned int i, count = dev->num_rx_queues; 9840 struct netdev_rx_queue *rx; 9841 size_t sz = count * sizeof(*rx); 9842 int err = 0; 9843 9844 BUG_ON(count < 1); 9845 9846 rx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL); 9847 if (!rx) 9848 return -ENOMEM; 9849 9850 dev->_rx = rx; 9851 9852 for (i = 0; i < count; i++) { 9853 rx[i].dev = dev; 9854 9855 /* XDP RX-queue setup */ 9856 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0); 9857 if (err < 0) 9858 goto err_rxq_info; 9859 } 9860 return 0; 9861 9862 err_rxq_info: 9863 /* Rollback successful reg's and free other resources */ 9864 while (i--) 9865 xdp_rxq_info_unreg(&rx[i].xdp_rxq); 9866 kvfree(dev->_rx); 9867 dev->_rx = NULL; 9868 return err; 9869 } 9870 9871 static void netif_free_rx_queues(struct net_device *dev) 9872 { 9873 unsigned int i, count = dev->num_rx_queues; 9874 9875 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */ 9876 if (!dev->_rx) 9877 return; 9878 9879 for (i = 0; i < count; i++) 9880 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq); 9881 9882 kvfree(dev->_rx); 9883 } 9884 9885 static void netdev_init_one_queue(struct net_device *dev, 9886 struct netdev_queue *queue, void *_unused) 9887 { 9888 /* Initialize queue lock */ 9889 spin_lock_init(&queue->_xmit_lock); 9890 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); 9891 queue->xmit_lock_owner = -1; 9892 netdev_queue_numa_node_write(queue, NUMA_NO_NODE); 9893 queue->dev = dev; 9894 #ifdef CONFIG_BQL 9895 dql_init(&queue->dql, HZ); 9896 #endif 9897 } 9898 9899 static void netif_free_tx_queues(struct net_device *dev) 9900 { 9901 kvfree(dev->_tx); 9902 } 9903 9904 static int netif_alloc_netdev_queues(struct net_device *dev) 9905 { 9906 unsigned int count = dev->num_tx_queues; 9907 struct netdev_queue *tx; 9908 size_t sz = count * sizeof(*tx); 9909 9910 if (count < 1 || count > 0xffff) 9911 return -EINVAL; 9912 9913 tx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL); 9914 if (!tx) 9915 return -ENOMEM; 9916 9917 dev->_tx = tx; 9918 9919 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); 9920 spin_lock_init(&dev->tx_global_lock); 9921 9922 return 0; 9923 } 9924 9925 void netif_tx_stop_all_queues(struct net_device *dev) 9926 { 9927 unsigned int i; 9928 9929 for (i = 0; i < dev->num_tx_queues; i++) { 9930 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 9931 9932 netif_tx_stop_queue(txq); 9933 } 9934 } 9935 EXPORT_SYMBOL(netif_tx_stop_all_queues); 9936 9937 /** 9938 * register_netdevice - register a network device 9939 * @dev: device to register 9940 * 9941 * Take a completed network device structure and add it to the kernel 9942 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 9943 * chain. 0 is returned on success. A negative errno code is returned 9944 * on a failure to set up the device, or if the name is a duplicate. 9945 * 9946 * Callers must hold the rtnl semaphore. You may want 9947 * register_netdev() instead of this. 9948 * 9949 * BUGS: 9950 * The locking appears insufficient to guarantee two parallel registers 9951 * will not get the same name. 9952 */ 9953 9954 int register_netdevice(struct net_device *dev) 9955 { 9956 int ret; 9957 struct net *net = dev_net(dev); 9958 9959 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE < 9960 NETDEV_FEATURE_COUNT); 9961 BUG_ON(dev_boot_phase); 9962 ASSERT_RTNL(); 9963 9964 might_sleep(); 9965 9966 /* When net_device's are persistent, this will be fatal. */ 9967 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); 9968 BUG_ON(!net); 9969 9970 ret = ethtool_check_ops(dev->ethtool_ops); 9971 if (ret) 9972 return ret; 9973 9974 spin_lock_init(&dev->addr_list_lock); 9975 netdev_set_addr_lockdep_class(dev); 9976 9977 ret = dev_get_valid_name(net, dev, dev->name); 9978 if (ret < 0) 9979 goto out; 9980 9981 ret = -ENOMEM; 9982 dev->name_node = netdev_name_node_head_alloc(dev); 9983 if (!dev->name_node) 9984 goto out; 9985 9986 /* Init, if this function is available */ 9987 if (dev->netdev_ops->ndo_init) { 9988 ret = dev->netdev_ops->ndo_init(dev); 9989 if (ret) { 9990 if (ret > 0) 9991 ret = -EIO; 9992 goto err_free_name; 9993 } 9994 } 9995 9996 if (((dev->hw_features | dev->features) & 9997 NETIF_F_HW_VLAN_CTAG_FILTER) && 9998 (!dev->netdev_ops->ndo_vlan_rx_add_vid || 9999 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { 10000 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); 10001 ret = -EINVAL; 10002 goto err_uninit; 10003 } 10004 10005 ret = -EBUSY; 10006 if (!dev->ifindex) 10007 dev->ifindex = dev_new_index(net); 10008 else if (__dev_get_by_index(net, dev->ifindex)) 10009 goto err_uninit; 10010 10011 /* Transfer changeable features to wanted_features and enable 10012 * software offloads (GSO and GRO). 10013 */ 10014 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF); 10015 dev->features |= NETIF_F_SOFT_FEATURES; 10016 10017 if (dev->netdev_ops->ndo_udp_tunnel_add) { 10018 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10019 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10020 } 10021 10022 dev->wanted_features = dev->features & dev->hw_features; 10023 10024 if (!(dev->flags & IFF_LOOPBACK)) 10025 dev->hw_features |= NETIF_F_NOCACHE_COPY; 10026 10027 /* If IPv4 TCP segmentation offload is supported we should also 10028 * allow the device to enable segmenting the frame with the option 10029 * of ignoring a static IP ID value. This doesn't enable the 10030 * feature itself but allows the user to enable it later. 10031 */ 10032 if (dev->hw_features & NETIF_F_TSO) 10033 dev->hw_features |= NETIF_F_TSO_MANGLEID; 10034 if (dev->vlan_features & NETIF_F_TSO) 10035 dev->vlan_features |= NETIF_F_TSO_MANGLEID; 10036 if (dev->mpls_features & NETIF_F_TSO) 10037 dev->mpls_features |= NETIF_F_TSO_MANGLEID; 10038 if (dev->hw_enc_features & NETIF_F_TSO) 10039 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; 10040 10041 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. 10042 */ 10043 dev->vlan_features |= NETIF_F_HIGHDMA; 10044 10045 /* Make NETIF_F_SG inheritable to tunnel devices. 10046 */ 10047 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; 10048 10049 /* Make NETIF_F_SG inheritable to MPLS. 10050 */ 10051 dev->mpls_features |= NETIF_F_SG; 10052 10053 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); 10054 ret = notifier_to_errno(ret); 10055 if (ret) 10056 goto err_uninit; 10057 10058 ret = netdev_register_kobject(dev); 10059 if (ret) { 10060 dev->reg_state = NETREG_UNREGISTERED; 10061 goto err_uninit; 10062 } 10063 dev->reg_state = NETREG_REGISTERED; 10064 10065 __netdev_update_features(dev); 10066 10067 /* 10068 * Default initial state at registry is that the 10069 * device is present. 10070 */ 10071 10072 set_bit(__LINK_STATE_PRESENT, &dev->state); 10073 10074 linkwatch_init_dev(dev); 10075 10076 dev_init_scheduler(dev); 10077 dev_hold(dev); 10078 list_netdevice(dev); 10079 add_device_randomness(dev->dev_addr, dev->addr_len); 10080 10081 /* If the device has permanent device address, driver should 10082 * set dev_addr and also addr_assign_type should be set to 10083 * NET_ADDR_PERM (default value). 10084 */ 10085 if (dev->addr_assign_type == NET_ADDR_PERM) 10086 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); 10087 10088 /* Notify protocols, that a new device appeared. */ 10089 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); 10090 ret = notifier_to_errno(ret); 10091 if (ret) { 10092 /* Expect explicit free_netdev() on failure */ 10093 dev->needs_free_netdev = false; 10094 rollback_registered(dev); 10095 net_set_todo(dev); 10096 goto out; 10097 } 10098 /* 10099 * Prevent userspace races by waiting until the network 10100 * device is fully setup before sending notifications. 10101 */ 10102 if (!dev->rtnl_link_ops || 10103 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 10104 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); 10105 10106 out: 10107 return ret; 10108 10109 err_uninit: 10110 if (dev->netdev_ops->ndo_uninit) 10111 dev->netdev_ops->ndo_uninit(dev); 10112 if (dev->priv_destructor) 10113 dev->priv_destructor(dev); 10114 err_free_name: 10115 netdev_name_node_free(dev->name_node); 10116 goto out; 10117 } 10118 EXPORT_SYMBOL(register_netdevice); 10119 10120 /** 10121 * init_dummy_netdev - init a dummy network device for NAPI 10122 * @dev: device to init 10123 * 10124 * This takes a network device structure and initialize the minimum 10125 * amount of fields so it can be used to schedule NAPI polls without 10126 * registering a full blown interface. This is to be used by drivers 10127 * that need to tie several hardware interfaces to a single NAPI 10128 * poll scheduler due to HW limitations. 10129 */ 10130 int init_dummy_netdev(struct net_device *dev) 10131 { 10132 /* Clear everything. Note we don't initialize spinlocks 10133 * are they aren't supposed to be taken by any of the 10134 * NAPI code and this dummy netdev is supposed to be 10135 * only ever used for NAPI polls 10136 */ 10137 memset(dev, 0, sizeof(struct net_device)); 10138 10139 /* make sure we BUG if trying to hit standard 10140 * register/unregister code path 10141 */ 10142 dev->reg_state = NETREG_DUMMY; 10143 10144 /* NAPI wants this */ 10145 INIT_LIST_HEAD(&dev->napi_list); 10146 10147 /* a dummy interface is started by default */ 10148 set_bit(__LINK_STATE_PRESENT, &dev->state); 10149 set_bit(__LINK_STATE_START, &dev->state); 10150 10151 /* napi_busy_loop stats accounting wants this */ 10152 dev_net_set(dev, &init_net); 10153 10154 /* Note : We dont allocate pcpu_refcnt for dummy devices, 10155 * because users of this 'device' dont need to change 10156 * its refcount. 10157 */ 10158 10159 return 0; 10160 } 10161 EXPORT_SYMBOL_GPL(init_dummy_netdev); 10162 10163 10164 /** 10165 * register_netdev - register a network device 10166 * @dev: device to register 10167 * 10168 * Take a completed network device structure and add it to the kernel 10169 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 10170 * chain. 0 is returned on success. A negative errno code is returned 10171 * on a failure to set up the device, or if the name is a duplicate. 10172 * 10173 * This is a wrapper around register_netdevice that takes the rtnl semaphore 10174 * and expands the device name if you passed a format string to 10175 * alloc_netdev. 10176 */ 10177 int register_netdev(struct net_device *dev) 10178 { 10179 int err; 10180 10181 if (rtnl_lock_killable()) 10182 return -EINTR; 10183 err = register_netdevice(dev); 10184 rtnl_unlock(); 10185 return err; 10186 } 10187 EXPORT_SYMBOL(register_netdev); 10188 10189 int netdev_refcnt_read(const struct net_device *dev) 10190 { 10191 int i, refcnt = 0; 10192 10193 for_each_possible_cpu(i) 10194 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); 10195 return refcnt; 10196 } 10197 EXPORT_SYMBOL(netdev_refcnt_read); 10198 10199 #define WAIT_REFS_MIN_MSECS 1 10200 #define WAIT_REFS_MAX_MSECS 250 10201 /** 10202 * netdev_wait_allrefs - wait until all references are gone. 10203 * @dev: target net_device 10204 * 10205 * This is called when unregistering network devices. 10206 * 10207 * Any protocol or device that holds a reference should register 10208 * for netdevice notification, and cleanup and put back the 10209 * reference if they receive an UNREGISTER event. 10210 * We can get stuck here if buggy protocols don't correctly 10211 * call dev_put. 10212 */ 10213 static void netdev_wait_allrefs(struct net_device *dev) 10214 { 10215 unsigned long rebroadcast_time, warning_time; 10216 int wait = 0, refcnt; 10217 10218 linkwatch_forget_dev(dev); 10219 10220 rebroadcast_time = warning_time = jiffies; 10221 refcnt = netdev_refcnt_read(dev); 10222 10223 while (refcnt != 0) { 10224 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { 10225 rtnl_lock(); 10226 10227 /* Rebroadcast unregister notification */ 10228 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10229 10230 __rtnl_unlock(); 10231 rcu_barrier(); 10232 rtnl_lock(); 10233 10234 if (test_bit(__LINK_STATE_LINKWATCH_PENDING, 10235 &dev->state)) { 10236 /* We must not have linkwatch events 10237 * pending on unregister. If this 10238 * happens, we simply run the queue 10239 * unscheduled, resulting in a noop 10240 * for this device. 10241 */ 10242 linkwatch_run_queue(); 10243 } 10244 10245 __rtnl_unlock(); 10246 10247 rebroadcast_time = jiffies; 10248 } 10249 10250 if (!wait) { 10251 rcu_barrier(); 10252 wait = WAIT_REFS_MIN_MSECS; 10253 } else { 10254 msleep(wait); 10255 wait = min(wait << 1, WAIT_REFS_MAX_MSECS); 10256 } 10257 10258 refcnt = netdev_refcnt_read(dev); 10259 10260 if (refcnt && time_after(jiffies, warning_time + 10 * HZ)) { 10261 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", 10262 dev->name, refcnt); 10263 warning_time = jiffies; 10264 } 10265 } 10266 } 10267 10268 /* The sequence is: 10269 * 10270 * rtnl_lock(); 10271 * ... 10272 * register_netdevice(x1); 10273 * register_netdevice(x2); 10274 * ... 10275 * unregister_netdevice(y1); 10276 * unregister_netdevice(y2); 10277 * ... 10278 * rtnl_unlock(); 10279 * free_netdev(y1); 10280 * free_netdev(y2); 10281 * 10282 * We are invoked by rtnl_unlock(). 10283 * This allows us to deal with problems: 10284 * 1) We can delete sysfs objects which invoke hotplug 10285 * without deadlocking with linkwatch via keventd. 10286 * 2) Since we run with the RTNL semaphore not held, we can sleep 10287 * safely in order to wait for the netdev refcnt to drop to zero. 10288 * 10289 * We must not return until all unregister events added during 10290 * the interval the lock was held have been completed. 10291 */ 10292 void netdev_run_todo(void) 10293 { 10294 struct list_head list; 10295 #ifdef CONFIG_LOCKDEP 10296 struct list_head unlink_list; 10297 10298 list_replace_init(&net_unlink_list, &unlink_list); 10299 10300 while (!list_empty(&unlink_list)) { 10301 struct net_device *dev = list_first_entry(&unlink_list, 10302 struct net_device, 10303 unlink_list); 10304 list_del_init(&dev->unlink_list); 10305 dev->nested_level = dev->lower_level - 1; 10306 } 10307 #endif 10308 10309 /* Snapshot list, allow later requests */ 10310 list_replace_init(&net_todo_list, &list); 10311 10312 __rtnl_unlock(); 10313 10314 10315 /* Wait for rcu callbacks to finish before next phase */ 10316 if (!list_empty(&list)) 10317 rcu_barrier(); 10318 10319 while (!list_empty(&list)) { 10320 struct net_device *dev 10321 = list_first_entry(&list, struct net_device, todo_list); 10322 list_del(&dev->todo_list); 10323 10324 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { 10325 pr_err("network todo '%s' but state %d\n", 10326 dev->name, dev->reg_state); 10327 dump_stack(); 10328 continue; 10329 } 10330 10331 dev->reg_state = NETREG_UNREGISTERED; 10332 10333 netdev_wait_allrefs(dev); 10334 10335 /* paranoia */ 10336 BUG_ON(netdev_refcnt_read(dev)); 10337 BUG_ON(!list_empty(&dev->ptype_all)); 10338 BUG_ON(!list_empty(&dev->ptype_specific)); 10339 WARN_ON(rcu_access_pointer(dev->ip_ptr)); 10340 WARN_ON(rcu_access_pointer(dev->ip6_ptr)); 10341 #if IS_ENABLED(CONFIG_DECNET) 10342 WARN_ON(dev->dn_ptr); 10343 #endif 10344 if (dev->priv_destructor) 10345 dev->priv_destructor(dev); 10346 if (dev->needs_free_netdev) 10347 free_netdev(dev); 10348 10349 /* Report a network device has been unregistered */ 10350 rtnl_lock(); 10351 dev_net(dev)->dev_unreg_count--; 10352 __rtnl_unlock(); 10353 wake_up(&netdev_unregistering_wq); 10354 10355 /* Free network device */ 10356 kobject_put(&dev->dev.kobj); 10357 } 10358 } 10359 10360 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has 10361 * all the same fields in the same order as net_device_stats, with only 10362 * the type differing, but rtnl_link_stats64 may have additional fields 10363 * at the end for newer counters. 10364 */ 10365 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, 10366 const struct net_device_stats *netdev_stats) 10367 { 10368 #if BITS_PER_LONG == 64 10369 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats)); 10370 memcpy(stats64, netdev_stats, sizeof(*netdev_stats)); 10371 /* zero out counters that only exist in rtnl_link_stats64 */ 10372 memset((char *)stats64 + sizeof(*netdev_stats), 0, 10373 sizeof(*stats64) - sizeof(*netdev_stats)); 10374 #else 10375 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long); 10376 const unsigned long *src = (const unsigned long *)netdev_stats; 10377 u64 *dst = (u64 *)stats64; 10378 10379 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); 10380 for (i = 0; i < n; i++) 10381 dst[i] = src[i]; 10382 /* zero out counters that only exist in rtnl_link_stats64 */ 10383 memset((char *)stats64 + n * sizeof(u64), 0, 10384 sizeof(*stats64) - n * sizeof(u64)); 10385 #endif 10386 } 10387 EXPORT_SYMBOL(netdev_stats_to_stats64); 10388 10389 /** 10390 * dev_get_stats - get network device statistics 10391 * @dev: device to get statistics from 10392 * @storage: place to store stats 10393 * 10394 * Get network statistics from device. Return @storage. 10395 * The device driver may provide its own method by setting 10396 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; 10397 * otherwise the internal statistics structure is used. 10398 */ 10399 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, 10400 struct rtnl_link_stats64 *storage) 10401 { 10402 const struct net_device_ops *ops = dev->netdev_ops; 10403 10404 if (ops->ndo_get_stats64) { 10405 memset(storage, 0, sizeof(*storage)); 10406 ops->ndo_get_stats64(dev, storage); 10407 } else if (ops->ndo_get_stats) { 10408 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); 10409 } else { 10410 netdev_stats_to_stats64(storage, &dev->stats); 10411 } 10412 storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped); 10413 storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped); 10414 storage->rx_nohandler += (unsigned long)atomic_long_read(&dev->rx_nohandler); 10415 return storage; 10416 } 10417 EXPORT_SYMBOL(dev_get_stats); 10418 10419 /** 10420 * dev_fetch_sw_netstats - get per-cpu network device statistics 10421 * @s: place to store stats 10422 * @netstats: per-cpu network stats to read from 10423 * 10424 * Read per-cpu network statistics and populate the related fields in @s. 10425 */ 10426 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s, 10427 const struct pcpu_sw_netstats __percpu *netstats) 10428 { 10429 int cpu; 10430 10431 for_each_possible_cpu(cpu) { 10432 const struct pcpu_sw_netstats *stats; 10433 struct pcpu_sw_netstats tmp; 10434 unsigned int start; 10435 10436 stats = per_cpu_ptr(netstats, cpu); 10437 do { 10438 start = u64_stats_fetch_begin_irq(&stats->syncp); 10439 tmp.rx_packets = stats->rx_packets; 10440 tmp.rx_bytes = stats->rx_bytes; 10441 tmp.tx_packets = stats->tx_packets; 10442 tmp.tx_bytes = stats->tx_bytes; 10443 } while (u64_stats_fetch_retry_irq(&stats->syncp, start)); 10444 10445 s->rx_packets += tmp.rx_packets; 10446 s->rx_bytes += tmp.rx_bytes; 10447 s->tx_packets += tmp.tx_packets; 10448 s->tx_bytes += tmp.tx_bytes; 10449 } 10450 } 10451 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats); 10452 10453 /** 10454 * dev_get_tstats64 - ndo_get_stats64 implementation 10455 * @dev: device to get statistics from 10456 * @s: place to store stats 10457 * 10458 * Populate @s from dev->stats and dev->tstats. Can be used as 10459 * ndo_get_stats64() callback. 10460 */ 10461 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s) 10462 { 10463 netdev_stats_to_stats64(s, &dev->stats); 10464 dev_fetch_sw_netstats(s, dev->tstats); 10465 } 10466 EXPORT_SYMBOL_GPL(dev_get_tstats64); 10467 10468 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) 10469 { 10470 struct netdev_queue *queue = dev_ingress_queue(dev); 10471 10472 #ifdef CONFIG_NET_CLS_ACT 10473 if (queue) 10474 return queue; 10475 queue = kzalloc(sizeof(*queue), GFP_KERNEL); 10476 if (!queue) 10477 return NULL; 10478 netdev_init_one_queue(dev, queue, NULL); 10479 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); 10480 queue->qdisc_sleeping = &noop_qdisc; 10481 rcu_assign_pointer(dev->ingress_queue, queue); 10482 #endif 10483 return queue; 10484 } 10485 10486 static const struct ethtool_ops default_ethtool_ops; 10487 10488 void netdev_set_default_ethtool_ops(struct net_device *dev, 10489 const struct ethtool_ops *ops) 10490 { 10491 if (dev->ethtool_ops == &default_ethtool_ops) 10492 dev->ethtool_ops = ops; 10493 } 10494 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); 10495 10496 void netdev_freemem(struct net_device *dev) 10497 { 10498 char *addr = (char *)dev - dev->padded; 10499 10500 kvfree(addr); 10501 } 10502 10503 /** 10504 * alloc_netdev_mqs - allocate network device 10505 * @sizeof_priv: size of private data to allocate space for 10506 * @name: device name format string 10507 * @name_assign_type: origin of device name 10508 * @setup: callback to initialize device 10509 * @txqs: the number of TX subqueues to allocate 10510 * @rxqs: the number of RX subqueues to allocate 10511 * 10512 * Allocates a struct net_device with private data area for driver use 10513 * and performs basic initialization. Also allocates subqueue structs 10514 * for each queue on the device. 10515 */ 10516 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, 10517 unsigned char name_assign_type, 10518 void (*setup)(struct net_device *), 10519 unsigned int txqs, unsigned int rxqs) 10520 { 10521 struct net_device *dev; 10522 unsigned int alloc_size; 10523 struct net_device *p; 10524 10525 BUG_ON(strlen(name) >= sizeof(dev->name)); 10526 10527 if (txqs < 1) { 10528 pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); 10529 return NULL; 10530 } 10531 10532 if (rxqs < 1) { 10533 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); 10534 return NULL; 10535 } 10536 10537 alloc_size = sizeof(struct net_device); 10538 if (sizeof_priv) { 10539 /* ensure 32-byte alignment of private area */ 10540 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); 10541 alloc_size += sizeof_priv; 10542 } 10543 /* ensure 32-byte alignment of whole construct */ 10544 alloc_size += NETDEV_ALIGN - 1; 10545 10546 p = kvzalloc(alloc_size, GFP_KERNEL | __GFP_RETRY_MAYFAIL); 10547 if (!p) 10548 return NULL; 10549 10550 dev = PTR_ALIGN(p, NETDEV_ALIGN); 10551 dev->padded = (char *)dev - (char *)p; 10552 10553 dev->pcpu_refcnt = alloc_percpu(int); 10554 if (!dev->pcpu_refcnt) 10555 goto free_dev; 10556 10557 if (dev_addr_init(dev)) 10558 goto free_pcpu; 10559 10560 dev_mc_init(dev); 10561 dev_uc_init(dev); 10562 10563 dev_net_set(dev, &init_net); 10564 10565 dev->gso_max_size = GSO_MAX_SIZE; 10566 dev->gso_max_segs = GSO_MAX_SEGS; 10567 dev->upper_level = 1; 10568 dev->lower_level = 1; 10569 #ifdef CONFIG_LOCKDEP 10570 dev->nested_level = 0; 10571 INIT_LIST_HEAD(&dev->unlink_list); 10572 #endif 10573 10574 INIT_LIST_HEAD(&dev->napi_list); 10575 INIT_LIST_HEAD(&dev->unreg_list); 10576 INIT_LIST_HEAD(&dev->close_list); 10577 INIT_LIST_HEAD(&dev->link_watch_list); 10578 INIT_LIST_HEAD(&dev->adj_list.upper); 10579 INIT_LIST_HEAD(&dev->adj_list.lower); 10580 INIT_LIST_HEAD(&dev->ptype_all); 10581 INIT_LIST_HEAD(&dev->ptype_specific); 10582 INIT_LIST_HEAD(&dev->net_notifier_list); 10583 #ifdef CONFIG_NET_SCHED 10584 hash_init(dev->qdisc_hash); 10585 #endif 10586 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; 10587 setup(dev); 10588 10589 if (!dev->tx_queue_len) { 10590 dev->priv_flags |= IFF_NO_QUEUE; 10591 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; 10592 } 10593 10594 dev->num_tx_queues = txqs; 10595 dev->real_num_tx_queues = txqs; 10596 if (netif_alloc_netdev_queues(dev)) 10597 goto free_all; 10598 10599 dev->num_rx_queues = rxqs; 10600 dev->real_num_rx_queues = rxqs; 10601 if (netif_alloc_rx_queues(dev)) 10602 goto free_all; 10603 10604 strcpy(dev->name, name); 10605 dev->name_assign_type = name_assign_type; 10606 dev->group = INIT_NETDEV_GROUP; 10607 if (!dev->ethtool_ops) 10608 dev->ethtool_ops = &default_ethtool_ops; 10609 10610 nf_hook_ingress_init(dev); 10611 10612 return dev; 10613 10614 free_all: 10615 free_netdev(dev); 10616 return NULL; 10617 10618 free_pcpu: 10619 free_percpu(dev->pcpu_refcnt); 10620 free_dev: 10621 netdev_freemem(dev); 10622 return NULL; 10623 } 10624 EXPORT_SYMBOL(alloc_netdev_mqs); 10625 10626 /** 10627 * free_netdev - free network device 10628 * @dev: device 10629 * 10630 * This function does the last stage of destroying an allocated device 10631 * interface. The reference to the device object is released. If this 10632 * is the last reference then it will be freed.Must be called in process 10633 * context. 10634 */ 10635 void free_netdev(struct net_device *dev) 10636 { 10637 struct napi_struct *p, *n; 10638 10639 might_sleep(); 10640 10641 /* When called immediately after register_netdevice() failed the unwind 10642 * handling may still be dismantling the device. Handle that case by 10643 * deferring the free. 10644 */ 10645 if (dev->reg_state == NETREG_UNREGISTERING) { 10646 ASSERT_RTNL(); 10647 dev->needs_free_netdev = true; 10648 return; 10649 } 10650 10651 netif_free_tx_queues(dev); 10652 netif_free_rx_queues(dev); 10653 10654 kfree(rcu_dereference_protected(dev->ingress_queue, 1)); 10655 10656 /* Flush device addresses */ 10657 dev_addr_flush(dev); 10658 10659 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) 10660 netif_napi_del(p); 10661 10662 free_percpu(dev->pcpu_refcnt); 10663 dev->pcpu_refcnt = NULL; 10664 free_percpu(dev->xdp_bulkq); 10665 dev->xdp_bulkq = NULL; 10666 10667 /* Compatibility with error handling in drivers */ 10668 if (dev->reg_state == NETREG_UNINITIALIZED) { 10669 netdev_freemem(dev); 10670 return; 10671 } 10672 10673 BUG_ON(dev->reg_state != NETREG_UNREGISTERED); 10674 dev->reg_state = NETREG_RELEASED; 10675 10676 /* will free via device release */ 10677 put_device(&dev->dev); 10678 } 10679 EXPORT_SYMBOL(free_netdev); 10680 10681 /** 10682 * synchronize_net - Synchronize with packet receive processing 10683 * 10684 * Wait for packets currently being received to be done. 10685 * Does not block later packets from starting. 10686 */ 10687 void synchronize_net(void) 10688 { 10689 might_sleep(); 10690 if (rtnl_is_locked()) 10691 synchronize_rcu_expedited(); 10692 else 10693 synchronize_rcu(); 10694 } 10695 EXPORT_SYMBOL(synchronize_net); 10696 10697 /** 10698 * unregister_netdevice_queue - remove device from the kernel 10699 * @dev: device 10700 * @head: list 10701 * 10702 * This function shuts down a device interface and removes it 10703 * from the kernel tables. 10704 * If head not NULL, device is queued to be unregistered later. 10705 * 10706 * Callers must hold the rtnl semaphore. You may want 10707 * unregister_netdev() instead of this. 10708 */ 10709 10710 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) 10711 { 10712 ASSERT_RTNL(); 10713 10714 if (head) { 10715 list_move_tail(&dev->unreg_list, head); 10716 } else { 10717 rollback_registered(dev); 10718 /* Finish processing unregister after unlock */ 10719 net_set_todo(dev); 10720 } 10721 } 10722 EXPORT_SYMBOL(unregister_netdevice_queue); 10723 10724 /** 10725 * unregister_netdevice_many - unregister many devices 10726 * @head: list of devices 10727 * 10728 * Note: As most callers use a stack allocated list_head, 10729 * we force a list_del() to make sure stack wont be corrupted later. 10730 */ 10731 void unregister_netdevice_many(struct list_head *head) 10732 { 10733 struct net_device *dev; 10734 10735 if (!list_empty(head)) { 10736 rollback_registered_many(head); 10737 list_for_each_entry(dev, head, unreg_list) 10738 net_set_todo(dev); 10739 list_del(head); 10740 } 10741 } 10742 EXPORT_SYMBOL(unregister_netdevice_many); 10743 10744 /** 10745 * unregister_netdev - remove device from the kernel 10746 * @dev: device 10747 * 10748 * This function shuts down a device interface and removes it 10749 * from the kernel tables. 10750 * 10751 * This is just a wrapper for unregister_netdevice that takes 10752 * the rtnl semaphore. In general you want to use this and not 10753 * unregister_netdevice. 10754 */ 10755 void unregister_netdev(struct net_device *dev) 10756 { 10757 rtnl_lock(); 10758 unregister_netdevice(dev); 10759 rtnl_unlock(); 10760 } 10761 EXPORT_SYMBOL(unregister_netdev); 10762 10763 /** 10764 * dev_change_net_namespace - move device to different nethost namespace 10765 * @dev: device 10766 * @net: network namespace 10767 * @pat: If not NULL name pattern to try if the current device name 10768 * is already taken in the destination network namespace. 10769 * 10770 * This function shuts down a device interface and moves it 10771 * to a new network namespace. On success 0 is returned, on 10772 * a failure a netagive errno code is returned. 10773 * 10774 * Callers must hold the rtnl semaphore. 10775 */ 10776 10777 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat) 10778 { 10779 struct net *net_old = dev_net(dev); 10780 int err, new_nsid, new_ifindex; 10781 10782 ASSERT_RTNL(); 10783 10784 /* Don't allow namespace local devices to be moved. */ 10785 err = -EINVAL; 10786 if (dev->features & NETIF_F_NETNS_LOCAL) 10787 goto out; 10788 10789 /* Ensure the device has been registrered */ 10790 if (dev->reg_state != NETREG_REGISTERED) 10791 goto out; 10792 10793 /* Get out if there is nothing todo */ 10794 err = 0; 10795 if (net_eq(net_old, net)) 10796 goto out; 10797 10798 /* Pick the destination device name, and ensure 10799 * we can use it in the destination network namespace. 10800 */ 10801 err = -EEXIST; 10802 if (__dev_get_by_name(net, dev->name)) { 10803 /* We get here if we can't use the current device name */ 10804 if (!pat) 10805 goto out; 10806 err = dev_get_valid_name(net, dev, pat); 10807 if (err < 0) 10808 goto out; 10809 } 10810 10811 /* 10812 * And now a mini version of register_netdevice unregister_netdevice. 10813 */ 10814 10815 /* If device is running close it first. */ 10816 dev_close(dev); 10817 10818 /* And unlink it from device chain */ 10819 unlist_netdevice(dev); 10820 10821 synchronize_net(); 10822 10823 /* Shutdown queueing discipline. */ 10824 dev_shutdown(dev); 10825 10826 /* Notify protocols, that we are about to destroy 10827 * this device. They should clean all the things. 10828 * 10829 * Note that dev->reg_state stays at NETREG_REGISTERED. 10830 * This is wanted because this way 8021q and macvlan know 10831 * the device is just moving and can keep their slaves up. 10832 */ 10833 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10834 rcu_barrier(); 10835 10836 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL); 10837 /* If there is an ifindex conflict assign a new one */ 10838 if (__dev_get_by_index(net, dev->ifindex)) 10839 new_ifindex = dev_new_index(net); 10840 else 10841 new_ifindex = dev->ifindex; 10842 10843 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid, 10844 new_ifindex); 10845 10846 /* 10847 * Flush the unicast and multicast chains 10848 */ 10849 dev_uc_flush(dev); 10850 dev_mc_flush(dev); 10851 10852 /* Send a netdev-removed uevent to the old namespace */ 10853 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); 10854 netdev_adjacent_del_links(dev); 10855 10856 /* Move per-net netdevice notifiers that are following the netdevice */ 10857 move_netdevice_notifiers_dev_net(dev, net); 10858 10859 /* Actually switch the network namespace */ 10860 dev_net_set(dev, net); 10861 dev->ifindex = new_ifindex; 10862 10863 /* Send a netdev-add uevent to the new namespace */ 10864 kobject_uevent(&dev->dev.kobj, KOBJ_ADD); 10865 netdev_adjacent_add_links(dev); 10866 10867 /* Fixup kobjects */ 10868 err = device_rename(&dev->dev, dev->name); 10869 WARN_ON(err); 10870 10871 /* Adapt owner in case owning user namespace of target network 10872 * namespace is different from the original one. 10873 */ 10874 err = netdev_change_owner(dev, net_old, net); 10875 WARN_ON(err); 10876 10877 /* Add the device back in the hashes */ 10878 list_netdevice(dev); 10879 10880 /* Notify protocols, that a new device appeared. */ 10881 call_netdevice_notifiers(NETDEV_REGISTER, dev); 10882 10883 /* 10884 * Prevent userspace races by waiting until the network 10885 * device is fully setup before sending notifications. 10886 */ 10887 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); 10888 10889 synchronize_net(); 10890 err = 0; 10891 out: 10892 return err; 10893 } 10894 EXPORT_SYMBOL_GPL(dev_change_net_namespace); 10895 10896 static int dev_cpu_dead(unsigned int oldcpu) 10897 { 10898 struct sk_buff **list_skb; 10899 struct sk_buff *skb; 10900 unsigned int cpu; 10901 struct softnet_data *sd, *oldsd, *remsd = NULL; 10902 10903 local_irq_disable(); 10904 cpu = smp_processor_id(); 10905 sd = &per_cpu(softnet_data, cpu); 10906 oldsd = &per_cpu(softnet_data, oldcpu); 10907 10908 /* Find end of our completion_queue. */ 10909 list_skb = &sd->completion_queue; 10910 while (*list_skb) 10911 list_skb = &(*list_skb)->next; 10912 /* Append completion queue from offline CPU. */ 10913 *list_skb = oldsd->completion_queue; 10914 oldsd->completion_queue = NULL; 10915 10916 /* Append output queue from offline CPU. */ 10917 if (oldsd->output_queue) { 10918 *sd->output_queue_tailp = oldsd->output_queue; 10919 sd->output_queue_tailp = oldsd->output_queue_tailp; 10920 oldsd->output_queue = NULL; 10921 oldsd->output_queue_tailp = &oldsd->output_queue; 10922 } 10923 /* Append NAPI poll list from offline CPU, with one exception : 10924 * process_backlog() must be called by cpu owning percpu backlog. 10925 * We properly handle process_queue & input_pkt_queue later. 10926 */ 10927 while (!list_empty(&oldsd->poll_list)) { 10928 struct napi_struct *napi = list_first_entry(&oldsd->poll_list, 10929 struct napi_struct, 10930 poll_list); 10931 10932 list_del_init(&napi->poll_list); 10933 if (napi->poll == process_backlog) 10934 napi->state = 0; 10935 else 10936 ____napi_schedule(sd, napi); 10937 } 10938 10939 raise_softirq_irqoff(NET_TX_SOFTIRQ); 10940 local_irq_enable(); 10941 10942 #ifdef CONFIG_RPS 10943 remsd = oldsd->rps_ipi_list; 10944 oldsd->rps_ipi_list = NULL; 10945 #endif 10946 /* send out pending IPI's on offline CPU */ 10947 net_rps_send_ipi(remsd); 10948 10949 /* Process offline CPU's input_pkt_queue */ 10950 while ((skb = __skb_dequeue(&oldsd->process_queue))) { 10951 netif_rx_ni(skb); 10952 input_queue_head_incr(oldsd); 10953 } 10954 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { 10955 netif_rx_ni(skb); 10956 input_queue_head_incr(oldsd); 10957 } 10958 10959 return 0; 10960 } 10961 10962 /** 10963 * netdev_increment_features - increment feature set by one 10964 * @all: current feature set 10965 * @one: new feature set 10966 * @mask: mask feature set 10967 * 10968 * Computes a new feature set after adding a device with feature set 10969 * @one to the master device with current feature set @all. Will not 10970 * enable anything that is off in @mask. Returns the new feature set. 10971 */ 10972 netdev_features_t netdev_increment_features(netdev_features_t all, 10973 netdev_features_t one, netdev_features_t mask) 10974 { 10975 if (mask & NETIF_F_HW_CSUM) 10976 mask |= NETIF_F_CSUM_MASK; 10977 mask |= NETIF_F_VLAN_CHALLENGED; 10978 10979 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; 10980 all &= one | ~NETIF_F_ALL_FOR_ALL; 10981 10982 /* If one device supports hw checksumming, set for all. */ 10983 if (all & NETIF_F_HW_CSUM) 10984 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); 10985 10986 return all; 10987 } 10988 EXPORT_SYMBOL(netdev_increment_features); 10989 10990 static struct hlist_head * __net_init netdev_create_hash(void) 10991 { 10992 int i; 10993 struct hlist_head *hash; 10994 10995 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL); 10996 if (hash != NULL) 10997 for (i = 0; i < NETDEV_HASHENTRIES; i++) 10998 INIT_HLIST_HEAD(&hash[i]); 10999 11000 return hash; 11001 } 11002 11003 /* Initialize per network namespace state */ 11004 static int __net_init netdev_init(struct net *net) 11005 { 11006 BUILD_BUG_ON(GRO_HASH_BUCKETS > 11007 8 * sizeof_field(struct napi_struct, gro_bitmask)); 11008 11009 if (net != &init_net) 11010 INIT_LIST_HEAD(&net->dev_base_head); 11011 11012 net->dev_name_head = netdev_create_hash(); 11013 if (net->dev_name_head == NULL) 11014 goto err_name; 11015 11016 net->dev_index_head = netdev_create_hash(); 11017 if (net->dev_index_head == NULL) 11018 goto err_idx; 11019 11020 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain); 11021 11022 return 0; 11023 11024 err_idx: 11025 kfree(net->dev_name_head); 11026 err_name: 11027 return -ENOMEM; 11028 } 11029 11030 /** 11031 * netdev_drivername - network driver for the device 11032 * @dev: network device 11033 * 11034 * Determine network driver for device. 11035 */ 11036 const char *netdev_drivername(const struct net_device *dev) 11037 { 11038 const struct device_driver *driver; 11039 const struct device *parent; 11040 const char *empty = ""; 11041 11042 parent = dev->dev.parent; 11043 if (!parent) 11044 return empty; 11045 11046 driver = parent->driver; 11047 if (driver && driver->name) 11048 return driver->name; 11049 return empty; 11050 } 11051 11052 static void __netdev_printk(const char *level, const struct net_device *dev, 11053 struct va_format *vaf) 11054 { 11055 if (dev && dev->dev.parent) { 11056 dev_printk_emit(level[1] - '0', 11057 dev->dev.parent, 11058 "%s %s %s%s: %pV", 11059 dev_driver_string(dev->dev.parent), 11060 dev_name(dev->dev.parent), 11061 netdev_name(dev), netdev_reg_state(dev), 11062 vaf); 11063 } else if (dev) { 11064 printk("%s%s%s: %pV", 11065 level, netdev_name(dev), netdev_reg_state(dev), vaf); 11066 } else { 11067 printk("%s(NULL net_device): %pV", level, vaf); 11068 } 11069 } 11070 11071 void netdev_printk(const char *level, const struct net_device *dev, 11072 const char *format, ...) 11073 { 11074 struct va_format vaf; 11075 va_list args; 11076 11077 va_start(args, format); 11078 11079 vaf.fmt = format; 11080 vaf.va = &args; 11081 11082 __netdev_printk(level, dev, &vaf); 11083 11084 va_end(args); 11085 } 11086 EXPORT_SYMBOL(netdev_printk); 11087 11088 #define define_netdev_printk_level(func, level) \ 11089 void func(const struct net_device *dev, const char *fmt, ...) \ 11090 { \ 11091 struct va_format vaf; \ 11092 va_list args; \ 11093 \ 11094 va_start(args, fmt); \ 11095 \ 11096 vaf.fmt = fmt; \ 11097 vaf.va = &args; \ 11098 \ 11099 __netdev_printk(level, dev, &vaf); \ 11100 \ 11101 va_end(args); \ 11102 } \ 11103 EXPORT_SYMBOL(func); 11104 11105 define_netdev_printk_level(netdev_emerg, KERN_EMERG); 11106 define_netdev_printk_level(netdev_alert, KERN_ALERT); 11107 define_netdev_printk_level(netdev_crit, KERN_CRIT); 11108 define_netdev_printk_level(netdev_err, KERN_ERR); 11109 define_netdev_printk_level(netdev_warn, KERN_WARNING); 11110 define_netdev_printk_level(netdev_notice, KERN_NOTICE); 11111 define_netdev_printk_level(netdev_info, KERN_INFO); 11112 11113 static void __net_exit netdev_exit(struct net *net) 11114 { 11115 kfree(net->dev_name_head); 11116 kfree(net->dev_index_head); 11117 if (net != &init_net) 11118 WARN_ON_ONCE(!list_empty(&net->dev_base_head)); 11119 } 11120 11121 static struct pernet_operations __net_initdata netdev_net_ops = { 11122 .init = netdev_init, 11123 .exit = netdev_exit, 11124 }; 11125 11126 static void __net_exit default_device_exit(struct net *net) 11127 { 11128 struct net_device *dev, *aux; 11129 /* 11130 * Push all migratable network devices back to the 11131 * initial network namespace 11132 */ 11133 rtnl_lock(); 11134 for_each_netdev_safe(net, dev, aux) { 11135 int err; 11136 char fb_name[IFNAMSIZ]; 11137 11138 /* Ignore unmoveable devices (i.e. loopback) */ 11139 if (dev->features & NETIF_F_NETNS_LOCAL) 11140 continue; 11141 11142 /* Leave virtual devices for the generic cleanup */ 11143 if (dev->rtnl_link_ops) 11144 continue; 11145 11146 /* Push remaining network devices to init_net */ 11147 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); 11148 if (__dev_get_by_name(&init_net, fb_name)) 11149 snprintf(fb_name, IFNAMSIZ, "dev%%d"); 11150 err = dev_change_net_namespace(dev, &init_net, fb_name); 11151 if (err) { 11152 pr_emerg("%s: failed to move %s to init_net: %d\n", 11153 __func__, dev->name, err); 11154 BUG(); 11155 } 11156 } 11157 rtnl_unlock(); 11158 } 11159 11160 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list) 11161 { 11162 /* Return with the rtnl_lock held when there are no network 11163 * devices unregistering in any network namespace in net_list. 11164 */ 11165 struct net *net; 11166 bool unregistering; 11167 DEFINE_WAIT_FUNC(wait, woken_wake_function); 11168 11169 add_wait_queue(&netdev_unregistering_wq, &wait); 11170 for (;;) { 11171 unregistering = false; 11172 rtnl_lock(); 11173 list_for_each_entry(net, net_list, exit_list) { 11174 if (net->dev_unreg_count > 0) { 11175 unregistering = true; 11176 break; 11177 } 11178 } 11179 if (!unregistering) 11180 break; 11181 __rtnl_unlock(); 11182 11183 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); 11184 } 11185 remove_wait_queue(&netdev_unregistering_wq, &wait); 11186 } 11187 11188 static void __net_exit default_device_exit_batch(struct list_head *net_list) 11189 { 11190 /* At exit all network devices most be removed from a network 11191 * namespace. Do this in the reverse order of registration. 11192 * Do this across as many network namespaces as possible to 11193 * improve batching efficiency. 11194 */ 11195 struct net_device *dev; 11196 struct net *net; 11197 LIST_HEAD(dev_kill_list); 11198 11199 /* To prevent network device cleanup code from dereferencing 11200 * loopback devices or network devices that have been freed 11201 * wait here for all pending unregistrations to complete, 11202 * before unregistring the loopback device and allowing the 11203 * network namespace be freed. 11204 * 11205 * The netdev todo list containing all network devices 11206 * unregistrations that happen in default_device_exit_batch 11207 * will run in the rtnl_unlock() at the end of 11208 * default_device_exit_batch. 11209 */ 11210 rtnl_lock_unregistering(net_list); 11211 list_for_each_entry(net, net_list, exit_list) { 11212 for_each_netdev_reverse(net, dev) { 11213 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) 11214 dev->rtnl_link_ops->dellink(dev, &dev_kill_list); 11215 else 11216 unregister_netdevice_queue(dev, &dev_kill_list); 11217 } 11218 } 11219 unregister_netdevice_many(&dev_kill_list); 11220 rtnl_unlock(); 11221 } 11222 11223 static struct pernet_operations __net_initdata default_device_ops = { 11224 .exit = default_device_exit, 11225 .exit_batch = default_device_exit_batch, 11226 }; 11227 11228 /* 11229 * Initialize the DEV module. At boot time this walks the device list and 11230 * unhooks any devices that fail to initialise (normally hardware not 11231 * present) and leaves us with a valid list of present and active devices. 11232 * 11233 */ 11234 11235 /* 11236 * This is called single threaded during boot, so no need 11237 * to take the rtnl semaphore. 11238 */ 11239 static int __init net_dev_init(void) 11240 { 11241 int i, rc = -ENOMEM; 11242 11243 BUG_ON(!dev_boot_phase); 11244 11245 if (dev_proc_init()) 11246 goto out; 11247 11248 if (netdev_kobject_init()) 11249 goto out; 11250 11251 INIT_LIST_HEAD(&ptype_all); 11252 for (i = 0; i < PTYPE_HASH_SIZE; i++) 11253 INIT_LIST_HEAD(&ptype_base[i]); 11254 11255 INIT_LIST_HEAD(&offload_base); 11256 11257 if (register_pernet_subsys(&netdev_net_ops)) 11258 goto out; 11259 11260 /* 11261 * Initialise the packet receive queues. 11262 */ 11263 11264 for_each_possible_cpu(i) { 11265 struct work_struct *flush = per_cpu_ptr(&flush_works, i); 11266 struct softnet_data *sd = &per_cpu(softnet_data, i); 11267 11268 INIT_WORK(flush, flush_backlog); 11269 11270 skb_queue_head_init(&sd->input_pkt_queue); 11271 skb_queue_head_init(&sd->process_queue); 11272 #ifdef CONFIG_XFRM_OFFLOAD 11273 skb_queue_head_init(&sd->xfrm_backlog); 11274 #endif 11275 INIT_LIST_HEAD(&sd->poll_list); 11276 sd->output_queue_tailp = &sd->output_queue; 11277 #ifdef CONFIG_RPS 11278 INIT_CSD(&sd->csd, rps_trigger_softirq, sd); 11279 sd->cpu = i; 11280 #endif 11281 11282 init_gro_hash(&sd->backlog); 11283 sd->backlog.poll = process_backlog; 11284 sd->backlog.weight = weight_p; 11285 } 11286 11287 dev_boot_phase = 0; 11288 11289 /* The loopback device is special if any other network devices 11290 * is present in a network namespace the loopback device must 11291 * be present. Since we now dynamically allocate and free the 11292 * loopback device ensure this invariant is maintained by 11293 * keeping the loopback device as the first device on the 11294 * list of network devices. Ensuring the loopback devices 11295 * is the first device that appears and the last network device 11296 * that disappears. 11297 */ 11298 if (register_pernet_device(&loopback_net_ops)) 11299 goto out; 11300 11301 if (register_pernet_device(&default_device_ops)) 11302 goto out; 11303 11304 open_softirq(NET_TX_SOFTIRQ, net_tx_action); 11305 open_softirq(NET_RX_SOFTIRQ, net_rx_action); 11306 11307 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", 11308 NULL, dev_cpu_dead); 11309 WARN_ON(rc < 0); 11310 rc = 0; 11311 out: 11312 return rc; 11313 } 11314 11315 subsys_initcall(net_dev_init); 11316