1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * NET3 Protocol independent device support routines. 4 * 5 * Derived from the non IP parts of dev.c 1.0.19 6 * Authors: Ross Biro 7 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 8 * Mark Evans, <evansmp@uhura.aston.ac.uk> 9 * 10 * Additional Authors: 11 * Florian la Roche <rzsfl@rz.uni-sb.de> 12 * Alan Cox <gw4pts@gw4pts.ampr.org> 13 * David Hinds <dahinds@users.sourceforge.net> 14 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru> 15 * Adam Sulmicki <adam@cfar.umd.edu> 16 * Pekka Riikonen <priikone@poesidon.pspt.fi> 17 * 18 * Changes: 19 * D.J. Barrow : Fixed bug where dev->refcnt gets set 20 * to 2 if register_netdev gets called 21 * before net_dev_init & also removed a 22 * few lines of code in the process. 23 * Alan Cox : device private ioctl copies fields back. 24 * Alan Cox : Transmit queue code does relevant 25 * stunts to keep the queue safe. 26 * Alan Cox : Fixed double lock. 27 * Alan Cox : Fixed promisc NULL pointer trap 28 * ???????? : Support the full private ioctl range 29 * Alan Cox : Moved ioctl permission check into 30 * drivers 31 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI 32 * Alan Cox : 100 backlog just doesn't cut it when 33 * you start doing multicast video 8) 34 * Alan Cox : Rewrote net_bh and list manager. 35 * Alan Cox : Fix ETH_P_ALL echoback lengths. 36 * Alan Cox : Took out transmit every packet pass 37 * Saved a few bytes in the ioctl handler 38 * Alan Cox : Network driver sets packet type before 39 * calling netif_rx. Saves a function 40 * call a packet. 41 * Alan Cox : Hashed net_bh() 42 * Richard Kooijman: Timestamp fixes. 43 * Alan Cox : Wrong field in SIOCGIFDSTADDR 44 * Alan Cox : Device lock protection. 45 * Alan Cox : Fixed nasty side effect of device close 46 * changes. 47 * Rudi Cilibrasi : Pass the right thing to 48 * set_mac_address() 49 * Dave Miller : 32bit quantity for the device lock to 50 * make it work out on a Sparc. 51 * Bjorn Ekwall : Added KERNELD hack. 52 * Alan Cox : Cleaned up the backlog initialise. 53 * Craig Metz : SIOCGIFCONF fix if space for under 54 * 1 device. 55 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there 56 * is no device open function. 57 * Andi Kleen : Fix error reporting for SIOCGIFCONF 58 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF 59 * Cyrus Durgin : Cleaned for KMOD 60 * Adam Sulmicki : Bug Fix : Network Device Unload 61 * A network device unload needs to purge 62 * the backlog queue. 63 * Paul Rusty Russell : SIOCSIFNAME 64 * Pekka Riikonen : Netdev boot-time settings code 65 * Andrew Morton : Make unregister_netdevice wait 66 * indefinitely on dev->refcnt 67 * J Hadi Salim : - Backlog queue sampling 68 * - netif_rx() feedback 69 */ 70 71 #include <linux/uaccess.h> 72 #include <linux/bitops.h> 73 #include <linux/capability.h> 74 #include <linux/cpu.h> 75 #include <linux/types.h> 76 #include <linux/kernel.h> 77 #include <linux/hash.h> 78 #include <linux/slab.h> 79 #include <linux/sched.h> 80 #include <linux/sched/mm.h> 81 #include <linux/mutex.h> 82 #include <linux/rwsem.h> 83 #include <linux/string.h> 84 #include <linux/mm.h> 85 #include <linux/socket.h> 86 #include <linux/sockios.h> 87 #include <linux/errno.h> 88 #include <linux/interrupt.h> 89 #include <linux/if_ether.h> 90 #include <linux/netdevice.h> 91 #include <linux/etherdevice.h> 92 #include <linux/ethtool.h> 93 #include <linux/skbuff.h> 94 #include <linux/kthread.h> 95 #include <linux/bpf.h> 96 #include <linux/bpf_trace.h> 97 #include <net/net_namespace.h> 98 #include <net/sock.h> 99 #include <net/busy_poll.h> 100 #include <linux/rtnetlink.h> 101 #include <linux/stat.h> 102 #include <net/dsa.h> 103 #include <net/dst.h> 104 #include <net/dst_metadata.h> 105 #include <net/gro.h> 106 #include <net/pkt_sched.h> 107 #include <net/pkt_cls.h> 108 #include <net/checksum.h> 109 #include <net/xfrm.h> 110 #include <linux/highmem.h> 111 #include <linux/init.h> 112 #include <linux/module.h> 113 #include <linux/netpoll.h> 114 #include <linux/rcupdate.h> 115 #include <linux/delay.h> 116 #include <net/iw_handler.h> 117 #include <asm/current.h> 118 #include <linux/audit.h> 119 #include <linux/dmaengine.h> 120 #include <linux/err.h> 121 #include <linux/ctype.h> 122 #include <linux/if_arp.h> 123 #include <linux/if_vlan.h> 124 #include <linux/ip.h> 125 #include <net/ip.h> 126 #include <net/mpls.h> 127 #include <linux/ipv6.h> 128 #include <linux/in.h> 129 #include <linux/jhash.h> 130 #include <linux/random.h> 131 #include <trace/events/napi.h> 132 #include <trace/events/net.h> 133 #include <trace/events/skb.h> 134 #include <trace/events/qdisc.h> 135 #include <linux/inetdevice.h> 136 #include <linux/cpu_rmap.h> 137 #include <linux/static_key.h> 138 #include <linux/hashtable.h> 139 #include <linux/vmalloc.h> 140 #include <linux/if_macvlan.h> 141 #include <linux/errqueue.h> 142 #include <linux/hrtimer.h> 143 #include <linux/netfilter_ingress.h> 144 #include <linux/crash_dump.h> 145 #include <linux/sctp.h> 146 #include <net/udp_tunnel.h> 147 #include <linux/net_namespace.h> 148 #include <linux/indirect_call_wrapper.h> 149 #include <net/devlink.h> 150 #include <linux/pm_runtime.h> 151 #include <linux/prandom.h> 152 #include <linux/once_lite.h> 153 154 #include "net-sysfs.h" 155 156 #define MAX_GRO_SKBS 8 157 158 /* This should be increased if a protocol with a bigger head is added. */ 159 #define GRO_MAX_HEAD (MAX_HEADER + 128) 160 161 static DEFINE_SPINLOCK(ptype_lock); 162 static DEFINE_SPINLOCK(offload_lock); 163 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; 164 struct list_head ptype_all __read_mostly; /* Taps */ 165 static struct list_head offload_base __read_mostly; 166 167 static int netif_rx_internal(struct sk_buff *skb); 168 static int call_netdevice_notifiers_info(unsigned long val, 169 struct netdev_notifier_info *info); 170 static int call_netdevice_notifiers_extack(unsigned long val, 171 struct net_device *dev, 172 struct netlink_ext_ack *extack); 173 static struct napi_struct *napi_by_id(unsigned int napi_id); 174 175 /* 176 * The @dev_base_head list is protected by @dev_base_lock and the rtnl 177 * semaphore. 178 * 179 * Pure readers hold dev_base_lock for reading, or rcu_read_lock() 180 * 181 * Writers must hold the rtnl semaphore while they loop through the 182 * dev_base_head list, and hold dev_base_lock for writing when they do the 183 * actual updates. This allows pure readers to access the list even 184 * while a writer is preparing to update it. 185 * 186 * To put it another way, dev_base_lock is held for writing only to 187 * protect against pure readers; the rtnl semaphore provides the 188 * protection against other writers. 189 * 190 * See, for example usages, register_netdevice() and 191 * unregister_netdevice(), which must be called with the rtnl 192 * semaphore held. 193 */ 194 DEFINE_RWLOCK(dev_base_lock); 195 EXPORT_SYMBOL(dev_base_lock); 196 197 static DEFINE_MUTEX(ifalias_mutex); 198 199 /* protects napi_hash addition/deletion and napi_gen_id */ 200 static DEFINE_SPINLOCK(napi_hash_lock); 201 202 static unsigned int napi_gen_id = NR_CPUS; 203 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8); 204 205 static DECLARE_RWSEM(devnet_rename_sem); 206 207 static inline void dev_base_seq_inc(struct net *net) 208 { 209 while (++net->dev_base_seq == 0) 210 ; 211 } 212 213 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) 214 { 215 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ)); 216 217 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; 218 } 219 220 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) 221 { 222 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; 223 } 224 225 static inline void rps_lock(struct softnet_data *sd) 226 { 227 #ifdef CONFIG_RPS 228 spin_lock(&sd->input_pkt_queue.lock); 229 #endif 230 } 231 232 static inline void rps_unlock(struct softnet_data *sd) 233 { 234 #ifdef CONFIG_RPS 235 spin_unlock(&sd->input_pkt_queue.lock); 236 #endif 237 } 238 239 static struct netdev_name_node *netdev_name_node_alloc(struct net_device *dev, 240 const char *name) 241 { 242 struct netdev_name_node *name_node; 243 244 name_node = kmalloc(sizeof(*name_node), GFP_KERNEL); 245 if (!name_node) 246 return NULL; 247 INIT_HLIST_NODE(&name_node->hlist); 248 name_node->dev = dev; 249 name_node->name = name; 250 return name_node; 251 } 252 253 static struct netdev_name_node * 254 netdev_name_node_head_alloc(struct net_device *dev) 255 { 256 struct netdev_name_node *name_node; 257 258 name_node = netdev_name_node_alloc(dev, dev->name); 259 if (!name_node) 260 return NULL; 261 INIT_LIST_HEAD(&name_node->list); 262 return name_node; 263 } 264 265 static void netdev_name_node_free(struct netdev_name_node *name_node) 266 { 267 kfree(name_node); 268 } 269 270 static void netdev_name_node_add(struct net *net, 271 struct netdev_name_node *name_node) 272 { 273 hlist_add_head_rcu(&name_node->hlist, 274 dev_name_hash(net, name_node->name)); 275 } 276 277 static void netdev_name_node_del(struct netdev_name_node *name_node) 278 { 279 hlist_del_rcu(&name_node->hlist); 280 } 281 282 static struct netdev_name_node *netdev_name_node_lookup(struct net *net, 283 const char *name) 284 { 285 struct hlist_head *head = dev_name_hash(net, name); 286 struct netdev_name_node *name_node; 287 288 hlist_for_each_entry(name_node, head, hlist) 289 if (!strcmp(name_node->name, name)) 290 return name_node; 291 return NULL; 292 } 293 294 static struct netdev_name_node *netdev_name_node_lookup_rcu(struct net *net, 295 const char *name) 296 { 297 struct hlist_head *head = dev_name_hash(net, name); 298 struct netdev_name_node *name_node; 299 300 hlist_for_each_entry_rcu(name_node, head, hlist) 301 if (!strcmp(name_node->name, name)) 302 return name_node; 303 return NULL; 304 } 305 306 int netdev_name_node_alt_create(struct net_device *dev, const char *name) 307 { 308 struct netdev_name_node *name_node; 309 struct net *net = dev_net(dev); 310 311 name_node = netdev_name_node_lookup(net, name); 312 if (name_node) 313 return -EEXIST; 314 name_node = netdev_name_node_alloc(dev, name); 315 if (!name_node) 316 return -ENOMEM; 317 netdev_name_node_add(net, name_node); 318 /* The node that holds dev->name acts as a head of per-device list. */ 319 list_add_tail(&name_node->list, &dev->name_node->list); 320 321 return 0; 322 } 323 EXPORT_SYMBOL(netdev_name_node_alt_create); 324 325 static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node) 326 { 327 list_del(&name_node->list); 328 netdev_name_node_del(name_node); 329 kfree(name_node->name); 330 netdev_name_node_free(name_node); 331 } 332 333 int netdev_name_node_alt_destroy(struct net_device *dev, const char *name) 334 { 335 struct netdev_name_node *name_node; 336 struct net *net = dev_net(dev); 337 338 name_node = netdev_name_node_lookup(net, name); 339 if (!name_node) 340 return -ENOENT; 341 /* lookup might have found our primary name or a name belonging 342 * to another device. 343 */ 344 if (name_node == dev->name_node || name_node->dev != dev) 345 return -EINVAL; 346 347 __netdev_name_node_alt_destroy(name_node); 348 349 return 0; 350 } 351 EXPORT_SYMBOL(netdev_name_node_alt_destroy); 352 353 static void netdev_name_node_alt_flush(struct net_device *dev) 354 { 355 struct netdev_name_node *name_node, *tmp; 356 357 list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list) 358 __netdev_name_node_alt_destroy(name_node); 359 } 360 361 /* Device list insertion */ 362 static void list_netdevice(struct net_device *dev) 363 { 364 struct net *net = dev_net(dev); 365 366 ASSERT_RTNL(); 367 368 write_lock_bh(&dev_base_lock); 369 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); 370 netdev_name_node_add(net, dev->name_node); 371 hlist_add_head_rcu(&dev->index_hlist, 372 dev_index_hash(net, dev->ifindex)); 373 write_unlock_bh(&dev_base_lock); 374 375 dev_base_seq_inc(net); 376 } 377 378 /* Device list removal 379 * caller must respect a RCU grace period before freeing/reusing dev 380 */ 381 static void unlist_netdevice(struct net_device *dev) 382 { 383 ASSERT_RTNL(); 384 385 /* Unlink dev from the device chain */ 386 write_lock_bh(&dev_base_lock); 387 list_del_rcu(&dev->dev_list); 388 netdev_name_node_del(dev->name_node); 389 hlist_del_rcu(&dev->index_hlist); 390 write_unlock_bh(&dev_base_lock); 391 392 dev_base_seq_inc(dev_net(dev)); 393 } 394 395 /* 396 * Our notifier list 397 */ 398 399 static RAW_NOTIFIER_HEAD(netdev_chain); 400 401 /* 402 * Device drivers call our routines to queue packets here. We empty the 403 * queue in the local softnet handler. 404 */ 405 406 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); 407 EXPORT_PER_CPU_SYMBOL(softnet_data); 408 409 #ifdef CONFIG_LOCKDEP 410 /* 411 * register_netdevice() inits txq->_xmit_lock and sets lockdep class 412 * according to dev->type 413 */ 414 static const unsigned short netdev_lock_type[] = { 415 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, 416 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, 417 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, 418 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, 419 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, 420 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, 421 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, 422 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, 423 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, 424 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, 425 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, 426 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, 427 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, 428 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, 429 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; 430 431 static const char *const netdev_lock_name[] = { 432 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", 433 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", 434 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", 435 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", 436 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", 437 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", 438 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", 439 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", 440 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", 441 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", 442 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", 443 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", 444 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", 445 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", 446 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; 447 448 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; 449 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; 450 451 static inline unsigned short netdev_lock_pos(unsigned short dev_type) 452 { 453 int i; 454 455 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) 456 if (netdev_lock_type[i] == dev_type) 457 return i; 458 /* the last key is used by default */ 459 return ARRAY_SIZE(netdev_lock_type) - 1; 460 } 461 462 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 463 unsigned short dev_type) 464 { 465 int i; 466 467 i = netdev_lock_pos(dev_type); 468 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], 469 netdev_lock_name[i]); 470 } 471 472 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 473 { 474 int i; 475 476 i = netdev_lock_pos(dev->type); 477 lockdep_set_class_and_name(&dev->addr_list_lock, 478 &netdev_addr_lock_key[i], 479 netdev_lock_name[i]); 480 } 481 #else 482 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 483 unsigned short dev_type) 484 { 485 } 486 487 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 488 { 489 } 490 #endif 491 492 /******************************************************************************* 493 * 494 * Protocol management and registration routines 495 * 496 *******************************************************************************/ 497 498 499 /* 500 * Add a protocol ID to the list. Now that the input handler is 501 * smarter we can dispense with all the messy stuff that used to be 502 * here. 503 * 504 * BEWARE!!! Protocol handlers, mangling input packets, 505 * MUST BE last in hash buckets and checking protocol handlers 506 * MUST start from promiscuous ptype_all chain in net_bh. 507 * It is true now, do not change it. 508 * Explanation follows: if protocol handler, mangling packet, will 509 * be the first on list, it is not able to sense, that packet 510 * is cloned and should be copied-on-write, so that it will 511 * change it and subsequent readers will get broken packet. 512 * --ANK (980803) 513 */ 514 515 static inline struct list_head *ptype_head(const struct packet_type *pt) 516 { 517 if (pt->type == htons(ETH_P_ALL)) 518 return pt->dev ? &pt->dev->ptype_all : &ptype_all; 519 else 520 return pt->dev ? &pt->dev->ptype_specific : 521 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; 522 } 523 524 /** 525 * dev_add_pack - add packet handler 526 * @pt: packet type declaration 527 * 528 * Add a protocol handler to the networking stack. The passed &packet_type 529 * is linked into kernel lists and may not be freed until it has been 530 * removed from the kernel lists. 531 * 532 * This call does not sleep therefore it can not 533 * guarantee all CPU's that are in middle of receiving packets 534 * will see the new packet type (until the next received packet). 535 */ 536 537 void dev_add_pack(struct packet_type *pt) 538 { 539 struct list_head *head = ptype_head(pt); 540 541 spin_lock(&ptype_lock); 542 list_add_rcu(&pt->list, head); 543 spin_unlock(&ptype_lock); 544 } 545 EXPORT_SYMBOL(dev_add_pack); 546 547 /** 548 * __dev_remove_pack - remove packet handler 549 * @pt: packet type declaration 550 * 551 * Remove a protocol handler that was previously added to the kernel 552 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 553 * from the kernel lists and can be freed or reused once this function 554 * returns. 555 * 556 * The packet type might still be in use by receivers 557 * and must not be freed until after all the CPU's have gone 558 * through a quiescent state. 559 */ 560 void __dev_remove_pack(struct packet_type *pt) 561 { 562 struct list_head *head = ptype_head(pt); 563 struct packet_type *pt1; 564 565 spin_lock(&ptype_lock); 566 567 list_for_each_entry(pt1, head, list) { 568 if (pt == pt1) { 569 list_del_rcu(&pt->list); 570 goto out; 571 } 572 } 573 574 pr_warn("dev_remove_pack: %p not found\n", pt); 575 out: 576 spin_unlock(&ptype_lock); 577 } 578 EXPORT_SYMBOL(__dev_remove_pack); 579 580 /** 581 * dev_remove_pack - remove packet handler 582 * @pt: packet type declaration 583 * 584 * Remove a protocol handler that was previously added to the kernel 585 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 586 * from the kernel lists and can be freed or reused once this function 587 * returns. 588 * 589 * This call sleeps to guarantee that no CPU is looking at the packet 590 * type after return. 591 */ 592 void dev_remove_pack(struct packet_type *pt) 593 { 594 __dev_remove_pack(pt); 595 596 synchronize_net(); 597 } 598 EXPORT_SYMBOL(dev_remove_pack); 599 600 601 /** 602 * dev_add_offload - register offload handlers 603 * @po: protocol offload declaration 604 * 605 * Add protocol offload handlers to the networking stack. The passed 606 * &proto_offload is linked into kernel lists and may not be freed until 607 * it has been removed from the kernel lists. 608 * 609 * This call does not sleep therefore it can not 610 * guarantee all CPU's that are in middle of receiving packets 611 * will see the new offload handlers (until the next received packet). 612 */ 613 void dev_add_offload(struct packet_offload *po) 614 { 615 struct packet_offload *elem; 616 617 spin_lock(&offload_lock); 618 list_for_each_entry(elem, &offload_base, list) { 619 if (po->priority < elem->priority) 620 break; 621 } 622 list_add_rcu(&po->list, elem->list.prev); 623 spin_unlock(&offload_lock); 624 } 625 EXPORT_SYMBOL(dev_add_offload); 626 627 /** 628 * __dev_remove_offload - remove offload handler 629 * @po: packet offload declaration 630 * 631 * Remove a protocol offload handler that was previously added to the 632 * kernel offload handlers by dev_add_offload(). The passed &offload_type 633 * is removed from the kernel lists and can be freed or reused once this 634 * function returns. 635 * 636 * The packet type might still be in use by receivers 637 * and must not be freed until after all the CPU's have gone 638 * through a quiescent state. 639 */ 640 static void __dev_remove_offload(struct packet_offload *po) 641 { 642 struct list_head *head = &offload_base; 643 struct packet_offload *po1; 644 645 spin_lock(&offload_lock); 646 647 list_for_each_entry(po1, head, list) { 648 if (po == po1) { 649 list_del_rcu(&po->list); 650 goto out; 651 } 652 } 653 654 pr_warn("dev_remove_offload: %p not found\n", po); 655 out: 656 spin_unlock(&offload_lock); 657 } 658 659 /** 660 * dev_remove_offload - remove packet offload handler 661 * @po: packet offload declaration 662 * 663 * Remove a packet offload handler that was previously added to the kernel 664 * offload handlers by dev_add_offload(). The passed &offload_type is 665 * removed from the kernel lists and can be freed or reused once this 666 * function returns. 667 * 668 * This call sleeps to guarantee that no CPU is looking at the packet 669 * type after return. 670 */ 671 void dev_remove_offload(struct packet_offload *po) 672 { 673 __dev_remove_offload(po); 674 675 synchronize_net(); 676 } 677 EXPORT_SYMBOL(dev_remove_offload); 678 679 /****************************************************************************** 680 * 681 * Device Boot-time Settings Routines 682 * 683 ******************************************************************************/ 684 685 /* Boot time configuration table */ 686 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX]; 687 688 /** 689 * netdev_boot_setup_add - add new setup entry 690 * @name: name of the device 691 * @map: configured settings for the device 692 * 693 * Adds new setup entry to the dev_boot_setup list. The function 694 * returns 0 on error and 1 on success. This is a generic routine to 695 * all netdevices. 696 */ 697 static int netdev_boot_setup_add(char *name, struct ifmap *map) 698 { 699 struct netdev_boot_setup *s; 700 int i; 701 702 s = dev_boot_setup; 703 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { 704 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') { 705 memset(s[i].name, 0, sizeof(s[i].name)); 706 strlcpy(s[i].name, name, IFNAMSIZ); 707 memcpy(&s[i].map, map, sizeof(s[i].map)); 708 break; 709 } 710 } 711 712 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1; 713 } 714 715 /** 716 * netdev_boot_setup_check - check boot time settings 717 * @dev: the netdevice 718 * 719 * Check boot time settings for the device. 720 * The found settings are set for the device to be used 721 * later in the device probing. 722 * Returns 0 if no settings found, 1 if they are. 723 */ 724 int netdev_boot_setup_check(struct net_device *dev) 725 { 726 struct netdev_boot_setup *s = dev_boot_setup; 727 int i; 728 729 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { 730 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' && 731 !strcmp(dev->name, s[i].name)) { 732 dev->irq = s[i].map.irq; 733 dev->base_addr = s[i].map.base_addr; 734 dev->mem_start = s[i].map.mem_start; 735 dev->mem_end = s[i].map.mem_end; 736 return 1; 737 } 738 } 739 return 0; 740 } 741 EXPORT_SYMBOL(netdev_boot_setup_check); 742 743 744 /** 745 * netdev_boot_base - get address from boot time settings 746 * @prefix: prefix for network device 747 * @unit: id for network device 748 * 749 * Check boot time settings for the base address of device. 750 * The found settings are set for the device to be used 751 * later in the device probing. 752 * Returns 0 if no settings found. 753 */ 754 unsigned long netdev_boot_base(const char *prefix, int unit) 755 { 756 const struct netdev_boot_setup *s = dev_boot_setup; 757 char name[IFNAMSIZ]; 758 int i; 759 760 sprintf(name, "%s%d", prefix, unit); 761 762 /* 763 * If device already registered then return base of 1 764 * to indicate not to probe for this interface 765 */ 766 if (__dev_get_by_name(&init_net, name)) 767 return 1; 768 769 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) 770 if (!strcmp(name, s[i].name)) 771 return s[i].map.base_addr; 772 return 0; 773 } 774 775 /* 776 * Saves at boot time configured settings for any netdevice. 777 */ 778 int __init netdev_boot_setup(char *str) 779 { 780 int ints[5]; 781 struct ifmap map; 782 783 str = get_options(str, ARRAY_SIZE(ints), ints); 784 if (!str || !*str) 785 return 0; 786 787 /* Save settings */ 788 memset(&map, 0, sizeof(map)); 789 if (ints[0] > 0) 790 map.irq = ints[1]; 791 if (ints[0] > 1) 792 map.base_addr = ints[2]; 793 if (ints[0] > 2) 794 map.mem_start = ints[3]; 795 if (ints[0] > 3) 796 map.mem_end = ints[4]; 797 798 /* Add new entry to the list */ 799 return netdev_boot_setup_add(str, &map); 800 } 801 802 __setup("netdev=", netdev_boot_setup); 803 804 /******************************************************************************* 805 * 806 * Device Interface Subroutines 807 * 808 *******************************************************************************/ 809 810 /** 811 * dev_get_iflink - get 'iflink' value of a interface 812 * @dev: targeted interface 813 * 814 * Indicates the ifindex the interface is linked to. 815 * Physical interfaces have the same 'ifindex' and 'iflink' values. 816 */ 817 818 int dev_get_iflink(const struct net_device *dev) 819 { 820 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink) 821 return dev->netdev_ops->ndo_get_iflink(dev); 822 823 return dev->ifindex; 824 } 825 EXPORT_SYMBOL(dev_get_iflink); 826 827 /** 828 * dev_fill_metadata_dst - Retrieve tunnel egress information. 829 * @dev: targeted interface 830 * @skb: The packet. 831 * 832 * For better visibility of tunnel traffic OVS needs to retrieve 833 * egress tunnel information for a packet. Following API allows 834 * user to get this info. 835 */ 836 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) 837 { 838 struct ip_tunnel_info *info; 839 840 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst) 841 return -EINVAL; 842 843 info = skb_tunnel_info_unclone(skb); 844 if (!info) 845 return -ENOMEM; 846 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX))) 847 return -EINVAL; 848 849 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb); 850 } 851 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst); 852 853 static struct net_device_path *dev_fwd_path(struct net_device_path_stack *stack) 854 { 855 int k = stack->num_paths++; 856 857 if (WARN_ON_ONCE(k >= NET_DEVICE_PATH_STACK_MAX)) 858 return NULL; 859 860 return &stack->path[k]; 861 } 862 863 int dev_fill_forward_path(const struct net_device *dev, const u8 *daddr, 864 struct net_device_path_stack *stack) 865 { 866 const struct net_device *last_dev; 867 struct net_device_path_ctx ctx = { 868 .dev = dev, 869 .daddr = daddr, 870 }; 871 struct net_device_path *path; 872 int ret = 0; 873 874 stack->num_paths = 0; 875 while (ctx.dev && ctx.dev->netdev_ops->ndo_fill_forward_path) { 876 last_dev = ctx.dev; 877 path = dev_fwd_path(stack); 878 if (!path) 879 return -1; 880 881 memset(path, 0, sizeof(struct net_device_path)); 882 ret = ctx.dev->netdev_ops->ndo_fill_forward_path(&ctx, path); 883 if (ret < 0) 884 return -1; 885 886 if (WARN_ON_ONCE(last_dev == ctx.dev)) 887 return -1; 888 } 889 path = dev_fwd_path(stack); 890 if (!path) 891 return -1; 892 path->type = DEV_PATH_ETHERNET; 893 path->dev = ctx.dev; 894 895 return ret; 896 } 897 EXPORT_SYMBOL_GPL(dev_fill_forward_path); 898 899 /** 900 * __dev_get_by_name - find a device by its name 901 * @net: the applicable net namespace 902 * @name: name to find 903 * 904 * Find an interface by name. Must be called under RTNL semaphore 905 * or @dev_base_lock. If the name is found a pointer to the device 906 * is returned. If the name is not found then %NULL is returned. The 907 * reference counters are not incremented so the caller must be 908 * careful with locks. 909 */ 910 911 struct net_device *__dev_get_by_name(struct net *net, const char *name) 912 { 913 struct netdev_name_node *node_name; 914 915 node_name = netdev_name_node_lookup(net, name); 916 return node_name ? node_name->dev : NULL; 917 } 918 EXPORT_SYMBOL(__dev_get_by_name); 919 920 /** 921 * dev_get_by_name_rcu - find a device by its name 922 * @net: the applicable net namespace 923 * @name: name to find 924 * 925 * Find an interface by name. 926 * If the name is found a pointer to the device is returned. 927 * If the name is not found then %NULL is returned. 928 * The reference counters are not incremented so the caller must be 929 * careful with locks. The caller must hold RCU lock. 930 */ 931 932 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) 933 { 934 struct netdev_name_node *node_name; 935 936 node_name = netdev_name_node_lookup_rcu(net, name); 937 return node_name ? node_name->dev : NULL; 938 } 939 EXPORT_SYMBOL(dev_get_by_name_rcu); 940 941 /** 942 * dev_get_by_name - find a device by its name 943 * @net: the applicable net namespace 944 * @name: name to find 945 * 946 * Find an interface by name. This can be called from any 947 * context and does its own locking. The returned handle has 948 * the usage count incremented and the caller must use dev_put() to 949 * release it when it is no longer needed. %NULL is returned if no 950 * matching device is found. 951 */ 952 953 struct net_device *dev_get_by_name(struct net *net, const char *name) 954 { 955 struct net_device *dev; 956 957 rcu_read_lock(); 958 dev = dev_get_by_name_rcu(net, name); 959 if (dev) 960 dev_hold(dev); 961 rcu_read_unlock(); 962 return dev; 963 } 964 EXPORT_SYMBOL(dev_get_by_name); 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 has not 973 * had its reference counter increased so the caller must be careful 974 * about locking. The caller must hold either the RTNL semaphore 975 * or @dev_base_lock. 976 */ 977 978 struct net_device *__dev_get_by_index(struct net *net, int ifindex) 979 { 980 struct net_device *dev; 981 struct hlist_head *head = dev_index_hash(net, ifindex); 982 983 hlist_for_each_entry(dev, head, index_hlist) 984 if (dev->ifindex == ifindex) 985 return dev; 986 987 return NULL; 988 } 989 EXPORT_SYMBOL(__dev_get_by_index); 990 991 /** 992 * dev_get_by_index_rcu - find a device by its ifindex 993 * @net: the applicable net namespace 994 * @ifindex: index of device 995 * 996 * Search for an interface by index. Returns %NULL if the device 997 * is not found or a pointer to the device. The device has not 998 * had its reference counter increased so the caller must be careful 999 * about locking. The caller must hold RCU lock. 1000 */ 1001 1002 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) 1003 { 1004 struct net_device *dev; 1005 struct hlist_head *head = dev_index_hash(net, ifindex); 1006 1007 hlist_for_each_entry_rcu(dev, head, index_hlist) 1008 if (dev->ifindex == ifindex) 1009 return dev; 1010 1011 return NULL; 1012 } 1013 EXPORT_SYMBOL(dev_get_by_index_rcu); 1014 1015 1016 /** 1017 * dev_get_by_index - find a device by its ifindex 1018 * @net: the applicable net namespace 1019 * @ifindex: index of device 1020 * 1021 * Search for an interface by index. Returns NULL if the device 1022 * is not found or a pointer to the device. The device returned has 1023 * had a reference added and the pointer is safe until the user calls 1024 * dev_put to indicate they have finished with it. 1025 */ 1026 1027 struct net_device *dev_get_by_index(struct net *net, int ifindex) 1028 { 1029 struct net_device *dev; 1030 1031 rcu_read_lock(); 1032 dev = dev_get_by_index_rcu(net, ifindex); 1033 if (dev) 1034 dev_hold(dev); 1035 rcu_read_unlock(); 1036 return dev; 1037 } 1038 EXPORT_SYMBOL(dev_get_by_index); 1039 1040 /** 1041 * dev_get_by_napi_id - find a device by napi_id 1042 * @napi_id: ID of the NAPI struct 1043 * 1044 * Search for an interface by NAPI ID. Returns %NULL if the device 1045 * is not found or a pointer to the device. The device has not had 1046 * its reference counter increased so the caller must be careful 1047 * about locking. The caller must hold RCU lock. 1048 */ 1049 1050 struct net_device *dev_get_by_napi_id(unsigned int napi_id) 1051 { 1052 struct napi_struct *napi; 1053 1054 WARN_ON_ONCE(!rcu_read_lock_held()); 1055 1056 if (napi_id < MIN_NAPI_ID) 1057 return NULL; 1058 1059 napi = napi_by_id(napi_id); 1060 1061 return napi ? napi->dev : NULL; 1062 } 1063 EXPORT_SYMBOL(dev_get_by_napi_id); 1064 1065 /** 1066 * netdev_get_name - get a netdevice name, knowing its ifindex. 1067 * @net: network namespace 1068 * @name: a pointer to the buffer where the name will be stored. 1069 * @ifindex: the ifindex of the interface to get the name from. 1070 */ 1071 int netdev_get_name(struct net *net, char *name, int ifindex) 1072 { 1073 struct net_device *dev; 1074 int ret; 1075 1076 down_read(&devnet_rename_sem); 1077 rcu_read_lock(); 1078 1079 dev = dev_get_by_index_rcu(net, ifindex); 1080 if (!dev) { 1081 ret = -ENODEV; 1082 goto out; 1083 } 1084 1085 strcpy(name, dev->name); 1086 1087 ret = 0; 1088 out: 1089 rcu_read_unlock(); 1090 up_read(&devnet_rename_sem); 1091 return ret; 1092 } 1093 1094 /** 1095 * dev_getbyhwaddr_rcu - find a device by its hardware address 1096 * @net: the applicable net namespace 1097 * @type: media type of device 1098 * @ha: hardware address 1099 * 1100 * Search for an interface by MAC address. Returns NULL if the device 1101 * is not found or a pointer to the device. 1102 * The caller must hold RCU or RTNL. 1103 * The returned device has not had its ref count increased 1104 * and the caller must therefore be careful about locking 1105 * 1106 */ 1107 1108 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, 1109 const char *ha) 1110 { 1111 struct net_device *dev; 1112 1113 for_each_netdev_rcu(net, dev) 1114 if (dev->type == type && 1115 !memcmp(dev->dev_addr, ha, dev->addr_len)) 1116 return dev; 1117 1118 return NULL; 1119 } 1120 EXPORT_SYMBOL(dev_getbyhwaddr_rcu); 1121 1122 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) 1123 { 1124 struct net_device *dev, *ret = NULL; 1125 1126 rcu_read_lock(); 1127 for_each_netdev_rcu(net, dev) 1128 if (dev->type == type) { 1129 dev_hold(dev); 1130 ret = dev; 1131 break; 1132 } 1133 rcu_read_unlock(); 1134 return ret; 1135 } 1136 EXPORT_SYMBOL(dev_getfirstbyhwtype); 1137 1138 /** 1139 * __dev_get_by_flags - find any device with given flags 1140 * @net: the applicable net namespace 1141 * @if_flags: IFF_* values 1142 * @mask: bitmask of bits in if_flags to check 1143 * 1144 * Search for any interface with the given flags. Returns NULL if a device 1145 * is not found or a pointer to the device. Must be called inside 1146 * rtnl_lock(), and result refcount is unchanged. 1147 */ 1148 1149 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags, 1150 unsigned short mask) 1151 { 1152 struct net_device *dev, *ret; 1153 1154 ASSERT_RTNL(); 1155 1156 ret = NULL; 1157 for_each_netdev(net, dev) { 1158 if (((dev->flags ^ if_flags) & mask) == 0) { 1159 ret = dev; 1160 break; 1161 } 1162 } 1163 return ret; 1164 } 1165 EXPORT_SYMBOL(__dev_get_by_flags); 1166 1167 /** 1168 * dev_valid_name - check if name is okay for network device 1169 * @name: name string 1170 * 1171 * Network device names need to be valid file names to 1172 * allow sysfs to work. We also disallow any kind of 1173 * whitespace. 1174 */ 1175 bool dev_valid_name(const char *name) 1176 { 1177 if (*name == '\0') 1178 return false; 1179 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ) 1180 return false; 1181 if (!strcmp(name, ".") || !strcmp(name, "..")) 1182 return false; 1183 1184 while (*name) { 1185 if (*name == '/' || *name == ':' || isspace(*name)) 1186 return false; 1187 name++; 1188 } 1189 return true; 1190 } 1191 EXPORT_SYMBOL(dev_valid_name); 1192 1193 /** 1194 * __dev_alloc_name - allocate a name for a device 1195 * @net: network namespace to allocate the device name in 1196 * @name: name format string 1197 * @buf: scratch buffer and result name string 1198 * 1199 * Passed a format string - eg "lt%d" it will try and find a suitable 1200 * id. It scans list of devices to build up a free map, then chooses 1201 * the first empty slot. The caller must hold the dev_base or rtnl lock 1202 * while allocating the name and adding the device in order to avoid 1203 * duplicates. 1204 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1205 * Returns the number of the unit assigned or a negative errno code. 1206 */ 1207 1208 static int __dev_alloc_name(struct net *net, const char *name, char *buf) 1209 { 1210 int i = 0; 1211 const char *p; 1212 const int max_netdevices = 8*PAGE_SIZE; 1213 unsigned long *inuse; 1214 struct net_device *d; 1215 1216 if (!dev_valid_name(name)) 1217 return -EINVAL; 1218 1219 p = strchr(name, '%'); 1220 if (p) { 1221 /* 1222 * Verify the string as this thing may have come from 1223 * the user. There must be either one "%d" and no other "%" 1224 * characters. 1225 */ 1226 if (p[1] != 'd' || strchr(p + 2, '%')) 1227 return -EINVAL; 1228 1229 /* Use one page as a bit array of possible slots */ 1230 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); 1231 if (!inuse) 1232 return -ENOMEM; 1233 1234 for_each_netdev(net, d) { 1235 struct netdev_name_node *name_node; 1236 list_for_each_entry(name_node, &d->name_node->list, list) { 1237 if (!sscanf(name_node->name, name, &i)) 1238 continue; 1239 if (i < 0 || i >= max_netdevices) 1240 continue; 1241 1242 /* avoid cases where sscanf is not exact inverse of printf */ 1243 snprintf(buf, IFNAMSIZ, name, i); 1244 if (!strncmp(buf, name_node->name, IFNAMSIZ)) 1245 set_bit(i, inuse); 1246 } 1247 if (!sscanf(d->name, name, &i)) 1248 continue; 1249 if (i < 0 || i >= max_netdevices) 1250 continue; 1251 1252 /* avoid cases where sscanf is not exact inverse of printf */ 1253 snprintf(buf, IFNAMSIZ, name, i); 1254 if (!strncmp(buf, d->name, IFNAMSIZ)) 1255 set_bit(i, inuse); 1256 } 1257 1258 i = find_first_zero_bit(inuse, max_netdevices); 1259 free_page((unsigned long) inuse); 1260 } 1261 1262 snprintf(buf, IFNAMSIZ, name, i); 1263 if (!__dev_get_by_name(net, buf)) 1264 return i; 1265 1266 /* It is possible to run out of possible slots 1267 * when the name is long and there isn't enough space left 1268 * for the digits, or if all bits are used. 1269 */ 1270 return -ENFILE; 1271 } 1272 1273 static int dev_alloc_name_ns(struct net *net, 1274 struct net_device *dev, 1275 const char *name) 1276 { 1277 char buf[IFNAMSIZ]; 1278 int ret; 1279 1280 BUG_ON(!net); 1281 ret = __dev_alloc_name(net, name, buf); 1282 if (ret >= 0) 1283 strlcpy(dev->name, buf, IFNAMSIZ); 1284 return ret; 1285 } 1286 1287 /** 1288 * dev_alloc_name - allocate a name for a device 1289 * @dev: device 1290 * @name: name format string 1291 * 1292 * Passed a format string - eg "lt%d" it will try and find a suitable 1293 * id. It scans list of devices to build up a free map, then chooses 1294 * the first empty slot. The caller must hold the dev_base or rtnl lock 1295 * while allocating the name and adding the device in order to avoid 1296 * duplicates. 1297 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1298 * Returns the number of the unit assigned or a negative errno code. 1299 */ 1300 1301 int dev_alloc_name(struct net_device *dev, const char *name) 1302 { 1303 return dev_alloc_name_ns(dev_net(dev), dev, name); 1304 } 1305 EXPORT_SYMBOL(dev_alloc_name); 1306 1307 static int dev_get_valid_name(struct net *net, struct net_device *dev, 1308 const char *name) 1309 { 1310 BUG_ON(!net); 1311 1312 if (!dev_valid_name(name)) 1313 return -EINVAL; 1314 1315 if (strchr(name, '%')) 1316 return dev_alloc_name_ns(net, dev, name); 1317 else if (__dev_get_by_name(net, name)) 1318 return -EEXIST; 1319 else if (dev->name != name) 1320 strlcpy(dev->name, name, IFNAMSIZ); 1321 1322 return 0; 1323 } 1324 1325 /** 1326 * dev_change_name - change name of a device 1327 * @dev: device 1328 * @newname: name (or format string) must be at least IFNAMSIZ 1329 * 1330 * Change name of a device, can pass format strings "eth%d". 1331 * for wildcarding. 1332 */ 1333 int dev_change_name(struct net_device *dev, const char *newname) 1334 { 1335 unsigned char old_assign_type; 1336 char oldname[IFNAMSIZ]; 1337 int err = 0; 1338 int ret; 1339 struct net *net; 1340 1341 ASSERT_RTNL(); 1342 BUG_ON(!dev_net(dev)); 1343 1344 net = dev_net(dev); 1345 1346 /* Some auto-enslaved devices e.g. failover slaves are 1347 * special, as userspace might rename the device after 1348 * the interface had been brought up and running since 1349 * the point kernel initiated auto-enslavement. Allow 1350 * live name change even when these slave devices are 1351 * up and running. 1352 * 1353 * Typically, users of these auto-enslaving devices 1354 * don't actually care about slave name change, as 1355 * they are supposed to operate on master interface 1356 * directly. 1357 */ 1358 if (dev->flags & IFF_UP && 1359 likely(!(dev->priv_flags & IFF_LIVE_RENAME_OK))) 1360 return -EBUSY; 1361 1362 down_write(&devnet_rename_sem); 1363 1364 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { 1365 up_write(&devnet_rename_sem); 1366 return 0; 1367 } 1368 1369 memcpy(oldname, dev->name, IFNAMSIZ); 1370 1371 err = dev_get_valid_name(net, dev, newname); 1372 if (err < 0) { 1373 up_write(&devnet_rename_sem); 1374 return err; 1375 } 1376 1377 if (oldname[0] && !strchr(oldname, '%')) 1378 netdev_info(dev, "renamed from %s\n", oldname); 1379 1380 old_assign_type = dev->name_assign_type; 1381 dev->name_assign_type = NET_NAME_RENAMED; 1382 1383 rollback: 1384 ret = device_rename(&dev->dev, dev->name); 1385 if (ret) { 1386 memcpy(dev->name, oldname, IFNAMSIZ); 1387 dev->name_assign_type = old_assign_type; 1388 up_write(&devnet_rename_sem); 1389 return ret; 1390 } 1391 1392 up_write(&devnet_rename_sem); 1393 1394 netdev_adjacent_rename_links(dev, oldname); 1395 1396 write_lock_bh(&dev_base_lock); 1397 netdev_name_node_del(dev->name_node); 1398 write_unlock_bh(&dev_base_lock); 1399 1400 synchronize_rcu(); 1401 1402 write_lock_bh(&dev_base_lock); 1403 netdev_name_node_add(net, dev->name_node); 1404 write_unlock_bh(&dev_base_lock); 1405 1406 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); 1407 ret = notifier_to_errno(ret); 1408 1409 if (ret) { 1410 /* err >= 0 after dev_alloc_name() or stores the first errno */ 1411 if (err >= 0) { 1412 err = ret; 1413 down_write(&devnet_rename_sem); 1414 memcpy(dev->name, oldname, IFNAMSIZ); 1415 memcpy(oldname, newname, IFNAMSIZ); 1416 dev->name_assign_type = old_assign_type; 1417 old_assign_type = NET_NAME_RENAMED; 1418 goto rollback; 1419 } else { 1420 pr_err("%s: name change rollback failed: %d\n", 1421 dev->name, ret); 1422 } 1423 } 1424 1425 return err; 1426 } 1427 1428 /** 1429 * dev_set_alias - change ifalias of a device 1430 * @dev: device 1431 * @alias: name up to IFALIASZ 1432 * @len: limit of bytes to copy from info 1433 * 1434 * Set ifalias for a device, 1435 */ 1436 int dev_set_alias(struct net_device *dev, const char *alias, size_t len) 1437 { 1438 struct dev_ifalias *new_alias = NULL; 1439 1440 if (len >= IFALIASZ) 1441 return -EINVAL; 1442 1443 if (len) { 1444 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL); 1445 if (!new_alias) 1446 return -ENOMEM; 1447 1448 memcpy(new_alias->ifalias, alias, len); 1449 new_alias->ifalias[len] = 0; 1450 } 1451 1452 mutex_lock(&ifalias_mutex); 1453 new_alias = rcu_replace_pointer(dev->ifalias, new_alias, 1454 mutex_is_locked(&ifalias_mutex)); 1455 mutex_unlock(&ifalias_mutex); 1456 1457 if (new_alias) 1458 kfree_rcu(new_alias, rcuhead); 1459 1460 return len; 1461 } 1462 EXPORT_SYMBOL(dev_set_alias); 1463 1464 /** 1465 * dev_get_alias - get ifalias of a device 1466 * @dev: device 1467 * @name: buffer to store name of ifalias 1468 * @len: size of buffer 1469 * 1470 * get ifalias for a device. Caller must make sure dev cannot go 1471 * away, e.g. rcu read lock or own a reference count to device. 1472 */ 1473 int dev_get_alias(const struct net_device *dev, char *name, size_t len) 1474 { 1475 const struct dev_ifalias *alias; 1476 int ret = 0; 1477 1478 rcu_read_lock(); 1479 alias = rcu_dereference(dev->ifalias); 1480 if (alias) 1481 ret = snprintf(name, len, "%s", alias->ifalias); 1482 rcu_read_unlock(); 1483 1484 return ret; 1485 } 1486 1487 /** 1488 * netdev_features_change - device changes features 1489 * @dev: device to cause notification 1490 * 1491 * Called to indicate a device has changed features. 1492 */ 1493 void netdev_features_change(struct net_device *dev) 1494 { 1495 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); 1496 } 1497 EXPORT_SYMBOL(netdev_features_change); 1498 1499 /** 1500 * netdev_state_change - device changes state 1501 * @dev: device to cause notification 1502 * 1503 * Called to indicate a device has changed state. This function calls 1504 * the notifier chains for netdev_chain and sends a NEWLINK message 1505 * to the routing socket. 1506 */ 1507 void netdev_state_change(struct net_device *dev) 1508 { 1509 if (dev->flags & IFF_UP) { 1510 struct netdev_notifier_change_info change_info = { 1511 .info.dev = dev, 1512 }; 1513 1514 call_netdevice_notifiers_info(NETDEV_CHANGE, 1515 &change_info.info); 1516 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL); 1517 } 1518 } 1519 EXPORT_SYMBOL(netdev_state_change); 1520 1521 /** 1522 * __netdev_notify_peers - notify network peers about existence of @dev, 1523 * to be called when rtnl lock is already held. 1524 * @dev: network device 1525 * 1526 * Generate traffic such that interested network peers are aware of 1527 * @dev, such as by generating a gratuitous ARP. This may be used when 1528 * a device wants to inform the rest of the network about some sort of 1529 * reconfiguration such as a failover event or virtual machine 1530 * migration. 1531 */ 1532 void __netdev_notify_peers(struct net_device *dev) 1533 { 1534 ASSERT_RTNL(); 1535 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); 1536 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev); 1537 } 1538 EXPORT_SYMBOL(__netdev_notify_peers); 1539 1540 /** 1541 * netdev_notify_peers - notify network peers about existence of @dev 1542 * @dev: network device 1543 * 1544 * Generate traffic such that interested network peers are aware of 1545 * @dev, such as by generating a gratuitous ARP. This may be used when 1546 * a device wants to inform the rest of the network about some sort of 1547 * reconfiguration such as a failover event or virtual machine 1548 * migration. 1549 */ 1550 void netdev_notify_peers(struct net_device *dev) 1551 { 1552 rtnl_lock(); 1553 __netdev_notify_peers(dev); 1554 rtnl_unlock(); 1555 } 1556 EXPORT_SYMBOL(netdev_notify_peers); 1557 1558 static int napi_threaded_poll(void *data); 1559 1560 static int napi_kthread_create(struct napi_struct *n) 1561 { 1562 int err = 0; 1563 1564 /* Create and wake up the kthread once to put it in 1565 * TASK_INTERRUPTIBLE mode to avoid the blocked task 1566 * warning and work with loadavg. 1567 */ 1568 n->thread = kthread_run(napi_threaded_poll, n, "napi/%s-%d", 1569 n->dev->name, n->napi_id); 1570 if (IS_ERR(n->thread)) { 1571 err = PTR_ERR(n->thread); 1572 pr_err("kthread_run failed with err %d\n", err); 1573 n->thread = NULL; 1574 } 1575 1576 return err; 1577 } 1578 1579 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1580 { 1581 const struct net_device_ops *ops = dev->netdev_ops; 1582 int ret; 1583 1584 ASSERT_RTNL(); 1585 1586 if (!netif_device_present(dev)) { 1587 /* may be detached because parent is runtime-suspended */ 1588 if (dev->dev.parent) 1589 pm_runtime_resume(dev->dev.parent); 1590 if (!netif_device_present(dev)) 1591 return -ENODEV; 1592 } 1593 1594 /* Block netpoll from trying to do any rx path servicing. 1595 * If we don't do this there is a chance ndo_poll_controller 1596 * or ndo_poll may be running while we open the device 1597 */ 1598 netpoll_poll_disable(dev); 1599 1600 ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack); 1601 ret = notifier_to_errno(ret); 1602 if (ret) 1603 return ret; 1604 1605 set_bit(__LINK_STATE_START, &dev->state); 1606 1607 if (ops->ndo_validate_addr) 1608 ret = ops->ndo_validate_addr(dev); 1609 1610 if (!ret && ops->ndo_open) 1611 ret = ops->ndo_open(dev); 1612 1613 netpoll_poll_enable(dev); 1614 1615 if (ret) 1616 clear_bit(__LINK_STATE_START, &dev->state); 1617 else { 1618 dev->flags |= IFF_UP; 1619 dev_set_rx_mode(dev); 1620 dev_activate(dev); 1621 add_device_randomness(dev->dev_addr, dev->addr_len); 1622 } 1623 1624 return ret; 1625 } 1626 1627 /** 1628 * dev_open - prepare an interface for use. 1629 * @dev: device to open 1630 * @extack: netlink extended ack 1631 * 1632 * Takes a device from down to up state. The device's private open 1633 * function is invoked and then the multicast lists are loaded. Finally 1634 * the device is moved into the up state and a %NETDEV_UP message is 1635 * sent to the netdev notifier chain. 1636 * 1637 * Calling this function on an active interface is a nop. On a failure 1638 * a negative errno code is returned. 1639 */ 1640 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1641 { 1642 int ret; 1643 1644 if (dev->flags & IFF_UP) 1645 return 0; 1646 1647 ret = __dev_open(dev, extack); 1648 if (ret < 0) 1649 return ret; 1650 1651 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); 1652 call_netdevice_notifiers(NETDEV_UP, dev); 1653 1654 return ret; 1655 } 1656 EXPORT_SYMBOL(dev_open); 1657 1658 static void __dev_close_many(struct list_head *head) 1659 { 1660 struct net_device *dev; 1661 1662 ASSERT_RTNL(); 1663 might_sleep(); 1664 1665 list_for_each_entry(dev, head, close_list) { 1666 /* Temporarily disable netpoll until the interface is down */ 1667 netpoll_poll_disable(dev); 1668 1669 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); 1670 1671 clear_bit(__LINK_STATE_START, &dev->state); 1672 1673 /* Synchronize to scheduled poll. We cannot touch poll list, it 1674 * can be even on different cpu. So just clear netif_running(). 1675 * 1676 * dev->stop() will invoke napi_disable() on all of it's 1677 * napi_struct instances on this device. 1678 */ 1679 smp_mb__after_atomic(); /* Commit netif_running(). */ 1680 } 1681 1682 dev_deactivate_many(head); 1683 1684 list_for_each_entry(dev, head, close_list) { 1685 const struct net_device_ops *ops = dev->netdev_ops; 1686 1687 /* 1688 * Call the device specific close. This cannot fail. 1689 * Only if device is UP 1690 * 1691 * We allow it to be called even after a DETACH hot-plug 1692 * event. 1693 */ 1694 if (ops->ndo_stop) 1695 ops->ndo_stop(dev); 1696 1697 dev->flags &= ~IFF_UP; 1698 netpoll_poll_enable(dev); 1699 } 1700 } 1701 1702 static void __dev_close(struct net_device *dev) 1703 { 1704 LIST_HEAD(single); 1705 1706 list_add(&dev->close_list, &single); 1707 __dev_close_many(&single); 1708 list_del(&single); 1709 } 1710 1711 void dev_close_many(struct list_head *head, bool unlink) 1712 { 1713 struct net_device *dev, *tmp; 1714 1715 /* Remove the devices that don't need to be closed */ 1716 list_for_each_entry_safe(dev, tmp, head, close_list) 1717 if (!(dev->flags & IFF_UP)) 1718 list_del_init(&dev->close_list); 1719 1720 __dev_close_many(head); 1721 1722 list_for_each_entry_safe(dev, tmp, head, close_list) { 1723 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); 1724 call_netdevice_notifiers(NETDEV_DOWN, dev); 1725 if (unlink) 1726 list_del_init(&dev->close_list); 1727 } 1728 } 1729 EXPORT_SYMBOL(dev_close_many); 1730 1731 /** 1732 * dev_close - shutdown an interface. 1733 * @dev: device to shutdown 1734 * 1735 * This function moves an active device into down state. A 1736 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device 1737 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier 1738 * chain. 1739 */ 1740 void dev_close(struct net_device *dev) 1741 { 1742 if (dev->flags & IFF_UP) { 1743 LIST_HEAD(single); 1744 1745 list_add(&dev->close_list, &single); 1746 dev_close_many(&single, true); 1747 list_del(&single); 1748 } 1749 } 1750 EXPORT_SYMBOL(dev_close); 1751 1752 1753 /** 1754 * dev_disable_lro - disable Large Receive Offload on a device 1755 * @dev: device 1756 * 1757 * Disable Large Receive Offload (LRO) on a net device. Must be 1758 * called under RTNL. This is needed if received packets may be 1759 * forwarded to another interface. 1760 */ 1761 void dev_disable_lro(struct net_device *dev) 1762 { 1763 struct net_device *lower_dev; 1764 struct list_head *iter; 1765 1766 dev->wanted_features &= ~NETIF_F_LRO; 1767 netdev_update_features(dev); 1768 1769 if (unlikely(dev->features & NETIF_F_LRO)) 1770 netdev_WARN(dev, "failed to disable LRO!\n"); 1771 1772 netdev_for_each_lower_dev(dev, lower_dev, iter) 1773 dev_disable_lro(lower_dev); 1774 } 1775 EXPORT_SYMBOL(dev_disable_lro); 1776 1777 /** 1778 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device 1779 * @dev: device 1780 * 1781 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be 1782 * called under RTNL. This is needed if Generic XDP is installed on 1783 * the device. 1784 */ 1785 static void dev_disable_gro_hw(struct net_device *dev) 1786 { 1787 dev->wanted_features &= ~NETIF_F_GRO_HW; 1788 netdev_update_features(dev); 1789 1790 if (unlikely(dev->features & NETIF_F_GRO_HW)) 1791 netdev_WARN(dev, "failed to disable GRO_HW!\n"); 1792 } 1793 1794 const char *netdev_cmd_to_name(enum netdev_cmd cmd) 1795 { 1796 #define N(val) \ 1797 case NETDEV_##val: \ 1798 return "NETDEV_" __stringify(val); 1799 switch (cmd) { 1800 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER) 1801 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE) 1802 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE) 1803 N(POST_INIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) N(CHANGEUPPER) 1804 N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) N(BONDING_INFO) 1805 N(PRECHANGEUPPER) N(CHANGELOWERSTATE) N(UDP_TUNNEL_PUSH_INFO) 1806 N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN) 1807 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO) 1808 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO) 1809 N(PRE_CHANGEADDR) 1810 } 1811 #undef N 1812 return "UNKNOWN_NETDEV_EVENT"; 1813 } 1814 EXPORT_SYMBOL_GPL(netdev_cmd_to_name); 1815 1816 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, 1817 struct net_device *dev) 1818 { 1819 struct netdev_notifier_info info = { 1820 .dev = dev, 1821 }; 1822 1823 return nb->notifier_call(nb, val, &info); 1824 } 1825 1826 static int call_netdevice_register_notifiers(struct notifier_block *nb, 1827 struct net_device *dev) 1828 { 1829 int err; 1830 1831 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); 1832 err = notifier_to_errno(err); 1833 if (err) 1834 return err; 1835 1836 if (!(dev->flags & IFF_UP)) 1837 return 0; 1838 1839 call_netdevice_notifier(nb, NETDEV_UP, dev); 1840 return 0; 1841 } 1842 1843 static void call_netdevice_unregister_notifiers(struct notifier_block *nb, 1844 struct net_device *dev) 1845 { 1846 if (dev->flags & IFF_UP) { 1847 call_netdevice_notifier(nb, NETDEV_GOING_DOWN, 1848 dev); 1849 call_netdevice_notifier(nb, NETDEV_DOWN, dev); 1850 } 1851 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); 1852 } 1853 1854 static int call_netdevice_register_net_notifiers(struct notifier_block *nb, 1855 struct net *net) 1856 { 1857 struct net_device *dev; 1858 int err; 1859 1860 for_each_netdev(net, dev) { 1861 err = call_netdevice_register_notifiers(nb, dev); 1862 if (err) 1863 goto rollback; 1864 } 1865 return 0; 1866 1867 rollback: 1868 for_each_netdev_continue_reverse(net, dev) 1869 call_netdevice_unregister_notifiers(nb, dev); 1870 return err; 1871 } 1872 1873 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb, 1874 struct net *net) 1875 { 1876 struct net_device *dev; 1877 1878 for_each_netdev(net, dev) 1879 call_netdevice_unregister_notifiers(nb, dev); 1880 } 1881 1882 static int dev_boot_phase = 1; 1883 1884 /** 1885 * register_netdevice_notifier - register a network notifier block 1886 * @nb: notifier 1887 * 1888 * Register a notifier to be called when network device events occur. 1889 * The notifier passed is linked into the kernel structures and must 1890 * not be reused until it has been unregistered. A negative errno code 1891 * is returned on a failure. 1892 * 1893 * When registered all registration and up events are replayed 1894 * to the new notifier to allow device to have a race free 1895 * view of the network device list. 1896 */ 1897 1898 int register_netdevice_notifier(struct notifier_block *nb) 1899 { 1900 struct net *net; 1901 int err; 1902 1903 /* Close race with setup_net() and cleanup_net() */ 1904 down_write(&pernet_ops_rwsem); 1905 rtnl_lock(); 1906 err = raw_notifier_chain_register(&netdev_chain, nb); 1907 if (err) 1908 goto unlock; 1909 if (dev_boot_phase) 1910 goto unlock; 1911 for_each_net(net) { 1912 err = call_netdevice_register_net_notifiers(nb, net); 1913 if (err) 1914 goto rollback; 1915 } 1916 1917 unlock: 1918 rtnl_unlock(); 1919 up_write(&pernet_ops_rwsem); 1920 return err; 1921 1922 rollback: 1923 for_each_net_continue_reverse(net) 1924 call_netdevice_unregister_net_notifiers(nb, net); 1925 1926 raw_notifier_chain_unregister(&netdev_chain, nb); 1927 goto unlock; 1928 } 1929 EXPORT_SYMBOL(register_netdevice_notifier); 1930 1931 /** 1932 * unregister_netdevice_notifier - unregister a network notifier block 1933 * @nb: notifier 1934 * 1935 * Unregister a notifier previously registered by 1936 * register_netdevice_notifier(). The notifier is unlinked into the 1937 * kernel structures and may then be reused. A negative errno code 1938 * is returned on a failure. 1939 * 1940 * After unregistering unregister and down device events are synthesized 1941 * for all devices on the device list to the removed notifier to remove 1942 * the need for special case cleanup code. 1943 */ 1944 1945 int unregister_netdevice_notifier(struct notifier_block *nb) 1946 { 1947 struct net *net; 1948 int err; 1949 1950 /* Close race with setup_net() and cleanup_net() */ 1951 down_write(&pernet_ops_rwsem); 1952 rtnl_lock(); 1953 err = raw_notifier_chain_unregister(&netdev_chain, nb); 1954 if (err) 1955 goto unlock; 1956 1957 for_each_net(net) 1958 call_netdevice_unregister_net_notifiers(nb, net); 1959 1960 unlock: 1961 rtnl_unlock(); 1962 up_write(&pernet_ops_rwsem); 1963 return err; 1964 } 1965 EXPORT_SYMBOL(unregister_netdevice_notifier); 1966 1967 static int __register_netdevice_notifier_net(struct net *net, 1968 struct notifier_block *nb, 1969 bool ignore_call_fail) 1970 { 1971 int err; 1972 1973 err = raw_notifier_chain_register(&net->netdev_chain, nb); 1974 if (err) 1975 return err; 1976 if (dev_boot_phase) 1977 return 0; 1978 1979 err = call_netdevice_register_net_notifiers(nb, net); 1980 if (err && !ignore_call_fail) 1981 goto chain_unregister; 1982 1983 return 0; 1984 1985 chain_unregister: 1986 raw_notifier_chain_unregister(&net->netdev_chain, nb); 1987 return err; 1988 } 1989 1990 static int __unregister_netdevice_notifier_net(struct net *net, 1991 struct notifier_block *nb) 1992 { 1993 int err; 1994 1995 err = raw_notifier_chain_unregister(&net->netdev_chain, nb); 1996 if (err) 1997 return err; 1998 1999 call_netdevice_unregister_net_notifiers(nb, net); 2000 return 0; 2001 } 2002 2003 /** 2004 * register_netdevice_notifier_net - register a per-netns network notifier block 2005 * @net: network namespace 2006 * @nb: notifier 2007 * 2008 * Register a notifier to be called when network device events occur. 2009 * The notifier passed is linked into the kernel structures and must 2010 * not be reused until it has been unregistered. A negative errno code 2011 * is returned on a failure. 2012 * 2013 * When registered all registration and up events are replayed 2014 * to the new notifier to allow device to have a race free 2015 * view of the network device list. 2016 */ 2017 2018 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb) 2019 { 2020 int err; 2021 2022 rtnl_lock(); 2023 err = __register_netdevice_notifier_net(net, nb, false); 2024 rtnl_unlock(); 2025 return err; 2026 } 2027 EXPORT_SYMBOL(register_netdevice_notifier_net); 2028 2029 /** 2030 * unregister_netdevice_notifier_net - unregister a per-netns 2031 * network notifier block 2032 * @net: network namespace 2033 * @nb: notifier 2034 * 2035 * Unregister a notifier previously registered by 2036 * register_netdevice_notifier(). The notifier is unlinked into the 2037 * kernel structures and may then be reused. A negative errno code 2038 * is returned on a failure. 2039 * 2040 * After unregistering unregister and down device events are synthesized 2041 * for all devices on the device list to the removed notifier to remove 2042 * the need for special case cleanup code. 2043 */ 2044 2045 int unregister_netdevice_notifier_net(struct net *net, 2046 struct notifier_block *nb) 2047 { 2048 int err; 2049 2050 rtnl_lock(); 2051 err = __unregister_netdevice_notifier_net(net, nb); 2052 rtnl_unlock(); 2053 return err; 2054 } 2055 EXPORT_SYMBOL(unregister_netdevice_notifier_net); 2056 2057 int register_netdevice_notifier_dev_net(struct net_device *dev, 2058 struct notifier_block *nb, 2059 struct netdev_net_notifier *nn) 2060 { 2061 int err; 2062 2063 rtnl_lock(); 2064 err = __register_netdevice_notifier_net(dev_net(dev), nb, false); 2065 if (!err) { 2066 nn->nb = nb; 2067 list_add(&nn->list, &dev->net_notifier_list); 2068 } 2069 rtnl_unlock(); 2070 return err; 2071 } 2072 EXPORT_SYMBOL(register_netdevice_notifier_dev_net); 2073 2074 int unregister_netdevice_notifier_dev_net(struct net_device *dev, 2075 struct notifier_block *nb, 2076 struct netdev_net_notifier *nn) 2077 { 2078 int err; 2079 2080 rtnl_lock(); 2081 list_del(&nn->list); 2082 err = __unregister_netdevice_notifier_net(dev_net(dev), nb); 2083 rtnl_unlock(); 2084 return err; 2085 } 2086 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net); 2087 2088 static void move_netdevice_notifiers_dev_net(struct net_device *dev, 2089 struct net *net) 2090 { 2091 struct netdev_net_notifier *nn; 2092 2093 list_for_each_entry(nn, &dev->net_notifier_list, list) { 2094 __unregister_netdevice_notifier_net(dev_net(dev), nn->nb); 2095 __register_netdevice_notifier_net(net, nn->nb, true); 2096 } 2097 } 2098 2099 /** 2100 * call_netdevice_notifiers_info - call all network notifier blocks 2101 * @val: value passed unmodified to notifier function 2102 * @info: notifier information data 2103 * 2104 * Call all network notifier blocks. Parameters and return value 2105 * are as for raw_notifier_call_chain(). 2106 */ 2107 2108 static int call_netdevice_notifiers_info(unsigned long val, 2109 struct netdev_notifier_info *info) 2110 { 2111 struct net *net = dev_net(info->dev); 2112 int ret; 2113 2114 ASSERT_RTNL(); 2115 2116 /* Run per-netns notifier block chain first, then run the global one. 2117 * Hopefully, one day, the global one is going to be removed after 2118 * all notifier block registrators get converted to be per-netns. 2119 */ 2120 ret = raw_notifier_call_chain(&net->netdev_chain, val, info); 2121 if (ret & NOTIFY_STOP_MASK) 2122 return ret; 2123 return raw_notifier_call_chain(&netdev_chain, val, info); 2124 } 2125 2126 static int call_netdevice_notifiers_extack(unsigned long val, 2127 struct net_device *dev, 2128 struct netlink_ext_ack *extack) 2129 { 2130 struct netdev_notifier_info info = { 2131 .dev = dev, 2132 .extack = extack, 2133 }; 2134 2135 return call_netdevice_notifiers_info(val, &info); 2136 } 2137 2138 /** 2139 * call_netdevice_notifiers - call all network notifier blocks 2140 * @val: value passed unmodified to notifier function 2141 * @dev: net_device pointer passed unmodified to notifier function 2142 * 2143 * Call all network notifier blocks. Parameters and return value 2144 * are as for raw_notifier_call_chain(). 2145 */ 2146 2147 int call_netdevice_notifiers(unsigned long val, struct net_device *dev) 2148 { 2149 return call_netdevice_notifiers_extack(val, dev, NULL); 2150 } 2151 EXPORT_SYMBOL(call_netdevice_notifiers); 2152 2153 /** 2154 * call_netdevice_notifiers_mtu - call all network notifier blocks 2155 * @val: value passed unmodified to notifier function 2156 * @dev: net_device pointer passed unmodified to notifier function 2157 * @arg: additional u32 argument passed to the notifier function 2158 * 2159 * Call all network notifier blocks. Parameters and return value 2160 * are as for raw_notifier_call_chain(). 2161 */ 2162 static int call_netdevice_notifiers_mtu(unsigned long val, 2163 struct net_device *dev, u32 arg) 2164 { 2165 struct netdev_notifier_info_ext info = { 2166 .info.dev = dev, 2167 .ext.mtu = arg, 2168 }; 2169 2170 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0); 2171 2172 return call_netdevice_notifiers_info(val, &info.info); 2173 } 2174 2175 #ifdef CONFIG_NET_INGRESS 2176 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key); 2177 2178 void net_inc_ingress_queue(void) 2179 { 2180 static_branch_inc(&ingress_needed_key); 2181 } 2182 EXPORT_SYMBOL_GPL(net_inc_ingress_queue); 2183 2184 void net_dec_ingress_queue(void) 2185 { 2186 static_branch_dec(&ingress_needed_key); 2187 } 2188 EXPORT_SYMBOL_GPL(net_dec_ingress_queue); 2189 #endif 2190 2191 #ifdef CONFIG_NET_EGRESS 2192 static DEFINE_STATIC_KEY_FALSE(egress_needed_key); 2193 2194 void net_inc_egress_queue(void) 2195 { 2196 static_branch_inc(&egress_needed_key); 2197 } 2198 EXPORT_SYMBOL_GPL(net_inc_egress_queue); 2199 2200 void net_dec_egress_queue(void) 2201 { 2202 static_branch_dec(&egress_needed_key); 2203 } 2204 EXPORT_SYMBOL_GPL(net_dec_egress_queue); 2205 #endif 2206 2207 static DEFINE_STATIC_KEY_FALSE(netstamp_needed_key); 2208 #ifdef CONFIG_JUMP_LABEL 2209 static atomic_t netstamp_needed_deferred; 2210 static atomic_t netstamp_wanted; 2211 static void netstamp_clear(struct work_struct *work) 2212 { 2213 int deferred = atomic_xchg(&netstamp_needed_deferred, 0); 2214 int wanted; 2215 2216 wanted = atomic_add_return(deferred, &netstamp_wanted); 2217 if (wanted > 0) 2218 static_branch_enable(&netstamp_needed_key); 2219 else 2220 static_branch_disable(&netstamp_needed_key); 2221 } 2222 static DECLARE_WORK(netstamp_work, netstamp_clear); 2223 #endif 2224 2225 void net_enable_timestamp(void) 2226 { 2227 #ifdef CONFIG_JUMP_LABEL 2228 int wanted; 2229 2230 while (1) { 2231 wanted = atomic_read(&netstamp_wanted); 2232 if (wanted <= 0) 2233 break; 2234 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted) 2235 return; 2236 } 2237 atomic_inc(&netstamp_needed_deferred); 2238 schedule_work(&netstamp_work); 2239 #else 2240 static_branch_inc(&netstamp_needed_key); 2241 #endif 2242 } 2243 EXPORT_SYMBOL(net_enable_timestamp); 2244 2245 void net_disable_timestamp(void) 2246 { 2247 #ifdef CONFIG_JUMP_LABEL 2248 int wanted; 2249 2250 while (1) { 2251 wanted = atomic_read(&netstamp_wanted); 2252 if (wanted <= 1) 2253 break; 2254 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted) 2255 return; 2256 } 2257 atomic_dec(&netstamp_needed_deferred); 2258 schedule_work(&netstamp_work); 2259 #else 2260 static_branch_dec(&netstamp_needed_key); 2261 #endif 2262 } 2263 EXPORT_SYMBOL(net_disable_timestamp); 2264 2265 static inline void net_timestamp_set(struct sk_buff *skb) 2266 { 2267 skb->tstamp = 0; 2268 if (static_branch_unlikely(&netstamp_needed_key)) 2269 __net_timestamp(skb); 2270 } 2271 2272 #define net_timestamp_check(COND, SKB) \ 2273 if (static_branch_unlikely(&netstamp_needed_key)) { \ 2274 if ((COND) && !(SKB)->tstamp) \ 2275 __net_timestamp(SKB); \ 2276 } \ 2277 2278 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb) 2279 { 2280 return __is_skb_forwardable(dev, skb, true); 2281 } 2282 EXPORT_SYMBOL_GPL(is_skb_forwardable); 2283 2284 static int __dev_forward_skb2(struct net_device *dev, struct sk_buff *skb, 2285 bool check_mtu) 2286 { 2287 int ret = ____dev_forward_skb(dev, skb, check_mtu); 2288 2289 if (likely(!ret)) { 2290 skb->protocol = eth_type_trans(skb, dev); 2291 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); 2292 } 2293 2294 return ret; 2295 } 2296 2297 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2298 { 2299 return __dev_forward_skb2(dev, skb, true); 2300 } 2301 EXPORT_SYMBOL_GPL(__dev_forward_skb); 2302 2303 /** 2304 * dev_forward_skb - loopback an skb to another netif 2305 * 2306 * @dev: destination network device 2307 * @skb: buffer to forward 2308 * 2309 * return values: 2310 * NET_RX_SUCCESS (no congestion) 2311 * NET_RX_DROP (packet was dropped, but freed) 2312 * 2313 * dev_forward_skb can be used for injecting an skb from the 2314 * start_xmit function of one device into the receive queue 2315 * of another device. 2316 * 2317 * The receiving device may be in another namespace, so 2318 * we have to clear all information in the skb that could 2319 * impact namespace isolation. 2320 */ 2321 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2322 { 2323 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); 2324 } 2325 EXPORT_SYMBOL_GPL(dev_forward_skb); 2326 2327 int dev_forward_skb_nomtu(struct net_device *dev, struct sk_buff *skb) 2328 { 2329 return __dev_forward_skb2(dev, skb, false) ?: netif_rx_internal(skb); 2330 } 2331 2332 static inline int deliver_skb(struct sk_buff *skb, 2333 struct packet_type *pt_prev, 2334 struct net_device *orig_dev) 2335 { 2336 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 2337 return -ENOMEM; 2338 refcount_inc(&skb->users); 2339 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 2340 } 2341 2342 static inline void deliver_ptype_list_skb(struct sk_buff *skb, 2343 struct packet_type **pt, 2344 struct net_device *orig_dev, 2345 __be16 type, 2346 struct list_head *ptype_list) 2347 { 2348 struct packet_type *ptype, *pt_prev = *pt; 2349 2350 list_for_each_entry_rcu(ptype, ptype_list, list) { 2351 if (ptype->type != type) 2352 continue; 2353 if (pt_prev) 2354 deliver_skb(skb, pt_prev, orig_dev); 2355 pt_prev = ptype; 2356 } 2357 *pt = pt_prev; 2358 } 2359 2360 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) 2361 { 2362 if (!ptype->af_packet_priv || !skb->sk) 2363 return false; 2364 2365 if (ptype->id_match) 2366 return ptype->id_match(ptype, skb->sk); 2367 else if ((struct sock *)ptype->af_packet_priv == skb->sk) 2368 return true; 2369 2370 return false; 2371 } 2372 2373 /** 2374 * dev_nit_active - return true if any network interface taps are in use 2375 * 2376 * @dev: network device to check for the presence of taps 2377 */ 2378 bool dev_nit_active(struct net_device *dev) 2379 { 2380 return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all); 2381 } 2382 EXPORT_SYMBOL_GPL(dev_nit_active); 2383 2384 /* 2385 * Support routine. Sends outgoing frames to any network 2386 * taps currently in use. 2387 */ 2388 2389 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) 2390 { 2391 struct packet_type *ptype; 2392 struct sk_buff *skb2 = NULL; 2393 struct packet_type *pt_prev = NULL; 2394 struct list_head *ptype_list = &ptype_all; 2395 2396 rcu_read_lock(); 2397 again: 2398 list_for_each_entry_rcu(ptype, ptype_list, list) { 2399 if (ptype->ignore_outgoing) 2400 continue; 2401 2402 /* Never send packets back to the socket 2403 * they originated from - MvS (miquels@drinkel.ow.org) 2404 */ 2405 if (skb_loop_sk(ptype, skb)) 2406 continue; 2407 2408 if (pt_prev) { 2409 deliver_skb(skb2, pt_prev, skb->dev); 2410 pt_prev = ptype; 2411 continue; 2412 } 2413 2414 /* need to clone skb, done only once */ 2415 skb2 = skb_clone(skb, GFP_ATOMIC); 2416 if (!skb2) 2417 goto out_unlock; 2418 2419 net_timestamp_set(skb2); 2420 2421 /* skb->nh should be correctly 2422 * set by sender, so that the second statement is 2423 * just protection against buggy protocols. 2424 */ 2425 skb_reset_mac_header(skb2); 2426 2427 if (skb_network_header(skb2) < skb2->data || 2428 skb_network_header(skb2) > skb_tail_pointer(skb2)) { 2429 net_crit_ratelimited("protocol %04x is buggy, dev %s\n", 2430 ntohs(skb2->protocol), 2431 dev->name); 2432 skb_reset_network_header(skb2); 2433 } 2434 2435 skb2->transport_header = skb2->network_header; 2436 skb2->pkt_type = PACKET_OUTGOING; 2437 pt_prev = ptype; 2438 } 2439 2440 if (ptype_list == &ptype_all) { 2441 ptype_list = &dev->ptype_all; 2442 goto again; 2443 } 2444 out_unlock: 2445 if (pt_prev) { 2446 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC)) 2447 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); 2448 else 2449 kfree_skb(skb2); 2450 } 2451 rcu_read_unlock(); 2452 } 2453 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit); 2454 2455 /** 2456 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change 2457 * @dev: Network device 2458 * @txq: number of queues available 2459 * 2460 * If real_num_tx_queues is changed the tc mappings may no longer be 2461 * valid. To resolve this verify the tc mapping remains valid and if 2462 * not NULL the mapping. With no priorities mapping to this 2463 * offset/count pair it will no longer be used. In the worst case TC0 2464 * is invalid nothing can be done so disable priority mappings. If is 2465 * expected that drivers will fix this mapping if they can before 2466 * calling netif_set_real_num_tx_queues. 2467 */ 2468 static void netif_setup_tc(struct net_device *dev, unsigned int txq) 2469 { 2470 int i; 2471 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2472 2473 /* If TC0 is invalidated disable TC mapping */ 2474 if (tc->offset + tc->count > txq) { 2475 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); 2476 dev->num_tc = 0; 2477 return; 2478 } 2479 2480 /* Invalidated prio to tc mappings set to TC0 */ 2481 for (i = 1; i < TC_BITMASK + 1; i++) { 2482 int q = netdev_get_prio_tc_map(dev, i); 2483 2484 tc = &dev->tc_to_txq[q]; 2485 if (tc->offset + tc->count > txq) { 2486 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n", 2487 i, q); 2488 netdev_set_prio_tc_map(dev, i, 0); 2489 } 2490 } 2491 } 2492 2493 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq) 2494 { 2495 if (dev->num_tc) { 2496 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2497 int i; 2498 2499 /* walk through the TCs and see if it falls into any of them */ 2500 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) { 2501 if ((txq - tc->offset) < tc->count) 2502 return i; 2503 } 2504 2505 /* didn't find it, just return -1 to indicate no match */ 2506 return -1; 2507 } 2508 2509 return 0; 2510 } 2511 EXPORT_SYMBOL(netdev_txq_to_tc); 2512 2513 #ifdef CONFIG_XPS 2514 static struct static_key xps_needed __read_mostly; 2515 static struct static_key xps_rxqs_needed __read_mostly; 2516 static DEFINE_MUTEX(xps_map_mutex); 2517 #define xmap_dereference(P) \ 2518 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) 2519 2520 static bool remove_xps_queue(struct xps_dev_maps *dev_maps, 2521 struct xps_dev_maps *old_maps, int tci, u16 index) 2522 { 2523 struct xps_map *map = NULL; 2524 int pos; 2525 2526 if (dev_maps) 2527 map = xmap_dereference(dev_maps->attr_map[tci]); 2528 if (!map) 2529 return false; 2530 2531 for (pos = map->len; pos--;) { 2532 if (map->queues[pos] != index) 2533 continue; 2534 2535 if (map->len > 1) { 2536 map->queues[pos] = map->queues[--map->len]; 2537 break; 2538 } 2539 2540 if (old_maps) 2541 RCU_INIT_POINTER(old_maps->attr_map[tci], NULL); 2542 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2543 kfree_rcu(map, rcu); 2544 return false; 2545 } 2546 2547 return true; 2548 } 2549 2550 static bool remove_xps_queue_cpu(struct net_device *dev, 2551 struct xps_dev_maps *dev_maps, 2552 int cpu, u16 offset, u16 count) 2553 { 2554 int num_tc = dev_maps->num_tc; 2555 bool active = false; 2556 int tci; 2557 2558 for (tci = cpu * num_tc; num_tc--; tci++) { 2559 int i, j; 2560 2561 for (i = count, j = offset; i--; j++) { 2562 if (!remove_xps_queue(dev_maps, NULL, tci, j)) 2563 break; 2564 } 2565 2566 active |= i < 0; 2567 } 2568 2569 return active; 2570 } 2571 2572 static void reset_xps_maps(struct net_device *dev, 2573 struct xps_dev_maps *dev_maps, 2574 enum xps_map_type type) 2575 { 2576 static_key_slow_dec_cpuslocked(&xps_needed); 2577 if (type == XPS_RXQS) 2578 static_key_slow_dec_cpuslocked(&xps_rxqs_needed); 2579 2580 RCU_INIT_POINTER(dev->xps_maps[type], NULL); 2581 2582 kfree_rcu(dev_maps, rcu); 2583 } 2584 2585 static void clean_xps_maps(struct net_device *dev, enum xps_map_type type, 2586 u16 offset, u16 count) 2587 { 2588 struct xps_dev_maps *dev_maps; 2589 bool active = false; 2590 int i, j; 2591 2592 dev_maps = xmap_dereference(dev->xps_maps[type]); 2593 if (!dev_maps) 2594 return; 2595 2596 for (j = 0; j < dev_maps->nr_ids; j++) 2597 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, count); 2598 if (!active) 2599 reset_xps_maps(dev, dev_maps, type); 2600 2601 if (type == XPS_CPUS) { 2602 for (i = offset + (count - 1); count--; i--) 2603 netdev_queue_numa_node_write( 2604 netdev_get_tx_queue(dev, i), NUMA_NO_NODE); 2605 } 2606 } 2607 2608 static void netif_reset_xps_queues(struct net_device *dev, u16 offset, 2609 u16 count) 2610 { 2611 if (!static_key_false(&xps_needed)) 2612 return; 2613 2614 cpus_read_lock(); 2615 mutex_lock(&xps_map_mutex); 2616 2617 if (static_key_false(&xps_rxqs_needed)) 2618 clean_xps_maps(dev, XPS_RXQS, offset, count); 2619 2620 clean_xps_maps(dev, XPS_CPUS, offset, count); 2621 2622 mutex_unlock(&xps_map_mutex); 2623 cpus_read_unlock(); 2624 } 2625 2626 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) 2627 { 2628 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index); 2629 } 2630 2631 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index, 2632 u16 index, bool is_rxqs_map) 2633 { 2634 struct xps_map *new_map; 2635 int alloc_len = XPS_MIN_MAP_ALLOC; 2636 int i, pos; 2637 2638 for (pos = 0; map && pos < map->len; pos++) { 2639 if (map->queues[pos] != index) 2640 continue; 2641 return map; 2642 } 2643 2644 /* Need to add tx-queue to this CPU's/rx-queue's existing map */ 2645 if (map) { 2646 if (pos < map->alloc_len) 2647 return map; 2648 2649 alloc_len = map->alloc_len * 2; 2650 } 2651 2652 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's 2653 * map 2654 */ 2655 if (is_rxqs_map) 2656 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL); 2657 else 2658 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, 2659 cpu_to_node(attr_index)); 2660 if (!new_map) 2661 return NULL; 2662 2663 for (i = 0; i < pos; i++) 2664 new_map->queues[i] = map->queues[i]; 2665 new_map->alloc_len = alloc_len; 2666 new_map->len = pos; 2667 2668 return new_map; 2669 } 2670 2671 /* Copy xps maps at a given index */ 2672 static void xps_copy_dev_maps(struct xps_dev_maps *dev_maps, 2673 struct xps_dev_maps *new_dev_maps, int index, 2674 int tc, bool skip_tc) 2675 { 2676 int i, tci = index * dev_maps->num_tc; 2677 struct xps_map *map; 2678 2679 /* copy maps belonging to foreign traffic classes */ 2680 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2681 if (i == tc && skip_tc) 2682 continue; 2683 2684 /* fill in the new device map from the old device map */ 2685 map = xmap_dereference(dev_maps->attr_map[tci]); 2686 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2687 } 2688 } 2689 2690 /* Must be called under cpus_read_lock */ 2691 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask, 2692 u16 index, enum xps_map_type type) 2693 { 2694 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL, *old_dev_maps = NULL; 2695 const unsigned long *online_mask = NULL; 2696 bool active = false, copy = false; 2697 int i, j, tci, numa_node_id = -2; 2698 int maps_sz, num_tc = 1, tc = 0; 2699 struct xps_map *map, *new_map; 2700 unsigned int nr_ids; 2701 2702 if (dev->num_tc) { 2703 /* Do not allow XPS on subordinate device directly */ 2704 num_tc = dev->num_tc; 2705 if (num_tc < 0) 2706 return -EINVAL; 2707 2708 /* If queue belongs to subordinate dev use its map */ 2709 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev; 2710 2711 tc = netdev_txq_to_tc(dev, index); 2712 if (tc < 0) 2713 return -EINVAL; 2714 } 2715 2716 mutex_lock(&xps_map_mutex); 2717 2718 dev_maps = xmap_dereference(dev->xps_maps[type]); 2719 if (type == XPS_RXQS) { 2720 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues); 2721 nr_ids = dev->num_rx_queues; 2722 } else { 2723 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc); 2724 if (num_possible_cpus() > 1) 2725 online_mask = cpumask_bits(cpu_online_mask); 2726 nr_ids = nr_cpu_ids; 2727 } 2728 2729 if (maps_sz < L1_CACHE_BYTES) 2730 maps_sz = L1_CACHE_BYTES; 2731 2732 /* The old dev_maps could be larger or smaller than the one we're 2733 * setting up now, as dev->num_tc or nr_ids could have been updated in 2734 * between. We could try to be smart, but let's be safe instead and only 2735 * copy foreign traffic classes if the two map sizes match. 2736 */ 2737 if (dev_maps && 2738 dev_maps->num_tc == num_tc && dev_maps->nr_ids == nr_ids) 2739 copy = true; 2740 2741 /* allocate memory for queue storage */ 2742 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids), 2743 j < nr_ids;) { 2744 if (!new_dev_maps) { 2745 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); 2746 if (!new_dev_maps) { 2747 mutex_unlock(&xps_map_mutex); 2748 return -ENOMEM; 2749 } 2750 2751 new_dev_maps->nr_ids = nr_ids; 2752 new_dev_maps->num_tc = num_tc; 2753 } 2754 2755 tci = j * num_tc + tc; 2756 map = copy ? xmap_dereference(dev_maps->attr_map[tci]) : NULL; 2757 2758 map = expand_xps_map(map, j, index, type == XPS_RXQS); 2759 if (!map) 2760 goto error; 2761 2762 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2763 } 2764 2765 if (!new_dev_maps) 2766 goto out_no_new_maps; 2767 2768 if (!dev_maps) { 2769 /* Increment static keys at most once per type */ 2770 static_key_slow_inc_cpuslocked(&xps_needed); 2771 if (type == XPS_RXQS) 2772 static_key_slow_inc_cpuslocked(&xps_rxqs_needed); 2773 } 2774 2775 for (j = 0; j < nr_ids; j++) { 2776 bool skip_tc = false; 2777 2778 tci = j * num_tc + tc; 2779 if (netif_attr_test_mask(j, mask, nr_ids) && 2780 netif_attr_test_online(j, online_mask, nr_ids)) { 2781 /* add tx-queue to CPU/rx-queue maps */ 2782 int pos = 0; 2783 2784 skip_tc = true; 2785 2786 map = xmap_dereference(new_dev_maps->attr_map[tci]); 2787 while ((pos < map->len) && (map->queues[pos] != index)) 2788 pos++; 2789 2790 if (pos == map->len) 2791 map->queues[map->len++] = index; 2792 #ifdef CONFIG_NUMA 2793 if (type == XPS_CPUS) { 2794 if (numa_node_id == -2) 2795 numa_node_id = cpu_to_node(j); 2796 else if (numa_node_id != cpu_to_node(j)) 2797 numa_node_id = -1; 2798 } 2799 #endif 2800 } 2801 2802 if (copy) 2803 xps_copy_dev_maps(dev_maps, new_dev_maps, j, tc, 2804 skip_tc); 2805 } 2806 2807 rcu_assign_pointer(dev->xps_maps[type], new_dev_maps); 2808 2809 /* Cleanup old maps */ 2810 if (!dev_maps) 2811 goto out_no_old_maps; 2812 2813 for (j = 0; j < dev_maps->nr_ids; j++) { 2814 for (i = num_tc, tci = j * dev_maps->num_tc; i--; tci++) { 2815 map = xmap_dereference(dev_maps->attr_map[tci]); 2816 if (!map) 2817 continue; 2818 2819 if (copy) { 2820 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2821 if (map == new_map) 2822 continue; 2823 } 2824 2825 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2826 kfree_rcu(map, rcu); 2827 } 2828 } 2829 2830 old_dev_maps = dev_maps; 2831 2832 out_no_old_maps: 2833 dev_maps = new_dev_maps; 2834 active = true; 2835 2836 out_no_new_maps: 2837 if (type == XPS_CPUS) 2838 /* update Tx queue numa node */ 2839 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), 2840 (numa_node_id >= 0) ? 2841 numa_node_id : NUMA_NO_NODE); 2842 2843 if (!dev_maps) 2844 goto out_no_maps; 2845 2846 /* removes tx-queue from unused CPUs/rx-queues */ 2847 for (j = 0; j < dev_maps->nr_ids; j++) { 2848 tci = j * dev_maps->num_tc; 2849 2850 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2851 if (i == tc && 2852 netif_attr_test_mask(j, mask, dev_maps->nr_ids) && 2853 netif_attr_test_online(j, online_mask, dev_maps->nr_ids)) 2854 continue; 2855 2856 active |= remove_xps_queue(dev_maps, 2857 copy ? old_dev_maps : NULL, 2858 tci, index); 2859 } 2860 } 2861 2862 if (old_dev_maps) 2863 kfree_rcu(old_dev_maps, rcu); 2864 2865 /* free map if not active */ 2866 if (!active) 2867 reset_xps_maps(dev, dev_maps, type); 2868 2869 out_no_maps: 2870 mutex_unlock(&xps_map_mutex); 2871 2872 return 0; 2873 error: 2874 /* remove any maps that we added */ 2875 for (j = 0; j < nr_ids; j++) { 2876 for (i = num_tc, tci = j * num_tc; i--; tci++) { 2877 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2878 map = copy ? 2879 xmap_dereference(dev_maps->attr_map[tci]) : 2880 NULL; 2881 if (new_map && new_map != map) 2882 kfree(new_map); 2883 } 2884 } 2885 2886 mutex_unlock(&xps_map_mutex); 2887 2888 kfree(new_dev_maps); 2889 return -ENOMEM; 2890 } 2891 EXPORT_SYMBOL_GPL(__netif_set_xps_queue); 2892 2893 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, 2894 u16 index) 2895 { 2896 int ret; 2897 2898 cpus_read_lock(); 2899 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, XPS_CPUS); 2900 cpus_read_unlock(); 2901 2902 return ret; 2903 } 2904 EXPORT_SYMBOL(netif_set_xps_queue); 2905 2906 #endif 2907 static void netdev_unbind_all_sb_channels(struct net_device *dev) 2908 { 2909 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2910 2911 /* Unbind any subordinate channels */ 2912 while (txq-- != &dev->_tx[0]) { 2913 if (txq->sb_dev) 2914 netdev_unbind_sb_channel(dev, txq->sb_dev); 2915 } 2916 } 2917 2918 void netdev_reset_tc(struct net_device *dev) 2919 { 2920 #ifdef CONFIG_XPS 2921 netif_reset_xps_queues_gt(dev, 0); 2922 #endif 2923 netdev_unbind_all_sb_channels(dev); 2924 2925 /* Reset TC configuration of device */ 2926 dev->num_tc = 0; 2927 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq)); 2928 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map)); 2929 } 2930 EXPORT_SYMBOL(netdev_reset_tc); 2931 2932 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset) 2933 { 2934 if (tc >= dev->num_tc) 2935 return -EINVAL; 2936 2937 #ifdef CONFIG_XPS 2938 netif_reset_xps_queues(dev, offset, count); 2939 #endif 2940 dev->tc_to_txq[tc].count = count; 2941 dev->tc_to_txq[tc].offset = offset; 2942 return 0; 2943 } 2944 EXPORT_SYMBOL(netdev_set_tc_queue); 2945 2946 int netdev_set_num_tc(struct net_device *dev, u8 num_tc) 2947 { 2948 if (num_tc > TC_MAX_QUEUE) 2949 return -EINVAL; 2950 2951 #ifdef CONFIG_XPS 2952 netif_reset_xps_queues_gt(dev, 0); 2953 #endif 2954 netdev_unbind_all_sb_channels(dev); 2955 2956 dev->num_tc = num_tc; 2957 return 0; 2958 } 2959 EXPORT_SYMBOL(netdev_set_num_tc); 2960 2961 void netdev_unbind_sb_channel(struct net_device *dev, 2962 struct net_device *sb_dev) 2963 { 2964 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2965 2966 #ifdef CONFIG_XPS 2967 netif_reset_xps_queues_gt(sb_dev, 0); 2968 #endif 2969 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq)); 2970 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map)); 2971 2972 while (txq-- != &dev->_tx[0]) { 2973 if (txq->sb_dev == sb_dev) 2974 txq->sb_dev = NULL; 2975 } 2976 } 2977 EXPORT_SYMBOL(netdev_unbind_sb_channel); 2978 2979 int netdev_bind_sb_channel_queue(struct net_device *dev, 2980 struct net_device *sb_dev, 2981 u8 tc, u16 count, u16 offset) 2982 { 2983 /* Make certain the sb_dev and dev are already configured */ 2984 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc) 2985 return -EINVAL; 2986 2987 /* We cannot hand out queues we don't have */ 2988 if ((offset + count) > dev->real_num_tx_queues) 2989 return -EINVAL; 2990 2991 /* Record the mapping */ 2992 sb_dev->tc_to_txq[tc].count = count; 2993 sb_dev->tc_to_txq[tc].offset = offset; 2994 2995 /* Provide a way for Tx queue to find the tc_to_txq map or 2996 * XPS map for itself. 2997 */ 2998 while (count--) 2999 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev; 3000 3001 return 0; 3002 } 3003 EXPORT_SYMBOL(netdev_bind_sb_channel_queue); 3004 3005 int netdev_set_sb_channel(struct net_device *dev, u16 channel) 3006 { 3007 /* Do not use a multiqueue device to represent a subordinate channel */ 3008 if (netif_is_multiqueue(dev)) 3009 return -ENODEV; 3010 3011 /* We allow channels 1 - 32767 to be used for subordinate channels. 3012 * Channel 0 is meant to be "native" mode and used only to represent 3013 * the main root device. We allow writing 0 to reset the device back 3014 * to normal mode after being used as a subordinate channel. 3015 */ 3016 if (channel > S16_MAX) 3017 return -EINVAL; 3018 3019 dev->num_tc = -channel; 3020 3021 return 0; 3022 } 3023 EXPORT_SYMBOL(netdev_set_sb_channel); 3024 3025 /* 3026 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues 3027 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed. 3028 */ 3029 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) 3030 { 3031 bool disabling; 3032 int rc; 3033 3034 disabling = txq < dev->real_num_tx_queues; 3035 3036 if (txq < 1 || txq > dev->num_tx_queues) 3037 return -EINVAL; 3038 3039 if (dev->reg_state == NETREG_REGISTERED || 3040 dev->reg_state == NETREG_UNREGISTERING) { 3041 ASSERT_RTNL(); 3042 3043 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, 3044 txq); 3045 if (rc) 3046 return rc; 3047 3048 if (dev->num_tc) 3049 netif_setup_tc(dev, txq); 3050 3051 dev->real_num_tx_queues = txq; 3052 3053 if (disabling) { 3054 synchronize_net(); 3055 qdisc_reset_all_tx_gt(dev, txq); 3056 #ifdef CONFIG_XPS 3057 netif_reset_xps_queues_gt(dev, txq); 3058 #endif 3059 } 3060 } else { 3061 dev->real_num_tx_queues = txq; 3062 } 3063 3064 return 0; 3065 } 3066 EXPORT_SYMBOL(netif_set_real_num_tx_queues); 3067 3068 #ifdef CONFIG_SYSFS 3069 /** 3070 * netif_set_real_num_rx_queues - set actual number of RX queues used 3071 * @dev: Network device 3072 * @rxq: Actual number of RX queues 3073 * 3074 * This must be called either with the rtnl_lock held or before 3075 * registration of the net device. Returns 0 on success, or a 3076 * negative error code. If called before registration, it always 3077 * succeeds. 3078 */ 3079 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) 3080 { 3081 int rc; 3082 3083 if (rxq < 1 || rxq > dev->num_rx_queues) 3084 return -EINVAL; 3085 3086 if (dev->reg_state == NETREG_REGISTERED) { 3087 ASSERT_RTNL(); 3088 3089 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, 3090 rxq); 3091 if (rc) 3092 return rc; 3093 } 3094 3095 dev->real_num_rx_queues = rxq; 3096 return 0; 3097 } 3098 EXPORT_SYMBOL(netif_set_real_num_rx_queues); 3099 #endif 3100 3101 /** 3102 * netif_get_num_default_rss_queues - default number of RSS queues 3103 * 3104 * This routine should set an upper limit on the number of RSS queues 3105 * used by default by multiqueue devices. 3106 */ 3107 int netif_get_num_default_rss_queues(void) 3108 { 3109 return is_kdump_kernel() ? 3110 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus()); 3111 } 3112 EXPORT_SYMBOL(netif_get_num_default_rss_queues); 3113 3114 static void __netif_reschedule(struct Qdisc *q) 3115 { 3116 struct softnet_data *sd; 3117 unsigned long flags; 3118 3119 local_irq_save(flags); 3120 sd = this_cpu_ptr(&softnet_data); 3121 q->next_sched = NULL; 3122 *sd->output_queue_tailp = q; 3123 sd->output_queue_tailp = &q->next_sched; 3124 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3125 local_irq_restore(flags); 3126 } 3127 3128 void __netif_schedule(struct Qdisc *q) 3129 { 3130 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) 3131 __netif_reschedule(q); 3132 } 3133 EXPORT_SYMBOL(__netif_schedule); 3134 3135 struct dev_kfree_skb_cb { 3136 enum skb_free_reason reason; 3137 }; 3138 3139 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) 3140 { 3141 return (struct dev_kfree_skb_cb *)skb->cb; 3142 } 3143 3144 void netif_schedule_queue(struct netdev_queue *txq) 3145 { 3146 rcu_read_lock(); 3147 if (!netif_xmit_stopped(txq)) { 3148 struct Qdisc *q = rcu_dereference(txq->qdisc); 3149 3150 __netif_schedule(q); 3151 } 3152 rcu_read_unlock(); 3153 } 3154 EXPORT_SYMBOL(netif_schedule_queue); 3155 3156 void netif_tx_wake_queue(struct netdev_queue *dev_queue) 3157 { 3158 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { 3159 struct Qdisc *q; 3160 3161 rcu_read_lock(); 3162 q = rcu_dereference(dev_queue->qdisc); 3163 __netif_schedule(q); 3164 rcu_read_unlock(); 3165 } 3166 } 3167 EXPORT_SYMBOL(netif_tx_wake_queue); 3168 3169 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason) 3170 { 3171 unsigned long flags; 3172 3173 if (unlikely(!skb)) 3174 return; 3175 3176 if (likely(refcount_read(&skb->users) == 1)) { 3177 smp_rmb(); 3178 refcount_set(&skb->users, 0); 3179 } else if (likely(!refcount_dec_and_test(&skb->users))) { 3180 return; 3181 } 3182 get_kfree_skb_cb(skb)->reason = reason; 3183 local_irq_save(flags); 3184 skb->next = __this_cpu_read(softnet_data.completion_queue); 3185 __this_cpu_write(softnet_data.completion_queue, skb); 3186 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3187 local_irq_restore(flags); 3188 } 3189 EXPORT_SYMBOL(__dev_kfree_skb_irq); 3190 3191 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason) 3192 { 3193 if (in_irq() || irqs_disabled()) 3194 __dev_kfree_skb_irq(skb, reason); 3195 else 3196 dev_kfree_skb(skb); 3197 } 3198 EXPORT_SYMBOL(__dev_kfree_skb_any); 3199 3200 3201 /** 3202 * netif_device_detach - mark device as removed 3203 * @dev: network device 3204 * 3205 * Mark device as removed from system and therefore no longer available. 3206 */ 3207 void netif_device_detach(struct net_device *dev) 3208 { 3209 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && 3210 netif_running(dev)) { 3211 netif_tx_stop_all_queues(dev); 3212 } 3213 } 3214 EXPORT_SYMBOL(netif_device_detach); 3215 3216 /** 3217 * netif_device_attach - mark device as attached 3218 * @dev: network device 3219 * 3220 * Mark device as attached from system and restart if needed. 3221 */ 3222 void netif_device_attach(struct net_device *dev) 3223 { 3224 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && 3225 netif_running(dev)) { 3226 netif_tx_wake_all_queues(dev); 3227 __netdev_watchdog_up(dev); 3228 } 3229 } 3230 EXPORT_SYMBOL(netif_device_attach); 3231 3232 /* 3233 * Returns a Tx hash based on the given packet descriptor a Tx queues' number 3234 * to be used as a distribution range. 3235 */ 3236 static u16 skb_tx_hash(const struct net_device *dev, 3237 const struct net_device *sb_dev, 3238 struct sk_buff *skb) 3239 { 3240 u32 hash; 3241 u16 qoffset = 0; 3242 u16 qcount = dev->real_num_tx_queues; 3243 3244 if (dev->num_tc) { 3245 u8 tc = netdev_get_prio_tc_map(dev, skb->priority); 3246 3247 qoffset = sb_dev->tc_to_txq[tc].offset; 3248 qcount = sb_dev->tc_to_txq[tc].count; 3249 } 3250 3251 if (skb_rx_queue_recorded(skb)) { 3252 hash = skb_get_rx_queue(skb); 3253 if (hash >= qoffset) 3254 hash -= qoffset; 3255 while (unlikely(hash >= qcount)) 3256 hash -= qcount; 3257 return hash + qoffset; 3258 } 3259 3260 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; 3261 } 3262 3263 static void skb_warn_bad_offload(const struct sk_buff *skb) 3264 { 3265 static const netdev_features_t null_features; 3266 struct net_device *dev = skb->dev; 3267 const char *name = ""; 3268 3269 if (!net_ratelimit()) 3270 return; 3271 3272 if (dev) { 3273 if (dev->dev.parent) 3274 name = dev_driver_string(dev->dev.parent); 3275 else 3276 name = netdev_name(dev); 3277 } 3278 skb_dump(KERN_WARNING, skb, false); 3279 WARN(1, "%s: caps=(%pNF, %pNF)\n", 3280 name, dev ? &dev->features : &null_features, 3281 skb->sk ? &skb->sk->sk_route_caps : &null_features); 3282 } 3283 3284 /* 3285 * Invalidate hardware checksum when packet is to be mangled, and 3286 * complete checksum manually on outgoing path. 3287 */ 3288 int skb_checksum_help(struct sk_buff *skb) 3289 { 3290 __wsum csum; 3291 int ret = 0, offset; 3292 3293 if (skb->ip_summed == CHECKSUM_COMPLETE) 3294 goto out_set_summed; 3295 3296 if (unlikely(skb_is_gso(skb))) { 3297 skb_warn_bad_offload(skb); 3298 return -EINVAL; 3299 } 3300 3301 /* Before computing a checksum, we should make sure no frag could 3302 * be modified by an external entity : checksum could be wrong. 3303 */ 3304 if (skb_has_shared_frag(skb)) { 3305 ret = __skb_linearize(skb); 3306 if (ret) 3307 goto out; 3308 } 3309 3310 offset = skb_checksum_start_offset(skb); 3311 BUG_ON(offset >= skb_headlen(skb)); 3312 csum = skb_checksum(skb, offset, skb->len - offset, 0); 3313 3314 offset += skb->csum_offset; 3315 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); 3316 3317 ret = skb_ensure_writable(skb, offset + sizeof(__sum16)); 3318 if (ret) 3319 goto out; 3320 3321 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0; 3322 out_set_summed: 3323 skb->ip_summed = CHECKSUM_NONE; 3324 out: 3325 return ret; 3326 } 3327 EXPORT_SYMBOL(skb_checksum_help); 3328 3329 int skb_crc32c_csum_help(struct sk_buff *skb) 3330 { 3331 __le32 crc32c_csum; 3332 int ret = 0, offset, start; 3333 3334 if (skb->ip_summed != CHECKSUM_PARTIAL) 3335 goto out; 3336 3337 if (unlikely(skb_is_gso(skb))) 3338 goto out; 3339 3340 /* Before computing a checksum, we should make sure no frag could 3341 * be modified by an external entity : checksum could be wrong. 3342 */ 3343 if (unlikely(skb_has_shared_frag(skb))) { 3344 ret = __skb_linearize(skb); 3345 if (ret) 3346 goto out; 3347 } 3348 start = skb_checksum_start_offset(skb); 3349 offset = start + offsetof(struct sctphdr, checksum); 3350 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { 3351 ret = -EINVAL; 3352 goto out; 3353 } 3354 3355 ret = skb_ensure_writable(skb, offset + sizeof(__le32)); 3356 if (ret) 3357 goto out; 3358 3359 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start, 3360 skb->len - start, ~(__u32)0, 3361 crc32c_csum_stub)); 3362 *(__le32 *)(skb->data + offset) = crc32c_csum; 3363 skb->ip_summed = CHECKSUM_NONE; 3364 skb->csum_not_inet = 0; 3365 out: 3366 return ret; 3367 } 3368 3369 __be16 skb_network_protocol(struct sk_buff *skb, int *depth) 3370 { 3371 __be16 type = skb->protocol; 3372 3373 /* Tunnel gso handlers can set protocol to ethernet. */ 3374 if (type == htons(ETH_P_TEB)) { 3375 struct ethhdr *eth; 3376 3377 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) 3378 return 0; 3379 3380 eth = (struct ethhdr *)skb->data; 3381 type = eth->h_proto; 3382 } 3383 3384 return __vlan_get_protocol(skb, type, depth); 3385 } 3386 3387 /** 3388 * skb_mac_gso_segment - mac layer segmentation handler. 3389 * @skb: buffer to segment 3390 * @features: features for the output path (see dev->features) 3391 */ 3392 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb, 3393 netdev_features_t features) 3394 { 3395 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); 3396 struct packet_offload *ptype; 3397 int vlan_depth = skb->mac_len; 3398 __be16 type = skb_network_protocol(skb, &vlan_depth); 3399 3400 if (unlikely(!type)) 3401 return ERR_PTR(-EINVAL); 3402 3403 __skb_pull(skb, vlan_depth); 3404 3405 rcu_read_lock(); 3406 list_for_each_entry_rcu(ptype, &offload_base, list) { 3407 if (ptype->type == type && ptype->callbacks.gso_segment) { 3408 segs = ptype->callbacks.gso_segment(skb, features); 3409 break; 3410 } 3411 } 3412 rcu_read_unlock(); 3413 3414 __skb_push(skb, skb->data - skb_mac_header(skb)); 3415 3416 return segs; 3417 } 3418 EXPORT_SYMBOL(skb_mac_gso_segment); 3419 3420 3421 /* openvswitch calls this on rx path, so we need a different check. 3422 */ 3423 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path) 3424 { 3425 if (tx_path) 3426 return skb->ip_summed != CHECKSUM_PARTIAL && 3427 skb->ip_summed != CHECKSUM_UNNECESSARY; 3428 3429 return skb->ip_summed == CHECKSUM_NONE; 3430 } 3431 3432 /** 3433 * __skb_gso_segment - Perform segmentation on skb. 3434 * @skb: buffer to segment 3435 * @features: features for the output path (see dev->features) 3436 * @tx_path: whether it is called in TX path 3437 * 3438 * This function segments the given skb and returns a list of segments. 3439 * 3440 * It may return NULL if the skb requires no segmentation. This is 3441 * only possible when GSO is used for verifying header integrity. 3442 * 3443 * Segmentation preserves SKB_GSO_CB_OFFSET bytes of previous skb cb. 3444 */ 3445 struct sk_buff *__skb_gso_segment(struct sk_buff *skb, 3446 netdev_features_t features, bool tx_path) 3447 { 3448 struct sk_buff *segs; 3449 3450 if (unlikely(skb_needs_check(skb, tx_path))) { 3451 int err; 3452 3453 /* We're going to init ->check field in TCP or UDP header */ 3454 err = skb_cow_head(skb, 0); 3455 if (err < 0) 3456 return ERR_PTR(err); 3457 } 3458 3459 /* Only report GSO partial support if it will enable us to 3460 * support segmentation on this frame without needing additional 3461 * work. 3462 */ 3463 if (features & NETIF_F_GSO_PARTIAL) { 3464 netdev_features_t partial_features = NETIF_F_GSO_ROBUST; 3465 struct net_device *dev = skb->dev; 3466 3467 partial_features |= dev->features & dev->gso_partial_features; 3468 if (!skb_gso_ok(skb, features | partial_features)) 3469 features &= ~NETIF_F_GSO_PARTIAL; 3470 } 3471 3472 BUILD_BUG_ON(SKB_GSO_CB_OFFSET + 3473 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb)); 3474 3475 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb); 3476 SKB_GSO_CB(skb)->encap_level = 0; 3477 3478 skb_reset_mac_header(skb); 3479 skb_reset_mac_len(skb); 3480 3481 segs = skb_mac_gso_segment(skb, features); 3482 3483 if (segs != skb && unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs))) 3484 skb_warn_bad_offload(skb); 3485 3486 return segs; 3487 } 3488 EXPORT_SYMBOL(__skb_gso_segment); 3489 3490 /* Take action when hardware reception checksum errors are detected. */ 3491 #ifdef CONFIG_BUG 3492 static void do_netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3493 { 3494 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>"); 3495 skb_dump(KERN_ERR, skb, true); 3496 dump_stack(); 3497 } 3498 3499 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3500 { 3501 DO_ONCE_LITE(do_netdev_rx_csum_fault, dev, skb); 3502 } 3503 EXPORT_SYMBOL(netdev_rx_csum_fault); 3504 #endif 3505 3506 /* XXX: check that highmem exists at all on the given machine. */ 3507 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) 3508 { 3509 #ifdef CONFIG_HIGHMEM 3510 int i; 3511 3512 if (!(dev->features & NETIF_F_HIGHDMA)) { 3513 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 3514 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 3515 3516 if (PageHighMem(skb_frag_page(frag))) 3517 return 1; 3518 } 3519 } 3520 #endif 3521 return 0; 3522 } 3523 3524 /* If MPLS offload request, verify we are testing hardware MPLS features 3525 * instead of standard features for the netdev. 3526 */ 3527 #if IS_ENABLED(CONFIG_NET_MPLS_GSO) 3528 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3529 netdev_features_t features, 3530 __be16 type) 3531 { 3532 if (eth_p_mpls(type)) 3533 features &= skb->dev->mpls_features; 3534 3535 return features; 3536 } 3537 #else 3538 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3539 netdev_features_t features, 3540 __be16 type) 3541 { 3542 return features; 3543 } 3544 #endif 3545 3546 static netdev_features_t harmonize_features(struct sk_buff *skb, 3547 netdev_features_t features) 3548 { 3549 __be16 type; 3550 3551 type = skb_network_protocol(skb, NULL); 3552 features = net_mpls_features(skb, features, type); 3553 3554 if (skb->ip_summed != CHECKSUM_NONE && 3555 !can_checksum_protocol(features, type)) { 3556 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); 3557 } 3558 if (illegal_highdma(skb->dev, skb)) 3559 features &= ~NETIF_F_SG; 3560 3561 return features; 3562 } 3563 3564 netdev_features_t passthru_features_check(struct sk_buff *skb, 3565 struct net_device *dev, 3566 netdev_features_t features) 3567 { 3568 return features; 3569 } 3570 EXPORT_SYMBOL(passthru_features_check); 3571 3572 static netdev_features_t dflt_features_check(struct sk_buff *skb, 3573 struct net_device *dev, 3574 netdev_features_t features) 3575 { 3576 return vlan_features_check(skb, features); 3577 } 3578 3579 static netdev_features_t gso_features_check(const struct sk_buff *skb, 3580 struct net_device *dev, 3581 netdev_features_t features) 3582 { 3583 u16 gso_segs = skb_shinfo(skb)->gso_segs; 3584 3585 if (gso_segs > dev->gso_max_segs) 3586 return features & ~NETIF_F_GSO_MASK; 3587 3588 if (!skb_shinfo(skb)->gso_type) { 3589 skb_warn_bad_offload(skb); 3590 return features & ~NETIF_F_GSO_MASK; 3591 } 3592 3593 /* Support for GSO partial features requires software 3594 * intervention before we can actually process the packets 3595 * so we need to strip support for any partial features now 3596 * and we can pull them back in after we have partially 3597 * segmented the frame. 3598 */ 3599 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)) 3600 features &= ~dev->gso_partial_features; 3601 3602 /* Make sure to clear the IPv4 ID mangling feature if the 3603 * IPv4 header has the potential to be fragmented. 3604 */ 3605 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { 3606 struct iphdr *iph = skb->encapsulation ? 3607 inner_ip_hdr(skb) : ip_hdr(skb); 3608 3609 if (!(iph->frag_off & htons(IP_DF))) 3610 features &= ~NETIF_F_TSO_MANGLEID; 3611 } 3612 3613 return features; 3614 } 3615 3616 netdev_features_t netif_skb_features(struct sk_buff *skb) 3617 { 3618 struct net_device *dev = skb->dev; 3619 netdev_features_t features = dev->features; 3620 3621 if (skb_is_gso(skb)) 3622 features = gso_features_check(skb, dev, features); 3623 3624 /* If encapsulation offload request, verify we are testing 3625 * hardware encapsulation features instead of standard 3626 * features for the netdev 3627 */ 3628 if (skb->encapsulation) 3629 features &= dev->hw_enc_features; 3630 3631 if (skb_vlan_tagged(skb)) 3632 features = netdev_intersect_features(features, 3633 dev->vlan_features | 3634 NETIF_F_HW_VLAN_CTAG_TX | 3635 NETIF_F_HW_VLAN_STAG_TX); 3636 3637 if (dev->netdev_ops->ndo_features_check) 3638 features &= dev->netdev_ops->ndo_features_check(skb, dev, 3639 features); 3640 else 3641 features &= dflt_features_check(skb, dev, features); 3642 3643 return harmonize_features(skb, features); 3644 } 3645 EXPORT_SYMBOL(netif_skb_features); 3646 3647 static int xmit_one(struct sk_buff *skb, struct net_device *dev, 3648 struct netdev_queue *txq, bool more) 3649 { 3650 unsigned int len; 3651 int rc; 3652 3653 if (dev_nit_active(dev)) 3654 dev_queue_xmit_nit(skb, dev); 3655 3656 len = skb->len; 3657 PRANDOM_ADD_NOISE(skb, dev, txq, len + jiffies); 3658 trace_net_dev_start_xmit(skb, dev); 3659 rc = netdev_start_xmit(skb, dev, txq, more); 3660 trace_net_dev_xmit(skb, rc, dev, len); 3661 3662 return rc; 3663 } 3664 3665 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, 3666 struct netdev_queue *txq, int *ret) 3667 { 3668 struct sk_buff *skb = first; 3669 int rc = NETDEV_TX_OK; 3670 3671 while (skb) { 3672 struct sk_buff *next = skb->next; 3673 3674 skb_mark_not_on_list(skb); 3675 rc = xmit_one(skb, dev, txq, next != NULL); 3676 if (unlikely(!dev_xmit_complete(rc))) { 3677 skb->next = next; 3678 goto out; 3679 } 3680 3681 skb = next; 3682 if (netif_tx_queue_stopped(txq) && skb) { 3683 rc = NETDEV_TX_BUSY; 3684 break; 3685 } 3686 } 3687 3688 out: 3689 *ret = rc; 3690 return skb; 3691 } 3692 3693 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, 3694 netdev_features_t features) 3695 { 3696 if (skb_vlan_tag_present(skb) && 3697 !vlan_hw_offload_capable(features, skb->vlan_proto)) 3698 skb = __vlan_hwaccel_push_inside(skb); 3699 return skb; 3700 } 3701 3702 int skb_csum_hwoffload_help(struct sk_buff *skb, 3703 const netdev_features_t features) 3704 { 3705 if (unlikely(skb_csum_is_sctp(skb))) 3706 return !!(features & NETIF_F_SCTP_CRC) ? 0 : 3707 skb_crc32c_csum_help(skb); 3708 3709 if (features & NETIF_F_HW_CSUM) 3710 return 0; 3711 3712 if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) { 3713 switch (skb->csum_offset) { 3714 case offsetof(struct tcphdr, check): 3715 case offsetof(struct udphdr, check): 3716 return 0; 3717 } 3718 } 3719 3720 return skb_checksum_help(skb); 3721 } 3722 EXPORT_SYMBOL(skb_csum_hwoffload_help); 3723 3724 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again) 3725 { 3726 netdev_features_t features; 3727 3728 features = netif_skb_features(skb); 3729 skb = validate_xmit_vlan(skb, features); 3730 if (unlikely(!skb)) 3731 goto out_null; 3732 3733 skb = sk_validate_xmit_skb(skb, dev); 3734 if (unlikely(!skb)) 3735 goto out_null; 3736 3737 if (netif_needs_gso(skb, features)) { 3738 struct sk_buff *segs; 3739 3740 segs = skb_gso_segment(skb, features); 3741 if (IS_ERR(segs)) { 3742 goto out_kfree_skb; 3743 } else if (segs) { 3744 consume_skb(skb); 3745 skb = segs; 3746 } 3747 } else { 3748 if (skb_needs_linearize(skb, features) && 3749 __skb_linearize(skb)) 3750 goto out_kfree_skb; 3751 3752 /* If packet is not checksummed and device does not 3753 * support checksumming for this protocol, complete 3754 * checksumming here. 3755 */ 3756 if (skb->ip_summed == CHECKSUM_PARTIAL) { 3757 if (skb->encapsulation) 3758 skb_set_inner_transport_header(skb, 3759 skb_checksum_start_offset(skb)); 3760 else 3761 skb_set_transport_header(skb, 3762 skb_checksum_start_offset(skb)); 3763 if (skb_csum_hwoffload_help(skb, features)) 3764 goto out_kfree_skb; 3765 } 3766 } 3767 3768 skb = validate_xmit_xfrm(skb, features, again); 3769 3770 return skb; 3771 3772 out_kfree_skb: 3773 kfree_skb(skb); 3774 out_null: 3775 atomic_long_inc(&dev->tx_dropped); 3776 return NULL; 3777 } 3778 3779 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again) 3780 { 3781 struct sk_buff *next, *head = NULL, *tail; 3782 3783 for (; skb != NULL; skb = next) { 3784 next = skb->next; 3785 skb_mark_not_on_list(skb); 3786 3787 /* in case skb wont be segmented, point to itself */ 3788 skb->prev = skb; 3789 3790 skb = validate_xmit_skb(skb, dev, again); 3791 if (!skb) 3792 continue; 3793 3794 if (!head) 3795 head = skb; 3796 else 3797 tail->next = skb; 3798 /* If skb was segmented, skb->prev points to 3799 * the last segment. If not, it still contains skb. 3800 */ 3801 tail = skb->prev; 3802 } 3803 return head; 3804 } 3805 EXPORT_SYMBOL_GPL(validate_xmit_skb_list); 3806 3807 static void qdisc_pkt_len_init(struct sk_buff *skb) 3808 { 3809 const struct skb_shared_info *shinfo = skb_shinfo(skb); 3810 3811 qdisc_skb_cb(skb)->pkt_len = skb->len; 3812 3813 /* To get more precise estimation of bytes sent on wire, 3814 * we add to pkt_len the headers size of all segments 3815 */ 3816 if (shinfo->gso_size && skb_transport_header_was_set(skb)) { 3817 unsigned int hdr_len; 3818 u16 gso_segs = shinfo->gso_segs; 3819 3820 /* mac layer + network layer */ 3821 hdr_len = skb_transport_header(skb) - skb_mac_header(skb); 3822 3823 /* + transport layer */ 3824 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { 3825 const struct tcphdr *th; 3826 struct tcphdr _tcphdr; 3827 3828 th = skb_header_pointer(skb, skb_transport_offset(skb), 3829 sizeof(_tcphdr), &_tcphdr); 3830 if (likely(th)) 3831 hdr_len += __tcp_hdrlen(th); 3832 } else { 3833 struct udphdr _udphdr; 3834 3835 if (skb_header_pointer(skb, skb_transport_offset(skb), 3836 sizeof(_udphdr), &_udphdr)) 3837 hdr_len += sizeof(struct udphdr); 3838 } 3839 3840 if (shinfo->gso_type & SKB_GSO_DODGY) 3841 gso_segs = DIV_ROUND_UP(skb->len - hdr_len, 3842 shinfo->gso_size); 3843 3844 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; 3845 } 3846 } 3847 3848 static int dev_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *q, 3849 struct sk_buff **to_free, 3850 struct netdev_queue *txq) 3851 { 3852 int rc; 3853 3854 rc = q->enqueue(skb, q, to_free) & NET_XMIT_MASK; 3855 if (rc == NET_XMIT_SUCCESS) 3856 trace_qdisc_enqueue(q, txq, skb); 3857 return rc; 3858 } 3859 3860 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, 3861 struct net_device *dev, 3862 struct netdev_queue *txq) 3863 { 3864 spinlock_t *root_lock = qdisc_lock(q); 3865 struct sk_buff *to_free = NULL; 3866 bool contended; 3867 int rc; 3868 3869 qdisc_calculate_pkt_len(skb, q); 3870 3871 if (q->flags & TCQ_F_NOLOCK) { 3872 if (q->flags & TCQ_F_CAN_BYPASS && nolock_qdisc_is_empty(q) && 3873 qdisc_run_begin(q)) { 3874 /* Retest nolock_qdisc_is_empty() within the protection 3875 * of q->seqlock to protect from racing with requeuing. 3876 */ 3877 if (unlikely(!nolock_qdisc_is_empty(q))) { 3878 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3879 __qdisc_run(q); 3880 qdisc_run_end(q); 3881 3882 goto no_lock_out; 3883 } 3884 3885 qdisc_bstats_cpu_update(q, skb); 3886 if (sch_direct_xmit(skb, q, dev, txq, NULL, true) && 3887 !nolock_qdisc_is_empty(q)) 3888 __qdisc_run(q); 3889 3890 qdisc_run_end(q); 3891 return NET_XMIT_SUCCESS; 3892 } 3893 3894 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3895 qdisc_run(q); 3896 3897 no_lock_out: 3898 if (unlikely(to_free)) 3899 kfree_skb_list(to_free); 3900 return rc; 3901 } 3902 3903 /* 3904 * Heuristic to force contended enqueues to serialize on a 3905 * separate lock before trying to get qdisc main lock. 3906 * This permits qdisc->running owner to get the lock more 3907 * often and dequeue packets faster. 3908 */ 3909 contended = qdisc_is_running(q); 3910 if (unlikely(contended)) 3911 spin_lock(&q->busylock); 3912 3913 spin_lock(root_lock); 3914 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { 3915 __qdisc_drop(skb, &to_free); 3916 rc = NET_XMIT_DROP; 3917 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && 3918 qdisc_run_begin(q)) { 3919 /* 3920 * This is a work-conserving queue; there are no old skbs 3921 * waiting to be sent out; and the qdisc is not running - 3922 * xmit the skb directly. 3923 */ 3924 3925 qdisc_bstats_update(q, skb); 3926 3927 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { 3928 if (unlikely(contended)) { 3929 spin_unlock(&q->busylock); 3930 contended = false; 3931 } 3932 __qdisc_run(q); 3933 } 3934 3935 qdisc_run_end(q); 3936 rc = NET_XMIT_SUCCESS; 3937 } else { 3938 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3939 if (qdisc_run_begin(q)) { 3940 if (unlikely(contended)) { 3941 spin_unlock(&q->busylock); 3942 contended = false; 3943 } 3944 __qdisc_run(q); 3945 qdisc_run_end(q); 3946 } 3947 } 3948 spin_unlock(root_lock); 3949 if (unlikely(to_free)) 3950 kfree_skb_list(to_free); 3951 if (unlikely(contended)) 3952 spin_unlock(&q->busylock); 3953 return rc; 3954 } 3955 3956 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) 3957 static void skb_update_prio(struct sk_buff *skb) 3958 { 3959 const struct netprio_map *map; 3960 const struct sock *sk; 3961 unsigned int prioidx; 3962 3963 if (skb->priority) 3964 return; 3965 map = rcu_dereference_bh(skb->dev->priomap); 3966 if (!map) 3967 return; 3968 sk = skb_to_full_sk(skb); 3969 if (!sk) 3970 return; 3971 3972 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data); 3973 3974 if (prioidx < map->priomap_len) 3975 skb->priority = map->priomap[prioidx]; 3976 } 3977 #else 3978 #define skb_update_prio(skb) 3979 #endif 3980 3981 /** 3982 * dev_loopback_xmit - loop back @skb 3983 * @net: network namespace this loopback is happening in 3984 * @sk: sk needed to be a netfilter okfn 3985 * @skb: buffer to transmit 3986 */ 3987 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) 3988 { 3989 skb_reset_mac_header(skb); 3990 __skb_pull(skb, skb_network_offset(skb)); 3991 skb->pkt_type = PACKET_LOOPBACK; 3992 skb->ip_summed = CHECKSUM_UNNECESSARY; 3993 WARN_ON(!skb_dst(skb)); 3994 skb_dst_force(skb); 3995 netif_rx_ni(skb); 3996 return 0; 3997 } 3998 EXPORT_SYMBOL(dev_loopback_xmit); 3999 4000 #ifdef CONFIG_NET_EGRESS 4001 static struct sk_buff * 4002 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 4003 { 4004 struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress); 4005 struct tcf_result cl_res; 4006 4007 if (!miniq) 4008 return skb; 4009 4010 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */ 4011 qdisc_skb_cb(skb)->mru = 0; 4012 qdisc_skb_cb(skb)->post_ct = false; 4013 mini_qdisc_bstats_cpu_update(miniq, skb); 4014 4015 switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) { 4016 case TC_ACT_OK: 4017 case TC_ACT_RECLASSIFY: 4018 skb->tc_index = TC_H_MIN(cl_res.classid); 4019 break; 4020 case TC_ACT_SHOT: 4021 mini_qdisc_qstats_cpu_drop(miniq); 4022 *ret = NET_XMIT_DROP; 4023 kfree_skb(skb); 4024 return NULL; 4025 case TC_ACT_STOLEN: 4026 case TC_ACT_QUEUED: 4027 case TC_ACT_TRAP: 4028 *ret = NET_XMIT_SUCCESS; 4029 consume_skb(skb); 4030 return NULL; 4031 case TC_ACT_REDIRECT: 4032 /* No need to push/pop skb's mac_header here on egress! */ 4033 skb_do_redirect(skb); 4034 *ret = NET_XMIT_SUCCESS; 4035 return NULL; 4036 default: 4037 break; 4038 } 4039 4040 return skb; 4041 } 4042 #endif /* CONFIG_NET_EGRESS */ 4043 4044 #ifdef CONFIG_XPS 4045 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb, 4046 struct xps_dev_maps *dev_maps, unsigned int tci) 4047 { 4048 int tc = netdev_get_prio_tc_map(dev, skb->priority); 4049 struct xps_map *map; 4050 int queue_index = -1; 4051 4052 if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids) 4053 return queue_index; 4054 4055 tci *= dev_maps->num_tc; 4056 tci += tc; 4057 4058 map = rcu_dereference(dev_maps->attr_map[tci]); 4059 if (map) { 4060 if (map->len == 1) 4061 queue_index = map->queues[0]; 4062 else 4063 queue_index = map->queues[reciprocal_scale( 4064 skb_get_hash(skb), map->len)]; 4065 if (unlikely(queue_index >= dev->real_num_tx_queues)) 4066 queue_index = -1; 4067 } 4068 return queue_index; 4069 } 4070 #endif 4071 4072 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev, 4073 struct sk_buff *skb) 4074 { 4075 #ifdef CONFIG_XPS 4076 struct xps_dev_maps *dev_maps; 4077 struct sock *sk = skb->sk; 4078 int queue_index = -1; 4079 4080 if (!static_key_false(&xps_needed)) 4081 return -1; 4082 4083 rcu_read_lock(); 4084 if (!static_key_false(&xps_rxqs_needed)) 4085 goto get_cpus_map; 4086 4087 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]); 4088 if (dev_maps) { 4089 int tci = sk_rx_queue_get(sk); 4090 4091 if (tci >= 0) 4092 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4093 tci); 4094 } 4095 4096 get_cpus_map: 4097 if (queue_index < 0) { 4098 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]); 4099 if (dev_maps) { 4100 unsigned int tci = skb->sender_cpu - 1; 4101 4102 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4103 tci); 4104 } 4105 } 4106 rcu_read_unlock(); 4107 4108 return queue_index; 4109 #else 4110 return -1; 4111 #endif 4112 } 4113 4114 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb, 4115 struct net_device *sb_dev) 4116 { 4117 return 0; 4118 } 4119 EXPORT_SYMBOL(dev_pick_tx_zero); 4120 4121 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb, 4122 struct net_device *sb_dev) 4123 { 4124 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues; 4125 } 4126 EXPORT_SYMBOL(dev_pick_tx_cpu_id); 4127 4128 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb, 4129 struct net_device *sb_dev) 4130 { 4131 struct sock *sk = skb->sk; 4132 int queue_index = sk_tx_queue_get(sk); 4133 4134 sb_dev = sb_dev ? : dev; 4135 4136 if (queue_index < 0 || skb->ooo_okay || 4137 queue_index >= dev->real_num_tx_queues) { 4138 int new_index = get_xps_queue(dev, sb_dev, skb); 4139 4140 if (new_index < 0) 4141 new_index = skb_tx_hash(dev, sb_dev, skb); 4142 4143 if (queue_index != new_index && sk && 4144 sk_fullsock(sk) && 4145 rcu_access_pointer(sk->sk_dst_cache)) 4146 sk_tx_queue_set(sk, new_index); 4147 4148 queue_index = new_index; 4149 } 4150 4151 return queue_index; 4152 } 4153 EXPORT_SYMBOL(netdev_pick_tx); 4154 4155 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev, 4156 struct sk_buff *skb, 4157 struct net_device *sb_dev) 4158 { 4159 int queue_index = 0; 4160 4161 #ifdef CONFIG_XPS 4162 u32 sender_cpu = skb->sender_cpu - 1; 4163 4164 if (sender_cpu >= (u32)NR_CPUS) 4165 skb->sender_cpu = raw_smp_processor_id() + 1; 4166 #endif 4167 4168 if (dev->real_num_tx_queues != 1) { 4169 const struct net_device_ops *ops = dev->netdev_ops; 4170 4171 if (ops->ndo_select_queue) 4172 queue_index = ops->ndo_select_queue(dev, skb, sb_dev); 4173 else 4174 queue_index = netdev_pick_tx(dev, skb, sb_dev); 4175 4176 queue_index = netdev_cap_txqueue(dev, queue_index); 4177 } 4178 4179 skb_set_queue_mapping(skb, queue_index); 4180 return netdev_get_tx_queue(dev, queue_index); 4181 } 4182 4183 /** 4184 * __dev_queue_xmit - transmit a buffer 4185 * @skb: buffer to transmit 4186 * @sb_dev: suboordinate device used for L2 forwarding offload 4187 * 4188 * Queue a buffer for transmission to a network device. The caller must 4189 * have set the device and priority and built the buffer before calling 4190 * this function. The function can be called from an interrupt. 4191 * 4192 * A negative errno code is returned on a failure. A success does not 4193 * guarantee the frame will be transmitted as it may be dropped due 4194 * to congestion or traffic shaping. 4195 * 4196 * ----------------------------------------------------------------------------------- 4197 * I notice this method can also return errors from the queue disciplines, 4198 * including NET_XMIT_DROP, which is a positive value. So, errors can also 4199 * be positive. 4200 * 4201 * Regardless of the return value, the skb is consumed, so it is currently 4202 * difficult to retry a send to this method. (You can bump the ref count 4203 * before sending to hold a reference for retry if you are careful.) 4204 * 4205 * When calling this method, interrupts MUST be enabled. This is because 4206 * the BH enable code must have IRQs enabled so that it will not deadlock. 4207 * --BLG 4208 */ 4209 static int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev) 4210 { 4211 struct net_device *dev = skb->dev; 4212 struct netdev_queue *txq; 4213 struct Qdisc *q; 4214 int rc = -ENOMEM; 4215 bool again = false; 4216 4217 skb_reset_mac_header(skb); 4218 4219 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) 4220 __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED); 4221 4222 /* Disable soft irqs for various locks below. Also 4223 * stops preemption for RCU. 4224 */ 4225 rcu_read_lock_bh(); 4226 4227 skb_update_prio(skb); 4228 4229 qdisc_pkt_len_init(skb); 4230 #ifdef CONFIG_NET_CLS_ACT 4231 skb->tc_at_ingress = 0; 4232 # ifdef CONFIG_NET_EGRESS 4233 if (static_branch_unlikely(&egress_needed_key)) { 4234 skb = sch_handle_egress(skb, &rc, dev); 4235 if (!skb) 4236 goto out; 4237 } 4238 # endif 4239 #endif 4240 /* If device/qdisc don't need skb->dst, release it right now while 4241 * its hot in this cpu cache. 4242 */ 4243 if (dev->priv_flags & IFF_XMIT_DST_RELEASE) 4244 skb_dst_drop(skb); 4245 else 4246 skb_dst_force(skb); 4247 4248 txq = netdev_core_pick_tx(dev, skb, sb_dev); 4249 q = rcu_dereference_bh(txq->qdisc); 4250 4251 trace_net_dev_queue(skb); 4252 if (q->enqueue) { 4253 rc = __dev_xmit_skb(skb, q, dev, txq); 4254 goto out; 4255 } 4256 4257 /* The device has no queue. Common case for software devices: 4258 * loopback, all the sorts of tunnels... 4259 4260 * Really, it is unlikely that netif_tx_lock protection is necessary 4261 * here. (f.e. loopback and IP tunnels are clean ignoring statistics 4262 * counters.) 4263 * However, it is possible, that they rely on protection 4264 * made by us here. 4265 4266 * Check this and shot the lock. It is not prone from deadlocks. 4267 *Either shot noqueue qdisc, it is even simpler 8) 4268 */ 4269 if (dev->flags & IFF_UP) { 4270 int cpu = smp_processor_id(); /* ok because BHs are off */ 4271 4272 if (txq->xmit_lock_owner != cpu) { 4273 if (dev_xmit_recursion()) 4274 goto recursion_alert; 4275 4276 skb = validate_xmit_skb(skb, dev, &again); 4277 if (!skb) 4278 goto out; 4279 4280 PRANDOM_ADD_NOISE(skb, dev, txq, jiffies); 4281 HARD_TX_LOCK(dev, txq, cpu); 4282 4283 if (!netif_xmit_stopped(txq)) { 4284 dev_xmit_recursion_inc(); 4285 skb = dev_hard_start_xmit(skb, dev, txq, &rc); 4286 dev_xmit_recursion_dec(); 4287 if (dev_xmit_complete(rc)) { 4288 HARD_TX_UNLOCK(dev, txq); 4289 goto out; 4290 } 4291 } 4292 HARD_TX_UNLOCK(dev, txq); 4293 net_crit_ratelimited("Virtual device %s asks to queue packet!\n", 4294 dev->name); 4295 } else { 4296 /* Recursion is detected! It is possible, 4297 * unfortunately 4298 */ 4299 recursion_alert: 4300 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", 4301 dev->name); 4302 } 4303 } 4304 4305 rc = -ENETDOWN; 4306 rcu_read_unlock_bh(); 4307 4308 atomic_long_inc(&dev->tx_dropped); 4309 kfree_skb_list(skb); 4310 return rc; 4311 out: 4312 rcu_read_unlock_bh(); 4313 return rc; 4314 } 4315 4316 int dev_queue_xmit(struct sk_buff *skb) 4317 { 4318 return __dev_queue_xmit(skb, NULL); 4319 } 4320 EXPORT_SYMBOL(dev_queue_xmit); 4321 4322 int dev_queue_xmit_accel(struct sk_buff *skb, struct net_device *sb_dev) 4323 { 4324 return __dev_queue_xmit(skb, sb_dev); 4325 } 4326 EXPORT_SYMBOL(dev_queue_xmit_accel); 4327 4328 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id) 4329 { 4330 struct net_device *dev = skb->dev; 4331 struct sk_buff *orig_skb = skb; 4332 struct netdev_queue *txq; 4333 int ret = NETDEV_TX_BUSY; 4334 bool again = false; 4335 4336 if (unlikely(!netif_running(dev) || 4337 !netif_carrier_ok(dev))) 4338 goto drop; 4339 4340 skb = validate_xmit_skb_list(skb, dev, &again); 4341 if (skb != orig_skb) 4342 goto drop; 4343 4344 skb_set_queue_mapping(skb, queue_id); 4345 txq = skb_get_tx_queue(dev, skb); 4346 PRANDOM_ADD_NOISE(skb, dev, txq, jiffies); 4347 4348 local_bh_disable(); 4349 4350 dev_xmit_recursion_inc(); 4351 HARD_TX_LOCK(dev, txq, smp_processor_id()); 4352 if (!netif_xmit_frozen_or_drv_stopped(txq)) 4353 ret = netdev_start_xmit(skb, dev, txq, false); 4354 HARD_TX_UNLOCK(dev, txq); 4355 dev_xmit_recursion_dec(); 4356 4357 local_bh_enable(); 4358 return ret; 4359 drop: 4360 atomic_long_inc(&dev->tx_dropped); 4361 kfree_skb_list(skb); 4362 return NET_XMIT_DROP; 4363 } 4364 EXPORT_SYMBOL(__dev_direct_xmit); 4365 4366 /************************************************************************* 4367 * Receiver routines 4368 *************************************************************************/ 4369 4370 int netdev_max_backlog __read_mostly = 1000; 4371 EXPORT_SYMBOL(netdev_max_backlog); 4372 4373 int netdev_tstamp_prequeue __read_mostly = 1; 4374 int netdev_budget __read_mostly = 300; 4375 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */ 4376 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ; 4377 int weight_p __read_mostly = 64; /* old backlog weight */ 4378 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ 4379 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ 4380 int dev_rx_weight __read_mostly = 64; 4381 int dev_tx_weight __read_mostly = 64; 4382 /* Maximum number of GRO_NORMAL skbs to batch up for list-RX */ 4383 int gro_normal_batch __read_mostly = 8; 4384 4385 /* Called with irq disabled */ 4386 static inline void ____napi_schedule(struct softnet_data *sd, 4387 struct napi_struct *napi) 4388 { 4389 struct task_struct *thread; 4390 4391 if (test_bit(NAPI_STATE_THREADED, &napi->state)) { 4392 /* Paired with smp_mb__before_atomic() in 4393 * napi_enable()/dev_set_threaded(). 4394 * Use READ_ONCE() to guarantee a complete 4395 * read on napi->thread. Only call 4396 * wake_up_process() when it's not NULL. 4397 */ 4398 thread = READ_ONCE(napi->thread); 4399 if (thread) { 4400 /* Avoid doing set_bit() if the thread is in 4401 * INTERRUPTIBLE state, cause napi_thread_wait() 4402 * makes sure to proceed with napi polling 4403 * if the thread is explicitly woken from here. 4404 */ 4405 if (READ_ONCE(thread->__state) != TASK_INTERRUPTIBLE) 4406 set_bit(NAPI_STATE_SCHED_THREADED, &napi->state); 4407 wake_up_process(thread); 4408 return; 4409 } 4410 } 4411 4412 list_add_tail(&napi->poll_list, &sd->poll_list); 4413 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4414 } 4415 4416 #ifdef CONFIG_RPS 4417 4418 /* One global table that all flow-based protocols share. */ 4419 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; 4420 EXPORT_SYMBOL(rps_sock_flow_table); 4421 u32 rps_cpu_mask __read_mostly; 4422 EXPORT_SYMBOL(rps_cpu_mask); 4423 4424 struct static_key_false rps_needed __read_mostly; 4425 EXPORT_SYMBOL(rps_needed); 4426 struct static_key_false rfs_needed __read_mostly; 4427 EXPORT_SYMBOL(rfs_needed); 4428 4429 static struct rps_dev_flow * 4430 set_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4431 struct rps_dev_flow *rflow, u16 next_cpu) 4432 { 4433 if (next_cpu < nr_cpu_ids) { 4434 #ifdef CONFIG_RFS_ACCEL 4435 struct netdev_rx_queue *rxqueue; 4436 struct rps_dev_flow_table *flow_table; 4437 struct rps_dev_flow *old_rflow; 4438 u32 flow_id; 4439 u16 rxq_index; 4440 int rc; 4441 4442 /* Should we steer this flow to a different hardware queue? */ 4443 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || 4444 !(dev->features & NETIF_F_NTUPLE)) 4445 goto out; 4446 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); 4447 if (rxq_index == skb_get_rx_queue(skb)) 4448 goto out; 4449 4450 rxqueue = dev->_rx + rxq_index; 4451 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4452 if (!flow_table) 4453 goto out; 4454 flow_id = skb_get_hash(skb) & flow_table->mask; 4455 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, 4456 rxq_index, flow_id); 4457 if (rc < 0) 4458 goto out; 4459 old_rflow = rflow; 4460 rflow = &flow_table->flows[flow_id]; 4461 rflow->filter = rc; 4462 if (old_rflow->filter == rflow->filter) 4463 old_rflow->filter = RPS_NO_FILTER; 4464 out: 4465 #endif 4466 rflow->last_qtail = 4467 per_cpu(softnet_data, next_cpu).input_queue_head; 4468 } 4469 4470 rflow->cpu = next_cpu; 4471 return rflow; 4472 } 4473 4474 /* 4475 * get_rps_cpu is called from netif_receive_skb and returns the target 4476 * CPU from the RPS map of the receiving queue for a given skb. 4477 * rcu_read_lock must be held on entry. 4478 */ 4479 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4480 struct rps_dev_flow **rflowp) 4481 { 4482 const struct rps_sock_flow_table *sock_flow_table; 4483 struct netdev_rx_queue *rxqueue = dev->_rx; 4484 struct rps_dev_flow_table *flow_table; 4485 struct rps_map *map; 4486 int cpu = -1; 4487 u32 tcpu; 4488 u32 hash; 4489 4490 if (skb_rx_queue_recorded(skb)) { 4491 u16 index = skb_get_rx_queue(skb); 4492 4493 if (unlikely(index >= dev->real_num_rx_queues)) { 4494 WARN_ONCE(dev->real_num_rx_queues > 1, 4495 "%s received packet on queue %u, but number " 4496 "of RX queues is %u\n", 4497 dev->name, index, dev->real_num_rx_queues); 4498 goto done; 4499 } 4500 rxqueue += index; 4501 } 4502 4503 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ 4504 4505 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4506 map = rcu_dereference(rxqueue->rps_map); 4507 if (!flow_table && !map) 4508 goto done; 4509 4510 skb_reset_network_header(skb); 4511 hash = skb_get_hash(skb); 4512 if (!hash) 4513 goto done; 4514 4515 sock_flow_table = rcu_dereference(rps_sock_flow_table); 4516 if (flow_table && sock_flow_table) { 4517 struct rps_dev_flow *rflow; 4518 u32 next_cpu; 4519 u32 ident; 4520 4521 /* First check into global flow table if there is a match */ 4522 ident = sock_flow_table->ents[hash & sock_flow_table->mask]; 4523 if ((ident ^ hash) & ~rps_cpu_mask) 4524 goto try_rps; 4525 4526 next_cpu = ident & rps_cpu_mask; 4527 4528 /* OK, now we know there is a match, 4529 * we can look at the local (per receive queue) flow table 4530 */ 4531 rflow = &flow_table->flows[hash & flow_table->mask]; 4532 tcpu = rflow->cpu; 4533 4534 /* 4535 * If the desired CPU (where last recvmsg was done) is 4536 * different from current CPU (one in the rx-queue flow 4537 * table entry), switch if one of the following holds: 4538 * - Current CPU is unset (>= nr_cpu_ids). 4539 * - Current CPU is offline. 4540 * - The current CPU's queue tail has advanced beyond the 4541 * last packet that was enqueued using this table entry. 4542 * This guarantees that all previous packets for the flow 4543 * have been dequeued, thus preserving in order delivery. 4544 */ 4545 if (unlikely(tcpu != next_cpu) && 4546 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || 4547 ((int)(per_cpu(softnet_data, tcpu).input_queue_head - 4548 rflow->last_qtail)) >= 0)) { 4549 tcpu = next_cpu; 4550 rflow = set_rps_cpu(dev, skb, rflow, next_cpu); 4551 } 4552 4553 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { 4554 *rflowp = rflow; 4555 cpu = tcpu; 4556 goto done; 4557 } 4558 } 4559 4560 try_rps: 4561 4562 if (map) { 4563 tcpu = map->cpus[reciprocal_scale(hash, map->len)]; 4564 if (cpu_online(tcpu)) { 4565 cpu = tcpu; 4566 goto done; 4567 } 4568 } 4569 4570 done: 4571 return cpu; 4572 } 4573 4574 #ifdef CONFIG_RFS_ACCEL 4575 4576 /** 4577 * rps_may_expire_flow - check whether an RFS hardware filter may be removed 4578 * @dev: Device on which the filter was set 4579 * @rxq_index: RX queue index 4580 * @flow_id: Flow ID passed to ndo_rx_flow_steer() 4581 * @filter_id: Filter ID returned by ndo_rx_flow_steer() 4582 * 4583 * Drivers that implement ndo_rx_flow_steer() should periodically call 4584 * this function for each installed filter and remove the filters for 4585 * which it returns %true. 4586 */ 4587 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, 4588 u32 flow_id, u16 filter_id) 4589 { 4590 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; 4591 struct rps_dev_flow_table *flow_table; 4592 struct rps_dev_flow *rflow; 4593 bool expire = true; 4594 unsigned int cpu; 4595 4596 rcu_read_lock(); 4597 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4598 if (flow_table && flow_id <= flow_table->mask) { 4599 rflow = &flow_table->flows[flow_id]; 4600 cpu = READ_ONCE(rflow->cpu); 4601 if (rflow->filter == filter_id && cpu < nr_cpu_ids && 4602 ((int)(per_cpu(softnet_data, cpu).input_queue_head - 4603 rflow->last_qtail) < 4604 (int)(10 * flow_table->mask))) 4605 expire = false; 4606 } 4607 rcu_read_unlock(); 4608 return expire; 4609 } 4610 EXPORT_SYMBOL(rps_may_expire_flow); 4611 4612 #endif /* CONFIG_RFS_ACCEL */ 4613 4614 /* Called from hardirq (IPI) context */ 4615 static void rps_trigger_softirq(void *data) 4616 { 4617 struct softnet_data *sd = data; 4618 4619 ____napi_schedule(sd, &sd->backlog); 4620 sd->received_rps++; 4621 } 4622 4623 #endif /* CONFIG_RPS */ 4624 4625 /* 4626 * Check if this softnet_data structure is another cpu one 4627 * If yes, queue it to our IPI list and return 1 4628 * If no, return 0 4629 */ 4630 static int rps_ipi_queued(struct softnet_data *sd) 4631 { 4632 #ifdef CONFIG_RPS 4633 struct softnet_data *mysd = this_cpu_ptr(&softnet_data); 4634 4635 if (sd != mysd) { 4636 sd->rps_ipi_next = mysd->rps_ipi_list; 4637 mysd->rps_ipi_list = sd; 4638 4639 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4640 return 1; 4641 } 4642 #endif /* CONFIG_RPS */ 4643 return 0; 4644 } 4645 4646 #ifdef CONFIG_NET_FLOW_LIMIT 4647 int netdev_flow_limit_table_len __read_mostly = (1 << 12); 4648 #endif 4649 4650 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) 4651 { 4652 #ifdef CONFIG_NET_FLOW_LIMIT 4653 struct sd_flow_limit *fl; 4654 struct softnet_data *sd; 4655 unsigned int old_flow, new_flow; 4656 4657 if (qlen < (netdev_max_backlog >> 1)) 4658 return false; 4659 4660 sd = this_cpu_ptr(&softnet_data); 4661 4662 rcu_read_lock(); 4663 fl = rcu_dereference(sd->flow_limit); 4664 if (fl) { 4665 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); 4666 old_flow = fl->history[fl->history_head]; 4667 fl->history[fl->history_head] = new_flow; 4668 4669 fl->history_head++; 4670 fl->history_head &= FLOW_LIMIT_HISTORY - 1; 4671 4672 if (likely(fl->buckets[old_flow])) 4673 fl->buckets[old_flow]--; 4674 4675 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { 4676 fl->count++; 4677 rcu_read_unlock(); 4678 return true; 4679 } 4680 } 4681 rcu_read_unlock(); 4682 #endif 4683 return false; 4684 } 4685 4686 /* 4687 * enqueue_to_backlog is called to queue an skb to a per CPU backlog 4688 * queue (may be a remote CPU queue). 4689 */ 4690 static int enqueue_to_backlog(struct sk_buff *skb, int cpu, 4691 unsigned int *qtail) 4692 { 4693 struct softnet_data *sd; 4694 unsigned long flags; 4695 unsigned int qlen; 4696 4697 sd = &per_cpu(softnet_data, cpu); 4698 4699 local_irq_save(flags); 4700 4701 rps_lock(sd); 4702 if (!netif_running(skb->dev)) 4703 goto drop; 4704 qlen = skb_queue_len(&sd->input_pkt_queue); 4705 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) { 4706 if (qlen) { 4707 enqueue: 4708 __skb_queue_tail(&sd->input_pkt_queue, skb); 4709 input_queue_tail_incr_save(sd, qtail); 4710 rps_unlock(sd); 4711 local_irq_restore(flags); 4712 return NET_RX_SUCCESS; 4713 } 4714 4715 /* Schedule NAPI for backlog device 4716 * We can use non atomic operation since we own the queue lock 4717 */ 4718 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) { 4719 if (!rps_ipi_queued(sd)) 4720 ____napi_schedule(sd, &sd->backlog); 4721 } 4722 goto enqueue; 4723 } 4724 4725 drop: 4726 sd->dropped++; 4727 rps_unlock(sd); 4728 4729 local_irq_restore(flags); 4730 4731 atomic_long_inc(&skb->dev->rx_dropped); 4732 kfree_skb(skb); 4733 return NET_RX_DROP; 4734 } 4735 4736 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb) 4737 { 4738 struct net_device *dev = skb->dev; 4739 struct netdev_rx_queue *rxqueue; 4740 4741 rxqueue = dev->_rx; 4742 4743 if (skb_rx_queue_recorded(skb)) { 4744 u16 index = skb_get_rx_queue(skb); 4745 4746 if (unlikely(index >= dev->real_num_rx_queues)) { 4747 WARN_ONCE(dev->real_num_rx_queues > 1, 4748 "%s received packet on queue %u, but number " 4749 "of RX queues is %u\n", 4750 dev->name, index, dev->real_num_rx_queues); 4751 4752 return rxqueue; /* Return first rxqueue */ 4753 } 4754 rxqueue += index; 4755 } 4756 return rxqueue; 4757 } 4758 4759 static u32 netif_receive_generic_xdp(struct sk_buff *skb, 4760 struct xdp_buff *xdp, 4761 struct bpf_prog *xdp_prog) 4762 { 4763 void *orig_data, *orig_data_end, *hard_start; 4764 struct netdev_rx_queue *rxqueue; 4765 u32 metalen, act = XDP_DROP; 4766 bool orig_bcast, orig_host; 4767 u32 mac_len, frame_sz; 4768 __be16 orig_eth_type; 4769 struct ethhdr *eth; 4770 int off; 4771 4772 /* Reinjected packets coming from act_mirred or similar should 4773 * not get XDP generic processing. 4774 */ 4775 if (skb_is_redirected(skb)) 4776 return XDP_PASS; 4777 4778 /* XDP packets must be linear and must have sufficient headroom 4779 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also 4780 * native XDP provides, thus we need to do it here as well. 4781 */ 4782 if (skb_cloned(skb) || skb_is_nonlinear(skb) || 4783 skb_headroom(skb) < XDP_PACKET_HEADROOM) { 4784 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb); 4785 int troom = skb->tail + skb->data_len - skb->end; 4786 4787 /* In case we have to go down the path and also linearize, 4788 * then lets do the pskb_expand_head() work just once here. 4789 */ 4790 if (pskb_expand_head(skb, 4791 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0, 4792 troom > 0 ? troom + 128 : 0, GFP_ATOMIC)) 4793 goto do_drop; 4794 if (skb_linearize(skb)) 4795 goto do_drop; 4796 } 4797 4798 /* The XDP program wants to see the packet starting at the MAC 4799 * header. 4800 */ 4801 mac_len = skb->data - skb_mac_header(skb); 4802 hard_start = skb->data - skb_headroom(skb); 4803 4804 /* SKB "head" area always have tailroom for skb_shared_info */ 4805 frame_sz = (void *)skb_end_pointer(skb) - hard_start; 4806 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 4807 4808 rxqueue = netif_get_rxqueue(skb); 4809 xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq); 4810 xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len, 4811 skb_headlen(skb) + mac_len, true); 4812 4813 orig_data_end = xdp->data_end; 4814 orig_data = xdp->data; 4815 eth = (struct ethhdr *)xdp->data; 4816 orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr); 4817 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest); 4818 orig_eth_type = eth->h_proto; 4819 4820 act = bpf_prog_run_xdp(xdp_prog, xdp); 4821 4822 /* check if bpf_xdp_adjust_head was used */ 4823 off = xdp->data - orig_data; 4824 if (off) { 4825 if (off > 0) 4826 __skb_pull(skb, off); 4827 else if (off < 0) 4828 __skb_push(skb, -off); 4829 4830 skb->mac_header += off; 4831 skb_reset_network_header(skb); 4832 } 4833 4834 /* check if bpf_xdp_adjust_tail was used */ 4835 off = xdp->data_end - orig_data_end; 4836 if (off != 0) { 4837 skb_set_tail_pointer(skb, xdp->data_end - xdp->data); 4838 skb->len += off; /* positive on grow, negative on shrink */ 4839 } 4840 4841 /* check if XDP changed eth hdr such SKB needs update */ 4842 eth = (struct ethhdr *)xdp->data; 4843 if ((orig_eth_type != eth->h_proto) || 4844 (orig_host != ether_addr_equal_64bits(eth->h_dest, 4845 skb->dev->dev_addr)) || 4846 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) { 4847 __skb_push(skb, ETH_HLEN); 4848 skb->pkt_type = PACKET_HOST; 4849 skb->protocol = eth_type_trans(skb, skb->dev); 4850 } 4851 4852 switch (act) { 4853 case XDP_REDIRECT: 4854 case XDP_TX: 4855 __skb_push(skb, mac_len); 4856 break; 4857 case XDP_PASS: 4858 metalen = xdp->data - xdp->data_meta; 4859 if (metalen) 4860 skb_metadata_set(skb, metalen); 4861 break; 4862 default: 4863 bpf_warn_invalid_xdp_action(act); 4864 fallthrough; 4865 case XDP_ABORTED: 4866 trace_xdp_exception(skb->dev, xdp_prog, act); 4867 fallthrough; 4868 case XDP_DROP: 4869 do_drop: 4870 kfree_skb(skb); 4871 break; 4872 } 4873 4874 return act; 4875 } 4876 4877 /* When doing generic XDP we have to bypass the qdisc layer and the 4878 * network taps in order to match in-driver-XDP behavior. 4879 */ 4880 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog) 4881 { 4882 struct net_device *dev = skb->dev; 4883 struct netdev_queue *txq; 4884 bool free_skb = true; 4885 int cpu, rc; 4886 4887 txq = netdev_core_pick_tx(dev, skb, NULL); 4888 cpu = smp_processor_id(); 4889 HARD_TX_LOCK(dev, txq, cpu); 4890 if (!netif_xmit_stopped(txq)) { 4891 rc = netdev_start_xmit(skb, dev, txq, 0); 4892 if (dev_xmit_complete(rc)) 4893 free_skb = false; 4894 } 4895 HARD_TX_UNLOCK(dev, txq); 4896 if (free_skb) { 4897 trace_xdp_exception(dev, xdp_prog, XDP_TX); 4898 kfree_skb(skb); 4899 } 4900 } 4901 4902 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key); 4903 4904 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb) 4905 { 4906 if (xdp_prog) { 4907 struct xdp_buff xdp; 4908 u32 act; 4909 int err; 4910 4911 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog); 4912 if (act != XDP_PASS) { 4913 switch (act) { 4914 case XDP_REDIRECT: 4915 err = xdp_do_generic_redirect(skb->dev, skb, 4916 &xdp, xdp_prog); 4917 if (err) 4918 goto out_redir; 4919 break; 4920 case XDP_TX: 4921 generic_xdp_tx(skb, xdp_prog); 4922 break; 4923 } 4924 return XDP_DROP; 4925 } 4926 } 4927 return XDP_PASS; 4928 out_redir: 4929 kfree_skb(skb); 4930 return XDP_DROP; 4931 } 4932 EXPORT_SYMBOL_GPL(do_xdp_generic); 4933 4934 static int netif_rx_internal(struct sk_buff *skb) 4935 { 4936 int ret; 4937 4938 net_timestamp_check(netdev_tstamp_prequeue, skb); 4939 4940 trace_netif_rx(skb); 4941 4942 #ifdef CONFIG_RPS 4943 if (static_branch_unlikely(&rps_needed)) { 4944 struct rps_dev_flow voidflow, *rflow = &voidflow; 4945 int cpu; 4946 4947 preempt_disable(); 4948 rcu_read_lock(); 4949 4950 cpu = get_rps_cpu(skb->dev, skb, &rflow); 4951 if (cpu < 0) 4952 cpu = smp_processor_id(); 4953 4954 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 4955 4956 rcu_read_unlock(); 4957 preempt_enable(); 4958 } else 4959 #endif 4960 { 4961 unsigned int qtail; 4962 4963 ret = enqueue_to_backlog(skb, get_cpu(), &qtail); 4964 put_cpu(); 4965 } 4966 return ret; 4967 } 4968 4969 /** 4970 * netif_rx - post buffer to the network code 4971 * @skb: buffer to post 4972 * 4973 * This function receives a packet from a device driver and queues it for 4974 * the upper (protocol) levels to process. It always succeeds. The buffer 4975 * may be dropped during processing for congestion control or by the 4976 * protocol layers. 4977 * 4978 * return values: 4979 * NET_RX_SUCCESS (no congestion) 4980 * NET_RX_DROP (packet was dropped) 4981 * 4982 */ 4983 4984 int netif_rx(struct sk_buff *skb) 4985 { 4986 int ret; 4987 4988 trace_netif_rx_entry(skb); 4989 4990 ret = netif_rx_internal(skb); 4991 trace_netif_rx_exit(ret); 4992 4993 return ret; 4994 } 4995 EXPORT_SYMBOL(netif_rx); 4996 4997 int netif_rx_ni(struct sk_buff *skb) 4998 { 4999 int err; 5000 5001 trace_netif_rx_ni_entry(skb); 5002 5003 preempt_disable(); 5004 err = netif_rx_internal(skb); 5005 if (local_softirq_pending()) 5006 do_softirq(); 5007 preempt_enable(); 5008 trace_netif_rx_ni_exit(err); 5009 5010 return err; 5011 } 5012 EXPORT_SYMBOL(netif_rx_ni); 5013 5014 int netif_rx_any_context(struct sk_buff *skb) 5015 { 5016 /* 5017 * If invoked from contexts which do not invoke bottom half 5018 * processing either at return from interrupt or when softrqs are 5019 * reenabled, use netif_rx_ni() which invokes bottomhalf processing 5020 * directly. 5021 */ 5022 if (in_interrupt()) 5023 return netif_rx(skb); 5024 else 5025 return netif_rx_ni(skb); 5026 } 5027 EXPORT_SYMBOL(netif_rx_any_context); 5028 5029 static __latent_entropy void net_tx_action(struct softirq_action *h) 5030 { 5031 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 5032 5033 if (sd->completion_queue) { 5034 struct sk_buff *clist; 5035 5036 local_irq_disable(); 5037 clist = sd->completion_queue; 5038 sd->completion_queue = NULL; 5039 local_irq_enable(); 5040 5041 while (clist) { 5042 struct sk_buff *skb = clist; 5043 5044 clist = clist->next; 5045 5046 WARN_ON(refcount_read(&skb->users)); 5047 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED)) 5048 trace_consume_skb(skb); 5049 else 5050 trace_kfree_skb(skb, net_tx_action); 5051 5052 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) 5053 __kfree_skb(skb); 5054 else 5055 __kfree_skb_defer(skb); 5056 } 5057 } 5058 5059 if (sd->output_queue) { 5060 struct Qdisc *head; 5061 5062 local_irq_disable(); 5063 head = sd->output_queue; 5064 sd->output_queue = NULL; 5065 sd->output_queue_tailp = &sd->output_queue; 5066 local_irq_enable(); 5067 5068 rcu_read_lock(); 5069 5070 while (head) { 5071 struct Qdisc *q = head; 5072 spinlock_t *root_lock = NULL; 5073 5074 head = head->next_sched; 5075 5076 /* We need to make sure head->next_sched is read 5077 * before clearing __QDISC_STATE_SCHED 5078 */ 5079 smp_mb__before_atomic(); 5080 5081 if (!(q->flags & TCQ_F_NOLOCK)) { 5082 root_lock = qdisc_lock(q); 5083 spin_lock(root_lock); 5084 } else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, 5085 &q->state))) { 5086 /* There is a synchronize_net() between 5087 * STATE_DEACTIVATED flag being set and 5088 * qdisc_reset()/some_qdisc_is_busy() in 5089 * dev_deactivate(), so we can safely bail out 5090 * early here to avoid data race between 5091 * qdisc_deactivate() and some_qdisc_is_busy() 5092 * for lockless qdisc. 5093 */ 5094 clear_bit(__QDISC_STATE_SCHED, &q->state); 5095 continue; 5096 } 5097 5098 clear_bit(__QDISC_STATE_SCHED, &q->state); 5099 qdisc_run(q); 5100 if (root_lock) 5101 spin_unlock(root_lock); 5102 } 5103 5104 rcu_read_unlock(); 5105 } 5106 5107 xfrm_dev_backlog(sd); 5108 } 5109 5110 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) 5111 /* This hook is defined here for ATM LANE */ 5112 int (*br_fdb_test_addr_hook)(struct net_device *dev, 5113 unsigned char *addr) __read_mostly; 5114 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); 5115 #endif 5116 5117 static inline struct sk_buff * 5118 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 5119 struct net_device *orig_dev, bool *another) 5120 { 5121 #ifdef CONFIG_NET_CLS_ACT 5122 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress); 5123 struct tcf_result cl_res; 5124 5125 /* If there's at least one ingress present somewhere (so 5126 * we get here via enabled static key), remaining devices 5127 * that are not configured with an ingress qdisc will bail 5128 * out here. 5129 */ 5130 if (!miniq) 5131 return skb; 5132 5133 if (*pt_prev) { 5134 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5135 *pt_prev = NULL; 5136 } 5137 5138 qdisc_skb_cb(skb)->pkt_len = skb->len; 5139 qdisc_skb_cb(skb)->mru = 0; 5140 qdisc_skb_cb(skb)->post_ct = false; 5141 skb->tc_at_ingress = 1; 5142 mini_qdisc_bstats_cpu_update(miniq, skb); 5143 5144 switch (tcf_classify_ingress(skb, miniq->block, miniq->filter_list, 5145 &cl_res, false)) { 5146 case TC_ACT_OK: 5147 case TC_ACT_RECLASSIFY: 5148 skb->tc_index = TC_H_MIN(cl_res.classid); 5149 break; 5150 case TC_ACT_SHOT: 5151 mini_qdisc_qstats_cpu_drop(miniq); 5152 kfree_skb(skb); 5153 return NULL; 5154 case TC_ACT_STOLEN: 5155 case TC_ACT_QUEUED: 5156 case TC_ACT_TRAP: 5157 consume_skb(skb); 5158 return NULL; 5159 case TC_ACT_REDIRECT: 5160 /* skb_mac_header check was done by cls/act_bpf, so 5161 * we can safely push the L2 header back before 5162 * redirecting to another netdev 5163 */ 5164 __skb_push(skb, skb->mac_len); 5165 if (skb_do_redirect(skb) == -EAGAIN) { 5166 __skb_pull(skb, skb->mac_len); 5167 *another = true; 5168 break; 5169 } 5170 return NULL; 5171 case TC_ACT_CONSUMED: 5172 return NULL; 5173 default: 5174 break; 5175 } 5176 #endif /* CONFIG_NET_CLS_ACT */ 5177 return skb; 5178 } 5179 5180 /** 5181 * netdev_is_rx_handler_busy - check if receive handler is registered 5182 * @dev: device to check 5183 * 5184 * Check if a receive handler is already registered for a given device. 5185 * Return true if there one. 5186 * 5187 * The caller must hold the rtnl_mutex. 5188 */ 5189 bool netdev_is_rx_handler_busy(struct net_device *dev) 5190 { 5191 ASSERT_RTNL(); 5192 return dev && rtnl_dereference(dev->rx_handler); 5193 } 5194 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); 5195 5196 /** 5197 * netdev_rx_handler_register - register receive handler 5198 * @dev: device to register a handler for 5199 * @rx_handler: receive handler to register 5200 * @rx_handler_data: data pointer that is used by rx handler 5201 * 5202 * Register a receive handler for a device. This handler will then be 5203 * called from __netif_receive_skb. A negative errno code is returned 5204 * on a failure. 5205 * 5206 * The caller must hold the rtnl_mutex. 5207 * 5208 * For a general description of rx_handler, see enum rx_handler_result. 5209 */ 5210 int netdev_rx_handler_register(struct net_device *dev, 5211 rx_handler_func_t *rx_handler, 5212 void *rx_handler_data) 5213 { 5214 if (netdev_is_rx_handler_busy(dev)) 5215 return -EBUSY; 5216 5217 if (dev->priv_flags & IFF_NO_RX_HANDLER) 5218 return -EINVAL; 5219 5220 /* Note: rx_handler_data must be set before rx_handler */ 5221 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); 5222 rcu_assign_pointer(dev->rx_handler, rx_handler); 5223 5224 return 0; 5225 } 5226 EXPORT_SYMBOL_GPL(netdev_rx_handler_register); 5227 5228 /** 5229 * netdev_rx_handler_unregister - unregister receive handler 5230 * @dev: device to unregister a handler from 5231 * 5232 * Unregister a receive handler from a device. 5233 * 5234 * The caller must hold the rtnl_mutex. 5235 */ 5236 void netdev_rx_handler_unregister(struct net_device *dev) 5237 { 5238 5239 ASSERT_RTNL(); 5240 RCU_INIT_POINTER(dev->rx_handler, NULL); 5241 /* a reader seeing a non NULL rx_handler in a rcu_read_lock() 5242 * section has a guarantee to see a non NULL rx_handler_data 5243 * as well. 5244 */ 5245 synchronize_net(); 5246 RCU_INIT_POINTER(dev->rx_handler_data, NULL); 5247 } 5248 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); 5249 5250 /* 5251 * Limit the use of PFMEMALLOC reserves to those protocols that implement 5252 * the special handling of PFMEMALLOC skbs. 5253 */ 5254 static bool skb_pfmemalloc_protocol(struct sk_buff *skb) 5255 { 5256 switch (skb->protocol) { 5257 case htons(ETH_P_ARP): 5258 case htons(ETH_P_IP): 5259 case htons(ETH_P_IPV6): 5260 case htons(ETH_P_8021Q): 5261 case htons(ETH_P_8021AD): 5262 return true; 5263 default: 5264 return false; 5265 } 5266 } 5267 5268 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, 5269 int *ret, struct net_device *orig_dev) 5270 { 5271 if (nf_hook_ingress_active(skb)) { 5272 int ingress_retval; 5273 5274 if (*pt_prev) { 5275 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5276 *pt_prev = NULL; 5277 } 5278 5279 rcu_read_lock(); 5280 ingress_retval = nf_hook_ingress(skb); 5281 rcu_read_unlock(); 5282 return ingress_retval; 5283 } 5284 return 0; 5285 } 5286 5287 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc, 5288 struct packet_type **ppt_prev) 5289 { 5290 struct packet_type *ptype, *pt_prev; 5291 rx_handler_func_t *rx_handler; 5292 struct sk_buff *skb = *pskb; 5293 struct net_device *orig_dev; 5294 bool deliver_exact = false; 5295 int ret = NET_RX_DROP; 5296 __be16 type; 5297 5298 net_timestamp_check(!netdev_tstamp_prequeue, skb); 5299 5300 trace_netif_receive_skb(skb); 5301 5302 orig_dev = skb->dev; 5303 5304 skb_reset_network_header(skb); 5305 if (!skb_transport_header_was_set(skb)) 5306 skb_reset_transport_header(skb); 5307 skb_reset_mac_len(skb); 5308 5309 pt_prev = NULL; 5310 5311 another_round: 5312 skb->skb_iif = skb->dev->ifindex; 5313 5314 __this_cpu_inc(softnet_data.processed); 5315 5316 if (static_branch_unlikely(&generic_xdp_needed_key)) { 5317 int ret2; 5318 5319 migrate_disable(); 5320 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb); 5321 migrate_enable(); 5322 5323 if (ret2 != XDP_PASS) { 5324 ret = NET_RX_DROP; 5325 goto out; 5326 } 5327 skb_reset_mac_len(skb); 5328 } 5329 5330 if (eth_type_vlan(skb->protocol)) { 5331 skb = skb_vlan_untag(skb); 5332 if (unlikely(!skb)) 5333 goto out; 5334 } 5335 5336 if (skb_skip_tc_classify(skb)) 5337 goto skip_classify; 5338 5339 if (pfmemalloc) 5340 goto skip_taps; 5341 5342 list_for_each_entry_rcu(ptype, &ptype_all, list) { 5343 if (pt_prev) 5344 ret = deliver_skb(skb, pt_prev, orig_dev); 5345 pt_prev = ptype; 5346 } 5347 5348 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { 5349 if (pt_prev) 5350 ret = deliver_skb(skb, pt_prev, orig_dev); 5351 pt_prev = ptype; 5352 } 5353 5354 skip_taps: 5355 #ifdef CONFIG_NET_INGRESS 5356 if (static_branch_unlikely(&ingress_needed_key)) { 5357 bool another = false; 5358 5359 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev, 5360 &another); 5361 if (another) 5362 goto another_round; 5363 if (!skb) 5364 goto out; 5365 5366 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) 5367 goto out; 5368 } 5369 #endif 5370 skb_reset_redirect(skb); 5371 skip_classify: 5372 if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) 5373 goto drop; 5374 5375 if (skb_vlan_tag_present(skb)) { 5376 if (pt_prev) { 5377 ret = deliver_skb(skb, pt_prev, orig_dev); 5378 pt_prev = NULL; 5379 } 5380 if (vlan_do_receive(&skb)) 5381 goto another_round; 5382 else if (unlikely(!skb)) 5383 goto out; 5384 } 5385 5386 rx_handler = rcu_dereference(skb->dev->rx_handler); 5387 if (rx_handler) { 5388 if (pt_prev) { 5389 ret = deliver_skb(skb, pt_prev, orig_dev); 5390 pt_prev = NULL; 5391 } 5392 switch (rx_handler(&skb)) { 5393 case RX_HANDLER_CONSUMED: 5394 ret = NET_RX_SUCCESS; 5395 goto out; 5396 case RX_HANDLER_ANOTHER: 5397 goto another_round; 5398 case RX_HANDLER_EXACT: 5399 deliver_exact = true; 5400 break; 5401 case RX_HANDLER_PASS: 5402 break; 5403 default: 5404 BUG(); 5405 } 5406 } 5407 5408 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) { 5409 check_vlan_id: 5410 if (skb_vlan_tag_get_id(skb)) { 5411 /* Vlan id is non 0 and vlan_do_receive() above couldn't 5412 * find vlan device. 5413 */ 5414 skb->pkt_type = PACKET_OTHERHOST; 5415 } else if (eth_type_vlan(skb->protocol)) { 5416 /* Outer header is 802.1P with vlan 0, inner header is 5417 * 802.1Q or 802.1AD and vlan_do_receive() above could 5418 * not find vlan dev for vlan id 0. 5419 */ 5420 __vlan_hwaccel_clear_tag(skb); 5421 skb = skb_vlan_untag(skb); 5422 if (unlikely(!skb)) 5423 goto out; 5424 if (vlan_do_receive(&skb)) 5425 /* After stripping off 802.1P header with vlan 0 5426 * vlan dev is found for inner header. 5427 */ 5428 goto another_round; 5429 else if (unlikely(!skb)) 5430 goto out; 5431 else 5432 /* We have stripped outer 802.1P vlan 0 header. 5433 * But could not find vlan dev. 5434 * check again for vlan id to set OTHERHOST. 5435 */ 5436 goto check_vlan_id; 5437 } 5438 /* Note: we might in the future use prio bits 5439 * and set skb->priority like in vlan_do_receive() 5440 * For the time being, just ignore Priority Code Point 5441 */ 5442 __vlan_hwaccel_clear_tag(skb); 5443 } 5444 5445 type = skb->protocol; 5446 5447 /* deliver only exact match when indicated */ 5448 if (likely(!deliver_exact)) { 5449 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5450 &ptype_base[ntohs(type) & 5451 PTYPE_HASH_MASK]); 5452 } 5453 5454 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5455 &orig_dev->ptype_specific); 5456 5457 if (unlikely(skb->dev != orig_dev)) { 5458 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5459 &skb->dev->ptype_specific); 5460 } 5461 5462 if (pt_prev) { 5463 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 5464 goto drop; 5465 *ppt_prev = pt_prev; 5466 } else { 5467 drop: 5468 if (!deliver_exact) 5469 atomic_long_inc(&skb->dev->rx_dropped); 5470 else 5471 atomic_long_inc(&skb->dev->rx_nohandler); 5472 kfree_skb(skb); 5473 /* Jamal, now you will not able to escape explaining 5474 * me how you were going to use this. :-) 5475 */ 5476 ret = NET_RX_DROP; 5477 } 5478 5479 out: 5480 /* The invariant here is that if *ppt_prev is not NULL 5481 * then skb should also be non-NULL. 5482 * 5483 * Apparently *ppt_prev assignment above holds this invariant due to 5484 * skb dereferencing near it. 5485 */ 5486 *pskb = skb; 5487 return ret; 5488 } 5489 5490 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc) 5491 { 5492 struct net_device *orig_dev = skb->dev; 5493 struct packet_type *pt_prev = NULL; 5494 int ret; 5495 5496 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5497 if (pt_prev) 5498 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb, 5499 skb->dev, pt_prev, orig_dev); 5500 return ret; 5501 } 5502 5503 /** 5504 * netif_receive_skb_core - special purpose version of netif_receive_skb 5505 * @skb: buffer to process 5506 * 5507 * More direct receive version of netif_receive_skb(). It should 5508 * only be used by callers that have a need to skip RPS and Generic XDP. 5509 * Caller must also take care of handling if ``(page_is_)pfmemalloc``. 5510 * 5511 * This function may only be called from softirq context and interrupts 5512 * should be enabled. 5513 * 5514 * Return values (usually ignored): 5515 * NET_RX_SUCCESS: no congestion 5516 * NET_RX_DROP: packet was dropped 5517 */ 5518 int netif_receive_skb_core(struct sk_buff *skb) 5519 { 5520 int ret; 5521 5522 rcu_read_lock(); 5523 ret = __netif_receive_skb_one_core(skb, false); 5524 rcu_read_unlock(); 5525 5526 return ret; 5527 } 5528 EXPORT_SYMBOL(netif_receive_skb_core); 5529 5530 static inline void __netif_receive_skb_list_ptype(struct list_head *head, 5531 struct packet_type *pt_prev, 5532 struct net_device *orig_dev) 5533 { 5534 struct sk_buff *skb, *next; 5535 5536 if (!pt_prev) 5537 return; 5538 if (list_empty(head)) 5539 return; 5540 if (pt_prev->list_func != NULL) 5541 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv, 5542 ip_list_rcv, head, pt_prev, orig_dev); 5543 else 5544 list_for_each_entry_safe(skb, next, head, list) { 5545 skb_list_del_init(skb); 5546 pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 5547 } 5548 } 5549 5550 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc) 5551 { 5552 /* Fast-path assumptions: 5553 * - There is no RX handler. 5554 * - Only one packet_type matches. 5555 * If either of these fails, we will end up doing some per-packet 5556 * processing in-line, then handling the 'last ptype' for the whole 5557 * sublist. This can't cause out-of-order delivery to any single ptype, 5558 * because the 'last ptype' must be constant across the sublist, and all 5559 * other ptypes are handled per-packet. 5560 */ 5561 /* Current (common) ptype of sublist */ 5562 struct packet_type *pt_curr = NULL; 5563 /* Current (common) orig_dev of sublist */ 5564 struct net_device *od_curr = NULL; 5565 struct list_head sublist; 5566 struct sk_buff *skb, *next; 5567 5568 INIT_LIST_HEAD(&sublist); 5569 list_for_each_entry_safe(skb, next, head, list) { 5570 struct net_device *orig_dev = skb->dev; 5571 struct packet_type *pt_prev = NULL; 5572 5573 skb_list_del_init(skb); 5574 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5575 if (!pt_prev) 5576 continue; 5577 if (pt_curr != pt_prev || od_curr != orig_dev) { 5578 /* dispatch old sublist */ 5579 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5580 /* start new sublist */ 5581 INIT_LIST_HEAD(&sublist); 5582 pt_curr = pt_prev; 5583 od_curr = orig_dev; 5584 } 5585 list_add_tail(&skb->list, &sublist); 5586 } 5587 5588 /* dispatch final sublist */ 5589 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5590 } 5591 5592 static int __netif_receive_skb(struct sk_buff *skb) 5593 { 5594 int ret; 5595 5596 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { 5597 unsigned int noreclaim_flag; 5598 5599 /* 5600 * PFMEMALLOC skbs are special, they should 5601 * - be delivered to SOCK_MEMALLOC sockets only 5602 * - stay away from userspace 5603 * - have bounded memory usage 5604 * 5605 * Use PF_MEMALLOC as this saves us from propagating the allocation 5606 * context down to all allocation sites. 5607 */ 5608 noreclaim_flag = memalloc_noreclaim_save(); 5609 ret = __netif_receive_skb_one_core(skb, true); 5610 memalloc_noreclaim_restore(noreclaim_flag); 5611 } else 5612 ret = __netif_receive_skb_one_core(skb, false); 5613 5614 return ret; 5615 } 5616 5617 static void __netif_receive_skb_list(struct list_head *head) 5618 { 5619 unsigned long noreclaim_flag = 0; 5620 struct sk_buff *skb, *next; 5621 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */ 5622 5623 list_for_each_entry_safe(skb, next, head, list) { 5624 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) { 5625 struct list_head sublist; 5626 5627 /* Handle the previous sublist */ 5628 list_cut_before(&sublist, head, &skb->list); 5629 if (!list_empty(&sublist)) 5630 __netif_receive_skb_list_core(&sublist, pfmemalloc); 5631 pfmemalloc = !pfmemalloc; 5632 /* See comments in __netif_receive_skb */ 5633 if (pfmemalloc) 5634 noreclaim_flag = memalloc_noreclaim_save(); 5635 else 5636 memalloc_noreclaim_restore(noreclaim_flag); 5637 } 5638 } 5639 /* Handle the remaining sublist */ 5640 if (!list_empty(head)) 5641 __netif_receive_skb_list_core(head, pfmemalloc); 5642 /* Restore pflags */ 5643 if (pfmemalloc) 5644 memalloc_noreclaim_restore(noreclaim_flag); 5645 } 5646 5647 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp) 5648 { 5649 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog); 5650 struct bpf_prog *new = xdp->prog; 5651 int ret = 0; 5652 5653 if (new) { 5654 u32 i; 5655 5656 mutex_lock(&new->aux->used_maps_mutex); 5657 5658 /* generic XDP does not work with DEVMAPs that can 5659 * have a bpf_prog installed on an entry 5660 */ 5661 for (i = 0; i < new->aux->used_map_cnt; i++) { 5662 if (dev_map_can_have_prog(new->aux->used_maps[i]) || 5663 cpu_map_prog_allowed(new->aux->used_maps[i])) { 5664 mutex_unlock(&new->aux->used_maps_mutex); 5665 return -EINVAL; 5666 } 5667 } 5668 5669 mutex_unlock(&new->aux->used_maps_mutex); 5670 } 5671 5672 switch (xdp->command) { 5673 case XDP_SETUP_PROG: 5674 rcu_assign_pointer(dev->xdp_prog, new); 5675 if (old) 5676 bpf_prog_put(old); 5677 5678 if (old && !new) { 5679 static_branch_dec(&generic_xdp_needed_key); 5680 } else if (new && !old) { 5681 static_branch_inc(&generic_xdp_needed_key); 5682 dev_disable_lro(dev); 5683 dev_disable_gro_hw(dev); 5684 } 5685 break; 5686 5687 default: 5688 ret = -EINVAL; 5689 break; 5690 } 5691 5692 return ret; 5693 } 5694 5695 static int netif_receive_skb_internal(struct sk_buff *skb) 5696 { 5697 int ret; 5698 5699 net_timestamp_check(netdev_tstamp_prequeue, skb); 5700 5701 if (skb_defer_rx_timestamp(skb)) 5702 return NET_RX_SUCCESS; 5703 5704 rcu_read_lock(); 5705 #ifdef CONFIG_RPS 5706 if (static_branch_unlikely(&rps_needed)) { 5707 struct rps_dev_flow voidflow, *rflow = &voidflow; 5708 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5709 5710 if (cpu >= 0) { 5711 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5712 rcu_read_unlock(); 5713 return ret; 5714 } 5715 } 5716 #endif 5717 ret = __netif_receive_skb(skb); 5718 rcu_read_unlock(); 5719 return ret; 5720 } 5721 5722 static void netif_receive_skb_list_internal(struct list_head *head) 5723 { 5724 struct sk_buff *skb, *next; 5725 struct list_head sublist; 5726 5727 INIT_LIST_HEAD(&sublist); 5728 list_for_each_entry_safe(skb, next, head, list) { 5729 net_timestamp_check(netdev_tstamp_prequeue, skb); 5730 skb_list_del_init(skb); 5731 if (!skb_defer_rx_timestamp(skb)) 5732 list_add_tail(&skb->list, &sublist); 5733 } 5734 list_splice_init(&sublist, head); 5735 5736 rcu_read_lock(); 5737 #ifdef CONFIG_RPS 5738 if (static_branch_unlikely(&rps_needed)) { 5739 list_for_each_entry_safe(skb, next, head, list) { 5740 struct rps_dev_flow voidflow, *rflow = &voidflow; 5741 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5742 5743 if (cpu >= 0) { 5744 /* Will be handled, remove from list */ 5745 skb_list_del_init(skb); 5746 enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5747 } 5748 } 5749 } 5750 #endif 5751 __netif_receive_skb_list(head); 5752 rcu_read_unlock(); 5753 } 5754 5755 /** 5756 * netif_receive_skb - process receive buffer from network 5757 * @skb: buffer to process 5758 * 5759 * netif_receive_skb() is the main receive data processing function. 5760 * It always succeeds. The buffer may be dropped during processing 5761 * for congestion control or by the protocol layers. 5762 * 5763 * This function may only be called from softirq context and interrupts 5764 * should be enabled. 5765 * 5766 * Return values (usually ignored): 5767 * NET_RX_SUCCESS: no congestion 5768 * NET_RX_DROP: packet was dropped 5769 */ 5770 int netif_receive_skb(struct sk_buff *skb) 5771 { 5772 int ret; 5773 5774 trace_netif_receive_skb_entry(skb); 5775 5776 ret = netif_receive_skb_internal(skb); 5777 trace_netif_receive_skb_exit(ret); 5778 5779 return ret; 5780 } 5781 EXPORT_SYMBOL(netif_receive_skb); 5782 5783 /** 5784 * netif_receive_skb_list - process many receive buffers from network 5785 * @head: list of skbs to process. 5786 * 5787 * Since return value of netif_receive_skb() is normally ignored, and 5788 * wouldn't be meaningful for a list, this function returns void. 5789 * 5790 * This function may only be called from softirq context and interrupts 5791 * should be enabled. 5792 */ 5793 void netif_receive_skb_list(struct list_head *head) 5794 { 5795 struct sk_buff *skb; 5796 5797 if (list_empty(head)) 5798 return; 5799 if (trace_netif_receive_skb_list_entry_enabled()) { 5800 list_for_each_entry(skb, head, list) 5801 trace_netif_receive_skb_list_entry(skb); 5802 } 5803 netif_receive_skb_list_internal(head); 5804 trace_netif_receive_skb_list_exit(0); 5805 } 5806 EXPORT_SYMBOL(netif_receive_skb_list); 5807 5808 static DEFINE_PER_CPU(struct work_struct, flush_works); 5809 5810 /* Network device is going away, flush any packets still pending */ 5811 static void flush_backlog(struct work_struct *work) 5812 { 5813 struct sk_buff *skb, *tmp; 5814 struct softnet_data *sd; 5815 5816 local_bh_disable(); 5817 sd = this_cpu_ptr(&softnet_data); 5818 5819 local_irq_disable(); 5820 rps_lock(sd); 5821 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { 5822 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5823 __skb_unlink(skb, &sd->input_pkt_queue); 5824 dev_kfree_skb_irq(skb); 5825 input_queue_head_incr(sd); 5826 } 5827 } 5828 rps_unlock(sd); 5829 local_irq_enable(); 5830 5831 skb_queue_walk_safe(&sd->process_queue, skb, tmp) { 5832 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5833 __skb_unlink(skb, &sd->process_queue); 5834 kfree_skb(skb); 5835 input_queue_head_incr(sd); 5836 } 5837 } 5838 local_bh_enable(); 5839 } 5840 5841 static bool flush_required(int cpu) 5842 { 5843 #if IS_ENABLED(CONFIG_RPS) 5844 struct softnet_data *sd = &per_cpu(softnet_data, cpu); 5845 bool do_flush; 5846 5847 local_irq_disable(); 5848 rps_lock(sd); 5849 5850 /* as insertion into process_queue happens with the rps lock held, 5851 * process_queue access may race only with dequeue 5852 */ 5853 do_flush = !skb_queue_empty(&sd->input_pkt_queue) || 5854 !skb_queue_empty_lockless(&sd->process_queue); 5855 rps_unlock(sd); 5856 local_irq_enable(); 5857 5858 return do_flush; 5859 #endif 5860 /* without RPS we can't safely check input_pkt_queue: during a 5861 * concurrent remote skb_queue_splice() we can detect as empty both 5862 * input_pkt_queue and process_queue even if the latter could end-up 5863 * containing a lot of packets. 5864 */ 5865 return true; 5866 } 5867 5868 static void flush_all_backlogs(void) 5869 { 5870 static cpumask_t flush_cpus; 5871 unsigned int cpu; 5872 5873 /* since we are under rtnl lock protection we can use static data 5874 * for the cpumask and avoid allocating on stack the possibly 5875 * large mask 5876 */ 5877 ASSERT_RTNL(); 5878 5879 get_online_cpus(); 5880 5881 cpumask_clear(&flush_cpus); 5882 for_each_online_cpu(cpu) { 5883 if (flush_required(cpu)) { 5884 queue_work_on(cpu, system_highpri_wq, 5885 per_cpu_ptr(&flush_works, cpu)); 5886 cpumask_set_cpu(cpu, &flush_cpus); 5887 } 5888 } 5889 5890 /* we can have in flight packet[s] on the cpus we are not flushing, 5891 * synchronize_net() in unregister_netdevice_many() will take care of 5892 * them 5893 */ 5894 for_each_cpu(cpu, &flush_cpus) 5895 flush_work(per_cpu_ptr(&flush_works, cpu)); 5896 5897 put_online_cpus(); 5898 } 5899 5900 /* Pass the currently batched GRO_NORMAL SKBs up to the stack. */ 5901 static void gro_normal_list(struct napi_struct *napi) 5902 { 5903 if (!napi->rx_count) 5904 return; 5905 netif_receive_skb_list_internal(&napi->rx_list); 5906 INIT_LIST_HEAD(&napi->rx_list); 5907 napi->rx_count = 0; 5908 } 5909 5910 /* Queue one GRO_NORMAL SKB up for list processing. If batch size exceeded, 5911 * pass the whole batch up to the stack. 5912 */ 5913 static void gro_normal_one(struct napi_struct *napi, struct sk_buff *skb, int segs) 5914 { 5915 list_add_tail(&skb->list, &napi->rx_list); 5916 napi->rx_count += segs; 5917 if (napi->rx_count >= gro_normal_batch) 5918 gro_normal_list(napi); 5919 } 5920 5921 static int napi_gro_complete(struct napi_struct *napi, struct sk_buff *skb) 5922 { 5923 struct packet_offload *ptype; 5924 __be16 type = skb->protocol; 5925 struct list_head *head = &offload_base; 5926 int err = -ENOENT; 5927 5928 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb)); 5929 5930 if (NAPI_GRO_CB(skb)->count == 1) { 5931 skb_shinfo(skb)->gso_size = 0; 5932 goto out; 5933 } 5934 5935 rcu_read_lock(); 5936 list_for_each_entry_rcu(ptype, head, list) { 5937 if (ptype->type != type || !ptype->callbacks.gro_complete) 5938 continue; 5939 5940 err = INDIRECT_CALL_INET(ptype->callbacks.gro_complete, 5941 ipv6_gro_complete, inet_gro_complete, 5942 skb, 0); 5943 break; 5944 } 5945 rcu_read_unlock(); 5946 5947 if (err) { 5948 WARN_ON(&ptype->list == head); 5949 kfree_skb(skb); 5950 return NET_RX_SUCCESS; 5951 } 5952 5953 out: 5954 gro_normal_one(napi, skb, NAPI_GRO_CB(skb)->count); 5955 return NET_RX_SUCCESS; 5956 } 5957 5958 static void __napi_gro_flush_chain(struct napi_struct *napi, u32 index, 5959 bool flush_old) 5960 { 5961 struct list_head *head = &napi->gro_hash[index].list; 5962 struct sk_buff *skb, *p; 5963 5964 list_for_each_entry_safe_reverse(skb, p, head, list) { 5965 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies) 5966 return; 5967 skb_list_del_init(skb); 5968 napi_gro_complete(napi, skb); 5969 napi->gro_hash[index].count--; 5970 } 5971 5972 if (!napi->gro_hash[index].count) 5973 __clear_bit(index, &napi->gro_bitmask); 5974 } 5975 5976 /* napi->gro_hash[].list contains packets ordered by age. 5977 * youngest packets at the head of it. 5978 * Complete skbs in reverse order to reduce latencies. 5979 */ 5980 void napi_gro_flush(struct napi_struct *napi, bool flush_old) 5981 { 5982 unsigned long bitmask = napi->gro_bitmask; 5983 unsigned int i, base = ~0U; 5984 5985 while ((i = ffs(bitmask)) != 0) { 5986 bitmask >>= i; 5987 base += i; 5988 __napi_gro_flush_chain(napi, base, flush_old); 5989 } 5990 } 5991 EXPORT_SYMBOL(napi_gro_flush); 5992 5993 static void gro_list_prepare(const struct list_head *head, 5994 const struct sk_buff *skb) 5995 { 5996 unsigned int maclen = skb->dev->hard_header_len; 5997 u32 hash = skb_get_hash_raw(skb); 5998 struct sk_buff *p; 5999 6000 list_for_each_entry(p, head, list) { 6001 unsigned long diffs; 6002 6003 NAPI_GRO_CB(p)->flush = 0; 6004 6005 if (hash != skb_get_hash_raw(p)) { 6006 NAPI_GRO_CB(p)->same_flow = 0; 6007 continue; 6008 } 6009 6010 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev; 6011 diffs |= skb_vlan_tag_present(p) ^ skb_vlan_tag_present(skb); 6012 if (skb_vlan_tag_present(p)) 6013 diffs |= skb_vlan_tag_get(p) ^ skb_vlan_tag_get(skb); 6014 diffs |= skb_metadata_dst_cmp(p, skb); 6015 diffs |= skb_metadata_differs(p, skb); 6016 if (maclen == ETH_HLEN) 6017 diffs |= compare_ether_header(skb_mac_header(p), 6018 skb_mac_header(skb)); 6019 else if (!diffs) 6020 diffs = memcmp(skb_mac_header(p), 6021 skb_mac_header(skb), 6022 maclen); 6023 6024 diffs |= skb_get_nfct(p) ^ skb_get_nfct(skb); 6025 #if IS_ENABLED(CONFIG_SKB_EXTENSIONS) && IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 6026 if (!diffs) { 6027 struct tc_skb_ext *skb_ext = skb_ext_find(skb, TC_SKB_EXT); 6028 struct tc_skb_ext *p_ext = skb_ext_find(p, TC_SKB_EXT); 6029 6030 diffs |= (!!p_ext) ^ (!!skb_ext); 6031 if (!diffs && unlikely(skb_ext)) 6032 diffs |= p_ext->chain ^ skb_ext->chain; 6033 } 6034 #endif 6035 6036 NAPI_GRO_CB(p)->same_flow = !diffs; 6037 } 6038 } 6039 6040 static inline void skb_gro_reset_offset(struct sk_buff *skb, u32 nhoff) 6041 { 6042 const struct skb_shared_info *pinfo = skb_shinfo(skb); 6043 const skb_frag_t *frag0 = &pinfo->frags[0]; 6044 6045 NAPI_GRO_CB(skb)->data_offset = 0; 6046 NAPI_GRO_CB(skb)->frag0 = NULL; 6047 NAPI_GRO_CB(skb)->frag0_len = 0; 6048 6049 if (!skb_headlen(skb) && pinfo->nr_frags && 6050 !PageHighMem(skb_frag_page(frag0)) && 6051 (!NET_IP_ALIGN || !((skb_frag_off(frag0) + nhoff) & 3))) { 6052 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0); 6053 NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int, 6054 skb_frag_size(frag0), 6055 skb->end - skb->tail); 6056 } 6057 } 6058 6059 static void gro_pull_from_frag0(struct sk_buff *skb, int grow) 6060 { 6061 struct skb_shared_info *pinfo = skb_shinfo(skb); 6062 6063 BUG_ON(skb->end - skb->tail < grow); 6064 6065 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); 6066 6067 skb->data_len -= grow; 6068 skb->tail += grow; 6069 6070 skb_frag_off_add(&pinfo->frags[0], grow); 6071 skb_frag_size_sub(&pinfo->frags[0], grow); 6072 6073 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) { 6074 skb_frag_unref(skb, 0); 6075 memmove(pinfo->frags, pinfo->frags + 1, 6076 --pinfo->nr_frags * sizeof(pinfo->frags[0])); 6077 } 6078 } 6079 6080 static void gro_flush_oldest(struct napi_struct *napi, struct list_head *head) 6081 { 6082 struct sk_buff *oldest; 6083 6084 oldest = list_last_entry(head, struct sk_buff, list); 6085 6086 /* We are called with head length >= MAX_GRO_SKBS, so this is 6087 * impossible. 6088 */ 6089 if (WARN_ON_ONCE(!oldest)) 6090 return; 6091 6092 /* Do not adjust napi->gro_hash[].count, caller is adding a new 6093 * SKB to the chain. 6094 */ 6095 skb_list_del_init(oldest); 6096 napi_gro_complete(napi, oldest); 6097 } 6098 6099 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) 6100 { 6101 u32 bucket = skb_get_hash_raw(skb) & (GRO_HASH_BUCKETS - 1); 6102 struct gro_list *gro_list = &napi->gro_hash[bucket]; 6103 struct list_head *head = &offload_base; 6104 struct packet_offload *ptype; 6105 __be16 type = skb->protocol; 6106 struct sk_buff *pp = NULL; 6107 enum gro_result ret; 6108 int same_flow; 6109 int grow; 6110 6111 if (netif_elide_gro(skb->dev)) 6112 goto normal; 6113 6114 gro_list_prepare(&gro_list->list, skb); 6115 6116 rcu_read_lock(); 6117 list_for_each_entry_rcu(ptype, head, list) { 6118 if (ptype->type != type || !ptype->callbacks.gro_receive) 6119 continue; 6120 6121 skb_set_network_header(skb, skb_gro_offset(skb)); 6122 skb_reset_mac_len(skb); 6123 NAPI_GRO_CB(skb)->same_flow = 0; 6124 NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb); 6125 NAPI_GRO_CB(skb)->free = 0; 6126 NAPI_GRO_CB(skb)->encap_mark = 0; 6127 NAPI_GRO_CB(skb)->recursion_counter = 0; 6128 NAPI_GRO_CB(skb)->is_fou = 0; 6129 NAPI_GRO_CB(skb)->is_atomic = 1; 6130 NAPI_GRO_CB(skb)->gro_remcsum_start = 0; 6131 6132 /* Setup for GRO checksum validation */ 6133 switch (skb->ip_summed) { 6134 case CHECKSUM_COMPLETE: 6135 NAPI_GRO_CB(skb)->csum = skb->csum; 6136 NAPI_GRO_CB(skb)->csum_valid = 1; 6137 NAPI_GRO_CB(skb)->csum_cnt = 0; 6138 break; 6139 case CHECKSUM_UNNECESSARY: 6140 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1; 6141 NAPI_GRO_CB(skb)->csum_valid = 0; 6142 break; 6143 default: 6144 NAPI_GRO_CB(skb)->csum_cnt = 0; 6145 NAPI_GRO_CB(skb)->csum_valid = 0; 6146 } 6147 6148 pp = INDIRECT_CALL_INET(ptype->callbacks.gro_receive, 6149 ipv6_gro_receive, inet_gro_receive, 6150 &gro_list->list, skb); 6151 break; 6152 } 6153 rcu_read_unlock(); 6154 6155 if (&ptype->list == head) 6156 goto normal; 6157 6158 if (PTR_ERR(pp) == -EINPROGRESS) { 6159 ret = GRO_CONSUMED; 6160 goto ok; 6161 } 6162 6163 same_flow = NAPI_GRO_CB(skb)->same_flow; 6164 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; 6165 6166 if (pp) { 6167 skb_list_del_init(pp); 6168 napi_gro_complete(napi, pp); 6169 gro_list->count--; 6170 } 6171 6172 if (same_flow) 6173 goto ok; 6174 6175 if (NAPI_GRO_CB(skb)->flush) 6176 goto normal; 6177 6178 if (unlikely(gro_list->count >= MAX_GRO_SKBS)) 6179 gro_flush_oldest(napi, &gro_list->list); 6180 else 6181 gro_list->count++; 6182 6183 NAPI_GRO_CB(skb)->count = 1; 6184 NAPI_GRO_CB(skb)->age = jiffies; 6185 NAPI_GRO_CB(skb)->last = skb; 6186 skb_shinfo(skb)->gso_size = skb_gro_len(skb); 6187 list_add(&skb->list, &gro_list->list); 6188 ret = GRO_HELD; 6189 6190 pull: 6191 grow = skb_gro_offset(skb) - skb_headlen(skb); 6192 if (grow > 0) 6193 gro_pull_from_frag0(skb, grow); 6194 ok: 6195 if (gro_list->count) { 6196 if (!test_bit(bucket, &napi->gro_bitmask)) 6197 __set_bit(bucket, &napi->gro_bitmask); 6198 } else if (test_bit(bucket, &napi->gro_bitmask)) { 6199 __clear_bit(bucket, &napi->gro_bitmask); 6200 } 6201 6202 return ret; 6203 6204 normal: 6205 ret = GRO_NORMAL; 6206 goto pull; 6207 } 6208 6209 struct packet_offload *gro_find_receive_by_type(__be16 type) 6210 { 6211 struct list_head *offload_head = &offload_base; 6212 struct packet_offload *ptype; 6213 6214 list_for_each_entry_rcu(ptype, offload_head, list) { 6215 if (ptype->type != type || !ptype->callbacks.gro_receive) 6216 continue; 6217 return ptype; 6218 } 6219 return NULL; 6220 } 6221 EXPORT_SYMBOL(gro_find_receive_by_type); 6222 6223 struct packet_offload *gro_find_complete_by_type(__be16 type) 6224 { 6225 struct list_head *offload_head = &offload_base; 6226 struct packet_offload *ptype; 6227 6228 list_for_each_entry_rcu(ptype, offload_head, list) { 6229 if (ptype->type != type || !ptype->callbacks.gro_complete) 6230 continue; 6231 return ptype; 6232 } 6233 return NULL; 6234 } 6235 EXPORT_SYMBOL(gro_find_complete_by_type); 6236 6237 static gro_result_t napi_skb_finish(struct napi_struct *napi, 6238 struct sk_buff *skb, 6239 gro_result_t ret) 6240 { 6241 switch (ret) { 6242 case GRO_NORMAL: 6243 gro_normal_one(napi, skb, 1); 6244 break; 6245 6246 case GRO_MERGED_FREE: 6247 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) 6248 napi_skb_free_stolen_head(skb); 6249 else if (skb->fclone != SKB_FCLONE_UNAVAILABLE) 6250 __kfree_skb(skb); 6251 else 6252 __kfree_skb_defer(skb); 6253 break; 6254 6255 case GRO_HELD: 6256 case GRO_MERGED: 6257 case GRO_CONSUMED: 6258 break; 6259 } 6260 6261 return ret; 6262 } 6263 6264 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) 6265 { 6266 gro_result_t ret; 6267 6268 skb_mark_napi_id(skb, napi); 6269 trace_napi_gro_receive_entry(skb); 6270 6271 skb_gro_reset_offset(skb, 0); 6272 6273 ret = napi_skb_finish(napi, skb, dev_gro_receive(napi, skb)); 6274 trace_napi_gro_receive_exit(ret); 6275 6276 return ret; 6277 } 6278 EXPORT_SYMBOL(napi_gro_receive); 6279 6280 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) 6281 { 6282 if (unlikely(skb->pfmemalloc)) { 6283 consume_skb(skb); 6284 return; 6285 } 6286 __skb_pull(skb, skb_headlen(skb)); 6287 /* restore the reserve we had after netdev_alloc_skb_ip_align() */ 6288 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb)); 6289 __vlan_hwaccel_clear_tag(skb); 6290 skb->dev = napi->dev; 6291 skb->skb_iif = 0; 6292 6293 /* eth_type_trans() assumes pkt_type is PACKET_HOST */ 6294 skb->pkt_type = PACKET_HOST; 6295 6296 skb->encapsulation = 0; 6297 skb_shinfo(skb)->gso_type = 0; 6298 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); 6299 skb_ext_reset(skb); 6300 nf_reset_ct(skb); 6301 6302 napi->skb = skb; 6303 } 6304 6305 struct sk_buff *napi_get_frags(struct napi_struct *napi) 6306 { 6307 struct sk_buff *skb = napi->skb; 6308 6309 if (!skb) { 6310 skb = napi_alloc_skb(napi, GRO_MAX_HEAD); 6311 if (skb) { 6312 napi->skb = skb; 6313 skb_mark_napi_id(skb, napi); 6314 } 6315 } 6316 return skb; 6317 } 6318 EXPORT_SYMBOL(napi_get_frags); 6319 6320 static gro_result_t napi_frags_finish(struct napi_struct *napi, 6321 struct sk_buff *skb, 6322 gro_result_t ret) 6323 { 6324 switch (ret) { 6325 case GRO_NORMAL: 6326 case GRO_HELD: 6327 __skb_push(skb, ETH_HLEN); 6328 skb->protocol = eth_type_trans(skb, skb->dev); 6329 if (ret == GRO_NORMAL) 6330 gro_normal_one(napi, skb, 1); 6331 break; 6332 6333 case GRO_MERGED_FREE: 6334 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) 6335 napi_skb_free_stolen_head(skb); 6336 else 6337 napi_reuse_skb(napi, skb); 6338 break; 6339 6340 case GRO_MERGED: 6341 case GRO_CONSUMED: 6342 break; 6343 } 6344 6345 return ret; 6346 } 6347 6348 /* Upper GRO stack assumes network header starts at gro_offset=0 6349 * Drivers could call both napi_gro_frags() and napi_gro_receive() 6350 * We copy ethernet header into skb->data to have a common layout. 6351 */ 6352 static struct sk_buff *napi_frags_skb(struct napi_struct *napi) 6353 { 6354 struct sk_buff *skb = napi->skb; 6355 const struct ethhdr *eth; 6356 unsigned int hlen = sizeof(*eth); 6357 6358 napi->skb = NULL; 6359 6360 skb_reset_mac_header(skb); 6361 skb_gro_reset_offset(skb, hlen); 6362 6363 if (unlikely(skb_gro_header_hard(skb, hlen))) { 6364 eth = skb_gro_header_slow(skb, hlen, 0); 6365 if (unlikely(!eth)) { 6366 net_warn_ratelimited("%s: dropping impossible skb from %s\n", 6367 __func__, napi->dev->name); 6368 napi_reuse_skb(napi, skb); 6369 return NULL; 6370 } 6371 } else { 6372 eth = (const struct ethhdr *)skb->data; 6373 gro_pull_from_frag0(skb, hlen); 6374 NAPI_GRO_CB(skb)->frag0 += hlen; 6375 NAPI_GRO_CB(skb)->frag0_len -= hlen; 6376 } 6377 __skb_pull(skb, hlen); 6378 6379 /* 6380 * This works because the only protocols we care about don't require 6381 * special handling. 6382 * We'll fix it up properly in napi_frags_finish() 6383 */ 6384 skb->protocol = eth->h_proto; 6385 6386 return skb; 6387 } 6388 6389 gro_result_t napi_gro_frags(struct napi_struct *napi) 6390 { 6391 gro_result_t ret; 6392 struct sk_buff *skb = napi_frags_skb(napi); 6393 6394 trace_napi_gro_frags_entry(skb); 6395 6396 ret = napi_frags_finish(napi, skb, dev_gro_receive(napi, skb)); 6397 trace_napi_gro_frags_exit(ret); 6398 6399 return ret; 6400 } 6401 EXPORT_SYMBOL(napi_gro_frags); 6402 6403 /* Compute the checksum from gro_offset and return the folded value 6404 * after adding in any pseudo checksum. 6405 */ 6406 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb) 6407 { 6408 __wsum wsum; 6409 __sum16 sum; 6410 6411 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0); 6412 6413 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */ 6414 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum)); 6415 /* See comments in __skb_checksum_complete(). */ 6416 if (likely(!sum)) { 6417 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && 6418 !skb->csum_complete_sw) 6419 netdev_rx_csum_fault(skb->dev, skb); 6420 } 6421 6422 NAPI_GRO_CB(skb)->csum = wsum; 6423 NAPI_GRO_CB(skb)->csum_valid = 1; 6424 6425 return sum; 6426 } 6427 EXPORT_SYMBOL(__skb_gro_checksum_complete); 6428 6429 static void net_rps_send_ipi(struct softnet_data *remsd) 6430 { 6431 #ifdef CONFIG_RPS 6432 while (remsd) { 6433 struct softnet_data *next = remsd->rps_ipi_next; 6434 6435 if (cpu_online(remsd->cpu)) 6436 smp_call_function_single_async(remsd->cpu, &remsd->csd); 6437 remsd = next; 6438 } 6439 #endif 6440 } 6441 6442 /* 6443 * net_rps_action_and_irq_enable sends any pending IPI's for rps. 6444 * Note: called with local irq disabled, but exits with local irq enabled. 6445 */ 6446 static void net_rps_action_and_irq_enable(struct softnet_data *sd) 6447 { 6448 #ifdef CONFIG_RPS 6449 struct softnet_data *remsd = sd->rps_ipi_list; 6450 6451 if (remsd) { 6452 sd->rps_ipi_list = NULL; 6453 6454 local_irq_enable(); 6455 6456 /* Send pending IPI's to kick RPS processing on remote cpus. */ 6457 net_rps_send_ipi(remsd); 6458 } else 6459 #endif 6460 local_irq_enable(); 6461 } 6462 6463 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) 6464 { 6465 #ifdef CONFIG_RPS 6466 return sd->rps_ipi_list != NULL; 6467 #else 6468 return false; 6469 #endif 6470 } 6471 6472 static int process_backlog(struct napi_struct *napi, int quota) 6473 { 6474 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); 6475 bool again = true; 6476 int work = 0; 6477 6478 /* Check if we have pending ipi, its better to send them now, 6479 * not waiting net_rx_action() end. 6480 */ 6481 if (sd_has_rps_ipi_waiting(sd)) { 6482 local_irq_disable(); 6483 net_rps_action_and_irq_enable(sd); 6484 } 6485 6486 napi->weight = dev_rx_weight; 6487 while (again) { 6488 struct sk_buff *skb; 6489 6490 while ((skb = __skb_dequeue(&sd->process_queue))) { 6491 rcu_read_lock(); 6492 __netif_receive_skb(skb); 6493 rcu_read_unlock(); 6494 input_queue_head_incr(sd); 6495 if (++work >= quota) 6496 return work; 6497 6498 } 6499 6500 local_irq_disable(); 6501 rps_lock(sd); 6502 if (skb_queue_empty(&sd->input_pkt_queue)) { 6503 /* 6504 * Inline a custom version of __napi_complete(). 6505 * only current cpu owns and manipulates this napi, 6506 * and NAPI_STATE_SCHED is the only possible flag set 6507 * on backlog. 6508 * We can use a plain write instead of clear_bit(), 6509 * and we dont need an smp_mb() memory barrier. 6510 */ 6511 napi->state = 0; 6512 again = false; 6513 } else { 6514 skb_queue_splice_tail_init(&sd->input_pkt_queue, 6515 &sd->process_queue); 6516 } 6517 rps_unlock(sd); 6518 local_irq_enable(); 6519 } 6520 6521 return work; 6522 } 6523 6524 /** 6525 * __napi_schedule - schedule for receive 6526 * @n: entry to schedule 6527 * 6528 * The entry's receive function will be scheduled to run. 6529 * Consider using __napi_schedule_irqoff() if hard irqs are masked. 6530 */ 6531 void __napi_schedule(struct napi_struct *n) 6532 { 6533 unsigned long flags; 6534 6535 local_irq_save(flags); 6536 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6537 local_irq_restore(flags); 6538 } 6539 EXPORT_SYMBOL(__napi_schedule); 6540 6541 /** 6542 * napi_schedule_prep - check if napi can be scheduled 6543 * @n: napi context 6544 * 6545 * Test if NAPI routine is already running, and if not mark 6546 * it as running. This is used as a condition variable to 6547 * insure only one NAPI poll instance runs. We also make 6548 * sure there is no pending NAPI disable. 6549 */ 6550 bool napi_schedule_prep(struct napi_struct *n) 6551 { 6552 unsigned long val, new; 6553 6554 do { 6555 val = READ_ONCE(n->state); 6556 if (unlikely(val & NAPIF_STATE_DISABLE)) 6557 return false; 6558 new = val | NAPIF_STATE_SCHED; 6559 6560 /* Sets STATE_MISSED bit if STATE_SCHED was already set 6561 * This was suggested by Alexander Duyck, as compiler 6562 * emits better code than : 6563 * if (val & NAPIF_STATE_SCHED) 6564 * new |= NAPIF_STATE_MISSED; 6565 */ 6566 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * 6567 NAPIF_STATE_MISSED; 6568 } while (cmpxchg(&n->state, val, new) != val); 6569 6570 return !(val & NAPIF_STATE_SCHED); 6571 } 6572 EXPORT_SYMBOL(napi_schedule_prep); 6573 6574 /** 6575 * __napi_schedule_irqoff - schedule for receive 6576 * @n: entry to schedule 6577 * 6578 * Variant of __napi_schedule() assuming hard irqs are masked. 6579 * 6580 * On PREEMPT_RT enabled kernels this maps to __napi_schedule() 6581 * because the interrupt disabled assumption might not be true 6582 * due to force-threaded interrupts and spinlock substitution. 6583 */ 6584 void __napi_schedule_irqoff(struct napi_struct *n) 6585 { 6586 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6587 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6588 else 6589 __napi_schedule(n); 6590 } 6591 EXPORT_SYMBOL(__napi_schedule_irqoff); 6592 6593 bool napi_complete_done(struct napi_struct *n, int work_done) 6594 { 6595 unsigned long flags, val, new, timeout = 0; 6596 bool ret = true; 6597 6598 /* 6599 * 1) Don't let napi dequeue from the cpu poll list 6600 * just in case its running on a different cpu. 6601 * 2) If we are busy polling, do nothing here, we have 6602 * the guarantee we will be called later. 6603 */ 6604 if (unlikely(n->state & (NAPIF_STATE_NPSVC | 6605 NAPIF_STATE_IN_BUSY_POLL))) 6606 return false; 6607 6608 if (work_done) { 6609 if (n->gro_bitmask) 6610 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6611 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs); 6612 } 6613 if (n->defer_hard_irqs_count > 0) { 6614 n->defer_hard_irqs_count--; 6615 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6616 if (timeout) 6617 ret = false; 6618 } 6619 if (n->gro_bitmask) { 6620 /* When the NAPI instance uses a timeout and keeps postponing 6621 * it, we need to bound somehow the time packets are kept in 6622 * the GRO layer 6623 */ 6624 napi_gro_flush(n, !!timeout); 6625 } 6626 6627 gro_normal_list(n); 6628 6629 if (unlikely(!list_empty(&n->poll_list))) { 6630 /* If n->poll_list is not empty, we need to mask irqs */ 6631 local_irq_save(flags); 6632 list_del_init(&n->poll_list); 6633 local_irq_restore(flags); 6634 } 6635 6636 do { 6637 val = READ_ONCE(n->state); 6638 6639 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); 6640 6641 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED | 6642 NAPIF_STATE_SCHED_THREADED | 6643 NAPIF_STATE_PREFER_BUSY_POLL); 6644 6645 /* If STATE_MISSED was set, leave STATE_SCHED set, 6646 * because we will call napi->poll() one more time. 6647 * This C code was suggested by Alexander Duyck to help gcc. 6648 */ 6649 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * 6650 NAPIF_STATE_SCHED; 6651 } while (cmpxchg(&n->state, val, new) != val); 6652 6653 if (unlikely(val & NAPIF_STATE_MISSED)) { 6654 __napi_schedule(n); 6655 return false; 6656 } 6657 6658 if (timeout) 6659 hrtimer_start(&n->timer, ns_to_ktime(timeout), 6660 HRTIMER_MODE_REL_PINNED); 6661 return ret; 6662 } 6663 EXPORT_SYMBOL(napi_complete_done); 6664 6665 /* must be called under rcu_read_lock(), as we dont take a reference */ 6666 static struct napi_struct *napi_by_id(unsigned int napi_id) 6667 { 6668 unsigned int hash = napi_id % HASH_SIZE(napi_hash); 6669 struct napi_struct *napi; 6670 6671 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) 6672 if (napi->napi_id == napi_id) 6673 return napi; 6674 6675 return NULL; 6676 } 6677 6678 #if defined(CONFIG_NET_RX_BUSY_POLL) 6679 6680 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule) 6681 { 6682 if (!skip_schedule) { 6683 gro_normal_list(napi); 6684 __napi_schedule(napi); 6685 return; 6686 } 6687 6688 if (napi->gro_bitmask) { 6689 /* flush too old packets 6690 * If HZ < 1000, flush all packets. 6691 */ 6692 napi_gro_flush(napi, HZ >= 1000); 6693 } 6694 6695 gro_normal_list(napi); 6696 clear_bit(NAPI_STATE_SCHED, &napi->state); 6697 } 6698 6699 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll, 6700 u16 budget) 6701 { 6702 bool skip_schedule = false; 6703 unsigned long timeout; 6704 int rc; 6705 6706 /* Busy polling means there is a high chance device driver hard irq 6707 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was 6708 * set in napi_schedule_prep(). 6709 * Since we are about to call napi->poll() once more, we can safely 6710 * clear NAPI_STATE_MISSED. 6711 * 6712 * Note: x86 could use a single "lock and ..." instruction 6713 * to perform these two clear_bit() 6714 */ 6715 clear_bit(NAPI_STATE_MISSED, &napi->state); 6716 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); 6717 6718 local_bh_disable(); 6719 6720 if (prefer_busy_poll) { 6721 napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs); 6722 timeout = READ_ONCE(napi->dev->gro_flush_timeout); 6723 if (napi->defer_hard_irqs_count && timeout) { 6724 hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED); 6725 skip_schedule = true; 6726 } 6727 } 6728 6729 /* All we really want here is to re-enable device interrupts. 6730 * Ideally, a new ndo_busy_poll_stop() could avoid another round. 6731 */ 6732 rc = napi->poll(napi, budget); 6733 /* We can't gro_normal_list() here, because napi->poll() might have 6734 * rearmed the napi (napi_complete_done()) in which case it could 6735 * already be running on another CPU. 6736 */ 6737 trace_napi_poll(napi, rc, budget); 6738 netpoll_poll_unlock(have_poll_lock); 6739 if (rc == budget) 6740 __busy_poll_stop(napi, skip_schedule); 6741 local_bh_enable(); 6742 } 6743 6744 void napi_busy_loop(unsigned int napi_id, 6745 bool (*loop_end)(void *, unsigned long), 6746 void *loop_end_arg, bool prefer_busy_poll, u16 budget) 6747 { 6748 unsigned long start_time = loop_end ? busy_loop_current_time() : 0; 6749 int (*napi_poll)(struct napi_struct *napi, int budget); 6750 void *have_poll_lock = NULL; 6751 struct napi_struct *napi; 6752 6753 restart: 6754 napi_poll = NULL; 6755 6756 rcu_read_lock(); 6757 6758 napi = napi_by_id(napi_id); 6759 if (!napi) 6760 goto out; 6761 6762 preempt_disable(); 6763 for (;;) { 6764 int work = 0; 6765 6766 local_bh_disable(); 6767 if (!napi_poll) { 6768 unsigned long val = READ_ONCE(napi->state); 6769 6770 /* If multiple threads are competing for this napi, 6771 * we avoid dirtying napi->state as much as we can. 6772 */ 6773 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | 6774 NAPIF_STATE_IN_BUSY_POLL)) { 6775 if (prefer_busy_poll) 6776 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6777 goto count; 6778 } 6779 if (cmpxchg(&napi->state, val, 6780 val | NAPIF_STATE_IN_BUSY_POLL | 6781 NAPIF_STATE_SCHED) != val) { 6782 if (prefer_busy_poll) 6783 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6784 goto count; 6785 } 6786 have_poll_lock = netpoll_poll_lock(napi); 6787 napi_poll = napi->poll; 6788 } 6789 work = napi_poll(napi, budget); 6790 trace_napi_poll(napi, work, budget); 6791 gro_normal_list(napi); 6792 count: 6793 if (work > 0) 6794 __NET_ADD_STATS(dev_net(napi->dev), 6795 LINUX_MIB_BUSYPOLLRXPACKETS, work); 6796 local_bh_enable(); 6797 6798 if (!loop_end || loop_end(loop_end_arg, start_time)) 6799 break; 6800 6801 if (unlikely(need_resched())) { 6802 if (napi_poll) 6803 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6804 preempt_enable(); 6805 rcu_read_unlock(); 6806 cond_resched(); 6807 if (loop_end(loop_end_arg, start_time)) 6808 return; 6809 goto restart; 6810 } 6811 cpu_relax(); 6812 } 6813 if (napi_poll) 6814 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6815 preempt_enable(); 6816 out: 6817 rcu_read_unlock(); 6818 } 6819 EXPORT_SYMBOL(napi_busy_loop); 6820 6821 #endif /* CONFIG_NET_RX_BUSY_POLL */ 6822 6823 static void napi_hash_add(struct napi_struct *napi) 6824 { 6825 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state)) 6826 return; 6827 6828 spin_lock(&napi_hash_lock); 6829 6830 /* 0..NR_CPUS range is reserved for sender_cpu use */ 6831 do { 6832 if (unlikely(++napi_gen_id < MIN_NAPI_ID)) 6833 napi_gen_id = MIN_NAPI_ID; 6834 } while (napi_by_id(napi_gen_id)); 6835 napi->napi_id = napi_gen_id; 6836 6837 hlist_add_head_rcu(&napi->napi_hash_node, 6838 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); 6839 6840 spin_unlock(&napi_hash_lock); 6841 } 6842 6843 /* Warning : caller is responsible to make sure rcu grace period 6844 * is respected before freeing memory containing @napi 6845 */ 6846 static void napi_hash_del(struct napi_struct *napi) 6847 { 6848 spin_lock(&napi_hash_lock); 6849 6850 hlist_del_init_rcu(&napi->napi_hash_node); 6851 6852 spin_unlock(&napi_hash_lock); 6853 } 6854 6855 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) 6856 { 6857 struct napi_struct *napi; 6858 6859 napi = container_of(timer, struct napi_struct, timer); 6860 6861 /* Note : we use a relaxed variant of napi_schedule_prep() not setting 6862 * NAPI_STATE_MISSED, since we do not react to a device IRQ. 6863 */ 6864 if (!napi_disable_pending(napi) && 6865 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) { 6866 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6867 __napi_schedule_irqoff(napi); 6868 } 6869 6870 return HRTIMER_NORESTART; 6871 } 6872 6873 static void init_gro_hash(struct napi_struct *napi) 6874 { 6875 int i; 6876 6877 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6878 INIT_LIST_HEAD(&napi->gro_hash[i].list); 6879 napi->gro_hash[i].count = 0; 6880 } 6881 napi->gro_bitmask = 0; 6882 } 6883 6884 int dev_set_threaded(struct net_device *dev, bool threaded) 6885 { 6886 struct napi_struct *napi; 6887 int err = 0; 6888 6889 if (dev->threaded == threaded) 6890 return 0; 6891 6892 if (threaded) { 6893 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6894 if (!napi->thread) { 6895 err = napi_kthread_create(napi); 6896 if (err) { 6897 threaded = false; 6898 break; 6899 } 6900 } 6901 } 6902 } 6903 6904 dev->threaded = threaded; 6905 6906 /* Make sure kthread is created before THREADED bit 6907 * is set. 6908 */ 6909 smp_mb__before_atomic(); 6910 6911 /* Setting/unsetting threaded mode on a napi might not immediately 6912 * take effect, if the current napi instance is actively being 6913 * polled. In this case, the switch between threaded mode and 6914 * softirq mode will happen in the next round of napi_schedule(). 6915 * This should not cause hiccups/stalls to the live traffic. 6916 */ 6917 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6918 if (threaded) 6919 set_bit(NAPI_STATE_THREADED, &napi->state); 6920 else 6921 clear_bit(NAPI_STATE_THREADED, &napi->state); 6922 } 6923 6924 return err; 6925 } 6926 EXPORT_SYMBOL(dev_set_threaded); 6927 6928 void netif_napi_add(struct net_device *dev, struct napi_struct *napi, 6929 int (*poll)(struct napi_struct *, int), int weight) 6930 { 6931 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state))) 6932 return; 6933 6934 INIT_LIST_HEAD(&napi->poll_list); 6935 INIT_HLIST_NODE(&napi->napi_hash_node); 6936 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); 6937 napi->timer.function = napi_watchdog; 6938 init_gro_hash(napi); 6939 napi->skb = NULL; 6940 INIT_LIST_HEAD(&napi->rx_list); 6941 napi->rx_count = 0; 6942 napi->poll = poll; 6943 if (weight > NAPI_POLL_WEIGHT) 6944 netdev_err_once(dev, "%s() called with weight %d\n", __func__, 6945 weight); 6946 napi->weight = weight; 6947 napi->dev = dev; 6948 #ifdef CONFIG_NETPOLL 6949 napi->poll_owner = -1; 6950 #endif 6951 set_bit(NAPI_STATE_SCHED, &napi->state); 6952 set_bit(NAPI_STATE_NPSVC, &napi->state); 6953 list_add_rcu(&napi->dev_list, &dev->napi_list); 6954 napi_hash_add(napi); 6955 /* Create kthread for this napi if dev->threaded is set. 6956 * Clear dev->threaded if kthread creation failed so that 6957 * threaded mode will not be enabled in napi_enable(). 6958 */ 6959 if (dev->threaded && napi_kthread_create(napi)) 6960 dev->threaded = 0; 6961 } 6962 EXPORT_SYMBOL(netif_napi_add); 6963 6964 void napi_disable(struct napi_struct *n) 6965 { 6966 might_sleep(); 6967 set_bit(NAPI_STATE_DISABLE, &n->state); 6968 6969 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state)) 6970 msleep(1); 6971 while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state)) 6972 msleep(1); 6973 6974 hrtimer_cancel(&n->timer); 6975 6976 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &n->state); 6977 clear_bit(NAPI_STATE_DISABLE, &n->state); 6978 clear_bit(NAPI_STATE_THREADED, &n->state); 6979 } 6980 EXPORT_SYMBOL(napi_disable); 6981 6982 /** 6983 * napi_enable - enable NAPI scheduling 6984 * @n: NAPI context 6985 * 6986 * Resume NAPI from being scheduled on this context. 6987 * Must be paired with napi_disable. 6988 */ 6989 void napi_enable(struct napi_struct *n) 6990 { 6991 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state)); 6992 smp_mb__before_atomic(); 6993 clear_bit(NAPI_STATE_SCHED, &n->state); 6994 clear_bit(NAPI_STATE_NPSVC, &n->state); 6995 if (n->dev->threaded && n->thread) 6996 set_bit(NAPI_STATE_THREADED, &n->state); 6997 } 6998 EXPORT_SYMBOL(napi_enable); 6999 7000 static void flush_gro_hash(struct napi_struct *napi) 7001 { 7002 int i; 7003 7004 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 7005 struct sk_buff *skb, *n; 7006 7007 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list) 7008 kfree_skb(skb); 7009 napi->gro_hash[i].count = 0; 7010 } 7011 } 7012 7013 /* Must be called in process context */ 7014 void __netif_napi_del(struct napi_struct *napi) 7015 { 7016 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state)) 7017 return; 7018 7019 napi_hash_del(napi); 7020 list_del_rcu(&napi->dev_list); 7021 napi_free_frags(napi); 7022 7023 flush_gro_hash(napi); 7024 napi->gro_bitmask = 0; 7025 7026 if (napi->thread) { 7027 kthread_stop(napi->thread); 7028 napi->thread = NULL; 7029 } 7030 } 7031 EXPORT_SYMBOL(__netif_napi_del); 7032 7033 static int __napi_poll(struct napi_struct *n, bool *repoll) 7034 { 7035 int work, weight; 7036 7037 weight = n->weight; 7038 7039 /* This NAPI_STATE_SCHED test is for avoiding a race 7040 * with netpoll's poll_napi(). Only the entity which 7041 * obtains the lock and sees NAPI_STATE_SCHED set will 7042 * actually make the ->poll() call. Therefore we avoid 7043 * accidentally calling ->poll() when NAPI is not scheduled. 7044 */ 7045 work = 0; 7046 if (test_bit(NAPI_STATE_SCHED, &n->state)) { 7047 work = n->poll(n, weight); 7048 trace_napi_poll(n, work, weight); 7049 } 7050 7051 if (unlikely(work > weight)) 7052 pr_err_once("NAPI poll function %pS returned %d, exceeding its budget of %d.\n", 7053 n->poll, work, weight); 7054 7055 if (likely(work < weight)) 7056 return work; 7057 7058 /* Drivers must not modify the NAPI state if they 7059 * consume the entire weight. In such cases this code 7060 * still "owns" the NAPI instance and therefore can 7061 * move the instance around on the list at-will. 7062 */ 7063 if (unlikely(napi_disable_pending(n))) { 7064 napi_complete(n); 7065 return work; 7066 } 7067 7068 /* The NAPI context has more processing work, but busy-polling 7069 * is preferred. Exit early. 7070 */ 7071 if (napi_prefer_busy_poll(n)) { 7072 if (napi_complete_done(n, work)) { 7073 /* If timeout is not set, we need to make sure 7074 * that the NAPI is re-scheduled. 7075 */ 7076 napi_schedule(n); 7077 } 7078 return work; 7079 } 7080 7081 if (n->gro_bitmask) { 7082 /* flush too old packets 7083 * If HZ < 1000, flush all packets. 7084 */ 7085 napi_gro_flush(n, HZ >= 1000); 7086 } 7087 7088 gro_normal_list(n); 7089 7090 /* Some drivers may have called napi_schedule 7091 * prior to exhausting their budget. 7092 */ 7093 if (unlikely(!list_empty(&n->poll_list))) { 7094 pr_warn_once("%s: Budget exhausted after napi rescheduled\n", 7095 n->dev ? n->dev->name : "backlog"); 7096 return work; 7097 } 7098 7099 *repoll = true; 7100 7101 return work; 7102 } 7103 7104 static int napi_poll(struct napi_struct *n, struct list_head *repoll) 7105 { 7106 bool do_repoll = false; 7107 void *have; 7108 int work; 7109 7110 list_del_init(&n->poll_list); 7111 7112 have = netpoll_poll_lock(n); 7113 7114 work = __napi_poll(n, &do_repoll); 7115 7116 if (do_repoll) 7117 list_add_tail(&n->poll_list, repoll); 7118 7119 netpoll_poll_unlock(have); 7120 7121 return work; 7122 } 7123 7124 static int napi_thread_wait(struct napi_struct *napi) 7125 { 7126 bool woken = false; 7127 7128 set_current_state(TASK_INTERRUPTIBLE); 7129 7130 while (!kthread_should_stop()) { 7131 /* Testing SCHED_THREADED bit here to make sure the current 7132 * kthread owns this napi and could poll on this napi. 7133 * Testing SCHED bit is not enough because SCHED bit might be 7134 * set by some other busy poll thread or by napi_disable(). 7135 */ 7136 if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state) || woken) { 7137 WARN_ON(!list_empty(&napi->poll_list)); 7138 __set_current_state(TASK_RUNNING); 7139 return 0; 7140 } 7141 7142 schedule(); 7143 /* woken being true indicates this thread owns this napi. */ 7144 woken = true; 7145 set_current_state(TASK_INTERRUPTIBLE); 7146 } 7147 __set_current_state(TASK_RUNNING); 7148 7149 return -1; 7150 } 7151 7152 static int napi_threaded_poll(void *data) 7153 { 7154 struct napi_struct *napi = data; 7155 void *have; 7156 7157 while (!napi_thread_wait(napi)) { 7158 for (;;) { 7159 bool repoll = false; 7160 7161 local_bh_disable(); 7162 7163 have = netpoll_poll_lock(napi); 7164 __napi_poll(napi, &repoll); 7165 netpoll_poll_unlock(have); 7166 7167 local_bh_enable(); 7168 7169 if (!repoll) 7170 break; 7171 7172 cond_resched(); 7173 } 7174 } 7175 return 0; 7176 } 7177 7178 static __latent_entropy void net_rx_action(struct softirq_action *h) 7179 { 7180 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 7181 unsigned long time_limit = jiffies + 7182 usecs_to_jiffies(netdev_budget_usecs); 7183 int budget = netdev_budget; 7184 LIST_HEAD(list); 7185 LIST_HEAD(repoll); 7186 7187 local_irq_disable(); 7188 list_splice_init(&sd->poll_list, &list); 7189 local_irq_enable(); 7190 7191 for (;;) { 7192 struct napi_struct *n; 7193 7194 if (list_empty(&list)) { 7195 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll)) 7196 return; 7197 break; 7198 } 7199 7200 n = list_first_entry(&list, struct napi_struct, poll_list); 7201 budget -= napi_poll(n, &repoll); 7202 7203 /* If softirq window is exhausted then punt. 7204 * Allow this to run for 2 jiffies since which will allow 7205 * an average latency of 1.5/HZ. 7206 */ 7207 if (unlikely(budget <= 0 || 7208 time_after_eq(jiffies, time_limit))) { 7209 sd->time_squeeze++; 7210 break; 7211 } 7212 } 7213 7214 local_irq_disable(); 7215 7216 list_splice_tail_init(&sd->poll_list, &list); 7217 list_splice_tail(&repoll, &list); 7218 list_splice(&list, &sd->poll_list); 7219 if (!list_empty(&sd->poll_list)) 7220 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 7221 7222 net_rps_action_and_irq_enable(sd); 7223 } 7224 7225 struct netdev_adjacent { 7226 struct net_device *dev; 7227 7228 /* upper master flag, there can only be one master device per list */ 7229 bool master; 7230 7231 /* lookup ignore flag */ 7232 bool ignore; 7233 7234 /* counter for the number of times this device was added to us */ 7235 u16 ref_nr; 7236 7237 /* private field for the users */ 7238 void *private; 7239 7240 struct list_head list; 7241 struct rcu_head rcu; 7242 }; 7243 7244 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, 7245 struct list_head *adj_list) 7246 { 7247 struct netdev_adjacent *adj; 7248 7249 list_for_each_entry(adj, adj_list, list) { 7250 if (adj->dev == adj_dev) 7251 return adj; 7252 } 7253 return NULL; 7254 } 7255 7256 static int ____netdev_has_upper_dev(struct net_device *upper_dev, 7257 struct netdev_nested_priv *priv) 7258 { 7259 struct net_device *dev = (struct net_device *)priv->data; 7260 7261 return upper_dev == dev; 7262 } 7263 7264 /** 7265 * netdev_has_upper_dev - Check if device is linked to an upper device 7266 * @dev: device 7267 * @upper_dev: upper device to check 7268 * 7269 * Find out if a device is linked to specified upper device and return true 7270 * in case it is. Note that this checks only immediate upper device, 7271 * not through a complete stack of devices. The caller must hold the RTNL lock. 7272 */ 7273 bool netdev_has_upper_dev(struct net_device *dev, 7274 struct net_device *upper_dev) 7275 { 7276 struct netdev_nested_priv priv = { 7277 .data = (void *)upper_dev, 7278 }; 7279 7280 ASSERT_RTNL(); 7281 7282 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 7283 &priv); 7284 } 7285 EXPORT_SYMBOL(netdev_has_upper_dev); 7286 7287 /** 7288 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device 7289 * @dev: device 7290 * @upper_dev: upper device to check 7291 * 7292 * Find out if a device is linked to specified upper device and return true 7293 * in case it is. Note that this checks the entire upper device chain. 7294 * The caller must hold rcu lock. 7295 */ 7296 7297 bool netdev_has_upper_dev_all_rcu(struct net_device *dev, 7298 struct net_device *upper_dev) 7299 { 7300 struct netdev_nested_priv priv = { 7301 .data = (void *)upper_dev, 7302 }; 7303 7304 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 7305 &priv); 7306 } 7307 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); 7308 7309 /** 7310 * netdev_has_any_upper_dev - Check if device is linked to some device 7311 * @dev: device 7312 * 7313 * Find out if a device is linked to an upper device and return true in case 7314 * it is. The caller must hold the RTNL lock. 7315 */ 7316 bool netdev_has_any_upper_dev(struct net_device *dev) 7317 { 7318 ASSERT_RTNL(); 7319 7320 return !list_empty(&dev->adj_list.upper); 7321 } 7322 EXPORT_SYMBOL(netdev_has_any_upper_dev); 7323 7324 /** 7325 * netdev_master_upper_dev_get - Get master upper device 7326 * @dev: device 7327 * 7328 * Find a master upper device and return pointer to it or NULL in case 7329 * it's not there. The caller must hold the RTNL lock. 7330 */ 7331 struct net_device *netdev_master_upper_dev_get(struct net_device *dev) 7332 { 7333 struct netdev_adjacent *upper; 7334 7335 ASSERT_RTNL(); 7336 7337 if (list_empty(&dev->adj_list.upper)) 7338 return NULL; 7339 7340 upper = list_first_entry(&dev->adj_list.upper, 7341 struct netdev_adjacent, list); 7342 if (likely(upper->master)) 7343 return upper->dev; 7344 return NULL; 7345 } 7346 EXPORT_SYMBOL(netdev_master_upper_dev_get); 7347 7348 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev) 7349 { 7350 struct netdev_adjacent *upper; 7351 7352 ASSERT_RTNL(); 7353 7354 if (list_empty(&dev->adj_list.upper)) 7355 return NULL; 7356 7357 upper = list_first_entry(&dev->adj_list.upper, 7358 struct netdev_adjacent, list); 7359 if (likely(upper->master) && !upper->ignore) 7360 return upper->dev; 7361 return NULL; 7362 } 7363 7364 /** 7365 * netdev_has_any_lower_dev - Check if device is linked to some device 7366 * @dev: device 7367 * 7368 * Find out if a device is linked to a lower device and return true in case 7369 * it is. The caller must hold the RTNL lock. 7370 */ 7371 static bool netdev_has_any_lower_dev(struct net_device *dev) 7372 { 7373 ASSERT_RTNL(); 7374 7375 return !list_empty(&dev->adj_list.lower); 7376 } 7377 7378 void *netdev_adjacent_get_private(struct list_head *adj_list) 7379 { 7380 struct netdev_adjacent *adj; 7381 7382 adj = list_entry(adj_list, struct netdev_adjacent, list); 7383 7384 return adj->private; 7385 } 7386 EXPORT_SYMBOL(netdev_adjacent_get_private); 7387 7388 /** 7389 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list 7390 * @dev: device 7391 * @iter: list_head ** of the current position 7392 * 7393 * Gets the next device from the dev's upper list, starting from iter 7394 * position. The caller must hold RCU read lock. 7395 */ 7396 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, 7397 struct list_head **iter) 7398 { 7399 struct netdev_adjacent *upper; 7400 7401 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 7402 7403 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7404 7405 if (&upper->list == &dev->adj_list.upper) 7406 return NULL; 7407 7408 *iter = &upper->list; 7409 7410 return upper->dev; 7411 } 7412 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); 7413 7414 static struct net_device *__netdev_next_upper_dev(struct net_device *dev, 7415 struct list_head **iter, 7416 bool *ignore) 7417 { 7418 struct netdev_adjacent *upper; 7419 7420 upper = list_entry((*iter)->next, struct netdev_adjacent, list); 7421 7422 if (&upper->list == &dev->adj_list.upper) 7423 return NULL; 7424 7425 *iter = &upper->list; 7426 *ignore = upper->ignore; 7427 7428 return upper->dev; 7429 } 7430 7431 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, 7432 struct list_head **iter) 7433 { 7434 struct netdev_adjacent *upper; 7435 7436 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 7437 7438 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7439 7440 if (&upper->list == &dev->adj_list.upper) 7441 return NULL; 7442 7443 *iter = &upper->list; 7444 7445 return upper->dev; 7446 } 7447 7448 static int __netdev_walk_all_upper_dev(struct net_device *dev, 7449 int (*fn)(struct net_device *dev, 7450 struct netdev_nested_priv *priv), 7451 struct netdev_nested_priv *priv) 7452 { 7453 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7454 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7455 int ret, cur = 0; 7456 bool ignore; 7457 7458 now = dev; 7459 iter = &dev->adj_list.upper; 7460 7461 while (1) { 7462 if (now != dev) { 7463 ret = fn(now, priv); 7464 if (ret) 7465 return ret; 7466 } 7467 7468 next = NULL; 7469 while (1) { 7470 udev = __netdev_next_upper_dev(now, &iter, &ignore); 7471 if (!udev) 7472 break; 7473 if (ignore) 7474 continue; 7475 7476 next = udev; 7477 niter = &udev->adj_list.upper; 7478 dev_stack[cur] = now; 7479 iter_stack[cur++] = iter; 7480 break; 7481 } 7482 7483 if (!next) { 7484 if (!cur) 7485 return 0; 7486 next = dev_stack[--cur]; 7487 niter = iter_stack[cur]; 7488 } 7489 7490 now = next; 7491 iter = niter; 7492 } 7493 7494 return 0; 7495 } 7496 7497 int netdev_walk_all_upper_dev_rcu(struct net_device *dev, 7498 int (*fn)(struct net_device *dev, 7499 struct netdev_nested_priv *priv), 7500 struct netdev_nested_priv *priv) 7501 { 7502 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7503 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7504 int ret, cur = 0; 7505 7506 now = dev; 7507 iter = &dev->adj_list.upper; 7508 7509 while (1) { 7510 if (now != dev) { 7511 ret = fn(now, priv); 7512 if (ret) 7513 return ret; 7514 } 7515 7516 next = NULL; 7517 while (1) { 7518 udev = netdev_next_upper_dev_rcu(now, &iter); 7519 if (!udev) 7520 break; 7521 7522 next = udev; 7523 niter = &udev->adj_list.upper; 7524 dev_stack[cur] = now; 7525 iter_stack[cur++] = iter; 7526 break; 7527 } 7528 7529 if (!next) { 7530 if (!cur) 7531 return 0; 7532 next = dev_stack[--cur]; 7533 niter = iter_stack[cur]; 7534 } 7535 7536 now = next; 7537 iter = niter; 7538 } 7539 7540 return 0; 7541 } 7542 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); 7543 7544 static bool __netdev_has_upper_dev(struct net_device *dev, 7545 struct net_device *upper_dev) 7546 { 7547 struct netdev_nested_priv priv = { 7548 .flags = 0, 7549 .data = (void *)upper_dev, 7550 }; 7551 7552 ASSERT_RTNL(); 7553 7554 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev, 7555 &priv); 7556 } 7557 7558 /** 7559 * netdev_lower_get_next_private - Get the next ->private from the 7560 * lower neighbour list 7561 * @dev: device 7562 * @iter: list_head ** of the current position 7563 * 7564 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7565 * list, starting from iter position. The caller must hold either hold the 7566 * RTNL lock or its own locking that guarantees that the neighbour lower 7567 * list will remain unchanged. 7568 */ 7569 void *netdev_lower_get_next_private(struct net_device *dev, 7570 struct list_head **iter) 7571 { 7572 struct netdev_adjacent *lower; 7573 7574 lower = list_entry(*iter, struct netdev_adjacent, list); 7575 7576 if (&lower->list == &dev->adj_list.lower) 7577 return NULL; 7578 7579 *iter = lower->list.next; 7580 7581 return lower->private; 7582 } 7583 EXPORT_SYMBOL(netdev_lower_get_next_private); 7584 7585 /** 7586 * netdev_lower_get_next_private_rcu - Get the next ->private from the 7587 * lower neighbour list, RCU 7588 * variant 7589 * @dev: device 7590 * @iter: list_head ** of the current position 7591 * 7592 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7593 * list, starting from iter position. The caller must hold RCU read lock. 7594 */ 7595 void *netdev_lower_get_next_private_rcu(struct net_device *dev, 7596 struct list_head **iter) 7597 { 7598 struct netdev_adjacent *lower; 7599 7600 WARN_ON_ONCE(!rcu_read_lock_held()); 7601 7602 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7603 7604 if (&lower->list == &dev->adj_list.lower) 7605 return NULL; 7606 7607 *iter = &lower->list; 7608 7609 return lower->private; 7610 } 7611 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); 7612 7613 /** 7614 * netdev_lower_get_next - Get the next device from the lower neighbour 7615 * list 7616 * @dev: device 7617 * @iter: list_head ** of the current position 7618 * 7619 * Gets the next netdev_adjacent from the dev's lower neighbour 7620 * list, starting from iter position. The caller must hold RTNL lock or 7621 * its own locking that guarantees that the neighbour lower 7622 * list will remain unchanged. 7623 */ 7624 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) 7625 { 7626 struct netdev_adjacent *lower; 7627 7628 lower = list_entry(*iter, struct netdev_adjacent, list); 7629 7630 if (&lower->list == &dev->adj_list.lower) 7631 return NULL; 7632 7633 *iter = lower->list.next; 7634 7635 return lower->dev; 7636 } 7637 EXPORT_SYMBOL(netdev_lower_get_next); 7638 7639 static struct net_device *netdev_next_lower_dev(struct net_device *dev, 7640 struct list_head **iter) 7641 { 7642 struct netdev_adjacent *lower; 7643 7644 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7645 7646 if (&lower->list == &dev->adj_list.lower) 7647 return NULL; 7648 7649 *iter = &lower->list; 7650 7651 return lower->dev; 7652 } 7653 7654 static struct net_device *__netdev_next_lower_dev(struct net_device *dev, 7655 struct list_head **iter, 7656 bool *ignore) 7657 { 7658 struct netdev_adjacent *lower; 7659 7660 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7661 7662 if (&lower->list == &dev->adj_list.lower) 7663 return NULL; 7664 7665 *iter = &lower->list; 7666 *ignore = lower->ignore; 7667 7668 return lower->dev; 7669 } 7670 7671 int netdev_walk_all_lower_dev(struct net_device *dev, 7672 int (*fn)(struct net_device *dev, 7673 struct netdev_nested_priv *priv), 7674 struct netdev_nested_priv *priv) 7675 { 7676 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7677 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7678 int ret, cur = 0; 7679 7680 now = dev; 7681 iter = &dev->adj_list.lower; 7682 7683 while (1) { 7684 if (now != dev) { 7685 ret = fn(now, priv); 7686 if (ret) 7687 return ret; 7688 } 7689 7690 next = NULL; 7691 while (1) { 7692 ldev = netdev_next_lower_dev(now, &iter); 7693 if (!ldev) 7694 break; 7695 7696 next = ldev; 7697 niter = &ldev->adj_list.lower; 7698 dev_stack[cur] = now; 7699 iter_stack[cur++] = iter; 7700 break; 7701 } 7702 7703 if (!next) { 7704 if (!cur) 7705 return 0; 7706 next = dev_stack[--cur]; 7707 niter = iter_stack[cur]; 7708 } 7709 7710 now = next; 7711 iter = niter; 7712 } 7713 7714 return 0; 7715 } 7716 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); 7717 7718 static int __netdev_walk_all_lower_dev(struct net_device *dev, 7719 int (*fn)(struct net_device *dev, 7720 struct netdev_nested_priv *priv), 7721 struct netdev_nested_priv *priv) 7722 { 7723 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7724 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7725 int ret, cur = 0; 7726 bool ignore; 7727 7728 now = dev; 7729 iter = &dev->adj_list.lower; 7730 7731 while (1) { 7732 if (now != dev) { 7733 ret = fn(now, priv); 7734 if (ret) 7735 return ret; 7736 } 7737 7738 next = NULL; 7739 while (1) { 7740 ldev = __netdev_next_lower_dev(now, &iter, &ignore); 7741 if (!ldev) 7742 break; 7743 if (ignore) 7744 continue; 7745 7746 next = ldev; 7747 niter = &ldev->adj_list.lower; 7748 dev_stack[cur] = now; 7749 iter_stack[cur++] = iter; 7750 break; 7751 } 7752 7753 if (!next) { 7754 if (!cur) 7755 return 0; 7756 next = dev_stack[--cur]; 7757 niter = iter_stack[cur]; 7758 } 7759 7760 now = next; 7761 iter = niter; 7762 } 7763 7764 return 0; 7765 } 7766 7767 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, 7768 struct list_head **iter) 7769 { 7770 struct netdev_adjacent *lower; 7771 7772 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7773 if (&lower->list == &dev->adj_list.lower) 7774 return NULL; 7775 7776 *iter = &lower->list; 7777 7778 return lower->dev; 7779 } 7780 EXPORT_SYMBOL(netdev_next_lower_dev_rcu); 7781 7782 static u8 __netdev_upper_depth(struct net_device *dev) 7783 { 7784 struct net_device *udev; 7785 struct list_head *iter; 7786 u8 max_depth = 0; 7787 bool ignore; 7788 7789 for (iter = &dev->adj_list.upper, 7790 udev = __netdev_next_upper_dev(dev, &iter, &ignore); 7791 udev; 7792 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) { 7793 if (ignore) 7794 continue; 7795 if (max_depth < udev->upper_level) 7796 max_depth = udev->upper_level; 7797 } 7798 7799 return max_depth; 7800 } 7801 7802 static u8 __netdev_lower_depth(struct net_device *dev) 7803 { 7804 struct net_device *ldev; 7805 struct list_head *iter; 7806 u8 max_depth = 0; 7807 bool ignore; 7808 7809 for (iter = &dev->adj_list.lower, 7810 ldev = __netdev_next_lower_dev(dev, &iter, &ignore); 7811 ldev; 7812 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) { 7813 if (ignore) 7814 continue; 7815 if (max_depth < ldev->lower_level) 7816 max_depth = ldev->lower_level; 7817 } 7818 7819 return max_depth; 7820 } 7821 7822 static int __netdev_update_upper_level(struct net_device *dev, 7823 struct netdev_nested_priv *__unused) 7824 { 7825 dev->upper_level = __netdev_upper_depth(dev) + 1; 7826 return 0; 7827 } 7828 7829 static int __netdev_update_lower_level(struct net_device *dev, 7830 struct netdev_nested_priv *priv) 7831 { 7832 dev->lower_level = __netdev_lower_depth(dev) + 1; 7833 7834 #ifdef CONFIG_LOCKDEP 7835 if (!priv) 7836 return 0; 7837 7838 if (priv->flags & NESTED_SYNC_IMM) 7839 dev->nested_level = dev->lower_level - 1; 7840 if (priv->flags & NESTED_SYNC_TODO) 7841 net_unlink_todo(dev); 7842 #endif 7843 return 0; 7844 } 7845 7846 int netdev_walk_all_lower_dev_rcu(struct net_device *dev, 7847 int (*fn)(struct net_device *dev, 7848 struct netdev_nested_priv *priv), 7849 struct netdev_nested_priv *priv) 7850 { 7851 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7852 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7853 int ret, cur = 0; 7854 7855 now = dev; 7856 iter = &dev->adj_list.lower; 7857 7858 while (1) { 7859 if (now != dev) { 7860 ret = fn(now, priv); 7861 if (ret) 7862 return ret; 7863 } 7864 7865 next = NULL; 7866 while (1) { 7867 ldev = netdev_next_lower_dev_rcu(now, &iter); 7868 if (!ldev) 7869 break; 7870 7871 next = ldev; 7872 niter = &ldev->adj_list.lower; 7873 dev_stack[cur] = now; 7874 iter_stack[cur++] = iter; 7875 break; 7876 } 7877 7878 if (!next) { 7879 if (!cur) 7880 return 0; 7881 next = dev_stack[--cur]; 7882 niter = iter_stack[cur]; 7883 } 7884 7885 now = next; 7886 iter = niter; 7887 } 7888 7889 return 0; 7890 } 7891 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); 7892 7893 /** 7894 * netdev_lower_get_first_private_rcu - Get the first ->private from the 7895 * lower neighbour list, RCU 7896 * variant 7897 * @dev: device 7898 * 7899 * Gets the first netdev_adjacent->private from the dev's lower neighbour 7900 * list. The caller must hold RCU read lock. 7901 */ 7902 void *netdev_lower_get_first_private_rcu(struct net_device *dev) 7903 { 7904 struct netdev_adjacent *lower; 7905 7906 lower = list_first_or_null_rcu(&dev->adj_list.lower, 7907 struct netdev_adjacent, list); 7908 if (lower) 7909 return lower->private; 7910 return NULL; 7911 } 7912 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); 7913 7914 /** 7915 * netdev_master_upper_dev_get_rcu - Get master upper device 7916 * @dev: device 7917 * 7918 * Find a master upper device and return pointer to it or NULL in case 7919 * it's not there. The caller must hold the RCU read lock. 7920 */ 7921 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) 7922 { 7923 struct netdev_adjacent *upper; 7924 7925 upper = list_first_or_null_rcu(&dev->adj_list.upper, 7926 struct netdev_adjacent, list); 7927 if (upper && likely(upper->master)) 7928 return upper->dev; 7929 return NULL; 7930 } 7931 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); 7932 7933 static int netdev_adjacent_sysfs_add(struct net_device *dev, 7934 struct net_device *adj_dev, 7935 struct list_head *dev_list) 7936 { 7937 char linkname[IFNAMSIZ+7]; 7938 7939 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7940 "upper_%s" : "lower_%s", adj_dev->name); 7941 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), 7942 linkname); 7943 } 7944 static void netdev_adjacent_sysfs_del(struct net_device *dev, 7945 char *name, 7946 struct list_head *dev_list) 7947 { 7948 char linkname[IFNAMSIZ+7]; 7949 7950 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7951 "upper_%s" : "lower_%s", name); 7952 sysfs_remove_link(&(dev->dev.kobj), linkname); 7953 } 7954 7955 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, 7956 struct net_device *adj_dev, 7957 struct list_head *dev_list) 7958 { 7959 return (dev_list == &dev->adj_list.upper || 7960 dev_list == &dev->adj_list.lower) && 7961 net_eq(dev_net(dev), dev_net(adj_dev)); 7962 } 7963 7964 static int __netdev_adjacent_dev_insert(struct net_device *dev, 7965 struct net_device *adj_dev, 7966 struct list_head *dev_list, 7967 void *private, bool master) 7968 { 7969 struct netdev_adjacent *adj; 7970 int ret; 7971 7972 adj = __netdev_find_adj(adj_dev, dev_list); 7973 7974 if (adj) { 7975 adj->ref_nr += 1; 7976 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", 7977 dev->name, adj_dev->name, adj->ref_nr); 7978 7979 return 0; 7980 } 7981 7982 adj = kmalloc(sizeof(*adj), GFP_KERNEL); 7983 if (!adj) 7984 return -ENOMEM; 7985 7986 adj->dev = adj_dev; 7987 adj->master = master; 7988 adj->ref_nr = 1; 7989 adj->private = private; 7990 adj->ignore = false; 7991 dev_hold(adj_dev); 7992 7993 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", 7994 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); 7995 7996 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { 7997 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); 7998 if (ret) 7999 goto free_adj; 8000 } 8001 8002 /* Ensure that master link is always the first item in list. */ 8003 if (master) { 8004 ret = sysfs_create_link(&(dev->dev.kobj), 8005 &(adj_dev->dev.kobj), "master"); 8006 if (ret) 8007 goto remove_symlinks; 8008 8009 list_add_rcu(&adj->list, dev_list); 8010 } else { 8011 list_add_tail_rcu(&adj->list, dev_list); 8012 } 8013 8014 return 0; 8015 8016 remove_symlinks: 8017 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 8018 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 8019 free_adj: 8020 kfree(adj); 8021 dev_put(adj_dev); 8022 8023 return ret; 8024 } 8025 8026 static void __netdev_adjacent_dev_remove(struct net_device *dev, 8027 struct net_device *adj_dev, 8028 u16 ref_nr, 8029 struct list_head *dev_list) 8030 { 8031 struct netdev_adjacent *adj; 8032 8033 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", 8034 dev->name, adj_dev->name, ref_nr); 8035 8036 adj = __netdev_find_adj(adj_dev, dev_list); 8037 8038 if (!adj) { 8039 pr_err("Adjacency does not exist for device %s from %s\n", 8040 dev->name, adj_dev->name); 8041 WARN_ON(1); 8042 return; 8043 } 8044 8045 if (adj->ref_nr > ref_nr) { 8046 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", 8047 dev->name, adj_dev->name, ref_nr, 8048 adj->ref_nr - ref_nr); 8049 adj->ref_nr -= ref_nr; 8050 return; 8051 } 8052 8053 if (adj->master) 8054 sysfs_remove_link(&(dev->dev.kobj), "master"); 8055 8056 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 8057 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 8058 8059 list_del_rcu(&adj->list); 8060 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", 8061 adj_dev->name, dev->name, adj_dev->name); 8062 dev_put(adj_dev); 8063 kfree_rcu(adj, rcu); 8064 } 8065 8066 static int __netdev_adjacent_dev_link_lists(struct net_device *dev, 8067 struct net_device *upper_dev, 8068 struct list_head *up_list, 8069 struct list_head *down_list, 8070 void *private, bool master) 8071 { 8072 int ret; 8073 8074 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, 8075 private, master); 8076 if (ret) 8077 return ret; 8078 8079 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, 8080 private, false); 8081 if (ret) { 8082 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); 8083 return ret; 8084 } 8085 8086 return 0; 8087 } 8088 8089 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, 8090 struct net_device *upper_dev, 8091 u16 ref_nr, 8092 struct list_head *up_list, 8093 struct list_head *down_list) 8094 { 8095 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); 8096 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); 8097 } 8098 8099 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, 8100 struct net_device *upper_dev, 8101 void *private, bool master) 8102 { 8103 return __netdev_adjacent_dev_link_lists(dev, upper_dev, 8104 &dev->adj_list.upper, 8105 &upper_dev->adj_list.lower, 8106 private, master); 8107 } 8108 8109 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, 8110 struct net_device *upper_dev) 8111 { 8112 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, 8113 &dev->adj_list.upper, 8114 &upper_dev->adj_list.lower); 8115 } 8116 8117 static int __netdev_upper_dev_link(struct net_device *dev, 8118 struct net_device *upper_dev, bool master, 8119 void *upper_priv, void *upper_info, 8120 struct netdev_nested_priv *priv, 8121 struct netlink_ext_ack *extack) 8122 { 8123 struct netdev_notifier_changeupper_info changeupper_info = { 8124 .info = { 8125 .dev = dev, 8126 .extack = extack, 8127 }, 8128 .upper_dev = upper_dev, 8129 .master = master, 8130 .linking = true, 8131 .upper_info = upper_info, 8132 }; 8133 struct net_device *master_dev; 8134 int ret = 0; 8135 8136 ASSERT_RTNL(); 8137 8138 if (dev == upper_dev) 8139 return -EBUSY; 8140 8141 /* To prevent loops, check if dev is not upper device to upper_dev. */ 8142 if (__netdev_has_upper_dev(upper_dev, dev)) 8143 return -EBUSY; 8144 8145 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV) 8146 return -EMLINK; 8147 8148 if (!master) { 8149 if (__netdev_has_upper_dev(dev, upper_dev)) 8150 return -EEXIST; 8151 } else { 8152 master_dev = __netdev_master_upper_dev_get(dev); 8153 if (master_dev) 8154 return master_dev == upper_dev ? -EEXIST : -EBUSY; 8155 } 8156 8157 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 8158 &changeupper_info.info); 8159 ret = notifier_to_errno(ret); 8160 if (ret) 8161 return ret; 8162 8163 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, 8164 master); 8165 if (ret) 8166 return ret; 8167 8168 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 8169 &changeupper_info.info); 8170 ret = notifier_to_errno(ret); 8171 if (ret) 8172 goto rollback; 8173 8174 __netdev_update_upper_level(dev, NULL); 8175 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 8176 8177 __netdev_update_lower_level(upper_dev, priv); 8178 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 8179 priv); 8180 8181 return 0; 8182 8183 rollback: 8184 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 8185 8186 return ret; 8187 } 8188 8189 /** 8190 * netdev_upper_dev_link - Add a link to the upper device 8191 * @dev: device 8192 * @upper_dev: new upper device 8193 * @extack: netlink extended ack 8194 * 8195 * Adds a link to device which is upper to this one. The caller must hold 8196 * the RTNL lock. On a failure a negative errno code is returned. 8197 * On success the reference counts are adjusted and the function 8198 * returns zero. 8199 */ 8200 int netdev_upper_dev_link(struct net_device *dev, 8201 struct net_device *upper_dev, 8202 struct netlink_ext_ack *extack) 8203 { 8204 struct netdev_nested_priv priv = { 8205 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 8206 .data = NULL, 8207 }; 8208 8209 return __netdev_upper_dev_link(dev, upper_dev, false, 8210 NULL, NULL, &priv, extack); 8211 } 8212 EXPORT_SYMBOL(netdev_upper_dev_link); 8213 8214 /** 8215 * netdev_master_upper_dev_link - Add a master link to the upper device 8216 * @dev: device 8217 * @upper_dev: new upper device 8218 * @upper_priv: upper device private 8219 * @upper_info: upper info to be passed down via notifier 8220 * @extack: netlink extended ack 8221 * 8222 * Adds a link to device which is upper to this one. In this case, only 8223 * one master upper device can be linked, although other non-master devices 8224 * might be linked as well. The caller must hold the RTNL lock. 8225 * On a failure a negative errno code is returned. On success the reference 8226 * counts are adjusted and the function returns zero. 8227 */ 8228 int netdev_master_upper_dev_link(struct net_device *dev, 8229 struct net_device *upper_dev, 8230 void *upper_priv, void *upper_info, 8231 struct netlink_ext_ack *extack) 8232 { 8233 struct netdev_nested_priv priv = { 8234 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 8235 .data = NULL, 8236 }; 8237 8238 return __netdev_upper_dev_link(dev, upper_dev, true, 8239 upper_priv, upper_info, &priv, extack); 8240 } 8241 EXPORT_SYMBOL(netdev_master_upper_dev_link); 8242 8243 static void __netdev_upper_dev_unlink(struct net_device *dev, 8244 struct net_device *upper_dev, 8245 struct netdev_nested_priv *priv) 8246 { 8247 struct netdev_notifier_changeupper_info changeupper_info = { 8248 .info = { 8249 .dev = dev, 8250 }, 8251 .upper_dev = upper_dev, 8252 .linking = false, 8253 }; 8254 8255 ASSERT_RTNL(); 8256 8257 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; 8258 8259 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 8260 &changeupper_info.info); 8261 8262 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 8263 8264 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 8265 &changeupper_info.info); 8266 8267 __netdev_update_upper_level(dev, NULL); 8268 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 8269 8270 __netdev_update_lower_level(upper_dev, priv); 8271 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 8272 priv); 8273 } 8274 8275 /** 8276 * netdev_upper_dev_unlink - Removes a link to upper device 8277 * @dev: device 8278 * @upper_dev: new upper device 8279 * 8280 * Removes a link to device which is upper to this one. The caller must hold 8281 * the RTNL lock. 8282 */ 8283 void netdev_upper_dev_unlink(struct net_device *dev, 8284 struct net_device *upper_dev) 8285 { 8286 struct netdev_nested_priv priv = { 8287 .flags = NESTED_SYNC_TODO, 8288 .data = NULL, 8289 }; 8290 8291 __netdev_upper_dev_unlink(dev, upper_dev, &priv); 8292 } 8293 EXPORT_SYMBOL(netdev_upper_dev_unlink); 8294 8295 static void __netdev_adjacent_dev_set(struct net_device *upper_dev, 8296 struct net_device *lower_dev, 8297 bool val) 8298 { 8299 struct netdev_adjacent *adj; 8300 8301 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower); 8302 if (adj) 8303 adj->ignore = val; 8304 8305 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper); 8306 if (adj) 8307 adj->ignore = val; 8308 } 8309 8310 static void netdev_adjacent_dev_disable(struct net_device *upper_dev, 8311 struct net_device *lower_dev) 8312 { 8313 __netdev_adjacent_dev_set(upper_dev, lower_dev, true); 8314 } 8315 8316 static void netdev_adjacent_dev_enable(struct net_device *upper_dev, 8317 struct net_device *lower_dev) 8318 { 8319 __netdev_adjacent_dev_set(upper_dev, lower_dev, false); 8320 } 8321 8322 int netdev_adjacent_change_prepare(struct net_device *old_dev, 8323 struct net_device *new_dev, 8324 struct net_device *dev, 8325 struct netlink_ext_ack *extack) 8326 { 8327 struct netdev_nested_priv priv = { 8328 .flags = 0, 8329 .data = NULL, 8330 }; 8331 int err; 8332 8333 if (!new_dev) 8334 return 0; 8335 8336 if (old_dev && new_dev != old_dev) 8337 netdev_adjacent_dev_disable(dev, old_dev); 8338 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv, 8339 extack); 8340 if (err) { 8341 if (old_dev && new_dev != old_dev) 8342 netdev_adjacent_dev_enable(dev, old_dev); 8343 return err; 8344 } 8345 8346 return 0; 8347 } 8348 EXPORT_SYMBOL(netdev_adjacent_change_prepare); 8349 8350 void netdev_adjacent_change_commit(struct net_device *old_dev, 8351 struct net_device *new_dev, 8352 struct net_device *dev) 8353 { 8354 struct netdev_nested_priv priv = { 8355 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 8356 .data = NULL, 8357 }; 8358 8359 if (!new_dev || !old_dev) 8360 return; 8361 8362 if (new_dev == old_dev) 8363 return; 8364 8365 netdev_adjacent_dev_enable(dev, old_dev); 8366 __netdev_upper_dev_unlink(old_dev, dev, &priv); 8367 } 8368 EXPORT_SYMBOL(netdev_adjacent_change_commit); 8369 8370 void netdev_adjacent_change_abort(struct net_device *old_dev, 8371 struct net_device *new_dev, 8372 struct net_device *dev) 8373 { 8374 struct netdev_nested_priv priv = { 8375 .flags = 0, 8376 .data = NULL, 8377 }; 8378 8379 if (!new_dev) 8380 return; 8381 8382 if (old_dev && new_dev != old_dev) 8383 netdev_adjacent_dev_enable(dev, old_dev); 8384 8385 __netdev_upper_dev_unlink(new_dev, dev, &priv); 8386 } 8387 EXPORT_SYMBOL(netdev_adjacent_change_abort); 8388 8389 /** 8390 * netdev_bonding_info_change - Dispatch event about slave change 8391 * @dev: device 8392 * @bonding_info: info to dispatch 8393 * 8394 * Send NETDEV_BONDING_INFO to netdev notifiers with info. 8395 * The caller must hold the RTNL lock. 8396 */ 8397 void netdev_bonding_info_change(struct net_device *dev, 8398 struct netdev_bonding_info *bonding_info) 8399 { 8400 struct netdev_notifier_bonding_info info = { 8401 .info.dev = dev, 8402 }; 8403 8404 memcpy(&info.bonding_info, bonding_info, 8405 sizeof(struct netdev_bonding_info)); 8406 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, 8407 &info.info); 8408 } 8409 EXPORT_SYMBOL(netdev_bonding_info_change); 8410 8411 /** 8412 * netdev_get_xmit_slave - Get the xmit slave of master device 8413 * @dev: device 8414 * @skb: The packet 8415 * @all_slaves: assume all the slaves are active 8416 * 8417 * The reference counters are not incremented so the caller must be 8418 * careful with locks. The caller must hold RCU lock. 8419 * %NULL is returned if no slave is found. 8420 */ 8421 8422 struct net_device *netdev_get_xmit_slave(struct net_device *dev, 8423 struct sk_buff *skb, 8424 bool all_slaves) 8425 { 8426 const struct net_device_ops *ops = dev->netdev_ops; 8427 8428 if (!ops->ndo_get_xmit_slave) 8429 return NULL; 8430 return ops->ndo_get_xmit_slave(dev, skb, all_slaves); 8431 } 8432 EXPORT_SYMBOL(netdev_get_xmit_slave); 8433 8434 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev, 8435 struct sock *sk) 8436 { 8437 const struct net_device_ops *ops = dev->netdev_ops; 8438 8439 if (!ops->ndo_sk_get_lower_dev) 8440 return NULL; 8441 return ops->ndo_sk_get_lower_dev(dev, sk); 8442 } 8443 8444 /** 8445 * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket 8446 * @dev: device 8447 * @sk: the socket 8448 * 8449 * %NULL is returned if no lower device is found. 8450 */ 8451 8452 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev, 8453 struct sock *sk) 8454 { 8455 struct net_device *lower; 8456 8457 lower = netdev_sk_get_lower_dev(dev, sk); 8458 while (lower) { 8459 dev = lower; 8460 lower = netdev_sk_get_lower_dev(dev, sk); 8461 } 8462 8463 return dev; 8464 } 8465 EXPORT_SYMBOL(netdev_sk_get_lowest_dev); 8466 8467 static void netdev_adjacent_add_links(struct net_device *dev) 8468 { 8469 struct netdev_adjacent *iter; 8470 8471 struct net *net = dev_net(dev); 8472 8473 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8474 if (!net_eq(net, dev_net(iter->dev))) 8475 continue; 8476 netdev_adjacent_sysfs_add(iter->dev, dev, 8477 &iter->dev->adj_list.lower); 8478 netdev_adjacent_sysfs_add(dev, iter->dev, 8479 &dev->adj_list.upper); 8480 } 8481 8482 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8483 if (!net_eq(net, dev_net(iter->dev))) 8484 continue; 8485 netdev_adjacent_sysfs_add(iter->dev, dev, 8486 &iter->dev->adj_list.upper); 8487 netdev_adjacent_sysfs_add(dev, iter->dev, 8488 &dev->adj_list.lower); 8489 } 8490 } 8491 8492 static void netdev_adjacent_del_links(struct net_device *dev) 8493 { 8494 struct netdev_adjacent *iter; 8495 8496 struct net *net = dev_net(dev); 8497 8498 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8499 if (!net_eq(net, dev_net(iter->dev))) 8500 continue; 8501 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8502 &iter->dev->adj_list.lower); 8503 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8504 &dev->adj_list.upper); 8505 } 8506 8507 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8508 if (!net_eq(net, dev_net(iter->dev))) 8509 continue; 8510 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8511 &iter->dev->adj_list.upper); 8512 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8513 &dev->adj_list.lower); 8514 } 8515 } 8516 8517 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) 8518 { 8519 struct netdev_adjacent *iter; 8520 8521 struct net *net = dev_net(dev); 8522 8523 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8524 if (!net_eq(net, dev_net(iter->dev))) 8525 continue; 8526 netdev_adjacent_sysfs_del(iter->dev, oldname, 8527 &iter->dev->adj_list.lower); 8528 netdev_adjacent_sysfs_add(iter->dev, dev, 8529 &iter->dev->adj_list.lower); 8530 } 8531 8532 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8533 if (!net_eq(net, dev_net(iter->dev))) 8534 continue; 8535 netdev_adjacent_sysfs_del(iter->dev, oldname, 8536 &iter->dev->adj_list.upper); 8537 netdev_adjacent_sysfs_add(iter->dev, dev, 8538 &iter->dev->adj_list.upper); 8539 } 8540 } 8541 8542 void *netdev_lower_dev_get_private(struct net_device *dev, 8543 struct net_device *lower_dev) 8544 { 8545 struct netdev_adjacent *lower; 8546 8547 if (!lower_dev) 8548 return NULL; 8549 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); 8550 if (!lower) 8551 return NULL; 8552 8553 return lower->private; 8554 } 8555 EXPORT_SYMBOL(netdev_lower_dev_get_private); 8556 8557 8558 /** 8559 * netdev_lower_state_changed - Dispatch event about lower device state change 8560 * @lower_dev: device 8561 * @lower_state_info: state to dispatch 8562 * 8563 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. 8564 * The caller must hold the RTNL lock. 8565 */ 8566 void netdev_lower_state_changed(struct net_device *lower_dev, 8567 void *lower_state_info) 8568 { 8569 struct netdev_notifier_changelowerstate_info changelowerstate_info = { 8570 .info.dev = lower_dev, 8571 }; 8572 8573 ASSERT_RTNL(); 8574 changelowerstate_info.lower_state_info = lower_state_info; 8575 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, 8576 &changelowerstate_info.info); 8577 } 8578 EXPORT_SYMBOL(netdev_lower_state_changed); 8579 8580 static void dev_change_rx_flags(struct net_device *dev, int flags) 8581 { 8582 const struct net_device_ops *ops = dev->netdev_ops; 8583 8584 if (ops->ndo_change_rx_flags) 8585 ops->ndo_change_rx_flags(dev, flags); 8586 } 8587 8588 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) 8589 { 8590 unsigned int old_flags = dev->flags; 8591 kuid_t uid; 8592 kgid_t gid; 8593 8594 ASSERT_RTNL(); 8595 8596 dev->flags |= IFF_PROMISC; 8597 dev->promiscuity += inc; 8598 if (dev->promiscuity == 0) { 8599 /* 8600 * Avoid overflow. 8601 * If inc causes overflow, untouch promisc and return error. 8602 */ 8603 if (inc < 0) 8604 dev->flags &= ~IFF_PROMISC; 8605 else { 8606 dev->promiscuity -= inc; 8607 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n", 8608 dev->name); 8609 return -EOVERFLOW; 8610 } 8611 } 8612 if (dev->flags != old_flags) { 8613 pr_info("device %s %s promiscuous mode\n", 8614 dev->name, 8615 dev->flags & IFF_PROMISC ? "entered" : "left"); 8616 if (audit_enabled) { 8617 current_uid_gid(&uid, &gid); 8618 audit_log(audit_context(), GFP_ATOMIC, 8619 AUDIT_ANOM_PROMISCUOUS, 8620 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", 8621 dev->name, (dev->flags & IFF_PROMISC), 8622 (old_flags & IFF_PROMISC), 8623 from_kuid(&init_user_ns, audit_get_loginuid(current)), 8624 from_kuid(&init_user_ns, uid), 8625 from_kgid(&init_user_ns, gid), 8626 audit_get_sessionid(current)); 8627 } 8628 8629 dev_change_rx_flags(dev, IFF_PROMISC); 8630 } 8631 if (notify) 8632 __dev_notify_flags(dev, old_flags, IFF_PROMISC); 8633 return 0; 8634 } 8635 8636 /** 8637 * dev_set_promiscuity - update promiscuity count on a device 8638 * @dev: device 8639 * @inc: modifier 8640 * 8641 * Add or remove promiscuity from a device. While the count in the device 8642 * remains above zero the interface remains promiscuous. Once it hits zero 8643 * the device reverts back to normal filtering operation. A negative inc 8644 * value is used to drop promiscuity on the device. 8645 * Return 0 if successful or a negative errno code on error. 8646 */ 8647 int dev_set_promiscuity(struct net_device *dev, int inc) 8648 { 8649 unsigned int old_flags = dev->flags; 8650 int err; 8651 8652 err = __dev_set_promiscuity(dev, inc, true); 8653 if (err < 0) 8654 return err; 8655 if (dev->flags != old_flags) 8656 dev_set_rx_mode(dev); 8657 return err; 8658 } 8659 EXPORT_SYMBOL(dev_set_promiscuity); 8660 8661 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) 8662 { 8663 unsigned int old_flags = dev->flags, old_gflags = dev->gflags; 8664 8665 ASSERT_RTNL(); 8666 8667 dev->flags |= IFF_ALLMULTI; 8668 dev->allmulti += inc; 8669 if (dev->allmulti == 0) { 8670 /* 8671 * Avoid overflow. 8672 * If inc causes overflow, untouch allmulti and return error. 8673 */ 8674 if (inc < 0) 8675 dev->flags &= ~IFF_ALLMULTI; 8676 else { 8677 dev->allmulti -= inc; 8678 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n", 8679 dev->name); 8680 return -EOVERFLOW; 8681 } 8682 } 8683 if (dev->flags ^ old_flags) { 8684 dev_change_rx_flags(dev, IFF_ALLMULTI); 8685 dev_set_rx_mode(dev); 8686 if (notify) 8687 __dev_notify_flags(dev, old_flags, 8688 dev->gflags ^ old_gflags); 8689 } 8690 return 0; 8691 } 8692 8693 /** 8694 * dev_set_allmulti - update allmulti count on a device 8695 * @dev: device 8696 * @inc: modifier 8697 * 8698 * Add or remove reception of all multicast frames to a device. While the 8699 * count in the device remains above zero the interface remains listening 8700 * to all interfaces. Once it hits zero the device reverts back to normal 8701 * filtering operation. A negative @inc value is used to drop the counter 8702 * when releasing a resource needing all multicasts. 8703 * Return 0 if successful or a negative errno code on error. 8704 */ 8705 8706 int dev_set_allmulti(struct net_device *dev, int inc) 8707 { 8708 return __dev_set_allmulti(dev, inc, true); 8709 } 8710 EXPORT_SYMBOL(dev_set_allmulti); 8711 8712 /* 8713 * Upload unicast and multicast address lists to device and 8714 * configure RX filtering. When the device doesn't support unicast 8715 * filtering it is put in promiscuous mode while unicast addresses 8716 * are present. 8717 */ 8718 void __dev_set_rx_mode(struct net_device *dev) 8719 { 8720 const struct net_device_ops *ops = dev->netdev_ops; 8721 8722 /* dev_open will call this function so the list will stay sane. */ 8723 if (!(dev->flags&IFF_UP)) 8724 return; 8725 8726 if (!netif_device_present(dev)) 8727 return; 8728 8729 if (!(dev->priv_flags & IFF_UNICAST_FLT)) { 8730 /* Unicast addresses changes may only happen under the rtnl, 8731 * therefore calling __dev_set_promiscuity here is safe. 8732 */ 8733 if (!netdev_uc_empty(dev) && !dev->uc_promisc) { 8734 __dev_set_promiscuity(dev, 1, false); 8735 dev->uc_promisc = true; 8736 } else if (netdev_uc_empty(dev) && dev->uc_promisc) { 8737 __dev_set_promiscuity(dev, -1, false); 8738 dev->uc_promisc = false; 8739 } 8740 } 8741 8742 if (ops->ndo_set_rx_mode) 8743 ops->ndo_set_rx_mode(dev); 8744 } 8745 8746 void dev_set_rx_mode(struct net_device *dev) 8747 { 8748 netif_addr_lock_bh(dev); 8749 __dev_set_rx_mode(dev); 8750 netif_addr_unlock_bh(dev); 8751 } 8752 8753 /** 8754 * dev_get_flags - get flags reported to userspace 8755 * @dev: device 8756 * 8757 * Get the combination of flag bits exported through APIs to userspace. 8758 */ 8759 unsigned int dev_get_flags(const struct net_device *dev) 8760 { 8761 unsigned int flags; 8762 8763 flags = (dev->flags & ~(IFF_PROMISC | 8764 IFF_ALLMULTI | 8765 IFF_RUNNING | 8766 IFF_LOWER_UP | 8767 IFF_DORMANT)) | 8768 (dev->gflags & (IFF_PROMISC | 8769 IFF_ALLMULTI)); 8770 8771 if (netif_running(dev)) { 8772 if (netif_oper_up(dev)) 8773 flags |= IFF_RUNNING; 8774 if (netif_carrier_ok(dev)) 8775 flags |= IFF_LOWER_UP; 8776 if (netif_dormant(dev)) 8777 flags |= IFF_DORMANT; 8778 } 8779 8780 return flags; 8781 } 8782 EXPORT_SYMBOL(dev_get_flags); 8783 8784 int __dev_change_flags(struct net_device *dev, unsigned int flags, 8785 struct netlink_ext_ack *extack) 8786 { 8787 unsigned int old_flags = dev->flags; 8788 int ret; 8789 8790 ASSERT_RTNL(); 8791 8792 /* 8793 * Set the flags on our device. 8794 */ 8795 8796 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | 8797 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | 8798 IFF_AUTOMEDIA)) | 8799 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | 8800 IFF_ALLMULTI)); 8801 8802 /* 8803 * Load in the correct multicast list now the flags have changed. 8804 */ 8805 8806 if ((old_flags ^ flags) & IFF_MULTICAST) 8807 dev_change_rx_flags(dev, IFF_MULTICAST); 8808 8809 dev_set_rx_mode(dev); 8810 8811 /* 8812 * Have we downed the interface. We handle IFF_UP ourselves 8813 * according to user attempts to set it, rather than blindly 8814 * setting it. 8815 */ 8816 8817 ret = 0; 8818 if ((old_flags ^ flags) & IFF_UP) { 8819 if (old_flags & IFF_UP) 8820 __dev_close(dev); 8821 else 8822 ret = __dev_open(dev, extack); 8823 } 8824 8825 if ((flags ^ dev->gflags) & IFF_PROMISC) { 8826 int inc = (flags & IFF_PROMISC) ? 1 : -1; 8827 unsigned int old_flags = dev->flags; 8828 8829 dev->gflags ^= IFF_PROMISC; 8830 8831 if (__dev_set_promiscuity(dev, inc, false) >= 0) 8832 if (dev->flags != old_flags) 8833 dev_set_rx_mode(dev); 8834 } 8835 8836 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI 8837 * is important. Some (broken) drivers set IFF_PROMISC, when 8838 * IFF_ALLMULTI is requested not asking us and not reporting. 8839 */ 8840 if ((flags ^ dev->gflags) & IFF_ALLMULTI) { 8841 int inc = (flags & IFF_ALLMULTI) ? 1 : -1; 8842 8843 dev->gflags ^= IFF_ALLMULTI; 8844 __dev_set_allmulti(dev, inc, false); 8845 } 8846 8847 return ret; 8848 } 8849 8850 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, 8851 unsigned int gchanges) 8852 { 8853 unsigned int changes = dev->flags ^ old_flags; 8854 8855 if (gchanges) 8856 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC); 8857 8858 if (changes & IFF_UP) { 8859 if (dev->flags & IFF_UP) 8860 call_netdevice_notifiers(NETDEV_UP, dev); 8861 else 8862 call_netdevice_notifiers(NETDEV_DOWN, dev); 8863 } 8864 8865 if (dev->flags & IFF_UP && 8866 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { 8867 struct netdev_notifier_change_info change_info = { 8868 .info = { 8869 .dev = dev, 8870 }, 8871 .flags_changed = changes, 8872 }; 8873 8874 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info); 8875 } 8876 } 8877 8878 /** 8879 * dev_change_flags - change device settings 8880 * @dev: device 8881 * @flags: device state flags 8882 * @extack: netlink extended ack 8883 * 8884 * Change settings on device based state flags. The flags are 8885 * in the userspace exported format. 8886 */ 8887 int dev_change_flags(struct net_device *dev, unsigned int flags, 8888 struct netlink_ext_ack *extack) 8889 { 8890 int ret; 8891 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; 8892 8893 ret = __dev_change_flags(dev, flags, extack); 8894 if (ret < 0) 8895 return ret; 8896 8897 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); 8898 __dev_notify_flags(dev, old_flags, changes); 8899 return ret; 8900 } 8901 EXPORT_SYMBOL(dev_change_flags); 8902 8903 int __dev_set_mtu(struct net_device *dev, int new_mtu) 8904 { 8905 const struct net_device_ops *ops = dev->netdev_ops; 8906 8907 if (ops->ndo_change_mtu) 8908 return ops->ndo_change_mtu(dev, new_mtu); 8909 8910 /* Pairs with all the lockless reads of dev->mtu in the stack */ 8911 WRITE_ONCE(dev->mtu, new_mtu); 8912 return 0; 8913 } 8914 EXPORT_SYMBOL(__dev_set_mtu); 8915 8916 int dev_validate_mtu(struct net_device *dev, int new_mtu, 8917 struct netlink_ext_ack *extack) 8918 { 8919 /* MTU must be positive, and in range */ 8920 if (new_mtu < 0 || new_mtu < dev->min_mtu) { 8921 NL_SET_ERR_MSG(extack, "mtu less than device minimum"); 8922 return -EINVAL; 8923 } 8924 8925 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { 8926 NL_SET_ERR_MSG(extack, "mtu greater than device maximum"); 8927 return -EINVAL; 8928 } 8929 return 0; 8930 } 8931 8932 /** 8933 * dev_set_mtu_ext - Change maximum transfer unit 8934 * @dev: device 8935 * @new_mtu: new transfer unit 8936 * @extack: netlink extended ack 8937 * 8938 * Change the maximum transfer size of the network device. 8939 */ 8940 int dev_set_mtu_ext(struct net_device *dev, int new_mtu, 8941 struct netlink_ext_ack *extack) 8942 { 8943 int err, orig_mtu; 8944 8945 if (new_mtu == dev->mtu) 8946 return 0; 8947 8948 err = dev_validate_mtu(dev, new_mtu, extack); 8949 if (err) 8950 return err; 8951 8952 if (!netif_device_present(dev)) 8953 return -ENODEV; 8954 8955 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); 8956 err = notifier_to_errno(err); 8957 if (err) 8958 return err; 8959 8960 orig_mtu = dev->mtu; 8961 err = __dev_set_mtu(dev, new_mtu); 8962 8963 if (!err) { 8964 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8965 orig_mtu); 8966 err = notifier_to_errno(err); 8967 if (err) { 8968 /* setting mtu back and notifying everyone again, 8969 * so that they have a chance to revert changes. 8970 */ 8971 __dev_set_mtu(dev, orig_mtu); 8972 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8973 new_mtu); 8974 } 8975 } 8976 return err; 8977 } 8978 8979 int dev_set_mtu(struct net_device *dev, int new_mtu) 8980 { 8981 struct netlink_ext_ack extack; 8982 int err; 8983 8984 memset(&extack, 0, sizeof(extack)); 8985 err = dev_set_mtu_ext(dev, new_mtu, &extack); 8986 if (err && extack._msg) 8987 net_err_ratelimited("%s: %s\n", dev->name, extack._msg); 8988 return err; 8989 } 8990 EXPORT_SYMBOL(dev_set_mtu); 8991 8992 /** 8993 * dev_change_tx_queue_len - Change TX queue length of a netdevice 8994 * @dev: device 8995 * @new_len: new tx queue length 8996 */ 8997 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len) 8998 { 8999 unsigned int orig_len = dev->tx_queue_len; 9000 int res; 9001 9002 if (new_len != (unsigned int)new_len) 9003 return -ERANGE; 9004 9005 if (new_len != orig_len) { 9006 dev->tx_queue_len = new_len; 9007 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev); 9008 res = notifier_to_errno(res); 9009 if (res) 9010 goto err_rollback; 9011 res = dev_qdisc_change_tx_queue_len(dev); 9012 if (res) 9013 goto err_rollback; 9014 } 9015 9016 return 0; 9017 9018 err_rollback: 9019 netdev_err(dev, "refused to change device tx_queue_len\n"); 9020 dev->tx_queue_len = orig_len; 9021 return res; 9022 } 9023 9024 /** 9025 * dev_set_group - Change group this device belongs to 9026 * @dev: device 9027 * @new_group: group this device should belong to 9028 */ 9029 void dev_set_group(struct net_device *dev, int new_group) 9030 { 9031 dev->group = new_group; 9032 } 9033 EXPORT_SYMBOL(dev_set_group); 9034 9035 /** 9036 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR. 9037 * @dev: device 9038 * @addr: new address 9039 * @extack: netlink extended ack 9040 */ 9041 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr, 9042 struct netlink_ext_ack *extack) 9043 { 9044 struct netdev_notifier_pre_changeaddr_info info = { 9045 .info.dev = dev, 9046 .info.extack = extack, 9047 .dev_addr = addr, 9048 }; 9049 int rc; 9050 9051 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info); 9052 return notifier_to_errno(rc); 9053 } 9054 EXPORT_SYMBOL(dev_pre_changeaddr_notify); 9055 9056 /** 9057 * dev_set_mac_address - Change Media Access Control Address 9058 * @dev: device 9059 * @sa: new address 9060 * @extack: netlink extended ack 9061 * 9062 * Change the hardware (MAC) address of the device 9063 */ 9064 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa, 9065 struct netlink_ext_ack *extack) 9066 { 9067 const struct net_device_ops *ops = dev->netdev_ops; 9068 int err; 9069 9070 if (!ops->ndo_set_mac_address) 9071 return -EOPNOTSUPP; 9072 if (sa->sa_family != dev->type) 9073 return -EINVAL; 9074 if (!netif_device_present(dev)) 9075 return -ENODEV; 9076 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack); 9077 if (err) 9078 return err; 9079 err = ops->ndo_set_mac_address(dev, sa); 9080 if (err) 9081 return err; 9082 dev->addr_assign_type = NET_ADDR_SET; 9083 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); 9084 add_device_randomness(dev->dev_addr, dev->addr_len); 9085 return 0; 9086 } 9087 EXPORT_SYMBOL(dev_set_mac_address); 9088 9089 static DECLARE_RWSEM(dev_addr_sem); 9090 9091 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa, 9092 struct netlink_ext_ack *extack) 9093 { 9094 int ret; 9095 9096 down_write(&dev_addr_sem); 9097 ret = dev_set_mac_address(dev, sa, extack); 9098 up_write(&dev_addr_sem); 9099 return ret; 9100 } 9101 EXPORT_SYMBOL(dev_set_mac_address_user); 9102 9103 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name) 9104 { 9105 size_t size = sizeof(sa->sa_data); 9106 struct net_device *dev; 9107 int ret = 0; 9108 9109 down_read(&dev_addr_sem); 9110 rcu_read_lock(); 9111 9112 dev = dev_get_by_name_rcu(net, dev_name); 9113 if (!dev) { 9114 ret = -ENODEV; 9115 goto unlock; 9116 } 9117 if (!dev->addr_len) 9118 memset(sa->sa_data, 0, size); 9119 else 9120 memcpy(sa->sa_data, dev->dev_addr, 9121 min_t(size_t, size, dev->addr_len)); 9122 sa->sa_family = dev->type; 9123 9124 unlock: 9125 rcu_read_unlock(); 9126 up_read(&dev_addr_sem); 9127 return ret; 9128 } 9129 EXPORT_SYMBOL(dev_get_mac_address); 9130 9131 /** 9132 * dev_change_carrier - Change device carrier 9133 * @dev: device 9134 * @new_carrier: new value 9135 * 9136 * Change device carrier 9137 */ 9138 int dev_change_carrier(struct net_device *dev, bool new_carrier) 9139 { 9140 const struct net_device_ops *ops = dev->netdev_ops; 9141 9142 if (!ops->ndo_change_carrier) 9143 return -EOPNOTSUPP; 9144 if (!netif_device_present(dev)) 9145 return -ENODEV; 9146 return ops->ndo_change_carrier(dev, new_carrier); 9147 } 9148 EXPORT_SYMBOL(dev_change_carrier); 9149 9150 /** 9151 * dev_get_phys_port_id - Get device physical port ID 9152 * @dev: device 9153 * @ppid: port ID 9154 * 9155 * Get device physical port ID 9156 */ 9157 int dev_get_phys_port_id(struct net_device *dev, 9158 struct netdev_phys_item_id *ppid) 9159 { 9160 const struct net_device_ops *ops = dev->netdev_ops; 9161 9162 if (!ops->ndo_get_phys_port_id) 9163 return -EOPNOTSUPP; 9164 return ops->ndo_get_phys_port_id(dev, ppid); 9165 } 9166 EXPORT_SYMBOL(dev_get_phys_port_id); 9167 9168 /** 9169 * dev_get_phys_port_name - Get device physical port name 9170 * @dev: device 9171 * @name: port name 9172 * @len: limit of bytes to copy to name 9173 * 9174 * Get device physical port name 9175 */ 9176 int dev_get_phys_port_name(struct net_device *dev, 9177 char *name, size_t len) 9178 { 9179 const struct net_device_ops *ops = dev->netdev_ops; 9180 int err; 9181 9182 if (ops->ndo_get_phys_port_name) { 9183 err = ops->ndo_get_phys_port_name(dev, name, len); 9184 if (err != -EOPNOTSUPP) 9185 return err; 9186 } 9187 return devlink_compat_phys_port_name_get(dev, name, len); 9188 } 9189 EXPORT_SYMBOL(dev_get_phys_port_name); 9190 9191 /** 9192 * dev_get_port_parent_id - Get the device's port parent identifier 9193 * @dev: network device 9194 * @ppid: pointer to a storage for the port's parent identifier 9195 * @recurse: allow/disallow recursion to lower devices 9196 * 9197 * Get the devices's port parent identifier 9198 */ 9199 int dev_get_port_parent_id(struct net_device *dev, 9200 struct netdev_phys_item_id *ppid, 9201 bool recurse) 9202 { 9203 const struct net_device_ops *ops = dev->netdev_ops; 9204 struct netdev_phys_item_id first = { }; 9205 struct net_device *lower_dev; 9206 struct list_head *iter; 9207 int err; 9208 9209 if (ops->ndo_get_port_parent_id) { 9210 err = ops->ndo_get_port_parent_id(dev, ppid); 9211 if (err != -EOPNOTSUPP) 9212 return err; 9213 } 9214 9215 err = devlink_compat_switch_id_get(dev, ppid); 9216 if (!err || err != -EOPNOTSUPP) 9217 return err; 9218 9219 if (!recurse) 9220 return -EOPNOTSUPP; 9221 9222 netdev_for_each_lower_dev(dev, lower_dev, iter) { 9223 err = dev_get_port_parent_id(lower_dev, ppid, recurse); 9224 if (err) 9225 break; 9226 if (!first.id_len) 9227 first = *ppid; 9228 else if (memcmp(&first, ppid, sizeof(*ppid))) 9229 return -EOPNOTSUPP; 9230 } 9231 9232 return err; 9233 } 9234 EXPORT_SYMBOL(dev_get_port_parent_id); 9235 9236 /** 9237 * netdev_port_same_parent_id - Indicate if two network devices have 9238 * the same port parent identifier 9239 * @a: first network device 9240 * @b: second network device 9241 */ 9242 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b) 9243 { 9244 struct netdev_phys_item_id a_id = { }; 9245 struct netdev_phys_item_id b_id = { }; 9246 9247 if (dev_get_port_parent_id(a, &a_id, true) || 9248 dev_get_port_parent_id(b, &b_id, true)) 9249 return false; 9250 9251 return netdev_phys_item_id_same(&a_id, &b_id); 9252 } 9253 EXPORT_SYMBOL(netdev_port_same_parent_id); 9254 9255 /** 9256 * dev_change_proto_down - update protocol port state information 9257 * @dev: device 9258 * @proto_down: new value 9259 * 9260 * This info can be used by switch drivers to set the phys state of the 9261 * port. 9262 */ 9263 int dev_change_proto_down(struct net_device *dev, bool proto_down) 9264 { 9265 const struct net_device_ops *ops = dev->netdev_ops; 9266 9267 if (!ops->ndo_change_proto_down) 9268 return -EOPNOTSUPP; 9269 if (!netif_device_present(dev)) 9270 return -ENODEV; 9271 return ops->ndo_change_proto_down(dev, proto_down); 9272 } 9273 EXPORT_SYMBOL(dev_change_proto_down); 9274 9275 /** 9276 * dev_change_proto_down_generic - generic implementation for 9277 * ndo_change_proto_down that sets carrier according to 9278 * proto_down. 9279 * 9280 * @dev: device 9281 * @proto_down: new value 9282 */ 9283 int dev_change_proto_down_generic(struct net_device *dev, bool proto_down) 9284 { 9285 if (proto_down) 9286 netif_carrier_off(dev); 9287 else 9288 netif_carrier_on(dev); 9289 dev->proto_down = proto_down; 9290 return 0; 9291 } 9292 EXPORT_SYMBOL(dev_change_proto_down_generic); 9293 9294 /** 9295 * dev_change_proto_down_reason - proto down reason 9296 * 9297 * @dev: device 9298 * @mask: proto down mask 9299 * @value: proto down value 9300 */ 9301 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask, 9302 u32 value) 9303 { 9304 int b; 9305 9306 if (!mask) { 9307 dev->proto_down_reason = value; 9308 } else { 9309 for_each_set_bit(b, &mask, 32) { 9310 if (value & (1 << b)) 9311 dev->proto_down_reason |= BIT(b); 9312 else 9313 dev->proto_down_reason &= ~BIT(b); 9314 } 9315 } 9316 } 9317 EXPORT_SYMBOL(dev_change_proto_down_reason); 9318 9319 struct bpf_xdp_link { 9320 struct bpf_link link; 9321 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */ 9322 int flags; 9323 }; 9324 9325 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags) 9326 { 9327 if (flags & XDP_FLAGS_HW_MODE) 9328 return XDP_MODE_HW; 9329 if (flags & XDP_FLAGS_DRV_MODE) 9330 return XDP_MODE_DRV; 9331 if (flags & XDP_FLAGS_SKB_MODE) 9332 return XDP_MODE_SKB; 9333 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB; 9334 } 9335 9336 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode) 9337 { 9338 switch (mode) { 9339 case XDP_MODE_SKB: 9340 return generic_xdp_install; 9341 case XDP_MODE_DRV: 9342 case XDP_MODE_HW: 9343 return dev->netdev_ops->ndo_bpf; 9344 default: 9345 return NULL; 9346 } 9347 } 9348 9349 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev, 9350 enum bpf_xdp_mode mode) 9351 { 9352 return dev->xdp_state[mode].link; 9353 } 9354 9355 static struct bpf_prog *dev_xdp_prog(struct net_device *dev, 9356 enum bpf_xdp_mode mode) 9357 { 9358 struct bpf_xdp_link *link = dev_xdp_link(dev, mode); 9359 9360 if (link) 9361 return link->link.prog; 9362 return dev->xdp_state[mode].prog; 9363 } 9364 9365 static u8 dev_xdp_prog_count(struct net_device *dev) 9366 { 9367 u8 count = 0; 9368 int i; 9369 9370 for (i = 0; i < __MAX_XDP_MODE; i++) 9371 if (dev->xdp_state[i].prog || dev->xdp_state[i].link) 9372 count++; 9373 return count; 9374 } 9375 9376 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode) 9377 { 9378 struct bpf_prog *prog = dev_xdp_prog(dev, mode); 9379 9380 return prog ? prog->aux->id : 0; 9381 } 9382 9383 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode, 9384 struct bpf_xdp_link *link) 9385 { 9386 dev->xdp_state[mode].link = link; 9387 dev->xdp_state[mode].prog = NULL; 9388 } 9389 9390 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode, 9391 struct bpf_prog *prog) 9392 { 9393 dev->xdp_state[mode].link = NULL; 9394 dev->xdp_state[mode].prog = prog; 9395 } 9396 9397 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode, 9398 bpf_op_t bpf_op, struct netlink_ext_ack *extack, 9399 u32 flags, struct bpf_prog *prog) 9400 { 9401 struct netdev_bpf xdp; 9402 int err; 9403 9404 memset(&xdp, 0, sizeof(xdp)); 9405 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG; 9406 xdp.extack = extack; 9407 xdp.flags = flags; 9408 xdp.prog = prog; 9409 9410 /* Drivers assume refcnt is already incremented (i.e, prog pointer is 9411 * "moved" into driver), so they don't increment it on their own, but 9412 * they do decrement refcnt when program is detached or replaced. 9413 * Given net_device also owns link/prog, we need to bump refcnt here 9414 * to prevent drivers from underflowing it. 9415 */ 9416 if (prog) 9417 bpf_prog_inc(prog); 9418 err = bpf_op(dev, &xdp); 9419 if (err) { 9420 if (prog) 9421 bpf_prog_put(prog); 9422 return err; 9423 } 9424 9425 if (mode != XDP_MODE_HW) 9426 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog); 9427 9428 return 0; 9429 } 9430 9431 static void dev_xdp_uninstall(struct net_device *dev) 9432 { 9433 struct bpf_xdp_link *link; 9434 struct bpf_prog *prog; 9435 enum bpf_xdp_mode mode; 9436 bpf_op_t bpf_op; 9437 9438 ASSERT_RTNL(); 9439 9440 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) { 9441 prog = dev_xdp_prog(dev, mode); 9442 if (!prog) 9443 continue; 9444 9445 bpf_op = dev_xdp_bpf_op(dev, mode); 9446 if (!bpf_op) 9447 continue; 9448 9449 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9450 9451 /* auto-detach link from net device */ 9452 link = dev_xdp_link(dev, mode); 9453 if (link) 9454 link->dev = NULL; 9455 else 9456 bpf_prog_put(prog); 9457 9458 dev_xdp_set_link(dev, mode, NULL); 9459 } 9460 } 9461 9462 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack, 9463 struct bpf_xdp_link *link, struct bpf_prog *new_prog, 9464 struct bpf_prog *old_prog, u32 flags) 9465 { 9466 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES); 9467 struct bpf_prog *cur_prog; 9468 enum bpf_xdp_mode mode; 9469 bpf_op_t bpf_op; 9470 int err; 9471 9472 ASSERT_RTNL(); 9473 9474 /* either link or prog attachment, never both */ 9475 if (link && (new_prog || old_prog)) 9476 return -EINVAL; 9477 /* link supports only XDP mode flags */ 9478 if (link && (flags & ~XDP_FLAGS_MODES)) { 9479 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment"); 9480 return -EINVAL; 9481 } 9482 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */ 9483 if (num_modes > 1) { 9484 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set"); 9485 return -EINVAL; 9486 } 9487 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */ 9488 if (!num_modes && dev_xdp_prog_count(dev) > 1) { 9489 NL_SET_ERR_MSG(extack, 9490 "More than one program loaded, unset mode is ambiguous"); 9491 return -EINVAL; 9492 } 9493 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */ 9494 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) { 9495 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified"); 9496 return -EINVAL; 9497 } 9498 9499 mode = dev_xdp_mode(dev, flags); 9500 /* can't replace attached link */ 9501 if (dev_xdp_link(dev, mode)) { 9502 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link"); 9503 return -EBUSY; 9504 } 9505 9506 cur_prog = dev_xdp_prog(dev, mode); 9507 /* can't replace attached prog with link */ 9508 if (link && cur_prog) { 9509 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link"); 9510 return -EBUSY; 9511 } 9512 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) { 9513 NL_SET_ERR_MSG(extack, "Active program does not match expected"); 9514 return -EEXIST; 9515 } 9516 9517 /* put effective new program into new_prog */ 9518 if (link) 9519 new_prog = link->link.prog; 9520 9521 if (new_prog) { 9522 bool offload = mode == XDP_MODE_HW; 9523 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB 9524 ? XDP_MODE_DRV : XDP_MODE_SKB; 9525 9526 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) { 9527 NL_SET_ERR_MSG(extack, "XDP program already attached"); 9528 return -EBUSY; 9529 } 9530 if (!offload && dev_xdp_prog(dev, other_mode)) { 9531 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time"); 9532 return -EEXIST; 9533 } 9534 if (!offload && bpf_prog_is_dev_bound(new_prog->aux)) { 9535 NL_SET_ERR_MSG(extack, "Using device-bound program without HW_MODE flag is not supported"); 9536 return -EINVAL; 9537 } 9538 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) { 9539 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device"); 9540 return -EINVAL; 9541 } 9542 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) { 9543 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device"); 9544 return -EINVAL; 9545 } 9546 } 9547 9548 /* don't call drivers if the effective program didn't change */ 9549 if (new_prog != cur_prog) { 9550 bpf_op = dev_xdp_bpf_op(dev, mode); 9551 if (!bpf_op) { 9552 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode"); 9553 return -EOPNOTSUPP; 9554 } 9555 9556 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog); 9557 if (err) 9558 return err; 9559 } 9560 9561 if (link) 9562 dev_xdp_set_link(dev, mode, link); 9563 else 9564 dev_xdp_set_prog(dev, mode, new_prog); 9565 if (cur_prog) 9566 bpf_prog_put(cur_prog); 9567 9568 return 0; 9569 } 9570 9571 static int dev_xdp_attach_link(struct net_device *dev, 9572 struct netlink_ext_ack *extack, 9573 struct bpf_xdp_link *link) 9574 { 9575 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags); 9576 } 9577 9578 static int dev_xdp_detach_link(struct net_device *dev, 9579 struct netlink_ext_ack *extack, 9580 struct bpf_xdp_link *link) 9581 { 9582 enum bpf_xdp_mode mode; 9583 bpf_op_t bpf_op; 9584 9585 ASSERT_RTNL(); 9586 9587 mode = dev_xdp_mode(dev, link->flags); 9588 if (dev_xdp_link(dev, mode) != link) 9589 return -EINVAL; 9590 9591 bpf_op = dev_xdp_bpf_op(dev, mode); 9592 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9593 dev_xdp_set_link(dev, mode, NULL); 9594 return 0; 9595 } 9596 9597 static void bpf_xdp_link_release(struct bpf_link *link) 9598 { 9599 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9600 9601 rtnl_lock(); 9602 9603 /* if racing with net_device's tear down, xdp_link->dev might be 9604 * already NULL, in which case link was already auto-detached 9605 */ 9606 if (xdp_link->dev) { 9607 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link)); 9608 xdp_link->dev = NULL; 9609 } 9610 9611 rtnl_unlock(); 9612 } 9613 9614 static int bpf_xdp_link_detach(struct bpf_link *link) 9615 { 9616 bpf_xdp_link_release(link); 9617 return 0; 9618 } 9619 9620 static void bpf_xdp_link_dealloc(struct bpf_link *link) 9621 { 9622 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9623 9624 kfree(xdp_link); 9625 } 9626 9627 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link, 9628 struct seq_file *seq) 9629 { 9630 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9631 u32 ifindex = 0; 9632 9633 rtnl_lock(); 9634 if (xdp_link->dev) 9635 ifindex = xdp_link->dev->ifindex; 9636 rtnl_unlock(); 9637 9638 seq_printf(seq, "ifindex:\t%u\n", ifindex); 9639 } 9640 9641 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link, 9642 struct bpf_link_info *info) 9643 { 9644 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9645 u32 ifindex = 0; 9646 9647 rtnl_lock(); 9648 if (xdp_link->dev) 9649 ifindex = xdp_link->dev->ifindex; 9650 rtnl_unlock(); 9651 9652 info->xdp.ifindex = ifindex; 9653 return 0; 9654 } 9655 9656 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog, 9657 struct bpf_prog *old_prog) 9658 { 9659 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9660 enum bpf_xdp_mode mode; 9661 bpf_op_t bpf_op; 9662 int err = 0; 9663 9664 rtnl_lock(); 9665 9666 /* link might have been auto-released already, so fail */ 9667 if (!xdp_link->dev) { 9668 err = -ENOLINK; 9669 goto out_unlock; 9670 } 9671 9672 if (old_prog && link->prog != old_prog) { 9673 err = -EPERM; 9674 goto out_unlock; 9675 } 9676 old_prog = link->prog; 9677 if (old_prog == new_prog) { 9678 /* no-op, don't disturb drivers */ 9679 bpf_prog_put(new_prog); 9680 goto out_unlock; 9681 } 9682 9683 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags); 9684 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode); 9685 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL, 9686 xdp_link->flags, new_prog); 9687 if (err) 9688 goto out_unlock; 9689 9690 old_prog = xchg(&link->prog, new_prog); 9691 bpf_prog_put(old_prog); 9692 9693 out_unlock: 9694 rtnl_unlock(); 9695 return err; 9696 } 9697 9698 static const struct bpf_link_ops bpf_xdp_link_lops = { 9699 .release = bpf_xdp_link_release, 9700 .dealloc = bpf_xdp_link_dealloc, 9701 .detach = bpf_xdp_link_detach, 9702 .show_fdinfo = bpf_xdp_link_show_fdinfo, 9703 .fill_link_info = bpf_xdp_link_fill_link_info, 9704 .update_prog = bpf_xdp_link_update, 9705 }; 9706 9707 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog) 9708 { 9709 struct net *net = current->nsproxy->net_ns; 9710 struct bpf_link_primer link_primer; 9711 struct bpf_xdp_link *link; 9712 struct net_device *dev; 9713 int err, fd; 9714 9715 rtnl_lock(); 9716 dev = dev_get_by_index(net, attr->link_create.target_ifindex); 9717 if (!dev) { 9718 rtnl_unlock(); 9719 return -EINVAL; 9720 } 9721 9722 link = kzalloc(sizeof(*link), GFP_USER); 9723 if (!link) { 9724 err = -ENOMEM; 9725 goto unlock; 9726 } 9727 9728 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog); 9729 link->dev = dev; 9730 link->flags = attr->link_create.flags; 9731 9732 err = bpf_link_prime(&link->link, &link_primer); 9733 if (err) { 9734 kfree(link); 9735 goto unlock; 9736 } 9737 9738 err = dev_xdp_attach_link(dev, NULL, link); 9739 rtnl_unlock(); 9740 9741 if (err) { 9742 link->dev = NULL; 9743 bpf_link_cleanup(&link_primer); 9744 goto out_put_dev; 9745 } 9746 9747 fd = bpf_link_settle(&link_primer); 9748 /* link itself doesn't hold dev's refcnt to not complicate shutdown */ 9749 dev_put(dev); 9750 return fd; 9751 9752 unlock: 9753 rtnl_unlock(); 9754 9755 out_put_dev: 9756 dev_put(dev); 9757 return err; 9758 } 9759 9760 /** 9761 * dev_change_xdp_fd - set or clear a bpf program for a device rx path 9762 * @dev: device 9763 * @extack: netlink extended ack 9764 * @fd: new program fd or negative value to clear 9765 * @expected_fd: old program fd that userspace expects to replace or clear 9766 * @flags: xdp-related flags 9767 * 9768 * Set or clear a bpf program for a device 9769 */ 9770 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, 9771 int fd, int expected_fd, u32 flags) 9772 { 9773 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags); 9774 struct bpf_prog *new_prog = NULL, *old_prog = NULL; 9775 int err; 9776 9777 ASSERT_RTNL(); 9778 9779 if (fd >= 0) { 9780 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP, 9781 mode != XDP_MODE_SKB); 9782 if (IS_ERR(new_prog)) 9783 return PTR_ERR(new_prog); 9784 } 9785 9786 if (expected_fd >= 0) { 9787 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP, 9788 mode != XDP_MODE_SKB); 9789 if (IS_ERR(old_prog)) { 9790 err = PTR_ERR(old_prog); 9791 old_prog = NULL; 9792 goto err_out; 9793 } 9794 } 9795 9796 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags); 9797 9798 err_out: 9799 if (err && new_prog) 9800 bpf_prog_put(new_prog); 9801 if (old_prog) 9802 bpf_prog_put(old_prog); 9803 return err; 9804 } 9805 9806 /** 9807 * dev_new_index - allocate an ifindex 9808 * @net: the applicable net namespace 9809 * 9810 * Returns a suitable unique value for a new device interface 9811 * number. The caller must hold the rtnl semaphore or the 9812 * dev_base_lock to be sure it remains unique. 9813 */ 9814 static int dev_new_index(struct net *net) 9815 { 9816 int ifindex = net->ifindex; 9817 9818 for (;;) { 9819 if (++ifindex <= 0) 9820 ifindex = 1; 9821 if (!__dev_get_by_index(net, ifindex)) 9822 return net->ifindex = ifindex; 9823 } 9824 } 9825 9826 /* Delayed registration/unregisteration */ 9827 static LIST_HEAD(net_todo_list); 9828 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); 9829 9830 static void net_set_todo(struct net_device *dev) 9831 { 9832 list_add_tail(&dev->todo_list, &net_todo_list); 9833 dev_net(dev)->dev_unreg_count++; 9834 } 9835 9836 static netdev_features_t netdev_sync_upper_features(struct net_device *lower, 9837 struct net_device *upper, netdev_features_t features) 9838 { 9839 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9840 netdev_features_t feature; 9841 int feature_bit; 9842 9843 for_each_netdev_feature(upper_disables, feature_bit) { 9844 feature = __NETIF_F_BIT(feature_bit); 9845 if (!(upper->wanted_features & feature) 9846 && (features & feature)) { 9847 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", 9848 &feature, upper->name); 9849 features &= ~feature; 9850 } 9851 } 9852 9853 return features; 9854 } 9855 9856 static void netdev_sync_lower_features(struct net_device *upper, 9857 struct net_device *lower, netdev_features_t features) 9858 { 9859 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9860 netdev_features_t feature; 9861 int feature_bit; 9862 9863 for_each_netdev_feature(upper_disables, feature_bit) { 9864 feature = __NETIF_F_BIT(feature_bit); 9865 if (!(features & feature) && (lower->features & feature)) { 9866 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", 9867 &feature, lower->name); 9868 lower->wanted_features &= ~feature; 9869 __netdev_update_features(lower); 9870 9871 if (unlikely(lower->features & feature)) 9872 netdev_WARN(upper, "failed to disable %pNF on %s!\n", 9873 &feature, lower->name); 9874 else 9875 netdev_features_change(lower); 9876 } 9877 } 9878 } 9879 9880 static netdev_features_t netdev_fix_features(struct net_device *dev, 9881 netdev_features_t features) 9882 { 9883 /* Fix illegal checksum combinations */ 9884 if ((features & NETIF_F_HW_CSUM) && 9885 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { 9886 netdev_warn(dev, "mixed HW and IP checksum settings.\n"); 9887 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); 9888 } 9889 9890 /* TSO requires that SG is present as well. */ 9891 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { 9892 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); 9893 features &= ~NETIF_F_ALL_TSO; 9894 } 9895 9896 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && 9897 !(features & NETIF_F_IP_CSUM)) { 9898 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); 9899 features &= ~NETIF_F_TSO; 9900 features &= ~NETIF_F_TSO_ECN; 9901 } 9902 9903 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && 9904 !(features & NETIF_F_IPV6_CSUM)) { 9905 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); 9906 features &= ~NETIF_F_TSO6; 9907 } 9908 9909 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ 9910 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) 9911 features &= ~NETIF_F_TSO_MANGLEID; 9912 9913 /* TSO ECN requires that TSO is present as well. */ 9914 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) 9915 features &= ~NETIF_F_TSO_ECN; 9916 9917 /* Software GSO depends on SG. */ 9918 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { 9919 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); 9920 features &= ~NETIF_F_GSO; 9921 } 9922 9923 /* GSO partial features require GSO partial be set */ 9924 if ((features & dev->gso_partial_features) && 9925 !(features & NETIF_F_GSO_PARTIAL)) { 9926 netdev_dbg(dev, 9927 "Dropping partially supported GSO features since no GSO partial.\n"); 9928 features &= ~dev->gso_partial_features; 9929 } 9930 9931 if (!(features & NETIF_F_RXCSUM)) { 9932 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet 9933 * successfully merged by hardware must also have the 9934 * checksum verified by hardware. If the user does not 9935 * want to enable RXCSUM, logically, we should disable GRO_HW. 9936 */ 9937 if (features & NETIF_F_GRO_HW) { 9938 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n"); 9939 features &= ~NETIF_F_GRO_HW; 9940 } 9941 } 9942 9943 /* LRO/HW-GRO features cannot be combined with RX-FCS */ 9944 if (features & NETIF_F_RXFCS) { 9945 if (features & NETIF_F_LRO) { 9946 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n"); 9947 features &= ~NETIF_F_LRO; 9948 } 9949 9950 if (features & NETIF_F_GRO_HW) { 9951 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n"); 9952 features &= ~NETIF_F_GRO_HW; 9953 } 9954 } 9955 9956 if (features & NETIF_F_HW_TLS_TX) { 9957 bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) == 9958 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM); 9959 bool hw_csum = features & NETIF_F_HW_CSUM; 9960 9961 if (!ip_csum && !hw_csum) { 9962 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n"); 9963 features &= ~NETIF_F_HW_TLS_TX; 9964 } 9965 } 9966 9967 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) { 9968 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n"); 9969 features &= ~NETIF_F_HW_TLS_RX; 9970 } 9971 9972 return features; 9973 } 9974 9975 int __netdev_update_features(struct net_device *dev) 9976 { 9977 struct net_device *upper, *lower; 9978 netdev_features_t features; 9979 struct list_head *iter; 9980 int err = -1; 9981 9982 ASSERT_RTNL(); 9983 9984 features = netdev_get_wanted_features(dev); 9985 9986 if (dev->netdev_ops->ndo_fix_features) 9987 features = dev->netdev_ops->ndo_fix_features(dev, features); 9988 9989 /* driver might be less strict about feature dependencies */ 9990 features = netdev_fix_features(dev, features); 9991 9992 /* some features can't be enabled if they're off on an upper device */ 9993 netdev_for_each_upper_dev_rcu(dev, upper, iter) 9994 features = netdev_sync_upper_features(dev, upper, features); 9995 9996 if (dev->features == features) 9997 goto sync_lower; 9998 9999 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", 10000 &dev->features, &features); 10001 10002 if (dev->netdev_ops->ndo_set_features) 10003 err = dev->netdev_ops->ndo_set_features(dev, features); 10004 else 10005 err = 0; 10006 10007 if (unlikely(err < 0)) { 10008 netdev_err(dev, 10009 "set_features() failed (%d); wanted %pNF, left %pNF\n", 10010 err, &features, &dev->features); 10011 /* return non-0 since some features might have changed and 10012 * it's better to fire a spurious notification than miss it 10013 */ 10014 return -1; 10015 } 10016 10017 sync_lower: 10018 /* some features must be disabled on lower devices when disabled 10019 * on an upper device (think: bonding master or bridge) 10020 */ 10021 netdev_for_each_lower_dev(dev, lower, iter) 10022 netdev_sync_lower_features(dev, lower, features); 10023 10024 if (!err) { 10025 netdev_features_t diff = features ^ dev->features; 10026 10027 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) { 10028 /* udp_tunnel_{get,drop}_rx_info both need 10029 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the 10030 * device, or they won't do anything. 10031 * Thus we need to update dev->features 10032 * *before* calling udp_tunnel_get_rx_info, 10033 * but *after* calling udp_tunnel_drop_rx_info. 10034 */ 10035 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) { 10036 dev->features = features; 10037 udp_tunnel_get_rx_info(dev); 10038 } else { 10039 udp_tunnel_drop_rx_info(dev); 10040 } 10041 } 10042 10043 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) { 10044 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) { 10045 dev->features = features; 10046 err |= vlan_get_rx_ctag_filter_info(dev); 10047 } else { 10048 vlan_drop_rx_ctag_filter_info(dev); 10049 } 10050 } 10051 10052 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) { 10053 if (features & NETIF_F_HW_VLAN_STAG_FILTER) { 10054 dev->features = features; 10055 err |= vlan_get_rx_stag_filter_info(dev); 10056 } else { 10057 vlan_drop_rx_stag_filter_info(dev); 10058 } 10059 } 10060 10061 dev->features = features; 10062 } 10063 10064 return err < 0 ? 0 : 1; 10065 } 10066 10067 /** 10068 * netdev_update_features - recalculate device features 10069 * @dev: the device to check 10070 * 10071 * Recalculate dev->features set and send notifications if it 10072 * has changed. Should be called after driver or hardware dependent 10073 * conditions might have changed that influence the features. 10074 */ 10075 void netdev_update_features(struct net_device *dev) 10076 { 10077 if (__netdev_update_features(dev)) 10078 netdev_features_change(dev); 10079 } 10080 EXPORT_SYMBOL(netdev_update_features); 10081 10082 /** 10083 * netdev_change_features - recalculate device features 10084 * @dev: the device to check 10085 * 10086 * Recalculate dev->features set and send notifications even 10087 * if they have not changed. Should be called instead of 10088 * netdev_update_features() if also dev->vlan_features might 10089 * have changed to allow the changes to be propagated to stacked 10090 * VLAN devices. 10091 */ 10092 void netdev_change_features(struct net_device *dev) 10093 { 10094 __netdev_update_features(dev); 10095 netdev_features_change(dev); 10096 } 10097 EXPORT_SYMBOL(netdev_change_features); 10098 10099 /** 10100 * netif_stacked_transfer_operstate - transfer operstate 10101 * @rootdev: the root or lower level device to transfer state from 10102 * @dev: the device to transfer operstate to 10103 * 10104 * Transfer operational state from root to device. This is normally 10105 * called when a stacking relationship exists between the root 10106 * device and the device(a leaf device). 10107 */ 10108 void netif_stacked_transfer_operstate(const struct net_device *rootdev, 10109 struct net_device *dev) 10110 { 10111 if (rootdev->operstate == IF_OPER_DORMANT) 10112 netif_dormant_on(dev); 10113 else 10114 netif_dormant_off(dev); 10115 10116 if (rootdev->operstate == IF_OPER_TESTING) 10117 netif_testing_on(dev); 10118 else 10119 netif_testing_off(dev); 10120 10121 if (netif_carrier_ok(rootdev)) 10122 netif_carrier_on(dev); 10123 else 10124 netif_carrier_off(dev); 10125 } 10126 EXPORT_SYMBOL(netif_stacked_transfer_operstate); 10127 10128 static int netif_alloc_rx_queues(struct net_device *dev) 10129 { 10130 unsigned int i, count = dev->num_rx_queues; 10131 struct netdev_rx_queue *rx; 10132 size_t sz = count * sizeof(*rx); 10133 int err = 0; 10134 10135 BUG_ON(count < 1); 10136 10137 rx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL); 10138 if (!rx) 10139 return -ENOMEM; 10140 10141 dev->_rx = rx; 10142 10143 for (i = 0; i < count; i++) { 10144 rx[i].dev = dev; 10145 10146 /* XDP RX-queue setup */ 10147 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0); 10148 if (err < 0) 10149 goto err_rxq_info; 10150 } 10151 return 0; 10152 10153 err_rxq_info: 10154 /* Rollback successful reg's and free other resources */ 10155 while (i--) 10156 xdp_rxq_info_unreg(&rx[i].xdp_rxq); 10157 kvfree(dev->_rx); 10158 dev->_rx = NULL; 10159 return err; 10160 } 10161 10162 static void netif_free_rx_queues(struct net_device *dev) 10163 { 10164 unsigned int i, count = dev->num_rx_queues; 10165 10166 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */ 10167 if (!dev->_rx) 10168 return; 10169 10170 for (i = 0; i < count; i++) 10171 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq); 10172 10173 kvfree(dev->_rx); 10174 } 10175 10176 static void netdev_init_one_queue(struct net_device *dev, 10177 struct netdev_queue *queue, void *_unused) 10178 { 10179 /* Initialize queue lock */ 10180 spin_lock_init(&queue->_xmit_lock); 10181 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); 10182 queue->xmit_lock_owner = -1; 10183 netdev_queue_numa_node_write(queue, NUMA_NO_NODE); 10184 queue->dev = dev; 10185 #ifdef CONFIG_BQL 10186 dql_init(&queue->dql, HZ); 10187 #endif 10188 } 10189 10190 static void netif_free_tx_queues(struct net_device *dev) 10191 { 10192 kvfree(dev->_tx); 10193 } 10194 10195 static int netif_alloc_netdev_queues(struct net_device *dev) 10196 { 10197 unsigned int count = dev->num_tx_queues; 10198 struct netdev_queue *tx; 10199 size_t sz = count * sizeof(*tx); 10200 10201 if (count < 1 || count > 0xffff) 10202 return -EINVAL; 10203 10204 tx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL); 10205 if (!tx) 10206 return -ENOMEM; 10207 10208 dev->_tx = tx; 10209 10210 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); 10211 spin_lock_init(&dev->tx_global_lock); 10212 10213 return 0; 10214 } 10215 10216 void netif_tx_stop_all_queues(struct net_device *dev) 10217 { 10218 unsigned int i; 10219 10220 for (i = 0; i < dev->num_tx_queues; i++) { 10221 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 10222 10223 netif_tx_stop_queue(txq); 10224 } 10225 } 10226 EXPORT_SYMBOL(netif_tx_stop_all_queues); 10227 10228 /** 10229 * register_netdevice - register a network device 10230 * @dev: device to register 10231 * 10232 * Take a completed network device structure and add it to the kernel 10233 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 10234 * chain. 0 is returned on success. A negative errno code is returned 10235 * on a failure to set up the device, or if the name is a duplicate. 10236 * 10237 * Callers must hold the rtnl semaphore. You may want 10238 * register_netdev() instead of this. 10239 * 10240 * BUGS: 10241 * The locking appears insufficient to guarantee two parallel registers 10242 * will not get the same name. 10243 */ 10244 10245 int register_netdevice(struct net_device *dev) 10246 { 10247 int ret; 10248 struct net *net = dev_net(dev); 10249 10250 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE < 10251 NETDEV_FEATURE_COUNT); 10252 BUG_ON(dev_boot_phase); 10253 ASSERT_RTNL(); 10254 10255 might_sleep(); 10256 10257 /* When net_device's are persistent, this will be fatal. */ 10258 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); 10259 BUG_ON(!net); 10260 10261 ret = ethtool_check_ops(dev->ethtool_ops); 10262 if (ret) 10263 return ret; 10264 10265 spin_lock_init(&dev->addr_list_lock); 10266 netdev_set_addr_lockdep_class(dev); 10267 10268 ret = dev_get_valid_name(net, dev, dev->name); 10269 if (ret < 0) 10270 goto out; 10271 10272 ret = -ENOMEM; 10273 dev->name_node = netdev_name_node_head_alloc(dev); 10274 if (!dev->name_node) 10275 goto out; 10276 10277 /* Init, if this function is available */ 10278 if (dev->netdev_ops->ndo_init) { 10279 ret = dev->netdev_ops->ndo_init(dev); 10280 if (ret) { 10281 if (ret > 0) 10282 ret = -EIO; 10283 goto err_free_name; 10284 } 10285 } 10286 10287 if (((dev->hw_features | dev->features) & 10288 NETIF_F_HW_VLAN_CTAG_FILTER) && 10289 (!dev->netdev_ops->ndo_vlan_rx_add_vid || 10290 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { 10291 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); 10292 ret = -EINVAL; 10293 goto err_uninit; 10294 } 10295 10296 ret = -EBUSY; 10297 if (!dev->ifindex) 10298 dev->ifindex = dev_new_index(net); 10299 else if (__dev_get_by_index(net, dev->ifindex)) 10300 goto err_uninit; 10301 10302 /* Transfer changeable features to wanted_features and enable 10303 * software offloads (GSO and GRO). 10304 */ 10305 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF); 10306 dev->features |= NETIF_F_SOFT_FEATURES; 10307 10308 if (dev->udp_tunnel_nic_info) { 10309 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10310 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10311 } 10312 10313 dev->wanted_features = dev->features & dev->hw_features; 10314 10315 if (!(dev->flags & IFF_LOOPBACK)) 10316 dev->hw_features |= NETIF_F_NOCACHE_COPY; 10317 10318 /* If IPv4 TCP segmentation offload is supported we should also 10319 * allow the device to enable segmenting the frame with the option 10320 * of ignoring a static IP ID value. This doesn't enable the 10321 * feature itself but allows the user to enable it later. 10322 */ 10323 if (dev->hw_features & NETIF_F_TSO) 10324 dev->hw_features |= NETIF_F_TSO_MANGLEID; 10325 if (dev->vlan_features & NETIF_F_TSO) 10326 dev->vlan_features |= NETIF_F_TSO_MANGLEID; 10327 if (dev->mpls_features & NETIF_F_TSO) 10328 dev->mpls_features |= NETIF_F_TSO_MANGLEID; 10329 if (dev->hw_enc_features & NETIF_F_TSO) 10330 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; 10331 10332 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. 10333 */ 10334 dev->vlan_features |= NETIF_F_HIGHDMA; 10335 10336 /* Make NETIF_F_SG inheritable to tunnel devices. 10337 */ 10338 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; 10339 10340 /* Make NETIF_F_SG inheritable to MPLS. 10341 */ 10342 dev->mpls_features |= NETIF_F_SG; 10343 10344 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); 10345 ret = notifier_to_errno(ret); 10346 if (ret) 10347 goto err_uninit; 10348 10349 ret = netdev_register_kobject(dev); 10350 if (ret) { 10351 dev->reg_state = NETREG_UNREGISTERED; 10352 goto err_uninit; 10353 } 10354 dev->reg_state = NETREG_REGISTERED; 10355 10356 __netdev_update_features(dev); 10357 10358 /* 10359 * Default initial state at registry is that the 10360 * device is present. 10361 */ 10362 10363 set_bit(__LINK_STATE_PRESENT, &dev->state); 10364 10365 linkwatch_init_dev(dev); 10366 10367 dev_init_scheduler(dev); 10368 dev_hold(dev); 10369 list_netdevice(dev); 10370 add_device_randomness(dev->dev_addr, dev->addr_len); 10371 10372 /* If the device has permanent device address, driver should 10373 * set dev_addr and also addr_assign_type should be set to 10374 * NET_ADDR_PERM (default value). 10375 */ 10376 if (dev->addr_assign_type == NET_ADDR_PERM) 10377 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); 10378 10379 /* Notify protocols, that a new device appeared. */ 10380 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); 10381 ret = notifier_to_errno(ret); 10382 if (ret) { 10383 /* Expect explicit free_netdev() on failure */ 10384 dev->needs_free_netdev = false; 10385 unregister_netdevice_queue(dev, NULL); 10386 goto out; 10387 } 10388 /* 10389 * Prevent userspace races by waiting until the network 10390 * device is fully setup before sending notifications. 10391 */ 10392 if (!dev->rtnl_link_ops || 10393 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 10394 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); 10395 10396 out: 10397 return ret; 10398 10399 err_uninit: 10400 if (dev->netdev_ops->ndo_uninit) 10401 dev->netdev_ops->ndo_uninit(dev); 10402 if (dev->priv_destructor) 10403 dev->priv_destructor(dev); 10404 err_free_name: 10405 netdev_name_node_free(dev->name_node); 10406 goto out; 10407 } 10408 EXPORT_SYMBOL(register_netdevice); 10409 10410 /** 10411 * init_dummy_netdev - init a dummy network device for NAPI 10412 * @dev: device to init 10413 * 10414 * This takes a network device structure and initialize the minimum 10415 * amount of fields so it can be used to schedule NAPI polls without 10416 * registering a full blown interface. This is to be used by drivers 10417 * that need to tie several hardware interfaces to a single NAPI 10418 * poll scheduler due to HW limitations. 10419 */ 10420 int init_dummy_netdev(struct net_device *dev) 10421 { 10422 /* Clear everything. Note we don't initialize spinlocks 10423 * are they aren't supposed to be taken by any of the 10424 * NAPI code and this dummy netdev is supposed to be 10425 * only ever used for NAPI polls 10426 */ 10427 memset(dev, 0, sizeof(struct net_device)); 10428 10429 /* make sure we BUG if trying to hit standard 10430 * register/unregister code path 10431 */ 10432 dev->reg_state = NETREG_DUMMY; 10433 10434 /* NAPI wants this */ 10435 INIT_LIST_HEAD(&dev->napi_list); 10436 10437 /* a dummy interface is started by default */ 10438 set_bit(__LINK_STATE_PRESENT, &dev->state); 10439 set_bit(__LINK_STATE_START, &dev->state); 10440 10441 /* napi_busy_loop stats accounting wants this */ 10442 dev_net_set(dev, &init_net); 10443 10444 /* Note : We dont allocate pcpu_refcnt for dummy devices, 10445 * because users of this 'device' dont need to change 10446 * its refcount. 10447 */ 10448 10449 return 0; 10450 } 10451 EXPORT_SYMBOL_GPL(init_dummy_netdev); 10452 10453 10454 /** 10455 * register_netdev - register a network device 10456 * @dev: device to register 10457 * 10458 * Take a completed network device structure and add it to the kernel 10459 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 10460 * chain. 0 is returned on success. A negative errno code is returned 10461 * on a failure to set up the device, or if the name is a duplicate. 10462 * 10463 * This is a wrapper around register_netdevice that takes the rtnl semaphore 10464 * and expands the device name if you passed a format string to 10465 * alloc_netdev. 10466 */ 10467 int register_netdev(struct net_device *dev) 10468 { 10469 int err; 10470 10471 if (rtnl_lock_killable()) 10472 return -EINTR; 10473 err = register_netdevice(dev); 10474 rtnl_unlock(); 10475 return err; 10476 } 10477 EXPORT_SYMBOL(register_netdev); 10478 10479 int netdev_refcnt_read(const struct net_device *dev) 10480 { 10481 #ifdef CONFIG_PCPU_DEV_REFCNT 10482 int i, refcnt = 0; 10483 10484 for_each_possible_cpu(i) 10485 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); 10486 return refcnt; 10487 #else 10488 return refcount_read(&dev->dev_refcnt); 10489 #endif 10490 } 10491 EXPORT_SYMBOL(netdev_refcnt_read); 10492 10493 int netdev_unregister_timeout_secs __read_mostly = 10; 10494 10495 #define WAIT_REFS_MIN_MSECS 1 10496 #define WAIT_REFS_MAX_MSECS 250 10497 /** 10498 * netdev_wait_allrefs - wait until all references are gone. 10499 * @dev: target net_device 10500 * 10501 * This is called when unregistering network devices. 10502 * 10503 * Any protocol or device that holds a reference should register 10504 * for netdevice notification, and cleanup and put back the 10505 * reference if they receive an UNREGISTER event. 10506 * We can get stuck here if buggy protocols don't correctly 10507 * call dev_put. 10508 */ 10509 static void netdev_wait_allrefs(struct net_device *dev) 10510 { 10511 unsigned long rebroadcast_time, warning_time; 10512 int wait = 0, refcnt; 10513 10514 linkwatch_forget_dev(dev); 10515 10516 rebroadcast_time = warning_time = jiffies; 10517 refcnt = netdev_refcnt_read(dev); 10518 10519 while (refcnt != 1) { 10520 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { 10521 rtnl_lock(); 10522 10523 /* Rebroadcast unregister notification */ 10524 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10525 10526 __rtnl_unlock(); 10527 rcu_barrier(); 10528 rtnl_lock(); 10529 10530 if (test_bit(__LINK_STATE_LINKWATCH_PENDING, 10531 &dev->state)) { 10532 /* We must not have linkwatch events 10533 * pending on unregister. If this 10534 * happens, we simply run the queue 10535 * unscheduled, resulting in a noop 10536 * for this device. 10537 */ 10538 linkwatch_run_queue(); 10539 } 10540 10541 __rtnl_unlock(); 10542 10543 rebroadcast_time = jiffies; 10544 } 10545 10546 if (!wait) { 10547 rcu_barrier(); 10548 wait = WAIT_REFS_MIN_MSECS; 10549 } else { 10550 msleep(wait); 10551 wait = min(wait << 1, WAIT_REFS_MAX_MSECS); 10552 } 10553 10554 refcnt = netdev_refcnt_read(dev); 10555 10556 if (refcnt != 1 && 10557 time_after(jiffies, warning_time + 10558 netdev_unregister_timeout_secs * HZ)) { 10559 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", 10560 dev->name, refcnt); 10561 warning_time = jiffies; 10562 } 10563 } 10564 } 10565 10566 /* The sequence is: 10567 * 10568 * rtnl_lock(); 10569 * ... 10570 * register_netdevice(x1); 10571 * register_netdevice(x2); 10572 * ... 10573 * unregister_netdevice(y1); 10574 * unregister_netdevice(y2); 10575 * ... 10576 * rtnl_unlock(); 10577 * free_netdev(y1); 10578 * free_netdev(y2); 10579 * 10580 * We are invoked by rtnl_unlock(). 10581 * This allows us to deal with problems: 10582 * 1) We can delete sysfs objects which invoke hotplug 10583 * without deadlocking with linkwatch via keventd. 10584 * 2) Since we run with the RTNL semaphore not held, we can sleep 10585 * safely in order to wait for the netdev refcnt to drop to zero. 10586 * 10587 * We must not return until all unregister events added during 10588 * the interval the lock was held have been completed. 10589 */ 10590 void netdev_run_todo(void) 10591 { 10592 struct list_head list; 10593 #ifdef CONFIG_LOCKDEP 10594 struct list_head unlink_list; 10595 10596 list_replace_init(&net_unlink_list, &unlink_list); 10597 10598 while (!list_empty(&unlink_list)) { 10599 struct net_device *dev = list_first_entry(&unlink_list, 10600 struct net_device, 10601 unlink_list); 10602 list_del_init(&dev->unlink_list); 10603 dev->nested_level = dev->lower_level - 1; 10604 } 10605 #endif 10606 10607 /* Snapshot list, allow later requests */ 10608 list_replace_init(&net_todo_list, &list); 10609 10610 __rtnl_unlock(); 10611 10612 10613 /* Wait for rcu callbacks to finish before next phase */ 10614 if (!list_empty(&list)) 10615 rcu_barrier(); 10616 10617 while (!list_empty(&list)) { 10618 struct net_device *dev 10619 = list_first_entry(&list, struct net_device, todo_list); 10620 list_del(&dev->todo_list); 10621 10622 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { 10623 pr_err("network todo '%s' but state %d\n", 10624 dev->name, dev->reg_state); 10625 dump_stack(); 10626 continue; 10627 } 10628 10629 dev->reg_state = NETREG_UNREGISTERED; 10630 10631 netdev_wait_allrefs(dev); 10632 10633 /* paranoia */ 10634 BUG_ON(netdev_refcnt_read(dev) != 1); 10635 BUG_ON(!list_empty(&dev->ptype_all)); 10636 BUG_ON(!list_empty(&dev->ptype_specific)); 10637 WARN_ON(rcu_access_pointer(dev->ip_ptr)); 10638 WARN_ON(rcu_access_pointer(dev->ip6_ptr)); 10639 #if IS_ENABLED(CONFIG_DECNET) 10640 WARN_ON(dev->dn_ptr); 10641 #endif 10642 if (dev->priv_destructor) 10643 dev->priv_destructor(dev); 10644 if (dev->needs_free_netdev) 10645 free_netdev(dev); 10646 10647 /* Report a network device has been unregistered */ 10648 rtnl_lock(); 10649 dev_net(dev)->dev_unreg_count--; 10650 __rtnl_unlock(); 10651 wake_up(&netdev_unregistering_wq); 10652 10653 /* Free network device */ 10654 kobject_put(&dev->dev.kobj); 10655 } 10656 } 10657 10658 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has 10659 * all the same fields in the same order as net_device_stats, with only 10660 * the type differing, but rtnl_link_stats64 may have additional fields 10661 * at the end for newer counters. 10662 */ 10663 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, 10664 const struct net_device_stats *netdev_stats) 10665 { 10666 #if BITS_PER_LONG == 64 10667 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats)); 10668 memcpy(stats64, netdev_stats, sizeof(*netdev_stats)); 10669 /* zero out counters that only exist in rtnl_link_stats64 */ 10670 memset((char *)stats64 + sizeof(*netdev_stats), 0, 10671 sizeof(*stats64) - sizeof(*netdev_stats)); 10672 #else 10673 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long); 10674 const unsigned long *src = (const unsigned long *)netdev_stats; 10675 u64 *dst = (u64 *)stats64; 10676 10677 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); 10678 for (i = 0; i < n; i++) 10679 dst[i] = src[i]; 10680 /* zero out counters that only exist in rtnl_link_stats64 */ 10681 memset((char *)stats64 + n * sizeof(u64), 0, 10682 sizeof(*stats64) - n * sizeof(u64)); 10683 #endif 10684 } 10685 EXPORT_SYMBOL(netdev_stats_to_stats64); 10686 10687 /** 10688 * dev_get_stats - get network device statistics 10689 * @dev: device to get statistics from 10690 * @storage: place to store stats 10691 * 10692 * Get network statistics from device. Return @storage. 10693 * The device driver may provide its own method by setting 10694 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; 10695 * otherwise the internal statistics structure is used. 10696 */ 10697 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, 10698 struct rtnl_link_stats64 *storage) 10699 { 10700 const struct net_device_ops *ops = dev->netdev_ops; 10701 10702 if (ops->ndo_get_stats64) { 10703 memset(storage, 0, sizeof(*storage)); 10704 ops->ndo_get_stats64(dev, storage); 10705 } else if (ops->ndo_get_stats) { 10706 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); 10707 } else { 10708 netdev_stats_to_stats64(storage, &dev->stats); 10709 } 10710 storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped); 10711 storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped); 10712 storage->rx_nohandler += (unsigned long)atomic_long_read(&dev->rx_nohandler); 10713 return storage; 10714 } 10715 EXPORT_SYMBOL(dev_get_stats); 10716 10717 /** 10718 * dev_fetch_sw_netstats - get per-cpu network device statistics 10719 * @s: place to store stats 10720 * @netstats: per-cpu network stats to read from 10721 * 10722 * Read per-cpu network statistics and populate the related fields in @s. 10723 */ 10724 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s, 10725 const struct pcpu_sw_netstats __percpu *netstats) 10726 { 10727 int cpu; 10728 10729 for_each_possible_cpu(cpu) { 10730 const struct pcpu_sw_netstats *stats; 10731 struct pcpu_sw_netstats tmp; 10732 unsigned int start; 10733 10734 stats = per_cpu_ptr(netstats, cpu); 10735 do { 10736 start = u64_stats_fetch_begin_irq(&stats->syncp); 10737 tmp.rx_packets = stats->rx_packets; 10738 tmp.rx_bytes = stats->rx_bytes; 10739 tmp.tx_packets = stats->tx_packets; 10740 tmp.tx_bytes = stats->tx_bytes; 10741 } while (u64_stats_fetch_retry_irq(&stats->syncp, start)); 10742 10743 s->rx_packets += tmp.rx_packets; 10744 s->rx_bytes += tmp.rx_bytes; 10745 s->tx_packets += tmp.tx_packets; 10746 s->tx_bytes += tmp.tx_bytes; 10747 } 10748 } 10749 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats); 10750 10751 /** 10752 * dev_get_tstats64 - ndo_get_stats64 implementation 10753 * @dev: device to get statistics from 10754 * @s: place to store stats 10755 * 10756 * Populate @s from dev->stats and dev->tstats. Can be used as 10757 * ndo_get_stats64() callback. 10758 */ 10759 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s) 10760 { 10761 netdev_stats_to_stats64(s, &dev->stats); 10762 dev_fetch_sw_netstats(s, dev->tstats); 10763 } 10764 EXPORT_SYMBOL_GPL(dev_get_tstats64); 10765 10766 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) 10767 { 10768 struct netdev_queue *queue = dev_ingress_queue(dev); 10769 10770 #ifdef CONFIG_NET_CLS_ACT 10771 if (queue) 10772 return queue; 10773 queue = kzalloc(sizeof(*queue), GFP_KERNEL); 10774 if (!queue) 10775 return NULL; 10776 netdev_init_one_queue(dev, queue, NULL); 10777 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); 10778 queue->qdisc_sleeping = &noop_qdisc; 10779 rcu_assign_pointer(dev->ingress_queue, queue); 10780 #endif 10781 return queue; 10782 } 10783 10784 static const struct ethtool_ops default_ethtool_ops; 10785 10786 void netdev_set_default_ethtool_ops(struct net_device *dev, 10787 const struct ethtool_ops *ops) 10788 { 10789 if (dev->ethtool_ops == &default_ethtool_ops) 10790 dev->ethtool_ops = ops; 10791 } 10792 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); 10793 10794 void netdev_freemem(struct net_device *dev) 10795 { 10796 char *addr = (char *)dev - dev->padded; 10797 10798 kvfree(addr); 10799 } 10800 10801 /** 10802 * alloc_netdev_mqs - allocate network device 10803 * @sizeof_priv: size of private data to allocate space for 10804 * @name: device name format string 10805 * @name_assign_type: origin of device name 10806 * @setup: callback to initialize device 10807 * @txqs: the number of TX subqueues to allocate 10808 * @rxqs: the number of RX subqueues to allocate 10809 * 10810 * Allocates a struct net_device with private data area for driver use 10811 * and performs basic initialization. Also allocates subqueue structs 10812 * for each queue on the device. 10813 */ 10814 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, 10815 unsigned char name_assign_type, 10816 void (*setup)(struct net_device *), 10817 unsigned int txqs, unsigned int rxqs) 10818 { 10819 struct net_device *dev; 10820 unsigned int alloc_size; 10821 struct net_device *p; 10822 10823 BUG_ON(strlen(name) >= sizeof(dev->name)); 10824 10825 if (txqs < 1) { 10826 pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); 10827 return NULL; 10828 } 10829 10830 if (rxqs < 1) { 10831 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); 10832 return NULL; 10833 } 10834 10835 alloc_size = sizeof(struct net_device); 10836 if (sizeof_priv) { 10837 /* ensure 32-byte alignment of private area */ 10838 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); 10839 alloc_size += sizeof_priv; 10840 } 10841 /* ensure 32-byte alignment of whole construct */ 10842 alloc_size += NETDEV_ALIGN - 1; 10843 10844 p = kvzalloc(alloc_size, GFP_KERNEL | __GFP_RETRY_MAYFAIL); 10845 if (!p) 10846 return NULL; 10847 10848 dev = PTR_ALIGN(p, NETDEV_ALIGN); 10849 dev->padded = (char *)dev - (char *)p; 10850 10851 #ifdef CONFIG_PCPU_DEV_REFCNT 10852 dev->pcpu_refcnt = alloc_percpu(int); 10853 if (!dev->pcpu_refcnt) 10854 goto free_dev; 10855 dev_hold(dev); 10856 #else 10857 refcount_set(&dev->dev_refcnt, 1); 10858 #endif 10859 10860 if (dev_addr_init(dev)) 10861 goto free_pcpu; 10862 10863 dev_mc_init(dev); 10864 dev_uc_init(dev); 10865 10866 dev_net_set(dev, &init_net); 10867 10868 dev->gso_max_size = GSO_MAX_SIZE; 10869 dev->gso_max_segs = GSO_MAX_SEGS; 10870 dev->upper_level = 1; 10871 dev->lower_level = 1; 10872 #ifdef CONFIG_LOCKDEP 10873 dev->nested_level = 0; 10874 INIT_LIST_HEAD(&dev->unlink_list); 10875 #endif 10876 10877 INIT_LIST_HEAD(&dev->napi_list); 10878 INIT_LIST_HEAD(&dev->unreg_list); 10879 INIT_LIST_HEAD(&dev->close_list); 10880 INIT_LIST_HEAD(&dev->link_watch_list); 10881 INIT_LIST_HEAD(&dev->adj_list.upper); 10882 INIT_LIST_HEAD(&dev->adj_list.lower); 10883 INIT_LIST_HEAD(&dev->ptype_all); 10884 INIT_LIST_HEAD(&dev->ptype_specific); 10885 INIT_LIST_HEAD(&dev->net_notifier_list); 10886 #ifdef CONFIG_NET_SCHED 10887 hash_init(dev->qdisc_hash); 10888 #endif 10889 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; 10890 setup(dev); 10891 10892 if (!dev->tx_queue_len) { 10893 dev->priv_flags |= IFF_NO_QUEUE; 10894 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; 10895 } 10896 10897 dev->num_tx_queues = txqs; 10898 dev->real_num_tx_queues = txqs; 10899 if (netif_alloc_netdev_queues(dev)) 10900 goto free_all; 10901 10902 dev->num_rx_queues = rxqs; 10903 dev->real_num_rx_queues = rxqs; 10904 if (netif_alloc_rx_queues(dev)) 10905 goto free_all; 10906 10907 strcpy(dev->name, name); 10908 dev->name_assign_type = name_assign_type; 10909 dev->group = INIT_NETDEV_GROUP; 10910 if (!dev->ethtool_ops) 10911 dev->ethtool_ops = &default_ethtool_ops; 10912 10913 nf_hook_ingress_init(dev); 10914 10915 return dev; 10916 10917 free_all: 10918 free_netdev(dev); 10919 return NULL; 10920 10921 free_pcpu: 10922 #ifdef CONFIG_PCPU_DEV_REFCNT 10923 free_percpu(dev->pcpu_refcnt); 10924 free_dev: 10925 #endif 10926 netdev_freemem(dev); 10927 return NULL; 10928 } 10929 EXPORT_SYMBOL(alloc_netdev_mqs); 10930 10931 /** 10932 * free_netdev - free network device 10933 * @dev: device 10934 * 10935 * This function does the last stage of destroying an allocated device 10936 * interface. The reference to the device object is released. If this 10937 * is the last reference then it will be freed.Must be called in process 10938 * context. 10939 */ 10940 void free_netdev(struct net_device *dev) 10941 { 10942 struct napi_struct *p, *n; 10943 10944 might_sleep(); 10945 10946 /* When called immediately after register_netdevice() failed the unwind 10947 * handling may still be dismantling the device. Handle that case by 10948 * deferring the free. 10949 */ 10950 if (dev->reg_state == NETREG_UNREGISTERING) { 10951 ASSERT_RTNL(); 10952 dev->needs_free_netdev = true; 10953 return; 10954 } 10955 10956 netif_free_tx_queues(dev); 10957 netif_free_rx_queues(dev); 10958 10959 kfree(rcu_dereference_protected(dev->ingress_queue, 1)); 10960 10961 /* Flush device addresses */ 10962 dev_addr_flush(dev); 10963 10964 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) 10965 netif_napi_del(p); 10966 10967 #ifdef CONFIG_PCPU_DEV_REFCNT 10968 free_percpu(dev->pcpu_refcnt); 10969 dev->pcpu_refcnt = NULL; 10970 #endif 10971 free_percpu(dev->xdp_bulkq); 10972 dev->xdp_bulkq = NULL; 10973 10974 /* Compatibility with error handling in drivers */ 10975 if (dev->reg_state == NETREG_UNINITIALIZED) { 10976 netdev_freemem(dev); 10977 return; 10978 } 10979 10980 BUG_ON(dev->reg_state != NETREG_UNREGISTERED); 10981 dev->reg_state = NETREG_RELEASED; 10982 10983 /* will free via device release */ 10984 put_device(&dev->dev); 10985 } 10986 EXPORT_SYMBOL(free_netdev); 10987 10988 /** 10989 * synchronize_net - Synchronize with packet receive processing 10990 * 10991 * Wait for packets currently being received to be done. 10992 * Does not block later packets from starting. 10993 */ 10994 void synchronize_net(void) 10995 { 10996 might_sleep(); 10997 if (rtnl_is_locked()) 10998 synchronize_rcu_expedited(); 10999 else 11000 synchronize_rcu(); 11001 } 11002 EXPORT_SYMBOL(synchronize_net); 11003 11004 /** 11005 * unregister_netdevice_queue - remove device from the kernel 11006 * @dev: device 11007 * @head: list 11008 * 11009 * This function shuts down a device interface and removes it 11010 * from the kernel tables. 11011 * If head not NULL, device is queued to be unregistered later. 11012 * 11013 * Callers must hold the rtnl semaphore. You may want 11014 * unregister_netdev() instead of this. 11015 */ 11016 11017 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) 11018 { 11019 ASSERT_RTNL(); 11020 11021 if (head) { 11022 list_move_tail(&dev->unreg_list, head); 11023 } else { 11024 LIST_HEAD(single); 11025 11026 list_add(&dev->unreg_list, &single); 11027 unregister_netdevice_many(&single); 11028 } 11029 } 11030 EXPORT_SYMBOL(unregister_netdevice_queue); 11031 11032 /** 11033 * unregister_netdevice_many - unregister many devices 11034 * @head: list of devices 11035 * 11036 * Note: As most callers use a stack allocated list_head, 11037 * we force a list_del() to make sure stack wont be corrupted later. 11038 */ 11039 void unregister_netdevice_many(struct list_head *head) 11040 { 11041 struct net_device *dev, *tmp; 11042 LIST_HEAD(close_head); 11043 11044 BUG_ON(dev_boot_phase); 11045 ASSERT_RTNL(); 11046 11047 if (list_empty(head)) 11048 return; 11049 11050 list_for_each_entry_safe(dev, tmp, head, unreg_list) { 11051 /* Some devices call without registering 11052 * for initialization unwind. Remove those 11053 * devices and proceed with the remaining. 11054 */ 11055 if (dev->reg_state == NETREG_UNINITIALIZED) { 11056 pr_debug("unregister_netdevice: device %s/%p never was registered\n", 11057 dev->name, dev); 11058 11059 WARN_ON(1); 11060 list_del(&dev->unreg_list); 11061 continue; 11062 } 11063 dev->dismantle = true; 11064 BUG_ON(dev->reg_state != NETREG_REGISTERED); 11065 } 11066 11067 /* If device is running, close it first. */ 11068 list_for_each_entry(dev, head, unreg_list) 11069 list_add_tail(&dev->close_list, &close_head); 11070 dev_close_many(&close_head, true); 11071 11072 list_for_each_entry(dev, head, unreg_list) { 11073 /* And unlink it from device chain. */ 11074 unlist_netdevice(dev); 11075 11076 dev->reg_state = NETREG_UNREGISTERING; 11077 } 11078 flush_all_backlogs(); 11079 11080 synchronize_net(); 11081 11082 list_for_each_entry(dev, head, unreg_list) { 11083 struct sk_buff *skb = NULL; 11084 11085 /* Shutdown queueing discipline. */ 11086 dev_shutdown(dev); 11087 11088 dev_xdp_uninstall(dev); 11089 11090 /* Notify protocols, that we are about to destroy 11091 * this device. They should clean all the things. 11092 */ 11093 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 11094 11095 if (!dev->rtnl_link_ops || 11096 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 11097 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0, 11098 GFP_KERNEL, NULL, 0); 11099 11100 /* 11101 * Flush the unicast and multicast chains 11102 */ 11103 dev_uc_flush(dev); 11104 dev_mc_flush(dev); 11105 11106 netdev_name_node_alt_flush(dev); 11107 netdev_name_node_free(dev->name_node); 11108 11109 if (dev->netdev_ops->ndo_uninit) 11110 dev->netdev_ops->ndo_uninit(dev); 11111 11112 if (skb) 11113 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL); 11114 11115 /* Notifier chain MUST detach us all upper devices. */ 11116 WARN_ON(netdev_has_any_upper_dev(dev)); 11117 WARN_ON(netdev_has_any_lower_dev(dev)); 11118 11119 /* Remove entries from kobject tree */ 11120 netdev_unregister_kobject(dev); 11121 #ifdef CONFIG_XPS 11122 /* Remove XPS queueing entries */ 11123 netif_reset_xps_queues_gt(dev, 0); 11124 #endif 11125 } 11126 11127 synchronize_net(); 11128 11129 list_for_each_entry(dev, head, unreg_list) { 11130 dev_put(dev); 11131 net_set_todo(dev); 11132 } 11133 11134 list_del(head); 11135 } 11136 EXPORT_SYMBOL(unregister_netdevice_many); 11137 11138 /** 11139 * unregister_netdev - remove device from the kernel 11140 * @dev: device 11141 * 11142 * This function shuts down a device interface and removes it 11143 * from the kernel tables. 11144 * 11145 * This is just a wrapper for unregister_netdevice that takes 11146 * the rtnl semaphore. In general you want to use this and not 11147 * unregister_netdevice. 11148 */ 11149 void unregister_netdev(struct net_device *dev) 11150 { 11151 rtnl_lock(); 11152 unregister_netdevice(dev); 11153 rtnl_unlock(); 11154 } 11155 EXPORT_SYMBOL(unregister_netdev); 11156 11157 /** 11158 * __dev_change_net_namespace - move device to different nethost namespace 11159 * @dev: device 11160 * @net: network namespace 11161 * @pat: If not NULL name pattern to try if the current device name 11162 * is already taken in the destination network namespace. 11163 * @new_ifindex: If not zero, specifies device index in the target 11164 * namespace. 11165 * 11166 * This function shuts down a device interface and moves it 11167 * to a new network namespace. On success 0 is returned, on 11168 * a failure a netagive errno code is returned. 11169 * 11170 * Callers must hold the rtnl semaphore. 11171 */ 11172 11173 int __dev_change_net_namespace(struct net_device *dev, struct net *net, 11174 const char *pat, int new_ifindex) 11175 { 11176 struct net *net_old = dev_net(dev); 11177 int err, new_nsid; 11178 11179 ASSERT_RTNL(); 11180 11181 /* Don't allow namespace local devices to be moved. */ 11182 err = -EINVAL; 11183 if (dev->features & NETIF_F_NETNS_LOCAL) 11184 goto out; 11185 11186 /* Ensure the device has been registrered */ 11187 if (dev->reg_state != NETREG_REGISTERED) 11188 goto out; 11189 11190 /* Get out if there is nothing todo */ 11191 err = 0; 11192 if (net_eq(net_old, net)) 11193 goto out; 11194 11195 /* Pick the destination device name, and ensure 11196 * we can use it in the destination network namespace. 11197 */ 11198 err = -EEXIST; 11199 if (__dev_get_by_name(net, dev->name)) { 11200 /* We get here if we can't use the current device name */ 11201 if (!pat) 11202 goto out; 11203 err = dev_get_valid_name(net, dev, pat); 11204 if (err < 0) 11205 goto out; 11206 } 11207 11208 /* Check that new_ifindex isn't used yet. */ 11209 err = -EBUSY; 11210 if (new_ifindex && __dev_get_by_index(net, new_ifindex)) 11211 goto out; 11212 11213 /* 11214 * And now a mini version of register_netdevice unregister_netdevice. 11215 */ 11216 11217 /* If device is running close it first. */ 11218 dev_close(dev); 11219 11220 /* And unlink it from device chain */ 11221 unlist_netdevice(dev); 11222 11223 synchronize_net(); 11224 11225 /* Shutdown queueing discipline. */ 11226 dev_shutdown(dev); 11227 11228 /* Notify protocols, that we are about to destroy 11229 * this device. They should clean all the things. 11230 * 11231 * Note that dev->reg_state stays at NETREG_REGISTERED. 11232 * This is wanted because this way 8021q and macvlan know 11233 * the device is just moving and can keep their slaves up. 11234 */ 11235 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 11236 rcu_barrier(); 11237 11238 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL); 11239 /* If there is an ifindex conflict assign a new one */ 11240 if (!new_ifindex) { 11241 if (__dev_get_by_index(net, dev->ifindex)) 11242 new_ifindex = dev_new_index(net); 11243 else 11244 new_ifindex = dev->ifindex; 11245 } 11246 11247 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid, 11248 new_ifindex); 11249 11250 /* 11251 * Flush the unicast and multicast chains 11252 */ 11253 dev_uc_flush(dev); 11254 dev_mc_flush(dev); 11255 11256 /* Send a netdev-removed uevent to the old namespace */ 11257 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); 11258 netdev_adjacent_del_links(dev); 11259 11260 /* Move per-net netdevice notifiers that are following the netdevice */ 11261 move_netdevice_notifiers_dev_net(dev, net); 11262 11263 /* Actually switch the network namespace */ 11264 dev_net_set(dev, net); 11265 dev->ifindex = new_ifindex; 11266 11267 /* Send a netdev-add uevent to the new namespace */ 11268 kobject_uevent(&dev->dev.kobj, KOBJ_ADD); 11269 netdev_adjacent_add_links(dev); 11270 11271 /* Fixup kobjects */ 11272 err = device_rename(&dev->dev, dev->name); 11273 WARN_ON(err); 11274 11275 /* Adapt owner in case owning user namespace of target network 11276 * namespace is different from the original one. 11277 */ 11278 err = netdev_change_owner(dev, net_old, net); 11279 WARN_ON(err); 11280 11281 /* Add the device back in the hashes */ 11282 list_netdevice(dev); 11283 11284 /* Notify protocols, that a new device appeared. */ 11285 call_netdevice_notifiers(NETDEV_REGISTER, dev); 11286 11287 /* 11288 * Prevent userspace races by waiting until the network 11289 * device is fully setup before sending notifications. 11290 */ 11291 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); 11292 11293 synchronize_net(); 11294 err = 0; 11295 out: 11296 return err; 11297 } 11298 EXPORT_SYMBOL_GPL(__dev_change_net_namespace); 11299 11300 static int dev_cpu_dead(unsigned int oldcpu) 11301 { 11302 struct sk_buff **list_skb; 11303 struct sk_buff *skb; 11304 unsigned int cpu; 11305 struct softnet_data *sd, *oldsd, *remsd = NULL; 11306 11307 local_irq_disable(); 11308 cpu = smp_processor_id(); 11309 sd = &per_cpu(softnet_data, cpu); 11310 oldsd = &per_cpu(softnet_data, oldcpu); 11311 11312 /* Find end of our completion_queue. */ 11313 list_skb = &sd->completion_queue; 11314 while (*list_skb) 11315 list_skb = &(*list_skb)->next; 11316 /* Append completion queue from offline CPU. */ 11317 *list_skb = oldsd->completion_queue; 11318 oldsd->completion_queue = NULL; 11319 11320 /* Append output queue from offline CPU. */ 11321 if (oldsd->output_queue) { 11322 *sd->output_queue_tailp = oldsd->output_queue; 11323 sd->output_queue_tailp = oldsd->output_queue_tailp; 11324 oldsd->output_queue = NULL; 11325 oldsd->output_queue_tailp = &oldsd->output_queue; 11326 } 11327 /* Append NAPI poll list from offline CPU, with one exception : 11328 * process_backlog() must be called by cpu owning percpu backlog. 11329 * We properly handle process_queue & input_pkt_queue later. 11330 */ 11331 while (!list_empty(&oldsd->poll_list)) { 11332 struct napi_struct *napi = list_first_entry(&oldsd->poll_list, 11333 struct napi_struct, 11334 poll_list); 11335 11336 list_del_init(&napi->poll_list); 11337 if (napi->poll == process_backlog) 11338 napi->state = 0; 11339 else 11340 ____napi_schedule(sd, napi); 11341 } 11342 11343 raise_softirq_irqoff(NET_TX_SOFTIRQ); 11344 local_irq_enable(); 11345 11346 #ifdef CONFIG_RPS 11347 remsd = oldsd->rps_ipi_list; 11348 oldsd->rps_ipi_list = NULL; 11349 #endif 11350 /* send out pending IPI's on offline CPU */ 11351 net_rps_send_ipi(remsd); 11352 11353 /* Process offline CPU's input_pkt_queue */ 11354 while ((skb = __skb_dequeue(&oldsd->process_queue))) { 11355 netif_rx_ni(skb); 11356 input_queue_head_incr(oldsd); 11357 } 11358 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { 11359 netif_rx_ni(skb); 11360 input_queue_head_incr(oldsd); 11361 } 11362 11363 return 0; 11364 } 11365 11366 /** 11367 * netdev_increment_features - increment feature set by one 11368 * @all: current feature set 11369 * @one: new feature set 11370 * @mask: mask feature set 11371 * 11372 * Computes a new feature set after adding a device with feature set 11373 * @one to the master device with current feature set @all. Will not 11374 * enable anything that is off in @mask. Returns the new feature set. 11375 */ 11376 netdev_features_t netdev_increment_features(netdev_features_t all, 11377 netdev_features_t one, netdev_features_t mask) 11378 { 11379 if (mask & NETIF_F_HW_CSUM) 11380 mask |= NETIF_F_CSUM_MASK; 11381 mask |= NETIF_F_VLAN_CHALLENGED; 11382 11383 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; 11384 all &= one | ~NETIF_F_ALL_FOR_ALL; 11385 11386 /* If one device supports hw checksumming, set for all. */ 11387 if (all & NETIF_F_HW_CSUM) 11388 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); 11389 11390 return all; 11391 } 11392 EXPORT_SYMBOL(netdev_increment_features); 11393 11394 static struct hlist_head * __net_init netdev_create_hash(void) 11395 { 11396 int i; 11397 struct hlist_head *hash; 11398 11399 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL); 11400 if (hash != NULL) 11401 for (i = 0; i < NETDEV_HASHENTRIES; i++) 11402 INIT_HLIST_HEAD(&hash[i]); 11403 11404 return hash; 11405 } 11406 11407 /* Initialize per network namespace state */ 11408 static int __net_init netdev_init(struct net *net) 11409 { 11410 BUILD_BUG_ON(GRO_HASH_BUCKETS > 11411 8 * sizeof_field(struct napi_struct, gro_bitmask)); 11412 11413 if (net != &init_net) 11414 INIT_LIST_HEAD(&net->dev_base_head); 11415 11416 net->dev_name_head = netdev_create_hash(); 11417 if (net->dev_name_head == NULL) 11418 goto err_name; 11419 11420 net->dev_index_head = netdev_create_hash(); 11421 if (net->dev_index_head == NULL) 11422 goto err_idx; 11423 11424 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain); 11425 11426 return 0; 11427 11428 err_idx: 11429 kfree(net->dev_name_head); 11430 err_name: 11431 return -ENOMEM; 11432 } 11433 11434 /** 11435 * netdev_drivername - network driver for the device 11436 * @dev: network device 11437 * 11438 * Determine network driver for device. 11439 */ 11440 const char *netdev_drivername(const struct net_device *dev) 11441 { 11442 const struct device_driver *driver; 11443 const struct device *parent; 11444 const char *empty = ""; 11445 11446 parent = dev->dev.parent; 11447 if (!parent) 11448 return empty; 11449 11450 driver = parent->driver; 11451 if (driver && driver->name) 11452 return driver->name; 11453 return empty; 11454 } 11455 11456 static void __netdev_printk(const char *level, const struct net_device *dev, 11457 struct va_format *vaf) 11458 { 11459 if (dev && dev->dev.parent) { 11460 dev_printk_emit(level[1] - '0', 11461 dev->dev.parent, 11462 "%s %s %s%s: %pV", 11463 dev_driver_string(dev->dev.parent), 11464 dev_name(dev->dev.parent), 11465 netdev_name(dev), netdev_reg_state(dev), 11466 vaf); 11467 } else if (dev) { 11468 printk("%s%s%s: %pV", 11469 level, netdev_name(dev), netdev_reg_state(dev), vaf); 11470 } else { 11471 printk("%s(NULL net_device): %pV", level, vaf); 11472 } 11473 } 11474 11475 void netdev_printk(const char *level, const struct net_device *dev, 11476 const char *format, ...) 11477 { 11478 struct va_format vaf; 11479 va_list args; 11480 11481 va_start(args, format); 11482 11483 vaf.fmt = format; 11484 vaf.va = &args; 11485 11486 __netdev_printk(level, dev, &vaf); 11487 11488 va_end(args); 11489 } 11490 EXPORT_SYMBOL(netdev_printk); 11491 11492 #define define_netdev_printk_level(func, level) \ 11493 void func(const struct net_device *dev, const char *fmt, ...) \ 11494 { \ 11495 struct va_format vaf; \ 11496 va_list args; \ 11497 \ 11498 va_start(args, fmt); \ 11499 \ 11500 vaf.fmt = fmt; \ 11501 vaf.va = &args; \ 11502 \ 11503 __netdev_printk(level, dev, &vaf); \ 11504 \ 11505 va_end(args); \ 11506 } \ 11507 EXPORT_SYMBOL(func); 11508 11509 define_netdev_printk_level(netdev_emerg, KERN_EMERG); 11510 define_netdev_printk_level(netdev_alert, KERN_ALERT); 11511 define_netdev_printk_level(netdev_crit, KERN_CRIT); 11512 define_netdev_printk_level(netdev_err, KERN_ERR); 11513 define_netdev_printk_level(netdev_warn, KERN_WARNING); 11514 define_netdev_printk_level(netdev_notice, KERN_NOTICE); 11515 define_netdev_printk_level(netdev_info, KERN_INFO); 11516 11517 static void __net_exit netdev_exit(struct net *net) 11518 { 11519 kfree(net->dev_name_head); 11520 kfree(net->dev_index_head); 11521 if (net != &init_net) 11522 WARN_ON_ONCE(!list_empty(&net->dev_base_head)); 11523 } 11524 11525 static struct pernet_operations __net_initdata netdev_net_ops = { 11526 .init = netdev_init, 11527 .exit = netdev_exit, 11528 }; 11529 11530 static void __net_exit default_device_exit(struct net *net) 11531 { 11532 struct net_device *dev, *aux; 11533 /* 11534 * Push all migratable network devices back to the 11535 * initial network namespace 11536 */ 11537 rtnl_lock(); 11538 for_each_netdev_safe(net, dev, aux) { 11539 int err; 11540 char fb_name[IFNAMSIZ]; 11541 11542 /* Ignore unmoveable devices (i.e. loopback) */ 11543 if (dev->features & NETIF_F_NETNS_LOCAL) 11544 continue; 11545 11546 /* Leave virtual devices for the generic cleanup */ 11547 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund) 11548 continue; 11549 11550 /* Push remaining network devices to init_net */ 11551 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); 11552 if (__dev_get_by_name(&init_net, fb_name)) 11553 snprintf(fb_name, IFNAMSIZ, "dev%%d"); 11554 err = dev_change_net_namespace(dev, &init_net, fb_name); 11555 if (err) { 11556 pr_emerg("%s: failed to move %s to init_net: %d\n", 11557 __func__, dev->name, err); 11558 BUG(); 11559 } 11560 } 11561 rtnl_unlock(); 11562 } 11563 11564 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list) 11565 { 11566 /* Return with the rtnl_lock held when there are no network 11567 * devices unregistering in any network namespace in net_list. 11568 */ 11569 struct net *net; 11570 bool unregistering; 11571 DEFINE_WAIT_FUNC(wait, woken_wake_function); 11572 11573 add_wait_queue(&netdev_unregistering_wq, &wait); 11574 for (;;) { 11575 unregistering = false; 11576 rtnl_lock(); 11577 list_for_each_entry(net, net_list, exit_list) { 11578 if (net->dev_unreg_count > 0) { 11579 unregistering = true; 11580 break; 11581 } 11582 } 11583 if (!unregistering) 11584 break; 11585 __rtnl_unlock(); 11586 11587 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); 11588 } 11589 remove_wait_queue(&netdev_unregistering_wq, &wait); 11590 } 11591 11592 static void __net_exit default_device_exit_batch(struct list_head *net_list) 11593 { 11594 /* At exit all network devices most be removed from a network 11595 * namespace. Do this in the reverse order of registration. 11596 * Do this across as many network namespaces as possible to 11597 * improve batching efficiency. 11598 */ 11599 struct net_device *dev; 11600 struct net *net; 11601 LIST_HEAD(dev_kill_list); 11602 11603 /* To prevent network device cleanup code from dereferencing 11604 * loopback devices or network devices that have been freed 11605 * wait here for all pending unregistrations to complete, 11606 * before unregistring the loopback device and allowing the 11607 * network namespace be freed. 11608 * 11609 * The netdev todo list containing all network devices 11610 * unregistrations that happen in default_device_exit_batch 11611 * will run in the rtnl_unlock() at the end of 11612 * default_device_exit_batch. 11613 */ 11614 rtnl_lock_unregistering(net_list); 11615 list_for_each_entry(net, net_list, exit_list) { 11616 for_each_netdev_reverse(net, dev) { 11617 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) 11618 dev->rtnl_link_ops->dellink(dev, &dev_kill_list); 11619 else 11620 unregister_netdevice_queue(dev, &dev_kill_list); 11621 } 11622 } 11623 unregister_netdevice_many(&dev_kill_list); 11624 rtnl_unlock(); 11625 } 11626 11627 static struct pernet_operations __net_initdata default_device_ops = { 11628 .exit = default_device_exit, 11629 .exit_batch = default_device_exit_batch, 11630 }; 11631 11632 /* 11633 * Initialize the DEV module. At boot time this walks the device list and 11634 * unhooks any devices that fail to initialise (normally hardware not 11635 * present) and leaves us with a valid list of present and active devices. 11636 * 11637 */ 11638 11639 /* 11640 * This is called single threaded during boot, so no need 11641 * to take the rtnl semaphore. 11642 */ 11643 static int __init net_dev_init(void) 11644 { 11645 int i, rc = -ENOMEM; 11646 11647 BUG_ON(!dev_boot_phase); 11648 11649 if (dev_proc_init()) 11650 goto out; 11651 11652 if (netdev_kobject_init()) 11653 goto out; 11654 11655 INIT_LIST_HEAD(&ptype_all); 11656 for (i = 0; i < PTYPE_HASH_SIZE; i++) 11657 INIT_LIST_HEAD(&ptype_base[i]); 11658 11659 INIT_LIST_HEAD(&offload_base); 11660 11661 if (register_pernet_subsys(&netdev_net_ops)) 11662 goto out; 11663 11664 /* 11665 * Initialise the packet receive queues. 11666 */ 11667 11668 for_each_possible_cpu(i) { 11669 struct work_struct *flush = per_cpu_ptr(&flush_works, i); 11670 struct softnet_data *sd = &per_cpu(softnet_data, i); 11671 11672 INIT_WORK(flush, flush_backlog); 11673 11674 skb_queue_head_init(&sd->input_pkt_queue); 11675 skb_queue_head_init(&sd->process_queue); 11676 #ifdef CONFIG_XFRM_OFFLOAD 11677 skb_queue_head_init(&sd->xfrm_backlog); 11678 #endif 11679 INIT_LIST_HEAD(&sd->poll_list); 11680 sd->output_queue_tailp = &sd->output_queue; 11681 #ifdef CONFIG_RPS 11682 INIT_CSD(&sd->csd, rps_trigger_softirq, sd); 11683 sd->cpu = i; 11684 #endif 11685 11686 init_gro_hash(&sd->backlog); 11687 sd->backlog.poll = process_backlog; 11688 sd->backlog.weight = weight_p; 11689 } 11690 11691 dev_boot_phase = 0; 11692 11693 /* The loopback device is special if any other network devices 11694 * is present in a network namespace the loopback device must 11695 * be present. Since we now dynamically allocate and free the 11696 * loopback device ensure this invariant is maintained by 11697 * keeping the loopback device as the first device on the 11698 * list of network devices. Ensuring the loopback devices 11699 * is the first device that appears and the last network device 11700 * that disappears. 11701 */ 11702 if (register_pernet_device(&loopback_net_ops)) 11703 goto out; 11704 11705 if (register_pernet_device(&default_device_ops)) 11706 goto out; 11707 11708 open_softirq(NET_TX_SOFTIRQ, net_tx_action); 11709 open_softirq(NET_RX_SOFTIRQ, net_rx_action); 11710 11711 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", 11712 NULL, dev_cpu_dead); 11713 WARN_ON(rc < 0); 11714 rc = 0; 11715 out: 11716 return rc; 11717 } 11718 11719 subsys_initcall(net_dev_init); 11720