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