1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * NET3 Protocol independent device support routines. 4 * 5 * Derived from the non IP parts of dev.c 1.0.19 6 * Authors: Ross Biro 7 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 8 * Mark Evans, <evansmp@uhura.aston.ac.uk> 9 * 10 * Additional Authors: 11 * Florian la Roche <rzsfl@rz.uni-sb.de> 12 * Alan Cox <gw4pts@gw4pts.ampr.org> 13 * David Hinds <dahinds@users.sourceforge.net> 14 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru> 15 * Adam Sulmicki <adam@cfar.umd.edu> 16 * Pekka Riikonen <priikone@poesidon.pspt.fi> 17 * 18 * Changes: 19 * D.J. Barrow : Fixed bug where dev->refcnt gets set 20 * to 2 if register_netdev gets called 21 * before net_dev_init & also removed a 22 * few lines of code in the process. 23 * Alan Cox : device private ioctl copies fields back. 24 * Alan Cox : Transmit queue code does relevant 25 * stunts to keep the queue safe. 26 * Alan Cox : Fixed double lock. 27 * Alan Cox : Fixed promisc NULL pointer trap 28 * ???????? : Support the full private ioctl range 29 * Alan Cox : Moved ioctl permission check into 30 * drivers 31 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI 32 * Alan Cox : 100 backlog just doesn't cut it when 33 * you start doing multicast video 8) 34 * Alan Cox : Rewrote net_bh and list manager. 35 * Alan Cox : Fix ETH_P_ALL echoback lengths. 36 * Alan Cox : Took out transmit every packet pass 37 * Saved a few bytes in the ioctl handler 38 * Alan Cox : Network driver sets packet type before 39 * calling netif_rx. Saves a function 40 * call a packet. 41 * Alan Cox : Hashed net_bh() 42 * Richard Kooijman: Timestamp fixes. 43 * Alan Cox : Wrong field in SIOCGIFDSTADDR 44 * Alan Cox : Device lock protection. 45 * Alan Cox : Fixed nasty side effect of device close 46 * changes. 47 * Rudi Cilibrasi : Pass the right thing to 48 * set_mac_address() 49 * Dave Miller : 32bit quantity for the device lock to 50 * make it work out on a Sparc. 51 * Bjorn Ekwall : Added KERNELD hack. 52 * Alan Cox : Cleaned up the backlog initialise. 53 * Craig Metz : SIOCGIFCONF fix if space for under 54 * 1 device. 55 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there 56 * is no device open function. 57 * Andi Kleen : Fix error reporting for SIOCGIFCONF 58 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF 59 * Cyrus Durgin : Cleaned for KMOD 60 * Adam Sulmicki : Bug Fix : Network Device Unload 61 * A network device unload needs to purge 62 * the backlog queue. 63 * Paul Rusty Russell : SIOCSIFNAME 64 * Pekka Riikonen : Netdev boot-time settings code 65 * Andrew Morton : Make unregister_netdevice wait 66 * indefinitely on dev->refcnt 67 * J Hadi Salim : - Backlog queue sampling 68 * - netif_rx() feedback 69 */ 70 71 #include <linux/uaccess.h> 72 #include <linux/bitops.h> 73 #include <linux/capability.h> 74 #include <linux/cpu.h> 75 #include <linux/types.h> 76 #include <linux/kernel.h> 77 #include <linux/hash.h> 78 #include <linux/slab.h> 79 #include <linux/sched.h> 80 #include <linux/sched/mm.h> 81 #include <linux/mutex.h> 82 #include <linux/rwsem.h> 83 #include <linux/string.h> 84 #include <linux/mm.h> 85 #include <linux/socket.h> 86 #include <linux/sockios.h> 87 #include <linux/errno.h> 88 #include <linux/interrupt.h> 89 #include <linux/if_ether.h> 90 #include <linux/netdevice.h> 91 #include <linux/etherdevice.h> 92 #include <linux/ethtool.h> 93 #include <linux/skbuff.h> 94 #include <linux/kthread.h> 95 #include <linux/bpf.h> 96 #include <linux/bpf_trace.h> 97 #include <net/net_namespace.h> 98 #include <net/sock.h> 99 #include <net/busy_poll.h> 100 #include <linux/rtnetlink.h> 101 #include <linux/stat.h> 102 #include <net/dsa.h> 103 #include <net/dst.h> 104 #include <net/dst_metadata.h> 105 #include <net/gro.h> 106 #include <net/pkt_sched.h> 107 #include <net/pkt_cls.h> 108 #include <net/checksum.h> 109 #include <net/xfrm.h> 110 #include <linux/highmem.h> 111 #include <linux/init.h> 112 #include <linux/module.h> 113 #include <linux/netpoll.h> 114 #include <linux/rcupdate.h> 115 #include <linux/delay.h> 116 #include <net/iw_handler.h> 117 #include <asm/current.h> 118 #include <linux/audit.h> 119 #include <linux/dmaengine.h> 120 #include <linux/err.h> 121 #include <linux/ctype.h> 122 #include <linux/if_arp.h> 123 #include <linux/if_vlan.h> 124 #include <linux/ip.h> 125 #include <net/ip.h> 126 #include <net/mpls.h> 127 #include <linux/ipv6.h> 128 #include <linux/in.h> 129 #include <linux/jhash.h> 130 #include <linux/random.h> 131 #include <trace/events/napi.h> 132 #include <trace/events/net.h> 133 #include <trace/events/skb.h> 134 #include <trace/events/qdisc.h> 135 #include <linux/inetdevice.h> 136 #include <linux/cpu_rmap.h> 137 #include <linux/static_key.h> 138 #include <linux/hashtable.h> 139 #include <linux/vmalloc.h> 140 #include <linux/if_macvlan.h> 141 #include <linux/errqueue.h> 142 #include <linux/hrtimer.h> 143 #include <linux/netfilter_netdev.h> 144 #include <linux/crash_dump.h> 145 #include <linux/sctp.h> 146 #include <net/udp_tunnel.h> 147 #include <linux/net_namespace.h> 148 #include <linux/indirect_call_wrapper.h> 149 #include <net/devlink.h> 150 #include <linux/pm_runtime.h> 151 #include <linux/prandom.h> 152 #include <linux/once_lite.h> 153 154 #include "net-sysfs.h" 155 156 157 static DEFINE_SPINLOCK(ptype_lock); 158 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; 159 struct list_head ptype_all __read_mostly; /* Taps */ 160 161 static int netif_rx_internal(struct sk_buff *skb); 162 static int call_netdevice_notifiers_info(unsigned long val, 163 struct netdev_notifier_info *info); 164 static int call_netdevice_notifiers_extack(unsigned long val, 165 struct net_device *dev, 166 struct netlink_ext_ack *extack); 167 static struct napi_struct *napi_by_id(unsigned int napi_id); 168 169 /* 170 * The @dev_base_head list is protected by @dev_base_lock and the rtnl 171 * semaphore. 172 * 173 * Pure readers hold dev_base_lock for reading, or rcu_read_lock() 174 * 175 * Writers must hold the rtnl semaphore while they loop through the 176 * dev_base_head list, and hold dev_base_lock for writing when they do the 177 * actual updates. This allows pure readers to access the list even 178 * while a writer is preparing to update it. 179 * 180 * To put it another way, dev_base_lock is held for writing only to 181 * protect against pure readers; the rtnl semaphore provides the 182 * protection against other writers. 183 * 184 * See, for example usages, register_netdevice() and 185 * unregister_netdevice(), which must be called with the rtnl 186 * semaphore held. 187 */ 188 DEFINE_RWLOCK(dev_base_lock); 189 EXPORT_SYMBOL(dev_base_lock); 190 191 static DEFINE_MUTEX(ifalias_mutex); 192 193 /* protects napi_hash addition/deletion and napi_gen_id */ 194 static DEFINE_SPINLOCK(napi_hash_lock); 195 196 static unsigned int napi_gen_id = NR_CPUS; 197 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8); 198 199 static DECLARE_RWSEM(devnet_rename_sem); 200 201 static inline void dev_base_seq_inc(struct net *net) 202 { 203 while (++net->dev_base_seq == 0) 204 ; 205 } 206 207 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) 208 { 209 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ)); 210 211 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; 212 } 213 214 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) 215 { 216 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; 217 } 218 219 static inline void rps_lock(struct softnet_data *sd) 220 { 221 #ifdef CONFIG_RPS 222 spin_lock(&sd->input_pkt_queue.lock); 223 #endif 224 } 225 226 static inline void rps_unlock(struct softnet_data *sd) 227 { 228 #ifdef CONFIG_RPS 229 spin_unlock(&sd->input_pkt_queue.lock); 230 #endif 231 } 232 233 static struct netdev_name_node *netdev_name_node_alloc(struct net_device *dev, 234 const char *name) 235 { 236 struct netdev_name_node *name_node; 237 238 name_node = kmalloc(sizeof(*name_node), GFP_KERNEL); 239 if (!name_node) 240 return NULL; 241 INIT_HLIST_NODE(&name_node->hlist); 242 name_node->dev = dev; 243 name_node->name = name; 244 return name_node; 245 } 246 247 static struct netdev_name_node * 248 netdev_name_node_head_alloc(struct net_device *dev) 249 { 250 struct netdev_name_node *name_node; 251 252 name_node = netdev_name_node_alloc(dev, dev->name); 253 if (!name_node) 254 return NULL; 255 INIT_LIST_HEAD(&name_node->list); 256 return name_node; 257 } 258 259 static void netdev_name_node_free(struct netdev_name_node *name_node) 260 { 261 kfree(name_node); 262 } 263 264 static void netdev_name_node_add(struct net *net, 265 struct netdev_name_node *name_node) 266 { 267 hlist_add_head_rcu(&name_node->hlist, 268 dev_name_hash(net, name_node->name)); 269 } 270 271 static void netdev_name_node_del(struct netdev_name_node *name_node) 272 { 273 hlist_del_rcu(&name_node->hlist); 274 } 275 276 static struct netdev_name_node *netdev_name_node_lookup(struct net *net, 277 const char *name) 278 { 279 struct hlist_head *head = dev_name_hash(net, name); 280 struct netdev_name_node *name_node; 281 282 hlist_for_each_entry(name_node, head, hlist) 283 if (!strcmp(name_node->name, name)) 284 return name_node; 285 return NULL; 286 } 287 288 static struct netdev_name_node *netdev_name_node_lookup_rcu(struct net *net, 289 const char *name) 290 { 291 struct hlist_head *head = dev_name_hash(net, name); 292 struct netdev_name_node *name_node; 293 294 hlist_for_each_entry_rcu(name_node, head, hlist) 295 if (!strcmp(name_node->name, name)) 296 return name_node; 297 return NULL; 298 } 299 300 bool netdev_name_in_use(struct net *net, const char *name) 301 { 302 return netdev_name_node_lookup(net, name); 303 } 304 EXPORT_SYMBOL(netdev_name_in_use); 305 306 int netdev_name_node_alt_create(struct net_device *dev, const char *name) 307 { 308 struct netdev_name_node *name_node; 309 struct net *net = dev_net(dev); 310 311 name_node = netdev_name_node_lookup(net, name); 312 if (name_node) 313 return -EEXIST; 314 name_node = netdev_name_node_alloc(dev, name); 315 if (!name_node) 316 return -ENOMEM; 317 netdev_name_node_add(net, name_node); 318 /* The node that holds dev->name acts as a head of per-device list. */ 319 list_add_tail(&name_node->list, &dev->name_node->list); 320 321 return 0; 322 } 323 EXPORT_SYMBOL(netdev_name_node_alt_create); 324 325 static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node) 326 { 327 list_del(&name_node->list); 328 netdev_name_node_del(name_node); 329 kfree(name_node->name); 330 netdev_name_node_free(name_node); 331 } 332 333 int netdev_name_node_alt_destroy(struct net_device *dev, const char *name) 334 { 335 struct netdev_name_node *name_node; 336 struct net *net = dev_net(dev); 337 338 name_node = netdev_name_node_lookup(net, name); 339 if (!name_node) 340 return -ENOENT; 341 /* lookup might have found our primary name or a name belonging 342 * to another device. 343 */ 344 if (name_node == dev->name_node || name_node->dev != dev) 345 return -EINVAL; 346 347 __netdev_name_node_alt_destroy(name_node); 348 349 return 0; 350 } 351 EXPORT_SYMBOL(netdev_name_node_alt_destroy); 352 353 static void netdev_name_node_alt_flush(struct net_device *dev) 354 { 355 struct netdev_name_node *name_node, *tmp; 356 357 list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list) 358 __netdev_name_node_alt_destroy(name_node); 359 } 360 361 /* Device list insertion */ 362 static void list_netdevice(struct net_device *dev) 363 { 364 struct net *net = dev_net(dev); 365 366 ASSERT_RTNL(); 367 368 write_lock(&dev_base_lock); 369 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); 370 netdev_name_node_add(net, dev->name_node); 371 hlist_add_head_rcu(&dev->index_hlist, 372 dev_index_hash(net, dev->ifindex)); 373 write_unlock(&dev_base_lock); 374 375 dev_base_seq_inc(net); 376 } 377 378 /* Device list removal 379 * caller must respect a RCU grace period before freeing/reusing dev 380 */ 381 static void unlist_netdevice(struct net_device *dev) 382 { 383 ASSERT_RTNL(); 384 385 /* Unlink dev from the device chain */ 386 write_lock(&dev_base_lock); 387 list_del_rcu(&dev->dev_list); 388 netdev_name_node_del(dev->name_node); 389 hlist_del_rcu(&dev->index_hlist); 390 write_unlock(&dev_base_lock); 391 392 dev_base_seq_inc(dev_net(dev)); 393 } 394 395 /* 396 * Our notifier list 397 */ 398 399 static RAW_NOTIFIER_HEAD(netdev_chain); 400 401 /* 402 * Device drivers call our routines to queue packets here. We empty the 403 * queue in the local softnet handler. 404 */ 405 406 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); 407 EXPORT_PER_CPU_SYMBOL(softnet_data); 408 409 #ifdef CONFIG_LOCKDEP 410 /* 411 * register_netdevice() inits txq->_xmit_lock and sets lockdep class 412 * according to dev->type 413 */ 414 static const unsigned short netdev_lock_type[] = { 415 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, 416 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, 417 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, 418 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, 419 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, 420 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, 421 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, 422 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, 423 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, 424 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, 425 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, 426 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, 427 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, 428 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, 429 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; 430 431 static const char *const netdev_lock_name[] = { 432 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", 433 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", 434 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", 435 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", 436 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", 437 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", 438 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", 439 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", 440 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", 441 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", 442 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", 443 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", 444 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", 445 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", 446 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; 447 448 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; 449 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; 450 451 static inline unsigned short netdev_lock_pos(unsigned short dev_type) 452 { 453 int i; 454 455 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) 456 if (netdev_lock_type[i] == dev_type) 457 return i; 458 /* the last key is used by default */ 459 return ARRAY_SIZE(netdev_lock_type) - 1; 460 } 461 462 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 463 unsigned short dev_type) 464 { 465 int i; 466 467 i = netdev_lock_pos(dev_type); 468 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], 469 netdev_lock_name[i]); 470 } 471 472 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 473 { 474 int i; 475 476 i = netdev_lock_pos(dev->type); 477 lockdep_set_class_and_name(&dev->addr_list_lock, 478 &netdev_addr_lock_key[i], 479 netdev_lock_name[i]); 480 } 481 #else 482 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 483 unsigned short dev_type) 484 { 485 } 486 487 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 488 { 489 } 490 #endif 491 492 /******************************************************************************* 493 * 494 * Protocol management and registration routines 495 * 496 *******************************************************************************/ 497 498 499 /* 500 * Add a protocol ID to the list. Now that the input handler is 501 * smarter we can dispense with all the messy stuff that used to be 502 * here. 503 * 504 * BEWARE!!! Protocol handlers, mangling input packets, 505 * MUST BE last in hash buckets and checking protocol handlers 506 * MUST start from promiscuous ptype_all chain in net_bh. 507 * It is true now, do not change it. 508 * Explanation follows: if protocol handler, mangling packet, will 509 * be the first on list, it is not able to sense, that packet 510 * is cloned and should be copied-on-write, so that it will 511 * change it and subsequent readers will get broken packet. 512 * --ANK (980803) 513 */ 514 515 static inline struct list_head *ptype_head(const struct packet_type *pt) 516 { 517 if (pt->type == htons(ETH_P_ALL)) 518 return pt->dev ? &pt->dev->ptype_all : &ptype_all; 519 else 520 return pt->dev ? &pt->dev->ptype_specific : 521 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; 522 } 523 524 /** 525 * dev_add_pack - add packet handler 526 * @pt: packet type declaration 527 * 528 * Add a protocol handler to the networking stack. The passed &packet_type 529 * is linked into kernel lists and may not be freed until it has been 530 * removed from the kernel lists. 531 * 532 * This call does not sleep therefore it can not 533 * guarantee all CPU's that are in middle of receiving packets 534 * will see the new packet type (until the next received packet). 535 */ 536 537 void dev_add_pack(struct packet_type *pt) 538 { 539 struct list_head *head = ptype_head(pt); 540 541 spin_lock(&ptype_lock); 542 list_add_rcu(&pt->list, head); 543 spin_unlock(&ptype_lock); 544 } 545 EXPORT_SYMBOL(dev_add_pack); 546 547 /** 548 * __dev_remove_pack - remove packet handler 549 * @pt: packet type declaration 550 * 551 * Remove a protocol handler that was previously added to the kernel 552 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 553 * from the kernel lists and can be freed or reused once this function 554 * returns. 555 * 556 * The packet type might still be in use by receivers 557 * and must not be freed until after all the CPU's have gone 558 * through a quiescent state. 559 */ 560 void __dev_remove_pack(struct packet_type *pt) 561 { 562 struct list_head *head = ptype_head(pt); 563 struct packet_type *pt1; 564 565 spin_lock(&ptype_lock); 566 567 list_for_each_entry(pt1, head, list) { 568 if (pt == pt1) { 569 list_del_rcu(&pt->list); 570 goto out; 571 } 572 } 573 574 pr_warn("dev_remove_pack: %p not found\n", pt); 575 out: 576 spin_unlock(&ptype_lock); 577 } 578 EXPORT_SYMBOL(__dev_remove_pack); 579 580 /** 581 * dev_remove_pack - remove packet handler 582 * @pt: packet type declaration 583 * 584 * Remove a protocol handler that was previously added to the kernel 585 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 586 * from the kernel lists and can be freed or reused once this function 587 * returns. 588 * 589 * This call sleeps to guarantee that no CPU is looking at the packet 590 * type after return. 591 */ 592 void dev_remove_pack(struct packet_type *pt) 593 { 594 __dev_remove_pack(pt); 595 596 synchronize_net(); 597 } 598 EXPORT_SYMBOL(dev_remove_pack); 599 600 601 /******************************************************************************* 602 * 603 * Device Interface Subroutines 604 * 605 *******************************************************************************/ 606 607 /** 608 * dev_get_iflink - get 'iflink' value of a interface 609 * @dev: targeted interface 610 * 611 * Indicates the ifindex the interface is linked to. 612 * Physical interfaces have the same 'ifindex' and 'iflink' values. 613 */ 614 615 int dev_get_iflink(const struct net_device *dev) 616 { 617 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink) 618 return dev->netdev_ops->ndo_get_iflink(dev); 619 620 return dev->ifindex; 621 } 622 EXPORT_SYMBOL(dev_get_iflink); 623 624 /** 625 * dev_fill_metadata_dst - Retrieve tunnel egress information. 626 * @dev: targeted interface 627 * @skb: The packet. 628 * 629 * For better visibility of tunnel traffic OVS needs to retrieve 630 * egress tunnel information for a packet. Following API allows 631 * user to get this info. 632 */ 633 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) 634 { 635 struct ip_tunnel_info *info; 636 637 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst) 638 return -EINVAL; 639 640 info = skb_tunnel_info_unclone(skb); 641 if (!info) 642 return -ENOMEM; 643 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX))) 644 return -EINVAL; 645 646 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb); 647 } 648 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst); 649 650 static struct net_device_path *dev_fwd_path(struct net_device_path_stack *stack) 651 { 652 int k = stack->num_paths++; 653 654 if (WARN_ON_ONCE(k >= NET_DEVICE_PATH_STACK_MAX)) 655 return NULL; 656 657 return &stack->path[k]; 658 } 659 660 int dev_fill_forward_path(const struct net_device *dev, const u8 *daddr, 661 struct net_device_path_stack *stack) 662 { 663 const struct net_device *last_dev; 664 struct net_device_path_ctx ctx = { 665 .dev = dev, 666 .daddr = daddr, 667 }; 668 struct net_device_path *path; 669 int ret = 0; 670 671 stack->num_paths = 0; 672 while (ctx.dev && ctx.dev->netdev_ops->ndo_fill_forward_path) { 673 last_dev = ctx.dev; 674 path = dev_fwd_path(stack); 675 if (!path) 676 return -1; 677 678 memset(path, 0, sizeof(struct net_device_path)); 679 ret = ctx.dev->netdev_ops->ndo_fill_forward_path(&ctx, path); 680 if (ret < 0) 681 return -1; 682 683 if (WARN_ON_ONCE(last_dev == ctx.dev)) 684 return -1; 685 } 686 path = dev_fwd_path(stack); 687 if (!path) 688 return -1; 689 path->type = DEV_PATH_ETHERNET; 690 path->dev = ctx.dev; 691 692 return ret; 693 } 694 EXPORT_SYMBOL_GPL(dev_fill_forward_path); 695 696 /** 697 * __dev_get_by_name - find a device by its name 698 * @net: the applicable net namespace 699 * @name: name to find 700 * 701 * Find an interface by name. Must be called under RTNL semaphore 702 * or @dev_base_lock. If the name is found a pointer to the device 703 * is returned. If the name is not found then %NULL is returned. The 704 * reference counters are not incremented so the caller must be 705 * careful with locks. 706 */ 707 708 struct net_device *__dev_get_by_name(struct net *net, const char *name) 709 { 710 struct netdev_name_node *node_name; 711 712 node_name = netdev_name_node_lookup(net, name); 713 return node_name ? node_name->dev : NULL; 714 } 715 EXPORT_SYMBOL(__dev_get_by_name); 716 717 /** 718 * dev_get_by_name_rcu - find a device by its name 719 * @net: the applicable net namespace 720 * @name: name to find 721 * 722 * Find an interface by name. 723 * If the name is found a pointer to the device is returned. 724 * If the name is not found then %NULL is returned. 725 * The reference counters are not incremented so the caller must be 726 * careful with locks. The caller must hold RCU lock. 727 */ 728 729 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) 730 { 731 struct netdev_name_node *node_name; 732 733 node_name = netdev_name_node_lookup_rcu(net, name); 734 return node_name ? node_name->dev : NULL; 735 } 736 EXPORT_SYMBOL(dev_get_by_name_rcu); 737 738 /** 739 * dev_get_by_name - find a device by its name 740 * @net: the applicable net namespace 741 * @name: name to find 742 * 743 * Find an interface by name. This can be called from any 744 * context and does its own locking. The returned handle has 745 * the usage count incremented and the caller must use dev_put() to 746 * release it when it is no longer needed. %NULL is returned if no 747 * matching device is found. 748 */ 749 750 struct net_device *dev_get_by_name(struct net *net, const char *name) 751 { 752 struct net_device *dev; 753 754 rcu_read_lock(); 755 dev = dev_get_by_name_rcu(net, name); 756 dev_hold(dev); 757 rcu_read_unlock(); 758 return dev; 759 } 760 EXPORT_SYMBOL(dev_get_by_name); 761 762 /** 763 * __dev_get_by_index - find a device by its ifindex 764 * @net: the applicable net namespace 765 * @ifindex: index of device 766 * 767 * Search for an interface by index. Returns %NULL if the device 768 * is not found or a pointer to the device. The device has not 769 * had its reference counter increased so the caller must be careful 770 * about locking. The caller must hold either the RTNL semaphore 771 * or @dev_base_lock. 772 */ 773 774 struct net_device *__dev_get_by_index(struct net *net, int ifindex) 775 { 776 struct net_device *dev; 777 struct hlist_head *head = dev_index_hash(net, ifindex); 778 779 hlist_for_each_entry(dev, head, index_hlist) 780 if (dev->ifindex == ifindex) 781 return dev; 782 783 return NULL; 784 } 785 EXPORT_SYMBOL(__dev_get_by_index); 786 787 /** 788 * dev_get_by_index_rcu - find a device by its ifindex 789 * @net: the applicable net namespace 790 * @ifindex: index of device 791 * 792 * Search for an interface by index. Returns %NULL if the device 793 * is not found or a pointer to the device. The device has not 794 * had its reference counter increased so the caller must be careful 795 * about locking. The caller must hold RCU lock. 796 */ 797 798 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) 799 { 800 struct net_device *dev; 801 struct hlist_head *head = dev_index_hash(net, ifindex); 802 803 hlist_for_each_entry_rcu(dev, head, index_hlist) 804 if (dev->ifindex == ifindex) 805 return dev; 806 807 return NULL; 808 } 809 EXPORT_SYMBOL(dev_get_by_index_rcu); 810 811 812 /** 813 * dev_get_by_index - find a device by its ifindex 814 * @net: the applicable net namespace 815 * @ifindex: index of device 816 * 817 * Search for an interface by index. Returns NULL if the device 818 * is not found or a pointer to the device. The device returned has 819 * had a reference added and the pointer is safe until the user calls 820 * dev_put to indicate they have finished with it. 821 */ 822 823 struct net_device *dev_get_by_index(struct net *net, int ifindex) 824 { 825 struct net_device *dev; 826 827 rcu_read_lock(); 828 dev = dev_get_by_index_rcu(net, ifindex); 829 dev_hold(dev); 830 rcu_read_unlock(); 831 return dev; 832 } 833 EXPORT_SYMBOL(dev_get_by_index); 834 835 /** 836 * dev_get_by_napi_id - find a device by napi_id 837 * @napi_id: ID of the NAPI struct 838 * 839 * Search for an interface by NAPI ID. Returns %NULL if the device 840 * is not found or a pointer to the device. The device has not had 841 * its reference counter increased so the caller must be careful 842 * about locking. The caller must hold RCU lock. 843 */ 844 845 struct net_device *dev_get_by_napi_id(unsigned int napi_id) 846 { 847 struct napi_struct *napi; 848 849 WARN_ON_ONCE(!rcu_read_lock_held()); 850 851 if (napi_id < MIN_NAPI_ID) 852 return NULL; 853 854 napi = napi_by_id(napi_id); 855 856 return napi ? napi->dev : NULL; 857 } 858 EXPORT_SYMBOL(dev_get_by_napi_id); 859 860 /** 861 * netdev_get_name - get a netdevice name, knowing its ifindex. 862 * @net: network namespace 863 * @name: a pointer to the buffer where the name will be stored. 864 * @ifindex: the ifindex of the interface to get the name from. 865 */ 866 int netdev_get_name(struct net *net, char *name, int ifindex) 867 { 868 struct net_device *dev; 869 int ret; 870 871 down_read(&devnet_rename_sem); 872 rcu_read_lock(); 873 874 dev = dev_get_by_index_rcu(net, ifindex); 875 if (!dev) { 876 ret = -ENODEV; 877 goto out; 878 } 879 880 strcpy(name, dev->name); 881 882 ret = 0; 883 out: 884 rcu_read_unlock(); 885 up_read(&devnet_rename_sem); 886 return ret; 887 } 888 889 /** 890 * dev_getbyhwaddr_rcu - find a device by its hardware address 891 * @net: the applicable net namespace 892 * @type: media type of device 893 * @ha: hardware address 894 * 895 * Search for an interface by MAC address. Returns NULL if the device 896 * is not found or a pointer to the device. 897 * The caller must hold RCU or RTNL. 898 * The returned device has not had its ref count increased 899 * and the caller must therefore be careful about locking 900 * 901 */ 902 903 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, 904 const char *ha) 905 { 906 struct net_device *dev; 907 908 for_each_netdev_rcu(net, dev) 909 if (dev->type == type && 910 !memcmp(dev->dev_addr, ha, dev->addr_len)) 911 return dev; 912 913 return NULL; 914 } 915 EXPORT_SYMBOL(dev_getbyhwaddr_rcu); 916 917 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) 918 { 919 struct net_device *dev, *ret = NULL; 920 921 rcu_read_lock(); 922 for_each_netdev_rcu(net, dev) 923 if (dev->type == type) { 924 dev_hold(dev); 925 ret = dev; 926 break; 927 } 928 rcu_read_unlock(); 929 return ret; 930 } 931 EXPORT_SYMBOL(dev_getfirstbyhwtype); 932 933 /** 934 * __dev_get_by_flags - find any device with given flags 935 * @net: the applicable net namespace 936 * @if_flags: IFF_* values 937 * @mask: bitmask of bits in if_flags to check 938 * 939 * Search for any interface with the given flags. Returns NULL if a device 940 * is not found or a pointer to the device. Must be called inside 941 * rtnl_lock(), and result refcount is unchanged. 942 */ 943 944 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags, 945 unsigned short mask) 946 { 947 struct net_device *dev, *ret; 948 949 ASSERT_RTNL(); 950 951 ret = NULL; 952 for_each_netdev(net, dev) { 953 if (((dev->flags ^ if_flags) & mask) == 0) { 954 ret = dev; 955 break; 956 } 957 } 958 return ret; 959 } 960 EXPORT_SYMBOL(__dev_get_by_flags); 961 962 /** 963 * dev_valid_name - check if name is okay for network device 964 * @name: name string 965 * 966 * Network device names need to be valid file names to 967 * allow sysfs to work. We also disallow any kind of 968 * whitespace. 969 */ 970 bool dev_valid_name(const char *name) 971 { 972 if (*name == '\0') 973 return false; 974 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ) 975 return false; 976 if (!strcmp(name, ".") || !strcmp(name, "..")) 977 return false; 978 979 while (*name) { 980 if (*name == '/' || *name == ':' || isspace(*name)) 981 return false; 982 name++; 983 } 984 return true; 985 } 986 EXPORT_SYMBOL(dev_valid_name); 987 988 /** 989 * __dev_alloc_name - allocate a name for a device 990 * @net: network namespace to allocate the device name in 991 * @name: name format string 992 * @buf: scratch buffer and result name string 993 * 994 * Passed a format string - eg "lt%d" it will try and find a suitable 995 * id. It scans list of devices to build up a free map, then chooses 996 * the first empty slot. The caller must hold the dev_base or rtnl lock 997 * while allocating the name and adding the device in order to avoid 998 * duplicates. 999 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1000 * Returns the number of the unit assigned or a negative errno code. 1001 */ 1002 1003 static int __dev_alloc_name(struct net *net, const char *name, char *buf) 1004 { 1005 int i = 0; 1006 const char *p; 1007 const int max_netdevices = 8*PAGE_SIZE; 1008 unsigned long *inuse; 1009 struct net_device *d; 1010 1011 if (!dev_valid_name(name)) 1012 return -EINVAL; 1013 1014 p = strchr(name, '%'); 1015 if (p) { 1016 /* 1017 * Verify the string as this thing may have come from 1018 * the user. There must be either one "%d" and no other "%" 1019 * characters. 1020 */ 1021 if (p[1] != 'd' || strchr(p + 2, '%')) 1022 return -EINVAL; 1023 1024 /* Use one page as a bit array of possible slots */ 1025 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); 1026 if (!inuse) 1027 return -ENOMEM; 1028 1029 for_each_netdev(net, d) { 1030 struct netdev_name_node *name_node; 1031 list_for_each_entry(name_node, &d->name_node->list, list) { 1032 if (!sscanf(name_node->name, name, &i)) 1033 continue; 1034 if (i < 0 || i >= max_netdevices) 1035 continue; 1036 1037 /* avoid cases where sscanf is not exact inverse of printf */ 1038 snprintf(buf, IFNAMSIZ, name, i); 1039 if (!strncmp(buf, name_node->name, IFNAMSIZ)) 1040 set_bit(i, inuse); 1041 } 1042 if (!sscanf(d->name, name, &i)) 1043 continue; 1044 if (i < 0 || i >= max_netdevices) 1045 continue; 1046 1047 /* avoid cases where sscanf is not exact inverse of printf */ 1048 snprintf(buf, IFNAMSIZ, name, i); 1049 if (!strncmp(buf, d->name, IFNAMSIZ)) 1050 set_bit(i, inuse); 1051 } 1052 1053 i = find_first_zero_bit(inuse, max_netdevices); 1054 free_page((unsigned long) inuse); 1055 } 1056 1057 snprintf(buf, IFNAMSIZ, name, i); 1058 if (!netdev_name_in_use(net, buf)) 1059 return i; 1060 1061 /* It is possible to run out of possible slots 1062 * when the name is long and there isn't enough space left 1063 * for the digits, or if all bits are used. 1064 */ 1065 return -ENFILE; 1066 } 1067 1068 static int dev_alloc_name_ns(struct net *net, 1069 struct net_device *dev, 1070 const char *name) 1071 { 1072 char buf[IFNAMSIZ]; 1073 int ret; 1074 1075 BUG_ON(!net); 1076 ret = __dev_alloc_name(net, name, buf); 1077 if (ret >= 0) 1078 strlcpy(dev->name, buf, IFNAMSIZ); 1079 return ret; 1080 } 1081 1082 /** 1083 * dev_alloc_name - allocate a name for a device 1084 * @dev: device 1085 * @name: name format string 1086 * 1087 * Passed a format string - eg "lt%d" it will try and find a suitable 1088 * id. It scans list of devices to build up a free map, then chooses 1089 * the first empty slot. The caller must hold the dev_base or rtnl lock 1090 * while allocating the name and adding the device in order to avoid 1091 * duplicates. 1092 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1093 * Returns the number of the unit assigned or a negative errno code. 1094 */ 1095 1096 int dev_alloc_name(struct net_device *dev, const char *name) 1097 { 1098 return dev_alloc_name_ns(dev_net(dev), dev, name); 1099 } 1100 EXPORT_SYMBOL(dev_alloc_name); 1101 1102 static int dev_get_valid_name(struct net *net, struct net_device *dev, 1103 const char *name) 1104 { 1105 BUG_ON(!net); 1106 1107 if (!dev_valid_name(name)) 1108 return -EINVAL; 1109 1110 if (strchr(name, '%')) 1111 return dev_alloc_name_ns(net, dev, name); 1112 else if (netdev_name_in_use(net, name)) 1113 return -EEXIST; 1114 else if (dev->name != name) 1115 strlcpy(dev->name, name, IFNAMSIZ); 1116 1117 return 0; 1118 } 1119 1120 /** 1121 * dev_change_name - change name of a device 1122 * @dev: device 1123 * @newname: name (or format string) must be at least IFNAMSIZ 1124 * 1125 * Change name of a device, can pass format strings "eth%d". 1126 * for wildcarding. 1127 */ 1128 int dev_change_name(struct net_device *dev, const char *newname) 1129 { 1130 unsigned char old_assign_type; 1131 char oldname[IFNAMSIZ]; 1132 int err = 0; 1133 int ret; 1134 struct net *net; 1135 1136 ASSERT_RTNL(); 1137 BUG_ON(!dev_net(dev)); 1138 1139 net = dev_net(dev); 1140 1141 /* Some auto-enslaved devices e.g. failover slaves are 1142 * special, as userspace might rename the device after 1143 * the interface had been brought up and running since 1144 * the point kernel initiated auto-enslavement. Allow 1145 * live name change even when these slave devices are 1146 * up and running. 1147 * 1148 * Typically, users of these auto-enslaving devices 1149 * don't actually care about slave name change, as 1150 * they are supposed to operate on master interface 1151 * directly. 1152 */ 1153 if (dev->flags & IFF_UP && 1154 likely(!(dev->priv_flags & IFF_LIVE_RENAME_OK))) 1155 return -EBUSY; 1156 1157 down_write(&devnet_rename_sem); 1158 1159 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { 1160 up_write(&devnet_rename_sem); 1161 return 0; 1162 } 1163 1164 memcpy(oldname, dev->name, IFNAMSIZ); 1165 1166 err = dev_get_valid_name(net, dev, newname); 1167 if (err < 0) { 1168 up_write(&devnet_rename_sem); 1169 return err; 1170 } 1171 1172 if (oldname[0] && !strchr(oldname, '%')) 1173 netdev_info(dev, "renamed from %s\n", oldname); 1174 1175 old_assign_type = dev->name_assign_type; 1176 dev->name_assign_type = NET_NAME_RENAMED; 1177 1178 rollback: 1179 ret = device_rename(&dev->dev, dev->name); 1180 if (ret) { 1181 memcpy(dev->name, oldname, IFNAMSIZ); 1182 dev->name_assign_type = old_assign_type; 1183 up_write(&devnet_rename_sem); 1184 return ret; 1185 } 1186 1187 up_write(&devnet_rename_sem); 1188 1189 netdev_adjacent_rename_links(dev, oldname); 1190 1191 write_lock(&dev_base_lock); 1192 netdev_name_node_del(dev->name_node); 1193 write_unlock(&dev_base_lock); 1194 1195 synchronize_rcu(); 1196 1197 write_lock(&dev_base_lock); 1198 netdev_name_node_add(net, dev->name_node); 1199 write_unlock(&dev_base_lock); 1200 1201 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); 1202 ret = notifier_to_errno(ret); 1203 1204 if (ret) { 1205 /* err >= 0 after dev_alloc_name() or stores the first errno */ 1206 if (err >= 0) { 1207 err = ret; 1208 down_write(&devnet_rename_sem); 1209 memcpy(dev->name, oldname, IFNAMSIZ); 1210 memcpy(oldname, newname, IFNAMSIZ); 1211 dev->name_assign_type = old_assign_type; 1212 old_assign_type = NET_NAME_RENAMED; 1213 goto rollback; 1214 } else { 1215 netdev_err(dev, "name change rollback failed: %d\n", 1216 ret); 1217 } 1218 } 1219 1220 return err; 1221 } 1222 1223 /** 1224 * dev_set_alias - change ifalias of a device 1225 * @dev: device 1226 * @alias: name up to IFALIASZ 1227 * @len: limit of bytes to copy from info 1228 * 1229 * Set ifalias for a device, 1230 */ 1231 int dev_set_alias(struct net_device *dev, const char *alias, size_t len) 1232 { 1233 struct dev_ifalias *new_alias = NULL; 1234 1235 if (len >= IFALIASZ) 1236 return -EINVAL; 1237 1238 if (len) { 1239 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL); 1240 if (!new_alias) 1241 return -ENOMEM; 1242 1243 memcpy(new_alias->ifalias, alias, len); 1244 new_alias->ifalias[len] = 0; 1245 } 1246 1247 mutex_lock(&ifalias_mutex); 1248 new_alias = rcu_replace_pointer(dev->ifalias, new_alias, 1249 mutex_is_locked(&ifalias_mutex)); 1250 mutex_unlock(&ifalias_mutex); 1251 1252 if (new_alias) 1253 kfree_rcu(new_alias, rcuhead); 1254 1255 return len; 1256 } 1257 EXPORT_SYMBOL(dev_set_alias); 1258 1259 /** 1260 * dev_get_alias - get ifalias of a device 1261 * @dev: device 1262 * @name: buffer to store name of ifalias 1263 * @len: size of buffer 1264 * 1265 * get ifalias for a device. Caller must make sure dev cannot go 1266 * away, e.g. rcu read lock or own a reference count to device. 1267 */ 1268 int dev_get_alias(const struct net_device *dev, char *name, size_t len) 1269 { 1270 const struct dev_ifalias *alias; 1271 int ret = 0; 1272 1273 rcu_read_lock(); 1274 alias = rcu_dereference(dev->ifalias); 1275 if (alias) 1276 ret = snprintf(name, len, "%s", alias->ifalias); 1277 rcu_read_unlock(); 1278 1279 return ret; 1280 } 1281 1282 /** 1283 * netdev_features_change - device changes features 1284 * @dev: device to cause notification 1285 * 1286 * Called to indicate a device has changed features. 1287 */ 1288 void netdev_features_change(struct net_device *dev) 1289 { 1290 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); 1291 } 1292 EXPORT_SYMBOL(netdev_features_change); 1293 1294 /** 1295 * netdev_state_change - device changes state 1296 * @dev: device to cause notification 1297 * 1298 * Called to indicate a device has changed state. This function calls 1299 * the notifier chains for netdev_chain and sends a NEWLINK message 1300 * to the routing socket. 1301 */ 1302 void netdev_state_change(struct net_device *dev) 1303 { 1304 if (dev->flags & IFF_UP) { 1305 struct netdev_notifier_change_info change_info = { 1306 .info.dev = dev, 1307 }; 1308 1309 call_netdevice_notifiers_info(NETDEV_CHANGE, 1310 &change_info.info); 1311 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL); 1312 } 1313 } 1314 EXPORT_SYMBOL(netdev_state_change); 1315 1316 /** 1317 * __netdev_notify_peers - notify network peers about existence of @dev, 1318 * to be called when rtnl lock is already held. 1319 * @dev: network device 1320 * 1321 * Generate traffic such that interested network peers are aware of 1322 * @dev, such as by generating a gratuitous ARP. This may be used when 1323 * a device wants to inform the rest of the network about some sort of 1324 * reconfiguration such as a failover event or virtual machine 1325 * migration. 1326 */ 1327 void __netdev_notify_peers(struct net_device *dev) 1328 { 1329 ASSERT_RTNL(); 1330 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); 1331 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev); 1332 } 1333 EXPORT_SYMBOL(__netdev_notify_peers); 1334 1335 /** 1336 * netdev_notify_peers - notify network peers about existence of @dev 1337 * @dev: network device 1338 * 1339 * Generate traffic such that interested network peers are aware of 1340 * @dev, such as by generating a gratuitous ARP. This may be used when 1341 * a device wants to inform the rest of the network about some sort of 1342 * reconfiguration such as a failover event or virtual machine 1343 * migration. 1344 */ 1345 void netdev_notify_peers(struct net_device *dev) 1346 { 1347 rtnl_lock(); 1348 __netdev_notify_peers(dev); 1349 rtnl_unlock(); 1350 } 1351 EXPORT_SYMBOL(netdev_notify_peers); 1352 1353 static int napi_threaded_poll(void *data); 1354 1355 static int napi_kthread_create(struct napi_struct *n) 1356 { 1357 int err = 0; 1358 1359 /* Create and wake up the kthread once to put it in 1360 * TASK_INTERRUPTIBLE mode to avoid the blocked task 1361 * warning and work with loadavg. 1362 */ 1363 n->thread = kthread_run(napi_threaded_poll, n, "napi/%s-%d", 1364 n->dev->name, n->napi_id); 1365 if (IS_ERR(n->thread)) { 1366 err = PTR_ERR(n->thread); 1367 pr_err("kthread_run failed with err %d\n", err); 1368 n->thread = NULL; 1369 } 1370 1371 return err; 1372 } 1373 1374 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1375 { 1376 const struct net_device_ops *ops = dev->netdev_ops; 1377 int ret; 1378 1379 ASSERT_RTNL(); 1380 dev_addr_check(dev); 1381 1382 if (!netif_device_present(dev)) { 1383 /* may be detached because parent is runtime-suspended */ 1384 if (dev->dev.parent) 1385 pm_runtime_resume(dev->dev.parent); 1386 if (!netif_device_present(dev)) 1387 return -ENODEV; 1388 } 1389 1390 /* Block netpoll from trying to do any rx path servicing. 1391 * If we don't do this there is a chance ndo_poll_controller 1392 * or ndo_poll may be running while we open the device 1393 */ 1394 netpoll_poll_disable(dev); 1395 1396 ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack); 1397 ret = notifier_to_errno(ret); 1398 if (ret) 1399 return ret; 1400 1401 set_bit(__LINK_STATE_START, &dev->state); 1402 1403 if (ops->ndo_validate_addr) 1404 ret = ops->ndo_validate_addr(dev); 1405 1406 if (!ret && ops->ndo_open) 1407 ret = ops->ndo_open(dev); 1408 1409 netpoll_poll_enable(dev); 1410 1411 if (ret) 1412 clear_bit(__LINK_STATE_START, &dev->state); 1413 else { 1414 dev->flags |= IFF_UP; 1415 dev_set_rx_mode(dev); 1416 dev_activate(dev); 1417 add_device_randomness(dev->dev_addr, dev->addr_len); 1418 } 1419 1420 return ret; 1421 } 1422 1423 /** 1424 * dev_open - prepare an interface for use. 1425 * @dev: device to open 1426 * @extack: netlink extended ack 1427 * 1428 * Takes a device from down to up state. The device's private open 1429 * function is invoked and then the multicast lists are loaded. Finally 1430 * the device is moved into the up state and a %NETDEV_UP message is 1431 * sent to the netdev notifier chain. 1432 * 1433 * Calling this function on an active interface is a nop. On a failure 1434 * a negative errno code is returned. 1435 */ 1436 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1437 { 1438 int ret; 1439 1440 if (dev->flags & IFF_UP) 1441 return 0; 1442 1443 ret = __dev_open(dev, extack); 1444 if (ret < 0) 1445 return ret; 1446 1447 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); 1448 call_netdevice_notifiers(NETDEV_UP, dev); 1449 1450 return ret; 1451 } 1452 EXPORT_SYMBOL(dev_open); 1453 1454 static void __dev_close_many(struct list_head *head) 1455 { 1456 struct net_device *dev; 1457 1458 ASSERT_RTNL(); 1459 might_sleep(); 1460 1461 list_for_each_entry(dev, head, close_list) { 1462 /* Temporarily disable netpoll until the interface is down */ 1463 netpoll_poll_disable(dev); 1464 1465 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); 1466 1467 clear_bit(__LINK_STATE_START, &dev->state); 1468 1469 /* Synchronize to scheduled poll. We cannot touch poll list, it 1470 * can be even on different cpu. So just clear netif_running(). 1471 * 1472 * dev->stop() will invoke napi_disable() on all of it's 1473 * napi_struct instances on this device. 1474 */ 1475 smp_mb__after_atomic(); /* Commit netif_running(). */ 1476 } 1477 1478 dev_deactivate_many(head); 1479 1480 list_for_each_entry(dev, head, close_list) { 1481 const struct net_device_ops *ops = dev->netdev_ops; 1482 1483 /* 1484 * Call the device specific close. This cannot fail. 1485 * Only if device is UP 1486 * 1487 * We allow it to be called even after a DETACH hot-plug 1488 * event. 1489 */ 1490 if (ops->ndo_stop) 1491 ops->ndo_stop(dev); 1492 1493 dev->flags &= ~IFF_UP; 1494 netpoll_poll_enable(dev); 1495 } 1496 } 1497 1498 static void __dev_close(struct net_device *dev) 1499 { 1500 LIST_HEAD(single); 1501 1502 list_add(&dev->close_list, &single); 1503 __dev_close_many(&single); 1504 list_del(&single); 1505 } 1506 1507 void dev_close_many(struct list_head *head, bool unlink) 1508 { 1509 struct net_device *dev, *tmp; 1510 1511 /* Remove the devices that don't need to be closed */ 1512 list_for_each_entry_safe(dev, tmp, head, close_list) 1513 if (!(dev->flags & IFF_UP)) 1514 list_del_init(&dev->close_list); 1515 1516 __dev_close_many(head); 1517 1518 list_for_each_entry_safe(dev, tmp, head, close_list) { 1519 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); 1520 call_netdevice_notifiers(NETDEV_DOWN, dev); 1521 if (unlink) 1522 list_del_init(&dev->close_list); 1523 } 1524 } 1525 EXPORT_SYMBOL(dev_close_many); 1526 1527 /** 1528 * dev_close - shutdown an interface. 1529 * @dev: device to shutdown 1530 * 1531 * This function moves an active device into down state. A 1532 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device 1533 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier 1534 * chain. 1535 */ 1536 void dev_close(struct net_device *dev) 1537 { 1538 if (dev->flags & IFF_UP) { 1539 LIST_HEAD(single); 1540 1541 list_add(&dev->close_list, &single); 1542 dev_close_many(&single, true); 1543 list_del(&single); 1544 } 1545 } 1546 EXPORT_SYMBOL(dev_close); 1547 1548 1549 /** 1550 * dev_disable_lro - disable Large Receive Offload on a device 1551 * @dev: device 1552 * 1553 * Disable Large Receive Offload (LRO) on a net device. Must be 1554 * called under RTNL. This is needed if received packets may be 1555 * forwarded to another interface. 1556 */ 1557 void dev_disable_lro(struct net_device *dev) 1558 { 1559 struct net_device *lower_dev; 1560 struct list_head *iter; 1561 1562 dev->wanted_features &= ~NETIF_F_LRO; 1563 netdev_update_features(dev); 1564 1565 if (unlikely(dev->features & NETIF_F_LRO)) 1566 netdev_WARN(dev, "failed to disable LRO!\n"); 1567 1568 netdev_for_each_lower_dev(dev, lower_dev, iter) 1569 dev_disable_lro(lower_dev); 1570 } 1571 EXPORT_SYMBOL(dev_disable_lro); 1572 1573 /** 1574 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device 1575 * @dev: device 1576 * 1577 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be 1578 * called under RTNL. This is needed if Generic XDP is installed on 1579 * the device. 1580 */ 1581 static void dev_disable_gro_hw(struct net_device *dev) 1582 { 1583 dev->wanted_features &= ~NETIF_F_GRO_HW; 1584 netdev_update_features(dev); 1585 1586 if (unlikely(dev->features & NETIF_F_GRO_HW)) 1587 netdev_WARN(dev, "failed to disable GRO_HW!\n"); 1588 } 1589 1590 const char *netdev_cmd_to_name(enum netdev_cmd cmd) 1591 { 1592 #define N(val) \ 1593 case NETDEV_##val: \ 1594 return "NETDEV_" __stringify(val); 1595 switch (cmd) { 1596 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER) 1597 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE) 1598 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE) 1599 N(POST_INIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) N(CHANGEUPPER) 1600 N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) N(BONDING_INFO) 1601 N(PRECHANGEUPPER) N(CHANGELOWERSTATE) N(UDP_TUNNEL_PUSH_INFO) 1602 N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN) 1603 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO) 1604 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO) 1605 N(PRE_CHANGEADDR) 1606 } 1607 #undef N 1608 return "UNKNOWN_NETDEV_EVENT"; 1609 } 1610 EXPORT_SYMBOL_GPL(netdev_cmd_to_name); 1611 1612 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, 1613 struct net_device *dev) 1614 { 1615 struct netdev_notifier_info info = { 1616 .dev = dev, 1617 }; 1618 1619 return nb->notifier_call(nb, val, &info); 1620 } 1621 1622 static int call_netdevice_register_notifiers(struct notifier_block *nb, 1623 struct net_device *dev) 1624 { 1625 int err; 1626 1627 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); 1628 err = notifier_to_errno(err); 1629 if (err) 1630 return err; 1631 1632 if (!(dev->flags & IFF_UP)) 1633 return 0; 1634 1635 call_netdevice_notifier(nb, NETDEV_UP, dev); 1636 return 0; 1637 } 1638 1639 static void call_netdevice_unregister_notifiers(struct notifier_block *nb, 1640 struct net_device *dev) 1641 { 1642 if (dev->flags & IFF_UP) { 1643 call_netdevice_notifier(nb, NETDEV_GOING_DOWN, 1644 dev); 1645 call_netdevice_notifier(nb, NETDEV_DOWN, dev); 1646 } 1647 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); 1648 } 1649 1650 static int call_netdevice_register_net_notifiers(struct notifier_block *nb, 1651 struct net *net) 1652 { 1653 struct net_device *dev; 1654 int err; 1655 1656 for_each_netdev(net, dev) { 1657 err = call_netdevice_register_notifiers(nb, dev); 1658 if (err) 1659 goto rollback; 1660 } 1661 return 0; 1662 1663 rollback: 1664 for_each_netdev_continue_reverse(net, dev) 1665 call_netdevice_unregister_notifiers(nb, dev); 1666 return err; 1667 } 1668 1669 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb, 1670 struct net *net) 1671 { 1672 struct net_device *dev; 1673 1674 for_each_netdev(net, dev) 1675 call_netdevice_unregister_notifiers(nb, dev); 1676 } 1677 1678 static int dev_boot_phase = 1; 1679 1680 /** 1681 * register_netdevice_notifier - register a network notifier block 1682 * @nb: notifier 1683 * 1684 * Register a notifier to be called when network device events occur. 1685 * The notifier passed is linked into the kernel structures and must 1686 * not be reused until it has been unregistered. A negative errno code 1687 * is returned on a failure. 1688 * 1689 * When registered all registration and up events are replayed 1690 * to the new notifier to allow device to have a race free 1691 * view of the network device list. 1692 */ 1693 1694 int register_netdevice_notifier(struct notifier_block *nb) 1695 { 1696 struct net *net; 1697 int err; 1698 1699 /* Close race with setup_net() and cleanup_net() */ 1700 down_write(&pernet_ops_rwsem); 1701 rtnl_lock(); 1702 err = raw_notifier_chain_register(&netdev_chain, nb); 1703 if (err) 1704 goto unlock; 1705 if (dev_boot_phase) 1706 goto unlock; 1707 for_each_net(net) { 1708 err = call_netdevice_register_net_notifiers(nb, net); 1709 if (err) 1710 goto rollback; 1711 } 1712 1713 unlock: 1714 rtnl_unlock(); 1715 up_write(&pernet_ops_rwsem); 1716 return err; 1717 1718 rollback: 1719 for_each_net_continue_reverse(net) 1720 call_netdevice_unregister_net_notifiers(nb, net); 1721 1722 raw_notifier_chain_unregister(&netdev_chain, nb); 1723 goto unlock; 1724 } 1725 EXPORT_SYMBOL(register_netdevice_notifier); 1726 1727 /** 1728 * unregister_netdevice_notifier - unregister a network notifier block 1729 * @nb: notifier 1730 * 1731 * Unregister a notifier previously registered by 1732 * register_netdevice_notifier(). The notifier is unlinked into the 1733 * kernel structures and may then be reused. A negative errno code 1734 * is returned on a failure. 1735 * 1736 * After unregistering unregister and down device events are synthesized 1737 * for all devices on the device list to the removed notifier to remove 1738 * the need for special case cleanup code. 1739 */ 1740 1741 int unregister_netdevice_notifier(struct notifier_block *nb) 1742 { 1743 struct net *net; 1744 int err; 1745 1746 /* Close race with setup_net() and cleanup_net() */ 1747 down_write(&pernet_ops_rwsem); 1748 rtnl_lock(); 1749 err = raw_notifier_chain_unregister(&netdev_chain, nb); 1750 if (err) 1751 goto unlock; 1752 1753 for_each_net(net) 1754 call_netdevice_unregister_net_notifiers(nb, net); 1755 1756 unlock: 1757 rtnl_unlock(); 1758 up_write(&pernet_ops_rwsem); 1759 return err; 1760 } 1761 EXPORT_SYMBOL(unregister_netdevice_notifier); 1762 1763 static int __register_netdevice_notifier_net(struct net *net, 1764 struct notifier_block *nb, 1765 bool ignore_call_fail) 1766 { 1767 int err; 1768 1769 err = raw_notifier_chain_register(&net->netdev_chain, nb); 1770 if (err) 1771 return err; 1772 if (dev_boot_phase) 1773 return 0; 1774 1775 err = call_netdevice_register_net_notifiers(nb, net); 1776 if (err && !ignore_call_fail) 1777 goto chain_unregister; 1778 1779 return 0; 1780 1781 chain_unregister: 1782 raw_notifier_chain_unregister(&net->netdev_chain, nb); 1783 return err; 1784 } 1785 1786 static int __unregister_netdevice_notifier_net(struct net *net, 1787 struct notifier_block *nb) 1788 { 1789 int err; 1790 1791 err = raw_notifier_chain_unregister(&net->netdev_chain, nb); 1792 if (err) 1793 return err; 1794 1795 call_netdevice_unregister_net_notifiers(nb, net); 1796 return 0; 1797 } 1798 1799 /** 1800 * register_netdevice_notifier_net - register a per-netns network notifier block 1801 * @net: network namespace 1802 * @nb: notifier 1803 * 1804 * Register a notifier to be called when network device events occur. 1805 * The notifier passed is linked into the kernel structures and must 1806 * not be reused until it has been unregistered. A negative errno code 1807 * is returned on a failure. 1808 * 1809 * When registered all registration and up events are replayed 1810 * to the new notifier to allow device to have a race free 1811 * view of the network device list. 1812 */ 1813 1814 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb) 1815 { 1816 int err; 1817 1818 rtnl_lock(); 1819 err = __register_netdevice_notifier_net(net, nb, false); 1820 rtnl_unlock(); 1821 return err; 1822 } 1823 EXPORT_SYMBOL(register_netdevice_notifier_net); 1824 1825 /** 1826 * unregister_netdevice_notifier_net - unregister a per-netns 1827 * network notifier block 1828 * @net: network namespace 1829 * @nb: notifier 1830 * 1831 * Unregister a notifier previously registered by 1832 * register_netdevice_notifier(). The notifier is unlinked into the 1833 * kernel structures and may then be reused. A negative errno code 1834 * is returned on a failure. 1835 * 1836 * After unregistering unregister and down device events are synthesized 1837 * for all devices on the device list to the removed notifier to remove 1838 * the need for special case cleanup code. 1839 */ 1840 1841 int unregister_netdevice_notifier_net(struct net *net, 1842 struct notifier_block *nb) 1843 { 1844 int err; 1845 1846 rtnl_lock(); 1847 err = __unregister_netdevice_notifier_net(net, nb); 1848 rtnl_unlock(); 1849 return err; 1850 } 1851 EXPORT_SYMBOL(unregister_netdevice_notifier_net); 1852 1853 int register_netdevice_notifier_dev_net(struct net_device *dev, 1854 struct notifier_block *nb, 1855 struct netdev_net_notifier *nn) 1856 { 1857 int err; 1858 1859 rtnl_lock(); 1860 err = __register_netdevice_notifier_net(dev_net(dev), nb, false); 1861 if (!err) { 1862 nn->nb = nb; 1863 list_add(&nn->list, &dev->net_notifier_list); 1864 } 1865 rtnl_unlock(); 1866 return err; 1867 } 1868 EXPORT_SYMBOL(register_netdevice_notifier_dev_net); 1869 1870 int unregister_netdevice_notifier_dev_net(struct net_device *dev, 1871 struct notifier_block *nb, 1872 struct netdev_net_notifier *nn) 1873 { 1874 int err; 1875 1876 rtnl_lock(); 1877 list_del(&nn->list); 1878 err = __unregister_netdevice_notifier_net(dev_net(dev), nb); 1879 rtnl_unlock(); 1880 return err; 1881 } 1882 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net); 1883 1884 static void move_netdevice_notifiers_dev_net(struct net_device *dev, 1885 struct net *net) 1886 { 1887 struct netdev_net_notifier *nn; 1888 1889 list_for_each_entry(nn, &dev->net_notifier_list, list) { 1890 __unregister_netdevice_notifier_net(dev_net(dev), nn->nb); 1891 __register_netdevice_notifier_net(net, nn->nb, true); 1892 } 1893 } 1894 1895 /** 1896 * call_netdevice_notifiers_info - call all network notifier blocks 1897 * @val: value passed unmodified to notifier function 1898 * @info: notifier information data 1899 * 1900 * Call all network notifier blocks. Parameters and return value 1901 * are as for raw_notifier_call_chain(). 1902 */ 1903 1904 static int call_netdevice_notifiers_info(unsigned long val, 1905 struct netdev_notifier_info *info) 1906 { 1907 struct net *net = dev_net(info->dev); 1908 int ret; 1909 1910 ASSERT_RTNL(); 1911 1912 /* Run per-netns notifier block chain first, then run the global one. 1913 * Hopefully, one day, the global one is going to be removed after 1914 * all notifier block registrators get converted to be per-netns. 1915 */ 1916 ret = raw_notifier_call_chain(&net->netdev_chain, val, info); 1917 if (ret & NOTIFY_STOP_MASK) 1918 return ret; 1919 return raw_notifier_call_chain(&netdev_chain, val, info); 1920 } 1921 1922 static int call_netdevice_notifiers_extack(unsigned long val, 1923 struct net_device *dev, 1924 struct netlink_ext_ack *extack) 1925 { 1926 struct netdev_notifier_info info = { 1927 .dev = dev, 1928 .extack = extack, 1929 }; 1930 1931 return call_netdevice_notifiers_info(val, &info); 1932 } 1933 1934 /** 1935 * call_netdevice_notifiers - call all network notifier blocks 1936 * @val: value passed unmodified to notifier function 1937 * @dev: net_device pointer passed unmodified to notifier function 1938 * 1939 * Call all network notifier blocks. Parameters and return value 1940 * are as for raw_notifier_call_chain(). 1941 */ 1942 1943 int call_netdevice_notifiers(unsigned long val, struct net_device *dev) 1944 { 1945 return call_netdevice_notifiers_extack(val, dev, NULL); 1946 } 1947 EXPORT_SYMBOL(call_netdevice_notifiers); 1948 1949 /** 1950 * call_netdevice_notifiers_mtu - call all network notifier blocks 1951 * @val: value passed unmodified to notifier function 1952 * @dev: net_device pointer passed unmodified to notifier function 1953 * @arg: additional u32 argument passed to the notifier function 1954 * 1955 * Call all network notifier blocks. Parameters and return value 1956 * are as for raw_notifier_call_chain(). 1957 */ 1958 static int call_netdevice_notifiers_mtu(unsigned long val, 1959 struct net_device *dev, u32 arg) 1960 { 1961 struct netdev_notifier_info_ext info = { 1962 .info.dev = dev, 1963 .ext.mtu = arg, 1964 }; 1965 1966 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0); 1967 1968 return call_netdevice_notifiers_info(val, &info.info); 1969 } 1970 1971 #ifdef CONFIG_NET_INGRESS 1972 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key); 1973 1974 void net_inc_ingress_queue(void) 1975 { 1976 static_branch_inc(&ingress_needed_key); 1977 } 1978 EXPORT_SYMBOL_GPL(net_inc_ingress_queue); 1979 1980 void net_dec_ingress_queue(void) 1981 { 1982 static_branch_dec(&ingress_needed_key); 1983 } 1984 EXPORT_SYMBOL_GPL(net_dec_ingress_queue); 1985 #endif 1986 1987 #ifdef CONFIG_NET_EGRESS 1988 static DEFINE_STATIC_KEY_FALSE(egress_needed_key); 1989 1990 void net_inc_egress_queue(void) 1991 { 1992 static_branch_inc(&egress_needed_key); 1993 } 1994 EXPORT_SYMBOL_GPL(net_inc_egress_queue); 1995 1996 void net_dec_egress_queue(void) 1997 { 1998 static_branch_dec(&egress_needed_key); 1999 } 2000 EXPORT_SYMBOL_GPL(net_dec_egress_queue); 2001 #endif 2002 2003 static DEFINE_STATIC_KEY_FALSE(netstamp_needed_key); 2004 #ifdef CONFIG_JUMP_LABEL 2005 static atomic_t netstamp_needed_deferred; 2006 static atomic_t netstamp_wanted; 2007 static void netstamp_clear(struct work_struct *work) 2008 { 2009 int deferred = atomic_xchg(&netstamp_needed_deferred, 0); 2010 int wanted; 2011 2012 wanted = atomic_add_return(deferred, &netstamp_wanted); 2013 if (wanted > 0) 2014 static_branch_enable(&netstamp_needed_key); 2015 else 2016 static_branch_disable(&netstamp_needed_key); 2017 } 2018 static DECLARE_WORK(netstamp_work, netstamp_clear); 2019 #endif 2020 2021 void net_enable_timestamp(void) 2022 { 2023 #ifdef CONFIG_JUMP_LABEL 2024 int wanted; 2025 2026 while (1) { 2027 wanted = atomic_read(&netstamp_wanted); 2028 if (wanted <= 0) 2029 break; 2030 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted) 2031 return; 2032 } 2033 atomic_inc(&netstamp_needed_deferred); 2034 schedule_work(&netstamp_work); 2035 #else 2036 static_branch_inc(&netstamp_needed_key); 2037 #endif 2038 } 2039 EXPORT_SYMBOL(net_enable_timestamp); 2040 2041 void net_disable_timestamp(void) 2042 { 2043 #ifdef CONFIG_JUMP_LABEL 2044 int wanted; 2045 2046 while (1) { 2047 wanted = atomic_read(&netstamp_wanted); 2048 if (wanted <= 1) 2049 break; 2050 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted) 2051 return; 2052 } 2053 atomic_dec(&netstamp_needed_deferred); 2054 schedule_work(&netstamp_work); 2055 #else 2056 static_branch_dec(&netstamp_needed_key); 2057 #endif 2058 } 2059 EXPORT_SYMBOL(net_disable_timestamp); 2060 2061 static inline void net_timestamp_set(struct sk_buff *skb) 2062 { 2063 skb->tstamp = 0; 2064 if (static_branch_unlikely(&netstamp_needed_key)) 2065 __net_timestamp(skb); 2066 } 2067 2068 #define net_timestamp_check(COND, SKB) \ 2069 if (static_branch_unlikely(&netstamp_needed_key)) { \ 2070 if ((COND) && !(SKB)->tstamp) \ 2071 __net_timestamp(SKB); \ 2072 } \ 2073 2074 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb) 2075 { 2076 return __is_skb_forwardable(dev, skb, true); 2077 } 2078 EXPORT_SYMBOL_GPL(is_skb_forwardable); 2079 2080 static int __dev_forward_skb2(struct net_device *dev, struct sk_buff *skb, 2081 bool check_mtu) 2082 { 2083 int ret = ____dev_forward_skb(dev, skb, check_mtu); 2084 2085 if (likely(!ret)) { 2086 skb->protocol = eth_type_trans(skb, dev); 2087 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); 2088 } 2089 2090 return ret; 2091 } 2092 2093 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2094 { 2095 return __dev_forward_skb2(dev, skb, true); 2096 } 2097 EXPORT_SYMBOL_GPL(__dev_forward_skb); 2098 2099 /** 2100 * dev_forward_skb - loopback an skb to another netif 2101 * 2102 * @dev: destination network device 2103 * @skb: buffer to forward 2104 * 2105 * return values: 2106 * NET_RX_SUCCESS (no congestion) 2107 * NET_RX_DROP (packet was dropped, but freed) 2108 * 2109 * dev_forward_skb can be used for injecting an skb from the 2110 * start_xmit function of one device into the receive queue 2111 * of another device. 2112 * 2113 * The receiving device may be in another namespace, so 2114 * we have to clear all information in the skb that could 2115 * impact namespace isolation. 2116 */ 2117 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2118 { 2119 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); 2120 } 2121 EXPORT_SYMBOL_GPL(dev_forward_skb); 2122 2123 int dev_forward_skb_nomtu(struct net_device *dev, struct sk_buff *skb) 2124 { 2125 return __dev_forward_skb2(dev, skb, false) ?: netif_rx_internal(skb); 2126 } 2127 2128 static inline int deliver_skb(struct sk_buff *skb, 2129 struct packet_type *pt_prev, 2130 struct net_device *orig_dev) 2131 { 2132 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 2133 return -ENOMEM; 2134 refcount_inc(&skb->users); 2135 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 2136 } 2137 2138 static inline void deliver_ptype_list_skb(struct sk_buff *skb, 2139 struct packet_type **pt, 2140 struct net_device *orig_dev, 2141 __be16 type, 2142 struct list_head *ptype_list) 2143 { 2144 struct packet_type *ptype, *pt_prev = *pt; 2145 2146 list_for_each_entry_rcu(ptype, ptype_list, list) { 2147 if (ptype->type != type) 2148 continue; 2149 if (pt_prev) 2150 deliver_skb(skb, pt_prev, orig_dev); 2151 pt_prev = ptype; 2152 } 2153 *pt = pt_prev; 2154 } 2155 2156 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) 2157 { 2158 if (!ptype->af_packet_priv || !skb->sk) 2159 return false; 2160 2161 if (ptype->id_match) 2162 return ptype->id_match(ptype, skb->sk); 2163 else if ((struct sock *)ptype->af_packet_priv == skb->sk) 2164 return true; 2165 2166 return false; 2167 } 2168 2169 /** 2170 * dev_nit_active - return true if any network interface taps are in use 2171 * 2172 * @dev: network device to check for the presence of taps 2173 */ 2174 bool dev_nit_active(struct net_device *dev) 2175 { 2176 return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all); 2177 } 2178 EXPORT_SYMBOL_GPL(dev_nit_active); 2179 2180 /* 2181 * Support routine. Sends outgoing frames to any network 2182 * taps currently in use. 2183 */ 2184 2185 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) 2186 { 2187 struct packet_type *ptype; 2188 struct sk_buff *skb2 = NULL; 2189 struct packet_type *pt_prev = NULL; 2190 struct list_head *ptype_list = &ptype_all; 2191 2192 rcu_read_lock(); 2193 again: 2194 list_for_each_entry_rcu(ptype, ptype_list, list) { 2195 if (ptype->ignore_outgoing) 2196 continue; 2197 2198 /* Never send packets back to the socket 2199 * they originated from - MvS (miquels@drinkel.ow.org) 2200 */ 2201 if (skb_loop_sk(ptype, skb)) 2202 continue; 2203 2204 if (pt_prev) { 2205 deliver_skb(skb2, pt_prev, skb->dev); 2206 pt_prev = ptype; 2207 continue; 2208 } 2209 2210 /* need to clone skb, done only once */ 2211 skb2 = skb_clone(skb, GFP_ATOMIC); 2212 if (!skb2) 2213 goto out_unlock; 2214 2215 net_timestamp_set(skb2); 2216 2217 /* skb->nh should be correctly 2218 * set by sender, so that the second statement is 2219 * just protection against buggy protocols. 2220 */ 2221 skb_reset_mac_header(skb2); 2222 2223 if (skb_network_header(skb2) < skb2->data || 2224 skb_network_header(skb2) > skb_tail_pointer(skb2)) { 2225 net_crit_ratelimited("protocol %04x is buggy, dev %s\n", 2226 ntohs(skb2->protocol), 2227 dev->name); 2228 skb_reset_network_header(skb2); 2229 } 2230 2231 skb2->transport_header = skb2->network_header; 2232 skb2->pkt_type = PACKET_OUTGOING; 2233 pt_prev = ptype; 2234 } 2235 2236 if (ptype_list == &ptype_all) { 2237 ptype_list = &dev->ptype_all; 2238 goto again; 2239 } 2240 out_unlock: 2241 if (pt_prev) { 2242 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC)) 2243 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); 2244 else 2245 kfree_skb(skb2); 2246 } 2247 rcu_read_unlock(); 2248 } 2249 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit); 2250 2251 /** 2252 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change 2253 * @dev: Network device 2254 * @txq: number of queues available 2255 * 2256 * If real_num_tx_queues is changed the tc mappings may no longer be 2257 * valid. To resolve this verify the tc mapping remains valid and if 2258 * not NULL the mapping. With no priorities mapping to this 2259 * offset/count pair it will no longer be used. In the worst case TC0 2260 * is invalid nothing can be done so disable priority mappings. If is 2261 * expected that drivers will fix this mapping if they can before 2262 * calling netif_set_real_num_tx_queues. 2263 */ 2264 static void netif_setup_tc(struct net_device *dev, unsigned int txq) 2265 { 2266 int i; 2267 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2268 2269 /* If TC0 is invalidated disable TC mapping */ 2270 if (tc->offset + tc->count > txq) { 2271 netdev_warn(dev, "Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); 2272 dev->num_tc = 0; 2273 return; 2274 } 2275 2276 /* Invalidated prio to tc mappings set to TC0 */ 2277 for (i = 1; i < TC_BITMASK + 1; i++) { 2278 int q = netdev_get_prio_tc_map(dev, i); 2279 2280 tc = &dev->tc_to_txq[q]; 2281 if (tc->offset + tc->count > txq) { 2282 netdev_warn(dev, "Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n", 2283 i, q); 2284 netdev_set_prio_tc_map(dev, i, 0); 2285 } 2286 } 2287 } 2288 2289 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq) 2290 { 2291 if (dev->num_tc) { 2292 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2293 int i; 2294 2295 /* walk through the TCs and see if it falls into any of them */ 2296 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) { 2297 if ((txq - tc->offset) < tc->count) 2298 return i; 2299 } 2300 2301 /* didn't find it, just return -1 to indicate no match */ 2302 return -1; 2303 } 2304 2305 return 0; 2306 } 2307 EXPORT_SYMBOL(netdev_txq_to_tc); 2308 2309 #ifdef CONFIG_XPS 2310 static struct static_key xps_needed __read_mostly; 2311 static struct static_key xps_rxqs_needed __read_mostly; 2312 static DEFINE_MUTEX(xps_map_mutex); 2313 #define xmap_dereference(P) \ 2314 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) 2315 2316 static bool remove_xps_queue(struct xps_dev_maps *dev_maps, 2317 struct xps_dev_maps *old_maps, int tci, u16 index) 2318 { 2319 struct xps_map *map = NULL; 2320 int pos; 2321 2322 if (dev_maps) 2323 map = xmap_dereference(dev_maps->attr_map[tci]); 2324 if (!map) 2325 return false; 2326 2327 for (pos = map->len; pos--;) { 2328 if (map->queues[pos] != index) 2329 continue; 2330 2331 if (map->len > 1) { 2332 map->queues[pos] = map->queues[--map->len]; 2333 break; 2334 } 2335 2336 if (old_maps) 2337 RCU_INIT_POINTER(old_maps->attr_map[tci], NULL); 2338 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2339 kfree_rcu(map, rcu); 2340 return false; 2341 } 2342 2343 return true; 2344 } 2345 2346 static bool remove_xps_queue_cpu(struct net_device *dev, 2347 struct xps_dev_maps *dev_maps, 2348 int cpu, u16 offset, u16 count) 2349 { 2350 int num_tc = dev_maps->num_tc; 2351 bool active = false; 2352 int tci; 2353 2354 for (tci = cpu * num_tc; num_tc--; tci++) { 2355 int i, j; 2356 2357 for (i = count, j = offset; i--; j++) { 2358 if (!remove_xps_queue(dev_maps, NULL, tci, j)) 2359 break; 2360 } 2361 2362 active |= i < 0; 2363 } 2364 2365 return active; 2366 } 2367 2368 static void reset_xps_maps(struct net_device *dev, 2369 struct xps_dev_maps *dev_maps, 2370 enum xps_map_type type) 2371 { 2372 static_key_slow_dec_cpuslocked(&xps_needed); 2373 if (type == XPS_RXQS) 2374 static_key_slow_dec_cpuslocked(&xps_rxqs_needed); 2375 2376 RCU_INIT_POINTER(dev->xps_maps[type], NULL); 2377 2378 kfree_rcu(dev_maps, rcu); 2379 } 2380 2381 static void clean_xps_maps(struct net_device *dev, enum xps_map_type type, 2382 u16 offset, u16 count) 2383 { 2384 struct xps_dev_maps *dev_maps; 2385 bool active = false; 2386 int i, j; 2387 2388 dev_maps = xmap_dereference(dev->xps_maps[type]); 2389 if (!dev_maps) 2390 return; 2391 2392 for (j = 0; j < dev_maps->nr_ids; j++) 2393 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, count); 2394 if (!active) 2395 reset_xps_maps(dev, dev_maps, type); 2396 2397 if (type == XPS_CPUS) { 2398 for (i = offset + (count - 1); count--; i--) 2399 netdev_queue_numa_node_write( 2400 netdev_get_tx_queue(dev, i), NUMA_NO_NODE); 2401 } 2402 } 2403 2404 static void netif_reset_xps_queues(struct net_device *dev, u16 offset, 2405 u16 count) 2406 { 2407 if (!static_key_false(&xps_needed)) 2408 return; 2409 2410 cpus_read_lock(); 2411 mutex_lock(&xps_map_mutex); 2412 2413 if (static_key_false(&xps_rxqs_needed)) 2414 clean_xps_maps(dev, XPS_RXQS, offset, count); 2415 2416 clean_xps_maps(dev, XPS_CPUS, offset, count); 2417 2418 mutex_unlock(&xps_map_mutex); 2419 cpus_read_unlock(); 2420 } 2421 2422 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) 2423 { 2424 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index); 2425 } 2426 2427 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index, 2428 u16 index, bool is_rxqs_map) 2429 { 2430 struct xps_map *new_map; 2431 int alloc_len = XPS_MIN_MAP_ALLOC; 2432 int i, pos; 2433 2434 for (pos = 0; map && pos < map->len; pos++) { 2435 if (map->queues[pos] != index) 2436 continue; 2437 return map; 2438 } 2439 2440 /* Need to add tx-queue to this CPU's/rx-queue's existing map */ 2441 if (map) { 2442 if (pos < map->alloc_len) 2443 return map; 2444 2445 alloc_len = map->alloc_len * 2; 2446 } 2447 2448 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's 2449 * map 2450 */ 2451 if (is_rxqs_map) 2452 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL); 2453 else 2454 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, 2455 cpu_to_node(attr_index)); 2456 if (!new_map) 2457 return NULL; 2458 2459 for (i = 0; i < pos; i++) 2460 new_map->queues[i] = map->queues[i]; 2461 new_map->alloc_len = alloc_len; 2462 new_map->len = pos; 2463 2464 return new_map; 2465 } 2466 2467 /* Copy xps maps at a given index */ 2468 static void xps_copy_dev_maps(struct xps_dev_maps *dev_maps, 2469 struct xps_dev_maps *new_dev_maps, int index, 2470 int tc, bool skip_tc) 2471 { 2472 int i, tci = index * dev_maps->num_tc; 2473 struct xps_map *map; 2474 2475 /* copy maps belonging to foreign traffic classes */ 2476 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2477 if (i == tc && skip_tc) 2478 continue; 2479 2480 /* fill in the new device map from the old device map */ 2481 map = xmap_dereference(dev_maps->attr_map[tci]); 2482 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2483 } 2484 } 2485 2486 /* Must be called under cpus_read_lock */ 2487 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask, 2488 u16 index, enum xps_map_type type) 2489 { 2490 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL, *old_dev_maps = NULL; 2491 const unsigned long *online_mask = NULL; 2492 bool active = false, copy = false; 2493 int i, j, tci, numa_node_id = -2; 2494 int maps_sz, num_tc = 1, tc = 0; 2495 struct xps_map *map, *new_map; 2496 unsigned int nr_ids; 2497 2498 if (dev->num_tc) { 2499 /* Do not allow XPS on subordinate device directly */ 2500 num_tc = dev->num_tc; 2501 if (num_tc < 0) 2502 return -EINVAL; 2503 2504 /* If queue belongs to subordinate dev use its map */ 2505 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev; 2506 2507 tc = netdev_txq_to_tc(dev, index); 2508 if (tc < 0) 2509 return -EINVAL; 2510 } 2511 2512 mutex_lock(&xps_map_mutex); 2513 2514 dev_maps = xmap_dereference(dev->xps_maps[type]); 2515 if (type == XPS_RXQS) { 2516 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues); 2517 nr_ids = dev->num_rx_queues; 2518 } else { 2519 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc); 2520 if (num_possible_cpus() > 1) 2521 online_mask = cpumask_bits(cpu_online_mask); 2522 nr_ids = nr_cpu_ids; 2523 } 2524 2525 if (maps_sz < L1_CACHE_BYTES) 2526 maps_sz = L1_CACHE_BYTES; 2527 2528 /* The old dev_maps could be larger or smaller than the one we're 2529 * setting up now, as dev->num_tc or nr_ids could have been updated in 2530 * between. We could try to be smart, but let's be safe instead and only 2531 * copy foreign traffic classes if the two map sizes match. 2532 */ 2533 if (dev_maps && 2534 dev_maps->num_tc == num_tc && dev_maps->nr_ids == nr_ids) 2535 copy = true; 2536 2537 /* allocate memory for queue storage */ 2538 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids), 2539 j < nr_ids;) { 2540 if (!new_dev_maps) { 2541 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); 2542 if (!new_dev_maps) { 2543 mutex_unlock(&xps_map_mutex); 2544 return -ENOMEM; 2545 } 2546 2547 new_dev_maps->nr_ids = nr_ids; 2548 new_dev_maps->num_tc = num_tc; 2549 } 2550 2551 tci = j * num_tc + tc; 2552 map = copy ? xmap_dereference(dev_maps->attr_map[tci]) : NULL; 2553 2554 map = expand_xps_map(map, j, index, type == XPS_RXQS); 2555 if (!map) 2556 goto error; 2557 2558 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2559 } 2560 2561 if (!new_dev_maps) 2562 goto out_no_new_maps; 2563 2564 if (!dev_maps) { 2565 /* Increment static keys at most once per type */ 2566 static_key_slow_inc_cpuslocked(&xps_needed); 2567 if (type == XPS_RXQS) 2568 static_key_slow_inc_cpuslocked(&xps_rxqs_needed); 2569 } 2570 2571 for (j = 0; j < nr_ids; j++) { 2572 bool skip_tc = false; 2573 2574 tci = j * num_tc + tc; 2575 if (netif_attr_test_mask(j, mask, nr_ids) && 2576 netif_attr_test_online(j, online_mask, nr_ids)) { 2577 /* add tx-queue to CPU/rx-queue maps */ 2578 int pos = 0; 2579 2580 skip_tc = true; 2581 2582 map = xmap_dereference(new_dev_maps->attr_map[tci]); 2583 while ((pos < map->len) && (map->queues[pos] != index)) 2584 pos++; 2585 2586 if (pos == map->len) 2587 map->queues[map->len++] = index; 2588 #ifdef CONFIG_NUMA 2589 if (type == XPS_CPUS) { 2590 if (numa_node_id == -2) 2591 numa_node_id = cpu_to_node(j); 2592 else if (numa_node_id != cpu_to_node(j)) 2593 numa_node_id = -1; 2594 } 2595 #endif 2596 } 2597 2598 if (copy) 2599 xps_copy_dev_maps(dev_maps, new_dev_maps, j, tc, 2600 skip_tc); 2601 } 2602 2603 rcu_assign_pointer(dev->xps_maps[type], new_dev_maps); 2604 2605 /* Cleanup old maps */ 2606 if (!dev_maps) 2607 goto out_no_old_maps; 2608 2609 for (j = 0; j < dev_maps->nr_ids; j++) { 2610 for (i = num_tc, tci = j * dev_maps->num_tc; i--; tci++) { 2611 map = xmap_dereference(dev_maps->attr_map[tci]); 2612 if (!map) 2613 continue; 2614 2615 if (copy) { 2616 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2617 if (map == new_map) 2618 continue; 2619 } 2620 2621 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2622 kfree_rcu(map, rcu); 2623 } 2624 } 2625 2626 old_dev_maps = dev_maps; 2627 2628 out_no_old_maps: 2629 dev_maps = new_dev_maps; 2630 active = true; 2631 2632 out_no_new_maps: 2633 if (type == XPS_CPUS) 2634 /* update Tx queue numa node */ 2635 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), 2636 (numa_node_id >= 0) ? 2637 numa_node_id : NUMA_NO_NODE); 2638 2639 if (!dev_maps) 2640 goto out_no_maps; 2641 2642 /* removes tx-queue from unused CPUs/rx-queues */ 2643 for (j = 0; j < dev_maps->nr_ids; j++) { 2644 tci = j * dev_maps->num_tc; 2645 2646 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2647 if (i == tc && 2648 netif_attr_test_mask(j, mask, dev_maps->nr_ids) && 2649 netif_attr_test_online(j, online_mask, dev_maps->nr_ids)) 2650 continue; 2651 2652 active |= remove_xps_queue(dev_maps, 2653 copy ? old_dev_maps : NULL, 2654 tci, index); 2655 } 2656 } 2657 2658 if (old_dev_maps) 2659 kfree_rcu(old_dev_maps, rcu); 2660 2661 /* free map if not active */ 2662 if (!active) 2663 reset_xps_maps(dev, dev_maps, type); 2664 2665 out_no_maps: 2666 mutex_unlock(&xps_map_mutex); 2667 2668 return 0; 2669 error: 2670 /* remove any maps that we added */ 2671 for (j = 0; j < nr_ids; j++) { 2672 for (i = num_tc, tci = j * num_tc; i--; tci++) { 2673 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2674 map = copy ? 2675 xmap_dereference(dev_maps->attr_map[tci]) : 2676 NULL; 2677 if (new_map && new_map != map) 2678 kfree(new_map); 2679 } 2680 } 2681 2682 mutex_unlock(&xps_map_mutex); 2683 2684 kfree(new_dev_maps); 2685 return -ENOMEM; 2686 } 2687 EXPORT_SYMBOL_GPL(__netif_set_xps_queue); 2688 2689 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, 2690 u16 index) 2691 { 2692 int ret; 2693 2694 cpus_read_lock(); 2695 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, XPS_CPUS); 2696 cpus_read_unlock(); 2697 2698 return ret; 2699 } 2700 EXPORT_SYMBOL(netif_set_xps_queue); 2701 2702 #endif 2703 static void netdev_unbind_all_sb_channels(struct net_device *dev) 2704 { 2705 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2706 2707 /* Unbind any subordinate channels */ 2708 while (txq-- != &dev->_tx[0]) { 2709 if (txq->sb_dev) 2710 netdev_unbind_sb_channel(dev, txq->sb_dev); 2711 } 2712 } 2713 2714 void netdev_reset_tc(struct net_device *dev) 2715 { 2716 #ifdef CONFIG_XPS 2717 netif_reset_xps_queues_gt(dev, 0); 2718 #endif 2719 netdev_unbind_all_sb_channels(dev); 2720 2721 /* Reset TC configuration of device */ 2722 dev->num_tc = 0; 2723 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq)); 2724 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map)); 2725 } 2726 EXPORT_SYMBOL(netdev_reset_tc); 2727 2728 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset) 2729 { 2730 if (tc >= dev->num_tc) 2731 return -EINVAL; 2732 2733 #ifdef CONFIG_XPS 2734 netif_reset_xps_queues(dev, offset, count); 2735 #endif 2736 dev->tc_to_txq[tc].count = count; 2737 dev->tc_to_txq[tc].offset = offset; 2738 return 0; 2739 } 2740 EXPORT_SYMBOL(netdev_set_tc_queue); 2741 2742 int netdev_set_num_tc(struct net_device *dev, u8 num_tc) 2743 { 2744 if (num_tc > TC_MAX_QUEUE) 2745 return -EINVAL; 2746 2747 #ifdef CONFIG_XPS 2748 netif_reset_xps_queues_gt(dev, 0); 2749 #endif 2750 netdev_unbind_all_sb_channels(dev); 2751 2752 dev->num_tc = num_tc; 2753 return 0; 2754 } 2755 EXPORT_SYMBOL(netdev_set_num_tc); 2756 2757 void netdev_unbind_sb_channel(struct net_device *dev, 2758 struct net_device *sb_dev) 2759 { 2760 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2761 2762 #ifdef CONFIG_XPS 2763 netif_reset_xps_queues_gt(sb_dev, 0); 2764 #endif 2765 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq)); 2766 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map)); 2767 2768 while (txq-- != &dev->_tx[0]) { 2769 if (txq->sb_dev == sb_dev) 2770 txq->sb_dev = NULL; 2771 } 2772 } 2773 EXPORT_SYMBOL(netdev_unbind_sb_channel); 2774 2775 int netdev_bind_sb_channel_queue(struct net_device *dev, 2776 struct net_device *sb_dev, 2777 u8 tc, u16 count, u16 offset) 2778 { 2779 /* Make certain the sb_dev and dev are already configured */ 2780 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc) 2781 return -EINVAL; 2782 2783 /* We cannot hand out queues we don't have */ 2784 if ((offset + count) > dev->real_num_tx_queues) 2785 return -EINVAL; 2786 2787 /* Record the mapping */ 2788 sb_dev->tc_to_txq[tc].count = count; 2789 sb_dev->tc_to_txq[tc].offset = offset; 2790 2791 /* Provide a way for Tx queue to find the tc_to_txq map or 2792 * XPS map for itself. 2793 */ 2794 while (count--) 2795 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev; 2796 2797 return 0; 2798 } 2799 EXPORT_SYMBOL(netdev_bind_sb_channel_queue); 2800 2801 int netdev_set_sb_channel(struct net_device *dev, u16 channel) 2802 { 2803 /* Do not use a multiqueue device to represent a subordinate channel */ 2804 if (netif_is_multiqueue(dev)) 2805 return -ENODEV; 2806 2807 /* We allow channels 1 - 32767 to be used for subordinate channels. 2808 * Channel 0 is meant to be "native" mode and used only to represent 2809 * the main root device. We allow writing 0 to reset the device back 2810 * to normal mode after being used as a subordinate channel. 2811 */ 2812 if (channel > S16_MAX) 2813 return -EINVAL; 2814 2815 dev->num_tc = -channel; 2816 2817 return 0; 2818 } 2819 EXPORT_SYMBOL(netdev_set_sb_channel); 2820 2821 /* 2822 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues 2823 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed. 2824 */ 2825 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) 2826 { 2827 bool disabling; 2828 int rc; 2829 2830 disabling = txq < dev->real_num_tx_queues; 2831 2832 if (txq < 1 || txq > dev->num_tx_queues) 2833 return -EINVAL; 2834 2835 if (dev->reg_state == NETREG_REGISTERED || 2836 dev->reg_state == NETREG_UNREGISTERING) { 2837 ASSERT_RTNL(); 2838 2839 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, 2840 txq); 2841 if (rc) 2842 return rc; 2843 2844 if (dev->num_tc) 2845 netif_setup_tc(dev, txq); 2846 2847 dev_qdisc_change_real_num_tx(dev, txq); 2848 2849 dev->real_num_tx_queues = txq; 2850 2851 if (disabling) { 2852 synchronize_net(); 2853 qdisc_reset_all_tx_gt(dev, txq); 2854 #ifdef CONFIG_XPS 2855 netif_reset_xps_queues_gt(dev, txq); 2856 #endif 2857 } 2858 } else { 2859 dev->real_num_tx_queues = txq; 2860 } 2861 2862 return 0; 2863 } 2864 EXPORT_SYMBOL(netif_set_real_num_tx_queues); 2865 2866 #ifdef CONFIG_SYSFS 2867 /** 2868 * netif_set_real_num_rx_queues - set actual number of RX queues used 2869 * @dev: Network device 2870 * @rxq: Actual number of RX queues 2871 * 2872 * This must be called either with the rtnl_lock held or before 2873 * registration of the net device. Returns 0 on success, or a 2874 * negative error code. If called before registration, it always 2875 * succeeds. 2876 */ 2877 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) 2878 { 2879 int rc; 2880 2881 if (rxq < 1 || rxq > dev->num_rx_queues) 2882 return -EINVAL; 2883 2884 if (dev->reg_state == NETREG_REGISTERED) { 2885 ASSERT_RTNL(); 2886 2887 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, 2888 rxq); 2889 if (rc) 2890 return rc; 2891 } 2892 2893 dev->real_num_rx_queues = rxq; 2894 return 0; 2895 } 2896 EXPORT_SYMBOL(netif_set_real_num_rx_queues); 2897 #endif 2898 2899 /** 2900 * netif_set_real_num_queues - set actual number of RX and TX queues used 2901 * @dev: Network device 2902 * @txq: Actual number of TX queues 2903 * @rxq: Actual number of RX queues 2904 * 2905 * Set the real number of both TX and RX queues. 2906 * Does nothing if the number of queues is already correct. 2907 */ 2908 int netif_set_real_num_queues(struct net_device *dev, 2909 unsigned int txq, unsigned int rxq) 2910 { 2911 unsigned int old_rxq = dev->real_num_rx_queues; 2912 int err; 2913 2914 if (txq < 1 || txq > dev->num_tx_queues || 2915 rxq < 1 || rxq > dev->num_rx_queues) 2916 return -EINVAL; 2917 2918 /* Start from increases, so the error path only does decreases - 2919 * decreases can't fail. 2920 */ 2921 if (rxq > dev->real_num_rx_queues) { 2922 err = netif_set_real_num_rx_queues(dev, rxq); 2923 if (err) 2924 return err; 2925 } 2926 if (txq > dev->real_num_tx_queues) { 2927 err = netif_set_real_num_tx_queues(dev, txq); 2928 if (err) 2929 goto undo_rx; 2930 } 2931 if (rxq < dev->real_num_rx_queues) 2932 WARN_ON(netif_set_real_num_rx_queues(dev, rxq)); 2933 if (txq < dev->real_num_tx_queues) 2934 WARN_ON(netif_set_real_num_tx_queues(dev, txq)); 2935 2936 return 0; 2937 undo_rx: 2938 WARN_ON(netif_set_real_num_rx_queues(dev, old_rxq)); 2939 return err; 2940 } 2941 EXPORT_SYMBOL(netif_set_real_num_queues); 2942 2943 /** 2944 * netif_get_num_default_rss_queues - default number of RSS queues 2945 * 2946 * This routine should set an upper limit on the number of RSS queues 2947 * used by default by multiqueue devices. 2948 */ 2949 int netif_get_num_default_rss_queues(void) 2950 { 2951 return is_kdump_kernel() ? 2952 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus()); 2953 } 2954 EXPORT_SYMBOL(netif_get_num_default_rss_queues); 2955 2956 static void __netif_reschedule(struct Qdisc *q) 2957 { 2958 struct softnet_data *sd; 2959 unsigned long flags; 2960 2961 local_irq_save(flags); 2962 sd = this_cpu_ptr(&softnet_data); 2963 q->next_sched = NULL; 2964 *sd->output_queue_tailp = q; 2965 sd->output_queue_tailp = &q->next_sched; 2966 raise_softirq_irqoff(NET_TX_SOFTIRQ); 2967 local_irq_restore(flags); 2968 } 2969 2970 void __netif_schedule(struct Qdisc *q) 2971 { 2972 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) 2973 __netif_reschedule(q); 2974 } 2975 EXPORT_SYMBOL(__netif_schedule); 2976 2977 struct dev_kfree_skb_cb { 2978 enum skb_free_reason reason; 2979 }; 2980 2981 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) 2982 { 2983 return (struct dev_kfree_skb_cb *)skb->cb; 2984 } 2985 2986 void netif_schedule_queue(struct netdev_queue *txq) 2987 { 2988 rcu_read_lock(); 2989 if (!netif_xmit_stopped(txq)) { 2990 struct Qdisc *q = rcu_dereference(txq->qdisc); 2991 2992 __netif_schedule(q); 2993 } 2994 rcu_read_unlock(); 2995 } 2996 EXPORT_SYMBOL(netif_schedule_queue); 2997 2998 void netif_tx_wake_queue(struct netdev_queue *dev_queue) 2999 { 3000 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { 3001 struct Qdisc *q; 3002 3003 rcu_read_lock(); 3004 q = rcu_dereference(dev_queue->qdisc); 3005 __netif_schedule(q); 3006 rcu_read_unlock(); 3007 } 3008 } 3009 EXPORT_SYMBOL(netif_tx_wake_queue); 3010 3011 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason) 3012 { 3013 unsigned long flags; 3014 3015 if (unlikely(!skb)) 3016 return; 3017 3018 if (likely(refcount_read(&skb->users) == 1)) { 3019 smp_rmb(); 3020 refcount_set(&skb->users, 0); 3021 } else if (likely(!refcount_dec_and_test(&skb->users))) { 3022 return; 3023 } 3024 get_kfree_skb_cb(skb)->reason = reason; 3025 local_irq_save(flags); 3026 skb->next = __this_cpu_read(softnet_data.completion_queue); 3027 __this_cpu_write(softnet_data.completion_queue, skb); 3028 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3029 local_irq_restore(flags); 3030 } 3031 EXPORT_SYMBOL(__dev_kfree_skb_irq); 3032 3033 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason) 3034 { 3035 if (in_hardirq() || irqs_disabled()) 3036 __dev_kfree_skb_irq(skb, reason); 3037 else 3038 dev_kfree_skb(skb); 3039 } 3040 EXPORT_SYMBOL(__dev_kfree_skb_any); 3041 3042 3043 /** 3044 * netif_device_detach - mark device as removed 3045 * @dev: network device 3046 * 3047 * Mark device as removed from system and therefore no longer available. 3048 */ 3049 void netif_device_detach(struct net_device *dev) 3050 { 3051 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && 3052 netif_running(dev)) { 3053 netif_tx_stop_all_queues(dev); 3054 } 3055 } 3056 EXPORT_SYMBOL(netif_device_detach); 3057 3058 /** 3059 * netif_device_attach - mark device as attached 3060 * @dev: network device 3061 * 3062 * Mark device as attached from system and restart if needed. 3063 */ 3064 void netif_device_attach(struct net_device *dev) 3065 { 3066 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && 3067 netif_running(dev)) { 3068 netif_tx_wake_all_queues(dev); 3069 __netdev_watchdog_up(dev); 3070 } 3071 } 3072 EXPORT_SYMBOL(netif_device_attach); 3073 3074 /* 3075 * Returns a Tx hash based on the given packet descriptor a Tx queues' number 3076 * to be used as a distribution range. 3077 */ 3078 static u16 skb_tx_hash(const struct net_device *dev, 3079 const struct net_device *sb_dev, 3080 struct sk_buff *skb) 3081 { 3082 u32 hash; 3083 u16 qoffset = 0; 3084 u16 qcount = dev->real_num_tx_queues; 3085 3086 if (dev->num_tc) { 3087 u8 tc = netdev_get_prio_tc_map(dev, skb->priority); 3088 3089 qoffset = sb_dev->tc_to_txq[tc].offset; 3090 qcount = sb_dev->tc_to_txq[tc].count; 3091 if (unlikely(!qcount)) { 3092 net_warn_ratelimited("%s: invalid qcount, qoffset %u for tc %u\n", 3093 sb_dev->name, qoffset, tc); 3094 qoffset = 0; 3095 qcount = dev->real_num_tx_queues; 3096 } 3097 } 3098 3099 if (skb_rx_queue_recorded(skb)) { 3100 hash = skb_get_rx_queue(skb); 3101 if (hash >= qoffset) 3102 hash -= qoffset; 3103 while (unlikely(hash >= qcount)) 3104 hash -= qcount; 3105 return hash + qoffset; 3106 } 3107 3108 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; 3109 } 3110 3111 static void skb_warn_bad_offload(const struct sk_buff *skb) 3112 { 3113 static const netdev_features_t null_features; 3114 struct net_device *dev = skb->dev; 3115 const char *name = ""; 3116 3117 if (!net_ratelimit()) 3118 return; 3119 3120 if (dev) { 3121 if (dev->dev.parent) 3122 name = dev_driver_string(dev->dev.parent); 3123 else 3124 name = netdev_name(dev); 3125 } 3126 skb_dump(KERN_WARNING, skb, false); 3127 WARN(1, "%s: caps=(%pNF, %pNF)\n", 3128 name, dev ? &dev->features : &null_features, 3129 skb->sk ? &skb->sk->sk_route_caps : &null_features); 3130 } 3131 3132 /* 3133 * Invalidate hardware checksum when packet is to be mangled, and 3134 * complete checksum manually on outgoing path. 3135 */ 3136 int skb_checksum_help(struct sk_buff *skb) 3137 { 3138 __wsum csum; 3139 int ret = 0, offset; 3140 3141 if (skb->ip_summed == CHECKSUM_COMPLETE) 3142 goto out_set_summed; 3143 3144 if (unlikely(skb_is_gso(skb))) { 3145 skb_warn_bad_offload(skb); 3146 return -EINVAL; 3147 } 3148 3149 /* Before computing a checksum, we should make sure no frag could 3150 * be modified by an external entity : checksum could be wrong. 3151 */ 3152 if (skb_has_shared_frag(skb)) { 3153 ret = __skb_linearize(skb); 3154 if (ret) 3155 goto out; 3156 } 3157 3158 offset = skb_checksum_start_offset(skb); 3159 BUG_ON(offset >= skb_headlen(skb)); 3160 csum = skb_checksum(skb, offset, skb->len - offset, 0); 3161 3162 offset += skb->csum_offset; 3163 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); 3164 3165 ret = skb_ensure_writable(skb, offset + sizeof(__sum16)); 3166 if (ret) 3167 goto out; 3168 3169 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0; 3170 out_set_summed: 3171 skb->ip_summed = CHECKSUM_NONE; 3172 out: 3173 return ret; 3174 } 3175 EXPORT_SYMBOL(skb_checksum_help); 3176 3177 int skb_crc32c_csum_help(struct sk_buff *skb) 3178 { 3179 __le32 crc32c_csum; 3180 int ret = 0, offset, start; 3181 3182 if (skb->ip_summed != CHECKSUM_PARTIAL) 3183 goto out; 3184 3185 if (unlikely(skb_is_gso(skb))) 3186 goto out; 3187 3188 /* Before computing a checksum, we should make sure no frag could 3189 * be modified by an external entity : checksum could be wrong. 3190 */ 3191 if (unlikely(skb_has_shared_frag(skb))) { 3192 ret = __skb_linearize(skb); 3193 if (ret) 3194 goto out; 3195 } 3196 start = skb_checksum_start_offset(skb); 3197 offset = start + offsetof(struct sctphdr, checksum); 3198 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { 3199 ret = -EINVAL; 3200 goto out; 3201 } 3202 3203 ret = skb_ensure_writable(skb, offset + sizeof(__le32)); 3204 if (ret) 3205 goto out; 3206 3207 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start, 3208 skb->len - start, ~(__u32)0, 3209 crc32c_csum_stub)); 3210 *(__le32 *)(skb->data + offset) = crc32c_csum; 3211 skb->ip_summed = CHECKSUM_NONE; 3212 skb->csum_not_inet = 0; 3213 out: 3214 return ret; 3215 } 3216 3217 __be16 skb_network_protocol(struct sk_buff *skb, int *depth) 3218 { 3219 __be16 type = skb->protocol; 3220 3221 /* Tunnel gso handlers can set protocol to ethernet. */ 3222 if (type == htons(ETH_P_TEB)) { 3223 struct ethhdr *eth; 3224 3225 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) 3226 return 0; 3227 3228 eth = (struct ethhdr *)skb->data; 3229 type = eth->h_proto; 3230 } 3231 3232 return __vlan_get_protocol(skb, type, depth); 3233 } 3234 3235 /* openvswitch calls this on rx path, so we need a different check. 3236 */ 3237 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path) 3238 { 3239 if (tx_path) 3240 return skb->ip_summed != CHECKSUM_PARTIAL && 3241 skb->ip_summed != CHECKSUM_UNNECESSARY; 3242 3243 return skb->ip_summed == CHECKSUM_NONE; 3244 } 3245 3246 /** 3247 * __skb_gso_segment - Perform segmentation on skb. 3248 * @skb: buffer to segment 3249 * @features: features for the output path (see dev->features) 3250 * @tx_path: whether it is called in TX path 3251 * 3252 * This function segments the given skb and returns a list of segments. 3253 * 3254 * It may return NULL if the skb requires no segmentation. This is 3255 * only possible when GSO is used for verifying header integrity. 3256 * 3257 * Segmentation preserves SKB_GSO_CB_OFFSET bytes of previous skb cb. 3258 */ 3259 struct sk_buff *__skb_gso_segment(struct sk_buff *skb, 3260 netdev_features_t features, bool tx_path) 3261 { 3262 struct sk_buff *segs; 3263 3264 if (unlikely(skb_needs_check(skb, tx_path))) { 3265 int err; 3266 3267 /* We're going to init ->check field in TCP or UDP header */ 3268 err = skb_cow_head(skb, 0); 3269 if (err < 0) 3270 return ERR_PTR(err); 3271 } 3272 3273 /* Only report GSO partial support if it will enable us to 3274 * support segmentation on this frame without needing additional 3275 * work. 3276 */ 3277 if (features & NETIF_F_GSO_PARTIAL) { 3278 netdev_features_t partial_features = NETIF_F_GSO_ROBUST; 3279 struct net_device *dev = skb->dev; 3280 3281 partial_features |= dev->features & dev->gso_partial_features; 3282 if (!skb_gso_ok(skb, features | partial_features)) 3283 features &= ~NETIF_F_GSO_PARTIAL; 3284 } 3285 3286 BUILD_BUG_ON(SKB_GSO_CB_OFFSET + 3287 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb)); 3288 3289 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb); 3290 SKB_GSO_CB(skb)->encap_level = 0; 3291 3292 skb_reset_mac_header(skb); 3293 skb_reset_mac_len(skb); 3294 3295 segs = skb_mac_gso_segment(skb, features); 3296 3297 if (segs != skb && unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs))) 3298 skb_warn_bad_offload(skb); 3299 3300 return segs; 3301 } 3302 EXPORT_SYMBOL(__skb_gso_segment); 3303 3304 /* Take action when hardware reception checksum errors are detected. */ 3305 #ifdef CONFIG_BUG 3306 static void do_netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3307 { 3308 netdev_err(dev, "hw csum failure\n"); 3309 skb_dump(KERN_ERR, skb, true); 3310 dump_stack(); 3311 } 3312 3313 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3314 { 3315 DO_ONCE_LITE(do_netdev_rx_csum_fault, dev, skb); 3316 } 3317 EXPORT_SYMBOL(netdev_rx_csum_fault); 3318 #endif 3319 3320 /* XXX: check that highmem exists at all on the given machine. */ 3321 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) 3322 { 3323 #ifdef CONFIG_HIGHMEM 3324 int i; 3325 3326 if (!(dev->features & NETIF_F_HIGHDMA)) { 3327 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 3328 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 3329 3330 if (PageHighMem(skb_frag_page(frag))) 3331 return 1; 3332 } 3333 } 3334 #endif 3335 return 0; 3336 } 3337 3338 /* If MPLS offload request, verify we are testing hardware MPLS features 3339 * instead of standard features for the netdev. 3340 */ 3341 #if IS_ENABLED(CONFIG_NET_MPLS_GSO) 3342 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3343 netdev_features_t features, 3344 __be16 type) 3345 { 3346 if (eth_p_mpls(type)) 3347 features &= skb->dev->mpls_features; 3348 3349 return features; 3350 } 3351 #else 3352 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3353 netdev_features_t features, 3354 __be16 type) 3355 { 3356 return features; 3357 } 3358 #endif 3359 3360 static netdev_features_t harmonize_features(struct sk_buff *skb, 3361 netdev_features_t features) 3362 { 3363 __be16 type; 3364 3365 type = skb_network_protocol(skb, NULL); 3366 features = net_mpls_features(skb, features, type); 3367 3368 if (skb->ip_summed != CHECKSUM_NONE && 3369 !can_checksum_protocol(features, type)) { 3370 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); 3371 } 3372 if (illegal_highdma(skb->dev, skb)) 3373 features &= ~NETIF_F_SG; 3374 3375 return features; 3376 } 3377 3378 netdev_features_t passthru_features_check(struct sk_buff *skb, 3379 struct net_device *dev, 3380 netdev_features_t features) 3381 { 3382 return features; 3383 } 3384 EXPORT_SYMBOL(passthru_features_check); 3385 3386 static netdev_features_t dflt_features_check(struct sk_buff *skb, 3387 struct net_device *dev, 3388 netdev_features_t features) 3389 { 3390 return vlan_features_check(skb, features); 3391 } 3392 3393 static netdev_features_t gso_features_check(const struct sk_buff *skb, 3394 struct net_device *dev, 3395 netdev_features_t features) 3396 { 3397 u16 gso_segs = skb_shinfo(skb)->gso_segs; 3398 3399 if (gso_segs > READ_ONCE(dev->gso_max_segs)) 3400 return features & ~NETIF_F_GSO_MASK; 3401 3402 if (!skb_shinfo(skb)->gso_type) { 3403 skb_warn_bad_offload(skb); 3404 return features & ~NETIF_F_GSO_MASK; 3405 } 3406 3407 /* Support for GSO partial features requires software 3408 * intervention before we can actually process the packets 3409 * so we need to strip support for any partial features now 3410 * and we can pull them back in after we have partially 3411 * segmented the frame. 3412 */ 3413 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)) 3414 features &= ~dev->gso_partial_features; 3415 3416 /* Make sure to clear the IPv4 ID mangling feature if the 3417 * IPv4 header has the potential to be fragmented. 3418 */ 3419 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { 3420 struct iphdr *iph = skb->encapsulation ? 3421 inner_ip_hdr(skb) : ip_hdr(skb); 3422 3423 if (!(iph->frag_off & htons(IP_DF))) 3424 features &= ~NETIF_F_TSO_MANGLEID; 3425 } 3426 3427 return features; 3428 } 3429 3430 netdev_features_t netif_skb_features(struct sk_buff *skb) 3431 { 3432 struct net_device *dev = skb->dev; 3433 netdev_features_t features = dev->features; 3434 3435 if (skb_is_gso(skb)) 3436 features = gso_features_check(skb, dev, features); 3437 3438 /* If encapsulation offload request, verify we are testing 3439 * hardware encapsulation features instead of standard 3440 * features for the netdev 3441 */ 3442 if (skb->encapsulation) 3443 features &= dev->hw_enc_features; 3444 3445 if (skb_vlan_tagged(skb)) 3446 features = netdev_intersect_features(features, 3447 dev->vlan_features | 3448 NETIF_F_HW_VLAN_CTAG_TX | 3449 NETIF_F_HW_VLAN_STAG_TX); 3450 3451 if (dev->netdev_ops->ndo_features_check) 3452 features &= dev->netdev_ops->ndo_features_check(skb, dev, 3453 features); 3454 else 3455 features &= dflt_features_check(skb, dev, features); 3456 3457 return harmonize_features(skb, features); 3458 } 3459 EXPORT_SYMBOL(netif_skb_features); 3460 3461 static int xmit_one(struct sk_buff *skb, struct net_device *dev, 3462 struct netdev_queue *txq, bool more) 3463 { 3464 unsigned int len; 3465 int rc; 3466 3467 if (dev_nit_active(dev)) 3468 dev_queue_xmit_nit(skb, dev); 3469 3470 len = skb->len; 3471 PRANDOM_ADD_NOISE(skb, dev, txq, len + jiffies); 3472 trace_net_dev_start_xmit(skb, dev); 3473 rc = netdev_start_xmit(skb, dev, txq, more); 3474 trace_net_dev_xmit(skb, rc, dev, len); 3475 3476 return rc; 3477 } 3478 3479 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, 3480 struct netdev_queue *txq, int *ret) 3481 { 3482 struct sk_buff *skb = first; 3483 int rc = NETDEV_TX_OK; 3484 3485 while (skb) { 3486 struct sk_buff *next = skb->next; 3487 3488 skb_mark_not_on_list(skb); 3489 rc = xmit_one(skb, dev, txq, next != NULL); 3490 if (unlikely(!dev_xmit_complete(rc))) { 3491 skb->next = next; 3492 goto out; 3493 } 3494 3495 skb = next; 3496 if (netif_tx_queue_stopped(txq) && skb) { 3497 rc = NETDEV_TX_BUSY; 3498 break; 3499 } 3500 } 3501 3502 out: 3503 *ret = rc; 3504 return skb; 3505 } 3506 3507 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, 3508 netdev_features_t features) 3509 { 3510 if (skb_vlan_tag_present(skb) && 3511 !vlan_hw_offload_capable(features, skb->vlan_proto)) 3512 skb = __vlan_hwaccel_push_inside(skb); 3513 return skb; 3514 } 3515 3516 int skb_csum_hwoffload_help(struct sk_buff *skb, 3517 const netdev_features_t features) 3518 { 3519 if (unlikely(skb_csum_is_sctp(skb))) 3520 return !!(features & NETIF_F_SCTP_CRC) ? 0 : 3521 skb_crc32c_csum_help(skb); 3522 3523 if (features & NETIF_F_HW_CSUM) 3524 return 0; 3525 3526 if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) { 3527 switch (skb->csum_offset) { 3528 case offsetof(struct tcphdr, check): 3529 case offsetof(struct udphdr, check): 3530 return 0; 3531 } 3532 } 3533 3534 return skb_checksum_help(skb); 3535 } 3536 EXPORT_SYMBOL(skb_csum_hwoffload_help); 3537 3538 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again) 3539 { 3540 netdev_features_t features; 3541 3542 features = netif_skb_features(skb); 3543 skb = validate_xmit_vlan(skb, features); 3544 if (unlikely(!skb)) 3545 goto out_null; 3546 3547 skb = sk_validate_xmit_skb(skb, dev); 3548 if (unlikely(!skb)) 3549 goto out_null; 3550 3551 if (netif_needs_gso(skb, features)) { 3552 struct sk_buff *segs; 3553 3554 segs = skb_gso_segment(skb, features); 3555 if (IS_ERR(segs)) { 3556 goto out_kfree_skb; 3557 } else if (segs) { 3558 consume_skb(skb); 3559 skb = segs; 3560 } 3561 } else { 3562 if (skb_needs_linearize(skb, features) && 3563 __skb_linearize(skb)) 3564 goto out_kfree_skb; 3565 3566 /* If packet is not checksummed and device does not 3567 * support checksumming for this protocol, complete 3568 * checksumming here. 3569 */ 3570 if (skb->ip_summed == CHECKSUM_PARTIAL) { 3571 if (skb->encapsulation) 3572 skb_set_inner_transport_header(skb, 3573 skb_checksum_start_offset(skb)); 3574 else 3575 skb_set_transport_header(skb, 3576 skb_checksum_start_offset(skb)); 3577 if (skb_csum_hwoffload_help(skb, features)) 3578 goto out_kfree_skb; 3579 } 3580 } 3581 3582 skb = validate_xmit_xfrm(skb, features, again); 3583 3584 return skb; 3585 3586 out_kfree_skb: 3587 kfree_skb(skb); 3588 out_null: 3589 atomic_long_inc(&dev->tx_dropped); 3590 return NULL; 3591 } 3592 3593 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again) 3594 { 3595 struct sk_buff *next, *head = NULL, *tail; 3596 3597 for (; skb != NULL; skb = next) { 3598 next = skb->next; 3599 skb_mark_not_on_list(skb); 3600 3601 /* in case skb wont be segmented, point to itself */ 3602 skb->prev = skb; 3603 3604 skb = validate_xmit_skb(skb, dev, again); 3605 if (!skb) 3606 continue; 3607 3608 if (!head) 3609 head = skb; 3610 else 3611 tail->next = skb; 3612 /* If skb was segmented, skb->prev points to 3613 * the last segment. If not, it still contains skb. 3614 */ 3615 tail = skb->prev; 3616 } 3617 return head; 3618 } 3619 EXPORT_SYMBOL_GPL(validate_xmit_skb_list); 3620 3621 static void qdisc_pkt_len_init(struct sk_buff *skb) 3622 { 3623 const struct skb_shared_info *shinfo = skb_shinfo(skb); 3624 3625 qdisc_skb_cb(skb)->pkt_len = skb->len; 3626 3627 /* To get more precise estimation of bytes sent on wire, 3628 * we add to pkt_len the headers size of all segments 3629 */ 3630 if (shinfo->gso_size && skb_transport_header_was_set(skb)) { 3631 unsigned int hdr_len; 3632 u16 gso_segs = shinfo->gso_segs; 3633 3634 /* mac layer + network layer */ 3635 hdr_len = skb_transport_header(skb) - skb_mac_header(skb); 3636 3637 /* + transport layer */ 3638 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { 3639 const struct tcphdr *th; 3640 struct tcphdr _tcphdr; 3641 3642 th = skb_header_pointer(skb, skb_transport_offset(skb), 3643 sizeof(_tcphdr), &_tcphdr); 3644 if (likely(th)) 3645 hdr_len += __tcp_hdrlen(th); 3646 } else { 3647 struct udphdr _udphdr; 3648 3649 if (skb_header_pointer(skb, skb_transport_offset(skb), 3650 sizeof(_udphdr), &_udphdr)) 3651 hdr_len += sizeof(struct udphdr); 3652 } 3653 3654 if (shinfo->gso_type & SKB_GSO_DODGY) 3655 gso_segs = DIV_ROUND_UP(skb->len - hdr_len, 3656 shinfo->gso_size); 3657 3658 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; 3659 } 3660 } 3661 3662 static int dev_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *q, 3663 struct sk_buff **to_free, 3664 struct netdev_queue *txq) 3665 { 3666 int rc; 3667 3668 rc = q->enqueue(skb, q, to_free) & NET_XMIT_MASK; 3669 if (rc == NET_XMIT_SUCCESS) 3670 trace_qdisc_enqueue(q, txq, skb); 3671 return rc; 3672 } 3673 3674 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, 3675 struct net_device *dev, 3676 struct netdev_queue *txq) 3677 { 3678 spinlock_t *root_lock = qdisc_lock(q); 3679 struct sk_buff *to_free = NULL; 3680 bool contended; 3681 int rc; 3682 3683 qdisc_calculate_pkt_len(skb, q); 3684 3685 if (q->flags & TCQ_F_NOLOCK) { 3686 if (q->flags & TCQ_F_CAN_BYPASS && nolock_qdisc_is_empty(q) && 3687 qdisc_run_begin(q)) { 3688 /* Retest nolock_qdisc_is_empty() within the protection 3689 * of q->seqlock to protect from racing with requeuing. 3690 */ 3691 if (unlikely(!nolock_qdisc_is_empty(q))) { 3692 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3693 __qdisc_run(q); 3694 qdisc_run_end(q); 3695 3696 goto no_lock_out; 3697 } 3698 3699 qdisc_bstats_cpu_update(q, skb); 3700 if (sch_direct_xmit(skb, q, dev, txq, NULL, true) && 3701 !nolock_qdisc_is_empty(q)) 3702 __qdisc_run(q); 3703 3704 qdisc_run_end(q); 3705 return NET_XMIT_SUCCESS; 3706 } 3707 3708 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3709 qdisc_run(q); 3710 3711 no_lock_out: 3712 if (unlikely(to_free)) 3713 kfree_skb_list(to_free); 3714 return rc; 3715 } 3716 3717 /* 3718 * Heuristic to force contended enqueues to serialize on a 3719 * separate lock before trying to get qdisc main lock. 3720 * This permits qdisc->running owner to get the lock more 3721 * often and dequeue packets faster. 3722 * On PREEMPT_RT it is possible to preempt the qdisc owner during xmit 3723 * and then other tasks will only enqueue packets. The packets will be 3724 * sent after the qdisc owner is scheduled again. To prevent this 3725 * scenario the task always serialize on the lock. 3726 */ 3727 contended = qdisc_is_running(q) || IS_ENABLED(CONFIG_PREEMPT_RT); 3728 if (unlikely(contended)) 3729 spin_lock(&q->busylock); 3730 3731 spin_lock(root_lock); 3732 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { 3733 __qdisc_drop(skb, &to_free); 3734 rc = NET_XMIT_DROP; 3735 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && 3736 qdisc_run_begin(q)) { 3737 /* 3738 * This is a work-conserving queue; there are no old skbs 3739 * waiting to be sent out; and the qdisc is not running - 3740 * xmit the skb directly. 3741 */ 3742 3743 qdisc_bstats_update(q, skb); 3744 3745 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { 3746 if (unlikely(contended)) { 3747 spin_unlock(&q->busylock); 3748 contended = false; 3749 } 3750 __qdisc_run(q); 3751 } 3752 3753 qdisc_run_end(q); 3754 rc = NET_XMIT_SUCCESS; 3755 } else { 3756 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3757 if (qdisc_run_begin(q)) { 3758 if (unlikely(contended)) { 3759 spin_unlock(&q->busylock); 3760 contended = false; 3761 } 3762 __qdisc_run(q); 3763 qdisc_run_end(q); 3764 } 3765 } 3766 spin_unlock(root_lock); 3767 if (unlikely(to_free)) 3768 kfree_skb_list(to_free); 3769 if (unlikely(contended)) 3770 spin_unlock(&q->busylock); 3771 return rc; 3772 } 3773 3774 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) 3775 static void skb_update_prio(struct sk_buff *skb) 3776 { 3777 const struct netprio_map *map; 3778 const struct sock *sk; 3779 unsigned int prioidx; 3780 3781 if (skb->priority) 3782 return; 3783 map = rcu_dereference_bh(skb->dev->priomap); 3784 if (!map) 3785 return; 3786 sk = skb_to_full_sk(skb); 3787 if (!sk) 3788 return; 3789 3790 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data); 3791 3792 if (prioidx < map->priomap_len) 3793 skb->priority = map->priomap[prioidx]; 3794 } 3795 #else 3796 #define skb_update_prio(skb) 3797 #endif 3798 3799 /** 3800 * dev_loopback_xmit - loop back @skb 3801 * @net: network namespace this loopback is happening in 3802 * @sk: sk needed to be a netfilter okfn 3803 * @skb: buffer to transmit 3804 */ 3805 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) 3806 { 3807 skb_reset_mac_header(skb); 3808 __skb_pull(skb, skb_network_offset(skb)); 3809 skb->pkt_type = PACKET_LOOPBACK; 3810 if (skb->ip_summed == CHECKSUM_NONE) 3811 skb->ip_summed = CHECKSUM_UNNECESSARY; 3812 WARN_ON(!skb_dst(skb)); 3813 skb_dst_force(skb); 3814 netif_rx_ni(skb); 3815 return 0; 3816 } 3817 EXPORT_SYMBOL(dev_loopback_xmit); 3818 3819 #ifdef CONFIG_NET_EGRESS 3820 static struct sk_buff * 3821 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 3822 { 3823 #ifdef CONFIG_NET_CLS_ACT 3824 struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress); 3825 struct tcf_result cl_res; 3826 3827 if (!miniq) 3828 return skb; 3829 3830 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */ 3831 tc_skb_cb(skb)->mru = 0; 3832 tc_skb_cb(skb)->post_ct = false; 3833 mini_qdisc_bstats_cpu_update(miniq, skb); 3834 3835 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) { 3836 case TC_ACT_OK: 3837 case TC_ACT_RECLASSIFY: 3838 skb->tc_index = TC_H_MIN(cl_res.classid); 3839 break; 3840 case TC_ACT_SHOT: 3841 mini_qdisc_qstats_cpu_drop(miniq); 3842 *ret = NET_XMIT_DROP; 3843 kfree_skb(skb); 3844 return NULL; 3845 case TC_ACT_STOLEN: 3846 case TC_ACT_QUEUED: 3847 case TC_ACT_TRAP: 3848 *ret = NET_XMIT_SUCCESS; 3849 consume_skb(skb); 3850 return NULL; 3851 case TC_ACT_REDIRECT: 3852 /* No need to push/pop skb's mac_header here on egress! */ 3853 skb_do_redirect(skb); 3854 *ret = NET_XMIT_SUCCESS; 3855 return NULL; 3856 default: 3857 break; 3858 } 3859 #endif /* CONFIG_NET_CLS_ACT */ 3860 3861 return skb; 3862 } 3863 #endif /* CONFIG_NET_EGRESS */ 3864 3865 #ifdef CONFIG_XPS 3866 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb, 3867 struct xps_dev_maps *dev_maps, unsigned int tci) 3868 { 3869 int tc = netdev_get_prio_tc_map(dev, skb->priority); 3870 struct xps_map *map; 3871 int queue_index = -1; 3872 3873 if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids) 3874 return queue_index; 3875 3876 tci *= dev_maps->num_tc; 3877 tci += tc; 3878 3879 map = rcu_dereference(dev_maps->attr_map[tci]); 3880 if (map) { 3881 if (map->len == 1) 3882 queue_index = map->queues[0]; 3883 else 3884 queue_index = map->queues[reciprocal_scale( 3885 skb_get_hash(skb), map->len)]; 3886 if (unlikely(queue_index >= dev->real_num_tx_queues)) 3887 queue_index = -1; 3888 } 3889 return queue_index; 3890 } 3891 #endif 3892 3893 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev, 3894 struct sk_buff *skb) 3895 { 3896 #ifdef CONFIG_XPS 3897 struct xps_dev_maps *dev_maps; 3898 struct sock *sk = skb->sk; 3899 int queue_index = -1; 3900 3901 if (!static_key_false(&xps_needed)) 3902 return -1; 3903 3904 rcu_read_lock(); 3905 if (!static_key_false(&xps_rxqs_needed)) 3906 goto get_cpus_map; 3907 3908 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]); 3909 if (dev_maps) { 3910 int tci = sk_rx_queue_get(sk); 3911 3912 if (tci >= 0) 3913 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 3914 tci); 3915 } 3916 3917 get_cpus_map: 3918 if (queue_index < 0) { 3919 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]); 3920 if (dev_maps) { 3921 unsigned int tci = skb->sender_cpu - 1; 3922 3923 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 3924 tci); 3925 } 3926 } 3927 rcu_read_unlock(); 3928 3929 return queue_index; 3930 #else 3931 return -1; 3932 #endif 3933 } 3934 3935 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb, 3936 struct net_device *sb_dev) 3937 { 3938 return 0; 3939 } 3940 EXPORT_SYMBOL(dev_pick_tx_zero); 3941 3942 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb, 3943 struct net_device *sb_dev) 3944 { 3945 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues; 3946 } 3947 EXPORT_SYMBOL(dev_pick_tx_cpu_id); 3948 3949 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb, 3950 struct net_device *sb_dev) 3951 { 3952 struct sock *sk = skb->sk; 3953 int queue_index = sk_tx_queue_get(sk); 3954 3955 sb_dev = sb_dev ? : dev; 3956 3957 if (queue_index < 0 || skb->ooo_okay || 3958 queue_index >= dev->real_num_tx_queues) { 3959 int new_index = get_xps_queue(dev, sb_dev, skb); 3960 3961 if (new_index < 0) 3962 new_index = skb_tx_hash(dev, sb_dev, skb); 3963 3964 if (queue_index != new_index && sk && 3965 sk_fullsock(sk) && 3966 rcu_access_pointer(sk->sk_dst_cache)) 3967 sk_tx_queue_set(sk, new_index); 3968 3969 queue_index = new_index; 3970 } 3971 3972 return queue_index; 3973 } 3974 EXPORT_SYMBOL(netdev_pick_tx); 3975 3976 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev, 3977 struct sk_buff *skb, 3978 struct net_device *sb_dev) 3979 { 3980 int queue_index = 0; 3981 3982 #ifdef CONFIG_XPS 3983 u32 sender_cpu = skb->sender_cpu - 1; 3984 3985 if (sender_cpu >= (u32)NR_CPUS) 3986 skb->sender_cpu = raw_smp_processor_id() + 1; 3987 #endif 3988 3989 if (dev->real_num_tx_queues != 1) { 3990 const struct net_device_ops *ops = dev->netdev_ops; 3991 3992 if (ops->ndo_select_queue) 3993 queue_index = ops->ndo_select_queue(dev, skb, sb_dev); 3994 else 3995 queue_index = netdev_pick_tx(dev, skb, sb_dev); 3996 3997 queue_index = netdev_cap_txqueue(dev, queue_index); 3998 } 3999 4000 skb_set_queue_mapping(skb, queue_index); 4001 return netdev_get_tx_queue(dev, queue_index); 4002 } 4003 4004 /** 4005 * __dev_queue_xmit - transmit a buffer 4006 * @skb: buffer to transmit 4007 * @sb_dev: suboordinate device used for L2 forwarding offload 4008 * 4009 * Queue a buffer for transmission to a network device. The caller must 4010 * have set the device and priority and built the buffer before calling 4011 * this function. The function can be called from an interrupt. 4012 * 4013 * A negative errno code is returned on a failure. A success does not 4014 * guarantee the frame will be transmitted as it may be dropped due 4015 * to congestion or traffic shaping. 4016 * 4017 * ----------------------------------------------------------------------------------- 4018 * I notice this method can also return errors from the queue disciplines, 4019 * including NET_XMIT_DROP, which is a positive value. So, errors can also 4020 * be positive. 4021 * 4022 * Regardless of the return value, the skb is consumed, so it is currently 4023 * difficult to retry a send to this method. (You can bump the ref count 4024 * before sending to hold a reference for retry if you are careful.) 4025 * 4026 * When calling this method, interrupts MUST be enabled. This is because 4027 * the BH enable code must have IRQs enabled so that it will not deadlock. 4028 * --BLG 4029 */ 4030 static int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev) 4031 { 4032 struct net_device *dev = skb->dev; 4033 struct netdev_queue *txq; 4034 struct Qdisc *q; 4035 int rc = -ENOMEM; 4036 bool again = false; 4037 4038 skb_reset_mac_header(skb); 4039 4040 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) 4041 __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED); 4042 4043 /* Disable soft irqs for various locks below. Also 4044 * stops preemption for RCU. 4045 */ 4046 rcu_read_lock_bh(); 4047 4048 skb_update_prio(skb); 4049 4050 qdisc_pkt_len_init(skb); 4051 #ifdef CONFIG_NET_CLS_ACT 4052 skb->tc_at_ingress = 0; 4053 #endif 4054 #ifdef CONFIG_NET_EGRESS 4055 if (static_branch_unlikely(&egress_needed_key)) { 4056 if (nf_hook_egress_active()) { 4057 skb = nf_hook_egress(skb, &rc, dev); 4058 if (!skb) 4059 goto out; 4060 } 4061 nf_skip_egress(skb, true); 4062 skb = sch_handle_egress(skb, &rc, dev); 4063 if (!skb) 4064 goto out; 4065 nf_skip_egress(skb, false); 4066 } 4067 #endif 4068 /* If device/qdisc don't need skb->dst, release it right now while 4069 * its hot in this cpu cache. 4070 */ 4071 if (dev->priv_flags & IFF_XMIT_DST_RELEASE) 4072 skb_dst_drop(skb); 4073 else 4074 skb_dst_force(skb); 4075 4076 txq = netdev_core_pick_tx(dev, skb, sb_dev); 4077 q = rcu_dereference_bh(txq->qdisc); 4078 4079 trace_net_dev_queue(skb); 4080 if (q->enqueue) { 4081 rc = __dev_xmit_skb(skb, q, dev, txq); 4082 goto out; 4083 } 4084 4085 /* The device has no queue. Common case for software devices: 4086 * loopback, all the sorts of tunnels... 4087 4088 * Really, it is unlikely that netif_tx_lock protection is necessary 4089 * here. (f.e. loopback and IP tunnels are clean ignoring statistics 4090 * counters.) 4091 * However, it is possible, that they rely on protection 4092 * made by us here. 4093 4094 * Check this and shot the lock. It is not prone from deadlocks. 4095 *Either shot noqueue qdisc, it is even simpler 8) 4096 */ 4097 if (dev->flags & IFF_UP) { 4098 int cpu = smp_processor_id(); /* ok because BHs are off */ 4099 4100 /* Other cpus might concurrently change txq->xmit_lock_owner 4101 * to -1 or to their cpu id, but not to our id. 4102 */ 4103 if (READ_ONCE(txq->xmit_lock_owner) != cpu) { 4104 if (dev_xmit_recursion()) 4105 goto recursion_alert; 4106 4107 skb = validate_xmit_skb(skb, dev, &again); 4108 if (!skb) 4109 goto out; 4110 4111 PRANDOM_ADD_NOISE(skb, dev, txq, jiffies); 4112 HARD_TX_LOCK(dev, txq, cpu); 4113 4114 if (!netif_xmit_stopped(txq)) { 4115 dev_xmit_recursion_inc(); 4116 skb = dev_hard_start_xmit(skb, dev, txq, &rc); 4117 dev_xmit_recursion_dec(); 4118 if (dev_xmit_complete(rc)) { 4119 HARD_TX_UNLOCK(dev, txq); 4120 goto out; 4121 } 4122 } 4123 HARD_TX_UNLOCK(dev, txq); 4124 net_crit_ratelimited("Virtual device %s asks to queue packet!\n", 4125 dev->name); 4126 } else { 4127 /* Recursion is detected! It is possible, 4128 * unfortunately 4129 */ 4130 recursion_alert: 4131 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", 4132 dev->name); 4133 } 4134 } 4135 4136 rc = -ENETDOWN; 4137 rcu_read_unlock_bh(); 4138 4139 atomic_long_inc(&dev->tx_dropped); 4140 kfree_skb_list(skb); 4141 return rc; 4142 out: 4143 rcu_read_unlock_bh(); 4144 return rc; 4145 } 4146 4147 int dev_queue_xmit(struct sk_buff *skb) 4148 { 4149 return __dev_queue_xmit(skb, NULL); 4150 } 4151 EXPORT_SYMBOL(dev_queue_xmit); 4152 4153 int dev_queue_xmit_accel(struct sk_buff *skb, struct net_device *sb_dev) 4154 { 4155 return __dev_queue_xmit(skb, sb_dev); 4156 } 4157 EXPORT_SYMBOL(dev_queue_xmit_accel); 4158 4159 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id) 4160 { 4161 struct net_device *dev = skb->dev; 4162 struct sk_buff *orig_skb = skb; 4163 struct netdev_queue *txq; 4164 int ret = NETDEV_TX_BUSY; 4165 bool again = false; 4166 4167 if (unlikely(!netif_running(dev) || 4168 !netif_carrier_ok(dev))) 4169 goto drop; 4170 4171 skb = validate_xmit_skb_list(skb, dev, &again); 4172 if (skb != orig_skb) 4173 goto drop; 4174 4175 skb_set_queue_mapping(skb, queue_id); 4176 txq = skb_get_tx_queue(dev, skb); 4177 PRANDOM_ADD_NOISE(skb, dev, txq, jiffies); 4178 4179 local_bh_disable(); 4180 4181 dev_xmit_recursion_inc(); 4182 HARD_TX_LOCK(dev, txq, smp_processor_id()); 4183 if (!netif_xmit_frozen_or_drv_stopped(txq)) 4184 ret = netdev_start_xmit(skb, dev, txq, false); 4185 HARD_TX_UNLOCK(dev, txq); 4186 dev_xmit_recursion_dec(); 4187 4188 local_bh_enable(); 4189 return ret; 4190 drop: 4191 atomic_long_inc(&dev->tx_dropped); 4192 kfree_skb_list(skb); 4193 return NET_XMIT_DROP; 4194 } 4195 EXPORT_SYMBOL(__dev_direct_xmit); 4196 4197 /************************************************************************* 4198 * Receiver routines 4199 *************************************************************************/ 4200 4201 int netdev_max_backlog __read_mostly = 1000; 4202 EXPORT_SYMBOL(netdev_max_backlog); 4203 4204 int netdev_tstamp_prequeue __read_mostly = 1; 4205 int netdev_budget __read_mostly = 300; 4206 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */ 4207 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ; 4208 int weight_p __read_mostly = 64; /* old backlog weight */ 4209 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ 4210 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ 4211 int dev_rx_weight __read_mostly = 64; 4212 int dev_tx_weight __read_mostly = 64; 4213 4214 /* Called with irq disabled */ 4215 static inline void ____napi_schedule(struct softnet_data *sd, 4216 struct napi_struct *napi) 4217 { 4218 struct task_struct *thread; 4219 4220 if (test_bit(NAPI_STATE_THREADED, &napi->state)) { 4221 /* Paired with smp_mb__before_atomic() in 4222 * napi_enable()/dev_set_threaded(). 4223 * Use READ_ONCE() to guarantee a complete 4224 * read on napi->thread. Only call 4225 * wake_up_process() when it's not NULL. 4226 */ 4227 thread = READ_ONCE(napi->thread); 4228 if (thread) { 4229 /* Avoid doing set_bit() if the thread is in 4230 * INTERRUPTIBLE state, cause napi_thread_wait() 4231 * makes sure to proceed with napi polling 4232 * if the thread is explicitly woken from here. 4233 */ 4234 if (READ_ONCE(thread->__state) != TASK_INTERRUPTIBLE) 4235 set_bit(NAPI_STATE_SCHED_THREADED, &napi->state); 4236 wake_up_process(thread); 4237 return; 4238 } 4239 } 4240 4241 list_add_tail(&napi->poll_list, &sd->poll_list); 4242 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4243 } 4244 4245 #ifdef CONFIG_RPS 4246 4247 /* One global table that all flow-based protocols share. */ 4248 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; 4249 EXPORT_SYMBOL(rps_sock_flow_table); 4250 u32 rps_cpu_mask __read_mostly; 4251 EXPORT_SYMBOL(rps_cpu_mask); 4252 4253 struct static_key_false rps_needed __read_mostly; 4254 EXPORT_SYMBOL(rps_needed); 4255 struct static_key_false rfs_needed __read_mostly; 4256 EXPORT_SYMBOL(rfs_needed); 4257 4258 static struct rps_dev_flow * 4259 set_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4260 struct rps_dev_flow *rflow, u16 next_cpu) 4261 { 4262 if (next_cpu < nr_cpu_ids) { 4263 #ifdef CONFIG_RFS_ACCEL 4264 struct netdev_rx_queue *rxqueue; 4265 struct rps_dev_flow_table *flow_table; 4266 struct rps_dev_flow *old_rflow; 4267 u32 flow_id; 4268 u16 rxq_index; 4269 int rc; 4270 4271 /* Should we steer this flow to a different hardware queue? */ 4272 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || 4273 !(dev->features & NETIF_F_NTUPLE)) 4274 goto out; 4275 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); 4276 if (rxq_index == skb_get_rx_queue(skb)) 4277 goto out; 4278 4279 rxqueue = dev->_rx + rxq_index; 4280 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4281 if (!flow_table) 4282 goto out; 4283 flow_id = skb_get_hash(skb) & flow_table->mask; 4284 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, 4285 rxq_index, flow_id); 4286 if (rc < 0) 4287 goto out; 4288 old_rflow = rflow; 4289 rflow = &flow_table->flows[flow_id]; 4290 rflow->filter = rc; 4291 if (old_rflow->filter == rflow->filter) 4292 old_rflow->filter = RPS_NO_FILTER; 4293 out: 4294 #endif 4295 rflow->last_qtail = 4296 per_cpu(softnet_data, next_cpu).input_queue_head; 4297 } 4298 4299 rflow->cpu = next_cpu; 4300 return rflow; 4301 } 4302 4303 /* 4304 * get_rps_cpu is called from netif_receive_skb and returns the target 4305 * CPU from the RPS map of the receiving queue for a given skb. 4306 * rcu_read_lock must be held on entry. 4307 */ 4308 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4309 struct rps_dev_flow **rflowp) 4310 { 4311 const struct rps_sock_flow_table *sock_flow_table; 4312 struct netdev_rx_queue *rxqueue = dev->_rx; 4313 struct rps_dev_flow_table *flow_table; 4314 struct rps_map *map; 4315 int cpu = -1; 4316 u32 tcpu; 4317 u32 hash; 4318 4319 if (skb_rx_queue_recorded(skb)) { 4320 u16 index = skb_get_rx_queue(skb); 4321 4322 if (unlikely(index >= dev->real_num_rx_queues)) { 4323 WARN_ONCE(dev->real_num_rx_queues > 1, 4324 "%s received packet on queue %u, but number " 4325 "of RX queues is %u\n", 4326 dev->name, index, dev->real_num_rx_queues); 4327 goto done; 4328 } 4329 rxqueue += index; 4330 } 4331 4332 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ 4333 4334 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4335 map = rcu_dereference(rxqueue->rps_map); 4336 if (!flow_table && !map) 4337 goto done; 4338 4339 skb_reset_network_header(skb); 4340 hash = skb_get_hash(skb); 4341 if (!hash) 4342 goto done; 4343 4344 sock_flow_table = rcu_dereference(rps_sock_flow_table); 4345 if (flow_table && sock_flow_table) { 4346 struct rps_dev_flow *rflow; 4347 u32 next_cpu; 4348 u32 ident; 4349 4350 /* First check into global flow table if there is a match */ 4351 ident = sock_flow_table->ents[hash & sock_flow_table->mask]; 4352 if ((ident ^ hash) & ~rps_cpu_mask) 4353 goto try_rps; 4354 4355 next_cpu = ident & rps_cpu_mask; 4356 4357 /* OK, now we know there is a match, 4358 * we can look at the local (per receive queue) flow table 4359 */ 4360 rflow = &flow_table->flows[hash & flow_table->mask]; 4361 tcpu = rflow->cpu; 4362 4363 /* 4364 * If the desired CPU (where last recvmsg was done) is 4365 * different from current CPU (one in the rx-queue flow 4366 * table entry), switch if one of the following holds: 4367 * - Current CPU is unset (>= nr_cpu_ids). 4368 * - Current CPU is offline. 4369 * - The current CPU's queue tail has advanced beyond the 4370 * last packet that was enqueued using this table entry. 4371 * This guarantees that all previous packets for the flow 4372 * have been dequeued, thus preserving in order delivery. 4373 */ 4374 if (unlikely(tcpu != next_cpu) && 4375 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || 4376 ((int)(per_cpu(softnet_data, tcpu).input_queue_head - 4377 rflow->last_qtail)) >= 0)) { 4378 tcpu = next_cpu; 4379 rflow = set_rps_cpu(dev, skb, rflow, next_cpu); 4380 } 4381 4382 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { 4383 *rflowp = rflow; 4384 cpu = tcpu; 4385 goto done; 4386 } 4387 } 4388 4389 try_rps: 4390 4391 if (map) { 4392 tcpu = map->cpus[reciprocal_scale(hash, map->len)]; 4393 if (cpu_online(tcpu)) { 4394 cpu = tcpu; 4395 goto done; 4396 } 4397 } 4398 4399 done: 4400 return cpu; 4401 } 4402 4403 #ifdef CONFIG_RFS_ACCEL 4404 4405 /** 4406 * rps_may_expire_flow - check whether an RFS hardware filter may be removed 4407 * @dev: Device on which the filter was set 4408 * @rxq_index: RX queue index 4409 * @flow_id: Flow ID passed to ndo_rx_flow_steer() 4410 * @filter_id: Filter ID returned by ndo_rx_flow_steer() 4411 * 4412 * Drivers that implement ndo_rx_flow_steer() should periodically call 4413 * this function for each installed filter and remove the filters for 4414 * which it returns %true. 4415 */ 4416 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, 4417 u32 flow_id, u16 filter_id) 4418 { 4419 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; 4420 struct rps_dev_flow_table *flow_table; 4421 struct rps_dev_flow *rflow; 4422 bool expire = true; 4423 unsigned int cpu; 4424 4425 rcu_read_lock(); 4426 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4427 if (flow_table && flow_id <= flow_table->mask) { 4428 rflow = &flow_table->flows[flow_id]; 4429 cpu = READ_ONCE(rflow->cpu); 4430 if (rflow->filter == filter_id && cpu < nr_cpu_ids && 4431 ((int)(per_cpu(softnet_data, cpu).input_queue_head - 4432 rflow->last_qtail) < 4433 (int)(10 * flow_table->mask))) 4434 expire = false; 4435 } 4436 rcu_read_unlock(); 4437 return expire; 4438 } 4439 EXPORT_SYMBOL(rps_may_expire_flow); 4440 4441 #endif /* CONFIG_RFS_ACCEL */ 4442 4443 /* Called from hardirq (IPI) context */ 4444 static void rps_trigger_softirq(void *data) 4445 { 4446 struct softnet_data *sd = data; 4447 4448 ____napi_schedule(sd, &sd->backlog); 4449 sd->received_rps++; 4450 } 4451 4452 #endif /* CONFIG_RPS */ 4453 4454 /* 4455 * Check if this softnet_data structure is another cpu one 4456 * If yes, queue it to our IPI list and return 1 4457 * If no, return 0 4458 */ 4459 static int rps_ipi_queued(struct softnet_data *sd) 4460 { 4461 #ifdef CONFIG_RPS 4462 struct softnet_data *mysd = this_cpu_ptr(&softnet_data); 4463 4464 if (sd != mysd) { 4465 sd->rps_ipi_next = mysd->rps_ipi_list; 4466 mysd->rps_ipi_list = sd; 4467 4468 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4469 return 1; 4470 } 4471 #endif /* CONFIG_RPS */ 4472 return 0; 4473 } 4474 4475 #ifdef CONFIG_NET_FLOW_LIMIT 4476 int netdev_flow_limit_table_len __read_mostly = (1 << 12); 4477 #endif 4478 4479 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) 4480 { 4481 #ifdef CONFIG_NET_FLOW_LIMIT 4482 struct sd_flow_limit *fl; 4483 struct softnet_data *sd; 4484 unsigned int old_flow, new_flow; 4485 4486 if (qlen < (netdev_max_backlog >> 1)) 4487 return false; 4488 4489 sd = this_cpu_ptr(&softnet_data); 4490 4491 rcu_read_lock(); 4492 fl = rcu_dereference(sd->flow_limit); 4493 if (fl) { 4494 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); 4495 old_flow = fl->history[fl->history_head]; 4496 fl->history[fl->history_head] = new_flow; 4497 4498 fl->history_head++; 4499 fl->history_head &= FLOW_LIMIT_HISTORY - 1; 4500 4501 if (likely(fl->buckets[old_flow])) 4502 fl->buckets[old_flow]--; 4503 4504 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { 4505 fl->count++; 4506 rcu_read_unlock(); 4507 return true; 4508 } 4509 } 4510 rcu_read_unlock(); 4511 #endif 4512 return false; 4513 } 4514 4515 /* 4516 * enqueue_to_backlog is called to queue an skb to a per CPU backlog 4517 * queue (may be a remote CPU queue). 4518 */ 4519 static int enqueue_to_backlog(struct sk_buff *skb, int cpu, 4520 unsigned int *qtail) 4521 { 4522 struct softnet_data *sd; 4523 unsigned long flags; 4524 unsigned int qlen; 4525 4526 sd = &per_cpu(softnet_data, cpu); 4527 4528 local_irq_save(flags); 4529 4530 rps_lock(sd); 4531 if (!netif_running(skb->dev)) 4532 goto drop; 4533 qlen = skb_queue_len(&sd->input_pkt_queue); 4534 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) { 4535 if (qlen) { 4536 enqueue: 4537 __skb_queue_tail(&sd->input_pkt_queue, skb); 4538 input_queue_tail_incr_save(sd, qtail); 4539 rps_unlock(sd); 4540 local_irq_restore(flags); 4541 return NET_RX_SUCCESS; 4542 } 4543 4544 /* Schedule NAPI for backlog device 4545 * We can use non atomic operation since we own the queue lock 4546 */ 4547 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) { 4548 if (!rps_ipi_queued(sd)) 4549 ____napi_schedule(sd, &sd->backlog); 4550 } 4551 goto enqueue; 4552 } 4553 4554 drop: 4555 sd->dropped++; 4556 rps_unlock(sd); 4557 4558 local_irq_restore(flags); 4559 4560 atomic_long_inc(&skb->dev->rx_dropped); 4561 kfree_skb(skb); 4562 return NET_RX_DROP; 4563 } 4564 4565 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb) 4566 { 4567 struct net_device *dev = skb->dev; 4568 struct netdev_rx_queue *rxqueue; 4569 4570 rxqueue = dev->_rx; 4571 4572 if (skb_rx_queue_recorded(skb)) { 4573 u16 index = skb_get_rx_queue(skb); 4574 4575 if (unlikely(index >= dev->real_num_rx_queues)) { 4576 WARN_ONCE(dev->real_num_rx_queues > 1, 4577 "%s received packet on queue %u, but number " 4578 "of RX queues is %u\n", 4579 dev->name, index, dev->real_num_rx_queues); 4580 4581 return rxqueue; /* Return first rxqueue */ 4582 } 4583 rxqueue += index; 4584 } 4585 return rxqueue; 4586 } 4587 4588 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp, 4589 struct bpf_prog *xdp_prog) 4590 { 4591 void *orig_data, *orig_data_end, *hard_start; 4592 struct netdev_rx_queue *rxqueue; 4593 bool orig_bcast, orig_host; 4594 u32 mac_len, frame_sz; 4595 __be16 orig_eth_type; 4596 struct ethhdr *eth; 4597 u32 metalen, act; 4598 int off; 4599 4600 /* The XDP program wants to see the packet starting at the MAC 4601 * header. 4602 */ 4603 mac_len = skb->data - skb_mac_header(skb); 4604 hard_start = skb->data - skb_headroom(skb); 4605 4606 /* SKB "head" area always have tailroom for skb_shared_info */ 4607 frame_sz = (void *)skb_end_pointer(skb) - hard_start; 4608 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 4609 4610 rxqueue = netif_get_rxqueue(skb); 4611 xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq); 4612 xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len, 4613 skb_headlen(skb) + mac_len, true); 4614 4615 orig_data_end = xdp->data_end; 4616 orig_data = xdp->data; 4617 eth = (struct ethhdr *)xdp->data; 4618 orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr); 4619 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest); 4620 orig_eth_type = eth->h_proto; 4621 4622 act = bpf_prog_run_xdp(xdp_prog, xdp); 4623 4624 /* check if bpf_xdp_adjust_head was used */ 4625 off = xdp->data - orig_data; 4626 if (off) { 4627 if (off > 0) 4628 __skb_pull(skb, off); 4629 else if (off < 0) 4630 __skb_push(skb, -off); 4631 4632 skb->mac_header += off; 4633 skb_reset_network_header(skb); 4634 } 4635 4636 /* check if bpf_xdp_adjust_tail was used */ 4637 off = xdp->data_end - orig_data_end; 4638 if (off != 0) { 4639 skb_set_tail_pointer(skb, xdp->data_end - xdp->data); 4640 skb->len += off; /* positive on grow, negative on shrink */ 4641 } 4642 4643 /* check if XDP changed eth hdr such SKB needs update */ 4644 eth = (struct ethhdr *)xdp->data; 4645 if ((orig_eth_type != eth->h_proto) || 4646 (orig_host != ether_addr_equal_64bits(eth->h_dest, 4647 skb->dev->dev_addr)) || 4648 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) { 4649 __skb_push(skb, ETH_HLEN); 4650 skb->pkt_type = PACKET_HOST; 4651 skb->protocol = eth_type_trans(skb, skb->dev); 4652 } 4653 4654 /* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull 4655 * before calling us again on redirect path. We do not call do_redirect 4656 * as we leave that up to the caller. 4657 * 4658 * Caller is responsible for managing lifetime of skb (i.e. calling 4659 * kfree_skb in response to actions it cannot handle/XDP_DROP). 4660 */ 4661 switch (act) { 4662 case XDP_REDIRECT: 4663 case XDP_TX: 4664 __skb_push(skb, mac_len); 4665 break; 4666 case XDP_PASS: 4667 metalen = xdp->data - xdp->data_meta; 4668 if (metalen) 4669 skb_metadata_set(skb, metalen); 4670 break; 4671 } 4672 4673 return act; 4674 } 4675 4676 static u32 netif_receive_generic_xdp(struct sk_buff *skb, 4677 struct xdp_buff *xdp, 4678 struct bpf_prog *xdp_prog) 4679 { 4680 u32 act = XDP_DROP; 4681 4682 /* Reinjected packets coming from act_mirred or similar should 4683 * not get XDP generic processing. 4684 */ 4685 if (skb_is_redirected(skb)) 4686 return XDP_PASS; 4687 4688 /* XDP packets must be linear and must have sufficient headroom 4689 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also 4690 * native XDP provides, thus we need to do it here as well. 4691 */ 4692 if (skb_cloned(skb) || skb_is_nonlinear(skb) || 4693 skb_headroom(skb) < XDP_PACKET_HEADROOM) { 4694 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb); 4695 int troom = skb->tail + skb->data_len - skb->end; 4696 4697 /* In case we have to go down the path and also linearize, 4698 * then lets do the pskb_expand_head() work just once here. 4699 */ 4700 if (pskb_expand_head(skb, 4701 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0, 4702 troom > 0 ? troom + 128 : 0, GFP_ATOMIC)) 4703 goto do_drop; 4704 if (skb_linearize(skb)) 4705 goto do_drop; 4706 } 4707 4708 act = bpf_prog_run_generic_xdp(skb, xdp, xdp_prog); 4709 switch (act) { 4710 case XDP_REDIRECT: 4711 case XDP_TX: 4712 case XDP_PASS: 4713 break; 4714 default: 4715 bpf_warn_invalid_xdp_action(skb->dev, xdp_prog, act); 4716 fallthrough; 4717 case XDP_ABORTED: 4718 trace_xdp_exception(skb->dev, xdp_prog, act); 4719 fallthrough; 4720 case XDP_DROP: 4721 do_drop: 4722 kfree_skb(skb); 4723 break; 4724 } 4725 4726 return act; 4727 } 4728 4729 /* When doing generic XDP we have to bypass the qdisc layer and the 4730 * network taps in order to match in-driver-XDP behavior. 4731 */ 4732 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog) 4733 { 4734 struct net_device *dev = skb->dev; 4735 struct netdev_queue *txq; 4736 bool free_skb = true; 4737 int cpu, rc; 4738 4739 txq = netdev_core_pick_tx(dev, skb, NULL); 4740 cpu = smp_processor_id(); 4741 HARD_TX_LOCK(dev, txq, cpu); 4742 if (!netif_xmit_stopped(txq)) { 4743 rc = netdev_start_xmit(skb, dev, txq, 0); 4744 if (dev_xmit_complete(rc)) 4745 free_skb = false; 4746 } 4747 HARD_TX_UNLOCK(dev, txq); 4748 if (free_skb) { 4749 trace_xdp_exception(dev, xdp_prog, XDP_TX); 4750 kfree_skb(skb); 4751 } 4752 } 4753 4754 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key); 4755 4756 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb) 4757 { 4758 if (xdp_prog) { 4759 struct xdp_buff xdp; 4760 u32 act; 4761 int err; 4762 4763 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog); 4764 if (act != XDP_PASS) { 4765 switch (act) { 4766 case XDP_REDIRECT: 4767 err = xdp_do_generic_redirect(skb->dev, skb, 4768 &xdp, xdp_prog); 4769 if (err) 4770 goto out_redir; 4771 break; 4772 case XDP_TX: 4773 generic_xdp_tx(skb, xdp_prog); 4774 break; 4775 } 4776 return XDP_DROP; 4777 } 4778 } 4779 return XDP_PASS; 4780 out_redir: 4781 kfree_skb(skb); 4782 return XDP_DROP; 4783 } 4784 EXPORT_SYMBOL_GPL(do_xdp_generic); 4785 4786 static int netif_rx_internal(struct sk_buff *skb) 4787 { 4788 int ret; 4789 4790 net_timestamp_check(netdev_tstamp_prequeue, skb); 4791 4792 trace_netif_rx(skb); 4793 4794 #ifdef CONFIG_RPS 4795 if (static_branch_unlikely(&rps_needed)) { 4796 struct rps_dev_flow voidflow, *rflow = &voidflow; 4797 int cpu; 4798 4799 preempt_disable(); 4800 rcu_read_lock(); 4801 4802 cpu = get_rps_cpu(skb->dev, skb, &rflow); 4803 if (cpu < 0) 4804 cpu = smp_processor_id(); 4805 4806 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 4807 4808 rcu_read_unlock(); 4809 preempt_enable(); 4810 } else 4811 #endif 4812 { 4813 unsigned int qtail; 4814 4815 ret = enqueue_to_backlog(skb, get_cpu(), &qtail); 4816 put_cpu(); 4817 } 4818 return ret; 4819 } 4820 4821 /** 4822 * netif_rx - post buffer to the network code 4823 * @skb: buffer to post 4824 * 4825 * This function receives a packet from a device driver and queues it for 4826 * the upper (protocol) levels to process. It always succeeds. The buffer 4827 * may be dropped during processing for congestion control or by the 4828 * protocol layers. 4829 * 4830 * return values: 4831 * NET_RX_SUCCESS (no congestion) 4832 * NET_RX_DROP (packet was dropped) 4833 * 4834 */ 4835 4836 int netif_rx(struct sk_buff *skb) 4837 { 4838 int ret; 4839 4840 trace_netif_rx_entry(skb); 4841 4842 ret = netif_rx_internal(skb); 4843 trace_netif_rx_exit(ret); 4844 4845 return ret; 4846 } 4847 EXPORT_SYMBOL(netif_rx); 4848 4849 int netif_rx_ni(struct sk_buff *skb) 4850 { 4851 int err; 4852 4853 trace_netif_rx_ni_entry(skb); 4854 4855 preempt_disable(); 4856 err = netif_rx_internal(skb); 4857 if (local_softirq_pending()) 4858 do_softirq(); 4859 preempt_enable(); 4860 trace_netif_rx_ni_exit(err); 4861 4862 return err; 4863 } 4864 EXPORT_SYMBOL(netif_rx_ni); 4865 4866 int netif_rx_any_context(struct sk_buff *skb) 4867 { 4868 /* 4869 * If invoked from contexts which do not invoke bottom half 4870 * processing either at return from interrupt or when softrqs are 4871 * reenabled, use netif_rx_ni() which invokes bottomhalf processing 4872 * directly. 4873 */ 4874 if (in_interrupt()) 4875 return netif_rx(skb); 4876 else 4877 return netif_rx_ni(skb); 4878 } 4879 EXPORT_SYMBOL(netif_rx_any_context); 4880 4881 static __latent_entropy void net_tx_action(struct softirq_action *h) 4882 { 4883 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 4884 4885 if (sd->completion_queue) { 4886 struct sk_buff *clist; 4887 4888 local_irq_disable(); 4889 clist = sd->completion_queue; 4890 sd->completion_queue = NULL; 4891 local_irq_enable(); 4892 4893 while (clist) { 4894 struct sk_buff *skb = clist; 4895 4896 clist = clist->next; 4897 4898 WARN_ON(refcount_read(&skb->users)); 4899 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED)) 4900 trace_consume_skb(skb); 4901 else 4902 trace_kfree_skb(skb, net_tx_action, 4903 SKB_DROP_REASON_NOT_SPECIFIED); 4904 4905 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) 4906 __kfree_skb(skb); 4907 else 4908 __kfree_skb_defer(skb); 4909 } 4910 } 4911 4912 if (sd->output_queue) { 4913 struct Qdisc *head; 4914 4915 local_irq_disable(); 4916 head = sd->output_queue; 4917 sd->output_queue = NULL; 4918 sd->output_queue_tailp = &sd->output_queue; 4919 local_irq_enable(); 4920 4921 rcu_read_lock(); 4922 4923 while (head) { 4924 struct Qdisc *q = head; 4925 spinlock_t *root_lock = NULL; 4926 4927 head = head->next_sched; 4928 4929 /* We need to make sure head->next_sched is read 4930 * before clearing __QDISC_STATE_SCHED 4931 */ 4932 smp_mb__before_atomic(); 4933 4934 if (!(q->flags & TCQ_F_NOLOCK)) { 4935 root_lock = qdisc_lock(q); 4936 spin_lock(root_lock); 4937 } else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, 4938 &q->state))) { 4939 /* There is a synchronize_net() between 4940 * STATE_DEACTIVATED flag being set and 4941 * qdisc_reset()/some_qdisc_is_busy() in 4942 * dev_deactivate(), so we can safely bail out 4943 * early here to avoid data race between 4944 * qdisc_deactivate() and some_qdisc_is_busy() 4945 * for lockless qdisc. 4946 */ 4947 clear_bit(__QDISC_STATE_SCHED, &q->state); 4948 continue; 4949 } 4950 4951 clear_bit(__QDISC_STATE_SCHED, &q->state); 4952 qdisc_run(q); 4953 if (root_lock) 4954 spin_unlock(root_lock); 4955 } 4956 4957 rcu_read_unlock(); 4958 } 4959 4960 xfrm_dev_backlog(sd); 4961 } 4962 4963 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) 4964 /* This hook is defined here for ATM LANE */ 4965 int (*br_fdb_test_addr_hook)(struct net_device *dev, 4966 unsigned char *addr) __read_mostly; 4967 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); 4968 #endif 4969 4970 static inline struct sk_buff * 4971 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 4972 struct net_device *orig_dev, bool *another) 4973 { 4974 #ifdef CONFIG_NET_CLS_ACT 4975 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress); 4976 struct tcf_result cl_res; 4977 4978 /* If there's at least one ingress present somewhere (so 4979 * we get here via enabled static key), remaining devices 4980 * that are not configured with an ingress qdisc will bail 4981 * out here. 4982 */ 4983 if (!miniq) 4984 return skb; 4985 4986 if (*pt_prev) { 4987 *ret = deliver_skb(skb, *pt_prev, orig_dev); 4988 *pt_prev = NULL; 4989 } 4990 4991 qdisc_skb_cb(skb)->pkt_len = skb->len; 4992 tc_skb_cb(skb)->mru = 0; 4993 tc_skb_cb(skb)->post_ct = false; 4994 skb->tc_at_ingress = 1; 4995 mini_qdisc_bstats_cpu_update(miniq, skb); 4996 4997 switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) { 4998 case TC_ACT_OK: 4999 case TC_ACT_RECLASSIFY: 5000 skb->tc_index = TC_H_MIN(cl_res.classid); 5001 break; 5002 case TC_ACT_SHOT: 5003 mini_qdisc_qstats_cpu_drop(miniq); 5004 kfree_skb(skb); 5005 return NULL; 5006 case TC_ACT_STOLEN: 5007 case TC_ACT_QUEUED: 5008 case TC_ACT_TRAP: 5009 consume_skb(skb); 5010 return NULL; 5011 case TC_ACT_REDIRECT: 5012 /* skb_mac_header check was done by cls/act_bpf, so 5013 * we can safely push the L2 header back before 5014 * redirecting to another netdev 5015 */ 5016 __skb_push(skb, skb->mac_len); 5017 if (skb_do_redirect(skb) == -EAGAIN) { 5018 __skb_pull(skb, skb->mac_len); 5019 *another = true; 5020 break; 5021 } 5022 return NULL; 5023 case TC_ACT_CONSUMED: 5024 return NULL; 5025 default: 5026 break; 5027 } 5028 #endif /* CONFIG_NET_CLS_ACT */ 5029 return skb; 5030 } 5031 5032 /** 5033 * netdev_is_rx_handler_busy - check if receive handler is registered 5034 * @dev: device to check 5035 * 5036 * Check if a receive handler is already registered for a given device. 5037 * Return true if there one. 5038 * 5039 * The caller must hold the rtnl_mutex. 5040 */ 5041 bool netdev_is_rx_handler_busy(struct net_device *dev) 5042 { 5043 ASSERT_RTNL(); 5044 return dev && rtnl_dereference(dev->rx_handler); 5045 } 5046 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); 5047 5048 /** 5049 * netdev_rx_handler_register - register receive handler 5050 * @dev: device to register a handler for 5051 * @rx_handler: receive handler to register 5052 * @rx_handler_data: data pointer that is used by rx handler 5053 * 5054 * Register a receive handler for a device. This handler will then be 5055 * called from __netif_receive_skb. A negative errno code is returned 5056 * on a failure. 5057 * 5058 * The caller must hold the rtnl_mutex. 5059 * 5060 * For a general description of rx_handler, see enum rx_handler_result. 5061 */ 5062 int netdev_rx_handler_register(struct net_device *dev, 5063 rx_handler_func_t *rx_handler, 5064 void *rx_handler_data) 5065 { 5066 if (netdev_is_rx_handler_busy(dev)) 5067 return -EBUSY; 5068 5069 if (dev->priv_flags & IFF_NO_RX_HANDLER) 5070 return -EINVAL; 5071 5072 /* Note: rx_handler_data must be set before rx_handler */ 5073 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); 5074 rcu_assign_pointer(dev->rx_handler, rx_handler); 5075 5076 return 0; 5077 } 5078 EXPORT_SYMBOL_GPL(netdev_rx_handler_register); 5079 5080 /** 5081 * netdev_rx_handler_unregister - unregister receive handler 5082 * @dev: device to unregister a handler from 5083 * 5084 * Unregister a receive handler from a device. 5085 * 5086 * The caller must hold the rtnl_mutex. 5087 */ 5088 void netdev_rx_handler_unregister(struct net_device *dev) 5089 { 5090 5091 ASSERT_RTNL(); 5092 RCU_INIT_POINTER(dev->rx_handler, NULL); 5093 /* a reader seeing a non NULL rx_handler in a rcu_read_lock() 5094 * section has a guarantee to see a non NULL rx_handler_data 5095 * as well. 5096 */ 5097 synchronize_net(); 5098 RCU_INIT_POINTER(dev->rx_handler_data, NULL); 5099 } 5100 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); 5101 5102 /* 5103 * Limit the use of PFMEMALLOC reserves to those protocols that implement 5104 * the special handling of PFMEMALLOC skbs. 5105 */ 5106 static bool skb_pfmemalloc_protocol(struct sk_buff *skb) 5107 { 5108 switch (skb->protocol) { 5109 case htons(ETH_P_ARP): 5110 case htons(ETH_P_IP): 5111 case htons(ETH_P_IPV6): 5112 case htons(ETH_P_8021Q): 5113 case htons(ETH_P_8021AD): 5114 return true; 5115 default: 5116 return false; 5117 } 5118 } 5119 5120 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, 5121 int *ret, struct net_device *orig_dev) 5122 { 5123 if (nf_hook_ingress_active(skb)) { 5124 int ingress_retval; 5125 5126 if (*pt_prev) { 5127 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5128 *pt_prev = NULL; 5129 } 5130 5131 rcu_read_lock(); 5132 ingress_retval = nf_hook_ingress(skb); 5133 rcu_read_unlock(); 5134 return ingress_retval; 5135 } 5136 return 0; 5137 } 5138 5139 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc, 5140 struct packet_type **ppt_prev) 5141 { 5142 struct packet_type *ptype, *pt_prev; 5143 rx_handler_func_t *rx_handler; 5144 struct sk_buff *skb = *pskb; 5145 struct net_device *orig_dev; 5146 bool deliver_exact = false; 5147 int ret = NET_RX_DROP; 5148 __be16 type; 5149 5150 net_timestamp_check(!netdev_tstamp_prequeue, skb); 5151 5152 trace_netif_receive_skb(skb); 5153 5154 orig_dev = skb->dev; 5155 5156 skb_reset_network_header(skb); 5157 if (!skb_transport_header_was_set(skb)) 5158 skb_reset_transport_header(skb); 5159 skb_reset_mac_len(skb); 5160 5161 pt_prev = NULL; 5162 5163 another_round: 5164 skb->skb_iif = skb->dev->ifindex; 5165 5166 __this_cpu_inc(softnet_data.processed); 5167 5168 if (static_branch_unlikely(&generic_xdp_needed_key)) { 5169 int ret2; 5170 5171 migrate_disable(); 5172 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb); 5173 migrate_enable(); 5174 5175 if (ret2 != XDP_PASS) { 5176 ret = NET_RX_DROP; 5177 goto out; 5178 } 5179 } 5180 5181 if (eth_type_vlan(skb->protocol)) { 5182 skb = skb_vlan_untag(skb); 5183 if (unlikely(!skb)) 5184 goto out; 5185 } 5186 5187 if (skb_skip_tc_classify(skb)) 5188 goto skip_classify; 5189 5190 if (pfmemalloc) 5191 goto skip_taps; 5192 5193 list_for_each_entry_rcu(ptype, &ptype_all, list) { 5194 if (pt_prev) 5195 ret = deliver_skb(skb, pt_prev, orig_dev); 5196 pt_prev = ptype; 5197 } 5198 5199 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { 5200 if (pt_prev) 5201 ret = deliver_skb(skb, pt_prev, orig_dev); 5202 pt_prev = ptype; 5203 } 5204 5205 skip_taps: 5206 #ifdef CONFIG_NET_INGRESS 5207 if (static_branch_unlikely(&ingress_needed_key)) { 5208 bool another = false; 5209 5210 nf_skip_egress(skb, true); 5211 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev, 5212 &another); 5213 if (another) 5214 goto another_round; 5215 if (!skb) 5216 goto out; 5217 5218 nf_skip_egress(skb, false); 5219 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) 5220 goto out; 5221 } 5222 #endif 5223 skb_reset_redirect(skb); 5224 skip_classify: 5225 if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) 5226 goto drop; 5227 5228 if (skb_vlan_tag_present(skb)) { 5229 if (pt_prev) { 5230 ret = deliver_skb(skb, pt_prev, orig_dev); 5231 pt_prev = NULL; 5232 } 5233 if (vlan_do_receive(&skb)) 5234 goto another_round; 5235 else if (unlikely(!skb)) 5236 goto out; 5237 } 5238 5239 rx_handler = rcu_dereference(skb->dev->rx_handler); 5240 if (rx_handler) { 5241 if (pt_prev) { 5242 ret = deliver_skb(skb, pt_prev, orig_dev); 5243 pt_prev = NULL; 5244 } 5245 switch (rx_handler(&skb)) { 5246 case RX_HANDLER_CONSUMED: 5247 ret = NET_RX_SUCCESS; 5248 goto out; 5249 case RX_HANDLER_ANOTHER: 5250 goto another_round; 5251 case RX_HANDLER_EXACT: 5252 deliver_exact = true; 5253 break; 5254 case RX_HANDLER_PASS: 5255 break; 5256 default: 5257 BUG(); 5258 } 5259 } 5260 5261 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) { 5262 check_vlan_id: 5263 if (skb_vlan_tag_get_id(skb)) { 5264 /* Vlan id is non 0 and vlan_do_receive() above couldn't 5265 * find vlan device. 5266 */ 5267 skb->pkt_type = PACKET_OTHERHOST; 5268 } else if (eth_type_vlan(skb->protocol)) { 5269 /* Outer header is 802.1P with vlan 0, inner header is 5270 * 802.1Q or 802.1AD and vlan_do_receive() above could 5271 * not find vlan dev for vlan id 0. 5272 */ 5273 __vlan_hwaccel_clear_tag(skb); 5274 skb = skb_vlan_untag(skb); 5275 if (unlikely(!skb)) 5276 goto out; 5277 if (vlan_do_receive(&skb)) 5278 /* After stripping off 802.1P header with vlan 0 5279 * vlan dev is found for inner header. 5280 */ 5281 goto another_round; 5282 else if (unlikely(!skb)) 5283 goto out; 5284 else 5285 /* We have stripped outer 802.1P vlan 0 header. 5286 * But could not find vlan dev. 5287 * check again for vlan id to set OTHERHOST. 5288 */ 5289 goto check_vlan_id; 5290 } 5291 /* Note: we might in the future use prio bits 5292 * and set skb->priority like in vlan_do_receive() 5293 * For the time being, just ignore Priority Code Point 5294 */ 5295 __vlan_hwaccel_clear_tag(skb); 5296 } 5297 5298 type = skb->protocol; 5299 5300 /* deliver only exact match when indicated */ 5301 if (likely(!deliver_exact)) { 5302 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5303 &ptype_base[ntohs(type) & 5304 PTYPE_HASH_MASK]); 5305 } 5306 5307 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5308 &orig_dev->ptype_specific); 5309 5310 if (unlikely(skb->dev != orig_dev)) { 5311 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5312 &skb->dev->ptype_specific); 5313 } 5314 5315 if (pt_prev) { 5316 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 5317 goto drop; 5318 *ppt_prev = pt_prev; 5319 } else { 5320 drop: 5321 if (!deliver_exact) 5322 atomic_long_inc(&skb->dev->rx_dropped); 5323 else 5324 atomic_long_inc(&skb->dev->rx_nohandler); 5325 kfree_skb(skb); 5326 /* Jamal, now you will not able to escape explaining 5327 * me how you were going to use this. :-) 5328 */ 5329 ret = NET_RX_DROP; 5330 } 5331 5332 out: 5333 /* The invariant here is that if *ppt_prev is not NULL 5334 * then skb should also be non-NULL. 5335 * 5336 * Apparently *ppt_prev assignment above holds this invariant due to 5337 * skb dereferencing near it. 5338 */ 5339 *pskb = skb; 5340 return ret; 5341 } 5342 5343 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc) 5344 { 5345 struct net_device *orig_dev = skb->dev; 5346 struct packet_type *pt_prev = NULL; 5347 int ret; 5348 5349 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5350 if (pt_prev) 5351 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb, 5352 skb->dev, pt_prev, orig_dev); 5353 return ret; 5354 } 5355 5356 /** 5357 * netif_receive_skb_core - special purpose version of netif_receive_skb 5358 * @skb: buffer to process 5359 * 5360 * More direct receive version of netif_receive_skb(). It should 5361 * only be used by callers that have a need to skip RPS and Generic XDP. 5362 * Caller must also take care of handling if ``(page_is_)pfmemalloc``. 5363 * 5364 * This function may only be called from softirq context and interrupts 5365 * should be enabled. 5366 * 5367 * Return values (usually ignored): 5368 * NET_RX_SUCCESS: no congestion 5369 * NET_RX_DROP: packet was dropped 5370 */ 5371 int netif_receive_skb_core(struct sk_buff *skb) 5372 { 5373 int ret; 5374 5375 rcu_read_lock(); 5376 ret = __netif_receive_skb_one_core(skb, false); 5377 rcu_read_unlock(); 5378 5379 return ret; 5380 } 5381 EXPORT_SYMBOL(netif_receive_skb_core); 5382 5383 static inline void __netif_receive_skb_list_ptype(struct list_head *head, 5384 struct packet_type *pt_prev, 5385 struct net_device *orig_dev) 5386 { 5387 struct sk_buff *skb, *next; 5388 5389 if (!pt_prev) 5390 return; 5391 if (list_empty(head)) 5392 return; 5393 if (pt_prev->list_func != NULL) 5394 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv, 5395 ip_list_rcv, head, pt_prev, orig_dev); 5396 else 5397 list_for_each_entry_safe(skb, next, head, list) { 5398 skb_list_del_init(skb); 5399 pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 5400 } 5401 } 5402 5403 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc) 5404 { 5405 /* Fast-path assumptions: 5406 * - There is no RX handler. 5407 * - Only one packet_type matches. 5408 * If either of these fails, we will end up doing some per-packet 5409 * processing in-line, then handling the 'last ptype' for the whole 5410 * sublist. This can't cause out-of-order delivery to any single ptype, 5411 * because the 'last ptype' must be constant across the sublist, and all 5412 * other ptypes are handled per-packet. 5413 */ 5414 /* Current (common) ptype of sublist */ 5415 struct packet_type *pt_curr = NULL; 5416 /* Current (common) orig_dev of sublist */ 5417 struct net_device *od_curr = NULL; 5418 struct list_head sublist; 5419 struct sk_buff *skb, *next; 5420 5421 INIT_LIST_HEAD(&sublist); 5422 list_for_each_entry_safe(skb, next, head, list) { 5423 struct net_device *orig_dev = skb->dev; 5424 struct packet_type *pt_prev = NULL; 5425 5426 skb_list_del_init(skb); 5427 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5428 if (!pt_prev) 5429 continue; 5430 if (pt_curr != pt_prev || od_curr != orig_dev) { 5431 /* dispatch old sublist */ 5432 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5433 /* start new sublist */ 5434 INIT_LIST_HEAD(&sublist); 5435 pt_curr = pt_prev; 5436 od_curr = orig_dev; 5437 } 5438 list_add_tail(&skb->list, &sublist); 5439 } 5440 5441 /* dispatch final sublist */ 5442 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5443 } 5444 5445 static int __netif_receive_skb(struct sk_buff *skb) 5446 { 5447 int ret; 5448 5449 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { 5450 unsigned int noreclaim_flag; 5451 5452 /* 5453 * PFMEMALLOC skbs are special, they should 5454 * - be delivered to SOCK_MEMALLOC sockets only 5455 * - stay away from userspace 5456 * - have bounded memory usage 5457 * 5458 * Use PF_MEMALLOC as this saves us from propagating the allocation 5459 * context down to all allocation sites. 5460 */ 5461 noreclaim_flag = memalloc_noreclaim_save(); 5462 ret = __netif_receive_skb_one_core(skb, true); 5463 memalloc_noreclaim_restore(noreclaim_flag); 5464 } else 5465 ret = __netif_receive_skb_one_core(skb, false); 5466 5467 return ret; 5468 } 5469 5470 static void __netif_receive_skb_list(struct list_head *head) 5471 { 5472 unsigned long noreclaim_flag = 0; 5473 struct sk_buff *skb, *next; 5474 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */ 5475 5476 list_for_each_entry_safe(skb, next, head, list) { 5477 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) { 5478 struct list_head sublist; 5479 5480 /* Handle the previous sublist */ 5481 list_cut_before(&sublist, head, &skb->list); 5482 if (!list_empty(&sublist)) 5483 __netif_receive_skb_list_core(&sublist, pfmemalloc); 5484 pfmemalloc = !pfmemalloc; 5485 /* See comments in __netif_receive_skb */ 5486 if (pfmemalloc) 5487 noreclaim_flag = memalloc_noreclaim_save(); 5488 else 5489 memalloc_noreclaim_restore(noreclaim_flag); 5490 } 5491 } 5492 /* Handle the remaining sublist */ 5493 if (!list_empty(head)) 5494 __netif_receive_skb_list_core(head, pfmemalloc); 5495 /* Restore pflags */ 5496 if (pfmemalloc) 5497 memalloc_noreclaim_restore(noreclaim_flag); 5498 } 5499 5500 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp) 5501 { 5502 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog); 5503 struct bpf_prog *new = xdp->prog; 5504 int ret = 0; 5505 5506 switch (xdp->command) { 5507 case XDP_SETUP_PROG: 5508 rcu_assign_pointer(dev->xdp_prog, new); 5509 if (old) 5510 bpf_prog_put(old); 5511 5512 if (old && !new) { 5513 static_branch_dec(&generic_xdp_needed_key); 5514 } else if (new && !old) { 5515 static_branch_inc(&generic_xdp_needed_key); 5516 dev_disable_lro(dev); 5517 dev_disable_gro_hw(dev); 5518 } 5519 break; 5520 5521 default: 5522 ret = -EINVAL; 5523 break; 5524 } 5525 5526 return ret; 5527 } 5528 5529 static int netif_receive_skb_internal(struct sk_buff *skb) 5530 { 5531 int ret; 5532 5533 net_timestamp_check(netdev_tstamp_prequeue, skb); 5534 5535 if (skb_defer_rx_timestamp(skb)) 5536 return NET_RX_SUCCESS; 5537 5538 rcu_read_lock(); 5539 #ifdef CONFIG_RPS 5540 if (static_branch_unlikely(&rps_needed)) { 5541 struct rps_dev_flow voidflow, *rflow = &voidflow; 5542 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5543 5544 if (cpu >= 0) { 5545 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5546 rcu_read_unlock(); 5547 return ret; 5548 } 5549 } 5550 #endif 5551 ret = __netif_receive_skb(skb); 5552 rcu_read_unlock(); 5553 return ret; 5554 } 5555 5556 void netif_receive_skb_list_internal(struct list_head *head) 5557 { 5558 struct sk_buff *skb, *next; 5559 struct list_head sublist; 5560 5561 INIT_LIST_HEAD(&sublist); 5562 list_for_each_entry_safe(skb, next, head, list) { 5563 net_timestamp_check(netdev_tstamp_prequeue, skb); 5564 skb_list_del_init(skb); 5565 if (!skb_defer_rx_timestamp(skb)) 5566 list_add_tail(&skb->list, &sublist); 5567 } 5568 list_splice_init(&sublist, head); 5569 5570 rcu_read_lock(); 5571 #ifdef CONFIG_RPS 5572 if (static_branch_unlikely(&rps_needed)) { 5573 list_for_each_entry_safe(skb, next, head, list) { 5574 struct rps_dev_flow voidflow, *rflow = &voidflow; 5575 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5576 5577 if (cpu >= 0) { 5578 /* Will be handled, remove from list */ 5579 skb_list_del_init(skb); 5580 enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5581 } 5582 } 5583 } 5584 #endif 5585 __netif_receive_skb_list(head); 5586 rcu_read_unlock(); 5587 } 5588 5589 /** 5590 * netif_receive_skb - process receive buffer from network 5591 * @skb: buffer to process 5592 * 5593 * netif_receive_skb() is the main receive data processing function. 5594 * It always succeeds. The buffer may be dropped during processing 5595 * for congestion control or by the protocol layers. 5596 * 5597 * This function may only be called from softirq context and interrupts 5598 * should be enabled. 5599 * 5600 * Return values (usually ignored): 5601 * NET_RX_SUCCESS: no congestion 5602 * NET_RX_DROP: packet was dropped 5603 */ 5604 int netif_receive_skb(struct sk_buff *skb) 5605 { 5606 int ret; 5607 5608 trace_netif_receive_skb_entry(skb); 5609 5610 ret = netif_receive_skb_internal(skb); 5611 trace_netif_receive_skb_exit(ret); 5612 5613 return ret; 5614 } 5615 EXPORT_SYMBOL(netif_receive_skb); 5616 5617 /** 5618 * netif_receive_skb_list - process many receive buffers from network 5619 * @head: list of skbs to process. 5620 * 5621 * Since return value of netif_receive_skb() is normally ignored, and 5622 * wouldn't be meaningful for a list, this function returns void. 5623 * 5624 * This function may only be called from softirq context and interrupts 5625 * should be enabled. 5626 */ 5627 void netif_receive_skb_list(struct list_head *head) 5628 { 5629 struct sk_buff *skb; 5630 5631 if (list_empty(head)) 5632 return; 5633 if (trace_netif_receive_skb_list_entry_enabled()) { 5634 list_for_each_entry(skb, head, list) 5635 trace_netif_receive_skb_list_entry(skb); 5636 } 5637 netif_receive_skb_list_internal(head); 5638 trace_netif_receive_skb_list_exit(0); 5639 } 5640 EXPORT_SYMBOL(netif_receive_skb_list); 5641 5642 static DEFINE_PER_CPU(struct work_struct, flush_works); 5643 5644 /* Network device is going away, flush any packets still pending */ 5645 static void flush_backlog(struct work_struct *work) 5646 { 5647 struct sk_buff *skb, *tmp; 5648 struct softnet_data *sd; 5649 5650 local_bh_disable(); 5651 sd = this_cpu_ptr(&softnet_data); 5652 5653 local_irq_disable(); 5654 rps_lock(sd); 5655 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { 5656 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5657 __skb_unlink(skb, &sd->input_pkt_queue); 5658 dev_kfree_skb_irq(skb); 5659 input_queue_head_incr(sd); 5660 } 5661 } 5662 rps_unlock(sd); 5663 local_irq_enable(); 5664 5665 skb_queue_walk_safe(&sd->process_queue, skb, tmp) { 5666 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5667 __skb_unlink(skb, &sd->process_queue); 5668 kfree_skb(skb); 5669 input_queue_head_incr(sd); 5670 } 5671 } 5672 local_bh_enable(); 5673 } 5674 5675 static bool flush_required(int cpu) 5676 { 5677 #if IS_ENABLED(CONFIG_RPS) 5678 struct softnet_data *sd = &per_cpu(softnet_data, cpu); 5679 bool do_flush; 5680 5681 local_irq_disable(); 5682 rps_lock(sd); 5683 5684 /* as insertion into process_queue happens with the rps lock held, 5685 * process_queue access may race only with dequeue 5686 */ 5687 do_flush = !skb_queue_empty(&sd->input_pkt_queue) || 5688 !skb_queue_empty_lockless(&sd->process_queue); 5689 rps_unlock(sd); 5690 local_irq_enable(); 5691 5692 return do_flush; 5693 #endif 5694 /* without RPS we can't safely check input_pkt_queue: during a 5695 * concurrent remote skb_queue_splice() we can detect as empty both 5696 * input_pkt_queue and process_queue even if the latter could end-up 5697 * containing a lot of packets. 5698 */ 5699 return true; 5700 } 5701 5702 static void flush_all_backlogs(void) 5703 { 5704 static cpumask_t flush_cpus; 5705 unsigned int cpu; 5706 5707 /* since we are under rtnl lock protection we can use static data 5708 * for the cpumask and avoid allocating on stack the possibly 5709 * large mask 5710 */ 5711 ASSERT_RTNL(); 5712 5713 cpus_read_lock(); 5714 5715 cpumask_clear(&flush_cpus); 5716 for_each_online_cpu(cpu) { 5717 if (flush_required(cpu)) { 5718 queue_work_on(cpu, system_highpri_wq, 5719 per_cpu_ptr(&flush_works, cpu)); 5720 cpumask_set_cpu(cpu, &flush_cpus); 5721 } 5722 } 5723 5724 /* we can have in flight packet[s] on the cpus we are not flushing, 5725 * synchronize_net() in unregister_netdevice_many() will take care of 5726 * them 5727 */ 5728 for_each_cpu(cpu, &flush_cpus) 5729 flush_work(per_cpu_ptr(&flush_works, cpu)); 5730 5731 cpus_read_unlock(); 5732 } 5733 5734 static void net_rps_send_ipi(struct softnet_data *remsd) 5735 { 5736 #ifdef CONFIG_RPS 5737 while (remsd) { 5738 struct softnet_data *next = remsd->rps_ipi_next; 5739 5740 if (cpu_online(remsd->cpu)) 5741 smp_call_function_single_async(remsd->cpu, &remsd->csd); 5742 remsd = next; 5743 } 5744 #endif 5745 } 5746 5747 /* 5748 * net_rps_action_and_irq_enable sends any pending IPI's for rps. 5749 * Note: called with local irq disabled, but exits with local irq enabled. 5750 */ 5751 static void net_rps_action_and_irq_enable(struct softnet_data *sd) 5752 { 5753 #ifdef CONFIG_RPS 5754 struct softnet_data *remsd = sd->rps_ipi_list; 5755 5756 if (remsd) { 5757 sd->rps_ipi_list = NULL; 5758 5759 local_irq_enable(); 5760 5761 /* Send pending IPI's to kick RPS processing on remote cpus. */ 5762 net_rps_send_ipi(remsd); 5763 } else 5764 #endif 5765 local_irq_enable(); 5766 } 5767 5768 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) 5769 { 5770 #ifdef CONFIG_RPS 5771 return sd->rps_ipi_list != NULL; 5772 #else 5773 return false; 5774 #endif 5775 } 5776 5777 static int process_backlog(struct napi_struct *napi, int quota) 5778 { 5779 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); 5780 bool again = true; 5781 int work = 0; 5782 5783 /* Check if we have pending ipi, its better to send them now, 5784 * not waiting net_rx_action() end. 5785 */ 5786 if (sd_has_rps_ipi_waiting(sd)) { 5787 local_irq_disable(); 5788 net_rps_action_and_irq_enable(sd); 5789 } 5790 5791 napi->weight = dev_rx_weight; 5792 while (again) { 5793 struct sk_buff *skb; 5794 5795 while ((skb = __skb_dequeue(&sd->process_queue))) { 5796 rcu_read_lock(); 5797 __netif_receive_skb(skb); 5798 rcu_read_unlock(); 5799 input_queue_head_incr(sd); 5800 if (++work >= quota) 5801 return work; 5802 5803 } 5804 5805 local_irq_disable(); 5806 rps_lock(sd); 5807 if (skb_queue_empty(&sd->input_pkt_queue)) { 5808 /* 5809 * Inline a custom version of __napi_complete(). 5810 * only current cpu owns and manipulates this napi, 5811 * and NAPI_STATE_SCHED is the only possible flag set 5812 * on backlog. 5813 * We can use a plain write instead of clear_bit(), 5814 * and we dont need an smp_mb() memory barrier. 5815 */ 5816 napi->state = 0; 5817 again = false; 5818 } else { 5819 skb_queue_splice_tail_init(&sd->input_pkt_queue, 5820 &sd->process_queue); 5821 } 5822 rps_unlock(sd); 5823 local_irq_enable(); 5824 } 5825 5826 return work; 5827 } 5828 5829 /** 5830 * __napi_schedule - schedule for receive 5831 * @n: entry to schedule 5832 * 5833 * The entry's receive function will be scheduled to run. 5834 * Consider using __napi_schedule_irqoff() if hard irqs are masked. 5835 */ 5836 void __napi_schedule(struct napi_struct *n) 5837 { 5838 unsigned long flags; 5839 5840 local_irq_save(flags); 5841 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 5842 local_irq_restore(flags); 5843 } 5844 EXPORT_SYMBOL(__napi_schedule); 5845 5846 /** 5847 * napi_schedule_prep - check if napi can be scheduled 5848 * @n: napi context 5849 * 5850 * Test if NAPI routine is already running, and if not mark 5851 * it as running. This is used as a condition variable to 5852 * insure only one NAPI poll instance runs. We also make 5853 * sure there is no pending NAPI disable. 5854 */ 5855 bool napi_schedule_prep(struct napi_struct *n) 5856 { 5857 unsigned long val, new; 5858 5859 do { 5860 val = READ_ONCE(n->state); 5861 if (unlikely(val & NAPIF_STATE_DISABLE)) 5862 return false; 5863 new = val | NAPIF_STATE_SCHED; 5864 5865 /* Sets STATE_MISSED bit if STATE_SCHED was already set 5866 * This was suggested by Alexander Duyck, as compiler 5867 * emits better code than : 5868 * if (val & NAPIF_STATE_SCHED) 5869 * new |= NAPIF_STATE_MISSED; 5870 */ 5871 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * 5872 NAPIF_STATE_MISSED; 5873 } while (cmpxchg(&n->state, val, new) != val); 5874 5875 return !(val & NAPIF_STATE_SCHED); 5876 } 5877 EXPORT_SYMBOL(napi_schedule_prep); 5878 5879 /** 5880 * __napi_schedule_irqoff - schedule for receive 5881 * @n: entry to schedule 5882 * 5883 * Variant of __napi_schedule() assuming hard irqs are masked. 5884 * 5885 * On PREEMPT_RT enabled kernels this maps to __napi_schedule() 5886 * because the interrupt disabled assumption might not be true 5887 * due to force-threaded interrupts and spinlock substitution. 5888 */ 5889 void __napi_schedule_irqoff(struct napi_struct *n) 5890 { 5891 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 5892 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 5893 else 5894 __napi_schedule(n); 5895 } 5896 EXPORT_SYMBOL(__napi_schedule_irqoff); 5897 5898 bool napi_complete_done(struct napi_struct *n, int work_done) 5899 { 5900 unsigned long flags, val, new, timeout = 0; 5901 bool ret = true; 5902 5903 /* 5904 * 1) Don't let napi dequeue from the cpu poll list 5905 * just in case its running on a different cpu. 5906 * 2) If we are busy polling, do nothing here, we have 5907 * the guarantee we will be called later. 5908 */ 5909 if (unlikely(n->state & (NAPIF_STATE_NPSVC | 5910 NAPIF_STATE_IN_BUSY_POLL))) 5911 return false; 5912 5913 if (work_done) { 5914 if (n->gro_bitmask) 5915 timeout = READ_ONCE(n->dev->gro_flush_timeout); 5916 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs); 5917 } 5918 if (n->defer_hard_irqs_count > 0) { 5919 n->defer_hard_irqs_count--; 5920 timeout = READ_ONCE(n->dev->gro_flush_timeout); 5921 if (timeout) 5922 ret = false; 5923 } 5924 if (n->gro_bitmask) { 5925 /* When the NAPI instance uses a timeout and keeps postponing 5926 * it, we need to bound somehow the time packets are kept in 5927 * the GRO layer 5928 */ 5929 napi_gro_flush(n, !!timeout); 5930 } 5931 5932 gro_normal_list(n); 5933 5934 if (unlikely(!list_empty(&n->poll_list))) { 5935 /* If n->poll_list is not empty, we need to mask irqs */ 5936 local_irq_save(flags); 5937 list_del_init(&n->poll_list); 5938 local_irq_restore(flags); 5939 } 5940 5941 do { 5942 val = READ_ONCE(n->state); 5943 5944 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); 5945 5946 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED | 5947 NAPIF_STATE_SCHED_THREADED | 5948 NAPIF_STATE_PREFER_BUSY_POLL); 5949 5950 /* If STATE_MISSED was set, leave STATE_SCHED set, 5951 * because we will call napi->poll() one more time. 5952 * This C code was suggested by Alexander Duyck to help gcc. 5953 */ 5954 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * 5955 NAPIF_STATE_SCHED; 5956 } while (cmpxchg(&n->state, val, new) != val); 5957 5958 if (unlikely(val & NAPIF_STATE_MISSED)) { 5959 __napi_schedule(n); 5960 return false; 5961 } 5962 5963 if (timeout) 5964 hrtimer_start(&n->timer, ns_to_ktime(timeout), 5965 HRTIMER_MODE_REL_PINNED); 5966 return ret; 5967 } 5968 EXPORT_SYMBOL(napi_complete_done); 5969 5970 /* must be called under rcu_read_lock(), as we dont take a reference */ 5971 static struct napi_struct *napi_by_id(unsigned int napi_id) 5972 { 5973 unsigned int hash = napi_id % HASH_SIZE(napi_hash); 5974 struct napi_struct *napi; 5975 5976 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) 5977 if (napi->napi_id == napi_id) 5978 return napi; 5979 5980 return NULL; 5981 } 5982 5983 #if defined(CONFIG_NET_RX_BUSY_POLL) 5984 5985 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule) 5986 { 5987 if (!skip_schedule) { 5988 gro_normal_list(napi); 5989 __napi_schedule(napi); 5990 return; 5991 } 5992 5993 if (napi->gro_bitmask) { 5994 /* flush too old packets 5995 * If HZ < 1000, flush all packets. 5996 */ 5997 napi_gro_flush(napi, HZ >= 1000); 5998 } 5999 6000 gro_normal_list(napi); 6001 clear_bit(NAPI_STATE_SCHED, &napi->state); 6002 } 6003 6004 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll, 6005 u16 budget) 6006 { 6007 bool skip_schedule = false; 6008 unsigned long timeout; 6009 int rc; 6010 6011 /* Busy polling means there is a high chance device driver hard irq 6012 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was 6013 * set in napi_schedule_prep(). 6014 * Since we are about to call napi->poll() once more, we can safely 6015 * clear NAPI_STATE_MISSED. 6016 * 6017 * Note: x86 could use a single "lock and ..." instruction 6018 * to perform these two clear_bit() 6019 */ 6020 clear_bit(NAPI_STATE_MISSED, &napi->state); 6021 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); 6022 6023 local_bh_disable(); 6024 6025 if (prefer_busy_poll) { 6026 napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs); 6027 timeout = READ_ONCE(napi->dev->gro_flush_timeout); 6028 if (napi->defer_hard_irqs_count && timeout) { 6029 hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED); 6030 skip_schedule = true; 6031 } 6032 } 6033 6034 /* All we really want here is to re-enable device interrupts. 6035 * Ideally, a new ndo_busy_poll_stop() could avoid another round. 6036 */ 6037 rc = napi->poll(napi, budget); 6038 /* We can't gro_normal_list() here, because napi->poll() might have 6039 * rearmed the napi (napi_complete_done()) in which case it could 6040 * already be running on another CPU. 6041 */ 6042 trace_napi_poll(napi, rc, budget); 6043 netpoll_poll_unlock(have_poll_lock); 6044 if (rc == budget) 6045 __busy_poll_stop(napi, skip_schedule); 6046 local_bh_enable(); 6047 } 6048 6049 void napi_busy_loop(unsigned int napi_id, 6050 bool (*loop_end)(void *, unsigned long), 6051 void *loop_end_arg, bool prefer_busy_poll, u16 budget) 6052 { 6053 unsigned long start_time = loop_end ? busy_loop_current_time() : 0; 6054 int (*napi_poll)(struct napi_struct *napi, int budget); 6055 void *have_poll_lock = NULL; 6056 struct napi_struct *napi; 6057 6058 restart: 6059 napi_poll = NULL; 6060 6061 rcu_read_lock(); 6062 6063 napi = napi_by_id(napi_id); 6064 if (!napi) 6065 goto out; 6066 6067 preempt_disable(); 6068 for (;;) { 6069 int work = 0; 6070 6071 local_bh_disable(); 6072 if (!napi_poll) { 6073 unsigned long val = READ_ONCE(napi->state); 6074 6075 /* If multiple threads are competing for this napi, 6076 * we avoid dirtying napi->state as much as we can. 6077 */ 6078 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | 6079 NAPIF_STATE_IN_BUSY_POLL)) { 6080 if (prefer_busy_poll) 6081 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6082 goto count; 6083 } 6084 if (cmpxchg(&napi->state, val, 6085 val | NAPIF_STATE_IN_BUSY_POLL | 6086 NAPIF_STATE_SCHED) != val) { 6087 if (prefer_busy_poll) 6088 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6089 goto count; 6090 } 6091 have_poll_lock = netpoll_poll_lock(napi); 6092 napi_poll = napi->poll; 6093 } 6094 work = napi_poll(napi, budget); 6095 trace_napi_poll(napi, work, budget); 6096 gro_normal_list(napi); 6097 count: 6098 if (work > 0) 6099 __NET_ADD_STATS(dev_net(napi->dev), 6100 LINUX_MIB_BUSYPOLLRXPACKETS, work); 6101 local_bh_enable(); 6102 6103 if (!loop_end || loop_end(loop_end_arg, start_time)) 6104 break; 6105 6106 if (unlikely(need_resched())) { 6107 if (napi_poll) 6108 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6109 preempt_enable(); 6110 rcu_read_unlock(); 6111 cond_resched(); 6112 if (loop_end(loop_end_arg, start_time)) 6113 return; 6114 goto restart; 6115 } 6116 cpu_relax(); 6117 } 6118 if (napi_poll) 6119 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget); 6120 preempt_enable(); 6121 out: 6122 rcu_read_unlock(); 6123 } 6124 EXPORT_SYMBOL(napi_busy_loop); 6125 6126 #endif /* CONFIG_NET_RX_BUSY_POLL */ 6127 6128 static void napi_hash_add(struct napi_struct *napi) 6129 { 6130 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state)) 6131 return; 6132 6133 spin_lock(&napi_hash_lock); 6134 6135 /* 0..NR_CPUS range is reserved for sender_cpu use */ 6136 do { 6137 if (unlikely(++napi_gen_id < MIN_NAPI_ID)) 6138 napi_gen_id = MIN_NAPI_ID; 6139 } while (napi_by_id(napi_gen_id)); 6140 napi->napi_id = napi_gen_id; 6141 6142 hlist_add_head_rcu(&napi->napi_hash_node, 6143 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); 6144 6145 spin_unlock(&napi_hash_lock); 6146 } 6147 6148 /* Warning : caller is responsible to make sure rcu grace period 6149 * is respected before freeing memory containing @napi 6150 */ 6151 static void napi_hash_del(struct napi_struct *napi) 6152 { 6153 spin_lock(&napi_hash_lock); 6154 6155 hlist_del_init_rcu(&napi->napi_hash_node); 6156 6157 spin_unlock(&napi_hash_lock); 6158 } 6159 6160 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) 6161 { 6162 struct napi_struct *napi; 6163 6164 napi = container_of(timer, struct napi_struct, timer); 6165 6166 /* Note : we use a relaxed variant of napi_schedule_prep() not setting 6167 * NAPI_STATE_MISSED, since we do not react to a device IRQ. 6168 */ 6169 if (!napi_disable_pending(napi) && 6170 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) { 6171 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6172 __napi_schedule_irqoff(napi); 6173 } 6174 6175 return HRTIMER_NORESTART; 6176 } 6177 6178 static void init_gro_hash(struct napi_struct *napi) 6179 { 6180 int i; 6181 6182 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6183 INIT_LIST_HEAD(&napi->gro_hash[i].list); 6184 napi->gro_hash[i].count = 0; 6185 } 6186 napi->gro_bitmask = 0; 6187 } 6188 6189 int dev_set_threaded(struct net_device *dev, bool threaded) 6190 { 6191 struct napi_struct *napi; 6192 int err = 0; 6193 6194 if (dev->threaded == threaded) 6195 return 0; 6196 6197 if (threaded) { 6198 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6199 if (!napi->thread) { 6200 err = napi_kthread_create(napi); 6201 if (err) { 6202 threaded = false; 6203 break; 6204 } 6205 } 6206 } 6207 } 6208 6209 dev->threaded = threaded; 6210 6211 /* Make sure kthread is created before THREADED bit 6212 * is set. 6213 */ 6214 smp_mb__before_atomic(); 6215 6216 /* Setting/unsetting threaded mode on a napi might not immediately 6217 * take effect, if the current napi instance is actively being 6218 * polled. In this case, the switch between threaded mode and 6219 * softirq mode will happen in the next round of napi_schedule(). 6220 * This should not cause hiccups/stalls to the live traffic. 6221 */ 6222 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6223 if (threaded) 6224 set_bit(NAPI_STATE_THREADED, &napi->state); 6225 else 6226 clear_bit(NAPI_STATE_THREADED, &napi->state); 6227 } 6228 6229 return err; 6230 } 6231 EXPORT_SYMBOL(dev_set_threaded); 6232 6233 void netif_napi_add(struct net_device *dev, struct napi_struct *napi, 6234 int (*poll)(struct napi_struct *, int), int weight) 6235 { 6236 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state))) 6237 return; 6238 6239 INIT_LIST_HEAD(&napi->poll_list); 6240 INIT_HLIST_NODE(&napi->napi_hash_node); 6241 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); 6242 napi->timer.function = napi_watchdog; 6243 init_gro_hash(napi); 6244 napi->skb = NULL; 6245 INIT_LIST_HEAD(&napi->rx_list); 6246 napi->rx_count = 0; 6247 napi->poll = poll; 6248 if (weight > NAPI_POLL_WEIGHT) 6249 netdev_err_once(dev, "%s() called with weight %d\n", __func__, 6250 weight); 6251 napi->weight = weight; 6252 napi->dev = dev; 6253 #ifdef CONFIG_NETPOLL 6254 napi->poll_owner = -1; 6255 #endif 6256 set_bit(NAPI_STATE_SCHED, &napi->state); 6257 set_bit(NAPI_STATE_NPSVC, &napi->state); 6258 list_add_rcu(&napi->dev_list, &dev->napi_list); 6259 napi_hash_add(napi); 6260 /* Create kthread for this napi if dev->threaded is set. 6261 * Clear dev->threaded if kthread creation failed so that 6262 * threaded mode will not be enabled in napi_enable(). 6263 */ 6264 if (dev->threaded && napi_kthread_create(napi)) 6265 dev->threaded = 0; 6266 } 6267 EXPORT_SYMBOL(netif_napi_add); 6268 6269 void napi_disable(struct napi_struct *n) 6270 { 6271 unsigned long val, new; 6272 6273 might_sleep(); 6274 set_bit(NAPI_STATE_DISABLE, &n->state); 6275 6276 for ( ; ; ) { 6277 val = READ_ONCE(n->state); 6278 if (val & (NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC)) { 6279 usleep_range(20, 200); 6280 continue; 6281 } 6282 6283 new = val | NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC; 6284 new &= ~(NAPIF_STATE_THREADED | NAPIF_STATE_PREFER_BUSY_POLL); 6285 6286 if (cmpxchg(&n->state, val, new) == val) 6287 break; 6288 } 6289 6290 hrtimer_cancel(&n->timer); 6291 6292 clear_bit(NAPI_STATE_DISABLE, &n->state); 6293 } 6294 EXPORT_SYMBOL(napi_disable); 6295 6296 /** 6297 * napi_enable - enable NAPI scheduling 6298 * @n: NAPI context 6299 * 6300 * Resume NAPI from being scheduled on this context. 6301 * Must be paired with napi_disable. 6302 */ 6303 void napi_enable(struct napi_struct *n) 6304 { 6305 unsigned long val, new; 6306 6307 do { 6308 val = READ_ONCE(n->state); 6309 BUG_ON(!test_bit(NAPI_STATE_SCHED, &val)); 6310 6311 new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC); 6312 if (n->dev->threaded && n->thread) 6313 new |= NAPIF_STATE_THREADED; 6314 } while (cmpxchg(&n->state, val, new) != val); 6315 } 6316 EXPORT_SYMBOL(napi_enable); 6317 6318 static void flush_gro_hash(struct napi_struct *napi) 6319 { 6320 int i; 6321 6322 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6323 struct sk_buff *skb, *n; 6324 6325 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list) 6326 kfree_skb(skb); 6327 napi->gro_hash[i].count = 0; 6328 } 6329 } 6330 6331 /* Must be called in process context */ 6332 void __netif_napi_del(struct napi_struct *napi) 6333 { 6334 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state)) 6335 return; 6336 6337 napi_hash_del(napi); 6338 list_del_rcu(&napi->dev_list); 6339 napi_free_frags(napi); 6340 6341 flush_gro_hash(napi); 6342 napi->gro_bitmask = 0; 6343 6344 if (napi->thread) { 6345 kthread_stop(napi->thread); 6346 napi->thread = NULL; 6347 } 6348 } 6349 EXPORT_SYMBOL(__netif_napi_del); 6350 6351 static int __napi_poll(struct napi_struct *n, bool *repoll) 6352 { 6353 int work, weight; 6354 6355 weight = n->weight; 6356 6357 /* This NAPI_STATE_SCHED test is for avoiding a race 6358 * with netpoll's poll_napi(). Only the entity which 6359 * obtains the lock and sees NAPI_STATE_SCHED set will 6360 * actually make the ->poll() call. Therefore we avoid 6361 * accidentally calling ->poll() when NAPI is not scheduled. 6362 */ 6363 work = 0; 6364 if (test_bit(NAPI_STATE_SCHED, &n->state)) { 6365 work = n->poll(n, weight); 6366 trace_napi_poll(n, work, weight); 6367 } 6368 6369 if (unlikely(work > weight)) 6370 netdev_err_once(n->dev, "NAPI poll function %pS returned %d, exceeding its budget of %d.\n", 6371 n->poll, work, weight); 6372 6373 if (likely(work < weight)) 6374 return work; 6375 6376 /* Drivers must not modify the NAPI state if they 6377 * consume the entire weight. In such cases this code 6378 * still "owns" the NAPI instance and therefore can 6379 * move the instance around on the list at-will. 6380 */ 6381 if (unlikely(napi_disable_pending(n))) { 6382 napi_complete(n); 6383 return work; 6384 } 6385 6386 /* The NAPI context has more processing work, but busy-polling 6387 * is preferred. Exit early. 6388 */ 6389 if (napi_prefer_busy_poll(n)) { 6390 if (napi_complete_done(n, work)) { 6391 /* If timeout is not set, we need to make sure 6392 * that the NAPI is re-scheduled. 6393 */ 6394 napi_schedule(n); 6395 } 6396 return work; 6397 } 6398 6399 if (n->gro_bitmask) { 6400 /* flush too old packets 6401 * If HZ < 1000, flush all packets. 6402 */ 6403 napi_gro_flush(n, HZ >= 1000); 6404 } 6405 6406 gro_normal_list(n); 6407 6408 /* Some drivers may have called napi_schedule 6409 * prior to exhausting their budget. 6410 */ 6411 if (unlikely(!list_empty(&n->poll_list))) { 6412 pr_warn_once("%s: Budget exhausted after napi rescheduled\n", 6413 n->dev ? n->dev->name : "backlog"); 6414 return work; 6415 } 6416 6417 *repoll = true; 6418 6419 return work; 6420 } 6421 6422 static int napi_poll(struct napi_struct *n, struct list_head *repoll) 6423 { 6424 bool do_repoll = false; 6425 void *have; 6426 int work; 6427 6428 list_del_init(&n->poll_list); 6429 6430 have = netpoll_poll_lock(n); 6431 6432 work = __napi_poll(n, &do_repoll); 6433 6434 if (do_repoll) 6435 list_add_tail(&n->poll_list, repoll); 6436 6437 netpoll_poll_unlock(have); 6438 6439 return work; 6440 } 6441 6442 static int napi_thread_wait(struct napi_struct *napi) 6443 { 6444 bool woken = false; 6445 6446 set_current_state(TASK_INTERRUPTIBLE); 6447 6448 while (!kthread_should_stop()) { 6449 /* Testing SCHED_THREADED bit here to make sure the current 6450 * kthread owns this napi and could poll on this napi. 6451 * Testing SCHED bit is not enough because SCHED bit might be 6452 * set by some other busy poll thread or by napi_disable(). 6453 */ 6454 if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state) || woken) { 6455 WARN_ON(!list_empty(&napi->poll_list)); 6456 __set_current_state(TASK_RUNNING); 6457 return 0; 6458 } 6459 6460 schedule(); 6461 /* woken being true indicates this thread owns this napi. */ 6462 woken = true; 6463 set_current_state(TASK_INTERRUPTIBLE); 6464 } 6465 __set_current_state(TASK_RUNNING); 6466 6467 return -1; 6468 } 6469 6470 static int napi_threaded_poll(void *data) 6471 { 6472 struct napi_struct *napi = data; 6473 void *have; 6474 6475 while (!napi_thread_wait(napi)) { 6476 for (;;) { 6477 bool repoll = false; 6478 6479 local_bh_disable(); 6480 6481 have = netpoll_poll_lock(napi); 6482 __napi_poll(napi, &repoll); 6483 netpoll_poll_unlock(have); 6484 6485 local_bh_enable(); 6486 6487 if (!repoll) 6488 break; 6489 6490 cond_resched(); 6491 } 6492 } 6493 return 0; 6494 } 6495 6496 static __latent_entropy void net_rx_action(struct softirq_action *h) 6497 { 6498 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 6499 unsigned long time_limit = jiffies + 6500 usecs_to_jiffies(netdev_budget_usecs); 6501 int budget = netdev_budget; 6502 LIST_HEAD(list); 6503 LIST_HEAD(repoll); 6504 6505 local_irq_disable(); 6506 list_splice_init(&sd->poll_list, &list); 6507 local_irq_enable(); 6508 6509 for (;;) { 6510 struct napi_struct *n; 6511 6512 if (list_empty(&list)) { 6513 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll)) 6514 return; 6515 break; 6516 } 6517 6518 n = list_first_entry(&list, struct napi_struct, poll_list); 6519 budget -= napi_poll(n, &repoll); 6520 6521 /* If softirq window is exhausted then punt. 6522 * Allow this to run for 2 jiffies since which will allow 6523 * an average latency of 1.5/HZ. 6524 */ 6525 if (unlikely(budget <= 0 || 6526 time_after_eq(jiffies, time_limit))) { 6527 sd->time_squeeze++; 6528 break; 6529 } 6530 } 6531 6532 local_irq_disable(); 6533 6534 list_splice_tail_init(&sd->poll_list, &list); 6535 list_splice_tail(&repoll, &list); 6536 list_splice(&list, &sd->poll_list); 6537 if (!list_empty(&sd->poll_list)) 6538 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 6539 6540 net_rps_action_and_irq_enable(sd); 6541 } 6542 6543 struct netdev_adjacent { 6544 struct net_device *dev; 6545 netdevice_tracker dev_tracker; 6546 6547 /* upper master flag, there can only be one master device per list */ 6548 bool master; 6549 6550 /* lookup ignore flag */ 6551 bool ignore; 6552 6553 /* counter for the number of times this device was added to us */ 6554 u16 ref_nr; 6555 6556 /* private field for the users */ 6557 void *private; 6558 6559 struct list_head list; 6560 struct rcu_head rcu; 6561 }; 6562 6563 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, 6564 struct list_head *adj_list) 6565 { 6566 struct netdev_adjacent *adj; 6567 6568 list_for_each_entry(adj, adj_list, list) { 6569 if (adj->dev == adj_dev) 6570 return adj; 6571 } 6572 return NULL; 6573 } 6574 6575 static int ____netdev_has_upper_dev(struct net_device *upper_dev, 6576 struct netdev_nested_priv *priv) 6577 { 6578 struct net_device *dev = (struct net_device *)priv->data; 6579 6580 return upper_dev == dev; 6581 } 6582 6583 /** 6584 * netdev_has_upper_dev - Check if device is linked to an upper device 6585 * @dev: device 6586 * @upper_dev: upper device to check 6587 * 6588 * Find out if a device is linked to specified upper device and return true 6589 * in case it is. Note that this checks only immediate upper device, 6590 * not through a complete stack of devices. The caller must hold the RTNL lock. 6591 */ 6592 bool netdev_has_upper_dev(struct net_device *dev, 6593 struct net_device *upper_dev) 6594 { 6595 struct netdev_nested_priv priv = { 6596 .data = (void *)upper_dev, 6597 }; 6598 6599 ASSERT_RTNL(); 6600 6601 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 6602 &priv); 6603 } 6604 EXPORT_SYMBOL(netdev_has_upper_dev); 6605 6606 /** 6607 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device 6608 * @dev: device 6609 * @upper_dev: upper device to check 6610 * 6611 * Find out if a device is linked to specified upper device and return true 6612 * in case it is. Note that this checks the entire upper device chain. 6613 * The caller must hold rcu lock. 6614 */ 6615 6616 bool netdev_has_upper_dev_all_rcu(struct net_device *dev, 6617 struct net_device *upper_dev) 6618 { 6619 struct netdev_nested_priv priv = { 6620 .data = (void *)upper_dev, 6621 }; 6622 6623 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 6624 &priv); 6625 } 6626 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); 6627 6628 /** 6629 * netdev_has_any_upper_dev - Check if device is linked to some device 6630 * @dev: device 6631 * 6632 * Find out if a device is linked to an upper device and return true in case 6633 * it is. The caller must hold the RTNL lock. 6634 */ 6635 bool netdev_has_any_upper_dev(struct net_device *dev) 6636 { 6637 ASSERT_RTNL(); 6638 6639 return !list_empty(&dev->adj_list.upper); 6640 } 6641 EXPORT_SYMBOL(netdev_has_any_upper_dev); 6642 6643 /** 6644 * netdev_master_upper_dev_get - Get master upper device 6645 * @dev: device 6646 * 6647 * Find a master upper device and return pointer to it or NULL in case 6648 * it's not there. The caller must hold the RTNL lock. 6649 */ 6650 struct net_device *netdev_master_upper_dev_get(struct net_device *dev) 6651 { 6652 struct netdev_adjacent *upper; 6653 6654 ASSERT_RTNL(); 6655 6656 if (list_empty(&dev->adj_list.upper)) 6657 return NULL; 6658 6659 upper = list_first_entry(&dev->adj_list.upper, 6660 struct netdev_adjacent, list); 6661 if (likely(upper->master)) 6662 return upper->dev; 6663 return NULL; 6664 } 6665 EXPORT_SYMBOL(netdev_master_upper_dev_get); 6666 6667 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev) 6668 { 6669 struct netdev_adjacent *upper; 6670 6671 ASSERT_RTNL(); 6672 6673 if (list_empty(&dev->adj_list.upper)) 6674 return NULL; 6675 6676 upper = list_first_entry(&dev->adj_list.upper, 6677 struct netdev_adjacent, list); 6678 if (likely(upper->master) && !upper->ignore) 6679 return upper->dev; 6680 return NULL; 6681 } 6682 6683 /** 6684 * netdev_has_any_lower_dev - Check if device is linked to some device 6685 * @dev: device 6686 * 6687 * Find out if a device is linked to a lower device and return true in case 6688 * it is. The caller must hold the RTNL lock. 6689 */ 6690 static bool netdev_has_any_lower_dev(struct net_device *dev) 6691 { 6692 ASSERT_RTNL(); 6693 6694 return !list_empty(&dev->adj_list.lower); 6695 } 6696 6697 void *netdev_adjacent_get_private(struct list_head *adj_list) 6698 { 6699 struct netdev_adjacent *adj; 6700 6701 adj = list_entry(adj_list, struct netdev_adjacent, list); 6702 6703 return adj->private; 6704 } 6705 EXPORT_SYMBOL(netdev_adjacent_get_private); 6706 6707 /** 6708 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list 6709 * @dev: device 6710 * @iter: list_head ** of the current position 6711 * 6712 * Gets the next device from the dev's upper list, starting from iter 6713 * position. The caller must hold RCU read lock. 6714 */ 6715 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, 6716 struct list_head **iter) 6717 { 6718 struct netdev_adjacent *upper; 6719 6720 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 6721 6722 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 6723 6724 if (&upper->list == &dev->adj_list.upper) 6725 return NULL; 6726 6727 *iter = &upper->list; 6728 6729 return upper->dev; 6730 } 6731 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); 6732 6733 static struct net_device *__netdev_next_upper_dev(struct net_device *dev, 6734 struct list_head **iter, 6735 bool *ignore) 6736 { 6737 struct netdev_adjacent *upper; 6738 6739 upper = list_entry((*iter)->next, struct netdev_adjacent, list); 6740 6741 if (&upper->list == &dev->adj_list.upper) 6742 return NULL; 6743 6744 *iter = &upper->list; 6745 *ignore = upper->ignore; 6746 6747 return upper->dev; 6748 } 6749 6750 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, 6751 struct list_head **iter) 6752 { 6753 struct netdev_adjacent *upper; 6754 6755 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 6756 6757 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 6758 6759 if (&upper->list == &dev->adj_list.upper) 6760 return NULL; 6761 6762 *iter = &upper->list; 6763 6764 return upper->dev; 6765 } 6766 6767 static int __netdev_walk_all_upper_dev(struct net_device *dev, 6768 int (*fn)(struct net_device *dev, 6769 struct netdev_nested_priv *priv), 6770 struct netdev_nested_priv *priv) 6771 { 6772 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 6773 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 6774 int ret, cur = 0; 6775 bool ignore; 6776 6777 now = dev; 6778 iter = &dev->adj_list.upper; 6779 6780 while (1) { 6781 if (now != dev) { 6782 ret = fn(now, priv); 6783 if (ret) 6784 return ret; 6785 } 6786 6787 next = NULL; 6788 while (1) { 6789 udev = __netdev_next_upper_dev(now, &iter, &ignore); 6790 if (!udev) 6791 break; 6792 if (ignore) 6793 continue; 6794 6795 next = udev; 6796 niter = &udev->adj_list.upper; 6797 dev_stack[cur] = now; 6798 iter_stack[cur++] = iter; 6799 break; 6800 } 6801 6802 if (!next) { 6803 if (!cur) 6804 return 0; 6805 next = dev_stack[--cur]; 6806 niter = iter_stack[cur]; 6807 } 6808 6809 now = next; 6810 iter = niter; 6811 } 6812 6813 return 0; 6814 } 6815 6816 int netdev_walk_all_upper_dev_rcu(struct net_device *dev, 6817 int (*fn)(struct net_device *dev, 6818 struct netdev_nested_priv *priv), 6819 struct netdev_nested_priv *priv) 6820 { 6821 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 6822 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 6823 int ret, cur = 0; 6824 6825 now = dev; 6826 iter = &dev->adj_list.upper; 6827 6828 while (1) { 6829 if (now != dev) { 6830 ret = fn(now, priv); 6831 if (ret) 6832 return ret; 6833 } 6834 6835 next = NULL; 6836 while (1) { 6837 udev = netdev_next_upper_dev_rcu(now, &iter); 6838 if (!udev) 6839 break; 6840 6841 next = udev; 6842 niter = &udev->adj_list.upper; 6843 dev_stack[cur] = now; 6844 iter_stack[cur++] = iter; 6845 break; 6846 } 6847 6848 if (!next) { 6849 if (!cur) 6850 return 0; 6851 next = dev_stack[--cur]; 6852 niter = iter_stack[cur]; 6853 } 6854 6855 now = next; 6856 iter = niter; 6857 } 6858 6859 return 0; 6860 } 6861 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); 6862 6863 static bool __netdev_has_upper_dev(struct net_device *dev, 6864 struct net_device *upper_dev) 6865 { 6866 struct netdev_nested_priv priv = { 6867 .flags = 0, 6868 .data = (void *)upper_dev, 6869 }; 6870 6871 ASSERT_RTNL(); 6872 6873 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev, 6874 &priv); 6875 } 6876 6877 /** 6878 * netdev_lower_get_next_private - Get the next ->private from the 6879 * lower neighbour list 6880 * @dev: device 6881 * @iter: list_head ** of the current position 6882 * 6883 * Gets the next netdev_adjacent->private from the dev's lower neighbour 6884 * list, starting from iter position. The caller must hold either hold the 6885 * RTNL lock or its own locking that guarantees that the neighbour lower 6886 * list will remain unchanged. 6887 */ 6888 void *netdev_lower_get_next_private(struct net_device *dev, 6889 struct list_head **iter) 6890 { 6891 struct netdev_adjacent *lower; 6892 6893 lower = list_entry(*iter, struct netdev_adjacent, list); 6894 6895 if (&lower->list == &dev->adj_list.lower) 6896 return NULL; 6897 6898 *iter = lower->list.next; 6899 6900 return lower->private; 6901 } 6902 EXPORT_SYMBOL(netdev_lower_get_next_private); 6903 6904 /** 6905 * netdev_lower_get_next_private_rcu - Get the next ->private from the 6906 * lower neighbour list, RCU 6907 * variant 6908 * @dev: device 6909 * @iter: list_head ** of the current position 6910 * 6911 * Gets the next netdev_adjacent->private from the dev's lower neighbour 6912 * list, starting from iter position. The caller must hold RCU read lock. 6913 */ 6914 void *netdev_lower_get_next_private_rcu(struct net_device *dev, 6915 struct list_head **iter) 6916 { 6917 struct netdev_adjacent *lower; 6918 6919 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 6920 6921 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 6922 6923 if (&lower->list == &dev->adj_list.lower) 6924 return NULL; 6925 6926 *iter = &lower->list; 6927 6928 return lower->private; 6929 } 6930 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); 6931 6932 /** 6933 * netdev_lower_get_next - Get the next device from the lower neighbour 6934 * list 6935 * @dev: device 6936 * @iter: list_head ** of the current position 6937 * 6938 * Gets the next netdev_adjacent from the dev's lower neighbour 6939 * list, starting from iter position. The caller must hold RTNL lock or 6940 * its own locking that guarantees that the neighbour lower 6941 * list will remain unchanged. 6942 */ 6943 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) 6944 { 6945 struct netdev_adjacent *lower; 6946 6947 lower = list_entry(*iter, struct netdev_adjacent, list); 6948 6949 if (&lower->list == &dev->adj_list.lower) 6950 return NULL; 6951 6952 *iter = lower->list.next; 6953 6954 return lower->dev; 6955 } 6956 EXPORT_SYMBOL(netdev_lower_get_next); 6957 6958 static struct net_device *netdev_next_lower_dev(struct net_device *dev, 6959 struct list_head **iter) 6960 { 6961 struct netdev_adjacent *lower; 6962 6963 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 6964 6965 if (&lower->list == &dev->adj_list.lower) 6966 return NULL; 6967 6968 *iter = &lower->list; 6969 6970 return lower->dev; 6971 } 6972 6973 static struct net_device *__netdev_next_lower_dev(struct net_device *dev, 6974 struct list_head **iter, 6975 bool *ignore) 6976 { 6977 struct netdev_adjacent *lower; 6978 6979 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 6980 6981 if (&lower->list == &dev->adj_list.lower) 6982 return NULL; 6983 6984 *iter = &lower->list; 6985 *ignore = lower->ignore; 6986 6987 return lower->dev; 6988 } 6989 6990 int netdev_walk_all_lower_dev(struct net_device *dev, 6991 int (*fn)(struct net_device *dev, 6992 struct netdev_nested_priv *priv), 6993 struct netdev_nested_priv *priv) 6994 { 6995 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 6996 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 6997 int ret, cur = 0; 6998 6999 now = dev; 7000 iter = &dev->adj_list.lower; 7001 7002 while (1) { 7003 if (now != dev) { 7004 ret = fn(now, priv); 7005 if (ret) 7006 return ret; 7007 } 7008 7009 next = NULL; 7010 while (1) { 7011 ldev = netdev_next_lower_dev(now, &iter); 7012 if (!ldev) 7013 break; 7014 7015 next = ldev; 7016 niter = &ldev->adj_list.lower; 7017 dev_stack[cur] = now; 7018 iter_stack[cur++] = iter; 7019 break; 7020 } 7021 7022 if (!next) { 7023 if (!cur) 7024 return 0; 7025 next = dev_stack[--cur]; 7026 niter = iter_stack[cur]; 7027 } 7028 7029 now = next; 7030 iter = niter; 7031 } 7032 7033 return 0; 7034 } 7035 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); 7036 7037 static int __netdev_walk_all_lower_dev(struct net_device *dev, 7038 int (*fn)(struct net_device *dev, 7039 struct netdev_nested_priv *priv), 7040 struct netdev_nested_priv *priv) 7041 { 7042 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7043 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7044 int ret, cur = 0; 7045 bool ignore; 7046 7047 now = dev; 7048 iter = &dev->adj_list.lower; 7049 7050 while (1) { 7051 if (now != dev) { 7052 ret = fn(now, priv); 7053 if (ret) 7054 return ret; 7055 } 7056 7057 next = NULL; 7058 while (1) { 7059 ldev = __netdev_next_lower_dev(now, &iter, &ignore); 7060 if (!ldev) 7061 break; 7062 if (ignore) 7063 continue; 7064 7065 next = ldev; 7066 niter = &ldev->adj_list.lower; 7067 dev_stack[cur] = now; 7068 iter_stack[cur++] = iter; 7069 break; 7070 } 7071 7072 if (!next) { 7073 if (!cur) 7074 return 0; 7075 next = dev_stack[--cur]; 7076 niter = iter_stack[cur]; 7077 } 7078 7079 now = next; 7080 iter = niter; 7081 } 7082 7083 return 0; 7084 } 7085 7086 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, 7087 struct list_head **iter) 7088 { 7089 struct netdev_adjacent *lower; 7090 7091 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7092 if (&lower->list == &dev->adj_list.lower) 7093 return NULL; 7094 7095 *iter = &lower->list; 7096 7097 return lower->dev; 7098 } 7099 EXPORT_SYMBOL(netdev_next_lower_dev_rcu); 7100 7101 static u8 __netdev_upper_depth(struct net_device *dev) 7102 { 7103 struct net_device *udev; 7104 struct list_head *iter; 7105 u8 max_depth = 0; 7106 bool ignore; 7107 7108 for (iter = &dev->adj_list.upper, 7109 udev = __netdev_next_upper_dev(dev, &iter, &ignore); 7110 udev; 7111 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) { 7112 if (ignore) 7113 continue; 7114 if (max_depth < udev->upper_level) 7115 max_depth = udev->upper_level; 7116 } 7117 7118 return max_depth; 7119 } 7120 7121 static u8 __netdev_lower_depth(struct net_device *dev) 7122 { 7123 struct net_device *ldev; 7124 struct list_head *iter; 7125 u8 max_depth = 0; 7126 bool ignore; 7127 7128 for (iter = &dev->adj_list.lower, 7129 ldev = __netdev_next_lower_dev(dev, &iter, &ignore); 7130 ldev; 7131 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) { 7132 if (ignore) 7133 continue; 7134 if (max_depth < ldev->lower_level) 7135 max_depth = ldev->lower_level; 7136 } 7137 7138 return max_depth; 7139 } 7140 7141 static int __netdev_update_upper_level(struct net_device *dev, 7142 struct netdev_nested_priv *__unused) 7143 { 7144 dev->upper_level = __netdev_upper_depth(dev) + 1; 7145 return 0; 7146 } 7147 7148 static int __netdev_update_lower_level(struct net_device *dev, 7149 struct netdev_nested_priv *priv) 7150 { 7151 dev->lower_level = __netdev_lower_depth(dev) + 1; 7152 7153 #ifdef CONFIG_LOCKDEP 7154 if (!priv) 7155 return 0; 7156 7157 if (priv->flags & NESTED_SYNC_IMM) 7158 dev->nested_level = dev->lower_level - 1; 7159 if (priv->flags & NESTED_SYNC_TODO) 7160 net_unlink_todo(dev); 7161 #endif 7162 return 0; 7163 } 7164 7165 int netdev_walk_all_lower_dev_rcu(struct net_device *dev, 7166 int (*fn)(struct net_device *dev, 7167 struct netdev_nested_priv *priv), 7168 struct netdev_nested_priv *priv) 7169 { 7170 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7171 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7172 int ret, cur = 0; 7173 7174 now = dev; 7175 iter = &dev->adj_list.lower; 7176 7177 while (1) { 7178 if (now != dev) { 7179 ret = fn(now, priv); 7180 if (ret) 7181 return ret; 7182 } 7183 7184 next = NULL; 7185 while (1) { 7186 ldev = netdev_next_lower_dev_rcu(now, &iter); 7187 if (!ldev) 7188 break; 7189 7190 next = ldev; 7191 niter = &ldev->adj_list.lower; 7192 dev_stack[cur] = now; 7193 iter_stack[cur++] = iter; 7194 break; 7195 } 7196 7197 if (!next) { 7198 if (!cur) 7199 return 0; 7200 next = dev_stack[--cur]; 7201 niter = iter_stack[cur]; 7202 } 7203 7204 now = next; 7205 iter = niter; 7206 } 7207 7208 return 0; 7209 } 7210 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); 7211 7212 /** 7213 * netdev_lower_get_first_private_rcu - Get the first ->private from the 7214 * lower neighbour list, RCU 7215 * variant 7216 * @dev: device 7217 * 7218 * Gets the first netdev_adjacent->private from the dev's lower neighbour 7219 * list. The caller must hold RCU read lock. 7220 */ 7221 void *netdev_lower_get_first_private_rcu(struct net_device *dev) 7222 { 7223 struct netdev_adjacent *lower; 7224 7225 lower = list_first_or_null_rcu(&dev->adj_list.lower, 7226 struct netdev_adjacent, list); 7227 if (lower) 7228 return lower->private; 7229 return NULL; 7230 } 7231 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); 7232 7233 /** 7234 * netdev_master_upper_dev_get_rcu - Get master upper device 7235 * @dev: device 7236 * 7237 * Find a master upper device and return pointer to it or NULL in case 7238 * it's not there. The caller must hold the RCU read lock. 7239 */ 7240 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) 7241 { 7242 struct netdev_adjacent *upper; 7243 7244 upper = list_first_or_null_rcu(&dev->adj_list.upper, 7245 struct netdev_adjacent, list); 7246 if (upper && likely(upper->master)) 7247 return upper->dev; 7248 return NULL; 7249 } 7250 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); 7251 7252 static int netdev_adjacent_sysfs_add(struct net_device *dev, 7253 struct net_device *adj_dev, 7254 struct list_head *dev_list) 7255 { 7256 char linkname[IFNAMSIZ+7]; 7257 7258 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7259 "upper_%s" : "lower_%s", adj_dev->name); 7260 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), 7261 linkname); 7262 } 7263 static void netdev_adjacent_sysfs_del(struct net_device *dev, 7264 char *name, 7265 struct list_head *dev_list) 7266 { 7267 char linkname[IFNAMSIZ+7]; 7268 7269 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7270 "upper_%s" : "lower_%s", name); 7271 sysfs_remove_link(&(dev->dev.kobj), linkname); 7272 } 7273 7274 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, 7275 struct net_device *adj_dev, 7276 struct list_head *dev_list) 7277 { 7278 return (dev_list == &dev->adj_list.upper || 7279 dev_list == &dev->adj_list.lower) && 7280 net_eq(dev_net(dev), dev_net(adj_dev)); 7281 } 7282 7283 static int __netdev_adjacent_dev_insert(struct net_device *dev, 7284 struct net_device *adj_dev, 7285 struct list_head *dev_list, 7286 void *private, bool master) 7287 { 7288 struct netdev_adjacent *adj; 7289 int ret; 7290 7291 adj = __netdev_find_adj(adj_dev, dev_list); 7292 7293 if (adj) { 7294 adj->ref_nr += 1; 7295 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", 7296 dev->name, adj_dev->name, adj->ref_nr); 7297 7298 return 0; 7299 } 7300 7301 adj = kmalloc(sizeof(*adj), GFP_KERNEL); 7302 if (!adj) 7303 return -ENOMEM; 7304 7305 adj->dev = adj_dev; 7306 adj->master = master; 7307 adj->ref_nr = 1; 7308 adj->private = private; 7309 adj->ignore = false; 7310 dev_hold_track(adj_dev, &adj->dev_tracker, GFP_KERNEL); 7311 7312 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", 7313 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); 7314 7315 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { 7316 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); 7317 if (ret) 7318 goto free_adj; 7319 } 7320 7321 /* Ensure that master link is always the first item in list. */ 7322 if (master) { 7323 ret = sysfs_create_link(&(dev->dev.kobj), 7324 &(adj_dev->dev.kobj), "master"); 7325 if (ret) 7326 goto remove_symlinks; 7327 7328 list_add_rcu(&adj->list, dev_list); 7329 } else { 7330 list_add_tail_rcu(&adj->list, dev_list); 7331 } 7332 7333 return 0; 7334 7335 remove_symlinks: 7336 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7337 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7338 free_adj: 7339 dev_put_track(adj_dev, &adj->dev_tracker); 7340 kfree(adj); 7341 7342 return ret; 7343 } 7344 7345 static void __netdev_adjacent_dev_remove(struct net_device *dev, 7346 struct net_device *adj_dev, 7347 u16 ref_nr, 7348 struct list_head *dev_list) 7349 { 7350 struct netdev_adjacent *adj; 7351 7352 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", 7353 dev->name, adj_dev->name, ref_nr); 7354 7355 adj = __netdev_find_adj(adj_dev, dev_list); 7356 7357 if (!adj) { 7358 pr_err("Adjacency does not exist for device %s from %s\n", 7359 dev->name, adj_dev->name); 7360 WARN_ON(1); 7361 return; 7362 } 7363 7364 if (adj->ref_nr > ref_nr) { 7365 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", 7366 dev->name, adj_dev->name, ref_nr, 7367 adj->ref_nr - ref_nr); 7368 adj->ref_nr -= ref_nr; 7369 return; 7370 } 7371 7372 if (adj->master) 7373 sysfs_remove_link(&(dev->dev.kobj), "master"); 7374 7375 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7376 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7377 7378 list_del_rcu(&adj->list); 7379 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", 7380 adj_dev->name, dev->name, adj_dev->name); 7381 dev_put_track(adj_dev, &adj->dev_tracker); 7382 kfree_rcu(adj, rcu); 7383 } 7384 7385 static int __netdev_adjacent_dev_link_lists(struct net_device *dev, 7386 struct net_device *upper_dev, 7387 struct list_head *up_list, 7388 struct list_head *down_list, 7389 void *private, bool master) 7390 { 7391 int ret; 7392 7393 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, 7394 private, master); 7395 if (ret) 7396 return ret; 7397 7398 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, 7399 private, false); 7400 if (ret) { 7401 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); 7402 return ret; 7403 } 7404 7405 return 0; 7406 } 7407 7408 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, 7409 struct net_device *upper_dev, 7410 u16 ref_nr, 7411 struct list_head *up_list, 7412 struct list_head *down_list) 7413 { 7414 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); 7415 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); 7416 } 7417 7418 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, 7419 struct net_device *upper_dev, 7420 void *private, bool master) 7421 { 7422 return __netdev_adjacent_dev_link_lists(dev, upper_dev, 7423 &dev->adj_list.upper, 7424 &upper_dev->adj_list.lower, 7425 private, master); 7426 } 7427 7428 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, 7429 struct net_device *upper_dev) 7430 { 7431 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, 7432 &dev->adj_list.upper, 7433 &upper_dev->adj_list.lower); 7434 } 7435 7436 static int __netdev_upper_dev_link(struct net_device *dev, 7437 struct net_device *upper_dev, bool master, 7438 void *upper_priv, void *upper_info, 7439 struct netdev_nested_priv *priv, 7440 struct netlink_ext_ack *extack) 7441 { 7442 struct netdev_notifier_changeupper_info changeupper_info = { 7443 .info = { 7444 .dev = dev, 7445 .extack = extack, 7446 }, 7447 .upper_dev = upper_dev, 7448 .master = master, 7449 .linking = true, 7450 .upper_info = upper_info, 7451 }; 7452 struct net_device *master_dev; 7453 int ret = 0; 7454 7455 ASSERT_RTNL(); 7456 7457 if (dev == upper_dev) 7458 return -EBUSY; 7459 7460 /* To prevent loops, check if dev is not upper device to upper_dev. */ 7461 if (__netdev_has_upper_dev(upper_dev, dev)) 7462 return -EBUSY; 7463 7464 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV) 7465 return -EMLINK; 7466 7467 if (!master) { 7468 if (__netdev_has_upper_dev(dev, upper_dev)) 7469 return -EEXIST; 7470 } else { 7471 master_dev = __netdev_master_upper_dev_get(dev); 7472 if (master_dev) 7473 return master_dev == upper_dev ? -EEXIST : -EBUSY; 7474 } 7475 7476 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7477 &changeupper_info.info); 7478 ret = notifier_to_errno(ret); 7479 if (ret) 7480 return ret; 7481 7482 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, 7483 master); 7484 if (ret) 7485 return ret; 7486 7487 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 7488 &changeupper_info.info); 7489 ret = notifier_to_errno(ret); 7490 if (ret) 7491 goto rollback; 7492 7493 __netdev_update_upper_level(dev, NULL); 7494 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 7495 7496 __netdev_update_lower_level(upper_dev, priv); 7497 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 7498 priv); 7499 7500 return 0; 7501 7502 rollback: 7503 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 7504 7505 return ret; 7506 } 7507 7508 /** 7509 * netdev_upper_dev_link - Add a link to the upper device 7510 * @dev: device 7511 * @upper_dev: new upper device 7512 * @extack: netlink extended ack 7513 * 7514 * Adds a link to device which is upper to this one. The caller must hold 7515 * the RTNL lock. On a failure a negative errno code is returned. 7516 * On success the reference counts are adjusted and the function 7517 * returns zero. 7518 */ 7519 int netdev_upper_dev_link(struct net_device *dev, 7520 struct net_device *upper_dev, 7521 struct netlink_ext_ack *extack) 7522 { 7523 struct netdev_nested_priv priv = { 7524 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7525 .data = NULL, 7526 }; 7527 7528 return __netdev_upper_dev_link(dev, upper_dev, false, 7529 NULL, NULL, &priv, extack); 7530 } 7531 EXPORT_SYMBOL(netdev_upper_dev_link); 7532 7533 /** 7534 * netdev_master_upper_dev_link - Add a master link to the upper device 7535 * @dev: device 7536 * @upper_dev: new upper device 7537 * @upper_priv: upper device private 7538 * @upper_info: upper info to be passed down via notifier 7539 * @extack: netlink extended ack 7540 * 7541 * Adds a link to device which is upper to this one. In this case, only 7542 * one master upper device can be linked, although other non-master devices 7543 * might be linked as well. The caller must hold the RTNL lock. 7544 * On a failure a negative errno code is returned. On success the reference 7545 * counts are adjusted and the function returns zero. 7546 */ 7547 int netdev_master_upper_dev_link(struct net_device *dev, 7548 struct net_device *upper_dev, 7549 void *upper_priv, void *upper_info, 7550 struct netlink_ext_ack *extack) 7551 { 7552 struct netdev_nested_priv priv = { 7553 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7554 .data = NULL, 7555 }; 7556 7557 return __netdev_upper_dev_link(dev, upper_dev, true, 7558 upper_priv, upper_info, &priv, extack); 7559 } 7560 EXPORT_SYMBOL(netdev_master_upper_dev_link); 7561 7562 static void __netdev_upper_dev_unlink(struct net_device *dev, 7563 struct net_device *upper_dev, 7564 struct netdev_nested_priv *priv) 7565 { 7566 struct netdev_notifier_changeupper_info changeupper_info = { 7567 .info = { 7568 .dev = dev, 7569 }, 7570 .upper_dev = upper_dev, 7571 .linking = false, 7572 }; 7573 7574 ASSERT_RTNL(); 7575 7576 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; 7577 7578 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7579 &changeupper_info.info); 7580 7581 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 7582 7583 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 7584 &changeupper_info.info); 7585 7586 __netdev_update_upper_level(dev, NULL); 7587 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 7588 7589 __netdev_update_lower_level(upper_dev, priv); 7590 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 7591 priv); 7592 } 7593 7594 /** 7595 * netdev_upper_dev_unlink - Removes a link to upper device 7596 * @dev: device 7597 * @upper_dev: new upper device 7598 * 7599 * Removes a link to device which is upper to this one. The caller must hold 7600 * the RTNL lock. 7601 */ 7602 void netdev_upper_dev_unlink(struct net_device *dev, 7603 struct net_device *upper_dev) 7604 { 7605 struct netdev_nested_priv priv = { 7606 .flags = NESTED_SYNC_TODO, 7607 .data = NULL, 7608 }; 7609 7610 __netdev_upper_dev_unlink(dev, upper_dev, &priv); 7611 } 7612 EXPORT_SYMBOL(netdev_upper_dev_unlink); 7613 7614 static void __netdev_adjacent_dev_set(struct net_device *upper_dev, 7615 struct net_device *lower_dev, 7616 bool val) 7617 { 7618 struct netdev_adjacent *adj; 7619 7620 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower); 7621 if (adj) 7622 adj->ignore = val; 7623 7624 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper); 7625 if (adj) 7626 adj->ignore = val; 7627 } 7628 7629 static void netdev_adjacent_dev_disable(struct net_device *upper_dev, 7630 struct net_device *lower_dev) 7631 { 7632 __netdev_adjacent_dev_set(upper_dev, lower_dev, true); 7633 } 7634 7635 static void netdev_adjacent_dev_enable(struct net_device *upper_dev, 7636 struct net_device *lower_dev) 7637 { 7638 __netdev_adjacent_dev_set(upper_dev, lower_dev, false); 7639 } 7640 7641 int netdev_adjacent_change_prepare(struct net_device *old_dev, 7642 struct net_device *new_dev, 7643 struct net_device *dev, 7644 struct netlink_ext_ack *extack) 7645 { 7646 struct netdev_nested_priv priv = { 7647 .flags = 0, 7648 .data = NULL, 7649 }; 7650 int err; 7651 7652 if (!new_dev) 7653 return 0; 7654 7655 if (old_dev && new_dev != old_dev) 7656 netdev_adjacent_dev_disable(dev, old_dev); 7657 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv, 7658 extack); 7659 if (err) { 7660 if (old_dev && new_dev != old_dev) 7661 netdev_adjacent_dev_enable(dev, old_dev); 7662 return err; 7663 } 7664 7665 return 0; 7666 } 7667 EXPORT_SYMBOL(netdev_adjacent_change_prepare); 7668 7669 void netdev_adjacent_change_commit(struct net_device *old_dev, 7670 struct net_device *new_dev, 7671 struct net_device *dev) 7672 { 7673 struct netdev_nested_priv priv = { 7674 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7675 .data = NULL, 7676 }; 7677 7678 if (!new_dev || !old_dev) 7679 return; 7680 7681 if (new_dev == old_dev) 7682 return; 7683 7684 netdev_adjacent_dev_enable(dev, old_dev); 7685 __netdev_upper_dev_unlink(old_dev, dev, &priv); 7686 } 7687 EXPORT_SYMBOL(netdev_adjacent_change_commit); 7688 7689 void netdev_adjacent_change_abort(struct net_device *old_dev, 7690 struct net_device *new_dev, 7691 struct net_device *dev) 7692 { 7693 struct netdev_nested_priv priv = { 7694 .flags = 0, 7695 .data = NULL, 7696 }; 7697 7698 if (!new_dev) 7699 return; 7700 7701 if (old_dev && new_dev != old_dev) 7702 netdev_adjacent_dev_enable(dev, old_dev); 7703 7704 __netdev_upper_dev_unlink(new_dev, dev, &priv); 7705 } 7706 EXPORT_SYMBOL(netdev_adjacent_change_abort); 7707 7708 /** 7709 * netdev_bonding_info_change - Dispatch event about slave change 7710 * @dev: device 7711 * @bonding_info: info to dispatch 7712 * 7713 * Send NETDEV_BONDING_INFO to netdev notifiers with info. 7714 * The caller must hold the RTNL lock. 7715 */ 7716 void netdev_bonding_info_change(struct net_device *dev, 7717 struct netdev_bonding_info *bonding_info) 7718 { 7719 struct netdev_notifier_bonding_info info = { 7720 .info.dev = dev, 7721 }; 7722 7723 memcpy(&info.bonding_info, bonding_info, 7724 sizeof(struct netdev_bonding_info)); 7725 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, 7726 &info.info); 7727 } 7728 EXPORT_SYMBOL(netdev_bonding_info_change); 7729 7730 /** 7731 * netdev_get_xmit_slave - Get the xmit slave of master device 7732 * @dev: device 7733 * @skb: The packet 7734 * @all_slaves: assume all the slaves are active 7735 * 7736 * The reference counters are not incremented so the caller must be 7737 * careful with locks. The caller must hold RCU lock. 7738 * %NULL is returned if no slave is found. 7739 */ 7740 7741 struct net_device *netdev_get_xmit_slave(struct net_device *dev, 7742 struct sk_buff *skb, 7743 bool all_slaves) 7744 { 7745 const struct net_device_ops *ops = dev->netdev_ops; 7746 7747 if (!ops->ndo_get_xmit_slave) 7748 return NULL; 7749 return ops->ndo_get_xmit_slave(dev, skb, all_slaves); 7750 } 7751 EXPORT_SYMBOL(netdev_get_xmit_slave); 7752 7753 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev, 7754 struct sock *sk) 7755 { 7756 const struct net_device_ops *ops = dev->netdev_ops; 7757 7758 if (!ops->ndo_sk_get_lower_dev) 7759 return NULL; 7760 return ops->ndo_sk_get_lower_dev(dev, sk); 7761 } 7762 7763 /** 7764 * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket 7765 * @dev: device 7766 * @sk: the socket 7767 * 7768 * %NULL is returned if no lower device is found. 7769 */ 7770 7771 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev, 7772 struct sock *sk) 7773 { 7774 struct net_device *lower; 7775 7776 lower = netdev_sk_get_lower_dev(dev, sk); 7777 while (lower) { 7778 dev = lower; 7779 lower = netdev_sk_get_lower_dev(dev, sk); 7780 } 7781 7782 return dev; 7783 } 7784 EXPORT_SYMBOL(netdev_sk_get_lowest_dev); 7785 7786 static void netdev_adjacent_add_links(struct net_device *dev) 7787 { 7788 struct netdev_adjacent *iter; 7789 7790 struct net *net = dev_net(dev); 7791 7792 list_for_each_entry(iter, &dev->adj_list.upper, list) { 7793 if (!net_eq(net, dev_net(iter->dev))) 7794 continue; 7795 netdev_adjacent_sysfs_add(iter->dev, dev, 7796 &iter->dev->adj_list.lower); 7797 netdev_adjacent_sysfs_add(dev, iter->dev, 7798 &dev->adj_list.upper); 7799 } 7800 7801 list_for_each_entry(iter, &dev->adj_list.lower, list) { 7802 if (!net_eq(net, dev_net(iter->dev))) 7803 continue; 7804 netdev_adjacent_sysfs_add(iter->dev, dev, 7805 &iter->dev->adj_list.upper); 7806 netdev_adjacent_sysfs_add(dev, iter->dev, 7807 &dev->adj_list.lower); 7808 } 7809 } 7810 7811 static void netdev_adjacent_del_links(struct net_device *dev) 7812 { 7813 struct netdev_adjacent *iter; 7814 7815 struct net *net = dev_net(dev); 7816 7817 list_for_each_entry(iter, &dev->adj_list.upper, list) { 7818 if (!net_eq(net, dev_net(iter->dev))) 7819 continue; 7820 netdev_adjacent_sysfs_del(iter->dev, dev->name, 7821 &iter->dev->adj_list.lower); 7822 netdev_adjacent_sysfs_del(dev, iter->dev->name, 7823 &dev->adj_list.upper); 7824 } 7825 7826 list_for_each_entry(iter, &dev->adj_list.lower, list) { 7827 if (!net_eq(net, dev_net(iter->dev))) 7828 continue; 7829 netdev_adjacent_sysfs_del(iter->dev, dev->name, 7830 &iter->dev->adj_list.upper); 7831 netdev_adjacent_sysfs_del(dev, iter->dev->name, 7832 &dev->adj_list.lower); 7833 } 7834 } 7835 7836 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) 7837 { 7838 struct netdev_adjacent *iter; 7839 7840 struct net *net = dev_net(dev); 7841 7842 list_for_each_entry(iter, &dev->adj_list.upper, list) { 7843 if (!net_eq(net, dev_net(iter->dev))) 7844 continue; 7845 netdev_adjacent_sysfs_del(iter->dev, oldname, 7846 &iter->dev->adj_list.lower); 7847 netdev_adjacent_sysfs_add(iter->dev, dev, 7848 &iter->dev->adj_list.lower); 7849 } 7850 7851 list_for_each_entry(iter, &dev->adj_list.lower, list) { 7852 if (!net_eq(net, dev_net(iter->dev))) 7853 continue; 7854 netdev_adjacent_sysfs_del(iter->dev, oldname, 7855 &iter->dev->adj_list.upper); 7856 netdev_adjacent_sysfs_add(iter->dev, dev, 7857 &iter->dev->adj_list.upper); 7858 } 7859 } 7860 7861 void *netdev_lower_dev_get_private(struct net_device *dev, 7862 struct net_device *lower_dev) 7863 { 7864 struct netdev_adjacent *lower; 7865 7866 if (!lower_dev) 7867 return NULL; 7868 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); 7869 if (!lower) 7870 return NULL; 7871 7872 return lower->private; 7873 } 7874 EXPORT_SYMBOL(netdev_lower_dev_get_private); 7875 7876 7877 /** 7878 * netdev_lower_state_changed - Dispatch event about lower device state change 7879 * @lower_dev: device 7880 * @lower_state_info: state to dispatch 7881 * 7882 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. 7883 * The caller must hold the RTNL lock. 7884 */ 7885 void netdev_lower_state_changed(struct net_device *lower_dev, 7886 void *lower_state_info) 7887 { 7888 struct netdev_notifier_changelowerstate_info changelowerstate_info = { 7889 .info.dev = lower_dev, 7890 }; 7891 7892 ASSERT_RTNL(); 7893 changelowerstate_info.lower_state_info = lower_state_info; 7894 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, 7895 &changelowerstate_info.info); 7896 } 7897 EXPORT_SYMBOL(netdev_lower_state_changed); 7898 7899 static void dev_change_rx_flags(struct net_device *dev, int flags) 7900 { 7901 const struct net_device_ops *ops = dev->netdev_ops; 7902 7903 if (ops->ndo_change_rx_flags) 7904 ops->ndo_change_rx_flags(dev, flags); 7905 } 7906 7907 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) 7908 { 7909 unsigned int old_flags = dev->flags; 7910 kuid_t uid; 7911 kgid_t gid; 7912 7913 ASSERT_RTNL(); 7914 7915 dev->flags |= IFF_PROMISC; 7916 dev->promiscuity += inc; 7917 if (dev->promiscuity == 0) { 7918 /* 7919 * Avoid overflow. 7920 * If inc causes overflow, untouch promisc and return error. 7921 */ 7922 if (inc < 0) 7923 dev->flags &= ~IFF_PROMISC; 7924 else { 7925 dev->promiscuity -= inc; 7926 netdev_warn(dev, "promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n"); 7927 return -EOVERFLOW; 7928 } 7929 } 7930 if (dev->flags != old_flags) { 7931 pr_info("device %s %s promiscuous mode\n", 7932 dev->name, 7933 dev->flags & IFF_PROMISC ? "entered" : "left"); 7934 if (audit_enabled) { 7935 current_uid_gid(&uid, &gid); 7936 audit_log(audit_context(), GFP_ATOMIC, 7937 AUDIT_ANOM_PROMISCUOUS, 7938 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", 7939 dev->name, (dev->flags & IFF_PROMISC), 7940 (old_flags & IFF_PROMISC), 7941 from_kuid(&init_user_ns, audit_get_loginuid(current)), 7942 from_kuid(&init_user_ns, uid), 7943 from_kgid(&init_user_ns, gid), 7944 audit_get_sessionid(current)); 7945 } 7946 7947 dev_change_rx_flags(dev, IFF_PROMISC); 7948 } 7949 if (notify) 7950 __dev_notify_flags(dev, old_flags, IFF_PROMISC); 7951 return 0; 7952 } 7953 7954 /** 7955 * dev_set_promiscuity - update promiscuity count on a device 7956 * @dev: device 7957 * @inc: modifier 7958 * 7959 * Add or remove promiscuity from a device. While the count in the device 7960 * remains above zero the interface remains promiscuous. Once it hits zero 7961 * the device reverts back to normal filtering operation. A negative inc 7962 * value is used to drop promiscuity on the device. 7963 * Return 0 if successful or a negative errno code on error. 7964 */ 7965 int dev_set_promiscuity(struct net_device *dev, int inc) 7966 { 7967 unsigned int old_flags = dev->flags; 7968 int err; 7969 7970 err = __dev_set_promiscuity(dev, inc, true); 7971 if (err < 0) 7972 return err; 7973 if (dev->flags != old_flags) 7974 dev_set_rx_mode(dev); 7975 return err; 7976 } 7977 EXPORT_SYMBOL(dev_set_promiscuity); 7978 7979 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) 7980 { 7981 unsigned int old_flags = dev->flags, old_gflags = dev->gflags; 7982 7983 ASSERT_RTNL(); 7984 7985 dev->flags |= IFF_ALLMULTI; 7986 dev->allmulti += inc; 7987 if (dev->allmulti == 0) { 7988 /* 7989 * Avoid overflow. 7990 * If inc causes overflow, untouch allmulti and return error. 7991 */ 7992 if (inc < 0) 7993 dev->flags &= ~IFF_ALLMULTI; 7994 else { 7995 dev->allmulti -= inc; 7996 netdev_warn(dev, "allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n"); 7997 return -EOVERFLOW; 7998 } 7999 } 8000 if (dev->flags ^ old_flags) { 8001 dev_change_rx_flags(dev, IFF_ALLMULTI); 8002 dev_set_rx_mode(dev); 8003 if (notify) 8004 __dev_notify_flags(dev, old_flags, 8005 dev->gflags ^ old_gflags); 8006 } 8007 return 0; 8008 } 8009 8010 /** 8011 * dev_set_allmulti - update allmulti count on a device 8012 * @dev: device 8013 * @inc: modifier 8014 * 8015 * Add or remove reception of all multicast frames to a device. While the 8016 * count in the device remains above zero the interface remains listening 8017 * to all interfaces. Once it hits zero the device reverts back to normal 8018 * filtering operation. A negative @inc value is used to drop the counter 8019 * when releasing a resource needing all multicasts. 8020 * Return 0 if successful or a negative errno code on error. 8021 */ 8022 8023 int dev_set_allmulti(struct net_device *dev, int inc) 8024 { 8025 return __dev_set_allmulti(dev, inc, true); 8026 } 8027 EXPORT_SYMBOL(dev_set_allmulti); 8028 8029 /* 8030 * Upload unicast and multicast address lists to device and 8031 * configure RX filtering. When the device doesn't support unicast 8032 * filtering it is put in promiscuous mode while unicast addresses 8033 * are present. 8034 */ 8035 void __dev_set_rx_mode(struct net_device *dev) 8036 { 8037 const struct net_device_ops *ops = dev->netdev_ops; 8038 8039 /* dev_open will call this function so the list will stay sane. */ 8040 if (!(dev->flags&IFF_UP)) 8041 return; 8042 8043 if (!netif_device_present(dev)) 8044 return; 8045 8046 if (!(dev->priv_flags & IFF_UNICAST_FLT)) { 8047 /* Unicast addresses changes may only happen under the rtnl, 8048 * therefore calling __dev_set_promiscuity here is safe. 8049 */ 8050 if (!netdev_uc_empty(dev) && !dev->uc_promisc) { 8051 __dev_set_promiscuity(dev, 1, false); 8052 dev->uc_promisc = true; 8053 } else if (netdev_uc_empty(dev) && dev->uc_promisc) { 8054 __dev_set_promiscuity(dev, -1, false); 8055 dev->uc_promisc = false; 8056 } 8057 } 8058 8059 if (ops->ndo_set_rx_mode) 8060 ops->ndo_set_rx_mode(dev); 8061 } 8062 8063 void dev_set_rx_mode(struct net_device *dev) 8064 { 8065 netif_addr_lock_bh(dev); 8066 __dev_set_rx_mode(dev); 8067 netif_addr_unlock_bh(dev); 8068 } 8069 8070 /** 8071 * dev_get_flags - get flags reported to userspace 8072 * @dev: device 8073 * 8074 * Get the combination of flag bits exported through APIs to userspace. 8075 */ 8076 unsigned int dev_get_flags(const struct net_device *dev) 8077 { 8078 unsigned int flags; 8079 8080 flags = (dev->flags & ~(IFF_PROMISC | 8081 IFF_ALLMULTI | 8082 IFF_RUNNING | 8083 IFF_LOWER_UP | 8084 IFF_DORMANT)) | 8085 (dev->gflags & (IFF_PROMISC | 8086 IFF_ALLMULTI)); 8087 8088 if (netif_running(dev)) { 8089 if (netif_oper_up(dev)) 8090 flags |= IFF_RUNNING; 8091 if (netif_carrier_ok(dev)) 8092 flags |= IFF_LOWER_UP; 8093 if (netif_dormant(dev)) 8094 flags |= IFF_DORMANT; 8095 } 8096 8097 return flags; 8098 } 8099 EXPORT_SYMBOL(dev_get_flags); 8100 8101 int __dev_change_flags(struct net_device *dev, unsigned int flags, 8102 struct netlink_ext_ack *extack) 8103 { 8104 unsigned int old_flags = dev->flags; 8105 int ret; 8106 8107 ASSERT_RTNL(); 8108 8109 /* 8110 * Set the flags on our device. 8111 */ 8112 8113 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | 8114 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | 8115 IFF_AUTOMEDIA)) | 8116 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | 8117 IFF_ALLMULTI)); 8118 8119 /* 8120 * Load in the correct multicast list now the flags have changed. 8121 */ 8122 8123 if ((old_flags ^ flags) & IFF_MULTICAST) 8124 dev_change_rx_flags(dev, IFF_MULTICAST); 8125 8126 dev_set_rx_mode(dev); 8127 8128 /* 8129 * Have we downed the interface. We handle IFF_UP ourselves 8130 * according to user attempts to set it, rather than blindly 8131 * setting it. 8132 */ 8133 8134 ret = 0; 8135 if ((old_flags ^ flags) & IFF_UP) { 8136 if (old_flags & IFF_UP) 8137 __dev_close(dev); 8138 else 8139 ret = __dev_open(dev, extack); 8140 } 8141 8142 if ((flags ^ dev->gflags) & IFF_PROMISC) { 8143 int inc = (flags & IFF_PROMISC) ? 1 : -1; 8144 unsigned int old_flags = dev->flags; 8145 8146 dev->gflags ^= IFF_PROMISC; 8147 8148 if (__dev_set_promiscuity(dev, inc, false) >= 0) 8149 if (dev->flags != old_flags) 8150 dev_set_rx_mode(dev); 8151 } 8152 8153 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI 8154 * is important. Some (broken) drivers set IFF_PROMISC, when 8155 * IFF_ALLMULTI is requested not asking us and not reporting. 8156 */ 8157 if ((flags ^ dev->gflags) & IFF_ALLMULTI) { 8158 int inc = (flags & IFF_ALLMULTI) ? 1 : -1; 8159 8160 dev->gflags ^= IFF_ALLMULTI; 8161 __dev_set_allmulti(dev, inc, false); 8162 } 8163 8164 return ret; 8165 } 8166 8167 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, 8168 unsigned int gchanges) 8169 { 8170 unsigned int changes = dev->flags ^ old_flags; 8171 8172 if (gchanges) 8173 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC); 8174 8175 if (changes & IFF_UP) { 8176 if (dev->flags & IFF_UP) 8177 call_netdevice_notifiers(NETDEV_UP, dev); 8178 else 8179 call_netdevice_notifiers(NETDEV_DOWN, dev); 8180 } 8181 8182 if (dev->flags & IFF_UP && 8183 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { 8184 struct netdev_notifier_change_info change_info = { 8185 .info = { 8186 .dev = dev, 8187 }, 8188 .flags_changed = changes, 8189 }; 8190 8191 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info); 8192 } 8193 } 8194 8195 /** 8196 * dev_change_flags - change device settings 8197 * @dev: device 8198 * @flags: device state flags 8199 * @extack: netlink extended ack 8200 * 8201 * Change settings on device based state flags. The flags are 8202 * in the userspace exported format. 8203 */ 8204 int dev_change_flags(struct net_device *dev, unsigned int flags, 8205 struct netlink_ext_ack *extack) 8206 { 8207 int ret; 8208 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; 8209 8210 ret = __dev_change_flags(dev, flags, extack); 8211 if (ret < 0) 8212 return ret; 8213 8214 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); 8215 __dev_notify_flags(dev, old_flags, changes); 8216 return ret; 8217 } 8218 EXPORT_SYMBOL(dev_change_flags); 8219 8220 int __dev_set_mtu(struct net_device *dev, int new_mtu) 8221 { 8222 const struct net_device_ops *ops = dev->netdev_ops; 8223 8224 if (ops->ndo_change_mtu) 8225 return ops->ndo_change_mtu(dev, new_mtu); 8226 8227 /* Pairs with all the lockless reads of dev->mtu in the stack */ 8228 WRITE_ONCE(dev->mtu, new_mtu); 8229 return 0; 8230 } 8231 EXPORT_SYMBOL(__dev_set_mtu); 8232 8233 int dev_validate_mtu(struct net_device *dev, int new_mtu, 8234 struct netlink_ext_ack *extack) 8235 { 8236 /* MTU must be positive, and in range */ 8237 if (new_mtu < 0 || new_mtu < dev->min_mtu) { 8238 NL_SET_ERR_MSG(extack, "mtu less than device minimum"); 8239 return -EINVAL; 8240 } 8241 8242 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { 8243 NL_SET_ERR_MSG(extack, "mtu greater than device maximum"); 8244 return -EINVAL; 8245 } 8246 return 0; 8247 } 8248 8249 /** 8250 * dev_set_mtu_ext - Change maximum transfer unit 8251 * @dev: device 8252 * @new_mtu: new transfer unit 8253 * @extack: netlink extended ack 8254 * 8255 * Change the maximum transfer size of the network device. 8256 */ 8257 int dev_set_mtu_ext(struct net_device *dev, int new_mtu, 8258 struct netlink_ext_ack *extack) 8259 { 8260 int err, orig_mtu; 8261 8262 if (new_mtu == dev->mtu) 8263 return 0; 8264 8265 err = dev_validate_mtu(dev, new_mtu, extack); 8266 if (err) 8267 return err; 8268 8269 if (!netif_device_present(dev)) 8270 return -ENODEV; 8271 8272 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); 8273 err = notifier_to_errno(err); 8274 if (err) 8275 return err; 8276 8277 orig_mtu = dev->mtu; 8278 err = __dev_set_mtu(dev, new_mtu); 8279 8280 if (!err) { 8281 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8282 orig_mtu); 8283 err = notifier_to_errno(err); 8284 if (err) { 8285 /* setting mtu back and notifying everyone again, 8286 * so that they have a chance to revert changes. 8287 */ 8288 __dev_set_mtu(dev, orig_mtu); 8289 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8290 new_mtu); 8291 } 8292 } 8293 return err; 8294 } 8295 8296 int dev_set_mtu(struct net_device *dev, int new_mtu) 8297 { 8298 struct netlink_ext_ack extack; 8299 int err; 8300 8301 memset(&extack, 0, sizeof(extack)); 8302 err = dev_set_mtu_ext(dev, new_mtu, &extack); 8303 if (err && extack._msg) 8304 net_err_ratelimited("%s: %s\n", dev->name, extack._msg); 8305 return err; 8306 } 8307 EXPORT_SYMBOL(dev_set_mtu); 8308 8309 /** 8310 * dev_change_tx_queue_len - Change TX queue length of a netdevice 8311 * @dev: device 8312 * @new_len: new tx queue length 8313 */ 8314 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len) 8315 { 8316 unsigned int orig_len = dev->tx_queue_len; 8317 int res; 8318 8319 if (new_len != (unsigned int)new_len) 8320 return -ERANGE; 8321 8322 if (new_len != orig_len) { 8323 dev->tx_queue_len = new_len; 8324 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev); 8325 res = notifier_to_errno(res); 8326 if (res) 8327 goto err_rollback; 8328 res = dev_qdisc_change_tx_queue_len(dev); 8329 if (res) 8330 goto err_rollback; 8331 } 8332 8333 return 0; 8334 8335 err_rollback: 8336 netdev_err(dev, "refused to change device tx_queue_len\n"); 8337 dev->tx_queue_len = orig_len; 8338 return res; 8339 } 8340 8341 /** 8342 * dev_set_group - Change group this device belongs to 8343 * @dev: device 8344 * @new_group: group this device should belong to 8345 */ 8346 void dev_set_group(struct net_device *dev, int new_group) 8347 { 8348 dev->group = new_group; 8349 } 8350 EXPORT_SYMBOL(dev_set_group); 8351 8352 /** 8353 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR. 8354 * @dev: device 8355 * @addr: new address 8356 * @extack: netlink extended ack 8357 */ 8358 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr, 8359 struct netlink_ext_ack *extack) 8360 { 8361 struct netdev_notifier_pre_changeaddr_info info = { 8362 .info.dev = dev, 8363 .info.extack = extack, 8364 .dev_addr = addr, 8365 }; 8366 int rc; 8367 8368 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info); 8369 return notifier_to_errno(rc); 8370 } 8371 EXPORT_SYMBOL(dev_pre_changeaddr_notify); 8372 8373 /** 8374 * dev_set_mac_address - Change Media Access Control Address 8375 * @dev: device 8376 * @sa: new address 8377 * @extack: netlink extended ack 8378 * 8379 * Change the hardware (MAC) address of the device 8380 */ 8381 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa, 8382 struct netlink_ext_ack *extack) 8383 { 8384 const struct net_device_ops *ops = dev->netdev_ops; 8385 int err; 8386 8387 if (!ops->ndo_set_mac_address) 8388 return -EOPNOTSUPP; 8389 if (sa->sa_family != dev->type) 8390 return -EINVAL; 8391 if (!netif_device_present(dev)) 8392 return -ENODEV; 8393 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack); 8394 if (err) 8395 return err; 8396 err = ops->ndo_set_mac_address(dev, sa); 8397 if (err) 8398 return err; 8399 dev->addr_assign_type = NET_ADDR_SET; 8400 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); 8401 add_device_randomness(dev->dev_addr, dev->addr_len); 8402 return 0; 8403 } 8404 EXPORT_SYMBOL(dev_set_mac_address); 8405 8406 static DECLARE_RWSEM(dev_addr_sem); 8407 8408 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa, 8409 struct netlink_ext_ack *extack) 8410 { 8411 int ret; 8412 8413 down_write(&dev_addr_sem); 8414 ret = dev_set_mac_address(dev, sa, extack); 8415 up_write(&dev_addr_sem); 8416 return ret; 8417 } 8418 EXPORT_SYMBOL(dev_set_mac_address_user); 8419 8420 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name) 8421 { 8422 size_t size = sizeof(sa->sa_data); 8423 struct net_device *dev; 8424 int ret = 0; 8425 8426 down_read(&dev_addr_sem); 8427 rcu_read_lock(); 8428 8429 dev = dev_get_by_name_rcu(net, dev_name); 8430 if (!dev) { 8431 ret = -ENODEV; 8432 goto unlock; 8433 } 8434 if (!dev->addr_len) 8435 memset(sa->sa_data, 0, size); 8436 else 8437 memcpy(sa->sa_data, dev->dev_addr, 8438 min_t(size_t, size, dev->addr_len)); 8439 sa->sa_family = dev->type; 8440 8441 unlock: 8442 rcu_read_unlock(); 8443 up_read(&dev_addr_sem); 8444 return ret; 8445 } 8446 EXPORT_SYMBOL(dev_get_mac_address); 8447 8448 /** 8449 * dev_change_carrier - Change device carrier 8450 * @dev: device 8451 * @new_carrier: new value 8452 * 8453 * Change device carrier 8454 */ 8455 int dev_change_carrier(struct net_device *dev, bool new_carrier) 8456 { 8457 const struct net_device_ops *ops = dev->netdev_ops; 8458 8459 if (!ops->ndo_change_carrier) 8460 return -EOPNOTSUPP; 8461 if (!netif_device_present(dev)) 8462 return -ENODEV; 8463 return ops->ndo_change_carrier(dev, new_carrier); 8464 } 8465 EXPORT_SYMBOL(dev_change_carrier); 8466 8467 /** 8468 * dev_get_phys_port_id - Get device physical port ID 8469 * @dev: device 8470 * @ppid: port ID 8471 * 8472 * Get device physical port ID 8473 */ 8474 int dev_get_phys_port_id(struct net_device *dev, 8475 struct netdev_phys_item_id *ppid) 8476 { 8477 const struct net_device_ops *ops = dev->netdev_ops; 8478 8479 if (!ops->ndo_get_phys_port_id) 8480 return -EOPNOTSUPP; 8481 return ops->ndo_get_phys_port_id(dev, ppid); 8482 } 8483 EXPORT_SYMBOL(dev_get_phys_port_id); 8484 8485 /** 8486 * dev_get_phys_port_name - Get device physical port name 8487 * @dev: device 8488 * @name: port name 8489 * @len: limit of bytes to copy to name 8490 * 8491 * Get device physical port name 8492 */ 8493 int dev_get_phys_port_name(struct net_device *dev, 8494 char *name, size_t len) 8495 { 8496 const struct net_device_ops *ops = dev->netdev_ops; 8497 int err; 8498 8499 if (ops->ndo_get_phys_port_name) { 8500 err = ops->ndo_get_phys_port_name(dev, name, len); 8501 if (err != -EOPNOTSUPP) 8502 return err; 8503 } 8504 return devlink_compat_phys_port_name_get(dev, name, len); 8505 } 8506 EXPORT_SYMBOL(dev_get_phys_port_name); 8507 8508 /** 8509 * dev_get_port_parent_id - Get the device's port parent identifier 8510 * @dev: network device 8511 * @ppid: pointer to a storage for the port's parent identifier 8512 * @recurse: allow/disallow recursion to lower devices 8513 * 8514 * Get the devices's port parent identifier 8515 */ 8516 int dev_get_port_parent_id(struct net_device *dev, 8517 struct netdev_phys_item_id *ppid, 8518 bool recurse) 8519 { 8520 const struct net_device_ops *ops = dev->netdev_ops; 8521 struct netdev_phys_item_id first = { }; 8522 struct net_device *lower_dev; 8523 struct list_head *iter; 8524 int err; 8525 8526 if (ops->ndo_get_port_parent_id) { 8527 err = ops->ndo_get_port_parent_id(dev, ppid); 8528 if (err != -EOPNOTSUPP) 8529 return err; 8530 } 8531 8532 err = devlink_compat_switch_id_get(dev, ppid); 8533 if (!recurse || err != -EOPNOTSUPP) 8534 return err; 8535 8536 netdev_for_each_lower_dev(dev, lower_dev, iter) { 8537 err = dev_get_port_parent_id(lower_dev, ppid, true); 8538 if (err) 8539 break; 8540 if (!first.id_len) 8541 first = *ppid; 8542 else if (memcmp(&first, ppid, sizeof(*ppid))) 8543 return -EOPNOTSUPP; 8544 } 8545 8546 return err; 8547 } 8548 EXPORT_SYMBOL(dev_get_port_parent_id); 8549 8550 /** 8551 * netdev_port_same_parent_id - Indicate if two network devices have 8552 * the same port parent identifier 8553 * @a: first network device 8554 * @b: second network device 8555 */ 8556 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b) 8557 { 8558 struct netdev_phys_item_id a_id = { }; 8559 struct netdev_phys_item_id b_id = { }; 8560 8561 if (dev_get_port_parent_id(a, &a_id, true) || 8562 dev_get_port_parent_id(b, &b_id, true)) 8563 return false; 8564 8565 return netdev_phys_item_id_same(&a_id, &b_id); 8566 } 8567 EXPORT_SYMBOL(netdev_port_same_parent_id); 8568 8569 /** 8570 * dev_change_proto_down - set carrier according to proto_down. 8571 * 8572 * @dev: device 8573 * @proto_down: new value 8574 */ 8575 int dev_change_proto_down(struct net_device *dev, bool proto_down) 8576 { 8577 if (!(dev->priv_flags & IFF_CHANGE_PROTO_DOWN)) 8578 return -EOPNOTSUPP; 8579 if (!netif_device_present(dev)) 8580 return -ENODEV; 8581 if (proto_down) 8582 netif_carrier_off(dev); 8583 else 8584 netif_carrier_on(dev); 8585 dev->proto_down = proto_down; 8586 return 0; 8587 } 8588 EXPORT_SYMBOL(dev_change_proto_down); 8589 8590 /** 8591 * dev_change_proto_down_reason - proto down reason 8592 * 8593 * @dev: device 8594 * @mask: proto down mask 8595 * @value: proto down value 8596 */ 8597 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask, 8598 u32 value) 8599 { 8600 int b; 8601 8602 if (!mask) { 8603 dev->proto_down_reason = value; 8604 } else { 8605 for_each_set_bit(b, &mask, 32) { 8606 if (value & (1 << b)) 8607 dev->proto_down_reason |= BIT(b); 8608 else 8609 dev->proto_down_reason &= ~BIT(b); 8610 } 8611 } 8612 } 8613 EXPORT_SYMBOL(dev_change_proto_down_reason); 8614 8615 struct bpf_xdp_link { 8616 struct bpf_link link; 8617 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */ 8618 int flags; 8619 }; 8620 8621 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags) 8622 { 8623 if (flags & XDP_FLAGS_HW_MODE) 8624 return XDP_MODE_HW; 8625 if (flags & XDP_FLAGS_DRV_MODE) 8626 return XDP_MODE_DRV; 8627 if (flags & XDP_FLAGS_SKB_MODE) 8628 return XDP_MODE_SKB; 8629 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB; 8630 } 8631 8632 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode) 8633 { 8634 switch (mode) { 8635 case XDP_MODE_SKB: 8636 return generic_xdp_install; 8637 case XDP_MODE_DRV: 8638 case XDP_MODE_HW: 8639 return dev->netdev_ops->ndo_bpf; 8640 default: 8641 return NULL; 8642 } 8643 } 8644 8645 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev, 8646 enum bpf_xdp_mode mode) 8647 { 8648 return dev->xdp_state[mode].link; 8649 } 8650 8651 static struct bpf_prog *dev_xdp_prog(struct net_device *dev, 8652 enum bpf_xdp_mode mode) 8653 { 8654 struct bpf_xdp_link *link = dev_xdp_link(dev, mode); 8655 8656 if (link) 8657 return link->link.prog; 8658 return dev->xdp_state[mode].prog; 8659 } 8660 8661 u8 dev_xdp_prog_count(struct net_device *dev) 8662 { 8663 u8 count = 0; 8664 int i; 8665 8666 for (i = 0; i < __MAX_XDP_MODE; i++) 8667 if (dev->xdp_state[i].prog || dev->xdp_state[i].link) 8668 count++; 8669 return count; 8670 } 8671 EXPORT_SYMBOL_GPL(dev_xdp_prog_count); 8672 8673 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode) 8674 { 8675 struct bpf_prog *prog = dev_xdp_prog(dev, mode); 8676 8677 return prog ? prog->aux->id : 0; 8678 } 8679 8680 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode, 8681 struct bpf_xdp_link *link) 8682 { 8683 dev->xdp_state[mode].link = link; 8684 dev->xdp_state[mode].prog = NULL; 8685 } 8686 8687 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode, 8688 struct bpf_prog *prog) 8689 { 8690 dev->xdp_state[mode].link = NULL; 8691 dev->xdp_state[mode].prog = prog; 8692 } 8693 8694 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode, 8695 bpf_op_t bpf_op, struct netlink_ext_ack *extack, 8696 u32 flags, struct bpf_prog *prog) 8697 { 8698 struct netdev_bpf xdp; 8699 int err; 8700 8701 memset(&xdp, 0, sizeof(xdp)); 8702 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG; 8703 xdp.extack = extack; 8704 xdp.flags = flags; 8705 xdp.prog = prog; 8706 8707 /* Drivers assume refcnt is already incremented (i.e, prog pointer is 8708 * "moved" into driver), so they don't increment it on their own, but 8709 * they do decrement refcnt when program is detached or replaced. 8710 * Given net_device also owns link/prog, we need to bump refcnt here 8711 * to prevent drivers from underflowing it. 8712 */ 8713 if (prog) 8714 bpf_prog_inc(prog); 8715 err = bpf_op(dev, &xdp); 8716 if (err) { 8717 if (prog) 8718 bpf_prog_put(prog); 8719 return err; 8720 } 8721 8722 if (mode != XDP_MODE_HW) 8723 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog); 8724 8725 return 0; 8726 } 8727 8728 static void dev_xdp_uninstall(struct net_device *dev) 8729 { 8730 struct bpf_xdp_link *link; 8731 struct bpf_prog *prog; 8732 enum bpf_xdp_mode mode; 8733 bpf_op_t bpf_op; 8734 8735 ASSERT_RTNL(); 8736 8737 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) { 8738 prog = dev_xdp_prog(dev, mode); 8739 if (!prog) 8740 continue; 8741 8742 bpf_op = dev_xdp_bpf_op(dev, mode); 8743 if (!bpf_op) 8744 continue; 8745 8746 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 8747 8748 /* auto-detach link from net device */ 8749 link = dev_xdp_link(dev, mode); 8750 if (link) 8751 link->dev = NULL; 8752 else 8753 bpf_prog_put(prog); 8754 8755 dev_xdp_set_link(dev, mode, NULL); 8756 } 8757 } 8758 8759 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack, 8760 struct bpf_xdp_link *link, struct bpf_prog *new_prog, 8761 struct bpf_prog *old_prog, u32 flags) 8762 { 8763 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES); 8764 struct bpf_prog *cur_prog; 8765 struct net_device *upper; 8766 struct list_head *iter; 8767 enum bpf_xdp_mode mode; 8768 bpf_op_t bpf_op; 8769 int err; 8770 8771 ASSERT_RTNL(); 8772 8773 /* either link or prog attachment, never both */ 8774 if (link && (new_prog || old_prog)) 8775 return -EINVAL; 8776 /* link supports only XDP mode flags */ 8777 if (link && (flags & ~XDP_FLAGS_MODES)) { 8778 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment"); 8779 return -EINVAL; 8780 } 8781 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */ 8782 if (num_modes > 1) { 8783 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set"); 8784 return -EINVAL; 8785 } 8786 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */ 8787 if (!num_modes && dev_xdp_prog_count(dev) > 1) { 8788 NL_SET_ERR_MSG(extack, 8789 "More than one program loaded, unset mode is ambiguous"); 8790 return -EINVAL; 8791 } 8792 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */ 8793 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) { 8794 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified"); 8795 return -EINVAL; 8796 } 8797 8798 mode = dev_xdp_mode(dev, flags); 8799 /* can't replace attached link */ 8800 if (dev_xdp_link(dev, mode)) { 8801 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link"); 8802 return -EBUSY; 8803 } 8804 8805 /* don't allow if an upper device already has a program */ 8806 netdev_for_each_upper_dev_rcu(dev, upper, iter) { 8807 if (dev_xdp_prog_count(upper) > 0) { 8808 NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program"); 8809 return -EEXIST; 8810 } 8811 } 8812 8813 cur_prog = dev_xdp_prog(dev, mode); 8814 /* can't replace attached prog with link */ 8815 if (link && cur_prog) { 8816 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link"); 8817 return -EBUSY; 8818 } 8819 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) { 8820 NL_SET_ERR_MSG(extack, "Active program does not match expected"); 8821 return -EEXIST; 8822 } 8823 8824 /* put effective new program into new_prog */ 8825 if (link) 8826 new_prog = link->link.prog; 8827 8828 if (new_prog) { 8829 bool offload = mode == XDP_MODE_HW; 8830 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB 8831 ? XDP_MODE_DRV : XDP_MODE_SKB; 8832 8833 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) { 8834 NL_SET_ERR_MSG(extack, "XDP program already attached"); 8835 return -EBUSY; 8836 } 8837 if (!offload && dev_xdp_prog(dev, other_mode)) { 8838 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time"); 8839 return -EEXIST; 8840 } 8841 if (!offload && bpf_prog_is_dev_bound(new_prog->aux)) { 8842 NL_SET_ERR_MSG(extack, "Using device-bound program without HW_MODE flag is not supported"); 8843 return -EINVAL; 8844 } 8845 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) { 8846 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device"); 8847 return -EINVAL; 8848 } 8849 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) { 8850 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device"); 8851 return -EINVAL; 8852 } 8853 } 8854 8855 /* don't call drivers if the effective program didn't change */ 8856 if (new_prog != cur_prog) { 8857 bpf_op = dev_xdp_bpf_op(dev, mode); 8858 if (!bpf_op) { 8859 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode"); 8860 return -EOPNOTSUPP; 8861 } 8862 8863 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog); 8864 if (err) 8865 return err; 8866 } 8867 8868 if (link) 8869 dev_xdp_set_link(dev, mode, link); 8870 else 8871 dev_xdp_set_prog(dev, mode, new_prog); 8872 if (cur_prog) 8873 bpf_prog_put(cur_prog); 8874 8875 return 0; 8876 } 8877 8878 static int dev_xdp_attach_link(struct net_device *dev, 8879 struct netlink_ext_ack *extack, 8880 struct bpf_xdp_link *link) 8881 { 8882 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags); 8883 } 8884 8885 static int dev_xdp_detach_link(struct net_device *dev, 8886 struct netlink_ext_ack *extack, 8887 struct bpf_xdp_link *link) 8888 { 8889 enum bpf_xdp_mode mode; 8890 bpf_op_t bpf_op; 8891 8892 ASSERT_RTNL(); 8893 8894 mode = dev_xdp_mode(dev, link->flags); 8895 if (dev_xdp_link(dev, mode) != link) 8896 return -EINVAL; 8897 8898 bpf_op = dev_xdp_bpf_op(dev, mode); 8899 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 8900 dev_xdp_set_link(dev, mode, NULL); 8901 return 0; 8902 } 8903 8904 static void bpf_xdp_link_release(struct bpf_link *link) 8905 { 8906 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 8907 8908 rtnl_lock(); 8909 8910 /* if racing with net_device's tear down, xdp_link->dev might be 8911 * already NULL, in which case link was already auto-detached 8912 */ 8913 if (xdp_link->dev) { 8914 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link)); 8915 xdp_link->dev = NULL; 8916 } 8917 8918 rtnl_unlock(); 8919 } 8920 8921 static int bpf_xdp_link_detach(struct bpf_link *link) 8922 { 8923 bpf_xdp_link_release(link); 8924 return 0; 8925 } 8926 8927 static void bpf_xdp_link_dealloc(struct bpf_link *link) 8928 { 8929 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 8930 8931 kfree(xdp_link); 8932 } 8933 8934 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link, 8935 struct seq_file *seq) 8936 { 8937 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 8938 u32 ifindex = 0; 8939 8940 rtnl_lock(); 8941 if (xdp_link->dev) 8942 ifindex = xdp_link->dev->ifindex; 8943 rtnl_unlock(); 8944 8945 seq_printf(seq, "ifindex:\t%u\n", ifindex); 8946 } 8947 8948 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link, 8949 struct bpf_link_info *info) 8950 { 8951 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 8952 u32 ifindex = 0; 8953 8954 rtnl_lock(); 8955 if (xdp_link->dev) 8956 ifindex = xdp_link->dev->ifindex; 8957 rtnl_unlock(); 8958 8959 info->xdp.ifindex = ifindex; 8960 return 0; 8961 } 8962 8963 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog, 8964 struct bpf_prog *old_prog) 8965 { 8966 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 8967 enum bpf_xdp_mode mode; 8968 bpf_op_t bpf_op; 8969 int err = 0; 8970 8971 rtnl_lock(); 8972 8973 /* link might have been auto-released already, so fail */ 8974 if (!xdp_link->dev) { 8975 err = -ENOLINK; 8976 goto out_unlock; 8977 } 8978 8979 if (old_prog && link->prog != old_prog) { 8980 err = -EPERM; 8981 goto out_unlock; 8982 } 8983 old_prog = link->prog; 8984 if (old_prog->type != new_prog->type || 8985 old_prog->expected_attach_type != new_prog->expected_attach_type) { 8986 err = -EINVAL; 8987 goto out_unlock; 8988 } 8989 8990 if (old_prog == new_prog) { 8991 /* no-op, don't disturb drivers */ 8992 bpf_prog_put(new_prog); 8993 goto out_unlock; 8994 } 8995 8996 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags); 8997 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode); 8998 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL, 8999 xdp_link->flags, new_prog); 9000 if (err) 9001 goto out_unlock; 9002 9003 old_prog = xchg(&link->prog, new_prog); 9004 bpf_prog_put(old_prog); 9005 9006 out_unlock: 9007 rtnl_unlock(); 9008 return err; 9009 } 9010 9011 static const struct bpf_link_ops bpf_xdp_link_lops = { 9012 .release = bpf_xdp_link_release, 9013 .dealloc = bpf_xdp_link_dealloc, 9014 .detach = bpf_xdp_link_detach, 9015 .show_fdinfo = bpf_xdp_link_show_fdinfo, 9016 .fill_link_info = bpf_xdp_link_fill_link_info, 9017 .update_prog = bpf_xdp_link_update, 9018 }; 9019 9020 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog) 9021 { 9022 struct net *net = current->nsproxy->net_ns; 9023 struct bpf_link_primer link_primer; 9024 struct bpf_xdp_link *link; 9025 struct net_device *dev; 9026 int err, fd; 9027 9028 rtnl_lock(); 9029 dev = dev_get_by_index(net, attr->link_create.target_ifindex); 9030 if (!dev) { 9031 rtnl_unlock(); 9032 return -EINVAL; 9033 } 9034 9035 link = kzalloc(sizeof(*link), GFP_USER); 9036 if (!link) { 9037 err = -ENOMEM; 9038 goto unlock; 9039 } 9040 9041 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog); 9042 link->dev = dev; 9043 link->flags = attr->link_create.flags; 9044 9045 err = bpf_link_prime(&link->link, &link_primer); 9046 if (err) { 9047 kfree(link); 9048 goto unlock; 9049 } 9050 9051 err = dev_xdp_attach_link(dev, NULL, link); 9052 rtnl_unlock(); 9053 9054 if (err) { 9055 link->dev = NULL; 9056 bpf_link_cleanup(&link_primer); 9057 goto out_put_dev; 9058 } 9059 9060 fd = bpf_link_settle(&link_primer); 9061 /* link itself doesn't hold dev's refcnt to not complicate shutdown */ 9062 dev_put(dev); 9063 return fd; 9064 9065 unlock: 9066 rtnl_unlock(); 9067 9068 out_put_dev: 9069 dev_put(dev); 9070 return err; 9071 } 9072 9073 /** 9074 * dev_change_xdp_fd - set or clear a bpf program for a device rx path 9075 * @dev: device 9076 * @extack: netlink extended ack 9077 * @fd: new program fd or negative value to clear 9078 * @expected_fd: old program fd that userspace expects to replace or clear 9079 * @flags: xdp-related flags 9080 * 9081 * Set or clear a bpf program for a device 9082 */ 9083 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, 9084 int fd, int expected_fd, u32 flags) 9085 { 9086 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags); 9087 struct bpf_prog *new_prog = NULL, *old_prog = NULL; 9088 int err; 9089 9090 ASSERT_RTNL(); 9091 9092 if (fd >= 0) { 9093 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP, 9094 mode != XDP_MODE_SKB); 9095 if (IS_ERR(new_prog)) 9096 return PTR_ERR(new_prog); 9097 } 9098 9099 if (expected_fd >= 0) { 9100 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP, 9101 mode != XDP_MODE_SKB); 9102 if (IS_ERR(old_prog)) { 9103 err = PTR_ERR(old_prog); 9104 old_prog = NULL; 9105 goto err_out; 9106 } 9107 } 9108 9109 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags); 9110 9111 err_out: 9112 if (err && new_prog) 9113 bpf_prog_put(new_prog); 9114 if (old_prog) 9115 bpf_prog_put(old_prog); 9116 return err; 9117 } 9118 9119 /** 9120 * dev_new_index - allocate an ifindex 9121 * @net: the applicable net namespace 9122 * 9123 * Returns a suitable unique value for a new device interface 9124 * number. The caller must hold the rtnl semaphore or the 9125 * dev_base_lock to be sure it remains unique. 9126 */ 9127 static int dev_new_index(struct net *net) 9128 { 9129 int ifindex = net->ifindex; 9130 9131 for (;;) { 9132 if (++ifindex <= 0) 9133 ifindex = 1; 9134 if (!__dev_get_by_index(net, ifindex)) 9135 return net->ifindex = ifindex; 9136 } 9137 } 9138 9139 /* Delayed registration/unregisteration */ 9140 static LIST_HEAD(net_todo_list); 9141 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); 9142 9143 static void net_set_todo(struct net_device *dev) 9144 { 9145 list_add_tail(&dev->todo_list, &net_todo_list); 9146 dev_net(dev)->dev_unreg_count++; 9147 } 9148 9149 static netdev_features_t netdev_sync_upper_features(struct net_device *lower, 9150 struct net_device *upper, netdev_features_t features) 9151 { 9152 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9153 netdev_features_t feature; 9154 int feature_bit; 9155 9156 for_each_netdev_feature(upper_disables, feature_bit) { 9157 feature = __NETIF_F_BIT(feature_bit); 9158 if (!(upper->wanted_features & feature) 9159 && (features & feature)) { 9160 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", 9161 &feature, upper->name); 9162 features &= ~feature; 9163 } 9164 } 9165 9166 return features; 9167 } 9168 9169 static void netdev_sync_lower_features(struct net_device *upper, 9170 struct net_device *lower, netdev_features_t features) 9171 { 9172 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9173 netdev_features_t feature; 9174 int feature_bit; 9175 9176 for_each_netdev_feature(upper_disables, feature_bit) { 9177 feature = __NETIF_F_BIT(feature_bit); 9178 if (!(features & feature) && (lower->features & feature)) { 9179 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", 9180 &feature, lower->name); 9181 lower->wanted_features &= ~feature; 9182 __netdev_update_features(lower); 9183 9184 if (unlikely(lower->features & feature)) 9185 netdev_WARN(upper, "failed to disable %pNF on %s!\n", 9186 &feature, lower->name); 9187 else 9188 netdev_features_change(lower); 9189 } 9190 } 9191 } 9192 9193 static netdev_features_t netdev_fix_features(struct net_device *dev, 9194 netdev_features_t features) 9195 { 9196 /* Fix illegal checksum combinations */ 9197 if ((features & NETIF_F_HW_CSUM) && 9198 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { 9199 netdev_warn(dev, "mixed HW and IP checksum settings.\n"); 9200 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); 9201 } 9202 9203 /* TSO requires that SG is present as well. */ 9204 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { 9205 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); 9206 features &= ~NETIF_F_ALL_TSO; 9207 } 9208 9209 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && 9210 !(features & NETIF_F_IP_CSUM)) { 9211 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); 9212 features &= ~NETIF_F_TSO; 9213 features &= ~NETIF_F_TSO_ECN; 9214 } 9215 9216 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && 9217 !(features & NETIF_F_IPV6_CSUM)) { 9218 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); 9219 features &= ~NETIF_F_TSO6; 9220 } 9221 9222 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ 9223 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) 9224 features &= ~NETIF_F_TSO_MANGLEID; 9225 9226 /* TSO ECN requires that TSO is present as well. */ 9227 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) 9228 features &= ~NETIF_F_TSO_ECN; 9229 9230 /* Software GSO depends on SG. */ 9231 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { 9232 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); 9233 features &= ~NETIF_F_GSO; 9234 } 9235 9236 /* GSO partial features require GSO partial be set */ 9237 if ((features & dev->gso_partial_features) && 9238 !(features & NETIF_F_GSO_PARTIAL)) { 9239 netdev_dbg(dev, 9240 "Dropping partially supported GSO features since no GSO partial.\n"); 9241 features &= ~dev->gso_partial_features; 9242 } 9243 9244 if (!(features & NETIF_F_RXCSUM)) { 9245 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet 9246 * successfully merged by hardware must also have the 9247 * checksum verified by hardware. If the user does not 9248 * want to enable RXCSUM, logically, we should disable GRO_HW. 9249 */ 9250 if (features & NETIF_F_GRO_HW) { 9251 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n"); 9252 features &= ~NETIF_F_GRO_HW; 9253 } 9254 } 9255 9256 /* LRO/HW-GRO features cannot be combined with RX-FCS */ 9257 if (features & NETIF_F_RXFCS) { 9258 if (features & NETIF_F_LRO) { 9259 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n"); 9260 features &= ~NETIF_F_LRO; 9261 } 9262 9263 if (features & NETIF_F_GRO_HW) { 9264 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n"); 9265 features &= ~NETIF_F_GRO_HW; 9266 } 9267 } 9268 9269 if ((features & NETIF_F_GRO_HW) && (features & NETIF_F_LRO)) { 9270 netdev_dbg(dev, "Dropping LRO feature since HW-GRO is requested.\n"); 9271 features &= ~NETIF_F_LRO; 9272 } 9273 9274 if (features & NETIF_F_HW_TLS_TX) { 9275 bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) == 9276 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM); 9277 bool hw_csum = features & NETIF_F_HW_CSUM; 9278 9279 if (!ip_csum && !hw_csum) { 9280 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n"); 9281 features &= ~NETIF_F_HW_TLS_TX; 9282 } 9283 } 9284 9285 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) { 9286 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n"); 9287 features &= ~NETIF_F_HW_TLS_RX; 9288 } 9289 9290 return features; 9291 } 9292 9293 int __netdev_update_features(struct net_device *dev) 9294 { 9295 struct net_device *upper, *lower; 9296 netdev_features_t features; 9297 struct list_head *iter; 9298 int err = -1; 9299 9300 ASSERT_RTNL(); 9301 9302 features = netdev_get_wanted_features(dev); 9303 9304 if (dev->netdev_ops->ndo_fix_features) 9305 features = dev->netdev_ops->ndo_fix_features(dev, features); 9306 9307 /* driver might be less strict about feature dependencies */ 9308 features = netdev_fix_features(dev, features); 9309 9310 /* some features can't be enabled if they're off on an upper device */ 9311 netdev_for_each_upper_dev_rcu(dev, upper, iter) 9312 features = netdev_sync_upper_features(dev, upper, features); 9313 9314 if (dev->features == features) 9315 goto sync_lower; 9316 9317 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", 9318 &dev->features, &features); 9319 9320 if (dev->netdev_ops->ndo_set_features) 9321 err = dev->netdev_ops->ndo_set_features(dev, features); 9322 else 9323 err = 0; 9324 9325 if (unlikely(err < 0)) { 9326 netdev_err(dev, 9327 "set_features() failed (%d); wanted %pNF, left %pNF\n", 9328 err, &features, &dev->features); 9329 /* return non-0 since some features might have changed and 9330 * it's better to fire a spurious notification than miss it 9331 */ 9332 return -1; 9333 } 9334 9335 sync_lower: 9336 /* some features must be disabled on lower devices when disabled 9337 * on an upper device (think: bonding master or bridge) 9338 */ 9339 netdev_for_each_lower_dev(dev, lower, iter) 9340 netdev_sync_lower_features(dev, lower, features); 9341 9342 if (!err) { 9343 netdev_features_t diff = features ^ dev->features; 9344 9345 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) { 9346 /* udp_tunnel_{get,drop}_rx_info both need 9347 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the 9348 * device, or they won't do anything. 9349 * Thus we need to update dev->features 9350 * *before* calling udp_tunnel_get_rx_info, 9351 * but *after* calling udp_tunnel_drop_rx_info. 9352 */ 9353 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) { 9354 dev->features = features; 9355 udp_tunnel_get_rx_info(dev); 9356 } else { 9357 udp_tunnel_drop_rx_info(dev); 9358 } 9359 } 9360 9361 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) { 9362 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) { 9363 dev->features = features; 9364 err |= vlan_get_rx_ctag_filter_info(dev); 9365 } else { 9366 vlan_drop_rx_ctag_filter_info(dev); 9367 } 9368 } 9369 9370 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) { 9371 if (features & NETIF_F_HW_VLAN_STAG_FILTER) { 9372 dev->features = features; 9373 err |= vlan_get_rx_stag_filter_info(dev); 9374 } else { 9375 vlan_drop_rx_stag_filter_info(dev); 9376 } 9377 } 9378 9379 dev->features = features; 9380 } 9381 9382 return err < 0 ? 0 : 1; 9383 } 9384 9385 /** 9386 * netdev_update_features - recalculate device features 9387 * @dev: the device to check 9388 * 9389 * Recalculate dev->features set and send notifications if it 9390 * has changed. Should be called after driver or hardware dependent 9391 * conditions might have changed that influence the features. 9392 */ 9393 void netdev_update_features(struct net_device *dev) 9394 { 9395 if (__netdev_update_features(dev)) 9396 netdev_features_change(dev); 9397 } 9398 EXPORT_SYMBOL(netdev_update_features); 9399 9400 /** 9401 * netdev_change_features - recalculate device features 9402 * @dev: the device to check 9403 * 9404 * Recalculate dev->features set and send notifications even 9405 * if they have not changed. Should be called instead of 9406 * netdev_update_features() if also dev->vlan_features might 9407 * have changed to allow the changes to be propagated to stacked 9408 * VLAN devices. 9409 */ 9410 void netdev_change_features(struct net_device *dev) 9411 { 9412 __netdev_update_features(dev); 9413 netdev_features_change(dev); 9414 } 9415 EXPORT_SYMBOL(netdev_change_features); 9416 9417 /** 9418 * netif_stacked_transfer_operstate - transfer operstate 9419 * @rootdev: the root or lower level device to transfer state from 9420 * @dev: the device to transfer operstate to 9421 * 9422 * Transfer operational state from root to device. This is normally 9423 * called when a stacking relationship exists between the root 9424 * device and the device(a leaf device). 9425 */ 9426 void netif_stacked_transfer_operstate(const struct net_device *rootdev, 9427 struct net_device *dev) 9428 { 9429 if (rootdev->operstate == IF_OPER_DORMANT) 9430 netif_dormant_on(dev); 9431 else 9432 netif_dormant_off(dev); 9433 9434 if (rootdev->operstate == IF_OPER_TESTING) 9435 netif_testing_on(dev); 9436 else 9437 netif_testing_off(dev); 9438 9439 if (netif_carrier_ok(rootdev)) 9440 netif_carrier_on(dev); 9441 else 9442 netif_carrier_off(dev); 9443 } 9444 EXPORT_SYMBOL(netif_stacked_transfer_operstate); 9445 9446 static int netif_alloc_rx_queues(struct net_device *dev) 9447 { 9448 unsigned int i, count = dev->num_rx_queues; 9449 struct netdev_rx_queue *rx; 9450 size_t sz = count * sizeof(*rx); 9451 int err = 0; 9452 9453 BUG_ON(count < 1); 9454 9455 rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 9456 if (!rx) 9457 return -ENOMEM; 9458 9459 dev->_rx = rx; 9460 9461 for (i = 0; i < count; i++) { 9462 rx[i].dev = dev; 9463 9464 /* XDP RX-queue setup */ 9465 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0); 9466 if (err < 0) 9467 goto err_rxq_info; 9468 } 9469 return 0; 9470 9471 err_rxq_info: 9472 /* Rollback successful reg's and free other resources */ 9473 while (i--) 9474 xdp_rxq_info_unreg(&rx[i].xdp_rxq); 9475 kvfree(dev->_rx); 9476 dev->_rx = NULL; 9477 return err; 9478 } 9479 9480 static void netif_free_rx_queues(struct net_device *dev) 9481 { 9482 unsigned int i, count = dev->num_rx_queues; 9483 9484 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */ 9485 if (!dev->_rx) 9486 return; 9487 9488 for (i = 0; i < count; i++) 9489 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq); 9490 9491 kvfree(dev->_rx); 9492 } 9493 9494 static void netdev_init_one_queue(struct net_device *dev, 9495 struct netdev_queue *queue, void *_unused) 9496 { 9497 /* Initialize queue lock */ 9498 spin_lock_init(&queue->_xmit_lock); 9499 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); 9500 queue->xmit_lock_owner = -1; 9501 netdev_queue_numa_node_write(queue, NUMA_NO_NODE); 9502 queue->dev = dev; 9503 #ifdef CONFIG_BQL 9504 dql_init(&queue->dql, HZ); 9505 #endif 9506 } 9507 9508 static void netif_free_tx_queues(struct net_device *dev) 9509 { 9510 kvfree(dev->_tx); 9511 } 9512 9513 static int netif_alloc_netdev_queues(struct net_device *dev) 9514 { 9515 unsigned int count = dev->num_tx_queues; 9516 struct netdev_queue *tx; 9517 size_t sz = count * sizeof(*tx); 9518 9519 if (count < 1 || count > 0xffff) 9520 return -EINVAL; 9521 9522 tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 9523 if (!tx) 9524 return -ENOMEM; 9525 9526 dev->_tx = tx; 9527 9528 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); 9529 spin_lock_init(&dev->tx_global_lock); 9530 9531 return 0; 9532 } 9533 9534 void netif_tx_stop_all_queues(struct net_device *dev) 9535 { 9536 unsigned int i; 9537 9538 for (i = 0; i < dev->num_tx_queues; i++) { 9539 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 9540 9541 netif_tx_stop_queue(txq); 9542 } 9543 } 9544 EXPORT_SYMBOL(netif_tx_stop_all_queues); 9545 9546 /** 9547 * register_netdevice - register a network device 9548 * @dev: device to register 9549 * 9550 * Take a completed network device structure and add it to the kernel 9551 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 9552 * chain. 0 is returned on success. A negative errno code is returned 9553 * on a failure to set up the device, or if the name is a duplicate. 9554 * 9555 * Callers must hold the rtnl semaphore. You may want 9556 * register_netdev() instead of this. 9557 * 9558 * BUGS: 9559 * The locking appears insufficient to guarantee two parallel registers 9560 * will not get the same name. 9561 */ 9562 9563 int register_netdevice(struct net_device *dev) 9564 { 9565 int ret; 9566 struct net *net = dev_net(dev); 9567 9568 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE < 9569 NETDEV_FEATURE_COUNT); 9570 BUG_ON(dev_boot_phase); 9571 ASSERT_RTNL(); 9572 9573 might_sleep(); 9574 9575 /* When net_device's are persistent, this will be fatal. */ 9576 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); 9577 BUG_ON(!net); 9578 9579 ret = ethtool_check_ops(dev->ethtool_ops); 9580 if (ret) 9581 return ret; 9582 9583 spin_lock_init(&dev->addr_list_lock); 9584 netdev_set_addr_lockdep_class(dev); 9585 9586 ret = dev_get_valid_name(net, dev, dev->name); 9587 if (ret < 0) 9588 goto out; 9589 9590 ret = -ENOMEM; 9591 dev->name_node = netdev_name_node_head_alloc(dev); 9592 if (!dev->name_node) 9593 goto out; 9594 9595 /* Init, if this function is available */ 9596 if (dev->netdev_ops->ndo_init) { 9597 ret = dev->netdev_ops->ndo_init(dev); 9598 if (ret) { 9599 if (ret > 0) 9600 ret = -EIO; 9601 goto err_free_name; 9602 } 9603 } 9604 9605 if (((dev->hw_features | dev->features) & 9606 NETIF_F_HW_VLAN_CTAG_FILTER) && 9607 (!dev->netdev_ops->ndo_vlan_rx_add_vid || 9608 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { 9609 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); 9610 ret = -EINVAL; 9611 goto err_uninit; 9612 } 9613 9614 ret = -EBUSY; 9615 if (!dev->ifindex) 9616 dev->ifindex = dev_new_index(net); 9617 else if (__dev_get_by_index(net, dev->ifindex)) 9618 goto err_uninit; 9619 9620 /* Transfer changeable features to wanted_features and enable 9621 * software offloads (GSO and GRO). 9622 */ 9623 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF); 9624 dev->features |= NETIF_F_SOFT_FEATURES; 9625 9626 if (dev->udp_tunnel_nic_info) { 9627 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT; 9628 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT; 9629 } 9630 9631 dev->wanted_features = dev->features & dev->hw_features; 9632 9633 if (!(dev->flags & IFF_LOOPBACK)) 9634 dev->hw_features |= NETIF_F_NOCACHE_COPY; 9635 9636 /* If IPv4 TCP segmentation offload is supported we should also 9637 * allow the device to enable segmenting the frame with the option 9638 * of ignoring a static IP ID value. This doesn't enable the 9639 * feature itself but allows the user to enable it later. 9640 */ 9641 if (dev->hw_features & NETIF_F_TSO) 9642 dev->hw_features |= NETIF_F_TSO_MANGLEID; 9643 if (dev->vlan_features & NETIF_F_TSO) 9644 dev->vlan_features |= NETIF_F_TSO_MANGLEID; 9645 if (dev->mpls_features & NETIF_F_TSO) 9646 dev->mpls_features |= NETIF_F_TSO_MANGLEID; 9647 if (dev->hw_enc_features & NETIF_F_TSO) 9648 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; 9649 9650 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. 9651 */ 9652 dev->vlan_features |= NETIF_F_HIGHDMA; 9653 9654 /* Make NETIF_F_SG inheritable to tunnel devices. 9655 */ 9656 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; 9657 9658 /* Make NETIF_F_SG inheritable to MPLS. 9659 */ 9660 dev->mpls_features |= NETIF_F_SG; 9661 9662 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); 9663 ret = notifier_to_errno(ret); 9664 if (ret) 9665 goto err_uninit; 9666 9667 ret = netdev_register_kobject(dev); 9668 if (ret) { 9669 dev->reg_state = NETREG_UNREGISTERED; 9670 goto err_uninit; 9671 } 9672 dev->reg_state = NETREG_REGISTERED; 9673 9674 __netdev_update_features(dev); 9675 9676 /* 9677 * Default initial state at registry is that the 9678 * device is present. 9679 */ 9680 9681 set_bit(__LINK_STATE_PRESENT, &dev->state); 9682 9683 linkwatch_init_dev(dev); 9684 9685 dev_init_scheduler(dev); 9686 dev_hold(dev); 9687 list_netdevice(dev); 9688 add_device_randomness(dev->dev_addr, dev->addr_len); 9689 9690 /* If the device has permanent device address, driver should 9691 * set dev_addr and also addr_assign_type should be set to 9692 * NET_ADDR_PERM (default value). 9693 */ 9694 if (dev->addr_assign_type == NET_ADDR_PERM) 9695 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); 9696 9697 /* Notify protocols, that a new device appeared. */ 9698 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); 9699 ret = notifier_to_errno(ret); 9700 if (ret) { 9701 /* Expect explicit free_netdev() on failure */ 9702 dev->needs_free_netdev = false; 9703 unregister_netdevice_queue(dev, NULL); 9704 goto out; 9705 } 9706 /* 9707 * Prevent userspace races by waiting until the network 9708 * device is fully setup before sending notifications. 9709 */ 9710 if (!dev->rtnl_link_ops || 9711 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 9712 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); 9713 9714 out: 9715 return ret; 9716 9717 err_uninit: 9718 if (dev->netdev_ops->ndo_uninit) 9719 dev->netdev_ops->ndo_uninit(dev); 9720 if (dev->priv_destructor) 9721 dev->priv_destructor(dev); 9722 err_free_name: 9723 netdev_name_node_free(dev->name_node); 9724 goto out; 9725 } 9726 EXPORT_SYMBOL(register_netdevice); 9727 9728 /** 9729 * init_dummy_netdev - init a dummy network device for NAPI 9730 * @dev: device to init 9731 * 9732 * This takes a network device structure and initialize the minimum 9733 * amount of fields so it can be used to schedule NAPI polls without 9734 * registering a full blown interface. This is to be used by drivers 9735 * that need to tie several hardware interfaces to a single NAPI 9736 * poll scheduler due to HW limitations. 9737 */ 9738 int init_dummy_netdev(struct net_device *dev) 9739 { 9740 /* Clear everything. Note we don't initialize spinlocks 9741 * are they aren't supposed to be taken by any of the 9742 * NAPI code and this dummy netdev is supposed to be 9743 * only ever used for NAPI polls 9744 */ 9745 memset(dev, 0, sizeof(struct net_device)); 9746 9747 /* make sure we BUG if trying to hit standard 9748 * register/unregister code path 9749 */ 9750 dev->reg_state = NETREG_DUMMY; 9751 9752 /* NAPI wants this */ 9753 INIT_LIST_HEAD(&dev->napi_list); 9754 9755 /* a dummy interface is started by default */ 9756 set_bit(__LINK_STATE_PRESENT, &dev->state); 9757 set_bit(__LINK_STATE_START, &dev->state); 9758 9759 /* napi_busy_loop stats accounting wants this */ 9760 dev_net_set(dev, &init_net); 9761 9762 /* Note : We dont allocate pcpu_refcnt for dummy devices, 9763 * because users of this 'device' dont need to change 9764 * its refcount. 9765 */ 9766 9767 return 0; 9768 } 9769 EXPORT_SYMBOL_GPL(init_dummy_netdev); 9770 9771 9772 /** 9773 * register_netdev - register a network device 9774 * @dev: device to register 9775 * 9776 * Take a completed network device structure and add it to the kernel 9777 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 9778 * chain. 0 is returned on success. A negative errno code is returned 9779 * on a failure to set up the device, or if the name is a duplicate. 9780 * 9781 * This is a wrapper around register_netdevice that takes the rtnl semaphore 9782 * and expands the device name if you passed a format string to 9783 * alloc_netdev. 9784 */ 9785 int register_netdev(struct net_device *dev) 9786 { 9787 int err; 9788 9789 if (rtnl_lock_killable()) 9790 return -EINTR; 9791 err = register_netdevice(dev); 9792 rtnl_unlock(); 9793 return err; 9794 } 9795 EXPORT_SYMBOL(register_netdev); 9796 9797 int netdev_refcnt_read(const struct net_device *dev) 9798 { 9799 #ifdef CONFIG_PCPU_DEV_REFCNT 9800 int i, refcnt = 0; 9801 9802 for_each_possible_cpu(i) 9803 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); 9804 return refcnt; 9805 #else 9806 return refcount_read(&dev->dev_refcnt); 9807 #endif 9808 } 9809 EXPORT_SYMBOL(netdev_refcnt_read); 9810 9811 int netdev_unregister_timeout_secs __read_mostly = 10; 9812 9813 #define WAIT_REFS_MIN_MSECS 1 9814 #define WAIT_REFS_MAX_MSECS 250 9815 /** 9816 * netdev_wait_allrefs - wait until all references are gone. 9817 * @dev: target net_device 9818 * 9819 * This is called when unregistering network devices. 9820 * 9821 * Any protocol or device that holds a reference should register 9822 * for netdevice notification, and cleanup and put back the 9823 * reference if they receive an UNREGISTER event. 9824 * We can get stuck here if buggy protocols don't correctly 9825 * call dev_put. 9826 */ 9827 static void netdev_wait_allrefs(struct net_device *dev) 9828 { 9829 unsigned long rebroadcast_time, warning_time; 9830 int wait = 0, refcnt; 9831 9832 linkwatch_forget_dev(dev); 9833 9834 rebroadcast_time = warning_time = jiffies; 9835 refcnt = netdev_refcnt_read(dev); 9836 9837 while (refcnt != 1) { 9838 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { 9839 rtnl_lock(); 9840 9841 /* Rebroadcast unregister notification */ 9842 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 9843 9844 __rtnl_unlock(); 9845 rcu_barrier(); 9846 rtnl_lock(); 9847 9848 if (test_bit(__LINK_STATE_LINKWATCH_PENDING, 9849 &dev->state)) { 9850 /* We must not have linkwatch events 9851 * pending on unregister. If this 9852 * happens, we simply run the queue 9853 * unscheduled, resulting in a noop 9854 * for this device. 9855 */ 9856 linkwatch_run_queue(); 9857 } 9858 9859 __rtnl_unlock(); 9860 9861 rebroadcast_time = jiffies; 9862 } 9863 9864 if (!wait) { 9865 rcu_barrier(); 9866 wait = WAIT_REFS_MIN_MSECS; 9867 } else { 9868 msleep(wait); 9869 wait = min(wait << 1, WAIT_REFS_MAX_MSECS); 9870 } 9871 9872 refcnt = netdev_refcnt_read(dev); 9873 9874 if (refcnt != 1 && 9875 time_after(jiffies, warning_time + 9876 netdev_unregister_timeout_secs * HZ)) { 9877 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", 9878 dev->name, refcnt); 9879 ref_tracker_dir_print(&dev->refcnt_tracker, 10); 9880 warning_time = jiffies; 9881 } 9882 } 9883 } 9884 9885 /* The sequence is: 9886 * 9887 * rtnl_lock(); 9888 * ... 9889 * register_netdevice(x1); 9890 * register_netdevice(x2); 9891 * ... 9892 * unregister_netdevice(y1); 9893 * unregister_netdevice(y2); 9894 * ... 9895 * rtnl_unlock(); 9896 * free_netdev(y1); 9897 * free_netdev(y2); 9898 * 9899 * We are invoked by rtnl_unlock(). 9900 * This allows us to deal with problems: 9901 * 1) We can delete sysfs objects which invoke hotplug 9902 * without deadlocking with linkwatch via keventd. 9903 * 2) Since we run with the RTNL semaphore not held, we can sleep 9904 * safely in order to wait for the netdev refcnt to drop to zero. 9905 * 9906 * We must not return until all unregister events added during 9907 * the interval the lock was held have been completed. 9908 */ 9909 void netdev_run_todo(void) 9910 { 9911 struct list_head list; 9912 #ifdef CONFIG_LOCKDEP 9913 struct list_head unlink_list; 9914 9915 list_replace_init(&net_unlink_list, &unlink_list); 9916 9917 while (!list_empty(&unlink_list)) { 9918 struct net_device *dev = list_first_entry(&unlink_list, 9919 struct net_device, 9920 unlink_list); 9921 list_del_init(&dev->unlink_list); 9922 dev->nested_level = dev->lower_level - 1; 9923 } 9924 #endif 9925 9926 /* Snapshot list, allow later requests */ 9927 list_replace_init(&net_todo_list, &list); 9928 9929 __rtnl_unlock(); 9930 9931 9932 /* Wait for rcu callbacks to finish before next phase */ 9933 if (!list_empty(&list)) 9934 rcu_barrier(); 9935 9936 while (!list_empty(&list)) { 9937 struct net_device *dev 9938 = list_first_entry(&list, struct net_device, todo_list); 9939 list_del(&dev->todo_list); 9940 9941 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { 9942 pr_err("network todo '%s' but state %d\n", 9943 dev->name, dev->reg_state); 9944 dump_stack(); 9945 continue; 9946 } 9947 9948 dev->reg_state = NETREG_UNREGISTERED; 9949 9950 netdev_wait_allrefs(dev); 9951 9952 /* paranoia */ 9953 BUG_ON(netdev_refcnt_read(dev) != 1); 9954 BUG_ON(!list_empty(&dev->ptype_all)); 9955 BUG_ON(!list_empty(&dev->ptype_specific)); 9956 WARN_ON(rcu_access_pointer(dev->ip_ptr)); 9957 WARN_ON(rcu_access_pointer(dev->ip6_ptr)); 9958 #if IS_ENABLED(CONFIG_DECNET) 9959 WARN_ON(dev->dn_ptr); 9960 #endif 9961 if (dev->priv_destructor) 9962 dev->priv_destructor(dev); 9963 if (dev->needs_free_netdev) 9964 free_netdev(dev); 9965 9966 /* Report a network device has been unregistered */ 9967 rtnl_lock(); 9968 dev_net(dev)->dev_unreg_count--; 9969 __rtnl_unlock(); 9970 wake_up(&netdev_unregistering_wq); 9971 9972 /* Free network device */ 9973 kobject_put(&dev->dev.kobj); 9974 } 9975 } 9976 9977 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has 9978 * all the same fields in the same order as net_device_stats, with only 9979 * the type differing, but rtnl_link_stats64 may have additional fields 9980 * at the end for newer counters. 9981 */ 9982 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, 9983 const struct net_device_stats *netdev_stats) 9984 { 9985 #if BITS_PER_LONG == 64 9986 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats)); 9987 memcpy(stats64, netdev_stats, sizeof(*netdev_stats)); 9988 /* zero out counters that only exist in rtnl_link_stats64 */ 9989 memset((char *)stats64 + sizeof(*netdev_stats), 0, 9990 sizeof(*stats64) - sizeof(*netdev_stats)); 9991 #else 9992 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long); 9993 const unsigned long *src = (const unsigned long *)netdev_stats; 9994 u64 *dst = (u64 *)stats64; 9995 9996 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); 9997 for (i = 0; i < n; i++) 9998 dst[i] = src[i]; 9999 /* zero out counters that only exist in rtnl_link_stats64 */ 10000 memset((char *)stats64 + n * sizeof(u64), 0, 10001 sizeof(*stats64) - n * sizeof(u64)); 10002 #endif 10003 } 10004 EXPORT_SYMBOL(netdev_stats_to_stats64); 10005 10006 /** 10007 * dev_get_stats - get network device statistics 10008 * @dev: device to get statistics from 10009 * @storage: place to store stats 10010 * 10011 * Get network statistics from device. Return @storage. 10012 * The device driver may provide its own method by setting 10013 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; 10014 * otherwise the internal statistics structure is used. 10015 */ 10016 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, 10017 struct rtnl_link_stats64 *storage) 10018 { 10019 const struct net_device_ops *ops = dev->netdev_ops; 10020 10021 if (ops->ndo_get_stats64) { 10022 memset(storage, 0, sizeof(*storage)); 10023 ops->ndo_get_stats64(dev, storage); 10024 } else if (ops->ndo_get_stats) { 10025 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); 10026 } else { 10027 netdev_stats_to_stats64(storage, &dev->stats); 10028 } 10029 storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped); 10030 storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped); 10031 storage->rx_nohandler += (unsigned long)atomic_long_read(&dev->rx_nohandler); 10032 return storage; 10033 } 10034 EXPORT_SYMBOL(dev_get_stats); 10035 10036 /** 10037 * dev_fetch_sw_netstats - get per-cpu network device statistics 10038 * @s: place to store stats 10039 * @netstats: per-cpu network stats to read from 10040 * 10041 * Read per-cpu network statistics and populate the related fields in @s. 10042 */ 10043 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s, 10044 const struct pcpu_sw_netstats __percpu *netstats) 10045 { 10046 int cpu; 10047 10048 for_each_possible_cpu(cpu) { 10049 const struct pcpu_sw_netstats *stats; 10050 struct pcpu_sw_netstats tmp; 10051 unsigned int start; 10052 10053 stats = per_cpu_ptr(netstats, cpu); 10054 do { 10055 start = u64_stats_fetch_begin_irq(&stats->syncp); 10056 tmp.rx_packets = stats->rx_packets; 10057 tmp.rx_bytes = stats->rx_bytes; 10058 tmp.tx_packets = stats->tx_packets; 10059 tmp.tx_bytes = stats->tx_bytes; 10060 } while (u64_stats_fetch_retry_irq(&stats->syncp, start)); 10061 10062 s->rx_packets += tmp.rx_packets; 10063 s->rx_bytes += tmp.rx_bytes; 10064 s->tx_packets += tmp.tx_packets; 10065 s->tx_bytes += tmp.tx_bytes; 10066 } 10067 } 10068 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats); 10069 10070 /** 10071 * dev_get_tstats64 - ndo_get_stats64 implementation 10072 * @dev: device to get statistics from 10073 * @s: place to store stats 10074 * 10075 * Populate @s from dev->stats and dev->tstats. Can be used as 10076 * ndo_get_stats64() callback. 10077 */ 10078 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s) 10079 { 10080 netdev_stats_to_stats64(s, &dev->stats); 10081 dev_fetch_sw_netstats(s, dev->tstats); 10082 } 10083 EXPORT_SYMBOL_GPL(dev_get_tstats64); 10084 10085 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) 10086 { 10087 struct netdev_queue *queue = dev_ingress_queue(dev); 10088 10089 #ifdef CONFIG_NET_CLS_ACT 10090 if (queue) 10091 return queue; 10092 queue = kzalloc(sizeof(*queue), GFP_KERNEL); 10093 if (!queue) 10094 return NULL; 10095 netdev_init_one_queue(dev, queue, NULL); 10096 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); 10097 queue->qdisc_sleeping = &noop_qdisc; 10098 rcu_assign_pointer(dev->ingress_queue, queue); 10099 #endif 10100 return queue; 10101 } 10102 10103 static const struct ethtool_ops default_ethtool_ops; 10104 10105 void netdev_set_default_ethtool_ops(struct net_device *dev, 10106 const struct ethtool_ops *ops) 10107 { 10108 if (dev->ethtool_ops == &default_ethtool_ops) 10109 dev->ethtool_ops = ops; 10110 } 10111 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); 10112 10113 void netdev_freemem(struct net_device *dev) 10114 { 10115 char *addr = (char *)dev - dev->padded; 10116 10117 kvfree(addr); 10118 } 10119 10120 /** 10121 * alloc_netdev_mqs - allocate network device 10122 * @sizeof_priv: size of private data to allocate space for 10123 * @name: device name format string 10124 * @name_assign_type: origin of device name 10125 * @setup: callback to initialize device 10126 * @txqs: the number of TX subqueues to allocate 10127 * @rxqs: the number of RX subqueues to allocate 10128 * 10129 * Allocates a struct net_device with private data area for driver use 10130 * and performs basic initialization. Also allocates subqueue structs 10131 * for each queue on the device. 10132 */ 10133 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, 10134 unsigned char name_assign_type, 10135 void (*setup)(struct net_device *), 10136 unsigned int txqs, unsigned int rxqs) 10137 { 10138 struct net_device *dev; 10139 unsigned int alloc_size; 10140 struct net_device *p; 10141 10142 BUG_ON(strlen(name) >= sizeof(dev->name)); 10143 10144 if (txqs < 1) { 10145 pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); 10146 return NULL; 10147 } 10148 10149 if (rxqs < 1) { 10150 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); 10151 return NULL; 10152 } 10153 10154 alloc_size = sizeof(struct net_device); 10155 if (sizeof_priv) { 10156 /* ensure 32-byte alignment of private area */ 10157 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); 10158 alloc_size += sizeof_priv; 10159 } 10160 /* ensure 32-byte alignment of whole construct */ 10161 alloc_size += NETDEV_ALIGN - 1; 10162 10163 p = kvzalloc(alloc_size, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10164 if (!p) 10165 return NULL; 10166 10167 dev = PTR_ALIGN(p, NETDEV_ALIGN); 10168 dev->padded = (char *)dev - (char *)p; 10169 10170 ref_tracker_dir_init(&dev->refcnt_tracker, 128); 10171 #ifdef CONFIG_PCPU_DEV_REFCNT 10172 dev->pcpu_refcnt = alloc_percpu(int); 10173 if (!dev->pcpu_refcnt) 10174 goto free_dev; 10175 dev_hold(dev); 10176 #else 10177 refcount_set(&dev->dev_refcnt, 1); 10178 #endif 10179 10180 if (dev_addr_init(dev)) 10181 goto free_pcpu; 10182 10183 dev_mc_init(dev); 10184 dev_uc_init(dev); 10185 10186 dev_net_set(dev, &init_net); 10187 10188 dev->gso_max_size = GSO_MAX_SIZE; 10189 dev->gso_max_segs = GSO_MAX_SEGS; 10190 dev->gro_max_size = GRO_MAX_SIZE; 10191 dev->upper_level = 1; 10192 dev->lower_level = 1; 10193 #ifdef CONFIG_LOCKDEP 10194 dev->nested_level = 0; 10195 INIT_LIST_HEAD(&dev->unlink_list); 10196 #endif 10197 10198 INIT_LIST_HEAD(&dev->napi_list); 10199 INIT_LIST_HEAD(&dev->unreg_list); 10200 INIT_LIST_HEAD(&dev->close_list); 10201 INIT_LIST_HEAD(&dev->link_watch_list); 10202 INIT_LIST_HEAD(&dev->adj_list.upper); 10203 INIT_LIST_HEAD(&dev->adj_list.lower); 10204 INIT_LIST_HEAD(&dev->ptype_all); 10205 INIT_LIST_HEAD(&dev->ptype_specific); 10206 INIT_LIST_HEAD(&dev->net_notifier_list); 10207 #ifdef CONFIG_NET_SCHED 10208 hash_init(dev->qdisc_hash); 10209 #endif 10210 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; 10211 setup(dev); 10212 10213 if (!dev->tx_queue_len) { 10214 dev->priv_flags |= IFF_NO_QUEUE; 10215 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; 10216 } 10217 10218 dev->num_tx_queues = txqs; 10219 dev->real_num_tx_queues = txqs; 10220 if (netif_alloc_netdev_queues(dev)) 10221 goto free_all; 10222 10223 dev->num_rx_queues = rxqs; 10224 dev->real_num_rx_queues = rxqs; 10225 if (netif_alloc_rx_queues(dev)) 10226 goto free_all; 10227 10228 strcpy(dev->name, name); 10229 dev->name_assign_type = name_assign_type; 10230 dev->group = INIT_NETDEV_GROUP; 10231 if (!dev->ethtool_ops) 10232 dev->ethtool_ops = &default_ethtool_ops; 10233 10234 nf_hook_netdev_init(dev); 10235 10236 return dev; 10237 10238 free_all: 10239 free_netdev(dev); 10240 return NULL; 10241 10242 free_pcpu: 10243 #ifdef CONFIG_PCPU_DEV_REFCNT 10244 free_percpu(dev->pcpu_refcnt); 10245 free_dev: 10246 #endif 10247 netdev_freemem(dev); 10248 return NULL; 10249 } 10250 EXPORT_SYMBOL(alloc_netdev_mqs); 10251 10252 /** 10253 * free_netdev - free network device 10254 * @dev: device 10255 * 10256 * This function does the last stage of destroying an allocated device 10257 * interface. The reference to the device object is released. If this 10258 * is the last reference then it will be freed.Must be called in process 10259 * context. 10260 */ 10261 void free_netdev(struct net_device *dev) 10262 { 10263 struct napi_struct *p, *n; 10264 10265 might_sleep(); 10266 10267 /* When called immediately after register_netdevice() failed the unwind 10268 * handling may still be dismantling the device. Handle that case by 10269 * deferring the free. 10270 */ 10271 if (dev->reg_state == NETREG_UNREGISTERING) { 10272 ASSERT_RTNL(); 10273 dev->needs_free_netdev = true; 10274 return; 10275 } 10276 10277 netif_free_tx_queues(dev); 10278 netif_free_rx_queues(dev); 10279 10280 kfree(rcu_dereference_protected(dev->ingress_queue, 1)); 10281 10282 /* Flush device addresses */ 10283 dev_addr_flush(dev); 10284 10285 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) 10286 netif_napi_del(p); 10287 10288 ref_tracker_dir_exit(&dev->refcnt_tracker); 10289 #ifdef CONFIG_PCPU_DEV_REFCNT 10290 free_percpu(dev->pcpu_refcnt); 10291 dev->pcpu_refcnt = NULL; 10292 #endif 10293 free_percpu(dev->xdp_bulkq); 10294 dev->xdp_bulkq = NULL; 10295 10296 /* Compatibility with error handling in drivers */ 10297 if (dev->reg_state == NETREG_UNINITIALIZED) { 10298 netdev_freemem(dev); 10299 return; 10300 } 10301 10302 BUG_ON(dev->reg_state != NETREG_UNREGISTERED); 10303 dev->reg_state = NETREG_RELEASED; 10304 10305 /* will free via device release */ 10306 put_device(&dev->dev); 10307 } 10308 EXPORT_SYMBOL(free_netdev); 10309 10310 /** 10311 * synchronize_net - Synchronize with packet receive processing 10312 * 10313 * Wait for packets currently being received to be done. 10314 * Does not block later packets from starting. 10315 */ 10316 void synchronize_net(void) 10317 { 10318 might_sleep(); 10319 if (rtnl_is_locked()) 10320 synchronize_rcu_expedited(); 10321 else 10322 synchronize_rcu(); 10323 } 10324 EXPORT_SYMBOL(synchronize_net); 10325 10326 /** 10327 * unregister_netdevice_queue - remove device from the kernel 10328 * @dev: device 10329 * @head: list 10330 * 10331 * This function shuts down a device interface and removes it 10332 * from the kernel tables. 10333 * If head not NULL, device is queued to be unregistered later. 10334 * 10335 * Callers must hold the rtnl semaphore. You may want 10336 * unregister_netdev() instead of this. 10337 */ 10338 10339 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) 10340 { 10341 ASSERT_RTNL(); 10342 10343 if (head) { 10344 list_move_tail(&dev->unreg_list, head); 10345 } else { 10346 LIST_HEAD(single); 10347 10348 list_add(&dev->unreg_list, &single); 10349 unregister_netdevice_many(&single); 10350 } 10351 } 10352 EXPORT_SYMBOL(unregister_netdevice_queue); 10353 10354 /** 10355 * unregister_netdevice_many - unregister many devices 10356 * @head: list of devices 10357 * 10358 * Note: As most callers use a stack allocated list_head, 10359 * we force a list_del() to make sure stack wont be corrupted later. 10360 */ 10361 void unregister_netdevice_many(struct list_head *head) 10362 { 10363 struct net_device *dev, *tmp; 10364 LIST_HEAD(close_head); 10365 10366 BUG_ON(dev_boot_phase); 10367 ASSERT_RTNL(); 10368 10369 if (list_empty(head)) 10370 return; 10371 10372 list_for_each_entry_safe(dev, tmp, head, unreg_list) { 10373 /* Some devices call without registering 10374 * for initialization unwind. Remove those 10375 * devices and proceed with the remaining. 10376 */ 10377 if (dev->reg_state == NETREG_UNINITIALIZED) { 10378 pr_debug("unregister_netdevice: device %s/%p never was registered\n", 10379 dev->name, dev); 10380 10381 WARN_ON(1); 10382 list_del(&dev->unreg_list); 10383 continue; 10384 } 10385 dev->dismantle = true; 10386 BUG_ON(dev->reg_state != NETREG_REGISTERED); 10387 } 10388 10389 /* If device is running, close it first. */ 10390 list_for_each_entry(dev, head, unreg_list) 10391 list_add_tail(&dev->close_list, &close_head); 10392 dev_close_many(&close_head, true); 10393 10394 list_for_each_entry(dev, head, unreg_list) { 10395 /* And unlink it from device chain. */ 10396 unlist_netdevice(dev); 10397 10398 dev->reg_state = NETREG_UNREGISTERING; 10399 } 10400 flush_all_backlogs(); 10401 10402 synchronize_net(); 10403 10404 list_for_each_entry(dev, head, unreg_list) { 10405 struct sk_buff *skb = NULL; 10406 10407 /* Shutdown queueing discipline. */ 10408 dev_shutdown(dev); 10409 10410 dev_xdp_uninstall(dev); 10411 10412 /* Notify protocols, that we are about to destroy 10413 * this device. They should clean all the things. 10414 */ 10415 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10416 10417 if (!dev->rtnl_link_ops || 10418 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 10419 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0, 10420 GFP_KERNEL, NULL, 0); 10421 10422 /* 10423 * Flush the unicast and multicast chains 10424 */ 10425 dev_uc_flush(dev); 10426 dev_mc_flush(dev); 10427 10428 netdev_name_node_alt_flush(dev); 10429 netdev_name_node_free(dev->name_node); 10430 10431 if (dev->netdev_ops->ndo_uninit) 10432 dev->netdev_ops->ndo_uninit(dev); 10433 10434 if (skb) 10435 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL); 10436 10437 /* Notifier chain MUST detach us all upper devices. */ 10438 WARN_ON(netdev_has_any_upper_dev(dev)); 10439 WARN_ON(netdev_has_any_lower_dev(dev)); 10440 10441 /* Remove entries from kobject tree */ 10442 netdev_unregister_kobject(dev); 10443 #ifdef CONFIG_XPS 10444 /* Remove XPS queueing entries */ 10445 netif_reset_xps_queues_gt(dev, 0); 10446 #endif 10447 } 10448 10449 synchronize_net(); 10450 10451 list_for_each_entry(dev, head, unreg_list) { 10452 dev_put(dev); 10453 net_set_todo(dev); 10454 } 10455 10456 list_del(head); 10457 } 10458 EXPORT_SYMBOL(unregister_netdevice_many); 10459 10460 /** 10461 * unregister_netdev - remove device from the kernel 10462 * @dev: device 10463 * 10464 * This function shuts down a device interface and removes it 10465 * from the kernel tables. 10466 * 10467 * This is just a wrapper for unregister_netdevice that takes 10468 * the rtnl semaphore. In general you want to use this and not 10469 * unregister_netdevice. 10470 */ 10471 void unregister_netdev(struct net_device *dev) 10472 { 10473 rtnl_lock(); 10474 unregister_netdevice(dev); 10475 rtnl_unlock(); 10476 } 10477 EXPORT_SYMBOL(unregister_netdev); 10478 10479 /** 10480 * __dev_change_net_namespace - move device to different nethost namespace 10481 * @dev: device 10482 * @net: network namespace 10483 * @pat: If not NULL name pattern to try if the current device name 10484 * is already taken in the destination network namespace. 10485 * @new_ifindex: If not zero, specifies device index in the target 10486 * namespace. 10487 * 10488 * This function shuts down a device interface and moves it 10489 * to a new network namespace. On success 0 is returned, on 10490 * a failure a netagive errno code is returned. 10491 * 10492 * Callers must hold the rtnl semaphore. 10493 */ 10494 10495 int __dev_change_net_namespace(struct net_device *dev, struct net *net, 10496 const char *pat, int new_ifindex) 10497 { 10498 struct net *net_old = dev_net(dev); 10499 int err, new_nsid; 10500 10501 ASSERT_RTNL(); 10502 10503 /* Don't allow namespace local devices to be moved. */ 10504 err = -EINVAL; 10505 if (dev->features & NETIF_F_NETNS_LOCAL) 10506 goto out; 10507 10508 /* Ensure the device has been registrered */ 10509 if (dev->reg_state != NETREG_REGISTERED) 10510 goto out; 10511 10512 /* Get out if there is nothing todo */ 10513 err = 0; 10514 if (net_eq(net_old, net)) 10515 goto out; 10516 10517 /* Pick the destination device name, and ensure 10518 * we can use it in the destination network namespace. 10519 */ 10520 err = -EEXIST; 10521 if (netdev_name_in_use(net, dev->name)) { 10522 /* We get here if we can't use the current device name */ 10523 if (!pat) 10524 goto out; 10525 err = dev_get_valid_name(net, dev, pat); 10526 if (err < 0) 10527 goto out; 10528 } 10529 10530 /* Check that new_ifindex isn't used yet. */ 10531 err = -EBUSY; 10532 if (new_ifindex && __dev_get_by_index(net, new_ifindex)) 10533 goto out; 10534 10535 /* 10536 * And now a mini version of register_netdevice unregister_netdevice. 10537 */ 10538 10539 /* If device is running close it first. */ 10540 dev_close(dev); 10541 10542 /* And unlink it from device chain */ 10543 unlist_netdevice(dev); 10544 10545 synchronize_net(); 10546 10547 /* Shutdown queueing discipline. */ 10548 dev_shutdown(dev); 10549 10550 /* Notify protocols, that we are about to destroy 10551 * this device. They should clean all the things. 10552 * 10553 * Note that dev->reg_state stays at NETREG_REGISTERED. 10554 * This is wanted because this way 8021q and macvlan know 10555 * the device is just moving and can keep their slaves up. 10556 */ 10557 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10558 rcu_barrier(); 10559 10560 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL); 10561 /* If there is an ifindex conflict assign a new one */ 10562 if (!new_ifindex) { 10563 if (__dev_get_by_index(net, dev->ifindex)) 10564 new_ifindex = dev_new_index(net); 10565 else 10566 new_ifindex = dev->ifindex; 10567 } 10568 10569 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid, 10570 new_ifindex); 10571 10572 /* 10573 * Flush the unicast and multicast chains 10574 */ 10575 dev_uc_flush(dev); 10576 dev_mc_flush(dev); 10577 10578 /* Send a netdev-removed uevent to the old namespace */ 10579 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); 10580 netdev_adjacent_del_links(dev); 10581 10582 /* Move per-net netdevice notifiers that are following the netdevice */ 10583 move_netdevice_notifiers_dev_net(dev, net); 10584 10585 /* Actually switch the network namespace */ 10586 dev_net_set(dev, net); 10587 dev->ifindex = new_ifindex; 10588 10589 /* Send a netdev-add uevent to the new namespace */ 10590 kobject_uevent(&dev->dev.kobj, KOBJ_ADD); 10591 netdev_adjacent_add_links(dev); 10592 10593 /* Fixup kobjects */ 10594 err = device_rename(&dev->dev, dev->name); 10595 WARN_ON(err); 10596 10597 /* Adapt owner in case owning user namespace of target network 10598 * namespace is different from the original one. 10599 */ 10600 err = netdev_change_owner(dev, net_old, net); 10601 WARN_ON(err); 10602 10603 /* Add the device back in the hashes */ 10604 list_netdevice(dev); 10605 10606 /* Notify protocols, that a new device appeared. */ 10607 call_netdevice_notifiers(NETDEV_REGISTER, dev); 10608 10609 /* 10610 * Prevent userspace races by waiting until the network 10611 * device is fully setup before sending notifications. 10612 */ 10613 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); 10614 10615 synchronize_net(); 10616 err = 0; 10617 out: 10618 return err; 10619 } 10620 EXPORT_SYMBOL_GPL(__dev_change_net_namespace); 10621 10622 static int dev_cpu_dead(unsigned int oldcpu) 10623 { 10624 struct sk_buff **list_skb; 10625 struct sk_buff *skb; 10626 unsigned int cpu; 10627 struct softnet_data *sd, *oldsd, *remsd = NULL; 10628 10629 local_irq_disable(); 10630 cpu = smp_processor_id(); 10631 sd = &per_cpu(softnet_data, cpu); 10632 oldsd = &per_cpu(softnet_data, oldcpu); 10633 10634 /* Find end of our completion_queue. */ 10635 list_skb = &sd->completion_queue; 10636 while (*list_skb) 10637 list_skb = &(*list_skb)->next; 10638 /* Append completion queue from offline CPU. */ 10639 *list_skb = oldsd->completion_queue; 10640 oldsd->completion_queue = NULL; 10641 10642 /* Append output queue from offline CPU. */ 10643 if (oldsd->output_queue) { 10644 *sd->output_queue_tailp = oldsd->output_queue; 10645 sd->output_queue_tailp = oldsd->output_queue_tailp; 10646 oldsd->output_queue = NULL; 10647 oldsd->output_queue_tailp = &oldsd->output_queue; 10648 } 10649 /* Append NAPI poll list from offline CPU, with one exception : 10650 * process_backlog() must be called by cpu owning percpu backlog. 10651 * We properly handle process_queue & input_pkt_queue later. 10652 */ 10653 while (!list_empty(&oldsd->poll_list)) { 10654 struct napi_struct *napi = list_first_entry(&oldsd->poll_list, 10655 struct napi_struct, 10656 poll_list); 10657 10658 list_del_init(&napi->poll_list); 10659 if (napi->poll == process_backlog) 10660 napi->state = 0; 10661 else 10662 ____napi_schedule(sd, napi); 10663 } 10664 10665 raise_softirq_irqoff(NET_TX_SOFTIRQ); 10666 local_irq_enable(); 10667 10668 #ifdef CONFIG_RPS 10669 remsd = oldsd->rps_ipi_list; 10670 oldsd->rps_ipi_list = NULL; 10671 #endif 10672 /* send out pending IPI's on offline CPU */ 10673 net_rps_send_ipi(remsd); 10674 10675 /* Process offline CPU's input_pkt_queue */ 10676 while ((skb = __skb_dequeue(&oldsd->process_queue))) { 10677 netif_rx_ni(skb); 10678 input_queue_head_incr(oldsd); 10679 } 10680 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { 10681 netif_rx_ni(skb); 10682 input_queue_head_incr(oldsd); 10683 } 10684 10685 return 0; 10686 } 10687 10688 /** 10689 * netdev_increment_features - increment feature set by one 10690 * @all: current feature set 10691 * @one: new feature set 10692 * @mask: mask feature set 10693 * 10694 * Computes a new feature set after adding a device with feature set 10695 * @one to the master device with current feature set @all. Will not 10696 * enable anything that is off in @mask. Returns the new feature set. 10697 */ 10698 netdev_features_t netdev_increment_features(netdev_features_t all, 10699 netdev_features_t one, netdev_features_t mask) 10700 { 10701 if (mask & NETIF_F_HW_CSUM) 10702 mask |= NETIF_F_CSUM_MASK; 10703 mask |= NETIF_F_VLAN_CHALLENGED; 10704 10705 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; 10706 all &= one | ~NETIF_F_ALL_FOR_ALL; 10707 10708 /* If one device supports hw checksumming, set for all. */ 10709 if (all & NETIF_F_HW_CSUM) 10710 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); 10711 10712 return all; 10713 } 10714 EXPORT_SYMBOL(netdev_increment_features); 10715 10716 static struct hlist_head * __net_init netdev_create_hash(void) 10717 { 10718 int i; 10719 struct hlist_head *hash; 10720 10721 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL); 10722 if (hash != NULL) 10723 for (i = 0; i < NETDEV_HASHENTRIES; i++) 10724 INIT_HLIST_HEAD(&hash[i]); 10725 10726 return hash; 10727 } 10728 10729 /* Initialize per network namespace state */ 10730 static int __net_init netdev_init(struct net *net) 10731 { 10732 BUILD_BUG_ON(GRO_HASH_BUCKETS > 10733 8 * sizeof_field(struct napi_struct, gro_bitmask)); 10734 10735 if (net != &init_net) 10736 INIT_LIST_HEAD(&net->dev_base_head); 10737 10738 net->dev_name_head = netdev_create_hash(); 10739 if (net->dev_name_head == NULL) 10740 goto err_name; 10741 10742 net->dev_index_head = netdev_create_hash(); 10743 if (net->dev_index_head == NULL) 10744 goto err_idx; 10745 10746 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain); 10747 10748 return 0; 10749 10750 err_idx: 10751 kfree(net->dev_name_head); 10752 err_name: 10753 return -ENOMEM; 10754 } 10755 10756 /** 10757 * netdev_drivername - network driver for the device 10758 * @dev: network device 10759 * 10760 * Determine network driver for device. 10761 */ 10762 const char *netdev_drivername(const struct net_device *dev) 10763 { 10764 const struct device_driver *driver; 10765 const struct device *parent; 10766 const char *empty = ""; 10767 10768 parent = dev->dev.parent; 10769 if (!parent) 10770 return empty; 10771 10772 driver = parent->driver; 10773 if (driver && driver->name) 10774 return driver->name; 10775 return empty; 10776 } 10777 10778 static void __netdev_printk(const char *level, const struct net_device *dev, 10779 struct va_format *vaf) 10780 { 10781 if (dev && dev->dev.parent) { 10782 dev_printk_emit(level[1] - '0', 10783 dev->dev.parent, 10784 "%s %s %s%s: %pV", 10785 dev_driver_string(dev->dev.parent), 10786 dev_name(dev->dev.parent), 10787 netdev_name(dev), netdev_reg_state(dev), 10788 vaf); 10789 } else if (dev) { 10790 printk("%s%s%s: %pV", 10791 level, netdev_name(dev), netdev_reg_state(dev), vaf); 10792 } else { 10793 printk("%s(NULL net_device): %pV", level, vaf); 10794 } 10795 } 10796 10797 void netdev_printk(const char *level, const struct net_device *dev, 10798 const char *format, ...) 10799 { 10800 struct va_format vaf; 10801 va_list args; 10802 10803 va_start(args, format); 10804 10805 vaf.fmt = format; 10806 vaf.va = &args; 10807 10808 __netdev_printk(level, dev, &vaf); 10809 10810 va_end(args); 10811 } 10812 EXPORT_SYMBOL(netdev_printk); 10813 10814 #define define_netdev_printk_level(func, level) \ 10815 void func(const struct net_device *dev, const char *fmt, ...) \ 10816 { \ 10817 struct va_format vaf; \ 10818 va_list args; \ 10819 \ 10820 va_start(args, fmt); \ 10821 \ 10822 vaf.fmt = fmt; \ 10823 vaf.va = &args; \ 10824 \ 10825 __netdev_printk(level, dev, &vaf); \ 10826 \ 10827 va_end(args); \ 10828 } \ 10829 EXPORT_SYMBOL(func); 10830 10831 define_netdev_printk_level(netdev_emerg, KERN_EMERG); 10832 define_netdev_printk_level(netdev_alert, KERN_ALERT); 10833 define_netdev_printk_level(netdev_crit, KERN_CRIT); 10834 define_netdev_printk_level(netdev_err, KERN_ERR); 10835 define_netdev_printk_level(netdev_warn, KERN_WARNING); 10836 define_netdev_printk_level(netdev_notice, KERN_NOTICE); 10837 define_netdev_printk_level(netdev_info, KERN_INFO); 10838 10839 static void __net_exit netdev_exit(struct net *net) 10840 { 10841 kfree(net->dev_name_head); 10842 kfree(net->dev_index_head); 10843 if (net != &init_net) 10844 WARN_ON_ONCE(!list_empty(&net->dev_base_head)); 10845 } 10846 10847 static struct pernet_operations __net_initdata netdev_net_ops = { 10848 .init = netdev_init, 10849 .exit = netdev_exit, 10850 }; 10851 10852 static void __net_exit default_device_exit(struct net *net) 10853 { 10854 struct net_device *dev, *aux; 10855 /* 10856 * Push all migratable network devices back to the 10857 * initial network namespace 10858 */ 10859 rtnl_lock(); 10860 for_each_netdev_safe(net, dev, aux) { 10861 int err; 10862 char fb_name[IFNAMSIZ]; 10863 10864 /* Ignore unmoveable devices (i.e. loopback) */ 10865 if (dev->features & NETIF_F_NETNS_LOCAL) 10866 continue; 10867 10868 /* Leave virtual devices for the generic cleanup */ 10869 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund) 10870 continue; 10871 10872 /* Push remaining network devices to init_net */ 10873 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); 10874 if (netdev_name_in_use(&init_net, fb_name)) 10875 snprintf(fb_name, IFNAMSIZ, "dev%%d"); 10876 err = dev_change_net_namespace(dev, &init_net, fb_name); 10877 if (err) { 10878 pr_emerg("%s: failed to move %s to init_net: %d\n", 10879 __func__, dev->name, err); 10880 BUG(); 10881 } 10882 } 10883 rtnl_unlock(); 10884 } 10885 10886 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list) 10887 { 10888 /* Return with the rtnl_lock held when there are no network 10889 * devices unregistering in any network namespace in net_list. 10890 */ 10891 struct net *net; 10892 bool unregistering; 10893 DEFINE_WAIT_FUNC(wait, woken_wake_function); 10894 10895 add_wait_queue(&netdev_unregistering_wq, &wait); 10896 for (;;) { 10897 unregistering = false; 10898 rtnl_lock(); 10899 list_for_each_entry(net, net_list, exit_list) { 10900 if (net->dev_unreg_count > 0) { 10901 unregistering = true; 10902 break; 10903 } 10904 } 10905 if (!unregistering) 10906 break; 10907 __rtnl_unlock(); 10908 10909 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); 10910 } 10911 remove_wait_queue(&netdev_unregistering_wq, &wait); 10912 } 10913 10914 static void __net_exit default_device_exit_batch(struct list_head *net_list) 10915 { 10916 /* At exit all network devices most be removed from a network 10917 * namespace. Do this in the reverse order of registration. 10918 * Do this across as many network namespaces as possible to 10919 * improve batching efficiency. 10920 */ 10921 struct net_device *dev; 10922 struct net *net; 10923 LIST_HEAD(dev_kill_list); 10924 10925 /* To prevent network device cleanup code from dereferencing 10926 * loopback devices or network devices that have been freed 10927 * wait here for all pending unregistrations to complete, 10928 * before unregistring the loopback device and allowing the 10929 * network namespace be freed. 10930 * 10931 * The netdev todo list containing all network devices 10932 * unregistrations that happen in default_device_exit_batch 10933 * will run in the rtnl_unlock() at the end of 10934 * default_device_exit_batch. 10935 */ 10936 rtnl_lock_unregistering(net_list); 10937 list_for_each_entry(net, net_list, exit_list) { 10938 for_each_netdev_reverse(net, dev) { 10939 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) 10940 dev->rtnl_link_ops->dellink(dev, &dev_kill_list); 10941 else 10942 unregister_netdevice_queue(dev, &dev_kill_list); 10943 } 10944 } 10945 unregister_netdevice_many(&dev_kill_list); 10946 rtnl_unlock(); 10947 } 10948 10949 static struct pernet_operations __net_initdata default_device_ops = { 10950 .exit = default_device_exit, 10951 .exit_batch = default_device_exit_batch, 10952 }; 10953 10954 /* 10955 * Initialize the DEV module. At boot time this walks the device list and 10956 * unhooks any devices that fail to initialise (normally hardware not 10957 * present) and leaves us with a valid list of present and active devices. 10958 * 10959 */ 10960 10961 /* 10962 * This is called single threaded during boot, so no need 10963 * to take the rtnl semaphore. 10964 */ 10965 static int __init net_dev_init(void) 10966 { 10967 int i, rc = -ENOMEM; 10968 10969 BUG_ON(!dev_boot_phase); 10970 10971 if (dev_proc_init()) 10972 goto out; 10973 10974 if (netdev_kobject_init()) 10975 goto out; 10976 10977 INIT_LIST_HEAD(&ptype_all); 10978 for (i = 0; i < PTYPE_HASH_SIZE; i++) 10979 INIT_LIST_HEAD(&ptype_base[i]); 10980 10981 if (register_pernet_subsys(&netdev_net_ops)) 10982 goto out; 10983 10984 /* 10985 * Initialise the packet receive queues. 10986 */ 10987 10988 for_each_possible_cpu(i) { 10989 struct work_struct *flush = per_cpu_ptr(&flush_works, i); 10990 struct softnet_data *sd = &per_cpu(softnet_data, i); 10991 10992 INIT_WORK(flush, flush_backlog); 10993 10994 skb_queue_head_init(&sd->input_pkt_queue); 10995 skb_queue_head_init(&sd->process_queue); 10996 #ifdef CONFIG_XFRM_OFFLOAD 10997 skb_queue_head_init(&sd->xfrm_backlog); 10998 #endif 10999 INIT_LIST_HEAD(&sd->poll_list); 11000 sd->output_queue_tailp = &sd->output_queue; 11001 #ifdef CONFIG_RPS 11002 INIT_CSD(&sd->csd, rps_trigger_softirq, sd); 11003 sd->cpu = i; 11004 #endif 11005 11006 init_gro_hash(&sd->backlog); 11007 sd->backlog.poll = process_backlog; 11008 sd->backlog.weight = weight_p; 11009 } 11010 11011 dev_boot_phase = 0; 11012 11013 /* The loopback device is special if any other network devices 11014 * is present in a network namespace the loopback device must 11015 * be present. Since we now dynamically allocate and free the 11016 * loopback device ensure this invariant is maintained by 11017 * keeping the loopback device as the first device on the 11018 * list of network devices. Ensuring the loopback devices 11019 * is the first device that appears and the last network device 11020 * that disappears. 11021 */ 11022 if (register_pernet_device(&loopback_net_ops)) 11023 goto out; 11024 11025 if (register_pernet_device(&default_device_ops)) 11026 goto out; 11027 11028 open_softirq(NET_TX_SOFTIRQ, net_tx_action); 11029 open_softirq(NET_RX_SOFTIRQ, net_rx_action); 11030 11031 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", 11032 NULL, dev_cpu_dead); 11033 WARN_ON(rc < 0); 11034 rc = 0; 11035 out: 11036 return rc; 11037 } 11038 11039 subsys_initcall(net_dev_init); 11040