1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * NET3 Protocol independent device support routines. 4 * 5 * Derived from the non IP parts of dev.c 1.0.19 6 * Authors: Ross Biro 7 * Fred N. van Kempen, <[email protected]> 8 * Mark Evans, <[email protected]> 9 * 10 * Additional Authors: 11 * Florian la Roche <[email protected]> 12 * Alan Cox <[email protected]> 13 * David Hinds <[email protected]> 14 * Alexey Kuznetsov <[email protected]> 15 * Adam Sulmicki <[email protected]> 16 * Pekka Riikonen <[email protected]> 17 * 18 * Changes: 19 * D.J. Barrow : Fixed bug where dev->refcnt gets set 20 * to 2 if register_netdev gets called 21 * before net_dev_init & also removed a 22 * few lines of code in the process. 23 * Alan Cox : device private ioctl copies fields back. 24 * Alan Cox : Transmit queue code does relevant 25 * stunts to keep the queue safe. 26 * Alan Cox : Fixed double lock. 27 * Alan Cox : Fixed promisc NULL pointer trap 28 * ???????? : Support the full private ioctl range 29 * Alan Cox : Moved ioctl permission check into 30 * drivers 31 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI 32 * Alan Cox : 100 backlog just doesn't cut it when 33 * you start doing multicast video 8) 34 * Alan Cox : Rewrote net_bh and list manager. 35 * Alan Cox : Fix ETH_P_ALL echoback lengths. 36 * Alan Cox : Took out transmit every packet pass 37 * Saved a few bytes in the ioctl handler 38 * Alan Cox : Network driver sets packet type before 39 * calling netif_rx. Saves a function 40 * call a packet. 41 * Alan Cox : Hashed net_bh() 42 * Richard Kooijman: Timestamp fixes. 43 * Alan Cox : Wrong field in SIOCGIFDSTADDR 44 * Alan Cox : Device lock protection. 45 * Alan Cox : Fixed nasty side effect of device close 46 * changes. 47 * Rudi Cilibrasi : Pass the right thing to 48 * set_mac_address() 49 * Dave Miller : 32bit quantity for the device lock to 50 * make it work out on a Sparc. 51 * Bjorn Ekwall : Added KERNELD hack. 52 * Alan Cox : Cleaned up the backlog initialise. 53 * Craig Metz : SIOCGIFCONF fix if space for under 54 * 1 device. 55 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there 56 * is no device open function. 57 * Andi Kleen : Fix error reporting for SIOCGIFCONF 58 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF 59 * Cyrus Durgin : Cleaned for KMOD 60 * Adam Sulmicki : Bug Fix : Network Device Unload 61 * A network device unload needs to purge 62 * the backlog queue. 63 * Paul Rusty Russell : SIOCSIFNAME 64 * Pekka Riikonen : Netdev boot-time settings code 65 * Andrew Morton : Make unregister_netdevice wait 66 * indefinitely on dev->refcnt 67 * J Hadi Salim : - Backlog queue sampling 68 * - netif_rx() feedback 69 */ 70 71 #include <linux/uaccess.h> 72 #include <linux/bitmap.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 <net/tcx.h> 111 #include <linux/highmem.h> 112 #include <linux/init.h> 113 #include <linux/module.h> 114 #include <linux/netpoll.h> 115 #include <linux/rcupdate.h> 116 #include <linux/delay.h> 117 #include <net/iw_handler.h> 118 #include <asm/current.h> 119 #include <linux/audit.h> 120 #include <linux/dmaengine.h> 121 #include <linux/err.h> 122 #include <linux/ctype.h> 123 #include <linux/if_arp.h> 124 #include <linux/if_vlan.h> 125 #include <linux/ip.h> 126 #include <net/ip.h> 127 #include <net/mpls.h> 128 #include <linux/ipv6.h> 129 #include <linux/in.h> 130 #include <linux/jhash.h> 131 #include <linux/random.h> 132 #include <trace/events/napi.h> 133 #include <trace/events/net.h> 134 #include <trace/events/skb.h> 135 #include <trace/events/qdisc.h> 136 #include <trace/events/xdp.h> 137 #include <linux/inetdevice.h> 138 #include <linux/cpu_rmap.h> 139 #include <linux/static_key.h> 140 #include <linux/hashtable.h> 141 #include <linux/vmalloc.h> 142 #include <linux/if_macvlan.h> 143 #include <linux/errqueue.h> 144 #include <linux/hrtimer.h> 145 #include <linux/netfilter_netdev.h> 146 #include <linux/crash_dump.h> 147 #include <linux/sctp.h> 148 #include <net/udp_tunnel.h> 149 #include <linux/net_namespace.h> 150 #include <linux/indirect_call_wrapper.h> 151 #include <net/devlink.h> 152 #include <linux/pm_runtime.h> 153 #include <linux/prandom.h> 154 #include <linux/once_lite.h> 155 #include <net/netdev_rx_queue.h> 156 #include <net/page_pool/types.h> 157 #include <net/page_pool/helpers.h> 158 159 #include "dev.h" 160 #include "net-sysfs.h" 161 162 static DEFINE_SPINLOCK(ptype_lock); 163 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; 164 struct list_head ptype_all __read_mostly; /* Taps */ 165 166 static int netif_rx_internal(struct sk_buff *skb); 167 static int call_netdevice_notifiers_extack(unsigned long val, 168 struct net_device *dev, 169 struct netlink_ext_ack *extack); 170 171 static DEFINE_MUTEX(ifalias_mutex); 172 173 /* protects napi_hash addition/deletion and napi_gen_id */ 174 static DEFINE_SPINLOCK(napi_hash_lock); 175 176 static unsigned int napi_gen_id = NR_CPUS; 177 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8); 178 179 static DECLARE_RWSEM(devnet_rename_sem); 180 181 static inline void dev_base_seq_inc(struct net *net) 182 { 183 while (++net->dev_base_seq == 0) 184 ; 185 } 186 187 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) 188 { 189 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ)); 190 191 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; 192 } 193 194 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) 195 { 196 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; 197 } 198 199 static inline void rps_lock_irqsave(struct softnet_data *sd, 200 unsigned long *flags) 201 { 202 if (IS_ENABLED(CONFIG_RPS)) 203 spin_lock_irqsave(&sd->input_pkt_queue.lock, *flags); 204 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 205 local_irq_save(*flags); 206 } 207 208 static inline void rps_lock_irq_disable(struct softnet_data *sd) 209 { 210 if (IS_ENABLED(CONFIG_RPS)) 211 spin_lock_irq(&sd->input_pkt_queue.lock); 212 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 213 local_irq_disable(); 214 } 215 216 static inline void rps_unlock_irq_restore(struct softnet_data *sd, 217 unsigned long *flags) 218 { 219 if (IS_ENABLED(CONFIG_RPS)) 220 spin_unlock_irqrestore(&sd->input_pkt_queue.lock, *flags); 221 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 222 local_irq_restore(*flags); 223 } 224 225 static inline void rps_unlock_irq_enable(struct softnet_data *sd) 226 { 227 if (IS_ENABLED(CONFIG_RPS)) 228 spin_unlock_irq(&sd->input_pkt_queue.lock); 229 else if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 230 local_irq_enable(); 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_rcu(&name_node->list, &dev->name_node->list); 320 321 return 0; 322 } 323 324 static void netdev_name_node_alt_free(struct rcu_head *head) 325 { 326 struct netdev_name_node *name_node = 327 container_of(head, struct netdev_name_node, rcu); 328 329 kfree(name_node->name); 330 netdev_name_node_free(name_node); 331 } 332 333 static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node) 334 { 335 netdev_name_node_del(name_node); 336 list_del(&name_node->list); 337 call_rcu(&name_node->rcu, netdev_name_node_alt_free); 338 } 339 340 int netdev_name_node_alt_destroy(struct net_device *dev, const char *name) 341 { 342 struct netdev_name_node *name_node; 343 struct net *net = dev_net(dev); 344 345 name_node = netdev_name_node_lookup(net, name); 346 if (!name_node) 347 return -ENOENT; 348 /* lookup might have found our primary name or a name belonging 349 * to another device. 350 */ 351 if (name_node == dev->name_node || name_node->dev != dev) 352 return -EINVAL; 353 354 __netdev_name_node_alt_destroy(name_node); 355 return 0; 356 } 357 358 static void netdev_name_node_alt_flush(struct net_device *dev) 359 { 360 struct netdev_name_node *name_node, *tmp; 361 362 list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list) { 363 list_del(&name_node->list); 364 netdev_name_node_alt_free(&name_node->rcu); 365 } 366 } 367 368 /* Device list insertion */ 369 static void list_netdevice(struct net_device *dev) 370 { 371 struct netdev_name_node *name_node; 372 struct net *net = dev_net(dev); 373 374 ASSERT_RTNL(); 375 376 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); 377 netdev_name_node_add(net, dev->name_node); 378 hlist_add_head_rcu(&dev->index_hlist, 379 dev_index_hash(net, dev->ifindex)); 380 381 netdev_for_each_altname(dev, name_node) 382 netdev_name_node_add(net, name_node); 383 384 /* We reserved the ifindex, this can't fail */ 385 WARN_ON(xa_store(&net->dev_by_index, dev->ifindex, dev, GFP_KERNEL)); 386 387 dev_base_seq_inc(net); 388 } 389 390 /* Device list removal 391 * caller must respect a RCU grace period before freeing/reusing dev 392 */ 393 static void unlist_netdevice(struct net_device *dev) 394 { 395 struct netdev_name_node *name_node; 396 struct net *net = dev_net(dev); 397 398 ASSERT_RTNL(); 399 400 xa_erase(&net->dev_by_index, dev->ifindex); 401 402 netdev_for_each_altname(dev, name_node) 403 netdev_name_node_del(name_node); 404 405 /* Unlink dev from the device chain */ 406 list_del_rcu(&dev->dev_list); 407 netdev_name_node_del(dev->name_node); 408 hlist_del_rcu(&dev->index_hlist); 409 410 dev_base_seq_inc(dev_net(dev)); 411 } 412 413 /* 414 * Our notifier list 415 */ 416 417 static RAW_NOTIFIER_HEAD(netdev_chain); 418 419 /* 420 * Device drivers call our routines to queue packets here. We empty the 421 * queue in the local softnet handler. 422 */ 423 424 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); 425 EXPORT_PER_CPU_SYMBOL(softnet_data); 426 427 /* Page_pool has a lockless array/stack to alloc/recycle pages. 428 * PP consumers must pay attention to run APIs in the appropriate context 429 * (e.g. NAPI context). 430 */ 431 static DEFINE_PER_CPU_ALIGNED(struct page_pool *, system_page_pool); 432 433 #ifdef CONFIG_LOCKDEP 434 /* 435 * register_netdevice() inits txq->_xmit_lock and sets lockdep class 436 * according to dev->type 437 */ 438 static const unsigned short netdev_lock_type[] = { 439 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, 440 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, 441 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, 442 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, 443 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, 444 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, 445 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, 446 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, 447 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, 448 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, 449 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, 450 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, 451 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, 452 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, 453 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; 454 455 static const char *const netdev_lock_name[] = { 456 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", 457 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", 458 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", 459 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", 460 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", 461 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", 462 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", 463 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", 464 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", 465 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", 466 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", 467 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", 468 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", 469 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", 470 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; 471 472 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; 473 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; 474 475 static inline unsigned short netdev_lock_pos(unsigned short dev_type) 476 { 477 int i; 478 479 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) 480 if (netdev_lock_type[i] == dev_type) 481 return i; 482 /* the last key is used by default */ 483 return ARRAY_SIZE(netdev_lock_type) - 1; 484 } 485 486 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 487 unsigned short dev_type) 488 { 489 int i; 490 491 i = netdev_lock_pos(dev_type); 492 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], 493 netdev_lock_name[i]); 494 } 495 496 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 497 { 498 int i; 499 500 i = netdev_lock_pos(dev->type); 501 lockdep_set_class_and_name(&dev->addr_list_lock, 502 &netdev_addr_lock_key[i], 503 netdev_lock_name[i]); 504 } 505 #else 506 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, 507 unsigned short dev_type) 508 { 509 } 510 511 static inline void netdev_set_addr_lockdep_class(struct net_device *dev) 512 { 513 } 514 #endif 515 516 /******************************************************************************* 517 * 518 * Protocol management and registration routines 519 * 520 *******************************************************************************/ 521 522 523 /* 524 * Add a protocol ID to the list. Now that the input handler is 525 * smarter we can dispense with all the messy stuff that used to be 526 * here. 527 * 528 * BEWARE!!! Protocol handlers, mangling input packets, 529 * MUST BE last in hash buckets and checking protocol handlers 530 * MUST start from promiscuous ptype_all chain in net_bh. 531 * It is true now, do not change it. 532 * Explanation follows: if protocol handler, mangling packet, will 533 * be the first on list, it is not able to sense, that packet 534 * is cloned and should be copied-on-write, so that it will 535 * change it and subsequent readers will get broken packet. 536 * --ANK (980803) 537 */ 538 539 static inline struct list_head *ptype_head(const struct packet_type *pt) 540 { 541 if (pt->type == htons(ETH_P_ALL)) 542 return pt->dev ? &pt->dev->ptype_all : &ptype_all; 543 else 544 return pt->dev ? &pt->dev->ptype_specific : 545 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; 546 } 547 548 /** 549 * dev_add_pack - add packet handler 550 * @pt: packet type declaration 551 * 552 * Add a protocol handler to the networking stack. The passed &packet_type 553 * is linked into kernel lists and may not be freed until it has been 554 * removed from the kernel lists. 555 * 556 * This call does not sleep therefore it can not 557 * guarantee all CPU's that are in middle of receiving packets 558 * will see the new packet type (until the next received packet). 559 */ 560 561 void dev_add_pack(struct packet_type *pt) 562 { 563 struct list_head *head = ptype_head(pt); 564 565 spin_lock(&ptype_lock); 566 list_add_rcu(&pt->list, head); 567 spin_unlock(&ptype_lock); 568 } 569 EXPORT_SYMBOL(dev_add_pack); 570 571 /** 572 * __dev_remove_pack - remove packet handler 573 * @pt: packet type declaration 574 * 575 * Remove a protocol handler that was previously added to the kernel 576 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 577 * from the kernel lists and can be freed or reused once this function 578 * returns. 579 * 580 * The packet type might still be in use by receivers 581 * and must not be freed until after all the CPU's have gone 582 * through a quiescent state. 583 */ 584 void __dev_remove_pack(struct packet_type *pt) 585 { 586 struct list_head *head = ptype_head(pt); 587 struct packet_type *pt1; 588 589 spin_lock(&ptype_lock); 590 591 list_for_each_entry(pt1, head, list) { 592 if (pt == pt1) { 593 list_del_rcu(&pt->list); 594 goto out; 595 } 596 } 597 598 pr_warn("dev_remove_pack: %p not found\n", pt); 599 out: 600 spin_unlock(&ptype_lock); 601 } 602 EXPORT_SYMBOL(__dev_remove_pack); 603 604 /** 605 * dev_remove_pack - remove packet handler 606 * @pt: packet type declaration 607 * 608 * Remove a protocol handler that was previously added to the kernel 609 * protocol handlers by dev_add_pack(). The passed &packet_type is removed 610 * from the kernel lists and can be freed or reused once this function 611 * returns. 612 * 613 * This call sleeps to guarantee that no CPU is looking at the packet 614 * type after return. 615 */ 616 void dev_remove_pack(struct packet_type *pt) 617 { 618 __dev_remove_pack(pt); 619 620 synchronize_net(); 621 } 622 EXPORT_SYMBOL(dev_remove_pack); 623 624 625 /******************************************************************************* 626 * 627 * Device Interface Subroutines 628 * 629 *******************************************************************************/ 630 631 /** 632 * dev_get_iflink - get 'iflink' value of a interface 633 * @dev: targeted interface 634 * 635 * Indicates the ifindex the interface is linked to. 636 * Physical interfaces have the same 'ifindex' and 'iflink' values. 637 */ 638 639 int dev_get_iflink(const struct net_device *dev) 640 { 641 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink) 642 return dev->netdev_ops->ndo_get_iflink(dev); 643 644 return READ_ONCE(dev->ifindex); 645 } 646 EXPORT_SYMBOL(dev_get_iflink); 647 648 /** 649 * dev_fill_metadata_dst - Retrieve tunnel egress information. 650 * @dev: targeted interface 651 * @skb: The packet. 652 * 653 * For better visibility of tunnel traffic OVS needs to retrieve 654 * egress tunnel information for a packet. Following API allows 655 * user to get this info. 656 */ 657 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) 658 { 659 struct ip_tunnel_info *info; 660 661 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst) 662 return -EINVAL; 663 664 info = skb_tunnel_info_unclone(skb); 665 if (!info) 666 return -ENOMEM; 667 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX))) 668 return -EINVAL; 669 670 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb); 671 } 672 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst); 673 674 static struct net_device_path *dev_fwd_path(struct net_device_path_stack *stack) 675 { 676 int k = stack->num_paths++; 677 678 if (WARN_ON_ONCE(k >= NET_DEVICE_PATH_STACK_MAX)) 679 return NULL; 680 681 return &stack->path[k]; 682 } 683 684 int dev_fill_forward_path(const struct net_device *dev, const u8 *daddr, 685 struct net_device_path_stack *stack) 686 { 687 const struct net_device *last_dev; 688 struct net_device_path_ctx ctx = { 689 .dev = dev, 690 }; 691 struct net_device_path *path; 692 int ret = 0; 693 694 memcpy(ctx.daddr, daddr, sizeof(ctx.daddr)); 695 stack->num_paths = 0; 696 while (ctx.dev && ctx.dev->netdev_ops->ndo_fill_forward_path) { 697 last_dev = ctx.dev; 698 path = dev_fwd_path(stack); 699 if (!path) 700 return -1; 701 702 memset(path, 0, sizeof(struct net_device_path)); 703 ret = ctx.dev->netdev_ops->ndo_fill_forward_path(&ctx, path); 704 if (ret < 0) 705 return -1; 706 707 if (WARN_ON_ONCE(last_dev == ctx.dev)) 708 return -1; 709 } 710 711 if (!ctx.dev) 712 return ret; 713 714 path = dev_fwd_path(stack); 715 if (!path) 716 return -1; 717 path->type = DEV_PATH_ETHERNET; 718 path->dev = ctx.dev; 719 720 return ret; 721 } 722 EXPORT_SYMBOL_GPL(dev_fill_forward_path); 723 724 /** 725 * __dev_get_by_name - find a device by its name 726 * @net: the applicable net namespace 727 * @name: name to find 728 * 729 * Find an interface by name. Must be called under RTNL semaphore. 730 * If the name is found a pointer to the device is returned. 731 * If the name is not found then %NULL is returned. The 732 * reference counters are not incremented so the caller must be 733 * careful with locks. 734 */ 735 736 struct net_device *__dev_get_by_name(struct net *net, const char *name) 737 { 738 struct netdev_name_node *node_name; 739 740 node_name = netdev_name_node_lookup(net, name); 741 return node_name ? node_name->dev : NULL; 742 } 743 EXPORT_SYMBOL(__dev_get_by_name); 744 745 /** 746 * dev_get_by_name_rcu - find a device by its name 747 * @net: the applicable net namespace 748 * @name: name to find 749 * 750 * Find an interface by name. 751 * If the name is found a pointer to the device is returned. 752 * If the name is not found then %NULL is returned. 753 * The reference counters are not incremented so the caller must be 754 * careful with locks. The caller must hold RCU lock. 755 */ 756 757 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) 758 { 759 struct netdev_name_node *node_name; 760 761 node_name = netdev_name_node_lookup_rcu(net, name); 762 return node_name ? node_name->dev : NULL; 763 } 764 EXPORT_SYMBOL(dev_get_by_name_rcu); 765 766 /* Deprecated for new users, call netdev_get_by_name() instead */ 767 struct net_device *dev_get_by_name(struct net *net, const char *name) 768 { 769 struct net_device *dev; 770 771 rcu_read_lock(); 772 dev = dev_get_by_name_rcu(net, name); 773 dev_hold(dev); 774 rcu_read_unlock(); 775 return dev; 776 } 777 EXPORT_SYMBOL(dev_get_by_name); 778 779 /** 780 * netdev_get_by_name() - find a device by its name 781 * @net: the applicable net namespace 782 * @name: name to find 783 * @tracker: tracking object for the acquired reference 784 * @gfp: allocation flags for the tracker 785 * 786 * Find an interface by name. This can be called from any 787 * context and does its own locking. The returned handle has 788 * the usage count incremented and the caller must use netdev_put() to 789 * release it when it is no longer needed. %NULL is returned if no 790 * matching device is found. 791 */ 792 struct net_device *netdev_get_by_name(struct net *net, const char *name, 793 netdevice_tracker *tracker, gfp_t gfp) 794 { 795 struct net_device *dev; 796 797 dev = dev_get_by_name(net, name); 798 if (dev) 799 netdev_tracker_alloc(dev, tracker, gfp); 800 return dev; 801 } 802 EXPORT_SYMBOL(netdev_get_by_name); 803 804 /** 805 * __dev_get_by_index - find a device by its ifindex 806 * @net: the applicable net namespace 807 * @ifindex: index of device 808 * 809 * Search for an interface by index. Returns %NULL if the device 810 * is not found or a pointer to the device. The device has not 811 * had its reference counter increased so the caller must be careful 812 * about locking. The caller must hold the RTNL semaphore. 813 */ 814 815 struct net_device *__dev_get_by_index(struct net *net, int ifindex) 816 { 817 struct net_device *dev; 818 struct hlist_head *head = dev_index_hash(net, ifindex); 819 820 hlist_for_each_entry(dev, head, index_hlist) 821 if (dev->ifindex == ifindex) 822 return dev; 823 824 return NULL; 825 } 826 EXPORT_SYMBOL(__dev_get_by_index); 827 828 /** 829 * dev_get_by_index_rcu - find a device by its ifindex 830 * @net: the applicable net namespace 831 * @ifindex: index of device 832 * 833 * Search for an interface by index. Returns %NULL if the device 834 * is not found or a pointer to the device. The device has not 835 * had its reference counter increased so the caller must be careful 836 * about locking. The caller must hold RCU lock. 837 */ 838 839 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) 840 { 841 struct net_device *dev; 842 struct hlist_head *head = dev_index_hash(net, ifindex); 843 844 hlist_for_each_entry_rcu(dev, head, index_hlist) 845 if (dev->ifindex == ifindex) 846 return dev; 847 848 return NULL; 849 } 850 EXPORT_SYMBOL(dev_get_by_index_rcu); 851 852 /* Deprecated for new users, call netdev_get_by_index() instead */ 853 struct net_device *dev_get_by_index(struct net *net, int ifindex) 854 { 855 struct net_device *dev; 856 857 rcu_read_lock(); 858 dev = dev_get_by_index_rcu(net, ifindex); 859 dev_hold(dev); 860 rcu_read_unlock(); 861 return dev; 862 } 863 EXPORT_SYMBOL(dev_get_by_index); 864 865 /** 866 * netdev_get_by_index() - find a device by its ifindex 867 * @net: the applicable net namespace 868 * @ifindex: index of device 869 * @tracker: tracking object for the acquired reference 870 * @gfp: allocation flags for the tracker 871 * 872 * Search for an interface by index. Returns NULL if the device 873 * is not found or a pointer to the device. The device returned has 874 * had a reference added and the pointer is safe until the user calls 875 * netdev_put() to indicate they have finished with it. 876 */ 877 struct net_device *netdev_get_by_index(struct net *net, int ifindex, 878 netdevice_tracker *tracker, gfp_t gfp) 879 { 880 struct net_device *dev; 881 882 dev = dev_get_by_index(net, ifindex); 883 if (dev) 884 netdev_tracker_alloc(dev, tracker, gfp); 885 return dev; 886 } 887 EXPORT_SYMBOL(netdev_get_by_index); 888 889 /** 890 * dev_get_by_napi_id - find a device by napi_id 891 * @napi_id: ID of the NAPI struct 892 * 893 * Search for an interface by NAPI ID. Returns %NULL if the device 894 * is not found or a pointer to the device. The device has not had 895 * its reference counter increased so the caller must be careful 896 * about locking. The caller must hold RCU lock. 897 */ 898 899 struct net_device *dev_get_by_napi_id(unsigned int napi_id) 900 { 901 struct napi_struct *napi; 902 903 WARN_ON_ONCE(!rcu_read_lock_held()); 904 905 if (napi_id < MIN_NAPI_ID) 906 return NULL; 907 908 napi = napi_by_id(napi_id); 909 910 return napi ? napi->dev : NULL; 911 } 912 EXPORT_SYMBOL(dev_get_by_napi_id); 913 914 /** 915 * netdev_get_name - get a netdevice name, knowing its ifindex. 916 * @net: network namespace 917 * @name: a pointer to the buffer where the name will be stored. 918 * @ifindex: the ifindex of the interface to get the name from. 919 */ 920 int netdev_get_name(struct net *net, char *name, int ifindex) 921 { 922 struct net_device *dev; 923 int ret; 924 925 down_read(&devnet_rename_sem); 926 rcu_read_lock(); 927 928 dev = dev_get_by_index_rcu(net, ifindex); 929 if (!dev) { 930 ret = -ENODEV; 931 goto out; 932 } 933 934 strcpy(name, dev->name); 935 936 ret = 0; 937 out: 938 rcu_read_unlock(); 939 up_read(&devnet_rename_sem); 940 return ret; 941 } 942 943 /** 944 * dev_getbyhwaddr_rcu - find a device by its hardware address 945 * @net: the applicable net namespace 946 * @type: media type of device 947 * @ha: hardware address 948 * 949 * Search for an interface by MAC address. Returns NULL if the device 950 * is not found or a pointer to the device. 951 * The caller must hold RCU or RTNL. 952 * The returned device has not had its ref count increased 953 * and the caller must therefore be careful about locking 954 * 955 */ 956 957 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, 958 const char *ha) 959 { 960 struct net_device *dev; 961 962 for_each_netdev_rcu(net, dev) 963 if (dev->type == type && 964 !memcmp(dev->dev_addr, ha, dev->addr_len)) 965 return dev; 966 967 return NULL; 968 } 969 EXPORT_SYMBOL(dev_getbyhwaddr_rcu); 970 971 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) 972 { 973 struct net_device *dev, *ret = NULL; 974 975 rcu_read_lock(); 976 for_each_netdev_rcu(net, dev) 977 if (dev->type == type) { 978 dev_hold(dev); 979 ret = dev; 980 break; 981 } 982 rcu_read_unlock(); 983 return ret; 984 } 985 EXPORT_SYMBOL(dev_getfirstbyhwtype); 986 987 /** 988 * __dev_get_by_flags - find any device with given flags 989 * @net: the applicable net namespace 990 * @if_flags: IFF_* values 991 * @mask: bitmask of bits in if_flags to check 992 * 993 * Search for any interface with the given flags. Returns NULL if a device 994 * is not found or a pointer to the device. Must be called inside 995 * rtnl_lock(), and result refcount is unchanged. 996 */ 997 998 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags, 999 unsigned short mask) 1000 { 1001 struct net_device *dev, *ret; 1002 1003 ASSERT_RTNL(); 1004 1005 ret = NULL; 1006 for_each_netdev(net, dev) { 1007 if (((dev->flags ^ if_flags) & mask) == 0) { 1008 ret = dev; 1009 break; 1010 } 1011 } 1012 return ret; 1013 } 1014 EXPORT_SYMBOL(__dev_get_by_flags); 1015 1016 /** 1017 * dev_valid_name - check if name is okay for network device 1018 * @name: name string 1019 * 1020 * Network device names need to be valid file names to 1021 * allow sysfs to work. We also disallow any kind of 1022 * whitespace. 1023 */ 1024 bool dev_valid_name(const char *name) 1025 { 1026 if (*name == '\0') 1027 return false; 1028 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ) 1029 return false; 1030 if (!strcmp(name, ".") || !strcmp(name, "..")) 1031 return false; 1032 1033 while (*name) { 1034 if (*name == '/' || *name == ':' || isspace(*name)) 1035 return false; 1036 name++; 1037 } 1038 return true; 1039 } 1040 EXPORT_SYMBOL(dev_valid_name); 1041 1042 /** 1043 * __dev_alloc_name - allocate a name for a device 1044 * @net: network namespace to allocate the device name in 1045 * @name: name format string 1046 * @res: result name string 1047 * 1048 * Passed a format string - eg "lt%d" it will try and find a suitable 1049 * id. It scans list of devices to build up a free map, then chooses 1050 * the first empty slot. The caller must hold the dev_base or rtnl lock 1051 * while allocating the name and adding the device in order to avoid 1052 * duplicates. 1053 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1054 * Returns the number of the unit assigned or a negative errno code. 1055 */ 1056 1057 static int __dev_alloc_name(struct net *net, const char *name, char *res) 1058 { 1059 int i = 0; 1060 const char *p; 1061 const int max_netdevices = 8*PAGE_SIZE; 1062 unsigned long *inuse; 1063 struct net_device *d; 1064 char buf[IFNAMSIZ]; 1065 1066 /* Verify the string as this thing may have come from the user. 1067 * There must be one "%d" and no other "%" characters. 1068 */ 1069 p = strchr(name, '%'); 1070 if (!p || p[1] != 'd' || strchr(p + 2, '%')) 1071 return -EINVAL; 1072 1073 /* Use one page as a bit array of possible slots */ 1074 inuse = bitmap_zalloc(max_netdevices, GFP_ATOMIC); 1075 if (!inuse) 1076 return -ENOMEM; 1077 1078 for_each_netdev(net, d) { 1079 struct netdev_name_node *name_node; 1080 1081 netdev_for_each_altname(d, name_node) { 1082 if (!sscanf(name_node->name, name, &i)) 1083 continue; 1084 if (i < 0 || i >= max_netdevices) 1085 continue; 1086 1087 /* avoid cases where sscanf is not exact inverse of printf */ 1088 snprintf(buf, IFNAMSIZ, name, i); 1089 if (!strncmp(buf, name_node->name, IFNAMSIZ)) 1090 __set_bit(i, inuse); 1091 } 1092 if (!sscanf(d->name, name, &i)) 1093 continue; 1094 if (i < 0 || i >= max_netdevices) 1095 continue; 1096 1097 /* avoid cases where sscanf is not exact inverse of printf */ 1098 snprintf(buf, IFNAMSIZ, name, i); 1099 if (!strncmp(buf, d->name, IFNAMSIZ)) 1100 __set_bit(i, inuse); 1101 } 1102 1103 i = find_first_zero_bit(inuse, max_netdevices); 1104 bitmap_free(inuse); 1105 if (i == max_netdevices) 1106 return -ENFILE; 1107 1108 /* 'res' and 'name' could overlap, use 'buf' as an intermediate buffer */ 1109 strscpy(buf, name, IFNAMSIZ); 1110 snprintf(res, IFNAMSIZ, buf, i); 1111 return i; 1112 } 1113 1114 /* Returns negative errno or allocated unit id (see __dev_alloc_name()) */ 1115 static int dev_prep_valid_name(struct net *net, struct net_device *dev, 1116 const char *want_name, char *out_name, 1117 int dup_errno) 1118 { 1119 if (!dev_valid_name(want_name)) 1120 return -EINVAL; 1121 1122 if (strchr(want_name, '%')) 1123 return __dev_alloc_name(net, want_name, out_name); 1124 1125 if (netdev_name_in_use(net, want_name)) 1126 return -dup_errno; 1127 if (out_name != want_name) 1128 strscpy(out_name, want_name, IFNAMSIZ); 1129 return 0; 1130 } 1131 1132 /** 1133 * dev_alloc_name - allocate a name for a device 1134 * @dev: device 1135 * @name: name format string 1136 * 1137 * Passed a format string - eg "lt%d" it will try and find a suitable 1138 * id. It scans list of devices to build up a free map, then chooses 1139 * the first empty slot. The caller must hold the dev_base or rtnl lock 1140 * while allocating the name and adding the device in order to avoid 1141 * duplicates. 1142 * Limited to bits_per_byte * page size devices (ie 32K on most platforms). 1143 * Returns the number of the unit assigned or a negative errno code. 1144 */ 1145 1146 int dev_alloc_name(struct net_device *dev, const char *name) 1147 { 1148 return dev_prep_valid_name(dev_net(dev), dev, name, dev->name, ENFILE); 1149 } 1150 EXPORT_SYMBOL(dev_alloc_name); 1151 1152 static int dev_get_valid_name(struct net *net, struct net_device *dev, 1153 const char *name) 1154 { 1155 int ret; 1156 1157 ret = dev_prep_valid_name(net, dev, name, dev->name, EEXIST); 1158 return ret < 0 ? ret : 0; 1159 } 1160 1161 /** 1162 * dev_change_name - change name of a device 1163 * @dev: device 1164 * @newname: name (or format string) must be at least IFNAMSIZ 1165 * 1166 * Change name of a device, can pass format strings "eth%d". 1167 * for wildcarding. 1168 */ 1169 int dev_change_name(struct net_device *dev, const char *newname) 1170 { 1171 unsigned char old_assign_type; 1172 char oldname[IFNAMSIZ]; 1173 int err = 0; 1174 int ret; 1175 struct net *net; 1176 1177 ASSERT_RTNL(); 1178 BUG_ON(!dev_net(dev)); 1179 1180 net = dev_net(dev); 1181 1182 down_write(&devnet_rename_sem); 1183 1184 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { 1185 up_write(&devnet_rename_sem); 1186 return 0; 1187 } 1188 1189 memcpy(oldname, dev->name, IFNAMSIZ); 1190 1191 err = dev_get_valid_name(net, dev, newname); 1192 if (err < 0) { 1193 up_write(&devnet_rename_sem); 1194 return err; 1195 } 1196 1197 if (oldname[0] && !strchr(oldname, '%')) 1198 netdev_info(dev, "renamed from %s%s\n", oldname, 1199 dev->flags & IFF_UP ? " (while UP)" : ""); 1200 1201 old_assign_type = dev->name_assign_type; 1202 WRITE_ONCE(dev->name_assign_type, NET_NAME_RENAMED); 1203 1204 rollback: 1205 ret = device_rename(&dev->dev, dev->name); 1206 if (ret) { 1207 memcpy(dev->name, oldname, IFNAMSIZ); 1208 WRITE_ONCE(dev->name_assign_type, old_assign_type); 1209 up_write(&devnet_rename_sem); 1210 return ret; 1211 } 1212 1213 up_write(&devnet_rename_sem); 1214 1215 netdev_adjacent_rename_links(dev, oldname); 1216 1217 netdev_name_node_del(dev->name_node); 1218 1219 synchronize_net(); 1220 1221 netdev_name_node_add(net, dev->name_node); 1222 1223 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); 1224 ret = notifier_to_errno(ret); 1225 1226 if (ret) { 1227 /* err >= 0 after dev_alloc_name() or stores the first errno */ 1228 if (err >= 0) { 1229 err = ret; 1230 down_write(&devnet_rename_sem); 1231 memcpy(dev->name, oldname, IFNAMSIZ); 1232 memcpy(oldname, newname, IFNAMSIZ); 1233 WRITE_ONCE(dev->name_assign_type, old_assign_type); 1234 old_assign_type = NET_NAME_RENAMED; 1235 goto rollback; 1236 } else { 1237 netdev_err(dev, "name change rollback failed: %d\n", 1238 ret); 1239 } 1240 } 1241 1242 return err; 1243 } 1244 1245 /** 1246 * dev_set_alias - change ifalias of a device 1247 * @dev: device 1248 * @alias: name up to IFALIASZ 1249 * @len: limit of bytes to copy from info 1250 * 1251 * Set ifalias for a device, 1252 */ 1253 int dev_set_alias(struct net_device *dev, const char *alias, size_t len) 1254 { 1255 struct dev_ifalias *new_alias = NULL; 1256 1257 if (len >= IFALIASZ) 1258 return -EINVAL; 1259 1260 if (len) { 1261 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL); 1262 if (!new_alias) 1263 return -ENOMEM; 1264 1265 memcpy(new_alias->ifalias, alias, len); 1266 new_alias->ifalias[len] = 0; 1267 } 1268 1269 mutex_lock(&ifalias_mutex); 1270 new_alias = rcu_replace_pointer(dev->ifalias, new_alias, 1271 mutex_is_locked(&ifalias_mutex)); 1272 mutex_unlock(&ifalias_mutex); 1273 1274 if (new_alias) 1275 kfree_rcu(new_alias, rcuhead); 1276 1277 return len; 1278 } 1279 EXPORT_SYMBOL(dev_set_alias); 1280 1281 /** 1282 * dev_get_alias - get ifalias of a device 1283 * @dev: device 1284 * @name: buffer to store name of ifalias 1285 * @len: size of buffer 1286 * 1287 * get ifalias for a device. Caller must make sure dev cannot go 1288 * away, e.g. rcu read lock or own a reference count to device. 1289 */ 1290 int dev_get_alias(const struct net_device *dev, char *name, size_t len) 1291 { 1292 const struct dev_ifalias *alias; 1293 int ret = 0; 1294 1295 rcu_read_lock(); 1296 alias = rcu_dereference(dev->ifalias); 1297 if (alias) 1298 ret = snprintf(name, len, "%s", alias->ifalias); 1299 rcu_read_unlock(); 1300 1301 return ret; 1302 } 1303 1304 /** 1305 * netdev_features_change - device changes features 1306 * @dev: device to cause notification 1307 * 1308 * Called to indicate a device has changed features. 1309 */ 1310 void netdev_features_change(struct net_device *dev) 1311 { 1312 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); 1313 } 1314 EXPORT_SYMBOL(netdev_features_change); 1315 1316 /** 1317 * netdev_state_change - device changes state 1318 * @dev: device to cause notification 1319 * 1320 * Called to indicate a device has changed state. This function calls 1321 * the notifier chains for netdev_chain and sends a NEWLINK message 1322 * to the routing socket. 1323 */ 1324 void netdev_state_change(struct net_device *dev) 1325 { 1326 if (dev->flags & IFF_UP) { 1327 struct netdev_notifier_change_info change_info = { 1328 .info.dev = dev, 1329 }; 1330 1331 call_netdevice_notifiers_info(NETDEV_CHANGE, 1332 &change_info.info); 1333 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL, 0, NULL); 1334 } 1335 } 1336 EXPORT_SYMBOL(netdev_state_change); 1337 1338 /** 1339 * __netdev_notify_peers - notify network peers about existence of @dev, 1340 * to be called when rtnl lock is already held. 1341 * @dev: network device 1342 * 1343 * Generate traffic such that interested network peers are aware of 1344 * @dev, such as by generating a gratuitous ARP. This may be used when 1345 * a device wants to inform the rest of the network about some sort of 1346 * reconfiguration such as a failover event or virtual machine 1347 * migration. 1348 */ 1349 void __netdev_notify_peers(struct net_device *dev) 1350 { 1351 ASSERT_RTNL(); 1352 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); 1353 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev); 1354 } 1355 EXPORT_SYMBOL(__netdev_notify_peers); 1356 1357 /** 1358 * netdev_notify_peers - notify network peers about existence of @dev 1359 * @dev: network device 1360 * 1361 * Generate traffic such that interested network peers are aware of 1362 * @dev, such as by generating a gratuitous ARP. This may be used when 1363 * a device wants to inform the rest of the network about some sort of 1364 * reconfiguration such as a failover event or virtual machine 1365 * migration. 1366 */ 1367 void netdev_notify_peers(struct net_device *dev) 1368 { 1369 rtnl_lock(); 1370 __netdev_notify_peers(dev); 1371 rtnl_unlock(); 1372 } 1373 EXPORT_SYMBOL(netdev_notify_peers); 1374 1375 static int napi_threaded_poll(void *data); 1376 1377 static int napi_kthread_create(struct napi_struct *n) 1378 { 1379 int err = 0; 1380 1381 /* Create and wake up the kthread once to put it in 1382 * TASK_INTERRUPTIBLE mode to avoid the blocked task 1383 * warning and work with loadavg. 1384 */ 1385 n->thread = kthread_run(napi_threaded_poll, n, "napi/%s-%d", 1386 n->dev->name, n->napi_id); 1387 if (IS_ERR(n->thread)) { 1388 err = PTR_ERR(n->thread); 1389 pr_err("kthread_run failed with err %d\n", err); 1390 n->thread = NULL; 1391 } 1392 1393 return err; 1394 } 1395 1396 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1397 { 1398 const struct net_device_ops *ops = dev->netdev_ops; 1399 int ret; 1400 1401 ASSERT_RTNL(); 1402 dev_addr_check(dev); 1403 1404 if (!netif_device_present(dev)) { 1405 /* may be detached because parent is runtime-suspended */ 1406 if (dev->dev.parent) 1407 pm_runtime_resume(dev->dev.parent); 1408 if (!netif_device_present(dev)) 1409 return -ENODEV; 1410 } 1411 1412 /* Block netpoll from trying to do any rx path servicing. 1413 * If we don't do this there is a chance ndo_poll_controller 1414 * or ndo_poll may be running while we open the device 1415 */ 1416 netpoll_poll_disable(dev); 1417 1418 ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack); 1419 ret = notifier_to_errno(ret); 1420 if (ret) 1421 return ret; 1422 1423 set_bit(__LINK_STATE_START, &dev->state); 1424 1425 if (ops->ndo_validate_addr) 1426 ret = ops->ndo_validate_addr(dev); 1427 1428 if (!ret && ops->ndo_open) 1429 ret = ops->ndo_open(dev); 1430 1431 netpoll_poll_enable(dev); 1432 1433 if (ret) 1434 clear_bit(__LINK_STATE_START, &dev->state); 1435 else { 1436 dev->flags |= IFF_UP; 1437 dev_set_rx_mode(dev); 1438 dev_activate(dev); 1439 add_device_randomness(dev->dev_addr, dev->addr_len); 1440 } 1441 1442 return ret; 1443 } 1444 1445 /** 1446 * dev_open - prepare an interface for use. 1447 * @dev: device to open 1448 * @extack: netlink extended ack 1449 * 1450 * Takes a device from down to up state. The device's private open 1451 * function is invoked and then the multicast lists are loaded. Finally 1452 * the device is moved into the up state and a %NETDEV_UP message is 1453 * sent to the netdev notifier chain. 1454 * 1455 * Calling this function on an active interface is a nop. On a failure 1456 * a negative errno code is returned. 1457 */ 1458 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack) 1459 { 1460 int ret; 1461 1462 if (dev->flags & IFF_UP) 1463 return 0; 1464 1465 ret = __dev_open(dev, extack); 1466 if (ret < 0) 1467 return ret; 1468 1469 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL); 1470 call_netdevice_notifiers(NETDEV_UP, dev); 1471 1472 return ret; 1473 } 1474 EXPORT_SYMBOL(dev_open); 1475 1476 static void __dev_close_many(struct list_head *head) 1477 { 1478 struct net_device *dev; 1479 1480 ASSERT_RTNL(); 1481 might_sleep(); 1482 1483 list_for_each_entry(dev, head, close_list) { 1484 /* Temporarily disable netpoll until the interface is down */ 1485 netpoll_poll_disable(dev); 1486 1487 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); 1488 1489 clear_bit(__LINK_STATE_START, &dev->state); 1490 1491 /* Synchronize to scheduled poll. We cannot touch poll list, it 1492 * can be even on different cpu. So just clear netif_running(). 1493 * 1494 * dev->stop() will invoke napi_disable() on all of it's 1495 * napi_struct instances on this device. 1496 */ 1497 smp_mb__after_atomic(); /* Commit netif_running(). */ 1498 } 1499 1500 dev_deactivate_many(head); 1501 1502 list_for_each_entry(dev, head, close_list) { 1503 const struct net_device_ops *ops = dev->netdev_ops; 1504 1505 /* 1506 * Call the device specific close. This cannot fail. 1507 * Only if device is UP 1508 * 1509 * We allow it to be called even after a DETACH hot-plug 1510 * event. 1511 */ 1512 if (ops->ndo_stop) 1513 ops->ndo_stop(dev); 1514 1515 dev->flags &= ~IFF_UP; 1516 netpoll_poll_enable(dev); 1517 } 1518 } 1519 1520 static void __dev_close(struct net_device *dev) 1521 { 1522 LIST_HEAD(single); 1523 1524 list_add(&dev->close_list, &single); 1525 __dev_close_many(&single); 1526 list_del(&single); 1527 } 1528 1529 void dev_close_many(struct list_head *head, bool unlink) 1530 { 1531 struct net_device *dev, *tmp; 1532 1533 /* Remove the devices that don't need to be closed */ 1534 list_for_each_entry_safe(dev, tmp, head, close_list) 1535 if (!(dev->flags & IFF_UP)) 1536 list_del_init(&dev->close_list); 1537 1538 __dev_close_many(head); 1539 1540 list_for_each_entry_safe(dev, tmp, head, close_list) { 1541 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL); 1542 call_netdevice_notifiers(NETDEV_DOWN, dev); 1543 if (unlink) 1544 list_del_init(&dev->close_list); 1545 } 1546 } 1547 EXPORT_SYMBOL(dev_close_many); 1548 1549 /** 1550 * dev_close - shutdown an interface. 1551 * @dev: device to shutdown 1552 * 1553 * This function moves an active device into down state. A 1554 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device 1555 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier 1556 * chain. 1557 */ 1558 void dev_close(struct net_device *dev) 1559 { 1560 if (dev->flags & IFF_UP) { 1561 LIST_HEAD(single); 1562 1563 list_add(&dev->close_list, &single); 1564 dev_close_many(&single, true); 1565 list_del(&single); 1566 } 1567 } 1568 EXPORT_SYMBOL(dev_close); 1569 1570 1571 /** 1572 * dev_disable_lro - disable Large Receive Offload on a device 1573 * @dev: device 1574 * 1575 * Disable Large Receive Offload (LRO) on a net device. Must be 1576 * called under RTNL. This is needed if received packets may be 1577 * forwarded to another interface. 1578 */ 1579 void dev_disable_lro(struct net_device *dev) 1580 { 1581 struct net_device *lower_dev; 1582 struct list_head *iter; 1583 1584 dev->wanted_features &= ~NETIF_F_LRO; 1585 netdev_update_features(dev); 1586 1587 if (unlikely(dev->features & NETIF_F_LRO)) 1588 netdev_WARN(dev, "failed to disable LRO!\n"); 1589 1590 netdev_for_each_lower_dev(dev, lower_dev, iter) 1591 dev_disable_lro(lower_dev); 1592 } 1593 EXPORT_SYMBOL(dev_disable_lro); 1594 1595 /** 1596 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device 1597 * @dev: device 1598 * 1599 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be 1600 * called under RTNL. This is needed if Generic XDP is installed on 1601 * the device. 1602 */ 1603 static void dev_disable_gro_hw(struct net_device *dev) 1604 { 1605 dev->wanted_features &= ~NETIF_F_GRO_HW; 1606 netdev_update_features(dev); 1607 1608 if (unlikely(dev->features & NETIF_F_GRO_HW)) 1609 netdev_WARN(dev, "failed to disable GRO_HW!\n"); 1610 } 1611 1612 const char *netdev_cmd_to_name(enum netdev_cmd cmd) 1613 { 1614 #define N(val) \ 1615 case NETDEV_##val: \ 1616 return "NETDEV_" __stringify(val); 1617 switch (cmd) { 1618 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER) 1619 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE) 1620 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE) 1621 N(POST_INIT) N(PRE_UNINIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) 1622 N(CHANGEUPPER) N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) 1623 N(BONDING_INFO) N(PRECHANGEUPPER) N(CHANGELOWERSTATE) 1624 N(UDP_TUNNEL_PUSH_INFO) N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN) 1625 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO) 1626 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO) 1627 N(PRE_CHANGEADDR) N(OFFLOAD_XSTATS_ENABLE) N(OFFLOAD_XSTATS_DISABLE) 1628 N(OFFLOAD_XSTATS_REPORT_USED) N(OFFLOAD_XSTATS_REPORT_DELTA) 1629 N(XDP_FEAT_CHANGE) 1630 } 1631 #undef N 1632 return "UNKNOWN_NETDEV_EVENT"; 1633 } 1634 EXPORT_SYMBOL_GPL(netdev_cmd_to_name); 1635 1636 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, 1637 struct net_device *dev) 1638 { 1639 struct netdev_notifier_info info = { 1640 .dev = dev, 1641 }; 1642 1643 return nb->notifier_call(nb, val, &info); 1644 } 1645 1646 static int call_netdevice_register_notifiers(struct notifier_block *nb, 1647 struct net_device *dev) 1648 { 1649 int err; 1650 1651 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); 1652 err = notifier_to_errno(err); 1653 if (err) 1654 return err; 1655 1656 if (!(dev->flags & IFF_UP)) 1657 return 0; 1658 1659 call_netdevice_notifier(nb, NETDEV_UP, dev); 1660 return 0; 1661 } 1662 1663 static void call_netdevice_unregister_notifiers(struct notifier_block *nb, 1664 struct net_device *dev) 1665 { 1666 if (dev->flags & IFF_UP) { 1667 call_netdevice_notifier(nb, NETDEV_GOING_DOWN, 1668 dev); 1669 call_netdevice_notifier(nb, NETDEV_DOWN, dev); 1670 } 1671 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); 1672 } 1673 1674 static int call_netdevice_register_net_notifiers(struct notifier_block *nb, 1675 struct net *net) 1676 { 1677 struct net_device *dev; 1678 int err; 1679 1680 for_each_netdev(net, dev) { 1681 err = call_netdevice_register_notifiers(nb, dev); 1682 if (err) 1683 goto rollback; 1684 } 1685 return 0; 1686 1687 rollback: 1688 for_each_netdev_continue_reverse(net, dev) 1689 call_netdevice_unregister_notifiers(nb, dev); 1690 return err; 1691 } 1692 1693 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb, 1694 struct net *net) 1695 { 1696 struct net_device *dev; 1697 1698 for_each_netdev(net, dev) 1699 call_netdevice_unregister_notifiers(nb, dev); 1700 } 1701 1702 static int dev_boot_phase = 1; 1703 1704 /** 1705 * register_netdevice_notifier - register a network notifier block 1706 * @nb: notifier 1707 * 1708 * Register a notifier to be called when network device events occur. 1709 * The notifier passed is linked into the kernel structures and must 1710 * not be reused until it has been unregistered. A negative errno code 1711 * is returned on a failure. 1712 * 1713 * When registered all registration and up events are replayed 1714 * to the new notifier to allow device to have a race free 1715 * view of the network device list. 1716 */ 1717 1718 int register_netdevice_notifier(struct notifier_block *nb) 1719 { 1720 struct net *net; 1721 int err; 1722 1723 /* Close race with setup_net() and cleanup_net() */ 1724 down_write(&pernet_ops_rwsem); 1725 rtnl_lock(); 1726 err = raw_notifier_chain_register(&netdev_chain, nb); 1727 if (err) 1728 goto unlock; 1729 if (dev_boot_phase) 1730 goto unlock; 1731 for_each_net(net) { 1732 err = call_netdevice_register_net_notifiers(nb, net); 1733 if (err) 1734 goto rollback; 1735 } 1736 1737 unlock: 1738 rtnl_unlock(); 1739 up_write(&pernet_ops_rwsem); 1740 return err; 1741 1742 rollback: 1743 for_each_net_continue_reverse(net) 1744 call_netdevice_unregister_net_notifiers(nb, net); 1745 1746 raw_notifier_chain_unregister(&netdev_chain, nb); 1747 goto unlock; 1748 } 1749 EXPORT_SYMBOL(register_netdevice_notifier); 1750 1751 /** 1752 * unregister_netdevice_notifier - unregister a network notifier block 1753 * @nb: notifier 1754 * 1755 * Unregister a notifier previously registered by 1756 * register_netdevice_notifier(). The notifier is unlinked into the 1757 * kernel structures and may then be reused. A negative errno code 1758 * is returned on a failure. 1759 * 1760 * After unregistering unregister and down device events are synthesized 1761 * for all devices on the device list to the removed notifier to remove 1762 * the need for special case cleanup code. 1763 */ 1764 1765 int unregister_netdevice_notifier(struct notifier_block *nb) 1766 { 1767 struct net *net; 1768 int err; 1769 1770 /* Close race with setup_net() and cleanup_net() */ 1771 down_write(&pernet_ops_rwsem); 1772 rtnl_lock(); 1773 err = raw_notifier_chain_unregister(&netdev_chain, nb); 1774 if (err) 1775 goto unlock; 1776 1777 for_each_net(net) 1778 call_netdevice_unregister_net_notifiers(nb, net); 1779 1780 unlock: 1781 rtnl_unlock(); 1782 up_write(&pernet_ops_rwsem); 1783 return err; 1784 } 1785 EXPORT_SYMBOL(unregister_netdevice_notifier); 1786 1787 static int __register_netdevice_notifier_net(struct net *net, 1788 struct notifier_block *nb, 1789 bool ignore_call_fail) 1790 { 1791 int err; 1792 1793 err = raw_notifier_chain_register(&net->netdev_chain, nb); 1794 if (err) 1795 return err; 1796 if (dev_boot_phase) 1797 return 0; 1798 1799 err = call_netdevice_register_net_notifiers(nb, net); 1800 if (err && !ignore_call_fail) 1801 goto chain_unregister; 1802 1803 return 0; 1804 1805 chain_unregister: 1806 raw_notifier_chain_unregister(&net->netdev_chain, nb); 1807 return err; 1808 } 1809 1810 static int __unregister_netdevice_notifier_net(struct net *net, 1811 struct notifier_block *nb) 1812 { 1813 int err; 1814 1815 err = raw_notifier_chain_unregister(&net->netdev_chain, nb); 1816 if (err) 1817 return err; 1818 1819 call_netdevice_unregister_net_notifiers(nb, net); 1820 return 0; 1821 } 1822 1823 /** 1824 * register_netdevice_notifier_net - register a per-netns network notifier block 1825 * @net: network namespace 1826 * @nb: notifier 1827 * 1828 * Register a notifier to be called when network device events occur. 1829 * The notifier passed is linked into the kernel structures and must 1830 * not be reused until it has been unregistered. A negative errno code 1831 * is returned on a failure. 1832 * 1833 * When registered all registration and up events are replayed 1834 * to the new notifier to allow device to have a race free 1835 * view of the network device list. 1836 */ 1837 1838 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb) 1839 { 1840 int err; 1841 1842 rtnl_lock(); 1843 err = __register_netdevice_notifier_net(net, nb, false); 1844 rtnl_unlock(); 1845 return err; 1846 } 1847 EXPORT_SYMBOL(register_netdevice_notifier_net); 1848 1849 /** 1850 * unregister_netdevice_notifier_net - unregister a per-netns 1851 * network notifier block 1852 * @net: network namespace 1853 * @nb: notifier 1854 * 1855 * Unregister a notifier previously registered by 1856 * register_netdevice_notifier_net(). The notifier is unlinked from the 1857 * kernel structures and may then be reused. A negative errno code 1858 * is returned on a failure. 1859 * 1860 * After unregistering unregister and down device events are synthesized 1861 * for all devices on the device list to the removed notifier to remove 1862 * the need for special case cleanup code. 1863 */ 1864 1865 int unregister_netdevice_notifier_net(struct net *net, 1866 struct notifier_block *nb) 1867 { 1868 int err; 1869 1870 rtnl_lock(); 1871 err = __unregister_netdevice_notifier_net(net, nb); 1872 rtnl_unlock(); 1873 return err; 1874 } 1875 EXPORT_SYMBOL(unregister_netdevice_notifier_net); 1876 1877 static void __move_netdevice_notifier_net(struct net *src_net, 1878 struct net *dst_net, 1879 struct notifier_block *nb) 1880 { 1881 __unregister_netdevice_notifier_net(src_net, nb); 1882 __register_netdevice_notifier_net(dst_net, nb, true); 1883 } 1884 1885 int register_netdevice_notifier_dev_net(struct net_device *dev, 1886 struct notifier_block *nb, 1887 struct netdev_net_notifier *nn) 1888 { 1889 int err; 1890 1891 rtnl_lock(); 1892 err = __register_netdevice_notifier_net(dev_net(dev), nb, false); 1893 if (!err) { 1894 nn->nb = nb; 1895 list_add(&nn->list, &dev->net_notifier_list); 1896 } 1897 rtnl_unlock(); 1898 return err; 1899 } 1900 EXPORT_SYMBOL(register_netdevice_notifier_dev_net); 1901 1902 int unregister_netdevice_notifier_dev_net(struct net_device *dev, 1903 struct notifier_block *nb, 1904 struct netdev_net_notifier *nn) 1905 { 1906 int err; 1907 1908 rtnl_lock(); 1909 list_del(&nn->list); 1910 err = __unregister_netdevice_notifier_net(dev_net(dev), nb); 1911 rtnl_unlock(); 1912 return err; 1913 } 1914 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net); 1915 1916 static void move_netdevice_notifiers_dev_net(struct net_device *dev, 1917 struct net *net) 1918 { 1919 struct netdev_net_notifier *nn; 1920 1921 list_for_each_entry(nn, &dev->net_notifier_list, list) 1922 __move_netdevice_notifier_net(dev_net(dev), net, nn->nb); 1923 } 1924 1925 /** 1926 * call_netdevice_notifiers_info - call all network notifier blocks 1927 * @val: value passed unmodified to notifier function 1928 * @info: notifier information data 1929 * 1930 * Call all network notifier blocks. Parameters and return value 1931 * are as for raw_notifier_call_chain(). 1932 */ 1933 1934 int call_netdevice_notifiers_info(unsigned long val, 1935 struct netdev_notifier_info *info) 1936 { 1937 struct net *net = dev_net(info->dev); 1938 int ret; 1939 1940 ASSERT_RTNL(); 1941 1942 /* Run per-netns notifier block chain first, then run the global one. 1943 * Hopefully, one day, the global one is going to be removed after 1944 * all notifier block registrators get converted to be per-netns. 1945 */ 1946 ret = raw_notifier_call_chain(&net->netdev_chain, val, info); 1947 if (ret & NOTIFY_STOP_MASK) 1948 return ret; 1949 return raw_notifier_call_chain(&netdev_chain, val, info); 1950 } 1951 1952 /** 1953 * call_netdevice_notifiers_info_robust - call per-netns notifier blocks 1954 * for and rollback on error 1955 * @val_up: value passed unmodified to notifier function 1956 * @val_down: value passed unmodified to the notifier function when 1957 * recovering from an error on @val_up 1958 * @info: notifier information data 1959 * 1960 * Call all per-netns network notifier blocks, but not notifier blocks on 1961 * the global notifier chain. Parameters and return value are as for 1962 * raw_notifier_call_chain_robust(). 1963 */ 1964 1965 static int 1966 call_netdevice_notifiers_info_robust(unsigned long val_up, 1967 unsigned long val_down, 1968 struct netdev_notifier_info *info) 1969 { 1970 struct net *net = dev_net(info->dev); 1971 1972 ASSERT_RTNL(); 1973 1974 return raw_notifier_call_chain_robust(&net->netdev_chain, 1975 val_up, val_down, info); 1976 } 1977 1978 static int call_netdevice_notifiers_extack(unsigned long val, 1979 struct net_device *dev, 1980 struct netlink_ext_ack *extack) 1981 { 1982 struct netdev_notifier_info info = { 1983 .dev = dev, 1984 .extack = extack, 1985 }; 1986 1987 return call_netdevice_notifiers_info(val, &info); 1988 } 1989 1990 /** 1991 * call_netdevice_notifiers - call all network notifier blocks 1992 * @val: value passed unmodified to notifier function 1993 * @dev: net_device pointer passed unmodified to notifier function 1994 * 1995 * Call all network notifier blocks. Parameters and return value 1996 * are as for raw_notifier_call_chain(). 1997 */ 1998 1999 int call_netdevice_notifiers(unsigned long val, struct net_device *dev) 2000 { 2001 return call_netdevice_notifiers_extack(val, dev, NULL); 2002 } 2003 EXPORT_SYMBOL(call_netdevice_notifiers); 2004 2005 /** 2006 * call_netdevice_notifiers_mtu - call all network notifier blocks 2007 * @val: value passed unmodified to notifier function 2008 * @dev: net_device pointer passed unmodified to notifier function 2009 * @arg: additional u32 argument passed to the notifier function 2010 * 2011 * Call all network notifier blocks. Parameters and return value 2012 * are as for raw_notifier_call_chain(). 2013 */ 2014 static int call_netdevice_notifiers_mtu(unsigned long val, 2015 struct net_device *dev, u32 arg) 2016 { 2017 struct netdev_notifier_info_ext info = { 2018 .info.dev = dev, 2019 .ext.mtu = arg, 2020 }; 2021 2022 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0); 2023 2024 return call_netdevice_notifiers_info(val, &info.info); 2025 } 2026 2027 #ifdef CONFIG_NET_INGRESS 2028 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key); 2029 2030 void net_inc_ingress_queue(void) 2031 { 2032 static_branch_inc(&ingress_needed_key); 2033 } 2034 EXPORT_SYMBOL_GPL(net_inc_ingress_queue); 2035 2036 void net_dec_ingress_queue(void) 2037 { 2038 static_branch_dec(&ingress_needed_key); 2039 } 2040 EXPORT_SYMBOL_GPL(net_dec_ingress_queue); 2041 #endif 2042 2043 #ifdef CONFIG_NET_EGRESS 2044 static DEFINE_STATIC_KEY_FALSE(egress_needed_key); 2045 2046 void net_inc_egress_queue(void) 2047 { 2048 static_branch_inc(&egress_needed_key); 2049 } 2050 EXPORT_SYMBOL_GPL(net_inc_egress_queue); 2051 2052 void net_dec_egress_queue(void) 2053 { 2054 static_branch_dec(&egress_needed_key); 2055 } 2056 EXPORT_SYMBOL_GPL(net_dec_egress_queue); 2057 #endif 2058 2059 DEFINE_STATIC_KEY_FALSE(netstamp_needed_key); 2060 EXPORT_SYMBOL(netstamp_needed_key); 2061 #ifdef CONFIG_JUMP_LABEL 2062 static atomic_t netstamp_needed_deferred; 2063 static atomic_t netstamp_wanted; 2064 static void netstamp_clear(struct work_struct *work) 2065 { 2066 int deferred = atomic_xchg(&netstamp_needed_deferred, 0); 2067 int wanted; 2068 2069 wanted = atomic_add_return(deferred, &netstamp_wanted); 2070 if (wanted > 0) 2071 static_branch_enable(&netstamp_needed_key); 2072 else 2073 static_branch_disable(&netstamp_needed_key); 2074 } 2075 static DECLARE_WORK(netstamp_work, netstamp_clear); 2076 #endif 2077 2078 void net_enable_timestamp(void) 2079 { 2080 #ifdef CONFIG_JUMP_LABEL 2081 int wanted = atomic_read(&netstamp_wanted); 2082 2083 while (wanted > 0) { 2084 if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted + 1)) 2085 return; 2086 } 2087 atomic_inc(&netstamp_needed_deferred); 2088 schedule_work(&netstamp_work); 2089 #else 2090 static_branch_inc(&netstamp_needed_key); 2091 #endif 2092 } 2093 EXPORT_SYMBOL(net_enable_timestamp); 2094 2095 void net_disable_timestamp(void) 2096 { 2097 #ifdef CONFIG_JUMP_LABEL 2098 int wanted = atomic_read(&netstamp_wanted); 2099 2100 while (wanted > 1) { 2101 if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted - 1)) 2102 return; 2103 } 2104 atomic_dec(&netstamp_needed_deferred); 2105 schedule_work(&netstamp_work); 2106 #else 2107 static_branch_dec(&netstamp_needed_key); 2108 #endif 2109 } 2110 EXPORT_SYMBOL(net_disable_timestamp); 2111 2112 static inline void net_timestamp_set(struct sk_buff *skb) 2113 { 2114 skb->tstamp = 0; 2115 skb->mono_delivery_time = 0; 2116 if (static_branch_unlikely(&netstamp_needed_key)) 2117 skb->tstamp = ktime_get_real(); 2118 } 2119 2120 #define net_timestamp_check(COND, SKB) \ 2121 if (static_branch_unlikely(&netstamp_needed_key)) { \ 2122 if ((COND) && !(SKB)->tstamp) \ 2123 (SKB)->tstamp = ktime_get_real(); \ 2124 } \ 2125 2126 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb) 2127 { 2128 return __is_skb_forwardable(dev, skb, true); 2129 } 2130 EXPORT_SYMBOL_GPL(is_skb_forwardable); 2131 2132 static int __dev_forward_skb2(struct net_device *dev, struct sk_buff *skb, 2133 bool check_mtu) 2134 { 2135 int ret = ____dev_forward_skb(dev, skb, check_mtu); 2136 2137 if (likely(!ret)) { 2138 skb->protocol = eth_type_trans(skb, dev); 2139 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); 2140 } 2141 2142 return ret; 2143 } 2144 2145 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2146 { 2147 return __dev_forward_skb2(dev, skb, true); 2148 } 2149 EXPORT_SYMBOL_GPL(__dev_forward_skb); 2150 2151 /** 2152 * dev_forward_skb - loopback an skb to another netif 2153 * 2154 * @dev: destination network device 2155 * @skb: buffer to forward 2156 * 2157 * return values: 2158 * NET_RX_SUCCESS (no congestion) 2159 * NET_RX_DROP (packet was dropped, but freed) 2160 * 2161 * dev_forward_skb can be used for injecting an skb from the 2162 * start_xmit function of one device into the receive queue 2163 * of another device. 2164 * 2165 * The receiving device may be in another namespace, so 2166 * we have to clear all information in the skb that could 2167 * impact namespace isolation. 2168 */ 2169 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) 2170 { 2171 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); 2172 } 2173 EXPORT_SYMBOL_GPL(dev_forward_skb); 2174 2175 int dev_forward_skb_nomtu(struct net_device *dev, struct sk_buff *skb) 2176 { 2177 return __dev_forward_skb2(dev, skb, false) ?: netif_rx_internal(skb); 2178 } 2179 2180 static inline int deliver_skb(struct sk_buff *skb, 2181 struct packet_type *pt_prev, 2182 struct net_device *orig_dev) 2183 { 2184 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 2185 return -ENOMEM; 2186 refcount_inc(&skb->users); 2187 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 2188 } 2189 2190 static inline void deliver_ptype_list_skb(struct sk_buff *skb, 2191 struct packet_type **pt, 2192 struct net_device *orig_dev, 2193 __be16 type, 2194 struct list_head *ptype_list) 2195 { 2196 struct packet_type *ptype, *pt_prev = *pt; 2197 2198 list_for_each_entry_rcu(ptype, ptype_list, list) { 2199 if (ptype->type != type) 2200 continue; 2201 if (pt_prev) 2202 deliver_skb(skb, pt_prev, orig_dev); 2203 pt_prev = ptype; 2204 } 2205 *pt = pt_prev; 2206 } 2207 2208 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) 2209 { 2210 if (!ptype->af_packet_priv || !skb->sk) 2211 return false; 2212 2213 if (ptype->id_match) 2214 return ptype->id_match(ptype, skb->sk); 2215 else if ((struct sock *)ptype->af_packet_priv == skb->sk) 2216 return true; 2217 2218 return false; 2219 } 2220 2221 /** 2222 * dev_nit_active - return true if any network interface taps are in use 2223 * 2224 * @dev: network device to check for the presence of taps 2225 */ 2226 bool dev_nit_active(struct net_device *dev) 2227 { 2228 return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all); 2229 } 2230 EXPORT_SYMBOL_GPL(dev_nit_active); 2231 2232 /* 2233 * Support routine. Sends outgoing frames to any network 2234 * taps currently in use. 2235 */ 2236 2237 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) 2238 { 2239 struct packet_type *ptype; 2240 struct sk_buff *skb2 = NULL; 2241 struct packet_type *pt_prev = NULL; 2242 struct list_head *ptype_list = &ptype_all; 2243 2244 rcu_read_lock(); 2245 again: 2246 list_for_each_entry_rcu(ptype, ptype_list, list) { 2247 if (ptype->ignore_outgoing) 2248 continue; 2249 2250 /* Never send packets back to the socket 2251 * they originated from - MvS ([email protected]) 2252 */ 2253 if (skb_loop_sk(ptype, skb)) 2254 continue; 2255 2256 if (pt_prev) { 2257 deliver_skb(skb2, pt_prev, skb->dev); 2258 pt_prev = ptype; 2259 continue; 2260 } 2261 2262 /* need to clone skb, done only once */ 2263 skb2 = skb_clone(skb, GFP_ATOMIC); 2264 if (!skb2) 2265 goto out_unlock; 2266 2267 net_timestamp_set(skb2); 2268 2269 /* skb->nh should be correctly 2270 * set by sender, so that the second statement is 2271 * just protection against buggy protocols. 2272 */ 2273 skb_reset_mac_header(skb2); 2274 2275 if (skb_network_header(skb2) < skb2->data || 2276 skb_network_header(skb2) > skb_tail_pointer(skb2)) { 2277 net_crit_ratelimited("protocol %04x is buggy, dev %s\n", 2278 ntohs(skb2->protocol), 2279 dev->name); 2280 skb_reset_network_header(skb2); 2281 } 2282 2283 skb2->transport_header = skb2->network_header; 2284 skb2->pkt_type = PACKET_OUTGOING; 2285 pt_prev = ptype; 2286 } 2287 2288 if (ptype_list == &ptype_all) { 2289 ptype_list = &dev->ptype_all; 2290 goto again; 2291 } 2292 out_unlock: 2293 if (pt_prev) { 2294 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC)) 2295 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); 2296 else 2297 kfree_skb(skb2); 2298 } 2299 rcu_read_unlock(); 2300 } 2301 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit); 2302 2303 /** 2304 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change 2305 * @dev: Network device 2306 * @txq: number of queues available 2307 * 2308 * If real_num_tx_queues is changed the tc mappings may no longer be 2309 * valid. To resolve this verify the tc mapping remains valid and if 2310 * not NULL the mapping. With no priorities mapping to this 2311 * offset/count pair it will no longer be used. In the worst case TC0 2312 * is invalid nothing can be done so disable priority mappings. If is 2313 * expected that drivers will fix this mapping if they can before 2314 * calling netif_set_real_num_tx_queues. 2315 */ 2316 static void netif_setup_tc(struct net_device *dev, unsigned int txq) 2317 { 2318 int i; 2319 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2320 2321 /* If TC0 is invalidated disable TC mapping */ 2322 if (tc->offset + tc->count > txq) { 2323 netdev_warn(dev, "Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); 2324 dev->num_tc = 0; 2325 return; 2326 } 2327 2328 /* Invalidated prio to tc mappings set to TC0 */ 2329 for (i = 1; i < TC_BITMASK + 1; i++) { 2330 int q = netdev_get_prio_tc_map(dev, i); 2331 2332 tc = &dev->tc_to_txq[q]; 2333 if (tc->offset + tc->count > txq) { 2334 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", 2335 i, q); 2336 netdev_set_prio_tc_map(dev, i, 0); 2337 } 2338 } 2339 } 2340 2341 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq) 2342 { 2343 if (dev->num_tc) { 2344 struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; 2345 int i; 2346 2347 /* walk through the TCs and see if it falls into any of them */ 2348 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) { 2349 if ((txq - tc->offset) < tc->count) 2350 return i; 2351 } 2352 2353 /* didn't find it, just return -1 to indicate no match */ 2354 return -1; 2355 } 2356 2357 return 0; 2358 } 2359 EXPORT_SYMBOL(netdev_txq_to_tc); 2360 2361 #ifdef CONFIG_XPS 2362 static struct static_key xps_needed __read_mostly; 2363 static struct static_key xps_rxqs_needed __read_mostly; 2364 static DEFINE_MUTEX(xps_map_mutex); 2365 #define xmap_dereference(P) \ 2366 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) 2367 2368 static bool remove_xps_queue(struct xps_dev_maps *dev_maps, 2369 struct xps_dev_maps *old_maps, int tci, u16 index) 2370 { 2371 struct xps_map *map = NULL; 2372 int pos; 2373 2374 map = xmap_dereference(dev_maps->attr_map[tci]); 2375 if (!map) 2376 return false; 2377 2378 for (pos = map->len; pos--;) { 2379 if (map->queues[pos] != index) 2380 continue; 2381 2382 if (map->len > 1) { 2383 map->queues[pos] = map->queues[--map->len]; 2384 break; 2385 } 2386 2387 if (old_maps) 2388 RCU_INIT_POINTER(old_maps->attr_map[tci], NULL); 2389 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2390 kfree_rcu(map, rcu); 2391 return false; 2392 } 2393 2394 return true; 2395 } 2396 2397 static bool remove_xps_queue_cpu(struct net_device *dev, 2398 struct xps_dev_maps *dev_maps, 2399 int cpu, u16 offset, u16 count) 2400 { 2401 int num_tc = dev_maps->num_tc; 2402 bool active = false; 2403 int tci; 2404 2405 for (tci = cpu * num_tc; num_tc--; tci++) { 2406 int i, j; 2407 2408 for (i = count, j = offset; i--; j++) { 2409 if (!remove_xps_queue(dev_maps, NULL, tci, j)) 2410 break; 2411 } 2412 2413 active |= i < 0; 2414 } 2415 2416 return active; 2417 } 2418 2419 static void reset_xps_maps(struct net_device *dev, 2420 struct xps_dev_maps *dev_maps, 2421 enum xps_map_type type) 2422 { 2423 static_key_slow_dec_cpuslocked(&xps_needed); 2424 if (type == XPS_RXQS) 2425 static_key_slow_dec_cpuslocked(&xps_rxqs_needed); 2426 2427 RCU_INIT_POINTER(dev->xps_maps[type], NULL); 2428 2429 kfree_rcu(dev_maps, rcu); 2430 } 2431 2432 static void clean_xps_maps(struct net_device *dev, enum xps_map_type type, 2433 u16 offset, u16 count) 2434 { 2435 struct xps_dev_maps *dev_maps; 2436 bool active = false; 2437 int i, j; 2438 2439 dev_maps = xmap_dereference(dev->xps_maps[type]); 2440 if (!dev_maps) 2441 return; 2442 2443 for (j = 0; j < dev_maps->nr_ids; j++) 2444 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, count); 2445 if (!active) 2446 reset_xps_maps(dev, dev_maps, type); 2447 2448 if (type == XPS_CPUS) { 2449 for (i = offset + (count - 1); count--; i--) 2450 netdev_queue_numa_node_write( 2451 netdev_get_tx_queue(dev, i), NUMA_NO_NODE); 2452 } 2453 } 2454 2455 static void netif_reset_xps_queues(struct net_device *dev, u16 offset, 2456 u16 count) 2457 { 2458 if (!static_key_false(&xps_needed)) 2459 return; 2460 2461 cpus_read_lock(); 2462 mutex_lock(&xps_map_mutex); 2463 2464 if (static_key_false(&xps_rxqs_needed)) 2465 clean_xps_maps(dev, XPS_RXQS, offset, count); 2466 2467 clean_xps_maps(dev, XPS_CPUS, offset, count); 2468 2469 mutex_unlock(&xps_map_mutex); 2470 cpus_read_unlock(); 2471 } 2472 2473 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) 2474 { 2475 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index); 2476 } 2477 2478 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index, 2479 u16 index, bool is_rxqs_map) 2480 { 2481 struct xps_map *new_map; 2482 int alloc_len = XPS_MIN_MAP_ALLOC; 2483 int i, pos; 2484 2485 for (pos = 0; map && pos < map->len; pos++) { 2486 if (map->queues[pos] != index) 2487 continue; 2488 return map; 2489 } 2490 2491 /* Need to add tx-queue to this CPU's/rx-queue's existing map */ 2492 if (map) { 2493 if (pos < map->alloc_len) 2494 return map; 2495 2496 alloc_len = map->alloc_len * 2; 2497 } 2498 2499 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's 2500 * map 2501 */ 2502 if (is_rxqs_map) 2503 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL); 2504 else 2505 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, 2506 cpu_to_node(attr_index)); 2507 if (!new_map) 2508 return NULL; 2509 2510 for (i = 0; i < pos; i++) 2511 new_map->queues[i] = map->queues[i]; 2512 new_map->alloc_len = alloc_len; 2513 new_map->len = pos; 2514 2515 return new_map; 2516 } 2517 2518 /* Copy xps maps at a given index */ 2519 static void xps_copy_dev_maps(struct xps_dev_maps *dev_maps, 2520 struct xps_dev_maps *new_dev_maps, int index, 2521 int tc, bool skip_tc) 2522 { 2523 int i, tci = index * dev_maps->num_tc; 2524 struct xps_map *map; 2525 2526 /* copy maps belonging to foreign traffic classes */ 2527 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2528 if (i == tc && skip_tc) 2529 continue; 2530 2531 /* fill in the new device map from the old device map */ 2532 map = xmap_dereference(dev_maps->attr_map[tci]); 2533 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2534 } 2535 } 2536 2537 /* Must be called under cpus_read_lock */ 2538 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask, 2539 u16 index, enum xps_map_type type) 2540 { 2541 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL, *old_dev_maps = NULL; 2542 const unsigned long *online_mask = NULL; 2543 bool active = false, copy = false; 2544 int i, j, tci, numa_node_id = -2; 2545 int maps_sz, num_tc = 1, tc = 0; 2546 struct xps_map *map, *new_map; 2547 unsigned int nr_ids; 2548 2549 WARN_ON_ONCE(index >= dev->num_tx_queues); 2550 2551 if (dev->num_tc) { 2552 /* Do not allow XPS on subordinate device directly */ 2553 num_tc = dev->num_tc; 2554 if (num_tc < 0) 2555 return -EINVAL; 2556 2557 /* If queue belongs to subordinate dev use its map */ 2558 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev; 2559 2560 tc = netdev_txq_to_tc(dev, index); 2561 if (tc < 0) 2562 return -EINVAL; 2563 } 2564 2565 mutex_lock(&xps_map_mutex); 2566 2567 dev_maps = xmap_dereference(dev->xps_maps[type]); 2568 if (type == XPS_RXQS) { 2569 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues); 2570 nr_ids = dev->num_rx_queues; 2571 } else { 2572 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc); 2573 if (num_possible_cpus() > 1) 2574 online_mask = cpumask_bits(cpu_online_mask); 2575 nr_ids = nr_cpu_ids; 2576 } 2577 2578 if (maps_sz < L1_CACHE_BYTES) 2579 maps_sz = L1_CACHE_BYTES; 2580 2581 /* The old dev_maps could be larger or smaller than the one we're 2582 * setting up now, as dev->num_tc or nr_ids could have been updated in 2583 * between. We could try to be smart, but let's be safe instead and only 2584 * copy foreign traffic classes if the two map sizes match. 2585 */ 2586 if (dev_maps && 2587 dev_maps->num_tc == num_tc && dev_maps->nr_ids == nr_ids) 2588 copy = true; 2589 2590 /* allocate memory for queue storage */ 2591 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids), 2592 j < nr_ids;) { 2593 if (!new_dev_maps) { 2594 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); 2595 if (!new_dev_maps) { 2596 mutex_unlock(&xps_map_mutex); 2597 return -ENOMEM; 2598 } 2599 2600 new_dev_maps->nr_ids = nr_ids; 2601 new_dev_maps->num_tc = num_tc; 2602 } 2603 2604 tci = j * num_tc + tc; 2605 map = copy ? xmap_dereference(dev_maps->attr_map[tci]) : NULL; 2606 2607 map = expand_xps_map(map, j, index, type == XPS_RXQS); 2608 if (!map) 2609 goto error; 2610 2611 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); 2612 } 2613 2614 if (!new_dev_maps) 2615 goto out_no_new_maps; 2616 2617 if (!dev_maps) { 2618 /* Increment static keys at most once per type */ 2619 static_key_slow_inc_cpuslocked(&xps_needed); 2620 if (type == XPS_RXQS) 2621 static_key_slow_inc_cpuslocked(&xps_rxqs_needed); 2622 } 2623 2624 for (j = 0; j < nr_ids; j++) { 2625 bool skip_tc = false; 2626 2627 tci = j * num_tc + tc; 2628 if (netif_attr_test_mask(j, mask, nr_ids) && 2629 netif_attr_test_online(j, online_mask, nr_ids)) { 2630 /* add tx-queue to CPU/rx-queue maps */ 2631 int pos = 0; 2632 2633 skip_tc = true; 2634 2635 map = xmap_dereference(new_dev_maps->attr_map[tci]); 2636 while ((pos < map->len) && (map->queues[pos] != index)) 2637 pos++; 2638 2639 if (pos == map->len) 2640 map->queues[map->len++] = index; 2641 #ifdef CONFIG_NUMA 2642 if (type == XPS_CPUS) { 2643 if (numa_node_id == -2) 2644 numa_node_id = cpu_to_node(j); 2645 else if (numa_node_id != cpu_to_node(j)) 2646 numa_node_id = -1; 2647 } 2648 #endif 2649 } 2650 2651 if (copy) 2652 xps_copy_dev_maps(dev_maps, new_dev_maps, j, tc, 2653 skip_tc); 2654 } 2655 2656 rcu_assign_pointer(dev->xps_maps[type], new_dev_maps); 2657 2658 /* Cleanup old maps */ 2659 if (!dev_maps) 2660 goto out_no_old_maps; 2661 2662 for (j = 0; j < dev_maps->nr_ids; j++) { 2663 for (i = num_tc, tci = j * dev_maps->num_tc; i--; tci++) { 2664 map = xmap_dereference(dev_maps->attr_map[tci]); 2665 if (!map) 2666 continue; 2667 2668 if (copy) { 2669 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2670 if (map == new_map) 2671 continue; 2672 } 2673 2674 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); 2675 kfree_rcu(map, rcu); 2676 } 2677 } 2678 2679 old_dev_maps = dev_maps; 2680 2681 out_no_old_maps: 2682 dev_maps = new_dev_maps; 2683 active = true; 2684 2685 out_no_new_maps: 2686 if (type == XPS_CPUS) 2687 /* update Tx queue numa node */ 2688 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), 2689 (numa_node_id >= 0) ? 2690 numa_node_id : NUMA_NO_NODE); 2691 2692 if (!dev_maps) 2693 goto out_no_maps; 2694 2695 /* removes tx-queue from unused CPUs/rx-queues */ 2696 for (j = 0; j < dev_maps->nr_ids; j++) { 2697 tci = j * dev_maps->num_tc; 2698 2699 for (i = 0; i < dev_maps->num_tc; i++, tci++) { 2700 if (i == tc && 2701 netif_attr_test_mask(j, mask, dev_maps->nr_ids) && 2702 netif_attr_test_online(j, online_mask, dev_maps->nr_ids)) 2703 continue; 2704 2705 active |= remove_xps_queue(dev_maps, 2706 copy ? old_dev_maps : NULL, 2707 tci, index); 2708 } 2709 } 2710 2711 if (old_dev_maps) 2712 kfree_rcu(old_dev_maps, rcu); 2713 2714 /* free map if not active */ 2715 if (!active) 2716 reset_xps_maps(dev, dev_maps, type); 2717 2718 out_no_maps: 2719 mutex_unlock(&xps_map_mutex); 2720 2721 return 0; 2722 error: 2723 /* remove any maps that we added */ 2724 for (j = 0; j < nr_ids; j++) { 2725 for (i = num_tc, tci = j * num_tc; i--; tci++) { 2726 new_map = xmap_dereference(new_dev_maps->attr_map[tci]); 2727 map = copy ? 2728 xmap_dereference(dev_maps->attr_map[tci]) : 2729 NULL; 2730 if (new_map && new_map != map) 2731 kfree(new_map); 2732 } 2733 } 2734 2735 mutex_unlock(&xps_map_mutex); 2736 2737 kfree(new_dev_maps); 2738 return -ENOMEM; 2739 } 2740 EXPORT_SYMBOL_GPL(__netif_set_xps_queue); 2741 2742 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, 2743 u16 index) 2744 { 2745 int ret; 2746 2747 cpus_read_lock(); 2748 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, XPS_CPUS); 2749 cpus_read_unlock(); 2750 2751 return ret; 2752 } 2753 EXPORT_SYMBOL(netif_set_xps_queue); 2754 2755 #endif 2756 static void netdev_unbind_all_sb_channels(struct net_device *dev) 2757 { 2758 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2759 2760 /* Unbind any subordinate channels */ 2761 while (txq-- != &dev->_tx[0]) { 2762 if (txq->sb_dev) 2763 netdev_unbind_sb_channel(dev, txq->sb_dev); 2764 } 2765 } 2766 2767 void netdev_reset_tc(struct net_device *dev) 2768 { 2769 #ifdef CONFIG_XPS 2770 netif_reset_xps_queues_gt(dev, 0); 2771 #endif 2772 netdev_unbind_all_sb_channels(dev); 2773 2774 /* Reset TC configuration of device */ 2775 dev->num_tc = 0; 2776 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq)); 2777 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map)); 2778 } 2779 EXPORT_SYMBOL(netdev_reset_tc); 2780 2781 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset) 2782 { 2783 if (tc >= dev->num_tc) 2784 return -EINVAL; 2785 2786 #ifdef CONFIG_XPS 2787 netif_reset_xps_queues(dev, offset, count); 2788 #endif 2789 dev->tc_to_txq[tc].count = count; 2790 dev->tc_to_txq[tc].offset = offset; 2791 return 0; 2792 } 2793 EXPORT_SYMBOL(netdev_set_tc_queue); 2794 2795 int netdev_set_num_tc(struct net_device *dev, u8 num_tc) 2796 { 2797 if (num_tc > TC_MAX_QUEUE) 2798 return -EINVAL; 2799 2800 #ifdef CONFIG_XPS 2801 netif_reset_xps_queues_gt(dev, 0); 2802 #endif 2803 netdev_unbind_all_sb_channels(dev); 2804 2805 dev->num_tc = num_tc; 2806 return 0; 2807 } 2808 EXPORT_SYMBOL(netdev_set_num_tc); 2809 2810 void netdev_unbind_sb_channel(struct net_device *dev, 2811 struct net_device *sb_dev) 2812 { 2813 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; 2814 2815 #ifdef CONFIG_XPS 2816 netif_reset_xps_queues_gt(sb_dev, 0); 2817 #endif 2818 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq)); 2819 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map)); 2820 2821 while (txq-- != &dev->_tx[0]) { 2822 if (txq->sb_dev == sb_dev) 2823 txq->sb_dev = NULL; 2824 } 2825 } 2826 EXPORT_SYMBOL(netdev_unbind_sb_channel); 2827 2828 int netdev_bind_sb_channel_queue(struct net_device *dev, 2829 struct net_device *sb_dev, 2830 u8 tc, u16 count, u16 offset) 2831 { 2832 /* Make certain the sb_dev and dev are already configured */ 2833 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc) 2834 return -EINVAL; 2835 2836 /* We cannot hand out queues we don't have */ 2837 if ((offset + count) > dev->real_num_tx_queues) 2838 return -EINVAL; 2839 2840 /* Record the mapping */ 2841 sb_dev->tc_to_txq[tc].count = count; 2842 sb_dev->tc_to_txq[tc].offset = offset; 2843 2844 /* Provide a way for Tx queue to find the tc_to_txq map or 2845 * XPS map for itself. 2846 */ 2847 while (count--) 2848 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev; 2849 2850 return 0; 2851 } 2852 EXPORT_SYMBOL(netdev_bind_sb_channel_queue); 2853 2854 int netdev_set_sb_channel(struct net_device *dev, u16 channel) 2855 { 2856 /* Do not use a multiqueue device to represent a subordinate channel */ 2857 if (netif_is_multiqueue(dev)) 2858 return -ENODEV; 2859 2860 /* We allow channels 1 - 32767 to be used for subordinate channels. 2861 * Channel 0 is meant to be "native" mode and used only to represent 2862 * the main root device. We allow writing 0 to reset the device back 2863 * to normal mode after being used as a subordinate channel. 2864 */ 2865 if (channel > S16_MAX) 2866 return -EINVAL; 2867 2868 dev->num_tc = -channel; 2869 2870 return 0; 2871 } 2872 EXPORT_SYMBOL(netdev_set_sb_channel); 2873 2874 /* 2875 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues 2876 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed. 2877 */ 2878 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) 2879 { 2880 bool disabling; 2881 int rc; 2882 2883 disabling = txq < dev->real_num_tx_queues; 2884 2885 if (txq < 1 || txq > dev->num_tx_queues) 2886 return -EINVAL; 2887 2888 if (dev->reg_state == NETREG_REGISTERED || 2889 dev->reg_state == NETREG_UNREGISTERING) { 2890 ASSERT_RTNL(); 2891 2892 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, 2893 txq); 2894 if (rc) 2895 return rc; 2896 2897 if (dev->num_tc) 2898 netif_setup_tc(dev, txq); 2899 2900 dev_qdisc_change_real_num_tx(dev, txq); 2901 2902 dev->real_num_tx_queues = txq; 2903 2904 if (disabling) { 2905 synchronize_net(); 2906 qdisc_reset_all_tx_gt(dev, txq); 2907 #ifdef CONFIG_XPS 2908 netif_reset_xps_queues_gt(dev, txq); 2909 #endif 2910 } 2911 } else { 2912 dev->real_num_tx_queues = txq; 2913 } 2914 2915 return 0; 2916 } 2917 EXPORT_SYMBOL(netif_set_real_num_tx_queues); 2918 2919 #ifdef CONFIG_SYSFS 2920 /** 2921 * netif_set_real_num_rx_queues - set actual number of RX queues used 2922 * @dev: Network device 2923 * @rxq: Actual number of RX queues 2924 * 2925 * This must be called either with the rtnl_lock held or before 2926 * registration of the net device. Returns 0 on success, or a 2927 * negative error code. If called before registration, it always 2928 * succeeds. 2929 */ 2930 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) 2931 { 2932 int rc; 2933 2934 if (rxq < 1 || rxq > dev->num_rx_queues) 2935 return -EINVAL; 2936 2937 if (dev->reg_state == NETREG_REGISTERED) { 2938 ASSERT_RTNL(); 2939 2940 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, 2941 rxq); 2942 if (rc) 2943 return rc; 2944 } 2945 2946 dev->real_num_rx_queues = rxq; 2947 return 0; 2948 } 2949 EXPORT_SYMBOL(netif_set_real_num_rx_queues); 2950 #endif 2951 2952 /** 2953 * netif_set_real_num_queues - set actual number of RX and TX queues used 2954 * @dev: Network device 2955 * @txq: Actual number of TX queues 2956 * @rxq: Actual number of RX queues 2957 * 2958 * Set the real number of both TX and RX queues. 2959 * Does nothing if the number of queues is already correct. 2960 */ 2961 int netif_set_real_num_queues(struct net_device *dev, 2962 unsigned int txq, unsigned int rxq) 2963 { 2964 unsigned int old_rxq = dev->real_num_rx_queues; 2965 int err; 2966 2967 if (txq < 1 || txq > dev->num_tx_queues || 2968 rxq < 1 || rxq > dev->num_rx_queues) 2969 return -EINVAL; 2970 2971 /* Start from increases, so the error path only does decreases - 2972 * decreases can't fail. 2973 */ 2974 if (rxq > dev->real_num_rx_queues) { 2975 err = netif_set_real_num_rx_queues(dev, rxq); 2976 if (err) 2977 return err; 2978 } 2979 if (txq > dev->real_num_tx_queues) { 2980 err = netif_set_real_num_tx_queues(dev, txq); 2981 if (err) 2982 goto undo_rx; 2983 } 2984 if (rxq < dev->real_num_rx_queues) 2985 WARN_ON(netif_set_real_num_rx_queues(dev, rxq)); 2986 if (txq < dev->real_num_tx_queues) 2987 WARN_ON(netif_set_real_num_tx_queues(dev, txq)); 2988 2989 return 0; 2990 undo_rx: 2991 WARN_ON(netif_set_real_num_rx_queues(dev, old_rxq)); 2992 return err; 2993 } 2994 EXPORT_SYMBOL(netif_set_real_num_queues); 2995 2996 /** 2997 * netif_set_tso_max_size() - set the max size of TSO frames supported 2998 * @dev: netdev to update 2999 * @size: max skb->len of a TSO frame 3000 * 3001 * Set the limit on the size of TSO super-frames the device can handle. 3002 * Unless explicitly set the stack will assume the value of 3003 * %GSO_LEGACY_MAX_SIZE. 3004 */ 3005 void netif_set_tso_max_size(struct net_device *dev, unsigned int size) 3006 { 3007 dev->tso_max_size = min(GSO_MAX_SIZE, size); 3008 if (size < READ_ONCE(dev->gso_max_size)) 3009 netif_set_gso_max_size(dev, size); 3010 if (size < READ_ONCE(dev->gso_ipv4_max_size)) 3011 netif_set_gso_ipv4_max_size(dev, size); 3012 } 3013 EXPORT_SYMBOL(netif_set_tso_max_size); 3014 3015 /** 3016 * netif_set_tso_max_segs() - set the max number of segs supported for TSO 3017 * @dev: netdev to update 3018 * @segs: max number of TCP segments 3019 * 3020 * Set the limit on the number of TCP segments the device can generate from 3021 * a single TSO super-frame. 3022 * Unless explicitly set the stack will assume the value of %GSO_MAX_SEGS. 3023 */ 3024 void netif_set_tso_max_segs(struct net_device *dev, unsigned int segs) 3025 { 3026 dev->tso_max_segs = segs; 3027 if (segs < READ_ONCE(dev->gso_max_segs)) 3028 netif_set_gso_max_segs(dev, segs); 3029 } 3030 EXPORT_SYMBOL(netif_set_tso_max_segs); 3031 3032 /** 3033 * netif_inherit_tso_max() - copy all TSO limits from a lower device to an upper 3034 * @to: netdev to update 3035 * @from: netdev from which to copy the limits 3036 */ 3037 void netif_inherit_tso_max(struct net_device *to, const struct net_device *from) 3038 { 3039 netif_set_tso_max_size(to, from->tso_max_size); 3040 netif_set_tso_max_segs(to, from->tso_max_segs); 3041 } 3042 EXPORT_SYMBOL(netif_inherit_tso_max); 3043 3044 /** 3045 * netif_get_num_default_rss_queues - default number of RSS queues 3046 * 3047 * Default value is the number of physical cores if there are only 1 or 2, or 3048 * divided by 2 if there are more. 3049 */ 3050 int netif_get_num_default_rss_queues(void) 3051 { 3052 cpumask_var_t cpus; 3053 int cpu, count = 0; 3054 3055 if (unlikely(is_kdump_kernel() || !zalloc_cpumask_var(&cpus, GFP_KERNEL))) 3056 return 1; 3057 3058 cpumask_copy(cpus, cpu_online_mask); 3059 for_each_cpu(cpu, cpus) { 3060 ++count; 3061 cpumask_andnot(cpus, cpus, topology_sibling_cpumask(cpu)); 3062 } 3063 free_cpumask_var(cpus); 3064 3065 return count > 2 ? DIV_ROUND_UP(count, 2) : count; 3066 } 3067 EXPORT_SYMBOL(netif_get_num_default_rss_queues); 3068 3069 static void __netif_reschedule(struct Qdisc *q) 3070 { 3071 struct softnet_data *sd; 3072 unsigned long flags; 3073 3074 local_irq_save(flags); 3075 sd = this_cpu_ptr(&softnet_data); 3076 q->next_sched = NULL; 3077 *sd->output_queue_tailp = q; 3078 sd->output_queue_tailp = &q->next_sched; 3079 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3080 local_irq_restore(flags); 3081 } 3082 3083 void __netif_schedule(struct Qdisc *q) 3084 { 3085 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) 3086 __netif_reschedule(q); 3087 } 3088 EXPORT_SYMBOL(__netif_schedule); 3089 3090 struct dev_kfree_skb_cb { 3091 enum skb_drop_reason reason; 3092 }; 3093 3094 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) 3095 { 3096 return (struct dev_kfree_skb_cb *)skb->cb; 3097 } 3098 3099 void netif_schedule_queue(struct netdev_queue *txq) 3100 { 3101 rcu_read_lock(); 3102 if (!netif_xmit_stopped(txq)) { 3103 struct Qdisc *q = rcu_dereference(txq->qdisc); 3104 3105 __netif_schedule(q); 3106 } 3107 rcu_read_unlock(); 3108 } 3109 EXPORT_SYMBOL(netif_schedule_queue); 3110 3111 void netif_tx_wake_queue(struct netdev_queue *dev_queue) 3112 { 3113 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { 3114 struct Qdisc *q; 3115 3116 rcu_read_lock(); 3117 q = rcu_dereference(dev_queue->qdisc); 3118 __netif_schedule(q); 3119 rcu_read_unlock(); 3120 } 3121 } 3122 EXPORT_SYMBOL(netif_tx_wake_queue); 3123 3124 void dev_kfree_skb_irq_reason(struct sk_buff *skb, enum skb_drop_reason reason) 3125 { 3126 unsigned long flags; 3127 3128 if (unlikely(!skb)) 3129 return; 3130 3131 if (likely(refcount_read(&skb->users) == 1)) { 3132 smp_rmb(); 3133 refcount_set(&skb->users, 0); 3134 } else if (likely(!refcount_dec_and_test(&skb->users))) { 3135 return; 3136 } 3137 get_kfree_skb_cb(skb)->reason = reason; 3138 local_irq_save(flags); 3139 skb->next = __this_cpu_read(softnet_data.completion_queue); 3140 __this_cpu_write(softnet_data.completion_queue, skb); 3141 raise_softirq_irqoff(NET_TX_SOFTIRQ); 3142 local_irq_restore(flags); 3143 } 3144 EXPORT_SYMBOL(dev_kfree_skb_irq_reason); 3145 3146 void dev_kfree_skb_any_reason(struct sk_buff *skb, enum skb_drop_reason reason) 3147 { 3148 if (in_hardirq() || irqs_disabled()) 3149 dev_kfree_skb_irq_reason(skb, reason); 3150 else 3151 kfree_skb_reason(skb, reason); 3152 } 3153 EXPORT_SYMBOL(dev_kfree_skb_any_reason); 3154 3155 3156 /** 3157 * netif_device_detach - mark device as removed 3158 * @dev: network device 3159 * 3160 * Mark device as removed from system and therefore no longer available. 3161 */ 3162 void netif_device_detach(struct net_device *dev) 3163 { 3164 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && 3165 netif_running(dev)) { 3166 netif_tx_stop_all_queues(dev); 3167 } 3168 } 3169 EXPORT_SYMBOL(netif_device_detach); 3170 3171 /** 3172 * netif_device_attach - mark device as attached 3173 * @dev: network device 3174 * 3175 * Mark device as attached from system and restart if needed. 3176 */ 3177 void netif_device_attach(struct net_device *dev) 3178 { 3179 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && 3180 netif_running(dev)) { 3181 netif_tx_wake_all_queues(dev); 3182 __netdev_watchdog_up(dev); 3183 } 3184 } 3185 EXPORT_SYMBOL(netif_device_attach); 3186 3187 /* 3188 * Returns a Tx hash based on the given packet descriptor a Tx queues' number 3189 * to be used as a distribution range. 3190 */ 3191 static u16 skb_tx_hash(const struct net_device *dev, 3192 const struct net_device *sb_dev, 3193 struct sk_buff *skb) 3194 { 3195 u32 hash; 3196 u16 qoffset = 0; 3197 u16 qcount = dev->real_num_tx_queues; 3198 3199 if (dev->num_tc) { 3200 u8 tc = netdev_get_prio_tc_map(dev, skb->priority); 3201 3202 qoffset = sb_dev->tc_to_txq[tc].offset; 3203 qcount = sb_dev->tc_to_txq[tc].count; 3204 if (unlikely(!qcount)) { 3205 net_warn_ratelimited("%s: invalid qcount, qoffset %u for tc %u\n", 3206 sb_dev->name, qoffset, tc); 3207 qoffset = 0; 3208 qcount = dev->real_num_tx_queues; 3209 } 3210 } 3211 3212 if (skb_rx_queue_recorded(skb)) { 3213 DEBUG_NET_WARN_ON_ONCE(qcount == 0); 3214 hash = skb_get_rx_queue(skb); 3215 if (hash >= qoffset) 3216 hash -= qoffset; 3217 while (unlikely(hash >= qcount)) 3218 hash -= qcount; 3219 return hash + qoffset; 3220 } 3221 3222 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; 3223 } 3224 3225 void skb_warn_bad_offload(const struct sk_buff *skb) 3226 { 3227 static const netdev_features_t null_features; 3228 struct net_device *dev = skb->dev; 3229 const char *name = ""; 3230 3231 if (!net_ratelimit()) 3232 return; 3233 3234 if (dev) { 3235 if (dev->dev.parent) 3236 name = dev_driver_string(dev->dev.parent); 3237 else 3238 name = netdev_name(dev); 3239 } 3240 skb_dump(KERN_WARNING, skb, false); 3241 WARN(1, "%s: caps=(%pNF, %pNF)\n", 3242 name, dev ? &dev->features : &null_features, 3243 skb->sk ? &skb->sk->sk_route_caps : &null_features); 3244 } 3245 3246 /* 3247 * Invalidate hardware checksum when packet is to be mangled, and 3248 * complete checksum manually on outgoing path. 3249 */ 3250 int skb_checksum_help(struct sk_buff *skb) 3251 { 3252 __wsum csum; 3253 int ret = 0, offset; 3254 3255 if (skb->ip_summed == CHECKSUM_COMPLETE) 3256 goto out_set_summed; 3257 3258 if (unlikely(skb_is_gso(skb))) { 3259 skb_warn_bad_offload(skb); 3260 return -EINVAL; 3261 } 3262 3263 /* Before computing a checksum, we should make sure no frag could 3264 * be modified by an external entity : checksum could be wrong. 3265 */ 3266 if (skb_has_shared_frag(skb)) { 3267 ret = __skb_linearize(skb); 3268 if (ret) 3269 goto out; 3270 } 3271 3272 offset = skb_checksum_start_offset(skb); 3273 ret = -EINVAL; 3274 if (unlikely(offset >= skb_headlen(skb))) { 3275 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false); 3276 WARN_ONCE(true, "offset (%d) >= skb_headlen() (%u)\n", 3277 offset, skb_headlen(skb)); 3278 goto out; 3279 } 3280 csum = skb_checksum(skb, offset, skb->len - offset, 0); 3281 3282 offset += skb->csum_offset; 3283 if (unlikely(offset + sizeof(__sum16) > skb_headlen(skb))) { 3284 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false); 3285 WARN_ONCE(true, "offset+2 (%zu) > skb_headlen() (%u)\n", 3286 offset + sizeof(__sum16), skb_headlen(skb)); 3287 goto out; 3288 } 3289 ret = skb_ensure_writable(skb, offset + sizeof(__sum16)); 3290 if (ret) 3291 goto out; 3292 3293 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0; 3294 out_set_summed: 3295 skb->ip_summed = CHECKSUM_NONE; 3296 out: 3297 return ret; 3298 } 3299 EXPORT_SYMBOL(skb_checksum_help); 3300 3301 int skb_crc32c_csum_help(struct sk_buff *skb) 3302 { 3303 __le32 crc32c_csum; 3304 int ret = 0, offset, start; 3305 3306 if (skb->ip_summed != CHECKSUM_PARTIAL) 3307 goto out; 3308 3309 if (unlikely(skb_is_gso(skb))) 3310 goto out; 3311 3312 /* Before computing a checksum, we should make sure no frag could 3313 * be modified by an external entity : checksum could be wrong. 3314 */ 3315 if (unlikely(skb_has_shared_frag(skb))) { 3316 ret = __skb_linearize(skb); 3317 if (ret) 3318 goto out; 3319 } 3320 start = skb_checksum_start_offset(skb); 3321 offset = start + offsetof(struct sctphdr, checksum); 3322 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { 3323 ret = -EINVAL; 3324 goto out; 3325 } 3326 3327 ret = skb_ensure_writable(skb, offset + sizeof(__le32)); 3328 if (ret) 3329 goto out; 3330 3331 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start, 3332 skb->len - start, ~(__u32)0, 3333 crc32c_csum_stub)); 3334 *(__le32 *)(skb->data + offset) = crc32c_csum; 3335 skb_reset_csum_not_inet(skb); 3336 out: 3337 return ret; 3338 } 3339 3340 __be16 skb_network_protocol(struct sk_buff *skb, int *depth) 3341 { 3342 __be16 type = skb->protocol; 3343 3344 /* Tunnel gso handlers can set protocol to ethernet. */ 3345 if (type == htons(ETH_P_TEB)) { 3346 struct ethhdr *eth; 3347 3348 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) 3349 return 0; 3350 3351 eth = (struct ethhdr *)skb->data; 3352 type = eth->h_proto; 3353 } 3354 3355 return vlan_get_protocol_and_depth(skb, type, depth); 3356 } 3357 3358 3359 /* Take action when hardware reception checksum errors are detected. */ 3360 #ifdef CONFIG_BUG 3361 static void do_netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3362 { 3363 netdev_err(dev, "hw csum failure\n"); 3364 skb_dump(KERN_ERR, skb, true); 3365 dump_stack(); 3366 } 3367 3368 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) 3369 { 3370 DO_ONCE_LITE(do_netdev_rx_csum_fault, dev, skb); 3371 } 3372 EXPORT_SYMBOL(netdev_rx_csum_fault); 3373 #endif 3374 3375 /* XXX: check that highmem exists at all on the given machine. */ 3376 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) 3377 { 3378 #ifdef CONFIG_HIGHMEM 3379 int i; 3380 3381 if (!(dev->features & NETIF_F_HIGHDMA)) { 3382 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 3383 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 3384 3385 if (PageHighMem(skb_frag_page(frag))) 3386 return 1; 3387 } 3388 } 3389 #endif 3390 return 0; 3391 } 3392 3393 /* If MPLS offload request, verify we are testing hardware MPLS features 3394 * instead of standard features for the netdev. 3395 */ 3396 #if IS_ENABLED(CONFIG_NET_MPLS_GSO) 3397 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3398 netdev_features_t features, 3399 __be16 type) 3400 { 3401 if (eth_p_mpls(type)) 3402 features &= skb->dev->mpls_features; 3403 3404 return features; 3405 } 3406 #else 3407 static netdev_features_t net_mpls_features(struct sk_buff *skb, 3408 netdev_features_t features, 3409 __be16 type) 3410 { 3411 return features; 3412 } 3413 #endif 3414 3415 static netdev_features_t harmonize_features(struct sk_buff *skb, 3416 netdev_features_t features) 3417 { 3418 __be16 type; 3419 3420 type = skb_network_protocol(skb, NULL); 3421 features = net_mpls_features(skb, features, type); 3422 3423 if (skb->ip_summed != CHECKSUM_NONE && 3424 !can_checksum_protocol(features, type)) { 3425 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); 3426 } 3427 if (illegal_highdma(skb->dev, skb)) 3428 features &= ~NETIF_F_SG; 3429 3430 return features; 3431 } 3432 3433 netdev_features_t passthru_features_check(struct sk_buff *skb, 3434 struct net_device *dev, 3435 netdev_features_t features) 3436 { 3437 return features; 3438 } 3439 EXPORT_SYMBOL(passthru_features_check); 3440 3441 static netdev_features_t dflt_features_check(struct sk_buff *skb, 3442 struct net_device *dev, 3443 netdev_features_t features) 3444 { 3445 return vlan_features_check(skb, features); 3446 } 3447 3448 static netdev_features_t gso_features_check(const struct sk_buff *skb, 3449 struct net_device *dev, 3450 netdev_features_t features) 3451 { 3452 u16 gso_segs = skb_shinfo(skb)->gso_segs; 3453 3454 if (gso_segs > READ_ONCE(dev->gso_max_segs)) 3455 return features & ~NETIF_F_GSO_MASK; 3456 3457 if (unlikely(skb->len >= READ_ONCE(dev->gso_max_size))) 3458 return features & ~NETIF_F_GSO_MASK; 3459 3460 if (!skb_shinfo(skb)->gso_type) { 3461 skb_warn_bad_offload(skb); 3462 return features & ~NETIF_F_GSO_MASK; 3463 } 3464 3465 /* Support for GSO partial features requires software 3466 * intervention before we can actually process the packets 3467 * so we need to strip support for any partial features now 3468 * and we can pull them back in after we have partially 3469 * segmented the frame. 3470 */ 3471 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)) 3472 features &= ~dev->gso_partial_features; 3473 3474 /* Make sure to clear the IPv4 ID mangling feature if the 3475 * IPv4 header has the potential to be fragmented. 3476 */ 3477 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { 3478 struct iphdr *iph = skb->encapsulation ? 3479 inner_ip_hdr(skb) : ip_hdr(skb); 3480 3481 if (!(iph->frag_off & htons(IP_DF))) 3482 features &= ~NETIF_F_TSO_MANGLEID; 3483 } 3484 3485 return features; 3486 } 3487 3488 netdev_features_t netif_skb_features(struct sk_buff *skb) 3489 { 3490 struct net_device *dev = skb->dev; 3491 netdev_features_t features = dev->features; 3492 3493 if (skb_is_gso(skb)) 3494 features = gso_features_check(skb, dev, features); 3495 3496 /* If encapsulation offload request, verify we are testing 3497 * hardware encapsulation features instead of standard 3498 * features for the netdev 3499 */ 3500 if (skb->encapsulation) 3501 features &= dev->hw_enc_features; 3502 3503 if (skb_vlan_tagged(skb)) 3504 features = netdev_intersect_features(features, 3505 dev->vlan_features | 3506 NETIF_F_HW_VLAN_CTAG_TX | 3507 NETIF_F_HW_VLAN_STAG_TX); 3508 3509 if (dev->netdev_ops->ndo_features_check) 3510 features &= dev->netdev_ops->ndo_features_check(skb, dev, 3511 features); 3512 else 3513 features &= dflt_features_check(skb, dev, features); 3514 3515 return harmonize_features(skb, features); 3516 } 3517 EXPORT_SYMBOL(netif_skb_features); 3518 3519 static int xmit_one(struct sk_buff *skb, struct net_device *dev, 3520 struct netdev_queue *txq, bool more) 3521 { 3522 unsigned int len; 3523 int rc; 3524 3525 if (dev_nit_active(dev)) 3526 dev_queue_xmit_nit(skb, dev); 3527 3528 len = skb->len; 3529 trace_net_dev_start_xmit(skb, dev); 3530 rc = netdev_start_xmit(skb, dev, txq, more); 3531 trace_net_dev_xmit(skb, rc, dev, len); 3532 3533 return rc; 3534 } 3535 3536 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, 3537 struct netdev_queue *txq, int *ret) 3538 { 3539 struct sk_buff *skb = first; 3540 int rc = NETDEV_TX_OK; 3541 3542 while (skb) { 3543 struct sk_buff *next = skb->next; 3544 3545 skb_mark_not_on_list(skb); 3546 rc = xmit_one(skb, dev, txq, next != NULL); 3547 if (unlikely(!dev_xmit_complete(rc))) { 3548 skb->next = next; 3549 goto out; 3550 } 3551 3552 skb = next; 3553 if (netif_tx_queue_stopped(txq) && skb) { 3554 rc = NETDEV_TX_BUSY; 3555 break; 3556 } 3557 } 3558 3559 out: 3560 *ret = rc; 3561 return skb; 3562 } 3563 3564 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, 3565 netdev_features_t features) 3566 { 3567 if (skb_vlan_tag_present(skb) && 3568 !vlan_hw_offload_capable(features, skb->vlan_proto)) 3569 skb = __vlan_hwaccel_push_inside(skb); 3570 return skb; 3571 } 3572 3573 int skb_csum_hwoffload_help(struct sk_buff *skb, 3574 const netdev_features_t features) 3575 { 3576 if (unlikely(skb_csum_is_sctp(skb))) 3577 return !!(features & NETIF_F_SCTP_CRC) ? 0 : 3578 skb_crc32c_csum_help(skb); 3579 3580 if (features & NETIF_F_HW_CSUM) 3581 return 0; 3582 3583 if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) { 3584 switch (skb->csum_offset) { 3585 case offsetof(struct tcphdr, check): 3586 case offsetof(struct udphdr, check): 3587 return 0; 3588 } 3589 } 3590 3591 return skb_checksum_help(skb); 3592 } 3593 EXPORT_SYMBOL(skb_csum_hwoffload_help); 3594 3595 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again) 3596 { 3597 netdev_features_t features; 3598 3599 features = netif_skb_features(skb); 3600 skb = validate_xmit_vlan(skb, features); 3601 if (unlikely(!skb)) 3602 goto out_null; 3603 3604 skb = sk_validate_xmit_skb(skb, dev); 3605 if (unlikely(!skb)) 3606 goto out_null; 3607 3608 if (netif_needs_gso(skb, features)) { 3609 struct sk_buff *segs; 3610 3611 segs = skb_gso_segment(skb, features); 3612 if (IS_ERR(segs)) { 3613 goto out_kfree_skb; 3614 } else if (segs) { 3615 consume_skb(skb); 3616 skb = segs; 3617 } 3618 } else { 3619 if (skb_needs_linearize(skb, features) && 3620 __skb_linearize(skb)) 3621 goto out_kfree_skb; 3622 3623 /* If packet is not checksummed and device does not 3624 * support checksumming for this protocol, complete 3625 * checksumming here. 3626 */ 3627 if (skb->ip_summed == CHECKSUM_PARTIAL) { 3628 if (skb->encapsulation) 3629 skb_set_inner_transport_header(skb, 3630 skb_checksum_start_offset(skb)); 3631 else 3632 skb_set_transport_header(skb, 3633 skb_checksum_start_offset(skb)); 3634 if (skb_csum_hwoffload_help(skb, features)) 3635 goto out_kfree_skb; 3636 } 3637 } 3638 3639 skb = validate_xmit_xfrm(skb, features, again); 3640 3641 return skb; 3642 3643 out_kfree_skb: 3644 kfree_skb(skb); 3645 out_null: 3646 dev_core_stats_tx_dropped_inc(dev); 3647 return NULL; 3648 } 3649 3650 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again) 3651 { 3652 struct sk_buff *next, *head = NULL, *tail; 3653 3654 for (; skb != NULL; skb = next) { 3655 next = skb->next; 3656 skb_mark_not_on_list(skb); 3657 3658 /* in case skb wont be segmented, point to itself */ 3659 skb->prev = skb; 3660 3661 skb = validate_xmit_skb(skb, dev, again); 3662 if (!skb) 3663 continue; 3664 3665 if (!head) 3666 head = skb; 3667 else 3668 tail->next = skb; 3669 /* If skb was segmented, skb->prev points to 3670 * the last segment. If not, it still contains skb. 3671 */ 3672 tail = skb->prev; 3673 } 3674 return head; 3675 } 3676 EXPORT_SYMBOL_GPL(validate_xmit_skb_list); 3677 3678 static void qdisc_pkt_len_init(struct sk_buff *skb) 3679 { 3680 const struct skb_shared_info *shinfo = skb_shinfo(skb); 3681 3682 qdisc_skb_cb(skb)->pkt_len = skb->len; 3683 3684 /* To get more precise estimation of bytes sent on wire, 3685 * we add to pkt_len the headers size of all segments 3686 */ 3687 if (shinfo->gso_size && skb_transport_header_was_set(skb)) { 3688 u16 gso_segs = shinfo->gso_segs; 3689 unsigned int hdr_len; 3690 3691 /* mac layer + network layer */ 3692 hdr_len = skb_transport_offset(skb); 3693 3694 /* + transport layer */ 3695 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { 3696 const struct tcphdr *th; 3697 struct tcphdr _tcphdr; 3698 3699 th = skb_header_pointer(skb, hdr_len, 3700 sizeof(_tcphdr), &_tcphdr); 3701 if (likely(th)) 3702 hdr_len += __tcp_hdrlen(th); 3703 } else { 3704 struct udphdr _udphdr; 3705 3706 if (skb_header_pointer(skb, hdr_len, 3707 sizeof(_udphdr), &_udphdr)) 3708 hdr_len += sizeof(struct udphdr); 3709 } 3710 3711 if (shinfo->gso_type & SKB_GSO_DODGY) 3712 gso_segs = DIV_ROUND_UP(skb->len - hdr_len, 3713 shinfo->gso_size); 3714 3715 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; 3716 } 3717 } 3718 3719 static int dev_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *q, 3720 struct sk_buff **to_free, 3721 struct netdev_queue *txq) 3722 { 3723 int rc; 3724 3725 rc = q->enqueue(skb, q, to_free) & NET_XMIT_MASK; 3726 if (rc == NET_XMIT_SUCCESS) 3727 trace_qdisc_enqueue(q, txq, skb); 3728 return rc; 3729 } 3730 3731 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, 3732 struct net_device *dev, 3733 struct netdev_queue *txq) 3734 { 3735 spinlock_t *root_lock = qdisc_lock(q); 3736 struct sk_buff *to_free = NULL; 3737 bool contended; 3738 int rc; 3739 3740 qdisc_calculate_pkt_len(skb, q); 3741 3742 tcf_set_drop_reason(skb, SKB_DROP_REASON_QDISC_DROP); 3743 3744 if (q->flags & TCQ_F_NOLOCK) { 3745 if (q->flags & TCQ_F_CAN_BYPASS && nolock_qdisc_is_empty(q) && 3746 qdisc_run_begin(q)) { 3747 /* Retest nolock_qdisc_is_empty() within the protection 3748 * of q->seqlock to protect from racing with requeuing. 3749 */ 3750 if (unlikely(!nolock_qdisc_is_empty(q))) { 3751 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3752 __qdisc_run(q); 3753 qdisc_run_end(q); 3754 3755 goto no_lock_out; 3756 } 3757 3758 qdisc_bstats_cpu_update(q, skb); 3759 if (sch_direct_xmit(skb, q, dev, txq, NULL, true) && 3760 !nolock_qdisc_is_empty(q)) 3761 __qdisc_run(q); 3762 3763 qdisc_run_end(q); 3764 return NET_XMIT_SUCCESS; 3765 } 3766 3767 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3768 qdisc_run(q); 3769 3770 no_lock_out: 3771 if (unlikely(to_free)) 3772 kfree_skb_list_reason(to_free, 3773 tcf_get_drop_reason(to_free)); 3774 return rc; 3775 } 3776 3777 /* 3778 * Heuristic to force contended enqueues to serialize on a 3779 * separate lock before trying to get qdisc main lock. 3780 * This permits qdisc->running owner to get the lock more 3781 * often and dequeue packets faster. 3782 * On PREEMPT_RT it is possible to preempt the qdisc owner during xmit 3783 * and then other tasks will only enqueue packets. The packets will be 3784 * sent after the qdisc owner is scheduled again. To prevent this 3785 * scenario the task always serialize on the lock. 3786 */ 3787 contended = qdisc_is_running(q) || IS_ENABLED(CONFIG_PREEMPT_RT); 3788 if (unlikely(contended)) 3789 spin_lock(&q->busylock); 3790 3791 spin_lock(root_lock); 3792 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { 3793 __qdisc_drop(skb, &to_free); 3794 rc = NET_XMIT_DROP; 3795 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && 3796 qdisc_run_begin(q)) { 3797 /* 3798 * This is a work-conserving queue; there are no old skbs 3799 * waiting to be sent out; and the qdisc is not running - 3800 * xmit the skb directly. 3801 */ 3802 3803 qdisc_bstats_update(q, skb); 3804 3805 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { 3806 if (unlikely(contended)) { 3807 spin_unlock(&q->busylock); 3808 contended = false; 3809 } 3810 __qdisc_run(q); 3811 } 3812 3813 qdisc_run_end(q); 3814 rc = NET_XMIT_SUCCESS; 3815 } else { 3816 rc = dev_qdisc_enqueue(skb, q, &to_free, txq); 3817 if (qdisc_run_begin(q)) { 3818 if (unlikely(contended)) { 3819 spin_unlock(&q->busylock); 3820 contended = false; 3821 } 3822 __qdisc_run(q); 3823 qdisc_run_end(q); 3824 } 3825 } 3826 spin_unlock(root_lock); 3827 if (unlikely(to_free)) 3828 kfree_skb_list_reason(to_free, 3829 tcf_get_drop_reason(to_free)); 3830 if (unlikely(contended)) 3831 spin_unlock(&q->busylock); 3832 return rc; 3833 } 3834 3835 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) 3836 static void skb_update_prio(struct sk_buff *skb) 3837 { 3838 const struct netprio_map *map; 3839 const struct sock *sk; 3840 unsigned int prioidx; 3841 3842 if (skb->priority) 3843 return; 3844 map = rcu_dereference_bh(skb->dev->priomap); 3845 if (!map) 3846 return; 3847 sk = skb_to_full_sk(skb); 3848 if (!sk) 3849 return; 3850 3851 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data); 3852 3853 if (prioidx < map->priomap_len) 3854 skb->priority = map->priomap[prioidx]; 3855 } 3856 #else 3857 #define skb_update_prio(skb) 3858 #endif 3859 3860 /** 3861 * dev_loopback_xmit - loop back @skb 3862 * @net: network namespace this loopback is happening in 3863 * @sk: sk needed to be a netfilter okfn 3864 * @skb: buffer to transmit 3865 */ 3866 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) 3867 { 3868 skb_reset_mac_header(skb); 3869 __skb_pull(skb, skb_network_offset(skb)); 3870 skb->pkt_type = PACKET_LOOPBACK; 3871 if (skb->ip_summed == CHECKSUM_NONE) 3872 skb->ip_summed = CHECKSUM_UNNECESSARY; 3873 DEBUG_NET_WARN_ON_ONCE(!skb_dst(skb)); 3874 skb_dst_force(skb); 3875 netif_rx(skb); 3876 return 0; 3877 } 3878 EXPORT_SYMBOL(dev_loopback_xmit); 3879 3880 #ifdef CONFIG_NET_EGRESS 3881 static struct netdev_queue * 3882 netdev_tx_queue_mapping(struct net_device *dev, struct sk_buff *skb) 3883 { 3884 int qm = skb_get_queue_mapping(skb); 3885 3886 return netdev_get_tx_queue(dev, netdev_cap_txqueue(dev, qm)); 3887 } 3888 3889 static bool netdev_xmit_txqueue_skipped(void) 3890 { 3891 return __this_cpu_read(softnet_data.xmit.skip_txqueue); 3892 } 3893 3894 void netdev_xmit_skip_txqueue(bool skip) 3895 { 3896 __this_cpu_write(softnet_data.xmit.skip_txqueue, skip); 3897 } 3898 EXPORT_SYMBOL_GPL(netdev_xmit_skip_txqueue); 3899 #endif /* CONFIG_NET_EGRESS */ 3900 3901 #ifdef CONFIG_NET_XGRESS 3902 static int tc_run(struct tcx_entry *entry, struct sk_buff *skb, 3903 enum skb_drop_reason *drop_reason) 3904 { 3905 int ret = TC_ACT_UNSPEC; 3906 #ifdef CONFIG_NET_CLS_ACT 3907 struct mini_Qdisc *miniq = rcu_dereference_bh(entry->miniq); 3908 struct tcf_result res; 3909 3910 if (!miniq) 3911 return ret; 3912 3913 tc_skb_cb(skb)->mru = 0; 3914 tc_skb_cb(skb)->post_ct = false; 3915 tcf_set_drop_reason(skb, *drop_reason); 3916 3917 mini_qdisc_bstats_cpu_update(miniq, skb); 3918 ret = tcf_classify(skb, miniq->block, miniq->filter_list, &res, false); 3919 /* Only tcf related quirks below. */ 3920 switch (ret) { 3921 case TC_ACT_SHOT: 3922 *drop_reason = tcf_get_drop_reason(skb); 3923 mini_qdisc_qstats_cpu_drop(miniq); 3924 break; 3925 case TC_ACT_OK: 3926 case TC_ACT_RECLASSIFY: 3927 skb->tc_index = TC_H_MIN(res.classid); 3928 break; 3929 } 3930 #endif /* CONFIG_NET_CLS_ACT */ 3931 return ret; 3932 } 3933 3934 static DEFINE_STATIC_KEY_FALSE(tcx_needed_key); 3935 3936 void tcx_inc(void) 3937 { 3938 static_branch_inc(&tcx_needed_key); 3939 } 3940 3941 void tcx_dec(void) 3942 { 3943 static_branch_dec(&tcx_needed_key); 3944 } 3945 3946 static __always_inline enum tcx_action_base 3947 tcx_run(const struct bpf_mprog_entry *entry, struct sk_buff *skb, 3948 const bool needs_mac) 3949 { 3950 const struct bpf_mprog_fp *fp; 3951 const struct bpf_prog *prog; 3952 int ret = TCX_NEXT; 3953 3954 if (needs_mac) 3955 __skb_push(skb, skb->mac_len); 3956 bpf_mprog_foreach_prog(entry, fp, prog) { 3957 bpf_compute_data_pointers(skb); 3958 ret = bpf_prog_run(prog, skb); 3959 if (ret != TCX_NEXT) 3960 break; 3961 } 3962 if (needs_mac) 3963 __skb_pull(skb, skb->mac_len); 3964 return tcx_action_code(skb, ret); 3965 } 3966 3967 static __always_inline struct sk_buff * 3968 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 3969 struct net_device *orig_dev, bool *another) 3970 { 3971 struct bpf_mprog_entry *entry = rcu_dereference_bh(skb->dev->tcx_ingress); 3972 enum skb_drop_reason drop_reason = SKB_DROP_REASON_TC_INGRESS; 3973 int sch_ret; 3974 3975 if (!entry) 3976 return skb; 3977 if (*pt_prev) { 3978 *ret = deliver_skb(skb, *pt_prev, orig_dev); 3979 *pt_prev = NULL; 3980 } 3981 3982 qdisc_skb_cb(skb)->pkt_len = skb->len; 3983 tcx_set_ingress(skb, true); 3984 3985 if (static_branch_unlikely(&tcx_needed_key)) { 3986 sch_ret = tcx_run(entry, skb, true); 3987 if (sch_ret != TC_ACT_UNSPEC) 3988 goto ingress_verdict; 3989 } 3990 sch_ret = tc_run(tcx_entry(entry), skb, &drop_reason); 3991 ingress_verdict: 3992 switch (sch_ret) { 3993 case TC_ACT_REDIRECT: 3994 /* skb_mac_header check was done by BPF, so we can safely 3995 * push the L2 header back before redirecting to another 3996 * netdev. 3997 */ 3998 __skb_push(skb, skb->mac_len); 3999 if (skb_do_redirect(skb) == -EAGAIN) { 4000 __skb_pull(skb, skb->mac_len); 4001 *another = true; 4002 break; 4003 } 4004 *ret = NET_RX_SUCCESS; 4005 return NULL; 4006 case TC_ACT_SHOT: 4007 kfree_skb_reason(skb, drop_reason); 4008 *ret = NET_RX_DROP; 4009 return NULL; 4010 /* used by tc_run */ 4011 case TC_ACT_STOLEN: 4012 case TC_ACT_QUEUED: 4013 case TC_ACT_TRAP: 4014 consume_skb(skb); 4015 fallthrough; 4016 case TC_ACT_CONSUMED: 4017 *ret = NET_RX_SUCCESS; 4018 return NULL; 4019 } 4020 4021 return skb; 4022 } 4023 4024 static __always_inline struct sk_buff * 4025 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 4026 { 4027 struct bpf_mprog_entry *entry = rcu_dereference_bh(dev->tcx_egress); 4028 enum skb_drop_reason drop_reason = SKB_DROP_REASON_TC_EGRESS; 4029 int sch_ret; 4030 4031 if (!entry) 4032 return skb; 4033 4034 /* qdisc_skb_cb(skb)->pkt_len & tcx_set_ingress() was 4035 * already set by the caller. 4036 */ 4037 if (static_branch_unlikely(&tcx_needed_key)) { 4038 sch_ret = tcx_run(entry, skb, false); 4039 if (sch_ret != TC_ACT_UNSPEC) 4040 goto egress_verdict; 4041 } 4042 sch_ret = tc_run(tcx_entry(entry), skb, &drop_reason); 4043 egress_verdict: 4044 switch (sch_ret) { 4045 case TC_ACT_REDIRECT: 4046 /* No need to push/pop skb's mac_header here on egress! */ 4047 skb_do_redirect(skb); 4048 *ret = NET_XMIT_SUCCESS; 4049 return NULL; 4050 case TC_ACT_SHOT: 4051 kfree_skb_reason(skb, drop_reason); 4052 *ret = NET_XMIT_DROP; 4053 return NULL; 4054 /* used by tc_run */ 4055 case TC_ACT_STOLEN: 4056 case TC_ACT_QUEUED: 4057 case TC_ACT_TRAP: 4058 consume_skb(skb); 4059 fallthrough; 4060 case TC_ACT_CONSUMED: 4061 *ret = NET_XMIT_SUCCESS; 4062 return NULL; 4063 } 4064 4065 return skb; 4066 } 4067 #else 4068 static __always_inline struct sk_buff * 4069 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, 4070 struct net_device *orig_dev, bool *another) 4071 { 4072 return skb; 4073 } 4074 4075 static __always_inline struct sk_buff * 4076 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) 4077 { 4078 return skb; 4079 } 4080 #endif /* CONFIG_NET_XGRESS */ 4081 4082 #ifdef CONFIG_XPS 4083 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb, 4084 struct xps_dev_maps *dev_maps, unsigned int tci) 4085 { 4086 int tc = netdev_get_prio_tc_map(dev, skb->priority); 4087 struct xps_map *map; 4088 int queue_index = -1; 4089 4090 if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids) 4091 return queue_index; 4092 4093 tci *= dev_maps->num_tc; 4094 tci += tc; 4095 4096 map = rcu_dereference(dev_maps->attr_map[tci]); 4097 if (map) { 4098 if (map->len == 1) 4099 queue_index = map->queues[0]; 4100 else 4101 queue_index = map->queues[reciprocal_scale( 4102 skb_get_hash(skb), map->len)]; 4103 if (unlikely(queue_index >= dev->real_num_tx_queues)) 4104 queue_index = -1; 4105 } 4106 return queue_index; 4107 } 4108 #endif 4109 4110 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev, 4111 struct sk_buff *skb) 4112 { 4113 #ifdef CONFIG_XPS 4114 struct xps_dev_maps *dev_maps; 4115 struct sock *sk = skb->sk; 4116 int queue_index = -1; 4117 4118 if (!static_key_false(&xps_needed)) 4119 return -1; 4120 4121 rcu_read_lock(); 4122 if (!static_key_false(&xps_rxqs_needed)) 4123 goto get_cpus_map; 4124 4125 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]); 4126 if (dev_maps) { 4127 int tci = sk_rx_queue_get(sk); 4128 4129 if (tci >= 0) 4130 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4131 tci); 4132 } 4133 4134 get_cpus_map: 4135 if (queue_index < 0) { 4136 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]); 4137 if (dev_maps) { 4138 unsigned int tci = skb->sender_cpu - 1; 4139 4140 queue_index = __get_xps_queue_idx(dev, skb, dev_maps, 4141 tci); 4142 } 4143 } 4144 rcu_read_unlock(); 4145 4146 return queue_index; 4147 #else 4148 return -1; 4149 #endif 4150 } 4151 4152 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb, 4153 struct net_device *sb_dev) 4154 { 4155 return 0; 4156 } 4157 EXPORT_SYMBOL(dev_pick_tx_zero); 4158 4159 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb, 4160 struct net_device *sb_dev) 4161 { 4162 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues; 4163 } 4164 EXPORT_SYMBOL(dev_pick_tx_cpu_id); 4165 4166 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb, 4167 struct net_device *sb_dev) 4168 { 4169 struct sock *sk = skb->sk; 4170 int queue_index = sk_tx_queue_get(sk); 4171 4172 sb_dev = sb_dev ? : dev; 4173 4174 if (queue_index < 0 || skb->ooo_okay || 4175 queue_index >= dev->real_num_tx_queues) { 4176 int new_index = get_xps_queue(dev, sb_dev, skb); 4177 4178 if (new_index < 0) 4179 new_index = skb_tx_hash(dev, sb_dev, skb); 4180 4181 if (queue_index != new_index && sk && 4182 sk_fullsock(sk) && 4183 rcu_access_pointer(sk->sk_dst_cache)) 4184 sk_tx_queue_set(sk, new_index); 4185 4186 queue_index = new_index; 4187 } 4188 4189 return queue_index; 4190 } 4191 EXPORT_SYMBOL(netdev_pick_tx); 4192 4193 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev, 4194 struct sk_buff *skb, 4195 struct net_device *sb_dev) 4196 { 4197 int queue_index = 0; 4198 4199 #ifdef CONFIG_XPS 4200 u32 sender_cpu = skb->sender_cpu - 1; 4201 4202 if (sender_cpu >= (u32)NR_CPUS) 4203 skb->sender_cpu = raw_smp_processor_id() + 1; 4204 #endif 4205 4206 if (dev->real_num_tx_queues != 1) { 4207 const struct net_device_ops *ops = dev->netdev_ops; 4208 4209 if (ops->ndo_select_queue) 4210 queue_index = ops->ndo_select_queue(dev, skb, sb_dev); 4211 else 4212 queue_index = netdev_pick_tx(dev, skb, sb_dev); 4213 4214 queue_index = netdev_cap_txqueue(dev, queue_index); 4215 } 4216 4217 skb_set_queue_mapping(skb, queue_index); 4218 return netdev_get_tx_queue(dev, queue_index); 4219 } 4220 4221 /** 4222 * __dev_queue_xmit() - transmit a buffer 4223 * @skb: buffer to transmit 4224 * @sb_dev: suboordinate device used for L2 forwarding offload 4225 * 4226 * Queue a buffer for transmission to a network device. The caller must 4227 * have set the device and priority and built the buffer before calling 4228 * this function. The function can be called from an interrupt. 4229 * 4230 * When calling this method, interrupts MUST be enabled. This is because 4231 * the BH enable code must have IRQs enabled so that it will not deadlock. 4232 * 4233 * Regardless of the return value, the skb is consumed, so it is currently 4234 * difficult to retry a send to this method. (You can bump the ref count 4235 * before sending to hold a reference for retry if you are careful.) 4236 * 4237 * Return: 4238 * * 0 - buffer successfully transmitted 4239 * * positive qdisc return code - NET_XMIT_DROP etc. 4240 * * negative errno - other errors 4241 */ 4242 int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev) 4243 { 4244 struct net_device *dev = skb->dev; 4245 struct netdev_queue *txq = NULL; 4246 struct Qdisc *q; 4247 int rc = -ENOMEM; 4248 bool again = false; 4249 4250 skb_reset_mac_header(skb); 4251 skb_assert_len(skb); 4252 4253 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) 4254 __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED); 4255 4256 /* Disable soft irqs for various locks below. Also 4257 * stops preemption for RCU. 4258 */ 4259 rcu_read_lock_bh(); 4260 4261 skb_update_prio(skb); 4262 4263 qdisc_pkt_len_init(skb); 4264 tcx_set_ingress(skb, false); 4265 #ifdef CONFIG_NET_EGRESS 4266 if (static_branch_unlikely(&egress_needed_key)) { 4267 if (nf_hook_egress_active()) { 4268 skb = nf_hook_egress(skb, &rc, dev); 4269 if (!skb) 4270 goto out; 4271 } 4272 4273 netdev_xmit_skip_txqueue(false); 4274 4275 nf_skip_egress(skb, true); 4276 skb = sch_handle_egress(skb, &rc, dev); 4277 if (!skb) 4278 goto out; 4279 nf_skip_egress(skb, false); 4280 4281 if (netdev_xmit_txqueue_skipped()) 4282 txq = netdev_tx_queue_mapping(dev, skb); 4283 } 4284 #endif 4285 /* If device/qdisc don't need skb->dst, release it right now while 4286 * its hot in this cpu cache. 4287 */ 4288 if (dev->priv_flags & IFF_XMIT_DST_RELEASE) 4289 skb_dst_drop(skb); 4290 else 4291 skb_dst_force(skb); 4292 4293 if (!txq) 4294 txq = netdev_core_pick_tx(dev, skb, sb_dev); 4295 4296 q = rcu_dereference_bh(txq->qdisc); 4297 4298 trace_net_dev_queue(skb); 4299 if (q->enqueue) { 4300 rc = __dev_xmit_skb(skb, q, dev, txq); 4301 goto out; 4302 } 4303 4304 /* The device has no queue. Common case for software devices: 4305 * loopback, all the sorts of tunnels... 4306 4307 * Really, it is unlikely that netif_tx_lock protection is necessary 4308 * here. (f.e. loopback and IP tunnels are clean ignoring statistics 4309 * counters.) 4310 * However, it is possible, that they rely on protection 4311 * made by us here. 4312 4313 * Check this and shot the lock. It is not prone from deadlocks. 4314 *Either shot noqueue qdisc, it is even simpler 8) 4315 */ 4316 if (dev->flags & IFF_UP) { 4317 int cpu = smp_processor_id(); /* ok because BHs are off */ 4318 4319 /* Other cpus might concurrently change txq->xmit_lock_owner 4320 * to -1 or to their cpu id, but not to our id. 4321 */ 4322 if (READ_ONCE(txq->xmit_lock_owner) != cpu) { 4323 if (dev_xmit_recursion()) 4324 goto recursion_alert; 4325 4326 skb = validate_xmit_skb(skb, dev, &again); 4327 if (!skb) 4328 goto out; 4329 4330 HARD_TX_LOCK(dev, txq, cpu); 4331 4332 if (!netif_xmit_stopped(txq)) { 4333 dev_xmit_recursion_inc(); 4334 skb = dev_hard_start_xmit(skb, dev, txq, &rc); 4335 dev_xmit_recursion_dec(); 4336 if (dev_xmit_complete(rc)) { 4337 HARD_TX_UNLOCK(dev, txq); 4338 goto out; 4339 } 4340 } 4341 HARD_TX_UNLOCK(dev, txq); 4342 net_crit_ratelimited("Virtual device %s asks to queue packet!\n", 4343 dev->name); 4344 } else { 4345 /* Recursion is detected! It is possible, 4346 * unfortunately 4347 */ 4348 recursion_alert: 4349 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", 4350 dev->name); 4351 } 4352 } 4353 4354 rc = -ENETDOWN; 4355 rcu_read_unlock_bh(); 4356 4357 dev_core_stats_tx_dropped_inc(dev); 4358 kfree_skb_list(skb); 4359 return rc; 4360 out: 4361 rcu_read_unlock_bh(); 4362 return rc; 4363 } 4364 EXPORT_SYMBOL(__dev_queue_xmit); 4365 4366 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id) 4367 { 4368 struct net_device *dev = skb->dev; 4369 struct sk_buff *orig_skb = skb; 4370 struct netdev_queue *txq; 4371 int ret = NETDEV_TX_BUSY; 4372 bool again = false; 4373 4374 if (unlikely(!netif_running(dev) || 4375 !netif_carrier_ok(dev))) 4376 goto drop; 4377 4378 skb = validate_xmit_skb_list(skb, dev, &again); 4379 if (skb != orig_skb) 4380 goto drop; 4381 4382 skb_set_queue_mapping(skb, queue_id); 4383 txq = skb_get_tx_queue(dev, skb); 4384 4385 local_bh_disable(); 4386 4387 dev_xmit_recursion_inc(); 4388 HARD_TX_LOCK(dev, txq, smp_processor_id()); 4389 if (!netif_xmit_frozen_or_drv_stopped(txq)) 4390 ret = netdev_start_xmit(skb, dev, txq, false); 4391 HARD_TX_UNLOCK(dev, txq); 4392 dev_xmit_recursion_dec(); 4393 4394 local_bh_enable(); 4395 return ret; 4396 drop: 4397 dev_core_stats_tx_dropped_inc(dev); 4398 kfree_skb_list(skb); 4399 return NET_XMIT_DROP; 4400 } 4401 EXPORT_SYMBOL(__dev_direct_xmit); 4402 4403 /************************************************************************* 4404 * Receiver routines 4405 *************************************************************************/ 4406 4407 int netdev_max_backlog __read_mostly = 1000; 4408 EXPORT_SYMBOL(netdev_max_backlog); 4409 4410 int netdev_tstamp_prequeue __read_mostly = 1; 4411 unsigned int sysctl_skb_defer_max __read_mostly = 64; 4412 int netdev_budget __read_mostly = 300; 4413 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */ 4414 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ; 4415 int weight_p __read_mostly = 64; /* old backlog weight */ 4416 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ 4417 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ 4418 int dev_rx_weight __read_mostly = 64; 4419 int dev_tx_weight __read_mostly = 64; 4420 4421 /* Called with irq disabled */ 4422 static inline void ____napi_schedule(struct softnet_data *sd, 4423 struct napi_struct *napi) 4424 { 4425 struct task_struct *thread; 4426 4427 lockdep_assert_irqs_disabled(); 4428 4429 if (test_bit(NAPI_STATE_THREADED, &napi->state)) { 4430 /* Paired with smp_mb__before_atomic() in 4431 * napi_enable()/dev_set_threaded(). 4432 * Use READ_ONCE() to guarantee a complete 4433 * read on napi->thread. Only call 4434 * wake_up_process() when it's not NULL. 4435 */ 4436 thread = READ_ONCE(napi->thread); 4437 if (thread) { 4438 /* Avoid doing set_bit() if the thread is in 4439 * INTERRUPTIBLE state, cause napi_thread_wait() 4440 * makes sure to proceed with napi polling 4441 * if the thread is explicitly woken from here. 4442 */ 4443 if (READ_ONCE(thread->__state) != TASK_INTERRUPTIBLE) 4444 set_bit(NAPI_STATE_SCHED_THREADED, &napi->state); 4445 wake_up_process(thread); 4446 return; 4447 } 4448 } 4449 4450 list_add_tail(&napi->poll_list, &sd->poll_list); 4451 WRITE_ONCE(napi->list_owner, smp_processor_id()); 4452 /* If not called from net_rx_action() 4453 * we have to raise NET_RX_SOFTIRQ. 4454 */ 4455 if (!sd->in_net_rx_action) 4456 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4457 } 4458 4459 #ifdef CONFIG_RPS 4460 4461 /* One global table that all flow-based protocols share. */ 4462 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; 4463 EXPORT_SYMBOL(rps_sock_flow_table); 4464 u32 rps_cpu_mask __read_mostly; 4465 EXPORT_SYMBOL(rps_cpu_mask); 4466 4467 struct static_key_false rps_needed __read_mostly; 4468 EXPORT_SYMBOL(rps_needed); 4469 struct static_key_false rfs_needed __read_mostly; 4470 EXPORT_SYMBOL(rfs_needed); 4471 4472 static struct rps_dev_flow * 4473 set_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4474 struct rps_dev_flow *rflow, u16 next_cpu) 4475 { 4476 if (next_cpu < nr_cpu_ids) { 4477 #ifdef CONFIG_RFS_ACCEL 4478 struct netdev_rx_queue *rxqueue; 4479 struct rps_dev_flow_table *flow_table; 4480 struct rps_dev_flow *old_rflow; 4481 u32 flow_id; 4482 u16 rxq_index; 4483 int rc; 4484 4485 /* Should we steer this flow to a different hardware queue? */ 4486 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || 4487 !(dev->features & NETIF_F_NTUPLE)) 4488 goto out; 4489 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); 4490 if (rxq_index == skb_get_rx_queue(skb)) 4491 goto out; 4492 4493 rxqueue = dev->_rx + rxq_index; 4494 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4495 if (!flow_table) 4496 goto out; 4497 flow_id = skb_get_hash(skb) & flow_table->mask; 4498 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, 4499 rxq_index, flow_id); 4500 if (rc < 0) 4501 goto out; 4502 old_rflow = rflow; 4503 rflow = &flow_table->flows[flow_id]; 4504 rflow->filter = rc; 4505 if (old_rflow->filter == rflow->filter) 4506 old_rflow->filter = RPS_NO_FILTER; 4507 out: 4508 #endif 4509 rflow->last_qtail = 4510 per_cpu(softnet_data, next_cpu).input_queue_head; 4511 } 4512 4513 rflow->cpu = next_cpu; 4514 return rflow; 4515 } 4516 4517 /* 4518 * get_rps_cpu is called from netif_receive_skb and returns the target 4519 * CPU from the RPS map of the receiving queue for a given skb. 4520 * rcu_read_lock must be held on entry. 4521 */ 4522 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, 4523 struct rps_dev_flow **rflowp) 4524 { 4525 const struct rps_sock_flow_table *sock_flow_table; 4526 struct netdev_rx_queue *rxqueue = dev->_rx; 4527 struct rps_dev_flow_table *flow_table; 4528 struct rps_map *map; 4529 int cpu = -1; 4530 u32 tcpu; 4531 u32 hash; 4532 4533 if (skb_rx_queue_recorded(skb)) { 4534 u16 index = skb_get_rx_queue(skb); 4535 4536 if (unlikely(index >= dev->real_num_rx_queues)) { 4537 WARN_ONCE(dev->real_num_rx_queues > 1, 4538 "%s received packet on queue %u, but number " 4539 "of RX queues is %u\n", 4540 dev->name, index, dev->real_num_rx_queues); 4541 goto done; 4542 } 4543 rxqueue += index; 4544 } 4545 4546 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ 4547 4548 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4549 map = rcu_dereference(rxqueue->rps_map); 4550 if (!flow_table && !map) 4551 goto done; 4552 4553 skb_reset_network_header(skb); 4554 hash = skb_get_hash(skb); 4555 if (!hash) 4556 goto done; 4557 4558 sock_flow_table = rcu_dereference(rps_sock_flow_table); 4559 if (flow_table && sock_flow_table) { 4560 struct rps_dev_flow *rflow; 4561 u32 next_cpu; 4562 u32 ident; 4563 4564 /* First check into global flow table if there is a match. 4565 * This READ_ONCE() pairs with WRITE_ONCE() from rps_record_sock_flow(). 4566 */ 4567 ident = READ_ONCE(sock_flow_table->ents[hash & sock_flow_table->mask]); 4568 if ((ident ^ hash) & ~rps_cpu_mask) 4569 goto try_rps; 4570 4571 next_cpu = ident & rps_cpu_mask; 4572 4573 /* OK, now we know there is a match, 4574 * we can look at the local (per receive queue) flow table 4575 */ 4576 rflow = &flow_table->flows[hash & flow_table->mask]; 4577 tcpu = rflow->cpu; 4578 4579 /* 4580 * If the desired CPU (where last recvmsg was done) is 4581 * different from current CPU (one in the rx-queue flow 4582 * table entry), switch if one of the following holds: 4583 * - Current CPU is unset (>= nr_cpu_ids). 4584 * - Current CPU is offline. 4585 * - The current CPU's queue tail has advanced beyond the 4586 * last packet that was enqueued using this table entry. 4587 * This guarantees that all previous packets for the flow 4588 * have been dequeued, thus preserving in order delivery. 4589 */ 4590 if (unlikely(tcpu != next_cpu) && 4591 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || 4592 ((int)(per_cpu(softnet_data, tcpu).input_queue_head - 4593 rflow->last_qtail)) >= 0)) { 4594 tcpu = next_cpu; 4595 rflow = set_rps_cpu(dev, skb, rflow, next_cpu); 4596 } 4597 4598 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { 4599 *rflowp = rflow; 4600 cpu = tcpu; 4601 goto done; 4602 } 4603 } 4604 4605 try_rps: 4606 4607 if (map) { 4608 tcpu = map->cpus[reciprocal_scale(hash, map->len)]; 4609 if (cpu_online(tcpu)) { 4610 cpu = tcpu; 4611 goto done; 4612 } 4613 } 4614 4615 done: 4616 return cpu; 4617 } 4618 4619 #ifdef CONFIG_RFS_ACCEL 4620 4621 /** 4622 * rps_may_expire_flow - check whether an RFS hardware filter may be removed 4623 * @dev: Device on which the filter was set 4624 * @rxq_index: RX queue index 4625 * @flow_id: Flow ID passed to ndo_rx_flow_steer() 4626 * @filter_id: Filter ID returned by ndo_rx_flow_steer() 4627 * 4628 * Drivers that implement ndo_rx_flow_steer() should periodically call 4629 * this function for each installed filter and remove the filters for 4630 * which it returns %true. 4631 */ 4632 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, 4633 u32 flow_id, u16 filter_id) 4634 { 4635 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; 4636 struct rps_dev_flow_table *flow_table; 4637 struct rps_dev_flow *rflow; 4638 bool expire = true; 4639 unsigned int cpu; 4640 4641 rcu_read_lock(); 4642 flow_table = rcu_dereference(rxqueue->rps_flow_table); 4643 if (flow_table && flow_id <= flow_table->mask) { 4644 rflow = &flow_table->flows[flow_id]; 4645 cpu = READ_ONCE(rflow->cpu); 4646 if (rflow->filter == filter_id && cpu < nr_cpu_ids && 4647 ((int)(per_cpu(softnet_data, cpu).input_queue_head - 4648 rflow->last_qtail) < 4649 (int)(10 * flow_table->mask))) 4650 expire = false; 4651 } 4652 rcu_read_unlock(); 4653 return expire; 4654 } 4655 EXPORT_SYMBOL(rps_may_expire_flow); 4656 4657 #endif /* CONFIG_RFS_ACCEL */ 4658 4659 /* Called from hardirq (IPI) context */ 4660 static void rps_trigger_softirq(void *data) 4661 { 4662 struct softnet_data *sd = data; 4663 4664 ____napi_schedule(sd, &sd->backlog); 4665 sd->received_rps++; 4666 } 4667 4668 #endif /* CONFIG_RPS */ 4669 4670 /* Called from hardirq (IPI) context */ 4671 static void trigger_rx_softirq(void *data) 4672 { 4673 struct softnet_data *sd = data; 4674 4675 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4676 smp_store_release(&sd->defer_ipi_scheduled, 0); 4677 } 4678 4679 /* 4680 * After we queued a packet into sd->input_pkt_queue, 4681 * we need to make sure this queue is serviced soon. 4682 * 4683 * - If this is another cpu queue, link it to our rps_ipi_list, 4684 * and make sure we will process rps_ipi_list from net_rx_action(). 4685 * 4686 * - If this is our own queue, NAPI schedule our backlog. 4687 * Note that this also raises NET_RX_SOFTIRQ. 4688 */ 4689 static void napi_schedule_rps(struct softnet_data *sd) 4690 { 4691 struct softnet_data *mysd = this_cpu_ptr(&softnet_data); 4692 4693 #ifdef CONFIG_RPS 4694 if (sd != mysd) { 4695 sd->rps_ipi_next = mysd->rps_ipi_list; 4696 mysd->rps_ipi_list = sd; 4697 4698 /* If not called from net_rx_action() or napi_threaded_poll() 4699 * we have to raise NET_RX_SOFTIRQ. 4700 */ 4701 if (!mysd->in_net_rx_action && !mysd->in_napi_threaded_poll) 4702 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 4703 return; 4704 } 4705 #endif /* CONFIG_RPS */ 4706 __napi_schedule_irqoff(&mysd->backlog); 4707 } 4708 4709 #ifdef CONFIG_NET_FLOW_LIMIT 4710 int netdev_flow_limit_table_len __read_mostly = (1 << 12); 4711 #endif 4712 4713 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) 4714 { 4715 #ifdef CONFIG_NET_FLOW_LIMIT 4716 struct sd_flow_limit *fl; 4717 struct softnet_data *sd; 4718 unsigned int old_flow, new_flow; 4719 4720 if (qlen < (READ_ONCE(netdev_max_backlog) >> 1)) 4721 return false; 4722 4723 sd = this_cpu_ptr(&softnet_data); 4724 4725 rcu_read_lock(); 4726 fl = rcu_dereference(sd->flow_limit); 4727 if (fl) { 4728 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); 4729 old_flow = fl->history[fl->history_head]; 4730 fl->history[fl->history_head] = new_flow; 4731 4732 fl->history_head++; 4733 fl->history_head &= FLOW_LIMIT_HISTORY - 1; 4734 4735 if (likely(fl->buckets[old_flow])) 4736 fl->buckets[old_flow]--; 4737 4738 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { 4739 fl->count++; 4740 rcu_read_unlock(); 4741 return true; 4742 } 4743 } 4744 rcu_read_unlock(); 4745 #endif 4746 return false; 4747 } 4748 4749 /* 4750 * enqueue_to_backlog is called to queue an skb to a per CPU backlog 4751 * queue (may be a remote CPU queue). 4752 */ 4753 static int enqueue_to_backlog(struct sk_buff *skb, int cpu, 4754 unsigned int *qtail) 4755 { 4756 enum skb_drop_reason reason; 4757 struct softnet_data *sd; 4758 unsigned long flags; 4759 unsigned int qlen; 4760 4761 reason = SKB_DROP_REASON_NOT_SPECIFIED; 4762 sd = &per_cpu(softnet_data, cpu); 4763 4764 rps_lock_irqsave(sd, &flags); 4765 if (!netif_running(skb->dev)) 4766 goto drop; 4767 qlen = skb_queue_len(&sd->input_pkt_queue); 4768 if (qlen <= READ_ONCE(netdev_max_backlog) && !skb_flow_limit(skb, qlen)) { 4769 if (qlen) { 4770 enqueue: 4771 __skb_queue_tail(&sd->input_pkt_queue, skb); 4772 input_queue_tail_incr_save(sd, qtail); 4773 rps_unlock_irq_restore(sd, &flags); 4774 return NET_RX_SUCCESS; 4775 } 4776 4777 /* Schedule NAPI for backlog device 4778 * We can use non atomic operation since we own the queue lock 4779 */ 4780 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) 4781 napi_schedule_rps(sd); 4782 goto enqueue; 4783 } 4784 reason = SKB_DROP_REASON_CPU_BACKLOG; 4785 4786 drop: 4787 sd->dropped++; 4788 rps_unlock_irq_restore(sd, &flags); 4789 4790 dev_core_stats_rx_dropped_inc(skb->dev); 4791 kfree_skb_reason(skb, reason); 4792 return NET_RX_DROP; 4793 } 4794 4795 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb) 4796 { 4797 struct net_device *dev = skb->dev; 4798 struct netdev_rx_queue *rxqueue; 4799 4800 rxqueue = dev->_rx; 4801 4802 if (skb_rx_queue_recorded(skb)) { 4803 u16 index = skb_get_rx_queue(skb); 4804 4805 if (unlikely(index >= dev->real_num_rx_queues)) { 4806 WARN_ONCE(dev->real_num_rx_queues > 1, 4807 "%s received packet on queue %u, but number " 4808 "of RX queues is %u\n", 4809 dev->name, index, dev->real_num_rx_queues); 4810 4811 return rxqueue; /* Return first rxqueue */ 4812 } 4813 rxqueue += index; 4814 } 4815 return rxqueue; 4816 } 4817 4818 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp, 4819 struct bpf_prog *xdp_prog) 4820 { 4821 void *orig_data, *orig_data_end, *hard_start; 4822 struct netdev_rx_queue *rxqueue; 4823 bool orig_bcast, orig_host; 4824 u32 mac_len, frame_sz; 4825 __be16 orig_eth_type; 4826 struct ethhdr *eth; 4827 u32 metalen, act; 4828 int off; 4829 4830 /* The XDP program wants to see the packet starting at the MAC 4831 * header. 4832 */ 4833 mac_len = skb->data - skb_mac_header(skb); 4834 hard_start = skb->data - skb_headroom(skb); 4835 4836 /* SKB "head" area always have tailroom for skb_shared_info */ 4837 frame_sz = (void *)skb_end_pointer(skb) - hard_start; 4838 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 4839 4840 rxqueue = netif_get_rxqueue(skb); 4841 xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq); 4842 xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len, 4843 skb_headlen(skb) + mac_len, true); 4844 if (skb_is_nonlinear(skb)) { 4845 skb_shinfo(skb)->xdp_frags_size = skb->data_len; 4846 xdp_buff_set_frags_flag(xdp); 4847 } else { 4848 xdp_buff_clear_frags_flag(xdp); 4849 } 4850 4851 orig_data_end = xdp->data_end; 4852 orig_data = xdp->data; 4853 eth = (struct ethhdr *)xdp->data; 4854 orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr); 4855 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest); 4856 orig_eth_type = eth->h_proto; 4857 4858 act = bpf_prog_run_xdp(xdp_prog, xdp); 4859 4860 /* check if bpf_xdp_adjust_head was used */ 4861 off = xdp->data - orig_data; 4862 if (off) { 4863 if (off > 0) 4864 __skb_pull(skb, off); 4865 else if (off < 0) 4866 __skb_push(skb, -off); 4867 4868 skb->mac_header += off; 4869 skb_reset_network_header(skb); 4870 } 4871 4872 /* check if bpf_xdp_adjust_tail was used */ 4873 off = xdp->data_end - orig_data_end; 4874 if (off != 0) { 4875 skb_set_tail_pointer(skb, xdp->data_end - xdp->data); 4876 skb->len += off; /* positive on grow, negative on shrink */ 4877 } 4878 4879 /* XDP frag metadata (e.g. nr_frags) are updated in eBPF helpers 4880 * (e.g. bpf_xdp_adjust_tail), we need to update data_len here. 4881 */ 4882 if (xdp_buff_has_frags(xdp)) 4883 skb->data_len = skb_shinfo(skb)->xdp_frags_size; 4884 else 4885 skb->data_len = 0; 4886 4887 /* check if XDP changed eth hdr such SKB needs update */ 4888 eth = (struct ethhdr *)xdp->data; 4889 if ((orig_eth_type != eth->h_proto) || 4890 (orig_host != ether_addr_equal_64bits(eth->h_dest, 4891 skb->dev->dev_addr)) || 4892 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) { 4893 __skb_push(skb, ETH_HLEN); 4894 skb->pkt_type = PACKET_HOST; 4895 skb->protocol = eth_type_trans(skb, skb->dev); 4896 } 4897 4898 /* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull 4899 * before calling us again on redirect path. We do not call do_redirect 4900 * as we leave that up to the caller. 4901 * 4902 * Caller is responsible for managing lifetime of skb (i.e. calling 4903 * kfree_skb in response to actions it cannot handle/XDP_DROP). 4904 */ 4905 switch (act) { 4906 case XDP_REDIRECT: 4907 case XDP_TX: 4908 __skb_push(skb, mac_len); 4909 break; 4910 case XDP_PASS: 4911 metalen = xdp->data - xdp->data_meta; 4912 if (metalen) 4913 skb_metadata_set(skb, metalen); 4914 break; 4915 } 4916 4917 return act; 4918 } 4919 4920 static int 4921 netif_skb_check_for_xdp(struct sk_buff **pskb, struct bpf_prog *prog) 4922 { 4923 struct sk_buff *skb = *pskb; 4924 int err, hroom, troom; 4925 4926 if (!skb_cow_data_for_xdp(this_cpu_read(system_page_pool), pskb, prog)) 4927 return 0; 4928 4929 /* In case we have to go down the path and also linearize, 4930 * then lets do the pskb_expand_head() work just once here. 4931 */ 4932 hroom = XDP_PACKET_HEADROOM - skb_headroom(skb); 4933 troom = skb->tail + skb->data_len - skb->end; 4934 err = pskb_expand_head(skb, 4935 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0, 4936 troom > 0 ? troom + 128 : 0, GFP_ATOMIC); 4937 if (err) 4938 return err; 4939 4940 return skb_linearize(skb); 4941 } 4942 4943 static u32 netif_receive_generic_xdp(struct sk_buff **pskb, 4944 struct xdp_buff *xdp, 4945 struct bpf_prog *xdp_prog) 4946 { 4947 struct sk_buff *skb = *pskb; 4948 u32 mac_len, act = XDP_DROP; 4949 4950 /* Reinjected packets coming from act_mirred or similar should 4951 * not get XDP generic processing. 4952 */ 4953 if (skb_is_redirected(skb)) 4954 return XDP_PASS; 4955 4956 /* XDP packets must have sufficient headroom of XDP_PACKET_HEADROOM 4957 * bytes. This is the guarantee that also native XDP provides, 4958 * thus we need to do it here as well. 4959 */ 4960 mac_len = skb->data - skb_mac_header(skb); 4961 __skb_push(skb, mac_len); 4962 4963 if (skb_cloned(skb) || skb_is_nonlinear(skb) || 4964 skb_headroom(skb) < XDP_PACKET_HEADROOM) { 4965 if (netif_skb_check_for_xdp(pskb, xdp_prog)) 4966 goto do_drop; 4967 } 4968 4969 __skb_pull(*pskb, mac_len); 4970 4971 act = bpf_prog_run_generic_xdp(*pskb, xdp, xdp_prog); 4972 switch (act) { 4973 case XDP_REDIRECT: 4974 case XDP_TX: 4975 case XDP_PASS: 4976 break; 4977 default: 4978 bpf_warn_invalid_xdp_action((*pskb)->dev, xdp_prog, act); 4979 fallthrough; 4980 case XDP_ABORTED: 4981 trace_xdp_exception((*pskb)->dev, xdp_prog, act); 4982 fallthrough; 4983 case XDP_DROP: 4984 do_drop: 4985 kfree_skb(*pskb); 4986 break; 4987 } 4988 4989 return act; 4990 } 4991 4992 /* When doing generic XDP we have to bypass the qdisc layer and the 4993 * network taps in order to match in-driver-XDP behavior. This also means 4994 * that XDP packets are able to starve other packets going through a qdisc, 4995 * and DDOS attacks will be more effective. In-driver-XDP use dedicated TX 4996 * queues, so they do not have this starvation issue. 4997 */ 4998 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog) 4999 { 5000 struct net_device *dev = skb->dev; 5001 struct netdev_queue *txq; 5002 bool free_skb = true; 5003 int cpu, rc; 5004 5005 txq = netdev_core_pick_tx(dev, skb, NULL); 5006 cpu = smp_processor_id(); 5007 HARD_TX_LOCK(dev, txq, cpu); 5008 if (!netif_xmit_frozen_or_drv_stopped(txq)) { 5009 rc = netdev_start_xmit(skb, dev, txq, 0); 5010 if (dev_xmit_complete(rc)) 5011 free_skb = false; 5012 } 5013 HARD_TX_UNLOCK(dev, txq); 5014 if (free_skb) { 5015 trace_xdp_exception(dev, xdp_prog, XDP_TX); 5016 dev_core_stats_tx_dropped_inc(dev); 5017 kfree_skb(skb); 5018 } 5019 } 5020 5021 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key); 5022 5023 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff **pskb) 5024 { 5025 if (xdp_prog) { 5026 struct xdp_buff xdp; 5027 u32 act; 5028 int err; 5029 5030 act = netif_receive_generic_xdp(pskb, &xdp, xdp_prog); 5031 if (act != XDP_PASS) { 5032 switch (act) { 5033 case XDP_REDIRECT: 5034 err = xdp_do_generic_redirect((*pskb)->dev, *pskb, 5035 &xdp, xdp_prog); 5036 if (err) 5037 goto out_redir; 5038 break; 5039 case XDP_TX: 5040 generic_xdp_tx(*pskb, xdp_prog); 5041 break; 5042 } 5043 return XDP_DROP; 5044 } 5045 } 5046 return XDP_PASS; 5047 out_redir: 5048 kfree_skb_reason(*pskb, SKB_DROP_REASON_XDP); 5049 return XDP_DROP; 5050 } 5051 EXPORT_SYMBOL_GPL(do_xdp_generic); 5052 5053 static int netif_rx_internal(struct sk_buff *skb) 5054 { 5055 int ret; 5056 5057 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb); 5058 5059 trace_netif_rx(skb); 5060 5061 #ifdef CONFIG_RPS 5062 if (static_branch_unlikely(&rps_needed)) { 5063 struct rps_dev_flow voidflow, *rflow = &voidflow; 5064 int cpu; 5065 5066 rcu_read_lock(); 5067 5068 cpu = get_rps_cpu(skb->dev, skb, &rflow); 5069 if (cpu < 0) 5070 cpu = smp_processor_id(); 5071 5072 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5073 5074 rcu_read_unlock(); 5075 } else 5076 #endif 5077 { 5078 unsigned int qtail; 5079 5080 ret = enqueue_to_backlog(skb, smp_processor_id(), &qtail); 5081 } 5082 return ret; 5083 } 5084 5085 /** 5086 * __netif_rx - Slightly optimized version of netif_rx 5087 * @skb: buffer to post 5088 * 5089 * This behaves as netif_rx except that it does not disable bottom halves. 5090 * As a result this function may only be invoked from the interrupt context 5091 * (either hard or soft interrupt). 5092 */ 5093 int __netif_rx(struct sk_buff *skb) 5094 { 5095 int ret; 5096 5097 lockdep_assert_once(hardirq_count() | softirq_count()); 5098 5099 trace_netif_rx_entry(skb); 5100 ret = netif_rx_internal(skb); 5101 trace_netif_rx_exit(ret); 5102 return ret; 5103 } 5104 EXPORT_SYMBOL(__netif_rx); 5105 5106 /** 5107 * netif_rx - post buffer to the network code 5108 * @skb: buffer to post 5109 * 5110 * This function receives a packet from a device driver and queues it for 5111 * the upper (protocol) levels to process via the backlog NAPI device. It 5112 * always succeeds. The buffer may be dropped during processing for 5113 * congestion control or by the protocol layers. 5114 * The network buffer is passed via the backlog NAPI device. Modern NIC 5115 * driver should use NAPI and GRO. 5116 * This function can used from interrupt and from process context. The 5117 * caller from process context must not disable interrupts before invoking 5118 * this function. 5119 * 5120 * return values: 5121 * NET_RX_SUCCESS (no congestion) 5122 * NET_RX_DROP (packet was dropped) 5123 * 5124 */ 5125 int netif_rx(struct sk_buff *skb) 5126 { 5127 bool need_bh_off = !(hardirq_count() | softirq_count()); 5128 int ret; 5129 5130 if (need_bh_off) 5131 local_bh_disable(); 5132 trace_netif_rx_entry(skb); 5133 ret = netif_rx_internal(skb); 5134 trace_netif_rx_exit(ret); 5135 if (need_bh_off) 5136 local_bh_enable(); 5137 return ret; 5138 } 5139 EXPORT_SYMBOL(netif_rx); 5140 5141 static __latent_entropy void net_tx_action(struct softirq_action *h) 5142 { 5143 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 5144 5145 if (sd->completion_queue) { 5146 struct sk_buff *clist; 5147 5148 local_irq_disable(); 5149 clist = sd->completion_queue; 5150 sd->completion_queue = NULL; 5151 local_irq_enable(); 5152 5153 while (clist) { 5154 struct sk_buff *skb = clist; 5155 5156 clist = clist->next; 5157 5158 WARN_ON(refcount_read(&skb->users)); 5159 if (likely(get_kfree_skb_cb(skb)->reason == SKB_CONSUMED)) 5160 trace_consume_skb(skb, net_tx_action); 5161 else 5162 trace_kfree_skb(skb, net_tx_action, 5163 get_kfree_skb_cb(skb)->reason); 5164 5165 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) 5166 __kfree_skb(skb); 5167 else 5168 __napi_kfree_skb(skb, 5169 get_kfree_skb_cb(skb)->reason); 5170 } 5171 } 5172 5173 if (sd->output_queue) { 5174 struct Qdisc *head; 5175 5176 local_irq_disable(); 5177 head = sd->output_queue; 5178 sd->output_queue = NULL; 5179 sd->output_queue_tailp = &sd->output_queue; 5180 local_irq_enable(); 5181 5182 rcu_read_lock(); 5183 5184 while (head) { 5185 struct Qdisc *q = head; 5186 spinlock_t *root_lock = NULL; 5187 5188 head = head->next_sched; 5189 5190 /* We need to make sure head->next_sched is read 5191 * before clearing __QDISC_STATE_SCHED 5192 */ 5193 smp_mb__before_atomic(); 5194 5195 if (!(q->flags & TCQ_F_NOLOCK)) { 5196 root_lock = qdisc_lock(q); 5197 spin_lock(root_lock); 5198 } else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, 5199 &q->state))) { 5200 /* There is a synchronize_net() between 5201 * STATE_DEACTIVATED flag being set and 5202 * qdisc_reset()/some_qdisc_is_busy() in 5203 * dev_deactivate(), so we can safely bail out 5204 * early here to avoid data race between 5205 * qdisc_deactivate() and some_qdisc_is_busy() 5206 * for lockless qdisc. 5207 */ 5208 clear_bit(__QDISC_STATE_SCHED, &q->state); 5209 continue; 5210 } 5211 5212 clear_bit(__QDISC_STATE_SCHED, &q->state); 5213 qdisc_run(q); 5214 if (root_lock) 5215 spin_unlock(root_lock); 5216 } 5217 5218 rcu_read_unlock(); 5219 } 5220 5221 xfrm_dev_backlog(sd); 5222 } 5223 5224 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) 5225 /* This hook is defined here for ATM LANE */ 5226 int (*br_fdb_test_addr_hook)(struct net_device *dev, 5227 unsigned char *addr) __read_mostly; 5228 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); 5229 #endif 5230 5231 /** 5232 * netdev_is_rx_handler_busy - check if receive handler is registered 5233 * @dev: device to check 5234 * 5235 * Check if a receive handler is already registered for a given device. 5236 * Return true if there one. 5237 * 5238 * The caller must hold the rtnl_mutex. 5239 */ 5240 bool netdev_is_rx_handler_busy(struct net_device *dev) 5241 { 5242 ASSERT_RTNL(); 5243 return dev && rtnl_dereference(dev->rx_handler); 5244 } 5245 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); 5246 5247 /** 5248 * netdev_rx_handler_register - register receive handler 5249 * @dev: device to register a handler for 5250 * @rx_handler: receive handler to register 5251 * @rx_handler_data: data pointer that is used by rx handler 5252 * 5253 * Register a receive handler for a device. This handler will then be 5254 * called from __netif_receive_skb. A negative errno code is returned 5255 * on a failure. 5256 * 5257 * The caller must hold the rtnl_mutex. 5258 * 5259 * For a general description of rx_handler, see enum rx_handler_result. 5260 */ 5261 int netdev_rx_handler_register(struct net_device *dev, 5262 rx_handler_func_t *rx_handler, 5263 void *rx_handler_data) 5264 { 5265 if (netdev_is_rx_handler_busy(dev)) 5266 return -EBUSY; 5267 5268 if (dev->priv_flags & IFF_NO_RX_HANDLER) 5269 return -EINVAL; 5270 5271 /* Note: rx_handler_data must be set before rx_handler */ 5272 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); 5273 rcu_assign_pointer(dev->rx_handler, rx_handler); 5274 5275 return 0; 5276 } 5277 EXPORT_SYMBOL_GPL(netdev_rx_handler_register); 5278 5279 /** 5280 * netdev_rx_handler_unregister - unregister receive handler 5281 * @dev: device to unregister a handler from 5282 * 5283 * Unregister a receive handler from a device. 5284 * 5285 * The caller must hold the rtnl_mutex. 5286 */ 5287 void netdev_rx_handler_unregister(struct net_device *dev) 5288 { 5289 5290 ASSERT_RTNL(); 5291 RCU_INIT_POINTER(dev->rx_handler, NULL); 5292 /* a reader seeing a non NULL rx_handler in a rcu_read_lock() 5293 * section has a guarantee to see a non NULL rx_handler_data 5294 * as well. 5295 */ 5296 synchronize_net(); 5297 RCU_INIT_POINTER(dev->rx_handler_data, NULL); 5298 } 5299 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); 5300 5301 /* 5302 * Limit the use of PFMEMALLOC reserves to those protocols that implement 5303 * the special handling of PFMEMALLOC skbs. 5304 */ 5305 static bool skb_pfmemalloc_protocol(struct sk_buff *skb) 5306 { 5307 switch (skb->protocol) { 5308 case htons(ETH_P_ARP): 5309 case htons(ETH_P_IP): 5310 case htons(ETH_P_IPV6): 5311 case htons(ETH_P_8021Q): 5312 case htons(ETH_P_8021AD): 5313 return true; 5314 default: 5315 return false; 5316 } 5317 } 5318 5319 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, 5320 int *ret, struct net_device *orig_dev) 5321 { 5322 if (nf_hook_ingress_active(skb)) { 5323 int ingress_retval; 5324 5325 if (*pt_prev) { 5326 *ret = deliver_skb(skb, *pt_prev, orig_dev); 5327 *pt_prev = NULL; 5328 } 5329 5330 rcu_read_lock(); 5331 ingress_retval = nf_hook_ingress(skb); 5332 rcu_read_unlock(); 5333 return ingress_retval; 5334 } 5335 return 0; 5336 } 5337 5338 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc, 5339 struct packet_type **ppt_prev) 5340 { 5341 struct packet_type *ptype, *pt_prev; 5342 rx_handler_func_t *rx_handler; 5343 struct sk_buff *skb = *pskb; 5344 struct net_device *orig_dev; 5345 bool deliver_exact = false; 5346 int ret = NET_RX_DROP; 5347 __be16 type; 5348 5349 net_timestamp_check(!READ_ONCE(netdev_tstamp_prequeue), skb); 5350 5351 trace_netif_receive_skb(skb); 5352 5353 orig_dev = skb->dev; 5354 5355 skb_reset_network_header(skb); 5356 if (!skb_transport_header_was_set(skb)) 5357 skb_reset_transport_header(skb); 5358 skb_reset_mac_len(skb); 5359 5360 pt_prev = NULL; 5361 5362 another_round: 5363 skb->skb_iif = skb->dev->ifindex; 5364 5365 __this_cpu_inc(softnet_data.processed); 5366 5367 if (static_branch_unlikely(&generic_xdp_needed_key)) { 5368 int ret2; 5369 5370 migrate_disable(); 5371 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), 5372 &skb); 5373 migrate_enable(); 5374 5375 if (ret2 != XDP_PASS) { 5376 ret = NET_RX_DROP; 5377 goto out; 5378 } 5379 } 5380 5381 if (eth_type_vlan(skb->protocol)) { 5382 skb = skb_vlan_untag(skb); 5383 if (unlikely(!skb)) 5384 goto out; 5385 } 5386 5387 if (skb_skip_tc_classify(skb)) 5388 goto skip_classify; 5389 5390 if (pfmemalloc) 5391 goto skip_taps; 5392 5393 list_for_each_entry_rcu(ptype, &ptype_all, list) { 5394 if (pt_prev) 5395 ret = deliver_skb(skb, pt_prev, orig_dev); 5396 pt_prev = ptype; 5397 } 5398 5399 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { 5400 if (pt_prev) 5401 ret = deliver_skb(skb, pt_prev, orig_dev); 5402 pt_prev = ptype; 5403 } 5404 5405 skip_taps: 5406 #ifdef CONFIG_NET_INGRESS 5407 if (static_branch_unlikely(&ingress_needed_key)) { 5408 bool another = false; 5409 5410 nf_skip_egress(skb, true); 5411 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev, 5412 &another); 5413 if (another) 5414 goto another_round; 5415 if (!skb) 5416 goto out; 5417 5418 nf_skip_egress(skb, false); 5419 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) 5420 goto out; 5421 } 5422 #endif 5423 skb_reset_redirect(skb); 5424 skip_classify: 5425 if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) 5426 goto drop; 5427 5428 if (skb_vlan_tag_present(skb)) { 5429 if (pt_prev) { 5430 ret = deliver_skb(skb, pt_prev, orig_dev); 5431 pt_prev = NULL; 5432 } 5433 if (vlan_do_receive(&skb)) 5434 goto another_round; 5435 else if (unlikely(!skb)) 5436 goto out; 5437 } 5438 5439 rx_handler = rcu_dereference(skb->dev->rx_handler); 5440 if (rx_handler) { 5441 if (pt_prev) { 5442 ret = deliver_skb(skb, pt_prev, orig_dev); 5443 pt_prev = NULL; 5444 } 5445 switch (rx_handler(&skb)) { 5446 case RX_HANDLER_CONSUMED: 5447 ret = NET_RX_SUCCESS; 5448 goto out; 5449 case RX_HANDLER_ANOTHER: 5450 goto another_round; 5451 case RX_HANDLER_EXACT: 5452 deliver_exact = true; 5453 break; 5454 case RX_HANDLER_PASS: 5455 break; 5456 default: 5457 BUG(); 5458 } 5459 } 5460 5461 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) { 5462 check_vlan_id: 5463 if (skb_vlan_tag_get_id(skb)) { 5464 /* Vlan id is non 0 and vlan_do_receive() above couldn't 5465 * find vlan device. 5466 */ 5467 skb->pkt_type = PACKET_OTHERHOST; 5468 } else if (eth_type_vlan(skb->protocol)) { 5469 /* Outer header is 802.1P with vlan 0, inner header is 5470 * 802.1Q or 802.1AD and vlan_do_receive() above could 5471 * not find vlan dev for vlan id 0. 5472 */ 5473 __vlan_hwaccel_clear_tag(skb); 5474 skb = skb_vlan_untag(skb); 5475 if (unlikely(!skb)) 5476 goto out; 5477 if (vlan_do_receive(&skb)) 5478 /* After stripping off 802.1P header with vlan 0 5479 * vlan dev is found for inner header. 5480 */ 5481 goto another_round; 5482 else if (unlikely(!skb)) 5483 goto out; 5484 else 5485 /* We have stripped outer 802.1P vlan 0 header. 5486 * But could not find vlan dev. 5487 * check again for vlan id to set OTHERHOST. 5488 */ 5489 goto check_vlan_id; 5490 } 5491 /* Note: we might in the future use prio bits 5492 * and set skb->priority like in vlan_do_receive() 5493 * For the time being, just ignore Priority Code Point 5494 */ 5495 __vlan_hwaccel_clear_tag(skb); 5496 } 5497 5498 type = skb->protocol; 5499 5500 /* deliver only exact match when indicated */ 5501 if (likely(!deliver_exact)) { 5502 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5503 &ptype_base[ntohs(type) & 5504 PTYPE_HASH_MASK]); 5505 } 5506 5507 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5508 &orig_dev->ptype_specific); 5509 5510 if (unlikely(skb->dev != orig_dev)) { 5511 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, 5512 &skb->dev->ptype_specific); 5513 } 5514 5515 if (pt_prev) { 5516 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) 5517 goto drop; 5518 *ppt_prev = pt_prev; 5519 } else { 5520 drop: 5521 if (!deliver_exact) 5522 dev_core_stats_rx_dropped_inc(skb->dev); 5523 else 5524 dev_core_stats_rx_nohandler_inc(skb->dev); 5525 kfree_skb_reason(skb, SKB_DROP_REASON_UNHANDLED_PROTO); 5526 /* Jamal, now you will not able to escape explaining 5527 * me how you were going to use this. :-) 5528 */ 5529 ret = NET_RX_DROP; 5530 } 5531 5532 out: 5533 /* The invariant here is that if *ppt_prev is not NULL 5534 * then skb should also be non-NULL. 5535 * 5536 * Apparently *ppt_prev assignment above holds this invariant due to 5537 * skb dereferencing near it. 5538 */ 5539 *pskb = skb; 5540 return ret; 5541 } 5542 5543 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc) 5544 { 5545 struct net_device *orig_dev = skb->dev; 5546 struct packet_type *pt_prev = NULL; 5547 int ret; 5548 5549 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5550 if (pt_prev) 5551 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb, 5552 skb->dev, pt_prev, orig_dev); 5553 return ret; 5554 } 5555 5556 /** 5557 * netif_receive_skb_core - special purpose version of netif_receive_skb 5558 * @skb: buffer to process 5559 * 5560 * More direct receive version of netif_receive_skb(). It should 5561 * only be used by callers that have a need to skip RPS and Generic XDP. 5562 * Caller must also take care of handling if ``(page_is_)pfmemalloc``. 5563 * 5564 * This function may only be called from softirq context and interrupts 5565 * should be enabled. 5566 * 5567 * Return values (usually ignored): 5568 * NET_RX_SUCCESS: no congestion 5569 * NET_RX_DROP: packet was dropped 5570 */ 5571 int netif_receive_skb_core(struct sk_buff *skb) 5572 { 5573 int ret; 5574 5575 rcu_read_lock(); 5576 ret = __netif_receive_skb_one_core(skb, false); 5577 rcu_read_unlock(); 5578 5579 return ret; 5580 } 5581 EXPORT_SYMBOL(netif_receive_skb_core); 5582 5583 static inline void __netif_receive_skb_list_ptype(struct list_head *head, 5584 struct packet_type *pt_prev, 5585 struct net_device *orig_dev) 5586 { 5587 struct sk_buff *skb, *next; 5588 5589 if (!pt_prev) 5590 return; 5591 if (list_empty(head)) 5592 return; 5593 if (pt_prev->list_func != NULL) 5594 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv, 5595 ip_list_rcv, head, pt_prev, orig_dev); 5596 else 5597 list_for_each_entry_safe(skb, next, head, list) { 5598 skb_list_del_init(skb); 5599 pt_prev->func(skb, skb->dev, pt_prev, orig_dev); 5600 } 5601 } 5602 5603 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc) 5604 { 5605 /* Fast-path assumptions: 5606 * - There is no RX handler. 5607 * - Only one packet_type matches. 5608 * If either of these fails, we will end up doing some per-packet 5609 * processing in-line, then handling the 'last ptype' for the whole 5610 * sublist. This can't cause out-of-order delivery to any single ptype, 5611 * because the 'last ptype' must be constant across the sublist, and all 5612 * other ptypes are handled per-packet. 5613 */ 5614 /* Current (common) ptype of sublist */ 5615 struct packet_type *pt_curr = NULL; 5616 /* Current (common) orig_dev of sublist */ 5617 struct net_device *od_curr = NULL; 5618 struct list_head sublist; 5619 struct sk_buff *skb, *next; 5620 5621 INIT_LIST_HEAD(&sublist); 5622 list_for_each_entry_safe(skb, next, head, list) { 5623 struct net_device *orig_dev = skb->dev; 5624 struct packet_type *pt_prev = NULL; 5625 5626 skb_list_del_init(skb); 5627 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev); 5628 if (!pt_prev) 5629 continue; 5630 if (pt_curr != pt_prev || od_curr != orig_dev) { 5631 /* dispatch old sublist */ 5632 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5633 /* start new sublist */ 5634 INIT_LIST_HEAD(&sublist); 5635 pt_curr = pt_prev; 5636 od_curr = orig_dev; 5637 } 5638 list_add_tail(&skb->list, &sublist); 5639 } 5640 5641 /* dispatch final sublist */ 5642 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); 5643 } 5644 5645 static int __netif_receive_skb(struct sk_buff *skb) 5646 { 5647 int ret; 5648 5649 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { 5650 unsigned int noreclaim_flag; 5651 5652 /* 5653 * PFMEMALLOC skbs are special, they should 5654 * - be delivered to SOCK_MEMALLOC sockets only 5655 * - stay away from userspace 5656 * - have bounded memory usage 5657 * 5658 * Use PF_MEMALLOC as this saves us from propagating the allocation 5659 * context down to all allocation sites. 5660 */ 5661 noreclaim_flag = memalloc_noreclaim_save(); 5662 ret = __netif_receive_skb_one_core(skb, true); 5663 memalloc_noreclaim_restore(noreclaim_flag); 5664 } else 5665 ret = __netif_receive_skb_one_core(skb, false); 5666 5667 return ret; 5668 } 5669 5670 static void __netif_receive_skb_list(struct list_head *head) 5671 { 5672 unsigned long noreclaim_flag = 0; 5673 struct sk_buff *skb, *next; 5674 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */ 5675 5676 list_for_each_entry_safe(skb, next, head, list) { 5677 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) { 5678 struct list_head sublist; 5679 5680 /* Handle the previous sublist */ 5681 list_cut_before(&sublist, head, &skb->list); 5682 if (!list_empty(&sublist)) 5683 __netif_receive_skb_list_core(&sublist, pfmemalloc); 5684 pfmemalloc = !pfmemalloc; 5685 /* See comments in __netif_receive_skb */ 5686 if (pfmemalloc) 5687 noreclaim_flag = memalloc_noreclaim_save(); 5688 else 5689 memalloc_noreclaim_restore(noreclaim_flag); 5690 } 5691 } 5692 /* Handle the remaining sublist */ 5693 if (!list_empty(head)) 5694 __netif_receive_skb_list_core(head, pfmemalloc); 5695 /* Restore pflags */ 5696 if (pfmemalloc) 5697 memalloc_noreclaim_restore(noreclaim_flag); 5698 } 5699 5700 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp) 5701 { 5702 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog); 5703 struct bpf_prog *new = xdp->prog; 5704 int ret = 0; 5705 5706 switch (xdp->command) { 5707 case XDP_SETUP_PROG: 5708 rcu_assign_pointer(dev->xdp_prog, new); 5709 if (old) 5710 bpf_prog_put(old); 5711 5712 if (old && !new) { 5713 static_branch_dec(&generic_xdp_needed_key); 5714 } else if (new && !old) { 5715 static_branch_inc(&generic_xdp_needed_key); 5716 dev_disable_lro(dev); 5717 dev_disable_gro_hw(dev); 5718 } 5719 break; 5720 5721 default: 5722 ret = -EINVAL; 5723 break; 5724 } 5725 5726 return ret; 5727 } 5728 5729 static int netif_receive_skb_internal(struct sk_buff *skb) 5730 { 5731 int ret; 5732 5733 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb); 5734 5735 if (skb_defer_rx_timestamp(skb)) 5736 return NET_RX_SUCCESS; 5737 5738 rcu_read_lock(); 5739 #ifdef CONFIG_RPS 5740 if (static_branch_unlikely(&rps_needed)) { 5741 struct rps_dev_flow voidflow, *rflow = &voidflow; 5742 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5743 5744 if (cpu >= 0) { 5745 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5746 rcu_read_unlock(); 5747 return ret; 5748 } 5749 } 5750 #endif 5751 ret = __netif_receive_skb(skb); 5752 rcu_read_unlock(); 5753 return ret; 5754 } 5755 5756 void netif_receive_skb_list_internal(struct list_head *head) 5757 { 5758 struct sk_buff *skb, *next; 5759 struct list_head sublist; 5760 5761 INIT_LIST_HEAD(&sublist); 5762 list_for_each_entry_safe(skb, next, head, list) { 5763 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb); 5764 skb_list_del_init(skb); 5765 if (!skb_defer_rx_timestamp(skb)) 5766 list_add_tail(&skb->list, &sublist); 5767 } 5768 list_splice_init(&sublist, head); 5769 5770 rcu_read_lock(); 5771 #ifdef CONFIG_RPS 5772 if (static_branch_unlikely(&rps_needed)) { 5773 list_for_each_entry_safe(skb, next, head, list) { 5774 struct rps_dev_flow voidflow, *rflow = &voidflow; 5775 int cpu = get_rps_cpu(skb->dev, skb, &rflow); 5776 5777 if (cpu >= 0) { 5778 /* Will be handled, remove from list */ 5779 skb_list_del_init(skb); 5780 enqueue_to_backlog(skb, cpu, &rflow->last_qtail); 5781 } 5782 } 5783 } 5784 #endif 5785 __netif_receive_skb_list(head); 5786 rcu_read_unlock(); 5787 } 5788 5789 /** 5790 * netif_receive_skb - process receive buffer from network 5791 * @skb: buffer to process 5792 * 5793 * netif_receive_skb() is the main receive data processing function. 5794 * It always succeeds. The buffer may be dropped during processing 5795 * for congestion control or by the protocol layers. 5796 * 5797 * This function may only be called from softirq context and interrupts 5798 * should be enabled. 5799 * 5800 * Return values (usually ignored): 5801 * NET_RX_SUCCESS: no congestion 5802 * NET_RX_DROP: packet was dropped 5803 */ 5804 int netif_receive_skb(struct sk_buff *skb) 5805 { 5806 int ret; 5807 5808 trace_netif_receive_skb_entry(skb); 5809 5810 ret = netif_receive_skb_internal(skb); 5811 trace_netif_receive_skb_exit(ret); 5812 5813 return ret; 5814 } 5815 EXPORT_SYMBOL(netif_receive_skb); 5816 5817 /** 5818 * netif_receive_skb_list - process many receive buffers from network 5819 * @head: list of skbs to process. 5820 * 5821 * Since return value of netif_receive_skb() is normally ignored, and 5822 * wouldn't be meaningful for a list, this function returns void. 5823 * 5824 * This function may only be called from softirq context and interrupts 5825 * should be enabled. 5826 */ 5827 void netif_receive_skb_list(struct list_head *head) 5828 { 5829 struct sk_buff *skb; 5830 5831 if (list_empty(head)) 5832 return; 5833 if (trace_netif_receive_skb_list_entry_enabled()) { 5834 list_for_each_entry(skb, head, list) 5835 trace_netif_receive_skb_list_entry(skb); 5836 } 5837 netif_receive_skb_list_internal(head); 5838 trace_netif_receive_skb_list_exit(0); 5839 } 5840 EXPORT_SYMBOL(netif_receive_skb_list); 5841 5842 static DEFINE_PER_CPU(struct work_struct, flush_works); 5843 5844 /* Network device is going away, flush any packets still pending */ 5845 static void flush_backlog(struct work_struct *work) 5846 { 5847 struct sk_buff *skb, *tmp; 5848 struct softnet_data *sd; 5849 5850 local_bh_disable(); 5851 sd = this_cpu_ptr(&softnet_data); 5852 5853 rps_lock_irq_disable(sd); 5854 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { 5855 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5856 __skb_unlink(skb, &sd->input_pkt_queue); 5857 dev_kfree_skb_irq(skb); 5858 input_queue_head_incr(sd); 5859 } 5860 } 5861 rps_unlock_irq_enable(sd); 5862 5863 skb_queue_walk_safe(&sd->process_queue, skb, tmp) { 5864 if (skb->dev->reg_state == NETREG_UNREGISTERING) { 5865 __skb_unlink(skb, &sd->process_queue); 5866 kfree_skb(skb); 5867 input_queue_head_incr(sd); 5868 } 5869 } 5870 local_bh_enable(); 5871 } 5872 5873 static bool flush_required(int cpu) 5874 { 5875 #if IS_ENABLED(CONFIG_RPS) 5876 struct softnet_data *sd = &per_cpu(softnet_data, cpu); 5877 bool do_flush; 5878 5879 rps_lock_irq_disable(sd); 5880 5881 /* as insertion into process_queue happens with the rps lock held, 5882 * process_queue access may race only with dequeue 5883 */ 5884 do_flush = !skb_queue_empty(&sd->input_pkt_queue) || 5885 !skb_queue_empty_lockless(&sd->process_queue); 5886 rps_unlock_irq_enable(sd); 5887 5888 return do_flush; 5889 #endif 5890 /* without RPS we can't safely check input_pkt_queue: during a 5891 * concurrent remote skb_queue_splice() we can detect as empty both 5892 * input_pkt_queue and process_queue even if the latter could end-up 5893 * containing a lot of packets. 5894 */ 5895 return true; 5896 } 5897 5898 static void flush_all_backlogs(void) 5899 { 5900 static cpumask_t flush_cpus; 5901 unsigned int cpu; 5902 5903 /* since we are under rtnl lock protection we can use static data 5904 * for the cpumask and avoid allocating on stack the possibly 5905 * large mask 5906 */ 5907 ASSERT_RTNL(); 5908 5909 cpus_read_lock(); 5910 5911 cpumask_clear(&flush_cpus); 5912 for_each_online_cpu(cpu) { 5913 if (flush_required(cpu)) { 5914 queue_work_on(cpu, system_highpri_wq, 5915 per_cpu_ptr(&flush_works, cpu)); 5916 cpumask_set_cpu(cpu, &flush_cpus); 5917 } 5918 } 5919 5920 /* we can have in flight packet[s] on the cpus we are not flushing, 5921 * synchronize_net() in unregister_netdevice_many() will take care of 5922 * them 5923 */ 5924 for_each_cpu(cpu, &flush_cpus) 5925 flush_work(per_cpu_ptr(&flush_works, cpu)); 5926 5927 cpus_read_unlock(); 5928 } 5929 5930 static void net_rps_send_ipi(struct softnet_data *remsd) 5931 { 5932 #ifdef CONFIG_RPS 5933 while (remsd) { 5934 struct softnet_data *next = remsd->rps_ipi_next; 5935 5936 if (cpu_online(remsd->cpu)) 5937 smp_call_function_single_async(remsd->cpu, &remsd->csd); 5938 remsd = next; 5939 } 5940 #endif 5941 } 5942 5943 /* 5944 * net_rps_action_and_irq_enable sends any pending IPI's for rps. 5945 * Note: called with local irq disabled, but exits with local irq enabled. 5946 */ 5947 static void net_rps_action_and_irq_enable(struct softnet_data *sd) 5948 { 5949 #ifdef CONFIG_RPS 5950 struct softnet_data *remsd = sd->rps_ipi_list; 5951 5952 if (remsd) { 5953 sd->rps_ipi_list = NULL; 5954 5955 local_irq_enable(); 5956 5957 /* Send pending IPI's to kick RPS processing on remote cpus. */ 5958 net_rps_send_ipi(remsd); 5959 } else 5960 #endif 5961 local_irq_enable(); 5962 } 5963 5964 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) 5965 { 5966 #ifdef CONFIG_RPS 5967 return sd->rps_ipi_list != NULL; 5968 #else 5969 return false; 5970 #endif 5971 } 5972 5973 static int process_backlog(struct napi_struct *napi, int quota) 5974 { 5975 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); 5976 bool again = true; 5977 int work = 0; 5978 5979 /* Check if we have pending ipi, its better to send them now, 5980 * not waiting net_rx_action() end. 5981 */ 5982 if (sd_has_rps_ipi_waiting(sd)) { 5983 local_irq_disable(); 5984 net_rps_action_and_irq_enable(sd); 5985 } 5986 5987 napi->weight = READ_ONCE(dev_rx_weight); 5988 while (again) { 5989 struct sk_buff *skb; 5990 5991 while ((skb = __skb_dequeue(&sd->process_queue))) { 5992 rcu_read_lock(); 5993 __netif_receive_skb(skb); 5994 rcu_read_unlock(); 5995 input_queue_head_incr(sd); 5996 if (++work >= quota) 5997 return work; 5998 5999 } 6000 6001 rps_lock_irq_disable(sd); 6002 if (skb_queue_empty(&sd->input_pkt_queue)) { 6003 /* 6004 * Inline a custom version of __napi_complete(). 6005 * only current cpu owns and manipulates this napi, 6006 * and NAPI_STATE_SCHED is the only possible flag set 6007 * on backlog. 6008 * We can use a plain write instead of clear_bit(), 6009 * and we dont need an smp_mb() memory barrier. 6010 */ 6011 napi->state = 0; 6012 again = false; 6013 } else { 6014 skb_queue_splice_tail_init(&sd->input_pkt_queue, 6015 &sd->process_queue); 6016 } 6017 rps_unlock_irq_enable(sd); 6018 } 6019 6020 return work; 6021 } 6022 6023 /** 6024 * __napi_schedule - schedule for receive 6025 * @n: entry to schedule 6026 * 6027 * The entry's receive function will be scheduled to run. 6028 * Consider using __napi_schedule_irqoff() if hard irqs are masked. 6029 */ 6030 void __napi_schedule(struct napi_struct *n) 6031 { 6032 unsigned long flags; 6033 6034 local_irq_save(flags); 6035 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6036 local_irq_restore(flags); 6037 } 6038 EXPORT_SYMBOL(__napi_schedule); 6039 6040 /** 6041 * napi_schedule_prep - check if napi can be scheduled 6042 * @n: napi context 6043 * 6044 * Test if NAPI routine is already running, and if not mark 6045 * it as running. This is used as a condition variable to 6046 * insure only one NAPI poll instance runs. We also make 6047 * sure there is no pending NAPI disable. 6048 */ 6049 bool napi_schedule_prep(struct napi_struct *n) 6050 { 6051 unsigned long new, val = READ_ONCE(n->state); 6052 6053 do { 6054 if (unlikely(val & NAPIF_STATE_DISABLE)) 6055 return false; 6056 new = val | NAPIF_STATE_SCHED; 6057 6058 /* Sets STATE_MISSED bit if STATE_SCHED was already set 6059 * This was suggested by Alexander Duyck, as compiler 6060 * emits better code than : 6061 * if (val & NAPIF_STATE_SCHED) 6062 * new |= NAPIF_STATE_MISSED; 6063 */ 6064 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * 6065 NAPIF_STATE_MISSED; 6066 } while (!try_cmpxchg(&n->state, &val, new)); 6067 6068 return !(val & NAPIF_STATE_SCHED); 6069 } 6070 EXPORT_SYMBOL(napi_schedule_prep); 6071 6072 /** 6073 * __napi_schedule_irqoff - schedule for receive 6074 * @n: entry to schedule 6075 * 6076 * Variant of __napi_schedule() assuming hard irqs are masked. 6077 * 6078 * On PREEMPT_RT enabled kernels this maps to __napi_schedule() 6079 * because the interrupt disabled assumption might not be true 6080 * due to force-threaded interrupts and spinlock substitution. 6081 */ 6082 void __napi_schedule_irqoff(struct napi_struct *n) 6083 { 6084 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6085 ____napi_schedule(this_cpu_ptr(&softnet_data), n); 6086 else 6087 __napi_schedule(n); 6088 } 6089 EXPORT_SYMBOL(__napi_schedule_irqoff); 6090 6091 bool napi_complete_done(struct napi_struct *n, int work_done) 6092 { 6093 unsigned long flags, val, new, timeout = 0; 6094 bool ret = true; 6095 6096 /* 6097 * 1) Don't let napi dequeue from the cpu poll list 6098 * just in case its running on a different cpu. 6099 * 2) If we are busy polling, do nothing here, we have 6100 * the guarantee we will be called later. 6101 */ 6102 if (unlikely(n->state & (NAPIF_STATE_NPSVC | 6103 NAPIF_STATE_IN_BUSY_POLL))) 6104 return false; 6105 6106 if (work_done) { 6107 if (n->gro_bitmask) 6108 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6109 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs); 6110 } 6111 if (n->defer_hard_irqs_count > 0) { 6112 n->defer_hard_irqs_count--; 6113 timeout = READ_ONCE(n->dev->gro_flush_timeout); 6114 if (timeout) 6115 ret = false; 6116 } 6117 if (n->gro_bitmask) { 6118 /* When the NAPI instance uses a timeout and keeps postponing 6119 * it, we need to bound somehow the time packets are kept in 6120 * the GRO layer 6121 */ 6122 napi_gro_flush(n, !!timeout); 6123 } 6124 6125 gro_normal_list(n); 6126 6127 if (unlikely(!list_empty(&n->poll_list))) { 6128 /* If n->poll_list is not empty, we need to mask irqs */ 6129 local_irq_save(flags); 6130 list_del_init(&n->poll_list); 6131 local_irq_restore(flags); 6132 } 6133 WRITE_ONCE(n->list_owner, -1); 6134 6135 val = READ_ONCE(n->state); 6136 do { 6137 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); 6138 6139 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED | 6140 NAPIF_STATE_SCHED_THREADED | 6141 NAPIF_STATE_PREFER_BUSY_POLL); 6142 6143 /* If STATE_MISSED was set, leave STATE_SCHED set, 6144 * because we will call napi->poll() one more time. 6145 * This C code was suggested by Alexander Duyck to help gcc. 6146 */ 6147 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * 6148 NAPIF_STATE_SCHED; 6149 } while (!try_cmpxchg(&n->state, &val, new)); 6150 6151 if (unlikely(val & NAPIF_STATE_MISSED)) { 6152 __napi_schedule(n); 6153 return false; 6154 } 6155 6156 if (timeout) 6157 hrtimer_start(&n->timer, ns_to_ktime(timeout), 6158 HRTIMER_MODE_REL_PINNED); 6159 return ret; 6160 } 6161 EXPORT_SYMBOL(napi_complete_done); 6162 6163 /* must be called under rcu_read_lock(), as we dont take a reference */ 6164 struct napi_struct *napi_by_id(unsigned int napi_id) 6165 { 6166 unsigned int hash = napi_id % HASH_SIZE(napi_hash); 6167 struct napi_struct *napi; 6168 6169 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) 6170 if (napi->napi_id == napi_id) 6171 return napi; 6172 6173 return NULL; 6174 } 6175 6176 static void skb_defer_free_flush(struct softnet_data *sd) 6177 { 6178 struct sk_buff *skb, *next; 6179 6180 /* Paired with WRITE_ONCE() in skb_attempt_defer_free() */ 6181 if (!READ_ONCE(sd->defer_list)) 6182 return; 6183 6184 spin_lock(&sd->defer_lock); 6185 skb = sd->defer_list; 6186 sd->defer_list = NULL; 6187 sd->defer_count = 0; 6188 spin_unlock(&sd->defer_lock); 6189 6190 while (skb != NULL) { 6191 next = skb->next; 6192 napi_consume_skb(skb, 1); 6193 skb = next; 6194 } 6195 } 6196 6197 #if defined(CONFIG_NET_RX_BUSY_POLL) 6198 6199 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule) 6200 { 6201 if (!skip_schedule) { 6202 gro_normal_list(napi); 6203 __napi_schedule(napi); 6204 return; 6205 } 6206 6207 if (napi->gro_bitmask) { 6208 /* flush too old packets 6209 * If HZ < 1000, flush all packets. 6210 */ 6211 napi_gro_flush(napi, HZ >= 1000); 6212 } 6213 6214 gro_normal_list(napi); 6215 clear_bit(NAPI_STATE_SCHED, &napi->state); 6216 } 6217 6218 enum { 6219 NAPI_F_PREFER_BUSY_POLL = 1, 6220 NAPI_F_END_ON_RESCHED = 2, 6221 }; 6222 6223 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, 6224 unsigned flags, u16 budget) 6225 { 6226 bool skip_schedule = false; 6227 unsigned long timeout; 6228 int rc; 6229 6230 /* Busy polling means there is a high chance device driver hard irq 6231 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was 6232 * set in napi_schedule_prep(). 6233 * Since we are about to call napi->poll() once more, we can safely 6234 * clear NAPI_STATE_MISSED. 6235 * 6236 * Note: x86 could use a single "lock and ..." instruction 6237 * to perform these two clear_bit() 6238 */ 6239 clear_bit(NAPI_STATE_MISSED, &napi->state); 6240 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); 6241 6242 local_bh_disable(); 6243 6244 if (flags & NAPI_F_PREFER_BUSY_POLL) { 6245 napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs); 6246 timeout = READ_ONCE(napi->dev->gro_flush_timeout); 6247 if (napi->defer_hard_irqs_count && timeout) { 6248 hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED); 6249 skip_schedule = true; 6250 } 6251 } 6252 6253 /* All we really want here is to re-enable device interrupts. 6254 * Ideally, a new ndo_busy_poll_stop() could avoid another round. 6255 */ 6256 rc = napi->poll(napi, budget); 6257 /* We can't gro_normal_list() here, because napi->poll() might have 6258 * rearmed the napi (napi_complete_done()) in which case it could 6259 * already be running on another CPU. 6260 */ 6261 trace_napi_poll(napi, rc, budget); 6262 netpoll_poll_unlock(have_poll_lock); 6263 if (rc == budget) 6264 __busy_poll_stop(napi, skip_schedule); 6265 local_bh_enable(); 6266 } 6267 6268 static void __napi_busy_loop(unsigned int napi_id, 6269 bool (*loop_end)(void *, unsigned long), 6270 void *loop_end_arg, unsigned flags, u16 budget) 6271 { 6272 unsigned long start_time = loop_end ? busy_loop_current_time() : 0; 6273 int (*napi_poll)(struct napi_struct *napi, int budget); 6274 void *have_poll_lock = NULL; 6275 struct napi_struct *napi; 6276 6277 WARN_ON_ONCE(!rcu_read_lock_held()); 6278 6279 restart: 6280 napi_poll = NULL; 6281 6282 napi = napi_by_id(napi_id); 6283 if (!napi) 6284 return; 6285 6286 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6287 preempt_disable(); 6288 for (;;) { 6289 int work = 0; 6290 6291 local_bh_disable(); 6292 if (!napi_poll) { 6293 unsigned long val = READ_ONCE(napi->state); 6294 6295 /* If multiple threads are competing for this napi, 6296 * we avoid dirtying napi->state as much as we can. 6297 */ 6298 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | 6299 NAPIF_STATE_IN_BUSY_POLL)) { 6300 if (flags & NAPI_F_PREFER_BUSY_POLL) 6301 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6302 goto count; 6303 } 6304 if (cmpxchg(&napi->state, val, 6305 val | NAPIF_STATE_IN_BUSY_POLL | 6306 NAPIF_STATE_SCHED) != val) { 6307 if (flags & NAPI_F_PREFER_BUSY_POLL) 6308 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6309 goto count; 6310 } 6311 have_poll_lock = netpoll_poll_lock(napi); 6312 napi_poll = napi->poll; 6313 } 6314 work = napi_poll(napi, budget); 6315 trace_napi_poll(napi, work, budget); 6316 gro_normal_list(napi); 6317 count: 6318 if (work > 0) 6319 __NET_ADD_STATS(dev_net(napi->dev), 6320 LINUX_MIB_BUSYPOLLRXPACKETS, work); 6321 skb_defer_free_flush(this_cpu_ptr(&softnet_data)); 6322 local_bh_enable(); 6323 6324 if (!loop_end || loop_end(loop_end_arg, start_time)) 6325 break; 6326 6327 if (unlikely(need_resched())) { 6328 if (flags & NAPI_F_END_ON_RESCHED) 6329 break; 6330 if (napi_poll) 6331 busy_poll_stop(napi, have_poll_lock, flags, budget); 6332 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6333 preempt_enable(); 6334 rcu_read_unlock(); 6335 cond_resched(); 6336 rcu_read_lock(); 6337 if (loop_end(loop_end_arg, start_time)) 6338 return; 6339 goto restart; 6340 } 6341 cpu_relax(); 6342 } 6343 if (napi_poll) 6344 busy_poll_stop(napi, have_poll_lock, flags, budget); 6345 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 6346 preempt_enable(); 6347 } 6348 6349 void napi_busy_loop_rcu(unsigned int napi_id, 6350 bool (*loop_end)(void *, unsigned long), 6351 void *loop_end_arg, bool prefer_busy_poll, u16 budget) 6352 { 6353 unsigned flags = NAPI_F_END_ON_RESCHED; 6354 6355 if (prefer_busy_poll) 6356 flags |= NAPI_F_PREFER_BUSY_POLL; 6357 6358 __napi_busy_loop(napi_id, loop_end, loop_end_arg, flags, budget); 6359 } 6360 6361 void napi_busy_loop(unsigned int napi_id, 6362 bool (*loop_end)(void *, unsigned long), 6363 void *loop_end_arg, bool prefer_busy_poll, u16 budget) 6364 { 6365 unsigned flags = prefer_busy_poll ? NAPI_F_PREFER_BUSY_POLL : 0; 6366 6367 rcu_read_lock(); 6368 __napi_busy_loop(napi_id, loop_end, loop_end_arg, flags, budget); 6369 rcu_read_unlock(); 6370 } 6371 EXPORT_SYMBOL(napi_busy_loop); 6372 6373 #endif /* CONFIG_NET_RX_BUSY_POLL */ 6374 6375 static void napi_hash_add(struct napi_struct *napi) 6376 { 6377 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state)) 6378 return; 6379 6380 spin_lock(&napi_hash_lock); 6381 6382 /* 0..NR_CPUS range is reserved for sender_cpu use */ 6383 do { 6384 if (unlikely(++napi_gen_id < MIN_NAPI_ID)) 6385 napi_gen_id = MIN_NAPI_ID; 6386 } while (napi_by_id(napi_gen_id)); 6387 napi->napi_id = napi_gen_id; 6388 6389 hlist_add_head_rcu(&napi->napi_hash_node, 6390 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); 6391 6392 spin_unlock(&napi_hash_lock); 6393 } 6394 6395 /* Warning : caller is responsible to make sure rcu grace period 6396 * is respected before freeing memory containing @napi 6397 */ 6398 static void napi_hash_del(struct napi_struct *napi) 6399 { 6400 spin_lock(&napi_hash_lock); 6401 6402 hlist_del_init_rcu(&napi->napi_hash_node); 6403 6404 spin_unlock(&napi_hash_lock); 6405 } 6406 6407 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) 6408 { 6409 struct napi_struct *napi; 6410 6411 napi = container_of(timer, struct napi_struct, timer); 6412 6413 /* Note : we use a relaxed variant of napi_schedule_prep() not setting 6414 * NAPI_STATE_MISSED, since we do not react to a device IRQ. 6415 */ 6416 if (!napi_disable_pending(napi) && 6417 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) { 6418 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state); 6419 __napi_schedule_irqoff(napi); 6420 } 6421 6422 return HRTIMER_NORESTART; 6423 } 6424 6425 static void init_gro_hash(struct napi_struct *napi) 6426 { 6427 int i; 6428 6429 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6430 INIT_LIST_HEAD(&napi->gro_hash[i].list); 6431 napi->gro_hash[i].count = 0; 6432 } 6433 napi->gro_bitmask = 0; 6434 } 6435 6436 int dev_set_threaded(struct net_device *dev, bool threaded) 6437 { 6438 struct napi_struct *napi; 6439 int err = 0; 6440 6441 if (dev->threaded == threaded) 6442 return 0; 6443 6444 if (threaded) { 6445 list_for_each_entry(napi, &dev->napi_list, dev_list) { 6446 if (!napi->thread) { 6447 err = napi_kthread_create(napi); 6448 if (err) { 6449 threaded = false; 6450 break; 6451 } 6452 } 6453 } 6454 } 6455 6456 dev->threaded = threaded; 6457 6458 /* Make sure kthread is created before THREADED bit 6459 * is set. 6460 */ 6461 smp_mb__before_atomic(); 6462 6463 /* Setting/unsetting threaded mode on a napi might not immediately 6464 * take effect, if the current napi instance is actively being 6465 * polled. In this case, the switch between threaded mode and 6466 * softirq mode will happen in the next round of napi_schedule(). 6467 * This should not cause hiccups/stalls to the live traffic. 6468 */ 6469 list_for_each_entry(napi, &dev->napi_list, dev_list) 6470 assign_bit(NAPI_STATE_THREADED, &napi->state, threaded); 6471 6472 return err; 6473 } 6474 EXPORT_SYMBOL(dev_set_threaded); 6475 6476 /** 6477 * netif_queue_set_napi - Associate queue with the napi 6478 * @dev: device to which NAPI and queue belong 6479 * @queue_index: Index of queue 6480 * @type: queue type as RX or TX 6481 * @napi: NAPI context, pass NULL to clear previously set NAPI 6482 * 6483 * Set queue with its corresponding napi context. This should be done after 6484 * registering the NAPI handler for the queue-vector and the queues have been 6485 * mapped to the corresponding interrupt vector. 6486 */ 6487 void netif_queue_set_napi(struct net_device *dev, unsigned int queue_index, 6488 enum netdev_queue_type type, struct napi_struct *napi) 6489 { 6490 struct netdev_rx_queue *rxq; 6491 struct netdev_queue *txq; 6492 6493 if (WARN_ON_ONCE(napi && !napi->dev)) 6494 return; 6495 if (dev->reg_state >= NETREG_REGISTERED) 6496 ASSERT_RTNL(); 6497 6498 switch (type) { 6499 case NETDEV_QUEUE_TYPE_RX: 6500 rxq = __netif_get_rx_queue(dev, queue_index); 6501 rxq->napi = napi; 6502 return; 6503 case NETDEV_QUEUE_TYPE_TX: 6504 txq = netdev_get_tx_queue(dev, queue_index); 6505 txq->napi = napi; 6506 return; 6507 default: 6508 return; 6509 } 6510 } 6511 EXPORT_SYMBOL(netif_queue_set_napi); 6512 6513 void netif_napi_add_weight(struct net_device *dev, struct napi_struct *napi, 6514 int (*poll)(struct napi_struct *, int), int weight) 6515 { 6516 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state))) 6517 return; 6518 6519 INIT_LIST_HEAD(&napi->poll_list); 6520 INIT_HLIST_NODE(&napi->napi_hash_node); 6521 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); 6522 napi->timer.function = napi_watchdog; 6523 init_gro_hash(napi); 6524 napi->skb = NULL; 6525 INIT_LIST_HEAD(&napi->rx_list); 6526 napi->rx_count = 0; 6527 napi->poll = poll; 6528 if (weight > NAPI_POLL_WEIGHT) 6529 netdev_err_once(dev, "%s() called with weight %d\n", __func__, 6530 weight); 6531 napi->weight = weight; 6532 napi->dev = dev; 6533 #ifdef CONFIG_NETPOLL 6534 napi->poll_owner = -1; 6535 #endif 6536 napi->list_owner = -1; 6537 set_bit(NAPI_STATE_SCHED, &napi->state); 6538 set_bit(NAPI_STATE_NPSVC, &napi->state); 6539 list_add_rcu(&napi->dev_list, &dev->napi_list); 6540 napi_hash_add(napi); 6541 napi_get_frags_check(napi); 6542 /* Create kthread for this napi if dev->threaded is set. 6543 * Clear dev->threaded if kthread creation failed so that 6544 * threaded mode will not be enabled in napi_enable(). 6545 */ 6546 if (dev->threaded && napi_kthread_create(napi)) 6547 dev->threaded = 0; 6548 netif_napi_set_irq(napi, -1); 6549 } 6550 EXPORT_SYMBOL(netif_napi_add_weight); 6551 6552 void napi_disable(struct napi_struct *n) 6553 { 6554 unsigned long val, new; 6555 6556 might_sleep(); 6557 set_bit(NAPI_STATE_DISABLE, &n->state); 6558 6559 val = READ_ONCE(n->state); 6560 do { 6561 while (val & (NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC)) { 6562 usleep_range(20, 200); 6563 val = READ_ONCE(n->state); 6564 } 6565 6566 new = val | NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC; 6567 new &= ~(NAPIF_STATE_THREADED | NAPIF_STATE_PREFER_BUSY_POLL); 6568 } while (!try_cmpxchg(&n->state, &val, new)); 6569 6570 hrtimer_cancel(&n->timer); 6571 6572 clear_bit(NAPI_STATE_DISABLE, &n->state); 6573 } 6574 EXPORT_SYMBOL(napi_disable); 6575 6576 /** 6577 * napi_enable - enable NAPI scheduling 6578 * @n: NAPI context 6579 * 6580 * Resume NAPI from being scheduled on this context. 6581 * Must be paired with napi_disable. 6582 */ 6583 void napi_enable(struct napi_struct *n) 6584 { 6585 unsigned long new, val = READ_ONCE(n->state); 6586 6587 do { 6588 BUG_ON(!test_bit(NAPI_STATE_SCHED, &val)); 6589 6590 new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC); 6591 if (n->dev->threaded && n->thread) 6592 new |= NAPIF_STATE_THREADED; 6593 } while (!try_cmpxchg(&n->state, &val, new)); 6594 } 6595 EXPORT_SYMBOL(napi_enable); 6596 6597 static void flush_gro_hash(struct napi_struct *napi) 6598 { 6599 int i; 6600 6601 for (i = 0; i < GRO_HASH_BUCKETS; i++) { 6602 struct sk_buff *skb, *n; 6603 6604 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list) 6605 kfree_skb(skb); 6606 napi->gro_hash[i].count = 0; 6607 } 6608 } 6609 6610 /* Must be called in process context */ 6611 void __netif_napi_del(struct napi_struct *napi) 6612 { 6613 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state)) 6614 return; 6615 6616 napi_hash_del(napi); 6617 list_del_rcu(&napi->dev_list); 6618 napi_free_frags(napi); 6619 6620 flush_gro_hash(napi); 6621 napi->gro_bitmask = 0; 6622 6623 if (napi->thread) { 6624 kthread_stop(napi->thread); 6625 napi->thread = NULL; 6626 } 6627 } 6628 EXPORT_SYMBOL(__netif_napi_del); 6629 6630 static int __napi_poll(struct napi_struct *n, bool *repoll) 6631 { 6632 int work, weight; 6633 6634 weight = n->weight; 6635 6636 /* This NAPI_STATE_SCHED test is for avoiding a race 6637 * with netpoll's poll_napi(). Only the entity which 6638 * obtains the lock and sees NAPI_STATE_SCHED set will 6639 * actually make the ->poll() call. Therefore we avoid 6640 * accidentally calling ->poll() when NAPI is not scheduled. 6641 */ 6642 work = 0; 6643 if (napi_is_scheduled(n)) { 6644 work = n->poll(n, weight); 6645 trace_napi_poll(n, work, weight); 6646 6647 xdp_do_check_flushed(n); 6648 } 6649 6650 if (unlikely(work > weight)) 6651 netdev_err_once(n->dev, "NAPI poll function %pS returned %d, exceeding its budget of %d.\n", 6652 n->poll, work, weight); 6653 6654 if (likely(work < weight)) 6655 return work; 6656 6657 /* Drivers must not modify the NAPI state if they 6658 * consume the entire weight. In such cases this code 6659 * still "owns" the NAPI instance and therefore can 6660 * move the instance around on the list at-will. 6661 */ 6662 if (unlikely(napi_disable_pending(n))) { 6663 napi_complete(n); 6664 return work; 6665 } 6666 6667 /* The NAPI context has more processing work, but busy-polling 6668 * is preferred. Exit early. 6669 */ 6670 if (napi_prefer_busy_poll(n)) { 6671 if (napi_complete_done(n, work)) { 6672 /* If timeout is not set, we need to make sure 6673 * that the NAPI is re-scheduled. 6674 */ 6675 napi_schedule(n); 6676 } 6677 return work; 6678 } 6679 6680 if (n->gro_bitmask) { 6681 /* flush too old packets 6682 * If HZ < 1000, flush all packets. 6683 */ 6684 napi_gro_flush(n, HZ >= 1000); 6685 } 6686 6687 gro_normal_list(n); 6688 6689 /* Some drivers may have called napi_schedule 6690 * prior to exhausting their budget. 6691 */ 6692 if (unlikely(!list_empty(&n->poll_list))) { 6693 pr_warn_once("%s: Budget exhausted after napi rescheduled\n", 6694 n->dev ? n->dev->name : "backlog"); 6695 return work; 6696 } 6697 6698 *repoll = true; 6699 6700 return work; 6701 } 6702 6703 static int napi_poll(struct napi_struct *n, struct list_head *repoll) 6704 { 6705 bool do_repoll = false; 6706 void *have; 6707 int work; 6708 6709 list_del_init(&n->poll_list); 6710 6711 have = netpoll_poll_lock(n); 6712 6713 work = __napi_poll(n, &do_repoll); 6714 6715 if (do_repoll) 6716 list_add_tail(&n->poll_list, repoll); 6717 6718 netpoll_poll_unlock(have); 6719 6720 return work; 6721 } 6722 6723 static int napi_thread_wait(struct napi_struct *napi) 6724 { 6725 bool woken = false; 6726 6727 set_current_state(TASK_INTERRUPTIBLE); 6728 6729 while (!kthread_should_stop()) { 6730 /* Testing SCHED_THREADED bit here to make sure the current 6731 * kthread owns this napi and could poll on this napi. 6732 * Testing SCHED bit is not enough because SCHED bit might be 6733 * set by some other busy poll thread or by napi_disable(). 6734 */ 6735 if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state) || woken) { 6736 WARN_ON(!list_empty(&napi->poll_list)); 6737 __set_current_state(TASK_RUNNING); 6738 return 0; 6739 } 6740 6741 schedule(); 6742 /* woken being true indicates this thread owns this napi. */ 6743 woken = true; 6744 set_current_state(TASK_INTERRUPTIBLE); 6745 } 6746 __set_current_state(TASK_RUNNING); 6747 6748 return -1; 6749 } 6750 6751 static int napi_threaded_poll(void *data) 6752 { 6753 struct napi_struct *napi = data; 6754 struct softnet_data *sd; 6755 void *have; 6756 6757 while (!napi_thread_wait(napi)) { 6758 for (;;) { 6759 bool repoll = false; 6760 6761 local_bh_disable(); 6762 sd = this_cpu_ptr(&softnet_data); 6763 sd->in_napi_threaded_poll = true; 6764 6765 have = netpoll_poll_lock(napi); 6766 __napi_poll(napi, &repoll); 6767 netpoll_poll_unlock(have); 6768 6769 sd->in_napi_threaded_poll = false; 6770 barrier(); 6771 6772 if (sd_has_rps_ipi_waiting(sd)) { 6773 local_irq_disable(); 6774 net_rps_action_and_irq_enable(sd); 6775 } 6776 skb_defer_free_flush(sd); 6777 local_bh_enable(); 6778 6779 if (!repoll) 6780 break; 6781 6782 cond_resched(); 6783 } 6784 } 6785 return 0; 6786 } 6787 6788 static __latent_entropy void net_rx_action(struct softirq_action *h) 6789 { 6790 struct softnet_data *sd = this_cpu_ptr(&softnet_data); 6791 unsigned long time_limit = jiffies + 6792 usecs_to_jiffies(READ_ONCE(netdev_budget_usecs)); 6793 int budget = READ_ONCE(netdev_budget); 6794 LIST_HEAD(list); 6795 LIST_HEAD(repoll); 6796 6797 start: 6798 sd->in_net_rx_action = true; 6799 local_irq_disable(); 6800 list_splice_init(&sd->poll_list, &list); 6801 local_irq_enable(); 6802 6803 for (;;) { 6804 struct napi_struct *n; 6805 6806 skb_defer_free_flush(sd); 6807 6808 if (list_empty(&list)) { 6809 if (list_empty(&repoll)) { 6810 sd->in_net_rx_action = false; 6811 barrier(); 6812 /* We need to check if ____napi_schedule() 6813 * had refilled poll_list while 6814 * sd->in_net_rx_action was true. 6815 */ 6816 if (!list_empty(&sd->poll_list)) 6817 goto start; 6818 if (!sd_has_rps_ipi_waiting(sd)) 6819 goto end; 6820 } 6821 break; 6822 } 6823 6824 n = list_first_entry(&list, struct napi_struct, poll_list); 6825 budget -= napi_poll(n, &repoll); 6826 6827 /* If softirq window is exhausted then punt. 6828 * Allow this to run for 2 jiffies since which will allow 6829 * an average latency of 1.5/HZ. 6830 */ 6831 if (unlikely(budget <= 0 || 6832 time_after_eq(jiffies, time_limit))) { 6833 sd->time_squeeze++; 6834 break; 6835 } 6836 } 6837 6838 local_irq_disable(); 6839 6840 list_splice_tail_init(&sd->poll_list, &list); 6841 list_splice_tail(&repoll, &list); 6842 list_splice(&list, &sd->poll_list); 6843 if (!list_empty(&sd->poll_list)) 6844 __raise_softirq_irqoff(NET_RX_SOFTIRQ); 6845 else 6846 sd->in_net_rx_action = false; 6847 6848 net_rps_action_and_irq_enable(sd); 6849 end:; 6850 } 6851 6852 struct netdev_adjacent { 6853 struct net_device *dev; 6854 netdevice_tracker dev_tracker; 6855 6856 /* upper master flag, there can only be one master device per list */ 6857 bool master; 6858 6859 /* lookup ignore flag */ 6860 bool ignore; 6861 6862 /* counter for the number of times this device was added to us */ 6863 u16 ref_nr; 6864 6865 /* private field for the users */ 6866 void *private; 6867 6868 struct list_head list; 6869 struct rcu_head rcu; 6870 }; 6871 6872 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, 6873 struct list_head *adj_list) 6874 { 6875 struct netdev_adjacent *adj; 6876 6877 list_for_each_entry(adj, adj_list, list) { 6878 if (adj->dev == adj_dev) 6879 return adj; 6880 } 6881 return NULL; 6882 } 6883 6884 static int ____netdev_has_upper_dev(struct net_device *upper_dev, 6885 struct netdev_nested_priv *priv) 6886 { 6887 struct net_device *dev = (struct net_device *)priv->data; 6888 6889 return upper_dev == dev; 6890 } 6891 6892 /** 6893 * netdev_has_upper_dev - Check if device is linked to an upper device 6894 * @dev: device 6895 * @upper_dev: upper device to check 6896 * 6897 * Find out if a device is linked to specified upper device and return true 6898 * in case it is. Note that this checks only immediate upper device, 6899 * not through a complete stack of devices. The caller must hold the RTNL lock. 6900 */ 6901 bool netdev_has_upper_dev(struct net_device *dev, 6902 struct net_device *upper_dev) 6903 { 6904 struct netdev_nested_priv priv = { 6905 .data = (void *)upper_dev, 6906 }; 6907 6908 ASSERT_RTNL(); 6909 6910 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 6911 &priv); 6912 } 6913 EXPORT_SYMBOL(netdev_has_upper_dev); 6914 6915 /** 6916 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device 6917 * @dev: device 6918 * @upper_dev: upper device to check 6919 * 6920 * Find out if a device is linked to specified upper device and return true 6921 * in case it is. Note that this checks the entire upper device chain. 6922 * The caller must hold rcu lock. 6923 */ 6924 6925 bool netdev_has_upper_dev_all_rcu(struct net_device *dev, 6926 struct net_device *upper_dev) 6927 { 6928 struct netdev_nested_priv priv = { 6929 .data = (void *)upper_dev, 6930 }; 6931 6932 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev, 6933 &priv); 6934 } 6935 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); 6936 6937 /** 6938 * netdev_has_any_upper_dev - Check if device is linked to some device 6939 * @dev: device 6940 * 6941 * Find out if a device is linked to an upper device and return true in case 6942 * it is. The caller must hold the RTNL lock. 6943 */ 6944 bool netdev_has_any_upper_dev(struct net_device *dev) 6945 { 6946 ASSERT_RTNL(); 6947 6948 return !list_empty(&dev->adj_list.upper); 6949 } 6950 EXPORT_SYMBOL(netdev_has_any_upper_dev); 6951 6952 /** 6953 * netdev_master_upper_dev_get - Get master upper device 6954 * @dev: device 6955 * 6956 * Find a master upper device and return pointer to it or NULL in case 6957 * it's not there. The caller must hold the RTNL lock. 6958 */ 6959 struct net_device *netdev_master_upper_dev_get(struct net_device *dev) 6960 { 6961 struct netdev_adjacent *upper; 6962 6963 ASSERT_RTNL(); 6964 6965 if (list_empty(&dev->adj_list.upper)) 6966 return NULL; 6967 6968 upper = list_first_entry(&dev->adj_list.upper, 6969 struct netdev_adjacent, list); 6970 if (likely(upper->master)) 6971 return upper->dev; 6972 return NULL; 6973 } 6974 EXPORT_SYMBOL(netdev_master_upper_dev_get); 6975 6976 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev) 6977 { 6978 struct netdev_adjacent *upper; 6979 6980 ASSERT_RTNL(); 6981 6982 if (list_empty(&dev->adj_list.upper)) 6983 return NULL; 6984 6985 upper = list_first_entry(&dev->adj_list.upper, 6986 struct netdev_adjacent, list); 6987 if (likely(upper->master) && !upper->ignore) 6988 return upper->dev; 6989 return NULL; 6990 } 6991 6992 /** 6993 * netdev_has_any_lower_dev - Check if device is linked to some device 6994 * @dev: device 6995 * 6996 * Find out if a device is linked to a lower device and return true in case 6997 * it is. The caller must hold the RTNL lock. 6998 */ 6999 static bool netdev_has_any_lower_dev(struct net_device *dev) 7000 { 7001 ASSERT_RTNL(); 7002 7003 return !list_empty(&dev->adj_list.lower); 7004 } 7005 7006 void *netdev_adjacent_get_private(struct list_head *adj_list) 7007 { 7008 struct netdev_adjacent *adj; 7009 7010 adj = list_entry(adj_list, struct netdev_adjacent, list); 7011 7012 return adj->private; 7013 } 7014 EXPORT_SYMBOL(netdev_adjacent_get_private); 7015 7016 /** 7017 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list 7018 * @dev: device 7019 * @iter: list_head ** of the current position 7020 * 7021 * Gets the next device from the dev's upper list, starting from iter 7022 * position. The caller must hold RCU read lock. 7023 */ 7024 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, 7025 struct list_head **iter) 7026 { 7027 struct netdev_adjacent *upper; 7028 7029 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 7030 7031 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7032 7033 if (&upper->list == &dev->adj_list.upper) 7034 return NULL; 7035 7036 *iter = &upper->list; 7037 7038 return upper->dev; 7039 } 7040 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); 7041 7042 static struct net_device *__netdev_next_upper_dev(struct net_device *dev, 7043 struct list_head **iter, 7044 bool *ignore) 7045 { 7046 struct netdev_adjacent *upper; 7047 7048 upper = list_entry((*iter)->next, struct netdev_adjacent, list); 7049 7050 if (&upper->list == &dev->adj_list.upper) 7051 return NULL; 7052 7053 *iter = &upper->list; 7054 *ignore = upper->ignore; 7055 7056 return upper->dev; 7057 } 7058 7059 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, 7060 struct list_head **iter) 7061 { 7062 struct netdev_adjacent *upper; 7063 7064 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); 7065 7066 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7067 7068 if (&upper->list == &dev->adj_list.upper) 7069 return NULL; 7070 7071 *iter = &upper->list; 7072 7073 return upper->dev; 7074 } 7075 7076 static int __netdev_walk_all_upper_dev(struct net_device *dev, 7077 int (*fn)(struct net_device *dev, 7078 struct netdev_nested_priv *priv), 7079 struct netdev_nested_priv *priv) 7080 { 7081 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7082 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7083 int ret, cur = 0; 7084 bool ignore; 7085 7086 now = dev; 7087 iter = &dev->adj_list.upper; 7088 7089 while (1) { 7090 if (now != dev) { 7091 ret = fn(now, priv); 7092 if (ret) 7093 return ret; 7094 } 7095 7096 next = NULL; 7097 while (1) { 7098 udev = __netdev_next_upper_dev(now, &iter, &ignore); 7099 if (!udev) 7100 break; 7101 if (ignore) 7102 continue; 7103 7104 next = udev; 7105 niter = &udev->adj_list.upper; 7106 dev_stack[cur] = now; 7107 iter_stack[cur++] = iter; 7108 break; 7109 } 7110 7111 if (!next) { 7112 if (!cur) 7113 return 0; 7114 next = dev_stack[--cur]; 7115 niter = iter_stack[cur]; 7116 } 7117 7118 now = next; 7119 iter = niter; 7120 } 7121 7122 return 0; 7123 } 7124 7125 int netdev_walk_all_upper_dev_rcu(struct net_device *dev, 7126 int (*fn)(struct net_device *dev, 7127 struct netdev_nested_priv *priv), 7128 struct netdev_nested_priv *priv) 7129 { 7130 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7131 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7132 int ret, cur = 0; 7133 7134 now = dev; 7135 iter = &dev->adj_list.upper; 7136 7137 while (1) { 7138 if (now != dev) { 7139 ret = fn(now, priv); 7140 if (ret) 7141 return ret; 7142 } 7143 7144 next = NULL; 7145 while (1) { 7146 udev = netdev_next_upper_dev_rcu(now, &iter); 7147 if (!udev) 7148 break; 7149 7150 next = udev; 7151 niter = &udev->adj_list.upper; 7152 dev_stack[cur] = now; 7153 iter_stack[cur++] = iter; 7154 break; 7155 } 7156 7157 if (!next) { 7158 if (!cur) 7159 return 0; 7160 next = dev_stack[--cur]; 7161 niter = iter_stack[cur]; 7162 } 7163 7164 now = next; 7165 iter = niter; 7166 } 7167 7168 return 0; 7169 } 7170 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); 7171 7172 static bool __netdev_has_upper_dev(struct net_device *dev, 7173 struct net_device *upper_dev) 7174 { 7175 struct netdev_nested_priv priv = { 7176 .flags = 0, 7177 .data = (void *)upper_dev, 7178 }; 7179 7180 ASSERT_RTNL(); 7181 7182 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev, 7183 &priv); 7184 } 7185 7186 /** 7187 * netdev_lower_get_next_private - Get the next ->private from the 7188 * lower neighbour list 7189 * @dev: device 7190 * @iter: list_head ** of the current position 7191 * 7192 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7193 * list, starting from iter position. The caller must hold either hold the 7194 * RTNL lock or its own locking that guarantees that the neighbour lower 7195 * list will remain unchanged. 7196 */ 7197 void *netdev_lower_get_next_private(struct net_device *dev, 7198 struct list_head **iter) 7199 { 7200 struct netdev_adjacent *lower; 7201 7202 lower = list_entry(*iter, struct netdev_adjacent, list); 7203 7204 if (&lower->list == &dev->adj_list.lower) 7205 return NULL; 7206 7207 *iter = lower->list.next; 7208 7209 return lower->private; 7210 } 7211 EXPORT_SYMBOL(netdev_lower_get_next_private); 7212 7213 /** 7214 * netdev_lower_get_next_private_rcu - Get the next ->private from the 7215 * lower neighbour list, RCU 7216 * variant 7217 * @dev: device 7218 * @iter: list_head ** of the current position 7219 * 7220 * Gets the next netdev_adjacent->private from the dev's lower neighbour 7221 * list, starting from iter position. The caller must hold RCU read lock. 7222 */ 7223 void *netdev_lower_get_next_private_rcu(struct net_device *dev, 7224 struct list_head **iter) 7225 { 7226 struct netdev_adjacent *lower; 7227 7228 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 7229 7230 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7231 7232 if (&lower->list == &dev->adj_list.lower) 7233 return NULL; 7234 7235 *iter = &lower->list; 7236 7237 return lower->private; 7238 } 7239 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); 7240 7241 /** 7242 * netdev_lower_get_next - Get the next device from the lower neighbour 7243 * list 7244 * @dev: device 7245 * @iter: list_head ** of the current position 7246 * 7247 * Gets the next netdev_adjacent from the dev's lower neighbour 7248 * list, starting from iter position. The caller must hold RTNL lock or 7249 * its own locking that guarantees that the neighbour lower 7250 * list will remain unchanged. 7251 */ 7252 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) 7253 { 7254 struct netdev_adjacent *lower; 7255 7256 lower = list_entry(*iter, struct netdev_adjacent, list); 7257 7258 if (&lower->list == &dev->adj_list.lower) 7259 return NULL; 7260 7261 *iter = lower->list.next; 7262 7263 return lower->dev; 7264 } 7265 EXPORT_SYMBOL(netdev_lower_get_next); 7266 7267 static struct net_device *netdev_next_lower_dev(struct net_device *dev, 7268 struct list_head **iter) 7269 { 7270 struct netdev_adjacent *lower; 7271 7272 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7273 7274 if (&lower->list == &dev->adj_list.lower) 7275 return NULL; 7276 7277 *iter = &lower->list; 7278 7279 return lower->dev; 7280 } 7281 7282 static struct net_device *__netdev_next_lower_dev(struct net_device *dev, 7283 struct list_head **iter, 7284 bool *ignore) 7285 { 7286 struct netdev_adjacent *lower; 7287 7288 lower = list_entry((*iter)->next, struct netdev_adjacent, list); 7289 7290 if (&lower->list == &dev->adj_list.lower) 7291 return NULL; 7292 7293 *iter = &lower->list; 7294 *ignore = lower->ignore; 7295 7296 return lower->dev; 7297 } 7298 7299 int netdev_walk_all_lower_dev(struct net_device *dev, 7300 int (*fn)(struct net_device *dev, 7301 struct netdev_nested_priv *priv), 7302 struct netdev_nested_priv *priv) 7303 { 7304 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7305 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7306 int ret, cur = 0; 7307 7308 now = dev; 7309 iter = &dev->adj_list.lower; 7310 7311 while (1) { 7312 if (now != dev) { 7313 ret = fn(now, priv); 7314 if (ret) 7315 return ret; 7316 } 7317 7318 next = NULL; 7319 while (1) { 7320 ldev = netdev_next_lower_dev(now, &iter); 7321 if (!ldev) 7322 break; 7323 7324 next = ldev; 7325 niter = &ldev->adj_list.lower; 7326 dev_stack[cur] = now; 7327 iter_stack[cur++] = iter; 7328 break; 7329 } 7330 7331 if (!next) { 7332 if (!cur) 7333 return 0; 7334 next = dev_stack[--cur]; 7335 niter = iter_stack[cur]; 7336 } 7337 7338 now = next; 7339 iter = niter; 7340 } 7341 7342 return 0; 7343 } 7344 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); 7345 7346 static int __netdev_walk_all_lower_dev(struct net_device *dev, 7347 int (*fn)(struct net_device *dev, 7348 struct netdev_nested_priv *priv), 7349 struct netdev_nested_priv *priv) 7350 { 7351 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7352 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7353 int ret, cur = 0; 7354 bool ignore; 7355 7356 now = dev; 7357 iter = &dev->adj_list.lower; 7358 7359 while (1) { 7360 if (now != dev) { 7361 ret = fn(now, priv); 7362 if (ret) 7363 return ret; 7364 } 7365 7366 next = NULL; 7367 while (1) { 7368 ldev = __netdev_next_lower_dev(now, &iter, &ignore); 7369 if (!ldev) 7370 break; 7371 if (ignore) 7372 continue; 7373 7374 next = ldev; 7375 niter = &ldev->adj_list.lower; 7376 dev_stack[cur] = now; 7377 iter_stack[cur++] = iter; 7378 break; 7379 } 7380 7381 if (!next) { 7382 if (!cur) 7383 return 0; 7384 next = dev_stack[--cur]; 7385 niter = iter_stack[cur]; 7386 } 7387 7388 now = next; 7389 iter = niter; 7390 } 7391 7392 return 0; 7393 } 7394 7395 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, 7396 struct list_head **iter) 7397 { 7398 struct netdev_adjacent *lower; 7399 7400 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); 7401 if (&lower->list == &dev->adj_list.lower) 7402 return NULL; 7403 7404 *iter = &lower->list; 7405 7406 return lower->dev; 7407 } 7408 EXPORT_SYMBOL(netdev_next_lower_dev_rcu); 7409 7410 static u8 __netdev_upper_depth(struct net_device *dev) 7411 { 7412 struct net_device *udev; 7413 struct list_head *iter; 7414 u8 max_depth = 0; 7415 bool ignore; 7416 7417 for (iter = &dev->adj_list.upper, 7418 udev = __netdev_next_upper_dev(dev, &iter, &ignore); 7419 udev; 7420 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) { 7421 if (ignore) 7422 continue; 7423 if (max_depth < udev->upper_level) 7424 max_depth = udev->upper_level; 7425 } 7426 7427 return max_depth; 7428 } 7429 7430 static u8 __netdev_lower_depth(struct net_device *dev) 7431 { 7432 struct net_device *ldev; 7433 struct list_head *iter; 7434 u8 max_depth = 0; 7435 bool ignore; 7436 7437 for (iter = &dev->adj_list.lower, 7438 ldev = __netdev_next_lower_dev(dev, &iter, &ignore); 7439 ldev; 7440 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) { 7441 if (ignore) 7442 continue; 7443 if (max_depth < ldev->lower_level) 7444 max_depth = ldev->lower_level; 7445 } 7446 7447 return max_depth; 7448 } 7449 7450 static int __netdev_update_upper_level(struct net_device *dev, 7451 struct netdev_nested_priv *__unused) 7452 { 7453 dev->upper_level = __netdev_upper_depth(dev) + 1; 7454 return 0; 7455 } 7456 7457 #ifdef CONFIG_LOCKDEP 7458 static LIST_HEAD(net_unlink_list); 7459 7460 static void net_unlink_todo(struct net_device *dev) 7461 { 7462 if (list_empty(&dev->unlink_list)) 7463 list_add_tail(&dev->unlink_list, &net_unlink_list); 7464 } 7465 #endif 7466 7467 static int __netdev_update_lower_level(struct net_device *dev, 7468 struct netdev_nested_priv *priv) 7469 { 7470 dev->lower_level = __netdev_lower_depth(dev) + 1; 7471 7472 #ifdef CONFIG_LOCKDEP 7473 if (!priv) 7474 return 0; 7475 7476 if (priv->flags & NESTED_SYNC_IMM) 7477 dev->nested_level = dev->lower_level - 1; 7478 if (priv->flags & NESTED_SYNC_TODO) 7479 net_unlink_todo(dev); 7480 #endif 7481 return 0; 7482 } 7483 7484 int netdev_walk_all_lower_dev_rcu(struct net_device *dev, 7485 int (*fn)(struct net_device *dev, 7486 struct netdev_nested_priv *priv), 7487 struct netdev_nested_priv *priv) 7488 { 7489 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 7490 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 7491 int ret, cur = 0; 7492 7493 now = dev; 7494 iter = &dev->adj_list.lower; 7495 7496 while (1) { 7497 if (now != dev) { 7498 ret = fn(now, priv); 7499 if (ret) 7500 return ret; 7501 } 7502 7503 next = NULL; 7504 while (1) { 7505 ldev = netdev_next_lower_dev_rcu(now, &iter); 7506 if (!ldev) 7507 break; 7508 7509 next = ldev; 7510 niter = &ldev->adj_list.lower; 7511 dev_stack[cur] = now; 7512 iter_stack[cur++] = iter; 7513 break; 7514 } 7515 7516 if (!next) { 7517 if (!cur) 7518 return 0; 7519 next = dev_stack[--cur]; 7520 niter = iter_stack[cur]; 7521 } 7522 7523 now = next; 7524 iter = niter; 7525 } 7526 7527 return 0; 7528 } 7529 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); 7530 7531 /** 7532 * netdev_lower_get_first_private_rcu - Get the first ->private from the 7533 * lower neighbour list, RCU 7534 * variant 7535 * @dev: device 7536 * 7537 * Gets the first netdev_adjacent->private from the dev's lower neighbour 7538 * list. The caller must hold RCU read lock. 7539 */ 7540 void *netdev_lower_get_first_private_rcu(struct net_device *dev) 7541 { 7542 struct netdev_adjacent *lower; 7543 7544 lower = list_first_or_null_rcu(&dev->adj_list.lower, 7545 struct netdev_adjacent, list); 7546 if (lower) 7547 return lower->private; 7548 return NULL; 7549 } 7550 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); 7551 7552 /** 7553 * netdev_master_upper_dev_get_rcu - Get master upper device 7554 * @dev: device 7555 * 7556 * Find a master upper device and return pointer to it or NULL in case 7557 * it's not there. The caller must hold the RCU read lock. 7558 */ 7559 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) 7560 { 7561 struct netdev_adjacent *upper; 7562 7563 upper = list_first_or_null_rcu(&dev->adj_list.upper, 7564 struct netdev_adjacent, list); 7565 if (upper && likely(upper->master)) 7566 return upper->dev; 7567 return NULL; 7568 } 7569 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); 7570 7571 static int netdev_adjacent_sysfs_add(struct net_device *dev, 7572 struct net_device *adj_dev, 7573 struct list_head *dev_list) 7574 { 7575 char linkname[IFNAMSIZ+7]; 7576 7577 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7578 "upper_%s" : "lower_%s", adj_dev->name); 7579 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), 7580 linkname); 7581 } 7582 static void netdev_adjacent_sysfs_del(struct net_device *dev, 7583 char *name, 7584 struct list_head *dev_list) 7585 { 7586 char linkname[IFNAMSIZ+7]; 7587 7588 sprintf(linkname, dev_list == &dev->adj_list.upper ? 7589 "upper_%s" : "lower_%s", name); 7590 sysfs_remove_link(&(dev->dev.kobj), linkname); 7591 } 7592 7593 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, 7594 struct net_device *adj_dev, 7595 struct list_head *dev_list) 7596 { 7597 return (dev_list == &dev->adj_list.upper || 7598 dev_list == &dev->adj_list.lower) && 7599 net_eq(dev_net(dev), dev_net(adj_dev)); 7600 } 7601 7602 static int __netdev_adjacent_dev_insert(struct net_device *dev, 7603 struct net_device *adj_dev, 7604 struct list_head *dev_list, 7605 void *private, bool master) 7606 { 7607 struct netdev_adjacent *adj; 7608 int ret; 7609 7610 adj = __netdev_find_adj(adj_dev, dev_list); 7611 7612 if (adj) { 7613 adj->ref_nr += 1; 7614 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", 7615 dev->name, adj_dev->name, adj->ref_nr); 7616 7617 return 0; 7618 } 7619 7620 adj = kmalloc(sizeof(*adj), GFP_KERNEL); 7621 if (!adj) 7622 return -ENOMEM; 7623 7624 adj->dev = adj_dev; 7625 adj->master = master; 7626 adj->ref_nr = 1; 7627 adj->private = private; 7628 adj->ignore = false; 7629 netdev_hold(adj_dev, &adj->dev_tracker, GFP_KERNEL); 7630 7631 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", 7632 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); 7633 7634 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { 7635 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); 7636 if (ret) 7637 goto free_adj; 7638 } 7639 7640 /* Ensure that master link is always the first item in list. */ 7641 if (master) { 7642 ret = sysfs_create_link(&(dev->dev.kobj), 7643 &(adj_dev->dev.kobj), "master"); 7644 if (ret) 7645 goto remove_symlinks; 7646 7647 list_add_rcu(&adj->list, dev_list); 7648 } else { 7649 list_add_tail_rcu(&adj->list, dev_list); 7650 } 7651 7652 return 0; 7653 7654 remove_symlinks: 7655 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7656 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7657 free_adj: 7658 netdev_put(adj_dev, &adj->dev_tracker); 7659 kfree(adj); 7660 7661 return ret; 7662 } 7663 7664 static void __netdev_adjacent_dev_remove(struct net_device *dev, 7665 struct net_device *adj_dev, 7666 u16 ref_nr, 7667 struct list_head *dev_list) 7668 { 7669 struct netdev_adjacent *adj; 7670 7671 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", 7672 dev->name, adj_dev->name, ref_nr); 7673 7674 adj = __netdev_find_adj(adj_dev, dev_list); 7675 7676 if (!adj) { 7677 pr_err("Adjacency does not exist for device %s from %s\n", 7678 dev->name, adj_dev->name); 7679 WARN_ON(1); 7680 return; 7681 } 7682 7683 if (adj->ref_nr > ref_nr) { 7684 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", 7685 dev->name, adj_dev->name, ref_nr, 7686 adj->ref_nr - ref_nr); 7687 adj->ref_nr -= ref_nr; 7688 return; 7689 } 7690 7691 if (adj->master) 7692 sysfs_remove_link(&(dev->dev.kobj), "master"); 7693 7694 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) 7695 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); 7696 7697 list_del_rcu(&adj->list); 7698 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", 7699 adj_dev->name, dev->name, adj_dev->name); 7700 netdev_put(adj_dev, &adj->dev_tracker); 7701 kfree_rcu(adj, rcu); 7702 } 7703 7704 static int __netdev_adjacent_dev_link_lists(struct net_device *dev, 7705 struct net_device *upper_dev, 7706 struct list_head *up_list, 7707 struct list_head *down_list, 7708 void *private, bool master) 7709 { 7710 int ret; 7711 7712 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, 7713 private, master); 7714 if (ret) 7715 return ret; 7716 7717 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, 7718 private, false); 7719 if (ret) { 7720 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); 7721 return ret; 7722 } 7723 7724 return 0; 7725 } 7726 7727 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, 7728 struct net_device *upper_dev, 7729 u16 ref_nr, 7730 struct list_head *up_list, 7731 struct list_head *down_list) 7732 { 7733 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); 7734 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); 7735 } 7736 7737 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, 7738 struct net_device *upper_dev, 7739 void *private, bool master) 7740 { 7741 return __netdev_adjacent_dev_link_lists(dev, upper_dev, 7742 &dev->adj_list.upper, 7743 &upper_dev->adj_list.lower, 7744 private, master); 7745 } 7746 7747 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, 7748 struct net_device *upper_dev) 7749 { 7750 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, 7751 &dev->adj_list.upper, 7752 &upper_dev->adj_list.lower); 7753 } 7754 7755 static int __netdev_upper_dev_link(struct net_device *dev, 7756 struct net_device *upper_dev, bool master, 7757 void *upper_priv, void *upper_info, 7758 struct netdev_nested_priv *priv, 7759 struct netlink_ext_ack *extack) 7760 { 7761 struct netdev_notifier_changeupper_info changeupper_info = { 7762 .info = { 7763 .dev = dev, 7764 .extack = extack, 7765 }, 7766 .upper_dev = upper_dev, 7767 .master = master, 7768 .linking = true, 7769 .upper_info = upper_info, 7770 }; 7771 struct net_device *master_dev; 7772 int ret = 0; 7773 7774 ASSERT_RTNL(); 7775 7776 if (dev == upper_dev) 7777 return -EBUSY; 7778 7779 /* To prevent loops, check if dev is not upper device to upper_dev. */ 7780 if (__netdev_has_upper_dev(upper_dev, dev)) 7781 return -EBUSY; 7782 7783 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV) 7784 return -EMLINK; 7785 7786 if (!master) { 7787 if (__netdev_has_upper_dev(dev, upper_dev)) 7788 return -EEXIST; 7789 } else { 7790 master_dev = __netdev_master_upper_dev_get(dev); 7791 if (master_dev) 7792 return master_dev == upper_dev ? -EEXIST : -EBUSY; 7793 } 7794 7795 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7796 &changeupper_info.info); 7797 ret = notifier_to_errno(ret); 7798 if (ret) 7799 return ret; 7800 7801 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, 7802 master); 7803 if (ret) 7804 return ret; 7805 7806 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 7807 &changeupper_info.info); 7808 ret = notifier_to_errno(ret); 7809 if (ret) 7810 goto rollback; 7811 7812 __netdev_update_upper_level(dev, NULL); 7813 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 7814 7815 __netdev_update_lower_level(upper_dev, priv); 7816 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 7817 priv); 7818 7819 return 0; 7820 7821 rollback: 7822 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 7823 7824 return ret; 7825 } 7826 7827 /** 7828 * netdev_upper_dev_link - Add a link to the upper device 7829 * @dev: device 7830 * @upper_dev: new upper device 7831 * @extack: netlink extended ack 7832 * 7833 * Adds a link to device which is upper to this one. The caller must hold 7834 * the RTNL lock. On a failure a negative errno code is returned. 7835 * On success the reference counts are adjusted and the function 7836 * returns zero. 7837 */ 7838 int netdev_upper_dev_link(struct net_device *dev, 7839 struct net_device *upper_dev, 7840 struct netlink_ext_ack *extack) 7841 { 7842 struct netdev_nested_priv priv = { 7843 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7844 .data = NULL, 7845 }; 7846 7847 return __netdev_upper_dev_link(dev, upper_dev, false, 7848 NULL, NULL, &priv, extack); 7849 } 7850 EXPORT_SYMBOL(netdev_upper_dev_link); 7851 7852 /** 7853 * netdev_master_upper_dev_link - Add a master link to the upper device 7854 * @dev: device 7855 * @upper_dev: new upper device 7856 * @upper_priv: upper device private 7857 * @upper_info: upper info to be passed down via notifier 7858 * @extack: netlink extended ack 7859 * 7860 * Adds a link to device which is upper to this one. In this case, only 7861 * one master upper device can be linked, although other non-master devices 7862 * might be linked as well. The caller must hold the RTNL lock. 7863 * On a failure a negative errno code is returned. On success the reference 7864 * counts are adjusted and the function returns zero. 7865 */ 7866 int netdev_master_upper_dev_link(struct net_device *dev, 7867 struct net_device *upper_dev, 7868 void *upper_priv, void *upper_info, 7869 struct netlink_ext_ack *extack) 7870 { 7871 struct netdev_nested_priv priv = { 7872 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7873 .data = NULL, 7874 }; 7875 7876 return __netdev_upper_dev_link(dev, upper_dev, true, 7877 upper_priv, upper_info, &priv, extack); 7878 } 7879 EXPORT_SYMBOL(netdev_master_upper_dev_link); 7880 7881 static void __netdev_upper_dev_unlink(struct net_device *dev, 7882 struct net_device *upper_dev, 7883 struct netdev_nested_priv *priv) 7884 { 7885 struct netdev_notifier_changeupper_info changeupper_info = { 7886 .info = { 7887 .dev = dev, 7888 }, 7889 .upper_dev = upper_dev, 7890 .linking = false, 7891 }; 7892 7893 ASSERT_RTNL(); 7894 7895 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; 7896 7897 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, 7898 &changeupper_info.info); 7899 7900 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); 7901 7902 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, 7903 &changeupper_info.info); 7904 7905 __netdev_update_upper_level(dev, NULL); 7906 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL); 7907 7908 __netdev_update_lower_level(upper_dev, priv); 7909 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, 7910 priv); 7911 } 7912 7913 /** 7914 * netdev_upper_dev_unlink - Removes a link to upper device 7915 * @dev: device 7916 * @upper_dev: new upper device 7917 * 7918 * Removes a link to device which is upper to this one. The caller must hold 7919 * the RTNL lock. 7920 */ 7921 void netdev_upper_dev_unlink(struct net_device *dev, 7922 struct net_device *upper_dev) 7923 { 7924 struct netdev_nested_priv priv = { 7925 .flags = NESTED_SYNC_TODO, 7926 .data = NULL, 7927 }; 7928 7929 __netdev_upper_dev_unlink(dev, upper_dev, &priv); 7930 } 7931 EXPORT_SYMBOL(netdev_upper_dev_unlink); 7932 7933 static void __netdev_adjacent_dev_set(struct net_device *upper_dev, 7934 struct net_device *lower_dev, 7935 bool val) 7936 { 7937 struct netdev_adjacent *adj; 7938 7939 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower); 7940 if (adj) 7941 adj->ignore = val; 7942 7943 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper); 7944 if (adj) 7945 adj->ignore = val; 7946 } 7947 7948 static void netdev_adjacent_dev_disable(struct net_device *upper_dev, 7949 struct net_device *lower_dev) 7950 { 7951 __netdev_adjacent_dev_set(upper_dev, lower_dev, true); 7952 } 7953 7954 static void netdev_adjacent_dev_enable(struct net_device *upper_dev, 7955 struct net_device *lower_dev) 7956 { 7957 __netdev_adjacent_dev_set(upper_dev, lower_dev, false); 7958 } 7959 7960 int netdev_adjacent_change_prepare(struct net_device *old_dev, 7961 struct net_device *new_dev, 7962 struct net_device *dev, 7963 struct netlink_ext_ack *extack) 7964 { 7965 struct netdev_nested_priv priv = { 7966 .flags = 0, 7967 .data = NULL, 7968 }; 7969 int err; 7970 7971 if (!new_dev) 7972 return 0; 7973 7974 if (old_dev && new_dev != old_dev) 7975 netdev_adjacent_dev_disable(dev, old_dev); 7976 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv, 7977 extack); 7978 if (err) { 7979 if (old_dev && new_dev != old_dev) 7980 netdev_adjacent_dev_enable(dev, old_dev); 7981 return err; 7982 } 7983 7984 return 0; 7985 } 7986 EXPORT_SYMBOL(netdev_adjacent_change_prepare); 7987 7988 void netdev_adjacent_change_commit(struct net_device *old_dev, 7989 struct net_device *new_dev, 7990 struct net_device *dev) 7991 { 7992 struct netdev_nested_priv priv = { 7993 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO, 7994 .data = NULL, 7995 }; 7996 7997 if (!new_dev || !old_dev) 7998 return; 7999 8000 if (new_dev == old_dev) 8001 return; 8002 8003 netdev_adjacent_dev_enable(dev, old_dev); 8004 __netdev_upper_dev_unlink(old_dev, dev, &priv); 8005 } 8006 EXPORT_SYMBOL(netdev_adjacent_change_commit); 8007 8008 void netdev_adjacent_change_abort(struct net_device *old_dev, 8009 struct net_device *new_dev, 8010 struct net_device *dev) 8011 { 8012 struct netdev_nested_priv priv = { 8013 .flags = 0, 8014 .data = NULL, 8015 }; 8016 8017 if (!new_dev) 8018 return; 8019 8020 if (old_dev && new_dev != old_dev) 8021 netdev_adjacent_dev_enable(dev, old_dev); 8022 8023 __netdev_upper_dev_unlink(new_dev, dev, &priv); 8024 } 8025 EXPORT_SYMBOL(netdev_adjacent_change_abort); 8026 8027 /** 8028 * netdev_bonding_info_change - Dispatch event about slave change 8029 * @dev: device 8030 * @bonding_info: info to dispatch 8031 * 8032 * Send NETDEV_BONDING_INFO to netdev notifiers with info. 8033 * The caller must hold the RTNL lock. 8034 */ 8035 void netdev_bonding_info_change(struct net_device *dev, 8036 struct netdev_bonding_info *bonding_info) 8037 { 8038 struct netdev_notifier_bonding_info info = { 8039 .info.dev = dev, 8040 }; 8041 8042 memcpy(&info.bonding_info, bonding_info, 8043 sizeof(struct netdev_bonding_info)); 8044 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, 8045 &info.info); 8046 } 8047 EXPORT_SYMBOL(netdev_bonding_info_change); 8048 8049 static int netdev_offload_xstats_enable_l3(struct net_device *dev, 8050 struct netlink_ext_ack *extack) 8051 { 8052 struct netdev_notifier_offload_xstats_info info = { 8053 .info.dev = dev, 8054 .info.extack = extack, 8055 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, 8056 }; 8057 int err; 8058 int rc; 8059 8060 dev->offload_xstats_l3 = kzalloc(sizeof(*dev->offload_xstats_l3), 8061 GFP_KERNEL); 8062 if (!dev->offload_xstats_l3) 8063 return -ENOMEM; 8064 8065 rc = call_netdevice_notifiers_info_robust(NETDEV_OFFLOAD_XSTATS_ENABLE, 8066 NETDEV_OFFLOAD_XSTATS_DISABLE, 8067 &info.info); 8068 err = notifier_to_errno(rc); 8069 if (err) 8070 goto free_stats; 8071 8072 return 0; 8073 8074 free_stats: 8075 kfree(dev->offload_xstats_l3); 8076 dev->offload_xstats_l3 = NULL; 8077 return err; 8078 } 8079 8080 int netdev_offload_xstats_enable(struct net_device *dev, 8081 enum netdev_offload_xstats_type type, 8082 struct netlink_ext_ack *extack) 8083 { 8084 ASSERT_RTNL(); 8085 8086 if (netdev_offload_xstats_enabled(dev, type)) 8087 return -EALREADY; 8088 8089 switch (type) { 8090 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 8091 return netdev_offload_xstats_enable_l3(dev, extack); 8092 } 8093 8094 WARN_ON(1); 8095 return -EINVAL; 8096 } 8097 EXPORT_SYMBOL(netdev_offload_xstats_enable); 8098 8099 static void netdev_offload_xstats_disable_l3(struct net_device *dev) 8100 { 8101 struct netdev_notifier_offload_xstats_info info = { 8102 .info.dev = dev, 8103 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3, 8104 }; 8105 8106 call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_DISABLE, 8107 &info.info); 8108 kfree(dev->offload_xstats_l3); 8109 dev->offload_xstats_l3 = NULL; 8110 } 8111 8112 int netdev_offload_xstats_disable(struct net_device *dev, 8113 enum netdev_offload_xstats_type type) 8114 { 8115 ASSERT_RTNL(); 8116 8117 if (!netdev_offload_xstats_enabled(dev, type)) 8118 return -EALREADY; 8119 8120 switch (type) { 8121 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 8122 netdev_offload_xstats_disable_l3(dev); 8123 return 0; 8124 } 8125 8126 WARN_ON(1); 8127 return -EINVAL; 8128 } 8129 EXPORT_SYMBOL(netdev_offload_xstats_disable); 8130 8131 static void netdev_offload_xstats_disable_all(struct net_device *dev) 8132 { 8133 netdev_offload_xstats_disable(dev, NETDEV_OFFLOAD_XSTATS_TYPE_L3); 8134 } 8135 8136 static struct rtnl_hw_stats64 * 8137 netdev_offload_xstats_get_ptr(const struct net_device *dev, 8138 enum netdev_offload_xstats_type type) 8139 { 8140 switch (type) { 8141 case NETDEV_OFFLOAD_XSTATS_TYPE_L3: 8142 return dev->offload_xstats_l3; 8143 } 8144 8145 WARN_ON(1); 8146 return NULL; 8147 } 8148 8149 bool netdev_offload_xstats_enabled(const struct net_device *dev, 8150 enum netdev_offload_xstats_type type) 8151 { 8152 ASSERT_RTNL(); 8153 8154 return netdev_offload_xstats_get_ptr(dev, type); 8155 } 8156 EXPORT_SYMBOL(netdev_offload_xstats_enabled); 8157 8158 struct netdev_notifier_offload_xstats_ru { 8159 bool used; 8160 }; 8161 8162 struct netdev_notifier_offload_xstats_rd { 8163 struct rtnl_hw_stats64 stats; 8164 bool used; 8165 }; 8166 8167 static void netdev_hw_stats64_add(struct rtnl_hw_stats64 *dest, 8168 const struct rtnl_hw_stats64 *src) 8169 { 8170 dest->rx_packets += src->rx_packets; 8171 dest->tx_packets += src->tx_packets; 8172 dest->rx_bytes += src->rx_bytes; 8173 dest->tx_bytes += src->tx_bytes; 8174 dest->rx_errors += src->rx_errors; 8175 dest->tx_errors += src->tx_errors; 8176 dest->rx_dropped += src->rx_dropped; 8177 dest->tx_dropped += src->tx_dropped; 8178 dest->multicast += src->multicast; 8179 } 8180 8181 static int netdev_offload_xstats_get_used(struct net_device *dev, 8182 enum netdev_offload_xstats_type type, 8183 bool *p_used, 8184 struct netlink_ext_ack *extack) 8185 { 8186 struct netdev_notifier_offload_xstats_ru report_used = {}; 8187 struct netdev_notifier_offload_xstats_info info = { 8188 .info.dev = dev, 8189 .info.extack = extack, 8190 .type = type, 8191 .report_used = &report_used, 8192 }; 8193 int rc; 8194 8195 WARN_ON(!netdev_offload_xstats_enabled(dev, type)); 8196 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_USED, 8197 &info.info); 8198 *p_used = report_used.used; 8199 return notifier_to_errno(rc); 8200 } 8201 8202 static int netdev_offload_xstats_get_stats(struct net_device *dev, 8203 enum netdev_offload_xstats_type type, 8204 struct rtnl_hw_stats64 *p_stats, 8205 bool *p_used, 8206 struct netlink_ext_ack *extack) 8207 { 8208 struct netdev_notifier_offload_xstats_rd report_delta = {}; 8209 struct netdev_notifier_offload_xstats_info info = { 8210 .info.dev = dev, 8211 .info.extack = extack, 8212 .type = type, 8213 .report_delta = &report_delta, 8214 }; 8215 struct rtnl_hw_stats64 *stats; 8216 int rc; 8217 8218 stats = netdev_offload_xstats_get_ptr(dev, type); 8219 if (WARN_ON(!stats)) 8220 return -EINVAL; 8221 8222 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_DELTA, 8223 &info.info); 8224 8225 /* Cache whatever we got, even if there was an error, otherwise the 8226 * successful stats retrievals would get lost. 8227 */ 8228 netdev_hw_stats64_add(stats, &report_delta.stats); 8229 8230 if (p_stats) 8231 *p_stats = *stats; 8232 *p_used = report_delta.used; 8233 8234 return notifier_to_errno(rc); 8235 } 8236 8237 int netdev_offload_xstats_get(struct net_device *dev, 8238 enum netdev_offload_xstats_type type, 8239 struct rtnl_hw_stats64 *p_stats, bool *p_used, 8240 struct netlink_ext_ack *extack) 8241 { 8242 ASSERT_RTNL(); 8243 8244 if (p_stats) 8245 return netdev_offload_xstats_get_stats(dev, type, p_stats, 8246 p_used, extack); 8247 else 8248 return netdev_offload_xstats_get_used(dev, type, p_used, 8249 extack); 8250 } 8251 EXPORT_SYMBOL(netdev_offload_xstats_get); 8252 8253 void 8254 netdev_offload_xstats_report_delta(struct netdev_notifier_offload_xstats_rd *report_delta, 8255 const struct rtnl_hw_stats64 *stats) 8256 { 8257 report_delta->used = true; 8258 netdev_hw_stats64_add(&report_delta->stats, stats); 8259 } 8260 EXPORT_SYMBOL(netdev_offload_xstats_report_delta); 8261 8262 void 8263 netdev_offload_xstats_report_used(struct netdev_notifier_offload_xstats_ru *report_used) 8264 { 8265 report_used->used = true; 8266 } 8267 EXPORT_SYMBOL(netdev_offload_xstats_report_used); 8268 8269 void netdev_offload_xstats_push_delta(struct net_device *dev, 8270 enum netdev_offload_xstats_type type, 8271 const struct rtnl_hw_stats64 *p_stats) 8272 { 8273 struct rtnl_hw_stats64 *stats; 8274 8275 ASSERT_RTNL(); 8276 8277 stats = netdev_offload_xstats_get_ptr(dev, type); 8278 if (WARN_ON(!stats)) 8279 return; 8280 8281 netdev_hw_stats64_add(stats, p_stats); 8282 } 8283 EXPORT_SYMBOL(netdev_offload_xstats_push_delta); 8284 8285 /** 8286 * netdev_get_xmit_slave - Get the xmit slave of master device 8287 * @dev: device 8288 * @skb: The packet 8289 * @all_slaves: assume all the slaves are active 8290 * 8291 * The reference counters are not incremented so the caller must be 8292 * careful with locks. The caller must hold RCU lock. 8293 * %NULL is returned if no slave is found. 8294 */ 8295 8296 struct net_device *netdev_get_xmit_slave(struct net_device *dev, 8297 struct sk_buff *skb, 8298 bool all_slaves) 8299 { 8300 const struct net_device_ops *ops = dev->netdev_ops; 8301 8302 if (!ops->ndo_get_xmit_slave) 8303 return NULL; 8304 return ops->ndo_get_xmit_slave(dev, skb, all_slaves); 8305 } 8306 EXPORT_SYMBOL(netdev_get_xmit_slave); 8307 8308 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev, 8309 struct sock *sk) 8310 { 8311 const struct net_device_ops *ops = dev->netdev_ops; 8312 8313 if (!ops->ndo_sk_get_lower_dev) 8314 return NULL; 8315 return ops->ndo_sk_get_lower_dev(dev, sk); 8316 } 8317 8318 /** 8319 * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket 8320 * @dev: device 8321 * @sk: the socket 8322 * 8323 * %NULL is returned if no lower device is found. 8324 */ 8325 8326 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev, 8327 struct sock *sk) 8328 { 8329 struct net_device *lower; 8330 8331 lower = netdev_sk_get_lower_dev(dev, sk); 8332 while (lower) { 8333 dev = lower; 8334 lower = netdev_sk_get_lower_dev(dev, sk); 8335 } 8336 8337 return dev; 8338 } 8339 EXPORT_SYMBOL(netdev_sk_get_lowest_dev); 8340 8341 static void netdev_adjacent_add_links(struct net_device *dev) 8342 { 8343 struct netdev_adjacent *iter; 8344 8345 struct net *net = dev_net(dev); 8346 8347 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8348 if (!net_eq(net, dev_net(iter->dev))) 8349 continue; 8350 netdev_adjacent_sysfs_add(iter->dev, dev, 8351 &iter->dev->adj_list.lower); 8352 netdev_adjacent_sysfs_add(dev, iter->dev, 8353 &dev->adj_list.upper); 8354 } 8355 8356 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8357 if (!net_eq(net, dev_net(iter->dev))) 8358 continue; 8359 netdev_adjacent_sysfs_add(iter->dev, dev, 8360 &iter->dev->adj_list.upper); 8361 netdev_adjacent_sysfs_add(dev, iter->dev, 8362 &dev->adj_list.lower); 8363 } 8364 } 8365 8366 static void netdev_adjacent_del_links(struct net_device *dev) 8367 { 8368 struct netdev_adjacent *iter; 8369 8370 struct net *net = dev_net(dev); 8371 8372 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8373 if (!net_eq(net, dev_net(iter->dev))) 8374 continue; 8375 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8376 &iter->dev->adj_list.lower); 8377 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8378 &dev->adj_list.upper); 8379 } 8380 8381 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8382 if (!net_eq(net, dev_net(iter->dev))) 8383 continue; 8384 netdev_adjacent_sysfs_del(iter->dev, dev->name, 8385 &iter->dev->adj_list.upper); 8386 netdev_adjacent_sysfs_del(dev, iter->dev->name, 8387 &dev->adj_list.lower); 8388 } 8389 } 8390 8391 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) 8392 { 8393 struct netdev_adjacent *iter; 8394 8395 struct net *net = dev_net(dev); 8396 8397 list_for_each_entry(iter, &dev->adj_list.upper, list) { 8398 if (!net_eq(net, dev_net(iter->dev))) 8399 continue; 8400 netdev_adjacent_sysfs_del(iter->dev, oldname, 8401 &iter->dev->adj_list.lower); 8402 netdev_adjacent_sysfs_add(iter->dev, dev, 8403 &iter->dev->adj_list.lower); 8404 } 8405 8406 list_for_each_entry(iter, &dev->adj_list.lower, list) { 8407 if (!net_eq(net, dev_net(iter->dev))) 8408 continue; 8409 netdev_adjacent_sysfs_del(iter->dev, oldname, 8410 &iter->dev->adj_list.upper); 8411 netdev_adjacent_sysfs_add(iter->dev, dev, 8412 &iter->dev->adj_list.upper); 8413 } 8414 } 8415 8416 void *netdev_lower_dev_get_private(struct net_device *dev, 8417 struct net_device *lower_dev) 8418 { 8419 struct netdev_adjacent *lower; 8420 8421 if (!lower_dev) 8422 return NULL; 8423 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); 8424 if (!lower) 8425 return NULL; 8426 8427 return lower->private; 8428 } 8429 EXPORT_SYMBOL(netdev_lower_dev_get_private); 8430 8431 8432 /** 8433 * netdev_lower_state_changed - Dispatch event about lower device state change 8434 * @lower_dev: device 8435 * @lower_state_info: state to dispatch 8436 * 8437 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. 8438 * The caller must hold the RTNL lock. 8439 */ 8440 void netdev_lower_state_changed(struct net_device *lower_dev, 8441 void *lower_state_info) 8442 { 8443 struct netdev_notifier_changelowerstate_info changelowerstate_info = { 8444 .info.dev = lower_dev, 8445 }; 8446 8447 ASSERT_RTNL(); 8448 changelowerstate_info.lower_state_info = lower_state_info; 8449 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, 8450 &changelowerstate_info.info); 8451 } 8452 EXPORT_SYMBOL(netdev_lower_state_changed); 8453 8454 static void dev_change_rx_flags(struct net_device *dev, int flags) 8455 { 8456 const struct net_device_ops *ops = dev->netdev_ops; 8457 8458 if (ops->ndo_change_rx_flags) 8459 ops->ndo_change_rx_flags(dev, flags); 8460 } 8461 8462 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) 8463 { 8464 unsigned int old_flags = dev->flags; 8465 kuid_t uid; 8466 kgid_t gid; 8467 8468 ASSERT_RTNL(); 8469 8470 dev->flags |= IFF_PROMISC; 8471 dev->promiscuity += inc; 8472 if (dev->promiscuity == 0) { 8473 /* 8474 * Avoid overflow. 8475 * If inc causes overflow, untouch promisc and return error. 8476 */ 8477 if (inc < 0) 8478 dev->flags &= ~IFF_PROMISC; 8479 else { 8480 dev->promiscuity -= inc; 8481 netdev_warn(dev, "promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n"); 8482 return -EOVERFLOW; 8483 } 8484 } 8485 if (dev->flags != old_flags) { 8486 netdev_info(dev, "%s promiscuous mode\n", 8487 dev->flags & IFF_PROMISC ? "entered" : "left"); 8488 if (audit_enabled) { 8489 current_uid_gid(&uid, &gid); 8490 audit_log(audit_context(), GFP_ATOMIC, 8491 AUDIT_ANOM_PROMISCUOUS, 8492 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", 8493 dev->name, (dev->flags & IFF_PROMISC), 8494 (old_flags & IFF_PROMISC), 8495 from_kuid(&init_user_ns, audit_get_loginuid(current)), 8496 from_kuid(&init_user_ns, uid), 8497 from_kgid(&init_user_ns, gid), 8498 audit_get_sessionid(current)); 8499 } 8500 8501 dev_change_rx_flags(dev, IFF_PROMISC); 8502 } 8503 if (notify) 8504 __dev_notify_flags(dev, old_flags, IFF_PROMISC, 0, NULL); 8505 return 0; 8506 } 8507 8508 /** 8509 * dev_set_promiscuity - update promiscuity count on a device 8510 * @dev: device 8511 * @inc: modifier 8512 * 8513 * Add or remove promiscuity from a device. While the count in the device 8514 * remains above zero the interface remains promiscuous. Once it hits zero 8515 * the device reverts back to normal filtering operation. A negative inc 8516 * value is used to drop promiscuity on the device. 8517 * Return 0 if successful or a negative errno code on error. 8518 */ 8519 int dev_set_promiscuity(struct net_device *dev, int inc) 8520 { 8521 unsigned int old_flags = dev->flags; 8522 int err; 8523 8524 err = __dev_set_promiscuity(dev, inc, true); 8525 if (err < 0) 8526 return err; 8527 if (dev->flags != old_flags) 8528 dev_set_rx_mode(dev); 8529 return err; 8530 } 8531 EXPORT_SYMBOL(dev_set_promiscuity); 8532 8533 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) 8534 { 8535 unsigned int old_flags = dev->flags, old_gflags = dev->gflags; 8536 8537 ASSERT_RTNL(); 8538 8539 dev->flags |= IFF_ALLMULTI; 8540 dev->allmulti += inc; 8541 if (dev->allmulti == 0) { 8542 /* 8543 * Avoid overflow. 8544 * If inc causes overflow, untouch allmulti and return error. 8545 */ 8546 if (inc < 0) 8547 dev->flags &= ~IFF_ALLMULTI; 8548 else { 8549 dev->allmulti -= inc; 8550 netdev_warn(dev, "allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n"); 8551 return -EOVERFLOW; 8552 } 8553 } 8554 if (dev->flags ^ old_flags) { 8555 netdev_info(dev, "%s allmulticast mode\n", 8556 dev->flags & IFF_ALLMULTI ? "entered" : "left"); 8557 dev_change_rx_flags(dev, IFF_ALLMULTI); 8558 dev_set_rx_mode(dev); 8559 if (notify) 8560 __dev_notify_flags(dev, old_flags, 8561 dev->gflags ^ old_gflags, 0, NULL); 8562 } 8563 return 0; 8564 } 8565 8566 /** 8567 * dev_set_allmulti - update allmulti count on a device 8568 * @dev: device 8569 * @inc: modifier 8570 * 8571 * Add or remove reception of all multicast frames to a device. While the 8572 * count in the device remains above zero the interface remains listening 8573 * to all interfaces. Once it hits zero the device reverts back to normal 8574 * filtering operation. A negative @inc value is used to drop the counter 8575 * when releasing a resource needing all multicasts. 8576 * Return 0 if successful or a negative errno code on error. 8577 */ 8578 8579 int dev_set_allmulti(struct net_device *dev, int inc) 8580 { 8581 return __dev_set_allmulti(dev, inc, true); 8582 } 8583 EXPORT_SYMBOL(dev_set_allmulti); 8584 8585 /* 8586 * Upload unicast and multicast address lists to device and 8587 * configure RX filtering. When the device doesn't support unicast 8588 * filtering it is put in promiscuous mode while unicast addresses 8589 * are present. 8590 */ 8591 void __dev_set_rx_mode(struct net_device *dev) 8592 { 8593 const struct net_device_ops *ops = dev->netdev_ops; 8594 8595 /* dev_open will call this function so the list will stay sane. */ 8596 if (!(dev->flags&IFF_UP)) 8597 return; 8598 8599 if (!netif_device_present(dev)) 8600 return; 8601 8602 if (!(dev->priv_flags & IFF_UNICAST_FLT)) { 8603 /* Unicast addresses changes may only happen under the rtnl, 8604 * therefore calling __dev_set_promiscuity here is safe. 8605 */ 8606 if (!netdev_uc_empty(dev) && !dev->uc_promisc) { 8607 __dev_set_promiscuity(dev, 1, false); 8608 dev->uc_promisc = true; 8609 } else if (netdev_uc_empty(dev) && dev->uc_promisc) { 8610 __dev_set_promiscuity(dev, -1, false); 8611 dev->uc_promisc = false; 8612 } 8613 } 8614 8615 if (ops->ndo_set_rx_mode) 8616 ops->ndo_set_rx_mode(dev); 8617 } 8618 8619 void dev_set_rx_mode(struct net_device *dev) 8620 { 8621 netif_addr_lock_bh(dev); 8622 __dev_set_rx_mode(dev); 8623 netif_addr_unlock_bh(dev); 8624 } 8625 8626 /** 8627 * dev_get_flags - get flags reported to userspace 8628 * @dev: device 8629 * 8630 * Get the combination of flag bits exported through APIs to userspace. 8631 */ 8632 unsigned int dev_get_flags(const struct net_device *dev) 8633 { 8634 unsigned int flags; 8635 8636 flags = (READ_ONCE(dev->flags) & ~(IFF_PROMISC | 8637 IFF_ALLMULTI | 8638 IFF_RUNNING | 8639 IFF_LOWER_UP | 8640 IFF_DORMANT)) | 8641 (READ_ONCE(dev->gflags) & (IFF_PROMISC | 8642 IFF_ALLMULTI)); 8643 8644 if (netif_running(dev)) { 8645 if (netif_oper_up(dev)) 8646 flags |= IFF_RUNNING; 8647 if (netif_carrier_ok(dev)) 8648 flags |= IFF_LOWER_UP; 8649 if (netif_dormant(dev)) 8650 flags |= IFF_DORMANT; 8651 } 8652 8653 return flags; 8654 } 8655 EXPORT_SYMBOL(dev_get_flags); 8656 8657 int __dev_change_flags(struct net_device *dev, unsigned int flags, 8658 struct netlink_ext_ack *extack) 8659 { 8660 unsigned int old_flags = dev->flags; 8661 int ret; 8662 8663 ASSERT_RTNL(); 8664 8665 /* 8666 * Set the flags on our device. 8667 */ 8668 8669 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | 8670 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | 8671 IFF_AUTOMEDIA)) | 8672 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | 8673 IFF_ALLMULTI)); 8674 8675 /* 8676 * Load in the correct multicast list now the flags have changed. 8677 */ 8678 8679 if ((old_flags ^ flags) & IFF_MULTICAST) 8680 dev_change_rx_flags(dev, IFF_MULTICAST); 8681 8682 dev_set_rx_mode(dev); 8683 8684 /* 8685 * Have we downed the interface. We handle IFF_UP ourselves 8686 * according to user attempts to set it, rather than blindly 8687 * setting it. 8688 */ 8689 8690 ret = 0; 8691 if ((old_flags ^ flags) & IFF_UP) { 8692 if (old_flags & IFF_UP) 8693 __dev_close(dev); 8694 else 8695 ret = __dev_open(dev, extack); 8696 } 8697 8698 if ((flags ^ dev->gflags) & IFF_PROMISC) { 8699 int inc = (flags & IFF_PROMISC) ? 1 : -1; 8700 unsigned int old_flags = dev->flags; 8701 8702 dev->gflags ^= IFF_PROMISC; 8703 8704 if (__dev_set_promiscuity(dev, inc, false) >= 0) 8705 if (dev->flags != old_flags) 8706 dev_set_rx_mode(dev); 8707 } 8708 8709 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI 8710 * is important. Some (broken) drivers set IFF_PROMISC, when 8711 * IFF_ALLMULTI is requested not asking us and not reporting. 8712 */ 8713 if ((flags ^ dev->gflags) & IFF_ALLMULTI) { 8714 int inc = (flags & IFF_ALLMULTI) ? 1 : -1; 8715 8716 dev->gflags ^= IFF_ALLMULTI; 8717 __dev_set_allmulti(dev, inc, false); 8718 } 8719 8720 return ret; 8721 } 8722 8723 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, 8724 unsigned int gchanges, u32 portid, 8725 const struct nlmsghdr *nlh) 8726 { 8727 unsigned int changes = dev->flags ^ old_flags; 8728 8729 if (gchanges) 8730 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC, portid, nlh); 8731 8732 if (changes & IFF_UP) { 8733 if (dev->flags & IFF_UP) 8734 call_netdevice_notifiers(NETDEV_UP, dev); 8735 else 8736 call_netdevice_notifiers(NETDEV_DOWN, dev); 8737 } 8738 8739 if (dev->flags & IFF_UP && 8740 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { 8741 struct netdev_notifier_change_info change_info = { 8742 .info = { 8743 .dev = dev, 8744 }, 8745 .flags_changed = changes, 8746 }; 8747 8748 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info); 8749 } 8750 } 8751 8752 /** 8753 * dev_change_flags - change device settings 8754 * @dev: device 8755 * @flags: device state flags 8756 * @extack: netlink extended ack 8757 * 8758 * Change settings on device based state flags. The flags are 8759 * in the userspace exported format. 8760 */ 8761 int dev_change_flags(struct net_device *dev, unsigned int flags, 8762 struct netlink_ext_ack *extack) 8763 { 8764 int ret; 8765 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; 8766 8767 ret = __dev_change_flags(dev, flags, extack); 8768 if (ret < 0) 8769 return ret; 8770 8771 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); 8772 __dev_notify_flags(dev, old_flags, changes, 0, NULL); 8773 return ret; 8774 } 8775 EXPORT_SYMBOL(dev_change_flags); 8776 8777 int __dev_set_mtu(struct net_device *dev, int new_mtu) 8778 { 8779 const struct net_device_ops *ops = dev->netdev_ops; 8780 8781 if (ops->ndo_change_mtu) 8782 return ops->ndo_change_mtu(dev, new_mtu); 8783 8784 /* Pairs with all the lockless reads of dev->mtu in the stack */ 8785 WRITE_ONCE(dev->mtu, new_mtu); 8786 return 0; 8787 } 8788 EXPORT_SYMBOL(__dev_set_mtu); 8789 8790 int dev_validate_mtu(struct net_device *dev, int new_mtu, 8791 struct netlink_ext_ack *extack) 8792 { 8793 /* MTU must be positive, and in range */ 8794 if (new_mtu < 0 || new_mtu < dev->min_mtu) { 8795 NL_SET_ERR_MSG(extack, "mtu less than device minimum"); 8796 return -EINVAL; 8797 } 8798 8799 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { 8800 NL_SET_ERR_MSG(extack, "mtu greater than device maximum"); 8801 return -EINVAL; 8802 } 8803 return 0; 8804 } 8805 8806 /** 8807 * dev_set_mtu_ext - Change maximum transfer unit 8808 * @dev: device 8809 * @new_mtu: new transfer unit 8810 * @extack: netlink extended ack 8811 * 8812 * Change the maximum transfer size of the network device. 8813 */ 8814 int dev_set_mtu_ext(struct net_device *dev, int new_mtu, 8815 struct netlink_ext_ack *extack) 8816 { 8817 int err, orig_mtu; 8818 8819 if (new_mtu == dev->mtu) 8820 return 0; 8821 8822 err = dev_validate_mtu(dev, new_mtu, extack); 8823 if (err) 8824 return err; 8825 8826 if (!netif_device_present(dev)) 8827 return -ENODEV; 8828 8829 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); 8830 err = notifier_to_errno(err); 8831 if (err) 8832 return err; 8833 8834 orig_mtu = dev->mtu; 8835 err = __dev_set_mtu(dev, new_mtu); 8836 8837 if (!err) { 8838 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8839 orig_mtu); 8840 err = notifier_to_errno(err); 8841 if (err) { 8842 /* setting mtu back and notifying everyone again, 8843 * so that they have a chance to revert changes. 8844 */ 8845 __dev_set_mtu(dev, orig_mtu); 8846 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, 8847 new_mtu); 8848 } 8849 } 8850 return err; 8851 } 8852 8853 int dev_set_mtu(struct net_device *dev, int new_mtu) 8854 { 8855 struct netlink_ext_ack extack; 8856 int err; 8857 8858 memset(&extack, 0, sizeof(extack)); 8859 err = dev_set_mtu_ext(dev, new_mtu, &extack); 8860 if (err && extack._msg) 8861 net_err_ratelimited("%s: %s\n", dev->name, extack._msg); 8862 return err; 8863 } 8864 EXPORT_SYMBOL(dev_set_mtu); 8865 8866 /** 8867 * dev_change_tx_queue_len - Change TX queue length of a netdevice 8868 * @dev: device 8869 * @new_len: new tx queue length 8870 */ 8871 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len) 8872 { 8873 unsigned int orig_len = dev->tx_queue_len; 8874 int res; 8875 8876 if (new_len != (unsigned int)new_len) 8877 return -ERANGE; 8878 8879 if (new_len != orig_len) { 8880 dev->tx_queue_len = new_len; 8881 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev); 8882 res = notifier_to_errno(res); 8883 if (res) 8884 goto err_rollback; 8885 res = dev_qdisc_change_tx_queue_len(dev); 8886 if (res) 8887 goto err_rollback; 8888 } 8889 8890 return 0; 8891 8892 err_rollback: 8893 netdev_err(dev, "refused to change device tx_queue_len\n"); 8894 dev->tx_queue_len = orig_len; 8895 return res; 8896 } 8897 8898 /** 8899 * dev_set_group - Change group this device belongs to 8900 * @dev: device 8901 * @new_group: group this device should belong to 8902 */ 8903 void dev_set_group(struct net_device *dev, int new_group) 8904 { 8905 dev->group = new_group; 8906 } 8907 8908 /** 8909 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR. 8910 * @dev: device 8911 * @addr: new address 8912 * @extack: netlink extended ack 8913 */ 8914 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr, 8915 struct netlink_ext_ack *extack) 8916 { 8917 struct netdev_notifier_pre_changeaddr_info info = { 8918 .info.dev = dev, 8919 .info.extack = extack, 8920 .dev_addr = addr, 8921 }; 8922 int rc; 8923 8924 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info); 8925 return notifier_to_errno(rc); 8926 } 8927 EXPORT_SYMBOL(dev_pre_changeaddr_notify); 8928 8929 /** 8930 * dev_set_mac_address - Change Media Access Control Address 8931 * @dev: device 8932 * @sa: new address 8933 * @extack: netlink extended ack 8934 * 8935 * Change the hardware (MAC) address of the device 8936 */ 8937 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa, 8938 struct netlink_ext_ack *extack) 8939 { 8940 const struct net_device_ops *ops = dev->netdev_ops; 8941 int err; 8942 8943 if (!ops->ndo_set_mac_address) 8944 return -EOPNOTSUPP; 8945 if (sa->sa_family != dev->type) 8946 return -EINVAL; 8947 if (!netif_device_present(dev)) 8948 return -ENODEV; 8949 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack); 8950 if (err) 8951 return err; 8952 if (memcmp(dev->dev_addr, sa->sa_data, dev->addr_len)) { 8953 err = ops->ndo_set_mac_address(dev, sa); 8954 if (err) 8955 return err; 8956 } 8957 dev->addr_assign_type = NET_ADDR_SET; 8958 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); 8959 add_device_randomness(dev->dev_addr, dev->addr_len); 8960 return 0; 8961 } 8962 EXPORT_SYMBOL(dev_set_mac_address); 8963 8964 DECLARE_RWSEM(dev_addr_sem); 8965 8966 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa, 8967 struct netlink_ext_ack *extack) 8968 { 8969 int ret; 8970 8971 down_write(&dev_addr_sem); 8972 ret = dev_set_mac_address(dev, sa, extack); 8973 up_write(&dev_addr_sem); 8974 return ret; 8975 } 8976 EXPORT_SYMBOL(dev_set_mac_address_user); 8977 8978 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name) 8979 { 8980 size_t size = sizeof(sa->sa_data_min); 8981 struct net_device *dev; 8982 int ret = 0; 8983 8984 down_read(&dev_addr_sem); 8985 rcu_read_lock(); 8986 8987 dev = dev_get_by_name_rcu(net, dev_name); 8988 if (!dev) { 8989 ret = -ENODEV; 8990 goto unlock; 8991 } 8992 if (!dev->addr_len) 8993 memset(sa->sa_data, 0, size); 8994 else 8995 memcpy(sa->sa_data, dev->dev_addr, 8996 min_t(size_t, size, dev->addr_len)); 8997 sa->sa_family = dev->type; 8998 8999 unlock: 9000 rcu_read_unlock(); 9001 up_read(&dev_addr_sem); 9002 return ret; 9003 } 9004 EXPORT_SYMBOL(dev_get_mac_address); 9005 9006 /** 9007 * dev_change_carrier - Change device carrier 9008 * @dev: device 9009 * @new_carrier: new value 9010 * 9011 * Change device carrier 9012 */ 9013 int dev_change_carrier(struct net_device *dev, bool new_carrier) 9014 { 9015 const struct net_device_ops *ops = dev->netdev_ops; 9016 9017 if (!ops->ndo_change_carrier) 9018 return -EOPNOTSUPP; 9019 if (!netif_device_present(dev)) 9020 return -ENODEV; 9021 return ops->ndo_change_carrier(dev, new_carrier); 9022 } 9023 9024 /** 9025 * dev_get_phys_port_id - Get device physical port ID 9026 * @dev: device 9027 * @ppid: port ID 9028 * 9029 * Get device physical port ID 9030 */ 9031 int dev_get_phys_port_id(struct net_device *dev, 9032 struct netdev_phys_item_id *ppid) 9033 { 9034 const struct net_device_ops *ops = dev->netdev_ops; 9035 9036 if (!ops->ndo_get_phys_port_id) 9037 return -EOPNOTSUPP; 9038 return ops->ndo_get_phys_port_id(dev, ppid); 9039 } 9040 9041 /** 9042 * dev_get_phys_port_name - Get device physical port name 9043 * @dev: device 9044 * @name: port name 9045 * @len: limit of bytes to copy to name 9046 * 9047 * Get device physical port name 9048 */ 9049 int dev_get_phys_port_name(struct net_device *dev, 9050 char *name, size_t len) 9051 { 9052 const struct net_device_ops *ops = dev->netdev_ops; 9053 int err; 9054 9055 if (ops->ndo_get_phys_port_name) { 9056 err = ops->ndo_get_phys_port_name(dev, name, len); 9057 if (err != -EOPNOTSUPP) 9058 return err; 9059 } 9060 return devlink_compat_phys_port_name_get(dev, name, len); 9061 } 9062 9063 /** 9064 * dev_get_port_parent_id - Get the device's port parent identifier 9065 * @dev: network device 9066 * @ppid: pointer to a storage for the port's parent identifier 9067 * @recurse: allow/disallow recursion to lower devices 9068 * 9069 * Get the devices's port parent identifier 9070 */ 9071 int dev_get_port_parent_id(struct net_device *dev, 9072 struct netdev_phys_item_id *ppid, 9073 bool recurse) 9074 { 9075 const struct net_device_ops *ops = dev->netdev_ops; 9076 struct netdev_phys_item_id first = { }; 9077 struct net_device *lower_dev; 9078 struct list_head *iter; 9079 int err; 9080 9081 if (ops->ndo_get_port_parent_id) { 9082 err = ops->ndo_get_port_parent_id(dev, ppid); 9083 if (err != -EOPNOTSUPP) 9084 return err; 9085 } 9086 9087 err = devlink_compat_switch_id_get(dev, ppid); 9088 if (!recurse || err != -EOPNOTSUPP) 9089 return err; 9090 9091 netdev_for_each_lower_dev(dev, lower_dev, iter) { 9092 err = dev_get_port_parent_id(lower_dev, ppid, true); 9093 if (err) 9094 break; 9095 if (!first.id_len) 9096 first = *ppid; 9097 else if (memcmp(&first, ppid, sizeof(*ppid))) 9098 return -EOPNOTSUPP; 9099 } 9100 9101 return err; 9102 } 9103 EXPORT_SYMBOL(dev_get_port_parent_id); 9104 9105 /** 9106 * netdev_port_same_parent_id - Indicate if two network devices have 9107 * the same port parent identifier 9108 * @a: first network device 9109 * @b: second network device 9110 */ 9111 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b) 9112 { 9113 struct netdev_phys_item_id a_id = { }; 9114 struct netdev_phys_item_id b_id = { }; 9115 9116 if (dev_get_port_parent_id(a, &a_id, true) || 9117 dev_get_port_parent_id(b, &b_id, true)) 9118 return false; 9119 9120 return netdev_phys_item_id_same(&a_id, &b_id); 9121 } 9122 EXPORT_SYMBOL(netdev_port_same_parent_id); 9123 9124 static void netdev_dpll_pin_assign(struct net_device *dev, struct dpll_pin *dpll_pin) 9125 { 9126 #if IS_ENABLED(CONFIG_DPLL) 9127 rtnl_lock(); 9128 rcu_assign_pointer(dev->dpll_pin, dpll_pin); 9129 rtnl_unlock(); 9130 #endif 9131 } 9132 9133 void netdev_dpll_pin_set(struct net_device *dev, struct dpll_pin *dpll_pin) 9134 { 9135 WARN_ON(!dpll_pin); 9136 netdev_dpll_pin_assign(dev, dpll_pin); 9137 } 9138 EXPORT_SYMBOL(netdev_dpll_pin_set); 9139 9140 void netdev_dpll_pin_clear(struct net_device *dev) 9141 { 9142 netdev_dpll_pin_assign(dev, NULL); 9143 } 9144 EXPORT_SYMBOL(netdev_dpll_pin_clear); 9145 9146 /** 9147 * dev_change_proto_down - set carrier according to proto_down. 9148 * 9149 * @dev: device 9150 * @proto_down: new value 9151 */ 9152 int dev_change_proto_down(struct net_device *dev, bool proto_down) 9153 { 9154 if (!(dev->priv_flags & IFF_CHANGE_PROTO_DOWN)) 9155 return -EOPNOTSUPP; 9156 if (!netif_device_present(dev)) 9157 return -ENODEV; 9158 if (proto_down) 9159 netif_carrier_off(dev); 9160 else 9161 netif_carrier_on(dev); 9162 dev->proto_down = proto_down; 9163 return 0; 9164 } 9165 9166 /** 9167 * dev_change_proto_down_reason - proto down reason 9168 * 9169 * @dev: device 9170 * @mask: proto down mask 9171 * @value: proto down value 9172 */ 9173 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask, 9174 u32 value) 9175 { 9176 int b; 9177 9178 if (!mask) { 9179 dev->proto_down_reason = value; 9180 } else { 9181 for_each_set_bit(b, &mask, 32) { 9182 if (value & (1 << b)) 9183 dev->proto_down_reason |= BIT(b); 9184 else 9185 dev->proto_down_reason &= ~BIT(b); 9186 } 9187 } 9188 } 9189 9190 struct bpf_xdp_link { 9191 struct bpf_link link; 9192 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */ 9193 int flags; 9194 }; 9195 9196 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags) 9197 { 9198 if (flags & XDP_FLAGS_HW_MODE) 9199 return XDP_MODE_HW; 9200 if (flags & XDP_FLAGS_DRV_MODE) 9201 return XDP_MODE_DRV; 9202 if (flags & XDP_FLAGS_SKB_MODE) 9203 return XDP_MODE_SKB; 9204 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB; 9205 } 9206 9207 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode) 9208 { 9209 switch (mode) { 9210 case XDP_MODE_SKB: 9211 return generic_xdp_install; 9212 case XDP_MODE_DRV: 9213 case XDP_MODE_HW: 9214 return dev->netdev_ops->ndo_bpf; 9215 default: 9216 return NULL; 9217 } 9218 } 9219 9220 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev, 9221 enum bpf_xdp_mode mode) 9222 { 9223 return dev->xdp_state[mode].link; 9224 } 9225 9226 static struct bpf_prog *dev_xdp_prog(struct net_device *dev, 9227 enum bpf_xdp_mode mode) 9228 { 9229 struct bpf_xdp_link *link = dev_xdp_link(dev, mode); 9230 9231 if (link) 9232 return link->link.prog; 9233 return dev->xdp_state[mode].prog; 9234 } 9235 9236 u8 dev_xdp_prog_count(struct net_device *dev) 9237 { 9238 u8 count = 0; 9239 int i; 9240 9241 for (i = 0; i < __MAX_XDP_MODE; i++) 9242 if (dev->xdp_state[i].prog || dev->xdp_state[i].link) 9243 count++; 9244 return count; 9245 } 9246 EXPORT_SYMBOL_GPL(dev_xdp_prog_count); 9247 9248 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode) 9249 { 9250 struct bpf_prog *prog = dev_xdp_prog(dev, mode); 9251 9252 return prog ? prog->aux->id : 0; 9253 } 9254 9255 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode, 9256 struct bpf_xdp_link *link) 9257 { 9258 dev->xdp_state[mode].link = link; 9259 dev->xdp_state[mode].prog = NULL; 9260 } 9261 9262 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode, 9263 struct bpf_prog *prog) 9264 { 9265 dev->xdp_state[mode].link = NULL; 9266 dev->xdp_state[mode].prog = prog; 9267 } 9268 9269 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode, 9270 bpf_op_t bpf_op, struct netlink_ext_ack *extack, 9271 u32 flags, struct bpf_prog *prog) 9272 { 9273 struct netdev_bpf xdp; 9274 int err; 9275 9276 memset(&xdp, 0, sizeof(xdp)); 9277 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG; 9278 xdp.extack = extack; 9279 xdp.flags = flags; 9280 xdp.prog = prog; 9281 9282 /* Drivers assume refcnt is already incremented (i.e, prog pointer is 9283 * "moved" into driver), so they don't increment it on their own, but 9284 * they do decrement refcnt when program is detached or replaced. 9285 * Given net_device also owns link/prog, we need to bump refcnt here 9286 * to prevent drivers from underflowing it. 9287 */ 9288 if (prog) 9289 bpf_prog_inc(prog); 9290 err = bpf_op(dev, &xdp); 9291 if (err) { 9292 if (prog) 9293 bpf_prog_put(prog); 9294 return err; 9295 } 9296 9297 if (mode != XDP_MODE_HW) 9298 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog); 9299 9300 return 0; 9301 } 9302 9303 static void dev_xdp_uninstall(struct net_device *dev) 9304 { 9305 struct bpf_xdp_link *link; 9306 struct bpf_prog *prog; 9307 enum bpf_xdp_mode mode; 9308 bpf_op_t bpf_op; 9309 9310 ASSERT_RTNL(); 9311 9312 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) { 9313 prog = dev_xdp_prog(dev, mode); 9314 if (!prog) 9315 continue; 9316 9317 bpf_op = dev_xdp_bpf_op(dev, mode); 9318 if (!bpf_op) 9319 continue; 9320 9321 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9322 9323 /* auto-detach link from net device */ 9324 link = dev_xdp_link(dev, mode); 9325 if (link) 9326 link->dev = NULL; 9327 else 9328 bpf_prog_put(prog); 9329 9330 dev_xdp_set_link(dev, mode, NULL); 9331 } 9332 } 9333 9334 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack, 9335 struct bpf_xdp_link *link, struct bpf_prog *new_prog, 9336 struct bpf_prog *old_prog, u32 flags) 9337 { 9338 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES); 9339 struct bpf_prog *cur_prog; 9340 struct net_device *upper; 9341 struct list_head *iter; 9342 enum bpf_xdp_mode mode; 9343 bpf_op_t bpf_op; 9344 int err; 9345 9346 ASSERT_RTNL(); 9347 9348 /* either link or prog attachment, never both */ 9349 if (link && (new_prog || old_prog)) 9350 return -EINVAL; 9351 /* link supports only XDP mode flags */ 9352 if (link && (flags & ~XDP_FLAGS_MODES)) { 9353 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment"); 9354 return -EINVAL; 9355 } 9356 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */ 9357 if (num_modes > 1) { 9358 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set"); 9359 return -EINVAL; 9360 } 9361 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */ 9362 if (!num_modes && dev_xdp_prog_count(dev) > 1) { 9363 NL_SET_ERR_MSG(extack, 9364 "More than one program loaded, unset mode is ambiguous"); 9365 return -EINVAL; 9366 } 9367 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */ 9368 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) { 9369 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified"); 9370 return -EINVAL; 9371 } 9372 9373 mode = dev_xdp_mode(dev, flags); 9374 /* can't replace attached link */ 9375 if (dev_xdp_link(dev, mode)) { 9376 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link"); 9377 return -EBUSY; 9378 } 9379 9380 /* don't allow if an upper device already has a program */ 9381 netdev_for_each_upper_dev_rcu(dev, upper, iter) { 9382 if (dev_xdp_prog_count(upper) > 0) { 9383 NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program"); 9384 return -EEXIST; 9385 } 9386 } 9387 9388 cur_prog = dev_xdp_prog(dev, mode); 9389 /* can't replace attached prog with link */ 9390 if (link && cur_prog) { 9391 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link"); 9392 return -EBUSY; 9393 } 9394 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) { 9395 NL_SET_ERR_MSG(extack, "Active program does not match expected"); 9396 return -EEXIST; 9397 } 9398 9399 /* put effective new program into new_prog */ 9400 if (link) 9401 new_prog = link->link.prog; 9402 9403 if (new_prog) { 9404 bool offload = mode == XDP_MODE_HW; 9405 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB 9406 ? XDP_MODE_DRV : XDP_MODE_SKB; 9407 9408 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) { 9409 NL_SET_ERR_MSG(extack, "XDP program already attached"); 9410 return -EBUSY; 9411 } 9412 if (!offload && dev_xdp_prog(dev, other_mode)) { 9413 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time"); 9414 return -EEXIST; 9415 } 9416 if (!offload && bpf_prog_is_offloaded(new_prog->aux)) { 9417 NL_SET_ERR_MSG(extack, "Using offloaded program without HW_MODE flag is not supported"); 9418 return -EINVAL; 9419 } 9420 if (bpf_prog_is_dev_bound(new_prog->aux) && !bpf_offload_dev_match(new_prog, dev)) { 9421 NL_SET_ERR_MSG(extack, "Program bound to different device"); 9422 return -EINVAL; 9423 } 9424 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) { 9425 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device"); 9426 return -EINVAL; 9427 } 9428 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) { 9429 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device"); 9430 return -EINVAL; 9431 } 9432 } 9433 9434 /* don't call drivers if the effective program didn't change */ 9435 if (new_prog != cur_prog) { 9436 bpf_op = dev_xdp_bpf_op(dev, mode); 9437 if (!bpf_op) { 9438 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode"); 9439 return -EOPNOTSUPP; 9440 } 9441 9442 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog); 9443 if (err) 9444 return err; 9445 } 9446 9447 if (link) 9448 dev_xdp_set_link(dev, mode, link); 9449 else 9450 dev_xdp_set_prog(dev, mode, new_prog); 9451 if (cur_prog) 9452 bpf_prog_put(cur_prog); 9453 9454 return 0; 9455 } 9456 9457 static int dev_xdp_attach_link(struct net_device *dev, 9458 struct netlink_ext_ack *extack, 9459 struct bpf_xdp_link *link) 9460 { 9461 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags); 9462 } 9463 9464 static int dev_xdp_detach_link(struct net_device *dev, 9465 struct netlink_ext_ack *extack, 9466 struct bpf_xdp_link *link) 9467 { 9468 enum bpf_xdp_mode mode; 9469 bpf_op_t bpf_op; 9470 9471 ASSERT_RTNL(); 9472 9473 mode = dev_xdp_mode(dev, link->flags); 9474 if (dev_xdp_link(dev, mode) != link) 9475 return -EINVAL; 9476 9477 bpf_op = dev_xdp_bpf_op(dev, mode); 9478 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL)); 9479 dev_xdp_set_link(dev, mode, NULL); 9480 return 0; 9481 } 9482 9483 static void bpf_xdp_link_release(struct bpf_link *link) 9484 { 9485 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9486 9487 rtnl_lock(); 9488 9489 /* if racing with net_device's tear down, xdp_link->dev might be 9490 * already NULL, in which case link was already auto-detached 9491 */ 9492 if (xdp_link->dev) { 9493 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link)); 9494 xdp_link->dev = NULL; 9495 } 9496 9497 rtnl_unlock(); 9498 } 9499 9500 static int bpf_xdp_link_detach(struct bpf_link *link) 9501 { 9502 bpf_xdp_link_release(link); 9503 return 0; 9504 } 9505 9506 static void bpf_xdp_link_dealloc(struct bpf_link *link) 9507 { 9508 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9509 9510 kfree(xdp_link); 9511 } 9512 9513 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link, 9514 struct seq_file *seq) 9515 { 9516 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9517 u32 ifindex = 0; 9518 9519 rtnl_lock(); 9520 if (xdp_link->dev) 9521 ifindex = xdp_link->dev->ifindex; 9522 rtnl_unlock(); 9523 9524 seq_printf(seq, "ifindex:\t%u\n", ifindex); 9525 } 9526 9527 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link, 9528 struct bpf_link_info *info) 9529 { 9530 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9531 u32 ifindex = 0; 9532 9533 rtnl_lock(); 9534 if (xdp_link->dev) 9535 ifindex = xdp_link->dev->ifindex; 9536 rtnl_unlock(); 9537 9538 info->xdp.ifindex = ifindex; 9539 return 0; 9540 } 9541 9542 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog, 9543 struct bpf_prog *old_prog) 9544 { 9545 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link); 9546 enum bpf_xdp_mode mode; 9547 bpf_op_t bpf_op; 9548 int err = 0; 9549 9550 rtnl_lock(); 9551 9552 /* link might have been auto-released already, so fail */ 9553 if (!xdp_link->dev) { 9554 err = -ENOLINK; 9555 goto out_unlock; 9556 } 9557 9558 if (old_prog && link->prog != old_prog) { 9559 err = -EPERM; 9560 goto out_unlock; 9561 } 9562 old_prog = link->prog; 9563 if (old_prog->type != new_prog->type || 9564 old_prog->expected_attach_type != new_prog->expected_attach_type) { 9565 err = -EINVAL; 9566 goto out_unlock; 9567 } 9568 9569 if (old_prog == new_prog) { 9570 /* no-op, don't disturb drivers */ 9571 bpf_prog_put(new_prog); 9572 goto out_unlock; 9573 } 9574 9575 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags); 9576 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode); 9577 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL, 9578 xdp_link->flags, new_prog); 9579 if (err) 9580 goto out_unlock; 9581 9582 old_prog = xchg(&link->prog, new_prog); 9583 bpf_prog_put(old_prog); 9584 9585 out_unlock: 9586 rtnl_unlock(); 9587 return err; 9588 } 9589 9590 static const struct bpf_link_ops bpf_xdp_link_lops = { 9591 .release = bpf_xdp_link_release, 9592 .dealloc = bpf_xdp_link_dealloc, 9593 .detach = bpf_xdp_link_detach, 9594 .show_fdinfo = bpf_xdp_link_show_fdinfo, 9595 .fill_link_info = bpf_xdp_link_fill_link_info, 9596 .update_prog = bpf_xdp_link_update, 9597 }; 9598 9599 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog) 9600 { 9601 struct net *net = current->nsproxy->net_ns; 9602 struct bpf_link_primer link_primer; 9603 struct netlink_ext_ack extack = {}; 9604 struct bpf_xdp_link *link; 9605 struct net_device *dev; 9606 int err, fd; 9607 9608 rtnl_lock(); 9609 dev = dev_get_by_index(net, attr->link_create.target_ifindex); 9610 if (!dev) { 9611 rtnl_unlock(); 9612 return -EINVAL; 9613 } 9614 9615 link = kzalloc(sizeof(*link), GFP_USER); 9616 if (!link) { 9617 err = -ENOMEM; 9618 goto unlock; 9619 } 9620 9621 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog); 9622 link->dev = dev; 9623 link->flags = attr->link_create.flags; 9624 9625 err = bpf_link_prime(&link->link, &link_primer); 9626 if (err) { 9627 kfree(link); 9628 goto unlock; 9629 } 9630 9631 err = dev_xdp_attach_link(dev, &extack, link); 9632 rtnl_unlock(); 9633 9634 if (err) { 9635 link->dev = NULL; 9636 bpf_link_cleanup(&link_primer); 9637 trace_bpf_xdp_link_attach_failed(extack._msg); 9638 goto out_put_dev; 9639 } 9640 9641 fd = bpf_link_settle(&link_primer); 9642 /* link itself doesn't hold dev's refcnt to not complicate shutdown */ 9643 dev_put(dev); 9644 return fd; 9645 9646 unlock: 9647 rtnl_unlock(); 9648 9649 out_put_dev: 9650 dev_put(dev); 9651 return err; 9652 } 9653 9654 /** 9655 * dev_change_xdp_fd - set or clear a bpf program for a device rx path 9656 * @dev: device 9657 * @extack: netlink extended ack 9658 * @fd: new program fd or negative value to clear 9659 * @expected_fd: old program fd that userspace expects to replace or clear 9660 * @flags: xdp-related flags 9661 * 9662 * Set or clear a bpf program for a device 9663 */ 9664 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, 9665 int fd, int expected_fd, u32 flags) 9666 { 9667 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags); 9668 struct bpf_prog *new_prog = NULL, *old_prog = NULL; 9669 int err; 9670 9671 ASSERT_RTNL(); 9672 9673 if (fd >= 0) { 9674 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP, 9675 mode != XDP_MODE_SKB); 9676 if (IS_ERR(new_prog)) 9677 return PTR_ERR(new_prog); 9678 } 9679 9680 if (expected_fd >= 0) { 9681 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP, 9682 mode != XDP_MODE_SKB); 9683 if (IS_ERR(old_prog)) { 9684 err = PTR_ERR(old_prog); 9685 old_prog = NULL; 9686 goto err_out; 9687 } 9688 } 9689 9690 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags); 9691 9692 err_out: 9693 if (err && new_prog) 9694 bpf_prog_put(new_prog); 9695 if (old_prog) 9696 bpf_prog_put(old_prog); 9697 return err; 9698 } 9699 9700 /** 9701 * dev_index_reserve() - allocate an ifindex in a namespace 9702 * @net: the applicable net namespace 9703 * @ifindex: requested ifindex, pass %0 to get one allocated 9704 * 9705 * Allocate a ifindex for a new device. Caller must either use the ifindex 9706 * to store the device (via list_netdevice()) or call dev_index_release() 9707 * to give the index up. 9708 * 9709 * Return: a suitable unique value for a new device interface number or -errno. 9710 */ 9711 static int dev_index_reserve(struct net *net, u32 ifindex) 9712 { 9713 int err; 9714 9715 if (ifindex > INT_MAX) { 9716 DEBUG_NET_WARN_ON_ONCE(1); 9717 return -EINVAL; 9718 } 9719 9720 if (!ifindex) 9721 err = xa_alloc_cyclic(&net->dev_by_index, &ifindex, NULL, 9722 xa_limit_31b, &net->ifindex, GFP_KERNEL); 9723 else 9724 err = xa_insert(&net->dev_by_index, ifindex, NULL, GFP_KERNEL); 9725 if (err < 0) 9726 return err; 9727 9728 return ifindex; 9729 } 9730 9731 static void dev_index_release(struct net *net, int ifindex) 9732 { 9733 /* Expect only unused indexes, unlist_netdevice() removes the used */ 9734 WARN_ON(xa_erase(&net->dev_by_index, ifindex)); 9735 } 9736 9737 /* Delayed registration/unregisteration */ 9738 LIST_HEAD(net_todo_list); 9739 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); 9740 atomic_t dev_unreg_count = ATOMIC_INIT(0); 9741 9742 static void net_set_todo(struct net_device *dev) 9743 { 9744 list_add_tail(&dev->todo_list, &net_todo_list); 9745 } 9746 9747 static netdev_features_t netdev_sync_upper_features(struct net_device *lower, 9748 struct net_device *upper, netdev_features_t features) 9749 { 9750 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9751 netdev_features_t feature; 9752 int feature_bit; 9753 9754 for_each_netdev_feature(upper_disables, feature_bit) { 9755 feature = __NETIF_F_BIT(feature_bit); 9756 if (!(upper->wanted_features & feature) 9757 && (features & feature)) { 9758 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", 9759 &feature, upper->name); 9760 features &= ~feature; 9761 } 9762 } 9763 9764 return features; 9765 } 9766 9767 static void netdev_sync_lower_features(struct net_device *upper, 9768 struct net_device *lower, netdev_features_t features) 9769 { 9770 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; 9771 netdev_features_t feature; 9772 int feature_bit; 9773 9774 for_each_netdev_feature(upper_disables, feature_bit) { 9775 feature = __NETIF_F_BIT(feature_bit); 9776 if (!(features & feature) && (lower->features & feature)) { 9777 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", 9778 &feature, lower->name); 9779 lower->wanted_features &= ~feature; 9780 __netdev_update_features(lower); 9781 9782 if (unlikely(lower->features & feature)) 9783 netdev_WARN(upper, "failed to disable %pNF on %s!\n", 9784 &feature, lower->name); 9785 else 9786 netdev_features_change(lower); 9787 } 9788 } 9789 } 9790 9791 static netdev_features_t netdev_fix_features(struct net_device *dev, 9792 netdev_features_t features) 9793 { 9794 /* Fix illegal checksum combinations */ 9795 if ((features & NETIF_F_HW_CSUM) && 9796 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { 9797 netdev_warn(dev, "mixed HW and IP checksum settings.\n"); 9798 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); 9799 } 9800 9801 /* TSO requires that SG is present as well. */ 9802 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { 9803 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); 9804 features &= ~NETIF_F_ALL_TSO; 9805 } 9806 9807 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && 9808 !(features & NETIF_F_IP_CSUM)) { 9809 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); 9810 features &= ~NETIF_F_TSO; 9811 features &= ~NETIF_F_TSO_ECN; 9812 } 9813 9814 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && 9815 !(features & NETIF_F_IPV6_CSUM)) { 9816 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); 9817 features &= ~NETIF_F_TSO6; 9818 } 9819 9820 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ 9821 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) 9822 features &= ~NETIF_F_TSO_MANGLEID; 9823 9824 /* TSO ECN requires that TSO is present as well. */ 9825 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) 9826 features &= ~NETIF_F_TSO_ECN; 9827 9828 /* Software GSO depends on SG. */ 9829 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { 9830 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); 9831 features &= ~NETIF_F_GSO; 9832 } 9833 9834 /* GSO partial features require GSO partial be set */ 9835 if ((features & dev->gso_partial_features) && 9836 !(features & NETIF_F_GSO_PARTIAL)) { 9837 netdev_dbg(dev, 9838 "Dropping partially supported GSO features since no GSO partial.\n"); 9839 features &= ~dev->gso_partial_features; 9840 } 9841 9842 if (!(features & NETIF_F_RXCSUM)) { 9843 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet 9844 * successfully merged by hardware must also have the 9845 * checksum verified by hardware. If the user does not 9846 * want to enable RXCSUM, logically, we should disable GRO_HW. 9847 */ 9848 if (features & NETIF_F_GRO_HW) { 9849 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n"); 9850 features &= ~NETIF_F_GRO_HW; 9851 } 9852 } 9853 9854 /* LRO/HW-GRO features cannot be combined with RX-FCS */ 9855 if (features & NETIF_F_RXFCS) { 9856 if (features & NETIF_F_LRO) { 9857 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n"); 9858 features &= ~NETIF_F_LRO; 9859 } 9860 9861 if (features & NETIF_F_GRO_HW) { 9862 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n"); 9863 features &= ~NETIF_F_GRO_HW; 9864 } 9865 } 9866 9867 if ((features & NETIF_F_GRO_HW) && (features & NETIF_F_LRO)) { 9868 netdev_dbg(dev, "Dropping LRO feature since HW-GRO is requested.\n"); 9869 features &= ~NETIF_F_LRO; 9870 } 9871 9872 if (features & NETIF_F_HW_TLS_TX) { 9873 bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) == 9874 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM); 9875 bool hw_csum = features & NETIF_F_HW_CSUM; 9876 9877 if (!ip_csum && !hw_csum) { 9878 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n"); 9879 features &= ~NETIF_F_HW_TLS_TX; 9880 } 9881 } 9882 9883 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) { 9884 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n"); 9885 features &= ~NETIF_F_HW_TLS_RX; 9886 } 9887 9888 return features; 9889 } 9890 9891 int __netdev_update_features(struct net_device *dev) 9892 { 9893 struct net_device *upper, *lower; 9894 netdev_features_t features; 9895 struct list_head *iter; 9896 int err = -1; 9897 9898 ASSERT_RTNL(); 9899 9900 features = netdev_get_wanted_features(dev); 9901 9902 if (dev->netdev_ops->ndo_fix_features) 9903 features = dev->netdev_ops->ndo_fix_features(dev, features); 9904 9905 /* driver might be less strict about feature dependencies */ 9906 features = netdev_fix_features(dev, features); 9907 9908 /* some features can't be enabled if they're off on an upper device */ 9909 netdev_for_each_upper_dev_rcu(dev, upper, iter) 9910 features = netdev_sync_upper_features(dev, upper, features); 9911 9912 if (dev->features == features) 9913 goto sync_lower; 9914 9915 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", 9916 &dev->features, &features); 9917 9918 if (dev->netdev_ops->ndo_set_features) 9919 err = dev->netdev_ops->ndo_set_features(dev, features); 9920 else 9921 err = 0; 9922 9923 if (unlikely(err < 0)) { 9924 netdev_err(dev, 9925 "set_features() failed (%d); wanted %pNF, left %pNF\n", 9926 err, &features, &dev->features); 9927 /* return non-0 since some features might have changed and 9928 * it's better to fire a spurious notification than miss it 9929 */ 9930 return -1; 9931 } 9932 9933 sync_lower: 9934 /* some features must be disabled on lower devices when disabled 9935 * on an upper device (think: bonding master or bridge) 9936 */ 9937 netdev_for_each_lower_dev(dev, lower, iter) 9938 netdev_sync_lower_features(dev, lower, features); 9939 9940 if (!err) { 9941 netdev_features_t diff = features ^ dev->features; 9942 9943 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) { 9944 /* udp_tunnel_{get,drop}_rx_info both need 9945 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the 9946 * device, or they won't do anything. 9947 * Thus we need to update dev->features 9948 * *before* calling udp_tunnel_get_rx_info, 9949 * but *after* calling udp_tunnel_drop_rx_info. 9950 */ 9951 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) { 9952 dev->features = features; 9953 udp_tunnel_get_rx_info(dev); 9954 } else { 9955 udp_tunnel_drop_rx_info(dev); 9956 } 9957 } 9958 9959 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) { 9960 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) { 9961 dev->features = features; 9962 err |= vlan_get_rx_ctag_filter_info(dev); 9963 } else { 9964 vlan_drop_rx_ctag_filter_info(dev); 9965 } 9966 } 9967 9968 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) { 9969 if (features & NETIF_F_HW_VLAN_STAG_FILTER) { 9970 dev->features = features; 9971 err |= vlan_get_rx_stag_filter_info(dev); 9972 } else { 9973 vlan_drop_rx_stag_filter_info(dev); 9974 } 9975 } 9976 9977 dev->features = features; 9978 } 9979 9980 return err < 0 ? 0 : 1; 9981 } 9982 9983 /** 9984 * netdev_update_features - recalculate device features 9985 * @dev: the device to check 9986 * 9987 * Recalculate dev->features set and send notifications if it 9988 * has changed. Should be called after driver or hardware dependent 9989 * conditions might have changed that influence the features. 9990 */ 9991 void netdev_update_features(struct net_device *dev) 9992 { 9993 if (__netdev_update_features(dev)) 9994 netdev_features_change(dev); 9995 } 9996 EXPORT_SYMBOL(netdev_update_features); 9997 9998 /** 9999 * netdev_change_features - recalculate device features 10000 * @dev: the device to check 10001 * 10002 * Recalculate dev->features set and send notifications even 10003 * if they have not changed. Should be called instead of 10004 * netdev_update_features() if also dev->vlan_features might 10005 * have changed to allow the changes to be propagated to stacked 10006 * VLAN devices. 10007 */ 10008 void netdev_change_features(struct net_device *dev) 10009 { 10010 __netdev_update_features(dev); 10011 netdev_features_change(dev); 10012 } 10013 EXPORT_SYMBOL(netdev_change_features); 10014 10015 /** 10016 * netif_stacked_transfer_operstate - transfer operstate 10017 * @rootdev: the root or lower level device to transfer state from 10018 * @dev: the device to transfer operstate to 10019 * 10020 * Transfer operational state from root to device. This is normally 10021 * called when a stacking relationship exists between the root 10022 * device and the device(a leaf device). 10023 */ 10024 void netif_stacked_transfer_operstate(const struct net_device *rootdev, 10025 struct net_device *dev) 10026 { 10027 if (rootdev->operstate == IF_OPER_DORMANT) 10028 netif_dormant_on(dev); 10029 else 10030 netif_dormant_off(dev); 10031 10032 if (rootdev->operstate == IF_OPER_TESTING) 10033 netif_testing_on(dev); 10034 else 10035 netif_testing_off(dev); 10036 10037 if (netif_carrier_ok(rootdev)) 10038 netif_carrier_on(dev); 10039 else 10040 netif_carrier_off(dev); 10041 } 10042 EXPORT_SYMBOL(netif_stacked_transfer_operstate); 10043 10044 static int netif_alloc_rx_queues(struct net_device *dev) 10045 { 10046 unsigned int i, count = dev->num_rx_queues; 10047 struct netdev_rx_queue *rx; 10048 size_t sz = count * sizeof(*rx); 10049 int err = 0; 10050 10051 BUG_ON(count < 1); 10052 10053 rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10054 if (!rx) 10055 return -ENOMEM; 10056 10057 dev->_rx = rx; 10058 10059 for (i = 0; i < count; i++) { 10060 rx[i].dev = dev; 10061 10062 /* XDP RX-queue setup */ 10063 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0); 10064 if (err < 0) 10065 goto err_rxq_info; 10066 } 10067 return 0; 10068 10069 err_rxq_info: 10070 /* Rollback successful reg's and free other resources */ 10071 while (i--) 10072 xdp_rxq_info_unreg(&rx[i].xdp_rxq); 10073 kvfree(dev->_rx); 10074 dev->_rx = NULL; 10075 return err; 10076 } 10077 10078 static void netif_free_rx_queues(struct net_device *dev) 10079 { 10080 unsigned int i, count = dev->num_rx_queues; 10081 10082 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */ 10083 if (!dev->_rx) 10084 return; 10085 10086 for (i = 0; i < count; i++) 10087 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq); 10088 10089 kvfree(dev->_rx); 10090 } 10091 10092 static void netdev_init_one_queue(struct net_device *dev, 10093 struct netdev_queue *queue, void *_unused) 10094 { 10095 /* Initialize queue lock */ 10096 spin_lock_init(&queue->_xmit_lock); 10097 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); 10098 queue->xmit_lock_owner = -1; 10099 netdev_queue_numa_node_write(queue, NUMA_NO_NODE); 10100 queue->dev = dev; 10101 #ifdef CONFIG_BQL 10102 dql_init(&queue->dql, HZ); 10103 #endif 10104 } 10105 10106 static void netif_free_tx_queues(struct net_device *dev) 10107 { 10108 kvfree(dev->_tx); 10109 } 10110 10111 static int netif_alloc_netdev_queues(struct net_device *dev) 10112 { 10113 unsigned int count = dev->num_tx_queues; 10114 struct netdev_queue *tx; 10115 size_t sz = count * sizeof(*tx); 10116 10117 if (count < 1 || count > 0xffff) 10118 return -EINVAL; 10119 10120 tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10121 if (!tx) 10122 return -ENOMEM; 10123 10124 dev->_tx = tx; 10125 10126 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); 10127 spin_lock_init(&dev->tx_global_lock); 10128 10129 return 0; 10130 } 10131 10132 void netif_tx_stop_all_queues(struct net_device *dev) 10133 { 10134 unsigned int i; 10135 10136 for (i = 0; i < dev->num_tx_queues; i++) { 10137 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 10138 10139 netif_tx_stop_queue(txq); 10140 } 10141 } 10142 EXPORT_SYMBOL(netif_tx_stop_all_queues); 10143 10144 static int netdev_do_alloc_pcpu_stats(struct net_device *dev) 10145 { 10146 void __percpu *v; 10147 10148 /* Drivers implementing ndo_get_peer_dev must support tstat 10149 * accounting, so that skb_do_redirect() can bump the dev's 10150 * RX stats upon network namespace switch. 10151 */ 10152 if (dev->netdev_ops->ndo_get_peer_dev && 10153 dev->pcpu_stat_type != NETDEV_PCPU_STAT_TSTATS) 10154 return -EOPNOTSUPP; 10155 10156 switch (dev->pcpu_stat_type) { 10157 case NETDEV_PCPU_STAT_NONE: 10158 return 0; 10159 case NETDEV_PCPU_STAT_LSTATS: 10160 v = dev->lstats = netdev_alloc_pcpu_stats(struct pcpu_lstats); 10161 break; 10162 case NETDEV_PCPU_STAT_TSTATS: 10163 v = dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats); 10164 break; 10165 case NETDEV_PCPU_STAT_DSTATS: 10166 v = dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats); 10167 break; 10168 default: 10169 return -EINVAL; 10170 } 10171 10172 return v ? 0 : -ENOMEM; 10173 } 10174 10175 static void netdev_do_free_pcpu_stats(struct net_device *dev) 10176 { 10177 switch (dev->pcpu_stat_type) { 10178 case NETDEV_PCPU_STAT_NONE: 10179 return; 10180 case NETDEV_PCPU_STAT_LSTATS: 10181 free_percpu(dev->lstats); 10182 break; 10183 case NETDEV_PCPU_STAT_TSTATS: 10184 free_percpu(dev->tstats); 10185 break; 10186 case NETDEV_PCPU_STAT_DSTATS: 10187 free_percpu(dev->dstats); 10188 break; 10189 } 10190 } 10191 10192 /** 10193 * register_netdevice() - register a network device 10194 * @dev: device to register 10195 * 10196 * Take a prepared network device structure and make it externally accessible. 10197 * A %NETDEV_REGISTER message is sent to the netdev notifier chain. 10198 * Callers must hold the rtnl lock - you may want register_netdev() 10199 * instead of this. 10200 */ 10201 int register_netdevice(struct net_device *dev) 10202 { 10203 int ret; 10204 struct net *net = dev_net(dev); 10205 10206 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE < 10207 NETDEV_FEATURE_COUNT); 10208 BUG_ON(dev_boot_phase); 10209 ASSERT_RTNL(); 10210 10211 might_sleep(); 10212 10213 /* When net_device's are persistent, this will be fatal. */ 10214 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); 10215 BUG_ON(!net); 10216 10217 ret = ethtool_check_ops(dev->ethtool_ops); 10218 if (ret) 10219 return ret; 10220 10221 spin_lock_init(&dev->addr_list_lock); 10222 netdev_set_addr_lockdep_class(dev); 10223 10224 ret = dev_get_valid_name(net, dev, dev->name); 10225 if (ret < 0) 10226 goto out; 10227 10228 ret = -ENOMEM; 10229 dev->name_node = netdev_name_node_head_alloc(dev); 10230 if (!dev->name_node) 10231 goto out; 10232 10233 /* Init, if this function is available */ 10234 if (dev->netdev_ops->ndo_init) { 10235 ret = dev->netdev_ops->ndo_init(dev); 10236 if (ret) { 10237 if (ret > 0) 10238 ret = -EIO; 10239 goto err_free_name; 10240 } 10241 } 10242 10243 if (((dev->hw_features | dev->features) & 10244 NETIF_F_HW_VLAN_CTAG_FILTER) && 10245 (!dev->netdev_ops->ndo_vlan_rx_add_vid || 10246 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { 10247 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); 10248 ret = -EINVAL; 10249 goto err_uninit; 10250 } 10251 10252 ret = netdev_do_alloc_pcpu_stats(dev); 10253 if (ret) 10254 goto err_uninit; 10255 10256 ret = dev_index_reserve(net, dev->ifindex); 10257 if (ret < 0) 10258 goto err_free_pcpu; 10259 dev->ifindex = ret; 10260 10261 /* Transfer changeable features to wanted_features and enable 10262 * software offloads (GSO and GRO). 10263 */ 10264 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF); 10265 dev->features |= NETIF_F_SOFT_FEATURES; 10266 10267 if (dev->udp_tunnel_nic_info) { 10268 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10269 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT; 10270 } 10271 10272 dev->wanted_features = dev->features & dev->hw_features; 10273 10274 if (!(dev->flags & IFF_LOOPBACK)) 10275 dev->hw_features |= NETIF_F_NOCACHE_COPY; 10276 10277 /* If IPv4 TCP segmentation offload is supported we should also 10278 * allow the device to enable segmenting the frame with the option 10279 * of ignoring a static IP ID value. This doesn't enable the 10280 * feature itself but allows the user to enable it later. 10281 */ 10282 if (dev->hw_features & NETIF_F_TSO) 10283 dev->hw_features |= NETIF_F_TSO_MANGLEID; 10284 if (dev->vlan_features & NETIF_F_TSO) 10285 dev->vlan_features |= NETIF_F_TSO_MANGLEID; 10286 if (dev->mpls_features & NETIF_F_TSO) 10287 dev->mpls_features |= NETIF_F_TSO_MANGLEID; 10288 if (dev->hw_enc_features & NETIF_F_TSO) 10289 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; 10290 10291 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. 10292 */ 10293 dev->vlan_features |= NETIF_F_HIGHDMA; 10294 10295 /* Make NETIF_F_SG inheritable to tunnel devices. 10296 */ 10297 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; 10298 10299 /* Make NETIF_F_SG inheritable to MPLS. 10300 */ 10301 dev->mpls_features |= NETIF_F_SG; 10302 10303 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); 10304 ret = notifier_to_errno(ret); 10305 if (ret) 10306 goto err_ifindex_release; 10307 10308 ret = netdev_register_kobject(dev); 10309 10310 WRITE_ONCE(dev->reg_state, ret ? NETREG_UNREGISTERED : NETREG_REGISTERED); 10311 10312 if (ret) 10313 goto err_uninit_notify; 10314 10315 __netdev_update_features(dev); 10316 10317 /* 10318 * Default initial state at registry is that the 10319 * device is present. 10320 */ 10321 10322 set_bit(__LINK_STATE_PRESENT, &dev->state); 10323 10324 linkwatch_init_dev(dev); 10325 10326 dev_init_scheduler(dev); 10327 10328 netdev_hold(dev, &dev->dev_registered_tracker, GFP_KERNEL); 10329 list_netdevice(dev); 10330 10331 add_device_randomness(dev->dev_addr, dev->addr_len); 10332 10333 /* If the device has permanent device address, driver should 10334 * set dev_addr and also addr_assign_type should be set to 10335 * NET_ADDR_PERM (default value). 10336 */ 10337 if (dev->addr_assign_type == NET_ADDR_PERM) 10338 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); 10339 10340 /* Notify protocols, that a new device appeared. */ 10341 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); 10342 ret = notifier_to_errno(ret); 10343 if (ret) { 10344 /* Expect explicit free_netdev() on failure */ 10345 dev->needs_free_netdev = false; 10346 unregister_netdevice_queue(dev, NULL); 10347 goto out; 10348 } 10349 /* 10350 * Prevent userspace races by waiting until the network 10351 * device is fully setup before sending notifications. 10352 */ 10353 if (!dev->rtnl_link_ops || 10354 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 10355 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL); 10356 10357 out: 10358 return ret; 10359 10360 err_uninit_notify: 10361 call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev); 10362 err_ifindex_release: 10363 dev_index_release(net, dev->ifindex); 10364 err_free_pcpu: 10365 netdev_do_free_pcpu_stats(dev); 10366 err_uninit: 10367 if (dev->netdev_ops->ndo_uninit) 10368 dev->netdev_ops->ndo_uninit(dev); 10369 if (dev->priv_destructor) 10370 dev->priv_destructor(dev); 10371 err_free_name: 10372 netdev_name_node_free(dev->name_node); 10373 goto out; 10374 } 10375 EXPORT_SYMBOL(register_netdevice); 10376 10377 /** 10378 * init_dummy_netdev - init a dummy network device for NAPI 10379 * @dev: device to init 10380 * 10381 * This takes a network device structure and initialize the minimum 10382 * amount of fields so it can be used to schedule NAPI polls without 10383 * registering a full blown interface. This is to be used by drivers 10384 * that need to tie several hardware interfaces to a single NAPI 10385 * poll scheduler due to HW limitations. 10386 */ 10387 void init_dummy_netdev(struct net_device *dev) 10388 { 10389 /* Clear everything. Note we don't initialize spinlocks 10390 * are they aren't supposed to be taken by any of the 10391 * NAPI code and this dummy netdev is supposed to be 10392 * only ever used for NAPI polls 10393 */ 10394 memset(dev, 0, sizeof(struct net_device)); 10395 10396 /* make sure we BUG if trying to hit standard 10397 * register/unregister code path 10398 */ 10399 dev->reg_state = NETREG_DUMMY; 10400 10401 /* NAPI wants this */ 10402 INIT_LIST_HEAD(&dev->napi_list); 10403 10404 /* a dummy interface is started by default */ 10405 set_bit(__LINK_STATE_PRESENT, &dev->state); 10406 set_bit(__LINK_STATE_START, &dev->state); 10407 10408 /* napi_busy_loop stats accounting wants this */ 10409 dev_net_set(dev, &init_net); 10410 10411 /* Note : We dont allocate pcpu_refcnt for dummy devices, 10412 * because users of this 'device' dont need to change 10413 * its refcount. 10414 */ 10415 } 10416 EXPORT_SYMBOL_GPL(init_dummy_netdev); 10417 10418 10419 /** 10420 * register_netdev - register a network device 10421 * @dev: device to register 10422 * 10423 * Take a completed network device structure and add it to the kernel 10424 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier 10425 * chain. 0 is returned on success. A negative errno code is returned 10426 * on a failure to set up the device, or if the name is a duplicate. 10427 * 10428 * This is a wrapper around register_netdevice that takes the rtnl semaphore 10429 * and expands the device name if you passed a format string to 10430 * alloc_netdev. 10431 */ 10432 int register_netdev(struct net_device *dev) 10433 { 10434 int err; 10435 10436 if (rtnl_lock_killable()) 10437 return -EINTR; 10438 err = register_netdevice(dev); 10439 rtnl_unlock(); 10440 return err; 10441 } 10442 EXPORT_SYMBOL(register_netdev); 10443 10444 int netdev_refcnt_read(const struct net_device *dev) 10445 { 10446 #ifdef CONFIG_PCPU_DEV_REFCNT 10447 int i, refcnt = 0; 10448 10449 for_each_possible_cpu(i) 10450 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); 10451 return refcnt; 10452 #else 10453 return refcount_read(&dev->dev_refcnt); 10454 #endif 10455 } 10456 EXPORT_SYMBOL(netdev_refcnt_read); 10457 10458 int netdev_unregister_timeout_secs __read_mostly = 10; 10459 10460 #define WAIT_REFS_MIN_MSECS 1 10461 #define WAIT_REFS_MAX_MSECS 250 10462 /** 10463 * netdev_wait_allrefs_any - wait until all references are gone. 10464 * @list: list of net_devices to wait on 10465 * 10466 * This is called when unregistering network devices. 10467 * 10468 * Any protocol or device that holds a reference should register 10469 * for netdevice notification, and cleanup and put back the 10470 * reference if they receive an UNREGISTER event. 10471 * We can get stuck here if buggy protocols don't correctly 10472 * call dev_put. 10473 */ 10474 static struct net_device *netdev_wait_allrefs_any(struct list_head *list) 10475 { 10476 unsigned long rebroadcast_time, warning_time; 10477 struct net_device *dev; 10478 int wait = 0; 10479 10480 rebroadcast_time = warning_time = jiffies; 10481 10482 list_for_each_entry(dev, list, todo_list) 10483 if (netdev_refcnt_read(dev) == 1) 10484 return dev; 10485 10486 while (true) { 10487 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { 10488 rtnl_lock(); 10489 10490 /* Rebroadcast unregister notification */ 10491 list_for_each_entry(dev, list, todo_list) 10492 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 10493 10494 __rtnl_unlock(); 10495 rcu_barrier(); 10496 rtnl_lock(); 10497 10498 list_for_each_entry(dev, list, todo_list) 10499 if (test_bit(__LINK_STATE_LINKWATCH_PENDING, 10500 &dev->state)) { 10501 /* We must not have linkwatch events 10502 * pending on unregister. If this 10503 * happens, we simply run the queue 10504 * unscheduled, resulting in a noop 10505 * for this device. 10506 */ 10507 linkwatch_run_queue(); 10508 break; 10509 } 10510 10511 __rtnl_unlock(); 10512 10513 rebroadcast_time = jiffies; 10514 } 10515 10516 if (!wait) { 10517 rcu_barrier(); 10518 wait = WAIT_REFS_MIN_MSECS; 10519 } else { 10520 msleep(wait); 10521 wait = min(wait << 1, WAIT_REFS_MAX_MSECS); 10522 } 10523 10524 list_for_each_entry(dev, list, todo_list) 10525 if (netdev_refcnt_read(dev) == 1) 10526 return dev; 10527 10528 if (time_after(jiffies, warning_time + 10529 READ_ONCE(netdev_unregister_timeout_secs) * HZ)) { 10530 list_for_each_entry(dev, list, todo_list) { 10531 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", 10532 dev->name, netdev_refcnt_read(dev)); 10533 ref_tracker_dir_print(&dev->refcnt_tracker, 10); 10534 } 10535 10536 warning_time = jiffies; 10537 } 10538 } 10539 } 10540 10541 /* The sequence is: 10542 * 10543 * rtnl_lock(); 10544 * ... 10545 * register_netdevice(x1); 10546 * register_netdevice(x2); 10547 * ... 10548 * unregister_netdevice(y1); 10549 * unregister_netdevice(y2); 10550 * ... 10551 * rtnl_unlock(); 10552 * free_netdev(y1); 10553 * free_netdev(y2); 10554 * 10555 * We are invoked by rtnl_unlock(). 10556 * This allows us to deal with problems: 10557 * 1) We can delete sysfs objects which invoke hotplug 10558 * without deadlocking with linkwatch via keventd. 10559 * 2) Since we run with the RTNL semaphore not held, we can sleep 10560 * safely in order to wait for the netdev refcnt to drop to zero. 10561 * 10562 * We must not return until all unregister events added during 10563 * the interval the lock was held have been completed. 10564 */ 10565 void netdev_run_todo(void) 10566 { 10567 struct net_device *dev, *tmp; 10568 struct list_head list; 10569 int cnt; 10570 #ifdef CONFIG_LOCKDEP 10571 struct list_head unlink_list; 10572 10573 list_replace_init(&net_unlink_list, &unlink_list); 10574 10575 while (!list_empty(&unlink_list)) { 10576 struct net_device *dev = list_first_entry(&unlink_list, 10577 struct net_device, 10578 unlink_list); 10579 list_del_init(&dev->unlink_list); 10580 dev->nested_level = dev->lower_level - 1; 10581 } 10582 #endif 10583 10584 /* Snapshot list, allow later requests */ 10585 list_replace_init(&net_todo_list, &list); 10586 10587 __rtnl_unlock(); 10588 10589 /* Wait for rcu callbacks to finish before next phase */ 10590 if (!list_empty(&list)) 10591 rcu_barrier(); 10592 10593 list_for_each_entry_safe(dev, tmp, &list, todo_list) { 10594 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { 10595 netdev_WARN(dev, "run_todo but not unregistering\n"); 10596 list_del(&dev->todo_list); 10597 continue; 10598 } 10599 10600 WRITE_ONCE(dev->reg_state, NETREG_UNREGISTERED); 10601 linkwatch_sync_dev(dev); 10602 } 10603 10604 cnt = 0; 10605 while (!list_empty(&list)) { 10606 dev = netdev_wait_allrefs_any(&list); 10607 list_del(&dev->todo_list); 10608 10609 /* paranoia */ 10610 BUG_ON(netdev_refcnt_read(dev) != 1); 10611 BUG_ON(!list_empty(&dev->ptype_all)); 10612 BUG_ON(!list_empty(&dev->ptype_specific)); 10613 WARN_ON(rcu_access_pointer(dev->ip_ptr)); 10614 WARN_ON(rcu_access_pointer(dev->ip6_ptr)); 10615 10616 netdev_do_free_pcpu_stats(dev); 10617 if (dev->priv_destructor) 10618 dev->priv_destructor(dev); 10619 if (dev->needs_free_netdev) 10620 free_netdev(dev); 10621 10622 cnt++; 10623 10624 /* Free network device */ 10625 kobject_put(&dev->dev.kobj); 10626 } 10627 if (cnt && atomic_sub_and_test(cnt, &dev_unreg_count)) 10628 wake_up(&netdev_unregistering_wq); 10629 } 10630 10631 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has 10632 * all the same fields in the same order as net_device_stats, with only 10633 * the type differing, but rtnl_link_stats64 may have additional fields 10634 * at the end for newer counters. 10635 */ 10636 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, 10637 const struct net_device_stats *netdev_stats) 10638 { 10639 size_t i, n = sizeof(*netdev_stats) / sizeof(atomic_long_t); 10640 const atomic_long_t *src = (atomic_long_t *)netdev_stats; 10641 u64 *dst = (u64 *)stats64; 10642 10643 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); 10644 for (i = 0; i < n; i++) 10645 dst[i] = (unsigned long)atomic_long_read(&src[i]); 10646 /* zero out counters that only exist in rtnl_link_stats64 */ 10647 memset((char *)stats64 + n * sizeof(u64), 0, 10648 sizeof(*stats64) - n * sizeof(u64)); 10649 } 10650 EXPORT_SYMBOL(netdev_stats_to_stats64); 10651 10652 static __cold struct net_device_core_stats __percpu *netdev_core_stats_alloc( 10653 struct net_device *dev) 10654 { 10655 struct net_device_core_stats __percpu *p; 10656 10657 p = alloc_percpu_gfp(struct net_device_core_stats, 10658 GFP_ATOMIC | __GFP_NOWARN); 10659 10660 if (p && cmpxchg(&dev->core_stats, NULL, p)) 10661 free_percpu(p); 10662 10663 /* This READ_ONCE() pairs with the cmpxchg() above */ 10664 return READ_ONCE(dev->core_stats); 10665 } 10666 10667 noinline void netdev_core_stats_inc(struct net_device *dev, u32 offset) 10668 { 10669 /* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */ 10670 struct net_device_core_stats __percpu *p = READ_ONCE(dev->core_stats); 10671 unsigned long __percpu *field; 10672 10673 if (unlikely(!p)) { 10674 p = netdev_core_stats_alloc(dev); 10675 if (!p) 10676 return; 10677 } 10678 10679 field = (__force unsigned long __percpu *)((__force void *)p + offset); 10680 this_cpu_inc(*field); 10681 } 10682 EXPORT_SYMBOL_GPL(netdev_core_stats_inc); 10683 10684 /** 10685 * dev_get_stats - get network device statistics 10686 * @dev: device to get statistics from 10687 * @storage: place to store stats 10688 * 10689 * Get network statistics from device. Return @storage. 10690 * The device driver may provide its own method by setting 10691 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; 10692 * otherwise the internal statistics structure is used. 10693 */ 10694 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, 10695 struct rtnl_link_stats64 *storage) 10696 { 10697 const struct net_device_ops *ops = dev->netdev_ops; 10698 const struct net_device_core_stats __percpu *p; 10699 10700 if (ops->ndo_get_stats64) { 10701 memset(storage, 0, sizeof(*storage)); 10702 ops->ndo_get_stats64(dev, storage); 10703 } else if (ops->ndo_get_stats) { 10704 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); 10705 } else if (dev->pcpu_stat_type == NETDEV_PCPU_STAT_TSTATS) { 10706 dev_get_tstats64(dev, storage); 10707 } else { 10708 netdev_stats_to_stats64(storage, &dev->stats); 10709 } 10710 10711 /* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */ 10712 p = READ_ONCE(dev->core_stats); 10713 if (p) { 10714 const struct net_device_core_stats *core_stats; 10715 int i; 10716 10717 for_each_possible_cpu(i) { 10718 core_stats = per_cpu_ptr(p, i); 10719 storage->rx_dropped += READ_ONCE(core_stats->rx_dropped); 10720 storage->tx_dropped += READ_ONCE(core_stats->tx_dropped); 10721 storage->rx_nohandler += READ_ONCE(core_stats->rx_nohandler); 10722 storage->rx_otherhost_dropped += READ_ONCE(core_stats->rx_otherhost_dropped); 10723 } 10724 } 10725 return storage; 10726 } 10727 EXPORT_SYMBOL(dev_get_stats); 10728 10729 /** 10730 * dev_fetch_sw_netstats - get per-cpu network device statistics 10731 * @s: place to store stats 10732 * @netstats: per-cpu network stats to read from 10733 * 10734 * Read per-cpu network statistics and populate the related fields in @s. 10735 */ 10736 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s, 10737 const struct pcpu_sw_netstats __percpu *netstats) 10738 { 10739 int cpu; 10740 10741 for_each_possible_cpu(cpu) { 10742 u64 rx_packets, rx_bytes, tx_packets, tx_bytes; 10743 const struct pcpu_sw_netstats *stats; 10744 unsigned int start; 10745 10746 stats = per_cpu_ptr(netstats, cpu); 10747 do { 10748 start = u64_stats_fetch_begin(&stats->syncp); 10749 rx_packets = u64_stats_read(&stats->rx_packets); 10750 rx_bytes = u64_stats_read(&stats->rx_bytes); 10751 tx_packets = u64_stats_read(&stats->tx_packets); 10752 tx_bytes = u64_stats_read(&stats->tx_bytes); 10753 } while (u64_stats_fetch_retry(&stats->syncp, start)); 10754 10755 s->rx_packets += rx_packets; 10756 s->rx_bytes += rx_bytes; 10757 s->tx_packets += tx_packets; 10758 s->tx_bytes += tx_bytes; 10759 } 10760 } 10761 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats); 10762 10763 /** 10764 * dev_get_tstats64 - ndo_get_stats64 implementation 10765 * @dev: device to get statistics from 10766 * @s: place to store stats 10767 * 10768 * Populate @s from dev->stats and dev->tstats. Can be used as 10769 * ndo_get_stats64() callback. 10770 */ 10771 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s) 10772 { 10773 netdev_stats_to_stats64(s, &dev->stats); 10774 dev_fetch_sw_netstats(s, dev->tstats); 10775 } 10776 EXPORT_SYMBOL_GPL(dev_get_tstats64); 10777 10778 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) 10779 { 10780 struct netdev_queue *queue = dev_ingress_queue(dev); 10781 10782 #ifdef CONFIG_NET_CLS_ACT 10783 if (queue) 10784 return queue; 10785 queue = kzalloc(sizeof(*queue), GFP_KERNEL); 10786 if (!queue) 10787 return NULL; 10788 netdev_init_one_queue(dev, queue, NULL); 10789 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); 10790 RCU_INIT_POINTER(queue->qdisc_sleeping, &noop_qdisc); 10791 rcu_assign_pointer(dev->ingress_queue, queue); 10792 #endif 10793 return queue; 10794 } 10795 10796 static const struct ethtool_ops default_ethtool_ops; 10797 10798 void netdev_set_default_ethtool_ops(struct net_device *dev, 10799 const struct ethtool_ops *ops) 10800 { 10801 if (dev->ethtool_ops == &default_ethtool_ops) 10802 dev->ethtool_ops = ops; 10803 } 10804 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); 10805 10806 /** 10807 * netdev_sw_irq_coalesce_default_on() - enable SW IRQ coalescing by default 10808 * @dev: netdev to enable the IRQ coalescing on 10809 * 10810 * Sets a conservative default for SW IRQ coalescing. Users can use 10811 * sysfs attributes to override the default values. 10812 */ 10813 void netdev_sw_irq_coalesce_default_on(struct net_device *dev) 10814 { 10815 WARN_ON(dev->reg_state == NETREG_REGISTERED); 10816 10817 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) { 10818 dev->gro_flush_timeout = 20000; 10819 dev->napi_defer_hard_irqs = 1; 10820 } 10821 } 10822 EXPORT_SYMBOL_GPL(netdev_sw_irq_coalesce_default_on); 10823 10824 void netdev_freemem(struct net_device *dev) 10825 { 10826 char *addr = (char *)dev - dev->padded; 10827 10828 kvfree(addr); 10829 } 10830 10831 /** 10832 * alloc_netdev_mqs - allocate network device 10833 * @sizeof_priv: size of private data to allocate space for 10834 * @name: device name format string 10835 * @name_assign_type: origin of device name 10836 * @setup: callback to initialize device 10837 * @txqs: the number of TX subqueues to allocate 10838 * @rxqs: the number of RX subqueues to allocate 10839 * 10840 * Allocates a struct net_device with private data area for driver use 10841 * and performs basic initialization. Also allocates subqueue structs 10842 * for each queue on the device. 10843 */ 10844 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, 10845 unsigned char name_assign_type, 10846 void (*setup)(struct net_device *), 10847 unsigned int txqs, unsigned int rxqs) 10848 { 10849 struct net_device *dev; 10850 unsigned int alloc_size; 10851 struct net_device *p; 10852 10853 BUG_ON(strlen(name) >= sizeof(dev->name)); 10854 10855 if (txqs < 1) { 10856 pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); 10857 return NULL; 10858 } 10859 10860 if (rxqs < 1) { 10861 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); 10862 return NULL; 10863 } 10864 10865 alloc_size = sizeof(struct net_device); 10866 if (sizeof_priv) { 10867 /* ensure 32-byte alignment of private area */ 10868 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); 10869 alloc_size += sizeof_priv; 10870 } 10871 /* ensure 32-byte alignment of whole construct */ 10872 alloc_size += NETDEV_ALIGN - 1; 10873 10874 p = kvzalloc(alloc_size, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); 10875 if (!p) 10876 return NULL; 10877 10878 dev = PTR_ALIGN(p, NETDEV_ALIGN); 10879 dev->padded = (char *)dev - (char *)p; 10880 10881 ref_tracker_dir_init(&dev->refcnt_tracker, 128, name); 10882 #ifdef CONFIG_PCPU_DEV_REFCNT 10883 dev->pcpu_refcnt = alloc_percpu(int); 10884 if (!dev->pcpu_refcnt) 10885 goto free_dev; 10886 __dev_hold(dev); 10887 #else 10888 refcount_set(&dev->dev_refcnt, 1); 10889 #endif 10890 10891 if (dev_addr_init(dev)) 10892 goto free_pcpu; 10893 10894 dev_mc_init(dev); 10895 dev_uc_init(dev); 10896 10897 dev_net_set(dev, &init_net); 10898 10899 dev->gso_max_size = GSO_LEGACY_MAX_SIZE; 10900 dev->xdp_zc_max_segs = 1; 10901 dev->gso_max_segs = GSO_MAX_SEGS; 10902 dev->gro_max_size = GRO_LEGACY_MAX_SIZE; 10903 dev->gso_ipv4_max_size = GSO_LEGACY_MAX_SIZE; 10904 dev->gro_ipv4_max_size = GRO_LEGACY_MAX_SIZE; 10905 dev->tso_max_size = TSO_LEGACY_MAX_SIZE; 10906 dev->tso_max_segs = TSO_MAX_SEGS; 10907 dev->upper_level = 1; 10908 dev->lower_level = 1; 10909 #ifdef CONFIG_LOCKDEP 10910 dev->nested_level = 0; 10911 INIT_LIST_HEAD(&dev->unlink_list); 10912 #endif 10913 10914 INIT_LIST_HEAD(&dev->napi_list); 10915 INIT_LIST_HEAD(&dev->unreg_list); 10916 INIT_LIST_HEAD(&dev->close_list); 10917 INIT_LIST_HEAD(&dev->link_watch_list); 10918 INIT_LIST_HEAD(&dev->adj_list.upper); 10919 INIT_LIST_HEAD(&dev->adj_list.lower); 10920 INIT_LIST_HEAD(&dev->ptype_all); 10921 INIT_LIST_HEAD(&dev->ptype_specific); 10922 INIT_LIST_HEAD(&dev->net_notifier_list); 10923 #ifdef CONFIG_NET_SCHED 10924 hash_init(dev->qdisc_hash); 10925 #endif 10926 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; 10927 setup(dev); 10928 10929 if (!dev->tx_queue_len) { 10930 dev->priv_flags |= IFF_NO_QUEUE; 10931 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; 10932 } 10933 10934 dev->num_tx_queues = txqs; 10935 dev->real_num_tx_queues = txqs; 10936 if (netif_alloc_netdev_queues(dev)) 10937 goto free_all; 10938 10939 dev->num_rx_queues = rxqs; 10940 dev->real_num_rx_queues = rxqs; 10941 if (netif_alloc_rx_queues(dev)) 10942 goto free_all; 10943 10944 strcpy(dev->name, name); 10945 dev->name_assign_type = name_assign_type; 10946 dev->group = INIT_NETDEV_GROUP; 10947 if (!dev->ethtool_ops) 10948 dev->ethtool_ops = &default_ethtool_ops; 10949 10950 nf_hook_netdev_init(dev); 10951 10952 return dev; 10953 10954 free_all: 10955 free_netdev(dev); 10956 return NULL; 10957 10958 free_pcpu: 10959 #ifdef CONFIG_PCPU_DEV_REFCNT 10960 free_percpu(dev->pcpu_refcnt); 10961 free_dev: 10962 #endif 10963 netdev_freemem(dev); 10964 return NULL; 10965 } 10966 EXPORT_SYMBOL(alloc_netdev_mqs); 10967 10968 /** 10969 * free_netdev - free network device 10970 * @dev: device 10971 * 10972 * This function does the last stage of destroying an allocated device 10973 * interface. The reference to the device object is released. If this 10974 * is the last reference then it will be freed.Must be called in process 10975 * context. 10976 */ 10977 void free_netdev(struct net_device *dev) 10978 { 10979 struct napi_struct *p, *n; 10980 10981 might_sleep(); 10982 10983 /* When called immediately after register_netdevice() failed the unwind 10984 * handling may still be dismantling the device. Handle that case by 10985 * deferring the free. 10986 */ 10987 if (dev->reg_state == NETREG_UNREGISTERING) { 10988 ASSERT_RTNL(); 10989 dev->needs_free_netdev = true; 10990 return; 10991 } 10992 10993 netif_free_tx_queues(dev); 10994 netif_free_rx_queues(dev); 10995 10996 kfree(rcu_dereference_protected(dev->ingress_queue, 1)); 10997 10998 /* Flush device addresses */ 10999 dev_addr_flush(dev); 11000 11001 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) 11002 netif_napi_del(p); 11003 11004 ref_tracker_dir_exit(&dev->refcnt_tracker); 11005 #ifdef CONFIG_PCPU_DEV_REFCNT 11006 free_percpu(dev->pcpu_refcnt); 11007 dev->pcpu_refcnt = NULL; 11008 #endif 11009 free_percpu(dev->core_stats); 11010 dev->core_stats = NULL; 11011 free_percpu(dev->xdp_bulkq); 11012 dev->xdp_bulkq = NULL; 11013 11014 /* Compatibility with error handling in drivers */ 11015 if (dev->reg_state == NETREG_UNINITIALIZED) { 11016 netdev_freemem(dev); 11017 return; 11018 } 11019 11020 BUG_ON(dev->reg_state != NETREG_UNREGISTERED); 11021 WRITE_ONCE(dev->reg_state, NETREG_RELEASED); 11022 11023 /* will free via device release */ 11024 put_device(&dev->dev); 11025 } 11026 EXPORT_SYMBOL(free_netdev); 11027 11028 /** 11029 * synchronize_net - Synchronize with packet receive processing 11030 * 11031 * Wait for packets currently being received to be done. 11032 * Does not block later packets from starting. 11033 */ 11034 void synchronize_net(void) 11035 { 11036 might_sleep(); 11037 if (rtnl_is_locked()) 11038 synchronize_rcu_expedited(); 11039 else 11040 synchronize_rcu(); 11041 } 11042 EXPORT_SYMBOL(synchronize_net); 11043 11044 /** 11045 * unregister_netdevice_queue - remove device from the kernel 11046 * @dev: device 11047 * @head: list 11048 * 11049 * This function shuts down a device interface and removes it 11050 * from the kernel tables. 11051 * If head not NULL, device is queued to be unregistered later. 11052 * 11053 * Callers must hold the rtnl semaphore. You may want 11054 * unregister_netdev() instead of this. 11055 */ 11056 11057 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) 11058 { 11059 ASSERT_RTNL(); 11060 11061 if (head) { 11062 list_move_tail(&dev->unreg_list, head); 11063 } else { 11064 LIST_HEAD(single); 11065 11066 list_add(&dev->unreg_list, &single); 11067 unregister_netdevice_many(&single); 11068 } 11069 } 11070 EXPORT_SYMBOL(unregister_netdevice_queue); 11071 11072 void unregister_netdevice_many_notify(struct list_head *head, 11073 u32 portid, const struct nlmsghdr *nlh) 11074 { 11075 struct net_device *dev, *tmp; 11076 LIST_HEAD(close_head); 11077 int cnt = 0; 11078 11079 BUG_ON(dev_boot_phase); 11080 ASSERT_RTNL(); 11081 11082 if (list_empty(head)) 11083 return; 11084 11085 list_for_each_entry_safe(dev, tmp, head, unreg_list) { 11086 /* Some devices call without registering 11087 * for initialization unwind. Remove those 11088 * devices and proceed with the remaining. 11089 */ 11090 if (dev->reg_state == NETREG_UNINITIALIZED) { 11091 pr_debug("unregister_netdevice: device %s/%p never was registered\n", 11092 dev->name, dev); 11093 11094 WARN_ON(1); 11095 list_del(&dev->unreg_list); 11096 continue; 11097 } 11098 dev->dismantle = true; 11099 BUG_ON(dev->reg_state != NETREG_REGISTERED); 11100 } 11101 11102 /* If device is running, close it first. */ 11103 list_for_each_entry(dev, head, unreg_list) 11104 list_add_tail(&dev->close_list, &close_head); 11105 dev_close_many(&close_head, true); 11106 11107 list_for_each_entry(dev, head, unreg_list) { 11108 /* And unlink it from device chain. */ 11109 unlist_netdevice(dev); 11110 WRITE_ONCE(dev->reg_state, NETREG_UNREGISTERING); 11111 } 11112 flush_all_backlogs(); 11113 11114 synchronize_net(); 11115 11116 list_for_each_entry(dev, head, unreg_list) { 11117 struct sk_buff *skb = NULL; 11118 11119 /* Shutdown queueing discipline. */ 11120 dev_shutdown(dev); 11121 dev_tcx_uninstall(dev); 11122 dev_xdp_uninstall(dev); 11123 bpf_dev_bound_netdev_unregister(dev); 11124 11125 netdev_offload_xstats_disable_all(dev); 11126 11127 /* Notify protocols, that we are about to destroy 11128 * this device. They should clean all the things. 11129 */ 11130 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 11131 11132 if (!dev->rtnl_link_ops || 11133 dev->rtnl_link_state == RTNL_LINK_INITIALIZED) 11134 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0, 11135 GFP_KERNEL, NULL, 0, 11136 portid, nlh); 11137 11138 /* 11139 * Flush the unicast and multicast chains 11140 */ 11141 dev_uc_flush(dev); 11142 dev_mc_flush(dev); 11143 11144 netdev_name_node_alt_flush(dev); 11145 netdev_name_node_free(dev->name_node); 11146 11147 call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev); 11148 11149 if (dev->netdev_ops->ndo_uninit) 11150 dev->netdev_ops->ndo_uninit(dev); 11151 11152 if (skb) 11153 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL, portid, nlh); 11154 11155 /* Notifier chain MUST detach us all upper devices. */ 11156 WARN_ON(netdev_has_any_upper_dev(dev)); 11157 WARN_ON(netdev_has_any_lower_dev(dev)); 11158 11159 /* Remove entries from kobject tree */ 11160 netdev_unregister_kobject(dev); 11161 #ifdef CONFIG_XPS 11162 /* Remove XPS queueing entries */ 11163 netif_reset_xps_queues_gt(dev, 0); 11164 #endif 11165 } 11166 11167 synchronize_net(); 11168 11169 list_for_each_entry(dev, head, unreg_list) { 11170 netdev_put(dev, &dev->dev_registered_tracker); 11171 net_set_todo(dev); 11172 cnt++; 11173 } 11174 atomic_add(cnt, &dev_unreg_count); 11175 11176 list_del(head); 11177 } 11178 11179 /** 11180 * unregister_netdevice_many - unregister many devices 11181 * @head: list of devices 11182 * 11183 * Note: As most callers use a stack allocated list_head, 11184 * we force a list_del() to make sure stack wont be corrupted later. 11185 */ 11186 void unregister_netdevice_many(struct list_head *head) 11187 { 11188 unregister_netdevice_many_notify(head, 0, NULL); 11189 } 11190 EXPORT_SYMBOL(unregister_netdevice_many); 11191 11192 /** 11193 * unregister_netdev - remove device from the kernel 11194 * @dev: device 11195 * 11196 * This function shuts down a device interface and removes it 11197 * from the kernel tables. 11198 * 11199 * This is just a wrapper for unregister_netdevice that takes 11200 * the rtnl semaphore. In general you want to use this and not 11201 * unregister_netdevice. 11202 */ 11203 void unregister_netdev(struct net_device *dev) 11204 { 11205 rtnl_lock(); 11206 unregister_netdevice(dev); 11207 rtnl_unlock(); 11208 } 11209 EXPORT_SYMBOL(unregister_netdev); 11210 11211 /** 11212 * __dev_change_net_namespace - move device to different nethost namespace 11213 * @dev: device 11214 * @net: network namespace 11215 * @pat: If not NULL name pattern to try if the current device name 11216 * is already taken in the destination network namespace. 11217 * @new_ifindex: If not zero, specifies device index in the target 11218 * namespace. 11219 * 11220 * This function shuts down a device interface and moves it 11221 * to a new network namespace. On success 0 is returned, on 11222 * a failure a netagive errno code is returned. 11223 * 11224 * Callers must hold the rtnl semaphore. 11225 */ 11226 11227 int __dev_change_net_namespace(struct net_device *dev, struct net *net, 11228 const char *pat, int new_ifindex) 11229 { 11230 struct netdev_name_node *name_node; 11231 struct net *net_old = dev_net(dev); 11232 char new_name[IFNAMSIZ] = {}; 11233 int err, new_nsid; 11234 11235 ASSERT_RTNL(); 11236 11237 /* Don't allow namespace local devices to be moved. */ 11238 err = -EINVAL; 11239 if (dev->features & NETIF_F_NETNS_LOCAL) 11240 goto out; 11241 11242 /* Ensure the device has been registrered */ 11243 if (dev->reg_state != NETREG_REGISTERED) 11244 goto out; 11245 11246 /* Get out if there is nothing todo */ 11247 err = 0; 11248 if (net_eq(net_old, net)) 11249 goto out; 11250 11251 /* Pick the destination device name, and ensure 11252 * we can use it in the destination network namespace. 11253 */ 11254 err = -EEXIST; 11255 if (netdev_name_in_use(net, dev->name)) { 11256 /* We get here if we can't use the current device name */ 11257 if (!pat) 11258 goto out; 11259 err = dev_prep_valid_name(net, dev, pat, new_name, EEXIST); 11260 if (err < 0) 11261 goto out; 11262 } 11263 /* Check that none of the altnames conflicts. */ 11264 err = -EEXIST; 11265 netdev_for_each_altname(dev, name_node) 11266 if (netdev_name_in_use(net, name_node->name)) 11267 goto out; 11268 11269 /* Check that new_ifindex isn't used yet. */ 11270 if (new_ifindex) { 11271 err = dev_index_reserve(net, new_ifindex); 11272 if (err < 0) 11273 goto out; 11274 } else { 11275 /* If there is an ifindex conflict assign a new one */ 11276 err = dev_index_reserve(net, dev->ifindex); 11277 if (err == -EBUSY) 11278 err = dev_index_reserve(net, 0); 11279 if (err < 0) 11280 goto out; 11281 new_ifindex = err; 11282 } 11283 11284 /* 11285 * And now a mini version of register_netdevice unregister_netdevice. 11286 */ 11287 11288 /* If device is running close it first. */ 11289 dev_close(dev); 11290 11291 /* And unlink it from device chain */ 11292 unlist_netdevice(dev); 11293 11294 synchronize_net(); 11295 11296 /* Shutdown queueing discipline. */ 11297 dev_shutdown(dev); 11298 11299 /* Notify protocols, that we are about to destroy 11300 * this device. They should clean all the things. 11301 * 11302 * Note that dev->reg_state stays at NETREG_REGISTERED. 11303 * This is wanted because this way 8021q and macvlan know 11304 * the device is just moving and can keep their slaves up. 11305 */ 11306 call_netdevice_notifiers(NETDEV_UNREGISTER, dev); 11307 rcu_barrier(); 11308 11309 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL); 11310 11311 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid, 11312 new_ifindex); 11313 11314 /* 11315 * Flush the unicast and multicast chains 11316 */ 11317 dev_uc_flush(dev); 11318 dev_mc_flush(dev); 11319 11320 /* Send a netdev-removed uevent to the old namespace */ 11321 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); 11322 netdev_adjacent_del_links(dev); 11323 11324 /* Move per-net netdevice notifiers that are following the netdevice */ 11325 move_netdevice_notifiers_dev_net(dev, net); 11326 11327 /* Actually switch the network namespace */ 11328 dev_net_set(dev, net); 11329 dev->ifindex = new_ifindex; 11330 11331 if (new_name[0]) /* Rename the netdev to prepared name */ 11332 strscpy(dev->name, new_name, IFNAMSIZ); 11333 11334 /* Fixup kobjects */ 11335 dev_set_uevent_suppress(&dev->dev, 1); 11336 err = device_rename(&dev->dev, dev->name); 11337 dev_set_uevent_suppress(&dev->dev, 0); 11338 WARN_ON(err); 11339 11340 /* Send a netdev-add uevent to the new namespace */ 11341 kobject_uevent(&dev->dev.kobj, KOBJ_ADD); 11342 netdev_adjacent_add_links(dev); 11343 11344 /* Adapt owner in case owning user namespace of target network 11345 * namespace is different from the original one. 11346 */ 11347 err = netdev_change_owner(dev, net_old, net); 11348 WARN_ON(err); 11349 11350 /* Add the device back in the hashes */ 11351 list_netdevice(dev); 11352 11353 /* Notify protocols, that a new device appeared. */ 11354 call_netdevice_notifiers(NETDEV_REGISTER, dev); 11355 11356 /* 11357 * Prevent userspace races by waiting until the network 11358 * device is fully setup before sending notifications. 11359 */ 11360 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL); 11361 11362 synchronize_net(); 11363 err = 0; 11364 out: 11365 return err; 11366 } 11367 EXPORT_SYMBOL_GPL(__dev_change_net_namespace); 11368 11369 static int dev_cpu_dead(unsigned int oldcpu) 11370 { 11371 struct sk_buff **list_skb; 11372 struct sk_buff *skb; 11373 unsigned int cpu; 11374 struct softnet_data *sd, *oldsd, *remsd = NULL; 11375 11376 local_irq_disable(); 11377 cpu = smp_processor_id(); 11378 sd = &per_cpu(softnet_data, cpu); 11379 oldsd = &per_cpu(softnet_data, oldcpu); 11380 11381 /* Find end of our completion_queue. */ 11382 list_skb = &sd->completion_queue; 11383 while (*list_skb) 11384 list_skb = &(*list_skb)->next; 11385 /* Append completion queue from offline CPU. */ 11386 *list_skb = oldsd->completion_queue; 11387 oldsd->completion_queue = NULL; 11388 11389 /* Append output queue from offline CPU. */ 11390 if (oldsd->output_queue) { 11391 *sd->output_queue_tailp = oldsd->output_queue; 11392 sd->output_queue_tailp = oldsd->output_queue_tailp; 11393 oldsd->output_queue = NULL; 11394 oldsd->output_queue_tailp = &oldsd->output_queue; 11395 } 11396 /* Append NAPI poll list from offline CPU, with one exception : 11397 * process_backlog() must be called by cpu owning percpu backlog. 11398 * We properly handle process_queue & input_pkt_queue later. 11399 */ 11400 while (!list_empty(&oldsd->poll_list)) { 11401 struct napi_struct *napi = list_first_entry(&oldsd->poll_list, 11402 struct napi_struct, 11403 poll_list); 11404 11405 list_del_init(&napi->poll_list); 11406 if (napi->poll == process_backlog) 11407 napi->state = 0; 11408 else 11409 ____napi_schedule(sd, napi); 11410 } 11411 11412 raise_softirq_irqoff(NET_TX_SOFTIRQ); 11413 local_irq_enable(); 11414 11415 #ifdef CONFIG_RPS 11416 remsd = oldsd->rps_ipi_list; 11417 oldsd->rps_ipi_list = NULL; 11418 #endif 11419 /* send out pending IPI's on offline CPU */ 11420 net_rps_send_ipi(remsd); 11421 11422 /* Process offline CPU's input_pkt_queue */ 11423 while ((skb = __skb_dequeue(&oldsd->process_queue))) { 11424 netif_rx(skb); 11425 input_queue_head_incr(oldsd); 11426 } 11427 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { 11428 netif_rx(skb); 11429 input_queue_head_incr(oldsd); 11430 } 11431 11432 return 0; 11433 } 11434 11435 /** 11436 * netdev_increment_features - increment feature set by one 11437 * @all: current feature set 11438 * @one: new feature set 11439 * @mask: mask feature set 11440 * 11441 * Computes a new feature set after adding a device with feature set 11442 * @one to the master device with current feature set @all. Will not 11443 * enable anything that is off in @mask. Returns the new feature set. 11444 */ 11445 netdev_features_t netdev_increment_features(netdev_features_t all, 11446 netdev_features_t one, netdev_features_t mask) 11447 { 11448 if (mask & NETIF_F_HW_CSUM) 11449 mask |= NETIF_F_CSUM_MASK; 11450 mask |= NETIF_F_VLAN_CHALLENGED; 11451 11452 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; 11453 all &= one | ~NETIF_F_ALL_FOR_ALL; 11454 11455 /* If one device supports hw checksumming, set for all. */ 11456 if (all & NETIF_F_HW_CSUM) 11457 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); 11458 11459 return all; 11460 } 11461 EXPORT_SYMBOL(netdev_increment_features); 11462 11463 static struct hlist_head * __net_init netdev_create_hash(void) 11464 { 11465 int i; 11466 struct hlist_head *hash; 11467 11468 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL); 11469 if (hash != NULL) 11470 for (i = 0; i < NETDEV_HASHENTRIES; i++) 11471 INIT_HLIST_HEAD(&hash[i]); 11472 11473 return hash; 11474 } 11475 11476 /* Initialize per network namespace state */ 11477 static int __net_init netdev_init(struct net *net) 11478 { 11479 BUILD_BUG_ON(GRO_HASH_BUCKETS > 11480 8 * sizeof_field(struct napi_struct, gro_bitmask)); 11481 11482 INIT_LIST_HEAD(&net->dev_base_head); 11483 11484 net->dev_name_head = netdev_create_hash(); 11485 if (net->dev_name_head == NULL) 11486 goto err_name; 11487 11488 net->dev_index_head = netdev_create_hash(); 11489 if (net->dev_index_head == NULL) 11490 goto err_idx; 11491 11492 xa_init_flags(&net->dev_by_index, XA_FLAGS_ALLOC1); 11493 11494 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain); 11495 11496 return 0; 11497 11498 err_idx: 11499 kfree(net->dev_name_head); 11500 err_name: 11501 return -ENOMEM; 11502 } 11503 11504 /** 11505 * netdev_drivername - network driver for the device 11506 * @dev: network device 11507 * 11508 * Determine network driver for device. 11509 */ 11510 const char *netdev_drivername(const struct net_device *dev) 11511 { 11512 const struct device_driver *driver; 11513 const struct device *parent; 11514 const char *empty = ""; 11515 11516 parent = dev->dev.parent; 11517 if (!parent) 11518 return empty; 11519 11520 driver = parent->driver; 11521 if (driver && driver->name) 11522 return driver->name; 11523 return empty; 11524 } 11525 11526 static void __netdev_printk(const char *level, const struct net_device *dev, 11527 struct va_format *vaf) 11528 { 11529 if (dev && dev->dev.parent) { 11530 dev_printk_emit(level[1] - '0', 11531 dev->dev.parent, 11532 "%s %s %s%s: %pV", 11533 dev_driver_string(dev->dev.parent), 11534 dev_name(dev->dev.parent), 11535 netdev_name(dev), netdev_reg_state(dev), 11536 vaf); 11537 } else if (dev) { 11538 printk("%s%s%s: %pV", 11539 level, netdev_name(dev), netdev_reg_state(dev), vaf); 11540 } else { 11541 printk("%s(NULL net_device): %pV", level, vaf); 11542 } 11543 } 11544 11545 void netdev_printk(const char *level, const struct net_device *dev, 11546 const char *format, ...) 11547 { 11548 struct va_format vaf; 11549 va_list args; 11550 11551 va_start(args, format); 11552 11553 vaf.fmt = format; 11554 vaf.va = &args; 11555 11556 __netdev_printk(level, dev, &vaf); 11557 11558 va_end(args); 11559 } 11560 EXPORT_SYMBOL(netdev_printk); 11561 11562 #define define_netdev_printk_level(func, level) \ 11563 void func(const struct net_device *dev, const char *fmt, ...) \ 11564 { \ 11565 struct va_format vaf; \ 11566 va_list args; \ 11567 \ 11568 va_start(args, fmt); \ 11569 \ 11570 vaf.fmt = fmt; \ 11571 vaf.va = &args; \ 11572 \ 11573 __netdev_printk(level, dev, &vaf); \ 11574 \ 11575 va_end(args); \ 11576 } \ 11577 EXPORT_SYMBOL(func); 11578 11579 define_netdev_printk_level(netdev_emerg, KERN_EMERG); 11580 define_netdev_printk_level(netdev_alert, KERN_ALERT); 11581 define_netdev_printk_level(netdev_crit, KERN_CRIT); 11582 define_netdev_printk_level(netdev_err, KERN_ERR); 11583 define_netdev_printk_level(netdev_warn, KERN_WARNING); 11584 define_netdev_printk_level(netdev_notice, KERN_NOTICE); 11585 define_netdev_printk_level(netdev_info, KERN_INFO); 11586 11587 static void __net_exit netdev_exit(struct net *net) 11588 { 11589 kfree(net->dev_name_head); 11590 kfree(net->dev_index_head); 11591 xa_destroy(&net->dev_by_index); 11592 if (net != &init_net) 11593 WARN_ON_ONCE(!list_empty(&net->dev_base_head)); 11594 } 11595 11596 static struct pernet_operations __net_initdata netdev_net_ops = { 11597 .init = netdev_init, 11598 .exit = netdev_exit, 11599 }; 11600 11601 static void __net_exit default_device_exit_net(struct net *net) 11602 { 11603 struct netdev_name_node *name_node, *tmp; 11604 struct net_device *dev, *aux; 11605 /* 11606 * Push all migratable network devices back to the 11607 * initial network namespace 11608 */ 11609 ASSERT_RTNL(); 11610 for_each_netdev_safe(net, dev, aux) { 11611 int err; 11612 char fb_name[IFNAMSIZ]; 11613 11614 /* Ignore unmoveable devices (i.e. loopback) */ 11615 if (dev->features & NETIF_F_NETNS_LOCAL) 11616 continue; 11617 11618 /* Leave virtual devices for the generic cleanup */ 11619 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund) 11620 continue; 11621 11622 /* Push remaining network devices to init_net */ 11623 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); 11624 if (netdev_name_in_use(&init_net, fb_name)) 11625 snprintf(fb_name, IFNAMSIZ, "dev%%d"); 11626 11627 netdev_for_each_altname_safe(dev, name_node, tmp) 11628 if (netdev_name_in_use(&init_net, name_node->name)) 11629 __netdev_name_node_alt_destroy(name_node); 11630 11631 err = dev_change_net_namespace(dev, &init_net, fb_name); 11632 if (err) { 11633 pr_emerg("%s: failed to move %s to init_net: %d\n", 11634 __func__, dev->name, err); 11635 BUG(); 11636 } 11637 } 11638 } 11639 11640 static void __net_exit default_device_exit_batch(struct list_head *net_list) 11641 { 11642 /* At exit all network devices most be removed from a network 11643 * namespace. Do this in the reverse order of registration. 11644 * Do this across as many network namespaces as possible to 11645 * improve batching efficiency. 11646 */ 11647 struct net_device *dev; 11648 struct net *net; 11649 LIST_HEAD(dev_kill_list); 11650 11651 rtnl_lock(); 11652 list_for_each_entry(net, net_list, exit_list) { 11653 default_device_exit_net(net); 11654 cond_resched(); 11655 } 11656 11657 list_for_each_entry(net, net_list, exit_list) { 11658 for_each_netdev_reverse(net, dev) { 11659 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) 11660 dev->rtnl_link_ops->dellink(dev, &dev_kill_list); 11661 else 11662 unregister_netdevice_queue(dev, &dev_kill_list); 11663 } 11664 } 11665 unregister_netdevice_many(&dev_kill_list); 11666 rtnl_unlock(); 11667 } 11668 11669 static struct pernet_operations __net_initdata default_device_ops = { 11670 .exit_batch = default_device_exit_batch, 11671 }; 11672 11673 static void __init net_dev_struct_check(void) 11674 { 11675 /* TX read-mostly hotpath */ 11676 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, priv_flags); 11677 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, netdev_ops); 11678 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, header_ops); 11679 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, _tx); 11680 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, real_num_tx_queues); 11681 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_max_size); 11682 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_ipv4_max_size); 11683 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_max_segs); 11684 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_partial_features); 11685 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, num_tc); 11686 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, mtu); 11687 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, needed_headroom); 11688 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, tc_to_txq); 11689 #ifdef CONFIG_XPS 11690 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, xps_maps); 11691 #endif 11692 #ifdef CONFIG_NETFILTER_EGRESS 11693 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, nf_hooks_egress); 11694 #endif 11695 #ifdef CONFIG_NET_XGRESS 11696 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, tcx_egress); 11697 #endif 11698 CACHELINE_ASSERT_GROUP_SIZE(struct net_device, net_device_read_tx, 160); 11699 11700 /* TXRX read-mostly hotpath */ 11701 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, lstats); 11702 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, flags); 11703 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, hard_header_len); 11704 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, features); 11705 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, ip6_ptr); 11706 CACHELINE_ASSERT_GROUP_SIZE(struct net_device, net_device_read_txrx, 38); 11707 11708 /* RX read-mostly hotpath */ 11709 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, ptype_specific); 11710 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, ifindex); 11711 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, real_num_rx_queues); 11712 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, _rx); 11713 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, gro_flush_timeout); 11714 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, napi_defer_hard_irqs); 11715 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, gro_max_size); 11716 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, gro_ipv4_max_size); 11717 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, rx_handler); 11718 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, rx_handler_data); 11719 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, nd_net); 11720 #ifdef CONFIG_NETPOLL 11721 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, npinfo); 11722 #endif 11723 #ifdef CONFIG_NET_XGRESS 11724 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, tcx_ingress); 11725 #endif 11726 CACHELINE_ASSERT_GROUP_SIZE(struct net_device, net_device_read_rx, 104); 11727 } 11728 11729 /* 11730 * Initialize the DEV module. At boot time this walks the device list and 11731 * unhooks any devices that fail to initialise (normally hardware not 11732 * present) and leaves us with a valid list of present and active devices. 11733 * 11734 */ 11735 11736 /* We allocate 256 pages for each CPU if PAGE_SHIFT is 12 */ 11737 #define SYSTEM_PERCPU_PAGE_POOL_SIZE ((1 << 20) / PAGE_SIZE) 11738 11739 static int net_page_pool_create(int cpuid) 11740 { 11741 #if IS_ENABLED(CONFIG_PAGE_POOL) 11742 struct page_pool_params page_pool_params = { 11743 .pool_size = SYSTEM_PERCPU_PAGE_POOL_SIZE, 11744 .flags = PP_FLAG_SYSTEM_POOL, 11745 .nid = NUMA_NO_NODE, 11746 }; 11747 struct page_pool *pp_ptr; 11748 11749 pp_ptr = page_pool_create_percpu(&page_pool_params, cpuid); 11750 if (IS_ERR(pp_ptr)) 11751 return -ENOMEM; 11752 11753 per_cpu(system_page_pool, cpuid) = pp_ptr; 11754 #endif 11755 return 0; 11756 } 11757 11758 /* 11759 * This is called single threaded during boot, so no need 11760 * to take the rtnl semaphore. 11761 */ 11762 static int __init net_dev_init(void) 11763 { 11764 int i, rc = -ENOMEM; 11765 11766 BUG_ON(!dev_boot_phase); 11767 11768 net_dev_struct_check(); 11769 11770 if (dev_proc_init()) 11771 goto out; 11772 11773 if (netdev_kobject_init()) 11774 goto out; 11775 11776 INIT_LIST_HEAD(&ptype_all); 11777 for (i = 0; i < PTYPE_HASH_SIZE; i++) 11778 INIT_LIST_HEAD(&ptype_base[i]); 11779 11780 if (register_pernet_subsys(&netdev_net_ops)) 11781 goto out; 11782 11783 /* 11784 * Initialise the packet receive queues. 11785 */ 11786 11787 for_each_possible_cpu(i) { 11788 struct work_struct *flush = per_cpu_ptr(&flush_works, i); 11789 struct softnet_data *sd = &per_cpu(softnet_data, i); 11790 11791 INIT_WORK(flush, flush_backlog); 11792 11793 skb_queue_head_init(&sd->input_pkt_queue); 11794 skb_queue_head_init(&sd->process_queue); 11795 #ifdef CONFIG_XFRM_OFFLOAD 11796 skb_queue_head_init(&sd->xfrm_backlog); 11797 #endif 11798 INIT_LIST_HEAD(&sd->poll_list); 11799 sd->output_queue_tailp = &sd->output_queue; 11800 #ifdef CONFIG_RPS 11801 INIT_CSD(&sd->csd, rps_trigger_softirq, sd); 11802 sd->cpu = i; 11803 #endif 11804 INIT_CSD(&sd->defer_csd, trigger_rx_softirq, sd); 11805 spin_lock_init(&sd->defer_lock); 11806 11807 init_gro_hash(&sd->backlog); 11808 sd->backlog.poll = process_backlog; 11809 sd->backlog.weight = weight_p; 11810 11811 if (net_page_pool_create(i)) 11812 goto out; 11813 } 11814 11815 dev_boot_phase = 0; 11816 11817 /* The loopback device is special if any other network devices 11818 * is present in a network namespace the loopback device must 11819 * be present. Since we now dynamically allocate and free the 11820 * loopback device ensure this invariant is maintained by 11821 * keeping the loopback device as the first device on the 11822 * list of network devices. Ensuring the loopback devices 11823 * is the first device that appears and the last network device 11824 * that disappears. 11825 */ 11826 if (register_pernet_device(&loopback_net_ops)) 11827 goto out; 11828 11829 if (register_pernet_device(&default_device_ops)) 11830 goto out; 11831 11832 open_softirq(NET_TX_SOFTIRQ, net_tx_action); 11833 open_softirq(NET_RX_SOFTIRQ, net_rx_action); 11834 11835 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", 11836 NULL, dev_cpu_dead); 11837 WARN_ON(rc < 0); 11838 rc = 0; 11839 out: 11840 if (rc < 0) { 11841 for_each_possible_cpu(i) { 11842 struct page_pool *pp_ptr; 11843 11844 pp_ptr = per_cpu(system_page_pool, i); 11845 if (!pp_ptr) 11846 continue; 11847 11848 page_pool_destroy(pp_ptr); 11849 per_cpu(system_page_pool, i) = NULL; 11850 } 11851 } 11852 11853 return rc; 11854 } 11855 11856 subsys_initcall(net_dev_init); 11857