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