net/core/dev.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  *      NET3    Protocol independent device support routines.
   4  *
   5  *      Derived from the non IP parts of dev.c 1.0.19
   6  *              Authors:        Ross Biro
   7  *                              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
   8  *                              Mark Evans, <evansmp@uhura.aston.ac.uk>
   9  *
  10  *      Additional Authors:
  11  *              Florian la Roche <rzsfl@rz.uni-sb.de>
  12  *              Alan Cox <gw4pts@gw4pts.ampr.org>
  13  *              David Hinds <dahinds@users.sourceforge.net>
  14  *              Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
  15  *              Adam Sulmicki <adam@cfar.umd.edu>
  16  *              Pekka Riikonen <priikone@poesidon.pspt.fi>
  17  *
  18  *      Changes:
  19  *              D.J. Barrow     :       Fixed bug where dev->refcnt gets set
  20  *                                      to 2 if register_netdev gets called
  21  *                                      before net_dev_init & also removed a
  22  *                                      few lines of code in the process.
  23  *              Alan Cox        :       device private ioctl copies fields back.
  24  *              Alan Cox        :       Transmit queue code does relevant
  25  *                                      stunts to keep the queue safe.
  26  *              Alan Cox        :       Fixed double lock.
  27  *              Alan Cox        :       Fixed promisc NULL pointer trap
  28  *              ????????        :       Support the full private ioctl range
  29  *              Alan Cox        :       Moved ioctl permission check into
  30  *                                      drivers
  31  *              Tim Kordas      :       SIOCADDMULTI/SIOCDELMULTI
  32  *              Alan Cox        :       100 backlog just doesn't cut it when
  33  *                                      you start doing multicast video 8)
  34  *              Alan Cox        :       Rewrote net_bh and list manager.
  35  *              Alan Cox        :       Fix ETH_P_ALL echoback lengths.
  36  *              Alan Cox        :       Took out transmit every packet pass
  37  *                                      Saved a few bytes in the ioctl handler
  38  *              Alan Cox        :       Network driver sets packet type before
  39  *                                      calling netif_rx. Saves a function
  40  *                                      call a packet.
  41  *              Alan Cox        :       Hashed net_bh()
  42  *              Richard Kooijman:       Timestamp fixes.
  43  *              Alan Cox        :       Wrong field in SIOCGIFDSTADDR
  44  *              Alan Cox        :       Device lock protection.
  45  *              Alan Cox        :       Fixed nasty side effect of device close
  46  *                                      changes.
  47  *              Rudi Cilibrasi  :       Pass the right thing to
  48  *                                      set_mac_address()
  49  *              Dave Miller     :       32bit quantity for the device lock to
  50  *                                      make it work out on a Sparc.
  51  *              Bjorn Ekwall    :       Added KERNELD hack.
  52  *              Alan Cox        :       Cleaned up the backlog initialise.
  53  *              Craig Metz      :       SIOCGIFCONF fix if space for under
  54  *                                      1 device.
  55  *          Thomas Bogendoerfer :       Return ENODEV for dev_open, if there
  56  *                                      is no device open function.
  57  *              Andi Kleen      :       Fix error reporting for SIOCGIFCONF
  58  *          Michael Chastain    :       Fix signed/unsigned for SIOCGIFCONF
  59  *              Cyrus Durgin    :       Cleaned for KMOD
  60  *              Adam Sulmicki   :       Bug Fix : Network Device Unload
  61  *                                      A network device unload needs to purge
  62  *                                      the backlog queue.
  63  *      Paul Rusty Russell      :       SIOCSIFNAME
  64  *              Pekka Riikonen  :       Netdev boot-time settings code
  65  *              Andrew Morton   :       Make unregister_netdevice wait
  66  *                                      indefinitely on dev->refcnt
  67  *              J Hadi Salim    :       - Backlog queue sampling
  68  *                                      - netif_rx() feedback
  69  */
  70
  71 #include <linux/uaccess.h>
  72 #include <linux/bitops.h>
  73 #include <linux/capability.h>
  74 #include <linux/cpu.h>
  75 #include <linux/types.h>
  76 #include <linux/kernel.h>
  77 #include <linux/hash.h>
  78 #include <linux/slab.h>
  79 #include <linux/sched.h>
  80 #include <linux/sched/mm.h>
  81 #include <linux/mutex.h>
  82 #include <linux/rwsem.h>
  83 #include <linux/string.h>
  84 #include <linux/mm.h>
  85 #include <linux/socket.h>
  86 #include <linux/sockios.h>
  87 #include <linux/errno.h>
  88 #include <linux/interrupt.h>
  89 #include <linux/if_ether.h>
  90 #include <linux/netdevice.h>
  91 #include <linux/etherdevice.h>
  92 #include <linux/ethtool.h>
  93 #include <linux/skbuff.h>
  94 #include <linux/kthread.h>
  95 #include <linux/bpf.h>
  96 #include <linux/bpf_trace.h>
  97 #include <net/net_namespace.h>
  98 #include <net/sock.h>
  99 #include <net/busy_poll.h>
 100 #include <linux/rtnetlink.h>
 101 #include <linux/stat.h>
 102 #include <net/dsa.h>
 103 #include <net/dst.h>
 104 #include <net/dst_metadata.h>
 105 #include <net/gro.h>
 106 #include <net/pkt_sched.h>
 107 #include <net/pkt_cls.h>
 108 #include <net/checksum.h>
 109 #include <net/xfrm.h>
 110 #include <linux/highmem.h>
 111 #include <linux/init.h>
 112 #include <linux/module.h>
 113 #include <linux/netpoll.h>
 114 #include <linux/rcupdate.h>
 115 #include <linux/delay.h>
 116 #include <net/iw_handler.h>
 117 #include <asm/current.h>
 118 #include <linux/audit.h>
 119 #include <linux/dmaengine.h>
 120 #include <linux/err.h>
 121 #include <linux/ctype.h>
 122 #include <linux/if_arp.h>
 123 #include <linux/if_vlan.h>
 124 #include <linux/ip.h>
 125 #include <net/ip.h>
 126 #include <net/mpls.h>
 127 #include <linux/ipv6.h>
 128 #include <linux/in.h>
 129 #include <linux/jhash.h>
 130 #include <linux/random.h>
 131 #include <trace/events/napi.h>
 132 #include <trace/events/net.h>
 133 #include <trace/events/skb.h>
 134 #include <trace/events/qdisc.h>
 135 #include <linux/inetdevice.h>
 136 #include <linux/cpu_rmap.h>
 137 #include <linux/static_key.h>
 138 #include <linux/hashtable.h>
 139 #include <linux/vmalloc.h>
 140 #include <linux/if_macvlan.h>
 141 #include <linux/errqueue.h>
 142 #include <linux/hrtimer.h>
 143 #include <linux/netfilter_netdev.h>
 144 #include <linux/crash_dump.h>
 145 #include <linux/sctp.h>
 146 #include <net/udp_tunnel.h>
 147 #include <linux/net_namespace.h>
 148 #include <linux/indirect_call_wrapper.h>
 149 #include <net/devlink.h>
 150 #include <linux/pm_runtime.h>
 151 #include <linux/prandom.h>
 152 #include <linux/once_lite.h>
 153
 154 #include "dev.h"
 155 #include "net-sysfs.h"
 156
 157
 158 static DEFINE_SPINLOCK(ptype_lock);
 159 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
 160 struct list_head ptype_all __read_mostly;       /* Taps */
 161
 162 static int netif_rx_internal(struct sk_buff *skb);
 163 static int call_netdevice_notifiers_info(unsigned long val,
 164                                          struct netdev_notifier_info *info);
 165 static int call_netdevice_notifiers_extack(unsigned long val,
 166                                            struct net_device *dev,
 167                                            struct netlink_ext_ack *extack);
 168 static struct napi_struct *napi_by_id(unsigned int napi_id);
 169
 170 /*
 171  * The @dev_base_head list is protected by @dev_base_lock and the rtnl
 172  * semaphore.
 173  *
 174  * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
 175  *
 176  * Writers must hold the rtnl semaphore while they loop through the
 177  * dev_base_head list, and hold dev_base_lock for writing when they do the
 178  * actual updates.  This allows pure readers to access the list even
 179  * while a writer is preparing to update it.
 180  *
 181  * To put it another way, dev_base_lock is held for writing only to
 182  * protect against pure readers; the rtnl semaphore provides the
 183  * protection against other writers.
 184  *
 185  * See, for example usages, register_netdevice() and
 186  * unregister_netdevice(), which must be called with the rtnl
 187  * semaphore held.
 188  */
 189 DEFINE_RWLOCK(dev_base_lock);
 190 EXPORT_SYMBOL(dev_base_lock);
 191
 192 static DEFINE_MUTEX(ifalias_mutex);
 193
 194 /* protects napi_hash addition/deletion and napi_gen_id */
 195 static DEFINE_SPINLOCK(napi_hash_lock);
 196
 197 static unsigned int napi_gen_id = NR_CPUS;
 198 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8);
 199
 200 static DECLARE_RWSEM(devnet_rename_sem);
 201
 202 static inline void dev_base_seq_inc(struct net *net)
 203 {
 204         while (++net->dev_base_seq == 0)
 205                 ;
 206 }
 207
 208 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
 209 {
 210         unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ));
 211
 212         return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
 213 }
 214
 215 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
 216 {
 217         return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
 218 }
 219
 220 static inline void rps_lock_irqsave(struct softnet_data *sd,
 221                                     unsigned long *flags)
 222 {
 223         if (IS_ENABLED(CONFIG_RPS))
 224                 spin_lock_irqsave(&sd->input_pkt_queue.lock, *flags);
 225         else if (!IS_ENABLED(CONFIG_PREEMPT_RT))
 226                 local_irq_save(*flags);
 227 }
 228
 229 static inline void rps_lock_irq_disable(struct softnet_data *sd)
 230 {
 231         if (IS_ENABLED(CONFIG_RPS))
 232                 spin_lock_irq(&sd->input_pkt_queue.lock);
 233         else if (!IS_ENABLED(CONFIG_PREEMPT_RT))
 234                 local_irq_disable();
 235 }
 236
 237 static inline void rps_unlock_irq_restore(struct softnet_data *sd,
 238                                           unsigned long *flags)
 239 {
 240         if (IS_ENABLED(CONFIG_RPS))
 241                 spin_unlock_irqrestore(&sd->input_pkt_queue.lock, *flags);
 242         else if (!IS_ENABLED(CONFIG_PREEMPT_RT))
 243                 local_irq_restore(*flags);
 244 }
 245
 246 static inline void rps_unlock_irq_enable(struct softnet_data *sd)
 247 {
 248         if (IS_ENABLED(CONFIG_RPS))
 249                 spin_unlock_irq(&sd->input_pkt_queue.lock);
 250         else if (!IS_ENABLED(CONFIG_PREEMPT_RT))
 251                 local_irq_enable();
 252 }
 253
 254 static struct netdev_name_node *netdev_name_node_alloc(struct net_device *dev,
 255                                                        const char *name)
 256 {
 257         struct netdev_name_node *name_node;
 258
 259         name_node = kmalloc(sizeof(*name_node), GFP_KERNEL);
 260         if (!name_node)
 261                 return NULL;
 262         INIT_HLIST_NODE(&name_node->hlist);
 263         name_node->dev = dev;
 264         name_node->name = name;
 265         return name_node;
 266 }
 267
 268 static struct netdev_name_node *
 269 netdev_name_node_head_alloc(struct net_device *dev)
 270 {
 271         struct netdev_name_node *name_node;
 272
 273         name_node = netdev_name_node_alloc(dev, dev->name);
 274         if (!name_node)
 275                 return NULL;
 276         INIT_LIST_HEAD(&name_node->list);
 277         return name_node;
 278 }
 279
 280 static void netdev_name_node_free(struct netdev_name_node *name_node)
 281 {
 282         kfree(name_node);
 283 }
 284
 285 static void netdev_name_node_add(struct net *net,
 286                                  struct netdev_name_node *name_node)
 287 {
 288         hlist_add_head_rcu(&name_node->hlist,
 289                            dev_name_hash(net, name_node->name));
 290 }
 291
 292 static void netdev_name_node_del(struct netdev_name_node *name_node)
 293 {
 294         hlist_del_rcu(&name_node->hlist);
 295 }
 296
 297 static struct netdev_name_node *netdev_name_node_lookup(struct net *net,
 298                                                         const char *name)
 299 {
 300         struct hlist_head *head = dev_name_hash(net, name);
 301         struct netdev_name_node *name_node;
 302
 303         hlist_for_each_entry(name_node, head, hlist)
 304                 if (!strcmp(name_node->name, name))
 305                         return name_node;
 306         return NULL;
 307 }
 308
 309 static struct netdev_name_node *netdev_name_node_lookup_rcu(struct net *net,
 310                                                             const char *name)
 311 {
 312         struct hlist_head *head = dev_name_hash(net, name);
 313         struct netdev_name_node *name_node;
 314
 315         hlist_for_each_entry_rcu(name_node, head, hlist)
 316                 if (!strcmp(name_node->name, name))
 317                         return name_node;
 318         return NULL;
 319 }
 320
 321 bool netdev_name_in_use(struct net *net, const char *name)
 322 {
 323         return netdev_name_node_lookup(net, name);
 324 }
 325 EXPORT_SYMBOL(netdev_name_in_use);
 326
 327 int netdev_name_node_alt_create(struct net_device *dev, const char *name)
 328 {
 329         struct netdev_name_node *name_node;
 330         struct net *net = dev_net(dev);
 331
 332         name_node = netdev_name_node_lookup(net, name);
 333         if (name_node)
 334                 return -EEXIST;
 335         name_node = netdev_name_node_alloc(dev, name);
 336         if (!name_node)
 337                 return -ENOMEM;
 338         netdev_name_node_add(net, name_node);
 339         /* The node that holds dev->name acts as a head of per-device list. */
 340         list_add_tail(&name_node->list, &dev->name_node->list);
 341
 342         return 0;
 343 }
 344
 345 static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node)
 346 {
 347         list_del(&name_node->list);
 348         netdev_name_node_del(name_node);
 349         kfree(name_node->name);
 350         netdev_name_node_free(name_node);
 351 }
 352
 353 int netdev_name_node_alt_destroy(struct net_device *dev, const char *name)
 354 {
 355         struct netdev_name_node *name_node;
 356         struct net *net = dev_net(dev);
 357
 358         name_node = netdev_name_node_lookup(net, name);
 359         if (!name_node)
 360                 return -ENOENT;
 361         /* lookup might have found our primary name or a name belonging
 362          * to another device.
 363          */
 364         if (name_node == dev->name_node || name_node->dev != dev)
 365                 return -EINVAL;
 366
 367         __netdev_name_node_alt_destroy(name_node);
 368
 369         return 0;
 370 }
 371
 372 static void netdev_name_node_alt_flush(struct net_device *dev)
 373 {
 374         struct netdev_name_node *name_node, *tmp;
 375
 376         list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list)
 377                 __netdev_name_node_alt_destroy(name_node);
 378 }
 379
 380 /* Device list insertion */
 381 static void list_netdevice(struct net_device *dev)
 382 {
 383         struct net *net = dev_net(dev);
 384
 385         ASSERT_RTNL();
 386
 387         write_lock(&dev_base_lock);
 388         list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
 389         netdev_name_node_add(net, dev->name_node);
 390         hlist_add_head_rcu(&dev->index_hlist,
 391                            dev_index_hash(net, dev->ifindex));
 392         write_unlock(&dev_base_lock);
 393
 394         dev_base_seq_inc(net);
 395 }
 396
 397 /* Device list removal
 398  * caller must respect a RCU grace period before freeing/reusing dev
 399  */
 400 static void unlist_netdevice(struct net_device *dev, bool lock)
 401 {
 402         ASSERT_RTNL();
 403
 404         /* Unlink dev from the device chain */
 405         if (lock)
 406                 write_lock(&dev_base_lock);
 407         list_del_rcu(&dev->dev_list);
 408         netdev_name_node_del(dev->name_node);
 409         hlist_del_rcu(&dev->index_hlist);
 410         if (lock)
 411                 write_unlock(&dev_base_lock);
 412
 413         dev_base_seq_inc(dev_net(dev));
 414 }
 415
 416 /*
 417  *      Our notifier list
 418  */
 419
 420 static RAW_NOTIFIER_HEAD(netdev_chain);
 421
 422 /*
 423  *      Device drivers call our routines to queue packets here. We empty the
 424  *      queue in the local softnet handler.
 425  */
 426
 427 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
 428 EXPORT_PER_CPU_SYMBOL(softnet_data);
 429
 430 #ifdef CONFIG_LOCKDEP
 431 /*
 432  * register_netdevice() inits txq->_xmit_lock and sets lockdep class
 433  * according to dev->type
 434  */
 435 static const unsigned short netdev_lock_type[] = {
 436          ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
 437          ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
 438          ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
 439          ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
 440          ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
 441          ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
 442          ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
 443          ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
 444          ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
 445          ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
 446          ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
 447          ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
 448          ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
 449          ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
 450          ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
 451
 452 static const char *const netdev_lock_name[] = {
 453         "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
 454         "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
 455         "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
 456         "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
 457         "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
 458         "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
 459         "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
 460         "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
 461         "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
 462         "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
 463         "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
 464         "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
 465         "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
 466         "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
 467         "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
 468
 469 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
 470 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
 471
 472 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
 473 {
 474         int i;
 475
 476         for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
 477                 if (netdev_lock_type[i] == dev_type)
 478                         return i;
 479         /* the last key is used by default */
 480         return ARRAY_SIZE(netdev_lock_type) - 1;
 481 }
 482
 483 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
 484                                                  unsigned short dev_type)
 485 {
 486         int i;
 487
 488         i = netdev_lock_pos(dev_type);
 489         lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
 490                                    netdev_lock_name[i]);
 491 }
 492
 493 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
 494 {
 495         int i;
 496
 497         i = netdev_lock_pos(dev->type);
 498         lockdep_set_class_and_name(&dev->addr_list_lock,
 499                                    &netdev_addr_lock_key[i],
 500                                    netdev_lock_name[i]);
 501 }
 502 #else
 503 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
 504                                                  unsigned short dev_type)
 505 {
 506 }
 507
 508 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
 509 {
 510 }
 511 #endif
 512
 513 /*******************************************************************************
 514  *
 515  *              Protocol management and registration routines
 516  *
 517  *******************************************************************************/
 518
 519
 520 /*
 521  *      Add a protocol ID to the list. Now that the input handler is
 522  *      smarter we can dispense with all the messy stuff that used to be
 523  *      here.
 524  *
 525  *      BEWARE!!! Protocol handlers, mangling input packets,
 526  *      MUST BE last in hash buckets and checking protocol handlers
 527  *      MUST start from promiscuous ptype_all chain in net_bh.
 528  *      It is true now, do not change it.
 529  *      Explanation follows: if protocol handler, mangling packet, will
 530  *      be the first on list, it is not able to sense, that packet
 531  *      is cloned and should be copied-on-write, so that it will
 532  *      change it and subsequent readers will get broken packet.
 533  *                                                      --ANK (980803)
 534  */
 535
 536 static inline struct list_head *ptype_head(const struct packet_type *pt)
 537 {
 538         if (pt->type == htons(ETH_P_ALL))
 539                 return pt->dev ? &pt->dev->ptype_all : &ptype_all;
 540         else
 541                 return pt->dev ? &pt->dev->ptype_specific :
 542                                  &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
 543 }
 544
 545 /**
 546  *      dev_add_pack - add packet handler
 547  *      @pt: packet type declaration
 548  *
 549  *      Add a protocol handler to the networking stack. The passed &packet_type
 550  *      is linked into kernel lists and may not be freed until it has been
 551  *      removed from the kernel lists.
 552  *
 553  *      This call does not sleep therefore it can not
 554  *      guarantee all CPU's that are in middle of receiving packets
 555  *      will see the new packet type (until the next received packet).
 556  */
 557
 558 void dev_add_pack(struct packet_type *pt)
 559 {
 560         struct list_head *head = ptype_head(pt);
 561
 562         spin_lock(&ptype_lock);
 563         list_add_rcu(&pt->list, head);
 564         spin_unlock(&ptype_lock);
 565 }
 566 EXPORT_SYMBOL(dev_add_pack);
 567
 568 /**
 569  *      __dev_remove_pack        - remove packet handler
 570  *      @pt: packet type declaration
 571  *
 572  *      Remove a protocol handler that was previously added to the kernel
 573  *      protocol handlers by dev_add_pack(). The passed &packet_type is removed
 574  *      from the kernel lists and can be freed or reused once this function
 575  *      returns.
 576  *
 577  *      The packet type might still be in use by receivers
 578  *      and must not be freed until after all the CPU's have gone
 579  *      through a quiescent state.
 580  */
 581 void __dev_remove_pack(struct packet_type *pt)
 582 {
 583         struct list_head *head = ptype_head(pt);
 584         struct packet_type *pt1;
 585
 586         spin_lock(&ptype_lock);
 587
 588         list_for_each_entry(pt1, head, list) {
 589                 if (pt == pt1) {
 590                         list_del_rcu(&pt->list);
 591                         goto out;
 592                 }
 593         }
 594
 595         pr_warn("dev_remove_pack: %p not found\n", pt);
 596 out:
 597         spin_unlock(&ptype_lock);
 598 }
 599 EXPORT_SYMBOL(__dev_remove_pack);
 600
 601 /**
 602  *      dev_remove_pack  - remove packet handler
 603  *      @pt: packet type declaration
 604  *
 605  *      Remove a protocol handler that was previously added to the kernel
 606  *      protocol handlers by dev_add_pack(). The passed &packet_type is removed
 607  *      from the kernel lists and can be freed or reused once this function
 608  *      returns.
 609  *
 610  *      This call sleeps to guarantee that no CPU is looking at the packet
 611  *      type after return.
 612  */
 613 void dev_remove_pack(struct packet_type *pt)
 614 {
 615         __dev_remove_pack(pt);
 616
 617         synchronize_net();
 618 }
 619 EXPORT_SYMBOL(dev_remove_pack);
 620
 621
 622 /*******************************************************************************
 623  *
 624  *                          Device Interface Subroutines
 625  *
 626  *******************************************************************************/
 627
 628 /**
 629  *      dev_get_iflink  - get 'iflink' value of a interface
 630  *      @dev: targeted interface
 631  *
 632  *      Indicates the ifindex the interface is linked to.
 633  *      Physical interfaces have the same 'ifindex' and 'iflink' values.
 634  */
 635
 636 int dev_get_iflink(const struct net_device *dev)
 637 {
 638         if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink)
 639                 return dev->netdev_ops->ndo_get_iflink(dev);
 640
 641         return dev->ifindex;
 642 }
 643 EXPORT_SYMBOL(dev_get_iflink);
 644
 645 /**
 646  *      dev_fill_metadata_dst - Retrieve tunnel egress information.
 647  *      @dev: targeted interface
 648  *      @skb: The packet.
 649  *
 650  *      For better visibility of tunnel traffic OVS needs to retrieve
 651  *      egress tunnel information for a packet. Following API allows
 652  *      user to get this info.
 653  */
 654 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
 655 {
 656         struct ip_tunnel_info *info;
 657
 658         if (!dev->netdev_ops  || !dev->netdev_ops->ndo_fill_metadata_dst)
 659                 return -EINVAL;
 660
 661         info = skb_tunnel_info_unclone(skb);
 662         if (!info)
 663                 return -ENOMEM;
 664         if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX)))
 665                 return -EINVAL;
 666
 667         return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb);
 668 }
 669 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst);
 670
 671 static struct net_device_path *dev_fwd_path(struct net_device_path_stack *stack)
 672 {
 673         int k = stack->num_paths++;
 674
 675         if (WARN_ON_ONCE(k >= NET_DEVICE_PATH_STACK_MAX))
 676                 return NULL;
 677
 678         return &stack->path[k];
 679 }
 680
 681 int dev_fill_forward_path(const struct net_device *dev, const u8 *daddr,
 682                           struct net_device_path_stack *stack)
 683 {
 684         const struct net_device *last_dev;
 685         struct net_device_path_ctx ctx = {
 686                 .dev    = dev,
 687         };
 688         struct net_device_path *path;
 689         int ret = 0;
 690
 691         memcpy(ctx.daddr, daddr, sizeof(ctx.daddr));
 692         stack->num_paths = 0;
 693         while (ctx.dev && ctx.dev->netdev_ops->ndo_fill_forward_path) {
 694                 last_dev = ctx.dev;
 695                 path = dev_fwd_path(stack);
 696                 if (!path)
 697                         return -1;
 698
 699                 memset(path, 0, sizeof(struct net_device_path));
 700                 ret = ctx.dev->netdev_ops->ndo_fill_forward_path(&ctx, path);
 701                 if (ret < 0)
 702                         return -1;
 703
 704                 if (WARN_ON_ONCE(last_dev == ctx.dev))
 705                         return -1;
 706         }
 707
 708         if (!ctx.dev)
 709                 return ret;
 710
 711         path = dev_fwd_path(stack);
 712         if (!path)
 713                 return -1;
 714         path->type = DEV_PATH_ETHERNET;
 715         path->dev = ctx.dev;
 716
 717         return ret;
 718 }
 719 EXPORT_SYMBOL_GPL(dev_fill_forward_path);
 720
 721 /**
 722  *      __dev_get_by_name       - find a device by its name
 723  *      @net: the applicable net namespace
 724  *      @name: name to find
 725  *
 726  *      Find an interface by name. Must be called under RTNL semaphore
 727  *      or @dev_base_lock. If the name is found a pointer to the device
 728  *      is returned. If the name is not found then %NULL is returned. The
 729  *      reference counters are not incremented so the caller must be
 730  *      careful with locks.
 731  */
 732
 733 struct net_device *__dev_get_by_name(struct net *net, const char *name)
 734 {
 735         struct netdev_name_node *node_name;
 736
 737         node_name = netdev_name_node_lookup(net, name);
 738         return node_name ? node_name->dev : NULL;
 739 }
 740 EXPORT_SYMBOL(__dev_get_by_name);
 741
 742 /**
 743  * dev_get_by_name_rcu  - find a device by its name
 744  * @net: the applicable net namespace
 745  * @name: name to find
 746  *
 747  * Find an interface by name.
 748  * If the name is found a pointer to the device is returned.
 749  * If the name is not found then %NULL is returned.
 750  * The reference counters are not incremented so the caller must be
 751  * careful with locks. The caller must hold RCU lock.
 752  */
 753
 754 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
 755 {
 756         struct netdev_name_node *node_name;
 757
 758         node_name = netdev_name_node_lookup_rcu(net, name);
 759         return node_name ? node_name->dev : NULL;
 760 }
 761 EXPORT_SYMBOL(dev_get_by_name_rcu);
 762
 763 /**
 764  *      dev_get_by_name         - find a device by its name
 765  *      @net: the applicable net namespace
 766  *      @name: name to find
 767  *
 768  *      Find an interface by name. This can be called from any
 769  *      context and does its own locking. The returned handle has
 770  *      the usage count incremented and the caller must use dev_put() to
 771  *      release it when it is no longer needed. %NULL is returned if no
 772  *      matching device is found.
 773  */
 774
 775 struct net_device *dev_get_by_name(struct net *net, const char *name)
 776 {
 777         struct net_device *dev;
 778
 779         rcu_read_lock();
 780         dev = dev_get_by_name_rcu(net, name);
 781         dev_hold(dev);
 782         rcu_read_unlock();
 783         return dev;
 784 }
 785 EXPORT_SYMBOL(dev_get_by_name);
 786
 787 /**
 788  *      __dev_get_by_index - find a device by its ifindex
 789  *      @net: the applicable net namespace
 790  *      @ifindex: index of device
 791  *
 792  *      Search for an interface by index. Returns %NULL if the device
 793  *      is not found or a pointer to the device. The device has not
 794  *      had its reference counter increased so the caller must be careful
 795  *      about locking. The caller must hold either the RTNL semaphore
 796  *      or @dev_base_lock.
 797  */
 798
 799 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
 800 {
 801         struct net_device *dev;
 802         struct hlist_head *head = dev_index_hash(net, ifindex);
 803
 804         hlist_for_each_entry(dev, head, index_hlist)
 805                 if (dev->ifindex == ifindex)
 806                         return dev;
 807
 808         return NULL;
 809 }
 810 EXPORT_SYMBOL(__dev_get_by_index);
 811
 812 /**
 813  *      dev_get_by_index_rcu - find a device by its ifindex
 814  *      @net: the applicable net namespace
 815  *      @ifindex: index of device
 816  *
 817  *      Search for an interface by index. Returns %NULL if the device
 818  *      is not found or a pointer to the device. The device has not
 819  *      had its reference counter increased so the caller must be careful
 820  *      about locking. The caller must hold RCU lock.
 821  */
 822
 823 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
 824 {
 825         struct net_device *dev;
 826         struct hlist_head *head = dev_index_hash(net, ifindex);
 827
 828         hlist_for_each_entry_rcu(dev, head, index_hlist)
 829                 if (dev->ifindex == ifindex)
 830                         return dev;
 831
 832         return NULL;
 833 }
 834 EXPORT_SYMBOL(dev_get_by_index_rcu);
 835
 836
 837 /**
 838  *      dev_get_by_index - find a device by its ifindex
 839  *      @net: the applicable net namespace
 840  *      @ifindex: index of device
 841  *
 842  *      Search for an interface by index. Returns NULL if the device
 843  *      is not found or a pointer to the device. The device returned has
 844  *      had a reference added and the pointer is safe until the user calls
 845  *      dev_put to indicate they have finished with it.
 846  */
 847
 848 struct net_device *dev_get_by_index(struct net *net, int ifindex)
 849 {
 850         struct net_device *dev;
 851
 852         rcu_read_lock();
 853         dev = dev_get_by_index_rcu(net, ifindex);
 854         dev_hold(dev);
 855         rcu_read_unlock();
 856         return dev;
 857 }
 858 EXPORT_SYMBOL(dev_get_by_index);
 859
 860 /**
 861  *      dev_get_by_napi_id - find a device by napi_id
 862  *      @napi_id: ID of the NAPI struct
 863  *
 864  *      Search for an interface by NAPI ID. Returns %NULL if the device
 865  *      is not found or a pointer to the device. The device has not had
 866  *      its reference counter increased so the caller must be careful
 867  *      about locking. The caller must hold RCU lock.
 868  */
 869
 870 struct net_device *dev_get_by_napi_id(unsigned int napi_id)
 871 {
 872         struct napi_struct *napi;
 873
 874         WARN_ON_ONCE(!rcu_read_lock_held());
 875
 876         if (napi_id < MIN_NAPI_ID)
 877                 return NULL;
 878
 879         napi = napi_by_id(napi_id);
 880
 881         return napi ? napi->dev : NULL;
 882 }
 883 EXPORT_SYMBOL(dev_get_by_napi_id);
 884
 885 /**
 886  *      netdev_get_name - get a netdevice name, knowing its ifindex.
 887  *      @net: network namespace
 888  *      @name: a pointer to the buffer where the name will be stored.
 889  *      @ifindex: the ifindex of the interface to get the name from.
 890  */
 891 int netdev_get_name(struct net *net, char *name, int ifindex)
 892 {
 893         struct net_device *dev;
 894         int ret;
 895
 896         down_read(&devnet_rename_sem);
 897         rcu_read_lock();
 898
 899         dev = dev_get_by_index_rcu(net, ifindex);
 900         if (!dev) {
 901                 ret = -ENODEV;
 902                 goto out;
 903         }
 904
 905         strcpy(name, dev->name);
 906
 907         ret = 0;
 908 out:
 909         rcu_read_unlock();
 910         up_read(&devnet_rename_sem);
 911         return ret;
 912 }
 913
 914 /**
 915  *      dev_getbyhwaddr_rcu - find a device by its hardware address
 916  *      @net: the applicable net namespace
 917  *      @type: media type of device
 918  *      @ha: hardware address
 919  *
 920  *      Search for an interface by MAC address. Returns NULL if the device
 921  *      is not found or a pointer to the device.
 922  *      The caller must hold RCU or RTNL.
 923  *      The returned device has not had its ref count increased
 924  *      and the caller must therefore be careful about locking
 925  *
 926  */
 927
 928 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
 929                                        const char *ha)
 930 {
 931         struct net_device *dev;
 932
 933         for_each_netdev_rcu(net, dev)
 934                 if (dev->type == type &&
 935                     !memcmp(dev->dev_addr, ha, dev->addr_len))
 936                         return dev;
 937
 938         return NULL;
 939 }
 940 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
 941
 942 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
 943 {
 944         struct net_device *dev, *ret = NULL;
 945
 946         rcu_read_lock();
 947         for_each_netdev_rcu(net, dev)
 948                 if (dev->type == type) {
 949                         dev_hold(dev);
 950                         ret = dev;
 951                         break;
 952                 }
 953         rcu_read_unlock();
 954         return ret;
 955 }
 956 EXPORT_SYMBOL(dev_getfirstbyhwtype);
 957
 958 /**
 959  *      __dev_get_by_flags - find any device with given flags
 960  *      @net: the applicable net namespace
 961  *      @if_flags: IFF_* values
 962  *      @mask: bitmask of bits in if_flags to check
 963  *
 964  *      Search for any interface with the given flags. Returns NULL if a device
 965  *      is not found or a pointer to the device. Must be called inside
 966  *      rtnl_lock(), and result refcount is unchanged.
 967  */
 968
 969 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
 970                                       unsigned short mask)
 971 {
 972         struct net_device *dev, *ret;
 973
 974         ASSERT_RTNL();
 975
 976         ret = NULL;
 977         for_each_netdev(net, dev) {
 978                 if (((dev->flags ^ if_flags) & mask) == 0) {
 979                         ret = dev;
 980                         break;
 981                 }
 982         }
 983         return ret;
 984 }
 985 EXPORT_SYMBOL(__dev_get_by_flags);
 986
 987 /**
 988  *      dev_valid_name - check if name is okay for network device
 989  *      @name: name string
 990  *
 991  *      Network device names need to be valid file names to
 992  *      allow sysfs to work.  We also disallow any kind of
 993  *      whitespace.
 994  */
 995 bool dev_valid_name(const char *name)
 996 {
 997         if (*name == '\0')
 998                 return false;
 999         if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
1000                 return false;
1001         if (!strcmp(name, ".") || !strcmp(name, ".."))
1002                 return false;
1003
1004         while (*name) {
1005                 if (*name == '/' || *name == ':' || isspace(*name))
1006                         return false;
1007                 name++;
1008         }
1009         return true;
1010 }
1011 EXPORT_SYMBOL(dev_valid_name);
1012
1013 /**
1014  *      __dev_alloc_name - allocate a name for a device
1015  *      @net: network namespace to allocate the device name in
1016  *      @name: name format string
1017  *      @buf:  scratch buffer and result name string
1018  *
1019  *      Passed a format string - eg "lt%d" it will try and find a suitable
1020  *      id. It scans list of devices to build up a free map, then chooses
1021  *      the first empty slot. The caller must hold the dev_base or rtnl lock
1022  *      while allocating the name and adding the device in order to avoid
1023  *      duplicates.
1024  *      Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1025  *      Returns the number of the unit assigned or a negative errno code.
1026  */
1027
1028 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
1029 {
1030         int i = 0;
1031         const char *p;
1032         const int max_netdevices = 8*PAGE_SIZE;
1033         unsigned long *inuse;
1034         struct net_device *d;
1035
1036         if (!dev_valid_name(name))
1037                 return -EINVAL;
1038
1039         p = strchr(name, '%');
1040         if (p) {
1041                 /*
1042                  * Verify the string as this thing may have come from
1043                  * the user.  There must be either one "%d" and no other "%"
1044                  * characters.
1045                  */
1046                 if (p[1] != 'd' || strchr(p + 2, '%'))
1047                         return -EINVAL;
1048
1049                 /* Use one page as a bit array of possible slots */
1050                 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
1051                 if (!inuse)
1052                         return -ENOMEM;
1053
1054                 for_each_netdev(net, d) {
1055                         struct netdev_name_node *name_node;
1056                         list_for_each_entry(name_node, &d->name_node->list, list) {
1057                                 if (!sscanf(name_node->name, name, &i))
1058                                         continue;
1059                                 if (i < 0 || i >= max_netdevices)
1060                                         continue;
1061
1062                                 /*  avoid cases where sscanf is not exact inverse of printf */
1063                                 snprintf(buf, IFNAMSIZ, name, i);
1064                                 if (!strncmp(buf, name_node->name, IFNAMSIZ))
1065                                         __set_bit(i, inuse);
1066                         }
1067                         if (!sscanf(d->name, name, &i))
1068                                 continue;
1069                         if (i < 0 || i >= max_netdevices)
1070                                 continue;
1071
1072                         /*  avoid cases where sscanf is not exact inverse of printf */
1073                         snprintf(buf, IFNAMSIZ, name, i);
1074                         if (!strncmp(buf, d->name, IFNAMSIZ))
1075                                 __set_bit(i, inuse);
1076                 }
1077
1078                 i = find_first_zero_bit(inuse, max_netdevices);
1079                 free_page((unsigned long) inuse);
1080         }
1081
1082         snprintf(buf, IFNAMSIZ, name, i);
1083         if (!netdev_name_in_use(net, buf))
1084                 return i;
1085
1086         /* It is possible to run out of possible slots
1087          * when the name is long and there isn't enough space left
1088          * for the digits, or if all bits are used.
1089          */
1090         return -ENFILE;
1091 }
1092
1093 static int dev_alloc_name_ns(struct net *net,
1094                              struct net_device *dev,
1095                              const char *name)
1096 {
1097         char buf[IFNAMSIZ];
1098         int ret;
1099
1100         BUG_ON(!net);
1101         ret = __dev_alloc_name(net, name, buf);
1102         if (ret >= 0)
1103                 strscpy(dev->name, buf, IFNAMSIZ);
1104         return ret;
1105 }
1106
1107 /**
1108  *      dev_alloc_name - allocate a name for a device
1109  *      @dev: device
1110  *      @name: name format string
1111  *
1112  *      Passed a format string - eg "lt%d" it will try and find a suitable
1113  *      id. It scans list of devices to build up a free map, then chooses
1114  *      the first empty slot. The caller must hold the dev_base or rtnl lock
1115  *      while allocating the name and adding the device in order to avoid
1116  *      duplicates.
1117  *      Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1118  *      Returns the number of the unit assigned or a negative errno code.
1119  */
1120
1121 int dev_alloc_name(struct net_device *dev, const char *name)
1122 {
1123         return dev_alloc_name_ns(dev_net(dev), dev, name);
1124 }
1125 EXPORT_SYMBOL(dev_alloc_name);
1126
1127 static int dev_get_valid_name(struct net *net, struct net_device *dev,
1128                               const char *name)
1129 {
1130         BUG_ON(!net);
1131
1132         if (!dev_valid_name(name))
1133                 return -EINVAL;
1134
1135         if (strchr(name, '%'))
1136                 return dev_alloc_name_ns(net, dev, name);
1137         else if (netdev_name_in_use(net, name))
1138                 return -EEXIST;
1139         else if (dev->name != name)
1140                 strscpy(dev->name, name, IFNAMSIZ);
1141
1142         return 0;
1143 }
1144
1145 /**
1146  *      dev_change_name - change name of a device
1147  *      @dev: device
1148  *      @newname: name (or format string) must be at least IFNAMSIZ
1149  *
1150  *      Change name of a device, can pass format strings "eth%d".
1151  *      for wildcarding.
1152  */
1153 int dev_change_name(struct net_device *dev, const char *newname)
1154 {
1155         unsigned char old_assign_type;
1156         char oldname[IFNAMSIZ];
1157         int err = 0;
1158         int ret;
1159         struct net *net;
1160
1161         ASSERT_RTNL();
1162         BUG_ON(!dev_net(dev));
1163
1164         net = dev_net(dev);
1165
1166         down_write(&devnet_rename_sem);
1167
1168         if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1169                 up_write(&devnet_rename_sem);
1170                 return 0;
1171         }
1172
1173         memcpy(oldname, dev->name, IFNAMSIZ);
1174
1175         err = dev_get_valid_name(net, dev, newname);
1176         if (err < 0) {
1177                 up_write(&devnet_rename_sem);
1178                 return err;
1179         }
1180
1181         if (oldname[0] && !strchr(oldname, '%'))
1182                 netdev_info(dev, "renamed from %s%s\n", oldname,
1183                             dev->flags & IFF_UP ? " (while UP)" : "");
1184
1185         old_assign_type = dev->name_assign_type;
1186         dev->name_assign_type = NET_NAME_RENAMED;
1187
1188 rollback:
1189         ret = device_rename(&dev->dev, dev->name);
1190         if (ret) {
1191                 memcpy(dev->name, oldname, IFNAMSIZ);
1192                 dev->name_assign_type = old_assign_type;
1193                 up_write(&devnet_rename_sem);
1194                 return ret;
1195         }
1196
1197         up_write(&devnet_rename_sem);
1198
1199         netdev_adjacent_rename_links(dev, oldname);
1200
1201         write_lock(&dev_base_lock);
1202         netdev_name_node_del(dev->name_node);
1203         write_unlock(&dev_base_lock);
1204
1205         synchronize_rcu();
1206
1207         write_lock(&dev_base_lock);
1208         netdev_name_node_add(net, dev->name_node);
1209         write_unlock(&dev_base_lock);
1210
1211         ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1212         ret = notifier_to_errno(ret);
1213
1214         if (ret) {
1215                 /* err >= 0 after dev_alloc_name() or stores the first errno */
1216                 if (err >= 0) {
1217                         err = ret;
1218                         down_write(&devnet_rename_sem);
1219                         memcpy(dev->name, oldname, IFNAMSIZ);
1220                         memcpy(oldname, newname, IFNAMSIZ);
1221                         dev->name_assign_type = old_assign_type;
1222                         old_assign_type = NET_NAME_RENAMED;
1223                         goto rollback;
1224                 } else {
1225                         netdev_err(dev, "name change rollback failed: %d\n",
1226                                    ret);
1227                 }
1228         }
1229
1230         return err;
1231 }
1232
1233 /**
1234  *      dev_set_alias - change ifalias of a device
1235  *      @dev: device
1236  *      @alias: name up to IFALIASZ
1237  *      @len: limit of bytes to copy from info
1238  *
1239  *      Set ifalias for a device,
1240  */
1241 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1242 {
1243         struct dev_ifalias *new_alias = NULL;
1244
1245         if (len >= IFALIASZ)
1246                 return -EINVAL;
1247
1248         if (len) {
1249                 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL);
1250                 if (!new_alias)
1251                         return -ENOMEM;
1252
1253                 memcpy(new_alias->ifalias, alias, len);
1254                 new_alias->ifalias[len] = 0;
1255         }
1256
1257         mutex_lock(&ifalias_mutex);
1258         new_alias = rcu_replace_pointer(dev->ifalias, new_alias,
1259                                         mutex_is_locked(&ifalias_mutex));
1260         mutex_unlock(&ifalias_mutex);
1261
1262         if (new_alias)
1263                 kfree_rcu(new_alias, rcuhead);
1264
1265         return len;
1266 }
1267 EXPORT_SYMBOL(dev_set_alias);
1268
1269 /**
1270  *      dev_get_alias - get ifalias of a device
1271  *      @dev: device
1272  *      @name: buffer to store name of ifalias
1273  *      @len: size of buffer
1274  *
1275  *      get ifalias for a device.  Caller must make sure dev cannot go
1276  *      away,  e.g. rcu read lock or own a reference count to device.
1277  */
1278 int dev_get_alias(const struct net_device *dev, char *name, size_t len)
1279 {
1280         const struct dev_ifalias *alias;
1281         int ret = 0;
1282
1283         rcu_read_lock();
1284         alias = rcu_dereference(dev->ifalias);
1285         if (alias)
1286                 ret = snprintf(name, len, "%s", alias->ifalias);
1287         rcu_read_unlock();
1288
1289         return ret;
1290 }
1291
1292 /**
1293  *      netdev_features_change - device changes features
1294  *      @dev: device to cause notification
1295  *
1296  *      Called to indicate a device has changed features.
1297  */
1298 void netdev_features_change(struct net_device *dev)
1299 {
1300         call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1301 }
1302 EXPORT_SYMBOL(netdev_features_change);
1303
1304 /**
1305  *      netdev_state_change - device changes state
1306  *      @dev: device to cause notification
1307  *
1308  *      Called to indicate a device has changed state. This function calls
1309  *      the notifier chains for netdev_chain and sends a NEWLINK message
1310  *      to the routing socket.
1311  */
1312 void netdev_state_change(struct net_device *dev)
1313 {
1314         if (dev->flags & IFF_UP) {
1315                 struct netdev_notifier_change_info change_info = {
1316                         .info.dev = dev,
1317                 };
1318
1319                 call_netdevice_notifiers_info(NETDEV_CHANGE,
1320                                               &change_info.info);
1321                 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL, 0, NULL);
1322         }
1323 }
1324 EXPORT_SYMBOL(netdev_state_change);
1325
1326 /**
1327  * __netdev_notify_peers - notify network peers about existence of @dev,
1328  * to be called when rtnl lock is already held.
1329  * @dev: network device
1330  *
1331  * Generate traffic such that interested network peers are aware of
1332  * @dev, such as by generating a gratuitous ARP. This may be used when
1333  * a device wants to inform the rest of the network about some sort of
1334  * reconfiguration such as a failover event or virtual machine
1335  * migration.
1336  */
1337 void __netdev_notify_peers(struct net_device *dev)
1338 {
1339         ASSERT_RTNL();
1340         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1341         call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev);
1342 }
1343 EXPORT_SYMBOL(__netdev_notify_peers);
1344
1345 /**
1346  * netdev_notify_peers - notify network peers about existence of @dev
1347  * @dev: network device
1348  *
1349  * Generate traffic such that interested network peers are aware of
1350  * @dev, such as by generating a gratuitous ARP. This may be used when
1351  * a device wants to inform the rest of the network about some sort of
1352  * reconfiguration such as a failover event or virtual machine
1353  * migration.
1354  */
1355 void netdev_notify_peers(struct net_device *dev)
1356 {
1357         rtnl_lock();
1358         __netdev_notify_peers(dev);
1359         rtnl_unlock();
1360 }
1361 EXPORT_SYMBOL(netdev_notify_peers);
1362
1363 static int napi_threaded_poll(void *data);
1364
1365 static int napi_kthread_create(struct napi_struct *n)
1366 {
1367         int err = 0;
1368
1369         /* Create and wake up the kthread once to put it in
1370          * TASK_INTERRUPTIBLE mode to avoid the blocked task
1371          * warning and work with loadavg.
1372          */
1373         n->thread = kthread_run(napi_threaded_poll, n, "napi/%s-%d",
1374                                 n->dev->name, n->napi_id);
1375         if (IS_ERR(n->thread)) {
1376                 err = PTR_ERR(n->thread);
1377                 pr_err("kthread_run failed with err %d\n", err);
1378                 n->thread = NULL;
1379         }
1380
1381         return err;
1382 }
1383
1384 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1385 {
1386         const struct net_device_ops *ops = dev->netdev_ops;
1387         int ret;
1388
1389         ASSERT_RTNL();
1390         dev_addr_check(dev);
1391
1392         if (!netif_device_present(dev)) {
1393                 /* may be detached because parent is runtime-suspended */
1394                 if (dev->dev.parent)
1395                         pm_runtime_resume(dev->dev.parent);
1396                 if (!netif_device_present(dev))
1397                         return -ENODEV;
1398         }
1399
1400         /* Block netpoll from trying to do any rx path servicing.
1401          * If we don't do this there is a chance ndo_poll_controller
1402          * or ndo_poll may be running while we open the device
1403          */
1404         netpoll_poll_disable(dev);
1405
1406         ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack);
1407         ret = notifier_to_errno(ret);
1408         if (ret)
1409                 return ret;
1410
1411         set_bit(__LINK_STATE_START, &dev->state);
1412
1413         if (ops->ndo_validate_addr)
1414                 ret = ops->ndo_validate_addr(dev);
1415
1416         if (!ret && ops->ndo_open)
1417                 ret = ops->ndo_open(dev);
1418
1419         netpoll_poll_enable(dev);
1420
1421         if (ret)
1422                 clear_bit(__LINK_STATE_START, &dev->state);
1423         else {
1424                 dev->flags |= IFF_UP;
1425                 dev_set_rx_mode(dev);
1426                 dev_activate(dev);
1427                 add_device_randomness(dev->dev_addr, dev->addr_len);
1428         }
1429
1430         return ret;
1431 }
1432
1433 /**
1434  *      dev_open        - prepare an interface for use.
1435  *      @dev: device to open
1436  *      @extack: netlink extended ack
1437  *
1438  *      Takes a device from down to up state. The device's private open
1439  *      function is invoked and then the multicast lists are loaded. Finally
1440  *      the device is moved into the up state and a %NETDEV_UP message is
1441  *      sent to the netdev notifier chain.
1442  *
1443  *      Calling this function on an active interface is a nop. On a failure
1444  *      a negative errno code is returned.
1445  */
1446 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1447 {
1448         int ret;
1449
1450         if (dev->flags & IFF_UP)
1451                 return 0;
1452
1453         ret = __dev_open(dev, extack);
1454         if (ret < 0)
1455                 return ret;
1456
1457         rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL);
1458         call_netdevice_notifiers(NETDEV_UP, dev);
1459
1460         return ret;
1461 }
1462 EXPORT_SYMBOL(dev_open);
1463
1464 static void __dev_close_many(struct list_head *head)
1465 {
1466         struct net_device *dev;
1467
1468         ASSERT_RTNL();
1469         might_sleep();
1470
1471         list_for_each_entry(dev, head, close_list) {
1472                 /* Temporarily disable netpoll until the interface is down */
1473                 netpoll_poll_disable(dev);
1474
1475                 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1476
1477                 clear_bit(__LINK_STATE_START, &dev->state);
1478
1479                 /* Synchronize to scheduled poll. We cannot touch poll list, it
1480                  * can be even on different cpu. So just clear netif_running().
1481                  *
1482                  * dev->stop() will invoke napi_disable() on all of it's
1483                  * napi_struct instances on this device.
1484                  */
1485                 smp_mb__after_atomic(); /* Commit netif_running(). */
1486         }
1487
1488         dev_deactivate_many(head);
1489
1490         list_for_each_entry(dev, head, close_list) {
1491                 const struct net_device_ops *ops = dev->netdev_ops;
1492
1493                 /*
1494                  *      Call the device specific close. This cannot fail.
1495                  *      Only if device is UP
1496                  *
1497                  *      We allow it to be called even after a DETACH hot-plug
1498                  *      event.
1499                  */
1500                 if (ops->ndo_stop)
1501                         ops->ndo_stop(dev);
1502
1503                 dev->flags &= ~IFF_UP;
1504                 netpoll_poll_enable(dev);
1505         }
1506 }
1507
1508 static void __dev_close(struct net_device *dev)
1509 {
1510         LIST_HEAD(single);
1511
1512         list_add(&dev->close_list, &single);
1513         __dev_close_many(&single);
1514         list_del(&single);
1515 }
1516
1517 void dev_close_many(struct list_head *head, bool unlink)
1518 {
1519         struct net_device *dev, *tmp;
1520
1521         /* Remove the devices that don't need to be closed */
1522         list_for_each_entry_safe(dev, tmp, head, close_list)
1523                 if (!(dev->flags & IFF_UP))
1524                         list_del_init(&dev->close_list);
1525
1526         __dev_close_many(head);
1527
1528         list_for_each_entry_safe(dev, tmp, head, close_list) {
1529                 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL);
1530                 call_netdevice_notifiers(NETDEV_DOWN, dev);
1531                 if (unlink)
1532                         list_del_init(&dev->close_list);
1533         }
1534 }
1535 EXPORT_SYMBOL(dev_close_many);
1536
1537 /**
1538  *      dev_close - shutdown an interface.
1539  *      @dev: device to shutdown
1540  *
1541  *      This function moves an active device into down state. A
1542  *      %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1543  *      is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1544  *      chain.
1545  */
1546 void dev_close(struct net_device *dev)
1547 {
1548         if (dev->flags & IFF_UP) {
1549                 LIST_HEAD(single);
1550
1551                 list_add(&dev->close_list, &single);
1552                 dev_close_many(&single, true);
1553                 list_del(&single);
1554         }
1555 }
1556 EXPORT_SYMBOL(dev_close);
1557
1558
1559 /**
1560  *      dev_disable_lro - disable Large Receive Offload on a device
1561  *      @dev: device
1562  *
1563  *      Disable Large Receive Offload (LRO) on a net device.  Must be
1564  *      called under RTNL.  This is needed if received packets may be
1565  *      forwarded to another interface.
1566  */
1567 void dev_disable_lro(struct net_device *dev)
1568 {
1569         struct net_device *lower_dev;
1570         struct list_head *iter;
1571
1572         dev->wanted_features &= ~NETIF_F_LRO;
1573         netdev_update_features(dev);
1574
1575         if (unlikely(dev->features & NETIF_F_LRO))
1576                 netdev_WARN(dev, "failed to disable LRO!\n");
1577
1578         netdev_for_each_lower_dev(dev, lower_dev, iter)
1579                 dev_disable_lro(lower_dev);
1580 }
1581 EXPORT_SYMBOL(dev_disable_lro);
1582
1583 /**
1584  *      dev_disable_gro_hw - disable HW Generic Receive Offload on a device
1585  *      @dev: device
1586  *
1587  *      Disable HW Generic Receive Offload (GRO_HW) on a net device.  Must be
1588  *      called under RTNL.  This is needed if Generic XDP is installed on
1589  *      the device.
1590  */
1591 static void dev_disable_gro_hw(struct net_device *dev)
1592 {
1593         dev->wanted_features &= ~NETIF_F_GRO_HW;
1594         netdev_update_features(dev);
1595
1596         if (unlikely(dev->features & NETIF_F_GRO_HW))
1597                 netdev_WARN(dev, "failed to disable GRO_HW!\n");
1598 }
1599
1600 const char *netdev_cmd_to_name(enum netdev_cmd cmd)
1601 {
1602 #define N(val)                                          \
1603         case NETDEV_##val:                              \
1604                 return "NETDEV_" __stringify(val);
1605         switch (cmd) {
1606         N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER)
1607         N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE)
1608         N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE)
1609         N(POST_INIT) N(PRE_UNINIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN)
1610         N(CHANGEUPPER) N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA)
1611         N(BONDING_INFO) N(PRECHANGEUPPER) N(CHANGELOWERSTATE)
1612         N(UDP_TUNNEL_PUSH_INFO) N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN)
1613         N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO)
1614         N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO)
1615         N(PRE_CHANGEADDR) N(OFFLOAD_XSTATS_ENABLE) N(OFFLOAD_XSTATS_DISABLE)
1616         N(OFFLOAD_XSTATS_REPORT_USED) N(OFFLOAD_XSTATS_REPORT_DELTA)
1617         }
1618 #undef N
1619         return "UNKNOWN_NETDEV_EVENT";
1620 }
1621 EXPORT_SYMBOL_GPL(netdev_cmd_to_name);
1622
1623 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1624                                    struct net_device *dev)
1625 {
1626         struct netdev_notifier_info info = {
1627                 .dev = dev,
1628         };
1629
1630         return nb->notifier_call(nb, val, &info);
1631 }
1632
1633 static int call_netdevice_register_notifiers(struct notifier_block *nb,
1634                                              struct net_device *dev)
1635 {
1636         int err;
1637
1638         err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1639         err = notifier_to_errno(err);
1640         if (err)
1641                 return err;
1642
1643         if (!(dev->flags & IFF_UP))
1644                 return 0;
1645
1646         call_netdevice_notifier(nb, NETDEV_UP, dev);
1647         return 0;
1648 }
1649
1650 static void call_netdevice_unregister_notifiers(struct notifier_block *nb,
1651                                                 struct net_device *dev)
1652 {
1653         if (dev->flags & IFF_UP) {
1654                 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1655                                         dev);
1656                 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1657         }
1658         call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1659 }
1660
1661 static int call_netdevice_register_net_notifiers(struct notifier_block *nb,
1662                                                  struct net *net)
1663 {
1664         struct net_device *dev;
1665         int err;
1666
1667         for_each_netdev(net, dev) {
1668                 err = call_netdevice_register_notifiers(nb, dev);
1669                 if (err)
1670                         goto rollback;
1671         }
1672         return 0;
1673
1674 rollback:
1675         for_each_netdev_continue_reverse(net, dev)
1676                 call_netdevice_unregister_notifiers(nb, dev);
1677         return err;
1678 }
1679
1680 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb,
1681                                                     struct net *net)
1682 {
1683         struct net_device *dev;
1684
1685         for_each_netdev(net, dev)
1686                 call_netdevice_unregister_notifiers(nb, dev);
1687 }
1688
1689 static int dev_boot_phase = 1;
1690
1691 /**
1692  * register_netdevice_notifier - register a network notifier block
1693  * @nb: notifier
1694  *
1695  * Register a notifier to be called when network device events occur.
1696  * The notifier passed is linked into the kernel structures and must
1697  * not be reused until it has been unregistered. A negative errno code
1698  * is returned on a failure.
1699  *
1700  * When registered all registration and up events are replayed
1701  * to the new notifier to allow device to have a race free
1702  * view of the network device list.
1703  */
1704
1705 int register_netdevice_notifier(struct notifier_block *nb)
1706 {
1707         struct net *net;
1708         int err;
1709
1710         /* Close race with setup_net() and cleanup_net() */
1711         down_write(&pernet_ops_rwsem);
1712         rtnl_lock();
1713         err = raw_notifier_chain_register(&netdev_chain, nb);
1714         if (err)
1715                 goto unlock;
1716         if (dev_boot_phase)
1717                 goto unlock;
1718         for_each_net(net) {
1719                 err = call_netdevice_register_net_notifiers(nb, net);
1720                 if (err)
1721                         goto rollback;
1722         }
1723
1724 unlock:
1725         rtnl_unlock();
1726         up_write(&pernet_ops_rwsem);
1727         return err;
1728
1729 rollback:
1730         for_each_net_continue_reverse(net)
1731                 call_netdevice_unregister_net_notifiers(nb, net);
1732
1733         raw_notifier_chain_unregister(&netdev_chain, nb);
1734         goto unlock;
1735 }
1736 EXPORT_SYMBOL(register_netdevice_notifier);
1737
1738 /**
1739  * unregister_netdevice_notifier - unregister a network notifier block
1740  * @nb: notifier
1741  *
1742  * Unregister a notifier previously registered by
1743  * register_netdevice_notifier(). The notifier is unlinked into the
1744  * kernel structures and may then be reused. A negative errno code
1745  * is returned on a failure.
1746  *
1747  * After unregistering unregister and down device events are synthesized
1748  * for all devices on the device list to the removed notifier to remove
1749  * the need for special case cleanup code.
1750  */
1751
1752 int unregister_netdevice_notifier(struct notifier_block *nb)
1753 {
1754         struct net *net;
1755         int err;
1756
1757         /* Close race with setup_net() and cleanup_net() */
1758         down_write(&pernet_ops_rwsem);
1759         rtnl_lock();
1760         err = raw_notifier_chain_unregister(&netdev_chain, nb);
1761         if (err)
1762                 goto unlock;
1763
1764         for_each_net(net)
1765                 call_netdevice_unregister_net_notifiers(nb, net);
1766
1767 unlock:
1768         rtnl_unlock();
1769         up_write(&pernet_ops_rwsem);
1770         return err;
1771 }
1772 EXPORT_SYMBOL(unregister_netdevice_notifier);
1773
1774 static int __register_netdevice_notifier_net(struct net *net,
1775                                              struct notifier_block *nb,
1776                                              bool ignore_call_fail)
1777 {
1778         int err;
1779
1780         err = raw_notifier_chain_register(&net->netdev_chain, nb);
1781         if (err)
1782                 return err;
1783         if (dev_boot_phase)
1784                 return 0;
1785
1786         err = call_netdevice_register_net_notifiers(nb, net);
1787         if (err && !ignore_call_fail)
1788                 goto chain_unregister;
1789
1790         return 0;
1791
1792 chain_unregister:
1793         raw_notifier_chain_unregister(&net->netdev_chain, nb);
1794         return err;
1795 }
1796
1797 static int __unregister_netdevice_notifier_net(struct net *net,
1798                                                struct notifier_block *nb)
1799 {
1800         int err;
1801
1802         err = raw_notifier_chain_unregister(&net->netdev_chain, nb);
1803         if (err)
1804                 return err;
1805
1806         call_netdevice_unregister_net_notifiers(nb, net);
1807         return 0;
1808 }
1809
1810 /**
1811  * register_netdevice_notifier_net - register a per-netns network notifier block
1812  * @net: network namespace
1813  * @nb: notifier
1814  *
1815  * Register a notifier to be called when network device events occur.
1816  * The notifier passed is linked into the kernel structures and must
1817  * not be reused until it has been unregistered. A negative errno code
1818  * is returned on a failure.
1819  *
1820  * When registered all registration and up events are replayed
1821  * to the new notifier to allow device to have a race free
1822  * view of the network device list.
1823  */
1824
1825 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb)
1826 {
1827         int err;
1828
1829         rtnl_lock();
1830         err = __register_netdevice_notifier_net(net, nb, false);
1831         rtnl_unlock();
1832         return err;
1833 }
1834 EXPORT_SYMBOL(register_netdevice_notifier_net);
1835
1836 /**
1837  * unregister_netdevice_notifier_net - unregister a per-netns
1838  *                                     network notifier block
1839  * @net: network namespace
1840  * @nb: notifier
1841  *
1842  * Unregister a notifier previously registered by
1843  * register_netdevice_notifier(). The notifier is unlinked into the
1844  * kernel structures and may then be reused. A negative errno code
1845  * is returned on a failure.
1846  *
1847  * After unregistering unregister and down device events are synthesized
1848  * for all devices on the device list to the removed notifier to remove
1849  * the need for special case cleanup code.
1850  */
1851
1852 int unregister_netdevice_notifier_net(struct net *net,
1853                                       struct notifier_block *nb)
1854 {
1855         int err;
1856
1857         rtnl_lock();
1858         err = __unregister_netdevice_notifier_net(net, nb);
1859         rtnl_unlock();
1860         return err;
1861 }
1862 EXPORT_SYMBOL(unregister_netdevice_notifier_net);
1863
1864 static void __move_netdevice_notifier_net(struct net *src_net,
1865                                           struct net *dst_net,
1866                                           struct notifier_block *nb)
1867 {
1868         __unregister_netdevice_notifier_net(src_net, nb);
1869         __register_netdevice_notifier_net(dst_net, nb, true);
1870 }
1871
1872 void move_netdevice_notifier_net(struct net *src_net, struct net *dst_net,
1873                                  struct notifier_block *nb)
1874 {
1875         rtnl_lock();
1876         __move_netdevice_notifier_net(src_net, dst_net, nb);
1877         rtnl_unlock();
1878 }
1879
1880 int register_netdevice_notifier_dev_net(struct net_device *dev,
1881                                         struct notifier_block *nb,
1882                                         struct netdev_net_notifier *nn)
1883 {
1884         int err;
1885
1886         rtnl_lock();
1887         err = __register_netdevice_notifier_net(dev_net(dev), nb, false);
1888         if (!err) {
1889                 nn->nb = nb;
1890                 list_add(&nn->list, &dev->net_notifier_list);
1891         }
1892         rtnl_unlock();
1893         return err;
1894 }
1895 EXPORT_SYMBOL(register_netdevice_notifier_dev_net);
1896
1897 int unregister_netdevice_notifier_dev_net(struct net_device *dev,
1898                                           struct notifier_block *nb,
1899                                           struct netdev_net_notifier *nn)
1900 {
1901         int err;
1902
1903         rtnl_lock();
1904         list_del(&nn->list);
1905         err = __unregister_netdevice_notifier_net(dev_net(dev), nb);
1906         rtnl_unlock();
1907         return err;
1908 }
1909 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net);
1910
1911 static void move_netdevice_notifiers_dev_net(struct net_device *dev,
1912                                              struct net *net)
1913 {
1914         struct netdev_net_notifier *nn;
1915
1916         list_for_each_entry(nn, &dev->net_notifier_list, list)
1917                 __move_netdevice_notifier_net(dev_net(dev), net, nn->nb);
1918 }
1919
1920 /**
1921  *      call_netdevice_notifiers_info - call all network notifier blocks
1922  *      @val: value passed unmodified to notifier function
1923  *      @info: notifier information data
1924  *
1925  *      Call all network notifier blocks.  Parameters and return value
1926  *      are as for raw_notifier_call_chain().
1927  */
1928
1929 static int call_netdevice_notifiers_info(unsigned long val,
1930                                          struct netdev_notifier_info *info)
1931 {
1932         struct net *net = dev_net(info->dev);
1933         int ret;
1934
1935         ASSERT_RTNL();
1936
1937         /* Run per-netns notifier block chain first, then run the global one.
1938          * Hopefully, one day, the global one is going to be removed after
1939          * all notifier block registrators get converted to be per-netns.
1940          */
1941         ret = raw_notifier_call_chain(&net->netdev_chain, val, info);
1942         if (ret & NOTIFY_STOP_MASK)
1943                 return ret;
1944         return raw_notifier_call_chain(&netdev_chain, val, info);
1945 }
1946
1947 /**
1948  *      call_netdevice_notifiers_info_robust - call per-netns notifier blocks
1949  *                                             for and rollback on error
1950  *      @val_up: value passed unmodified to notifier function
1951  *      @val_down: value passed unmodified to the notifier function when
1952  *                 recovering from an error on @val_up
1953  *      @info: notifier information data
1954  *
1955  *      Call all per-netns network notifier blocks, but not notifier blocks on
1956  *      the global notifier chain. Parameters and return value are as for
1957  *      raw_notifier_call_chain_robust().
1958  */
1959
1960 static int
1961 call_netdevice_notifiers_info_robust(unsigned long val_up,
1962                                      unsigned long val_down,
1963                                      struct netdev_notifier_info *info)
1964 {
1965         struct net *net = dev_net(info->dev);
1966
1967         ASSERT_RTNL();
1968
1969         return raw_notifier_call_chain_robust(&net->netdev_chain,
1970                                               val_up, val_down, info);
1971 }
1972
1973 static int call_netdevice_notifiers_extack(unsigned long val,
1974                                            struct net_device *dev,
1975                                            struct netlink_ext_ack *extack)
1976 {
1977         struct netdev_notifier_info info = {
1978                 .dev = dev,
1979                 .extack = extack,
1980         };
1981
1982         return call_netdevice_notifiers_info(val, &info);
1983 }
1984
1985 /**
1986  *      call_netdevice_notifiers - call all network notifier blocks
1987  *      @val: value passed unmodified to notifier function
1988  *      @dev: net_device pointer passed unmodified to notifier function
1989  *
1990  *      Call all network notifier blocks.  Parameters and return value
1991  *      are as for raw_notifier_call_chain().
1992  */
1993
1994 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1995 {
1996         return call_netdevice_notifiers_extack(val, dev, NULL);
1997 }
1998 EXPORT_SYMBOL(call_netdevice_notifiers);
1999
2000 /**
2001  *      call_netdevice_notifiers_mtu - call all network notifier blocks
2002  *      @val: value passed unmodified to notifier function
2003  *      @dev: net_device pointer passed unmodified to notifier function
2004  *      @arg: additional u32 argument passed to the notifier function
2005  *
2006  *      Call all network notifier blocks.  Parameters and return value
2007  *      are as for raw_notifier_call_chain().
2008  */
2009 static int call_netdevice_notifiers_mtu(unsigned long val,
2010                                         struct net_device *dev, u32 arg)
2011 {
2012         struct netdev_notifier_info_ext info = {
2013                 .info.dev = dev,
2014                 .ext.mtu = arg,
2015         };
2016
2017         BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0);
2018
2019         return call_netdevice_notifiers_info(val, &info.info);
2020 }
2021
2022 #ifdef CONFIG_NET_INGRESS
2023 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key);
2024
2025 void net_inc_ingress_queue(void)
2026 {
2027         static_branch_inc(&ingress_needed_key);
2028 }
2029 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
2030
2031 void net_dec_ingress_queue(void)
2032 {
2033         static_branch_dec(&ingress_needed_key);
2034 }
2035 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
2036 #endif
2037
2038 #ifdef CONFIG_NET_EGRESS
2039 static DEFINE_STATIC_KEY_FALSE(egress_needed_key);
2040
2041 void net_inc_egress_queue(void)
2042 {
2043         static_branch_inc(&egress_needed_key);
2044 }
2045 EXPORT_SYMBOL_GPL(net_inc_egress_queue);
2046
2047 void net_dec_egress_queue(void)
2048 {
2049         static_branch_dec(&egress_needed_key);
2050 }
2051 EXPORT_SYMBOL_GPL(net_dec_egress_queue);
2052 #endif
2053
2054 DEFINE_STATIC_KEY_FALSE(netstamp_needed_key);
2055 EXPORT_SYMBOL(netstamp_needed_key);
2056 #ifdef CONFIG_JUMP_LABEL
2057 static atomic_t netstamp_needed_deferred;
2058 static atomic_t netstamp_wanted;
2059 static void netstamp_clear(struct work_struct *work)
2060 {
2061         int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
2062         int wanted;
2063
2064         wanted = atomic_add_return(deferred, &netstamp_wanted);
2065         if (wanted > 0)
2066                 static_branch_enable(&netstamp_needed_key);
2067         else
2068                 static_branch_disable(&netstamp_needed_key);
2069 }
2070 static DECLARE_WORK(netstamp_work, netstamp_clear);
2071 #endif
2072
2073 void net_enable_timestamp(void)
2074 {
2075 #ifdef CONFIG_JUMP_LABEL
2076         int wanted = atomic_read(&netstamp_wanted);
2077
2078         while (wanted > 0) {
2079                 if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted + 1))
2080                         return;
2081         }
2082         atomic_inc(&netstamp_needed_deferred);
2083         schedule_work(&netstamp_work);
2084 #else
2085         static_branch_inc(&netstamp_needed_key);
2086 #endif
2087 }
2088 EXPORT_SYMBOL(net_enable_timestamp);
2089
2090 void net_disable_timestamp(void)
2091 {
2092 #ifdef CONFIG_JUMP_LABEL
2093         int wanted = atomic_read(&netstamp_wanted);
2094
2095         while (wanted > 1) {
2096                 if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted - 1))
2097                         return;
2098         }
2099         atomic_dec(&netstamp_needed_deferred);
2100         schedule_work(&netstamp_work);
2101 #else
2102         static_branch_dec(&netstamp_needed_key);
2103 #endif
2104 }
2105 EXPORT_SYMBOL(net_disable_timestamp);
2106
2107 static inline void net_timestamp_set(struct sk_buff *skb)
2108 {
2109         skb->tstamp = 0;
2110         skb->mono_delivery_time = 0;
2111         if (static_branch_unlikely(&netstamp_needed_key))
2112                 skb->tstamp = ktime_get_real();
2113 }
2114
2115 #define net_timestamp_check(COND, SKB)                          \
2116         if (static_branch_unlikely(&netstamp_needed_key)) {     \
2117                 if ((COND) && !(SKB)->tstamp)                   \
2118                         (SKB)->tstamp = ktime_get_real();       \
2119         }                                                       \
2120
2121 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb)
2122 {
2123         return __is_skb_forwardable(dev, skb, true);
2124 }
2125 EXPORT_SYMBOL_GPL(is_skb_forwardable);
2126
2127 static int __dev_forward_skb2(struct net_device *dev, struct sk_buff *skb,
2128                               bool check_mtu)
2129 {
2130         int ret = ____dev_forward_skb(dev, skb, check_mtu);
2131
2132         if (likely(!ret)) {
2133                 skb->protocol = eth_type_trans(skb, dev);
2134                 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
2135         }
2136
2137         return ret;
2138 }
2139
2140 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2141 {
2142         return __dev_forward_skb2(dev, skb, true);
2143 }
2144 EXPORT_SYMBOL_GPL(__dev_forward_skb);
2145
2146 /**
2147  * dev_forward_skb - loopback an skb to another netif
2148  *
2149  * @dev: destination network device
2150  * @skb: buffer to forward
2151  *
2152  * return values:
2153  *      NET_RX_SUCCESS  (no congestion)
2154  *      NET_RX_DROP     (packet was dropped, but freed)
2155  *
2156  * dev_forward_skb can be used for injecting an skb from the
2157  * start_xmit function of one device into the receive queue
2158  * of another device.
2159  *
2160  * The receiving device may be in another namespace, so
2161  * we have to clear all information in the skb that could
2162  * impact namespace isolation.
2163  */
2164 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2165 {
2166         return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
2167 }
2168 EXPORT_SYMBOL_GPL(dev_forward_skb);
2169
2170 int dev_forward_skb_nomtu(struct net_device *dev, struct sk_buff *skb)
2171 {
2172         return __dev_forward_skb2(dev, skb, false) ?: netif_rx_internal(skb);
2173 }
2174
2175 static inline int deliver_skb(struct sk_buff *skb,
2176                               struct packet_type *pt_prev,
2177                               struct net_device *orig_dev)
2178 {
2179         if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
2180                 return -ENOMEM;
2181         refcount_inc(&skb->users);
2182         return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2183 }
2184
2185 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
2186                                           struct packet_type **pt,
2187                                           struct net_device *orig_dev,
2188                                           __be16 type,
2189                                           struct list_head *ptype_list)
2190 {
2191         struct packet_type *ptype, *pt_prev = *pt;
2192
2193         list_for_each_entry_rcu(ptype, ptype_list, list) {
2194                 if (ptype->type != type)
2195                         continue;
2196                 if (pt_prev)
2197                         deliver_skb(skb, pt_prev, orig_dev);
2198                 pt_prev = ptype;
2199         }
2200         *pt = pt_prev;
2201 }
2202
2203 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
2204 {
2205         if (!ptype->af_packet_priv || !skb->sk)
2206                 return false;
2207
2208         if (ptype->id_match)
2209                 return ptype->id_match(ptype, skb->sk);
2210         else if ((struct sock *)ptype->af_packet_priv == skb->sk)
2211                 return true;
2212
2213         return false;
2214 }
2215
2216 /**
2217  * dev_nit_active - return true if any network interface taps are in use
2218  *
2219  * @dev: network device to check for the presence of taps
2220  */
2221 bool dev_nit_active(struct net_device *dev)
2222 {
2223         return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all);
2224 }
2225 EXPORT_SYMBOL_GPL(dev_nit_active);
2226
2227 /*
2228  *      Support routine. Sends outgoing frames to any network
2229  *      taps currently in use.
2230  */
2231
2232 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
2233 {
2234         struct packet_type *ptype;
2235         struct sk_buff *skb2 = NULL;
2236         struct packet_type *pt_prev = NULL;
2237         struct list_head *ptype_list = &ptype_all;
2238
2239         rcu_read_lock();
2240 again:
2241         list_for_each_entry_rcu(ptype, ptype_list, list) {
2242                 if (ptype->ignore_outgoing)
2243                         continue;
2244
2245                 /* Never send packets back to the socket
2246                  * they originated from - MvS (miquels@drinkel.ow.org)
2247                  */
2248                 if (skb_loop_sk(ptype, skb))
2249                         continue;
2250
2251                 if (pt_prev) {
2252                         deliver_skb(skb2, pt_prev, skb->dev);
2253                         pt_prev = ptype;
2254                         continue;
2255                 }
2256
2257                 /* need to clone skb, done only once */
2258                 skb2 = skb_clone(skb, GFP_ATOMIC);
2259                 if (!skb2)
2260                         goto out_unlock;
2261
2262                 net_timestamp_set(skb2);
2263
2264                 /* skb->nh should be correctly
2265                  * set by sender, so that the second statement is
2266                  * just protection against buggy protocols.
2267                  */
2268                 skb_reset_mac_header(skb2);
2269
2270                 if (skb_network_header(skb2) < skb2->data ||
2271                     skb_network_header(skb2) > skb_tail_pointer(skb2)) {
2272                         net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
2273                                              ntohs(skb2->protocol),
2274                                              dev->name);
2275                         skb_reset_network_header(skb2);
2276                 }
2277
2278                 skb2->transport_header = skb2->network_header;
2279                 skb2->pkt_type = PACKET_OUTGOING;
2280                 pt_prev = ptype;
2281         }
2282
2283         if (ptype_list == &ptype_all) {
2284                 ptype_list = &dev->ptype_all;
2285                 goto again;
2286         }
2287 out_unlock:
2288         if (pt_prev) {
2289                 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC))
2290                         pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
2291                 else
2292                         kfree_skb(skb2);
2293         }
2294         rcu_read_unlock();
2295 }
2296 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit);
2297
2298 /**
2299  * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
2300  * @dev: Network device
2301  * @txq: number of queues available
2302  *
2303  * If real_num_tx_queues is changed the tc mappings may no longer be
2304  * valid. To resolve this verify the tc mapping remains valid and if
2305  * not NULL the mapping. With no priorities mapping to this
2306  * offset/count pair it will no longer be used. In the worst case TC0
2307  * is invalid nothing can be done so disable priority mappings. If is
2308  * expected that drivers will fix this mapping if they can before
2309  * calling netif_set_real_num_tx_queues.
2310  */
2311 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
2312 {
2313         int i;
2314         struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2315
2316         /* If TC0 is invalidated disable TC mapping */
2317         if (tc->offset + tc->count > txq) {
2318                 netdev_warn(dev, "Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
2319                 dev->num_tc = 0;
2320                 return;
2321         }
2322
2323         /* Invalidated prio to tc mappings set to TC0 */
2324         for (i = 1; i < TC_BITMASK + 1; i++) {
2325                 int q = netdev_get_prio_tc_map(dev, i);
2326
2327                 tc = &dev->tc_to_txq[q];
2328                 if (tc->offset + tc->count > txq) {
2329                         netdev_warn(dev, "Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
2330                                     i, q);
2331                         netdev_set_prio_tc_map(dev, i, 0);
2332                 }
2333         }
2334 }
2335
2336 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq)
2337 {
2338         if (dev->num_tc) {
2339                 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2340                 int i;
2341
2342                 /* walk through the TCs and see if it falls into any of them */
2343                 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) {
2344                         if ((txq - tc->offset) < tc->count)
2345                                 return i;
2346                 }
2347
2348                 /* didn't find it, just return -1 to indicate no match */
2349                 return -1;
2350         }
2351
2352         return 0;
2353 }
2354 EXPORT_SYMBOL(netdev_txq_to_tc);
2355
2356 #ifdef CONFIG_XPS
2357 static struct static_key xps_needed __read_mostly;
2358 static struct static_key xps_rxqs_needed __read_mostly;
2359 static DEFINE_MUTEX(xps_map_mutex);
2360 #define xmap_dereference(P)             \
2361         rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
2362
2363 static bool remove_xps_queue(struct xps_dev_maps *dev_maps,
2364                              struct xps_dev_maps *old_maps, int tci, u16 index)
2365 {
2366         struct xps_map *map = NULL;
2367         int pos;
2368
2369         if (dev_maps)
2370                 map = xmap_dereference(dev_maps->attr_map[tci]);
2371         if (!map)
2372                 return false;
2373
2374         for (pos = map->len; pos--;) {
2375                 if (map->queues[pos] != index)
2376                         continue;
2377
2378                 if (map->len > 1) {
2379                         map->queues[pos] = map->queues[--map->len];
2380                         break;
2381                 }
2382
2383                 if (old_maps)
2384                         RCU_INIT_POINTER(old_maps->attr_map[tci], NULL);
2385                 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2386                 kfree_rcu(map, rcu);
2387                 return false;
2388         }
2389
2390         return true;
2391 }
2392
2393 static bool remove_xps_queue_cpu(struct net_device *dev,
2394                                  struct xps_dev_maps *dev_maps,
2395                                  int cpu, u16 offset, u16 count)
2396 {
2397         int num_tc = dev_maps->num_tc;
2398         bool active = false;
2399         int tci;
2400
2401         for (tci = cpu * num_tc; num_tc--; tci++) {
2402                 int i, j;
2403
2404                 for (i = count, j = offset; i--; j++) {
2405                         if (!remove_xps_queue(dev_maps, NULL, tci, j))
2406                                 break;
2407                 }
2408
2409                 active |= i < 0;
2410         }
2411
2412         return active;
2413 }
2414
2415 static void reset_xps_maps(struct net_device *dev,
2416                            struct xps_dev_maps *dev_maps,
2417                            enum xps_map_type type)
2418 {
2419         static_key_slow_dec_cpuslocked(&xps_needed);
2420         if (type == XPS_RXQS)
2421                 static_key_slow_dec_cpuslocked(&xps_rxqs_needed);
2422
2423         RCU_INIT_POINTER(dev->xps_maps[type], NULL);
2424
2425         kfree_rcu(dev_maps, rcu);
2426 }
2427
2428 static void clean_xps_maps(struct net_device *dev, enum xps_map_type type,
2429                            u16 offset, u16 count)
2430 {
2431         struct xps_dev_maps *dev_maps;
2432         bool active = false;
2433         int i, j;
2434
2435         dev_maps = xmap_dereference(dev->xps_maps[type]);
2436         if (!dev_maps)
2437                 return;
2438
2439         for (j = 0; j < dev_maps->nr_ids; j++)
2440                 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, count);
2441         if (!active)
2442                 reset_xps_maps(dev, dev_maps, type);
2443
2444         if (type == XPS_CPUS) {
2445                 for (i = offset + (count - 1); count--; i--)
2446                         netdev_queue_numa_node_write(
2447                                 netdev_get_tx_queue(dev, i), NUMA_NO_NODE);
2448         }
2449 }
2450
2451 static void netif_reset_xps_queues(struct net_device *dev, u16 offset,
2452                                    u16 count)
2453 {
2454         if (!static_key_false(&xps_needed))
2455                 return;
2456
2457         cpus_read_lock();
2458         mutex_lock(&xps_map_mutex);
2459
2460         if (static_key_false(&xps_rxqs_needed))
2461                 clean_xps_maps(dev, XPS_RXQS, offset, count);
2462
2463         clean_xps_maps(dev, XPS_CPUS, offset, count);
2464
2465         mutex_unlock(&xps_map_mutex);
2466         cpus_read_unlock();
2467 }
2468
2469 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
2470 {
2471         netif_reset_xps_queues(dev, index, dev->num_tx_queues - index);
2472 }
2473
2474 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index,
2475                                       u16 index, bool is_rxqs_map)
2476 {
2477         struct xps_map *new_map;
2478         int alloc_len = XPS_MIN_MAP_ALLOC;
2479         int i, pos;
2480
2481         for (pos = 0; map && pos < map->len; pos++) {
2482                 if (map->queues[pos] != index)
2483                         continue;
2484                 return map;
2485         }
2486
2487         /* Need to add tx-queue to this CPU's/rx-queue's existing map */
2488         if (map) {
2489                 if (pos < map->alloc_len)
2490                         return map;
2491
2492                 alloc_len = map->alloc_len * 2;
2493         }
2494
2495         /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's
2496          *  map
2497          */
2498         if (is_rxqs_map)
2499                 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL);
2500         else
2501                 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2502                                        cpu_to_node(attr_index));
2503         if (!new_map)
2504                 return NULL;
2505
2506         for (i = 0; i < pos; i++)
2507                 new_map->queues[i] = map->queues[i];
2508         new_map->alloc_len = alloc_len;
2509         new_map->len = pos;
2510
2511         return new_map;
2512 }
2513
2514 /* Copy xps maps at a given index */
2515 static void xps_copy_dev_maps(struct xps_dev_maps *dev_maps,
2516                               struct xps_dev_maps *new_dev_maps, int index,
2517                               int tc, bool skip_tc)
2518 {
2519         int i, tci = index * dev_maps->num_tc;
2520         struct xps_map *map;
2521
2522         /* copy maps belonging to foreign traffic classes */
2523         for (i = 0; i < dev_maps->num_tc; i++, tci++) {
2524                 if (i == tc && skip_tc)
2525                         continue;
2526
2527                 /* fill in the new device map from the old device map */
2528                 map = xmap_dereference(dev_maps->attr_map[tci]);
2529                 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2530         }
2531 }
2532
2533 /* Must be called under cpus_read_lock */
2534 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask,
2535                           u16 index, enum xps_map_type type)
2536 {
2537         struct xps_dev_maps *dev_maps, *new_dev_maps = NULL, *old_dev_maps = NULL;
2538         const unsigned long *online_mask = NULL;
2539         bool active = false, copy = false;
2540         int i, j, tci, numa_node_id = -2;
2541         int maps_sz, num_tc = 1, tc = 0;
2542         struct xps_map *map, *new_map;
2543         unsigned int nr_ids;
2544
2545         if (dev->num_tc) {
2546                 /* Do not allow XPS on subordinate device directly */
2547                 num_tc = dev->num_tc;
2548                 if (num_tc < 0)
2549                         return -EINVAL;
2550
2551                 /* If queue belongs to subordinate dev use its map */
2552                 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
2553
2554                 tc = netdev_txq_to_tc(dev, index);
2555                 if (tc < 0)
2556                         return -EINVAL;
2557         }
2558
2559         mutex_lock(&xps_map_mutex);
2560
2561         dev_maps = xmap_dereference(dev->xps_maps[type]);
2562         if (type == XPS_RXQS) {
2563                 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues);
2564                 nr_ids = dev->num_rx_queues;
2565         } else {
2566                 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc);
2567                 if (num_possible_cpus() > 1)
2568                         online_mask = cpumask_bits(cpu_online_mask);
2569                 nr_ids = nr_cpu_ids;
2570         }
2571
2572         if (maps_sz < L1_CACHE_BYTES)
2573                 maps_sz = L1_CACHE_BYTES;
2574
2575         /* The old dev_maps could be larger or smaller than the one we're
2576          * setting up now, as dev->num_tc or nr_ids could have been updated in
2577          * between. We could try to be smart, but let's be safe instead and only
2578          * copy foreign traffic classes if the two map sizes match.
2579          */
2580         if (dev_maps &&
2581             dev_maps->num_tc == num_tc && dev_maps->nr_ids == nr_ids)
2582                 copy = true;
2583
2584         /* allocate memory for queue storage */
2585         for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids),
2586              j < nr_ids;) {
2587                 if (!new_dev_maps) {
2588                         new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2589                         if (!new_dev_maps) {
2590                                 mutex_unlock(&xps_map_mutex);
2591                                 return -ENOMEM;
2592                         }
2593
2594                         new_dev_maps->nr_ids = nr_ids;
2595                         new_dev_maps->num_tc = num_tc;
2596                 }
2597
2598                 tci = j * num_tc + tc;
2599                 map = copy ? xmap_dereference(dev_maps->attr_map[tci]) : NULL;
2600
2601                 map = expand_xps_map(map, j, index, type == XPS_RXQS);
2602                 if (!map)
2603                         goto error;
2604
2605                 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2606         }
2607
2608         if (!new_dev_maps)
2609                 goto out_no_new_maps;
2610
2611         if (!dev_maps) {
2612                 /* Increment static keys at most once per type */
2613                 static_key_slow_inc_cpuslocked(&xps_needed);
2614                 if (type == XPS_RXQS)
2615                         static_key_slow_inc_cpuslocked(&xps_rxqs_needed);
2616         }
2617
2618         for (j = 0; j < nr_ids; j++) {
2619                 bool skip_tc = false;
2620
2621                 tci = j * num_tc + tc;
2622                 if (netif_attr_test_mask(j, mask, nr_ids) &&
2623                     netif_attr_test_online(j, online_mask, nr_ids)) {
2624                         /* add tx-queue to CPU/rx-queue maps */
2625                         int pos = 0;
2626
2627                         skip_tc = true;
2628
2629                         map = xmap_dereference(new_dev_maps->attr_map[tci]);
2630                         while ((pos < map->len) && (map->queues[pos] != index))
2631                                 pos++;
2632
2633                         if (pos == map->len)
2634                                 map->queues[map->len++] = index;
2635 #ifdef CONFIG_NUMA
2636                         if (type == XPS_CPUS) {
2637                                 if (numa_node_id == -2)
2638                                         numa_node_id = cpu_to_node(j);
2639                                 else if (numa_node_id != cpu_to_node(j))
2640                                         numa_node_id = -1;
2641                         }
2642 #endif
2643                 }
2644
2645                 if (copy)
2646                         xps_copy_dev_maps(dev_maps, new_dev_maps, j, tc,
2647                                           skip_tc);
2648         }
2649
2650         rcu_assign_pointer(dev->xps_maps[type], new_dev_maps);
2651
2652         /* Cleanup old maps */
2653         if (!dev_maps)
2654                 goto out_no_old_maps;
2655
2656         for (j = 0; j < dev_maps->nr_ids; j++) {
2657                 for (i = num_tc, tci = j * dev_maps->num_tc; i--; tci++) {
2658                         map = xmap_dereference(dev_maps->attr_map[tci]);
2659                         if (!map)
2660                                 continue;
2661
2662                         if (copy) {
2663                                 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2664                                 if (map == new_map)
2665                                         continue;
2666                         }
2667
2668                         RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2669                         kfree_rcu(map, rcu);
2670                 }
2671         }
2672
2673         old_dev_maps = dev_maps;
2674
2675 out_no_old_maps:
2676         dev_maps = new_dev_maps;
2677         active = true;
2678
2679 out_no_new_maps:
2680         if (type == XPS_CPUS)
2681                 /* update Tx queue numa node */
2682                 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2683                                              (numa_node_id >= 0) ?
2684                                              numa_node_id : NUMA_NO_NODE);
2685
2686         if (!dev_maps)
2687                 goto out_no_maps;
2688
2689         /* removes tx-queue from unused CPUs/rx-queues */
2690         for (j = 0; j < dev_maps->nr_ids; j++) {
2691                 tci = j * dev_maps->num_tc;
2692
2693                 for (i = 0; i < dev_maps->num_tc; i++, tci++) {
2694                         if (i == tc &&
2695                             netif_attr_test_mask(j, mask, dev_maps->nr_ids) &&
2696                             netif_attr_test_online(j, online_mask, dev_maps->nr_ids))
2697                                 continue;
2698
2699                         active |= remove_xps_queue(dev_maps,
2700                                                    copy ? old_dev_maps : NULL,
2701                                                    tci, index);
2702                 }
2703         }
2704
2705         if (old_dev_maps)
2706                 kfree_rcu(old_dev_maps, rcu);
2707
2708         /* free map if not active */
2709         if (!active)
2710                 reset_xps_maps(dev, dev_maps, type);
2711
2712 out_no_maps:
2713         mutex_unlock(&xps_map_mutex);
2714
2715         return 0;
2716 error:
2717         /* remove any maps that we added */
2718         for (j = 0; j < nr_ids; j++) {
2719                 for (i = num_tc, tci = j * num_tc; i--; tci++) {
2720                         new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2721                         map = copy ?
2722                               xmap_dereference(dev_maps->attr_map[tci]) :
2723                               NULL;
2724                         if (new_map && new_map != map)
2725                                 kfree(new_map);
2726                 }
2727         }
2728
2729         mutex_unlock(&xps_map_mutex);
2730
2731         kfree(new_dev_maps);
2732         return -ENOMEM;
2733 }
2734 EXPORT_SYMBOL_GPL(__netif_set_xps_queue);
2735
2736 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2737                         u16 index)
2738 {
2739         int ret;
2740
2741         cpus_read_lock();
2742         ret =  __netif_set_xps_queue(dev, cpumask_bits(mask), index, XPS_CPUS);
2743         cpus_read_unlock();
2744
2745         return ret;
2746 }
2747 EXPORT_SYMBOL(netif_set_xps_queue);
2748
2749 #endif
2750 static void netdev_unbind_all_sb_channels(struct net_device *dev)
2751 {
2752         struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2753
2754         /* Unbind any subordinate channels */
2755         while (txq-- != &dev->_tx[0]) {
2756                 if (txq->sb_dev)
2757                         netdev_unbind_sb_channel(dev, txq->sb_dev);
2758         }
2759 }
2760
2761 void netdev_reset_tc(struct net_device *dev)
2762 {
2763 #ifdef CONFIG_XPS
2764         netif_reset_xps_queues_gt(dev, 0);
2765 #endif
2766         netdev_unbind_all_sb_channels(dev);
2767
2768         /* Reset TC configuration of device */
2769         dev->num_tc = 0;
2770         memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq));
2771         memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map));
2772 }
2773 EXPORT_SYMBOL(netdev_reset_tc);
2774
2775 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset)
2776 {
2777         if (tc >= dev->num_tc)
2778                 return -EINVAL;
2779
2780 #ifdef CONFIG_XPS
2781         netif_reset_xps_queues(dev, offset, count);
2782 #endif
2783         dev->tc_to_txq[tc].count = count;
2784         dev->tc_to_txq[tc].offset = offset;
2785         return 0;
2786 }
2787 EXPORT_SYMBOL(netdev_set_tc_queue);
2788
2789 int netdev_set_num_tc(struct net_device *dev, u8 num_tc)
2790 {
2791         if (num_tc > TC_MAX_QUEUE)
2792                 return -EINVAL;
2793
2794 #ifdef CONFIG_XPS
2795         netif_reset_xps_queues_gt(dev, 0);
2796 #endif
2797         netdev_unbind_all_sb_channels(dev);
2798
2799         dev->num_tc = num_tc;
2800         return 0;
2801 }
2802 EXPORT_SYMBOL(netdev_set_num_tc);
2803
2804 void netdev_unbind_sb_channel(struct net_device *dev,
2805                               struct net_device *sb_dev)
2806 {
2807         struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2808
2809 #ifdef CONFIG_XPS
2810         netif_reset_xps_queues_gt(sb_dev, 0);
2811 #endif
2812         memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq));
2813         memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map));
2814
2815         while (txq-- != &dev->_tx[0]) {
2816                 if (txq->sb_dev == sb_dev)
2817                         txq->sb_dev = NULL;
2818         }
2819 }
2820 EXPORT_SYMBOL(netdev_unbind_sb_channel);
2821
2822 int netdev_bind_sb_channel_queue(struct net_device *dev,
2823                                  struct net_device *sb_dev,
2824                                  u8 tc, u16 count, u16 offset)
2825 {
2826         /* Make certain the sb_dev and dev are already configured */
2827         if (sb_dev->num_tc >= 0 || tc >= dev->num_tc)
2828                 return -EINVAL;
2829
2830         /* We cannot hand out queues we don't have */
2831         if ((offset + count) > dev->real_num_tx_queues)
2832                 return -EINVAL;
2833
2834         /* Record the mapping */
2835         sb_dev->tc_to_txq[tc].count = count;
2836         sb_dev->tc_to_txq[tc].offset = offset;
2837
2838         /* Provide a way for Tx queue to find the tc_to_txq map or
2839          * XPS map for itself.
2840          */
2841         while (count--)
2842                 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev;
2843
2844         return 0;
2845 }
2846 EXPORT_SYMBOL(netdev_bind_sb_channel_queue);
2847
2848 int netdev_set_sb_channel(struct net_device *dev, u16 channel)
2849 {
2850         /* Do not use a multiqueue device to represent a subordinate channel */
2851         if (netif_is_multiqueue(dev))
2852                 return -ENODEV;
2853
2854         /* We allow channels 1 - 32767 to be used for subordinate channels.
2855          * Channel 0 is meant to be "native" mode and used only to represent
2856          * the main root device. We allow writing 0 to reset the device back
2857          * to normal mode after being used as a subordinate channel.
2858          */
2859         if (channel > S16_MAX)
2860                 return -EINVAL;
2861
2862         dev->num_tc = -channel;
2863
2864         return 0;
2865 }
2866 EXPORT_SYMBOL(netdev_set_sb_channel);
2867
2868 /*
2869  * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2870  * greater than real_num_tx_queues stale skbs on the qdisc must be flushed.
2871  */
2872 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2873 {
2874         bool disabling;
2875         int rc;
2876
2877         disabling = txq < dev->real_num_tx_queues;
2878
2879         if (txq < 1 || txq > dev->num_tx_queues)
2880                 return -EINVAL;
2881
2882         if (dev->reg_state == NETREG_REGISTERED ||
2883             dev->reg_state == NETREG_UNREGISTERING) {
2884                 ASSERT_RTNL();
2885
2886                 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2887                                                   txq);
2888                 if (rc)
2889                         return rc;
2890
2891                 if (dev->num_tc)
2892                         netif_setup_tc(dev, txq);
2893
2894                 dev_qdisc_change_real_num_tx(dev, txq);
2895
2896                 dev->real_num_tx_queues = txq;
2897
2898                 if (disabling) {
2899                         synchronize_net();
2900                         qdisc_reset_all_tx_gt(dev, txq);
2901 #ifdef CONFIG_XPS
2902                         netif_reset_xps_queues_gt(dev, txq);
2903 #endif
2904                 }
2905         } else {
2906                 dev->real_num_tx_queues = txq;
2907         }
2908
2909         return 0;
2910 }
2911 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2912
2913 #ifdef CONFIG_SYSFS
2914 /**
2915  *      netif_set_real_num_rx_queues - set actual number of RX queues used
2916  *      @dev: Network device
2917  *      @rxq: Actual number of RX queues
2918  *
2919  *      This must be called either with the rtnl_lock held or before
2920  *      registration of the net device.  Returns 0 on success, or a
2921  *      negative error code.  If called before registration, it always
2922  *      succeeds.
2923  */
2924 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2925 {
2926         int rc;
2927
2928         if (rxq < 1 || rxq > dev->num_rx_queues)
2929                 return -EINVAL;
2930
2931         if (dev->reg_state == NETREG_REGISTERED) {
2932                 ASSERT_RTNL();
2933
2934                 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2935                                                   rxq);
2936                 if (rc)
2937                         return rc;
2938         }
2939
2940         dev->real_num_rx_queues = rxq;
2941         return 0;
2942 }
2943 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2944 #endif
2945
2946 /**
2947  *      netif_set_real_num_queues - set actual number of RX and TX queues used
2948  *      @dev: Network device
2949  *      @txq: Actual number of TX queues
2950  *      @rxq: Actual number of RX queues
2951  *
2952  *      Set the real number of both TX and RX queues.
2953  *      Does nothing if the number of queues is already correct.
2954  */
2955 int netif_set_real_num_queues(struct net_device *dev,
2956                               unsigned int txq, unsigned int rxq)
2957 {
2958         unsigned int old_rxq = dev->real_num_rx_queues;
2959         int err;
2960
2961         if (txq < 1 || txq > dev->num_tx_queues ||
2962             rxq < 1 || rxq > dev->num_rx_queues)
2963                 return -EINVAL;
2964
2965         /* Start from increases, so the error path only does decreases -
2966          * decreases can't fail.
2967          */
2968         if (rxq > dev->real_num_rx_queues) {
2969                 err = netif_set_real_num_rx_queues(dev, rxq);
2970                 if (err)
2971                         return err;
2972         }
2973         if (txq > dev->real_num_tx_queues) {
2974                 err = netif_set_real_num_tx_queues(dev, txq);
2975                 if (err)
2976                         goto undo_rx;
2977         }
2978         if (rxq < dev->real_num_rx_queues)
2979                 WARN_ON(netif_set_real_num_rx_queues(dev, rxq));
2980         if (txq < dev->real_num_tx_queues)
2981                 WARN_ON(netif_set_real_num_tx_queues(dev, txq));
2982
2983         return 0;
2984 undo_rx:
2985         WARN_ON(netif_set_real_num_rx_queues(dev, old_rxq));
2986         return err;
2987 }
2988 EXPORT_SYMBOL(netif_set_real_num_queues);
2989
2990 /**
2991  * netif_set_tso_max_size() - set the max size of TSO frames supported
2992  * @dev:        netdev to update
2993  * @size:       max skb->len of a TSO frame
2994  *
2995  * Set the limit on the size of TSO super-frames the device can handle.
2996  * Unless explicitly set the stack will assume the value of
2997  * %GSO_LEGACY_MAX_SIZE.
2998  */
2999 void netif_set_tso_max_size(struct net_device *dev, unsigned int size)
3000 {
3001         dev->tso_max_size = min(GSO_MAX_SIZE, size);
3002         if (size < READ_ONCE(dev->gso_max_size))
3003                 netif_set_gso_max_size(dev, size);
3004 }
3005 EXPORT_SYMBOL(netif_set_tso_max_size);
3006
3007 /**
3008  * netif_set_tso_max_segs() - set the max number of segs supported for TSO
3009  * @dev:        netdev to update
3010  * @segs:       max number of TCP segments
3011  *
3012  * Set the limit on the number of TCP segments the device can generate from
3013  * a single TSO super-frame.
3014  * Unless explicitly set the stack will assume the value of %GSO_MAX_SEGS.
3015  */
3016 void netif_set_tso_max_segs(struct net_device *dev, unsigned int segs)
3017 {
3018         dev->tso_max_segs = segs;
3019         if (segs < READ_ONCE(dev->gso_max_segs))
3020                 netif_set_gso_max_segs(dev, segs);
3021 }
3022 EXPORT_SYMBOL(netif_set_tso_max_segs);
3023
3024 /**
3025  * netif_inherit_tso_max() - copy all TSO limits from a lower device to an upper
3026  * @to:         netdev to update
3027  * @from:       netdev from which to copy the limits
3028  */
3029 void netif_inherit_tso_max(struct net_device *to, const struct net_device *from)
3030 {
3031         netif_set_tso_max_size(to, from->tso_max_size);
3032         netif_set_tso_max_segs(to, from->tso_max_segs);
3033 }
3034 EXPORT_SYMBOL(netif_inherit_tso_max);
3035
3036 /**
3037  * netif_get_num_default_rss_queues - default number of RSS queues
3038  *
3039  * Default value is the number of physical cores if there are only 1 or 2, or
3040  * divided by 2 if there are more.
3041  */
3042 int netif_get_num_default_rss_queues(void)
3043 {
3044         cpumask_var_t cpus;
3045         int cpu, count = 0;
3046
3047         if (unlikely(is_kdump_kernel() || !zalloc_cpumask_var(&cpus, GFP_KERNEL)))
3048                 return 1;
3049
3050         cpumask_copy(cpus, cpu_online_mask);
3051         for_each_cpu(cpu, cpus) {
3052                 ++count;
3053                 cpumask_andnot(cpus, cpus, topology_sibling_cpumask(cpu));
3054         }
3055         free_cpumask_var(cpus);
3056
3057         return count > 2 ? DIV_ROUND_UP(count, 2) : count;
3058 }
3059 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
3060
3061 static void __netif_reschedule(struct Qdisc *q)
3062 {
3063         struct softnet_data *sd;
3064         unsigned long flags;
3065
3066         local_irq_save(flags);
3067         sd = this_cpu_ptr(&softnet_data);
3068         q->next_sched = NULL;
3069         *sd->output_queue_tailp = q;
3070         sd->output_queue_tailp = &q->next_sched;
3071         raise_softirq_irqoff(NET_TX_SOFTIRQ);
3072         local_irq_restore(flags);
3073 }
3074
3075 void __netif_schedule(struct Qdisc *q)
3076 {
3077         if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
3078                 __netif_reschedule(q);
3079 }
3080 EXPORT_SYMBOL(__netif_schedule);
3081
3082 struct dev_kfree_skb_cb {
3083         enum skb_free_reason reason;
3084 };
3085
3086 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
3087 {
3088         return (struct dev_kfree_skb_cb *)skb->cb;
3089 }
3090
3091 void netif_schedule_queue(struct netdev_queue *txq)
3092 {
3093         rcu_read_lock();
3094         if (!netif_xmit_stopped(txq)) {
3095                 struct Qdisc *q = rcu_dereference(txq->qdisc);
3096
3097                 __netif_schedule(q);
3098         }
3099         rcu_read_unlock();
3100 }
3101 EXPORT_SYMBOL(netif_schedule_queue);
3102
3103 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
3104 {
3105         if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
3106                 struct Qdisc *q;
3107
3108                 rcu_read_lock();
3109                 q = rcu_dereference(dev_queue->qdisc);
3110                 __netif_schedule(q);
3111                 rcu_read_unlock();
3112         }
3113 }
3114 EXPORT_SYMBOL(netif_tx_wake_queue);
3115
3116 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
3117 {
3118         unsigned long flags;
3119
3120         if (unlikely(!skb))
3121                 return;
3122
3123         if (likely(refcount_read(&skb->users) == 1)) {
3124                 smp_rmb();
3125                 refcount_set(&skb->users, 0);
3126         } else if (likely(!refcount_dec_and_test(&skb->users))) {
3127                 return;
3128         }
3129         get_kfree_skb_cb(skb)->reason = reason;
3130         local_irq_save(flags);
3131         skb->next = __this_cpu_read(softnet_data.completion_queue);
3132         __this_cpu_write(softnet_data.completion_queue, skb);
3133         raise_softirq_irqoff(NET_TX_SOFTIRQ);
3134         local_irq_restore(flags);
3135 }
3136 EXPORT_SYMBOL(__dev_kfree_skb_irq);
3137
3138 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
3139 {
3140         if (in_hardirq() || irqs_disabled())
3141                 __dev_kfree_skb_irq(skb, reason);
3142         else
3143                 dev_kfree_skb(skb);
3144 }
3145 EXPORT_SYMBOL(__dev_kfree_skb_any);
3146
3147
3148 /**
3149  * netif_device_detach - mark device as removed
3150  * @dev: network device
3151  *
3152  * Mark device as removed from system and therefore no longer available.
3153  */
3154 void netif_device_detach(struct net_device *dev)
3155 {
3156         if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
3157             netif_running(dev)) {
3158                 netif_tx_stop_all_queues(dev);
3159         }
3160 }
3161 EXPORT_SYMBOL(netif_device_detach);
3162
3163 /**
3164  * netif_device_attach - mark device as attached
3165  * @dev: network device
3166  *
3167  * Mark device as attached from system and restart if needed.
3168  */
3169 void netif_device_attach(struct net_device *dev)
3170 {
3171         if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
3172             netif_running(dev)) {
3173                 netif_tx_wake_all_queues(dev);
3174                 __netdev_watchdog_up(dev);
3175         }
3176 }
3177 EXPORT_SYMBOL(netif_device_attach);
3178
3179 /*
3180  * Returns a Tx hash based on the given packet descriptor a Tx queues' number
3181  * to be used as a distribution range.
3182  */
3183 static u16 skb_tx_hash(const struct net_device *dev,
3184                        const struct net_device *sb_dev,
3185                        struct sk_buff *skb)
3186 {
3187         u32 hash;
3188         u16 qoffset = 0;
3189         u16 qcount = dev->real_num_tx_queues;
3190
3191         if (dev->num_tc) {
3192                 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
3193
3194                 qoffset = sb_dev->tc_to_txq[tc].offset;
3195                 qcount = sb_dev->tc_to_txq[tc].count;
3196                 if (unlikely(!qcount)) {
3197                         net_warn_ratelimited("%s: invalid qcount, qoffset %u for tc %u\n",
3198                                              sb_dev->name, qoffset, tc);
3199                         qoffset = 0;
3200                         qcount = dev->real_num_tx_queues;
3201                 }
3202         }
3203
3204         if (skb_rx_queue_recorded(skb)) {
3205                 hash = skb_get_rx_queue(skb);
3206                 if (hash >= qoffset)
3207                         hash -= qoffset;
3208                 while (unlikely(hash >= qcount))
3209                         hash -= qcount;
3210                 return hash + qoffset;
3211         }
3212
3213         return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
3214 }
3215
3216 static void skb_warn_bad_offload(const struct sk_buff *skb)
3217 {
3218         static const netdev_features_t null_features;
3219         struct net_device *dev = skb->dev;
3220         const char *name = "";
3221
3222         if (!net_ratelimit())
3223                 return;
3224
3225         if (dev) {
3226                 if (dev->dev.parent)
3227                         name = dev_driver_string(dev->dev.parent);
3228                 else
3229                         name = netdev_name(dev);
3230         }
3231         skb_dump(KERN_WARNING, skb, false);
3232         WARN(1, "%s: caps=(%pNF, %pNF)\n",
3233              name, dev ? &dev->features : &null_features,
3234              skb->sk ? &skb->sk->sk_route_caps : &null_features);
3235 }
3236
3237 /*
3238  * Invalidate hardware checksum when packet is to be mangled, and
3239  * complete checksum manually on outgoing path.
3240  */
3241 int skb_checksum_help(struct sk_buff *skb)
3242 {
3243         __wsum csum;
3244         int ret = 0, offset;
3245
3246         if (skb->ip_summed == CHECKSUM_COMPLETE)
3247                 goto out_set_summed;
3248
3249         if (unlikely(skb_is_gso(skb))) {
3250                 skb_warn_bad_offload(skb);
3251                 return -EINVAL;
3252         }
3253
3254         /* Before computing a checksum, we should make sure no frag could
3255          * be modified by an external entity : checksum could be wrong.
3256          */
3257         if (skb_has_shared_frag(skb)) {
3258                 ret = __skb_linearize(skb);
3259                 if (ret)
3260                         goto out;
3261         }
3262
3263         offset = skb_checksum_start_offset(skb);
3264         ret = -EINVAL;
3265         if (WARN_ON_ONCE(offset >= skb_headlen(skb))) {
3266                 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false);
3267                 goto out;
3268         }
3269         csum = skb_checksum(skb, offset, skb->len - offset, 0);
3270
3271         offset += skb->csum_offset;
3272         if (WARN_ON_ONCE(offset + sizeof(__sum16) > skb_headlen(skb))) {
3273                 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false);
3274                 goto out;
3275         }
3276         ret = skb_ensure_writable(skb, offset + sizeof(__sum16));
3277         if (ret)
3278                 goto out;
3279
3280         *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0;
3281 out_set_summed:
3282         skb->ip_summed = CHECKSUM_NONE;
3283 out:
3284         return ret;
3285 }
3286 EXPORT_SYMBOL(skb_checksum_help);
3287
3288 int skb_crc32c_csum_help(struct sk_buff *skb)
3289 {
3290         __le32 crc32c_csum;
3291         int ret = 0, offset, start;
3292
3293         if (skb->ip_summed != CHECKSUM_PARTIAL)
3294                 goto out;
3295
3296         if (unlikely(skb_is_gso(skb)))
3297                 goto out;
3298
3299         /* Before computing a checksum, we should make sure no frag could
3300          * be modified by an external entity : checksum could be wrong.
3301          */
3302         if (unlikely(skb_has_shared_frag(skb))) {
3303                 ret = __skb_linearize(skb);
3304                 if (ret)
3305                         goto out;
3306         }
3307         start = skb_checksum_start_offset(skb);
3308         offset = start + offsetof(struct sctphdr, checksum);
3309         if (WARN_ON_ONCE(offset >= skb_headlen(skb))) {
3310                 ret = -EINVAL;
3311                 goto out;
3312         }
3313
3314         ret = skb_ensure_writable(skb, offset + sizeof(__le32));
3315         if (ret)
3316                 goto out;
3317
3318         crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start,
3319                                                   skb->len - start, ~(__u32)0,
3320                                                   crc32c_csum_stub));
3321         *(__le32 *)(skb->data + offset) = crc32c_csum;
3322         skb->ip_summed = CHECKSUM_NONE;
3323         skb->csum_not_inet = 0;
3324 out:
3325         return ret;
3326 }
3327
3328 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
3329 {
3330         __be16 type = skb->protocol;
3331
3332         /* Tunnel gso handlers can set protocol to ethernet. */
3333         if (type == htons(ETH_P_TEB)) {
3334                 struct ethhdr *eth;
3335
3336                 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
3337                         return 0;
3338
3339                 eth = (struct ethhdr *)skb->data;
3340                 type = eth->h_proto;
3341         }
3342
3343         return __vlan_get_protocol(skb, type, depth);
3344 }
3345
3346 /* openvswitch calls this on rx path, so we need a different check.
3347  */
3348 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
3349 {
3350         if (tx_path)
3351                 return skb->ip_summed != CHECKSUM_PARTIAL &&
3352                        skb->ip_summed != CHECKSUM_UNNECESSARY;
3353
3354         return skb->ip_summed == CHECKSUM_NONE;
3355 }
3356
3357 /**
3358  *      __skb_gso_segment - Perform segmentation on skb.
3359  *      @skb: buffer to segment
3360  *      @features: features for the output path (see dev->features)
3361  *      @tx_path: whether it is called in TX path
3362  *
3363  *      This function segments the given skb and returns a list of segments.
3364  *
3365  *      It may return NULL if the skb requires no segmentation.  This is
3366  *      only possible when GSO is used for verifying header integrity.
3367  *
3368  *      Segmentation preserves SKB_GSO_CB_OFFSET bytes of previous skb cb.
3369  */
3370 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
3371                                   netdev_features_t features, bool tx_path)
3372 {
3373         struct sk_buff *segs;
3374
3375         if (unlikely(skb_needs_check(skb, tx_path))) {
3376                 int err;
3377
3378                 /* We're going to init ->check field in TCP or UDP header */
3379                 err = skb_cow_head(skb, 0);
3380                 if (err < 0)
3381                         return ERR_PTR(err);
3382         }
3383
3384         /* Only report GSO partial support if it will enable us to
3385          * support segmentation on this frame without needing additional
3386          * work.
3387          */
3388         if (features & NETIF_F_GSO_PARTIAL) {
3389                 netdev_features_t partial_features = NETIF_F_GSO_ROBUST;
3390                 struct net_device *dev = skb->dev;
3391
3392                 partial_features |= dev->features & dev->gso_partial_features;
3393                 if (!skb_gso_ok(skb, features | partial_features))
3394                         features &= ~NETIF_F_GSO_PARTIAL;
3395         }
3396
3397         BUILD_BUG_ON(SKB_GSO_CB_OFFSET +
3398                      sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb));
3399
3400         SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
3401         SKB_GSO_CB(skb)->encap_level = 0;
3402
3403         skb_reset_mac_header(skb);
3404         skb_reset_mac_len(skb);
3405
3406         segs = skb_mac_gso_segment(skb, features);
3407
3408         if (segs != skb && unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs)))
3409                 skb_warn_bad_offload(skb);
3410
3411         return segs;
3412 }
3413 EXPORT_SYMBOL(__skb_gso_segment);
3414
3415 /* Take action when hardware reception checksum errors are detected. */
3416 #ifdef CONFIG_BUG
3417 static void do_netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3418 {
3419         netdev_err(dev, "hw csum failure\n");
3420         skb_dump(KERN_ERR, skb, true);
3421         dump_stack();
3422 }
3423
3424 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3425 {
3426         DO_ONCE_LITE(do_netdev_rx_csum_fault, dev, skb);
3427 }
3428 EXPORT_SYMBOL(netdev_rx_csum_fault);
3429 #endif
3430
3431 /* XXX: check that highmem exists at all on the given machine. */
3432 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
3433 {
3434 #ifdef CONFIG_HIGHMEM
3435         int i;
3436
3437         if (!(dev->features & NETIF_F_HIGHDMA)) {
3438                 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3439                         skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3440
3441                         if (PageHighMem(skb_frag_page(frag)))
3442                                 return 1;
3443                 }
3444         }
3445 #endif
3446         return 0;
3447 }
3448
3449 /* If MPLS offload request, verify we are testing hardware MPLS features
3450  * instead of standard features for the netdev.
3451  */
3452 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
3453 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3454                                            netdev_features_t features,
3455                                            __be16 type)
3456 {
3457         if (eth_p_mpls(type))
3458                 features &= skb->dev->mpls_features;
3459
3460         return features;
3461 }
3462 #else
3463 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3464                                            netdev_features_t features,
3465                                            __be16 type)
3466 {
3467         return features;
3468 }
3469 #endif
3470
3471 static netdev_features_t harmonize_features(struct sk_buff *skb,
3472         netdev_features_t features)
3473 {
3474         __be16 type;
3475
3476         type = skb_network_protocol(skb, NULL);
3477         features = net_mpls_features(skb, features, type);
3478
3479         if (skb->ip_summed != CHECKSUM_NONE &&
3480             !can_checksum_protocol(features, type)) {
3481                 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3482         }
3483         if (illegal_highdma(skb->dev, skb))
3484                 features &= ~NETIF_F_SG;
3485
3486         return features;
3487 }
3488
3489 netdev_features_t passthru_features_check(struct sk_buff *skb,
3490                                           struct net_device *dev,
3491                                           netdev_features_t features)
3492 {
3493         return features;
3494 }
3495 EXPORT_SYMBOL(passthru_features_check);
3496
3497 static netdev_features_t dflt_features_check(struct sk_buff *skb,
3498                                              struct net_device *dev,
3499                                              netdev_features_t features)
3500 {
3501         return vlan_features_check(skb, features);
3502 }
3503
3504 static netdev_features_t gso_features_check(const struct sk_buff *skb,
3505                                             struct net_device *dev,
3506                                             netdev_features_t features)
3507 {
3508         u16 gso_segs = skb_shinfo(skb)->gso_segs;
3509
3510         if (gso_segs > READ_ONCE(dev->gso_max_segs))
3511                 return features & ~NETIF_F_GSO_MASK;
3512
3513         if (!skb_shinfo(skb)->gso_type) {
3514                 skb_warn_bad_offload(skb);
3515                 return features & ~NETIF_F_GSO_MASK;
3516         }
3517
3518         /* Support for GSO partial features requires software
3519          * intervention before we can actually process the packets
3520          * so we need to strip support for any partial features now
3521          * and we can pull them back in after we have partially
3522          * segmented the frame.
3523          */
3524         if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL))
3525                 features &= ~dev->gso_partial_features;
3526
3527         /* Make sure to clear the IPv4 ID mangling feature if the
3528          * IPv4 header has the potential to be fragmented.
3529          */
3530         if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
3531                 struct iphdr *iph = skb->encapsulation ?
3532                                     inner_ip_hdr(skb) : ip_hdr(skb);
3533
3534                 if (!(iph->frag_off & htons(IP_DF)))
3535                         features &= ~NETIF_F_TSO_MANGLEID;
3536         }
3537
3538         return features;
3539 }
3540
3541 netdev_features_t netif_skb_features(struct sk_buff *skb)
3542 {
3543         struct net_device *dev = skb->dev;
3544         netdev_features_t features = dev->features;
3545
3546         if (skb_is_gso(skb))
3547                 features = gso_features_check(skb, dev, features);
3548
3549         /* If encapsulation offload request, verify we are testing
3550          * hardware encapsulation features instead of standard
3551          * features for the netdev
3552          */
3553         if (skb->encapsulation)
3554                 features &= dev->hw_enc_features;
3555
3556         if (skb_vlan_tagged(skb))
3557                 features = netdev_intersect_features(features,
3558                                                      dev->vlan_features |
3559                                                      NETIF_F_HW_VLAN_CTAG_TX |
3560                                                      NETIF_F_HW_VLAN_STAG_TX);
3561
3562         if (dev->netdev_ops->ndo_features_check)
3563                 features &= dev->netdev_ops->ndo_features_check(skb, dev,
3564                                                                 features);
3565         else
3566                 features &= dflt_features_check(skb, dev, features);
3567
3568         return harmonize_features(skb, features);
3569 }
3570 EXPORT_SYMBOL(netif_skb_features);
3571
3572 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
3573                     struct netdev_queue *txq, bool more)
3574 {
3575         unsigned int len;
3576         int rc;
3577
3578         if (dev_nit_active(dev))
3579                 dev_queue_xmit_nit(skb, dev);
3580
3581         len = skb->len;
3582         trace_net_dev_start_xmit(skb, dev);
3583         rc = netdev_start_xmit(skb, dev, txq, more);
3584         trace_net_dev_xmit(skb, rc, dev, len);
3585
3586         return rc;
3587 }
3588
3589 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
3590                                     struct netdev_queue *txq, int *ret)
3591 {
3592         struct sk_buff *skb = first;
3593         int rc = NETDEV_TX_OK;
3594
3595         while (skb) {
3596                 struct sk_buff *next = skb->next;
3597
3598                 skb_mark_not_on_list(skb);
3599                 rc = xmit_one(skb, dev, txq, next != NULL);
3600                 if (unlikely(!dev_xmit_complete(rc))) {
3601                         skb->next = next;
3602                         goto out;
3603                 }
3604
3605                 skb = next;
3606                 if (netif_tx_queue_stopped(txq) && skb) {
3607                         rc = NETDEV_TX_BUSY;
3608                         break;
3609                 }
3610         }
3611
3612 out:
3613         *ret = rc;
3614         return skb;
3615 }
3616
3617 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
3618                                           netdev_features_t features)
3619 {
3620         if (skb_vlan_tag_present(skb) &&
3621             !vlan_hw_offload_capable(features, skb->vlan_proto))
3622                 skb = __vlan_hwaccel_push_inside(skb);
3623         return skb;
3624 }
3625
3626 int skb_csum_hwoffload_help(struct sk_buff *skb,
3627                             const netdev_features_t features)
3628 {
3629         if (unlikely(skb_csum_is_sctp(skb)))
3630                 return !!(features & NETIF_F_SCTP_CRC) ? 0 :
3631                         skb_crc32c_csum_help(skb);
3632
3633         if (features & NETIF_F_HW_CSUM)
3634                 return 0;
3635
3636         if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) {
3637                 switch (skb->csum_offset) {
3638                 case offsetof(struct tcphdr, check):
3639                 case offsetof(struct udphdr, check):
3640                         return 0;
3641                 }
3642         }
3643
3644         return skb_checksum_help(skb);
3645 }
3646 EXPORT_SYMBOL(skb_csum_hwoffload_help);
3647
3648 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again)
3649 {
3650         netdev_features_t features;
3651
3652         features = netif_skb_features(skb);
3653         skb = validate_xmit_vlan(skb, features);
3654         if (unlikely(!skb))
3655                 goto out_null;
3656
3657         skb = sk_validate_xmit_skb(skb, dev);
3658         if (unlikely(!skb))
3659                 goto out_null;
3660
3661         if (netif_needs_gso(skb, features)) {
3662                 struct sk_buff *segs;
3663
3664                 segs = skb_gso_segment(skb, features);
3665                 if (IS_ERR(segs)) {
3666                         goto out_kfree_skb;
3667                 } else if (segs) {
3668                         consume_skb(skb);
3669                         skb = segs;
3670                 }
3671         } else {
3672                 if (skb_needs_linearize(skb, features) &&
3673                     __skb_linearize(skb))
3674                         goto out_kfree_skb;
3675
3676                 /* If packet is not checksummed and device does not
3677                  * support checksumming for this protocol, complete
3678                  * checksumming here.
3679                  */
3680                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3681                         if (skb->encapsulation)
3682                                 skb_set_inner_transport_header(skb,
3683                                                                skb_checksum_start_offset(skb));
3684                         else
3685                                 skb_set_transport_header(skb,
3686                                                          skb_checksum_start_offset(skb));
3687                         if (skb_csum_hwoffload_help(skb, features))
3688                                 goto out_kfree_skb;
3689                 }
3690         }
3691
3692         skb = validate_xmit_xfrm(skb, features, again);
3693
3694         return skb;
3695
3696 out_kfree_skb:
3697         kfree_skb(skb);
3698 out_null:
3699         dev_core_stats_tx_dropped_inc(dev);
3700         return NULL;
3701 }
3702
3703 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again)
3704 {
3705         struct sk_buff *next, *head = NULL, *tail;
3706
3707         for (; skb != NULL; skb = next) {
3708                 next = skb->next;
3709                 skb_mark_not_on_list(skb);
3710
3711                 /* in case skb wont be segmented, point to itself */
3712                 skb->prev = skb;
3713
3714                 skb = validate_xmit_skb(skb, dev, again);
3715                 if (!skb)
3716                         continue;
3717
3718                 if (!head)
3719                         head = skb;
3720                 else
3721                         tail->next = skb;
3722                 /* If skb was segmented, skb->prev points to
3723                  * the last segment. If not, it still contains skb.
3724                  */
3725                 tail = skb->prev;
3726         }
3727         return head;
3728 }
3729 EXPORT_SYMBOL_GPL(validate_xmit_skb_list);
3730
3731 static void qdisc_pkt_len_init(struct sk_buff *skb)
3732 {
3733         const struct skb_shared_info *shinfo = skb_shinfo(skb);
3734
3735         qdisc_skb_cb(skb)->pkt_len = skb->len;
3736
3737         /* To get more precise estimation of bytes sent on wire,
3738          * we add to pkt_len the headers size of all segments
3739          */
3740         if (shinfo->gso_size && skb_transport_header_was_set(skb)) {
3741                 unsigned int hdr_len;
3742                 u16 gso_segs = shinfo->gso_segs;
3743
3744                 /* mac layer + network layer */
3745                 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
3746
3747                 /* + transport layer */
3748                 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
3749                         const struct tcphdr *th;
3750                         struct tcphdr _tcphdr;
3751
3752                         th = skb_header_pointer(skb, skb_transport_offset(skb),
3753                                                 sizeof(_tcphdr), &_tcphdr);
3754                         if (likely(th))
3755                                 hdr_len += __tcp_hdrlen(th);
3756                 } else {
3757                         struct udphdr _udphdr;
3758
3759                         if (skb_header_pointer(skb, skb_transport_offset(skb),
3760                                                sizeof(_udphdr), &_udphdr))
3761                                 hdr_len += sizeof(struct udphdr);
3762                 }
3763
3764                 if (shinfo->gso_type & SKB_GSO_DODGY)
3765                         gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
3766                                                 shinfo->gso_size);
3767
3768                 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
3769         }
3770 }
3771
3772 static int dev_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *q,
3773                              struct sk_buff **to_free,
3774                              struct netdev_queue *txq)
3775 {
3776         int rc;
3777
3778         rc = q->enqueue(skb, q, to_free) & NET_XMIT_MASK;
3779         if (rc == NET_XMIT_SUCCESS)
3780                 trace_qdisc_enqueue(q, txq, skb);
3781         return rc;
3782 }
3783
3784 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
3785                                  struct net_device *dev,
3786                                  struct netdev_queue *txq)
3787 {
3788         spinlock_t *root_lock = qdisc_lock(q);
3789         struct sk_buff *to_free = NULL;
3790         bool contended;
3791         int rc;
3792
3793         qdisc_calculate_pkt_len(skb, q);
3794
3795         if (q->flags & TCQ_F_NOLOCK) {
3796                 if (q->flags & TCQ_F_CAN_BYPASS && nolock_qdisc_is_empty(q) &&
3797                     qdisc_run_begin(q)) {
3798                         /* Retest nolock_qdisc_is_empty() within the protection
3799                          * of q->seqlock to protect from racing with requeuing.
3800                          */
3801                         if (unlikely(!nolock_qdisc_is_empty(q))) {
3802                                 rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3803                                 __qdisc_run(q);
3804                                 qdisc_run_end(q);
3805
3806                                 goto no_lock_out;
3807                         }
3808
3809                         qdisc_bstats_cpu_update(q, skb);
3810                         if (sch_direct_xmit(skb, q, dev, txq, NULL, true) &&
3811                             !nolock_qdisc_is_empty(q))
3812                                 __qdisc_run(q);
3813
3814                         qdisc_run_end(q);
3815                         return NET_XMIT_SUCCESS;
3816                 }
3817
3818                 rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3819                 qdisc_run(q);
3820
3821 no_lock_out:
3822                 if (unlikely(to_free))
3823                         kfree_skb_list_reason(to_free,
3824                                               SKB_DROP_REASON_QDISC_DROP);
3825                 return rc;
3826         }
3827
3828         /*
3829          * Heuristic to force contended enqueues to serialize on a
3830          * separate lock before trying to get qdisc main lock.
3831          * This permits qdisc->running owner to get the lock more
3832          * often and dequeue packets faster.
3833          * On PREEMPT_RT it is possible to preempt the qdisc owner during xmit
3834          * and then other tasks will only enqueue packets. The packets will be
3835          * sent after the qdisc owner is scheduled again. To prevent this
3836          * scenario the task always serialize on the lock.
3837          */
3838         contended = qdisc_is_running(q) || IS_ENABLED(CONFIG_PREEMPT_RT);
3839         if (unlikely(contended))
3840                 spin_lock(&q->busylock);
3841
3842         spin_lock(root_lock);
3843         if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3844                 __qdisc_drop(skb, &to_free);
3845                 rc = NET_XMIT_DROP;
3846         } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
3847                    qdisc_run_begin(q)) {
3848                 /*
3849                  * This is a work-conserving queue; there are no old skbs
3850                  * waiting to be sent out; and the qdisc is not running -
3851                  * xmit the skb directly.
3852                  */
3853
3854                 qdisc_bstats_update(q, skb);
3855
3856                 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
3857                         if (unlikely(contended)) {
3858                                 spin_unlock(&q->busylock);
3859                                 contended = false;
3860                         }
3861                         __qdisc_run(q);
3862                 }
3863
3864                 qdisc_run_end(q);
3865                 rc = NET_XMIT_SUCCESS;
3866         } else {
3867                 rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3868                 if (qdisc_run_begin(q)) {
3869                         if (unlikely(contended)) {
3870                                 spin_unlock(&q->busylock);
3871                                 contended = false;
3872                         }
3873                         __qdisc_run(q);
3874                         qdisc_run_end(q);
3875                 }
3876         }
3877         spin_unlock(root_lock);
3878         if (unlikely(to_free))
3879                 kfree_skb_list_reason(to_free, SKB_DROP_REASON_QDISC_DROP);
3880         if (unlikely(contended))
3881                 spin_unlock(&q->busylock);
3882         return rc;
3883 }
3884
3885 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
3886 static void skb_update_prio(struct sk_buff *skb)
3887 {
3888         const struct netprio_map *map;
3889         const struct sock *sk;
3890         unsigned int prioidx;
3891
3892         if (skb->priority)
3893                 return;
3894         map = rcu_dereference_bh(skb->dev->priomap);
3895         if (!map)
3896                 return;
3897         sk = skb_to_full_sk(skb);
3898         if (!sk)
3899                 return;
3900
3901         prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data);
3902
3903         if (prioidx < map->priomap_len)
3904                 skb->priority = map->priomap[prioidx];
3905 }
3906 #else
3907 #define skb_update_prio(skb)
3908 #endif
3909
3910 /**
3911  *      dev_loopback_xmit - loop back @skb
3912  *      @net: network namespace this loopback is happening in
3913  *      @sk:  sk needed to be a netfilter okfn
3914  *      @skb: buffer to transmit
3915  */
3916 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
3917 {
3918         skb_reset_mac_header(skb);
3919         __skb_pull(skb, skb_network_offset(skb));
3920         skb->pkt_type = PACKET_LOOPBACK;
3921         if (skb->ip_summed == CHECKSUM_NONE)
3922                 skb->ip_summed = CHECKSUM_UNNECESSARY;
3923         DEBUG_NET_WARN_ON_ONCE(!skb_dst(skb));
3924         skb_dst_force(skb);
3925         netif_rx(skb);
3926         return 0;
3927 }
3928 EXPORT_SYMBOL(dev_loopback_xmit);
3929
3930 #ifdef CONFIG_NET_EGRESS
3931 static struct sk_buff *
3932 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
3933 {
3934 #ifdef CONFIG_NET_CLS_ACT
3935         struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress);
3936         struct tcf_result cl_res;
3937
3938         if (!miniq)
3939                 return skb;
3940
3941         /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */
3942         tc_skb_cb(skb)->mru = 0;
3943         tc_skb_cb(skb)->post_ct = false;
3944         mini_qdisc_bstats_cpu_update(miniq, skb);
3945
3946         switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) {
3947         case TC_ACT_OK:
3948         case TC_ACT_RECLASSIFY:
3949                 skb->tc_index = TC_H_MIN(cl_res.classid);
3950                 break;
3951         case TC_ACT_SHOT:
3952                 mini_qdisc_qstats_cpu_drop(miniq);
3953                 *ret = NET_XMIT_DROP;
3954                 kfree_skb_reason(skb, SKB_DROP_REASON_TC_EGRESS);
3955                 return NULL;
3956         case TC_ACT_STOLEN:
3957         case TC_ACT_QUEUED:
3958         case TC_ACT_TRAP:
3959                 *ret = NET_XMIT_SUCCESS;
3960                 consume_skb(skb);
3961                 return NULL;
3962         case TC_ACT_REDIRECT:
3963                 /* No need to push/pop skb's mac_header here on egress! */
3964                 skb_do_redirect(skb);
3965                 *ret = NET_XMIT_SUCCESS;
3966                 return NULL;
3967         default:
3968                 break;
3969         }
3970 #endif /* CONFIG_NET_CLS_ACT */
3971
3972         return skb;
3973 }
3974
3975 static struct netdev_queue *
3976 netdev_tx_queue_mapping(struct net_device *dev, struct sk_buff *skb)
3977 {
3978         int qm = skb_get_queue_mapping(skb);
3979
3980         return netdev_get_tx_queue(dev, netdev_cap_txqueue(dev, qm));
3981 }
3982
3983 static bool netdev_xmit_txqueue_skipped(void)
3984 {
3985         return __this_cpu_read(softnet_data.xmit.skip_txqueue);
3986 }
3987
3988 void netdev_xmit_skip_txqueue(bool skip)
3989 {
3990         __this_cpu_write(softnet_data.xmit.skip_txqueue, skip);
3991 }
3992 EXPORT_SYMBOL_GPL(netdev_xmit_skip_txqueue);
3993 #endif /* CONFIG_NET_EGRESS */
3994
3995 #ifdef CONFIG_XPS
3996 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb,
3997                                struct xps_dev_maps *dev_maps, unsigned int tci)
3998 {
3999         int tc = netdev_get_prio_tc_map(dev, skb->priority);
4000         struct xps_map *map;
4001         int queue_index = -1;
4002
4003         if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids)
4004                 return queue_index;
4005
4006         tci *= dev_maps->num_tc;
4007         tci += tc;
4008
4009         map = rcu_dereference(dev_maps->attr_map[tci]);
4010         if (map) {
4011                 if (map->len == 1)
4012                         queue_index = map->queues[0];
4013                 else
4014                         queue_index = map->queues[reciprocal_scale(
4015                                                 skb_get_hash(skb), map->len)];
4016                 if (unlikely(queue_index >= dev->real_num_tx_queues))
4017                         queue_index = -1;
4018         }
4019         return queue_index;
4020 }
4021 #endif
4022
4023 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev,
4024                          struct sk_buff *skb)
4025 {
4026 #ifdef CONFIG_XPS
4027         struct xps_dev_maps *dev_maps;
4028         struct sock *sk = skb->sk;
4029         int queue_index = -1;
4030
4031         if (!static_key_false(&xps_needed))
4032                 return -1;
4033
4034         rcu_read_lock();
4035         if (!static_key_false(&xps_rxqs_needed))
4036                 goto get_cpus_map;
4037
4038         dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]);
4039         if (dev_maps) {
4040                 int tci = sk_rx_queue_get(sk);
4041
4042                 if (tci >= 0)
4043                         queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
4044                                                           tci);
4045         }
4046
4047 get_cpus_map:
4048         if (queue_index < 0) {
4049                 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]);
4050                 if (dev_maps) {
4051                         unsigned int tci = skb->sender_cpu - 1;
4052
4053                         queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
4054                                                           tci);
4055                 }
4056         }
4057         rcu_read_unlock();
4058
4059         return queue_index;
4060 #else
4061         return -1;
4062 #endif
4063 }
4064
4065 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb,
4066                      struct net_device *sb_dev)
4067 {
4068         return 0;
4069 }
4070 EXPORT_SYMBOL(dev_pick_tx_zero);
4071
4072 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb,
4073                        struct net_device *sb_dev)
4074 {
4075         return (u16)raw_smp_processor_id() % dev->real_num_tx_queues;
4076 }
4077 EXPORT_SYMBOL(dev_pick_tx_cpu_id);
4078
4079 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb,
4080                      struct net_device *sb_dev)
4081 {
4082         struct sock *sk = skb->sk;
4083         int queue_index = sk_tx_queue_get(sk);
4084
4085         sb_dev = sb_dev ? : dev;
4086
4087         if (queue_index < 0 || skb->ooo_okay ||
4088             queue_index >= dev->real_num_tx_queues) {
4089                 int new_index = get_xps_queue(dev, sb_dev, skb);
4090
4091                 if (new_index < 0)
4092                         new_index = skb_tx_hash(dev, sb_dev, skb);
4093
4094                 if (queue_index != new_index && sk &&
4095                     sk_fullsock(sk) &&
4096                     rcu_access_pointer(sk->sk_dst_cache))
4097                         sk_tx_queue_set(sk, new_index);
4098
4099                 queue_index = new_index;
4100         }
4101
4102         return queue_index;
4103 }
4104 EXPORT_SYMBOL(netdev_pick_tx);
4105
4106 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev,
4107                                          struct sk_buff *skb,
4108                                          struct net_device *sb_dev)
4109 {
4110         int queue_index = 0;
4111
4112 #ifdef CONFIG_XPS
4113         u32 sender_cpu = skb->sender_cpu - 1;
4114
4115         if (sender_cpu >= (u32)NR_CPUS)
4116                 skb->sender_cpu = raw_smp_processor_id() + 1;
4117 #endif
4118
4119         if (dev->real_num_tx_queues != 1) {
4120                 const struct net_device_ops *ops = dev->netdev_ops;
4121
4122                 if (ops->ndo_select_queue)
4123                         queue_index = ops->ndo_select_queue(dev, skb, sb_dev);
4124                 else
4125                         queue_index = netdev_pick_tx(dev, skb, sb_dev);
4126
4127                 queue_index = netdev_cap_txqueue(dev, queue_index);
4128         }
4129
4130         skb_set_queue_mapping(skb, queue_index);
4131         return netdev_get_tx_queue(dev, queue_index);
4132 }
4133
4134 /**
4135  * __dev_queue_xmit() - transmit a buffer
4136  * @skb:        buffer to transmit
4137  * @sb_dev:     suboordinate device used for L2 forwarding offload
4138  *
4139  * Queue a buffer for transmission to a network device. The caller must
4140  * have set the device and priority and built the buffer before calling
4141  * this function. The function can be called from an interrupt.
4142  *
4143  * When calling this method, interrupts MUST be enabled. This is because
4144  * the BH enable code must have IRQs enabled so that it will not deadlock.
4145  *
4146  * Regardless of the return value, the skb is consumed, so it is currently
4147  * difficult to retry a send to this method. (You can bump the ref count
4148  * before sending to hold a reference for retry if you are careful.)
4149  *
4150  * Return:
4151  * * 0                          - buffer successfully transmitted
4152  * * positive qdisc return code - NET_XMIT_DROP etc.
4153  * * negative errno             - other errors
4154  */
4155 int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev)
4156 {
4157         struct net_device *dev = skb->dev;
4158         struct netdev_queue *txq = NULL;
4159         struct Qdisc *q;
4160         int rc = -ENOMEM;
4161         bool again = false;
4162
4163         skb_reset_mac_header(skb);
4164         skb_assert_len(skb);
4165
4166         if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
4167                 __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED);
4168
4169         /* Disable soft irqs for various locks below. Also
4170          * stops preemption for RCU.
4171          */
4172         rcu_read_lock_bh();
4173
4174         skb_update_prio(skb);
4175
4176         qdisc_pkt_len_init(skb);
4177 #ifdef CONFIG_NET_CLS_ACT
4178         skb->tc_at_ingress = 0;
4179 #endif
4180 #ifdef CONFIG_NET_EGRESS
4181         if (static_branch_unlikely(&egress_needed_key)) {
4182                 if (nf_hook_egress_active()) {
4183                         skb = nf_hook_egress(skb, &rc, dev);
4184                         if (!skb)
4185                                 goto out;
4186                 }
4187
4188                 netdev_xmit_skip_txqueue(false);
4189
4190                 nf_skip_egress(skb, true);
4191                 skb = sch_handle_egress(skb, &rc, dev);
4192                 if (!skb)
4193                         goto out;
4194                 nf_skip_egress(skb, false);
4195
4196                 if (netdev_xmit_txqueue_skipped())
4197                         txq = netdev_tx_queue_mapping(dev, skb);
4198         }
4199 #endif
4200         /* If device/qdisc don't need skb->dst, release it right now while
4201          * its hot in this cpu cache.
4202          */
4203         if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
4204                 skb_dst_drop(skb);
4205         else
4206                 skb_dst_force(skb);
4207
4208         if (!txq)
4209                 txq = netdev_core_pick_tx(dev, skb, sb_dev);
4210
4211         q = rcu_dereference_bh(txq->qdisc);
4212
4213         trace_net_dev_queue(skb);
4214         if (q->enqueue) {
4215                 rc = __dev_xmit_skb(skb, q, dev, txq);
4216                 goto out;
4217         }
4218
4219         /* The device has no queue. Common case for software devices:
4220          * loopback, all the sorts of tunnels...
4221
4222          * Really, it is unlikely that netif_tx_lock protection is necessary
4223          * here.  (f.e. loopback and IP tunnels are clean ignoring statistics
4224          * counters.)
4225          * However, it is possible, that they rely on protection
4226          * made by us here.
4227
4228          * Check this and shot the lock. It is not prone from deadlocks.
4229          *Either shot noqueue qdisc, it is even simpler 8)
4230          */
4231         if (dev->flags & IFF_UP) {
4232                 int cpu = smp_processor_id(); /* ok because BHs are off */
4233
4234                 /* Other cpus might concurrently change txq->xmit_lock_owner
4235                  * to -1 or to their cpu id, but not to our id.
4236                  */
4237                 if (READ_ONCE(txq->xmit_lock_owner) != cpu) {
4238                         if (dev_xmit_recursion())
4239                                 goto recursion_alert;
4240
4241                         skb = validate_xmit_skb(skb, dev, &again);
4242                         if (!skb)
4243                                 goto out;
4244
4245                         HARD_TX_LOCK(dev, txq, cpu);
4246
4247                         if (!netif_xmit_stopped(txq)) {
4248                                 dev_xmit_recursion_inc();
4249                                 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
4250                                 dev_xmit_recursion_dec();
4251                                 if (dev_xmit_complete(rc)) {
4252                                         HARD_TX_UNLOCK(dev, txq);
4253                                         goto out;
4254                                 }
4255                         }
4256                         HARD_TX_UNLOCK(dev, txq);
4257                         net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
4258                                              dev->name);
4259                 } else {
4260                         /* Recursion is detected! It is possible,
4261                          * unfortunately
4262                          */
4263 recursion_alert:
4264                         net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
4265                                              dev->name);
4266                 }
4267         }
4268
4269         rc = -ENETDOWN;
4270         rcu_read_unlock_bh();
4271
4272         dev_core_stats_tx_dropped_inc(dev);
4273         kfree_skb_list(skb);
4274         return rc;
4275 out:
4276         rcu_read_unlock_bh();
4277         return rc;
4278 }
4279 EXPORT_SYMBOL(__dev_queue_xmit);
4280
4281 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id)
4282 {
4283         struct net_device *dev = skb->dev;
4284         struct sk_buff *orig_skb = skb;
4285         struct netdev_queue *txq;
4286         int ret = NETDEV_TX_BUSY;
4287         bool again = false;
4288
4289         if (unlikely(!netif_running(dev) ||
4290                      !netif_carrier_ok(dev)))
4291                 goto drop;
4292
4293         skb = validate_xmit_skb_list(skb, dev, &again);
4294         if (skb != orig_skb)
4295                 goto drop;
4296
4297         skb_set_queue_mapping(skb, queue_id);
4298         txq = skb_get_tx_queue(dev, skb);
4299
4300         local_bh_disable();
4301
4302         dev_xmit_recursion_inc();
4303         HARD_TX_LOCK(dev, txq, smp_processor_id());
4304         if (!netif_xmit_frozen_or_drv_stopped(txq))
4305                 ret = netdev_start_xmit(skb, dev, txq, false);
4306         HARD_TX_UNLOCK(dev, txq);
4307         dev_xmit_recursion_dec();
4308
4309         local_bh_enable();
4310         return ret;
4311 drop:
4312         dev_core_stats_tx_dropped_inc(dev);
4313         kfree_skb_list(skb);
4314         return NET_XMIT_DROP;
4315 }
4316 EXPORT_SYMBOL(__dev_direct_xmit);
4317
4318 /*************************************************************************
4319  *                      Receiver routines
4320  *************************************************************************/
4321
4322 int netdev_max_backlog __read_mostly = 1000;
4323 EXPORT_SYMBOL(netdev_max_backlog);
4324
4325 int netdev_tstamp_prequeue __read_mostly = 1;
4326 unsigned int sysctl_skb_defer_max __read_mostly = 64;
4327 int netdev_budget __read_mostly = 300;
4328 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */
4329 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ;
4330 int weight_p __read_mostly = 64;           /* old backlog weight */
4331 int dev_weight_rx_bias __read_mostly = 1;  /* bias for backlog weight */
4332 int dev_weight_tx_bias __read_mostly = 1;  /* bias for output_queue quota */
4333 int dev_rx_weight __read_mostly = 64;
4334 int dev_tx_weight __read_mostly = 64;
4335
4336 /* Called with irq disabled */
4337 static inline void ____napi_schedule(struct softnet_data *sd,
4338                                      struct napi_struct *napi)
4339 {
4340         struct task_struct *thread;
4341
4342         lockdep_assert_irqs_disabled();
4343
4344         if (test_bit(NAPI_STATE_THREADED, &napi->state)) {
4345                 /* Paired with smp_mb__before_atomic() in
4346                  * napi_enable()/dev_set_threaded().
4347                  * Use READ_ONCE() to guarantee a complete
4348                  * read on napi->thread. Only call
4349                  * wake_up_process() when it's not NULL.
4350                  */
4351                 thread = READ_ONCE(napi->thread);
4352                 if (thread) {
4353                         /* Avoid doing set_bit() if the thread is in
4354                          * INTERRUPTIBLE state, cause napi_thread_wait()
4355                          * makes sure to proceed with napi polling
4356                          * if the thread is explicitly woken from here.
4357                          */
4358                         if (READ_ONCE(thread->__state) != TASK_INTERRUPTIBLE)
4359                                 set_bit(NAPI_STATE_SCHED_THREADED, &napi->state);
4360                         wake_up_process(thread);
4361                         return;
4362                 }
4363         }
4364
4365         list_add_tail(&napi->poll_list, &sd->poll_list);
4366         __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4367 }
4368
4369 #ifdef CONFIG_RPS
4370
4371 /* One global table that all flow-based protocols share. */
4372 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
4373 EXPORT_SYMBOL(rps_sock_flow_table);
4374 u32 rps_cpu_mask __read_mostly;
4375 EXPORT_SYMBOL(rps_cpu_mask);
4376
4377 struct static_key_false rps_needed __read_mostly;
4378 EXPORT_SYMBOL(rps_needed);
4379 struct static_key_false rfs_needed __read_mostly;
4380 EXPORT_SYMBOL(rfs_needed);
4381
4382 static struct rps_dev_flow *
4383 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4384             struct rps_dev_flow *rflow, u16 next_cpu)
4385 {
4386         if (next_cpu < nr_cpu_ids) {
4387 #ifdef CONFIG_RFS_ACCEL
4388                 struct netdev_rx_queue *rxqueue;
4389                 struct rps_dev_flow_table *flow_table;
4390                 struct rps_dev_flow *old_rflow;
4391                 u32 flow_id;
4392                 u16 rxq_index;
4393                 int rc;
4394
4395                 /* Should we steer this flow to a different hardware queue? */
4396                 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
4397                     !(dev->features & NETIF_F_NTUPLE))
4398                         goto out;
4399                 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
4400                 if (rxq_index == skb_get_rx_queue(skb))
4401                         goto out;
4402
4403                 rxqueue = dev->_rx + rxq_index;
4404                 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4405                 if (!flow_table)
4406                         goto out;
4407                 flow_id = skb_get_hash(skb) & flow_table->mask;
4408                 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
4409                                                         rxq_index, flow_id);
4410                 if (rc < 0)
4411                         goto out;
4412                 old_rflow = rflow;
4413                 rflow = &flow_table->flows[flow_id];
4414                 rflow->filter = rc;
4415                 if (old_rflow->filter == rflow->filter)
4416                         old_rflow->filter = RPS_NO_FILTER;
4417         out:
4418 #endif
4419                 rflow->last_qtail =
4420                         per_cpu(softnet_data, next_cpu).input_queue_head;
4421         }
4422
4423         rflow->cpu = next_cpu;
4424         return rflow;
4425 }
4426
4427 /*
4428  * get_rps_cpu is called from netif_receive_skb and returns the target
4429  * CPU from the RPS map of the receiving queue for a given skb.
4430  * rcu_read_lock must be held on entry.
4431  */
4432 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4433                        struct rps_dev_flow **rflowp)
4434 {
4435         const struct rps_sock_flow_table *sock_flow_table;
4436         struct netdev_rx_queue *rxqueue = dev->_rx;
4437         struct rps_dev_flow_table *flow_table;
4438         struct rps_map *map;
4439         int cpu = -1;
4440         u32 tcpu;
4441         u32 hash;
4442
4443         if (skb_rx_queue_recorded(skb)) {
4444                 u16 index = skb_get_rx_queue(skb);
4445
4446                 if (unlikely(index >= dev->real_num_rx_queues)) {
4447                         WARN_ONCE(dev->real_num_rx_queues > 1,
4448                                   "%s received packet on queue %u, but number "
4449                                   "of RX queues is %u\n",
4450                                   dev->name, index, dev->real_num_rx_queues);
4451                         goto done;
4452                 }
4453                 rxqueue += index;
4454         }
4455
4456         /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
4457
4458         flow_table = rcu_dereference(rxqueue->rps_flow_table);
4459         map = rcu_dereference(rxqueue->rps_map);
4460         if (!flow_table && !map)
4461                 goto done;
4462
4463         skb_reset_network_header(skb);
4464         hash = skb_get_hash(skb);
4465         if (!hash)
4466                 goto done;
4467
4468         sock_flow_table = rcu_dereference(rps_sock_flow_table);
4469         if (flow_table && sock_flow_table) {
4470                 struct rps_dev_flow *rflow;
4471                 u32 next_cpu;
4472                 u32 ident;
4473
4474                 /* First check into global flow table if there is a match */
4475                 ident = sock_flow_table->ents[hash & sock_flow_table->mask];
4476                 if ((ident ^ hash) & ~rps_cpu_mask)
4477                         goto try_rps;
4478
4479                 next_cpu = ident & rps_cpu_mask;
4480
4481                 /* OK, now we know there is a match,
4482                  * we can look at the local (per receive queue) flow table
4483                  */
4484                 rflow = &flow_table->flows[hash & flow_table->mask];
4485                 tcpu = rflow->cpu;
4486
4487                 /*
4488                  * If the desired CPU (where last recvmsg was done) is
4489                  * different from current CPU (one in the rx-queue flow
4490                  * table entry), switch if one of the following holds:
4491                  *   - Current CPU is unset (>= nr_cpu_ids).
4492                  *   - Current CPU is offline.
4493                  *   - The current CPU's queue tail has advanced beyond the
4494                  *     last packet that was enqueued using this table entry.
4495                  *     This guarantees that all previous packets for the flow
4496                  *     have been dequeued, thus preserving in order delivery.
4497                  */
4498                 if (unlikely(tcpu != next_cpu) &&
4499                     (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
4500                      ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
4501                       rflow->last_qtail)) >= 0)) {
4502                         tcpu = next_cpu;
4503                         rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
4504                 }
4505
4506                 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
4507                         *rflowp = rflow;
4508                         cpu = tcpu;
4509                         goto done;
4510                 }
4511         }
4512
4513 try_rps:
4514
4515         if (map) {
4516                 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
4517                 if (cpu_online(tcpu)) {
4518                         cpu = tcpu;
4519                         goto done;
4520                 }
4521         }
4522
4523 done:
4524         return cpu;
4525 }
4526
4527 #ifdef CONFIG_RFS_ACCEL
4528
4529 /**
4530  * rps_may_expire_flow - check whether an RFS hardware filter may be removed
4531  * @dev: Device on which the filter was set
4532  * @rxq_index: RX queue index
4533  * @flow_id: Flow ID passed to ndo_rx_flow_steer()
4534  * @filter_id: Filter ID returned by ndo_rx_flow_steer()
4535  *
4536  * Drivers that implement ndo_rx_flow_steer() should periodically call
4537  * this function for each installed filter and remove the filters for
4538  * which it returns %true.
4539  */
4540 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
4541                          u32 flow_id, u16 filter_id)
4542 {
4543         struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
4544         struct rps_dev_flow_table *flow_table;
4545         struct rps_dev_flow *rflow;
4546         bool expire = true;
4547         unsigned int cpu;
4548
4549         rcu_read_lock();
4550         flow_table = rcu_dereference(rxqueue->rps_flow_table);
4551         if (flow_table && flow_id <= flow_table->mask) {
4552                 rflow = &flow_table->flows[flow_id];
4553                 cpu = READ_ONCE(rflow->cpu);
4554                 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
4555                     ((int)(per_cpu(softnet_data, cpu).input_queue_head -
4556                            rflow->last_qtail) <
4557                      (int)(10 * flow_table->mask)))
4558                         expire = false;
4559         }
4560         rcu_read_unlock();
4561         return expire;
4562 }
4563 EXPORT_SYMBOL(rps_may_expire_flow);
4564
4565 #endif /* CONFIG_RFS_ACCEL */
4566
4567 /* Called from hardirq (IPI) context */
4568 static void rps_trigger_softirq(void *data)
4569 {
4570         struct softnet_data *sd = data;
4571
4572         ____napi_schedule(sd, &sd->backlog);
4573         sd->received_rps++;
4574 }
4575
4576 #endif /* CONFIG_RPS */
4577
4578 /* Called from hardirq (IPI) context */
4579 static void trigger_rx_softirq(void *data)
4580 {
4581         struct softnet_data *sd = data;
4582
4583         __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4584         smp_store_release(&sd->defer_ipi_scheduled, 0);
4585 }
4586
4587 /*
4588  * Check if this softnet_data structure is another cpu one
4589  * If yes, queue it to our IPI list and return 1
4590  * If no, return 0
4591  */
4592 static int napi_schedule_rps(struct softnet_data *sd)
4593 {
4594         struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
4595
4596 #ifdef CONFIG_RPS
4597         if (sd != mysd) {
4598                 sd->rps_ipi_next = mysd->rps_ipi_list;
4599                 mysd->rps_ipi_list = sd;
4600
4601                 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4602                 return 1;
4603         }
4604 #endif /* CONFIG_RPS */
4605         __napi_schedule_irqoff(&mysd->backlog);
4606         return 0;
4607 }
4608
4609 #ifdef CONFIG_NET_FLOW_LIMIT
4610 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
4611 #endif
4612
4613 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
4614 {
4615 #ifdef CONFIG_NET_FLOW_LIMIT
4616         struct sd_flow_limit *fl;
4617         struct softnet_data *sd;
4618         unsigned int old_flow, new_flow;
4619
4620         if (qlen < (READ_ONCE(netdev_max_backlog) >> 1))
4621                 return false;
4622
4623         sd = this_cpu_ptr(&softnet_data);
4624
4625         rcu_read_lock();
4626         fl = rcu_dereference(sd->flow_limit);
4627         if (fl) {
4628                 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
4629                 old_flow = fl->history[fl->history_head];
4630                 fl->history[fl->history_head] = new_flow;
4631
4632                 fl->history_head++;
4633                 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
4634
4635                 if (likely(fl->buckets[old_flow]))
4636                         fl->buckets[old_flow]--;
4637
4638                 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
4639                         fl->count++;
4640                         rcu_read_unlock();
4641                         return true;
4642                 }
4643         }
4644         rcu_read_unlock();
4645 #endif
4646         return false;
4647 }
4648
4649 /*
4650  * enqueue_to_backlog is called to queue an skb to a per CPU backlog
4651  * queue (may be a remote CPU queue).
4652  */
4653 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
4654                               unsigned int *qtail)
4655 {
4656         enum skb_drop_reason reason;
4657         struct softnet_data *sd;
4658         unsigned long flags;
4659         unsigned int qlen;
4660
4661         reason = SKB_DROP_REASON_NOT_SPECIFIED;
4662         sd = &per_cpu(softnet_data, cpu);
4663
4664         rps_lock_irqsave(sd, &flags);
4665         if (!netif_running(skb->dev))
4666                 goto drop;
4667         qlen = skb_queue_len(&sd->input_pkt_queue);
4668         if (qlen <= READ_ONCE(netdev_max_backlog) && !skb_flow_limit(skb, qlen)) {
4669                 if (qlen) {
4670 enqueue:
4671                         __skb_queue_tail(&sd->input_pkt_queue, skb);
4672                         input_queue_tail_incr_save(sd, qtail);
4673                         rps_unlock_irq_restore(sd, &flags);
4674                         return NET_RX_SUCCESS;
4675                 }
4676
4677                 /* Schedule NAPI for backlog device
4678                  * We can use non atomic operation since we own the queue lock
4679                  */
4680                 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state))
4681                         napi_schedule_rps(sd);
4682                 goto enqueue;
4683         }
4684         reason = SKB_DROP_REASON_CPU_BACKLOG;
4685
4686 drop:
4687         sd->dropped++;
4688         rps_unlock_irq_restore(sd, &flags);
4689
4690         dev_core_stats_rx_dropped_inc(skb->dev);
4691         kfree_skb_reason(skb, reason);
4692         return NET_RX_DROP;
4693 }
4694
4695 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb)
4696 {
4697         struct net_device *dev = skb->dev;
4698         struct netdev_rx_queue *rxqueue;
4699
4700         rxqueue = dev->_rx;
4701
4702         if (skb_rx_queue_recorded(skb)) {
4703                 u16 index = skb_get_rx_queue(skb);
4704
4705                 if (unlikely(index >= dev->real_num_rx_queues)) {
4706                         WARN_ONCE(dev->real_num_rx_queues > 1,
4707                                   "%s received packet on queue %u, but number "
4708                                   "of RX queues is %u\n",
4709                                   dev->name, index, dev->real_num_rx_queues);
4710
4711                         return rxqueue; /* Return first rxqueue */
4712                 }
4713                 rxqueue += index;
4714         }
4715         return rxqueue;
4716 }
4717
4718 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp,
4719                              struct bpf_prog *xdp_prog)
4720 {
4721         void *orig_data, *orig_data_end, *hard_start;
4722         struct netdev_rx_queue *rxqueue;
4723         bool orig_bcast, orig_host;
4724         u32 mac_len, frame_sz;
4725         __be16 orig_eth_type;
4726         struct ethhdr *eth;
4727         u32 metalen, act;
4728         int off;
4729
4730         /* The XDP program wants to see the packet starting at the MAC
4731          * header.
4732          */
4733         mac_len = skb->data - skb_mac_header(skb);
4734         hard_start = skb->data - skb_headroom(skb);
4735
4736         /* SKB "head" area always have tailroom for skb_shared_info */
4737         frame_sz = (void *)skb_end_pointer(skb) - hard_start;
4738         frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
4739
4740         rxqueue = netif_get_rxqueue(skb);
4741         xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq);
4742         xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len,
4743                          skb_headlen(skb) + mac_len, true);
4744
4745         orig_data_end = xdp->data_end;
4746         orig_data = xdp->data;
4747         eth = (struct ethhdr *)xdp->data;
4748         orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr);
4749         orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest);
4750         orig_eth_type = eth->h_proto;
4751
4752         act = bpf_prog_run_xdp(xdp_prog, xdp);
4753
4754         /* check if bpf_xdp_adjust_head was used */
4755         off = xdp->data - orig_data;
4756         if (off) {
4757                 if (off > 0)
4758                         __skb_pull(skb, off);
4759                 else if (off < 0)
4760                         __skb_push(skb, -off);
4761
4762                 skb->mac_header += off;
4763                 skb_reset_network_header(skb);
4764         }
4765
4766         /* check if bpf_xdp_adjust_tail was used */
4767         off = xdp->data_end - orig_data_end;
4768         if (off != 0) {
4769                 skb_set_tail_pointer(skb, xdp->data_end - xdp->data);
4770                 skb->len += off; /* positive on grow, negative on shrink */
4771         }
4772
4773         /* check if XDP changed eth hdr such SKB needs update */
4774         eth = (struct ethhdr *)xdp->data;
4775         if ((orig_eth_type != eth->h_proto) ||
4776             (orig_host != ether_addr_equal_64bits(eth->h_dest,
4777                                                   skb->dev->dev_addr)) ||
4778             (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) {
4779                 __skb_push(skb, ETH_HLEN);
4780                 skb->pkt_type = PACKET_HOST;
4781                 skb->protocol = eth_type_trans(skb, skb->dev);
4782         }
4783
4784         /* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull
4785          * before calling us again on redirect path. We do not call do_redirect
4786          * as we leave that up to the caller.
4787          *
4788          * Caller is responsible for managing lifetime of skb (i.e. calling
4789          * kfree_skb in response to actions it cannot handle/XDP_DROP).
4790          */
4791         switch (act) {
4792         case XDP_REDIRECT:
4793         case XDP_TX:
4794                 __skb_push(skb, mac_len);
4795                 break;
4796         case XDP_PASS:
4797                 metalen = xdp->data - xdp->data_meta;
4798                 if (metalen)
4799                         skb_metadata_set(skb, metalen);
4800                 break;
4801         }
4802
4803         return act;
4804 }
4805
4806 static u32 netif_receive_generic_xdp(struct sk_buff *skb,
4807                                      struct xdp_buff *xdp,
4808                                      struct bpf_prog *xdp_prog)
4809 {
4810         u32 act = XDP_DROP;
4811
4812         /* Reinjected packets coming from act_mirred or similar should
4813          * not get XDP generic processing.
4814          */
4815         if (skb_is_redirected(skb))
4816                 return XDP_PASS;
4817
4818         /* XDP packets must be linear and must have sufficient headroom
4819          * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also
4820          * native XDP provides, thus we need to do it here as well.
4821          */
4822         if (skb_cloned(skb) || skb_is_nonlinear(skb) ||
4823             skb_headroom(skb) < XDP_PACKET_HEADROOM) {
4824                 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb);
4825                 int troom = skb->tail + skb->data_len - skb->end;
4826
4827                 /* In case we have to go down the path and also linearize,
4828                  * then lets do the pskb_expand_head() work just once here.
4829                  */
4830                 if (pskb_expand_head(skb,
4831                                      hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0,
4832                                      troom > 0 ? troom + 128 : 0, GFP_ATOMIC))
4833                         goto do_drop;
4834                 if (skb_linearize(skb))
4835                         goto do_drop;
4836         }
4837
4838         act = bpf_prog_run_generic_xdp(skb, xdp, xdp_prog);
4839         switch (act) {
4840         case XDP_REDIRECT:
4841         case XDP_TX:
4842         case XDP_PASS:
4843                 break;
4844         default:
4845                 bpf_warn_invalid_xdp_action(skb->dev, xdp_prog, act);
4846                 fallthrough;
4847         case XDP_ABORTED:
4848                 trace_xdp_exception(skb->dev, xdp_prog, act);
4849                 fallthrough;
4850         case XDP_DROP:
4851         do_drop:
4852                 kfree_skb(skb);
4853                 break;
4854         }
4855
4856         return act;
4857 }
4858
4859 /* When doing generic XDP we have to bypass the qdisc layer and the
4860  * network taps in order to match in-driver-XDP behavior. This also means
4861  * that XDP packets are able to starve other packets going through a qdisc,
4862  * and DDOS attacks will be more effective. In-driver-XDP use dedicated TX
4863  * queues, so they do not have this starvation issue.
4864  */
4865 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog)
4866 {
4867         struct net_device *dev = skb->dev;
4868         struct netdev_queue *txq;
4869         bool free_skb = true;
4870         int cpu, rc;
4871
4872         txq = netdev_core_pick_tx(dev, skb, NULL);
4873         cpu = smp_processor_id();
4874         HARD_TX_LOCK(dev, txq, cpu);
4875         if (!netif_xmit_frozen_or_drv_stopped(txq)) {
4876                 rc = netdev_start_xmit(skb, dev, txq, 0);
4877                 if (dev_xmit_complete(rc))
4878                         free_skb = false;
4879         }
4880         HARD_TX_UNLOCK(dev, txq);
4881         if (free_skb) {
4882                 trace_xdp_exception(dev, xdp_prog, XDP_TX);
4883                 dev_core_stats_tx_dropped_inc(dev);
4884                 kfree_skb(skb);
4885         }
4886 }
4887
4888 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key);
4889
4890 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb)
4891 {
4892         if (xdp_prog) {
4893                 struct xdp_buff xdp;
4894                 u32 act;
4895                 int err;
4896
4897                 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog);
4898                 if (act != XDP_PASS) {
4899                         switch (act) {
4900                         case XDP_REDIRECT:
4901                                 err = xdp_do_generic_redirect(skb->dev, skb,
4902                                                               &xdp, xdp_prog);
4903                                 if (err)
4904                                         goto out_redir;
4905                                 break;
4906                         case XDP_TX:
4907                                 generic_xdp_tx(skb, xdp_prog);
4908                                 break;
4909                         }
4910                         return XDP_DROP;
4911                 }
4912         }
4913         return XDP_PASS;
4914 out_redir:
4915         kfree_skb_reason(skb, SKB_DROP_REASON_XDP);
4916         return XDP_DROP;
4917 }
4918 EXPORT_SYMBOL_GPL(do_xdp_generic);
4919
4920 static int netif_rx_internal(struct sk_buff *skb)
4921 {
4922         int ret;
4923
4924         net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb);
4925
4926         trace_netif_rx(skb);
4927
4928 #ifdef CONFIG_RPS
4929         if (static_branch_unlikely(&rps_needed)) {
4930                 struct rps_dev_flow voidflow, *rflow = &voidflow;
4931                 int cpu;
4932
4933                 rcu_read_lock();
4934
4935                 cpu = get_rps_cpu(skb->dev, skb, &rflow);
4936                 if (cpu < 0)
4937                         cpu = smp_processor_id();
4938
4939                 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4940
4941                 rcu_read_unlock();
4942         } else
4943 #endif
4944         {
4945                 unsigned int qtail;
4946
4947                 ret = enqueue_to_backlog(skb, smp_processor_id(), &qtail);
4948         }
4949         return ret;
4950 }
4951
4952 /**
4953  *      __netif_rx      -       Slightly optimized version of netif_rx
4954  *      @skb: buffer to post
4955  *
4956  *      This behaves as netif_rx except that it does not disable bottom halves.
4957  *      As a result this function may only be invoked from the interrupt context
4958  *      (either hard or soft interrupt).
4959  */
4960 int __netif_rx(struct sk_buff *skb)
4961 {
4962         int ret;
4963
4964         lockdep_assert_once(hardirq_count() | softirq_count());
4965
4966         trace_netif_rx_entry(skb);
4967         ret = netif_rx_internal(skb);
4968         trace_netif_rx_exit(ret);
4969         return ret;
4970 }
4971 EXPORT_SYMBOL(__netif_rx);
4972
4973 /**
4974  *      netif_rx        -       post buffer to the network code
4975  *      @skb: buffer to post
4976  *
4977  *      This function receives a packet from a device driver and queues it for
4978  *      the upper (protocol) levels to process via the backlog NAPI device. It
4979  *      always succeeds. The buffer may be dropped during processing for
4980  *      congestion control or by the protocol layers.
4981  *      The network buffer is passed via the backlog NAPI device. Modern NIC
4982  *      driver should use NAPI and GRO.
4983  *      This function can used from interrupt and from process context. The
4984  *      caller from process context must not disable interrupts before invoking
4985  *      this function.
4986  *
4987  *      return values:
4988  *      NET_RX_SUCCESS  (no congestion)
4989  *      NET_RX_DROP     (packet was dropped)
4990  *
4991  */
4992 int netif_rx(struct sk_buff *skb)
4993 {
4994         bool need_bh_off = !(hardirq_count() | softirq_count());
4995         int ret;
4996
4997         if (need_bh_off)
4998                 local_bh_disable();
4999         trace_netif_rx_entry(skb);
5000         ret = netif_rx_internal(skb);
5001         trace_netif_rx_exit(ret);
5002         if (need_bh_off)
5003                 local_bh_enable();
5004         return ret;
5005 }
5006 EXPORT_SYMBOL(netif_rx);
5007
5008 static __latent_entropy void net_tx_action(struct softirq_action *h)
5009 {
5010         struct softnet_data *sd = this_cpu_ptr(&softnet_data);
5011
5012         if (sd->completion_queue) {
5013                 struct sk_buff *clist;
5014
5015                 local_irq_disable();
5016                 clist = sd->completion_queue;
5017                 sd->completion_queue = NULL;
5018                 local_irq_enable();
5019
5020                 while (clist) {
5021                         struct sk_buff *skb = clist;
5022
5023                         clist = clist->next;
5024
5025                         WARN_ON(refcount_read(&skb->users));
5026                         if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
5027                                 trace_consume_skb(skb);
5028                         else
5029                                 trace_kfree_skb(skb, net_tx_action,
5030                                                 SKB_DROP_REASON_NOT_SPECIFIED);
5031
5032                         if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
5033                                 __kfree_skb(skb);
5034                         else
5035                                 __kfree_skb_defer(skb);
5036                 }
5037         }
5038
5039         if (sd->output_queue) {
5040                 struct Qdisc *head;
5041
5042                 local_irq_disable();
5043                 head = sd->output_queue;
5044                 sd->output_queue = NULL;
5045                 sd->output_queue_tailp = &sd->output_queue;
5046                 local_irq_enable();
5047
5048                 rcu_read_lock();
5049
5050                 while (head) {
5051                         struct Qdisc *q = head;
5052                         spinlock_t *root_lock = NULL;
5053
5054                         head = head->next_sched;
5055
5056                         /* We need to make sure head->next_sched is read
5057                          * before clearing __QDISC_STATE_SCHED
5058                          */
5059                         smp_mb__before_atomic();
5060
5061                         if (!(q->flags & TCQ_F_NOLOCK)) {
5062                                 root_lock = qdisc_lock(q);
5063                                 spin_lock(root_lock);
5064                         } else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED,
5065                                                      &q->state))) {
5066                                 /* There is a synchronize_net() between
5067                                  * STATE_DEACTIVATED flag being set and
5068                                  * qdisc_reset()/some_qdisc_is_busy() in
5069                                  * dev_deactivate(), so we can safely bail out
5070                                  * early here to avoid data race between
5071                                  * qdisc_deactivate() and some_qdisc_is_busy()
5072                                  * for lockless qdisc.
5073                                  */
5074                                 clear_bit(__QDISC_STATE_SCHED, &q->state);
5075                                 continue;
5076                         }
5077
5078                         clear_bit(__QDISC_STATE_SCHED, &q->state);
5079                         qdisc_run(q);
5080                         if (root_lock)
5081                                 spin_unlock(root_lock);
5082                 }
5083
5084                 rcu_read_unlock();
5085         }
5086
5087         xfrm_dev_backlog(sd);
5088 }
5089
5090 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE)
5091 /* This hook is defined here for ATM LANE */
5092 int (*br_fdb_test_addr_hook)(struct net_device *dev,
5093                              unsigned char *addr) __read_mostly;
5094 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
5095 #endif
5096
5097 static inline struct sk_buff *
5098 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
5099                    struct net_device *orig_dev, bool *another)
5100 {
5101 #ifdef CONFIG_NET_CLS_ACT
5102         struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress);
5103         struct tcf_result cl_res;
5104
5105         /* If there's at least one ingress present somewhere (so
5106          * we get here via enabled static key), remaining devices
5107          * that are not configured with an ingress qdisc will bail
5108          * out here.
5109          */
5110         if (!miniq)
5111                 return skb;
5112
5113         if (*pt_prev) {
5114                 *ret = deliver_skb(skb, *pt_prev, orig_dev);
5115                 *pt_prev = NULL;
5116         }
5117
5118         qdisc_skb_cb(skb)->pkt_len = skb->len;
5119         tc_skb_cb(skb)->mru = 0;
5120         tc_skb_cb(skb)->post_ct = false;
5121         skb->tc_at_ingress = 1;
5122         mini_qdisc_bstats_cpu_update(miniq, skb);
5123
5124         switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) {
5125         case TC_ACT_OK:
5126         case TC_ACT_RECLASSIFY:
5127                 skb->tc_index = TC_H_MIN(cl_res.classid);
5128                 break;
5129         case TC_ACT_SHOT:
5130                 mini_qdisc_qstats_cpu_drop(miniq);
5131                 kfree_skb_reason(skb, SKB_DROP_REASON_TC_INGRESS);
5132                 *ret = NET_RX_DROP;
5133                 return NULL;
5134         case TC_ACT_STOLEN:
5135         case TC_ACT_QUEUED:
5136         case TC_ACT_TRAP:
5137                 consume_skb(skb);
5138                 *ret = NET_RX_SUCCESS;
5139                 return NULL;
5140         case TC_ACT_REDIRECT:
5141                 /* skb_mac_header check was done by cls/act_bpf, so
5142                  * we can safely push the L2 header back before
5143                  * redirecting to another netdev
5144                  */
5145                 __skb_push(skb, skb->mac_len);
5146                 if (skb_do_redirect(skb) == -EAGAIN) {
5147                         __skb_pull(skb, skb->mac_len);
5148                         *another = true;
5149                         break;
5150                 }
5151                 *ret = NET_RX_SUCCESS;
5152                 return NULL;
5153         case TC_ACT_CONSUMED:
5154                 *ret = NET_RX_SUCCESS;
5155                 return NULL;
5156         default:
5157                 break;
5158         }
5159 #endif /* CONFIG_NET_CLS_ACT */
5160         return skb;
5161 }
5162
5163 /**
5164  *      netdev_is_rx_handler_busy - check if receive handler is registered
5165  *      @dev: device to check
5166  *
5167  *      Check if a receive handler is already registered for a given device.
5168  *      Return true if there one.
5169  *
5170  *      The caller must hold the rtnl_mutex.
5171  */
5172 bool netdev_is_rx_handler_busy(struct net_device *dev)
5173 {
5174         ASSERT_RTNL();
5175         return dev && rtnl_dereference(dev->rx_handler);
5176 }
5177 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy);
5178
5179 /**
5180  *      netdev_rx_handler_register - register receive handler
5181  *      @dev: device to register a handler for
5182  *      @rx_handler: receive handler to register
5183  *      @rx_handler_data: data pointer that is used by rx handler
5184  *
5185  *      Register a receive handler for a device. This handler will then be
5186  *      called from __netif_receive_skb. A negative errno code is returned
5187  *      on a failure.
5188  *
5189  *      The caller must hold the rtnl_mutex.
5190  *
5191  *      For a general description of rx_handler, see enum rx_handler_result.
5192  */
5193 int netdev_rx_handler_register(struct net_device *dev,
5194                                rx_handler_func_t *rx_handler,
5195                                void *rx_handler_data)
5196 {
5197         if (netdev_is_rx_handler_busy(dev))
5198                 return -EBUSY;
5199
5200         if (dev->priv_flags & IFF_NO_RX_HANDLER)
5201                 return -EINVAL;
5202
5203         /* Note: rx_handler_data must be set before rx_handler */
5204         rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
5205         rcu_assign_pointer(dev->rx_handler, rx_handler);
5206
5207         return 0;
5208 }
5209 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
5210
5211 /**
5212  *      netdev_rx_handler_unregister - unregister receive handler
5213  *      @dev: device to unregister a handler from
5214  *
5215  *      Unregister a receive handler from a device.
5216  *
5217  *      The caller must hold the rtnl_mutex.
5218  */
5219 void netdev_rx_handler_unregister(struct net_device *dev)
5220 {
5221
5222         ASSERT_RTNL();
5223         RCU_INIT_POINTER(dev->rx_handler, NULL);
5224         /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
5225          * section has a guarantee to see a non NULL rx_handler_data
5226          * as well.
5227          */
5228         synchronize_net();
5229         RCU_INIT_POINTER(dev->rx_handler_data, NULL);
5230 }
5231 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
5232
5233 /*
5234  * Limit the use of PFMEMALLOC reserves to those protocols that implement
5235  * the special handling of PFMEMALLOC skbs.
5236  */
5237 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
5238 {
5239         switch (skb->protocol) {
5240         case htons(ETH_P_ARP):
5241         case htons(ETH_P_IP):
5242         case htons(ETH_P_IPV6):
5243         case htons(ETH_P_8021Q):
5244         case htons(ETH_P_8021AD):
5245                 return true;
5246         default:
5247                 return false;
5248         }
5249 }
5250
5251 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
5252                              int *ret, struct net_device *orig_dev)
5253 {
5254         if (nf_hook_ingress_active(skb)) {
5255                 int ingress_retval;
5256
5257                 if (*pt_prev) {
5258                         *ret = deliver_skb(skb, *pt_prev, orig_dev);
5259                         *pt_prev = NULL;
5260                 }
5261
5262                 rcu_read_lock();
5263                 ingress_retval = nf_hook_ingress(skb);
5264                 rcu_read_unlock();
5265                 return ingress_retval;
5266         }
5267         return 0;
5268 }
5269
5270 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc,
5271                                     struct packet_type **ppt_prev)
5272 {
5273         struct packet_type *ptype, *pt_prev;
5274         rx_handler_func_t *rx_handler;
5275         struct sk_buff *skb = *pskb;
5276         struct net_device *orig_dev;
5277         bool deliver_exact = false;
5278         int ret = NET_RX_DROP;
5279         __be16 type;
5280
5281         net_timestamp_check(!READ_ONCE(netdev_tstamp_prequeue), skb);
5282
5283         trace_netif_receive_skb(skb);
5284
5285         orig_dev = skb->dev;
5286
5287         skb_reset_network_header(skb);
5288         if (!skb_transport_header_was_set(skb))
5289                 skb_reset_transport_header(skb);
5290         skb_reset_mac_len(skb);
5291
5292         pt_prev = NULL;
5293
5294 another_round:
5295         skb->skb_iif = skb->dev->ifindex;
5296
5297         __this_cpu_inc(softnet_data.processed);
5298
5299         if (static_branch_unlikely(&generic_xdp_needed_key)) {
5300                 int ret2;
5301
5302                 migrate_disable();
5303                 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb);
5304                 migrate_enable();
5305
5306                 if (ret2 != XDP_PASS) {
5307                         ret = NET_RX_DROP;
5308                         goto out;
5309                 }
5310         }
5311
5312         if (eth_type_vlan(skb->protocol)) {
5313                 skb = skb_vlan_untag(skb);
5314                 if (unlikely(!skb))
5315                         goto out;
5316         }
5317
5318         if (skb_skip_tc_classify(skb))
5319                 goto skip_classify;
5320
5321         if (pfmemalloc)
5322                 goto skip_taps;
5323
5324         list_for_each_entry_rcu(ptype, &ptype_all, list) {
5325                 if (pt_prev)
5326                         ret = deliver_skb(skb, pt_prev, orig_dev);
5327                 pt_prev = ptype;
5328         }
5329
5330         list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
5331                 if (pt_prev)
5332                         ret = deliver_skb(skb, pt_prev, orig_dev);
5333                 pt_prev = ptype;
5334         }
5335
5336 skip_taps:
5337 #ifdef CONFIG_NET_INGRESS
5338         if (static_branch_unlikely(&ingress_needed_key)) {
5339                 bool another = false;
5340
5341                 nf_skip_egress(skb, true);
5342                 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev,
5343                                          &another);
5344                 if (another)
5345                         goto another_round;
5346                 if (!skb)
5347                         goto out;
5348
5349                 nf_skip_egress(skb, false);
5350                 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
5351                         goto out;
5352         }
5353 #endif
5354         skb_reset_redirect(skb);
5355 skip_classify:
5356         if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
5357                 goto drop;
5358
5359         if (skb_vlan_tag_present(skb)) {
5360                 if (pt_prev) {
5361                         ret = deliver_skb(skb, pt_prev, orig_dev);
5362                         pt_prev = NULL;
5363                 }
5364                 if (vlan_do_receive(&skb))
5365                         goto another_round;
5366                 else if (unlikely(!skb))
5367                         goto out;
5368         }
5369
5370         rx_handler = rcu_dereference(skb->dev->rx_handler);
5371         if (rx_handler) {
5372                 if (pt_prev) {
5373                         ret = deliver_skb(skb, pt_prev, orig_dev);
5374                         pt_prev = NULL;
5375                 }
5376                 switch (rx_handler(&skb)) {
5377                 case RX_HANDLER_CONSUMED:
5378                         ret = NET_RX_SUCCESS;
5379                         goto out;
5380                 case RX_HANDLER_ANOTHER:
5381                         goto another_round;
5382                 case RX_HANDLER_EXACT:
5383                         deliver_exact = true;
5384                         break;
5385                 case RX_HANDLER_PASS:
5386                         break;
5387                 default:
5388                         BUG();
5389                 }
5390         }
5391
5392         if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) {
5393 check_vlan_id:
5394                 if (skb_vlan_tag_get_id(skb)) {
5395                         /* Vlan id is non 0 and vlan_do_receive() above couldn't
5396                          * find vlan device.
5397                          */
5398                         skb->pkt_type = PACKET_OTHERHOST;
5399                 } else if (eth_type_vlan(skb->protocol)) {
5400                         /* Outer header is 802.1P with vlan 0, inner header is
5401                          * 802.1Q or 802.1AD and vlan_do_receive() above could
5402                          * not find vlan dev for vlan id 0.
5403                          */
5404                         __vlan_hwaccel_clear_tag(skb);
5405                         skb = skb_vlan_untag(skb);
5406                         if (unlikely(!skb))
5407                                 goto out;
5408                         if (vlan_do_receive(&skb))
5409                                 /* After stripping off 802.1P header with vlan 0
5410                                  * vlan dev is found for inner header.
5411                                  */
5412                                 goto another_round;
5413                         else if (unlikely(!skb))
5414                                 goto out;
5415                         else
5416                                 /* We have stripped outer 802.1P vlan 0 header.
5417                                  * But could not find vlan dev.
5418                                  * check again for vlan id to set OTHERHOST.
5419                                  */
5420                                 goto check_vlan_id;
5421                 }
5422                 /* Note: we might in the future use prio bits
5423                  * and set skb->priority like in vlan_do_receive()
5424                  * For the time being, just ignore Priority Code Point
5425                  */
5426                 __vlan_hwaccel_clear_tag(skb);
5427         }
5428
5429         type = skb->protocol;
5430
5431         /* deliver only exact match when indicated */
5432         if (likely(!deliver_exact)) {
5433                 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5434                                        &ptype_base[ntohs(type) &
5435                                                    PTYPE_HASH_MASK]);
5436         }
5437
5438         deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5439                                &orig_dev->ptype_specific);
5440
5441         if (unlikely(skb->dev != orig_dev)) {
5442                 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5443                                        &skb->dev->ptype_specific);
5444         }
5445
5446         if (pt_prev) {
5447                 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
5448                         goto drop;
5449                 *ppt_prev = pt_prev;
5450         } else {
5451 drop:
5452                 if (!deliver_exact)
5453                         dev_core_stats_rx_dropped_inc(skb->dev);
5454                 else
5455                         dev_core_stats_rx_nohandler_inc(skb->dev);
5456                 kfree_skb_reason(skb, SKB_DROP_REASON_UNHANDLED_PROTO);
5457                 /* Jamal, now you will not able to escape explaining
5458                  * me how you were going to use this. :-)
5459                  */
5460                 ret = NET_RX_DROP;
5461         }
5462
5463 out:
5464         /* The invariant here is that if *ppt_prev is not NULL
5465          * then skb should also be non-NULL.
5466          *
5467          * Apparently *ppt_prev assignment above holds this invariant due to
5468          * skb dereferencing near it.
5469          */
5470         *pskb = skb;
5471         return ret;
5472 }
5473
5474 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc)
5475 {
5476         struct net_device *orig_dev = skb->dev;
5477         struct packet_type *pt_prev = NULL;
5478         int ret;
5479
5480         ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5481         if (pt_prev)
5482                 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb,
5483                                          skb->dev, pt_prev, orig_dev);
5484         return ret;
5485 }
5486
5487 /**
5488  *      netif_receive_skb_core - special purpose version of netif_receive_skb
5489  *      @skb: buffer to process
5490  *
5491  *      More direct receive version of netif_receive_skb().  It should
5492  *      only be used by callers that have a need to skip RPS and Generic XDP.
5493  *      Caller must also take care of handling if ``(page_is_)pfmemalloc``.
5494  *
5495  *      This function may only be called from softirq context and interrupts
5496  *      should be enabled.
5497  *
5498  *      Return values (usually ignored):
5499  *      NET_RX_SUCCESS: no congestion
5500  *      NET_RX_DROP: packet was dropped
5501  */
5502 int netif_receive_skb_core(struct sk_buff *skb)
5503 {
5504         int ret;
5505
5506         rcu_read_lock();
5507         ret = __netif_receive_skb_one_core(skb, false);
5508         rcu_read_unlock();
5509
5510         return ret;
5511 }
5512 EXPORT_SYMBOL(netif_receive_skb_core);
5513
5514 static inline void __netif_receive_skb_list_ptype(struct list_head *head,
5515                                                   struct packet_type *pt_prev,
5516                                                   struct net_device *orig_dev)
5517 {
5518         struct sk_buff *skb, *next;
5519
5520         if (!pt_prev)
5521                 return;
5522         if (list_empty(head))
5523                 return;
5524         if (pt_prev->list_func != NULL)
5525                 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv,
5526                                    ip_list_rcv, head, pt_prev, orig_dev);
5527         else
5528                 list_for_each_entry_safe(skb, next, head, list) {
5529                         skb_list_del_init(skb);
5530                         pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
5531                 }
5532 }
5533
5534 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc)
5535 {
5536         /* Fast-path assumptions:
5537          * - There is no RX handler.
5538          * - Only one packet_type matches.
5539          * If either of these fails, we will end up doing some per-packet
5540          * processing in-line, then handling the 'last ptype' for the whole
5541          * sublist.  This can't cause out-of-order delivery to any single ptype,
5542          * because the 'last ptype' must be constant across the sublist, and all
5543          * other ptypes are handled per-packet.
5544          */
5545         /* Current (common) ptype of sublist */
5546         struct packet_type *pt_curr = NULL;
5547         /* Current (common) orig_dev of sublist */
5548         struct net_device *od_curr = NULL;
5549         struct list_head sublist;
5550         struct sk_buff *skb, *next;
5551
5552         INIT_LIST_HEAD(&sublist);
5553         list_for_each_entry_safe(skb, next, head, list) {
5554                 struct net_device *orig_dev = skb->dev;
5555                 struct packet_type *pt_prev = NULL;
5556
5557                 skb_list_del_init(skb);
5558                 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5559                 if (!pt_prev)
5560                         continue;
5561                 if (pt_curr != pt_prev || od_curr != orig_dev) {
5562                         /* dispatch old sublist */
5563                         __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5564                         /* start new sublist */
5565                         INIT_LIST_HEAD(&sublist);
5566                         pt_curr = pt_prev;
5567                         od_curr = orig_dev;
5568                 }
5569                 list_add_tail(&skb->list, &sublist);
5570         }
5571
5572         /* dispatch final sublist */
5573         __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5574 }
5575
5576 static int __netif_receive_skb(struct sk_buff *skb)
5577 {
5578         int ret;
5579
5580         if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
5581                 unsigned int noreclaim_flag;
5582
5583                 /*
5584                  * PFMEMALLOC skbs are special, they should
5585                  * - be delivered to SOCK_MEMALLOC sockets only
5586                  * - stay away from userspace
5587                  * - have bounded memory usage
5588                  *
5589                  * Use PF_MEMALLOC as this saves us from propagating the allocation
5590                  * context down to all allocation sites.
5591                  */
5592                 noreclaim_flag = memalloc_noreclaim_save();
5593                 ret = __netif_receive_skb_one_core(skb, true);
5594                 memalloc_noreclaim_restore(noreclaim_flag);
5595         } else
5596                 ret = __netif_receive_skb_one_core(skb, false);
5597
5598         return ret;
5599 }
5600
5601 static void __netif_receive_skb_list(struct list_head *head)
5602 {
5603         unsigned long noreclaim_flag = 0;
5604         struct sk_buff *skb, *next;
5605         bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */
5606
5607         list_for_each_entry_safe(skb, next, head, list) {
5608                 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) {
5609                         struct list_head sublist;
5610
5611                         /* Handle the previous sublist */
5612                         list_cut_before(&sublist, head, &skb->list);
5613                         if (!list_empty(&sublist))
5614                                 __netif_receive_skb_list_core(&sublist, pfmemalloc);
5615                         pfmemalloc = !pfmemalloc;
5616                         /* See comments in __netif_receive_skb */
5617                         if (pfmemalloc)
5618                                 noreclaim_flag = memalloc_noreclaim_save();
5619                         else
5620                                 memalloc_noreclaim_restore(noreclaim_flag);
5621                 }
5622         }
5623         /* Handle the remaining sublist */
5624         if (!list_empty(head))
5625                 __netif_receive_skb_list_core(head, pfmemalloc);
5626         /* Restore pflags */
5627         if (pfmemalloc)
5628                 memalloc_noreclaim_restore(noreclaim_flag);
5629 }
5630
5631 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp)
5632 {
5633         struct bpf_prog *old = rtnl_dereference(dev->xdp_prog);
5634         struct bpf_prog *new = xdp->prog;
5635         int ret = 0;
5636
5637         switch (xdp->command) {
5638         case XDP_SETUP_PROG:
5639                 rcu_assign_pointer(dev->xdp_prog, new);
5640                 if (old)
5641                         bpf_prog_put(old);
5642
5643                 if (old && !new) {
5644                         static_branch_dec(&generic_xdp_needed_key);
5645                 } else if (new && !old) {
5646                         static_branch_inc(&generic_xdp_needed_key);
5647                         dev_disable_lro(dev);
5648                         dev_disable_gro_hw(dev);
5649                 }
5650                 break;
5651
5652         default:
5653                 ret = -EINVAL;
5654                 break;
5655         }
5656
5657         return ret;
5658 }
5659
5660 static int netif_receive_skb_internal(struct sk_buff *skb)
5661 {
5662         int ret;
5663
5664         net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb);
5665
5666         if (skb_defer_rx_timestamp(skb))
5667                 return NET_RX_SUCCESS;
5668
5669         rcu_read_lock();
5670 #ifdef CONFIG_RPS
5671         if (static_branch_unlikely(&rps_needed)) {
5672                 struct rps_dev_flow voidflow, *rflow = &voidflow;
5673                 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5674
5675                 if (cpu >= 0) {
5676                         ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5677                         rcu_read_unlock();
5678                         return ret;
5679                 }
5680         }
5681 #endif
5682         ret = __netif_receive_skb(skb);
5683         rcu_read_unlock();
5684         return ret;
5685 }
5686
5687 void netif_receive_skb_list_internal(struct list_head *head)
5688 {
5689         struct sk_buff *skb, *next;
5690         struct list_head sublist;
5691
5692         INIT_LIST_HEAD(&sublist);
5693         list_for_each_entry_safe(skb, next, head, list) {
5694                 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb);
5695                 skb_list_del_init(skb);
5696                 if (!skb_defer_rx_timestamp(skb))
5697                         list_add_tail(&skb->list, &sublist);
5698         }
5699         list_splice_init(&sublist, head);
5700
5701         rcu_read_lock();
5702 #ifdef CONFIG_RPS
5703         if (static_branch_unlikely(&rps_needed)) {
5704                 list_for_each_entry_safe(skb, next, head, list) {
5705                         struct rps_dev_flow voidflow, *rflow = &voidflow;
5706                         int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5707
5708                         if (cpu >= 0) {
5709                                 /* Will be handled, remove from list */
5710                                 skb_list_del_init(skb);
5711                                 enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5712                         }
5713                 }
5714         }
5715 #endif
5716         __netif_receive_skb_list(head);
5717         rcu_read_unlock();
5718 }
5719
5720 /**
5721  *      netif_receive_skb - process receive buffer from network
5722  *      @skb: buffer to process
5723  *
5724  *      netif_receive_skb() is the main receive data processing function.
5725  *      It always succeeds. The buffer may be dropped during processing
5726  *      for congestion control or by the protocol layers.
5727  *
5728  *      This function may only be called from softirq context and interrupts
5729  *      should be enabled.
5730  *
5731  *      Return values (usually ignored):
5732  *      NET_RX_SUCCESS: no congestion
5733  *      NET_RX_DROP: packet was dropped
5734  */
5735 int netif_receive_skb(struct sk_buff *skb)
5736 {
5737         int ret;
5738
5739         trace_netif_receive_skb_entry(skb);
5740
5741         ret = netif_receive_skb_internal(skb);
5742         trace_netif_receive_skb_exit(ret);
5743
5744         return ret;
5745 }
5746 EXPORT_SYMBOL(netif_receive_skb);
5747
5748 /**
5749  *      netif_receive_skb_list - process many receive buffers from network
5750  *      @head: list of skbs to process.
5751  *
5752  *      Since return value of netif_receive_skb() is normally ignored, and
5753  *      wouldn't be meaningful for a list, this function returns void.
5754  *
5755  *      This function may only be called from softirq context and interrupts
5756  *      should be enabled.
5757  */
5758 void netif_receive_skb_list(struct list_head *head)
5759 {
5760         struct sk_buff *skb;
5761
5762         if (list_empty(head))
5763                 return;
5764         if (trace_netif_receive_skb_list_entry_enabled()) {
5765                 list_for_each_entry(skb, head, list)
5766                         trace_netif_receive_skb_list_entry(skb);
5767         }
5768         netif_receive_skb_list_internal(head);
5769         trace_netif_receive_skb_list_exit(0);
5770 }
5771 EXPORT_SYMBOL(netif_receive_skb_list);
5772
5773 static DEFINE_PER_CPU(struct work_struct, flush_works);
5774
5775 /* Network device is going away, flush any packets still pending */
5776 static void flush_backlog(struct work_struct *work)
5777 {
5778         struct sk_buff *skb, *tmp;
5779         struct softnet_data *sd;
5780
5781         local_bh_disable();
5782         sd = this_cpu_ptr(&softnet_data);
5783
5784         rps_lock_irq_disable(sd);
5785         skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
5786                 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5787                         __skb_unlink(skb, &sd->input_pkt_queue);
5788                         dev_kfree_skb_irq(skb);
5789                         input_queue_head_incr(sd);
5790                 }
5791         }
5792         rps_unlock_irq_enable(sd);
5793
5794         skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
5795                 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5796                         __skb_unlink(skb, &sd->process_queue);
5797                         kfree_skb(skb);
5798                         input_queue_head_incr(sd);
5799                 }
5800         }
5801         local_bh_enable();
5802 }
5803
5804 static bool flush_required(int cpu)
5805 {
5806 #if IS_ENABLED(CONFIG_RPS)
5807         struct softnet_data *sd = &per_cpu(softnet_data, cpu);
5808         bool do_flush;
5809
5810         rps_lock_irq_disable(sd);
5811
5812         /* as insertion into process_queue happens with the rps lock held,
5813          * process_queue access may race only with dequeue
5814          */
5815         do_flush = !skb_queue_empty(&sd->input_pkt_queue) ||
5816                    !skb_queue_empty_lockless(&sd->process_queue);
5817         rps_unlock_irq_enable(sd);
5818
5819         return do_flush;
5820 #endif
5821         /* without RPS we can't safely check input_pkt_queue: during a
5822          * concurrent remote skb_queue_splice() we can detect as empty both
5823          * input_pkt_queue and process_queue even if the latter could end-up
5824          * containing a lot of packets.
5825          */
5826         return true;
5827 }
5828
5829 static void flush_all_backlogs(void)
5830 {
5831         static cpumask_t flush_cpus;
5832         unsigned int cpu;
5833
5834         /* since we are under rtnl lock protection we can use static data
5835          * for the cpumask and avoid allocating on stack the possibly
5836          * large mask
5837          */
5838         ASSERT_RTNL();
5839
5840         cpus_read_lock();
5841
5842         cpumask_clear(&flush_cpus);
5843         for_each_online_cpu(cpu) {
5844                 if (flush_required(cpu)) {
5845                         queue_work_on(cpu, system_highpri_wq,
5846                                       per_cpu_ptr(&flush_works, cpu));
5847                         cpumask_set_cpu(cpu, &flush_cpus);
5848                 }
5849         }
5850
5851         /* we can have in flight packet[s] on the cpus we are not flushing,
5852          * synchronize_net() in unregister_netdevice_many() will take care of
5853          * them
5854          */
5855         for_each_cpu(cpu, &flush_cpus)
5856                 flush_work(per_cpu_ptr(&flush_works, cpu));
5857
5858         cpus_read_unlock();
5859 }
5860
5861 static void net_rps_send_ipi(struct softnet_data *remsd)
5862 {
5863 #ifdef CONFIG_RPS
5864         while (remsd) {
5865                 struct softnet_data *next = remsd->rps_ipi_next;
5866
5867                 if (cpu_online(remsd->cpu))
5868                         smp_call_function_single_async(remsd->cpu, &remsd->csd);
5869                 remsd = next;
5870         }
5871 #endif
5872 }
5873
5874 /*
5875  * net_rps_action_and_irq_enable sends any pending IPI's for rps.
5876  * Note: called with local irq disabled, but exits with local irq enabled.
5877  */
5878 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
5879 {
5880 #ifdef CONFIG_RPS
5881         struct softnet_data *remsd = sd->rps_ipi_list;
5882
5883         if (remsd) {
5884                 sd->rps_ipi_list = NULL;
5885
5886                 local_irq_enable();
5887
5888                 /* Send pending IPI's to kick RPS processing on remote cpus. */
5889                 net_rps_send_ipi(remsd);
5890         } else
5891 #endif
5892                 local_irq_enable();
5893 }
5894
5895 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
5896 {
5897 #ifdef CONFIG_RPS
5898         return sd->rps_ipi_list != NULL;
5899 #else
5900         return false;
5901 #endif
5902 }
5903
5904 static int process_backlog(struct napi_struct *napi, int quota)
5905 {
5906         struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
5907         bool again = true;
5908         int work = 0;
5909
5910         /* Check if we have pending ipi, its better to send them now,
5911          * not waiting net_rx_action() end.
5912          */
5913         if (sd_has_rps_ipi_waiting(sd)) {
5914                 local_irq_disable();
5915                 net_rps_action_and_irq_enable(sd);
5916         }
5917
5918         napi->weight = READ_ONCE(dev_rx_weight);
5919         while (again) {
5920                 struct sk_buff *skb;
5921
5922                 while ((skb = __skb_dequeue(&sd->process_queue))) {
5923                         rcu_read_lock();
5924                         __netif_receive_skb(skb);
5925                         rcu_read_unlock();
5926                         input_queue_head_incr(sd);
5927                         if (++work >= quota)
5928                                 return work;
5929
5930                 }
5931
5932                 rps_lock_irq_disable(sd);
5933                 if (skb_queue_empty(&sd->input_pkt_queue)) {
5934                         /*
5935                          * Inline a custom version of __napi_complete().
5936                          * only current cpu owns and manipulates this napi,
5937                          * and NAPI_STATE_SCHED is the only possible flag set
5938                          * on backlog.
5939                          * We can use a plain write instead of clear_bit(),
5940                          * and we dont need an smp_mb() memory barrier.
5941                          */
5942                         napi->state = 0;
5943                         again = false;
5944                 } else {
5945                         skb_queue_splice_tail_init(&sd->input_pkt_queue,
5946                                                    &sd->process_queue);
5947                 }
5948                 rps_unlock_irq_enable(sd);
5949         }
5950
5951         return work;
5952 }
5953
5954 /**
5955  * __napi_schedule - schedule for receive
5956  * @n: entry to schedule
5957  *
5958  * The entry's receive function will be scheduled to run.
5959  * Consider using __napi_schedule_irqoff() if hard irqs are masked.
5960  */
5961 void __napi_schedule(struct napi_struct *n)
5962 {
5963         unsigned long flags;
5964
5965         local_irq_save(flags);
5966         ____napi_schedule(this_cpu_ptr(&softnet_data), n);
5967         local_irq_restore(flags);
5968 }
5969 EXPORT_SYMBOL(__napi_schedule);
5970
5971 /**
5972  *      napi_schedule_prep - check if napi can be scheduled
5973  *      @n: napi context
5974  *
5975  * Test if NAPI routine is already running, and if not mark
5976  * it as running.  This is used as a condition variable to
5977  * insure only one NAPI poll instance runs.  We also make
5978  * sure there is no pending NAPI disable.
5979  */
5980 bool napi_schedule_prep(struct napi_struct *n)
5981 {
5982         unsigned long new, val = READ_ONCE(n->state);
5983
5984         do {
5985                 if (unlikely(val & NAPIF_STATE_DISABLE))
5986                         return false;
5987                 new = val | NAPIF_STATE_SCHED;
5988
5989                 /* Sets STATE_MISSED bit if STATE_SCHED was already set
5990                  * This was suggested by Alexander Duyck, as compiler
5991                  * emits better code than :
5992                  * if (val & NAPIF_STATE_SCHED)
5993                  *     new |= NAPIF_STATE_MISSED;
5994                  */
5995                 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED *
5996                                                    NAPIF_STATE_MISSED;
5997         } while (!try_cmpxchg(&n->state, &val, new));
5998
5999         return !(val & NAPIF_STATE_SCHED);
6000 }
6001 EXPORT_SYMBOL(napi_schedule_prep);
6002
6003 /**
6004  * __napi_schedule_irqoff - schedule for receive
6005  * @n: entry to schedule
6006  *
6007  * Variant of __napi_schedule() assuming hard irqs are masked.
6008  *
6009  * On PREEMPT_RT enabled kernels this maps to __napi_schedule()
6010  * because the interrupt disabled assumption might not be true
6011  * due to force-threaded interrupts and spinlock substitution.
6012  */
6013 void __napi_schedule_irqoff(struct napi_struct *n)
6014 {
6015         if (!IS_ENABLED(CONFIG_PREEMPT_RT))
6016                 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
6017         else
6018                 __napi_schedule(n);
6019 }
6020 EXPORT_SYMBOL(__napi_schedule_irqoff);
6021
6022 bool napi_complete_done(struct napi_struct *n, int work_done)
6023 {
6024         unsigned long flags, val, new, timeout = 0;
6025         bool ret = true;
6026
6027         /*
6028          * 1) Don't let napi dequeue from the cpu poll list
6029          *    just in case its running on a different cpu.
6030          * 2) If we are busy polling, do nothing here, we have
6031          *    the guarantee we will be called later.
6032          */
6033         if (unlikely(n->state & (NAPIF_STATE_NPSVC |
6034                                  NAPIF_STATE_IN_BUSY_POLL)))
6035                 return false;
6036
6037         if (work_done) {
6038                 if (n->gro_bitmask)
6039                         timeout = READ_ONCE(n->dev->gro_flush_timeout);
6040                 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs);
6041         }
6042         if (n->defer_hard_irqs_count > 0) {
6043                 n->defer_hard_irqs_count--;
6044                 timeout = READ_ONCE(n->dev->gro_flush_timeout);
6045                 if (timeout)
6046                         ret = false;
6047         }
6048         if (n->gro_bitmask) {
6049                 /* When the NAPI instance uses a timeout and keeps postponing
6050                  * it, we need to bound somehow the time packets are kept in
6051                  * the GRO layer
6052                  */
6053                 napi_gro_flush(n, !!timeout);
6054         }
6055
6056         gro_normal_list(n);
6057
6058         if (unlikely(!list_empty(&n->poll_list))) {
6059                 /* If n->poll_list is not empty, we need to mask irqs */
6060                 local_irq_save(flags);
6061                 list_del_init(&n->poll_list);
6062                 local_irq_restore(flags);
6063         }
6064
6065         val = READ_ONCE(n->state);
6066         do {
6067                 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED));
6068
6069                 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED |
6070                               NAPIF_STATE_SCHED_THREADED |
6071                               NAPIF_STATE_PREFER_BUSY_POLL);
6072
6073                 /* If STATE_MISSED was set, leave STATE_SCHED set,
6074                  * because we will call napi->poll() one more time.
6075                  * This C code was suggested by Alexander Duyck to help gcc.
6076                  */
6077                 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED *
6078                                                     NAPIF_STATE_SCHED;
6079         } while (!try_cmpxchg(&n->state, &val, new));
6080
6081         if (unlikely(val & NAPIF_STATE_MISSED)) {
6082                 __napi_schedule(n);
6083                 return false;
6084         }
6085
6086         if (timeout)
6087                 hrtimer_start(&n->timer, ns_to_ktime(timeout),
6088                               HRTIMER_MODE_REL_PINNED);
6089         return ret;
6090 }
6091 EXPORT_SYMBOL(napi_complete_done);
6092
6093 /* must be called under rcu_read_lock(), as we dont take a reference */
6094 static struct napi_struct *napi_by_id(unsigned int napi_id)
6095 {
6096         unsigned int hash = napi_id % HASH_SIZE(napi_hash);
6097         struct napi_struct *napi;
6098
6099         hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
6100                 if (napi->napi_id == napi_id)
6101                         return napi;
6102
6103         return NULL;
6104 }
6105
6106 #if defined(CONFIG_NET_RX_BUSY_POLL)
6107
6108 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule)
6109 {
6110         if (!skip_schedule) {
6111                 gro_normal_list(napi);
6112                 __napi_schedule(napi);
6113                 return;
6114         }
6115
6116         if (napi->gro_bitmask) {
6117                 /* flush too old packets
6118                  * If HZ < 1000, flush all packets.
6119                  */
6120                 napi_gro_flush(napi, HZ >= 1000);
6121         }
6122
6123         gro_normal_list(napi);
6124         clear_bit(NAPI_STATE_SCHED, &napi->state);
6125 }
6126
6127 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll,
6128                            u16 budget)
6129 {
6130         bool skip_schedule = false;
6131         unsigned long timeout;
6132         int rc;
6133
6134         /* Busy polling means there is a high chance device driver hard irq
6135          * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was
6136          * set in napi_schedule_prep().
6137          * Since we are about to call napi->poll() once more, we can safely
6138          * clear NAPI_STATE_MISSED.
6139          *
6140          * Note: x86 could use a single "lock and ..." instruction
6141          * to perform these two clear_bit()
6142          */
6143         clear_bit(NAPI_STATE_MISSED, &napi->state);
6144         clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state);
6145
6146         local_bh_disable();
6147
6148         if (prefer_busy_poll) {
6149                 napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs);
6150                 timeout = READ_ONCE(napi->dev->gro_flush_timeout);
6151                 if (napi->defer_hard_irqs_count && timeout) {
6152                         hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED);
6153                         skip_schedule = true;
6154                 }
6155         }
6156
6157         /* All we really want here is to re-enable device interrupts.
6158          * Ideally, a new ndo_busy_poll_stop() could avoid another round.
6159          */
6160         rc = napi->poll(napi, budget);
6161         /* We can't gro_normal_list() here, because napi->poll() might have
6162          * rearmed the napi (napi_complete_done()) in which case it could
6163          * already be running on another CPU.
6164          */
6165         trace_napi_poll(napi, rc, budget);
6166         netpoll_poll_unlock(have_poll_lock);
6167         if (rc == budget)
6168                 __busy_poll_stop(napi, skip_schedule);
6169         local_bh_enable();
6170 }
6171
6172 void napi_busy_loop(unsigned int napi_id,
6173                     bool (*loop_end)(void *, unsigned long),
6174                     void *loop_end_arg, bool prefer_busy_poll, u16 budget)
6175 {
6176         unsigned long start_time = loop_end ? busy_loop_current_time() : 0;
6177         int (*napi_poll)(struct napi_struct *napi, int budget);
6178         void *have_poll_lock = NULL;
6179         struct napi_struct *napi;
6180
6181 restart:
6182         napi_poll = NULL;
6183
6184         rcu_read_lock();
6185
6186         napi = napi_by_id(napi_id);
6187         if (!napi)
6188                 goto out;
6189
6190         preempt_disable();
6191         for (;;) {
6192                 int work = 0;
6193
6194                 local_bh_disable();
6195                 if (!napi_poll) {
6196                         unsigned long val = READ_ONCE(napi->state);
6197
6198                         /* If multiple threads are competing for this napi,
6199                          * we avoid dirtying napi->state as much as we can.
6200                          */
6201                         if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED |
6202                                    NAPIF_STATE_IN_BUSY_POLL)) {
6203                                 if (prefer_busy_poll)
6204                                         set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6205                                 goto count;
6206                         }
6207                         if (cmpxchg(&napi->state, val,
6208                                     val | NAPIF_STATE_IN_BUSY_POLL |
6209                                           NAPIF_STATE_SCHED) != val) {
6210                                 if (prefer_busy_poll)
6211                                         set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6212                                 goto count;
6213                         }
6214                         have_poll_lock = netpoll_poll_lock(napi);
6215                         napi_poll = napi->poll;
6216                 }
6217                 work = napi_poll(napi, budget);
6218                 trace_napi_poll(napi, work, budget);
6219                 gro_normal_list(napi);
6220 count:
6221                 if (work > 0)
6222                         __NET_ADD_STATS(dev_net(napi->dev),
6223                                         LINUX_MIB_BUSYPOLLRXPACKETS, work);
6224                 local_bh_enable();
6225
6226                 if (!loop_end || loop_end(loop_end_arg, start_time))
6227                         break;
6228
6229                 if (unlikely(need_resched())) {
6230                         if (napi_poll)
6231                                 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget);
6232                         preempt_enable();
6233                         rcu_read_unlock();
6234                         cond_resched();
6235                         if (loop_end(loop_end_arg, start_time))
6236                                 return;
6237                         goto restart;
6238                 }
6239                 cpu_relax();
6240         }
6241         if (napi_poll)
6242                 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget);
6243         preempt_enable();
6244 out:
6245         rcu_read_unlock();
6246 }
6247 EXPORT_SYMBOL(napi_busy_loop);
6248
6249 #endif /* CONFIG_NET_RX_BUSY_POLL */
6250
6251 static void napi_hash_add(struct napi_struct *napi)
6252 {
6253         if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state))
6254                 return;
6255
6256         spin_lock(&napi_hash_lock);
6257
6258         /* 0..NR_CPUS range is reserved for sender_cpu use */
6259         do {
6260                 if (unlikely(++napi_gen_id < MIN_NAPI_ID))
6261                         napi_gen_id = MIN_NAPI_ID;
6262         } while (napi_by_id(napi_gen_id));
6263         napi->napi_id = napi_gen_id;
6264
6265         hlist_add_head_rcu(&napi->napi_hash_node,
6266                            &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
6267
6268         spin_unlock(&napi_hash_lock);
6269 }
6270
6271 /* Warning : caller is responsible to make sure rcu grace period
6272  * is respected before freeing memory containing @napi
6273  */
6274 static void napi_hash_del(struct napi_struct *napi)
6275 {
6276         spin_lock(&napi_hash_lock);
6277
6278         hlist_del_init_rcu(&napi->napi_hash_node);
6279
6280         spin_unlock(&napi_hash_lock);
6281 }
6282
6283 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
6284 {
6285         struct napi_struct *napi;
6286
6287         napi = container_of(timer, struct napi_struct, timer);
6288
6289         /* Note : we use a relaxed variant of napi_schedule_prep() not setting
6290          * NAPI_STATE_MISSED, since we do not react to a device IRQ.
6291          */
6292         if (!napi_disable_pending(napi) &&
6293             !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) {
6294                 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6295                 __napi_schedule_irqoff(napi);
6296         }
6297
6298         return HRTIMER_NORESTART;
6299 }
6300
6301 static void init_gro_hash(struct napi_struct *napi)
6302 {
6303         int i;
6304
6305         for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6306                 INIT_LIST_HEAD(&napi->gro_hash[i].list);
6307                 napi->gro_hash[i].count = 0;
6308         }
6309         napi->gro_bitmask = 0;
6310 }
6311
6312 int dev_set_threaded(struct net_device *dev, bool threaded)
6313 {
6314         struct napi_struct *napi;
6315         int err = 0;
6316
6317         if (dev->threaded == threaded)
6318                 return 0;
6319
6320         if (threaded) {
6321                 list_for_each_entry(napi, &dev->napi_list, dev_list) {
6322                         if (!napi->thread) {
6323                                 err = napi_kthread_create(napi);
6324                                 if (err) {
6325                                         threaded = false;
6326                                         break;
6327                                 }
6328                         }
6329                 }
6330         }
6331
6332         dev->threaded = threaded;
6333
6334         /* Make sure kthread is created before THREADED bit
6335          * is set.
6336          */
6337         smp_mb__before_atomic();
6338
6339         /* Setting/unsetting threaded mode on a napi might not immediately
6340          * take effect, if the current napi instance is actively being
6341          * polled. In this case, the switch between threaded mode and
6342          * softirq mode will happen in the next round of napi_schedule().
6343          * This should not cause hiccups/stalls to the live traffic.
6344          */
6345         list_for_each_entry(napi, &dev->napi_list, dev_list) {
6346                 if (threaded)
6347                         set_bit(NAPI_STATE_THREADED, &napi->state);
6348                 else
6349                         clear_bit(NAPI_STATE_THREADED, &napi->state);
6350         }
6351
6352         return err;
6353 }
6354 EXPORT_SYMBOL(dev_set_threaded);
6355
6356 void netif_napi_add_weight(struct net_device *dev, struct napi_struct *napi,
6357                            int (*poll)(struct napi_struct *, int), int weight)
6358 {
6359         if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state)))
6360                 return;
6361
6362         INIT_LIST_HEAD(&napi->poll_list);
6363         INIT_HLIST_NODE(&napi->napi_hash_node);
6364         hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
6365         napi->timer.function = napi_watchdog;
6366         init_gro_hash(napi);
6367         napi->skb = NULL;
6368         INIT_LIST_HEAD(&napi->rx_list);
6369         napi->rx_count = 0;
6370         napi->poll = poll;
6371         if (weight > NAPI_POLL_WEIGHT)
6372                 netdev_err_once(dev, "%s() called with weight %d\n", __func__,
6373                                 weight);
6374         napi->weight = weight;
6375         napi->dev = dev;
6376 #ifdef CONFIG_NETPOLL
6377         napi->poll_owner = -1;
6378 #endif
6379         set_bit(NAPI_STATE_SCHED, &napi->state);
6380         set_bit(NAPI_STATE_NPSVC, &napi->state);
6381         list_add_rcu(&napi->dev_list, &dev->napi_list);
6382         napi_hash_add(napi);
6383         napi_get_frags_check(napi);
6384         /* Create kthread for this napi if dev->threaded is set.
6385          * Clear dev->threaded if kthread creation failed so that
6386          * threaded mode will not be enabled in napi_enable().
6387          */
6388         if (dev->threaded && napi_kthread_create(napi))
6389                 dev->threaded = 0;
6390 }
6391 EXPORT_SYMBOL(netif_napi_add_weight);
6392
6393 void napi_disable(struct napi_struct *n)
6394 {
6395         unsigned long val, new;
6396
6397         might_sleep();
6398         set_bit(NAPI_STATE_DISABLE, &n->state);
6399
6400         val = READ_ONCE(n->state);
6401         do {
6402                 while (val & (NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC)) {
6403                         usleep_range(20, 200);
6404                         val = READ_ONCE(n->state);
6405                 }
6406
6407                 new = val | NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC;
6408                 new &= ~(NAPIF_STATE_THREADED | NAPIF_STATE_PREFER_BUSY_POLL);
6409         } while (!try_cmpxchg(&n->state, &val, new));
6410
6411         hrtimer_cancel(&n->timer);
6412
6413         clear_bit(NAPI_STATE_DISABLE, &n->state);
6414 }
6415 EXPORT_SYMBOL(napi_disable);
6416
6417 /**
6418  *      napi_enable - enable NAPI scheduling
6419  *      @n: NAPI context
6420  *
6421  * Resume NAPI from being scheduled on this context.
6422  * Must be paired with napi_disable.
6423  */
6424 void napi_enable(struct napi_struct *n)
6425 {
6426         unsigned long new, val = READ_ONCE(n->state);
6427
6428         do {
6429                 BUG_ON(!test_bit(NAPI_STATE_SCHED, &val));
6430
6431                 new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC);
6432                 if (n->dev->threaded && n->thread)
6433                         new |= NAPIF_STATE_THREADED;
6434         } while (!try_cmpxchg(&n->state, &val, new));
6435 }
6436 EXPORT_SYMBOL(napi_enable);
6437
6438 static void flush_gro_hash(struct napi_struct *napi)
6439 {
6440         int i;
6441
6442         for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6443                 struct sk_buff *skb, *n;
6444
6445                 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list)
6446                         kfree_skb(skb);
6447                 napi->gro_hash[i].count = 0;
6448         }
6449 }
6450
6451 /* Must be called in process context */
6452 void __netif_napi_del(struct napi_struct *napi)
6453 {
6454         if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state))
6455                 return;
6456
6457         napi_hash_del(napi);
6458         list_del_rcu(&napi->dev_list);
6459         napi_free_frags(napi);
6460
6461         flush_gro_hash(napi);
6462         napi->gro_bitmask = 0;
6463
6464         if (napi->thread) {
6465                 kthread_stop(napi->thread);
6466                 napi->thread = NULL;
6467         }
6468 }
6469 EXPORT_SYMBOL(__netif_napi_del);
6470
6471 static int __napi_poll(struct napi_struct *n, bool *repoll)
6472 {
6473         int work, weight;
6474
6475         weight = n->weight;
6476
6477         /* This NAPI_STATE_SCHED test is for avoiding a race
6478          * with netpoll's poll_napi().  Only the entity which
6479          * obtains the lock and sees NAPI_STATE_SCHED set will
6480          * actually make the ->poll() call.  Therefore we avoid
6481          * accidentally calling ->poll() when NAPI is not scheduled.
6482          */
6483         work = 0;
6484         if (test_bit(NAPI_STATE_SCHED, &n->state)) {
6485                 work = n->poll(n, weight);
6486                 trace_napi_poll(n, work, weight);
6487         }
6488
6489         if (unlikely(work > weight))
6490                 netdev_err_once(n->dev, "NAPI poll function %pS returned %d, exceeding its budget of %d.\n",
6491                                 n->poll, work, weight);
6492
6493         if (likely(work < weight))
6494                 return work;
6495
6496         /* Drivers must not modify the NAPI state if they
6497          * consume the entire weight.  In such cases this code
6498          * still "owns" the NAPI instance and therefore can
6499          * move the instance around on the list at-will.
6500          */
6501         if (unlikely(napi_disable_pending(n))) {
6502                 napi_complete(n);
6503                 return work;
6504         }
6505
6506         /* The NAPI context has more processing work, but busy-polling
6507          * is preferred. Exit early.
6508          */
6509         if (napi_prefer_busy_poll(n)) {
6510                 if (napi_complete_done(n, work)) {
6511                         /* If timeout is not set, we need to make sure
6512                          * that the NAPI is re-scheduled.
6513                          */
6514                         napi_schedule(n);
6515                 }
6516                 return work;
6517         }
6518
6519         if (n->gro_bitmask) {
6520                 /* flush too old packets
6521                  * If HZ < 1000, flush all packets.
6522                  */
6523                 napi_gro_flush(n, HZ >= 1000);
6524         }
6525
6526         gro_normal_list(n);
6527
6528         /* Some drivers may have called napi_schedule
6529          * prior to exhausting their budget.
6530          */
6531         if (unlikely(!list_empty(&n->poll_list))) {
6532                 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
6533                              n->dev ? n->dev->name : "backlog");
6534                 return work;
6535         }
6536
6537         *repoll = true;
6538
6539         return work;
6540 }
6541
6542 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
6543 {
6544         bool do_repoll = false;
6545         void *have;
6546         int work;
6547
6548         list_del_init(&n->poll_list);
6549
6550         have = netpoll_poll_lock(n);
6551
6552         work = __napi_poll(n, &do_repoll);
6553
6554         if (do_repoll)
6555                 list_add_tail(&n->poll_list, repoll);
6556
6557         netpoll_poll_unlock(have);
6558
6559         return work;
6560 }
6561
6562 static int napi_thread_wait(struct napi_struct *napi)
6563 {
6564         bool woken = false;
6565
6566         set_current_state(TASK_INTERRUPTIBLE);
6567
6568         while (!kthread_should_stop()) {
6569                 /* Testing SCHED_THREADED bit here to make sure the current
6570                  * kthread owns this napi and could poll on this napi.
6571                  * Testing SCHED bit is not enough because SCHED bit might be
6572                  * set by some other busy poll thread or by napi_disable().
6573                  */
6574                 if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state) || woken) {
6575                         WARN_ON(!list_empty(&napi->poll_list));
6576                         __set_current_state(TASK_RUNNING);
6577                         return 0;
6578                 }
6579
6580                 schedule();
6581                 /* woken being true indicates this thread owns this napi. */
6582                 woken = true;
6583                 set_current_state(TASK_INTERRUPTIBLE);
6584         }
6585         __set_current_state(TASK_RUNNING);
6586
6587         return -1;
6588 }
6589
6590 static int napi_threaded_poll(void *data)
6591 {
6592         struct napi_struct *napi = data;
6593         void *have;
6594
6595         while (!napi_thread_wait(napi)) {
6596                 for (;;) {
6597                         bool repoll = false;
6598
6599                         local_bh_disable();
6600
6601                         have = netpoll_poll_lock(napi);
6602                         __napi_poll(napi, &repoll);
6603                         netpoll_poll_unlock(have);
6604
6605                         local_bh_enable();
6606
6607                         if (!repoll)
6608                                 break;
6609
6610                         cond_resched();
6611                 }
6612         }
6613         return 0;
6614 }
6615
6616 static void skb_defer_free_flush(struct softnet_data *sd)
6617 {
6618         struct sk_buff *skb, *next;
6619         unsigned long flags;
6620
6621         /* Paired with WRITE_ONCE() in skb_attempt_defer_free() */
6622         if (!READ_ONCE(sd->defer_list))
6623                 return;
6624
6625         spin_lock_irqsave(&sd->defer_lock, flags);
6626         skb = sd->defer_list;
6627         sd->defer_list = NULL;
6628         sd->defer_count = 0;
6629         spin_unlock_irqrestore(&sd->defer_lock, flags);
6630
6631         while (skb != NULL) {
6632                 next = skb->next;
6633                 napi_consume_skb(skb, 1);
6634                 skb = next;
6635         }
6636 }
6637
6638 static __latent_entropy void net_rx_action(struct softirq_action *h)
6639 {
6640         struct softnet_data *sd = this_cpu_ptr(&softnet_data);
6641         unsigned long time_limit = jiffies +
6642                 usecs_to_jiffies(READ_ONCE(netdev_budget_usecs));
6643         int budget = READ_ONCE(netdev_budget);
6644         LIST_HEAD(list);
6645         LIST_HEAD(repoll);
6646
6647         local_irq_disable();
6648         list_splice_init(&sd->poll_list, &list);
6649         local_irq_enable();
6650
6651         for (;;) {
6652                 struct napi_struct *n;
6653
6654                 skb_defer_free_flush(sd);
6655
6656                 if (list_empty(&list)) {
6657                         if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
6658                                 goto end;
6659                         break;
6660                 }
6661
6662                 n = list_first_entry(&list, struct napi_struct, poll_list);
6663                 budget -= napi_poll(n, &repoll);
6664
6665                 /* If softirq window is exhausted then punt.
6666                  * Allow this to run for 2 jiffies since which will allow
6667                  * an average latency of 1.5/HZ.
6668                  */
6669                 if (unlikely(budget <= 0 ||
6670                              time_after_eq(jiffies, time_limit))) {
6671                         sd->time_squeeze++;
6672                         break;
6673                 }
6674         }
6675
6676         local_irq_disable();
6677
6678         list_splice_tail_init(&sd->poll_list, &list);
6679         list_splice_tail(&repoll, &list);
6680         list_splice(&list, &sd->poll_list);
6681         if (!list_empty(&sd->poll_list))
6682                 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
6683
6684         net_rps_action_and_irq_enable(sd);
6685 end:;
6686 }
6687
6688 struct netdev_adjacent {
6689         struct net_device *dev;
6690         netdevice_tracker dev_tracker;
6691
6692         /* upper master flag, there can only be one master device per list */
6693         bool master;
6694
6695         /* lookup ignore flag */
6696         bool ignore;
6697
6698         /* counter for the number of times this device was added to us */
6699         u16 ref_nr;
6700
6701         /* private field for the users */
6702         void *private;
6703
6704         struct list_head list;
6705         struct rcu_head rcu;
6706 };
6707
6708 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
6709                                                  struct list_head *adj_list)
6710 {
6711         struct netdev_adjacent *adj;
6712
6713         list_for_each_entry(adj, adj_list, list) {
6714                 if (adj->dev == adj_dev)
6715                         return adj;
6716         }
6717         return NULL;
6718 }
6719
6720 static int ____netdev_has_upper_dev(struct net_device *upper_dev,
6721                                     struct netdev_nested_priv *priv)
6722 {
6723         struct net_device *dev = (struct net_device *)priv->data;
6724
6725         return upper_dev == dev;
6726 }
6727
6728 /**
6729  * netdev_has_upper_dev - Check if device is linked to an upper device
6730  * @dev: device
6731  * @upper_dev: upper device to check
6732  *
6733  * Find out if a device is linked to specified upper device and return true
6734  * in case it is. Note that this checks only immediate upper device,
6735  * not through a complete stack of devices. The caller must hold the RTNL lock.
6736  */
6737 bool netdev_has_upper_dev(struct net_device *dev,
6738                           struct net_device *upper_dev)
6739 {
6740         struct netdev_nested_priv priv = {
6741                 .data = (void *)upper_dev,
6742         };
6743
6744         ASSERT_RTNL();
6745
6746         return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
6747                                              &priv);
6748 }
6749 EXPORT_SYMBOL(netdev_has_upper_dev);
6750
6751 /**
6752  * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device
6753  * @dev: device
6754  * @upper_dev: upper device to check
6755  *
6756  * Find out if a device is linked to specified upper device and return true
6757  * in case it is. Note that this checks the entire upper device chain.
6758  * The caller must hold rcu lock.
6759  */
6760
6761 bool netdev_has_upper_dev_all_rcu(struct net_device *dev,
6762                                   struct net_device *upper_dev)
6763 {
6764         struct netdev_nested_priv priv = {
6765                 .data = (void *)upper_dev,
6766         };
6767
6768         return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
6769                                                &priv);
6770 }
6771 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu);
6772
6773 /**
6774  * netdev_has_any_upper_dev - Check if device is linked to some device
6775  * @dev: device
6776  *
6777  * Find out if a device is linked to an upper device and return true in case
6778  * it is. The caller must hold the RTNL lock.
6779  */
6780 bool netdev_has_any_upper_dev(struct net_device *dev)
6781 {
6782         ASSERT_RTNL();
6783
6784         return !list_empty(&dev->adj_list.upper);
6785 }
6786 EXPORT_SYMBOL(netdev_has_any_upper_dev);
6787
6788 /**
6789  * netdev_master_upper_dev_get - Get master upper device
6790  * @dev: device
6791  *
6792  * Find a master upper device and return pointer to it or NULL in case
6793  * it's not there. The caller must hold the RTNL lock.
6794  */
6795 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
6796 {
6797         struct netdev_adjacent *upper;
6798
6799         ASSERT_RTNL();
6800
6801         if (list_empty(&dev->adj_list.upper))
6802                 return NULL;
6803
6804         upper = list_first_entry(&dev->adj_list.upper,
6805                                  struct netdev_adjacent, list);
6806         if (likely(upper->master))
6807                 return upper->dev;
6808         return NULL;
6809 }
6810 EXPORT_SYMBOL(netdev_master_upper_dev_get);
6811
6812 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev)
6813 {
6814         struct netdev_adjacent *upper;
6815
6816         ASSERT_RTNL();
6817
6818         if (list_empty(&dev->adj_list.upper))
6819                 return NULL;
6820
6821         upper = list_first_entry(&dev->adj_list.upper,
6822                                  struct netdev_adjacent, list);
6823         if (likely(upper->master) && !upper->ignore)
6824                 return upper->dev;
6825         return NULL;
6826 }
6827
6828 /**
6829  * netdev_has_any_lower_dev - Check if device is linked to some device
6830  * @dev: device
6831  *
6832  * Find out if a device is linked to a lower device and return true in case
6833  * it is. The caller must hold the RTNL lock.
6834  */
6835 static bool netdev_has_any_lower_dev(struct net_device *dev)
6836 {
6837         ASSERT_RTNL();
6838
6839         return !list_empty(&dev->adj_list.lower);
6840 }
6841
6842 void *netdev_adjacent_get_private(struct list_head *adj_list)
6843 {
6844         struct netdev_adjacent *adj;
6845
6846         adj = list_entry(adj_list, struct netdev_adjacent, list);
6847
6848         return adj->private;
6849 }
6850 EXPORT_SYMBOL(netdev_adjacent_get_private);
6851
6852 /**
6853  * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
6854  * @dev: device
6855  * @iter: list_head ** of the current position
6856  *
6857  * Gets the next device from the dev's upper list, starting from iter
6858  * position. The caller must hold RCU read lock.
6859  */
6860 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
6861                                                  struct list_head **iter)
6862 {
6863         struct netdev_adjacent *upper;
6864
6865         WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6866
6867         upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6868
6869         if (&upper->list == &dev->adj_list.upper)
6870                 return NULL;
6871
6872         *iter = &upper->list;
6873
6874         return upper->dev;
6875 }
6876 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
6877
6878 static struct net_device *__netdev_next_upper_dev(struct net_device *dev,
6879                                                   struct list_head **iter,
6880                                                   bool *ignore)
6881 {
6882         struct netdev_adjacent *upper;
6883
6884         upper = list_entry((*iter)->next, struct netdev_adjacent, list);
6885
6886         if (&upper->list == &dev->adj_list.upper)
6887                 return NULL;
6888
6889         *iter = &upper->list;
6890         *ignore = upper->ignore;
6891
6892         return upper->dev;
6893 }
6894
6895 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev,
6896                                                     struct list_head **iter)
6897 {
6898         struct netdev_adjacent *upper;
6899
6900         WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6901
6902         upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6903
6904         if (&upper->list == &dev->adj_list.upper)
6905                 return NULL;
6906
6907         *iter = &upper->list;
6908
6909         return upper->dev;
6910 }
6911
6912 static int __netdev_walk_all_upper_dev(struct net_device *dev,
6913                                        int (*fn)(struct net_device *dev,
6914                                          struct netdev_nested_priv *priv),
6915                                        struct netdev_nested_priv *priv)
6916 {
6917         struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
6918         struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
6919         int ret, cur = 0;
6920         bool ignore;
6921
6922         now = dev;
6923         iter = &dev->adj_list.upper;
6924
6925         while (1) {
6926                 if (now != dev) {
6927                         ret = fn(now, priv);
6928                         if (ret)
6929                                 return ret;
6930                 }
6931
6932                 next = NULL;
6933                 while (1) {
6934                         udev = __netdev_next_upper_dev(now, &iter, &ignore);
6935                         if (!udev)
6936                                 break;
6937                         if (ignore)
6938                                 continue;
6939
6940                         next = udev;
6941                         niter = &udev->adj_list.upper;
6942                         dev_stack[cur] = now;
6943                         iter_stack[cur++] = iter;
6944                         break;
6945                 }
6946
6947                 if (!next) {
6948                         if (!cur)
6949                                 return 0;
6950                         next = dev_stack[--cur];
6951                         niter = iter_stack[cur];
6952                 }
6953
6954                 now = next;
6955                 iter = niter;
6956         }
6957
6958         return 0;
6959 }
6960
6961 int netdev_walk_all_upper_dev_rcu(struct net_device *dev,
6962                                   int (*fn)(struct net_device *dev,
6963                                             struct netdev_nested_priv *priv),
6964                                   struct netdev_nested_priv *priv)
6965 {
6966         struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
6967         struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
6968         int ret, cur = 0;
6969
6970         now = dev;
6971         iter = &dev->adj_list.upper;
6972
6973         while (1) {
6974                 if (now != dev) {
6975                         ret = fn(now, priv);
6976                         if (ret)
6977                                 return ret;
6978                 }
6979
6980                 next = NULL;
6981                 while (1) {
6982                         udev = netdev_next_upper_dev_rcu(now, &iter);
6983                         if (!udev)
6984                                 break;
6985
6986                         next = udev;
6987                         niter = &udev->adj_list.upper;
6988                         dev_stack[cur] = now;
6989                         iter_stack[cur++] = iter;
6990                         break;
6991                 }
6992
6993                 if (!next) {
6994                         if (!cur)
6995                                 return 0;
6996                         next = dev_stack[--cur];
6997                         niter = iter_stack[cur];
6998                 }
6999
7000                 now = next;
7001                 iter = niter;
7002         }
7003
7004         return 0;
7005 }
7006 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu);
7007
7008 static bool __netdev_has_upper_dev(struct net_device *dev,
7009                                    struct net_device *upper_dev)
7010 {
7011         struct netdev_nested_priv priv = {
7012                 .flags = 0,
7013                 .data = (void *)upper_dev,
7014         };
7015
7016         ASSERT_RTNL();
7017
7018         return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev,
7019                                            &priv);
7020 }
7021
7022 /**
7023  * netdev_lower_get_next_private - Get the next ->private from the
7024  *                                 lower neighbour list
7025  * @dev: device
7026  * @iter: list_head ** of the current position
7027  *
7028  * Gets the next netdev_adjacent->private from the dev's lower neighbour
7029  * list, starting from iter position. The caller must hold either hold the
7030  * RTNL lock or its own locking that guarantees that the neighbour lower
7031  * list will remain unchanged.
7032  */
7033 void *netdev_lower_get_next_private(struct net_device *dev,
7034                                     struct list_head **iter)
7035 {
7036         struct netdev_adjacent *lower;
7037
7038         lower = list_entry(*iter, struct netdev_adjacent, list);
7039
7040         if (&lower->list == &dev->adj_list.lower)
7041                 return NULL;
7042
7043         *iter = lower->list.next;
7044
7045         return lower->private;
7046 }
7047 EXPORT_SYMBOL(netdev_lower_get_next_private);
7048
7049 /**
7050  * netdev_lower_get_next_private_rcu - Get the next ->private from the
7051  *                                     lower neighbour list, RCU
7052  *                                     variant
7053  * @dev: device
7054  * @iter: list_head ** of the current position
7055  *
7056  * Gets the next netdev_adjacent->private from the dev's lower neighbour
7057  * list, starting from iter position. The caller must hold RCU read lock.
7058  */
7059 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
7060                                         struct list_head **iter)
7061 {
7062         struct netdev_adjacent *lower;
7063
7064         WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
7065
7066         lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7067
7068         if (&lower->list == &dev->adj_list.lower)
7069                 return NULL;
7070
7071         *iter = &lower->list;
7072
7073         return lower->private;
7074 }
7075 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
7076
7077 /**
7078  * netdev_lower_get_next - Get the next device from the lower neighbour
7079  *                         list
7080  * @dev: device
7081  * @iter: list_head ** of the current position
7082  *
7083  * Gets the next netdev_adjacent from the dev's lower neighbour
7084  * list, starting from iter position. The caller must hold RTNL lock or
7085  * its own locking that guarantees that the neighbour lower
7086  * list will remain unchanged.
7087  */
7088 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
7089 {
7090         struct netdev_adjacent *lower;
7091
7092         lower = list_entry(*iter, struct netdev_adjacent, list);
7093
7094         if (&lower->list == &dev->adj_list.lower)
7095                 return NULL;
7096
7097         *iter = lower->list.next;
7098
7099         return lower->dev;
7100 }
7101 EXPORT_SYMBOL(netdev_lower_get_next);
7102
7103 static struct net_device *netdev_next_lower_dev(struct net_device *dev,
7104                                                 struct list_head **iter)
7105 {
7106         struct netdev_adjacent *lower;
7107
7108         lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7109
7110         if (&lower->list == &dev->adj_list.lower)
7111                 return NULL;
7112
7113         *iter = &lower->list;
7114
7115         return lower->dev;
7116 }
7117
7118 static struct net_device *__netdev_next_lower_dev(struct net_device *dev,
7119                                                   struct list_head **iter,
7120                                                   bool *ignore)
7121 {
7122         struct netdev_adjacent *lower;
7123
7124         lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7125
7126         if (&lower->list == &dev->adj_list.lower)
7127                 return NULL;
7128
7129         *iter = &lower->list;
7130         *ignore = lower->ignore;
7131
7132         return lower->dev;
7133 }
7134
7135 int netdev_walk_all_lower_dev(struct net_device *dev,
7136                               int (*fn)(struct net_device *dev,
7137                                         struct netdev_nested_priv *priv),
7138                               struct netdev_nested_priv *priv)
7139 {
7140         struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7141         struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7142         int ret, cur = 0;
7143
7144         now = dev;
7145         iter = &dev->adj_list.lower;
7146
7147         while (1) {
7148                 if (now != dev) {
7149                         ret = fn(now, priv);
7150                         if (ret)
7151                                 return ret;
7152                 }
7153
7154                 next = NULL;
7155                 while (1) {
7156                         ldev = netdev_next_lower_dev(now, &iter);
7157                         if (!ldev)
7158                                 break;
7159
7160                         next = ldev;
7161                         niter = &ldev->adj_list.lower;
7162                         dev_stack[cur] = now;
7163                         iter_stack[cur++] = iter;
7164                         break;
7165                 }
7166
7167                 if (!next) {
7168                         if (!cur)
7169                                 return 0;
7170                         next = dev_stack[--cur];
7171                         niter = iter_stack[cur];
7172                 }
7173
7174                 now = next;
7175                 iter = niter;
7176         }
7177
7178         return 0;
7179 }
7180 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev);
7181
7182 static int __netdev_walk_all_lower_dev(struct net_device *dev,
7183                                        int (*fn)(struct net_device *dev,
7184                                          struct netdev_nested_priv *priv),
7185                                        struct netdev_nested_priv *priv)
7186 {
7187         struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7188         struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7189         int ret, cur = 0;
7190         bool ignore;
7191
7192         now = dev;
7193         iter = &dev->adj_list.lower;
7194
7195         while (1) {
7196                 if (now != dev) {
7197                         ret = fn(now, priv);
7198                         if (ret)
7199                                 return ret;
7200                 }
7201
7202                 next = NULL;
7203                 while (1) {
7204                         ldev = __netdev_next_lower_dev(now, &iter, &ignore);
7205                         if (!ldev)
7206                                 break;
7207                         if (ignore)
7208                                 continue;
7209
7210                         next = ldev;
7211                         niter = &ldev->adj_list.lower;
7212                         dev_stack[cur] = now;
7213                         iter_stack[cur++] = iter;
7214                         break;
7215                 }
7216
7217                 if (!next) {
7218                         if (!cur)
7219                                 return 0;
7220                         next = dev_stack[--cur];
7221                         niter = iter_stack[cur];
7222                 }
7223
7224                 now = next;
7225                 iter = niter;
7226         }
7227
7228         return 0;
7229 }
7230
7231 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev,
7232                                              struct list_head **iter)
7233 {
7234         struct netdev_adjacent *lower;
7235
7236         lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7237         if (&lower->list == &dev->adj_list.lower)
7238                 return NULL;
7239
7240         *iter = &lower->list;
7241
7242         return lower->dev;
7243 }
7244 EXPORT_SYMBOL(netdev_next_lower_dev_rcu);
7245
7246 static u8 __netdev_upper_depth(struct net_device *dev)
7247 {
7248         struct net_device *udev;
7249         struct list_head *iter;
7250         u8 max_depth = 0;
7251         bool ignore;
7252
7253         for (iter = &dev->adj_list.upper,
7254              udev = __netdev_next_upper_dev(dev, &iter, &ignore);
7255              udev;
7256              udev = __netdev_next_upper_dev(dev, &iter, &ignore)) {
7257                 if (ignore)
7258                         continue;
7259                 if (max_depth < udev->upper_level)
7260                         max_depth = udev->upper_level;
7261         }
7262
7263         return max_depth;
7264 }
7265
7266 static u8 __netdev_lower_depth(struct net_device *dev)
7267 {
7268         struct net_device *ldev;
7269         struct list_head *iter;
7270         u8 max_depth = 0;
7271         bool ignore;
7272
7273         for (iter = &dev->adj_list.lower,
7274              ldev = __netdev_next_lower_dev(dev, &iter, &ignore);
7275              ldev;
7276              ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) {
7277                 if (ignore)
7278                         continue;
7279                 if (max_depth < ldev->lower_level)
7280                         max_depth = ldev->lower_level;
7281         }
7282
7283         return max_depth;
7284 }
7285
7286 static int __netdev_update_upper_level(struct net_device *dev,
7287                                        struct netdev_nested_priv *__unused)
7288 {
7289         dev->upper_level = __netdev_upper_depth(dev) + 1;
7290         return 0;
7291 }
7292
7293 #ifdef CONFIG_LOCKDEP
7294 static LIST_HEAD(net_unlink_list);
7295
7296 static void net_unlink_todo(struct net_device *dev)
7297 {
7298         if (list_empty(&dev->unlink_list))
7299                 list_add_tail(&dev->unlink_list, &net_unlink_list);
7300 }
7301 #endif
7302
7303 static int __netdev_update_lower_level(struct net_device *dev,
7304                                        struct netdev_nested_priv *priv)
7305 {
7306         dev->lower_level = __netdev_lower_depth(dev) + 1;
7307
7308 #ifdef CONFIG_LOCKDEP
7309         if (!priv)
7310                 return 0;
7311
7312         if (priv->flags & NESTED_SYNC_IMM)
7313                 dev->nested_level = dev->lower_level - 1;
7314         if (priv->flags & NESTED_SYNC_TODO)
7315                 net_unlink_todo(dev);
7316 #endif
7317         return 0;
7318 }
7319
7320 int netdev_walk_all_lower_dev_rcu(struct net_device *dev,
7321                                   int (*fn)(struct net_device *dev,
7322                                             struct netdev_nested_priv *priv),
7323                                   struct netdev_nested_priv *priv)
7324 {
7325         struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7326         struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7327         int ret, cur = 0;
7328
7329         now = dev;
7330         iter = &dev->adj_list.lower;
7331
7332         while (1) {
7333                 if (now != dev) {
7334                         ret = fn(now, priv);
7335                         if (ret)
7336                                 return ret;
7337                 }
7338
7339                 next = NULL;
7340                 while (1) {
7341                         ldev = netdev_next_lower_dev_rcu(now, &iter);
7342                         if (!ldev)
7343                                 break;
7344
7345                         next = ldev;
7346                         niter = &ldev->adj_list.lower;
7347                         dev_stack[cur] = now;
7348                         iter_stack[cur++] = iter;
7349                         break;
7350                 }
7351
7352                 if (!next) {
7353                         if (!cur)
7354                                 return 0;
7355                         next = dev_stack[--cur];
7356                         niter = iter_stack[cur];
7357                 }
7358
7359                 now = next;
7360                 iter = niter;
7361         }
7362
7363         return 0;
7364 }
7365 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu);
7366
7367 /**
7368  * netdev_lower_get_first_private_rcu - Get the first ->private from the
7369  *                                     lower neighbour list, RCU
7370  *                                     variant
7371  * @dev: device
7372  *
7373  * Gets the first netdev_adjacent->private from the dev's lower neighbour
7374  * list. The caller must hold RCU read lock.
7375  */
7376 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
7377 {
7378         struct netdev_adjacent *lower;
7379
7380         lower = list_first_or_null_rcu(&dev->adj_list.lower,
7381                         struct netdev_adjacent, list);
7382         if (lower)
7383                 return lower->private;
7384         return NULL;
7385 }
7386 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
7387
7388 /**
7389  * netdev_master_upper_dev_get_rcu - Get master upper device
7390  * @dev: device
7391  *
7392  * Find a master upper device and return pointer to it or NULL in case
7393  * it's not there. The caller must hold the RCU read lock.
7394  */
7395 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
7396 {
7397         struct netdev_adjacent *upper;
7398
7399         upper = list_first_or_null_rcu(&dev->adj_list.upper,
7400                                        struct netdev_adjacent, list);
7401         if (upper && likely(upper->master))
7402                 return upper->dev;
7403         return NULL;
7404 }
7405 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
7406
7407 static int netdev_adjacent_sysfs_add(struct net_device *dev,
7408                               struct net_device *adj_dev,
7409                               struct list_head *dev_list)
7410 {
7411         char linkname[IFNAMSIZ+7];
7412
7413         sprintf(linkname, dev_list == &dev->adj_list.upper ?
7414                 "upper_%s" : "lower_%s", adj_dev->name);
7415         return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
7416                                  linkname);
7417 }
7418 static void netdev_adjacent_sysfs_del(struct net_device *dev,
7419                                char *name,
7420                                struct list_head *dev_list)
7421 {
7422         char linkname[IFNAMSIZ+7];
7423
7424         sprintf(linkname, dev_list == &dev->adj_list.upper ?
7425                 "upper_%s" : "lower_%s", name);
7426         sysfs_remove_link(&(dev->dev.kobj), linkname);
7427 }
7428
7429 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
7430                                                  struct net_device *adj_dev,
7431                                                  struct list_head *dev_list)
7432 {
7433         return (dev_list == &dev->adj_list.upper ||
7434                 dev_list == &dev->adj_list.lower) &&
7435                 net_eq(dev_net(dev), dev_net(adj_dev));
7436 }
7437
7438 static int __netdev_adjacent_dev_insert(struct net_device *dev,
7439                                         struct net_device *adj_dev,
7440                                         struct list_head *dev_list,
7441                                         void *private, bool master)
7442 {
7443         struct netdev_adjacent *adj;
7444         int ret;
7445
7446         adj = __netdev_find_adj(adj_dev, dev_list);
7447
7448         if (adj) {
7449                 adj->ref_nr += 1;
7450                 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n",
7451                          dev->name, adj_dev->name, adj->ref_nr);
7452
7453                 return 0;
7454         }
7455
7456         adj = kmalloc(sizeof(*adj), GFP_KERNEL);
7457         if (!adj)
7458                 return -ENOMEM;
7459
7460         adj->dev = adj_dev;
7461         adj->master = master;
7462         adj->ref_nr = 1;
7463         adj->private = private;
7464         adj->ignore = false;
7465         netdev_hold(adj_dev, &adj->dev_tracker, GFP_KERNEL);
7466
7467         pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n",
7468                  dev->name, adj_dev->name, adj->ref_nr, adj_dev->name);
7469
7470         if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
7471                 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
7472                 if (ret)
7473                         goto free_adj;
7474         }
7475
7476         /* Ensure that master link is always the first item in list. */
7477         if (master) {
7478                 ret = sysfs_create_link(&(dev->dev.kobj),
7479                                         &(adj_dev->dev.kobj), "master");
7480                 if (ret)
7481                         goto remove_symlinks;
7482
7483                 list_add_rcu(&adj->list, dev_list);
7484         } else {
7485                 list_add_tail_rcu(&adj->list, dev_list);
7486         }
7487
7488         return 0;
7489
7490 remove_symlinks:
7491         if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7492                 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7493 free_adj:
7494         netdev_put(adj_dev, &adj->dev_tracker);
7495         kfree(adj);
7496
7497         return ret;
7498 }
7499
7500 static void __netdev_adjacent_dev_remove(struct net_device *dev,
7501                                          struct net_device *adj_dev,
7502                                          u16 ref_nr,
7503                                          struct list_head *dev_list)
7504 {
7505         struct netdev_adjacent *adj;
7506
7507         pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n",
7508                  dev->name, adj_dev->name, ref_nr);
7509
7510         adj = __netdev_find_adj(adj_dev, dev_list);
7511
7512         if (!adj) {
7513                 pr_err("Adjacency does not exist for device %s from %s\n",
7514                        dev->name, adj_dev->name);
7515                 WARN_ON(1);
7516                 return;
7517         }
7518
7519         if (adj->ref_nr > ref_nr) {
7520                 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n",
7521                          dev->name, adj_dev->name, ref_nr,
7522                          adj->ref_nr - ref_nr);
7523                 adj->ref_nr -= ref_nr;
7524                 return;
7525         }
7526
7527         if (adj->master)
7528                 sysfs_remove_link(&(dev->dev.kobj), "master");
7529
7530         if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7531                 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7532
7533         list_del_rcu(&adj->list);
7534         pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n",
7535                  adj_dev->name, dev->name, adj_dev->name);
7536         netdev_put(adj_dev, &adj->dev_tracker);
7537         kfree_rcu(adj, rcu);
7538 }
7539
7540 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
7541                                             struct net_device *upper_dev,
7542                                             struct list_head *up_list,
7543                                             struct list_head *down_list,
7544                                             void *private, bool master)
7545 {
7546         int ret;
7547
7548         ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list,
7549                                            private, master);
7550         if (ret)
7551                 return ret;
7552
7553         ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list,
7554                                            private, false);
7555         if (ret) {
7556                 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list);
7557                 return ret;
7558         }
7559
7560         return 0;
7561 }
7562
7563 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
7564                                                struct net_device *upper_dev,
7565                                                u16 ref_nr,
7566                                                struct list_head *up_list,
7567                                                struct list_head *down_list)
7568 {
7569         __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
7570         __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
7571 }
7572
7573 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
7574                                                 struct net_device *upper_dev,
7575                                                 void *private, bool master)
7576 {
7577         return __netdev_adjacent_dev_link_lists(dev, upper_dev,
7578                                                 &dev->adj_list.upper,
7579                                                 &upper_dev->adj_list.lower,
7580                                                 private, master);
7581 }
7582
7583 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
7584                                                    struct net_device *upper_dev)
7585 {
7586         __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
7587                                            &dev->adj_list.upper,
7588                                            &upper_dev->adj_list.lower);
7589 }
7590
7591 static int __netdev_upper_dev_link(struct net_device *dev,
7592                                    struct net_device *upper_dev, bool master,
7593                                    void *upper_priv, void *upper_info,
7594                                    struct netdev_nested_priv *priv,
7595                                    struct netlink_ext_ack *extack)
7596 {
7597         struct netdev_notifier_changeupper_info changeupper_info = {
7598                 .info = {
7599                         .dev = dev,
7600                         .extack = extack,
7601                 },
7602                 .upper_dev = upper_dev,
7603                 .master = master,
7604                 .linking = true,
7605                 .upper_info = upper_info,
7606         };
7607         struct net_device *master_dev;
7608         int ret = 0;
7609
7610         ASSERT_RTNL();
7611
7612         if (dev == upper_dev)
7613                 return -EBUSY;
7614
7615         /* To prevent loops, check if dev is not upper device to upper_dev. */
7616         if (__netdev_has_upper_dev(upper_dev, dev))
7617                 return -EBUSY;
7618
7619         if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV)
7620                 return -EMLINK;
7621
7622         if (!master) {
7623                 if (__netdev_has_upper_dev(dev, upper_dev))
7624                         return -EEXIST;
7625         } else {
7626                 master_dev = __netdev_master_upper_dev_get(dev);
7627                 if (master_dev)
7628                         return master_dev == upper_dev ? -EEXIST : -EBUSY;
7629         }
7630
7631         ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7632                                             &changeupper_info.info);
7633         ret = notifier_to_errno(ret);
7634         if (ret)
7635                 return ret;
7636
7637         ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
7638                                                    master);
7639         if (ret)
7640                 return ret;
7641
7642         ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7643                                             &changeupper_info.info);
7644         ret = notifier_to_errno(ret);
7645         if (ret)
7646                 goto rollback;
7647
7648         __netdev_update_upper_level(dev, NULL);
7649         __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
7650
7651         __netdev_update_lower_level(upper_dev, priv);
7652         __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
7653                                     priv);
7654
7655         return 0;
7656
7657 rollback:
7658         __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7659
7660         return ret;
7661 }
7662
7663 /**
7664  * netdev_upper_dev_link - Add a link to the upper device
7665  * @dev: device
7666  * @upper_dev: new upper device
7667  * @extack: netlink extended ack
7668  *
7669  * Adds a link to device which is upper to this one. The caller must hold
7670  * the RTNL lock. On a failure a negative errno code is returned.
7671  * On success the reference counts are adjusted and the function
7672  * returns zero.
7673  */
7674 int netdev_upper_dev_link(struct net_device *dev,
7675                           struct net_device *upper_dev,
7676                           struct netlink_ext_ack *extack)
7677 {
7678         struct netdev_nested_priv priv = {
7679                 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
7680                 .data = NULL,
7681         };
7682
7683         return __netdev_upper_dev_link(dev, upper_dev, false,
7684                                        NULL, NULL, &priv, extack);
7685 }
7686 EXPORT_SYMBOL(netdev_upper_dev_link);
7687
7688 /**
7689  * netdev_master_upper_dev_link - Add a master link to the upper device
7690  * @dev: device
7691  * @upper_dev: new upper device
7692  * @upper_priv: upper device private
7693  * @upper_info: upper info to be passed down via notifier
7694  * @extack: netlink extended ack
7695  *
7696  * Adds a link to device which is upper to this one. In this case, only
7697  * one master upper device can be linked, although other non-master devices
7698  * might be linked as well. The caller must hold the RTNL lock.
7699  * On a failure a negative errno code is returned. On success the reference
7700  * counts are adjusted and the function returns zero.
7701  */
7702 int netdev_master_upper_dev_link(struct net_device *dev,
7703                                  struct net_device *upper_dev,
7704                                  void *upper_priv, void *upper_info,
7705                                  struct netlink_ext_ack *extack)
7706 {
7707         struct netdev_nested_priv priv = {
7708                 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
7709                 .data = NULL,
7710         };
7711
7712         return __netdev_upper_dev_link(dev, upper_dev, true,
7713                                        upper_priv, upper_info, &priv, extack);
7714 }
7715 EXPORT_SYMBOL(netdev_master_upper_dev_link);
7716
7717 static void __netdev_upper_dev_unlink(struct net_device *dev,
7718                                       struct net_device *upper_dev,
7719                                       struct netdev_nested_priv *priv)
7720 {
7721         struct netdev_notifier_changeupper_info changeupper_info = {
7722                 .info = {
7723                         .dev = dev,
7724                 },
7725                 .upper_dev = upper_dev,
7726                 .linking = false,
7727         };
7728
7729         ASSERT_RTNL();
7730
7731         changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
7732
7733         call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7734                                       &changeupper_info.info);
7735
7736         __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7737
7738         call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7739                                       &changeupper_info.info);
7740
7741         __netdev_update_upper_level(dev, NULL);
7742         __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
7743
7744         __netdev_update_lower_level(upper_dev, priv);
7745         __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
7746                                     priv);
7747 }
7748
7749 /**
7750  * netdev_upper_dev_unlink - Removes a link to upper device
7751  * @dev: device
7752  * @upper_dev: new upper device
7753  *
7754  * Removes a link to device which is upper to this one. The caller must hold
7755  * the RTNL lock.
7756  */
7757 void netdev_upper_dev_unlink(struct net_device *dev,
7758                              struct net_device *upper_dev)
7759 {
7760         struct netdev_nested_priv priv = {
7761                 .flags = NESTED_SYNC_TODO,
7762                 .data = NULL,
7763         };
7764
7765         __netdev_upper_dev_unlink(dev, upper_dev, &priv);
7766 }
7767 EXPORT_SYMBOL(netdev_upper_dev_unlink);
7768
7769 static void __netdev_adjacent_dev_set(struct net_device *upper_dev,
7770                                       struct net_device *lower_dev,
7771                                       bool val)
7772 {
7773         struct netdev_adjacent *adj;
7774
7775         adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower);
7776         if (adj)
7777                 adj->ignore = val;
7778
7779         adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper);
7780         if (adj)
7781                 adj->ignore = val;
7782 }
7783
7784 static void netdev_adjacent_dev_disable(struct net_device *upper_dev,
7785                                         struct net_device *lower_dev)
7786 {
7787         __netdev_adjacent_dev_set(upper_dev, lower_dev, true);
7788 }
7789
7790 static void netdev_adjacent_dev_enable(struct net_device *upper_dev,
7791                                        struct net_device *lower_dev)
7792 {
7793         __netdev_adjacent_dev_set(upper_dev, lower_dev, false);
7794 }
7795
7796 int netdev_adjacent_change_prepare(struct net_device *old_dev,
7797                                    struct net_device *new_dev,
7798                                    struct net_device *dev,
7799                                    struct netlink_ext_ack *extack)
7800 {
7801         struct netdev_nested_priv priv = {
7802                 .flags = 0,
7803                 .data = NULL,
7804         };
7805         int err;
7806
7807         if (!new_dev)
7808                 return 0;
7809
7810         if (old_dev && new_dev != old_dev)
7811                 netdev_adjacent_dev_disable(dev, old_dev);
7812         err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv,
7813                                       extack);
7814         if (err) {
7815                 if (old_dev && new_dev != old_dev)
7816                         netdev_adjacent_dev_enable(dev, old_dev);
7817                 return err;
7818         }
7819
7820         return 0;
7821 }
7822 EXPORT_SYMBOL(netdev_adjacent_change_prepare);
7823
7824 void netdev_adjacent_change_commit(struct net_device *old_dev,
7825                                    struct net_device *new_dev,
7826                                    struct net_device *dev)
7827 {
7828         struct netdev_nested_priv priv = {
7829                 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
7830                 .data = NULL,
7831         };
7832
7833         if (!new_dev || !old_dev)
7834                 return;
7835
7836         if (new_dev == old_dev)
7837                 return;
7838
7839         netdev_adjacent_dev_enable(dev, old_dev);
7840         __netdev_upper_dev_unlink(old_dev, dev, &priv);
7841 }
7842 EXPORT_SYMBOL(netdev_adjacent_change_commit);
7843
7844 void netdev_adjacent_change_abort(struct net_device *old_dev,
7845                                   struct net_device *new_dev,
7846                                   struct net_device *dev)
7847 {
7848         struct netdev_nested_priv priv = {
7849                 .flags = 0,
7850                 .data = NULL,
7851         };
7852
7853         if (!new_dev)
7854                 return;
7855
7856         if (old_dev && new_dev != old_dev)
7857                 netdev_adjacent_dev_enable(dev, old_dev);
7858
7859         __netdev_upper_dev_unlink(new_dev, dev, &priv);
7860 }
7861 EXPORT_SYMBOL(netdev_adjacent_change_abort);
7862
7863 /**
7864  * netdev_bonding_info_change - Dispatch event about slave change
7865  * @dev: device
7866  * @bonding_info: info to dispatch
7867  *
7868  * Send NETDEV_BONDING_INFO to netdev notifiers with info.
7869  * The caller must hold the RTNL lock.
7870  */
7871 void netdev_bonding_info_change(struct net_device *dev,
7872                                 struct netdev_bonding_info *bonding_info)
7873 {
7874         struct netdev_notifier_bonding_info info = {
7875                 .info.dev = dev,
7876         };
7877
7878         memcpy(&info.bonding_info, bonding_info,
7879                sizeof(struct netdev_bonding_info));
7880         call_netdevice_notifiers_info(NETDEV_BONDING_INFO,
7881                                       &info.info);
7882 }
7883 EXPORT_SYMBOL(netdev_bonding_info_change);
7884
7885 static int netdev_offload_xstats_enable_l3(struct net_device *dev,
7886                                            struct netlink_ext_ack *extack)
7887 {
7888         struct netdev_notifier_offload_xstats_info info = {
7889                 .info.dev = dev,
7890                 .info.extack = extack,
7891                 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3,
7892         };
7893         int err;
7894         int rc;
7895
7896         dev->offload_xstats_l3 = kzalloc(sizeof(*dev->offload_xstats_l3),
7897                                          GFP_KERNEL);
7898         if (!dev->offload_xstats_l3)
7899                 return -ENOMEM;
7900
7901         rc = call_netdevice_notifiers_info_robust(NETDEV_OFFLOAD_XSTATS_ENABLE,
7902                                                   NETDEV_OFFLOAD_XSTATS_DISABLE,
7903                                                   &info.info);
7904         err = notifier_to_errno(rc);
7905         if (err)
7906                 goto free_stats;
7907
7908         return 0;
7909
7910 free_stats:
7911         kfree(dev->offload_xstats_l3);
7912         dev->offload_xstats_l3 = NULL;
7913         return err;
7914 }
7915
7916 int netdev_offload_xstats_enable(struct net_device *dev,
7917                                  enum netdev_offload_xstats_type type,
7918                                  struct netlink_ext_ack *extack)
7919 {
7920         ASSERT_RTNL();
7921
7922         if (netdev_offload_xstats_enabled(dev, type))
7923                 return -EALREADY;
7924
7925         switch (type) {
7926         case NETDEV_OFFLOAD_XSTATS_TYPE_L3:
7927                 return netdev_offload_xstats_enable_l3(dev, extack);
7928         }
7929
7930         WARN_ON(1);
7931         return -EINVAL;
7932 }
7933 EXPORT_SYMBOL(netdev_offload_xstats_enable);
7934
7935 static void netdev_offload_xstats_disable_l3(struct net_device *dev)
7936 {
7937         struct netdev_notifier_offload_xstats_info info = {
7938                 .info.dev = dev,
7939                 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3,
7940         };
7941
7942         call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_DISABLE,
7943                                       &info.info);
7944         kfree(dev->offload_xstats_l3);
7945         dev->offload_xstats_l3 = NULL;
7946 }
7947
7948 int netdev_offload_xstats_disable(struct net_device *dev,
7949                                   enum netdev_offload_xstats_type type)
7950 {
7951         ASSERT_RTNL();
7952
7953         if (!netdev_offload_xstats_enabled(dev, type))
7954                 return -EALREADY;
7955
7956         switch (type) {
7957         case NETDEV_OFFLOAD_XSTATS_TYPE_L3:
7958                 netdev_offload_xstats_disable_l3(dev);
7959                 return 0;
7960         }
7961
7962         WARN_ON(1);
7963         return -EINVAL;
7964 }
7965 EXPORT_SYMBOL(netdev_offload_xstats_disable);
7966
7967 static void netdev_offload_xstats_disable_all(struct net_device *dev)
7968 {
7969         netdev_offload_xstats_disable(dev, NETDEV_OFFLOAD_XSTATS_TYPE_L3);
7970 }
7971
7972 static struct rtnl_hw_stats64 *
7973 netdev_offload_xstats_get_ptr(const struct net_device *dev,
7974                               enum netdev_offload_xstats_type type)
7975 {
7976         switch (type) {
7977         case NETDEV_OFFLOAD_XSTATS_TYPE_L3:
7978                 return dev->offload_xstats_l3;
7979         }
7980
7981         WARN_ON(1);
7982         return NULL;
7983 }
7984
7985 bool netdev_offload_xstats_enabled(const struct net_device *dev,
7986                                    enum netdev_offload_xstats_type type)
7987 {
7988         ASSERT_RTNL();
7989
7990         return netdev_offload_xstats_get_ptr(dev, type);
7991 }
7992 EXPORT_SYMBOL(netdev_offload_xstats_enabled);
7993
7994 struct netdev_notifier_offload_xstats_ru {
7995         bool used;
7996 };
7997
7998 struct netdev_notifier_offload_xstats_rd {
7999         struct rtnl_hw_stats64 stats;
8000         bool used;
8001 };
8002
8003 static void netdev_hw_stats64_add(struct rtnl_hw_stats64 *dest,
8004                                   const struct rtnl_hw_stats64 *src)
8005 {
8006         dest->rx_packets          += src->rx_packets;
8007         dest->tx_packets          += src->tx_packets;
8008         dest->rx_bytes            += src->rx_bytes;
8009         dest->tx_bytes            += src->tx_bytes;
8010         dest->rx_errors           += src->rx_errors;
8011         dest->tx_errors           += src->tx_errors;
8012         dest->rx_dropped          += src->rx_dropped;
8013         dest->tx_dropped          += src->tx_dropped;
8014         dest->multicast           += src->multicast;
8015 }
8016
8017 static int netdev_offload_xstats_get_used(struct net_device *dev,
8018                                           enum netdev_offload_xstats_type type,
8019                                           bool *p_used,
8020                                           struct netlink_ext_ack *extack)
8021 {
8022         struct netdev_notifier_offload_xstats_ru report_used = {};
8023         struct netdev_notifier_offload_xstats_info info = {
8024                 .info.dev = dev,
8025                 .info.extack = extack,
8026                 .type = type,
8027                 .report_used = &report_used,
8028         };
8029         int rc;
8030
8031         WARN_ON(!netdev_offload_xstats_enabled(dev, type));
8032         rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_USED,
8033                                            &info.info);
8034         *p_used = report_used.used;
8035         return notifier_to_errno(rc);
8036 }
8037
8038 static int netdev_offload_xstats_get_stats(struct net_device *dev,
8039                                            enum netdev_offload_xstats_type type,
8040                                            struct rtnl_hw_stats64 *p_stats,
8041                                            bool *p_used,
8042                                            struct netlink_ext_ack *extack)
8043 {
8044         struct netdev_notifier_offload_xstats_rd report_delta = {};
8045         struct netdev_notifier_offload_xstats_info info = {
8046                 .info.dev = dev,
8047                 .info.extack = extack,
8048                 .type = type,
8049                 .report_delta = &report_delta,
8050         };
8051         struct rtnl_hw_stats64 *stats;
8052         int rc;
8053
8054         stats = netdev_offload_xstats_get_ptr(dev, type);
8055         if (WARN_ON(!stats))
8056                 return -EINVAL;
8057
8058         rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_DELTA,
8059                                            &info.info);
8060
8061         /* Cache whatever we got, even if there was an error, otherwise the
8062          * successful stats retrievals would get lost.
8063          */
8064         netdev_hw_stats64_add(stats, &report_delta.stats);
8065
8066         if (p_stats)
8067                 *p_stats = *stats;
8068         *p_used = report_delta.used;
8069
8070         return notifier_to_errno(rc);
8071 }
8072
8073 int netdev_offload_xstats_get(struct net_device *dev,
8074                               enum netdev_offload_xstats_type type,
8075                               struct rtnl_hw_stats64 *p_stats, bool *p_used,
8076                               struct netlink_ext_ack *extack)
8077 {
8078         ASSERT_RTNL();
8079
8080         if (p_stats)
8081                 return netdev_offload_xstats_get_stats(dev, type, p_stats,
8082                                                        p_used, extack);
8083         else
8084                 return netdev_offload_xstats_get_used(dev, type, p_used,
8085                                                       extack);
8086 }
8087 EXPORT_SYMBOL(netdev_offload_xstats_get);
8088
8089 void
8090 netdev_offload_xstats_report_delta(struct netdev_notifier_offload_xstats_rd *report_delta,
8091                                    const struct rtnl_hw_stats64 *stats)
8092 {
8093         report_delta->used = true;
8094         netdev_hw_stats64_add(&report_delta->stats, stats);
8095 }
8096 EXPORT_SYMBOL(netdev_offload_xstats_report_delta);
8097
8098 void
8099 netdev_offload_xstats_report_used(struct netdev_notifier_offload_xstats_ru *report_used)
8100 {
8101         report_used->used = true;
8102 }
8103 EXPORT_SYMBOL(netdev_offload_xstats_report_used);
8104
8105 void netdev_offload_xstats_push_delta(struct net_device *dev,
8106                                       enum netdev_offload_xstats_type type,
8107                                       const struct rtnl_hw_stats64 *p_stats)
8108 {
8109         struct rtnl_hw_stats64 *stats;
8110
8111         ASSERT_RTNL();
8112
8113         stats = netdev_offload_xstats_get_ptr(dev, type);
8114         if (WARN_ON(!stats))
8115                 return;
8116
8117         netdev_hw_stats64_add(stats, p_stats);
8118 }
8119 EXPORT_SYMBOL(netdev_offload_xstats_push_delta);
8120
8121 /**
8122  * netdev_get_xmit_slave - Get the xmit slave of master device
8123  * @dev: device
8124  * @skb: The packet
8125  * @all_slaves: assume all the slaves are active
8126  *
8127  * The reference counters are not incremented so the caller must be
8128  * careful with locks. The caller must hold RCU lock.
8129  * %NULL is returned if no slave is found.
8130  */
8131
8132 struct net_device *netdev_get_xmit_slave(struct net_device *dev,
8133                                          struct sk_buff *skb,
8134                                          bool all_slaves)
8135 {
8136         const struct net_device_ops *ops = dev->netdev_ops;
8137
8138         if (!ops->ndo_get_xmit_slave)
8139                 return NULL;
8140         return ops->ndo_get_xmit_slave(dev, skb, all_slaves);
8141 }
8142 EXPORT_SYMBOL(netdev_get_xmit_slave);
8143
8144 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev,
8145                                                   struct sock *sk)
8146 {
8147         const struct net_device_ops *ops = dev->netdev_ops;
8148
8149         if (!ops->ndo_sk_get_lower_dev)
8150                 return NULL;
8151         return ops->ndo_sk_get_lower_dev(dev, sk);
8152 }
8153
8154 /**
8155  * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket
8156  * @dev: device
8157  * @sk: the socket
8158  *
8159  * %NULL is returned if no lower device is found.
8160  */
8161
8162 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev,
8163                                             struct sock *sk)
8164 {
8165         struct net_device *lower;
8166
8167         lower = netdev_sk_get_lower_dev(dev, sk);
8168         while (lower) {
8169                 dev = lower;
8170                 lower = netdev_sk_get_lower_dev(dev, sk);
8171         }
8172
8173         return dev;
8174 }
8175 EXPORT_SYMBOL(netdev_sk_get_lowest_dev);
8176
8177 static void netdev_adjacent_add_links(struct net_device *dev)
8178 {
8179         struct netdev_adjacent *iter;
8180
8181         struct net *net = dev_net(dev);
8182
8183         list_for_each_entry(iter, &dev->adj_list.upper, list) {
8184                 if (!net_eq(net, dev_net(iter->dev)))
8185                         continue;
8186                 netdev_adjacent_sysfs_add(iter->dev, dev,
8187                                           &iter->dev->adj_list.lower);
8188                 netdev_adjacent_sysfs_add(dev, iter->dev,
8189                                           &dev->adj_list.upper);
8190         }
8191
8192         list_for_each_entry(iter, &dev->adj_list.lower, list) {
8193                 if (!net_eq(net, dev_net(iter->dev)))
8194                         continue;
8195                 netdev_adjacent_sysfs_add(iter->dev, dev,
8196                                           &iter->dev->adj_list.upper);
8197                 netdev_adjacent_sysfs_add(dev, iter->dev,
8198                                           &dev->adj_list.lower);
8199         }
8200 }
8201
8202 static void netdev_adjacent_del_links(struct net_device *dev)
8203 {
8204         struct netdev_adjacent *iter;
8205
8206         struct net *net = dev_net(dev);
8207
8208         list_for_each_entry(iter, &dev->adj_list.upper, list) {
8209                 if (!net_eq(net, dev_net(iter->dev)))
8210                         continue;
8211                 netdev_adjacent_sysfs_del(iter->dev, dev->name,
8212                                           &iter->dev->adj_list.lower);
8213                 netdev_adjacent_sysfs_del(dev, iter->dev->name,
8214                                           &dev->adj_list.upper);
8215         }
8216
8217         list_for_each_entry(iter, &dev->adj_list.lower, list) {
8218                 if (!net_eq(net, dev_net(iter->dev)))
8219                         continue;
8220                 netdev_adjacent_sysfs_del(iter->dev, dev->name,
8221                                           &iter->dev->adj_list.upper);
8222                 netdev_adjacent_sysfs_del(dev, iter->dev->name,
8223                                           &dev->adj_list.lower);
8224         }
8225 }
8226
8227 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
8228 {
8229         struct netdev_adjacent *iter;
8230
8231         struct net *net = dev_net(dev);
8232
8233         list_for_each_entry(iter, &dev->adj_list.upper, list) {
8234                 if (!net_eq(net, dev_net(iter->dev)))
8235                         continue;
8236                 netdev_adjacent_sysfs_del(iter->dev, oldname,
8237                                           &iter->dev->adj_list.lower);
8238                 netdev_adjacent_sysfs_add(iter->dev, dev,
8239                                           &iter->dev->adj_list.lower);
8240         }
8241
8242         list_for_each_entry(iter, &dev->adj_list.lower, list) {
8243                 if (!net_eq(net, dev_net(iter->dev)))
8244                         continue;
8245                 netdev_adjacent_sysfs_del(iter->dev, oldname,
8246                                           &iter->dev->adj_list.upper);
8247                 netdev_adjacent_sysfs_add(iter->dev, dev,
8248                                           &iter->dev->adj_list.upper);
8249         }
8250 }
8251
8252 void *netdev_lower_dev_get_private(struct net_device *dev,
8253                                    struct net_device *lower_dev)
8254 {
8255         struct netdev_adjacent *lower;
8256
8257         if (!lower_dev)
8258                 return NULL;
8259         lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
8260         if (!lower)
8261                 return NULL;
8262
8263         return lower->private;
8264 }
8265 EXPORT_SYMBOL(netdev_lower_dev_get_private);
8266
8267
8268 /**
8269  * netdev_lower_state_changed - Dispatch event about lower device state change
8270  * @lower_dev: device
8271  * @lower_state_info: state to dispatch
8272  *
8273  * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
8274  * The caller must hold the RTNL lock.
8275  */
8276 void netdev_lower_state_changed(struct net_device *lower_dev,
8277                                 void *lower_state_info)
8278 {
8279         struct netdev_notifier_changelowerstate_info changelowerstate_info = {
8280                 .info.dev = lower_dev,
8281         };
8282
8283         ASSERT_RTNL();
8284         changelowerstate_info.lower_state_info = lower_state_info;
8285         call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE,
8286                                       &changelowerstate_info.info);
8287 }
8288 EXPORT_SYMBOL(netdev_lower_state_changed);
8289
8290 static void dev_change_rx_flags(struct net_device *dev, int flags)
8291 {
8292         const struct net_device_ops *ops = dev->netdev_ops;
8293
8294         if (ops->ndo_change_rx_flags)
8295                 ops->ndo_change_rx_flags(dev, flags);
8296 }
8297
8298 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
8299 {
8300         unsigned int old_flags = dev->flags;
8301         kuid_t uid;
8302         kgid_t gid;
8303
8304         ASSERT_RTNL();
8305
8306         dev->flags |= IFF_PROMISC;
8307         dev->promiscuity += inc;
8308         if (dev->promiscuity == 0) {
8309                 /*
8310                  * Avoid overflow.
8311                  * If inc causes overflow, untouch promisc and return error.
8312                  */
8313                 if (inc < 0)
8314                         dev->flags &= ~IFF_PROMISC;
8315                 else {
8316                         dev->promiscuity -= inc;
8317                         netdev_warn(dev, "promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n");
8318                         return -EOVERFLOW;
8319                 }
8320         }
8321         if (dev->flags != old_flags) {
8322                 pr_info("device %s %s promiscuous mode\n",
8323                         dev->name,
8324                         dev->flags & IFF_PROMISC ? "entered" : "left");
8325                 if (audit_enabled) {
8326                         current_uid_gid(&uid, &gid);
8327                         audit_log(audit_context(), GFP_ATOMIC,
8328                                   AUDIT_ANOM_PROMISCUOUS,
8329                                   "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
8330                                   dev->name, (dev->flags & IFF_PROMISC),
8331                                   (old_flags & IFF_PROMISC),
8332                                   from_kuid(&init_user_ns, audit_get_loginuid(current)),
8333                                   from_kuid(&init_user_ns, uid),
8334                                   from_kgid(&init_user_ns, gid),
8335                                   audit_get_sessionid(current));
8336                 }
8337
8338                 dev_change_rx_flags(dev, IFF_PROMISC);
8339         }
8340         if (notify)
8341                 __dev_notify_flags(dev, old_flags, IFF_PROMISC, 0, NULL);
8342         return 0;
8343 }
8344
8345 /**
8346  *      dev_set_promiscuity     - update promiscuity count on a device
8347  *      @dev: device
8348  *      @inc: modifier
8349  *
8350  *      Add or remove promiscuity from a device. While the count in the device
8351  *      remains above zero the interface remains promiscuous. Once it hits zero
8352  *      the device reverts back to normal filtering operation. A negative inc
8353  *      value is used to drop promiscuity on the device.
8354  *      Return 0 if successful or a negative errno code on error.
8355  */
8356 int dev_set_promiscuity(struct net_device *dev, int inc)
8357 {
8358         unsigned int old_flags = dev->flags;
8359         int err;
8360
8361         err = __dev_set_promiscuity(dev, inc, true);
8362         if (err < 0)
8363                 return err;
8364         if (dev->flags != old_flags)
8365                 dev_set_rx_mode(dev);
8366         return err;
8367 }
8368 EXPORT_SYMBOL(dev_set_promiscuity);
8369
8370 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
8371 {
8372         unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
8373
8374         ASSERT_RTNL();
8375
8376         dev->flags |= IFF_ALLMULTI;
8377         dev->allmulti += inc;
8378         if (dev->allmulti == 0) {
8379                 /*
8380                  * Avoid overflow.
8381                  * If inc causes overflow, untouch allmulti and return error.
8382                  */
8383                 if (inc < 0)
8384                         dev->flags &= ~IFF_ALLMULTI;
8385                 else {
8386                         dev->allmulti -= inc;
8387                         netdev_warn(dev, "allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n");
8388                         return -EOVERFLOW;
8389                 }
8390         }
8391         if (dev->flags ^ old_flags) {
8392                 dev_change_rx_flags(dev, IFF_ALLMULTI);
8393                 dev_set_rx_mode(dev);
8394                 if (notify)
8395                         __dev_notify_flags(dev, old_flags,
8396                                            dev->gflags ^ old_gflags, 0, NULL);
8397         }
8398         return 0;
8399 }
8400
8401 /**
8402  *      dev_set_allmulti        - update allmulti count on a device
8403  *      @dev: device
8404  *      @inc: modifier
8405  *
8406  *      Add or remove reception of all multicast frames to a device. While the
8407  *      count in the device remains above zero the interface remains listening
8408  *      to all interfaces. Once it hits zero the device reverts back to normal
8409  *      filtering operation. A negative @inc value is used to drop the counter
8410  *      when releasing a resource needing all multicasts.
8411  *      Return 0 if successful or a negative errno code on error.
8412  */
8413
8414 int dev_set_allmulti(struct net_device *dev, int inc)
8415 {
8416         return __dev_set_allmulti(dev, inc, true);
8417 }
8418 EXPORT_SYMBOL(dev_set_allmulti);
8419
8420 /*
8421  *      Upload unicast and multicast address lists to device and
8422  *      configure RX filtering. When the device doesn't support unicast
8423  *      filtering it is put in promiscuous mode while unicast addresses
8424  *      are present.
8425  */
8426 void __dev_set_rx_mode(struct net_device *dev)
8427 {
8428         const struct net_device_ops *ops = dev->netdev_ops;
8429
8430         /* dev_open will call this function so the list will stay sane. */
8431         if (!(dev->flags&IFF_UP))
8432                 return;
8433
8434         if (!netif_device_present(dev))
8435                 return;
8436
8437         if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
8438                 /* Unicast addresses changes may only happen under the rtnl,
8439                  * therefore calling __dev_set_promiscuity here is safe.
8440                  */
8441                 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
8442                         __dev_set_promiscuity(dev, 1, false);
8443                         dev->uc_promisc = true;
8444                 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
8445                         __dev_set_promiscuity(dev, -1, false);
8446                         dev->uc_promisc = false;
8447                 }
8448         }
8449
8450         if (ops->ndo_set_rx_mode)
8451                 ops->ndo_set_rx_mode(dev);
8452 }
8453
8454 void dev_set_rx_mode(struct net_device *dev)
8455 {
8456         netif_addr_lock_bh(dev);
8457         __dev_set_rx_mode(dev);
8458         netif_addr_unlock_bh(dev);
8459 }
8460
8461 /**
8462  *      dev_get_flags - get flags reported to userspace
8463  *      @dev: device
8464  *
8465  *      Get the combination of flag bits exported through APIs to userspace.
8466  */
8467 unsigned int dev_get_flags(const struct net_device *dev)
8468 {
8469         unsigned int flags;
8470
8471         flags = (dev->flags & ~(IFF_PROMISC |
8472                                 IFF_ALLMULTI |
8473                                 IFF_RUNNING |
8474                                 IFF_LOWER_UP |
8475                                 IFF_DORMANT)) |
8476                 (dev->gflags & (IFF_PROMISC |
8477                                 IFF_ALLMULTI));
8478
8479         if (netif_running(dev)) {
8480                 if (netif_oper_up(dev))
8481                         flags |= IFF_RUNNING;
8482                 if (netif_carrier_ok(dev))
8483                         flags |= IFF_LOWER_UP;
8484                 if (netif_dormant(dev))
8485                         flags |= IFF_DORMANT;
8486         }
8487
8488         return flags;
8489 }
8490 EXPORT_SYMBOL(dev_get_flags);
8491
8492 int __dev_change_flags(struct net_device *dev, unsigned int flags,
8493                        struct netlink_ext_ack *extack)
8494 {
8495         unsigned int old_flags = dev->flags;
8496         int ret;
8497
8498         ASSERT_RTNL();
8499
8500         /*
8501          *      Set the flags on our device.
8502          */
8503
8504         dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
8505                                IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
8506                                IFF_AUTOMEDIA)) |
8507                      (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
8508                                     IFF_ALLMULTI));
8509
8510         /*
8511          *      Load in the correct multicast list now the flags have changed.
8512          */
8513
8514         if ((old_flags ^ flags) & IFF_MULTICAST)
8515                 dev_change_rx_flags(dev, IFF_MULTICAST);
8516
8517         dev_set_rx_mode(dev);
8518
8519         /*
8520          *      Have we downed the interface. We handle IFF_UP ourselves
8521          *      according to user attempts to set it, rather than blindly
8522          *      setting it.
8523          */
8524
8525         ret = 0;
8526         if ((old_flags ^ flags) & IFF_UP) {
8527                 if (old_flags & IFF_UP)
8528                         __dev_close(dev);
8529                 else
8530                         ret = __dev_open(dev, extack);
8531         }
8532
8533         if ((flags ^ dev->gflags) & IFF_PROMISC) {
8534                 int inc = (flags & IFF_PROMISC) ? 1 : -1;
8535                 unsigned int old_flags = dev->flags;
8536
8537                 dev->gflags ^= IFF_PROMISC;
8538
8539                 if (__dev_set_promiscuity(dev, inc, false) >= 0)
8540                         if (dev->flags != old_flags)
8541                                 dev_set_rx_mode(dev);
8542         }
8543
8544         /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
8545          * is important. Some (broken) drivers set IFF_PROMISC, when
8546          * IFF_ALLMULTI is requested not asking us and not reporting.
8547          */
8548         if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
8549                 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
8550
8551                 dev->gflags ^= IFF_ALLMULTI;
8552                 __dev_set_allmulti(dev, inc, false);
8553         }
8554
8555         return ret;
8556 }
8557
8558 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
8559                         unsigned int gchanges, u32 portid,
8560                         const struct nlmsghdr *nlh)
8561 {
8562         unsigned int changes = dev->flags ^ old_flags;
8563
8564         if (gchanges)
8565                 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC, portid, nlh);
8566
8567         if (changes & IFF_UP) {
8568                 if (dev->flags & IFF_UP)
8569                         call_netdevice_notifiers(NETDEV_UP, dev);
8570                 else
8571                         call_netdevice_notifiers(NETDEV_DOWN, dev);
8572         }
8573
8574         if (dev->flags & IFF_UP &&
8575             (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
8576                 struct netdev_notifier_change_info change_info = {
8577                         .info = {
8578                                 .dev = dev,
8579                         },
8580                         .flags_changed = changes,
8581                 };
8582
8583                 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info);
8584         }
8585 }
8586
8587 /**
8588  *      dev_change_flags - change device settings
8589  *      @dev: device
8590  *      @flags: device state flags
8591  *      @extack: netlink extended ack
8592  *
8593  *      Change settings on device based state flags. The flags are
8594  *      in the userspace exported format.
8595  */
8596 int dev_change_flags(struct net_device *dev, unsigned int flags,
8597                      struct netlink_ext_ack *extack)
8598 {
8599         int ret;
8600         unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
8601
8602         ret = __dev_change_flags(dev, flags, extack);
8603         if (ret < 0)
8604                 return ret;
8605
8606         changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
8607         __dev_notify_flags(dev, old_flags, changes, 0, NULL);
8608         return ret;
8609 }
8610 EXPORT_SYMBOL(dev_change_flags);
8611
8612 int __dev_set_mtu(struct net_device *dev, int new_mtu)
8613 {
8614         const struct net_device_ops *ops = dev->netdev_ops;
8615
8616         if (ops->ndo_change_mtu)
8617                 return ops->ndo_change_mtu(dev, new_mtu);
8618
8619         /* Pairs with all the lockless reads of dev->mtu in the stack */
8620         WRITE_ONCE(dev->mtu, new_mtu);
8621         return 0;
8622 }
8623 EXPORT_SYMBOL(__dev_set_mtu);
8624
8625 int dev_validate_mtu(struct net_device *dev, int new_mtu,
8626                      struct netlink_ext_ack *extack)
8627 {
8628         /* MTU must be positive, and in range */
8629         if (new_mtu < 0 || new_mtu < dev->min_mtu) {
8630                 NL_SET_ERR_MSG(extack, "mtu less than device minimum");
8631                 return -EINVAL;
8632         }
8633
8634         if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) {
8635                 NL_SET_ERR_MSG(extack, "mtu greater than device maximum");
8636                 return -EINVAL;
8637         }
8638         return 0;
8639 }
8640
8641 /**
8642  *      dev_set_mtu_ext - Change maximum transfer unit
8643  *      @dev: device
8644  *      @new_mtu: new transfer unit
8645  *      @extack: netlink extended ack
8646  *
8647  *      Change the maximum transfer size of the network device.
8648  */
8649 int dev_set_mtu_ext(struct net_device *dev, int new_mtu,
8650                     struct netlink_ext_ack *extack)
8651 {
8652         int err, orig_mtu;
8653
8654         if (new_mtu == dev->mtu)
8655                 return 0;
8656
8657         err = dev_validate_mtu(dev, new_mtu, extack);
8658         if (err)
8659                 return err;
8660
8661         if (!netif_device_present(dev))
8662                 return -ENODEV;
8663
8664         err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
8665         err = notifier_to_errno(err);
8666         if (err)
8667                 return err;
8668
8669         orig_mtu = dev->mtu;
8670         err = __dev_set_mtu(dev, new_mtu);
8671
8672         if (!err) {
8673                 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8674                                                    orig_mtu);
8675                 err = notifier_to_errno(err);
8676                 if (err) {
8677                         /* setting mtu back and notifying everyone again,
8678                          * so that they have a chance to revert changes.
8679                          */
8680                         __dev_set_mtu(dev, orig_mtu);
8681                         call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8682                                                      new_mtu);
8683                 }
8684         }
8685         return err;
8686 }
8687
8688 int dev_set_mtu(struct net_device *dev, int new_mtu)
8689 {
8690         struct netlink_ext_ack extack;
8691         int err;
8692
8693         memset(&extack, 0, sizeof(extack));
8694         err = dev_set_mtu_ext(dev, new_mtu, &extack);
8695         if (err && extack._msg)
8696                 net_err_ratelimited("%s: %s\n", dev->name, extack._msg);
8697         return err;
8698 }
8699 EXPORT_SYMBOL(dev_set_mtu);
8700
8701 /**
8702  *      dev_change_tx_queue_len - Change TX queue length of a netdevice
8703  *      @dev: device
8704  *      @new_len: new tx queue length
8705  */
8706 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len)
8707 {
8708         unsigned int orig_len = dev->tx_queue_len;
8709         int res;
8710
8711         if (new_len != (unsigned int)new_len)
8712                 return -ERANGE;
8713
8714         if (new_len != orig_len) {
8715                 dev->tx_queue_len = new_len;
8716                 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev);
8717                 res = notifier_to_errno(res);
8718                 if (res)
8719                         goto err_rollback;
8720                 res = dev_qdisc_change_tx_queue_len(dev);
8721                 if (res)
8722                         goto err_rollback;
8723         }
8724
8725         return 0;
8726
8727 err_rollback:
8728         netdev_err(dev, "refused to change device tx_queue_len\n");
8729         dev->tx_queue_len = orig_len;
8730         return res;
8731 }
8732
8733 /**
8734  *      dev_set_group - Change group this device belongs to
8735  *      @dev: device
8736  *      @new_group: group this device should belong to
8737  */
8738 void dev_set_group(struct net_device *dev, int new_group)
8739 {
8740         dev->group = new_group;
8741 }
8742
8743 /**
8744  *      dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR.
8745  *      @dev: device
8746  *      @addr: new address
8747  *      @extack: netlink extended ack
8748  */
8749 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr,
8750                               struct netlink_ext_ack *extack)
8751 {
8752         struct netdev_notifier_pre_changeaddr_info info = {
8753                 .info.dev = dev,
8754                 .info.extack = extack,
8755                 .dev_addr = addr,
8756         };
8757         int rc;
8758
8759         rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info);
8760         return notifier_to_errno(rc);
8761 }
8762 EXPORT_SYMBOL(dev_pre_changeaddr_notify);
8763
8764 /**
8765  *      dev_set_mac_address - Change Media Access Control Address
8766  *      @dev: device
8767  *      @sa: new address
8768  *      @extack: netlink extended ack
8769  *
8770  *      Change the hardware (MAC) address of the device
8771  */
8772 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa,
8773                         struct netlink_ext_ack *extack)
8774 {
8775         const struct net_device_ops *ops = dev->netdev_ops;
8776         int err;
8777
8778         if (!ops->ndo_set_mac_address)
8779                 return -EOPNOTSUPP;
8780         if (sa->sa_family != dev->type)
8781                 return -EINVAL;
8782         if (!netif_device_present(dev))
8783                 return -ENODEV;
8784         err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack);
8785         if (err)
8786                 return err;
8787         err = ops->ndo_set_mac_address(dev, sa);
8788         if (err)
8789                 return err;
8790         dev->addr_assign_type = NET_ADDR_SET;
8791         call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
8792         add_device_randomness(dev->dev_addr, dev->addr_len);
8793         return 0;
8794 }
8795 EXPORT_SYMBOL(dev_set_mac_address);
8796
8797 static DECLARE_RWSEM(dev_addr_sem);
8798
8799 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa,
8800                              struct netlink_ext_ack *extack)
8801 {
8802         int ret;
8803
8804         down_write(&dev_addr_sem);
8805         ret = dev_set_mac_address(dev, sa, extack);
8806         up_write(&dev_addr_sem);
8807         return ret;
8808 }
8809 EXPORT_SYMBOL(dev_set_mac_address_user);
8810
8811 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name)
8812 {
8813         size_t size = sizeof(sa->sa_data_min);
8814         struct net_device *dev;
8815         int ret = 0;
8816
8817         down_read(&dev_addr_sem);
8818         rcu_read_lock();
8819
8820         dev = dev_get_by_name_rcu(net, dev_name);
8821         if (!dev) {
8822                 ret = -ENODEV;
8823                 goto unlock;
8824         }
8825         if (!dev->addr_len)
8826                 memset(sa->sa_data, 0, size);
8827         else
8828                 memcpy(sa->sa_data, dev->dev_addr,
8829                        min_t(size_t, size, dev->addr_len));
8830         sa->sa_family = dev->type;
8831
8832 unlock:
8833         rcu_read_unlock();
8834         up_read(&dev_addr_sem);
8835         return ret;
8836 }
8837 EXPORT_SYMBOL(dev_get_mac_address);
8838
8839 /**
8840  *      dev_change_carrier - Change device carrier
8841  *      @dev: device
8842  *      @new_carrier: new value
8843  *
8844  *      Change device carrier
8845  */
8846 int dev_change_carrier(struct net_device *dev, bool new_carrier)
8847 {
8848         const struct net_device_ops *ops = dev->netdev_ops;
8849
8850         if (!ops->ndo_change_carrier)
8851                 return -EOPNOTSUPP;
8852         if (!netif_device_present(dev))
8853                 return -ENODEV;
8854         return ops->ndo_change_carrier(dev, new_carrier);
8855 }
8856
8857 /**
8858  *      dev_get_phys_port_id - Get device physical port ID
8859  *      @dev: device
8860  *      @ppid: port ID
8861  *
8862  *      Get device physical port ID
8863  */
8864 int dev_get_phys_port_id(struct net_device *dev,
8865                          struct netdev_phys_item_id *ppid)
8866 {
8867         const struct net_device_ops *ops = dev->netdev_ops;
8868
8869         if (!ops->ndo_get_phys_port_id)
8870                 return -EOPNOTSUPP;
8871         return ops->ndo_get_phys_port_id(dev, ppid);
8872 }
8873
8874 /**
8875  *      dev_get_phys_port_name - Get device physical port name
8876  *      @dev: device
8877  *      @name: port name
8878  *      @len: limit of bytes to copy to name
8879  *
8880  *      Get device physical port name
8881  */
8882 int dev_get_phys_port_name(struct net_device *dev,
8883                            char *name, size_t len)
8884 {
8885         const struct net_device_ops *ops = dev->netdev_ops;
8886         int err;
8887
8888         if (ops->ndo_get_phys_port_name) {
8889                 err = ops->ndo_get_phys_port_name(dev, name, len);
8890                 if (err != -EOPNOTSUPP)
8891                         return err;
8892         }
8893         return devlink_compat_phys_port_name_get(dev, name, len);
8894 }
8895
8896 /**
8897  *      dev_get_port_parent_id - Get the device's port parent identifier
8898  *      @dev: network device
8899  *      @ppid: pointer to a storage for the port's parent identifier
8900  *      @recurse: allow/disallow recursion to lower devices
8901  *
8902  *      Get the devices's port parent identifier
8903  */
8904 int dev_get_port_parent_id(struct net_device *dev,
8905                            struct netdev_phys_item_id *ppid,
8906                            bool recurse)
8907 {
8908         const struct net_device_ops *ops = dev->netdev_ops;
8909         struct netdev_phys_item_id first = { };
8910         struct net_device *lower_dev;
8911         struct list_head *iter;
8912         int err;
8913
8914         if (ops->ndo_get_port_parent_id) {
8915                 err = ops->ndo_get_port_parent_id(dev, ppid);
8916                 if (err != -EOPNOTSUPP)
8917                         return err;
8918         }
8919
8920         err = devlink_compat_switch_id_get(dev, ppid);
8921         if (!recurse || err != -EOPNOTSUPP)
8922                 return err;
8923
8924         netdev_for_each_lower_dev(dev, lower_dev, iter) {
8925                 err = dev_get_port_parent_id(lower_dev, ppid, true);
8926                 if (err)
8927                         break;
8928                 if (!first.id_len)
8929                         first = *ppid;
8930                 else if (memcmp(&first, ppid, sizeof(*ppid)))
8931                         return -EOPNOTSUPP;
8932         }
8933
8934         return err;
8935 }
8936 EXPORT_SYMBOL(dev_get_port_parent_id);
8937
8938 /**
8939  *      netdev_port_same_parent_id - Indicate if two network devices have
8940  *      the same port parent identifier
8941  *      @a: first network device
8942  *      @b: second network device
8943  */
8944 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b)
8945 {
8946         struct netdev_phys_item_id a_id = { };
8947         struct netdev_phys_item_id b_id = { };
8948
8949         if (dev_get_port_parent_id(a, &a_id, true) ||
8950             dev_get_port_parent_id(b, &b_id, true))
8951                 return false;
8952
8953         return netdev_phys_item_id_same(&a_id, &b_id);
8954 }
8955 EXPORT_SYMBOL(netdev_port_same_parent_id);
8956
8957 /**
8958  *      dev_change_proto_down - set carrier according to proto_down.
8959  *
8960  *      @dev: device
8961  *      @proto_down: new value
8962  */
8963 int dev_change_proto_down(struct net_device *dev, bool proto_down)
8964 {
8965         if (!(dev->priv_flags & IFF_CHANGE_PROTO_DOWN))
8966                 return -EOPNOTSUPP;
8967         if (!netif_device_present(dev))
8968                 return -ENODEV;
8969         if (proto_down)
8970                 netif_carrier_off(dev);
8971         else
8972                 netif_carrier_on(dev);
8973         dev->proto_down = proto_down;
8974         return 0;
8975 }
8976
8977 /**
8978  *      dev_change_proto_down_reason - proto down reason
8979  *
8980  *      @dev: device
8981  *      @mask: proto down mask
8982  *      @value: proto down value
8983  */
8984 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask,
8985                                   u32 value)
8986 {
8987         int b;
8988
8989         if (!mask) {
8990                 dev->proto_down_reason = value;
8991         } else {
8992                 for_each_set_bit(b, &mask, 32) {
8993                         if (value & (1 << b))
8994                                 dev->proto_down_reason |= BIT(b);
8995                         else
8996                                 dev->proto_down_reason &= ~BIT(b);
8997                 }
8998         }
8999 }
9000
9001 struct bpf_xdp_link {
9002         struct bpf_link link;
9003         struct net_device *dev; /* protected by rtnl_lock, no refcnt held */
9004         int flags;
9005 };
9006
9007 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags)
9008 {
9009         if (flags & XDP_FLAGS_HW_MODE)
9010                 return XDP_MODE_HW;
9011         if (flags & XDP_FLAGS_DRV_MODE)
9012                 return XDP_MODE_DRV;
9013         if (flags & XDP_FLAGS_SKB_MODE)
9014                 return XDP_MODE_SKB;
9015         return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB;
9016 }
9017
9018 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode)
9019 {
9020         switch (mode) {
9021         case XDP_MODE_SKB:
9022                 return generic_xdp_install;
9023         case XDP_MODE_DRV:
9024         case XDP_MODE_HW:
9025                 return dev->netdev_ops->ndo_bpf;
9026         default:
9027                 return NULL;
9028         }
9029 }
9030
9031 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev,
9032                                          enum bpf_xdp_mode mode)
9033 {
9034         return dev->xdp_state[mode].link;
9035 }
9036
9037 static struct bpf_prog *dev_xdp_prog(struct net_device *dev,
9038                                      enum bpf_xdp_mode mode)
9039 {
9040         struct bpf_xdp_link *link = dev_xdp_link(dev, mode);
9041
9042         if (link)
9043                 return link->link.prog;
9044         return dev->xdp_state[mode].prog;
9045 }
9046
9047 u8 dev_xdp_prog_count(struct net_device *dev)
9048 {
9049         u8 count = 0;
9050         int i;
9051
9052         for (i = 0; i < __MAX_XDP_MODE; i++)
9053                 if (dev->xdp_state[i].prog || dev->xdp_state[i].link)
9054                         count++;
9055         return count;
9056 }
9057 EXPORT_SYMBOL_GPL(dev_xdp_prog_count);
9058
9059 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode)
9060 {
9061         struct bpf_prog *prog = dev_xdp_prog(dev, mode);
9062
9063         return prog ? prog->aux->id : 0;
9064 }
9065
9066 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode,
9067                              struct bpf_xdp_link *link)
9068 {
9069         dev->xdp_state[mode].link = link;
9070         dev->xdp_state[mode].prog = NULL;
9071 }
9072
9073 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode,
9074                              struct bpf_prog *prog)
9075 {
9076         dev->xdp_state[mode].link = NULL;
9077         dev->xdp_state[mode].prog = prog;
9078 }
9079
9080 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode,
9081                            bpf_op_t bpf_op, struct netlink_ext_ack *extack,
9082                            u32 flags, struct bpf_prog *prog)
9083 {
9084         struct netdev_bpf xdp;
9085         int err;
9086
9087         memset(&xdp, 0, sizeof(xdp));
9088         xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG;
9089         xdp.extack = extack;
9090         xdp.flags = flags;
9091         xdp.prog = prog;
9092
9093         /* Drivers assume refcnt is already incremented (i.e, prog pointer is
9094          * "moved" into driver), so they don't increment it on their own, but
9095          * they do decrement refcnt when program is detached or replaced.
9096          * Given net_device also owns link/prog, we need to bump refcnt here
9097          * to prevent drivers from underflowing it.
9098          */
9099         if (prog)
9100                 bpf_prog_inc(prog);
9101         err = bpf_op(dev, &xdp);
9102         if (err) {
9103                 if (prog)
9104                         bpf_prog_put(prog);
9105                 return err;
9106         }
9107
9108         if (mode != XDP_MODE_HW)
9109                 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog);
9110
9111         return 0;
9112 }
9113
9114 static void dev_xdp_uninstall(struct net_device *dev)
9115 {
9116         struct bpf_xdp_link *link;
9117         struct bpf_prog *prog;
9118         enum bpf_xdp_mode mode;
9119         bpf_op_t bpf_op;
9120
9121         ASSERT_RTNL();
9122
9123         for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) {
9124                 prog = dev_xdp_prog(dev, mode);
9125                 if (!prog)
9126                         continue;
9127
9128                 bpf_op = dev_xdp_bpf_op(dev, mode);
9129                 if (!bpf_op)
9130                         continue;
9131
9132                 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
9133
9134                 /* auto-detach link from net device */
9135                 link = dev_xdp_link(dev, mode);
9136                 if (link)
9137                         link->dev = NULL;
9138                 else
9139                         bpf_prog_put(prog);
9140
9141                 dev_xdp_set_link(dev, mode, NULL);
9142         }
9143 }
9144
9145 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack,
9146                           struct bpf_xdp_link *link, struct bpf_prog *new_prog,
9147                           struct bpf_prog *old_prog, u32 flags)
9148 {
9149         unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES);
9150         struct bpf_prog *cur_prog;
9151         struct net_device *upper;
9152         struct list_head *iter;
9153         enum bpf_xdp_mode mode;
9154         bpf_op_t bpf_op;
9155         int err;
9156
9157         ASSERT_RTNL();
9158
9159         /* either link or prog attachment, never both */
9160         if (link && (new_prog || old_prog))
9161                 return -EINVAL;
9162         /* link supports only XDP mode flags */
9163         if (link && (flags & ~XDP_FLAGS_MODES)) {
9164                 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment");
9165                 return -EINVAL;
9166         }
9167         /* just one XDP mode bit should be set, zero defaults to drv/skb mode */
9168         if (num_modes > 1) {
9169                 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set");
9170                 return -EINVAL;
9171         }
9172         /* avoid ambiguity if offload + drv/skb mode progs are both loaded */
9173         if (!num_modes && dev_xdp_prog_count(dev) > 1) {
9174                 NL_SET_ERR_MSG(extack,
9175                                "More than one program loaded, unset mode is ambiguous");
9176                 return -EINVAL;
9177         }
9178         /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */
9179         if (old_prog && !(flags & XDP_FLAGS_REPLACE)) {
9180                 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified");
9181                 return -EINVAL;
9182         }
9183
9184         mode = dev_xdp_mode(dev, flags);
9185         /* can't replace attached link */
9186         if (dev_xdp_link(dev, mode)) {
9187                 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link");
9188                 return -EBUSY;
9189         }
9190
9191         /* don't allow if an upper device already has a program */
9192         netdev_for_each_upper_dev_rcu(dev, upper, iter) {
9193                 if (dev_xdp_prog_count(upper) > 0) {
9194                         NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program");
9195                         return -EEXIST;
9196                 }
9197         }
9198
9199         cur_prog = dev_xdp_prog(dev, mode);
9200         /* can't replace attached prog with link */
9201         if (link && cur_prog) {
9202                 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link");
9203                 return -EBUSY;
9204         }
9205         if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) {
9206                 NL_SET_ERR_MSG(extack, "Active program does not match expected");
9207                 return -EEXIST;
9208         }
9209
9210         /* put effective new program into new_prog */
9211         if (link)
9212                 new_prog = link->link.prog;
9213
9214         if (new_prog) {
9215                 bool offload = mode == XDP_MODE_HW;
9216                 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB
9217                                                ? XDP_MODE_DRV : XDP_MODE_SKB;
9218
9219                 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) {
9220                         NL_SET_ERR_MSG(extack, "XDP program already attached");
9221                         return -EBUSY;
9222                 }
9223                 if (!offload && dev_xdp_prog(dev, other_mode)) {
9224                         NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time");
9225                         return -EEXIST;
9226                 }
9227                 if (!offload && bpf_prog_is_dev_bound(new_prog->aux)) {
9228                         NL_SET_ERR_MSG(extack, "Using device-bound program without HW_MODE flag is not supported");
9229                         return -EINVAL;
9230                 }
9231                 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) {
9232                         NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device");
9233                         return -EINVAL;
9234                 }
9235                 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) {
9236                         NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device");
9237                         return -EINVAL;
9238                 }
9239         }
9240
9241         /* don't call drivers if the effective program didn't change */
9242         if (new_prog != cur_prog) {
9243                 bpf_op = dev_xdp_bpf_op(dev, mode);
9244                 if (!bpf_op) {
9245                         NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode");
9246                         return -EOPNOTSUPP;
9247                 }
9248
9249                 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog);
9250                 if (err)
9251                         return err;
9252         }
9253
9254         if (link)
9255                 dev_xdp_set_link(dev, mode, link);
9256         else
9257                 dev_xdp_set_prog(dev, mode, new_prog);
9258         if (cur_prog)
9259                 bpf_prog_put(cur_prog);
9260
9261         return 0;
9262 }
9263
9264 static int dev_xdp_attach_link(struct net_device *dev,
9265                                struct netlink_ext_ack *extack,
9266                                struct bpf_xdp_link *link)
9267 {
9268         return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags);
9269 }
9270
9271 static int dev_xdp_detach_link(struct net_device *dev,
9272                                struct netlink_ext_ack *extack,
9273                                struct bpf_xdp_link *link)
9274 {
9275         enum bpf_xdp_mode mode;
9276         bpf_op_t bpf_op;
9277
9278         ASSERT_RTNL();
9279
9280         mode = dev_xdp_mode(dev, link->flags);
9281         if (dev_xdp_link(dev, mode) != link)
9282                 return -EINVAL;
9283
9284         bpf_op = dev_xdp_bpf_op(dev, mode);
9285         WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
9286         dev_xdp_set_link(dev, mode, NULL);
9287         return 0;
9288 }
9289
9290 static void bpf_xdp_link_release(struct bpf_link *link)
9291 {
9292         struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9293
9294         rtnl_lock();
9295
9296         /* if racing with net_device's tear down, xdp_link->dev might be
9297          * already NULL, in which case link was already auto-detached
9298          */
9299         if (xdp_link->dev) {
9300                 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link));
9301                 xdp_link->dev = NULL;
9302         }
9303
9304         rtnl_unlock();
9305 }
9306
9307 static int bpf_xdp_link_detach(struct bpf_link *link)
9308 {
9309         bpf_xdp_link_release(link);
9310         return 0;
9311 }
9312
9313 static void bpf_xdp_link_dealloc(struct bpf_link *link)
9314 {
9315         struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9316
9317         kfree(xdp_link);
9318 }
9319
9320 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link,
9321                                      struct seq_file *seq)
9322 {
9323         struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9324         u32 ifindex = 0;
9325
9326         rtnl_lock();
9327         if (xdp_link->dev)
9328                 ifindex = xdp_link->dev->ifindex;
9329         rtnl_unlock();
9330
9331         seq_printf(seq, "ifindex:\t%u\n", ifindex);
9332 }
9333
9334 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link,
9335                                        struct bpf_link_info *info)
9336 {
9337         struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9338         u32 ifindex = 0;
9339
9340         rtnl_lock();
9341         if (xdp_link->dev)
9342                 ifindex = xdp_link->dev->ifindex;
9343         rtnl_unlock();
9344
9345         info->xdp.ifindex = ifindex;
9346         return 0;
9347 }
9348
9349 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog,
9350                                struct bpf_prog *old_prog)
9351 {
9352         struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9353         enum bpf_xdp_mode mode;
9354         bpf_op_t bpf_op;
9355         int err = 0;
9356
9357         rtnl_lock();
9358
9359         /* link might have been auto-released already, so fail */
9360         if (!xdp_link->dev) {
9361                 err = -ENOLINK;
9362                 goto out_unlock;
9363         }
9364
9365         if (old_prog && link->prog != old_prog) {
9366                 err = -EPERM;
9367                 goto out_unlock;
9368         }
9369         old_prog = link->prog;
9370         if (old_prog->type != new_prog->type ||
9371             old_prog->expected_attach_type != new_prog->expected_attach_type) {
9372                 err = -EINVAL;
9373                 goto out_unlock;
9374         }
9375
9376         if (old_prog == new_prog) {
9377                 /* no-op, don't disturb drivers */
9378                 bpf_prog_put(new_prog);
9379                 goto out_unlock;
9380         }
9381
9382         mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags);
9383         bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode);
9384         err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL,
9385                               xdp_link->flags, new_prog);
9386         if (err)
9387                 goto out_unlock;
9388
9389         old_prog = xchg(&link->prog, new_prog);
9390         bpf_prog_put(old_prog);
9391
9392 out_unlock:
9393         rtnl_unlock();
9394         return err;
9395 }
9396
9397 static const struct bpf_link_ops bpf_xdp_link_lops = {
9398         .release = bpf_xdp_link_release,
9399         .dealloc = bpf_xdp_link_dealloc,
9400         .detach = bpf_xdp_link_detach,
9401         .show_fdinfo = bpf_xdp_link_show_fdinfo,
9402         .fill_link_info = bpf_xdp_link_fill_link_info,
9403         .update_prog = bpf_xdp_link_update,
9404 };
9405
9406 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
9407 {
9408         struct net *net = current->nsproxy->net_ns;
9409         struct bpf_link_primer link_primer;
9410         struct bpf_xdp_link *link;
9411         struct net_device *dev;
9412         int err, fd;
9413
9414         rtnl_lock();
9415         dev = dev_get_by_index(net, attr->link_create.target_ifindex);
9416         if (!dev) {
9417                 rtnl_unlock();
9418                 return -EINVAL;
9419         }
9420
9421         link = kzalloc(sizeof(*link), GFP_USER);
9422         if (!link) {
9423                 err = -ENOMEM;
9424                 goto unlock;
9425         }
9426
9427         bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog);
9428         link->dev = dev;
9429         link->flags = attr->link_create.flags;
9430
9431         err = bpf_link_prime(&link->link, &link_primer);
9432         if (err) {
9433                 kfree(link);
9434                 goto unlock;
9435         }
9436
9437         err = dev_xdp_attach_link(dev, NULL, link);
9438         rtnl_unlock();
9439
9440         if (err) {
9441                 link->dev = NULL;
9442                 bpf_link_cleanup(&link_primer);
9443                 goto out_put_dev;
9444         }
9445
9446         fd = bpf_link_settle(&link_primer);
9447         /* link itself doesn't hold dev's refcnt to not complicate shutdown */
9448         dev_put(dev);
9449         return fd;
9450
9451 unlock:
9452         rtnl_unlock();
9453
9454 out_put_dev:
9455         dev_put(dev);
9456         return err;
9457 }
9458
9459 /**
9460  *      dev_change_xdp_fd - set or clear a bpf program for a device rx path
9461  *      @dev: device
9462  *      @extack: netlink extended ack
9463  *      @fd: new program fd or negative value to clear
9464  *      @expected_fd: old program fd that userspace expects to replace or clear
9465  *      @flags: xdp-related flags
9466  *
9467  *      Set or clear a bpf program for a device
9468  */
9469 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack,
9470                       int fd, int expected_fd, u32 flags)
9471 {
9472         enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags);
9473         struct bpf_prog *new_prog = NULL, *old_prog = NULL;
9474         int err;
9475
9476         ASSERT_RTNL();
9477
9478         if (fd >= 0) {
9479                 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP,
9480                                                  mode != XDP_MODE_SKB);
9481                 if (IS_ERR(new_prog))
9482                         return PTR_ERR(new_prog);
9483         }
9484
9485         if (expected_fd >= 0) {
9486                 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP,
9487                                                  mode != XDP_MODE_SKB);
9488                 if (IS_ERR(old_prog)) {
9489                         err = PTR_ERR(old_prog);
9490                         old_prog = NULL;
9491                         goto err_out;
9492                 }
9493         }
9494
9495         err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags);
9496
9497 err_out:
9498         if (err && new_prog)
9499                 bpf_prog_put(new_prog);
9500         if (old_prog)
9501                 bpf_prog_put(old_prog);
9502         return err;
9503 }
9504
9505 /**
9506  *      dev_new_index   -       allocate an ifindex
9507  *      @net: the applicable net namespace
9508  *
9509  *      Returns a suitable unique value for a new device interface
9510  *      number.  The caller must hold the rtnl semaphore or the
9511  *      dev_base_lock to be sure it remains unique.
9512  */
9513 static int dev_new_index(struct net *net)
9514 {
9515         int ifindex = net->ifindex;
9516
9517         for (;;) {
9518                 if (++ifindex <= 0)
9519                         ifindex = 1;
9520                 if (!__dev_get_by_index(net, ifindex))
9521                         return net->ifindex = ifindex;
9522         }
9523 }
9524
9525 /* Delayed registration/unregisteration */
9526 LIST_HEAD(net_todo_list);
9527 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
9528
9529 static void net_set_todo(struct net_device *dev)
9530 {
9531         list_add_tail(&dev->todo_list, &net_todo_list);
9532         atomic_inc(&dev_net(dev)->dev_unreg_count);
9533 }
9534
9535 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
9536         struct net_device *upper, netdev_features_t features)
9537 {
9538         netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9539         netdev_features_t feature;
9540         int feature_bit;
9541
9542         for_each_netdev_feature(upper_disables, feature_bit) {
9543                 feature = __NETIF_F_BIT(feature_bit);
9544                 if (!(upper->wanted_features & feature)
9545                     && (features & feature)) {
9546                         netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
9547                                    &feature, upper->name);
9548                         features &= ~feature;
9549                 }
9550         }
9551
9552         return features;
9553 }
9554
9555 static void netdev_sync_lower_features(struct net_device *upper,
9556         struct net_device *lower, netdev_features_t features)
9557 {
9558         netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9559         netdev_features_t feature;
9560         int feature_bit;
9561
9562         for_each_netdev_feature(upper_disables, feature_bit) {
9563                 feature = __NETIF_F_BIT(feature_bit);
9564                 if (!(features & feature) && (lower->features & feature)) {
9565                         netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
9566                                    &feature, lower->name);
9567                         lower->wanted_features &= ~feature;
9568                         __netdev_update_features(lower);
9569
9570                         if (unlikely(lower->features & feature))
9571                                 netdev_WARN(upper, "failed to disable %pNF on %s!\n",
9572                                             &feature, lower->name);
9573                         else
9574                                 netdev_features_change(lower);
9575                 }
9576         }
9577 }
9578
9579 static netdev_features_t netdev_fix_features(struct net_device *dev,
9580         netdev_features_t features)
9581 {
9582         /* Fix illegal checksum combinations */
9583         if ((features & NETIF_F_HW_CSUM) &&
9584             (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
9585                 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
9586                 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
9587         }
9588
9589         /* TSO requires that SG is present as well. */
9590         if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
9591                 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
9592                 features &= ~NETIF_F_ALL_TSO;
9593         }
9594
9595         if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
9596                                         !(features & NETIF_F_IP_CSUM)) {
9597                 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
9598                 features &= ~NETIF_F_TSO;
9599                 features &= ~NETIF_F_TSO_ECN;
9600         }
9601
9602         if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
9603                                          !(features & NETIF_F_IPV6_CSUM)) {
9604                 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
9605                 features &= ~NETIF_F_TSO6;
9606         }
9607
9608         /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */
9609         if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO))
9610                 features &= ~NETIF_F_TSO_MANGLEID;
9611
9612         /* TSO ECN requires that TSO is present as well. */
9613         if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
9614                 features &= ~NETIF_F_TSO_ECN;
9615
9616         /* Software GSO depends on SG. */
9617         if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
9618                 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
9619                 features &= ~NETIF_F_GSO;
9620         }
9621
9622         /* GSO partial features require GSO partial be set */
9623         if ((features & dev->gso_partial_features) &&
9624             !(features & NETIF_F_GSO_PARTIAL)) {
9625                 netdev_dbg(dev,
9626                            "Dropping partially supported GSO features since no GSO partial.\n");
9627                 features &= ~dev->gso_partial_features;
9628         }
9629
9630         if (!(features & NETIF_F_RXCSUM)) {
9631                 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet
9632                  * successfully merged by hardware must also have the
9633                  * checksum verified by hardware.  If the user does not
9634                  * want to enable RXCSUM, logically, we should disable GRO_HW.
9635                  */
9636                 if (features & NETIF_F_GRO_HW) {
9637                         netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n");
9638                         features &= ~NETIF_F_GRO_HW;
9639                 }
9640         }
9641
9642         /* LRO/HW-GRO features cannot be combined with RX-FCS */
9643         if (features & NETIF_F_RXFCS) {
9644                 if (features & NETIF_F_LRO) {
9645                         netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n");
9646                         features &= ~NETIF_F_LRO;
9647                 }
9648
9649                 if (features & NETIF_F_GRO_HW) {
9650                         netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n");
9651                         features &= ~NETIF_F_GRO_HW;
9652                 }
9653         }
9654
9655         if ((features & NETIF_F_GRO_HW) && (features & NETIF_F_LRO)) {
9656                 netdev_dbg(dev, "Dropping LRO feature since HW-GRO is requested.\n");
9657                 features &= ~NETIF_F_LRO;
9658         }
9659
9660         if (features & NETIF_F_HW_TLS_TX) {
9661                 bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) ==
9662                         (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
9663                 bool hw_csum = features & NETIF_F_HW_CSUM;
9664
9665                 if (!ip_csum && !hw_csum) {
9666                         netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n");
9667                         features &= ~NETIF_F_HW_TLS_TX;
9668                 }
9669         }
9670
9671         if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) {
9672                 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n");
9673                 features &= ~NETIF_F_HW_TLS_RX;
9674         }
9675
9676         return features;
9677 }
9678
9679 int __netdev_update_features(struct net_device *dev)
9680 {
9681         struct net_device *upper, *lower;
9682         netdev_features_t features;
9683         struct list_head *iter;
9684         int err = -1;
9685
9686         ASSERT_RTNL();
9687
9688         features = netdev_get_wanted_features(dev);
9689
9690         if (dev->netdev_ops->ndo_fix_features)
9691                 features = dev->netdev_ops->ndo_fix_features(dev, features);
9692
9693         /* driver might be less strict about feature dependencies */
9694         features = netdev_fix_features(dev, features);
9695
9696         /* some features can't be enabled if they're off on an upper device */
9697         netdev_for_each_upper_dev_rcu(dev, upper, iter)
9698                 features = netdev_sync_upper_features(dev, upper, features);
9699
9700         if (dev->features == features)
9701                 goto sync_lower;
9702
9703         netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
9704                 &dev->features, &features);
9705
9706         if (dev->netdev_ops->ndo_set_features)
9707                 err = dev->netdev_ops->ndo_set_features(dev, features);
9708         else
9709                 err = 0;
9710
9711         if (unlikely(err < 0)) {
9712                 netdev_err(dev,
9713                         "set_features() failed (%d); wanted %pNF, left %pNF\n",
9714                         err, &features, &dev->features);
9715                 /* return non-0 since some features might have changed and
9716                  * it's better to fire a spurious notification than miss it
9717                  */
9718                 return -1;
9719         }
9720
9721 sync_lower:
9722         /* some features must be disabled on lower devices when disabled
9723          * on an upper device (think: bonding master or bridge)
9724          */
9725         netdev_for_each_lower_dev(dev, lower, iter)
9726                 netdev_sync_lower_features(dev, lower, features);
9727
9728         if (!err) {
9729                 netdev_features_t diff = features ^ dev->features;
9730
9731                 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) {
9732                         /* udp_tunnel_{get,drop}_rx_info both need
9733                          * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the
9734                          * device, or they won't do anything.
9735                          * Thus we need to update dev->features
9736                          * *before* calling udp_tunnel_get_rx_info,
9737                          * but *after* calling udp_tunnel_drop_rx_info.
9738                          */
9739                         if (features & NETIF_F_RX_UDP_TUNNEL_PORT) {
9740                                 dev->features = features;
9741                                 udp_tunnel_get_rx_info(dev);
9742                         } else {
9743                                 udp_tunnel_drop_rx_info(dev);
9744                         }
9745                 }
9746
9747                 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) {
9748                         if (features & NETIF_F_HW_VLAN_CTAG_FILTER) {
9749                                 dev->features = features;
9750                                 err |= vlan_get_rx_ctag_filter_info(dev);
9751                         } else {
9752                                 vlan_drop_rx_ctag_filter_info(dev);
9753                         }
9754                 }
9755
9756                 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) {
9757                         if (features & NETIF_F_HW_VLAN_STAG_FILTER) {
9758                                 dev->features = features;
9759                                 err |= vlan_get_rx_stag_filter_info(dev);
9760                         } else {
9761                                 vlan_drop_rx_stag_filter_info(dev);
9762                         }
9763                 }
9764
9765                 dev->features = features;
9766         }
9767
9768         return err < 0 ? 0 : 1;
9769 }
9770
9771 /**
9772  *      netdev_update_features - recalculate device features
9773  *      @dev: the device to check
9774  *
9775  *      Recalculate dev->features set and send notifications if it
9776  *      has changed. Should be called after driver or hardware dependent
9777  *      conditions might have changed that influence the features.
9778  */
9779 void netdev_update_features(struct net_device *dev)
9780 {
9781         if (__netdev_update_features(dev))
9782                 netdev_features_change(dev);
9783 }
9784 EXPORT_SYMBOL(netdev_update_features);
9785
9786 /**
9787  *      netdev_change_features - recalculate device features
9788  *      @dev: the device to check
9789  *
9790  *      Recalculate dev->features set and send notifications even
9791  *      if they have not changed. Should be called instead of
9792  *      netdev_update_features() if also dev->vlan_features might
9793  *      have changed to allow the changes to be propagated to stacked
9794  *      VLAN devices.
9795  */
9796 void netdev_change_features(struct net_device *dev)
9797 {
9798         __netdev_update_features(dev);
9799         netdev_features_change(dev);
9800 }
9801 EXPORT_SYMBOL(netdev_change_features);
9802
9803 /**
9804  *      netif_stacked_transfer_operstate -      transfer operstate
9805  *      @rootdev: the root or lower level device to transfer state from
9806  *      @dev: the device to transfer operstate to
9807  *
9808  *      Transfer operational state from root to device. This is normally
9809  *      called when a stacking relationship exists between the root
9810  *      device and the device(a leaf device).
9811  */
9812 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
9813                                         struct net_device *dev)
9814 {
9815         if (rootdev->operstate == IF_OPER_DORMANT)
9816                 netif_dormant_on(dev);
9817         else
9818                 netif_dormant_off(dev);
9819
9820         if (rootdev->operstate == IF_OPER_TESTING)
9821                 netif_testing_on(dev);
9822         else
9823                 netif_testing_off(dev);
9824
9825         if (netif_carrier_ok(rootdev))
9826                 netif_carrier_on(dev);
9827         else
9828                 netif_carrier_off(dev);
9829 }
9830 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
9831
9832 static int netif_alloc_rx_queues(struct net_device *dev)
9833 {
9834         unsigned int i, count = dev->num_rx_queues;
9835         struct netdev_rx_queue *rx;
9836         size_t sz = count * sizeof(*rx);
9837         int err = 0;
9838
9839         BUG_ON(count < 1);
9840
9841         rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
9842         if (!rx)
9843                 return -ENOMEM;
9844
9845         dev->_rx = rx;
9846
9847         for (i = 0; i < count; i++) {
9848                 rx[i].dev = dev;
9849
9850                 /* XDP RX-queue setup */
9851                 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0);
9852                 if (err < 0)
9853                         goto err_rxq_info;
9854         }
9855         return 0;
9856
9857 err_rxq_info:
9858         /* Rollback successful reg's and free other resources */
9859         while (i--)
9860                 xdp_rxq_info_unreg(&rx[i].xdp_rxq);
9861         kvfree(dev->_rx);
9862         dev->_rx = NULL;
9863         return err;
9864 }
9865
9866 static void netif_free_rx_queues(struct net_device *dev)
9867 {
9868         unsigned int i, count = dev->num_rx_queues;
9869
9870         /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */
9871         if (!dev->_rx)
9872                 return;
9873
9874         for (i = 0; i < count; i++)
9875                 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq);
9876
9877         kvfree(dev->_rx);
9878 }
9879
9880 static void netdev_init_one_queue(struct net_device *dev,
9881                                   struct netdev_queue *queue, void *_unused)
9882 {
9883         /* Initialize queue lock */
9884         spin_lock_init(&queue->_xmit_lock);
9885         netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
9886         queue->xmit_lock_owner = -1;
9887         netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
9888         queue->dev = dev;
9889 #ifdef CONFIG_BQL
9890         dql_init(&queue->dql, HZ);
9891 #endif
9892 }
9893
9894 static void netif_free_tx_queues(struct net_device *dev)
9895 {
9896         kvfree(dev->_tx);
9897 }
9898
9899 static int netif_alloc_netdev_queues(struct net_device *dev)
9900 {
9901         unsigned int count = dev->num_tx_queues;
9902         struct netdev_queue *tx;
9903         size_t sz = count * sizeof(*tx);
9904
9905         if (count < 1 || count > 0xffff)
9906                 return -EINVAL;
9907
9908         tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
9909         if (!tx)
9910                 return -ENOMEM;
9911
9912         dev->_tx = tx;
9913
9914         netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
9915         spin_lock_init(&dev->tx_global_lock);
9916
9917         return 0;
9918 }
9919
9920 void netif_tx_stop_all_queues(struct net_device *dev)
9921 {
9922         unsigned int i;
9923
9924         for (i = 0; i < dev->num_tx_queues; i++) {
9925                 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
9926
9927                 netif_tx_stop_queue(txq);
9928         }
9929 }
9930 EXPORT_SYMBOL(netif_tx_stop_all_queues);
9931
9932 /**
9933  * register_netdevice() - register a network device
9934  * @dev: device to register
9935  *
9936  * Take a prepared network device structure and make it externally accessible.
9937  * A %NETDEV_REGISTER message is sent to the netdev notifier chain.
9938  * Callers must hold the rtnl lock - you may want register_netdev()
9939  * instead of this.
9940  */
9941 int register_netdevice(struct net_device *dev)
9942 {
9943         int ret;
9944         struct net *net = dev_net(dev);
9945
9946         BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE <
9947                      NETDEV_FEATURE_COUNT);
9948         BUG_ON(dev_boot_phase);
9949         ASSERT_RTNL();
9950
9951         might_sleep();
9952
9953         /* When net_device's are persistent, this will be fatal. */
9954         BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
9955         BUG_ON(!net);
9956
9957         ret = ethtool_check_ops(dev->ethtool_ops);
9958         if (ret)
9959                 return ret;
9960
9961         spin_lock_init(&dev->addr_list_lock);
9962         netdev_set_addr_lockdep_class(dev);
9963
9964         ret = dev_get_valid_name(net, dev, dev->name);
9965         if (ret < 0)
9966                 goto out;
9967
9968         ret = -ENOMEM;
9969         dev->name_node = netdev_name_node_head_alloc(dev);
9970         if (!dev->name_node)
9971                 goto out;
9972
9973         /* Init, if this function is available */
9974         if (dev->netdev_ops->ndo_init) {
9975                 ret = dev->netdev_ops->ndo_init(dev);
9976                 if (ret) {
9977                         if (ret > 0)
9978                                 ret = -EIO;
9979                         goto err_free_name;
9980                 }
9981         }
9982
9983         if (((dev->hw_features | dev->features) &
9984              NETIF_F_HW_VLAN_CTAG_FILTER) &&
9985             (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
9986              !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
9987                 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
9988                 ret = -EINVAL;
9989                 goto err_uninit;
9990         }
9991
9992         ret = -EBUSY;
9993         if (!dev->ifindex)
9994                 dev->ifindex = dev_new_index(net);
9995         else if (__dev_get_by_index(net, dev->ifindex))
9996                 goto err_uninit;
9997
9998         /* Transfer changeable features to wanted_features and enable
9999          * software offloads (GSO and GRO).
10000          */
10001         dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF);
10002         dev->features |= NETIF_F_SOFT_FEATURES;
10003
10004         if (dev->udp_tunnel_nic_info) {
10005                 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT;
10006                 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT;
10007         }
10008
10009         dev->wanted_features = dev->features & dev->hw_features;
10010
10011         if (!(dev->flags & IFF_LOOPBACK))
10012                 dev->hw_features |= NETIF_F_NOCACHE_COPY;
10013
10014         /* If IPv4 TCP segmentation offload is supported we should also
10015          * allow the device to enable segmenting the frame with the option
10016          * of ignoring a static IP ID value.  This doesn't enable the
10017          * feature itself but allows the user to enable it later.
10018          */
10019         if (dev->hw_features & NETIF_F_TSO)
10020                 dev->hw_features |= NETIF_F_TSO_MANGLEID;
10021         if (dev->vlan_features & NETIF_F_TSO)
10022                 dev->vlan_features |= NETIF_F_TSO_MANGLEID;
10023         if (dev->mpls_features & NETIF_F_TSO)
10024                 dev->mpls_features |= NETIF_F_TSO_MANGLEID;
10025         if (dev->hw_enc_features & NETIF_F_TSO)
10026                 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
10027
10028         /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
10029          */
10030         dev->vlan_features |= NETIF_F_HIGHDMA;
10031
10032         /* Make NETIF_F_SG inheritable to tunnel devices.
10033          */
10034         dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL;
10035
10036         /* Make NETIF_F_SG inheritable to MPLS.
10037          */
10038         dev->mpls_features |= NETIF_F_SG;
10039
10040         ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
10041         ret = notifier_to_errno(ret);
10042         if (ret)
10043                 goto err_uninit;
10044
10045         ret = netdev_register_kobject(dev);
10046         write_lock(&dev_base_lock);
10047         dev->reg_state = ret ? NETREG_UNREGISTERED : NETREG_REGISTERED;
10048         write_unlock(&dev_base_lock);
10049         if (ret)
10050                 goto err_uninit_notify;
10051
10052         __netdev_update_features(dev);
10053
10054         /*
10055          *      Default initial state at registry is that the
10056          *      device is present.
10057          */
10058
10059         set_bit(__LINK_STATE_PRESENT, &dev->state);
10060
10061         linkwatch_init_dev(dev);
10062
10063         dev_init_scheduler(dev);
10064
10065         netdev_hold(dev, &dev->dev_registered_tracker, GFP_KERNEL);
10066         list_netdevice(dev);
10067
10068         add_device_randomness(dev->dev_addr, dev->addr_len);
10069
10070         /* If the device has permanent device address, driver should
10071          * set dev_addr and also addr_assign_type should be set to
10072          * NET_ADDR_PERM (default value).
10073          */
10074         if (dev->addr_assign_type == NET_ADDR_PERM)
10075                 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
10076
10077         /* Notify protocols, that a new device appeared. */
10078         ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
10079         ret = notifier_to_errno(ret);
10080         if (ret) {
10081                 /* Expect explicit free_netdev() on failure */
10082                 dev->needs_free_netdev = false;
10083                 unregister_netdevice_queue(dev, NULL);
10084                 goto out;
10085         }
10086         /*
10087          *      Prevent userspace races by waiting until the network
10088          *      device is fully setup before sending notifications.
10089          */
10090         if (!dev->rtnl_link_ops ||
10091             dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
10092                 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL);
10093
10094 out:
10095         return ret;
10096
10097 err_uninit_notify:
10098         call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev);
10099 err_uninit:
10100         if (dev->netdev_ops->ndo_uninit)
10101                 dev->netdev_ops->ndo_uninit(dev);
10102         if (dev->priv_destructor)
10103                 dev->priv_destructor(dev);
10104 err_free_name:
10105         netdev_name_node_free(dev->name_node);
10106         goto out;
10107 }
10108 EXPORT_SYMBOL(register_netdevice);
10109
10110 /**
10111  *      init_dummy_netdev       - init a dummy network device for NAPI
10112  *      @dev: device to init
10113  *
10114  *      This takes a network device structure and initialize the minimum
10115  *      amount of fields so it can be used to schedule NAPI polls without
10116  *      registering a full blown interface. This is to be used by drivers
10117  *      that need to tie several hardware interfaces to a single NAPI
10118  *      poll scheduler due to HW limitations.
10119  */
10120 int init_dummy_netdev(struct net_device *dev)
10121 {
10122         /* Clear everything. Note we don't initialize spinlocks
10123          * are they aren't supposed to be taken by any of the
10124          * NAPI code and this dummy netdev is supposed to be
10125          * only ever used for NAPI polls
10126          */
10127         memset(dev, 0, sizeof(struct net_device));
10128
10129         /* make sure we BUG if trying to hit standard
10130          * register/unregister code path
10131          */
10132         dev->reg_state = NETREG_DUMMY;
10133
10134         /* NAPI wants this */
10135         INIT_LIST_HEAD(&dev->napi_list);
10136
10137         /* a dummy interface is started by default */
10138         set_bit(__LINK_STATE_PRESENT, &dev->state);
10139         set_bit(__LINK_STATE_START, &dev->state);
10140
10141         /* napi_busy_loop stats accounting wants this */
10142         dev_net_set(dev, &init_net);
10143
10144         /* Note : We dont allocate pcpu_refcnt for dummy devices,
10145          * because users of this 'device' dont need to change
10146          * its refcount.
10147          */
10148
10149         return 0;
10150 }
10151 EXPORT_SYMBOL_GPL(init_dummy_netdev);
10152
10153
10154 /**
10155  *      register_netdev - register a network device
10156  *      @dev: device to register
10157  *
10158  *      Take a completed network device structure and add it to the kernel
10159  *      interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
10160  *      chain. 0 is returned on success. A negative errno code is returned
10161  *      on a failure to set up the device, or if the name is a duplicate.
10162  *
10163  *      This is a wrapper around register_netdevice that takes the rtnl semaphore
10164  *      and expands the device name if you passed a format string to
10165  *      alloc_netdev.
10166  */
10167 int register_netdev(struct net_device *dev)
10168 {
10169         int err;
10170
10171         if (rtnl_lock_killable())
10172                 return -EINTR;
10173         err = register_netdevice(dev);
10174         rtnl_unlock();
10175         return err;
10176 }
10177 EXPORT_SYMBOL(register_netdev);
10178
10179 int netdev_refcnt_read(const struct net_device *dev)
10180 {
10181 #ifdef CONFIG_PCPU_DEV_REFCNT
10182         int i, refcnt = 0;
10183
10184         for_each_possible_cpu(i)
10185                 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
10186         return refcnt;
10187 #else
10188         return refcount_read(&dev->dev_refcnt);
10189 #endif
10190 }
10191 EXPORT_SYMBOL(netdev_refcnt_read);
10192
10193 int netdev_unregister_timeout_secs __read_mostly = 10;
10194
10195 #define WAIT_REFS_MIN_MSECS 1
10196 #define WAIT_REFS_MAX_MSECS 250
10197 /**
10198  * netdev_wait_allrefs_any - wait until all references are gone.
10199  * @list: list of net_devices to wait on
10200  *
10201  * This is called when unregistering network devices.
10202  *
10203  * Any protocol or device that holds a reference should register
10204  * for netdevice notification, and cleanup and put back the
10205  * reference if they receive an UNREGISTER event.
10206  * We can get stuck here if buggy protocols don't correctly
10207  * call dev_put.
10208  */
10209 static struct net_device *netdev_wait_allrefs_any(struct list_head *list)
10210 {
10211         unsigned long rebroadcast_time, warning_time;
10212         struct net_device *dev;
10213         int wait = 0;
10214
10215         rebroadcast_time = warning_time = jiffies;
10216
10217         list_for_each_entry(dev, list, todo_list)
10218                 if (netdev_refcnt_read(dev) == 1)
10219                         return dev;
10220
10221         while (true) {
10222                 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
10223                         rtnl_lock();
10224
10225                         /* Rebroadcast unregister notification */
10226                         list_for_each_entry(dev, list, todo_list)
10227                                 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
10228
10229                         __rtnl_unlock();
10230                         rcu_barrier();
10231                         rtnl_lock();
10232
10233                         list_for_each_entry(dev, list, todo_list)
10234                                 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
10235                                              &dev->state)) {
10236                                         /* We must not have linkwatch events
10237                                          * pending on unregister. If this
10238                                          * happens, we simply run the queue
10239                                          * unscheduled, resulting in a noop
10240                                          * for this device.
10241                                          */
10242                                         linkwatch_run_queue();
10243                                         break;
10244                                 }
10245
10246                         __rtnl_unlock();
10247
10248                         rebroadcast_time = jiffies;
10249                 }
10250
10251                 if (!wait) {
10252                         rcu_barrier();
10253                         wait = WAIT_REFS_MIN_MSECS;
10254                 } else {
10255                         msleep(wait);
10256                         wait = min(wait << 1, WAIT_REFS_MAX_MSECS);
10257                 }
10258
10259                 list_for_each_entry(dev, list, todo_list)
10260                         if (netdev_refcnt_read(dev) == 1)
10261                                 return dev;
10262
10263                 if (time_after(jiffies, warning_time +
10264                                READ_ONCE(netdev_unregister_timeout_secs) * HZ)) {
10265                         list_for_each_entry(dev, list, todo_list) {
10266                                 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
10267                                          dev->name, netdev_refcnt_read(dev));
10268                                 ref_tracker_dir_print(&dev->refcnt_tracker, 10);
10269                         }
10270
10271                         warning_time = jiffies;
10272                 }
10273         }
10274 }
10275
10276 /* The sequence is:
10277  *
10278  *      rtnl_lock();
10279  *      ...
10280  *      register_netdevice(x1);
10281  *      register_netdevice(x2);
10282  *      ...
10283  *      unregister_netdevice(y1);
10284  *      unregister_netdevice(y2);
10285  *      ...
10286  *      rtnl_unlock();
10287  *      free_netdev(y1);
10288  *      free_netdev(y2);
10289  *
10290  * We are invoked by rtnl_unlock().
10291  * This allows us to deal with problems:
10292  * 1) We can delete sysfs objects which invoke hotplug
10293  *    without deadlocking with linkwatch via keventd.
10294  * 2) Since we run with the RTNL semaphore not held, we can sleep
10295  *    safely in order to wait for the netdev refcnt to drop to zero.
10296  *
10297  * We must not return until all unregister events added during
10298  * the interval the lock was held have been completed.
10299  */
10300 void netdev_run_todo(void)
10301 {
10302         struct net_device *dev, *tmp;
10303         struct list_head list;
10304 #ifdef CONFIG_LOCKDEP
10305         struct list_head unlink_list;
10306
10307         list_replace_init(&net_unlink_list, &unlink_list);
10308
10309         while (!list_empty(&unlink_list)) {
10310                 struct net_device *dev = list_first_entry(&unlink_list,
10311                                                           struct net_device,
10312                                                           unlink_list);
10313                 list_del_init(&dev->unlink_list);
10314                 dev->nested_level = dev->lower_level - 1;
10315         }
10316 #endif
10317
10318         /* Snapshot list, allow later requests */
10319         list_replace_init(&net_todo_list, &list);
10320
10321         __rtnl_unlock();
10322
10323         /* Wait for rcu callbacks to finish before next phase */
10324         if (!list_empty(&list))
10325                 rcu_barrier();
10326
10327         list_for_each_entry_safe(dev, tmp, &list, todo_list) {
10328                 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
10329                         netdev_WARN(dev, "run_todo but not unregistering\n");
10330                         list_del(&dev->todo_list);
10331                         continue;
10332                 }
10333
10334                 write_lock(&dev_base_lock);
10335                 dev->reg_state = NETREG_UNREGISTERED;
10336                 write_unlock(&dev_base_lock);
10337                 linkwatch_forget_dev(dev);
10338         }
10339
10340         while (!list_empty(&list)) {
10341                 dev = netdev_wait_allrefs_any(&list);
10342                 list_del(&dev->todo_list);
10343
10344                 /* paranoia */
10345                 BUG_ON(netdev_refcnt_read(dev) != 1);
10346                 BUG_ON(!list_empty(&dev->ptype_all));
10347                 BUG_ON(!list_empty(&dev->ptype_specific));
10348                 WARN_ON(rcu_access_pointer(dev->ip_ptr));
10349                 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
10350
10351                 if (dev->priv_destructor)
10352                         dev->priv_destructor(dev);
10353                 if (dev->needs_free_netdev)
10354                         free_netdev(dev);
10355
10356                 if (atomic_dec_and_test(&dev_net(dev)->dev_unreg_count))
10357                         wake_up(&netdev_unregistering_wq);
10358
10359                 /* Free network device */
10360                 kobject_put(&dev->dev.kobj);
10361         }
10362 }
10363
10364 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has
10365  * all the same fields in the same order as net_device_stats, with only
10366  * the type differing, but rtnl_link_stats64 may have additional fields
10367  * at the end for newer counters.
10368  */
10369 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
10370                              const struct net_device_stats *netdev_stats)
10371 {
10372         size_t i, n = sizeof(*netdev_stats) / sizeof(atomic_long_t);
10373         const atomic_long_t *src = (atomic_long_t *)netdev_stats;
10374         u64 *dst = (u64 *)stats64;
10375
10376         BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64));
10377         for (i = 0; i < n; i++)
10378                 dst[i] = (unsigned long)atomic_long_read(&src[i]);
10379         /* zero out counters that only exist in rtnl_link_stats64 */
10380         memset((char *)stats64 + n * sizeof(u64), 0,
10381                sizeof(*stats64) - n * sizeof(u64));
10382 }
10383 EXPORT_SYMBOL(netdev_stats_to_stats64);
10384
10385 struct net_device_core_stats __percpu *netdev_core_stats_alloc(struct net_device *dev)
10386 {
10387         struct net_device_core_stats __percpu *p;
10388
10389         p = alloc_percpu_gfp(struct net_device_core_stats,
10390                              GFP_ATOMIC | __GFP_NOWARN);
10391
10392         if (p && cmpxchg(&dev->core_stats, NULL, p))
10393                 free_percpu(p);
10394
10395         /* This READ_ONCE() pairs with the cmpxchg() above */
10396         return READ_ONCE(dev->core_stats);
10397 }
10398 EXPORT_SYMBOL(netdev_core_stats_alloc);
10399
10400 /**
10401  *      dev_get_stats   - get network device statistics
10402  *      @dev: device to get statistics from
10403  *      @storage: place to store stats
10404  *
10405  *      Get network statistics from device. Return @storage.
10406  *      The device driver may provide its own method by setting
10407  *      dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
10408  *      otherwise the internal statistics structure is used.
10409  */
10410 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
10411                                         struct rtnl_link_stats64 *storage)
10412 {
10413         const struct net_device_ops *ops = dev->netdev_ops;
10414         const struct net_device_core_stats __percpu *p;
10415
10416         if (ops->ndo_get_stats64) {
10417                 memset(storage, 0, sizeof(*storage));
10418                 ops->ndo_get_stats64(dev, storage);
10419         } else if (ops->ndo_get_stats) {
10420                 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
10421         } else {
10422                 netdev_stats_to_stats64(storage, &dev->stats);
10423         }
10424
10425         /* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */
10426         p = READ_ONCE(dev->core_stats);
10427         if (p) {
10428                 const struct net_device_core_stats *core_stats;
10429                 int i;
10430
10431                 for_each_possible_cpu(i) {
10432                         core_stats = per_cpu_ptr(p, i);
10433                         storage->rx_dropped += READ_ONCE(core_stats->rx_dropped);
10434                         storage->tx_dropped += READ_ONCE(core_stats->tx_dropped);
10435                         storage->rx_nohandler += READ_ONCE(core_stats->rx_nohandler);
10436                         storage->rx_otherhost_dropped += READ_ONCE(core_stats->rx_otherhost_dropped);
10437                 }
10438         }
10439         return storage;
10440 }
10441 EXPORT_SYMBOL(dev_get_stats);
10442
10443 /**
10444  *      dev_fetch_sw_netstats - get per-cpu network device statistics
10445  *      @s: place to store stats
10446  *      @netstats: per-cpu network stats to read from
10447  *
10448  *      Read per-cpu network statistics and populate the related fields in @s.
10449  */
10450 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s,
10451                            const struct pcpu_sw_netstats __percpu *netstats)
10452 {
10453         int cpu;
10454
10455         for_each_possible_cpu(cpu) {
10456                 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
10457                 const struct pcpu_sw_netstats *stats;
10458                 unsigned int start;
10459
10460                 stats = per_cpu_ptr(netstats, cpu);
10461                 do {
10462                         start = u64_stats_fetch_begin(&stats->syncp);
10463                         rx_packets = u64_stats_read(&stats->rx_packets);
10464                         rx_bytes   = u64_stats_read(&stats->rx_bytes);
10465                         tx_packets = u64_stats_read(&stats->tx_packets);
10466                         tx_bytes   = u64_stats_read(&stats->tx_bytes);
10467                 } while (u64_stats_fetch_retry(&stats->syncp, start));
10468
10469                 s->rx_packets += rx_packets;
10470                 s->rx_bytes   += rx_bytes;
10471                 s->tx_packets += tx_packets;
10472                 s->tx_bytes   += tx_bytes;
10473         }
10474 }
10475 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats);
10476
10477 /**
10478  *      dev_get_tstats64 - ndo_get_stats64 implementation
10479  *      @dev: device to get statistics from
10480  *      @s: place to store stats
10481  *
10482  *      Populate @s from dev->stats and dev->tstats. Can be used as
10483  *      ndo_get_stats64() callback.
10484  */
10485 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s)
10486 {
10487         netdev_stats_to_stats64(s, &dev->stats);
10488         dev_fetch_sw_netstats(s, dev->tstats);
10489 }
10490 EXPORT_SYMBOL_GPL(dev_get_tstats64);
10491
10492 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
10493 {
10494         struct netdev_queue *queue = dev_ingress_queue(dev);
10495
10496 #ifdef CONFIG_NET_CLS_ACT
10497         if (queue)
10498                 return queue;
10499         queue = kzalloc(sizeof(*queue), GFP_KERNEL);
10500         if (!queue)
10501                 return NULL;
10502         netdev_init_one_queue(dev, queue, NULL);
10503         RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
10504         queue->qdisc_sleeping = &noop_qdisc;
10505         rcu_assign_pointer(dev->ingress_queue, queue);
10506 #endif
10507         return queue;
10508 }
10509
10510 static const struct ethtool_ops default_ethtool_ops;
10511
10512 void netdev_set_default_ethtool_ops(struct net_device *dev,
10513                                     const struct ethtool_ops *ops)
10514 {
10515         if (dev->ethtool_ops == &default_ethtool_ops)
10516                 dev->ethtool_ops = ops;
10517 }
10518 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
10519
10520 /**
10521  * netdev_sw_irq_coalesce_default_on() - enable SW IRQ coalescing by default
10522  * @dev: netdev to enable the IRQ coalescing on
10523  *
10524  * Sets a conservative default for SW IRQ coalescing. Users can use
10525  * sysfs attributes to override the default values.
10526  */
10527 void netdev_sw_irq_coalesce_default_on(struct net_device *dev)
10528 {
10529         WARN_ON(dev->reg_state == NETREG_REGISTERED);
10530
10531         dev->gro_flush_timeout = 20000;
10532         dev->napi_defer_hard_irqs = 1;
10533 }
10534 EXPORT_SYMBOL_GPL(netdev_sw_irq_coalesce_default_on);
10535
10536 void netdev_freemem(struct net_device *dev)
10537 {
10538         char *addr = (char *)dev - dev->padded;
10539
10540         kvfree(addr);
10541 }
10542
10543 /**
10544  * alloc_netdev_mqs - allocate network device
10545  * @sizeof_priv: size of private data to allocate space for
10546  * @name: device name format string
10547  * @name_assign_type: origin of device name
10548  * @setup: callback to initialize device
10549  * @txqs: the number of TX subqueues to allocate
10550  * @rxqs: the number of RX subqueues to allocate
10551  *
10552  * Allocates a struct net_device with private data area for driver use
10553  * and performs basic initialization.  Also allocates subqueue structs
10554  * for each queue on the device.
10555  */
10556 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
10557                 unsigned char name_assign_type,
10558                 void (*setup)(struct net_device *),
10559                 unsigned int txqs, unsigned int rxqs)
10560 {
10561         struct net_device *dev;
10562         unsigned int alloc_size;
10563         struct net_device *p;
10564
10565         BUG_ON(strlen(name) >= sizeof(dev->name));
10566
10567         if (txqs < 1) {
10568                 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
10569                 return NULL;
10570         }
10571
10572         if (rxqs < 1) {
10573                 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
10574                 return NULL;
10575         }
10576
10577         alloc_size = sizeof(struct net_device);
10578         if (sizeof_priv) {
10579                 /* ensure 32-byte alignment of private area */
10580                 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
10581                 alloc_size += sizeof_priv;
10582         }
10583         /* ensure 32-byte alignment of whole construct */
10584         alloc_size += NETDEV_ALIGN - 1;
10585
10586         p = kvzalloc(alloc_size, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
10587         if (!p)
10588                 return NULL;
10589
10590         dev = PTR_ALIGN(p, NETDEV_ALIGN);
10591         dev->padded = (char *)dev - (char *)p;
10592
10593         ref_tracker_dir_init(&dev->refcnt_tracker, 128);
10594 #ifdef CONFIG_PCPU_DEV_REFCNT
10595         dev->pcpu_refcnt = alloc_percpu(int);
10596         if (!dev->pcpu_refcnt)
10597                 goto free_dev;
10598         __dev_hold(dev);
10599 #else
10600         refcount_set(&dev->dev_refcnt, 1);
10601 #endif
10602
10603         if (dev_addr_init(dev))
10604                 goto free_pcpu;
10605
10606         dev_mc_init(dev);
10607         dev_uc_init(dev);
10608
10609         dev_net_set(dev, &init_net);
10610
10611         dev->gso_max_size = GSO_LEGACY_MAX_SIZE;
10612         dev->gso_max_segs = GSO_MAX_SEGS;
10613         dev->gro_max_size = GRO_LEGACY_MAX_SIZE;
10614         dev->tso_max_size = TSO_LEGACY_MAX_SIZE;
10615         dev->tso_max_segs = TSO_MAX_SEGS;
10616         dev->upper_level = 1;
10617         dev->lower_level = 1;
10618 #ifdef CONFIG_LOCKDEP
10619         dev->nested_level = 0;
10620         INIT_LIST_HEAD(&dev->unlink_list);
10621 #endif
10622
10623         INIT_LIST_HEAD(&dev->napi_list);
10624         INIT_LIST_HEAD(&dev->unreg_list);
10625         INIT_LIST_HEAD(&dev->close_list);
10626         INIT_LIST_HEAD(&dev->link_watch_list);
10627         INIT_LIST_HEAD(&dev->adj_list.upper);
10628         INIT_LIST_HEAD(&dev->adj_list.lower);
10629         INIT_LIST_HEAD(&dev->ptype_all);
10630         INIT_LIST_HEAD(&dev->ptype_specific);
10631         INIT_LIST_HEAD(&dev->net_notifier_list);
10632 #ifdef CONFIG_NET_SCHED
10633         hash_init(dev->qdisc_hash);
10634 #endif
10635         dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
10636         setup(dev);
10637
10638         if (!dev->tx_queue_len) {
10639                 dev->priv_flags |= IFF_NO_QUEUE;
10640                 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
10641         }
10642
10643         dev->num_tx_queues = txqs;
10644         dev->real_num_tx_queues = txqs;
10645         if (netif_alloc_netdev_queues(dev))
10646                 goto free_all;
10647
10648         dev->num_rx_queues = rxqs;
10649         dev->real_num_rx_queues = rxqs;
10650         if (netif_alloc_rx_queues(dev))
10651                 goto free_all;
10652
10653         strcpy(dev->name, name);
10654         dev->name_assign_type = name_assign_type;
10655         dev->group = INIT_NETDEV_GROUP;
10656         if (!dev->ethtool_ops)
10657                 dev->ethtool_ops = &default_ethtool_ops;
10658
10659         nf_hook_netdev_init(dev);
10660
10661         return dev;
10662
10663 free_all:
10664         free_netdev(dev);
10665         return NULL;
10666
10667 free_pcpu:
10668 #ifdef CONFIG_PCPU_DEV_REFCNT
10669         free_percpu(dev->pcpu_refcnt);
10670 free_dev:
10671 #endif
10672         netdev_freemem(dev);
10673         return NULL;
10674 }
10675 EXPORT_SYMBOL(alloc_netdev_mqs);
10676
10677 /**
10678  * free_netdev - free network device
10679  * @dev: device
10680  *
10681  * This function does the last stage of destroying an allocated device
10682  * interface. The reference to the device object is released. If this
10683  * is the last reference then it will be freed.Must be called in process
10684  * context.
10685  */
10686 void free_netdev(struct net_device *dev)
10687 {
10688         struct napi_struct *p, *n;
10689
10690         might_sleep();
10691
10692         /* When called immediately after register_netdevice() failed the unwind
10693          * handling may still be dismantling the device. Handle that case by
10694          * deferring the free.
10695          */
10696         if (dev->reg_state == NETREG_UNREGISTERING) {
10697                 ASSERT_RTNL();
10698                 dev->needs_free_netdev = true;
10699                 return;
10700         }
10701
10702         netif_free_tx_queues(dev);
10703         netif_free_rx_queues(dev);
10704
10705         kfree(rcu_dereference_protected(dev->ingress_queue, 1));
10706
10707         /* Flush device addresses */
10708         dev_addr_flush(dev);
10709
10710         list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
10711                 netif_napi_del(p);
10712
10713         ref_tracker_dir_exit(&dev->refcnt_tracker);
10714 #ifdef CONFIG_PCPU_DEV_REFCNT
10715         free_percpu(dev->pcpu_refcnt);
10716         dev->pcpu_refcnt = NULL;
10717 #endif
10718         free_percpu(dev->core_stats);
10719         dev->core_stats = NULL;
10720         free_percpu(dev->xdp_bulkq);
10721         dev->xdp_bulkq = NULL;
10722
10723         /*  Compatibility with error handling in drivers */
10724         if (dev->reg_state == NETREG_UNINITIALIZED) {
10725                 netdev_freemem(dev);
10726                 return;
10727         }
10728
10729         BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
10730         dev->reg_state = NETREG_RELEASED;
10731
10732         /* will free via device release */
10733         put_device(&dev->dev);
10734 }
10735 EXPORT_SYMBOL(free_netdev);
10736
10737 /**
10738  *      synchronize_net -  Synchronize with packet receive processing
10739  *
10740  *      Wait for packets currently being received to be done.
10741  *      Does not block later packets from starting.
10742  */
10743 void synchronize_net(void)
10744 {
10745         might_sleep();
10746         if (rtnl_is_locked())
10747                 synchronize_rcu_expedited();
10748         else
10749                 synchronize_rcu();
10750 }
10751 EXPORT_SYMBOL(synchronize_net);
10752
10753 /**
10754  *      unregister_netdevice_queue - remove device from the kernel
10755  *      @dev: device
10756  *      @head: list
10757  *
10758  *      This function shuts down a device interface and removes it
10759  *      from the kernel tables.
10760  *      If head not NULL, device is queued to be unregistered later.
10761  *
10762  *      Callers must hold the rtnl semaphore.  You may want
10763  *      unregister_netdev() instead of this.
10764  */
10765
10766 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
10767 {
10768         ASSERT_RTNL();
10769
10770         if (head) {
10771                 list_move_tail(&dev->unreg_list, head);
10772         } else {
10773                 LIST_HEAD(single);
10774
10775                 list_add(&dev->unreg_list, &single);
10776                 unregister_netdevice_many(&single);
10777         }
10778 }
10779 EXPORT_SYMBOL(unregister_netdevice_queue);
10780
10781 void unregister_netdevice_many_notify(struct list_head *head,
10782                                       u32 portid, const struct nlmsghdr *nlh)
10783 {
10784         struct net_device *dev, *tmp;
10785         LIST_HEAD(close_head);
10786
10787         BUG_ON(dev_boot_phase);
10788         ASSERT_RTNL();
10789
10790         if (list_empty(head))
10791                 return;
10792
10793         list_for_each_entry_safe(dev, tmp, head, unreg_list) {
10794                 /* Some devices call without registering
10795                  * for initialization unwind. Remove those
10796                  * devices and proceed with the remaining.
10797                  */
10798                 if (dev->reg_state == NETREG_UNINITIALIZED) {
10799                         pr_debug("unregister_netdevice: device %s/%p never was registered\n",
10800                                  dev->name, dev);
10801
10802                         WARN_ON(1);
10803                         list_del(&dev->unreg_list);
10804                         continue;
10805                 }
10806                 dev->dismantle = true;
10807                 BUG_ON(dev->reg_state != NETREG_REGISTERED);
10808         }
10809
10810         /* If device is running, close it first. */
10811         list_for_each_entry(dev, head, unreg_list)
10812                 list_add_tail(&dev->close_list, &close_head);
10813         dev_close_many(&close_head, true);
10814
10815         list_for_each_entry(dev, head, unreg_list) {
10816                 /* And unlink it from device chain. */
10817                 write_lock(&dev_base_lock);
10818                 unlist_netdevice(dev, false);
10819                 dev->reg_state = NETREG_UNREGISTERING;
10820                 write_unlock(&dev_base_lock);
10821         }
10822         flush_all_backlogs();
10823
10824         synchronize_net();
10825
10826         list_for_each_entry(dev, head, unreg_list) {
10827                 struct sk_buff *skb = NULL;
10828
10829                 /* Shutdown queueing discipline. */
10830                 dev_shutdown(dev);
10831
10832                 dev_xdp_uninstall(dev);
10833
10834                 netdev_offload_xstats_disable_all(dev);
10835
10836                 /* Notify protocols, that we are about to destroy
10837                  * this device. They should clean all the things.
10838                  */
10839                 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
10840
10841                 if (!dev->rtnl_link_ops ||
10842                     dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
10843                         skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0,
10844                                                      GFP_KERNEL, NULL, 0,
10845                                                      portid, nlmsg_seq(nlh));
10846
10847                 /*
10848                  *      Flush the unicast and multicast chains
10849                  */
10850                 dev_uc_flush(dev);
10851                 dev_mc_flush(dev);
10852
10853                 netdev_name_node_alt_flush(dev);
10854                 netdev_name_node_free(dev->name_node);
10855
10856                 call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev);
10857
10858                 if (dev->netdev_ops->ndo_uninit)
10859                         dev->netdev_ops->ndo_uninit(dev);
10860
10861                 if (skb)
10862                         rtmsg_ifinfo_send(skb, dev, GFP_KERNEL, portid, nlh);
10863
10864                 /* Notifier chain MUST detach us all upper devices. */
10865                 WARN_ON(netdev_has_any_upper_dev(dev));
10866                 WARN_ON(netdev_has_any_lower_dev(dev));
10867
10868                 /* Remove entries from kobject tree */
10869                 netdev_unregister_kobject(dev);
10870 #ifdef CONFIG_XPS
10871                 /* Remove XPS queueing entries */
10872                 netif_reset_xps_queues_gt(dev, 0);
10873 #endif
10874         }
10875
10876         synchronize_net();
10877
10878         list_for_each_entry(dev, head, unreg_list) {
10879                 netdev_put(dev, &dev->dev_registered_tracker);
10880                 net_set_todo(dev);
10881         }
10882
10883         list_del(head);
10884 }
10885
10886 /**
10887  *      unregister_netdevice_many - unregister many devices
10888  *      @head: list of devices
10889  *
10890  *  Note: As most callers use a stack allocated list_head,
10891  *  we force a list_del() to make sure stack wont be corrupted later.
10892  */
10893 void unregister_netdevice_many(struct list_head *head)
10894 {
10895         unregister_netdevice_many_notify(head, 0, NULL);
10896 }
10897 EXPORT_SYMBOL(unregister_netdevice_many);
10898
10899 /**
10900  *      unregister_netdev - remove device from the kernel
10901  *      @dev: device
10902  *
10903  *      This function shuts down a device interface and removes it
10904  *      from the kernel tables.
10905  *
10906  *      This is just a wrapper for unregister_netdevice that takes
10907  *      the rtnl semaphore.  In general you want to use this and not
10908  *      unregister_netdevice.
10909  */
10910 void unregister_netdev(struct net_device *dev)
10911 {
10912         rtnl_lock();
10913         unregister_netdevice(dev);
10914         rtnl_unlock();
10915 }
10916 EXPORT_SYMBOL(unregister_netdev);
10917
10918 /**
10919  *      __dev_change_net_namespace - move device to different nethost namespace
10920  *      @dev: device
10921  *      @net: network namespace
10922  *      @pat: If not NULL name pattern to try if the current device name
10923  *            is already taken in the destination network namespace.
10924  *      @new_ifindex: If not zero, specifies device index in the target
10925  *                    namespace.
10926  *
10927  *      This function shuts down a device interface and moves it
10928  *      to a new network namespace. On success 0 is returned, on
10929  *      a failure a netagive errno code is returned.
10930  *
10931  *      Callers must hold the rtnl semaphore.
10932  */
10933
10934 int __dev_change_net_namespace(struct net_device *dev, struct net *net,
10935                                const char *pat, int new_ifindex)
10936 {
10937         struct net *net_old = dev_net(dev);
10938         int err, new_nsid;
10939
10940         ASSERT_RTNL();
10941
10942         /* Don't allow namespace local devices to be moved. */
10943         err = -EINVAL;
10944         if (dev->features & NETIF_F_NETNS_LOCAL)
10945                 goto out;
10946
10947         /* Ensure the device has been registrered */
10948         if (dev->reg_state != NETREG_REGISTERED)
10949                 goto out;
10950
10951         /* Get out if there is nothing todo */
10952         err = 0;
10953         if (net_eq(net_old, net))
10954                 goto out;
10955
10956         /* Pick the destination device name, and ensure
10957          * we can use it in the destination network namespace.
10958          */
10959         err = -EEXIST;
10960         if (netdev_name_in_use(net, dev->name)) {
10961                 /* We get here if we can't use the current device name */
10962                 if (!pat)
10963                         goto out;
10964                 err = dev_get_valid_name(net, dev, pat);
10965                 if (err < 0)
10966                         goto out;
10967         }
10968
10969         /* Check that new_ifindex isn't used yet. */
10970         err = -EBUSY;
10971         if (new_ifindex && __dev_get_by_index(net, new_ifindex))
10972                 goto out;
10973
10974         /*
10975          * And now a mini version of register_netdevice unregister_netdevice.
10976          */
10977
10978         /* If device is running close it first. */
10979         dev_close(dev);
10980
10981         /* And unlink it from device chain */
10982         unlist_netdevice(dev, true);
10983
10984         synchronize_net();
10985
10986         /* Shutdown queueing discipline. */
10987         dev_shutdown(dev);
10988
10989         /* Notify protocols, that we are about to destroy
10990          * this device. They should clean all the things.
10991          *
10992          * Note that dev->reg_state stays at NETREG_REGISTERED.
10993          * This is wanted because this way 8021q and macvlan know
10994          * the device is just moving and can keep their slaves up.
10995          */
10996         call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
10997         rcu_barrier();
10998
10999         new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL);
11000         /* If there is an ifindex conflict assign a new one */
11001         if (!new_ifindex) {
11002                 if (__dev_get_by_index(net, dev->ifindex))
11003                         new_ifindex = dev_new_index(net);
11004                 else
11005                         new_ifindex = dev->ifindex;
11006         }
11007
11008         rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid,
11009                             new_ifindex);
11010
11011         /*
11012          *      Flush the unicast and multicast chains
11013          */
11014         dev_uc_flush(dev);
11015         dev_mc_flush(dev);
11016
11017         /* Send a netdev-removed uevent to the old namespace */
11018         kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
11019         netdev_adjacent_del_links(dev);
11020
11021         /* Move per-net netdevice notifiers that are following the netdevice */
11022         move_netdevice_notifiers_dev_net(dev, net);
11023
11024         /* Actually switch the network namespace */
11025         dev_net_set(dev, net);
11026         dev->ifindex = new_ifindex;
11027
11028         /* Send a netdev-add uevent to the new namespace */
11029         kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
11030         netdev_adjacent_add_links(dev);
11031
11032         /* Fixup kobjects */
11033         err = device_rename(&dev->dev, dev->name);
11034         WARN_ON(err);
11035
11036         /* Adapt owner in case owning user namespace of target network
11037          * namespace is different from the original one.
11038          */
11039         err = netdev_change_owner(dev, net_old, net);
11040         WARN_ON(err);
11041
11042         /* Add the device back in the hashes */
11043         list_netdevice(dev);
11044
11045         /* Notify protocols, that a new device appeared. */
11046         call_netdevice_notifiers(NETDEV_REGISTER, dev);
11047
11048         /*
11049          *      Prevent userspace races by waiting until the network
11050          *      device is fully setup before sending notifications.
11051          */
11052         rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL);
11053
11054         synchronize_net();
11055         err = 0;
11056 out:
11057         return err;
11058 }
11059 EXPORT_SYMBOL_GPL(__dev_change_net_namespace);
11060
11061 static int dev_cpu_dead(unsigned int oldcpu)
11062 {
11063         struct sk_buff **list_skb;
11064         struct sk_buff *skb;
11065         unsigned int cpu;
11066         struct softnet_data *sd, *oldsd, *remsd = NULL;
11067
11068         local_irq_disable();
11069         cpu = smp_processor_id();
11070         sd = &per_cpu(softnet_data, cpu);
11071         oldsd = &per_cpu(softnet_data, oldcpu);
11072
11073         /* Find end of our completion_queue. */
11074         list_skb = &sd->completion_queue;
11075         while (*list_skb)
11076                 list_skb = &(*list_skb)->next;
11077         /* Append completion queue from offline CPU. */
11078         *list_skb = oldsd->completion_queue;
11079         oldsd->completion_queue = NULL;
11080
11081         /* Append output queue from offline CPU. */
11082         if (oldsd->output_queue) {
11083                 *sd->output_queue_tailp = oldsd->output_queue;
11084                 sd->output_queue_tailp = oldsd->output_queue_tailp;
11085                 oldsd->output_queue = NULL;
11086                 oldsd->output_queue_tailp = &oldsd->output_queue;
11087         }
11088         /* Append NAPI poll list from offline CPU, with one exception :
11089          * process_backlog() must be called by cpu owning percpu backlog.
11090          * We properly handle process_queue & input_pkt_queue later.
11091          */
11092         while (!list_empty(&oldsd->poll_list)) {
11093                 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
11094                                                             struct napi_struct,
11095                                                             poll_list);
11096
11097                 list_del_init(&napi->poll_list);
11098                 if (napi->poll == process_backlog)
11099                         napi->state = 0;
11100                 else
11101                         ____napi_schedule(sd, napi);
11102         }
11103
11104         raise_softirq_irqoff(NET_TX_SOFTIRQ);
11105         local_irq_enable();
11106
11107 #ifdef CONFIG_RPS
11108         remsd = oldsd->rps_ipi_list;
11109         oldsd->rps_ipi_list = NULL;
11110 #endif
11111         /* send out pending IPI's on offline CPU */
11112         net_rps_send_ipi(remsd);
11113
11114         /* Process offline CPU's input_pkt_queue */
11115         while ((skb = __skb_dequeue(&oldsd->process_queue))) {
11116                 netif_rx(skb);
11117                 input_queue_head_incr(oldsd);
11118         }
11119         while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
11120                 netif_rx(skb);
11121                 input_queue_head_incr(oldsd);
11122         }
11123
11124         return 0;
11125 }
11126
11127 /**
11128  *      netdev_increment_features - increment feature set by one
11129  *      @all: current feature set
11130  *      @one: new feature set
11131  *      @mask: mask feature set
11132  *
11133  *      Computes a new feature set after adding a device with feature set
11134  *      @one to the master device with current feature set @all.  Will not
11135  *      enable anything that is off in @mask. Returns the new feature set.
11136  */
11137 netdev_features_t netdev_increment_features(netdev_features_t all,
11138         netdev_features_t one, netdev_features_t mask)
11139 {
11140         if (mask & NETIF_F_HW_CSUM)
11141                 mask |= NETIF_F_CSUM_MASK;
11142         mask |= NETIF_F_VLAN_CHALLENGED;
11143
11144         all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask;
11145         all &= one | ~NETIF_F_ALL_FOR_ALL;
11146
11147         /* If one device supports hw checksumming, set for all. */
11148         if (all & NETIF_F_HW_CSUM)
11149                 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM);
11150
11151         return all;
11152 }
11153 EXPORT_SYMBOL(netdev_increment_features);
11154
11155 static struct hlist_head * __net_init netdev_create_hash(void)
11156 {
11157         int i;
11158         struct hlist_head *hash;
11159
11160         hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL);
11161         if (hash != NULL)
11162                 for (i = 0; i < NETDEV_HASHENTRIES; i++)
11163                         INIT_HLIST_HEAD(&hash[i]);
11164
11165         return hash;
11166 }
11167
11168 /* Initialize per network namespace state */
11169 static int __net_init netdev_init(struct net *net)
11170 {
11171         BUILD_BUG_ON(GRO_HASH_BUCKETS >
11172                      8 * sizeof_field(struct napi_struct, gro_bitmask));
11173
11174         INIT_LIST_HEAD(&net->dev_base_head);
11175
11176         net->dev_name_head = netdev_create_hash();
11177         if (net->dev_name_head == NULL)
11178                 goto err_name;
11179
11180         net->dev_index_head = netdev_create_hash();
11181         if (net->dev_index_head == NULL)
11182                 goto err_idx;
11183
11184         RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain);
11185
11186         return 0;
11187
11188 err_idx:
11189         kfree(net->dev_name_head);
11190 err_name:
11191         return -ENOMEM;
11192 }
11193
11194 /**
11195  *      netdev_drivername - network driver for the device
11196  *      @dev: network device
11197  *
11198  *      Determine network driver for device.
11199  */
11200 const char *netdev_drivername(const struct net_device *dev)
11201 {
11202         const struct device_driver *driver;
11203         const struct device *parent;
11204         const char *empty = "";
11205
11206         parent = dev->dev.parent;
11207         if (!parent)
11208                 return empty;
11209
11210         driver = parent->driver;
11211         if (driver && driver->name)
11212                 return driver->name;
11213         return empty;
11214 }
11215
11216 static void __netdev_printk(const char *level, const struct net_device *dev,
11217                             struct va_format *vaf)
11218 {
11219         if (dev && dev->dev.parent) {
11220                 dev_printk_emit(level[1] - '0',
11221                                 dev->dev.parent,
11222                                 "%s %s %s%s: %pV",
11223                                 dev_driver_string(dev->dev.parent),
11224                                 dev_name(dev->dev.parent),
11225                                 netdev_name(dev), netdev_reg_state(dev),
11226                                 vaf);
11227         } else if (dev) {
11228                 printk("%s%s%s: %pV",
11229                        level, netdev_name(dev), netdev_reg_state(dev), vaf);
11230         } else {
11231                 printk("%s(NULL net_device): %pV", level, vaf);
11232         }
11233 }
11234
11235 void netdev_printk(const char *level, const struct net_device *dev,
11236                    const char *format, ...)
11237 {
11238         struct va_format vaf;
11239         va_list args;
11240
11241         va_start(args, format);
11242
11243         vaf.fmt = format;
11244         vaf.va = &args;
11245
11246         __netdev_printk(level, dev, &vaf);
11247
11248         va_end(args);
11249 }
11250 EXPORT_SYMBOL(netdev_printk);
11251
11252 #define define_netdev_printk_level(func, level)                 \
11253 void func(const struct net_device *dev, const char *fmt, ...)   \
11254 {                                                               \
11255         struct va_format vaf;                                   \
11256         va_list args;                                           \
11257                                                                 \
11258         va_start(args, fmt);                                    \
11259                                                                 \
11260         vaf.fmt = fmt;                                          \
11261         vaf.va = &args;                                         \
11262                                                                 \
11263         __netdev_printk(level, dev, &vaf);                      \
11264                                                                 \
11265         va_end(args);                                           \
11266 }                                                               \
11267 EXPORT_SYMBOL(func);
11268
11269 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
11270 define_netdev_printk_level(netdev_alert, KERN_ALERT);
11271 define_netdev_printk_level(netdev_crit, KERN_CRIT);
11272 define_netdev_printk_level(netdev_err, KERN_ERR);
11273 define_netdev_printk_level(netdev_warn, KERN_WARNING);
11274 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
11275 define_netdev_printk_level(netdev_info, KERN_INFO);
11276
11277 static void __net_exit netdev_exit(struct net *net)
11278 {
11279         kfree(net->dev_name_head);
11280         kfree(net->dev_index_head);
11281         if (net != &init_net)
11282                 WARN_ON_ONCE(!list_empty(&net->dev_base_head));
11283 }
11284
11285 static struct pernet_operations __net_initdata netdev_net_ops = {
11286         .init = netdev_init,
11287         .exit = netdev_exit,
11288 };
11289
11290 static void __net_exit default_device_exit_net(struct net *net)
11291 {
11292         struct net_device *dev, *aux;
11293         /*
11294          * Push all migratable network devices back to the
11295          * initial network namespace
11296          */
11297         ASSERT_RTNL();
11298         for_each_netdev_safe(net, dev, aux) {
11299                 int err;
11300                 char fb_name[IFNAMSIZ];
11301
11302                 /* Ignore unmoveable devices (i.e. loopback) */
11303                 if (dev->features & NETIF_F_NETNS_LOCAL)
11304                         continue;
11305
11306                 /* Leave virtual devices for the generic cleanup */
11307                 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund)
11308                         continue;
11309
11310                 /* Push remaining network devices to init_net */
11311                 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
11312                 if (netdev_name_in_use(&init_net, fb_name))
11313                         snprintf(fb_name, IFNAMSIZ, "dev%%d");
11314                 err = dev_change_net_namespace(dev, &init_net, fb_name);
11315                 if (err) {
11316                         pr_emerg("%s: failed to move %s to init_net: %d\n",
11317                                  __func__, dev->name, err);
11318                         BUG();
11319                 }
11320         }
11321 }
11322
11323 static void __net_exit default_device_exit_batch(struct list_head *net_list)
11324 {
11325         /* At exit all network devices most be removed from a network
11326          * namespace.  Do this in the reverse order of registration.
11327          * Do this across as many network namespaces as possible to
11328          * improve batching efficiency.
11329          */
11330         struct net_device *dev;
11331         struct net *net;
11332         LIST_HEAD(dev_kill_list);
11333
11334         rtnl_lock();
11335         list_for_each_entry(net, net_list, exit_list) {
11336                 default_device_exit_net(net);
11337                 cond_resched();
11338         }
11339
11340         list_for_each_entry(net, net_list, exit_list) {
11341                 for_each_netdev_reverse(net, dev) {
11342                         if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
11343                                 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
11344                         else
11345                                 unregister_netdevice_queue(dev, &dev_kill_list);
11346                 }
11347         }
11348         unregister_netdevice_many(&dev_kill_list);
11349         rtnl_unlock();
11350 }
11351
11352 static struct pernet_operations __net_initdata default_device_ops = {
11353         .exit_batch = default_device_exit_batch,
11354 };
11355
11356 /*
11357  *      Initialize the DEV module. At boot time this walks the device list and
11358  *      unhooks any devices that fail to initialise (normally hardware not
11359  *      present) and leaves us with a valid list of present and active devices.
11360  *
11361  */
11362
11363 /*
11364  *       This is called single threaded during boot, so no need
11365  *       to take the rtnl semaphore.
11366  */
11367 static int __init net_dev_init(void)
11368 {
11369         int i, rc = -ENOMEM;
11370
11371         BUG_ON(!dev_boot_phase);
11372
11373         if (dev_proc_init())
11374                 goto out;
11375
11376         if (netdev_kobject_init())
11377                 goto out;
11378
11379         INIT_LIST_HEAD(&ptype_all);
11380         for (i = 0; i < PTYPE_HASH_SIZE; i++)
11381                 INIT_LIST_HEAD(&ptype_base[i]);
11382
11383         if (register_pernet_subsys(&netdev_net_ops))
11384                 goto out;
11385
11386         /*
11387          *      Initialise the packet receive queues.
11388          */
11389
11390         for_each_possible_cpu(i) {
11391                 struct work_struct *flush = per_cpu_ptr(&flush_works, i);
11392                 struct softnet_data *sd = &per_cpu(softnet_data, i);
11393
11394                 INIT_WORK(flush, flush_backlog);
11395
11396                 skb_queue_head_init(&sd->input_pkt_queue);
11397                 skb_queue_head_init(&sd->process_queue);
11398 #ifdef CONFIG_XFRM_OFFLOAD
11399                 skb_queue_head_init(&sd->xfrm_backlog);
11400 #endif
11401                 INIT_LIST_HEAD(&sd->poll_list);
11402                 sd->output_queue_tailp = &sd->output_queue;
11403 #ifdef CONFIG_RPS
11404                 INIT_CSD(&sd->csd, rps_trigger_softirq, sd);
11405                 sd->cpu = i;
11406 #endif
11407                 INIT_CSD(&sd->defer_csd, trigger_rx_softirq, sd);
11408                 spin_lock_init(&sd->defer_lock);
11409
11410                 init_gro_hash(&sd->backlog);
11411                 sd->backlog.poll = process_backlog;
11412                 sd->backlog.weight = weight_p;
11413         }
11414
11415         dev_boot_phase = 0;
11416
11417         /* The loopback device is special if any other network devices
11418          * is present in a network namespace the loopback device must
11419          * be present. Since we now dynamically allocate and free the
11420          * loopback device ensure this invariant is maintained by
11421          * keeping the loopback device as the first device on the
11422          * list of network devices.  Ensuring the loopback devices
11423          * is the first device that appears and the last network device
11424          * that disappears.
11425          */
11426         if (register_pernet_device(&loopback_net_ops))
11427                 goto out;
11428
11429         if (register_pernet_device(&default_device_ops))
11430                 goto out;
11431
11432         open_softirq(NET_TX_SOFTIRQ, net_tx_action);
11433         open_softirq(NET_RX_SOFTIRQ, net_rx_action);
11434
11435         rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead",
11436                                        NULL, dev_cpu_dead);
11437         WARN_ON(rc < 0);
11438         rc = 0;
11439 out:
11440         return rc;
11441 }
11442
11443 subsys_initcall(net_dev_init);