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_extack(unsigned long val,
 164                                            struct net_device *dev,
 165                                            struct netlink_ext_ack *extack);
 166 static struct napi_struct *napi_by_id(unsigned int napi_id);
 167
 168 /*
 169  * The @dev_base_head list is protected by @dev_base_lock and the rtnl
 170  * semaphore.
 171  *
 172  * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
 173  *
 174  * Writers must hold the rtnl semaphore while they loop through the
 175  * dev_base_head list, and hold dev_base_lock for writing when they do the
 176  * actual updates.  This allows pure readers to access the list even
 177  * while a writer is preparing to update it.
 178  *
 179  * To put it another way, dev_base_lock is held for writing only to
 180  * protect against pure readers; the rtnl semaphore provides the
 181  * protection against other writers.
 182  *
 183  * See, for example usages, register_netdevice() and
 184  * unregister_netdevice(), which must be called with the rtnl
 185  * semaphore held.
 186  */
 187 DEFINE_RWLOCK(dev_base_lock);
 188 EXPORT_SYMBOL(dev_base_lock);
 189
 190 static DEFINE_MUTEX(ifalias_mutex);
 191
 192 /* protects napi_hash addition/deletion and napi_gen_id */
 193 static DEFINE_SPINLOCK(napi_hash_lock);
 194
 195 static unsigned int napi_gen_id = NR_CPUS;
 196 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8);
 197
 198 static DECLARE_RWSEM(devnet_rename_sem);
 199
 200 static inline void dev_base_seq_inc(struct net *net)
 201 {
 202         while (++net->dev_base_seq == 0)
 203                 ;
 204 }
 205
 206 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
 207 {
 208         unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ));
 209
 210         return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
 211 }
 212
 213 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
 214 {
 215         return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
 216 }
 217
 218 static inline void rps_lock_irqsave(struct softnet_data *sd,
 219                                     unsigned long *flags)
 220 {
 221         if (IS_ENABLED(CONFIG_RPS))
 222                 spin_lock_irqsave(&sd->input_pkt_queue.lock, *flags);
 223         else if (!IS_ENABLED(CONFIG_PREEMPT_RT))
 224                 local_irq_save(*flags);
 225 }
 226
 227 static inline void rps_lock_irq_disable(struct softnet_data *sd)
 228 {
 229         if (IS_ENABLED(CONFIG_RPS))
 230                 spin_lock_irq(&sd->input_pkt_queue.lock);
 231         else if (!IS_ENABLED(CONFIG_PREEMPT_RT))
 232                 local_irq_disable();
 233 }
 234
 235 static inline void rps_unlock_irq_restore(struct softnet_data *sd,
 236                                           unsigned long *flags)
 237 {
 238         if (IS_ENABLED(CONFIG_RPS))
 239                 spin_unlock_irqrestore(&sd->input_pkt_queue.lock, *flags);
 240         else if (!IS_ENABLED(CONFIG_PREEMPT_RT))
 241                 local_irq_restore(*flags);
 242 }
 243
 244 static inline void rps_unlock_irq_enable(struct softnet_data *sd)
 245 {
 246         if (IS_ENABLED(CONFIG_RPS))
 247                 spin_unlock_irq(&sd->input_pkt_queue.lock);
 248         else if (!IS_ENABLED(CONFIG_PREEMPT_RT))
 249                 local_irq_enable();
 250 }
 251
 252 static struct netdev_name_node *netdev_name_node_alloc(struct net_device *dev,
 253                                                        const char *name)
 254 {
 255         struct netdev_name_node *name_node;
 256
 257         name_node = kmalloc(sizeof(*name_node), GFP_KERNEL);
 258         if (!name_node)
 259                 return NULL;
 260         INIT_HLIST_NODE(&name_node->hlist);
 261         name_node->dev = dev;
 262         name_node->name = name;
 263         return name_node;
 264 }
 265
 266 static struct netdev_name_node *
 267 netdev_name_node_head_alloc(struct net_device *dev)
 268 {
 269         struct netdev_name_node *name_node;
 270
 271         name_node = netdev_name_node_alloc(dev, dev->name);
 272         if (!name_node)
 273                 return NULL;
 274         INIT_LIST_HEAD(&name_node->list);
 275         return name_node;
 276 }
 277
 278 static void netdev_name_node_free(struct netdev_name_node *name_node)
 279 {
 280         kfree(name_node);
 281 }
 282
 283 static void netdev_name_node_add(struct net *net,
 284                                  struct netdev_name_node *name_node)
 285 {
 286         hlist_add_head_rcu(&name_node->hlist,
 287                            dev_name_hash(net, name_node->name));
 288 }
 289
 290 static void netdev_name_node_del(struct netdev_name_node *name_node)
 291 {
 292         hlist_del_rcu(&name_node->hlist);
 293 }
 294
 295 static struct netdev_name_node *netdev_name_node_lookup(struct net *net,
 296                                                         const char *name)
 297 {
 298         struct hlist_head *head = dev_name_hash(net, name);
 299         struct netdev_name_node *name_node;
 300
 301         hlist_for_each_entry(name_node, head, hlist)
 302                 if (!strcmp(name_node->name, name))
 303                         return name_node;
 304         return NULL;
 305 }
 306
 307 static struct netdev_name_node *netdev_name_node_lookup_rcu(struct net *net,
 308                                                             const char *name)
 309 {
 310         struct hlist_head *head = dev_name_hash(net, name);
 311         struct netdev_name_node *name_node;
 312
 313         hlist_for_each_entry_rcu(name_node, head, hlist)
 314                 if (!strcmp(name_node->name, name))
 315                         return name_node;
 316         return NULL;
 317 }
 318
 319 bool netdev_name_in_use(struct net *net, const char *name)
 320 {
 321         return netdev_name_node_lookup(net, name);
 322 }
 323 EXPORT_SYMBOL(netdev_name_in_use);
 324
 325 int netdev_name_node_alt_create(struct net_device *dev, const char *name)
 326 {
 327         struct netdev_name_node *name_node;
 328         struct net *net = dev_net(dev);
 329
 330         name_node = netdev_name_node_lookup(net, name);
 331         if (name_node)
 332                 return -EEXIST;
 333         name_node = netdev_name_node_alloc(dev, name);
 334         if (!name_node)
 335                 return -ENOMEM;
 336         netdev_name_node_add(net, name_node);
 337         /* The node that holds dev->name acts as a head of per-device list. */
 338         list_add_tail(&name_node->list, &dev->name_node->list);
 339
 340         return 0;
 341 }
 342
 343 static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node)
 344 {
 345         list_del(&name_node->list);
 346         netdev_name_node_del(name_node);
 347         kfree(name_node->name);
 348         netdev_name_node_free(name_node);
 349 }
 350
 351 int netdev_name_node_alt_destroy(struct net_device *dev, const char *name)
 352 {
 353         struct netdev_name_node *name_node;
 354         struct net *net = dev_net(dev);
 355
 356         name_node = netdev_name_node_lookup(net, name);
 357         if (!name_node)
 358                 return -ENOENT;
 359         /* lookup might have found our primary name or a name belonging
 360          * to another device.
 361          */
 362         if (name_node == dev->name_node || name_node->dev != dev)
 363                 return -EINVAL;
 364
 365         __netdev_name_node_alt_destroy(name_node);
 366
 367         return 0;
 368 }
 369
 370 static void netdev_name_node_alt_flush(struct net_device *dev)
 371 {
 372         struct netdev_name_node *name_node, *tmp;
 373
 374         list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list)
 375                 __netdev_name_node_alt_destroy(name_node);
 376 }
 377
 378 /* Device list insertion */
 379 static void list_netdevice(struct net_device *dev)
 380 {
 381         struct net *net = dev_net(dev);
 382
 383         ASSERT_RTNL();
 384
 385         write_lock(&dev_base_lock);
 386         list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
 387         netdev_name_node_add(net, dev->name_node);
 388         hlist_add_head_rcu(&dev->index_hlist,
 389                            dev_index_hash(net, dev->ifindex));
 390         write_unlock(&dev_base_lock);
 391
 392         dev_base_seq_inc(net);
 393 }
 394
 395 /* Device list removal
 396  * caller must respect a RCU grace period before freeing/reusing dev
 397  */
 398 static void unlist_netdevice(struct net_device *dev, bool lock)
 399 {
 400         ASSERT_RTNL();
 401
 402         /* Unlink dev from the device chain */
 403         if (lock)
 404                 write_lock(&dev_base_lock);
 405         list_del_rcu(&dev->dev_list);
 406         netdev_name_node_del(dev->name_node);
 407         hlist_del_rcu(&dev->index_hlist);
 408         if (lock)
 409                 write_unlock(&dev_base_lock);
 410
 411         dev_base_seq_inc(dev_net(dev));
 412 }
 413
 414 /*
 415  *      Our notifier list
 416  */
 417
 418 static RAW_NOTIFIER_HEAD(netdev_chain);
 419
 420 /*
 421  *      Device drivers call our routines to queue packets here. We empty the
 422  *      queue in the local softnet handler.
 423  */
 424
 425 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
 426 EXPORT_PER_CPU_SYMBOL(softnet_data);
 427
 428 #ifdef CONFIG_LOCKDEP
 429 /*
 430  * register_netdevice() inits txq->_xmit_lock and sets lockdep class
 431  * according to dev->type
 432  */
 433 static const unsigned short netdev_lock_type[] = {
 434          ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
 435          ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
 436          ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
 437          ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
 438          ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
 439          ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
 440          ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
 441          ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
 442          ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
 443          ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
 444          ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
 445          ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
 446          ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
 447          ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
 448          ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
 449
 450 static const char *const netdev_lock_name[] = {
 451         "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
 452         "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
 453         "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
 454         "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
 455         "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
 456         "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
 457         "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
 458         "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
 459         "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
 460         "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
 461         "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
 462         "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
 463         "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
 464         "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
 465         "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
 466
 467 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
 468 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
 469
 470 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
 471 {
 472         int i;
 473
 474         for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
 475                 if (netdev_lock_type[i] == dev_type)
 476                         return i;
 477         /* the last key is used by default */
 478         return ARRAY_SIZE(netdev_lock_type) - 1;
 479 }
 480
 481 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
 482                                                  unsigned short dev_type)
 483 {
 484         int i;
 485
 486         i = netdev_lock_pos(dev_type);
 487         lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
 488                                    netdev_lock_name[i]);
 489 }
 490
 491 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
 492 {
 493         int i;
 494
 495         i = netdev_lock_pos(dev->type);
 496         lockdep_set_class_and_name(&dev->addr_list_lock,
 497                                    &netdev_addr_lock_key[i],
 498                                    netdev_lock_name[i]);
 499 }
 500 #else
 501 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
 502                                                  unsigned short dev_type)
 503 {
 504 }
 505
 506 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
 507 {
 508 }
 509 #endif
 510
 511 /*******************************************************************************
 512  *
 513  *              Protocol management and registration routines
 514  *
 515  *******************************************************************************/
 516
 517
 518 /*
 519  *      Add a protocol ID to the list. Now that the input handler is
 520  *      smarter we can dispense with all the messy stuff that used to be
 521  *      here.
 522  *
 523  *      BEWARE!!! Protocol handlers, mangling input packets,
 524  *      MUST BE last in hash buckets and checking protocol handlers
 525  *      MUST start from promiscuous ptype_all chain in net_bh.
 526  *      It is true now, do not change it.
 527  *      Explanation follows: if protocol handler, mangling packet, will
 528  *      be the first on list, it is not able to sense, that packet
 529  *      is cloned and should be copied-on-write, so that it will
 530  *      change it and subsequent readers will get broken packet.
 531  *                                                      --ANK (980803)
 532  */
 533
 534 static inline struct list_head *ptype_head(const struct packet_type *pt)
 535 {
 536         if (pt->type == htons(ETH_P_ALL))
 537                 return pt->dev ? &pt->dev->ptype_all : &ptype_all;
 538         else
 539                 return pt->dev ? &pt->dev->ptype_specific :
 540                                  &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
 541 }
 542
 543 /**
 544  *      dev_add_pack - add packet handler
 545  *      @pt: packet type declaration
 546  *
 547  *      Add a protocol handler to the networking stack. The passed &packet_type
 548  *      is linked into kernel lists and may not be freed until it has been
 549  *      removed from the kernel lists.
 550  *
 551  *      This call does not sleep therefore it can not
 552  *      guarantee all CPU's that are in middle of receiving packets
 553  *      will see the new packet type (until the next received packet).
 554  */
 555
 556 void dev_add_pack(struct packet_type *pt)
 557 {
 558         struct list_head *head = ptype_head(pt);
 559
 560         spin_lock(&ptype_lock);
 561         list_add_rcu(&pt->list, head);
 562         spin_unlock(&ptype_lock);
 563 }
 564 EXPORT_SYMBOL(dev_add_pack);
 565
 566 /**
 567  *      __dev_remove_pack        - remove packet handler
 568  *      @pt: packet type declaration
 569  *
 570  *      Remove a protocol handler that was previously added to the kernel
 571  *      protocol handlers by dev_add_pack(). The passed &packet_type is removed
 572  *      from the kernel lists and can be freed or reused once this function
 573  *      returns.
 574  *
 575  *      The packet type might still be in use by receivers
 576  *      and must not be freed until after all the CPU's have gone
 577  *      through a quiescent state.
 578  */
 579 void __dev_remove_pack(struct packet_type *pt)
 580 {
 581         struct list_head *head = ptype_head(pt);
 582         struct packet_type *pt1;
 583
 584         spin_lock(&ptype_lock);
 585
 586         list_for_each_entry(pt1, head, list) {
 587                 if (pt == pt1) {
 588                         list_del_rcu(&pt->list);
 589                         goto out;
 590                 }
 591         }
 592
 593         pr_warn("dev_remove_pack: %p not found\n", pt);
 594 out:
 595         spin_unlock(&ptype_lock);
 596 }
 597 EXPORT_SYMBOL(__dev_remove_pack);
 598
 599 /**
 600  *      dev_remove_pack  - remove packet handler
 601  *      @pt: packet type declaration
 602  *
 603  *      Remove a protocol handler that was previously added to the kernel
 604  *      protocol handlers by dev_add_pack(). The passed &packet_type is removed
 605  *      from the kernel lists and can be freed or reused once this function
 606  *      returns.
 607  *
 608  *      This call sleeps to guarantee that no CPU is looking at the packet
 609  *      type after return.
 610  */
 611 void dev_remove_pack(struct packet_type *pt)
 612 {
 613         __dev_remove_pack(pt);
 614
 615         synchronize_net();
 616 }
 617 EXPORT_SYMBOL(dev_remove_pack);
 618
 619
 620 /*******************************************************************************
 621  *
 622  *                          Device Interface Subroutines
 623  *
 624  *******************************************************************************/
 625
 626 /**
 627  *      dev_get_iflink  - get 'iflink' value of a interface
 628  *      @dev: targeted interface
 629  *
 630  *      Indicates the ifindex the interface is linked to.
 631  *      Physical interfaces have the same 'ifindex' and 'iflink' values.
 632  */
 633
 634 int dev_get_iflink(const struct net_device *dev)
 635 {
 636         if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink)
 637                 return dev->netdev_ops->ndo_get_iflink(dev);
 638
 639         return dev->ifindex;
 640 }
 641 EXPORT_SYMBOL(dev_get_iflink);
 642
 643 /**
 644  *      dev_fill_metadata_dst - Retrieve tunnel egress information.
 645  *      @dev: targeted interface
 646  *      @skb: The packet.
 647  *
 648  *      For better visibility of tunnel traffic OVS needs to retrieve
 649  *      egress tunnel information for a packet. Following API allows
 650  *      user to get this info.
 651  */
 652 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
 653 {
 654         struct ip_tunnel_info *info;
 655
 656         if (!dev->netdev_ops  || !dev->netdev_ops->ndo_fill_metadata_dst)
 657                 return -EINVAL;
 658
 659         info = skb_tunnel_info_unclone(skb);
 660         if (!info)
 661                 return -ENOMEM;
 662         if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX)))
 663                 return -EINVAL;
 664
 665         return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb);
 666 }
 667 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst);
 668
 669 static struct net_device_path *dev_fwd_path(struct net_device_path_stack *stack)
 670 {
 671         int k = stack->num_paths++;
 672
 673         if (WARN_ON_ONCE(k >= NET_DEVICE_PATH_STACK_MAX))
 674                 return NULL;
 675
 676         return &stack->path[k];
 677 }
 678
 679 int dev_fill_forward_path(const struct net_device *dev, const u8 *daddr,
 680                           struct net_device_path_stack *stack)
 681 {
 682         const struct net_device *last_dev;
 683         struct net_device_path_ctx ctx = {
 684                 .dev    = dev,
 685         };
 686         struct net_device_path *path;
 687         int ret = 0;
 688
 689         memcpy(ctx.daddr, daddr, sizeof(ctx.daddr));
 690         stack->num_paths = 0;
 691         while (ctx.dev && ctx.dev->netdev_ops->ndo_fill_forward_path) {
 692                 last_dev = ctx.dev;
 693                 path = dev_fwd_path(stack);
 694                 if (!path)
 695                         return -1;
 696
 697                 memset(path, 0, sizeof(struct net_device_path));
 698                 ret = ctx.dev->netdev_ops->ndo_fill_forward_path(&ctx, path);
 699                 if (ret < 0)
 700                         return -1;
 701
 702                 if (WARN_ON_ONCE(last_dev == ctx.dev))
 703                         return -1;
 704         }
 705
 706         if (!ctx.dev)
 707                 return ret;
 708
 709         path = dev_fwd_path(stack);
 710         if (!path)
 711                 return -1;
 712         path->type = DEV_PATH_ETHERNET;
 713         path->dev = ctx.dev;
 714
 715         return ret;
 716 }
 717 EXPORT_SYMBOL_GPL(dev_fill_forward_path);
 718
 719 /**
 720  *      __dev_get_by_name       - find a device by its name
 721  *      @net: the applicable net namespace
 722  *      @name: name to find
 723  *
 724  *      Find an interface by name. Must be called under RTNL semaphore
 725  *      or @dev_base_lock. If the name is found a pointer to the device
 726  *      is returned. If the name is not found then %NULL is returned. The
 727  *      reference counters are not incremented so the caller must be
 728  *      careful with locks.
 729  */
 730
 731 struct net_device *__dev_get_by_name(struct net *net, const char *name)
 732 {
 733         struct netdev_name_node *node_name;
 734
 735         node_name = netdev_name_node_lookup(net, name);
 736         return node_name ? node_name->dev : NULL;
 737 }
 738 EXPORT_SYMBOL(__dev_get_by_name);
 739
 740 /**
 741  * dev_get_by_name_rcu  - find a device by its name
 742  * @net: the applicable net namespace
 743  * @name: name to find
 744  *
 745  * Find an interface by name.
 746  * If the name is found a pointer to the device is returned.
 747  * If the name is not found then %NULL is returned.
 748  * The reference counters are not incremented so the caller must be
 749  * careful with locks. The caller must hold RCU lock.
 750  */
 751
 752 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
 753 {
 754         struct netdev_name_node *node_name;
 755
 756         node_name = netdev_name_node_lookup_rcu(net, name);
 757         return node_name ? node_name->dev : NULL;
 758 }
 759 EXPORT_SYMBOL(dev_get_by_name_rcu);
 760
 761 /**
 762  *      dev_get_by_name         - find a device by its name
 763  *      @net: the applicable net namespace
 764  *      @name: name to find
 765  *
 766  *      Find an interface by name. This can be called from any
 767  *      context and does its own locking. The returned handle has
 768  *      the usage count incremented and the caller must use dev_put() to
 769  *      release it when it is no longer needed. %NULL is returned if no
 770  *      matching device is found.
 771  */
 772
 773 struct net_device *dev_get_by_name(struct net *net, const char *name)
 774 {
 775         struct net_device *dev;
 776
 777         rcu_read_lock();
 778         dev = dev_get_by_name_rcu(net, name);
 779         dev_hold(dev);
 780         rcu_read_unlock();
 781         return dev;
 782 }
 783 EXPORT_SYMBOL(dev_get_by_name);
 784
 785 /**
 786  *      __dev_get_by_index - find a device by its ifindex
 787  *      @net: the applicable net namespace
 788  *      @ifindex: index of device
 789  *
 790  *      Search for an interface by index. Returns %NULL if the device
 791  *      is not found or a pointer to the device. The device has not
 792  *      had its reference counter increased so the caller must be careful
 793  *      about locking. The caller must hold either the RTNL semaphore
 794  *      or @dev_base_lock.
 795  */
 796
 797 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
 798 {
 799         struct net_device *dev;
 800         struct hlist_head *head = dev_index_hash(net, ifindex);
 801
 802         hlist_for_each_entry(dev, head, index_hlist)
 803                 if (dev->ifindex == ifindex)
 804                         return dev;
 805
 806         return NULL;
 807 }
 808 EXPORT_SYMBOL(__dev_get_by_index);
 809
 810 /**
 811  *      dev_get_by_index_rcu - find a device by its ifindex
 812  *      @net: the applicable net namespace
 813  *      @ifindex: index of device
 814  *
 815  *      Search for an interface by index. Returns %NULL if the device
 816  *      is not found or a pointer to the device. The device has not
 817  *      had its reference counter increased so the caller must be careful
 818  *      about locking. The caller must hold RCU lock.
 819  */
 820
 821 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
 822 {
 823         struct net_device *dev;
 824         struct hlist_head *head = dev_index_hash(net, ifindex);
 825
 826         hlist_for_each_entry_rcu(dev, head, index_hlist)
 827                 if (dev->ifindex == ifindex)
 828                         return dev;
 829
 830         return NULL;
 831 }
 832 EXPORT_SYMBOL(dev_get_by_index_rcu);
 833
 834
 835 /**
 836  *      dev_get_by_index - find a device by its ifindex
 837  *      @net: the applicable net namespace
 838  *      @ifindex: index of device
 839  *
 840  *      Search for an interface by index. Returns NULL if the device
 841  *      is not found or a pointer to the device. The device returned has
 842  *      had a reference added and the pointer is safe until the user calls
 843  *      dev_put to indicate they have finished with it.
 844  */
 845
 846 struct net_device *dev_get_by_index(struct net *net, int ifindex)
 847 {
 848         struct net_device *dev;
 849
 850         rcu_read_lock();
 851         dev = dev_get_by_index_rcu(net, ifindex);
 852         dev_hold(dev);
 853         rcu_read_unlock();
 854         return dev;
 855 }
 856 EXPORT_SYMBOL(dev_get_by_index);
 857
 858 /**
 859  *      dev_get_by_napi_id - find a device by napi_id
 860  *      @napi_id: ID of the NAPI struct
 861  *
 862  *      Search for an interface by NAPI ID. Returns %NULL if the device
 863  *      is not found or a pointer to the device. The device has not had
 864  *      its reference counter increased so the caller must be careful
 865  *      about locking. The caller must hold RCU lock.
 866  */
 867
 868 struct net_device *dev_get_by_napi_id(unsigned int napi_id)
 869 {
 870         struct napi_struct *napi;
 871
 872         WARN_ON_ONCE(!rcu_read_lock_held());
 873
 874         if (napi_id < MIN_NAPI_ID)
 875                 return NULL;
 876
 877         napi = napi_by_id(napi_id);
 878
 879         return napi ? napi->dev : NULL;
 880 }
 881 EXPORT_SYMBOL(dev_get_by_napi_id);
 882
 883 /**
 884  *      netdev_get_name - get a netdevice name, knowing its ifindex.
 885  *      @net: network namespace
 886  *      @name: a pointer to the buffer where the name will be stored.
 887  *      @ifindex: the ifindex of the interface to get the name from.
 888  */
 889 int netdev_get_name(struct net *net, char *name, int ifindex)
 890 {
 891         struct net_device *dev;
 892         int ret;
 893
 894         down_read(&devnet_rename_sem);
 895         rcu_read_lock();
 896
 897         dev = dev_get_by_index_rcu(net, ifindex);
 898         if (!dev) {
 899                 ret = -ENODEV;
 900                 goto out;
 901         }
 902
 903         strcpy(name, dev->name);
 904
 905         ret = 0;
 906 out:
 907         rcu_read_unlock();
 908         up_read(&devnet_rename_sem);
 909         return ret;
 910 }
 911
 912 /**
 913  *      dev_getbyhwaddr_rcu - find a device by its hardware address
 914  *      @net: the applicable net namespace
 915  *      @type: media type of device
 916  *      @ha: hardware address
 917  *
 918  *      Search for an interface by MAC address. Returns NULL if the device
 919  *      is not found or a pointer to the device.
 920  *      The caller must hold RCU or RTNL.
 921  *      The returned device has not had its ref count increased
 922  *      and the caller must therefore be careful about locking
 923  *
 924  */
 925
 926 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
 927                                        const char *ha)
 928 {
 929         struct net_device *dev;
 930
 931         for_each_netdev_rcu(net, dev)
 932                 if (dev->type == type &&
 933                     !memcmp(dev->dev_addr, ha, dev->addr_len))
 934                         return dev;
 935
 936         return NULL;
 937 }
 938 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
 939
 940 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
 941 {
 942         struct net_device *dev, *ret = NULL;
 943
 944         rcu_read_lock();
 945         for_each_netdev_rcu(net, dev)
 946                 if (dev->type == type) {
 947                         dev_hold(dev);
 948                         ret = dev;
 949                         break;
 950                 }
 951         rcu_read_unlock();
 952         return ret;
 953 }
 954 EXPORT_SYMBOL(dev_getfirstbyhwtype);
 955
 956 /**
 957  *      __dev_get_by_flags - find any device with given flags
 958  *      @net: the applicable net namespace
 959  *      @if_flags: IFF_* values
 960  *      @mask: bitmask of bits in if_flags to check
 961  *
 962  *      Search for any interface with the given flags. Returns NULL if a device
 963  *      is not found or a pointer to the device. Must be called inside
 964  *      rtnl_lock(), and result refcount is unchanged.
 965  */
 966
 967 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
 968                                       unsigned short mask)
 969 {
 970         struct net_device *dev, *ret;
 971
 972         ASSERT_RTNL();
 973
 974         ret = NULL;
 975         for_each_netdev(net, dev) {
 976                 if (((dev->flags ^ if_flags) & mask) == 0) {
 977                         ret = dev;
 978                         break;
 979                 }
 980         }
 981         return ret;
 982 }
 983 EXPORT_SYMBOL(__dev_get_by_flags);
 984
 985 /**
 986  *      dev_valid_name - check if name is okay for network device
 987  *      @name: name string
 988  *
 989  *      Network device names need to be valid file names to
 990  *      allow sysfs to work.  We also disallow any kind of
 991  *      whitespace.
 992  */
 993 bool dev_valid_name(const char *name)
 994 {
 995         if (*name == '\0')
 996                 return false;
 997         if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
 998                 return false;
 999         if (!strcmp(name, ".") || !strcmp(name, ".."))
1000                 return false;
1001
1002         while (*name) {
1003                 if (*name == '/' || *name == ':' || isspace(*name))
1004                         return false;
1005                 name++;
1006         }
1007         return true;
1008 }
1009 EXPORT_SYMBOL(dev_valid_name);
1010
1011 /**
1012  *      __dev_alloc_name - allocate a name for a device
1013  *      @net: network namespace to allocate the device name in
1014  *      @name: name format string
1015  *      @buf:  scratch buffer and result name string
1016  *
1017  *      Passed a format string - eg "lt%d" it will try and find a suitable
1018  *      id. It scans list of devices to build up a free map, then chooses
1019  *      the first empty slot. The caller must hold the dev_base or rtnl lock
1020  *      while allocating the name and adding the device in order to avoid
1021  *      duplicates.
1022  *      Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1023  *      Returns the number of the unit assigned or a negative errno code.
1024  */
1025
1026 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
1027 {
1028         int i = 0;
1029         const char *p;
1030         const int max_netdevices = 8*PAGE_SIZE;
1031         unsigned long *inuse;
1032         struct net_device *d;
1033
1034         if (!dev_valid_name(name))
1035                 return -EINVAL;
1036
1037         p = strchr(name, '%');
1038         if (p) {
1039                 /*
1040                  * Verify the string as this thing may have come from
1041                  * the user.  There must be either one "%d" and no other "%"
1042                  * characters.
1043                  */
1044                 if (p[1] != 'd' || strchr(p + 2, '%'))
1045                         return -EINVAL;
1046
1047                 /* Use one page as a bit array of possible slots */
1048                 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
1049                 if (!inuse)
1050                         return -ENOMEM;
1051
1052                 for_each_netdev(net, d) {
1053                         struct netdev_name_node *name_node;
1054                         list_for_each_entry(name_node, &d->name_node->list, list) {
1055                                 if (!sscanf(name_node->name, name, &i))
1056                                         continue;
1057                                 if (i < 0 || i >= max_netdevices)
1058                                         continue;
1059
1060                                 /*  avoid cases where sscanf is not exact inverse of printf */
1061                                 snprintf(buf, IFNAMSIZ, name, i);
1062                                 if (!strncmp(buf, name_node->name, IFNAMSIZ))
1063                                         __set_bit(i, inuse);
1064                         }
1065                         if (!sscanf(d->name, name, &i))
1066                                 continue;
1067                         if (i < 0 || i >= max_netdevices)
1068                                 continue;
1069
1070                         /*  avoid cases where sscanf is not exact inverse of printf */
1071                         snprintf(buf, IFNAMSIZ, name, i);
1072                         if (!strncmp(buf, d->name, IFNAMSIZ))
1073                                 __set_bit(i, inuse);
1074                 }
1075
1076                 i = find_first_zero_bit(inuse, max_netdevices);
1077                 free_page((unsigned long) inuse);
1078         }
1079
1080         snprintf(buf, IFNAMSIZ, name, i);
1081         if (!netdev_name_in_use(net, buf))
1082                 return i;
1083
1084         /* It is possible to run out of possible slots
1085          * when the name is long and there isn't enough space left
1086          * for the digits, or if all bits are used.
1087          */
1088         return -ENFILE;
1089 }
1090
1091 static int dev_alloc_name_ns(struct net *net,
1092                              struct net_device *dev,
1093                              const char *name)
1094 {
1095         char buf[IFNAMSIZ];
1096         int ret;
1097
1098         BUG_ON(!net);
1099         ret = __dev_alloc_name(net, name, buf);
1100         if (ret >= 0)
1101                 strscpy(dev->name, buf, IFNAMSIZ);
1102         return ret;
1103 }
1104
1105 /**
1106  *      dev_alloc_name - allocate a name for a device
1107  *      @dev: device
1108  *      @name: name format string
1109  *
1110  *      Passed a format string - eg "lt%d" it will try and find a suitable
1111  *      id. It scans list of devices to build up a free map, then chooses
1112  *      the first empty slot. The caller must hold the dev_base or rtnl lock
1113  *      while allocating the name and adding the device in order to avoid
1114  *      duplicates.
1115  *      Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1116  *      Returns the number of the unit assigned or a negative errno code.
1117  */
1118
1119 int dev_alloc_name(struct net_device *dev, const char *name)
1120 {
1121         return dev_alloc_name_ns(dev_net(dev), dev, name);
1122 }
1123 EXPORT_SYMBOL(dev_alloc_name);
1124
1125 static int dev_get_valid_name(struct net *net, struct net_device *dev,
1126                               const char *name)
1127 {
1128         BUG_ON(!net);
1129
1130         if (!dev_valid_name(name))
1131                 return -EINVAL;
1132
1133         if (strchr(name, '%'))
1134                 return dev_alloc_name_ns(net, dev, name);
1135         else if (netdev_name_in_use(net, name))
1136                 return -EEXIST;
1137         else if (dev->name != name)
1138                 strscpy(dev->name, name, IFNAMSIZ);
1139
1140         return 0;
1141 }
1142
1143 /**
1144  *      dev_change_name - change name of a device
1145  *      @dev: device
1146  *      @newname: name (or format string) must be at least IFNAMSIZ
1147  *
1148  *      Change name of a device, can pass format strings "eth%d".
1149  *      for wildcarding.
1150  */
1151 int dev_change_name(struct net_device *dev, const char *newname)
1152 {
1153         unsigned char old_assign_type;
1154         char oldname[IFNAMSIZ];
1155         int err = 0;
1156         int ret;
1157         struct net *net;
1158
1159         ASSERT_RTNL();
1160         BUG_ON(!dev_net(dev));
1161
1162         net = dev_net(dev);
1163
1164         down_write(&devnet_rename_sem);
1165
1166         if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1167                 up_write(&devnet_rename_sem);
1168                 return 0;
1169         }
1170
1171         memcpy(oldname, dev->name, IFNAMSIZ);
1172
1173         err = dev_get_valid_name(net, dev, newname);
1174         if (err < 0) {
1175                 up_write(&devnet_rename_sem);
1176                 return err;
1177         }
1178
1179         if (oldname[0] && !strchr(oldname, '%'))
1180                 netdev_info(dev, "renamed from %s%s\n", oldname,
1181                             dev->flags & IFF_UP ? " (while UP)" : "");
1182
1183         old_assign_type = dev->name_assign_type;
1184         dev->name_assign_type = NET_NAME_RENAMED;
1185
1186 rollback:
1187         ret = device_rename(&dev->dev, dev->name);
1188         if (ret) {
1189                 memcpy(dev->name, oldname, IFNAMSIZ);
1190                 dev->name_assign_type = old_assign_type;
1191                 up_write(&devnet_rename_sem);
1192                 return ret;
1193         }
1194
1195         up_write(&devnet_rename_sem);
1196
1197         netdev_adjacent_rename_links(dev, oldname);
1198
1199         write_lock(&dev_base_lock);
1200         netdev_name_node_del(dev->name_node);
1201         write_unlock(&dev_base_lock);
1202
1203         synchronize_rcu();
1204
1205         write_lock(&dev_base_lock);
1206         netdev_name_node_add(net, dev->name_node);
1207         write_unlock(&dev_base_lock);
1208
1209         ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1210         ret = notifier_to_errno(ret);
1211
1212         if (ret) {
1213                 /* err >= 0 after dev_alloc_name() or stores the first errno */
1214                 if (err >= 0) {
1215                         err = ret;
1216                         down_write(&devnet_rename_sem);
1217                         memcpy(dev->name, oldname, IFNAMSIZ);
1218                         memcpy(oldname, newname, IFNAMSIZ);
1219                         dev->name_assign_type = old_assign_type;
1220                         old_assign_type = NET_NAME_RENAMED;
1221                         goto rollback;
1222                 } else {
1223                         netdev_err(dev, "name change rollback failed: %d\n",
1224                                    ret);
1225                 }
1226         }
1227
1228         return err;
1229 }
1230
1231 /**
1232  *      dev_set_alias - change ifalias of a device
1233  *      @dev: device
1234  *      @alias: name up to IFALIASZ
1235  *      @len: limit of bytes to copy from info
1236  *
1237  *      Set ifalias for a device,
1238  */
1239 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1240 {
1241         struct dev_ifalias *new_alias = NULL;
1242
1243         if (len >= IFALIASZ)
1244                 return -EINVAL;
1245
1246         if (len) {
1247                 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL);
1248                 if (!new_alias)
1249                         return -ENOMEM;
1250
1251                 memcpy(new_alias->ifalias, alias, len);
1252                 new_alias->ifalias[len] = 0;
1253         }
1254
1255         mutex_lock(&ifalias_mutex);
1256         new_alias = rcu_replace_pointer(dev->ifalias, new_alias,
1257                                         mutex_is_locked(&ifalias_mutex));
1258         mutex_unlock(&ifalias_mutex);
1259
1260         if (new_alias)
1261                 kfree_rcu(new_alias, rcuhead);
1262
1263         return len;
1264 }
1265 EXPORT_SYMBOL(dev_set_alias);
1266
1267 /**
1268  *      dev_get_alias - get ifalias of a device
1269  *      @dev: device
1270  *      @name: buffer to store name of ifalias
1271  *      @len: size of buffer
1272  *
1273  *      get ifalias for a device.  Caller must make sure dev cannot go
1274  *      away,  e.g. rcu read lock or own a reference count to device.
1275  */
1276 int dev_get_alias(const struct net_device *dev, char *name, size_t len)
1277 {
1278         const struct dev_ifalias *alias;
1279         int ret = 0;
1280
1281         rcu_read_lock();
1282         alias = rcu_dereference(dev->ifalias);
1283         if (alias)
1284                 ret = snprintf(name, len, "%s", alias->ifalias);
1285         rcu_read_unlock();
1286
1287         return ret;
1288 }
1289
1290 /**
1291  *      netdev_features_change - device changes features
1292  *      @dev: device to cause notification
1293  *
1294  *      Called to indicate a device has changed features.
1295  */
1296 void netdev_features_change(struct net_device *dev)
1297 {
1298         call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1299 }
1300 EXPORT_SYMBOL(netdev_features_change);
1301
1302 /**
1303  *      netdev_state_change - device changes state
1304  *      @dev: device to cause notification
1305  *
1306  *      Called to indicate a device has changed state. This function calls
1307  *      the notifier chains for netdev_chain and sends a NEWLINK message
1308  *      to the routing socket.
1309  */
1310 void netdev_state_change(struct net_device *dev)
1311 {
1312         if (dev->flags & IFF_UP) {
1313                 struct netdev_notifier_change_info change_info = {
1314                         .info.dev = dev,
1315                 };
1316
1317                 call_netdevice_notifiers_info(NETDEV_CHANGE,
1318                                               &change_info.info);
1319                 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL, 0, NULL);
1320         }
1321 }
1322 EXPORT_SYMBOL(netdev_state_change);
1323
1324 /**
1325  * __netdev_notify_peers - notify network peers about existence of @dev,
1326  * to be called when rtnl lock is already held.
1327  * @dev: network device
1328  *
1329  * Generate traffic such that interested network peers are aware of
1330  * @dev, such as by generating a gratuitous ARP. This may be used when
1331  * a device wants to inform the rest of the network about some sort of
1332  * reconfiguration such as a failover event or virtual machine
1333  * migration.
1334  */
1335 void __netdev_notify_peers(struct net_device *dev)
1336 {
1337         ASSERT_RTNL();
1338         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1339         call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev);
1340 }
1341 EXPORT_SYMBOL(__netdev_notify_peers);
1342
1343 /**
1344  * netdev_notify_peers - notify network peers about existence of @dev
1345  * @dev: network device
1346  *
1347  * Generate traffic such that interested network peers are aware of
1348  * @dev, such as by generating a gratuitous ARP. This may be used when
1349  * a device wants to inform the rest of the network about some sort of
1350  * reconfiguration such as a failover event or virtual machine
1351  * migration.
1352  */
1353 void netdev_notify_peers(struct net_device *dev)
1354 {
1355         rtnl_lock();
1356         __netdev_notify_peers(dev);
1357         rtnl_unlock();
1358 }
1359 EXPORT_SYMBOL(netdev_notify_peers);
1360
1361 static int napi_threaded_poll(void *data);
1362
1363 static int napi_kthread_create(struct napi_struct *n)
1364 {
1365         int err = 0;
1366
1367         /* Create and wake up the kthread once to put it in
1368          * TASK_INTERRUPTIBLE mode to avoid the blocked task
1369          * warning and work with loadavg.
1370          */
1371         n->thread = kthread_run(napi_threaded_poll, n, "napi/%s-%d",
1372                                 n->dev->name, n->napi_id);
1373         if (IS_ERR(n->thread)) {
1374                 err = PTR_ERR(n->thread);
1375                 pr_err("kthread_run failed with err %d\n", err);
1376                 n->thread = NULL;
1377         }
1378
1379         return err;
1380 }
1381
1382 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1383 {
1384         const struct net_device_ops *ops = dev->netdev_ops;
1385         int ret;
1386
1387         ASSERT_RTNL();
1388         dev_addr_check(dev);
1389
1390         if (!netif_device_present(dev)) {
1391                 /* may be detached because parent is runtime-suspended */
1392                 if (dev->dev.parent)
1393                         pm_runtime_resume(dev->dev.parent);
1394                 if (!netif_device_present(dev))
1395                         return -ENODEV;
1396         }
1397
1398         /* Block netpoll from trying to do any rx path servicing.
1399          * If we don't do this there is a chance ndo_poll_controller
1400          * or ndo_poll may be running while we open the device
1401          */
1402         netpoll_poll_disable(dev);
1403
1404         ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack);
1405         ret = notifier_to_errno(ret);
1406         if (ret)
1407                 return ret;
1408
1409         set_bit(__LINK_STATE_START, &dev->state);
1410
1411         if (ops->ndo_validate_addr)
1412                 ret = ops->ndo_validate_addr(dev);
1413
1414         if (!ret && ops->ndo_open)
1415                 ret = ops->ndo_open(dev);
1416
1417         netpoll_poll_enable(dev);
1418
1419         if (ret)
1420                 clear_bit(__LINK_STATE_START, &dev->state);
1421         else {
1422                 dev->flags |= IFF_UP;
1423                 dev_set_rx_mode(dev);
1424                 dev_activate(dev);
1425                 add_device_randomness(dev->dev_addr, dev->addr_len);
1426         }
1427
1428         return ret;
1429 }
1430
1431 /**
1432  *      dev_open        - prepare an interface for use.
1433  *      @dev: device to open
1434  *      @extack: netlink extended ack
1435  *
1436  *      Takes a device from down to up state. The device's private open
1437  *      function is invoked and then the multicast lists are loaded. Finally
1438  *      the device is moved into the up state and a %NETDEV_UP message is
1439  *      sent to the netdev notifier chain.
1440  *
1441  *      Calling this function on an active interface is a nop. On a failure
1442  *      a negative errno code is returned.
1443  */
1444 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1445 {
1446         int ret;
1447
1448         if (dev->flags & IFF_UP)
1449                 return 0;
1450
1451         ret = __dev_open(dev, extack);
1452         if (ret < 0)
1453                 return ret;
1454
1455         rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL);
1456         call_netdevice_notifiers(NETDEV_UP, dev);
1457
1458         return ret;
1459 }
1460 EXPORT_SYMBOL(dev_open);
1461
1462 static void __dev_close_many(struct list_head *head)
1463 {
1464         struct net_device *dev;
1465
1466         ASSERT_RTNL();
1467         might_sleep();
1468
1469         list_for_each_entry(dev, head, close_list) {
1470                 /* Temporarily disable netpoll until the interface is down */
1471                 netpoll_poll_disable(dev);
1472
1473                 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1474
1475                 clear_bit(__LINK_STATE_START, &dev->state);
1476
1477                 /* Synchronize to scheduled poll. We cannot touch poll list, it
1478                  * can be even on different cpu. So just clear netif_running().
1479                  *
1480                  * dev->stop() will invoke napi_disable() on all of it's
1481                  * napi_struct instances on this device.
1482                  */
1483                 smp_mb__after_atomic(); /* Commit netif_running(). */
1484         }
1485
1486         dev_deactivate_many(head);
1487
1488         list_for_each_entry(dev, head, close_list) {
1489                 const struct net_device_ops *ops = dev->netdev_ops;
1490
1491                 /*
1492                  *      Call the device specific close. This cannot fail.
1493                  *      Only if device is UP
1494                  *
1495                  *      We allow it to be called even after a DETACH hot-plug
1496                  *      event.
1497                  */
1498                 if (ops->ndo_stop)
1499                         ops->ndo_stop(dev);
1500
1501                 dev->flags &= ~IFF_UP;
1502                 netpoll_poll_enable(dev);
1503         }
1504 }
1505
1506 static void __dev_close(struct net_device *dev)
1507 {
1508         LIST_HEAD(single);
1509
1510         list_add(&dev->close_list, &single);
1511         __dev_close_many(&single);
1512         list_del(&single);
1513 }
1514
1515 void dev_close_many(struct list_head *head, bool unlink)
1516 {
1517         struct net_device *dev, *tmp;
1518
1519         /* Remove the devices that don't need to be closed */
1520         list_for_each_entry_safe(dev, tmp, head, close_list)
1521                 if (!(dev->flags & IFF_UP))
1522                         list_del_init(&dev->close_list);
1523
1524         __dev_close_many(head);
1525
1526         list_for_each_entry_safe(dev, tmp, head, close_list) {
1527                 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL);
1528                 call_netdevice_notifiers(NETDEV_DOWN, dev);
1529                 if (unlink)
1530                         list_del_init(&dev->close_list);
1531         }
1532 }
1533 EXPORT_SYMBOL(dev_close_many);
1534
1535 /**
1536  *      dev_close - shutdown an interface.
1537  *      @dev: device to shutdown
1538  *
1539  *      This function moves an active device into down state. A
1540  *      %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1541  *      is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1542  *      chain.
1543  */
1544 void dev_close(struct net_device *dev)
1545 {
1546         if (dev->flags & IFF_UP) {
1547                 LIST_HEAD(single);
1548
1549                 list_add(&dev->close_list, &single);
1550                 dev_close_many(&single, true);
1551                 list_del(&single);
1552         }
1553 }
1554 EXPORT_SYMBOL(dev_close);
1555
1556
1557 /**
1558  *      dev_disable_lro - disable Large Receive Offload on a device
1559  *      @dev: device
1560  *
1561  *      Disable Large Receive Offload (LRO) on a net device.  Must be
1562  *      called under RTNL.  This is needed if received packets may be
1563  *      forwarded to another interface.
1564  */
1565 void dev_disable_lro(struct net_device *dev)
1566 {
1567         struct net_device *lower_dev;
1568         struct list_head *iter;
1569
1570         dev->wanted_features &= ~NETIF_F_LRO;
1571         netdev_update_features(dev);
1572
1573         if (unlikely(dev->features & NETIF_F_LRO))
1574                 netdev_WARN(dev, "failed to disable LRO!\n");
1575
1576         netdev_for_each_lower_dev(dev, lower_dev, iter)
1577                 dev_disable_lro(lower_dev);
1578 }
1579 EXPORT_SYMBOL(dev_disable_lro);
1580
1581 /**
1582  *      dev_disable_gro_hw - disable HW Generic Receive Offload on a device
1583  *      @dev: device
1584  *
1585  *      Disable HW Generic Receive Offload (GRO_HW) on a net device.  Must be
1586  *      called under RTNL.  This is needed if Generic XDP is installed on
1587  *      the device.
1588  */
1589 static void dev_disable_gro_hw(struct net_device *dev)
1590 {
1591         dev->wanted_features &= ~NETIF_F_GRO_HW;
1592         netdev_update_features(dev);
1593
1594         if (unlikely(dev->features & NETIF_F_GRO_HW))
1595                 netdev_WARN(dev, "failed to disable GRO_HW!\n");
1596 }
1597
1598 const char *netdev_cmd_to_name(enum netdev_cmd cmd)
1599 {
1600 #define N(val)                                          \
1601         case NETDEV_##val:                              \
1602                 return "NETDEV_" __stringify(val);
1603         switch (cmd) {
1604         N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER)
1605         N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE)
1606         N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE)
1607         N(POST_INIT) N(PRE_UNINIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN)
1608         N(CHANGEUPPER) N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA)
1609         N(BONDING_INFO) N(PRECHANGEUPPER) N(CHANGELOWERSTATE)
1610         N(UDP_TUNNEL_PUSH_INFO) N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN)
1611         N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO)
1612         N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO)
1613         N(PRE_CHANGEADDR) N(OFFLOAD_XSTATS_ENABLE) N(OFFLOAD_XSTATS_DISABLE)
1614         N(OFFLOAD_XSTATS_REPORT_USED) N(OFFLOAD_XSTATS_REPORT_DELTA)
1615         N(XDP_FEAT_CHANGE)
1616         }
1617 #undef N
1618         return "UNKNOWN_NETDEV_EVENT";
1619 }
1620 EXPORT_SYMBOL_GPL(netdev_cmd_to_name);
1621
1622 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1623                                    struct net_device *dev)
1624 {
1625         struct netdev_notifier_info info = {
1626                 .dev = dev,
1627         };
1628
1629         return nb->notifier_call(nb, val, &info);
1630 }
1631
1632 static int call_netdevice_register_notifiers(struct notifier_block *nb,
1633                                              struct net_device *dev)
1634 {
1635         int err;
1636
1637         err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1638         err = notifier_to_errno(err);
1639         if (err)
1640                 return err;
1641
1642         if (!(dev->flags & IFF_UP))
1643                 return 0;
1644
1645         call_netdevice_notifier(nb, NETDEV_UP, dev);
1646         return 0;
1647 }
1648
1649 static void call_netdevice_unregister_notifiers(struct notifier_block *nb,
1650                                                 struct net_device *dev)
1651 {
1652         if (dev->flags & IFF_UP) {
1653                 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1654                                         dev);
1655                 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1656         }
1657         call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1658 }
1659
1660 static int call_netdevice_register_net_notifiers(struct notifier_block *nb,
1661                                                  struct net *net)
1662 {
1663         struct net_device *dev;
1664         int err;
1665
1666         for_each_netdev(net, dev) {
1667                 err = call_netdevice_register_notifiers(nb, dev);
1668                 if (err)
1669                         goto rollback;
1670         }
1671         return 0;
1672
1673 rollback:
1674         for_each_netdev_continue_reverse(net, dev)
1675                 call_netdevice_unregister_notifiers(nb, dev);
1676         return err;
1677 }
1678
1679 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb,
1680                                                     struct net *net)
1681 {
1682         struct net_device *dev;
1683
1684         for_each_netdev(net, dev)
1685                 call_netdevice_unregister_notifiers(nb, dev);
1686 }
1687
1688 static int dev_boot_phase = 1;
1689
1690 /**
1691  * register_netdevice_notifier - register a network notifier block
1692  * @nb: notifier
1693  *
1694  * Register a notifier to be called when network device events occur.
1695  * The notifier passed is linked into the kernel structures and must
1696  * not be reused until it has been unregistered. A negative errno code
1697  * is returned on a failure.
1698  *
1699  * When registered all registration and up events are replayed
1700  * to the new notifier to allow device to have a race free
1701  * view of the network device list.
1702  */
1703
1704 int register_netdevice_notifier(struct notifier_block *nb)
1705 {
1706         struct net *net;
1707         int err;
1708
1709         /* Close race with setup_net() and cleanup_net() */
1710         down_write(&pernet_ops_rwsem);
1711         rtnl_lock();
1712         err = raw_notifier_chain_register(&netdev_chain, nb);
1713         if (err)
1714                 goto unlock;
1715         if (dev_boot_phase)
1716                 goto unlock;
1717         for_each_net(net) {
1718                 err = call_netdevice_register_net_notifiers(nb, net);
1719                 if (err)
1720                         goto rollback;
1721         }
1722
1723 unlock:
1724         rtnl_unlock();
1725         up_write(&pernet_ops_rwsem);
1726         return err;
1727
1728 rollback:
1729         for_each_net_continue_reverse(net)
1730                 call_netdevice_unregister_net_notifiers(nb, net);
1731
1732         raw_notifier_chain_unregister(&netdev_chain, nb);
1733         goto unlock;
1734 }
1735 EXPORT_SYMBOL(register_netdevice_notifier);
1736
1737 /**
1738  * unregister_netdevice_notifier - unregister a network notifier block
1739  * @nb: notifier
1740  *
1741  * Unregister a notifier previously registered by
1742  * register_netdevice_notifier(). The notifier is unlinked into the
1743  * kernel structures and may then be reused. A negative errno code
1744  * is returned on a failure.
1745  *
1746  * After unregistering unregister and down device events are synthesized
1747  * for all devices on the device list to the removed notifier to remove
1748  * the need for special case cleanup code.
1749  */
1750
1751 int unregister_netdevice_notifier(struct notifier_block *nb)
1752 {
1753         struct net *net;
1754         int err;
1755
1756         /* Close race with setup_net() and cleanup_net() */
1757         down_write(&pernet_ops_rwsem);
1758         rtnl_lock();
1759         err = raw_notifier_chain_unregister(&netdev_chain, nb);
1760         if (err)
1761                 goto unlock;
1762
1763         for_each_net(net)
1764                 call_netdevice_unregister_net_notifiers(nb, net);
1765
1766 unlock:
1767         rtnl_unlock();
1768         up_write(&pernet_ops_rwsem);
1769         return err;
1770 }
1771 EXPORT_SYMBOL(unregister_netdevice_notifier);
1772
1773 static int __register_netdevice_notifier_net(struct net *net,
1774                                              struct notifier_block *nb,
1775                                              bool ignore_call_fail)
1776 {
1777         int err;
1778
1779         err = raw_notifier_chain_register(&net->netdev_chain, nb);
1780         if (err)
1781                 return err;
1782         if (dev_boot_phase)
1783                 return 0;
1784
1785         err = call_netdevice_register_net_notifiers(nb, net);
1786         if (err && !ignore_call_fail)
1787                 goto chain_unregister;
1788
1789         return 0;
1790
1791 chain_unregister:
1792         raw_notifier_chain_unregister(&net->netdev_chain, nb);
1793         return err;
1794 }
1795
1796 static int __unregister_netdevice_notifier_net(struct net *net,
1797                                                struct notifier_block *nb)
1798 {
1799         int err;
1800
1801         err = raw_notifier_chain_unregister(&net->netdev_chain, nb);
1802         if (err)
1803                 return err;
1804
1805         call_netdevice_unregister_net_notifiers(nb, net);
1806         return 0;
1807 }
1808
1809 /**
1810  * register_netdevice_notifier_net - register a per-netns network notifier block
1811  * @net: network namespace
1812  * @nb: notifier
1813  *
1814  * Register a notifier to be called when network device events occur.
1815  * The notifier passed is linked into the kernel structures and must
1816  * not be reused until it has been unregistered. A negative errno code
1817  * is returned on a failure.
1818  *
1819  * When registered all registration and up events are replayed
1820  * to the new notifier to allow device to have a race free
1821  * view of the network device list.
1822  */
1823
1824 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb)
1825 {
1826         int err;
1827
1828         rtnl_lock();
1829         err = __register_netdevice_notifier_net(net, nb, false);
1830         rtnl_unlock();
1831         return err;
1832 }
1833 EXPORT_SYMBOL(register_netdevice_notifier_net);
1834
1835 /**
1836  * unregister_netdevice_notifier_net - unregister a per-netns
1837  *                                     network notifier block
1838  * @net: network namespace
1839  * @nb: notifier
1840  *
1841  * Unregister a notifier previously registered by
1842  * register_netdevice_notifier_net(). The notifier is unlinked from the
1843  * kernel structures and may then be reused. A negative errno code
1844  * is returned on a failure.
1845  *
1846  * After unregistering unregister and down device events are synthesized
1847  * for all devices on the device list to the removed notifier to remove
1848  * the need for special case cleanup code.
1849  */
1850
1851 int unregister_netdevice_notifier_net(struct net *net,
1852                                       struct notifier_block *nb)
1853 {
1854         int err;
1855
1856         rtnl_lock();
1857         err = __unregister_netdevice_notifier_net(net, nb);
1858         rtnl_unlock();
1859         return err;
1860 }
1861 EXPORT_SYMBOL(unregister_netdevice_notifier_net);
1862
1863 static void __move_netdevice_notifier_net(struct net *src_net,
1864                                           struct net *dst_net,
1865                                           struct notifier_block *nb)
1866 {
1867         __unregister_netdevice_notifier_net(src_net, nb);
1868         __register_netdevice_notifier_net(dst_net, nb, true);
1869 }
1870
1871 int register_netdevice_notifier_dev_net(struct net_device *dev,
1872                                         struct notifier_block *nb,
1873                                         struct netdev_net_notifier *nn)
1874 {
1875         int err;
1876
1877         rtnl_lock();
1878         err = __register_netdevice_notifier_net(dev_net(dev), nb, false);
1879         if (!err) {
1880                 nn->nb = nb;
1881                 list_add(&nn->list, &dev->net_notifier_list);
1882         }
1883         rtnl_unlock();
1884         return err;
1885 }
1886 EXPORT_SYMBOL(register_netdevice_notifier_dev_net);
1887
1888 int unregister_netdevice_notifier_dev_net(struct net_device *dev,
1889                                           struct notifier_block *nb,
1890                                           struct netdev_net_notifier *nn)
1891 {
1892         int err;
1893
1894         rtnl_lock();
1895         list_del(&nn->list);
1896         err = __unregister_netdevice_notifier_net(dev_net(dev), nb);
1897         rtnl_unlock();
1898         return err;
1899 }
1900 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net);
1901
1902 static void move_netdevice_notifiers_dev_net(struct net_device *dev,
1903                                              struct net *net)
1904 {
1905         struct netdev_net_notifier *nn;
1906
1907         list_for_each_entry(nn, &dev->net_notifier_list, list)
1908                 __move_netdevice_notifier_net(dev_net(dev), net, nn->nb);
1909 }
1910
1911 /**
1912  *      call_netdevice_notifiers_info - call all network notifier blocks
1913  *      @val: value passed unmodified to notifier function
1914  *      @info: notifier information data
1915  *
1916  *      Call all network notifier blocks.  Parameters and return value
1917  *      are as for raw_notifier_call_chain().
1918  */
1919
1920 int call_netdevice_notifiers_info(unsigned long val,
1921                                   struct netdev_notifier_info *info)
1922 {
1923         struct net *net = dev_net(info->dev);
1924         int ret;
1925
1926         ASSERT_RTNL();
1927
1928         /* Run per-netns notifier block chain first, then run the global one.
1929          * Hopefully, one day, the global one is going to be removed after
1930          * all notifier block registrators get converted to be per-netns.
1931          */
1932         ret = raw_notifier_call_chain(&net->netdev_chain, val, info);
1933         if (ret & NOTIFY_STOP_MASK)
1934                 return ret;
1935         return raw_notifier_call_chain(&netdev_chain, val, info);
1936 }
1937
1938 /**
1939  *      call_netdevice_notifiers_info_robust - call per-netns notifier blocks
1940  *                                             for and rollback on error
1941  *      @val_up: value passed unmodified to notifier function
1942  *      @val_down: value passed unmodified to the notifier function when
1943  *                 recovering from an error on @val_up
1944  *      @info: notifier information data
1945  *
1946  *      Call all per-netns network notifier blocks, but not notifier blocks on
1947  *      the global notifier chain. Parameters and return value are as for
1948  *      raw_notifier_call_chain_robust().
1949  */
1950
1951 static int
1952 call_netdevice_notifiers_info_robust(unsigned long val_up,
1953                                      unsigned long val_down,
1954                                      struct netdev_notifier_info *info)
1955 {
1956         struct net *net = dev_net(info->dev);
1957
1958         ASSERT_RTNL();
1959
1960         return raw_notifier_call_chain_robust(&net->netdev_chain,
1961                                               val_up, val_down, info);
1962 }
1963
1964 static int call_netdevice_notifiers_extack(unsigned long val,
1965                                            struct net_device *dev,
1966                                            struct netlink_ext_ack *extack)
1967 {
1968         struct netdev_notifier_info info = {
1969                 .dev = dev,
1970                 .extack = extack,
1971         };
1972
1973         return call_netdevice_notifiers_info(val, &info);
1974 }
1975
1976 /**
1977  *      call_netdevice_notifiers - call all network notifier blocks
1978  *      @val: value passed unmodified to notifier function
1979  *      @dev: net_device pointer passed unmodified to notifier function
1980  *
1981  *      Call all network notifier blocks.  Parameters and return value
1982  *      are as for raw_notifier_call_chain().
1983  */
1984
1985 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1986 {
1987         return call_netdevice_notifiers_extack(val, dev, NULL);
1988 }
1989 EXPORT_SYMBOL(call_netdevice_notifiers);
1990
1991 /**
1992  *      call_netdevice_notifiers_mtu - call all network notifier blocks
1993  *      @val: value passed unmodified to notifier function
1994  *      @dev: net_device pointer passed unmodified to notifier function
1995  *      @arg: additional u32 argument passed to the notifier function
1996  *
1997  *      Call all network notifier blocks.  Parameters and return value
1998  *      are as for raw_notifier_call_chain().
1999  */
2000 static int call_netdevice_notifiers_mtu(unsigned long val,
2001                                         struct net_device *dev, u32 arg)
2002 {
2003         struct netdev_notifier_info_ext info = {
2004                 .info.dev = dev,
2005                 .ext.mtu = arg,
2006         };
2007
2008         BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0);
2009
2010         return call_netdevice_notifiers_info(val, &info.info);
2011 }
2012
2013 #ifdef CONFIG_NET_INGRESS
2014 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key);
2015
2016 void net_inc_ingress_queue(void)
2017 {
2018         static_branch_inc(&ingress_needed_key);
2019 }
2020 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
2021
2022 void net_dec_ingress_queue(void)
2023 {
2024         static_branch_dec(&ingress_needed_key);
2025 }
2026 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
2027 #endif
2028
2029 #ifdef CONFIG_NET_EGRESS
2030 static DEFINE_STATIC_KEY_FALSE(egress_needed_key);
2031
2032 void net_inc_egress_queue(void)
2033 {
2034         static_branch_inc(&egress_needed_key);
2035 }
2036 EXPORT_SYMBOL_GPL(net_inc_egress_queue);
2037
2038 void net_dec_egress_queue(void)
2039 {
2040         static_branch_dec(&egress_needed_key);
2041 }
2042 EXPORT_SYMBOL_GPL(net_dec_egress_queue);
2043 #endif
2044
2045 DEFINE_STATIC_KEY_FALSE(netstamp_needed_key);
2046 EXPORT_SYMBOL(netstamp_needed_key);
2047 #ifdef CONFIG_JUMP_LABEL
2048 static atomic_t netstamp_needed_deferred;
2049 static atomic_t netstamp_wanted;
2050 static void netstamp_clear(struct work_struct *work)
2051 {
2052         int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
2053         int wanted;
2054
2055         wanted = atomic_add_return(deferred, &netstamp_wanted);
2056         if (wanted > 0)
2057                 static_branch_enable(&netstamp_needed_key);
2058         else
2059                 static_branch_disable(&netstamp_needed_key);
2060 }
2061 static DECLARE_WORK(netstamp_work, netstamp_clear);
2062 #endif
2063
2064 void net_enable_timestamp(void)
2065 {
2066 #ifdef CONFIG_JUMP_LABEL
2067         int wanted = atomic_read(&netstamp_wanted);
2068
2069         while (wanted > 0) {
2070                 if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted + 1))
2071                         return;
2072         }
2073         atomic_inc(&netstamp_needed_deferred);
2074         schedule_work(&netstamp_work);
2075 #else
2076         static_branch_inc(&netstamp_needed_key);
2077 #endif
2078 }
2079 EXPORT_SYMBOL(net_enable_timestamp);
2080
2081 void net_disable_timestamp(void)
2082 {
2083 #ifdef CONFIG_JUMP_LABEL
2084         int wanted = atomic_read(&netstamp_wanted);
2085
2086         while (wanted > 1) {
2087                 if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted - 1))
2088                         return;
2089         }
2090         atomic_dec(&netstamp_needed_deferred);
2091         schedule_work(&netstamp_work);
2092 #else
2093         static_branch_dec(&netstamp_needed_key);
2094 #endif
2095 }
2096 EXPORT_SYMBOL(net_disable_timestamp);
2097
2098 static inline void net_timestamp_set(struct sk_buff *skb)
2099 {
2100         skb->tstamp = 0;
2101         skb->mono_delivery_time = 0;
2102         if (static_branch_unlikely(&netstamp_needed_key))
2103                 skb->tstamp = ktime_get_real();
2104 }
2105
2106 #define net_timestamp_check(COND, SKB)                          \
2107         if (static_branch_unlikely(&netstamp_needed_key)) {     \
2108                 if ((COND) && !(SKB)->tstamp)                   \
2109                         (SKB)->tstamp = ktime_get_real();       \
2110         }                                                       \
2111
2112 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb)
2113 {
2114         return __is_skb_forwardable(dev, skb, true);
2115 }
2116 EXPORT_SYMBOL_GPL(is_skb_forwardable);
2117
2118 static int __dev_forward_skb2(struct net_device *dev, struct sk_buff *skb,
2119                               bool check_mtu)
2120 {
2121         int ret = ____dev_forward_skb(dev, skb, check_mtu);
2122
2123         if (likely(!ret)) {
2124                 skb->protocol = eth_type_trans(skb, dev);
2125                 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
2126         }
2127
2128         return ret;
2129 }
2130
2131 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2132 {
2133         return __dev_forward_skb2(dev, skb, true);
2134 }
2135 EXPORT_SYMBOL_GPL(__dev_forward_skb);
2136
2137 /**
2138  * dev_forward_skb - loopback an skb to another netif
2139  *
2140  * @dev: destination network device
2141  * @skb: buffer to forward
2142  *
2143  * return values:
2144  *      NET_RX_SUCCESS  (no congestion)
2145  *      NET_RX_DROP     (packet was dropped, but freed)
2146  *
2147  * dev_forward_skb can be used for injecting an skb from the
2148  * start_xmit function of one device into the receive queue
2149  * of another device.
2150  *
2151  * The receiving device may be in another namespace, so
2152  * we have to clear all information in the skb that could
2153  * impact namespace isolation.
2154  */
2155 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2156 {
2157         return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
2158 }
2159 EXPORT_SYMBOL_GPL(dev_forward_skb);
2160
2161 int dev_forward_skb_nomtu(struct net_device *dev, struct sk_buff *skb)
2162 {
2163         return __dev_forward_skb2(dev, skb, false) ?: netif_rx_internal(skb);
2164 }
2165
2166 static inline int deliver_skb(struct sk_buff *skb,
2167                               struct packet_type *pt_prev,
2168                               struct net_device *orig_dev)
2169 {
2170         if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
2171                 return -ENOMEM;
2172         refcount_inc(&skb->users);
2173         return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2174 }
2175
2176 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
2177                                           struct packet_type **pt,
2178                                           struct net_device *orig_dev,
2179                                           __be16 type,
2180                                           struct list_head *ptype_list)
2181 {
2182         struct packet_type *ptype, *pt_prev = *pt;
2183
2184         list_for_each_entry_rcu(ptype, ptype_list, list) {
2185                 if (ptype->type != type)
2186                         continue;
2187                 if (pt_prev)
2188                         deliver_skb(skb, pt_prev, orig_dev);
2189                 pt_prev = ptype;
2190         }
2191         *pt = pt_prev;
2192 }
2193
2194 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
2195 {
2196         if (!ptype->af_packet_priv || !skb->sk)
2197                 return false;
2198
2199         if (ptype->id_match)
2200                 return ptype->id_match(ptype, skb->sk);
2201         else if ((struct sock *)ptype->af_packet_priv == skb->sk)
2202                 return true;
2203
2204         return false;
2205 }
2206
2207 /**
2208  * dev_nit_active - return true if any network interface taps are in use
2209  *
2210  * @dev: network device to check for the presence of taps
2211  */
2212 bool dev_nit_active(struct net_device *dev)
2213 {
2214         return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all);
2215 }
2216 EXPORT_SYMBOL_GPL(dev_nit_active);
2217
2218 /*
2219  *      Support routine. Sends outgoing frames to any network
2220  *      taps currently in use.
2221  */
2222
2223 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
2224 {
2225         struct packet_type *ptype;
2226         struct sk_buff *skb2 = NULL;
2227         struct packet_type *pt_prev = NULL;
2228         struct list_head *ptype_list = &ptype_all;
2229
2230         rcu_read_lock();
2231 again:
2232         list_for_each_entry_rcu(ptype, ptype_list, list) {
2233                 if (ptype->ignore_outgoing)
2234                         continue;
2235
2236                 /* Never send packets back to the socket
2237                  * they originated from - MvS (miquels@drinkel.ow.org)
2238                  */
2239                 if (skb_loop_sk(ptype, skb))
2240                         continue;
2241
2242                 if (pt_prev) {
2243                         deliver_skb(skb2, pt_prev, skb->dev);
2244                         pt_prev = ptype;
2245                         continue;
2246                 }
2247
2248                 /* need to clone skb, done only once */
2249                 skb2 = skb_clone(skb, GFP_ATOMIC);
2250                 if (!skb2)
2251                         goto out_unlock;
2252
2253                 net_timestamp_set(skb2);
2254
2255                 /* skb->nh should be correctly
2256                  * set by sender, so that the second statement is
2257                  * just protection against buggy protocols.
2258                  */
2259                 skb_reset_mac_header(skb2);
2260
2261                 if (skb_network_header(skb2) < skb2->data ||
2262                     skb_network_header(skb2) > skb_tail_pointer(skb2)) {
2263                         net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
2264                                              ntohs(skb2->protocol),
2265                                              dev->name);
2266                         skb_reset_network_header(skb2);
2267                 }
2268
2269                 skb2->transport_header = skb2->network_header;
2270                 skb2->pkt_type = PACKET_OUTGOING;
2271                 pt_prev = ptype;
2272         }
2273
2274         if (ptype_list == &ptype_all) {
2275                 ptype_list = &dev->ptype_all;
2276                 goto again;
2277         }
2278 out_unlock:
2279         if (pt_prev) {
2280                 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC))
2281                         pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
2282                 else
2283                         kfree_skb(skb2);
2284         }
2285         rcu_read_unlock();
2286 }
2287 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit);
2288
2289 /**
2290  * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
2291  * @dev: Network device
2292  * @txq: number of queues available
2293  *
2294  * If real_num_tx_queues is changed the tc mappings may no longer be
2295  * valid. To resolve this verify the tc mapping remains valid and if
2296  * not NULL the mapping. With no priorities mapping to this
2297  * offset/count pair it will no longer be used. In the worst case TC0
2298  * is invalid nothing can be done so disable priority mappings. If is
2299  * expected that drivers will fix this mapping if they can before
2300  * calling netif_set_real_num_tx_queues.
2301  */
2302 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
2303 {
2304         int i;
2305         struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2306
2307         /* If TC0 is invalidated disable TC mapping */
2308         if (tc->offset + tc->count > txq) {
2309                 netdev_warn(dev, "Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
2310                 dev->num_tc = 0;
2311                 return;
2312         }
2313
2314         /* Invalidated prio to tc mappings set to TC0 */
2315         for (i = 1; i < TC_BITMASK + 1; i++) {
2316                 int q = netdev_get_prio_tc_map(dev, i);
2317
2318                 tc = &dev->tc_to_txq[q];
2319                 if (tc->offset + tc->count > txq) {
2320                         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",
2321                                     i, q);
2322                         netdev_set_prio_tc_map(dev, i, 0);
2323                 }
2324         }
2325 }
2326
2327 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq)
2328 {
2329         if (dev->num_tc) {
2330                 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2331                 int i;
2332
2333                 /* walk through the TCs and see if it falls into any of them */
2334                 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) {
2335                         if ((txq - tc->offset) < tc->count)
2336                                 return i;
2337                 }
2338
2339                 /* didn't find it, just return -1 to indicate no match */
2340                 return -1;
2341         }
2342
2343         return 0;
2344 }
2345 EXPORT_SYMBOL(netdev_txq_to_tc);
2346
2347 #ifdef CONFIG_XPS
2348 static struct static_key xps_needed __read_mostly;
2349 static struct static_key xps_rxqs_needed __read_mostly;
2350 static DEFINE_MUTEX(xps_map_mutex);
2351 #define xmap_dereference(P)             \
2352         rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
2353
2354 static bool remove_xps_queue(struct xps_dev_maps *dev_maps,
2355                              struct xps_dev_maps *old_maps, int tci, u16 index)
2356 {
2357         struct xps_map *map = NULL;
2358         int pos;
2359
2360         if (dev_maps)
2361                 map = xmap_dereference(dev_maps->attr_map[tci]);
2362         if (!map)
2363                 return false;
2364
2365         for (pos = map->len; pos--;) {
2366                 if (map->queues[pos] != index)
2367                         continue;
2368
2369                 if (map->len > 1) {
2370                         map->queues[pos] = map->queues[--map->len];
2371                         break;
2372                 }
2373
2374                 if (old_maps)
2375                         RCU_INIT_POINTER(old_maps->attr_map[tci], NULL);
2376                 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2377                 kfree_rcu(map, rcu);
2378                 return false;
2379         }
2380
2381         return true;
2382 }
2383
2384 static bool remove_xps_queue_cpu(struct net_device *dev,
2385                                  struct xps_dev_maps *dev_maps,
2386                                  int cpu, u16 offset, u16 count)
2387 {
2388         int num_tc = dev_maps->num_tc;
2389         bool active = false;
2390         int tci;
2391
2392         for (tci = cpu * num_tc; num_tc--; tci++) {
2393                 int i, j;
2394
2395                 for (i = count, j = offset; i--; j++) {
2396                         if (!remove_xps_queue(dev_maps, NULL, tci, j))
2397                                 break;
2398                 }
2399
2400                 active |= i < 0;
2401         }
2402
2403         return active;
2404 }
2405
2406 static void reset_xps_maps(struct net_device *dev,
2407                            struct xps_dev_maps *dev_maps,
2408                            enum xps_map_type type)
2409 {
2410         static_key_slow_dec_cpuslocked(&xps_needed);
2411         if (type == XPS_RXQS)
2412                 static_key_slow_dec_cpuslocked(&xps_rxqs_needed);
2413
2414         RCU_INIT_POINTER(dev->xps_maps[type], NULL);
2415
2416         kfree_rcu(dev_maps, rcu);
2417 }
2418
2419 static void clean_xps_maps(struct net_device *dev, enum xps_map_type type,
2420                            u16 offset, u16 count)
2421 {
2422         struct xps_dev_maps *dev_maps;
2423         bool active = false;
2424         int i, j;
2425
2426         dev_maps = xmap_dereference(dev->xps_maps[type]);
2427         if (!dev_maps)
2428                 return;
2429
2430         for (j = 0; j < dev_maps->nr_ids; j++)
2431                 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, count);
2432         if (!active)
2433                 reset_xps_maps(dev, dev_maps, type);
2434
2435         if (type == XPS_CPUS) {
2436                 for (i = offset + (count - 1); count--; i--)
2437                         netdev_queue_numa_node_write(
2438                                 netdev_get_tx_queue(dev, i), NUMA_NO_NODE);
2439         }
2440 }
2441
2442 static void netif_reset_xps_queues(struct net_device *dev, u16 offset,
2443                                    u16 count)
2444 {
2445         if (!static_key_false(&xps_needed))
2446                 return;
2447
2448         cpus_read_lock();
2449         mutex_lock(&xps_map_mutex);
2450
2451         if (static_key_false(&xps_rxqs_needed))
2452                 clean_xps_maps(dev, XPS_RXQS, offset, count);
2453
2454         clean_xps_maps(dev, XPS_CPUS, offset, count);
2455
2456         mutex_unlock(&xps_map_mutex);
2457         cpus_read_unlock();
2458 }
2459
2460 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
2461 {
2462         netif_reset_xps_queues(dev, index, dev->num_tx_queues - index);
2463 }
2464
2465 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index,
2466                                       u16 index, bool is_rxqs_map)
2467 {
2468         struct xps_map *new_map;
2469         int alloc_len = XPS_MIN_MAP_ALLOC;
2470         int i, pos;
2471
2472         for (pos = 0; map && pos < map->len; pos++) {
2473                 if (map->queues[pos] != index)
2474                         continue;
2475                 return map;
2476         }
2477
2478         /* Need to add tx-queue to this CPU's/rx-queue's existing map */
2479         if (map) {
2480                 if (pos < map->alloc_len)
2481                         return map;
2482
2483                 alloc_len = map->alloc_len * 2;
2484         }
2485
2486         /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's
2487          *  map
2488          */
2489         if (is_rxqs_map)
2490                 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL);
2491         else
2492                 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2493                                        cpu_to_node(attr_index));
2494         if (!new_map)
2495                 return NULL;
2496
2497         for (i = 0; i < pos; i++)
2498                 new_map->queues[i] = map->queues[i];
2499         new_map->alloc_len = alloc_len;
2500         new_map->len = pos;
2501
2502         return new_map;
2503 }
2504
2505 /* Copy xps maps at a given index */
2506 static void xps_copy_dev_maps(struct xps_dev_maps *dev_maps,
2507                               struct xps_dev_maps *new_dev_maps, int index,
2508                               int tc, bool skip_tc)
2509 {
2510         int i, tci = index * dev_maps->num_tc;
2511         struct xps_map *map;
2512
2513         /* copy maps belonging to foreign traffic classes */
2514         for (i = 0; i < dev_maps->num_tc; i++, tci++) {
2515                 if (i == tc && skip_tc)
2516                         continue;
2517
2518                 /* fill in the new device map from the old device map */
2519                 map = xmap_dereference(dev_maps->attr_map[tci]);
2520                 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2521         }
2522 }
2523
2524 /* Must be called under cpus_read_lock */
2525 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask,
2526                           u16 index, enum xps_map_type type)
2527 {
2528         struct xps_dev_maps *dev_maps, *new_dev_maps = NULL, *old_dev_maps = NULL;
2529         const unsigned long *online_mask = NULL;
2530         bool active = false, copy = false;
2531         int i, j, tci, numa_node_id = -2;
2532         int maps_sz, num_tc = 1, tc = 0;
2533         struct xps_map *map, *new_map;
2534         unsigned int nr_ids;
2535
2536         WARN_ON_ONCE(index >= dev->num_tx_queues);
2537
2538         if (dev->num_tc) {
2539                 /* Do not allow XPS on subordinate device directly */
2540                 num_tc = dev->num_tc;
2541                 if (num_tc < 0)
2542                         return -EINVAL;
2543
2544                 /* If queue belongs to subordinate dev use its map */
2545                 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
2546
2547                 tc = netdev_txq_to_tc(dev, index);
2548                 if (tc < 0)
2549                         return -EINVAL;
2550         }
2551
2552         mutex_lock(&xps_map_mutex);
2553
2554         dev_maps = xmap_dereference(dev->xps_maps[type]);
2555         if (type == XPS_RXQS) {
2556                 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues);
2557                 nr_ids = dev->num_rx_queues;
2558         } else {
2559                 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc);
2560                 if (num_possible_cpus() > 1)
2561                         online_mask = cpumask_bits(cpu_online_mask);
2562                 nr_ids = nr_cpu_ids;
2563         }
2564
2565         if (maps_sz < L1_CACHE_BYTES)
2566                 maps_sz = L1_CACHE_BYTES;
2567
2568         /* The old dev_maps could be larger or smaller than the one we're
2569          * setting up now, as dev->num_tc or nr_ids could have been updated in
2570          * between. We could try to be smart, but let's be safe instead and only
2571          * copy foreign traffic classes if the two map sizes match.
2572          */
2573         if (dev_maps &&
2574             dev_maps->num_tc == num_tc && dev_maps->nr_ids == nr_ids)
2575                 copy = true;
2576
2577         /* allocate memory for queue storage */
2578         for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids),
2579              j < nr_ids;) {
2580                 if (!new_dev_maps) {
2581                         new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2582                         if (!new_dev_maps) {
2583                                 mutex_unlock(&xps_map_mutex);
2584                                 return -ENOMEM;
2585                         }
2586
2587                         new_dev_maps->nr_ids = nr_ids;
2588                         new_dev_maps->num_tc = num_tc;
2589                 }
2590
2591                 tci = j * num_tc + tc;
2592                 map = copy ? xmap_dereference(dev_maps->attr_map[tci]) : NULL;
2593
2594                 map = expand_xps_map(map, j, index, type == XPS_RXQS);
2595                 if (!map)
2596                         goto error;
2597
2598                 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2599         }
2600
2601         if (!new_dev_maps)
2602                 goto out_no_new_maps;
2603
2604         if (!dev_maps) {
2605                 /* Increment static keys at most once per type */
2606                 static_key_slow_inc_cpuslocked(&xps_needed);
2607                 if (type == XPS_RXQS)
2608                         static_key_slow_inc_cpuslocked(&xps_rxqs_needed);
2609         }
2610
2611         for (j = 0; j < nr_ids; j++) {
2612                 bool skip_tc = false;
2613
2614                 tci = j * num_tc + tc;
2615                 if (netif_attr_test_mask(j, mask, nr_ids) &&
2616                     netif_attr_test_online(j, online_mask, nr_ids)) {
2617                         /* add tx-queue to CPU/rx-queue maps */
2618                         int pos = 0;
2619
2620                         skip_tc = true;
2621
2622                         map = xmap_dereference(new_dev_maps->attr_map[tci]);
2623                         while ((pos < map->len) && (map->queues[pos] != index))
2624                                 pos++;
2625
2626                         if (pos == map->len)
2627                                 map->queues[map->len++] = index;
2628 #ifdef CONFIG_NUMA
2629                         if (type == XPS_CPUS) {
2630                                 if (numa_node_id == -2)
2631                                         numa_node_id = cpu_to_node(j);
2632                                 else if (numa_node_id != cpu_to_node(j))
2633                                         numa_node_id = -1;
2634                         }
2635 #endif
2636                 }
2637
2638                 if (copy)
2639                         xps_copy_dev_maps(dev_maps, new_dev_maps, j, tc,
2640                                           skip_tc);
2641         }
2642
2643         rcu_assign_pointer(dev->xps_maps[type], new_dev_maps);
2644
2645         /* Cleanup old maps */
2646         if (!dev_maps)
2647                 goto out_no_old_maps;
2648
2649         for (j = 0; j < dev_maps->nr_ids; j++) {
2650                 for (i = num_tc, tci = j * dev_maps->num_tc; i--; tci++) {
2651                         map = xmap_dereference(dev_maps->attr_map[tci]);
2652                         if (!map)
2653                                 continue;
2654
2655                         if (copy) {
2656                                 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2657                                 if (map == new_map)
2658                                         continue;
2659                         }
2660
2661                         RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2662                         kfree_rcu(map, rcu);
2663                 }
2664         }
2665
2666         old_dev_maps = dev_maps;
2667
2668 out_no_old_maps:
2669         dev_maps = new_dev_maps;
2670         active = true;
2671
2672 out_no_new_maps:
2673         if (type == XPS_CPUS)
2674                 /* update Tx queue numa node */
2675                 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2676                                              (numa_node_id >= 0) ?
2677                                              numa_node_id : NUMA_NO_NODE);
2678
2679         if (!dev_maps)
2680                 goto out_no_maps;
2681
2682         /* removes tx-queue from unused CPUs/rx-queues */
2683         for (j = 0; j < dev_maps->nr_ids; j++) {
2684                 tci = j * dev_maps->num_tc;
2685
2686                 for (i = 0; i < dev_maps->num_tc; i++, tci++) {
2687                         if (i == tc &&
2688                             netif_attr_test_mask(j, mask, dev_maps->nr_ids) &&
2689                             netif_attr_test_online(j, online_mask, dev_maps->nr_ids))
2690                                 continue;
2691
2692                         active |= remove_xps_queue(dev_maps,
2693                                                    copy ? old_dev_maps : NULL,
2694                                                    tci, index);
2695                 }
2696         }
2697
2698         if (old_dev_maps)
2699                 kfree_rcu(old_dev_maps, rcu);
2700
2701         /* free map if not active */
2702         if (!active)
2703                 reset_xps_maps(dev, dev_maps, type);
2704
2705 out_no_maps:
2706         mutex_unlock(&xps_map_mutex);
2707
2708         return 0;
2709 error:
2710         /* remove any maps that we added */
2711         for (j = 0; j < nr_ids; j++) {
2712                 for (i = num_tc, tci = j * num_tc; i--; tci++) {
2713                         new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2714                         map = copy ?
2715                               xmap_dereference(dev_maps->attr_map[tci]) :
2716                               NULL;
2717                         if (new_map && new_map != map)
2718                                 kfree(new_map);
2719                 }
2720         }
2721
2722         mutex_unlock(&xps_map_mutex);
2723
2724         kfree(new_dev_maps);
2725         return -ENOMEM;
2726 }
2727 EXPORT_SYMBOL_GPL(__netif_set_xps_queue);
2728
2729 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2730                         u16 index)
2731 {
2732         int ret;
2733
2734         cpus_read_lock();
2735         ret =  __netif_set_xps_queue(dev, cpumask_bits(mask), index, XPS_CPUS);
2736         cpus_read_unlock();
2737
2738         return ret;
2739 }
2740 EXPORT_SYMBOL(netif_set_xps_queue);
2741
2742 #endif
2743 static void netdev_unbind_all_sb_channels(struct net_device *dev)
2744 {
2745         struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2746
2747         /* Unbind any subordinate channels */
2748         while (txq-- != &dev->_tx[0]) {
2749                 if (txq->sb_dev)
2750                         netdev_unbind_sb_channel(dev, txq->sb_dev);
2751         }
2752 }
2753
2754 void netdev_reset_tc(struct net_device *dev)
2755 {
2756 #ifdef CONFIG_XPS
2757         netif_reset_xps_queues_gt(dev, 0);
2758 #endif
2759         netdev_unbind_all_sb_channels(dev);
2760
2761         /* Reset TC configuration of device */
2762         dev->num_tc = 0;
2763         memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq));
2764         memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map));
2765 }
2766 EXPORT_SYMBOL(netdev_reset_tc);
2767
2768 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset)
2769 {
2770         if (tc >= dev->num_tc)
2771                 return -EINVAL;
2772
2773 #ifdef CONFIG_XPS
2774         netif_reset_xps_queues(dev, offset, count);
2775 #endif
2776         dev->tc_to_txq[tc].count = count;
2777         dev->tc_to_txq[tc].offset = offset;
2778         return 0;
2779 }
2780 EXPORT_SYMBOL(netdev_set_tc_queue);
2781
2782 int netdev_set_num_tc(struct net_device *dev, u8 num_tc)
2783 {
2784         if (num_tc > TC_MAX_QUEUE)
2785                 return -EINVAL;
2786
2787 #ifdef CONFIG_XPS
2788         netif_reset_xps_queues_gt(dev, 0);
2789 #endif
2790         netdev_unbind_all_sb_channels(dev);
2791
2792         dev->num_tc = num_tc;
2793         return 0;
2794 }
2795 EXPORT_SYMBOL(netdev_set_num_tc);
2796
2797 void netdev_unbind_sb_channel(struct net_device *dev,
2798                               struct net_device *sb_dev)
2799 {
2800         struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2801
2802 #ifdef CONFIG_XPS
2803         netif_reset_xps_queues_gt(sb_dev, 0);
2804 #endif
2805         memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq));
2806         memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map));
2807
2808         while (txq-- != &dev->_tx[0]) {
2809                 if (txq->sb_dev == sb_dev)
2810                         txq->sb_dev = NULL;
2811         }
2812 }
2813 EXPORT_SYMBOL(netdev_unbind_sb_channel);
2814
2815 int netdev_bind_sb_channel_queue(struct net_device *dev,
2816                                  struct net_device *sb_dev,
2817                                  u8 tc, u16 count, u16 offset)
2818 {
2819         /* Make certain the sb_dev and dev are already configured */
2820         if (sb_dev->num_tc >= 0 || tc >= dev->num_tc)
2821                 return -EINVAL;
2822
2823         /* We cannot hand out queues we don't have */
2824         if ((offset + count) > dev->real_num_tx_queues)
2825                 return -EINVAL;
2826
2827         /* Record the mapping */
2828         sb_dev->tc_to_txq[tc].count = count;
2829         sb_dev->tc_to_txq[tc].offset = offset;
2830
2831         /* Provide a way for Tx queue to find the tc_to_txq map or
2832          * XPS map for itself.
2833          */
2834         while (count--)
2835                 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev;
2836
2837         return 0;
2838 }
2839 EXPORT_SYMBOL(netdev_bind_sb_channel_queue);
2840
2841 int netdev_set_sb_channel(struct net_device *dev, u16 channel)
2842 {
2843         /* Do not use a multiqueue device to represent a subordinate channel */
2844         if (netif_is_multiqueue(dev))
2845                 return -ENODEV;
2846
2847         /* We allow channels 1 - 32767 to be used for subordinate channels.
2848          * Channel 0 is meant to be "native" mode and used only to represent
2849          * the main root device. We allow writing 0 to reset the device back
2850          * to normal mode after being used as a subordinate channel.
2851          */
2852         if (channel > S16_MAX)
2853                 return -EINVAL;
2854
2855         dev->num_tc = -channel;
2856
2857         return 0;
2858 }
2859 EXPORT_SYMBOL(netdev_set_sb_channel);
2860
2861 /*
2862  * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2863  * greater than real_num_tx_queues stale skbs on the qdisc must be flushed.
2864  */
2865 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2866 {
2867         bool disabling;
2868         int rc;
2869
2870         disabling = txq < dev->real_num_tx_queues;
2871
2872         if (txq < 1 || txq > dev->num_tx_queues)
2873                 return -EINVAL;
2874
2875         if (dev->reg_state == NETREG_REGISTERED ||
2876             dev->reg_state == NETREG_UNREGISTERING) {
2877                 ASSERT_RTNL();
2878
2879                 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2880                                                   txq);
2881                 if (rc)
2882                         return rc;
2883
2884                 if (dev->num_tc)
2885                         netif_setup_tc(dev, txq);
2886
2887                 dev_qdisc_change_real_num_tx(dev, txq);
2888
2889                 dev->real_num_tx_queues = txq;
2890
2891                 if (disabling) {
2892                         synchronize_net();
2893                         qdisc_reset_all_tx_gt(dev, txq);
2894 #ifdef CONFIG_XPS
2895                         netif_reset_xps_queues_gt(dev, txq);
2896 #endif
2897                 }
2898         } else {
2899                 dev->real_num_tx_queues = txq;
2900         }
2901
2902         return 0;
2903 }
2904 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2905
2906 #ifdef CONFIG_SYSFS
2907 /**
2908  *      netif_set_real_num_rx_queues - set actual number of RX queues used
2909  *      @dev: Network device
2910  *      @rxq: Actual number of RX queues
2911  *
2912  *      This must be called either with the rtnl_lock held or before
2913  *      registration of the net device.  Returns 0 on success, or a
2914  *      negative error code.  If called before registration, it always
2915  *      succeeds.
2916  */
2917 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2918 {
2919         int rc;
2920
2921         if (rxq < 1 || rxq > dev->num_rx_queues)
2922                 return -EINVAL;
2923
2924         if (dev->reg_state == NETREG_REGISTERED) {
2925                 ASSERT_RTNL();
2926
2927                 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2928                                                   rxq);
2929                 if (rc)
2930                         return rc;
2931         }
2932
2933         dev->real_num_rx_queues = rxq;
2934         return 0;
2935 }
2936 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2937 #endif
2938
2939 /**
2940  *      netif_set_real_num_queues - set actual number of RX and TX queues used
2941  *      @dev: Network device
2942  *      @txq: Actual number of TX queues
2943  *      @rxq: Actual number of RX queues
2944  *
2945  *      Set the real number of both TX and RX queues.
2946  *      Does nothing if the number of queues is already correct.
2947  */
2948 int netif_set_real_num_queues(struct net_device *dev,
2949                               unsigned int txq, unsigned int rxq)
2950 {
2951         unsigned int old_rxq = dev->real_num_rx_queues;
2952         int err;
2953
2954         if (txq < 1 || txq > dev->num_tx_queues ||
2955             rxq < 1 || rxq > dev->num_rx_queues)
2956                 return -EINVAL;
2957
2958         /* Start from increases, so the error path only does decreases -
2959          * decreases can't fail.
2960          */
2961         if (rxq > dev->real_num_rx_queues) {
2962                 err = netif_set_real_num_rx_queues(dev, rxq);
2963                 if (err)
2964                         return err;
2965         }
2966         if (txq > dev->real_num_tx_queues) {
2967                 err = netif_set_real_num_tx_queues(dev, txq);
2968                 if (err)
2969                         goto undo_rx;
2970         }
2971         if (rxq < dev->real_num_rx_queues)
2972                 WARN_ON(netif_set_real_num_rx_queues(dev, rxq));
2973         if (txq < dev->real_num_tx_queues)
2974                 WARN_ON(netif_set_real_num_tx_queues(dev, txq));
2975
2976         return 0;
2977 undo_rx:
2978         WARN_ON(netif_set_real_num_rx_queues(dev, old_rxq));
2979         return err;
2980 }
2981 EXPORT_SYMBOL(netif_set_real_num_queues);
2982
2983 /**
2984  * netif_set_tso_max_size() - set the max size of TSO frames supported
2985  * @dev:        netdev to update
2986  * @size:       max skb->len of a TSO frame
2987  *
2988  * Set the limit on the size of TSO super-frames the device can handle.
2989  * Unless explicitly set the stack will assume the value of
2990  * %GSO_LEGACY_MAX_SIZE.
2991  */
2992 void netif_set_tso_max_size(struct net_device *dev, unsigned int size)
2993 {
2994         dev->tso_max_size = min(GSO_MAX_SIZE, size);
2995         if (size < READ_ONCE(dev->gso_max_size))
2996                 netif_set_gso_max_size(dev, size);
2997         if (size < READ_ONCE(dev->gso_ipv4_max_size))
2998                 netif_set_gso_ipv4_max_size(dev, size);
2999 }
3000 EXPORT_SYMBOL(netif_set_tso_max_size);
3001
3002 /**
3003  * netif_set_tso_max_segs() - set the max number of segs supported for TSO
3004  * @dev:        netdev to update
3005  * @segs:       max number of TCP segments
3006  *
3007  * Set the limit on the number of TCP segments the device can generate from
3008  * a single TSO super-frame.
3009  * Unless explicitly set the stack will assume the value of %GSO_MAX_SEGS.
3010  */
3011 void netif_set_tso_max_segs(struct net_device *dev, unsigned int segs)
3012 {
3013         dev->tso_max_segs = segs;
3014         if (segs < READ_ONCE(dev->gso_max_segs))
3015                 netif_set_gso_max_segs(dev, segs);
3016 }
3017 EXPORT_SYMBOL(netif_set_tso_max_segs);
3018
3019 /**
3020  * netif_inherit_tso_max() - copy all TSO limits from a lower device to an upper
3021  * @to:         netdev to update
3022  * @from:       netdev from which to copy the limits
3023  */
3024 void netif_inherit_tso_max(struct net_device *to, const struct net_device *from)
3025 {
3026         netif_set_tso_max_size(to, from->tso_max_size);
3027         netif_set_tso_max_segs(to, from->tso_max_segs);
3028 }
3029 EXPORT_SYMBOL(netif_inherit_tso_max);
3030
3031 /**
3032  * netif_get_num_default_rss_queues - default number of RSS queues
3033  *
3034  * Default value is the number of physical cores if there are only 1 or 2, or
3035  * divided by 2 if there are more.
3036  */
3037 int netif_get_num_default_rss_queues(void)
3038 {
3039         cpumask_var_t cpus;
3040         int cpu, count = 0;
3041
3042         if (unlikely(is_kdump_kernel() || !zalloc_cpumask_var(&cpus, GFP_KERNEL)))
3043                 return 1;
3044
3045         cpumask_copy(cpus, cpu_online_mask);
3046         for_each_cpu(cpu, cpus) {
3047                 ++count;
3048                 cpumask_andnot(cpus, cpus, topology_sibling_cpumask(cpu));
3049         }
3050         free_cpumask_var(cpus);
3051
3052         return count > 2 ? DIV_ROUND_UP(count, 2) : count;
3053 }
3054 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
3055
3056 static void __netif_reschedule(struct Qdisc *q)
3057 {
3058         struct softnet_data *sd;
3059         unsigned long flags;
3060
3061         local_irq_save(flags);
3062         sd = this_cpu_ptr(&softnet_data);
3063         q->next_sched = NULL;
3064         *sd->output_queue_tailp = q;
3065         sd->output_queue_tailp = &q->next_sched;
3066         raise_softirq_irqoff(NET_TX_SOFTIRQ);
3067         local_irq_restore(flags);
3068 }
3069
3070 void __netif_schedule(struct Qdisc *q)
3071 {
3072         if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
3073                 __netif_reschedule(q);
3074 }
3075 EXPORT_SYMBOL(__netif_schedule);
3076
3077 struct dev_kfree_skb_cb {
3078         enum skb_drop_reason reason;
3079 };
3080
3081 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
3082 {
3083         return (struct dev_kfree_skb_cb *)skb->cb;
3084 }
3085
3086 void netif_schedule_queue(struct netdev_queue *txq)
3087 {
3088         rcu_read_lock();
3089         if (!netif_xmit_stopped(txq)) {
3090                 struct Qdisc *q = rcu_dereference(txq->qdisc);
3091
3092                 __netif_schedule(q);
3093         }
3094         rcu_read_unlock();
3095 }
3096 EXPORT_SYMBOL(netif_schedule_queue);
3097
3098 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
3099 {
3100         if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
3101                 struct Qdisc *q;
3102
3103                 rcu_read_lock();
3104                 q = rcu_dereference(dev_queue->qdisc);
3105                 __netif_schedule(q);
3106                 rcu_read_unlock();
3107         }
3108 }
3109 EXPORT_SYMBOL(netif_tx_wake_queue);
3110
3111 void dev_kfree_skb_irq_reason(struct sk_buff *skb, enum skb_drop_reason reason)
3112 {
3113         unsigned long flags;
3114
3115         if (unlikely(!skb))
3116                 return;
3117
3118         if (likely(refcount_read(&skb->users) == 1)) {
3119                 smp_rmb();
3120                 refcount_set(&skb->users, 0);
3121         } else if (likely(!refcount_dec_and_test(&skb->users))) {
3122                 return;
3123         }
3124         get_kfree_skb_cb(skb)->reason = reason;
3125         local_irq_save(flags);
3126         skb->next = __this_cpu_read(softnet_data.completion_queue);
3127         __this_cpu_write(softnet_data.completion_queue, skb);
3128         raise_softirq_irqoff(NET_TX_SOFTIRQ);
3129         local_irq_restore(flags);
3130 }
3131 EXPORT_SYMBOL(dev_kfree_skb_irq_reason);
3132
3133 void dev_kfree_skb_any_reason(struct sk_buff *skb, enum skb_drop_reason reason)
3134 {
3135         if (in_hardirq() || irqs_disabled())
3136                 dev_kfree_skb_irq_reason(skb, reason);
3137         else
3138                 kfree_skb_reason(skb, reason);
3139 }
3140 EXPORT_SYMBOL(dev_kfree_skb_any_reason);
3141
3142
3143 /**
3144  * netif_device_detach - mark device as removed
3145  * @dev: network device
3146  *
3147  * Mark device as removed from system and therefore no longer available.
3148  */
3149 void netif_device_detach(struct net_device *dev)
3150 {
3151         if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
3152             netif_running(dev)) {
3153                 netif_tx_stop_all_queues(dev);
3154         }
3155 }
3156 EXPORT_SYMBOL(netif_device_detach);
3157
3158 /**
3159  * netif_device_attach - mark device as attached
3160  * @dev: network device
3161  *
3162  * Mark device as attached from system and restart if needed.
3163  */
3164 void netif_device_attach(struct net_device *dev)
3165 {
3166         if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
3167             netif_running(dev)) {
3168                 netif_tx_wake_all_queues(dev);
3169                 __netdev_watchdog_up(dev);
3170         }
3171 }
3172 EXPORT_SYMBOL(netif_device_attach);
3173
3174 /*
3175  * Returns a Tx hash based on the given packet descriptor a Tx queues' number
3176  * to be used as a distribution range.
3177  */
3178 static u16 skb_tx_hash(const struct net_device *dev,
3179                        const struct net_device *sb_dev,
3180                        struct sk_buff *skb)
3181 {
3182         u32 hash;
3183         u16 qoffset = 0;
3184         u16 qcount = dev->real_num_tx_queues;
3185
3186         if (dev->num_tc) {
3187                 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
3188
3189                 qoffset = sb_dev->tc_to_txq[tc].offset;
3190                 qcount = sb_dev->tc_to_txq[tc].count;
3191                 if (unlikely(!qcount)) {
3192                         net_warn_ratelimited("%s: invalid qcount, qoffset %u for tc %u\n",
3193                                              sb_dev->name, qoffset, tc);
3194                         qoffset = 0;
3195                         qcount = dev->real_num_tx_queues;
3196                 }
3197         }
3198
3199         if (skb_rx_queue_recorded(skb)) {
3200                 DEBUG_NET_WARN_ON_ONCE(qcount == 0);
3201                 hash = skb_get_rx_queue(skb);
3202                 if (hash >= qoffset)
3203                         hash -= qoffset;
3204                 while (unlikely(hash >= qcount))
3205                         hash -= qcount;
3206                 return hash + qoffset;
3207         }
3208
3209         return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
3210 }
3211
3212 static void skb_warn_bad_offload(const struct sk_buff *skb)
3213 {
3214         static const netdev_features_t null_features;
3215         struct net_device *dev = skb->dev;
3216         const char *name = "";
3217
3218         if (!net_ratelimit())
3219                 return;
3220
3221         if (dev) {
3222                 if (dev->dev.parent)
3223                         name = dev_driver_string(dev->dev.parent);
3224                 else
3225                         name = netdev_name(dev);
3226         }
3227         skb_dump(KERN_WARNING, skb, false);
3228         WARN(1, "%s: caps=(%pNF, %pNF)\n",
3229              name, dev ? &dev->features : &null_features,
3230              skb->sk ? &skb->sk->sk_route_caps : &null_features);
3231 }
3232
3233 /*
3234  * Invalidate hardware checksum when packet is to be mangled, and
3235  * complete checksum manually on outgoing path.
3236  */
3237 int skb_checksum_help(struct sk_buff *skb)
3238 {
3239         __wsum csum;
3240         int ret = 0, offset;
3241
3242         if (skb->ip_summed == CHECKSUM_COMPLETE)
3243                 goto out_set_summed;
3244
3245         if (unlikely(skb_is_gso(skb))) {
3246                 skb_warn_bad_offload(skb);
3247                 return -EINVAL;
3248         }
3249
3250         /* Before computing a checksum, we should make sure no frag could
3251          * be modified by an external entity : checksum could be wrong.
3252          */
3253         if (skb_has_shared_frag(skb)) {
3254                 ret = __skb_linearize(skb);
3255                 if (ret)
3256                         goto out;
3257         }
3258
3259         offset = skb_checksum_start_offset(skb);
3260         ret = -EINVAL;
3261         if (WARN_ON_ONCE(offset >= skb_headlen(skb))) {
3262                 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false);
3263                 goto out;
3264         }
3265         csum = skb_checksum(skb, offset, skb->len - offset, 0);
3266
3267         offset += skb->csum_offset;
3268         if (WARN_ON_ONCE(offset + sizeof(__sum16) > skb_headlen(skb))) {
3269                 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false);
3270                 goto out;
3271         }
3272         ret = skb_ensure_writable(skb, offset + sizeof(__sum16));
3273         if (ret)
3274                 goto out;
3275
3276         *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0;
3277 out_set_summed:
3278         skb->ip_summed = CHECKSUM_NONE;
3279 out:
3280         return ret;
3281 }
3282 EXPORT_SYMBOL(skb_checksum_help);
3283
3284 int skb_crc32c_csum_help(struct sk_buff *skb)
3285 {
3286         __le32 crc32c_csum;
3287         int ret = 0, offset, start;
3288
3289         if (skb->ip_summed != CHECKSUM_PARTIAL)
3290                 goto out;
3291
3292         if (unlikely(skb_is_gso(skb)))
3293                 goto out;
3294
3295         /* Before computing a checksum, we should make sure no frag could
3296          * be modified by an external entity : checksum could be wrong.
3297          */
3298         if (unlikely(skb_has_shared_frag(skb))) {
3299                 ret = __skb_linearize(skb);
3300                 if (ret)
3301                         goto out;
3302         }
3303         start = skb_checksum_start_offset(skb);
3304         offset = start + offsetof(struct sctphdr, checksum);
3305         if (WARN_ON_ONCE(offset >= skb_headlen(skb))) {
3306                 ret = -EINVAL;
3307                 goto out;
3308         }
3309
3310         ret = skb_ensure_writable(skb, offset + sizeof(__le32));
3311         if (ret)
3312                 goto out;
3313
3314         crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start,
3315                                                   skb->len - start, ~(__u32)0,
3316                                                   crc32c_csum_stub));
3317         *(__le32 *)(skb->data + offset) = crc32c_csum;
3318         skb_reset_csum_not_inet(skb);
3319 out:
3320         return ret;
3321 }
3322
3323 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
3324 {
3325         __be16 type = skb->protocol;
3326
3327         /* Tunnel gso handlers can set protocol to ethernet. */
3328         if (type == htons(ETH_P_TEB)) {
3329                 struct ethhdr *eth;
3330
3331                 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
3332                         return 0;
3333
3334                 eth = (struct ethhdr *)skb->data;
3335                 type = eth->h_proto;
3336         }
3337
3338         return __vlan_get_protocol(skb, type, depth);
3339 }
3340
3341 /* openvswitch calls this on rx path, so we need a different check.
3342  */
3343 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
3344 {
3345         if (tx_path)
3346                 return skb->ip_summed != CHECKSUM_PARTIAL &&
3347                        skb->ip_summed != CHECKSUM_UNNECESSARY;
3348
3349         return skb->ip_summed == CHECKSUM_NONE;
3350 }
3351
3352 /**
3353  *      __skb_gso_segment - Perform segmentation on skb.
3354  *      @skb: buffer to segment
3355  *      @features: features for the output path (see dev->features)
3356  *      @tx_path: whether it is called in TX path
3357  *
3358  *      This function segments the given skb and returns a list of segments.
3359  *
3360  *      It may return NULL if the skb requires no segmentation.  This is
3361  *      only possible when GSO is used for verifying header integrity.
3362  *
3363  *      Segmentation preserves SKB_GSO_CB_OFFSET bytes of previous skb cb.
3364  */
3365 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
3366                                   netdev_features_t features, bool tx_path)
3367 {
3368         struct sk_buff *segs;
3369
3370         if (unlikely(skb_needs_check(skb, tx_path))) {
3371                 int err;
3372
3373                 /* We're going to init ->check field in TCP or UDP header */
3374                 err = skb_cow_head(skb, 0);
3375                 if (err < 0)
3376                         return ERR_PTR(err);
3377         }
3378
3379         /* Only report GSO partial support if it will enable us to
3380          * support segmentation on this frame without needing additional
3381          * work.
3382          */
3383         if (features & NETIF_F_GSO_PARTIAL) {
3384                 netdev_features_t partial_features = NETIF_F_GSO_ROBUST;
3385                 struct net_device *dev = skb->dev;
3386
3387                 partial_features |= dev->features & dev->gso_partial_features;
3388                 if (!skb_gso_ok(skb, features | partial_features))
3389                         features &= ~NETIF_F_GSO_PARTIAL;
3390         }
3391
3392         BUILD_BUG_ON(SKB_GSO_CB_OFFSET +
3393                      sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb));
3394
3395         SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
3396         SKB_GSO_CB(skb)->encap_level = 0;
3397
3398         skb_reset_mac_header(skb);
3399         skb_reset_mac_len(skb);
3400
3401         segs = skb_mac_gso_segment(skb, features);
3402
3403         if (segs != skb && unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs)))
3404                 skb_warn_bad_offload(skb);
3405
3406         return segs;
3407 }
3408 EXPORT_SYMBOL(__skb_gso_segment);
3409
3410 /* Take action when hardware reception checksum errors are detected. */
3411 #ifdef CONFIG_BUG
3412 static void do_netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3413 {
3414         netdev_err(dev, "hw csum failure\n");
3415         skb_dump(KERN_ERR, skb, true);
3416         dump_stack();
3417 }
3418
3419 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3420 {
3421         DO_ONCE_LITE(do_netdev_rx_csum_fault, dev, skb);
3422 }
3423 EXPORT_SYMBOL(netdev_rx_csum_fault);
3424 #endif
3425
3426 /* XXX: check that highmem exists at all on the given machine. */
3427 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
3428 {
3429 #ifdef CONFIG_HIGHMEM
3430         int i;
3431
3432         if (!(dev->features & NETIF_F_HIGHDMA)) {
3433                 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3434                         skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3435
3436                         if (PageHighMem(skb_frag_page(frag)))
3437                                 return 1;
3438                 }
3439         }
3440 #endif
3441         return 0;
3442 }
3443
3444 /* If MPLS offload request, verify we are testing hardware MPLS features
3445  * instead of standard features for the netdev.
3446  */
3447 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
3448 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3449                                            netdev_features_t features,
3450                                            __be16 type)
3451 {
3452         if (eth_p_mpls(type))
3453                 features &= skb->dev->mpls_features;
3454
3455         return features;
3456 }
3457 #else
3458 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3459                                            netdev_features_t features,
3460                                            __be16 type)
3461 {
3462         return features;
3463 }
3464 #endif
3465
3466 static netdev_features_t harmonize_features(struct sk_buff *skb,
3467         netdev_features_t features)
3468 {
3469         __be16 type;
3470
3471         type = skb_network_protocol(skb, NULL);
3472         features = net_mpls_features(skb, features, type);
3473
3474         if (skb->ip_summed != CHECKSUM_NONE &&
3475             !can_checksum_protocol(features, type)) {
3476                 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3477         }
3478         if (illegal_highdma(skb->dev, skb))
3479                 features &= ~NETIF_F_SG;
3480
3481         return features;
3482 }
3483
3484 netdev_features_t passthru_features_check(struct sk_buff *skb,
3485                                           struct net_device *dev,
3486                                           netdev_features_t features)
3487 {
3488         return features;
3489 }
3490 EXPORT_SYMBOL(passthru_features_check);
3491
3492 static netdev_features_t dflt_features_check(struct sk_buff *skb,
3493                                              struct net_device *dev,
3494                                              netdev_features_t features)
3495 {
3496         return vlan_features_check(skb, features);
3497 }
3498
3499 static netdev_features_t gso_features_check(const struct sk_buff *skb,
3500                                             struct net_device *dev,
3501                                             netdev_features_t features)
3502 {
3503         u16 gso_segs = skb_shinfo(skb)->gso_segs;
3504
3505         if (gso_segs > READ_ONCE(dev->gso_max_segs))
3506                 return features & ~NETIF_F_GSO_MASK;
3507
3508         if (!skb_shinfo(skb)->gso_type) {
3509                 skb_warn_bad_offload(skb);
3510                 return features & ~NETIF_F_GSO_MASK;
3511         }
3512
3513         /* Support for GSO partial features requires software
3514          * intervention before we can actually process the packets
3515          * so we need to strip support for any partial features now
3516          * and we can pull them back in after we have partially
3517          * segmented the frame.
3518          */
3519         if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL))
3520                 features &= ~dev->gso_partial_features;
3521
3522         /* Make sure to clear the IPv4 ID mangling feature if the
3523          * IPv4 header has the potential to be fragmented.
3524          */
3525         if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
3526                 struct iphdr *iph = skb->encapsulation ?
3527                                     inner_ip_hdr(skb) : ip_hdr(skb);
3528
3529                 if (!(iph->frag_off & htons(IP_DF)))
3530                         features &= ~NETIF_F_TSO_MANGLEID;
3531         }
3532
3533         return features;
3534 }
3535
3536 netdev_features_t netif_skb_features(struct sk_buff *skb)
3537 {
3538         struct net_device *dev = skb->dev;
3539         netdev_features_t features = dev->features;
3540
3541         if (skb_is_gso(skb))
3542                 features = gso_features_check(skb, dev, features);
3543
3544         /* If encapsulation offload request, verify we are testing
3545          * hardware encapsulation features instead of standard
3546          * features for the netdev
3547          */
3548         if (skb->encapsulation)
3549                 features &= dev->hw_enc_features;
3550
3551         if (skb_vlan_tagged(skb))
3552                 features = netdev_intersect_features(features,
3553                                                      dev->vlan_features |
3554                                                      NETIF_F_HW_VLAN_CTAG_TX |
3555                                                      NETIF_F_HW_VLAN_STAG_TX);
3556
3557         if (dev->netdev_ops->ndo_features_check)
3558                 features &= dev->netdev_ops->ndo_features_check(skb, dev,
3559                                                                 features);
3560         else
3561                 features &= dflt_features_check(skb, dev, features);
3562
3563         return harmonize_features(skb, features);
3564 }
3565 EXPORT_SYMBOL(netif_skb_features);
3566
3567 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
3568                     struct netdev_queue *txq, bool more)
3569 {
3570         unsigned int len;
3571         int rc;
3572
3573         if (dev_nit_active(dev))
3574                 dev_queue_xmit_nit(skb, dev);
3575
3576         len = skb->len;
3577         trace_net_dev_start_xmit(skb, dev);
3578         rc = netdev_start_xmit(skb, dev, txq, more);
3579         trace_net_dev_xmit(skb, rc, dev, len);
3580
3581         return rc;
3582 }
3583
3584 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
3585                                     struct netdev_queue *txq, int *ret)
3586 {
3587         struct sk_buff *skb = first;
3588         int rc = NETDEV_TX_OK;
3589
3590         while (skb) {
3591                 struct sk_buff *next = skb->next;
3592
3593                 skb_mark_not_on_list(skb);
3594                 rc = xmit_one(skb, dev, txq, next != NULL);
3595                 if (unlikely(!dev_xmit_complete(rc))) {
3596                         skb->next = next;
3597                         goto out;
3598                 }
3599
3600                 skb = next;
3601                 if (netif_tx_queue_stopped(txq) && skb) {
3602                         rc = NETDEV_TX_BUSY;
3603                         break;
3604                 }
3605         }
3606
3607 out:
3608         *ret = rc;
3609         return skb;
3610 }
3611
3612 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
3613                                           netdev_features_t features)
3614 {
3615         if (skb_vlan_tag_present(skb) &&
3616             !vlan_hw_offload_capable(features, skb->vlan_proto))
3617                 skb = __vlan_hwaccel_push_inside(skb);
3618         return skb;
3619 }
3620
3621 int skb_csum_hwoffload_help(struct sk_buff *skb,
3622                             const netdev_features_t features)
3623 {
3624         if (unlikely(skb_csum_is_sctp(skb)))
3625                 return !!(features & NETIF_F_SCTP_CRC) ? 0 :
3626                         skb_crc32c_csum_help(skb);
3627
3628         if (features & NETIF_F_HW_CSUM)
3629                 return 0;
3630
3631         if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) {
3632                 switch (skb->csum_offset) {
3633                 case offsetof(struct tcphdr, check):
3634                 case offsetof(struct udphdr, check):
3635                         return 0;
3636                 }
3637         }
3638
3639         return skb_checksum_help(skb);
3640 }
3641 EXPORT_SYMBOL(skb_csum_hwoffload_help);
3642
3643 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again)
3644 {
3645         netdev_features_t features;
3646
3647         features = netif_skb_features(skb);
3648         skb = validate_xmit_vlan(skb, features);
3649         if (unlikely(!skb))
3650                 goto out_null;
3651
3652         skb = sk_validate_xmit_skb(skb, dev);
3653         if (unlikely(!skb))
3654                 goto out_null;
3655
3656         if (netif_needs_gso(skb, features)) {
3657                 struct sk_buff *segs;
3658
3659                 segs = skb_gso_segment(skb, features);
3660                 if (IS_ERR(segs)) {
3661                         goto out_kfree_skb;
3662                 } else if (segs) {
3663                         consume_skb(skb);
3664                         skb = segs;
3665                 }
3666         } else {
3667                 if (skb_needs_linearize(skb, features) &&
3668                     __skb_linearize(skb))
3669                         goto out_kfree_skb;
3670
3671                 /* If packet is not checksummed and device does not
3672                  * support checksumming for this protocol, complete
3673                  * checksumming here.
3674                  */
3675                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3676                         if (skb->encapsulation)
3677                                 skb_set_inner_transport_header(skb,
3678                                                                skb_checksum_start_offset(skb));
3679                         else
3680                                 skb_set_transport_header(skb,
3681                                                          skb_checksum_start_offset(skb));
3682                         if (skb_csum_hwoffload_help(skb, features))
3683                                 goto out_kfree_skb;
3684                 }
3685         }
3686
3687         skb = validate_xmit_xfrm(skb, features, again);
3688
3689         return skb;
3690
3691 out_kfree_skb:
3692         kfree_skb(skb);
3693 out_null:
3694         dev_core_stats_tx_dropped_inc(dev);
3695         return NULL;
3696 }
3697
3698 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again)
3699 {
3700         struct sk_buff *next, *head = NULL, *tail;
3701
3702         for (; skb != NULL; skb = next) {
3703                 next = skb->next;
3704                 skb_mark_not_on_list(skb);
3705
3706                 /* in case skb wont be segmented, point to itself */
3707                 skb->prev = skb;
3708
3709                 skb = validate_xmit_skb(skb, dev, again);
3710                 if (!skb)
3711                         continue;
3712
3713                 if (!head)
3714                         head = skb;
3715                 else
3716                         tail->next = skb;
3717                 /* If skb was segmented, skb->prev points to
3718                  * the last segment. If not, it still contains skb.
3719                  */
3720                 tail = skb->prev;
3721         }
3722         return head;
3723 }
3724 EXPORT_SYMBOL_GPL(validate_xmit_skb_list);
3725
3726 static void qdisc_pkt_len_init(struct sk_buff *skb)
3727 {
3728         const struct skb_shared_info *shinfo = skb_shinfo(skb);
3729
3730         qdisc_skb_cb(skb)->pkt_len = skb->len;
3731
3732         /* To get more precise estimation of bytes sent on wire,
3733          * we add to pkt_len the headers size of all segments
3734          */
3735         if (shinfo->gso_size && skb_transport_header_was_set(skb)) {
3736                 u16 gso_segs = shinfo->gso_segs;
3737                 unsigned int hdr_len;
3738
3739                 /* mac layer + network layer */
3740                 hdr_len = skb_transport_offset(skb);
3741
3742                 /* + transport layer */
3743                 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
3744                         const struct tcphdr *th;
3745                         struct tcphdr _tcphdr;
3746
3747                         th = skb_header_pointer(skb, hdr_len,
3748                                                 sizeof(_tcphdr), &_tcphdr);
3749                         if (likely(th))
3750                                 hdr_len += __tcp_hdrlen(th);
3751                 } else {
3752                         struct udphdr _udphdr;
3753
3754                         if (skb_header_pointer(skb, hdr_len,
3755                                                sizeof(_udphdr), &_udphdr))
3756                                 hdr_len += sizeof(struct udphdr);
3757                 }
3758
3759                 if (shinfo->gso_type & SKB_GSO_DODGY)
3760                         gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
3761                                                 shinfo->gso_size);
3762
3763                 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
3764         }
3765 }
3766
3767 static int dev_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *q,
3768                              struct sk_buff **to_free,
3769                              struct netdev_queue *txq)
3770 {
3771         int rc;
3772
3773         rc = q->enqueue(skb, q, to_free) & NET_XMIT_MASK;
3774         if (rc == NET_XMIT_SUCCESS)
3775                 trace_qdisc_enqueue(q, txq, skb);
3776         return rc;
3777 }
3778
3779 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
3780                                  struct net_device *dev,
3781                                  struct netdev_queue *txq)
3782 {
3783         spinlock_t *root_lock = qdisc_lock(q);
3784         struct sk_buff *to_free = NULL;
3785         bool contended;
3786         int rc;
3787
3788         qdisc_calculate_pkt_len(skb, q);
3789
3790         if (q->flags & TCQ_F_NOLOCK) {
3791                 if (q->flags & TCQ_F_CAN_BYPASS && nolock_qdisc_is_empty(q) &&
3792                     qdisc_run_begin(q)) {
3793                         /* Retest nolock_qdisc_is_empty() within the protection
3794                          * of q->seqlock to protect from racing with requeuing.
3795                          */
3796                         if (unlikely(!nolock_qdisc_is_empty(q))) {
3797                                 rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3798                                 __qdisc_run(q);
3799                                 qdisc_run_end(q);
3800
3801                                 goto no_lock_out;
3802                         }
3803
3804                         qdisc_bstats_cpu_update(q, skb);
3805                         if (sch_direct_xmit(skb, q, dev, txq, NULL, true) &&
3806                             !nolock_qdisc_is_empty(q))
3807                                 __qdisc_run(q);
3808
3809                         qdisc_run_end(q);
3810                         return NET_XMIT_SUCCESS;
3811                 }
3812
3813                 rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3814                 qdisc_run(q);
3815
3816 no_lock_out:
3817                 if (unlikely(to_free))
3818                         kfree_skb_list_reason(to_free,
3819                                               SKB_DROP_REASON_QDISC_DROP);
3820                 return rc;
3821         }
3822
3823         /*
3824          * Heuristic to force contended enqueues to serialize on a
3825          * separate lock before trying to get qdisc main lock.
3826          * This permits qdisc->running owner to get the lock more
3827          * often and dequeue packets faster.
3828          * On PREEMPT_RT it is possible to preempt the qdisc owner during xmit
3829          * and then other tasks will only enqueue packets. The packets will be
3830          * sent after the qdisc owner is scheduled again. To prevent this
3831          * scenario the task always serialize on the lock.
3832          */
3833         contended = qdisc_is_running(q) || IS_ENABLED(CONFIG_PREEMPT_RT);
3834         if (unlikely(contended))
3835                 spin_lock(&q->busylock);
3836
3837         spin_lock(root_lock);
3838         if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3839                 __qdisc_drop(skb, &to_free);
3840                 rc = NET_XMIT_DROP;
3841         } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
3842                    qdisc_run_begin(q)) {
3843                 /*
3844                  * This is a work-conserving queue; there are no old skbs
3845                  * waiting to be sent out; and the qdisc is not running -
3846                  * xmit the skb directly.
3847                  */
3848
3849                 qdisc_bstats_update(q, skb);
3850
3851                 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
3852                         if (unlikely(contended)) {
3853                                 spin_unlock(&q->busylock);
3854                                 contended = false;
3855                         }
3856                         __qdisc_run(q);
3857                 }
3858
3859                 qdisc_run_end(q);
3860                 rc = NET_XMIT_SUCCESS;
3861         } else {
3862                 rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3863                 if (qdisc_run_begin(q)) {
3864                         if (unlikely(contended)) {
3865                                 spin_unlock(&q->busylock);
3866                                 contended = false;
3867                         }
3868                         __qdisc_run(q);
3869                         qdisc_run_end(q);
3870                 }
3871         }
3872         spin_unlock(root_lock);
3873         if (unlikely(to_free))
3874                 kfree_skb_list_reason(to_free, SKB_DROP_REASON_QDISC_DROP);
3875         if (unlikely(contended))
3876                 spin_unlock(&q->busylock);
3877         return rc;
3878 }
3879
3880 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
3881 static void skb_update_prio(struct sk_buff *skb)
3882 {
3883         const struct netprio_map *map;
3884         const struct sock *sk;
3885         unsigned int prioidx;
3886
3887         if (skb->priority)
3888                 return;
3889         map = rcu_dereference_bh(skb->dev->priomap);
3890         if (!map)
3891                 return;
3892         sk = skb_to_full_sk(skb);
3893         if (!sk)
3894                 return;
3895
3896         prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data);
3897
3898         if (prioidx < map->priomap_len)
3899                 skb->priority = map->priomap[prioidx];
3900 }
3901 #else
3902 #define skb_update_prio(skb)
3903 #endif
3904
3905 /**
3906  *      dev_loopback_xmit - loop back @skb
3907  *      @net: network namespace this loopback is happening in
3908  *      @sk:  sk needed to be a netfilter okfn
3909  *      @skb: buffer to transmit
3910  */
3911 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
3912 {
3913         skb_reset_mac_header(skb);
3914         __skb_pull(skb, skb_network_offset(skb));
3915         skb->pkt_type = PACKET_LOOPBACK;
3916         if (skb->ip_summed == CHECKSUM_NONE)
3917                 skb->ip_summed = CHECKSUM_UNNECESSARY;
3918         DEBUG_NET_WARN_ON_ONCE(!skb_dst(skb));
3919         skb_dst_force(skb);
3920         netif_rx(skb);
3921         return 0;
3922 }
3923 EXPORT_SYMBOL(dev_loopback_xmit);
3924
3925 #ifdef CONFIG_NET_EGRESS
3926 static struct sk_buff *
3927 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
3928 {
3929 #ifdef CONFIG_NET_CLS_ACT
3930         struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress);
3931         struct tcf_result cl_res;
3932
3933         if (!miniq)
3934                 return skb;
3935
3936         /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */
3937         tc_skb_cb(skb)->mru = 0;
3938         tc_skb_cb(skb)->post_ct = false;
3939         mini_qdisc_bstats_cpu_update(miniq, skb);
3940
3941         switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) {
3942         case TC_ACT_OK:
3943         case TC_ACT_RECLASSIFY:
3944                 skb->tc_index = TC_H_MIN(cl_res.classid);
3945                 break;
3946         case TC_ACT_SHOT:
3947                 mini_qdisc_qstats_cpu_drop(miniq);
3948                 *ret = NET_XMIT_DROP;
3949                 kfree_skb_reason(skb, SKB_DROP_REASON_TC_EGRESS);
3950                 return NULL;
3951         case TC_ACT_STOLEN:
3952         case TC_ACT_QUEUED:
3953         case TC_ACT_TRAP:
3954                 *ret = NET_XMIT_SUCCESS;
3955                 consume_skb(skb);
3956                 return NULL;
3957         case TC_ACT_REDIRECT:
3958                 /* No need to push/pop skb's mac_header here on egress! */
3959                 skb_do_redirect(skb);
3960                 *ret = NET_XMIT_SUCCESS;
3961                 return NULL;
3962         default:
3963                 break;
3964         }
3965 #endif /* CONFIG_NET_CLS_ACT */
3966
3967         return skb;
3968 }
3969
3970 static struct netdev_queue *
3971 netdev_tx_queue_mapping(struct net_device *dev, struct sk_buff *skb)
3972 {
3973         int qm = skb_get_queue_mapping(skb);
3974
3975         return netdev_get_tx_queue(dev, netdev_cap_txqueue(dev, qm));
3976 }
3977
3978 static bool netdev_xmit_txqueue_skipped(void)
3979 {
3980         return __this_cpu_read(softnet_data.xmit.skip_txqueue);
3981 }
3982
3983 void netdev_xmit_skip_txqueue(bool skip)
3984 {
3985         __this_cpu_write(softnet_data.xmit.skip_txqueue, skip);
3986 }
3987 EXPORT_SYMBOL_GPL(netdev_xmit_skip_txqueue);
3988 #endif /* CONFIG_NET_EGRESS */
3989
3990 #ifdef CONFIG_XPS
3991 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb,
3992                                struct xps_dev_maps *dev_maps, unsigned int tci)
3993 {
3994         int tc = netdev_get_prio_tc_map(dev, skb->priority);
3995         struct xps_map *map;
3996         int queue_index = -1;
3997
3998         if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids)
3999                 return queue_index;
4000
4001         tci *= dev_maps->num_tc;
4002         tci += tc;
4003
4004         map = rcu_dereference(dev_maps->attr_map[tci]);
4005         if (map) {
4006                 if (map->len == 1)
4007                         queue_index = map->queues[0];
4008                 else
4009                         queue_index = map->queues[reciprocal_scale(
4010                                                 skb_get_hash(skb), map->len)];
4011                 if (unlikely(queue_index >= dev->real_num_tx_queues))
4012                         queue_index = -1;
4013         }
4014         return queue_index;
4015 }
4016 #endif
4017
4018 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev,
4019                          struct sk_buff *skb)
4020 {
4021 #ifdef CONFIG_XPS
4022         struct xps_dev_maps *dev_maps;
4023         struct sock *sk = skb->sk;
4024         int queue_index = -1;
4025
4026         if (!static_key_false(&xps_needed))
4027                 return -1;
4028
4029         rcu_read_lock();
4030         if (!static_key_false(&xps_rxqs_needed))
4031                 goto get_cpus_map;
4032
4033         dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]);
4034         if (dev_maps) {
4035                 int tci = sk_rx_queue_get(sk);
4036
4037                 if (tci >= 0)
4038                         queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
4039                                                           tci);
4040         }
4041
4042 get_cpus_map:
4043         if (queue_index < 0) {
4044                 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]);
4045                 if (dev_maps) {
4046                         unsigned int tci = skb->sender_cpu - 1;
4047
4048                         queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
4049                                                           tci);
4050                 }
4051         }
4052         rcu_read_unlock();
4053
4054         return queue_index;
4055 #else
4056         return -1;
4057 #endif
4058 }
4059
4060 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb,
4061                      struct net_device *sb_dev)
4062 {
4063         return 0;
4064 }
4065 EXPORT_SYMBOL(dev_pick_tx_zero);
4066
4067 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb,
4068                        struct net_device *sb_dev)
4069 {
4070         return (u16)raw_smp_processor_id() % dev->real_num_tx_queues;
4071 }
4072 EXPORT_SYMBOL(dev_pick_tx_cpu_id);
4073
4074 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb,
4075                      struct net_device *sb_dev)
4076 {
4077         struct sock *sk = skb->sk;
4078         int queue_index = sk_tx_queue_get(sk);
4079
4080         sb_dev = sb_dev ? : dev;
4081
4082         if (queue_index < 0 || skb->ooo_okay ||
4083             queue_index >= dev->real_num_tx_queues) {
4084                 int new_index = get_xps_queue(dev, sb_dev, skb);
4085
4086                 if (new_index < 0)
4087                         new_index = skb_tx_hash(dev, sb_dev, skb);
4088
4089                 if (queue_index != new_index && sk &&
4090                     sk_fullsock(sk) &&
4091                     rcu_access_pointer(sk->sk_dst_cache))
4092                         sk_tx_queue_set(sk, new_index);
4093
4094                 queue_index = new_index;
4095         }
4096
4097         return queue_index;
4098 }
4099 EXPORT_SYMBOL(netdev_pick_tx);
4100
4101 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev,
4102                                          struct sk_buff *skb,
4103                                          struct net_device *sb_dev)
4104 {
4105         int queue_index = 0;
4106
4107 #ifdef CONFIG_XPS
4108         u32 sender_cpu = skb->sender_cpu - 1;
4109
4110         if (sender_cpu >= (u32)NR_CPUS)
4111                 skb->sender_cpu = raw_smp_processor_id() + 1;
4112 #endif
4113
4114         if (dev->real_num_tx_queues != 1) {
4115                 const struct net_device_ops *ops = dev->netdev_ops;
4116
4117                 if (ops->ndo_select_queue)
4118                         queue_index = ops->ndo_select_queue(dev, skb, sb_dev);
4119                 else
4120                         queue_index = netdev_pick_tx(dev, skb, sb_dev);
4121
4122                 queue_index = netdev_cap_txqueue(dev, queue_index);
4123         }
4124
4125         skb_set_queue_mapping(skb, queue_index);
4126         return netdev_get_tx_queue(dev, queue_index);
4127 }
4128
4129 /**
4130  * __dev_queue_xmit() - transmit a buffer
4131  * @skb:        buffer to transmit
4132  * @sb_dev:     suboordinate device used for L2 forwarding offload
4133  *
4134  * Queue a buffer for transmission to a network device. The caller must
4135  * have set the device and priority and built the buffer before calling
4136  * this function. The function can be called from an interrupt.
4137  *
4138  * When calling this method, interrupts MUST be enabled. This is because
4139  * the BH enable code must have IRQs enabled so that it will not deadlock.
4140  *
4141  * Regardless of the return value, the skb is consumed, so it is currently
4142  * difficult to retry a send to this method. (You can bump the ref count
4143  * before sending to hold a reference for retry if you are careful.)
4144  *
4145  * Return:
4146  * * 0                          - buffer successfully transmitted
4147  * * positive qdisc return code - NET_XMIT_DROP etc.
4148  * * negative errno             - other errors
4149  */
4150 int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev)
4151 {
4152         struct net_device *dev = skb->dev;
4153         struct netdev_queue *txq = NULL;
4154         struct Qdisc *q;
4155         int rc = -ENOMEM;
4156         bool again = false;
4157
4158         skb_reset_mac_header(skb);
4159         skb_assert_len(skb);
4160
4161         if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
4162                 __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED);
4163
4164         /* Disable soft irqs for various locks below. Also
4165          * stops preemption for RCU.
4166          */
4167         rcu_read_lock_bh();
4168
4169         skb_update_prio(skb);
4170
4171         qdisc_pkt_len_init(skb);
4172 #ifdef CONFIG_NET_CLS_ACT
4173         skb->tc_at_ingress = 0;
4174 #endif
4175 #ifdef CONFIG_NET_EGRESS
4176         if (static_branch_unlikely(&egress_needed_key)) {
4177                 if (nf_hook_egress_active()) {
4178                         skb = nf_hook_egress(skb, &rc, dev);
4179                         if (!skb)
4180                                 goto out;
4181                 }
4182
4183                 netdev_xmit_skip_txqueue(false);
4184
4185                 nf_skip_egress(skb, true);
4186                 skb = sch_handle_egress(skb, &rc, dev);
4187                 if (!skb)
4188                         goto out;
4189                 nf_skip_egress(skb, false);
4190
4191                 if (netdev_xmit_txqueue_skipped())
4192                         txq = netdev_tx_queue_mapping(dev, skb);
4193         }
4194 #endif
4195         /* If device/qdisc don't need skb->dst, release it right now while
4196          * its hot in this cpu cache.
4197          */
4198         if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
4199                 skb_dst_drop(skb);
4200         else
4201                 skb_dst_force(skb);
4202
4203         if (!txq)
4204                 txq = netdev_core_pick_tx(dev, skb, sb_dev);
4205
4206         q = rcu_dereference_bh(txq->qdisc);
4207
4208         trace_net_dev_queue(skb);
4209         if (q->enqueue) {
4210                 rc = __dev_xmit_skb(skb, q, dev, txq);
4211                 goto out;
4212         }
4213
4214         /* The device has no queue. Common case for software devices:
4215          * loopback, all the sorts of tunnels...
4216
4217          * Really, it is unlikely that netif_tx_lock protection is necessary
4218          * here.  (f.e. loopback and IP tunnels are clean ignoring statistics
4219          * counters.)
4220          * However, it is possible, that they rely on protection
4221          * made by us here.
4222
4223          * Check this and shot the lock. It is not prone from deadlocks.
4224          *Either shot noqueue qdisc, it is even simpler 8)
4225          */
4226         if (dev->flags & IFF_UP) {
4227                 int cpu = smp_processor_id(); /* ok because BHs are off */
4228
4229                 /* Other cpus might concurrently change txq->xmit_lock_owner
4230                  * to -1 or to their cpu id, but not to our id.
4231                  */
4232                 if (READ_ONCE(txq->xmit_lock_owner) != cpu) {
4233                         if (dev_xmit_recursion())
4234                                 goto recursion_alert;
4235
4236                         skb = validate_xmit_skb(skb, dev, &again);
4237                         if (!skb)
4238                                 goto out;
4239
4240                         HARD_TX_LOCK(dev, txq, cpu);
4241
4242                         if (!netif_xmit_stopped(txq)) {
4243                                 dev_xmit_recursion_inc();
4244                                 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
4245                                 dev_xmit_recursion_dec();
4246                                 if (dev_xmit_complete(rc)) {
4247                                         HARD_TX_UNLOCK(dev, txq);
4248                                         goto out;
4249                                 }
4250                         }
4251                         HARD_TX_UNLOCK(dev, txq);
4252                         net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
4253                                              dev->name);
4254                 } else {
4255                         /* Recursion is detected! It is possible,
4256                          * unfortunately
4257                          */
4258 recursion_alert:
4259                         net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
4260                                              dev->name);
4261                 }
4262         }
4263
4264         rc = -ENETDOWN;
4265         rcu_read_unlock_bh();
4266
4267         dev_core_stats_tx_dropped_inc(dev);
4268         kfree_skb_list(skb);
4269         return rc;
4270 out:
4271         rcu_read_unlock_bh();
4272         return rc;
4273 }
4274 EXPORT_SYMBOL(__dev_queue_xmit);
4275
4276 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id)
4277 {
4278         struct net_device *dev = skb->dev;
4279         struct sk_buff *orig_skb = skb;
4280         struct netdev_queue *txq;
4281         int ret = NETDEV_TX_BUSY;
4282         bool again = false;
4283
4284         if (unlikely(!netif_running(dev) ||
4285                      !netif_carrier_ok(dev)))
4286                 goto drop;
4287
4288         skb = validate_xmit_skb_list(skb, dev, &again);
4289         if (skb != orig_skb)
4290                 goto drop;
4291
4292         skb_set_queue_mapping(skb, queue_id);
4293         txq = skb_get_tx_queue(dev, skb);
4294
4295         local_bh_disable();
4296
4297         dev_xmit_recursion_inc();
4298         HARD_TX_LOCK(dev, txq, smp_processor_id());
4299         if (!netif_xmit_frozen_or_drv_stopped(txq))
4300                 ret = netdev_start_xmit(skb, dev, txq, false);
4301         HARD_TX_UNLOCK(dev, txq);
4302         dev_xmit_recursion_dec();
4303
4304         local_bh_enable();
4305         return ret;
4306 drop:
4307         dev_core_stats_tx_dropped_inc(dev);
4308         kfree_skb_list(skb);
4309         return NET_XMIT_DROP;
4310 }
4311 EXPORT_SYMBOL(__dev_direct_xmit);
4312
4313 /*************************************************************************
4314  *                      Receiver routines
4315  *************************************************************************/
4316
4317 int netdev_max_backlog __read_mostly = 1000;
4318 EXPORT_SYMBOL(netdev_max_backlog);
4319
4320 int netdev_tstamp_prequeue __read_mostly = 1;
4321 unsigned int sysctl_skb_defer_max __read_mostly = 64;
4322 int netdev_budget __read_mostly = 300;
4323 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */
4324 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ;
4325 int weight_p __read_mostly = 64;           /* old backlog weight */
4326 int dev_weight_rx_bias __read_mostly = 1;  /* bias for backlog weight */
4327 int dev_weight_tx_bias __read_mostly = 1;  /* bias for output_queue quota */
4328 int dev_rx_weight __read_mostly = 64;
4329 int dev_tx_weight __read_mostly = 64;
4330
4331 /* Called with irq disabled */
4332 static inline void ____napi_schedule(struct softnet_data *sd,
4333                                      struct napi_struct *napi)
4334 {
4335         struct task_struct *thread;
4336
4337         lockdep_assert_irqs_disabled();
4338
4339         if (test_bit(NAPI_STATE_THREADED, &napi->state)) {
4340                 /* Paired with smp_mb__before_atomic() in
4341                  * napi_enable()/dev_set_threaded().
4342                  * Use READ_ONCE() to guarantee a complete
4343                  * read on napi->thread. Only call
4344                  * wake_up_process() when it's not NULL.
4345                  */
4346                 thread = READ_ONCE(napi->thread);
4347                 if (thread) {
4348                         /* Avoid doing set_bit() if the thread is in
4349                          * INTERRUPTIBLE state, cause napi_thread_wait()
4350                          * makes sure to proceed with napi polling
4351                          * if the thread is explicitly woken from here.
4352                          */
4353                         if (READ_ONCE(thread->__state) != TASK_INTERRUPTIBLE)
4354                                 set_bit(NAPI_STATE_SCHED_THREADED, &napi->state);
4355                         wake_up_process(thread);
4356                         return;
4357                 }
4358         }
4359
4360         list_add_tail(&napi->poll_list, &sd->poll_list);
4361         WRITE_ONCE(napi->list_owner, smp_processor_id());
4362         /* If not called from net_rx_action()
4363          * we have to raise NET_RX_SOFTIRQ.
4364          */
4365         if (!sd->in_net_rx_action)
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  * After we queued a packet into sd->input_pkt_queue,
4589  * we need to make sure this queue is serviced soon.
4590  *
4591  * - If this is another cpu queue, link it to our rps_ipi_list,
4592  *   and make sure we will process rps_ipi_list from net_rx_action().
4593  *
4594  * - If this is our own queue, NAPI schedule our backlog.
4595  *   Note that this also raises NET_RX_SOFTIRQ.
4596  */
4597 static void napi_schedule_rps(struct softnet_data *sd)
4598 {
4599         struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
4600
4601 #ifdef CONFIG_RPS
4602         if (sd != mysd) {
4603                 sd->rps_ipi_next = mysd->rps_ipi_list;
4604                 mysd->rps_ipi_list = sd;
4605
4606                 /* If not called from net_rx_action()
4607                  * we have to raise NET_RX_SOFTIRQ.
4608                  */
4609                 if (!mysd->in_net_rx_action)
4610                         __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4611                 return;
4612         }
4613 #endif /* CONFIG_RPS */
4614         __napi_schedule_irqoff(&mysd->backlog);
4615 }
4616
4617 #ifdef CONFIG_NET_FLOW_LIMIT
4618 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
4619 #endif
4620
4621 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
4622 {
4623 #ifdef CONFIG_NET_FLOW_LIMIT
4624         struct sd_flow_limit *fl;
4625         struct softnet_data *sd;
4626         unsigned int old_flow, new_flow;
4627
4628         if (qlen < (READ_ONCE(netdev_max_backlog) >> 1))
4629                 return false;
4630
4631         sd = this_cpu_ptr(&softnet_data);
4632
4633         rcu_read_lock();
4634         fl = rcu_dereference(sd->flow_limit);
4635         if (fl) {
4636                 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
4637                 old_flow = fl->history[fl->history_head];
4638                 fl->history[fl->history_head] = new_flow;
4639
4640                 fl->history_head++;
4641                 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
4642
4643                 if (likely(fl->buckets[old_flow]))
4644                         fl->buckets[old_flow]--;
4645
4646                 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
4647                         fl->count++;
4648                         rcu_read_unlock();
4649                         return true;
4650                 }
4651         }
4652         rcu_read_unlock();
4653 #endif
4654         return false;
4655 }
4656
4657 /*
4658  * enqueue_to_backlog is called to queue an skb to a per CPU backlog
4659  * queue (may be a remote CPU queue).
4660  */
4661 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
4662                               unsigned int *qtail)
4663 {
4664         enum skb_drop_reason reason;
4665         struct softnet_data *sd;
4666         unsigned long flags;
4667         unsigned int qlen;
4668
4669         reason = SKB_DROP_REASON_NOT_SPECIFIED;
4670         sd = &per_cpu(softnet_data, cpu);
4671
4672         rps_lock_irqsave(sd, &flags);
4673         if (!netif_running(skb->dev))
4674                 goto drop;
4675         qlen = skb_queue_len(&sd->input_pkt_queue);
4676         if (qlen <= READ_ONCE(netdev_max_backlog) && !skb_flow_limit(skb, qlen)) {
4677                 if (qlen) {
4678 enqueue:
4679                         __skb_queue_tail(&sd->input_pkt_queue, skb);
4680                         input_queue_tail_incr_save(sd, qtail);
4681                         rps_unlock_irq_restore(sd, &flags);
4682                         return NET_RX_SUCCESS;
4683                 }
4684
4685                 /* Schedule NAPI for backlog device
4686                  * We can use non atomic operation since we own the queue lock
4687                  */
4688                 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state))
4689                         napi_schedule_rps(sd);
4690                 goto enqueue;
4691         }
4692         reason = SKB_DROP_REASON_CPU_BACKLOG;
4693
4694 drop:
4695         sd->dropped++;
4696         rps_unlock_irq_restore(sd, &flags);
4697
4698         dev_core_stats_rx_dropped_inc(skb->dev);
4699         kfree_skb_reason(skb, reason);
4700         return NET_RX_DROP;
4701 }
4702
4703 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb)
4704 {
4705         struct net_device *dev = skb->dev;
4706         struct netdev_rx_queue *rxqueue;
4707
4708         rxqueue = dev->_rx;
4709
4710         if (skb_rx_queue_recorded(skb)) {
4711                 u16 index = skb_get_rx_queue(skb);
4712
4713                 if (unlikely(index >= dev->real_num_rx_queues)) {
4714                         WARN_ONCE(dev->real_num_rx_queues > 1,
4715                                   "%s received packet on queue %u, but number "
4716                                   "of RX queues is %u\n",
4717                                   dev->name, index, dev->real_num_rx_queues);
4718
4719                         return rxqueue; /* Return first rxqueue */
4720                 }
4721                 rxqueue += index;
4722         }
4723         return rxqueue;
4724 }
4725
4726 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp,
4727                              struct bpf_prog *xdp_prog)
4728 {
4729         void *orig_data, *orig_data_end, *hard_start;
4730         struct netdev_rx_queue *rxqueue;
4731         bool orig_bcast, orig_host;
4732         u32 mac_len, frame_sz;
4733         __be16 orig_eth_type;
4734         struct ethhdr *eth;
4735         u32 metalen, act;
4736         int off;
4737
4738         /* The XDP program wants to see the packet starting at the MAC
4739          * header.
4740          */
4741         mac_len = skb->data - skb_mac_header(skb);
4742         hard_start = skb->data - skb_headroom(skb);
4743
4744         /* SKB "head" area always have tailroom for skb_shared_info */
4745         frame_sz = (void *)skb_end_pointer(skb) - hard_start;
4746         frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
4747
4748         rxqueue = netif_get_rxqueue(skb);
4749         xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq);
4750         xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len,
4751                          skb_headlen(skb) + mac_len, true);
4752
4753         orig_data_end = xdp->data_end;
4754         orig_data = xdp->data;
4755         eth = (struct ethhdr *)xdp->data;
4756         orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr);
4757         orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest);
4758         orig_eth_type = eth->h_proto;
4759
4760         act = bpf_prog_run_xdp(xdp_prog, xdp);
4761
4762         /* check if bpf_xdp_adjust_head was used */
4763         off = xdp->data - orig_data;
4764         if (off) {
4765                 if (off > 0)
4766                         __skb_pull(skb, off);
4767                 else if (off < 0)
4768                         __skb_push(skb, -off);
4769
4770                 skb->mac_header += off;
4771                 skb_reset_network_header(skb);
4772         }
4773
4774         /* check if bpf_xdp_adjust_tail was used */
4775         off = xdp->data_end - orig_data_end;
4776         if (off != 0) {
4777                 skb_set_tail_pointer(skb, xdp->data_end - xdp->data);
4778                 skb->len += off; /* positive on grow, negative on shrink */
4779         }
4780
4781         /* check if XDP changed eth hdr such SKB needs update */
4782         eth = (struct ethhdr *)xdp->data;
4783         if ((orig_eth_type != eth->h_proto) ||
4784             (orig_host != ether_addr_equal_64bits(eth->h_dest,
4785                                                   skb->dev->dev_addr)) ||
4786             (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) {
4787                 __skb_push(skb, ETH_HLEN);
4788                 skb->pkt_type = PACKET_HOST;
4789                 skb->protocol = eth_type_trans(skb, skb->dev);
4790         }
4791
4792         /* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull
4793          * before calling us again on redirect path. We do not call do_redirect
4794          * as we leave that up to the caller.
4795          *
4796          * Caller is responsible for managing lifetime of skb (i.e. calling
4797          * kfree_skb in response to actions it cannot handle/XDP_DROP).
4798          */
4799         switch (act) {
4800         case XDP_REDIRECT:
4801         case XDP_TX:
4802                 __skb_push(skb, mac_len);
4803                 break;
4804         case XDP_PASS:
4805                 metalen = xdp->data - xdp->data_meta;
4806                 if (metalen)
4807                         skb_metadata_set(skb, metalen);
4808                 break;
4809         }
4810
4811         return act;
4812 }
4813
4814 static u32 netif_receive_generic_xdp(struct sk_buff *skb,
4815                                      struct xdp_buff *xdp,
4816                                      struct bpf_prog *xdp_prog)
4817 {
4818         u32 act = XDP_DROP;
4819
4820         /* Reinjected packets coming from act_mirred or similar should
4821          * not get XDP generic processing.
4822          */
4823         if (skb_is_redirected(skb))
4824                 return XDP_PASS;
4825
4826         /* XDP packets must be linear and must have sufficient headroom
4827          * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also
4828          * native XDP provides, thus we need to do it here as well.
4829          */
4830         if (skb_cloned(skb) || skb_is_nonlinear(skb) ||
4831             skb_headroom(skb) < XDP_PACKET_HEADROOM) {
4832                 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb);
4833                 int troom = skb->tail + skb->data_len - skb->end;
4834
4835                 /* In case we have to go down the path and also linearize,
4836                  * then lets do the pskb_expand_head() work just once here.
4837                  */
4838                 if (pskb_expand_head(skb,
4839                                      hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0,
4840                                      troom > 0 ? troom + 128 : 0, GFP_ATOMIC))
4841                         goto do_drop;
4842                 if (skb_linearize(skb))
4843                         goto do_drop;
4844         }
4845
4846         act = bpf_prog_run_generic_xdp(skb, xdp, xdp_prog);
4847         switch (act) {
4848         case XDP_REDIRECT:
4849         case XDP_TX:
4850         case XDP_PASS:
4851                 break;
4852         default:
4853                 bpf_warn_invalid_xdp_action(skb->dev, xdp_prog, act);
4854                 fallthrough;
4855         case XDP_ABORTED:
4856                 trace_xdp_exception(skb->dev, xdp_prog, act);
4857                 fallthrough;
4858         case XDP_DROP:
4859         do_drop:
4860                 kfree_skb(skb);
4861                 break;
4862         }
4863
4864         return act;
4865 }
4866
4867 /* When doing generic XDP we have to bypass the qdisc layer and the
4868  * network taps in order to match in-driver-XDP behavior. This also means
4869  * that XDP packets are able to starve other packets going through a qdisc,
4870  * and DDOS attacks will be more effective. In-driver-XDP use dedicated TX
4871  * queues, so they do not have this starvation issue.
4872  */
4873 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog)
4874 {
4875         struct net_device *dev = skb->dev;
4876         struct netdev_queue *txq;
4877         bool free_skb = true;
4878         int cpu, rc;
4879
4880         txq = netdev_core_pick_tx(dev, skb, NULL);
4881         cpu = smp_processor_id();
4882         HARD_TX_LOCK(dev, txq, cpu);
4883         if (!netif_xmit_frozen_or_drv_stopped(txq)) {
4884                 rc = netdev_start_xmit(skb, dev, txq, 0);
4885                 if (dev_xmit_complete(rc))
4886                         free_skb = false;
4887         }
4888         HARD_TX_UNLOCK(dev, txq);
4889         if (free_skb) {
4890                 trace_xdp_exception(dev, xdp_prog, XDP_TX);
4891                 dev_core_stats_tx_dropped_inc(dev);
4892                 kfree_skb(skb);
4893         }
4894 }
4895
4896 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key);
4897
4898 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb)
4899 {
4900         if (xdp_prog) {
4901                 struct xdp_buff xdp;
4902                 u32 act;
4903                 int err;
4904
4905                 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog);
4906                 if (act != XDP_PASS) {
4907                         switch (act) {
4908                         case XDP_REDIRECT:
4909                                 err = xdp_do_generic_redirect(skb->dev, skb,
4910                                                               &xdp, xdp_prog);
4911                                 if (err)
4912                                         goto out_redir;
4913                                 break;
4914                         case XDP_TX:
4915                                 generic_xdp_tx(skb, xdp_prog);
4916                                 break;
4917                         }
4918                         return XDP_DROP;
4919                 }
4920         }
4921         return XDP_PASS;
4922 out_redir:
4923         kfree_skb_reason(skb, SKB_DROP_REASON_XDP);
4924         return XDP_DROP;
4925 }
4926 EXPORT_SYMBOL_GPL(do_xdp_generic);
4927
4928 static int netif_rx_internal(struct sk_buff *skb)
4929 {
4930         int ret;
4931
4932         net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb);
4933
4934         trace_netif_rx(skb);
4935
4936 #ifdef CONFIG_RPS
4937         if (static_branch_unlikely(&rps_needed)) {
4938                 struct rps_dev_flow voidflow, *rflow = &voidflow;
4939                 int cpu;
4940
4941                 rcu_read_lock();
4942
4943                 cpu = get_rps_cpu(skb->dev, skb, &rflow);
4944                 if (cpu < 0)
4945                         cpu = smp_processor_id();
4946
4947                 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4948
4949                 rcu_read_unlock();
4950         } else
4951 #endif
4952         {
4953                 unsigned int qtail;
4954
4955                 ret = enqueue_to_backlog(skb, smp_processor_id(), &qtail);
4956         }
4957         return ret;
4958 }
4959
4960 /**
4961  *      __netif_rx      -       Slightly optimized version of netif_rx
4962  *      @skb: buffer to post
4963  *
4964  *      This behaves as netif_rx except that it does not disable bottom halves.
4965  *      As a result this function may only be invoked from the interrupt context
4966  *      (either hard or soft interrupt).
4967  */
4968 int __netif_rx(struct sk_buff *skb)
4969 {
4970         int ret;
4971
4972         lockdep_assert_once(hardirq_count() | softirq_count());
4973
4974         trace_netif_rx_entry(skb);
4975         ret = netif_rx_internal(skb);
4976         trace_netif_rx_exit(ret);
4977         return ret;
4978 }
4979 EXPORT_SYMBOL(__netif_rx);
4980
4981 /**
4982  *      netif_rx        -       post buffer to the network code
4983  *      @skb: buffer to post
4984  *
4985  *      This function receives a packet from a device driver and queues it for
4986  *      the upper (protocol) levels to process via the backlog NAPI device. It
4987  *      always succeeds. The buffer may be dropped during processing for
4988  *      congestion control or by the protocol layers.
4989  *      The network buffer is passed via the backlog NAPI device. Modern NIC
4990  *      driver should use NAPI and GRO.
4991  *      This function can used from interrupt and from process context. The
4992  *      caller from process context must not disable interrupts before invoking
4993  *      this function.
4994  *
4995  *      return values:
4996  *      NET_RX_SUCCESS  (no congestion)
4997  *      NET_RX_DROP     (packet was dropped)
4998  *
4999  */
5000 int netif_rx(struct sk_buff *skb)
5001 {
5002         bool need_bh_off = !(hardirq_count() | softirq_count());
5003         int ret;
5004
5005         if (need_bh_off)
5006                 local_bh_disable();
5007         trace_netif_rx_entry(skb);
5008         ret = netif_rx_internal(skb);
5009         trace_netif_rx_exit(ret);
5010         if (need_bh_off)
5011                 local_bh_enable();
5012         return ret;
5013 }
5014 EXPORT_SYMBOL(netif_rx);
5015
5016 static __latent_entropy void net_tx_action(struct softirq_action *h)
5017 {
5018         struct softnet_data *sd = this_cpu_ptr(&softnet_data);
5019
5020         if (sd->completion_queue) {
5021                 struct sk_buff *clist;
5022
5023                 local_irq_disable();
5024                 clist = sd->completion_queue;
5025                 sd->completion_queue = NULL;
5026                 local_irq_enable();
5027
5028                 while (clist) {
5029                         struct sk_buff *skb = clist;
5030
5031                         clist = clist->next;
5032
5033                         WARN_ON(refcount_read(&skb->users));
5034                         if (likely(get_kfree_skb_cb(skb)->reason == SKB_CONSUMED))
5035                                 trace_consume_skb(skb, net_tx_action);
5036                         else
5037                                 trace_kfree_skb(skb, net_tx_action,
5038                                                 get_kfree_skb_cb(skb)->reason);
5039
5040                         if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
5041                                 __kfree_skb(skb);
5042                         else
5043                                 __napi_kfree_skb(skb,
5044                                                  get_kfree_skb_cb(skb)->reason);
5045                 }
5046         }
5047
5048         if (sd->output_queue) {
5049                 struct Qdisc *head;
5050
5051                 local_irq_disable();
5052                 head = sd->output_queue;
5053                 sd->output_queue = NULL;
5054                 sd->output_queue_tailp = &sd->output_queue;
5055                 local_irq_enable();
5056
5057                 rcu_read_lock();
5058
5059                 while (head) {
5060                         struct Qdisc *q = head;
5061                         spinlock_t *root_lock = NULL;
5062
5063                         head = head->next_sched;
5064
5065                         /* We need to make sure head->next_sched is read
5066                          * before clearing __QDISC_STATE_SCHED
5067                          */
5068                         smp_mb__before_atomic();
5069
5070                         if (!(q->flags & TCQ_F_NOLOCK)) {
5071                                 root_lock = qdisc_lock(q);
5072                                 spin_lock(root_lock);
5073                         } else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED,
5074                                                      &q->state))) {
5075                                 /* There is a synchronize_net() between
5076                                  * STATE_DEACTIVATED flag being set and
5077                                  * qdisc_reset()/some_qdisc_is_busy() in
5078                                  * dev_deactivate(), so we can safely bail out
5079                                  * early here to avoid data race between
5080                                  * qdisc_deactivate() and some_qdisc_is_busy()
5081                                  * for lockless qdisc.
5082                                  */
5083                                 clear_bit(__QDISC_STATE_SCHED, &q->state);
5084                                 continue;
5085                         }
5086
5087                         clear_bit(__QDISC_STATE_SCHED, &q->state);
5088                         qdisc_run(q);
5089                         if (root_lock)
5090                                 spin_unlock(root_lock);
5091                 }
5092
5093                 rcu_read_unlock();
5094         }
5095
5096         xfrm_dev_backlog(sd);
5097 }
5098
5099 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE)
5100 /* This hook is defined here for ATM LANE */
5101 int (*br_fdb_test_addr_hook)(struct net_device *dev,
5102                              unsigned char *addr) __read_mostly;
5103 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
5104 #endif
5105
5106 static inline struct sk_buff *
5107 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
5108                    struct net_device *orig_dev, bool *another)
5109 {
5110 #ifdef CONFIG_NET_CLS_ACT
5111         struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress);
5112         struct tcf_result cl_res;
5113
5114         /* If there's at least one ingress present somewhere (so
5115          * we get here via enabled static key), remaining devices
5116          * that are not configured with an ingress qdisc will bail
5117          * out here.
5118          */
5119         if (!miniq)
5120                 return skb;
5121
5122         if (*pt_prev) {
5123                 *ret = deliver_skb(skb, *pt_prev, orig_dev);
5124                 *pt_prev = NULL;
5125         }
5126
5127         qdisc_skb_cb(skb)->pkt_len = skb->len;
5128         tc_skb_cb(skb)->mru = 0;
5129         tc_skb_cb(skb)->post_ct = false;
5130         skb->tc_at_ingress = 1;
5131         mini_qdisc_bstats_cpu_update(miniq, skb);
5132
5133         switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) {
5134         case TC_ACT_OK:
5135         case TC_ACT_RECLASSIFY:
5136                 skb->tc_index = TC_H_MIN(cl_res.classid);
5137                 break;
5138         case TC_ACT_SHOT:
5139                 mini_qdisc_qstats_cpu_drop(miniq);
5140                 kfree_skb_reason(skb, SKB_DROP_REASON_TC_INGRESS);
5141                 *ret = NET_RX_DROP;
5142                 return NULL;
5143         case TC_ACT_STOLEN:
5144         case TC_ACT_QUEUED:
5145         case TC_ACT_TRAP:
5146                 consume_skb(skb);
5147                 *ret = NET_RX_SUCCESS;
5148                 return NULL;
5149         case TC_ACT_REDIRECT:
5150                 /* skb_mac_header check was done by cls/act_bpf, so
5151                  * we can safely push the L2 header back before
5152                  * redirecting to another netdev
5153                  */
5154                 __skb_push(skb, skb->mac_len);
5155                 if (skb_do_redirect(skb) == -EAGAIN) {
5156                         __skb_pull(skb, skb->mac_len);
5157                         *another = true;
5158                         break;
5159                 }
5160                 *ret = NET_RX_SUCCESS;
5161                 return NULL;
5162         case TC_ACT_CONSUMED:
5163                 *ret = NET_RX_SUCCESS;
5164                 return NULL;
5165         default:
5166                 break;
5167         }
5168 #endif /* CONFIG_NET_CLS_ACT */
5169         return skb;
5170 }
5171
5172 /**
5173  *      netdev_is_rx_handler_busy - check if receive handler is registered
5174  *      @dev: device to check
5175  *
5176  *      Check if a receive handler is already registered for a given device.
5177  *      Return true if there one.
5178  *
5179  *      The caller must hold the rtnl_mutex.
5180  */
5181 bool netdev_is_rx_handler_busy(struct net_device *dev)
5182 {
5183         ASSERT_RTNL();
5184         return dev && rtnl_dereference(dev->rx_handler);
5185 }
5186 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy);
5187
5188 /**
5189  *      netdev_rx_handler_register - register receive handler
5190  *      @dev: device to register a handler for
5191  *      @rx_handler: receive handler to register
5192  *      @rx_handler_data: data pointer that is used by rx handler
5193  *
5194  *      Register a receive handler for a device. This handler will then be
5195  *      called from __netif_receive_skb. A negative errno code is returned
5196  *      on a failure.
5197  *
5198  *      The caller must hold the rtnl_mutex.
5199  *
5200  *      For a general description of rx_handler, see enum rx_handler_result.
5201  */
5202 int netdev_rx_handler_register(struct net_device *dev,
5203                                rx_handler_func_t *rx_handler,
5204                                void *rx_handler_data)
5205 {
5206         if (netdev_is_rx_handler_busy(dev))
5207                 return -EBUSY;
5208
5209         if (dev->priv_flags & IFF_NO_RX_HANDLER)
5210                 return -EINVAL;
5211
5212         /* Note: rx_handler_data must be set before rx_handler */
5213         rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
5214         rcu_assign_pointer(dev->rx_handler, rx_handler);
5215
5216         return 0;
5217 }
5218 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
5219
5220 /**
5221  *      netdev_rx_handler_unregister - unregister receive handler
5222  *      @dev: device to unregister a handler from
5223  *
5224  *      Unregister a receive handler from a device.
5225  *
5226  *      The caller must hold the rtnl_mutex.
5227  */
5228 void netdev_rx_handler_unregister(struct net_device *dev)
5229 {
5230
5231         ASSERT_RTNL();
5232         RCU_INIT_POINTER(dev->rx_handler, NULL);
5233         /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
5234          * section has a guarantee to see a non NULL rx_handler_data
5235          * as well.
5236          */
5237         synchronize_net();
5238         RCU_INIT_POINTER(dev->rx_handler_data, NULL);
5239 }
5240 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
5241
5242 /*
5243  * Limit the use of PFMEMALLOC reserves to those protocols that implement
5244  * the special handling of PFMEMALLOC skbs.
5245  */
5246 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
5247 {
5248         switch (skb->protocol) {
5249         case htons(ETH_P_ARP):
5250         case htons(ETH_P_IP):
5251         case htons(ETH_P_IPV6):
5252         case htons(ETH_P_8021Q):
5253         case htons(ETH_P_8021AD):
5254                 return true;
5255         default:
5256                 return false;
5257         }
5258 }
5259
5260 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
5261                              int *ret, struct net_device *orig_dev)
5262 {
5263         if (nf_hook_ingress_active(skb)) {
5264                 int ingress_retval;
5265
5266                 if (*pt_prev) {
5267                         *ret = deliver_skb(skb, *pt_prev, orig_dev);
5268                         *pt_prev = NULL;
5269                 }
5270
5271                 rcu_read_lock();
5272                 ingress_retval = nf_hook_ingress(skb);
5273                 rcu_read_unlock();
5274                 return ingress_retval;
5275         }
5276         return 0;
5277 }
5278
5279 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc,
5280                                     struct packet_type **ppt_prev)
5281 {
5282         struct packet_type *ptype, *pt_prev;
5283         rx_handler_func_t *rx_handler;
5284         struct sk_buff *skb = *pskb;
5285         struct net_device *orig_dev;
5286         bool deliver_exact = false;
5287         int ret = NET_RX_DROP;
5288         __be16 type;
5289
5290         net_timestamp_check(!READ_ONCE(netdev_tstamp_prequeue), skb);
5291
5292         trace_netif_receive_skb(skb);
5293
5294         orig_dev = skb->dev;
5295
5296         skb_reset_network_header(skb);
5297         if (!skb_transport_header_was_set(skb))
5298                 skb_reset_transport_header(skb);
5299         skb_reset_mac_len(skb);
5300
5301         pt_prev = NULL;
5302
5303 another_round:
5304         skb->skb_iif = skb->dev->ifindex;
5305
5306         __this_cpu_inc(softnet_data.processed);
5307
5308         if (static_branch_unlikely(&generic_xdp_needed_key)) {
5309                 int ret2;
5310
5311                 migrate_disable();
5312                 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb);
5313                 migrate_enable();
5314
5315                 if (ret2 != XDP_PASS) {
5316                         ret = NET_RX_DROP;
5317                         goto out;
5318                 }
5319         }
5320
5321         if (eth_type_vlan(skb->protocol)) {
5322                 skb = skb_vlan_untag(skb);
5323                 if (unlikely(!skb))
5324                         goto out;
5325         }
5326
5327         if (skb_skip_tc_classify(skb))
5328                 goto skip_classify;
5329
5330         if (pfmemalloc)
5331                 goto skip_taps;
5332
5333         list_for_each_entry_rcu(ptype, &ptype_all, list) {
5334                 if (pt_prev)
5335                         ret = deliver_skb(skb, pt_prev, orig_dev);
5336                 pt_prev = ptype;
5337         }
5338
5339         list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
5340                 if (pt_prev)
5341                         ret = deliver_skb(skb, pt_prev, orig_dev);
5342                 pt_prev = ptype;
5343         }
5344
5345 skip_taps:
5346 #ifdef CONFIG_NET_INGRESS
5347         if (static_branch_unlikely(&ingress_needed_key)) {
5348                 bool another = false;
5349
5350                 nf_skip_egress(skb, true);
5351                 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev,
5352                                          &another);
5353                 if (another)
5354                         goto another_round;
5355                 if (!skb)
5356                         goto out;
5357
5358                 nf_skip_egress(skb, false);
5359                 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
5360                         goto out;
5361         }
5362 #endif
5363         skb_reset_redirect(skb);
5364 skip_classify:
5365         if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
5366                 goto drop;
5367
5368         if (skb_vlan_tag_present(skb)) {
5369                 if (pt_prev) {
5370                         ret = deliver_skb(skb, pt_prev, orig_dev);
5371                         pt_prev = NULL;
5372                 }
5373                 if (vlan_do_receive(&skb))
5374                         goto another_round;
5375                 else if (unlikely(!skb))
5376                         goto out;
5377         }
5378
5379         rx_handler = rcu_dereference(skb->dev->rx_handler);
5380         if (rx_handler) {
5381                 if (pt_prev) {
5382                         ret = deliver_skb(skb, pt_prev, orig_dev);
5383                         pt_prev = NULL;
5384                 }
5385                 switch (rx_handler(&skb)) {
5386                 case RX_HANDLER_CONSUMED:
5387                         ret = NET_RX_SUCCESS;
5388                         goto out;
5389                 case RX_HANDLER_ANOTHER:
5390                         goto another_round;
5391                 case RX_HANDLER_EXACT:
5392                         deliver_exact = true;
5393                         break;
5394                 case RX_HANDLER_PASS:
5395                         break;
5396                 default:
5397                         BUG();
5398                 }
5399         }
5400
5401         if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) {
5402 check_vlan_id:
5403                 if (skb_vlan_tag_get_id(skb)) {
5404                         /* Vlan id is non 0 and vlan_do_receive() above couldn't
5405                          * find vlan device.
5406                          */
5407                         skb->pkt_type = PACKET_OTHERHOST;
5408                 } else if (eth_type_vlan(skb->protocol)) {
5409                         /* Outer header is 802.1P with vlan 0, inner header is
5410                          * 802.1Q or 802.1AD and vlan_do_receive() above could
5411                          * not find vlan dev for vlan id 0.
5412                          */
5413                         __vlan_hwaccel_clear_tag(skb);
5414                         skb = skb_vlan_untag(skb);
5415                         if (unlikely(!skb))
5416                                 goto out;
5417                         if (vlan_do_receive(&skb))
5418                                 /* After stripping off 802.1P header with vlan 0
5419                                  * vlan dev is found for inner header.
5420                                  */
5421                                 goto another_round;
5422                         else if (unlikely(!skb))
5423                                 goto out;
5424                         else
5425                                 /* We have stripped outer 802.1P vlan 0 header.
5426                                  * But could not find vlan dev.
5427                                  * check again for vlan id to set OTHERHOST.
5428                                  */
5429                                 goto check_vlan_id;
5430                 }
5431                 /* Note: we might in the future use prio bits
5432                  * and set skb->priority like in vlan_do_receive()
5433                  * For the time being, just ignore Priority Code Point
5434                  */
5435                 __vlan_hwaccel_clear_tag(skb);
5436         }
5437
5438         type = skb->protocol;
5439
5440         /* deliver only exact match when indicated */
5441         if (likely(!deliver_exact)) {
5442                 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5443                                        &ptype_base[ntohs(type) &
5444                                                    PTYPE_HASH_MASK]);
5445         }
5446
5447         deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5448                                &orig_dev->ptype_specific);
5449
5450         if (unlikely(skb->dev != orig_dev)) {
5451                 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5452                                        &skb->dev->ptype_specific);
5453         }
5454
5455         if (pt_prev) {
5456                 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
5457                         goto drop;
5458                 *ppt_prev = pt_prev;
5459         } else {
5460 drop:
5461                 if (!deliver_exact)
5462                         dev_core_stats_rx_dropped_inc(skb->dev);
5463                 else
5464                         dev_core_stats_rx_nohandler_inc(skb->dev);
5465                 kfree_skb_reason(skb, SKB_DROP_REASON_UNHANDLED_PROTO);
5466                 /* Jamal, now you will not able to escape explaining
5467                  * me how you were going to use this. :-)
5468                  */
5469                 ret = NET_RX_DROP;
5470         }
5471
5472 out:
5473         /* The invariant here is that if *ppt_prev is not NULL
5474          * then skb should also be non-NULL.
5475          *
5476          * Apparently *ppt_prev assignment above holds this invariant due to
5477          * skb dereferencing near it.
5478          */
5479         *pskb = skb;
5480         return ret;
5481 }
5482
5483 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc)
5484 {
5485         struct net_device *orig_dev = skb->dev;
5486         struct packet_type *pt_prev = NULL;
5487         int ret;
5488
5489         ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5490         if (pt_prev)
5491                 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb,
5492                                          skb->dev, pt_prev, orig_dev);
5493         return ret;
5494 }
5495
5496 /**
5497  *      netif_receive_skb_core - special purpose version of netif_receive_skb
5498  *      @skb: buffer to process
5499  *
5500  *      More direct receive version of netif_receive_skb().  It should
5501  *      only be used by callers that have a need to skip RPS and Generic XDP.
5502  *      Caller must also take care of handling if ``(page_is_)pfmemalloc``.
5503  *
5504  *      This function may only be called from softirq context and interrupts
5505  *      should be enabled.
5506  *
5507  *      Return values (usually ignored):
5508  *      NET_RX_SUCCESS: no congestion
5509  *      NET_RX_DROP: packet was dropped
5510  */
5511 int netif_receive_skb_core(struct sk_buff *skb)
5512 {
5513         int ret;
5514
5515         rcu_read_lock();
5516         ret = __netif_receive_skb_one_core(skb, false);
5517         rcu_read_unlock();
5518
5519         return ret;
5520 }
5521 EXPORT_SYMBOL(netif_receive_skb_core);
5522
5523 static inline void __netif_receive_skb_list_ptype(struct list_head *head,
5524                                                   struct packet_type *pt_prev,
5525                                                   struct net_device *orig_dev)
5526 {
5527         struct sk_buff *skb, *next;
5528
5529         if (!pt_prev)
5530                 return;
5531         if (list_empty(head))
5532                 return;
5533         if (pt_prev->list_func != NULL)
5534                 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv,
5535                                    ip_list_rcv, head, pt_prev, orig_dev);
5536         else
5537                 list_for_each_entry_safe(skb, next, head, list) {
5538                         skb_list_del_init(skb);
5539                         pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
5540                 }
5541 }
5542
5543 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc)
5544 {
5545         /* Fast-path assumptions:
5546          * - There is no RX handler.
5547          * - Only one packet_type matches.
5548          * If either of these fails, we will end up doing some per-packet
5549          * processing in-line, then handling the 'last ptype' for the whole
5550          * sublist.  This can't cause out-of-order delivery to any single ptype,
5551          * because the 'last ptype' must be constant across the sublist, and all
5552          * other ptypes are handled per-packet.
5553          */
5554         /* Current (common) ptype of sublist */
5555         struct packet_type *pt_curr = NULL;
5556         /* Current (common) orig_dev of sublist */
5557         struct net_device *od_curr = NULL;
5558         struct list_head sublist;
5559         struct sk_buff *skb, *next;
5560
5561         INIT_LIST_HEAD(&sublist);
5562         list_for_each_entry_safe(skb, next, head, list) {
5563                 struct net_device *orig_dev = skb->dev;
5564                 struct packet_type *pt_prev = NULL;
5565
5566                 skb_list_del_init(skb);
5567                 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5568                 if (!pt_prev)
5569                         continue;
5570                 if (pt_curr != pt_prev || od_curr != orig_dev) {
5571                         /* dispatch old sublist */
5572                         __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5573                         /* start new sublist */
5574                         INIT_LIST_HEAD(&sublist);
5575                         pt_curr = pt_prev;
5576                         od_curr = orig_dev;
5577                 }
5578                 list_add_tail(&skb->list, &sublist);
5579         }
5580
5581         /* dispatch final sublist */
5582         __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5583 }
5584
5585 static int __netif_receive_skb(struct sk_buff *skb)
5586 {
5587         int ret;
5588
5589         if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
5590                 unsigned int noreclaim_flag;
5591
5592                 /*
5593                  * PFMEMALLOC skbs are special, they should
5594                  * - be delivered to SOCK_MEMALLOC sockets only
5595                  * - stay away from userspace
5596                  * - have bounded memory usage
5597                  *
5598                  * Use PF_MEMALLOC as this saves us from propagating the allocation
5599                  * context down to all allocation sites.
5600                  */
5601                 noreclaim_flag = memalloc_noreclaim_save();
5602                 ret = __netif_receive_skb_one_core(skb, true);
5603                 memalloc_noreclaim_restore(noreclaim_flag);
5604         } else
5605                 ret = __netif_receive_skb_one_core(skb, false);
5606
5607         return ret;
5608 }
5609
5610 static void __netif_receive_skb_list(struct list_head *head)
5611 {
5612         unsigned long noreclaim_flag = 0;
5613         struct sk_buff *skb, *next;
5614         bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */
5615
5616         list_for_each_entry_safe(skb, next, head, list) {
5617                 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) {
5618                         struct list_head sublist;
5619
5620                         /* Handle the previous sublist */
5621                         list_cut_before(&sublist, head, &skb->list);
5622                         if (!list_empty(&sublist))
5623                                 __netif_receive_skb_list_core(&sublist, pfmemalloc);
5624                         pfmemalloc = !pfmemalloc;
5625                         /* See comments in __netif_receive_skb */
5626                         if (pfmemalloc)
5627                                 noreclaim_flag = memalloc_noreclaim_save();
5628                         else
5629                                 memalloc_noreclaim_restore(noreclaim_flag);
5630                 }
5631         }
5632         /* Handle the remaining sublist */
5633         if (!list_empty(head))
5634                 __netif_receive_skb_list_core(head, pfmemalloc);
5635         /* Restore pflags */
5636         if (pfmemalloc)
5637                 memalloc_noreclaim_restore(noreclaim_flag);
5638 }
5639
5640 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp)
5641 {
5642         struct bpf_prog *old = rtnl_dereference(dev->xdp_prog);
5643         struct bpf_prog *new = xdp->prog;
5644         int ret = 0;
5645
5646         switch (xdp->command) {
5647         case XDP_SETUP_PROG:
5648                 rcu_assign_pointer(dev->xdp_prog, new);
5649                 if (old)
5650                         bpf_prog_put(old);
5651
5652                 if (old && !new) {
5653                         static_branch_dec(&generic_xdp_needed_key);
5654                 } else if (new && !old) {
5655                         static_branch_inc(&generic_xdp_needed_key);
5656                         dev_disable_lro(dev);
5657                         dev_disable_gro_hw(dev);
5658                 }
5659                 break;
5660
5661         default:
5662                 ret = -EINVAL;
5663                 break;
5664         }
5665
5666         return ret;
5667 }
5668
5669 static int netif_receive_skb_internal(struct sk_buff *skb)
5670 {
5671         int ret;
5672
5673         net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb);
5674
5675         if (skb_defer_rx_timestamp(skb))
5676                 return NET_RX_SUCCESS;
5677
5678         rcu_read_lock();
5679 #ifdef CONFIG_RPS
5680         if (static_branch_unlikely(&rps_needed)) {
5681                 struct rps_dev_flow voidflow, *rflow = &voidflow;
5682                 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5683
5684                 if (cpu >= 0) {
5685                         ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5686                         rcu_read_unlock();
5687                         return ret;
5688                 }
5689         }
5690 #endif
5691         ret = __netif_receive_skb(skb);
5692         rcu_read_unlock();
5693         return ret;
5694 }
5695
5696 void netif_receive_skb_list_internal(struct list_head *head)
5697 {
5698         struct sk_buff *skb, *next;
5699         struct list_head sublist;
5700
5701         INIT_LIST_HEAD(&sublist);
5702         list_for_each_entry_safe(skb, next, head, list) {
5703                 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb);
5704                 skb_list_del_init(skb);
5705                 if (!skb_defer_rx_timestamp(skb))
5706                         list_add_tail(&skb->list, &sublist);
5707         }
5708         list_splice_init(&sublist, head);
5709
5710         rcu_read_lock();
5711 #ifdef CONFIG_RPS
5712         if (static_branch_unlikely(&rps_needed)) {
5713                 list_for_each_entry_safe(skb, next, head, list) {
5714                         struct rps_dev_flow voidflow, *rflow = &voidflow;
5715                         int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5716
5717                         if (cpu >= 0) {
5718                                 /* Will be handled, remove from list */
5719                                 skb_list_del_init(skb);
5720                                 enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5721                         }
5722                 }
5723         }
5724 #endif
5725         __netif_receive_skb_list(head);
5726         rcu_read_unlock();
5727 }
5728
5729 /**
5730  *      netif_receive_skb - process receive buffer from network
5731  *      @skb: buffer to process
5732  *
5733  *      netif_receive_skb() is the main receive data processing function.
5734  *      It always succeeds. The buffer may be dropped during processing
5735  *      for congestion control or by the protocol layers.
5736  *
5737  *      This function may only be called from softirq context and interrupts
5738  *      should be enabled.
5739  *
5740  *      Return values (usually ignored):
5741  *      NET_RX_SUCCESS: no congestion
5742  *      NET_RX_DROP: packet was dropped
5743  */
5744 int netif_receive_skb(struct sk_buff *skb)
5745 {
5746         int ret;
5747
5748         trace_netif_receive_skb_entry(skb);
5749
5750         ret = netif_receive_skb_internal(skb);
5751         trace_netif_receive_skb_exit(ret);
5752
5753         return ret;
5754 }
5755 EXPORT_SYMBOL(netif_receive_skb);
5756
5757 /**
5758  *      netif_receive_skb_list - process many receive buffers from network
5759  *      @head: list of skbs to process.
5760  *
5761  *      Since return value of netif_receive_skb() is normally ignored, and
5762  *      wouldn't be meaningful for a list, this function returns void.
5763  *
5764  *      This function may only be called from softirq context and interrupts
5765  *      should be enabled.
5766  */
5767 void netif_receive_skb_list(struct list_head *head)
5768 {
5769         struct sk_buff *skb;
5770
5771         if (list_empty(head))
5772                 return;
5773         if (trace_netif_receive_skb_list_entry_enabled()) {
5774                 list_for_each_entry(skb, head, list)
5775                         trace_netif_receive_skb_list_entry(skb);
5776         }
5777         netif_receive_skb_list_internal(head);
5778         trace_netif_receive_skb_list_exit(0);
5779 }
5780 EXPORT_SYMBOL(netif_receive_skb_list);
5781
5782 static DEFINE_PER_CPU(struct work_struct, flush_works);
5783
5784 /* Network device is going away, flush any packets still pending */
5785 static void flush_backlog(struct work_struct *work)
5786 {
5787         struct sk_buff *skb, *tmp;
5788         struct softnet_data *sd;
5789
5790         local_bh_disable();
5791         sd = this_cpu_ptr(&softnet_data);
5792
5793         rps_lock_irq_disable(sd);
5794         skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
5795                 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5796                         __skb_unlink(skb, &sd->input_pkt_queue);
5797                         dev_kfree_skb_irq(skb);
5798                         input_queue_head_incr(sd);
5799                 }
5800         }
5801         rps_unlock_irq_enable(sd);
5802
5803         skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
5804                 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5805                         __skb_unlink(skb, &sd->process_queue);
5806                         kfree_skb(skb);
5807                         input_queue_head_incr(sd);
5808                 }
5809         }
5810         local_bh_enable();
5811 }
5812
5813 static bool flush_required(int cpu)
5814 {
5815 #if IS_ENABLED(CONFIG_RPS)
5816         struct softnet_data *sd = &per_cpu(softnet_data, cpu);
5817         bool do_flush;
5818
5819         rps_lock_irq_disable(sd);
5820
5821         /* as insertion into process_queue happens with the rps lock held,
5822          * process_queue access may race only with dequeue
5823          */
5824         do_flush = !skb_queue_empty(&sd->input_pkt_queue) ||
5825                    !skb_queue_empty_lockless(&sd->process_queue);
5826         rps_unlock_irq_enable(sd);
5827
5828         return do_flush;
5829 #endif
5830         /* without RPS we can't safely check input_pkt_queue: during a
5831          * concurrent remote skb_queue_splice() we can detect as empty both
5832          * input_pkt_queue and process_queue even if the latter could end-up
5833          * containing a lot of packets.
5834          */
5835         return true;
5836 }
5837
5838 static void flush_all_backlogs(void)
5839 {
5840         static cpumask_t flush_cpus;
5841         unsigned int cpu;
5842
5843         /* since we are under rtnl lock protection we can use static data
5844          * for the cpumask and avoid allocating on stack the possibly
5845          * large mask
5846          */
5847         ASSERT_RTNL();
5848
5849         cpus_read_lock();
5850
5851         cpumask_clear(&flush_cpus);
5852         for_each_online_cpu(cpu) {
5853                 if (flush_required(cpu)) {
5854                         queue_work_on(cpu, system_highpri_wq,
5855                                       per_cpu_ptr(&flush_works, cpu));
5856                         cpumask_set_cpu(cpu, &flush_cpus);
5857                 }
5858         }
5859
5860         /* we can have in flight packet[s] on the cpus we are not flushing,
5861          * synchronize_net() in unregister_netdevice_many() will take care of
5862          * them
5863          */
5864         for_each_cpu(cpu, &flush_cpus)
5865                 flush_work(per_cpu_ptr(&flush_works, cpu));
5866
5867         cpus_read_unlock();
5868 }
5869
5870 static void net_rps_send_ipi(struct softnet_data *remsd)
5871 {
5872 #ifdef CONFIG_RPS
5873         while (remsd) {
5874                 struct softnet_data *next = remsd->rps_ipi_next;
5875
5876                 if (cpu_online(remsd->cpu))
5877                         smp_call_function_single_async(remsd->cpu, &remsd->csd);
5878                 remsd = next;
5879         }
5880 #endif
5881 }
5882
5883 /*
5884  * net_rps_action_and_irq_enable sends any pending IPI's for rps.
5885  * Note: called with local irq disabled, but exits with local irq enabled.
5886  */
5887 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
5888 {
5889 #ifdef CONFIG_RPS
5890         struct softnet_data *remsd = sd->rps_ipi_list;
5891
5892         if (remsd) {
5893                 sd->rps_ipi_list = NULL;
5894
5895                 local_irq_enable();
5896
5897                 /* Send pending IPI's to kick RPS processing on remote cpus. */
5898                 net_rps_send_ipi(remsd);
5899         } else
5900 #endif
5901                 local_irq_enable();
5902 }
5903
5904 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
5905 {
5906 #ifdef CONFIG_RPS
5907         return sd->rps_ipi_list != NULL;
5908 #else
5909         return false;
5910 #endif
5911 }
5912
5913 static int process_backlog(struct napi_struct *napi, int quota)
5914 {
5915         struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
5916         bool again = true;
5917         int work = 0;
5918
5919         /* Check if we have pending ipi, its better to send them now,
5920          * not waiting net_rx_action() end.
5921          */
5922         if (sd_has_rps_ipi_waiting(sd)) {
5923                 local_irq_disable();
5924                 net_rps_action_and_irq_enable(sd);
5925         }
5926
5927         napi->weight = READ_ONCE(dev_rx_weight);
5928         while (again) {
5929                 struct sk_buff *skb;
5930
5931                 while ((skb = __skb_dequeue(&sd->process_queue))) {
5932                         rcu_read_lock();
5933                         __netif_receive_skb(skb);
5934                         rcu_read_unlock();
5935                         input_queue_head_incr(sd);
5936                         if (++work >= quota)
5937                                 return work;
5938
5939                 }
5940
5941                 rps_lock_irq_disable(sd);
5942                 if (skb_queue_empty(&sd->input_pkt_queue)) {
5943                         /*
5944                          * Inline a custom version of __napi_complete().
5945                          * only current cpu owns and manipulates this napi,
5946                          * and NAPI_STATE_SCHED is the only possible flag set
5947                          * on backlog.
5948                          * We can use a plain write instead of clear_bit(),
5949                          * and we dont need an smp_mb() memory barrier.
5950                          */
5951                         napi->state = 0;
5952                         again = false;
5953                 } else {
5954                         skb_queue_splice_tail_init(&sd->input_pkt_queue,
5955                                                    &sd->process_queue);
5956                 }
5957                 rps_unlock_irq_enable(sd);
5958         }
5959
5960         return work;
5961 }
5962
5963 /**
5964  * __napi_schedule - schedule for receive
5965  * @n: entry to schedule
5966  *
5967  * The entry's receive function will be scheduled to run.
5968  * Consider using __napi_schedule_irqoff() if hard irqs are masked.
5969  */
5970 void __napi_schedule(struct napi_struct *n)
5971 {
5972         unsigned long flags;
5973
5974         local_irq_save(flags);
5975         ____napi_schedule(this_cpu_ptr(&softnet_data), n);
5976         local_irq_restore(flags);
5977 }
5978 EXPORT_SYMBOL(__napi_schedule);
5979
5980 /**
5981  *      napi_schedule_prep - check if napi can be scheduled
5982  *      @n: napi context
5983  *
5984  * Test if NAPI routine is already running, and if not mark
5985  * it as running.  This is used as a condition variable to
5986  * insure only one NAPI poll instance runs.  We also make
5987  * sure there is no pending NAPI disable.
5988  */
5989 bool napi_schedule_prep(struct napi_struct *n)
5990 {
5991         unsigned long new, val = READ_ONCE(n->state);
5992
5993         do {
5994                 if (unlikely(val & NAPIF_STATE_DISABLE))
5995                         return false;
5996                 new = val | NAPIF_STATE_SCHED;
5997
5998                 /* Sets STATE_MISSED bit if STATE_SCHED was already set
5999                  * This was suggested by Alexander Duyck, as compiler
6000                  * emits better code than :
6001                  * if (val & NAPIF_STATE_SCHED)
6002                  *     new |= NAPIF_STATE_MISSED;
6003                  */
6004                 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED *
6005                                                    NAPIF_STATE_MISSED;
6006         } while (!try_cmpxchg(&n->state, &val, new));
6007
6008         return !(val & NAPIF_STATE_SCHED);
6009 }
6010 EXPORT_SYMBOL(napi_schedule_prep);
6011
6012 /**
6013  * __napi_schedule_irqoff - schedule for receive
6014  * @n: entry to schedule
6015  *
6016  * Variant of __napi_schedule() assuming hard irqs are masked.
6017  *
6018  * On PREEMPT_RT enabled kernels this maps to __napi_schedule()
6019  * because the interrupt disabled assumption might not be true
6020  * due to force-threaded interrupts and spinlock substitution.
6021  */
6022 void __napi_schedule_irqoff(struct napi_struct *n)
6023 {
6024         if (!IS_ENABLED(CONFIG_PREEMPT_RT))
6025                 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
6026         else
6027                 __napi_schedule(n);
6028 }
6029 EXPORT_SYMBOL(__napi_schedule_irqoff);
6030
6031 bool napi_complete_done(struct napi_struct *n, int work_done)
6032 {
6033         unsigned long flags, val, new, timeout = 0;
6034         bool ret = true;
6035
6036         /*
6037          * 1) Don't let napi dequeue from the cpu poll list
6038          *    just in case its running on a different cpu.
6039          * 2) If we are busy polling, do nothing here, we have
6040          *    the guarantee we will be called later.
6041          */
6042         if (unlikely(n->state & (NAPIF_STATE_NPSVC |
6043                                  NAPIF_STATE_IN_BUSY_POLL)))
6044                 return false;
6045
6046         if (work_done) {
6047                 if (n->gro_bitmask)
6048                         timeout = READ_ONCE(n->dev->gro_flush_timeout);
6049                 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs);
6050         }
6051         if (n->defer_hard_irqs_count > 0) {
6052                 n->defer_hard_irqs_count--;
6053                 timeout = READ_ONCE(n->dev->gro_flush_timeout);
6054                 if (timeout)
6055                         ret = false;
6056         }
6057         if (n->gro_bitmask) {
6058                 /* When the NAPI instance uses a timeout and keeps postponing
6059                  * it, we need to bound somehow the time packets are kept in
6060                  * the GRO layer
6061                  */
6062                 napi_gro_flush(n, !!timeout);
6063         }
6064
6065         gro_normal_list(n);
6066
6067         if (unlikely(!list_empty(&n->poll_list))) {
6068                 /* If n->poll_list is not empty, we need to mask irqs */
6069                 local_irq_save(flags);
6070                 list_del_init(&n->poll_list);
6071                 local_irq_restore(flags);
6072         }
6073         WRITE_ONCE(n->list_owner, -1);
6074
6075         val = READ_ONCE(n->state);
6076         do {
6077                 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED));
6078
6079                 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED |
6080                               NAPIF_STATE_SCHED_THREADED |
6081                               NAPIF_STATE_PREFER_BUSY_POLL);
6082
6083                 /* If STATE_MISSED was set, leave STATE_SCHED set,
6084                  * because we will call napi->poll() one more time.
6085                  * This C code was suggested by Alexander Duyck to help gcc.
6086                  */
6087                 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED *
6088                                                     NAPIF_STATE_SCHED;
6089         } while (!try_cmpxchg(&n->state, &val, new));
6090
6091         if (unlikely(val & NAPIF_STATE_MISSED)) {
6092                 __napi_schedule(n);
6093                 return false;
6094         }
6095
6096         if (timeout)
6097                 hrtimer_start(&n->timer, ns_to_ktime(timeout),
6098                               HRTIMER_MODE_REL_PINNED);
6099         return ret;
6100 }
6101 EXPORT_SYMBOL(napi_complete_done);
6102
6103 /* must be called under rcu_read_lock(), as we dont take a reference */
6104 static struct napi_struct *napi_by_id(unsigned int napi_id)
6105 {
6106         unsigned int hash = napi_id % HASH_SIZE(napi_hash);
6107         struct napi_struct *napi;
6108
6109         hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
6110                 if (napi->napi_id == napi_id)
6111                         return napi;
6112
6113         return NULL;
6114 }
6115
6116 #if defined(CONFIG_NET_RX_BUSY_POLL)
6117
6118 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule)
6119 {
6120         if (!skip_schedule) {
6121                 gro_normal_list(napi);
6122                 __napi_schedule(napi);
6123                 return;
6124         }
6125
6126         if (napi->gro_bitmask) {
6127                 /* flush too old packets
6128                  * If HZ < 1000, flush all packets.
6129                  */
6130                 napi_gro_flush(napi, HZ >= 1000);
6131         }
6132
6133         gro_normal_list(napi);
6134         clear_bit(NAPI_STATE_SCHED, &napi->state);
6135 }
6136
6137 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll,
6138                            u16 budget)
6139 {
6140         bool skip_schedule = false;
6141         unsigned long timeout;
6142         int rc;
6143
6144         /* Busy polling means there is a high chance device driver hard irq
6145          * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was
6146          * set in napi_schedule_prep().
6147          * Since we are about to call napi->poll() once more, we can safely
6148          * clear NAPI_STATE_MISSED.
6149          *
6150          * Note: x86 could use a single "lock and ..." instruction
6151          * to perform these two clear_bit()
6152          */
6153         clear_bit(NAPI_STATE_MISSED, &napi->state);
6154         clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state);
6155
6156         local_bh_disable();
6157
6158         if (prefer_busy_poll) {
6159                 napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs);
6160                 timeout = READ_ONCE(napi->dev->gro_flush_timeout);
6161                 if (napi->defer_hard_irqs_count && timeout) {
6162                         hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED);
6163                         skip_schedule = true;
6164                 }
6165         }
6166
6167         /* All we really want here is to re-enable device interrupts.
6168          * Ideally, a new ndo_busy_poll_stop() could avoid another round.
6169          */
6170         rc = napi->poll(napi, budget);
6171         /* We can't gro_normal_list() here, because napi->poll() might have
6172          * rearmed the napi (napi_complete_done()) in which case it could
6173          * already be running on another CPU.
6174          */
6175         trace_napi_poll(napi, rc, budget);
6176         netpoll_poll_unlock(have_poll_lock);
6177         if (rc == budget)
6178                 __busy_poll_stop(napi, skip_schedule);
6179         local_bh_enable();
6180 }
6181
6182 void napi_busy_loop(unsigned int napi_id,
6183                     bool (*loop_end)(void *, unsigned long),
6184                     void *loop_end_arg, bool prefer_busy_poll, u16 budget)
6185 {
6186         unsigned long start_time = loop_end ? busy_loop_current_time() : 0;
6187         int (*napi_poll)(struct napi_struct *napi, int budget);
6188         void *have_poll_lock = NULL;
6189         struct napi_struct *napi;
6190
6191 restart:
6192         napi_poll = NULL;
6193
6194         rcu_read_lock();
6195
6196         napi = napi_by_id(napi_id);
6197         if (!napi)
6198                 goto out;
6199
6200         preempt_disable();
6201         for (;;) {
6202                 int work = 0;
6203
6204                 local_bh_disable();
6205                 if (!napi_poll) {
6206                         unsigned long val = READ_ONCE(napi->state);
6207
6208                         /* If multiple threads are competing for this napi,
6209                          * we avoid dirtying napi->state as much as we can.
6210                          */
6211                         if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED |
6212                                    NAPIF_STATE_IN_BUSY_POLL)) {
6213                                 if (prefer_busy_poll)
6214                                         set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6215                                 goto count;
6216                         }
6217                         if (cmpxchg(&napi->state, val,
6218                                     val | NAPIF_STATE_IN_BUSY_POLL |
6219                                           NAPIF_STATE_SCHED) != val) {
6220                                 if (prefer_busy_poll)
6221                                         set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6222                                 goto count;
6223                         }
6224                         have_poll_lock = netpoll_poll_lock(napi);
6225                         napi_poll = napi->poll;
6226                 }
6227                 work = napi_poll(napi, budget);
6228                 trace_napi_poll(napi, work, budget);
6229                 gro_normal_list(napi);
6230 count:
6231                 if (work > 0)
6232                         __NET_ADD_STATS(dev_net(napi->dev),
6233                                         LINUX_MIB_BUSYPOLLRXPACKETS, work);
6234                 local_bh_enable();
6235
6236                 if (!loop_end || loop_end(loop_end_arg, start_time))
6237                         break;
6238
6239                 if (unlikely(need_resched())) {
6240                         if (napi_poll)
6241                                 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget);
6242                         preempt_enable();
6243                         rcu_read_unlock();
6244                         cond_resched();
6245                         if (loop_end(loop_end_arg, start_time))
6246                                 return;
6247                         goto restart;
6248                 }
6249                 cpu_relax();
6250         }
6251         if (napi_poll)
6252                 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget);
6253         preempt_enable();
6254 out:
6255         rcu_read_unlock();
6256 }
6257 EXPORT_SYMBOL(napi_busy_loop);
6258
6259 #endif /* CONFIG_NET_RX_BUSY_POLL */
6260
6261 static void napi_hash_add(struct napi_struct *napi)
6262 {
6263         if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state))
6264                 return;
6265
6266         spin_lock(&napi_hash_lock);
6267
6268         /* 0..NR_CPUS range is reserved for sender_cpu use */
6269         do {
6270                 if (unlikely(++napi_gen_id < MIN_NAPI_ID))
6271                         napi_gen_id = MIN_NAPI_ID;
6272         } while (napi_by_id(napi_gen_id));
6273         napi->napi_id = napi_gen_id;
6274
6275         hlist_add_head_rcu(&napi->napi_hash_node,
6276                            &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
6277
6278         spin_unlock(&napi_hash_lock);
6279 }
6280
6281 /* Warning : caller is responsible to make sure rcu grace period
6282  * is respected before freeing memory containing @napi
6283  */
6284 static void napi_hash_del(struct napi_struct *napi)
6285 {
6286         spin_lock(&napi_hash_lock);
6287
6288         hlist_del_init_rcu(&napi->napi_hash_node);
6289
6290         spin_unlock(&napi_hash_lock);
6291 }
6292
6293 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
6294 {
6295         struct napi_struct *napi;
6296
6297         napi = container_of(timer, struct napi_struct, timer);
6298
6299         /* Note : we use a relaxed variant of napi_schedule_prep() not setting
6300          * NAPI_STATE_MISSED, since we do not react to a device IRQ.
6301          */
6302         if (!napi_disable_pending(napi) &&
6303             !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) {
6304                 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6305                 __napi_schedule_irqoff(napi);
6306         }
6307
6308         return HRTIMER_NORESTART;
6309 }
6310
6311 static void init_gro_hash(struct napi_struct *napi)
6312 {
6313         int i;
6314
6315         for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6316                 INIT_LIST_HEAD(&napi->gro_hash[i].list);
6317                 napi->gro_hash[i].count = 0;
6318         }
6319         napi->gro_bitmask = 0;
6320 }
6321
6322 int dev_set_threaded(struct net_device *dev, bool threaded)
6323 {
6324         struct napi_struct *napi;
6325         int err = 0;
6326
6327         if (dev->threaded == threaded)
6328                 return 0;
6329
6330         if (threaded) {
6331                 list_for_each_entry(napi, &dev->napi_list, dev_list) {
6332                         if (!napi->thread) {
6333                                 err = napi_kthread_create(napi);
6334                                 if (err) {
6335                                         threaded = false;
6336                                         break;
6337                                 }
6338                         }
6339                 }
6340         }
6341
6342         dev->threaded = threaded;
6343
6344         /* Make sure kthread is created before THREADED bit
6345          * is set.
6346          */
6347         smp_mb__before_atomic();
6348
6349         /* Setting/unsetting threaded mode on a napi might not immediately
6350          * take effect, if the current napi instance is actively being
6351          * polled. In this case, the switch between threaded mode and
6352          * softirq mode will happen in the next round of napi_schedule().
6353          * This should not cause hiccups/stalls to the live traffic.
6354          */
6355         list_for_each_entry(napi, &dev->napi_list, dev_list) {
6356                 if (threaded)
6357                         set_bit(NAPI_STATE_THREADED, &napi->state);
6358                 else
6359                         clear_bit(NAPI_STATE_THREADED, &napi->state);
6360         }
6361
6362         return err;
6363 }
6364 EXPORT_SYMBOL(dev_set_threaded);
6365
6366 void netif_napi_add_weight(struct net_device *dev, struct napi_struct *napi,
6367                            int (*poll)(struct napi_struct *, int), int weight)
6368 {
6369         if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state)))
6370                 return;
6371
6372         INIT_LIST_HEAD(&napi->poll_list);
6373         INIT_HLIST_NODE(&napi->napi_hash_node);
6374         hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
6375         napi->timer.function = napi_watchdog;
6376         init_gro_hash(napi);
6377         napi->skb = NULL;
6378         INIT_LIST_HEAD(&napi->rx_list);
6379         napi->rx_count = 0;
6380         napi->poll = poll;
6381         if (weight > NAPI_POLL_WEIGHT)
6382                 netdev_err_once(dev, "%s() called with weight %d\n", __func__,
6383                                 weight);
6384         napi->weight = weight;
6385         napi->dev = dev;
6386 #ifdef CONFIG_NETPOLL
6387         napi->poll_owner = -1;
6388 #endif
6389         napi->list_owner = -1;
6390         set_bit(NAPI_STATE_SCHED, &napi->state);
6391         set_bit(NAPI_STATE_NPSVC, &napi->state);
6392         list_add_rcu(&napi->dev_list, &dev->napi_list);
6393         napi_hash_add(napi);
6394         napi_get_frags_check(napi);
6395         /* Create kthread for this napi if dev->threaded is set.
6396          * Clear dev->threaded if kthread creation failed so that
6397          * threaded mode will not be enabled in napi_enable().
6398          */
6399         if (dev->threaded && napi_kthread_create(napi))
6400                 dev->threaded = 0;
6401 }
6402 EXPORT_SYMBOL(netif_napi_add_weight);
6403
6404 void napi_disable(struct napi_struct *n)
6405 {
6406         unsigned long val, new;
6407
6408         might_sleep();
6409         set_bit(NAPI_STATE_DISABLE, &n->state);
6410
6411         val = READ_ONCE(n->state);
6412         do {
6413                 while (val & (NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC)) {
6414                         usleep_range(20, 200);
6415                         val = READ_ONCE(n->state);
6416                 }
6417
6418                 new = val | NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC;
6419                 new &= ~(NAPIF_STATE_THREADED | NAPIF_STATE_PREFER_BUSY_POLL);
6420         } while (!try_cmpxchg(&n->state, &val, new));
6421
6422         hrtimer_cancel(&n->timer);
6423
6424         clear_bit(NAPI_STATE_DISABLE, &n->state);
6425 }
6426 EXPORT_SYMBOL(napi_disable);
6427
6428 /**
6429  *      napi_enable - enable NAPI scheduling
6430  *      @n: NAPI context
6431  *
6432  * Resume NAPI from being scheduled on this context.
6433  * Must be paired with napi_disable.
6434  */
6435 void napi_enable(struct napi_struct *n)
6436 {
6437         unsigned long new, val = READ_ONCE(n->state);
6438
6439         do {
6440                 BUG_ON(!test_bit(NAPI_STATE_SCHED, &val));
6441
6442                 new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC);
6443                 if (n->dev->threaded && n->thread)
6444                         new |= NAPIF_STATE_THREADED;
6445         } while (!try_cmpxchg(&n->state, &val, new));
6446 }
6447 EXPORT_SYMBOL(napi_enable);
6448
6449 static void flush_gro_hash(struct napi_struct *napi)
6450 {
6451         int i;
6452
6453         for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6454                 struct sk_buff *skb, *n;
6455
6456                 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list)
6457                         kfree_skb(skb);
6458                 napi->gro_hash[i].count = 0;
6459         }
6460 }
6461
6462 /* Must be called in process context */
6463 void __netif_napi_del(struct napi_struct *napi)
6464 {
6465         if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state))
6466                 return;
6467
6468         napi_hash_del(napi);
6469         list_del_rcu(&napi->dev_list);
6470         napi_free_frags(napi);
6471
6472         flush_gro_hash(napi);
6473         napi->gro_bitmask = 0;
6474
6475         if (napi->thread) {
6476                 kthread_stop(napi->thread);
6477                 napi->thread = NULL;
6478         }
6479 }
6480 EXPORT_SYMBOL(__netif_napi_del);
6481
6482 static int __napi_poll(struct napi_struct *n, bool *repoll)
6483 {
6484         int work, weight;
6485
6486         weight = n->weight;
6487
6488         /* This NAPI_STATE_SCHED test is for avoiding a race
6489          * with netpoll's poll_napi().  Only the entity which
6490          * obtains the lock and sees NAPI_STATE_SCHED set will
6491          * actually make the ->poll() call.  Therefore we avoid
6492          * accidentally calling ->poll() when NAPI is not scheduled.
6493          */
6494         work = 0;
6495         if (test_bit(NAPI_STATE_SCHED, &n->state)) {
6496                 work = n->poll(n, weight);
6497                 trace_napi_poll(n, work, weight);
6498         }
6499
6500         if (unlikely(work > weight))
6501                 netdev_err_once(n->dev, "NAPI poll function %pS returned %d, exceeding its budget of %d.\n",
6502                                 n->poll, work, weight);
6503
6504         if (likely(work < weight))
6505                 return work;
6506
6507         /* Drivers must not modify the NAPI state if they
6508          * consume the entire weight.  In such cases this code
6509          * still "owns" the NAPI instance and therefore can
6510          * move the instance around on the list at-will.
6511          */
6512         if (unlikely(napi_disable_pending(n))) {
6513                 napi_complete(n);
6514                 return work;
6515         }
6516
6517         /* The NAPI context has more processing work, but busy-polling
6518          * is preferred. Exit early.
6519          */
6520         if (napi_prefer_busy_poll(n)) {
6521                 if (napi_complete_done(n, work)) {
6522                         /* If timeout is not set, we need to make sure
6523                          * that the NAPI is re-scheduled.
6524                          */
6525                         napi_schedule(n);
6526                 }
6527                 return work;
6528         }
6529
6530         if (n->gro_bitmask) {
6531                 /* flush too old packets
6532                  * If HZ < 1000, flush all packets.
6533                  */
6534                 napi_gro_flush(n, HZ >= 1000);
6535         }
6536
6537         gro_normal_list(n);
6538
6539         /* Some drivers may have called napi_schedule
6540          * prior to exhausting their budget.
6541          */
6542         if (unlikely(!list_empty(&n->poll_list))) {
6543                 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
6544                              n->dev ? n->dev->name : "backlog");
6545                 return work;
6546         }
6547
6548         *repoll = true;
6549
6550         return work;
6551 }
6552
6553 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
6554 {
6555         bool do_repoll = false;
6556         void *have;
6557         int work;
6558
6559         list_del_init(&n->poll_list);
6560
6561         have = netpoll_poll_lock(n);
6562
6563         work = __napi_poll(n, &do_repoll);
6564
6565         if (do_repoll)
6566                 list_add_tail(&n->poll_list, repoll);
6567
6568         netpoll_poll_unlock(have);
6569
6570         return work;
6571 }
6572
6573 static int napi_thread_wait(struct napi_struct *napi)
6574 {
6575         bool woken = false;
6576
6577         set_current_state(TASK_INTERRUPTIBLE);
6578
6579         while (!kthread_should_stop()) {
6580                 /* Testing SCHED_THREADED bit here to make sure the current
6581                  * kthread owns this napi and could poll on this napi.
6582                  * Testing SCHED bit is not enough because SCHED bit might be
6583                  * set by some other busy poll thread or by napi_disable().
6584                  */
6585                 if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state) || woken) {
6586                         WARN_ON(!list_empty(&napi->poll_list));
6587                         __set_current_state(TASK_RUNNING);
6588                         return 0;
6589                 }
6590
6591                 schedule();
6592                 /* woken being true indicates this thread owns this napi. */
6593                 woken = true;
6594                 set_current_state(TASK_INTERRUPTIBLE);
6595         }
6596         __set_current_state(TASK_RUNNING);
6597
6598         return -1;
6599 }
6600
6601 static void skb_defer_free_flush(struct softnet_data *sd)
6602 {
6603         struct sk_buff *skb, *next;
6604
6605         /* Paired with WRITE_ONCE() in skb_attempt_defer_free() */
6606         if (!READ_ONCE(sd->defer_list))
6607                 return;
6608
6609         spin_lock(&sd->defer_lock);
6610         skb = sd->defer_list;
6611         sd->defer_list = NULL;
6612         sd->defer_count = 0;
6613         spin_unlock(&sd->defer_lock);
6614
6615         while (skb != NULL) {
6616                 next = skb->next;
6617                 napi_consume_skb(skb, 1);
6618                 skb = next;
6619         }
6620 }
6621
6622 static int napi_threaded_poll(void *data)
6623 {
6624         struct napi_struct *napi = data;
6625         struct softnet_data *sd;
6626         void *have;
6627
6628         while (!napi_thread_wait(napi)) {
6629                 for (;;) {
6630                         bool repoll = false;
6631
6632                         local_bh_disable();
6633                         sd = this_cpu_ptr(&softnet_data);
6634
6635                         have = netpoll_poll_lock(napi);
6636                         __napi_poll(napi, &repoll);
6637                         netpoll_poll_unlock(have);
6638
6639                         skb_defer_free_flush(sd);
6640                         local_bh_enable();
6641
6642                         if (!repoll)
6643                                 break;
6644
6645                         cond_resched();
6646                 }
6647         }
6648         return 0;
6649 }
6650
6651 static __latent_entropy void net_rx_action(struct softirq_action *h)
6652 {
6653         struct softnet_data *sd = this_cpu_ptr(&softnet_data);
6654         unsigned long time_limit = jiffies +
6655                 usecs_to_jiffies(READ_ONCE(netdev_budget_usecs));
6656         int budget = READ_ONCE(netdev_budget);
6657         LIST_HEAD(list);
6658         LIST_HEAD(repoll);
6659
6660 start:
6661         sd->in_net_rx_action = true;
6662         local_irq_disable();
6663         list_splice_init(&sd->poll_list, &list);
6664         local_irq_enable();
6665
6666         for (;;) {
6667                 struct napi_struct *n;
6668
6669                 skb_defer_free_flush(sd);
6670
6671                 if (list_empty(&list)) {
6672                         if (list_empty(&repoll)) {
6673                                 sd->in_net_rx_action = false;
6674                                 barrier();
6675                                 /* We need to check if ____napi_schedule()
6676                                  * had refilled poll_list while
6677                                  * sd->in_net_rx_action was true.
6678                                  */
6679                                 if (!list_empty(&sd->poll_list))
6680                                         goto start;
6681                                 if (!sd_has_rps_ipi_waiting(sd))
6682                                         goto end;
6683                         }
6684                         break;
6685                 }
6686
6687                 n = list_first_entry(&list, struct napi_struct, poll_list);
6688                 budget -= napi_poll(n, &repoll);
6689
6690                 /* If softirq window is exhausted then punt.
6691                  * Allow this to run for 2 jiffies since which will allow
6692                  * an average latency of 1.5/HZ.
6693                  */
6694                 if (unlikely(budget <= 0 ||
6695                              time_after_eq(jiffies, time_limit))) {
6696                         sd->time_squeeze++;
6697                         break;
6698                 }
6699         }
6700
6701         local_irq_disable();
6702
6703         list_splice_tail_init(&sd->poll_list, &list);
6704         list_splice_tail(&repoll, &list);
6705         list_splice(&list, &sd->poll_list);
6706         if (!list_empty(&sd->poll_list))
6707                 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
6708         else
6709                 sd->in_net_rx_action = false;
6710
6711         net_rps_action_and_irq_enable(sd);
6712 end:;
6713 }
6714
6715 struct netdev_adjacent {
6716         struct net_device *dev;
6717         netdevice_tracker dev_tracker;
6718
6719         /* upper master flag, there can only be one master device per list */
6720         bool master;
6721
6722         /* lookup ignore flag */
6723         bool ignore;
6724
6725         /* counter for the number of times this device was added to us */
6726         u16 ref_nr;
6727
6728         /* private field for the users */
6729         void *private;
6730
6731         struct list_head list;
6732         struct rcu_head rcu;
6733 };
6734
6735 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
6736                                                  struct list_head *adj_list)
6737 {
6738         struct netdev_adjacent *adj;
6739
6740         list_for_each_entry(adj, adj_list, list) {
6741                 if (adj->dev == adj_dev)
6742                         return adj;
6743         }
6744         return NULL;
6745 }
6746
6747 static int ____netdev_has_upper_dev(struct net_device *upper_dev,
6748                                     struct netdev_nested_priv *priv)
6749 {
6750         struct net_device *dev = (struct net_device *)priv->data;
6751
6752         return upper_dev == dev;
6753 }
6754
6755 /**
6756  * netdev_has_upper_dev - Check if device is linked to an upper device
6757  * @dev: device
6758  * @upper_dev: upper device to check
6759  *
6760  * Find out if a device is linked to specified upper device and return true
6761  * in case it is. Note that this checks only immediate upper device,
6762  * not through a complete stack of devices. The caller must hold the RTNL lock.
6763  */
6764 bool netdev_has_upper_dev(struct net_device *dev,
6765                           struct net_device *upper_dev)
6766 {
6767         struct netdev_nested_priv priv = {
6768                 .data = (void *)upper_dev,
6769         };
6770
6771         ASSERT_RTNL();
6772
6773         return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
6774                                              &priv);
6775 }
6776 EXPORT_SYMBOL(netdev_has_upper_dev);
6777
6778 /**
6779  * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device
6780  * @dev: device
6781  * @upper_dev: upper device to check
6782  *
6783  * Find out if a device is linked to specified upper device and return true
6784  * in case it is. Note that this checks the entire upper device chain.
6785  * The caller must hold rcu lock.
6786  */
6787
6788 bool netdev_has_upper_dev_all_rcu(struct net_device *dev,
6789                                   struct net_device *upper_dev)
6790 {
6791         struct netdev_nested_priv priv = {
6792                 .data = (void *)upper_dev,
6793         };
6794
6795         return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
6796                                                &priv);
6797 }
6798 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu);
6799
6800 /**
6801  * netdev_has_any_upper_dev - Check if device is linked to some device
6802  * @dev: device
6803  *
6804  * Find out if a device is linked to an upper device and return true in case
6805  * it is. The caller must hold the RTNL lock.
6806  */
6807 bool netdev_has_any_upper_dev(struct net_device *dev)
6808 {
6809         ASSERT_RTNL();
6810
6811         return !list_empty(&dev->adj_list.upper);
6812 }
6813 EXPORT_SYMBOL(netdev_has_any_upper_dev);
6814
6815 /**
6816  * netdev_master_upper_dev_get - Get master upper device
6817  * @dev: device
6818  *
6819  * Find a master upper device and return pointer to it or NULL in case
6820  * it's not there. The caller must hold the RTNL lock.
6821  */
6822 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
6823 {
6824         struct netdev_adjacent *upper;
6825
6826         ASSERT_RTNL();
6827
6828         if (list_empty(&dev->adj_list.upper))
6829                 return NULL;
6830
6831         upper = list_first_entry(&dev->adj_list.upper,
6832                                  struct netdev_adjacent, list);
6833         if (likely(upper->master))
6834                 return upper->dev;
6835         return NULL;
6836 }
6837 EXPORT_SYMBOL(netdev_master_upper_dev_get);
6838
6839 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev)
6840 {
6841         struct netdev_adjacent *upper;
6842
6843         ASSERT_RTNL();
6844
6845         if (list_empty(&dev->adj_list.upper))
6846                 return NULL;
6847
6848         upper = list_first_entry(&dev->adj_list.upper,
6849                                  struct netdev_adjacent, list);
6850         if (likely(upper->master) && !upper->ignore)
6851                 return upper->dev;
6852         return NULL;
6853 }
6854
6855 /**
6856  * netdev_has_any_lower_dev - Check if device is linked to some device
6857  * @dev: device
6858  *
6859  * Find out if a device is linked to a lower device and return true in case
6860  * it is. The caller must hold the RTNL lock.
6861  */
6862 static bool netdev_has_any_lower_dev(struct net_device *dev)
6863 {
6864         ASSERT_RTNL();
6865
6866         return !list_empty(&dev->adj_list.lower);
6867 }
6868
6869 void *netdev_adjacent_get_private(struct list_head *adj_list)
6870 {
6871         struct netdev_adjacent *adj;
6872
6873         adj = list_entry(adj_list, struct netdev_adjacent, list);
6874
6875         return adj->private;
6876 }
6877 EXPORT_SYMBOL(netdev_adjacent_get_private);
6878
6879 /**
6880  * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
6881  * @dev: device
6882  * @iter: list_head ** of the current position
6883  *
6884  * Gets the next device from the dev's upper list, starting from iter
6885  * position. The caller must hold RCU read lock.
6886  */
6887 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
6888                                                  struct list_head **iter)
6889 {
6890         struct netdev_adjacent *upper;
6891
6892         WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6893
6894         upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6895
6896         if (&upper->list == &dev->adj_list.upper)
6897                 return NULL;
6898
6899         *iter = &upper->list;
6900
6901         return upper->dev;
6902 }
6903 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
6904
6905 static struct net_device *__netdev_next_upper_dev(struct net_device *dev,
6906                                                   struct list_head **iter,
6907                                                   bool *ignore)
6908 {
6909         struct netdev_adjacent *upper;
6910
6911         upper = list_entry((*iter)->next, struct netdev_adjacent, list);
6912
6913         if (&upper->list == &dev->adj_list.upper)
6914                 return NULL;
6915
6916         *iter = &upper->list;
6917         *ignore = upper->ignore;
6918
6919         return upper->dev;
6920 }
6921
6922 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev,
6923                                                     struct list_head **iter)
6924 {
6925         struct netdev_adjacent *upper;
6926
6927         WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6928
6929         upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6930
6931         if (&upper->list == &dev->adj_list.upper)
6932                 return NULL;
6933
6934         *iter = &upper->list;
6935
6936         return upper->dev;
6937 }
6938
6939 static int __netdev_walk_all_upper_dev(struct net_device *dev,
6940                                        int (*fn)(struct net_device *dev,
6941                                          struct netdev_nested_priv *priv),
6942                                        struct netdev_nested_priv *priv)
6943 {
6944         struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
6945         struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
6946         int ret, cur = 0;
6947         bool ignore;
6948
6949         now = dev;
6950         iter = &dev->adj_list.upper;
6951
6952         while (1) {
6953                 if (now != dev) {
6954                         ret = fn(now, priv);
6955                         if (ret)
6956                                 return ret;
6957                 }
6958
6959                 next = NULL;
6960                 while (1) {
6961                         udev = __netdev_next_upper_dev(now, &iter, &ignore);
6962                         if (!udev)
6963                                 break;
6964                         if (ignore)
6965                                 continue;
6966
6967                         next = udev;
6968                         niter = &udev->adj_list.upper;
6969                         dev_stack[cur] = now;
6970                         iter_stack[cur++] = iter;
6971                         break;
6972                 }
6973
6974                 if (!next) {
6975                         if (!cur)
6976                                 return 0;
6977                         next = dev_stack[--cur];
6978                         niter = iter_stack[cur];
6979                 }
6980
6981                 now = next;
6982                 iter = niter;
6983         }
6984
6985         return 0;
6986 }
6987
6988 int netdev_walk_all_upper_dev_rcu(struct net_device *dev,
6989                                   int (*fn)(struct net_device *dev,
6990                                             struct netdev_nested_priv *priv),
6991                                   struct netdev_nested_priv *priv)
6992 {
6993         struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
6994         struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
6995         int ret, cur = 0;
6996
6997         now = dev;
6998         iter = &dev->adj_list.upper;
6999
7000         while (1) {
7001                 if (now != dev) {
7002                         ret = fn(now, priv);
7003                         if (ret)
7004                                 return ret;
7005                 }
7006
7007                 next = NULL;
7008                 while (1) {
7009                         udev = netdev_next_upper_dev_rcu(now, &iter);
7010                         if (!udev)
7011                                 break;
7012
7013                         next = udev;
7014                         niter = &udev->adj_list.upper;
7015                         dev_stack[cur] = now;
7016                         iter_stack[cur++] = iter;
7017                         break;
7018                 }
7019
7020                 if (!next) {
7021                         if (!cur)
7022                                 return 0;
7023                         next = dev_stack[--cur];
7024                         niter = iter_stack[cur];
7025                 }
7026
7027                 now = next;
7028                 iter = niter;
7029         }
7030
7031         return 0;
7032 }
7033 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu);
7034
7035 static bool __netdev_has_upper_dev(struct net_device *dev,
7036                                    struct net_device *upper_dev)
7037 {
7038         struct netdev_nested_priv priv = {
7039                 .flags = 0,
7040                 .data = (void *)upper_dev,
7041         };
7042
7043         ASSERT_RTNL();
7044
7045         return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev,
7046                                            &priv);
7047 }
7048
7049 /**
7050  * netdev_lower_get_next_private - Get the next ->private from the
7051  *                                 lower neighbour list
7052  * @dev: device
7053  * @iter: list_head ** of the current position
7054  *
7055  * Gets the next netdev_adjacent->private from the dev's lower neighbour
7056  * list, starting from iter position. The caller must hold either hold the
7057  * RTNL lock or its own locking that guarantees that the neighbour lower
7058  * list will remain unchanged.
7059  */
7060 void *netdev_lower_get_next_private(struct net_device *dev,
7061                                     struct list_head **iter)
7062 {
7063         struct netdev_adjacent *lower;
7064
7065         lower = list_entry(*iter, struct netdev_adjacent, list);
7066
7067         if (&lower->list == &dev->adj_list.lower)
7068                 return NULL;
7069
7070         *iter = lower->list.next;
7071
7072         return lower->private;
7073 }
7074 EXPORT_SYMBOL(netdev_lower_get_next_private);
7075
7076 /**
7077  * netdev_lower_get_next_private_rcu - Get the next ->private from the
7078  *                                     lower neighbour list, RCU
7079  *                                     variant
7080  * @dev: device
7081  * @iter: list_head ** of the current position
7082  *
7083  * Gets the next netdev_adjacent->private from the dev's lower neighbour
7084  * list, starting from iter position. The caller must hold RCU read lock.
7085  */
7086 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
7087                                         struct list_head **iter)
7088 {
7089         struct netdev_adjacent *lower;
7090
7091         WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
7092
7093         lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7094
7095         if (&lower->list == &dev->adj_list.lower)
7096                 return NULL;
7097
7098         *iter = &lower->list;
7099
7100         return lower->private;
7101 }
7102 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
7103
7104 /**
7105  * netdev_lower_get_next - Get the next device from the lower neighbour
7106  *                         list
7107  * @dev: device
7108  * @iter: list_head ** of the current position
7109  *
7110  * Gets the next netdev_adjacent from the dev's lower neighbour
7111  * list, starting from iter position. The caller must hold RTNL lock or
7112  * its own locking that guarantees that the neighbour lower
7113  * list will remain unchanged.
7114  */
7115 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
7116 {
7117         struct netdev_adjacent *lower;
7118
7119         lower = list_entry(*iter, struct netdev_adjacent, list);
7120
7121         if (&lower->list == &dev->adj_list.lower)
7122                 return NULL;
7123
7124         *iter = lower->list.next;
7125
7126         return lower->dev;
7127 }
7128 EXPORT_SYMBOL(netdev_lower_get_next);
7129
7130 static struct net_device *netdev_next_lower_dev(struct net_device *dev,
7131                                                 struct list_head **iter)
7132 {
7133         struct netdev_adjacent *lower;
7134
7135         lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7136
7137         if (&lower->list == &dev->adj_list.lower)
7138                 return NULL;
7139
7140         *iter = &lower->list;
7141
7142         return lower->dev;
7143 }
7144
7145 static struct net_device *__netdev_next_lower_dev(struct net_device *dev,
7146                                                   struct list_head **iter,
7147                                                   bool *ignore)
7148 {
7149         struct netdev_adjacent *lower;
7150
7151         lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7152
7153         if (&lower->list == &dev->adj_list.lower)
7154                 return NULL;
7155
7156         *iter = &lower->list;
7157         *ignore = lower->ignore;
7158
7159         return lower->dev;
7160 }
7161
7162 int netdev_walk_all_lower_dev(struct net_device *dev,
7163                               int (*fn)(struct net_device *dev,
7164                                         struct netdev_nested_priv *priv),
7165                               struct netdev_nested_priv *priv)
7166 {
7167         struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7168         struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7169         int ret, cur = 0;
7170
7171         now = dev;
7172         iter = &dev->adj_list.lower;
7173
7174         while (1) {
7175                 if (now != dev) {
7176                         ret = fn(now, priv);
7177                         if (ret)
7178                                 return ret;
7179                 }
7180
7181                 next = NULL;
7182                 while (1) {
7183                         ldev = netdev_next_lower_dev(now, &iter);
7184                         if (!ldev)
7185                                 break;
7186
7187                         next = ldev;
7188                         niter = &ldev->adj_list.lower;
7189                         dev_stack[cur] = now;
7190                         iter_stack[cur++] = iter;
7191                         break;
7192                 }
7193
7194                 if (!next) {
7195                         if (!cur)
7196                                 return 0;
7197                         next = dev_stack[--cur];
7198                         niter = iter_stack[cur];
7199                 }
7200
7201                 now = next;
7202                 iter = niter;
7203         }
7204
7205         return 0;
7206 }
7207 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev);
7208
7209 static int __netdev_walk_all_lower_dev(struct net_device *dev,
7210                                        int (*fn)(struct net_device *dev,
7211                                          struct netdev_nested_priv *priv),
7212                                        struct netdev_nested_priv *priv)
7213 {
7214         struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7215         struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7216         int ret, cur = 0;
7217         bool ignore;
7218
7219         now = dev;
7220         iter = &dev->adj_list.lower;
7221
7222         while (1) {
7223                 if (now != dev) {
7224                         ret = fn(now, priv);
7225                         if (ret)
7226                                 return ret;
7227                 }
7228
7229                 next = NULL;
7230                 while (1) {
7231                         ldev = __netdev_next_lower_dev(now, &iter, &ignore);
7232                         if (!ldev)
7233                                 break;
7234                         if (ignore)
7235                                 continue;
7236
7237                         next = ldev;
7238                         niter = &ldev->adj_list.lower;
7239                         dev_stack[cur] = now;
7240                         iter_stack[cur++] = iter;
7241                         break;
7242                 }
7243
7244                 if (!next) {
7245                         if (!cur)
7246                                 return 0;
7247                         next = dev_stack[--cur];
7248                         niter = iter_stack[cur];
7249                 }
7250
7251                 now = next;
7252                 iter = niter;
7253         }
7254
7255         return 0;
7256 }
7257
7258 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev,
7259                                              struct list_head **iter)
7260 {
7261         struct netdev_adjacent *lower;
7262
7263         lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7264         if (&lower->list == &dev->adj_list.lower)
7265                 return NULL;
7266
7267         *iter = &lower->list;
7268
7269         return lower->dev;
7270 }
7271 EXPORT_SYMBOL(netdev_next_lower_dev_rcu);
7272
7273 static u8 __netdev_upper_depth(struct net_device *dev)
7274 {
7275         struct net_device *udev;
7276         struct list_head *iter;
7277         u8 max_depth = 0;
7278         bool ignore;
7279
7280         for (iter = &dev->adj_list.upper,
7281              udev = __netdev_next_upper_dev(dev, &iter, &ignore);
7282              udev;
7283              udev = __netdev_next_upper_dev(dev, &iter, &ignore)) {
7284                 if (ignore)
7285                         continue;
7286                 if (max_depth < udev->upper_level)
7287                         max_depth = udev->upper_level;
7288         }
7289
7290         return max_depth;
7291 }
7292
7293 static u8 __netdev_lower_depth(struct net_device *dev)
7294 {
7295         struct net_device *ldev;
7296         struct list_head *iter;
7297         u8 max_depth = 0;
7298         bool ignore;
7299
7300         for (iter = &dev->adj_list.lower,
7301              ldev = __netdev_next_lower_dev(dev, &iter, &ignore);
7302              ldev;
7303              ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) {
7304                 if (ignore)
7305                         continue;
7306                 if (max_depth < ldev->lower_level)
7307                         max_depth = ldev->lower_level;
7308         }
7309
7310         return max_depth;
7311 }
7312
7313 static int __netdev_update_upper_level(struct net_device *dev,
7314                                        struct netdev_nested_priv *__unused)
7315 {
7316         dev->upper_level = __netdev_upper_depth(dev) + 1;
7317         return 0;
7318 }
7319
7320 #ifdef CONFIG_LOCKDEP
7321 static LIST_HEAD(net_unlink_list);
7322
7323 static void net_unlink_todo(struct net_device *dev)
7324 {
7325         if (list_empty(&dev->unlink_list))
7326                 list_add_tail(&dev->unlink_list, &net_unlink_list);
7327 }
7328 #endif
7329
7330 static int __netdev_update_lower_level(struct net_device *dev,
7331                                        struct netdev_nested_priv *priv)
7332 {
7333         dev->lower_level = __netdev_lower_depth(dev) + 1;
7334
7335 #ifdef CONFIG_LOCKDEP
7336         if (!priv)
7337                 return 0;
7338
7339         if (priv->flags & NESTED_SYNC_IMM)
7340                 dev->nested_level = dev->lower_level - 1;
7341         if (priv->flags & NESTED_SYNC_TODO)
7342                 net_unlink_todo(dev);
7343 #endif
7344         return 0;
7345 }
7346
7347 int netdev_walk_all_lower_dev_rcu(struct net_device *dev,
7348                                   int (*fn)(struct net_device *dev,
7349                                             struct netdev_nested_priv *priv),
7350                                   struct netdev_nested_priv *priv)
7351 {
7352         struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7353         struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7354         int ret, cur = 0;
7355
7356         now = dev;
7357         iter = &dev->adj_list.lower;
7358
7359         while (1) {
7360                 if (now != dev) {
7361                         ret = fn(now, priv);
7362                         if (ret)
7363                                 return ret;
7364                 }
7365
7366                 next = NULL;
7367                 while (1) {
7368                         ldev = netdev_next_lower_dev_rcu(now, &iter);
7369                         if (!ldev)
7370                                 break;
7371
7372                         next = ldev;
7373                         niter = &ldev->adj_list.lower;
7374                         dev_stack[cur] = now;
7375                         iter_stack[cur++] = iter;
7376                         break;
7377                 }
7378
7379                 if (!next) {
7380                         if (!cur)
7381                                 return 0;
7382                         next = dev_stack[--cur];
7383                         niter = iter_stack[cur];
7384                 }
7385
7386                 now = next;
7387                 iter = niter;
7388         }
7389
7390         return 0;
7391 }
7392 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu);
7393
7394 /**
7395  * netdev_lower_get_first_private_rcu - Get the first ->private from the
7396  *                                     lower neighbour list, RCU
7397  *                                     variant
7398  * @dev: device
7399  *
7400  * Gets the first netdev_adjacent->private from the dev's lower neighbour
7401  * list. The caller must hold RCU read lock.
7402  */
7403 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
7404 {
7405         struct netdev_adjacent *lower;
7406
7407         lower = list_first_or_null_rcu(&dev->adj_list.lower,
7408                         struct netdev_adjacent, list);
7409         if (lower)
7410                 return lower->private;
7411         return NULL;
7412 }
7413 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
7414
7415 /**
7416  * netdev_master_upper_dev_get_rcu - Get master upper device
7417  * @dev: device
7418  *
7419  * Find a master upper device and return pointer to it or NULL in case
7420  * it's not there. The caller must hold the RCU read lock.
7421  */
7422 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
7423 {
7424         struct netdev_adjacent *upper;
7425
7426         upper = list_first_or_null_rcu(&dev->adj_list.upper,
7427                                        struct netdev_adjacent, list);
7428         if (upper && likely(upper->master))
7429                 return upper->dev;
7430         return NULL;
7431 }
7432 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
7433
7434 static int netdev_adjacent_sysfs_add(struct net_device *dev,
7435                               struct net_device *adj_dev,
7436                               struct list_head *dev_list)
7437 {
7438         char linkname[IFNAMSIZ+7];
7439
7440         sprintf(linkname, dev_list == &dev->adj_list.upper ?
7441                 "upper_%s" : "lower_%s", adj_dev->name);
7442         return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
7443                                  linkname);
7444 }
7445 static void netdev_adjacent_sysfs_del(struct net_device *dev,
7446                                char *name,
7447                                struct list_head *dev_list)
7448 {
7449         char linkname[IFNAMSIZ+7];
7450
7451         sprintf(linkname, dev_list == &dev->adj_list.upper ?
7452                 "upper_%s" : "lower_%s", name);
7453         sysfs_remove_link(&(dev->dev.kobj), linkname);
7454 }
7455
7456 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
7457                                                  struct net_device *adj_dev,
7458                                                  struct list_head *dev_list)
7459 {
7460         return (dev_list == &dev->adj_list.upper ||
7461                 dev_list == &dev->adj_list.lower) &&
7462                 net_eq(dev_net(dev), dev_net(adj_dev));
7463 }
7464
7465 static int __netdev_adjacent_dev_insert(struct net_device *dev,
7466                                         struct net_device *adj_dev,
7467                                         struct list_head *dev_list,
7468                                         void *private, bool master)
7469 {
7470         struct netdev_adjacent *adj;
7471         int ret;
7472
7473         adj = __netdev_find_adj(adj_dev, dev_list);
7474
7475         if (adj) {
7476                 adj->ref_nr += 1;
7477                 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n",
7478                          dev->name, adj_dev->name, adj->ref_nr);
7479
7480                 return 0;
7481         }
7482
7483         adj = kmalloc(sizeof(*adj), GFP_KERNEL);
7484         if (!adj)
7485                 return -ENOMEM;
7486
7487         adj->dev = adj_dev;
7488         adj->master = master;
7489         adj->ref_nr = 1;
7490         adj->private = private;
7491         adj->ignore = false;
7492         netdev_hold(adj_dev, &adj->dev_tracker, GFP_KERNEL);
7493
7494         pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n",
7495                  dev->name, adj_dev->name, adj->ref_nr, adj_dev->name);
7496
7497         if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
7498                 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
7499                 if (ret)
7500                         goto free_adj;
7501         }
7502
7503         /* Ensure that master link is always the first item in list. */
7504         if (master) {
7505                 ret = sysfs_create_link(&(dev->dev.kobj),
7506                                         &(adj_dev->dev.kobj), "master");
7507                 if (ret)
7508                         goto remove_symlinks;
7509
7510                 list_add_rcu(&adj->list, dev_list);
7511         } else {
7512                 list_add_tail_rcu(&adj->list, dev_list);
7513         }
7514
7515         return 0;
7516
7517 remove_symlinks:
7518         if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7519                 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7520 free_adj:
7521         netdev_put(adj_dev, &adj->dev_tracker);
7522         kfree(adj);
7523
7524         return ret;
7525 }
7526
7527 static void __netdev_adjacent_dev_remove(struct net_device *dev,
7528                                          struct net_device *adj_dev,
7529                                          u16 ref_nr,
7530                                          struct list_head *dev_list)
7531 {
7532         struct netdev_adjacent *adj;
7533
7534         pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n",
7535                  dev->name, adj_dev->name, ref_nr);
7536
7537         adj = __netdev_find_adj(adj_dev, dev_list);
7538
7539         if (!adj) {
7540                 pr_err("Adjacency does not exist for device %s from %s\n",
7541                        dev->name, adj_dev->name);
7542                 WARN_ON(1);
7543                 return;
7544         }
7545
7546         if (adj->ref_nr > ref_nr) {
7547                 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n",
7548                          dev->name, adj_dev->name, ref_nr,
7549                          adj->ref_nr - ref_nr);
7550                 adj->ref_nr -= ref_nr;
7551                 return;
7552         }
7553
7554         if (adj->master)
7555                 sysfs_remove_link(&(dev->dev.kobj), "master");
7556
7557         if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7558                 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7559
7560         list_del_rcu(&adj->list);
7561         pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n",
7562                  adj_dev->name, dev->name, adj_dev->name);
7563         netdev_put(adj_dev, &adj->dev_tracker);
7564         kfree_rcu(adj, rcu);
7565 }
7566
7567 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
7568                                             struct net_device *upper_dev,
7569                                             struct list_head *up_list,
7570                                             struct list_head *down_list,
7571                                             void *private, bool master)
7572 {
7573         int ret;
7574
7575         ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list,
7576                                            private, master);
7577         if (ret)
7578                 return ret;
7579
7580         ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list,
7581                                            private, false);
7582         if (ret) {
7583                 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list);
7584                 return ret;
7585         }
7586
7587         return 0;
7588 }
7589
7590 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
7591                                                struct net_device *upper_dev,
7592                                                u16 ref_nr,
7593                                                struct list_head *up_list,
7594                                                struct list_head *down_list)
7595 {
7596         __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
7597         __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
7598 }
7599
7600 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
7601                                                 struct net_device *upper_dev,
7602                                                 void *private, bool master)
7603 {
7604         return __netdev_adjacent_dev_link_lists(dev, upper_dev,
7605                                                 &dev->adj_list.upper,
7606                                                 &upper_dev->adj_list.lower,
7607                                                 private, master);
7608 }
7609
7610 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
7611                                                    struct net_device *upper_dev)
7612 {
7613         __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
7614                                            &dev->adj_list.upper,
7615                                            &upper_dev->adj_list.lower);
7616 }
7617
7618 static int __netdev_upper_dev_link(struct net_device *dev,
7619                                    struct net_device *upper_dev, bool master,
7620                                    void *upper_priv, void *upper_info,
7621                                    struct netdev_nested_priv *priv,
7622                                    struct netlink_ext_ack *extack)
7623 {
7624         struct netdev_notifier_changeupper_info changeupper_info = {
7625                 .info = {
7626                         .dev = dev,
7627                         .extack = extack,
7628                 },
7629                 .upper_dev = upper_dev,
7630                 .master = master,
7631                 .linking = true,
7632                 .upper_info = upper_info,
7633         };
7634         struct net_device *master_dev;
7635         int ret = 0;
7636
7637         ASSERT_RTNL();
7638
7639         if (dev == upper_dev)
7640                 return -EBUSY;
7641
7642         /* To prevent loops, check if dev is not upper device to upper_dev. */
7643         if (__netdev_has_upper_dev(upper_dev, dev))
7644                 return -EBUSY;
7645
7646         if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV)
7647                 return -EMLINK;
7648
7649         if (!master) {
7650                 if (__netdev_has_upper_dev(dev, upper_dev))
7651                         return -EEXIST;
7652         } else {
7653                 master_dev = __netdev_master_upper_dev_get(dev);
7654                 if (master_dev)
7655                         return master_dev == upper_dev ? -EEXIST : -EBUSY;
7656         }
7657
7658         ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7659                                             &changeupper_info.info);
7660         ret = notifier_to_errno(ret);
7661         if (ret)
7662                 return ret;
7663
7664         ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
7665                                                    master);
7666         if (ret)
7667                 return ret;
7668
7669         ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7670                                             &changeupper_info.info);
7671         ret = notifier_to_errno(ret);
7672         if (ret)
7673                 goto rollback;
7674
7675         __netdev_update_upper_level(dev, NULL);
7676         __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
7677
7678         __netdev_update_lower_level(upper_dev, priv);
7679         __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
7680                                     priv);
7681
7682         return 0;
7683
7684 rollback:
7685         __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7686
7687         return ret;
7688 }
7689
7690 /**
7691  * netdev_upper_dev_link - Add a link to the upper device
7692  * @dev: device
7693  * @upper_dev: new upper device
7694  * @extack: netlink extended ack
7695  *
7696  * Adds a link to device which is upper to this one. The caller must hold
7697  * the RTNL lock. On a failure a negative errno code is returned.
7698  * On success the reference counts are adjusted and the function
7699  * returns zero.
7700  */
7701 int netdev_upper_dev_link(struct net_device *dev,
7702                           struct net_device *upper_dev,
7703                           struct netlink_ext_ack *extack)
7704 {
7705         struct netdev_nested_priv priv = {
7706                 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
7707                 .data = NULL,
7708         };
7709
7710         return __netdev_upper_dev_link(dev, upper_dev, false,
7711                                        NULL, NULL, &priv, extack);
7712 }
7713 EXPORT_SYMBOL(netdev_upper_dev_link);
7714
7715 /**
7716  * netdev_master_upper_dev_link - Add a master link to the upper device
7717  * @dev: device
7718  * @upper_dev: new upper device
7719  * @upper_priv: upper device private
7720  * @upper_info: upper info to be passed down via notifier
7721  * @extack: netlink extended ack
7722  *
7723  * Adds a link to device which is upper to this one. In this case, only
7724  * one master upper device can be linked, although other non-master devices
7725  * might be linked as well. The caller must hold the RTNL lock.
7726  * On a failure a negative errno code is returned. On success the reference
7727  * counts are adjusted and the function returns zero.
7728  */
7729 int netdev_master_upper_dev_link(struct net_device *dev,
7730                                  struct net_device *upper_dev,
7731                                  void *upper_priv, void *upper_info,
7732                                  struct netlink_ext_ack *extack)
7733 {
7734         struct netdev_nested_priv priv = {
7735                 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
7736                 .data = NULL,
7737         };
7738
7739         return __netdev_upper_dev_link(dev, upper_dev, true,
7740                                        upper_priv, upper_info, &priv, extack);
7741 }
7742 EXPORT_SYMBOL(netdev_master_upper_dev_link);
7743
7744 static void __netdev_upper_dev_unlink(struct net_device *dev,
7745                                       struct net_device *upper_dev,
7746                                       struct netdev_nested_priv *priv)
7747 {
7748         struct netdev_notifier_changeupper_info changeupper_info = {
7749                 .info = {
7750                         .dev = dev,
7751                 },
7752                 .upper_dev = upper_dev,
7753                 .linking = false,
7754         };
7755
7756         ASSERT_RTNL();
7757
7758         changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
7759
7760         call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7761                                       &changeupper_info.info);
7762
7763         __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7764
7765         call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7766                                       &changeupper_info.info);
7767
7768         __netdev_update_upper_level(dev, NULL);
7769         __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
7770
7771         __netdev_update_lower_level(upper_dev, priv);
7772         __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
7773                                     priv);
7774 }
7775
7776 /**
7777  * netdev_upper_dev_unlink - Removes a link to upper device
7778  * @dev: device
7779  * @upper_dev: new upper device
7780  *
7781  * Removes a link to device which is upper to this one. The caller must hold
7782  * the RTNL lock.
7783  */
7784 void netdev_upper_dev_unlink(struct net_device *dev,
7785                              struct net_device *upper_dev)
7786 {
7787         struct netdev_nested_priv priv = {
7788                 .flags = NESTED_SYNC_TODO,
7789                 .data = NULL,
7790         };
7791
7792         __netdev_upper_dev_unlink(dev, upper_dev, &priv);
7793 }
7794 EXPORT_SYMBOL(netdev_upper_dev_unlink);
7795
7796 static void __netdev_adjacent_dev_set(struct net_device *upper_dev,
7797                                       struct net_device *lower_dev,
7798                                       bool val)
7799 {
7800         struct netdev_adjacent *adj;
7801
7802         adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower);
7803         if (adj)
7804                 adj->ignore = val;
7805
7806         adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper);
7807         if (adj)
7808                 adj->ignore = val;
7809 }
7810
7811 static void netdev_adjacent_dev_disable(struct net_device *upper_dev,
7812                                         struct net_device *lower_dev)
7813 {
7814         __netdev_adjacent_dev_set(upper_dev, lower_dev, true);
7815 }
7816
7817 static void netdev_adjacent_dev_enable(struct net_device *upper_dev,
7818                                        struct net_device *lower_dev)
7819 {
7820         __netdev_adjacent_dev_set(upper_dev, lower_dev, false);
7821 }
7822
7823 int netdev_adjacent_change_prepare(struct net_device *old_dev,
7824                                    struct net_device *new_dev,
7825                                    struct net_device *dev,
7826                                    struct netlink_ext_ack *extack)
7827 {
7828         struct netdev_nested_priv priv = {
7829                 .flags = 0,
7830                 .data = NULL,
7831         };
7832         int err;
7833
7834         if (!new_dev)
7835                 return 0;
7836
7837         if (old_dev && new_dev != old_dev)
7838                 netdev_adjacent_dev_disable(dev, old_dev);
7839         err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv,
7840                                       extack);
7841         if (err) {
7842                 if (old_dev && new_dev != old_dev)
7843                         netdev_adjacent_dev_enable(dev, old_dev);
7844                 return err;
7845         }
7846
7847         return 0;
7848 }
7849 EXPORT_SYMBOL(netdev_adjacent_change_prepare);
7850
7851 void netdev_adjacent_change_commit(struct net_device *old_dev,
7852                                    struct net_device *new_dev,
7853                                    struct net_device *dev)
7854 {
7855         struct netdev_nested_priv priv = {
7856                 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
7857                 .data = NULL,
7858         };
7859
7860         if (!new_dev || !old_dev)
7861                 return;
7862
7863         if (new_dev == old_dev)
7864                 return;
7865
7866         netdev_adjacent_dev_enable(dev, old_dev);
7867         __netdev_upper_dev_unlink(old_dev, dev, &priv);
7868 }
7869 EXPORT_SYMBOL(netdev_adjacent_change_commit);
7870
7871 void netdev_adjacent_change_abort(struct net_device *old_dev,
7872                                   struct net_device *new_dev,
7873                                   struct net_device *dev)
7874 {
7875         struct netdev_nested_priv priv = {
7876                 .flags = 0,
7877                 .data = NULL,
7878         };
7879
7880         if (!new_dev)
7881                 return;
7882
7883         if (old_dev && new_dev != old_dev)
7884                 netdev_adjacent_dev_enable(dev, old_dev);
7885
7886         __netdev_upper_dev_unlink(new_dev, dev, &priv);
7887 }
7888 EXPORT_SYMBOL(netdev_adjacent_change_abort);
7889
7890 /**
7891  * netdev_bonding_info_change - Dispatch event about slave change
7892  * @dev: device
7893  * @bonding_info: info to dispatch
7894  *
7895  * Send NETDEV_BONDING_INFO to netdev notifiers with info.
7896  * The caller must hold the RTNL lock.
7897  */
7898 void netdev_bonding_info_change(struct net_device *dev,
7899                                 struct netdev_bonding_info *bonding_info)
7900 {
7901         struct netdev_notifier_bonding_info info = {
7902                 .info.dev = dev,
7903         };
7904
7905         memcpy(&info.bonding_info, bonding_info,
7906                sizeof(struct netdev_bonding_info));
7907         call_netdevice_notifiers_info(NETDEV_BONDING_INFO,
7908                                       &info.info);
7909 }
7910 EXPORT_SYMBOL(netdev_bonding_info_change);
7911
7912 static int netdev_offload_xstats_enable_l3(struct net_device *dev,
7913                                            struct netlink_ext_ack *extack)
7914 {
7915         struct netdev_notifier_offload_xstats_info info = {
7916                 .info.dev = dev,
7917                 .info.extack = extack,
7918                 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3,
7919         };
7920         int err;
7921         int rc;
7922
7923         dev->offload_xstats_l3 = kzalloc(sizeof(*dev->offload_xstats_l3),
7924                                          GFP_KERNEL);
7925         if (!dev->offload_xstats_l3)
7926                 return -ENOMEM;
7927
7928         rc = call_netdevice_notifiers_info_robust(NETDEV_OFFLOAD_XSTATS_ENABLE,
7929                                                   NETDEV_OFFLOAD_XSTATS_DISABLE,
7930                                                   &info.info);
7931         err = notifier_to_errno(rc);
7932         if (err)
7933                 goto free_stats;
7934
7935         return 0;
7936
7937 free_stats:
7938         kfree(dev->offload_xstats_l3);
7939         dev->offload_xstats_l3 = NULL;
7940         return err;
7941 }
7942
7943 int netdev_offload_xstats_enable(struct net_device *dev,
7944                                  enum netdev_offload_xstats_type type,
7945                                  struct netlink_ext_ack *extack)
7946 {
7947         ASSERT_RTNL();
7948
7949         if (netdev_offload_xstats_enabled(dev, type))
7950                 return -EALREADY;
7951
7952         switch (type) {
7953         case NETDEV_OFFLOAD_XSTATS_TYPE_L3:
7954                 return netdev_offload_xstats_enable_l3(dev, extack);
7955         }
7956
7957         WARN_ON(1);
7958         return -EINVAL;
7959 }
7960 EXPORT_SYMBOL(netdev_offload_xstats_enable);
7961
7962 static void netdev_offload_xstats_disable_l3(struct net_device *dev)
7963 {
7964         struct netdev_notifier_offload_xstats_info info = {
7965                 .info.dev = dev,
7966                 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3,
7967         };
7968
7969         call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_DISABLE,
7970                                       &info.info);
7971         kfree(dev->offload_xstats_l3);
7972         dev->offload_xstats_l3 = NULL;
7973 }
7974
7975 int netdev_offload_xstats_disable(struct net_device *dev,
7976                                   enum netdev_offload_xstats_type type)
7977 {
7978         ASSERT_RTNL();
7979
7980         if (!netdev_offload_xstats_enabled(dev, type))
7981                 return -EALREADY;
7982
7983         switch (type) {
7984         case NETDEV_OFFLOAD_XSTATS_TYPE_L3:
7985                 netdev_offload_xstats_disable_l3(dev);
7986                 return 0;
7987         }
7988
7989         WARN_ON(1);
7990         return -EINVAL;
7991 }
7992 EXPORT_SYMBOL(netdev_offload_xstats_disable);
7993
7994 static void netdev_offload_xstats_disable_all(struct net_device *dev)
7995 {
7996         netdev_offload_xstats_disable(dev, NETDEV_OFFLOAD_XSTATS_TYPE_L3);
7997 }
7998
7999 static struct rtnl_hw_stats64 *
8000 netdev_offload_xstats_get_ptr(const struct net_device *dev,
8001                               enum netdev_offload_xstats_type type)
8002 {
8003         switch (type) {
8004         case NETDEV_OFFLOAD_XSTATS_TYPE_L3:
8005                 return dev->offload_xstats_l3;
8006         }
8007
8008         WARN_ON(1);
8009         return NULL;
8010 }
8011
8012 bool netdev_offload_xstats_enabled(const struct net_device *dev,
8013                                    enum netdev_offload_xstats_type type)
8014 {
8015         ASSERT_RTNL();
8016
8017         return netdev_offload_xstats_get_ptr(dev, type);
8018 }
8019 EXPORT_SYMBOL(netdev_offload_xstats_enabled);
8020
8021 struct netdev_notifier_offload_xstats_ru {
8022         bool used;
8023 };
8024
8025 struct netdev_notifier_offload_xstats_rd {
8026         struct rtnl_hw_stats64 stats;
8027         bool used;
8028 };
8029
8030 static void netdev_hw_stats64_add(struct rtnl_hw_stats64 *dest,
8031                                   const struct rtnl_hw_stats64 *src)
8032 {
8033         dest->rx_packets          += src->rx_packets;
8034         dest->tx_packets          += src->tx_packets;
8035         dest->rx_bytes            += src->rx_bytes;
8036         dest->tx_bytes            += src->tx_bytes;
8037         dest->rx_errors           += src->rx_errors;
8038         dest->tx_errors           += src->tx_errors;
8039         dest->rx_dropped          += src->rx_dropped;
8040         dest->tx_dropped          += src->tx_dropped;
8041         dest->multicast           += src->multicast;
8042 }
8043
8044 static int netdev_offload_xstats_get_used(struct net_device *dev,
8045                                           enum netdev_offload_xstats_type type,
8046                                           bool *p_used,
8047                                           struct netlink_ext_ack *extack)
8048 {
8049         struct netdev_notifier_offload_xstats_ru report_used = {};
8050         struct netdev_notifier_offload_xstats_info info = {
8051                 .info.dev = dev,
8052                 .info.extack = extack,
8053                 .type = type,
8054                 .report_used = &report_used,
8055         };
8056         int rc;
8057
8058         WARN_ON(!netdev_offload_xstats_enabled(dev, type));
8059         rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_USED,
8060                                            &info.info);
8061         *p_used = report_used.used;
8062         return notifier_to_errno(rc);
8063 }
8064
8065 static int netdev_offload_xstats_get_stats(struct net_device *dev,
8066                                            enum netdev_offload_xstats_type type,
8067                                            struct rtnl_hw_stats64 *p_stats,
8068                                            bool *p_used,
8069                                            struct netlink_ext_ack *extack)
8070 {
8071         struct netdev_notifier_offload_xstats_rd report_delta = {};
8072         struct netdev_notifier_offload_xstats_info info = {
8073                 .info.dev = dev,
8074                 .info.extack = extack,
8075                 .type = type,
8076                 .report_delta = &report_delta,
8077         };
8078         struct rtnl_hw_stats64 *stats;
8079         int rc;
8080
8081         stats = netdev_offload_xstats_get_ptr(dev, type);
8082         if (WARN_ON(!stats))
8083                 return -EINVAL;
8084
8085         rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_DELTA,
8086                                            &info.info);
8087
8088         /* Cache whatever we got, even if there was an error, otherwise the
8089          * successful stats retrievals would get lost.
8090          */
8091         netdev_hw_stats64_add(stats, &report_delta.stats);
8092
8093         if (p_stats)
8094                 *p_stats = *stats;
8095         *p_used = report_delta.used;
8096
8097         return notifier_to_errno(rc);
8098 }
8099
8100 int netdev_offload_xstats_get(struct net_device *dev,
8101                               enum netdev_offload_xstats_type type,
8102                               struct rtnl_hw_stats64 *p_stats, bool *p_used,
8103                               struct netlink_ext_ack *extack)
8104 {
8105         ASSERT_RTNL();
8106
8107         if (p_stats)
8108                 return netdev_offload_xstats_get_stats(dev, type, p_stats,
8109                                                        p_used, extack);
8110         else
8111                 return netdev_offload_xstats_get_used(dev, type, p_used,
8112                                                       extack);
8113 }
8114 EXPORT_SYMBOL(netdev_offload_xstats_get);
8115
8116 void
8117 netdev_offload_xstats_report_delta(struct netdev_notifier_offload_xstats_rd *report_delta,
8118                                    const struct rtnl_hw_stats64 *stats)
8119 {
8120         report_delta->used = true;
8121         netdev_hw_stats64_add(&report_delta->stats, stats);
8122 }
8123 EXPORT_SYMBOL(netdev_offload_xstats_report_delta);
8124
8125 void
8126 netdev_offload_xstats_report_used(struct netdev_notifier_offload_xstats_ru *report_used)
8127 {
8128         report_used->used = true;
8129 }
8130 EXPORT_SYMBOL(netdev_offload_xstats_report_used);
8131
8132 void netdev_offload_xstats_push_delta(struct net_device *dev,
8133                                       enum netdev_offload_xstats_type type,
8134                                       const struct rtnl_hw_stats64 *p_stats)
8135 {
8136         struct rtnl_hw_stats64 *stats;
8137
8138         ASSERT_RTNL();
8139
8140         stats = netdev_offload_xstats_get_ptr(dev, type);
8141         if (WARN_ON(!stats))
8142                 return;
8143
8144         netdev_hw_stats64_add(stats, p_stats);
8145 }
8146 EXPORT_SYMBOL(netdev_offload_xstats_push_delta);
8147
8148 /**
8149  * netdev_get_xmit_slave - Get the xmit slave of master device
8150  * @dev: device
8151  * @skb: The packet
8152  * @all_slaves: assume all the slaves are active
8153  *
8154  * The reference counters are not incremented so the caller must be
8155  * careful with locks. The caller must hold RCU lock.
8156  * %NULL is returned if no slave is found.
8157  */
8158
8159 struct net_device *netdev_get_xmit_slave(struct net_device *dev,
8160                                          struct sk_buff *skb,
8161                                          bool all_slaves)
8162 {
8163         const struct net_device_ops *ops = dev->netdev_ops;
8164
8165         if (!ops->ndo_get_xmit_slave)
8166                 return NULL;
8167         return ops->ndo_get_xmit_slave(dev, skb, all_slaves);
8168 }
8169 EXPORT_SYMBOL(netdev_get_xmit_slave);
8170
8171 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev,
8172                                                   struct sock *sk)
8173 {
8174         const struct net_device_ops *ops = dev->netdev_ops;
8175
8176         if (!ops->ndo_sk_get_lower_dev)
8177                 return NULL;
8178         return ops->ndo_sk_get_lower_dev(dev, sk);
8179 }
8180
8181 /**
8182  * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket
8183  * @dev: device
8184  * @sk: the socket
8185  *
8186  * %NULL is returned if no lower device is found.
8187  */
8188
8189 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev,
8190                                             struct sock *sk)
8191 {
8192         struct net_device *lower;
8193
8194         lower = netdev_sk_get_lower_dev(dev, sk);
8195         while (lower) {
8196                 dev = lower;
8197                 lower = netdev_sk_get_lower_dev(dev, sk);
8198         }
8199
8200         return dev;
8201 }
8202 EXPORT_SYMBOL(netdev_sk_get_lowest_dev);
8203
8204 static void netdev_adjacent_add_links(struct net_device *dev)
8205 {
8206         struct netdev_adjacent *iter;
8207
8208         struct net *net = dev_net(dev);
8209
8210         list_for_each_entry(iter, &dev->adj_list.upper, list) {
8211                 if (!net_eq(net, dev_net(iter->dev)))
8212                         continue;
8213                 netdev_adjacent_sysfs_add(iter->dev, dev,
8214                                           &iter->dev->adj_list.lower);
8215                 netdev_adjacent_sysfs_add(dev, iter->dev,
8216                                           &dev->adj_list.upper);
8217         }
8218
8219         list_for_each_entry(iter, &dev->adj_list.lower, list) {
8220                 if (!net_eq(net, dev_net(iter->dev)))
8221                         continue;
8222                 netdev_adjacent_sysfs_add(iter->dev, dev,
8223                                           &iter->dev->adj_list.upper);
8224                 netdev_adjacent_sysfs_add(dev, iter->dev,
8225                                           &dev->adj_list.lower);
8226         }
8227 }
8228
8229 static void netdev_adjacent_del_links(struct net_device *dev)
8230 {
8231         struct netdev_adjacent *iter;
8232
8233         struct net *net = dev_net(dev);
8234
8235         list_for_each_entry(iter, &dev->adj_list.upper, list) {
8236                 if (!net_eq(net, dev_net(iter->dev)))
8237                         continue;
8238                 netdev_adjacent_sysfs_del(iter->dev, dev->name,
8239                                           &iter->dev->adj_list.lower);
8240                 netdev_adjacent_sysfs_del(dev, iter->dev->name,
8241                                           &dev->adj_list.upper);
8242         }
8243
8244         list_for_each_entry(iter, &dev->adj_list.lower, list) {
8245                 if (!net_eq(net, dev_net(iter->dev)))
8246                         continue;
8247                 netdev_adjacent_sysfs_del(iter->dev, dev->name,
8248                                           &iter->dev->adj_list.upper);
8249                 netdev_adjacent_sysfs_del(dev, iter->dev->name,
8250                                           &dev->adj_list.lower);
8251         }
8252 }
8253
8254 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
8255 {
8256         struct netdev_adjacent *iter;
8257
8258         struct net *net = dev_net(dev);
8259
8260         list_for_each_entry(iter, &dev->adj_list.upper, list) {
8261                 if (!net_eq(net, dev_net(iter->dev)))
8262                         continue;
8263                 netdev_adjacent_sysfs_del(iter->dev, oldname,
8264                                           &iter->dev->adj_list.lower);
8265                 netdev_adjacent_sysfs_add(iter->dev, dev,
8266                                           &iter->dev->adj_list.lower);
8267         }
8268
8269         list_for_each_entry(iter, &dev->adj_list.lower, list) {
8270                 if (!net_eq(net, dev_net(iter->dev)))
8271                         continue;
8272                 netdev_adjacent_sysfs_del(iter->dev, oldname,
8273                                           &iter->dev->adj_list.upper);
8274                 netdev_adjacent_sysfs_add(iter->dev, dev,
8275                                           &iter->dev->adj_list.upper);
8276         }
8277 }
8278
8279 void *netdev_lower_dev_get_private(struct net_device *dev,
8280                                    struct net_device *lower_dev)
8281 {
8282         struct netdev_adjacent *lower;
8283
8284         if (!lower_dev)
8285                 return NULL;
8286         lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
8287         if (!lower)
8288                 return NULL;
8289
8290         return lower->private;
8291 }
8292 EXPORT_SYMBOL(netdev_lower_dev_get_private);
8293
8294
8295 /**
8296  * netdev_lower_state_changed - Dispatch event about lower device state change
8297  * @lower_dev: device
8298  * @lower_state_info: state to dispatch
8299  *
8300  * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
8301  * The caller must hold the RTNL lock.
8302  */
8303 void netdev_lower_state_changed(struct net_device *lower_dev,
8304                                 void *lower_state_info)
8305 {
8306         struct netdev_notifier_changelowerstate_info changelowerstate_info = {
8307                 .info.dev = lower_dev,
8308         };
8309
8310         ASSERT_RTNL();
8311         changelowerstate_info.lower_state_info = lower_state_info;
8312         call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE,
8313                                       &changelowerstate_info.info);
8314 }
8315 EXPORT_SYMBOL(netdev_lower_state_changed);
8316
8317 static void dev_change_rx_flags(struct net_device *dev, int flags)
8318 {
8319         const struct net_device_ops *ops = dev->netdev_ops;
8320
8321         if (ops->ndo_change_rx_flags)
8322                 ops->ndo_change_rx_flags(dev, flags);
8323 }
8324
8325 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
8326 {
8327         unsigned int old_flags = dev->flags;
8328         kuid_t uid;
8329         kgid_t gid;
8330
8331         ASSERT_RTNL();
8332
8333         dev->flags |= IFF_PROMISC;
8334         dev->promiscuity += inc;
8335         if (dev->promiscuity == 0) {
8336                 /*
8337                  * Avoid overflow.
8338                  * If inc causes overflow, untouch promisc and return error.
8339                  */
8340                 if (inc < 0)
8341                         dev->flags &= ~IFF_PROMISC;
8342                 else {
8343                         dev->promiscuity -= inc;
8344                         netdev_warn(dev, "promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n");
8345                         return -EOVERFLOW;
8346                 }
8347         }
8348         if (dev->flags != old_flags) {
8349                 netdev_info(dev, "%s promiscuous mode\n",
8350                             dev->flags & IFF_PROMISC ? "entered" : "left");
8351                 if (audit_enabled) {
8352                         current_uid_gid(&uid, &gid);
8353                         audit_log(audit_context(), GFP_ATOMIC,
8354                                   AUDIT_ANOM_PROMISCUOUS,
8355                                   "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
8356                                   dev->name, (dev->flags & IFF_PROMISC),
8357                                   (old_flags & IFF_PROMISC),
8358                                   from_kuid(&init_user_ns, audit_get_loginuid(current)),
8359                                   from_kuid(&init_user_ns, uid),
8360                                   from_kgid(&init_user_ns, gid),
8361                                   audit_get_sessionid(current));
8362                 }
8363
8364                 dev_change_rx_flags(dev, IFF_PROMISC);
8365         }
8366         if (notify)
8367                 __dev_notify_flags(dev, old_flags, IFF_PROMISC, 0, NULL);
8368         return 0;
8369 }
8370
8371 /**
8372  *      dev_set_promiscuity     - update promiscuity count on a device
8373  *      @dev: device
8374  *      @inc: modifier
8375  *
8376  *      Add or remove promiscuity from a device. While the count in the device
8377  *      remains above zero the interface remains promiscuous. Once it hits zero
8378  *      the device reverts back to normal filtering operation. A negative inc
8379  *      value is used to drop promiscuity on the device.
8380  *      Return 0 if successful or a negative errno code on error.
8381  */
8382 int dev_set_promiscuity(struct net_device *dev, int inc)
8383 {
8384         unsigned int old_flags = dev->flags;
8385         int err;
8386
8387         err = __dev_set_promiscuity(dev, inc, true);
8388         if (err < 0)
8389                 return err;
8390         if (dev->flags != old_flags)
8391                 dev_set_rx_mode(dev);
8392         return err;
8393 }
8394 EXPORT_SYMBOL(dev_set_promiscuity);
8395
8396 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
8397 {
8398         unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
8399
8400         ASSERT_RTNL();
8401
8402         dev->flags |= IFF_ALLMULTI;
8403         dev->allmulti += inc;
8404         if (dev->allmulti == 0) {
8405                 /*
8406                  * Avoid overflow.
8407                  * If inc causes overflow, untouch allmulti and return error.
8408                  */
8409                 if (inc < 0)
8410                         dev->flags &= ~IFF_ALLMULTI;
8411                 else {
8412                         dev->allmulti -= inc;
8413                         netdev_warn(dev, "allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n");
8414                         return -EOVERFLOW;
8415                 }
8416         }
8417         if (dev->flags ^ old_flags) {
8418                 netdev_info(dev, "%s allmulticast mode\n",
8419                             dev->flags & IFF_ALLMULTI ? "entered" : "left");
8420                 dev_change_rx_flags(dev, IFF_ALLMULTI);
8421                 dev_set_rx_mode(dev);
8422                 if (notify)
8423                         __dev_notify_flags(dev, old_flags,
8424                                            dev->gflags ^ old_gflags, 0, NULL);
8425         }
8426         return 0;
8427 }
8428
8429 /**
8430  *      dev_set_allmulti        - update allmulti count on a device
8431  *      @dev: device
8432  *      @inc: modifier
8433  *
8434  *      Add or remove reception of all multicast frames to a device. While the
8435  *      count in the device remains above zero the interface remains listening
8436  *      to all interfaces. Once it hits zero the device reverts back to normal
8437  *      filtering operation. A negative @inc value is used to drop the counter
8438  *      when releasing a resource needing all multicasts.
8439  *      Return 0 if successful or a negative errno code on error.
8440  */
8441
8442 int dev_set_allmulti(struct net_device *dev, int inc)
8443 {
8444         return __dev_set_allmulti(dev, inc, true);
8445 }
8446 EXPORT_SYMBOL(dev_set_allmulti);
8447
8448 /*
8449  *      Upload unicast and multicast address lists to device and
8450  *      configure RX filtering. When the device doesn't support unicast
8451  *      filtering it is put in promiscuous mode while unicast addresses
8452  *      are present.
8453  */
8454 void __dev_set_rx_mode(struct net_device *dev)
8455 {
8456         const struct net_device_ops *ops = dev->netdev_ops;
8457
8458         /* dev_open will call this function so the list will stay sane. */
8459         if (!(dev->flags&IFF_UP))
8460                 return;
8461
8462         if (!netif_device_present(dev))
8463                 return;
8464
8465         if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
8466                 /* Unicast addresses changes may only happen under the rtnl,
8467                  * therefore calling __dev_set_promiscuity here is safe.
8468                  */
8469                 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
8470                         __dev_set_promiscuity(dev, 1, false);
8471                         dev->uc_promisc = true;
8472                 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
8473                         __dev_set_promiscuity(dev, -1, false);
8474                         dev->uc_promisc = false;
8475                 }
8476         }
8477
8478         if (ops->ndo_set_rx_mode)
8479                 ops->ndo_set_rx_mode(dev);
8480 }
8481
8482 void dev_set_rx_mode(struct net_device *dev)
8483 {
8484         netif_addr_lock_bh(dev);
8485         __dev_set_rx_mode(dev);
8486         netif_addr_unlock_bh(dev);
8487 }
8488
8489 /**
8490  *      dev_get_flags - get flags reported to userspace
8491  *      @dev: device
8492  *
8493  *      Get the combination of flag bits exported through APIs to userspace.
8494  */
8495 unsigned int dev_get_flags(const struct net_device *dev)
8496 {
8497         unsigned int flags;
8498
8499         flags = (dev->flags & ~(IFF_PROMISC |
8500                                 IFF_ALLMULTI |
8501                                 IFF_RUNNING |
8502                                 IFF_LOWER_UP |
8503                                 IFF_DORMANT)) |
8504                 (dev->gflags & (IFF_PROMISC |
8505                                 IFF_ALLMULTI));
8506
8507         if (netif_running(dev)) {
8508                 if (netif_oper_up(dev))
8509                         flags |= IFF_RUNNING;
8510                 if (netif_carrier_ok(dev))
8511                         flags |= IFF_LOWER_UP;
8512                 if (netif_dormant(dev))
8513                         flags |= IFF_DORMANT;
8514         }
8515
8516         return flags;
8517 }
8518 EXPORT_SYMBOL(dev_get_flags);
8519
8520 int __dev_change_flags(struct net_device *dev, unsigned int flags,
8521                        struct netlink_ext_ack *extack)
8522 {
8523         unsigned int old_flags = dev->flags;
8524         int ret;
8525
8526         ASSERT_RTNL();
8527
8528         /*
8529          *      Set the flags on our device.
8530          */
8531
8532         dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
8533                                IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
8534                                IFF_AUTOMEDIA)) |
8535                      (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
8536                                     IFF_ALLMULTI));
8537
8538         /*
8539          *      Load in the correct multicast list now the flags have changed.
8540          */
8541
8542         if ((old_flags ^ flags) & IFF_MULTICAST)
8543                 dev_change_rx_flags(dev, IFF_MULTICAST);
8544
8545         dev_set_rx_mode(dev);
8546
8547         /*
8548          *      Have we downed the interface. We handle IFF_UP ourselves
8549          *      according to user attempts to set it, rather than blindly
8550          *      setting it.
8551          */
8552
8553         ret = 0;
8554         if ((old_flags ^ flags) & IFF_UP) {
8555                 if (old_flags & IFF_UP)
8556                         __dev_close(dev);
8557                 else
8558                         ret = __dev_open(dev, extack);
8559         }
8560
8561         if ((flags ^ dev->gflags) & IFF_PROMISC) {
8562                 int inc = (flags & IFF_PROMISC) ? 1 : -1;
8563                 unsigned int old_flags = dev->flags;
8564
8565                 dev->gflags ^= IFF_PROMISC;
8566
8567                 if (__dev_set_promiscuity(dev, inc, false) >= 0)
8568                         if (dev->flags != old_flags)
8569                                 dev_set_rx_mode(dev);
8570         }
8571
8572         /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
8573          * is important. Some (broken) drivers set IFF_PROMISC, when
8574          * IFF_ALLMULTI is requested not asking us and not reporting.
8575          */
8576         if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
8577                 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
8578
8579                 dev->gflags ^= IFF_ALLMULTI;
8580                 __dev_set_allmulti(dev, inc, false);
8581         }
8582
8583         return ret;
8584 }
8585
8586 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
8587                         unsigned int gchanges, u32 portid,
8588                         const struct nlmsghdr *nlh)
8589 {
8590         unsigned int changes = dev->flags ^ old_flags;
8591
8592         if (gchanges)
8593                 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC, portid, nlh);
8594
8595         if (changes & IFF_UP) {
8596                 if (dev->flags & IFF_UP)
8597                         call_netdevice_notifiers(NETDEV_UP, dev);
8598                 else
8599                         call_netdevice_notifiers(NETDEV_DOWN, dev);
8600         }
8601
8602         if (dev->flags & IFF_UP &&
8603             (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
8604                 struct netdev_notifier_change_info change_info = {
8605                         .info = {
8606                                 .dev = dev,
8607                         },
8608                         .flags_changed = changes,
8609                 };
8610
8611                 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info);
8612         }
8613 }
8614
8615 /**
8616  *      dev_change_flags - change device settings
8617  *      @dev: device
8618  *      @flags: device state flags
8619  *      @extack: netlink extended ack
8620  *
8621  *      Change settings on device based state flags. The flags are
8622  *      in the userspace exported format.
8623  */
8624 int dev_change_flags(struct net_device *dev, unsigned int flags,
8625                      struct netlink_ext_ack *extack)
8626 {
8627         int ret;
8628         unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
8629
8630         ret = __dev_change_flags(dev, flags, extack);
8631         if (ret < 0)
8632                 return ret;
8633
8634         changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
8635         __dev_notify_flags(dev, old_flags, changes, 0, NULL);
8636         return ret;
8637 }
8638 EXPORT_SYMBOL(dev_change_flags);
8639
8640 int __dev_set_mtu(struct net_device *dev, int new_mtu)
8641 {
8642         const struct net_device_ops *ops = dev->netdev_ops;
8643
8644         if (ops->ndo_change_mtu)
8645                 return ops->ndo_change_mtu(dev, new_mtu);
8646
8647         /* Pairs with all the lockless reads of dev->mtu in the stack */
8648         WRITE_ONCE(dev->mtu, new_mtu);
8649         return 0;
8650 }
8651 EXPORT_SYMBOL(__dev_set_mtu);
8652
8653 int dev_validate_mtu(struct net_device *dev, int new_mtu,
8654                      struct netlink_ext_ack *extack)
8655 {
8656         /* MTU must be positive, and in range */
8657         if (new_mtu < 0 || new_mtu < dev->min_mtu) {
8658                 NL_SET_ERR_MSG(extack, "mtu less than device minimum");
8659                 return -EINVAL;
8660         }
8661
8662         if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) {
8663                 NL_SET_ERR_MSG(extack, "mtu greater than device maximum");
8664                 return -EINVAL;
8665         }
8666         return 0;
8667 }
8668
8669 /**
8670  *      dev_set_mtu_ext - Change maximum transfer unit
8671  *      @dev: device
8672  *      @new_mtu: new transfer unit
8673  *      @extack: netlink extended ack
8674  *
8675  *      Change the maximum transfer size of the network device.
8676  */
8677 int dev_set_mtu_ext(struct net_device *dev, int new_mtu,
8678                     struct netlink_ext_ack *extack)
8679 {
8680         int err, orig_mtu;
8681
8682         if (new_mtu == dev->mtu)
8683                 return 0;
8684
8685         err = dev_validate_mtu(dev, new_mtu, extack);
8686         if (err)
8687                 return err;
8688
8689         if (!netif_device_present(dev))
8690                 return -ENODEV;
8691
8692         err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
8693         err = notifier_to_errno(err);
8694         if (err)
8695                 return err;
8696
8697         orig_mtu = dev->mtu;
8698         err = __dev_set_mtu(dev, new_mtu);
8699
8700         if (!err) {
8701                 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8702                                                    orig_mtu);
8703                 err = notifier_to_errno(err);
8704                 if (err) {
8705                         /* setting mtu back and notifying everyone again,
8706                          * so that they have a chance to revert changes.
8707                          */
8708                         __dev_set_mtu(dev, orig_mtu);
8709                         call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8710                                                      new_mtu);
8711                 }
8712         }
8713         return err;
8714 }
8715
8716 int dev_set_mtu(struct net_device *dev, int new_mtu)
8717 {
8718         struct netlink_ext_ack extack;
8719         int err;
8720
8721         memset(&extack, 0, sizeof(extack));
8722         err = dev_set_mtu_ext(dev, new_mtu, &extack);
8723         if (err && extack._msg)
8724                 net_err_ratelimited("%s: %s\n", dev->name, extack._msg);
8725         return err;
8726 }
8727 EXPORT_SYMBOL(dev_set_mtu);
8728
8729 /**
8730  *      dev_change_tx_queue_len - Change TX queue length of a netdevice
8731  *      @dev: device
8732  *      @new_len: new tx queue length
8733  */
8734 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len)
8735 {
8736         unsigned int orig_len = dev->tx_queue_len;
8737         int res;
8738
8739         if (new_len != (unsigned int)new_len)
8740                 return -ERANGE;
8741
8742         if (new_len != orig_len) {
8743                 dev->tx_queue_len = new_len;
8744                 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev);
8745                 res = notifier_to_errno(res);
8746                 if (res)
8747                         goto err_rollback;
8748                 res = dev_qdisc_change_tx_queue_len(dev);
8749                 if (res)
8750                         goto err_rollback;
8751         }
8752
8753         return 0;
8754
8755 err_rollback:
8756         netdev_err(dev, "refused to change device tx_queue_len\n");
8757         dev->tx_queue_len = orig_len;
8758         return res;
8759 }
8760
8761 /**
8762  *      dev_set_group - Change group this device belongs to
8763  *      @dev: device
8764  *      @new_group: group this device should belong to
8765  */
8766 void dev_set_group(struct net_device *dev, int new_group)
8767 {
8768         dev->group = new_group;
8769 }
8770
8771 /**
8772  *      dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR.
8773  *      @dev: device
8774  *      @addr: new address
8775  *      @extack: netlink extended ack
8776  */
8777 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr,
8778                               struct netlink_ext_ack *extack)
8779 {
8780         struct netdev_notifier_pre_changeaddr_info info = {
8781                 .info.dev = dev,
8782                 .info.extack = extack,
8783                 .dev_addr = addr,
8784         };
8785         int rc;
8786
8787         rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info);
8788         return notifier_to_errno(rc);
8789 }
8790 EXPORT_SYMBOL(dev_pre_changeaddr_notify);
8791
8792 /**
8793  *      dev_set_mac_address - Change Media Access Control Address
8794  *      @dev: device
8795  *      @sa: new address
8796  *      @extack: netlink extended ack
8797  *
8798  *      Change the hardware (MAC) address of the device
8799  */
8800 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa,
8801                         struct netlink_ext_ack *extack)
8802 {
8803         const struct net_device_ops *ops = dev->netdev_ops;
8804         int err;
8805
8806         if (!ops->ndo_set_mac_address)
8807                 return -EOPNOTSUPP;
8808         if (sa->sa_family != dev->type)
8809                 return -EINVAL;
8810         if (!netif_device_present(dev))
8811                 return -ENODEV;
8812         err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack);
8813         if (err)
8814                 return err;
8815         err = ops->ndo_set_mac_address(dev, sa);
8816         if (err)
8817                 return err;
8818         dev->addr_assign_type = NET_ADDR_SET;
8819         call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
8820         add_device_randomness(dev->dev_addr, dev->addr_len);
8821         return 0;
8822 }
8823 EXPORT_SYMBOL(dev_set_mac_address);
8824
8825 static DECLARE_RWSEM(dev_addr_sem);
8826
8827 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa,
8828                              struct netlink_ext_ack *extack)
8829 {
8830         int ret;
8831
8832         down_write(&dev_addr_sem);
8833         ret = dev_set_mac_address(dev, sa, extack);
8834         up_write(&dev_addr_sem);
8835         return ret;
8836 }
8837 EXPORT_SYMBOL(dev_set_mac_address_user);
8838
8839 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name)
8840 {
8841         size_t size = sizeof(sa->sa_data_min);
8842         struct net_device *dev;
8843         int ret = 0;
8844
8845         down_read(&dev_addr_sem);
8846         rcu_read_lock();
8847
8848         dev = dev_get_by_name_rcu(net, dev_name);
8849         if (!dev) {
8850                 ret = -ENODEV;
8851                 goto unlock;
8852         }
8853         if (!dev->addr_len)
8854                 memset(sa->sa_data, 0, size);
8855         else
8856                 memcpy(sa->sa_data, dev->dev_addr,
8857                        min_t(size_t, size, dev->addr_len));
8858         sa->sa_family = dev->type;
8859
8860 unlock:
8861         rcu_read_unlock();
8862         up_read(&dev_addr_sem);
8863         return ret;
8864 }
8865 EXPORT_SYMBOL(dev_get_mac_address);
8866
8867 /**
8868  *      dev_change_carrier - Change device carrier
8869  *      @dev: device
8870  *      @new_carrier: new value
8871  *
8872  *      Change device carrier
8873  */
8874 int dev_change_carrier(struct net_device *dev, bool new_carrier)
8875 {
8876         const struct net_device_ops *ops = dev->netdev_ops;
8877
8878         if (!ops->ndo_change_carrier)
8879                 return -EOPNOTSUPP;
8880         if (!netif_device_present(dev))
8881                 return -ENODEV;
8882         return ops->ndo_change_carrier(dev, new_carrier);
8883 }
8884
8885 /**
8886  *      dev_get_phys_port_id - Get device physical port ID
8887  *      @dev: device
8888  *      @ppid: port ID
8889  *
8890  *      Get device physical port ID
8891  */
8892 int dev_get_phys_port_id(struct net_device *dev,
8893                          struct netdev_phys_item_id *ppid)
8894 {
8895         const struct net_device_ops *ops = dev->netdev_ops;
8896
8897         if (!ops->ndo_get_phys_port_id)
8898                 return -EOPNOTSUPP;
8899         return ops->ndo_get_phys_port_id(dev, ppid);
8900 }
8901
8902 /**
8903  *      dev_get_phys_port_name - Get device physical port name
8904  *      @dev: device
8905  *      @name: port name
8906  *      @len: limit of bytes to copy to name
8907  *
8908  *      Get device physical port name
8909  */
8910 int dev_get_phys_port_name(struct net_device *dev,
8911                            char *name, size_t len)
8912 {
8913         const struct net_device_ops *ops = dev->netdev_ops;
8914         int err;
8915
8916         if (ops->ndo_get_phys_port_name) {
8917                 err = ops->ndo_get_phys_port_name(dev, name, len);
8918                 if (err != -EOPNOTSUPP)
8919                         return err;
8920         }
8921         return devlink_compat_phys_port_name_get(dev, name, len);
8922 }
8923
8924 /**
8925  *      dev_get_port_parent_id - Get the device's port parent identifier
8926  *      @dev: network device
8927  *      @ppid: pointer to a storage for the port's parent identifier
8928  *      @recurse: allow/disallow recursion to lower devices
8929  *
8930  *      Get the devices's port parent identifier
8931  */
8932 int dev_get_port_parent_id(struct net_device *dev,
8933                            struct netdev_phys_item_id *ppid,
8934                            bool recurse)
8935 {
8936         const struct net_device_ops *ops = dev->netdev_ops;
8937         struct netdev_phys_item_id first = { };
8938         struct net_device *lower_dev;
8939         struct list_head *iter;
8940         int err;
8941
8942         if (ops->ndo_get_port_parent_id) {
8943                 err = ops->ndo_get_port_parent_id(dev, ppid);
8944                 if (err != -EOPNOTSUPP)
8945                         return err;
8946         }
8947
8948         err = devlink_compat_switch_id_get(dev, ppid);
8949         if (!recurse || err != -EOPNOTSUPP)
8950                 return err;
8951
8952         netdev_for_each_lower_dev(dev, lower_dev, iter) {
8953                 err = dev_get_port_parent_id(lower_dev, ppid, true);
8954                 if (err)
8955                         break;
8956                 if (!first.id_len)
8957                         first = *ppid;
8958                 else if (memcmp(&first, ppid, sizeof(*ppid)))
8959                         return -EOPNOTSUPP;
8960         }
8961
8962         return err;
8963 }
8964 EXPORT_SYMBOL(dev_get_port_parent_id);
8965
8966 /**
8967  *      netdev_port_same_parent_id - Indicate if two network devices have
8968  *      the same port parent identifier
8969  *      @a: first network device
8970  *      @b: second network device
8971  */
8972 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b)
8973 {
8974         struct netdev_phys_item_id a_id = { };
8975         struct netdev_phys_item_id b_id = { };
8976
8977         if (dev_get_port_parent_id(a, &a_id, true) ||
8978             dev_get_port_parent_id(b, &b_id, true))
8979                 return false;
8980
8981         return netdev_phys_item_id_same(&a_id, &b_id);
8982 }
8983 EXPORT_SYMBOL(netdev_port_same_parent_id);
8984
8985 /**
8986  *      dev_change_proto_down - set carrier according to proto_down.
8987  *
8988  *      @dev: device
8989  *      @proto_down: new value
8990  */
8991 int dev_change_proto_down(struct net_device *dev, bool proto_down)
8992 {
8993         if (!(dev->priv_flags & IFF_CHANGE_PROTO_DOWN))
8994                 return -EOPNOTSUPP;
8995         if (!netif_device_present(dev))
8996                 return -ENODEV;
8997         if (proto_down)
8998                 netif_carrier_off(dev);
8999         else
9000                 netif_carrier_on(dev);
9001         dev->proto_down = proto_down;
9002         return 0;
9003 }
9004
9005 /**
9006  *      dev_change_proto_down_reason - proto down reason
9007  *
9008  *      @dev: device
9009  *      @mask: proto down mask
9010  *      @value: proto down value
9011  */
9012 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask,
9013                                   u32 value)
9014 {
9015         int b;
9016
9017         if (!mask) {
9018                 dev->proto_down_reason = value;
9019         } else {
9020                 for_each_set_bit(b, &mask, 32) {
9021                         if (value & (1 << b))
9022                                 dev->proto_down_reason |= BIT(b);
9023                         else
9024                                 dev->proto_down_reason &= ~BIT(b);
9025                 }
9026         }
9027 }
9028
9029 struct bpf_xdp_link {
9030         struct bpf_link link;
9031         struct net_device *dev; /* protected by rtnl_lock, no refcnt held */
9032         int flags;
9033 };
9034
9035 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags)
9036 {
9037         if (flags & XDP_FLAGS_HW_MODE)
9038                 return XDP_MODE_HW;
9039         if (flags & XDP_FLAGS_DRV_MODE)
9040                 return XDP_MODE_DRV;
9041         if (flags & XDP_FLAGS_SKB_MODE)
9042                 return XDP_MODE_SKB;
9043         return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB;
9044 }
9045
9046 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode)
9047 {
9048         switch (mode) {
9049         case XDP_MODE_SKB:
9050                 return generic_xdp_install;
9051         case XDP_MODE_DRV:
9052         case XDP_MODE_HW:
9053                 return dev->netdev_ops->ndo_bpf;
9054         default:
9055                 return NULL;
9056         }
9057 }
9058
9059 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev,
9060                                          enum bpf_xdp_mode mode)
9061 {
9062         return dev->xdp_state[mode].link;
9063 }
9064
9065 static struct bpf_prog *dev_xdp_prog(struct net_device *dev,
9066                                      enum bpf_xdp_mode mode)
9067 {
9068         struct bpf_xdp_link *link = dev_xdp_link(dev, mode);
9069
9070         if (link)
9071                 return link->link.prog;
9072         return dev->xdp_state[mode].prog;
9073 }
9074
9075 u8 dev_xdp_prog_count(struct net_device *dev)
9076 {
9077         u8 count = 0;
9078         int i;
9079
9080         for (i = 0; i < __MAX_XDP_MODE; i++)
9081                 if (dev->xdp_state[i].prog || dev->xdp_state[i].link)
9082                         count++;
9083         return count;
9084 }
9085 EXPORT_SYMBOL_GPL(dev_xdp_prog_count);
9086
9087 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode)
9088 {
9089         struct bpf_prog *prog = dev_xdp_prog(dev, mode);
9090
9091         return prog ? prog->aux->id : 0;
9092 }
9093
9094 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode,
9095                              struct bpf_xdp_link *link)
9096 {
9097         dev->xdp_state[mode].link = link;
9098         dev->xdp_state[mode].prog = NULL;
9099 }
9100
9101 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode,
9102                              struct bpf_prog *prog)
9103 {
9104         dev->xdp_state[mode].link = NULL;
9105         dev->xdp_state[mode].prog = prog;
9106 }
9107
9108 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode,
9109                            bpf_op_t bpf_op, struct netlink_ext_ack *extack,
9110                            u32 flags, struct bpf_prog *prog)
9111 {
9112         struct netdev_bpf xdp;
9113         int err;
9114
9115         memset(&xdp, 0, sizeof(xdp));
9116         xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG;
9117         xdp.extack = extack;
9118         xdp.flags = flags;
9119         xdp.prog = prog;
9120
9121         /* Drivers assume refcnt is already incremented (i.e, prog pointer is
9122          * "moved" into driver), so they don't increment it on their own, but
9123          * they do decrement refcnt when program is detached or replaced.
9124          * Given net_device also owns link/prog, we need to bump refcnt here
9125          * to prevent drivers from underflowing it.
9126          */
9127         if (prog)
9128                 bpf_prog_inc(prog);
9129         err = bpf_op(dev, &xdp);
9130         if (err) {
9131                 if (prog)
9132                         bpf_prog_put(prog);
9133                 return err;
9134         }
9135
9136         if (mode != XDP_MODE_HW)
9137                 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog);
9138
9139         return 0;
9140 }
9141
9142 static void dev_xdp_uninstall(struct net_device *dev)
9143 {
9144         struct bpf_xdp_link *link;
9145         struct bpf_prog *prog;
9146         enum bpf_xdp_mode mode;
9147         bpf_op_t bpf_op;
9148
9149         ASSERT_RTNL();
9150
9151         for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) {
9152                 prog = dev_xdp_prog(dev, mode);
9153                 if (!prog)
9154                         continue;
9155
9156                 bpf_op = dev_xdp_bpf_op(dev, mode);
9157                 if (!bpf_op)
9158                         continue;
9159
9160                 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
9161
9162                 /* auto-detach link from net device */
9163                 link = dev_xdp_link(dev, mode);
9164                 if (link)
9165                         link->dev = NULL;
9166                 else
9167                         bpf_prog_put(prog);
9168
9169                 dev_xdp_set_link(dev, mode, NULL);
9170         }
9171 }
9172
9173 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack,
9174                           struct bpf_xdp_link *link, struct bpf_prog *new_prog,
9175                           struct bpf_prog *old_prog, u32 flags)
9176 {
9177         unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES);
9178         struct bpf_prog *cur_prog;
9179         struct net_device *upper;
9180         struct list_head *iter;
9181         enum bpf_xdp_mode mode;
9182         bpf_op_t bpf_op;
9183         int err;
9184
9185         ASSERT_RTNL();
9186
9187         /* either link or prog attachment, never both */
9188         if (link && (new_prog || old_prog))
9189                 return -EINVAL;
9190         /* link supports only XDP mode flags */
9191         if (link && (flags & ~XDP_FLAGS_MODES)) {
9192                 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment");
9193                 return -EINVAL;
9194         }
9195         /* just one XDP mode bit should be set, zero defaults to drv/skb mode */
9196         if (num_modes > 1) {
9197                 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set");
9198                 return -EINVAL;
9199         }
9200         /* avoid ambiguity if offload + drv/skb mode progs are both loaded */
9201         if (!num_modes && dev_xdp_prog_count(dev) > 1) {
9202                 NL_SET_ERR_MSG(extack,
9203                                "More than one program loaded, unset mode is ambiguous");
9204                 return -EINVAL;
9205         }
9206         /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */
9207         if (old_prog && !(flags & XDP_FLAGS_REPLACE)) {
9208                 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified");
9209                 return -EINVAL;
9210         }
9211
9212         mode = dev_xdp_mode(dev, flags);
9213         /* can't replace attached link */
9214         if (dev_xdp_link(dev, mode)) {
9215                 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link");
9216                 return -EBUSY;
9217         }
9218
9219         /* don't allow if an upper device already has a program */
9220         netdev_for_each_upper_dev_rcu(dev, upper, iter) {
9221                 if (dev_xdp_prog_count(upper) > 0) {
9222                         NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program");
9223                         return -EEXIST;
9224                 }
9225         }
9226
9227         cur_prog = dev_xdp_prog(dev, mode);
9228         /* can't replace attached prog with link */
9229         if (link && cur_prog) {
9230                 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link");
9231                 return -EBUSY;
9232         }
9233         if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) {
9234                 NL_SET_ERR_MSG(extack, "Active program does not match expected");
9235                 return -EEXIST;
9236         }
9237
9238         /* put effective new program into new_prog */
9239         if (link)
9240                 new_prog = link->link.prog;
9241
9242         if (new_prog) {
9243                 bool offload = mode == XDP_MODE_HW;
9244                 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB
9245                                                ? XDP_MODE_DRV : XDP_MODE_SKB;
9246
9247                 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) {
9248                         NL_SET_ERR_MSG(extack, "XDP program already attached");
9249                         return -EBUSY;
9250                 }
9251                 if (!offload && dev_xdp_prog(dev, other_mode)) {
9252                         NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time");
9253                         return -EEXIST;
9254                 }
9255                 if (!offload && bpf_prog_is_offloaded(new_prog->aux)) {
9256                         NL_SET_ERR_MSG(extack, "Using offloaded program without HW_MODE flag is not supported");
9257                         return -EINVAL;
9258                 }
9259                 if (bpf_prog_is_dev_bound(new_prog->aux) && !bpf_offload_dev_match(new_prog, dev)) {
9260                         NL_SET_ERR_MSG(extack, "Program bound to different device");
9261                         return -EINVAL;
9262                 }
9263                 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) {
9264                         NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device");
9265                         return -EINVAL;
9266                 }
9267                 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) {
9268                         NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device");
9269                         return -EINVAL;
9270                 }
9271         }
9272
9273         /* don't call drivers if the effective program didn't change */
9274         if (new_prog != cur_prog) {
9275                 bpf_op = dev_xdp_bpf_op(dev, mode);
9276                 if (!bpf_op) {
9277                         NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode");
9278                         return -EOPNOTSUPP;
9279                 }
9280
9281                 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog);
9282                 if (err)
9283                         return err;
9284         }
9285
9286         if (link)
9287                 dev_xdp_set_link(dev, mode, link);
9288         else
9289                 dev_xdp_set_prog(dev, mode, new_prog);
9290         if (cur_prog)
9291                 bpf_prog_put(cur_prog);
9292
9293         return 0;
9294 }
9295
9296 static int dev_xdp_attach_link(struct net_device *dev,
9297                                struct netlink_ext_ack *extack,
9298                                struct bpf_xdp_link *link)
9299 {
9300         return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags);
9301 }
9302
9303 static int dev_xdp_detach_link(struct net_device *dev,
9304                                struct netlink_ext_ack *extack,
9305                                struct bpf_xdp_link *link)
9306 {
9307         enum bpf_xdp_mode mode;
9308         bpf_op_t bpf_op;
9309
9310         ASSERT_RTNL();
9311
9312         mode = dev_xdp_mode(dev, link->flags);
9313         if (dev_xdp_link(dev, mode) != link)
9314                 return -EINVAL;
9315
9316         bpf_op = dev_xdp_bpf_op(dev, mode);
9317         WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
9318         dev_xdp_set_link(dev, mode, NULL);
9319         return 0;
9320 }
9321
9322 static void bpf_xdp_link_release(struct bpf_link *link)
9323 {
9324         struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9325
9326         rtnl_lock();
9327
9328         /* if racing with net_device's tear down, xdp_link->dev might be
9329          * already NULL, in which case link was already auto-detached
9330          */
9331         if (xdp_link->dev) {
9332                 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link));
9333                 xdp_link->dev = NULL;
9334         }
9335
9336         rtnl_unlock();
9337 }
9338
9339 static int bpf_xdp_link_detach(struct bpf_link *link)
9340 {
9341         bpf_xdp_link_release(link);
9342         return 0;
9343 }
9344
9345 static void bpf_xdp_link_dealloc(struct bpf_link *link)
9346 {
9347         struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9348
9349         kfree(xdp_link);
9350 }
9351
9352 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link,
9353                                      struct seq_file *seq)
9354 {
9355         struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9356         u32 ifindex = 0;
9357
9358         rtnl_lock();
9359         if (xdp_link->dev)
9360                 ifindex = xdp_link->dev->ifindex;
9361         rtnl_unlock();
9362
9363         seq_printf(seq, "ifindex:\t%u\n", ifindex);
9364 }
9365
9366 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link,
9367                                        struct bpf_link_info *info)
9368 {
9369         struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9370         u32 ifindex = 0;
9371
9372         rtnl_lock();
9373         if (xdp_link->dev)
9374                 ifindex = xdp_link->dev->ifindex;
9375         rtnl_unlock();
9376
9377         info->xdp.ifindex = ifindex;
9378         return 0;
9379 }
9380
9381 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog,
9382                                struct bpf_prog *old_prog)
9383 {
9384         struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9385         enum bpf_xdp_mode mode;
9386         bpf_op_t bpf_op;
9387         int err = 0;
9388
9389         rtnl_lock();
9390
9391         /* link might have been auto-released already, so fail */
9392         if (!xdp_link->dev) {
9393                 err = -ENOLINK;
9394                 goto out_unlock;
9395         }
9396
9397         if (old_prog && link->prog != old_prog) {
9398                 err = -EPERM;
9399                 goto out_unlock;
9400         }
9401         old_prog = link->prog;
9402         if (old_prog->type != new_prog->type ||
9403             old_prog->expected_attach_type != new_prog->expected_attach_type) {
9404                 err = -EINVAL;
9405                 goto out_unlock;
9406         }
9407
9408         if (old_prog == new_prog) {
9409                 /* no-op, don't disturb drivers */
9410                 bpf_prog_put(new_prog);
9411                 goto out_unlock;
9412         }
9413
9414         mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags);
9415         bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode);
9416         err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL,
9417                               xdp_link->flags, new_prog);
9418         if (err)
9419                 goto out_unlock;
9420
9421         old_prog = xchg(&link->prog, new_prog);
9422         bpf_prog_put(old_prog);
9423
9424 out_unlock:
9425         rtnl_unlock();
9426         return err;
9427 }
9428
9429 static const struct bpf_link_ops bpf_xdp_link_lops = {
9430         .release = bpf_xdp_link_release,
9431         .dealloc = bpf_xdp_link_dealloc,
9432         .detach = bpf_xdp_link_detach,
9433         .show_fdinfo = bpf_xdp_link_show_fdinfo,
9434         .fill_link_info = bpf_xdp_link_fill_link_info,
9435         .update_prog = bpf_xdp_link_update,
9436 };
9437
9438 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
9439 {
9440         struct net *net = current->nsproxy->net_ns;
9441         struct bpf_link_primer link_primer;
9442         struct bpf_xdp_link *link;
9443         struct net_device *dev;
9444         int err, fd;
9445
9446         rtnl_lock();
9447         dev = dev_get_by_index(net, attr->link_create.target_ifindex);
9448         if (!dev) {
9449                 rtnl_unlock();
9450                 return -EINVAL;
9451         }
9452
9453         link = kzalloc(sizeof(*link), GFP_USER);
9454         if (!link) {
9455                 err = -ENOMEM;
9456                 goto unlock;
9457         }
9458
9459         bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog);
9460         link->dev = dev;
9461         link->flags = attr->link_create.flags;
9462
9463         err = bpf_link_prime(&link->link, &link_primer);
9464         if (err) {
9465                 kfree(link);
9466                 goto unlock;
9467         }
9468
9469         err = dev_xdp_attach_link(dev, NULL, link);
9470         rtnl_unlock();
9471
9472         if (err) {
9473                 link->dev = NULL;
9474                 bpf_link_cleanup(&link_primer);
9475                 goto out_put_dev;
9476         }
9477
9478         fd = bpf_link_settle(&link_primer);
9479         /* link itself doesn't hold dev's refcnt to not complicate shutdown */
9480         dev_put(dev);
9481         return fd;
9482
9483 unlock:
9484         rtnl_unlock();
9485
9486 out_put_dev:
9487         dev_put(dev);
9488         return err;
9489 }
9490
9491 /**
9492  *      dev_change_xdp_fd - set or clear a bpf program for a device rx path
9493  *      @dev: device
9494  *      @extack: netlink extended ack
9495  *      @fd: new program fd or negative value to clear
9496  *      @expected_fd: old program fd that userspace expects to replace or clear
9497  *      @flags: xdp-related flags
9498  *
9499  *      Set or clear a bpf program for a device
9500  */
9501 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack,
9502                       int fd, int expected_fd, u32 flags)
9503 {
9504         enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags);
9505         struct bpf_prog *new_prog = NULL, *old_prog = NULL;
9506         int err;
9507
9508         ASSERT_RTNL();
9509
9510         if (fd >= 0) {
9511                 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP,
9512                                                  mode != XDP_MODE_SKB);
9513                 if (IS_ERR(new_prog))
9514                         return PTR_ERR(new_prog);
9515         }
9516
9517         if (expected_fd >= 0) {
9518                 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP,
9519                                                  mode != XDP_MODE_SKB);
9520                 if (IS_ERR(old_prog)) {
9521                         err = PTR_ERR(old_prog);
9522                         old_prog = NULL;
9523                         goto err_out;
9524                 }
9525         }
9526
9527         err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags);
9528
9529 err_out:
9530         if (err && new_prog)
9531                 bpf_prog_put(new_prog);
9532         if (old_prog)
9533                 bpf_prog_put(old_prog);
9534         return err;
9535 }
9536
9537 /**
9538  *      dev_new_index   -       allocate an ifindex
9539  *      @net: the applicable net namespace
9540  *
9541  *      Returns a suitable unique value for a new device interface
9542  *      number.  The caller must hold the rtnl semaphore or the
9543  *      dev_base_lock to be sure it remains unique.
9544  */
9545 static int dev_new_index(struct net *net)
9546 {
9547         int ifindex = net->ifindex;
9548
9549         for (;;) {
9550                 if (++ifindex <= 0)
9551                         ifindex = 1;
9552                 if (!__dev_get_by_index(net, ifindex))
9553                         return net->ifindex = ifindex;
9554         }
9555 }
9556
9557 /* Delayed registration/unregisteration */
9558 LIST_HEAD(net_todo_list);
9559 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
9560
9561 static void net_set_todo(struct net_device *dev)
9562 {
9563         list_add_tail(&dev->todo_list, &net_todo_list);
9564         atomic_inc(&dev_net(dev)->dev_unreg_count);
9565 }
9566
9567 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
9568         struct net_device *upper, netdev_features_t features)
9569 {
9570         netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9571         netdev_features_t feature;
9572         int feature_bit;
9573
9574         for_each_netdev_feature(upper_disables, feature_bit) {
9575                 feature = __NETIF_F_BIT(feature_bit);
9576                 if (!(upper->wanted_features & feature)
9577                     && (features & feature)) {
9578                         netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
9579                                    &feature, upper->name);
9580                         features &= ~feature;
9581                 }
9582         }
9583
9584         return features;
9585 }
9586
9587 static void netdev_sync_lower_features(struct net_device *upper,
9588         struct net_device *lower, netdev_features_t features)
9589 {
9590         netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9591         netdev_features_t feature;
9592         int feature_bit;
9593
9594         for_each_netdev_feature(upper_disables, feature_bit) {
9595                 feature = __NETIF_F_BIT(feature_bit);
9596                 if (!(features & feature) && (lower->features & feature)) {
9597                         netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
9598                                    &feature, lower->name);
9599                         lower->wanted_features &= ~feature;
9600                         __netdev_update_features(lower);
9601
9602                         if (unlikely(lower->features & feature))
9603                                 netdev_WARN(upper, "failed to disable %pNF on %s!\n",
9604                                             &feature, lower->name);
9605                         else
9606                                 netdev_features_change(lower);
9607                 }
9608         }
9609 }
9610
9611 static netdev_features_t netdev_fix_features(struct net_device *dev,
9612         netdev_features_t features)
9613 {
9614         /* Fix illegal checksum combinations */
9615         if ((features & NETIF_F_HW_CSUM) &&
9616             (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
9617                 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
9618                 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
9619         }
9620
9621         /* TSO requires that SG is present as well. */
9622         if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
9623                 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
9624                 features &= ~NETIF_F_ALL_TSO;
9625         }
9626
9627         if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
9628                                         !(features & NETIF_F_IP_CSUM)) {
9629                 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
9630                 features &= ~NETIF_F_TSO;
9631                 features &= ~NETIF_F_TSO_ECN;
9632         }
9633
9634         if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
9635                                          !(features & NETIF_F_IPV6_CSUM)) {
9636                 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
9637                 features &= ~NETIF_F_TSO6;
9638         }
9639
9640         /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */
9641         if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO))
9642                 features &= ~NETIF_F_TSO_MANGLEID;
9643
9644         /* TSO ECN requires that TSO is present as well. */
9645         if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
9646                 features &= ~NETIF_F_TSO_ECN;
9647
9648         /* Software GSO depends on SG. */
9649         if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
9650                 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
9651                 features &= ~NETIF_F_GSO;
9652         }
9653
9654         /* GSO partial features require GSO partial be set */
9655         if ((features & dev->gso_partial_features) &&
9656             !(features & NETIF_F_GSO_PARTIAL)) {
9657                 netdev_dbg(dev,
9658                            "Dropping partially supported GSO features since no GSO partial.\n");
9659                 features &= ~dev->gso_partial_features;
9660         }
9661
9662         if (!(features & NETIF_F_RXCSUM)) {
9663                 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet
9664                  * successfully merged by hardware must also have the
9665                  * checksum verified by hardware.  If the user does not
9666                  * want to enable RXCSUM, logically, we should disable GRO_HW.
9667                  */
9668                 if (features & NETIF_F_GRO_HW) {
9669                         netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n");
9670                         features &= ~NETIF_F_GRO_HW;
9671                 }
9672         }
9673
9674         /* LRO/HW-GRO features cannot be combined with RX-FCS */
9675         if (features & NETIF_F_RXFCS) {
9676                 if (features & NETIF_F_LRO) {
9677                         netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n");
9678                         features &= ~NETIF_F_LRO;
9679                 }
9680
9681                 if (features & NETIF_F_GRO_HW) {
9682                         netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n");
9683                         features &= ~NETIF_F_GRO_HW;
9684                 }
9685         }
9686
9687         if ((features & NETIF_F_GRO_HW) && (features & NETIF_F_LRO)) {
9688                 netdev_dbg(dev, "Dropping LRO feature since HW-GRO is requested.\n");
9689                 features &= ~NETIF_F_LRO;
9690         }
9691
9692         if (features & NETIF_F_HW_TLS_TX) {
9693                 bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) ==
9694                         (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
9695                 bool hw_csum = features & NETIF_F_HW_CSUM;
9696
9697                 if (!ip_csum && !hw_csum) {
9698                         netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n");
9699                         features &= ~NETIF_F_HW_TLS_TX;
9700                 }
9701         }
9702
9703         if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) {
9704                 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n");
9705                 features &= ~NETIF_F_HW_TLS_RX;
9706         }
9707
9708         return features;
9709 }
9710
9711 int __netdev_update_features(struct net_device *dev)
9712 {
9713         struct net_device *upper, *lower;
9714         netdev_features_t features;
9715         struct list_head *iter;
9716         int err = -1;
9717
9718         ASSERT_RTNL();
9719
9720         features = netdev_get_wanted_features(dev);
9721
9722         if (dev->netdev_ops->ndo_fix_features)
9723                 features = dev->netdev_ops->ndo_fix_features(dev, features);
9724
9725         /* driver might be less strict about feature dependencies */
9726         features = netdev_fix_features(dev, features);
9727
9728         /* some features can't be enabled if they're off on an upper device */
9729         netdev_for_each_upper_dev_rcu(dev, upper, iter)
9730                 features = netdev_sync_upper_features(dev, upper, features);
9731
9732         if (dev->features == features)
9733                 goto sync_lower;
9734
9735         netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
9736                 &dev->features, &features);
9737
9738         if (dev->netdev_ops->ndo_set_features)
9739                 err = dev->netdev_ops->ndo_set_features(dev, features);
9740         else
9741                 err = 0;
9742
9743         if (unlikely(err < 0)) {
9744                 netdev_err(dev,
9745                         "set_features() failed (%d); wanted %pNF, left %pNF\n",
9746                         err, &features, &dev->features);
9747                 /* return non-0 since some features might have changed and
9748                  * it's better to fire a spurious notification than miss it
9749                  */
9750                 return -1;
9751         }
9752
9753 sync_lower:
9754         /* some features must be disabled on lower devices when disabled
9755          * on an upper device (think: bonding master or bridge)
9756          */
9757         netdev_for_each_lower_dev(dev, lower, iter)
9758                 netdev_sync_lower_features(dev, lower, features);
9759
9760         if (!err) {
9761                 netdev_features_t diff = features ^ dev->features;
9762
9763                 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) {
9764                         /* udp_tunnel_{get,drop}_rx_info both need
9765                          * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the
9766                          * device, or they won't do anything.
9767                          * Thus we need to update dev->features
9768                          * *before* calling udp_tunnel_get_rx_info,
9769                          * but *after* calling udp_tunnel_drop_rx_info.
9770                          */
9771                         if (features & NETIF_F_RX_UDP_TUNNEL_PORT) {
9772                                 dev->features = features;
9773                                 udp_tunnel_get_rx_info(dev);
9774                         } else {
9775                                 udp_tunnel_drop_rx_info(dev);
9776                         }
9777                 }
9778
9779                 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) {
9780                         if (features & NETIF_F_HW_VLAN_CTAG_FILTER) {
9781                                 dev->features = features;
9782                                 err |= vlan_get_rx_ctag_filter_info(dev);
9783                         } else {
9784                                 vlan_drop_rx_ctag_filter_info(dev);
9785                         }
9786                 }
9787
9788                 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) {
9789                         if (features & NETIF_F_HW_VLAN_STAG_FILTER) {
9790                                 dev->features = features;
9791                                 err |= vlan_get_rx_stag_filter_info(dev);
9792                         } else {
9793                                 vlan_drop_rx_stag_filter_info(dev);
9794                         }
9795                 }
9796
9797                 dev->features = features;
9798         }
9799
9800         return err < 0 ? 0 : 1;
9801 }
9802
9803 /**
9804  *      netdev_update_features - recalculate device features
9805  *      @dev: the device to check
9806  *
9807  *      Recalculate dev->features set and send notifications if it
9808  *      has changed. Should be called after driver or hardware dependent
9809  *      conditions might have changed that influence the features.
9810  */
9811 void netdev_update_features(struct net_device *dev)
9812 {
9813         if (__netdev_update_features(dev))
9814                 netdev_features_change(dev);
9815 }
9816 EXPORT_SYMBOL(netdev_update_features);
9817
9818 /**
9819  *      netdev_change_features - recalculate device features
9820  *      @dev: the device to check
9821  *
9822  *      Recalculate dev->features set and send notifications even
9823  *      if they have not changed. Should be called instead of
9824  *      netdev_update_features() if also dev->vlan_features might
9825  *      have changed to allow the changes to be propagated to stacked
9826  *      VLAN devices.
9827  */
9828 void netdev_change_features(struct net_device *dev)
9829 {
9830         __netdev_update_features(dev);
9831         netdev_features_change(dev);
9832 }
9833 EXPORT_SYMBOL(netdev_change_features);
9834
9835 /**
9836  *      netif_stacked_transfer_operstate -      transfer operstate
9837  *      @rootdev: the root or lower level device to transfer state from
9838  *      @dev: the device to transfer operstate to
9839  *
9840  *      Transfer operational state from root to device. This is normally
9841  *      called when a stacking relationship exists between the root
9842  *      device and the device(a leaf device).
9843  */
9844 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
9845                                         struct net_device *dev)
9846 {
9847         if (rootdev->operstate == IF_OPER_DORMANT)
9848                 netif_dormant_on(dev);
9849         else
9850                 netif_dormant_off(dev);
9851
9852         if (rootdev->operstate == IF_OPER_TESTING)
9853                 netif_testing_on(dev);
9854         else
9855                 netif_testing_off(dev);
9856
9857         if (netif_carrier_ok(rootdev))
9858                 netif_carrier_on(dev);
9859         else
9860                 netif_carrier_off(dev);
9861 }
9862 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
9863
9864 static int netif_alloc_rx_queues(struct net_device *dev)
9865 {
9866         unsigned int i, count = dev->num_rx_queues;
9867         struct netdev_rx_queue *rx;
9868         size_t sz = count * sizeof(*rx);
9869         int err = 0;
9870
9871         BUG_ON(count < 1);
9872
9873         rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
9874         if (!rx)
9875                 return -ENOMEM;
9876
9877         dev->_rx = rx;
9878
9879         for (i = 0; i < count; i++) {
9880                 rx[i].dev = dev;
9881
9882                 /* XDP RX-queue setup */
9883                 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0);
9884                 if (err < 0)
9885                         goto err_rxq_info;
9886         }
9887         return 0;
9888
9889 err_rxq_info:
9890         /* Rollback successful reg's and free other resources */
9891         while (i--)
9892                 xdp_rxq_info_unreg(&rx[i].xdp_rxq);
9893         kvfree(dev->_rx);
9894         dev->_rx = NULL;
9895         return err;
9896 }
9897
9898 static void netif_free_rx_queues(struct net_device *dev)
9899 {
9900         unsigned int i, count = dev->num_rx_queues;
9901
9902         /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */
9903         if (!dev->_rx)
9904                 return;
9905
9906         for (i = 0; i < count; i++)
9907                 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq);
9908
9909         kvfree(dev->_rx);
9910 }
9911
9912 static void netdev_init_one_queue(struct net_device *dev,
9913                                   struct netdev_queue *queue, void *_unused)
9914 {
9915         /* Initialize queue lock */
9916         spin_lock_init(&queue->_xmit_lock);
9917         netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
9918         queue->xmit_lock_owner = -1;
9919         netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
9920         queue->dev = dev;
9921 #ifdef CONFIG_BQL
9922         dql_init(&queue->dql, HZ);
9923 #endif
9924 }
9925
9926 static void netif_free_tx_queues(struct net_device *dev)
9927 {
9928         kvfree(dev->_tx);
9929 }
9930
9931 static int netif_alloc_netdev_queues(struct net_device *dev)
9932 {
9933         unsigned int count = dev->num_tx_queues;
9934         struct netdev_queue *tx;
9935         size_t sz = count * sizeof(*tx);
9936
9937         if (count < 1 || count > 0xffff)
9938                 return -EINVAL;
9939
9940         tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
9941         if (!tx)
9942                 return -ENOMEM;
9943
9944         dev->_tx = tx;
9945
9946         netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
9947         spin_lock_init(&dev->tx_global_lock);
9948
9949         return 0;
9950 }
9951
9952 void netif_tx_stop_all_queues(struct net_device *dev)
9953 {
9954         unsigned int i;
9955
9956         for (i = 0; i < dev->num_tx_queues; i++) {
9957                 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
9958
9959                 netif_tx_stop_queue(txq);
9960         }
9961 }
9962 EXPORT_SYMBOL(netif_tx_stop_all_queues);
9963
9964 /**
9965  * register_netdevice() - register a network device
9966  * @dev: device to register
9967  *
9968  * Take a prepared network device structure and make it externally accessible.
9969  * A %NETDEV_REGISTER message is sent to the netdev notifier chain.
9970  * Callers must hold the rtnl lock - you may want register_netdev()
9971  * instead of this.
9972  */
9973 int register_netdevice(struct net_device *dev)
9974 {
9975         int ret;
9976         struct net *net = dev_net(dev);
9977
9978         BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE <
9979                      NETDEV_FEATURE_COUNT);
9980         BUG_ON(dev_boot_phase);
9981         ASSERT_RTNL();
9982
9983         might_sleep();
9984
9985         /* When net_device's are persistent, this will be fatal. */
9986         BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
9987         BUG_ON(!net);
9988
9989         ret = ethtool_check_ops(dev->ethtool_ops);
9990         if (ret)
9991                 return ret;
9992
9993         spin_lock_init(&dev->addr_list_lock);
9994         netdev_set_addr_lockdep_class(dev);
9995
9996         ret = dev_get_valid_name(net, dev, dev->name);
9997         if (ret < 0)
9998                 goto out;
9999
10000         ret = -ENOMEM;
10001         dev->name_node = netdev_name_node_head_alloc(dev);
10002         if (!dev->name_node)
10003                 goto out;
10004
10005         /* Init, if this function is available */
10006         if (dev->netdev_ops->ndo_init) {
10007                 ret = dev->netdev_ops->ndo_init(dev);
10008                 if (ret) {
10009                         if (ret > 0)
10010                                 ret = -EIO;
10011                         goto err_free_name;
10012                 }
10013         }
10014
10015         if (((dev->hw_features | dev->features) &
10016              NETIF_F_HW_VLAN_CTAG_FILTER) &&
10017             (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
10018              !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
10019                 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
10020                 ret = -EINVAL;
10021                 goto err_uninit;
10022         }
10023
10024         ret = -EBUSY;
10025         if (!dev->ifindex)
10026                 dev->ifindex = dev_new_index(net);
10027         else if (__dev_get_by_index(net, dev->ifindex))
10028                 goto err_uninit;
10029
10030         /* Transfer changeable features to wanted_features and enable
10031          * software offloads (GSO and GRO).
10032          */
10033         dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF);
10034         dev->features |= NETIF_F_SOFT_FEATURES;
10035
10036         if (dev->udp_tunnel_nic_info) {
10037                 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT;
10038                 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT;
10039         }
10040
10041         dev->wanted_features = dev->features & dev->hw_features;
10042
10043         if (!(dev->flags & IFF_LOOPBACK))
10044                 dev->hw_features |= NETIF_F_NOCACHE_COPY;
10045
10046         /* If IPv4 TCP segmentation offload is supported we should also
10047          * allow the device to enable segmenting the frame with the option
10048          * of ignoring a static IP ID value.  This doesn't enable the
10049          * feature itself but allows the user to enable it later.
10050          */
10051         if (dev->hw_features & NETIF_F_TSO)
10052                 dev->hw_features |= NETIF_F_TSO_MANGLEID;
10053         if (dev->vlan_features & NETIF_F_TSO)
10054                 dev->vlan_features |= NETIF_F_TSO_MANGLEID;
10055         if (dev->mpls_features & NETIF_F_TSO)
10056                 dev->mpls_features |= NETIF_F_TSO_MANGLEID;
10057         if (dev->hw_enc_features & NETIF_F_TSO)
10058                 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
10059
10060         /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
10061          */
10062         dev->vlan_features |= NETIF_F_HIGHDMA;
10063
10064         /* Make NETIF_F_SG inheritable to tunnel devices.
10065          */
10066         dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL;
10067
10068         /* Make NETIF_F_SG inheritable to MPLS.
10069          */
10070         dev->mpls_features |= NETIF_F_SG;
10071
10072         ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
10073         ret = notifier_to_errno(ret);
10074         if (ret)
10075                 goto err_uninit;
10076
10077         ret = netdev_register_kobject(dev);
10078         write_lock(&dev_base_lock);
10079         dev->reg_state = ret ? NETREG_UNREGISTERED : NETREG_REGISTERED;
10080         write_unlock(&dev_base_lock);
10081         if (ret)
10082                 goto err_uninit_notify;
10083
10084         __netdev_update_features(dev);
10085
10086         /*
10087          *      Default initial state at registry is that the
10088          *      device is present.
10089          */
10090
10091         set_bit(__LINK_STATE_PRESENT, &dev->state);
10092
10093         linkwatch_init_dev(dev);
10094
10095         dev_init_scheduler(dev);
10096
10097         netdev_hold(dev, &dev->dev_registered_tracker, GFP_KERNEL);
10098         list_netdevice(dev);
10099
10100         add_device_randomness(dev->dev_addr, dev->addr_len);
10101
10102         /* If the device has permanent device address, driver should
10103          * set dev_addr and also addr_assign_type should be set to
10104          * NET_ADDR_PERM (default value).
10105          */
10106         if (dev->addr_assign_type == NET_ADDR_PERM)
10107                 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
10108
10109         /* Notify protocols, that a new device appeared. */
10110         ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
10111         ret = notifier_to_errno(ret);
10112         if (ret) {
10113                 /* Expect explicit free_netdev() on failure */
10114                 dev->needs_free_netdev = false;
10115                 unregister_netdevice_queue(dev, NULL);
10116                 goto out;
10117         }
10118         /*
10119          *      Prevent userspace races by waiting until the network
10120          *      device is fully setup before sending notifications.
10121          */
10122         if (!dev->rtnl_link_ops ||
10123             dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
10124                 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL);
10125
10126 out:
10127         return ret;
10128
10129 err_uninit_notify:
10130         call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev);
10131 err_uninit:
10132         if (dev->netdev_ops->ndo_uninit)
10133                 dev->netdev_ops->ndo_uninit(dev);
10134         if (dev->priv_destructor)
10135                 dev->priv_destructor(dev);
10136 err_free_name:
10137         netdev_name_node_free(dev->name_node);
10138         goto out;
10139 }
10140 EXPORT_SYMBOL(register_netdevice);
10141
10142 /**
10143  *      init_dummy_netdev       - init a dummy network device for NAPI
10144  *      @dev: device to init
10145  *
10146  *      This takes a network device structure and initialize the minimum
10147  *      amount of fields so it can be used to schedule NAPI polls without
10148  *      registering a full blown interface. This is to be used by drivers
10149  *      that need to tie several hardware interfaces to a single NAPI
10150  *      poll scheduler due to HW limitations.
10151  */
10152 int init_dummy_netdev(struct net_device *dev)
10153 {
10154         /* Clear everything. Note we don't initialize spinlocks
10155          * are they aren't supposed to be taken by any of the
10156          * NAPI code and this dummy netdev is supposed to be
10157          * only ever used for NAPI polls
10158          */
10159         memset(dev, 0, sizeof(struct net_device));
10160
10161         /* make sure we BUG if trying to hit standard
10162          * register/unregister code path
10163          */
10164         dev->reg_state = NETREG_DUMMY;
10165
10166         /* NAPI wants this */
10167         INIT_LIST_HEAD(&dev->napi_list);
10168
10169         /* a dummy interface is started by default */
10170         set_bit(__LINK_STATE_PRESENT, &dev->state);
10171         set_bit(__LINK_STATE_START, &dev->state);
10172
10173         /* napi_busy_loop stats accounting wants this */
10174         dev_net_set(dev, &init_net);
10175
10176         /* Note : We dont allocate pcpu_refcnt for dummy devices,
10177          * because users of this 'device' dont need to change
10178          * its refcount.
10179          */
10180
10181         return 0;
10182 }
10183 EXPORT_SYMBOL_GPL(init_dummy_netdev);
10184
10185
10186 /**
10187  *      register_netdev - register a network device
10188  *      @dev: device to register
10189  *
10190  *      Take a completed network device structure and add it to the kernel
10191  *      interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
10192  *      chain. 0 is returned on success. A negative errno code is returned
10193  *      on a failure to set up the device, or if the name is a duplicate.
10194  *
10195  *      This is a wrapper around register_netdevice that takes the rtnl semaphore
10196  *      and expands the device name if you passed a format string to
10197  *      alloc_netdev.
10198  */
10199 int register_netdev(struct net_device *dev)
10200 {
10201         int err;
10202
10203         if (rtnl_lock_killable())
10204                 return -EINTR;
10205         err = register_netdevice(dev);
10206         rtnl_unlock();
10207         return err;
10208 }
10209 EXPORT_SYMBOL(register_netdev);
10210
10211 int netdev_refcnt_read(const struct net_device *dev)
10212 {
10213 #ifdef CONFIG_PCPU_DEV_REFCNT
10214         int i, refcnt = 0;
10215
10216         for_each_possible_cpu(i)
10217                 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
10218         return refcnt;
10219 #else
10220         return refcount_read(&dev->dev_refcnt);
10221 #endif
10222 }
10223 EXPORT_SYMBOL(netdev_refcnt_read);
10224
10225 int netdev_unregister_timeout_secs __read_mostly = 10;
10226
10227 #define WAIT_REFS_MIN_MSECS 1
10228 #define WAIT_REFS_MAX_MSECS 250
10229 /**
10230  * netdev_wait_allrefs_any - wait until all references are gone.
10231  * @list: list of net_devices to wait on
10232  *
10233  * This is called when unregistering network devices.
10234  *
10235  * Any protocol or device that holds a reference should register
10236  * for netdevice notification, and cleanup and put back the
10237  * reference if they receive an UNREGISTER event.
10238  * We can get stuck here if buggy protocols don't correctly
10239  * call dev_put.
10240  */
10241 static struct net_device *netdev_wait_allrefs_any(struct list_head *list)
10242 {
10243         unsigned long rebroadcast_time, warning_time;
10244         struct net_device *dev;
10245         int wait = 0;
10246
10247         rebroadcast_time = warning_time = jiffies;
10248
10249         list_for_each_entry(dev, list, todo_list)
10250                 if (netdev_refcnt_read(dev) == 1)
10251                         return dev;
10252
10253         while (true) {
10254                 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
10255                         rtnl_lock();
10256
10257                         /* Rebroadcast unregister notification */
10258                         list_for_each_entry(dev, list, todo_list)
10259                                 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
10260
10261                         __rtnl_unlock();
10262                         rcu_barrier();
10263                         rtnl_lock();
10264
10265                         list_for_each_entry(dev, list, todo_list)
10266                                 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
10267                                              &dev->state)) {
10268                                         /* We must not have linkwatch events
10269                                          * pending on unregister. If this
10270                                          * happens, we simply run the queue
10271                                          * unscheduled, resulting in a noop
10272                                          * for this device.
10273                                          */
10274                                         linkwatch_run_queue();
10275                                         break;
10276                                 }
10277
10278                         __rtnl_unlock();
10279
10280                         rebroadcast_time = jiffies;
10281                 }
10282
10283                 if (!wait) {
10284                         rcu_barrier();
10285                         wait = WAIT_REFS_MIN_MSECS;
10286                 } else {
10287                         msleep(wait);
10288                         wait = min(wait << 1, WAIT_REFS_MAX_MSECS);
10289                 }
10290
10291                 list_for_each_entry(dev, list, todo_list)
10292                         if (netdev_refcnt_read(dev) == 1)
10293                                 return dev;
10294
10295                 if (time_after(jiffies, warning_time +
10296                                READ_ONCE(netdev_unregister_timeout_secs) * HZ)) {
10297                         list_for_each_entry(dev, list, todo_list) {
10298                                 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
10299                                          dev->name, netdev_refcnt_read(dev));
10300                                 ref_tracker_dir_print(&dev->refcnt_tracker, 10);
10301                         }
10302
10303                         warning_time = jiffies;
10304                 }
10305         }
10306 }
10307
10308 /* The sequence is:
10309  *
10310  *      rtnl_lock();
10311  *      ...
10312  *      register_netdevice(x1);
10313  *      register_netdevice(x2);
10314  *      ...
10315  *      unregister_netdevice(y1);
10316  *      unregister_netdevice(y2);
10317  *      ...
10318  *      rtnl_unlock();
10319  *      free_netdev(y1);
10320  *      free_netdev(y2);
10321  *
10322  * We are invoked by rtnl_unlock().
10323  * This allows us to deal with problems:
10324  * 1) We can delete sysfs objects which invoke hotplug
10325  *    without deadlocking with linkwatch via keventd.
10326  * 2) Since we run with the RTNL semaphore not held, we can sleep
10327  *    safely in order to wait for the netdev refcnt to drop to zero.
10328  *
10329  * We must not return until all unregister events added during
10330  * the interval the lock was held have been completed.
10331  */
10332 void netdev_run_todo(void)
10333 {
10334         struct net_device *dev, *tmp;
10335         struct list_head list;
10336 #ifdef CONFIG_LOCKDEP
10337         struct list_head unlink_list;
10338
10339         list_replace_init(&net_unlink_list, &unlink_list);
10340
10341         while (!list_empty(&unlink_list)) {
10342                 struct net_device *dev = list_first_entry(&unlink_list,
10343                                                           struct net_device,
10344                                                           unlink_list);
10345                 list_del_init(&dev->unlink_list);
10346                 dev->nested_level = dev->lower_level - 1;
10347         }
10348 #endif
10349
10350         /* Snapshot list, allow later requests */
10351         list_replace_init(&net_todo_list, &list);
10352
10353         __rtnl_unlock();
10354
10355         /* Wait for rcu callbacks to finish before next phase */
10356         if (!list_empty(&list))
10357                 rcu_barrier();
10358
10359         list_for_each_entry_safe(dev, tmp, &list, todo_list) {
10360                 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
10361                         netdev_WARN(dev, "run_todo but not unregistering\n");
10362                         list_del(&dev->todo_list);
10363                         continue;
10364                 }
10365
10366                 write_lock(&dev_base_lock);
10367                 dev->reg_state = NETREG_UNREGISTERED;
10368                 write_unlock(&dev_base_lock);
10369                 linkwatch_forget_dev(dev);
10370         }
10371
10372         while (!list_empty(&list)) {
10373                 dev = netdev_wait_allrefs_any(&list);
10374                 list_del(&dev->todo_list);
10375
10376                 /* paranoia */
10377                 BUG_ON(netdev_refcnt_read(dev) != 1);
10378                 BUG_ON(!list_empty(&dev->ptype_all));
10379                 BUG_ON(!list_empty(&dev->ptype_specific));
10380                 WARN_ON(rcu_access_pointer(dev->ip_ptr));
10381                 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
10382
10383                 if (dev->priv_destructor)
10384                         dev->priv_destructor(dev);
10385                 if (dev->needs_free_netdev)
10386                         free_netdev(dev);
10387
10388                 if (atomic_dec_and_test(&dev_net(dev)->dev_unreg_count))
10389                         wake_up(&netdev_unregistering_wq);
10390
10391                 /* Free network device */
10392                 kobject_put(&dev->dev.kobj);
10393         }
10394 }
10395
10396 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has
10397  * all the same fields in the same order as net_device_stats, with only
10398  * the type differing, but rtnl_link_stats64 may have additional fields
10399  * at the end for newer counters.
10400  */
10401 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
10402                              const struct net_device_stats *netdev_stats)
10403 {
10404         size_t i, n = sizeof(*netdev_stats) / sizeof(atomic_long_t);
10405         const atomic_long_t *src = (atomic_long_t *)netdev_stats;
10406         u64 *dst = (u64 *)stats64;
10407
10408         BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64));
10409         for (i = 0; i < n; i++)
10410                 dst[i] = (unsigned long)atomic_long_read(&src[i]);
10411         /* zero out counters that only exist in rtnl_link_stats64 */
10412         memset((char *)stats64 + n * sizeof(u64), 0,
10413                sizeof(*stats64) - n * sizeof(u64));
10414 }
10415 EXPORT_SYMBOL(netdev_stats_to_stats64);
10416
10417 struct net_device_core_stats __percpu *netdev_core_stats_alloc(struct net_device *dev)
10418 {
10419         struct net_device_core_stats __percpu *p;
10420
10421         p = alloc_percpu_gfp(struct net_device_core_stats,
10422                              GFP_ATOMIC | __GFP_NOWARN);
10423
10424         if (p && cmpxchg(&dev->core_stats, NULL, p))
10425                 free_percpu(p);
10426
10427         /* This READ_ONCE() pairs with the cmpxchg() above */
10428         return READ_ONCE(dev->core_stats);
10429 }
10430 EXPORT_SYMBOL(netdev_core_stats_alloc);
10431
10432 /**
10433  *      dev_get_stats   - get network device statistics
10434  *      @dev: device to get statistics from
10435  *      @storage: place to store stats
10436  *
10437  *      Get network statistics from device. Return @storage.
10438  *      The device driver may provide its own method by setting
10439  *      dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
10440  *      otherwise the internal statistics structure is used.
10441  */
10442 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
10443                                         struct rtnl_link_stats64 *storage)
10444 {
10445         const struct net_device_ops *ops = dev->netdev_ops;
10446         const struct net_device_core_stats __percpu *p;
10447
10448         if (ops->ndo_get_stats64) {
10449                 memset(storage, 0, sizeof(*storage));
10450                 ops->ndo_get_stats64(dev, storage);
10451         } else if (ops->ndo_get_stats) {
10452                 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
10453         } else {
10454                 netdev_stats_to_stats64(storage, &dev->stats);
10455         }
10456
10457         /* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */
10458         p = READ_ONCE(dev->core_stats);
10459         if (p) {
10460                 const struct net_device_core_stats *core_stats;
10461                 int i;
10462
10463                 for_each_possible_cpu(i) {
10464                         core_stats = per_cpu_ptr(p, i);
10465                         storage->rx_dropped += READ_ONCE(core_stats->rx_dropped);
10466                         storage->tx_dropped += READ_ONCE(core_stats->tx_dropped);
10467                         storage->rx_nohandler += READ_ONCE(core_stats->rx_nohandler);
10468                         storage->rx_otherhost_dropped += READ_ONCE(core_stats->rx_otherhost_dropped);
10469                 }
10470         }
10471         return storage;
10472 }
10473 EXPORT_SYMBOL(dev_get_stats);
10474
10475 /**
10476  *      dev_fetch_sw_netstats - get per-cpu network device statistics
10477  *      @s: place to store stats
10478  *      @netstats: per-cpu network stats to read from
10479  *
10480  *      Read per-cpu network statistics and populate the related fields in @s.
10481  */
10482 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s,
10483                            const struct pcpu_sw_netstats __percpu *netstats)
10484 {
10485         int cpu;
10486
10487         for_each_possible_cpu(cpu) {
10488                 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
10489                 const struct pcpu_sw_netstats *stats;
10490                 unsigned int start;
10491
10492                 stats = per_cpu_ptr(netstats, cpu);
10493                 do {
10494                         start = u64_stats_fetch_begin(&stats->syncp);
10495                         rx_packets = u64_stats_read(&stats->rx_packets);
10496                         rx_bytes   = u64_stats_read(&stats->rx_bytes);
10497                         tx_packets = u64_stats_read(&stats->tx_packets);
10498                         tx_bytes   = u64_stats_read(&stats->tx_bytes);
10499                 } while (u64_stats_fetch_retry(&stats->syncp, start));
10500
10501                 s->rx_packets += rx_packets;
10502                 s->rx_bytes   += rx_bytes;
10503                 s->tx_packets += tx_packets;
10504                 s->tx_bytes   += tx_bytes;
10505         }
10506 }
10507 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats);
10508
10509 /**
10510  *      dev_get_tstats64 - ndo_get_stats64 implementation
10511  *      @dev: device to get statistics from
10512  *      @s: place to store stats
10513  *
10514  *      Populate @s from dev->stats and dev->tstats. Can be used as
10515  *      ndo_get_stats64() callback.
10516  */
10517 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s)
10518 {
10519         netdev_stats_to_stats64(s, &dev->stats);
10520         dev_fetch_sw_netstats(s, dev->tstats);
10521 }
10522 EXPORT_SYMBOL_GPL(dev_get_tstats64);
10523
10524 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
10525 {
10526         struct netdev_queue *queue = dev_ingress_queue(dev);
10527
10528 #ifdef CONFIG_NET_CLS_ACT
10529         if (queue)
10530                 return queue;
10531         queue = kzalloc(sizeof(*queue), GFP_KERNEL);
10532         if (!queue)
10533                 return NULL;
10534         netdev_init_one_queue(dev, queue, NULL);
10535         RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
10536         queue->qdisc_sleeping = &noop_qdisc;
10537         rcu_assign_pointer(dev->ingress_queue, queue);
10538 #endif
10539         return queue;
10540 }
10541
10542 static const struct ethtool_ops default_ethtool_ops;
10543
10544 void netdev_set_default_ethtool_ops(struct net_device *dev,
10545                                     const struct ethtool_ops *ops)
10546 {
10547         if (dev->ethtool_ops == &default_ethtool_ops)
10548                 dev->ethtool_ops = ops;
10549 }
10550 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
10551
10552 /**
10553  * netdev_sw_irq_coalesce_default_on() - enable SW IRQ coalescing by default
10554  * @dev: netdev to enable the IRQ coalescing on
10555  *
10556  * Sets a conservative default for SW IRQ coalescing. Users can use
10557  * sysfs attributes to override the default values.
10558  */
10559 void netdev_sw_irq_coalesce_default_on(struct net_device *dev)
10560 {
10561         WARN_ON(dev->reg_state == NETREG_REGISTERED);
10562
10563         dev->gro_flush_timeout = 20000;
10564         dev->napi_defer_hard_irqs = 1;
10565 }
10566 EXPORT_SYMBOL_GPL(netdev_sw_irq_coalesce_default_on);
10567
10568 void netdev_freemem(struct net_device *dev)
10569 {
10570         char *addr = (char *)dev - dev->padded;
10571
10572         kvfree(addr);
10573 }
10574
10575 /**
10576  * alloc_netdev_mqs - allocate network device
10577  * @sizeof_priv: size of private data to allocate space for
10578  * @name: device name format string
10579  * @name_assign_type: origin of device name
10580  * @setup: callback to initialize device
10581  * @txqs: the number of TX subqueues to allocate
10582  * @rxqs: the number of RX subqueues to allocate
10583  *
10584  * Allocates a struct net_device with private data area for driver use
10585  * and performs basic initialization.  Also allocates subqueue structs
10586  * for each queue on the device.
10587  */
10588 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
10589                 unsigned char name_assign_type,
10590                 void (*setup)(struct net_device *),
10591                 unsigned int txqs, unsigned int rxqs)
10592 {
10593         struct net_device *dev;
10594         unsigned int alloc_size;
10595         struct net_device *p;
10596
10597         BUG_ON(strlen(name) >= sizeof(dev->name));
10598
10599         if (txqs < 1) {
10600                 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
10601                 return NULL;
10602         }
10603
10604         if (rxqs < 1) {
10605                 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
10606                 return NULL;
10607         }
10608
10609         alloc_size = sizeof(struct net_device);
10610         if (sizeof_priv) {
10611                 /* ensure 32-byte alignment of private area */
10612                 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
10613                 alloc_size += sizeof_priv;
10614         }
10615         /* ensure 32-byte alignment of whole construct */
10616         alloc_size += NETDEV_ALIGN - 1;
10617
10618         p = kvzalloc(alloc_size, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
10619         if (!p)
10620                 return NULL;
10621
10622         dev = PTR_ALIGN(p, NETDEV_ALIGN);
10623         dev->padded = (char *)dev - (char *)p;
10624
10625         ref_tracker_dir_init(&dev->refcnt_tracker, 128);
10626 #ifdef CONFIG_PCPU_DEV_REFCNT
10627         dev->pcpu_refcnt = alloc_percpu(int);
10628         if (!dev->pcpu_refcnt)
10629                 goto free_dev;
10630         __dev_hold(dev);
10631 #else
10632         refcount_set(&dev->dev_refcnt, 1);
10633 #endif
10634
10635         if (dev_addr_init(dev))
10636                 goto free_pcpu;
10637
10638         dev_mc_init(dev);
10639         dev_uc_init(dev);
10640
10641         dev_net_set(dev, &init_net);
10642
10643         dev->gso_max_size = GSO_LEGACY_MAX_SIZE;
10644         dev->gso_max_segs = GSO_MAX_SEGS;
10645         dev->gro_max_size = GRO_LEGACY_MAX_SIZE;
10646         dev->gso_ipv4_max_size = GSO_LEGACY_MAX_SIZE;
10647         dev->gro_ipv4_max_size = GRO_LEGACY_MAX_SIZE;
10648         dev->tso_max_size = TSO_LEGACY_MAX_SIZE;
10649         dev->tso_max_segs = TSO_MAX_SEGS;
10650         dev->upper_level = 1;
10651         dev->lower_level = 1;
10652 #ifdef CONFIG_LOCKDEP
10653         dev->nested_level = 0;
10654         INIT_LIST_HEAD(&dev->unlink_list);
10655 #endif
10656
10657         INIT_LIST_HEAD(&dev->napi_list);
10658         INIT_LIST_HEAD(&dev->unreg_list);
10659         INIT_LIST_HEAD(&dev->close_list);
10660         INIT_LIST_HEAD(&dev->link_watch_list);
10661         INIT_LIST_HEAD(&dev->adj_list.upper);
10662         INIT_LIST_HEAD(&dev->adj_list.lower);
10663         INIT_LIST_HEAD(&dev->ptype_all);
10664         INIT_LIST_HEAD(&dev->ptype_specific);
10665         INIT_LIST_HEAD(&dev->net_notifier_list);
10666 #ifdef CONFIG_NET_SCHED
10667         hash_init(dev->qdisc_hash);
10668 #endif
10669         dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
10670         setup(dev);
10671
10672         if (!dev->tx_queue_len) {
10673                 dev->priv_flags |= IFF_NO_QUEUE;
10674                 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
10675         }
10676
10677         dev->num_tx_queues = txqs;
10678         dev->real_num_tx_queues = txqs;
10679         if (netif_alloc_netdev_queues(dev))
10680                 goto free_all;
10681
10682         dev->num_rx_queues = rxqs;
10683         dev->real_num_rx_queues = rxqs;
10684         if (netif_alloc_rx_queues(dev))
10685                 goto free_all;
10686
10687         strcpy(dev->name, name);
10688         dev->name_assign_type = name_assign_type;
10689         dev->group = INIT_NETDEV_GROUP;
10690         if (!dev->ethtool_ops)
10691                 dev->ethtool_ops = &default_ethtool_ops;
10692
10693         nf_hook_netdev_init(dev);
10694
10695         return dev;
10696
10697 free_all:
10698         free_netdev(dev);
10699         return NULL;
10700
10701 free_pcpu:
10702 #ifdef CONFIG_PCPU_DEV_REFCNT
10703         free_percpu(dev->pcpu_refcnt);
10704 free_dev:
10705 #endif
10706         netdev_freemem(dev);
10707         return NULL;
10708 }
10709 EXPORT_SYMBOL(alloc_netdev_mqs);
10710
10711 /**
10712  * free_netdev - free network device
10713  * @dev: device
10714  *
10715  * This function does the last stage of destroying an allocated device
10716  * interface. The reference to the device object is released. If this
10717  * is the last reference then it will be freed.Must be called in process
10718  * context.
10719  */
10720 void free_netdev(struct net_device *dev)
10721 {
10722         struct napi_struct *p, *n;
10723
10724         might_sleep();
10725
10726         /* When called immediately after register_netdevice() failed the unwind
10727          * handling may still be dismantling the device. Handle that case by
10728          * deferring the free.
10729          */
10730         if (dev->reg_state == NETREG_UNREGISTERING) {
10731                 ASSERT_RTNL();
10732                 dev->needs_free_netdev = true;
10733                 return;
10734         }
10735
10736         netif_free_tx_queues(dev);
10737         netif_free_rx_queues(dev);
10738
10739         kfree(rcu_dereference_protected(dev->ingress_queue, 1));
10740
10741         /* Flush device addresses */
10742         dev_addr_flush(dev);
10743
10744         list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
10745                 netif_napi_del(p);
10746
10747         ref_tracker_dir_exit(&dev->refcnt_tracker);
10748 #ifdef CONFIG_PCPU_DEV_REFCNT
10749         free_percpu(dev->pcpu_refcnt);
10750         dev->pcpu_refcnt = NULL;
10751 #endif
10752         free_percpu(dev->core_stats);
10753         dev->core_stats = NULL;
10754         free_percpu(dev->xdp_bulkq);
10755         dev->xdp_bulkq = NULL;
10756
10757         /*  Compatibility with error handling in drivers */
10758         if (dev->reg_state == NETREG_UNINITIALIZED) {
10759                 netdev_freemem(dev);
10760                 return;
10761         }
10762
10763         BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
10764         dev->reg_state = NETREG_RELEASED;
10765
10766         /* will free via device release */
10767         put_device(&dev->dev);
10768 }
10769 EXPORT_SYMBOL(free_netdev);
10770
10771 /**
10772  *      synchronize_net -  Synchronize with packet receive processing
10773  *
10774  *      Wait for packets currently being received to be done.
10775  *      Does not block later packets from starting.
10776  */
10777 void synchronize_net(void)
10778 {
10779         might_sleep();
10780         if (rtnl_is_locked())
10781                 synchronize_rcu_expedited();
10782         else
10783                 synchronize_rcu();
10784 }
10785 EXPORT_SYMBOL(synchronize_net);
10786
10787 /**
10788  *      unregister_netdevice_queue - remove device from the kernel
10789  *      @dev: device
10790  *      @head: list
10791  *
10792  *      This function shuts down a device interface and removes it
10793  *      from the kernel tables.
10794  *      If head not NULL, device is queued to be unregistered later.
10795  *
10796  *      Callers must hold the rtnl semaphore.  You may want
10797  *      unregister_netdev() instead of this.
10798  */
10799
10800 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
10801 {
10802         ASSERT_RTNL();
10803
10804         if (head) {
10805                 list_move_tail(&dev->unreg_list, head);
10806         } else {
10807                 LIST_HEAD(single);
10808
10809                 list_add(&dev->unreg_list, &single);
10810                 unregister_netdevice_many(&single);
10811         }
10812 }
10813 EXPORT_SYMBOL(unregister_netdevice_queue);
10814
10815 void unregister_netdevice_many_notify(struct list_head *head,
10816                                       u32 portid, const struct nlmsghdr *nlh)
10817 {
10818         struct net_device *dev, *tmp;
10819         LIST_HEAD(close_head);
10820
10821         BUG_ON(dev_boot_phase);
10822         ASSERT_RTNL();
10823
10824         if (list_empty(head))
10825                 return;
10826
10827         list_for_each_entry_safe(dev, tmp, head, unreg_list) {
10828                 /* Some devices call without registering
10829                  * for initialization unwind. Remove those
10830                  * devices and proceed with the remaining.
10831                  */
10832                 if (dev->reg_state == NETREG_UNINITIALIZED) {
10833                         pr_debug("unregister_netdevice: device %s/%p never was registered\n",
10834                                  dev->name, dev);
10835
10836                         WARN_ON(1);
10837                         list_del(&dev->unreg_list);
10838                         continue;
10839                 }
10840                 dev->dismantle = true;
10841                 BUG_ON(dev->reg_state != NETREG_REGISTERED);
10842         }
10843
10844         /* If device is running, close it first. */
10845         list_for_each_entry(dev, head, unreg_list)
10846                 list_add_tail(&dev->close_list, &close_head);
10847         dev_close_many(&close_head, true);
10848
10849         list_for_each_entry(dev, head, unreg_list) {
10850                 /* And unlink it from device chain. */
10851                 write_lock(&dev_base_lock);
10852                 unlist_netdevice(dev, false);
10853                 dev->reg_state = NETREG_UNREGISTERING;
10854                 write_unlock(&dev_base_lock);
10855         }
10856         flush_all_backlogs();
10857
10858         synchronize_net();
10859
10860         list_for_each_entry(dev, head, unreg_list) {
10861                 struct sk_buff *skb = NULL;
10862
10863                 /* Shutdown queueing discipline. */
10864                 dev_shutdown(dev);
10865
10866                 dev_xdp_uninstall(dev);
10867                 bpf_dev_bound_netdev_unregister(dev);
10868
10869                 netdev_offload_xstats_disable_all(dev);
10870
10871                 /* Notify protocols, that we are about to destroy
10872                  * this device. They should clean all the things.
10873                  */
10874                 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
10875
10876                 if (!dev->rtnl_link_ops ||
10877                     dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
10878                         skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0,
10879                                                      GFP_KERNEL, NULL, 0,
10880                                                      portid, nlh);
10881
10882                 /*
10883                  *      Flush the unicast and multicast chains
10884                  */
10885                 dev_uc_flush(dev);
10886                 dev_mc_flush(dev);
10887
10888                 netdev_name_node_alt_flush(dev);
10889                 netdev_name_node_free(dev->name_node);
10890
10891                 call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev);
10892
10893                 if (dev->netdev_ops->ndo_uninit)
10894                         dev->netdev_ops->ndo_uninit(dev);
10895
10896                 if (skb)
10897                         rtmsg_ifinfo_send(skb, dev, GFP_KERNEL, portid, nlh);
10898
10899                 /* Notifier chain MUST detach us all upper devices. */
10900                 WARN_ON(netdev_has_any_upper_dev(dev));
10901                 WARN_ON(netdev_has_any_lower_dev(dev));
10902
10903                 /* Remove entries from kobject tree */
10904                 netdev_unregister_kobject(dev);
10905 #ifdef CONFIG_XPS
10906                 /* Remove XPS queueing entries */
10907                 netif_reset_xps_queues_gt(dev, 0);
10908 #endif
10909         }
10910
10911         synchronize_net();
10912
10913         list_for_each_entry(dev, head, unreg_list) {
10914                 netdev_put(dev, &dev->dev_registered_tracker);
10915                 net_set_todo(dev);
10916         }
10917
10918         list_del(head);
10919 }
10920
10921 /**
10922  *      unregister_netdevice_many - unregister many devices
10923  *      @head: list of devices
10924  *
10925  *  Note: As most callers use a stack allocated list_head,
10926  *  we force a list_del() to make sure stack wont be corrupted later.
10927  */
10928 void unregister_netdevice_many(struct list_head *head)
10929 {
10930         unregister_netdevice_many_notify(head, 0, NULL);
10931 }
10932 EXPORT_SYMBOL(unregister_netdevice_many);
10933
10934 /**
10935  *      unregister_netdev - remove device from the kernel
10936  *      @dev: device
10937  *
10938  *      This function shuts down a device interface and removes it
10939  *      from the kernel tables.
10940  *
10941  *      This is just a wrapper for unregister_netdevice that takes
10942  *      the rtnl semaphore.  In general you want to use this and not
10943  *      unregister_netdevice.
10944  */
10945 void unregister_netdev(struct net_device *dev)
10946 {
10947         rtnl_lock();
10948         unregister_netdevice(dev);
10949         rtnl_unlock();
10950 }
10951 EXPORT_SYMBOL(unregister_netdev);
10952
10953 /**
10954  *      __dev_change_net_namespace - move device to different nethost namespace
10955  *      @dev: device
10956  *      @net: network namespace
10957  *      @pat: If not NULL name pattern to try if the current device name
10958  *            is already taken in the destination network namespace.
10959  *      @new_ifindex: If not zero, specifies device index in the target
10960  *                    namespace.
10961  *
10962  *      This function shuts down a device interface and moves it
10963  *      to a new network namespace. On success 0 is returned, on
10964  *      a failure a netagive errno code is returned.
10965  *
10966  *      Callers must hold the rtnl semaphore.
10967  */
10968
10969 int __dev_change_net_namespace(struct net_device *dev, struct net *net,
10970                                const char *pat, int new_ifindex)
10971 {
10972         struct net *net_old = dev_net(dev);
10973         int err, new_nsid;
10974
10975         ASSERT_RTNL();
10976
10977         /* Don't allow namespace local devices to be moved. */
10978         err = -EINVAL;
10979         if (dev->features & NETIF_F_NETNS_LOCAL)
10980                 goto out;
10981
10982         /* Ensure the device has been registrered */
10983         if (dev->reg_state != NETREG_REGISTERED)
10984                 goto out;
10985
10986         /* Get out if there is nothing todo */
10987         err = 0;
10988         if (net_eq(net_old, net))
10989                 goto out;
10990
10991         /* Pick the destination device name, and ensure
10992          * we can use it in the destination network namespace.
10993          */
10994         err = -EEXIST;
10995         if (netdev_name_in_use(net, dev->name)) {
10996                 /* We get here if we can't use the current device name */
10997                 if (!pat)
10998                         goto out;
10999                 err = dev_get_valid_name(net, dev, pat);
11000                 if (err < 0)
11001                         goto out;
11002         }
11003
11004         /* Check that new_ifindex isn't used yet. */
11005         err = -EBUSY;
11006         if (new_ifindex && __dev_get_by_index(net, new_ifindex))
11007                 goto out;
11008
11009         /*
11010          * And now a mini version of register_netdevice unregister_netdevice.
11011          */
11012
11013         /* If device is running close it first. */
11014         dev_close(dev);
11015
11016         /* And unlink it from device chain */
11017         unlist_netdevice(dev, true);
11018
11019         synchronize_net();
11020
11021         /* Shutdown queueing discipline. */
11022         dev_shutdown(dev);
11023
11024         /* Notify protocols, that we are about to destroy
11025          * this device. They should clean all the things.
11026          *
11027          * Note that dev->reg_state stays at NETREG_REGISTERED.
11028          * This is wanted because this way 8021q and macvlan know
11029          * the device is just moving and can keep their slaves up.
11030          */
11031         call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
11032         rcu_barrier();
11033
11034         new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL);
11035         /* If there is an ifindex conflict assign a new one */
11036         if (!new_ifindex) {
11037                 if (__dev_get_by_index(net, dev->ifindex))
11038                         new_ifindex = dev_new_index(net);
11039                 else
11040                         new_ifindex = dev->ifindex;
11041         }
11042
11043         rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid,
11044                             new_ifindex);
11045
11046         /*
11047          *      Flush the unicast and multicast chains
11048          */
11049         dev_uc_flush(dev);
11050         dev_mc_flush(dev);
11051
11052         /* Send a netdev-removed uevent to the old namespace */
11053         kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
11054         netdev_adjacent_del_links(dev);
11055
11056         /* Move per-net netdevice notifiers that are following the netdevice */
11057         move_netdevice_notifiers_dev_net(dev, net);
11058
11059         /* Actually switch the network namespace */
11060         dev_net_set(dev, net);
11061         dev->ifindex = new_ifindex;
11062
11063         /* Send a netdev-add uevent to the new namespace */
11064         kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
11065         netdev_adjacent_add_links(dev);
11066
11067         /* Fixup kobjects */
11068         err = device_rename(&dev->dev, dev->name);
11069         WARN_ON(err);
11070
11071         /* Adapt owner in case owning user namespace of target network
11072          * namespace is different from the original one.
11073          */
11074         err = netdev_change_owner(dev, net_old, net);
11075         WARN_ON(err);
11076
11077         /* Add the device back in the hashes */
11078         list_netdevice(dev);
11079
11080         /* Notify protocols, that a new device appeared. */
11081         call_netdevice_notifiers(NETDEV_REGISTER, dev);
11082
11083         /*
11084          *      Prevent userspace races by waiting until the network
11085          *      device is fully setup before sending notifications.
11086          */
11087         rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL);
11088
11089         synchronize_net();
11090         err = 0;
11091 out:
11092         return err;
11093 }
11094 EXPORT_SYMBOL_GPL(__dev_change_net_namespace);
11095
11096 static int dev_cpu_dead(unsigned int oldcpu)
11097 {
11098         struct sk_buff **list_skb;
11099         struct sk_buff *skb;
11100         unsigned int cpu;
11101         struct softnet_data *sd, *oldsd, *remsd = NULL;
11102
11103         local_irq_disable();
11104         cpu = smp_processor_id();
11105         sd = &per_cpu(softnet_data, cpu);
11106         oldsd = &per_cpu(softnet_data, oldcpu);
11107
11108         /* Find end of our completion_queue. */
11109         list_skb = &sd->completion_queue;
11110         while (*list_skb)
11111                 list_skb = &(*list_skb)->next;
11112         /* Append completion queue from offline CPU. */
11113         *list_skb = oldsd->completion_queue;
11114         oldsd->completion_queue = NULL;
11115
11116         /* Append output queue from offline CPU. */
11117         if (oldsd->output_queue) {
11118                 *sd->output_queue_tailp = oldsd->output_queue;
11119                 sd->output_queue_tailp = oldsd->output_queue_tailp;
11120                 oldsd->output_queue = NULL;
11121                 oldsd->output_queue_tailp = &oldsd->output_queue;
11122         }
11123         /* Append NAPI poll list from offline CPU, with one exception :
11124          * process_backlog() must be called by cpu owning percpu backlog.
11125          * We properly handle process_queue & input_pkt_queue later.
11126          */
11127         while (!list_empty(&oldsd->poll_list)) {
11128                 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
11129                                                             struct napi_struct,
11130                                                             poll_list);
11131
11132                 list_del_init(&napi->poll_list);
11133                 if (napi->poll == process_backlog)
11134                         napi->state = 0;
11135                 else
11136                         ____napi_schedule(sd, napi);
11137         }
11138
11139         raise_softirq_irqoff(NET_TX_SOFTIRQ);
11140         local_irq_enable();
11141
11142 #ifdef CONFIG_RPS
11143         remsd = oldsd->rps_ipi_list;
11144         oldsd->rps_ipi_list = NULL;
11145 #endif
11146         /* send out pending IPI's on offline CPU */
11147         net_rps_send_ipi(remsd);
11148
11149         /* Process offline CPU's input_pkt_queue */
11150         while ((skb = __skb_dequeue(&oldsd->process_queue))) {
11151                 netif_rx(skb);
11152                 input_queue_head_incr(oldsd);
11153         }
11154         while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
11155                 netif_rx(skb);
11156                 input_queue_head_incr(oldsd);
11157         }
11158
11159         return 0;
11160 }
11161
11162 /**
11163  *      netdev_increment_features - increment feature set by one
11164  *      @all: current feature set
11165  *      @one: new feature set
11166  *      @mask: mask feature set
11167  *
11168  *      Computes a new feature set after adding a device with feature set
11169  *      @one to the master device with current feature set @all.  Will not
11170  *      enable anything that is off in @mask. Returns the new feature set.
11171  */
11172 netdev_features_t netdev_increment_features(netdev_features_t all,
11173         netdev_features_t one, netdev_features_t mask)
11174 {
11175         if (mask & NETIF_F_HW_CSUM)
11176                 mask |= NETIF_F_CSUM_MASK;
11177         mask |= NETIF_F_VLAN_CHALLENGED;
11178
11179         all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask;
11180         all &= one | ~NETIF_F_ALL_FOR_ALL;
11181
11182         /* If one device supports hw checksumming, set for all. */
11183         if (all & NETIF_F_HW_CSUM)
11184                 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM);
11185
11186         return all;
11187 }
11188 EXPORT_SYMBOL(netdev_increment_features);
11189
11190 static struct hlist_head * __net_init netdev_create_hash(void)
11191 {
11192         int i;
11193         struct hlist_head *hash;
11194
11195         hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL);
11196         if (hash != NULL)
11197                 for (i = 0; i < NETDEV_HASHENTRIES; i++)
11198                         INIT_HLIST_HEAD(&hash[i]);
11199
11200         return hash;
11201 }
11202
11203 /* Initialize per network namespace state */
11204 static int __net_init netdev_init(struct net *net)
11205 {
11206         BUILD_BUG_ON(GRO_HASH_BUCKETS >
11207                      8 * sizeof_field(struct napi_struct, gro_bitmask));
11208
11209         INIT_LIST_HEAD(&net->dev_base_head);
11210
11211         net->dev_name_head = netdev_create_hash();
11212         if (net->dev_name_head == NULL)
11213                 goto err_name;
11214
11215         net->dev_index_head = netdev_create_hash();
11216         if (net->dev_index_head == NULL)
11217                 goto err_idx;
11218
11219         RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain);
11220
11221         return 0;
11222
11223 err_idx:
11224         kfree(net->dev_name_head);
11225 err_name:
11226         return -ENOMEM;
11227 }
11228
11229 /**
11230  *      netdev_drivername - network driver for the device
11231  *      @dev: network device
11232  *
11233  *      Determine network driver for device.
11234  */
11235 const char *netdev_drivername(const struct net_device *dev)
11236 {
11237         const struct device_driver *driver;
11238         const struct device *parent;
11239         const char *empty = "";
11240
11241         parent = dev->dev.parent;
11242         if (!parent)
11243                 return empty;
11244
11245         driver = parent->driver;
11246         if (driver && driver->name)
11247                 return driver->name;
11248         return empty;
11249 }
11250
11251 static void __netdev_printk(const char *level, const struct net_device *dev,
11252                             struct va_format *vaf)
11253 {
11254         if (dev && dev->dev.parent) {
11255                 dev_printk_emit(level[1] - '0',
11256                                 dev->dev.parent,
11257                                 "%s %s %s%s: %pV",
11258                                 dev_driver_string(dev->dev.parent),
11259                                 dev_name(dev->dev.parent),
11260                                 netdev_name(dev), netdev_reg_state(dev),
11261                                 vaf);
11262         } else if (dev) {
11263                 printk("%s%s%s: %pV",
11264                        level, netdev_name(dev), netdev_reg_state(dev), vaf);
11265         } else {
11266                 printk("%s(NULL net_device): %pV", level, vaf);
11267         }
11268 }
11269
11270 void netdev_printk(const char *level, const struct net_device *dev,
11271                    const char *format, ...)
11272 {
11273         struct va_format vaf;
11274         va_list args;
11275
11276         va_start(args, format);
11277
11278         vaf.fmt = format;
11279         vaf.va = &args;
11280
11281         __netdev_printk(level, dev, &vaf);
11282
11283         va_end(args);
11284 }
11285 EXPORT_SYMBOL(netdev_printk);
11286
11287 #define define_netdev_printk_level(func, level)                 \
11288 void func(const struct net_device *dev, const char *fmt, ...)   \
11289 {                                                               \
11290         struct va_format vaf;                                   \
11291         va_list args;                                           \
11292                                                                 \
11293         va_start(args, fmt);                                    \
11294                                                                 \
11295         vaf.fmt = fmt;                                          \
11296         vaf.va = &args;                                         \
11297                                                                 \
11298         __netdev_printk(level, dev, &vaf);                      \
11299                                                                 \
11300         va_end(args);                                           \
11301 }                                                               \
11302 EXPORT_SYMBOL(func);
11303
11304 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
11305 define_netdev_printk_level(netdev_alert, KERN_ALERT);
11306 define_netdev_printk_level(netdev_crit, KERN_CRIT);
11307 define_netdev_printk_level(netdev_err, KERN_ERR);
11308 define_netdev_printk_level(netdev_warn, KERN_WARNING);
11309 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
11310 define_netdev_printk_level(netdev_info, KERN_INFO);
11311
11312 static void __net_exit netdev_exit(struct net *net)
11313 {
11314         kfree(net->dev_name_head);
11315         kfree(net->dev_index_head);
11316         if (net != &init_net)
11317                 WARN_ON_ONCE(!list_empty(&net->dev_base_head));
11318 }
11319
11320 static struct pernet_operations __net_initdata netdev_net_ops = {
11321         .init = netdev_init,
11322         .exit = netdev_exit,
11323 };
11324
11325 static void __net_exit default_device_exit_net(struct net *net)
11326 {
11327         struct net_device *dev, *aux;
11328         /*
11329          * Push all migratable network devices back to the
11330          * initial network namespace
11331          */
11332         ASSERT_RTNL();
11333         for_each_netdev_safe(net, dev, aux) {
11334                 int err;
11335                 char fb_name[IFNAMSIZ];
11336
11337                 /* Ignore unmoveable devices (i.e. loopback) */
11338                 if (dev->features & NETIF_F_NETNS_LOCAL)
11339                         continue;
11340
11341                 /* Leave virtual devices for the generic cleanup */
11342                 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund)
11343                         continue;
11344
11345                 /* Push remaining network devices to init_net */
11346                 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
11347                 if (netdev_name_in_use(&init_net, fb_name))
11348                         snprintf(fb_name, IFNAMSIZ, "dev%%d");
11349                 err = dev_change_net_namespace(dev, &init_net, fb_name);
11350                 if (err) {
11351                         pr_emerg("%s: failed to move %s to init_net: %d\n",
11352                                  __func__, dev->name, err);
11353                         BUG();
11354                 }
11355         }
11356 }
11357
11358 static void __net_exit default_device_exit_batch(struct list_head *net_list)
11359 {
11360         /* At exit all network devices most be removed from a network
11361          * namespace.  Do this in the reverse order of registration.
11362          * Do this across as many network namespaces as possible to
11363          * improve batching efficiency.
11364          */
11365         struct net_device *dev;
11366         struct net *net;
11367         LIST_HEAD(dev_kill_list);
11368
11369         rtnl_lock();
11370         list_for_each_entry(net, net_list, exit_list) {
11371                 default_device_exit_net(net);
11372                 cond_resched();
11373         }
11374
11375         list_for_each_entry(net, net_list, exit_list) {
11376                 for_each_netdev_reverse(net, dev) {
11377                         if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
11378                                 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
11379                         else
11380                                 unregister_netdevice_queue(dev, &dev_kill_list);
11381                 }
11382         }
11383         unregister_netdevice_many(&dev_kill_list);
11384         rtnl_unlock();
11385 }
11386
11387 static struct pernet_operations __net_initdata default_device_ops = {
11388         .exit_batch = default_device_exit_batch,
11389 };
11390
11391 /*
11392  *      Initialize the DEV module. At boot time this walks the device list and
11393  *      unhooks any devices that fail to initialise (normally hardware not
11394  *      present) and leaves us with a valid list of present and active devices.
11395  *
11396  */
11397
11398 /*
11399  *       This is called single threaded during boot, so no need
11400  *       to take the rtnl semaphore.
11401  */
11402 static int __init net_dev_init(void)
11403 {
11404         int i, rc = -ENOMEM;
11405
11406         BUG_ON(!dev_boot_phase);
11407
11408         if (dev_proc_init())
11409                 goto out;
11410
11411         if (netdev_kobject_init())
11412                 goto out;
11413
11414         INIT_LIST_HEAD(&ptype_all);
11415         for (i = 0; i < PTYPE_HASH_SIZE; i++)
11416                 INIT_LIST_HEAD(&ptype_base[i]);
11417
11418         if (register_pernet_subsys(&netdev_net_ops))
11419                 goto out;
11420
11421         /*
11422          *      Initialise the packet receive queues.
11423          */
11424
11425         for_each_possible_cpu(i) {
11426                 struct work_struct *flush = per_cpu_ptr(&flush_works, i);
11427                 struct softnet_data *sd = &per_cpu(softnet_data, i);
11428
11429                 INIT_WORK(flush, flush_backlog);
11430
11431                 skb_queue_head_init(&sd->input_pkt_queue);
11432                 skb_queue_head_init(&sd->process_queue);
11433 #ifdef CONFIG_XFRM_OFFLOAD
11434                 skb_queue_head_init(&sd->xfrm_backlog);
11435 #endif
11436                 INIT_LIST_HEAD(&sd->poll_list);
11437                 sd->output_queue_tailp = &sd->output_queue;
11438 #ifdef CONFIG_RPS
11439                 INIT_CSD(&sd->csd, rps_trigger_softirq, sd);
11440                 sd->cpu = i;
11441 #endif
11442                 INIT_CSD(&sd->defer_csd, trigger_rx_softirq, sd);
11443                 spin_lock_init(&sd->defer_lock);
11444
11445                 init_gro_hash(&sd->backlog);
11446                 sd->backlog.poll = process_backlog;
11447                 sd->backlog.weight = weight_p;
11448         }
11449
11450         dev_boot_phase = 0;
11451
11452         /* The loopback device is special if any other network devices
11453          * is present in a network namespace the loopback device must
11454          * be present. Since we now dynamically allocate and free the
11455          * loopback device ensure this invariant is maintained by
11456          * keeping the loopback device as the first device on the
11457          * list of network devices.  Ensuring the loopback devices
11458          * is the first device that appears and the last network device
11459          * that disappears.
11460          */
11461         if (register_pernet_device(&loopback_net_ops))
11462                 goto out;
11463
11464         if (register_pernet_device(&default_device_ops))
11465                 goto out;
11466
11467         open_softirq(NET_TX_SOFTIRQ, net_tx_action);
11468         open_softirq(NET_RX_SOFTIRQ, net_rx_action);
11469
11470         rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead",
11471                                        NULL, dev_cpu_dead);
11472         WARN_ON(rc < 0);
11473         rc = 0;
11474 out:
11475         return rc;
11476 }
11477
11478 subsys_initcall(net_dev_init);