net/core/neighbour.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  *      Generic address resolution entity
   4  *
   5  *      Authors:
   6  *      Pedro Roque             <roque@di.fc.ul.pt>
   7  *      Alexey Kuznetsov        <kuznet@ms2.inr.ac.ru>
   8  *
   9  *      Fixes:
  10  *      Vitaly E. Lavrov        releasing NULL neighbor in neigh_add.
  11  *      Harald Welte            Add neighbour cache statistics like rtstat
  12  */
  13
  14 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  15
  16 #include <linux/slab.h>
  17 #include <linux/kmemleak.h>
  18 #include <linux/types.h>
  19 #include <linux/kernel.h>
  20 #include <linux/module.h>
  21 #include <linux/socket.h>
  22 #include <linux/netdevice.h>
  23 #include <linux/proc_fs.h>
  24 #ifdef CONFIG_SYSCTL
  25 #include <linux/sysctl.h>
  26 #endif
  27 #include <linux/times.h>
  28 #include <net/net_namespace.h>
  29 #include <net/neighbour.h>
  30 #include <net/arp.h>
  31 #include <net/dst.h>
  32 #include <net/sock.h>
  33 #include <net/netevent.h>
  34 #include <net/netlink.h>
  35 #include <linux/rtnetlink.h>
  36 #include <linux/random.h>
  37 #include <linux/string.h>
  38 #include <linux/log2.h>
  39 #include <linux/inetdevice.h>
  40 #include <net/addrconf.h>
  41
  42 #include <trace/events/neigh.h>
  43
  44 #define DEBUG
  45 #define NEIGH_DEBUG 1
  46 #define neigh_dbg(level, fmt, ...)              \
  47 do {                                            \
  48         if (level <= NEIGH_DEBUG)               \
  49                 pr_debug(fmt, ##__VA_ARGS__);   \
  50 } while (0)
  51
  52 #define PNEIGH_HASHMASK         0xF
  53
  54 static void neigh_timer_handler(struct timer_list *t);
  55 static void __neigh_notify(struct neighbour *n, int type, int flags,
  56                            u32 pid);
  57 static void neigh_update_notify(struct neighbour *neigh, u32 nlmsg_pid);
  58 static int pneigh_ifdown_and_unlock(struct neigh_table *tbl,
  59                                     struct net_device *dev);
  60
  61 #ifdef CONFIG_PROC_FS
  62 static const struct seq_operations neigh_stat_seq_ops;
  63 #endif
  64
  65 /*
  66    Neighbour hash table buckets are protected with rwlock tbl->lock.
  67
  68    - All the scans/updates to hash buckets MUST be made under this lock.
  69    - NOTHING clever should be made under this lock: no callbacks
  70      to protocol backends, no attempts to send something to network.
  71      It will result in deadlocks, if backend/driver wants to use neighbour
  72      cache.
  73    - If the entry requires some non-trivial actions, increase
  74      its reference count and release table lock.
  75
  76    Neighbour entries are protected:
  77    - with reference count.
  78    - with rwlock neigh->lock
  79
  80    Reference count prevents destruction.
  81
  82    neigh->lock mainly serializes ll address data and its validity state.
  83    However, the same lock is used to protect another entry fields:
  84     - timer
  85     - resolution queue
  86
  87    Again, nothing clever shall be made under neigh->lock,
  88    the most complicated procedure, which we allow is dev->hard_header.
  89    It is supposed, that dev->hard_header is simplistic and does
  90    not make callbacks to neighbour tables.
  91  */
  92
  93 static int neigh_blackhole(struct neighbour *neigh, struct sk_buff *skb)
  94 {
  95         kfree_skb(skb);
  96         return -ENETDOWN;
  97 }
  98
  99 static void neigh_cleanup_and_release(struct neighbour *neigh)
 100 {
 101         trace_neigh_cleanup_and_release(neigh, 0);
 102         __neigh_notify(neigh, RTM_DELNEIGH, 0, 0);
 103         call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, neigh);
 104         neigh_release(neigh);
 105 }
 106
 107 /*
 108  * It is random distribution in the interval (1/2)*base...(3/2)*base.
 109  * It corresponds to default IPv6 settings and is not overridable,
 110  * because it is really reasonable choice.
 111  */
 112
 113 unsigned long neigh_rand_reach_time(unsigned long base)
 114 {
 115         return base ? (prandom_u32() % base) + (base >> 1) : 0;
 116 }
 117 EXPORT_SYMBOL(neigh_rand_reach_time);
 118
 119 static void neigh_mark_dead(struct neighbour *n)
 120 {
 121         n->dead = 1;
 122         if (!list_empty(&n->gc_list)) {
 123                 list_del_init(&n->gc_list);
 124                 atomic_dec(&n->tbl->gc_entries);
 125         }
 126 }
 127
 128 static void neigh_update_gc_list(struct neighbour *n)
 129 {
 130         bool on_gc_list, exempt_from_gc;
 131
 132         write_lock_bh(&n->tbl->lock);
 133         write_lock(&n->lock);
 134
 135         /* remove from the gc list if new state is permanent or if neighbor
 136          * is externally learned; otherwise entry should be on the gc list
 137          */
 138         exempt_from_gc = n->nud_state & NUD_PERMANENT ||
 139                          n->flags & NTF_EXT_LEARNED;
 140         on_gc_list = !list_empty(&n->gc_list);
 141
 142         if (exempt_from_gc && on_gc_list) {
 143                 list_del_init(&n->gc_list);
 144                 atomic_dec(&n->tbl->gc_entries);
 145         } else if (!exempt_from_gc && !on_gc_list) {
 146                 /* add entries to the tail; cleaning removes from the front */
 147                 list_add_tail(&n->gc_list, &n->tbl->gc_list);
 148                 atomic_inc(&n->tbl->gc_entries);
 149         }
 150
 151         write_unlock(&n->lock);
 152         write_unlock_bh(&n->tbl->lock);
 153 }
 154
 155 static bool neigh_update_ext_learned(struct neighbour *neigh, u32 flags,
 156                                      int *notify)
 157 {
 158         bool rc = false;
 159         u8 ndm_flags;
 160
 161         if (!(flags & NEIGH_UPDATE_F_ADMIN))
 162                 return rc;
 163
 164         ndm_flags = (flags & NEIGH_UPDATE_F_EXT_LEARNED) ? NTF_EXT_LEARNED : 0;
 165         if ((neigh->flags ^ ndm_flags) & NTF_EXT_LEARNED) {
 166                 if (ndm_flags & NTF_EXT_LEARNED)
 167                         neigh->flags |= NTF_EXT_LEARNED;
 168                 else
 169                         neigh->flags &= ~NTF_EXT_LEARNED;
 170                 rc = true;
 171                 *notify = 1;
 172         }
 173
 174         return rc;
 175 }
 176
 177 static bool neigh_del(struct neighbour *n, struct neighbour __rcu **np,
 178                       struct neigh_table *tbl)
 179 {
 180         bool retval = false;
 181
 182         write_lock(&n->lock);
 183         if (refcount_read(&n->refcnt) == 1) {
 184                 struct neighbour *neigh;
 185
 186                 neigh = rcu_dereference_protected(n->next,
 187                                                   lockdep_is_held(&tbl->lock));
 188                 rcu_assign_pointer(*np, neigh);
 189                 neigh_mark_dead(n);
 190                 retval = true;
 191         }
 192         write_unlock(&n->lock);
 193         if (retval)
 194                 neigh_cleanup_and_release(n);
 195         return retval;
 196 }
 197
 198 bool neigh_remove_one(struct neighbour *ndel, struct neigh_table *tbl)
 199 {
 200         struct neigh_hash_table *nht;
 201         void *pkey = ndel->primary_key;
 202         u32 hash_val;
 203         struct neighbour *n;
 204         struct neighbour __rcu **np;
 205
 206         nht = rcu_dereference_protected(tbl->nht,
 207                                         lockdep_is_held(&tbl->lock));
 208         hash_val = tbl->hash(pkey, ndel->dev, nht->hash_rnd);
 209         hash_val = hash_val >> (32 - nht->hash_shift);
 210
 211         np = &nht->hash_buckets[hash_val];
 212         while ((n = rcu_dereference_protected(*np,
 213                                               lockdep_is_held(&tbl->lock)))) {
 214                 if (n == ndel)
 215                         return neigh_del(n, np, tbl);
 216                 np = &n->next;
 217         }
 218         return false;
 219 }
 220
 221 static int neigh_forced_gc(struct neigh_table *tbl)
 222 {
 223         int max_clean = atomic_read(&tbl->gc_entries) - tbl->gc_thresh2;
 224         unsigned long tref = jiffies - 5 * HZ;
 225         struct neighbour *n, *tmp;
 226         int shrunk = 0;
 227
 228         NEIGH_CACHE_STAT_INC(tbl, forced_gc_runs);
 229
 230         write_lock_bh(&tbl->lock);
 231
 232         list_for_each_entry_safe(n, tmp, &tbl->gc_list, gc_list) {
 233                 if (refcount_read(&n->refcnt) == 1) {
 234                         bool remove = false;
 235
 236                         write_lock(&n->lock);
 237                         if ((n->nud_state == NUD_FAILED) ||
 238                             (tbl->is_multicast &&
 239                              tbl->is_multicast(n->primary_key)) ||
 240                             time_after(tref, n->updated))
 241                                 remove = true;
 242                         write_unlock(&n->lock);
 243
 244                         if (remove && neigh_remove_one(n, tbl))
 245                                 shrunk++;
 246                         if (shrunk >= max_clean)
 247                                 break;
 248                 }
 249         }
 250
 251         tbl->last_flush = jiffies;
 252
 253         write_unlock_bh(&tbl->lock);
 254
 255         return shrunk;
 256 }
 257
 258 static void neigh_add_timer(struct neighbour *n, unsigned long when)
 259 {
 260         neigh_hold(n);
 261         if (unlikely(mod_timer(&n->timer, when))) {
 262                 printk("NEIGH: BUG, double timer add, state is %x\n",
 263                        n->nud_state);
 264                 dump_stack();
 265         }
 266 }
 267
 268 static int neigh_del_timer(struct neighbour *n)
 269 {
 270         if ((n->nud_state & NUD_IN_TIMER) &&
 271             del_timer(&n->timer)) {
 272                 neigh_release(n);
 273                 return 1;
 274         }
 275         return 0;
 276 }
 277
 278 static void pneigh_queue_purge(struct sk_buff_head *list)
 279 {
 280         struct sk_buff *skb;
 281
 282         while ((skb = skb_dequeue(list)) != NULL) {
 283                 dev_put(skb->dev);
 284                 kfree_skb(skb);
 285         }
 286 }
 287
 288 static void neigh_flush_dev(struct neigh_table *tbl, struct net_device *dev,
 289                             bool skip_perm)
 290 {
 291         int i;
 292         struct neigh_hash_table *nht;
 293
 294         nht = rcu_dereference_protected(tbl->nht,
 295                                         lockdep_is_held(&tbl->lock));
 296
 297         for (i = 0; i < (1 << nht->hash_shift); i++) {
 298                 struct neighbour *n;
 299                 struct neighbour __rcu **np = &nht->hash_buckets[i];
 300
 301                 while ((n = rcu_dereference_protected(*np,
 302                                         lockdep_is_held(&tbl->lock))) != NULL) {
 303                         if (dev && n->dev != dev) {
 304                                 np = &n->next;
 305                                 continue;
 306                         }
 307                         if (skip_perm && n->nud_state & NUD_PERMANENT) {
 308                                 np = &n->next;
 309                                 continue;
 310                         }
 311                         rcu_assign_pointer(*np,
 312                                    rcu_dereference_protected(n->next,
 313                                                 lockdep_is_held(&tbl->lock)));
 314                         write_lock(&n->lock);
 315                         neigh_del_timer(n);
 316                         neigh_mark_dead(n);
 317                         if (refcount_read(&n->refcnt) != 1) {
 318                                 /* The most unpleasant situation.
 319                                    We must destroy neighbour entry,
 320                                    but someone still uses it.
 321
 322                                    The destroy will be delayed until
 323                                    the last user releases us, but
 324                                    we must kill timers etc. and move
 325                                    it to safe state.
 326                                  */
 327                                 __skb_queue_purge(&n->arp_queue);
 328                                 n->arp_queue_len_bytes = 0;
 329                                 n->output = neigh_blackhole;
 330                                 if (n->nud_state & NUD_VALID)
 331                                         n->nud_state = NUD_NOARP;
 332                                 else
 333                                         n->nud_state = NUD_NONE;
 334                                 neigh_dbg(2, "neigh %p is stray\n", n);
 335                         }
 336                         write_unlock(&n->lock);
 337                         neigh_cleanup_and_release(n);
 338                 }
 339         }
 340 }
 341
 342 void neigh_changeaddr(struct neigh_table *tbl, struct net_device *dev)
 343 {
 344         write_lock_bh(&tbl->lock);
 345         neigh_flush_dev(tbl, dev, false);
 346         write_unlock_bh(&tbl->lock);
 347 }
 348 EXPORT_SYMBOL(neigh_changeaddr);
 349
 350 static int __neigh_ifdown(struct neigh_table *tbl, struct net_device *dev,
 351                           bool skip_perm)
 352 {
 353         write_lock_bh(&tbl->lock);
 354         neigh_flush_dev(tbl, dev, skip_perm);
 355         pneigh_ifdown_and_unlock(tbl, dev);
 356
 357         del_timer_sync(&tbl->proxy_timer);
 358         pneigh_queue_purge(&tbl->proxy_queue);
 359         return 0;
 360 }
 361
 362 int neigh_carrier_down(struct neigh_table *tbl, struct net_device *dev)
 363 {
 364         __neigh_ifdown(tbl, dev, true);
 365         return 0;
 366 }
 367 EXPORT_SYMBOL(neigh_carrier_down);
 368
 369 int neigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
 370 {
 371         __neigh_ifdown(tbl, dev, false);
 372         return 0;
 373 }
 374 EXPORT_SYMBOL(neigh_ifdown);
 375
 376 static struct neighbour *neigh_alloc(struct neigh_table *tbl,
 377                                      struct net_device *dev,
 378                                      bool exempt_from_gc)
 379 {
 380         struct neighbour *n = NULL;
 381         unsigned long now = jiffies;
 382         int entries;
 383
 384         if (exempt_from_gc)
 385                 goto do_alloc;
 386
 387         entries = atomic_inc_return(&tbl->gc_entries) - 1;
 388         if (entries >= tbl->gc_thresh3 ||
 389             (entries >= tbl->gc_thresh2 &&
 390              time_after(now, tbl->last_flush + 5 * HZ))) {
 391                 if (!neigh_forced_gc(tbl) &&
 392                     entries >= tbl->gc_thresh3) {
 393                         net_info_ratelimited("%s: neighbor table overflow!\n",
 394                                              tbl->id);
 395                         NEIGH_CACHE_STAT_INC(tbl, table_fulls);
 396                         goto out_entries;
 397                 }
 398         }
 399
 400 do_alloc:
 401         n = kzalloc(tbl->entry_size + dev->neigh_priv_len, GFP_ATOMIC);
 402         if (!n)
 403                 goto out_entries;
 404
 405         __skb_queue_head_init(&n->arp_queue);
 406         rwlock_init(&n->lock);
 407         seqlock_init(&n->ha_lock);
 408         n->updated        = n->used = now;
 409         n->nud_state      = NUD_NONE;
 410         n->output         = neigh_blackhole;
 411         seqlock_init(&n->hh.hh_lock);
 412         n->parms          = neigh_parms_clone(&tbl->parms);
 413         timer_setup(&n->timer, neigh_timer_handler, 0);
 414
 415         NEIGH_CACHE_STAT_INC(tbl, allocs);
 416         n->tbl            = tbl;
 417         refcount_set(&n->refcnt, 1);
 418         n->dead           = 1;
 419         INIT_LIST_HEAD(&n->gc_list);
 420
 421         atomic_inc(&tbl->entries);
 422 out:
 423         return n;
 424
 425 out_entries:
 426         if (!exempt_from_gc)
 427                 atomic_dec(&tbl->gc_entries);
 428         goto out;
 429 }
 430
 431 static void neigh_get_hash_rnd(u32 *x)
 432 {
 433         *x = get_random_u32() | 1;
 434 }
 435
 436 static struct neigh_hash_table *neigh_hash_alloc(unsigned int shift)
 437 {
 438         size_t size = (1 << shift) * sizeof(struct neighbour *);
 439         struct neigh_hash_table *ret;
 440         struct neighbour __rcu **buckets;
 441         int i;
 442
 443         ret = kmalloc(sizeof(*ret), GFP_ATOMIC);
 444         if (!ret)
 445                 return NULL;
 446         if (size <= PAGE_SIZE) {
 447                 buckets = kzalloc(size, GFP_ATOMIC);
 448         } else {
 449                 buckets = (struct neighbour __rcu **)
 450                           __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
 451                                            get_order(size));
 452                 kmemleak_alloc(buckets, size, 1, GFP_ATOMIC);
 453         }
 454         if (!buckets) {
 455                 kfree(ret);
 456                 return NULL;
 457         }
 458         ret->hash_buckets = buckets;
 459         ret->hash_shift = shift;
 460         for (i = 0; i < NEIGH_NUM_HASH_RND; i++)
 461                 neigh_get_hash_rnd(&ret->hash_rnd[i]);
 462         return ret;
 463 }
 464
 465 static void neigh_hash_free_rcu(struct rcu_head *head)
 466 {
 467         struct neigh_hash_table *nht = container_of(head,
 468                                                     struct neigh_hash_table,
 469                                                     rcu);
 470         size_t size = (1 << nht->hash_shift) * sizeof(struct neighbour *);
 471         struct neighbour __rcu **buckets = nht->hash_buckets;
 472
 473         if (size <= PAGE_SIZE) {
 474                 kfree(buckets);
 475         } else {
 476                 kmemleak_free(buckets);
 477                 free_pages((unsigned long)buckets, get_order(size));
 478         }
 479         kfree(nht);
 480 }
 481
 482 static struct neigh_hash_table *neigh_hash_grow(struct neigh_table *tbl,
 483                                                 unsigned long new_shift)
 484 {
 485         unsigned int i, hash;
 486         struct neigh_hash_table *new_nht, *old_nht;
 487
 488         NEIGH_CACHE_STAT_INC(tbl, hash_grows);
 489
 490         old_nht = rcu_dereference_protected(tbl->nht,
 491                                             lockdep_is_held(&tbl->lock));
 492         new_nht = neigh_hash_alloc(new_shift);
 493         if (!new_nht)
 494                 return old_nht;
 495
 496         for (i = 0; i < (1 << old_nht->hash_shift); i++) {
 497                 struct neighbour *n, *next;
 498
 499                 for (n = rcu_dereference_protected(old_nht->hash_buckets[i],
 500                                                    lockdep_is_held(&tbl->lock));
 501                      n != NULL;
 502                      n = next) {
 503                         hash = tbl->hash(n->primary_key, n->dev,
 504                                          new_nht->hash_rnd);
 505
 506                         hash >>= (32 - new_nht->hash_shift);
 507                         next = rcu_dereference_protected(n->next,
 508                                                 lockdep_is_held(&tbl->lock));
 509
 510                         rcu_assign_pointer(n->next,
 511                                            rcu_dereference_protected(
 512                                                 new_nht->hash_buckets[hash],
 513                                                 lockdep_is_held(&tbl->lock)));
 514                         rcu_assign_pointer(new_nht->hash_buckets[hash], n);
 515                 }
 516         }
 517
 518         rcu_assign_pointer(tbl->nht, new_nht);
 519         call_rcu(&old_nht->rcu, neigh_hash_free_rcu);
 520         return new_nht;
 521 }
 522
 523 struct neighbour *neigh_lookup(struct neigh_table *tbl, const void *pkey,
 524                                struct net_device *dev)
 525 {
 526         struct neighbour *n;
 527
 528         NEIGH_CACHE_STAT_INC(tbl, lookups);
 529
 530         rcu_read_lock_bh();
 531         n = __neigh_lookup_noref(tbl, pkey, dev);
 532         if (n) {
 533                 if (!refcount_inc_not_zero(&n->refcnt))
 534                         n = NULL;
 535                 NEIGH_CACHE_STAT_INC(tbl, hits);
 536         }
 537
 538         rcu_read_unlock_bh();
 539         return n;
 540 }
 541 EXPORT_SYMBOL(neigh_lookup);
 542
 543 struct neighbour *neigh_lookup_nodev(struct neigh_table *tbl, struct net *net,
 544                                      const void *pkey)
 545 {
 546         struct neighbour *n;
 547         unsigned int key_len = tbl->key_len;
 548         u32 hash_val;
 549         struct neigh_hash_table *nht;
 550
 551         NEIGH_CACHE_STAT_INC(tbl, lookups);
 552
 553         rcu_read_lock_bh();
 554         nht = rcu_dereference_bh(tbl->nht);
 555         hash_val = tbl->hash(pkey, NULL, nht->hash_rnd) >> (32 - nht->hash_shift);
 556
 557         for (n = rcu_dereference_bh(nht->hash_buckets[hash_val]);
 558              n != NULL;
 559              n = rcu_dereference_bh(n->next)) {
 560                 if (!memcmp(n->primary_key, pkey, key_len) &&
 561                     net_eq(dev_net(n->dev), net)) {
 562                         if (!refcount_inc_not_zero(&n->refcnt))
 563                                 n = NULL;
 564                         NEIGH_CACHE_STAT_INC(tbl, hits);
 565                         break;
 566                 }
 567         }
 568
 569         rcu_read_unlock_bh();
 570         return n;
 571 }
 572 EXPORT_SYMBOL(neigh_lookup_nodev);
 573
 574 static struct neighbour *___neigh_create(struct neigh_table *tbl,
 575                                          const void *pkey,
 576                                          struct net_device *dev,
 577                                          bool exempt_from_gc, bool want_ref)
 578 {
 579         struct neighbour *n1, *rc, *n = neigh_alloc(tbl, dev, exempt_from_gc);
 580         u32 hash_val;
 581         unsigned int key_len = tbl->key_len;
 582         int error;
 583         struct neigh_hash_table *nht;
 584
 585         trace_neigh_create(tbl, dev, pkey, n, exempt_from_gc);
 586
 587         if (!n) {
 588                 rc = ERR_PTR(-ENOBUFS);
 589                 goto out;
 590         }
 591
 592         memcpy(n->primary_key, pkey, key_len);
 593         n->dev = dev;
 594         dev_hold(dev);
 595
 596         /* Protocol specific setup. */
 597         if (tbl->constructor && (error = tbl->constructor(n)) < 0) {
 598                 rc = ERR_PTR(error);
 599                 goto out_neigh_release;
 600         }
 601
 602         if (dev->netdev_ops->ndo_neigh_construct) {
 603                 error = dev->netdev_ops->ndo_neigh_construct(dev, n);
 604                 if (error < 0) {
 605                         rc = ERR_PTR(error);
 606                         goto out_neigh_release;
 607                 }
 608         }
 609
 610         /* Device specific setup. */
 611         if (n->parms->neigh_setup &&
 612             (error = n->parms->neigh_setup(n)) < 0) {
 613                 rc = ERR_PTR(error);
 614                 goto out_neigh_release;
 615         }
 616
 617         n->confirmed = jiffies - (NEIGH_VAR(n->parms, BASE_REACHABLE_TIME) << 1);
 618
 619         write_lock_bh(&tbl->lock);
 620         nht = rcu_dereference_protected(tbl->nht,
 621                                         lockdep_is_held(&tbl->lock));
 622
 623         if (atomic_read(&tbl->entries) > (1 << nht->hash_shift))
 624                 nht = neigh_hash_grow(tbl, nht->hash_shift + 1);
 625
 626         hash_val = tbl->hash(n->primary_key, dev, nht->hash_rnd) >> (32 - nht->hash_shift);
 627
 628         if (n->parms->dead) {
 629                 rc = ERR_PTR(-EINVAL);
 630                 goto out_tbl_unlock;
 631         }
 632
 633         for (n1 = rcu_dereference_protected(nht->hash_buckets[hash_val],
 634                                             lockdep_is_held(&tbl->lock));
 635              n1 != NULL;
 636              n1 = rcu_dereference_protected(n1->next,
 637                         lockdep_is_held(&tbl->lock))) {
 638                 if (dev == n1->dev && !memcmp(n1->primary_key, n->primary_key, key_len)) {
 639                         if (want_ref)
 640                                 neigh_hold(n1);
 641                         rc = n1;
 642                         goto out_tbl_unlock;
 643                 }
 644         }
 645
 646         n->dead = 0;
 647         if (!exempt_from_gc)
 648                 list_add_tail(&n->gc_list, &n->tbl->gc_list);
 649
 650         if (want_ref)
 651                 neigh_hold(n);
 652         rcu_assign_pointer(n->next,
 653                            rcu_dereference_protected(nht->hash_buckets[hash_val],
 654                                                      lockdep_is_held(&tbl->lock)));
 655         rcu_assign_pointer(nht->hash_buckets[hash_val], n);
 656         write_unlock_bh(&tbl->lock);
 657         neigh_dbg(2, "neigh %p is created\n", n);
 658         rc = n;
 659 out:
 660         return rc;
 661 out_tbl_unlock:
 662         write_unlock_bh(&tbl->lock);
 663 out_neigh_release:
 664         if (!exempt_from_gc)
 665                 atomic_dec(&tbl->gc_entries);
 666         neigh_release(n);
 667         goto out;
 668 }
 669
 670 struct neighbour *__neigh_create(struct neigh_table *tbl, const void *pkey,
 671                                  struct net_device *dev, bool want_ref)
 672 {
 673         return ___neigh_create(tbl, pkey, dev, false, want_ref);
 674 }
 675 EXPORT_SYMBOL(__neigh_create);
 676
 677 static u32 pneigh_hash(const void *pkey, unsigned int key_len)
 678 {
 679         u32 hash_val = *(u32 *)(pkey + key_len - 4);
 680         hash_val ^= (hash_val >> 16);
 681         hash_val ^= hash_val >> 8;
 682         hash_val ^= hash_val >> 4;
 683         hash_val &= PNEIGH_HASHMASK;
 684         return hash_val;
 685 }
 686
 687 static struct pneigh_entry *__pneigh_lookup_1(struct pneigh_entry *n,
 688                                               struct net *net,
 689                                               const void *pkey,
 690                                               unsigned int key_len,
 691                                               struct net_device *dev)
 692 {
 693         while (n) {
 694                 if (!memcmp(n->key, pkey, key_len) &&
 695                     net_eq(pneigh_net(n), net) &&
 696                     (n->dev == dev || !n->dev))
 697                         return n;
 698                 n = n->next;
 699         }
 700         return NULL;
 701 }
 702
 703 struct pneigh_entry *__pneigh_lookup(struct neigh_table *tbl,
 704                 struct net *net, const void *pkey, struct net_device *dev)
 705 {
 706         unsigned int key_len = tbl->key_len;
 707         u32 hash_val = pneigh_hash(pkey, key_len);
 708
 709         return __pneigh_lookup_1(tbl->phash_buckets[hash_val],
 710                                  net, pkey, key_len, dev);
 711 }
 712 EXPORT_SYMBOL_GPL(__pneigh_lookup);
 713
 714 struct pneigh_entry * pneigh_lookup(struct neigh_table *tbl,
 715                                     struct net *net, const void *pkey,
 716                                     struct net_device *dev, int creat)
 717 {
 718         struct pneigh_entry *n;
 719         unsigned int key_len = tbl->key_len;
 720         u32 hash_val = pneigh_hash(pkey, key_len);
 721
 722         read_lock_bh(&tbl->lock);
 723         n = __pneigh_lookup_1(tbl->phash_buckets[hash_val],
 724                               net, pkey, key_len, dev);
 725         read_unlock_bh(&tbl->lock);
 726
 727         if (n || !creat)
 728                 goto out;
 729
 730         ASSERT_RTNL();
 731
 732         n = kmalloc(sizeof(*n) + key_len, GFP_KERNEL);
 733         if (!n)
 734                 goto out;
 735
 736         n->protocol = 0;
 737         write_pnet(&n->net, net);
 738         memcpy(n->key, pkey, key_len);
 739         n->dev = dev;
 740         if (dev)
 741                 dev_hold(dev);
 742
 743         if (tbl->pconstructor && tbl->pconstructor(n)) {
 744                 if (dev)
 745                         dev_put(dev);
 746                 kfree(n);
 747                 n = NULL;
 748                 goto out;
 749         }
 750
 751         write_lock_bh(&tbl->lock);
 752         n->next = tbl->phash_buckets[hash_val];
 753         tbl->phash_buckets[hash_val] = n;
 754         write_unlock_bh(&tbl->lock);
 755 out:
 756         return n;
 757 }
 758 EXPORT_SYMBOL(pneigh_lookup);
 759
 760
 761 int pneigh_delete(struct neigh_table *tbl, struct net *net, const void *pkey,
 762                   struct net_device *dev)
 763 {
 764         struct pneigh_entry *n, **np;
 765         unsigned int key_len = tbl->key_len;
 766         u32 hash_val = pneigh_hash(pkey, key_len);
 767
 768         write_lock_bh(&tbl->lock);
 769         for (np = &tbl->phash_buckets[hash_val]; (n = *np) != NULL;
 770              np = &n->next) {
 771                 if (!memcmp(n->key, pkey, key_len) && n->dev == dev &&
 772                     net_eq(pneigh_net(n), net)) {
 773                         *np = n->next;
 774                         write_unlock_bh(&tbl->lock);
 775                         if (tbl->pdestructor)
 776                                 tbl->pdestructor(n);
 777                         if (n->dev)
 778                                 dev_put(n->dev);
 779                         kfree(n);
 780                         return 0;
 781                 }
 782         }
 783         write_unlock_bh(&tbl->lock);
 784         return -ENOENT;
 785 }
 786
 787 static int pneigh_ifdown_and_unlock(struct neigh_table *tbl,
 788                                     struct net_device *dev)
 789 {
 790         struct pneigh_entry *n, **np, *freelist = NULL;
 791         u32 h;
 792
 793         for (h = 0; h <= PNEIGH_HASHMASK; h++) {
 794                 np = &tbl->phash_buckets[h];
 795                 while ((n = *np) != NULL) {
 796                         if (!dev || n->dev == dev) {
 797                                 *np = n->next;
 798                                 n->next = freelist;
 799                                 freelist = n;
 800                                 continue;
 801                         }
 802                         np = &n->next;
 803                 }
 804         }
 805         write_unlock_bh(&tbl->lock);
 806         while ((n = freelist)) {
 807                 freelist = n->next;
 808                 n->next = NULL;
 809                 if (tbl->pdestructor)
 810                         tbl->pdestructor(n);
 811                 if (n->dev)
 812                         dev_put(n->dev);
 813                 kfree(n);
 814         }
 815         return -ENOENT;
 816 }
 817
 818 static void neigh_parms_destroy(struct neigh_parms *parms);
 819
 820 static inline void neigh_parms_put(struct neigh_parms *parms)
 821 {
 822         if (refcount_dec_and_test(&parms->refcnt))
 823                 neigh_parms_destroy(parms);
 824 }
 825
 826 /*
 827  *      neighbour must already be out of the table;
 828  *
 829  */
 830 void neigh_destroy(struct neighbour *neigh)
 831 {
 832         struct net_device *dev = neigh->dev;
 833
 834         NEIGH_CACHE_STAT_INC(neigh->tbl, destroys);
 835
 836         if (!neigh->dead) {
 837                 pr_warn("Destroying alive neighbour %p\n", neigh);
 838                 dump_stack();
 839                 return;
 840         }
 841
 842         if (neigh_del_timer(neigh))
 843                 pr_warn("Impossible event\n");
 844
 845         write_lock_bh(&neigh->lock);
 846         __skb_queue_purge(&neigh->arp_queue);
 847         write_unlock_bh(&neigh->lock);
 848         neigh->arp_queue_len_bytes = 0;
 849
 850         if (dev->netdev_ops->ndo_neigh_destroy)
 851                 dev->netdev_ops->ndo_neigh_destroy(dev, neigh);
 852
 853         dev_put(dev);
 854         neigh_parms_put(neigh->parms);
 855
 856         neigh_dbg(2, "neigh %p is destroyed\n", neigh);
 857
 858         atomic_dec(&neigh->tbl->entries);
 859         kfree_rcu(neigh, rcu);
 860 }
 861 EXPORT_SYMBOL(neigh_destroy);
 862
 863 /* Neighbour state is suspicious;
 864    disable fast path.
 865
 866    Called with write_locked neigh.
 867  */
 868 static void neigh_suspect(struct neighbour *neigh)
 869 {
 870         neigh_dbg(2, "neigh %p is suspected\n", neigh);
 871
 872         neigh->output = neigh->ops->output;
 873 }
 874
 875 /* Neighbour state is OK;
 876    enable fast path.
 877
 878    Called with write_locked neigh.
 879  */
 880 static void neigh_connect(struct neighbour *neigh)
 881 {
 882         neigh_dbg(2, "neigh %p is connected\n", neigh);
 883
 884         neigh->output = neigh->ops->connected_output;
 885 }
 886
 887 static void neigh_periodic_work(struct work_struct *work)
 888 {
 889         struct neigh_table *tbl = container_of(work, struct neigh_table, gc_work.work);
 890         struct neighbour *n;
 891         struct neighbour __rcu **np;
 892         unsigned int i;
 893         struct neigh_hash_table *nht;
 894
 895         NEIGH_CACHE_STAT_INC(tbl, periodic_gc_runs);
 896
 897         write_lock_bh(&tbl->lock);
 898         nht = rcu_dereference_protected(tbl->nht,
 899                                         lockdep_is_held(&tbl->lock));
 900
 901         /*
 902          *      periodically recompute ReachableTime from random function
 903          */
 904
 905         if (time_after(jiffies, tbl->last_rand + 300 * HZ)) {
 906                 struct neigh_parms *p;
 907                 tbl->last_rand = jiffies;
 908                 list_for_each_entry(p, &tbl->parms_list, list)
 909                         p->reachable_time =
 910                                 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
 911         }
 912
 913         if (atomic_read(&tbl->entries) < tbl->gc_thresh1)
 914                 goto out;
 915
 916         for (i = 0 ; i < (1 << nht->hash_shift); i++) {
 917                 np = &nht->hash_buckets[i];
 918
 919                 while ((n = rcu_dereference_protected(*np,
 920                                 lockdep_is_held(&tbl->lock))) != NULL) {
 921                         unsigned int state;
 922
 923                         write_lock(&n->lock);
 924
 925                         state = n->nud_state;
 926                         if ((state & (NUD_PERMANENT | NUD_IN_TIMER)) ||
 927                             (n->flags & NTF_EXT_LEARNED)) {
 928                                 write_unlock(&n->lock);
 929                                 goto next_elt;
 930                         }
 931
 932                         if (time_before(n->used, n->confirmed))
 933                                 n->used = n->confirmed;
 934
 935                         if (refcount_read(&n->refcnt) == 1 &&
 936                             (state == NUD_FAILED ||
 937                              time_after(jiffies, n->used + NEIGH_VAR(n->parms, GC_STALETIME)))) {
 938                                 *np = n->next;
 939                                 neigh_mark_dead(n);
 940                                 write_unlock(&n->lock);
 941                                 neigh_cleanup_and_release(n);
 942                                 continue;
 943                         }
 944                         write_unlock(&n->lock);
 945
 946 next_elt:
 947                         np = &n->next;
 948                 }
 949                 /*
 950                  * It's fine to release lock here, even if hash table
 951                  * grows while we are preempted.
 952                  */
 953                 write_unlock_bh(&tbl->lock);
 954                 cond_resched();
 955                 write_lock_bh(&tbl->lock);
 956                 nht = rcu_dereference_protected(tbl->nht,
 957                                                 lockdep_is_held(&tbl->lock));
 958         }
 959 out:
 960         /* Cycle through all hash buckets every BASE_REACHABLE_TIME/2 ticks.
 961          * ARP entry timeouts range from 1/2 BASE_REACHABLE_TIME to 3/2
 962          * BASE_REACHABLE_TIME.
 963          */
 964         queue_delayed_work(system_power_efficient_wq, &tbl->gc_work,
 965                               NEIGH_VAR(&tbl->parms, BASE_REACHABLE_TIME) >> 1);
 966         write_unlock_bh(&tbl->lock);
 967 }
 968
 969 static __inline__ int neigh_max_probes(struct neighbour *n)
 970 {
 971         struct neigh_parms *p = n->parms;
 972         return NEIGH_VAR(p, UCAST_PROBES) + NEIGH_VAR(p, APP_PROBES) +
 973                (n->nud_state & NUD_PROBE ? NEIGH_VAR(p, MCAST_REPROBES) :
 974                 NEIGH_VAR(p, MCAST_PROBES));
 975 }
 976
 977 static void neigh_invalidate(struct neighbour *neigh)
 978         __releases(neigh->lock)
 979         __acquires(neigh->lock)
 980 {
 981         struct sk_buff *skb;
 982
 983         NEIGH_CACHE_STAT_INC(neigh->tbl, res_failed);
 984         neigh_dbg(2, "neigh %p is failed\n", neigh);
 985         neigh->updated = jiffies;
 986
 987         /* It is very thin place. report_unreachable is very complicated
 988            routine. Particularly, it can hit the same neighbour entry!
 989
 990            So that, we try to be accurate and avoid dead loop. --ANK
 991          */
 992         while (neigh->nud_state == NUD_FAILED &&
 993                (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
 994                 write_unlock(&neigh->lock);
 995                 neigh->ops->error_report(neigh, skb);
 996                 write_lock(&neigh->lock);
 997         }
 998         __skb_queue_purge(&neigh->arp_queue);
 999         neigh->arp_queue_len_bytes = 0;
1000 }
1001
1002 static void neigh_probe(struct neighbour *neigh)
1003         __releases(neigh->lock)
1004 {
1005         struct sk_buff *skb = skb_peek_tail(&neigh->arp_queue);
1006         /* keep skb alive even if arp_queue overflows */
1007         if (skb)
1008                 skb = skb_clone(skb, GFP_ATOMIC);
1009         write_unlock(&neigh->lock);
1010         if (neigh->ops->solicit)
1011                 neigh->ops->solicit(neigh, skb);
1012         atomic_inc(&neigh->probes);
1013         consume_skb(skb);
1014 }
1015
1016 /* Called when a timer expires for a neighbour entry. */
1017
1018 static void neigh_timer_handler(struct timer_list *t)
1019 {
1020         unsigned long now, next;
1021         struct neighbour *neigh = from_timer(neigh, t, timer);
1022         unsigned int state;
1023         int notify = 0;
1024
1025         write_lock(&neigh->lock);
1026
1027         state = neigh->nud_state;
1028         now = jiffies;
1029         next = now + HZ;
1030
1031         if (!(state & NUD_IN_TIMER))
1032                 goto out;
1033
1034         if (state & NUD_REACHABLE) {
1035                 if (time_before_eq(now,
1036                                    neigh->confirmed + neigh->parms->reachable_time)) {
1037                         neigh_dbg(2, "neigh %p is still alive\n", neigh);
1038                         next = neigh->confirmed + neigh->parms->reachable_time;
1039                 } else if (time_before_eq(now,
1040                                           neigh->used +
1041                                           NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME))) {
1042                         neigh_dbg(2, "neigh %p is delayed\n", neigh);
1043                         neigh->nud_state = NUD_DELAY;
1044                         neigh->updated = jiffies;
1045                         neigh_suspect(neigh);
1046                         next = now + NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME);
1047                 } else {
1048                         neigh_dbg(2, "neigh %p is suspected\n", neigh);
1049                         neigh->nud_state = NUD_STALE;
1050                         neigh->updated = jiffies;
1051                         neigh_suspect(neigh);
1052                         notify = 1;
1053                 }
1054         } else if (state & NUD_DELAY) {
1055                 if (time_before_eq(now,
1056                                    neigh->confirmed +
1057                                    NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME))) {
1058                         neigh_dbg(2, "neigh %p is now reachable\n", neigh);
1059                         neigh->nud_state = NUD_REACHABLE;
1060                         neigh->updated = jiffies;
1061                         neigh_connect(neigh);
1062                         notify = 1;
1063                         next = neigh->confirmed + neigh->parms->reachable_time;
1064                 } else {
1065                         neigh_dbg(2, "neigh %p is probed\n", neigh);
1066                         neigh->nud_state = NUD_PROBE;
1067                         neigh->updated = jiffies;
1068                         atomic_set(&neigh->probes, 0);
1069                         notify = 1;
1070                         next = now + max(NEIGH_VAR(neigh->parms, RETRANS_TIME),
1071                                          HZ/100);
1072                 }
1073         } else {
1074                 /* NUD_PROBE|NUD_INCOMPLETE */
1075                 next = now + max(NEIGH_VAR(neigh->parms, RETRANS_TIME), HZ/100);
1076         }
1077
1078         if ((neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) &&
1079             atomic_read(&neigh->probes) >= neigh_max_probes(neigh)) {
1080                 neigh->nud_state = NUD_FAILED;
1081                 notify = 1;
1082                 neigh_invalidate(neigh);
1083                 goto out;
1084         }
1085
1086         if (neigh->nud_state & NUD_IN_TIMER) {
1087                 if (time_before(next, jiffies + HZ/100))
1088                         next = jiffies + HZ/100;
1089                 if (!mod_timer(&neigh->timer, next))
1090                         neigh_hold(neigh);
1091         }
1092         if (neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) {
1093                 neigh_probe(neigh);
1094         } else {
1095 out:
1096                 write_unlock(&neigh->lock);
1097         }
1098
1099         if (notify)
1100                 neigh_update_notify(neigh, 0);
1101
1102         trace_neigh_timer_handler(neigh, 0);
1103
1104         neigh_release(neigh);
1105 }
1106
1107 int __neigh_event_send(struct neighbour *neigh, struct sk_buff *skb)
1108 {
1109         int rc;
1110         bool immediate_probe = false;
1111
1112         write_lock_bh(&neigh->lock);
1113
1114         rc = 0;
1115         if (neigh->nud_state & (NUD_CONNECTED | NUD_DELAY | NUD_PROBE))
1116                 goto out_unlock_bh;
1117         if (neigh->dead)
1118                 goto out_dead;
1119
1120         if (!(neigh->nud_state & (NUD_STALE | NUD_INCOMPLETE))) {
1121                 if (NEIGH_VAR(neigh->parms, MCAST_PROBES) +
1122                     NEIGH_VAR(neigh->parms, APP_PROBES)) {
1123                         unsigned long next, now = jiffies;
1124
1125                         atomic_set(&neigh->probes,
1126                                    NEIGH_VAR(neigh->parms, UCAST_PROBES));
1127                         neigh_del_timer(neigh);
1128                         neigh->nud_state     = NUD_INCOMPLETE;
1129                         neigh->updated = now;
1130                         next = now + max(NEIGH_VAR(neigh->parms, RETRANS_TIME),
1131                                          HZ/100);
1132                         neigh_add_timer(neigh, next);
1133                         immediate_probe = true;
1134                 } else {
1135                         neigh->nud_state = NUD_FAILED;
1136                         neigh->updated = jiffies;
1137                         write_unlock_bh(&neigh->lock);
1138
1139                         kfree_skb(skb);
1140                         return 1;
1141                 }
1142         } else if (neigh->nud_state & NUD_STALE) {
1143                 neigh_dbg(2, "neigh %p is delayed\n", neigh);
1144                 neigh_del_timer(neigh);
1145                 neigh->nud_state = NUD_DELAY;
1146                 neigh->updated = jiffies;
1147                 neigh_add_timer(neigh, jiffies +
1148                                 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME));
1149         }
1150
1151         if (neigh->nud_state == NUD_INCOMPLETE) {
1152                 if (skb) {
1153                         while (neigh->arp_queue_len_bytes + skb->truesize >
1154                                NEIGH_VAR(neigh->parms, QUEUE_LEN_BYTES)) {
1155                                 struct sk_buff *buff;
1156
1157                                 buff = __skb_dequeue(&neigh->arp_queue);
1158                                 if (!buff)
1159                                         break;
1160                                 neigh->arp_queue_len_bytes -= buff->truesize;
1161                                 kfree_skb(buff);
1162                                 NEIGH_CACHE_STAT_INC(neigh->tbl, unres_discards);
1163                         }
1164                         skb_dst_force(skb);
1165                         __skb_queue_tail(&neigh->arp_queue, skb);
1166                         neigh->arp_queue_len_bytes += skb->truesize;
1167                 }
1168                 rc = 1;
1169         }
1170 out_unlock_bh:
1171         if (immediate_probe)
1172                 neigh_probe(neigh);
1173         else
1174                 write_unlock(&neigh->lock);
1175         local_bh_enable();
1176         trace_neigh_event_send_done(neigh, rc);
1177         return rc;
1178
1179 out_dead:
1180         if (neigh->nud_state & NUD_STALE)
1181                 goto out_unlock_bh;
1182         write_unlock_bh(&neigh->lock);
1183         kfree_skb(skb);
1184         trace_neigh_event_send_dead(neigh, 1);
1185         return 1;
1186 }
1187 EXPORT_SYMBOL(__neigh_event_send);
1188
1189 static void neigh_update_hhs(struct neighbour *neigh)
1190 {
1191         struct hh_cache *hh;
1192         void (*update)(struct hh_cache*, const struct net_device*, const unsigned char *)
1193                 = NULL;
1194
1195         if (neigh->dev->header_ops)
1196                 update = neigh->dev->header_ops->cache_update;
1197
1198         if (update) {
1199                 hh = &neigh->hh;
1200                 if (READ_ONCE(hh->hh_len)) {
1201                         write_seqlock_bh(&hh->hh_lock);
1202                         update(hh, neigh->dev, neigh->ha);
1203                         write_sequnlock_bh(&hh->hh_lock);
1204                 }
1205         }
1206 }
1207
1208
1209
1210 /* Generic update routine.
1211    -- lladdr is new lladdr or NULL, if it is not supplied.
1212    -- new    is new state.
1213    -- flags
1214         NEIGH_UPDATE_F_OVERRIDE allows to override existing lladdr,
1215                                 if it is different.
1216         NEIGH_UPDATE_F_WEAK_OVERRIDE will suspect existing "connected"
1217                                 lladdr instead of overriding it
1218                                 if it is different.
1219         NEIGH_UPDATE_F_ADMIN    means that the change is administrative.
1220
1221         NEIGH_UPDATE_F_OVERRIDE_ISROUTER allows to override existing
1222                                 NTF_ROUTER flag.
1223         NEIGH_UPDATE_F_ISROUTER indicates if the neighbour is known as
1224                                 a router.
1225
1226    Caller MUST hold reference count on the entry.
1227  */
1228
1229 static int __neigh_update(struct neighbour *neigh, const u8 *lladdr,
1230                           u8 new, u32 flags, u32 nlmsg_pid,
1231                           struct netlink_ext_ack *extack)
1232 {
1233         bool ext_learn_change = false;
1234         u8 old;
1235         int err;
1236         int notify = 0;
1237         struct net_device *dev;
1238         int update_isrouter = 0;
1239
1240         trace_neigh_update(neigh, lladdr, new, flags, nlmsg_pid);
1241
1242         write_lock_bh(&neigh->lock);
1243
1244         dev    = neigh->dev;
1245         old    = neigh->nud_state;
1246         err    = -EPERM;
1247
1248         if (!(flags & NEIGH_UPDATE_F_ADMIN) &&
1249             (old & (NUD_NOARP | NUD_PERMANENT)))
1250                 goto out;
1251         if (neigh->dead) {
1252                 NL_SET_ERR_MSG(extack, "Neighbor entry is now dead");
1253                 goto out;
1254         }
1255
1256         ext_learn_change = neigh_update_ext_learned(neigh, flags, &notify);
1257
1258         if (!(new & NUD_VALID)) {
1259                 neigh_del_timer(neigh);
1260                 if (old & NUD_CONNECTED)
1261                         neigh_suspect(neigh);
1262                 neigh->nud_state = new;
1263                 err = 0;
1264                 notify = old & NUD_VALID;
1265                 if ((old & (NUD_INCOMPLETE | NUD_PROBE)) &&
1266                     (new & NUD_FAILED)) {
1267                         neigh_invalidate(neigh);
1268                         notify = 1;
1269                 }
1270                 goto out;
1271         }
1272
1273         /* Compare new lladdr with cached one */
1274         if (!dev->addr_len) {
1275                 /* First case: device needs no address. */
1276                 lladdr = neigh->ha;
1277         } else if (lladdr) {
1278                 /* The second case: if something is already cached
1279                    and a new address is proposed:
1280                    - compare new & old
1281                    - if they are different, check override flag
1282                  */
1283                 if ((old & NUD_VALID) &&
1284                     !memcmp(lladdr, neigh->ha, dev->addr_len))
1285                         lladdr = neigh->ha;
1286         } else {
1287                 /* No address is supplied; if we know something,
1288                    use it, otherwise discard the request.
1289                  */
1290                 err = -EINVAL;
1291                 if (!(old & NUD_VALID)) {
1292                         NL_SET_ERR_MSG(extack, "No link layer address given");
1293                         goto out;
1294                 }
1295                 lladdr = neigh->ha;
1296         }
1297
1298         /* Update confirmed timestamp for neighbour entry after we
1299          * received ARP packet even if it doesn't change IP to MAC binding.
1300          */
1301         if (new & NUD_CONNECTED)
1302                 neigh->confirmed = jiffies;
1303
1304         /* If entry was valid and address is not changed,
1305            do not change entry state, if new one is STALE.
1306          */
1307         err = 0;
1308         update_isrouter = flags & NEIGH_UPDATE_F_OVERRIDE_ISROUTER;
1309         if (old & NUD_VALID) {
1310                 if (lladdr != neigh->ha && !(flags & NEIGH_UPDATE_F_OVERRIDE)) {
1311                         update_isrouter = 0;
1312                         if ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) &&
1313                             (old & NUD_CONNECTED)) {
1314                                 lladdr = neigh->ha;
1315                                 new = NUD_STALE;
1316                         } else
1317                                 goto out;
1318                 } else {
1319                         if (lladdr == neigh->ha && new == NUD_STALE &&
1320                             !(flags & NEIGH_UPDATE_F_ADMIN))
1321                                 new = old;
1322                 }
1323         }
1324
1325         /* Update timestamp only once we know we will make a change to the
1326          * neighbour entry. Otherwise we risk to move the locktime window with
1327          * noop updates and ignore relevant ARP updates.
1328          */
1329         if (new != old || lladdr != neigh->ha)
1330                 neigh->updated = jiffies;
1331
1332         if (new != old) {
1333                 neigh_del_timer(neigh);
1334                 if (new & NUD_PROBE)
1335                         atomic_set(&neigh->probes, 0);
1336                 if (new & NUD_IN_TIMER)
1337                         neigh_add_timer(neigh, (jiffies +
1338                                                 ((new & NUD_REACHABLE) ?
1339                                                  neigh->parms->reachable_time :
1340                                                  0)));
1341                 neigh->nud_state = new;
1342                 notify = 1;
1343         }
1344
1345         if (lladdr != neigh->ha) {
1346                 write_seqlock(&neigh->ha_lock);
1347                 memcpy(&neigh->ha, lladdr, dev->addr_len);
1348                 write_sequnlock(&neigh->ha_lock);
1349                 neigh_update_hhs(neigh);
1350                 if (!(new & NUD_CONNECTED))
1351                         neigh->confirmed = jiffies -
1352                                       (NEIGH_VAR(neigh->parms, BASE_REACHABLE_TIME) << 1);
1353                 notify = 1;
1354         }
1355         if (new == old)
1356                 goto out;
1357         if (new & NUD_CONNECTED)
1358                 neigh_connect(neigh);
1359         else
1360                 neigh_suspect(neigh);
1361         if (!(old & NUD_VALID)) {
1362                 struct sk_buff *skb;
1363
1364                 /* Again: avoid dead loop if something went wrong */
1365
1366                 while (neigh->nud_state & NUD_VALID &&
1367                        (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
1368                         struct dst_entry *dst = skb_dst(skb);
1369                         struct neighbour *n2, *n1 = neigh;
1370                         write_unlock_bh(&neigh->lock);
1371
1372                         rcu_read_lock();
1373
1374                         /* Why not just use 'neigh' as-is?  The problem is that
1375                          * things such as shaper, eql, and sch_teql can end up
1376                          * using alternative, different, neigh objects to output
1377                          * the packet in the output path.  So what we need to do
1378                          * here is re-lookup the top-level neigh in the path so
1379                          * we can reinject the packet there.
1380                          */
1381                         n2 = NULL;
1382                         if (dst) {
1383                                 n2 = dst_neigh_lookup_skb(dst, skb);
1384                                 if (n2)
1385                                         n1 = n2;
1386                         }
1387                         n1->output(n1, skb);
1388                         if (n2)
1389                                 neigh_release(n2);
1390                         rcu_read_unlock();
1391
1392                         write_lock_bh(&neigh->lock);
1393                 }
1394                 __skb_queue_purge(&neigh->arp_queue);
1395                 neigh->arp_queue_len_bytes = 0;
1396         }
1397 out:
1398         if (update_isrouter)
1399                 neigh_update_is_router(neigh, flags, &notify);
1400         write_unlock_bh(&neigh->lock);
1401
1402         if (((new ^ old) & NUD_PERMANENT) || ext_learn_change)
1403                 neigh_update_gc_list(neigh);
1404
1405         if (notify)
1406                 neigh_update_notify(neigh, nlmsg_pid);
1407
1408         trace_neigh_update_done(neigh, err);
1409
1410         return err;
1411 }
1412
1413 int neigh_update(struct neighbour *neigh, const u8 *lladdr, u8 new,
1414                  u32 flags, u32 nlmsg_pid)
1415 {
1416         return __neigh_update(neigh, lladdr, new, flags, nlmsg_pid, NULL);
1417 }
1418 EXPORT_SYMBOL(neigh_update);
1419
1420 /* Update the neigh to listen temporarily for probe responses, even if it is
1421  * in a NUD_FAILED state. The caller has to hold neigh->lock for writing.
1422  */
1423 void __neigh_set_probe_once(struct neighbour *neigh)
1424 {
1425         if (neigh->dead)
1426                 return;
1427         neigh->updated = jiffies;
1428         if (!(neigh->nud_state & NUD_FAILED))
1429                 return;
1430         neigh->nud_state = NUD_INCOMPLETE;
1431         atomic_set(&neigh->probes, neigh_max_probes(neigh));
1432         neigh_add_timer(neigh,
1433                         jiffies + max(NEIGH_VAR(neigh->parms, RETRANS_TIME),
1434                                       HZ/100));
1435 }
1436 EXPORT_SYMBOL(__neigh_set_probe_once);
1437
1438 struct neighbour *neigh_event_ns(struct neigh_table *tbl,
1439                                  u8 *lladdr, void *saddr,
1440                                  struct net_device *dev)
1441 {
1442         struct neighbour *neigh = __neigh_lookup(tbl, saddr, dev,
1443                                                  lladdr || !dev->addr_len);
1444         if (neigh)
1445                 neigh_update(neigh, lladdr, NUD_STALE,
1446                              NEIGH_UPDATE_F_OVERRIDE, 0);
1447         return neigh;
1448 }
1449 EXPORT_SYMBOL(neigh_event_ns);
1450
1451 /* called with read_lock_bh(&n->lock); */
1452 static void neigh_hh_init(struct neighbour *n)
1453 {
1454         struct net_device *dev = n->dev;
1455         __be16 prot = n->tbl->protocol;
1456         struct hh_cache *hh = &n->hh;
1457
1458         write_lock_bh(&n->lock);
1459
1460         /* Only one thread can come in here and initialize the
1461          * hh_cache entry.
1462          */
1463         if (!hh->hh_len)
1464                 dev->header_ops->cache(n, hh, prot);
1465
1466         write_unlock_bh(&n->lock);
1467 }
1468
1469 /* Slow and careful. */
1470
1471 int neigh_resolve_output(struct neighbour *neigh, struct sk_buff *skb)
1472 {
1473         int rc = 0;
1474
1475         if (!neigh_event_send(neigh, skb)) {
1476                 int err;
1477                 struct net_device *dev = neigh->dev;
1478                 unsigned int seq;
1479
1480                 if (dev->header_ops->cache && !READ_ONCE(neigh->hh.hh_len))
1481                         neigh_hh_init(neigh);
1482
1483                 do {
1484                         __skb_pull(skb, skb_network_offset(skb));
1485                         seq = read_seqbegin(&neigh->ha_lock);
1486                         err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1487                                               neigh->ha, NULL, skb->len);
1488                 } while (read_seqretry(&neigh->ha_lock, seq));
1489
1490                 if (err >= 0)
1491                         rc = dev_queue_xmit(skb);
1492                 else
1493                         goto out_kfree_skb;
1494         }
1495 out:
1496         return rc;
1497 out_kfree_skb:
1498         rc = -EINVAL;
1499         kfree_skb(skb);
1500         goto out;
1501 }
1502 EXPORT_SYMBOL(neigh_resolve_output);
1503
1504 /* As fast as possible without hh cache */
1505
1506 int neigh_connected_output(struct neighbour *neigh, struct sk_buff *skb)
1507 {
1508         struct net_device *dev = neigh->dev;
1509         unsigned int seq;
1510         int err;
1511
1512         do {
1513                 __skb_pull(skb, skb_network_offset(skb));
1514                 seq = read_seqbegin(&neigh->ha_lock);
1515                 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1516                                       neigh->ha, NULL, skb->len);
1517         } while (read_seqretry(&neigh->ha_lock, seq));
1518
1519         if (err >= 0)
1520                 err = dev_queue_xmit(skb);
1521         else {
1522                 err = -EINVAL;
1523                 kfree_skb(skb);
1524         }
1525         return err;
1526 }
1527 EXPORT_SYMBOL(neigh_connected_output);
1528
1529 int neigh_direct_output(struct neighbour *neigh, struct sk_buff *skb)
1530 {
1531         return dev_queue_xmit(skb);
1532 }
1533 EXPORT_SYMBOL(neigh_direct_output);
1534
1535 static void neigh_proxy_process(struct timer_list *t)
1536 {
1537         struct neigh_table *tbl = from_timer(tbl, t, proxy_timer);
1538         long sched_next = 0;
1539         unsigned long now = jiffies;
1540         struct sk_buff *skb, *n;
1541
1542         spin_lock(&tbl->proxy_queue.lock);
1543
1544         skb_queue_walk_safe(&tbl->proxy_queue, skb, n) {
1545                 long tdif = NEIGH_CB(skb)->sched_next - now;
1546
1547                 if (tdif <= 0) {
1548                         struct net_device *dev = skb->dev;
1549
1550                         __skb_unlink(skb, &tbl->proxy_queue);
1551                         if (tbl->proxy_redo && netif_running(dev)) {
1552                                 rcu_read_lock();
1553                                 tbl->proxy_redo(skb);
1554                                 rcu_read_unlock();
1555                         } else {
1556                                 kfree_skb(skb);
1557                         }
1558
1559                         dev_put(dev);
1560                 } else if (!sched_next || tdif < sched_next)
1561                         sched_next = tdif;
1562         }
1563         del_timer(&tbl->proxy_timer);
1564         if (sched_next)
1565                 mod_timer(&tbl->proxy_timer, jiffies + sched_next);
1566         spin_unlock(&tbl->proxy_queue.lock);
1567 }
1568
1569 void pneigh_enqueue(struct neigh_table *tbl, struct neigh_parms *p,
1570                     struct sk_buff *skb)
1571 {
1572         unsigned long sched_next = jiffies +
1573                         prandom_u32_max(NEIGH_VAR(p, PROXY_DELAY));
1574
1575         if (tbl->proxy_queue.qlen > NEIGH_VAR(p, PROXY_QLEN)) {
1576                 kfree_skb(skb);
1577                 return;
1578         }
1579
1580         NEIGH_CB(skb)->sched_next = sched_next;
1581         NEIGH_CB(skb)->flags |= LOCALLY_ENQUEUED;
1582
1583         spin_lock(&tbl->proxy_queue.lock);
1584         if (del_timer(&tbl->proxy_timer)) {
1585                 if (time_before(tbl->proxy_timer.expires, sched_next))
1586                         sched_next = tbl->proxy_timer.expires;
1587         }
1588         skb_dst_drop(skb);
1589         dev_hold(skb->dev);
1590         __skb_queue_tail(&tbl->proxy_queue, skb);
1591         mod_timer(&tbl->proxy_timer, sched_next);
1592         spin_unlock(&tbl->proxy_queue.lock);
1593 }
1594 EXPORT_SYMBOL(pneigh_enqueue);
1595
1596 static inline struct neigh_parms *lookup_neigh_parms(struct neigh_table *tbl,
1597                                                       struct net *net, int ifindex)
1598 {
1599         struct neigh_parms *p;
1600
1601         list_for_each_entry(p, &tbl->parms_list, list) {
1602                 if ((p->dev && p->dev->ifindex == ifindex && net_eq(neigh_parms_net(p), net)) ||
1603                     (!p->dev && !ifindex && net_eq(net, &init_net)))
1604                         return p;
1605         }
1606
1607         return NULL;
1608 }
1609
1610 struct neigh_parms *neigh_parms_alloc(struct net_device *dev,
1611                                       struct neigh_table *tbl)
1612 {
1613         struct neigh_parms *p;
1614         struct net *net = dev_net(dev);
1615         const struct net_device_ops *ops = dev->netdev_ops;
1616
1617         p = kmemdup(&tbl->parms, sizeof(*p), GFP_KERNEL);
1618         if (p) {
1619                 p->tbl            = tbl;
1620                 refcount_set(&p->refcnt, 1);
1621                 p->reachable_time =
1622                                 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
1623                 dev_hold(dev);
1624                 p->dev = dev;
1625                 write_pnet(&p->net, net);
1626                 p->sysctl_table = NULL;
1627
1628                 if (ops->ndo_neigh_setup && ops->ndo_neigh_setup(dev, p)) {
1629                         dev_put(dev);
1630                         kfree(p);
1631                         return NULL;
1632                 }
1633
1634                 write_lock_bh(&tbl->lock);
1635                 list_add(&p->list, &tbl->parms.list);
1636                 write_unlock_bh(&tbl->lock);
1637
1638                 neigh_parms_data_state_cleanall(p);
1639         }
1640         return p;
1641 }
1642 EXPORT_SYMBOL(neigh_parms_alloc);
1643
1644 static void neigh_rcu_free_parms(struct rcu_head *head)
1645 {
1646         struct neigh_parms *parms =
1647                 container_of(head, struct neigh_parms, rcu_head);
1648
1649         neigh_parms_put(parms);
1650 }
1651
1652 void neigh_parms_release(struct neigh_table *tbl, struct neigh_parms *parms)
1653 {
1654         if (!parms || parms == &tbl->parms)
1655                 return;
1656         write_lock_bh(&tbl->lock);
1657         list_del(&parms->list);
1658         parms->dead = 1;
1659         write_unlock_bh(&tbl->lock);
1660         if (parms->dev)
1661                 dev_put(parms->dev);
1662         call_rcu(&parms->rcu_head, neigh_rcu_free_parms);
1663 }
1664 EXPORT_SYMBOL(neigh_parms_release);
1665
1666 static void neigh_parms_destroy(struct neigh_parms *parms)
1667 {
1668         kfree(parms);
1669 }
1670
1671 static struct lock_class_key neigh_table_proxy_queue_class;
1672
1673 static struct neigh_table *neigh_tables[NEIGH_NR_TABLES] __read_mostly;
1674
1675 void neigh_table_init(int index, struct neigh_table *tbl)
1676 {
1677         unsigned long now = jiffies;
1678         unsigned long phsize;
1679
1680         INIT_LIST_HEAD(&tbl->parms_list);
1681         INIT_LIST_HEAD(&tbl->gc_list);
1682         list_add(&tbl->parms.list, &tbl->parms_list);
1683         write_pnet(&tbl->parms.net, &init_net);
1684         refcount_set(&tbl->parms.refcnt, 1);
1685         tbl->parms.reachable_time =
1686                           neigh_rand_reach_time(NEIGH_VAR(&tbl->parms, BASE_REACHABLE_TIME));
1687
1688         tbl->stats = alloc_percpu(struct neigh_statistics);
1689         if (!tbl->stats)
1690                 panic("cannot create neighbour cache statistics");
1691
1692 #ifdef CONFIG_PROC_FS
1693         if (!proc_create_seq_data(tbl->id, 0, init_net.proc_net_stat,
1694                               &neigh_stat_seq_ops, tbl))
1695                 panic("cannot create neighbour proc dir entry");
1696 #endif
1697
1698         RCU_INIT_POINTER(tbl->nht, neigh_hash_alloc(3));
1699
1700         phsize = (PNEIGH_HASHMASK + 1) * sizeof(struct pneigh_entry *);
1701         tbl->phash_buckets = kzalloc(phsize, GFP_KERNEL);
1702
1703         if (!tbl->nht || !tbl->phash_buckets)
1704                 panic("cannot allocate neighbour cache hashes");
1705
1706         if (!tbl->entry_size)
1707                 tbl->entry_size = ALIGN(offsetof(struct neighbour, primary_key) +
1708                                         tbl->key_len, NEIGH_PRIV_ALIGN);
1709         else
1710                 WARN_ON(tbl->entry_size % NEIGH_PRIV_ALIGN);
1711
1712         rwlock_init(&tbl->lock);
1713         INIT_DEFERRABLE_WORK(&tbl->gc_work, neigh_periodic_work);
1714         queue_delayed_work(system_power_efficient_wq, &tbl->gc_work,
1715                         tbl->parms.reachable_time);
1716         timer_setup(&tbl->proxy_timer, neigh_proxy_process, 0);
1717         skb_queue_head_init_class(&tbl->proxy_queue,
1718                         &neigh_table_proxy_queue_class);
1719
1720         tbl->last_flush = now;
1721         tbl->last_rand  = now + tbl->parms.reachable_time * 20;
1722
1723         neigh_tables[index] = tbl;
1724 }
1725 EXPORT_SYMBOL(neigh_table_init);
1726
1727 int neigh_table_clear(int index, struct neigh_table *tbl)
1728 {
1729         neigh_tables[index] = NULL;
1730         /* It is not clean... Fix it to unload IPv6 module safely */
1731         cancel_delayed_work_sync(&tbl->gc_work);
1732         del_timer_sync(&tbl->proxy_timer);
1733         pneigh_queue_purge(&tbl->proxy_queue);
1734         neigh_ifdown(tbl, NULL);
1735         if (atomic_read(&tbl->entries))
1736                 pr_crit("neighbour leakage\n");
1737
1738         call_rcu(&rcu_dereference_protected(tbl->nht, 1)->rcu,
1739                  neigh_hash_free_rcu);
1740         tbl->nht = NULL;
1741
1742         kfree(tbl->phash_buckets);
1743         tbl->phash_buckets = NULL;
1744
1745         remove_proc_entry(tbl->id, init_net.proc_net_stat);
1746
1747         free_percpu(tbl->stats);
1748         tbl->stats = NULL;
1749
1750         return 0;
1751 }
1752 EXPORT_SYMBOL(neigh_table_clear);
1753
1754 static struct neigh_table *neigh_find_table(int family)
1755 {
1756         struct neigh_table *tbl = NULL;
1757
1758         switch (family) {
1759         case AF_INET:
1760                 tbl = neigh_tables[NEIGH_ARP_TABLE];
1761                 break;
1762         case AF_INET6:
1763                 tbl = neigh_tables[NEIGH_ND_TABLE];
1764                 break;
1765         case AF_DECnet:
1766                 tbl = neigh_tables[NEIGH_DN_TABLE];
1767                 break;
1768         }
1769
1770         return tbl;
1771 }
1772
1773 const struct nla_policy nda_policy[NDA_MAX+1] = {
1774         [NDA_UNSPEC]            = { .strict_start_type = NDA_NH_ID },
1775         [NDA_DST]               = { .type = NLA_BINARY, .len = MAX_ADDR_LEN },
1776         [NDA_LLADDR]            = { .type = NLA_BINARY, .len = MAX_ADDR_LEN },
1777         [NDA_CACHEINFO]         = { .len = sizeof(struct nda_cacheinfo) },
1778         [NDA_PROBES]            = { .type = NLA_U32 },
1779         [NDA_VLAN]              = { .type = NLA_U16 },
1780         [NDA_PORT]              = { .type = NLA_U16 },
1781         [NDA_VNI]               = { .type = NLA_U32 },
1782         [NDA_IFINDEX]           = { .type = NLA_U32 },
1783         [NDA_MASTER]            = { .type = NLA_U32 },
1784         [NDA_PROTOCOL]          = { .type = NLA_U8 },
1785         [NDA_NH_ID]             = { .type = NLA_U32 },
1786         [NDA_FDB_EXT_ATTRS]     = { .type = NLA_NESTED },
1787 };
1788
1789 static int neigh_delete(struct sk_buff *skb, struct nlmsghdr *nlh,
1790                         struct netlink_ext_ack *extack)
1791 {
1792         struct net *net = sock_net(skb->sk);
1793         struct ndmsg *ndm;
1794         struct nlattr *dst_attr;
1795         struct neigh_table *tbl;
1796         struct neighbour *neigh;
1797         struct net_device *dev = NULL;
1798         int err = -EINVAL;
1799
1800         ASSERT_RTNL();
1801         if (nlmsg_len(nlh) < sizeof(*ndm))
1802                 goto out;
1803
1804         dst_attr = nlmsg_find_attr(nlh, sizeof(*ndm), NDA_DST);
1805         if (!dst_attr) {
1806                 NL_SET_ERR_MSG(extack, "Network address not specified");
1807                 goto out;
1808         }
1809
1810         ndm = nlmsg_data(nlh);
1811         if (ndm->ndm_ifindex) {
1812                 dev = __dev_get_by_index(net, ndm->ndm_ifindex);
1813                 if (dev == NULL) {
1814                         err = -ENODEV;
1815                         goto out;
1816                 }
1817         }
1818
1819         tbl = neigh_find_table(ndm->ndm_family);
1820         if (tbl == NULL)
1821                 return -EAFNOSUPPORT;
1822
1823         if (nla_len(dst_attr) < (int)tbl->key_len) {
1824                 NL_SET_ERR_MSG(extack, "Invalid network address");
1825                 goto out;
1826         }
1827
1828         if (ndm->ndm_flags & NTF_PROXY) {
1829                 err = pneigh_delete(tbl, net, nla_data(dst_attr), dev);
1830                 goto out;
1831         }
1832
1833         if (dev == NULL)
1834                 goto out;
1835
1836         neigh = neigh_lookup(tbl, nla_data(dst_attr), dev);
1837         if (neigh == NULL) {
1838                 err = -ENOENT;
1839                 goto out;
1840         }
1841
1842         err = __neigh_update(neigh, NULL, NUD_FAILED,
1843                              NEIGH_UPDATE_F_OVERRIDE | NEIGH_UPDATE_F_ADMIN,
1844                              NETLINK_CB(skb).portid, extack);
1845         write_lock_bh(&tbl->lock);
1846         neigh_release(neigh);
1847         neigh_remove_one(neigh, tbl);
1848         write_unlock_bh(&tbl->lock);
1849
1850 out:
1851         return err;
1852 }
1853
1854 static int neigh_add(struct sk_buff *skb, struct nlmsghdr *nlh,
1855                      struct netlink_ext_ack *extack)
1856 {
1857         int flags = NEIGH_UPDATE_F_ADMIN | NEIGH_UPDATE_F_OVERRIDE |
1858                 NEIGH_UPDATE_F_OVERRIDE_ISROUTER;
1859         struct net *net = sock_net(skb->sk);
1860         struct ndmsg *ndm;
1861         struct nlattr *tb[NDA_MAX+1];
1862         struct neigh_table *tbl;
1863         struct net_device *dev = NULL;
1864         struct neighbour *neigh;
1865         void *dst, *lladdr;
1866         u8 protocol = 0;
1867         int err;
1868
1869         ASSERT_RTNL();
1870         err = nlmsg_parse_deprecated(nlh, sizeof(*ndm), tb, NDA_MAX,
1871                                      nda_policy, extack);
1872         if (err < 0)
1873                 goto out;
1874
1875         err = -EINVAL;
1876         if (!tb[NDA_DST]) {
1877                 NL_SET_ERR_MSG(extack, "Network address not specified");
1878                 goto out;
1879         }
1880
1881         ndm = nlmsg_data(nlh);
1882         if (ndm->ndm_ifindex) {
1883                 dev = __dev_get_by_index(net, ndm->ndm_ifindex);
1884                 if (dev == NULL) {
1885                         err = -ENODEV;
1886                         goto out;
1887                 }
1888
1889                 if (tb[NDA_LLADDR] && nla_len(tb[NDA_LLADDR]) < dev->addr_len) {
1890                         NL_SET_ERR_MSG(extack, "Invalid link address");
1891                         goto out;
1892                 }
1893         }
1894
1895         tbl = neigh_find_table(ndm->ndm_family);
1896         if (tbl == NULL)
1897                 return -EAFNOSUPPORT;
1898
1899         if (nla_len(tb[NDA_DST]) < (int)tbl->key_len) {
1900                 NL_SET_ERR_MSG(extack, "Invalid network address");
1901                 goto out;
1902         }
1903
1904         dst = nla_data(tb[NDA_DST]);
1905         lladdr = tb[NDA_LLADDR] ? nla_data(tb[NDA_LLADDR]) : NULL;
1906
1907         if (tb[NDA_PROTOCOL])
1908                 protocol = nla_get_u8(tb[NDA_PROTOCOL]);
1909
1910         if (ndm->ndm_flags & NTF_PROXY) {
1911                 struct pneigh_entry *pn;
1912
1913                 err = -ENOBUFS;
1914                 pn = pneigh_lookup(tbl, net, dst, dev, 1);
1915                 if (pn) {
1916                         pn->flags = ndm->ndm_flags;
1917                         if (protocol)
1918                                 pn->protocol = protocol;
1919                         err = 0;
1920                 }
1921                 goto out;
1922         }
1923
1924         if (!dev) {
1925                 NL_SET_ERR_MSG(extack, "Device not specified");
1926                 goto out;
1927         }
1928
1929         if (tbl->allow_add && !tbl->allow_add(dev, extack)) {
1930                 err = -EINVAL;
1931                 goto out;
1932         }
1933
1934         neigh = neigh_lookup(tbl, dst, dev);
1935         if (neigh == NULL) {
1936                 bool exempt_from_gc;
1937
1938                 if (!(nlh->nlmsg_flags & NLM_F_CREATE)) {
1939                         err = -ENOENT;
1940                         goto out;
1941                 }
1942
1943                 exempt_from_gc = ndm->ndm_state & NUD_PERMANENT ||
1944                                  ndm->ndm_flags & NTF_EXT_LEARNED;
1945                 neigh = ___neigh_create(tbl, dst, dev, exempt_from_gc, true);
1946                 if (IS_ERR(neigh)) {
1947                         err = PTR_ERR(neigh);
1948                         goto out;
1949                 }
1950         } else {
1951                 if (nlh->nlmsg_flags & NLM_F_EXCL) {
1952                         err = -EEXIST;
1953                         neigh_release(neigh);
1954                         goto out;
1955                 }
1956
1957                 if (!(nlh->nlmsg_flags & NLM_F_REPLACE))
1958                         flags &= ~(NEIGH_UPDATE_F_OVERRIDE |
1959                                    NEIGH_UPDATE_F_OVERRIDE_ISROUTER);
1960         }
1961
1962         if (protocol)
1963                 neigh->protocol = protocol;
1964
1965         if (ndm->ndm_flags & NTF_EXT_LEARNED)
1966                 flags |= NEIGH_UPDATE_F_EXT_LEARNED;
1967
1968         if (ndm->ndm_flags & NTF_ROUTER)
1969                 flags |= NEIGH_UPDATE_F_ISROUTER;
1970
1971         if (ndm->ndm_flags & NTF_USE) {
1972                 neigh_event_send(neigh, NULL);
1973                 err = 0;
1974         } else
1975                 err = __neigh_update(neigh, lladdr, ndm->ndm_state, flags,
1976                                      NETLINK_CB(skb).portid, extack);
1977
1978         neigh_release(neigh);
1979
1980 out:
1981         return err;
1982 }
1983
1984 static int neightbl_fill_parms(struct sk_buff *skb, struct neigh_parms *parms)
1985 {
1986         struct nlattr *nest;
1987
1988         nest = nla_nest_start_noflag(skb, NDTA_PARMS);
1989         if (nest == NULL)
1990                 return -ENOBUFS;
1991
1992         if ((parms->dev &&
1993              nla_put_u32(skb, NDTPA_IFINDEX, parms->dev->ifindex)) ||
1994             nla_put_u32(skb, NDTPA_REFCNT, refcount_read(&parms->refcnt)) ||
1995             nla_put_u32(skb, NDTPA_QUEUE_LENBYTES,
1996                         NEIGH_VAR(parms, QUEUE_LEN_BYTES)) ||
1997             /* approximative value for deprecated QUEUE_LEN (in packets) */
1998             nla_put_u32(skb, NDTPA_QUEUE_LEN,
1999                         NEIGH_VAR(parms, QUEUE_LEN_BYTES) / SKB_TRUESIZE(ETH_FRAME_LEN)) ||
2000             nla_put_u32(skb, NDTPA_PROXY_QLEN, NEIGH_VAR(parms, PROXY_QLEN)) ||
2001             nla_put_u32(skb, NDTPA_APP_PROBES, NEIGH_VAR(parms, APP_PROBES)) ||
2002             nla_put_u32(skb, NDTPA_UCAST_PROBES,
2003                         NEIGH_VAR(parms, UCAST_PROBES)) ||
2004             nla_put_u32(skb, NDTPA_MCAST_PROBES,
2005                         NEIGH_VAR(parms, MCAST_PROBES)) ||
2006             nla_put_u32(skb, NDTPA_MCAST_REPROBES,
2007                         NEIGH_VAR(parms, MCAST_REPROBES)) ||
2008             nla_put_msecs(skb, NDTPA_REACHABLE_TIME, parms->reachable_time,
2009                           NDTPA_PAD) ||
2010             nla_put_msecs(skb, NDTPA_BASE_REACHABLE_TIME,
2011                           NEIGH_VAR(parms, BASE_REACHABLE_TIME), NDTPA_PAD) ||
2012             nla_put_msecs(skb, NDTPA_GC_STALETIME,
2013                           NEIGH_VAR(parms, GC_STALETIME), NDTPA_PAD) ||
2014             nla_put_msecs(skb, NDTPA_DELAY_PROBE_TIME,
2015                           NEIGH_VAR(parms, DELAY_PROBE_TIME), NDTPA_PAD) ||
2016             nla_put_msecs(skb, NDTPA_RETRANS_TIME,
2017                           NEIGH_VAR(parms, RETRANS_TIME), NDTPA_PAD) ||
2018             nla_put_msecs(skb, NDTPA_ANYCAST_DELAY,
2019                           NEIGH_VAR(parms, ANYCAST_DELAY), NDTPA_PAD) ||
2020             nla_put_msecs(skb, NDTPA_PROXY_DELAY,
2021                           NEIGH_VAR(parms, PROXY_DELAY), NDTPA_PAD) ||
2022             nla_put_msecs(skb, NDTPA_LOCKTIME,
2023                           NEIGH_VAR(parms, LOCKTIME), NDTPA_PAD))
2024                 goto nla_put_failure;
2025         return nla_nest_end(skb, nest);
2026
2027 nla_put_failure:
2028         nla_nest_cancel(skb, nest);
2029         return -EMSGSIZE;
2030 }
2031
2032 static int neightbl_fill_info(struct sk_buff *skb, struct neigh_table *tbl,
2033                               u32 pid, u32 seq, int type, int flags)
2034 {
2035         struct nlmsghdr *nlh;
2036         struct ndtmsg *ndtmsg;
2037
2038         nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
2039         if (nlh == NULL)
2040                 return -EMSGSIZE;
2041
2042         ndtmsg = nlmsg_data(nlh);
2043
2044         read_lock_bh(&tbl->lock);
2045         ndtmsg->ndtm_family = tbl->family;
2046         ndtmsg->ndtm_pad1   = 0;
2047         ndtmsg->ndtm_pad2   = 0;
2048
2049         if (nla_put_string(skb, NDTA_NAME, tbl->id) ||
2050             nla_put_msecs(skb, NDTA_GC_INTERVAL, tbl->gc_interval, NDTA_PAD) ||
2051             nla_put_u32(skb, NDTA_THRESH1, tbl->gc_thresh1) ||
2052             nla_put_u32(skb, NDTA_THRESH2, tbl->gc_thresh2) ||
2053             nla_put_u32(skb, NDTA_THRESH3, tbl->gc_thresh3))
2054                 goto nla_put_failure;
2055         {
2056                 unsigned long now = jiffies;
2057                 long flush_delta = now - tbl->last_flush;
2058                 long rand_delta = now - tbl->last_rand;
2059                 struct neigh_hash_table *nht;
2060                 struct ndt_config ndc = {
2061                         .ndtc_key_len           = tbl->key_len,
2062                         .ndtc_entry_size        = tbl->entry_size,
2063                         .ndtc_entries           = atomic_read(&tbl->entries),
2064                         .ndtc_last_flush        = jiffies_to_msecs(flush_delta),
2065                         .ndtc_last_rand         = jiffies_to_msecs(rand_delta),
2066                         .ndtc_proxy_qlen        = tbl->proxy_queue.qlen,
2067                 };
2068
2069                 rcu_read_lock_bh();
2070                 nht = rcu_dereference_bh(tbl->nht);
2071                 ndc.ndtc_hash_rnd = nht->hash_rnd[0];
2072                 ndc.ndtc_hash_mask = ((1 << nht->hash_shift) - 1);
2073                 rcu_read_unlock_bh();
2074
2075                 if (nla_put(skb, NDTA_CONFIG, sizeof(ndc), &ndc))
2076                         goto nla_put_failure;
2077         }
2078
2079         {
2080                 int cpu;
2081                 struct ndt_stats ndst;
2082
2083                 memset(&ndst, 0, sizeof(ndst));
2084
2085                 for_each_possible_cpu(cpu) {
2086                         struct neigh_statistics *st;
2087
2088                         st = per_cpu_ptr(tbl->stats, cpu);
2089                         ndst.ndts_allocs                += st->allocs;
2090                         ndst.ndts_destroys              += st->destroys;
2091                         ndst.ndts_hash_grows            += st->hash_grows;
2092                         ndst.ndts_res_failed            += st->res_failed;
2093                         ndst.ndts_lookups               += st->lookups;
2094                         ndst.ndts_hits                  += st->hits;
2095                         ndst.ndts_rcv_probes_mcast      += st->rcv_probes_mcast;
2096                         ndst.ndts_rcv_probes_ucast      += st->rcv_probes_ucast;
2097                         ndst.ndts_periodic_gc_runs      += st->periodic_gc_runs;
2098                         ndst.ndts_forced_gc_runs        += st->forced_gc_runs;
2099                         ndst.ndts_table_fulls           += st->table_fulls;
2100                 }
2101
2102                 if (nla_put_64bit(skb, NDTA_STATS, sizeof(ndst), &ndst,
2103                                   NDTA_PAD))
2104                         goto nla_put_failure;
2105         }
2106
2107         BUG_ON(tbl->parms.dev);
2108         if (neightbl_fill_parms(skb, &tbl->parms) < 0)
2109                 goto nla_put_failure;
2110
2111         read_unlock_bh(&tbl->lock);
2112         nlmsg_end(skb, nlh);
2113         return 0;
2114
2115 nla_put_failure:
2116         read_unlock_bh(&tbl->lock);
2117         nlmsg_cancel(skb, nlh);
2118         return -EMSGSIZE;
2119 }
2120
2121 static int neightbl_fill_param_info(struct sk_buff *skb,
2122                                     struct neigh_table *tbl,
2123                                     struct neigh_parms *parms,
2124                                     u32 pid, u32 seq, int type,
2125                                     unsigned int flags)
2126 {
2127         struct ndtmsg *ndtmsg;
2128         struct nlmsghdr *nlh;
2129
2130         nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
2131         if (nlh == NULL)
2132                 return -EMSGSIZE;
2133
2134         ndtmsg = nlmsg_data(nlh);
2135
2136         read_lock_bh(&tbl->lock);
2137         ndtmsg->ndtm_family = tbl->family;
2138         ndtmsg->ndtm_pad1   = 0;
2139         ndtmsg->ndtm_pad2   = 0;
2140
2141         if (nla_put_string(skb, NDTA_NAME, tbl->id) < 0 ||
2142             neightbl_fill_parms(skb, parms) < 0)
2143                 goto errout;
2144
2145         read_unlock_bh(&tbl->lock);
2146         nlmsg_end(skb, nlh);
2147         return 0;
2148 errout:
2149         read_unlock_bh(&tbl->lock);
2150         nlmsg_cancel(skb, nlh);
2151         return -EMSGSIZE;
2152 }
2153
2154 static const struct nla_policy nl_neightbl_policy[NDTA_MAX+1] = {
2155         [NDTA_NAME]             = { .type = NLA_STRING },
2156         [NDTA_THRESH1]          = { .type = NLA_U32 },
2157         [NDTA_THRESH2]          = { .type = NLA_U32 },
2158         [NDTA_THRESH3]          = { .type = NLA_U32 },
2159         [NDTA_GC_INTERVAL]      = { .type = NLA_U64 },
2160         [NDTA_PARMS]            = { .type = NLA_NESTED },
2161 };
2162
2163 static const struct nla_policy nl_ntbl_parm_policy[NDTPA_MAX+1] = {
2164         [NDTPA_IFINDEX]                 = { .type = NLA_U32 },
2165         [NDTPA_QUEUE_LEN]               = { .type = NLA_U32 },
2166         [NDTPA_PROXY_QLEN]              = { .type = NLA_U32 },
2167         [NDTPA_APP_PROBES]              = { .type = NLA_U32 },
2168         [NDTPA_UCAST_PROBES]            = { .type = NLA_U32 },
2169         [NDTPA_MCAST_PROBES]            = { .type = NLA_U32 },
2170         [NDTPA_MCAST_REPROBES]          = { .type = NLA_U32 },
2171         [NDTPA_BASE_REACHABLE_TIME]     = { .type = NLA_U64 },
2172         [NDTPA_GC_STALETIME]            = { .type = NLA_U64 },
2173         [NDTPA_DELAY_PROBE_TIME]        = { .type = NLA_U64 },
2174         [NDTPA_RETRANS_TIME]            = { .type = NLA_U64 },
2175         [NDTPA_ANYCAST_DELAY]           = { .type = NLA_U64 },
2176         [NDTPA_PROXY_DELAY]             = { .type = NLA_U64 },
2177         [NDTPA_LOCKTIME]                = { .type = NLA_U64 },
2178 };
2179
2180 static int neightbl_set(struct sk_buff *skb, struct nlmsghdr *nlh,
2181                         struct netlink_ext_ack *extack)
2182 {
2183         struct net *net = sock_net(skb->sk);
2184         struct neigh_table *tbl;
2185         struct ndtmsg *ndtmsg;
2186         struct nlattr *tb[NDTA_MAX+1];
2187         bool found = false;
2188         int err, tidx;
2189
2190         err = nlmsg_parse_deprecated(nlh, sizeof(*ndtmsg), tb, NDTA_MAX,
2191                                      nl_neightbl_policy, extack);
2192         if (err < 0)
2193                 goto errout;
2194
2195         if (tb[NDTA_NAME] == NULL) {
2196                 err = -EINVAL;
2197                 goto errout;
2198         }
2199
2200         ndtmsg = nlmsg_data(nlh);
2201
2202         for (tidx = 0; tidx < NEIGH_NR_TABLES; tidx++) {
2203                 tbl = neigh_tables[tidx];
2204                 if (!tbl)
2205                         continue;
2206                 if (ndtmsg->ndtm_family && tbl->family != ndtmsg->ndtm_family)
2207                         continue;
2208                 if (nla_strcmp(tb[NDTA_NAME], tbl->id) == 0) {
2209                         found = true;
2210                         break;
2211                 }
2212         }
2213
2214         if (!found)
2215                 return -ENOENT;
2216
2217         /*
2218          * We acquire tbl->lock to be nice to the periodic timers and
2219          * make sure they always see a consistent set of values.
2220          */
2221         write_lock_bh(&tbl->lock);
2222
2223         if (tb[NDTA_PARMS]) {
2224                 struct nlattr *tbp[NDTPA_MAX+1];
2225                 struct neigh_parms *p;
2226                 int i, ifindex = 0;
2227
2228                 err = nla_parse_nested_deprecated(tbp, NDTPA_MAX,
2229                                                   tb[NDTA_PARMS],
2230                                                   nl_ntbl_parm_policy, extack);
2231                 if (err < 0)
2232                         goto errout_tbl_lock;
2233
2234                 if (tbp[NDTPA_IFINDEX])
2235                         ifindex = nla_get_u32(tbp[NDTPA_IFINDEX]);
2236
2237                 p = lookup_neigh_parms(tbl, net, ifindex);
2238                 if (p == NULL) {
2239                         err = -ENOENT;
2240                         goto errout_tbl_lock;
2241                 }
2242
2243                 for (i = 1; i <= NDTPA_MAX; i++) {
2244                         if (tbp[i] == NULL)
2245                                 continue;
2246
2247                         switch (i) {
2248                         case NDTPA_QUEUE_LEN:
2249                                 NEIGH_VAR_SET(p, QUEUE_LEN_BYTES,
2250                                               nla_get_u32(tbp[i]) *
2251                                               SKB_TRUESIZE(ETH_FRAME_LEN));
2252                                 break;
2253                         case NDTPA_QUEUE_LENBYTES:
2254                                 NEIGH_VAR_SET(p, QUEUE_LEN_BYTES,
2255                                               nla_get_u32(tbp[i]));
2256                                 break;
2257                         case NDTPA_PROXY_QLEN:
2258                                 NEIGH_VAR_SET(p, PROXY_QLEN,
2259                                               nla_get_u32(tbp[i]));
2260                                 break;
2261                         case NDTPA_APP_PROBES:
2262                                 NEIGH_VAR_SET(p, APP_PROBES,
2263                                               nla_get_u32(tbp[i]));
2264                                 break;
2265                         case NDTPA_UCAST_PROBES:
2266                                 NEIGH_VAR_SET(p, UCAST_PROBES,
2267                                               nla_get_u32(tbp[i]));
2268                                 break;
2269                         case NDTPA_MCAST_PROBES:
2270                                 NEIGH_VAR_SET(p, MCAST_PROBES,
2271                                               nla_get_u32(tbp[i]));
2272                                 break;
2273                         case NDTPA_MCAST_REPROBES:
2274                                 NEIGH_VAR_SET(p, MCAST_REPROBES,
2275                                               nla_get_u32(tbp[i]));
2276                                 break;
2277                         case NDTPA_BASE_REACHABLE_TIME:
2278                                 NEIGH_VAR_SET(p, BASE_REACHABLE_TIME,
2279                                               nla_get_msecs(tbp[i]));
2280                                 /* update reachable_time as well, otherwise, the change will
2281                                  * only be effective after the next time neigh_periodic_work
2282                                  * decides to recompute it (can be multiple minutes)
2283                                  */
2284                                 p->reachable_time =
2285                                         neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
2286                                 break;
2287                         case NDTPA_GC_STALETIME:
2288                                 NEIGH_VAR_SET(p, GC_STALETIME,
2289                                               nla_get_msecs(tbp[i]));
2290                                 break;
2291                         case NDTPA_DELAY_PROBE_TIME:
2292                                 NEIGH_VAR_SET(p, DELAY_PROBE_TIME,
2293                                               nla_get_msecs(tbp[i]));
2294                                 call_netevent_notifiers(NETEVENT_DELAY_PROBE_TIME_UPDATE, p);
2295                                 break;
2296                         case NDTPA_RETRANS_TIME:
2297                                 NEIGH_VAR_SET(p, RETRANS_TIME,
2298                                               nla_get_msecs(tbp[i]));
2299                                 break;
2300                         case NDTPA_ANYCAST_DELAY:
2301                                 NEIGH_VAR_SET(p, ANYCAST_DELAY,
2302                                               nla_get_msecs(tbp[i]));
2303                                 break;
2304                         case NDTPA_PROXY_DELAY:
2305                                 NEIGH_VAR_SET(p, PROXY_DELAY,
2306                                               nla_get_msecs(tbp[i]));
2307                                 break;
2308                         case NDTPA_LOCKTIME:
2309                                 NEIGH_VAR_SET(p, LOCKTIME,
2310                                               nla_get_msecs(tbp[i]));
2311                                 break;
2312                         }
2313                 }
2314         }
2315
2316         err = -ENOENT;
2317         if ((tb[NDTA_THRESH1] || tb[NDTA_THRESH2] ||
2318              tb[NDTA_THRESH3] || tb[NDTA_GC_INTERVAL]) &&
2319             !net_eq(net, &init_net))
2320                 goto errout_tbl_lock;
2321
2322         if (tb[NDTA_THRESH1])
2323                 tbl->gc_thresh1 = nla_get_u32(tb[NDTA_THRESH1]);
2324
2325         if (tb[NDTA_THRESH2])
2326                 tbl->gc_thresh2 = nla_get_u32(tb[NDTA_THRESH2]);
2327
2328         if (tb[NDTA_THRESH3])
2329                 tbl->gc_thresh3 = nla_get_u32(tb[NDTA_THRESH3]);
2330
2331         if (tb[NDTA_GC_INTERVAL])
2332                 tbl->gc_interval = nla_get_msecs(tb[NDTA_GC_INTERVAL]);
2333
2334         err = 0;
2335
2336 errout_tbl_lock:
2337         write_unlock_bh(&tbl->lock);
2338 errout:
2339         return err;
2340 }
2341
2342 static int neightbl_valid_dump_info(const struct nlmsghdr *nlh,
2343                                     struct netlink_ext_ack *extack)
2344 {
2345         struct ndtmsg *ndtm;
2346
2347         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ndtm))) {
2348                 NL_SET_ERR_MSG(extack, "Invalid header for neighbor table dump request");
2349                 return -EINVAL;
2350         }
2351
2352         ndtm = nlmsg_data(nlh);
2353         if (ndtm->ndtm_pad1  || ndtm->ndtm_pad2) {
2354                 NL_SET_ERR_MSG(extack, "Invalid values in header for neighbor table dump request");
2355                 return -EINVAL;
2356         }
2357
2358         if (nlmsg_attrlen(nlh, sizeof(*ndtm))) {
2359                 NL_SET_ERR_MSG(extack, "Invalid data after header in neighbor table dump request");
2360                 return -EINVAL;
2361         }
2362
2363         return 0;
2364 }
2365
2366 static int neightbl_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2367 {
2368         const struct nlmsghdr *nlh = cb->nlh;
2369         struct net *net = sock_net(skb->sk);
2370         int family, tidx, nidx = 0;
2371         int tbl_skip = cb->args[0];
2372         int neigh_skip = cb->args[1];
2373         struct neigh_table *tbl;
2374
2375         if (cb->strict_check) {
2376                 int err = neightbl_valid_dump_info(nlh, cb->extack);
2377
2378                 if (err < 0)
2379                         return err;
2380         }
2381
2382         family = ((struct rtgenmsg *)nlmsg_data(nlh))->rtgen_family;
2383
2384         for (tidx = 0; tidx < NEIGH_NR_TABLES; tidx++) {
2385                 struct neigh_parms *p;
2386
2387                 tbl = neigh_tables[tidx];
2388                 if (!tbl)
2389                         continue;
2390
2391                 if (tidx < tbl_skip || (family && tbl->family != family))
2392                         continue;
2393
2394                 if (neightbl_fill_info(skb, tbl, NETLINK_CB(cb->skb).portid,
2395                                        nlh->nlmsg_seq, RTM_NEWNEIGHTBL,
2396                                        NLM_F_MULTI) < 0)
2397                         break;
2398
2399                 nidx = 0;
2400                 p = list_next_entry(&tbl->parms, list);
2401                 list_for_each_entry_from(p, &tbl->parms_list, list) {
2402                         if (!net_eq(neigh_parms_net(p), net))
2403                                 continue;
2404
2405                         if (nidx < neigh_skip)
2406                                 goto next;
2407
2408                         if (neightbl_fill_param_info(skb, tbl, p,
2409                                                      NETLINK_CB(cb->skb).portid,
2410                                                      nlh->nlmsg_seq,
2411                                                      RTM_NEWNEIGHTBL,
2412                                                      NLM_F_MULTI) < 0)
2413                                 goto out;
2414                 next:
2415                         nidx++;
2416                 }
2417
2418                 neigh_skip = 0;
2419         }
2420 out:
2421         cb->args[0] = tidx;
2422         cb->args[1] = nidx;
2423
2424         return skb->len;
2425 }
2426
2427 static int neigh_fill_info(struct sk_buff *skb, struct neighbour *neigh,
2428                            u32 pid, u32 seq, int type, unsigned int flags)
2429 {
2430         unsigned long now = jiffies;
2431         struct nda_cacheinfo ci;
2432         struct nlmsghdr *nlh;
2433         struct ndmsg *ndm;
2434
2435         nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2436         if (nlh == NULL)
2437                 return -EMSGSIZE;
2438
2439         ndm = nlmsg_data(nlh);
2440         ndm->ndm_family  = neigh->ops->family;
2441         ndm->ndm_pad1    = 0;
2442         ndm->ndm_pad2    = 0;
2443         ndm->ndm_flags   = neigh->flags;
2444         ndm->ndm_type    = neigh->type;
2445         ndm->ndm_ifindex = neigh->dev->ifindex;
2446
2447         if (nla_put(skb, NDA_DST, neigh->tbl->key_len, neigh->primary_key))
2448                 goto nla_put_failure;
2449
2450         read_lock_bh(&neigh->lock);
2451         ndm->ndm_state   = neigh->nud_state;
2452         if (neigh->nud_state & NUD_VALID) {
2453                 char haddr[MAX_ADDR_LEN];
2454
2455                 neigh_ha_snapshot(haddr, neigh, neigh->dev);
2456                 if (nla_put(skb, NDA_LLADDR, neigh->dev->addr_len, haddr) < 0) {
2457                         read_unlock_bh(&neigh->lock);
2458                         goto nla_put_failure;
2459                 }
2460         }
2461
2462         ci.ndm_used      = jiffies_to_clock_t(now - neigh->used);
2463         ci.ndm_confirmed = jiffies_to_clock_t(now - neigh->confirmed);
2464         ci.ndm_updated   = jiffies_to_clock_t(now - neigh->updated);
2465         ci.ndm_refcnt    = refcount_read(&neigh->refcnt) - 1;
2466         read_unlock_bh(&neigh->lock);
2467
2468         if (nla_put_u32(skb, NDA_PROBES, atomic_read(&neigh->probes)) ||
2469             nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
2470                 goto nla_put_failure;
2471
2472         if (neigh->protocol && nla_put_u8(skb, NDA_PROTOCOL, neigh->protocol))
2473                 goto nla_put_failure;
2474
2475         nlmsg_end(skb, nlh);
2476         return 0;
2477
2478 nla_put_failure:
2479         nlmsg_cancel(skb, nlh);
2480         return -EMSGSIZE;
2481 }
2482
2483 static int pneigh_fill_info(struct sk_buff *skb, struct pneigh_entry *pn,
2484                             u32 pid, u32 seq, int type, unsigned int flags,
2485                             struct neigh_table *tbl)
2486 {
2487         struct nlmsghdr *nlh;
2488         struct ndmsg *ndm;
2489
2490         nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2491         if (nlh == NULL)
2492                 return -EMSGSIZE;
2493
2494         ndm = nlmsg_data(nlh);
2495         ndm->ndm_family  = tbl->family;
2496         ndm->ndm_pad1    = 0;
2497         ndm->ndm_pad2    = 0;
2498         ndm->ndm_flags   = pn->flags | NTF_PROXY;
2499         ndm->ndm_type    = RTN_UNICAST;
2500         ndm->ndm_ifindex = pn->dev ? pn->dev->ifindex : 0;
2501         ndm->ndm_state   = NUD_NONE;
2502
2503         if (nla_put(skb, NDA_DST, tbl->key_len, pn->key))
2504                 goto nla_put_failure;
2505
2506         if (pn->protocol && nla_put_u8(skb, NDA_PROTOCOL, pn->protocol))
2507                 goto nla_put_failure;
2508
2509         nlmsg_end(skb, nlh);
2510         return 0;
2511
2512 nla_put_failure:
2513         nlmsg_cancel(skb, nlh);
2514         return -EMSGSIZE;
2515 }
2516
2517 static void neigh_update_notify(struct neighbour *neigh, u32 nlmsg_pid)
2518 {
2519         call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, neigh);
2520         __neigh_notify(neigh, RTM_NEWNEIGH, 0, nlmsg_pid);
2521 }
2522
2523 static bool neigh_master_filtered(struct net_device *dev, int master_idx)
2524 {
2525         struct net_device *master;
2526
2527         if (!master_idx)
2528                 return false;
2529
2530         master = dev ? netdev_master_upper_dev_get(dev) : NULL;
2531         if (!master || master->ifindex != master_idx)
2532                 return true;
2533
2534         return false;
2535 }
2536
2537 static bool neigh_ifindex_filtered(struct net_device *dev, int filter_idx)
2538 {
2539         if (filter_idx && (!dev || dev->ifindex != filter_idx))
2540                 return true;
2541
2542         return false;
2543 }
2544
2545 struct neigh_dump_filter {
2546         int master_idx;
2547         int dev_idx;
2548 };
2549
2550 static int neigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2551                             struct netlink_callback *cb,
2552                             struct neigh_dump_filter *filter)
2553 {
2554         struct net *net = sock_net(skb->sk);
2555         struct neighbour *n;
2556         int rc, h, s_h = cb->args[1];
2557         int idx, s_idx = idx = cb->args[2];
2558         struct neigh_hash_table *nht;
2559         unsigned int flags = NLM_F_MULTI;
2560
2561         if (filter->dev_idx || filter->master_idx)
2562                 flags |= NLM_F_DUMP_FILTERED;
2563
2564         rcu_read_lock_bh();
2565         nht = rcu_dereference_bh(tbl->nht);
2566
2567         for (h = s_h; h < (1 << nht->hash_shift); h++) {
2568                 if (h > s_h)
2569                         s_idx = 0;
2570                 for (n = rcu_dereference_bh(nht->hash_buckets[h]), idx = 0;
2571                      n != NULL;
2572                      n = rcu_dereference_bh(n->next)) {
2573                         if (idx < s_idx || !net_eq(dev_net(n->dev), net))
2574                                 goto next;
2575                         if (neigh_ifindex_filtered(n->dev, filter->dev_idx) ||
2576                             neigh_master_filtered(n->dev, filter->master_idx))
2577                                 goto next;
2578                         if (neigh_fill_info(skb, n, NETLINK_CB(cb->skb).portid,
2579                                             cb->nlh->nlmsg_seq,
2580                                             RTM_NEWNEIGH,
2581                                             flags) < 0) {
2582                                 rc = -1;
2583                                 goto out;
2584                         }
2585 next:
2586                         idx++;
2587                 }
2588         }
2589         rc = skb->len;
2590 out:
2591         rcu_read_unlock_bh();
2592         cb->args[1] = h;
2593         cb->args[2] = idx;
2594         return rc;
2595 }
2596
2597 static int pneigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2598                              struct netlink_callback *cb,
2599                              struct neigh_dump_filter *filter)
2600 {
2601         struct pneigh_entry *n;
2602         struct net *net = sock_net(skb->sk);
2603         int rc, h, s_h = cb->args[3];
2604         int idx, s_idx = idx = cb->args[4];
2605         unsigned int flags = NLM_F_MULTI;
2606
2607         if (filter->dev_idx || filter->master_idx)
2608                 flags |= NLM_F_DUMP_FILTERED;
2609
2610         read_lock_bh(&tbl->lock);
2611
2612         for (h = s_h; h <= PNEIGH_HASHMASK; h++) {
2613                 if (h > s_h)
2614                         s_idx = 0;
2615                 for (n = tbl->phash_buckets[h], idx = 0; n; n = n->next) {
2616                         if (idx < s_idx || pneigh_net(n) != net)
2617                                 goto next;
2618                         if (neigh_ifindex_filtered(n->dev, filter->dev_idx) ||
2619                             neigh_master_filtered(n->dev, filter->master_idx))
2620                                 goto next;
2621                         if (pneigh_fill_info(skb, n, NETLINK_CB(cb->skb).portid,
2622                                             cb->nlh->nlmsg_seq,
2623                                             RTM_NEWNEIGH, flags, tbl) < 0) {
2624                                 read_unlock_bh(&tbl->lock);
2625                                 rc = -1;
2626                                 goto out;
2627                         }
2628                 next:
2629                         idx++;
2630                 }
2631         }
2632
2633         read_unlock_bh(&tbl->lock);
2634         rc = skb->len;
2635 out:
2636         cb->args[3] = h;
2637         cb->args[4] = idx;
2638         return rc;
2639
2640 }
2641
2642 static int neigh_valid_dump_req(const struct nlmsghdr *nlh,
2643                                 bool strict_check,
2644                                 struct neigh_dump_filter *filter,
2645                                 struct netlink_ext_ack *extack)
2646 {
2647         struct nlattr *tb[NDA_MAX + 1];
2648         int err, i;
2649
2650         if (strict_check) {
2651                 struct ndmsg *ndm;
2652
2653                 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ndm))) {
2654                         NL_SET_ERR_MSG(extack, "Invalid header for neighbor dump request");
2655                         return -EINVAL;
2656                 }
2657
2658                 ndm = nlmsg_data(nlh);
2659                 if (ndm->ndm_pad1  || ndm->ndm_pad2  || ndm->ndm_ifindex ||
2660                     ndm->ndm_state || ndm->ndm_type) {
2661                         NL_SET_ERR_MSG(extack, "Invalid values in header for neighbor dump request");
2662                         return -EINVAL;
2663                 }
2664
2665                 if (ndm->ndm_flags & ~NTF_PROXY) {
2666                         NL_SET_ERR_MSG(extack, "Invalid flags in header for neighbor dump request");
2667                         return -EINVAL;
2668                 }
2669
2670                 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct ndmsg),
2671                                                     tb, NDA_MAX, nda_policy,
2672                                                     extack);
2673         } else {
2674                 err = nlmsg_parse_deprecated(nlh, sizeof(struct ndmsg), tb,
2675                                              NDA_MAX, nda_policy, extack);
2676         }
2677         if (err < 0)
2678                 return err;
2679
2680         for (i = 0; i <= NDA_MAX; ++i) {
2681                 if (!tb[i])
2682                         continue;
2683
2684                 /* all new attributes should require strict_check */
2685                 switch (i) {
2686                 case NDA_IFINDEX:
2687                         filter->dev_idx = nla_get_u32(tb[i]);
2688                         break;
2689                 case NDA_MASTER:
2690                         filter->master_idx = nla_get_u32(tb[i]);
2691                         break;
2692                 default:
2693                         if (strict_check) {
2694                                 NL_SET_ERR_MSG(extack, "Unsupported attribute in neighbor dump request");
2695                                 return -EINVAL;
2696                         }
2697                 }
2698         }
2699
2700         return 0;
2701 }
2702
2703 static int neigh_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2704 {
2705         const struct nlmsghdr *nlh = cb->nlh;
2706         struct neigh_dump_filter filter = {};
2707         struct neigh_table *tbl;
2708         int t, family, s_t;
2709         int proxy = 0;
2710         int err;
2711
2712         family = ((struct rtgenmsg *)nlmsg_data(nlh))->rtgen_family;
2713
2714         /* check for full ndmsg structure presence, family member is
2715          * the same for both structures
2716          */
2717         if (nlmsg_len(nlh) >= sizeof(struct ndmsg) &&
2718             ((struct ndmsg *)nlmsg_data(nlh))->ndm_flags == NTF_PROXY)
2719                 proxy = 1;
2720
2721         err = neigh_valid_dump_req(nlh, cb->strict_check, &filter, cb->extack);
2722         if (err < 0 && cb->strict_check)
2723                 return err;
2724
2725         s_t = cb->args[0];
2726
2727         for (t = 0; t < NEIGH_NR_TABLES; t++) {
2728                 tbl = neigh_tables[t];
2729
2730                 if (!tbl)
2731                         continue;
2732                 if (t < s_t || (family && tbl->family != family))
2733                         continue;
2734                 if (t > s_t)
2735                         memset(&cb->args[1], 0, sizeof(cb->args) -
2736                                                 sizeof(cb->args[0]));
2737                 if (proxy)
2738                         err = pneigh_dump_table(tbl, skb, cb, &filter);
2739                 else
2740                         err = neigh_dump_table(tbl, skb, cb, &filter);
2741                 if (err < 0)
2742                         break;
2743         }
2744
2745         cb->args[0] = t;
2746         return skb->len;
2747 }
2748
2749 static int neigh_valid_get_req(const struct nlmsghdr *nlh,
2750                                struct neigh_table **tbl,
2751                                void **dst, int *dev_idx, u8 *ndm_flags,
2752                                struct netlink_ext_ack *extack)
2753 {
2754         struct nlattr *tb[NDA_MAX + 1];
2755         struct ndmsg *ndm;
2756         int err, i;
2757
2758         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ndm))) {
2759                 NL_SET_ERR_MSG(extack, "Invalid header for neighbor get request");
2760                 return -EINVAL;
2761         }
2762
2763         ndm = nlmsg_data(nlh);
2764         if (ndm->ndm_pad1  || ndm->ndm_pad2  || ndm->ndm_state ||
2765             ndm->ndm_type) {
2766                 NL_SET_ERR_MSG(extack, "Invalid values in header for neighbor get request");
2767                 return -EINVAL;
2768         }
2769
2770         if (ndm->ndm_flags & ~NTF_PROXY) {
2771                 NL_SET_ERR_MSG(extack, "Invalid flags in header for neighbor get request");
2772                 return -EINVAL;
2773         }
2774
2775         err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct ndmsg), tb,
2776                                             NDA_MAX, nda_policy, extack);
2777         if (err < 0)
2778                 return err;
2779
2780         *ndm_flags = ndm->ndm_flags;
2781         *dev_idx = ndm->ndm_ifindex;
2782         *tbl = neigh_find_table(ndm->ndm_family);
2783         if (*tbl == NULL) {
2784                 NL_SET_ERR_MSG(extack, "Unsupported family in header for neighbor get request");
2785                 return -EAFNOSUPPORT;
2786         }
2787
2788         for (i = 0; i <= NDA_MAX; ++i) {
2789                 if (!tb[i])
2790                         continue;
2791
2792                 switch (i) {
2793                 case NDA_DST:
2794                         if (nla_len(tb[i]) != (int)(*tbl)->key_len) {
2795                                 NL_SET_ERR_MSG(extack, "Invalid network address in neighbor get request");
2796                                 return -EINVAL;
2797                         }
2798                         *dst = nla_data(tb[i]);
2799                         break;
2800                 default:
2801                         NL_SET_ERR_MSG(extack, "Unsupported attribute in neighbor get request");
2802                         return -EINVAL;
2803                 }
2804         }
2805
2806         return 0;
2807 }
2808
2809 static inline size_t neigh_nlmsg_size(void)
2810 {
2811         return NLMSG_ALIGN(sizeof(struct ndmsg))
2812                + nla_total_size(MAX_ADDR_LEN) /* NDA_DST */
2813                + nla_total_size(MAX_ADDR_LEN) /* NDA_LLADDR */
2814                + nla_total_size(sizeof(struct nda_cacheinfo))
2815                + nla_total_size(4)  /* NDA_PROBES */
2816                + nla_total_size(1); /* NDA_PROTOCOL */
2817 }
2818
2819 static int neigh_get_reply(struct net *net, struct neighbour *neigh,
2820                            u32 pid, u32 seq)
2821 {
2822         struct sk_buff *skb;
2823         int err = 0;
2824
2825         skb = nlmsg_new(neigh_nlmsg_size(), GFP_KERNEL);
2826         if (!skb)
2827                 return -ENOBUFS;
2828
2829         err = neigh_fill_info(skb, neigh, pid, seq, RTM_NEWNEIGH, 0);
2830         if (err) {
2831                 kfree_skb(skb);
2832                 goto errout;
2833         }
2834
2835         err = rtnl_unicast(skb, net, pid);
2836 errout:
2837         return err;
2838 }
2839
2840 static inline size_t pneigh_nlmsg_size(void)
2841 {
2842         return NLMSG_ALIGN(sizeof(struct ndmsg))
2843                + nla_total_size(MAX_ADDR_LEN) /* NDA_DST */
2844                + nla_total_size(1); /* NDA_PROTOCOL */
2845 }
2846
2847 static int pneigh_get_reply(struct net *net, struct pneigh_entry *neigh,
2848                             u32 pid, u32 seq, struct neigh_table *tbl)
2849 {
2850         struct sk_buff *skb;
2851         int err = 0;
2852
2853         skb = nlmsg_new(pneigh_nlmsg_size(), GFP_KERNEL);
2854         if (!skb)
2855                 return -ENOBUFS;
2856
2857         err = pneigh_fill_info(skb, neigh, pid, seq, RTM_NEWNEIGH, 0, tbl);
2858         if (err) {
2859                 kfree_skb(skb);
2860                 goto errout;
2861         }
2862
2863         err = rtnl_unicast(skb, net, pid);
2864 errout:
2865         return err;
2866 }
2867
2868 static int neigh_get(struct sk_buff *in_skb, struct nlmsghdr *nlh,
2869                      struct netlink_ext_ack *extack)
2870 {
2871         struct net *net = sock_net(in_skb->sk);
2872         struct net_device *dev = NULL;
2873         struct neigh_table *tbl = NULL;
2874         struct neighbour *neigh;
2875         void *dst = NULL;
2876         u8 ndm_flags = 0;
2877         int dev_idx = 0;
2878         int err;
2879
2880         err = neigh_valid_get_req(nlh, &tbl, &dst, &dev_idx, &ndm_flags,
2881                                   extack);
2882         if (err < 0)
2883                 return err;
2884
2885         if (dev_idx) {
2886                 dev = __dev_get_by_index(net, dev_idx);
2887                 if (!dev) {
2888                         NL_SET_ERR_MSG(extack, "Unknown device ifindex");
2889                         return -ENODEV;
2890                 }
2891         }
2892
2893         if (!dst) {
2894                 NL_SET_ERR_MSG(extack, "Network address not specified");
2895                 return -EINVAL;
2896         }
2897
2898         if (ndm_flags & NTF_PROXY) {
2899                 struct pneigh_entry *pn;
2900
2901                 pn = pneigh_lookup(tbl, net, dst, dev, 0);
2902                 if (!pn) {
2903                         NL_SET_ERR_MSG(extack, "Proxy neighbour entry not found");
2904                         return -ENOENT;
2905                 }
2906                 return pneigh_get_reply(net, pn, NETLINK_CB(in_skb).portid,
2907                                         nlh->nlmsg_seq, tbl);
2908         }
2909
2910         if (!dev) {
2911                 NL_SET_ERR_MSG(extack, "No device specified");
2912                 return -EINVAL;
2913         }
2914
2915         neigh = neigh_lookup(tbl, dst, dev);
2916         if (!neigh) {
2917                 NL_SET_ERR_MSG(extack, "Neighbour entry not found");
2918                 return -ENOENT;
2919         }
2920
2921         err = neigh_get_reply(net, neigh, NETLINK_CB(in_skb).portid,
2922                               nlh->nlmsg_seq);
2923
2924         neigh_release(neigh);
2925
2926         return err;
2927 }
2928
2929 void neigh_for_each(struct neigh_table *tbl, void (*cb)(struct neighbour *, void *), void *cookie)
2930 {
2931         int chain;
2932         struct neigh_hash_table *nht;
2933
2934         rcu_read_lock_bh();
2935         nht = rcu_dereference_bh(tbl->nht);
2936
2937         read_lock(&tbl->lock); /* avoid resizes */
2938         for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
2939                 struct neighbour *n;
2940
2941                 for (n = rcu_dereference_bh(nht->hash_buckets[chain]);
2942                      n != NULL;
2943                      n = rcu_dereference_bh(n->next))
2944                         cb(n, cookie);
2945         }
2946         read_unlock(&tbl->lock);
2947         rcu_read_unlock_bh();
2948 }
2949 EXPORT_SYMBOL(neigh_for_each);
2950
2951 /* The tbl->lock must be held as a writer and BH disabled. */
2952 void __neigh_for_each_release(struct neigh_table *tbl,
2953                               int (*cb)(struct neighbour *))
2954 {
2955         int chain;
2956         struct neigh_hash_table *nht;
2957
2958         nht = rcu_dereference_protected(tbl->nht,
2959                                         lockdep_is_held(&tbl->lock));
2960         for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
2961                 struct neighbour *n;
2962                 struct neighbour __rcu **np;
2963
2964                 np = &nht->hash_buckets[chain];
2965                 while ((n = rcu_dereference_protected(*np,
2966                                         lockdep_is_held(&tbl->lock))) != NULL) {
2967                         int release;
2968
2969                         write_lock(&n->lock);
2970                         release = cb(n);
2971                         if (release) {
2972                                 rcu_assign_pointer(*np,
2973                                         rcu_dereference_protected(n->next,
2974                                                 lockdep_is_held(&tbl->lock)));
2975                                 neigh_mark_dead(n);
2976                         } else
2977                                 np = &n->next;
2978                         write_unlock(&n->lock);
2979                         if (release)
2980                                 neigh_cleanup_and_release(n);
2981                 }
2982         }
2983 }
2984 EXPORT_SYMBOL(__neigh_for_each_release);
2985
2986 int neigh_xmit(int index, struct net_device *dev,
2987                const void *addr, struct sk_buff *skb)
2988 {
2989         int err = -EAFNOSUPPORT;
2990         if (likely(index < NEIGH_NR_TABLES)) {
2991                 struct neigh_table *tbl;
2992                 struct neighbour *neigh;
2993
2994                 tbl = neigh_tables[index];
2995                 if (!tbl)
2996                         goto out;
2997                 rcu_read_lock_bh();
2998                 if (index == NEIGH_ARP_TABLE) {
2999                         u32 key = *((u32 *)addr);
3000
3001                         neigh = __ipv4_neigh_lookup_noref(dev, key);
3002                 } else {
3003                         neigh = __neigh_lookup_noref(tbl, addr, dev);
3004                 }
3005                 if (!neigh)
3006                         neigh = __neigh_create(tbl, addr, dev, false);
3007                 err = PTR_ERR(neigh);
3008                 if (IS_ERR(neigh)) {
3009                         rcu_read_unlock_bh();
3010                         goto out_kfree_skb;
3011                 }
3012                 err = neigh->output(neigh, skb);
3013                 rcu_read_unlock_bh();
3014         }
3015         else if (index == NEIGH_LINK_TABLE) {
3016                 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
3017                                       addr, NULL, skb->len);
3018                 if (err < 0)
3019                         goto out_kfree_skb;
3020                 err = dev_queue_xmit(skb);
3021         }
3022 out:
3023         return err;
3024 out_kfree_skb:
3025         kfree_skb(skb);
3026         goto out;
3027 }
3028 EXPORT_SYMBOL(neigh_xmit);
3029
3030 #ifdef CONFIG_PROC_FS
3031
3032 static struct neighbour *neigh_get_first(struct seq_file *seq)
3033 {
3034         struct neigh_seq_state *state = seq->private;
3035         struct net *net = seq_file_net(seq);
3036         struct neigh_hash_table *nht = state->nht;
3037         struct neighbour *n = NULL;
3038         int bucket;
3039
3040         state->flags &= ~NEIGH_SEQ_IS_PNEIGH;
3041         for (bucket = 0; bucket < (1 << nht->hash_shift); bucket++) {
3042                 n = rcu_dereference_bh(nht->hash_buckets[bucket]);
3043
3044                 while (n) {
3045                         if (!net_eq(dev_net(n->dev), net))
3046                                 goto next;
3047                         if (state->neigh_sub_iter) {
3048                                 loff_t fakep = 0;
3049                                 void *v;
3050
3051                                 v = state->neigh_sub_iter(state, n, &fakep);
3052                                 if (!v)
3053                                         goto next;
3054                         }
3055                         if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
3056                                 break;
3057                         if (n->nud_state & ~NUD_NOARP)
3058                                 break;
3059 next:
3060                         n = rcu_dereference_bh(n->next);
3061                 }
3062
3063                 if (n)
3064                         break;
3065         }
3066         state->bucket = bucket;
3067
3068         return n;
3069 }
3070
3071 static struct neighbour *neigh_get_next(struct seq_file *seq,
3072                                         struct neighbour *n,
3073                                         loff_t *pos)
3074 {
3075         struct neigh_seq_state *state = seq->private;
3076         struct net *net = seq_file_net(seq);
3077         struct neigh_hash_table *nht = state->nht;
3078
3079         if (state->neigh_sub_iter) {
3080                 void *v = state->neigh_sub_iter(state, n, pos);
3081                 if (v)
3082                         return n;
3083         }
3084         n = rcu_dereference_bh(n->next);
3085
3086         while (1) {
3087                 while (n) {
3088                         if (!net_eq(dev_net(n->dev), net))
3089                                 goto next;
3090                         if (state->neigh_sub_iter) {
3091                                 void *v = state->neigh_sub_iter(state, n, pos);
3092                                 if (v)
3093                                         return n;
3094                                 goto next;
3095                         }
3096                         if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
3097                                 break;
3098
3099                         if (n->nud_state & ~NUD_NOARP)
3100                                 break;
3101 next:
3102                         n = rcu_dereference_bh(n->next);
3103                 }
3104
3105                 if (n)
3106                         break;
3107
3108                 if (++state->bucket >= (1 << nht->hash_shift))
3109                         break;
3110
3111                 n = rcu_dereference_bh(nht->hash_buckets[state->bucket]);
3112         }
3113
3114         if (n && pos)
3115                 --(*pos);
3116         return n;
3117 }
3118
3119 static struct neighbour *neigh_get_idx(struct seq_file *seq, loff_t *pos)
3120 {
3121         struct neighbour *n = neigh_get_first(seq);
3122
3123         if (n) {
3124                 --(*pos);
3125                 while (*pos) {
3126                         n = neigh_get_next(seq, n, pos);
3127                         if (!n)
3128                                 break;
3129                 }
3130         }
3131         return *pos ? NULL : n;
3132 }
3133
3134 static struct pneigh_entry *pneigh_get_first(struct seq_file *seq)
3135 {
3136         struct neigh_seq_state *state = seq->private;
3137         struct net *net = seq_file_net(seq);
3138         struct neigh_table *tbl = state->tbl;
3139         struct pneigh_entry *pn = NULL;
3140         int bucket = state->bucket;
3141
3142         state->flags |= NEIGH_SEQ_IS_PNEIGH;
3143         for (bucket = 0; bucket <= PNEIGH_HASHMASK; bucket++) {
3144                 pn = tbl->phash_buckets[bucket];
3145                 while (pn && !net_eq(pneigh_net(pn), net))
3146                         pn = pn->next;
3147                 if (pn)
3148                         break;
3149         }
3150         state->bucket = bucket;
3151
3152         return pn;
3153 }
3154
3155 static struct pneigh_entry *pneigh_get_next(struct seq_file *seq,
3156                                             struct pneigh_entry *pn,
3157                                             loff_t *pos)
3158 {
3159         struct neigh_seq_state *state = seq->private;
3160         struct net *net = seq_file_net(seq);
3161         struct neigh_table *tbl = state->tbl;
3162
3163         do {
3164                 pn = pn->next;
3165         } while (pn && !net_eq(pneigh_net(pn), net));
3166
3167         while (!pn) {
3168                 if (++state->bucket > PNEIGH_HASHMASK)
3169                         break;
3170                 pn = tbl->phash_buckets[state->bucket];
3171                 while (pn && !net_eq(pneigh_net(pn), net))
3172                         pn = pn->next;
3173                 if (pn)
3174                         break;
3175         }
3176
3177         if (pn && pos)
3178                 --(*pos);
3179
3180         return pn;
3181 }
3182
3183 static struct pneigh_entry *pneigh_get_idx(struct seq_file *seq, loff_t *pos)
3184 {
3185         struct pneigh_entry *pn = pneigh_get_first(seq);
3186
3187         if (pn) {
3188                 --(*pos);
3189                 while (*pos) {
3190                         pn = pneigh_get_next(seq, pn, pos);
3191                         if (!pn)
3192                                 break;
3193                 }
3194         }
3195         return *pos ? NULL : pn;
3196 }
3197
3198 static void *neigh_get_idx_any(struct seq_file *seq, loff_t *pos)
3199 {
3200         struct neigh_seq_state *state = seq->private;
3201         void *rc;
3202         loff_t idxpos = *pos;
3203
3204         rc = neigh_get_idx(seq, &idxpos);
3205         if (!rc && !(state->flags & NEIGH_SEQ_NEIGH_ONLY))
3206                 rc = pneigh_get_idx(seq, &idxpos);
3207
3208         return rc;
3209 }
3210
3211 void *neigh_seq_start(struct seq_file *seq, loff_t *pos, struct neigh_table *tbl, unsigned int neigh_seq_flags)
3212         __acquires(tbl->lock)
3213         __acquires(rcu_bh)
3214 {
3215         struct neigh_seq_state *state = seq->private;
3216
3217         state->tbl = tbl;
3218         state->bucket = 0;
3219         state->flags = (neigh_seq_flags & ~NEIGH_SEQ_IS_PNEIGH);
3220
3221         rcu_read_lock_bh();
3222         state->nht = rcu_dereference_bh(tbl->nht);
3223         read_lock(&tbl->lock);
3224
3225         return *pos ? neigh_get_idx_any(seq, pos) : SEQ_START_TOKEN;
3226 }
3227 EXPORT_SYMBOL(neigh_seq_start);
3228
3229 void *neigh_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3230 {
3231         struct neigh_seq_state *state;
3232         void *rc;
3233
3234         if (v == SEQ_START_TOKEN) {
3235                 rc = neigh_get_first(seq);
3236                 goto out;
3237         }
3238
3239         state = seq->private;
3240         if (!(state->flags & NEIGH_SEQ_IS_PNEIGH)) {
3241                 rc = neigh_get_next(seq, v, NULL);
3242                 if (rc)
3243                         goto out;
3244                 if (!(state->flags & NEIGH_SEQ_NEIGH_ONLY))
3245                         rc = pneigh_get_first(seq);
3246         } else {
3247                 BUG_ON(state->flags & NEIGH_SEQ_NEIGH_ONLY);
3248                 rc = pneigh_get_next(seq, v, NULL);
3249         }
3250 out:
3251         ++(*pos);
3252         return rc;
3253 }
3254 EXPORT_SYMBOL(neigh_seq_next);
3255
3256 void neigh_seq_stop(struct seq_file *seq, void *v)
3257         __releases(tbl->lock)
3258         __releases(rcu_bh)
3259 {
3260         struct neigh_seq_state *state = seq->private;
3261         struct neigh_table *tbl = state->tbl;
3262
3263         read_unlock(&tbl->lock);
3264         rcu_read_unlock_bh();
3265 }
3266 EXPORT_SYMBOL(neigh_seq_stop);
3267
3268 /* statistics via seq_file */
3269
3270 static void *neigh_stat_seq_start(struct seq_file *seq, loff_t *pos)
3271 {
3272         struct neigh_table *tbl = PDE_DATA(file_inode(seq->file));
3273         int cpu;
3274
3275         if (*pos == 0)
3276                 return SEQ_START_TOKEN;
3277
3278         for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
3279                 if (!cpu_possible(cpu))
3280                         continue;
3281                 *pos = cpu+1;
3282                 return per_cpu_ptr(tbl->stats, cpu);
3283         }
3284         return NULL;
3285 }
3286
3287 static void *neigh_stat_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3288 {
3289         struct neigh_table *tbl = PDE_DATA(file_inode(seq->file));
3290         int cpu;
3291
3292         for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
3293                 if (!cpu_possible(cpu))
3294                         continue;
3295                 *pos = cpu+1;
3296                 return per_cpu_ptr(tbl->stats, cpu);
3297         }
3298         (*pos)++;
3299         return NULL;
3300 }
3301
3302 static void neigh_stat_seq_stop(struct seq_file *seq, void *v)
3303 {
3304
3305 }
3306
3307 static int neigh_stat_seq_show(struct seq_file *seq, void *v)
3308 {
3309         struct neigh_table *tbl = PDE_DATA(file_inode(seq->file));
3310         struct neigh_statistics *st = v;
3311
3312         if (v == SEQ_START_TOKEN) {
3313                 seq_printf(seq, "entries  allocs destroys hash_grows  lookups hits  res_failed  rcv_probes_mcast rcv_probes_ucast  periodic_gc_runs forced_gc_runs unresolved_discards table_fulls\n");
3314                 return 0;
3315         }
3316
3317         seq_printf(seq, "%08x  %08lx %08lx %08lx  %08lx %08lx  %08lx  "
3318                         "%08lx %08lx  %08lx %08lx %08lx %08lx\n",
3319                    atomic_read(&tbl->entries),
3320
3321                    st->allocs,
3322                    st->destroys,
3323                    st->hash_grows,
3324
3325                    st->lookups,
3326                    st->hits,
3327
3328                    st->res_failed,
3329
3330                    st->rcv_probes_mcast,
3331                    st->rcv_probes_ucast,
3332
3333                    st->periodic_gc_runs,
3334                    st->forced_gc_runs,
3335                    st->unres_discards,
3336                    st->table_fulls
3337                    );
3338
3339         return 0;
3340 }
3341
3342 static const struct seq_operations neigh_stat_seq_ops = {
3343         .start  = neigh_stat_seq_start,
3344         .next   = neigh_stat_seq_next,
3345         .stop   = neigh_stat_seq_stop,
3346         .show   = neigh_stat_seq_show,
3347 };
3348 #endif /* CONFIG_PROC_FS */
3349
3350 static void __neigh_notify(struct neighbour *n, int type, int flags,
3351                            u32 pid)
3352 {
3353         struct net *net = dev_net(n->dev);
3354         struct sk_buff *skb;
3355         int err = -ENOBUFS;
3356
3357         skb = nlmsg_new(neigh_nlmsg_size(), GFP_ATOMIC);
3358         if (skb == NULL)
3359                 goto errout;
3360
3361         err = neigh_fill_info(skb, n, pid, 0, type, flags);
3362         if (err < 0) {
3363                 /* -EMSGSIZE implies BUG in neigh_nlmsg_size() */
3364                 WARN_ON(err == -EMSGSIZE);
3365                 kfree_skb(skb);
3366                 goto errout;
3367         }
3368         rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
3369         return;
3370 errout:
3371         if (err < 0)
3372                 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
3373 }
3374
3375 void neigh_app_ns(struct neighbour *n)
3376 {
3377         __neigh_notify(n, RTM_GETNEIGH, NLM_F_REQUEST, 0);
3378 }
3379 EXPORT_SYMBOL(neigh_app_ns);
3380
3381 #ifdef CONFIG_SYSCTL
3382 static int unres_qlen_max = INT_MAX / SKB_TRUESIZE(ETH_FRAME_LEN);
3383
3384 static int proc_unres_qlen(struct ctl_table *ctl, int write,
3385                            void *buffer, size_t *lenp, loff_t *ppos)
3386 {
3387         int size, ret;
3388         struct ctl_table tmp = *ctl;
3389
3390         tmp.extra1 = SYSCTL_ZERO;
3391         tmp.extra2 = &unres_qlen_max;
3392         tmp.data = &size;
3393
3394         size = *(int *)ctl->data / SKB_TRUESIZE(ETH_FRAME_LEN);
3395         ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
3396
3397         if (write && !ret)
3398                 *(int *)ctl->data = size * SKB_TRUESIZE(ETH_FRAME_LEN);
3399         return ret;
3400 }
3401
3402 static struct neigh_parms *neigh_get_dev_parms_rcu(struct net_device *dev,
3403                                                    int family)
3404 {
3405         switch (family) {
3406         case AF_INET:
3407                 return __in_dev_arp_parms_get_rcu(dev);
3408         case AF_INET6:
3409                 return __in6_dev_nd_parms_get_rcu(dev);
3410         }
3411         return NULL;
3412 }
3413
3414 static void neigh_copy_dflt_parms(struct net *net, struct neigh_parms *p,
3415                                   int index)
3416 {
3417         struct net_device *dev;
3418         int family = neigh_parms_family(p);
3419
3420         rcu_read_lock();
3421         for_each_netdev_rcu(net, dev) {
3422                 struct neigh_parms *dst_p =
3423                                 neigh_get_dev_parms_rcu(dev, family);
3424
3425                 if (dst_p && !test_bit(index, dst_p->data_state))
3426                         dst_p->data[index] = p->data[index];
3427         }
3428         rcu_read_unlock();
3429 }
3430
3431 static void neigh_proc_update(struct ctl_table *ctl, int write)
3432 {
3433         struct net_device *dev = ctl->extra1;
3434         struct neigh_parms *p = ctl->extra2;
3435         struct net *net = neigh_parms_net(p);
3436         int index = (int *) ctl->data - p->data;
3437
3438         if (!write)
3439                 return;
3440
3441         set_bit(index, p->data_state);
3442         if (index == NEIGH_VAR_DELAY_PROBE_TIME)
3443                 call_netevent_notifiers(NETEVENT_DELAY_PROBE_TIME_UPDATE, p);
3444         if (!dev) /* NULL dev means this is default value */
3445                 neigh_copy_dflt_parms(net, p, index);
3446 }
3447
3448 static int neigh_proc_dointvec_zero_intmax(struct ctl_table *ctl, int write,
3449                                            void *buffer, size_t *lenp,
3450                                            loff_t *ppos)
3451 {
3452         struct ctl_table tmp = *ctl;
3453         int ret;
3454
3455         tmp.extra1 = SYSCTL_ZERO;
3456         tmp.extra2 = SYSCTL_INT_MAX;
3457
3458         ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
3459         neigh_proc_update(ctl, write);
3460         return ret;
3461 }
3462
3463 int neigh_proc_dointvec(struct ctl_table *ctl, int write, void *buffer,
3464                         size_t *lenp, loff_t *ppos)
3465 {
3466         int ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
3467
3468         neigh_proc_update(ctl, write);
3469         return ret;
3470 }
3471 EXPORT_SYMBOL(neigh_proc_dointvec);
3472
3473 int neigh_proc_dointvec_jiffies(struct ctl_table *ctl, int write, void *buffer,
3474                                 size_t *lenp, loff_t *ppos)
3475 {
3476         int ret = proc_dointvec_jiffies(ctl, write, buffer, lenp, ppos);
3477
3478         neigh_proc_update(ctl, write);
3479         return ret;
3480 }
3481 EXPORT_SYMBOL(neigh_proc_dointvec_jiffies);
3482
3483 static int neigh_proc_dointvec_userhz_jiffies(struct ctl_table *ctl, int write,
3484                                               void *buffer, size_t *lenp,
3485                                               loff_t *ppos)
3486 {
3487         int ret = proc_dointvec_userhz_jiffies(ctl, write, buffer, lenp, ppos);
3488
3489         neigh_proc_update(ctl, write);
3490         return ret;
3491 }
3492
3493 int neigh_proc_dointvec_ms_jiffies(struct ctl_table *ctl, int write,
3494                                    void *buffer, size_t *lenp, loff_t *ppos)
3495 {
3496         int ret = proc_dointvec_ms_jiffies(ctl, write, buffer, lenp, ppos);
3497
3498         neigh_proc_update(ctl, write);
3499         return ret;
3500 }
3501 EXPORT_SYMBOL(neigh_proc_dointvec_ms_jiffies);
3502
3503 static int neigh_proc_dointvec_unres_qlen(struct ctl_table *ctl, int write,
3504                                           void *buffer, size_t *lenp,
3505                                           loff_t *ppos)
3506 {
3507         int ret = proc_unres_qlen(ctl, write, buffer, lenp, ppos);
3508
3509         neigh_proc_update(ctl, write);
3510         return ret;
3511 }
3512
3513 static int neigh_proc_base_reachable_time(struct ctl_table *ctl, int write,
3514                                           void *buffer, size_t *lenp,
3515                                           loff_t *ppos)
3516 {
3517         struct neigh_parms *p = ctl->extra2;
3518         int ret;
3519
3520         if (strcmp(ctl->procname, "base_reachable_time") == 0)
3521                 ret = neigh_proc_dointvec_jiffies(ctl, write, buffer, lenp, ppos);
3522         else if (strcmp(ctl->procname, "base_reachable_time_ms") == 0)
3523                 ret = neigh_proc_dointvec_ms_jiffies(ctl, write, buffer, lenp, ppos);
3524         else
3525                 ret = -1;
3526
3527         if (write && ret == 0) {
3528                 /* update reachable_time as well, otherwise, the change will
3529                  * only be effective after the next time neigh_periodic_work
3530                  * decides to recompute it
3531                  */
3532                 p->reachable_time =
3533                         neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
3534         }
3535         return ret;
3536 }
3537
3538 #define NEIGH_PARMS_DATA_OFFSET(index)  \
3539         (&((struct neigh_parms *) 0)->data[index])
3540
3541 #define NEIGH_SYSCTL_ENTRY(attr, data_attr, name, mval, proc) \
3542         [NEIGH_VAR_ ## attr] = { \
3543                 .procname       = name, \
3544                 .data           = NEIGH_PARMS_DATA_OFFSET(NEIGH_VAR_ ## data_attr), \
3545                 .maxlen         = sizeof(int), \
3546                 .mode           = mval, \
3547                 .proc_handler   = proc, \
3548         }
3549
3550 #define NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(attr, name) \
3551         NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_zero_intmax)
3552
3553 #define NEIGH_SYSCTL_JIFFIES_ENTRY(attr, name) \
3554         NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_jiffies)
3555
3556 #define NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(attr, name) \
3557         NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_userhz_jiffies)
3558
3559 #define NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(attr, data_attr, name) \
3560         NEIGH_SYSCTL_ENTRY(attr, data_attr, name, 0644, neigh_proc_dointvec_ms_jiffies)
3561
3562 #define NEIGH_SYSCTL_UNRES_QLEN_REUSED_ENTRY(attr, data_attr, name) \
3563         NEIGH_SYSCTL_ENTRY(attr, data_attr, name, 0644, neigh_proc_dointvec_unres_qlen)
3564
3565 static struct neigh_sysctl_table {
3566         struct ctl_table_header *sysctl_header;
3567         struct ctl_table neigh_vars[NEIGH_VAR_MAX + 1];
3568 } neigh_sysctl_template __read_mostly = {
3569         .neigh_vars = {
3570                 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(MCAST_PROBES, "mcast_solicit"),
3571                 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(UCAST_PROBES, "ucast_solicit"),
3572                 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(APP_PROBES, "app_solicit"),
3573                 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(MCAST_REPROBES, "mcast_resolicit"),
3574                 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(RETRANS_TIME, "retrans_time"),
3575                 NEIGH_SYSCTL_JIFFIES_ENTRY(BASE_REACHABLE_TIME, "base_reachable_time"),
3576                 NEIGH_SYSCTL_JIFFIES_ENTRY(DELAY_PROBE_TIME, "delay_first_probe_time"),
3577                 NEIGH_SYSCTL_JIFFIES_ENTRY(GC_STALETIME, "gc_stale_time"),
3578                 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(QUEUE_LEN_BYTES, "unres_qlen_bytes"),
3579                 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(PROXY_QLEN, "proxy_qlen"),
3580                 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(ANYCAST_DELAY, "anycast_delay"),
3581                 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(PROXY_DELAY, "proxy_delay"),
3582                 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(LOCKTIME, "locktime"),
3583                 NEIGH_SYSCTL_UNRES_QLEN_REUSED_ENTRY(QUEUE_LEN, QUEUE_LEN_BYTES, "unres_qlen"),
3584                 NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(RETRANS_TIME_MS, RETRANS_TIME, "retrans_time_ms"),
3585                 NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(BASE_REACHABLE_TIME_MS, BASE_REACHABLE_TIME, "base_reachable_time_ms"),
3586                 [NEIGH_VAR_GC_INTERVAL] = {
3587                         .procname       = "gc_interval",
3588                         .maxlen         = sizeof(int),
3589                         .mode           = 0644,
3590                         .proc_handler   = proc_dointvec_jiffies,
3591                 },
3592                 [NEIGH_VAR_GC_THRESH1] = {
3593                         .procname       = "gc_thresh1",
3594                         .maxlen         = sizeof(int),
3595                         .mode           = 0644,
3596                         .extra1         = SYSCTL_ZERO,
3597                         .extra2         = SYSCTL_INT_MAX,
3598                         .proc_handler   = proc_dointvec_minmax,
3599                 },
3600                 [NEIGH_VAR_GC_THRESH2] = {
3601                         .procname       = "gc_thresh2",
3602                         .maxlen         = sizeof(int),
3603                         .mode           = 0644,
3604                         .extra1         = SYSCTL_ZERO,
3605                         .extra2         = SYSCTL_INT_MAX,
3606                         .proc_handler   = proc_dointvec_minmax,
3607                 },
3608                 [NEIGH_VAR_GC_THRESH3] = {
3609                         .procname       = "gc_thresh3",
3610                         .maxlen         = sizeof(int),
3611                         .mode           = 0644,
3612                         .extra1         = SYSCTL_ZERO,
3613                         .extra2         = SYSCTL_INT_MAX,
3614                         .proc_handler   = proc_dointvec_minmax,
3615                 },
3616                 {},
3617         },
3618 };
3619
3620 int neigh_sysctl_register(struct net_device *dev, struct neigh_parms *p,
3621                           proc_handler *handler)
3622 {
3623         int i;
3624         struct neigh_sysctl_table *t;
3625         const char *dev_name_source;
3626         char neigh_path[ sizeof("net//neigh/") + IFNAMSIZ + IFNAMSIZ ];
3627         char *p_name;
3628
3629         t = kmemdup(&neigh_sysctl_template, sizeof(*t), GFP_KERNEL);
3630         if (!t)
3631                 goto err;
3632
3633         for (i = 0; i < NEIGH_VAR_GC_INTERVAL; i++) {
3634                 t->neigh_vars[i].data += (long) p;
3635                 t->neigh_vars[i].extra1 = dev;
3636                 t->neigh_vars[i].extra2 = p;
3637         }
3638
3639         if (dev) {
3640                 dev_name_source = dev->name;
3641                 /* Terminate the table early */
3642                 memset(&t->neigh_vars[NEIGH_VAR_GC_INTERVAL], 0,
3643                        sizeof(t->neigh_vars[NEIGH_VAR_GC_INTERVAL]));
3644         } else {
3645                 struct neigh_table *tbl = p->tbl;
3646                 dev_name_source = "default";
3647                 t->neigh_vars[NEIGH_VAR_GC_INTERVAL].data = &tbl->gc_interval;
3648                 t->neigh_vars[NEIGH_VAR_GC_THRESH1].data = &tbl->gc_thresh1;
3649                 t->neigh_vars[NEIGH_VAR_GC_THRESH2].data = &tbl->gc_thresh2;
3650                 t->neigh_vars[NEIGH_VAR_GC_THRESH3].data = &tbl->gc_thresh3;
3651         }
3652
3653         if (handler) {
3654                 /* RetransTime */
3655                 t->neigh_vars[NEIGH_VAR_RETRANS_TIME].proc_handler = handler;
3656                 /* ReachableTime */
3657                 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].proc_handler = handler;
3658                 /* RetransTime (in milliseconds)*/
3659                 t->neigh_vars[NEIGH_VAR_RETRANS_TIME_MS].proc_handler = handler;
3660                 /* ReachableTime (in milliseconds) */
3661                 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].proc_handler = handler;
3662         } else {
3663                 /* Those handlers will update p->reachable_time after
3664                  * base_reachable_time(_ms) is set to ensure the new timer starts being
3665                  * applied after the next neighbour update instead of waiting for
3666                  * neigh_periodic_work to update its value (can be multiple minutes)
3667                  * So any handler that replaces them should do this as well
3668                  */
3669                 /* ReachableTime */
3670                 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].proc_handler =
3671                         neigh_proc_base_reachable_time;
3672                 /* ReachableTime (in milliseconds) */
3673                 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].proc_handler =
3674                         neigh_proc_base_reachable_time;
3675         }
3676
3677         /* Don't export sysctls to unprivileged users */
3678         if (neigh_parms_net(p)->user_ns != &init_user_ns)
3679                 t->neigh_vars[0].procname = NULL;
3680
3681         switch (neigh_parms_family(p)) {
3682         case AF_INET:
3683               p_name = "ipv4";
3684               break;
3685         case AF_INET6:
3686               p_name = "ipv6";
3687               break;
3688         default:
3689               BUG();
3690         }
3691
3692         snprintf(neigh_path, sizeof(neigh_path), "net/%s/neigh/%s",
3693                 p_name, dev_name_source);
3694         t->sysctl_header =
3695                 register_net_sysctl(neigh_parms_net(p), neigh_path, t->neigh_vars);
3696         if (!t->sysctl_header)
3697                 goto free;
3698
3699         p->sysctl_table = t;
3700         return 0;
3701
3702 free:
3703         kfree(t);
3704 err:
3705         return -ENOBUFS;
3706 }
3707 EXPORT_SYMBOL(neigh_sysctl_register);
3708
3709 void neigh_sysctl_unregister(struct neigh_parms *p)
3710 {
3711         if (p->sysctl_table) {
3712                 struct neigh_sysctl_table *t = p->sysctl_table;
3713                 p->sysctl_table = NULL;
3714                 unregister_net_sysctl_table(t->sysctl_header);
3715                 kfree(t);
3716         }
3717 }
3718 EXPORT_SYMBOL(neigh_sysctl_unregister);
3719
3720 #endif  /* CONFIG_SYSCTL */
3721
3722 static int __init neigh_init(void)
3723 {
3724         rtnl_register(PF_UNSPEC, RTM_NEWNEIGH, neigh_add, NULL, 0);
3725         rtnl_register(PF_UNSPEC, RTM_DELNEIGH, neigh_delete, NULL, 0);
3726         rtnl_register(PF_UNSPEC, RTM_GETNEIGH, neigh_get, neigh_dump_info, 0);
3727
3728         rtnl_register(PF_UNSPEC, RTM_GETNEIGHTBL, NULL, neightbl_dump_info,
3729                       0);
3730         rtnl_register(PF_UNSPEC, RTM_SETNEIGHTBL, neightbl_set, NULL, 0);
3731
3732         return 0;
3733 }
3734
3735 subsys_initcall(neigh_init);