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 NEIGH_DEBUG 1
  45 #define neigh_dbg(level, fmt, ...)              \
  46 do {                                            \
  47         if (level <= NEIGH_DEBUG)               \
  48                 pr_debug(fmt, ##__VA_ARGS__);   \
  49 } while (0)
  50
  51 #define PNEIGH_HASHMASK         0xF
  52
  53 static void neigh_timer_handler(struct timer_list *t);
  54 static void __neigh_notify(struct neighbour *n, int type, int flags,
  55                            u32 pid);
  56 static void neigh_update_notify(struct neighbour *neigh, u32 nlmsg_pid);
  57 static int pneigh_ifdown_and_unlock(struct neigh_table *tbl,
  58                                     struct net_device *dev);
  59
  60 #ifdef CONFIG_PROC_FS
  61 static const struct seq_operations neigh_stat_seq_ops;
  62 #endif
  63
  64 /*
  65    Neighbour hash table buckets are protected with rwlock tbl->lock.
  66
  67    - All the scans/updates to hash buckets MUST be made under this lock.
  68    - NOTHING clever should be made under this lock: no callbacks
  69      to protocol backends, no attempts to send something to network.
  70      It will result in deadlocks, if backend/driver wants to use neighbour
  71      cache.
  72    - If the entry requires some non-trivial actions, increase
  73      its reference count and release table lock.
  74
  75    Neighbour entries are protected:
  76    - with reference count.
  77    - with rwlock neigh->lock
  78
  79    Reference count prevents destruction.
  80
  81    neigh->lock mainly serializes ll address data and its validity state.
  82    However, the same lock is used to protect another entry fields:
  83     - timer
  84     - resolution queue
  85
  86    Again, nothing clever shall be made under neigh->lock,
  87    the most complicated procedure, which we allow is dev->hard_header.
  88    It is supposed, that dev->hard_header is simplistic and does
  89    not make callbacks to neighbour tables.
  90  */
  91
  92 static int neigh_blackhole(struct neighbour *neigh, struct sk_buff *skb)
  93 {
  94         kfree_skb(skb);
  95         return -ENETDOWN;
  96 }
  97
  98 static void neigh_cleanup_and_release(struct neighbour *neigh)
  99 {
 100         trace_neigh_cleanup_and_release(neigh, 0);
 101         __neigh_notify(neigh, RTM_DELNEIGH, 0, 0);
 102         call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, neigh);
 103         neigh_release(neigh);
 104 }
 105
 106 /*
 107  * It is random distribution in the interval (1/2)*base...(3/2)*base.
 108  * It corresponds to default IPv6 settings and is not overridable,
 109  * because it is really reasonable choice.
 110  */
 111
 112 unsigned long neigh_rand_reach_time(unsigned long base)
 113 {
 114         return base ? (prandom_u32() % base) + (base >> 1) : 0;
 115 }
 116 EXPORT_SYMBOL(neigh_rand_reach_time);
 117
 118 static void neigh_mark_dead(struct neighbour *n)
 119 {
 120         n->dead = 1;
 121         if (!list_empty(&n->gc_list)) {
 122                 list_del_init(&n->gc_list);
 123                 atomic_dec(&n->tbl->gc_entries);
 124         }
 125 }
 126
 127 static void neigh_update_gc_list(struct neighbour *n)
 128 {
 129         bool on_gc_list, exempt_from_gc;
 130
 131         write_lock_bh(&n->tbl->lock);
 132         write_lock(&n->lock);
 133
 134         if (n->dead)
 135                 goto out;
 136
 137         /* remove from the gc list if new state is permanent or if neighbor
 138          * is externally learned; otherwise entry should be on the gc list
 139          */
 140         exempt_from_gc = n->nud_state & NUD_PERMANENT ||
 141                          n->flags & NTF_EXT_LEARNED;
 142         on_gc_list = !list_empty(&n->gc_list);
 143
 144         if (exempt_from_gc && on_gc_list) {
 145                 list_del_init(&n->gc_list);
 146                 atomic_dec(&n->tbl->gc_entries);
 147         } else if (!exempt_from_gc && !on_gc_list) {
 148                 /* add entries to the tail; cleaning removes from the front */
 149                 list_add_tail(&n->gc_list, &n->tbl->gc_list);
 150                 atomic_inc(&n->tbl->gc_entries);
 151         }
 152
 153 out:
 154         write_unlock(&n->lock);
 155         write_unlock_bh(&n->tbl->lock);
 156 }
 157
 158 static bool neigh_update_ext_learned(struct neighbour *neigh, u32 flags,
 159                                      int *notify)
 160 {
 161         bool rc = false;
 162         u8 ndm_flags;
 163
 164         if (!(flags & NEIGH_UPDATE_F_ADMIN))
 165                 return rc;
 166
 167         ndm_flags = (flags & NEIGH_UPDATE_F_EXT_LEARNED) ? NTF_EXT_LEARNED : 0;
 168         if ((neigh->flags ^ ndm_flags) & NTF_EXT_LEARNED) {
 169                 if (ndm_flags & NTF_EXT_LEARNED)
 170                         neigh->flags |= NTF_EXT_LEARNED;
 171                 else
 172                         neigh->flags &= ~NTF_EXT_LEARNED;
 173                 rc = true;
 174                 *notify = 1;
 175         }
 176
 177         return rc;
 178 }
 179
 180 static bool neigh_del(struct neighbour *n, struct neighbour __rcu **np,
 181                       struct neigh_table *tbl)
 182 {
 183         bool retval = false;
 184
 185         write_lock(&n->lock);
 186         if (refcount_read(&n->refcnt) == 1) {
 187                 struct neighbour *neigh;
 188
 189                 neigh = rcu_dereference_protected(n->next,
 190                                                   lockdep_is_held(&tbl->lock));
 191                 rcu_assign_pointer(*np, neigh);
 192                 neigh_mark_dead(n);
 193                 retval = true;
 194         }
 195         write_unlock(&n->lock);
 196         if (retval)
 197                 neigh_cleanup_and_release(n);
 198         return retval;
 199 }
 200
 201 bool neigh_remove_one(struct neighbour *ndel, struct neigh_table *tbl)
 202 {
 203         struct neigh_hash_table *nht;
 204         void *pkey = ndel->primary_key;
 205         u32 hash_val;
 206         struct neighbour *n;
 207         struct neighbour __rcu **np;
 208
 209         nht = rcu_dereference_protected(tbl->nht,
 210                                         lockdep_is_held(&tbl->lock));
 211         hash_val = tbl->hash(pkey, ndel->dev, nht->hash_rnd);
 212         hash_val = hash_val >> (32 - nht->hash_shift);
 213
 214         np = &nht->hash_buckets[hash_val];
 215         while ((n = rcu_dereference_protected(*np,
 216                                               lockdep_is_held(&tbl->lock)))) {
 217                 if (n == ndel)
 218                         return neigh_del(n, np, tbl);
 219                 np = &n->next;
 220         }
 221         return false;
 222 }
 223
 224 static int neigh_forced_gc(struct neigh_table *tbl)
 225 {
 226         int max_clean = atomic_read(&tbl->gc_entries) - tbl->gc_thresh2;
 227         unsigned long tref = jiffies - 5 * HZ;
 228         struct neighbour *n, *tmp;
 229         int shrunk = 0;
 230
 231         NEIGH_CACHE_STAT_INC(tbl, forced_gc_runs);
 232
 233         write_lock_bh(&tbl->lock);
 234
 235         list_for_each_entry_safe(n, tmp, &tbl->gc_list, gc_list) {
 236                 if (refcount_read(&n->refcnt) == 1) {
 237                         bool remove = false;
 238
 239                         write_lock(&n->lock);
 240                         if ((n->nud_state == NUD_FAILED) ||
 241                             (n->nud_state == NUD_NOARP) ||
 242                             (tbl->is_multicast &&
 243                              tbl->is_multicast(n->primary_key)) ||
 244                             time_after(tref, n->updated))
 245                                 remove = true;
 246                         write_unlock(&n->lock);
 247
 248                         if (remove && neigh_remove_one(n, tbl))
 249                                 shrunk++;
 250                         if (shrunk >= max_clean)
 251                                 break;
 252                 }
 253         }
 254
 255         tbl->last_flush = jiffies;
 256
 257         write_unlock_bh(&tbl->lock);
 258
 259         return shrunk;
 260 }
 261
 262 static void neigh_add_timer(struct neighbour *n, unsigned long when)
 263 {
 264         neigh_hold(n);
 265         if (unlikely(mod_timer(&n->timer, when))) {
 266                 printk("NEIGH: BUG, double timer add, state is %x\n",
 267                        n->nud_state);
 268                 dump_stack();
 269         }
 270 }
 271
 272 static int neigh_del_timer(struct neighbour *n)
 273 {
 274         if ((n->nud_state & NUD_IN_TIMER) &&
 275             del_timer(&n->timer)) {
 276                 neigh_release(n);
 277                 return 1;
 278         }
 279         return 0;
 280 }
 281
 282 static void pneigh_queue_purge(struct sk_buff_head *list)
 283 {
 284         struct sk_buff *skb;
 285
 286         while ((skb = skb_dequeue(list)) != NULL) {
 287                 dev_put(skb->dev);
 288                 kfree_skb(skb);
 289         }
 290 }
 291
 292 static void neigh_flush_dev(struct neigh_table *tbl, struct net_device *dev,
 293                             bool skip_perm)
 294 {
 295         int i;
 296         struct neigh_hash_table *nht;
 297
 298         nht = rcu_dereference_protected(tbl->nht,
 299                                         lockdep_is_held(&tbl->lock));
 300
 301         for (i = 0; i < (1 << nht->hash_shift); i++) {
 302                 struct neighbour *n;
 303                 struct neighbour __rcu **np = &nht->hash_buckets[i];
 304
 305                 while ((n = rcu_dereference_protected(*np,
 306                                         lockdep_is_held(&tbl->lock))) != NULL) {
 307                         if (dev && n->dev != dev) {
 308                                 np = &n->next;
 309                                 continue;
 310                         }
 311                         if (skip_perm && n->nud_state & NUD_PERMANENT) {
 312                                 np = &n->next;
 313                                 continue;
 314                         }
 315                         rcu_assign_pointer(*np,
 316                                    rcu_dereference_protected(n->next,
 317                                                 lockdep_is_held(&tbl->lock)));
 318                         write_lock(&n->lock);
 319                         neigh_del_timer(n);
 320                         neigh_mark_dead(n);
 321                         if (refcount_read(&n->refcnt) != 1) {
 322                                 /* The most unpleasant situation.
 323                                    We must destroy neighbour entry,
 324                                    but someone still uses it.
 325
 326                                    The destroy will be delayed until
 327                                    the last user releases us, but
 328                                    we must kill timers etc. and move
 329                                    it to safe state.
 330                                  */
 331                                 __skb_queue_purge(&n->arp_queue);
 332                                 n->arp_queue_len_bytes = 0;
 333                                 n->output = neigh_blackhole;
 334                                 if (n->nud_state & NUD_VALID)
 335                                         n->nud_state = NUD_NOARP;
 336                                 else
 337                                         n->nud_state = NUD_NONE;
 338                                 neigh_dbg(2, "neigh %p is stray\n", n);
 339                         }
 340                         write_unlock(&n->lock);
 341                         neigh_cleanup_and_release(n);
 342                 }
 343         }
 344 }
 345
 346 void neigh_changeaddr(struct neigh_table *tbl, struct net_device *dev)
 347 {
 348         write_lock_bh(&tbl->lock);
 349         neigh_flush_dev(tbl, dev, false);
 350         write_unlock_bh(&tbl->lock);
 351 }
 352 EXPORT_SYMBOL(neigh_changeaddr);
 353
 354 static int __neigh_ifdown(struct neigh_table *tbl, struct net_device *dev,
 355                           bool skip_perm)
 356 {
 357         write_lock_bh(&tbl->lock);
 358         neigh_flush_dev(tbl, dev, skip_perm);
 359         pneigh_ifdown_and_unlock(tbl, dev);
 360
 361         del_timer_sync(&tbl->proxy_timer);
 362         pneigh_queue_purge(&tbl->proxy_queue);
 363         return 0;
 364 }
 365
 366 int neigh_carrier_down(struct neigh_table *tbl, struct net_device *dev)
 367 {
 368         __neigh_ifdown(tbl, dev, true);
 369         return 0;
 370 }
 371 EXPORT_SYMBOL(neigh_carrier_down);
 372
 373 int neigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
 374 {
 375         __neigh_ifdown(tbl, dev, false);
 376         return 0;
 377 }
 378 EXPORT_SYMBOL(neigh_ifdown);
 379
 380 static struct neighbour *neigh_alloc(struct neigh_table *tbl,
 381                                      struct net_device *dev,
 382                                      bool exempt_from_gc)
 383 {
 384         struct neighbour *n = NULL;
 385         unsigned long now = jiffies;
 386         int entries;
 387
 388         if (exempt_from_gc)
 389                 goto do_alloc;
 390
 391         entries = atomic_inc_return(&tbl->gc_entries) - 1;
 392         if (entries >= tbl->gc_thresh3 ||
 393             (entries >= tbl->gc_thresh2 &&
 394              time_after(now, tbl->last_flush + 5 * HZ))) {
 395                 if (!neigh_forced_gc(tbl) &&
 396                     entries >= tbl->gc_thresh3) {
 397                         net_info_ratelimited("%s: neighbor table overflow!\n",
 398                                              tbl->id);
 399                         NEIGH_CACHE_STAT_INC(tbl, table_fulls);
 400                         goto out_entries;
 401                 }
 402         }
 403
 404 do_alloc:
 405         n = kzalloc(tbl->entry_size + dev->neigh_priv_len, GFP_ATOMIC);
 406         if (!n)
 407                 goto out_entries;
 408
 409         __skb_queue_head_init(&n->arp_queue);
 410         rwlock_init(&n->lock);
 411         seqlock_init(&n->ha_lock);
 412         n->updated        = n->used = now;
 413         n->nud_state      = NUD_NONE;
 414         n->output         = neigh_blackhole;
 415         seqlock_init(&n->hh.hh_lock);
 416         n->parms          = neigh_parms_clone(&tbl->parms);
 417         timer_setup(&n->timer, neigh_timer_handler, 0);
 418
 419         NEIGH_CACHE_STAT_INC(tbl, allocs);
 420         n->tbl            = tbl;
 421         refcount_set(&n->refcnt, 1);
 422         n->dead           = 1;
 423         INIT_LIST_HEAD(&n->gc_list);
 424
 425         atomic_inc(&tbl->entries);
 426 out:
 427         return n;
 428
 429 out_entries:
 430         if (!exempt_from_gc)
 431                 atomic_dec(&tbl->gc_entries);
 432         goto out;
 433 }
 434
 435 static void neigh_get_hash_rnd(u32 *x)
 436 {
 437         *x = get_random_u32() | 1;
 438 }
 439
 440 static struct neigh_hash_table *neigh_hash_alloc(unsigned int shift)
 441 {
 442         size_t size = (1 << shift) * sizeof(struct neighbour *);
 443         struct neigh_hash_table *ret;
 444         struct neighbour __rcu **buckets;
 445         int i;
 446
 447         ret = kmalloc(sizeof(*ret), GFP_ATOMIC);
 448         if (!ret)
 449                 return NULL;
 450         if (size <= PAGE_SIZE) {
 451                 buckets = kzalloc(size, GFP_ATOMIC);
 452         } else {
 453                 buckets = (struct neighbour __rcu **)
 454                           __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
 455                                            get_order(size));
 456                 kmemleak_alloc(buckets, size, 1, GFP_ATOMIC);
 457         }
 458         if (!buckets) {
 459                 kfree(ret);
 460                 return NULL;
 461         }
 462         ret->hash_buckets = buckets;
 463         ret->hash_shift = shift;
 464         for (i = 0; i < NEIGH_NUM_HASH_RND; i++)
 465                 neigh_get_hash_rnd(&ret->hash_rnd[i]);
 466         return ret;
 467 }
 468
 469 static void neigh_hash_free_rcu(struct rcu_head *head)
 470 {
 471         struct neigh_hash_table *nht = container_of(head,
 472                                                     struct neigh_hash_table,
 473                                                     rcu);
 474         size_t size = (1 << nht->hash_shift) * sizeof(struct neighbour *);
 475         struct neighbour __rcu **buckets = nht->hash_buckets;
 476
 477         if (size <= PAGE_SIZE) {
 478                 kfree(buckets);
 479         } else {
 480                 kmemleak_free(buckets);
 481                 free_pages((unsigned long)buckets, get_order(size));
 482         }
 483         kfree(nht);
 484 }
 485
 486 static struct neigh_hash_table *neigh_hash_grow(struct neigh_table *tbl,
 487                                                 unsigned long new_shift)
 488 {
 489         unsigned int i, hash;
 490         struct neigh_hash_table *new_nht, *old_nht;
 491
 492         NEIGH_CACHE_STAT_INC(tbl, hash_grows);
 493
 494         old_nht = rcu_dereference_protected(tbl->nht,
 495                                             lockdep_is_held(&tbl->lock));
 496         new_nht = neigh_hash_alloc(new_shift);
 497         if (!new_nht)
 498                 return old_nht;
 499
 500         for (i = 0; i < (1 << old_nht->hash_shift); i++) {
 501                 struct neighbour *n, *next;
 502
 503                 for (n = rcu_dereference_protected(old_nht->hash_buckets[i],
 504                                                    lockdep_is_held(&tbl->lock));
 505                      n != NULL;
 506                      n = next) {
 507                         hash = tbl->hash(n->primary_key, n->dev,
 508                                          new_nht->hash_rnd);
 509
 510                         hash >>= (32 - new_nht->hash_shift);
 511                         next = rcu_dereference_protected(n->next,
 512                                                 lockdep_is_held(&tbl->lock));
 513
 514                         rcu_assign_pointer(n->next,
 515                                            rcu_dereference_protected(
 516                                                 new_nht->hash_buckets[hash],
 517                                                 lockdep_is_held(&tbl->lock)));
 518                         rcu_assign_pointer(new_nht->hash_buckets[hash], n);
 519                 }
 520         }
 521
 522         rcu_assign_pointer(tbl->nht, new_nht);
 523         call_rcu(&old_nht->rcu, neigh_hash_free_rcu);
 524         return new_nht;
 525 }
 526
 527 struct neighbour *neigh_lookup(struct neigh_table *tbl, const void *pkey,
 528                                struct net_device *dev)
 529 {
 530         struct neighbour *n;
 531
 532         NEIGH_CACHE_STAT_INC(tbl, lookups);
 533
 534         rcu_read_lock_bh();
 535         n = __neigh_lookup_noref(tbl, pkey, dev);
 536         if (n) {
 537                 if (!refcount_inc_not_zero(&n->refcnt))
 538                         n = NULL;
 539                 NEIGH_CACHE_STAT_INC(tbl, hits);
 540         }
 541
 542         rcu_read_unlock_bh();
 543         return n;
 544 }
 545 EXPORT_SYMBOL(neigh_lookup);
 546
 547 struct neighbour *neigh_lookup_nodev(struct neigh_table *tbl, struct net *net,
 548                                      const void *pkey)
 549 {
 550         struct neighbour *n;
 551         unsigned int key_len = tbl->key_len;
 552         u32 hash_val;
 553         struct neigh_hash_table *nht;
 554
 555         NEIGH_CACHE_STAT_INC(tbl, lookups);
 556
 557         rcu_read_lock_bh();
 558         nht = rcu_dereference_bh(tbl->nht);
 559         hash_val = tbl->hash(pkey, NULL, nht->hash_rnd) >> (32 - nht->hash_shift);
 560
 561         for (n = rcu_dereference_bh(nht->hash_buckets[hash_val]);
 562              n != NULL;
 563              n = rcu_dereference_bh(n->next)) {
 564                 if (!memcmp(n->primary_key, pkey, key_len) &&
 565                     net_eq(dev_net(n->dev), net)) {
 566                         if (!refcount_inc_not_zero(&n->refcnt))
 567                                 n = NULL;
 568                         NEIGH_CACHE_STAT_INC(tbl, hits);
 569                         break;
 570                 }
 571         }
 572
 573         rcu_read_unlock_bh();
 574         return n;
 575 }
 576 EXPORT_SYMBOL(neigh_lookup_nodev);
 577
 578 static struct neighbour *___neigh_create(struct neigh_table *tbl,
 579                                          const void *pkey,
 580                                          struct net_device *dev,
 581                                          bool exempt_from_gc, bool want_ref)
 582 {
 583         struct neighbour *n1, *rc, *n = neigh_alloc(tbl, dev, exempt_from_gc);
 584         u32 hash_val;
 585         unsigned int key_len = tbl->key_len;
 586         int error;
 587         struct neigh_hash_table *nht;
 588
 589         trace_neigh_create(tbl, dev, pkey, n, exempt_from_gc);
 590
 591         if (!n) {
 592                 rc = ERR_PTR(-ENOBUFS);
 593                 goto out;
 594         }
 595
 596         memcpy(n->primary_key, pkey, key_len);
 597         n->dev = dev;
 598         dev_hold(dev);
 599
 600         /* Protocol specific setup. */
 601         if (tbl->constructor && (error = tbl->constructor(n)) < 0) {
 602                 rc = ERR_PTR(error);
 603                 goto out_neigh_release;
 604         }
 605
 606         if (dev->netdev_ops->ndo_neigh_construct) {
 607                 error = dev->netdev_ops->ndo_neigh_construct(dev, n);
 608                 if (error < 0) {
 609                         rc = ERR_PTR(error);
 610                         goto out_neigh_release;
 611                 }
 612         }
 613
 614         /* Device specific setup. */
 615         if (n->parms->neigh_setup &&
 616             (error = n->parms->neigh_setup(n)) < 0) {
 617                 rc = ERR_PTR(error);
 618                 goto out_neigh_release;
 619         }
 620
 621         n->confirmed = jiffies - (NEIGH_VAR(n->parms, BASE_REACHABLE_TIME) << 1);
 622
 623         write_lock_bh(&tbl->lock);
 624         nht = rcu_dereference_protected(tbl->nht,
 625                                         lockdep_is_held(&tbl->lock));
 626
 627         if (atomic_read(&tbl->entries) > (1 << nht->hash_shift))
 628                 nht = neigh_hash_grow(tbl, nht->hash_shift + 1);
 629
 630         hash_val = tbl->hash(n->primary_key, dev, nht->hash_rnd) >> (32 - nht->hash_shift);
 631
 632         if (n->parms->dead) {
 633                 rc = ERR_PTR(-EINVAL);
 634                 goto out_tbl_unlock;
 635         }
 636
 637         for (n1 = rcu_dereference_protected(nht->hash_buckets[hash_val],
 638                                             lockdep_is_held(&tbl->lock));
 639              n1 != NULL;
 640              n1 = rcu_dereference_protected(n1->next,
 641                         lockdep_is_held(&tbl->lock))) {
 642                 if (dev == n1->dev && !memcmp(n1->primary_key, n->primary_key, key_len)) {
 643                         if (want_ref)
 644                                 neigh_hold(n1);
 645                         rc = n1;
 646                         goto out_tbl_unlock;
 647                 }
 648         }
 649
 650         n->dead = 0;
 651         if (!exempt_from_gc)
 652                 list_add_tail(&n->gc_list, &n->tbl->gc_list);
 653
 654         if (want_ref)
 655                 neigh_hold(n);
 656         rcu_assign_pointer(n->next,
 657                            rcu_dereference_protected(nht->hash_buckets[hash_val],
 658                                                      lockdep_is_held(&tbl->lock)));
 659         rcu_assign_pointer(nht->hash_buckets[hash_val], n);
 660         write_unlock_bh(&tbl->lock);
 661         neigh_dbg(2, "neigh %p is created\n", n);
 662         rc = n;
 663 out:
 664         return rc;
 665 out_tbl_unlock:
 666         write_unlock_bh(&tbl->lock);
 667 out_neigh_release:
 668         if (!exempt_from_gc)
 669                 atomic_dec(&tbl->gc_entries);
 670         neigh_release(n);
 671         goto out;
 672 }
 673
 674 struct neighbour *__neigh_create(struct neigh_table *tbl, const void *pkey,
 675                                  struct net_device *dev, bool want_ref)
 676 {
 677         return ___neigh_create(tbl, pkey, dev, false, want_ref);
 678 }
 679 EXPORT_SYMBOL(__neigh_create);
 680
 681 static u32 pneigh_hash(const void *pkey, unsigned int key_len)
 682 {
 683         u32 hash_val = *(u32 *)(pkey + key_len - 4);
 684         hash_val ^= (hash_val >> 16);
 685         hash_val ^= hash_val >> 8;
 686         hash_val ^= hash_val >> 4;
 687         hash_val &= PNEIGH_HASHMASK;
 688         return hash_val;
 689 }
 690
 691 static struct pneigh_entry *__pneigh_lookup_1(struct pneigh_entry *n,
 692                                               struct net *net,
 693                                               const void *pkey,
 694                                               unsigned int key_len,
 695                                               struct net_device *dev)
 696 {
 697         while (n) {
 698                 if (!memcmp(n->key, pkey, key_len) &&
 699                     net_eq(pneigh_net(n), net) &&
 700                     (n->dev == dev || !n->dev))
 701                         return n;
 702                 n = n->next;
 703         }
 704         return NULL;
 705 }
 706
 707 struct pneigh_entry *__pneigh_lookup(struct neigh_table *tbl,
 708                 struct net *net, const void *pkey, struct net_device *dev)
 709 {
 710         unsigned int key_len = tbl->key_len;
 711         u32 hash_val = pneigh_hash(pkey, key_len);
 712
 713         return __pneigh_lookup_1(tbl->phash_buckets[hash_val],
 714                                  net, pkey, key_len, dev);
 715 }
 716 EXPORT_SYMBOL_GPL(__pneigh_lookup);
 717
 718 struct pneigh_entry * pneigh_lookup(struct neigh_table *tbl,
 719                                     struct net *net, const void *pkey,
 720                                     struct net_device *dev, int creat)
 721 {
 722         struct pneigh_entry *n;
 723         unsigned int key_len = tbl->key_len;
 724         u32 hash_val = pneigh_hash(pkey, key_len);
 725
 726         read_lock_bh(&tbl->lock);
 727         n = __pneigh_lookup_1(tbl->phash_buckets[hash_val],
 728                               net, pkey, key_len, dev);
 729         read_unlock_bh(&tbl->lock);
 730
 731         if (n || !creat)
 732                 goto out;
 733
 734         ASSERT_RTNL();
 735
 736         n = kmalloc(sizeof(*n) + key_len, GFP_KERNEL);
 737         if (!n)
 738                 goto out;
 739
 740         n->protocol = 0;
 741         write_pnet(&n->net, net);
 742         memcpy(n->key, pkey, key_len);
 743         n->dev = dev;
 744         dev_hold(dev);
 745
 746         if (tbl->pconstructor && tbl->pconstructor(n)) {
 747                 dev_put(dev);
 748                 kfree(n);
 749                 n = NULL;
 750                 goto out;
 751         }
 752
 753         write_lock_bh(&tbl->lock);
 754         n->next = tbl->phash_buckets[hash_val];
 755         tbl->phash_buckets[hash_val] = n;
 756         write_unlock_bh(&tbl->lock);
 757 out:
 758         return n;
 759 }
 760 EXPORT_SYMBOL(pneigh_lookup);
 761
 762
 763 int pneigh_delete(struct neigh_table *tbl, struct net *net, const void *pkey,
 764                   struct net_device *dev)
 765 {
 766         struct pneigh_entry *n, **np;
 767         unsigned int key_len = tbl->key_len;
 768         u32 hash_val = pneigh_hash(pkey, key_len);
 769
 770         write_lock_bh(&tbl->lock);
 771         for (np = &tbl->phash_buckets[hash_val]; (n = *np) != NULL;
 772              np = &n->next) {
 773                 if (!memcmp(n->key, pkey, key_len) && n->dev == dev &&
 774                     net_eq(pneigh_net(n), net)) {
 775                         *np = n->next;
 776                         write_unlock_bh(&tbl->lock);
 777                         if (tbl->pdestructor)
 778                                 tbl->pdestructor(n);
 779                         dev_put(n->dev);
 780                         kfree(n);
 781                         return 0;
 782                 }
 783         }
 784         write_unlock_bh(&tbl->lock);
 785         return -ENOENT;
 786 }
 787
 788 static int pneigh_ifdown_and_unlock(struct neigh_table *tbl,
 789                                     struct net_device *dev)
 790 {
 791         struct pneigh_entry *n, **np, *freelist = NULL;
 792         u32 h;
 793
 794         for (h = 0; h <= PNEIGH_HASHMASK; h++) {
 795                 np = &tbl->phash_buckets[h];
 796                 while ((n = *np) != NULL) {
 797                         if (!dev || n->dev == dev) {
 798                                 *np = n->next;
 799                                 n->next = freelist;
 800                                 freelist = n;
 801                                 continue;
 802                         }
 803                         np = &n->next;
 804                 }
 805         }
 806         write_unlock_bh(&tbl->lock);
 807         while ((n = freelist)) {
 808                 freelist = n->next;
 809                 n->next = NULL;
 810                 if (tbl->pdestructor)
 811                         tbl->pdestructor(n);
 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 (neigh->dead) {
1249                 NL_SET_ERR_MSG(extack, "Neighbor entry is now dead");
1250                 new = old;
1251                 goto out;
1252         }
1253         if (!(flags & NEIGH_UPDATE_F_ADMIN) &&
1254             (old & (NUD_NOARP | NUD_PERMANENT)))
1255                 goto out;
1256
1257         ext_learn_change = neigh_update_ext_learned(neigh, flags, &notify);
1258
1259         if (!(new & NUD_VALID)) {
1260                 neigh_del_timer(neigh);
1261                 if (old & NUD_CONNECTED)
1262                         neigh_suspect(neigh);
1263                 neigh->nud_state = new;
1264                 err = 0;
1265                 notify = old & NUD_VALID;
1266                 if ((old & (NUD_INCOMPLETE | NUD_PROBE)) &&
1267                     (new & NUD_FAILED)) {
1268                         neigh_invalidate(neigh);
1269                         notify = 1;
1270                 }
1271                 goto out;
1272         }
1273
1274         /* Compare new lladdr with cached one */
1275         if (!dev->addr_len) {
1276                 /* First case: device needs no address. */
1277                 lladdr = neigh->ha;
1278         } else if (lladdr) {
1279                 /* The second case: if something is already cached
1280                    and a new address is proposed:
1281                    - compare new & old
1282                    - if they are different, check override flag
1283                  */
1284                 if ((old & NUD_VALID) &&
1285                     !memcmp(lladdr, neigh->ha, dev->addr_len))
1286                         lladdr = neigh->ha;
1287         } else {
1288                 /* No address is supplied; if we know something,
1289                    use it, otherwise discard the request.
1290                  */
1291                 err = -EINVAL;
1292                 if (!(old & NUD_VALID)) {
1293                         NL_SET_ERR_MSG(extack, "No link layer address given");
1294                         goto out;
1295                 }
1296                 lladdr = neigh->ha;
1297         }
1298
1299         /* Update confirmed timestamp for neighbour entry after we
1300          * received ARP packet even if it doesn't change IP to MAC binding.
1301          */
1302         if (new & NUD_CONNECTED)
1303                 neigh->confirmed = jiffies;
1304
1305         /* If entry was valid and address is not changed,
1306            do not change entry state, if new one is STALE.
1307          */
1308         err = 0;
1309         update_isrouter = flags & NEIGH_UPDATE_F_OVERRIDE_ISROUTER;
1310         if (old & NUD_VALID) {
1311                 if (lladdr != neigh->ha && !(flags & NEIGH_UPDATE_F_OVERRIDE)) {
1312                         update_isrouter = 0;
1313                         if ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) &&
1314                             (old & NUD_CONNECTED)) {
1315                                 lladdr = neigh->ha;
1316                                 new = NUD_STALE;
1317                         } else
1318                                 goto out;
1319                 } else {
1320                         if (lladdr == neigh->ha && new == NUD_STALE &&
1321                             !(flags & NEIGH_UPDATE_F_ADMIN))
1322                                 new = old;
1323                 }
1324         }
1325
1326         /* Update timestamp only once we know we will make a change to the
1327          * neighbour entry. Otherwise we risk to move the locktime window with
1328          * noop updates and ignore relevant ARP updates.
1329          */
1330         if (new != old || lladdr != neigh->ha)
1331                 neigh->updated = jiffies;
1332
1333         if (new != old) {
1334                 neigh_del_timer(neigh);
1335                 if (new & NUD_PROBE)
1336                         atomic_set(&neigh->probes, 0);
1337                 if (new & NUD_IN_TIMER)
1338                         neigh_add_timer(neigh, (jiffies +
1339                                                 ((new & NUD_REACHABLE) ?
1340                                                  neigh->parms->reachable_time :
1341                                                  0)));
1342                 neigh->nud_state = new;
1343                 notify = 1;
1344         }
1345
1346         if (lladdr != neigh->ha) {
1347                 write_seqlock(&neigh->ha_lock);
1348                 memcpy(&neigh->ha, lladdr, dev->addr_len);
1349                 write_sequnlock(&neigh->ha_lock);
1350                 neigh_update_hhs(neigh);
1351                 if (!(new & NUD_CONNECTED))
1352                         neigh->confirmed = jiffies -
1353                                       (NEIGH_VAR(neigh->parms, BASE_REACHABLE_TIME) << 1);
1354                 notify = 1;
1355         }
1356         if (new == old)
1357                 goto out;
1358         if (new & NUD_CONNECTED)
1359                 neigh_connect(neigh);
1360         else
1361                 neigh_suspect(neigh);
1362         if (!(old & NUD_VALID)) {
1363                 struct sk_buff *skb;
1364
1365                 /* Again: avoid dead loop if something went wrong */
1366
1367                 while (neigh->nud_state & NUD_VALID &&
1368                        (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
1369                         struct dst_entry *dst = skb_dst(skb);
1370                         struct neighbour *n2, *n1 = neigh;
1371                         write_unlock_bh(&neigh->lock);
1372
1373                         rcu_read_lock();
1374
1375                         /* Why not just use 'neigh' as-is?  The problem is that
1376                          * things such as shaper, eql, and sch_teql can end up
1377                          * using alternative, different, neigh objects to output
1378                          * the packet in the output path.  So what we need to do
1379                          * here is re-lookup the top-level neigh in the path so
1380                          * we can reinject the packet there.
1381                          */
1382                         n2 = NULL;
1383                         if (dst && dst->obsolete != DST_OBSOLETE_DEAD) {
1384                                 n2 = dst_neigh_lookup_skb(dst, skb);
1385                                 if (n2)
1386                                         n1 = n2;
1387                         }
1388                         n1->output(n1, skb);
1389                         if (n2)
1390                                 neigh_release(n2);
1391                         rcu_read_unlock();
1392
1393                         write_lock_bh(&neigh->lock);
1394                 }
1395                 __skb_queue_purge(&neigh->arp_queue);
1396                 neigh->arp_queue_len_bytes = 0;
1397         }
1398 out:
1399         if (update_isrouter)
1400                 neigh_update_is_router(neigh, flags, &notify);
1401         write_unlock_bh(&neigh->lock);
1402
1403         if (((new ^ old) & NUD_PERMANENT) || ext_learn_change)
1404                 neigh_update_gc_list(neigh);
1405
1406         if (notify)
1407                 neigh_update_notify(neigh, nlmsg_pid);
1408
1409         trace_neigh_update_done(neigh, err);
1410
1411         return err;
1412 }
1413
1414 int neigh_update(struct neighbour *neigh, const u8 *lladdr, u8 new,
1415                  u32 flags, u32 nlmsg_pid)
1416 {
1417         return __neigh_update(neigh, lladdr, new, flags, nlmsg_pid, NULL);
1418 }
1419 EXPORT_SYMBOL(neigh_update);
1420
1421 /* Update the neigh to listen temporarily for probe responses, even if it is
1422  * in a NUD_FAILED state. The caller has to hold neigh->lock for writing.
1423  */
1424 void __neigh_set_probe_once(struct neighbour *neigh)
1425 {
1426         if (neigh->dead)
1427                 return;
1428         neigh->updated = jiffies;
1429         if (!(neigh->nud_state & NUD_FAILED))
1430                 return;
1431         neigh->nud_state = NUD_INCOMPLETE;
1432         atomic_set(&neigh->probes, neigh_max_probes(neigh));
1433         neigh_add_timer(neigh,
1434                         jiffies + max(NEIGH_VAR(neigh->parms, RETRANS_TIME),
1435                                       HZ/100));
1436 }
1437 EXPORT_SYMBOL(__neigh_set_probe_once);
1438
1439 struct neighbour *neigh_event_ns(struct neigh_table *tbl,
1440                                  u8 *lladdr, void *saddr,
1441                                  struct net_device *dev)
1442 {
1443         struct neighbour *neigh = __neigh_lookup(tbl, saddr, dev,
1444                                                  lladdr || !dev->addr_len);
1445         if (neigh)
1446                 neigh_update(neigh, lladdr, NUD_STALE,
1447                              NEIGH_UPDATE_F_OVERRIDE, 0);
1448         return neigh;
1449 }
1450 EXPORT_SYMBOL(neigh_event_ns);
1451
1452 /* called with read_lock_bh(&n->lock); */
1453 static void neigh_hh_init(struct neighbour *n)
1454 {
1455         struct net_device *dev = n->dev;
1456         __be16 prot = n->tbl->protocol;
1457         struct hh_cache *hh = &n->hh;
1458
1459         write_lock_bh(&n->lock);
1460
1461         /* Only one thread can come in here and initialize the
1462          * hh_cache entry.
1463          */
1464         if (!hh->hh_len)
1465                 dev->header_ops->cache(n, hh, prot);
1466
1467         write_unlock_bh(&n->lock);
1468 }
1469
1470 /* Slow and careful. */
1471
1472 int neigh_resolve_output(struct neighbour *neigh, struct sk_buff *skb)
1473 {
1474         int rc = 0;
1475
1476         if (!neigh_event_send(neigh, skb)) {
1477                 int err;
1478                 struct net_device *dev = neigh->dev;
1479                 unsigned int seq;
1480
1481                 if (dev->header_ops->cache && !READ_ONCE(neigh->hh.hh_len))
1482                         neigh_hh_init(neigh);
1483
1484                 do {
1485                         __skb_pull(skb, skb_network_offset(skb));
1486                         seq = read_seqbegin(&neigh->ha_lock);
1487                         err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1488                                               neigh->ha, NULL, skb->len);
1489                 } while (read_seqretry(&neigh->ha_lock, seq));
1490
1491                 if (err >= 0)
1492                         rc = dev_queue_xmit(skb);
1493                 else
1494                         goto out_kfree_skb;
1495         }
1496 out:
1497         return rc;
1498 out_kfree_skb:
1499         rc = -EINVAL;
1500         kfree_skb(skb);
1501         goto out;
1502 }
1503 EXPORT_SYMBOL(neigh_resolve_output);
1504
1505 /* As fast as possible without hh cache */
1506
1507 int neigh_connected_output(struct neighbour *neigh, struct sk_buff *skb)
1508 {
1509         struct net_device *dev = neigh->dev;
1510         unsigned int seq;
1511         int err;
1512
1513         do {
1514                 __skb_pull(skb, skb_network_offset(skb));
1515                 seq = read_seqbegin(&neigh->ha_lock);
1516                 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1517                                       neigh->ha, NULL, skb->len);
1518         } while (read_seqretry(&neigh->ha_lock, seq));
1519
1520         if (err >= 0)
1521                 err = dev_queue_xmit(skb);
1522         else {
1523                 err = -EINVAL;
1524                 kfree_skb(skb);
1525         }
1526         return err;
1527 }
1528 EXPORT_SYMBOL(neigh_connected_output);
1529
1530 int neigh_direct_output(struct neighbour *neigh, struct sk_buff *skb)
1531 {
1532         return dev_queue_xmit(skb);
1533 }
1534 EXPORT_SYMBOL(neigh_direct_output);
1535
1536 static void neigh_proxy_process(struct timer_list *t)
1537 {
1538         struct neigh_table *tbl = from_timer(tbl, t, proxy_timer);
1539         long sched_next = 0;
1540         unsigned long now = jiffies;
1541         struct sk_buff *skb, *n;
1542
1543         spin_lock(&tbl->proxy_queue.lock);
1544
1545         skb_queue_walk_safe(&tbl->proxy_queue, skb, n) {
1546                 long tdif = NEIGH_CB(skb)->sched_next - now;
1547
1548                 if (tdif <= 0) {
1549                         struct net_device *dev = skb->dev;
1550
1551                         __skb_unlink(skb, &tbl->proxy_queue);
1552                         if (tbl->proxy_redo && netif_running(dev)) {
1553                                 rcu_read_lock();
1554                                 tbl->proxy_redo(skb);
1555                                 rcu_read_unlock();
1556                         } else {
1557                                 kfree_skb(skb);
1558                         }
1559
1560                         dev_put(dev);
1561                 } else if (!sched_next || tdif < sched_next)
1562                         sched_next = tdif;
1563         }
1564         del_timer(&tbl->proxy_timer);
1565         if (sched_next)
1566                 mod_timer(&tbl->proxy_timer, jiffies + sched_next);
1567         spin_unlock(&tbl->proxy_queue.lock);
1568 }
1569
1570 void pneigh_enqueue(struct neigh_table *tbl, struct neigh_parms *p,
1571                     struct sk_buff *skb)
1572 {
1573         unsigned long sched_next = jiffies +
1574                         prandom_u32_max(NEIGH_VAR(p, PROXY_DELAY));
1575
1576         if (tbl->proxy_queue.qlen > NEIGH_VAR(p, PROXY_QLEN)) {
1577                 kfree_skb(skb);
1578                 return;
1579         }
1580
1581         NEIGH_CB(skb)->sched_next = sched_next;
1582         NEIGH_CB(skb)->flags |= LOCALLY_ENQUEUED;
1583
1584         spin_lock(&tbl->proxy_queue.lock);
1585         if (del_timer(&tbl->proxy_timer)) {
1586                 if (time_before(tbl->proxy_timer.expires, sched_next))
1587                         sched_next = tbl->proxy_timer.expires;
1588         }
1589         skb_dst_drop(skb);
1590         dev_hold(skb->dev);
1591         __skb_queue_tail(&tbl->proxy_queue, skb);
1592         mod_timer(&tbl->proxy_timer, sched_next);
1593         spin_unlock(&tbl->proxy_queue.lock);
1594 }
1595 EXPORT_SYMBOL(pneigh_enqueue);
1596
1597 static inline struct neigh_parms *lookup_neigh_parms(struct neigh_table *tbl,
1598                                                       struct net *net, int ifindex)
1599 {
1600         struct neigh_parms *p;
1601
1602         list_for_each_entry(p, &tbl->parms_list, list) {
1603                 if ((p->dev && p->dev->ifindex == ifindex && net_eq(neigh_parms_net(p), net)) ||
1604                     (!p->dev && !ifindex && net_eq(net, &init_net)))
1605                         return p;
1606         }
1607
1608         return NULL;
1609 }
1610
1611 struct neigh_parms *neigh_parms_alloc(struct net_device *dev,
1612                                       struct neigh_table *tbl)
1613 {
1614         struct neigh_parms *p;
1615         struct net *net = dev_net(dev);
1616         const struct net_device_ops *ops = dev->netdev_ops;
1617
1618         p = kmemdup(&tbl->parms, sizeof(*p), GFP_KERNEL);
1619         if (p) {
1620                 p->tbl            = tbl;
1621                 refcount_set(&p->refcnt, 1);
1622                 p->reachable_time =
1623                                 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
1624                 dev_hold(dev);
1625                 p->dev = dev;
1626                 write_pnet(&p->net, net);
1627                 p->sysctl_table = NULL;
1628
1629                 if (ops->ndo_neigh_setup && ops->ndo_neigh_setup(dev, p)) {
1630                         dev_put(dev);
1631                         kfree(p);
1632                         return NULL;
1633                 }
1634
1635                 write_lock_bh(&tbl->lock);
1636                 list_add(&p->list, &tbl->parms.list);
1637                 write_unlock_bh(&tbl->lock);
1638
1639                 neigh_parms_data_state_cleanall(p);
1640         }
1641         return p;
1642 }
1643 EXPORT_SYMBOL(neigh_parms_alloc);
1644
1645 static void neigh_rcu_free_parms(struct rcu_head *head)
1646 {
1647         struct neigh_parms *parms =
1648                 container_of(head, struct neigh_parms, rcu_head);
1649
1650         neigh_parms_put(parms);
1651 }
1652
1653 void neigh_parms_release(struct neigh_table *tbl, struct neigh_parms *parms)
1654 {
1655         if (!parms || parms == &tbl->parms)
1656                 return;
1657         write_lock_bh(&tbl->lock);
1658         list_del(&parms->list);
1659         parms->dead = 1;
1660         write_unlock_bh(&tbl->lock);
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
2532         /* 0 is already used to denote NDA_MASTER wasn't passed, therefore need another
2533          * invalid value for ifindex to denote "no master".
2534          */
2535         if (master_idx == -1)
2536                 return !!master;
2537
2538         if (!master || master->ifindex != master_idx)
2539                 return true;
2540
2541         return false;
2542 }
2543
2544 static bool neigh_ifindex_filtered(struct net_device *dev, int filter_idx)
2545 {
2546         if (filter_idx && (!dev || dev->ifindex != filter_idx))
2547                 return true;
2548
2549         return false;
2550 }
2551
2552 struct neigh_dump_filter {
2553         int master_idx;
2554         int dev_idx;
2555 };
2556
2557 static int neigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2558                             struct netlink_callback *cb,
2559                             struct neigh_dump_filter *filter)
2560 {
2561         struct net *net = sock_net(skb->sk);
2562         struct neighbour *n;
2563         int rc, h, s_h = cb->args[1];
2564         int idx, s_idx = idx = cb->args[2];
2565         struct neigh_hash_table *nht;
2566         unsigned int flags = NLM_F_MULTI;
2567
2568         if (filter->dev_idx || filter->master_idx)
2569                 flags |= NLM_F_DUMP_FILTERED;
2570
2571         rcu_read_lock_bh();
2572         nht = rcu_dereference_bh(tbl->nht);
2573
2574         for (h = s_h; h < (1 << nht->hash_shift); h++) {
2575                 if (h > s_h)
2576                         s_idx = 0;
2577                 for (n = rcu_dereference_bh(nht->hash_buckets[h]), idx = 0;
2578                      n != NULL;
2579                      n = rcu_dereference_bh(n->next)) {
2580                         if (idx < s_idx || !net_eq(dev_net(n->dev), net))
2581                                 goto next;
2582                         if (neigh_ifindex_filtered(n->dev, filter->dev_idx) ||
2583                             neigh_master_filtered(n->dev, filter->master_idx))
2584                                 goto next;
2585                         if (neigh_fill_info(skb, n, NETLINK_CB(cb->skb).portid,
2586                                             cb->nlh->nlmsg_seq,
2587                                             RTM_NEWNEIGH,
2588                                             flags) < 0) {
2589                                 rc = -1;
2590                                 goto out;
2591                         }
2592 next:
2593                         idx++;
2594                 }
2595         }
2596         rc = skb->len;
2597 out:
2598         rcu_read_unlock_bh();
2599         cb->args[1] = h;
2600         cb->args[2] = idx;
2601         return rc;
2602 }
2603
2604 static int pneigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2605                              struct netlink_callback *cb,
2606                              struct neigh_dump_filter *filter)
2607 {
2608         struct pneigh_entry *n;
2609         struct net *net = sock_net(skb->sk);
2610         int rc, h, s_h = cb->args[3];
2611         int idx, s_idx = idx = cb->args[4];
2612         unsigned int flags = NLM_F_MULTI;
2613
2614         if (filter->dev_idx || filter->master_idx)
2615                 flags |= NLM_F_DUMP_FILTERED;
2616
2617         read_lock_bh(&tbl->lock);
2618
2619         for (h = s_h; h <= PNEIGH_HASHMASK; h++) {
2620                 if (h > s_h)
2621                         s_idx = 0;
2622                 for (n = tbl->phash_buckets[h], idx = 0; n; n = n->next) {
2623                         if (idx < s_idx || pneigh_net(n) != net)
2624                                 goto next;
2625                         if (neigh_ifindex_filtered(n->dev, filter->dev_idx) ||
2626                             neigh_master_filtered(n->dev, filter->master_idx))
2627                                 goto next;
2628                         if (pneigh_fill_info(skb, n, NETLINK_CB(cb->skb).portid,
2629                                             cb->nlh->nlmsg_seq,
2630                                             RTM_NEWNEIGH, flags, tbl) < 0) {
2631                                 read_unlock_bh(&tbl->lock);
2632                                 rc = -1;
2633                                 goto out;
2634                         }
2635                 next:
2636                         idx++;
2637                 }
2638         }
2639
2640         read_unlock_bh(&tbl->lock);
2641         rc = skb->len;
2642 out:
2643         cb->args[3] = h;
2644         cb->args[4] = idx;
2645         return rc;
2646
2647 }
2648
2649 static int neigh_valid_dump_req(const struct nlmsghdr *nlh,
2650                                 bool strict_check,
2651                                 struct neigh_dump_filter *filter,
2652                                 struct netlink_ext_ack *extack)
2653 {
2654         struct nlattr *tb[NDA_MAX + 1];
2655         int err, i;
2656
2657         if (strict_check) {
2658                 struct ndmsg *ndm;
2659
2660                 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ndm))) {
2661                         NL_SET_ERR_MSG(extack, "Invalid header for neighbor dump request");
2662                         return -EINVAL;
2663                 }
2664
2665                 ndm = nlmsg_data(nlh);
2666                 if (ndm->ndm_pad1  || ndm->ndm_pad2  || ndm->ndm_ifindex ||
2667                     ndm->ndm_state || ndm->ndm_type) {
2668                         NL_SET_ERR_MSG(extack, "Invalid values in header for neighbor dump request");
2669                         return -EINVAL;
2670                 }
2671
2672                 if (ndm->ndm_flags & ~NTF_PROXY) {
2673                         NL_SET_ERR_MSG(extack, "Invalid flags in header for neighbor dump request");
2674                         return -EINVAL;
2675                 }
2676
2677                 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct ndmsg),
2678                                                     tb, NDA_MAX, nda_policy,
2679                                                     extack);
2680         } else {
2681                 err = nlmsg_parse_deprecated(nlh, sizeof(struct ndmsg), tb,
2682                                              NDA_MAX, nda_policy, extack);
2683         }
2684         if (err < 0)
2685                 return err;
2686
2687         for (i = 0; i <= NDA_MAX; ++i) {
2688                 if (!tb[i])
2689                         continue;
2690
2691                 /* all new attributes should require strict_check */
2692                 switch (i) {
2693                 case NDA_IFINDEX:
2694                         filter->dev_idx = nla_get_u32(tb[i]);
2695                         break;
2696                 case NDA_MASTER:
2697                         filter->master_idx = nla_get_u32(tb[i]);
2698                         break;
2699                 default:
2700                         if (strict_check) {
2701                                 NL_SET_ERR_MSG(extack, "Unsupported attribute in neighbor dump request");
2702                                 return -EINVAL;
2703                         }
2704                 }
2705         }
2706
2707         return 0;
2708 }
2709
2710 static int neigh_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2711 {
2712         const struct nlmsghdr *nlh = cb->nlh;
2713         struct neigh_dump_filter filter = {};
2714         struct neigh_table *tbl;
2715         int t, family, s_t;
2716         int proxy = 0;
2717         int err;
2718
2719         family = ((struct rtgenmsg *)nlmsg_data(nlh))->rtgen_family;
2720
2721         /* check for full ndmsg structure presence, family member is
2722          * the same for both structures
2723          */
2724         if (nlmsg_len(nlh) >= sizeof(struct ndmsg) &&
2725             ((struct ndmsg *)nlmsg_data(nlh))->ndm_flags == NTF_PROXY)
2726                 proxy = 1;
2727
2728         err = neigh_valid_dump_req(nlh, cb->strict_check, &filter, cb->extack);
2729         if (err < 0 && cb->strict_check)
2730                 return err;
2731
2732         s_t = cb->args[0];
2733
2734         for (t = 0; t < NEIGH_NR_TABLES; t++) {
2735                 tbl = neigh_tables[t];
2736
2737                 if (!tbl)
2738                         continue;
2739                 if (t < s_t || (family && tbl->family != family))
2740                         continue;
2741                 if (t > s_t)
2742                         memset(&cb->args[1], 0, sizeof(cb->args) -
2743                                                 sizeof(cb->args[0]));
2744                 if (proxy)
2745                         err = pneigh_dump_table(tbl, skb, cb, &filter);
2746                 else
2747                         err = neigh_dump_table(tbl, skb, cb, &filter);
2748                 if (err < 0)
2749                         break;
2750         }
2751
2752         cb->args[0] = t;
2753         return skb->len;
2754 }
2755
2756 static int neigh_valid_get_req(const struct nlmsghdr *nlh,
2757                                struct neigh_table **tbl,
2758                                void **dst, int *dev_idx, u8 *ndm_flags,
2759                                struct netlink_ext_ack *extack)
2760 {
2761         struct nlattr *tb[NDA_MAX + 1];
2762         struct ndmsg *ndm;
2763         int err, i;
2764
2765         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ndm))) {
2766                 NL_SET_ERR_MSG(extack, "Invalid header for neighbor get request");
2767                 return -EINVAL;
2768         }
2769
2770         ndm = nlmsg_data(nlh);
2771         if (ndm->ndm_pad1  || ndm->ndm_pad2  || ndm->ndm_state ||
2772             ndm->ndm_type) {
2773                 NL_SET_ERR_MSG(extack, "Invalid values in header for neighbor get request");
2774                 return -EINVAL;
2775         }
2776
2777         if (ndm->ndm_flags & ~NTF_PROXY) {
2778                 NL_SET_ERR_MSG(extack, "Invalid flags in header for neighbor get request");
2779                 return -EINVAL;
2780         }
2781
2782         err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct ndmsg), tb,
2783                                             NDA_MAX, nda_policy, extack);
2784         if (err < 0)
2785                 return err;
2786
2787         *ndm_flags = ndm->ndm_flags;
2788         *dev_idx = ndm->ndm_ifindex;
2789         *tbl = neigh_find_table(ndm->ndm_family);
2790         if (*tbl == NULL) {
2791                 NL_SET_ERR_MSG(extack, "Unsupported family in header for neighbor get request");
2792                 return -EAFNOSUPPORT;
2793         }
2794
2795         for (i = 0; i <= NDA_MAX; ++i) {
2796                 if (!tb[i])
2797                         continue;
2798
2799                 switch (i) {
2800                 case NDA_DST:
2801                         if (nla_len(tb[i]) != (int)(*tbl)->key_len) {
2802                                 NL_SET_ERR_MSG(extack, "Invalid network address in neighbor get request");
2803                                 return -EINVAL;
2804                         }
2805                         *dst = nla_data(tb[i]);
2806                         break;
2807                 default:
2808                         NL_SET_ERR_MSG(extack, "Unsupported attribute in neighbor get request");
2809                         return -EINVAL;
2810                 }
2811         }
2812
2813         return 0;
2814 }
2815
2816 static inline size_t neigh_nlmsg_size(void)
2817 {
2818         return NLMSG_ALIGN(sizeof(struct ndmsg))
2819                + nla_total_size(MAX_ADDR_LEN) /* NDA_DST */
2820                + nla_total_size(MAX_ADDR_LEN) /* NDA_LLADDR */
2821                + nla_total_size(sizeof(struct nda_cacheinfo))
2822                + nla_total_size(4)  /* NDA_PROBES */
2823                + nla_total_size(1); /* NDA_PROTOCOL */
2824 }
2825
2826 static int neigh_get_reply(struct net *net, struct neighbour *neigh,
2827                            u32 pid, u32 seq)
2828 {
2829         struct sk_buff *skb;
2830         int err = 0;
2831
2832         skb = nlmsg_new(neigh_nlmsg_size(), GFP_KERNEL);
2833         if (!skb)
2834                 return -ENOBUFS;
2835
2836         err = neigh_fill_info(skb, neigh, pid, seq, RTM_NEWNEIGH, 0);
2837         if (err) {
2838                 kfree_skb(skb);
2839                 goto errout;
2840         }
2841
2842         err = rtnl_unicast(skb, net, pid);
2843 errout:
2844         return err;
2845 }
2846
2847 static inline size_t pneigh_nlmsg_size(void)
2848 {
2849         return NLMSG_ALIGN(sizeof(struct ndmsg))
2850                + nla_total_size(MAX_ADDR_LEN) /* NDA_DST */
2851                + nla_total_size(1); /* NDA_PROTOCOL */
2852 }
2853
2854 static int pneigh_get_reply(struct net *net, struct pneigh_entry *neigh,
2855                             u32 pid, u32 seq, struct neigh_table *tbl)
2856 {
2857         struct sk_buff *skb;
2858         int err = 0;
2859
2860         skb = nlmsg_new(pneigh_nlmsg_size(), GFP_KERNEL);
2861         if (!skb)
2862                 return -ENOBUFS;
2863
2864         err = pneigh_fill_info(skb, neigh, pid, seq, RTM_NEWNEIGH, 0, tbl);
2865         if (err) {
2866                 kfree_skb(skb);
2867                 goto errout;
2868         }
2869
2870         err = rtnl_unicast(skb, net, pid);
2871 errout:
2872         return err;
2873 }
2874
2875 static int neigh_get(struct sk_buff *in_skb, struct nlmsghdr *nlh,
2876                      struct netlink_ext_ack *extack)
2877 {
2878         struct net *net = sock_net(in_skb->sk);
2879         struct net_device *dev = NULL;
2880         struct neigh_table *tbl = NULL;
2881         struct neighbour *neigh;
2882         void *dst = NULL;
2883         u8 ndm_flags = 0;
2884         int dev_idx = 0;
2885         int err;
2886
2887         err = neigh_valid_get_req(nlh, &tbl, &dst, &dev_idx, &ndm_flags,
2888                                   extack);
2889         if (err < 0)
2890                 return err;
2891
2892         if (dev_idx) {
2893                 dev = __dev_get_by_index(net, dev_idx);
2894                 if (!dev) {
2895                         NL_SET_ERR_MSG(extack, "Unknown device ifindex");
2896                         return -ENODEV;
2897                 }
2898         }
2899
2900         if (!dst) {
2901                 NL_SET_ERR_MSG(extack, "Network address not specified");
2902                 return -EINVAL;
2903         }
2904
2905         if (ndm_flags & NTF_PROXY) {
2906                 struct pneigh_entry *pn;
2907
2908                 pn = pneigh_lookup(tbl, net, dst, dev, 0);
2909                 if (!pn) {
2910                         NL_SET_ERR_MSG(extack, "Proxy neighbour entry not found");
2911                         return -ENOENT;
2912                 }
2913                 return pneigh_get_reply(net, pn, NETLINK_CB(in_skb).portid,
2914                                         nlh->nlmsg_seq, tbl);
2915         }
2916
2917         if (!dev) {
2918                 NL_SET_ERR_MSG(extack, "No device specified");
2919                 return -EINVAL;
2920         }
2921
2922         neigh = neigh_lookup(tbl, dst, dev);
2923         if (!neigh) {
2924                 NL_SET_ERR_MSG(extack, "Neighbour entry not found");
2925                 return -ENOENT;
2926         }
2927
2928         err = neigh_get_reply(net, neigh, NETLINK_CB(in_skb).portid,
2929                               nlh->nlmsg_seq);
2930
2931         neigh_release(neigh);
2932
2933         return err;
2934 }
2935
2936 void neigh_for_each(struct neigh_table *tbl, void (*cb)(struct neighbour *, void *), void *cookie)
2937 {
2938         int chain;
2939         struct neigh_hash_table *nht;
2940
2941         rcu_read_lock_bh();
2942         nht = rcu_dereference_bh(tbl->nht);
2943
2944         read_lock(&tbl->lock); /* avoid resizes */
2945         for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
2946                 struct neighbour *n;
2947
2948                 for (n = rcu_dereference_bh(nht->hash_buckets[chain]);
2949                      n != NULL;
2950                      n = rcu_dereference_bh(n->next))
2951                         cb(n, cookie);
2952         }
2953         read_unlock(&tbl->lock);
2954         rcu_read_unlock_bh();
2955 }
2956 EXPORT_SYMBOL(neigh_for_each);
2957
2958 /* The tbl->lock must be held as a writer and BH disabled. */
2959 void __neigh_for_each_release(struct neigh_table *tbl,
2960                               int (*cb)(struct neighbour *))
2961 {
2962         int chain;
2963         struct neigh_hash_table *nht;
2964
2965         nht = rcu_dereference_protected(tbl->nht,
2966                                         lockdep_is_held(&tbl->lock));
2967         for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
2968                 struct neighbour *n;
2969                 struct neighbour __rcu **np;
2970
2971                 np = &nht->hash_buckets[chain];
2972                 while ((n = rcu_dereference_protected(*np,
2973                                         lockdep_is_held(&tbl->lock))) != NULL) {
2974                         int release;
2975
2976                         write_lock(&n->lock);
2977                         release = cb(n);
2978                         if (release) {
2979                                 rcu_assign_pointer(*np,
2980                                         rcu_dereference_protected(n->next,
2981                                                 lockdep_is_held(&tbl->lock)));
2982                                 neigh_mark_dead(n);
2983                         } else
2984                                 np = &n->next;
2985                         write_unlock(&n->lock);
2986                         if (release)
2987                                 neigh_cleanup_and_release(n);
2988                 }
2989         }
2990 }
2991 EXPORT_SYMBOL(__neigh_for_each_release);
2992
2993 int neigh_xmit(int index, struct net_device *dev,
2994                const void *addr, struct sk_buff *skb)
2995 {
2996         int err = -EAFNOSUPPORT;
2997         if (likely(index < NEIGH_NR_TABLES)) {
2998                 struct neigh_table *tbl;
2999                 struct neighbour *neigh;
3000
3001                 tbl = neigh_tables[index];
3002                 if (!tbl)
3003                         goto out;
3004                 rcu_read_lock_bh();
3005                 if (index == NEIGH_ARP_TABLE) {
3006                         u32 key = *((u32 *)addr);
3007
3008                         neigh = __ipv4_neigh_lookup_noref(dev, key);
3009                 } else {
3010                         neigh = __neigh_lookup_noref(tbl, addr, dev);
3011                 }
3012                 if (!neigh)
3013                         neigh = __neigh_create(tbl, addr, dev, false);
3014                 err = PTR_ERR(neigh);
3015                 if (IS_ERR(neigh)) {
3016                         rcu_read_unlock_bh();
3017                         goto out_kfree_skb;
3018                 }
3019                 err = neigh->output(neigh, skb);
3020                 rcu_read_unlock_bh();
3021         }
3022         else if (index == NEIGH_LINK_TABLE) {
3023                 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
3024                                       addr, NULL, skb->len);
3025                 if (err < 0)
3026                         goto out_kfree_skb;
3027                 err = dev_queue_xmit(skb);
3028         }
3029 out:
3030         return err;
3031 out_kfree_skb:
3032         kfree_skb(skb);
3033         goto out;
3034 }
3035 EXPORT_SYMBOL(neigh_xmit);
3036
3037 #ifdef CONFIG_PROC_FS
3038
3039 static struct neighbour *neigh_get_first(struct seq_file *seq)
3040 {
3041         struct neigh_seq_state *state = seq->private;
3042         struct net *net = seq_file_net(seq);
3043         struct neigh_hash_table *nht = state->nht;
3044         struct neighbour *n = NULL;
3045         int bucket;
3046
3047         state->flags &= ~NEIGH_SEQ_IS_PNEIGH;
3048         for (bucket = 0; bucket < (1 << nht->hash_shift); bucket++) {
3049                 n = rcu_dereference_bh(nht->hash_buckets[bucket]);
3050
3051                 while (n) {
3052                         if (!net_eq(dev_net(n->dev), net))
3053                                 goto next;
3054                         if (state->neigh_sub_iter) {
3055                                 loff_t fakep = 0;
3056                                 void *v;
3057
3058                                 v = state->neigh_sub_iter(state, n, &fakep);
3059                                 if (!v)
3060                                         goto next;
3061                         }
3062                         if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
3063                                 break;
3064                         if (n->nud_state & ~NUD_NOARP)
3065                                 break;
3066 next:
3067                         n = rcu_dereference_bh(n->next);
3068                 }
3069
3070                 if (n)
3071                         break;
3072         }
3073         state->bucket = bucket;
3074
3075         return n;
3076 }
3077
3078 static struct neighbour *neigh_get_next(struct seq_file *seq,
3079                                         struct neighbour *n,
3080                                         loff_t *pos)
3081 {
3082         struct neigh_seq_state *state = seq->private;
3083         struct net *net = seq_file_net(seq);
3084         struct neigh_hash_table *nht = state->nht;
3085
3086         if (state->neigh_sub_iter) {
3087                 void *v = state->neigh_sub_iter(state, n, pos);
3088                 if (v)
3089                         return n;
3090         }
3091         n = rcu_dereference_bh(n->next);
3092
3093         while (1) {
3094                 while (n) {
3095                         if (!net_eq(dev_net(n->dev), net))
3096                                 goto next;
3097                         if (state->neigh_sub_iter) {
3098                                 void *v = state->neigh_sub_iter(state, n, pos);
3099                                 if (v)
3100                                         return n;
3101                                 goto next;
3102                         }
3103                         if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
3104                                 break;
3105
3106                         if (n->nud_state & ~NUD_NOARP)
3107                                 break;
3108 next:
3109                         n = rcu_dereference_bh(n->next);
3110                 }
3111
3112                 if (n)
3113                         break;
3114
3115                 if (++state->bucket >= (1 << nht->hash_shift))
3116                         break;
3117
3118                 n = rcu_dereference_bh(nht->hash_buckets[state->bucket]);
3119         }
3120
3121         if (n && pos)
3122                 --(*pos);
3123         return n;
3124 }
3125
3126 static struct neighbour *neigh_get_idx(struct seq_file *seq, loff_t *pos)
3127 {
3128         struct neighbour *n = neigh_get_first(seq);
3129
3130         if (n) {
3131                 --(*pos);
3132                 while (*pos) {
3133                         n = neigh_get_next(seq, n, pos);
3134                         if (!n)
3135                                 break;
3136                 }
3137         }
3138         return *pos ? NULL : n;
3139 }
3140
3141 static struct pneigh_entry *pneigh_get_first(struct seq_file *seq)
3142 {
3143         struct neigh_seq_state *state = seq->private;
3144         struct net *net = seq_file_net(seq);
3145         struct neigh_table *tbl = state->tbl;
3146         struct pneigh_entry *pn = NULL;
3147         int bucket;
3148
3149         state->flags |= NEIGH_SEQ_IS_PNEIGH;
3150         for (bucket = 0; bucket <= PNEIGH_HASHMASK; bucket++) {
3151                 pn = tbl->phash_buckets[bucket];
3152                 while (pn && !net_eq(pneigh_net(pn), net))
3153                         pn = pn->next;
3154                 if (pn)
3155                         break;
3156         }
3157         state->bucket = bucket;
3158
3159         return pn;
3160 }
3161
3162 static struct pneigh_entry *pneigh_get_next(struct seq_file *seq,
3163                                             struct pneigh_entry *pn,
3164                                             loff_t *pos)
3165 {
3166         struct neigh_seq_state *state = seq->private;
3167         struct net *net = seq_file_net(seq);
3168         struct neigh_table *tbl = state->tbl;
3169
3170         do {
3171                 pn = pn->next;
3172         } while (pn && !net_eq(pneigh_net(pn), net));
3173
3174         while (!pn) {
3175                 if (++state->bucket > PNEIGH_HASHMASK)
3176                         break;
3177                 pn = tbl->phash_buckets[state->bucket];
3178                 while (pn && !net_eq(pneigh_net(pn), net))
3179                         pn = pn->next;
3180                 if (pn)
3181                         break;
3182         }
3183
3184         if (pn && pos)
3185                 --(*pos);
3186
3187         return pn;
3188 }
3189
3190 static struct pneigh_entry *pneigh_get_idx(struct seq_file *seq, loff_t *pos)
3191 {
3192         struct pneigh_entry *pn = pneigh_get_first(seq);
3193
3194         if (pn) {
3195                 --(*pos);
3196                 while (*pos) {
3197                         pn = pneigh_get_next(seq, pn, pos);
3198                         if (!pn)
3199                                 break;
3200                 }
3201         }
3202         return *pos ? NULL : pn;
3203 }
3204
3205 static void *neigh_get_idx_any(struct seq_file *seq, loff_t *pos)
3206 {
3207         struct neigh_seq_state *state = seq->private;
3208         void *rc;
3209         loff_t idxpos = *pos;
3210
3211         rc = neigh_get_idx(seq, &idxpos);
3212         if (!rc && !(state->flags & NEIGH_SEQ_NEIGH_ONLY))
3213                 rc = pneigh_get_idx(seq, &idxpos);
3214
3215         return rc;
3216 }
3217
3218 void *neigh_seq_start(struct seq_file *seq, loff_t *pos, struct neigh_table *tbl, unsigned int neigh_seq_flags)
3219         __acquires(tbl->lock)
3220         __acquires(rcu_bh)
3221 {
3222         struct neigh_seq_state *state = seq->private;
3223
3224         state->tbl = tbl;
3225         state->bucket = 0;
3226         state->flags = (neigh_seq_flags & ~NEIGH_SEQ_IS_PNEIGH);
3227
3228         rcu_read_lock_bh();
3229         state->nht = rcu_dereference_bh(tbl->nht);
3230         read_lock(&tbl->lock);
3231
3232         return *pos ? neigh_get_idx_any(seq, pos) : SEQ_START_TOKEN;
3233 }
3234 EXPORT_SYMBOL(neigh_seq_start);
3235
3236 void *neigh_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3237 {
3238         struct neigh_seq_state *state;
3239         void *rc;
3240
3241         if (v == SEQ_START_TOKEN) {
3242                 rc = neigh_get_first(seq);
3243                 goto out;
3244         }
3245
3246         state = seq->private;
3247         if (!(state->flags & NEIGH_SEQ_IS_PNEIGH)) {
3248                 rc = neigh_get_next(seq, v, NULL);
3249                 if (rc)
3250                         goto out;
3251                 if (!(state->flags & NEIGH_SEQ_NEIGH_ONLY))
3252                         rc = pneigh_get_first(seq);
3253         } else {
3254                 BUG_ON(state->flags & NEIGH_SEQ_NEIGH_ONLY);
3255                 rc = pneigh_get_next(seq, v, NULL);
3256         }
3257 out:
3258         ++(*pos);
3259         return rc;
3260 }
3261 EXPORT_SYMBOL(neigh_seq_next);
3262
3263 void neigh_seq_stop(struct seq_file *seq, void *v)
3264         __releases(tbl->lock)
3265         __releases(rcu_bh)
3266 {
3267         struct neigh_seq_state *state = seq->private;
3268         struct neigh_table *tbl = state->tbl;
3269
3270         read_unlock(&tbl->lock);
3271         rcu_read_unlock_bh();
3272 }
3273 EXPORT_SYMBOL(neigh_seq_stop);
3274
3275 /* statistics via seq_file */
3276
3277 static void *neigh_stat_seq_start(struct seq_file *seq, loff_t *pos)
3278 {
3279         struct neigh_table *tbl = PDE_DATA(file_inode(seq->file));
3280         int cpu;
3281
3282         if (*pos == 0)
3283                 return SEQ_START_TOKEN;
3284
3285         for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
3286                 if (!cpu_possible(cpu))
3287                         continue;
3288                 *pos = cpu+1;
3289                 return per_cpu_ptr(tbl->stats, cpu);
3290         }
3291         return NULL;
3292 }
3293
3294 static void *neigh_stat_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3295 {
3296         struct neigh_table *tbl = PDE_DATA(file_inode(seq->file));
3297         int cpu;
3298
3299         for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
3300                 if (!cpu_possible(cpu))
3301                         continue;
3302                 *pos = cpu+1;
3303                 return per_cpu_ptr(tbl->stats, cpu);
3304         }
3305         (*pos)++;
3306         return NULL;
3307 }
3308
3309 static void neigh_stat_seq_stop(struct seq_file *seq, void *v)
3310 {
3311
3312 }
3313
3314 static int neigh_stat_seq_show(struct seq_file *seq, void *v)
3315 {
3316         struct neigh_table *tbl = PDE_DATA(file_inode(seq->file));
3317         struct neigh_statistics *st = v;
3318
3319         if (v == SEQ_START_TOKEN) {
3320                 seq_puts(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");
3321                 return 0;
3322         }
3323
3324         seq_printf(seq, "%08x %08lx %08lx %08lx   %08lx %08lx %08lx   "
3325                         "%08lx         %08lx         %08lx         "
3326                         "%08lx       %08lx            %08lx\n",
3327                    atomic_read(&tbl->entries),
3328
3329                    st->allocs,
3330                    st->destroys,
3331                    st->hash_grows,
3332
3333                    st->lookups,
3334                    st->hits,
3335
3336                    st->res_failed,
3337
3338                    st->rcv_probes_mcast,
3339                    st->rcv_probes_ucast,
3340
3341                    st->periodic_gc_runs,
3342                    st->forced_gc_runs,
3343                    st->unres_discards,
3344                    st->table_fulls
3345                    );
3346
3347         return 0;
3348 }
3349
3350 static const struct seq_operations neigh_stat_seq_ops = {
3351         .start  = neigh_stat_seq_start,
3352         .next   = neigh_stat_seq_next,
3353         .stop   = neigh_stat_seq_stop,
3354         .show   = neigh_stat_seq_show,
3355 };
3356 #endif /* CONFIG_PROC_FS */
3357
3358 static void __neigh_notify(struct neighbour *n, int type, int flags,
3359                            u32 pid)
3360 {
3361         struct net *net = dev_net(n->dev);
3362         struct sk_buff *skb;
3363         int err = -ENOBUFS;
3364
3365         skb = nlmsg_new(neigh_nlmsg_size(), GFP_ATOMIC);
3366         if (skb == NULL)
3367                 goto errout;
3368
3369         err = neigh_fill_info(skb, n, pid, 0, type, flags);
3370         if (err < 0) {
3371                 /* -EMSGSIZE implies BUG in neigh_nlmsg_size() */
3372                 WARN_ON(err == -EMSGSIZE);
3373                 kfree_skb(skb);
3374                 goto errout;
3375         }
3376         rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
3377         return;
3378 errout:
3379         if (err < 0)
3380                 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
3381 }
3382
3383 void neigh_app_ns(struct neighbour *n)
3384 {
3385         __neigh_notify(n, RTM_GETNEIGH, NLM_F_REQUEST, 0);
3386 }
3387 EXPORT_SYMBOL(neigh_app_ns);
3388
3389 #ifdef CONFIG_SYSCTL
3390 static int unres_qlen_max = INT_MAX / SKB_TRUESIZE(ETH_FRAME_LEN);
3391
3392 static int proc_unres_qlen(struct ctl_table *ctl, int write,
3393                            void *buffer, size_t *lenp, loff_t *ppos)
3394 {
3395         int size, ret;
3396         struct ctl_table tmp = *ctl;
3397
3398         tmp.extra1 = SYSCTL_ZERO;
3399         tmp.extra2 = &unres_qlen_max;
3400         tmp.data = &size;
3401
3402         size = *(int *)ctl->data / SKB_TRUESIZE(ETH_FRAME_LEN);
3403         ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
3404
3405         if (write && !ret)
3406                 *(int *)ctl->data = size * SKB_TRUESIZE(ETH_FRAME_LEN);
3407         return ret;
3408 }
3409
3410 static struct neigh_parms *neigh_get_dev_parms_rcu(struct net_device *dev,
3411                                                    int family)
3412 {
3413         switch (family) {
3414         case AF_INET:
3415                 return __in_dev_arp_parms_get_rcu(dev);
3416         case AF_INET6:
3417                 return __in6_dev_nd_parms_get_rcu(dev);
3418         }
3419         return NULL;
3420 }
3421
3422 static void neigh_copy_dflt_parms(struct net *net, struct neigh_parms *p,
3423                                   int index)
3424 {
3425         struct net_device *dev;
3426         int family = neigh_parms_family(p);
3427
3428         rcu_read_lock();
3429         for_each_netdev_rcu(net, dev) {
3430                 struct neigh_parms *dst_p =
3431                                 neigh_get_dev_parms_rcu(dev, family);
3432
3433                 if (dst_p && !test_bit(index, dst_p->data_state))
3434                         dst_p->data[index] = p->data[index];
3435         }
3436         rcu_read_unlock();
3437 }
3438
3439 static void neigh_proc_update(struct ctl_table *ctl, int write)
3440 {
3441         struct net_device *dev = ctl->extra1;
3442         struct neigh_parms *p = ctl->extra2;
3443         struct net *net = neigh_parms_net(p);
3444         int index = (int *) ctl->data - p->data;
3445
3446         if (!write)
3447                 return;
3448
3449         set_bit(index, p->data_state);
3450         if (index == NEIGH_VAR_DELAY_PROBE_TIME)
3451                 call_netevent_notifiers(NETEVENT_DELAY_PROBE_TIME_UPDATE, p);
3452         if (!dev) /* NULL dev means this is default value */
3453                 neigh_copy_dflt_parms(net, p, index);
3454 }
3455
3456 static int neigh_proc_dointvec_zero_intmax(struct ctl_table *ctl, int write,
3457                                            void *buffer, size_t *lenp,
3458                                            loff_t *ppos)
3459 {
3460         struct ctl_table tmp = *ctl;
3461         int ret;
3462
3463         tmp.extra1 = SYSCTL_ZERO;
3464         tmp.extra2 = SYSCTL_INT_MAX;
3465
3466         ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
3467         neigh_proc_update(ctl, write);
3468         return ret;
3469 }
3470
3471 int neigh_proc_dointvec(struct ctl_table *ctl, int write, void *buffer,
3472                         size_t *lenp, loff_t *ppos)
3473 {
3474         int ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
3475
3476         neigh_proc_update(ctl, write);
3477         return ret;
3478 }
3479 EXPORT_SYMBOL(neigh_proc_dointvec);
3480
3481 int neigh_proc_dointvec_jiffies(struct ctl_table *ctl, int write, void *buffer,
3482                                 size_t *lenp, loff_t *ppos)
3483 {
3484         int ret = proc_dointvec_jiffies(ctl, write, buffer, lenp, ppos);
3485
3486         neigh_proc_update(ctl, write);
3487         return ret;
3488 }
3489 EXPORT_SYMBOL(neigh_proc_dointvec_jiffies);
3490
3491 static int neigh_proc_dointvec_userhz_jiffies(struct ctl_table *ctl, int write,
3492                                               void *buffer, size_t *lenp,
3493                                               loff_t *ppos)
3494 {
3495         int ret = proc_dointvec_userhz_jiffies(ctl, write, buffer, lenp, ppos);
3496
3497         neigh_proc_update(ctl, write);
3498         return ret;
3499 }
3500
3501 int neigh_proc_dointvec_ms_jiffies(struct ctl_table *ctl, int write,
3502                                    void *buffer, size_t *lenp, loff_t *ppos)
3503 {
3504         int ret = proc_dointvec_ms_jiffies(ctl, write, buffer, lenp, ppos);
3505
3506         neigh_proc_update(ctl, write);
3507         return ret;
3508 }
3509 EXPORT_SYMBOL(neigh_proc_dointvec_ms_jiffies);
3510
3511 static int neigh_proc_dointvec_unres_qlen(struct ctl_table *ctl, int write,
3512                                           void *buffer, size_t *lenp,
3513                                           loff_t *ppos)
3514 {
3515         int ret = proc_unres_qlen(ctl, write, buffer, lenp, ppos);
3516
3517         neigh_proc_update(ctl, write);
3518         return ret;
3519 }
3520
3521 static int neigh_proc_base_reachable_time(struct ctl_table *ctl, int write,
3522                                           void *buffer, size_t *lenp,
3523                                           loff_t *ppos)
3524 {
3525         struct neigh_parms *p = ctl->extra2;
3526         int ret;
3527
3528         if (strcmp(ctl->procname, "base_reachable_time") == 0)
3529                 ret = neigh_proc_dointvec_jiffies(ctl, write, buffer, lenp, ppos);
3530         else if (strcmp(ctl->procname, "base_reachable_time_ms") == 0)
3531                 ret = neigh_proc_dointvec_ms_jiffies(ctl, write, buffer, lenp, ppos);
3532         else
3533                 ret = -1;
3534
3535         if (write && ret == 0) {
3536                 /* update reachable_time as well, otherwise, the change will
3537                  * only be effective after the next time neigh_periodic_work
3538                  * decides to recompute it
3539                  */
3540                 p->reachable_time =
3541                         neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
3542         }
3543         return ret;
3544 }
3545
3546 #define NEIGH_PARMS_DATA_OFFSET(index)  \
3547         (&((struct neigh_parms *) 0)->data[index])
3548
3549 #define NEIGH_SYSCTL_ENTRY(attr, data_attr, name, mval, proc) \
3550         [NEIGH_VAR_ ## attr] = { \
3551                 .procname       = name, \
3552                 .data           = NEIGH_PARMS_DATA_OFFSET(NEIGH_VAR_ ## data_attr), \
3553                 .maxlen         = sizeof(int), \
3554                 .mode           = mval, \
3555                 .proc_handler   = proc, \
3556         }
3557
3558 #define NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(attr, name) \
3559         NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_zero_intmax)
3560
3561 #define NEIGH_SYSCTL_JIFFIES_ENTRY(attr, name) \
3562         NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_jiffies)
3563
3564 #define NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(attr, name) \
3565         NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_userhz_jiffies)
3566
3567 #define NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(attr, data_attr, name) \
3568         NEIGH_SYSCTL_ENTRY(attr, data_attr, name, 0644, neigh_proc_dointvec_ms_jiffies)
3569
3570 #define NEIGH_SYSCTL_UNRES_QLEN_REUSED_ENTRY(attr, data_attr, name) \
3571         NEIGH_SYSCTL_ENTRY(attr, data_attr, name, 0644, neigh_proc_dointvec_unres_qlen)
3572
3573 static struct neigh_sysctl_table {
3574         struct ctl_table_header *sysctl_header;
3575         struct ctl_table neigh_vars[NEIGH_VAR_MAX + 1];
3576 } neigh_sysctl_template __read_mostly = {
3577         .neigh_vars = {
3578                 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(MCAST_PROBES, "mcast_solicit"),
3579                 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(UCAST_PROBES, "ucast_solicit"),
3580                 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(APP_PROBES, "app_solicit"),
3581                 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(MCAST_REPROBES, "mcast_resolicit"),
3582                 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(RETRANS_TIME, "retrans_time"),
3583                 NEIGH_SYSCTL_JIFFIES_ENTRY(BASE_REACHABLE_TIME, "base_reachable_time"),
3584                 NEIGH_SYSCTL_JIFFIES_ENTRY(DELAY_PROBE_TIME, "delay_first_probe_time"),
3585                 NEIGH_SYSCTL_JIFFIES_ENTRY(GC_STALETIME, "gc_stale_time"),
3586                 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(QUEUE_LEN_BYTES, "unres_qlen_bytes"),
3587                 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(PROXY_QLEN, "proxy_qlen"),
3588                 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(ANYCAST_DELAY, "anycast_delay"),
3589                 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(PROXY_DELAY, "proxy_delay"),
3590                 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(LOCKTIME, "locktime"),
3591                 NEIGH_SYSCTL_UNRES_QLEN_REUSED_ENTRY(QUEUE_LEN, QUEUE_LEN_BYTES, "unres_qlen"),
3592                 NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(RETRANS_TIME_MS, RETRANS_TIME, "retrans_time_ms"),
3593                 NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(BASE_REACHABLE_TIME_MS, BASE_REACHABLE_TIME, "base_reachable_time_ms"),
3594                 [NEIGH_VAR_GC_INTERVAL] = {
3595                         .procname       = "gc_interval",
3596                         .maxlen         = sizeof(int),
3597                         .mode           = 0644,
3598                         .proc_handler   = proc_dointvec_jiffies,
3599                 },
3600                 [NEIGH_VAR_GC_THRESH1] = {
3601                         .procname       = "gc_thresh1",
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_THRESH2] = {
3609                         .procname       = "gc_thresh2",
3610                         .maxlen         = sizeof(int),
3611                         .mode           = 0644,
3612                         .extra1         = SYSCTL_ZERO,
3613                         .extra2         = SYSCTL_INT_MAX,
3614                         .proc_handler   = proc_dointvec_minmax,
3615                 },
3616                 [NEIGH_VAR_GC_THRESH3] = {
3617                         .procname       = "gc_thresh3",
3618                         .maxlen         = sizeof(int),
3619                         .mode           = 0644,
3620                         .extra1         = SYSCTL_ZERO,
3621                         .extra2         = SYSCTL_INT_MAX,
3622                         .proc_handler   = proc_dointvec_minmax,
3623                 },
3624                 {},
3625         },
3626 };
3627
3628 int neigh_sysctl_register(struct net_device *dev, struct neigh_parms *p,
3629                           proc_handler *handler)
3630 {
3631         int i;
3632         struct neigh_sysctl_table *t;
3633         const char *dev_name_source;
3634         char neigh_path[ sizeof("net//neigh/") + IFNAMSIZ + IFNAMSIZ ];
3635         char *p_name;
3636
3637         t = kmemdup(&neigh_sysctl_template, sizeof(*t), GFP_KERNEL);
3638         if (!t)
3639                 goto err;
3640
3641         for (i = 0; i < NEIGH_VAR_GC_INTERVAL; i++) {
3642                 t->neigh_vars[i].data += (long) p;
3643                 t->neigh_vars[i].extra1 = dev;
3644                 t->neigh_vars[i].extra2 = p;
3645         }
3646
3647         if (dev) {
3648                 dev_name_source = dev->name;
3649                 /* Terminate the table early */
3650                 memset(&t->neigh_vars[NEIGH_VAR_GC_INTERVAL], 0,
3651                        sizeof(t->neigh_vars[NEIGH_VAR_GC_INTERVAL]));
3652         } else {
3653                 struct neigh_table *tbl = p->tbl;
3654                 dev_name_source = "default";
3655                 t->neigh_vars[NEIGH_VAR_GC_INTERVAL].data = &tbl->gc_interval;
3656                 t->neigh_vars[NEIGH_VAR_GC_THRESH1].data = &tbl->gc_thresh1;
3657                 t->neigh_vars[NEIGH_VAR_GC_THRESH2].data = &tbl->gc_thresh2;
3658                 t->neigh_vars[NEIGH_VAR_GC_THRESH3].data = &tbl->gc_thresh3;
3659         }
3660
3661         if (handler) {
3662                 /* RetransTime */
3663                 t->neigh_vars[NEIGH_VAR_RETRANS_TIME].proc_handler = handler;
3664                 /* ReachableTime */
3665                 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].proc_handler = handler;
3666                 /* RetransTime (in milliseconds)*/
3667                 t->neigh_vars[NEIGH_VAR_RETRANS_TIME_MS].proc_handler = handler;
3668                 /* ReachableTime (in milliseconds) */
3669                 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].proc_handler = handler;
3670         } else {
3671                 /* Those handlers will update p->reachable_time after
3672                  * base_reachable_time(_ms) is set to ensure the new timer starts being
3673                  * applied after the next neighbour update instead of waiting for
3674                  * neigh_periodic_work to update its value (can be multiple minutes)
3675                  * So any handler that replaces them should do this as well
3676                  */
3677                 /* ReachableTime */
3678                 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].proc_handler =
3679                         neigh_proc_base_reachable_time;
3680                 /* ReachableTime (in milliseconds) */
3681                 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].proc_handler =
3682                         neigh_proc_base_reachable_time;
3683         }
3684
3685         /* Don't export sysctls to unprivileged users */
3686         if (neigh_parms_net(p)->user_ns != &init_user_ns)
3687                 t->neigh_vars[0].procname = NULL;
3688
3689         switch (neigh_parms_family(p)) {
3690         case AF_INET:
3691               p_name = "ipv4";
3692               break;
3693         case AF_INET6:
3694               p_name = "ipv6";
3695               break;
3696         default:
3697               BUG();
3698         }
3699
3700         snprintf(neigh_path, sizeof(neigh_path), "net/%s/neigh/%s",
3701                 p_name, dev_name_source);
3702         t->sysctl_header =
3703                 register_net_sysctl(neigh_parms_net(p), neigh_path, t->neigh_vars);
3704         if (!t->sysctl_header)
3705                 goto free;
3706
3707         p->sysctl_table = t;
3708         return 0;
3709
3710 free:
3711         kfree(t);
3712 err:
3713         return -ENOBUFS;
3714 }
3715 EXPORT_SYMBOL(neigh_sysctl_register);
3716
3717 void neigh_sysctl_unregister(struct neigh_parms *p)
3718 {
3719         if (p->sysctl_table) {
3720                 struct neigh_sysctl_table *t = p->sysctl_table;
3721                 p->sysctl_table = NULL;
3722                 unregister_net_sysctl_table(t->sysctl_header);
3723                 kfree(t);
3724         }
3725 }
3726 EXPORT_SYMBOL(neigh_sysctl_unregister);
3727
3728 #endif  /* CONFIG_SYSCTL */
3729
3730 static int __init neigh_init(void)
3731 {
3732         rtnl_register(PF_UNSPEC, RTM_NEWNEIGH, neigh_add, NULL, 0);
3733         rtnl_register(PF_UNSPEC, RTM_DELNEIGH, neigh_delete, NULL, 0);
3734         rtnl_register(PF_UNSPEC, RTM_GETNEIGH, neigh_get, neigh_dump_info, 0);
3735
3736         rtnl_register(PF_UNSPEC, RTM_GETNEIGHTBL, NULL, neightbl_dump_info,
3737                       0);
3738         rtnl_register(PF_UNSPEC, RTM_SETNEIGHTBL, neightbl_set, NULL, 0);
3739
3740         return 0;
3741 }
3742
3743 subsys_initcall(neigh_init);