net/core/neighbour.c

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