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