Pileus Git - ~andy/linux/blob - 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
  42 #define NEIGH_DEBUG 1
  43
  44 #define NEIGH_PRINTK(x...) printk(x)
  45 #define NEIGH_NOPRINTK(x...) do { ; } while(0)
  46 #define NEIGH_PRINTK1 NEIGH_NOPRINTK
  47 #define NEIGH_PRINTK2 NEIGH_NOPRINTK
  48
  49 #if NEIGH_DEBUG >= 1
  50 #undef NEIGH_PRINTK1
  51 #define NEIGH_PRINTK1 NEIGH_PRINTK
  52 #endif
  53 #if NEIGH_DEBUG >= 2
  54 #undef NEIGH_PRINTK2
  55 #define NEIGH_PRINTK2 NEIGH_PRINTK
  56 #endif
  57
  58 #define PNEIGH_HASHMASK         0xF
  59
  60 static void neigh_timer_handler(unsigned long arg);
  61 static void __neigh_notify(struct neighbour *n, int type, int flags);
  62 static void neigh_update_notify(struct neighbour *neigh);
  63 static int pneigh_ifdown(struct neigh_table *tbl, struct net_device *dev);
  64
  65 static struct neigh_table *neigh_tables;
  66 #ifdef CONFIG_PROC_FS
  67 static const struct file_operations neigh_stat_seq_fops;
  68 #endif
  69
  70 /*
  71    Neighbour hash table buckets are protected with rwlock tbl->lock.
  72
  73    - All the scans/updates to hash buckets MUST be made under this lock.
  74    - NOTHING clever should be made under this lock: no callbacks
  75      to protocol backends, no attempts to send something to network.
  76      It will result in deadlocks, if backend/driver wants to use neighbour
  77      cache.
  78    - If the entry requires some non-trivial actions, increase
  79      its reference count and release table lock.
  80
  81    Neighbour entries are protected:
  82    - with reference count.
  83    - with rwlock neigh->lock
  84
  85    Reference count prevents destruction.
  86
  87    neigh->lock mainly serializes ll address data and its validity state.
  88    However, the same lock is used to protect another entry fields:
  89     - timer
  90     - resolution queue
  91
  92    Again, nothing clever shall be made under neigh->lock,
  93    the most complicated procedure, which we allow is dev->hard_header.
  94    It is supposed, that dev->hard_header is simplistic and does
  95    not make callbacks to neighbour tables.
  96
  97    The last lock is neigh_tbl_lock. It is pure SMP lock, protecting
  98    list of neighbour tables. This list is used only in process context,
  99  */
 100
 101 static DEFINE_RWLOCK(neigh_tbl_lock);
 102
 103 static int neigh_blackhole(struct neighbour *neigh, struct sk_buff *skb)
 104 {
 105         kfree_skb(skb);
 106         return -ENETDOWN;
 107 }
 108
 109 static void neigh_cleanup_and_release(struct neighbour *neigh)
 110 {
 111         if (neigh->parms->neigh_cleanup)
 112                 neigh->parms->neigh_cleanup(neigh);
 113
 114         __neigh_notify(neigh, RTM_DELNEIGH, 0);
 115         neigh_release(neigh);
 116 }
 117
 118 /*
 119  * It is random distribution in the interval (1/2)*base...(3/2)*base.
 120  * It corresponds to default IPv6 settings and is not overridable,
 121  * because it is really reasonable choice.
 122  */
 123
 124 unsigned long neigh_rand_reach_time(unsigned long base)
 125 {
 126         return base ? (net_random() % base) + (base >> 1) : 0;
 127 }
 128 EXPORT_SYMBOL(neigh_rand_reach_time);
 129
 130
 131 static int neigh_forced_gc(struct neigh_table *tbl)
 132 {
 133         int shrunk = 0;
 134         int i;
 135         struct neigh_hash_table *nht;
 136
 137         NEIGH_CACHE_STAT_INC(tbl, forced_gc_runs);
 138
 139         write_lock_bh(&tbl->lock);
 140         nht = rcu_dereference_protected(tbl->nht,
 141                                         lockdep_is_held(&tbl->lock));
 142         for (i = 0; i < (1 << nht->hash_shift); i++) {
 143                 struct neighbour *n;
 144                 struct neighbour __rcu **np;
 145
 146                 np = &nht->hash_buckets[i];
 147                 while ((n = rcu_dereference_protected(*np,
 148                                         lockdep_is_held(&tbl->lock))) != NULL) {
 149                         /* Neighbour record may be discarded if:
 150                          * - nobody refers to it.
 151                          * - it is not permanent
 152                          */
 153                         write_lock(&n->lock);
 154                         if (atomic_read(&n->refcnt) == 1 &&
 155                             !(n->nud_state & NUD_PERMANENT)) {
 156                                 rcu_assign_pointer(*np,
 157                                         rcu_dereference_protected(n->next,
 158                                                   lockdep_is_held(&tbl->lock)));
 159                                 n->dead = 1;
 160                                 shrunk  = 1;
 161                                 write_unlock(&n->lock);
 162                                 neigh_cleanup_and_release(n);
 163                                 continue;
 164                         }
 165                         write_unlock(&n->lock);
 166                         np = &n->next;
 167                 }
 168         }
 169
 170         tbl->last_flush = jiffies;
 171
 172         write_unlock_bh(&tbl->lock);
 173
 174         return shrunk;
 175 }
 176
 177 static void neigh_add_timer(struct neighbour *n, unsigned long when)
 178 {
 179         neigh_hold(n);
 180         if (unlikely(mod_timer(&n->timer, when))) {
 181                 printk("NEIGH: BUG, double timer add, state is %x\n",
 182                        n->nud_state);
 183                 dump_stack();
 184         }
 185 }
 186
 187 static int neigh_del_timer(struct neighbour *n)
 188 {
 189         if ((n->nud_state & NUD_IN_TIMER) &&
 190             del_timer(&n->timer)) {
 191                 neigh_release(n);
 192                 return 1;
 193         }
 194         return 0;
 195 }
 196
 197 static void pneigh_queue_purge(struct sk_buff_head *list)
 198 {
 199         struct sk_buff *skb;
 200
 201         while ((skb = skb_dequeue(list)) != NULL) {
 202                 dev_put(skb->dev);
 203                 kfree_skb(skb);
 204         }
 205 }
 206
 207 static void neigh_flush_dev(struct neigh_table *tbl, struct net_device *dev)
 208 {
 209         int i;
 210         struct neigh_hash_table *nht;
 211
 212         nht = rcu_dereference_protected(tbl->nht,
 213                                         lockdep_is_held(&tbl->lock));
 214
 215         for (i = 0; i < (1 << nht->hash_shift); i++) {
 216                 struct neighbour *n;
 217                 struct neighbour __rcu **np = &nht->hash_buckets[i];
 218
 219                 while ((n = rcu_dereference_protected(*np,
 220                                         lockdep_is_held(&tbl->lock))) != NULL) {
 221                         if (dev && n->dev != dev) {
 222                                 np = &n->next;
 223                                 continue;
 224                         }
 225                         rcu_assign_pointer(*np,
 226                                    rcu_dereference_protected(n->next,
 227                                                 lockdep_is_held(&tbl->lock)));
 228                         write_lock(&n->lock);
 229                         neigh_del_timer(n);
 230                         n->dead = 1;
 231
 232                         if (atomic_read(&n->refcnt) != 1) {
 233                                 /* The most unpleasant situation.
 234                                    We must destroy neighbour entry,
 235                                    but someone still uses it.
 236
 237                                    The destroy will be delayed until
 238                                    the last user releases us, but
 239                                    we must kill timers etc. and move
 240                                    it to safe state.
 241                                  */
 242                                 skb_queue_purge(&n->arp_queue);
 243                                 n->arp_queue_len_bytes = 0;
 244                                 n->output = neigh_blackhole;
 245                                 if (n->nud_state & NUD_VALID)
 246                                         n->nud_state = NUD_NOARP;
 247                                 else
 248                                         n->nud_state = NUD_NONE;
 249                                 NEIGH_PRINTK2("neigh %p is stray.\n", n);
 250                         }
 251                         write_unlock(&n->lock);
 252                         neigh_cleanup_and_release(n);
 253                 }
 254         }
 255 }
 256
 257 void neigh_changeaddr(struct neigh_table *tbl, struct net_device *dev)
 258 {
 259         write_lock_bh(&tbl->lock);
 260         neigh_flush_dev(tbl, dev);
 261         write_unlock_bh(&tbl->lock);
 262 }
 263 EXPORT_SYMBOL(neigh_changeaddr);
 264
 265 int neigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
 266 {
 267         write_lock_bh(&tbl->lock);
 268         neigh_flush_dev(tbl, dev);
 269         pneigh_ifdown(tbl, dev);
 270         write_unlock_bh(&tbl->lock);
 271
 272         del_timer_sync(&tbl->proxy_timer);
 273         pneigh_queue_purge(&tbl->proxy_queue);
 274         return 0;
 275 }
 276 EXPORT_SYMBOL(neigh_ifdown);
 277
 278 static struct neighbour *neigh_alloc(struct neigh_table *tbl, struct net_device *dev)
 279 {
 280         struct neighbour *n = NULL;
 281         unsigned long now = jiffies;
 282         int entries;
 283
 284         entries = atomic_inc_return(&tbl->entries) - 1;
 285         if (entries >= tbl->gc_thresh3 ||
 286             (entries >= tbl->gc_thresh2 &&
 287              time_after(now, tbl->last_flush + 5 * HZ))) {
 288                 if (!neigh_forced_gc(tbl) &&
 289                     entries >= tbl->gc_thresh3)
 290                         goto out_entries;
 291         }
 292
 293         if (tbl->entry_size)
 294                 n = kzalloc(tbl->entry_size, GFP_ATOMIC);
 295         else {
 296                 int sz = sizeof(*n) + tbl->key_len;
 297
 298                 sz = ALIGN(sz, NEIGH_PRIV_ALIGN);
 299                 sz += dev->neigh_priv_len;
 300                 n = kzalloc(sz, GFP_ATOMIC);
 301         }
 302         if (!n)
 303                 goto out_entries;
 304
 305         skb_queue_head_init(&n->arp_queue);
 306         rwlock_init(&n->lock);
 307         seqlock_init(&n->ha_lock);
 308         n->updated        = n->used = now;
 309         n->nud_state      = NUD_NONE;
 310         n->output         = neigh_blackhole;
 311         seqlock_init(&n->hh.hh_lock);
 312         n->parms          = neigh_parms_clone(&tbl->parms);
 313         setup_timer(&n->timer, neigh_timer_handler, (unsigned long)n);
 314
 315         NEIGH_CACHE_STAT_INC(tbl, allocs);
 316         n->tbl            = tbl;
 317         atomic_set(&n->refcnt, 1);
 318         n->dead           = 1;
 319 out:
 320         return n;
 321
 322 out_entries:
 323         atomic_dec(&tbl->entries);
 324         goto out;
 325 }
 326
 327 static void neigh_get_hash_rnd(u32 *x)
 328 {
 329         get_random_bytes(x, sizeof(*x));
 330         *x |= 1;
 331 }
 332
 333 static struct neigh_hash_table *neigh_hash_alloc(unsigned int shift)
 334 {
 335         size_t size = (1 << shift) * sizeof(struct neighbour *);
 336         struct neigh_hash_table *ret;
 337         struct neighbour __rcu **buckets;
 338         int i;
 339
 340         ret = kmalloc(sizeof(*ret), GFP_ATOMIC);
 341         if (!ret)
 342                 return NULL;
 343         if (size <= PAGE_SIZE)
 344                 buckets = kzalloc(size, GFP_ATOMIC);
 345         else
 346                 buckets = (struct neighbour __rcu **)
 347                           __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
 348                                            get_order(size));
 349         if (!buckets) {
 350                 kfree(ret);
 351                 return NULL;
 352         }
 353         ret->hash_buckets = buckets;
 354         ret->hash_shift = shift;
 355         for (i = 0; i < NEIGH_NUM_HASH_RND; i++)
 356                 neigh_get_hash_rnd(&ret->hash_rnd[i]);
 357         return ret;
 358 }
 359
 360 static void neigh_hash_free_rcu(struct rcu_head *head)
 361 {
 362         struct neigh_hash_table *nht = container_of(head,
 363                                                     struct neigh_hash_table,
 364                                                     rcu);
 365         size_t size = (1 << nht->hash_shift) * sizeof(struct neighbour *);
 366         struct neighbour __rcu **buckets = nht->hash_buckets;
 367
 368         if (size <= PAGE_SIZE)
 369                 kfree(buckets);
 370         else
 371                 free_pages((unsigned long)buckets, get_order(size));
 372         kfree(nht);
 373 }
 374
 375 static struct neigh_hash_table *neigh_hash_grow(struct neigh_table *tbl,
 376                                                 unsigned long new_shift)
 377 {
 378         unsigned int i, hash;
 379         struct neigh_hash_table *new_nht, *old_nht;
 380
 381         NEIGH_CACHE_STAT_INC(tbl, hash_grows);
 382
 383         old_nht = rcu_dereference_protected(tbl->nht,
 384                                             lockdep_is_held(&tbl->lock));
 385         new_nht = neigh_hash_alloc(new_shift);
 386         if (!new_nht)
 387                 return old_nht;
 388
 389         for (i = 0; i < (1 << old_nht->hash_shift); i++) {
 390                 struct neighbour *n, *next;
 391
 392                 for (n = rcu_dereference_protected(old_nht->hash_buckets[i],
 393                                                    lockdep_is_held(&tbl->lock));
 394                      n != NULL;
 395                      n = next) {
 396                         hash = tbl->hash(n->primary_key, n->dev,
 397                                          new_nht->hash_rnd);
 398
 399                         hash >>= (32 - new_nht->hash_shift);
 400                         next = rcu_dereference_protected(n->next,
 401                                                 lockdep_is_held(&tbl->lock));
 402
 403                         rcu_assign_pointer(n->next,
 404                                            rcu_dereference_protected(
 405                                                 new_nht->hash_buckets[hash],
 406                                                 lockdep_is_held(&tbl->lock)));
 407                         rcu_assign_pointer(new_nht->hash_buckets[hash], n);
 408                 }
 409         }
 410
 411         rcu_assign_pointer(tbl->nht, new_nht);
 412         call_rcu(&old_nht->rcu, neigh_hash_free_rcu);
 413         return new_nht;
 414 }
 415
 416 struct neighbour *neigh_lookup(struct neigh_table *tbl, const void *pkey,
 417                                struct net_device *dev)
 418 {
 419         struct neighbour *n;
 420         int key_len = tbl->key_len;
 421         u32 hash_val;
 422         struct neigh_hash_table *nht;
 423
 424         NEIGH_CACHE_STAT_INC(tbl, lookups);
 425
 426         rcu_read_lock_bh();
 427         nht = rcu_dereference_bh(tbl->nht);
 428         hash_val = tbl->hash(pkey, dev, nht->hash_rnd) >> (32 - nht->hash_shift);
 429
 430         for (n = rcu_dereference_bh(nht->hash_buckets[hash_val]);
 431              n != NULL;
 432              n = rcu_dereference_bh(n->next)) {
 433                 if (dev == n->dev && !memcmp(n->primary_key, pkey, key_len)) {
 434                         if (!atomic_inc_not_zero(&n->refcnt))
 435                                 n = NULL;
 436                         NEIGH_CACHE_STAT_INC(tbl, hits);
 437                         break;
 438                 }
 439         }
 440
 441         rcu_read_unlock_bh();
 442         return n;
 443 }
 444 EXPORT_SYMBOL(neigh_lookup);
 445
 446 struct neighbour *neigh_lookup_nodev(struct neigh_table *tbl, struct net *net,
 447                                      const void *pkey)
 448 {
 449         struct neighbour *n;
 450         int key_len = tbl->key_len;
 451         u32 hash_val;
 452         struct neigh_hash_table *nht;
 453
 454         NEIGH_CACHE_STAT_INC(tbl, lookups);
 455
 456         rcu_read_lock_bh();
 457         nht = rcu_dereference_bh(tbl->nht);
 458         hash_val = tbl->hash(pkey, NULL, nht->hash_rnd) >> (32 - nht->hash_shift);
 459
 460         for (n = rcu_dereference_bh(nht->hash_buckets[hash_val]);
 461              n != NULL;
 462              n = rcu_dereference_bh(n->next)) {
 463                 if (!memcmp(n->primary_key, pkey, key_len) &&
 464                     net_eq(dev_net(n->dev), net)) {
 465                         if (!atomic_inc_not_zero(&n->refcnt))
 466                                 n = NULL;
 467                         NEIGH_CACHE_STAT_INC(tbl, hits);
 468                         break;
 469                 }
 470         }
 471
 472         rcu_read_unlock_bh();
 473         return n;
 474 }
 475 EXPORT_SYMBOL(neigh_lookup_nodev);
 476
 477 struct neighbour *neigh_create(struct neigh_table *tbl, const void *pkey,
 478                                struct net_device *dev)
 479 {
 480         u32 hash_val;
 481         int key_len = tbl->key_len;
 482         int error;
 483         struct neighbour *n1, *rc, *n = neigh_alloc(tbl, dev);
 484         struct neigh_hash_table *nht;
 485
 486         if (!n) {
 487                 rc = ERR_PTR(-ENOBUFS);
 488                 goto out;
 489         }
 490
 491         memcpy(n->primary_key, pkey, key_len);
 492         n->dev = dev;
 493         dev_hold(dev);
 494
 495         /* Protocol specific setup. */
 496         if (tbl->constructor && (error = tbl->constructor(n)) < 0) {
 497                 rc = ERR_PTR(error);
 498                 goto out_neigh_release;
 499         }
 500
 501         if (dev->netdev_ops->ndo_neigh_construct) {
 502                 error = dev->netdev_ops->ndo_neigh_construct(n);
 503                 if (error < 0) {
 504                         rc = ERR_PTR(error);
 505                         goto out_neigh_release;
 506                 }
 507         }
 508
 509         /* Device specific setup. */
 510         if (n->parms->neigh_setup &&
 511             (error = n->parms->neigh_setup(n)) < 0) {
 512                 rc = ERR_PTR(error);
 513                 goto out_neigh_release;
 514         }
 515
 516         n->confirmed = jiffies - (n->parms->base_reachable_time << 1);
 517
 518         write_lock_bh(&tbl->lock);
 519         nht = rcu_dereference_protected(tbl->nht,
 520                                         lockdep_is_held(&tbl->lock));
 521
 522         if (atomic_read(&tbl->entries) > (1 << nht->hash_shift))
 523                 nht = neigh_hash_grow(tbl, nht->hash_shift + 1);
 524
 525         hash_val = tbl->hash(pkey, dev, nht->hash_rnd) >> (32 - nht->hash_shift);
 526
 527         if (n->parms->dead) {
 528                 rc = ERR_PTR(-EINVAL);
 529                 goto out_tbl_unlock;
 530         }
 531
 532         for (n1 = rcu_dereference_protected(nht->hash_buckets[hash_val],
 533                                             lockdep_is_held(&tbl->lock));
 534              n1 != NULL;
 535              n1 = rcu_dereference_protected(n1->next,
 536                         lockdep_is_held(&tbl->lock))) {
 537                 if (dev == n1->dev && !memcmp(n1->primary_key, pkey, key_len)) {
 538                         neigh_hold(n1);
 539                         rc = n1;
 540                         goto out_tbl_unlock;
 541                 }
 542         }
 543
 544         n->dead = 0;
 545         neigh_hold(n);
 546         rcu_assign_pointer(n->next,
 547                            rcu_dereference_protected(nht->hash_buckets[hash_val],
 548                                                      lockdep_is_held(&tbl->lock)));
 549         rcu_assign_pointer(nht->hash_buckets[hash_val], n);
 550         write_unlock_bh(&tbl->lock);
 551         NEIGH_PRINTK2("neigh %p is created.\n", n);
 552         rc = n;
 553 out:
 554         return rc;
 555 out_tbl_unlock:
 556         write_unlock_bh(&tbl->lock);
 557 out_neigh_release:
 558         neigh_release(n);
 559         goto out;
 560 }
 561 EXPORT_SYMBOL(neigh_create);
 562
 563 static u32 pneigh_hash(const void *pkey, int key_len)
 564 {
 565         u32 hash_val = *(u32 *)(pkey + key_len - 4);
 566         hash_val ^= (hash_val >> 16);
 567         hash_val ^= hash_val >> 8;
 568         hash_val ^= hash_val >> 4;
 569         hash_val &= PNEIGH_HASHMASK;
 570         return hash_val;
 571 }
 572
 573 static struct pneigh_entry *__pneigh_lookup_1(struct pneigh_entry *n,
 574                                               struct net *net,
 575                                               const void *pkey,
 576                                               int key_len,
 577                                               struct net_device *dev)
 578 {
 579         while (n) {
 580                 if (!memcmp(n->key, pkey, key_len) &&
 581                     net_eq(pneigh_net(n), net) &&
 582                     (n->dev == dev || !n->dev))
 583                         return n;
 584                 n = n->next;
 585         }
 586         return NULL;
 587 }
 588
 589 struct pneigh_entry *__pneigh_lookup(struct neigh_table *tbl,
 590                 struct net *net, const void *pkey, struct net_device *dev)
 591 {
 592         int key_len = tbl->key_len;
 593         u32 hash_val = pneigh_hash(pkey, key_len);
 594
 595         return __pneigh_lookup_1(tbl->phash_buckets[hash_val],
 596                                  net, pkey, key_len, dev);
 597 }
 598 EXPORT_SYMBOL_GPL(__pneigh_lookup);
 599
 600 struct pneigh_entry * pneigh_lookup(struct neigh_table *tbl,
 601                                     struct net *net, const void *pkey,
 602                                     struct net_device *dev, int creat)
 603 {
 604         struct pneigh_entry *n;
 605         int key_len = tbl->key_len;
 606         u32 hash_val = pneigh_hash(pkey, key_len);
 607
 608         read_lock_bh(&tbl->lock);
 609         n = __pneigh_lookup_1(tbl->phash_buckets[hash_val],
 610                               net, pkey, key_len, dev);
 611         read_unlock_bh(&tbl->lock);
 612
 613         if (n || !creat)
 614                 goto out;
 615
 616         ASSERT_RTNL();
 617
 618         n = kmalloc(sizeof(*n) + key_len, GFP_KERNEL);
 619         if (!n)
 620                 goto out;
 621
 622         write_pnet(&n->net, hold_net(net));
 623         memcpy(n->key, pkey, key_len);
 624         n->dev = dev;
 625         if (dev)
 626                 dev_hold(dev);
 627
 628         if (tbl->pconstructor && tbl->pconstructor(n)) {
 629                 if (dev)
 630                         dev_put(dev);
 631                 release_net(net);
 632                 kfree(n);
 633                 n = NULL;
 634                 goto out;
 635         }
 636
 637         write_lock_bh(&tbl->lock);
 638         n->next = tbl->phash_buckets[hash_val];
 639         tbl->phash_buckets[hash_val] = n;
 640         write_unlock_bh(&tbl->lock);
 641 out:
 642         return n;
 643 }
 644 EXPORT_SYMBOL(pneigh_lookup);
 645
 646
 647 int pneigh_delete(struct neigh_table *tbl, struct net *net, const void *pkey,
 648                   struct net_device *dev)
 649 {
 650         struct pneigh_entry *n, **np;
 651         int key_len = tbl->key_len;
 652         u32 hash_val = pneigh_hash(pkey, key_len);
 653
 654         write_lock_bh(&tbl->lock);
 655         for (np = &tbl->phash_buckets[hash_val]; (n = *np) != NULL;
 656              np = &n->next) {
 657                 if (!memcmp(n->key, pkey, key_len) && n->dev == dev &&
 658                     net_eq(pneigh_net(n), net)) {
 659                         *np = n->next;
 660                         write_unlock_bh(&tbl->lock);
 661                         if (tbl->pdestructor)
 662                                 tbl->pdestructor(n);
 663                         if (n->dev)
 664                                 dev_put(n->dev);
 665                         release_net(pneigh_net(n));
 666                         kfree(n);
 667                         return 0;
 668                 }
 669         }
 670         write_unlock_bh(&tbl->lock);
 671         return -ENOENT;
 672 }
 673
 674 static int pneigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
 675 {
 676         struct pneigh_entry *n, **np;
 677         u32 h;
 678
 679         for (h = 0; h <= PNEIGH_HASHMASK; h++) {
 680                 np = &tbl->phash_buckets[h];
 681                 while ((n = *np) != NULL) {
 682                         if (!dev || n->dev == dev) {
 683                                 *np = n->next;
 684                                 if (tbl->pdestructor)
 685                                         tbl->pdestructor(n);
 686                                 if (n->dev)
 687                                         dev_put(n->dev);
 688                                 release_net(pneigh_net(n));
 689                                 kfree(n);
 690                                 continue;
 691                         }
 692                         np = &n->next;
 693                 }
 694         }
 695         return -ENOENT;
 696 }
 697
 698 static void neigh_parms_destroy(struct neigh_parms *parms);
 699
 700 static inline void neigh_parms_put(struct neigh_parms *parms)
 701 {
 702         if (atomic_dec_and_test(&parms->refcnt))
 703                 neigh_parms_destroy(parms);
 704 }
 705
 706 /*
 707  *      neighbour must already be out of the table;
 708  *
 709  */
 710 void neigh_destroy(struct neighbour *neigh)
 711 {
 712         struct net_device *dev = neigh->dev;
 713
 714         NEIGH_CACHE_STAT_INC(neigh->tbl, destroys);
 715
 716         if (!neigh->dead) {
 717                 pr_warn("Destroying alive neighbour %p\n", neigh);
 718                 dump_stack();
 719                 return;
 720         }
 721
 722         if (neigh_del_timer(neigh))
 723                 pr_warn("Impossible event\n");
 724
 725         skb_queue_purge(&neigh->arp_queue);
 726         neigh->arp_queue_len_bytes = 0;
 727
 728         if (dev->netdev_ops->ndo_neigh_destroy)
 729                 dev->netdev_ops->ndo_neigh_destroy(neigh);
 730
 731         dev_put(dev);
 732         neigh_parms_put(neigh->parms);
 733
 734         NEIGH_PRINTK2("neigh %p is destroyed.\n", neigh);
 735
 736         atomic_dec(&neigh->tbl->entries);
 737         kfree_rcu(neigh, rcu);
 738 }
 739 EXPORT_SYMBOL(neigh_destroy);
 740
 741 /* Neighbour state is suspicious;
 742    disable fast path.
 743
 744    Called with write_locked neigh.
 745  */
 746 static void neigh_suspect(struct neighbour *neigh)
 747 {
 748         NEIGH_PRINTK2("neigh %p is suspected.\n", neigh);
 749
 750         neigh->output = neigh->ops->output;
 751 }
 752
 753 /* Neighbour state is OK;
 754    enable fast path.
 755
 756    Called with write_locked neigh.
 757  */
 758 static void neigh_connect(struct neighbour *neigh)
 759 {
 760         NEIGH_PRINTK2("neigh %p is connected.\n", neigh);
 761
 762         neigh->output = neigh->ops->connected_output;
 763 }
 764
 765 static void neigh_periodic_work(struct work_struct *work)
 766 {
 767         struct neigh_table *tbl = container_of(work, struct neigh_table, gc_work.work);
 768         struct neighbour *n;
 769         struct neighbour __rcu **np;
 770         unsigned int i;
 771         struct neigh_hash_table *nht;
 772
 773         NEIGH_CACHE_STAT_INC(tbl, periodic_gc_runs);
 774
 775         write_lock_bh(&tbl->lock);
 776         nht = rcu_dereference_protected(tbl->nht,
 777                                         lockdep_is_held(&tbl->lock));
 778
 779         /*
 780          *      periodically recompute ReachableTime from random function
 781          */
 782
 783         if (time_after(jiffies, tbl->last_rand + 300 * HZ)) {
 784                 struct neigh_parms *p;
 785                 tbl->last_rand = jiffies;
 786                 for (p = &tbl->parms; p; p = p->next)
 787                         p->reachable_time =
 788                                 neigh_rand_reach_time(p->base_reachable_time);
 789         }
 790
 791         for (i = 0 ; i < (1 << nht->hash_shift); i++) {
 792                 np = &nht->hash_buckets[i];
 793
 794                 while ((n = rcu_dereference_protected(*np,
 795                                 lockdep_is_held(&tbl->lock))) != NULL) {
 796                         unsigned int state;
 797
 798                         write_lock(&n->lock);
 799
 800                         state = n->nud_state;
 801                         if (state & (NUD_PERMANENT | NUD_IN_TIMER)) {
 802                                 write_unlock(&n->lock);
 803                                 goto next_elt;
 804                         }
 805
 806                         if (time_before(n->used, n->confirmed))
 807                                 n->used = n->confirmed;
 808
 809                         if (atomic_read(&n->refcnt) == 1 &&
 810                             (state == NUD_FAILED ||
 811                              time_after(jiffies, n->used + n->parms->gc_staletime))) {
 812                                 *np = n->next;
 813                                 n->dead = 1;
 814                                 write_unlock(&n->lock);
 815                                 neigh_cleanup_and_release(n);
 816                                 continue;
 817                         }
 818                         write_unlock(&n->lock);
 819
 820 next_elt:
 821                         np = &n->next;
 822                 }
 823                 /*
 824                  * It's fine to release lock here, even if hash table
 825                  * grows while we are preempted.
 826                  */
 827                 write_unlock_bh(&tbl->lock);
 828                 cond_resched();
 829                 write_lock_bh(&tbl->lock);
 830                 nht = rcu_dereference_protected(tbl->nht,
 831                                                 lockdep_is_held(&tbl->lock));
 832         }
 833         /* Cycle through all hash buckets every base_reachable_time/2 ticks.
 834          * ARP entry timeouts range from 1/2 base_reachable_time to 3/2
 835          * base_reachable_time.
 836          */
 837         schedule_delayed_work(&tbl->gc_work,
 838                               tbl->parms.base_reachable_time >> 1);
 839         write_unlock_bh(&tbl->lock);
 840 }
 841
 842 static __inline__ int neigh_max_probes(struct neighbour *n)
 843 {
 844         struct neigh_parms *p = n->parms;
 845         return (n->nud_state & NUD_PROBE) ?
 846                 p->ucast_probes :
 847                 p->ucast_probes + p->app_probes + p->mcast_probes;
 848 }
 849
 850 static void neigh_invalidate(struct neighbour *neigh)
 851         __releases(neigh->lock)
 852         __acquires(neigh->lock)
 853 {
 854         struct sk_buff *skb;
 855
 856         NEIGH_CACHE_STAT_INC(neigh->tbl, res_failed);
 857         NEIGH_PRINTK2("neigh %p is failed.\n", neigh);
 858         neigh->updated = jiffies;
 859
 860         /* It is very thin place. report_unreachable is very complicated
 861            routine. Particularly, it can hit the same neighbour entry!
 862
 863            So that, we try to be accurate and avoid dead loop. --ANK
 864          */
 865         while (neigh->nud_state == NUD_FAILED &&
 866                (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
 867                 write_unlock(&neigh->lock);
 868                 neigh->ops->error_report(neigh, skb);
 869                 write_lock(&neigh->lock);
 870         }
 871         skb_queue_purge(&neigh->arp_queue);
 872         neigh->arp_queue_len_bytes = 0;
 873 }
 874
 875 static void neigh_probe(struct neighbour *neigh)
 876         __releases(neigh->lock)
 877 {
 878         struct sk_buff *skb = skb_peek(&neigh->arp_queue);
 879         /* keep skb alive even if arp_queue overflows */
 880         if (skb)
 881                 skb = skb_copy(skb, GFP_ATOMIC);
 882         write_unlock(&neigh->lock);
 883         neigh->ops->solicit(neigh, skb);
 884         atomic_inc(&neigh->probes);
 885         kfree_skb(skb);
 886 }
 887
 888 /* Called when a timer expires for a neighbour entry. */
 889
 890 static void neigh_timer_handler(unsigned long arg)
 891 {
 892         unsigned long now, next;
 893         struct neighbour *neigh = (struct neighbour *)arg;
 894         unsigned int state;
 895         int notify = 0;
 896
 897         write_lock(&neigh->lock);
 898
 899         state = neigh->nud_state;
 900         now = jiffies;
 901         next = now + HZ;
 902
 903         if (!(state & NUD_IN_TIMER))
 904                 goto out;
 905
 906         if (state & NUD_REACHABLE) {
 907                 if (time_before_eq(now,
 908                                    neigh->confirmed + neigh->parms->reachable_time)) {
 909                         NEIGH_PRINTK2("neigh %p is still alive.\n", neigh);
 910                         next = neigh->confirmed + neigh->parms->reachable_time;
 911                 } else if (time_before_eq(now,
 912                                           neigh->used + neigh->parms->delay_probe_time)) {
 913                         NEIGH_PRINTK2("neigh %p is delayed.\n", neigh);
 914                         neigh->nud_state = NUD_DELAY;
 915                         neigh->updated = jiffies;
 916                         neigh_suspect(neigh);
 917                         next = now + neigh->parms->delay_probe_time;
 918                 } else {
 919                         NEIGH_PRINTK2("neigh %p is suspected.\n", neigh);
 920                         neigh->nud_state = NUD_STALE;
 921                         neigh->updated = jiffies;
 922                         neigh_suspect(neigh);
 923                         notify = 1;
 924                 }
 925         } else if (state & NUD_DELAY) {
 926                 if (time_before_eq(now,
 927                                    neigh->confirmed + neigh->parms->delay_probe_time)) {
 928                         NEIGH_PRINTK2("neigh %p is now reachable.\n", neigh);
 929                         neigh->nud_state = NUD_REACHABLE;
 930                         neigh->updated = jiffies;
 931                         neigh_connect(neigh);
 932                         notify = 1;
 933                         next = neigh->confirmed + neigh->parms->reachable_time;
 934                 } else {
 935                         NEIGH_PRINTK2("neigh %p is probed.\n", neigh);
 936                         neigh->nud_state = NUD_PROBE;
 937                         neigh->updated = jiffies;
 938                         atomic_set(&neigh->probes, 0);
 939                         next = now + neigh->parms->retrans_time;
 940                 }
 941         } else {
 942                 /* NUD_PROBE|NUD_INCOMPLETE */
 943                 next = now + neigh->parms->retrans_time;
 944         }
 945
 946         if ((neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) &&
 947             atomic_read(&neigh->probes) >= neigh_max_probes(neigh)) {
 948                 neigh->nud_state = NUD_FAILED;
 949                 notify = 1;
 950                 neigh_invalidate(neigh);
 951         }
 952
 953         if (neigh->nud_state & NUD_IN_TIMER) {
 954                 if (time_before(next, jiffies + HZ/2))
 955                         next = jiffies + HZ/2;
 956                 if (!mod_timer(&neigh->timer, next))
 957                         neigh_hold(neigh);
 958         }
 959         if (neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) {
 960                 neigh_probe(neigh);
 961         } else {
 962 out:
 963                 write_unlock(&neigh->lock);
 964         }
 965
 966         if (notify)
 967                 neigh_update_notify(neigh);
 968
 969         neigh_release(neigh);
 970 }
 971
 972 int __neigh_event_send(struct neighbour *neigh, struct sk_buff *skb)
 973 {
 974         int rc;
 975         bool immediate_probe = false;
 976
 977         write_lock_bh(&neigh->lock);
 978
 979         rc = 0;
 980         if (neigh->nud_state & (NUD_CONNECTED | NUD_DELAY | NUD_PROBE))
 981                 goto out_unlock_bh;
 982
 983         if (!(neigh->nud_state & (NUD_STALE | NUD_INCOMPLETE))) {
 984                 if (neigh->parms->mcast_probes + neigh->parms->app_probes) {
 985                         unsigned long next, now = jiffies;
 986
 987                         atomic_set(&neigh->probes, neigh->parms->ucast_probes);
 988                         neigh->nud_state     = NUD_INCOMPLETE;
 989                         neigh->updated = now;
 990                         next = now + max(neigh->parms->retrans_time, 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_PRINTK2("neigh %p is delayed.\n", neigh);
1003                 neigh->nud_state = NUD_DELAY;
1004                 neigh->updated = jiffies;
1005                 neigh_add_timer(neigh,
1006                                 jiffies + 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->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         }
1173
1174         if (lladdr != neigh->ha) {
1175                 write_seqlock(&neigh->ha_lock);
1176                 memcpy(&neigh->ha, lladdr, dev->addr_len);
1177                 write_sequnlock(&neigh->ha_lock);
1178                 neigh_update_hhs(neigh);
1179                 if (!(new & NUD_CONNECTED))
1180                         neigh->confirmed = jiffies -
1181                                       (neigh->parms->base_reachable_time << 1);
1182                 notify = 1;
1183         }
1184         if (new == old)
1185                 goto out;
1186         if (new & NUD_CONNECTED)
1187                 neigh_connect(neigh);
1188         else
1189                 neigh_suspect(neigh);
1190         if (!(old & NUD_VALID)) {
1191                 struct sk_buff *skb;
1192
1193                 /* Again: avoid dead loop if something went wrong */
1194
1195                 while (neigh->nud_state & NUD_VALID &&
1196                        (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
1197                         struct dst_entry *dst = skb_dst(skb);
1198                         struct neighbour *n2, *n1 = neigh;
1199                         write_unlock_bh(&neigh->lock);
1200
1201                         rcu_read_lock();
1202                         /* On shaper/eql skb->dst->neighbour != neigh :( */
1203                         if (dst && (n2 = dst_get_neighbour_noref(dst)) != NULL)
1204                                 n1 = n2;
1205                         n1->output(n1, skb);
1206                         rcu_read_unlock();
1207
1208                         write_lock_bh(&neigh->lock);
1209                 }
1210                 skb_queue_purge(&neigh->arp_queue);
1211                 neigh->arp_queue_len_bytes = 0;
1212         }
1213 out:
1214         if (update_isrouter) {
1215                 neigh->flags = (flags & NEIGH_UPDATE_F_ISROUTER) ?
1216                         (neigh->flags | NTF_ROUTER) :
1217                         (neigh->flags & ~NTF_ROUTER);
1218         }
1219         write_unlock_bh(&neigh->lock);
1220
1221         if (notify)
1222                 neigh_update_notify(neigh);
1223
1224         return err;
1225 }
1226 EXPORT_SYMBOL(neigh_update);
1227
1228 struct neighbour *neigh_event_ns(struct neigh_table *tbl,
1229                                  u8 *lladdr, void *saddr,
1230                                  struct net_device *dev)
1231 {
1232         struct neighbour *neigh = __neigh_lookup(tbl, saddr, dev,
1233                                                  lladdr || !dev->addr_len);
1234         if (neigh)
1235                 neigh_update(neigh, lladdr, NUD_STALE,
1236                              NEIGH_UPDATE_F_OVERRIDE);
1237         return neigh;
1238 }
1239 EXPORT_SYMBOL(neigh_event_ns);
1240
1241 /* called with read_lock_bh(&n->lock); */
1242 static void neigh_hh_init(struct neighbour *n, struct dst_entry *dst)
1243 {
1244         struct net_device *dev = dst->dev;
1245         __be16 prot = dst->ops->protocol;
1246         struct hh_cache *hh = &n->hh;
1247
1248         write_lock_bh(&n->lock);
1249
1250         /* Only one thread can come in here and initialize the
1251          * hh_cache entry.
1252          */
1253         if (!hh->hh_len)
1254                 dev->header_ops->cache(n, hh, prot);
1255
1256         write_unlock_bh(&n->lock);
1257 }
1258
1259 /* This function can be used in contexts, where only old dev_queue_xmit
1260  * worked, f.e. if you want to override normal output path (eql, shaper),
1261  * but resolution is not made yet.
1262  */
1263
1264 int neigh_compat_output(struct neighbour *neigh, struct sk_buff *skb)
1265 {
1266         struct net_device *dev = skb->dev;
1267
1268         __skb_pull(skb, skb_network_offset(skb));
1269
1270         if (dev_hard_header(skb, dev, ntohs(skb->protocol), NULL, NULL,
1271                             skb->len) < 0 &&
1272             dev->header_ops->rebuild(skb))
1273                 return 0;
1274
1275         return dev_queue_xmit(skb);
1276 }
1277 EXPORT_SYMBOL(neigh_compat_output);
1278
1279 /* Slow and careful. */
1280
1281 int neigh_resolve_output(struct neighbour *neigh, struct sk_buff *skb)
1282 {
1283         struct dst_entry *dst = skb_dst(skb);
1284         int rc = 0;
1285
1286         if (!dst)
1287                 goto discard;
1288
1289         __skb_pull(skb, skb_network_offset(skb));
1290
1291         if (!neigh_event_send(neigh, skb)) {
1292                 int err;
1293                 struct net_device *dev = neigh->dev;
1294                 unsigned int seq;
1295
1296                 if (dev->header_ops->cache && !neigh->hh.hh_len)
1297                         neigh_hh_init(neigh, dst);
1298
1299                 do {
1300                         seq = read_seqbegin(&neigh->ha_lock);
1301                         err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1302                                               neigh->ha, NULL, skb->len);
1303                 } while (read_seqretry(&neigh->ha_lock, seq));
1304
1305                 if (err >= 0)
1306                         rc = dev_queue_xmit(skb);
1307                 else
1308                         goto out_kfree_skb;
1309         }
1310 out:
1311         return rc;
1312 discard:
1313         NEIGH_PRINTK1("neigh_resolve_output: dst=%p neigh=%p\n",
1314                       dst, neigh);
1315 out_kfree_skb:
1316         rc = -EINVAL;
1317         kfree_skb(skb);
1318         goto out;
1319 }
1320 EXPORT_SYMBOL(neigh_resolve_output);
1321
1322 /* As fast as possible without hh cache */
1323
1324 int neigh_connected_output(struct neighbour *neigh, struct sk_buff *skb)
1325 {
1326         struct net_device *dev = neigh->dev;
1327         unsigned int seq;
1328         int err;
1329
1330         __skb_pull(skb, skb_network_offset(skb));
1331
1332         do {
1333                 seq = read_seqbegin(&neigh->ha_lock);
1334                 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1335                                       neigh->ha, NULL, skb->len);
1336         } while (read_seqretry(&neigh->ha_lock, seq));
1337
1338         if (err >= 0)
1339                 err = dev_queue_xmit(skb);
1340         else {
1341                 err = -EINVAL;
1342                 kfree_skb(skb);
1343         }
1344         return err;
1345 }
1346 EXPORT_SYMBOL(neigh_connected_output);
1347
1348 int neigh_direct_output(struct neighbour *neigh, struct sk_buff *skb)
1349 {
1350         return dev_queue_xmit(skb);
1351 }
1352 EXPORT_SYMBOL(neigh_direct_output);
1353
1354 static void neigh_proxy_process(unsigned long arg)
1355 {
1356         struct neigh_table *tbl = (struct neigh_table *)arg;
1357         long sched_next = 0;
1358         unsigned long now = jiffies;
1359         struct sk_buff *skb, *n;
1360
1361         spin_lock(&tbl->proxy_queue.lock);
1362
1363         skb_queue_walk_safe(&tbl->proxy_queue, skb, n) {
1364                 long tdif = NEIGH_CB(skb)->sched_next - now;
1365
1366                 if (tdif <= 0) {
1367                         struct net_device *dev = skb->dev;
1368
1369                         __skb_unlink(skb, &tbl->proxy_queue);
1370                         if (tbl->proxy_redo && netif_running(dev)) {
1371                                 rcu_read_lock();
1372                                 tbl->proxy_redo(skb);
1373                                 rcu_read_unlock();
1374                         } else {
1375                                 kfree_skb(skb);
1376                         }
1377
1378                         dev_put(dev);
1379                 } else if (!sched_next || tdif < sched_next)
1380                         sched_next = tdif;
1381         }
1382         del_timer(&tbl->proxy_timer);
1383         if (sched_next)
1384                 mod_timer(&tbl->proxy_timer, jiffies + sched_next);
1385         spin_unlock(&tbl->proxy_queue.lock);
1386 }
1387
1388 void pneigh_enqueue(struct neigh_table *tbl, struct neigh_parms *p,
1389                     struct sk_buff *skb)
1390 {
1391         unsigned long now = jiffies;
1392         unsigned long sched_next = now + (net_random() % p->proxy_delay);
1393
1394         if (tbl->proxy_queue.qlen > p->proxy_qlen) {
1395                 kfree_skb(skb);
1396                 return;
1397         }
1398
1399         NEIGH_CB(skb)->sched_next = sched_next;
1400         NEIGH_CB(skb)->flags |= LOCALLY_ENQUEUED;
1401
1402         spin_lock(&tbl->proxy_queue.lock);
1403         if (del_timer(&tbl->proxy_timer)) {
1404                 if (time_before(tbl->proxy_timer.expires, sched_next))
1405                         sched_next = tbl->proxy_timer.expires;
1406         }
1407         skb_dst_drop(skb);
1408         dev_hold(skb->dev);
1409         __skb_queue_tail(&tbl->proxy_queue, skb);
1410         mod_timer(&tbl->proxy_timer, sched_next);
1411         spin_unlock(&tbl->proxy_queue.lock);
1412 }
1413 EXPORT_SYMBOL(pneigh_enqueue);
1414
1415 static inline struct neigh_parms *lookup_neigh_parms(struct neigh_table *tbl,
1416                                                       struct net *net, int ifindex)
1417 {
1418         struct neigh_parms *p;
1419
1420         for (p = &tbl->parms; p; p = p->next) {
1421                 if ((p->dev && p->dev->ifindex == ifindex && net_eq(neigh_parms_net(p), net)) ||
1422                     (!p->dev && !ifindex))
1423                         return p;
1424         }
1425
1426         return NULL;
1427 }
1428
1429 struct neigh_parms *neigh_parms_alloc(struct net_device *dev,
1430                                       struct neigh_table *tbl)
1431 {
1432         struct neigh_parms *p, *ref;
1433         struct net *net = dev_net(dev);
1434         const struct net_device_ops *ops = dev->netdev_ops;
1435
1436         ref = lookup_neigh_parms(tbl, net, 0);
1437         if (!ref)
1438                 return NULL;
1439
1440         p = kmemdup(ref, sizeof(*p), GFP_KERNEL);
1441         if (p) {
1442                 p->tbl            = tbl;
1443                 atomic_set(&p->refcnt, 1);
1444                 p->reachable_time =
1445                                 neigh_rand_reach_time(p->base_reachable_time);
1446
1447                 if (ops->ndo_neigh_setup && ops->ndo_neigh_setup(dev, p)) {
1448                         kfree(p);
1449                         return NULL;
1450                 }
1451
1452                 dev_hold(dev);
1453                 p->dev = dev;
1454                 write_pnet(&p->net, hold_net(net));
1455                 p->sysctl_table = NULL;
1456                 write_lock_bh(&tbl->lock);
1457                 p->next         = tbl->parms.next;
1458                 tbl->parms.next = p;
1459                 write_unlock_bh(&tbl->lock);
1460         }
1461         return p;
1462 }
1463 EXPORT_SYMBOL(neigh_parms_alloc);
1464
1465 static void neigh_rcu_free_parms(struct rcu_head *head)
1466 {
1467         struct neigh_parms *parms =
1468                 container_of(head, struct neigh_parms, rcu_head);
1469
1470         neigh_parms_put(parms);
1471 }
1472
1473 void neigh_parms_release(struct neigh_table *tbl, struct neigh_parms *parms)
1474 {
1475         struct neigh_parms **p;
1476
1477         if (!parms || parms == &tbl->parms)
1478                 return;
1479         write_lock_bh(&tbl->lock);
1480         for (p = &tbl->parms.next; *p; p = &(*p)->next) {
1481                 if (*p == parms) {
1482                         *p = parms->next;
1483                         parms->dead = 1;
1484                         write_unlock_bh(&tbl->lock);
1485                         if (parms->dev)
1486                                 dev_put(parms->dev);
1487                         call_rcu(&parms->rcu_head, neigh_rcu_free_parms);
1488                         return;
1489                 }
1490         }
1491         write_unlock_bh(&tbl->lock);
1492         NEIGH_PRINTK1("neigh_parms_release: not found\n");
1493 }
1494 EXPORT_SYMBOL(neigh_parms_release);
1495
1496 static void neigh_parms_destroy(struct neigh_parms *parms)
1497 {
1498         release_net(neigh_parms_net(parms));
1499         kfree(parms);
1500 }
1501
1502 static struct lock_class_key neigh_table_proxy_queue_class;
1503
1504 static void neigh_table_init_no_netlink(struct neigh_table *tbl)
1505 {
1506         unsigned long now = jiffies;
1507         unsigned long phsize;
1508
1509         write_pnet(&tbl->parms.net, &init_net);
1510         atomic_set(&tbl->parms.refcnt, 1);
1511         tbl->parms.reachable_time =
1512                           neigh_rand_reach_time(tbl->parms.base_reachable_time);
1513
1514         tbl->stats = alloc_percpu(struct neigh_statistics);
1515         if (!tbl->stats)
1516                 panic("cannot create neighbour cache statistics");
1517
1518 #ifdef CONFIG_PROC_FS
1519         if (!proc_create_data(tbl->id, 0, init_net.proc_net_stat,
1520                               &neigh_stat_seq_fops, tbl))
1521                 panic("cannot create neighbour proc dir entry");
1522 #endif
1523
1524         RCU_INIT_POINTER(tbl->nht, neigh_hash_alloc(3));
1525
1526         phsize = (PNEIGH_HASHMASK + 1) * sizeof(struct pneigh_entry *);
1527         tbl->phash_buckets = kzalloc(phsize, GFP_KERNEL);
1528
1529         if (!tbl->nht || !tbl->phash_buckets)
1530                 panic("cannot allocate neighbour cache hashes");
1531
1532         rwlock_init(&tbl->lock);
1533         INIT_DELAYED_WORK_DEFERRABLE(&tbl->gc_work, neigh_periodic_work);
1534         schedule_delayed_work(&tbl->gc_work, tbl->parms.reachable_time);
1535         setup_timer(&tbl->proxy_timer, neigh_proxy_process, (unsigned long)tbl);
1536         skb_queue_head_init_class(&tbl->proxy_queue,
1537                         &neigh_table_proxy_queue_class);
1538
1539         tbl->last_flush = now;
1540         tbl->last_rand  = now + tbl->parms.reachable_time * 20;
1541 }
1542
1543 void neigh_table_init(struct neigh_table *tbl)
1544 {
1545         struct neigh_table *tmp;
1546
1547         neigh_table_init_no_netlink(tbl);
1548         write_lock(&neigh_tbl_lock);
1549         for (tmp = neigh_tables; tmp; tmp = tmp->next) {
1550                 if (tmp->family == tbl->family)
1551                         break;
1552         }
1553         tbl->next       = neigh_tables;
1554         neigh_tables    = tbl;
1555         write_unlock(&neigh_tbl_lock);
1556
1557         if (unlikely(tmp)) {
1558                 pr_err("Registering multiple tables for family %d\n",
1559                        tbl->family);
1560                 dump_stack();
1561         }
1562 }
1563 EXPORT_SYMBOL(neigh_table_init);
1564
1565 int neigh_table_clear(struct neigh_table *tbl)
1566 {
1567         struct neigh_table **tp;
1568
1569         /* It is not clean... Fix it to unload IPv6 module safely */
1570         cancel_delayed_work_sync(&tbl->gc_work);
1571         del_timer_sync(&tbl->proxy_timer);
1572         pneigh_queue_purge(&tbl->proxy_queue);
1573         neigh_ifdown(tbl, NULL);
1574         if (atomic_read(&tbl->entries))
1575                 pr_crit("neighbour leakage\n");
1576         write_lock(&neigh_tbl_lock);
1577         for (tp = &neigh_tables; *tp; tp = &(*tp)->next) {
1578                 if (*tp == tbl) {
1579                         *tp = tbl->next;
1580                         break;
1581                 }
1582         }
1583         write_unlock(&neigh_tbl_lock);
1584
1585         call_rcu(&rcu_dereference_protected(tbl->nht, 1)->rcu,
1586                  neigh_hash_free_rcu);
1587         tbl->nht = NULL;
1588
1589         kfree(tbl->phash_buckets);
1590         tbl->phash_buckets = NULL;
1591
1592         remove_proc_entry(tbl->id, init_net.proc_net_stat);
1593
1594         free_percpu(tbl->stats);
1595         tbl->stats = NULL;
1596
1597         return 0;
1598 }
1599 EXPORT_SYMBOL(neigh_table_clear);
1600
1601 static int neigh_delete(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
1602 {
1603         struct net *net = sock_net(skb->sk);
1604         struct ndmsg *ndm;
1605         struct nlattr *dst_attr;
1606         struct neigh_table *tbl;
1607         struct net_device *dev = NULL;
1608         int err = -EINVAL;
1609
1610         ASSERT_RTNL();
1611         if (nlmsg_len(nlh) < sizeof(*ndm))
1612                 goto out;
1613
1614         dst_attr = nlmsg_find_attr(nlh, sizeof(*ndm), NDA_DST);
1615         if (dst_attr == NULL)
1616                 goto out;
1617
1618         ndm = nlmsg_data(nlh);
1619         if (ndm->ndm_ifindex) {
1620                 dev = __dev_get_by_index(net, ndm->ndm_ifindex);
1621                 if (dev == NULL) {
1622                         err = -ENODEV;
1623                         goto out;
1624                 }
1625         }
1626
1627         read_lock(&neigh_tbl_lock);
1628         for (tbl = neigh_tables; tbl; tbl = tbl->next) {
1629                 struct neighbour *neigh;
1630
1631                 if (tbl->family != ndm->ndm_family)
1632                         continue;
1633                 read_unlock(&neigh_tbl_lock);
1634
1635                 if (nla_len(dst_attr) < tbl->key_len)
1636                         goto out;
1637
1638                 if (ndm->ndm_flags & NTF_PROXY) {
1639                         err = pneigh_delete(tbl, net, nla_data(dst_attr), dev);
1640                         goto out;
1641                 }
1642
1643                 if (dev == NULL)
1644                         goto out;
1645
1646                 neigh = neigh_lookup(tbl, nla_data(dst_attr), dev);
1647                 if (neigh == NULL) {
1648                         err = -ENOENT;
1649                         goto out;
1650                 }
1651
1652                 err = neigh_update(neigh, NULL, NUD_FAILED,
1653                                    NEIGH_UPDATE_F_OVERRIDE |
1654                                    NEIGH_UPDATE_F_ADMIN);
1655                 neigh_release(neigh);
1656                 goto out;
1657         }
1658         read_unlock(&neigh_tbl_lock);
1659         err = -EAFNOSUPPORT;
1660
1661 out:
1662         return err;
1663 }
1664
1665 static int neigh_add(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
1666 {
1667         struct net *net = sock_net(skb->sk);
1668         struct ndmsg *ndm;
1669         struct nlattr *tb[NDA_MAX+1];
1670         struct neigh_table *tbl;
1671         struct net_device *dev = NULL;
1672         int err;
1673
1674         ASSERT_RTNL();
1675         err = nlmsg_parse(nlh, sizeof(*ndm), tb, NDA_MAX, NULL);
1676         if (err < 0)
1677                 goto out;
1678
1679         err = -EINVAL;
1680         if (tb[NDA_DST] == NULL)
1681                 goto out;
1682
1683         ndm = nlmsg_data(nlh);
1684         if (ndm->ndm_ifindex) {
1685                 dev = __dev_get_by_index(net, ndm->ndm_ifindex);
1686                 if (dev == NULL) {
1687                         err = -ENODEV;
1688                         goto out;
1689                 }
1690
1691                 if (tb[NDA_LLADDR] && nla_len(tb[NDA_LLADDR]) < dev->addr_len)
1692                         goto out;
1693         }
1694
1695         read_lock(&neigh_tbl_lock);
1696         for (tbl = neigh_tables; tbl; tbl = tbl->next) {
1697                 int flags = NEIGH_UPDATE_F_ADMIN | NEIGH_UPDATE_F_OVERRIDE;
1698                 struct neighbour *neigh;
1699                 void *dst, *lladdr;
1700
1701                 if (tbl->family != ndm->ndm_family)
1702                         continue;
1703                 read_unlock(&neigh_tbl_lock);
1704
1705                 if (nla_len(tb[NDA_DST]) < tbl->key_len)
1706                         goto out;
1707                 dst = nla_data(tb[NDA_DST]);
1708                 lladdr = tb[NDA_LLADDR] ? nla_data(tb[NDA_LLADDR]) : NULL;
1709
1710                 if (ndm->ndm_flags & NTF_PROXY) {
1711                         struct pneigh_entry *pn;
1712
1713                         err = -ENOBUFS;
1714                         pn = pneigh_lookup(tbl, net, dst, dev, 1);
1715                         if (pn) {
1716                                 pn->flags = ndm->ndm_flags;
1717                                 err = 0;
1718                         }
1719                         goto out;
1720                 }
1721
1722                 if (dev == NULL)
1723                         goto out;
1724
1725                 neigh = neigh_lookup(tbl, dst, dev);
1726                 if (neigh == NULL) {
1727                         if (!(nlh->nlmsg_flags & NLM_F_CREATE)) {
1728                                 err = -ENOENT;
1729                                 goto out;
1730                         }
1731
1732                         neigh = __neigh_lookup_errno(tbl, dst, dev);
1733                         if (IS_ERR(neigh)) {
1734                                 err = PTR_ERR(neigh);
1735                                 goto out;
1736                         }
1737                 } else {
1738                         if (nlh->nlmsg_flags & NLM_F_EXCL) {
1739                                 err = -EEXIST;
1740                                 neigh_release(neigh);
1741                                 goto out;
1742                         }
1743
1744                         if (!(nlh->nlmsg_flags & NLM_F_REPLACE))
1745                                 flags &= ~NEIGH_UPDATE_F_OVERRIDE;
1746                 }
1747
1748                 if (ndm->ndm_flags & NTF_USE) {
1749                         neigh_event_send(neigh, NULL);
1750                         err = 0;
1751                 } else
1752                         err = neigh_update(neigh, lladdr, ndm->ndm_state, flags);
1753                 neigh_release(neigh);
1754                 goto out;
1755         }
1756
1757         read_unlock(&neigh_tbl_lock);
1758         err = -EAFNOSUPPORT;
1759 out:
1760         return err;
1761 }
1762
1763 static int neightbl_fill_parms(struct sk_buff *skb, struct neigh_parms *parms)
1764 {
1765         struct nlattr *nest;
1766
1767         nest = nla_nest_start(skb, NDTA_PARMS);
1768         if (nest == NULL)
1769                 return -ENOBUFS;
1770
1771         if ((parms->dev &&
1772              nla_put_u32(skb, NDTPA_IFINDEX, parms->dev->ifindex)) ||
1773             nla_put_u32(skb, NDTPA_REFCNT, atomic_read(&parms->refcnt)) ||
1774             nla_put_u32(skb, NDTPA_QUEUE_LENBYTES, parms->queue_len_bytes) ||
1775             /* approximative value for deprecated QUEUE_LEN (in packets) */
1776             nla_put_u32(skb, NDTPA_QUEUE_LEN,
1777                         DIV_ROUND_UP(parms->queue_len_bytes,
1778                                      SKB_TRUESIZE(ETH_FRAME_LEN))) ||
1779             nla_put_u32(skb, NDTPA_PROXY_QLEN, parms->proxy_qlen) ||
1780             nla_put_u32(skb, NDTPA_APP_PROBES, parms->app_probes) ||
1781             nla_put_u32(skb, NDTPA_UCAST_PROBES, parms->ucast_probes) ||
1782             nla_put_u32(skb, NDTPA_MCAST_PROBES, parms->mcast_probes) ||
1783             nla_put_msecs(skb, NDTPA_REACHABLE_TIME, parms->reachable_time) ||
1784             nla_put_msecs(skb, NDTPA_BASE_REACHABLE_TIME,
1785                           parms->base_reachable_time) ||
1786             nla_put_msecs(skb, NDTPA_GC_STALETIME, parms->gc_staletime) ||
1787             nla_put_msecs(skb, NDTPA_DELAY_PROBE_TIME,
1788                           parms->delay_probe_time) ||
1789             nla_put_msecs(skb, NDTPA_RETRANS_TIME, parms->retrans_time) ||
1790             nla_put_msecs(skb, NDTPA_ANYCAST_DELAY, parms->anycast_delay) ||
1791             nla_put_msecs(skb, NDTPA_PROXY_DELAY, parms->proxy_delay) ||
1792             nla_put_msecs(skb, NDTPA_LOCKTIME, parms->locktime))
1793                 goto nla_put_failure;
1794         return nla_nest_end(skb, nest);
1795
1796 nla_put_failure:
1797         nla_nest_cancel(skb, nest);
1798         return -EMSGSIZE;
1799 }
1800
1801 static int neightbl_fill_info(struct sk_buff *skb, struct neigh_table *tbl,
1802                               u32 pid, u32 seq, int type, int flags)
1803 {
1804         struct nlmsghdr *nlh;
1805         struct ndtmsg *ndtmsg;
1806
1807         nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
1808         if (nlh == NULL)
1809                 return -EMSGSIZE;
1810
1811         ndtmsg = nlmsg_data(nlh);
1812
1813         read_lock_bh(&tbl->lock);
1814         ndtmsg->ndtm_family = tbl->family;
1815         ndtmsg->ndtm_pad1   = 0;
1816         ndtmsg->ndtm_pad2   = 0;
1817
1818         if (nla_put_string(skb, NDTA_NAME, tbl->id) ||
1819             nla_put_msecs(skb, NDTA_GC_INTERVAL, tbl->gc_interval) ||
1820             nla_put_u32(skb, NDTA_THRESH1, tbl->gc_thresh1) ||
1821             nla_put_u32(skb, NDTA_THRESH2, tbl->gc_thresh2) ||
1822             nla_put_u32(skb, NDTA_THRESH3, tbl->gc_thresh3))
1823                 goto nla_put_failure;
1824         {
1825                 unsigned long now = jiffies;
1826                 unsigned int flush_delta = now - tbl->last_flush;
1827                 unsigned int rand_delta = now - tbl->last_rand;
1828                 struct neigh_hash_table *nht;
1829                 struct ndt_config ndc = {
1830                         .ndtc_key_len           = tbl->key_len,
1831                         .ndtc_entry_size        = tbl->entry_size,
1832                         .ndtc_entries           = atomic_read(&tbl->entries),
1833                         .ndtc_last_flush        = jiffies_to_msecs(flush_delta),
1834                         .ndtc_last_rand         = jiffies_to_msecs(rand_delta),
1835                         .ndtc_proxy_qlen        = tbl->proxy_queue.qlen,
1836                 };
1837
1838                 rcu_read_lock_bh();
1839                 nht = rcu_dereference_bh(tbl->nht);
1840                 ndc.ndtc_hash_rnd = nht->hash_rnd[0];
1841                 ndc.ndtc_hash_mask = ((1 << nht->hash_shift) - 1);
1842                 rcu_read_unlock_bh();
1843
1844                 if (nla_put(skb, NDTA_CONFIG, sizeof(ndc), &ndc))
1845                         goto nla_put_failure;
1846         }
1847
1848         {
1849                 int cpu;
1850                 struct ndt_stats ndst;
1851
1852                 memset(&ndst, 0, sizeof(ndst));
1853
1854                 for_each_possible_cpu(cpu) {
1855                         struct neigh_statistics *st;
1856
1857                         st = per_cpu_ptr(tbl->stats, cpu);
1858                         ndst.ndts_allocs                += st->allocs;
1859                         ndst.ndts_destroys              += st->destroys;
1860                         ndst.ndts_hash_grows            += st->hash_grows;
1861                         ndst.ndts_res_failed            += st->res_failed;
1862                         ndst.ndts_lookups               += st->lookups;
1863                         ndst.ndts_hits                  += st->hits;
1864                         ndst.ndts_rcv_probes_mcast      += st->rcv_probes_mcast;
1865                         ndst.ndts_rcv_probes_ucast      += st->rcv_probes_ucast;
1866                         ndst.ndts_periodic_gc_runs      += st->periodic_gc_runs;
1867                         ndst.ndts_forced_gc_runs        += st->forced_gc_runs;
1868                 }
1869
1870                 if (nla_put(skb, NDTA_STATS, sizeof(ndst), &ndst))
1871                         goto nla_put_failure;
1872         }
1873
1874         BUG_ON(tbl->parms.dev);
1875         if (neightbl_fill_parms(skb, &tbl->parms) < 0)
1876                 goto nla_put_failure;
1877
1878         read_unlock_bh(&tbl->lock);
1879         return nlmsg_end(skb, nlh);
1880
1881 nla_put_failure:
1882         read_unlock_bh(&tbl->lock);
1883         nlmsg_cancel(skb, nlh);
1884         return -EMSGSIZE;
1885 }
1886
1887 static int neightbl_fill_param_info(struct sk_buff *skb,
1888                                     struct neigh_table *tbl,
1889                                     struct neigh_parms *parms,
1890                                     u32 pid, u32 seq, int type,
1891                                     unsigned int flags)
1892 {
1893         struct ndtmsg *ndtmsg;
1894         struct nlmsghdr *nlh;
1895
1896         nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
1897         if (nlh == NULL)
1898                 return -EMSGSIZE;
1899
1900         ndtmsg = nlmsg_data(nlh);
1901
1902         read_lock_bh(&tbl->lock);
1903         ndtmsg->ndtm_family = tbl->family;
1904         ndtmsg->ndtm_pad1   = 0;
1905         ndtmsg->ndtm_pad2   = 0;
1906
1907         if (nla_put_string(skb, NDTA_NAME, tbl->id) < 0 ||
1908             neightbl_fill_parms(skb, parms) < 0)
1909                 goto errout;
1910
1911         read_unlock_bh(&tbl->lock);
1912         return nlmsg_end(skb, nlh);
1913 errout:
1914         read_unlock_bh(&tbl->lock);
1915         nlmsg_cancel(skb, nlh);
1916         return -EMSGSIZE;
1917 }
1918
1919 static const struct nla_policy nl_neightbl_policy[NDTA_MAX+1] = {
1920         [NDTA_NAME]             = { .type = NLA_STRING },
1921         [NDTA_THRESH1]          = { .type = NLA_U32 },
1922         [NDTA_THRESH2]          = { .type = NLA_U32 },
1923         [NDTA_THRESH3]          = { .type = NLA_U32 },
1924         [NDTA_GC_INTERVAL]      = { .type = NLA_U64 },
1925         [NDTA_PARMS]            = { .type = NLA_NESTED },
1926 };
1927
1928 static const struct nla_policy nl_ntbl_parm_policy[NDTPA_MAX+1] = {
1929         [NDTPA_IFINDEX]                 = { .type = NLA_U32 },
1930         [NDTPA_QUEUE_LEN]               = { .type = NLA_U32 },
1931         [NDTPA_PROXY_QLEN]              = { .type = NLA_U32 },
1932         [NDTPA_APP_PROBES]              = { .type = NLA_U32 },
1933         [NDTPA_UCAST_PROBES]            = { .type = NLA_U32 },
1934         [NDTPA_MCAST_PROBES]            = { .type = NLA_U32 },
1935         [NDTPA_BASE_REACHABLE_TIME]     = { .type = NLA_U64 },
1936         [NDTPA_GC_STALETIME]            = { .type = NLA_U64 },
1937         [NDTPA_DELAY_PROBE_TIME]        = { .type = NLA_U64 },
1938         [NDTPA_RETRANS_TIME]            = { .type = NLA_U64 },
1939         [NDTPA_ANYCAST_DELAY]           = { .type = NLA_U64 },
1940         [NDTPA_PROXY_DELAY]             = { .type = NLA_U64 },
1941         [NDTPA_LOCKTIME]                = { .type = NLA_U64 },
1942 };
1943
1944 static int neightbl_set(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
1945 {
1946         struct net *net = sock_net(skb->sk);
1947         struct neigh_table *tbl;
1948         struct ndtmsg *ndtmsg;
1949         struct nlattr *tb[NDTA_MAX+1];
1950         int err;
1951
1952         err = nlmsg_parse(nlh, sizeof(*ndtmsg), tb, NDTA_MAX,
1953                           nl_neightbl_policy);
1954         if (err < 0)
1955                 goto errout;
1956
1957         if (tb[NDTA_NAME] == NULL) {
1958                 err = -EINVAL;
1959                 goto errout;
1960         }
1961
1962         ndtmsg = nlmsg_data(nlh);
1963         read_lock(&neigh_tbl_lock);
1964         for (tbl = neigh_tables; tbl; tbl = tbl->next) {
1965                 if (ndtmsg->ndtm_family && tbl->family != ndtmsg->ndtm_family)
1966                         continue;
1967
1968                 if (nla_strcmp(tb[NDTA_NAME], tbl->id) == 0)
1969                         break;
1970         }
1971
1972         if (tbl == NULL) {
1973                 err = -ENOENT;
1974                 goto errout_locked;
1975         }
1976
1977         /*
1978          * We acquire tbl->lock to be nice to the periodic timers and
1979          * make sure they always see a consistent set of values.
1980          */
1981         write_lock_bh(&tbl->lock);
1982
1983         if (tb[NDTA_PARMS]) {
1984                 struct nlattr *tbp[NDTPA_MAX+1];
1985                 struct neigh_parms *p;
1986                 int i, ifindex = 0;
1987
1988                 err = nla_parse_nested(tbp, NDTPA_MAX, tb[NDTA_PARMS],
1989                                        nl_ntbl_parm_policy);
1990                 if (err < 0)
1991                         goto errout_tbl_lock;
1992
1993                 if (tbp[NDTPA_IFINDEX])
1994                         ifindex = nla_get_u32(tbp[NDTPA_IFINDEX]);
1995
1996                 p = lookup_neigh_parms(tbl, net, ifindex);
1997                 if (p == NULL) {
1998                         err = -ENOENT;
1999                         goto errout_tbl_lock;
2000                 }
2001
2002                 for (i = 1; i <= NDTPA_MAX; i++) {
2003                         if (tbp[i] == NULL)
2004                                 continue;
2005
2006                         switch (i) {
2007                         case NDTPA_QUEUE_LEN:
2008                                 p->queue_len_bytes = nla_get_u32(tbp[i]) *
2009                                                      SKB_TRUESIZE(ETH_FRAME_LEN);
2010                                 break;
2011                         case NDTPA_QUEUE_LENBYTES:
2012                                 p->queue_len_bytes = nla_get_u32(tbp[i]);
2013                                 break;
2014                         case NDTPA_PROXY_QLEN:
2015                                 p->proxy_qlen = nla_get_u32(tbp[i]);
2016                                 break;
2017                         case NDTPA_APP_PROBES:
2018                                 p->app_probes = nla_get_u32(tbp[i]);
2019                                 break;
2020                         case NDTPA_UCAST_PROBES:
2021                                 p->ucast_probes = nla_get_u32(tbp[i]);
2022                                 break;
2023                         case NDTPA_MCAST_PROBES:
2024                                 p->mcast_probes = nla_get_u32(tbp[i]);
2025                                 break;
2026                         case NDTPA_BASE_REACHABLE_TIME:
2027                                 p->base_reachable_time = nla_get_msecs(tbp[i]);
2028                                 break;
2029                         case NDTPA_GC_STALETIME:
2030                                 p->gc_staletime = nla_get_msecs(tbp[i]);
2031                                 break;
2032                         case NDTPA_DELAY_PROBE_TIME:
2033                                 p->delay_probe_time = nla_get_msecs(tbp[i]);
2034                                 break;
2035                         case NDTPA_RETRANS_TIME:
2036                                 p->retrans_time = nla_get_msecs(tbp[i]);
2037                                 break;
2038                         case NDTPA_ANYCAST_DELAY:
2039                                 p->anycast_delay = nla_get_msecs(tbp[i]);
2040                                 break;
2041                         case NDTPA_PROXY_DELAY:
2042                                 p->proxy_delay = nla_get_msecs(tbp[i]);
2043                                 break;
2044                         case NDTPA_LOCKTIME:
2045                                 p->locktime = nla_get_msecs(tbp[i]);
2046                                 break;
2047                         }
2048                 }
2049         }
2050
2051         if (tb[NDTA_THRESH1])
2052                 tbl->gc_thresh1 = nla_get_u32(tb[NDTA_THRESH1]);
2053
2054         if (tb[NDTA_THRESH2])
2055                 tbl->gc_thresh2 = nla_get_u32(tb[NDTA_THRESH2]);
2056
2057         if (tb[NDTA_THRESH3])
2058                 tbl->gc_thresh3 = nla_get_u32(tb[NDTA_THRESH3]);
2059
2060         if (tb[NDTA_GC_INTERVAL])
2061                 tbl->gc_interval = nla_get_msecs(tb[NDTA_GC_INTERVAL]);
2062
2063         err = 0;
2064
2065 errout_tbl_lock:
2066         write_unlock_bh(&tbl->lock);
2067 errout_locked:
2068         read_unlock(&neigh_tbl_lock);
2069 errout:
2070         return err;
2071 }
2072
2073 static int neightbl_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2074 {
2075         struct net *net = sock_net(skb->sk);
2076         int family, tidx, nidx = 0;
2077         int tbl_skip = cb->args[0];
2078         int neigh_skip = cb->args[1];
2079         struct neigh_table *tbl;
2080
2081         family = ((struct rtgenmsg *) nlmsg_data(cb->nlh))->rtgen_family;
2082
2083         read_lock(&neigh_tbl_lock);
2084         for (tbl = neigh_tables, tidx = 0; tbl; tbl = tbl->next, tidx++) {
2085                 struct neigh_parms *p;
2086
2087                 if (tidx < tbl_skip || (family && tbl->family != family))
2088                         continue;
2089
2090                 if (neightbl_fill_info(skb, tbl, NETLINK_CB(cb->skb).pid,
2091                                        cb->nlh->nlmsg_seq, RTM_NEWNEIGHTBL,
2092                                        NLM_F_MULTI) <= 0)
2093                         break;
2094
2095                 for (nidx = 0, p = tbl->parms.next; p; p = p->next) {
2096                         if (!net_eq(neigh_parms_net(p), net))
2097                                 continue;
2098
2099                         if (nidx < neigh_skip)
2100                                 goto next;
2101
2102                         if (neightbl_fill_param_info(skb, tbl, p,
2103                                                      NETLINK_CB(cb->skb).pid,
2104                                                      cb->nlh->nlmsg_seq,
2105                                                      RTM_NEWNEIGHTBL,
2106                                                      NLM_F_MULTI) <= 0)
2107                                 goto out;
2108                 next:
2109                         nidx++;
2110                 }
2111
2112                 neigh_skip = 0;
2113         }
2114 out:
2115         read_unlock(&neigh_tbl_lock);
2116         cb->args[0] = tidx;
2117         cb->args[1] = nidx;
2118
2119         return skb->len;
2120 }
2121
2122 static int neigh_fill_info(struct sk_buff *skb, struct neighbour *neigh,
2123                            u32 pid, u32 seq, int type, unsigned int flags)
2124 {
2125         unsigned long now = jiffies;
2126         struct nda_cacheinfo ci;
2127         struct nlmsghdr *nlh;
2128         struct ndmsg *ndm;
2129
2130         nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2131         if (nlh == NULL)
2132                 return -EMSGSIZE;
2133
2134         ndm = nlmsg_data(nlh);
2135         ndm->ndm_family  = neigh->ops->family;
2136         ndm->ndm_pad1    = 0;
2137         ndm->ndm_pad2    = 0;
2138         ndm->ndm_flags   = neigh->flags;
2139         ndm->ndm_type    = neigh->type;
2140         ndm->ndm_ifindex = neigh->dev->ifindex;
2141
2142         if (nla_put(skb, NDA_DST, neigh->tbl->key_len, neigh->primary_key))
2143                 goto nla_put_failure;
2144
2145         read_lock_bh(&neigh->lock);
2146         ndm->ndm_state   = neigh->nud_state;
2147         if (neigh->nud_state & NUD_VALID) {
2148                 char haddr[MAX_ADDR_LEN];
2149
2150                 neigh_ha_snapshot(haddr, neigh, neigh->dev);
2151                 if (nla_put(skb, NDA_LLADDR, neigh->dev->addr_len, haddr) < 0) {
2152                         read_unlock_bh(&neigh->lock);
2153                         goto nla_put_failure;
2154                 }
2155         }
2156
2157         ci.ndm_used      = jiffies_to_clock_t(now - neigh->used);
2158         ci.ndm_confirmed = jiffies_to_clock_t(now - neigh->confirmed);
2159         ci.ndm_updated   = jiffies_to_clock_t(now - neigh->updated);
2160         ci.ndm_refcnt    = atomic_read(&neigh->refcnt) - 1;
2161         read_unlock_bh(&neigh->lock);
2162
2163         if (nla_put_u32(skb, NDA_PROBES, atomic_read(&neigh->probes)) ||
2164             nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
2165                 goto nla_put_failure;
2166
2167         return nlmsg_end(skb, nlh);
2168
2169 nla_put_failure:
2170         nlmsg_cancel(skb, nlh);
2171         return -EMSGSIZE;
2172 }
2173
2174 static int pneigh_fill_info(struct sk_buff *skb, struct pneigh_entry *pn,
2175                             u32 pid, u32 seq, int type, unsigned int flags,
2176                             struct neigh_table *tbl)
2177 {
2178         struct nlmsghdr *nlh;
2179         struct ndmsg *ndm;
2180
2181         nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2182         if (nlh == NULL)
2183                 return -EMSGSIZE;
2184
2185         ndm = nlmsg_data(nlh);
2186         ndm->ndm_family  = tbl->family;
2187         ndm->ndm_pad1    = 0;
2188         ndm->ndm_pad2    = 0;
2189         ndm->ndm_flags   = pn->flags | NTF_PROXY;
2190         ndm->ndm_type    = NDA_DST;
2191         ndm->ndm_ifindex = pn->dev->ifindex;
2192         ndm->ndm_state   = NUD_NONE;
2193
2194         if (nla_put(skb, NDA_DST, tbl->key_len, pn->key))
2195                 goto nla_put_failure;
2196
2197         return nlmsg_end(skb, nlh);
2198
2199 nla_put_failure:
2200         nlmsg_cancel(skb, nlh);
2201         return -EMSGSIZE;
2202 }
2203
2204 static void neigh_update_notify(struct neighbour *neigh)
2205 {
2206         call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, neigh);
2207         __neigh_notify(neigh, RTM_NEWNEIGH, 0);
2208 }
2209
2210 static int neigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2211                             struct netlink_callback *cb)
2212 {
2213         struct net *net = sock_net(skb->sk);
2214         struct neighbour *n;
2215         int rc, h, s_h = cb->args[1];
2216         int idx, s_idx = idx = cb->args[2];
2217         struct neigh_hash_table *nht;
2218
2219         rcu_read_lock_bh();
2220         nht = rcu_dereference_bh(tbl->nht);
2221
2222         for (h = 0; h < (1 << nht->hash_shift); h++) {
2223                 if (h < s_h)
2224                         continue;
2225                 if (h > s_h)
2226                         s_idx = 0;
2227                 for (n = rcu_dereference_bh(nht->hash_buckets[h]), idx = 0;
2228                      n != NULL;
2229                      n = rcu_dereference_bh(n->next)) {
2230                         if (!net_eq(dev_net(n->dev), net))
2231                                 continue;
2232                         if (idx < s_idx)
2233                                 goto next;
2234                         if (neigh_fill_info(skb, n, NETLINK_CB(cb->skb).pid,
2235                                             cb->nlh->nlmsg_seq,
2236                                             RTM_NEWNEIGH,
2237                                             NLM_F_MULTI) <= 0) {
2238                                 rc = -1;
2239                                 goto out;
2240                         }
2241 next:
2242                         idx++;
2243                 }
2244         }
2245         rc = skb->len;
2246 out:
2247         rcu_read_unlock_bh();
2248         cb->args[1] = h;
2249         cb->args[2] = idx;
2250         return rc;
2251 }
2252
2253 static int pneigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2254                              struct netlink_callback *cb)
2255 {
2256         struct pneigh_entry *n;
2257         struct net *net = sock_net(skb->sk);
2258         int rc, h, s_h = cb->args[3];
2259         int idx, s_idx = idx = cb->args[4];
2260
2261         read_lock_bh(&tbl->lock);
2262
2263         for (h = 0; h <= PNEIGH_HASHMASK; h++) {
2264                 if (h < s_h)
2265                         continue;
2266                 if (h > s_h)
2267                         s_idx = 0;
2268                 for (n = tbl->phash_buckets[h], idx = 0; n; n = n->next) {
2269                         if (dev_net(n->dev) != net)
2270                                 continue;
2271                         if (idx < s_idx)
2272                                 goto next;
2273                         if (pneigh_fill_info(skb, n, NETLINK_CB(cb->skb).pid,
2274                                             cb->nlh->nlmsg_seq,
2275                                             RTM_NEWNEIGH,
2276                                             NLM_F_MULTI, tbl) <= 0) {
2277                                 read_unlock_bh(&tbl->lock);
2278                                 rc = -1;
2279                                 goto out;
2280                         }
2281                 next:
2282                         idx++;
2283                 }
2284         }
2285
2286         read_unlock_bh(&tbl->lock);
2287         rc = skb->len;
2288 out:
2289         cb->args[3] = h;
2290         cb->args[4] = idx;
2291         return rc;
2292
2293 }
2294
2295 static int neigh_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2296 {
2297         struct neigh_table *tbl;
2298         int t, family, s_t;
2299         int proxy = 0;
2300         int err = 0;
2301
2302         read_lock(&neigh_tbl_lock);
2303         family = ((struct rtgenmsg *) nlmsg_data(cb->nlh))->rtgen_family;
2304
2305         /* check for full ndmsg structure presence, family member is
2306          * the same for both structures
2307          */
2308         if (nlmsg_len(cb->nlh) >= sizeof(struct ndmsg) &&
2309             ((struct ndmsg *) nlmsg_data(cb->nlh))->ndm_flags == NTF_PROXY)
2310                 proxy = 1;
2311
2312         s_t = cb->args[0];
2313
2314         for (tbl = neigh_tables, t = 0; tbl && (err >= 0);
2315              tbl = tbl->next, t++) {
2316                 if (t < s_t || (family && tbl->family != family))
2317                         continue;
2318                 if (t > s_t)
2319                         memset(&cb->args[1], 0, sizeof(cb->args) -
2320                                                 sizeof(cb->args[0]));
2321                 if (proxy)
2322                         err = pneigh_dump_table(tbl, skb, cb);
2323                 else
2324                         err = neigh_dump_table(tbl, skb, cb);
2325         }
2326         read_unlock(&neigh_tbl_lock);
2327
2328         cb->args[0] = t;
2329         return skb->len;
2330 }
2331
2332 void neigh_for_each(struct neigh_table *tbl, void (*cb)(struct neighbour *, void *), void *cookie)
2333 {
2334         int chain;
2335         struct neigh_hash_table *nht;
2336
2337         rcu_read_lock_bh();
2338         nht = rcu_dereference_bh(tbl->nht);
2339
2340         read_lock(&tbl->lock); /* avoid resizes */
2341         for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
2342                 struct neighbour *n;
2343
2344                 for (n = rcu_dereference_bh(nht->hash_buckets[chain]);
2345                      n != NULL;
2346                      n = rcu_dereference_bh(n->next))
2347                         cb(n, cookie);
2348         }
2349         read_unlock(&tbl->lock);
2350         rcu_read_unlock_bh();
2351 }
2352 EXPORT_SYMBOL(neigh_for_each);
2353
2354 /* The tbl->lock must be held as a writer and BH disabled. */
2355 void __neigh_for_each_release(struct neigh_table *tbl,
2356                               int (*cb)(struct neighbour *))
2357 {
2358         int chain;
2359         struct neigh_hash_table *nht;
2360
2361         nht = rcu_dereference_protected(tbl->nht,
2362                                         lockdep_is_held(&tbl->lock));
2363         for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
2364                 struct neighbour *n;
2365                 struct neighbour __rcu **np;
2366
2367                 np = &nht->hash_buckets[chain];
2368                 while ((n = rcu_dereference_protected(*np,
2369                                         lockdep_is_held(&tbl->lock))) != NULL) {
2370                         int release;
2371
2372                         write_lock(&n->lock);
2373                         release = cb(n);
2374                         if (release) {
2375                                 rcu_assign_pointer(*np,
2376                                         rcu_dereference_protected(n->next,
2377                                                 lockdep_is_held(&tbl->lock)));
2378                                 n->dead = 1;
2379                         } else
2380                                 np = &n->next;
2381                         write_unlock(&n->lock);
2382                         if (release)
2383                                 neigh_cleanup_and_release(n);
2384                 }
2385         }
2386 }
2387 EXPORT_SYMBOL(__neigh_for_each_release);
2388
2389 #ifdef CONFIG_PROC_FS
2390
2391 static struct neighbour *neigh_get_first(struct seq_file *seq)
2392 {
2393         struct neigh_seq_state *state = seq->private;
2394         struct net *net = seq_file_net(seq);
2395         struct neigh_hash_table *nht = state->nht;
2396         struct neighbour *n = NULL;
2397         int bucket = state->bucket;
2398
2399         state->flags &= ~NEIGH_SEQ_IS_PNEIGH;
2400         for (bucket = 0; bucket < (1 << nht->hash_shift); bucket++) {
2401                 n = rcu_dereference_bh(nht->hash_buckets[bucket]);
2402
2403                 while (n) {
2404                         if (!net_eq(dev_net(n->dev), net))
2405                                 goto next;
2406                         if (state->neigh_sub_iter) {
2407                                 loff_t fakep = 0;
2408                                 void *v;
2409
2410                                 v = state->neigh_sub_iter(state, n, &fakep);
2411                                 if (!v)
2412                                         goto next;
2413                         }
2414                         if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
2415                                 break;
2416                         if (n->nud_state & ~NUD_NOARP)
2417                                 break;
2418 next:
2419                         n = rcu_dereference_bh(n->next);
2420                 }
2421
2422                 if (n)
2423                         break;
2424         }
2425         state->bucket = bucket;
2426
2427         return n;
2428 }
2429
2430 static struct neighbour *neigh_get_next(struct seq_file *seq,
2431                                         struct neighbour *n,
2432                                         loff_t *pos)
2433 {
2434         struct neigh_seq_state *state = seq->private;
2435         struct net *net = seq_file_net(seq);
2436         struct neigh_hash_table *nht = state->nht;
2437
2438         if (state->neigh_sub_iter) {
2439                 void *v = state->neigh_sub_iter(state, n, pos);
2440                 if (v)
2441                         return n;
2442         }
2443         n = rcu_dereference_bh(n->next);
2444
2445         while (1) {
2446                 while (n) {
2447                         if (!net_eq(dev_net(n->dev), net))
2448                                 goto next;
2449                         if (state->neigh_sub_iter) {
2450                                 void *v = state->neigh_sub_iter(state, n, pos);
2451                                 if (v)
2452                                         return n;
2453                                 goto next;
2454                         }
2455                         if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
2456                                 break;
2457
2458                         if (n->nud_state & ~NUD_NOARP)
2459                                 break;
2460 next:
2461                         n = rcu_dereference_bh(n->next);
2462                 }
2463
2464                 if (n)
2465                         break;
2466
2467                 if (++state->bucket >= (1 << nht->hash_shift))
2468                         break;
2469
2470                 n = rcu_dereference_bh(nht->hash_buckets[state->bucket]);
2471         }
2472
2473         if (n && pos)
2474                 --(*pos);
2475         return n;
2476 }
2477
2478 static struct neighbour *neigh_get_idx(struct seq_file *seq, loff_t *pos)
2479 {
2480         struct neighbour *n = neigh_get_first(seq);
2481
2482         if (n) {
2483                 --(*pos);
2484                 while (*pos) {
2485                         n = neigh_get_next(seq, n, pos);
2486                         if (!n)
2487                                 break;
2488                 }
2489         }
2490         return *pos ? NULL : n;
2491 }
2492
2493 static struct pneigh_entry *pneigh_get_first(struct seq_file *seq)
2494 {
2495         struct neigh_seq_state *state = seq->private;
2496         struct net *net = seq_file_net(seq);
2497         struct neigh_table *tbl = state->tbl;
2498         struct pneigh_entry *pn = NULL;
2499         int bucket = state->bucket;
2500
2501         state->flags |= NEIGH_SEQ_IS_PNEIGH;
2502         for (bucket = 0; bucket <= PNEIGH_HASHMASK; bucket++) {
2503                 pn = tbl->phash_buckets[bucket];
2504                 while (pn && !net_eq(pneigh_net(pn), net))
2505                         pn = pn->next;
2506                 if (pn)
2507                         break;
2508         }
2509         state->bucket = bucket;
2510
2511         return pn;
2512 }
2513
2514 static struct pneigh_entry *pneigh_get_next(struct seq_file *seq,
2515                                             struct pneigh_entry *pn,
2516                                             loff_t *pos)
2517 {
2518         struct neigh_seq_state *state = seq->private;
2519         struct net *net = seq_file_net(seq);
2520         struct neigh_table *tbl = state->tbl;
2521
2522         do {
2523                 pn = pn->next;
2524         } while (pn && !net_eq(pneigh_net(pn), net));
2525
2526         while (!pn) {
2527                 if (++state->bucket > PNEIGH_HASHMASK)
2528                         break;
2529                 pn = tbl->phash_buckets[state->bucket];
2530                 while (pn && !net_eq(pneigh_net(pn), net))
2531                         pn = pn->next;
2532                 if (pn)
2533                         break;
2534         }
2535
2536         if (pn && pos)
2537                 --(*pos);
2538
2539         return pn;
2540 }
2541
2542 static struct pneigh_entry *pneigh_get_idx(struct seq_file *seq, loff_t *pos)
2543 {
2544         struct pneigh_entry *pn = pneigh_get_first(seq);
2545
2546         if (pn) {
2547                 --(*pos);
2548                 while (*pos) {
2549                         pn = pneigh_get_next(seq, pn, pos);
2550                         if (!pn)
2551                                 break;
2552                 }
2553         }
2554         return *pos ? NULL : pn;
2555 }
2556
2557 static void *neigh_get_idx_any(struct seq_file *seq, loff_t *pos)
2558 {
2559         struct neigh_seq_state *state = seq->private;
2560         void *rc;
2561         loff_t idxpos = *pos;
2562
2563         rc = neigh_get_idx(seq, &idxpos);
2564         if (!rc && !(state->flags & NEIGH_SEQ_NEIGH_ONLY))
2565                 rc = pneigh_get_idx(seq, &idxpos);
2566
2567         return rc;
2568 }
2569
2570 void *neigh_seq_start(struct seq_file *seq, loff_t *pos, struct neigh_table *tbl, unsigned int neigh_seq_flags)
2571         __acquires(rcu_bh)
2572 {
2573         struct neigh_seq_state *state = seq->private;
2574
2575         state->tbl = tbl;
2576         state->bucket = 0;
2577         state->flags = (neigh_seq_flags & ~NEIGH_SEQ_IS_PNEIGH);
2578
2579         rcu_read_lock_bh();
2580         state->nht = rcu_dereference_bh(tbl->nht);
2581
2582         return *pos ? neigh_get_idx_any(seq, pos) : SEQ_START_TOKEN;
2583 }
2584 EXPORT_SYMBOL(neigh_seq_start);
2585
2586 void *neigh_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2587 {
2588         struct neigh_seq_state *state;
2589         void *rc;
2590
2591         if (v == SEQ_START_TOKEN) {
2592                 rc = neigh_get_first(seq);
2593                 goto out;
2594         }
2595
2596         state = seq->private;
2597         if (!(state->flags & NEIGH_SEQ_IS_PNEIGH)) {
2598                 rc = neigh_get_next(seq, v, NULL);
2599                 if (rc)
2600                         goto out;
2601                 if (!(state->flags & NEIGH_SEQ_NEIGH_ONLY))
2602                         rc = pneigh_get_first(seq);
2603         } else {
2604                 BUG_ON(state->flags & NEIGH_SEQ_NEIGH_ONLY);
2605                 rc = pneigh_get_next(seq, v, NULL);
2606         }
2607 out:
2608         ++(*pos);
2609         return rc;
2610 }
2611 EXPORT_SYMBOL(neigh_seq_next);
2612
2613 void neigh_seq_stop(struct seq_file *seq, void *v)
2614         __releases(rcu_bh)
2615 {
2616         rcu_read_unlock_bh();
2617 }
2618 EXPORT_SYMBOL(neigh_seq_stop);
2619
2620 /* statistics via seq_file */
2621
2622 static void *neigh_stat_seq_start(struct seq_file *seq, loff_t *pos)
2623 {
2624         struct neigh_table *tbl = seq->private;
2625         int cpu;
2626
2627         if (*pos == 0)
2628                 return SEQ_START_TOKEN;
2629
2630         for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
2631                 if (!cpu_possible(cpu))
2632                         continue;
2633                 *pos = cpu+1;
2634                 return per_cpu_ptr(tbl->stats, cpu);
2635         }
2636         return NULL;
2637 }
2638
2639 static void *neigh_stat_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2640 {
2641         struct neigh_table *tbl = seq->private;
2642         int cpu;
2643
2644         for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
2645                 if (!cpu_possible(cpu))
2646                         continue;
2647                 *pos = cpu+1;
2648                 return per_cpu_ptr(tbl->stats, cpu);
2649         }
2650         return NULL;
2651 }
2652
2653 static void neigh_stat_seq_stop(struct seq_file *seq, void *v)
2654 {
2655
2656 }
2657
2658 static int neigh_stat_seq_show(struct seq_file *seq, void *v)
2659 {
2660         struct neigh_table *tbl = seq->private;
2661         struct neigh_statistics *st = v;
2662
2663         if (v == SEQ_START_TOKEN) {
2664                 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");
2665                 return 0;
2666         }
2667
2668         seq_printf(seq, "%08x  %08lx %08lx %08lx  %08lx %08lx  %08lx  "
2669                         "%08lx %08lx  %08lx %08lx %08lx\n",
2670                    atomic_read(&tbl->entries),
2671
2672                    st->allocs,
2673                    st->destroys,
2674                    st->hash_grows,
2675
2676                    st->lookups,
2677                    st->hits,
2678
2679                    st->res_failed,
2680
2681                    st->rcv_probes_mcast,
2682                    st->rcv_probes_ucast,
2683
2684                    st->periodic_gc_runs,
2685                    st->forced_gc_runs,
2686                    st->unres_discards
2687                    );
2688
2689         return 0;
2690 }
2691
2692 static const struct seq_operations neigh_stat_seq_ops = {
2693         .start  = neigh_stat_seq_start,
2694         .next   = neigh_stat_seq_next,
2695         .stop   = neigh_stat_seq_stop,
2696         .show   = neigh_stat_seq_show,
2697 };
2698
2699 static int neigh_stat_seq_open(struct inode *inode, struct file *file)
2700 {
2701         int ret = seq_open(file, &neigh_stat_seq_ops);
2702
2703         if (!ret) {
2704                 struct seq_file *sf = file->private_data;
2705                 sf->private = PDE(inode)->data;
2706         }
2707         return ret;
2708 };
2709
2710 static const struct file_operations neigh_stat_seq_fops = {
2711         .owner   = THIS_MODULE,
2712         .open    = neigh_stat_seq_open,
2713         .read    = seq_read,
2714         .llseek  = seq_lseek,
2715         .release = seq_release,
2716 };
2717
2718 #endif /* CONFIG_PROC_FS */
2719
2720 static inline size_t neigh_nlmsg_size(void)
2721 {
2722         return NLMSG_ALIGN(sizeof(struct ndmsg))
2723                + nla_total_size(MAX_ADDR_LEN) /* NDA_DST */
2724                + nla_total_size(MAX_ADDR_LEN) /* NDA_LLADDR */
2725                + nla_total_size(sizeof(struct nda_cacheinfo))
2726                + nla_total_size(4); /* NDA_PROBES */
2727 }
2728
2729 static void __neigh_notify(struct neighbour *n, int type, int flags)
2730 {
2731         struct net *net = dev_net(n->dev);
2732         struct sk_buff *skb;
2733         int err = -ENOBUFS;
2734
2735         skb = nlmsg_new(neigh_nlmsg_size(), GFP_ATOMIC);
2736         if (skb == NULL)
2737                 goto errout;
2738
2739         err = neigh_fill_info(skb, n, 0, 0, type, flags);
2740         if (err < 0) {
2741                 /* -EMSGSIZE implies BUG in neigh_nlmsg_size() */
2742                 WARN_ON(err == -EMSGSIZE);
2743                 kfree_skb(skb);
2744                 goto errout;
2745         }
2746         rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
2747         return;
2748 errout:
2749         if (err < 0)
2750                 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
2751 }
2752
2753 #ifdef CONFIG_ARPD
2754 void neigh_app_ns(struct neighbour *n)
2755 {
2756         __neigh_notify(n, RTM_GETNEIGH, NLM_F_REQUEST);
2757 }
2758 EXPORT_SYMBOL(neigh_app_ns);
2759 #endif /* CONFIG_ARPD */
2760
2761 #ifdef CONFIG_SYSCTL
2762
2763 static int proc_unres_qlen(ctl_table *ctl, int write, void __user *buffer,
2764                            size_t *lenp, loff_t *ppos)
2765 {
2766         int size, ret;
2767         ctl_table tmp = *ctl;
2768
2769         tmp.data = &size;
2770         size = DIV_ROUND_UP(*(int *)ctl->data, SKB_TRUESIZE(ETH_FRAME_LEN));
2771         ret = proc_dointvec(&tmp, write, buffer, lenp, ppos);
2772         if (write && !ret)
2773                 *(int *)ctl->data = size * SKB_TRUESIZE(ETH_FRAME_LEN);
2774         return ret;
2775 }
2776
2777 enum {
2778         NEIGH_VAR_MCAST_PROBE,
2779         NEIGH_VAR_UCAST_PROBE,
2780         NEIGH_VAR_APP_PROBE,
2781         NEIGH_VAR_RETRANS_TIME,
2782         NEIGH_VAR_BASE_REACHABLE_TIME,
2783         NEIGH_VAR_DELAY_PROBE_TIME,
2784         NEIGH_VAR_GC_STALETIME,
2785         NEIGH_VAR_QUEUE_LEN,
2786         NEIGH_VAR_QUEUE_LEN_BYTES,
2787         NEIGH_VAR_PROXY_QLEN,
2788         NEIGH_VAR_ANYCAST_DELAY,
2789         NEIGH_VAR_PROXY_DELAY,
2790         NEIGH_VAR_LOCKTIME,
2791         NEIGH_VAR_RETRANS_TIME_MS,
2792         NEIGH_VAR_BASE_REACHABLE_TIME_MS,
2793         NEIGH_VAR_GC_INTERVAL,
2794         NEIGH_VAR_GC_THRESH1,
2795         NEIGH_VAR_GC_THRESH2,
2796         NEIGH_VAR_GC_THRESH3,
2797         NEIGH_VAR_MAX
2798 };
2799
2800 static struct neigh_sysctl_table {
2801         struct ctl_table_header *sysctl_header;
2802         struct ctl_table neigh_vars[NEIGH_VAR_MAX + 1];
2803 } neigh_sysctl_template __read_mostly = {
2804         .neigh_vars = {
2805                 [NEIGH_VAR_MCAST_PROBE] = {
2806                         .procname       = "mcast_solicit",
2807                         .maxlen         = sizeof(int),
2808                         .mode           = 0644,
2809                         .proc_handler   = proc_dointvec,
2810                 },
2811                 [NEIGH_VAR_UCAST_PROBE] = {
2812                         .procname       = "ucast_solicit",
2813                         .maxlen         = sizeof(int),
2814                         .mode           = 0644,
2815                         .proc_handler   = proc_dointvec,
2816                 },
2817                 [NEIGH_VAR_APP_PROBE] = {
2818                         .procname       = "app_solicit",
2819                         .maxlen         = sizeof(int),
2820                         .mode           = 0644,
2821                         .proc_handler   = proc_dointvec,
2822                 },
2823                 [NEIGH_VAR_RETRANS_TIME] = {
2824                         .procname       = "retrans_time",
2825                         .maxlen         = sizeof(int),
2826                         .mode           = 0644,
2827                         .proc_handler   = proc_dointvec_userhz_jiffies,
2828                 },
2829                 [NEIGH_VAR_BASE_REACHABLE_TIME] = {
2830                         .procname       = "base_reachable_time",
2831                         .maxlen         = sizeof(int),
2832                         .mode           = 0644,
2833                         .proc_handler   = proc_dointvec_jiffies,
2834                 },
2835                 [NEIGH_VAR_DELAY_PROBE_TIME] = {
2836                         .procname       = "delay_first_probe_time",
2837                         .maxlen         = sizeof(int),
2838                         .mode           = 0644,
2839                         .proc_handler   = proc_dointvec_jiffies,
2840                 },
2841                 [NEIGH_VAR_GC_STALETIME] = {
2842                         .procname       = "gc_stale_time",
2843                         .maxlen         = sizeof(int),
2844                         .mode           = 0644,
2845                         .proc_handler   = proc_dointvec_jiffies,
2846                 },
2847                 [NEIGH_VAR_QUEUE_LEN] = {
2848                         .procname       = "unres_qlen",
2849                         .maxlen         = sizeof(int),
2850                         .mode           = 0644,
2851                         .proc_handler   = proc_unres_qlen,
2852                 },
2853                 [NEIGH_VAR_QUEUE_LEN_BYTES] = {
2854                         .procname       = "unres_qlen_bytes",
2855                         .maxlen         = sizeof(int),
2856                         .mode           = 0644,
2857                         .proc_handler   = proc_dointvec,
2858                 },
2859                 [NEIGH_VAR_PROXY_QLEN] = {
2860                         .procname       = "proxy_qlen",
2861                         .maxlen         = sizeof(int),
2862                         .mode           = 0644,
2863                         .proc_handler   = proc_dointvec,
2864                 },
2865                 [NEIGH_VAR_ANYCAST_DELAY] = {
2866                         .procname       = "anycast_delay",
2867                         .maxlen         = sizeof(int),
2868                         .mode           = 0644,
2869                         .proc_handler   = proc_dointvec_userhz_jiffies,
2870                 },
2871                 [NEIGH_VAR_PROXY_DELAY] = {
2872                         .procname       = "proxy_delay",
2873                         .maxlen         = sizeof(int),
2874                         .mode           = 0644,
2875                         .proc_handler   = proc_dointvec_userhz_jiffies,
2876                 },
2877                 [NEIGH_VAR_LOCKTIME] = {
2878                         .procname       = "locktime",
2879                         .maxlen         = sizeof(int),
2880                         .mode           = 0644,
2881                         .proc_handler   = proc_dointvec_userhz_jiffies,
2882                 },
2883                 [NEIGH_VAR_RETRANS_TIME_MS] = {
2884                         .procname       = "retrans_time_ms",
2885                         .maxlen         = sizeof(int),
2886                         .mode           = 0644,
2887                         .proc_handler   = proc_dointvec_ms_jiffies,
2888                 },
2889                 [NEIGH_VAR_BASE_REACHABLE_TIME_MS] = {
2890                         .procname       = "base_reachable_time_ms",
2891                         .maxlen         = sizeof(int),
2892                         .mode           = 0644,
2893                         .proc_handler   = proc_dointvec_ms_jiffies,
2894                 },
2895                 [NEIGH_VAR_GC_INTERVAL] = {
2896                         .procname       = "gc_interval",
2897                         .maxlen         = sizeof(int),
2898                         .mode           = 0644,
2899                         .proc_handler   = proc_dointvec_jiffies,
2900                 },
2901                 [NEIGH_VAR_GC_THRESH1] = {
2902                         .procname       = "gc_thresh1",
2903                         .maxlen         = sizeof(int),
2904                         .mode           = 0644,
2905                         .proc_handler   = proc_dointvec,
2906                 },
2907                 [NEIGH_VAR_GC_THRESH2] = {
2908                         .procname       = "gc_thresh2",
2909                         .maxlen         = sizeof(int),
2910                         .mode           = 0644,
2911                         .proc_handler   = proc_dointvec,
2912                 },
2913                 [NEIGH_VAR_GC_THRESH3] = {
2914                         .procname       = "gc_thresh3",
2915                         .maxlen         = sizeof(int),
2916                         .mode           = 0644,
2917                         .proc_handler   = proc_dointvec,
2918                 },
2919                 {},
2920         },
2921 };
2922
2923 int neigh_sysctl_register(struct net_device *dev, struct neigh_parms *p,
2924                           char *p_name, proc_handler *handler)
2925 {
2926         struct neigh_sysctl_table *t;
2927         const char *dev_name_source = NULL;
2928         char neigh_path[ sizeof("net//neigh/") + IFNAMSIZ + IFNAMSIZ ];
2929
2930         t = kmemdup(&neigh_sysctl_template, sizeof(*t), GFP_KERNEL);
2931         if (!t)
2932                 goto err;
2933
2934         t->neigh_vars[NEIGH_VAR_MCAST_PROBE].data  = &p->mcast_probes;
2935         t->neigh_vars[NEIGH_VAR_UCAST_PROBE].data  = &p->ucast_probes;
2936         t->neigh_vars[NEIGH_VAR_APP_PROBE].data  = &p->app_probes;
2937         t->neigh_vars[NEIGH_VAR_RETRANS_TIME].data  = &p->retrans_time;
2938         t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].data  = &p->base_reachable_time;
2939         t->neigh_vars[NEIGH_VAR_DELAY_PROBE_TIME].data  = &p->delay_probe_time;
2940         t->neigh_vars[NEIGH_VAR_GC_STALETIME].data  = &p->gc_staletime;
2941         t->neigh_vars[NEIGH_VAR_QUEUE_LEN].data  = &p->queue_len_bytes;
2942         t->neigh_vars[NEIGH_VAR_QUEUE_LEN_BYTES].data  = &p->queue_len_bytes;
2943         t->neigh_vars[NEIGH_VAR_PROXY_QLEN].data  = &p->proxy_qlen;
2944         t->neigh_vars[NEIGH_VAR_ANYCAST_DELAY].data  = &p->anycast_delay;
2945         t->neigh_vars[NEIGH_VAR_PROXY_DELAY].data = &p->proxy_delay;
2946         t->neigh_vars[NEIGH_VAR_LOCKTIME].data = &p->locktime;
2947         t->neigh_vars[NEIGH_VAR_RETRANS_TIME_MS].data  = &p->retrans_time;
2948         t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].data  = &p->base_reachable_time;
2949
2950         if (dev) {
2951                 dev_name_source = dev->name;
2952                 /* Terminate the table early */
2953                 memset(&t->neigh_vars[NEIGH_VAR_GC_INTERVAL], 0,
2954                        sizeof(t->neigh_vars[NEIGH_VAR_GC_INTERVAL]));
2955         } else {
2956                 dev_name_source = "default";
2957                 t->neigh_vars[NEIGH_VAR_GC_INTERVAL].data = (int *)(p + 1);
2958                 t->neigh_vars[NEIGH_VAR_GC_THRESH1].data = (int *)(p + 1) + 1;
2959                 t->neigh_vars[NEIGH_VAR_GC_THRESH2].data = (int *)(p + 1) + 2;
2960                 t->neigh_vars[NEIGH_VAR_GC_THRESH3].data = (int *)(p + 1) + 3;
2961         }
2962
2963
2964         if (handler) {
2965                 /* RetransTime */
2966                 t->neigh_vars[NEIGH_VAR_RETRANS_TIME].proc_handler = handler;
2967                 t->neigh_vars[NEIGH_VAR_RETRANS_TIME].extra1 = dev;
2968                 /* ReachableTime */
2969                 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].proc_handler = handler;
2970                 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].extra1 = dev;
2971                 /* RetransTime (in milliseconds)*/
2972                 t->neigh_vars[NEIGH_VAR_RETRANS_TIME_MS].proc_handler = handler;
2973                 t->neigh_vars[NEIGH_VAR_RETRANS_TIME_MS].extra1 = dev;
2974                 /* ReachableTime (in milliseconds) */
2975                 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].proc_handler = handler;
2976                 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].extra1 = dev;
2977         }
2978
2979         snprintf(neigh_path, sizeof(neigh_path), "net/%s/neigh/%s",
2980                 p_name, dev_name_source);
2981         t->sysctl_header =
2982                 register_net_sysctl(neigh_parms_net(p), neigh_path, t->neigh_vars);
2983         if (!t->sysctl_header)
2984                 goto free;
2985
2986         p->sysctl_table = t;
2987         return 0;
2988
2989 free:
2990         kfree(t);
2991 err:
2992         return -ENOBUFS;
2993 }
2994 EXPORT_SYMBOL(neigh_sysctl_register);
2995
2996 void neigh_sysctl_unregister(struct neigh_parms *p)
2997 {
2998         if (p->sysctl_table) {
2999                 struct neigh_sysctl_table *t = p->sysctl_table;
3000                 p->sysctl_table = NULL;
3001                 unregister_net_sysctl_table(t->sysctl_header);
3002                 kfree(t);
3003         }
3004 }
3005 EXPORT_SYMBOL(neigh_sysctl_unregister);
3006
3007 #endif  /* CONFIG_SYSCTL */
3008
3009 static int __init neigh_init(void)
3010 {
3011         rtnl_register(PF_UNSPEC, RTM_NEWNEIGH, neigh_add, NULL, NULL);
3012         rtnl_register(PF_UNSPEC, RTM_DELNEIGH, neigh_delete, NULL, NULL);
3013         rtnl_register(PF_UNSPEC, RTM_GETNEIGH, NULL, neigh_dump_info, NULL);
3014
3015         rtnl_register(PF_UNSPEC, RTM_GETNEIGHTBL, NULL, neightbl_dump_info,
3016                       NULL);
3017         rtnl_register(PF_UNSPEC, RTM_SETNEIGHTBL, neightbl_set, NULL, NULL);
3018
3019         return 0;
3020 }
3021
3022 subsys_initcall(neigh_init);
3023