kernel/net/core/neighbour.c

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