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 #include <linux/types.h>
  19 #include <linux/kernel.h>
  20 #include <linux/module.h>
  21 #include <linux/socket.h>
  22 #include <linux/netdevice.h>
  23 #include <linux/proc_fs.h>
  24 #ifdef CONFIG_SYSCTL
  25 #include <linux/sysctl.h>
  26 #endif
  27 #include <linux/times.h>
  28 #include <net/net_namespace.h>
  29 #include <net/neighbour.h>
  30 #include <net/dst.h>
  31 #include <net/sock.h>
  32 #include <net/netevent.h>
  33 #include <net/netlink.h>
  34 #include <linux/rtnetlink.h>
  35 #include <linux/random.h>
  36 #include <linux/string.h>
  37 #include <linux/log2.h>
  38
  39 #define NEIGH_DEBUG 1
  40
  41 #define NEIGH_PRINTK(x...) printk(x)
  42 #define NEIGH_NOPRINTK(x...) do { ; } while(0)
  43 #define NEIGH_PRINTK0 NEIGH_PRINTK
  44 #define NEIGH_PRINTK1 NEIGH_NOPRINTK
  45 #define NEIGH_PRINTK2 NEIGH_NOPRINTK
  46
  47 #if NEIGH_DEBUG >= 1
  48 #undef NEIGH_PRINTK1
  49 #define NEIGH_PRINTK1 NEIGH_PRINTK
  50 #endif
  51 #if NEIGH_DEBUG >= 2
  52 #undef NEIGH_PRINTK2
  53 #define NEIGH_PRINTK2 NEIGH_PRINTK
  54 #endif
  55
  56 #define PNEIGH_HASHMASK         0xF
  57
  58 static void neigh_timer_handler(unsigned long arg);
  59 static void __neigh_notify(struct neighbour *n, int type, int flags);
  60 static void neigh_update_notify(struct neighbour *neigh);
  61 static int pneigh_ifdown(struct neigh_table *tbl, struct net_device *dev);
  62
  63 static struct neigh_table *neigh_tables;
  64 #ifdef CONFIG_PROC_FS
  65 static const struct file_operations neigh_stat_seq_fops;
  66 #endif
  67
  68 /*
  69    Neighbour hash table buckets are protected with rwlock tbl->lock.
  70
  71    - All the scans/updates to hash buckets MUST be made under this lock.
  72    - NOTHING clever should be made under this lock: no callbacks
  73      to protocol backends, no attempts to send something to network.
  74      It will result in deadlocks, if backend/driver wants to use neighbour
  75      cache.
  76    - If the entry requires some non-trivial actions, increase
  77      its reference count and release table lock.
  78
  79    Neighbour entries are protected:
  80    - with reference count.
  81    - with rwlock neigh->lock
  82
  83    Reference count prevents destruction.
  84
  85    neigh->lock mainly serializes ll address data and its validity state.
  86    However, the same lock is used to protect another entry fields:
  87     - timer
  88     - resolution queue
  89
  90    Again, nothing clever shall be made under neigh->lock,
  91    the most complicated procedure, which we allow is dev->hard_header.
  92    It is supposed, that dev->hard_header is simplistic and does
  93    not make callbacks to neighbour tables.
  94
  95    The last lock is neigh_tbl_lock. It is pure SMP lock, protecting
  96    list of neighbour tables. This list is used only in process context,
  97  */
  98
  99 static DEFINE_RWLOCK(neigh_tbl_lock);
 100
 101 static int neigh_blackhole(struct sk_buff *skb)
 102 {
 103         kfree_skb(skb);
 104         return -ENETDOWN;
 105 }
 106
 107 static void neigh_cleanup_and_release(struct neighbour *neigh)
 108 {
 109         if (neigh->parms->neigh_cleanup)
 110                 neigh->parms->neigh_cleanup(neigh);
 111
 112         __neigh_notify(neigh, RTM_DELNEIGH, 0);
 113         neigh_release(neigh);
 114 }
 115
 116 /*
 117  * It is random distribution in the interval (1/2)*base...(3/2)*base.
 118  * It corresponds to default IPv6 settings and is not overridable,
 119  * because it is really reasonable choice.
 120  */
 121
 122 unsigned long neigh_rand_reach_time(unsigned long base)
 123 {
 124         return (base ? (net_random() % base) + (base >> 1) : 0);
 125 }
 126 EXPORT_SYMBOL(neigh_rand_reach_time);
 127
 128
 129 static int neigh_forced_gc(struct neigh_table *tbl)
 130 {
 131         int shrunk = 0;
 132         int i;
 133
 134         NEIGH_CACHE_STAT_INC(tbl, forced_gc_runs);
 135
 136         write_lock_bh(&tbl->lock);
 137         for (i = 0; i <= tbl->hash_mask; i++) {
 138                 struct neighbour *n, **np;
 139
 140                 np = &tbl->hash_buckets[i];
 141                 while ((n = *np) != NULL) {
 142                         /* Neighbour record may be discarded if:
 143                          * - nobody refers to it.
 144                          * - it is not permanent
 145                          */
 146                         write_lock(&n->lock);
 147                         if (atomic_read(&n->refcnt) == 1 &&
 148                             !(n->nud_state & NUD_PERMANENT)) {
 149                                 *np     = n->next;
 150                                 n->dead = 1;
 151                                 shrunk  = 1;
 152                                 write_unlock(&n->lock);
 153                                 neigh_cleanup_and_release(n);
 154                                 continue;
 155                         }
 156                         write_unlock(&n->lock);
 157                         np = &n->next;
 158                 }
 159         }
 160
 161         tbl->last_flush = jiffies;
 162
 163         write_unlock_bh(&tbl->lock);
 164
 165         return shrunk;
 166 }
 167
 168 static void neigh_add_timer(struct neighbour *n, unsigned long when)
 169 {
 170         neigh_hold(n);
 171         if (unlikely(mod_timer(&n->timer, when))) {
 172                 printk("NEIGH: BUG, double timer add, state is %x\n",
 173                        n->nud_state);
 174                 dump_stack();
 175         }
 176 }
 177
 178 static int neigh_del_timer(struct neighbour *n)
 179 {
 180         if ((n->nud_state & NUD_IN_TIMER) &&
 181             del_timer(&n->timer)) {
 182                 neigh_release(n);
 183                 return 1;
 184         }
 185         return 0;
 186 }
 187
 188 static void pneigh_queue_purge(struct sk_buff_head *list)
 189 {
 190         struct sk_buff *skb;
 191
 192         while ((skb = skb_dequeue(list)) != NULL) {
 193                 dev_put(skb->dev);
 194                 kfree_skb(skb);
 195         }
 196 }
 197
 198 static void neigh_flush_dev(struct neigh_table *tbl, struct net_device *dev)
 199 {
 200         int i;
 201
 202         for (i = 0; i <= tbl->hash_mask; i++) {
 203                 struct neighbour *n, **np = &tbl->hash_buckets[i];
 204
 205                 while ((n = *np) != NULL) {
 206                         if (dev && n->dev != dev) {
 207                                 np = &n->next;
 208                                 continue;
 209                         }
 210                         *np = n->next;
 211                         write_lock(&n->lock);
 212                         neigh_del_timer(n);
 213                         n->dead = 1;
 214
 215                         if (atomic_read(&n->refcnt) != 1) {
 216                                 /* The most unpleasant situation.
 217                                    We must destroy neighbour entry,
 218                                    but someone still uses it.
 219
 220                                    The destroy will be delayed until
 221                                    the last user releases us, but
 222                                    we must kill timers etc. and move
 223                                    it to safe state.
 224                                  */
 225                                 skb_queue_purge(&n->arp_queue);
 226                                 n->output = neigh_blackhole;
 227                                 if (n->nud_state & NUD_VALID)
 228                                         n->nud_state = NUD_NOARP;
 229                                 else
 230                                         n->nud_state = NUD_NONE;
 231                                 NEIGH_PRINTK2("neigh %p is stray.\n", n);
 232                         }
 233                         write_unlock(&n->lock);
 234                         neigh_cleanup_and_release(n);
 235                 }
 236         }
 237 }
 238
 239 void neigh_changeaddr(struct neigh_table *tbl, struct net_device *dev)
 240 {
 241         write_lock_bh(&tbl->lock);
 242         neigh_flush_dev(tbl, dev);
 243         write_unlock_bh(&tbl->lock);
 244 }
 245 EXPORT_SYMBOL(neigh_changeaddr);
 246
 247 int neigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
 248 {
 249         write_lock_bh(&tbl->lock);
 250         neigh_flush_dev(tbl, dev);
 251         pneigh_ifdown(tbl, dev);
 252         write_unlock_bh(&tbl->lock);
 253
 254         del_timer_sync(&tbl->proxy_timer);
 255         pneigh_queue_purge(&tbl->proxy_queue);
 256         return 0;
 257 }
 258 EXPORT_SYMBOL(neigh_ifdown);
 259
 260 static struct neighbour *neigh_alloc(struct neigh_table *tbl)
 261 {
 262         struct neighbour *n = NULL;
 263         unsigned long now = jiffies;
 264         int entries;
 265
 266         entries = atomic_inc_return(&tbl->entries) - 1;
 267         if (entries >= tbl->gc_thresh3 ||
 268             (entries >= tbl->gc_thresh2 &&
 269              time_after(now, tbl->last_flush + 5 * HZ))) {
 270                 if (!neigh_forced_gc(tbl) &&
 271                     entries >= tbl->gc_thresh3)
 272                         goto out_entries;
 273         }
 274
 275         n = kmem_cache_zalloc(tbl->kmem_cachep, GFP_ATOMIC);
 276         if (!n)
 277                 goto out_entries;
 278
 279         skb_queue_head_init(&n->arp_queue);
 280         rwlock_init(&n->lock);
 281         n->updated        = n->used = now;
 282         n->nud_state      = NUD_NONE;
 283         n->output         = neigh_blackhole;
 284         n->parms          = neigh_parms_clone(&tbl->parms);
 285         setup_timer(&n->timer, neigh_timer_handler, (unsigned long)n);
 286
 287         NEIGH_CACHE_STAT_INC(tbl, allocs);
 288         n->tbl            = tbl;
 289         atomic_set(&n->refcnt, 1);
 290         n->dead           = 1;
 291 out:
 292         return n;
 293
 294 out_entries:
 295         atomic_dec(&tbl->entries);
 296         goto out;
 297 }
 298
 299 static struct neighbour **neigh_hash_alloc(unsigned int entries)
 300 {
 301         unsigned long size = entries * sizeof(struct neighbour *);
 302         struct neighbour **ret;
 303
 304         if (size <= PAGE_SIZE) {
 305                 ret = kzalloc(size, GFP_ATOMIC);
 306         } else {
 307                 ret = (struct neighbour **)
 308                       __get_free_pages(GFP_ATOMIC|__GFP_ZERO, get_order(size));
 309         }
 310         return ret;
 311 }
 312
 313 static void neigh_hash_free(struct neighbour **hash, unsigned int entries)
 314 {
 315         unsigned long size = entries * sizeof(struct neighbour *);
 316
 317         if (size <= PAGE_SIZE)
 318                 kfree(hash);
 319         else
 320                 free_pages((unsigned long)hash, get_order(size));
 321 }
 322
 323 static void neigh_hash_grow(struct neigh_table *tbl, unsigned long new_entries)
 324 {
 325         struct neighbour **new_hash, **old_hash;
 326         unsigned int i, new_hash_mask, old_entries;
 327
 328         NEIGH_CACHE_STAT_INC(tbl, hash_grows);
 329
 330         BUG_ON(!is_power_of_2(new_entries));
 331         new_hash = neigh_hash_alloc(new_entries);
 332         if (!new_hash)
 333                 return;
 334
 335         old_entries = tbl->hash_mask + 1;
 336         new_hash_mask = new_entries - 1;
 337         old_hash = tbl->hash_buckets;
 338
 339         get_random_bytes(&tbl->hash_rnd, sizeof(tbl->hash_rnd));
 340         for (i = 0; i < old_entries; i++) {
 341                 struct neighbour *n, *next;
 342
 343                 for (n = old_hash[i]; n; n = next) {
 344                         unsigned int hash_val = tbl->hash(n->primary_key, n->dev);
 345
 346                         hash_val &= new_hash_mask;
 347                         next = n->next;
 348
 349                         n->next = new_hash[hash_val];
 350                         new_hash[hash_val] = n;
 351                 }
 352         }
 353         tbl->hash_buckets = new_hash;
 354         tbl->hash_mask = new_hash_mask;
 355
 356         neigh_hash_free(old_hash, old_entries);
 357 }
 358
 359 struct neighbour *neigh_lookup(struct neigh_table *tbl, const void *pkey,
 360                                struct net_device *dev)
 361 {
 362         struct neighbour *n;
 363         int key_len = tbl->key_len;
 364         u32 hash_val;
 365
 366         NEIGH_CACHE_STAT_INC(tbl, lookups);
 367
 368         read_lock_bh(&tbl->lock);
 369         hash_val = tbl->hash(pkey, dev);
 370         for (n = tbl->hash_buckets[hash_val & tbl->hash_mask]; n; n = n->next) {
 371                 if (dev == n->dev && !memcmp(n->primary_key, pkey, key_len)) {
 372                         neigh_hold(n);
 373                         NEIGH_CACHE_STAT_INC(tbl, hits);
 374                         break;
 375                 }
 376         }
 377         read_unlock_bh(&tbl->lock);
 378         return n;
 379 }
 380 EXPORT_SYMBOL(neigh_lookup);
 381
 382 struct neighbour *neigh_lookup_nodev(struct neigh_table *tbl, struct net *net,
 383                                      const void *pkey)
 384 {
 385         struct neighbour *n;
 386         int key_len = tbl->key_len;
 387         u32 hash_val;
 388
 389         NEIGH_CACHE_STAT_INC(tbl, lookups);
 390
 391         read_lock_bh(&tbl->lock);
 392         hash_val = tbl->hash(pkey, NULL);
 393         for (n = tbl->hash_buckets[hash_val & tbl->hash_mask]; n; n = n->next) {
 394                 if (!memcmp(n->primary_key, pkey, key_len) &&
 395                     net_eq(dev_net(n->dev), net)) {
 396                         neigh_hold(n);
 397                         NEIGH_CACHE_STAT_INC(tbl, hits);
 398                         break;
 399                 }
 400         }
 401         read_unlock_bh(&tbl->lock);
 402         return n;
 403 }
 404 EXPORT_SYMBOL(neigh_lookup_nodev);
 405
 406 struct neighbour *neigh_create(struct neigh_table *tbl, const void *pkey,
 407                                struct net_device *dev)
 408 {
 409         u32 hash_val;
 410         int key_len = tbl->key_len;
 411         int error;
 412         struct neighbour *n1, *rc, *n = neigh_alloc(tbl);
 413
 414         if (!n) {
 415                 rc = ERR_PTR(-ENOBUFS);
 416                 goto out;
 417         }
 418
 419         memcpy(n->primary_key, pkey, key_len);
 420         n->dev = dev;
 421         dev_hold(dev);
 422
 423         /* Protocol specific setup. */
 424         if (tbl->constructor && (error = tbl->constructor(n)) < 0) {
 425                 rc = ERR_PTR(error);
 426                 goto out_neigh_release;
 427         }
 428
 429         /* Device specific setup. */
 430         if (n->parms->neigh_setup &&
 431             (error = n->parms->neigh_setup(n)) < 0) {
 432                 rc = ERR_PTR(error);
 433                 goto out_neigh_release;
 434         }
 435
 436         n->confirmed = jiffies - (n->parms->base_reachable_time << 1);
 437
 438         write_lock_bh(&tbl->lock);
 439
 440         if (atomic_read(&tbl->entries) > (tbl->hash_mask + 1))
 441                 neigh_hash_grow(tbl, (tbl->hash_mask + 1) << 1);
 442
 443         hash_val = tbl->hash(pkey, dev) & tbl->hash_mask;
 444
 445         if (n->parms->dead) {
 446                 rc = ERR_PTR(-EINVAL);
 447                 goto out_tbl_unlock;
 448         }
 449
 450         for (n1 = tbl->hash_buckets[hash_val]; n1; n1 = n1->next) {
 451                 if (dev == n1->dev && !memcmp(n1->primary_key, pkey, key_len)) {
 452                         neigh_hold(n1);
 453                         rc = n1;
 454                         goto out_tbl_unlock;
 455                 }
 456         }
 457
 458         n->next = tbl->hash_buckets[hash_val];
 459         tbl->hash_buckets[hash_val] = n;
 460         n->dead = 0;
 461         neigh_hold(n);
 462         write_unlock_bh(&tbl->lock);
 463         NEIGH_PRINTK2("neigh %p is created.\n", n);
 464         rc = n;
 465 out:
 466         return rc;
 467 out_tbl_unlock:
 468         write_unlock_bh(&tbl->lock);
 469 out_neigh_release:
 470         neigh_release(n);
 471         goto out;
 472 }
 473 EXPORT_SYMBOL(neigh_create);
 474
 475 static u32 pneigh_hash(const void *pkey, int key_len)
 476 {
 477         u32 hash_val = *(u32 *)(pkey + key_len - 4);
 478         hash_val ^= (hash_val >> 16);
 479         hash_val ^= hash_val >> 8;
 480         hash_val ^= hash_val >> 4;
 481         hash_val &= PNEIGH_HASHMASK;
 482         return hash_val;
 483 }
 484
 485 static struct pneigh_entry *__pneigh_lookup_1(struct pneigh_entry *n,
 486                                               struct net *net,
 487                                               const void *pkey,
 488                                               int key_len,
 489                                               struct net_device *dev)
 490 {
 491         while (n) {
 492                 if (!memcmp(n->key, pkey, key_len) &&
 493                     net_eq(pneigh_net(n), net) &&
 494                     (n->dev == dev || !n->dev))
 495                         return n;
 496                 n = n->next;
 497         }
 498         return NULL;
 499 }
 500
 501 struct pneigh_entry *__pneigh_lookup(struct neigh_table *tbl,
 502                 struct net *net, const void *pkey, struct net_device *dev)
 503 {
 504         int key_len = tbl->key_len;
 505         u32 hash_val = pneigh_hash(pkey, key_len);
 506
 507         return __pneigh_lookup_1(tbl->phash_buckets[hash_val],
 508                                  net, pkey, key_len, dev);
 509 }
 510 EXPORT_SYMBOL_GPL(__pneigh_lookup);
 511
 512 struct pneigh_entry * pneigh_lookup(struct neigh_table *tbl,
 513                                     struct net *net, const void *pkey,
 514                                     struct net_device *dev, int creat)
 515 {
 516         struct pneigh_entry *n;
 517         int key_len = tbl->key_len;
 518         u32 hash_val = pneigh_hash(pkey, key_len);
 519
 520         read_lock_bh(&tbl->lock);
 521         n = __pneigh_lookup_1(tbl->phash_buckets[hash_val],
 522                               net, pkey, key_len, dev);
 523         read_unlock_bh(&tbl->lock);
 524
 525         if (n || !creat)
 526                 goto out;
 527
 528         ASSERT_RTNL();
 529
 530         n = kmalloc(sizeof(*n) + key_len, GFP_KERNEL);
 531         if (!n)
 532                 goto out;
 533
 534         write_pnet(&n->net, hold_net(net));
 535         memcpy(n->key, pkey, key_len);
 536         n->dev = dev;
 537         if (dev)
 538                 dev_hold(dev);
 539
 540         if (tbl->pconstructor && tbl->pconstructor(n)) {
 541                 if (dev)
 542                         dev_put(dev);
 543                 release_net(net);
 544                 kfree(n);
 545                 n = NULL;
 546                 goto out;
 547         }
 548
 549         write_lock_bh(&tbl->lock);
 550         n->next = tbl->phash_buckets[hash_val];
 551         tbl->phash_buckets[hash_val] = n;
 552         write_unlock_bh(&tbl->lock);
 553 out:
 554         return n;
 555 }
 556 EXPORT_SYMBOL(pneigh_lookup);
 557
 558
 559 int pneigh_delete(struct neigh_table *tbl, struct net *net, const void *pkey,
 560                   struct net_device *dev)
 561 {
 562         struct pneigh_entry *n, **np;
 563         int key_len = tbl->key_len;
 564         u32 hash_val = pneigh_hash(pkey, key_len);
 565
 566         write_lock_bh(&tbl->lock);
 567         for (np = &tbl->phash_buckets[hash_val]; (n = *np) != NULL;
 568              np = &n->next) {
 569                 if (!memcmp(n->key, pkey, key_len) && n->dev == dev &&
 570                     net_eq(pneigh_net(n), net)) {
 571                         *np = n->next;
 572                         write_unlock_bh(&tbl->lock);
 573                         if (tbl->pdestructor)
 574                                 tbl->pdestructor(n);
 575                         if (n->dev)
 576                                 dev_put(n->dev);
 577                         release_net(pneigh_net(n));
 578                         kfree(n);
 579                         return 0;
 580                 }
 581         }
 582         write_unlock_bh(&tbl->lock);
 583         return -ENOENT;
 584 }
 585
 586 static int pneigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
 587 {
 588         struct pneigh_entry *n, **np;
 589         u32 h;
 590
 591         for (h = 0; h <= PNEIGH_HASHMASK; h++) {
 592                 np = &tbl->phash_buckets[h];
 593                 while ((n = *np) != NULL) {
 594                         if (!dev || n->dev == dev) {
 595                                 *np = n->next;
 596                                 if (tbl->pdestructor)
 597                                         tbl->pdestructor(n);
 598                                 if (n->dev)
 599                                         dev_put(n->dev);
 600                                 release_net(pneigh_net(n));
 601                                 kfree(n);
 602                                 continue;
 603                         }
 604                         np = &n->next;
 605                 }
 606         }
 607         return -ENOENT;
 608 }
 609
 610 static void neigh_parms_destroy(struct neigh_parms *parms);
 611
 612 static inline void neigh_parms_put(struct neigh_parms *parms)
 613 {
 614         if (atomic_dec_and_test(&parms->refcnt))
 615                 neigh_parms_destroy(parms);
 616 }
 617
 618 /*
 619  *      neighbour must already be out of the table;
 620  *
 621  */
 622 void neigh_destroy(struct neighbour *neigh)
 623 {
 624         struct hh_cache *hh;
 625
 626         NEIGH_CACHE_STAT_INC(neigh->tbl, destroys);
 627
 628         if (!neigh->dead) {
 629                 printk(KERN_WARNING
 630                        "Destroying alive neighbour %p\n", neigh);
 631                 dump_stack();
 632                 return;
 633         }
 634
 635         if (neigh_del_timer(neigh))
 636                 printk(KERN_WARNING "Impossible event.\n");
 637
 638         while ((hh = neigh->hh) != NULL) {
 639                 neigh->hh = hh->hh_next;
 640                 hh->hh_next = NULL;
 641
 642                 write_seqlock_bh(&hh->hh_lock);
 643                 hh->hh_output = neigh_blackhole;
 644                 write_sequnlock_bh(&hh->hh_lock);
 645                 if (atomic_dec_and_test(&hh->hh_refcnt))
 646                         kfree(hh);
 647         }
 648
 649         skb_queue_purge(&neigh->arp_queue);
 650
 651         dev_put(neigh->dev);
 652         neigh_parms_put(neigh->parms);
 653
 654         NEIGH_PRINTK2("neigh %p is destroyed.\n", neigh);
 655
 656         atomic_dec(&neigh->tbl->entries);
 657         kmem_cache_free(neigh->tbl->kmem_cachep, neigh);
 658 }
 659 EXPORT_SYMBOL(neigh_destroy);
 660
 661 /* Neighbour state is suspicious;
 662    disable fast path.
 663
 664    Called with write_locked neigh.
 665  */
 666 static void neigh_suspect(struct neighbour *neigh)
 667 {
 668         struct hh_cache *hh;
 669
 670         NEIGH_PRINTK2("neigh %p is suspected.\n", neigh);
 671
 672         neigh->output = neigh->ops->output;
 673
 674         for (hh = neigh->hh; hh; hh = hh->hh_next)
 675                 hh->hh_output = neigh->ops->output;
 676 }
 677
 678 /* Neighbour state is OK;
 679    enable fast path.
 680
 681    Called with write_locked neigh.
 682  */
 683 static void neigh_connect(struct neighbour *neigh)
 684 {
 685         struct hh_cache *hh;
 686
 687         NEIGH_PRINTK2("neigh %p is connected.\n", neigh);
 688
 689         neigh->output = neigh->ops->connected_output;
 690
 691         for (hh = neigh->hh; hh; hh = hh->hh_next)
 692                 hh->hh_output = neigh->ops->hh_output;
 693 }
 694
 695 static void neigh_periodic_work(struct work_struct *work)
 696 {
 697         struct neigh_table *tbl = container_of(work, struct neigh_table, gc_work.work);
 698         struct neighbour *n, **np;
 699         unsigned int i;
 700
 701         NEIGH_CACHE_STAT_INC(tbl, periodic_gc_runs);
 702
 703         write_lock_bh(&tbl->lock);
 704
 705         /*
 706          *      periodically recompute ReachableTime from random function
 707          */
 708
 709         if (time_after(jiffies, tbl->last_rand + 300 * HZ)) {
 710                 struct neigh_parms *p;
 711                 tbl->last_rand = jiffies;
 712                 for (p = &tbl->parms; p; p = p->next)
 713                         p->reachable_time =
 714                                 neigh_rand_reach_time(p->base_reachable_time);
 715         }
 716
 717         for (i = 0 ; i <= tbl->hash_mask; i++) {
 718                 np = &tbl->hash_buckets[i];
 719
 720                 while ((n = *np) != NULL) {
 721                         unsigned int state;
 722
 723                         write_lock(&n->lock);
 724
 725                         state = n->nud_state;
 726                         if (state & (NUD_PERMANENT | NUD_IN_TIMER)) {
 727                                 write_unlock(&n->lock);
 728                                 goto next_elt;
 729                         }
 730
 731                         if (time_before(n->used, n->confirmed))
 732                                 n->used = n->confirmed;
 733
 734                         if (atomic_read(&n->refcnt) == 1 &&
 735                             (state == NUD_FAILED ||
 736                              time_after(jiffies, n->used + n->parms->gc_staletime))) {
 737                                 *np = n->next;
 738                                 n->dead = 1;
 739                                 write_unlock(&n->lock);
 740                                 neigh_cleanup_and_release(n);
 741                                 continue;
 742                         }
 743                         write_unlock(&n->lock);
 744
 745 next_elt:
 746                         np = &n->next;
 747                 }
 748                 /*
 749                  * It's fine to release lock here, even if hash table
 750                  * grows while we are preempted.
 751                  */
 752                 write_unlock_bh(&tbl->lock);
 753                 cond_resched();
 754                 write_lock_bh(&tbl->lock);
 755         }
 756         /* Cycle through all hash buckets every base_reachable_time/2 ticks.
 757          * ARP entry timeouts range from 1/2 base_reachable_time to 3/2
 758          * base_reachable_time.
 759          */
 760         schedule_delayed_work(&tbl->gc_work,
 761                               tbl->parms.base_reachable_time >> 1);
 762         write_unlock_bh(&tbl->lock);
 763 }
 764
 765 static __inline__ int neigh_max_probes(struct neighbour *n)
 766 {
 767         struct neigh_parms *p = n->parms;
 768         return (n->nud_state & NUD_PROBE ?
 769                 p->ucast_probes :
 770                 p->ucast_probes + p->app_probes + p->mcast_probes);
 771 }
 772
 773 static void neigh_invalidate(struct neighbour *neigh)
 774         __releases(neigh->lock)
 775         __acquires(neigh->lock)
 776 {
 777         struct sk_buff *skb;
 778
 779         NEIGH_CACHE_STAT_INC(neigh->tbl, res_failed);
 780         NEIGH_PRINTK2("neigh %p is failed.\n", neigh);
 781         neigh->updated = jiffies;
 782
 783         /* It is very thin place. report_unreachable is very complicated
 784            routine. Particularly, it can hit the same neighbour entry!
 785
 786            So that, we try to be accurate and avoid dead loop. --ANK
 787          */
 788         while (neigh->nud_state == NUD_FAILED &&
 789                (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
 790                 write_unlock(&neigh->lock);
 791                 neigh->ops->error_report(neigh, skb);
 792                 write_lock(&neigh->lock);
 793         }
 794         skb_queue_purge(&neigh->arp_queue);
 795 }
 796
 797 /* Called when a timer expires for a neighbour entry. */
 798
 799 static void neigh_timer_handler(unsigned long arg)
 800 {
 801         unsigned long now, next;
 802         struct neighbour *neigh = (struct neighbour *)arg;
 803         unsigned state;
 804         int notify = 0;
 805
 806         write_lock(&neigh->lock);
 807
 808         state = neigh->nud_state;
 809         now = jiffies;
 810         next = now + HZ;
 811
 812         if (!(state & NUD_IN_TIMER)) {
 813 #ifndef CONFIG_SMP
 814                 printk(KERN_WARNING "neigh: timer & !nud_in_timer\n");
 815 #endif
 816                 goto out;
 817         }
 818
 819         if (state & NUD_REACHABLE) {
 820                 if (time_before_eq(now,
 821                                    neigh->confirmed + neigh->parms->reachable_time)) {
 822                         NEIGH_PRINTK2("neigh %p is still alive.\n", neigh);
 823                         next = neigh->confirmed + neigh->parms->reachable_time;
 824                 } else if (time_before_eq(now,
 825                                           neigh->used + neigh->parms->delay_probe_time)) {
 826                         NEIGH_PRINTK2("neigh %p is delayed.\n", neigh);
 827                         neigh->nud_state = NUD_DELAY;
 828                         neigh->updated = jiffies;
 829                         neigh_suspect(neigh);
 830                         next = now + neigh->parms->delay_probe_time;
 831                 } else {
 832                         NEIGH_PRINTK2("neigh %p is suspected.\n", neigh);
 833                         neigh->nud_state = NUD_STALE;
 834                         neigh->updated = jiffies;
 835                         neigh_suspect(neigh);
 836                         notify = 1;
 837                 }
 838         } else if (state & NUD_DELAY) {
 839                 if (time_before_eq(now,
 840                                    neigh->confirmed + neigh->parms->delay_probe_time)) {
 841                         NEIGH_PRINTK2("neigh %p is now reachable.\n", neigh);
 842                         neigh->nud_state = NUD_REACHABLE;
 843                         neigh->updated = jiffies;
 844                         neigh_connect(neigh);
 845                         notify = 1;
 846                         next = neigh->confirmed + neigh->parms->reachable_time;
 847                 } else {
 848                         NEIGH_PRINTK2("neigh %p is probed.\n", neigh);
 849                         neigh->nud_state = NUD_PROBE;
 850                         neigh->updated = jiffies;
 851                         atomic_set(&neigh->probes, 0);
 852                         next = now + neigh->parms->retrans_time;
 853                 }
 854         } else {
 855                 /* NUD_PROBE|NUD_INCOMPLETE */
 856                 next = now + neigh->parms->retrans_time;
 857         }
 858
 859         if ((neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) &&
 860             atomic_read(&neigh->probes) >= neigh_max_probes(neigh)) {
 861                 neigh->nud_state = NUD_FAILED;
 862                 notify = 1;
 863                 neigh_invalidate(neigh);
 864         }
 865
 866         if (neigh->nud_state & NUD_IN_TIMER) {
 867                 if (time_before(next, jiffies + HZ/2))
 868                         next = jiffies + HZ/2;
 869                 if (!mod_timer(&neigh->timer, next))
 870                         neigh_hold(neigh);
 871         }
 872         if (neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) {
 873                 struct sk_buff *skb = skb_peek(&neigh->arp_queue);
 874                 /* keep skb alive even if arp_queue overflows */
 875                 if (skb)
 876                         skb = skb_copy(skb, GFP_ATOMIC);
 877                 write_unlock(&neigh->lock);
 878                 neigh->ops->solicit(neigh, skb);
 879                 atomic_inc(&neigh->probes);
 880                 kfree_skb(skb);
 881         } else {
 882 out:
 883                 write_unlock(&neigh->lock);
 884         }
 885
 886         if (notify)
 887                 neigh_update_notify(neigh);
 888
 889         neigh_release(neigh);
 890 }
 891
 892 int __neigh_event_send(struct neighbour *neigh, struct sk_buff *skb)
 893 {
 894         int rc;
 895         unsigned long now;
 896
 897         write_lock_bh(&neigh->lock);
 898
 899         rc = 0;
 900         if (neigh->nud_state & (NUD_CONNECTED | NUD_DELAY | NUD_PROBE))
 901                 goto out_unlock_bh;
 902
 903         now = jiffies;
 904
 905         if (!(neigh->nud_state & (NUD_STALE | NUD_INCOMPLETE))) {
 906                 if (neigh->parms->mcast_probes + neigh->parms->app_probes) {
 907                         atomic_set(&neigh->probes, neigh->parms->ucast_probes);
 908                         neigh->nud_state     = NUD_INCOMPLETE;
 909                         neigh->updated = jiffies;
 910                         neigh_add_timer(neigh, now + 1);
 911                 } else {
 912                         neigh->nud_state = NUD_FAILED;
 913                         neigh->updated = jiffies;
 914                         write_unlock_bh(&neigh->lock);
 915
 916                         kfree_skb(skb);
 917                         return 1;
 918                 }
 919         } else if (neigh->nud_state & NUD_STALE) {
 920                 NEIGH_PRINTK2("neigh %p is delayed.\n", neigh);
 921                 neigh->nud_state = NUD_DELAY;
 922                 neigh->updated = jiffies;
 923                 neigh_add_timer(neigh,
 924                                 jiffies + neigh->parms->delay_probe_time);
 925         }
 926
 927         if (neigh->nud_state == NUD_INCOMPLETE) {
 928                 if (skb) {
 929                         if (skb_queue_len(&neigh->arp_queue) >=
 930                             neigh->parms->queue_len) {
 931                                 struct sk_buff *buff;
 932                                 buff = __skb_dequeue(&neigh->arp_queue);
 933                                 kfree_skb(buff);
 934                                 NEIGH_CACHE_STAT_INC(neigh->tbl, unres_discards);
 935                         }
 936                         __skb_queue_tail(&neigh->arp_queue, skb);
 937                 }
 938                 rc = 1;
 939         }
 940 out_unlock_bh:
 941         write_unlock_bh(&neigh->lock);
 942         return rc;
 943 }
 944 EXPORT_SYMBOL(__neigh_event_send);
 945
 946 static void neigh_update_hhs(struct neighbour *neigh)
 947 {
 948         struct hh_cache *hh;
 949         void (*update)(struct hh_cache*, const struct net_device*, const unsigned char *)
 950                 = neigh->dev->header_ops->cache_update;
 951
 952         if (update) {
 953                 for (hh = neigh->hh; hh; hh = hh->hh_next) {
 954                         write_seqlock_bh(&hh->hh_lock);
 955                         update(hh, neigh->dev, neigh->ha);
 956                         write_sequnlock_bh(&hh->hh_lock);
 957                 }
 958         }
 959 }
 960
 961
 962
 963 /* Generic update routine.
 964    -- lladdr is new lladdr or NULL, if it is not supplied.
 965    -- new    is new state.
 966    -- flags
 967         NEIGH_UPDATE_F_OVERRIDE allows to override existing lladdr,
 968                                 if it is different.
 969         NEIGH_UPDATE_F_WEAK_OVERRIDE will suspect existing "connected"
 970                                 lladdr instead of overriding it
 971                                 if it is different.
 972                                 It also allows to retain current state
 973                                 if lladdr is unchanged.
 974         NEIGH_UPDATE_F_ADMIN    means that the change is administrative.
 975
 976         NEIGH_UPDATE_F_OVERRIDE_ISROUTER allows to override existing
 977                                 NTF_ROUTER flag.
 978         NEIGH_UPDATE_F_ISROUTER indicates if the neighbour is known as
 979                                 a router.
 980
 981    Caller MUST hold reference count on the entry.
 982  */
 983
 984 int neigh_update(struct neighbour *neigh, const u8 *lladdr, u8 new,
 985                  u32 flags)
 986 {
 987         u8 old;
 988         int err;
 989         int notify = 0;
 990         struct net_device *dev;
 991         int update_isrouter = 0;
 992
 993         write_lock_bh(&neigh->lock);
 994
 995         dev    = neigh->dev;
 996         old    = neigh->nud_state;
 997         err    = -EPERM;
 998
 999         if (!(flags & NEIGH_UPDATE_F_ADMIN) &&
1000             (old & (NUD_NOARP | NUD_PERMANENT)))
1001                 goto out;
1002
1003         if (!(new & NUD_VALID)) {
1004                 neigh_del_timer(neigh);
1005                 if (old & NUD_CONNECTED)
1006                         neigh_suspect(neigh);
1007                 neigh->nud_state = new;
1008                 err = 0;
1009                 notify = old & NUD_VALID;
1010                 if ((old & (NUD_INCOMPLETE | NUD_PROBE)) &&
1011                     (new & NUD_FAILED)) {
1012                         neigh_invalidate(neigh);
1013                         notify = 1;
1014                 }
1015                 goto out;
1016         }
1017
1018         /* Compare new lladdr with cached one */
1019         if (!dev->addr_len) {
1020                 /* First case: device needs no address. */
1021                 lladdr = neigh->ha;
1022         } else if (lladdr) {
1023                 /* The second case: if something is already cached
1024                    and a new address is proposed:
1025                    - compare new & old
1026                    - if they are different, check override flag
1027                  */
1028                 if ((old & NUD_VALID) &&
1029                     !memcmp(lladdr, neigh->ha, dev->addr_len))
1030                         lladdr = neigh->ha;
1031         } else {
1032                 /* No address is supplied; if we know something,
1033                    use it, otherwise discard the request.
1034                  */
1035                 err = -EINVAL;
1036                 if (!(old & NUD_VALID))
1037                         goto out;
1038                 lladdr = neigh->ha;
1039         }
1040
1041         if (new & NUD_CONNECTED)
1042                 neigh->confirmed = jiffies;
1043         neigh->updated = jiffies;
1044
1045         /* If entry was valid and address is not changed,
1046            do not change entry state, if new one is STALE.
1047          */
1048         err = 0;
1049         update_isrouter = flags & NEIGH_UPDATE_F_OVERRIDE_ISROUTER;
1050         if (old & NUD_VALID) {
1051                 if (lladdr != neigh->ha && !(flags & NEIGH_UPDATE_F_OVERRIDE)) {
1052                         update_isrouter = 0;
1053                         if ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) &&
1054                             (old & NUD_CONNECTED)) {
1055                                 lladdr = neigh->ha;
1056                                 new = NUD_STALE;
1057                         } else
1058                                 goto out;
1059                 } else {
1060                         if (lladdr == neigh->ha && new == NUD_STALE &&
1061                             ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) ||
1062                              (old & NUD_CONNECTED))
1063                             )
1064                                 new = old;
1065                 }
1066         }
1067
1068         if (new != old) {
1069                 neigh_del_timer(neigh);
1070                 if (new & NUD_IN_TIMER)
1071                         neigh_add_timer(neigh, (jiffies +
1072                                                 ((new & NUD_REACHABLE) ?
1073                                                  neigh->parms->reachable_time :
1074                                                  0)));
1075                 neigh->nud_state = new;
1076         }
1077
1078         if (lladdr != neigh->ha) {
1079                 memcpy(&neigh->ha, lladdr, dev->addr_len);
1080                 neigh_update_hhs(neigh);
1081                 if (!(new & NUD_CONNECTED))
1082                         neigh->confirmed = jiffies -
1083                                       (neigh->parms->base_reachable_time << 1);
1084                 notify = 1;
1085         }
1086         if (new == old)
1087                 goto out;
1088         if (new & NUD_CONNECTED)
1089                 neigh_connect(neigh);
1090         else
1091                 neigh_suspect(neigh);
1092         if (!(old & NUD_VALID)) {
1093                 struct sk_buff *skb;
1094
1095                 /* Again: avoid dead loop if something went wrong */
1096
1097                 while (neigh->nud_state & NUD_VALID &&
1098                        (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
1099                         struct neighbour *n1 = neigh;
1100                         write_unlock_bh(&neigh->lock);
1101                         /* On shaper/eql skb->dst->neighbour != neigh :( */
1102                         if (skb_dst(skb) && skb_dst(skb)->neighbour)
1103                                 n1 = skb_dst(skb)->neighbour;
1104                         n1->output(skb);
1105                         write_lock_bh(&neigh->lock);
1106                 }
1107                 skb_queue_purge(&neigh->arp_queue);
1108         }
1109 out:
1110         if (update_isrouter) {
1111                 neigh->flags = (flags & NEIGH_UPDATE_F_ISROUTER) ?
1112                         (neigh->flags | NTF_ROUTER) :
1113                         (neigh->flags & ~NTF_ROUTER);
1114         }
1115         write_unlock_bh(&neigh->lock);
1116
1117         if (notify)
1118                 neigh_update_notify(neigh);
1119
1120         return err;
1121 }
1122 EXPORT_SYMBOL(neigh_update);
1123
1124 struct neighbour *neigh_event_ns(struct neigh_table *tbl,
1125                                  u8 *lladdr, void *saddr,
1126                                  struct net_device *dev)
1127 {
1128         struct neighbour *neigh = __neigh_lookup(tbl, saddr, dev,
1129                                                  lladdr || !dev->addr_len);
1130         if (neigh)
1131                 neigh_update(neigh, lladdr, NUD_STALE,
1132                              NEIGH_UPDATE_F_OVERRIDE);
1133         return neigh;
1134 }
1135 EXPORT_SYMBOL(neigh_event_ns);
1136
1137 static void neigh_hh_init(struct neighbour *n, struct dst_entry *dst,
1138                           __be16 protocol)
1139 {
1140         struct hh_cache *hh;
1141         struct net_device *dev = dst->dev;
1142
1143         for (hh = n->hh; hh; hh = hh->hh_next)
1144                 if (hh->hh_type == protocol)
1145                         break;
1146
1147         if (!hh && (hh = kzalloc(sizeof(*hh), GFP_ATOMIC)) != NULL) {
1148                 seqlock_init(&hh->hh_lock);
1149                 hh->hh_type = protocol;
1150                 atomic_set(&hh->hh_refcnt, 0);
1151                 hh->hh_next = NULL;
1152
1153                 if (dev->header_ops->cache(n, hh)) {
1154                         kfree(hh);
1155                         hh = NULL;
1156                 } else {
1157                         atomic_inc(&hh->hh_refcnt);
1158                         hh->hh_next = n->hh;
1159                         n->hh       = hh;
1160                         if (n->nud_state & NUD_CONNECTED)
1161                                 hh->hh_output = n->ops->hh_output;
1162                         else
1163                                 hh->hh_output = n->ops->output;
1164                 }
1165         }
1166         if (hh) {
1167                 atomic_inc(&hh->hh_refcnt);
1168                 dst->hh = hh;
1169         }
1170 }
1171
1172 /* This function can be used in contexts, where only old dev_queue_xmit
1173    worked, f.e. if you want to override normal output path (eql, shaper),
1174    but resolution is not made yet.
1175  */
1176
1177 int neigh_compat_output(struct sk_buff *skb)
1178 {
1179         struct net_device *dev = skb->dev;
1180
1181         __skb_pull(skb, skb_network_offset(skb));
1182
1183         if (dev_hard_header(skb, dev, ntohs(skb->protocol), NULL, NULL,
1184                             skb->len) < 0 &&
1185             dev->header_ops->rebuild(skb))
1186                 return 0;
1187
1188         return dev_queue_xmit(skb);
1189 }
1190 EXPORT_SYMBOL(neigh_compat_output);
1191
1192 /* Slow and careful. */
1193
1194 int neigh_resolve_output(struct sk_buff *skb)
1195 {
1196         struct dst_entry *dst = skb_dst(skb);
1197         struct neighbour *neigh;
1198         int rc = 0;
1199
1200         if (!dst || !(neigh = dst->neighbour))
1201                 goto discard;
1202
1203         __skb_pull(skb, skb_network_offset(skb));
1204
1205         if (!neigh_event_send(neigh, skb)) {
1206                 int err;
1207                 struct net_device *dev = neigh->dev;
1208                 if (dev->header_ops->cache && !dst->hh) {
1209                         write_lock_bh(&neigh->lock);
1210                         if (!dst->hh)
1211                                 neigh_hh_init(neigh, dst, dst->ops->protocol);
1212                         err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1213                                               neigh->ha, NULL, skb->len);
1214                         write_unlock_bh(&neigh->lock);
1215                 } else {
1216                         read_lock_bh(&neigh->lock);
1217                         err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1218                                               neigh->ha, NULL, skb->len);
1219                         read_unlock_bh(&neigh->lock);
1220                 }
1221                 if (err >= 0)
1222                         rc = neigh->ops->queue_xmit(skb);
1223                 else
1224                         goto out_kfree_skb;
1225         }
1226 out:
1227         return rc;
1228 discard:
1229         NEIGH_PRINTK1("neigh_resolve_output: dst=%p neigh=%p\n",
1230                       dst, dst ? dst->neighbour : NULL);
1231 out_kfree_skb:
1232         rc = -EINVAL;
1233         kfree_skb(skb);
1234         goto out;
1235 }
1236 EXPORT_SYMBOL(neigh_resolve_output);
1237
1238 /* As fast as possible without hh cache */
1239
1240 int neigh_connected_output(struct sk_buff *skb)
1241 {
1242         int err;
1243         struct dst_entry *dst = skb_dst(skb);
1244         struct neighbour *neigh = dst->neighbour;
1245         struct net_device *dev = neigh->dev;
1246
1247         __skb_pull(skb, skb_network_offset(skb));
1248
1249         read_lock_bh(&neigh->lock);
1250         err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1251                               neigh->ha, NULL, skb->len);
1252         read_unlock_bh(&neigh->lock);
1253         if (err >= 0)
1254                 err = neigh->ops->queue_xmit(skb);
1255         else {
1256                 err = -EINVAL;
1257                 kfree_skb(skb);
1258         }
1259         return err;
1260 }
1261 EXPORT_SYMBOL(neigh_connected_output);
1262
1263 static void neigh_proxy_process(unsigned long arg)
1264 {
1265         struct neigh_table *tbl = (struct neigh_table *)arg;
1266         long sched_next = 0;
1267         unsigned long now = jiffies;
1268         struct sk_buff *skb, *n;
1269
1270         spin_lock(&tbl->proxy_queue.lock);
1271
1272         skb_queue_walk_safe(&tbl->proxy_queue, skb, n) {
1273                 long tdif = NEIGH_CB(skb)->sched_next - now;
1274
1275                 if (tdif <= 0) {
1276                         struct net_device *dev = skb->dev;
1277                         __skb_unlink(skb, &tbl->proxy_queue);
1278                         if (tbl->proxy_redo && netif_running(dev))
1279                                 tbl->proxy_redo(skb);
1280                         else
1281                                 kfree_skb(skb);
1282
1283                         dev_put(dev);
1284                 } else if (!sched_next || tdif < sched_next)
1285                         sched_next = tdif;
1286         }
1287         del_timer(&tbl->proxy_timer);
1288         if (sched_next)
1289                 mod_timer(&tbl->proxy_timer, jiffies + sched_next);
1290         spin_unlock(&tbl->proxy_queue.lock);
1291 }
1292
1293 void pneigh_enqueue(struct neigh_table *tbl, struct neigh_parms *p,
1294                     struct sk_buff *skb)
1295 {
1296         unsigned long now = jiffies;
1297         unsigned long sched_next = now + (net_random() % p->proxy_delay);
1298
1299         if (tbl->proxy_queue.qlen > p->proxy_qlen) {
1300                 kfree_skb(skb);
1301                 return;
1302         }
1303
1304         NEIGH_CB(skb)->sched_next = sched_next;
1305         NEIGH_CB(skb)->flags |= LOCALLY_ENQUEUED;
1306
1307         spin_lock(&tbl->proxy_queue.lock);
1308         if (del_timer(&tbl->proxy_timer)) {
1309                 if (time_before(tbl->proxy_timer.expires, sched_next))
1310                         sched_next = tbl->proxy_timer.expires;
1311         }
1312         skb_dst_drop(skb);
1313         dev_hold(skb->dev);
1314         __skb_queue_tail(&tbl->proxy_queue, skb);
1315         mod_timer(&tbl->proxy_timer, sched_next);
1316         spin_unlock(&tbl->proxy_queue.lock);
1317 }
1318 EXPORT_SYMBOL(pneigh_enqueue);
1319
1320 static inline struct neigh_parms *lookup_neigh_parms(struct neigh_table *tbl,
1321                                                       struct net *net, int ifindex)
1322 {
1323         struct neigh_parms *p;
1324
1325         for (p = &tbl->parms; p; p = p->next) {
1326                 if ((p->dev && p->dev->ifindex == ifindex && net_eq(neigh_parms_net(p), net)) ||
1327                     (!p->dev && !ifindex))
1328                         return p;
1329         }
1330
1331         return NULL;
1332 }
1333
1334 struct neigh_parms *neigh_parms_alloc(struct net_device *dev,
1335                                       struct neigh_table *tbl)
1336 {
1337         struct neigh_parms *p, *ref;
1338         struct net *net = dev_net(dev);
1339         const struct net_device_ops *ops = dev->netdev_ops;
1340
1341         ref = lookup_neigh_parms(tbl, net, 0);
1342         if (!ref)
1343                 return NULL;
1344
1345         p = kmemdup(ref, sizeof(*p), GFP_KERNEL);
1346         if (p) {
1347                 p->tbl            = tbl;
1348                 atomic_set(&p->refcnt, 1);
1349                 p->reachable_time =
1350                                 neigh_rand_reach_time(p->base_reachable_time);
1351
1352                 if (ops->ndo_neigh_setup && ops->ndo_neigh_setup(dev, p)) {
1353                         kfree(p);
1354                         return NULL;
1355                 }
1356
1357                 dev_hold(dev);
1358                 p->dev = dev;
1359                 write_pnet(&p->net, hold_net(net));
1360                 p->sysctl_table = NULL;
1361                 write_lock_bh(&tbl->lock);
1362                 p->next         = tbl->parms.next;
1363                 tbl->parms.next = p;
1364                 write_unlock_bh(&tbl->lock);
1365         }
1366         return p;
1367 }
1368 EXPORT_SYMBOL(neigh_parms_alloc);
1369
1370 static void neigh_rcu_free_parms(struct rcu_head *head)
1371 {
1372         struct neigh_parms *parms =
1373                 container_of(head, struct neigh_parms, rcu_head);
1374
1375         neigh_parms_put(parms);
1376 }
1377
1378 void neigh_parms_release(struct neigh_table *tbl, struct neigh_parms *parms)
1379 {
1380         struct neigh_parms **p;
1381
1382         if (!parms || parms == &tbl->parms)
1383                 return;
1384         write_lock_bh(&tbl->lock);
1385         for (p = &tbl->parms.next; *p; p = &(*p)->next) {
1386                 if (*p == parms) {
1387                         *p = parms->next;
1388                         parms->dead = 1;
1389                         write_unlock_bh(&tbl->lock);
1390                         if (parms->dev)
1391                                 dev_put(parms->dev);
1392                         call_rcu(&parms->rcu_head, neigh_rcu_free_parms);
1393                         return;
1394                 }
1395         }
1396         write_unlock_bh(&tbl->lock);
1397         NEIGH_PRINTK1("neigh_parms_release: not found\n");
1398 }
1399 EXPORT_SYMBOL(neigh_parms_release);
1400
1401 static void neigh_parms_destroy(struct neigh_parms *parms)
1402 {
1403         release_net(neigh_parms_net(parms));
1404         kfree(parms);
1405 }
1406
1407 static struct lock_class_key neigh_table_proxy_queue_class;
1408
1409 void neigh_table_init_no_netlink(struct neigh_table *tbl)
1410 {
1411         unsigned long now = jiffies;
1412         unsigned long phsize;
1413
1414         write_pnet(&tbl->parms.net, &init_net);
1415         atomic_set(&tbl->parms.refcnt, 1);
1416         tbl->parms.reachable_time =
1417                           neigh_rand_reach_time(tbl->parms.base_reachable_time);
1418
1419         if (!tbl->kmem_cachep)
1420                 tbl->kmem_cachep =
1421                         kmem_cache_create(tbl->id, tbl->entry_size, 0,
1422                                           SLAB_HWCACHE_ALIGN|SLAB_PANIC,
1423                                           NULL);
1424         tbl->stats = alloc_percpu(struct neigh_statistics);
1425         if (!tbl->stats)
1426                 panic("cannot create neighbour cache statistics");
1427
1428 #ifdef CONFIG_PROC_FS
1429         if (!proc_create_data(tbl->id, 0, init_net.proc_net_stat,
1430                               &neigh_stat_seq_fops, tbl))
1431                 panic("cannot create neighbour proc dir entry");
1432 #endif
1433
1434         tbl->hash_mask = 1;
1435         tbl->hash_buckets = neigh_hash_alloc(tbl->hash_mask + 1);
1436
1437         phsize = (PNEIGH_HASHMASK + 1) * sizeof(struct pneigh_entry *);
1438         tbl->phash_buckets = kzalloc(phsize, GFP_KERNEL);
1439
1440         if (!tbl->hash_buckets || !tbl->phash_buckets)
1441                 panic("cannot allocate neighbour cache hashes");
1442
1443         get_random_bytes(&tbl->hash_rnd, sizeof(tbl->hash_rnd));
1444
1445         rwlock_init(&tbl->lock);
1446         INIT_DELAYED_WORK_DEFERRABLE(&tbl->gc_work, neigh_periodic_work);
1447         schedule_delayed_work(&tbl->gc_work, tbl->parms.reachable_time);
1448         setup_timer(&tbl->proxy_timer, neigh_proxy_process, (unsigned long)tbl);
1449         skb_queue_head_init_class(&tbl->proxy_queue,
1450                         &neigh_table_proxy_queue_class);
1451
1452         tbl->last_flush = now;
1453         tbl->last_rand  = now + tbl->parms.reachable_time * 20;
1454 }
1455 EXPORT_SYMBOL(neigh_table_init_no_netlink);
1456
1457 void neigh_table_init(struct neigh_table *tbl)
1458 {
1459         struct neigh_table *tmp;
1460
1461         neigh_table_init_no_netlink(tbl);
1462         write_lock(&neigh_tbl_lock);
1463         for (tmp = neigh_tables; tmp; tmp = tmp->next) {
1464                 if (tmp->family == tbl->family)
1465                         break;
1466         }
1467         tbl->next       = neigh_tables;
1468         neigh_tables    = tbl;
1469         write_unlock(&neigh_tbl_lock);
1470
1471         if (unlikely(tmp)) {
1472                 printk(KERN_ERR "NEIGH: Registering multiple tables for "
1473                        "family %d\n", tbl->family);
1474                 dump_stack();
1475         }
1476 }
1477 EXPORT_SYMBOL(neigh_table_init);
1478
1479 int neigh_table_clear(struct neigh_table *tbl)
1480 {
1481         struct neigh_table **tp;
1482
1483         /* It is not clean... Fix it to unload IPv6 module safely */
1484         cancel_delayed_work(&tbl->gc_work);
1485         flush_scheduled_work();
1486         del_timer_sync(&tbl->proxy_timer);
1487         pneigh_queue_purge(&tbl->proxy_queue);
1488         neigh_ifdown(tbl, NULL);
1489         if (atomic_read(&tbl->entries))
1490                 printk(KERN_CRIT "neighbour leakage\n");
1491         write_lock(&neigh_tbl_lock);
1492         for (tp = &neigh_tables; *tp; tp = &(*tp)->next) {
1493                 if (*tp == tbl) {
1494                         *tp = tbl->next;
1495                         break;
1496                 }
1497         }
1498         write_unlock(&neigh_tbl_lock);
1499
1500         neigh_hash_free(tbl->hash_buckets, tbl->hash_mask + 1);
1501         tbl->hash_buckets = NULL;
1502
1503         kfree(tbl->phash_buckets);
1504         tbl->phash_buckets = NULL;
1505
1506         remove_proc_entry(tbl->id, init_net.proc_net_stat);
1507
1508         free_percpu(tbl->stats);
1509         tbl->stats = NULL;
1510
1511         kmem_cache_destroy(tbl->kmem_cachep);
1512         tbl->kmem_cachep = NULL;
1513
1514         return 0;
1515 }
1516 EXPORT_SYMBOL(neigh_table_clear);
1517
1518 static int neigh_delete(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
1519 {
1520         struct net *net = sock_net(skb->sk);
1521         struct ndmsg *ndm;
1522         struct nlattr *dst_attr;
1523         struct neigh_table *tbl;
1524         struct net_device *dev = NULL;
1525         int err = -EINVAL;
1526
1527         if (nlmsg_len(nlh) < sizeof(*ndm))
1528                 goto out;
1529
1530         dst_attr = nlmsg_find_attr(nlh, sizeof(*ndm), NDA_DST);
1531         if (dst_attr == NULL)
1532                 goto out;
1533
1534         ndm = nlmsg_data(nlh);
1535         if (ndm->ndm_ifindex) {
1536                 dev = dev_get_by_index(net, ndm->ndm_ifindex);
1537                 if (dev == NULL) {
1538                         err = -ENODEV;
1539                         goto out;
1540                 }
1541         }
1542
1543         read_lock(&neigh_tbl_lock);
1544         for (tbl = neigh_tables; tbl; tbl = tbl->next) {
1545                 struct neighbour *neigh;
1546
1547                 if (tbl->family != ndm->ndm_family)
1548                         continue;
1549                 read_unlock(&neigh_tbl_lock);
1550
1551                 if (nla_len(dst_attr) < tbl->key_len)
1552                         goto out_dev_put;
1553
1554                 if (ndm->ndm_flags & NTF_PROXY) {
1555                         err = pneigh_delete(tbl, net, nla_data(dst_attr), dev);
1556                         goto out_dev_put;
1557                 }
1558
1559                 if (dev == NULL)
1560                         goto out_dev_put;
1561
1562                 neigh = neigh_lookup(tbl, nla_data(dst_attr), dev);
1563                 if (neigh == NULL) {
1564                         err = -ENOENT;
1565                         goto out_dev_put;
1566                 }
1567
1568                 err = neigh_update(neigh, NULL, NUD_FAILED,
1569                                    NEIGH_UPDATE_F_OVERRIDE |
1570                                    NEIGH_UPDATE_F_ADMIN);
1571                 neigh_release(neigh);
1572                 goto out_dev_put;
1573         }
1574         read_unlock(&neigh_tbl_lock);
1575         err = -EAFNOSUPPORT;
1576
1577 out_dev_put:
1578         if (dev)
1579                 dev_put(dev);
1580 out:
1581         return err;
1582 }
1583
1584 static int neigh_add(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
1585 {
1586         struct net *net = sock_net(skb->sk);
1587         struct ndmsg *ndm;
1588         struct nlattr *tb[NDA_MAX+1];
1589         struct neigh_table *tbl;
1590         struct net_device *dev = NULL;
1591         int err;
1592
1593         err = nlmsg_parse(nlh, sizeof(*ndm), tb, NDA_MAX, NULL);
1594         if (err < 0)
1595                 goto out;
1596
1597         err = -EINVAL;
1598         if (tb[NDA_DST] == NULL)
1599                 goto out;
1600
1601         ndm = nlmsg_data(nlh);
1602         if (ndm->ndm_ifindex) {
1603                 dev = dev_get_by_index(net, ndm->ndm_ifindex);
1604                 if (dev == NULL) {
1605                         err = -ENODEV;
1606                         goto out;
1607                 }
1608
1609                 if (tb[NDA_LLADDR] && nla_len(tb[NDA_LLADDR]) < dev->addr_len)
1610                         goto out_dev_put;
1611         }
1612
1613         read_lock(&neigh_tbl_lock);
1614         for (tbl = neigh_tables; tbl; tbl = tbl->next) {
1615                 int flags = NEIGH_UPDATE_F_ADMIN | NEIGH_UPDATE_F_OVERRIDE;
1616                 struct neighbour *neigh;
1617                 void *dst, *lladdr;
1618
1619                 if (tbl->family != ndm->ndm_family)
1620                         continue;
1621                 read_unlock(&neigh_tbl_lock);
1622
1623                 if (nla_len(tb[NDA_DST]) < tbl->key_len)
1624                         goto out_dev_put;
1625                 dst = nla_data(tb[NDA_DST]);
1626                 lladdr = tb[NDA_LLADDR] ? nla_data(tb[NDA_LLADDR]) : NULL;
1627
1628                 if (ndm->ndm_flags & NTF_PROXY) {
1629                         struct pneigh_entry *pn;
1630
1631                         err = -ENOBUFS;
1632                         pn = pneigh_lookup(tbl, net, dst, dev, 1);
1633                         if (pn) {
1634                                 pn->flags = ndm->ndm_flags;
1635                                 err = 0;
1636                         }
1637                         goto out_dev_put;
1638                 }
1639
1640                 if (dev == NULL)
1641                         goto out_dev_put;
1642
1643                 neigh = neigh_lookup(tbl, dst, dev);
1644                 if (neigh == NULL) {
1645                         if (!(nlh->nlmsg_flags & NLM_F_CREATE)) {
1646                                 err = -ENOENT;
1647                                 goto out_dev_put;
1648                         }
1649
1650                         neigh = __neigh_lookup_errno(tbl, dst, dev);
1651                         if (IS_ERR(neigh)) {
1652                                 err = PTR_ERR(neigh);
1653                                 goto out_dev_put;
1654                         }
1655                 } else {
1656                         if (nlh->nlmsg_flags & NLM_F_EXCL) {
1657                                 err = -EEXIST;
1658                                 neigh_release(neigh);
1659                                 goto out_dev_put;
1660                         }
1661
1662                         if (!(nlh->nlmsg_flags & NLM_F_REPLACE))
1663                                 flags &= ~NEIGH_UPDATE_F_OVERRIDE;
1664                 }
1665
1666                 if (ndm->ndm_flags & NTF_USE) {
1667                         neigh_event_send(neigh, NULL);
1668                         err = 0;
1669                 } else
1670                         err = neigh_update(neigh, lladdr, ndm->ndm_state, flags);
1671                 neigh_release(neigh);
1672                 goto out_dev_put;
1673         }
1674
1675         read_unlock(&neigh_tbl_lock);
1676         err = -EAFNOSUPPORT;
1677
1678 out_dev_put:
1679         if (dev)
1680                 dev_put(dev);
1681 out:
1682         return err;
1683 }
1684
1685 static int neightbl_fill_parms(struct sk_buff *skb, struct neigh_parms *parms)
1686 {
1687         struct nlattr *nest;
1688
1689         nest = nla_nest_start(skb, NDTA_PARMS);
1690         if (nest == NULL)
1691                 return -ENOBUFS;
1692
1693         if (parms->dev)
1694                 NLA_PUT_U32(skb, NDTPA_IFINDEX, parms->dev->ifindex);
1695
1696         NLA_PUT_U32(skb, NDTPA_REFCNT, atomic_read(&parms->refcnt));
1697         NLA_PUT_U32(skb, NDTPA_QUEUE_LEN, parms->queue_len);
1698         NLA_PUT_U32(skb, NDTPA_PROXY_QLEN, parms->proxy_qlen);
1699         NLA_PUT_U32(skb, NDTPA_APP_PROBES, parms->app_probes);
1700         NLA_PUT_U32(skb, NDTPA_UCAST_PROBES, parms->ucast_probes);
1701         NLA_PUT_U32(skb, NDTPA_MCAST_PROBES, parms->mcast_probes);
1702         NLA_PUT_MSECS(skb, NDTPA_REACHABLE_TIME, parms->reachable_time);
1703         NLA_PUT_MSECS(skb, NDTPA_BASE_REACHABLE_TIME,
1704                       parms->base_reachable_time);
1705         NLA_PUT_MSECS(skb, NDTPA_GC_STALETIME, parms->gc_staletime);
1706         NLA_PUT_MSECS(skb, NDTPA_DELAY_PROBE_TIME, parms->delay_probe_time);
1707         NLA_PUT_MSECS(skb, NDTPA_RETRANS_TIME, parms->retrans_time);
1708         NLA_PUT_MSECS(skb, NDTPA_ANYCAST_DELAY, parms->anycast_delay);
1709         NLA_PUT_MSECS(skb, NDTPA_PROXY_DELAY, parms->proxy_delay);
1710         NLA_PUT_MSECS(skb, NDTPA_LOCKTIME, parms->locktime);
1711
1712         return nla_nest_end(skb, nest);
1713
1714 nla_put_failure:
1715         nla_nest_cancel(skb, nest);
1716         return -EMSGSIZE;
1717 }
1718
1719 static int neightbl_fill_info(struct sk_buff *skb, struct neigh_table *tbl,
1720                               u32 pid, u32 seq, int type, int flags)
1721 {
1722         struct nlmsghdr *nlh;
1723         struct ndtmsg *ndtmsg;
1724
1725         nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
1726         if (nlh == NULL)
1727                 return -EMSGSIZE;
1728
1729         ndtmsg = nlmsg_data(nlh);
1730
1731         read_lock_bh(&tbl->lock);
1732         ndtmsg->ndtm_family = tbl->family;
1733         ndtmsg->ndtm_pad1   = 0;
1734         ndtmsg->ndtm_pad2   = 0;
1735
1736         NLA_PUT_STRING(skb, NDTA_NAME, tbl->id);
1737         NLA_PUT_MSECS(skb, NDTA_GC_INTERVAL, tbl->gc_interval);
1738         NLA_PUT_U32(skb, NDTA_THRESH1, tbl->gc_thresh1);
1739         NLA_PUT_U32(skb, NDTA_THRESH2, tbl->gc_thresh2);
1740         NLA_PUT_U32(skb, NDTA_THRESH3, tbl->gc_thresh3);
1741
1742         {
1743                 unsigned long now = jiffies;
1744                 unsigned int flush_delta = now - tbl->last_flush;
1745                 unsigned int rand_delta = now - tbl->last_rand;
1746
1747                 struct ndt_config ndc = {
1748                         .ndtc_key_len           = tbl->key_len,
1749                         .ndtc_entry_size        = tbl->entry_size,
1750                         .ndtc_entries           = atomic_read(&tbl->entries),
1751                         .ndtc_last_flush        = jiffies_to_msecs(flush_delta),
1752                         .ndtc_last_rand         = jiffies_to_msecs(rand_delta),
1753                         .ndtc_hash_rnd          = tbl->hash_rnd,
1754                         .ndtc_hash_mask         = tbl->hash_mask,
1755                         .ndtc_proxy_qlen        = tbl->proxy_queue.qlen,
1756                 };
1757
1758                 NLA_PUT(skb, NDTA_CONFIG, sizeof(ndc), &ndc);
1759         }
1760
1761         {
1762                 int cpu;
1763                 struct ndt_stats ndst;
1764
1765                 memset(&ndst, 0, sizeof(ndst));
1766
1767                 for_each_possible_cpu(cpu) {
1768                         struct neigh_statistics *st;
1769
1770                         st = per_cpu_ptr(tbl->stats, cpu);
1771                         ndst.ndts_allocs                += st->allocs;
1772                         ndst.ndts_destroys              += st->destroys;
1773                         ndst.ndts_hash_grows            += st->hash_grows;
1774                         ndst.ndts_res_failed            += st->res_failed;
1775                         ndst.ndts_lookups               += st->lookups;
1776                         ndst.ndts_hits                  += st->hits;
1777                         ndst.ndts_rcv_probes_mcast      += st->rcv_probes_mcast;
1778                         ndst.ndts_rcv_probes_ucast      += st->rcv_probes_ucast;
1779                         ndst.ndts_periodic_gc_runs      += st->periodic_gc_runs;
1780                         ndst.ndts_forced_gc_runs        += st->forced_gc_runs;
1781                 }
1782
1783                 NLA_PUT(skb, NDTA_STATS, sizeof(ndst), &ndst);
1784         }
1785
1786         BUG_ON(tbl->parms.dev);
1787         if (neightbl_fill_parms(skb, &tbl->parms) < 0)
1788                 goto nla_put_failure;
1789
1790         read_unlock_bh(&tbl->lock);
1791         return nlmsg_end(skb, nlh);
1792
1793 nla_put_failure:
1794         read_unlock_bh(&tbl->lock);
1795         nlmsg_cancel(skb, nlh);
1796         return -EMSGSIZE;
1797 }
1798
1799 static int neightbl_fill_param_info(struct sk_buff *skb,
1800                                     struct neigh_table *tbl,
1801                                     struct neigh_parms *parms,
1802                                     u32 pid, u32 seq, int type,
1803                                     unsigned int flags)
1804 {
1805         struct ndtmsg *ndtmsg;
1806         struct nlmsghdr *nlh;
1807
1808         nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
1809         if (nlh == NULL)
1810                 return -EMSGSIZE;
1811
1812         ndtmsg = nlmsg_data(nlh);
1813
1814         read_lock_bh(&tbl->lock);
1815         ndtmsg->ndtm_family = tbl->family;
1816         ndtmsg->ndtm_pad1   = 0;
1817         ndtmsg->ndtm_pad2   = 0;
1818
1819         if (nla_put_string(skb, NDTA_NAME, tbl->id) < 0 ||
1820             neightbl_fill_parms(skb, parms) < 0)
1821                 goto errout;
1822
1823         read_unlock_bh(&tbl->lock);
1824         return nlmsg_end(skb, nlh);
1825 errout:
1826         read_unlock_bh(&tbl->lock);
1827         nlmsg_cancel(skb, nlh);
1828         return -EMSGSIZE;
1829 }
1830
1831 static const struct nla_policy nl_neightbl_policy[NDTA_MAX+1] = {
1832         [NDTA_NAME]             = { .type = NLA_STRING },
1833         [NDTA_THRESH1]          = { .type = NLA_U32 },
1834         [NDTA_THRESH2]          = { .type = NLA_U32 },
1835         [NDTA_THRESH3]          = { .type = NLA_U32 },
1836         [NDTA_GC_INTERVAL]      = { .type = NLA_U64 },
1837         [NDTA_PARMS]            = { .type = NLA_NESTED },
1838 };
1839
1840 static const struct nla_policy nl_ntbl_parm_policy[NDTPA_MAX+1] = {
1841         [NDTPA_IFINDEX]                 = { .type = NLA_U32 },
1842         [NDTPA_QUEUE_LEN]               = { .type = NLA_U32 },
1843         [NDTPA_PROXY_QLEN]              = { .type = NLA_U32 },
1844         [NDTPA_APP_PROBES]              = { .type = NLA_U32 },
1845         [NDTPA_UCAST_PROBES]            = { .type = NLA_U32 },
1846         [NDTPA_MCAST_PROBES]            = { .type = NLA_U32 },
1847         [NDTPA_BASE_REACHABLE_TIME]     = { .type = NLA_U64 },
1848         [NDTPA_GC_STALETIME]            = { .type = NLA_U64 },
1849         [NDTPA_DELAY_PROBE_TIME]        = { .type = NLA_U64 },
1850         [NDTPA_RETRANS_TIME]            = { .type = NLA_U64 },
1851         [NDTPA_ANYCAST_DELAY]           = { .type = NLA_U64 },
1852         [NDTPA_PROXY_DELAY]             = { .type = NLA_U64 },
1853         [NDTPA_LOCKTIME]                = { .type = NLA_U64 },
1854 };
1855
1856 static int neightbl_set(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
1857 {
1858         struct net *net = sock_net(skb->sk);
1859         struct neigh_table *tbl;
1860         struct ndtmsg *ndtmsg;
1861         struct nlattr *tb[NDTA_MAX+1];
1862         int err;
1863
1864         err = nlmsg_parse(nlh, sizeof(*ndtmsg), tb, NDTA_MAX,
1865                           nl_neightbl_policy);
1866         if (err < 0)
1867                 goto errout;
1868
1869         if (tb[NDTA_NAME] == NULL) {
1870                 err = -EINVAL;
1871                 goto errout;
1872         }
1873
1874         ndtmsg = nlmsg_data(nlh);
1875         read_lock(&neigh_tbl_lock);
1876         for (tbl = neigh_tables; tbl; tbl = tbl->next) {
1877                 if (ndtmsg->ndtm_family && tbl->family != ndtmsg->ndtm_family)
1878                         continue;
1879
1880                 if (nla_strcmp(tb[NDTA_NAME], tbl->id) == 0)
1881                         break;
1882         }
1883
1884         if (tbl == NULL) {
1885                 err = -ENOENT;
1886                 goto errout_locked;
1887         }
1888
1889         /*
1890          * We acquire tbl->lock to be nice to the periodic timers and
1891          * make sure they always see a consistent set of values.
1892          */
1893         write_lock_bh(&tbl->lock);
1894
1895         if (tb[NDTA_PARMS]) {
1896                 struct nlattr *tbp[NDTPA_MAX+1];
1897                 struct neigh_parms *p;
1898                 int i, ifindex = 0;
1899
1900                 err = nla_parse_nested(tbp, NDTPA_MAX, tb[NDTA_PARMS],
1901                                        nl_ntbl_parm_policy);
1902                 if (err < 0)
1903                         goto errout_tbl_lock;
1904
1905                 if (tbp[NDTPA_IFINDEX])
1906                         ifindex = nla_get_u32(tbp[NDTPA_IFINDEX]);
1907
1908                 p = lookup_neigh_parms(tbl, net, ifindex);
1909                 if (p == NULL) {
1910                         err = -ENOENT;
1911                         goto errout_tbl_lock;
1912                 }
1913
1914                 for (i = 1; i <= NDTPA_MAX; i++) {
1915                         if (tbp[i] == NULL)
1916                                 continue;
1917
1918                         switch (i) {
1919                         case NDTPA_QUEUE_LEN:
1920                                 p->queue_len = nla_get_u32(tbp[i]);
1921                                 break;
1922                         case NDTPA_PROXY_QLEN:
1923                                 p->proxy_qlen = nla_get_u32(tbp[i]);
1924                                 break;
1925                         case NDTPA_APP_PROBES:
1926                                 p->app_probes = nla_get_u32(tbp[i]);
1927                                 break;
1928                         case NDTPA_UCAST_PROBES:
1929                                 p->ucast_probes = nla_get_u32(tbp[i]);
1930                                 break;
1931                         case NDTPA_MCAST_PROBES:
1932                                 p->mcast_probes = nla_get_u32(tbp[i]);
1933                                 break;
1934                         case NDTPA_BASE_REACHABLE_TIME:
1935                                 p->base_reachable_time = nla_get_msecs(tbp[i]);
1936                                 break;
1937                         case NDTPA_GC_STALETIME:
1938                                 p->gc_staletime = nla_get_msecs(tbp[i]);
1939                                 break;
1940                         case NDTPA_DELAY_PROBE_TIME:
1941                                 p->delay_probe_time = nla_get_msecs(tbp[i]);
1942                                 break;
1943                         case NDTPA_RETRANS_TIME:
1944                                 p->retrans_time = nla_get_msecs(tbp[i]);
1945                                 break;
1946                         case NDTPA_ANYCAST_DELAY:
1947                                 p->anycast_delay = nla_get_msecs(tbp[i]);
1948                                 break;
1949                         case NDTPA_PROXY_DELAY:
1950                                 p->proxy_delay = nla_get_msecs(tbp[i]);
1951                                 break;
1952                         case NDTPA_LOCKTIME:
1953                                 p->locktime = nla_get_msecs(tbp[i]);
1954                                 break;
1955                         }
1956                 }
1957         }
1958
1959         if (tb[NDTA_THRESH1])
1960                 tbl->gc_thresh1 = nla_get_u32(tb[NDTA_THRESH1]);
1961
1962         if (tb[NDTA_THRESH2])
1963                 tbl->gc_thresh2 = nla_get_u32(tb[NDTA_THRESH2]);
1964
1965         if (tb[NDTA_THRESH3])
1966                 tbl->gc_thresh3 = nla_get_u32(tb[NDTA_THRESH3]);
1967
1968         if (tb[NDTA_GC_INTERVAL])
1969                 tbl->gc_interval = nla_get_msecs(tb[NDTA_GC_INTERVAL]);
1970
1971         err = 0;
1972
1973 errout_tbl_lock:
1974         write_unlock_bh(&tbl->lock);
1975 errout_locked:
1976         read_unlock(&neigh_tbl_lock);
1977 errout:
1978         return err;
1979 }
1980
1981 static int neightbl_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
1982 {
1983         struct net *net = sock_net(skb->sk);
1984         int family, tidx, nidx = 0;
1985         int tbl_skip = cb->args[0];
1986         int neigh_skip = cb->args[1];
1987         struct neigh_table *tbl;
1988
1989         family = ((struct rtgenmsg *) nlmsg_data(cb->nlh))->rtgen_family;
1990
1991         read_lock(&neigh_tbl_lock);
1992         for (tbl = neigh_tables, tidx = 0; tbl; tbl = tbl->next, tidx++) {
1993                 struct neigh_parms *p;
1994
1995                 if (tidx < tbl_skip || (family && tbl->family != family))
1996                         continue;
1997
1998                 if (neightbl_fill_info(skb, tbl, NETLINK_CB(cb->skb).pid,
1999                                        cb->nlh->nlmsg_seq, RTM_NEWNEIGHTBL,
2000                                        NLM_F_MULTI) <= 0)
2001                         break;
2002
2003                 for (nidx = 0, p = tbl->parms.next; p; p = p->next) {
2004                         if (!net_eq(neigh_parms_net(p), net))
2005                                 continue;
2006
2007                         if (nidx < neigh_skip)
2008                                 goto next;
2009
2010                         if (neightbl_fill_param_info(skb, tbl, p,
2011                                                      NETLINK_CB(cb->skb).pid,
2012                                                      cb->nlh->nlmsg_seq,
2013                                                      RTM_NEWNEIGHTBL,
2014                                                      NLM_F_MULTI) <= 0)
2015                                 goto out;
2016                 next:
2017                         nidx++;
2018                 }
2019
2020                 neigh_skip = 0;
2021         }
2022 out:
2023         read_unlock(&neigh_tbl_lock);
2024         cb->args[0] = tidx;
2025         cb->args[1] = nidx;
2026
2027         return skb->len;
2028 }
2029
2030 static int neigh_fill_info(struct sk_buff *skb, struct neighbour *neigh,
2031                            u32 pid, u32 seq, int type, unsigned int flags)
2032 {
2033         unsigned long now = jiffies;
2034         struct nda_cacheinfo ci;
2035         struct nlmsghdr *nlh;
2036         struct ndmsg *ndm;
2037
2038         nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2039         if (nlh == NULL)
2040                 return -EMSGSIZE;
2041
2042         ndm = nlmsg_data(nlh);
2043         ndm->ndm_family  = neigh->ops->family;
2044         ndm->ndm_pad1    = 0;
2045         ndm->ndm_pad2    = 0;
2046         ndm->ndm_flags   = neigh->flags;
2047         ndm->ndm_type    = neigh->type;
2048         ndm->ndm_ifindex = neigh->dev->ifindex;
2049
2050         NLA_PUT(skb, NDA_DST, neigh->tbl->key_len, neigh->primary_key);
2051
2052         read_lock_bh(&neigh->lock);
2053         ndm->ndm_state   = neigh->nud_state;
2054         if ((neigh->nud_state & NUD_VALID) &&
2055             nla_put(skb, NDA_LLADDR, neigh->dev->addr_len, neigh->ha) < 0) {
2056                 read_unlock_bh(&neigh->lock);
2057                 goto nla_put_failure;
2058         }
2059
2060         ci.ndm_used      = jiffies_to_clock_t(now - neigh->used);
2061         ci.ndm_confirmed = jiffies_to_clock_t(now - neigh->confirmed);
2062         ci.ndm_updated   = jiffies_to_clock_t(now - neigh->updated);
2063         ci.ndm_refcnt    = atomic_read(&neigh->refcnt) - 1;
2064         read_unlock_bh(&neigh->lock);
2065
2066         NLA_PUT_U32(skb, NDA_PROBES, atomic_read(&neigh->probes));
2067         NLA_PUT(skb, NDA_CACHEINFO, sizeof(ci), &ci);
2068
2069         return nlmsg_end(skb, nlh);
2070
2071 nla_put_failure:
2072         nlmsg_cancel(skb, nlh);
2073         return -EMSGSIZE;
2074 }
2075
2076 static void neigh_update_notify(struct neighbour *neigh)
2077 {
2078         call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, neigh);
2079         __neigh_notify(neigh, RTM_NEWNEIGH, 0);
2080 }
2081
2082 static int neigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2083                             struct netlink_callback *cb)
2084 {
2085         struct net * net = sock_net(skb->sk);
2086         struct neighbour *n;
2087         int rc, h, s_h = cb->args[1];
2088         int idx, s_idx = idx = cb->args[2];
2089
2090         read_lock_bh(&tbl->lock);
2091         for (h = 0; h <= tbl->hash_mask; h++) {
2092                 if (h < s_h)
2093                         continue;
2094                 if (h > s_h)
2095                         s_idx = 0;
2096                 for (n = tbl->hash_buckets[h], idx = 0; n; n = n->next) {
2097                         if (!net_eq(dev_net(n->dev), net))
2098                                 continue;
2099                         if (idx < s_idx)
2100                                 goto next;
2101                         if (neigh_fill_info(skb, n, NETLINK_CB(cb->skb).pid,
2102                                             cb->nlh->nlmsg_seq,
2103                                             RTM_NEWNEIGH,
2104                                             NLM_F_MULTI) <= 0) {
2105                                 read_unlock_bh(&tbl->lock);
2106                                 rc = -1;
2107                                 goto out;
2108                         }
2109                 next:
2110                         idx++;
2111                 }
2112         }
2113         read_unlock_bh(&tbl->lock);
2114         rc = skb->len;
2115 out:
2116         cb->args[1] = h;
2117         cb->args[2] = idx;
2118         return rc;
2119 }
2120
2121 static int neigh_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2122 {
2123         struct neigh_table *tbl;
2124         int t, family, s_t;
2125
2126         read_lock(&neigh_tbl_lock);
2127         family = ((struct rtgenmsg *) nlmsg_data(cb->nlh))->rtgen_family;
2128         s_t = cb->args[0];
2129
2130         for (tbl = neigh_tables, t = 0; tbl; tbl = tbl->next, t++) {
2131                 if (t < s_t || (family && tbl->family != family))
2132                         continue;
2133                 if (t > s_t)
2134                         memset(&cb->args[1], 0, sizeof(cb->args) -
2135                                                 sizeof(cb->args[0]));
2136                 if (neigh_dump_table(tbl, skb, cb) < 0)
2137                         break;
2138         }
2139         read_unlock(&neigh_tbl_lock);
2140
2141         cb->args[0] = t;
2142         return skb->len;
2143 }
2144
2145 void neigh_for_each(struct neigh_table *tbl, void (*cb)(struct neighbour *, void *), void *cookie)
2146 {
2147         int chain;
2148
2149         read_lock_bh(&tbl->lock);
2150         for (chain = 0; chain <= tbl->hash_mask; chain++) {
2151                 struct neighbour *n;
2152
2153                 for (n = tbl->hash_buckets[chain]; n; n = n->next)
2154                         cb(n, cookie);
2155         }
2156         read_unlock_bh(&tbl->lock);
2157 }
2158 EXPORT_SYMBOL(neigh_for_each);
2159
2160 /* The tbl->lock must be held as a writer and BH disabled. */
2161 void __neigh_for_each_release(struct neigh_table *tbl,
2162                               int (*cb)(struct neighbour *))
2163 {
2164         int chain;
2165
2166         for (chain = 0; chain <= tbl->hash_mask; chain++) {
2167                 struct neighbour *n, **np;
2168
2169                 np = &tbl->hash_buckets[chain];
2170                 while ((n = *np) != NULL) {
2171                         int release;
2172
2173                         write_lock(&n->lock);
2174                         release = cb(n);
2175                         if (release) {
2176                                 *np = n->next;
2177                                 n->dead = 1;
2178                         } else
2179                                 np = &n->next;
2180                         write_unlock(&n->lock);
2181                         if (release)
2182                                 neigh_cleanup_and_release(n);
2183                 }
2184         }
2185 }
2186 EXPORT_SYMBOL(__neigh_for_each_release);
2187
2188 #ifdef CONFIG_PROC_FS
2189
2190 static struct neighbour *neigh_get_first(struct seq_file *seq)
2191 {
2192         struct neigh_seq_state *state = seq->private;
2193         struct net *net = seq_file_net(seq);
2194         struct neigh_table *tbl = state->tbl;
2195         struct neighbour *n = NULL;
2196         int bucket = state->bucket;
2197
2198         state->flags &= ~NEIGH_SEQ_IS_PNEIGH;
2199         for (bucket = 0; bucket <= tbl->hash_mask; bucket++) {
2200                 n = tbl->hash_buckets[bucket];
2201
2202                 while (n) {
2203                         if (!net_eq(dev_net(n->dev), net))
2204                                 goto next;
2205                         if (state->neigh_sub_iter) {
2206                                 loff_t fakep = 0;
2207                                 void *v;
2208
2209                                 v = state->neigh_sub_iter(state, n, &fakep);
2210                                 if (!v)
2211                                         goto next;
2212                         }
2213                         if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
2214                                 break;
2215                         if (n->nud_state & ~NUD_NOARP)
2216                                 break;
2217                 next:
2218                         n = n->next;
2219                 }
2220
2221                 if (n)
2222                         break;
2223         }
2224         state->bucket = bucket;
2225
2226         return n;
2227 }
2228
2229 static struct neighbour *neigh_get_next(struct seq_file *seq,
2230                                         struct neighbour *n,
2231                                         loff_t *pos)
2232 {
2233         struct neigh_seq_state *state = seq->private;
2234         struct net *net = seq_file_net(seq);
2235         struct neigh_table *tbl = state->tbl;
2236
2237         if (state->neigh_sub_iter) {
2238                 void *v = state->neigh_sub_iter(state, n, pos);
2239                 if (v)
2240                         return n;
2241         }
2242         n = n->next;
2243
2244         while (1) {
2245                 while (n) {
2246                         if (!net_eq(dev_net(n->dev), net))
2247                                 goto next;
2248                         if (state->neigh_sub_iter) {
2249                                 void *v = state->neigh_sub_iter(state, n, pos);
2250                                 if (v)
2251                                         return n;
2252                                 goto next;
2253                         }
2254                         if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
2255                                 break;
2256
2257                         if (n->nud_state & ~NUD_NOARP)
2258                                 break;
2259                 next:
2260                         n = n->next;
2261                 }
2262
2263                 if (n)
2264                         break;
2265
2266                 if (++state->bucket > tbl->hash_mask)
2267                         break;
2268
2269                 n = tbl->hash_buckets[state->bucket];
2270         }
2271
2272         if (n && pos)
2273                 --(*pos);
2274         return n;
2275 }
2276
2277 static struct neighbour *neigh_get_idx(struct seq_file *seq, loff_t *pos)
2278 {
2279         struct neighbour *n = neigh_get_first(seq);
2280
2281         if (n) {
2282                 --(*pos);
2283                 while (*pos) {
2284                         n = neigh_get_next(seq, n, pos);
2285                         if (!n)
2286                                 break;
2287                 }
2288         }
2289         return *pos ? NULL : n;
2290 }
2291
2292 static struct pneigh_entry *pneigh_get_first(struct seq_file *seq)
2293 {
2294         struct neigh_seq_state *state = seq->private;
2295         struct net *net = seq_file_net(seq);
2296         struct neigh_table *tbl = state->tbl;
2297         struct pneigh_entry *pn = NULL;
2298         int bucket = state->bucket;
2299
2300         state->flags |= NEIGH_SEQ_IS_PNEIGH;
2301         for (bucket = 0; bucket <= PNEIGH_HASHMASK; bucket++) {
2302                 pn = tbl->phash_buckets[bucket];
2303                 while (pn && !net_eq(pneigh_net(pn), net))
2304                         pn = pn->next;
2305                 if (pn)
2306                         break;
2307         }
2308         state->bucket = bucket;
2309
2310         return pn;
2311 }
2312
2313 static struct pneigh_entry *pneigh_get_next(struct seq_file *seq,
2314                                             struct pneigh_entry *pn,
2315                                             loff_t *pos)
2316 {
2317         struct neigh_seq_state *state = seq->private;
2318         struct net *net = seq_file_net(seq);
2319         struct neigh_table *tbl = state->tbl;
2320
2321         pn = pn->next;
2322         while (!pn) {
2323                 if (++state->bucket > PNEIGH_HASHMASK)
2324                         break;
2325                 pn = tbl->phash_buckets[state->bucket];
2326                 while (pn && !net_eq(pneigh_net(pn), net))
2327                         pn = pn->next;
2328                 if (pn)
2329                         break;
2330         }
2331
2332         if (pn && pos)
2333                 --(*pos);
2334
2335         return pn;
2336 }
2337
2338 static struct pneigh_entry *pneigh_get_idx(struct seq_file *seq, loff_t *pos)
2339 {
2340         struct pneigh_entry *pn = pneigh_get_first(seq);
2341
2342         if (pn) {
2343                 --(*pos);
2344                 while (*pos) {
2345                         pn = pneigh_get_next(seq, pn, pos);
2346                         if (!pn)
2347                                 break;
2348                 }
2349         }
2350         return *pos ? NULL : pn;
2351 }
2352
2353 static void *neigh_get_idx_any(struct seq_file *seq, loff_t *pos)
2354 {
2355         struct neigh_seq_state *state = seq->private;
2356         void *rc;
2357         loff_t idxpos = *pos;
2358
2359         rc = neigh_get_idx(seq, &idxpos);
2360         if (!rc && !(state->flags & NEIGH_SEQ_NEIGH_ONLY))
2361                 rc = pneigh_get_idx(seq, &idxpos);
2362
2363         return rc;
2364 }
2365
2366 void *neigh_seq_start(struct seq_file *seq, loff_t *pos, struct neigh_table *tbl, unsigned int neigh_seq_flags)
2367         __acquires(tbl->lock)
2368 {
2369         struct neigh_seq_state *state = seq->private;
2370
2371         state->tbl = tbl;
2372         state->bucket = 0;
2373         state->flags = (neigh_seq_flags & ~NEIGH_SEQ_IS_PNEIGH);
2374
2375         read_lock_bh(&tbl->lock);
2376
2377         return *pos ? neigh_get_idx_any(seq, pos) : SEQ_START_TOKEN;
2378 }
2379 EXPORT_SYMBOL(neigh_seq_start);
2380
2381 void *neigh_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2382 {
2383         struct neigh_seq_state *state;
2384         void *rc;
2385
2386         if (v == SEQ_START_TOKEN) {
2387                 rc = neigh_get_first(seq);
2388                 goto out;
2389         }
2390
2391         state = seq->private;
2392         if (!(state->flags & NEIGH_SEQ_IS_PNEIGH)) {
2393                 rc = neigh_get_next(seq, v, NULL);
2394                 if (rc)
2395                         goto out;
2396                 if (!(state->flags & NEIGH_SEQ_NEIGH_ONLY))
2397                         rc = pneigh_get_first(seq);
2398         } else {
2399                 BUG_ON(state->flags & NEIGH_SEQ_NEIGH_ONLY);
2400                 rc = pneigh_get_next(seq, v, NULL);
2401         }
2402 out:
2403         ++(*pos);
2404         return rc;
2405 }
2406 EXPORT_SYMBOL(neigh_seq_next);
2407
2408 void neigh_seq_stop(struct seq_file *seq, void *v)
2409         __releases(tbl->lock)
2410 {
2411         struct neigh_seq_state *state = seq->private;
2412         struct neigh_table *tbl = state->tbl;
2413
2414         read_unlock_bh(&tbl->lock);
2415 }
2416 EXPORT_SYMBOL(neigh_seq_stop);
2417
2418 /* statistics via seq_file */
2419
2420 static void *neigh_stat_seq_start(struct seq_file *seq, loff_t *pos)
2421 {
2422         struct neigh_table *tbl = seq->private;
2423         int cpu;
2424
2425         if (*pos == 0)
2426                 return SEQ_START_TOKEN;
2427
2428         for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
2429                 if (!cpu_possible(cpu))
2430                         continue;
2431                 *pos = cpu+1;
2432                 return per_cpu_ptr(tbl->stats, cpu);
2433         }
2434         return NULL;
2435 }
2436
2437 static void *neigh_stat_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2438 {
2439         struct neigh_table *tbl = seq->private;
2440         int cpu;
2441
2442         for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
2443                 if (!cpu_possible(cpu))
2444                         continue;
2445                 *pos = cpu+1;
2446                 return per_cpu_ptr(tbl->stats, cpu);
2447         }
2448         return NULL;
2449 }
2450
2451 static void neigh_stat_seq_stop(struct seq_file *seq, void *v)
2452 {
2453
2454 }
2455
2456 static int neigh_stat_seq_show(struct seq_file *seq, void *v)
2457 {
2458         struct neigh_table *tbl = seq->private;
2459         struct neigh_statistics *st = v;
2460
2461         if (v == SEQ_START_TOKEN) {
2462                 seq_printf(seq, "entries  allocs destroys hash_grows  lookups hits  res_failed  rcv_probes_mcast rcv_probes_ucast  periodic_gc_runs forced_gc_runs unresolved_discards\n");
2463                 return 0;
2464         }
2465
2466         seq_printf(seq, "%08x  %08lx %08lx %08lx  %08lx %08lx  %08lx  "
2467                         "%08lx %08lx  %08lx %08lx %08lx\n",
2468                    atomic_read(&tbl->entries),
2469
2470                    st->allocs,
2471                    st->destroys,
2472                    st->hash_grows,
2473
2474                    st->lookups,
2475                    st->hits,
2476
2477                    st->res_failed,
2478
2479                    st->rcv_probes_mcast,
2480                    st->rcv_probes_ucast,
2481
2482                    st->periodic_gc_runs,
2483                    st->forced_gc_runs,
2484                    st->unres_discards
2485                    );
2486
2487         return 0;
2488 }
2489
2490 static const struct seq_operations neigh_stat_seq_ops = {
2491         .start  = neigh_stat_seq_start,
2492         .next   = neigh_stat_seq_next,
2493         .stop   = neigh_stat_seq_stop,
2494         .show   = neigh_stat_seq_show,
2495 };
2496
2497 static int neigh_stat_seq_open(struct inode *inode, struct file *file)
2498 {
2499         int ret = seq_open(file, &neigh_stat_seq_ops);
2500
2501         if (!ret) {
2502                 struct seq_file *sf = file->private_data;
2503                 sf->private = PDE(inode)->data;
2504         }
2505         return ret;
2506 };
2507
2508 static const struct file_operations neigh_stat_seq_fops = {
2509         .owner   = THIS_MODULE,
2510         .open    = neigh_stat_seq_open,
2511         .read    = seq_read,
2512         .llseek  = seq_lseek,
2513         .release = seq_release,
2514 };
2515
2516 #endif /* CONFIG_PROC_FS */
2517
2518 static inline size_t neigh_nlmsg_size(void)
2519 {
2520         return NLMSG_ALIGN(sizeof(struct ndmsg))
2521                + nla_total_size(MAX_ADDR_LEN) /* NDA_DST */
2522                + nla_total_size(MAX_ADDR_LEN) /* NDA_LLADDR */
2523                + nla_total_size(sizeof(struct nda_cacheinfo))
2524                + nla_total_size(4); /* NDA_PROBES */
2525 }
2526
2527 static void __neigh_notify(struct neighbour *n, int type, int flags)
2528 {
2529         struct net *net = dev_net(n->dev);
2530         struct sk_buff *skb;
2531         int err = -ENOBUFS;
2532
2533         skb = nlmsg_new(neigh_nlmsg_size(), GFP_ATOMIC);
2534         if (skb == NULL)
2535                 goto errout;
2536
2537         err = neigh_fill_info(skb, n, 0, 0, type, flags);
2538         if (err < 0) {
2539                 /* -EMSGSIZE implies BUG in neigh_nlmsg_size() */
2540                 WARN_ON(err == -EMSGSIZE);
2541                 kfree_skb(skb);
2542                 goto errout;
2543         }
2544         rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
2545         return;
2546 errout:
2547         if (err < 0)
2548                 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
2549 }
2550
2551 #ifdef CONFIG_ARPD
2552 void neigh_app_ns(struct neighbour *n)
2553 {
2554         __neigh_notify(n, RTM_GETNEIGH, NLM_F_REQUEST);
2555 }
2556 EXPORT_SYMBOL(neigh_app_ns);
2557 #endif /* CONFIG_ARPD */
2558
2559 #ifdef CONFIG_SYSCTL
2560
2561 #define NEIGH_VARS_MAX 19
2562
2563 static struct neigh_sysctl_table {
2564         struct ctl_table_header *sysctl_header;
2565         struct ctl_table neigh_vars[NEIGH_VARS_MAX];
2566         char *dev_name;
2567 } neigh_sysctl_template __read_mostly = {
2568         .neigh_vars = {
2569                 {
2570                         .procname       = "mcast_solicit",
2571                         .maxlen         = sizeof(int),
2572                         .mode           = 0644,
2573                         .proc_handler   = proc_dointvec,
2574                 },
2575                 {
2576                         .procname       = "ucast_solicit",
2577                         .maxlen         = sizeof(int),
2578                         .mode           = 0644,
2579                         .proc_handler   = proc_dointvec,
2580                 },
2581                 {
2582                         .procname       = "app_solicit",
2583                         .maxlen         = sizeof(int),
2584                         .mode           = 0644,
2585                         .proc_handler   = proc_dointvec,
2586                 },
2587                 {
2588                         .procname       = "retrans_time",
2589                         .maxlen         = sizeof(int),
2590                         .mode           = 0644,
2591                         .proc_handler   = proc_dointvec_userhz_jiffies,
2592                 },
2593                 {
2594                         .procname       = "base_reachable_time",
2595                         .maxlen         = sizeof(int),
2596                         .mode           = 0644,
2597                         .proc_handler   = proc_dointvec_jiffies,
2598                 },
2599                 {
2600                         .procname       = "delay_first_probe_time",
2601                         .maxlen         = sizeof(int),
2602                         .mode           = 0644,
2603                         .proc_handler   = proc_dointvec_jiffies,
2604                 },
2605                 {
2606                         .procname       = "gc_stale_time",
2607                         .maxlen         = sizeof(int),
2608                         .mode           = 0644,
2609                         .proc_handler   = proc_dointvec_jiffies,
2610                 },
2611                 {
2612                         .procname       = "unres_qlen",
2613                         .maxlen         = sizeof(int),
2614                         .mode           = 0644,
2615                         .proc_handler   = proc_dointvec,
2616                 },
2617                 {
2618                         .procname       = "proxy_qlen",
2619                         .maxlen         = sizeof(int),
2620                         .mode           = 0644,
2621                         .proc_handler   = proc_dointvec,
2622                 },
2623                 {
2624                         .procname       = "anycast_delay",
2625                         .maxlen         = sizeof(int),
2626                         .mode           = 0644,
2627                         .proc_handler   = proc_dointvec_userhz_jiffies,
2628                 },
2629                 {
2630                         .procname       = "proxy_delay",
2631                         .maxlen         = sizeof(int),
2632                         .mode           = 0644,
2633                         .proc_handler   = proc_dointvec_userhz_jiffies,
2634                 },
2635                 {
2636                         .procname       = "locktime",
2637                         .maxlen         = sizeof(int),
2638                         .mode           = 0644,
2639                         .proc_handler   = proc_dointvec_userhz_jiffies,
2640                 },
2641                 {
2642                         .procname       = "retrans_time_ms",
2643                         .maxlen         = sizeof(int),
2644                         .mode           = 0644,
2645                         .proc_handler   = proc_dointvec_ms_jiffies,
2646                 },
2647                 {
2648                         .procname       = "base_reachable_time_ms",
2649                         .maxlen         = sizeof(int),
2650                         .mode           = 0644,
2651                         .proc_handler   = proc_dointvec_ms_jiffies,
2652                 },
2653                 {
2654                         .procname       = "gc_interval",
2655                         .maxlen         = sizeof(int),
2656                         .mode           = 0644,
2657                         .proc_handler   = proc_dointvec_jiffies,
2658                 },
2659                 {
2660                         .procname       = "gc_thresh1",
2661                         .maxlen         = sizeof(int),
2662                         .mode           = 0644,
2663                         .proc_handler   = proc_dointvec,
2664                 },
2665                 {
2666                         .procname       = "gc_thresh2",
2667                         .maxlen         = sizeof(int),
2668                         .mode           = 0644,
2669                         .proc_handler   = proc_dointvec,
2670                 },
2671                 {
2672                         .procname       = "gc_thresh3",
2673                         .maxlen         = sizeof(int),
2674                         .mode           = 0644,
2675                         .proc_handler   = proc_dointvec,
2676                 },
2677                 {},
2678         },
2679 };
2680
2681 int neigh_sysctl_register(struct net_device *dev, struct neigh_parms *p,
2682                           char *p_name, proc_handler *handler)
2683 {
2684         struct neigh_sysctl_table *t;
2685         const char *dev_name_source = NULL;
2686
2687 #define NEIGH_CTL_PATH_ROOT     0
2688 #define NEIGH_CTL_PATH_PROTO    1
2689 #define NEIGH_CTL_PATH_NEIGH    2
2690 #define NEIGH_CTL_PATH_DEV      3
2691
2692         struct ctl_path neigh_path[] = {
2693                 { .procname = "net",     },
2694                 { .procname = "proto",   },
2695                 { .procname = "neigh",   },
2696                 { .procname = "default", },
2697                 { },
2698         };
2699
2700         t = kmemdup(&neigh_sysctl_template, sizeof(*t), GFP_KERNEL);
2701         if (!t)
2702                 goto err;
2703
2704         t->neigh_vars[0].data  = &p->mcast_probes;
2705         t->neigh_vars[1].data  = &p->ucast_probes;
2706         t->neigh_vars[2].data  = &p->app_probes;
2707         t->neigh_vars[3].data  = &p->retrans_time;
2708         t->neigh_vars[4].data  = &p->base_reachable_time;
2709         t->neigh_vars[5].data  = &p->delay_probe_time;
2710         t->neigh_vars[6].data  = &p->gc_staletime;
2711         t->neigh_vars[7].data  = &p->queue_len;
2712         t->neigh_vars[8].data  = &p->proxy_qlen;
2713         t->neigh_vars[9].data  = &p->anycast_delay;
2714         t->neigh_vars[10].data = &p->proxy_delay;
2715         t->neigh_vars[11].data = &p->locktime;
2716         t->neigh_vars[12].data  = &p->retrans_time;
2717         t->neigh_vars[13].data  = &p->base_reachable_time;
2718
2719         if (dev) {
2720                 dev_name_source = dev->name;
2721                 /* Terminate the table early */
2722                 memset(&t->neigh_vars[14], 0, sizeof(t->neigh_vars[14]));
2723         } else {
2724                 dev_name_source = neigh_path[NEIGH_CTL_PATH_DEV].procname;
2725                 t->neigh_vars[14].data = (int *)(p + 1);
2726                 t->neigh_vars[15].data = (int *)(p + 1) + 1;
2727                 t->neigh_vars[16].data = (int *)(p + 1) + 2;
2728                 t->neigh_vars[17].data = (int *)(p + 1) + 3;
2729         }
2730
2731
2732         if (handler) {
2733                 /* RetransTime */
2734                 t->neigh_vars[3].proc_handler = handler;
2735                 t->neigh_vars[3].extra1 = dev;
2736                 /* ReachableTime */
2737                 t->neigh_vars[4].proc_handler = handler;
2738                 t->neigh_vars[4].extra1 = dev;
2739                 /* RetransTime (in milliseconds)*/
2740                 t->neigh_vars[12].proc_handler = handler;
2741                 t->neigh_vars[12].extra1 = dev;
2742                 /* ReachableTime (in milliseconds) */
2743                 t->neigh_vars[13].proc_handler = handler;
2744                 t->neigh_vars[13].extra1 = dev;
2745         }
2746
2747         t->dev_name = kstrdup(dev_name_source, GFP_KERNEL);
2748         if (!t->dev_name)
2749                 goto free;
2750
2751         neigh_path[NEIGH_CTL_PATH_DEV].procname = t->dev_name;
2752         neigh_path[NEIGH_CTL_PATH_PROTO].procname = p_name;
2753
2754         t->sysctl_header =
2755                 register_net_sysctl_table(neigh_parms_net(p), neigh_path, t->neigh_vars);
2756         if (!t->sysctl_header)
2757                 goto free_procname;
2758
2759         p->sysctl_table = t;
2760         return 0;
2761
2762 free_procname:
2763         kfree(t->dev_name);
2764 free:
2765         kfree(t);
2766 err:
2767         return -ENOBUFS;
2768 }
2769 EXPORT_SYMBOL(neigh_sysctl_register);
2770
2771 void neigh_sysctl_unregister(struct neigh_parms *p)
2772 {
2773         if (p->sysctl_table) {
2774                 struct neigh_sysctl_table *t = p->sysctl_table;
2775                 p->sysctl_table = NULL;
2776                 unregister_sysctl_table(t->sysctl_header);
2777                 kfree(t->dev_name);
2778                 kfree(t);
2779         }
2780 }
2781 EXPORT_SYMBOL(neigh_sysctl_unregister);
2782
2783 #endif  /* CONFIG_SYSCTL */
2784
2785 static int __init neigh_init(void)
2786 {
2787         rtnl_register(PF_UNSPEC, RTM_NEWNEIGH, neigh_add, NULL);
2788         rtnl_register(PF_UNSPEC, RTM_DELNEIGH, neigh_delete, NULL);
2789         rtnl_register(PF_UNSPEC, RTM_GETNEIGH, NULL, neigh_dump_info);
2790
2791         rtnl_register(PF_UNSPEC, RTM_GETNEIGHTBL, NULL, neightbl_dump_info);
2792         rtnl_register(PF_UNSPEC, RTM_SETNEIGHTBL, neightbl_set, NULL);
2793
2794         return 0;
2795 }
2796
2797 subsys_initcall(neigh_init);
2798