git.oblomov.eu Git - linux-2.6/blob - net/core/neighbour.c

   1 /*
   2  *      Generic address resolution entity
   3  *
   4  *      Authors:
   5  *      Pedro Roque             <roque@di.fc.ul.pt>
   6  *      Alexey Kuznetsov        <kuznet@ms2.inr.ac.ru>
   7  *
   8  *      This program is free software; you can redistribute it and/or
   9  *      modify it under the terms of the GNU General Public License
  10  *      as published by the Free Software Foundation; either version
  11  *      2 of the License, or (at your option) any later version.
  12  *
  13  *      Fixes:
  14  *      Vitaly E. Lavrov        releasing NULL neighbor in neigh_add.
  15  *      Harald Welte            Add neighbour cache statistics like rtstat
  16  */
  17
  18 #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 #ifdef CONFIG_NET_NS
 535         n->net = hold_net(net);
 536 #endif
 537         memcpy(n->key, pkey, key_len);
 538         n->dev = dev;
 539         if (dev)
 540                 dev_hold(dev);
 541
 542         if (tbl->pconstructor && tbl->pconstructor(n)) {
 543                 if (dev)
 544                         dev_put(dev);
 545                 release_net(net);
 546                 kfree(n);
 547                 n = NULL;
 548                 goto out;
 549         }
 550
 551         write_lock_bh(&tbl->lock);
 552         n->next = tbl->phash_buckets[hash_val];
 553         tbl->phash_buckets[hash_val] = n;
 554         write_unlock_bh(&tbl->lock);
 555 out:
 556         return n;
 557 }
 558 EXPORT_SYMBOL(pneigh_lookup);
 559
 560
 561 int pneigh_delete(struct neigh_table *tbl, struct net *net, const void *pkey,
 562                   struct net_device *dev)
 563 {
 564         struct pneigh_entry *n, **np;
 565         int key_len = tbl->key_len;
 566         u32 hash_val = pneigh_hash(pkey, key_len);
 567
 568         write_lock_bh(&tbl->lock);
 569         for (np = &tbl->phash_buckets[hash_val]; (n = *np) != NULL;
 570              np = &n->next) {
 571                 if (!memcmp(n->key, pkey, key_len) && n->dev == dev &&
 572                     net_eq(pneigh_net(n), net)) {
 573                         *np = n->next;
 574                         write_unlock_bh(&tbl->lock);
 575                         if (tbl->pdestructor)
 576                                 tbl->pdestructor(n);
 577                         if (n->dev)
 578                                 dev_put(n->dev);
 579                         release_net(pneigh_net(n));
 580                         kfree(n);
 581                         return 0;
 582                 }
 583         }
 584         write_unlock_bh(&tbl->lock);
 585         return -ENOENT;
 586 }
 587
 588 static int pneigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
 589 {
 590         struct pneigh_entry *n, **np;
 591         u32 h;
 592
 593         for (h = 0; h <= PNEIGH_HASHMASK; h++) {
 594                 np = &tbl->phash_buckets[h];
 595                 while ((n = *np) != NULL) {
 596                         if (!dev || n->dev == dev) {
 597                                 *np = n->next;
 598                                 if (tbl->pdestructor)
 599                                         tbl->pdestructor(n);
 600                                 if (n->dev)
 601                                         dev_put(n->dev);
 602                                 release_net(pneigh_net(n));
 603                                 kfree(n);
 604                                 continue;
 605                         }
 606                         np = &n->next;
 607                 }
 608         }
 609         return -ENOENT;
 610 }
 611
 612 static void neigh_parms_destroy(struct neigh_parms *parms);
 613
 614 static inline void neigh_parms_put(struct neigh_parms *parms)
 615 {
 616         if (atomic_dec_and_test(&parms->refcnt))
 617                 neigh_parms_destroy(parms);
 618 }
 619
 620 /*
 621  *      neighbour must already be out of the table;
 622  *
 623  */
 624 void neigh_destroy(struct neighbour *neigh)
 625 {
 626         struct hh_cache *hh;
 627
 628         NEIGH_CACHE_STAT_INC(neigh->tbl, destroys);
 629
 630         if (!neigh->dead) {
 631                 printk(KERN_WARNING
 632                        "Destroying alive neighbour %p\n", neigh);
 633                 dump_stack();
 634                 return;
 635         }
 636
 637         if (neigh_del_timer(neigh))
 638                 printk(KERN_WARNING "Impossible event.\n");
 639
 640         while ((hh = neigh->hh) != NULL) {
 641                 neigh->hh = hh->hh_next;
 642                 hh->hh_next = NULL;
 643
 644                 write_seqlock_bh(&hh->hh_lock);
 645                 hh->hh_output = neigh_blackhole;
 646                 write_sequnlock_bh(&hh->hh_lock);
 647                 if (atomic_dec_and_test(&hh->hh_refcnt))
 648                         kfree(hh);
 649         }
 650
 651         skb_queue_purge(&neigh->arp_queue);
 652
 653         dev_put(neigh->dev);
 654         neigh_parms_put(neigh->parms);
 655
 656         NEIGH_PRINTK2("neigh %p is destroyed.\n", neigh);
 657
 658         atomic_dec(&neigh->tbl->entries);
 659         kmem_cache_free(neigh->tbl->kmem_cachep, neigh);
 660 }
 661 EXPORT_SYMBOL(neigh_destroy);
 662
 663 /* Neighbour state is suspicious;
 664    disable fast path.
 665
 666    Called with write_locked neigh.
 667  */
 668 static void neigh_suspect(struct neighbour *neigh)
 669 {
 670         struct hh_cache *hh;
 671
 672         NEIGH_PRINTK2("neigh %p is suspected.\n", neigh);
 673
 674         neigh->output = neigh->ops->output;
 675
 676         for (hh = neigh->hh; hh; hh = hh->hh_next)
 677                 hh->hh_output = neigh->ops->output;
 678 }
 679
 680 /* Neighbour state is OK;
 681    enable fast path.
 682
 683    Called with write_locked neigh.
 684  */
 685 static void neigh_connect(struct neighbour *neigh)
 686 {
 687         struct hh_cache *hh;
 688
 689         NEIGH_PRINTK2("neigh %p is connected.\n", neigh);
 690
 691         neigh->output = neigh->ops->connected_output;
 692
 693         for (hh = neigh->hh; hh; hh = hh->hh_next)
 694                 hh->hh_output = neigh->ops->hh_output;
 695 }
 696
 697 static void neigh_periodic_timer(unsigned long arg)
 698 {
 699         struct neigh_table *tbl = (struct neigh_table *)arg;
 700         struct neighbour *n, **np;
 701         unsigned long expire, now = jiffies;
 702
 703         NEIGH_CACHE_STAT_INC(tbl, periodic_gc_runs);
 704
 705         write_lock(&tbl->lock);
 706
 707         /*
 708          *      periodically recompute ReachableTime from random function
 709          */
 710
 711         if (time_after(now, tbl->last_rand + 300 * HZ)) {
 712                 struct neigh_parms *p;
 713                 tbl->last_rand = now;
 714                 for (p = &tbl->parms; p; p = p->next)
 715                         p->reachable_time =
 716                                 neigh_rand_reach_time(p->base_reachable_time);
 717         }
 718
 719         np = &tbl->hash_buckets[tbl->hash_chain_gc];
 720         tbl->hash_chain_gc = ((tbl->hash_chain_gc + 1) & tbl->hash_mask);
 721
 722         while ((n = *np) != NULL) {
 723                 unsigned int state;
 724
 725                 write_lock(&n->lock);
 726
 727                 state = n->nud_state;
 728                 if (state & (NUD_PERMANENT | NUD_IN_TIMER)) {
 729                         write_unlock(&n->lock);
 730                         goto next_elt;
 731                 }
 732
 733                 if (time_before(n->used, n->confirmed))
 734                         n->used = n->confirmed;
 735
 736                 if (atomic_read(&n->refcnt) == 1 &&
 737                     (state == NUD_FAILED ||
 738                      time_after(now, n->used + n->parms->gc_staletime))) {
 739                         *np = n->next;
 740                         n->dead = 1;
 741                         write_unlock(&n->lock);
 742                         neigh_cleanup_and_release(n);
 743                         continue;
 744                 }
 745                 write_unlock(&n->lock);
 746
 747 next_elt:
 748                 np = &n->next;
 749         }
 750
 751         /* Cycle through all hash buckets every base_reachable_time/2 ticks.
 752          * ARP entry timeouts range from 1/2 base_reachable_time to 3/2
 753          * base_reachable_time.
 754          */
 755         expire = tbl->parms.base_reachable_time >> 1;
 756         expire /= (tbl->hash_mask + 1);
 757         if (!expire)
 758                 expire = 1;
 759
 760         if (expire>HZ)
 761                 mod_timer(&tbl->gc_timer, round_jiffies(now + expire));
 762         else
 763                 mod_timer(&tbl->gc_timer, now + expire);
 764
 765         write_unlock(&tbl->lock);
 766 }
 767
 768 static __inline__ int neigh_max_probes(struct neighbour *n)
 769 {
 770         struct neigh_parms *p = n->parms;
 771         return (n->nud_state & NUD_PROBE ?
 772                 p->ucast_probes :
 773                 p->ucast_probes + p->app_probes + p->mcast_probes);
 774 }
 775
 776 /* Called when a timer expires for a neighbour entry. */
 777
 778 static void neigh_timer_handler(unsigned long arg)
 779 {
 780         unsigned long now, next;
 781         struct neighbour *neigh = (struct neighbour *)arg;
 782         unsigned state;
 783         int notify = 0;
 784
 785         write_lock(&neigh->lock);
 786
 787         state = neigh->nud_state;
 788         now = jiffies;
 789         next = now + HZ;
 790
 791         if (!(state & NUD_IN_TIMER)) {
 792 #ifndef CONFIG_SMP
 793                 printk(KERN_WARNING "neigh: timer & !nud_in_timer\n");
 794 #endif
 795                 goto out;
 796         }
 797
 798         if (state & NUD_REACHABLE) {
 799                 if (time_before_eq(now,
 800                                    neigh->confirmed + neigh->parms->reachable_time)) {
 801                         NEIGH_PRINTK2("neigh %p is still alive.\n", neigh);
 802                         next = neigh->confirmed + neigh->parms->reachable_time;
 803                 } else if (time_before_eq(now,
 804                                           neigh->used + neigh->parms->delay_probe_time)) {
 805                         NEIGH_PRINTK2("neigh %p is delayed.\n", neigh);
 806                         neigh->nud_state = NUD_DELAY;
 807                         neigh->updated = jiffies;
 808                         neigh_suspect(neigh);
 809                         next = now + neigh->parms->delay_probe_time;
 810                 } else {
 811                         NEIGH_PRINTK2("neigh %p is suspected.\n", neigh);
 812                         neigh->nud_state = NUD_STALE;
 813                         neigh->updated = jiffies;
 814                         neigh_suspect(neigh);
 815                         notify = 1;
 816                 }
 817         } else if (state & NUD_DELAY) {
 818                 if (time_before_eq(now,
 819                                    neigh->confirmed + neigh->parms->delay_probe_time)) {
 820                         NEIGH_PRINTK2("neigh %p is now reachable.\n", neigh);
 821                         neigh->nud_state = NUD_REACHABLE;
 822                         neigh->updated = jiffies;
 823                         neigh_connect(neigh);
 824                         notify = 1;
 825                         next = neigh->confirmed + neigh->parms->reachable_time;
 826                 } else {
 827                         NEIGH_PRINTK2("neigh %p is probed.\n", neigh);
 828                         neigh->nud_state = NUD_PROBE;
 829                         neigh->updated = jiffies;
 830                         atomic_set(&neigh->probes, 0);
 831                         next = now + neigh->parms->retrans_time;
 832                 }
 833         } else {
 834                 /* NUD_PROBE|NUD_INCOMPLETE */
 835                 next = now + neigh->parms->retrans_time;
 836         }
 837
 838         if ((neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) &&
 839             atomic_read(&neigh->probes) >= neigh_max_probes(neigh)) {
 840                 struct sk_buff *skb;
 841
 842                 neigh->nud_state = NUD_FAILED;
 843                 neigh->updated = jiffies;
 844                 notify = 1;
 845                 NEIGH_CACHE_STAT_INC(neigh->tbl, res_failed);
 846                 NEIGH_PRINTK2("neigh %p is failed.\n", neigh);
 847
 848                 /* It is very thin place. report_unreachable is very complicated
 849                    routine. Particularly, it can hit the same neighbour entry!
 850
 851                    So that, we try to be accurate and avoid dead loop. --ANK
 852                  */
 853                 while (neigh->nud_state == NUD_FAILED &&
 854                        (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
 855                         write_unlock(&neigh->lock);
 856                         neigh->ops->error_report(neigh, skb);
 857                         write_lock(&neigh->lock);
 858                 }
 859                 skb_queue_purge(&neigh->arp_queue);
 860         }
 861
 862         if (neigh->nud_state & NUD_IN_TIMER) {
 863                 if (time_before(next, jiffies + HZ/2))
 864                         next = jiffies + HZ/2;
 865                 if (!mod_timer(&neigh->timer, next))
 866                         neigh_hold(neigh);
 867         }
 868         if (neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) {
 869                 struct sk_buff *skb = skb_peek(&neigh->arp_queue);
 870                 /* keep skb alive even if arp_queue overflows */
 871                 if (skb)
 872                         skb = skb_copy(skb, GFP_ATOMIC);
 873                 write_unlock(&neigh->lock);
 874                 neigh->ops->solicit(neigh, skb);
 875                 atomic_inc(&neigh->probes);
 876                 if (skb)
 877                         kfree_skb(skb);
 878         } else {
 879 out:
 880                 write_unlock(&neigh->lock);
 881         }
 882
 883         if (notify)
 884                 neigh_update_notify(neigh);
 885
 886         neigh_release(neigh);
 887 }
 888
 889 int __neigh_event_send(struct neighbour *neigh, struct sk_buff *skb)
 890 {
 891         int rc;
 892         unsigned long now;
 893
 894         write_lock_bh(&neigh->lock);
 895
 896         rc = 0;
 897         if (neigh->nud_state & (NUD_CONNECTED | NUD_DELAY | NUD_PROBE))
 898                 goto out_unlock_bh;
 899
 900         now = jiffies;
 901
 902         if (!(neigh->nud_state & (NUD_STALE | NUD_INCOMPLETE))) {
 903                 if (neigh->parms->mcast_probes + neigh->parms->app_probes) {
 904                         atomic_set(&neigh->probes, neigh->parms->ucast_probes);
 905                         neigh->nud_state     = NUD_INCOMPLETE;
 906                         neigh->updated = jiffies;
 907                         neigh_add_timer(neigh, now + 1);
 908                 } else {
 909                         neigh->nud_state = NUD_FAILED;
 910                         neigh->updated = jiffies;
 911                         write_unlock_bh(&neigh->lock);
 912
 913                         if (skb)
 914                                 kfree_skb(skb);
 915                         return 1;
 916                 }
 917         } else if (neigh->nud_state & NUD_STALE) {
 918                 NEIGH_PRINTK2("neigh %p is delayed.\n", neigh);
 919                 neigh->nud_state = NUD_DELAY;
 920                 neigh->updated = jiffies;
 921                 neigh_add_timer(neigh,
 922                                 jiffies + neigh->parms->delay_probe_time);
 923         }
 924
 925         if (neigh->nud_state == NUD_INCOMPLETE) {
 926                 if (skb) {
 927                         if (skb_queue_len(&neigh->arp_queue) >=
 928                             neigh->parms->queue_len) {
 929                                 struct sk_buff *buff;
 930                                 buff = __skb_dequeue(&neigh->arp_queue);
 931                                 kfree_skb(buff);
 932                                 NEIGH_CACHE_STAT_INC(neigh->tbl, unres_discards);
 933                         }
 934                         __skb_queue_tail(&neigh->arp_queue, skb);
 935                 }
 936                 rc = 1;
 937         }
 938 out_unlock_bh:
 939         write_unlock_bh(&neigh->lock);
 940         return rc;
 941 }
 942 EXPORT_SYMBOL(__neigh_event_send);
 943
 944 static void neigh_update_hhs(struct neighbour *neigh)
 945 {
 946         struct hh_cache *hh;
 947         void (*update)(struct hh_cache*, const struct net_device*, const unsigned char *)
 948                 = neigh->dev->header_ops->cache_update;
 949
 950         if (update) {
 951                 for (hh = neigh->hh; hh; hh = hh->hh_next) {
 952                         write_seqlock_bh(&hh->hh_lock);
 953                         update(hh, neigh->dev, neigh->ha);
 954                         write_sequnlock_bh(&hh->hh_lock);
 955                 }
 956         }
 957 }
 958
 959
 960
 961 /* Generic update routine.
 962    -- lladdr is new lladdr or NULL, if it is not supplied.
 963    -- new    is new state.
 964    -- flags
 965         NEIGH_UPDATE_F_OVERRIDE allows to override existing lladdr,
 966                                 if it is different.
 967         NEIGH_UPDATE_F_WEAK_OVERRIDE will suspect existing "connected"
 968                                 lladdr instead of overriding it
 969                                 if it is different.
 970                                 It also allows to retain current state
 971                                 if lladdr is unchanged.
 972         NEIGH_UPDATE_F_ADMIN    means that the change is administrative.
 973
 974         NEIGH_UPDATE_F_OVERRIDE_ISROUTER allows to override existing
 975                                 NTF_ROUTER flag.
 976         NEIGH_UPDATE_F_ISROUTER indicates if the neighbour is known as
 977                                 a router.
 978
 979    Caller MUST hold reference count on the entry.
 980  */
 981
 982 int neigh_update(struct neighbour *neigh, const u8 *lladdr, u8 new,
 983                  u32 flags)
 984 {
 985         u8 old;
 986         int err;
 987         int notify = 0;
 988         struct net_device *dev;
 989         int update_isrouter = 0;
 990
 991         write_lock_bh(&neigh->lock);
 992
 993         dev    = neigh->dev;
 994         old    = neigh->nud_state;
 995         err    = -EPERM;
 996
 997         if (!(flags & NEIGH_UPDATE_F_ADMIN) &&
 998             (old & (NUD_NOARP | NUD_PERMANENT)))
 999                 goto out;
1000
1001         if (!(new & NUD_VALID)) {
1002                 neigh_del_timer(neigh);
1003                 if (old & NUD_CONNECTED)
1004                         neigh_suspect(neigh);
1005                 neigh->nud_state = new;
1006                 err = 0;
1007                 notify = old & NUD_VALID;
1008                 goto out;
1009         }
1010
1011         /* Compare new lladdr with cached one */
1012         if (!dev->addr_len) {
1013                 /* First case: device needs no address. */
1014                 lladdr = neigh->ha;
1015         } else if (lladdr) {
1016                 /* The second case: if something is already cached
1017                    and a new address is proposed:
1018                    - compare new & old
1019                    - if they are different, check override flag
1020                  */
1021                 if ((old & NUD_VALID) &&
1022                     !memcmp(lladdr, neigh->ha, dev->addr_len))
1023                         lladdr = neigh->ha;
1024         } else {
1025                 /* No address is supplied; if we know something,
1026                    use it, otherwise discard the request.
1027                  */
1028                 err = -EINVAL;
1029                 if (!(old & NUD_VALID))
1030                         goto out;
1031                 lladdr = neigh->ha;
1032         }
1033
1034         if (new & NUD_CONNECTED)
1035                 neigh->confirmed = jiffies;
1036         neigh->updated = jiffies;
1037
1038         /* If entry was valid and address is not changed,
1039            do not change entry state, if new one is STALE.
1040          */
1041         err = 0;
1042         update_isrouter = flags & NEIGH_UPDATE_F_OVERRIDE_ISROUTER;
1043         if (old & NUD_VALID) {
1044                 if (lladdr != neigh->ha && !(flags & NEIGH_UPDATE_F_OVERRIDE)) {
1045                         update_isrouter = 0;
1046                         if ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) &&
1047                             (old & NUD_CONNECTED)) {
1048                                 lladdr = neigh->ha;
1049                                 new = NUD_STALE;
1050                         } else
1051                                 goto out;
1052                 } else {
1053                         if (lladdr == neigh->ha && new == NUD_STALE &&
1054                             ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) ||
1055                              (old & NUD_CONNECTED))
1056                             )
1057                                 new = old;
1058                 }
1059         }
1060
1061         if (new != old) {
1062                 neigh_del_timer(neigh);
1063                 if (new & NUD_IN_TIMER)
1064                         neigh_add_timer(neigh, (jiffies +
1065                                                 ((new & NUD_REACHABLE) ?
1066                                                  neigh->parms->reachable_time :
1067                                                  0)));
1068                 neigh->nud_state = new;
1069         }
1070
1071         if (lladdr != neigh->ha) {
1072                 memcpy(&neigh->ha, lladdr, dev->addr_len);
1073                 neigh_update_hhs(neigh);
1074                 if (!(new & NUD_CONNECTED))
1075                         neigh->confirmed = jiffies -
1076                                       (neigh->parms->base_reachable_time << 1);
1077                 notify = 1;
1078         }
1079         if (new == old)
1080                 goto out;
1081         if (new & NUD_CONNECTED)
1082                 neigh_connect(neigh);
1083         else
1084                 neigh_suspect(neigh);
1085         if (!(old & NUD_VALID)) {
1086                 struct sk_buff *skb;
1087
1088                 /* Again: avoid dead loop if something went wrong */
1089
1090                 while (neigh->nud_state & NUD_VALID &&
1091                        (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
1092                         struct neighbour *n1 = neigh;
1093                         write_unlock_bh(&neigh->lock);
1094                         /* On shaper/eql skb->dst->neighbour != neigh :( */
1095                         if (skb->dst && skb->dst->neighbour)
1096                                 n1 = skb->dst->neighbour;
1097                         n1->output(skb);
1098                         write_lock_bh(&neigh->lock);
1099                 }
1100                 skb_queue_purge(&neigh->arp_queue);
1101         }
1102 out:
1103         if (update_isrouter) {
1104                 neigh->flags = (flags & NEIGH_UPDATE_F_ISROUTER) ?
1105                         (neigh->flags | NTF_ROUTER) :
1106                         (neigh->flags & ~NTF_ROUTER);
1107         }
1108         write_unlock_bh(&neigh->lock);
1109
1110         if (notify)
1111                 neigh_update_notify(neigh);
1112
1113         return err;
1114 }
1115 EXPORT_SYMBOL(neigh_update);
1116
1117 struct neighbour *neigh_event_ns(struct neigh_table *tbl,
1118                                  u8 *lladdr, void *saddr,
1119                                  struct net_device *dev)
1120 {
1121         struct neighbour *neigh = __neigh_lookup(tbl, saddr, dev,
1122                                                  lladdr || !dev->addr_len);
1123         if (neigh)
1124                 neigh_update(neigh, lladdr, NUD_STALE,
1125                              NEIGH_UPDATE_F_OVERRIDE);
1126         return neigh;
1127 }
1128 EXPORT_SYMBOL(neigh_event_ns);
1129
1130 static void neigh_hh_init(struct neighbour *n, struct dst_entry *dst,
1131                           __be16 protocol)
1132 {
1133         struct hh_cache *hh;
1134         struct net_device *dev = dst->dev;
1135
1136         for (hh = n->hh; hh; hh = hh->hh_next)
1137                 if (hh->hh_type == protocol)
1138                         break;
1139
1140         if (!hh && (hh = kzalloc(sizeof(*hh), GFP_ATOMIC)) != NULL) {
1141                 seqlock_init(&hh->hh_lock);
1142                 hh->hh_type = protocol;
1143                 atomic_set(&hh->hh_refcnt, 0);
1144                 hh->hh_next = NULL;
1145
1146                 if (dev->header_ops->cache(n, hh)) {
1147                         kfree(hh);
1148                         hh = NULL;
1149                 } else {
1150                         atomic_inc(&hh->hh_refcnt);
1151                         hh->hh_next = n->hh;
1152                         n->hh       = hh;
1153                         if (n->nud_state & NUD_CONNECTED)
1154                                 hh->hh_output = n->ops->hh_output;
1155                         else
1156                                 hh->hh_output = n->ops->output;
1157                 }
1158         }
1159         if (hh) {
1160                 atomic_inc(&hh->hh_refcnt);
1161                 dst->hh = hh;
1162         }
1163 }
1164
1165 /* This function can be used in contexts, where only old dev_queue_xmit
1166    worked, f.e. if you want to override normal output path (eql, shaper),
1167    but resolution is not made yet.
1168  */
1169
1170 int neigh_compat_output(struct sk_buff *skb)
1171 {
1172         struct net_device *dev = skb->dev;
1173
1174         __skb_pull(skb, skb_network_offset(skb));
1175
1176         if (dev_hard_header(skb, dev, ntohs(skb->protocol), NULL, NULL,
1177                             skb->len) < 0 &&
1178             dev->header_ops->rebuild(skb))
1179                 return 0;
1180
1181         return dev_queue_xmit(skb);
1182 }
1183 EXPORT_SYMBOL(neigh_compat_output);
1184
1185 /* Slow and careful. */
1186
1187 int neigh_resolve_output(struct sk_buff *skb)
1188 {
1189         struct dst_entry *dst = skb->dst;
1190         struct neighbour *neigh;
1191         int rc = 0;
1192
1193         if (!dst || !(neigh = dst->neighbour))
1194                 goto discard;
1195
1196         __skb_pull(skb, skb_network_offset(skb));
1197
1198         if (!neigh_event_send(neigh, skb)) {
1199                 int err;
1200                 struct net_device *dev = neigh->dev;
1201                 if (dev->header_ops->cache && !dst->hh) {
1202                         write_lock_bh(&neigh->lock);
1203                         if (!dst->hh)
1204                                 neigh_hh_init(neigh, dst, dst->ops->protocol);
1205                         err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1206                                               neigh->ha, NULL, skb->len);
1207                         write_unlock_bh(&neigh->lock);
1208                 } else {
1209                         read_lock_bh(&neigh->lock);
1210                         err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1211                                               neigh->ha, NULL, skb->len);
1212                         read_unlock_bh(&neigh->lock);
1213                 }
1214                 if (err >= 0)
1215                         rc = neigh->ops->queue_xmit(skb);
1216                 else
1217                         goto out_kfree_skb;
1218         }
1219 out:
1220         return rc;
1221 discard:
1222         NEIGH_PRINTK1("neigh_resolve_output: dst=%p neigh=%p\n",
1223                       dst, dst ? dst->neighbour : NULL);
1224 out_kfree_skb:
1225         rc = -EINVAL;
1226         kfree_skb(skb);
1227         goto out;
1228 }
1229 EXPORT_SYMBOL(neigh_resolve_output);
1230
1231 /* As fast as possible without hh cache */
1232
1233 int neigh_connected_output(struct sk_buff *skb)
1234 {
1235         int err;
1236         struct dst_entry *dst = skb->dst;
1237         struct neighbour *neigh = dst->neighbour;
1238         struct net_device *dev = neigh->dev;
1239
1240         __skb_pull(skb, skb_network_offset(skb));
1241
1242         read_lock_bh(&neigh->lock);
1243         err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1244                               neigh->ha, NULL, skb->len);
1245         read_unlock_bh(&neigh->lock);
1246         if (err >= 0)
1247                 err = neigh->ops->queue_xmit(skb);
1248         else {
1249                 err = -EINVAL;
1250                 kfree_skb(skb);
1251         }
1252         return err;
1253 }
1254 EXPORT_SYMBOL(neigh_connected_output);
1255
1256 static void neigh_proxy_process(unsigned long arg)
1257 {
1258         struct neigh_table *tbl = (struct neigh_table *)arg;
1259         long sched_next = 0;
1260         unsigned long now = jiffies;
1261         struct sk_buff *skb, *n;
1262
1263         spin_lock(&tbl->proxy_queue.lock);
1264
1265         skb_queue_walk_safe(&tbl->proxy_queue, skb, n) {
1266                 long tdif = NEIGH_CB(skb)->sched_next - now;
1267
1268                 if (tdif <= 0) {
1269                         struct net_device *dev = skb->dev;
1270                         __skb_unlink(skb, &tbl->proxy_queue);
1271                         if (tbl->proxy_redo && netif_running(dev))
1272                                 tbl->proxy_redo(skb);
1273                         else
1274                                 kfree_skb(skb);
1275
1276                         dev_put(dev);
1277                 } else if (!sched_next || tdif < sched_next)
1278                         sched_next = tdif;
1279         }
1280         del_timer(&tbl->proxy_timer);
1281         if (sched_next)
1282                 mod_timer(&tbl->proxy_timer, jiffies + sched_next);
1283         spin_unlock(&tbl->proxy_queue.lock);
1284 }
1285
1286 void pneigh_enqueue(struct neigh_table *tbl, struct neigh_parms *p,
1287                     struct sk_buff *skb)
1288 {
1289         unsigned long now = jiffies;
1290         unsigned long sched_next = now + (net_random() % p->proxy_delay);
1291
1292         if (tbl->proxy_queue.qlen > p->proxy_qlen) {
1293                 kfree_skb(skb);
1294                 return;
1295         }
1296
1297         NEIGH_CB(skb)->sched_next = sched_next;
1298         NEIGH_CB(skb)->flags |= LOCALLY_ENQUEUED;
1299
1300         spin_lock(&tbl->proxy_queue.lock);
1301         if (del_timer(&tbl->proxy_timer)) {
1302                 if (time_before(tbl->proxy_timer.expires, sched_next))
1303                         sched_next = tbl->proxy_timer.expires;
1304         }
1305         dst_release(skb->dst);
1306         skb->dst = NULL;
1307         dev_hold(skb->dev);
1308         __skb_queue_tail(&tbl->proxy_queue, skb);
1309         mod_timer(&tbl->proxy_timer, sched_next);
1310         spin_unlock(&tbl->proxy_queue.lock);
1311 }
1312 EXPORT_SYMBOL(pneigh_enqueue);
1313
1314 static inline struct neigh_parms *lookup_neigh_params(struct neigh_table *tbl,
1315                                                       struct net *net, int ifindex)
1316 {
1317         struct neigh_parms *p;
1318
1319         for (p = &tbl->parms; p; p = p->next) {
1320                 if ((p->dev && p->dev->ifindex == ifindex && net_eq(neigh_parms_net(p), net)) ||
1321                     (!p->dev && !ifindex))
1322                         return p;
1323         }
1324
1325         return NULL;
1326 }
1327
1328 struct neigh_parms *neigh_parms_alloc(struct net_device *dev,
1329                                       struct neigh_table *tbl)
1330 {
1331         struct neigh_parms *p, *ref;
1332         struct net *net;
1333
1334         net = dev_net(dev);
1335         ref = lookup_neigh_params(tbl, net, 0);
1336         if (!ref)
1337                 return NULL;
1338
1339         p = kmemdup(ref, sizeof(*p), GFP_KERNEL);
1340         if (p) {
1341                 p->tbl            = tbl;
1342                 atomic_set(&p->refcnt, 1);
1343                 INIT_RCU_HEAD(&p->rcu_head);
1344                 p->reachable_time =
1345                                 neigh_rand_reach_time(p->base_reachable_time);
1346
1347                 if (dev->neigh_setup && dev->neigh_setup(dev, p)) {
1348                         kfree(p);
1349                         return NULL;
1350                 }
1351
1352                 dev_hold(dev);
1353                 p->dev = dev;
1354 #ifdef CONFIG_NET_NS
1355                 p->net = hold_net(net);
1356 #endif
1357                 p->sysctl_table = NULL;
1358                 write_lock_bh(&tbl->lock);
1359                 p->next         = tbl->parms.next;
1360                 tbl->parms.next = p;
1361                 write_unlock_bh(&tbl->lock);
1362         }
1363         return p;
1364 }
1365 EXPORT_SYMBOL(neigh_parms_alloc);
1366
1367 static void neigh_rcu_free_parms(struct rcu_head *head)
1368 {
1369         struct neigh_parms *parms =
1370                 container_of(head, struct neigh_parms, rcu_head);
1371
1372         neigh_parms_put(parms);
1373 }
1374
1375 void neigh_parms_release(struct neigh_table *tbl, struct neigh_parms *parms)
1376 {
1377         struct neigh_parms **p;
1378
1379         if (!parms || parms == &tbl->parms)
1380                 return;
1381         write_lock_bh(&tbl->lock);
1382         for (p = &tbl->parms.next; *p; p = &(*p)->next) {
1383                 if (*p == parms) {
1384                         *p = parms->next;
1385                         parms->dead = 1;
1386                         write_unlock_bh(&tbl->lock);
1387                         if (parms->dev)
1388                                 dev_put(parms->dev);
1389                         call_rcu(&parms->rcu_head, neigh_rcu_free_parms);
1390                         return;
1391                 }
1392         }
1393         write_unlock_bh(&tbl->lock);
1394         NEIGH_PRINTK1("neigh_parms_release: not found\n");
1395 }
1396 EXPORT_SYMBOL(neigh_parms_release);
1397
1398 static void neigh_parms_destroy(struct neigh_parms *parms)
1399 {
1400         release_net(neigh_parms_net(parms));
1401         kfree(parms);
1402 }
1403
1404 static struct lock_class_key neigh_table_proxy_queue_class;
1405
1406 void neigh_table_init_no_netlink(struct neigh_table *tbl)
1407 {
1408         unsigned long now = jiffies;
1409         unsigned long phsize;
1410
1411 #ifdef CONFIG_NET_NS
1412         tbl->parms.net = &init_net;
1413 #endif
1414         atomic_set(&tbl->parms.refcnt, 1);
1415         INIT_RCU_HEAD(&tbl->parms.rcu_head);
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         tbl->pde = proc_create_data(tbl->id, 0, init_net.proc_net_stat,
1430                                     &neigh_stat_seq_fops, tbl);
1431         if (!tbl->pde)
1432                 panic("cannot create neighbour proc dir entry");
1433 #endif
1434
1435         tbl->hash_mask = 1;
1436         tbl->hash_buckets = neigh_hash_alloc(tbl->hash_mask + 1);
1437
1438         phsize = (PNEIGH_HASHMASK + 1) * sizeof(struct pneigh_entry *);
1439         tbl->phash_buckets = kzalloc(phsize, GFP_KERNEL);
1440
1441         if (!tbl->hash_buckets || !tbl->phash_buckets)
1442                 panic("cannot allocate neighbour cache hashes");
1443
1444         get_random_bytes(&tbl->hash_rnd, sizeof(tbl->hash_rnd));
1445
1446         rwlock_init(&tbl->lock);
1447         setup_timer(&tbl->gc_timer, neigh_periodic_timer, (unsigned long)tbl);
1448         tbl->gc_timer.expires  = now + 1;
1449         add_timer(&tbl->gc_timer);
1450
1451         setup_timer(&tbl->proxy_timer, neigh_proxy_process, (unsigned long)tbl);
1452         skb_queue_head_init_class(&tbl->proxy_queue,
1453                         &neigh_table_proxy_queue_class);
1454
1455         tbl->last_flush = now;
1456         tbl->last_rand  = now + tbl->parms.reachable_time * 20;
1457 }
1458 EXPORT_SYMBOL(neigh_table_init_no_netlink);
1459
1460 void neigh_table_init(struct neigh_table *tbl)
1461 {
1462         struct neigh_table *tmp;
1463
1464         neigh_table_init_no_netlink(tbl);
1465         write_lock(&neigh_tbl_lock);
1466         for (tmp = neigh_tables; tmp; tmp = tmp->next) {
1467                 if (tmp->family == tbl->family)
1468                         break;
1469         }
1470         tbl->next       = neigh_tables;
1471         neigh_tables    = tbl;
1472         write_unlock(&neigh_tbl_lock);
1473
1474         if (unlikely(tmp)) {
1475                 printk(KERN_ERR "NEIGH: Registering multiple tables for "
1476                        "family %d\n", tbl->family);
1477                 dump_stack();
1478         }
1479 }
1480 EXPORT_SYMBOL(neigh_table_init);
1481
1482 int neigh_table_clear(struct neigh_table *tbl)
1483 {
1484         struct neigh_table **tp;
1485
1486         /* It is not clean... Fix it to unload IPv6 module safely */
1487         del_timer_sync(&tbl->gc_timer);
1488         del_timer_sync(&tbl->proxy_timer);
1489         pneigh_queue_purge(&tbl->proxy_queue);
1490         neigh_ifdown(tbl, NULL);
1491         if (atomic_read(&tbl->entries))
1492                 printk(KERN_CRIT "neighbour leakage\n");
1493         write_lock(&neigh_tbl_lock);
1494         for (tp = &neigh_tables; *tp; tp = &(*tp)->next) {
1495                 if (*tp == tbl) {
1496                         *tp = tbl->next;
1497                         break;
1498                 }
1499         }
1500         write_unlock(&neigh_tbl_lock);
1501
1502         neigh_hash_free(tbl->hash_buckets, tbl->hash_mask + 1);
1503         tbl->hash_buckets = NULL;
1504
1505         kfree(tbl->phash_buckets);
1506         tbl->phash_buckets = NULL;
1507
1508         remove_proc_entry(tbl->id, init_net.proc_net_stat);
1509
1510         free_percpu(tbl->stats);
1511         tbl->stats = NULL;
1512
1513         kmem_cache_destroy(tbl->kmem_cachep);
1514         tbl->kmem_cachep = NULL;
1515
1516         return 0;
1517 }
1518 EXPORT_SYMBOL(neigh_table_clear);
1519
1520 static int neigh_delete(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
1521 {
1522         struct net *net = sock_net(skb->sk);
1523         struct ndmsg *ndm;
1524         struct nlattr *dst_attr;
1525         struct neigh_table *tbl;
1526         struct net_device *dev = NULL;
1527         int err = -EINVAL;
1528
1529         if (nlmsg_len(nlh) < sizeof(*ndm))
1530                 goto out;
1531
1532         dst_attr = nlmsg_find_attr(nlh, sizeof(*ndm), NDA_DST);
1533         if (dst_attr == NULL)
1534                 goto out;
1535
1536         ndm = nlmsg_data(nlh);
1537         if (ndm->ndm_ifindex) {
1538                 dev = dev_get_by_index(net, ndm->ndm_ifindex);
1539                 if (dev == NULL) {
1540                         err = -ENODEV;
1541                         goto out;
1542                 }
1543         }
1544
1545         read_lock(&neigh_tbl_lock);
1546         for (tbl = neigh_tables; tbl; tbl = tbl->next) {
1547                 struct neighbour *neigh;
1548
1549                 if (tbl->family != ndm->ndm_family)
1550                         continue;
1551                 read_unlock(&neigh_tbl_lock);
1552
1553                 if (nla_len(dst_attr) < tbl->key_len)
1554                         goto out_dev_put;
1555
1556                 if (ndm->ndm_flags & NTF_PROXY) {
1557                         err = pneigh_delete(tbl, net, nla_data(dst_attr), dev);
1558                         goto out_dev_put;
1559                 }
1560
1561                 if (dev == NULL)
1562                         goto out_dev_put;
1563
1564                 neigh = neigh_lookup(tbl, nla_data(dst_attr), dev);
1565                 if (neigh == NULL) {
1566                         err = -ENOENT;
1567                         goto out_dev_put;
1568                 }
1569
1570                 err = neigh_update(neigh, NULL, NUD_FAILED,
1571                                    NEIGH_UPDATE_F_OVERRIDE |
1572                                    NEIGH_UPDATE_F_ADMIN);
1573                 neigh_release(neigh);
1574                 goto out_dev_put;
1575         }
1576         read_unlock(&neigh_tbl_lock);
1577         err = -EAFNOSUPPORT;
1578
1579 out_dev_put:
1580         if (dev)
1581                 dev_put(dev);
1582 out:
1583         return err;
1584 }
1585
1586 static int neigh_add(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
1587 {
1588         struct net *net = sock_net(skb->sk);
1589         struct ndmsg *ndm;
1590         struct nlattr *tb[NDA_MAX+1];
1591         struct neigh_table *tbl;
1592         struct net_device *dev = NULL;
1593         int err;
1594
1595         err = nlmsg_parse(nlh, sizeof(*ndm), tb, NDA_MAX, NULL);
1596         if (err < 0)
1597                 goto out;
1598
1599         err = -EINVAL;
1600         if (tb[NDA_DST] == NULL)
1601                 goto out;
1602
1603         ndm = nlmsg_data(nlh);
1604         if (ndm->ndm_ifindex) {
1605                 dev = dev_get_by_index(net, ndm->ndm_ifindex);
1606                 if (dev == NULL) {
1607                         err = -ENODEV;
1608                         goto out;
1609                 }
1610
1611                 if (tb[NDA_LLADDR] && nla_len(tb[NDA_LLADDR]) < dev->addr_len)
1612                         goto out_dev_put;
1613         }
1614
1615         read_lock(&neigh_tbl_lock);
1616         for (tbl = neigh_tables; tbl; tbl = tbl->next) {
1617                 int flags = NEIGH_UPDATE_F_ADMIN | NEIGH_UPDATE_F_OVERRIDE;
1618                 struct neighbour *neigh;
1619                 void *dst, *lladdr;
1620
1621                 if (tbl->family != ndm->ndm_family)
1622                         continue;
1623                 read_unlock(&neigh_tbl_lock);
1624
1625                 if (nla_len(tb[NDA_DST]) < tbl->key_len)
1626                         goto out_dev_put;
1627                 dst = nla_data(tb[NDA_DST]);
1628                 lladdr = tb[NDA_LLADDR] ? nla_data(tb[NDA_LLADDR]) : NULL;
1629
1630                 if (ndm->ndm_flags & NTF_PROXY) {
1631                         struct pneigh_entry *pn;
1632
1633                         err = -ENOBUFS;
1634                         pn = pneigh_lookup(tbl, net, dst, dev, 1);
1635                         if (pn) {
1636                                 pn->flags = ndm->ndm_flags;
1637                                 err = 0;
1638                         }
1639                         goto out_dev_put;
1640                 }
1641
1642                 if (dev == NULL)
1643                         goto out_dev_put;
1644
1645                 neigh = neigh_lookup(tbl, dst, dev);
1646                 if (neigh == NULL) {
1647                         if (!(nlh->nlmsg_flags & NLM_F_CREATE)) {
1648                                 err = -ENOENT;
1649                                 goto out_dev_put;
1650                         }
1651
1652                         neigh = __neigh_lookup_errno(tbl, dst, dev);
1653                         if (IS_ERR(neigh)) {
1654                                 err = PTR_ERR(neigh);
1655                                 goto out_dev_put;
1656                         }
1657                 } else {
1658                         if (nlh->nlmsg_flags & NLM_F_EXCL) {
1659                                 err = -EEXIST;
1660                                 neigh_release(neigh);
1661                                 goto out_dev_put;
1662                         }
1663
1664                         if (!(nlh->nlmsg_flags & NLM_F_REPLACE))
1665                                 flags &= ~NEIGH_UPDATE_F_OVERRIDE;
1666                 }
1667
1668                 err = neigh_update(neigh, lladdr, ndm->ndm_state, flags);
1669                 neigh_release(neigh);
1670                 goto out_dev_put;
1671         }
1672
1673         read_unlock(&neigh_tbl_lock);
1674         err = -EAFNOSUPPORT;
1675
1676 out_dev_put:
1677         if (dev)
1678                 dev_put(dev);
1679 out:
1680         return err;
1681 }
1682
1683 static int neightbl_fill_parms(struct sk_buff *skb, struct neigh_parms *parms)
1684 {
1685         struct nlattr *nest;
1686
1687         nest = nla_nest_start(skb, NDTA_PARMS);
1688         if (nest == NULL)
1689                 return -ENOBUFS;
1690
1691         if (parms->dev)
1692                 NLA_PUT_U32(skb, NDTPA_IFINDEX, parms->dev->ifindex);
1693
1694         NLA_PUT_U32(skb, NDTPA_REFCNT, atomic_read(&parms->refcnt));
1695         NLA_PUT_U32(skb, NDTPA_QUEUE_LEN, parms->queue_len);
1696         NLA_PUT_U32(skb, NDTPA_PROXY_QLEN, parms->proxy_qlen);
1697         NLA_PUT_U32(skb, NDTPA_APP_PROBES, parms->app_probes);
1698         NLA_PUT_U32(skb, NDTPA_UCAST_PROBES, parms->ucast_probes);
1699         NLA_PUT_U32(skb, NDTPA_MCAST_PROBES, parms->mcast_probes);
1700         NLA_PUT_MSECS(skb, NDTPA_REACHABLE_TIME, parms->reachable_time);
1701         NLA_PUT_MSECS(skb, NDTPA_BASE_REACHABLE_TIME,
1702                       parms->base_reachable_time);
1703         NLA_PUT_MSECS(skb, NDTPA_GC_STALETIME, parms->gc_staletime);
1704         NLA_PUT_MSECS(skb, NDTPA_DELAY_PROBE_TIME, parms->delay_probe_time);
1705         NLA_PUT_MSECS(skb, NDTPA_RETRANS_TIME, parms->retrans_time);
1706         NLA_PUT_MSECS(skb, NDTPA_ANYCAST_DELAY, parms->anycast_delay);
1707         NLA_PUT_MSECS(skb, NDTPA_PROXY_DELAY, parms->proxy_delay);
1708         NLA_PUT_MSECS(skb, NDTPA_LOCKTIME, parms->locktime);
1709
1710         return nla_nest_end(skb, nest);
1711
1712 nla_put_failure:
1713         nla_nest_cancel(skb, nest);
1714         return -EMSGSIZE;
1715 }
1716
1717 static int neightbl_fill_info(struct sk_buff *skb, struct neigh_table *tbl,
1718                               u32 pid, u32 seq, int type, int flags)
1719 {
1720         struct nlmsghdr *nlh;
1721         struct ndtmsg *ndtmsg;
1722
1723         nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
1724         if (nlh == NULL)
1725                 return -EMSGSIZE;
1726
1727         ndtmsg = nlmsg_data(nlh);
1728
1729         read_lock_bh(&tbl->lock);
1730         ndtmsg->ndtm_family = tbl->family;
1731         ndtmsg->ndtm_pad1   = 0;
1732         ndtmsg->ndtm_pad2   = 0;
1733
1734         NLA_PUT_STRING(skb, NDTA_NAME, tbl->id);
1735         NLA_PUT_MSECS(skb, NDTA_GC_INTERVAL, tbl->gc_interval);
1736         NLA_PUT_U32(skb, NDTA_THRESH1, tbl->gc_thresh1);
1737         NLA_PUT_U32(skb, NDTA_THRESH2, tbl->gc_thresh2);
1738         NLA_PUT_U32(skb, NDTA_THRESH3, tbl->gc_thresh3);
1739
1740         {
1741                 unsigned long now = jiffies;
1742                 unsigned int flush_delta = now - tbl->last_flush;
1743                 unsigned int rand_delta = now - tbl->last_rand;
1744
1745                 struct ndt_config ndc = {
1746                         .ndtc_key_len           = tbl->key_len,
1747                         .ndtc_entry_size        = tbl->entry_size,
1748                         .ndtc_entries           = atomic_read(&tbl->entries),
1749                         .ndtc_last_flush        = jiffies_to_msecs(flush_delta),
1750                         .ndtc_last_rand         = jiffies_to_msecs(rand_delta),
1751                         .ndtc_hash_rnd          = tbl->hash_rnd,
1752                         .ndtc_hash_mask         = tbl->hash_mask,
1753                         .ndtc_hash_chain_gc     = tbl->hash_chain_gc,
1754                         .ndtc_proxy_qlen        = tbl->proxy_queue.qlen,
1755                 };
1756
1757                 NLA_PUT(skb, NDTA_CONFIG, sizeof(ndc), &ndc);
1758         }
1759
1760         {
1761                 int cpu;
1762                 struct ndt_stats ndst;
1763
1764                 memset(&ndst, 0, sizeof(ndst));
1765
1766                 for_each_possible_cpu(cpu) {
1767                         struct neigh_statistics *st;
1768
1769                         st = per_cpu_ptr(tbl->stats, cpu);
1770                         ndst.ndts_allocs                += st->allocs;
1771                         ndst.ndts_destroys              += st->destroys;
1772                         ndst.ndts_hash_grows            += st->hash_grows;
1773                         ndst.ndts_res_failed            += st->res_failed;
1774                         ndst.ndts_lookups               += st->lookups;
1775                         ndst.ndts_hits                  += st->hits;
1776                         ndst.ndts_rcv_probes_mcast      += st->rcv_probes_mcast;
1777                         ndst.ndts_rcv_probes_ucast      += st->rcv_probes_ucast;
1778                         ndst.ndts_periodic_gc_runs      += st->periodic_gc_runs;
1779                         ndst.ndts_forced_gc_runs        += st->forced_gc_runs;
1780                 }
1781
1782                 NLA_PUT(skb, NDTA_STATS, sizeof(ndst), &ndst);
1783         }
1784
1785         BUG_ON(tbl->parms.dev);
1786         if (neightbl_fill_parms(skb, &tbl->parms) < 0)
1787                 goto nla_put_failure;
1788
1789         read_unlock_bh(&tbl->lock);
1790         return nlmsg_end(skb, nlh);
1791
1792 nla_put_failure:
1793         read_unlock_bh(&tbl->lock);
1794         nlmsg_cancel(skb, nlh);
1795         return -EMSGSIZE;
1796 }
1797
1798 static int neightbl_fill_param_info(struct sk_buff *skb,
1799                                     struct neigh_table *tbl,
1800                                     struct neigh_parms *parms,
1801                                     u32 pid, u32 seq, int type,
1802                                     unsigned int flags)
1803 {
1804         struct ndtmsg *ndtmsg;
1805         struct nlmsghdr *nlh;
1806
1807         nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
1808         if (nlh == NULL)
1809                 return -EMSGSIZE;
1810
1811         ndtmsg = nlmsg_data(nlh);
1812
1813         read_lock_bh(&tbl->lock);
1814         ndtmsg->ndtm_family = tbl->family;
1815         ndtmsg->ndtm_pad1   = 0;
1816         ndtmsg->ndtm_pad2   = 0;
1817
1818         if (nla_put_string(skb, NDTA_NAME, tbl->id) < 0 ||
1819             neightbl_fill_parms(skb, parms) < 0)
1820                 goto errout;
1821
1822         read_unlock_bh(&tbl->lock);
1823         return nlmsg_end(skb, nlh);
1824 errout:
1825         read_unlock_bh(&tbl->lock);
1826         nlmsg_cancel(skb, nlh);
1827         return -EMSGSIZE;
1828 }
1829
1830 static const struct nla_policy nl_neightbl_policy[NDTA_MAX+1] = {
1831         [NDTA_NAME]             = { .type = NLA_STRING },
1832         [NDTA_THRESH1]          = { .type = NLA_U32 },
1833         [NDTA_THRESH2]          = { .type = NLA_U32 },
1834         [NDTA_THRESH3]          = { .type = NLA_U32 },
1835         [NDTA_GC_INTERVAL]      = { .type = NLA_U64 },
1836         [NDTA_PARMS]            = { .type = NLA_NESTED },
1837 };
1838
1839 static const struct nla_policy nl_ntbl_parm_policy[NDTPA_MAX+1] = {
1840         [NDTPA_IFINDEX]                 = { .type = NLA_U32 },
1841         [NDTPA_QUEUE_LEN]               = { .type = NLA_U32 },
1842         [NDTPA_PROXY_QLEN]              = { .type = NLA_U32 },
1843         [NDTPA_APP_PROBES]              = { .type = NLA_U32 },
1844         [NDTPA_UCAST_PROBES]            = { .type = NLA_U32 },
1845         [NDTPA_MCAST_PROBES]            = { .type = NLA_U32 },
1846         [NDTPA_BASE_REACHABLE_TIME]     = { .type = NLA_U64 },
1847         [NDTPA_GC_STALETIME]            = { .type = NLA_U64 },
1848         [NDTPA_DELAY_PROBE_TIME]        = { .type = NLA_U64 },
1849         [NDTPA_RETRANS_TIME]            = { .type = NLA_U64 },
1850         [NDTPA_ANYCAST_DELAY]           = { .type = NLA_U64 },
1851         [NDTPA_PROXY_DELAY]             = { .type = NLA_U64 },
1852         [NDTPA_LOCKTIME]                = { .type = NLA_U64 },
1853 };
1854
1855 static int neightbl_set(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
1856 {
1857         struct net *net = sock_net(skb->sk);
1858         struct neigh_table *tbl;
1859         struct ndtmsg *ndtmsg;
1860         struct nlattr *tb[NDTA_MAX+1];
1861         int err;
1862
1863         err = nlmsg_parse(nlh, sizeof(*ndtmsg), tb, NDTA_MAX,
1864                           nl_neightbl_policy);
1865         if (err < 0)
1866                 goto errout;
1867
1868         if (tb[NDTA_NAME] == NULL) {
1869                 err = -EINVAL;
1870                 goto errout;
1871         }
1872
1873         ndtmsg = nlmsg_data(nlh);
1874         read_lock(&neigh_tbl_lock);
1875         for (tbl = neigh_tables; tbl; tbl = tbl->next) {
1876                 if (ndtmsg->ndtm_family && tbl->family != ndtmsg->ndtm_family)
1877                         continue;
1878
1879                 if (nla_strcmp(tb[NDTA_NAME], tbl->id) == 0)
1880                         break;
1881         }
1882
1883         if (tbl == NULL) {
1884                 err = -ENOENT;
1885                 goto errout_locked;
1886         }
1887
1888         /*
1889          * We acquire tbl->lock to be nice to the periodic timers and
1890          * make sure they always see a consistent set of values.
1891          */
1892         write_lock_bh(&tbl->lock);
1893
1894         if (tb[NDTA_PARMS]) {
1895                 struct nlattr *tbp[NDTPA_MAX+1];
1896                 struct neigh_parms *p;
1897                 int i, ifindex = 0;
1898
1899                 err = nla_parse_nested(tbp, NDTPA_MAX, tb[NDTA_PARMS],
1900                                        nl_ntbl_parm_policy);
1901                 if (err < 0)
1902                         goto errout_tbl_lock;
1903
1904                 if (tbp[NDTPA_IFINDEX])
1905                         ifindex = nla_get_u32(tbp[NDTPA_IFINDEX]);
1906
1907                 p = lookup_neigh_params(tbl, net, ifindex);
1908                 if (p == NULL) {
1909                         err = -ENOENT;
1910                         goto errout_tbl_lock;
1911                 }
1912
1913                 for (i = 1; i <= NDTPA_MAX; i++) {
1914                         if (tbp[i] == NULL)
1915                                 continue;
1916
1917                         switch (i) {
1918                         case NDTPA_QUEUE_LEN:
1919                                 p->queue_len = nla_get_u32(tbp[i]);
1920                                 break;
1921                         case NDTPA_PROXY_QLEN:
1922                                 p->proxy_qlen = nla_get_u32(tbp[i]);
1923                                 break;
1924                         case NDTPA_APP_PROBES:
1925                                 p->app_probes = nla_get_u32(tbp[i]);
1926                                 break;
1927                         case NDTPA_UCAST_PROBES:
1928                                 p->ucast_probes = nla_get_u32(tbp[i]);
1929                                 break;
1930                         case NDTPA_MCAST_PROBES:
1931                                 p->mcast_probes = nla_get_u32(tbp[i]);
1932                                 break;
1933                         case NDTPA_BASE_REACHABLE_TIME:
1934                                 p->base_reachable_time = nla_get_msecs(tbp[i]);
1935                                 break;
1936                         case NDTPA_GC_STALETIME:
1937                                 p->gc_staletime = nla_get_msecs(tbp[i]);
1938                                 break;
1939                         case NDTPA_DELAY_PROBE_TIME:
1940                                 p->delay_probe_time = nla_get_msecs(tbp[i]);
1941                                 break;
1942                         case NDTPA_RETRANS_TIME:
1943                                 p->retrans_time = nla_get_msecs(tbp[i]);
1944                                 break;
1945                         case NDTPA_ANYCAST_DELAY:
1946                                 p->anycast_delay = nla_get_msecs(tbp[i]);
1947                                 break;
1948                         case NDTPA_PROXY_DELAY:
1949                                 p->proxy_delay = nla_get_msecs(tbp[i]);
1950                                 break;
1951                         case NDTPA_LOCKTIME:
1952                                 p->locktime = nla_get_msecs(tbp[i]);
1953                                 break;
1954                         }
1955                 }
1956         }
1957
1958         if (tb[NDTA_THRESH1])
1959                 tbl->gc_thresh1 = nla_get_u32(tb[NDTA_THRESH1]);
1960
1961         if (tb[NDTA_THRESH2])
1962                 tbl->gc_thresh2 = nla_get_u32(tb[NDTA_THRESH2]);
1963
1964         if (tb[NDTA_THRESH3])
1965                 tbl->gc_thresh3 = nla_get_u32(tb[NDTA_THRESH3]);
1966
1967         if (tb[NDTA_GC_INTERVAL])
1968                 tbl->gc_interval = nla_get_msecs(tb[NDTA_GC_INTERVAL]);
1969
1970         err = 0;
1971
1972 errout_tbl_lock:
1973         write_unlock_bh(&tbl->lock);
1974 errout_locked:
1975         read_unlock(&neigh_tbl_lock);
1976 errout:
1977         return err;
1978 }
1979
1980 static int neightbl_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
1981 {
1982         struct net *net = sock_net(skb->sk);
1983         int family, tidx, nidx = 0;
1984         int tbl_skip = cb->args[0];
1985         int neigh_skip = cb->args[1];
1986         struct neigh_table *tbl;
1987
1988         family = ((struct rtgenmsg *) nlmsg_data(cb->nlh))->rtgen_family;
1989
1990         read_lock(&neigh_tbl_lock);
1991         for (tbl = neigh_tables, tidx = 0; tbl; tbl = tbl->next, tidx++) {
1992                 struct neigh_parms *p;
1993
1994                 if (tidx < tbl_skip || (family && tbl->family != family))
1995                         continue;
1996
1997                 if (neightbl_fill_info(skb, tbl, NETLINK_CB(cb->skb).pid,
1998                                        cb->nlh->nlmsg_seq, RTM_NEWNEIGHTBL,
1999                                        NLM_F_MULTI) <= 0)
2000                         break;
2001
2002                 for (nidx = 0, p = tbl->parms.next; p; p = p->next) {
2003                         if (!net_eq(neigh_parms_net(p), net))
2004                                 continue;
2005
2006                         if (nidx++ < neigh_skip)
2007                                 continue;
2008
2009                         if (neightbl_fill_param_info(skb, tbl, p,
2010                                                      NETLINK_CB(cb->skb).pid,
2011                                                      cb->nlh->nlmsg_seq,
2012                                                      RTM_NEWNEIGHTBL,
2013                                                      NLM_F_MULTI) <= 0)
2014                                 goto out;
2015                 }
2016
2017                 neigh_skip = 0;
2018         }
2019 out:
2020         read_unlock(&neigh_tbl_lock);
2021         cb->args[0] = tidx;
2022         cb->args[1] = nidx;
2023
2024         return skb->len;
2025 }
2026
2027 static int neigh_fill_info(struct sk_buff *skb, struct neighbour *neigh,
2028                            u32 pid, u32 seq, int type, unsigned int flags)
2029 {
2030         unsigned long now = jiffies;
2031         struct nda_cacheinfo ci;
2032         struct nlmsghdr *nlh;
2033         struct ndmsg *ndm;
2034
2035         nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2036         if (nlh == NULL)
2037                 return -EMSGSIZE;
2038
2039         ndm = nlmsg_data(nlh);
2040         ndm->ndm_family  = neigh->ops->family;
2041         ndm->ndm_pad1    = 0;
2042         ndm->ndm_pad2    = 0;
2043         ndm->ndm_flags   = neigh->flags;
2044         ndm->ndm_type    = neigh->type;
2045         ndm->ndm_ifindex = neigh->dev->ifindex;
2046
2047         NLA_PUT(skb, NDA_DST, neigh->tbl->key_len, neigh->primary_key);
2048
2049         read_lock_bh(&neigh->lock);
2050         ndm->ndm_state   = neigh->nud_state;
2051         if ((neigh->nud_state & NUD_VALID) &&
2052             nla_put(skb, NDA_LLADDR, neigh->dev->addr_len, neigh->ha) < 0) {
2053                 read_unlock_bh(&neigh->lock);
2054                 goto nla_put_failure;
2055         }
2056
2057         ci.ndm_used      = jiffies_to_clock_t(now - neigh->used);
2058         ci.ndm_confirmed = jiffies_to_clock_t(now - neigh->confirmed);
2059         ci.ndm_updated   = jiffies_to_clock_t(now - neigh->updated);
2060         ci.ndm_refcnt    = atomic_read(&neigh->refcnt) - 1;
2061         read_unlock_bh(&neigh->lock);
2062
2063         NLA_PUT_U32(skb, NDA_PROBES, atomic_read(&neigh->probes));
2064         NLA_PUT(skb, NDA_CACHEINFO, sizeof(ci), &ci);
2065
2066         return nlmsg_end(skb, nlh);
2067
2068 nla_put_failure:
2069         nlmsg_cancel(skb, nlh);
2070         return -EMSGSIZE;
2071 }
2072
2073 static void neigh_update_notify(struct neighbour *neigh)
2074 {
2075         call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, neigh);
2076         __neigh_notify(neigh, RTM_NEWNEIGH, 0);
2077 }
2078
2079 static int neigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2080                             struct netlink_callback *cb)
2081 {
2082         struct net * net = sock_net(skb->sk);
2083         struct neighbour *n;
2084         int rc, h, s_h = cb->args[1];
2085         int idx, s_idx = idx = cb->args[2];
2086
2087         read_lock_bh(&tbl->lock);
2088         for (h = 0; h <= tbl->hash_mask; h++) {
2089                 if (h < s_h)
2090                         continue;
2091                 if (h > s_h)
2092                         s_idx = 0;
2093                 for (n = tbl->hash_buckets[h], idx = 0; n; n = n->next) {
2094                         int lidx;
2095                         if (dev_net(n->dev) != net)
2096                                 continue;
2097                         lidx = idx++;
2098                         if (lidx < s_idx)
2099                                 continue;
2100                         if (neigh_fill_info(skb, n, NETLINK_CB(cb->skb).pid,
2101                                             cb->nlh->nlmsg_seq,
2102                                             RTM_NEWNEIGH,
2103                                             NLM_F_MULTI) <= 0) {
2104                                 read_unlock_bh(&tbl->lock);
2105                                 rc = -1;
2106                                 goto out;
2107                         }
2108                 }
2109         }
2110         read_unlock_bh(&tbl->lock);
2111         rc = skb->len;
2112 out:
2113         cb->args[1] = h;
2114         cb->args[2] = idx;
2115         return rc;
2116 }
2117
2118 static int neigh_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2119 {
2120         struct neigh_table *tbl;
2121         int t, family, s_t;
2122
2123         read_lock(&neigh_tbl_lock);
2124         family = ((struct rtgenmsg *) nlmsg_data(cb->nlh))->rtgen_family;
2125         s_t = cb->args[0];
2126
2127         for (tbl = neigh_tables, t = 0; tbl; tbl = tbl->next, t++) {
2128                 if (t < s_t || (family && tbl->family != family))
2129                         continue;
2130                 if (t > s_t)
2131                         memset(&cb->args[1], 0, sizeof(cb->args) -
2132                                                 sizeof(cb->args[0]));
2133                 if (neigh_dump_table(tbl, skb, cb) < 0)
2134                         break;
2135         }
2136         read_unlock(&neigh_tbl_lock);
2137
2138         cb->args[0] = t;
2139         return skb->len;
2140 }
2141
2142 void neigh_for_each(struct neigh_table *tbl, void (*cb)(struct neighbour *, void *), void *cookie)
2143 {
2144         int chain;
2145
2146         read_lock_bh(&tbl->lock);
2147         for (chain = 0; chain <= tbl->hash_mask; chain++) {
2148                 struct neighbour *n;
2149
2150                 for (n = tbl->hash_buckets[chain]; n; n = n->next)
2151                         cb(n, cookie);
2152         }
2153         read_unlock_bh(&tbl->lock);
2154 }
2155 EXPORT_SYMBOL(neigh_for_each);
2156
2157 /* The tbl->lock must be held as a writer and BH disabled. */
2158 void __neigh_for_each_release(struct neigh_table *tbl,
2159                               int (*cb)(struct neighbour *))
2160 {
2161         int chain;
2162
2163         for (chain = 0; chain <= tbl->hash_mask; chain++) {
2164                 struct neighbour *n, **np;
2165
2166                 np = &tbl->hash_buckets[chain];
2167                 while ((n = *np) != NULL) {
2168                         int release;
2169
2170                         write_lock(&n->lock);
2171                         release = cb(n);
2172                         if (release) {
2173                                 *np = n->next;
2174                                 n->dead = 1;
2175                         } else
2176                                 np = &n->next;
2177                         write_unlock(&n->lock);
2178                         if (release)
2179                                 neigh_cleanup_and_release(n);
2180                 }
2181         }
2182 }
2183 EXPORT_SYMBOL(__neigh_for_each_release);
2184
2185 #ifdef CONFIG_PROC_FS
2186
2187 static struct neighbour *neigh_get_first(struct seq_file *seq)
2188 {
2189         struct neigh_seq_state *state = seq->private;
2190         struct net *net = seq_file_net(seq);
2191         struct neigh_table *tbl = state->tbl;
2192         struct neighbour *n = NULL;
2193         int bucket = state->bucket;
2194
2195         state->flags &= ~NEIGH_SEQ_IS_PNEIGH;
2196         for (bucket = 0; bucket <= tbl->hash_mask; bucket++) {
2197                 n = tbl->hash_buckets[bucket];
2198
2199                 while (n) {
2200                         if (!net_eq(dev_net(n->dev), net))
2201                                 goto next;
2202                         if (state->neigh_sub_iter) {
2203                                 loff_t fakep = 0;
2204                                 void *v;
2205
2206                                 v = state->neigh_sub_iter(state, n, &fakep);
2207                                 if (!v)
2208                                         goto next;
2209                         }
2210                         if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
2211                                 break;
2212                         if (n->nud_state & ~NUD_NOARP)
2213                                 break;
2214                 next:
2215                         n = n->next;
2216                 }
2217
2218                 if (n)
2219                         break;
2220         }
2221         state->bucket = bucket;
2222
2223         return n;
2224 }
2225
2226 static struct neighbour *neigh_get_next(struct seq_file *seq,
2227                                         struct neighbour *n,
2228                                         loff_t *pos)
2229 {
2230         struct neigh_seq_state *state = seq->private;
2231         struct net *net = seq_file_net(seq);
2232         struct neigh_table *tbl = state->tbl;
2233
2234         if (state->neigh_sub_iter) {
2235                 void *v = state->neigh_sub_iter(state, n, pos);
2236                 if (v)
2237                         return n;
2238         }
2239         n = n->next;
2240
2241         while (1) {
2242                 while (n) {
2243                         if (!net_eq(dev_net(n->dev), net))
2244                                 goto next;
2245                         if (state->neigh_sub_iter) {
2246                                 void *v = state->neigh_sub_iter(state, n, pos);
2247                                 if (v)
2248                                         return n;
2249                                 goto next;
2250                         }
2251                         if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
2252                                 break;
2253
2254                         if (n->nud_state & ~NUD_NOARP)
2255                                 break;
2256                 next:
2257                         n = n->next;
2258                 }
2259
2260                 if (n)
2261                         break;
2262
2263                 if (++state->bucket > tbl->hash_mask)
2264                         break;
2265
2266                 n = tbl->hash_buckets[state->bucket];
2267         }
2268
2269         if (n && pos)
2270                 --(*pos);
2271         return n;
2272 }
2273
2274 static struct neighbour *neigh_get_idx(struct seq_file *seq, loff_t *pos)
2275 {
2276         struct neighbour *n = neigh_get_first(seq);
2277
2278         if (n) {
2279                 --(*pos);
2280                 while (*pos) {
2281                         n = neigh_get_next(seq, n, pos);
2282                         if (!n)
2283                                 break;
2284                 }
2285         }
2286         return *pos ? NULL : n;
2287 }
2288
2289 static struct pneigh_entry *pneigh_get_first(struct seq_file *seq)
2290 {
2291         struct neigh_seq_state *state = seq->private;
2292         struct net *net = seq_file_net(seq);
2293         struct neigh_table *tbl = state->tbl;
2294         struct pneigh_entry *pn = NULL;
2295         int bucket = state->bucket;
2296
2297         state->flags |= NEIGH_SEQ_IS_PNEIGH;
2298         for (bucket = 0; bucket <= PNEIGH_HASHMASK; bucket++) {
2299                 pn = tbl->phash_buckets[bucket];
2300                 while (pn && !net_eq(pneigh_net(pn), net))
2301                         pn = pn->next;
2302                 if (pn)
2303                         break;
2304         }
2305         state->bucket = bucket;
2306
2307         return pn;
2308 }
2309
2310 static struct pneigh_entry *pneigh_get_next(struct seq_file *seq,
2311                                             struct pneigh_entry *pn,
2312                                             loff_t *pos)
2313 {
2314         struct neigh_seq_state *state = seq->private;
2315         struct net *net = seq_file_net(seq);
2316         struct neigh_table *tbl = state->tbl;
2317
2318         pn = pn->next;
2319         while (!pn) {
2320                 if (++state->bucket > PNEIGH_HASHMASK)
2321                         break;
2322                 pn = tbl->phash_buckets[state->bucket];
2323                 while (pn && !net_eq(pneigh_net(pn), net))
2324                         pn = pn->next;
2325                 if (pn)
2326                         break;
2327         }
2328
2329         if (pn && pos)
2330                 --(*pos);
2331
2332         return pn;
2333 }
2334
2335 static struct pneigh_entry *pneigh_get_idx(struct seq_file *seq, loff_t *pos)
2336 {
2337         struct pneigh_entry *pn = pneigh_get_first(seq);
2338
2339         if (pn) {
2340                 --(*pos);
2341                 while (*pos) {
2342                         pn = pneigh_get_next(seq, pn, pos);
2343                         if (!pn)
2344                                 break;
2345                 }
2346         }
2347         return *pos ? NULL : pn;
2348 }
2349
2350 static void *neigh_get_idx_any(struct seq_file *seq, loff_t *pos)
2351 {
2352         struct neigh_seq_state *state = seq->private;
2353         void *rc;
2354         loff_t idxpos = *pos;
2355
2356         rc = neigh_get_idx(seq, &idxpos);
2357         if (!rc && !(state->flags & NEIGH_SEQ_NEIGH_ONLY))
2358                 rc = pneigh_get_idx(seq, &idxpos);
2359
2360         return rc;
2361 }
2362
2363 void *neigh_seq_start(struct seq_file *seq, loff_t *pos, struct neigh_table *tbl, unsigned int neigh_seq_flags)
2364         __acquires(tbl->lock)
2365 {
2366         struct neigh_seq_state *state = seq->private;
2367
2368         state->tbl = tbl;
2369         state->bucket = 0;
2370         state->flags = (neigh_seq_flags & ~NEIGH_SEQ_IS_PNEIGH);
2371
2372         read_lock_bh(&tbl->lock);
2373
2374         return *pos ? neigh_get_idx_any(seq, pos) : SEQ_START_TOKEN;
2375 }
2376 EXPORT_SYMBOL(neigh_seq_start);
2377
2378 void *neigh_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2379 {
2380         struct neigh_seq_state *state;
2381         void *rc;
2382
2383         if (v == SEQ_START_TOKEN) {
2384                 rc = neigh_get_first(seq);
2385                 goto out;
2386         }
2387
2388         state = seq->private;
2389         if (!(state->flags & NEIGH_SEQ_IS_PNEIGH)) {
2390                 rc = neigh_get_next(seq, v, NULL);
2391                 if (rc)
2392                         goto out;
2393                 if (!(state->flags & NEIGH_SEQ_NEIGH_ONLY))
2394                         rc = pneigh_get_first(seq);
2395         } else {
2396                 BUG_ON(state->flags & NEIGH_SEQ_NEIGH_ONLY);
2397                 rc = pneigh_get_next(seq, v, NULL);
2398         }
2399 out:
2400         ++(*pos);
2401         return rc;
2402 }
2403 EXPORT_SYMBOL(neigh_seq_next);
2404
2405 void neigh_seq_stop(struct seq_file *seq, void *v)
2406         __releases(tbl->lock)
2407 {
2408         struct neigh_seq_state *state = seq->private;
2409         struct neigh_table *tbl = state->tbl;
2410
2411         read_unlock_bh(&tbl->lock);
2412 }
2413 EXPORT_SYMBOL(neigh_seq_stop);
2414
2415 /* statistics via seq_file */
2416
2417 static void *neigh_stat_seq_start(struct seq_file *seq, loff_t *pos)
2418 {
2419         struct proc_dir_entry *pde = seq->private;
2420         struct neigh_table *tbl = pde->data;
2421         int cpu;
2422
2423         if (*pos == 0)
2424                 return SEQ_START_TOKEN;
2425
2426         for (cpu = *pos-1; cpu < NR_CPUS; ++cpu) {
2427                 if (!cpu_possible(cpu))
2428                         continue;
2429                 *pos = cpu+1;
2430                 return per_cpu_ptr(tbl->stats, cpu);
2431         }
2432         return NULL;
2433 }
2434
2435 static void *neigh_stat_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2436 {
2437         struct proc_dir_entry *pde = seq->private;
2438         struct neigh_table *tbl = pde->data;
2439         int cpu;
2440
2441         for (cpu = *pos; cpu < NR_CPUS; ++cpu) {
2442                 if (!cpu_possible(cpu))
2443                         continue;
2444                 *pos = cpu+1;
2445                 return per_cpu_ptr(tbl->stats, cpu);
2446         }
2447         return NULL;
2448 }
2449
2450 static void neigh_stat_seq_stop(struct seq_file *seq, void *v)
2451 {
2452
2453 }
2454
2455 static int neigh_stat_seq_show(struct seq_file *seq, void *v)
2456 {
2457         struct proc_dir_entry *pde = seq->private;
2458         struct neigh_table *tbl = pde->data;
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);
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         err = rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
2545 errout:
2546         if (err < 0)
2547                 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
2548 }
2549
2550 #ifdef CONFIG_ARPD
2551 void neigh_app_ns(struct neighbour *n)
2552 {
2553         __neigh_notify(n, RTM_GETNEIGH, NLM_F_REQUEST);
2554 }
2555 EXPORT_SYMBOL(neigh_app_ns);
2556 #endif /* CONFIG_ARPD */
2557
2558 #ifdef CONFIG_SYSCTL
2559
2560 static struct neigh_sysctl_table {
2561         struct ctl_table_header *sysctl_header;
2562         struct ctl_table neigh_vars[__NET_NEIGH_MAX];
2563         char *dev_name;
2564 } neigh_sysctl_template __read_mostly = {
2565         .neigh_vars = {
2566                 {
2567                         .ctl_name       = NET_NEIGH_MCAST_SOLICIT,
2568                         .procname       = "mcast_solicit",
2569                         .maxlen         = sizeof(int),
2570                         .mode           = 0644,
2571                         .proc_handler   = &proc_dointvec,
2572                 },
2573                 {
2574                         .ctl_name       = NET_NEIGH_UCAST_SOLICIT,
2575                         .procname       = "ucast_solicit",
2576                         .maxlen         = sizeof(int),
2577                         .mode           = 0644,
2578                         .proc_handler   = &proc_dointvec,
2579                 },
2580                 {
2581                         .ctl_name       = NET_NEIGH_APP_SOLICIT,
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                         .ctl_name       = NET_NEIGH_REACHABLE_TIME,
2595                         .procname       = "base_reachable_time",
2596                         .maxlen         = sizeof(int),
2597                         .mode           = 0644,
2598                         .proc_handler   = &proc_dointvec_jiffies,
2599                         .strategy       = &sysctl_jiffies,
2600                 },
2601                 {
2602                         .ctl_name       = NET_NEIGH_DELAY_PROBE_TIME,
2603                         .procname       = "delay_first_probe_time",
2604                         .maxlen         = sizeof(int),
2605                         .mode           = 0644,
2606                         .proc_handler   = &proc_dointvec_jiffies,
2607                         .strategy       = &sysctl_jiffies,
2608                 },
2609                 {
2610                         .ctl_name       = NET_NEIGH_GC_STALE_TIME,
2611                         .procname       = "gc_stale_time",
2612                         .maxlen         = sizeof(int),
2613                         .mode           = 0644,
2614                         .proc_handler   = &proc_dointvec_jiffies,
2615                         .strategy       = &sysctl_jiffies,
2616                 },
2617                 {
2618                         .ctl_name       = NET_NEIGH_UNRES_QLEN,
2619                         .procname       = "unres_qlen",
2620                         .maxlen         = sizeof(int),
2621                         .mode           = 0644,
2622                         .proc_handler   = &proc_dointvec,
2623                 },
2624                 {
2625                         .ctl_name       = NET_NEIGH_PROXY_QLEN,
2626                         .procname       = "proxy_qlen",
2627                         .maxlen         = sizeof(int),
2628                         .mode           = 0644,
2629                         .proc_handler   = &proc_dointvec,
2630                 },
2631                 {
2632                         .procname       = "anycast_delay",
2633                         .maxlen         = sizeof(int),
2634                         .mode           = 0644,
2635                         .proc_handler   = &proc_dointvec_userhz_jiffies,
2636                 },
2637                 {
2638                         .procname       = "proxy_delay",
2639                         .maxlen         = sizeof(int),
2640                         .mode           = 0644,
2641                         .proc_handler   = &proc_dointvec_userhz_jiffies,
2642                 },
2643                 {
2644                         .procname       = "locktime",
2645                         .maxlen         = sizeof(int),
2646                         .mode           = 0644,
2647                         .proc_handler   = &proc_dointvec_userhz_jiffies,
2648                 },
2649                 {
2650                         .ctl_name       = NET_NEIGH_RETRANS_TIME_MS,
2651                         .procname       = "retrans_time_ms",
2652                         .maxlen         = sizeof(int),
2653                         .mode           = 0644,
2654                         .proc_handler   = &proc_dointvec_ms_jiffies,
2655                         .strategy       = &sysctl_ms_jiffies,
2656                 },
2657                 {
2658                         .ctl_name       = NET_NEIGH_REACHABLE_TIME_MS,
2659                         .procname       = "base_reachable_time_ms",
2660                         .maxlen         = sizeof(int),
2661                         .mode           = 0644,
2662                         .proc_handler   = &proc_dointvec_ms_jiffies,
2663                         .strategy       = &sysctl_ms_jiffies,
2664                 },
2665                 {
2666                         .ctl_name       = NET_NEIGH_GC_INTERVAL,
2667                         .procname       = "gc_interval",
2668                         .maxlen         = sizeof(int),
2669                         .mode           = 0644,
2670                         .proc_handler   = &proc_dointvec_jiffies,
2671                         .strategy       = &sysctl_jiffies,
2672                 },
2673                 {
2674                         .ctl_name       = NET_NEIGH_GC_THRESH1,
2675                         .procname       = "gc_thresh1",
2676                         .maxlen         = sizeof(int),
2677                         .mode           = 0644,
2678                         .proc_handler   = &proc_dointvec,
2679                 },
2680                 {
2681                         .ctl_name       = NET_NEIGH_GC_THRESH2,
2682                         .procname       = "gc_thresh2",
2683                         .maxlen         = sizeof(int),
2684                         .mode           = 0644,
2685                         .proc_handler   = &proc_dointvec,
2686                 },
2687                 {
2688                         .ctl_name       = NET_NEIGH_GC_THRESH3,
2689                         .procname       = "gc_thresh3",
2690                         .maxlen         = sizeof(int),
2691                         .mode           = 0644,
2692                         .proc_handler   = &proc_dointvec,
2693                 },
2694                 {},
2695         },
2696 };
2697
2698 int neigh_sysctl_register(struct net_device *dev, struct neigh_parms *p,
2699                           int p_id, int pdev_id, char *p_name,
2700                           proc_handler *handler, ctl_handler *strategy)
2701 {
2702         struct neigh_sysctl_table *t;
2703         const char *dev_name_source = NULL;
2704
2705 #define NEIGH_CTL_PATH_ROOT     0
2706 #define NEIGH_CTL_PATH_PROTO    1
2707 #define NEIGH_CTL_PATH_NEIGH    2
2708 #define NEIGH_CTL_PATH_DEV      3
2709
2710         struct ctl_path neigh_path[] = {
2711                 { .procname = "net",     .ctl_name = CTL_NET, },
2712                 { .procname = "proto",   .ctl_name = 0, },
2713                 { .procname = "neigh",   .ctl_name = 0, },
2714                 { .procname = "default", .ctl_name = NET_PROTO_CONF_DEFAULT, },
2715                 { },
2716         };
2717
2718         t = kmemdup(&neigh_sysctl_template, sizeof(*t), GFP_KERNEL);
2719         if (!t)
2720                 goto err;
2721
2722         t->neigh_vars[0].data  = &p->mcast_probes;
2723         t->neigh_vars[1].data  = &p->ucast_probes;
2724         t->neigh_vars[2].data  = &p->app_probes;
2725         t->neigh_vars[3].data  = &p->retrans_time;
2726         t->neigh_vars[4].data  = &p->base_reachable_time;
2727         t->neigh_vars[5].data  = &p->delay_probe_time;
2728         t->neigh_vars[6].data  = &p->gc_staletime;
2729         t->neigh_vars[7].data  = &p->queue_len;
2730         t->neigh_vars[8].data  = &p->proxy_qlen;
2731         t->neigh_vars[9].data  = &p->anycast_delay;
2732         t->neigh_vars[10].data = &p->proxy_delay;
2733         t->neigh_vars[11].data = &p->locktime;
2734         t->neigh_vars[12].data  = &p->retrans_time;
2735         t->neigh_vars[13].data  = &p->base_reachable_time;
2736
2737         if (dev) {
2738                 dev_name_source = dev->name;
2739                 neigh_path[NEIGH_CTL_PATH_DEV].ctl_name = dev->ifindex;
2740                 /* Terminate the table early */
2741                 memset(&t->neigh_vars[14], 0, sizeof(t->neigh_vars[14]));
2742         } else {
2743                 dev_name_source = neigh_path[NEIGH_CTL_PATH_DEV].procname;
2744                 t->neigh_vars[14].data = (int *)(p + 1);
2745                 t->neigh_vars[15].data = (int *)(p + 1) + 1;
2746                 t->neigh_vars[16].data = (int *)(p + 1) + 2;
2747                 t->neigh_vars[17].data = (int *)(p + 1) + 3;
2748         }
2749
2750
2751         if (handler || strategy) {
2752                 /* RetransTime */
2753                 t->neigh_vars[3].proc_handler = handler;
2754                 t->neigh_vars[3].strategy = strategy;
2755                 t->neigh_vars[3].extra1 = dev;
2756                 if (!strategy)
2757                         t->neigh_vars[3].ctl_name = CTL_UNNUMBERED;
2758                 /* ReachableTime */
2759                 t->neigh_vars[4].proc_handler = handler;
2760                 t->neigh_vars[4].strategy = strategy;
2761                 t->neigh_vars[4].extra1 = dev;
2762                 if (!strategy)
2763                         t->neigh_vars[4].ctl_name = CTL_UNNUMBERED;
2764                 /* RetransTime (in milliseconds)*/
2765                 t->neigh_vars[12].proc_handler = handler;
2766                 t->neigh_vars[12].strategy = strategy;
2767                 t->neigh_vars[12].extra1 = dev;
2768                 if (!strategy)
2769                         t->neigh_vars[12].ctl_name = CTL_UNNUMBERED;
2770                 /* ReachableTime (in milliseconds) */
2771                 t->neigh_vars[13].proc_handler = handler;
2772                 t->neigh_vars[13].strategy = strategy;
2773                 t->neigh_vars[13].extra1 = dev;
2774                 if (!strategy)
2775                         t->neigh_vars[13].ctl_name = CTL_UNNUMBERED;
2776         }
2777
2778         t->dev_name = kstrdup(dev_name_source, GFP_KERNEL);
2779         if (!t->dev_name)
2780                 goto free;
2781
2782         neigh_path[NEIGH_CTL_PATH_DEV].procname = t->dev_name;
2783         neigh_path[NEIGH_CTL_PATH_NEIGH].ctl_name = pdev_id;
2784         neigh_path[NEIGH_CTL_PATH_PROTO].procname = p_name;
2785         neigh_path[NEIGH_CTL_PATH_PROTO].ctl_name = p_id;
2786
2787         t->sysctl_header =
2788                 register_net_sysctl_table(neigh_parms_net(p), neigh_path, t->neigh_vars);
2789         if (!t->sysctl_header)
2790                 goto free_procname;
2791
2792         p->sysctl_table = t;
2793         return 0;
2794
2795 free_procname:
2796         kfree(t->dev_name);
2797 free:
2798         kfree(t);
2799 err:
2800         return -ENOBUFS;
2801 }
2802 EXPORT_SYMBOL(neigh_sysctl_register);
2803
2804 void neigh_sysctl_unregister(struct neigh_parms *p)
2805 {
2806         if (p->sysctl_table) {
2807                 struct neigh_sysctl_table *t = p->sysctl_table;
2808                 p->sysctl_table = NULL;
2809                 unregister_sysctl_table(t->sysctl_header);
2810                 kfree(t->dev_name);
2811                 kfree(t);
2812         }
2813 }
2814 EXPORT_SYMBOL(neigh_sysctl_unregister);
2815
2816 #endif  /* CONFIG_SYSCTL */
2817
2818 static int __init neigh_init(void)
2819 {
2820         rtnl_register(PF_UNSPEC, RTM_NEWNEIGH, neigh_add, NULL);
2821         rtnl_register(PF_UNSPEC, RTM_DELNEIGH, neigh_delete, NULL);
2822         rtnl_register(PF_UNSPEC, RTM_GETNEIGH, NULL, neigh_dump_info);
2823
2824         rtnl_register(PF_UNSPEC, RTM_GETNEIGHTBL, NULL, neightbl_dump_info);
2825         rtnl_register(PF_UNSPEC, RTM_SETNEIGHTBL, neightbl_set, NULL);
2826
2827         return 0;
2828 }
2829
2830 subsys_initcall(neigh_init);
2831