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/sched.h>
  23 #include <linux/netdevice.h>
  24 #include <linux/proc_fs.h>
  25 #ifdef CONFIG_SYSCTL
  26 #include <linux/sysctl.h>
  27 #endif
  28 #include <linux/times.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
  38 #define NEIGH_DEBUG 1
  39
  40 #define NEIGH_PRINTK(x...) printk(x)
  41 #define NEIGH_NOPRINTK(x...) do { ; } while(0)
  42 #define NEIGH_PRINTK0 NEIGH_PRINTK
  43 #define NEIGH_PRINTK1 NEIGH_NOPRINTK
  44 #define NEIGH_PRINTK2 NEIGH_NOPRINTK
  45
  46 #if NEIGH_DEBUG >= 1
  47 #undef NEIGH_PRINTK1
  48 #define NEIGH_PRINTK1 NEIGH_PRINTK
  49 #endif
  50 #if NEIGH_DEBUG >= 2
  51 #undef NEIGH_PRINTK2
  52 #define NEIGH_PRINTK2 NEIGH_PRINTK
  53 #endif
  54
  55 #define PNEIGH_HASHMASK         0xF
  56
  57 static void neigh_timer_handler(unsigned long arg);
  58 #ifdef CONFIG_ARPD
  59 static void neigh_app_notify(struct neighbour *n);
  60 #endif
  61 static int pneigh_ifdown(struct neigh_table *tbl, struct net_device *dev);
  62 void neigh_changeaddr(struct neigh_table *tbl, struct net_device *dev);
  63
  64 static struct neigh_table *neigh_tables;
  65 #ifdef CONFIG_PROC_FS
  66 static struct file_operations neigh_stat_seq_fops;
  67 #endif
  68
  69 /*
  70    Neighbour hash table buckets are protected with rwlock tbl->lock.
  71
  72    - All the scans/updates to hash buckets MUST be made under this lock.
  73    - NOTHING clever should be made under this lock: no callbacks
  74      to protocol backends, no attempts to send something to network.
  75      It will result in deadlocks, if backend/driver wants to use neighbour
  76      cache.
  77    - If the entry requires some non-trivial actions, increase
  78      its reference count and release table lock.
  79
  80    Neighbour entries are protected:
  81    - with reference count.
  82    - with rwlock neigh->lock
  83
  84    Reference count prevents destruction.
  85
  86    neigh->lock mainly serializes ll address data and its validity state.
  87    However, the same lock is used to protect another entry fields:
  88     - timer
  89     - resolution queue
  90
  91    Again, nothing clever shall be made under neigh->lock,
  92    the most complicated procedure, which we allow is dev->hard_header.
  93    It is supposed, that dev->hard_header is simplistic and does
  94    not make callbacks to neighbour tables.
  95
  96    The last lock is neigh_tbl_lock. It is pure SMP lock, protecting
  97    list of neighbour tables. This list is used only in process context,
  98  */
  99
 100 static DEFINE_RWLOCK(neigh_tbl_lock);
 101
 102 static int neigh_blackhole(struct sk_buff *skb)
 103 {
 104         kfree_skb(skb);
 105         return -ENETDOWN;
 106 }
 107
 108 /*
 109  * It is random distribution in the interval (1/2)*base...(3/2)*base.
 110  * It corresponds to default IPv6 settings and is not overridable,
 111  * because it is really reasonable choice.
 112  */
 113
 114 unsigned long neigh_rand_reach_time(unsigned long base)
 115 {
 116         return (base ? (net_random() % base) + (base >> 1) : 0);
 117 }
 118
 119
 120 static int neigh_forced_gc(struct neigh_table *tbl)
 121 {
 122         int shrunk = 0;
 123         int i;
 124
 125         NEIGH_CACHE_STAT_INC(tbl, forced_gc_runs);
 126
 127         write_lock_bh(&tbl->lock);
 128         for (i = 0; i <= tbl->hash_mask; i++) {
 129                 struct neighbour *n, **np;
 130
 131                 np = &tbl->hash_buckets[i];
 132                 while ((n = *np) != NULL) {
 133                         /* Neighbour record may be discarded if:
 134                          * - nobody refers to it.
 135                          * - it is not permanent
 136                          */
 137                         write_lock(&n->lock);
 138                         if (atomic_read(&n->refcnt) == 1 &&
 139                             !(n->nud_state & NUD_PERMANENT)) {
 140                                 *np     = n->next;
 141                                 n->dead = 1;
 142                                 shrunk  = 1;
 143                                 write_unlock(&n->lock);
 144                                 neigh_release(n);
 145                                 continue;
 146                         }
 147                         write_unlock(&n->lock);
 148                         np = &n->next;
 149                 }
 150         }
 151
 152         tbl->last_flush = jiffies;
 153
 154         write_unlock_bh(&tbl->lock);
 155
 156         return shrunk;
 157 }
 158
 159 static int neigh_del_timer(struct neighbour *n)
 160 {
 161         if ((n->nud_state & NUD_IN_TIMER) &&
 162             del_timer(&n->timer)) {
 163                 neigh_release(n);
 164                 return 1;
 165         }
 166         return 0;
 167 }
 168
 169 static void pneigh_queue_purge(struct sk_buff_head *list)
 170 {
 171         struct sk_buff *skb;
 172
 173         while ((skb = skb_dequeue(list)) != NULL) {
 174                 dev_put(skb->dev);
 175                 kfree_skb(skb);
 176         }
 177 }
 178
 179 static void neigh_flush_dev(struct neigh_table *tbl, struct net_device *dev)
 180 {
 181         int i;
 182
 183         for (i = 0; i <= tbl->hash_mask; i++) {
 184                 struct neighbour *n, **np = &tbl->hash_buckets[i];
 185
 186                 while ((n = *np) != NULL) {
 187                         if (dev && n->dev != dev) {
 188                                 np = &n->next;
 189                                 continue;
 190                         }
 191                         *np = n->next;
 192                         write_lock(&n->lock);
 193                         neigh_del_timer(n);
 194                         n->dead = 1;
 195
 196                         if (atomic_read(&n->refcnt) != 1) {
 197                                 /* The most unpleasant situation.
 198                                    We must destroy neighbour entry,
 199                                    but someone still uses it.
 200
 201                                    The destroy will be delayed until
 202                                    the last user releases us, but
 203                                    we must kill timers etc. and move
 204                                    it to safe state.
 205                                  */
 206                                 skb_queue_purge(&n->arp_queue);
 207                                 n->output = neigh_blackhole;
 208                                 if (n->nud_state & NUD_VALID)
 209                                         n->nud_state = NUD_NOARP;
 210                                 else
 211                                         n->nud_state = NUD_NONE;
 212                                 NEIGH_PRINTK2("neigh %p is stray.\n", n);
 213                         }
 214                         write_unlock(&n->lock);
 215                         neigh_release(n);
 216                 }
 217         }
 218 }
 219
 220 void neigh_changeaddr(struct neigh_table *tbl, struct net_device *dev)
 221 {
 222         write_lock_bh(&tbl->lock);
 223         neigh_flush_dev(tbl, dev);
 224         write_unlock_bh(&tbl->lock);
 225 }
 226
 227 int neigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
 228 {
 229         write_lock_bh(&tbl->lock);
 230         neigh_flush_dev(tbl, dev);
 231         pneigh_ifdown(tbl, dev);
 232         write_unlock_bh(&tbl->lock);
 233
 234         del_timer_sync(&tbl->proxy_timer);
 235         pneigh_queue_purge(&tbl->proxy_queue);
 236         return 0;
 237 }
 238
 239 static struct neighbour *neigh_alloc(struct neigh_table *tbl)
 240 {
 241         struct neighbour *n = NULL;
 242         unsigned long now = jiffies;
 243         int entries;
 244
 245         entries = atomic_inc_return(&tbl->entries) - 1;
 246         if (entries >= tbl->gc_thresh3 ||
 247             (entries >= tbl->gc_thresh2 &&
 248              time_after(now, tbl->last_flush + 5 * HZ))) {
 249                 if (!neigh_forced_gc(tbl) &&
 250                     entries >= tbl->gc_thresh3)
 251                         goto out_entries;
 252         }
 253
 254         n = kmem_cache_alloc(tbl->kmem_cachep, SLAB_ATOMIC);
 255         if (!n)
 256                 goto out_entries;
 257
 258         memset(n, 0, tbl->entry_size);
 259
 260         skb_queue_head_init(&n->arp_queue);
 261         rwlock_init(&n->lock);
 262         n->updated        = n->used = now;
 263         n->nud_state      = NUD_NONE;
 264         n->output         = neigh_blackhole;
 265         n->parms          = neigh_parms_clone(&tbl->parms);
 266         init_timer(&n->timer);
 267         n->timer.function = neigh_timer_handler;
 268         n->timer.data     = (unsigned long)n;
 269
 270         NEIGH_CACHE_STAT_INC(tbl, allocs);
 271         n->tbl            = tbl;
 272         atomic_set(&n->refcnt, 1);
 273         n->dead           = 1;
 274 out:
 275         return n;
 276
 277 out_entries:
 278         atomic_dec(&tbl->entries);
 279         goto out;
 280 }
 281
 282 static struct neighbour **neigh_hash_alloc(unsigned int entries)
 283 {
 284         unsigned long size = entries * sizeof(struct neighbour *);
 285         struct neighbour **ret;
 286
 287         if (size <= PAGE_SIZE) {
 288                 ret = kzalloc(size, GFP_ATOMIC);
 289         } else {
 290                 ret = (struct neighbour **)
 291                       __get_free_pages(GFP_ATOMIC|__GFP_ZERO, get_order(size));
 292         }
 293         return ret;
 294 }
 295
 296 static void neigh_hash_free(struct neighbour **hash, unsigned int entries)
 297 {
 298         unsigned long size = entries * sizeof(struct neighbour *);
 299
 300         if (size <= PAGE_SIZE)
 301                 kfree(hash);
 302         else
 303                 free_pages((unsigned long)hash, get_order(size));
 304 }
 305
 306 static void neigh_hash_grow(struct neigh_table *tbl, unsigned long new_entries)
 307 {
 308         struct neighbour **new_hash, **old_hash;
 309         unsigned int i, new_hash_mask, old_entries;
 310
 311         NEIGH_CACHE_STAT_INC(tbl, hash_grows);
 312
 313         BUG_ON(new_entries & (new_entries - 1));
 314         new_hash = neigh_hash_alloc(new_entries);
 315         if (!new_hash)
 316                 return;
 317
 318         old_entries = tbl->hash_mask + 1;
 319         new_hash_mask = new_entries - 1;
 320         old_hash = tbl->hash_buckets;
 321
 322         get_random_bytes(&tbl->hash_rnd, sizeof(tbl->hash_rnd));
 323         for (i = 0; i < old_entries; i++) {
 324                 struct neighbour *n, *next;
 325
 326                 for (n = old_hash[i]; n; n = next) {
 327                         unsigned int hash_val = tbl->hash(n->primary_key, n->dev);
 328
 329                         hash_val &= new_hash_mask;
 330                         next = n->next;
 331
 332                         n->next = new_hash[hash_val];
 333                         new_hash[hash_val] = n;
 334                 }
 335         }
 336         tbl->hash_buckets = new_hash;
 337         tbl->hash_mask = new_hash_mask;
 338
 339         neigh_hash_free(old_hash, old_entries);
 340 }
 341
 342 struct neighbour *neigh_lookup(struct neigh_table *tbl, const void *pkey,
 343                                struct net_device *dev)
 344 {
 345         struct neighbour *n;
 346         int key_len = tbl->key_len;
 347         u32 hash_val = tbl->hash(pkey, dev) & tbl->hash_mask;
 348
 349         NEIGH_CACHE_STAT_INC(tbl, lookups);
 350
 351         read_lock_bh(&tbl->lock);
 352         for (n = tbl->hash_buckets[hash_val]; n; n = n->next) {
 353                 if (dev == n->dev && !memcmp(n->primary_key, pkey, key_len)) {
 354                         neigh_hold(n);
 355                         NEIGH_CACHE_STAT_INC(tbl, hits);
 356                         break;
 357                 }
 358         }
 359         read_unlock_bh(&tbl->lock);
 360         return n;
 361 }
 362
 363 struct neighbour *neigh_lookup_nodev(struct neigh_table *tbl, const void *pkey)
 364 {
 365         struct neighbour *n;
 366         int key_len = tbl->key_len;
 367         u32 hash_val = tbl->hash(pkey, NULL) & tbl->hash_mask;
 368
 369         NEIGH_CACHE_STAT_INC(tbl, lookups);
 370
 371         read_lock_bh(&tbl->lock);
 372         for (n = tbl->hash_buckets[hash_val]; n; n = n->next) {
 373                 if (!memcmp(n->primary_key, pkey, key_len)) {
 374                         neigh_hold(n);
 375                         NEIGH_CACHE_STAT_INC(tbl, hits);
 376                         break;
 377                 }
 378         }
 379         read_unlock_bh(&tbl->lock);
 380         return n;
 381 }
 382
 383 struct neighbour *neigh_create(struct neigh_table *tbl, const void *pkey,
 384                                struct net_device *dev)
 385 {
 386         u32 hash_val;
 387         int key_len = tbl->key_len;
 388         int error;
 389         struct neighbour *n1, *rc, *n = neigh_alloc(tbl);
 390
 391         if (!n) {
 392                 rc = ERR_PTR(-ENOBUFS);
 393                 goto out;
 394         }
 395
 396         memcpy(n->primary_key, pkey, key_len);
 397         n->dev = dev;
 398         dev_hold(dev);
 399
 400         /* Protocol specific setup. */
 401         if (tbl->constructor && (error = tbl->constructor(n)) < 0) {
 402                 rc = ERR_PTR(error);
 403                 goto out_neigh_release;
 404         }
 405
 406         /* Device specific setup. */
 407         if (n->parms->neigh_setup &&
 408             (error = n->parms->neigh_setup(n)) < 0) {
 409                 rc = ERR_PTR(error);
 410                 goto out_neigh_release;
 411         }
 412
 413         n->confirmed = jiffies - (n->parms->base_reachable_time << 1);
 414
 415         write_lock_bh(&tbl->lock);
 416
 417         if (atomic_read(&tbl->entries) > (tbl->hash_mask + 1))
 418                 neigh_hash_grow(tbl, (tbl->hash_mask + 1) << 1);
 419
 420         hash_val = tbl->hash(pkey, dev) & tbl->hash_mask;
 421
 422         if (n->parms->dead) {
 423                 rc = ERR_PTR(-EINVAL);
 424                 goto out_tbl_unlock;
 425         }
 426
 427         for (n1 = tbl->hash_buckets[hash_val]; n1; n1 = n1->next) {
 428                 if (dev == n1->dev && !memcmp(n1->primary_key, pkey, key_len)) {
 429                         neigh_hold(n1);
 430                         rc = n1;
 431                         goto out_tbl_unlock;
 432                 }
 433         }
 434
 435         n->next = tbl->hash_buckets[hash_val];
 436         tbl->hash_buckets[hash_val] = n;
 437         n->dead = 0;
 438         neigh_hold(n);
 439         write_unlock_bh(&tbl->lock);
 440         NEIGH_PRINTK2("neigh %p is created.\n", n);
 441         rc = n;
 442 out:
 443         return rc;
 444 out_tbl_unlock:
 445         write_unlock_bh(&tbl->lock);
 446 out_neigh_release:
 447         neigh_release(n);
 448         goto out;
 449 }
 450
 451 struct pneigh_entry * pneigh_lookup(struct neigh_table *tbl, const void *pkey,
 452                                     struct net_device *dev, int creat)
 453 {
 454         struct pneigh_entry *n;
 455         int key_len = tbl->key_len;
 456         u32 hash_val = *(u32 *)(pkey + key_len - 4);
 457
 458         hash_val ^= (hash_val >> 16);
 459         hash_val ^= hash_val >> 8;
 460         hash_val ^= hash_val >> 4;
 461         hash_val &= PNEIGH_HASHMASK;
 462
 463         read_lock_bh(&tbl->lock);
 464
 465         for (n = tbl->phash_buckets[hash_val]; n; n = n->next) {
 466                 if (!memcmp(n->key, pkey, key_len) &&
 467                     (n->dev == dev || !n->dev)) {
 468                         read_unlock_bh(&tbl->lock);
 469                         goto out;
 470                 }
 471         }
 472         read_unlock_bh(&tbl->lock);
 473         n = NULL;
 474         if (!creat)
 475                 goto out;
 476
 477         n = kmalloc(sizeof(*n) + key_len, GFP_KERNEL);
 478         if (!n)
 479                 goto out;
 480
 481         memcpy(n->key, pkey, key_len);
 482         n->dev = dev;
 483         if (dev)
 484                 dev_hold(dev);
 485
 486         if (tbl->pconstructor && tbl->pconstructor(n)) {
 487                 if (dev)
 488                         dev_put(dev);
 489                 kfree(n);
 490                 n = NULL;
 491                 goto out;
 492         }
 493
 494         write_lock_bh(&tbl->lock);
 495         n->next = tbl->phash_buckets[hash_val];
 496         tbl->phash_buckets[hash_val] = n;
 497         write_unlock_bh(&tbl->lock);
 498 out:
 499         return n;
 500 }
 501
 502
 503 int pneigh_delete(struct neigh_table *tbl, const void *pkey,
 504                   struct net_device *dev)
 505 {
 506         struct pneigh_entry *n, **np;
 507         int key_len = tbl->key_len;
 508         u32 hash_val = *(u32 *)(pkey + key_len - 4);
 509
 510         hash_val ^= (hash_val >> 16);
 511         hash_val ^= hash_val >> 8;
 512         hash_val ^= hash_val >> 4;
 513         hash_val &= PNEIGH_HASHMASK;
 514
 515         write_lock_bh(&tbl->lock);
 516         for (np = &tbl->phash_buckets[hash_val]; (n = *np) != NULL;
 517              np = &n->next) {
 518                 if (!memcmp(n->key, pkey, key_len) && n->dev == dev) {
 519                         *np = n->next;
 520                         write_unlock_bh(&tbl->lock);
 521                         if (tbl->pdestructor)
 522                                 tbl->pdestructor(n);
 523                         if (n->dev)
 524                                 dev_put(n->dev);
 525                         kfree(n);
 526                         return 0;
 527                 }
 528         }
 529         write_unlock_bh(&tbl->lock);
 530         return -ENOENT;
 531 }
 532
 533 static int pneigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
 534 {
 535         struct pneigh_entry *n, **np;
 536         u32 h;
 537
 538         for (h = 0; h <= PNEIGH_HASHMASK; h++) {
 539                 np = &tbl->phash_buckets[h];
 540                 while ((n = *np) != NULL) {
 541                         if (!dev || n->dev == dev) {
 542                                 *np = n->next;
 543                                 if (tbl->pdestructor)
 544                                         tbl->pdestructor(n);
 545                                 if (n->dev)
 546                                         dev_put(n->dev);
 547                                 kfree(n);
 548                                 continue;
 549                         }
 550                         np = &n->next;
 551                 }
 552         }
 553         return -ENOENT;
 554 }
 555
 556
 557 /*
 558  *      neighbour must already be out of the table;
 559  *
 560  */
 561 void neigh_destroy(struct neighbour *neigh)
 562 {
 563         struct hh_cache *hh;
 564
 565         NEIGH_CACHE_STAT_INC(neigh->tbl, destroys);
 566
 567         if (!neigh->dead) {
 568                 printk(KERN_WARNING
 569                        "Destroying alive neighbour %p\n", neigh);
 570                 dump_stack();
 571                 return;
 572         }
 573
 574         if (neigh_del_timer(neigh))
 575                 printk(KERN_WARNING "Impossible event.\n");
 576
 577         while ((hh = neigh->hh) != NULL) {
 578                 neigh->hh = hh->hh_next;
 579                 hh->hh_next = NULL;
 580                 write_lock_bh(&hh->hh_lock);
 581                 hh->hh_output = neigh_blackhole;
 582                 write_unlock_bh(&hh->hh_lock);
 583                 if (atomic_dec_and_test(&hh->hh_refcnt))
 584                         kfree(hh);
 585         }
 586
 587         if (neigh->parms->neigh_destructor)
 588                 (neigh->parms->neigh_destructor)(neigh);
 589
 590         skb_queue_purge(&neigh->arp_queue);
 591
 592         dev_put(neigh->dev);
 593         neigh_parms_put(neigh->parms);
 594
 595         NEIGH_PRINTK2("neigh %p is destroyed.\n", neigh);
 596
 597         atomic_dec(&neigh->tbl->entries);
 598         kmem_cache_free(neigh->tbl->kmem_cachep, neigh);
 599 }
 600
 601 /* Neighbour state is suspicious;
 602    disable fast path.
 603
 604    Called with write_locked neigh.
 605  */
 606 static void neigh_suspect(struct neighbour *neigh)
 607 {
 608         struct hh_cache *hh;
 609
 610         NEIGH_PRINTK2("neigh %p is suspected.\n", neigh);
 611
 612         neigh->output = neigh->ops->output;
 613
 614         for (hh = neigh->hh; hh; hh = hh->hh_next)
 615                 hh->hh_output = neigh->ops->output;
 616 }
 617
 618 /* Neighbour state is OK;
 619    enable fast path.
 620
 621    Called with write_locked neigh.
 622  */
 623 static void neigh_connect(struct neighbour *neigh)
 624 {
 625         struct hh_cache *hh;
 626
 627         NEIGH_PRINTK2("neigh %p is connected.\n", neigh);
 628
 629         neigh->output = neigh->ops->connected_output;
 630
 631         for (hh = neigh->hh; hh; hh = hh->hh_next)
 632                 hh->hh_output = neigh->ops->hh_output;
 633 }
 634
 635 static void neigh_periodic_timer(unsigned long arg)
 636 {
 637         struct neigh_table *tbl = (struct neigh_table *)arg;
 638         struct neighbour *n, **np;
 639         unsigned long expire, now = jiffies;
 640
 641         NEIGH_CACHE_STAT_INC(tbl, periodic_gc_runs);
 642
 643         write_lock(&tbl->lock);
 644
 645         /*
 646          *      periodically recompute ReachableTime from random function
 647          */
 648
 649         if (time_after(now, tbl->last_rand + 300 * HZ)) {
 650                 struct neigh_parms *p;
 651                 tbl->last_rand = now;
 652                 for (p = &tbl->parms; p; p = p->next)
 653                         p->reachable_time =
 654                                 neigh_rand_reach_time(p->base_reachable_time);
 655         }
 656
 657         np = &tbl->hash_buckets[tbl->hash_chain_gc];
 658         tbl->hash_chain_gc = ((tbl->hash_chain_gc + 1) & tbl->hash_mask);
 659
 660         while ((n = *np) != NULL) {
 661                 unsigned int state;
 662
 663                 write_lock(&n->lock);
 664
 665                 state = n->nud_state;
 666                 if (state & (NUD_PERMANENT | NUD_IN_TIMER)) {
 667                         write_unlock(&n->lock);
 668                         goto next_elt;
 669                 }
 670
 671                 if (time_before(n->used, n->confirmed))
 672                         n->used = n->confirmed;
 673
 674                 if (atomic_read(&n->refcnt) == 1 &&
 675                     (state == NUD_FAILED ||
 676                      time_after(now, n->used + n->parms->gc_staletime))) {
 677                         *np = n->next;
 678                         n->dead = 1;
 679                         write_unlock(&n->lock);
 680                         neigh_release(n);
 681                         continue;
 682                 }
 683                 write_unlock(&n->lock);
 684
 685 next_elt:
 686                 np = &n->next;
 687         }
 688
 689         /* Cycle through all hash buckets every base_reachable_time/2 ticks.
 690          * ARP entry timeouts range from 1/2 base_reachable_time to 3/2
 691          * base_reachable_time.
 692          */
 693         expire = tbl->parms.base_reachable_time >> 1;
 694         expire /= (tbl->hash_mask + 1);
 695         if (!expire)
 696                 expire = 1;
 697
 698         mod_timer(&tbl->gc_timer, now + expire);
 699
 700         write_unlock(&tbl->lock);
 701 }
 702
 703 static __inline__ int neigh_max_probes(struct neighbour *n)
 704 {
 705         struct neigh_parms *p = n->parms;
 706         return (n->nud_state & NUD_PROBE ?
 707                 p->ucast_probes :
 708                 p->ucast_probes + p->app_probes + p->mcast_probes);
 709 }
 710
 711 static inline void neigh_add_timer(struct neighbour *n, unsigned long when)
 712 {
 713         if (unlikely(mod_timer(&n->timer, when))) {
 714                 printk("NEIGH: BUG, double timer add, state is %x\n",
 715                        n->nud_state);
 716                 dump_stack();
 717         }
 718 }
 719
 720 /* Called when a timer expires for a neighbour entry. */
 721
 722 static void neigh_timer_handler(unsigned long arg)
 723 {
 724         unsigned long now, next;
 725         struct neighbour *neigh = (struct neighbour *)arg;
 726         unsigned state;
 727         int notify = 0;
 728
 729         write_lock(&neigh->lock);
 730
 731         state = neigh->nud_state;
 732         now = jiffies;
 733         next = now + HZ;
 734
 735         if (!(state & NUD_IN_TIMER)) {
 736 #ifndef CONFIG_SMP
 737                 printk(KERN_WARNING "neigh: timer & !nud_in_timer\n");
 738 #endif
 739                 goto out;
 740         }
 741
 742         if (state & NUD_REACHABLE) {
 743                 if (time_before_eq(now,
 744                                    neigh->confirmed + neigh->parms->reachable_time)) {
 745                         NEIGH_PRINTK2("neigh %p is still alive.\n", neigh);
 746                         next = neigh->confirmed + neigh->parms->reachable_time;
 747                 } else if (time_before_eq(now,
 748                                           neigh->used + neigh->parms->delay_probe_time)) {
 749                         NEIGH_PRINTK2("neigh %p is delayed.\n", neigh);
 750                         neigh->nud_state = NUD_DELAY;
 751                         neigh->updated = jiffies;
 752                         neigh_suspect(neigh);
 753                         next = now + neigh->parms->delay_probe_time;
 754                 } else {
 755                         NEIGH_PRINTK2("neigh %p is suspected.\n", neigh);
 756                         neigh->nud_state = NUD_STALE;
 757                         neigh->updated = jiffies;
 758                         neigh_suspect(neigh);
 759                         notify = 1;
 760                 }
 761         } else if (state & NUD_DELAY) {
 762                 if (time_before_eq(now,
 763                                    neigh->confirmed + neigh->parms->delay_probe_time)) {
 764                         NEIGH_PRINTK2("neigh %p is now reachable.\n", neigh);
 765                         neigh->nud_state = NUD_REACHABLE;
 766                         neigh->updated = jiffies;
 767                         neigh_connect(neigh);
 768                         notify = 1;
 769                         next = neigh->confirmed + neigh->parms->reachable_time;
 770                 } else {
 771                         NEIGH_PRINTK2("neigh %p is probed.\n", neigh);
 772                         neigh->nud_state = NUD_PROBE;
 773                         neigh->updated = jiffies;
 774                         atomic_set(&neigh->probes, 0);
 775                         next = now + neigh->parms->retrans_time;
 776                 }
 777         } else {
 778                 /* NUD_PROBE|NUD_INCOMPLETE */
 779                 next = now + neigh->parms->retrans_time;
 780         }
 781
 782         if ((neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) &&
 783             atomic_read(&neigh->probes) >= neigh_max_probes(neigh)) {
 784                 struct sk_buff *skb;
 785
 786                 neigh->nud_state = NUD_FAILED;
 787                 neigh->updated = jiffies;
 788                 notify = 1;
 789                 NEIGH_CACHE_STAT_INC(neigh->tbl, res_failed);
 790                 NEIGH_PRINTK2("neigh %p is failed.\n", neigh);
 791
 792                 /* It is very thin place. report_unreachable is very complicated
 793                    routine. Particularly, it can hit the same neighbour entry!
 794
 795                    So that, we try to be accurate and avoid dead loop. --ANK
 796                  */
 797                 while (neigh->nud_state == NUD_FAILED &&
 798                        (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
 799                         write_unlock(&neigh->lock);
 800                         neigh->ops->error_report(neigh, skb);
 801                         write_lock(&neigh->lock);
 802                 }
 803                 skb_queue_purge(&neigh->arp_queue);
 804         }
 805
 806         if (neigh->nud_state & NUD_IN_TIMER) {
 807                 if (time_before(next, jiffies + HZ/2))
 808                         next = jiffies + HZ/2;
 809                 if (!mod_timer(&neigh->timer, next))
 810                         neigh_hold(neigh);
 811         }
 812         if (neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) {
 813                 struct sk_buff *skb = skb_peek(&neigh->arp_queue);
 814                 /* keep skb alive even if arp_queue overflows */
 815                 if (skb)
 816                         skb_get(skb);
 817                 write_unlock(&neigh->lock);
 818                 neigh->ops->solicit(neigh, skb);
 819                 atomic_inc(&neigh->probes);
 820                 if (skb)
 821                         kfree_skb(skb);
 822         } else {
 823 out:
 824                 write_unlock(&neigh->lock);
 825         }
 826         if (notify)
 827                 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, neigh);
 828
 829 #ifdef CONFIG_ARPD
 830         if (notify && neigh->parms->app_probes)
 831                 neigh_app_notify(neigh);
 832 #endif
 833         neigh_release(neigh);
 834 }
 835
 836 int __neigh_event_send(struct neighbour *neigh, struct sk_buff *skb)
 837 {
 838         int rc;
 839         unsigned long now;
 840
 841         write_lock_bh(&neigh->lock);
 842
 843         rc = 0;
 844         if (neigh->nud_state & (NUD_CONNECTED | NUD_DELAY | NUD_PROBE))
 845                 goto out_unlock_bh;
 846
 847         now = jiffies;
 848
 849         if (!(neigh->nud_state & (NUD_STALE | NUD_INCOMPLETE))) {
 850                 if (neigh->parms->mcast_probes + neigh->parms->app_probes) {
 851                         atomic_set(&neigh->probes, neigh->parms->ucast_probes);
 852                         neigh->nud_state     = NUD_INCOMPLETE;
 853                         neigh->updated = jiffies;
 854                         neigh_hold(neigh);
 855                         neigh_add_timer(neigh, now + 1);
 856                 } else {
 857                         neigh->nud_state = NUD_FAILED;
 858                         neigh->updated = jiffies;
 859                         write_unlock_bh(&neigh->lock);
 860
 861                         if (skb)
 862                                 kfree_skb(skb);
 863                         return 1;
 864                 }
 865         } else if (neigh->nud_state & NUD_STALE) {
 866                 NEIGH_PRINTK2("neigh %p is delayed.\n", neigh);
 867                 neigh_hold(neigh);
 868                 neigh->nud_state = NUD_DELAY;
 869                 neigh->updated = jiffies;
 870                 neigh_add_timer(neigh,
 871                                 jiffies + neigh->parms->delay_probe_time);
 872         }
 873
 874         if (neigh->nud_state == NUD_INCOMPLETE) {
 875                 if (skb) {
 876                         if (skb_queue_len(&neigh->arp_queue) >=
 877                             neigh->parms->queue_len) {
 878                                 struct sk_buff *buff;
 879                                 buff = neigh->arp_queue.next;
 880                                 __skb_unlink(buff, &neigh->arp_queue);
 881                                 kfree_skb(buff);
 882                         }
 883                         __skb_queue_tail(&neigh->arp_queue, skb);
 884                 }
 885                 rc = 1;
 886         }
 887 out_unlock_bh:
 888         write_unlock_bh(&neigh->lock);
 889         return rc;
 890 }
 891
 892 static __inline__ void neigh_update_hhs(struct neighbour *neigh)
 893 {
 894         struct hh_cache *hh;
 895         void (*update)(struct hh_cache*, struct net_device*, unsigned char *) =
 896                 neigh->dev->header_cache_update;
 897
 898         if (update) {
 899                 for (hh = neigh->hh; hh; hh = hh->hh_next) {
 900                         write_lock_bh(&hh->hh_lock);
 901                         update(hh, neigh->dev, neigh->ha);
 902                         write_unlock_bh(&hh->hh_lock);
 903                 }
 904         }
 905 }
 906
 907
 908
 909 /* Generic update routine.
 910    -- lladdr is new lladdr or NULL, if it is not supplied.
 911    -- new    is new state.
 912    -- flags
 913         NEIGH_UPDATE_F_OVERRIDE allows to override existing lladdr,
 914                                 if it is different.
 915         NEIGH_UPDATE_F_WEAK_OVERRIDE will suspect existing "connected"
 916                                 lladdr instead of overriding it
 917                                 if it is different.
 918                                 It also allows to retain current state
 919                                 if lladdr is unchanged.
 920         NEIGH_UPDATE_F_ADMIN    means that the change is administrative.
 921
 922         NEIGH_UPDATE_F_OVERRIDE_ISROUTER allows to override existing
 923                                 NTF_ROUTER flag.
 924         NEIGH_UPDATE_F_ISROUTER indicates if the neighbour is known as
 925                                 a router.
 926
 927    Caller MUST hold reference count on the entry.
 928  */
 929
 930 int neigh_update(struct neighbour *neigh, const u8 *lladdr, u8 new,
 931                  u32 flags)
 932 {
 933         u8 old;
 934         int err;
 935         int notify = 0;
 936         struct net_device *dev;
 937         int update_isrouter = 0;
 938
 939         write_lock_bh(&neigh->lock);
 940
 941         dev    = neigh->dev;
 942         old    = neigh->nud_state;
 943         err    = -EPERM;
 944
 945         if (!(flags & NEIGH_UPDATE_F_ADMIN) &&
 946             (old & (NUD_NOARP | NUD_PERMANENT)))
 947                 goto out;
 948
 949         if (!(new & NUD_VALID)) {
 950                 neigh_del_timer(neigh);
 951                 if (old & NUD_CONNECTED)
 952                         neigh_suspect(neigh);
 953                 neigh->nud_state = new;
 954                 err = 0;
 955                 notify = old & NUD_VALID;
 956                 goto out;
 957         }
 958
 959         /* Compare new lladdr with cached one */
 960         if (!dev->addr_len) {
 961                 /* First case: device needs no address. */
 962                 lladdr = neigh->ha;
 963         } else if (lladdr) {
 964                 /* The second case: if something is already cached
 965                    and a new address is proposed:
 966                    - compare new & old
 967                    - if they are different, check override flag
 968                  */
 969                 if ((old & NUD_VALID) &&
 970                     !memcmp(lladdr, neigh->ha, dev->addr_len))
 971                         lladdr = neigh->ha;
 972         } else {
 973                 /* No address is supplied; if we know something,
 974                    use it, otherwise discard the request.
 975                  */
 976                 err = -EINVAL;
 977                 if (!(old & NUD_VALID))
 978                         goto out;
 979                 lladdr = neigh->ha;
 980         }
 981
 982         if (new & NUD_CONNECTED)
 983                 neigh->confirmed = jiffies;
 984         neigh->updated = jiffies;
 985
 986         /* If entry was valid and address is not changed,
 987            do not change entry state, if new one is STALE.
 988          */
 989         err = 0;
 990         update_isrouter = flags & NEIGH_UPDATE_F_OVERRIDE_ISROUTER;
 991         if (old & NUD_VALID) {
 992                 if (lladdr != neigh->ha && !(flags & NEIGH_UPDATE_F_OVERRIDE)) {
 993                         update_isrouter = 0;
 994                         if ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) &&
 995                             (old & NUD_CONNECTED)) {
 996                                 lladdr = neigh->ha;
 997                                 new = NUD_STALE;
 998                         } else
 999                                 goto out;
1000                 } else {
1001                         if (lladdr == neigh->ha && new == NUD_STALE &&
1002                             ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) ||
1003                              (old & NUD_CONNECTED))
1004                             )
1005                                 new = old;
1006                 }
1007         }
1008
1009         if (new != old) {
1010                 neigh_del_timer(neigh);
1011                 if (new & NUD_IN_TIMER) {
1012                         neigh_hold(neigh);
1013                         neigh_add_timer(neigh, (jiffies +
1014                                                 ((new & NUD_REACHABLE) ?
1015                                                  neigh->parms->reachable_time :
1016                                                  0)));
1017                 }
1018                 neigh->nud_state = new;
1019         }
1020
1021         if (lladdr != neigh->ha) {
1022                 memcpy(&neigh->ha, lladdr, dev->addr_len);
1023                 neigh_update_hhs(neigh);
1024                 if (!(new & NUD_CONNECTED))
1025                         neigh->confirmed = jiffies -
1026                                       (neigh->parms->base_reachable_time << 1);
1027                 notify = 1;
1028         }
1029         if (new == old)
1030                 goto out;
1031         if (new & NUD_CONNECTED)
1032                 neigh_connect(neigh);
1033         else
1034                 neigh_suspect(neigh);
1035         if (!(old & NUD_VALID)) {
1036                 struct sk_buff *skb;
1037
1038                 /* Again: avoid dead loop if something went wrong */
1039
1040                 while (neigh->nud_state & NUD_VALID &&
1041                        (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
1042                         struct neighbour *n1 = neigh;
1043                         write_unlock_bh(&neigh->lock);
1044                         /* On shaper/eql skb->dst->neighbour != neigh :( */
1045                         if (skb->dst && skb->dst->neighbour)
1046                                 n1 = skb->dst->neighbour;
1047                         n1->output(skb);
1048                         write_lock_bh(&neigh->lock);
1049                 }
1050                 skb_queue_purge(&neigh->arp_queue);
1051         }
1052 out:
1053         if (update_isrouter) {
1054                 neigh->flags = (flags & NEIGH_UPDATE_F_ISROUTER) ?
1055                         (neigh->flags | NTF_ROUTER) :
1056                         (neigh->flags & ~NTF_ROUTER);
1057         }
1058         write_unlock_bh(&neigh->lock);
1059
1060         if (notify)
1061                 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, neigh);
1062 #ifdef CONFIG_ARPD
1063         if (notify && neigh->parms->app_probes)
1064                 neigh_app_notify(neigh);
1065 #endif
1066         return err;
1067 }
1068
1069 struct neighbour *neigh_event_ns(struct neigh_table *tbl,
1070                                  u8 *lladdr, void *saddr,
1071                                  struct net_device *dev)
1072 {
1073         struct neighbour *neigh = __neigh_lookup(tbl, saddr, dev,
1074                                                  lladdr || !dev->addr_len);
1075         if (neigh)
1076                 neigh_update(neigh, lladdr, NUD_STALE,
1077                              NEIGH_UPDATE_F_OVERRIDE);
1078         return neigh;
1079 }
1080
1081 static void neigh_hh_init(struct neighbour *n, struct dst_entry *dst,
1082                           u16 protocol)
1083 {
1084         struct hh_cache *hh;
1085         struct net_device *dev = dst->dev;
1086
1087         for (hh = n->hh; hh; hh = hh->hh_next)
1088                 if (hh->hh_type == protocol)
1089                         break;
1090
1091         if (!hh && (hh = kzalloc(sizeof(*hh), GFP_ATOMIC)) != NULL) {
1092                 rwlock_init(&hh->hh_lock);
1093                 hh->hh_type = protocol;
1094                 atomic_set(&hh->hh_refcnt, 0);
1095                 hh->hh_next = NULL;
1096                 if (dev->hard_header_cache(n, hh)) {
1097                         kfree(hh);
1098                         hh = NULL;
1099                 } else {
1100                         atomic_inc(&hh->hh_refcnt);
1101                         hh->hh_next = n->hh;
1102                         n->hh       = hh;
1103                         if (n->nud_state & NUD_CONNECTED)
1104                                 hh->hh_output = n->ops->hh_output;
1105                         else
1106                                 hh->hh_output = n->ops->output;
1107                 }
1108         }
1109         if (hh) {
1110                 atomic_inc(&hh->hh_refcnt);
1111                 dst->hh = hh;
1112         }
1113 }
1114
1115 /* This function can be used in contexts, where only old dev_queue_xmit
1116    worked, f.e. if you want to override normal output path (eql, shaper),
1117    but resolution is not made yet.
1118  */
1119
1120 int neigh_compat_output(struct sk_buff *skb)
1121 {
1122         struct net_device *dev = skb->dev;
1123
1124         __skb_pull(skb, skb->nh.raw - skb->data);
1125
1126         if (dev->hard_header &&
1127             dev->hard_header(skb, dev, ntohs(skb->protocol), NULL, NULL,
1128                              skb->len) < 0 &&
1129             dev->rebuild_header(skb))
1130                 return 0;
1131
1132         return dev_queue_xmit(skb);
1133 }
1134
1135 /* Slow and careful. */
1136
1137 int neigh_resolve_output(struct sk_buff *skb)
1138 {
1139         struct dst_entry *dst = skb->dst;
1140         struct neighbour *neigh;
1141         int rc = 0;
1142
1143         if (!dst || !(neigh = dst->neighbour))
1144                 goto discard;
1145
1146         __skb_pull(skb, skb->nh.raw - skb->data);
1147
1148         if (!neigh_event_send(neigh, skb)) {
1149                 int err;
1150                 struct net_device *dev = neigh->dev;
1151                 if (dev->hard_header_cache && !dst->hh) {
1152                         write_lock_bh(&neigh->lock);
1153                         if (!dst->hh)
1154                                 neigh_hh_init(neigh, dst, dst->ops->protocol);
1155                         err = dev->hard_header(skb, dev, ntohs(skb->protocol),
1156                                                neigh->ha, NULL, skb->len);
1157                         write_unlock_bh(&neigh->lock);
1158                 } else {
1159                         read_lock_bh(&neigh->lock);
1160                         err = dev->hard_header(skb, dev, ntohs(skb->protocol),
1161                                                neigh->ha, NULL, skb->len);
1162                         read_unlock_bh(&neigh->lock);
1163                 }
1164                 if (err >= 0)
1165                         rc = neigh->ops->queue_xmit(skb);
1166                 else
1167                         goto out_kfree_skb;
1168         }
1169 out:
1170         return rc;
1171 discard:
1172         NEIGH_PRINTK1("neigh_resolve_output: dst=%p neigh=%p\n",
1173                       dst, dst ? dst->neighbour : NULL);
1174 out_kfree_skb:
1175         rc = -EINVAL;
1176         kfree_skb(skb);
1177         goto out;
1178 }
1179
1180 /* As fast as possible without hh cache */
1181
1182 int neigh_connected_output(struct sk_buff *skb)
1183 {
1184         int err;
1185         struct dst_entry *dst = skb->dst;
1186         struct neighbour *neigh = dst->neighbour;
1187         struct net_device *dev = neigh->dev;
1188
1189         __skb_pull(skb, skb->nh.raw - skb->data);
1190
1191         read_lock_bh(&neigh->lock);
1192         err = dev->hard_header(skb, dev, ntohs(skb->protocol),
1193                                neigh->ha, NULL, skb->len);
1194         read_unlock_bh(&neigh->lock);
1195         if (err >= 0)
1196                 err = neigh->ops->queue_xmit(skb);
1197         else {
1198                 err = -EINVAL;
1199                 kfree_skb(skb);
1200         }
1201         return err;
1202 }
1203
1204 static void neigh_proxy_process(unsigned long arg)
1205 {
1206         struct neigh_table *tbl = (struct neigh_table *)arg;
1207         long sched_next = 0;
1208         unsigned long now = jiffies;
1209         struct sk_buff *skb;
1210
1211         spin_lock(&tbl->proxy_queue.lock);
1212
1213         skb = tbl->proxy_queue.next;
1214
1215         while (skb != (struct sk_buff *)&tbl->proxy_queue) {
1216                 struct sk_buff *back = skb;
1217                 long tdif = NEIGH_CB(back)->sched_next - now;
1218
1219                 skb = skb->next;
1220                 if (tdif <= 0) {
1221                         struct net_device *dev = back->dev;
1222                         __skb_unlink(back, &tbl->proxy_queue);
1223                         if (tbl->proxy_redo && netif_running(dev))
1224                                 tbl->proxy_redo(back);
1225                         else
1226                                 kfree_skb(back);
1227
1228                         dev_put(dev);
1229                 } else if (!sched_next || tdif < sched_next)
1230                         sched_next = tdif;
1231         }
1232         del_timer(&tbl->proxy_timer);
1233         if (sched_next)
1234                 mod_timer(&tbl->proxy_timer, jiffies + sched_next);
1235         spin_unlock(&tbl->proxy_queue.lock);
1236 }
1237
1238 void pneigh_enqueue(struct neigh_table *tbl, struct neigh_parms *p,
1239                     struct sk_buff *skb)
1240 {
1241         unsigned long now = jiffies;
1242         unsigned long sched_next = now + (net_random() % p->proxy_delay);
1243
1244         if (tbl->proxy_queue.qlen > p->proxy_qlen) {
1245                 kfree_skb(skb);
1246                 return;
1247         }
1248
1249         NEIGH_CB(skb)->sched_next = sched_next;
1250         NEIGH_CB(skb)->flags |= LOCALLY_ENQUEUED;
1251
1252         spin_lock(&tbl->proxy_queue.lock);
1253         if (del_timer(&tbl->proxy_timer)) {
1254                 if (time_before(tbl->proxy_timer.expires, sched_next))
1255                         sched_next = tbl->proxy_timer.expires;
1256         }
1257         dst_release(skb->dst);
1258         skb->dst = NULL;
1259         dev_hold(skb->dev);
1260         __skb_queue_tail(&tbl->proxy_queue, skb);
1261         mod_timer(&tbl->proxy_timer, sched_next);
1262         spin_unlock(&tbl->proxy_queue.lock);
1263 }
1264
1265
1266 struct neigh_parms *neigh_parms_alloc(struct net_device *dev,
1267                                       struct neigh_table *tbl)
1268 {
1269         struct neigh_parms *p = kmalloc(sizeof(*p), GFP_KERNEL);
1270
1271         if (p) {
1272                 memcpy(p, &tbl->parms, sizeof(*p));
1273                 p->tbl            = tbl;
1274                 atomic_set(&p->refcnt, 1);
1275                 INIT_RCU_HEAD(&p->rcu_head);
1276                 p->reachable_time =
1277                                 neigh_rand_reach_time(p->base_reachable_time);
1278                 if (dev) {
1279                         if (dev->neigh_setup && dev->neigh_setup(dev, p)) {
1280                                 kfree(p);
1281                                 return NULL;
1282                         }
1283
1284                         dev_hold(dev);
1285                         p->dev = dev;
1286                 }
1287                 p->sysctl_table = NULL;
1288                 write_lock_bh(&tbl->lock);
1289                 p->next         = tbl->parms.next;
1290                 tbl->parms.next = p;
1291                 write_unlock_bh(&tbl->lock);
1292         }
1293         return p;
1294 }
1295
1296 static void neigh_rcu_free_parms(struct rcu_head *head)
1297 {
1298         struct neigh_parms *parms =
1299                 container_of(head, struct neigh_parms, rcu_head);
1300
1301         neigh_parms_put(parms);
1302 }
1303
1304 void neigh_parms_release(struct neigh_table *tbl, struct neigh_parms *parms)
1305 {
1306         struct neigh_parms **p;
1307
1308         if (!parms || parms == &tbl->parms)
1309                 return;
1310         write_lock_bh(&tbl->lock);
1311         for (p = &tbl->parms.next; *p; p = &(*p)->next) {
1312                 if (*p == parms) {
1313                         *p = parms->next;
1314                         parms->dead = 1;
1315                         write_unlock_bh(&tbl->lock);
1316                         if (parms->dev)
1317                                 dev_put(parms->dev);
1318                         call_rcu(&parms->rcu_head, neigh_rcu_free_parms);
1319                         return;
1320                 }
1321         }
1322         write_unlock_bh(&tbl->lock);
1323         NEIGH_PRINTK1("neigh_parms_release: not found\n");
1324 }
1325
1326 void neigh_parms_destroy(struct neigh_parms *parms)
1327 {
1328         kfree(parms);
1329 }
1330
1331 void neigh_table_init_no_netlink(struct neigh_table *tbl)
1332 {
1333         unsigned long now = jiffies;
1334         unsigned long phsize;
1335
1336         atomic_set(&tbl->parms.refcnt, 1);
1337         INIT_RCU_HEAD(&tbl->parms.rcu_head);
1338         tbl->parms.reachable_time =
1339                           neigh_rand_reach_time(tbl->parms.base_reachable_time);
1340
1341         if (!tbl->kmem_cachep)
1342                 tbl->kmem_cachep = kmem_cache_create(tbl->id,
1343                                                      tbl->entry_size,
1344                                                      0, SLAB_HWCACHE_ALIGN,
1345                                                      NULL, NULL);
1346
1347         if (!tbl->kmem_cachep)
1348                 panic("cannot create neighbour cache");
1349
1350         tbl->stats = alloc_percpu(struct neigh_statistics);
1351         if (!tbl->stats)
1352                 panic("cannot create neighbour cache statistics");
1353
1354 #ifdef CONFIG_PROC_FS
1355         tbl->pde = create_proc_entry(tbl->id, 0, proc_net_stat);
1356         if (!tbl->pde)
1357                 panic("cannot create neighbour proc dir entry");
1358         tbl->pde->proc_fops = &neigh_stat_seq_fops;
1359         tbl->pde->data = tbl;
1360 #endif
1361
1362         tbl->hash_mask = 1;
1363         tbl->hash_buckets = neigh_hash_alloc(tbl->hash_mask + 1);
1364
1365         phsize = (PNEIGH_HASHMASK + 1) * sizeof(struct pneigh_entry *);
1366         tbl->phash_buckets = kzalloc(phsize, GFP_KERNEL);
1367
1368         if (!tbl->hash_buckets || !tbl->phash_buckets)
1369                 panic("cannot allocate neighbour cache hashes");
1370
1371         get_random_bytes(&tbl->hash_rnd, sizeof(tbl->hash_rnd));
1372
1373         rwlock_init(&tbl->lock);
1374         init_timer(&tbl->gc_timer);
1375         tbl->gc_timer.data     = (unsigned long)tbl;
1376         tbl->gc_timer.function = neigh_periodic_timer;
1377         tbl->gc_timer.expires  = now + 1;
1378         add_timer(&tbl->gc_timer);
1379
1380         init_timer(&tbl->proxy_timer);
1381         tbl->proxy_timer.data     = (unsigned long)tbl;
1382         tbl->proxy_timer.function = neigh_proxy_process;
1383         skb_queue_head_init(&tbl->proxy_queue);
1384
1385         tbl->last_flush = now;
1386         tbl->last_rand  = now + tbl->parms.reachable_time * 20;
1387 }
1388
1389 void neigh_table_init(struct neigh_table *tbl)
1390 {
1391         struct neigh_table *tmp;
1392
1393         neigh_table_init_no_netlink(tbl);
1394         write_lock(&neigh_tbl_lock);
1395         for (tmp = neigh_tables; tmp; tmp = tmp->next) {
1396                 if (tmp->family == tbl->family)
1397                         break;
1398         }
1399         tbl->next       = neigh_tables;
1400         neigh_tables    = tbl;
1401         write_unlock(&neigh_tbl_lock);
1402
1403         if (unlikely(tmp)) {
1404                 printk(KERN_ERR "NEIGH: Registering multiple tables for "
1405                        "family %d\n", tbl->family);
1406                 dump_stack();
1407         }
1408 }
1409
1410 int neigh_table_clear(struct neigh_table *tbl)
1411 {
1412         struct neigh_table **tp;
1413
1414         /* It is not clean... Fix it to unload IPv6 module safely */
1415         del_timer_sync(&tbl->gc_timer);
1416         del_timer_sync(&tbl->proxy_timer);
1417         pneigh_queue_purge(&tbl->proxy_queue);
1418         neigh_ifdown(tbl, NULL);
1419         if (atomic_read(&tbl->entries))
1420                 printk(KERN_CRIT "neighbour leakage\n");
1421         write_lock(&neigh_tbl_lock);
1422         for (tp = &neigh_tables; *tp; tp = &(*tp)->next) {
1423                 if (*tp == tbl) {
1424                         *tp = tbl->next;
1425                         break;
1426                 }
1427         }
1428         write_unlock(&neigh_tbl_lock);
1429
1430         neigh_hash_free(tbl->hash_buckets, tbl->hash_mask + 1);
1431         tbl->hash_buckets = NULL;
1432
1433         kfree(tbl->phash_buckets);
1434         tbl->phash_buckets = NULL;
1435
1436         free_percpu(tbl->stats);
1437         tbl->stats = NULL;
1438
1439         return 0;
1440 }
1441
1442 int neigh_delete(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
1443 {
1444         struct ndmsg *ndm;
1445         struct nlattr *dst_attr;
1446         struct neigh_table *tbl;
1447         struct net_device *dev = NULL;
1448         int err = -EINVAL;
1449
1450         if (nlmsg_len(nlh) < sizeof(*ndm))
1451                 goto out;
1452
1453         dst_attr = nlmsg_find_attr(nlh, sizeof(*ndm), NDA_DST);
1454         if (dst_attr == NULL)
1455                 goto out;
1456
1457         ndm = nlmsg_data(nlh);
1458         if (ndm->ndm_ifindex) {
1459                 dev = dev_get_by_index(ndm->ndm_ifindex);
1460                 if (dev == NULL) {
1461                         err = -ENODEV;
1462                         goto out;
1463                 }
1464         }
1465
1466         read_lock(&neigh_tbl_lock);
1467         for (tbl = neigh_tables; tbl; tbl = tbl->next) {
1468                 struct neighbour *neigh;
1469
1470                 if (tbl->family != ndm->ndm_family)
1471                         continue;
1472                 read_unlock(&neigh_tbl_lock);
1473
1474                 if (nla_len(dst_attr) < tbl->key_len)
1475                         goto out_dev_put;
1476
1477                 if (ndm->ndm_flags & NTF_PROXY) {
1478                         err = pneigh_delete(tbl, nla_data(dst_attr), dev);
1479                         goto out_dev_put;
1480                 }
1481
1482                 if (dev == NULL)
1483                         goto out_dev_put;
1484
1485                 neigh = neigh_lookup(tbl, nla_data(dst_attr), dev);
1486                 if (neigh == NULL) {
1487                         err = -ENOENT;
1488                         goto out_dev_put;
1489                 }
1490
1491                 err = neigh_update(neigh, NULL, NUD_FAILED,
1492                                    NEIGH_UPDATE_F_OVERRIDE |
1493                                    NEIGH_UPDATE_F_ADMIN);
1494                 neigh_release(neigh);
1495                 goto out_dev_put;
1496         }
1497         read_unlock(&neigh_tbl_lock);
1498         err = -EAFNOSUPPORT;
1499
1500 out_dev_put:
1501         if (dev)
1502                 dev_put(dev);
1503 out:
1504         return err;
1505 }
1506
1507 int neigh_add(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
1508 {
1509         struct ndmsg *ndm;
1510         struct nlattr *tb[NDA_MAX+1];
1511         struct neigh_table *tbl;
1512         struct net_device *dev = NULL;
1513         int err;
1514
1515         err = nlmsg_parse(nlh, sizeof(*ndm), tb, NDA_MAX, NULL);
1516         if (err < 0)
1517                 goto out;
1518
1519         err = -EINVAL;
1520         if (tb[NDA_DST] == NULL)
1521                 goto out;
1522
1523         ndm = nlmsg_data(nlh);
1524         if (ndm->ndm_ifindex) {
1525                 dev = dev_get_by_index(ndm->ndm_ifindex);
1526                 if (dev == NULL) {
1527                         err = -ENODEV;
1528                         goto out;
1529                 }
1530
1531                 if (tb[NDA_LLADDR] && nla_len(tb[NDA_LLADDR]) < dev->addr_len)
1532                         goto out_dev_put;
1533         }
1534
1535         read_lock(&neigh_tbl_lock);
1536         for (tbl = neigh_tables; tbl; tbl = tbl->next) {
1537                 int flags = NEIGH_UPDATE_F_ADMIN | NEIGH_UPDATE_F_OVERRIDE;
1538                 struct neighbour *neigh;
1539                 void *dst, *lladdr;
1540
1541                 if (tbl->family != ndm->ndm_family)
1542                         continue;
1543                 read_unlock(&neigh_tbl_lock);
1544
1545                 if (nla_len(tb[NDA_DST]) < tbl->key_len)
1546                         goto out_dev_put;
1547                 dst = nla_data(tb[NDA_DST]);
1548                 lladdr = tb[NDA_LLADDR] ? nla_data(tb[NDA_LLADDR]) : NULL;
1549
1550                 if (ndm->ndm_flags & NTF_PROXY) {
1551                         err = 0;
1552                         if (pneigh_lookup(tbl, dst, dev, 1) == NULL)
1553                                 err = -ENOBUFS;
1554                         goto out_dev_put;
1555                 }
1556
1557                 if (dev == NULL)
1558                         goto out_dev_put;
1559
1560                 neigh = neigh_lookup(tbl, dst, dev);
1561                 if (neigh == NULL) {
1562                         if (!(nlh->nlmsg_flags & NLM_F_CREATE)) {
1563                                 err = -ENOENT;
1564                                 goto out_dev_put;
1565                         }
1566
1567                         neigh = __neigh_lookup_errno(tbl, dst, dev);
1568                         if (IS_ERR(neigh)) {
1569                                 err = PTR_ERR(neigh);
1570                                 goto out_dev_put;
1571                         }
1572                 } else {
1573                         if (nlh->nlmsg_flags & NLM_F_EXCL) {
1574                                 err = -EEXIST;
1575                                 neigh_release(neigh);
1576                                 goto out_dev_put;
1577                         }
1578
1579                         if (!(nlh->nlmsg_flags & NLM_F_REPLACE))
1580                                 flags &= ~NEIGH_UPDATE_F_OVERRIDE;
1581                 }
1582
1583                 err = neigh_update(neigh, lladdr, ndm->ndm_state, flags);
1584                 neigh_release(neigh);
1585                 goto out_dev_put;
1586         }
1587
1588         read_unlock(&neigh_tbl_lock);
1589         err = -EAFNOSUPPORT;
1590
1591 out_dev_put:
1592         if (dev)
1593                 dev_put(dev);
1594 out:
1595         return err;
1596 }
1597
1598 static int neightbl_fill_parms(struct sk_buff *skb, struct neigh_parms *parms)
1599 {
1600         struct rtattr *nest = NULL;
1601
1602         nest = RTA_NEST(skb, NDTA_PARMS);
1603
1604         if (parms->dev)
1605                 RTA_PUT_U32(skb, NDTPA_IFINDEX, parms->dev->ifindex);
1606
1607         RTA_PUT_U32(skb, NDTPA_REFCNT, atomic_read(&parms->refcnt));
1608         RTA_PUT_U32(skb, NDTPA_QUEUE_LEN, parms->queue_len);
1609         RTA_PUT_U32(skb, NDTPA_PROXY_QLEN, parms->proxy_qlen);
1610         RTA_PUT_U32(skb, NDTPA_APP_PROBES, parms->app_probes);
1611         RTA_PUT_U32(skb, NDTPA_UCAST_PROBES, parms->ucast_probes);
1612         RTA_PUT_U32(skb, NDTPA_MCAST_PROBES, parms->mcast_probes);
1613         RTA_PUT_MSECS(skb, NDTPA_REACHABLE_TIME, parms->reachable_time);
1614         RTA_PUT_MSECS(skb, NDTPA_BASE_REACHABLE_TIME,
1615                       parms->base_reachable_time);
1616         RTA_PUT_MSECS(skb, NDTPA_GC_STALETIME, parms->gc_staletime);
1617         RTA_PUT_MSECS(skb, NDTPA_DELAY_PROBE_TIME, parms->delay_probe_time);
1618         RTA_PUT_MSECS(skb, NDTPA_RETRANS_TIME, parms->retrans_time);
1619         RTA_PUT_MSECS(skb, NDTPA_ANYCAST_DELAY, parms->anycast_delay);
1620         RTA_PUT_MSECS(skb, NDTPA_PROXY_DELAY, parms->proxy_delay);
1621         RTA_PUT_MSECS(skb, NDTPA_LOCKTIME, parms->locktime);
1622
1623         return RTA_NEST_END(skb, nest);
1624
1625 rtattr_failure:
1626         return RTA_NEST_CANCEL(skb, nest);
1627 }
1628
1629 static int neightbl_fill_info(struct neigh_table *tbl, struct sk_buff *skb,
1630                               struct netlink_callback *cb)
1631 {
1632         struct nlmsghdr *nlh;
1633         struct ndtmsg *ndtmsg;
1634
1635         nlh = NLMSG_NEW_ANSWER(skb, cb, RTM_NEWNEIGHTBL, sizeof(struct ndtmsg),
1636                                NLM_F_MULTI);
1637
1638         ndtmsg = NLMSG_DATA(nlh);
1639
1640         read_lock_bh(&tbl->lock);
1641         ndtmsg->ndtm_family = tbl->family;
1642         ndtmsg->ndtm_pad1   = 0;
1643         ndtmsg->ndtm_pad2   = 0;
1644
1645         RTA_PUT_STRING(skb, NDTA_NAME, tbl->id);
1646         RTA_PUT_MSECS(skb, NDTA_GC_INTERVAL, tbl->gc_interval);
1647         RTA_PUT_U32(skb, NDTA_THRESH1, tbl->gc_thresh1);
1648         RTA_PUT_U32(skb, NDTA_THRESH2, tbl->gc_thresh2);
1649         RTA_PUT_U32(skb, NDTA_THRESH3, tbl->gc_thresh3);
1650
1651         {
1652                 unsigned long now = jiffies;
1653                 unsigned int flush_delta = now - tbl->last_flush;
1654                 unsigned int rand_delta = now - tbl->last_rand;
1655
1656                 struct ndt_config ndc = {
1657                         .ndtc_key_len           = tbl->key_len,
1658                         .ndtc_entry_size        = tbl->entry_size,
1659                         .ndtc_entries           = atomic_read(&tbl->entries),
1660                         .ndtc_last_flush        = jiffies_to_msecs(flush_delta),
1661                         .ndtc_last_rand         = jiffies_to_msecs(rand_delta),
1662                         .ndtc_hash_rnd          = tbl->hash_rnd,
1663                         .ndtc_hash_mask         = tbl->hash_mask,
1664                         .ndtc_hash_chain_gc     = tbl->hash_chain_gc,
1665                         .ndtc_proxy_qlen        = tbl->proxy_queue.qlen,
1666                 };
1667
1668                 RTA_PUT(skb, NDTA_CONFIG, sizeof(ndc), &ndc);
1669         }
1670
1671         {
1672                 int cpu;
1673                 struct ndt_stats ndst;
1674
1675                 memset(&ndst, 0, sizeof(ndst));
1676
1677                 for_each_possible_cpu(cpu) {
1678                         struct neigh_statistics *st;
1679
1680                         st = per_cpu_ptr(tbl->stats, cpu);
1681                         ndst.ndts_allocs                += st->allocs;
1682                         ndst.ndts_destroys              += st->destroys;
1683                         ndst.ndts_hash_grows            += st->hash_grows;
1684                         ndst.ndts_res_failed            += st->res_failed;
1685                         ndst.ndts_lookups               += st->lookups;
1686                         ndst.ndts_hits                  += st->hits;
1687                         ndst.ndts_rcv_probes_mcast      += st->rcv_probes_mcast;
1688                         ndst.ndts_rcv_probes_ucast      += st->rcv_probes_ucast;
1689                         ndst.ndts_periodic_gc_runs      += st->periodic_gc_runs;
1690                         ndst.ndts_forced_gc_runs        += st->forced_gc_runs;
1691                 }
1692
1693                 RTA_PUT(skb, NDTA_STATS, sizeof(ndst), &ndst);
1694         }
1695
1696         BUG_ON(tbl->parms.dev);
1697         if (neightbl_fill_parms(skb, &tbl->parms) < 0)
1698                 goto rtattr_failure;
1699
1700         read_unlock_bh(&tbl->lock);
1701         return NLMSG_END(skb, nlh);
1702
1703 rtattr_failure:
1704         read_unlock_bh(&tbl->lock);
1705         return NLMSG_CANCEL(skb, nlh);
1706
1707 nlmsg_failure:
1708         return -1;
1709 }
1710
1711 static int neightbl_fill_param_info(struct neigh_table *tbl,
1712                                     struct neigh_parms *parms,
1713                                     struct sk_buff *skb,
1714                                     struct netlink_callback *cb)
1715 {
1716         struct ndtmsg *ndtmsg;
1717         struct nlmsghdr *nlh;
1718
1719         nlh = NLMSG_NEW_ANSWER(skb, cb, RTM_NEWNEIGHTBL, sizeof(struct ndtmsg),
1720                                NLM_F_MULTI);
1721
1722         ndtmsg = NLMSG_DATA(nlh);
1723
1724         read_lock_bh(&tbl->lock);
1725         ndtmsg->ndtm_family = tbl->family;
1726         ndtmsg->ndtm_pad1   = 0;
1727         ndtmsg->ndtm_pad2   = 0;
1728         RTA_PUT_STRING(skb, NDTA_NAME, tbl->id);
1729
1730         if (neightbl_fill_parms(skb, parms) < 0)
1731                 goto rtattr_failure;
1732
1733         read_unlock_bh(&tbl->lock);
1734         return NLMSG_END(skb, nlh);
1735
1736 rtattr_failure:
1737         read_unlock_bh(&tbl->lock);
1738         return NLMSG_CANCEL(skb, nlh);
1739
1740 nlmsg_failure:
1741         return -1;
1742 }
1743
1744 static inline struct neigh_parms *lookup_neigh_params(struct neigh_table *tbl,
1745                                                       int ifindex)
1746 {
1747         struct neigh_parms *p;
1748
1749         for (p = &tbl->parms; p; p = p->next)
1750                 if ((p->dev && p->dev->ifindex == ifindex) ||
1751                     (!p->dev && !ifindex))
1752                         return p;
1753
1754         return NULL;
1755 }
1756
1757 static struct nla_policy nl_neightbl_policy[NDTA_MAX+1] __read_mostly = {
1758         [NDTA_NAME]             = { .type = NLA_STRING },
1759         [NDTA_THRESH1]          = { .type = NLA_U32 },
1760         [NDTA_THRESH2]          = { .type = NLA_U32 },
1761         [NDTA_THRESH3]          = { .type = NLA_U32 },
1762         [NDTA_GC_INTERVAL]      = { .type = NLA_U64 },
1763         [NDTA_PARMS]            = { .type = NLA_NESTED },
1764 };
1765
1766 static struct nla_policy nl_ntbl_parm_policy[NDTPA_MAX+1] __read_mostly = {
1767         [NDTPA_IFINDEX]                 = { .type = NLA_U32 },
1768         [NDTPA_QUEUE_LEN]               = { .type = NLA_U32 },
1769         [NDTPA_PROXY_QLEN]              = { .type = NLA_U32 },
1770         [NDTPA_APP_PROBES]              = { .type = NLA_U32 },
1771         [NDTPA_UCAST_PROBES]            = { .type = NLA_U32 },
1772         [NDTPA_MCAST_PROBES]            = { .type = NLA_U32 },
1773         [NDTPA_BASE_REACHABLE_TIME]     = { .type = NLA_U64 },
1774         [NDTPA_GC_STALETIME]            = { .type = NLA_U64 },
1775         [NDTPA_DELAY_PROBE_TIME]        = { .type = NLA_U64 },
1776         [NDTPA_RETRANS_TIME]            = { .type = NLA_U64 },
1777         [NDTPA_ANYCAST_DELAY]           = { .type = NLA_U64 },
1778         [NDTPA_PROXY_DELAY]             = { .type = NLA_U64 },
1779         [NDTPA_LOCKTIME]                = { .type = NLA_U64 },
1780 };
1781
1782 int neightbl_set(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
1783 {
1784         struct neigh_table *tbl;
1785         struct ndtmsg *ndtmsg;
1786         struct nlattr *tb[NDTA_MAX+1];
1787         int err;
1788
1789         err = nlmsg_parse(nlh, sizeof(*ndtmsg), tb, NDTA_MAX,
1790                           nl_neightbl_policy);
1791         if (err < 0)
1792                 goto errout;
1793
1794         if (tb[NDTA_NAME] == NULL) {
1795                 err = -EINVAL;
1796                 goto errout;
1797         }
1798
1799         ndtmsg = nlmsg_data(nlh);
1800         read_lock(&neigh_tbl_lock);
1801         for (tbl = neigh_tables; tbl; tbl = tbl->next) {
1802                 if (ndtmsg->ndtm_family && tbl->family != ndtmsg->ndtm_family)
1803                         continue;
1804
1805                 if (nla_strcmp(tb[NDTA_NAME], tbl->id) == 0)
1806                         break;
1807         }
1808
1809         if (tbl == NULL) {
1810                 err = -ENOENT;
1811                 goto errout_locked;
1812         }
1813
1814         /*
1815          * We acquire tbl->lock to be nice to the periodic timers and
1816          * make sure they always see a consistent set of values.
1817          */
1818         write_lock_bh(&tbl->lock);
1819
1820         if (tb[NDTA_PARMS]) {
1821                 struct nlattr *tbp[NDTPA_MAX+1];
1822                 struct neigh_parms *p;
1823                 int i, ifindex = 0;
1824
1825                 err = nla_parse_nested(tbp, NDTPA_MAX, tb[NDTA_PARMS],
1826                                        nl_ntbl_parm_policy);
1827                 if (err < 0)
1828                         goto errout_tbl_lock;
1829
1830                 if (tbp[NDTPA_IFINDEX])
1831                         ifindex = nla_get_u32(tbp[NDTPA_IFINDEX]);
1832
1833                 p = lookup_neigh_params(tbl, ifindex);
1834                 if (p == NULL) {
1835                         err = -ENOENT;
1836                         goto errout_tbl_lock;
1837                 }
1838
1839                 for (i = 1; i <= NDTPA_MAX; i++) {
1840                         if (tbp[i] == NULL)
1841                                 continue;
1842
1843                         switch (i) {
1844                         case NDTPA_QUEUE_LEN:
1845                                 p->queue_len = nla_get_u32(tbp[i]);
1846                                 break;
1847                         case NDTPA_PROXY_QLEN:
1848                                 p->proxy_qlen = nla_get_u32(tbp[i]);
1849                                 break;
1850                         case NDTPA_APP_PROBES:
1851                                 p->app_probes = nla_get_u32(tbp[i]);
1852                                 break;
1853                         case NDTPA_UCAST_PROBES:
1854                                 p->ucast_probes = nla_get_u32(tbp[i]);
1855                                 break;
1856                         case NDTPA_MCAST_PROBES:
1857                                 p->mcast_probes = nla_get_u32(tbp[i]);
1858                                 break;
1859                         case NDTPA_BASE_REACHABLE_TIME:
1860                                 p->base_reachable_time = nla_get_msecs(tbp[i]);
1861                                 break;
1862                         case NDTPA_GC_STALETIME:
1863                                 p->gc_staletime = nla_get_msecs(tbp[i]);
1864                                 break;
1865                         case NDTPA_DELAY_PROBE_TIME:
1866                                 p->delay_probe_time = nla_get_msecs(tbp[i]);
1867                                 break;
1868                         case NDTPA_RETRANS_TIME:
1869                                 p->retrans_time = nla_get_msecs(tbp[i]);
1870                                 break;
1871                         case NDTPA_ANYCAST_DELAY:
1872                                 p->anycast_delay = nla_get_msecs(tbp[i]);
1873                                 break;
1874                         case NDTPA_PROXY_DELAY:
1875                                 p->proxy_delay = nla_get_msecs(tbp[i]);
1876                                 break;
1877                         case NDTPA_LOCKTIME:
1878                                 p->locktime = nla_get_msecs(tbp[i]);
1879                                 break;
1880                         }
1881                 }
1882         }
1883
1884         if (tb[NDTA_THRESH1])
1885                 tbl->gc_thresh1 = nla_get_u32(tb[NDTA_THRESH1]);
1886
1887         if (tb[NDTA_THRESH2])
1888                 tbl->gc_thresh2 = nla_get_u32(tb[NDTA_THRESH2]);
1889
1890         if (tb[NDTA_THRESH3])
1891                 tbl->gc_thresh3 = nla_get_u32(tb[NDTA_THRESH3]);
1892
1893         if (tb[NDTA_GC_INTERVAL])
1894                 tbl->gc_interval = nla_get_msecs(tb[NDTA_GC_INTERVAL]);
1895
1896         err = 0;
1897
1898 errout_tbl_lock:
1899         write_unlock_bh(&tbl->lock);
1900 errout_locked:
1901         read_unlock(&neigh_tbl_lock);
1902 errout:
1903         return err;
1904 }
1905
1906 int neightbl_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
1907 {
1908         int idx, family;
1909         int s_idx = cb->args[0];
1910         struct neigh_table *tbl;
1911
1912         family = ((struct rtgenmsg *)NLMSG_DATA(cb->nlh))->rtgen_family;
1913
1914         read_lock(&neigh_tbl_lock);
1915         for (tbl = neigh_tables, idx = 0; tbl; tbl = tbl->next) {
1916                 struct neigh_parms *p;
1917
1918                 if (idx < s_idx || (family && tbl->family != family))
1919                         continue;
1920
1921                 if (neightbl_fill_info(tbl, skb, cb) <= 0)
1922                         break;
1923
1924                 for (++idx, p = tbl->parms.next; p; p = p->next, idx++) {
1925                         if (idx < s_idx)
1926                                 continue;
1927
1928                         if (neightbl_fill_param_info(tbl, p, skb, cb) <= 0)
1929                                 goto out;
1930                 }
1931
1932         }
1933 out:
1934         read_unlock(&neigh_tbl_lock);
1935         cb->args[0] = idx;
1936
1937         return skb->len;
1938 }
1939
1940 static int neigh_fill_info(struct sk_buff *skb, struct neighbour *neigh,
1941                            u32 pid, u32 seq, int type, unsigned int flags)
1942 {
1943         unsigned long now = jiffies;
1944         struct nda_cacheinfo ci;
1945         struct nlmsghdr *nlh;
1946         struct ndmsg *ndm;
1947
1948         nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
1949         if (nlh == NULL)
1950                 return -ENOBUFS;
1951
1952         ndm = nlmsg_data(nlh);
1953         ndm->ndm_family  = neigh->ops->family;
1954         ndm->ndm_pad1    = 0;
1955         ndm->ndm_pad2    = 0;
1956         ndm->ndm_flags   = neigh->flags;
1957         ndm->ndm_type    = neigh->type;
1958         ndm->ndm_ifindex = neigh->dev->ifindex;
1959
1960         NLA_PUT(skb, NDA_DST, neigh->tbl->key_len, neigh->primary_key);
1961
1962         read_lock_bh(&neigh->lock);
1963         ndm->ndm_state   = neigh->nud_state;
1964         if ((neigh->nud_state & NUD_VALID) &&
1965             nla_put(skb, NDA_LLADDR, neigh->dev->addr_len, neigh->ha) < 0) {
1966                 read_unlock_bh(&neigh->lock);
1967                 goto nla_put_failure;
1968         }
1969
1970         ci.ndm_used      = now - neigh->used;
1971         ci.ndm_confirmed = now - neigh->confirmed;
1972         ci.ndm_updated   = now - neigh->updated;
1973         ci.ndm_refcnt    = atomic_read(&neigh->refcnt) - 1;
1974         read_unlock_bh(&neigh->lock);
1975
1976         NLA_PUT_U32(skb, NDA_PROBES, atomic_read(&neigh->probes));
1977         NLA_PUT(skb, NDA_CACHEINFO, sizeof(ci), &ci);
1978
1979         return nlmsg_end(skb, nlh);
1980
1981 nla_put_failure:
1982         return nlmsg_cancel(skb, nlh);
1983 }
1984
1985
1986 static int neigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
1987                             struct netlink_callback *cb)
1988 {
1989         struct neighbour *n;
1990         int rc, h, s_h = cb->args[1];
1991         int idx, s_idx = idx = cb->args[2];
1992
1993         for (h = 0; h <= tbl->hash_mask; h++) {
1994                 if (h < s_h)
1995                         continue;
1996                 if (h > s_h)
1997                         s_idx = 0;
1998                 read_lock_bh(&tbl->lock);
1999                 for (n = tbl->hash_buckets[h], idx = 0; n; n = n->next, idx++) {
2000                         if (idx < s_idx)
2001                                 continue;
2002                         if (neigh_fill_info(skb, n, NETLINK_CB(cb->skb).pid,
2003                                             cb->nlh->nlmsg_seq,
2004                                             RTM_NEWNEIGH,
2005                                             NLM_F_MULTI) <= 0) {
2006                                 read_unlock_bh(&tbl->lock);
2007                                 rc = -1;
2008                                 goto out;
2009                         }
2010                 }
2011                 read_unlock_bh(&tbl->lock);
2012         }
2013         rc = skb->len;
2014 out:
2015         cb->args[1] = h;
2016         cb->args[2] = idx;
2017         return rc;
2018 }
2019
2020 int neigh_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2021 {
2022         struct neigh_table *tbl;
2023         int t, family, s_t;
2024
2025         read_lock(&neigh_tbl_lock);
2026         family = ((struct rtgenmsg *) nlmsg_data(cb->nlh))->rtgen_family;
2027         s_t = cb->args[0];
2028
2029         for (tbl = neigh_tables, t = 0; tbl; tbl = tbl->next, t++) {
2030                 if (t < s_t || (family && tbl->family != family))
2031                         continue;
2032                 if (t > s_t)
2033                         memset(&cb->args[1], 0, sizeof(cb->args) -
2034                                                 sizeof(cb->args[0]));
2035                 if (neigh_dump_table(tbl, skb, cb) < 0)
2036                         break;
2037         }
2038         read_unlock(&neigh_tbl_lock);
2039
2040         cb->args[0] = t;
2041         return skb->len;
2042 }
2043
2044 void neigh_for_each(struct neigh_table *tbl, void (*cb)(struct neighbour *, void *), void *cookie)
2045 {
2046         int chain;
2047
2048         read_lock_bh(&tbl->lock);
2049         for (chain = 0; chain <= tbl->hash_mask; chain++) {
2050                 struct neighbour *n;
2051
2052                 for (n = tbl->hash_buckets[chain]; n; n = n->next)
2053                         cb(n, cookie);
2054         }
2055         read_unlock_bh(&tbl->lock);
2056 }
2057 EXPORT_SYMBOL(neigh_for_each);
2058
2059 /* The tbl->lock must be held as a writer and BH disabled. */
2060 void __neigh_for_each_release(struct neigh_table *tbl,
2061                               int (*cb)(struct neighbour *))
2062 {
2063         int chain;
2064
2065         for (chain = 0; chain <= tbl->hash_mask; chain++) {
2066                 struct neighbour *n, **np;
2067
2068                 np = &tbl->hash_buckets[chain];
2069                 while ((n = *np) != NULL) {
2070                         int release;
2071
2072                         write_lock(&n->lock);
2073                         release = cb(n);
2074                         if (release) {
2075                                 *np = n->next;
2076                                 n->dead = 1;
2077                         } else
2078                                 np = &n->next;
2079                         write_unlock(&n->lock);
2080                         if (release)
2081                                 neigh_release(n);
2082                 }
2083         }
2084 }
2085 EXPORT_SYMBOL(__neigh_for_each_release);
2086
2087 #ifdef CONFIG_PROC_FS
2088
2089 static struct neighbour *neigh_get_first(struct seq_file *seq)
2090 {
2091         struct neigh_seq_state *state = seq->private;
2092         struct neigh_table *tbl = state->tbl;
2093         struct neighbour *n = NULL;
2094         int bucket = state->bucket;
2095
2096         state->flags &= ~NEIGH_SEQ_IS_PNEIGH;
2097         for (bucket = 0; bucket <= tbl->hash_mask; bucket++) {
2098                 n = tbl->hash_buckets[bucket];
2099
2100                 while (n) {
2101                         if (state->neigh_sub_iter) {
2102                                 loff_t fakep = 0;
2103                                 void *v;
2104
2105                                 v = state->neigh_sub_iter(state, n, &fakep);
2106                                 if (!v)
2107                                         goto next;
2108                         }
2109                         if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
2110                                 break;
2111                         if (n->nud_state & ~NUD_NOARP)
2112                                 break;
2113                 next:
2114                         n = n->next;
2115                 }
2116
2117                 if (n)
2118                         break;
2119         }
2120         state->bucket = bucket;
2121
2122         return n;
2123 }
2124
2125 static struct neighbour *neigh_get_next(struct seq_file *seq,
2126                                         struct neighbour *n,
2127                                         loff_t *pos)
2128 {
2129         struct neigh_seq_state *state = seq->private;
2130         struct neigh_table *tbl = state->tbl;
2131
2132         if (state->neigh_sub_iter) {
2133                 void *v = state->neigh_sub_iter(state, n, pos);
2134                 if (v)
2135                         return n;
2136         }
2137         n = n->next;
2138
2139         while (1) {
2140                 while (n) {
2141                         if (state->neigh_sub_iter) {
2142                                 void *v = state->neigh_sub_iter(state, n, pos);
2143                                 if (v)
2144                                         return n;
2145                                 goto next;
2146                         }
2147                         if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
2148                                 break;
2149
2150                         if (n->nud_state & ~NUD_NOARP)
2151                                 break;
2152                 next:
2153                         n = n->next;
2154                 }
2155
2156                 if (n)
2157                         break;
2158
2159                 if (++state->bucket > tbl->hash_mask)
2160                         break;
2161
2162                 n = tbl->hash_buckets[state->bucket];
2163         }
2164
2165         if (n && pos)
2166                 --(*pos);
2167         return n;
2168 }
2169
2170 static struct neighbour *neigh_get_idx(struct seq_file *seq, loff_t *pos)
2171 {
2172         struct neighbour *n = neigh_get_first(seq);
2173
2174         if (n) {
2175                 while (*pos) {
2176                         n = neigh_get_next(seq, n, pos);
2177                         if (!n)
2178                                 break;
2179                 }
2180         }
2181         return *pos ? NULL : n;
2182 }
2183
2184 static struct pneigh_entry *pneigh_get_first(struct seq_file *seq)
2185 {
2186         struct neigh_seq_state *state = seq->private;
2187         struct neigh_table *tbl = state->tbl;
2188         struct pneigh_entry *pn = NULL;
2189         int bucket = state->bucket;
2190
2191         state->flags |= NEIGH_SEQ_IS_PNEIGH;
2192         for (bucket = 0; bucket <= PNEIGH_HASHMASK; bucket++) {
2193                 pn = tbl->phash_buckets[bucket];
2194                 if (pn)
2195                         break;
2196         }
2197         state->bucket = bucket;
2198
2199         return pn;
2200 }
2201
2202 static struct pneigh_entry *pneigh_get_next(struct seq_file *seq,
2203                                             struct pneigh_entry *pn,
2204                                             loff_t *pos)
2205 {
2206         struct neigh_seq_state *state = seq->private;
2207         struct neigh_table *tbl = state->tbl;
2208
2209         pn = pn->next;
2210         while (!pn) {
2211                 if (++state->bucket > PNEIGH_HASHMASK)
2212                         break;
2213                 pn = tbl->phash_buckets[state->bucket];
2214                 if (pn)
2215                         break;
2216         }
2217
2218         if (pn && pos)
2219                 --(*pos);
2220
2221         return pn;
2222 }
2223
2224 static struct pneigh_entry *pneigh_get_idx(struct seq_file *seq, loff_t *pos)
2225 {
2226         struct pneigh_entry *pn = pneigh_get_first(seq);
2227
2228         if (pn) {
2229                 while (*pos) {
2230                         pn = pneigh_get_next(seq, pn, pos);
2231                         if (!pn)
2232                                 break;
2233                 }
2234         }
2235         return *pos ? NULL : pn;
2236 }
2237
2238 static void *neigh_get_idx_any(struct seq_file *seq, loff_t *pos)
2239 {
2240         struct neigh_seq_state *state = seq->private;
2241         void *rc;
2242
2243         rc = neigh_get_idx(seq, pos);
2244         if (!rc && !(state->flags & NEIGH_SEQ_NEIGH_ONLY))
2245                 rc = pneigh_get_idx(seq, pos);
2246
2247         return rc;
2248 }
2249
2250 void *neigh_seq_start(struct seq_file *seq, loff_t *pos, struct neigh_table *tbl, unsigned int neigh_seq_flags)
2251 {
2252         struct neigh_seq_state *state = seq->private;
2253         loff_t pos_minus_one;
2254
2255         state->tbl = tbl;
2256         state->bucket = 0;
2257         state->flags = (neigh_seq_flags & ~NEIGH_SEQ_IS_PNEIGH);
2258
2259         read_lock_bh(&tbl->lock);
2260
2261         pos_minus_one = *pos - 1;
2262         return *pos ? neigh_get_idx_any(seq, &pos_minus_one) : SEQ_START_TOKEN;
2263 }
2264 EXPORT_SYMBOL(neigh_seq_start);
2265
2266 void *neigh_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2267 {
2268         struct neigh_seq_state *state;
2269         void *rc;
2270
2271         if (v == SEQ_START_TOKEN) {
2272                 rc = neigh_get_idx(seq, pos);
2273                 goto out;
2274         }
2275
2276         state = seq->private;
2277         if (!(state->flags & NEIGH_SEQ_IS_PNEIGH)) {
2278                 rc = neigh_get_next(seq, v, NULL);
2279                 if (rc)
2280                         goto out;
2281                 if (!(state->flags & NEIGH_SEQ_NEIGH_ONLY))
2282                         rc = pneigh_get_first(seq);
2283         } else {
2284                 BUG_ON(state->flags & NEIGH_SEQ_NEIGH_ONLY);
2285                 rc = pneigh_get_next(seq, v, NULL);
2286         }
2287 out:
2288         ++(*pos);
2289         return rc;
2290 }
2291 EXPORT_SYMBOL(neigh_seq_next);
2292
2293 void neigh_seq_stop(struct seq_file *seq, void *v)
2294 {
2295         struct neigh_seq_state *state = seq->private;
2296         struct neigh_table *tbl = state->tbl;
2297
2298         read_unlock_bh(&tbl->lock);
2299 }
2300 EXPORT_SYMBOL(neigh_seq_stop);
2301
2302 /* statistics via seq_file */
2303
2304 static void *neigh_stat_seq_start(struct seq_file *seq, loff_t *pos)
2305 {
2306         struct proc_dir_entry *pde = seq->private;
2307         struct neigh_table *tbl = pde->data;
2308         int cpu;
2309
2310         if (*pos == 0)
2311                 return SEQ_START_TOKEN;
2312
2313         for (cpu = *pos-1; cpu < NR_CPUS; ++cpu) {
2314                 if (!cpu_possible(cpu))
2315                         continue;
2316                 *pos = cpu+1;
2317                 return per_cpu_ptr(tbl->stats, cpu);
2318         }
2319         return NULL;
2320 }
2321
2322 static void *neigh_stat_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2323 {
2324         struct proc_dir_entry *pde = seq->private;
2325         struct neigh_table *tbl = pde->data;
2326         int cpu;
2327
2328         for (cpu = *pos; cpu < NR_CPUS; ++cpu) {
2329                 if (!cpu_possible(cpu))
2330                         continue;
2331                 *pos = cpu+1;
2332                 return per_cpu_ptr(tbl->stats, cpu);
2333         }
2334         return NULL;
2335 }
2336
2337 static void neigh_stat_seq_stop(struct seq_file *seq, void *v)
2338 {
2339
2340 }
2341
2342 static int neigh_stat_seq_show(struct seq_file *seq, void *v)
2343 {
2344         struct proc_dir_entry *pde = seq->private;
2345         struct neigh_table *tbl = pde->data;
2346         struct neigh_statistics *st = v;
2347
2348         if (v == SEQ_START_TOKEN) {
2349                 seq_printf(seq, "entries  allocs destroys hash_grows  lookups hits  res_failed  rcv_probes_mcast rcv_probes_ucast  periodic_gc_runs forced_gc_runs\n");
2350                 return 0;
2351         }
2352
2353         seq_printf(seq, "%08x  %08lx %08lx %08lx  %08lx %08lx  %08lx  "
2354                         "%08lx %08lx  %08lx %08lx\n",
2355                    atomic_read(&tbl->entries),
2356
2357                    st->allocs,
2358                    st->destroys,
2359                    st->hash_grows,
2360
2361                    st->lookups,
2362                    st->hits,
2363
2364                    st->res_failed,
2365
2366                    st->rcv_probes_mcast,
2367                    st->rcv_probes_ucast,
2368
2369                    st->periodic_gc_runs,
2370                    st->forced_gc_runs
2371                    );
2372
2373         return 0;
2374 }
2375
2376 static struct seq_operations neigh_stat_seq_ops = {
2377         .start  = neigh_stat_seq_start,
2378         .next   = neigh_stat_seq_next,
2379         .stop   = neigh_stat_seq_stop,
2380         .show   = neigh_stat_seq_show,
2381 };
2382
2383 static int neigh_stat_seq_open(struct inode *inode, struct file *file)
2384 {
2385         int ret = seq_open(file, &neigh_stat_seq_ops);
2386
2387         if (!ret) {
2388                 struct seq_file *sf = file->private_data;
2389                 sf->private = PDE(inode);
2390         }
2391         return ret;
2392 };
2393
2394 static struct file_operations neigh_stat_seq_fops = {
2395         .owner   = THIS_MODULE,
2396         .open    = neigh_stat_seq_open,
2397         .read    = seq_read,
2398         .llseek  = seq_lseek,
2399         .release = seq_release,
2400 };
2401
2402 #endif /* CONFIG_PROC_FS */
2403
2404 #ifdef CONFIG_ARPD
2405 void neigh_app_ns(struct neighbour *n)
2406 {
2407         struct sk_buff *skb;
2408
2409         skb = nlmsg_new(NLMSG_GOODSIZE, GFP_ATOMIC);
2410         if (skb == NULL)
2411                 return;
2412
2413         if (neigh_fill_info(skb, n, 0, 0, RTM_GETNEIGH, NLM_F_REQUEST) <= 0)
2414                 kfree_skb(skb);
2415         else {
2416                 NETLINK_CB(skb).dst_group  = RTNLGRP_NEIGH;
2417                 netlink_broadcast(rtnl, skb, 0, RTNLGRP_NEIGH, GFP_ATOMIC);
2418         }
2419 }
2420
2421 static void neigh_app_notify(struct neighbour *n)
2422 {
2423         struct sk_buff *skb;
2424
2425         skb = nlmsg_new(NLMSG_GOODSIZE, GFP_ATOMIC);
2426         if (skb == NULL)
2427                 return;
2428
2429         if (neigh_fill_info(skb, n, 0, 0, RTM_NEWNEIGH, 0) <= 0)
2430                 kfree_skb(skb);
2431         else {
2432                 NETLINK_CB(skb).dst_group  = RTNLGRP_NEIGH;
2433                 netlink_broadcast(rtnl, skb, 0, RTNLGRP_NEIGH, GFP_ATOMIC);
2434         }
2435 }
2436
2437 #endif /* CONFIG_ARPD */
2438
2439 #ifdef CONFIG_SYSCTL
2440
2441 static struct neigh_sysctl_table {
2442         struct ctl_table_header *sysctl_header;
2443         ctl_table               neigh_vars[__NET_NEIGH_MAX];
2444         ctl_table               neigh_dev[2];
2445         ctl_table               neigh_neigh_dir[2];
2446         ctl_table               neigh_proto_dir[2];
2447         ctl_table               neigh_root_dir[2];
2448 } neigh_sysctl_template = {
2449         .neigh_vars = {
2450                 {
2451                         .ctl_name       = NET_NEIGH_MCAST_SOLICIT,
2452                         .procname       = "mcast_solicit",
2453                         .maxlen         = sizeof(int),
2454                         .mode           = 0644,
2455                         .proc_handler   = &proc_dointvec,
2456                 },
2457                 {
2458                         .ctl_name       = NET_NEIGH_UCAST_SOLICIT,
2459                         .procname       = "ucast_solicit",
2460                         .maxlen         = sizeof(int),
2461                         .mode           = 0644,
2462                         .proc_handler   = &proc_dointvec,
2463                 },
2464                 {
2465                         .ctl_name       = NET_NEIGH_APP_SOLICIT,
2466                         .procname       = "app_solicit",
2467                         .maxlen         = sizeof(int),
2468                         .mode           = 0644,
2469                         .proc_handler   = &proc_dointvec,
2470                 },
2471                 {
2472                         .ctl_name       = NET_NEIGH_RETRANS_TIME,
2473                         .procname       = "retrans_time",
2474                         .maxlen         = sizeof(int),
2475                         .mode           = 0644,
2476                         .proc_handler   = &proc_dointvec_userhz_jiffies,
2477                 },
2478                 {
2479                         .ctl_name       = NET_NEIGH_REACHABLE_TIME,
2480                         .procname       = "base_reachable_time",
2481                         .maxlen         = sizeof(int),
2482                         .mode           = 0644,
2483                         .proc_handler   = &proc_dointvec_jiffies,
2484                         .strategy       = &sysctl_jiffies,
2485                 },
2486                 {
2487                         .ctl_name       = NET_NEIGH_DELAY_PROBE_TIME,
2488                         .procname       = "delay_first_probe_time",
2489                         .maxlen         = sizeof(int),
2490                         .mode           = 0644,
2491                         .proc_handler   = &proc_dointvec_jiffies,
2492                         .strategy       = &sysctl_jiffies,
2493                 },
2494                 {
2495                         .ctl_name       = NET_NEIGH_GC_STALE_TIME,
2496                         .procname       = "gc_stale_time",
2497                         .maxlen         = sizeof(int),
2498                         .mode           = 0644,
2499                         .proc_handler   = &proc_dointvec_jiffies,
2500                         .strategy       = &sysctl_jiffies,
2501                 },
2502                 {
2503                         .ctl_name       = NET_NEIGH_UNRES_QLEN,
2504                         .procname       = "unres_qlen",
2505                         .maxlen         = sizeof(int),
2506                         .mode           = 0644,
2507                         .proc_handler   = &proc_dointvec,
2508                 },
2509                 {
2510                         .ctl_name       = NET_NEIGH_PROXY_QLEN,
2511                         .procname       = "proxy_qlen",
2512                         .maxlen         = sizeof(int),
2513                         .mode           = 0644,
2514                         .proc_handler   = &proc_dointvec,
2515                 },
2516                 {
2517                         .ctl_name       = NET_NEIGH_ANYCAST_DELAY,
2518                         .procname       = "anycast_delay",
2519                         .maxlen         = sizeof(int),
2520                         .mode           = 0644,
2521                         .proc_handler   = &proc_dointvec_userhz_jiffies,
2522                 },
2523                 {
2524                         .ctl_name       = NET_NEIGH_PROXY_DELAY,
2525                         .procname       = "proxy_delay",
2526                         .maxlen         = sizeof(int),
2527                         .mode           = 0644,
2528                         .proc_handler   = &proc_dointvec_userhz_jiffies,
2529                 },
2530                 {
2531                         .ctl_name       = NET_NEIGH_LOCKTIME,
2532                         .procname       = "locktime",
2533                         .maxlen         = sizeof(int),
2534                         .mode           = 0644,
2535                         .proc_handler   = &proc_dointvec_userhz_jiffies,
2536                 },
2537                 {
2538                         .ctl_name       = NET_NEIGH_GC_INTERVAL,
2539                         .procname       = "gc_interval",
2540                         .maxlen         = sizeof(int),
2541                         .mode           = 0644,
2542                         .proc_handler   = &proc_dointvec_jiffies,
2543                         .strategy       = &sysctl_jiffies,
2544                 },
2545                 {
2546                         .ctl_name       = NET_NEIGH_GC_THRESH1,
2547                         .procname       = "gc_thresh1",
2548                         .maxlen         = sizeof(int),
2549                         .mode           = 0644,
2550                         .proc_handler   = &proc_dointvec,
2551                 },
2552                 {
2553                         .ctl_name       = NET_NEIGH_GC_THRESH2,
2554                         .procname       = "gc_thresh2",
2555                         .maxlen         = sizeof(int),
2556                         .mode           = 0644,
2557                         .proc_handler   = &proc_dointvec,
2558                 },
2559                 {
2560                         .ctl_name       = NET_NEIGH_GC_THRESH3,
2561                         .procname       = "gc_thresh3",
2562                         .maxlen         = sizeof(int),
2563                         .mode           = 0644,
2564                         .proc_handler   = &proc_dointvec,
2565                 },
2566                 {
2567                         .ctl_name       = NET_NEIGH_RETRANS_TIME_MS,
2568                         .procname       = "retrans_time_ms",
2569                         .maxlen         = sizeof(int),
2570                         .mode           = 0644,
2571                         .proc_handler   = &proc_dointvec_ms_jiffies,
2572                         .strategy       = &sysctl_ms_jiffies,
2573                 },
2574                 {
2575                         .ctl_name       = NET_NEIGH_REACHABLE_TIME_MS,
2576                         .procname       = "base_reachable_time_ms",
2577                         .maxlen         = sizeof(int),
2578                         .mode           = 0644,
2579                         .proc_handler   = &proc_dointvec_ms_jiffies,
2580                         .strategy       = &sysctl_ms_jiffies,
2581                 },
2582         },
2583         .neigh_dev = {
2584                 {
2585                         .ctl_name       = NET_PROTO_CONF_DEFAULT,
2586                         .procname       = "default",
2587                         .mode           = 0555,
2588                 },
2589         },
2590         .neigh_neigh_dir = {
2591                 {
2592                         .procname       = "neigh",
2593                         .mode           = 0555,
2594                 },
2595         },
2596         .neigh_proto_dir = {
2597                 {
2598                         .mode           = 0555,
2599                 },
2600         },
2601         .neigh_root_dir = {
2602                 {
2603                         .ctl_name       = CTL_NET,
2604                         .procname       = "net",
2605                         .mode           = 0555,
2606                 },
2607         },
2608 };
2609
2610 int neigh_sysctl_register(struct net_device *dev, struct neigh_parms *p,
2611                           int p_id, int pdev_id, char *p_name,
2612                           proc_handler *handler, ctl_handler *strategy)
2613 {
2614         struct neigh_sysctl_table *t = kmalloc(sizeof(*t), GFP_KERNEL);
2615         const char *dev_name_source = NULL;
2616         char *dev_name = NULL;
2617         int err = 0;
2618
2619         if (!t)
2620                 return -ENOBUFS;
2621         memcpy(t, &neigh_sysctl_template, sizeof(*t));
2622         t->neigh_vars[0].data  = &p->mcast_probes;
2623         t->neigh_vars[1].data  = &p->ucast_probes;
2624         t->neigh_vars[2].data  = &p->app_probes;
2625         t->neigh_vars[3].data  = &p->retrans_time;
2626         t->neigh_vars[4].data  = &p->base_reachable_time;
2627         t->neigh_vars[5].data  = &p->delay_probe_time;
2628         t->neigh_vars[6].data  = &p->gc_staletime;
2629         t->neigh_vars[7].data  = &p->queue_len;
2630         t->neigh_vars[8].data  = &p->proxy_qlen;
2631         t->neigh_vars[9].data  = &p->anycast_delay;
2632         t->neigh_vars[10].data = &p->proxy_delay;
2633         t->neigh_vars[11].data = &p->locktime;
2634
2635         if (dev) {
2636                 dev_name_source = dev->name;
2637                 t->neigh_dev[0].ctl_name = dev->ifindex;
2638                 t->neigh_vars[12].procname = NULL;
2639                 t->neigh_vars[13].procname = NULL;
2640                 t->neigh_vars[14].procname = NULL;
2641                 t->neigh_vars[15].procname = NULL;
2642         } else {
2643                 dev_name_source = t->neigh_dev[0].procname;
2644                 t->neigh_vars[12].data = (int *)(p + 1);
2645                 t->neigh_vars[13].data = (int *)(p + 1) + 1;
2646                 t->neigh_vars[14].data = (int *)(p + 1) + 2;
2647                 t->neigh_vars[15].data = (int *)(p + 1) + 3;
2648         }
2649
2650         t->neigh_vars[16].data  = &p->retrans_time;
2651         t->neigh_vars[17].data  = &p->base_reachable_time;
2652
2653         if (handler || strategy) {
2654                 /* RetransTime */
2655                 t->neigh_vars[3].proc_handler = handler;
2656                 t->neigh_vars[3].strategy = strategy;
2657                 t->neigh_vars[3].extra1 = dev;
2658                 /* ReachableTime */
2659                 t->neigh_vars[4].proc_handler = handler;
2660                 t->neigh_vars[4].strategy = strategy;
2661                 t->neigh_vars[4].extra1 = dev;
2662                 /* RetransTime (in milliseconds)*/
2663                 t->neigh_vars[16].proc_handler = handler;
2664                 t->neigh_vars[16].strategy = strategy;
2665                 t->neigh_vars[16].extra1 = dev;
2666                 /* ReachableTime (in milliseconds) */
2667                 t->neigh_vars[17].proc_handler = handler;
2668                 t->neigh_vars[17].strategy = strategy;
2669                 t->neigh_vars[17].extra1 = dev;
2670         }
2671
2672         dev_name = kstrdup(dev_name_source, GFP_KERNEL);
2673         if (!dev_name) {
2674                 err = -ENOBUFS;
2675                 goto free;
2676         }
2677
2678         t->neigh_dev[0].procname = dev_name;
2679
2680         t->neigh_neigh_dir[0].ctl_name = pdev_id;
2681
2682         t->neigh_proto_dir[0].procname = p_name;
2683         t->neigh_proto_dir[0].ctl_name = p_id;
2684
2685         t->neigh_dev[0].child          = t->neigh_vars;
2686         t->neigh_neigh_dir[0].child    = t->neigh_dev;
2687         t->neigh_proto_dir[0].child    = t->neigh_neigh_dir;
2688         t->neigh_root_dir[0].child     = t->neigh_proto_dir;
2689
2690         t->sysctl_header = register_sysctl_table(t->neigh_root_dir, 0);
2691         if (!t->sysctl_header) {
2692                 err = -ENOBUFS;
2693                 goto free_procname;
2694         }
2695         p->sysctl_table = t;
2696         return 0;
2697
2698         /* error path */
2699  free_procname:
2700         kfree(dev_name);
2701  free:
2702         kfree(t);
2703
2704         return err;
2705 }
2706
2707 void neigh_sysctl_unregister(struct neigh_parms *p)
2708 {
2709         if (p->sysctl_table) {
2710                 struct neigh_sysctl_table *t = p->sysctl_table;
2711                 p->sysctl_table = NULL;
2712                 unregister_sysctl_table(t->sysctl_header);
2713                 kfree(t->neigh_dev[0].procname);
2714                 kfree(t);
2715         }
2716 }
2717
2718 #endif  /* CONFIG_SYSCTL */
2719
2720 EXPORT_SYMBOL(__neigh_event_send);
2721 EXPORT_SYMBOL(neigh_add);
2722 EXPORT_SYMBOL(neigh_changeaddr);
2723 EXPORT_SYMBOL(neigh_compat_output);
2724 EXPORT_SYMBOL(neigh_connected_output);
2725 EXPORT_SYMBOL(neigh_create);
2726 EXPORT_SYMBOL(neigh_delete);
2727 EXPORT_SYMBOL(neigh_destroy);
2728 EXPORT_SYMBOL(neigh_dump_info);
2729 EXPORT_SYMBOL(neigh_event_ns);
2730 EXPORT_SYMBOL(neigh_ifdown);
2731 EXPORT_SYMBOL(neigh_lookup);
2732 EXPORT_SYMBOL(neigh_lookup_nodev);
2733 EXPORT_SYMBOL(neigh_parms_alloc);
2734 EXPORT_SYMBOL(neigh_parms_release);
2735 EXPORT_SYMBOL(neigh_rand_reach_time);
2736 EXPORT_SYMBOL(neigh_resolve_output);
2737 EXPORT_SYMBOL(neigh_table_clear);
2738 EXPORT_SYMBOL(neigh_table_init);
2739 EXPORT_SYMBOL(neigh_table_init_no_netlink);
2740 EXPORT_SYMBOL(neigh_update);
2741 EXPORT_SYMBOL(neigh_update_hhs);
2742 EXPORT_SYMBOL(pneigh_enqueue);
2743 EXPORT_SYMBOL(pneigh_lookup);
2744 EXPORT_SYMBOL(neightbl_dump_info);
2745 EXPORT_SYMBOL(neightbl_set);
2746
2747 #ifdef CONFIG_ARPD
2748 EXPORT_SYMBOL(neigh_app_ns);
2749 #endif
2750 #ifdef CONFIG_SYSCTL
2751 EXPORT_SYMBOL(neigh_sysctl_register);
2752 EXPORT_SYMBOL(neigh_sysctl_unregister);
2753 #endif