net/core/neighbour.c

   1 /*
   2  *      Generic address resolution entity
   3  *
   4  *      Authors:
   5  *      Pedro Roque             <roque@di.fc.ul.pt>
   6  *      Alexey Kuznetsov        <kuznet@ms2.inr.ac.ru>
   7  *
   8  *      This program is free software; you can redistribute it and/or
   9  *      modify it under the terms of the GNU General Public License
  10  *      as published by the Free Software Foundation; either version
  11  *      2 of the License, or (at your option) any later version.
  12  *
  13  *      Fixes:
  14  *      Vitaly E. Lavrov        releasing NULL neighbor in neigh_add.
  15  *      Harald Welte            Add neighbour cache statistics like rtstat
  16  */
  17
  18 #include <linux/types.h>
  19 #include <linux/kernel.h>
  20 #include <linux/module.h>
  21 #include <linux/socket.h>
  22 #include <linux/netdevice.h>
  23 #include <linux/proc_fs.h>
  24 #ifdef CONFIG_SYSCTL
  25 #include <linux/sysctl.h>
  26 #endif
  27 #include <linux/times.h>
  28 #include <net/net_namespace.h>
  29 #include <net/neighbour.h>
  30 #include <net/dst.h>
  31 #include <net/sock.h>
  32 #include <net/netevent.h>
  33 #include <net/netlink.h>
  34 #include <linux/rtnetlink.h>
  35 #include <linux/random.h>
  36 #include <linux/string.h>
  37 #include <linux/log2.h>
  38
  39 #define NEIGH_DEBUG 1
  40
  41 #define NEIGH_PRINTK(x...) printk(x)
  42 #define NEIGH_NOPRINTK(x...) do { ; } while(0)
  43 #define NEIGH_PRINTK0 NEIGH_PRINTK
  44 #define NEIGH_PRINTK1 NEIGH_NOPRINTK
  45 #define NEIGH_PRINTK2 NEIGH_NOPRINTK
  46
  47 #if NEIGH_DEBUG >= 1
  48 #undef NEIGH_PRINTK1
  49 #define NEIGH_PRINTK1 NEIGH_PRINTK
  50 #endif
  51 #if NEIGH_DEBUG >= 2
  52 #undef NEIGH_PRINTK2
  53 #define NEIGH_PRINTK2 NEIGH_PRINTK
  54 #endif
  55
  56 #define PNEIGH_HASHMASK         0xF
  57
  58 static void neigh_timer_handler(unsigned long arg);
  59 static void __neigh_notify(struct neighbour *n, int type, int flags);
  60 static void neigh_update_notify(struct neighbour *neigh);
  61 static int pneigh_ifdown(struct neigh_table *tbl, struct net_device *dev);
  62 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 const 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 static void neigh_cleanup_and_release(struct neighbour *neigh)
 109 {
 110         if (neigh->parms->neigh_cleanup)
 111                 neigh->parms->neigh_cleanup(neigh);
 112
 113         __neigh_notify(neigh, RTM_DELNEIGH, 0);
 114         neigh_release(neigh);
 115 }
 116
 117 /*
 118  * It is random distribution in the interval (1/2)*base...(3/2)*base.
 119  * It corresponds to default IPv6 settings and is not overridable,
 120  * because it is really reasonable choice.
 121  */
 122
 123 unsigned long neigh_rand_reach_time(unsigned long base)
 124 {
 125         return (base ? (net_random() % base) + (base >> 1) : 0);
 126 }
 127
 128
 129 static int neigh_forced_gc(struct neigh_table *tbl)
 130 {
 131         int shrunk = 0;
 132         int i;
 133
 134         NEIGH_CACHE_STAT_INC(tbl, forced_gc_runs);
 135
 136         write_lock_bh(&tbl->lock);
 137         for (i = 0; i <= tbl->hash_mask; i++) {
 138                 struct neighbour *n, **np;
 139
 140                 np = &tbl->hash_buckets[i];
 141                 while ((n = *np) != NULL) {
 142                         /* Neighbour record may be discarded if:
 143                          * - nobody refers to it.
 144                          * - it is not permanent
 145                          */
 146                         write_lock(&n->lock);
 147                         if (atomic_read(&n->refcnt) == 1 &&
 148                             !(n->nud_state & NUD_PERMANENT)) {
 149                                 *np     = n->next;
 150                                 n->dead = 1;
 151                                 shrunk  = 1;
 152                                 write_unlock(&n->lock);
 153                                 neigh_cleanup_and_release(n);
 154                                 continue;
 155                         }
 156                         write_unlock(&n->lock);
 157                         np = &n->next;
 158                 }
 159         }
 160
 161         tbl->last_flush = jiffies;
 162
 163         write_unlock_bh(&tbl->lock);
 164
 165         return shrunk;
 166 }
 167
 168 static int neigh_del_timer(struct neighbour *n)
 169 {
 170         if ((n->nud_state & NUD_IN_TIMER) &&
 171             del_timer(&n->timer)) {
 172                 neigh_release(n);
 173                 return 1;
 174         }
 175         return 0;
 176 }
 177
 178 static void pneigh_queue_purge(struct sk_buff_head *list)
 179 {
 180         struct sk_buff *skb;
 181
 182         while ((skb = skb_dequeue(list)) != NULL) {
 183                 dev_put(skb->dev);
 184                 kfree_skb(skb);
 185         }
 186 }
 187
 188 static void neigh_flush_dev(struct neigh_table *tbl, struct net_device *dev)
 189 {
 190         int i;
 191
 192         for (i = 0; i <= tbl->hash_mask; i++) {
 193                 struct neighbour *n, **np = &tbl->hash_buckets[i];
 194
 195                 while ((n = *np) != NULL) {
 196                         if (dev && n->dev != dev) {
 197                                 np = &n->next;
 198                                 continue;
 199                         }
 200                         *np = n->next;
 201                         write_lock(&n->lock);
 202                         neigh_del_timer(n);
 203                         n->dead = 1;
 204
 205                         if (atomic_read(&n->refcnt) != 1) {
 206                                 /* The most unpleasant situation.
 207                                    We must destroy neighbour entry,
 208                                    but someone still uses it.
 209
 210                                    The destroy will be delayed until
 211                                    the last user releases us, but
 212                                    we must kill timers etc. and move
 213                                    it to safe state.
 214                                  */
 215                                 skb_queue_purge(&n->arp_queue);
 216                                 n->output = neigh_blackhole;
 217                                 if (n->nud_state & NUD_VALID)
 218                                         n->nud_state = NUD_NOARP;
 219                                 else
 220                                         n->nud_state = NUD_NONE;
 221                                 NEIGH_PRINTK2("neigh %p is stray.\n", n);
 222                         }
 223                         write_unlock(&n->lock);
 224                         neigh_cleanup_and_release(n);
 225                 }
 226         }
 227 }
 228
 229 void neigh_changeaddr(struct neigh_table *tbl, struct net_device *dev)
 230 {
 231         write_lock_bh(&tbl->lock);
 232         neigh_flush_dev(tbl, dev);
 233         write_unlock_bh(&tbl->lock);
 234 }
 235
 236 int neigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
 237 {
 238         write_lock_bh(&tbl->lock);
 239         neigh_flush_dev(tbl, dev);
 240         pneigh_ifdown(tbl, dev);
 241         write_unlock_bh(&tbl->lock);
 242
 243         del_timer_sync(&tbl->proxy_timer);
 244         pneigh_queue_purge(&tbl->proxy_queue);
 245         return 0;
 246 }
 247
 248 static struct neighbour *neigh_alloc(struct neigh_table *tbl)
 249 {
 250         struct neighbour *n = NULL;
 251         unsigned long now = jiffies;
 252         int entries;
 253
 254         entries = atomic_inc_return(&tbl->entries) - 1;
 255         if (entries >= tbl->gc_thresh3 ||
 256             (entries >= tbl->gc_thresh2 &&
 257              time_after(now, tbl->last_flush + 5 * HZ))) {
 258                 if (!neigh_forced_gc(tbl) &&
 259                     entries >= tbl->gc_thresh3)
 260                         goto out_entries;
 261         }
 262
 263         n = kmem_cache_zalloc(tbl->kmem_cachep, GFP_ATOMIC);
 264         if (!n)
 265                 goto out_entries;
 266
 267         skb_queue_head_init(&n->arp_queue);
 268         rwlock_init(&n->lock);
 269         n->updated        = n->used = now;
 270         n->nud_state      = NUD_NONE;
 271         n->output         = neigh_blackhole;
 272         n->parms          = neigh_parms_clone(&tbl->parms);
 273         init_timer(&n->timer);
 274         n->timer.function = neigh_timer_handler;
 275         n->timer.data     = (unsigned long)n;
 276
 277         NEIGH_CACHE_STAT_INC(tbl, allocs);
 278         n->tbl            = tbl;
 279         atomic_set(&n->refcnt, 1);
 280         n->dead           = 1;
 281 out:
 282         return n;
 283
 284 out_entries:
 285         atomic_dec(&tbl->entries);
 286         goto out;
 287 }
 288
 289 static struct neighbour **neigh_hash_alloc(unsigned int entries)
 290 {
 291         unsigned long size = entries * sizeof(struct neighbour *);
 292         struct neighbour **ret;
 293
 294         if (size <= PAGE_SIZE) {
 295                 ret = kzalloc(size, GFP_ATOMIC);
 296         } else {
 297                 ret = (struct neighbour **)
 298                       __get_free_pages(GFP_ATOMIC|__GFP_ZERO, get_order(size));
 299         }
 300         return ret;
 301 }
 302
 303 static void neigh_hash_free(struct neighbour **hash, unsigned int entries)
 304 {
 305         unsigned long size = entries * sizeof(struct neighbour *);
 306
 307         if (size <= PAGE_SIZE)
 308                 kfree(hash);
 309         else
 310                 free_pages((unsigned long)hash, get_order(size));
 311 }
 312
 313 static void neigh_hash_grow(struct neigh_table *tbl, unsigned long new_entries)
 314 {
 315         struct neighbour **new_hash, **old_hash;
 316         unsigned int i, new_hash_mask, old_entries;
 317
 318         NEIGH_CACHE_STAT_INC(tbl, hash_grows);
 319
 320         BUG_ON(!is_power_of_2(new_entries));
 321         new_hash = neigh_hash_alloc(new_entries);
 322         if (!new_hash)
 323                 return;
 324
 325         old_entries = tbl->hash_mask + 1;
 326         new_hash_mask = new_entries - 1;
 327         old_hash = tbl->hash_buckets;
 328
 329         get_random_bytes(&tbl->hash_rnd, sizeof(tbl->hash_rnd));
 330         for (i = 0; i < old_entries; i++) {
 331                 struct neighbour *n, *next;
 332
 333                 for (n = old_hash[i]; n; n = next) {
 334                         unsigned int hash_val = tbl->hash(n->primary_key, n->dev);
 335
 336                         hash_val &= new_hash_mask;
 337                         next = n->next;
 338
 339                         n->next = new_hash[hash_val];
 340                         new_hash[hash_val] = n;
 341                 }
 342         }
 343         tbl->hash_buckets = new_hash;
 344         tbl->hash_mask = new_hash_mask;
 345
 346         neigh_hash_free(old_hash, old_entries);
 347 }
 348
 349 struct neighbour *neigh_lookup(struct neigh_table *tbl, const void *pkey,
 350                                struct net_device *dev)
 351 {
 352         struct neighbour *n;
 353         int key_len = tbl->key_len;
 354         u32 hash_val = tbl->hash(pkey, dev);
 355
 356         NEIGH_CACHE_STAT_INC(tbl, lookups);
 357
 358         read_lock_bh(&tbl->lock);
 359         for (n = tbl->hash_buckets[hash_val & tbl->hash_mask]; n; n = n->next) {
 360                 if (dev == n->dev && !memcmp(n->primary_key, pkey, key_len)) {
 361                         neigh_hold(n);
 362                         NEIGH_CACHE_STAT_INC(tbl, hits);
 363                         break;
 364                 }
 365         }
 366         read_unlock_bh(&tbl->lock);
 367         return n;
 368 }
 369
 370 struct neighbour *neigh_lookup_nodev(struct neigh_table *tbl, const void *pkey)
 371 {
 372         struct neighbour *n;
 373         int key_len = tbl->key_len;
 374         u32 hash_val = tbl->hash(pkey, NULL);
 375
 376         NEIGH_CACHE_STAT_INC(tbl, lookups);
 377
 378         read_lock_bh(&tbl->lock);
 379         for (n = tbl->hash_buckets[hash_val & tbl->hash_mask]; n; n = n->next) {
 380                 if (!memcmp(n->primary_key, pkey, key_len)) {
 381                         neigh_hold(n);
 382                         NEIGH_CACHE_STAT_INC(tbl, hits);
 383                         break;
 384                 }
 385         }
 386         read_unlock_bh(&tbl->lock);
 387         return n;
 388 }
 389
 390 struct neighbour *neigh_create(struct neigh_table *tbl, const void *pkey,
 391                                struct net_device *dev)
 392 {
 393         u32 hash_val;
 394         int key_len = tbl->key_len;
 395         int error;
 396         struct neighbour *n1, *rc, *n = neigh_alloc(tbl);
 397
 398         if (!n) {
 399                 rc = ERR_PTR(-ENOBUFS);
 400                 goto out;
 401         }
 402
 403         memcpy(n->primary_key, pkey, key_len);
 404         n->dev = dev;
 405         dev_hold(dev);
 406
 407         /* Protocol specific setup. */
 408         if (tbl->constructor && (error = tbl->constructor(n)) < 0) {
 409                 rc = ERR_PTR(error);
 410                 goto out_neigh_release;
 411         }
 412
 413         /* Device specific setup. */
 414         if (n->parms->neigh_setup &&
 415             (error = n->parms->neigh_setup(n)) < 0) {
 416                 rc = ERR_PTR(error);
 417                 goto out_neigh_release;
 418         }
 419
 420         n->confirmed = jiffies - (n->parms->base_reachable_time << 1);
 421
 422         write_lock_bh(&tbl->lock);
 423
 424         if (atomic_read(&tbl->entries) > (tbl->hash_mask + 1))
 425                 neigh_hash_grow(tbl, (tbl->hash_mask + 1) << 1);
 426
 427         hash_val = tbl->hash(pkey, dev) & tbl->hash_mask;
 428
 429         if (n->parms->dead) {
 430                 rc = ERR_PTR(-EINVAL);
 431                 goto out_tbl_unlock;
 432         }
 433
 434         for (n1 = tbl->hash_buckets[hash_val]; n1; n1 = n1->next) {
 435                 if (dev == n1->dev && !memcmp(n1->primary_key, pkey, key_len)) {
 436                         neigh_hold(n1);
 437                         rc = n1;
 438                         goto out_tbl_unlock;
 439                 }
 440         }
 441
 442         n->next = tbl->hash_buckets[hash_val];
 443         tbl->hash_buckets[hash_val] = n;
 444         n->dead = 0;
 445         neigh_hold(n);
 446         write_unlock_bh(&tbl->lock);
 447         NEIGH_PRINTK2("neigh %p is created.\n", n);
 448         rc = n;
 449 out:
 450         return rc;
 451 out_tbl_unlock:
 452         write_unlock_bh(&tbl->lock);
 453 out_neigh_release:
 454         neigh_release(n);
 455         goto out;
 456 }
 457
 458 struct pneigh_entry * pneigh_lookup(struct neigh_table *tbl, const void *pkey,
 459                                     struct net_device *dev, int creat)
 460 {
 461         struct pneigh_entry *n;
 462         int key_len = tbl->key_len;
 463         u32 hash_val = *(u32 *)(pkey + key_len - 4);
 464
 465         hash_val ^= (hash_val >> 16);
 466         hash_val ^= hash_val >> 8;
 467         hash_val ^= hash_val >> 4;
 468         hash_val &= PNEIGH_HASHMASK;
 469
 470         read_lock_bh(&tbl->lock);
 471
 472         for (n = tbl->phash_buckets[hash_val]; n; n = n->next) {
 473                 if (!memcmp(n->key, pkey, key_len) &&
 474                     (n->dev == dev || !n->dev)) {
 475                         read_unlock_bh(&tbl->lock);
 476                         goto out;
 477                 }
 478         }
 479         read_unlock_bh(&tbl->lock);
 480         n = NULL;
 481         if (!creat)
 482                 goto out;
 483
 484         ASSERT_RTNL();
 485
 486         n = kmalloc(sizeof(*n) + key_len, GFP_KERNEL);
 487         if (!n)
 488                 goto out;
 489
 490         memcpy(n->key, pkey, key_len);
 491         n->dev = dev;
 492         if (dev)
 493                 dev_hold(dev);
 494
 495         if (tbl->pconstructor && tbl->pconstructor(n)) {
 496                 if (dev)
 497                         dev_put(dev);
 498                 kfree(n);
 499                 n = NULL;
 500                 goto out;
 501         }
 502
 503         write_lock_bh(&tbl->lock);
 504         n->next = tbl->phash_buckets[hash_val];
 505         tbl->phash_buckets[hash_val] = n;
 506         write_unlock_bh(&tbl->lock);
 507 out:
 508         return n;
 509 }
 510
 511
 512 int pneigh_delete(struct neigh_table *tbl, const void *pkey,
 513                   struct net_device *dev)
 514 {
 515         struct pneigh_entry *n, **np;
 516         int key_len = tbl->key_len;
 517         u32 hash_val = *(u32 *)(pkey + key_len - 4);
 518
 519         hash_val ^= (hash_val >> 16);
 520         hash_val ^= hash_val >> 8;
 521         hash_val ^= hash_val >> 4;
 522         hash_val &= PNEIGH_HASHMASK;
 523
 524         write_lock_bh(&tbl->lock);
 525         for (np = &tbl->phash_buckets[hash_val]; (n = *np) != NULL;
 526              np = &n->next) {
 527                 if (!memcmp(n->key, pkey, key_len) && n->dev == dev) {
 528                         *np = n->next;
 529                         write_unlock_bh(&tbl->lock);
 530                         if (tbl->pdestructor)
 531                                 tbl->pdestructor(n);
 532                         if (n->dev)
 533                                 dev_put(n->dev);
 534                         kfree(n);
 535                         return 0;
 536                 }
 537         }
 538         write_unlock_bh(&tbl->lock);
 539         return -ENOENT;
 540 }
 541
 542 static int pneigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
 543 {
 544         struct pneigh_entry *n, **np;
 545         u32 h;
 546
 547         for (h = 0; h <= PNEIGH_HASHMASK; h++) {
 548                 np = &tbl->phash_buckets[h];
 549                 while ((n = *np) != NULL) {
 550                         if (!dev || n->dev == dev) {
 551                                 *np = n->next;
 552                                 if (tbl->pdestructor)
 553                                         tbl->pdestructor(n);
 554                                 if (n->dev)
 555                                         dev_put(n->dev);
 556                                 kfree(n);
 557                                 continue;
 558                         }
 559                         np = &n->next;
 560                 }
 561         }
 562         return -ENOENT;
 563 }
 564
 565
 566 /*
 567  *      neighbour must already be out of the table;
 568  *
 569  */
 570 void neigh_destroy(struct neighbour *neigh)
 571 {
 572         struct hh_cache *hh;
 573
 574         NEIGH_CACHE_STAT_INC(neigh->tbl, destroys);
 575
 576         if (!neigh->dead) {
 577                 printk(KERN_WARNING
 578                        "Destroying alive neighbour %p\n", neigh);
 579                 dump_stack();
 580                 return;
 581         }
 582
 583         if (neigh_del_timer(neigh))
 584                 printk(KERN_WARNING "Impossible event.\n");
 585
 586         while ((hh = neigh->hh) != NULL) {
 587                 neigh->hh = hh->hh_next;
 588                 hh->hh_next = NULL;
 589
 590                 write_seqlock_bh(&hh->hh_lock);
 591                 hh->hh_output = neigh_blackhole;
 592                 write_sequnlock_bh(&hh->hh_lock);
 593                 if (atomic_dec_and_test(&hh->hh_refcnt))
 594                         kfree(hh);
 595         }
 596
 597         skb_queue_purge(&neigh->arp_queue);
 598
 599         dev_put(neigh->dev);
 600         neigh_parms_put(neigh->parms);
 601
 602         NEIGH_PRINTK2("neigh %p is destroyed.\n", neigh);
 603
 604         atomic_dec(&neigh->tbl->entries);
 605         kmem_cache_free(neigh->tbl->kmem_cachep, neigh);
 606 }
 607
 608 /* Neighbour state is suspicious;
 609    disable fast path.
 610
 611    Called with write_locked neigh.
 612  */
 613 static void neigh_suspect(struct neighbour *neigh)
 614 {
 615         struct hh_cache *hh;
 616
 617         NEIGH_PRINTK2("neigh %p is suspected.\n", neigh);
 618
 619         neigh->output = neigh->ops->output;
 620
 621         for (hh = neigh->hh; hh; hh = hh->hh_next)
 622                 hh->hh_output = neigh->ops->output;
 623 }
 624
 625 /* Neighbour state is OK;
 626    enable fast path.
 627
 628    Called with write_locked neigh.
 629  */
 630 static void neigh_connect(struct neighbour *neigh)
 631 {
 632         struct hh_cache *hh;
 633
 634         NEIGH_PRINTK2("neigh %p is connected.\n", neigh);
 635
 636         neigh->output = neigh->ops->connected_output;
 637
 638         for (hh = neigh->hh; hh; hh = hh->hh_next)
 639                 hh->hh_output = neigh->ops->hh_output;
 640 }
 641
 642 static void neigh_periodic_timer(unsigned long arg)
 643 {
 644         struct neigh_table *tbl = (struct neigh_table *)arg;
 645         struct neighbour *n, **np;
 646         unsigned long expire, now = jiffies;
 647
 648         NEIGH_CACHE_STAT_INC(tbl, periodic_gc_runs);
 649
 650         write_lock(&tbl->lock);
 651
 652         /*
 653          *      periodically recompute ReachableTime from random function
 654          */
 655
 656         if (time_after(now, tbl->last_rand + 300 * HZ)) {
 657                 struct neigh_parms *p;
 658                 tbl->last_rand = now;
 659                 for (p = &tbl->parms; p; p = p->next)
 660                         p->reachable_time =
 661                                 neigh_rand_reach_time(p->base_reachable_time);
 662         }
 663
 664         np = &tbl->hash_buckets[tbl->hash_chain_gc];
 665         tbl->hash_chain_gc = ((tbl->hash_chain_gc + 1) & tbl->hash_mask);
 666
 667         while ((n = *np) != NULL) {
 668                 unsigned int state;
 669
 670                 write_lock(&n->lock);
 671
 672                 state = n->nud_state;
 673                 if (state & (NUD_PERMANENT | NUD_IN_TIMER)) {
 674                         write_unlock(&n->lock);
 675                         goto next_elt;
 676                 }
 677
 678                 if (time_before(n->used, n->confirmed))
 679                         n->used = n->confirmed;
 680
 681                 if (atomic_read(&n->refcnt) == 1 &&
 682                     (state == NUD_FAILED ||
 683                      time_after(now, n->used + n->parms->gc_staletime))) {
 684                         *np = n->next;
 685                         n->dead = 1;
 686                         write_unlock(&n->lock);
 687                         neigh_cleanup_and_release(n);
 688                         continue;
 689                 }
 690                 write_unlock(&n->lock);
 691
 692 next_elt:
 693                 np = &n->next;
 694         }
 695
 696         /* Cycle through all hash buckets every base_reachable_time/2 ticks.
 697          * ARP entry timeouts range from 1/2 base_reachable_time to 3/2
 698          * base_reachable_time.
 699          */
 700         expire = tbl->parms.base_reachable_time >> 1;
 701         expire /= (tbl->hash_mask + 1);
 702         if (!expire)
 703                 expire = 1;
 704
 705         if (expire>HZ)
 706                 mod_timer(&tbl->gc_timer, round_jiffies(now + expire));
 707         else
 708                 mod_timer(&tbl->gc_timer, now + expire);
 709
 710         write_unlock(&tbl->lock);
 711 }
 712
 713 static __inline__ int neigh_max_probes(struct neighbour *n)
 714 {
 715         struct neigh_parms *p = n->parms;
 716         return (n->nud_state & NUD_PROBE ?
 717                 p->ucast_probes :
 718                 p->ucast_probes + p->app_probes + p->mcast_probes);
 719 }
 720
 721 static inline void neigh_add_timer(struct neighbour *n, unsigned long when)
 722 {
 723         if (unlikely(mod_timer(&n->timer, when))) {
 724                 printk("NEIGH: BUG, double timer add, state is %x\n",
 725                        n->nud_state);
 726                 dump_stack();
 727         }
 728 }
 729
 730 /* Called when a timer expires for a neighbour entry. */
 731
 732 static void neigh_timer_handler(unsigned long arg)
 733 {
 734         unsigned long now, next;
 735         struct neighbour *neigh = (struct neighbour *)arg;
 736         unsigned state;
 737         int notify = 0;
 738
 739         write_lock(&neigh->lock);
 740
 741         state = neigh->nud_state;
 742         now = jiffies;
 743         next = now + HZ;
 744
 745         if (!(state & NUD_IN_TIMER)) {
 746 #ifndef CONFIG_SMP
 747                 printk(KERN_WARNING "neigh: timer & !nud_in_timer\n");
 748 #endif
 749                 goto out;
 750         }
 751
 752         if (state & NUD_REACHABLE) {
 753                 if (time_before_eq(now,
 754                                    neigh->confirmed + neigh->parms->reachable_time)) {
 755                         NEIGH_PRINTK2("neigh %p is still alive.\n", neigh);
 756                         next = neigh->confirmed + neigh->parms->reachable_time;
 757                 } else if (time_before_eq(now,
 758                                           neigh->used + neigh->parms->delay_probe_time)) {
 759                         NEIGH_PRINTK2("neigh %p is delayed.\n", neigh);
 760                         neigh->nud_state = NUD_DELAY;
 761                         neigh->updated = jiffies;
 762                         neigh_suspect(neigh);
 763                         next = now + neigh->parms->delay_probe_time;
 764                 } else {
 765                         NEIGH_PRINTK2("neigh %p is suspected.\n", neigh);
 766                         neigh->nud_state = NUD_STALE;
 767                         neigh->updated = jiffies;
 768                         neigh_suspect(neigh);
 769                         notify = 1;
 770                 }
 771         } else if (state & NUD_DELAY) {
 772                 if (time_before_eq(now,
 773                                    neigh->confirmed + neigh->parms->delay_probe_time)) {
 774                         NEIGH_PRINTK2("neigh %p is now reachable.\n", neigh);
 775                         neigh->nud_state = NUD_REACHABLE;
 776                         neigh->updated = jiffies;
 777                         neigh_connect(neigh);
 778                         notify = 1;
 779                         next = neigh->confirmed + neigh->parms->reachable_time;
 780                 } else {
 781                         NEIGH_PRINTK2("neigh %p is probed.\n", neigh);
 782                         neigh->nud_state = NUD_PROBE;
 783                         neigh->updated = jiffies;
 784                         atomic_set(&neigh->probes, 0);
 785                         next = now + neigh->parms->retrans_time;
 786                 }
 787         } else {
 788                 /* NUD_PROBE|NUD_INCOMPLETE */
 789                 next = now + neigh->parms->retrans_time;
 790         }
 791
 792         if ((neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) &&
 793             atomic_read(&neigh->probes) >= neigh_max_probes(neigh)) {
 794                 struct sk_buff *skb;
 795
 796                 neigh->nud_state = NUD_FAILED;
 797                 neigh->updated = jiffies;
 798                 notify = 1;
 799                 NEIGH_CACHE_STAT_INC(neigh->tbl, res_failed);
 800                 NEIGH_PRINTK2("neigh %p is failed.\n", neigh);
 801
 802                 /* It is very thin place. report_unreachable is very complicated
 803                    routine. Particularly, it can hit the same neighbour entry!
 804
 805                    So that, we try to be accurate and avoid dead loop. --ANK
 806                  */
 807                 while (neigh->nud_state == NUD_FAILED &&
 808                        (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
 809                         write_unlock(&neigh->lock);
 810                         neigh->ops->error_report(neigh, skb);
 811                         write_lock(&neigh->lock);
 812                 }
 813                 skb_queue_purge(&neigh->arp_queue);
 814         }
 815
 816         if (neigh->nud_state & NUD_IN_TIMER) {
 817                 if (time_before(next, jiffies + HZ/2))
 818                         next = jiffies + HZ/2;
 819                 if (!mod_timer(&neigh->timer, next))
 820                         neigh_hold(neigh);
 821         }
 822         if (neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) {
 823                 struct sk_buff *skb = skb_peek(&neigh->arp_queue);
 824                 /* keep skb alive even if arp_queue overflows */
 825                 if (skb)
 826                         skb_get(skb);
 827                 write_unlock(&neigh->lock);
 828                 neigh->ops->solicit(neigh, skb);
 829                 atomic_inc(&neigh->probes);
 830                 if (skb)
 831                         kfree_skb(skb);
 832         } else {
 833 out:
 834                 write_unlock(&neigh->lock);
 835         }
 836
 837         if (notify)
 838                 neigh_update_notify(neigh);
 839
 840         neigh_release(neigh);
 841 }
 842
 843 int __neigh_event_send(struct neighbour *neigh, struct sk_buff *skb)
 844 {
 845         int rc;
 846         unsigned long now;
 847
 848         write_lock_bh(&neigh->lock);
 849
 850         rc = 0;
 851         if (neigh->nud_state & (NUD_CONNECTED | NUD_DELAY | NUD_PROBE))
 852                 goto out_unlock_bh;
 853
 854         now = jiffies;
 855
 856         if (!(neigh->nud_state & (NUD_STALE | NUD_INCOMPLETE))) {
 857                 if (neigh->parms->mcast_probes + neigh->parms->app_probes) {
 858                         atomic_set(&neigh->probes, neigh->parms->ucast_probes);
 859                         neigh->nud_state     = NUD_INCOMPLETE;
 860                         neigh->updated = jiffies;
 861                         neigh_hold(neigh);
 862                         neigh_add_timer(neigh, now + 1);
 863                 } else {
 864                         neigh->nud_state = NUD_FAILED;
 865                         neigh->updated = jiffies;
 866                         write_unlock_bh(&neigh->lock);
 867
 868                         if (skb)
 869                                 kfree_skb(skb);
 870                         return 1;
 871                 }
 872         } else if (neigh->nud_state & NUD_STALE) {
 873                 NEIGH_PRINTK2("neigh %p is delayed.\n", neigh);
 874                 neigh_hold(neigh);
 875                 neigh->nud_state = NUD_DELAY;
 876                 neigh->updated = jiffies;
 877                 neigh_add_timer(neigh,
 878                                 jiffies + neigh->parms->delay_probe_time);
 879         }
 880
 881         if (neigh->nud_state == NUD_INCOMPLETE) {
 882                 if (skb) {
 883                         if (skb_queue_len(&neigh->arp_queue) >=
 884                             neigh->parms->queue_len) {
 885                                 struct sk_buff *buff;
 886                                 buff = neigh->arp_queue.next;
 887                                 __skb_unlink(buff, &neigh->arp_queue);
 888                                 kfree_skb(buff);
 889                         }
 890                         __skb_queue_tail(&neigh->arp_queue, skb);
 891                 }
 892                 rc = 1;
 893         }
 894 out_unlock_bh:
 895         write_unlock_bh(&neigh->lock);
 896         return rc;
 897 }
 898
 899 static void neigh_update_hhs(struct neighbour *neigh)
 900 {
 901         struct hh_cache *hh;
 902         void (*update)(struct hh_cache*, const struct net_device*, const unsigned char *)
 903                 = neigh->dev->header_ops->cache_update;
 904
 905         if (update) {
 906                 for (hh = neigh->hh; hh; hh = hh->hh_next) {
 907                         write_seqlock_bh(&hh->hh_lock);
 908                         update(hh, neigh->dev, neigh->ha);
 909                         write_sequnlock_bh(&hh->hh_lock);
 910                 }
 911         }
 912 }
 913
 914
 915
 916 /* Generic update routine.
 917    -- lladdr is new lladdr or NULL, if it is not supplied.
 918    -- new    is new state.
 919    -- flags
 920         NEIGH_UPDATE_F_OVERRIDE allows to override existing lladdr,
 921                                 if it is different.
 922         NEIGH_UPDATE_F_WEAK_OVERRIDE will suspect existing "connected"
 923                                 lladdr instead of overriding it
 924                                 if it is different.
 925                                 It also allows to retain current state
 926                                 if lladdr is unchanged.
 927         NEIGH_UPDATE_F_ADMIN    means that the change is administrative.
 928
 929         NEIGH_UPDATE_F_OVERRIDE_ISROUTER allows to override existing
 930                                 NTF_ROUTER flag.
 931         NEIGH_UPDATE_F_ISROUTER indicates if the neighbour is known as
 932                                 a router.
 933
 934    Caller MUST hold reference count on the entry.
 935  */
 936
 937 int neigh_update(struct neighbour *neigh, const u8 *lladdr, u8 new,
 938                  u32 flags)
 939 {
 940         u8 old;
 941         int err;
 942         int notify = 0;
 943         struct net_device *dev;
 944         int update_isrouter = 0;
 945
 946         write_lock_bh(&neigh->lock);
 947
 948         dev    = neigh->dev;
 949         old    = neigh->nud_state;
 950         err    = -EPERM;
 951
 952         if (!(flags & NEIGH_UPDATE_F_ADMIN) &&
 953             (old & (NUD_NOARP | NUD_PERMANENT)))
 954                 goto out;
 955
 956         if (!(new & NUD_VALID)) {
 957                 neigh_del_timer(neigh);
 958                 if (old & NUD_CONNECTED)
 959                         neigh_suspect(neigh);
 960                 neigh->nud_state = new;
 961                 err = 0;
 962                 notify = old & NUD_VALID;
 963                 goto out;
 964         }
 965
 966         /* Compare new lladdr with cached one */
 967         if (!dev->addr_len) {
 968                 /* First case: device needs no address. */
 969                 lladdr = neigh->ha;
 970         } else if (lladdr) {
 971                 /* The second case: if something is already cached
 972                    and a new address is proposed:
 973                    - compare new & old
 974                    - if they are different, check override flag
 975                  */
 976                 if ((old & NUD_VALID) &&
 977                     !memcmp(lladdr, neigh->ha, dev->addr_len))
 978                         lladdr = neigh->ha;
 979         } else {
 980                 /* No address is supplied; if we know something,
 981                    use it, otherwise discard the request.
 982                  */
 983                 err = -EINVAL;
 984                 if (!(old & NUD_VALID))
 985                         goto out;
 986                 lladdr = neigh->ha;
 987         }
 988
 989         if (new & NUD_CONNECTED)
 990                 neigh->confirmed = jiffies;
 991         neigh->updated = jiffies;
 992
 993         /* If entry was valid and address is not changed,
 994            do not change entry state, if new one is STALE.
 995          */
 996         err = 0;
 997         update_isrouter = flags & NEIGH_UPDATE_F_OVERRIDE_ISROUTER;
 998         if (old & NUD_VALID) {
 999                 if (lladdr != neigh->ha && !(flags & NEIGH_UPDATE_F_OVERRIDE)) {
1000                         update_isrouter = 0;
1001                         if ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) &&
1002                             (old & NUD_CONNECTED)) {
1003                                 lladdr = neigh->ha;
1004                                 new = NUD_STALE;
1005                         } else
1006                                 goto out;
1007                 } else {
1008                         if (lladdr == neigh->ha && new == NUD_STALE &&
1009                             ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) ||
1010                              (old & NUD_CONNECTED))
1011                             )
1012                                 new = old;
1013                 }
1014         }
1015
1016         if (new != old) {
1017                 neigh_del_timer(neigh);
1018                 if (new & NUD_IN_TIMER) {
1019                         neigh_hold(neigh);
1020                         neigh_add_timer(neigh, (jiffies +
1021                                                 ((new & NUD_REACHABLE) ?
1022                                                  neigh->parms->reachable_time :
1023                                                  0)));
1024                 }
1025                 neigh->nud_state = new;
1026         }
1027
1028         if (lladdr != neigh->ha) {
1029                 memcpy(&neigh->ha, lladdr, dev->addr_len);
1030                 neigh_update_hhs(neigh);
1031                 if (!(new & NUD_CONNECTED))
1032                         neigh->confirmed = jiffies -
1033                                       (neigh->parms->base_reachable_time << 1);
1034                 notify = 1;
1035         }
1036         if (new == old)
1037                 goto out;
1038         if (new & NUD_CONNECTED)
1039                 neigh_connect(neigh);
1040         else
1041                 neigh_suspect(neigh);
1042         if (!(old & NUD_VALID)) {
1043                 struct sk_buff *skb;
1044
1045                 /* Again: avoid dead loop if something went wrong */
1046
1047                 while (neigh->nud_state & NUD_VALID &&
1048                        (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
1049                         struct neighbour *n1 = neigh;
1050                         write_unlock_bh(&neigh->lock);
1051                         /* On shaper/eql skb->dst->neighbour != neigh :( */
1052                         if (skb->dst && skb->dst->neighbour)
1053                                 n1 = skb->dst->neighbour;
1054                         n1->output(skb);
1055                         write_lock_bh(&neigh->lock);
1056                 }
1057                 skb_queue_purge(&neigh->arp_queue);
1058         }
1059 out:
1060         if (update_isrouter) {
1061                 neigh->flags = (flags & NEIGH_UPDATE_F_ISROUTER) ?
1062                         (neigh->flags | NTF_ROUTER) :
1063                         (neigh->flags & ~NTF_ROUTER);
1064         }
1065         write_unlock_bh(&neigh->lock);
1066
1067         if (notify)
1068                 neigh_update_notify(neigh);
1069
1070         return err;
1071 }
1072
1073 struct neighbour *neigh_event_ns(struct neigh_table *tbl,
1074                                  u8 *lladdr, void *saddr,
1075                                  struct net_device *dev)
1076 {
1077         struct neighbour *neigh = __neigh_lookup(tbl, saddr, dev,
1078                                                  lladdr || !dev->addr_len);
1079         if (neigh)
1080                 neigh_update(neigh, lladdr, NUD_STALE,
1081                              NEIGH_UPDATE_F_OVERRIDE);
1082         return neigh;
1083 }
1084
1085 static void neigh_hh_init(struct neighbour *n, struct dst_entry *dst,
1086                           __be16 protocol)
1087 {
1088         struct hh_cache *hh;
1089         struct net_device *dev = dst->dev;
1090
1091         for (hh = n->hh; hh; hh = hh->hh_next)
1092                 if (hh->hh_type == protocol)
1093                         break;
1094
1095         if (!hh && (hh = kzalloc(sizeof(*hh), GFP_ATOMIC)) != NULL) {
1096                 seqlock_init(&hh->hh_lock);
1097                 hh->hh_type = protocol;
1098                 atomic_set(&hh->hh_refcnt, 0);
1099                 hh->hh_next = NULL;
1100
1101                 if (dev->header_ops->cache(n, hh)) {
1102                         kfree(hh);
1103                         hh = NULL;
1104                 } else {
1105                         atomic_inc(&hh->hh_refcnt);
1106                         hh->hh_next = n->hh;
1107                         n->hh       = hh;
1108                         if (n->nud_state & NUD_CONNECTED)
1109                                 hh->hh_output = n->ops->hh_output;
1110                         else
1111                                 hh->hh_output = n->ops->output;
1112                 }
1113         }
1114         if (hh) {
1115                 atomic_inc(&hh->hh_refcnt);
1116                 dst->hh = hh;
1117         }
1118 }
1119
1120 /* This function can be used in contexts, where only old dev_queue_xmit
1121    worked, f.e. if you want to override normal output path (eql, shaper),
1122    but resolution is not made yet.
1123  */
1124
1125 int neigh_compat_output(struct sk_buff *skb)
1126 {
1127         struct net_device *dev = skb->dev;
1128
1129         __skb_pull(skb, skb_network_offset(skb));
1130
1131         if (dev_hard_header(skb, dev, ntohs(skb->protocol), NULL, NULL,
1132                             skb->len) < 0 &&
1133             dev->header_ops->rebuild(skb))
1134                 return 0;
1135
1136         return dev_queue_xmit(skb);
1137 }
1138
1139 /* Slow and careful. */
1140
1141 int neigh_resolve_output(struct sk_buff *skb)
1142 {
1143         struct dst_entry *dst = skb->dst;
1144         struct neighbour *neigh;
1145         int rc = 0;
1146
1147         if (!dst || !(neigh = dst->neighbour))
1148                 goto discard;
1149
1150         __skb_pull(skb, skb_network_offset(skb));
1151
1152         if (!neigh_event_send(neigh, skb)) {
1153                 int err;
1154                 struct net_device *dev = neigh->dev;
1155                 if (dev->header_ops->cache && !dst->hh) {
1156                         write_lock_bh(&neigh->lock);
1157                         if (!dst->hh)
1158                                 neigh_hh_init(neigh, dst, dst->ops->protocol);
1159                         err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1160                                               neigh->ha, NULL, skb->len);
1161                         write_unlock_bh(&neigh->lock);
1162                 } else {
1163                         read_lock_bh(&neigh->lock);
1164                         err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1165                                               neigh->ha, NULL, skb->len);
1166                         read_unlock_bh(&neigh->lock);
1167                 }
1168                 if (err >= 0)
1169                         rc = neigh->ops->queue_xmit(skb);
1170                 else
1171                         goto out_kfree_skb;
1172         }
1173 out:
1174         return rc;
1175 discard:
1176         NEIGH_PRINTK1("neigh_resolve_output: dst=%p neigh=%p\n",
1177                       dst, dst ? dst->neighbour : NULL);
1178 out_kfree_skb:
1179         rc = -EINVAL;
1180         kfree_skb(skb);
1181         goto out;
1182 }
1183
1184 /* As fast as possible without hh cache */
1185
1186 int neigh_connected_output(struct sk_buff *skb)
1187 {
1188         int err;
1189         struct dst_entry *dst = skb->dst;
1190         struct neighbour *neigh = dst->neighbour;
1191         struct net_device *dev = neigh->dev;
1192
1193         __skb_pull(skb, skb_network_offset(skb));
1194
1195         read_lock_bh(&neigh->lock);
1196         err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1197                               neigh->ha, NULL, skb->len);
1198         read_unlock_bh(&neigh->lock);
1199         if (err >= 0)
1200                 err = neigh->ops->queue_xmit(skb);
1201         else {
1202                 err = -EINVAL;
1203                 kfree_skb(skb);
1204         }
1205         return err;
1206 }
1207
1208 static void neigh_proxy_process(unsigned long arg)
1209 {
1210         struct neigh_table *tbl = (struct neigh_table *)arg;
1211         long sched_next = 0;
1212         unsigned long now = jiffies;
1213         struct sk_buff *skb;
1214
1215         spin_lock(&tbl->proxy_queue.lock);
1216
1217         skb = tbl->proxy_queue.next;
1218
1219         while (skb != (struct sk_buff *)&tbl->proxy_queue) {
1220                 struct sk_buff *back = skb;
1221                 long tdif = NEIGH_CB(back)->sched_next - now;
1222
1223                 skb = skb->next;
1224                 if (tdif <= 0) {
1225                         struct net_device *dev = back->dev;
1226                         __skb_unlink(back, &tbl->proxy_queue);
1227                         if (tbl->proxy_redo && netif_running(dev))
1228                                 tbl->proxy_redo(back);
1229                         else
1230                                 kfree_skb(back);
1231
1232                         dev_put(dev);
1233                 } else if (!sched_next || tdif < sched_next)
1234                         sched_next = tdif;
1235         }
1236         del_timer(&tbl->proxy_timer);
1237         if (sched_next)
1238                 mod_timer(&tbl->proxy_timer, jiffies + sched_next);
1239         spin_unlock(&tbl->proxy_queue.lock);
1240 }
1241
1242 void pneigh_enqueue(struct neigh_table *tbl, struct neigh_parms *p,
1243                     struct sk_buff *skb)
1244 {
1245         unsigned long now = jiffies;
1246         unsigned long sched_next = now + (net_random() % p->proxy_delay);
1247
1248         if (tbl->proxy_queue.qlen > p->proxy_qlen) {
1249                 kfree_skb(skb);
1250                 return;
1251         }
1252
1253         NEIGH_CB(skb)->sched_next = sched_next;
1254         NEIGH_CB(skb)->flags |= LOCALLY_ENQUEUED;
1255
1256         spin_lock(&tbl->proxy_queue.lock);
1257         if (del_timer(&tbl->proxy_timer)) {
1258                 if (time_before(tbl->proxy_timer.expires, sched_next))
1259                         sched_next = tbl->proxy_timer.expires;
1260         }
1261         dst_release(skb->dst);
1262         skb->dst = NULL;
1263         dev_hold(skb->dev);
1264         __skb_queue_tail(&tbl->proxy_queue, skb);
1265         mod_timer(&tbl->proxy_timer, sched_next);
1266         spin_unlock(&tbl->proxy_queue.lock);
1267 }
1268
1269
1270 struct neigh_parms *neigh_parms_alloc(struct net_device *dev,
1271                                       struct neigh_table *tbl)
1272 {
1273         struct neigh_parms *p = kmemdup(&tbl->parms, sizeof(*p), GFP_KERNEL);
1274
1275         if (p) {
1276                 p->tbl            = tbl;
1277                 atomic_set(&p->refcnt, 1);
1278                 INIT_RCU_HEAD(&p->rcu_head);
1279                 p->reachable_time =
1280                                 neigh_rand_reach_time(p->base_reachable_time);
1281                 if (dev) {
1282                         if (dev->neigh_setup && dev->neigh_setup(dev, p)) {
1283                                 kfree(p);
1284                                 return NULL;
1285                         }
1286
1287                         dev_hold(dev);
1288                         p->dev = dev;
1289                 }
1290                 p->sysctl_table = NULL;
1291                 write_lock_bh(&tbl->lock);
1292                 p->next         = tbl->parms.next;
1293                 tbl->parms.next = p;
1294                 write_unlock_bh(&tbl->lock);
1295         }
1296         return p;
1297 }
1298
1299 static void neigh_rcu_free_parms(struct rcu_head *head)
1300 {
1301         struct neigh_parms *parms =
1302                 container_of(head, struct neigh_parms, rcu_head);
1303
1304         neigh_parms_put(parms);
1305 }
1306
1307 void neigh_parms_release(struct neigh_table *tbl, struct neigh_parms *parms)
1308 {
1309         struct neigh_parms **p;
1310
1311         if (!parms || parms == &tbl->parms)
1312                 return;
1313         write_lock_bh(&tbl->lock);
1314         for (p = &tbl->parms.next; *p; p = &(*p)->next) {
1315                 if (*p == parms) {
1316                         *p = parms->next;
1317                         parms->dead = 1;
1318                         write_unlock_bh(&tbl->lock);
1319                         if (parms->dev)
1320                                 dev_put(parms->dev);
1321                         call_rcu(&parms->rcu_head, neigh_rcu_free_parms);
1322                         return;
1323                 }
1324         }
1325         write_unlock_bh(&tbl->lock);
1326         NEIGH_PRINTK1("neigh_parms_release: not found\n");
1327 }
1328
1329 void neigh_parms_destroy(struct neigh_parms *parms)
1330 {
1331         kfree(parms);
1332 }
1333
1334 static struct lock_class_key neigh_table_proxy_queue_class;
1335
1336 void neigh_table_init_no_netlink(struct neigh_table *tbl)
1337 {
1338         unsigned long now = jiffies;
1339         unsigned long phsize;
1340
1341         atomic_set(&tbl->parms.refcnt, 1);
1342         INIT_RCU_HEAD(&tbl->parms.rcu_head);
1343         tbl->parms.reachable_time =
1344                           neigh_rand_reach_time(tbl->parms.base_reachable_time);
1345
1346         if (!tbl->kmem_cachep)
1347                 tbl->kmem_cachep =
1348                         kmem_cache_create(tbl->id, tbl->entry_size, 0,
1349                                           SLAB_HWCACHE_ALIGN|SLAB_PANIC,
1350                                           NULL);
1351         tbl->stats = alloc_percpu(struct neigh_statistics);
1352         if (!tbl->stats)
1353                 panic("cannot create neighbour cache statistics");
1354
1355 #ifdef CONFIG_PROC_FS
1356         tbl->pde = create_proc_entry(tbl->id, 0, init_net.proc_net_stat);
1357         if (!tbl->pde)
1358                 panic("cannot create neighbour proc dir entry");
1359         tbl->pde->proc_fops = &neigh_stat_seq_fops;
1360         tbl->pde->data = tbl;
1361 #endif
1362
1363         tbl->hash_mask = 1;
1364         tbl->hash_buckets = neigh_hash_alloc(tbl->hash_mask + 1);
1365
1366         phsize = (PNEIGH_HASHMASK + 1) * sizeof(struct pneigh_entry *);
1367         tbl->phash_buckets = kzalloc(phsize, GFP_KERNEL);
1368
1369         if (!tbl->hash_buckets || !tbl->phash_buckets)
1370                 panic("cannot allocate neighbour cache hashes");
1371
1372         get_random_bytes(&tbl->hash_rnd, sizeof(tbl->hash_rnd));
1373
1374         rwlock_init(&tbl->lock);
1375         init_timer(&tbl->gc_timer);
1376         tbl->gc_timer.data     = (unsigned long)tbl;
1377         tbl->gc_timer.function = neigh_periodic_timer;
1378         tbl->gc_timer.expires  = now + 1;
1379         add_timer(&tbl->gc_timer);
1380
1381         init_timer(&tbl->proxy_timer);
1382         tbl->proxy_timer.data     = (unsigned long)tbl;
1383         tbl->proxy_timer.function = neigh_proxy_process;
1384         skb_queue_head_init_class(&tbl->proxy_queue,
1385                         &neigh_table_proxy_queue_class);
1386
1387         tbl->last_flush = now;
1388         tbl->last_rand  = now + tbl->parms.reachable_time * 20;
1389 }
1390
1391 void neigh_table_init(struct neigh_table *tbl)
1392 {
1393         struct neigh_table *tmp;
1394
1395         neigh_table_init_no_netlink(tbl);
1396         write_lock(&neigh_tbl_lock);
1397         for (tmp = neigh_tables; tmp; tmp = tmp->next) {
1398                 if (tmp->family == tbl->family)
1399                         break;
1400         }
1401         tbl->next       = neigh_tables;
1402         neigh_tables    = tbl;
1403         write_unlock(&neigh_tbl_lock);
1404
1405         if (unlikely(tmp)) {
1406                 printk(KERN_ERR "NEIGH: Registering multiple tables for "
1407                        "family %d\n", tbl->family);
1408                 dump_stack();
1409         }
1410 }
1411
1412 int neigh_table_clear(struct neigh_table *tbl)
1413 {
1414         struct neigh_table **tp;
1415
1416         /* It is not clean... Fix it to unload IPv6 module safely */
1417         del_timer_sync(&tbl->gc_timer);
1418         del_timer_sync(&tbl->proxy_timer);
1419         pneigh_queue_purge(&tbl->proxy_queue);
1420         neigh_ifdown(tbl, NULL);
1421         if (atomic_read(&tbl->entries))
1422                 printk(KERN_CRIT "neighbour leakage\n");
1423         write_lock(&neigh_tbl_lock);
1424         for (tp = &neigh_tables; *tp; tp = &(*tp)->next) {
1425                 if (*tp == tbl) {
1426                         *tp = tbl->next;
1427                         break;
1428                 }
1429         }
1430         write_unlock(&neigh_tbl_lock);
1431
1432         neigh_hash_free(tbl->hash_buckets, tbl->hash_mask + 1);
1433         tbl->hash_buckets = NULL;
1434
1435         kfree(tbl->phash_buckets);
1436         tbl->phash_buckets = NULL;
1437
1438         remove_proc_entry(tbl->id, init_net.proc_net_stat);
1439
1440         free_percpu(tbl->stats);
1441         tbl->stats = NULL;
1442
1443         kmem_cache_destroy(tbl->kmem_cachep);
1444         tbl->kmem_cachep = NULL;
1445
1446         return 0;
1447 }
1448
1449 static int neigh_delete(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
1450 {
1451         struct net *net = skb->sk->sk_net;
1452         struct ndmsg *ndm;
1453         struct nlattr *dst_attr;
1454         struct neigh_table *tbl;
1455         struct net_device *dev = NULL;
1456         int err = -EINVAL;
1457
1458         if (nlmsg_len(nlh) < sizeof(*ndm))
1459                 goto out;
1460
1461         dst_attr = nlmsg_find_attr(nlh, sizeof(*ndm), NDA_DST);
1462         if (dst_attr == NULL)
1463                 goto out;
1464
1465         ndm = nlmsg_data(nlh);
1466         if (ndm->ndm_ifindex) {
1467                 dev = dev_get_by_index(net, ndm->ndm_ifindex);
1468                 if (dev == NULL) {
1469                         err = -ENODEV;
1470                         goto out;
1471                 }
1472         }
1473
1474         read_lock(&neigh_tbl_lock);
1475         for (tbl = neigh_tables; tbl; tbl = tbl->next) {
1476                 struct neighbour *neigh;
1477
1478                 if (tbl->family != ndm->ndm_family)
1479                         continue;
1480                 read_unlock(&neigh_tbl_lock);
1481
1482                 if (nla_len(dst_attr) < tbl->key_len)
1483                         goto out_dev_put;
1484
1485                 if (ndm->ndm_flags & NTF_PROXY) {
1486                         err = pneigh_delete(tbl, nla_data(dst_attr), dev);
1487                         goto out_dev_put;
1488                 }
1489
1490                 if (dev == NULL)
1491                         goto out_dev_put;
1492
1493                 neigh = neigh_lookup(tbl, nla_data(dst_attr), dev);
1494                 if (neigh == NULL) {
1495                         err = -ENOENT;
1496                         goto out_dev_put;
1497                 }
1498
1499                 err = neigh_update(neigh, NULL, NUD_FAILED,
1500                                    NEIGH_UPDATE_F_OVERRIDE |
1501                                    NEIGH_UPDATE_F_ADMIN);
1502                 neigh_release(neigh);
1503                 goto out_dev_put;
1504         }
1505         read_unlock(&neigh_tbl_lock);
1506         err = -EAFNOSUPPORT;
1507
1508 out_dev_put:
1509         if (dev)
1510                 dev_put(dev);
1511 out:
1512         return err;
1513 }
1514
1515 static int neigh_add(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
1516 {
1517         struct net *net = skb->sk->sk_net;
1518         struct ndmsg *ndm;
1519         struct nlattr *tb[NDA_MAX+1];
1520         struct neigh_table *tbl;
1521         struct net_device *dev = NULL;
1522         int err;
1523
1524         err = nlmsg_parse(nlh, sizeof(*ndm), tb, NDA_MAX, NULL);
1525         if (err < 0)
1526                 goto out;
1527
1528         err = -EINVAL;
1529         if (tb[NDA_DST] == NULL)
1530                 goto out;
1531
1532         ndm = nlmsg_data(nlh);
1533         if (ndm->ndm_ifindex) {
1534                 dev = dev_get_by_index(net, ndm->ndm_ifindex);
1535                 if (dev == NULL) {
1536                         err = -ENODEV;
1537                         goto out;
1538                 }
1539
1540                 if (tb[NDA_LLADDR] && nla_len(tb[NDA_LLADDR]) < dev->addr_len)
1541                         goto out_dev_put;
1542         }
1543
1544         read_lock(&neigh_tbl_lock);
1545         for (tbl = neigh_tables; tbl; tbl = tbl->next) {
1546                 int flags = NEIGH_UPDATE_F_ADMIN | NEIGH_UPDATE_F_OVERRIDE;
1547                 struct neighbour *neigh;
1548                 void *dst, *lladdr;
1549
1550                 if (tbl->family != ndm->ndm_family)
1551                         continue;
1552                 read_unlock(&neigh_tbl_lock);
1553
1554                 if (nla_len(tb[NDA_DST]) < tbl->key_len)
1555                         goto out_dev_put;
1556                 dst = nla_data(tb[NDA_DST]);
1557                 lladdr = tb[NDA_LLADDR] ? nla_data(tb[NDA_LLADDR]) : NULL;
1558
1559                 if (ndm->ndm_flags & NTF_PROXY) {
1560                         struct pneigh_entry *pn;
1561
1562                         err = -ENOBUFS;
1563                         pn = pneigh_lookup(tbl, dst, dev, 1);
1564                         if (pn) {
1565                                 pn->flags = ndm->ndm_flags;
1566                                 err = 0;
1567                         }
1568                         goto out_dev_put;
1569                 }
1570
1571                 if (dev == NULL)
1572                         goto out_dev_put;
1573
1574                 neigh = neigh_lookup(tbl, dst, dev);
1575                 if (neigh == NULL) {
1576                         if (!(nlh->nlmsg_flags & NLM_F_CREATE)) {
1577                                 err = -ENOENT;
1578                                 goto out_dev_put;
1579                         }
1580
1581                         neigh = __neigh_lookup_errno(tbl, dst, dev);
1582                         if (IS_ERR(neigh)) {
1583                                 err = PTR_ERR(neigh);
1584                                 goto out_dev_put;
1585                         }
1586                 } else {
1587                         if (nlh->nlmsg_flags & NLM_F_EXCL) {
1588                                 err = -EEXIST;
1589                                 neigh_release(neigh);
1590                                 goto out_dev_put;
1591                         }
1592
1593                         if (!(nlh->nlmsg_flags & NLM_F_REPLACE))
1594                                 flags &= ~NEIGH_UPDATE_F_OVERRIDE;
1595                 }
1596
1597                 err = neigh_update(neigh, lladdr, ndm->ndm_state, flags);
1598                 neigh_release(neigh);
1599                 goto out_dev_put;
1600         }
1601
1602         read_unlock(&neigh_tbl_lock);
1603         err = -EAFNOSUPPORT;
1604
1605 out_dev_put:
1606         if (dev)
1607                 dev_put(dev);
1608 out:
1609         return err;
1610 }
1611
1612 static int neightbl_fill_parms(struct sk_buff *skb, struct neigh_parms *parms)
1613 {
1614         struct nlattr *nest;
1615
1616         nest = nla_nest_start(skb, NDTA_PARMS);
1617         if (nest == NULL)
1618                 return -ENOBUFS;
1619
1620         if (parms->dev)
1621                 NLA_PUT_U32(skb, NDTPA_IFINDEX, parms->dev->ifindex);
1622
1623         NLA_PUT_U32(skb, NDTPA_REFCNT, atomic_read(&parms->refcnt));
1624         NLA_PUT_U32(skb, NDTPA_QUEUE_LEN, parms->queue_len);
1625         NLA_PUT_U32(skb, NDTPA_PROXY_QLEN, parms->proxy_qlen);
1626         NLA_PUT_U32(skb, NDTPA_APP_PROBES, parms->app_probes);
1627         NLA_PUT_U32(skb, NDTPA_UCAST_PROBES, parms->ucast_probes);
1628         NLA_PUT_U32(skb, NDTPA_MCAST_PROBES, parms->mcast_probes);
1629         NLA_PUT_MSECS(skb, NDTPA_REACHABLE_TIME, parms->reachable_time);
1630         NLA_PUT_MSECS(skb, NDTPA_BASE_REACHABLE_TIME,
1631                       parms->base_reachable_time);
1632         NLA_PUT_MSECS(skb, NDTPA_GC_STALETIME, parms->gc_staletime);
1633         NLA_PUT_MSECS(skb, NDTPA_DELAY_PROBE_TIME, parms->delay_probe_time);
1634         NLA_PUT_MSECS(skb, NDTPA_RETRANS_TIME, parms->retrans_time);
1635         NLA_PUT_MSECS(skb, NDTPA_ANYCAST_DELAY, parms->anycast_delay);
1636         NLA_PUT_MSECS(skb, NDTPA_PROXY_DELAY, parms->proxy_delay);
1637         NLA_PUT_MSECS(skb, NDTPA_LOCKTIME, parms->locktime);
1638
1639         return nla_nest_end(skb, nest);
1640
1641 nla_put_failure:
1642         return nla_nest_cancel(skb, nest);
1643 }
1644
1645 static int neightbl_fill_info(struct sk_buff *skb, struct neigh_table *tbl,
1646                               u32 pid, u32 seq, int type, int flags)
1647 {
1648         struct nlmsghdr *nlh;
1649         struct ndtmsg *ndtmsg;
1650
1651         nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
1652         if (nlh == NULL)
1653                 return -EMSGSIZE;
1654
1655         ndtmsg = nlmsg_data(nlh);
1656
1657         read_lock_bh(&tbl->lock);
1658         ndtmsg->ndtm_family = tbl->family;
1659         ndtmsg->ndtm_pad1   = 0;
1660         ndtmsg->ndtm_pad2   = 0;
1661
1662         NLA_PUT_STRING(skb, NDTA_NAME, tbl->id);
1663         NLA_PUT_MSECS(skb, NDTA_GC_INTERVAL, tbl->gc_interval);
1664         NLA_PUT_U32(skb, NDTA_THRESH1, tbl->gc_thresh1);
1665         NLA_PUT_U32(skb, NDTA_THRESH2, tbl->gc_thresh2);
1666         NLA_PUT_U32(skb, NDTA_THRESH3, tbl->gc_thresh3);
1667
1668         {
1669                 unsigned long now = jiffies;
1670                 unsigned int flush_delta = now - tbl->last_flush;
1671                 unsigned int rand_delta = now - tbl->last_rand;
1672
1673                 struct ndt_config ndc = {
1674                         .ndtc_key_len           = tbl->key_len,
1675                         .ndtc_entry_size        = tbl->entry_size,
1676                         .ndtc_entries           = atomic_read(&tbl->entries),
1677                         .ndtc_last_flush        = jiffies_to_msecs(flush_delta),
1678                         .ndtc_last_rand         = jiffies_to_msecs(rand_delta),
1679                         .ndtc_hash_rnd          = tbl->hash_rnd,
1680                         .ndtc_hash_mask         = tbl->hash_mask,
1681                         .ndtc_hash_chain_gc     = tbl->hash_chain_gc,
1682                         .ndtc_proxy_qlen        = tbl->proxy_queue.qlen,
1683                 };
1684
1685                 NLA_PUT(skb, NDTA_CONFIG, sizeof(ndc), &ndc);
1686         }
1687
1688         {
1689                 int cpu;
1690                 struct ndt_stats ndst;
1691
1692                 memset(&ndst, 0, sizeof(ndst));
1693
1694                 for_each_possible_cpu(cpu) {
1695                         struct neigh_statistics *st;
1696
1697                         st = per_cpu_ptr(tbl->stats, cpu);
1698                         ndst.ndts_allocs                += st->allocs;
1699                         ndst.ndts_destroys              += st->destroys;
1700                         ndst.ndts_hash_grows            += st->hash_grows;
1701                         ndst.ndts_res_failed            += st->res_failed;
1702                         ndst.ndts_lookups               += st->lookups;
1703                         ndst.ndts_hits                  += st->hits;
1704                         ndst.ndts_rcv_probes_mcast      += st->rcv_probes_mcast;
1705                         ndst.ndts_rcv_probes_ucast      += st->rcv_probes_ucast;
1706                         ndst.ndts_periodic_gc_runs      += st->periodic_gc_runs;
1707                         ndst.ndts_forced_gc_runs        += st->forced_gc_runs;
1708                 }
1709
1710                 NLA_PUT(skb, NDTA_STATS, sizeof(ndst), &ndst);
1711         }
1712
1713         BUG_ON(tbl->parms.dev);
1714         if (neightbl_fill_parms(skb, &tbl->parms) < 0)
1715                 goto nla_put_failure;
1716
1717         read_unlock_bh(&tbl->lock);
1718         return nlmsg_end(skb, nlh);
1719
1720 nla_put_failure:
1721         read_unlock_bh(&tbl->lock);
1722         nlmsg_cancel(skb, nlh);
1723         return -EMSGSIZE;
1724 }
1725
1726 static int neightbl_fill_param_info(struct sk_buff *skb,
1727                                     struct neigh_table *tbl,
1728                                     struct neigh_parms *parms,
1729                                     u32 pid, u32 seq, int type,
1730                                     unsigned int flags)
1731 {
1732         struct ndtmsg *ndtmsg;
1733         struct nlmsghdr *nlh;
1734
1735         nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
1736         if (nlh == NULL)
1737                 return -EMSGSIZE;
1738
1739         ndtmsg = nlmsg_data(nlh);
1740
1741         read_lock_bh(&tbl->lock);
1742         ndtmsg->ndtm_family = tbl->family;
1743         ndtmsg->ndtm_pad1   = 0;
1744         ndtmsg->ndtm_pad2   = 0;
1745
1746         if (nla_put_string(skb, NDTA_NAME, tbl->id) < 0 ||
1747             neightbl_fill_parms(skb, parms) < 0)
1748                 goto errout;
1749
1750         read_unlock_bh(&tbl->lock);
1751         return nlmsg_end(skb, nlh);
1752 errout:
1753         read_unlock_bh(&tbl->lock);
1754         nlmsg_cancel(skb, nlh);
1755         return -EMSGSIZE;
1756 }
1757
1758 static inline struct neigh_parms *lookup_neigh_params(struct neigh_table *tbl,
1759                                                       int ifindex)
1760 {
1761         struct neigh_parms *p;
1762
1763         for (p = &tbl->parms; p; p = p->next)
1764                 if ((p->dev && p->dev->ifindex == ifindex) ||
1765                     (!p->dev && !ifindex))
1766                         return p;
1767
1768         return NULL;
1769 }
1770
1771 static const struct nla_policy nl_neightbl_policy[NDTA_MAX+1] = {
1772         [NDTA_NAME]             = { .type = NLA_STRING },
1773         [NDTA_THRESH1]          = { .type = NLA_U32 },
1774         [NDTA_THRESH2]          = { .type = NLA_U32 },
1775         [NDTA_THRESH3]          = { .type = NLA_U32 },
1776         [NDTA_GC_INTERVAL]      = { .type = NLA_U64 },
1777         [NDTA_PARMS]            = { .type = NLA_NESTED },
1778 };
1779
1780 static const struct nla_policy nl_ntbl_parm_policy[NDTPA_MAX+1] = {
1781         [NDTPA_IFINDEX]                 = { .type = NLA_U32 },
1782         [NDTPA_QUEUE_LEN]               = { .type = NLA_U32 },
1783         [NDTPA_PROXY_QLEN]              = { .type = NLA_U32 },
1784         [NDTPA_APP_PROBES]              = { .type = NLA_U32 },
1785         [NDTPA_UCAST_PROBES]            = { .type = NLA_U32 },
1786         [NDTPA_MCAST_PROBES]            = { .type = NLA_U32 },
1787         [NDTPA_BASE_REACHABLE_TIME]     = { .type = NLA_U64 },
1788         [NDTPA_GC_STALETIME]            = { .type = NLA_U64 },
1789         [NDTPA_DELAY_PROBE_TIME]        = { .type = NLA_U64 },
1790         [NDTPA_RETRANS_TIME]            = { .type = NLA_U64 },
1791         [NDTPA_ANYCAST_DELAY]           = { .type = NLA_U64 },
1792         [NDTPA_PROXY_DELAY]             = { .type = NLA_U64 },
1793         [NDTPA_LOCKTIME]                = { .type = NLA_U64 },
1794 };
1795
1796 static int neightbl_set(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
1797 {
1798         struct neigh_table *tbl;
1799         struct ndtmsg *ndtmsg;
1800         struct nlattr *tb[NDTA_MAX+1];
1801         int err;
1802
1803         err = nlmsg_parse(nlh, sizeof(*ndtmsg), tb, NDTA_MAX,
1804                           nl_neightbl_policy);
1805         if (err < 0)
1806                 goto errout;
1807
1808         if (tb[NDTA_NAME] == NULL) {
1809                 err = -EINVAL;
1810                 goto errout;
1811         }
1812
1813         ndtmsg = nlmsg_data(nlh);
1814         read_lock(&neigh_tbl_lock);
1815         for (tbl = neigh_tables; tbl; tbl = tbl->next) {
1816                 if (ndtmsg->ndtm_family && tbl->family != ndtmsg->ndtm_family)
1817                         continue;
1818
1819                 if (nla_strcmp(tb[NDTA_NAME], tbl->id) == 0)
1820                         break;
1821         }
1822
1823         if (tbl == NULL) {
1824                 err = -ENOENT;
1825                 goto errout_locked;
1826         }
1827
1828         /*
1829          * We acquire tbl->lock to be nice to the periodic timers and
1830          * make sure they always see a consistent set of values.
1831          */
1832         write_lock_bh(&tbl->lock);
1833
1834         if (tb[NDTA_PARMS]) {
1835                 struct nlattr *tbp[NDTPA_MAX+1];
1836                 struct neigh_parms *p;
1837                 int i, ifindex = 0;
1838
1839                 err = nla_parse_nested(tbp, NDTPA_MAX, tb[NDTA_PARMS],
1840                                        nl_ntbl_parm_policy);
1841                 if (err < 0)
1842                         goto errout_tbl_lock;
1843
1844                 if (tbp[NDTPA_IFINDEX])
1845                         ifindex = nla_get_u32(tbp[NDTPA_IFINDEX]);
1846
1847                 p = lookup_neigh_params(tbl, ifindex);
1848                 if (p == NULL) {
1849                         err = -ENOENT;
1850                         goto errout_tbl_lock;
1851                 }
1852
1853                 for (i = 1; i <= NDTPA_MAX; i++) {
1854                         if (tbp[i] == NULL)
1855                                 continue;
1856
1857                         switch (i) {
1858                         case NDTPA_QUEUE_LEN:
1859                                 p->queue_len = nla_get_u32(tbp[i]);
1860                                 break;
1861                         case NDTPA_PROXY_QLEN:
1862                                 p->proxy_qlen = nla_get_u32(tbp[i]);
1863                                 break;
1864                         case NDTPA_APP_PROBES:
1865                                 p->app_probes = nla_get_u32(tbp[i]);
1866                                 break;
1867                         case NDTPA_UCAST_PROBES:
1868                                 p->ucast_probes = nla_get_u32(tbp[i]);
1869                                 break;
1870                         case NDTPA_MCAST_PROBES:
1871                                 p->mcast_probes = nla_get_u32(tbp[i]);
1872                                 break;
1873                         case NDTPA_BASE_REACHABLE_TIME:
1874                                 p->base_reachable_time = nla_get_msecs(tbp[i]);
1875                                 break;
1876                         case NDTPA_GC_STALETIME:
1877                                 p->gc_staletime = nla_get_msecs(tbp[i]);
1878                                 break;
1879                         case NDTPA_DELAY_PROBE_TIME:
1880                                 p->delay_probe_time = nla_get_msecs(tbp[i]);
1881                                 break;
1882                         case NDTPA_RETRANS_TIME:
1883                                 p->retrans_time = nla_get_msecs(tbp[i]);
1884                                 break;
1885                         case NDTPA_ANYCAST_DELAY:
1886                                 p->anycast_delay = nla_get_msecs(tbp[i]);
1887                                 break;
1888                         case NDTPA_PROXY_DELAY:
1889                                 p->proxy_delay = nla_get_msecs(tbp[i]);
1890                                 break;
1891                         case NDTPA_LOCKTIME:
1892                                 p->locktime = nla_get_msecs(tbp[i]);
1893                                 break;
1894                         }
1895                 }
1896         }
1897
1898         if (tb[NDTA_THRESH1])
1899                 tbl->gc_thresh1 = nla_get_u32(tb[NDTA_THRESH1]);
1900
1901         if (tb[NDTA_THRESH2])
1902                 tbl->gc_thresh2 = nla_get_u32(tb[NDTA_THRESH2]);
1903
1904         if (tb[NDTA_THRESH3])
1905                 tbl->gc_thresh3 = nla_get_u32(tb[NDTA_THRESH3]);
1906
1907         if (tb[NDTA_GC_INTERVAL])
1908                 tbl->gc_interval = nla_get_msecs(tb[NDTA_GC_INTERVAL]);
1909
1910         err = 0;
1911
1912 errout_tbl_lock:
1913         write_unlock_bh(&tbl->lock);
1914 errout_locked:
1915         read_unlock(&neigh_tbl_lock);
1916 errout:
1917         return err;
1918 }
1919
1920 static int neightbl_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
1921 {
1922         int family, tidx, nidx = 0;
1923         int tbl_skip = cb->args[0];
1924         int neigh_skip = cb->args[1];
1925         struct neigh_table *tbl;
1926
1927         family = ((struct rtgenmsg *) nlmsg_data(cb->nlh))->rtgen_family;
1928
1929         read_lock(&neigh_tbl_lock);
1930         for (tbl = neigh_tables, tidx = 0; tbl; tbl = tbl->next, tidx++) {
1931                 struct neigh_parms *p;
1932
1933                 if (tidx < tbl_skip || (family && tbl->family != family))
1934                         continue;
1935
1936                 if (neightbl_fill_info(skb, tbl, NETLINK_CB(cb->skb).pid,
1937                                        cb->nlh->nlmsg_seq, RTM_NEWNEIGHTBL,
1938                                        NLM_F_MULTI) <= 0)
1939                         break;
1940
1941                 for (nidx = 0, p = tbl->parms.next; p; p = p->next, nidx++) {
1942                         if (nidx < neigh_skip)
1943                                 continue;
1944
1945                         if (neightbl_fill_param_info(skb, tbl, p,
1946                                                      NETLINK_CB(cb->skb).pid,
1947                                                      cb->nlh->nlmsg_seq,
1948                                                      RTM_NEWNEIGHTBL,
1949                                                      NLM_F_MULTI) <= 0)
1950                                 goto out;
1951                 }
1952
1953                 neigh_skip = 0;
1954         }
1955 out:
1956         read_unlock(&neigh_tbl_lock);
1957         cb->args[0] = tidx;
1958         cb->args[1] = nidx;
1959
1960         return skb->len;
1961 }
1962
1963 static int neigh_fill_info(struct sk_buff *skb, struct neighbour *neigh,
1964                            u32 pid, u32 seq, int type, unsigned int flags)
1965 {
1966         unsigned long now = jiffies;
1967         struct nda_cacheinfo ci;
1968         struct nlmsghdr *nlh;
1969         struct ndmsg *ndm;
1970
1971         nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
1972         if (nlh == NULL)
1973                 return -EMSGSIZE;
1974
1975         ndm = nlmsg_data(nlh);
1976         ndm->ndm_family  = neigh->ops->family;
1977         ndm->ndm_pad1    = 0;
1978         ndm->ndm_pad2    = 0;
1979         ndm->ndm_flags   = neigh->flags;
1980         ndm->ndm_type    = neigh->type;
1981         ndm->ndm_ifindex = neigh->dev->ifindex;
1982
1983         NLA_PUT(skb, NDA_DST, neigh->tbl->key_len, neigh->primary_key);
1984
1985         read_lock_bh(&neigh->lock);
1986         ndm->ndm_state   = neigh->nud_state;
1987         if ((neigh->nud_state & NUD_VALID) &&
1988             nla_put(skb, NDA_LLADDR, neigh->dev->addr_len, neigh->ha) < 0) {
1989                 read_unlock_bh(&neigh->lock);
1990                 goto nla_put_failure;
1991         }
1992
1993         ci.ndm_used      = now - neigh->used;
1994         ci.ndm_confirmed = now - neigh->confirmed;
1995         ci.ndm_updated   = now - neigh->updated;
1996         ci.ndm_refcnt    = atomic_read(&neigh->refcnt) - 1;
1997         read_unlock_bh(&neigh->lock);
1998
1999         NLA_PUT_U32(skb, NDA_PROBES, atomic_read(&neigh->probes));
2000         NLA_PUT(skb, NDA_CACHEINFO, sizeof(ci), &ci);
2001
2002         return nlmsg_end(skb, nlh);
2003
2004 nla_put_failure:
2005         nlmsg_cancel(skb, nlh);
2006         return -EMSGSIZE;
2007 }
2008
2009 static void neigh_update_notify(struct neighbour *neigh)
2010 {
2011         call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, neigh);
2012         __neigh_notify(neigh, RTM_NEWNEIGH, 0);
2013 }
2014
2015 static int neigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2016                             struct netlink_callback *cb)
2017 {
2018         struct neighbour *n;
2019         int rc, h, s_h = cb->args[1];
2020         int idx, s_idx = idx = cb->args[2];
2021
2022         read_lock_bh(&tbl->lock);
2023         for (h = 0; h <= tbl->hash_mask; h++) {
2024                 if (h < s_h)
2025                         continue;
2026                 if (h > s_h)
2027                         s_idx = 0;
2028                 for (n = tbl->hash_buckets[h], idx = 0; n; n = n->next, idx++) {
2029                         if (idx < s_idx)
2030                                 continue;
2031                         if (neigh_fill_info(skb, n, NETLINK_CB(cb->skb).pid,
2032                                             cb->nlh->nlmsg_seq,
2033                                             RTM_NEWNEIGH,
2034                                             NLM_F_MULTI) <= 0) {
2035                                 read_unlock_bh(&tbl->lock);
2036                                 rc = -1;
2037                                 goto out;
2038                         }
2039                 }
2040         }
2041         read_unlock_bh(&tbl->lock);
2042         rc = skb->len;
2043 out:
2044         cb->args[1] = h;
2045         cb->args[2] = idx;
2046         return rc;
2047 }
2048
2049 static int neigh_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2050 {
2051         struct neigh_table *tbl;
2052         int t, family, s_t;
2053
2054         read_lock(&neigh_tbl_lock);
2055         family = ((struct rtgenmsg *) nlmsg_data(cb->nlh))->rtgen_family;
2056         s_t = cb->args[0];
2057
2058         for (tbl = neigh_tables, t = 0; tbl; tbl = tbl->next, t++) {
2059                 if (t < s_t || (family && tbl->family != family))
2060                         continue;
2061                 if (t > s_t)
2062                         memset(&cb->args[1], 0, sizeof(cb->args) -
2063                                                 sizeof(cb->args[0]));
2064                 if (neigh_dump_table(tbl, skb, cb) < 0)
2065                         break;
2066         }
2067         read_unlock(&neigh_tbl_lock);
2068
2069         cb->args[0] = t;
2070         return skb->len;
2071 }
2072
2073 void neigh_for_each(struct neigh_table *tbl, void (*cb)(struct neighbour *, void *), void *cookie)
2074 {
2075         int chain;
2076
2077         read_lock_bh(&tbl->lock);
2078         for (chain = 0; chain <= tbl->hash_mask; chain++) {
2079                 struct neighbour *n;
2080
2081                 for (n = tbl->hash_buckets[chain]; n; n = n->next)
2082                         cb(n, cookie);
2083         }
2084         read_unlock_bh(&tbl->lock);
2085 }
2086 EXPORT_SYMBOL(neigh_for_each);
2087
2088 /* The tbl->lock must be held as a writer and BH disabled. */
2089 void __neigh_for_each_release(struct neigh_table *tbl,
2090                               int (*cb)(struct neighbour *))
2091 {
2092         int chain;
2093
2094         for (chain = 0; chain <= tbl->hash_mask; chain++) {
2095                 struct neighbour *n, **np;
2096
2097                 np = &tbl->hash_buckets[chain];
2098                 while ((n = *np) != NULL) {
2099                         int release;
2100
2101                         write_lock(&n->lock);
2102                         release = cb(n);
2103                         if (release) {
2104                                 *np = n->next;
2105                                 n->dead = 1;
2106                         } else
2107                                 np = &n->next;
2108                         write_unlock(&n->lock);
2109                         if (release)
2110                                 neigh_cleanup_and_release(n);
2111                 }
2112         }
2113 }
2114 EXPORT_SYMBOL(__neigh_for_each_release);
2115
2116 #ifdef CONFIG_PROC_FS
2117
2118 static struct neighbour *neigh_get_first(struct seq_file *seq)
2119 {
2120         struct neigh_seq_state *state = seq->private;
2121         struct neigh_table *tbl = state->tbl;
2122         struct neighbour *n = NULL;
2123         int bucket = state->bucket;
2124
2125         state->flags &= ~NEIGH_SEQ_IS_PNEIGH;
2126         for (bucket = 0; bucket <= tbl->hash_mask; bucket++) {
2127                 n = tbl->hash_buckets[bucket];
2128
2129                 while (n) {
2130                         if (state->neigh_sub_iter) {
2131                                 loff_t fakep = 0;
2132                                 void *v;
2133
2134                                 v = state->neigh_sub_iter(state, n, &fakep);
2135                                 if (!v)
2136                                         goto next;
2137                         }
2138                         if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
2139                                 break;
2140                         if (n->nud_state & ~NUD_NOARP)
2141                                 break;
2142                 next:
2143                         n = n->next;
2144                 }
2145
2146                 if (n)
2147                         break;
2148         }
2149         state->bucket = bucket;
2150
2151         return n;
2152 }
2153
2154 static struct neighbour *neigh_get_next(struct seq_file *seq,
2155                                         struct neighbour *n,
2156                                         loff_t *pos)
2157 {
2158         struct neigh_seq_state *state = seq->private;
2159         struct neigh_table *tbl = state->tbl;
2160
2161         if (state->neigh_sub_iter) {
2162                 void *v = state->neigh_sub_iter(state, n, pos);
2163                 if (v)
2164                         return n;
2165         }
2166         n = n->next;
2167
2168         while (1) {
2169                 while (n) {
2170                         if (state->neigh_sub_iter) {
2171                                 void *v = state->neigh_sub_iter(state, n, pos);
2172                                 if (v)
2173                                         return n;
2174                                 goto next;
2175                         }
2176                         if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
2177                                 break;
2178
2179                         if (n->nud_state & ~NUD_NOARP)
2180                                 break;
2181                 next:
2182                         n = n->next;
2183                 }
2184
2185                 if (n)
2186                         break;
2187
2188                 if (++state->bucket > tbl->hash_mask)
2189                         break;
2190
2191                 n = tbl->hash_buckets[state->bucket];
2192         }
2193
2194         if (n && pos)
2195                 --(*pos);
2196         return n;
2197 }
2198
2199 static struct neighbour *neigh_get_idx(struct seq_file *seq, loff_t *pos)
2200 {
2201         struct neighbour *n = neigh_get_first(seq);
2202
2203         if (n) {
2204                 while (*pos) {
2205                         n = neigh_get_next(seq, n, pos);
2206                         if (!n)
2207                                 break;
2208                 }
2209         }
2210         return *pos ? NULL : n;
2211 }
2212
2213 static struct pneigh_entry *pneigh_get_first(struct seq_file *seq)
2214 {
2215         struct neigh_seq_state *state = seq->private;
2216         struct neigh_table *tbl = state->tbl;
2217         struct pneigh_entry *pn = NULL;
2218         int bucket = state->bucket;
2219
2220         state->flags |= NEIGH_SEQ_IS_PNEIGH;
2221         for (bucket = 0; bucket <= PNEIGH_HASHMASK; bucket++) {
2222                 pn = tbl->phash_buckets[bucket];
2223                 if (pn)
2224                         break;
2225         }
2226         state->bucket = bucket;
2227
2228         return pn;
2229 }
2230
2231 static struct pneigh_entry *pneigh_get_next(struct seq_file *seq,
2232                                             struct pneigh_entry *pn,
2233                                             loff_t *pos)
2234 {
2235         struct neigh_seq_state *state = seq->private;
2236         struct neigh_table *tbl = state->tbl;
2237
2238         pn = pn->next;
2239         while (!pn) {
2240                 if (++state->bucket > PNEIGH_HASHMASK)
2241                         break;
2242                 pn = tbl->phash_buckets[state->bucket];
2243                 if (pn)
2244                         break;
2245         }
2246
2247         if (pn && pos)
2248                 --(*pos);
2249
2250         return pn;
2251 }
2252
2253 static struct pneigh_entry *pneigh_get_idx(struct seq_file *seq, loff_t *pos)
2254 {
2255         struct pneigh_entry *pn = pneigh_get_first(seq);
2256
2257         if (pn) {
2258                 while (*pos) {
2259                         pn = pneigh_get_next(seq, pn, pos);
2260                         if (!pn)
2261                                 break;
2262                 }
2263         }
2264         return *pos ? NULL : pn;
2265 }
2266
2267 static void *neigh_get_idx_any(struct seq_file *seq, loff_t *pos)
2268 {
2269         struct neigh_seq_state *state = seq->private;
2270         void *rc;
2271
2272         rc = neigh_get_idx(seq, pos);
2273         if (!rc && !(state->flags & NEIGH_SEQ_NEIGH_ONLY))
2274                 rc = pneigh_get_idx(seq, pos);
2275
2276         return rc;
2277 }
2278
2279 void *neigh_seq_start(struct seq_file *seq, loff_t *pos, struct neigh_table *tbl, unsigned int neigh_seq_flags)
2280 {
2281         struct neigh_seq_state *state = seq->private;
2282         loff_t pos_minus_one;
2283
2284         state->tbl = tbl;
2285         state->bucket = 0;
2286         state->flags = (neigh_seq_flags & ~NEIGH_SEQ_IS_PNEIGH);
2287
2288         read_lock_bh(&tbl->lock);
2289
2290         pos_minus_one = *pos - 1;
2291         return *pos ? neigh_get_idx_any(seq, &pos_minus_one) : SEQ_START_TOKEN;
2292 }
2293 EXPORT_SYMBOL(neigh_seq_start);
2294
2295 void *neigh_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2296 {
2297         struct neigh_seq_state *state;
2298         void *rc;
2299
2300         if (v == SEQ_START_TOKEN) {
2301                 rc = neigh_get_idx(seq, pos);
2302                 goto out;
2303         }
2304
2305         state = seq->private;
2306         if (!(state->flags & NEIGH_SEQ_IS_PNEIGH)) {
2307                 rc = neigh_get_next(seq, v, NULL);
2308                 if (rc)
2309                         goto out;
2310                 if (!(state->flags & NEIGH_SEQ_NEIGH_ONLY))
2311                         rc = pneigh_get_first(seq);
2312         } else {
2313                 BUG_ON(state->flags & NEIGH_SEQ_NEIGH_ONLY);
2314                 rc = pneigh_get_next(seq, v, NULL);
2315         }
2316 out:
2317         ++(*pos);
2318         return rc;
2319 }
2320 EXPORT_SYMBOL(neigh_seq_next);
2321
2322 void neigh_seq_stop(struct seq_file *seq, void *v)
2323 {
2324         struct neigh_seq_state *state = seq->private;
2325         struct neigh_table *tbl = state->tbl;
2326
2327         read_unlock_bh(&tbl->lock);
2328 }
2329 EXPORT_SYMBOL(neigh_seq_stop);
2330
2331 /* statistics via seq_file */
2332
2333 static void *neigh_stat_seq_start(struct seq_file *seq, loff_t *pos)
2334 {
2335         struct proc_dir_entry *pde = seq->private;
2336         struct neigh_table *tbl = pde->data;
2337         int cpu;
2338
2339         if (*pos == 0)
2340                 return SEQ_START_TOKEN;
2341
2342         for (cpu = *pos-1; cpu < NR_CPUS; ++cpu) {
2343                 if (!cpu_possible(cpu))
2344                         continue;
2345                 *pos = cpu+1;
2346                 return per_cpu_ptr(tbl->stats, cpu);
2347         }
2348         return NULL;
2349 }
2350
2351 static void *neigh_stat_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2352 {
2353         struct proc_dir_entry *pde = seq->private;
2354         struct neigh_table *tbl = pde->data;
2355         int cpu;
2356
2357         for (cpu = *pos; cpu < NR_CPUS; ++cpu) {
2358                 if (!cpu_possible(cpu))
2359                         continue;
2360                 *pos = cpu+1;
2361                 return per_cpu_ptr(tbl->stats, cpu);
2362         }
2363         return NULL;
2364 }
2365
2366 static void neigh_stat_seq_stop(struct seq_file *seq, void *v)
2367 {
2368
2369 }
2370
2371 static int neigh_stat_seq_show(struct seq_file *seq, void *v)
2372 {
2373         struct proc_dir_entry *pde = seq->private;
2374         struct neigh_table *tbl = pde->data;
2375         struct neigh_statistics *st = v;
2376
2377         if (v == SEQ_START_TOKEN) {
2378                 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");
2379                 return 0;
2380         }
2381
2382         seq_printf(seq, "%08x  %08lx %08lx %08lx  %08lx %08lx  %08lx  "
2383                         "%08lx %08lx  %08lx %08lx\n",
2384                    atomic_read(&tbl->entries),
2385
2386                    st->allocs,
2387                    st->destroys,
2388                    st->hash_grows,
2389
2390                    st->lookups,
2391                    st->hits,
2392
2393                    st->res_failed,
2394
2395                    st->rcv_probes_mcast,
2396                    st->rcv_probes_ucast,
2397
2398                    st->periodic_gc_runs,
2399                    st->forced_gc_runs
2400                    );
2401
2402         return 0;
2403 }
2404
2405 static const struct seq_operations neigh_stat_seq_ops = {
2406         .start  = neigh_stat_seq_start,
2407         .next   = neigh_stat_seq_next,
2408         .stop   = neigh_stat_seq_stop,
2409         .show   = neigh_stat_seq_show,
2410 };
2411
2412 static int neigh_stat_seq_open(struct inode *inode, struct file *file)
2413 {
2414         int ret = seq_open(file, &neigh_stat_seq_ops);
2415
2416         if (!ret) {
2417                 struct seq_file *sf = file->private_data;
2418                 sf->private = PDE(inode);
2419         }
2420         return ret;
2421 };
2422
2423 static const struct file_operations neigh_stat_seq_fops = {
2424         .owner   = THIS_MODULE,
2425         .open    = neigh_stat_seq_open,
2426         .read    = seq_read,
2427         .llseek  = seq_lseek,
2428         .release = seq_release,
2429 };
2430
2431 #endif /* CONFIG_PROC_FS */
2432
2433 static inline size_t neigh_nlmsg_size(void)
2434 {
2435         return NLMSG_ALIGN(sizeof(struct ndmsg))
2436                + nla_total_size(MAX_ADDR_LEN) /* NDA_DST */
2437                + nla_total_size(MAX_ADDR_LEN) /* NDA_LLADDR */
2438                + nla_total_size(sizeof(struct nda_cacheinfo))
2439                + nla_total_size(4); /* NDA_PROBES */
2440 }
2441
2442 static void __neigh_notify(struct neighbour *n, int type, int flags)
2443 {
2444         struct sk_buff *skb;
2445         int err = -ENOBUFS;
2446
2447         skb = nlmsg_new(neigh_nlmsg_size(), GFP_ATOMIC);
2448         if (skb == NULL)
2449                 goto errout;
2450
2451         err = neigh_fill_info(skb, n, 0, 0, type, flags);
2452         if (err < 0) {
2453                 /* -EMSGSIZE implies BUG in neigh_nlmsg_size() */
2454                 WARN_ON(err == -EMSGSIZE);
2455                 kfree_skb(skb);
2456                 goto errout;
2457         }
2458         err = rtnl_notify(skb, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
2459 errout:
2460         if (err < 0)
2461                 rtnl_set_sk_err(RTNLGRP_NEIGH, err);
2462 }
2463
2464 #ifdef CONFIG_ARPD
2465 void neigh_app_ns(struct neighbour *n)
2466 {
2467         __neigh_notify(n, RTM_GETNEIGH, NLM_F_REQUEST);
2468 }
2469 #endif /* CONFIG_ARPD */
2470
2471 #ifdef CONFIG_SYSCTL
2472
2473 static struct neigh_sysctl_table {
2474         struct ctl_table_header *sysctl_header;
2475         ctl_table               neigh_vars[__NET_NEIGH_MAX];
2476         ctl_table               neigh_dev[2];
2477         ctl_table               neigh_neigh_dir[2];
2478         ctl_table               neigh_proto_dir[2];
2479         ctl_table               neigh_root_dir[2];
2480 } neigh_sysctl_template __read_mostly = {
2481         .neigh_vars = {
2482                 {
2483                         .ctl_name       = NET_NEIGH_MCAST_SOLICIT,
2484                         .procname       = "mcast_solicit",
2485                         .maxlen         = sizeof(int),
2486                         .mode           = 0644,
2487                         .proc_handler   = &proc_dointvec,
2488                 },
2489                 {
2490                         .ctl_name       = NET_NEIGH_UCAST_SOLICIT,
2491                         .procname       = "ucast_solicit",
2492                         .maxlen         = sizeof(int),
2493                         .mode           = 0644,
2494                         .proc_handler   = &proc_dointvec,
2495                 },
2496                 {
2497                         .ctl_name       = NET_NEIGH_APP_SOLICIT,
2498                         .procname       = "app_solicit",
2499                         .maxlen         = sizeof(int),
2500                         .mode           = 0644,
2501                         .proc_handler   = &proc_dointvec,
2502                 },
2503                 {
2504                         .procname       = "retrans_time",
2505                         .maxlen         = sizeof(int),
2506                         .mode           = 0644,
2507                         .proc_handler   = &proc_dointvec_userhz_jiffies,
2508                 },
2509                 {
2510                         .ctl_name       = NET_NEIGH_REACHABLE_TIME,
2511                         .procname       = "base_reachable_time",
2512                         .maxlen         = sizeof(int),
2513                         .mode           = 0644,
2514                         .proc_handler   = &proc_dointvec_jiffies,
2515                         .strategy       = &sysctl_jiffies,
2516                 },
2517                 {
2518                         .ctl_name       = NET_NEIGH_DELAY_PROBE_TIME,
2519                         .procname       = "delay_first_probe_time",
2520                         .maxlen         = sizeof(int),
2521                         .mode           = 0644,
2522                         .proc_handler   = &proc_dointvec_jiffies,
2523                         .strategy       = &sysctl_jiffies,
2524                 },
2525                 {
2526                         .ctl_name       = NET_NEIGH_GC_STALE_TIME,
2527                         .procname       = "gc_stale_time",
2528                         .maxlen         = sizeof(int),
2529                         .mode           = 0644,
2530                         .proc_handler   = &proc_dointvec_jiffies,
2531                         .strategy       = &sysctl_jiffies,
2532                 },
2533                 {
2534                         .ctl_name       = NET_NEIGH_UNRES_QLEN,
2535                         .procname       = "unres_qlen",
2536                         .maxlen         = sizeof(int),
2537                         .mode           = 0644,
2538                         .proc_handler   = &proc_dointvec,
2539                 },
2540                 {
2541                         .ctl_name       = NET_NEIGH_PROXY_QLEN,
2542                         .procname       = "proxy_qlen",
2543                         .maxlen         = sizeof(int),
2544                         .mode           = 0644,
2545                         .proc_handler   = &proc_dointvec,
2546                 },
2547                 {
2548                         .procname       = "anycast_delay",
2549                         .maxlen         = sizeof(int),
2550                         .mode           = 0644,
2551                         .proc_handler   = &proc_dointvec_userhz_jiffies,
2552                 },
2553                 {
2554                         .procname       = "proxy_delay",
2555                         .maxlen         = sizeof(int),
2556                         .mode           = 0644,
2557                         .proc_handler   = &proc_dointvec_userhz_jiffies,
2558                 },
2559                 {
2560                         .procname       = "locktime",
2561                         .maxlen         = sizeof(int),
2562                         .mode           = 0644,
2563                         .proc_handler   = &proc_dointvec_userhz_jiffies,
2564                 },
2565                 {
2566                         .ctl_name       = NET_NEIGH_RETRANS_TIME_MS,
2567                         .procname       = "retrans_time_ms",
2568                         .maxlen         = sizeof(int),
2569                         .mode           = 0644,
2570                         .proc_handler   = &proc_dointvec_ms_jiffies,
2571                         .strategy       = &sysctl_ms_jiffies,
2572                 },
2573                 {
2574                         .ctl_name       = NET_NEIGH_REACHABLE_TIME_MS,
2575                         .procname       = "base_reachable_time_ms",
2576                         .maxlen         = sizeof(int),
2577                         .mode           = 0644,
2578                         .proc_handler   = &proc_dointvec_ms_jiffies,
2579                         .strategy       = &sysctl_ms_jiffies,
2580                 },
2581                 {
2582                         .ctl_name       = NET_NEIGH_GC_INTERVAL,
2583                         .procname       = "gc_interval",
2584                         .maxlen         = sizeof(int),
2585                         .mode           = 0644,
2586                         .proc_handler   = &proc_dointvec_jiffies,
2587                         .strategy       = &sysctl_jiffies,
2588                 },
2589                 {
2590                         .ctl_name       = NET_NEIGH_GC_THRESH1,
2591                         .procname       = "gc_thresh1",
2592                         .maxlen         = sizeof(int),
2593                         .mode           = 0644,
2594                         .proc_handler   = &proc_dointvec,
2595                 },
2596                 {
2597                         .ctl_name       = NET_NEIGH_GC_THRESH2,
2598                         .procname       = "gc_thresh2",
2599                         .maxlen         = sizeof(int),
2600                         .mode           = 0644,
2601                         .proc_handler   = &proc_dointvec,
2602                 },
2603                 {
2604                         .ctl_name       = NET_NEIGH_GC_THRESH3,
2605                         .procname       = "gc_thresh3",
2606                         .maxlen         = sizeof(int),
2607                         .mode           = 0644,
2608                         .proc_handler   = &proc_dointvec,
2609                 },
2610                 {}
2611         },
2612         .neigh_dev = {
2613                 {
2614                         .ctl_name       = NET_PROTO_CONF_DEFAULT,
2615                         .procname       = "default",
2616                         .mode           = 0555,
2617                 },
2618         },
2619         .neigh_neigh_dir = {
2620                 {
2621                         .procname       = "neigh",
2622                         .mode           = 0555,
2623                 },
2624         },
2625         .neigh_proto_dir = {
2626                 {
2627                         .mode           = 0555,
2628                 },
2629         },
2630         .neigh_root_dir = {
2631                 {
2632                         .ctl_name       = CTL_NET,
2633                         .procname       = "net",
2634                         .mode           = 0555,
2635                 },
2636         },
2637 };
2638
2639 int neigh_sysctl_register(struct net_device *dev, struct neigh_parms *p,
2640                           int p_id, int pdev_id, char *p_name,
2641                           proc_handler *handler, ctl_handler *strategy)
2642 {
2643         struct neigh_sysctl_table *t = kmemdup(&neigh_sysctl_template,
2644                                                sizeof(*t), GFP_KERNEL);
2645         const char *dev_name_source = NULL;
2646         char *dev_name = NULL;
2647         int err = 0;
2648
2649         if (!t)
2650                 return -ENOBUFS;
2651         t->neigh_vars[0].data  = &p->mcast_probes;
2652         t->neigh_vars[1].data  = &p->ucast_probes;
2653         t->neigh_vars[2].data  = &p->app_probes;
2654         t->neigh_vars[3].data  = &p->retrans_time;
2655         t->neigh_vars[4].data  = &p->base_reachable_time;
2656         t->neigh_vars[5].data  = &p->delay_probe_time;
2657         t->neigh_vars[6].data  = &p->gc_staletime;
2658         t->neigh_vars[7].data  = &p->queue_len;
2659         t->neigh_vars[8].data  = &p->proxy_qlen;
2660         t->neigh_vars[9].data  = &p->anycast_delay;
2661         t->neigh_vars[10].data = &p->proxy_delay;
2662         t->neigh_vars[11].data = &p->locktime;
2663         t->neigh_vars[12].data  = &p->retrans_time;
2664         t->neigh_vars[13].data  = &p->base_reachable_time;
2665
2666         if (dev) {
2667                 dev_name_source = dev->name;
2668                 t->neigh_dev[0].ctl_name = dev->ifindex;
2669                 /* Terminate the table early */
2670                 memset(&t->neigh_vars[14], 0, sizeof(t->neigh_vars[14]));
2671         } else {
2672                 dev_name_source = t->neigh_dev[0].procname;
2673                 t->neigh_vars[14].data = (int *)(p + 1);
2674                 t->neigh_vars[15].data = (int *)(p + 1) + 1;
2675                 t->neigh_vars[16].data = (int *)(p + 1) + 2;
2676                 t->neigh_vars[17].data = (int *)(p + 1) + 3;
2677         }
2678
2679
2680         if (handler || strategy) {
2681                 /* RetransTime */
2682                 t->neigh_vars[3].proc_handler = handler;
2683                 t->neigh_vars[3].strategy = strategy;
2684                 t->neigh_vars[3].extra1 = dev;
2685                 if (!strategy)
2686                         t->neigh_vars[3].ctl_name = CTL_UNNUMBERED;
2687                 /* ReachableTime */
2688                 t->neigh_vars[4].proc_handler = handler;
2689                 t->neigh_vars[4].strategy = strategy;
2690                 t->neigh_vars[4].extra1 = dev;
2691                 if (!strategy)
2692                         t->neigh_vars[4].ctl_name = CTL_UNNUMBERED;
2693                 /* RetransTime (in milliseconds)*/
2694                 t->neigh_vars[12].proc_handler = handler;
2695                 t->neigh_vars[12].strategy = strategy;
2696                 t->neigh_vars[12].extra1 = dev;
2697                 if (!strategy)
2698                         t->neigh_vars[12].ctl_name = CTL_UNNUMBERED;
2699                 /* ReachableTime (in milliseconds) */
2700                 t->neigh_vars[13].proc_handler = handler;
2701                 t->neigh_vars[13].strategy = strategy;
2702                 t->neigh_vars[13].extra1 = dev;
2703                 if (!strategy)
2704                         t->neigh_vars[13].ctl_name = CTL_UNNUMBERED;
2705         }
2706
2707         dev_name = kstrdup(dev_name_source, GFP_KERNEL);
2708         if (!dev_name) {
2709                 err = -ENOBUFS;
2710                 goto free;
2711         }
2712
2713         t->neigh_dev[0].procname = dev_name;
2714
2715         t->neigh_neigh_dir[0].ctl_name = pdev_id;
2716
2717         t->neigh_proto_dir[0].procname = p_name;
2718         t->neigh_proto_dir[0].ctl_name = p_id;
2719
2720         t->neigh_dev[0].child          = t->neigh_vars;
2721         t->neigh_neigh_dir[0].child    = t->neigh_dev;
2722         t->neigh_proto_dir[0].child    = t->neigh_neigh_dir;
2723         t->neigh_root_dir[0].child     = t->neigh_proto_dir;
2724
2725         t->sysctl_header = register_sysctl_table(t->neigh_root_dir);
2726         if (!t->sysctl_header) {
2727                 err = -ENOBUFS;
2728                 goto free_procname;
2729         }
2730         p->sysctl_table = t;
2731         return 0;
2732
2733         /* error path */
2734  free_procname:
2735         kfree(dev_name);
2736  free:
2737         kfree(t);
2738
2739         return err;
2740 }
2741
2742 void neigh_sysctl_unregister(struct neigh_parms *p)
2743 {
2744         if (p->sysctl_table) {
2745                 struct neigh_sysctl_table *t = p->sysctl_table;
2746                 p->sysctl_table = NULL;
2747                 unregister_sysctl_table(t->sysctl_header);
2748                 kfree(t->neigh_dev[0].procname);
2749                 kfree(t);
2750         }
2751 }
2752
2753 #endif  /* CONFIG_SYSCTL */
2754
2755 static int __init neigh_init(void)
2756 {
2757         rtnl_register(PF_UNSPEC, RTM_NEWNEIGH, neigh_add, NULL);
2758         rtnl_register(PF_UNSPEC, RTM_DELNEIGH, neigh_delete, NULL);
2759         rtnl_register(PF_UNSPEC, RTM_GETNEIGH, NULL, neigh_dump_info);
2760
2761         rtnl_register(PF_UNSPEC, RTM_GETNEIGHTBL, NULL, neightbl_dump_info);
2762         rtnl_register(PF_UNSPEC, RTM_SETNEIGHTBL, neightbl_set, NULL);
2763
2764         return 0;
2765 }
2766
2767 subsys_initcall(neigh_init);
2768
2769 EXPORT_SYMBOL(__neigh_event_send);
2770 EXPORT_SYMBOL(neigh_changeaddr);
2771 EXPORT_SYMBOL(neigh_compat_output);
2772 EXPORT_SYMBOL(neigh_connected_output);
2773 EXPORT_SYMBOL(neigh_create);
2774 EXPORT_SYMBOL(neigh_destroy);
2775 EXPORT_SYMBOL(neigh_event_ns);
2776 EXPORT_SYMBOL(neigh_ifdown);
2777 EXPORT_SYMBOL(neigh_lookup);
2778 EXPORT_SYMBOL(neigh_lookup_nodev);
2779 EXPORT_SYMBOL(neigh_parms_alloc);
2780 EXPORT_SYMBOL(neigh_parms_release);
2781 EXPORT_SYMBOL(neigh_rand_reach_time);
2782 EXPORT_SYMBOL(neigh_resolve_output);
2783 EXPORT_SYMBOL(neigh_table_clear);
2784 EXPORT_SYMBOL(neigh_table_init);
2785 EXPORT_SYMBOL(neigh_table_init_no_netlink);
2786 EXPORT_SYMBOL(neigh_update);
2787 EXPORT_SYMBOL(pneigh_enqueue);
2788 EXPORT_SYMBOL(pneigh_lookup);
2789
2790 #ifdef CONFIG_ARPD
2791 EXPORT_SYMBOL(neigh_app_ns);
2792 #endif
2793 #ifdef CONFIG_SYSCTL
2794 EXPORT_SYMBOL(neigh_sysctl_register);
2795 EXPORT_SYMBOL(neigh_sysctl_unregister);
2796 #endif