net/core/neighbour.c

   1 /*
   2  *      Generic address resolution entity
   3  *
   4  *      Authors:
   5  *      Pedro Roque             <roque@di.fc.ul.pt>
   6  *      Alexey Kuznetsov        <kuznet@ms2.inr.ac.ru>
   7  *
   8  *      This program is free software; you can redistribute it and/or
   9  *      modify it under the terms of the GNU General Public License
  10  *      as published by the Free Software Foundation; either version
  11  *      2 of the License, or (at your option) any later version.
  12  *
  13  *      Fixes:
  14  *      Vitaly E. Lavrov        releasing NULL neighbor in neigh_add.
  15  *      Harald Welte            Add neighbour cache statistics like rtstat
  16  */
  17
  18 #include <linux/types.h>
  19 #include <linux/kernel.h>
  20 #include <linux/module.h>
  21 #include <linux/socket.h>
  22 #include <linux/netdevice.h>
  23 #include <linux/proc_fs.h>
  24 #ifdef CONFIG_SYSCTL
  25 #include <linux/sysctl.h>
  26 #endif
  27 #include <linux/times.h>
  28 #include <net/net_namespace.h>
  29 #include <net/neighbour.h>
  30 #include <net/dst.h>
  31 #include <net/sock.h>
  32 #include <net/netevent.h>
  33 #include <net/netlink.h>
  34 #include <linux/rtnetlink.h>
  35 #include <linux/random.h>
  36 #include <linux/string.h>
  37 #include <linux/log2.h>
  38
  39 #define NEIGH_DEBUG 1
  40
  41 #define NEIGH_PRINTK(x...) printk(x)
  42 #define NEIGH_NOPRINTK(x...) do { ; } while(0)
  43 #define NEIGH_PRINTK0 NEIGH_PRINTK
  44 #define NEIGH_PRINTK1 NEIGH_NOPRINTK
  45 #define NEIGH_PRINTK2 NEIGH_NOPRINTK
  46
  47 #if NEIGH_DEBUG >= 1
  48 #undef NEIGH_PRINTK1
  49 #define NEIGH_PRINTK1 NEIGH_PRINTK
  50 #endif
  51 #if NEIGH_DEBUG >= 2
  52 #undef NEIGH_PRINTK2
  53 #define NEIGH_PRINTK2 NEIGH_PRINTK
  54 #endif
  55
  56 #define PNEIGH_HASHMASK         0xF
  57
  58 static void neigh_timer_handler(unsigned long arg);
  59 static void __neigh_notify(struct neighbour *n, int type, int flags);
  60 static void neigh_update_notify(struct neighbour *neigh);
  61 static int pneigh_ifdown(struct neigh_table *tbl, struct net_device *dev);
  62
  63 static struct neigh_table *neigh_tables;
  64 #ifdef CONFIG_PROC_FS
  65 static const struct file_operations neigh_stat_seq_fops;
  66 #endif
  67
  68 /*
  69    Neighbour hash table buckets are protected with rwlock tbl->lock.
  70
  71    - All the scans/updates to hash buckets MUST be made under this lock.
  72    - NOTHING clever should be made under this lock: no callbacks
  73      to protocol backends, no attempts to send something to network.
  74      It will result in deadlocks, if backend/driver wants to use neighbour
  75      cache.
  76    - If the entry requires some non-trivial actions, increase
  77      its reference count and release table lock.
  78
  79    Neighbour entries are protected:
  80    - with reference count.
  81    - with rwlock neigh->lock
  82
  83    Reference count prevents destruction.
  84
  85    neigh->lock mainly serializes ll address data and its validity state.
  86    However, the same lock is used to protect another entry fields:
  87     - timer
  88     - resolution queue
  89
  90    Again, nothing clever shall be made under neigh->lock,
  91    the most complicated procedure, which we allow is dev->hard_header.
  92    It is supposed, that dev->hard_header is simplistic and does
  93    not make callbacks to neighbour tables.
  94
  95    The last lock is neigh_tbl_lock. It is pure SMP lock, protecting
  96    list of neighbour tables. This list is used only in process context,
  97  */
  98
  99 static DEFINE_RWLOCK(neigh_tbl_lock);
 100
 101 static int neigh_blackhole(struct sk_buff *skb)
 102 {
 103         kfree_skb(skb);
 104         return -ENETDOWN;
 105 }
 106
 107 static void neigh_cleanup_and_release(struct neighbour *neigh)
 108 {
 109         if (neigh->parms->neigh_cleanup)
 110                 neigh->parms->neigh_cleanup(neigh);
 111
 112         __neigh_notify(neigh, RTM_DELNEIGH, 0);
 113         neigh_release(neigh);
 114 }
 115
 116 /*
 117  * It is random distribution in the interval (1/2)*base...(3/2)*base.
 118  * It corresponds to default IPv6 settings and is not overridable,
 119  * because it is really reasonable choice.
 120  */
 121
 122 unsigned long neigh_rand_reach_time(unsigned long base)
 123 {
 124         return (base ? (net_random() % base) + (base >> 1) : 0);
 125 }
 126 EXPORT_SYMBOL(neigh_rand_reach_time);
 127
 128
 129 static int neigh_forced_gc(struct neigh_table *tbl)
 130 {
 131         int shrunk = 0;
 132         int i;
 133
 134         NEIGH_CACHE_STAT_INC(tbl, forced_gc_runs);
 135
 136         write_lock_bh(&tbl->lock);
 137         for (i = 0; i <= tbl->hash_mask; i++) {
 138                 struct neighbour *n, **np;
 139
 140                 np = &tbl->hash_buckets[i];
 141                 while ((n = *np) != NULL) {
 142                         /* Neighbour record may be discarded if:
 143                          * - nobody refers to it.
 144                          * - it is not permanent
 145                          */
 146                         write_lock(&n->lock);
 147                         if (atomic_read(&n->refcnt) == 1 &&
 148                             !(n->nud_state & NUD_PERMANENT)) {
 149                                 *np     = n->next;
 150                                 n->dead = 1;
 151                                 shrunk  = 1;
 152                                 write_unlock(&n->lock);
 153                                 neigh_cleanup_and_release(n);
 154                                 continue;
 155                         }
 156                         write_unlock(&n->lock);
 157                         np = &n->next;
 158                 }
 159         }
 160
 161         tbl->last_flush = jiffies;
 162
 163         write_unlock_bh(&tbl->lock);
 164
 165         return shrunk;
 166 }
 167
 168 static void neigh_add_timer(struct neighbour *n, unsigned long when)
 169 {
 170         neigh_hold(n);
 171         if (unlikely(mod_timer(&n->timer, when))) {
 172                 printk("NEIGH: BUG, double timer add, state is %x\n",
 173                        n->nud_state);
 174                 dump_stack();
 175         }
 176 }
 177
 178 static int neigh_del_timer(struct neighbour *n)
 179 {
 180         if ((n->nud_state & NUD_IN_TIMER) &&
 181             del_timer(&n->timer)) {
 182                 neigh_release(n);
 183                 return 1;
 184         }
 185         return 0;
 186 }
 187
 188 static void pneigh_queue_purge(struct sk_buff_head *list)
 189 {
 190         struct sk_buff *skb;
 191
 192         while ((skb = skb_dequeue(list)) != NULL) {
 193                 dev_put(skb->dev);
 194                 kfree_skb(skb);
 195         }
 196 }
 197
 198 static void neigh_flush_dev(struct neigh_table *tbl, struct net_device *dev)
 199 {
 200         int i;
 201
 202         for (i = 0; i <= tbl->hash_mask; i++) {
 203                 struct neighbour *n, **np = &tbl->hash_buckets[i];
 204
 205                 while ((n = *np) != NULL) {
 206                         if (dev && n->dev != dev) {
 207                                 np = &n->next;
 208                                 continue;
 209                         }
 210                         *np = n->next;
 211                         write_lock(&n->lock);
 212                         neigh_del_timer(n);
 213                         n->dead = 1;
 214
 215                         if (atomic_read(&n->refcnt) != 1) {
 216                                 /* The most unpleasant situation.
 217                                    We must destroy neighbour entry,
 218                                    but someone still uses it.
 219
 220                                    The destroy will be delayed until
 221                                    the last user releases us, but
 222                                    we must kill timers etc. and move
 223                                    it to safe state.
 224                                  */
 225                                 skb_queue_purge(&n->arp_queue);
 226                                 n->output = neigh_blackhole;
 227                                 if (n->nud_state & NUD_VALID)
 228                                         n->nud_state = NUD_NOARP;
 229                                 else
 230                                         n->nud_state = NUD_NONE;
 231                                 NEIGH_PRINTK2("neigh %p is stray.\n", n);
 232                         }
 233                         write_unlock(&n->lock);
 234                         neigh_cleanup_and_release(n);
 235                 }
 236         }
 237 }
 238
 239 void neigh_changeaddr(struct neigh_table *tbl, struct net_device *dev)
 240 {
 241         write_lock_bh(&tbl->lock);
 242         neigh_flush_dev(tbl, dev);
 243         write_unlock_bh(&tbl->lock);
 244 }
 245 EXPORT_SYMBOL(neigh_changeaddr);
 246
 247 int neigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
 248 {
 249         write_lock_bh(&tbl->lock);
 250         neigh_flush_dev(tbl, dev);
 251         pneigh_ifdown(tbl, dev);
 252         write_unlock_bh(&tbl->lock);
 253
 254         del_timer_sync(&tbl->proxy_timer);
 255         pneigh_queue_purge(&tbl->proxy_queue);
 256         return 0;
 257 }
 258 EXPORT_SYMBOL(neigh_ifdown);
 259
 260 static struct neighbour *neigh_alloc(struct neigh_table *tbl)
 261 {
 262         struct neighbour *n = NULL;
 263         unsigned long now = jiffies;
 264         int entries;
 265
 266         entries = atomic_inc_return(&tbl->entries) - 1;
 267         if (entries >= tbl->gc_thresh3 ||
 268             (entries >= tbl->gc_thresh2 &&
 269              time_after(now, tbl->last_flush + 5 * HZ))) {
 270                 if (!neigh_forced_gc(tbl) &&
 271                     entries >= tbl->gc_thresh3)
 272                         goto out_entries;
 273         }
 274
 275         n = kmem_cache_zalloc(tbl->kmem_cachep, GFP_ATOMIC);
 276         if (!n)
 277                 goto out_entries;
 278
 279         skb_queue_head_init(&n->arp_queue);
 280         rwlock_init(&n->lock);
 281         n->updated        = n->used = now;
 282         n->nud_state      = NUD_NONE;
 283         n->output         = neigh_blackhole;
 284         n->parms          = neigh_parms_clone(&tbl->parms);
 285         setup_timer(&n->timer, neigh_timer_handler, (unsigned long)n);
 286
 287         NEIGH_CACHE_STAT_INC(tbl, allocs);
 288         n->tbl            = tbl;
 289         atomic_set(&n->refcnt, 1);
 290         n->dead           = 1;
 291 out:
 292         return n;
 293
 294 out_entries:
 295         atomic_dec(&tbl->entries);
 296         goto out;
 297 }
 298
 299 static struct neighbour **neigh_hash_alloc(unsigned int entries)
 300 {
 301         unsigned long size = entries * sizeof(struct neighbour *);
 302         struct neighbour **ret;
 303
 304         if (size <= PAGE_SIZE) {
 305                 ret = kzalloc(size, GFP_ATOMIC);
 306         } else {
 307                 ret = (struct neighbour **)
 308                       __get_free_pages(GFP_ATOMIC|__GFP_ZERO, get_order(size));
 309         }
 310         return ret;
 311 }
 312
 313 static void neigh_hash_free(struct neighbour **hash, unsigned int entries)
 314 {
 315         unsigned long size = entries * sizeof(struct neighbour *);
 316
 317         if (size <= PAGE_SIZE)
 318                 kfree(hash);
 319         else
 320                 free_pages((unsigned long)hash, get_order(size));
 321 }
 322
 323 static void neigh_hash_grow(struct neigh_table *tbl, unsigned long new_entries)
 324 {
 325         struct neighbour **new_hash, **old_hash;
 326         unsigned int i, new_hash_mask, old_entries;
 327
 328         NEIGH_CACHE_STAT_INC(tbl, hash_grows);
 329
 330         BUG_ON(!is_power_of_2(new_entries));
 331         new_hash = neigh_hash_alloc(new_entries);
 332         if (!new_hash)
 333                 return;
 334
 335         old_entries = tbl->hash_mask + 1;
 336         new_hash_mask = new_entries - 1;
 337         old_hash = tbl->hash_buckets;
 338
 339         get_random_bytes(&tbl->hash_rnd, sizeof(tbl->hash_rnd));
 340         for (i = 0; i < old_entries; i++) {
 341                 struct neighbour *n, *next;
 342
 343                 for (n = old_hash[i]; n; n = next) {
 344                         unsigned int hash_val = tbl->hash(n->primary_key, n->dev);
 345
 346                         hash_val &= new_hash_mask;
 347                         next = n->next;
 348
 349                         n->next = new_hash[hash_val];
 350                         new_hash[hash_val] = n;
 351                 }
 352         }
 353         tbl->hash_buckets = new_hash;
 354         tbl->hash_mask = new_hash_mask;
 355
 356         neigh_hash_free(old_hash, old_entries);
 357 }
 358
 359 struct neighbour *neigh_lookup(struct neigh_table *tbl, const void *pkey,
 360                                struct net_device *dev)
 361 {
 362         struct neighbour *n;
 363         int key_len = tbl->key_len;
 364         u32 hash_val;
 365
 366         NEIGH_CACHE_STAT_INC(tbl, lookups);
 367
 368         read_lock_bh(&tbl->lock);
 369         hash_val = tbl->hash(pkey, dev);
 370         for (n = tbl->hash_buckets[hash_val & tbl->hash_mask]; n; n = n->next) {
 371                 if (dev == n->dev && !memcmp(n->primary_key, pkey, key_len)) {
 372                         neigh_hold(n);
 373                         NEIGH_CACHE_STAT_INC(tbl, hits);
 374                         break;
 375                 }
 376         }
 377         read_unlock_bh(&tbl->lock);
 378         return n;
 379 }
 380 EXPORT_SYMBOL(neigh_lookup);
 381
 382 struct neighbour *neigh_lookup_nodev(struct neigh_table *tbl, struct net *net,
 383                                      const void *pkey)
 384 {
 385         struct neighbour *n;
 386         int key_len = tbl->key_len;
 387         u32 hash_val;
 388
 389         NEIGH_CACHE_STAT_INC(tbl, lookups);
 390
 391         read_lock_bh(&tbl->lock);
 392         hash_val = tbl->hash(pkey, NULL);
 393         for (n = tbl->hash_buckets[hash_val & tbl->hash_mask]; n; n = n->next) {
 394                 if (!memcmp(n->primary_key, pkey, key_len) &&
 395                     net_eq(dev_net(n->dev), net)) {
 396                         neigh_hold(n);
 397                         NEIGH_CACHE_STAT_INC(tbl, hits);
 398                         break;
 399                 }
 400         }
 401         read_unlock_bh(&tbl->lock);
 402         return n;
 403 }
 404 EXPORT_SYMBOL(neigh_lookup_nodev);
 405
 406 struct neighbour *neigh_create(struct neigh_table *tbl, const void *pkey,
 407                                struct net_device *dev)
 408 {
 409         u32 hash_val;
 410         int key_len = tbl->key_len;
 411         int error;
 412         struct neighbour *n1, *rc, *n = neigh_alloc(tbl);
 413
 414         if (!n) {
 415                 rc = ERR_PTR(-ENOBUFS);
 416                 goto out;
 417         }
 418
 419         memcpy(n->primary_key, pkey, key_len);
 420         n->dev = dev;
 421         dev_hold(dev);
 422
 423         /* Protocol specific setup. */
 424         if (tbl->constructor && (error = tbl->constructor(n)) < 0) {
 425                 rc = ERR_PTR(error);
 426                 goto out_neigh_release;
 427         }
 428
 429         /* Device specific setup. */
 430         if (n->parms->neigh_setup &&
 431             (error = n->parms->neigh_setup(n)) < 0) {
 432                 rc = ERR_PTR(error);
 433                 goto out_neigh_release;
 434         }
 435
 436         n->confirmed = jiffies - (n->parms->base_reachable_time << 1);
 437
 438         write_lock_bh(&tbl->lock);
 439
 440         if (atomic_read(&tbl->entries) > (tbl->hash_mask + 1))
 441                 neigh_hash_grow(tbl, (tbl->hash_mask + 1) << 1);
 442
 443         hash_val = tbl->hash(pkey, dev) & tbl->hash_mask;
 444
 445         if (n->parms->dead) {
 446                 rc = ERR_PTR(-EINVAL);
 447                 goto out_tbl_unlock;
 448         }
 449
 450         for (n1 = tbl->hash_buckets[hash_val]; n1; n1 = n1->next) {
 451                 if (dev == n1->dev && !memcmp(n1->primary_key, pkey, key_len)) {
 452                         neigh_hold(n1);
 453                         rc = n1;
 454                         goto out_tbl_unlock;
 455                 }
 456         }
 457
 458         n->next = tbl->hash_buckets[hash_val];
 459         tbl->hash_buckets[hash_val] = n;
 460         n->dead = 0;
 461         neigh_hold(n);
 462         write_unlock_bh(&tbl->lock);
 463         NEIGH_PRINTK2("neigh %p is created.\n", n);
 464         rc = n;
 465 out:
 466         return rc;
 467 out_tbl_unlock:
 468         write_unlock_bh(&tbl->lock);
 469 out_neigh_release:
 470         neigh_release(n);
 471         goto out;
 472 }
 473 EXPORT_SYMBOL(neigh_create);
 474
 475 static u32 pneigh_hash(const void *pkey, int key_len)
 476 {
 477         u32 hash_val = *(u32 *)(pkey + key_len - 4);
 478         hash_val ^= (hash_val >> 16);
 479         hash_val ^= hash_val >> 8;
 480         hash_val ^= hash_val >> 4;
 481         hash_val &= PNEIGH_HASHMASK;
 482         return hash_val;
 483 }
 484
 485 static struct pneigh_entry *__pneigh_lookup_1(struct pneigh_entry *n,
 486                                               struct net *net,
 487                                               const void *pkey,
 488                                               int key_len,
 489                                               struct net_device *dev)
 490 {
 491         while (n) {
 492                 if (!memcmp(n->key, pkey, key_len) &&
 493                     net_eq(pneigh_net(n), net) &&
 494                     (n->dev == dev || !n->dev))
 495                         return n;
 496                 n = n->next;
 497         }
 498         return NULL;
 499 }
 500
 501 struct pneigh_entry *__pneigh_lookup(struct neigh_table *tbl,
 502                 struct net *net, const void *pkey, struct net_device *dev)
 503 {
 504         int key_len = tbl->key_len;
 505         u32 hash_val = pneigh_hash(pkey, key_len);
 506
 507         return __pneigh_lookup_1(tbl->phash_buckets[hash_val],
 508                                  net, pkey, key_len, dev);
 509 }
 510 EXPORT_SYMBOL_GPL(__pneigh_lookup);
 511
 512 struct pneigh_entry * pneigh_lookup(struct neigh_table *tbl,
 513                                     struct net *net, const void *pkey,
 514                                     struct net_device *dev, int creat)
 515 {
 516         struct pneigh_entry *n;
 517         int key_len = tbl->key_len;
 518         u32 hash_val = pneigh_hash(pkey, key_len);
 519
 520         read_lock_bh(&tbl->lock);
 521         n = __pneigh_lookup_1(tbl->phash_buckets[hash_val],
 522                               net, pkey, key_len, dev);
 523         read_unlock_bh(&tbl->lock);
 524
 525         if (n || !creat)
 526                 goto out;
 527
 528         ASSERT_RTNL();
 529
 530         n = kmalloc(sizeof(*n) + key_len, GFP_KERNEL);
 531         if (!n)
 532                 goto out;
 533
 534 #ifdef CONFIG_NET_NS
 535         n->net = hold_net(net);
 536 #endif
 537         memcpy(n->key, pkey, key_len);
 538         n->dev = dev;
 539         if (dev)
 540                 dev_hold(dev);
 541
 542         if (tbl->pconstructor && tbl->pconstructor(n)) {
 543                 if (dev)
 544                         dev_put(dev);
 545                 release_net(net);
 546                 kfree(n);
 547                 n = NULL;
 548                 goto out;
 549         }
 550
 551         write_lock_bh(&tbl->lock);
 552         n->next = tbl->phash_buckets[hash_val];
 553         tbl->phash_buckets[hash_val] = n;
 554         write_unlock_bh(&tbl->lock);
 555 out:
 556         return n;
 557 }
 558 EXPORT_SYMBOL(pneigh_lookup);
 559
 560
 561 int pneigh_delete(struct neigh_table *tbl, struct net *net, const void *pkey,
 562                   struct net_device *dev)
 563 {
 564         struct pneigh_entry *n, **np;
 565         int key_len = tbl->key_len;
 566         u32 hash_val = pneigh_hash(pkey, key_len);
 567
 568         write_lock_bh(&tbl->lock);
 569         for (np = &tbl->phash_buckets[hash_val]; (n = *np) != NULL;
 570              np = &n->next) {
 571                 if (!memcmp(n->key, pkey, key_len) && n->dev == dev &&
 572                     net_eq(pneigh_net(n), net)) {
 573                         *np = n->next;
 574                         write_unlock_bh(&tbl->lock);
 575                         if (tbl->pdestructor)
 576                                 tbl->pdestructor(n);
 577                         if (n->dev)
 578                                 dev_put(n->dev);
 579                         release_net(pneigh_net(n));
 580                         kfree(n);
 581                         return 0;
 582                 }
 583         }
 584         write_unlock_bh(&tbl->lock);
 585         return -ENOENT;
 586 }
 587
 588 static int pneigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
 589 {
 590         struct pneigh_entry *n, **np;
 591         u32 h;
 592
 593         for (h = 0; h <= PNEIGH_HASHMASK; h++) {
 594                 np = &tbl->phash_buckets[h];
 595                 while ((n = *np) != NULL) {
 596                         if (!dev || n->dev == dev) {
 597                                 *np = n->next;
 598                                 if (tbl->pdestructor)
 599                                         tbl->pdestructor(n);
 600                                 if (n->dev)
 601                                         dev_put(n->dev);
 602                                 release_net(pneigh_net(n));
 603                                 kfree(n);
 604                                 continue;
 605                         }
 606                         np = &n->next;
 607                 }
 608         }
 609         return -ENOENT;
 610 }
 611
 612 static void neigh_parms_destroy(struct neigh_parms *parms);
 613
 614 static inline void neigh_parms_put(struct neigh_parms *parms)
 615 {
 616         if (atomic_dec_and_test(&parms->refcnt))
 617                 neigh_parms_destroy(parms);
 618 }
 619
 620 /*
 621  *      neighbour must already be out of the table;
 622  *
 623  */
 624 void neigh_destroy(struct neighbour *neigh)
 625 {
 626         struct hh_cache *hh;
 627
 628         NEIGH_CACHE_STAT_INC(neigh->tbl, destroys);
 629
 630         if (!neigh->dead) {
 631                 printk(KERN_WARNING
 632                        "Destroying alive neighbour %p\n", neigh);
 633                 dump_stack();
 634                 return;
 635         }
 636
 637         if (neigh_del_timer(neigh))
 638                 printk(KERN_WARNING "Impossible event.\n");
 639
 640         while ((hh = neigh->hh) != NULL) {
 641                 neigh->hh = hh->hh_next;
 642                 hh->hh_next = NULL;
 643
 644                 write_seqlock_bh(&hh->hh_lock);
 645                 hh->hh_output = neigh_blackhole;
 646                 write_sequnlock_bh(&hh->hh_lock);
 647                 if (atomic_dec_and_test(&hh->hh_refcnt))
 648                         kfree(hh);
 649         }
 650
 651         skb_queue_purge(&neigh->arp_queue);
 652
 653         dev_put(neigh->dev);
 654         neigh_parms_put(neigh->parms);
 655
 656         NEIGH_PRINTK2("neigh %p is destroyed.\n", neigh);
 657
 658         atomic_dec(&neigh->tbl->entries);
 659         kmem_cache_free(neigh->tbl->kmem_cachep, neigh);
 660 }
 661 EXPORT_SYMBOL(neigh_destroy);
 662
 663 /* Neighbour state is suspicious;
 664    disable fast path.
 665
 666    Called with write_locked neigh.
 667  */
 668 static void neigh_suspect(struct neighbour *neigh)
 669 {
 670         struct hh_cache *hh;
 671
 672         NEIGH_PRINTK2("neigh %p is suspected.\n", neigh);
 673
 674         neigh->output = neigh->ops->output;
 675
 676         for (hh = neigh->hh; hh; hh = hh->hh_next)
 677                 hh->hh_output = neigh->ops->output;
 678 }
 679
 680 /* Neighbour state is OK;
 681    enable fast path.
 682
 683    Called with write_locked neigh.
 684  */
 685 static void neigh_connect(struct neighbour *neigh)
 686 {
 687         struct hh_cache *hh;
 688
 689         NEIGH_PRINTK2("neigh %p is connected.\n", neigh);
 690
 691         neigh->output = neigh->ops->connected_output;
 692
 693         for (hh = neigh->hh; hh; hh = hh->hh_next)
 694                 hh->hh_output = neigh->ops->hh_output;
 695 }
 696
 697 static void neigh_periodic_timer(unsigned long arg)
 698 {
 699         struct neigh_table *tbl = (struct neigh_table *)arg;
 700         struct neighbour *n, **np;
 701         unsigned long expire, now = jiffies;
 702
 703         NEIGH_CACHE_STAT_INC(tbl, periodic_gc_runs);
 704
 705         write_lock(&tbl->lock);
 706
 707         /*
 708          *      periodically recompute ReachableTime from random function
 709          */
 710
 711         if (time_after(now, tbl->last_rand + 300 * HZ)) {
 712                 struct neigh_parms *p;
 713                 tbl->last_rand = now;
 714                 for (p = &tbl->parms; p; p = p->next)
 715                         p->reachable_time =
 716                                 neigh_rand_reach_time(p->base_reachable_time);
 717         }
 718
 719         np = &tbl->hash_buckets[tbl->hash_chain_gc];
 720         tbl->hash_chain_gc = ((tbl->hash_chain_gc + 1) & tbl->hash_mask);
 721
 722         while ((n = *np) != NULL) {
 723                 unsigned int state;
 724
 725                 write_lock(&n->lock);
 726
 727                 state = n->nud_state;
 728                 if (state & (NUD_PERMANENT | NUD_IN_TIMER)) {
 729                         write_unlock(&n->lock);
 730                         goto next_elt;
 731                 }
 732
 733                 if (time_before(n->used, n->confirmed))
 734                         n->used = n->confirmed;
 735
 736                 if (atomic_read(&n->refcnt) == 1 &&
 737                     (state == NUD_FAILED ||
 738                      time_after(now, n->used + n->parms->gc_staletime))) {
 739                         *np = n->next;
 740                         n->dead = 1;
 741                         write_unlock(&n->lock);
 742                         neigh_cleanup_and_release(n);
 743                         continue;
 744                 }
 745                 write_unlock(&n->lock);
 746
 747 next_elt:
 748                 np = &n->next;
 749         }
 750
 751         /* Cycle through all hash buckets every base_reachable_time/2 ticks.
 752          * ARP entry timeouts range from 1/2 base_reachable_time to 3/2
 753          * base_reachable_time.
 754          */
 755         expire = tbl->parms.base_reachable_time >> 1;
 756         expire /= (tbl->hash_mask + 1);
 757         if (!expire)
 758                 expire = 1;
 759
 760         if (expire>HZ)
 761                 mod_timer(&tbl->gc_timer, round_jiffies(now + expire));
 762         else
 763                 mod_timer(&tbl->gc_timer, now + expire);
 764
 765         write_unlock(&tbl->lock);
 766 }
 767
 768 static __inline__ int neigh_max_probes(struct neighbour *n)
 769 {
 770         struct neigh_parms *p = n->parms;
 771         return (n->nud_state & NUD_PROBE ?
 772                 p->ucast_probes :
 773                 p->ucast_probes + p->app_probes + p->mcast_probes);
 774 }
 775
 776 /* Called when a timer expires for a neighbour entry. */
 777
 778 static void neigh_timer_handler(unsigned long arg)
 779 {
 780         unsigned long now, next;
 781         struct neighbour *neigh = (struct neighbour *)arg;
 782         unsigned state;
 783         int notify = 0;
 784
 785         write_lock(&neigh->lock);
 786
 787         state = neigh->nud_state;
 788         now = jiffies;
 789         next = now + HZ;
 790
 791         if (!(state & NUD_IN_TIMER)) {
 792 #ifndef CONFIG_SMP
 793                 printk(KERN_WARNING "neigh: timer & !nud_in_timer\n");
 794 #endif
 795                 goto out;
 796         }
 797
 798         if (state & NUD_REACHABLE) {
 799                 if (time_before_eq(now,
 800                                    neigh->confirmed + neigh->parms->reachable_time)) {
 801                         NEIGH_PRINTK2("neigh %p is still alive.\n", neigh);
 802                         next = neigh->confirmed + neigh->parms->reachable_time;
 803                 } else if (time_before_eq(now,
 804                                           neigh->used + neigh->parms->delay_probe_time)) {
 805                         NEIGH_PRINTK2("neigh %p is delayed.\n", neigh);
 806                         neigh->nud_state = NUD_DELAY;
 807                         neigh->updated = jiffies;
 808                         neigh_suspect(neigh);
 809                         next = now + neigh->parms->delay_probe_time;
 810                 } else {
 811                         NEIGH_PRINTK2("neigh %p is suspected.\n", neigh);
 812                         neigh->nud_state = NUD_STALE;
 813                         neigh->updated = jiffies;
 814                         neigh_suspect(neigh);
 815                         notify = 1;
 816                 }
 817         } else if (state & NUD_DELAY) {
 818                 if (time_before_eq(now,
 819                                    neigh->confirmed + neigh->parms->delay_probe_time)) {
 820                         NEIGH_PRINTK2("neigh %p is now reachable.\n", neigh);
 821                         neigh->nud_state = NUD_REACHABLE;
 822                         neigh->updated = jiffies;
 823                         neigh_connect(neigh);
 824                         notify = 1;
 825                         next = neigh->confirmed + neigh->parms->reachable_time;
 826                 } else {
 827                         NEIGH_PRINTK2("neigh %p is probed.\n", neigh);
 828                         neigh->nud_state = NUD_PROBE;
 829                         neigh->updated = jiffies;
 830                         atomic_set(&neigh->probes, 0);
 831                         next = now + neigh->parms->retrans_time;
 832                 }
 833         } else {
 834                 /* NUD_PROBE|NUD_INCOMPLETE */
 835                 next = now + neigh->parms->retrans_time;
 836         }
 837
 838         if ((neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) &&
 839             atomic_read(&neigh->probes) >= neigh_max_probes(neigh)) {
 840                 struct sk_buff *skb;
 841
 842                 neigh->nud_state = NUD_FAILED;
 843                 neigh->updated = jiffies;
 844                 notify = 1;
 845                 NEIGH_CACHE_STAT_INC(neigh->tbl, res_failed);
 846                 NEIGH_PRINTK2("neigh %p is failed.\n", neigh);
 847
 848                 /* It is very thin place. report_unreachable is very complicated
 849                    routine. Particularly, it can hit the same neighbour entry!
 850
 851                    So that, we try to be accurate and avoid dead loop. --ANK
 852                  */
 853                 while (neigh->nud_state == NUD_FAILED &&
 854                        (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
 855                         write_unlock(&neigh->lock);
 856                         neigh->ops->error_report(neigh, skb);
 857                         write_lock(&neigh->lock);
 858                 }
 859                 skb_queue_purge(&neigh->arp_queue);
 860         }
 861
 862         if (neigh->nud_state & NUD_IN_TIMER) {
 863                 if (time_before(next, jiffies + HZ/2))
 864                         next = jiffies + HZ/2;
 865                 if (!mod_timer(&neigh->timer, next))
 866                         neigh_hold(neigh);
 867         }
 868         if (neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) {
 869                 struct sk_buff *skb = skb_peek(&neigh->arp_queue);
 870                 /* keep skb alive even if arp_queue overflows */
 871                 if (skb)
 872                         skb = skb_copy(skb, GFP_ATOMIC);
 873                 write_unlock(&neigh->lock);
 874                 neigh->ops->solicit(neigh, skb);
 875                 atomic_inc(&neigh->probes);
 876                 if (skb)
 877                         kfree_skb(skb);
 878         } else {
 879 out:
 880                 write_unlock(&neigh->lock);
 881         }
 882
 883         if (notify)
 884                 neigh_update_notify(neigh);
 885
 886         neigh_release(neigh);
 887 }
 888
 889 int __neigh_event_send(struct neighbour *neigh, struct sk_buff *skb)
 890 {
 891         int rc;
 892         unsigned long now;
 893
 894         write_lock_bh(&neigh->lock);
 895
 896         rc = 0;
 897         if (neigh->nud_state & (NUD_CONNECTED | NUD_DELAY | NUD_PROBE))
 898                 goto out_unlock_bh;
 899
 900         now = jiffies;
 901
 902         if (!(neigh->nud_state & (NUD_STALE | NUD_INCOMPLETE))) {
 903                 if (neigh->parms->mcast_probes + neigh->parms->app_probes) {
 904                         atomic_set(&neigh->probes, neigh->parms->ucast_probes);
 905                         neigh->nud_state     = NUD_INCOMPLETE;
 906                         neigh->updated = jiffies;
 907                         neigh_add_timer(neigh, now + 1);
 908                 } else {
 909                         neigh->nud_state = NUD_FAILED;
 910                         neigh->updated = jiffies;
 911                         write_unlock_bh(&neigh->lock);
 912
 913                         if (skb)
 914                                 kfree_skb(skb);
 915                         return 1;
 916                 }
 917         } else if (neigh->nud_state & NUD_STALE) {
 918                 NEIGH_PRINTK2("neigh %p is delayed.\n", neigh);
 919                 neigh->nud_state = NUD_DELAY;
 920                 neigh->updated = jiffies;
 921                 neigh_add_timer(neigh,
 922                                 jiffies + neigh->parms->delay_probe_time);
 923         }
 924
 925         if (neigh->nud_state == NUD_INCOMPLETE) {
 926                 if (skb) {
 927                         if (skb_queue_len(&neigh->arp_queue) >=
 928                             neigh->parms->queue_len) {
 929                                 struct sk_buff *buff;
 930                                 buff = neigh->arp_queue.next;
 931                                 __skb_unlink(buff, &neigh->arp_queue);
 932                                 kfree_skb(buff);
 933                         }
 934                         __skb_queue_tail(&neigh->arp_queue, skb);
 935                 }
 936                 rc = 1;
 937         }
 938 out_unlock_bh:
 939         write_unlock_bh(&neigh->lock);
 940         return rc;
 941 }
 942 EXPORT_SYMBOL(__neigh_event_send);
 943
 944 static void neigh_update_hhs(struct neighbour *neigh)
 945 {
 946         struct hh_cache *hh;
 947         void (*update)(struct hh_cache*, const struct net_device*, const unsigned char *)
 948                 = neigh->dev->header_ops->cache_update;
 949
 950         if (update) {
 951                 for (hh = neigh->hh; hh; hh = hh->hh_next) {
 952                         write_seqlock_bh(&hh->hh_lock);
 953                         update(hh, neigh->dev, neigh->ha);
 954                         write_sequnlock_bh(&hh->hh_lock);
 955                 }
 956         }
 957 }
 958
 959
 960
 961 /* Generic update routine.
 962    -- lladdr is new lladdr or NULL, if it is not supplied.
 963    -- new    is new state.
 964    -- flags
 965         NEIGH_UPDATE_F_OVERRIDE allows to override existing lladdr,
 966                                 if it is different.
 967         NEIGH_UPDATE_F_WEAK_OVERRIDE will suspect existing "connected"
 968                                 lladdr instead of overriding it
 969                                 if it is different.
 970                                 It also allows to retain current state
 971                                 if lladdr is unchanged.
 972         NEIGH_UPDATE_F_ADMIN    means that the change is administrative.
 973
 974         NEIGH_UPDATE_F_OVERRIDE_ISROUTER allows to override existing
 975                                 NTF_ROUTER flag.
 976         NEIGH_UPDATE_F_ISROUTER indicates if the neighbour is known as
 977                                 a router.
 978
 979    Caller MUST hold reference count on the entry.
 980  */
 981
 982 int neigh_update(struct neighbour *neigh, const u8 *lladdr, u8 new,
 983                  u32 flags)
 984 {
 985         u8 old;
 986         int err;
 987         int notify = 0;
 988         struct net_device *dev;
 989         int update_isrouter = 0;
 990
 991         write_lock_bh(&neigh->lock);
 992
 993         dev    = neigh->dev;
 994         old    = neigh->nud_state;
 995         err    = -EPERM;
 996
 997         if (!(flags & NEIGH_UPDATE_F_ADMIN) &&
 998             (old & (NUD_NOARP | NUD_PERMANENT)))
 999                 goto out;
1000
1001         if (!(new & NUD_VALID)) {
1002                 neigh_del_timer(neigh);
1003                 if (old & NUD_CONNECTED)
1004                         neigh_suspect(neigh);
1005                 neigh->nud_state = new;
1006                 err = 0;
1007                 notify = old & NUD_VALID;
1008                 goto out;
1009         }
1010
1011         /* Compare new lladdr with cached one */
1012         if (!dev->addr_len) {
1013                 /* First case: device needs no address. */
1014                 lladdr = neigh->ha;
1015         } else if (lladdr) {
1016                 /* The second case: if something is already cached
1017                    and a new address is proposed:
1018                    - compare new & old
1019                    - if they are different, check override flag
1020                  */
1021                 if ((old & NUD_VALID) &&
1022                     !memcmp(lladdr, neigh->ha, dev->addr_len))
1023                         lladdr = neigh->ha;
1024         } else {
1025                 /* No address is supplied; if we know something,
1026                    use it, otherwise discard the request.
1027                  */
1028                 err = -EINVAL;
1029                 if (!(old & NUD_VALID))
1030                         goto out;
1031                 lladdr = neigh->ha;
1032         }
1033
1034         if (new & NUD_CONNECTED)
1035                 neigh->confirmed = jiffies;
1036         neigh->updated = jiffies;
1037
1038         /* If entry was valid and address is not changed,
1039            do not change entry state, if new one is STALE.
1040          */
1041         err = 0;
1042         update_isrouter = flags & NEIGH_UPDATE_F_OVERRIDE_ISROUTER;
1043         if (old & NUD_VALID) {
1044                 if (lladdr != neigh->ha && !(flags & NEIGH_UPDATE_F_OVERRIDE)) {
1045                         update_isrouter = 0;
1046                         if ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) &&
1047                             (old & NUD_CONNECTED)) {
1048                                 lladdr = neigh->ha;
1049                                 new = NUD_STALE;
1050                         } else
1051                                 goto out;
1052                 } else {
1053                         if (lladdr == neigh->ha && new == NUD_STALE &&
1054                             ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) ||
1055                              (old & NUD_CONNECTED))
1056                             )
1057                                 new = old;
1058                 }
1059         }
1060
1061         if (new != old) {
1062                 neigh_del_timer(neigh);
1063                 if (new & NUD_IN_TIMER)
1064                         neigh_add_timer(neigh, (jiffies +
1065                                                 ((new & NUD_REACHABLE) ?
1066                                                  neigh->parms->reachable_time :
1067                                                  0)));
1068                 neigh->nud_state = new;
1069         }
1070
1071         if (lladdr != neigh->ha) {
1072                 memcpy(&neigh->ha, lladdr, dev->addr_len);
1073                 neigh_update_hhs(neigh);
1074                 if (!(new & NUD_CONNECTED))
1075                         neigh->confirmed = jiffies -
1076                                       (neigh->parms->base_reachable_time << 1);
1077                 notify = 1;
1078         }
1079         if (new == old)
1080                 goto out;
1081         if (new & NUD_CONNECTED)
1082                 neigh_connect(neigh);
1083         else
1084                 neigh_suspect(neigh);
1085         if (!(old & NUD_VALID)) {
1086                 struct sk_buff *skb;
1087
1088                 /* Again: avoid dead loop if something went wrong */
1089
1090                 while (neigh->nud_state & NUD_VALID &&
1091                        (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
1092                         struct neighbour *n1 = neigh;
1093                         write_unlock_bh(&neigh->lock);
1094                         /* On shaper/eql skb->dst->neighbour != neigh :( */
1095                         if (skb->dst && skb->dst->neighbour)
1096                                 n1 = skb->dst->neighbour;
1097                         n1->output(skb);
1098                         write_lock_bh(&neigh->lock);
1099                 }
1100                 skb_queue_purge(&neigh->arp_queue);
1101         }
1102 out:
1103         if (update_isrouter) {
1104                 neigh->flags = (flags & NEIGH_UPDATE_F_ISROUTER) ?
1105                         (neigh->flags | NTF_ROUTER) :
1106                         (neigh->flags & ~NTF_ROUTER);
1107         }
1108         write_unlock_bh(&neigh->lock);
1109
1110         if (notify)
1111                 neigh_update_notify(neigh);
1112
1113         return err;
1114 }
1115 EXPORT_SYMBOL(neigh_update);
1116
1117 struct neighbour *neigh_event_ns(struct neigh_table *tbl,
1118                                  u8 *lladdr, void *saddr,
1119                                  struct net_device *dev)
1120 {
1121         struct neighbour *neigh = __neigh_lookup(tbl, saddr, dev,
1122                                                  lladdr || !dev->addr_len);
1123         if (neigh)
1124                 neigh_update(neigh, lladdr, NUD_STALE,
1125                              NEIGH_UPDATE_F_OVERRIDE);
1126         return neigh;
1127 }
1128 EXPORT_SYMBOL(neigh_event_ns);
1129
1130 static void neigh_hh_init(struct neighbour *n, struct dst_entry *dst,
1131                           __be16 protocol)
1132 {
1133         struct hh_cache *hh;
1134         struct net_device *dev = dst->dev;
1135
1136         for (hh = n->hh; hh; hh = hh->hh_next)
1137                 if (hh->hh_type == protocol)
1138                         break;
1139
1140         if (!hh && (hh = kzalloc(sizeof(*hh), GFP_ATOMIC)) != NULL) {
1141                 seqlock_init(&hh->hh_lock);
1142                 hh->hh_type = protocol;
1143                 atomic_set(&hh->hh_refcnt, 0);
1144                 hh->hh_next = NULL;
1145
1146                 if (dev->header_ops->cache(n, hh)) {
1147                         kfree(hh);
1148                         hh = NULL;
1149                 } else {
1150                         atomic_inc(&hh->hh_refcnt);
1151                         hh->hh_next = n->hh;
1152                         n->hh       = hh;
1153                         if (n->nud_state & NUD_CONNECTED)
1154                                 hh->hh_output = n->ops->hh_output;
1155                         else
1156                                 hh->hh_output = n->ops->output;
1157                 }
1158         }
1159         if (hh) {
1160                 atomic_inc(&hh->hh_refcnt);
1161                 dst->hh = hh;
1162         }
1163 }
1164
1165 /* This function can be used in contexts, where only old dev_queue_xmit
1166    worked, f.e. if you want to override normal output path (eql, shaper),
1167    but resolution is not made yet.
1168  */
1169
1170 int neigh_compat_output(struct sk_buff *skb)
1171 {
1172         struct net_device *dev = skb->dev;
1173
1174         __skb_pull(skb, skb_network_offset(skb));
1175
1176         if (dev_hard_header(skb, dev, ntohs(skb->protocol), NULL, NULL,
1177                             skb->len) < 0 &&
1178             dev->header_ops->rebuild(skb))
1179                 return 0;
1180
1181         return dev_queue_xmit(skb);
1182 }
1183 EXPORT_SYMBOL(neigh_compat_output);
1184
1185 /* Slow and careful. */
1186
1187 int neigh_resolve_output(struct sk_buff *skb)
1188 {
1189         struct dst_entry *dst = skb->dst;
1190         struct neighbour *neigh;
1191         int rc = 0;
1192
1193         if (!dst || !(neigh = dst->neighbour))
1194                 goto discard;
1195
1196         __skb_pull(skb, skb_network_offset(skb));
1197
1198         if (!neigh_event_send(neigh, skb)) {
1199                 int err;
1200                 struct net_device *dev = neigh->dev;
1201                 if (dev->header_ops->cache && !dst->hh) {
1202                         write_lock_bh(&neigh->lock);
1203                         if (!dst->hh)
1204                                 neigh_hh_init(neigh, dst, dst->ops->protocol);
1205                         err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1206                                               neigh->ha, NULL, skb->len);
1207                         write_unlock_bh(&neigh->lock);
1208                 } else {
1209                         read_lock_bh(&neigh->lock);
1210                         err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1211                                               neigh->ha, NULL, skb->len);
1212                         read_unlock_bh(&neigh->lock);
1213                 }
1214                 if (err >= 0)
1215                         rc = neigh->ops->queue_xmit(skb);
1216                 else
1217                         goto out_kfree_skb;
1218         }
1219 out:
1220         return rc;
1221 discard:
1222         NEIGH_PRINTK1("neigh_resolve_output: dst=%p neigh=%p\n",
1223                       dst, dst ? dst->neighbour : NULL);
1224 out_kfree_skb:
1225         rc = -EINVAL;
1226         kfree_skb(skb);
1227         goto out;
1228 }
1229 EXPORT_SYMBOL(neigh_resolve_output);
1230
1231 /* As fast as possible without hh cache */
1232
1233 int neigh_connected_output(struct sk_buff *skb)
1234 {
1235         int err;
1236         struct dst_entry *dst = skb->dst;
1237         struct neighbour *neigh = dst->neighbour;
1238         struct net_device *dev = neigh->dev;
1239
1240         __skb_pull(skb, skb_network_offset(skb));
1241
1242         read_lock_bh(&neigh->lock);
1243         err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1244                               neigh->ha, NULL, skb->len);
1245         read_unlock_bh(&neigh->lock);
1246         if (err >= 0)
1247                 err = neigh->ops->queue_xmit(skb);
1248         else {
1249                 err = -EINVAL;
1250                 kfree_skb(skb);
1251         }
1252         return err;
1253 }
1254 EXPORT_SYMBOL(neigh_connected_output);
1255
1256 static void neigh_proxy_process(unsigned long arg)
1257 {
1258         struct neigh_table *tbl = (struct neigh_table *)arg;
1259         long sched_next = 0;
1260         unsigned long now = jiffies;
1261         struct sk_buff *skb;
1262
1263         spin_lock(&tbl->proxy_queue.lock);
1264
1265         skb = tbl->proxy_queue.next;
1266
1267         while (skb != (struct sk_buff *)&tbl->proxy_queue) {
1268                 struct sk_buff *back = skb;
1269                 long tdif = NEIGH_CB(back)->sched_next - now;
1270
1271                 skb = skb->next;
1272                 if (tdif <= 0) {
1273                         struct net_device *dev = back->dev;
1274                         __skb_unlink(back, &tbl->proxy_queue);
1275                         if (tbl->proxy_redo && netif_running(dev))
1276                                 tbl->proxy_redo(back);
1277                         else
1278                                 kfree_skb(back);
1279
1280                         dev_put(dev);
1281                 } else if (!sched_next || tdif < sched_next)
1282                         sched_next = tdif;
1283         }
1284         del_timer(&tbl->proxy_timer);
1285         if (sched_next)
1286                 mod_timer(&tbl->proxy_timer, jiffies + sched_next);
1287         spin_unlock(&tbl->proxy_queue.lock);
1288 }
1289
1290 void pneigh_enqueue(struct neigh_table *tbl, struct neigh_parms *p,
1291                     struct sk_buff *skb)
1292 {
1293         unsigned long now = jiffies;
1294         unsigned long sched_next = now + (net_random() % p->proxy_delay);
1295
1296         if (tbl->proxy_queue.qlen > p->proxy_qlen) {
1297                 kfree_skb(skb);
1298                 return;
1299         }
1300
1301         NEIGH_CB(skb)->sched_next = sched_next;
1302         NEIGH_CB(skb)->flags |= LOCALLY_ENQUEUED;
1303
1304         spin_lock(&tbl->proxy_queue.lock);
1305         if (del_timer(&tbl->proxy_timer)) {
1306                 if (time_before(tbl->proxy_timer.expires, sched_next))
1307                         sched_next = tbl->proxy_timer.expires;
1308         }
1309         dst_release(skb->dst);
1310         skb->dst = NULL;
1311         dev_hold(skb->dev);
1312         __skb_queue_tail(&tbl->proxy_queue, skb);
1313         mod_timer(&tbl->proxy_timer, sched_next);
1314         spin_unlock(&tbl->proxy_queue.lock);
1315 }
1316 EXPORT_SYMBOL(pneigh_enqueue);
1317
1318 static inline struct neigh_parms *lookup_neigh_params(struct neigh_table *tbl,
1319                                                       struct net *net, int ifindex)
1320 {
1321         struct neigh_parms *p;
1322
1323         for (p = &tbl->parms; p; p = p->next) {
1324                 if ((p->dev && p->dev->ifindex == ifindex && net_eq(neigh_parms_net(p), net)) ||
1325                     (!p->dev && !ifindex))
1326                         return p;
1327         }
1328
1329         return NULL;
1330 }
1331
1332 struct neigh_parms *neigh_parms_alloc(struct net_device *dev,
1333                                       struct neigh_table *tbl)
1334 {
1335         struct neigh_parms *p, *ref;
1336         struct net *net;
1337
1338         net = dev_net(dev);
1339         ref = lookup_neigh_params(tbl, net, 0);
1340         if (!ref)
1341                 return NULL;
1342
1343         p = kmemdup(ref, sizeof(*p), GFP_KERNEL);
1344         if (p) {
1345                 p->tbl            = tbl;
1346                 atomic_set(&p->refcnt, 1);
1347                 INIT_RCU_HEAD(&p->rcu_head);
1348                 p->reachable_time =
1349                                 neigh_rand_reach_time(p->base_reachable_time);
1350
1351                 if (dev->neigh_setup && dev->neigh_setup(dev, p)) {
1352                         kfree(p);
1353                         return NULL;
1354                 }
1355
1356                 dev_hold(dev);
1357                 p->dev = dev;
1358 #ifdef CONFIG_NET_NS
1359                 p->net = hold_net(net);
1360 #endif
1361                 p->sysctl_table = NULL;
1362                 write_lock_bh(&tbl->lock);
1363                 p->next         = tbl->parms.next;
1364                 tbl->parms.next = p;
1365                 write_unlock_bh(&tbl->lock);
1366         }
1367         return p;
1368 }
1369 EXPORT_SYMBOL(neigh_parms_alloc);
1370
1371 static void neigh_rcu_free_parms(struct rcu_head *head)
1372 {
1373         struct neigh_parms *parms =
1374                 container_of(head, struct neigh_parms, rcu_head);
1375
1376         neigh_parms_put(parms);
1377 }
1378
1379 void neigh_parms_release(struct neigh_table *tbl, struct neigh_parms *parms)
1380 {
1381         struct neigh_parms **p;
1382
1383         if (!parms || parms == &tbl->parms)
1384                 return;
1385         write_lock_bh(&tbl->lock);
1386         for (p = &tbl->parms.next; *p; p = &(*p)->next) {
1387                 if (*p == parms) {
1388                         *p = parms->next;
1389                         parms->dead = 1;
1390                         write_unlock_bh(&tbl->lock);
1391                         if (parms->dev)
1392                                 dev_put(parms->dev);
1393                         call_rcu(&parms->rcu_head, neigh_rcu_free_parms);
1394                         return;
1395                 }
1396         }
1397         write_unlock_bh(&tbl->lock);
1398         NEIGH_PRINTK1("neigh_parms_release: not found\n");
1399 }
1400 EXPORT_SYMBOL(neigh_parms_release);
1401
1402 static void neigh_parms_destroy(struct neigh_parms *parms)
1403 {
1404         release_net(neigh_parms_net(parms));
1405         kfree(parms);
1406 }
1407
1408 static struct lock_class_key neigh_table_proxy_queue_class;
1409
1410 void neigh_table_init_no_netlink(struct neigh_table *tbl)
1411 {
1412         unsigned long now = jiffies;
1413         unsigned long phsize;
1414
1415 #ifdef CONFIG_NET_NS
1416         tbl->parms.net = &init_net;
1417 #endif
1418         atomic_set(&tbl->parms.refcnt, 1);
1419         INIT_RCU_HEAD(&tbl->parms.rcu_head);
1420         tbl->parms.reachable_time =
1421                           neigh_rand_reach_time(tbl->parms.base_reachable_time);
1422
1423         if (!tbl->kmem_cachep)
1424                 tbl->kmem_cachep =
1425                         kmem_cache_create(tbl->id, tbl->entry_size, 0,
1426                                           SLAB_HWCACHE_ALIGN|SLAB_PANIC,
1427                                           NULL);
1428         tbl->stats = alloc_percpu(struct neigh_statistics);
1429         if (!tbl->stats)
1430                 panic("cannot create neighbour cache statistics");
1431
1432 #ifdef CONFIG_PROC_FS
1433         tbl->pde = proc_create_data(tbl->id, 0, init_net.proc_net_stat,
1434                                     &neigh_stat_seq_fops, tbl);
1435         if (!tbl->pde)
1436                 panic("cannot create neighbour proc dir entry");
1437 #endif
1438
1439         tbl->hash_mask = 1;
1440         tbl->hash_buckets = neigh_hash_alloc(tbl->hash_mask + 1);
1441
1442         phsize = (PNEIGH_HASHMASK + 1) * sizeof(struct pneigh_entry *);
1443         tbl->phash_buckets = kzalloc(phsize, GFP_KERNEL);
1444
1445         if (!tbl->hash_buckets || !tbl->phash_buckets)
1446                 panic("cannot allocate neighbour cache hashes");
1447
1448         get_random_bytes(&tbl->hash_rnd, sizeof(tbl->hash_rnd));
1449
1450         rwlock_init(&tbl->lock);
1451         setup_timer(&tbl->gc_timer, neigh_periodic_timer, (unsigned long)tbl);
1452         tbl->gc_timer.expires  = now + 1;
1453         add_timer(&tbl->gc_timer);
1454
1455         setup_timer(&tbl->proxy_timer, neigh_proxy_process, (unsigned long)tbl);
1456         skb_queue_head_init_class(&tbl->proxy_queue,
1457                         &neigh_table_proxy_queue_class);
1458
1459         tbl->last_flush = now;
1460         tbl->last_rand  = now + tbl->parms.reachable_time * 20;
1461 }
1462 EXPORT_SYMBOL(neigh_table_init_no_netlink);
1463
1464 void neigh_table_init(struct neigh_table *tbl)
1465 {
1466         struct neigh_table *tmp;
1467
1468         neigh_table_init_no_netlink(tbl);
1469         write_lock(&neigh_tbl_lock);
1470         for (tmp = neigh_tables; tmp; tmp = tmp->next) {
1471                 if (tmp->family == tbl->family)
1472                         break;
1473         }
1474         tbl->next       = neigh_tables;
1475         neigh_tables    = tbl;
1476         write_unlock(&neigh_tbl_lock);
1477
1478         if (unlikely(tmp)) {
1479                 printk(KERN_ERR "NEIGH: Registering multiple tables for "
1480                        "family %d\n", tbl->family);
1481                 dump_stack();
1482         }
1483 }
1484 EXPORT_SYMBOL(neigh_table_init);
1485
1486 int neigh_table_clear(struct neigh_table *tbl)
1487 {
1488         struct neigh_table **tp;
1489
1490         /* It is not clean... Fix it to unload IPv6 module safely */
1491         del_timer_sync(&tbl->gc_timer);
1492         del_timer_sync(&tbl->proxy_timer);
1493         pneigh_queue_purge(&tbl->proxy_queue);
1494         neigh_ifdown(tbl, NULL);
1495         if (atomic_read(&tbl->entries))
1496                 printk(KERN_CRIT "neighbour leakage\n");
1497         write_lock(&neigh_tbl_lock);
1498         for (tp = &neigh_tables; *tp; tp = &(*tp)->next) {
1499                 if (*tp == tbl) {
1500                         *tp = tbl->next;
1501                         break;
1502                 }
1503         }
1504         write_unlock(&neigh_tbl_lock);
1505
1506         neigh_hash_free(tbl->hash_buckets, tbl->hash_mask + 1);
1507         tbl->hash_buckets = NULL;
1508
1509         kfree(tbl->phash_buckets);
1510         tbl->phash_buckets = NULL;
1511
1512         remove_proc_entry(tbl->id, init_net.proc_net_stat);
1513
1514         free_percpu(tbl->stats);
1515         tbl->stats = NULL;
1516
1517         kmem_cache_destroy(tbl->kmem_cachep);
1518         tbl->kmem_cachep = NULL;
1519
1520         return 0;
1521 }
1522 EXPORT_SYMBOL(neigh_table_clear);
1523
1524 static int neigh_delete(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
1525 {
1526         struct net *net = sock_net(skb->sk);
1527         struct ndmsg *ndm;
1528         struct nlattr *dst_attr;
1529         struct neigh_table *tbl;
1530         struct net_device *dev = NULL;
1531         int err = -EINVAL;
1532
1533         if (nlmsg_len(nlh) < sizeof(*ndm))
1534                 goto out;
1535
1536         dst_attr = nlmsg_find_attr(nlh, sizeof(*ndm), NDA_DST);
1537         if (dst_attr == NULL)
1538                 goto out;
1539
1540         ndm = nlmsg_data(nlh);
1541         if (ndm->ndm_ifindex) {
1542                 dev = dev_get_by_index(net, ndm->ndm_ifindex);
1543                 if (dev == NULL) {
1544                         err = -ENODEV;
1545                         goto out;
1546                 }
1547         }
1548
1549         read_lock(&neigh_tbl_lock);
1550         for (tbl = neigh_tables; tbl; tbl = tbl->next) {
1551                 struct neighbour *neigh;
1552
1553                 if (tbl->family != ndm->ndm_family)
1554                         continue;
1555                 read_unlock(&neigh_tbl_lock);
1556
1557                 if (nla_len(dst_attr) < tbl->key_len)
1558                         goto out_dev_put;
1559
1560                 if (ndm->ndm_flags & NTF_PROXY) {
1561                         err = pneigh_delete(tbl, net, nla_data(dst_attr), dev);
1562                         goto out_dev_put;
1563                 }
1564
1565                 if (dev == NULL)
1566                         goto out_dev_put;
1567
1568                 neigh = neigh_lookup(tbl, nla_data(dst_attr), dev);
1569                 if (neigh == NULL) {
1570                         err = -ENOENT;
1571                         goto out_dev_put;
1572                 }
1573
1574                 err = neigh_update(neigh, NULL, NUD_FAILED,
1575                                    NEIGH_UPDATE_F_OVERRIDE |
1576                                    NEIGH_UPDATE_F_ADMIN);
1577                 neigh_release(neigh);
1578                 goto out_dev_put;
1579         }
1580         read_unlock(&neigh_tbl_lock);
1581         err = -EAFNOSUPPORT;
1582
1583 out_dev_put:
1584         if (dev)
1585                 dev_put(dev);
1586 out:
1587         return err;
1588 }
1589
1590 static int neigh_add(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
1591 {
1592         struct net *net = sock_net(skb->sk);
1593         struct ndmsg *ndm;
1594         struct nlattr *tb[NDA_MAX+1];
1595         struct neigh_table *tbl;
1596         struct net_device *dev = NULL;
1597         int err;
1598
1599         err = nlmsg_parse(nlh, sizeof(*ndm), tb, NDA_MAX, NULL);
1600         if (err < 0)
1601                 goto out;
1602
1603         err = -EINVAL;
1604         if (tb[NDA_DST] == NULL)
1605                 goto out;
1606
1607         ndm = nlmsg_data(nlh);
1608         if (ndm->ndm_ifindex) {
1609                 dev = dev_get_by_index(net, ndm->ndm_ifindex);
1610                 if (dev == NULL) {
1611                         err = -ENODEV;
1612                         goto out;
1613                 }
1614
1615                 if (tb[NDA_LLADDR] && nla_len(tb[NDA_LLADDR]) < dev->addr_len)
1616                         goto out_dev_put;
1617         }
1618
1619         read_lock(&neigh_tbl_lock);
1620         for (tbl = neigh_tables; tbl; tbl = tbl->next) {
1621                 int flags = NEIGH_UPDATE_F_ADMIN | NEIGH_UPDATE_F_OVERRIDE;
1622                 struct neighbour *neigh;
1623                 void *dst, *lladdr;
1624
1625                 if (tbl->family != ndm->ndm_family)
1626                         continue;
1627                 read_unlock(&neigh_tbl_lock);
1628
1629                 if (nla_len(tb[NDA_DST]) < tbl->key_len)
1630                         goto out_dev_put;
1631                 dst = nla_data(tb[NDA_DST]);
1632                 lladdr = tb[NDA_LLADDR] ? nla_data(tb[NDA_LLADDR]) : NULL;
1633
1634                 if (ndm->ndm_flags & NTF_PROXY) {
1635                         struct pneigh_entry *pn;
1636
1637                         err = -ENOBUFS;
1638                         pn = pneigh_lookup(tbl, net, dst, dev, 1);
1639                         if (pn) {
1640                                 pn->flags = ndm->ndm_flags;
1641                                 err = 0;
1642                         }
1643                         goto out_dev_put;
1644                 }
1645
1646                 if (dev == NULL)
1647                         goto out_dev_put;
1648
1649                 neigh = neigh_lookup(tbl, dst, dev);
1650                 if (neigh == NULL) {
1651                         if (!(nlh->nlmsg_flags & NLM_F_CREATE)) {
1652                                 err = -ENOENT;
1653                                 goto out_dev_put;
1654                         }
1655
1656                         neigh = __neigh_lookup_errno(tbl, dst, dev);
1657                         if (IS_ERR(neigh)) {
1658                                 err = PTR_ERR(neigh);
1659                                 goto out_dev_put;
1660                         }
1661                 } else {
1662                         if (nlh->nlmsg_flags & NLM_F_EXCL) {
1663                                 err = -EEXIST;
1664                                 neigh_release(neigh);
1665                                 goto out_dev_put;
1666                         }
1667
1668                         if (!(nlh->nlmsg_flags & NLM_F_REPLACE))
1669                                 flags &= ~NEIGH_UPDATE_F_OVERRIDE;
1670                 }
1671
1672                 err = neigh_update(neigh, lladdr, ndm->ndm_state, flags);
1673                 neigh_release(neigh);
1674                 goto out_dev_put;
1675         }
1676
1677         read_unlock(&neigh_tbl_lock);
1678         err = -EAFNOSUPPORT;
1679
1680 out_dev_put:
1681         if (dev)
1682                 dev_put(dev);
1683 out:
1684         return err;
1685 }
1686
1687 static int neightbl_fill_parms(struct sk_buff *skb, struct neigh_parms *parms)
1688 {
1689         struct nlattr *nest;
1690
1691         nest = nla_nest_start(skb, NDTA_PARMS);
1692         if (nest == NULL)
1693                 return -ENOBUFS;
1694
1695         if (parms->dev)
1696                 NLA_PUT_U32(skb, NDTPA_IFINDEX, parms->dev->ifindex);
1697
1698         NLA_PUT_U32(skb, NDTPA_REFCNT, atomic_read(&parms->refcnt));
1699         NLA_PUT_U32(skb, NDTPA_QUEUE_LEN, parms->queue_len);
1700         NLA_PUT_U32(skb, NDTPA_PROXY_QLEN, parms->proxy_qlen);
1701         NLA_PUT_U32(skb, NDTPA_APP_PROBES, parms->app_probes);
1702         NLA_PUT_U32(skb, NDTPA_UCAST_PROBES, parms->ucast_probes);
1703         NLA_PUT_U32(skb, NDTPA_MCAST_PROBES, parms->mcast_probes);
1704         NLA_PUT_MSECS(skb, NDTPA_REACHABLE_TIME, parms->reachable_time);
1705         NLA_PUT_MSECS(skb, NDTPA_BASE_REACHABLE_TIME,
1706                       parms->base_reachable_time);
1707         NLA_PUT_MSECS(skb, NDTPA_GC_STALETIME, parms->gc_staletime);
1708         NLA_PUT_MSECS(skb, NDTPA_DELAY_PROBE_TIME, parms->delay_probe_time);
1709         NLA_PUT_MSECS(skb, NDTPA_RETRANS_TIME, parms->retrans_time);
1710         NLA_PUT_MSECS(skb, NDTPA_ANYCAST_DELAY, parms->anycast_delay);
1711         NLA_PUT_MSECS(skb, NDTPA_PROXY_DELAY, parms->proxy_delay);
1712         NLA_PUT_MSECS(skb, NDTPA_LOCKTIME, parms->locktime);
1713
1714         return nla_nest_end(skb, nest);
1715
1716 nla_put_failure:
1717         nla_nest_cancel(skb, nest);
1718         return -EMSGSIZE;
1719 }
1720
1721 static int neightbl_fill_info(struct sk_buff *skb, struct neigh_table *tbl,
1722                               u32 pid, u32 seq, int type, int flags)
1723 {
1724         struct nlmsghdr *nlh;
1725         struct ndtmsg *ndtmsg;
1726
1727         nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
1728         if (nlh == NULL)
1729                 return -EMSGSIZE;
1730
1731         ndtmsg = nlmsg_data(nlh);
1732
1733         read_lock_bh(&tbl->lock);
1734         ndtmsg->ndtm_family = tbl->family;
1735         ndtmsg->ndtm_pad1   = 0;
1736         ndtmsg->ndtm_pad2   = 0;
1737
1738         NLA_PUT_STRING(skb, NDTA_NAME, tbl->id);
1739         NLA_PUT_MSECS(skb, NDTA_GC_INTERVAL, tbl->gc_interval);
1740         NLA_PUT_U32(skb, NDTA_THRESH1, tbl->gc_thresh1);
1741         NLA_PUT_U32(skb, NDTA_THRESH2, tbl->gc_thresh2);
1742         NLA_PUT_U32(skb, NDTA_THRESH3, tbl->gc_thresh3);
1743
1744         {
1745                 unsigned long now = jiffies;
1746                 unsigned int flush_delta = now - tbl->last_flush;
1747                 unsigned int rand_delta = now - tbl->last_rand;
1748
1749                 struct ndt_config ndc = {
1750                         .ndtc_key_len           = tbl->key_len,
1751                         .ndtc_entry_size        = tbl->entry_size,
1752                         .ndtc_entries           = atomic_read(&tbl->entries),
1753                         .ndtc_last_flush        = jiffies_to_msecs(flush_delta),
1754                         .ndtc_last_rand         = jiffies_to_msecs(rand_delta),
1755                         .ndtc_hash_rnd          = tbl->hash_rnd,
1756                         .ndtc_hash_mask         = tbl->hash_mask,
1757                         .ndtc_hash_chain_gc     = tbl->hash_chain_gc,
1758                         .ndtc_proxy_qlen        = tbl->proxy_queue.qlen,
1759                 };
1760
1761                 NLA_PUT(skb, NDTA_CONFIG, sizeof(ndc), &ndc);
1762         }
1763
1764         {
1765                 int cpu;
1766                 struct ndt_stats ndst;
1767
1768                 memset(&ndst, 0, sizeof(ndst));
1769
1770                 for_each_possible_cpu(cpu) {
1771                         struct neigh_statistics *st;
1772
1773                         st = per_cpu_ptr(tbl->stats, cpu);
1774                         ndst.ndts_allocs                += st->allocs;
1775                         ndst.ndts_destroys              += st->destroys;
1776                         ndst.ndts_hash_grows            += st->hash_grows;
1777                         ndst.ndts_res_failed            += st->res_failed;
1778                         ndst.ndts_lookups               += st->lookups;
1779                         ndst.ndts_hits                  += st->hits;
1780                         ndst.ndts_rcv_probes_mcast      += st->rcv_probes_mcast;
1781                         ndst.ndts_rcv_probes_ucast      += st->rcv_probes_ucast;
1782                         ndst.ndts_periodic_gc_runs      += st->periodic_gc_runs;
1783                         ndst.ndts_forced_gc_runs        += st->forced_gc_runs;
1784                 }
1785
1786                 NLA_PUT(skb, NDTA_STATS, sizeof(ndst), &ndst);
1787         }
1788
1789         BUG_ON(tbl->parms.dev);
1790         if (neightbl_fill_parms(skb, &tbl->parms) < 0)
1791                 goto nla_put_failure;
1792
1793         read_unlock_bh(&tbl->lock);
1794         return nlmsg_end(skb, nlh);
1795
1796 nla_put_failure:
1797         read_unlock_bh(&tbl->lock);
1798         nlmsg_cancel(skb, nlh);
1799         return -EMSGSIZE;
1800 }
1801
1802 static int neightbl_fill_param_info(struct sk_buff *skb,
1803                                     struct neigh_table *tbl,
1804                                     struct neigh_parms *parms,
1805                                     u32 pid, u32 seq, int type,
1806                                     unsigned int flags)
1807 {
1808         struct ndtmsg *ndtmsg;
1809         struct nlmsghdr *nlh;
1810
1811         nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
1812         if (nlh == NULL)
1813                 return -EMSGSIZE;
1814
1815         ndtmsg = nlmsg_data(nlh);
1816
1817         read_lock_bh(&tbl->lock);
1818         ndtmsg->ndtm_family = tbl->family;
1819         ndtmsg->ndtm_pad1   = 0;
1820         ndtmsg->ndtm_pad2   = 0;
1821
1822         if (nla_put_string(skb, NDTA_NAME, tbl->id) < 0 ||
1823             neightbl_fill_parms(skb, parms) < 0)
1824                 goto errout;
1825
1826         read_unlock_bh(&tbl->lock);
1827         return nlmsg_end(skb, nlh);
1828 errout:
1829         read_unlock_bh(&tbl->lock);
1830         nlmsg_cancel(skb, nlh);
1831         return -EMSGSIZE;
1832 }
1833
1834 static const struct nla_policy nl_neightbl_policy[NDTA_MAX+1] = {
1835         [NDTA_NAME]             = { .type = NLA_STRING },
1836         [NDTA_THRESH1]          = { .type = NLA_U32 },
1837         [NDTA_THRESH2]          = { .type = NLA_U32 },
1838         [NDTA_THRESH3]          = { .type = NLA_U32 },
1839         [NDTA_GC_INTERVAL]      = { .type = NLA_U64 },
1840         [NDTA_PARMS]            = { .type = NLA_NESTED },
1841 };
1842
1843 static const struct nla_policy nl_ntbl_parm_policy[NDTPA_MAX+1] = {
1844         [NDTPA_IFINDEX]                 = { .type = NLA_U32 },
1845         [NDTPA_QUEUE_LEN]               = { .type = NLA_U32 },
1846         [NDTPA_PROXY_QLEN]              = { .type = NLA_U32 },
1847         [NDTPA_APP_PROBES]              = { .type = NLA_U32 },
1848         [NDTPA_UCAST_PROBES]            = { .type = NLA_U32 },
1849         [NDTPA_MCAST_PROBES]            = { .type = NLA_U32 },
1850         [NDTPA_BASE_REACHABLE_TIME]     = { .type = NLA_U64 },
1851         [NDTPA_GC_STALETIME]            = { .type = NLA_U64 },
1852         [NDTPA_DELAY_PROBE_TIME]        = { .type = NLA_U64 },
1853         [NDTPA_RETRANS_TIME]            = { .type = NLA_U64 },
1854         [NDTPA_ANYCAST_DELAY]           = { .type = NLA_U64 },
1855         [NDTPA_PROXY_DELAY]             = { .type = NLA_U64 },
1856         [NDTPA_LOCKTIME]                = { .type = NLA_U64 },
1857 };
1858
1859 static int neightbl_set(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
1860 {
1861         struct net *net = sock_net(skb->sk);
1862         struct neigh_table *tbl;
1863         struct ndtmsg *ndtmsg;
1864         struct nlattr *tb[NDTA_MAX+1];
1865         int err;
1866
1867         err = nlmsg_parse(nlh, sizeof(*ndtmsg), tb, NDTA_MAX,
1868                           nl_neightbl_policy);
1869         if (err < 0)
1870                 goto errout;
1871
1872         if (tb[NDTA_NAME] == NULL) {
1873                 err = -EINVAL;
1874                 goto errout;
1875         }
1876
1877         ndtmsg = nlmsg_data(nlh);
1878         read_lock(&neigh_tbl_lock);
1879         for (tbl = neigh_tables; tbl; tbl = tbl->next) {
1880                 if (ndtmsg->ndtm_family && tbl->family != ndtmsg->ndtm_family)
1881                         continue;
1882
1883                 if (nla_strcmp(tb[NDTA_NAME], tbl->id) == 0)
1884                         break;
1885         }
1886
1887         if (tbl == NULL) {
1888                 err = -ENOENT;
1889                 goto errout_locked;
1890         }
1891
1892         /*
1893          * We acquire tbl->lock to be nice to the periodic timers and
1894          * make sure they always see a consistent set of values.
1895          */
1896         write_lock_bh(&tbl->lock);
1897
1898         if (tb[NDTA_PARMS]) {
1899                 struct nlattr *tbp[NDTPA_MAX+1];
1900                 struct neigh_parms *p;
1901                 int i, ifindex = 0;
1902
1903                 err = nla_parse_nested(tbp, NDTPA_MAX, tb[NDTA_PARMS],
1904                                        nl_ntbl_parm_policy);
1905                 if (err < 0)
1906                         goto errout_tbl_lock;
1907
1908                 if (tbp[NDTPA_IFINDEX])
1909                         ifindex = nla_get_u32(tbp[NDTPA_IFINDEX]);
1910
1911                 p = lookup_neigh_params(tbl, net, ifindex);
1912                 if (p == NULL) {
1913                         err = -ENOENT;
1914                         goto errout_tbl_lock;
1915                 }
1916
1917                 for (i = 1; i <= NDTPA_MAX; i++) {
1918                         if (tbp[i] == NULL)
1919                                 continue;
1920
1921                         switch (i) {
1922                         case NDTPA_QUEUE_LEN:
1923                                 p->queue_len = nla_get_u32(tbp[i]);
1924                                 break;
1925                         case NDTPA_PROXY_QLEN:
1926                                 p->proxy_qlen = nla_get_u32(tbp[i]);
1927                                 break;
1928                         case NDTPA_APP_PROBES:
1929                                 p->app_probes = nla_get_u32(tbp[i]);
1930                                 break;
1931                         case NDTPA_UCAST_PROBES:
1932                                 p->ucast_probes = nla_get_u32(tbp[i]);
1933                                 break;
1934                         case NDTPA_MCAST_PROBES:
1935                                 p->mcast_probes = nla_get_u32(tbp[i]);
1936                                 break;
1937                         case NDTPA_BASE_REACHABLE_TIME:
1938                                 p->base_reachable_time = nla_get_msecs(tbp[i]);
1939                                 break;
1940                         case NDTPA_GC_STALETIME:
1941                                 p->gc_staletime = nla_get_msecs(tbp[i]);
1942                                 break;
1943                         case NDTPA_DELAY_PROBE_TIME:
1944                                 p->delay_probe_time = nla_get_msecs(tbp[i]);
1945                                 break;
1946                         case NDTPA_RETRANS_TIME:
1947                                 p->retrans_time = nla_get_msecs(tbp[i]);
1948                                 break;
1949                         case NDTPA_ANYCAST_DELAY:
1950                                 p->anycast_delay = nla_get_msecs(tbp[i]);
1951                                 break;
1952                         case NDTPA_PROXY_DELAY:
1953                                 p->proxy_delay = nla_get_msecs(tbp[i]);
1954                                 break;
1955                         case NDTPA_LOCKTIME:
1956                                 p->locktime = nla_get_msecs(tbp[i]);
1957                                 break;
1958                         }
1959                 }
1960         }
1961
1962         if (tb[NDTA_THRESH1])
1963                 tbl->gc_thresh1 = nla_get_u32(tb[NDTA_THRESH1]);
1964
1965         if (tb[NDTA_THRESH2])
1966                 tbl->gc_thresh2 = nla_get_u32(tb[NDTA_THRESH2]);
1967
1968         if (tb[NDTA_THRESH3])
1969                 tbl->gc_thresh3 = nla_get_u32(tb[NDTA_THRESH3]);
1970
1971         if (tb[NDTA_GC_INTERVAL])
1972                 tbl->gc_interval = nla_get_msecs(tb[NDTA_GC_INTERVAL]);
1973
1974         err = 0;
1975
1976 errout_tbl_lock:
1977         write_unlock_bh(&tbl->lock);
1978 errout_locked:
1979         read_unlock(&neigh_tbl_lock);
1980 errout:
1981         return err;
1982 }
1983
1984 static int neightbl_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
1985 {
1986         struct net *net = sock_net(skb->sk);
1987         int family, tidx, nidx = 0;
1988         int tbl_skip = cb->args[0];
1989         int neigh_skip = cb->args[1];
1990         struct neigh_table *tbl;
1991
1992         family = ((struct rtgenmsg *) nlmsg_data(cb->nlh))->rtgen_family;
1993
1994         read_lock(&neigh_tbl_lock);
1995         for (tbl = neigh_tables, tidx = 0; tbl; tbl = tbl->next, tidx++) {
1996                 struct neigh_parms *p;
1997
1998                 if (tidx < tbl_skip || (family && tbl->family != family))
1999                         continue;
2000
2001                 if (neightbl_fill_info(skb, tbl, NETLINK_CB(cb->skb).pid,
2002                                        cb->nlh->nlmsg_seq, RTM_NEWNEIGHTBL,
2003                                        NLM_F_MULTI) <= 0)
2004                         break;
2005
2006                 for (nidx = 0, p = tbl->parms.next; p; p = p->next) {
2007                         if (!net_eq(neigh_parms_net(p), net))
2008                                 continue;
2009
2010                         if (nidx++ < neigh_skip)
2011                                 continue;
2012
2013                         if (neightbl_fill_param_info(skb, tbl, p,
2014                                                      NETLINK_CB(cb->skb).pid,
2015                                                      cb->nlh->nlmsg_seq,
2016                                                      RTM_NEWNEIGHTBL,
2017                                                      NLM_F_MULTI) <= 0)
2018                                 goto out;
2019                 }
2020
2021                 neigh_skip = 0;
2022         }
2023 out:
2024         read_unlock(&neigh_tbl_lock);
2025         cb->args[0] = tidx;
2026         cb->args[1] = nidx;
2027
2028         return skb->len;
2029 }
2030
2031 static int neigh_fill_info(struct sk_buff *skb, struct neighbour *neigh,
2032                            u32 pid, u32 seq, int type, unsigned int flags)
2033 {
2034         unsigned long now = jiffies;
2035         struct nda_cacheinfo ci;
2036         struct nlmsghdr *nlh;
2037         struct ndmsg *ndm;
2038
2039         nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2040         if (nlh == NULL)
2041                 return -EMSGSIZE;
2042
2043         ndm = nlmsg_data(nlh);
2044         ndm->ndm_family  = neigh->ops->family;
2045         ndm->ndm_pad1    = 0;
2046         ndm->ndm_pad2    = 0;
2047         ndm->ndm_flags   = neigh->flags;
2048         ndm->ndm_type    = neigh->type;
2049         ndm->ndm_ifindex = neigh->dev->ifindex;
2050
2051         NLA_PUT(skb, NDA_DST, neigh->tbl->key_len, neigh->primary_key);
2052
2053         read_lock_bh(&neigh->lock);
2054         ndm->ndm_state   = neigh->nud_state;
2055         if ((neigh->nud_state & NUD_VALID) &&
2056             nla_put(skb, NDA_LLADDR, neigh->dev->addr_len, neigh->ha) < 0) {
2057                 read_unlock_bh(&neigh->lock);
2058                 goto nla_put_failure;
2059         }
2060
2061         ci.ndm_used      = jiffies_to_clock_t(now - neigh->used);
2062         ci.ndm_confirmed = jiffies_to_clock_t(now - neigh->confirmed);
2063         ci.ndm_updated   = jiffies_to_clock_t(now - neigh->updated);
2064         ci.ndm_refcnt    = atomic_read(&neigh->refcnt) - 1;
2065         read_unlock_bh(&neigh->lock);
2066
2067         NLA_PUT_U32(skb, NDA_PROBES, atomic_read(&neigh->probes));
2068         NLA_PUT(skb, NDA_CACHEINFO, sizeof(ci), &ci);
2069
2070         return nlmsg_end(skb, nlh);
2071
2072 nla_put_failure:
2073         nlmsg_cancel(skb, nlh);
2074         return -EMSGSIZE;
2075 }
2076
2077 static void neigh_update_notify(struct neighbour *neigh)
2078 {
2079         call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, neigh);
2080         __neigh_notify(neigh, RTM_NEWNEIGH, 0);
2081 }
2082
2083 static int neigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2084                             struct netlink_callback *cb)
2085 {
2086         struct net * net = sock_net(skb->sk);
2087         struct neighbour *n;
2088         int rc, h, s_h = cb->args[1];
2089         int idx, s_idx = idx = cb->args[2];
2090
2091         read_lock_bh(&tbl->lock);
2092         for (h = 0; h <= tbl->hash_mask; h++) {
2093                 if (h < s_h)
2094                         continue;
2095                 if (h > s_h)
2096                         s_idx = 0;
2097                 for (n = tbl->hash_buckets[h], idx = 0; n; n = n->next) {
2098                         int lidx;
2099                         if (dev_net(n->dev) != net)
2100                                 continue;
2101                         lidx = idx++;
2102                         if (lidx < s_idx)
2103                                 continue;
2104                         if (neigh_fill_info(skb, n, NETLINK_CB(cb->skb).pid,
2105                                             cb->nlh->nlmsg_seq,
2106                                             RTM_NEWNEIGH,
2107                                             NLM_F_MULTI) <= 0) {
2108                                 read_unlock_bh(&tbl->lock);
2109                                 rc = -1;
2110                                 goto out;
2111                         }
2112                 }
2113         }
2114         read_unlock_bh(&tbl->lock);
2115         rc = skb->len;
2116 out:
2117         cb->args[1] = h;
2118         cb->args[2] = idx;
2119         return rc;
2120 }
2121
2122 static int neigh_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2123 {
2124         struct neigh_table *tbl;
2125         int t, family, s_t;
2126
2127         read_lock(&neigh_tbl_lock);
2128         family = ((struct rtgenmsg *) nlmsg_data(cb->nlh))->rtgen_family;
2129         s_t = cb->args[0];
2130
2131         for (tbl = neigh_tables, t = 0; tbl; tbl = tbl->next, t++) {
2132                 if (t < s_t || (family && tbl->family != family))
2133                         continue;
2134                 if (t > s_t)
2135                         memset(&cb->args[1], 0, sizeof(cb->args) -
2136                                                 sizeof(cb->args[0]));
2137                 if (neigh_dump_table(tbl, skb, cb) < 0)
2138                         break;
2139         }
2140         read_unlock(&neigh_tbl_lock);
2141
2142         cb->args[0] = t;
2143         return skb->len;
2144 }
2145
2146 void neigh_for_each(struct neigh_table *tbl, void (*cb)(struct neighbour *, void *), void *cookie)
2147 {
2148         int chain;
2149
2150         read_lock_bh(&tbl->lock);
2151         for (chain = 0; chain <= tbl->hash_mask; chain++) {
2152                 struct neighbour *n;
2153
2154                 for (n = tbl->hash_buckets[chain]; n; n = n->next)
2155                         cb(n, cookie);
2156         }
2157         read_unlock_bh(&tbl->lock);
2158 }
2159 EXPORT_SYMBOL(neigh_for_each);
2160
2161 /* The tbl->lock must be held as a writer and BH disabled. */
2162 void __neigh_for_each_release(struct neigh_table *tbl,
2163                               int (*cb)(struct neighbour *))
2164 {
2165         int chain;
2166
2167         for (chain = 0; chain <= tbl->hash_mask; chain++) {
2168                 struct neighbour *n, **np;
2169
2170                 np = &tbl->hash_buckets[chain];
2171                 while ((n = *np) != NULL) {
2172                         int release;
2173
2174                         write_lock(&n->lock);
2175                         release = cb(n);
2176                         if (release) {
2177                                 *np = n->next;
2178                                 n->dead = 1;
2179                         } else
2180                                 np = &n->next;
2181                         write_unlock(&n->lock);
2182                         if (release)
2183                                 neigh_cleanup_and_release(n);
2184                 }
2185         }
2186 }
2187 EXPORT_SYMBOL(__neigh_for_each_release);
2188
2189 #ifdef CONFIG_PROC_FS
2190
2191 static struct neighbour *neigh_get_first(struct seq_file *seq)
2192 {
2193         struct neigh_seq_state *state = seq->private;
2194         struct net *net = seq_file_net(seq);
2195         struct neigh_table *tbl = state->tbl;
2196         struct neighbour *n = NULL;
2197         int bucket = state->bucket;
2198
2199         state->flags &= ~NEIGH_SEQ_IS_PNEIGH;
2200         for (bucket = 0; bucket <= tbl->hash_mask; bucket++) {
2201                 n = tbl->hash_buckets[bucket];
2202
2203                 while (n) {
2204                         if (!net_eq(dev_net(n->dev), net))
2205                                 goto next;
2206                         if (state->neigh_sub_iter) {
2207                                 loff_t fakep = 0;
2208                                 void *v;
2209
2210                                 v = state->neigh_sub_iter(state, n, &fakep);
2211                                 if (!v)
2212                                         goto next;
2213                         }
2214                         if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
2215                                 break;
2216                         if (n->nud_state & ~NUD_NOARP)
2217                                 break;
2218                 next:
2219                         n = n->next;
2220                 }
2221
2222                 if (n)
2223                         break;
2224         }
2225         state->bucket = bucket;
2226
2227         return n;
2228 }
2229
2230 static struct neighbour *neigh_get_next(struct seq_file *seq,
2231                                         struct neighbour *n,
2232                                         loff_t *pos)
2233 {
2234         struct neigh_seq_state *state = seq->private;
2235         struct net *net = seq_file_net(seq);
2236         struct neigh_table *tbl = state->tbl;
2237
2238         if (state->neigh_sub_iter) {
2239                 void *v = state->neigh_sub_iter(state, n, pos);
2240                 if (v)
2241                         return n;
2242         }
2243         n = n->next;
2244
2245         while (1) {
2246                 while (n) {
2247                         if (!net_eq(dev_net(n->dev), net))
2248                                 goto next;
2249                         if (state->neigh_sub_iter) {
2250                                 void *v = state->neigh_sub_iter(state, n, pos);
2251                                 if (v)
2252                                         return n;
2253                                 goto next;
2254                         }
2255                         if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
2256                                 break;
2257
2258                         if (n->nud_state & ~NUD_NOARP)
2259                                 break;
2260                 next:
2261                         n = n->next;
2262                 }
2263
2264                 if (n)
2265                         break;
2266
2267                 if (++state->bucket > tbl->hash_mask)
2268                         break;
2269
2270                 n = tbl->hash_buckets[state->bucket];
2271         }
2272
2273         if (n && pos)
2274                 --(*pos);
2275         return n;
2276 }
2277
2278 static struct neighbour *neigh_get_idx(struct seq_file *seq, loff_t *pos)
2279 {
2280         struct neighbour *n = neigh_get_first(seq);
2281
2282         if (n) {
2283                 while (*pos) {
2284                         n = neigh_get_next(seq, n, pos);
2285                         if (!n)
2286                                 break;
2287                 }
2288         }
2289         return *pos ? NULL : n;
2290 }
2291
2292 static struct pneigh_entry *pneigh_get_first(struct seq_file *seq)
2293 {
2294         struct neigh_seq_state *state = seq->private;
2295         struct net *net = seq_file_net(seq);
2296         struct neigh_table *tbl = state->tbl;
2297         struct pneigh_entry *pn = NULL;
2298         int bucket = state->bucket;
2299
2300         state->flags |= NEIGH_SEQ_IS_PNEIGH;
2301         for (bucket = 0; bucket <= PNEIGH_HASHMASK; bucket++) {
2302                 pn = tbl->phash_buckets[bucket];
2303                 while (pn && !net_eq(pneigh_net(pn), net))
2304                         pn = pn->next;
2305                 if (pn)
2306                         break;
2307         }
2308         state->bucket = bucket;
2309
2310         return pn;
2311 }
2312
2313 static struct pneigh_entry *pneigh_get_next(struct seq_file *seq,
2314                                             struct pneigh_entry *pn,
2315                                             loff_t *pos)
2316 {
2317         struct neigh_seq_state *state = seq->private;
2318         struct net *net = seq_file_net(seq);
2319         struct neigh_table *tbl = state->tbl;
2320
2321         pn = pn->next;
2322         while (!pn) {
2323                 if (++state->bucket > PNEIGH_HASHMASK)
2324                         break;
2325                 pn = tbl->phash_buckets[state->bucket];
2326                 while (pn && !net_eq(pneigh_net(pn), net))
2327                         pn = pn->next;
2328                 if (pn)
2329                         break;
2330         }
2331
2332         if (pn && pos)
2333                 --(*pos);
2334
2335         return pn;
2336 }
2337
2338 static struct pneigh_entry *pneigh_get_idx(struct seq_file *seq, loff_t *pos)
2339 {
2340         struct pneigh_entry *pn = pneigh_get_first(seq);
2341
2342         if (pn) {
2343                 while (*pos) {
2344                         pn = pneigh_get_next(seq, pn, pos);
2345                         if (!pn)
2346                                 break;
2347                 }
2348         }
2349         return *pos ? NULL : pn;
2350 }
2351
2352 static void *neigh_get_idx_any(struct seq_file *seq, loff_t *pos)
2353 {
2354         struct neigh_seq_state *state = seq->private;
2355         void *rc;
2356
2357         rc = neigh_get_idx(seq, pos);
2358         if (!rc && !(state->flags & NEIGH_SEQ_NEIGH_ONLY))
2359                 rc = pneigh_get_idx(seq, pos);
2360
2361         return rc;
2362 }
2363
2364 void *neigh_seq_start(struct seq_file *seq, loff_t *pos, struct neigh_table *tbl, unsigned int neigh_seq_flags)
2365         __acquires(tbl->lock)
2366 {
2367         struct neigh_seq_state *state = seq->private;
2368         loff_t pos_minus_one;
2369
2370         state->tbl = tbl;
2371         state->bucket = 0;
2372         state->flags = (neigh_seq_flags & ~NEIGH_SEQ_IS_PNEIGH);
2373
2374         read_lock_bh(&tbl->lock);
2375
2376         pos_minus_one = *pos - 1;
2377         return *pos ? neigh_get_idx_any(seq, &pos_minus_one) : SEQ_START_TOKEN;
2378 }
2379 EXPORT_SYMBOL(neigh_seq_start);
2380
2381 void *neigh_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2382 {
2383         struct neigh_seq_state *state;
2384         void *rc;
2385
2386         if (v == SEQ_START_TOKEN) {
2387                 rc = neigh_get_idx(seq, pos);
2388                 goto out;
2389         }
2390
2391         state = seq->private;
2392         if (!(state->flags & NEIGH_SEQ_IS_PNEIGH)) {
2393                 rc = neigh_get_next(seq, v, NULL);
2394                 if (rc)
2395                         goto out;
2396                 if (!(state->flags & NEIGH_SEQ_NEIGH_ONLY))
2397                         rc = pneigh_get_first(seq);
2398         } else {
2399                 BUG_ON(state->flags & NEIGH_SEQ_NEIGH_ONLY);
2400                 rc = pneigh_get_next(seq, v, NULL);
2401         }
2402 out:
2403         ++(*pos);
2404         return rc;
2405 }
2406 EXPORT_SYMBOL(neigh_seq_next);
2407
2408 void neigh_seq_stop(struct seq_file *seq, void *v)
2409         __releases(tbl->lock)
2410 {
2411         struct neigh_seq_state *state = seq->private;
2412         struct neigh_table *tbl = state->tbl;
2413
2414         read_unlock_bh(&tbl->lock);
2415 }
2416 EXPORT_SYMBOL(neigh_seq_stop);
2417
2418 /* statistics via seq_file */
2419
2420 static void *neigh_stat_seq_start(struct seq_file *seq, loff_t *pos)
2421 {
2422         struct proc_dir_entry *pde = seq->private;
2423         struct neigh_table *tbl = pde->data;
2424         int cpu;
2425
2426         if (*pos == 0)
2427                 return SEQ_START_TOKEN;
2428
2429         for (cpu = *pos-1; cpu < NR_CPUS; ++cpu) {
2430                 if (!cpu_possible(cpu))
2431                         continue;
2432                 *pos = cpu+1;
2433                 return per_cpu_ptr(tbl->stats, cpu);
2434         }
2435         return NULL;
2436 }
2437
2438 static void *neigh_stat_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2439 {
2440         struct proc_dir_entry *pde = seq->private;
2441         struct neigh_table *tbl = pde->data;
2442         int cpu;
2443
2444         for (cpu = *pos; cpu < NR_CPUS; ++cpu) {
2445                 if (!cpu_possible(cpu))
2446                         continue;
2447                 *pos = cpu+1;
2448                 return per_cpu_ptr(tbl->stats, cpu);
2449         }
2450         return NULL;
2451 }
2452
2453 static void neigh_stat_seq_stop(struct seq_file *seq, void *v)
2454 {
2455
2456 }
2457
2458 static int neigh_stat_seq_show(struct seq_file *seq, void *v)
2459 {
2460         struct proc_dir_entry *pde = seq->private;
2461         struct neigh_table *tbl = pde->data;
2462         struct neigh_statistics *st = v;
2463
2464         if (v == SEQ_START_TOKEN) {
2465                 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");
2466                 return 0;
2467         }
2468
2469         seq_printf(seq, "%08x  %08lx %08lx %08lx  %08lx %08lx  %08lx  "
2470                         "%08lx %08lx  %08lx %08lx\n",
2471                    atomic_read(&tbl->entries),
2472
2473                    st->allocs,
2474                    st->destroys,
2475                    st->hash_grows,
2476
2477                    st->lookups,
2478                    st->hits,
2479
2480                    st->res_failed,
2481
2482                    st->rcv_probes_mcast,
2483                    st->rcv_probes_ucast,
2484
2485                    st->periodic_gc_runs,
2486                    st->forced_gc_runs
2487                    );
2488
2489         return 0;
2490 }
2491
2492 static const struct seq_operations neigh_stat_seq_ops = {
2493         .start  = neigh_stat_seq_start,
2494         .next   = neigh_stat_seq_next,
2495         .stop   = neigh_stat_seq_stop,
2496         .show   = neigh_stat_seq_show,
2497 };
2498
2499 static int neigh_stat_seq_open(struct inode *inode, struct file *file)
2500 {
2501         int ret = seq_open(file, &neigh_stat_seq_ops);
2502
2503         if (!ret) {
2504                 struct seq_file *sf = file->private_data;
2505                 sf->private = PDE(inode);
2506         }
2507         return ret;
2508 };
2509
2510 static const struct file_operations neigh_stat_seq_fops = {
2511         .owner   = THIS_MODULE,
2512         .open    = neigh_stat_seq_open,
2513         .read    = seq_read,
2514         .llseek  = seq_lseek,
2515         .release = seq_release,
2516 };
2517
2518 #endif /* CONFIG_PROC_FS */
2519
2520 static inline size_t neigh_nlmsg_size(void)
2521 {
2522         return NLMSG_ALIGN(sizeof(struct ndmsg))
2523                + nla_total_size(MAX_ADDR_LEN) /* NDA_DST */
2524                + nla_total_size(MAX_ADDR_LEN) /* NDA_LLADDR */
2525                + nla_total_size(sizeof(struct nda_cacheinfo))
2526                + nla_total_size(4); /* NDA_PROBES */
2527 }
2528
2529 static void __neigh_notify(struct neighbour *n, int type, int flags)
2530 {
2531         struct net *net = dev_net(n->dev);
2532         struct sk_buff *skb;
2533         int err = -ENOBUFS;
2534
2535         skb = nlmsg_new(neigh_nlmsg_size(), GFP_ATOMIC);
2536         if (skb == NULL)
2537                 goto errout;
2538
2539         err = neigh_fill_info(skb, n, 0, 0, type, flags);
2540         if (err < 0) {
2541                 /* -EMSGSIZE implies BUG in neigh_nlmsg_size() */
2542                 WARN_ON(err == -EMSGSIZE);
2543                 kfree_skb(skb);
2544                 goto errout;
2545         }
2546         err = rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
2547 errout:
2548         if (err < 0)
2549                 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
2550 }
2551
2552 #ifdef CONFIG_ARPD
2553 void neigh_app_ns(struct neighbour *n)
2554 {
2555         __neigh_notify(n, RTM_GETNEIGH, NLM_F_REQUEST);
2556 }
2557 EXPORT_SYMBOL(neigh_app_ns);
2558 #endif /* CONFIG_ARPD */
2559
2560 #ifdef CONFIG_SYSCTL
2561
2562 static struct neigh_sysctl_table {
2563         struct ctl_table_header *sysctl_header;
2564         struct ctl_table neigh_vars[__NET_NEIGH_MAX];
2565         char *dev_name;
2566 } neigh_sysctl_template __read_mostly = {
2567         .neigh_vars = {
2568                 {
2569                         .ctl_name       = NET_NEIGH_MCAST_SOLICIT,
2570                         .procname       = "mcast_solicit",
2571                         .maxlen         = sizeof(int),
2572                         .mode           = 0644,
2573                         .proc_handler   = &proc_dointvec,
2574                 },
2575                 {
2576                         .ctl_name       = NET_NEIGH_UCAST_SOLICIT,
2577                         .procname       = "ucast_solicit",
2578                         .maxlen         = sizeof(int),
2579                         .mode           = 0644,
2580                         .proc_handler   = &proc_dointvec,
2581                 },
2582                 {
2583                         .ctl_name       = NET_NEIGH_APP_SOLICIT,
2584                         .procname       = "app_solicit",
2585                         .maxlen         = sizeof(int),
2586                         .mode           = 0644,
2587                         .proc_handler   = &proc_dointvec,
2588                 },
2589                 {
2590                         .procname       = "retrans_time",
2591                         .maxlen         = sizeof(int),
2592                         .mode           = 0644,
2593                         .proc_handler   = &proc_dointvec_userhz_jiffies,
2594                 },
2595                 {
2596                         .ctl_name       = NET_NEIGH_REACHABLE_TIME,
2597                         .procname       = "base_reachable_time",
2598                         .maxlen         = sizeof(int),
2599                         .mode           = 0644,
2600                         .proc_handler   = &proc_dointvec_jiffies,
2601                         .strategy       = &sysctl_jiffies,
2602                 },
2603                 {
2604                         .ctl_name       = NET_NEIGH_DELAY_PROBE_TIME,
2605                         .procname       = "delay_first_probe_time",
2606                         .maxlen         = sizeof(int),
2607                         .mode           = 0644,
2608                         .proc_handler   = &proc_dointvec_jiffies,
2609                         .strategy       = &sysctl_jiffies,
2610                 },
2611                 {
2612                         .ctl_name       = NET_NEIGH_GC_STALE_TIME,
2613                         .procname       = "gc_stale_time",
2614                         .maxlen         = sizeof(int),
2615                         .mode           = 0644,
2616                         .proc_handler   = &proc_dointvec_jiffies,
2617                         .strategy       = &sysctl_jiffies,
2618                 },
2619                 {
2620                         .ctl_name       = NET_NEIGH_UNRES_QLEN,
2621                         .procname       = "unres_qlen",
2622                         .maxlen         = sizeof(int),
2623                         .mode           = 0644,
2624                         .proc_handler   = &proc_dointvec,
2625                 },
2626                 {
2627                         .ctl_name       = NET_NEIGH_PROXY_QLEN,
2628                         .procname       = "proxy_qlen",
2629                         .maxlen         = sizeof(int),
2630                         .mode           = 0644,
2631                         .proc_handler   = &proc_dointvec,
2632                 },
2633                 {
2634                         .procname       = "anycast_delay",
2635                         .maxlen         = sizeof(int),
2636                         .mode           = 0644,
2637                         .proc_handler   = &proc_dointvec_userhz_jiffies,
2638                 },
2639                 {
2640                         .procname       = "proxy_delay",
2641                         .maxlen         = sizeof(int),
2642                         .mode           = 0644,
2643                         .proc_handler   = &proc_dointvec_userhz_jiffies,
2644                 },
2645                 {
2646                         .procname       = "locktime",
2647                         .maxlen         = sizeof(int),
2648                         .mode           = 0644,
2649                         .proc_handler   = &proc_dointvec_userhz_jiffies,
2650                 },
2651                 {
2652                         .ctl_name       = NET_NEIGH_RETRANS_TIME_MS,
2653                         .procname       = "retrans_time_ms",
2654                         .maxlen         = sizeof(int),
2655                         .mode           = 0644,
2656                         .proc_handler   = &proc_dointvec_ms_jiffies,
2657                         .strategy       = &sysctl_ms_jiffies,
2658                 },
2659                 {
2660                         .ctl_name       = NET_NEIGH_REACHABLE_TIME_MS,
2661                         .procname       = "base_reachable_time_ms",
2662                         .maxlen         = sizeof(int),
2663                         .mode           = 0644,
2664                         .proc_handler   = &proc_dointvec_ms_jiffies,
2665                         .strategy       = &sysctl_ms_jiffies,
2666                 },
2667                 {
2668                         .ctl_name       = NET_NEIGH_GC_INTERVAL,
2669                         .procname       = "gc_interval",
2670                         .maxlen         = sizeof(int),
2671                         .mode           = 0644,
2672                         .proc_handler   = &proc_dointvec_jiffies,
2673                         .strategy       = &sysctl_jiffies,
2674                 },
2675                 {
2676                         .ctl_name       = NET_NEIGH_GC_THRESH1,
2677                         .procname       = "gc_thresh1",
2678                         .maxlen         = sizeof(int),
2679                         .mode           = 0644,
2680                         .proc_handler   = &proc_dointvec,
2681                 },
2682                 {
2683                         .ctl_name       = NET_NEIGH_GC_THRESH2,
2684                         .procname       = "gc_thresh2",
2685                         .maxlen         = sizeof(int),
2686                         .mode           = 0644,
2687                         .proc_handler   = &proc_dointvec,
2688                 },
2689                 {
2690                         .ctl_name       = NET_NEIGH_GC_THRESH3,
2691                         .procname       = "gc_thresh3",
2692                         .maxlen         = sizeof(int),
2693                         .mode           = 0644,
2694                         .proc_handler   = &proc_dointvec,
2695                 },
2696                 {},
2697         },
2698 };
2699
2700 int neigh_sysctl_register(struct net_device *dev, struct neigh_parms *p,
2701                           int p_id, int pdev_id, char *p_name,
2702                           proc_handler *handler, ctl_handler *strategy)
2703 {
2704         struct neigh_sysctl_table *t;
2705         const char *dev_name_source = NULL;
2706
2707 #define NEIGH_CTL_PATH_ROOT     0
2708 #define NEIGH_CTL_PATH_PROTO    1
2709 #define NEIGH_CTL_PATH_NEIGH    2
2710 #define NEIGH_CTL_PATH_DEV      3
2711
2712         struct ctl_path neigh_path[] = {
2713                 { .procname = "net",     .ctl_name = CTL_NET, },
2714                 { .procname = "proto",   .ctl_name = 0, },
2715                 { .procname = "neigh",   .ctl_name = 0, },
2716                 { .procname = "default", .ctl_name = NET_PROTO_CONF_DEFAULT, },
2717                 { },
2718         };
2719
2720         t = kmemdup(&neigh_sysctl_template, sizeof(*t), GFP_KERNEL);
2721         if (!t)
2722                 goto err;
2723
2724         t->neigh_vars[0].data  = &p->mcast_probes;
2725         t->neigh_vars[1].data  = &p->ucast_probes;
2726         t->neigh_vars[2].data  = &p->app_probes;
2727         t->neigh_vars[3].data  = &p->retrans_time;
2728         t->neigh_vars[4].data  = &p->base_reachable_time;
2729         t->neigh_vars[5].data  = &p->delay_probe_time;
2730         t->neigh_vars[6].data  = &p->gc_staletime;
2731         t->neigh_vars[7].data  = &p->queue_len;
2732         t->neigh_vars[8].data  = &p->proxy_qlen;
2733         t->neigh_vars[9].data  = &p->anycast_delay;
2734         t->neigh_vars[10].data = &p->proxy_delay;
2735         t->neigh_vars[11].data = &p->locktime;
2736         t->neigh_vars[12].data  = &p->retrans_time;
2737         t->neigh_vars[13].data  = &p->base_reachable_time;
2738
2739         if (dev) {
2740                 dev_name_source = dev->name;
2741                 neigh_path[NEIGH_CTL_PATH_DEV].ctl_name = dev->ifindex;
2742                 /* Terminate the table early */
2743                 memset(&t->neigh_vars[14], 0, sizeof(t->neigh_vars[14]));
2744         } else {
2745                 dev_name_source = neigh_path[NEIGH_CTL_PATH_DEV].procname;
2746                 t->neigh_vars[14].data = (int *)(p + 1);
2747                 t->neigh_vars[15].data = (int *)(p + 1) + 1;
2748                 t->neigh_vars[16].data = (int *)(p + 1) + 2;
2749                 t->neigh_vars[17].data = (int *)(p + 1) + 3;
2750         }
2751
2752
2753         if (handler || strategy) {
2754                 /* RetransTime */
2755                 t->neigh_vars[3].proc_handler = handler;
2756                 t->neigh_vars[3].strategy = strategy;
2757                 t->neigh_vars[3].extra1 = dev;
2758                 if (!strategy)
2759                         t->neigh_vars[3].ctl_name = CTL_UNNUMBERED;
2760                 /* ReachableTime */
2761                 t->neigh_vars[4].proc_handler = handler;
2762                 t->neigh_vars[4].strategy = strategy;
2763                 t->neigh_vars[4].extra1 = dev;
2764                 if (!strategy)
2765                         t->neigh_vars[4].ctl_name = CTL_UNNUMBERED;
2766                 /* RetransTime (in milliseconds)*/
2767                 t->neigh_vars[12].proc_handler = handler;
2768                 t->neigh_vars[12].strategy = strategy;
2769                 t->neigh_vars[12].extra1 = dev;
2770                 if (!strategy)
2771                         t->neigh_vars[12].ctl_name = CTL_UNNUMBERED;
2772                 /* ReachableTime (in milliseconds) */
2773                 t->neigh_vars[13].proc_handler = handler;
2774                 t->neigh_vars[13].strategy = strategy;
2775                 t->neigh_vars[13].extra1 = dev;
2776                 if (!strategy)
2777                         t->neigh_vars[13].ctl_name = CTL_UNNUMBERED;
2778         }
2779
2780         t->dev_name = kstrdup(dev_name_source, GFP_KERNEL);
2781         if (!t->dev_name)
2782                 goto free;
2783
2784         neigh_path[NEIGH_CTL_PATH_DEV].procname = t->dev_name;
2785         neigh_path[NEIGH_CTL_PATH_NEIGH].ctl_name = pdev_id;
2786         neigh_path[NEIGH_CTL_PATH_PROTO].procname = p_name;
2787         neigh_path[NEIGH_CTL_PATH_PROTO].ctl_name = p_id;
2788
2789         t->sysctl_header =
2790                 register_net_sysctl_table(neigh_parms_net(p), neigh_path, t->neigh_vars);
2791         if (!t->sysctl_header)
2792                 goto free_procname;
2793
2794         p->sysctl_table = t;
2795         return 0;
2796
2797 free_procname:
2798         kfree(t->dev_name);
2799 free:
2800         kfree(t);
2801 err:
2802         return -ENOBUFS;
2803 }
2804 EXPORT_SYMBOL(neigh_sysctl_register);
2805
2806 void neigh_sysctl_unregister(struct neigh_parms *p)
2807 {
2808         if (p->sysctl_table) {
2809                 struct neigh_sysctl_table *t = p->sysctl_table;
2810                 p->sysctl_table = NULL;
2811                 unregister_sysctl_table(t->sysctl_header);
2812                 kfree(t->dev_name);
2813                 kfree(t);
2814         }
2815 }
2816 EXPORT_SYMBOL(neigh_sysctl_unregister);
2817
2818 #endif  /* CONFIG_SYSCTL */
2819
2820 static int __init neigh_init(void)
2821 {
2822         rtnl_register(PF_UNSPEC, RTM_NEWNEIGH, neigh_add, NULL);
2823         rtnl_register(PF_UNSPEC, RTM_DELNEIGH, neigh_delete, NULL);
2824         rtnl_register(PF_UNSPEC, RTM_GETNEIGH, NULL, neigh_dump_info);
2825
2826         rtnl_register(PF_UNSPEC, RTM_GETNEIGHTBL, NULL, neightbl_dump_info);
2827         rtnl_register(PF_UNSPEC, RTM_SETNEIGHTBL, neightbl_set, NULL);
2828
2829         return 0;
2830 }
2831
2832 subsys_initcall(neigh_init);
2833