net/core/neighbour.c

   1 /*
   2  *      Generic address resolution entity
   3  *
   4  *      Authors:
   5  *      Pedro Roque             <roque@di.fc.ul.pt>
   6  *      Alexey Kuznetsov        <kuznet@ms2.inr.ac.ru>
   7  *
   8  *      This program is free software; you can redistribute it and/or
   9  *      modify it under the terms of the GNU General Public License
  10  *      as published by the Free Software Foundation; either version
  11  *      2 of the License, or (at your option) any later version.
  12  *
  13  *      Fixes:
  14  *      Vitaly E. Lavrov        releasing NULL neighbor in neigh_add.
  15  *      Harald Welte            Add neighbour cache statistics like rtstat
  16  */
  17
  18 #include <linux/types.h>
  19 #include <linux/kernel.h>
  20 #include <linux/module.h>
  21 #include <linux/socket.h>
  22 #include <linux/sched.h>
  23 #include <linux/netdevice.h>
  24 #include <linux/proc_fs.h>
  25 #ifdef CONFIG_SYSCTL
  26 #include <linux/sysctl.h>
  27 #endif
  28 #include <linux/times.h>
  29 #include <net/neighbour.h>
  30 #include <net/dst.h>
  31 #include <net/sock.h>
  32 #include <net/netevent.h>
  33 #include <linux/rtnetlink.h>
  34 #include <linux/random.h>
  35 #include <linux/string.h>
  36
  37 #define NEIGH_DEBUG 1
  38
  39 #define NEIGH_PRINTK(x...) printk(x)
  40 #define NEIGH_NOPRINTK(x...) do { ; } while(0)
  41 #define NEIGH_PRINTK0 NEIGH_PRINTK
  42 #define NEIGH_PRINTK1 NEIGH_NOPRINTK
  43 #define NEIGH_PRINTK2 NEIGH_NOPRINTK
  44
  45 #if NEIGH_DEBUG >= 1
  46 #undef NEIGH_PRINTK1
  47 #define NEIGH_PRINTK1 NEIGH_PRINTK
  48 #endif
  49 #if NEIGH_DEBUG >= 2
  50 #undef NEIGH_PRINTK2
  51 #define NEIGH_PRINTK2 NEIGH_PRINTK
  52 #endif
  53
  54 #define PNEIGH_HASHMASK         0xF
  55
  56 static void neigh_timer_handler(unsigned long arg);
  57 #ifdef CONFIG_ARPD
  58 static void neigh_app_notify(struct neighbour *n);
  59 #endif
  60 static int pneigh_ifdown(struct neigh_table *tbl, struct net_device *dev);
  61 void neigh_changeaddr(struct neigh_table *tbl, struct net_device *dev);
  62
  63 static struct neigh_table *neigh_tables;
  64 #ifdef CONFIG_PROC_FS
  65 static 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 /*
 108  * It is random distribution in the interval (1/2)*base...(3/2)*base.
 109  * It corresponds to default IPv6 settings and is not overridable,
 110  * because it is really reasonable choice.
 111  */
 112
 113 unsigned long neigh_rand_reach_time(unsigned long base)
 114 {
 115         return (base ? (net_random() % base) + (base >> 1) : 0);
 116 }
 117
 118
 119 static int neigh_forced_gc(struct neigh_table *tbl)
 120 {
 121         int shrunk = 0;
 122         int i;
 123
 124         NEIGH_CACHE_STAT_INC(tbl, forced_gc_runs);
 125
 126         write_lock_bh(&tbl->lock);
 127         for (i = 0; i <= tbl->hash_mask; i++) {
 128                 struct neighbour *n, **np;
 129
 130                 np = &tbl->hash_buckets[i];
 131                 while ((n = *np) != NULL) {
 132                         /* Neighbour record may be discarded if:
 133                          * - nobody refers to it.
 134                          * - it is not permanent
 135                          */
 136                         write_lock(&n->lock);
 137                         if (atomic_read(&n->refcnt) == 1 &&
 138                             !(n->nud_state & NUD_PERMANENT)) {
 139                                 *np     = n->next;
 140                                 n->dead = 1;
 141                                 shrunk  = 1;
 142                                 write_unlock(&n->lock);
 143                                 neigh_release(n);
 144                                 continue;
 145                         }
 146                         write_unlock(&n->lock);
 147                         np = &n->next;
 148                 }
 149         }
 150
 151         tbl->last_flush = jiffies;
 152
 153         write_unlock_bh(&tbl->lock);
 154
 155         return shrunk;
 156 }
 157
 158 static int neigh_del_timer(struct neighbour *n)
 159 {
 160         if ((n->nud_state & NUD_IN_TIMER) &&
 161             del_timer(&n->timer)) {
 162                 neigh_release(n);
 163                 return 1;
 164         }
 165         return 0;
 166 }
 167
 168 static void pneigh_queue_purge(struct sk_buff_head *list)
 169 {
 170         struct sk_buff *skb;
 171
 172         while ((skb = skb_dequeue(list)) != NULL) {
 173                 dev_put(skb->dev);
 174                 kfree_skb(skb);
 175         }
 176 }
 177
 178 static void neigh_flush_dev(struct neigh_table *tbl, struct net_device *dev)
 179 {
 180         int i;
 181
 182         for (i = 0; i <= tbl->hash_mask; i++) {
 183                 struct neighbour *n, **np = &tbl->hash_buckets[i];
 184
 185                 while ((n = *np) != NULL) {
 186                         if (dev && n->dev != dev) {
 187                                 np = &n->next;
 188                                 continue;
 189                         }
 190                         *np = n->next;
 191                         write_lock(&n->lock);
 192                         neigh_del_timer(n);
 193                         n->dead = 1;
 194
 195                         if (atomic_read(&n->refcnt) != 1) {
 196                                 /* The most unpleasant situation.
 197                                    We must destroy neighbour entry,
 198                                    but someone still uses it.
 199
 200                                    The destroy will be delayed until
 201                                    the last user releases us, but
 202                                    we must kill timers etc. and move
 203                                    it to safe state.
 204                                  */
 205                                 skb_queue_purge(&n->arp_queue);
 206                                 n->output = neigh_blackhole;
 207                                 if (n->nud_state & NUD_VALID)
 208                                         n->nud_state = NUD_NOARP;
 209                                 else
 210                                         n->nud_state = NUD_NONE;
 211                                 NEIGH_PRINTK2("neigh %p is stray.\n", n);
 212                         }
 213                         write_unlock(&n->lock);
 214                         neigh_release(n);
 215                 }
 216         }
 217 }
 218
 219 void neigh_changeaddr(struct neigh_table *tbl, struct net_device *dev)
 220 {
 221         write_lock_bh(&tbl->lock);
 222         neigh_flush_dev(tbl, dev);
 223         write_unlock_bh(&tbl->lock);
 224 }
 225
 226 int neigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
 227 {
 228         write_lock_bh(&tbl->lock);
 229         neigh_flush_dev(tbl, dev);
 230         pneigh_ifdown(tbl, dev);
 231         write_unlock_bh(&tbl->lock);
 232
 233         del_timer_sync(&tbl->proxy_timer);
 234         pneigh_queue_purge(&tbl->proxy_queue);
 235         return 0;
 236 }
 237
 238 static struct neighbour *neigh_alloc(struct neigh_table *tbl)
 239 {
 240         struct neighbour *n = NULL;
 241         unsigned long now = jiffies;
 242         int entries;
 243
 244         entries = atomic_inc_return(&tbl->entries) - 1;
 245         if (entries >= tbl->gc_thresh3 ||
 246             (entries >= tbl->gc_thresh2 &&
 247              time_after(now, tbl->last_flush + 5 * HZ))) {
 248                 if (!neigh_forced_gc(tbl) &&
 249                     entries >= tbl->gc_thresh3)
 250                         goto out_entries;
 251         }
 252
 253         n = kmem_cache_alloc(tbl->kmem_cachep, SLAB_ATOMIC);
 254         if (!n)
 255                 goto out_entries;
 256
 257         memset(n, 0, tbl->entry_size);
 258
 259         skb_queue_head_init(&n->arp_queue);
 260         rwlock_init(&n->lock);
 261         n->updated        = n->used = now;
 262         n->nud_state      = NUD_NONE;
 263         n->output         = neigh_blackhole;
 264         n->parms          = neigh_parms_clone(&tbl->parms);
 265         init_timer(&n->timer);
 266         n->timer.function = neigh_timer_handler;
 267         n->timer.data     = (unsigned long)n;
 268
 269         NEIGH_CACHE_STAT_INC(tbl, allocs);
 270         n->tbl            = tbl;
 271         atomic_set(&n->refcnt, 1);
 272         n->dead           = 1;
 273 out:
 274         return n;
 275
 276 out_entries:
 277         atomic_dec(&tbl->entries);
 278         goto out;
 279 }
 280
 281 static struct neighbour **neigh_hash_alloc(unsigned int entries)
 282 {
 283         unsigned long size = entries * sizeof(struct neighbour *);
 284         struct neighbour **ret;
 285
 286         if (size <= PAGE_SIZE) {
 287                 ret = kzalloc(size, GFP_ATOMIC);
 288         } else {
 289                 ret = (struct neighbour **)
 290                       __get_free_pages(GFP_ATOMIC|__GFP_ZERO, get_order(size));
 291         }
 292         return ret;
 293 }
 294
 295 static void neigh_hash_free(struct neighbour **hash, unsigned int entries)
 296 {
 297         unsigned long size = entries * sizeof(struct neighbour *);
 298
 299         if (size <= PAGE_SIZE)
 300                 kfree(hash);
 301         else
 302                 free_pages((unsigned long)hash, get_order(size));
 303 }
 304
 305 static void neigh_hash_grow(struct neigh_table *tbl, unsigned long new_entries)
 306 {
 307         struct neighbour **new_hash, **old_hash;
 308         unsigned int i, new_hash_mask, old_entries;
 309
 310         NEIGH_CACHE_STAT_INC(tbl, hash_grows);
 311
 312         BUG_ON(new_entries & (new_entries - 1));
 313         new_hash = neigh_hash_alloc(new_entries);
 314         if (!new_hash)
 315                 return;
 316
 317         old_entries = tbl->hash_mask + 1;
 318         new_hash_mask = new_entries - 1;
 319         old_hash = tbl->hash_buckets;
 320
 321         get_random_bytes(&tbl->hash_rnd, sizeof(tbl->hash_rnd));
 322         for (i = 0; i < old_entries; i++) {
 323                 struct neighbour *n, *next;
 324
 325                 for (n = old_hash[i]; n; n = next) {
 326                         unsigned int hash_val = tbl->hash(n->primary_key, n->dev);
 327
 328                         hash_val &= new_hash_mask;
 329                         next = n->next;
 330
 331                         n->next = new_hash[hash_val];
 332                         new_hash[hash_val] = n;
 333                 }
 334         }
 335         tbl->hash_buckets = new_hash;
 336         tbl->hash_mask = new_hash_mask;
 337
 338         neigh_hash_free(old_hash, old_entries);
 339 }
 340
 341 struct neighbour *neigh_lookup(struct neigh_table *tbl, const void *pkey,
 342                                struct net_device *dev)
 343 {
 344         struct neighbour *n;
 345         int key_len = tbl->key_len;
 346         u32 hash_val = tbl->hash(pkey, dev) & tbl->hash_mask;
 347
 348         NEIGH_CACHE_STAT_INC(tbl, lookups);
 349
 350         read_lock_bh(&tbl->lock);
 351         for (n = tbl->hash_buckets[hash_val]; n; n = n->next) {
 352                 if (dev == n->dev && !memcmp(n->primary_key, pkey, key_len)) {
 353                         neigh_hold(n);
 354                         NEIGH_CACHE_STAT_INC(tbl, hits);
 355                         break;
 356                 }
 357         }
 358         read_unlock_bh(&tbl->lock);
 359         return n;
 360 }
 361
 362 struct neighbour *neigh_lookup_nodev(struct neigh_table *tbl, const void *pkey)
 363 {
 364         struct neighbour *n;
 365         int key_len = tbl->key_len;
 366         u32 hash_val = tbl->hash(pkey, NULL) & tbl->hash_mask;
 367
 368         NEIGH_CACHE_STAT_INC(tbl, lookups);
 369
 370         read_lock_bh(&tbl->lock);
 371         for (n = tbl->hash_buckets[hash_val]; n; n = n->next) {
 372                 if (!memcmp(n->primary_key, pkey, key_len)) {
 373                         neigh_hold(n);
 374                         NEIGH_CACHE_STAT_INC(tbl, hits);
 375                         break;
 376                 }
 377         }
 378         read_unlock_bh(&tbl->lock);
 379         return n;
 380 }
 381
 382 struct neighbour *neigh_create(struct neigh_table *tbl, const void *pkey,
 383                                struct net_device *dev)
 384 {
 385         u32 hash_val;
 386         int key_len = tbl->key_len;
 387         int error;
 388         struct neighbour *n1, *rc, *n = neigh_alloc(tbl);
 389
 390         if (!n) {
 391                 rc = ERR_PTR(-ENOBUFS);
 392                 goto out;
 393         }
 394
 395         memcpy(n->primary_key, pkey, key_len);
 396         n->dev = dev;
 397         dev_hold(dev);
 398
 399         /* Protocol specific setup. */
 400         if (tbl->constructor && (error = tbl->constructor(n)) < 0) {
 401                 rc = ERR_PTR(error);
 402                 goto out_neigh_release;
 403         }
 404
 405         /* Device specific setup. */
 406         if (n->parms->neigh_setup &&
 407             (error = n->parms->neigh_setup(n)) < 0) {
 408                 rc = ERR_PTR(error);
 409                 goto out_neigh_release;
 410         }
 411
 412         n->confirmed = jiffies - (n->parms->base_reachable_time << 1);
 413
 414         write_lock_bh(&tbl->lock);
 415
 416         if (atomic_read(&tbl->entries) > (tbl->hash_mask + 1))
 417                 neigh_hash_grow(tbl, (tbl->hash_mask + 1) << 1);
 418
 419         hash_val = tbl->hash(pkey, dev) & tbl->hash_mask;
 420
 421         if (n->parms->dead) {
 422                 rc = ERR_PTR(-EINVAL);
 423                 goto out_tbl_unlock;
 424         }
 425
 426         for (n1 = tbl->hash_buckets[hash_val]; n1; n1 = n1->next) {
 427                 if (dev == n1->dev && !memcmp(n1->primary_key, pkey, key_len)) {
 428                         neigh_hold(n1);
 429                         rc = n1;
 430                         goto out_tbl_unlock;
 431                 }
 432         }
 433
 434         n->next = tbl->hash_buckets[hash_val];
 435         tbl->hash_buckets[hash_val] = n;
 436         n->dead = 0;
 437         neigh_hold(n);
 438         write_unlock_bh(&tbl->lock);
 439         NEIGH_PRINTK2("neigh %p is created.\n", n);
 440         rc = n;
 441 out:
 442         return rc;
 443 out_tbl_unlock:
 444         write_unlock_bh(&tbl->lock);
 445 out_neigh_release:
 446         neigh_release(n);
 447         goto out;
 448 }
 449
 450 struct pneigh_entry * pneigh_lookup(struct neigh_table *tbl, const void *pkey,
 451                                     struct net_device *dev, int creat)
 452 {
 453         struct pneigh_entry *n;
 454         int key_len = tbl->key_len;
 455         u32 hash_val = *(u32 *)(pkey + key_len - 4);
 456
 457         hash_val ^= (hash_val >> 16);
 458         hash_val ^= hash_val >> 8;
 459         hash_val ^= hash_val >> 4;
 460         hash_val &= PNEIGH_HASHMASK;
 461
 462         read_lock_bh(&tbl->lock);
 463
 464         for (n = tbl->phash_buckets[hash_val]; n; n = n->next) {
 465                 if (!memcmp(n->key, pkey, key_len) &&
 466                     (n->dev == dev || !n->dev)) {
 467                         read_unlock_bh(&tbl->lock);
 468                         goto out;
 469                 }
 470         }
 471         read_unlock_bh(&tbl->lock);
 472         n = NULL;
 473         if (!creat)
 474                 goto out;
 475
 476         n = kmalloc(sizeof(*n) + key_len, GFP_KERNEL);
 477         if (!n)
 478                 goto out;
 479
 480         memcpy(n->key, pkey, key_len);
 481         n->dev = dev;
 482         if (dev)
 483                 dev_hold(dev);
 484
 485         if (tbl->pconstructor && tbl->pconstructor(n)) {
 486                 if (dev)
 487                         dev_put(dev);
 488                 kfree(n);
 489                 n = NULL;
 490                 goto out;
 491         }
 492
 493         write_lock_bh(&tbl->lock);
 494         n->next = tbl->phash_buckets[hash_val];
 495         tbl->phash_buckets[hash_val] = n;
 496         write_unlock_bh(&tbl->lock);
 497 out:
 498         return n;
 499 }
 500
 501
 502 int pneigh_delete(struct neigh_table *tbl, const void *pkey,
 503                   struct net_device *dev)
 504 {
 505         struct pneigh_entry *n, **np;
 506         int key_len = tbl->key_len;
 507         u32 hash_val = *(u32 *)(pkey + key_len - 4);
 508
 509         hash_val ^= (hash_val >> 16);
 510         hash_val ^= hash_val >> 8;
 511         hash_val ^= hash_val >> 4;
 512         hash_val &= PNEIGH_HASHMASK;
 513
 514         write_lock_bh(&tbl->lock);
 515         for (np = &tbl->phash_buckets[hash_val]; (n = *np) != NULL;
 516              np = &n->next) {
 517                 if (!memcmp(n->key, pkey, key_len) && n->dev == dev) {
 518                         *np = n->next;
 519                         write_unlock_bh(&tbl->lock);
 520                         if (tbl->pdestructor)
 521                                 tbl->pdestructor(n);
 522                         if (n->dev)
 523                                 dev_put(n->dev);
 524                         kfree(n);
 525                         return 0;
 526                 }
 527         }
 528         write_unlock_bh(&tbl->lock);
 529         return -ENOENT;
 530 }
 531
 532 static int pneigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
 533 {
 534         struct pneigh_entry *n, **np;
 535         u32 h;
 536
 537         for (h = 0; h <= PNEIGH_HASHMASK; h++) {
 538                 np = &tbl->phash_buckets[h];
 539                 while ((n = *np) != NULL) {
 540                         if (!dev || n->dev == dev) {
 541                                 *np = n->next;
 542                                 if (tbl->pdestructor)
 543                                         tbl->pdestructor(n);
 544                                 if (n->dev)
 545                                         dev_put(n->dev);
 546                                 kfree(n);
 547                                 continue;
 548                         }
 549                         np = &n->next;
 550                 }
 551         }
 552         return -ENOENT;
 553 }
 554
 555
 556 /*
 557  *      neighbour must already be out of the table;
 558  *
 559  */
 560 void neigh_destroy(struct neighbour *neigh)
 561 {
 562         struct hh_cache *hh;
 563
 564         NEIGH_CACHE_STAT_INC(neigh->tbl, destroys);
 565
 566         if (!neigh->dead) {
 567                 printk(KERN_WARNING
 568                        "Destroying alive neighbour %p\n", neigh);
 569                 dump_stack();
 570                 return;
 571         }
 572
 573         if (neigh_del_timer(neigh))
 574                 printk(KERN_WARNING "Impossible event.\n");
 575
 576         while ((hh = neigh->hh) != NULL) {
 577                 neigh->hh = hh->hh_next;
 578                 hh->hh_next = NULL;
 579                 write_lock_bh(&hh->hh_lock);
 580                 hh->hh_output = neigh_blackhole;
 581                 write_unlock_bh(&hh->hh_lock);
 582                 if (atomic_dec_and_test(&hh->hh_refcnt))
 583                         kfree(hh);
 584         }
 585
 586         if (neigh->parms->neigh_destructor)
 587                 (neigh->parms->neigh_destructor)(neigh);
 588
 589         skb_queue_purge(&neigh->arp_queue);
 590
 591         dev_put(neigh->dev);
 592         neigh_parms_put(neigh->parms);
 593
 594         NEIGH_PRINTK2("neigh %p is destroyed.\n", neigh);
 595
 596         atomic_dec(&neigh->tbl->entries);
 597         kmem_cache_free(neigh->tbl->kmem_cachep, neigh);
 598 }
 599
 600 /* Neighbour state is suspicious;
 601    disable fast path.
 602
 603    Called with write_locked neigh.
 604  */
 605 static void neigh_suspect(struct neighbour *neigh)
 606 {
 607         struct hh_cache *hh;
 608
 609         NEIGH_PRINTK2("neigh %p is suspected.\n", neigh);
 610
 611         neigh->output = neigh->ops->output;
 612
 613         for (hh = neigh->hh; hh; hh = hh->hh_next)
 614                 hh->hh_output = neigh->ops->output;
 615 }
 616
 617 /* Neighbour state is OK;
 618    enable fast path.
 619
 620    Called with write_locked neigh.
 621  */
 622 static void neigh_connect(struct neighbour *neigh)
 623 {
 624         struct hh_cache *hh;
 625
 626         NEIGH_PRINTK2("neigh %p is connected.\n", neigh);
 627
 628         neigh->output = neigh->ops->connected_output;
 629
 630         for (hh = neigh->hh; hh; hh = hh->hh_next)
 631                 hh->hh_output = neigh->ops->hh_output;
 632 }
 633
 634 static void neigh_periodic_timer(unsigned long arg)
 635 {
 636         struct neigh_table *tbl = (struct neigh_table *)arg;
 637         struct neighbour *n, **np;
 638         unsigned long expire, now = jiffies;
 639
 640         NEIGH_CACHE_STAT_INC(tbl, periodic_gc_runs);
 641
 642         write_lock(&tbl->lock);
 643
 644         /*
 645          *      periodically recompute ReachableTime from random function
 646          */
 647
 648         if (time_after(now, tbl->last_rand + 300 * HZ)) {
 649                 struct neigh_parms *p;
 650                 tbl->last_rand = now;
 651                 for (p = &tbl->parms; p; p = p->next)
 652                         p->reachable_time =
 653                                 neigh_rand_reach_time(p->base_reachable_time);
 654         }
 655
 656         np = &tbl->hash_buckets[tbl->hash_chain_gc];
 657         tbl->hash_chain_gc = ((tbl->hash_chain_gc + 1) & tbl->hash_mask);
 658
 659         while ((n = *np) != NULL) {
 660                 unsigned int state;
 661
 662                 write_lock(&n->lock);
 663
 664                 state = n->nud_state;
 665                 if (state & (NUD_PERMANENT | NUD_IN_TIMER)) {
 666                         write_unlock(&n->lock);
 667                         goto next_elt;
 668                 }
 669
 670                 if (time_before(n->used, n->confirmed))
 671                         n->used = n->confirmed;
 672
 673                 if (atomic_read(&n->refcnt) == 1 &&
 674                     (state == NUD_FAILED ||
 675                      time_after(now, n->used + n->parms->gc_staletime))) {
 676                         *np = n->next;
 677                         n->dead = 1;
 678                         write_unlock(&n->lock);
 679                         neigh_release(n);
 680                         continue;
 681                 }
 682                 write_unlock(&n->lock);
 683
 684 next_elt:
 685                 np = &n->next;
 686         }
 687
 688         /* Cycle through all hash buckets every base_reachable_time/2 ticks.
 689          * ARP entry timeouts range from 1/2 base_reachable_time to 3/2
 690          * base_reachable_time.
 691          */
 692         expire = tbl->parms.base_reachable_time >> 1;
 693         expire /= (tbl->hash_mask + 1);
 694         if (!expire)
 695                 expire = 1;
 696
 697         mod_timer(&tbl->gc_timer, now + expire);
 698
 699         write_unlock(&tbl->lock);
 700 }
 701
 702 static __inline__ int neigh_max_probes(struct neighbour *n)
 703 {
 704         struct neigh_parms *p = n->parms;
 705         return (n->nud_state & NUD_PROBE ?
 706                 p->ucast_probes :
 707                 p->ucast_probes + p->app_probes + p->mcast_probes);
 708 }
 709
 710 static inline void neigh_add_timer(struct neighbour *n, unsigned long when)
 711 {
 712         if (unlikely(mod_timer(&n->timer, when))) {
 713                 printk("NEIGH: BUG, double timer add, state is %x\n",
 714                        n->nud_state);
 715                 dump_stack();
 716         }
 717 }
 718
 719 /* Called when a timer expires for a neighbour entry. */
 720
 721 static void neigh_timer_handler(unsigned long arg)
 722 {
 723         unsigned long now, next;
 724         struct neighbour *neigh = (struct neighbour *)arg;
 725         unsigned state;
 726         int notify = 0;
 727
 728         write_lock(&neigh->lock);
 729
 730         state = neigh->nud_state;
 731         now = jiffies;
 732         next = now + HZ;
 733
 734         if (!(state & NUD_IN_TIMER)) {
 735 #ifndef CONFIG_SMP
 736                 printk(KERN_WARNING "neigh: timer & !nud_in_timer\n");
 737 #endif
 738                 goto out;
 739         }
 740
 741         if (state & NUD_REACHABLE) {
 742                 if (time_before_eq(now,
 743                                    neigh->confirmed + neigh->parms->reachable_time)) {
 744                         NEIGH_PRINTK2("neigh %p is still alive.\n", neigh);
 745                         next = neigh->confirmed + neigh->parms->reachable_time;
 746                 } else if (time_before_eq(now,
 747                                           neigh->used + neigh->parms->delay_probe_time)) {
 748                         NEIGH_PRINTK2("neigh %p is delayed.\n", neigh);
 749                         neigh->nud_state = NUD_DELAY;
 750                         neigh->updated = jiffies;
 751                         neigh_suspect(neigh);
 752                         next = now + neigh->parms->delay_probe_time;
 753                 } else {
 754                         NEIGH_PRINTK2("neigh %p is suspected.\n", neigh);
 755                         neigh->nud_state = NUD_STALE;
 756                         neigh->updated = jiffies;
 757                         neigh_suspect(neigh);
 758                         notify = 1;
 759                 }
 760         } else if (state & NUD_DELAY) {
 761                 if (time_before_eq(now,
 762                                    neigh->confirmed + neigh->parms->delay_probe_time)) {
 763                         NEIGH_PRINTK2("neigh %p is now reachable.\n", neigh);
 764                         neigh->nud_state = NUD_REACHABLE;
 765                         neigh->updated = jiffies;
 766                         neigh_connect(neigh);
 767                         notify = 1;
 768                         next = neigh->confirmed + neigh->parms->reachable_time;
 769                 } else {
 770                         NEIGH_PRINTK2("neigh %p is probed.\n", neigh);
 771                         neigh->nud_state = NUD_PROBE;
 772                         neigh->updated = jiffies;
 773                         atomic_set(&neigh->probes, 0);
 774                         next = now + neigh->parms->retrans_time;
 775                 }
 776         } else {
 777                 /* NUD_PROBE|NUD_INCOMPLETE */
 778                 next = now + neigh->parms->retrans_time;
 779         }
 780
 781         if ((neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) &&
 782             atomic_read(&neigh->probes) >= neigh_max_probes(neigh)) {
 783                 struct sk_buff *skb;
 784
 785                 neigh->nud_state = NUD_FAILED;
 786                 neigh->updated = jiffies;
 787                 notify = 1;
 788                 NEIGH_CACHE_STAT_INC(neigh->tbl, res_failed);
 789                 NEIGH_PRINTK2("neigh %p is failed.\n", neigh);
 790
 791                 /* It is very thin place. report_unreachable is very complicated
 792                    routine. Particularly, it can hit the same neighbour entry!
 793
 794                    So that, we try to be accurate and avoid dead loop. --ANK
 795                  */
 796                 while (neigh->nud_state == NUD_FAILED &&
 797                        (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
 798                         write_unlock(&neigh->lock);
 799                         neigh->ops->error_report(neigh, skb);
 800                         write_lock(&neigh->lock);
 801                 }
 802                 skb_queue_purge(&neigh->arp_queue);
 803         }
 804
 805         if (neigh->nud_state & NUD_IN_TIMER) {
 806                 if (time_before(next, jiffies + HZ/2))
 807                         next = jiffies + HZ/2;
 808                 if (!mod_timer(&neigh->timer, next))
 809                         neigh_hold(neigh);
 810         }
 811         if (neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) {
 812                 struct sk_buff *skb = skb_peek(&neigh->arp_queue);
 813                 /* keep skb alive even if arp_queue overflows */
 814                 if (skb)
 815                         skb_get(skb);
 816                 write_unlock(&neigh->lock);
 817                 neigh->ops->solicit(neigh, skb);
 818                 atomic_inc(&neigh->probes);
 819                 if (skb)
 820                         kfree_skb(skb);
 821         } else {
 822 out:
 823                 write_unlock(&neigh->lock);
 824         }
 825         if (notify)
 826                 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, neigh);
 827
 828 #ifdef CONFIG_ARPD
 829         if (notify && neigh->parms->app_probes)
 830                 neigh_app_notify(neigh);
 831 #endif
 832         neigh_release(neigh);
 833 }
 834
 835 int __neigh_event_send(struct neighbour *neigh, struct sk_buff *skb)
 836 {
 837         int rc;
 838         unsigned long now;
 839
 840         write_lock_bh(&neigh->lock);
 841
 842         rc = 0;
 843         if (neigh->nud_state & (NUD_CONNECTED | NUD_DELAY | NUD_PROBE))
 844                 goto out_unlock_bh;
 845
 846         now = jiffies;
 847
 848         if (!(neigh->nud_state & (NUD_STALE | NUD_INCOMPLETE))) {
 849                 if (neigh->parms->mcast_probes + neigh->parms->app_probes) {
 850                         atomic_set(&neigh->probes, neigh->parms->ucast_probes);
 851                         neigh->nud_state     = NUD_INCOMPLETE;
 852                         neigh->updated = jiffies;
 853                         neigh_hold(neigh);
 854                         neigh_add_timer(neigh, now + 1);
 855                 } else {
 856                         neigh->nud_state = NUD_FAILED;
 857                         neigh->updated = jiffies;
 858                         write_unlock_bh(&neigh->lock);
 859
 860                         if (skb)
 861                                 kfree_skb(skb);
 862                         return 1;
 863                 }
 864         } else if (neigh->nud_state & NUD_STALE) {
 865                 NEIGH_PRINTK2("neigh %p is delayed.\n", neigh);
 866                 neigh_hold(neigh);
 867                 neigh->nud_state = NUD_DELAY;
 868                 neigh->updated = jiffies;
 869                 neigh_add_timer(neigh,
 870                                 jiffies + neigh->parms->delay_probe_time);
 871         }
 872
 873         if (neigh->nud_state == NUD_INCOMPLETE) {
 874                 if (skb) {
 875                         if (skb_queue_len(&neigh->arp_queue) >=
 876                             neigh->parms->queue_len) {
 877                                 struct sk_buff *buff;
 878                                 buff = neigh->arp_queue.next;
 879                                 __skb_unlink(buff, &neigh->arp_queue);
 880                                 kfree_skb(buff);
 881                         }
 882                         __skb_queue_tail(&neigh->arp_queue, skb);
 883                 }
 884                 rc = 1;
 885         }
 886 out_unlock_bh:
 887         write_unlock_bh(&neigh->lock);
 888         return rc;
 889 }
 890
 891 static __inline__ void neigh_update_hhs(struct neighbour *neigh)
 892 {
 893         struct hh_cache *hh;
 894         void (*update)(struct hh_cache*, struct net_device*, unsigned char *) =
 895                 neigh->dev->header_cache_update;
 896
 897         if (update) {
 898                 for (hh = neigh->hh; hh; hh = hh->hh_next) {
 899                         write_lock_bh(&hh->hh_lock);
 900                         update(hh, neigh->dev, neigh->ha);
 901                         write_unlock_bh(&hh->hh_lock);
 902                 }
 903         }
 904 }
 905
 906
 907
 908 /* Generic update routine.
 909    -- lladdr is new lladdr or NULL, if it is not supplied.
 910    -- new    is new state.
 911    -- flags
 912         NEIGH_UPDATE_F_OVERRIDE allows to override existing lladdr,
 913                                 if it is different.
 914         NEIGH_UPDATE_F_WEAK_OVERRIDE will suspect existing "connected"
 915                                 lladdr instead of overriding it
 916                                 if it is different.
 917                                 It also allows to retain current state
 918                                 if lladdr is unchanged.
 919         NEIGH_UPDATE_F_ADMIN    means that the change is administrative.
 920
 921         NEIGH_UPDATE_F_OVERRIDE_ISROUTER allows to override existing
 922                                 NTF_ROUTER flag.
 923         NEIGH_UPDATE_F_ISROUTER indicates if the neighbour is known as
 924                                 a router.
 925
 926    Caller MUST hold reference count on the entry.
 927  */
 928
 929 int neigh_update(struct neighbour *neigh, const u8 *lladdr, u8 new,
 930                  u32 flags)
 931 {
 932         u8 old;
 933         int err;
 934         int notify = 0;
 935         struct net_device *dev;
 936         int update_isrouter = 0;
 937
 938         write_lock_bh(&neigh->lock);
 939
 940         dev    = neigh->dev;
 941         old    = neigh->nud_state;
 942         err    = -EPERM;
 943
 944         if (!(flags & NEIGH_UPDATE_F_ADMIN) &&
 945             (old & (NUD_NOARP | NUD_PERMANENT)))
 946                 goto out;
 947
 948         if (!(new & NUD_VALID)) {
 949                 neigh_del_timer(neigh);
 950                 if (old & NUD_CONNECTED)
 951                         neigh_suspect(neigh);
 952                 neigh->nud_state = new;
 953                 err = 0;
 954                 notify = old & NUD_VALID;
 955                 goto out;
 956         }
 957
 958         /* Compare new lladdr with cached one */
 959         if (!dev->addr_len) {
 960                 /* First case: device needs no address. */
 961                 lladdr = neigh->ha;
 962         } else if (lladdr) {
 963                 /* The second case: if something is already cached
 964                    and a new address is proposed:
 965                    - compare new & old
 966                    - if they are different, check override flag
 967                  */
 968                 if ((old & NUD_VALID) &&
 969                     !memcmp(lladdr, neigh->ha, dev->addr_len))
 970                         lladdr = neigh->ha;
 971         } else {
 972                 /* No address is supplied; if we know something,
 973                    use it, otherwise discard the request.
 974                  */
 975                 err = -EINVAL;
 976                 if (!(old & NUD_VALID))
 977                         goto out;
 978                 lladdr = neigh->ha;
 979         }
 980
 981         if (new & NUD_CONNECTED)
 982                 neigh->confirmed = jiffies;
 983         neigh->updated = jiffies;
 984
 985         /* If entry was valid and address is not changed,
 986            do not change entry state, if new one is STALE.
 987          */
 988         err = 0;
 989         update_isrouter = flags & NEIGH_UPDATE_F_OVERRIDE_ISROUTER;
 990         if (old & NUD_VALID) {
 991                 if (lladdr != neigh->ha && !(flags & NEIGH_UPDATE_F_OVERRIDE)) {
 992                         update_isrouter = 0;
 993                         if ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) &&
 994                             (old & NUD_CONNECTED)) {
 995                                 lladdr = neigh->ha;
 996                                 new = NUD_STALE;
 997                         } else
 998                                 goto out;
 999                 } else {
1000                         if (lladdr == neigh->ha && new == NUD_STALE &&
1001                             ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) ||
1002                              (old & NUD_CONNECTED))
1003                             )
1004                                 new = old;
1005                 }
1006         }
1007
1008         if (new != old) {
1009                 neigh_del_timer(neigh);
1010                 if (new & NUD_IN_TIMER) {
1011                         neigh_hold(neigh);
1012                         neigh_add_timer(neigh, (jiffies +
1013                                                 ((new & NUD_REACHABLE) ?
1014                                                  neigh->parms->reachable_time :
1015                                                  0)));
1016                 }
1017                 neigh->nud_state = new;
1018         }
1019
1020         if (lladdr != neigh->ha) {
1021                 memcpy(&neigh->ha, lladdr, dev->addr_len);
1022                 neigh_update_hhs(neigh);
1023                 if (!(new & NUD_CONNECTED))
1024                         neigh->confirmed = jiffies -
1025                                       (neigh->parms->base_reachable_time << 1);
1026                 notify = 1;
1027         }
1028         if (new == old)
1029                 goto out;
1030         if (new & NUD_CONNECTED)
1031                 neigh_connect(neigh);
1032         else
1033                 neigh_suspect(neigh);
1034         if (!(old & NUD_VALID)) {
1035                 struct sk_buff *skb;
1036
1037                 /* Again: avoid dead loop if something went wrong */
1038
1039                 while (neigh->nud_state & NUD_VALID &&
1040                        (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
1041                         struct neighbour *n1 = neigh;
1042                         write_unlock_bh(&neigh->lock);
1043                         /* On shaper/eql skb->dst->neighbour != neigh :( */
1044                         if (skb->dst && skb->dst->neighbour)
1045                                 n1 = skb->dst->neighbour;
1046                         n1->output(skb);
1047                         write_lock_bh(&neigh->lock);
1048                 }
1049                 skb_queue_purge(&neigh->arp_queue);
1050         }
1051 out:
1052         if (update_isrouter) {
1053                 neigh->flags = (flags & NEIGH_UPDATE_F_ISROUTER) ?
1054                         (neigh->flags | NTF_ROUTER) :
1055                         (neigh->flags & ~NTF_ROUTER);
1056         }
1057         write_unlock_bh(&neigh->lock);
1058
1059         if (notify)
1060                 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, neigh);
1061 #ifdef CONFIG_ARPD
1062         if (notify && neigh->parms->app_probes)
1063                 neigh_app_notify(neigh);
1064 #endif
1065         return err;
1066 }
1067
1068 struct neighbour *neigh_event_ns(struct neigh_table *tbl,
1069                                  u8 *lladdr, void *saddr,
1070                                  struct net_device *dev)
1071 {
1072         struct neighbour *neigh = __neigh_lookup(tbl, saddr, dev,
1073                                                  lladdr || !dev->addr_len);
1074         if (neigh)
1075                 neigh_update(neigh, lladdr, NUD_STALE,
1076                              NEIGH_UPDATE_F_OVERRIDE);
1077         return neigh;
1078 }
1079
1080 static void neigh_hh_init(struct neighbour *n, struct dst_entry *dst,
1081                           u16 protocol)
1082 {
1083         struct hh_cache *hh;
1084         struct net_device *dev = dst->dev;
1085
1086         for (hh = n->hh; hh; hh = hh->hh_next)
1087                 if (hh->hh_type == protocol)
1088                         break;
1089
1090         if (!hh && (hh = kzalloc(sizeof(*hh), GFP_ATOMIC)) != NULL) {
1091                 rwlock_init(&hh->hh_lock);
1092                 hh->hh_type = protocol;
1093                 atomic_set(&hh->hh_refcnt, 0);
1094                 hh->hh_next = NULL;
1095                 if (dev->hard_header_cache(n, hh)) {
1096                         kfree(hh);
1097                         hh = NULL;
1098                 } else {
1099                         atomic_inc(&hh->hh_refcnt);
1100                         hh->hh_next = n->hh;
1101                         n->hh       = hh;
1102                         if (n->nud_state & NUD_CONNECTED)
1103                                 hh->hh_output = n->ops->hh_output;
1104                         else
1105                                 hh->hh_output = n->ops->output;
1106                 }
1107         }
1108         if (hh) {
1109                 atomic_inc(&hh->hh_refcnt);
1110                 dst->hh = hh;
1111         }
1112 }
1113
1114 /* This function can be used in contexts, where only old dev_queue_xmit
1115    worked, f.e. if you want to override normal output path (eql, shaper),
1116    but resolution is not made yet.
1117  */
1118
1119 int neigh_compat_output(struct sk_buff *skb)
1120 {
1121         struct net_device *dev = skb->dev;
1122
1123         __skb_pull(skb, skb->nh.raw - skb->data);
1124
1125         if (dev->hard_header &&
1126             dev->hard_header(skb, dev, ntohs(skb->protocol), NULL, NULL,
1127                              skb->len) < 0 &&
1128             dev->rebuild_header(skb))
1129                 return 0;
1130
1131         return dev_queue_xmit(skb);
1132 }
1133
1134 /* Slow and careful. */
1135
1136 int neigh_resolve_output(struct sk_buff *skb)
1137 {
1138         struct dst_entry *dst = skb->dst;
1139         struct neighbour *neigh;
1140         int rc = 0;
1141
1142         if (!dst || !(neigh = dst->neighbour))
1143                 goto discard;
1144
1145         __skb_pull(skb, skb->nh.raw - skb->data);
1146
1147         if (!neigh_event_send(neigh, skb)) {
1148                 int err;
1149                 struct net_device *dev = neigh->dev;
1150                 if (dev->hard_header_cache && !dst->hh) {
1151                         write_lock_bh(&neigh->lock);
1152                         if (!dst->hh)
1153                                 neigh_hh_init(neigh, dst, dst->ops->protocol);
1154                         err = dev->hard_header(skb, dev, ntohs(skb->protocol),
1155                                                neigh->ha, NULL, skb->len);
1156                         write_unlock_bh(&neigh->lock);
1157                 } else {
1158                         read_lock_bh(&neigh->lock);
1159                         err = dev->hard_header(skb, dev, ntohs(skb->protocol),
1160                                                neigh->ha, NULL, skb->len);
1161                         read_unlock_bh(&neigh->lock);
1162                 }
1163                 if (err >= 0)
1164                         rc = neigh->ops->queue_xmit(skb);
1165                 else
1166                         goto out_kfree_skb;
1167         }
1168 out:
1169         return rc;
1170 discard:
1171         NEIGH_PRINTK1("neigh_resolve_output: dst=%p neigh=%p\n",
1172                       dst, dst ? dst->neighbour : NULL);
1173 out_kfree_skb:
1174         rc = -EINVAL;
1175         kfree_skb(skb);
1176         goto out;
1177 }
1178
1179 /* As fast as possible without hh cache */
1180
1181 int neigh_connected_output(struct sk_buff *skb)
1182 {
1183         int err;
1184         struct dst_entry *dst = skb->dst;
1185         struct neighbour *neigh = dst->neighbour;
1186         struct net_device *dev = neigh->dev;
1187
1188         __skb_pull(skb, skb->nh.raw - skb->data);
1189
1190         read_lock_bh(&neigh->lock);
1191         err = dev->hard_header(skb, dev, ntohs(skb->protocol),
1192                                neigh->ha, NULL, skb->len);
1193         read_unlock_bh(&neigh->lock);
1194         if (err >= 0)
1195                 err = neigh->ops->queue_xmit(skb);
1196         else {
1197                 err = -EINVAL;
1198                 kfree_skb(skb);
1199         }
1200         return err;
1201 }
1202
1203 static void neigh_proxy_process(unsigned long arg)
1204 {
1205         struct neigh_table *tbl = (struct neigh_table *)arg;
1206         long sched_next = 0;
1207         unsigned long now = jiffies;
1208         struct sk_buff *skb;
1209
1210         spin_lock(&tbl->proxy_queue.lock);
1211
1212         skb = tbl->proxy_queue.next;
1213
1214         while (skb != (struct sk_buff *)&tbl->proxy_queue) {
1215                 struct sk_buff *back = skb;
1216                 long tdif = NEIGH_CB(back)->sched_next - now;
1217
1218                 skb = skb->next;
1219                 if (tdif <= 0) {
1220                         struct net_device *dev = back->dev;
1221                         __skb_unlink(back, &tbl->proxy_queue);
1222                         if (tbl->proxy_redo && netif_running(dev))
1223                                 tbl->proxy_redo(back);
1224                         else
1225                                 kfree_skb(back);
1226
1227                         dev_put(dev);
1228                 } else if (!sched_next || tdif < sched_next)
1229                         sched_next = tdif;
1230         }
1231         del_timer(&tbl->proxy_timer);
1232         if (sched_next)
1233                 mod_timer(&tbl->proxy_timer, jiffies + sched_next);
1234         spin_unlock(&tbl->proxy_queue.lock);
1235 }
1236
1237 void pneigh_enqueue(struct neigh_table *tbl, struct neigh_parms *p,
1238                     struct sk_buff *skb)
1239 {
1240         unsigned long now = jiffies;
1241         unsigned long sched_next = now + (net_random() % p->proxy_delay);
1242
1243         if (tbl->proxy_queue.qlen > p->proxy_qlen) {
1244                 kfree_skb(skb);
1245                 return;
1246         }
1247
1248         NEIGH_CB(skb)->sched_next = sched_next;
1249         NEIGH_CB(skb)->flags |= LOCALLY_ENQUEUED;
1250
1251         spin_lock(&tbl->proxy_queue.lock);
1252         if (del_timer(&tbl->proxy_timer)) {
1253                 if (time_before(tbl->proxy_timer.expires, sched_next))
1254                         sched_next = tbl->proxy_timer.expires;
1255         }
1256         dst_release(skb->dst);
1257         skb->dst = NULL;
1258         dev_hold(skb->dev);
1259         __skb_queue_tail(&tbl->proxy_queue, skb);
1260         mod_timer(&tbl->proxy_timer, sched_next);
1261         spin_unlock(&tbl->proxy_queue.lock);
1262 }
1263
1264
1265 struct neigh_parms *neigh_parms_alloc(struct net_device *dev,
1266                                       struct neigh_table *tbl)
1267 {
1268         struct neigh_parms *p = kmalloc(sizeof(*p), GFP_KERNEL);
1269
1270         if (p) {
1271                 memcpy(p, &tbl->parms, sizeof(*p));
1272                 p->tbl            = tbl;
1273                 atomic_set(&p->refcnt, 1);
1274                 INIT_RCU_HEAD(&p->rcu_head);
1275                 p->reachable_time =
1276                                 neigh_rand_reach_time(p->base_reachable_time);
1277                 if (dev) {
1278                         if (dev->neigh_setup && dev->neigh_setup(dev, p)) {
1279                                 kfree(p);
1280                                 return NULL;
1281                         }
1282
1283                         dev_hold(dev);
1284                         p->dev = dev;
1285                 }
1286                 p->sysctl_table = NULL;
1287                 write_lock_bh(&tbl->lock);
1288                 p->next         = tbl->parms.next;
1289                 tbl->parms.next = p;
1290                 write_unlock_bh(&tbl->lock);
1291         }
1292         return p;
1293 }
1294
1295 static void neigh_rcu_free_parms(struct rcu_head *head)
1296 {
1297         struct neigh_parms *parms =
1298                 container_of(head, struct neigh_parms, rcu_head);
1299
1300         neigh_parms_put(parms);
1301 }
1302
1303 void neigh_parms_release(struct neigh_table *tbl, struct neigh_parms *parms)
1304 {
1305         struct neigh_parms **p;
1306
1307         if (!parms || parms == &tbl->parms)
1308                 return;
1309         write_lock_bh(&tbl->lock);
1310         for (p = &tbl->parms.next; *p; p = &(*p)->next) {
1311                 if (*p == parms) {
1312                         *p = parms->next;
1313                         parms->dead = 1;
1314                         write_unlock_bh(&tbl->lock);
1315                         if (parms->dev)
1316                                 dev_put(parms->dev);
1317                         call_rcu(&parms->rcu_head, neigh_rcu_free_parms);
1318                         return;
1319                 }
1320         }
1321         write_unlock_bh(&tbl->lock);
1322         NEIGH_PRINTK1("neigh_parms_release: not found\n");
1323 }
1324
1325 void neigh_parms_destroy(struct neigh_parms *parms)
1326 {
1327         kfree(parms);
1328 }
1329
1330 void neigh_table_init_no_netlink(struct neigh_table *tbl)
1331 {
1332         unsigned long now = jiffies;
1333         unsigned long phsize;
1334
1335         atomic_set(&tbl->parms.refcnt, 1);
1336         INIT_RCU_HEAD(&tbl->parms.rcu_head);
1337         tbl->parms.reachable_time =
1338                           neigh_rand_reach_time(tbl->parms.base_reachable_time);
1339
1340         if (!tbl->kmem_cachep)
1341                 tbl->kmem_cachep = kmem_cache_create(tbl->id,
1342                                                      tbl->entry_size,
1343                                                      0, SLAB_HWCACHE_ALIGN,
1344                                                      NULL, NULL);
1345
1346         if (!tbl->kmem_cachep)
1347                 panic("cannot create neighbour cache");
1348
1349         tbl->stats = alloc_percpu(struct neigh_statistics);
1350         if (!tbl->stats)
1351                 panic("cannot create neighbour cache statistics");
1352
1353 #ifdef CONFIG_PROC_FS
1354         tbl->pde = create_proc_entry(tbl->id, 0, proc_net_stat);
1355         if (!tbl->pde)
1356                 panic("cannot create neighbour proc dir entry");
1357         tbl->pde->proc_fops = &neigh_stat_seq_fops;
1358         tbl->pde->data = tbl;
1359 #endif
1360
1361         tbl->hash_mask = 1;
1362         tbl->hash_buckets = neigh_hash_alloc(tbl->hash_mask + 1);
1363
1364         phsize = (PNEIGH_HASHMASK + 1) * sizeof(struct pneigh_entry *);
1365         tbl->phash_buckets = kzalloc(phsize, GFP_KERNEL);
1366
1367         if (!tbl->hash_buckets || !tbl->phash_buckets)
1368                 panic("cannot allocate neighbour cache hashes");
1369
1370         get_random_bytes(&tbl->hash_rnd, sizeof(tbl->hash_rnd));
1371
1372         rwlock_init(&tbl->lock);
1373         init_timer(&tbl->gc_timer);
1374         tbl->gc_timer.data     = (unsigned long)tbl;
1375         tbl->gc_timer.function = neigh_periodic_timer;
1376         tbl->gc_timer.expires  = now + 1;
1377         add_timer(&tbl->gc_timer);
1378
1379         init_timer(&tbl->proxy_timer);
1380         tbl->proxy_timer.data     = (unsigned long)tbl;
1381         tbl->proxy_timer.function = neigh_proxy_process;
1382         skb_queue_head_init(&tbl->proxy_queue);
1383
1384         tbl->last_flush = now;
1385         tbl->last_rand  = now + tbl->parms.reachable_time * 20;
1386 }
1387
1388 void neigh_table_init(struct neigh_table *tbl)
1389 {
1390         struct neigh_table *tmp;
1391
1392         neigh_table_init_no_netlink(tbl);
1393         write_lock(&neigh_tbl_lock);
1394         for (tmp = neigh_tables; tmp; tmp = tmp->next) {
1395                 if (tmp->family == tbl->family)
1396                         break;
1397         }
1398         tbl->next       = neigh_tables;
1399         neigh_tables    = tbl;
1400         write_unlock(&neigh_tbl_lock);
1401
1402         if (unlikely(tmp)) {
1403                 printk(KERN_ERR "NEIGH: Registering multiple tables for "
1404                        "family %d\n", tbl->family);
1405                 dump_stack();
1406         }
1407 }
1408
1409 int neigh_table_clear(struct neigh_table *tbl)
1410 {
1411         struct neigh_table **tp;
1412
1413         /* It is not clean... Fix it to unload IPv6 module safely */
1414         del_timer_sync(&tbl->gc_timer);
1415         del_timer_sync(&tbl->proxy_timer);
1416         pneigh_queue_purge(&tbl->proxy_queue);
1417         neigh_ifdown(tbl, NULL);
1418         if (atomic_read(&tbl->entries))
1419                 printk(KERN_CRIT "neighbour leakage\n");
1420         write_lock(&neigh_tbl_lock);
1421         for (tp = &neigh_tables; *tp; tp = &(*tp)->next) {
1422                 if (*tp == tbl) {
1423                         *tp = tbl->next;
1424                         break;
1425                 }
1426         }
1427         write_unlock(&neigh_tbl_lock);
1428
1429         neigh_hash_free(tbl->hash_buckets, tbl->hash_mask + 1);
1430         tbl->hash_buckets = NULL;
1431
1432         kfree(tbl->phash_buckets);
1433         tbl->phash_buckets = NULL;
1434
1435         return 0;
1436 }
1437
1438 int neigh_delete(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
1439 {
1440         struct ndmsg *ndm = NLMSG_DATA(nlh);
1441         struct rtattr **nda = arg;
1442         struct neigh_table *tbl;
1443         struct net_device *dev = NULL;
1444         int err = -ENODEV;
1445
1446         if (ndm->ndm_ifindex &&
1447             (dev = dev_get_by_index(ndm->ndm_ifindex)) == NULL)
1448                 goto out;
1449
1450         read_lock(&neigh_tbl_lock);
1451         for (tbl = neigh_tables; tbl; tbl = tbl->next) {
1452                 struct rtattr *dst_attr = nda[NDA_DST - 1];
1453                 struct neighbour *n;
1454
1455                 if (tbl->family != ndm->ndm_family)
1456                         continue;
1457                 read_unlock(&neigh_tbl_lock);
1458
1459                 err = -EINVAL;
1460                 if (!dst_attr || RTA_PAYLOAD(dst_attr) < tbl->key_len)
1461                         goto out_dev_put;
1462
1463                 if (ndm->ndm_flags & NTF_PROXY) {
1464                         err = pneigh_delete(tbl, RTA_DATA(dst_attr), dev);
1465                         goto out_dev_put;
1466                 }
1467
1468                 if (!dev)
1469                         goto out;
1470
1471                 n = neigh_lookup(tbl, RTA_DATA(dst_attr), dev);
1472                 if (n) {
1473                         err = neigh_update(n, NULL, NUD_FAILED,
1474                                            NEIGH_UPDATE_F_OVERRIDE|
1475                                            NEIGH_UPDATE_F_ADMIN);
1476                         neigh_release(n);
1477                 }
1478                 goto out_dev_put;
1479         }
1480         read_unlock(&neigh_tbl_lock);
1481         err = -EADDRNOTAVAIL;
1482 out_dev_put:
1483         if (dev)
1484                 dev_put(dev);
1485 out:
1486         return err;
1487 }
1488
1489 int neigh_add(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
1490 {
1491         struct ndmsg *ndm = NLMSG_DATA(nlh);
1492         struct rtattr **nda = arg;
1493         struct neigh_table *tbl;
1494         struct net_device *dev = NULL;
1495         int err = -ENODEV;
1496
1497         if (ndm->ndm_ifindex &&
1498             (dev = dev_get_by_index(ndm->ndm_ifindex)) == NULL)
1499                 goto out;
1500
1501         read_lock(&neigh_tbl_lock);
1502         for (tbl = neigh_tables; tbl; tbl = tbl->next) {
1503                 struct rtattr *lladdr_attr = nda[NDA_LLADDR - 1];
1504                 struct rtattr *dst_attr = nda[NDA_DST - 1];
1505                 int override = 1;
1506                 struct neighbour *n;
1507
1508                 if (tbl->family != ndm->ndm_family)
1509                         continue;
1510                 read_unlock(&neigh_tbl_lock);
1511
1512                 err = -EINVAL;
1513                 if (!dst_attr || RTA_PAYLOAD(dst_attr) < tbl->key_len)
1514                         goto out_dev_put;
1515
1516                 if (ndm->ndm_flags & NTF_PROXY) {
1517                         err = -ENOBUFS;
1518                         if (pneigh_lookup(tbl, RTA_DATA(dst_attr), dev, 1))
1519                                 err = 0;
1520                         goto out_dev_put;
1521                 }
1522
1523                 err = -EINVAL;
1524                 if (!dev)
1525                         goto out;
1526                 if (lladdr_attr && RTA_PAYLOAD(lladdr_attr) < dev->addr_len)
1527                         goto out_dev_put;
1528
1529                 n = neigh_lookup(tbl, RTA_DATA(dst_attr), dev);
1530                 if (n) {
1531                         if (nlh->nlmsg_flags & NLM_F_EXCL) {
1532                                 err = -EEXIST;
1533                                 neigh_release(n);
1534                                 goto out_dev_put;
1535                         }
1536
1537                         override = nlh->nlmsg_flags & NLM_F_REPLACE;
1538                 } else if (!(nlh->nlmsg_flags & NLM_F_CREATE)) {
1539                         err = -ENOENT;
1540                         goto out_dev_put;
1541                 } else {
1542                         n = __neigh_lookup_errno(tbl, RTA_DATA(dst_attr), dev);
1543                         if (IS_ERR(n)) {
1544                                 err = PTR_ERR(n);
1545                                 goto out_dev_put;
1546                         }
1547                 }
1548
1549                 err = neigh_update(n,
1550                                    lladdr_attr ? RTA_DATA(lladdr_attr) : NULL,
1551                                    ndm->ndm_state,
1552                                    (override ? NEIGH_UPDATE_F_OVERRIDE : 0) |
1553                                    NEIGH_UPDATE_F_ADMIN);
1554
1555                 neigh_release(n);
1556                 goto out_dev_put;
1557         }
1558
1559         read_unlock(&neigh_tbl_lock);
1560         err = -EADDRNOTAVAIL;
1561 out_dev_put:
1562         if (dev)
1563                 dev_put(dev);
1564 out:
1565         return err;
1566 }
1567
1568 static int neightbl_fill_parms(struct sk_buff *skb, struct neigh_parms *parms)
1569 {
1570         struct rtattr *nest = NULL;
1571
1572         nest = RTA_NEST(skb, NDTA_PARMS);
1573
1574         if (parms->dev)
1575                 RTA_PUT_U32(skb, NDTPA_IFINDEX, parms->dev->ifindex);
1576
1577         RTA_PUT_U32(skb, NDTPA_REFCNT, atomic_read(&parms->refcnt));
1578         RTA_PUT_U32(skb, NDTPA_QUEUE_LEN, parms->queue_len);
1579         RTA_PUT_U32(skb, NDTPA_PROXY_QLEN, parms->proxy_qlen);
1580         RTA_PUT_U32(skb, NDTPA_APP_PROBES, parms->app_probes);
1581         RTA_PUT_U32(skb, NDTPA_UCAST_PROBES, parms->ucast_probes);
1582         RTA_PUT_U32(skb, NDTPA_MCAST_PROBES, parms->mcast_probes);
1583         RTA_PUT_MSECS(skb, NDTPA_REACHABLE_TIME, parms->reachable_time);
1584         RTA_PUT_MSECS(skb, NDTPA_BASE_REACHABLE_TIME,
1585                       parms->base_reachable_time);
1586         RTA_PUT_MSECS(skb, NDTPA_GC_STALETIME, parms->gc_staletime);
1587         RTA_PUT_MSECS(skb, NDTPA_DELAY_PROBE_TIME, parms->delay_probe_time);
1588         RTA_PUT_MSECS(skb, NDTPA_RETRANS_TIME, parms->retrans_time);
1589         RTA_PUT_MSECS(skb, NDTPA_ANYCAST_DELAY, parms->anycast_delay);
1590         RTA_PUT_MSECS(skb, NDTPA_PROXY_DELAY, parms->proxy_delay);
1591         RTA_PUT_MSECS(skb, NDTPA_LOCKTIME, parms->locktime);
1592
1593         return RTA_NEST_END(skb, nest);
1594
1595 rtattr_failure:
1596         return RTA_NEST_CANCEL(skb, nest);
1597 }
1598
1599 static int neightbl_fill_info(struct neigh_table *tbl, struct sk_buff *skb,
1600                               struct netlink_callback *cb)
1601 {
1602         struct nlmsghdr *nlh;
1603         struct ndtmsg *ndtmsg;
1604
1605         nlh = NLMSG_NEW_ANSWER(skb, cb, RTM_NEWNEIGHTBL, sizeof(struct ndtmsg),
1606                                NLM_F_MULTI);
1607
1608         ndtmsg = NLMSG_DATA(nlh);
1609
1610         read_lock_bh(&tbl->lock);
1611         ndtmsg->ndtm_family = tbl->family;
1612         ndtmsg->ndtm_pad1   = 0;
1613         ndtmsg->ndtm_pad2   = 0;
1614
1615         RTA_PUT_STRING(skb, NDTA_NAME, tbl->id);
1616         RTA_PUT_MSECS(skb, NDTA_GC_INTERVAL, tbl->gc_interval);
1617         RTA_PUT_U32(skb, NDTA_THRESH1, tbl->gc_thresh1);
1618         RTA_PUT_U32(skb, NDTA_THRESH2, tbl->gc_thresh2);
1619         RTA_PUT_U32(skb, NDTA_THRESH3, tbl->gc_thresh3);
1620
1621         {
1622                 unsigned long now = jiffies;
1623                 unsigned int flush_delta = now - tbl->last_flush;
1624                 unsigned int rand_delta = now - tbl->last_rand;
1625
1626                 struct ndt_config ndc = {
1627                         .ndtc_key_len           = tbl->key_len,
1628                         .ndtc_entry_size        = tbl->entry_size,
1629                         .ndtc_entries           = atomic_read(&tbl->entries),
1630                         .ndtc_last_flush        = jiffies_to_msecs(flush_delta),
1631                         .ndtc_last_rand         = jiffies_to_msecs(rand_delta),
1632                         .ndtc_hash_rnd          = tbl->hash_rnd,
1633                         .ndtc_hash_mask         = tbl->hash_mask,
1634                         .ndtc_hash_chain_gc     = tbl->hash_chain_gc,
1635                         .ndtc_proxy_qlen        = tbl->proxy_queue.qlen,
1636                 };
1637
1638                 RTA_PUT(skb, NDTA_CONFIG, sizeof(ndc), &ndc);
1639         }
1640
1641         {
1642                 int cpu;
1643                 struct ndt_stats ndst;
1644
1645                 memset(&ndst, 0, sizeof(ndst));
1646
1647                 for_each_possible_cpu(cpu) {
1648                         struct neigh_statistics *st;
1649
1650                         st = per_cpu_ptr(tbl->stats, cpu);
1651                         ndst.ndts_allocs                += st->allocs;
1652                         ndst.ndts_destroys              += st->destroys;
1653                         ndst.ndts_hash_grows            += st->hash_grows;
1654                         ndst.ndts_res_failed            += st->res_failed;
1655                         ndst.ndts_lookups               += st->lookups;
1656                         ndst.ndts_hits                  += st->hits;
1657                         ndst.ndts_rcv_probes_mcast      += st->rcv_probes_mcast;
1658                         ndst.ndts_rcv_probes_ucast      += st->rcv_probes_ucast;
1659                         ndst.ndts_periodic_gc_runs      += st->periodic_gc_runs;
1660                         ndst.ndts_forced_gc_runs        += st->forced_gc_runs;
1661                 }
1662
1663                 RTA_PUT(skb, NDTA_STATS, sizeof(ndst), &ndst);
1664         }
1665
1666         BUG_ON(tbl->parms.dev);
1667         if (neightbl_fill_parms(skb, &tbl->parms) < 0)
1668                 goto rtattr_failure;
1669
1670         read_unlock_bh(&tbl->lock);
1671         return NLMSG_END(skb, nlh);
1672
1673 rtattr_failure:
1674         read_unlock_bh(&tbl->lock);
1675         return NLMSG_CANCEL(skb, nlh);
1676
1677 nlmsg_failure:
1678         return -1;
1679 }
1680
1681 static int neightbl_fill_param_info(struct neigh_table *tbl,
1682                                     struct neigh_parms *parms,
1683                                     struct sk_buff *skb,
1684                                     struct netlink_callback *cb)
1685 {
1686         struct ndtmsg *ndtmsg;
1687         struct nlmsghdr *nlh;
1688
1689         nlh = NLMSG_NEW_ANSWER(skb, cb, RTM_NEWNEIGHTBL, sizeof(struct ndtmsg),
1690                                NLM_F_MULTI);
1691
1692         ndtmsg = NLMSG_DATA(nlh);
1693
1694         read_lock_bh(&tbl->lock);
1695         ndtmsg->ndtm_family = tbl->family;
1696         ndtmsg->ndtm_pad1   = 0;
1697         ndtmsg->ndtm_pad2   = 0;
1698         RTA_PUT_STRING(skb, NDTA_NAME, tbl->id);
1699
1700         if (neightbl_fill_parms(skb, parms) < 0)
1701                 goto rtattr_failure;
1702
1703         read_unlock_bh(&tbl->lock);
1704         return NLMSG_END(skb, nlh);
1705
1706 rtattr_failure:
1707         read_unlock_bh(&tbl->lock);
1708         return NLMSG_CANCEL(skb, nlh);
1709
1710 nlmsg_failure:
1711         return -1;
1712 }
1713
1714 static inline struct neigh_parms *lookup_neigh_params(struct neigh_table *tbl,
1715                                                       int ifindex)
1716 {
1717         struct neigh_parms *p;
1718
1719         for (p = &tbl->parms; p; p = p->next)
1720                 if ((p->dev && p->dev->ifindex == ifindex) ||
1721                     (!p->dev && !ifindex))
1722                         return p;
1723
1724         return NULL;
1725 }
1726
1727 int neightbl_set(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
1728 {
1729         struct neigh_table *tbl;
1730         struct ndtmsg *ndtmsg = NLMSG_DATA(nlh);
1731         struct rtattr **tb = arg;
1732         int err = -EINVAL;
1733
1734         if (!tb[NDTA_NAME - 1] || !RTA_PAYLOAD(tb[NDTA_NAME - 1]))
1735                 return -EINVAL;
1736
1737         read_lock(&neigh_tbl_lock);
1738         for (tbl = neigh_tables; tbl; tbl = tbl->next) {
1739                 if (ndtmsg->ndtm_family && tbl->family != ndtmsg->ndtm_family)
1740                         continue;
1741
1742                 if (!rtattr_strcmp(tb[NDTA_NAME - 1], tbl->id))
1743                         break;
1744         }
1745
1746         if (tbl == NULL) {
1747                 err = -ENOENT;
1748                 goto errout;
1749         }
1750
1751         /*
1752          * We acquire tbl->lock to be nice to the periodic timers and
1753          * make sure they always see a consistent set of values.
1754          */
1755         write_lock_bh(&tbl->lock);
1756
1757         if (tb[NDTA_THRESH1 - 1])
1758                 tbl->gc_thresh1 = RTA_GET_U32(tb[NDTA_THRESH1 - 1]);
1759
1760         if (tb[NDTA_THRESH2 - 1])
1761                 tbl->gc_thresh2 = RTA_GET_U32(tb[NDTA_THRESH2 - 1]);
1762
1763         if (tb[NDTA_THRESH3 - 1])
1764                 tbl->gc_thresh3 = RTA_GET_U32(tb[NDTA_THRESH3 - 1]);
1765
1766         if (tb[NDTA_GC_INTERVAL - 1])
1767                 tbl->gc_interval = RTA_GET_MSECS(tb[NDTA_GC_INTERVAL - 1]);
1768
1769         if (tb[NDTA_PARMS - 1]) {
1770                 struct rtattr *tbp[NDTPA_MAX];
1771                 struct neigh_parms *p;
1772                 u32 ifindex = 0;
1773
1774                 if (rtattr_parse_nested(tbp, NDTPA_MAX, tb[NDTA_PARMS - 1]) < 0)
1775                         goto rtattr_failure;
1776
1777                 if (tbp[NDTPA_IFINDEX - 1])
1778                         ifindex = RTA_GET_U32(tbp[NDTPA_IFINDEX - 1]);
1779
1780                 p = lookup_neigh_params(tbl, ifindex);
1781                 if (p == NULL) {
1782                         err = -ENOENT;
1783                         goto rtattr_failure;
1784                 }
1785
1786                 if (tbp[NDTPA_QUEUE_LEN - 1])
1787                         p->queue_len = RTA_GET_U32(tbp[NDTPA_QUEUE_LEN - 1]);
1788
1789                 if (tbp[NDTPA_PROXY_QLEN - 1])
1790                         p->proxy_qlen = RTA_GET_U32(tbp[NDTPA_PROXY_QLEN - 1]);
1791
1792                 if (tbp[NDTPA_APP_PROBES - 1])
1793                         p->app_probes = RTA_GET_U32(tbp[NDTPA_APP_PROBES - 1]);
1794
1795                 if (tbp[NDTPA_UCAST_PROBES - 1])
1796                         p->ucast_probes =
1797                            RTA_GET_U32(tbp[NDTPA_UCAST_PROBES - 1]);
1798
1799                 if (tbp[NDTPA_MCAST_PROBES - 1])
1800                         p->mcast_probes =
1801                            RTA_GET_U32(tbp[NDTPA_MCAST_PROBES - 1]);
1802
1803                 if (tbp[NDTPA_BASE_REACHABLE_TIME - 1])
1804                         p->base_reachable_time =
1805                            RTA_GET_MSECS(tbp[NDTPA_BASE_REACHABLE_TIME - 1]);
1806
1807                 if (tbp[NDTPA_GC_STALETIME - 1])
1808                         p->gc_staletime =
1809                            RTA_GET_MSECS(tbp[NDTPA_GC_STALETIME - 1]);
1810
1811                 if (tbp[NDTPA_DELAY_PROBE_TIME - 1])
1812                         p->delay_probe_time =
1813                            RTA_GET_MSECS(tbp[NDTPA_DELAY_PROBE_TIME - 1]);
1814
1815                 if (tbp[NDTPA_RETRANS_TIME - 1])
1816                         p->retrans_time =
1817                            RTA_GET_MSECS(tbp[NDTPA_RETRANS_TIME - 1]);
1818
1819                 if (tbp[NDTPA_ANYCAST_DELAY - 1])
1820                         p->anycast_delay =
1821                            RTA_GET_MSECS(tbp[NDTPA_ANYCAST_DELAY - 1]);
1822
1823                 if (tbp[NDTPA_PROXY_DELAY - 1])
1824                         p->proxy_delay =
1825                            RTA_GET_MSECS(tbp[NDTPA_PROXY_DELAY - 1]);
1826
1827                 if (tbp[NDTPA_LOCKTIME - 1])
1828                         p->locktime = RTA_GET_MSECS(tbp[NDTPA_LOCKTIME - 1]);
1829         }
1830
1831         err = 0;
1832
1833 rtattr_failure:
1834         write_unlock_bh(&tbl->lock);
1835 errout:
1836         read_unlock(&neigh_tbl_lock);
1837         return err;
1838 }
1839
1840 int neightbl_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
1841 {
1842         int idx, family;
1843         int s_idx = cb->args[0];
1844         struct neigh_table *tbl;
1845
1846         family = ((struct rtgenmsg *)NLMSG_DATA(cb->nlh))->rtgen_family;
1847
1848         read_lock(&neigh_tbl_lock);
1849         for (tbl = neigh_tables, idx = 0; tbl; tbl = tbl->next) {
1850                 struct neigh_parms *p;
1851
1852                 if (idx < s_idx || (family && tbl->family != family))
1853                         continue;
1854
1855                 if (neightbl_fill_info(tbl, skb, cb) <= 0)
1856                         break;
1857
1858                 for (++idx, p = tbl->parms.next; p; p = p->next, idx++) {
1859                         if (idx < s_idx)
1860                                 continue;
1861
1862                         if (neightbl_fill_param_info(tbl, p, skb, cb) <= 0)
1863                                 goto out;
1864                 }
1865
1866         }
1867 out:
1868         read_unlock(&neigh_tbl_lock);
1869         cb->args[0] = idx;
1870
1871         return skb->len;
1872 }
1873
1874 static int neigh_fill_info(struct sk_buff *skb, struct neighbour *n,
1875                            u32 pid, u32 seq, int event, unsigned int flags)
1876 {
1877         unsigned long now = jiffies;
1878         unsigned char *b = skb->tail;
1879         struct nda_cacheinfo ci;
1880         int locked = 0;
1881         u32 probes;
1882         struct nlmsghdr *nlh = NLMSG_NEW(skb, pid, seq, event,
1883                                          sizeof(struct ndmsg), flags);
1884         struct ndmsg *ndm = NLMSG_DATA(nlh);
1885
1886         ndm->ndm_family  = n->ops->family;
1887         ndm->ndm_pad1    = 0;
1888         ndm->ndm_pad2    = 0;
1889         ndm->ndm_flags   = n->flags;
1890         ndm->ndm_type    = n->type;
1891         ndm->ndm_ifindex = n->dev->ifindex;
1892         RTA_PUT(skb, NDA_DST, n->tbl->key_len, n->primary_key);
1893         read_lock_bh(&n->lock);
1894         locked           = 1;
1895         ndm->ndm_state   = n->nud_state;
1896         if (n->nud_state & NUD_VALID)
1897                 RTA_PUT(skb, NDA_LLADDR, n->dev->addr_len, n->ha);
1898         ci.ndm_used      = now - n->used;
1899         ci.ndm_confirmed = now - n->confirmed;
1900         ci.ndm_updated   = now - n->updated;
1901         ci.ndm_refcnt    = atomic_read(&n->refcnt) - 1;
1902         probes = atomic_read(&n->probes);
1903         read_unlock_bh(&n->lock);
1904         locked           = 0;
1905         RTA_PUT(skb, NDA_CACHEINFO, sizeof(ci), &ci);
1906         RTA_PUT(skb, NDA_PROBES, sizeof(probes), &probes);
1907         nlh->nlmsg_len   = skb->tail - b;
1908         return skb->len;
1909
1910 nlmsg_failure:
1911 rtattr_failure:
1912         if (locked)
1913                 read_unlock_bh(&n->lock);
1914         skb_trim(skb, b - skb->data);
1915         return -1;
1916 }
1917
1918
1919 static int neigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
1920                             struct netlink_callback *cb)
1921 {
1922         struct neighbour *n;
1923         int rc, h, s_h = cb->args[1];
1924         int idx, s_idx = idx = cb->args[2];
1925
1926         for (h = 0; h <= tbl->hash_mask; h++) {
1927                 if (h < s_h)
1928                         continue;
1929                 if (h > s_h)
1930                         s_idx = 0;
1931                 read_lock_bh(&tbl->lock);
1932                 for (n = tbl->hash_buckets[h], idx = 0; n; n = n->next, idx++) {
1933                         if (idx < s_idx)
1934                                 continue;
1935                         if (neigh_fill_info(skb, n, NETLINK_CB(cb->skb).pid,
1936                                             cb->nlh->nlmsg_seq,
1937                                             RTM_NEWNEIGH,
1938                                             NLM_F_MULTI) <= 0) {
1939                                 read_unlock_bh(&tbl->lock);
1940                                 rc = -1;
1941                                 goto out;
1942                         }
1943                 }
1944                 read_unlock_bh(&tbl->lock);
1945         }
1946         rc = skb->len;
1947 out:
1948         cb->args[1] = h;
1949         cb->args[2] = idx;
1950         return rc;
1951 }
1952
1953 int neigh_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
1954 {
1955         struct neigh_table *tbl;
1956         int t, family, s_t;
1957
1958         read_lock(&neigh_tbl_lock);
1959         family = ((struct rtgenmsg *)NLMSG_DATA(cb->nlh))->rtgen_family;
1960         s_t = cb->args[0];
1961
1962         for (tbl = neigh_tables, t = 0; tbl; tbl = tbl->next, t++) {
1963                 if (t < s_t || (family && tbl->family != family))
1964                         continue;
1965                 if (t > s_t)
1966                         memset(&cb->args[1], 0, sizeof(cb->args) -
1967                                                 sizeof(cb->args[0]));
1968                 if (neigh_dump_table(tbl, skb, cb) < 0)
1969                         break;
1970         }
1971         read_unlock(&neigh_tbl_lock);
1972
1973         cb->args[0] = t;
1974         return skb->len;
1975 }
1976
1977 void neigh_for_each(struct neigh_table *tbl, void (*cb)(struct neighbour *, void *), void *cookie)
1978 {
1979         int chain;
1980
1981         read_lock_bh(&tbl->lock);
1982         for (chain = 0; chain <= tbl->hash_mask; chain++) {
1983                 struct neighbour *n;
1984
1985                 for (n = tbl->hash_buckets[chain]; n; n = n->next)
1986                         cb(n, cookie);
1987         }
1988         read_unlock_bh(&tbl->lock);
1989 }
1990 EXPORT_SYMBOL(neigh_for_each);
1991
1992 /* The tbl->lock must be held as a writer and BH disabled. */
1993 void __neigh_for_each_release(struct neigh_table *tbl,
1994                               int (*cb)(struct neighbour *))
1995 {
1996         int chain;
1997
1998         for (chain = 0; chain <= tbl->hash_mask; chain++) {
1999                 struct neighbour *n, **np;
2000
2001                 np = &tbl->hash_buckets[chain];
2002                 while ((n = *np) != NULL) {
2003                         int release;
2004
2005                         write_lock(&n->lock);
2006                         release = cb(n);
2007                         if (release) {
2008                                 *np = n->next;
2009                                 n->dead = 1;
2010                         } else
2011                                 np = &n->next;
2012                         write_unlock(&n->lock);
2013                         if (release)
2014                                 neigh_release(n);
2015                 }
2016         }
2017 }
2018 EXPORT_SYMBOL(__neigh_for_each_release);
2019
2020 #ifdef CONFIG_PROC_FS
2021
2022 static struct neighbour *neigh_get_first(struct seq_file *seq)
2023 {
2024         struct neigh_seq_state *state = seq->private;
2025         struct neigh_table *tbl = state->tbl;
2026         struct neighbour *n = NULL;
2027         int bucket = state->bucket;
2028
2029         state->flags &= ~NEIGH_SEQ_IS_PNEIGH;
2030         for (bucket = 0; bucket <= tbl->hash_mask; bucket++) {
2031                 n = tbl->hash_buckets[bucket];
2032
2033                 while (n) {
2034                         if (state->neigh_sub_iter) {
2035                                 loff_t fakep = 0;
2036                                 void *v;
2037
2038                                 v = state->neigh_sub_iter(state, n, &fakep);
2039                                 if (!v)
2040                                         goto next;
2041                         }
2042                         if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
2043                                 break;
2044                         if (n->nud_state & ~NUD_NOARP)
2045                                 break;
2046                 next:
2047                         n = n->next;
2048                 }
2049
2050                 if (n)
2051                         break;
2052         }
2053         state->bucket = bucket;
2054
2055         return n;
2056 }
2057
2058 static struct neighbour *neigh_get_next(struct seq_file *seq,
2059                                         struct neighbour *n,
2060                                         loff_t *pos)
2061 {
2062         struct neigh_seq_state *state = seq->private;
2063         struct neigh_table *tbl = state->tbl;
2064
2065         if (state->neigh_sub_iter) {
2066                 void *v = state->neigh_sub_iter(state, n, pos);
2067                 if (v)
2068                         return n;
2069         }
2070         n = n->next;
2071
2072         while (1) {
2073                 while (n) {
2074                         if (state->neigh_sub_iter) {
2075                                 void *v = state->neigh_sub_iter(state, n, pos);
2076                                 if (v)
2077                                         return n;
2078                                 goto next;
2079                         }
2080                         if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
2081                                 break;
2082
2083                         if (n->nud_state & ~NUD_NOARP)
2084                                 break;
2085                 next:
2086                         n = n->next;
2087                 }
2088
2089                 if (n)
2090                         break;
2091
2092                 if (++state->bucket > tbl->hash_mask)
2093                         break;
2094
2095                 n = tbl->hash_buckets[state->bucket];
2096         }
2097
2098         if (n && pos)
2099                 --(*pos);
2100         return n;
2101 }
2102
2103 static struct neighbour *neigh_get_idx(struct seq_file *seq, loff_t *pos)
2104 {
2105         struct neighbour *n = neigh_get_first(seq);
2106
2107         if (n) {
2108                 while (*pos) {
2109                         n = neigh_get_next(seq, n, pos);
2110                         if (!n)
2111                                 break;
2112                 }
2113         }
2114         return *pos ? NULL : n;
2115 }
2116
2117 static struct pneigh_entry *pneigh_get_first(struct seq_file *seq)
2118 {
2119         struct neigh_seq_state *state = seq->private;
2120         struct neigh_table *tbl = state->tbl;
2121         struct pneigh_entry *pn = NULL;
2122         int bucket = state->bucket;
2123
2124         state->flags |= NEIGH_SEQ_IS_PNEIGH;
2125         for (bucket = 0; bucket <= PNEIGH_HASHMASK; bucket++) {
2126                 pn = tbl->phash_buckets[bucket];
2127                 if (pn)
2128                         break;
2129         }
2130         state->bucket = bucket;
2131
2132         return pn;
2133 }
2134
2135 static struct pneigh_entry *pneigh_get_next(struct seq_file *seq,
2136                                             struct pneigh_entry *pn,
2137                                             loff_t *pos)
2138 {
2139         struct neigh_seq_state *state = seq->private;
2140         struct neigh_table *tbl = state->tbl;
2141
2142         pn = pn->next;
2143         while (!pn) {
2144                 if (++state->bucket > PNEIGH_HASHMASK)
2145                         break;
2146                 pn = tbl->phash_buckets[state->bucket];
2147                 if (pn)
2148                         break;
2149         }
2150
2151         if (pn && pos)
2152                 --(*pos);
2153
2154         return pn;
2155 }
2156
2157 static struct pneigh_entry *pneigh_get_idx(struct seq_file *seq, loff_t *pos)
2158 {
2159         struct pneigh_entry *pn = pneigh_get_first(seq);
2160
2161         if (pn) {
2162                 while (*pos) {
2163                         pn = pneigh_get_next(seq, pn, pos);
2164                         if (!pn)
2165                                 break;
2166                 }
2167         }
2168         return *pos ? NULL : pn;
2169 }
2170
2171 static void *neigh_get_idx_any(struct seq_file *seq, loff_t *pos)
2172 {
2173         struct neigh_seq_state *state = seq->private;
2174         void *rc;
2175
2176         rc = neigh_get_idx(seq, pos);
2177         if (!rc && !(state->flags & NEIGH_SEQ_NEIGH_ONLY))
2178                 rc = pneigh_get_idx(seq, pos);
2179
2180         return rc;
2181 }
2182
2183 void *neigh_seq_start(struct seq_file *seq, loff_t *pos, struct neigh_table *tbl, unsigned int neigh_seq_flags)
2184 {
2185         struct neigh_seq_state *state = seq->private;
2186         loff_t pos_minus_one;
2187
2188         state->tbl = tbl;
2189         state->bucket = 0;
2190         state->flags = (neigh_seq_flags & ~NEIGH_SEQ_IS_PNEIGH);
2191
2192         read_lock_bh(&tbl->lock);
2193
2194         pos_minus_one = *pos - 1;
2195         return *pos ? neigh_get_idx_any(seq, &pos_minus_one) : SEQ_START_TOKEN;
2196 }
2197 EXPORT_SYMBOL(neigh_seq_start);
2198
2199 void *neigh_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2200 {
2201         struct neigh_seq_state *state;
2202         void *rc;
2203
2204         if (v == SEQ_START_TOKEN) {
2205                 rc = neigh_get_idx(seq, pos);
2206                 goto out;
2207         }
2208
2209         state = seq->private;
2210         if (!(state->flags & NEIGH_SEQ_IS_PNEIGH)) {
2211                 rc = neigh_get_next(seq, v, NULL);
2212                 if (rc)
2213                         goto out;
2214                 if (!(state->flags & NEIGH_SEQ_NEIGH_ONLY))
2215                         rc = pneigh_get_first(seq);
2216         } else {
2217                 BUG_ON(state->flags & NEIGH_SEQ_NEIGH_ONLY);
2218                 rc = pneigh_get_next(seq, v, NULL);
2219         }
2220 out:
2221         ++(*pos);
2222         return rc;
2223 }
2224 EXPORT_SYMBOL(neigh_seq_next);
2225
2226 void neigh_seq_stop(struct seq_file *seq, void *v)
2227 {
2228         struct neigh_seq_state *state = seq->private;
2229         struct neigh_table *tbl = state->tbl;
2230
2231         read_unlock_bh(&tbl->lock);
2232 }
2233 EXPORT_SYMBOL(neigh_seq_stop);
2234
2235 /* statistics via seq_file */
2236
2237 static void *neigh_stat_seq_start(struct seq_file *seq, loff_t *pos)
2238 {
2239         struct proc_dir_entry *pde = seq->private;
2240         struct neigh_table *tbl = pde->data;
2241         int cpu;
2242
2243         if (*pos == 0)
2244                 return SEQ_START_TOKEN;
2245
2246         for (cpu = *pos-1; cpu < NR_CPUS; ++cpu) {
2247                 if (!cpu_possible(cpu))
2248                         continue;
2249                 *pos = cpu+1;
2250                 return per_cpu_ptr(tbl->stats, cpu);
2251         }
2252         return NULL;
2253 }
2254
2255 static void *neigh_stat_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2256 {
2257         struct proc_dir_entry *pde = seq->private;
2258         struct neigh_table *tbl = pde->data;
2259         int cpu;
2260
2261         for (cpu = *pos; cpu < NR_CPUS; ++cpu) {
2262                 if (!cpu_possible(cpu))
2263                         continue;
2264                 *pos = cpu+1;
2265                 return per_cpu_ptr(tbl->stats, cpu);
2266         }
2267         return NULL;
2268 }
2269
2270 static void neigh_stat_seq_stop(struct seq_file *seq, void *v)
2271 {
2272
2273 }
2274
2275 static int neigh_stat_seq_show(struct seq_file *seq, void *v)
2276 {
2277         struct proc_dir_entry *pde = seq->private;
2278         struct neigh_table *tbl = pde->data;
2279         struct neigh_statistics *st = v;
2280
2281         if (v == SEQ_START_TOKEN) {
2282                 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");
2283                 return 0;
2284         }
2285
2286         seq_printf(seq, "%08x  %08lx %08lx %08lx  %08lx %08lx  %08lx  "
2287                         "%08lx %08lx  %08lx %08lx\n",
2288                    atomic_read(&tbl->entries),
2289
2290                    st->allocs,
2291                    st->destroys,
2292                    st->hash_grows,
2293
2294                    st->lookups,
2295                    st->hits,
2296
2297                    st->res_failed,
2298
2299                    st->rcv_probes_mcast,
2300                    st->rcv_probes_ucast,
2301
2302                    st->periodic_gc_runs,
2303                    st->forced_gc_runs
2304                    );
2305
2306         return 0;
2307 }
2308
2309 static struct seq_operations neigh_stat_seq_ops = {
2310         .start  = neigh_stat_seq_start,
2311         .next   = neigh_stat_seq_next,
2312         .stop   = neigh_stat_seq_stop,
2313         .show   = neigh_stat_seq_show,
2314 };
2315
2316 static int neigh_stat_seq_open(struct inode *inode, struct file *file)
2317 {
2318         int ret = seq_open(file, &neigh_stat_seq_ops);
2319
2320         if (!ret) {
2321                 struct seq_file *sf = file->private_data;
2322                 sf->private = PDE(inode);
2323         }
2324         return ret;
2325 };
2326
2327 static struct file_operations neigh_stat_seq_fops = {
2328         .owner   = THIS_MODULE,
2329         .open    = neigh_stat_seq_open,
2330         .read    = seq_read,
2331         .llseek  = seq_lseek,
2332         .release = seq_release,
2333 };
2334
2335 #endif /* CONFIG_PROC_FS */
2336
2337 #ifdef CONFIG_ARPD
2338 void neigh_app_ns(struct neighbour *n)
2339 {
2340         struct nlmsghdr  *nlh;
2341         int size = NLMSG_SPACE(sizeof(struct ndmsg) + 256);
2342         struct sk_buff *skb = alloc_skb(size, GFP_ATOMIC);
2343
2344         if (!skb)
2345                 return;
2346
2347         if (neigh_fill_info(skb, n, 0, 0, RTM_GETNEIGH, 0) < 0) {
2348                 kfree_skb(skb);
2349                 return;
2350         }
2351         nlh                        = (struct nlmsghdr *)skb->data;
2352         nlh->nlmsg_flags           = NLM_F_REQUEST;
2353         NETLINK_CB(skb).dst_group  = RTNLGRP_NEIGH;
2354         netlink_broadcast(rtnl, skb, 0, RTNLGRP_NEIGH, GFP_ATOMIC);
2355 }
2356
2357 static void neigh_app_notify(struct neighbour *n)
2358 {
2359         struct nlmsghdr *nlh;
2360         int size = NLMSG_SPACE(sizeof(struct ndmsg) + 256);
2361         struct sk_buff *skb = alloc_skb(size, GFP_ATOMIC);
2362
2363         if (!skb)
2364                 return;
2365
2366         if (neigh_fill_info(skb, n, 0, 0, RTM_NEWNEIGH, 0) < 0) {
2367                 kfree_skb(skb);
2368                 return;
2369         }
2370         nlh                        = (struct nlmsghdr *)skb->data;
2371         NETLINK_CB(skb).dst_group  = RTNLGRP_NEIGH;
2372         netlink_broadcast(rtnl, skb, 0, RTNLGRP_NEIGH, GFP_ATOMIC);
2373 }
2374
2375 #endif /* CONFIG_ARPD */
2376
2377 #ifdef CONFIG_SYSCTL
2378
2379 static struct neigh_sysctl_table {
2380         struct ctl_table_header *sysctl_header;
2381         ctl_table               neigh_vars[__NET_NEIGH_MAX];
2382         ctl_table               neigh_dev[2];
2383         ctl_table               neigh_neigh_dir[2];
2384         ctl_table               neigh_proto_dir[2];
2385         ctl_table               neigh_root_dir[2];
2386 } neigh_sysctl_template = {
2387         .neigh_vars = {
2388                 {
2389                         .ctl_name       = NET_NEIGH_MCAST_SOLICIT,
2390                         .procname       = "mcast_solicit",
2391                         .maxlen         = sizeof(int),
2392                         .mode           = 0644,
2393                         .proc_handler   = &proc_dointvec,
2394                 },
2395                 {
2396                         .ctl_name       = NET_NEIGH_UCAST_SOLICIT,
2397                         .procname       = "ucast_solicit",
2398                         .maxlen         = sizeof(int),
2399                         .mode           = 0644,
2400                         .proc_handler   = &proc_dointvec,
2401                 },
2402                 {
2403                         .ctl_name       = NET_NEIGH_APP_SOLICIT,
2404                         .procname       = "app_solicit",
2405                         .maxlen         = sizeof(int),
2406                         .mode           = 0644,
2407                         .proc_handler   = &proc_dointvec,
2408                 },
2409                 {
2410                         .ctl_name       = NET_NEIGH_RETRANS_TIME,
2411                         .procname       = "retrans_time",
2412                         .maxlen         = sizeof(int),
2413                         .mode           = 0644,
2414                         .proc_handler   = &proc_dointvec_userhz_jiffies,
2415                 },
2416                 {
2417                         .ctl_name       = NET_NEIGH_REACHABLE_TIME,
2418                         .procname       = "base_reachable_time",
2419                         .maxlen         = sizeof(int),
2420                         .mode           = 0644,
2421                         .proc_handler   = &proc_dointvec_jiffies,
2422                         .strategy       = &sysctl_jiffies,
2423                 },
2424                 {
2425                         .ctl_name       = NET_NEIGH_DELAY_PROBE_TIME,
2426                         .procname       = "delay_first_probe_time",
2427                         .maxlen         = sizeof(int),
2428                         .mode           = 0644,
2429                         .proc_handler   = &proc_dointvec_jiffies,
2430                         .strategy       = &sysctl_jiffies,
2431                 },
2432                 {
2433                         .ctl_name       = NET_NEIGH_GC_STALE_TIME,
2434                         .procname       = "gc_stale_time",
2435                         .maxlen         = sizeof(int),
2436                         .mode           = 0644,
2437                         .proc_handler   = &proc_dointvec_jiffies,
2438                         .strategy       = &sysctl_jiffies,
2439                 },
2440                 {
2441                         .ctl_name       = NET_NEIGH_UNRES_QLEN,
2442                         .procname       = "unres_qlen",
2443                         .maxlen         = sizeof(int),
2444                         .mode           = 0644,
2445                         .proc_handler   = &proc_dointvec,
2446                 },
2447                 {
2448                         .ctl_name       = NET_NEIGH_PROXY_QLEN,
2449                         .procname       = "proxy_qlen",
2450                         .maxlen         = sizeof(int),
2451                         .mode           = 0644,
2452                         .proc_handler   = &proc_dointvec,
2453                 },
2454                 {
2455                         .ctl_name       = NET_NEIGH_ANYCAST_DELAY,
2456                         .procname       = "anycast_delay",
2457                         .maxlen         = sizeof(int),
2458                         .mode           = 0644,
2459                         .proc_handler   = &proc_dointvec_userhz_jiffies,
2460                 },
2461                 {
2462                         .ctl_name       = NET_NEIGH_PROXY_DELAY,
2463                         .procname       = "proxy_delay",
2464                         .maxlen         = sizeof(int),
2465                         .mode           = 0644,
2466                         .proc_handler   = &proc_dointvec_userhz_jiffies,
2467                 },
2468                 {
2469                         .ctl_name       = NET_NEIGH_LOCKTIME,
2470                         .procname       = "locktime",
2471                         .maxlen         = sizeof(int),
2472                         .mode           = 0644,
2473                         .proc_handler   = &proc_dointvec_userhz_jiffies,
2474                 },
2475                 {
2476                         .ctl_name       = NET_NEIGH_GC_INTERVAL,
2477                         .procname       = "gc_interval",
2478                         .maxlen         = sizeof(int),
2479                         .mode           = 0644,
2480                         .proc_handler   = &proc_dointvec_jiffies,
2481                         .strategy       = &sysctl_jiffies,
2482                 },
2483                 {
2484                         .ctl_name       = NET_NEIGH_GC_THRESH1,
2485                         .procname       = "gc_thresh1",
2486                         .maxlen         = sizeof(int),
2487                         .mode           = 0644,
2488                         .proc_handler   = &proc_dointvec,
2489                 },
2490                 {
2491                         .ctl_name       = NET_NEIGH_GC_THRESH2,
2492                         .procname       = "gc_thresh2",
2493                         .maxlen         = sizeof(int),
2494                         .mode           = 0644,
2495                         .proc_handler   = &proc_dointvec,
2496                 },
2497                 {
2498                         .ctl_name       = NET_NEIGH_GC_THRESH3,
2499                         .procname       = "gc_thresh3",
2500                         .maxlen         = sizeof(int),
2501                         .mode           = 0644,
2502                         .proc_handler   = &proc_dointvec,
2503                 },
2504                 {
2505                         .ctl_name       = NET_NEIGH_RETRANS_TIME_MS,
2506                         .procname       = "retrans_time_ms",
2507                         .maxlen         = sizeof(int),
2508                         .mode           = 0644,
2509                         .proc_handler   = &proc_dointvec_ms_jiffies,
2510                         .strategy       = &sysctl_ms_jiffies,
2511                 },
2512                 {
2513                         .ctl_name       = NET_NEIGH_REACHABLE_TIME_MS,
2514                         .procname       = "base_reachable_time_ms",
2515                         .maxlen         = sizeof(int),
2516                         .mode           = 0644,
2517                         .proc_handler   = &proc_dointvec_ms_jiffies,
2518                         .strategy       = &sysctl_ms_jiffies,
2519                 },
2520         },
2521         .neigh_dev = {
2522                 {
2523                         .ctl_name       = NET_PROTO_CONF_DEFAULT,
2524                         .procname       = "default",
2525                         .mode           = 0555,
2526                 },
2527         },
2528         .neigh_neigh_dir = {
2529                 {
2530                         .procname       = "neigh",
2531                         .mode           = 0555,
2532                 },
2533         },
2534         .neigh_proto_dir = {
2535                 {
2536                         .mode           = 0555,
2537                 },
2538         },
2539         .neigh_root_dir = {
2540                 {
2541                         .ctl_name       = CTL_NET,
2542                         .procname       = "net",
2543                         .mode           = 0555,
2544                 },
2545         },
2546 };
2547
2548 int neigh_sysctl_register(struct net_device *dev, struct neigh_parms *p,
2549                           int p_id, int pdev_id, char *p_name,
2550                           proc_handler *handler, ctl_handler *strategy)
2551 {
2552         struct neigh_sysctl_table *t = kmalloc(sizeof(*t), GFP_KERNEL);
2553         const char *dev_name_source = NULL;
2554         char *dev_name = NULL;
2555         int err = 0;
2556
2557         if (!t)
2558                 return -ENOBUFS;
2559         memcpy(t, &neigh_sysctl_template, sizeof(*t));
2560         t->neigh_vars[0].data  = &p->mcast_probes;
2561         t->neigh_vars[1].data  = &p->ucast_probes;
2562         t->neigh_vars[2].data  = &p->app_probes;
2563         t->neigh_vars[3].data  = &p->retrans_time;
2564         t->neigh_vars[4].data  = &p->base_reachable_time;
2565         t->neigh_vars[5].data  = &p->delay_probe_time;
2566         t->neigh_vars[6].data  = &p->gc_staletime;
2567         t->neigh_vars[7].data  = &p->queue_len;
2568         t->neigh_vars[8].data  = &p->proxy_qlen;
2569         t->neigh_vars[9].data  = &p->anycast_delay;
2570         t->neigh_vars[10].data = &p->proxy_delay;
2571         t->neigh_vars[11].data = &p->locktime;
2572
2573         if (dev) {
2574                 dev_name_source = dev->name;
2575                 t->neigh_dev[0].ctl_name = dev->ifindex;
2576                 t->neigh_vars[12].procname = NULL;
2577                 t->neigh_vars[13].procname = NULL;
2578                 t->neigh_vars[14].procname = NULL;
2579                 t->neigh_vars[15].procname = NULL;
2580         } else {
2581                 dev_name_source = t->neigh_dev[0].procname;
2582                 t->neigh_vars[12].data = (int *)(p + 1);
2583                 t->neigh_vars[13].data = (int *)(p + 1) + 1;
2584                 t->neigh_vars[14].data = (int *)(p + 1) + 2;
2585                 t->neigh_vars[15].data = (int *)(p + 1) + 3;
2586         }
2587
2588         t->neigh_vars[16].data  = &p->retrans_time;
2589         t->neigh_vars[17].data  = &p->base_reachable_time;
2590
2591         if (handler || strategy) {
2592                 /* RetransTime */
2593                 t->neigh_vars[3].proc_handler = handler;
2594                 t->neigh_vars[3].strategy = strategy;
2595                 t->neigh_vars[3].extra1 = dev;
2596                 /* ReachableTime */
2597                 t->neigh_vars[4].proc_handler = handler;
2598                 t->neigh_vars[4].strategy = strategy;
2599                 t->neigh_vars[4].extra1 = dev;
2600                 /* RetransTime (in milliseconds)*/
2601                 t->neigh_vars[16].proc_handler = handler;
2602                 t->neigh_vars[16].strategy = strategy;
2603                 t->neigh_vars[16].extra1 = dev;
2604                 /* ReachableTime (in milliseconds) */
2605                 t->neigh_vars[17].proc_handler = handler;
2606                 t->neigh_vars[17].strategy = strategy;
2607                 t->neigh_vars[17].extra1 = dev;
2608         }
2609
2610         dev_name = kstrdup(dev_name_source, GFP_KERNEL);
2611         if (!dev_name) {
2612                 err = -ENOBUFS;
2613                 goto free;
2614         }
2615
2616         t->neigh_dev[0].procname = dev_name;
2617
2618         t->neigh_neigh_dir[0].ctl_name = pdev_id;
2619
2620         t->neigh_proto_dir[0].procname = p_name;
2621         t->neigh_proto_dir[0].ctl_name = p_id;
2622
2623         t->neigh_dev[0].child          = t->neigh_vars;
2624         t->neigh_neigh_dir[0].child    = t->neigh_dev;
2625         t->neigh_proto_dir[0].child    = t->neigh_neigh_dir;
2626         t->neigh_root_dir[0].child     = t->neigh_proto_dir;
2627
2628         t->sysctl_header = register_sysctl_table(t->neigh_root_dir, 0);
2629         if (!t->sysctl_header) {
2630                 err = -ENOBUFS;
2631                 goto free_procname;
2632         }
2633         p->sysctl_table = t;
2634         return 0;
2635
2636         /* error path */
2637  free_procname:
2638         kfree(dev_name);
2639  free:
2640         kfree(t);
2641
2642         return err;
2643 }
2644
2645 void neigh_sysctl_unregister(struct neigh_parms *p)
2646 {
2647         if (p->sysctl_table) {
2648                 struct neigh_sysctl_table *t = p->sysctl_table;
2649                 p->sysctl_table = NULL;
2650                 unregister_sysctl_table(t->sysctl_header);
2651                 kfree(t->neigh_dev[0].procname);
2652                 kfree(t);
2653         }
2654 }
2655
2656 #endif  /* CONFIG_SYSCTL */
2657
2658 EXPORT_SYMBOL(__neigh_event_send);
2659 EXPORT_SYMBOL(neigh_add);
2660 EXPORT_SYMBOL(neigh_changeaddr);
2661 EXPORT_SYMBOL(neigh_compat_output);
2662 EXPORT_SYMBOL(neigh_connected_output);
2663 EXPORT_SYMBOL(neigh_create);
2664 EXPORT_SYMBOL(neigh_delete);
2665 EXPORT_SYMBOL(neigh_destroy);
2666 EXPORT_SYMBOL(neigh_dump_info);
2667 EXPORT_SYMBOL(neigh_event_ns);
2668 EXPORT_SYMBOL(neigh_ifdown);
2669 EXPORT_SYMBOL(neigh_lookup);
2670 EXPORT_SYMBOL(neigh_lookup_nodev);
2671 EXPORT_SYMBOL(neigh_parms_alloc);
2672 EXPORT_SYMBOL(neigh_parms_release);
2673 EXPORT_SYMBOL(neigh_rand_reach_time);
2674 EXPORT_SYMBOL(neigh_resolve_output);
2675 EXPORT_SYMBOL(neigh_table_clear);
2676 EXPORT_SYMBOL(neigh_table_init);
2677 EXPORT_SYMBOL(neigh_table_init_no_netlink);
2678 EXPORT_SYMBOL(neigh_update);
2679 EXPORT_SYMBOL(neigh_update_hhs);
2680 EXPORT_SYMBOL(pneigh_enqueue);
2681 EXPORT_SYMBOL(pneigh_lookup);
2682 EXPORT_SYMBOL(neightbl_dump_info);
2683 EXPORT_SYMBOL(neightbl_set);
2684
2685 #ifdef CONFIG_ARPD
2686 EXPORT_SYMBOL(neigh_app_ns);
2687 #endif
2688 #ifdef CONFIG_SYSCTL
2689 EXPORT_SYMBOL(neigh_sysctl_register);
2690 EXPORT_SYMBOL(neigh_sysctl_unregister);
2691 #endif