net/core/neighbour.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  *      Generic address resolution entity
   4  *
   5  *      Authors:
   6  *      Pedro Roque             <roque@di.fc.ul.pt>
   7  *      Alexey Kuznetsov        <kuznet@ms2.inr.ac.ru>
   8  *
   9  *      Fixes:
  10  *      Vitaly E. Lavrov        releasing NULL neighbor in neigh_add.
  11  *      Harald Welte            Add neighbour cache statistics like rtstat
  12  */
  13
  14 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  15
  16 #include <linux/slab.h>
  17 #include <linux/kmemleak.h>
  18 #include <linux/types.h>
  19 #include <linux/kernel.h>
  20 #include <linux/module.h>
  21 #include <linux/socket.h>
  22 #include <linux/netdevice.h>
  23 #include <linux/proc_fs.h>
  24 #ifdef CONFIG_SYSCTL
  25 #include <linux/sysctl.h>
  26 #endif
  27 #include <linux/times.h>
  28 #include <net/net_namespace.h>
  29 #include <net/neighbour.h>
  30 #include <net/arp.h>
  31 #include <net/dst.h>
  32 #include <net/sock.h>
  33 #include <net/netevent.h>
  34 #include <net/netlink.h>
  35 #include <linux/rtnetlink.h>
  36 #include <linux/random.h>
  37 #include <linux/string.h>
  38 #include <linux/log2.h>
  39 #include <linux/inetdevice.h>
  40 #include <net/addrconf.h>
  41
  42 #include <trace/events/neigh.h>
  43
  44 #define NEIGH_DEBUG 1
  45 #define neigh_dbg(level, fmt, ...)              \
  46 do {                                            \
  47         if (level <= NEIGH_DEBUG)               \
  48                 pr_debug(fmt, ##__VA_ARGS__);   \
  49 } while (0)
  50
  51 #define PNEIGH_HASHMASK         0xF
  52
  53 static void neigh_timer_handler(struct timer_list *t);
  54 static void __neigh_notify(struct neighbour *n, int type, int flags,
  55                            u32 pid);
  56 static void neigh_update_notify(struct neighbour *neigh, u32 nlmsg_pid);
  57 static int pneigh_ifdown_and_unlock(struct neigh_table *tbl,
  58                                     struct net_device *dev);
  59
  60 #ifdef CONFIG_PROC_FS
  61 static const struct seq_operations neigh_stat_seq_ops;
  62 #endif
  63
  64 /*
  65    Neighbour hash table buckets are protected with rwlock tbl->lock.
  66
  67    - All the scans/updates to hash buckets MUST be made under this lock.
  68    - NOTHING clever should be made under this lock: no callbacks
  69      to protocol backends, no attempts to send something to network.
  70      It will result in deadlocks, if backend/driver wants to use neighbour
  71      cache.
  72    - If the entry requires some non-trivial actions, increase
  73      its reference count and release table lock.
  74
  75    Neighbour entries are protected:
  76    - with reference count.
  77    - with rwlock neigh->lock
  78
  79    Reference count prevents destruction.
  80
  81    neigh->lock mainly serializes ll address data and its validity state.
  82    However, the same lock is used to protect another entry fields:
  83     - timer
  84     - resolution queue
  85
  86    Again, nothing clever shall be made under neigh->lock,
  87    the most complicated procedure, which we allow is dev->hard_header.
  88    It is supposed, that dev->hard_header is simplistic and does
  89    not make callbacks to neighbour tables.
  90  */
  91
  92 static int neigh_blackhole(struct neighbour *neigh, struct sk_buff *skb)
  93 {
  94         kfree_skb(skb);
  95         return -ENETDOWN;
  96 }
  97
  98 static void neigh_cleanup_and_release(struct neighbour *neigh)
  99 {
 100         trace_neigh_cleanup_and_release(neigh, 0);
 101         __neigh_notify(neigh, RTM_DELNEIGH, 0, 0);
 102         call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, neigh);
 103         neigh_release(neigh);
 104 }
 105
 106 /*
 107  * It is random distribution in the interval (1/2)*base...(3/2)*base.
 108  * It corresponds to default IPv6 settings and is not overridable,
 109  * because it is really reasonable choice.
 110  */
 111
 112 unsigned long neigh_rand_reach_time(unsigned long base)
 113 {
 114         return base ? get_random_u32_below(base) + (base >> 1) : 0;
 115 }
 116 EXPORT_SYMBOL(neigh_rand_reach_time);
 117
 118 static void neigh_mark_dead(struct neighbour *n)
 119 {
 120         n->dead = 1;
 121         if (!list_empty(&n->gc_list)) {
 122                 list_del_init(&n->gc_list);
 123                 atomic_dec(&n->tbl->gc_entries);
 124         }
 125         if (!list_empty(&n->managed_list))
 126                 list_del_init(&n->managed_list);
 127 }
 128
 129 static void neigh_update_gc_list(struct neighbour *n)
 130 {
 131         bool on_gc_list, exempt_from_gc;
 132
 133         write_lock_bh(&n->tbl->lock);
 134         write_lock(&n->lock);
 135         if (n->dead)
 136                 goto out;
 137
 138         /* remove from the gc list if new state is permanent or if neighbor
 139          * is externally learned; otherwise entry should be on the gc list
 140          */
 141         exempt_from_gc = n->nud_state & NUD_PERMANENT ||
 142                          n->flags & NTF_EXT_LEARNED;
 143         on_gc_list = !list_empty(&n->gc_list);
 144
 145         if (exempt_from_gc && on_gc_list) {
 146                 list_del_init(&n->gc_list);
 147                 atomic_dec(&n->tbl->gc_entries);
 148         } else if (!exempt_from_gc && !on_gc_list) {
 149                 /* add entries to the tail; cleaning removes from the front */
 150                 list_add_tail(&n->gc_list, &n->tbl->gc_list);
 151                 atomic_inc(&n->tbl->gc_entries);
 152         }
 153 out:
 154         write_unlock(&n->lock);
 155         write_unlock_bh(&n->tbl->lock);
 156 }
 157
 158 static void neigh_update_managed_list(struct neighbour *n)
 159 {
 160         bool on_managed_list, add_to_managed;
 161
 162         write_lock_bh(&n->tbl->lock);
 163         write_lock(&n->lock);
 164         if (n->dead)
 165                 goto out;
 166
 167         add_to_managed = n->flags & NTF_MANAGED;
 168         on_managed_list = !list_empty(&n->managed_list);
 169
 170         if (!add_to_managed && on_managed_list)
 171                 list_del_init(&n->managed_list);
 172         else if (add_to_managed && !on_managed_list)
 173                 list_add_tail(&n->managed_list, &n->tbl->managed_list);
 174 out:
 175         write_unlock(&n->lock);
 176         write_unlock_bh(&n->tbl->lock);
 177 }
 178
 179 static void neigh_update_flags(struct neighbour *neigh, u32 flags, int *notify,
 180                                bool *gc_update, bool *managed_update)
 181 {
 182         u32 ndm_flags, old_flags = neigh->flags;
 183
 184         if (!(flags & NEIGH_UPDATE_F_ADMIN))
 185                 return;
 186
 187         ndm_flags  = (flags & NEIGH_UPDATE_F_EXT_LEARNED) ? NTF_EXT_LEARNED : 0;
 188         ndm_flags |= (flags & NEIGH_UPDATE_F_MANAGED) ? NTF_MANAGED : 0;
 189
 190         if ((old_flags ^ ndm_flags) & NTF_EXT_LEARNED) {
 191                 if (ndm_flags & NTF_EXT_LEARNED)
 192                         neigh->flags |= NTF_EXT_LEARNED;
 193                 else
 194                         neigh->flags &= ~NTF_EXT_LEARNED;
 195                 *notify = 1;
 196                 *gc_update = true;
 197         }
 198         if ((old_flags ^ ndm_flags) & NTF_MANAGED) {
 199                 if (ndm_flags & NTF_MANAGED)
 200                         neigh->flags |= NTF_MANAGED;
 201                 else
 202                         neigh->flags &= ~NTF_MANAGED;
 203                 *notify = 1;
 204                 *managed_update = true;
 205         }
 206 }
 207
 208 static bool neigh_del(struct neighbour *n, struct neighbour __rcu **np,
 209                       struct neigh_table *tbl)
 210 {
 211         bool retval = false;
 212
 213         write_lock(&n->lock);
 214         if (refcount_read(&n->refcnt) == 1) {
 215                 struct neighbour *neigh;
 216
 217                 neigh = rcu_dereference_protected(n->next,
 218                                                   lockdep_is_held(&tbl->lock));
 219                 rcu_assign_pointer(*np, neigh);
 220                 neigh_mark_dead(n);
 221                 retval = true;
 222         }
 223         write_unlock(&n->lock);
 224         if (retval)
 225                 neigh_cleanup_and_release(n);
 226         return retval;
 227 }
 228
 229 bool neigh_remove_one(struct neighbour *ndel, struct neigh_table *tbl)
 230 {
 231         struct neigh_hash_table *nht;
 232         void *pkey = ndel->primary_key;
 233         u32 hash_val;
 234         struct neighbour *n;
 235         struct neighbour __rcu **np;
 236
 237         nht = rcu_dereference_protected(tbl->nht,
 238                                         lockdep_is_held(&tbl->lock));
 239         hash_val = tbl->hash(pkey, ndel->dev, nht->hash_rnd);
 240         hash_val = hash_val >> (32 - nht->hash_shift);
 241
 242         np = &nht->hash_buckets[hash_val];
 243         while ((n = rcu_dereference_protected(*np,
 244                                               lockdep_is_held(&tbl->lock)))) {
 245                 if (n == ndel)
 246                         return neigh_del(n, np, tbl);
 247                 np = &n->next;
 248         }
 249         return false;
 250 }
 251
 252 static int neigh_forced_gc(struct neigh_table *tbl)
 253 {
 254         int max_clean = atomic_read(&tbl->gc_entries) -
 255                         READ_ONCE(tbl->gc_thresh2);
 256         u64 tmax = ktime_get_ns() + NSEC_PER_MSEC;
 257         unsigned long tref = jiffies - 5 * HZ;
 258         struct neighbour *n, *tmp;
 259         int shrunk = 0;
 260         int loop = 0;
 261
 262         NEIGH_CACHE_STAT_INC(tbl, forced_gc_runs);
 263
 264         write_lock_bh(&tbl->lock);
 265
 266         list_for_each_entry_safe(n, tmp, &tbl->gc_list, gc_list) {
 267                 if (refcount_read(&n->refcnt) == 1) {
 268                         bool remove = false;
 269
 270                         write_lock(&n->lock);
 271                         if ((n->nud_state == NUD_FAILED) ||
 272                             (n->nud_state == NUD_NOARP) ||
 273                             (tbl->is_multicast &&
 274                              tbl->is_multicast(n->primary_key)) ||
 275                             !time_in_range(n->updated, tref, jiffies))
 276                                 remove = true;
 277                         write_unlock(&n->lock);
 278
 279                         if (remove && neigh_remove_one(n, tbl))
 280                                 shrunk++;
 281                         if (shrunk >= max_clean)
 282                                 break;
 283                         if (++loop == 16) {
 284                                 if (ktime_get_ns() > tmax)
 285                                         goto unlock;
 286                                 loop = 0;
 287                         }
 288                 }
 289         }
 290
 291         WRITE_ONCE(tbl->last_flush, jiffies);
 292 unlock:
 293         write_unlock_bh(&tbl->lock);
 294
 295         return shrunk;
 296 }
 297
 298 static void neigh_add_timer(struct neighbour *n, unsigned long when)
 299 {
 300         /* Use safe distance from the jiffies - LONG_MAX point while timer
 301          * is running in DELAY/PROBE state but still show to user space
 302          * large times in the past.
 303          */
 304         unsigned long mint = jiffies - (LONG_MAX - 86400 * HZ);
 305
 306         neigh_hold(n);
 307         if (!time_in_range(n->confirmed, mint, jiffies))
 308                 n->confirmed = mint;
 309         if (time_before(n->used, n->confirmed))
 310                 n->used = n->confirmed;
 311         if (unlikely(mod_timer(&n->timer, when))) {
 312                 printk("NEIGH: BUG, double timer add, state is %x\n",
 313                        n->nud_state);
 314                 dump_stack();
 315         }
 316 }
 317
 318 static int neigh_del_timer(struct neighbour *n)
 319 {
 320         if ((n->nud_state & NUD_IN_TIMER) &&
 321             del_timer(&n->timer)) {
 322                 neigh_release(n);
 323                 return 1;
 324         }
 325         return 0;
 326 }
 327
 328 static struct neigh_parms *neigh_get_dev_parms_rcu(struct net_device *dev,
 329                                                    int family)
 330 {
 331         switch (family) {
 332         case AF_INET:
 333                 return __in_dev_arp_parms_get_rcu(dev);
 334         case AF_INET6:
 335                 return __in6_dev_nd_parms_get_rcu(dev);
 336         }
 337         return NULL;
 338 }
 339
 340 static void neigh_parms_qlen_dec(struct net_device *dev, int family)
 341 {
 342         struct neigh_parms *p;
 343
 344         rcu_read_lock();
 345         p = neigh_get_dev_parms_rcu(dev, family);
 346         if (p)
 347                 p->qlen--;
 348         rcu_read_unlock();
 349 }
 350
 351 static void pneigh_queue_purge(struct sk_buff_head *list, struct net *net,
 352                                int family)
 353 {
 354         struct sk_buff_head tmp;
 355         unsigned long flags;
 356         struct sk_buff *skb;
 357
 358         skb_queue_head_init(&tmp);
 359         spin_lock_irqsave(&list->lock, flags);
 360         skb = skb_peek(list);
 361         while (skb != NULL) {
 362                 struct sk_buff *skb_next = skb_peek_next(skb, list);
 363                 struct net_device *dev = skb->dev;
 364
 365                 if (net == NULL || net_eq(dev_net(dev), net)) {
 366                         neigh_parms_qlen_dec(dev, family);
 367                         __skb_unlink(skb, list);
 368                         __skb_queue_tail(&tmp, skb);
 369                 }
 370                 skb = skb_next;
 371         }
 372         spin_unlock_irqrestore(&list->lock, flags);
 373
 374         while ((skb = __skb_dequeue(&tmp))) {
 375                 dev_put(skb->dev);
 376                 kfree_skb(skb);
 377         }
 378 }
 379
 380 static void neigh_flush_dev(struct neigh_table *tbl, struct net_device *dev,
 381                             bool skip_perm)
 382 {
 383         int i;
 384         struct neigh_hash_table *nht;
 385
 386         nht = rcu_dereference_protected(tbl->nht,
 387                                         lockdep_is_held(&tbl->lock));
 388
 389         for (i = 0; i < (1 << nht->hash_shift); i++) {
 390                 struct neighbour *n;
 391                 struct neighbour __rcu **np = &nht->hash_buckets[i];
 392
 393                 while ((n = rcu_dereference_protected(*np,
 394                                         lockdep_is_held(&tbl->lock))) != NULL) {
 395                         if (dev && n->dev != dev) {
 396                                 np = &n->next;
 397                                 continue;
 398                         }
 399                         if (skip_perm && n->nud_state & NUD_PERMANENT) {
 400                                 np = &n->next;
 401                                 continue;
 402                         }
 403                         rcu_assign_pointer(*np,
 404                                    rcu_dereference_protected(n->next,
 405                                                 lockdep_is_held(&tbl->lock)));
 406                         write_lock(&n->lock);
 407                         neigh_del_timer(n);
 408                         neigh_mark_dead(n);
 409                         if (refcount_read(&n->refcnt) != 1) {
 410                                 /* The most unpleasant situation.
 411                                    We must destroy neighbour entry,
 412                                    but someone still uses it.
 413
 414                                    The destroy will be delayed until
 415                                    the last user releases us, but
 416                                    we must kill timers etc. and move
 417                                    it to safe state.
 418                                  */
 419                                 __skb_queue_purge(&n->arp_queue);
 420                                 n->arp_queue_len_bytes = 0;
 421                                 WRITE_ONCE(n->output, neigh_blackhole);
 422                                 if (n->nud_state & NUD_VALID)
 423                                         n->nud_state = NUD_NOARP;
 424                                 else
 425                                         n->nud_state = NUD_NONE;
 426                                 neigh_dbg(2, "neigh %p is stray\n", n);
 427                         }
 428                         write_unlock(&n->lock);
 429                         neigh_cleanup_and_release(n);
 430                 }
 431         }
 432 }
 433
 434 void neigh_changeaddr(struct neigh_table *tbl, struct net_device *dev)
 435 {
 436         write_lock_bh(&tbl->lock);
 437         neigh_flush_dev(tbl, dev, false);
 438         write_unlock_bh(&tbl->lock);
 439 }
 440 EXPORT_SYMBOL(neigh_changeaddr);
 441
 442 static int __neigh_ifdown(struct neigh_table *tbl, struct net_device *dev,
 443                           bool skip_perm)
 444 {
 445         write_lock_bh(&tbl->lock);
 446         neigh_flush_dev(tbl, dev, skip_perm);
 447         pneigh_ifdown_and_unlock(tbl, dev);
 448         pneigh_queue_purge(&tbl->proxy_queue, dev ? dev_net(dev) : NULL,
 449                            tbl->family);
 450         if (skb_queue_empty_lockless(&tbl->proxy_queue))
 451                 del_timer_sync(&tbl->proxy_timer);
 452         return 0;
 453 }
 454
 455 int neigh_carrier_down(struct neigh_table *tbl, struct net_device *dev)
 456 {
 457         __neigh_ifdown(tbl, dev, true);
 458         return 0;
 459 }
 460 EXPORT_SYMBOL(neigh_carrier_down);
 461
 462 int neigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
 463 {
 464         __neigh_ifdown(tbl, dev, false);
 465         return 0;
 466 }
 467 EXPORT_SYMBOL(neigh_ifdown);
 468
 469 static struct neighbour *neigh_alloc(struct neigh_table *tbl,
 470                                      struct net_device *dev,
 471                                      u32 flags, bool exempt_from_gc)
 472 {
 473         struct neighbour *n = NULL;
 474         unsigned long now = jiffies;
 475         int entries, gc_thresh3;
 476
 477         if (exempt_from_gc)
 478                 goto do_alloc;
 479
 480         entries = atomic_inc_return(&tbl->gc_entries) - 1;
 481         gc_thresh3 = READ_ONCE(tbl->gc_thresh3);
 482         if (entries >= gc_thresh3 ||
 483             (entries >= READ_ONCE(tbl->gc_thresh2) &&
 484              time_after(now, READ_ONCE(tbl->last_flush) + 5 * HZ))) {
 485                 if (!neigh_forced_gc(tbl) && entries >= gc_thresh3) {
 486                         net_info_ratelimited("%s: neighbor table overflow!\n",
 487                                              tbl->id);
 488                         NEIGH_CACHE_STAT_INC(tbl, table_fulls);
 489                         goto out_entries;
 490                 }
 491         }
 492
 493 do_alloc:
 494         n = kzalloc(tbl->entry_size + dev->neigh_priv_len, GFP_ATOMIC);
 495         if (!n)
 496                 goto out_entries;
 497
 498         __skb_queue_head_init(&n->arp_queue);
 499         rwlock_init(&n->lock);
 500         seqlock_init(&n->ha_lock);
 501         n->updated        = n->used = now;
 502         n->nud_state      = NUD_NONE;
 503         n->output         = neigh_blackhole;
 504         n->flags          = flags;
 505         seqlock_init(&n->hh.hh_lock);
 506         n->parms          = neigh_parms_clone(&tbl->parms);
 507         timer_setup(&n->timer, neigh_timer_handler, 0);
 508
 509         NEIGH_CACHE_STAT_INC(tbl, allocs);
 510         n->tbl            = tbl;
 511         refcount_set(&n->refcnt, 1);
 512         n->dead           = 1;
 513         INIT_LIST_HEAD(&n->gc_list);
 514         INIT_LIST_HEAD(&n->managed_list);
 515
 516         atomic_inc(&tbl->entries);
 517 out:
 518         return n;
 519
 520 out_entries:
 521         if (!exempt_from_gc)
 522                 atomic_dec(&tbl->gc_entries);
 523         goto out;
 524 }
 525
 526 static void neigh_get_hash_rnd(u32 *x)
 527 {
 528         *x = get_random_u32() | 1;
 529 }
 530
 531 static struct neigh_hash_table *neigh_hash_alloc(unsigned int shift)
 532 {
 533         size_t size = (1 << shift) * sizeof(struct neighbour *);
 534         struct neigh_hash_table *ret;
 535         struct neighbour __rcu **buckets;
 536         int i;
 537
 538         ret = kmalloc(sizeof(*ret), GFP_ATOMIC);
 539         if (!ret)
 540                 return NULL;
 541         if (size <= PAGE_SIZE) {
 542                 buckets = kzalloc(size, GFP_ATOMIC);
 543         } else {
 544                 buckets = (struct neighbour __rcu **)
 545                           __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
 546                                            get_order(size));
 547                 kmemleak_alloc(buckets, size, 1, GFP_ATOMIC);
 548         }
 549         if (!buckets) {
 550                 kfree(ret);
 551                 return NULL;
 552         }
 553         ret->hash_buckets = buckets;
 554         ret->hash_shift = shift;
 555         for (i = 0; i < NEIGH_NUM_HASH_RND; i++)
 556                 neigh_get_hash_rnd(&ret->hash_rnd[i]);
 557         return ret;
 558 }
 559
 560 static void neigh_hash_free_rcu(struct rcu_head *head)
 561 {
 562         struct neigh_hash_table *nht = container_of(head,
 563                                                     struct neigh_hash_table,
 564                                                     rcu);
 565         size_t size = (1 << nht->hash_shift) * sizeof(struct neighbour *);
 566         struct neighbour __rcu **buckets = nht->hash_buckets;
 567
 568         if (size <= PAGE_SIZE) {
 569                 kfree(buckets);
 570         } else {
 571                 kmemleak_free(buckets);
 572                 free_pages((unsigned long)buckets, get_order(size));
 573         }
 574         kfree(nht);
 575 }
 576
 577 static struct neigh_hash_table *neigh_hash_grow(struct neigh_table *tbl,
 578                                                 unsigned long new_shift)
 579 {
 580         unsigned int i, hash;
 581         struct neigh_hash_table *new_nht, *old_nht;
 582
 583         NEIGH_CACHE_STAT_INC(tbl, hash_grows);
 584
 585         old_nht = rcu_dereference_protected(tbl->nht,
 586                                             lockdep_is_held(&tbl->lock));
 587         new_nht = neigh_hash_alloc(new_shift);
 588         if (!new_nht)
 589                 return old_nht;
 590
 591         for (i = 0; i < (1 << old_nht->hash_shift); i++) {
 592                 struct neighbour *n, *next;
 593
 594                 for (n = rcu_dereference_protected(old_nht->hash_buckets[i],
 595                                                    lockdep_is_held(&tbl->lock));
 596                      n != NULL;
 597                      n = next) {
 598                         hash = tbl->hash(n->primary_key, n->dev,
 599                                          new_nht->hash_rnd);
 600
 601                         hash >>= (32 - new_nht->hash_shift);
 602                         next = rcu_dereference_protected(n->next,
 603                                                 lockdep_is_held(&tbl->lock));
 604
 605                         rcu_assign_pointer(n->next,
 606                                            rcu_dereference_protected(
 607                                                 new_nht->hash_buckets[hash],
 608                                                 lockdep_is_held(&tbl->lock)));
 609                         rcu_assign_pointer(new_nht->hash_buckets[hash], n);
 610                 }
 611         }
 612
 613         rcu_assign_pointer(tbl->nht, new_nht);
 614         call_rcu(&old_nht->rcu, neigh_hash_free_rcu);
 615         return new_nht;
 616 }
 617
 618 struct neighbour *neigh_lookup(struct neigh_table *tbl, const void *pkey,
 619                                struct net_device *dev)
 620 {
 621         struct neighbour *n;
 622
 623         NEIGH_CACHE_STAT_INC(tbl, lookups);
 624
 625         rcu_read_lock();
 626         n = __neigh_lookup_noref(tbl, pkey, dev);
 627         if (n) {
 628                 if (!refcount_inc_not_zero(&n->refcnt))
 629                         n = NULL;
 630                 NEIGH_CACHE_STAT_INC(tbl, hits);
 631         }
 632
 633         rcu_read_unlock();
 634         return n;
 635 }
 636 EXPORT_SYMBOL(neigh_lookup);
 637
 638 static struct neighbour *
 639 ___neigh_create(struct neigh_table *tbl, const void *pkey,
 640                 struct net_device *dev, u32 flags,
 641                 bool exempt_from_gc, bool want_ref)
 642 {
 643         u32 hash_val, key_len = tbl->key_len;
 644         struct neighbour *n1, *rc, *n;
 645         struct neigh_hash_table *nht;
 646         int error;
 647
 648         n = neigh_alloc(tbl, dev, flags, exempt_from_gc);
 649         trace_neigh_create(tbl, dev, pkey, n, exempt_from_gc);
 650         if (!n) {
 651                 rc = ERR_PTR(-ENOBUFS);
 652                 goto out;
 653         }
 654
 655         memcpy(n->primary_key, pkey, key_len);
 656         n->dev = dev;
 657         netdev_hold(dev, &n->dev_tracker, GFP_ATOMIC);
 658
 659         /* Protocol specific setup. */
 660         if (tbl->constructor && (error = tbl->constructor(n)) < 0) {
 661                 rc = ERR_PTR(error);
 662                 goto out_neigh_release;
 663         }
 664
 665         if (dev->netdev_ops->ndo_neigh_construct) {
 666                 error = dev->netdev_ops->ndo_neigh_construct(dev, n);
 667                 if (error < 0) {
 668                         rc = ERR_PTR(error);
 669                         goto out_neigh_release;
 670                 }
 671         }
 672
 673         /* Device specific setup. */
 674         if (n->parms->neigh_setup &&
 675             (error = n->parms->neigh_setup(n)) < 0) {
 676                 rc = ERR_PTR(error);
 677                 goto out_neigh_release;
 678         }
 679
 680         n->confirmed = jiffies - (NEIGH_VAR(n->parms, BASE_REACHABLE_TIME) << 1);
 681
 682         write_lock_bh(&tbl->lock);
 683         nht = rcu_dereference_protected(tbl->nht,
 684                                         lockdep_is_held(&tbl->lock));
 685
 686         if (atomic_read(&tbl->entries) > (1 << nht->hash_shift))
 687                 nht = neigh_hash_grow(tbl, nht->hash_shift + 1);
 688
 689         hash_val = tbl->hash(n->primary_key, dev, nht->hash_rnd) >> (32 - nht->hash_shift);
 690
 691         if (n->parms->dead) {
 692                 rc = ERR_PTR(-EINVAL);
 693                 goto out_tbl_unlock;
 694         }
 695
 696         for (n1 = rcu_dereference_protected(nht->hash_buckets[hash_val],
 697                                             lockdep_is_held(&tbl->lock));
 698              n1 != NULL;
 699              n1 = rcu_dereference_protected(n1->next,
 700                         lockdep_is_held(&tbl->lock))) {
 701                 if (dev == n1->dev && !memcmp(n1->primary_key, n->primary_key, key_len)) {
 702                         if (want_ref)
 703                                 neigh_hold(n1);
 704                         rc = n1;
 705                         goto out_tbl_unlock;
 706                 }
 707         }
 708
 709         n->dead = 0;
 710         if (!exempt_from_gc)
 711                 list_add_tail(&n->gc_list, &n->tbl->gc_list);
 712         if (n->flags & NTF_MANAGED)
 713                 list_add_tail(&n->managed_list, &n->tbl->managed_list);
 714         if (want_ref)
 715                 neigh_hold(n);
 716         rcu_assign_pointer(n->next,
 717                            rcu_dereference_protected(nht->hash_buckets[hash_val],
 718                                                      lockdep_is_held(&tbl->lock)));
 719         rcu_assign_pointer(nht->hash_buckets[hash_val], n);
 720         write_unlock_bh(&tbl->lock);
 721         neigh_dbg(2, "neigh %p is created\n", n);
 722         rc = n;
 723 out:
 724         return rc;
 725 out_tbl_unlock:
 726         write_unlock_bh(&tbl->lock);
 727 out_neigh_release:
 728         if (!exempt_from_gc)
 729                 atomic_dec(&tbl->gc_entries);
 730         neigh_release(n);
 731         goto out;
 732 }
 733
 734 struct neighbour *__neigh_create(struct neigh_table *tbl, const void *pkey,
 735                                  struct net_device *dev, bool want_ref)
 736 {
 737         return ___neigh_create(tbl, pkey, dev, 0, false, want_ref);
 738 }
 739 EXPORT_SYMBOL(__neigh_create);
 740
 741 static u32 pneigh_hash(const void *pkey, unsigned int key_len)
 742 {
 743         u32 hash_val = *(u32 *)(pkey + key_len - 4);
 744         hash_val ^= (hash_val >> 16);
 745         hash_val ^= hash_val >> 8;
 746         hash_val ^= hash_val >> 4;
 747         hash_val &= PNEIGH_HASHMASK;
 748         return hash_val;
 749 }
 750
 751 static struct pneigh_entry *__pneigh_lookup_1(struct pneigh_entry *n,
 752                                               struct net *net,
 753                                               const void *pkey,
 754                                               unsigned int key_len,
 755                                               struct net_device *dev)
 756 {
 757         while (n) {
 758                 if (!memcmp(n->key, pkey, key_len) &&
 759                     net_eq(pneigh_net(n), net) &&
 760                     (n->dev == dev || !n->dev))
 761                         return n;
 762                 n = n->next;
 763         }
 764         return NULL;
 765 }
 766
 767 struct pneigh_entry *__pneigh_lookup(struct neigh_table *tbl,
 768                 struct net *net, const void *pkey, struct net_device *dev)
 769 {
 770         unsigned int key_len = tbl->key_len;
 771         u32 hash_val = pneigh_hash(pkey, key_len);
 772
 773         return __pneigh_lookup_1(tbl->phash_buckets[hash_val],
 774                                  net, pkey, key_len, dev);
 775 }
 776 EXPORT_SYMBOL_GPL(__pneigh_lookup);
 777
 778 struct pneigh_entry * pneigh_lookup(struct neigh_table *tbl,
 779                                     struct net *net, const void *pkey,
 780                                     struct net_device *dev, int creat)
 781 {
 782         struct pneigh_entry *n;
 783         unsigned int key_len = tbl->key_len;
 784         u32 hash_val = pneigh_hash(pkey, key_len);
 785
 786         read_lock_bh(&tbl->lock);
 787         n = __pneigh_lookup_1(tbl->phash_buckets[hash_val],
 788                               net, pkey, key_len, dev);
 789         read_unlock_bh(&tbl->lock);
 790
 791         if (n || !creat)
 792                 goto out;
 793
 794         ASSERT_RTNL();
 795
 796         n = kzalloc(sizeof(*n) + key_len, GFP_KERNEL);
 797         if (!n)
 798                 goto out;
 799
 800         write_pnet(&n->net, net);
 801         memcpy(n->key, pkey, key_len);
 802         n->dev = dev;
 803         netdev_hold(dev, &n->dev_tracker, GFP_KERNEL);
 804
 805         if (tbl->pconstructor && tbl->pconstructor(n)) {
 806                 netdev_put(dev, &n->dev_tracker);
 807                 kfree(n);
 808                 n = NULL;
 809                 goto out;
 810         }
 811
 812         write_lock_bh(&tbl->lock);
 813         n->next = tbl->phash_buckets[hash_val];
 814         tbl->phash_buckets[hash_val] = n;
 815         write_unlock_bh(&tbl->lock);
 816 out:
 817         return n;
 818 }
 819 EXPORT_SYMBOL(pneigh_lookup);
 820
 821
 822 int pneigh_delete(struct neigh_table *tbl, struct net *net, const void *pkey,
 823                   struct net_device *dev)
 824 {
 825         struct pneigh_entry *n, **np;
 826         unsigned int key_len = tbl->key_len;
 827         u32 hash_val = pneigh_hash(pkey, key_len);
 828
 829         write_lock_bh(&tbl->lock);
 830         for (np = &tbl->phash_buckets[hash_val]; (n = *np) != NULL;
 831              np = &n->next) {
 832                 if (!memcmp(n->key, pkey, key_len) && n->dev == dev &&
 833                     net_eq(pneigh_net(n), net)) {
 834                         *np = n->next;
 835                         write_unlock_bh(&tbl->lock);
 836                         if (tbl->pdestructor)
 837                                 tbl->pdestructor(n);
 838                         netdev_put(n->dev, &n->dev_tracker);
 839                         kfree(n);
 840                         return 0;
 841                 }
 842         }
 843         write_unlock_bh(&tbl->lock);
 844         return -ENOENT;
 845 }
 846
 847 static int pneigh_ifdown_and_unlock(struct neigh_table *tbl,
 848                                     struct net_device *dev)
 849 {
 850         struct pneigh_entry *n, **np, *freelist = NULL;
 851         u32 h;
 852
 853         for (h = 0; h <= PNEIGH_HASHMASK; h++) {
 854                 np = &tbl->phash_buckets[h];
 855                 while ((n = *np) != NULL) {
 856                         if (!dev || n->dev == dev) {
 857                                 *np = n->next;
 858                                 n->next = freelist;
 859                                 freelist = n;
 860                                 continue;
 861                         }
 862                         np = &n->next;
 863                 }
 864         }
 865         write_unlock_bh(&tbl->lock);
 866         while ((n = freelist)) {
 867                 freelist = n->next;
 868                 n->next = NULL;
 869                 if (tbl->pdestructor)
 870                         tbl->pdestructor(n);
 871                 netdev_put(n->dev, &n->dev_tracker);
 872                 kfree(n);
 873         }
 874         return -ENOENT;
 875 }
 876
 877 static void neigh_parms_destroy(struct neigh_parms *parms);
 878
 879 static inline void neigh_parms_put(struct neigh_parms *parms)
 880 {
 881         if (refcount_dec_and_test(&parms->refcnt))
 882                 neigh_parms_destroy(parms);
 883 }
 884
 885 /*
 886  *      neighbour must already be out of the table;
 887  *
 888  */
 889 void neigh_destroy(struct neighbour *neigh)
 890 {
 891         struct net_device *dev = neigh->dev;
 892
 893         NEIGH_CACHE_STAT_INC(neigh->tbl, destroys);
 894
 895         if (!neigh->dead) {
 896                 pr_warn("Destroying alive neighbour %p\n", neigh);
 897                 dump_stack();
 898                 return;
 899         }
 900
 901         if (neigh_del_timer(neigh))
 902                 pr_warn("Impossible event\n");
 903
 904         write_lock_bh(&neigh->lock);
 905         __skb_queue_purge(&neigh->arp_queue);
 906         write_unlock_bh(&neigh->lock);
 907         neigh->arp_queue_len_bytes = 0;
 908
 909         if (dev->netdev_ops->ndo_neigh_destroy)
 910                 dev->netdev_ops->ndo_neigh_destroy(dev, neigh);
 911
 912         netdev_put(dev, &neigh->dev_tracker);
 913         neigh_parms_put(neigh->parms);
 914
 915         neigh_dbg(2, "neigh %p is destroyed\n", neigh);
 916
 917         atomic_dec(&neigh->tbl->entries);
 918         kfree_rcu(neigh, rcu);
 919 }
 920 EXPORT_SYMBOL(neigh_destroy);
 921
 922 /* Neighbour state is suspicious;
 923    disable fast path.
 924
 925    Called with write_locked neigh.
 926  */
 927 static void neigh_suspect(struct neighbour *neigh)
 928 {
 929         neigh_dbg(2, "neigh %p is suspected\n", neigh);
 930
 931         WRITE_ONCE(neigh->output, neigh->ops->output);
 932 }
 933
 934 /* Neighbour state is OK;
 935    enable fast path.
 936
 937    Called with write_locked neigh.
 938  */
 939 static void neigh_connect(struct neighbour *neigh)
 940 {
 941         neigh_dbg(2, "neigh %p is connected\n", neigh);
 942
 943         WRITE_ONCE(neigh->output, neigh->ops->connected_output);
 944 }
 945
 946 static void neigh_periodic_work(struct work_struct *work)
 947 {
 948         struct neigh_table *tbl = container_of(work, struct neigh_table, gc_work.work);
 949         struct neighbour *n;
 950         struct neighbour __rcu **np;
 951         unsigned int i;
 952         struct neigh_hash_table *nht;
 953
 954         NEIGH_CACHE_STAT_INC(tbl, periodic_gc_runs);
 955
 956         write_lock_bh(&tbl->lock);
 957         nht = rcu_dereference_protected(tbl->nht,
 958                                         lockdep_is_held(&tbl->lock));
 959
 960         /*
 961          *      periodically recompute ReachableTime from random function
 962          */
 963
 964         if (time_after(jiffies, tbl->last_rand + 300 * HZ)) {
 965                 struct neigh_parms *p;
 966
 967                 WRITE_ONCE(tbl->last_rand, jiffies);
 968                 list_for_each_entry(p, &tbl->parms_list, list)
 969                         p->reachable_time =
 970                                 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
 971         }
 972
 973         if (atomic_read(&tbl->entries) < READ_ONCE(tbl->gc_thresh1))
 974                 goto out;
 975
 976         for (i = 0 ; i < (1 << nht->hash_shift); i++) {
 977                 np = &nht->hash_buckets[i];
 978
 979                 while ((n = rcu_dereference_protected(*np,
 980                                 lockdep_is_held(&tbl->lock))) != NULL) {
 981                         unsigned int state;
 982
 983                         write_lock(&n->lock);
 984
 985                         state = n->nud_state;
 986                         if ((state & (NUD_PERMANENT | NUD_IN_TIMER)) ||
 987                             (n->flags & NTF_EXT_LEARNED)) {
 988                                 write_unlock(&n->lock);
 989                                 goto next_elt;
 990                         }
 991
 992                         if (time_before(n->used, n->confirmed) &&
 993                             time_is_before_eq_jiffies(n->confirmed))
 994                                 n->used = n->confirmed;
 995
 996                         if (refcount_read(&n->refcnt) == 1 &&
 997                             (state == NUD_FAILED ||
 998                              !time_in_range_open(jiffies, n->used,
 999                                                  n->used + NEIGH_VAR(n->parms, GC_STALETIME)))) {
1000                                 rcu_assign_pointer(*np,
1001                                         rcu_dereference_protected(n->next,
1002                                                 lockdep_is_held(&tbl->lock)));
1003                                 neigh_mark_dead(n);
1004                                 write_unlock(&n->lock);
1005                                 neigh_cleanup_and_release(n);
1006                                 continue;
1007                         }
1008                         write_unlock(&n->lock);
1009
1010 next_elt:
1011                         np = &n->next;
1012                 }
1013                 /*
1014                  * It's fine to release lock here, even if hash table
1015                  * grows while we are preempted.
1016                  */
1017                 write_unlock_bh(&tbl->lock);
1018                 cond_resched();
1019                 write_lock_bh(&tbl->lock);
1020                 nht = rcu_dereference_protected(tbl->nht,
1021                                                 lockdep_is_held(&tbl->lock));
1022         }
1023 out:
1024         /* Cycle through all hash buckets every BASE_REACHABLE_TIME/2 ticks.
1025          * ARP entry timeouts range from 1/2 BASE_REACHABLE_TIME to 3/2
1026          * BASE_REACHABLE_TIME.
1027          */
1028         queue_delayed_work(system_power_efficient_wq, &tbl->gc_work,
1029                               NEIGH_VAR(&tbl->parms, BASE_REACHABLE_TIME) >> 1);
1030         write_unlock_bh(&tbl->lock);
1031 }
1032
1033 static __inline__ int neigh_max_probes(struct neighbour *n)
1034 {
1035         struct neigh_parms *p = n->parms;
1036         return NEIGH_VAR(p, UCAST_PROBES) + NEIGH_VAR(p, APP_PROBES) +
1037                (n->nud_state & NUD_PROBE ? NEIGH_VAR(p, MCAST_REPROBES) :
1038                 NEIGH_VAR(p, MCAST_PROBES));
1039 }
1040
1041 static void neigh_invalidate(struct neighbour *neigh)
1042         __releases(neigh->lock)
1043         __acquires(neigh->lock)
1044 {
1045         struct sk_buff *skb;
1046
1047         NEIGH_CACHE_STAT_INC(neigh->tbl, res_failed);
1048         neigh_dbg(2, "neigh %p is failed\n", neigh);
1049         neigh->updated = jiffies;
1050
1051         /* It is very thin place. report_unreachable is very complicated
1052            routine. Particularly, it can hit the same neighbour entry!
1053
1054            So that, we try to be accurate and avoid dead loop. --ANK
1055          */
1056         while (neigh->nud_state == NUD_FAILED &&
1057                (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
1058                 write_unlock(&neigh->lock);
1059                 neigh->ops->error_report(neigh, skb);
1060                 write_lock(&neigh->lock);
1061         }
1062         __skb_queue_purge(&neigh->arp_queue);
1063         neigh->arp_queue_len_bytes = 0;
1064 }
1065
1066 static void neigh_probe(struct neighbour *neigh)
1067         __releases(neigh->lock)
1068 {
1069         struct sk_buff *skb = skb_peek_tail(&neigh->arp_queue);
1070         /* keep skb alive even if arp_queue overflows */
1071         if (skb)
1072                 skb = skb_clone(skb, GFP_ATOMIC);
1073         write_unlock(&neigh->lock);
1074         if (neigh->ops->solicit)
1075                 neigh->ops->solicit(neigh, skb);
1076         atomic_inc(&neigh->probes);
1077         consume_skb(skb);
1078 }
1079
1080 /* Called when a timer expires for a neighbour entry. */
1081
1082 static void neigh_timer_handler(struct timer_list *t)
1083 {
1084         unsigned long now, next;
1085         struct neighbour *neigh = from_timer(neigh, t, timer);
1086         unsigned int state;
1087         int notify = 0;
1088
1089         write_lock(&neigh->lock);
1090
1091         state = neigh->nud_state;
1092         now = jiffies;
1093         next = now + HZ;
1094
1095         if (!(state & NUD_IN_TIMER))
1096                 goto out;
1097
1098         if (state & NUD_REACHABLE) {
1099                 if (time_before_eq(now,
1100                                    neigh->confirmed + neigh->parms->reachable_time)) {
1101                         neigh_dbg(2, "neigh %p is still alive\n", neigh);
1102                         next = neigh->confirmed + neigh->parms->reachable_time;
1103                 } else if (time_before_eq(now,
1104                                           neigh->used +
1105                                           NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME))) {
1106                         neigh_dbg(2, "neigh %p is delayed\n", neigh);
1107                         WRITE_ONCE(neigh->nud_state, NUD_DELAY);
1108                         neigh->updated = jiffies;
1109                         neigh_suspect(neigh);
1110                         next = now + NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME);
1111                 } else {
1112                         neigh_dbg(2, "neigh %p is suspected\n", neigh);
1113                         WRITE_ONCE(neigh->nud_state, NUD_STALE);
1114                         neigh->updated = jiffies;
1115                         neigh_suspect(neigh);
1116                         notify = 1;
1117                 }
1118         } else if (state & NUD_DELAY) {
1119                 if (time_before_eq(now,
1120                                    neigh->confirmed +
1121                                    NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME))) {
1122                         neigh_dbg(2, "neigh %p is now reachable\n", neigh);
1123                         WRITE_ONCE(neigh->nud_state, NUD_REACHABLE);
1124                         neigh->updated = jiffies;
1125                         neigh_connect(neigh);
1126                         notify = 1;
1127                         next = neigh->confirmed + neigh->parms->reachable_time;
1128                 } else {
1129                         neigh_dbg(2, "neigh %p is probed\n", neigh);
1130                         WRITE_ONCE(neigh->nud_state, NUD_PROBE);
1131                         neigh->updated = jiffies;
1132                         atomic_set(&neigh->probes, 0);
1133                         notify = 1;
1134                         next = now + max(NEIGH_VAR(neigh->parms, RETRANS_TIME),
1135                                          HZ/100);
1136                 }
1137         } else {
1138                 /* NUD_PROBE|NUD_INCOMPLETE */
1139                 next = now + max(NEIGH_VAR(neigh->parms, RETRANS_TIME), HZ/100);
1140         }
1141
1142         if ((neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) &&
1143             atomic_read(&neigh->probes) >= neigh_max_probes(neigh)) {
1144                 WRITE_ONCE(neigh->nud_state, NUD_FAILED);
1145                 notify = 1;
1146                 neigh_invalidate(neigh);
1147                 goto out;
1148         }
1149
1150         if (neigh->nud_state & NUD_IN_TIMER) {
1151                 if (time_before(next, jiffies + HZ/100))
1152                         next = jiffies + HZ/100;
1153                 if (!mod_timer(&neigh->timer, next))
1154                         neigh_hold(neigh);
1155         }
1156         if (neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) {
1157                 neigh_probe(neigh);
1158         } else {
1159 out:
1160                 write_unlock(&neigh->lock);
1161         }
1162
1163         if (notify)
1164                 neigh_update_notify(neigh, 0);
1165
1166         trace_neigh_timer_handler(neigh, 0);
1167
1168         neigh_release(neigh);
1169 }
1170
1171 int __neigh_event_send(struct neighbour *neigh, struct sk_buff *skb,
1172                        const bool immediate_ok)
1173 {
1174         int rc;
1175         bool immediate_probe = false;
1176
1177         write_lock_bh(&neigh->lock);
1178
1179         rc = 0;
1180         if (neigh->nud_state & (NUD_CONNECTED | NUD_DELAY | NUD_PROBE))
1181                 goto out_unlock_bh;
1182         if (neigh->dead)
1183                 goto out_dead;
1184
1185         if (!(neigh->nud_state & (NUD_STALE | NUD_INCOMPLETE))) {
1186                 if (NEIGH_VAR(neigh->parms, MCAST_PROBES) +
1187                     NEIGH_VAR(neigh->parms, APP_PROBES)) {
1188                         unsigned long next, now = jiffies;
1189
1190                         atomic_set(&neigh->probes,
1191                                    NEIGH_VAR(neigh->parms, UCAST_PROBES));
1192                         neigh_del_timer(neigh);
1193                         WRITE_ONCE(neigh->nud_state, NUD_INCOMPLETE);
1194                         neigh->updated = now;
1195                         if (!immediate_ok) {
1196                                 next = now + 1;
1197                         } else {
1198                                 immediate_probe = true;
1199                                 next = now + max(NEIGH_VAR(neigh->parms,
1200                                                            RETRANS_TIME),
1201                                                  HZ / 100);
1202                         }
1203                         neigh_add_timer(neigh, next);
1204                 } else {
1205                         WRITE_ONCE(neigh->nud_state, NUD_FAILED);
1206                         neigh->updated = jiffies;
1207                         write_unlock_bh(&neigh->lock);
1208
1209                         kfree_skb_reason(skb, SKB_DROP_REASON_NEIGH_FAILED);
1210                         return 1;
1211                 }
1212         } else if (neigh->nud_state & NUD_STALE) {
1213                 neigh_dbg(2, "neigh %p is delayed\n", neigh);
1214                 neigh_del_timer(neigh);
1215                 WRITE_ONCE(neigh->nud_state, NUD_DELAY);
1216                 neigh->updated = jiffies;
1217                 neigh_add_timer(neigh, jiffies +
1218                                 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME));
1219         }
1220
1221         if (neigh->nud_state == NUD_INCOMPLETE) {
1222                 if (skb) {
1223                         while (neigh->arp_queue_len_bytes + skb->truesize >
1224                                NEIGH_VAR(neigh->parms, QUEUE_LEN_BYTES)) {
1225                                 struct sk_buff *buff;
1226
1227                                 buff = __skb_dequeue(&neigh->arp_queue);
1228                                 if (!buff)
1229                                         break;
1230                                 neigh->arp_queue_len_bytes -= buff->truesize;
1231                                 kfree_skb_reason(buff, SKB_DROP_REASON_NEIGH_QUEUEFULL);
1232                                 NEIGH_CACHE_STAT_INC(neigh->tbl, unres_discards);
1233                         }
1234                         skb_dst_force(skb);
1235                         __skb_queue_tail(&neigh->arp_queue, skb);
1236                         neigh->arp_queue_len_bytes += skb->truesize;
1237                 }
1238                 rc = 1;
1239         }
1240 out_unlock_bh:
1241         if (immediate_probe)
1242                 neigh_probe(neigh);
1243         else
1244                 write_unlock(&neigh->lock);
1245         local_bh_enable();
1246         trace_neigh_event_send_done(neigh, rc);
1247         return rc;
1248
1249 out_dead:
1250         if (neigh->nud_state & NUD_STALE)
1251                 goto out_unlock_bh;
1252         write_unlock_bh(&neigh->lock);
1253         kfree_skb_reason(skb, SKB_DROP_REASON_NEIGH_DEAD);
1254         trace_neigh_event_send_dead(neigh, 1);
1255         return 1;
1256 }
1257 EXPORT_SYMBOL(__neigh_event_send);
1258
1259 static void neigh_update_hhs(struct neighbour *neigh)
1260 {
1261         struct hh_cache *hh;
1262         void (*update)(struct hh_cache*, const struct net_device*, const unsigned char *)
1263                 = NULL;
1264
1265         if (neigh->dev->header_ops)
1266                 update = neigh->dev->header_ops->cache_update;
1267
1268         if (update) {
1269                 hh = &neigh->hh;
1270                 if (READ_ONCE(hh->hh_len)) {
1271                         write_seqlock_bh(&hh->hh_lock);
1272                         update(hh, neigh->dev, neigh->ha);
1273                         write_sequnlock_bh(&hh->hh_lock);
1274                 }
1275         }
1276 }
1277
1278 /* Generic update routine.
1279    -- lladdr is new lladdr or NULL, if it is not supplied.
1280    -- new    is new state.
1281    -- flags
1282         NEIGH_UPDATE_F_OVERRIDE allows to override existing lladdr,
1283                                 if it is different.
1284         NEIGH_UPDATE_F_WEAK_OVERRIDE will suspect existing "connected"
1285                                 lladdr instead of overriding it
1286                                 if it is different.
1287         NEIGH_UPDATE_F_ADMIN    means that the change is administrative.
1288         NEIGH_UPDATE_F_USE      means that the entry is user triggered.
1289         NEIGH_UPDATE_F_MANAGED  means that the entry will be auto-refreshed.
1290         NEIGH_UPDATE_F_OVERRIDE_ISROUTER allows to override existing
1291                                 NTF_ROUTER flag.
1292         NEIGH_UPDATE_F_ISROUTER indicates if the neighbour is known as
1293                                 a router.
1294
1295    Caller MUST hold reference count on the entry.
1296  */
1297 static int __neigh_update(struct neighbour *neigh, const u8 *lladdr,
1298                           u8 new, u32 flags, u32 nlmsg_pid,
1299                           struct netlink_ext_ack *extack)
1300 {
1301         bool gc_update = false, managed_update = false;
1302         int update_isrouter = 0;
1303         struct net_device *dev;
1304         int err, notify = 0;
1305         u8 old;
1306
1307         trace_neigh_update(neigh, lladdr, new, flags, nlmsg_pid);
1308
1309         write_lock_bh(&neigh->lock);
1310
1311         dev    = neigh->dev;
1312         old    = neigh->nud_state;
1313         err    = -EPERM;
1314
1315         if (neigh->dead) {
1316                 NL_SET_ERR_MSG(extack, "Neighbor entry is now dead");
1317                 new = old;
1318                 goto out;
1319         }
1320         if (!(flags & NEIGH_UPDATE_F_ADMIN) &&
1321             (old & (NUD_NOARP | NUD_PERMANENT)))
1322                 goto out;
1323
1324         neigh_update_flags(neigh, flags, &notify, &gc_update, &managed_update);
1325         if (flags & (NEIGH_UPDATE_F_USE | NEIGH_UPDATE_F_MANAGED)) {
1326                 new = old & ~NUD_PERMANENT;
1327                 WRITE_ONCE(neigh->nud_state, new);
1328                 err = 0;
1329                 goto out;
1330         }
1331
1332         if (!(new & NUD_VALID)) {
1333                 neigh_del_timer(neigh);
1334                 if (old & NUD_CONNECTED)
1335                         neigh_suspect(neigh);
1336                 WRITE_ONCE(neigh->nud_state, new);
1337                 err = 0;
1338                 notify = old & NUD_VALID;
1339                 if ((old & (NUD_INCOMPLETE | NUD_PROBE)) &&
1340                     (new & NUD_FAILED)) {
1341                         neigh_invalidate(neigh);
1342                         notify = 1;
1343                 }
1344                 goto out;
1345         }
1346
1347         /* Compare new lladdr with cached one */
1348         if (!dev->addr_len) {
1349                 /* First case: device needs no address. */
1350                 lladdr = neigh->ha;
1351         } else if (lladdr) {
1352                 /* The second case: if something is already cached
1353                    and a new address is proposed:
1354                    - compare new & old
1355                    - if they are different, check override flag
1356                  */
1357                 if ((old & NUD_VALID) &&
1358                     !memcmp(lladdr, neigh->ha, dev->addr_len))
1359                         lladdr = neigh->ha;
1360         } else {
1361                 /* No address is supplied; if we know something,
1362                    use it, otherwise discard the request.
1363                  */
1364                 err = -EINVAL;
1365                 if (!(old & NUD_VALID)) {
1366                         NL_SET_ERR_MSG(extack, "No link layer address given");
1367                         goto out;
1368                 }
1369                 lladdr = neigh->ha;
1370         }
1371
1372         /* Update confirmed timestamp for neighbour entry after we
1373          * received ARP packet even if it doesn't change IP to MAC binding.
1374          */
1375         if (new & NUD_CONNECTED)
1376                 neigh->confirmed = jiffies;
1377
1378         /* If entry was valid and address is not changed,
1379            do not change entry state, if new one is STALE.
1380          */
1381         err = 0;
1382         update_isrouter = flags & NEIGH_UPDATE_F_OVERRIDE_ISROUTER;
1383         if (old & NUD_VALID) {
1384                 if (lladdr != neigh->ha && !(flags & NEIGH_UPDATE_F_OVERRIDE)) {
1385                         update_isrouter = 0;
1386                         if ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) &&
1387                             (old & NUD_CONNECTED)) {
1388                                 lladdr = neigh->ha;
1389                                 new = NUD_STALE;
1390                         } else
1391                                 goto out;
1392                 } else {
1393                         if (lladdr == neigh->ha && new == NUD_STALE &&
1394                             !(flags & NEIGH_UPDATE_F_ADMIN))
1395                                 new = old;
1396                 }
1397         }
1398
1399         /* Update timestamp only once we know we will make a change to the
1400          * neighbour entry. Otherwise we risk to move the locktime window with
1401          * noop updates and ignore relevant ARP updates.
1402          */
1403         if (new != old || lladdr != neigh->ha)
1404                 neigh->updated = jiffies;
1405
1406         if (new != old) {
1407                 neigh_del_timer(neigh);
1408                 if (new & NUD_PROBE)
1409                         atomic_set(&neigh->probes, 0);
1410                 if (new & NUD_IN_TIMER)
1411                         neigh_add_timer(neigh, (jiffies +
1412                                                 ((new & NUD_REACHABLE) ?
1413                                                  neigh->parms->reachable_time :
1414                                                  0)));
1415                 WRITE_ONCE(neigh->nud_state, new);
1416                 notify = 1;
1417         }
1418
1419         if (lladdr != neigh->ha) {
1420                 write_seqlock(&neigh->ha_lock);
1421                 memcpy(&neigh->ha, lladdr, dev->addr_len);
1422                 write_sequnlock(&neigh->ha_lock);
1423                 neigh_update_hhs(neigh);
1424                 if (!(new & NUD_CONNECTED))
1425                         neigh->confirmed = jiffies -
1426                                       (NEIGH_VAR(neigh->parms, BASE_REACHABLE_TIME) << 1);
1427                 notify = 1;
1428         }
1429         if (new == old)
1430                 goto out;
1431         if (new & NUD_CONNECTED)
1432                 neigh_connect(neigh);
1433         else
1434                 neigh_suspect(neigh);
1435         if (!(old & NUD_VALID)) {
1436                 struct sk_buff *skb;
1437
1438                 /* Again: avoid dead loop if something went wrong */
1439
1440                 while (neigh->nud_state & NUD_VALID &&
1441                        (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
1442                         struct dst_entry *dst = skb_dst(skb);
1443                         struct neighbour *n2, *n1 = neigh;
1444                         write_unlock_bh(&neigh->lock);
1445
1446                         rcu_read_lock();
1447
1448                         /* Why not just use 'neigh' as-is?  The problem is that
1449                          * things such as shaper, eql, and sch_teql can end up
1450                          * using alternative, different, neigh objects to output
1451                          * the packet in the output path.  So what we need to do
1452                          * here is re-lookup the top-level neigh in the path so
1453                          * we can reinject the packet there.
1454                          */
1455                         n2 = NULL;
1456                         if (dst && dst->obsolete != DST_OBSOLETE_DEAD) {
1457                                 n2 = dst_neigh_lookup_skb(dst, skb);
1458                                 if (n2)
1459                                         n1 = n2;
1460                         }
1461                         READ_ONCE(n1->output)(n1, skb);
1462                         if (n2)
1463                                 neigh_release(n2);
1464                         rcu_read_unlock();
1465
1466                         write_lock_bh(&neigh->lock);
1467                 }
1468                 __skb_queue_purge(&neigh->arp_queue);
1469                 neigh->arp_queue_len_bytes = 0;
1470         }
1471 out:
1472         if (update_isrouter)
1473                 neigh_update_is_router(neigh, flags, &notify);
1474         write_unlock_bh(&neigh->lock);
1475         if (((new ^ old) & NUD_PERMANENT) || gc_update)
1476                 neigh_update_gc_list(neigh);
1477         if (managed_update)
1478                 neigh_update_managed_list(neigh);
1479         if (notify)
1480                 neigh_update_notify(neigh, nlmsg_pid);
1481         trace_neigh_update_done(neigh, err);
1482         return err;
1483 }
1484
1485 int neigh_update(struct neighbour *neigh, const u8 *lladdr, u8 new,
1486                  u32 flags, u32 nlmsg_pid)
1487 {
1488         return __neigh_update(neigh, lladdr, new, flags, nlmsg_pid, NULL);
1489 }
1490 EXPORT_SYMBOL(neigh_update);
1491
1492 /* Update the neigh to listen temporarily for probe responses, even if it is
1493  * in a NUD_FAILED state. The caller has to hold neigh->lock for writing.
1494  */
1495 void __neigh_set_probe_once(struct neighbour *neigh)
1496 {
1497         if (neigh->dead)
1498                 return;
1499         neigh->updated = jiffies;
1500         if (!(neigh->nud_state & NUD_FAILED))
1501                 return;
1502         WRITE_ONCE(neigh->nud_state, NUD_INCOMPLETE);
1503         atomic_set(&neigh->probes, neigh_max_probes(neigh));
1504         neigh_add_timer(neigh,
1505                         jiffies + max(NEIGH_VAR(neigh->parms, RETRANS_TIME),
1506                                       HZ/100));
1507 }
1508 EXPORT_SYMBOL(__neigh_set_probe_once);
1509
1510 struct neighbour *neigh_event_ns(struct neigh_table *tbl,
1511                                  u8 *lladdr, void *saddr,
1512                                  struct net_device *dev)
1513 {
1514         struct neighbour *neigh = __neigh_lookup(tbl, saddr, dev,
1515                                                  lladdr || !dev->addr_len);
1516         if (neigh)
1517                 neigh_update(neigh, lladdr, NUD_STALE,
1518                              NEIGH_UPDATE_F_OVERRIDE, 0);
1519         return neigh;
1520 }
1521 EXPORT_SYMBOL(neigh_event_ns);
1522
1523 /* called with read_lock_bh(&n->lock); */
1524 static void neigh_hh_init(struct neighbour *n)
1525 {
1526         struct net_device *dev = n->dev;
1527         __be16 prot = n->tbl->protocol;
1528         struct hh_cache *hh = &n->hh;
1529
1530         write_lock_bh(&n->lock);
1531
1532         /* Only one thread can come in here and initialize the
1533          * hh_cache entry.
1534          */
1535         if (!hh->hh_len)
1536                 dev->header_ops->cache(n, hh, prot);
1537
1538         write_unlock_bh(&n->lock);
1539 }
1540
1541 /* Slow and careful. */
1542
1543 int neigh_resolve_output(struct neighbour *neigh, struct sk_buff *skb)
1544 {
1545         int rc = 0;
1546
1547         if (!neigh_event_send(neigh, skb)) {
1548                 int err;
1549                 struct net_device *dev = neigh->dev;
1550                 unsigned int seq;
1551
1552                 if (dev->header_ops->cache && !READ_ONCE(neigh->hh.hh_len))
1553                         neigh_hh_init(neigh);
1554
1555                 do {
1556                         __skb_pull(skb, skb_network_offset(skb));
1557                         seq = read_seqbegin(&neigh->ha_lock);
1558                         err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1559                                               neigh->ha, NULL, skb->len);
1560                 } while (read_seqretry(&neigh->ha_lock, seq));
1561
1562                 if (err >= 0)
1563                         rc = dev_queue_xmit(skb);
1564                 else
1565                         goto out_kfree_skb;
1566         }
1567 out:
1568         return rc;
1569 out_kfree_skb:
1570         rc = -EINVAL;
1571         kfree_skb(skb);
1572         goto out;
1573 }
1574 EXPORT_SYMBOL(neigh_resolve_output);
1575
1576 /* As fast as possible without hh cache */
1577
1578 int neigh_connected_output(struct neighbour *neigh, struct sk_buff *skb)
1579 {
1580         struct net_device *dev = neigh->dev;
1581         unsigned int seq;
1582         int err;
1583
1584         do {
1585                 __skb_pull(skb, skb_network_offset(skb));
1586                 seq = read_seqbegin(&neigh->ha_lock);
1587                 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1588                                       neigh->ha, NULL, skb->len);
1589         } while (read_seqretry(&neigh->ha_lock, seq));
1590
1591         if (err >= 0)
1592                 err = dev_queue_xmit(skb);
1593         else {
1594                 err = -EINVAL;
1595                 kfree_skb(skb);
1596         }
1597         return err;
1598 }
1599 EXPORT_SYMBOL(neigh_connected_output);
1600
1601 int neigh_direct_output(struct neighbour *neigh, struct sk_buff *skb)
1602 {
1603         return dev_queue_xmit(skb);
1604 }
1605 EXPORT_SYMBOL(neigh_direct_output);
1606
1607 static void neigh_managed_work(struct work_struct *work)
1608 {
1609         struct neigh_table *tbl = container_of(work, struct neigh_table,
1610                                                managed_work.work);
1611         struct neighbour *neigh;
1612
1613         write_lock_bh(&tbl->lock);
1614         list_for_each_entry(neigh, &tbl->managed_list, managed_list)
1615                 neigh_event_send_probe(neigh, NULL, false);
1616         queue_delayed_work(system_power_efficient_wq, &tbl->managed_work,
1617                            NEIGH_VAR(&tbl->parms, INTERVAL_PROBE_TIME_MS));
1618         write_unlock_bh(&tbl->lock);
1619 }
1620
1621 static void neigh_proxy_process(struct timer_list *t)
1622 {
1623         struct neigh_table *tbl = from_timer(tbl, t, proxy_timer);
1624         long sched_next = 0;
1625         unsigned long now = jiffies;
1626         struct sk_buff *skb, *n;
1627
1628         spin_lock(&tbl->proxy_queue.lock);
1629
1630         skb_queue_walk_safe(&tbl->proxy_queue, skb, n) {
1631                 long tdif = NEIGH_CB(skb)->sched_next - now;
1632
1633                 if (tdif <= 0) {
1634                         struct net_device *dev = skb->dev;
1635
1636                         neigh_parms_qlen_dec(dev, tbl->family);
1637                         __skb_unlink(skb, &tbl->proxy_queue);
1638
1639                         if (tbl->proxy_redo && netif_running(dev)) {
1640                                 rcu_read_lock();
1641                                 tbl->proxy_redo(skb);
1642                                 rcu_read_unlock();
1643                         } else {
1644                                 kfree_skb(skb);
1645                         }
1646
1647                         dev_put(dev);
1648                 } else if (!sched_next || tdif < sched_next)
1649                         sched_next = tdif;
1650         }
1651         del_timer(&tbl->proxy_timer);
1652         if (sched_next)
1653                 mod_timer(&tbl->proxy_timer, jiffies + sched_next);
1654         spin_unlock(&tbl->proxy_queue.lock);
1655 }
1656
1657 static unsigned long neigh_proxy_delay(struct neigh_parms *p)
1658 {
1659         /* If proxy_delay is zero, do not call get_random_u32_below()
1660          * as it is undefined behavior.
1661          */
1662         unsigned long proxy_delay = NEIGH_VAR(p, PROXY_DELAY);
1663
1664         return proxy_delay ?
1665                jiffies + get_random_u32_below(proxy_delay) : jiffies;
1666 }
1667
1668 void pneigh_enqueue(struct neigh_table *tbl, struct neigh_parms *p,
1669                     struct sk_buff *skb)
1670 {
1671         unsigned long sched_next = neigh_proxy_delay(p);
1672
1673         if (p->qlen > NEIGH_VAR(p, PROXY_QLEN)) {
1674                 kfree_skb(skb);
1675                 return;
1676         }
1677
1678         NEIGH_CB(skb)->sched_next = sched_next;
1679         NEIGH_CB(skb)->flags |= LOCALLY_ENQUEUED;
1680
1681         spin_lock(&tbl->proxy_queue.lock);
1682         if (del_timer(&tbl->proxy_timer)) {
1683                 if (time_before(tbl->proxy_timer.expires, sched_next))
1684                         sched_next = tbl->proxy_timer.expires;
1685         }
1686         skb_dst_drop(skb);
1687         dev_hold(skb->dev);
1688         __skb_queue_tail(&tbl->proxy_queue, skb);
1689         p->qlen++;
1690         mod_timer(&tbl->proxy_timer, sched_next);
1691         spin_unlock(&tbl->proxy_queue.lock);
1692 }
1693 EXPORT_SYMBOL(pneigh_enqueue);
1694
1695 static inline struct neigh_parms *lookup_neigh_parms(struct neigh_table *tbl,
1696                                                       struct net *net, int ifindex)
1697 {
1698         struct neigh_parms *p;
1699
1700         list_for_each_entry(p, &tbl->parms_list, list) {
1701                 if ((p->dev && p->dev->ifindex == ifindex && net_eq(neigh_parms_net(p), net)) ||
1702                     (!p->dev && !ifindex && net_eq(net, &init_net)))
1703                         return p;
1704         }
1705
1706         return NULL;
1707 }
1708
1709 struct neigh_parms *neigh_parms_alloc(struct net_device *dev,
1710                                       struct neigh_table *tbl)
1711 {
1712         struct neigh_parms *p;
1713         struct net *net = dev_net(dev);
1714         const struct net_device_ops *ops = dev->netdev_ops;
1715
1716         p = kmemdup(&tbl->parms, sizeof(*p), GFP_KERNEL);
1717         if (p) {
1718                 p->tbl            = tbl;
1719                 refcount_set(&p->refcnt, 1);
1720                 p->reachable_time =
1721                                 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
1722                 p->qlen = 0;
1723                 netdev_hold(dev, &p->dev_tracker, GFP_KERNEL);
1724                 p->dev = dev;
1725                 write_pnet(&p->net, net);
1726                 p->sysctl_table = NULL;
1727
1728                 if (ops->ndo_neigh_setup && ops->ndo_neigh_setup(dev, p)) {
1729                         netdev_put(dev, &p->dev_tracker);
1730                         kfree(p);
1731                         return NULL;
1732                 }
1733
1734                 write_lock_bh(&tbl->lock);
1735                 list_add(&p->list, &tbl->parms.list);
1736                 write_unlock_bh(&tbl->lock);
1737
1738                 neigh_parms_data_state_cleanall(p);
1739         }
1740         return p;
1741 }
1742 EXPORT_SYMBOL(neigh_parms_alloc);
1743
1744 static void neigh_rcu_free_parms(struct rcu_head *head)
1745 {
1746         struct neigh_parms *parms =
1747                 container_of(head, struct neigh_parms, rcu_head);
1748
1749         neigh_parms_put(parms);
1750 }
1751
1752 void neigh_parms_release(struct neigh_table *tbl, struct neigh_parms *parms)
1753 {
1754         if (!parms || parms == &tbl->parms)
1755                 return;
1756         write_lock_bh(&tbl->lock);
1757         list_del(&parms->list);
1758         parms->dead = 1;
1759         write_unlock_bh(&tbl->lock);
1760         netdev_put(parms->dev, &parms->dev_tracker);
1761         call_rcu(&parms->rcu_head, neigh_rcu_free_parms);
1762 }
1763 EXPORT_SYMBOL(neigh_parms_release);
1764
1765 static void neigh_parms_destroy(struct neigh_parms *parms)
1766 {
1767         kfree(parms);
1768 }
1769
1770 static struct lock_class_key neigh_table_proxy_queue_class;
1771
1772 static struct neigh_table *neigh_tables[NEIGH_NR_TABLES] __read_mostly;
1773
1774 void neigh_table_init(int index, struct neigh_table *tbl)
1775 {
1776         unsigned long now = jiffies;
1777         unsigned long phsize;
1778
1779         INIT_LIST_HEAD(&tbl->parms_list);
1780         INIT_LIST_HEAD(&tbl->gc_list);
1781         INIT_LIST_HEAD(&tbl->managed_list);
1782
1783         list_add(&tbl->parms.list, &tbl->parms_list);
1784         write_pnet(&tbl->parms.net, &init_net);
1785         refcount_set(&tbl->parms.refcnt, 1);
1786         tbl->parms.reachable_time =
1787                           neigh_rand_reach_time(NEIGH_VAR(&tbl->parms, BASE_REACHABLE_TIME));
1788         tbl->parms.qlen = 0;
1789
1790         tbl->stats = alloc_percpu(struct neigh_statistics);
1791         if (!tbl->stats)
1792                 panic("cannot create neighbour cache statistics");
1793
1794 #ifdef CONFIG_PROC_FS
1795         if (!proc_create_seq_data(tbl->id, 0, init_net.proc_net_stat,
1796                               &neigh_stat_seq_ops, tbl))
1797                 panic("cannot create neighbour proc dir entry");
1798 #endif
1799
1800         RCU_INIT_POINTER(tbl->nht, neigh_hash_alloc(3));
1801
1802         phsize = (PNEIGH_HASHMASK + 1) * sizeof(struct pneigh_entry *);
1803         tbl->phash_buckets = kzalloc(phsize, GFP_KERNEL);
1804
1805         if (!tbl->nht || !tbl->phash_buckets)
1806                 panic("cannot allocate neighbour cache hashes");
1807
1808         if (!tbl->entry_size)
1809                 tbl->entry_size = ALIGN(offsetof(struct neighbour, primary_key) +
1810                                         tbl->key_len, NEIGH_PRIV_ALIGN);
1811         else
1812                 WARN_ON(tbl->entry_size % NEIGH_PRIV_ALIGN);
1813
1814         rwlock_init(&tbl->lock);
1815
1816         INIT_DEFERRABLE_WORK(&tbl->gc_work, neigh_periodic_work);
1817         queue_delayed_work(system_power_efficient_wq, &tbl->gc_work,
1818                         tbl->parms.reachable_time);
1819         INIT_DEFERRABLE_WORK(&tbl->managed_work, neigh_managed_work);
1820         queue_delayed_work(system_power_efficient_wq, &tbl->managed_work, 0);
1821
1822         timer_setup(&tbl->proxy_timer, neigh_proxy_process, 0);
1823         skb_queue_head_init_class(&tbl->proxy_queue,
1824                         &neigh_table_proxy_queue_class);
1825
1826         tbl->last_flush = now;
1827         tbl->last_rand  = now + tbl->parms.reachable_time * 20;
1828
1829         neigh_tables[index] = tbl;
1830 }
1831 EXPORT_SYMBOL(neigh_table_init);
1832
1833 int neigh_table_clear(int index, struct neigh_table *tbl)
1834 {
1835         neigh_tables[index] = NULL;
1836         /* It is not clean... Fix it to unload IPv6 module safely */
1837         cancel_delayed_work_sync(&tbl->managed_work);
1838         cancel_delayed_work_sync(&tbl->gc_work);
1839         del_timer_sync(&tbl->proxy_timer);
1840         pneigh_queue_purge(&tbl->proxy_queue, NULL, tbl->family);
1841         neigh_ifdown(tbl, NULL);
1842         if (atomic_read(&tbl->entries))
1843                 pr_crit("neighbour leakage\n");
1844
1845         call_rcu(&rcu_dereference_protected(tbl->nht, 1)->rcu,
1846                  neigh_hash_free_rcu);
1847         tbl->nht = NULL;
1848
1849         kfree(tbl->phash_buckets);
1850         tbl->phash_buckets = NULL;
1851
1852         remove_proc_entry(tbl->id, init_net.proc_net_stat);
1853
1854         free_percpu(tbl->stats);
1855         tbl->stats = NULL;
1856
1857         return 0;
1858 }
1859 EXPORT_SYMBOL(neigh_table_clear);
1860
1861 static struct neigh_table *neigh_find_table(int family)
1862 {
1863         struct neigh_table *tbl = NULL;
1864
1865         switch (family) {
1866         case AF_INET:
1867                 tbl = neigh_tables[NEIGH_ARP_TABLE];
1868                 break;
1869         case AF_INET6:
1870                 tbl = neigh_tables[NEIGH_ND_TABLE];
1871                 break;
1872         }
1873
1874         return tbl;
1875 }
1876
1877 const struct nla_policy nda_policy[NDA_MAX+1] = {
1878         [NDA_UNSPEC]            = { .strict_start_type = NDA_NH_ID },
1879         [NDA_DST]               = { .type = NLA_BINARY, .len = MAX_ADDR_LEN },
1880         [NDA_LLADDR]            = { .type = NLA_BINARY, .len = MAX_ADDR_LEN },
1881         [NDA_CACHEINFO]         = { .len = sizeof(struct nda_cacheinfo) },
1882         [NDA_PROBES]            = { .type = NLA_U32 },
1883         [NDA_VLAN]              = { .type = NLA_U16 },
1884         [NDA_PORT]              = { .type = NLA_U16 },
1885         [NDA_VNI]               = { .type = NLA_U32 },
1886         [NDA_IFINDEX]           = { .type = NLA_U32 },
1887         [NDA_MASTER]            = { .type = NLA_U32 },
1888         [NDA_PROTOCOL]          = { .type = NLA_U8 },
1889         [NDA_NH_ID]             = { .type = NLA_U32 },
1890         [NDA_FLAGS_EXT]         = NLA_POLICY_MASK(NLA_U32, NTF_EXT_MASK),
1891         [NDA_FDB_EXT_ATTRS]     = { .type = NLA_NESTED },
1892 };
1893
1894 static int neigh_delete(struct sk_buff *skb, struct nlmsghdr *nlh,
1895                         struct netlink_ext_ack *extack)
1896 {
1897         struct net *net = sock_net(skb->sk);
1898         struct ndmsg *ndm;
1899         struct nlattr *dst_attr;
1900         struct neigh_table *tbl;
1901         struct neighbour *neigh;
1902         struct net_device *dev = NULL;
1903         int err = -EINVAL;
1904
1905         ASSERT_RTNL();
1906         if (nlmsg_len(nlh) < sizeof(*ndm))
1907                 goto out;
1908
1909         dst_attr = nlmsg_find_attr(nlh, sizeof(*ndm), NDA_DST);
1910         if (!dst_attr) {
1911                 NL_SET_ERR_MSG(extack, "Network address not specified");
1912                 goto out;
1913         }
1914
1915         ndm = nlmsg_data(nlh);
1916         if (ndm->ndm_ifindex) {
1917                 dev = __dev_get_by_index(net, ndm->ndm_ifindex);
1918                 if (dev == NULL) {
1919                         err = -ENODEV;
1920                         goto out;
1921                 }
1922         }
1923
1924         tbl = neigh_find_table(ndm->ndm_family);
1925         if (tbl == NULL)
1926                 return -EAFNOSUPPORT;
1927
1928         if (nla_len(dst_attr) < (int)tbl->key_len) {
1929                 NL_SET_ERR_MSG(extack, "Invalid network address");
1930                 goto out;
1931         }
1932
1933         if (ndm->ndm_flags & NTF_PROXY) {
1934                 err = pneigh_delete(tbl, net, nla_data(dst_attr), dev);
1935                 goto out;
1936         }
1937
1938         if (dev == NULL)
1939                 goto out;
1940
1941         neigh = neigh_lookup(tbl, nla_data(dst_attr), dev);
1942         if (neigh == NULL) {
1943                 err = -ENOENT;
1944                 goto out;
1945         }
1946
1947         err = __neigh_update(neigh, NULL, NUD_FAILED,
1948                              NEIGH_UPDATE_F_OVERRIDE | NEIGH_UPDATE_F_ADMIN,
1949                              NETLINK_CB(skb).portid, extack);
1950         write_lock_bh(&tbl->lock);
1951         neigh_release(neigh);
1952         neigh_remove_one(neigh, tbl);
1953         write_unlock_bh(&tbl->lock);
1954
1955 out:
1956         return err;
1957 }
1958
1959 static int neigh_add(struct sk_buff *skb, struct nlmsghdr *nlh,
1960                      struct netlink_ext_ack *extack)
1961 {
1962         int flags = NEIGH_UPDATE_F_ADMIN | NEIGH_UPDATE_F_OVERRIDE |
1963                     NEIGH_UPDATE_F_OVERRIDE_ISROUTER;
1964         struct net *net = sock_net(skb->sk);
1965         struct ndmsg *ndm;
1966         struct nlattr *tb[NDA_MAX+1];
1967         struct neigh_table *tbl;
1968         struct net_device *dev = NULL;
1969         struct neighbour *neigh;
1970         void *dst, *lladdr;
1971         u8 protocol = 0;
1972         u32 ndm_flags;
1973         int err;
1974
1975         ASSERT_RTNL();
1976         err = nlmsg_parse_deprecated(nlh, sizeof(*ndm), tb, NDA_MAX,
1977                                      nda_policy, extack);
1978         if (err < 0)
1979                 goto out;
1980
1981         err = -EINVAL;
1982         if (!tb[NDA_DST]) {
1983                 NL_SET_ERR_MSG(extack, "Network address not specified");
1984                 goto out;
1985         }
1986
1987         ndm = nlmsg_data(nlh);
1988         ndm_flags = ndm->ndm_flags;
1989         if (tb[NDA_FLAGS_EXT]) {
1990                 u32 ext = nla_get_u32(tb[NDA_FLAGS_EXT]);
1991
1992                 BUILD_BUG_ON(sizeof(neigh->flags) * BITS_PER_BYTE <
1993                              (sizeof(ndm->ndm_flags) * BITS_PER_BYTE +
1994                               hweight32(NTF_EXT_MASK)));
1995                 ndm_flags |= (ext << NTF_EXT_SHIFT);
1996         }
1997         if (ndm->ndm_ifindex) {
1998                 dev = __dev_get_by_index(net, ndm->ndm_ifindex);
1999                 if (dev == NULL) {
2000                         err = -ENODEV;
2001                         goto out;
2002                 }
2003
2004                 if (tb[NDA_LLADDR] && nla_len(tb[NDA_LLADDR]) < dev->addr_len) {
2005                         NL_SET_ERR_MSG(extack, "Invalid link address");
2006                         goto out;
2007                 }
2008         }
2009
2010         tbl = neigh_find_table(ndm->ndm_family);
2011         if (tbl == NULL)
2012                 return -EAFNOSUPPORT;
2013
2014         if (nla_len(tb[NDA_DST]) < (int)tbl->key_len) {
2015                 NL_SET_ERR_MSG(extack, "Invalid network address");
2016                 goto out;
2017         }
2018
2019         dst = nla_data(tb[NDA_DST]);
2020         lladdr = tb[NDA_LLADDR] ? nla_data(tb[NDA_LLADDR]) : NULL;
2021
2022         if (tb[NDA_PROTOCOL])
2023                 protocol = nla_get_u8(tb[NDA_PROTOCOL]);
2024         if (ndm_flags & NTF_PROXY) {
2025                 struct pneigh_entry *pn;
2026
2027                 if (ndm_flags & NTF_MANAGED) {
2028                         NL_SET_ERR_MSG(extack, "Invalid NTF_* flag combination");
2029                         goto out;
2030                 }
2031
2032                 err = -ENOBUFS;
2033                 pn = pneigh_lookup(tbl, net, dst, dev, 1);
2034                 if (pn) {
2035                         pn->flags = ndm_flags;
2036                         if (protocol)
2037                                 pn->protocol = protocol;
2038                         err = 0;
2039                 }
2040                 goto out;
2041         }
2042
2043         if (!dev) {
2044                 NL_SET_ERR_MSG(extack, "Device not specified");
2045                 goto out;
2046         }
2047
2048         if (tbl->allow_add && !tbl->allow_add(dev, extack)) {
2049                 err = -EINVAL;
2050                 goto out;
2051         }
2052
2053         neigh = neigh_lookup(tbl, dst, dev);
2054         if (neigh == NULL) {
2055                 bool ndm_permanent  = ndm->ndm_state & NUD_PERMANENT;
2056                 bool exempt_from_gc = ndm_permanent ||
2057                                       ndm_flags & NTF_EXT_LEARNED;
2058
2059                 if (!(nlh->nlmsg_flags & NLM_F_CREATE)) {
2060                         err = -ENOENT;
2061                         goto out;
2062                 }
2063                 if (ndm_permanent && (ndm_flags & NTF_MANAGED)) {
2064                         NL_SET_ERR_MSG(extack, "Invalid NTF_* flag for permanent entry");
2065                         err = -EINVAL;
2066                         goto out;
2067                 }
2068
2069                 neigh = ___neigh_create(tbl, dst, dev,
2070                                         ndm_flags &
2071                                         (NTF_EXT_LEARNED | NTF_MANAGED),
2072                                         exempt_from_gc, true);
2073                 if (IS_ERR(neigh)) {
2074                         err = PTR_ERR(neigh);
2075                         goto out;
2076                 }
2077         } else {
2078                 if (nlh->nlmsg_flags & NLM_F_EXCL) {
2079                         err = -EEXIST;
2080                         neigh_release(neigh);
2081                         goto out;
2082                 }
2083
2084                 if (!(nlh->nlmsg_flags & NLM_F_REPLACE))
2085                         flags &= ~(NEIGH_UPDATE_F_OVERRIDE |
2086                                    NEIGH_UPDATE_F_OVERRIDE_ISROUTER);
2087         }
2088
2089         if (protocol)
2090                 neigh->protocol = protocol;
2091         if (ndm_flags & NTF_EXT_LEARNED)
2092                 flags |= NEIGH_UPDATE_F_EXT_LEARNED;
2093         if (ndm_flags & NTF_ROUTER)
2094                 flags |= NEIGH_UPDATE_F_ISROUTER;
2095         if (ndm_flags & NTF_MANAGED)
2096                 flags |= NEIGH_UPDATE_F_MANAGED;
2097         if (ndm_flags & NTF_USE)
2098                 flags |= NEIGH_UPDATE_F_USE;
2099
2100         err = __neigh_update(neigh, lladdr, ndm->ndm_state, flags,
2101                              NETLINK_CB(skb).portid, extack);
2102         if (!err && ndm_flags & (NTF_USE | NTF_MANAGED)) {
2103                 neigh_event_send(neigh, NULL);
2104                 err = 0;
2105         }
2106         neigh_release(neigh);
2107 out:
2108         return err;
2109 }
2110
2111 static int neightbl_fill_parms(struct sk_buff *skb, struct neigh_parms *parms)
2112 {
2113         struct nlattr *nest;
2114
2115         nest = nla_nest_start_noflag(skb, NDTA_PARMS);
2116         if (nest == NULL)
2117                 return -ENOBUFS;
2118
2119         if ((parms->dev &&
2120              nla_put_u32(skb, NDTPA_IFINDEX, parms->dev->ifindex)) ||
2121             nla_put_u32(skb, NDTPA_REFCNT, refcount_read(&parms->refcnt)) ||
2122             nla_put_u32(skb, NDTPA_QUEUE_LENBYTES,
2123                         NEIGH_VAR(parms, QUEUE_LEN_BYTES)) ||
2124             /* approximative value for deprecated QUEUE_LEN (in packets) */
2125             nla_put_u32(skb, NDTPA_QUEUE_LEN,
2126                         NEIGH_VAR(parms, QUEUE_LEN_BYTES) / SKB_TRUESIZE(ETH_FRAME_LEN)) ||
2127             nla_put_u32(skb, NDTPA_PROXY_QLEN, NEIGH_VAR(parms, PROXY_QLEN)) ||
2128             nla_put_u32(skb, NDTPA_APP_PROBES, NEIGH_VAR(parms, APP_PROBES)) ||
2129             nla_put_u32(skb, NDTPA_UCAST_PROBES,
2130                         NEIGH_VAR(parms, UCAST_PROBES)) ||
2131             nla_put_u32(skb, NDTPA_MCAST_PROBES,
2132                         NEIGH_VAR(parms, MCAST_PROBES)) ||
2133             nla_put_u32(skb, NDTPA_MCAST_REPROBES,
2134                         NEIGH_VAR(parms, MCAST_REPROBES)) ||
2135             nla_put_msecs(skb, NDTPA_REACHABLE_TIME, parms->reachable_time,
2136                           NDTPA_PAD) ||
2137             nla_put_msecs(skb, NDTPA_BASE_REACHABLE_TIME,
2138                           NEIGH_VAR(parms, BASE_REACHABLE_TIME), NDTPA_PAD) ||
2139             nla_put_msecs(skb, NDTPA_GC_STALETIME,
2140                           NEIGH_VAR(parms, GC_STALETIME), NDTPA_PAD) ||
2141             nla_put_msecs(skb, NDTPA_DELAY_PROBE_TIME,
2142                           NEIGH_VAR(parms, DELAY_PROBE_TIME), NDTPA_PAD) ||
2143             nla_put_msecs(skb, NDTPA_RETRANS_TIME,
2144                           NEIGH_VAR(parms, RETRANS_TIME), NDTPA_PAD) ||
2145             nla_put_msecs(skb, NDTPA_ANYCAST_DELAY,
2146                           NEIGH_VAR(parms, ANYCAST_DELAY), NDTPA_PAD) ||
2147             nla_put_msecs(skb, NDTPA_PROXY_DELAY,
2148                           NEIGH_VAR(parms, PROXY_DELAY), NDTPA_PAD) ||
2149             nla_put_msecs(skb, NDTPA_LOCKTIME,
2150                           NEIGH_VAR(parms, LOCKTIME), NDTPA_PAD) ||
2151             nla_put_msecs(skb, NDTPA_INTERVAL_PROBE_TIME_MS,
2152                           NEIGH_VAR(parms, INTERVAL_PROBE_TIME_MS), NDTPA_PAD))
2153                 goto nla_put_failure;
2154         return nla_nest_end(skb, nest);
2155
2156 nla_put_failure:
2157         nla_nest_cancel(skb, nest);
2158         return -EMSGSIZE;
2159 }
2160
2161 static int neightbl_fill_info(struct sk_buff *skb, struct neigh_table *tbl,
2162                               u32 pid, u32 seq, int type, int flags)
2163 {
2164         struct nlmsghdr *nlh;
2165         struct ndtmsg *ndtmsg;
2166
2167         nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
2168         if (nlh == NULL)
2169                 return -EMSGSIZE;
2170
2171         ndtmsg = nlmsg_data(nlh);
2172
2173         read_lock_bh(&tbl->lock);
2174         ndtmsg->ndtm_family = tbl->family;
2175         ndtmsg->ndtm_pad1   = 0;
2176         ndtmsg->ndtm_pad2   = 0;
2177
2178         if (nla_put_string(skb, NDTA_NAME, tbl->id) ||
2179             nla_put_msecs(skb, NDTA_GC_INTERVAL, READ_ONCE(tbl->gc_interval),
2180                           NDTA_PAD) ||
2181             nla_put_u32(skb, NDTA_THRESH1, READ_ONCE(tbl->gc_thresh1)) ||
2182             nla_put_u32(skb, NDTA_THRESH2, READ_ONCE(tbl->gc_thresh2)) ||
2183             nla_put_u32(skb, NDTA_THRESH3, READ_ONCE(tbl->gc_thresh3)))
2184                 goto nla_put_failure;
2185         {
2186                 unsigned long now = jiffies;
2187                 long flush_delta = now - READ_ONCE(tbl->last_flush);
2188                 long rand_delta = now - READ_ONCE(tbl->last_rand);
2189                 struct neigh_hash_table *nht;
2190                 struct ndt_config ndc = {
2191                         .ndtc_key_len           = tbl->key_len,
2192                         .ndtc_entry_size        = tbl->entry_size,
2193                         .ndtc_entries           = atomic_read(&tbl->entries),
2194                         .ndtc_last_flush        = jiffies_to_msecs(flush_delta),
2195                         .ndtc_last_rand         = jiffies_to_msecs(rand_delta),
2196                         .ndtc_proxy_qlen        = READ_ONCE(tbl->proxy_queue.qlen),
2197                 };
2198
2199                 rcu_read_lock();
2200                 nht = rcu_dereference(tbl->nht);
2201                 ndc.ndtc_hash_rnd = nht->hash_rnd[0];
2202                 ndc.ndtc_hash_mask = ((1 << nht->hash_shift) - 1);
2203                 rcu_read_unlock();
2204
2205                 if (nla_put(skb, NDTA_CONFIG, sizeof(ndc), &ndc))
2206                         goto nla_put_failure;
2207         }
2208
2209         {
2210                 int cpu;
2211                 struct ndt_stats ndst;
2212
2213                 memset(&ndst, 0, sizeof(ndst));
2214
2215                 for_each_possible_cpu(cpu) {
2216                         struct neigh_statistics *st;
2217
2218                         st = per_cpu_ptr(tbl->stats, cpu);
2219                         ndst.ndts_allocs                += READ_ONCE(st->allocs);
2220                         ndst.ndts_destroys              += READ_ONCE(st->destroys);
2221                         ndst.ndts_hash_grows            += READ_ONCE(st->hash_grows);
2222                         ndst.ndts_res_failed            += READ_ONCE(st->res_failed);
2223                         ndst.ndts_lookups               += READ_ONCE(st->lookups);
2224                         ndst.ndts_hits                  += READ_ONCE(st->hits);
2225                         ndst.ndts_rcv_probes_mcast      += READ_ONCE(st->rcv_probes_mcast);
2226                         ndst.ndts_rcv_probes_ucast      += READ_ONCE(st->rcv_probes_ucast);
2227                         ndst.ndts_periodic_gc_runs      += READ_ONCE(st->periodic_gc_runs);
2228                         ndst.ndts_forced_gc_runs        += READ_ONCE(st->forced_gc_runs);
2229                         ndst.ndts_table_fulls           += READ_ONCE(st->table_fulls);
2230                 }
2231
2232                 if (nla_put_64bit(skb, NDTA_STATS, sizeof(ndst), &ndst,
2233                                   NDTA_PAD))
2234                         goto nla_put_failure;
2235         }
2236
2237         BUG_ON(tbl->parms.dev);
2238         if (neightbl_fill_parms(skb, &tbl->parms) < 0)
2239                 goto nla_put_failure;
2240
2241         read_unlock_bh(&tbl->lock);
2242         nlmsg_end(skb, nlh);
2243         return 0;
2244
2245 nla_put_failure:
2246         read_unlock_bh(&tbl->lock);
2247         nlmsg_cancel(skb, nlh);
2248         return -EMSGSIZE;
2249 }
2250
2251 static int neightbl_fill_param_info(struct sk_buff *skb,
2252                                     struct neigh_table *tbl,
2253                                     struct neigh_parms *parms,
2254                                     u32 pid, u32 seq, int type,
2255                                     unsigned int flags)
2256 {
2257         struct ndtmsg *ndtmsg;
2258         struct nlmsghdr *nlh;
2259
2260         nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
2261         if (nlh == NULL)
2262                 return -EMSGSIZE;
2263
2264         ndtmsg = nlmsg_data(nlh);
2265
2266         read_lock_bh(&tbl->lock);
2267         ndtmsg->ndtm_family = tbl->family;
2268         ndtmsg->ndtm_pad1   = 0;
2269         ndtmsg->ndtm_pad2   = 0;
2270
2271         if (nla_put_string(skb, NDTA_NAME, tbl->id) < 0 ||
2272             neightbl_fill_parms(skb, parms) < 0)
2273                 goto errout;
2274
2275         read_unlock_bh(&tbl->lock);
2276         nlmsg_end(skb, nlh);
2277         return 0;
2278 errout:
2279         read_unlock_bh(&tbl->lock);
2280         nlmsg_cancel(skb, nlh);
2281         return -EMSGSIZE;
2282 }
2283
2284 static const struct nla_policy nl_neightbl_policy[NDTA_MAX+1] = {
2285         [NDTA_NAME]             = { .type = NLA_STRING },
2286         [NDTA_THRESH1]          = { .type = NLA_U32 },
2287         [NDTA_THRESH2]          = { .type = NLA_U32 },
2288         [NDTA_THRESH3]          = { .type = NLA_U32 },
2289         [NDTA_GC_INTERVAL]      = { .type = NLA_U64 },
2290         [NDTA_PARMS]            = { .type = NLA_NESTED },
2291 };
2292
2293 static const struct nla_policy nl_ntbl_parm_policy[NDTPA_MAX+1] = {
2294         [NDTPA_IFINDEX]                 = { .type = NLA_U32 },
2295         [NDTPA_QUEUE_LEN]               = { .type = NLA_U32 },
2296         [NDTPA_PROXY_QLEN]              = { .type = NLA_U32 },
2297         [NDTPA_APP_PROBES]              = { .type = NLA_U32 },
2298         [NDTPA_UCAST_PROBES]            = { .type = NLA_U32 },
2299         [NDTPA_MCAST_PROBES]            = { .type = NLA_U32 },
2300         [NDTPA_MCAST_REPROBES]          = { .type = NLA_U32 },
2301         [NDTPA_BASE_REACHABLE_TIME]     = { .type = NLA_U64 },
2302         [NDTPA_GC_STALETIME]            = { .type = NLA_U64 },
2303         [NDTPA_DELAY_PROBE_TIME]        = { .type = NLA_U64 },
2304         [NDTPA_RETRANS_TIME]            = { .type = NLA_U64 },
2305         [NDTPA_ANYCAST_DELAY]           = { .type = NLA_U64 },
2306         [NDTPA_PROXY_DELAY]             = { .type = NLA_U64 },
2307         [NDTPA_LOCKTIME]                = { .type = NLA_U64 },
2308         [NDTPA_INTERVAL_PROBE_TIME_MS]  = { .type = NLA_U64, .min = 1 },
2309 };
2310
2311 static int neightbl_set(struct sk_buff *skb, struct nlmsghdr *nlh,
2312                         struct netlink_ext_ack *extack)
2313 {
2314         struct net *net = sock_net(skb->sk);
2315         struct neigh_table *tbl;
2316         struct ndtmsg *ndtmsg;
2317         struct nlattr *tb[NDTA_MAX+1];
2318         bool found = false;
2319         int err, tidx;
2320
2321         err = nlmsg_parse_deprecated(nlh, sizeof(*ndtmsg), tb, NDTA_MAX,
2322                                      nl_neightbl_policy, extack);
2323         if (err < 0)
2324                 goto errout;
2325
2326         if (tb[NDTA_NAME] == NULL) {
2327                 err = -EINVAL;
2328                 goto errout;
2329         }
2330
2331         ndtmsg = nlmsg_data(nlh);
2332
2333         for (tidx = 0; tidx < NEIGH_NR_TABLES; tidx++) {
2334                 tbl = neigh_tables[tidx];
2335                 if (!tbl)
2336                         continue;
2337                 if (ndtmsg->ndtm_family && tbl->family != ndtmsg->ndtm_family)
2338                         continue;
2339                 if (nla_strcmp(tb[NDTA_NAME], tbl->id) == 0) {
2340                         found = true;
2341                         break;
2342                 }
2343         }
2344
2345         if (!found)
2346                 return -ENOENT;
2347
2348         /*
2349          * We acquire tbl->lock to be nice to the periodic timers and
2350          * make sure they always see a consistent set of values.
2351          */
2352         write_lock_bh(&tbl->lock);
2353
2354         if (tb[NDTA_PARMS]) {
2355                 struct nlattr *tbp[NDTPA_MAX+1];
2356                 struct neigh_parms *p;
2357                 int i, ifindex = 0;
2358
2359                 err = nla_parse_nested_deprecated(tbp, NDTPA_MAX,
2360                                                   tb[NDTA_PARMS],
2361                                                   nl_ntbl_parm_policy, extack);
2362                 if (err < 0)
2363                         goto errout_tbl_lock;
2364
2365                 if (tbp[NDTPA_IFINDEX])
2366                         ifindex = nla_get_u32(tbp[NDTPA_IFINDEX]);
2367
2368                 p = lookup_neigh_parms(tbl, net, ifindex);
2369                 if (p == NULL) {
2370                         err = -ENOENT;
2371                         goto errout_tbl_lock;
2372                 }
2373
2374                 for (i = 1; i <= NDTPA_MAX; i++) {
2375                         if (tbp[i] == NULL)
2376                                 continue;
2377
2378                         switch (i) {
2379                         case NDTPA_QUEUE_LEN:
2380                                 NEIGH_VAR_SET(p, QUEUE_LEN_BYTES,
2381                                               nla_get_u32(tbp[i]) *
2382                                               SKB_TRUESIZE(ETH_FRAME_LEN));
2383                                 break;
2384                         case NDTPA_QUEUE_LENBYTES:
2385                                 NEIGH_VAR_SET(p, QUEUE_LEN_BYTES,
2386                                               nla_get_u32(tbp[i]));
2387                                 break;
2388                         case NDTPA_PROXY_QLEN:
2389                                 NEIGH_VAR_SET(p, PROXY_QLEN,
2390                                               nla_get_u32(tbp[i]));
2391                                 break;
2392                         case NDTPA_APP_PROBES:
2393                                 NEIGH_VAR_SET(p, APP_PROBES,
2394                                               nla_get_u32(tbp[i]));
2395                                 break;
2396                         case NDTPA_UCAST_PROBES:
2397                                 NEIGH_VAR_SET(p, UCAST_PROBES,
2398                                               nla_get_u32(tbp[i]));
2399                                 break;
2400                         case NDTPA_MCAST_PROBES:
2401                                 NEIGH_VAR_SET(p, MCAST_PROBES,
2402                                               nla_get_u32(tbp[i]));
2403                                 break;
2404                         case NDTPA_MCAST_REPROBES:
2405                                 NEIGH_VAR_SET(p, MCAST_REPROBES,
2406                                               nla_get_u32(tbp[i]));
2407                                 break;
2408                         case NDTPA_BASE_REACHABLE_TIME:
2409                                 NEIGH_VAR_SET(p, BASE_REACHABLE_TIME,
2410                                               nla_get_msecs(tbp[i]));
2411                                 /* update reachable_time as well, otherwise, the change will
2412                                  * only be effective after the next time neigh_periodic_work
2413                                  * decides to recompute it (can be multiple minutes)
2414                                  */
2415                                 p->reachable_time =
2416                                         neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
2417                                 break;
2418                         case NDTPA_GC_STALETIME:
2419                                 NEIGH_VAR_SET(p, GC_STALETIME,
2420                                               nla_get_msecs(tbp[i]));
2421                                 break;
2422                         case NDTPA_DELAY_PROBE_TIME:
2423                                 NEIGH_VAR_SET(p, DELAY_PROBE_TIME,
2424                                               nla_get_msecs(tbp[i]));
2425                                 call_netevent_notifiers(NETEVENT_DELAY_PROBE_TIME_UPDATE, p);
2426                                 break;
2427                         case NDTPA_INTERVAL_PROBE_TIME_MS:
2428                                 NEIGH_VAR_SET(p, INTERVAL_PROBE_TIME_MS,
2429                                               nla_get_msecs(tbp[i]));
2430                                 break;
2431                         case NDTPA_RETRANS_TIME:
2432                                 NEIGH_VAR_SET(p, RETRANS_TIME,
2433                                               nla_get_msecs(tbp[i]));
2434                                 break;
2435                         case NDTPA_ANYCAST_DELAY:
2436                                 NEIGH_VAR_SET(p, ANYCAST_DELAY,
2437                                               nla_get_msecs(tbp[i]));
2438                                 break;
2439                         case NDTPA_PROXY_DELAY:
2440                                 NEIGH_VAR_SET(p, PROXY_DELAY,
2441                                               nla_get_msecs(tbp[i]));
2442                                 break;
2443                         case NDTPA_LOCKTIME:
2444                                 NEIGH_VAR_SET(p, LOCKTIME,
2445                                               nla_get_msecs(tbp[i]));
2446                                 break;
2447                         }
2448                 }
2449         }
2450
2451         err = -ENOENT;
2452         if ((tb[NDTA_THRESH1] || tb[NDTA_THRESH2] ||
2453              tb[NDTA_THRESH3] || tb[NDTA_GC_INTERVAL]) &&
2454             !net_eq(net, &init_net))
2455                 goto errout_tbl_lock;
2456
2457         if (tb[NDTA_THRESH1])
2458                 WRITE_ONCE(tbl->gc_thresh1, nla_get_u32(tb[NDTA_THRESH1]));
2459
2460         if (tb[NDTA_THRESH2])
2461                 WRITE_ONCE(tbl->gc_thresh2, nla_get_u32(tb[NDTA_THRESH2]));
2462
2463         if (tb[NDTA_THRESH3])
2464                 WRITE_ONCE(tbl->gc_thresh3, nla_get_u32(tb[NDTA_THRESH3]));
2465
2466         if (tb[NDTA_GC_INTERVAL])
2467                 WRITE_ONCE(tbl->gc_interval, nla_get_msecs(tb[NDTA_GC_INTERVAL]));
2468
2469         err = 0;
2470
2471 errout_tbl_lock:
2472         write_unlock_bh(&tbl->lock);
2473 errout:
2474         return err;
2475 }
2476
2477 static int neightbl_valid_dump_info(const struct nlmsghdr *nlh,
2478                                     struct netlink_ext_ack *extack)
2479 {
2480         struct ndtmsg *ndtm;
2481
2482         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ndtm))) {
2483                 NL_SET_ERR_MSG(extack, "Invalid header for neighbor table dump request");
2484                 return -EINVAL;
2485         }
2486
2487         ndtm = nlmsg_data(nlh);
2488         if (ndtm->ndtm_pad1  || ndtm->ndtm_pad2) {
2489                 NL_SET_ERR_MSG(extack, "Invalid values in header for neighbor table dump request");
2490                 return -EINVAL;
2491         }
2492
2493         if (nlmsg_attrlen(nlh, sizeof(*ndtm))) {
2494                 NL_SET_ERR_MSG(extack, "Invalid data after header in neighbor table dump request");
2495                 return -EINVAL;
2496         }
2497
2498         return 0;
2499 }
2500
2501 static int neightbl_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2502 {
2503         const struct nlmsghdr *nlh = cb->nlh;
2504         struct net *net = sock_net(skb->sk);
2505         int family, tidx, nidx = 0;
2506         int tbl_skip = cb->args[0];
2507         int neigh_skip = cb->args[1];
2508         struct neigh_table *tbl;
2509
2510         if (cb->strict_check) {
2511                 int err = neightbl_valid_dump_info(nlh, cb->extack);
2512
2513                 if (err < 0)
2514                         return err;
2515         }
2516
2517         family = ((struct rtgenmsg *)nlmsg_data(nlh))->rtgen_family;
2518
2519         for (tidx = 0; tidx < NEIGH_NR_TABLES; tidx++) {
2520                 struct neigh_parms *p;
2521
2522                 tbl = neigh_tables[tidx];
2523                 if (!tbl)
2524                         continue;
2525
2526                 if (tidx < tbl_skip || (family && tbl->family != family))
2527                         continue;
2528
2529                 if (neightbl_fill_info(skb, tbl, NETLINK_CB(cb->skb).portid,
2530                                        nlh->nlmsg_seq, RTM_NEWNEIGHTBL,
2531                                        NLM_F_MULTI) < 0)
2532                         break;
2533
2534                 nidx = 0;
2535                 p = list_next_entry(&tbl->parms, list);
2536                 list_for_each_entry_from(p, &tbl->parms_list, list) {
2537                         if (!net_eq(neigh_parms_net(p), net))
2538                                 continue;
2539
2540                         if (nidx < neigh_skip)
2541                                 goto next;
2542
2543                         if (neightbl_fill_param_info(skb, tbl, p,
2544                                                      NETLINK_CB(cb->skb).portid,
2545                                                      nlh->nlmsg_seq,
2546                                                      RTM_NEWNEIGHTBL,
2547                                                      NLM_F_MULTI) < 0)
2548                                 goto out;
2549                 next:
2550                         nidx++;
2551                 }
2552
2553                 neigh_skip = 0;
2554         }
2555 out:
2556         cb->args[0] = tidx;
2557         cb->args[1] = nidx;
2558
2559         return skb->len;
2560 }
2561
2562 static int neigh_fill_info(struct sk_buff *skb, struct neighbour *neigh,
2563                            u32 pid, u32 seq, int type, unsigned int flags)
2564 {
2565         u32 neigh_flags, neigh_flags_ext;
2566         unsigned long now = jiffies;
2567         struct nda_cacheinfo ci;
2568         struct nlmsghdr *nlh;
2569         struct ndmsg *ndm;
2570
2571         nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2572         if (nlh == NULL)
2573                 return -EMSGSIZE;
2574
2575         neigh_flags_ext = neigh->flags >> NTF_EXT_SHIFT;
2576         neigh_flags     = neigh->flags & NTF_OLD_MASK;
2577
2578         ndm = nlmsg_data(nlh);
2579         ndm->ndm_family  = neigh->ops->family;
2580         ndm->ndm_pad1    = 0;
2581         ndm->ndm_pad2    = 0;
2582         ndm->ndm_flags   = neigh_flags;
2583         ndm->ndm_type    = neigh->type;
2584         ndm->ndm_ifindex = neigh->dev->ifindex;
2585
2586         if (nla_put(skb, NDA_DST, neigh->tbl->key_len, neigh->primary_key))
2587                 goto nla_put_failure;
2588
2589         read_lock_bh(&neigh->lock);
2590         ndm->ndm_state   = neigh->nud_state;
2591         if (neigh->nud_state & NUD_VALID) {
2592                 char haddr[MAX_ADDR_LEN];
2593
2594                 neigh_ha_snapshot(haddr, neigh, neigh->dev);
2595                 if (nla_put(skb, NDA_LLADDR, neigh->dev->addr_len, haddr) < 0) {
2596                         read_unlock_bh(&neigh->lock);
2597                         goto nla_put_failure;
2598                 }
2599         }
2600
2601         ci.ndm_used      = jiffies_to_clock_t(now - neigh->used);
2602         ci.ndm_confirmed = jiffies_to_clock_t(now - neigh->confirmed);
2603         ci.ndm_updated   = jiffies_to_clock_t(now - neigh->updated);
2604         ci.ndm_refcnt    = refcount_read(&neigh->refcnt) - 1;
2605         read_unlock_bh(&neigh->lock);
2606
2607         if (nla_put_u32(skb, NDA_PROBES, atomic_read(&neigh->probes)) ||
2608             nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
2609                 goto nla_put_failure;
2610
2611         if (neigh->protocol && nla_put_u8(skb, NDA_PROTOCOL, neigh->protocol))
2612                 goto nla_put_failure;
2613         if (neigh_flags_ext && nla_put_u32(skb, NDA_FLAGS_EXT, neigh_flags_ext))
2614                 goto nla_put_failure;
2615
2616         nlmsg_end(skb, nlh);
2617         return 0;
2618
2619 nla_put_failure:
2620         nlmsg_cancel(skb, nlh);
2621         return -EMSGSIZE;
2622 }
2623
2624 static int pneigh_fill_info(struct sk_buff *skb, struct pneigh_entry *pn,
2625                             u32 pid, u32 seq, int type, unsigned int flags,
2626                             struct neigh_table *tbl)
2627 {
2628         u32 neigh_flags, neigh_flags_ext;
2629         struct nlmsghdr *nlh;
2630         struct ndmsg *ndm;
2631
2632         nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2633         if (nlh == NULL)
2634                 return -EMSGSIZE;
2635
2636         neigh_flags_ext = pn->flags >> NTF_EXT_SHIFT;
2637         neigh_flags     = pn->flags & NTF_OLD_MASK;
2638
2639         ndm = nlmsg_data(nlh);
2640         ndm->ndm_family  = tbl->family;
2641         ndm->ndm_pad1    = 0;
2642         ndm->ndm_pad2    = 0;
2643         ndm->ndm_flags   = neigh_flags | NTF_PROXY;
2644         ndm->ndm_type    = RTN_UNICAST;
2645         ndm->ndm_ifindex = pn->dev ? pn->dev->ifindex : 0;
2646         ndm->ndm_state   = NUD_NONE;
2647
2648         if (nla_put(skb, NDA_DST, tbl->key_len, pn->key))
2649                 goto nla_put_failure;
2650
2651         if (pn->protocol && nla_put_u8(skb, NDA_PROTOCOL, pn->protocol))
2652                 goto nla_put_failure;
2653         if (neigh_flags_ext && nla_put_u32(skb, NDA_FLAGS_EXT, neigh_flags_ext))
2654                 goto nla_put_failure;
2655
2656         nlmsg_end(skb, nlh);
2657         return 0;
2658
2659 nla_put_failure:
2660         nlmsg_cancel(skb, nlh);
2661         return -EMSGSIZE;
2662 }
2663
2664 static void neigh_update_notify(struct neighbour *neigh, u32 nlmsg_pid)
2665 {
2666         call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, neigh);
2667         __neigh_notify(neigh, RTM_NEWNEIGH, 0, nlmsg_pid);
2668 }
2669
2670 static bool neigh_master_filtered(struct net_device *dev, int master_idx)
2671 {
2672         struct net_device *master;
2673
2674         if (!master_idx)
2675                 return false;
2676
2677         master = dev ? netdev_master_upper_dev_get(dev) : NULL;
2678
2679         /* 0 is already used to denote NDA_MASTER wasn't passed, therefore need another
2680          * invalid value for ifindex to denote "no master".
2681          */
2682         if (master_idx == -1)
2683                 return !!master;
2684
2685         if (!master || master->ifindex != master_idx)
2686                 return true;
2687
2688         return false;
2689 }
2690
2691 static bool neigh_ifindex_filtered(struct net_device *dev, int filter_idx)
2692 {
2693         if (filter_idx && (!dev || dev->ifindex != filter_idx))
2694                 return true;
2695
2696         return false;
2697 }
2698
2699 struct neigh_dump_filter {
2700         int master_idx;
2701         int dev_idx;
2702 };
2703
2704 static int neigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2705                             struct netlink_callback *cb,
2706                             struct neigh_dump_filter *filter)
2707 {
2708         struct net *net = sock_net(skb->sk);
2709         struct neighbour *n;
2710         int rc, h, s_h = cb->args[1];
2711         int idx, s_idx = idx = cb->args[2];
2712         struct neigh_hash_table *nht;
2713         unsigned int flags = NLM_F_MULTI;
2714
2715         if (filter->dev_idx || filter->master_idx)
2716                 flags |= NLM_F_DUMP_FILTERED;
2717
2718         rcu_read_lock();
2719         nht = rcu_dereference(tbl->nht);
2720
2721         for (h = s_h; h < (1 << nht->hash_shift); h++) {
2722                 if (h > s_h)
2723                         s_idx = 0;
2724                 for (n = rcu_dereference(nht->hash_buckets[h]), idx = 0;
2725                      n != NULL;
2726                      n = rcu_dereference(n->next)) {
2727                         if (idx < s_idx || !net_eq(dev_net(n->dev), net))
2728                                 goto next;
2729                         if (neigh_ifindex_filtered(n->dev, filter->dev_idx) ||
2730                             neigh_master_filtered(n->dev, filter->master_idx))
2731                                 goto next;
2732                         if (neigh_fill_info(skb, n, NETLINK_CB(cb->skb).portid,
2733                                             cb->nlh->nlmsg_seq,
2734                                             RTM_NEWNEIGH,
2735                                             flags) < 0) {
2736                                 rc = -1;
2737                                 goto out;
2738                         }
2739 next:
2740                         idx++;
2741                 }
2742         }
2743         rc = skb->len;
2744 out:
2745         rcu_read_unlock();
2746         cb->args[1] = h;
2747         cb->args[2] = idx;
2748         return rc;
2749 }
2750
2751 static int pneigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2752                              struct netlink_callback *cb,
2753                              struct neigh_dump_filter *filter)
2754 {
2755         struct pneigh_entry *n;
2756         struct net *net = sock_net(skb->sk);
2757         int rc, h, s_h = cb->args[3];
2758         int idx, s_idx = idx = cb->args[4];
2759         unsigned int flags = NLM_F_MULTI;
2760
2761         if (filter->dev_idx || filter->master_idx)
2762                 flags |= NLM_F_DUMP_FILTERED;
2763
2764         read_lock_bh(&tbl->lock);
2765
2766         for (h = s_h; h <= PNEIGH_HASHMASK; h++) {
2767                 if (h > s_h)
2768                         s_idx = 0;
2769                 for (n = tbl->phash_buckets[h], idx = 0; n; n = n->next) {
2770                         if (idx < s_idx || pneigh_net(n) != net)
2771                                 goto next;
2772                         if (neigh_ifindex_filtered(n->dev, filter->dev_idx) ||
2773                             neigh_master_filtered(n->dev, filter->master_idx))
2774                                 goto next;
2775                         if (pneigh_fill_info(skb, n, NETLINK_CB(cb->skb).portid,
2776                                             cb->nlh->nlmsg_seq,
2777                                             RTM_NEWNEIGH, flags, tbl) < 0) {
2778                                 read_unlock_bh(&tbl->lock);
2779                                 rc = -1;
2780                                 goto out;
2781                         }
2782                 next:
2783                         idx++;
2784                 }
2785         }
2786
2787         read_unlock_bh(&tbl->lock);
2788         rc = skb->len;
2789 out:
2790         cb->args[3] = h;
2791         cb->args[4] = idx;
2792         return rc;
2793
2794 }
2795
2796 static int neigh_valid_dump_req(const struct nlmsghdr *nlh,
2797                                 bool strict_check,
2798                                 struct neigh_dump_filter *filter,
2799                                 struct netlink_ext_ack *extack)
2800 {
2801         struct nlattr *tb[NDA_MAX + 1];
2802         int err, i;
2803
2804         if (strict_check) {
2805                 struct ndmsg *ndm;
2806
2807                 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ndm))) {
2808                         NL_SET_ERR_MSG(extack, "Invalid header for neighbor dump request");
2809                         return -EINVAL;
2810                 }
2811
2812                 ndm = nlmsg_data(nlh);
2813                 if (ndm->ndm_pad1  || ndm->ndm_pad2  || ndm->ndm_ifindex ||
2814                     ndm->ndm_state || ndm->ndm_type) {
2815                         NL_SET_ERR_MSG(extack, "Invalid values in header for neighbor dump request");
2816                         return -EINVAL;
2817                 }
2818
2819                 if (ndm->ndm_flags & ~NTF_PROXY) {
2820                         NL_SET_ERR_MSG(extack, "Invalid flags in header for neighbor dump request");
2821                         return -EINVAL;
2822                 }
2823
2824                 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct ndmsg),
2825                                                     tb, NDA_MAX, nda_policy,
2826                                                     extack);
2827         } else {
2828                 err = nlmsg_parse_deprecated(nlh, sizeof(struct ndmsg), tb,
2829                                              NDA_MAX, nda_policy, extack);
2830         }
2831         if (err < 0)
2832                 return err;
2833
2834         for (i = 0; i <= NDA_MAX; ++i) {
2835                 if (!tb[i])
2836                         continue;
2837
2838                 /* all new attributes should require strict_check */
2839                 switch (i) {
2840                 case NDA_IFINDEX:
2841                         filter->dev_idx = nla_get_u32(tb[i]);
2842                         break;
2843                 case NDA_MASTER:
2844                         filter->master_idx = nla_get_u32(tb[i]);
2845                         break;
2846                 default:
2847                         if (strict_check) {
2848                                 NL_SET_ERR_MSG(extack, "Unsupported attribute in neighbor dump request");
2849                                 return -EINVAL;
2850                         }
2851                 }
2852         }
2853
2854         return 0;
2855 }
2856
2857 static int neigh_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2858 {
2859         const struct nlmsghdr *nlh = cb->nlh;
2860         struct neigh_dump_filter filter = {};
2861         struct neigh_table *tbl;
2862         int t, family, s_t;
2863         int proxy = 0;
2864         int err;
2865
2866         family = ((struct rtgenmsg *)nlmsg_data(nlh))->rtgen_family;
2867
2868         /* check for full ndmsg structure presence, family member is
2869          * the same for both structures
2870          */
2871         if (nlmsg_len(nlh) >= sizeof(struct ndmsg) &&
2872             ((struct ndmsg *)nlmsg_data(nlh))->ndm_flags == NTF_PROXY)
2873                 proxy = 1;
2874
2875         err = neigh_valid_dump_req(nlh, cb->strict_check, &filter, cb->extack);
2876         if (err < 0 && cb->strict_check)
2877                 return err;
2878
2879         s_t = cb->args[0];
2880
2881         for (t = 0; t < NEIGH_NR_TABLES; t++) {
2882                 tbl = neigh_tables[t];
2883
2884                 if (!tbl)
2885                         continue;
2886                 if (t < s_t || (family && tbl->family != family))
2887                         continue;
2888                 if (t > s_t)
2889                         memset(&cb->args[1], 0, sizeof(cb->args) -
2890                                                 sizeof(cb->args[0]));
2891                 if (proxy)
2892                         err = pneigh_dump_table(tbl, skb, cb, &filter);
2893                 else
2894                         err = neigh_dump_table(tbl, skb, cb, &filter);
2895                 if (err < 0)
2896                         break;
2897         }
2898
2899         cb->args[0] = t;
2900         return skb->len;
2901 }
2902
2903 static int neigh_valid_get_req(const struct nlmsghdr *nlh,
2904                                struct neigh_table **tbl,
2905                                void **dst, int *dev_idx, u8 *ndm_flags,
2906                                struct netlink_ext_ack *extack)
2907 {
2908         struct nlattr *tb[NDA_MAX + 1];
2909         struct ndmsg *ndm;
2910         int err, i;
2911
2912         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ndm))) {
2913                 NL_SET_ERR_MSG(extack, "Invalid header for neighbor get request");
2914                 return -EINVAL;
2915         }
2916
2917         ndm = nlmsg_data(nlh);
2918         if (ndm->ndm_pad1  || ndm->ndm_pad2  || ndm->ndm_state ||
2919             ndm->ndm_type) {
2920                 NL_SET_ERR_MSG(extack, "Invalid values in header for neighbor get request");
2921                 return -EINVAL;
2922         }
2923
2924         if (ndm->ndm_flags & ~NTF_PROXY) {
2925                 NL_SET_ERR_MSG(extack, "Invalid flags in header for neighbor get request");
2926                 return -EINVAL;
2927         }
2928
2929         err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct ndmsg), tb,
2930                                             NDA_MAX, nda_policy, extack);
2931         if (err < 0)
2932                 return err;
2933
2934         *ndm_flags = ndm->ndm_flags;
2935         *dev_idx = ndm->ndm_ifindex;
2936         *tbl = neigh_find_table(ndm->ndm_family);
2937         if (*tbl == NULL) {
2938                 NL_SET_ERR_MSG(extack, "Unsupported family in header for neighbor get request");
2939                 return -EAFNOSUPPORT;
2940         }
2941
2942         for (i = 0; i <= NDA_MAX; ++i) {
2943                 if (!tb[i])
2944                         continue;
2945
2946                 switch (i) {
2947                 case NDA_DST:
2948                         if (nla_len(tb[i]) != (int)(*tbl)->key_len) {
2949                                 NL_SET_ERR_MSG(extack, "Invalid network address in neighbor get request");
2950                                 return -EINVAL;
2951                         }
2952                         *dst = nla_data(tb[i]);
2953                         break;
2954                 default:
2955                         NL_SET_ERR_MSG(extack, "Unsupported attribute in neighbor get request");
2956                         return -EINVAL;
2957                 }
2958         }
2959
2960         return 0;
2961 }
2962
2963 static inline size_t neigh_nlmsg_size(void)
2964 {
2965         return NLMSG_ALIGN(sizeof(struct ndmsg))
2966                + nla_total_size(MAX_ADDR_LEN) /* NDA_DST */
2967                + nla_total_size(MAX_ADDR_LEN) /* NDA_LLADDR */
2968                + nla_total_size(sizeof(struct nda_cacheinfo))
2969                + nla_total_size(4)  /* NDA_PROBES */
2970                + nla_total_size(4)  /* NDA_FLAGS_EXT */
2971                + nla_total_size(1); /* NDA_PROTOCOL */
2972 }
2973
2974 static int neigh_get_reply(struct net *net, struct neighbour *neigh,
2975                            u32 pid, u32 seq)
2976 {
2977         struct sk_buff *skb;
2978         int err = 0;
2979
2980         skb = nlmsg_new(neigh_nlmsg_size(), GFP_KERNEL);
2981         if (!skb)
2982                 return -ENOBUFS;
2983
2984         err = neigh_fill_info(skb, neigh, pid, seq, RTM_NEWNEIGH, 0);
2985         if (err) {
2986                 kfree_skb(skb);
2987                 goto errout;
2988         }
2989
2990         err = rtnl_unicast(skb, net, pid);
2991 errout:
2992         return err;
2993 }
2994
2995 static inline size_t pneigh_nlmsg_size(void)
2996 {
2997         return NLMSG_ALIGN(sizeof(struct ndmsg))
2998                + nla_total_size(MAX_ADDR_LEN) /* NDA_DST */
2999                + nla_total_size(4)  /* NDA_FLAGS_EXT */
3000                + nla_total_size(1); /* NDA_PROTOCOL */
3001 }
3002
3003 static int pneigh_get_reply(struct net *net, struct pneigh_entry *neigh,
3004                             u32 pid, u32 seq, struct neigh_table *tbl)
3005 {
3006         struct sk_buff *skb;
3007         int err = 0;
3008
3009         skb = nlmsg_new(pneigh_nlmsg_size(), GFP_KERNEL);
3010         if (!skb)
3011                 return -ENOBUFS;
3012
3013         err = pneigh_fill_info(skb, neigh, pid, seq, RTM_NEWNEIGH, 0, tbl);
3014         if (err) {
3015                 kfree_skb(skb);
3016                 goto errout;
3017         }
3018
3019         err = rtnl_unicast(skb, net, pid);
3020 errout:
3021         return err;
3022 }
3023
3024 static int neigh_get(struct sk_buff *in_skb, struct nlmsghdr *nlh,
3025                      struct netlink_ext_ack *extack)
3026 {
3027         struct net *net = sock_net(in_skb->sk);
3028         struct net_device *dev = NULL;
3029         struct neigh_table *tbl = NULL;
3030         struct neighbour *neigh;
3031         void *dst = NULL;
3032         u8 ndm_flags = 0;
3033         int dev_idx = 0;
3034         int err;
3035
3036         err = neigh_valid_get_req(nlh, &tbl, &dst, &dev_idx, &ndm_flags,
3037                                   extack);
3038         if (err < 0)
3039                 return err;
3040
3041         if (dev_idx) {
3042                 dev = __dev_get_by_index(net, dev_idx);
3043                 if (!dev) {
3044                         NL_SET_ERR_MSG(extack, "Unknown device ifindex");
3045                         return -ENODEV;
3046                 }
3047         }
3048
3049         if (!dst) {
3050                 NL_SET_ERR_MSG(extack, "Network address not specified");
3051                 return -EINVAL;
3052         }
3053
3054         if (ndm_flags & NTF_PROXY) {
3055                 struct pneigh_entry *pn;
3056
3057                 pn = pneigh_lookup(tbl, net, dst, dev, 0);
3058                 if (!pn) {
3059                         NL_SET_ERR_MSG(extack, "Proxy neighbour entry not found");
3060                         return -ENOENT;
3061                 }
3062                 return pneigh_get_reply(net, pn, NETLINK_CB(in_skb).portid,
3063                                         nlh->nlmsg_seq, tbl);
3064         }
3065
3066         if (!dev) {
3067                 NL_SET_ERR_MSG(extack, "No device specified");
3068                 return -EINVAL;
3069         }
3070
3071         neigh = neigh_lookup(tbl, dst, dev);
3072         if (!neigh) {
3073                 NL_SET_ERR_MSG(extack, "Neighbour entry not found");
3074                 return -ENOENT;
3075         }
3076
3077         err = neigh_get_reply(net, neigh, NETLINK_CB(in_skb).portid,
3078                               nlh->nlmsg_seq);
3079
3080         neigh_release(neigh);
3081
3082         return err;
3083 }
3084
3085 void neigh_for_each(struct neigh_table *tbl, void (*cb)(struct neighbour *, void *), void *cookie)
3086 {
3087         int chain;
3088         struct neigh_hash_table *nht;
3089
3090         rcu_read_lock();
3091         nht = rcu_dereference(tbl->nht);
3092
3093         read_lock_bh(&tbl->lock); /* avoid resizes */
3094         for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
3095                 struct neighbour *n;
3096
3097                 for (n = rcu_dereference(nht->hash_buckets[chain]);
3098                      n != NULL;
3099                      n = rcu_dereference(n->next))
3100                         cb(n, cookie);
3101         }
3102         read_unlock_bh(&tbl->lock);
3103         rcu_read_unlock();
3104 }
3105 EXPORT_SYMBOL(neigh_for_each);
3106
3107 /* The tbl->lock must be held as a writer and BH disabled. */
3108 void __neigh_for_each_release(struct neigh_table *tbl,
3109                               int (*cb)(struct neighbour *))
3110 {
3111         int chain;
3112         struct neigh_hash_table *nht;
3113
3114         nht = rcu_dereference_protected(tbl->nht,
3115                                         lockdep_is_held(&tbl->lock));
3116         for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
3117                 struct neighbour *n;
3118                 struct neighbour __rcu **np;
3119
3120                 np = &nht->hash_buckets[chain];
3121                 while ((n = rcu_dereference_protected(*np,
3122                                         lockdep_is_held(&tbl->lock))) != NULL) {
3123                         int release;
3124
3125                         write_lock(&n->lock);
3126                         release = cb(n);
3127                         if (release) {
3128                                 rcu_assign_pointer(*np,
3129                                         rcu_dereference_protected(n->next,
3130                                                 lockdep_is_held(&tbl->lock)));
3131                                 neigh_mark_dead(n);
3132                         } else
3133                                 np = &n->next;
3134                         write_unlock(&n->lock);
3135                         if (release)
3136                                 neigh_cleanup_and_release(n);
3137                 }
3138         }
3139 }
3140 EXPORT_SYMBOL(__neigh_for_each_release);
3141
3142 int neigh_xmit(int index, struct net_device *dev,
3143                const void *addr, struct sk_buff *skb)
3144 {
3145         int err = -EAFNOSUPPORT;
3146         if (likely(index < NEIGH_NR_TABLES)) {
3147                 struct neigh_table *tbl;
3148                 struct neighbour *neigh;
3149
3150                 tbl = neigh_tables[index];
3151                 if (!tbl)
3152                         goto out;
3153                 rcu_read_lock();
3154                 if (index == NEIGH_ARP_TABLE) {
3155                         u32 key = *((u32 *)addr);
3156
3157                         neigh = __ipv4_neigh_lookup_noref(dev, key);
3158                 } else {
3159                         neigh = __neigh_lookup_noref(tbl, addr, dev);
3160                 }
3161                 if (!neigh)
3162                         neigh = __neigh_create(tbl, addr, dev, false);
3163                 err = PTR_ERR(neigh);
3164                 if (IS_ERR(neigh)) {
3165                         rcu_read_unlock();
3166                         goto out_kfree_skb;
3167                 }
3168                 err = READ_ONCE(neigh->output)(neigh, skb);
3169                 rcu_read_unlock();
3170         }
3171         else if (index == NEIGH_LINK_TABLE) {
3172                 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
3173                                       addr, NULL, skb->len);
3174                 if (err < 0)
3175                         goto out_kfree_skb;
3176                 err = dev_queue_xmit(skb);
3177         }
3178 out:
3179         return err;
3180 out_kfree_skb:
3181         kfree_skb(skb);
3182         goto out;
3183 }
3184 EXPORT_SYMBOL(neigh_xmit);
3185
3186 #ifdef CONFIG_PROC_FS
3187
3188 static struct neighbour *neigh_get_first(struct seq_file *seq)
3189 {
3190         struct neigh_seq_state *state = seq->private;
3191         struct net *net = seq_file_net(seq);
3192         struct neigh_hash_table *nht = state->nht;
3193         struct neighbour *n = NULL;
3194         int bucket;
3195
3196         state->flags &= ~NEIGH_SEQ_IS_PNEIGH;
3197         for (bucket = 0; bucket < (1 << nht->hash_shift); bucket++) {
3198                 n = rcu_dereference(nht->hash_buckets[bucket]);
3199
3200                 while (n) {
3201                         if (!net_eq(dev_net(n->dev), net))
3202                                 goto next;
3203                         if (state->neigh_sub_iter) {
3204                                 loff_t fakep = 0;
3205                                 void *v;
3206
3207                                 v = state->neigh_sub_iter(state, n, &fakep);
3208                                 if (!v)
3209                                         goto next;
3210                         }
3211                         if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
3212                                 break;
3213                         if (READ_ONCE(n->nud_state) & ~NUD_NOARP)
3214                                 break;
3215 next:
3216                         n = rcu_dereference(n->next);
3217                 }
3218
3219                 if (n)
3220                         break;
3221         }
3222         state->bucket = bucket;
3223
3224         return n;
3225 }
3226
3227 static struct neighbour *neigh_get_next(struct seq_file *seq,
3228                                         struct neighbour *n,
3229                                         loff_t *pos)
3230 {
3231         struct neigh_seq_state *state = seq->private;
3232         struct net *net = seq_file_net(seq);
3233         struct neigh_hash_table *nht = state->nht;
3234
3235         if (state->neigh_sub_iter) {
3236                 void *v = state->neigh_sub_iter(state, n, pos);
3237                 if (v)
3238                         return n;
3239         }
3240         n = rcu_dereference(n->next);
3241
3242         while (1) {
3243                 while (n) {
3244                         if (!net_eq(dev_net(n->dev), net))
3245                                 goto next;
3246                         if (state->neigh_sub_iter) {
3247                                 void *v = state->neigh_sub_iter(state, n, pos);
3248                                 if (v)
3249                                         return n;
3250                                 goto next;
3251                         }
3252                         if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
3253                                 break;
3254
3255                         if (READ_ONCE(n->nud_state) & ~NUD_NOARP)
3256                                 break;
3257 next:
3258                         n = rcu_dereference(n->next);
3259                 }
3260
3261                 if (n)
3262                         break;
3263
3264                 if (++state->bucket >= (1 << nht->hash_shift))
3265                         break;
3266
3267                 n = rcu_dereference(nht->hash_buckets[state->bucket]);
3268         }
3269
3270         if (n && pos)
3271                 --(*pos);
3272         return n;
3273 }
3274
3275 static struct neighbour *neigh_get_idx(struct seq_file *seq, loff_t *pos)
3276 {
3277         struct neighbour *n = neigh_get_first(seq);
3278
3279         if (n) {
3280                 --(*pos);
3281                 while (*pos) {
3282                         n = neigh_get_next(seq, n, pos);
3283                         if (!n)
3284                                 break;
3285                 }
3286         }
3287         return *pos ? NULL : n;
3288 }
3289
3290 static struct pneigh_entry *pneigh_get_first(struct seq_file *seq)
3291 {
3292         struct neigh_seq_state *state = seq->private;
3293         struct net *net = seq_file_net(seq);
3294         struct neigh_table *tbl = state->tbl;
3295         struct pneigh_entry *pn = NULL;
3296         int bucket;
3297
3298         state->flags |= NEIGH_SEQ_IS_PNEIGH;
3299         for (bucket = 0; bucket <= PNEIGH_HASHMASK; bucket++) {
3300                 pn = tbl->phash_buckets[bucket];
3301                 while (pn && !net_eq(pneigh_net(pn), net))
3302                         pn = pn->next;
3303                 if (pn)
3304                         break;
3305         }
3306         state->bucket = bucket;
3307
3308         return pn;
3309 }
3310
3311 static struct pneigh_entry *pneigh_get_next(struct seq_file *seq,
3312                                             struct pneigh_entry *pn,
3313                                             loff_t *pos)
3314 {
3315         struct neigh_seq_state *state = seq->private;
3316         struct net *net = seq_file_net(seq);
3317         struct neigh_table *tbl = state->tbl;
3318
3319         do {
3320                 pn = pn->next;
3321         } while (pn && !net_eq(pneigh_net(pn), net));
3322
3323         while (!pn) {
3324                 if (++state->bucket > PNEIGH_HASHMASK)
3325                         break;
3326                 pn = tbl->phash_buckets[state->bucket];
3327                 while (pn && !net_eq(pneigh_net(pn), net))
3328                         pn = pn->next;
3329                 if (pn)
3330                         break;
3331         }
3332
3333         if (pn && pos)
3334                 --(*pos);
3335
3336         return pn;
3337 }
3338
3339 static struct pneigh_entry *pneigh_get_idx(struct seq_file *seq, loff_t *pos)
3340 {
3341         struct pneigh_entry *pn = pneigh_get_first(seq);
3342
3343         if (pn) {
3344                 --(*pos);
3345                 while (*pos) {
3346                         pn = pneigh_get_next(seq, pn, pos);
3347                         if (!pn)
3348                                 break;
3349                 }
3350         }
3351         return *pos ? NULL : pn;
3352 }
3353
3354 static void *neigh_get_idx_any(struct seq_file *seq, loff_t *pos)
3355 {
3356         struct neigh_seq_state *state = seq->private;
3357         void *rc;
3358         loff_t idxpos = *pos;
3359
3360         rc = neigh_get_idx(seq, &idxpos);
3361         if (!rc && !(state->flags & NEIGH_SEQ_NEIGH_ONLY))
3362                 rc = pneigh_get_idx(seq, &idxpos);
3363
3364         return rc;
3365 }
3366
3367 void *neigh_seq_start(struct seq_file *seq, loff_t *pos, struct neigh_table *tbl, unsigned int neigh_seq_flags)
3368         __acquires(tbl->lock)
3369         __acquires(rcu)
3370 {
3371         struct neigh_seq_state *state = seq->private;
3372
3373         state->tbl = tbl;
3374         state->bucket = 0;
3375         state->flags = (neigh_seq_flags & ~NEIGH_SEQ_IS_PNEIGH);
3376
3377         rcu_read_lock();
3378         state->nht = rcu_dereference(tbl->nht);
3379         read_lock_bh(&tbl->lock);
3380
3381         return *pos ? neigh_get_idx_any(seq, pos) : SEQ_START_TOKEN;
3382 }
3383 EXPORT_SYMBOL(neigh_seq_start);
3384
3385 void *neigh_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3386 {
3387         struct neigh_seq_state *state;
3388         void *rc;
3389
3390         if (v == SEQ_START_TOKEN) {
3391                 rc = neigh_get_first(seq);
3392                 goto out;
3393         }
3394
3395         state = seq->private;
3396         if (!(state->flags & NEIGH_SEQ_IS_PNEIGH)) {
3397                 rc = neigh_get_next(seq, v, NULL);
3398                 if (rc)
3399                         goto out;
3400                 if (!(state->flags & NEIGH_SEQ_NEIGH_ONLY))
3401                         rc = pneigh_get_first(seq);
3402         } else {
3403                 BUG_ON(state->flags & NEIGH_SEQ_NEIGH_ONLY);
3404                 rc = pneigh_get_next(seq, v, NULL);
3405         }
3406 out:
3407         ++(*pos);
3408         return rc;
3409 }
3410 EXPORT_SYMBOL(neigh_seq_next);
3411
3412 void neigh_seq_stop(struct seq_file *seq, void *v)
3413         __releases(tbl->lock)
3414         __releases(rcu)
3415 {
3416         struct neigh_seq_state *state = seq->private;
3417         struct neigh_table *tbl = state->tbl;
3418
3419         read_unlock_bh(&tbl->lock);
3420         rcu_read_unlock();
3421 }
3422 EXPORT_SYMBOL(neigh_seq_stop);
3423
3424 /* statistics via seq_file */
3425
3426 static void *neigh_stat_seq_start(struct seq_file *seq, loff_t *pos)
3427 {
3428         struct neigh_table *tbl = pde_data(file_inode(seq->file));
3429         int cpu;
3430
3431         if (*pos == 0)
3432                 return SEQ_START_TOKEN;
3433
3434         for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
3435                 if (!cpu_possible(cpu))
3436                         continue;
3437                 *pos = cpu+1;
3438                 return per_cpu_ptr(tbl->stats, cpu);
3439         }
3440         return NULL;
3441 }
3442
3443 static void *neigh_stat_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3444 {
3445         struct neigh_table *tbl = pde_data(file_inode(seq->file));
3446         int cpu;
3447
3448         for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
3449                 if (!cpu_possible(cpu))
3450                         continue;
3451                 *pos = cpu+1;
3452                 return per_cpu_ptr(tbl->stats, cpu);
3453         }
3454         (*pos)++;
3455         return NULL;
3456 }
3457
3458 static void neigh_stat_seq_stop(struct seq_file *seq, void *v)
3459 {
3460
3461 }
3462
3463 static int neigh_stat_seq_show(struct seq_file *seq, void *v)
3464 {
3465         struct neigh_table *tbl = pde_data(file_inode(seq->file));
3466         struct neigh_statistics *st = v;
3467
3468         if (v == SEQ_START_TOKEN) {
3469                 seq_puts(seq, "entries  allocs   destroys hash_grows lookups  hits     res_failed rcv_probes_mcast rcv_probes_ucast periodic_gc_runs forced_gc_runs unresolved_discards table_fulls\n");
3470                 return 0;
3471         }
3472
3473         seq_printf(seq, "%08x %08lx %08lx %08lx   %08lx %08lx %08lx   "
3474                         "%08lx         %08lx         %08lx         "
3475                         "%08lx       %08lx            %08lx\n",
3476                    atomic_read(&tbl->entries),
3477
3478                    st->allocs,
3479                    st->destroys,
3480                    st->hash_grows,
3481
3482                    st->lookups,
3483                    st->hits,
3484
3485                    st->res_failed,
3486
3487                    st->rcv_probes_mcast,
3488                    st->rcv_probes_ucast,
3489
3490                    st->periodic_gc_runs,
3491                    st->forced_gc_runs,
3492                    st->unres_discards,
3493                    st->table_fulls
3494                    );
3495
3496         return 0;
3497 }
3498
3499 static const struct seq_operations neigh_stat_seq_ops = {
3500         .start  = neigh_stat_seq_start,
3501         .next   = neigh_stat_seq_next,
3502         .stop   = neigh_stat_seq_stop,
3503         .show   = neigh_stat_seq_show,
3504 };
3505 #endif /* CONFIG_PROC_FS */
3506
3507 static void __neigh_notify(struct neighbour *n, int type, int flags,
3508                            u32 pid)
3509 {
3510         struct net *net = dev_net(n->dev);
3511         struct sk_buff *skb;
3512         int err = -ENOBUFS;
3513
3514         skb = nlmsg_new(neigh_nlmsg_size(), GFP_ATOMIC);
3515         if (skb == NULL)
3516                 goto errout;
3517
3518         err = neigh_fill_info(skb, n, pid, 0, type, flags);
3519         if (err < 0) {
3520                 /* -EMSGSIZE implies BUG in neigh_nlmsg_size() */
3521                 WARN_ON(err == -EMSGSIZE);
3522                 kfree_skb(skb);
3523                 goto errout;
3524         }
3525         rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
3526         return;
3527 errout:
3528         if (err < 0)
3529                 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
3530 }
3531
3532 void neigh_app_ns(struct neighbour *n)
3533 {
3534         __neigh_notify(n, RTM_GETNEIGH, NLM_F_REQUEST, 0);
3535 }
3536 EXPORT_SYMBOL(neigh_app_ns);
3537
3538 #ifdef CONFIG_SYSCTL
3539 static int unres_qlen_max = INT_MAX / SKB_TRUESIZE(ETH_FRAME_LEN);
3540
3541 static int proc_unres_qlen(struct ctl_table *ctl, int write,
3542                            void *buffer, size_t *lenp, loff_t *ppos)
3543 {
3544         int size, ret;
3545         struct ctl_table tmp = *ctl;
3546
3547         tmp.extra1 = SYSCTL_ZERO;
3548         tmp.extra2 = &unres_qlen_max;
3549         tmp.data = &size;
3550
3551         size = *(int *)ctl->data / SKB_TRUESIZE(ETH_FRAME_LEN);
3552         ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
3553
3554         if (write && !ret)
3555                 *(int *)ctl->data = size * SKB_TRUESIZE(ETH_FRAME_LEN);
3556         return ret;
3557 }
3558
3559 static void neigh_copy_dflt_parms(struct net *net, struct neigh_parms *p,
3560                                   int index)
3561 {
3562         struct net_device *dev;
3563         int family = neigh_parms_family(p);
3564
3565         rcu_read_lock();
3566         for_each_netdev_rcu(net, dev) {
3567                 struct neigh_parms *dst_p =
3568                                 neigh_get_dev_parms_rcu(dev, family);
3569
3570                 if (dst_p && !test_bit(index, dst_p->data_state))
3571                         dst_p->data[index] = p->data[index];
3572         }
3573         rcu_read_unlock();
3574 }
3575
3576 static void neigh_proc_update(struct ctl_table *ctl, int write)
3577 {
3578         struct net_device *dev = ctl->extra1;
3579         struct neigh_parms *p = ctl->extra2;
3580         struct net *net = neigh_parms_net(p);
3581         int index = (int *) ctl->data - p->data;
3582
3583         if (!write)
3584                 return;
3585
3586         set_bit(index, p->data_state);
3587         if (index == NEIGH_VAR_DELAY_PROBE_TIME)
3588                 call_netevent_notifiers(NETEVENT_DELAY_PROBE_TIME_UPDATE, p);
3589         if (!dev) /* NULL dev means this is default value */
3590                 neigh_copy_dflt_parms(net, p, index);
3591 }
3592
3593 static int neigh_proc_dointvec_zero_intmax(struct ctl_table *ctl, int write,
3594                                            void *buffer, size_t *lenp,
3595                                            loff_t *ppos)
3596 {
3597         struct ctl_table tmp = *ctl;
3598         int ret;
3599
3600         tmp.extra1 = SYSCTL_ZERO;
3601         tmp.extra2 = SYSCTL_INT_MAX;
3602
3603         ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
3604         neigh_proc_update(ctl, write);
3605         return ret;
3606 }
3607
3608 static int neigh_proc_dointvec_ms_jiffies_positive(struct ctl_table *ctl, int write,
3609                                                    void *buffer, size_t *lenp, loff_t *ppos)
3610 {
3611         struct ctl_table tmp = *ctl;
3612         int ret;
3613
3614         int min = msecs_to_jiffies(1);
3615
3616         tmp.extra1 = &min;
3617         tmp.extra2 = NULL;
3618
3619         ret = proc_dointvec_ms_jiffies_minmax(&tmp, write, buffer, lenp, ppos);
3620         neigh_proc_update(ctl, write);
3621         return ret;
3622 }
3623
3624 int neigh_proc_dointvec(struct ctl_table *ctl, int write, void *buffer,
3625                         size_t *lenp, loff_t *ppos)
3626 {
3627         int ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
3628
3629         neigh_proc_update(ctl, write);
3630         return ret;
3631 }
3632 EXPORT_SYMBOL(neigh_proc_dointvec);
3633
3634 int neigh_proc_dointvec_jiffies(struct ctl_table *ctl, int write, void *buffer,
3635                                 size_t *lenp, loff_t *ppos)
3636 {
3637         int ret = proc_dointvec_jiffies(ctl, write, buffer, lenp, ppos);
3638
3639         neigh_proc_update(ctl, write);
3640         return ret;
3641 }
3642 EXPORT_SYMBOL(neigh_proc_dointvec_jiffies);
3643
3644 static int neigh_proc_dointvec_userhz_jiffies(struct ctl_table *ctl, int write,
3645                                               void *buffer, size_t *lenp,
3646                                               loff_t *ppos)
3647 {
3648         int ret = proc_dointvec_userhz_jiffies(ctl, write, buffer, lenp, ppos);
3649
3650         neigh_proc_update(ctl, write);
3651         return ret;
3652 }
3653
3654 int neigh_proc_dointvec_ms_jiffies(struct ctl_table *ctl, int write,
3655                                    void *buffer, size_t *lenp, loff_t *ppos)
3656 {
3657         int ret = proc_dointvec_ms_jiffies(ctl, write, buffer, lenp, ppos);
3658
3659         neigh_proc_update(ctl, write);
3660         return ret;
3661 }
3662 EXPORT_SYMBOL(neigh_proc_dointvec_ms_jiffies);
3663
3664 static int neigh_proc_dointvec_unres_qlen(struct ctl_table *ctl, int write,
3665                                           void *buffer, size_t *lenp,
3666                                           loff_t *ppos)
3667 {
3668         int ret = proc_unres_qlen(ctl, write, buffer, lenp, ppos);
3669
3670         neigh_proc_update(ctl, write);
3671         return ret;
3672 }
3673
3674 static int neigh_proc_base_reachable_time(struct ctl_table *ctl, int write,
3675                                           void *buffer, size_t *lenp,
3676                                           loff_t *ppos)
3677 {
3678         struct neigh_parms *p = ctl->extra2;
3679         int ret;
3680
3681         if (strcmp(ctl->procname, "base_reachable_time") == 0)
3682                 ret = neigh_proc_dointvec_jiffies(ctl, write, buffer, lenp, ppos);
3683         else if (strcmp(ctl->procname, "base_reachable_time_ms") == 0)
3684                 ret = neigh_proc_dointvec_ms_jiffies(ctl, write, buffer, lenp, ppos);
3685         else
3686                 ret = -1;
3687
3688         if (write && ret == 0) {
3689                 /* update reachable_time as well, otherwise, the change will
3690                  * only be effective after the next time neigh_periodic_work
3691                  * decides to recompute it
3692                  */
3693                 p->reachable_time =
3694                         neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
3695         }
3696         return ret;
3697 }
3698
3699 #define NEIGH_PARMS_DATA_OFFSET(index)  \
3700         (&((struct neigh_parms *) 0)->data[index])
3701
3702 #define NEIGH_SYSCTL_ENTRY(attr, data_attr, name, mval, proc) \
3703         [NEIGH_VAR_ ## attr] = { \
3704                 .procname       = name, \
3705                 .data           = NEIGH_PARMS_DATA_OFFSET(NEIGH_VAR_ ## data_attr), \
3706                 .maxlen         = sizeof(int), \
3707                 .mode           = mval, \
3708                 .proc_handler   = proc, \
3709         }
3710
3711 #define NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(attr, name) \
3712         NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_zero_intmax)
3713
3714 #define NEIGH_SYSCTL_JIFFIES_ENTRY(attr, name) \
3715         NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_jiffies)
3716
3717 #define NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(attr, name) \
3718         NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_userhz_jiffies)
3719
3720 #define NEIGH_SYSCTL_MS_JIFFIES_POSITIVE_ENTRY(attr, name) \
3721         NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_ms_jiffies_positive)
3722
3723 #define NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(attr, data_attr, name) \
3724         NEIGH_SYSCTL_ENTRY(attr, data_attr, name, 0644, neigh_proc_dointvec_ms_jiffies)
3725
3726 #define NEIGH_SYSCTL_UNRES_QLEN_REUSED_ENTRY(attr, data_attr, name) \
3727         NEIGH_SYSCTL_ENTRY(attr, data_attr, name, 0644, neigh_proc_dointvec_unres_qlen)
3728
3729 static struct neigh_sysctl_table {
3730         struct ctl_table_header *sysctl_header;
3731         struct ctl_table neigh_vars[NEIGH_VAR_MAX + 1];
3732 } neigh_sysctl_template __read_mostly = {
3733         .neigh_vars = {
3734                 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(MCAST_PROBES, "mcast_solicit"),
3735                 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(UCAST_PROBES, "ucast_solicit"),
3736                 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(APP_PROBES, "app_solicit"),
3737                 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(MCAST_REPROBES, "mcast_resolicit"),
3738                 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(RETRANS_TIME, "retrans_time"),
3739                 NEIGH_SYSCTL_JIFFIES_ENTRY(BASE_REACHABLE_TIME, "base_reachable_time"),
3740                 NEIGH_SYSCTL_JIFFIES_ENTRY(DELAY_PROBE_TIME, "delay_first_probe_time"),
3741                 NEIGH_SYSCTL_MS_JIFFIES_POSITIVE_ENTRY(INTERVAL_PROBE_TIME_MS,
3742                                                        "interval_probe_time_ms"),
3743                 NEIGH_SYSCTL_JIFFIES_ENTRY(GC_STALETIME, "gc_stale_time"),
3744                 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(QUEUE_LEN_BYTES, "unres_qlen_bytes"),
3745                 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(PROXY_QLEN, "proxy_qlen"),
3746                 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(ANYCAST_DELAY, "anycast_delay"),
3747                 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(PROXY_DELAY, "proxy_delay"),
3748                 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(LOCKTIME, "locktime"),
3749                 NEIGH_SYSCTL_UNRES_QLEN_REUSED_ENTRY(QUEUE_LEN, QUEUE_LEN_BYTES, "unres_qlen"),
3750                 NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(RETRANS_TIME_MS, RETRANS_TIME, "retrans_time_ms"),
3751                 NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(BASE_REACHABLE_TIME_MS, BASE_REACHABLE_TIME, "base_reachable_time_ms"),
3752                 [NEIGH_VAR_GC_INTERVAL] = {
3753                         .procname       = "gc_interval",
3754                         .maxlen         = sizeof(int),
3755                         .mode           = 0644,
3756                         .proc_handler   = proc_dointvec_jiffies,
3757                 },
3758                 [NEIGH_VAR_GC_THRESH1] = {
3759                         .procname       = "gc_thresh1",
3760                         .maxlen         = sizeof(int),
3761                         .mode           = 0644,
3762                         .extra1         = SYSCTL_ZERO,
3763                         .extra2         = SYSCTL_INT_MAX,
3764                         .proc_handler   = proc_dointvec_minmax,
3765                 },
3766                 [NEIGH_VAR_GC_THRESH2] = {
3767                         .procname       = "gc_thresh2",
3768                         .maxlen         = sizeof(int),
3769                         .mode           = 0644,
3770                         .extra1         = SYSCTL_ZERO,
3771                         .extra2         = SYSCTL_INT_MAX,
3772                         .proc_handler   = proc_dointvec_minmax,
3773                 },
3774                 [NEIGH_VAR_GC_THRESH3] = {
3775                         .procname       = "gc_thresh3",
3776                         .maxlen         = sizeof(int),
3777                         .mode           = 0644,
3778                         .extra1         = SYSCTL_ZERO,
3779                         .extra2         = SYSCTL_INT_MAX,
3780                         .proc_handler   = proc_dointvec_minmax,
3781                 },
3782                 {},
3783         },
3784 };
3785
3786 int neigh_sysctl_register(struct net_device *dev, struct neigh_parms *p,
3787                           proc_handler *handler)
3788 {
3789         int i;
3790         struct neigh_sysctl_table *t;
3791         const char *dev_name_source;
3792         char neigh_path[ sizeof("net//neigh/") + IFNAMSIZ + IFNAMSIZ ];
3793         char *p_name;
3794         size_t neigh_vars_size;
3795
3796         t = kmemdup(&neigh_sysctl_template, sizeof(*t), GFP_KERNEL_ACCOUNT);
3797         if (!t)
3798                 goto err;
3799
3800         for (i = 0; i < NEIGH_VAR_GC_INTERVAL; i++) {
3801                 t->neigh_vars[i].data += (long) p;
3802                 t->neigh_vars[i].extra1 = dev;
3803                 t->neigh_vars[i].extra2 = p;
3804         }
3805
3806         neigh_vars_size = ARRAY_SIZE(t->neigh_vars);
3807         if (dev) {
3808                 dev_name_source = dev->name;
3809                 /* Terminate the table early */
3810                 memset(&t->neigh_vars[NEIGH_VAR_GC_INTERVAL], 0,
3811                        sizeof(t->neigh_vars[NEIGH_VAR_GC_INTERVAL]));
3812                 neigh_vars_size = NEIGH_VAR_BASE_REACHABLE_TIME_MS + 1;
3813         } else {
3814                 struct neigh_table *tbl = p->tbl;
3815                 dev_name_source = "default";
3816                 t->neigh_vars[NEIGH_VAR_GC_INTERVAL].data = &tbl->gc_interval;
3817                 t->neigh_vars[NEIGH_VAR_GC_THRESH1].data = &tbl->gc_thresh1;
3818                 t->neigh_vars[NEIGH_VAR_GC_THRESH2].data = &tbl->gc_thresh2;
3819                 t->neigh_vars[NEIGH_VAR_GC_THRESH3].data = &tbl->gc_thresh3;
3820         }
3821
3822         if (handler) {
3823                 /* RetransTime */
3824                 t->neigh_vars[NEIGH_VAR_RETRANS_TIME].proc_handler = handler;
3825                 /* ReachableTime */
3826                 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].proc_handler = handler;
3827                 /* RetransTime (in milliseconds)*/
3828                 t->neigh_vars[NEIGH_VAR_RETRANS_TIME_MS].proc_handler = handler;
3829                 /* ReachableTime (in milliseconds) */
3830                 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].proc_handler = handler;
3831         } else {
3832                 /* Those handlers will update p->reachable_time after
3833                  * base_reachable_time(_ms) is set to ensure the new timer starts being
3834                  * applied after the next neighbour update instead of waiting for
3835                  * neigh_periodic_work to update its value (can be multiple minutes)
3836                  * So any handler that replaces them should do this as well
3837                  */
3838                 /* ReachableTime */
3839                 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].proc_handler =
3840                         neigh_proc_base_reachable_time;
3841                 /* ReachableTime (in milliseconds) */
3842                 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].proc_handler =
3843                         neigh_proc_base_reachable_time;
3844         }
3845
3846         switch (neigh_parms_family(p)) {
3847         case AF_INET:
3848               p_name = "ipv4";
3849               break;
3850         case AF_INET6:
3851               p_name = "ipv6";
3852               break;
3853         default:
3854               BUG();
3855         }
3856
3857         snprintf(neigh_path, sizeof(neigh_path), "net/%s/neigh/%s",
3858                 p_name, dev_name_source);
3859         t->sysctl_header = register_net_sysctl_sz(neigh_parms_net(p),
3860                                                   neigh_path, t->neigh_vars,
3861                                                   neigh_vars_size);
3862         if (!t->sysctl_header)
3863                 goto free;
3864
3865         p->sysctl_table = t;
3866         return 0;
3867
3868 free:
3869         kfree(t);
3870 err:
3871         return -ENOBUFS;
3872 }
3873 EXPORT_SYMBOL(neigh_sysctl_register);
3874
3875 void neigh_sysctl_unregister(struct neigh_parms *p)
3876 {
3877         if (p->sysctl_table) {
3878                 struct neigh_sysctl_table *t = p->sysctl_table;
3879                 p->sysctl_table = NULL;
3880                 unregister_net_sysctl_table(t->sysctl_header);
3881                 kfree(t);
3882         }
3883 }
3884 EXPORT_SYMBOL(neigh_sysctl_unregister);
3885
3886 #endif  /* CONFIG_SYSCTL */
3887
3888 static int __init neigh_init(void)
3889 {
3890         rtnl_register(PF_UNSPEC, RTM_NEWNEIGH, neigh_add, NULL, 0);
3891         rtnl_register(PF_UNSPEC, RTM_DELNEIGH, neigh_delete, NULL, 0);
3892         rtnl_register(PF_UNSPEC, RTM_GETNEIGH, neigh_get, neigh_dump_info, 0);
3893
3894         rtnl_register(PF_UNSPEC, RTM_GETNEIGHTBL, NULL, neightbl_dump_info,
3895                       0);
3896         rtnl_register(PF_UNSPEC, RTM_SETNEIGHTBL, neightbl_set, NULL, 0);
3897
3898         return 0;
3899 }
3900
3901 subsys_initcall(neigh_init);