1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Generic address resolution entity
6 * Pedro Roque <roque@di.fc.ul.pt>
7 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
10 * Vitaly E. Lavrov releasing NULL neighbor in neigh_add.
11 * Harald Welte Add neighbour cache statistics like rtstat
14 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
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>
25 #include <linux/sysctl.h>
27 #include <linux/times.h>
28 #include <net/net_namespace.h>
29 #include <net/neighbour.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>
42 #include <trace/events/neigh.h>
45 #define neigh_dbg(level, fmt, ...) \
47 if (level <= NEIGH_DEBUG) \
48 pr_debug(fmt, ##__VA_ARGS__); \
51 #define PNEIGH_HASHMASK 0xF
53 static void neigh_timer_handler(struct timer_list *t);
54 static void __neigh_notify(struct neighbour *n, int type, int flags,
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);
61 static const struct seq_operations neigh_stat_seq_ops;
65 Neighbour hash table buckets are protected with rwlock tbl->lock.
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
72 - If the entry requires some non-trivial actions, increase
73 its reference count and release table lock.
75 Neighbour entries are protected:
76 - with reference count.
77 - with rwlock neigh->lock
79 Reference count prevents destruction.
81 neigh->lock mainly serializes ll address data and its validity state.
82 However, the same lock is used to protect another entry fields:
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.
92 static int neigh_blackhole(struct neighbour *neigh, struct sk_buff *skb)
98 static void neigh_cleanup_and_release(struct neighbour *neigh)
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);
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.
112 unsigned long neigh_rand_reach_time(unsigned long base)
114 return base ? (prandom_u32() % base) + (base >> 1) : 0;
116 EXPORT_SYMBOL(neigh_rand_reach_time);
118 static void neigh_mark_dead(struct neighbour *n)
121 if (!list_empty(&n->gc_list)) {
122 list_del_init(&n->gc_list);
123 atomic_dec(&n->tbl->gc_entries);
125 if (!list_empty(&n->managed_list))
126 list_del_init(&n->managed_list);
129 static void neigh_update_gc_list(struct neighbour *n)
131 bool on_gc_list, exempt_from_gc;
133 write_lock_bh(&n->tbl->lock);
134 write_lock(&n->lock);
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
141 exempt_from_gc = n->nud_state & NUD_PERMANENT ||
142 n->flags & NTF_EXT_LEARNED;
143 on_gc_list = !list_empty(&n->gc_list);
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);
154 write_unlock(&n->lock);
155 write_unlock_bh(&n->tbl->lock);
158 static void neigh_update_managed_list(struct neighbour *n)
160 bool on_managed_list, add_to_managed;
162 write_lock_bh(&n->tbl->lock);
163 write_lock(&n->lock);
167 add_to_managed = n->flags & NTF_MANAGED;
168 on_managed_list = !list_empty(&n->managed_list);
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);
175 write_unlock(&n->lock);
176 write_unlock_bh(&n->tbl->lock);
179 static void neigh_update_flags(struct neighbour *neigh, u32 flags, int *notify,
180 bool *gc_update, bool *managed_update)
182 u32 ndm_flags, old_flags = neigh->flags;
184 if (!(flags & NEIGH_UPDATE_F_ADMIN))
187 ndm_flags = (flags & NEIGH_UPDATE_F_EXT_LEARNED) ? NTF_EXT_LEARNED : 0;
188 ndm_flags |= (flags & NEIGH_UPDATE_F_MANAGED) ? NTF_MANAGED : 0;
190 if ((old_flags ^ ndm_flags) & NTF_EXT_LEARNED) {
191 if (ndm_flags & NTF_EXT_LEARNED)
192 neigh->flags |= NTF_EXT_LEARNED;
194 neigh->flags &= ~NTF_EXT_LEARNED;
198 if ((old_flags ^ ndm_flags) & NTF_MANAGED) {
199 if (ndm_flags & NTF_MANAGED)
200 neigh->flags |= NTF_MANAGED;
202 neigh->flags &= ~NTF_MANAGED;
204 *managed_update = true;
208 static bool neigh_del(struct neighbour *n, struct neighbour __rcu **np,
209 struct neigh_table *tbl)
213 write_lock(&n->lock);
214 if (refcount_read(&n->refcnt) == 1) {
215 struct neighbour *neigh;
217 neigh = rcu_dereference_protected(n->next,
218 lockdep_is_held(&tbl->lock));
219 rcu_assign_pointer(*np, neigh);
223 write_unlock(&n->lock);
225 neigh_cleanup_and_release(n);
229 bool neigh_remove_one(struct neighbour *ndel, struct neigh_table *tbl)
231 struct neigh_hash_table *nht;
232 void *pkey = ndel->primary_key;
235 struct neighbour __rcu **np;
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);
242 np = &nht->hash_buckets[hash_val];
243 while ((n = rcu_dereference_protected(*np,
244 lockdep_is_held(&tbl->lock)))) {
246 return neigh_del(n, np, tbl);
252 static int neigh_forced_gc(struct neigh_table *tbl)
254 int max_clean = atomic_read(&tbl->gc_entries) - tbl->gc_thresh2;
255 unsigned long tref = jiffies - 5 * HZ;
256 struct neighbour *n, *tmp;
259 NEIGH_CACHE_STAT_INC(tbl, forced_gc_runs);
261 write_lock_bh(&tbl->lock);
263 list_for_each_entry_safe(n, tmp, &tbl->gc_list, gc_list) {
264 if (refcount_read(&n->refcnt) == 1) {
267 write_lock(&n->lock);
268 if ((n->nud_state == NUD_FAILED) ||
269 (n->nud_state == NUD_NOARP) ||
270 (tbl->is_multicast &&
271 tbl->is_multicast(n->primary_key)) ||
272 time_after(tref, n->updated))
274 write_unlock(&n->lock);
276 if (remove && neigh_remove_one(n, tbl))
278 if (shrunk >= max_clean)
283 tbl->last_flush = jiffies;
285 write_unlock_bh(&tbl->lock);
290 static void neigh_add_timer(struct neighbour *n, unsigned long when)
293 if (unlikely(mod_timer(&n->timer, when))) {
294 printk("NEIGH: BUG, double timer add, state is %x\n",
300 static int neigh_del_timer(struct neighbour *n)
302 if ((n->nud_state & NUD_IN_TIMER) &&
303 del_timer(&n->timer)) {
310 static void pneigh_queue_purge(struct sk_buff_head *list)
314 while ((skb = skb_dequeue(list)) != NULL) {
320 static void neigh_flush_dev(struct neigh_table *tbl, struct net_device *dev,
324 struct neigh_hash_table *nht;
326 nht = rcu_dereference_protected(tbl->nht,
327 lockdep_is_held(&tbl->lock));
329 for (i = 0; i < (1 << nht->hash_shift); i++) {
331 struct neighbour __rcu **np = &nht->hash_buckets[i];
333 while ((n = rcu_dereference_protected(*np,
334 lockdep_is_held(&tbl->lock))) != NULL) {
335 if (dev && n->dev != dev) {
339 if (skip_perm && n->nud_state & NUD_PERMANENT) {
343 rcu_assign_pointer(*np,
344 rcu_dereference_protected(n->next,
345 lockdep_is_held(&tbl->lock)));
346 write_lock(&n->lock);
349 if (refcount_read(&n->refcnt) != 1) {
350 /* The most unpleasant situation.
351 We must destroy neighbour entry,
352 but someone still uses it.
354 The destroy will be delayed until
355 the last user releases us, but
356 we must kill timers etc. and move
359 __skb_queue_purge(&n->arp_queue);
360 n->arp_queue_len_bytes = 0;
361 n->output = neigh_blackhole;
362 if (n->nud_state & NUD_VALID)
363 n->nud_state = NUD_NOARP;
365 n->nud_state = NUD_NONE;
366 neigh_dbg(2, "neigh %p is stray\n", n);
368 write_unlock(&n->lock);
369 neigh_cleanup_and_release(n);
374 void neigh_changeaddr(struct neigh_table *tbl, struct net_device *dev)
376 write_lock_bh(&tbl->lock);
377 neigh_flush_dev(tbl, dev, false);
378 write_unlock_bh(&tbl->lock);
380 EXPORT_SYMBOL(neigh_changeaddr);
382 static int __neigh_ifdown(struct neigh_table *tbl, struct net_device *dev,
385 write_lock_bh(&tbl->lock);
386 neigh_flush_dev(tbl, dev, skip_perm);
387 pneigh_ifdown_and_unlock(tbl, dev);
389 del_timer_sync(&tbl->proxy_timer);
390 pneigh_queue_purge(&tbl->proxy_queue);
394 int neigh_carrier_down(struct neigh_table *tbl, struct net_device *dev)
396 __neigh_ifdown(tbl, dev, true);
399 EXPORT_SYMBOL(neigh_carrier_down);
401 int neigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
403 __neigh_ifdown(tbl, dev, false);
406 EXPORT_SYMBOL(neigh_ifdown);
408 static struct neighbour *neigh_alloc(struct neigh_table *tbl,
409 struct net_device *dev,
410 u32 flags, bool exempt_from_gc)
412 struct neighbour *n = NULL;
413 unsigned long now = jiffies;
419 entries = atomic_inc_return(&tbl->gc_entries) - 1;
420 if (entries >= tbl->gc_thresh3 ||
421 (entries >= tbl->gc_thresh2 &&
422 time_after(now, tbl->last_flush + 5 * HZ))) {
423 if (!neigh_forced_gc(tbl) &&
424 entries >= tbl->gc_thresh3) {
425 net_info_ratelimited("%s: neighbor table overflow!\n",
427 NEIGH_CACHE_STAT_INC(tbl, table_fulls);
433 n = kzalloc(tbl->entry_size + dev->neigh_priv_len, GFP_ATOMIC);
437 __skb_queue_head_init(&n->arp_queue);
438 rwlock_init(&n->lock);
439 seqlock_init(&n->ha_lock);
440 n->updated = n->used = now;
441 n->nud_state = NUD_NONE;
442 n->output = neigh_blackhole;
444 seqlock_init(&n->hh.hh_lock);
445 n->parms = neigh_parms_clone(&tbl->parms);
446 timer_setup(&n->timer, neigh_timer_handler, 0);
448 NEIGH_CACHE_STAT_INC(tbl, allocs);
450 refcount_set(&n->refcnt, 1);
452 INIT_LIST_HEAD(&n->gc_list);
453 INIT_LIST_HEAD(&n->managed_list);
455 atomic_inc(&tbl->entries);
461 atomic_dec(&tbl->gc_entries);
465 static void neigh_get_hash_rnd(u32 *x)
467 *x = get_random_u32() | 1;
470 static struct neigh_hash_table *neigh_hash_alloc(unsigned int shift)
472 size_t size = (1 << shift) * sizeof(struct neighbour *);
473 struct neigh_hash_table *ret;
474 struct neighbour __rcu **buckets;
477 ret = kmalloc(sizeof(*ret), GFP_ATOMIC);
480 if (size <= PAGE_SIZE) {
481 buckets = kzalloc(size, GFP_ATOMIC);
483 buckets = (struct neighbour __rcu **)
484 __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
486 kmemleak_alloc(buckets, size, 1, GFP_ATOMIC);
492 ret->hash_buckets = buckets;
493 ret->hash_shift = shift;
494 for (i = 0; i < NEIGH_NUM_HASH_RND; i++)
495 neigh_get_hash_rnd(&ret->hash_rnd[i]);
499 static void neigh_hash_free_rcu(struct rcu_head *head)
501 struct neigh_hash_table *nht = container_of(head,
502 struct neigh_hash_table,
504 size_t size = (1 << nht->hash_shift) * sizeof(struct neighbour *);
505 struct neighbour __rcu **buckets = nht->hash_buckets;
507 if (size <= PAGE_SIZE) {
510 kmemleak_free(buckets);
511 free_pages((unsigned long)buckets, get_order(size));
516 static struct neigh_hash_table *neigh_hash_grow(struct neigh_table *tbl,
517 unsigned long new_shift)
519 unsigned int i, hash;
520 struct neigh_hash_table *new_nht, *old_nht;
522 NEIGH_CACHE_STAT_INC(tbl, hash_grows);
524 old_nht = rcu_dereference_protected(tbl->nht,
525 lockdep_is_held(&tbl->lock));
526 new_nht = neigh_hash_alloc(new_shift);
530 for (i = 0; i < (1 << old_nht->hash_shift); i++) {
531 struct neighbour *n, *next;
533 for (n = rcu_dereference_protected(old_nht->hash_buckets[i],
534 lockdep_is_held(&tbl->lock));
537 hash = tbl->hash(n->primary_key, n->dev,
540 hash >>= (32 - new_nht->hash_shift);
541 next = rcu_dereference_protected(n->next,
542 lockdep_is_held(&tbl->lock));
544 rcu_assign_pointer(n->next,
545 rcu_dereference_protected(
546 new_nht->hash_buckets[hash],
547 lockdep_is_held(&tbl->lock)));
548 rcu_assign_pointer(new_nht->hash_buckets[hash], n);
552 rcu_assign_pointer(tbl->nht, new_nht);
553 call_rcu(&old_nht->rcu, neigh_hash_free_rcu);
557 struct neighbour *neigh_lookup(struct neigh_table *tbl, const void *pkey,
558 struct net_device *dev)
562 NEIGH_CACHE_STAT_INC(tbl, lookups);
565 n = __neigh_lookup_noref(tbl, pkey, dev);
567 if (!refcount_inc_not_zero(&n->refcnt))
569 NEIGH_CACHE_STAT_INC(tbl, hits);
572 rcu_read_unlock_bh();
575 EXPORT_SYMBOL(neigh_lookup);
577 struct neighbour *neigh_lookup_nodev(struct neigh_table *tbl, struct net *net,
581 unsigned int key_len = tbl->key_len;
583 struct neigh_hash_table *nht;
585 NEIGH_CACHE_STAT_INC(tbl, lookups);
588 nht = rcu_dereference_bh(tbl->nht);
589 hash_val = tbl->hash(pkey, NULL, nht->hash_rnd) >> (32 - nht->hash_shift);
591 for (n = rcu_dereference_bh(nht->hash_buckets[hash_val]);
593 n = rcu_dereference_bh(n->next)) {
594 if (!memcmp(n->primary_key, pkey, key_len) &&
595 net_eq(dev_net(n->dev), net)) {
596 if (!refcount_inc_not_zero(&n->refcnt))
598 NEIGH_CACHE_STAT_INC(tbl, hits);
603 rcu_read_unlock_bh();
606 EXPORT_SYMBOL(neigh_lookup_nodev);
608 static struct neighbour *
609 ___neigh_create(struct neigh_table *tbl, const void *pkey,
610 struct net_device *dev, u32 flags,
611 bool exempt_from_gc, bool want_ref)
613 u32 hash_val, key_len = tbl->key_len;
614 struct neighbour *n1, *rc, *n;
615 struct neigh_hash_table *nht;
618 n = neigh_alloc(tbl, dev, flags, exempt_from_gc);
619 trace_neigh_create(tbl, dev, pkey, n, exempt_from_gc);
621 rc = ERR_PTR(-ENOBUFS);
625 memcpy(n->primary_key, pkey, key_len);
627 dev_hold_track(dev, &n->dev_tracker, GFP_ATOMIC);
629 /* Protocol specific setup. */
630 if (tbl->constructor && (error = tbl->constructor(n)) < 0) {
632 goto out_neigh_release;
635 if (dev->netdev_ops->ndo_neigh_construct) {
636 error = dev->netdev_ops->ndo_neigh_construct(dev, n);
639 goto out_neigh_release;
643 /* Device specific setup. */
644 if (n->parms->neigh_setup &&
645 (error = n->parms->neigh_setup(n)) < 0) {
647 goto out_neigh_release;
650 n->confirmed = jiffies - (NEIGH_VAR(n->parms, BASE_REACHABLE_TIME) << 1);
652 write_lock_bh(&tbl->lock);
653 nht = rcu_dereference_protected(tbl->nht,
654 lockdep_is_held(&tbl->lock));
656 if (atomic_read(&tbl->entries) > (1 << nht->hash_shift))
657 nht = neigh_hash_grow(tbl, nht->hash_shift + 1);
659 hash_val = tbl->hash(n->primary_key, dev, nht->hash_rnd) >> (32 - nht->hash_shift);
661 if (n->parms->dead) {
662 rc = ERR_PTR(-EINVAL);
666 for (n1 = rcu_dereference_protected(nht->hash_buckets[hash_val],
667 lockdep_is_held(&tbl->lock));
669 n1 = rcu_dereference_protected(n1->next,
670 lockdep_is_held(&tbl->lock))) {
671 if (dev == n1->dev && !memcmp(n1->primary_key, n->primary_key, key_len)) {
681 list_add_tail(&n->gc_list, &n->tbl->gc_list);
682 if (n->flags & NTF_MANAGED)
683 list_add_tail(&n->managed_list, &n->tbl->managed_list);
686 rcu_assign_pointer(n->next,
687 rcu_dereference_protected(nht->hash_buckets[hash_val],
688 lockdep_is_held(&tbl->lock)));
689 rcu_assign_pointer(nht->hash_buckets[hash_val], n);
690 write_unlock_bh(&tbl->lock);
691 neigh_dbg(2, "neigh %p is created\n", n);
696 write_unlock_bh(&tbl->lock);
699 atomic_dec(&tbl->gc_entries);
704 struct neighbour *__neigh_create(struct neigh_table *tbl, const void *pkey,
705 struct net_device *dev, bool want_ref)
707 return ___neigh_create(tbl, pkey, dev, 0, false, want_ref);
709 EXPORT_SYMBOL(__neigh_create);
711 static u32 pneigh_hash(const void *pkey, unsigned int key_len)
713 u32 hash_val = *(u32 *)(pkey + key_len - 4);
714 hash_val ^= (hash_val >> 16);
715 hash_val ^= hash_val >> 8;
716 hash_val ^= hash_val >> 4;
717 hash_val &= PNEIGH_HASHMASK;
721 static struct pneigh_entry *__pneigh_lookup_1(struct pneigh_entry *n,
724 unsigned int key_len,
725 struct net_device *dev)
728 if (!memcmp(n->key, pkey, key_len) &&
729 net_eq(pneigh_net(n), net) &&
730 (n->dev == dev || !n->dev))
737 struct pneigh_entry *__pneigh_lookup(struct neigh_table *tbl,
738 struct net *net, const void *pkey, struct net_device *dev)
740 unsigned int key_len = tbl->key_len;
741 u32 hash_val = pneigh_hash(pkey, key_len);
743 return __pneigh_lookup_1(tbl->phash_buckets[hash_val],
744 net, pkey, key_len, dev);
746 EXPORT_SYMBOL_GPL(__pneigh_lookup);
748 struct pneigh_entry * pneigh_lookup(struct neigh_table *tbl,
749 struct net *net, const void *pkey,
750 struct net_device *dev, int creat)
752 struct pneigh_entry *n;
753 unsigned int key_len = tbl->key_len;
754 u32 hash_val = pneigh_hash(pkey, key_len);
756 read_lock_bh(&tbl->lock);
757 n = __pneigh_lookup_1(tbl->phash_buckets[hash_val],
758 net, pkey, key_len, dev);
759 read_unlock_bh(&tbl->lock);
766 n = kzalloc(sizeof(*n) + key_len, GFP_KERNEL);
770 write_pnet(&n->net, net);
771 memcpy(n->key, pkey, key_len);
773 dev_hold_track(dev, &n->dev_tracker, GFP_KERNEL);
775 if (tbl->pconstructor && tbl->pconstructor(n)) {
776 dev_put_track(dev, &n->dev_tracker);
782 write_lock_bh(&tbl->lock);
783 n->next = tbl->phash_buckets[hash_val];
784 tbl->phash_buckets[hash_val] = n;
785 write_unlock_bh(&tbl->lock);
789 EXPORT_SYMBOL(pneigh_lookup);
792 int pneigh_delete(struct neigh_table *tbl, struct net *net, const void *pkey,
793 struct net_device *dev)
795 struct pneigh_entry *n, **np;
796 unsigned int key_len = tbl->key_len;
797 u32 hash_val = pneigh_hash(pkey, key_len);
799 write_lock_bh(&tbl->lock);
800 for (np = &tbl->phash_buckets[hash_val]; (n = *np) != NULL;
802 if (!memcmp(n->key, pkey, key_len) && n->dev == dev &&
803 net_eq(pneigh_net(n), net)) {
805 write_unlock_bh(&tbl->lock);
806 if (tbl->pdestructor)
808 dev_put_track(n->dev, &n->dev_tracker);
813 write_unlock_bh(&tbl->lock);
817 static int pneigh_ifdown_and_unlock(struct neigh_table *tbl,
818 struct net_device *dev)
820 struct pneigh_entry *n, **np, *freelist = NULL;
823 for (h = 0; h <= PNEIGH_HASHMASK; h++) {
824 np = &tbl->phash_buckets[h];
825 while ((n = *np) != NULL) {
826 if (!dev || n->dev == dev) {
835 write_unlock_bh(&tbl->lock);
836 while ((n = freelist)) {
839 if (tbl->pdestructor)
841 dev_put_track(n->dev, &n->dev_tracker);
847 static void neigh_parms_destroy(struct neigh_parms *parms);
849 static inline void neigh_parms_put(struct neigh_parms *parms)
851 if (refcount_dec_and_test(&parms->refcnt))
852 neigh_parms_destroy(parms);
856 * neighbour must already be out of the table;
859 void neigh_destroy(struct neighbour *neigh)
861 struct net_device *dev = neigh->dev;
863 NEIGH_CACHE_STAT_INC(neigh->tbl, destroys);
866 pr_warn("Destroying alive neighbour %p\n", neigh);
871 if (neigh_del_timer(neigh))
872 pr_warn("Impossible event\n");
874 write_lock_bh(&neigh->lock);
875 __skb_queue_purge(&neigh->arp_queue);
876 write_unlock_bh(&neigh->lock);
877 neigh->arp_queue_len_bytes = 0;
879 if (dev->netdev_ops->ndo_neigh_destroy)
880 dev->netdev_ops->ndo_neigh_destroy(dev, neigh);
882 dev_put_track(dev, &neigh->dev_tracker);
883 neigh_parms_put(neigh->parms);
885 neigh_dbg(2, "neigh %p is destroyed\n", neigh);
887 atomic_dec(&neigh->tbl->entries);
888 kfree_rcu(neigh, rcu);
890 EXPORT_SYMBOL(neigh_destroy);
892 /* Neighbour state is suspicious;
895 Called with write_locked neigh.
897 static void neigh_suspect(struct neighbour *neigh)
899 neigh_dbg(2, "neigh %p is suspected\n", neigh);
901 neigh->output = neigh->ops->output;
904 /* Neighbour state is OK;
907 Called with write_locked neigh.
909 static void neigh_connect(struct neighbour *neigh)
911 neigh_dbg(2, "neigh %p is connected\n", neigh);
913 neigh->output = neigh->ops->connected_output;
916 static void neigh_periodic_work(struct work_struct *work)
918 struct neigh_table *tbl = container_of(work, struct neigh_table, gc_work.work);
920 struct neighbour __rcu **np;
922 struct neigh_hash_table *nht;
924 NEIGH_CACHE_STAT_INC(tbl, periodic_gc_runs);
926 write_lock_bh(&tbl->lock);
927 nht = rcu_dereference_protected(tbl->nht,
928 lockdep_is_held(&tbl->lock));
931 * periodically recompute ReachableTime from random function
934 if (time_after(jiffies, tbl->last_rand + 300 * HZ)) {
935 struct neigh_parms *p;
936 tbl->last_rand = jiffies;
937 list_for_each_entry(p, &tbl->parms_list, list)
939 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
942 if (atomic_read(&tbl->entries) < tbl->gc_thresh1)
945 for (i = 0 ; i < (1 << nht->hash_shift); i++) {
946 np = &nht->hash_buckets[i];
948 while ((n = rcu_dereference_protected(*np,
949 lockdep_is_held(&tbl->lock))) != NULL) {
952 write_lock(&n->lock);
954 state = n->nud_state;
955 if ((state & (NUD_PERMANENT | NUD_IN_TIMER)) ||
956 (n->flags & NTF_EXT_LEARNED)) {
957 write_unlock(&n->lock);
961 if (time_before(n->used, n->confirmed))
962 n->used = n->confirmed;
964 if (refcount_read(&n->refcnt) == 1 &&
965 (state == NUD_FAILED ||
966 time_after(jiffies, n->used + NEIGH_VAR(n->parms, GC_STALETIME)))) {
969 write_unlock(&n->lock);
970 neigh_cleanup_and_release(n);
973 write_unlock(&n->lock);
979 * It's fine to release lock here, even if hash table
980 * grows while we are preempted.
982 write_unlock_bh(&tbl->lock);
984 write_lock_bh(&tbl->lock);
985 nht = rcu_dereference_protected(tbl->nht,
986 lockdep_is_held(&tbl->lock));
989 /* Cycle through all hash buckets every BASE_REACHABLE_TIME/2 ticks.
990 * ARP entry timeouts range from 1/2 BASE_REACHABLE_TIME to 3/2
991 * BASE_REACHABLE_TIME.
993 queue_delayed_work(system_power_efficient_wq, &tbl->gc_work,
994 NEIGH_VAR(&tbl->parms, BASE_REACHABLE_TIME) >> 1);
995 write_unlock_bh(&tbl->lock);
998 static __inline__ int neigh_max_probes(struct neighbour *n)
1000 struct neigh_parms *p = n->parms;
1001 return NEIGH_VAR(p, UCAST_PROBES) + NEIGH_VAR(p, APP_PROBES) +
1002 (n->nud_state & NUD_PROBE ? NEIGH_VAR(p, MCAST_REPROBES) :
1003 NEIGH_VAR(p, MCAST_PROBES));
1006 static void neigh_invalidate(struct neighbour *neigh)
1007 __releases(neigh->lock)
1008 __acquires(neigh->lock)
1010 struct sk_buff *skb;
1012 NEIGH_CACHE_STAT_INC(neigh->tbl, res_failed);
1013 neigh_dbg(2, "neigh %p is failed\n", neigh);
1014 neigh->updated = jiffies;
1016 /* It is very thin place. report_unreachable is very complicated
1017 routine. Particularly, it can hit the same neighbour entry!
1019 So that, we try to be accurate and avoid dead loop. --ANK
1021 while (neigh->nud_state == NUD_FAILED &&
1022 (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
1023 write_unlock(&neigh->lock);
1024 neigh->ops->error_report(neigh, skb);
1025 write_lock(&neigh->lock);
1027 __skb_queue_purge(&neigh->arp_queue);
1028 neigh->arp_queue_len_bytes = 0;
1031 static void neigh_probe(struct neighbour *neigh)
1032 __releases(neigh->lock)
1034 struct sk_buff *skb = skb_peek_tail(&neigh->arp_queue);
1035 /* keep skb alive even if arp_queue overflows */
1037 skb = skb_clone(skb, GFP_ATOMIC);
1038 write_unlock(&neigh->lock);
1039 if (neigh->ops->solicit)
1040 neigh->ops->solicit(neigh, skb);
1041 atomic_inc(&neigh->probes);
1045 /* Called when a timer expires for a neighbour entry. */
1047 static void neigh_timer_handler(struct timer_list *t)
1049 unsigned long now, next;
1050 struct neighbour *neigh = from_timer(neigh, t, timer);
1054 write_lock(&neigh->lock);
1056 state = neigh->nud_state;
1060 if (!(state & NUD_IN_TIMER))
1063 if (state & NUD_REACHABLE) {
1064 if (time_before_eq(now,
1065 neigh->confirmed + neigh->parms->reachable_time)) {
1066 neigh_dbg(2, "neigh %p is still alive\n", neigh);
1067 next = neigh->confirmed + neigh->parms->reachable_time;
1068 } else if (time_before_eq(now,
1070 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME))) {
1071 neigh_dbg(2, "neigh %p is delayed\n", neigh);
1072 neigh->nud_state = NUD_DELAY;
1073 neigh->updated = jiffies;
1074 neigh_suspect(neigh);
1075 next = now + NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME);
1077 neigh_dbg(2, "neigh %p is suspected\n", neigh);
1078 neigh->nud_state = NUD_STALE;
1079 neigh->updated = jiffies;
1080 neigh_suspect(neigh);
1083 } else if (state & NUD_DELAY) {
1084 if (time_before_eq(now,
1086 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME))) {
1087 neigh_dbg(2, "neigh %p is now reachable\n", neigh);
1088 neigh->nud_state = NUD_REACHABLE;
1089 neigh->updated = jiffies;
1090 neigh_connect(neigh);
1092 next = neigh->confirmed + neigh->parms->reachable_time;
1094 neigh_dbg(2, "neigh %p is probed\n", neigh);
1095 neigh->nud_state = NUD_PROBE;
1096 neigh->updated = jiffies;
1097 atomic_set(&neigh->probes, 0);
1099 next = now + max(NEIGH_VAR(neigh->parms, RETRANS_TIME),
1103 /* NUD_PROBE|NUD_INCOMPLETE */
1104 next = now + max(NEIGH_VAR(neigh->parms, RETRANS_TIME), HZ/100);
1107 if ((neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) &&
1108 atomic_read(&neigh->probes) >= neigh_max_probes(neigh)) {
1109 neigh->nud_state = NUD_FAILED;
1111 neigh_invalidate(neigh);
1115 if (neigh->nud_state & NUD_IN_TIMER) {
1116 if (time_before(next, jiffies + HZ/100))
1117 next = jiffies + HZ/100;
1118 if (!mod_timer(&neigh->timer, next))
1121 if (neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) {
1125 write_unlock(&neigh->lock);
1129 neigh_update_notify(neigh, 0);
1131 trace_neigh_timer_handler(neigh, 0);
1133 neigh_release(neigh);
1136 int __neigh_event_send(struct neighbour *neigh, struct sk_buff *skb)
1139 bool immediate_probe = false;
1141 write_lock_bh(&neigh->lock);
1144 if (neigh->nud_state & (NUD_CONNECTED | NUD_DELAY | NUD_PROBE))
1149 if (!(neigh->nud_state & (NUD_STALE | NUD_INCOMPLETE))) {
1150 if (NEIGH_VAR(neigh->parms, MCAST_PROBES) +
1151 NEIGH_VAR(neigh->parms, APP_PROBES)) {
1152 unsigned long next, now = jiffies;
1154 atomic_set(&neigh->probes,
1155 NEIGH_VAR(neigh->parms, UCAST_PROBES));
1156 neigh_del_timer(neigh);
1157 neigh->nud_state = NUD_INCOMPLETE;
1158 neigh->updated = now;
1159 next = now + max(NEIGH_VAR(neigh->parms, RETRANS_TIME),
1161 neigh_add_timer(neigh, next);
1162 immediate_probe = true;
1164 neigh->nud_state = NUD_FAILED;
1165 neigh->updated = jiffies;
1166 write_unlock_bh(&neigh->lock);
1171 } else if (neigh->nud_state & NUD_STALE) {
1172 neigh_dbg(2, "neigh %p is delayed\n", neigh);
1173 neigh_del_timer(neigh);
1174 neigh->nud_state = NUD_DELAY;
1175 neigh->updated = jiffies;
1176 neigh_add_timer(neigh, jiffies +
1177 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME));
1180 if (neigh->nud_state == NUD_INCOMPLETE) {
1182 while (neigh->arp_queue_len_bytes + skb->truesize >
1183 NEIGH_VAR(neigh->parms, QUEUE_LEN_BYTES)) {
1184 struct sk_buff *buff;
1186 buff = __skb_dequeue(&neigh->arp_queue);
1189 neigh->arp_queue_len_bytes -= buff->truesize;
1191 NEIGH_CACHE_STAT_INC(neigh->tbl, unres_discards);
1194 __skb_queue_tail(&neigh->arp_queue, skb);
1195 neigh->arp_queue_len_bytes += skb->truesize;
1200 if (immediate_probe)
1203 write_unlock(&neigh->lock);
1205 trace_neigh_event_send_done(neigh, rc);
1209 if (neigh->nud_state & NUD_STALE)
1211 write_unlock_bh(&neigh->lock);
1213 trace_neigh_event_send_dead(neigh, 1);
1216 EXPORT_SYMBOL(__neigh_event_send);
1218 static void neigh_update_hhs(struct neighbour *neigh)
1220 struct hh_cache *hh;
1221 void (*update)(struct hh_cache*, const struct net_device*, const unsigned char *)
1224 if (neigh->dev->header_ops)
1225 update = neigh->dev->header_ops->cache_update;
1229 if (READ_ONCE(hh->hh_len)) {
1230 write_seqlock_bh(&hh->hh_lock);
1231 update(hh, neigh->dev, neigh->ha);
1232 write_sequnlock_bh(&hh->hh_lock);
1237 /* Generic update routine.
1238 -- lladdr is new lladdr or NULL, if it is not supplied.
1239 -- new is new state.
1241 NEIGH_UPDATE_F_OVERRIDE allows to override existing lladdr,
1243 NEIGH_UPDATE_F_WEAK_OVERRIDE will suspect existing "connected"
1244 lladdr instead of overriding it
1246 NEIGH_UPDATE_F_ADMIN means that the change is administrative.
1247 NEIGH_UPDATE_F_USE means that the entry is user triggered.
1248 NEIGH_UPDATE_F_MANAGED means that the entry will be auto-refreshed.
1249 NEIGH_UPDATE_F_OVERRIDE_ISROUTER allows to override existing
1251 NEIGH_UPDATE_F_ISROUTER indicates if the neighbour is known as
1254 Caller MUST hold reference count on the entry.
1256 static int __neigh_update(struct neighbour *neigh, const u8 *lladdr,
1257 u8 new, u32 flags, u32 nlmsg_pid,
1258 struct netlink_ext_ack *extack)
1260 bool gc_update = false, managed_update = false;
1261 int update_isrouter = 0;
1262 struct net_device *dev;
1263 int err, notify = 0;
1266 trace_neigh_update(neigh, lladdr, new, flags, nlmsg_pid);
1268 write_lock_bh(&neigh->lock);
1271 old = neigh->nud_state;
1275 NL_SET_ERR_MSG(extack, "Neighbor entry is now dead");
1279 if (!(flags & NEIGH_UPDATE_F_ADMIN) &&
1280 (old & (NUD_NOARP | NUD_PERMANENT)))
1283 neigh_update_flags(neigh, flags, ¬ify, &gc_update, &managed_update);
1284 if (flags & (NEIGH_UPDATE_F_USE | NEIGH_UPDATE_F_MANAGED)) {
1285 new = old & ~NUD_PERMANENT;
1286 neigh->nud_state = new;
1291 if (!(new & NUD_VALID)) {
1292 neigh_del_timer(neigh);
1293 if (old & NUD_CONNECTED)
1294 neigh_suspect(neigh);
1295 neigh->nud_state = new;
1297 notify = old & NUD_VALID;
1298 if ((old & (NUD_INCOMPLETE | NUD_PROBE)) &&
1299 (new & NUD_FAILED)) {
1300 neigh_invalidate(neigh);
1306 /* Compare new lladdr with cached one */
1307 if (!dev->addr_len) {
1308 /* First case: device needs no address. */
1310 } else if (lladdr) {
1311 /* The second case: if something is already cached
1312 and a new address is proposed:
1314 - if they are different, check override flag
1316 if ((old & NUD_VALID) &&
1317 !memcmp(lladdr, neigh->ha, dev->addr_len))
1320 /* No address is supplied; if we know something,
1321 use it, otherwise discard the request.
1324 if (!(old & NUD_VALID)) {
1325 NL_SET_ERR_MSG(extack, "No link layer address given");
1331 /* Update confirmed timestamp for neighbour entry after we
1332 * received ARP packet even if it doesn't change IP to MAC binding.
1334 if (new & NUD_CONNECTED)
1335 neigh->confirmed = jiffies;
1337 /* If entry was valid and address is not changed,
1338 do not change entry state, if new one is STALE.
1341 update_isrouter = flags & NEIGH_UPDATE_F_OVERRIDE_ISROUTER;
1342 if (old & NUD_VALID) {
1343 if (lladdr != neigh->ha && !(flags & NEIGH_UPDATE_F_OVERRIDE)) {
1344 update_isrouter = 0;
1345 if ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) &&
1346 (old & NUD_CONNECTED)) {
1352 if (lladdr == neigh->ha && new == NUD_STALE &&
1353 !(flags & NEIGH_UPDATE_F_ADMIN))
1358 /* Update timestamp only once we know we will make a change to the
1359 * neighbour entry. Otherwise we risk to move the locktime window with
1360 * noop updates and ignore relevant ARP updates.
1362 if (new != old || lladdr != neigh->ha)
1363 neigh->updated = jiffies;
1366 neigh_del_timer(neigh);
1367 if (new & NUD_PROBE)
1368 atomic_set(&neigh->probes, 0);
1369 if (new & NUD_IN_TIMER)
1370 neigh_add_timer(neigh, (jiffies +
1371 ((new & NUD_REACHABLE) ?
1372 neigh->parms->reachable_time :
1374 neigh->nud_state = new;
1378 if (lladdr != neigh->ha) {
1379 write_seqlock(&neigh->ha_lock);
1380 memcpy(&neigh->ha, lladdr, dev->addr_len);
1381 write_sequnlock(&neigh->ha_lock);
1382 neigh_update_hhs(neigh);
1383 if (!(new & NUD_CONNECTED))
1384 neigh->confirmed = jiffies -
1385 (NEIGH_VAR(neigh->parms, BASE_REACHABLE_TIME) << 1);
1390 if (new & NUD_CONNECTED)
1391 neigh_connect(neigh);
1393 neigh_suspect(neigh);
1394 if (!(old & NUD_VALID)) {
1395 struct sk_buff *skb;
1397 /* Again: avoid dead loop if something went wrong */
1399 while (neigh->nud_state & NUD_VALID &&
1400 (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
1401 struct dst_entry *dst = skb_dst(skb);
1402 struct neighbour *n2, *n1 = neigh;
1403 write_unlock_bh(&neigh->lock);
1407 /* Why not just use 'neigh' as-is? The problem is that
1408 * things such as shaper, eql, and sch_teql can end up
1409 * using alternative, different, neigh objects to output
1410 * the packet in the output path. So what we need to do
1411 * here is re-lookup the top-level neigh in the path so
1412 * we can reinject the packet there.
1415 if (dst && dst->obsolete != DST_OBSOLETE_DEAD) {
1416 n2 = dst_neigh_lookup_skb(dst, skb);
1420 n1->output(n1, skb);
1425 write_lock_bh(&neigh->lock);
1427 __skb_queue_purge(&neigh->arp_queue);
1428 neigh->arp_queue_len_bytes = 0;
1431 if (update_isrouter)
1432 neigh_update_is_router(neigh, flags, ¬ify);
1433 write_unlock_bh(&neigh->lock);
1434 if (((new ^ old) & NUD_PERMANENT) || gc_update)
1435 neigh_update_gc_list(neigh);
1437 neigh_update_managed_list(neigh);
1439 neigh_update_notify(neigh, nlmsg_pid);
1440 trace_neigh_update_done(neigh, err);
1444 int neigh_update(struct neighbour *neigh, const u8 *lladdr, u8 new,
1445 u32 flags, u32 nlmsg_pid)
1447 return __neigh_update(neigh, lladdr, new, flags, nlmsg_pid, NULL);
1449 EXPORT_SYMBOL(neigh_update);
1451 /* Update the neigh to listen temporarily for probe responses, even if it is
1452 * in a NUD_FAILED state. The caller has to hold neigh->lock for writing.
1454 void __neigh_set_probe_once(struct neighbour *neigh)
1458 neigh->updated = jiffies;
1459 if (!(neigh->nud_state & NUD_FAILED))
1461 neigh->nud_state = NUD_INCOMPLETE;
1462 atomic_set(&neigh->probes, neigh_max_probes(neigh));
1463 neigh_add_timer(neigh,
1464 jiffies + max(NEIGH_VAR(neigh->parms, RETRANS_TIME),
1467 EXPORT_SYMBOL(__neigh_set_probe_once);
1469 struct neighbour *neigh_event_ns(struct neigh_table *tbl,
1470 u8 *lladdr, void *saddr,
1471 struct net_device *dev)
1473 struct neighbour *neigh = __neigh_lookup(tbl, saddr, dev,
1474 lladdr || !dev->addr_len);
1476 neigh_update(neigh, lladdr, NUD_STALE,
1477 NEIGH_UPDATE_F_OVERRIDE, 0);
1480 EXPORT_SYMBOL(neigh_event_ns);
1482 /* called with read_lock_bh(&n->lock); */
1483 static void neigh_hh_init(struct neighbour *n)
1485 struct net_device *dev = n->dev;
1486 __be16 prot = n->tbl->protocol;
1487 struct hh_cache *hh = &n->hh;
1489 write_lock_bh(&n->lock);
1491 /* Only one thread can come in here and initialize the
1495 dev->header_ops->cache(n, hh, prot);
1497 write_unlock_bh(&n->lock);
1500 /* Slow and careful. */
1502 int neigh_resolve_output(struct neighbour *neigh, struct sk_buff *skb)
1506 if (!neigh_event_send(neigh, skb)) {
1508 struct net_device *dev = neigh->dev;
1511 if (dev->header_ops->cache && !READ_ONCE(neigh->hh.hh_len))
1512 neigh_hh_init(neigh);
1515 __skb_pull(skb, skb_network_offset(skb));
1516 seq = read_seqbegin(&neigh->ha_lock);
1517 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1518 neigh->ha, NULL, skb->len);
1519 } while (read_seqretry(&neigh->ha_lock, seq));
1522 rc = dev_queue_xmit(skb);
1533 EXPORT_SYMBOL(neigh_resolve_output);
1535 /* As fast as possible without hh cache */
1537 int neigh_connected_output(struct neighbour *neigh, struct sk_buff *skb)
1539 struct net_device *dev = neigh->dev;
1544 __skb_pull(skb, skb_network_offset(skb));
1545 seq = read_seqbegin(&neigh->ha_lock);
1546 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1547 neigh->ha, NULL, skb->len);
1548 } while (read_seqretry(&neigh->ha_lock, seq));
1551 err = dev_queue_xmit(skb);
1558 EXPORT_SYMBOL(neigh_connected_output);
1560 int neigh_direct_output(struct neighbour *neigh, struct sk_buff *skb)
1562 return dev_queue_xmit(skb);
1564 EXPORT_SYMBOL(neigh_direct_output);
1566 static void neigh_managed_work(struct work_struct *work)
1568 struct neigh_table *tbl = container_of(work, struct neigh_table,
1570 struct neighbour *neigh;
1572 write_lock_bh(&tbl->lock);
1573 list_for_each_entry(neigh, &tbl->managed_list, managed_list)
1574 neigh_event_send(neigh, NULL);
1575 queue_delayed_work(system_power_efficient_wq, &tbl->managed_work,
1576 NEIGH_VAR(&tbl->parms, DELAY_PROBE_TIME));
1577 write_unlock_bh(&tbl->lock);
1580 static void neigh_proxy_process(struct timer_list *t)
1582 struct neigh_table *tbl = from_timer(tbl, t, proxy_timer);
1583 long sched_next = 0;
1584 unsigned long now = jiffies;
1585 struct sk_buff *skb, *n;
1587 spin_lock(&tbl->proxy_queue.lock);
1589 skb_queue_walk_safe(&tbl->proxy_queue, skb, n) {
1590 long tdif = NEIGH_CB(skb)->sched_next - now;
1593 struct net_device *dev = skb->dev;
1595 __skb_unlink(skb, &tbl->proxy_queue);
1596 if (tbl->proxy_redo && netif_running(dev)) {
1598 tbl->proxy_redo(skb);
1605 } else if (!sched_next || tdif < sched_next)
1608 del_timer(&tbl->proxy_timer);
1610 mod_timer(&tbl->proxy_timer, jiffies + sched_next);
1611 spin_unlock(&tbl->proxy_queue.lock);
1614 void pneigh_enqueue(struct neigh_table *tbl, struct neigh_parms *p,
1615 struct sk_buff *skb)
1617 unsigned long sched_next = jiffies +
1618 prandom_u32_max(NEIGH_VAR(p, PROXY_DELAY));
1620 if (tbl->proxy_queue.qlen > NEIGH_VAR(p, PROXY_QLEN)) {
1625 NEIGH_CB(skb)->sched_next = sched_next;
1626 NEIGH_CB(skb)->flags |= LOCALLY_ENQUEUED;
1628 spin_lock(&tbl->proxy_queue.lock);
1629 if (del_timer(&tbl->proxy_timer)) {
1630 if (time_before(tbl->proxy_timer.expires, sched_next))
1631 sched_next = tbl->proxy_timer.expires;
1635 __skb_queue_tail(&tbl->proxy_queue, skb);
1636 mod_timer(&tbl->proxy_timer, sched_next);
1637 spin_unlock(&tbl->proxy_queue.lock);
1639 EXPORT_SYMBOL(pneigh_enqueue);
1641 static inline struct neigh_parms *lookup_neigh_parms(struct neigh_table *tbl,
1642 struct net *net, int ifindex)
1644 struct neigh_parms *p;
1646 list_for_each_entry(p, &tbl->parms_list, list) {
1647 if ((p->dev && p->dev->ifindex == ifindex && net_eq(neigh_parms_net(p), net)) ||
1648 (!p->dev && !ifindex && net_eq(net, &init_net)))
1655 struct neigh_parms *neigh_parms_alloc(struct net_device *dev,
1656 struct neigh_table *tbl)
1658 struct neigh_parms *p;
1659 struct net *net = dev_net(dev);
1660 const struct net_device_ops *ops = dev->netdev_ops;
1662 p = kmemdup(&tbl->parms, sizeof(*p), GFP_KERNEL);
1665 refcount_set(&p->refcnt, 1);
1667 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
1668 dev_hold_track(dev, &p->dev_tracker, GFP_KERNEL);
1670 write_pnet(&p->net, net);
1671 p->sysctl_table = NULL;
1673 if (ops->ndo_neigh_setup && ops->ndo_neigh_setup(dev, p)) {
1674 dev_put_track(dev, &p->dev_tracker);
1679 write_lock_bh(&tbl->lock);
1680 list_add(&p->list, &tbl->parms.list);
1681 write_unlock_bh(&tbl->lock);
1683 neigh_parms_data_state_cleanall(p);
1687 EXPORT_SYMBOL(neigh_parms_alloc);
1689 static void neigh_rcu_free_parms(struct rcu_head *head)
1691 struct neigh_parms *parms =
1692 container_of(head, struct neigh_parms, rcu_head);
1694 neigh_parms_put(parms);
1697 void neigh_parms_release(struct neigh_table *tbl, struct neigh_parms *parms)
1699 if (!parms || parms == &tbl->parms)
1701 write_lock_bh(&tbl->lock);
1702 list_del(&parms->list);
1704 write_unlock_bh(&tbl->lock);
1705 dev_put_track(parms->dev, &parms->dev_tracker);
1706 call_rcu(&parms->rcu_head, neigh_rcu_free_parms);
1708 EXPORT_SYMBOL(neigh_parms_release);
1710 static void neigh_parms_destroy(struct neigh_parms *parms)
1715 static struct lock_class_key neigh_table_proxy_queue_class;
1717 static struct neigh_table *neigh_tables[NEIGH_NR_TABLES] __read_mostly;
1719 void neigh_table_init(int index, struct neigh_table *tbl)
1721 unsigned long now = jiffies;
1722 unsigned long phsize;
1724 INIT_LIST_HEAD(&tbl->parms_list);
1725 INIT_LIST_HEAD(&tbl->gc_list);
1726 INIT_LIST_HEAD(&tbl->managed_list);
1728 list_add(&tbl->parms.list, &tbl->parms_list);
1729 write_pnet(&tbl->parms.net, &init_net);
1730 refcount_set(&tbl->parms.refcnt, 1);
1731 tbl->parms.reachable_time =
1732 neigh_rand_reach_time(NEIGH_VAR(&tbl->parms, BASE_REACHABLE_TIME));
1734 tbl->stats = alloc_percpu(struct neigh_statistics);
1736 panic("cannot create neighbour cache statistics");
1738 #ifdef CONFIG_PROC_FS
1739 if (!proc_create_seq_data(tbl->id, 0, init_net.proc_net_stat,
1740 &neigh_stat_seq_ops, tbl))
1741 panic("cannot create neighbour proc dir entry");
1744 RCU_INIT_POINTER(tbl->nht, neigh_hash_alloc(3));
1746 phsize = (PNEIGH_HASHMASK + 1) * sizeof(struct pneigh_entry *);
1747 tbl->phash_buckets = kzalloc(phsize, GFP_KERNEL);
1749 if (!tbl->nht || !tbl->phash_buckets)
1750 panic("cannot allocate neighbour cache hashes");
1752 if (!tbl->entry_size)
1753 tbl->entry_size = ALIGN(offsetof(struct neighbour, primary_key) +
1754 tbl->key_len, NEIGH_PRIV_ALIGN);
1756 WARN_ON(tbl->entry_size % NEIGH_PRIV_ALIGN);
1758 rwlock_init(&tbl->lock);
1760 INIT_DEFERRABLE_WORK(&tbl->gc_work, neigh_periodic_work);
1761 queue_delayed_work(system_power_efficient_wq, &tbl->gc_work,
1762 tbl->parms.reachable_time);
1763 INIT_DEFERRABLE_WORK(&tbl->managed_work, neigh_managed_work);
1764 queue_delayed_work(system_power_efficient_wq, &tbl->managed_work, 0);
1766 timer_setup(&tbl->proxy_timer, neigh_proxy_process, 0);
1767 skb_queue_head_init_class(&tbl->proxy_queue,
1768 &neigh_table_proxy_queue_class);
1770 tbl->last_flush = now;
1771 tbl->last_rand = now + tbl->parms.reachable_time * 20;
1773 neigh_tables[index] = tbl;
1775 EXPORT_SYMBOL(neigh_table_init);
1777 int neigh_table_clear(int index, struct neigh_table *tbl)
1779 neigh_tables[index] = NULL;
1780 /* It is not clean... Fix it to unload IPv6 module safely */
1781 cancel_delayed_work_sync(&tbl->managed_work);
1782 cancel_delayed_work_sync(&tbl->gc_work);
1783 del_timer_sync(&tbl->proxy_timer);
1784 pneigh_queue_purge(&tbl->proxy_queue);
1785 neigh_ifdown(tbl, NULL);
1786 if (atomic_read(&tbl->entries))
1787 pr_crit("neighbour leakage\n");
1789 call_rcu(&rcu_dereference_protected(tbl->nht, 1)->rcu,
1790 neigh_hash_free_rcu);
1793 kfree(tbl->phash_buckets);
1794 tbl->phash_buckets = NULL;
1796 remove_proc_entry(tbl->id, init_net.proc_net_stat);
1798 free_percpu(tbl->stats);
1803 EXPORT_SYMBOL(neigh_table_clear);
1805 static struct neigh_table *neigh_find_table(int family)
1807 struct neigh_table *tbl = NULL;
1811 tbl = neigh_tables[NEIGH_ARP_TABLE];
1814 tbl = neigh_tables[NEIGH_ND_TABLE];
1817 tbl = neigh_tables[NEIGH_DN_TABLE];
1824 const struct nla_policy nda_policy[NDA_MAX+1] = {
1825 [NDA_UNSPEC] = { .strict_start_type = NDA_NH_ID },
1826 [NDA_DST] = { .type = NLA_BINARY, .len = MAX_ADDR_LEN },
1827 [NDA_LLADDR] = { .type = NLA_BINARY, .len = MAX_ADDR_LEN },
1828 [NDA_CACHEINFO] = { .len = sizeof(struct nda_cacheinfo) },
1829 [NDA_PROBES] = { .type = NLA_U32 },
1830 [NDA_VLAN] = { .type = NLA_U16 },
1831 [NDA_PORT] = { .type = NLA_U16 },
1832 [NDA_VNI] = { .type = NLA_U32 },
1833 [NDA_IFINDEX] = { .type = NLA_U32 },
1834 [NDA_MASTER] = { .type = NLA_U32 },
1835 [NDA_PROTOCOL] = { .type = NLA_U8 },
1836 [NDA_NH_ID] = { .type = NLA_U32 },
1837 [NDA_FLAGS_EXT] = NLA_POLICY_MASK(NLA_U32, NTF_EXT_MASK),
1838 [NDA_FDB_EXT_ATTRS] = { .type = NLA_NESTED },
1841 static int neigh_delete(struct sk_buff *skb, struct nlmsghdr *nlh,
1842 struct netlink_ext_ack *extack)
1844 struct net *net = sock_net(skb->sk);
1846 struct nlattr *dst_attr;
1847 struct neigh_table *tbl;
1848 struct neighbour *neigh;
1849 struct net_device *dev = NULL;
1853 if (nlmsg_len(nlh) < sizeof(*ndm))
1856 dst_attr = nlmsg_find_attr(nlh, sizeof(*ndm), NDA_DST);
1858 NL_SET_ERR_MSG(extack, "Network address not specified");
1862 ndm = nlmsg_data(nlh);
1863 if (ndm->ndm_ifindex) {
1864 dev = __dev_get_by_index(net, ndm->ndm_ifindex);
1871 tbl = neigh_find_table(ndm->ndm_family);
1873 return -EAFNOSUPPORT;
1875 if (nla_len(dst_attr) < (int)tbl->key_len) {
1876 NL_SET_ERR_MSG(extack, "Invalid network address");
1880 if (ndm->ndm_flags & NTF_PROXY) {
1881 err = pneigh_delete(tbl, net, nla_data(dst_attr), dev);
1888 neigh = neigh_lookup(tbl, nla_data(dst_attr), dev);
1889 if (neigh == NULL) {
1894 err = __neigh_update(neigh, NULL, NUD_FAILED,
1895 NEIGH_UPDATE_F_OVERRIDE | NEIGH_UPDATE_F_ADMIN,
1896 NETLINK_CB(skb).portid, extack);
1897 write_lock_bh(&tbl->lock);
1898 neigh_release(neigh);
1899 neigh_remove_one(neigh, tbl);
1900 write_unlock_bh(&tbl->lock);
1906 static int neigh_add(struct sk_buff *skb, struct nlmsghdr *nlh,
1907 struct netlink_ext_ack *extack)
1909 int flags = NEIGH_UPDATE_F_ADMIN | NEIGH_UPDATE_F_OVERRIDE |
1910 NEIGH_UPDATE_F_OVERRIDE_ISROUTER;
1911 struct net *net = sock_net(skb->sk);
1913 struct nlattr *tb[NDA_MAX+1];
1914 struct neigh_table *tbl;
1915 struct net_device *dev = NULL;
1916 struct neighbour *neigh;
1923 err = nlmsg_parse_deprecated(nlh, sizeof(*ndm), tb, NDA_MAX,
1924 nda_policy, extack);
1930 NL_SET_ERR_MSG(extack, "Network address not specified");
1934 ndm = nlmsg_data(nlh);
1935 ndm_flags = ndm->ndm_flags;
1936 if (tb[NDA_FLAGS_EXT]) {
1937 u32 ext = nla_get_u32(tb[NDA_FLAGS_EXT]);
1939 BUILD_BUG_ON(sizeof(neigh->flags) * BITS_PER_BYTE <
1940 (sizeof(ndm->ndm_flags) * BITS_PER_BYTE +
1941 hweight32(NTF_EXT_MASK)));
1942 ndm_flags |= (ext << NTF_EXT_SHIFT);
1944 if (ndm->ndm_ifindex) {
1945 dev = __dev_get_by_index(net, ndm->ndm_ifindex);
1951 if (tb[NDA_LLADDR] && nla_len(tb[NDA_LLADDR]) < dev->addr_len) {
1952 NL_SET_ERR_MSG(extack, "Invalid link address");
1957 tbl = neigh_find_table(ndm->ndm_family);
1959 return -EAFNOSUPPORT;
1961 if (nla_len(tb[NDA_DST]) < (int)tbl->key_len) {
1962 NL_SET_ERR_MSG(extack, "Invalid network address");
1966 dst = nla_data(tb[NDA_DST]);
1967 lladdr = tb[NDA_LLADDR] ? nla_data(tb[NDA_LLADDR]) : NULL;
1969 if (tb[NDA_PROTOCOL])
1970 protocol = nla_get_u8(tb[NDA_PROTOCOL]);
1971 if (ndm_flags & NTF_PROXY) {
1972 struct pneigh_entry *pn;
1974 if (ndm_flags & NTF_MANAGED) {
1975 NL_SET_ERR_MSG(extack, "Invalid NTF_* flag combination");
1980 pn = pneigh_lookup(tbl, net, dst, dev, 1);
1982 pn->flags = ndm_flags;
1984 pn->protocol = protocol;
1991 NL_SET_ERR_MSG(extack, "Device not specified");
1995 if (tbl->allow_add && !tbl->allow_add(dev, extack)) {
2000 neigh = neigh_lookup(tbl, dst, dev);
2001 if (neigh == NULL) {
2002 bool ndm_permanent = ndm->ndm_state & NUD_PERMANENT;
2003 bool exempt_from_gc = ndm_permanent ||
2004 ndm_flags & NTF_EXT_LEARNED;
2006 if (!(nlh->nlmsg_flags & NLM_F_CREATE)) {
2010 if (ndm_permanent && (ndm_flags & NTF_MANAGED)) {
2011 NL_SET_ERR_MSG(extack, "Invalid NTF_* flag for permanent entry");
2016 neigh = ___neigh_create(tbl, dst, dev,
2018 (NTF_EXT_LEARNED | NTF_MANAGED),
2019 exempt_from_gc, true);
2020 if (IS_ERR(neigh)) {
2021 err = PTR_ERR(neigh);
2025 if (nlh->nlmsg_flags & NLM_F_EXCL) {
2027 neigh_release(neigh);
2031 if (!(nlh->nlmsg_flags & NLM_F_REPLACE))
2032 flags &= ~(NEIGH_UPDATE_F_OVERRIDE |
2033 NEIGH_UPDATE_F_OVERRIDE_ISROUTER);
2037 neigh->protocol = protocol;
2038 if (ndm_flags & NTF_EXT_LEARNED)
2039 flags |= NEIGH_UPDATE_F_EXT_LEARNED;
2040 if (ndm_flags & NTF_ROUTER)
2041 flags |= NEIGH_UPDATE_F_ISROUTER;
2042 if (ndm_flags & NTF_MANAGED)
2043 flags |= NEIGH_UPDATE_F_MANAGED;
2044 if (ndm_flags & NTF_USE)
2045 flags |= NEIGH_UPDATE_F_USE;
2047 err = __neigh_update(neigh, lladdr, ndm->ndm_state, flags,
2048 NETLINK_CB(skb).portid, extack);
2049 if (!err && ndm_flags & (NTF_USE | NTF_MANAGED)) {
2050 neigh_event_send(neigh, NULL);
2053 neigh_release(neigh);
2058 static int neightbl_fill_parms(struct sk_buff *skb, struct neigh_parms *parms)
2060 struct nlattr *nest;
2062 nest = nla_nest_start_noflag(skb, NDTA_PARMS);
2067 nla_put_u32(skb, NDTPA_IFINDEX, parms->dev->ifindex)) ||
2068 nla_put_u32(skb, NDTPA_REFCNT, refcount_read(&parms->refcnt)) ||
2069 nla_put_u32(skb, NDTPA_QUEUE_LENBYTES,
2070 NEIGH_VAR(parms, QUEUE_LEN_BYTES)) ||
2071 /* approximative value for deprecated QUEUE_LEN (in packets) */
2072 nla_put_u32(skb, NDTPA_QUEUE_LEN,
2073 NEIGH_VAR(parms, QUEUE_LEN_BYTES) / SKB_TRUESIZE(ETH_FRAME_LEN)) ||
2074 nla_put_u32(skb, NDTPA_PROXY_QLEN, NEIGH_VAR(parms, PROXY_QLEN)) ||
2075 nla_put_u32(skb, NDTPA_APP_PROBES, NEIGH_VAR(parms, APP_PROBES)) ||
2076 nla_put_u32(skb, NDTPA_UCAST_PROBES,
2077 NEIGH_VAR(parms, UCAST_PROBES)) ||
2078 nla_put_u32(skb, NDTPA_MCAST_PROBES,
2079 NEIGH_VAR(parms, MCAST_PROBES)) ||
2080 nla_put_u32(skb, NDTPA_MCAST_REPROBES,
2081 NEIGH_VAR(parms, MCAST_REPROBES)) ||
2082 nla_put_msecs(skb, NDTPA_REACHABLE_TIME, parms->reachable_time,
2084 nla_put_msecs(skb, NDTPA_BASE_REACHABLE_TIME,
2085 NEIGH_VAR(parms, BASE_REACHABLE_TIME), NDTPA_PAD) ||
2086 nla_put_msecs(skb, NDTPA_GC_STALETIME,
2087 NEIGH_VAR(parms, GC_STALETIME), NDTPA_PAD) ||
2088 nla_put_msecs(skb, NDTPA_DELAY_PROBE_TIME,
2089 NEIGH_VAR(parms, DELAY_PROBE_TIME), NDTPA_PAD) ||
2090 nla_put_msecs(skb, NDTPA_RETRANS_TIME,
2091 NEIGH_VAR(parms, RETRANS_TIME), NDTPA_PAD) ||
2092 nla_put_msecs(skb, NDTPA_ANYCAST_DELAY,
2093 NEIGH_VAR(parms, ANYCAST_DELAY), NDTPA_PAD) ||
2094 nla_put_msecs(skb, NDTPA_PROXY_DELAY,
2095 NEIGH_VAR(parms, PROXY_DELAY), NDTPA_PAD) ||
2096 nla_put_msecs(skb, NDTPA_LOCKTIME,
2097 NEIGH_VAR(parms, LOCKTIME), NDTPA_PAD))
2098 goto nla_put_failure;
2099 return nla_nest_end(skb, nest);
2102 nla_nest_cancel(skb, nest);
2106 static int neightbl_fill_info(struct sk_buff *skb, struct neigh_table *tbl,
2107 u32 pid, u32 seq, int type, int flags)
2109 struct nlmsghdr *nlh;
2110 struct ndtmsg *ndtmsg;
2112 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
2116 ndtmsg = nlmsg_data(nlh);
2118 read_lock_bh(&tbl->lock);
2119 ndtmsg->ndtm_family = tbl->family;
2120 ndtmsg->ndtm_pad1 = 0;
2121 ndtmsg->ndtm_pad2 = 0;
2123 if (nla_put_string(skb, NDTA_NAME, tbl->id) ||
2124 nla_put_msecs(skb, NDTA_GC_INTERVAL, tbl->gc_interval, NDTA_PAD) ||
2125 nla_put_u32(skb, NDTA_THRESH1, tbl->gc_thresh1) ||
2126 nla_put_u32(skb, NDTA_THRESH2, tbl->gc_thresh2) ||
2127 nla_put_u32(skb, NDTA_THRESH3, tbl->gc_thresh3))
2128 goto nla_put_failure;
2130 unsigned long now = jiffies;
2131 long flush_delta = now - tbl->last_flush;
2132 long rand_delta = now - tbl->last_rand;
2133 struct neigh_hash_table *nht;
2134 struct ndt_config ndc = {
2135 .ndtc_key_len = tbl->key_len,
2136 .ndtc_entry_size = tbl->entry_size,
2137 .ndtc_entries = atomic_read(&tbl->entries),
2138 .ndtc_last_flush = jiffies_to_msecs(flush_delta),
2139 .ndtc_last_rand = jiffies_to_msecs(rand_delta),
2140 .ndtc_proxy_qlen = tbl->proxy_queue.qlen,
2144 nht = rcu_dereference_bh(tbl->nht);
2145 ndc.ndtc_hash_rnd = nht->hash_rnd[0];
2146 ndc.ndtc_hash_mask = ((1 << nht->hash_shift) - 1);
2147 rcu_read_unlock_bh();
2149 if (nla_put(skb, NDTA_CONFIG, sizeof(ndc), &ndc))
2150 goto nla_put_failure;
2155 struct ndt_stats ndst;
2157 memset(&ndst, 0, sizeof(ndst));
2159 for_each_possible_cpu(cpu) {
2160 struct neigh_statistics *st;
2162 st = per_cpu_ptr(tbl->stats, cpu);
2163 ndst.ndts_allocs += st->allocs;
2164 ndst.ndts_destroys += st->destroys;
2165 ndst.ndts_hash_grows += st->hash_grows;
2166 ndst.ndts_res_failed += st->res_failed;
2167 ndst.ndts_lookups += st->lookups;
2168 ndst.ndts_hits += st->hits;
2169 ndst.ndts_rcv_probes_mcast += st->rcv_probes_mcast;
2170 ndst.ndts_rcv_probes_ucast += st->rcv_probes_ucast;
2171 ndst.ndts_periodic_gc_runs += st->periodic_gc_runs;
2172 ndst.ndts_forced_gc_runs += st->forced_gc_runs;
2173 ndst.ndts_table_fulls += st->table_fulls;
2176 if (nla_put_64bit(skb, NDTA_STATS, sizeof(ndst), &ndst,
2178 goto nla_put_failure;
2181 BUG_ON(tbl->parms.dev);
2182 if (neightbl_fill_parms(skb, &tbl->parms) < 0)
2183 goto nla_put_failure;
2185 read_unlock_bh(&tbl->lock);
2186 nlmsg_end(skb, nlh);
2190 read_unlock_bh(&tbl->lock);
2191 nlmsg_cancel(skb, nlh);
2195 static int neightbl_fill_param_info(struct sk_buff *skb,
2196 struct neigh_table *tbl,
2197 struct neigh_parms *parms,
2198 u32 pid, u32 seq, int type,
2201 struct ndtmsg *ndtmsg;
2202 struct nlmsghdr *nlh;
2204 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
2208 ndtmsg = nlmsg_data(nlh);
2210 read_lock_bh(&tbl->lock);
2211 ndtmsg->ndtm_family = tbl->family;
2212 ndtmsg->ndtm_pad1 = 0;
2213 ndtmsg->ndtm_pad2 = 0;
2215 if (nla_put_string(skb, NDTA_NAME, tbl->id) < 0 ||
2216 neightbl_fill_parms(skb, parms) < 0)
2219 read_unlock_bh(&tbl->lock);
2220 nlmsg_end(skb, nlh);
2223 read_unlock_bh(&tbl->lock);
2224 nlmsg_cancel(skb, nlh);
2228 static const struct nla_policy nl_neightbl_policy[NDTA_MAX+1] = {
2229 [NDTA_NAME] = { .type = NLA_STRING },
2230 [NDTA_THRESH1] = { .type = NLA_U32 },
2231 [NDTA_THRESH2] = { .type = NLA_U32 },
2232 [NDTA_THRESH3] = { .type = NLA_U32 },
2233 [NDTA_GC_INTERVAL] = { .type = NLA_U64 },
2234 [NDTA_PARMS] = { .type = NLA_NESTED },
2237 static const struct nla_policy nl_ntbl_parm_policy[NDTPA_MAX+1] = {
2238 [NDTPA_IFINDEX] = { .type = NLA_U32 },
2239 [NDTPA_QUEUE_LEN] = { .type = NLA_U32 },
2240 [NDTPA_PROXY_QLEN] = { .type = NLA_U32 },
2241 [NDTPA_APP_PROBES] = { .type = NLA_U32 },
2242 [NDTPA_UCAST_PROBES] = { .type = NLA_U32 },
2243 [NDTPA_MCAST_PROBES] = { .type = NLA_U32 },
2244 [NDTPA_MCAST_REPROBES] = { .type = NLA_U32 },
2245 [NDTPA_BASE_REACHABLE_TIME] = { .type = NLA_U64 },
2246 [NDTPA_GC_STALETIME] = { .type = NLA_U64 },
2247 [NDTPA_DELAY_PROBE_TIME] = { .type = NLA_U64 },
2248 [NDTPA_RETRANS_TIME] = { .type = NLA_U64 },
2249 [NDTPA_ANYCAST_DELAY] = { .type = NLA_U64 },
2250 [NDTPA_PROXY_DELAY] = { .type = NLA_U64 },
2251 [NDTPA_LOCKTIME] = { .type = NLA_U64 },
2254 static int neightbl_set(struct sk_buff *skb, struct nlmsghdr *nlh,
2255 struct netlink_ext_ack *extack)
2257 struct net *net = sock_net(skb->sk);
2258 struct neigh_table *tbl;
2259 struct ndtmsg *ndtmsg;
2260 struct nlattr *tb[NDTA_MAX+1];
2264 err = nlmsg_parse_deprecated(nlh, sizeof(*ndtmsg), tb, NDTA_MAX,
2265 nl_neightbl_policy, extack);
2269 if (tb[NDTA_NAME] == NULL) {
2274 ndtmsg = nlmsg_data(nlh);
2276 for (tidx = 0; tidx < NEIGH_NR_TABLES; tidx++) {
2277 tbl = neigh_tables[tidx];
2280 if (ndtmsg->ndtm_family && tbl->family != ndtmsg->ndtm_family)
2282 if (nla_strcmp(tb[NDTA_NAME], tbl->id) == 0) {
2292 * We acquire tbl->lock to be nice to the periodic timers and
2293 * make sure they always see a consistent set of values.
2295 write_lock_bh(&tbl->lock);
2297 if (tb[NDTA_PARMS]) {
2298 struct nlattr *tbp[NDTPA_MAX+1];
2299 struct neigh_parms *p;
2302 err = nla_parse_nested_deprecated(tbp, NDTPA_MAX,
2304 nl_ntbl_parm_policy, extack);
2306 goto errout_tbl_lock;
2308 if (tbp[NDTPA_IFINDEX])
2309 ifindex = nla_get_u32(tbp[NDTPA_IFINDEX]);
2311 p = lookup_neigh_parms(tbl, net, ifindex);
2314 goto errout_tbl_lock;
2317 for (i = 1; i <= NDTPA_MAX; i++) {
2322 case NDTPA_QUEUE_LEN:
2323 NEIGH_VAR_SET(p, QUEUE_LEN_BYTES,
2324 nla_get_u32(tbp[i]) *
2325 SKB_TRUESIZE(ETH_FRAME_LEN));
2327 case NDTPA_QUEUE_LENBYTES:
2328 NEIGH_VAR_SET(p, QUEUE_LEN_BYTES,
2329 nla_get_u32(tbp[i]));
2331 case NDTPA_PROXY_QLEN:
2332 NEIGH_VAR_SET(p, PROXY_QLEN,
2333 nla_get_u32(tbp[i]));
2335 case NDTPA_APP_PROBES:
2336 NEIGH_VAR_SET(p, APP_PROBES,
2337 nla_get_u32(tbp[i]));
2339 case NDTPA_UCAST_PROBES:
2340 NEIGH_VAR_SET(p, UCAST_PROBES,
2341 nla_get_u32(tbp[i]));
2343 case NDTPA_MCAST_PROBES:
2344 NEIGH_VAR_SET(p, MCAST_PROBES,
2345 nla_get_u32(tbp[i]));
2347 case NDTPA_MCAST_REPROBES:
2348 NEIGH_VAR_SET(p, MCAST_REPROBES,
2349 nla_get_u32(tbp[i]));
2351 case NDTPA_BASE_REACHABLE_TIME:
2352 NEIGH_VAR_SET(p, BASE_REACHABLE_TIME,
2353 nla_get_msecs(tbp[i]));
2354 /* update reachable_time as well, otherwise, the change will
2355 * only be effective after the next time neigh_periodic_work
2356 * decides to recompute it (can be multiple minutes)
2359 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
2361 case NDTPA_GC_STALETIME:
2362 NEIGH_VAR_SET(p, GC_STALETIME,
2363 nla_get_msecs(tbp[i]));
2365 case NDTPA_DELAY_PROBE_TIME:
2366 NEIGH_VAR_SET(p, DELAY_PROBE_TIME,
2367 nla_get_msecs(tbp[i]));
2368 call_netevent_notifiers(NETEVENT_DELAY_PROBE_TIME_UPDATE, p);
2370 case NDTPA_RETRANS_TIME:
2371 NEIGH_VAR_SET(p, RETRANS_TIME,
2372 nla_get_msecs(tbp[i]));
2374 case NDTPA_ANYCAST_DELAY:
2375 NEIGH_VAR_SET(p, ANYCAST_DELAY,
2376 nla_get_msecs(tbp[i]));
2378 case NDTPA_PROXY_DELAY:
2379 NEIGH_VAR_SET(p, PROXY_DELAY,
2380 nla_get_msecs(tbp[i]));
2382 case NDTPA_LOCKTIME:
2383 NEIGH_VAR_SET(p, LOCKTIME,
2384 nla_get_msecs(tbp[i]));
2391 if ((tb[NDTA_THRESH1] || tb[NDTA_THRESH2] ||
2392 tb[NDTA_THRESH3] || tb[NDTA_GC_INTERVAL]) &&
2393 !net_eq(net, &init_net))
2394 goto errout_tbl_lock;
2396 if (tb[NDTA_THRESH1])
2397 tbl->gc_thresh1 = nla_get_u32(tb[NDTA_THRESH1]);
2399 if (tb[NDTA_THRESH2])
2400 tbl->gc_thresh2 = nla_get_u32(tb[NDTA_THRESH2]);
2402 if (tb[NDTA_THRESH3])
2403 tbl->gc_thresh3 = nla_get_u32(tb[NDTA_THRESH3]);
2405 if (tb[NDTA_GC_INTERVAL])
2406 tbl->gc_interval = nla_get_msecs(tb[NDTA_GC_INTERVAL]);
2411 write_unlock_bh(&tbl->lock);
2416 static int neightbl_valid_dump_info(const struct nlmsghdr *nlh,
2417 struct netlink_ext_ack *extack)
2419 struct ndtmsg *ndtm;
2421 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ndtm))) {
2422 NL_SET_ERR_MSG(extack, "Invalid header for neighbor table dump request");
2426 ndtm = nlmsg_data(nlh);
2427 if (ndtm->ndtm_pad1 || ndtm->ndtm_pad2) {
2428 NL_SET_ERR_MSG(extack, "Invalid values in header for neighbor table dump request");
2432 if (nlmsg_attrlen(nlh, sizeof(*ndtm))) {
2433 NL_SET_ERR_MSG(extack, "Invalid data after header in neighbor table dump request");
2440 static int neightbl_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2442 const struct nlmsghdr *nlh = cb->nlh;
2443 struct net *net = sock_net(skb->sk);
2444 int family, tidx, nidx = 0;
2445 int tbl_skip = cb->args[0];
2446 int neigh_skip = cb->args[1];
2447 struct neigh_table *tbl;
2449 if (cb->strict_check) {
2450 int err = neightbl_valid_dump_info(nlh, cb->extack);
2456 family = ((struct rtgenmsg *)nlmsg_data(nlh))->rtgen_family;
2458 for (tidx = 0; tidx < NEIGH_NR_TABLES; tidx++) {
2459 struct neigh_parms *p;
2461 tbl = neigh_tables[tidx];
2465 if (tidx < tbl_skip || (family && tbl->family != family))
2468 if (neightbl_fill_info(skb, tbl, NETLINK_CB(cb->skb).portid,
2469 nlh->nlmsg_seq, RTM_NEWNEIGHTBL,
2474 p = list_next_entry(&tbl->parms, list);
2475 list_for_each_entry_from(p, &tbl->parms_list, list) {
2476 if (!net_eq(neigh_parms_net(p), net))
2479 if (nidx < neigh_skip)
2482 if (neightbl_fill_param_info(skb, tbl, p,
2483 NETLINK_CB(cb->skb).portid,
2501 static int neigh_fill_info(struct sk_buff *skb, struct neighbour *neigh,
2502 u32 pid, u32 seq, int type, unsigned int flags)
2504 u32 neigh_flags, neigh_flags_ext;
2505 unsigned long now = jiffies;
2506 struct nda_cacheinfo ci;
2507 struct nlmsghdr *nlh;
2510 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2514 neigh_flags_ext = neigh->flags >> NTF_EXT_SHIFT;
2515 neigh_flags = neigh->flags & NTF_OLD_MASK;
2517 ndm = nlmsg_data(nlh);
2518 ndm->ndm_family = neigh->ops->family;
2521 ndm->ndm_flags = neigh_flags;
2522 ndm->ndm_type = neigh->type;
2523 ndm->ndm_ifindex = neigh->dev->ifindex;
2525 if (nla_put(skb, NDA_DST, neigh->tbl->key_len, neigh->primary_key))
2526 goto nla_put_failure;
2528 read_lock_bh(&neigh->lock);
2529 ndm->ndm_state = neigh->nud_state;
2530 if (neigh->nud_state & NUD_VALID) {
2531 char haddr[MAX_ADDR_LEN];
2533 neigh_ha_snapshot(haddr, neigh, neigh->dev);
2534 if (nla_put(skb, NDA_LLADDR, neigh->dev->addr_len, haddr) < 0) {
2535 read_unlock_bh(&neigh->lock);
2536 goto nla_put_failure;
2540 ci.ndm_used = jiffies_to_clock_t(now - neigh->used);
2541 ci.ndm_confirmed = jiffies_to_clock_t(now - neigh->confirmed);
2542 ci.ndm_updated = jiffies_to_clock_t(now - neigh->updated);
2543 ci.ndm_refcnt = refcount_read(&neigh->refcnt) - 1;
2544 read_unlock_bh(&neigh->lock);
2546 if (nla_put_u32(skb, NDA_PROBES, atomic_read(&neigh->probes)) ||
2547 nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
2548 goto nla_put_failure;
2550 if (neigh->protocol && nla_put_u8(skb, NDA_PROTOCOL, neigh->protocol))
2551 goto nla_put_failure;
2552 if (neigh_flags_ext && nla_put_u32(skb, NDA_FLAGS_EXT, neigh_flags_ext))
2553 goto nla_put_failure;
2555 nlmsg_end(skb, nlh);
2559 nlmsg_cancel(skb, nlh);
2563 static int pneigh_fill_info(struct sk_buff *skb, struct pneigh_entry *pn,
2564 u32 pid, u32 seq, int type, unsigned int flags,
2565 struct neigh_table *tbl)
2567 u32 neigh_flags, neigh_flags_ext;
2568 struct nlmsghdr *nlh;
2571 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2575 neigh_flags_ext = pn->flags >> NTF_EXT_SHIFT;
2576 neigh_flags = pn->flags & NTF_OLD_MASK;
2578 ndm = nlmsg_data(nlh);
2579 ndm->ndm_family = tbl->family;
2582 ndm->ndm_flags = neigh_flags | NTF_PROXY;
2583 ndm->ndm_type = RTN_UNICAST;
2584 ndm->ndm_ifindex = pn->dev ? pn->dev->ifindex : 0;
2585 ndm->ndm_state = NUD_NONE;
2587 if (nla_put(skb, NDA_DST, tbl->key_len, pn->key))
2588 goto nla_put_failure;
2590 if (pn->protocol && nla_put_u8(skb, NDA_PROTOCOL, pn->protocol))
2591 goto nla_put_failure;
2592 if (neigh_flags_ext && nla_put_u32(skb, NDA_FLAGS_EXT, neigh_flags_ext))
2593 goto nla_put_failure;
2595 nlmsg_end(skb, nlh);
2599 nlmsg_cancel(skb, nlh);
2603 static void neigh_update_notify(struct neighbour *neigh, u32 nlmsg_pid)
2605 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, neigh);
2606 __neigh_notify(neigh, RTM_NEWNEIGH, 0, nlmsg_pid);
2609 static bool neigh_master_filtered(struct net_device *dev, int master_idx)
2611 struct net_device *master;
2616 master = dev ? netdev_master_upper_dev_get(dev) : NULL;
2618 /* 0 is already used to denote NDA_MASTER wasn't passed, therefore need another
2619 * invalid value for ifindex to denote "no master".
2621 if (master_idx == -1)
2624 if (!master || master->ifindex != master_idx)
2630 static bool neigh_ifindex_filtered(struct net_device *dev, int filter_idx)
2632 if (filter_idx && (!dev || dev->ifindex != filter_idx))
2638 struct neigh_dump_filter {
2643 static int neigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2644 struct netlink_callback *cb,
2645 struct neigh_dump_filter *filter)
2647 struct net *net = sock_net(skb->sk);
2648 struct neighbour *n;
2649 int rc, h, s_h = cb->args[1];
2650 int idx, s_idx = idx = cb->args[2];
2651 struct neigh_hash_table *nht;
2652 unsigned int flags = NLM_F_MULTI;
2654 if (filter->dev_idx || filter->master_idx)
2655 flags |= NLM_F_DUMP_FILTERED;
2658 nht = rcu_dereference_bh(tbl->nht);
2660 for (h = s_h; h < (1 << nht->hash_shift); h++) {
2663 for (n = rcu_dereference_bh(nht->hash_buckets[h]), idx = 0;
2665 n = rcu_dereference_bh(n->next)) {
2666 if (idx < s_idx || !net_eq(dev_net(n->dev), net))
2668 if (neigh_ifindex_filtered(n->dev, filter->dev_idx) ||
2669 neigh_master_filtered(n->dev, filter->master_idx))
2671 if (neigh_fill_info(skb, n, NETLINK_CB(cb->skb).portid,
2684 rcu_read_unlock_bh();
2690 static int pneigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2691 struct netlink_callback *cb,
2692 struct neigh_dump_filter *filter)
2694 struct pneigh_entry *n;
2695 struct net *net = sock_net(skb->sk);
2696 int rc, h, s_h = cb->args[3];
2697 int idx, s_idx = idx = cb->args[4];
2698 unsigned int flags = NLM_F_MULTI;
2700 if (filter->dev_idx || filter->master_idx)
2701 flags |= NLM_F_DUMP_FILTERED;
2703 read_lock_bh(&tbl->lock);
2705 for (h = s_h; h <= PNEIGH_HASHMASK; h++) {
2708 for (n = tbl->phash_buckets[h], idx = 0; n; n = n->next) {
2709 if (idx < s_idx || pneigh_net(n) != net)
2711 if (neigh_ifindex_filtered(n->dev, filter->dev_idx) ||
2712 neigh_master_filtered(n->dev, filter->master_idx))
2714 if (pneigh_fill_info(skb, n, NETLINK_CB(cb->skb).portid,
2716 RTM_NEWNEIGH, flags, tbl) < 0) {
2717 read_unlock_bh(&tbl->lock);
2726 read_unlock_bh(&tbl->lock);
2735 static int neigh_valid_dump_req(const struct nlmsghdr *nlh,
2737 struct neigh_dump_filter *filter,
2738 struct netlink_ext_ack *extack)
2740 struct nlattr *tb[NDA_MAX + 1];
2746 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ndm))) {
2747 NL_SET_ERR_MSG(extack, "Invalid header for neighbor dump request");
2751 ndm = nlmsg_data(nlh);
2752 if (ndm->ndm_pad1 || ndm->ndm_pad2 || ndm->ndm_ifindex ||
2753 ndm->ndm_state || ndm->ndm_type) {
2754 NL_SET_ERR_MSG(extack, "Invalid values in header for neighbor dump request");
2758 if (ndm->ndm_flags & ~NTF_PROXY) {
2759 NL_SET_ERR_MSG(extack, "Invalid flags in header for neighbor dump request");
2763 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct ndmsg),
2764 tb, NDA_MAX, nda_policy,
2767 err = nlmsg_parse_deprecated(nlh, sizeof(struct ndmsg), tb,
2768 NDA_MAX, nda_policy, extack);
2773 for (i = 0; i <= NDA_MAX; ++i) {
2777 /* all new attributes should require strict_check */
2780 filter->dev_idx = nla_get_u32(tb[i]);
2783 filter->master_idx = nla_get_u32(tb[i]);
2787 NL_SET_ERR_MSG(extack, "Unsupported attribute in neighbor dump request");
2796 static int neigh_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2798 const struct nlmsghdr *nlh = cb->nlh;
2799 struct neigh_dump_filter filter = {};
2800 struct neigh_table *tbl;
2805 family = ((struct rtgenmsg *)nlmsg_data(nlh))->rtgen_family;
2807 /* check for full ndmsg structure presence, family member is
2808 * the same for both structures
2810 if (nlmsg_len(nlh) >= sizeof(struct ndmsg) &&
2811 ((struct ndmsg *)nlmsg_data(nlh))->ndm_flags == NTF_PROXY)
2814 err = neigh_valid_dump_req(nlh, cb->strict_check, &filter, cb->extack);
2815 if (err < 0 && cb->strict_check)
2820 for (t = 0; t < NEIGH_NR_TABLES; t++) {
2821 tbl = neigh_tables[t];
2825 if (t < s_t || (family && tbl->family != family))
2828 memset(&cb->args[1], 0, sizeof(cb->args) -
2829 sizeof(cb->args[0]));
2831 err = pneigh_dump_table(tbl, skb, cb, &filter);
2833 err = neigh_dump_table(tbl, skb, cb, &filter);
2842 static int neigh_valid_get_req(const struct nlmsghdr *nlh,
2843 struct neigh_table **tbl,
2844 void **dst, int *dev_idx, u8 *ndm_flags,
2845 struct netlink_ext_ack *extack)
2847 struct nlattr *tb[NDA_MAX + 1];
2851 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ndm))) {
2852 NL_SET_ERR_MSG(extack, "Invalid header for neighbor get request");
2856 ndm = nlmsg_data(nlh);
2857 if (ndm->ndm_pad1 || ndm->ndm_pad2 || ndm->ndm_state ||
2859 NL_SET_ERR_MSG(extack, "Invalid values in header for neighbor get request");
2863 if (ndm->ndm_flags & ~NTF_PROXY) {
2864 NL_SET_ERR_MSG(extack, "Invalid flags in header for neighbor get request");
2868 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct ndmsg), tb,
2869 NDA_MAX, nda_policy, extack);
2873 *ndm_flags = ndm->ndm_flags;
2874 *dev_idx = ndm->ndm_ifindex;
2875 *tbl = neigh_find_table(ndm->ndm_family);
2877 NL_SET_ERR_MSG(extack, "Unsupported family in header for neighbor get request");
2878 return -EAFNOSUPPORT;
2881 for (i = 0; i <= NDA_MAX; ++i) {
2887 if (nla_len(tb[i]) != (int)(*tbl)->key_len) {
2888 NL_SET_ERR_MSG(extack, "Invalid network address in neighbor get request");
2891 *dst = nla_data(tb[i]);
2894 NL_SET_ERR_MSG(extack, "Unsupported attribute in neighbor get request");
2902 static inline size_t neigh_nlmsg_size(void)
2904 return NLMSG_ALIGN(sizeof(struct ndmsg))
2905 + nla_total_size(MAX_ADDR_LEN) /* NDA_DST */
2906 + nla_total_size(MAX_ADDR_LEN) /* NDA_LLADDR */
2907 + nla_total_size(sizeof(struct nda_cacheinfo))
2908 + nla_total_size(4) /* NDA_PROBES */
2909 + nla_total_size(4) /* NDA_FLAGS_EXT */
2910 + nla_total_size(1); /* NDA_PROTOCOL */
2913 static int neigh_get_reply(struct net *net, struct neighbour *neigh,
2916 struct sk_buff *skb;
2919 skb = nlmsg_new(neigh_nlmsg_size(), GFP_KERNEL);
2923 err = neigh_fill_info(skb, neigh, pid, seq, RTM_NEWNEIGH, 0);
2929 err = rtnl_unicast(skb, net, pid);
2934 static inline size_t pneigh_nlmsg_size(void)
2936 return NLMSG_ALIGN(sizeof(struct ndmsg))
2937 + nla_total_size(MAX_ADDR_LEN) /* NDA_DST */
2938 + nla_total_size(4) /* NDA_FLAGS_EXT */
2939 + nla_total_size(1); /* NDA_PROTOCOL */
2942 static int pneigh_get_reply(struct net *net, struct pneigh_entry *neigh,
2943 u32 pid, u32 seq, struct neigh_table *tbl)
2945 struct sk_buff *skb;
2948 skb = nlmsg_new(pneigh_nlmsg_size(), GFP_KERNEL);
2952 err = pneigh_fill_info(skb, neigh, pid, seq, RTM_NEWNEIGH, 0, tbl);
2958 err = rtnl_unicast(skb, net, pid);
2963 static int neigh_get(struct sk_buff *in_skb, struct nlmsghdr *nlh,
2964 struct netlink_ext_ack *extack)
2966 struct net *net = sock_net(in_skb->sk);
2967 struct net_device *dev = NULL;
2968 struct neigh_table *tbl = NULL;
2969 struct neighbour *neigh;
2975 err = neigh_valid_get_req(nlh, &tbl, &dst, &dev_idx, &ndm_flags,
2981 dev = __dev_get_by_index(net, dev_idx);
2983 NL_SET_ERR_MSG(extack, "Unknown device ifindex");
2989 NL_SET_ERR_MSG(extack, "Network address not specified");
2993 if (ndm_flags & NTF_PROXY) {
2994 struct pneigh_entry *pn;
2996 pn = pneigh_lookup(tbl, net, dst, dev, 0);
2998 NL_SET_ERR_MSG(extack, "Proxy neighbour entry not found");
3001 return pneigh_get_reply(net, pn, NETLINK_CB(in_skb).portid,
3002 nlh->nlmsg_seq, tbl);
3006 NL_SET_ERR_MSG(extack, "No device specified");
3010 neigh = neigh_lookup(tbl, dst, dev);
3012 NL_SET_ERR_MSG(extack, "Neighbour entry not found");
3016 err = neigh_get_reply(net, neigh, NETLINK_CB(in_skb).portid,
3019 neigh_release(neigh);
3024 void neigh_for_each(struct neigh_table *tbl, void (*cb)(struct neighbour *, void *), void *cookie)
3027 struct neigh_hash_table *nht;
3030 nht = rcu_dereference_bh(tbl->nht);
3032 read_lock(&tbl->lock); /* avoid resizes */
3033 for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
3034 struct neighbour *n;
3036 for (n = rcu_dereference_bh(nht->hash_buckets[chain]);
3038 n = rcu_dereference_bh(n->next))
3041 read_unlock(&tbl->lock);
3042 rcu_read_unlock_bh();
3044 EXPORT_SYMBOL(neigh_for_each);
3046 /* The tbl->lock must be held as a writer and BH disabled. */
3047 void __neigh_for_each_release(struct neigh_table *tbl,
3048 int (*cb)(struct neighbour *))
3051 struct neigh_hash_table *nht;
3053 nht = rcu_dereference_protected(tbl->nht,
3054 lockdep_is_held(&tbl->lock));
3055 for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
3056 struct neighbour *n;
3057 struct neighbour __rcu **np;
3059 np = &nht->hash_buckets[chain];
3060 while ((n = rcu_dereference_protected(*np,
3061 lockdep_is_held(&tbl->lock))) != NULL) {
3064 write_lock(&n->lock);
3067 rcu_assign_pointer(*np,
3068 rcu_dereference_protected(n->next,
3069 lockdep_is_held(&tbl->lock)));
3073 write_unlock(&n->lock);
3075 neigh_cleanup_and_release(n);
3079 EXPORT_SYMBOL(__neigh_for_each_release);
3081 int neigh_xmit(int index, struct net_device *dev,
3082 const void *addr, struct sk_buff *skb)
3084 int err = -EAFNOSUPPORT;
3085 if (likely(index < NEIGH_NR_TABLES)) {
3086 struct neigh_table *tbl;
3087 struct neighbour *neigh;
3089 tbl = neigh_tables[index];
3093 if (index == NEIGH_ARP_TABLE) {
3094 u32 key = *((u32 *)addr);
3096 neigh = __ipv4_neigh_lookup_noref(dev, key);
3098 neigh = __neigh_lookup_noref(tbl, addr, dev);
3101 neigh = __neigh_create(tbl, addr, dev, false);
3102 err = PTR_ERR(neigh);
3103 if (IS_ERR(neigh)) {
3104 rcu_read_unlock_bh();
3107 err = neigh->output(neigh, skb);
3108 rcu_read_unlock_bh();
3110 else if (index == NEIGH_LINK_TABLE) {
3111 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
3112 addr, NULL, skb->len);
3115 err = dev_queue_xmit(skb);
3123 EXPORT_SYMBOL(neigh_xmit);
3125 #ifdef CONFIG_PROC_FS
3127 static struct neighbour *neigh_get_first(struct seq_file *seq)
3129 struct neigh_seq_state *state = seq->private;
3130 struct net *net = seq_file_net(seq);
3131 struct neigh_hash_table *nht = state->nht;
3132 struct neighbour *n = NULL;
3135 state->flags &= ~NEIGH_SEQ_IS_PNEIGH;
3136 for (bucket = 0; bucket < (1 << nht->hash_shift); bucket++) {
3137 n = rcu_dereference_bh(nht->hash_buckets[bucket]);
3140 if (!net_eq(dev_net(n->dev), net))
3142 if (state->neigh_sub_iter) {
3146 v = state->neigh_sub_iter(state, n, &fakep);
3150 if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
3152 if (n->nud_state & ~NUD_NOARP)
3155 n = rcu_dereference_bh(n->next);
3161 state->bucket = bucket;
3166 static struct neighbour *neigh_get_next(struct seq_file *seq,
3167 struct neighbour *n,
3170 struct neigh_seq_state *state = seq->private;
3171 struct net *net = seq_file_net(seq);
3172 struct neigh_hash_table *nht = state->nht;
3174 if (state->neigh_sub_iter) {
3175 void *v = state->neigh_sub_iter(state, n, pos);
3179 n = rcu_dereference_bh(n->next);
3183 if (!net_eq(dev_net(n->dev), net))
3185 if (state->neigh_sub_iter) {
3186 void *v = state->neigh_sub_iter(state, n, pos);
3191 if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
3194 if (n->nud_state & ~NUD_NOARP)
3197 n = rcu_dereference_bh(n->next);
3203 if (++state->bucket >= (1 << nht->hash_shift))
3206 n = rcu_dereference_bh(nht->hash_buckets[state->bucket]);
3214 static struct neighbour *neigh_get_idx(struct seq_file *seq, loff_t *pos)
3216 struct neighbour *n = neigh_get_first(seq);
3221 n = neigh_get_next(seq, n, pos);
3226 return *pos ? NULL : n;
3229 static struct pneigh_entry *pneigh_get_first(struct seq_file *seq)
3231 struct neigh_seq_state *state = seq->private;
3232 struct net *net = seq_file_net(seq);
3233 struct neigh_table *tbl = state->tbl;
3234 struct pneigh_entry *pn = NULL;
3237 state->flags |= NEIGH_SEQ_IS_PNEIGH;
3238 for (bucket = 0; bucket <= PNEIGH_HASHMASK; bucket++) {
3239 pn = tbl->phash_buckets[bucket];
3240 while (pn && !net_eq(pneigh_net(pn), net))
3245 state->bucket = bucket;
3250 static struct pneigh_entry *pneigh_get_next(struct seq_file *seq,
3251 struct pneigh_entry *pn,
3254 struct neigh_seq_state *state = seq->private;
3255 struct net *net = seq_file_net(seq);
3256 struct neigh_table *tbl = state->tbl;
3260 } while (pn && !net_eq(pneigh_net(pn), net));
3263 if (++state->bucket > PNEIGH_HASHMASK)
3265 pn = tbl->phash_buckets[state->bucket];
3266 while (pn && !net_eq(pneigh_net(pn), net))
3278 static struct pneigh_entry *pneigh_get_idx(struct seq_file *seq, loff_t *pos)
3280 struct pneigh_entry *pn = pneigh_get_first(seq);
3285 pn = pneigh_get_next(seq, pn, pos);
3290 return *pos ? NULL : pn;
3293 static void *neigh_get_idx_any(struct seq_file *seq, loff_t *pos)
3295 struct neigh_seq_state *state = seq->private;
3297 loff_t idxpos = *pos;
3299 rc = neigh_get_idx(seq, &idxpos);
3300 if (!rc && !(state->flags & NEIGH_SEQ_NEIGH_ONLY))
3301 rc = pneigh_get_idx(seq, &idxpos);
3306 void *neigh_seq_start(struct seq_file *seq, loff_t *pos, struct neigh_table *tbl, unsigned int neigh_seq_flags)
3307 __acquires(tbl->lock)
3310 struct neigh_seq_state *state = seq->private;
3314 state->flags = (neigh_seq_flags & ~NEIGH_SEQ_IS_PNEIGH);
3317 state->nht = rcu_dereference_bh(tbl->nht);
3318 read_lock(&tbl->lock);
3320 return *pos ? neigh_get_idx_any(seq, pos) : SEQ_START_TOKEN;
3322 EXPORT_SYMBOL(neigh_seq_start);
3324 void *neigh_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3326 struct neigh_seq_state *state;
3329 if (v == SEQ_START_TOKEN) {
3330 rc = neigh_get_first(seq);
3334 state = seq->private;
3335 if (!(state->flags & NEIGH_SEQ_IS_PNEIGH)) {
3336 rc = neigh_get_next(seq, v, NULL);
3339 if (!(state->flags & NEIGH_SEQ_NEIGH_ONLY))
3340 rc = pneigh_get_first(seq);
3342 BUG_ON(state->flags & NEIGH_SEQ_NEIGH_ONLY);
3343 rc = pneigh_get_next(seq, v, NULL);
3349 EXPORT_SYMBOL(neigh_seq_next);
3351 void neigh_seq_stop(struct seq_file *seq, void *v)
3352 __releases(tbl->lock)
3355 struct neigh_seq_state *state = seq->private;
3356 struct neigh_table *tbl = state->tbl;
3358 read_unlock(&tbl->lock);
3359 rcu_read_unlock_bh();
3361 EXPORT_SYMBOL(neigh_seq_stop);
3363 /* statistics via seq_file */
3365 static void *neigh_stat_seq_start(struct seq_file *seq, loff_t *pos)
3367 struct neigh_table *tbl = pde_data(file_inode(seq->file));
3371 return SEQ_START_TOKEN;
3373 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
3374 if (!cpu_possible(cpu))
3377 return per_cpu_ptr(tbl->stats, cpu);
3382 static void *neigh_stat_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3384 struct neigh_table *tbl = pde_data(file_inode(seq->file));
3387 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
3388 if (!cpu_possible(cpu))
3391 return per_cpu_ptr(tbl->stats, cpu);
3397 static void neigh_stat_seq_stop(struct seq_file *seq, void *v)
3402 static int neigh_stat_seq_show(struct seq_file *seq, void *v)
3404 struct neigh_table *tbl = pde_data(file_inode(seq->file));
3405 struct neigh_statistics *st = v;
3407 if (v == SEQ_START_TOKEN) {
3408 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");
3412 seq_printf(seq, "%08x %08lx %08lx %08lx %08lx %08lx %08lx "
3413 "%08lx %08lx %08lx "
3414 "%08lx %08lx %08lx\n",
3415 atomic_read(&tbl->entries),
3426 st->rcv_probes_mcast,
3427 st->rcv_probes_ucast,
3429 st->periodic_gc_runs,
3438 static const struct seq_operations neigh_stat_seq_ops = {
3439 .start = neigh_stat_seq_start,
3440 .next = neigh_stat_seq_next,
3441 .stop = neigh_stat_seq_stop,
3442 .show = neigh_stat_seq_show,
3444 #endif /* CONFIG_PROC_FS */
3446 static void __neigh_notify(struct neighbour *n, int type, int flags,
3449 struct net *net = dev_net(n->dev);
3450 struct sk_buff *skb;
3453 skb = nlmsg_new(neigh_nlmsg_size(), GFP_ATOMIC);
3457 err = neigh_fill_info(skb, n, pid, 0, type, flags);
3459 /* -EMSGSIZE implies BUG in neigh_nlmsg_size() */
3460 WARN_ON(err == -EMSGSIZE);
3464 rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
3468 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
3471 void neigh_app_ns(struct neighbour *n)
3473 __neigh_notify(n, RTM_GETNEIGH, NLM_F_REQUEST, 0);
3475 EXPORT_SYMBOL(neigh_app_ns);
3477 #ifdef CONFIG_SYSCTL
3478 static int unres_qlen_max = INT_MAX / SKB_TRUESIZE(ETH_FRAME_LEN);
3480 static int proc_unres_qlen(struct ctl_table *ctl, int write,
3481 void *buffer, size_t *lenp, loff_t *ppos)
3484 struct ctl_table tmp = *ctl;
3486 tmp.extra1 = SYSCTL_ZERO;
3487 tmp.extra2 = &unres_qlen_max;
3490 size = *(int *)ctl->data / SKB_TRUESIZE(ETH_FRAME_LEN);
3491 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
3494 *(int *)ctl->data = size * SKB_TRUESIZE(ETH_FRAME_LEN);
3498 static struct neigh_parms *neigh_get_dev_parms_rcu(struct net_device *dev,
3503 return __in_dev_arp_parms_get_rcu(dev);
3505 return __in6_dev_nd_parms_get_rcu(dev);
3510 static void neigh_copy_dflt_parms(struct net *net, struct neigh_parms *p,
3513 struct net_device *dev;
3514 int family = neigh_parms_family(p);
3517 for_each_netdev_rcu(net, dev) {
3518 struct neigh_parms *dst_p =
3519 neigh_get_dev_parms_rcu(dev, family);
3521 if (dst_p && !test_bit(index, dst_p->data_state))
3522 dst_p->data[index] = p->data[index];
3527 static void neigh_proc_update(struct ctl_table *ctl, int write)
3529 struct net_device *dev = ctl->extra1;
3530 struct neigh_parms *p = ctl->extra2;
3531 struct net *net = neigh_parms_net(p);
3532 int index = (int *) ctl->data - p->data;
3537 set_bit(index, p->data_state);
3538 if (index == NEIGH_VAR_DELAY_PROBE_TIME)
3539 call_netevent_notifiers(NETEVENT_DELAY_PROBE_TIME_UPDATE, p);
3540 if (!dev) /* NULL dev means this is default value */
3541 neigh_copy_dflt_parms(net, p, index);
3544 static int neigh_proc_dointvec_zero_intmax(struct ctl_table *ctl, int write,
3545 void *buffer, size_t *lenp,
3548 struct ctl_table tmp = *ctl;
3551 tmp.extra1 = SYSCTL_ZERO;
3552 tmp.extra2 = SYSCTL_INT_MAX;
3554 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
3555 neigh_proc_update(ctl, write);
3559 int neigh_proc_dointvec(struct ctl_table *ctl, int write, void *buffer,
3560 size_t *lenp, loff_t *ppos)
3562 int ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
3564 neigh_proc_update(ctl, write);
3567 EXPORT_SYMBOL(neigh_proc_dointvec);
3569 int neigh_proc_dointvec_jiffies(struct ctl_table *ctl, int write, void *buffer,
3570 size_t *lenp, loff_t *ppos)
3572 int ret = proc_dointvec_jiffies(ctl, write, buffer, lenp, ppos);
3574 neigh_proc_update(ctl, write);
3577 EXPORT_SYMBOL(neigh_proc_dointvec_jiffies);
3579 static int neigh_proc_dointvec_userhz_jiffies(struct ctl_table *ctl, int write,
3580 void *buffer, size_t *lenp,
3583 int ret = proc_dointvec_userhz_jiffies(ctl, write, buffer, lenp, ppos);
3585 neigh_proc_update(ctl, write);
3589 int neigh_proc_dointvec_ms_jiffies(struct ctl_table *ctl, int write,
3590 void *buffer, size_t *lenp, loff_t *ppos)
3592 int ret = proc_dointvec_ms_jiffies(ctl, write, buffer, lenp, ppos);
3594 neigh_proc_update(ctl, write);
3597 EXPORT_SYMBOL(neigh_proc_dointvec_ms_jiffies);
3599 static int neigh_proc_dointvec_unres_qlen(struct ctl_table *ctl, int write,
3600 void *buffer, size_t *lenp,
3603 int ret = proc_unres_qlen(ctl, write, buffer, lenp, ppos);
3605 neigh_proc_update(ctl, write);
3609 static int neigh_proc_base_reachable_time(struct ctl_table *ctl, int write,
3610 void *buffer, size_t *lenp,
3613 struct neigh_parms *p = ctl->extra2;
3616 if (strcmp(ctl->procname, "base_reachable_time") == 0)
3617 ret = neigh_proc_dointvec_jiffies(ctl, write, buffer, lenp, ppos);
3618 else if (strcmp(ctl->procname, "base_reachable_time_ms") == 0)
3619 ret = neigh_proc_dointvec_ms_jiffies(ctl, write, buffer, lenp, ppos);
3623 if (write && ret == 0) {
3624 /* update reachable_time as well, otherwise, the change will
3625 * only be effective after the next time neigh_periodic_work
3626 * decides to recompute it
3629 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
3634 #define NEIGH_PARMS_DATA_OFFSET(index) \
3635 (&((struct neigh_parms *) 0)->data[index])
3637 #define NEIGH_SYSCTL_ENTRY(attr, data_attr, name, mval, proc) \
3638 [NEIGH_VAR_ ## attr] = { \
3640 .data = NEIGH_PARMS_DATA_OFFSET(NEIGH_VAR_ ## data_attr), \
3641 .maxlen = sizeof(int), \
3643 .proc_handler = proc, \
3646 #define NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(attr, name) \
3647 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_zero_intmax)
3649 #define NEIGH_SYSCTL_JIFFIES_ENTRY(attr, name) \
3650 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_jiffies)
3652 #define NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(attr, name) \
3653 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_userhz_jiffies)
3655 #define NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(attr, data_attr, name) \
3656 NEIGH_SYSCTL_ENTRY(attr, data_attr, name, 0644, neigh_proc_dointvec_ms_jiffies)
3658 #define NEIGH_SYSCTL_UNRES_QLEN_REUSED_ENTRY(attr, data_attr, name) \
3659 NEIGH_SYSCTL_ENTRY(attr, data_attr, name, 0644, neigh_proc_dointvec_unres_qlen)
3661 static struct neigh_sysctl_table {
3662 struct ctl_table_header *sysctl_header;
3663 struct ctl_table neigh_vars[NEIGH_VAR_MAX + 1];
3664 } neigh_sysctl_template __read_mostly = {
3666 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(MCAST_PROBES, "mcast_solicit"),
3667 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(UCAST_PROBES, "ucast_solicit"),
3668 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(APP_PROBES, "app_solicit"),
3669 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(MCAST_REPROBES, "mcast_resolicit"),
3670 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(RETRANS_TIME, "retrans_time"),
3671 NEIGH_SYSCTL_JIFFIES_ENTRY(BASE_REACHABLE_TIME, "base_reachable_time"),
3672 NEIGH_SYSCTL_JIFFIES_ENTRY(DELAY_PROBE_TIME, "delay_first_probe_time"),
3673 NEIGH_SYSCTL_JIFFIES_ENTRY(GC_STALETIME, "gc_stale_time"),
3674 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(QUEUE_LEN_BYTES, "unres_qlen_bytes"),
3675 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(PROXY_QLEN, "proxy_qlen"),
3676 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(ANYCAST_DELAY, "anycast_delay"),
3677 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(PROXY_DELAY, "proxy_delay"),
3678 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(LOCKTIME, "locktime"),
3679 NEIGH_SYSCTL_UNRES_QLEN_REUSED_ENTRY(QUEUE_LEN, QUEUE_LEN_BYTES, "unres_qlen"),
3680 NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(RETRANS_TIME_MS, RETRANS_TIME, "retrans_time_ms"),
3681 NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(BASE_REACHABLE_TIME_MS, BASE_REACHABLE_TIME, "base_reachable_time_ms"),
3682 [NEIGH_VAR_GC_INTERVAL] = {
3683 .procname = "gc_interval",
3684 .maxlen = sizeof(int),
3686 .proc_handler = proc_dointvec_jiffies,
3688 [NEIGH_VAR_GC_THRESH1] = {
3689 .procname = "gc_thresh1",
3690 .maxlen = sizeof(int),
3692 .extra1 = SYSCTL_ZERO,
3693 .extra2 = SYSCTL_INT_MAX,
3694 .proc_handler = proc_dointvec_minmax,
3696 [NEIGH_VAR_GC_THRESH2] = {
3697 .procname = "gc_thresh2",
3698 .maxlen = sizeof(int),
3700 .extra1 = SYSCTL_ZERO,
3701 .extra2 = SYSCTL_INT_MAX,
3702 .proc_handler = proc_dointvec_minmax,
3704 [NEIGH_VAR_GC_THRESH3] = {
3705 .procname = "gc_thresh3",
3706 .maxlen = sizeof(int),
3708 .extra1 = SYSCTL_ZERO,
3709 .extra2 = SYSCTL_INT_MAX,
3710 .proc_handler = proc_dointvec_minmax,
3716 int neigh_sysctl_register(struct net_device *dev, struct neigh_parms *p,
3717 proc_handler *handler)
3720 struct neigh_sysctl_table *t;
3721 const char *dev_name_source;
3722 char neigh_path[ sizeof("net//neigh/") + IFNAMSIZ + IFNAMSIZ ];
3725 t = kmemdup(&neigh_sysctl_template, sizeof(*t), GFP_KERNEL);
3729 for (i = 0; i < NEIGH_VAR_GC_INTERVAL; i++) {
3730 t->neigh_vars[i].data += (long) p;
3731 t->neigh_vars[i].extra1 = dev;
3732 t->neigh_vars[i].extra2 = p;
3736 dev_name_source = dev->name;
3737 /* Terminate the table early */
3738 memset(&t->neigh_vars[NEIGH_VAR_GC_INTERVAL], 0,
3739 sizeof(t->neigh_vars[NEIGH_VAR_GC_INTERVAL]));
3741 struct neigh_table *tbl = p->tbl;
3742 dev_name_source = "default";
3743 t->neigh_vars[NEIGH_VAR_GC_INTERVAL].data = &tbl->gc_interval;
3744 t->neigh_vars[NEIGH_VAR_GC_THRESH1].data = &tbl->gc_thresh1;
3745 t->neigh_vars[NEIGH_VAR_GC_THRESH2].data = &tbl->gc_thresh2;
3746 t->neigh_vars[NEIGH_VAR_GC_THRESH3].data = &tbl->gc_thresh3;
3751 t->neigh_vars[NEIGH_VAR_RETRANS_TIME].proc_handler = handler;
3753 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].proc_handler = handler;
3754 /* RetransTime (in milliseconds)*/
3755 t->neigh_vars[NEIGH_VAR_RETRANS_TIME_MS].proc_handler = handler;
3756 /* ReachableTime (in milliseconds) */
3757 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].proc_handler = handler;
3759 /* Those handlers will update p->reachable_time after
3760 * base_reachable_time(_ms) is set to ensure the new timer starts being
3761 * applied after the next neighbour update instead of waiting for
3762 * neigh_periodic_work to update its value (can be multiple minutes)
3763 * So any handler that replaces them should do this as well
3766 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].proc_handler =
3767 neigh_proc_base_reachable_time;
3768 /* ReachableTime (in milliseconds) */
3769 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].proc_handler =
3770 neigh_proc_base_reachable_time;
3773 switch (neigh_parms_family(p)) {
3784 snprintf(neigh_path, sizeof(neigh_path), "net/%s/neigh/%s",
3785 p_name, dev_name_source);
3787 register_net_sysctl(neigh_parms_net(p), neigh_path, t->neigh_vars);
3788 if (!t->sysctl_header)
3791 p->sysctl_table = t;
3799 EXPORT_SYMBOL(neigh_sysctl_register);
3801 void neigh_sysctl_unregister(struct neigh_parms *p)
3803 if (p->sysctl_table) {
3804 struct neigh_sysctl_table *t = p->sysctl_table;
3805 p->sysctl_table = NULL;
3806 unregister_net_sysctl_table(t->sysctl_header);
3810 EXPORT_SYMBOL(neigh_sysctl_unregister);
3812 #endif /* CONFIG_SYSCTL */
3814 static int __init neigh_init(void)
3816 rtnl_register(PF_UNSPEC, RTM_NEWNEIGH, neigh_add, NULL, 0);
3817 rtnl_register(PF_UNSPEC, RTM_DELNEIGH, neigh_delete, NULL, 0);
3818 rtnl_register(PF_UNSPEC, RTM_GETNEIGH, neigh_get, neigh_dump_info, 0);
3820 rtnl_register(PF_UNSPEC, RTM_GETNEIGHTBL, NULL, neightbl_dump_info,
3822 rtnl_register(PF_UNSPEC, RTM_SETNEIGHTBL, neightbl_set, NULL, 0);
3827 subsys_initcall(neigh_init);