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);
127 static void neigh_update_gc_list(struct neighbour *n)
129 bool on_gc_list, exempt_from_gc;
131 write_lock_bh(&n->tbl->lock);
132 write_lock(&n->lock);
137 /* remove from the gc list if new state is permanent or if neighbor
138 * is externally learned; otherwise entry should be on the gc list
140 exempt_from_gc = n->nud_state & NUD_PERMANENT ||
141 n->flags & NTF_EXT_LEARNED;
142 on_gc_list = !list_empty(&n->gc_list);
144 if (exempt_from_gc && on_gc_list) {
145 list_del_init(&n->gc_list);
146 atomic_dec(&n->tbl->gc_entries);
147 } else if (!exempt_from_gc && !on_gc_list) {
148 /* add entries to the tail; cleaning removes from the front */
149 list_add_tail(&n->gc_list, &n->tbl->gc_list);
150 atomic_inc(&n->tbl->gc_entries);
154 write_unlock(&n->lock);
155 write_unlock_bh(&n->tbl->lock);
158 static bool neigh_update_ext_learned(struct neighbour *neigh, u32 flags,
164 if (!(flags & NEIGH_UPDATE_F_ADMIN))
167 ndm_flags = (flags & NEIGH_UPDATE_F_EXT_LEARNED) ? NTF_EXT_LEARNED : 0;
168 if ((neigh->flags ^ ndm_flags) & NTF_EXT_LEARNED) {
169 if (ndm_flags & NTF_EXT_LEARNED)
170 neigh->flags |= NTF_EXT_LEARNED;
172 neigh->flags &= ~NTF_EXT_LEARNED;
180 static bool neigh_del(struct neighbour *n, struct neighbour __rcu **np,
181 struct neigh_table *tbl)
185 write_lock(&n->lock);
186 if (refcount_read(&n->refcnt) == 1) {
187 struct neighbour *neigh;
189 neigh = rcu_dereference_protected(n->next,
190 lockdep_is_held(&tbl->lock));
191 rcu_assign_pointer(*np, neigh);
195 write_unlock(&n->lock);
197 neigh_cleanup_and_release(n);
201 bool neigh_remove_one(struct neighbour *ndel, struct neigh_table *tbl)
203 struct neigh_hash_table *nht;
204 void *pkey = ndel->primary_key;
207 struct neighbour __rcu **np;
209 nht = rcu_dereference_protected(tbl->nht,
210 lockdep_is_held(&tbl->lock));
211 hash_val = tbl->hash(pkey, ndel->dev, nht->hash_rnd);
212 hash_val = hash_val >> (32 - nht->hash_shift);
214 np = &nht->hash_buckets[hash_val];
215 while ((n = rcu_dereference_protected(*np,
216 lockdep_is_held(&tbl->lock)))) {
218 return neigh_del(n, np, tbl);
224 static int neigh_forced_gc(struct neigh_table *tbl)
226 int max_clean = atomic_read(&tbl->gc_entries) - tbl->gc_thresh2;
227 unsigned long tref = jiffies - 5 * HZ;
228 struct neighbour *n, *tmp;
231 NEIGH_CACHE_STAT_INC(tbl, forced_gc_runs);
233 write_lock_bh(&tbl->lock);
235 list_for_each_entry_safe(n, tmp, &tbl->gc_list, gc_list) {
236 if (refcount_read(&n->refcnt) == 1) {
239 write_lock(&n->lock);
240 if ((n->nud_state == NUD_FAILED) ||
241 (n->nud_state == NUD_NOARP) ||
242 (tbl->is_multicast &&
243 tbl->is_multicast(n->primary_key)) ||
244 time_after(tref, n->updated))
246 write_unlock(&n->lock);
248 if (remove && neigh_remove_one(n, tbl))
250 if (shrunk >= max_clean)
255 tbl->last_flush = jiffies;
257 write_unlock_bh(&tbl->lock);
262 static void neigh_add_timer(struct neighbour *n, unsigned long when)
265 if (unlikely(mod_timer(&n->timer, when))) {
266 printk("NEIGH: BUG, double timer add, state is %x\n",
272 static int neigh_del_timer(struct neighbour *n)
274 if ((n->nud_state & NUD_IN_TIMER) &&
275 del_timer(&n->timer)) {
282 static void pneigh_queue_purge(struct sk_buff_head *list)
286 while ((skb = skb_dequeue(list)) != NULL) {
292 static void neigh_flush_dev(struct neigh_table *tbl, struct net_device *dev,
296 struct neigh_hash_table *nht;
298 nht = rcu_dereference_protected(tbl->nht,
299 lockdep_is_held(&tbl->lock));
301 for (i = 0; i < (1 << nht->hash_shift); i++) {
303 struct neighbour __rcu **np = &nht->hash_buckets[i];
305 while ((n = rcu_dereference_protected(*np,
306 lockdep_is_held(&tbl->lock))) != NULL) {
307 if (dev && n->dev != dev) {
311 if (skip_perm && n->nud_state & NUD_PERMANENT) {
315 rcu_assign_pointer(*np,
316 rcu_dereference_protected(n->next,
317 lockdep_is_held(&tbl->lock)));
318 write_lock(&n->lock);
321 if (refcount_read(&n->refcnt) != 1) {
322 /* The most unpleasant situation.
323 We must destroy neighbour entry,
324 but someone still uses it.
326 The destroy will be delayed until
327 the last user releases us, but
328 we must kill timers etc. and move
331 __skb_queue_purge(&n->arp_queue);
332 n->arp_queue_len_bytes = 0;
333 n->output = neigh_blackhole;
334 if (n->nud_state & NUD_VALID)
335 n->nud_state = NUD_NOARP;
337 n->nud_state = NUD_NONE;
338 neigh_dbg(2, "neigh %p is stray\n", n);
340 write_unlock(&n->lock);
341 neigh_cleanup_and_release(n);
346 void neigh_changeaddr(struct neigh_table *tbl, struct net_device *dev)
348 write_lock_bh(&tbl->lock);
349 neigh_flush_dev(tbl, dev, false);
350 write_unlock_bh(&tbl->lock);
352 EXPORT_SYMBOL(neigh_changeaddr);
354 static int __neigh_ifdown(struct neigh_table *tbl, struct net_device *dev,
357 write_lock_bh(&tbl->lock);
358 neigh_flush_dev(tbl, dev, skip_perm);
359 pneigh_ifdown_and_unlock(tbl, dev);
361 del_timer_sync(&tbl->proxy_timer);
362 pneigh_queue_purge(&tbl->proxy_queue);
366 int neigh_carrier_down(struct neigh_table *tbl, struct net_device *dev)
368 __neigh_ifdown(tbl, dev, true);
371 EXPORT_SYMBOL(neigh_carrier_down);
373 int neigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
375 __neigh_ifdown(tbl, dev, false);
378 EXPORT_SYMBOL(neigh_ifdown);
380 static struct neighbour *neigh_alloc(struct neigh_table *tbl,
381 struct net_device *dev,
382 u8 flags, bool exempt_from_gc)
384 struct neighbour *n = NULL;
385 unsigned long now = jiffies;
391 entries = atomic_inc_return(&tbl->gc_entries) - 1;
392 if (entries >= tbl->gc_thresh3 ||
393 (entries >= tbl->gc_thresh2 &&
394 time_after(now, tbl->last_flush + 5 * HZ))) {
395 if (!neigh_forced_gc(tbl) &&
396 entries >= tbl->gc_thresh3) {
397 net_info_ratelimited("%s: neighbor table overflow!\n",
399 NEIGH_CACHE_STAT_INC(tbl, table_fulls);
405 n = kzalloc(tbl->entry_size + dev->neigh_priv_len, GFP_ATOMIC);
409 __skb_queue_head_init(&n->arp_queue);
410 rwlock_init(&n->lock);
411 seqlock_init(&n->ha_lock);
412 n->updated = n->used = now;
413 n->nud_state = NUD_NONE;
414 n->output = neigh_blackhole;
416 seqlock_init(&n->hh.hh_lock);
417 n->parms = neigh_parms_clone(&tbl->parms);
418 timer_setup(&n->timer, neigh_timer_handler, 0);
420 NEIGH_CACHE_STAT_INC(tbl, allocs);
422 refcount_set(&n->refcnt, 1);
424 INIT_LIST_HEAD(&n->gc_list);
426 atomic_inc(&tbl->entries);
432 atomic_dec(&tbl->gc_entries);
436 static void neigh_get_hash_rnd(u32 *x)
438 *x = get_random_u32() | 1;
441 static struct neigh_hash_table *neigh_hash_alloc(unsigned int shift)
443 size_t size = (1 << shift) * sizeof(struct neighbour *);
444 struct neigh_hash_table *ret;
445 struct neighbour __rcu **buckets;
448 ret = kmalloc(sizeof(*ret), GFP_ATOMIC);
451 if (size <= PAGE_SIZE) {
452 buckets = kzalloc(size, GFP_ATOMIC);
454 buckets = (struct neighbour __rcu **)
455 __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
457 kmemleak_alloc(buckets, size, 1, GFP_ATOMIC);
463 ret->hash_buckets = buckets;
464 ret->hash_shift = shift;
465 for (i = 0; i < NEIGH_NUM_HASH_RND; i++)
466 neigh_get_hash_rnd(&ret->hash_rnd[i]);
470 static void neigh_hash_free_rcu(struct rcu_head *head)
472 struct neigh_hash_table *nht = container_of(head,
473 struct neigh_hash_table,
475 size_t size = (1 << nht->hash_shift) * sizeof(struct neighbour *);
476 struct neighbour __rcu **buckets = nht->hash_buckets;
478 if (size <= PAGE_SIZE) {
481 kmemleak_free(buckets);
482 free_pages((unsigned long)buckets, get_order(size));
487 static struct neigh_hash_table *neigh_hash_grow(struct neigh_table *tbl,
488 unsigned long new_shift)
490 unsigned int i, hash;
491 struct neigh_hash_table *new_nht, *old_nht;
493 NEIGH_CACHE_STAT_INC(tbl, hash_grows);
495 old_nht = rcu_dereference_protected(tbl->nht,
496 lockdep_is_held(&tbl->lock));
497 new_nht = neigh_hash_alloc(new_shift);
501 for (i = 0; i < (1 << old_nht->hash_shift); i++) {
502 struct neighbour *n, *next;
504 for (n = rcu_dereference_protected(old_nht->hash_buckets[i],
505 lockdep_is_held(&tbl->lock));
508 hash = tbl->hash(n->primary_key, n->dev,
511 hash >>= (32 - new_nht->hash_shift);
512 next = rcu_dereference_protected(n->next,
513 lockdep_is_held(&tbl->lock));
515 rcu_assign_pointer(n->next,
516 rcu_dereference_protected(
517 new_nht->hash_buckets[hash],
518 lockdep_is_held(&tbl->lock)));
519 rcu_assign_pointer(new_nht->hash_buckets[hash], n);
523 rcu_assign_pointer(tbl->nht, new_nht);
524 call_rcu(&old_nht->rcu, neigh_hash_free_rcu);
528 struct neighbour *neigh_lookup(struct neigh_table *tbl, const void *pkey,
529 struct net_device *dev)
533 NEIGH_CACHE_STAT_INC(tbl, lookups);
536 n = __neigh_lookup_noref(tbl, pkey, dev);
538 if (!refcount_inc_not_zero(&n->refcnt))
540 NEIGH_CACHE_STAT_INC(tbl, hits);
543 rcu_read_unlock_bh();
546 EXPORT_SYMBOL(neigh_lookup);
548 struct neighbour *neigh_lookup_nodev(struct neigh_table *tbl, struct net *net,
552 unsigned int key_len = tbl->key_len;
554 struct neigh_hash_table *nht;
556 NEIGH_CACHE_STAT_INC(tbl, lookups);
559 nht = rcu_dereference_bh(tbl->nht);
560 hash_val = tbl->hash(pkey, NULL, nht->hash_rnd) >> (32 - nht->hash_shift);
562 for (n = rcu_dereference_bh(nht->hash_buckets[hash_val]);
564 n = rcu_dereference_bh(n->next)) {
565 if (!memcmp(n->primary_key, pkey, key_len) &&
566 net_eq(dev_net(n->dev), net)) {
567 if (!refcount_inc_not_zero(&n->refcnt))
569 NEIGH_CACHE_STAT_INC(tbl, hits);
574 rcu_read_unlock_bh();
577 EXPORT_SYMBOL(neigh_lookup_nodev);
579 static struct neighbour *
580 ___neigh_create(struct neigh_table *tbl, const void *pkey,
581 struct net_device *dev, u8 flags,
582 bool exempt_from_gc, bool want_ref)
584 u32 hash_val, key_len = tbl->key_len;
585 struct neighbour *n1, *rc, *n;
586 struct neigh_hash_table *nht;
589 n = neigh_alloc(tbl, dev, flags, exempt_from_gc);
590 trace_neigh_create(tbl, dev, pkey, n, exempt_from_gc);
592 rc = ERR_PTR(-ENOBUFS);
596 memcpy(n->primary_key, pkey, key_len);
600 /* Protocol specific setup. */
601 if (tbl->constructor && (error = tbl->constructor(n)) < 0) {
603 goto out_neigh_release;
606 if (dev->netdev_ops->ndo_neigh_construct) {
607 error = dev->netdev_ops->ndo_neigh_construct(dev, n);
610 goto out_neigh_release;
614 /* Device specific setup. */
615 if (n->parms->neigh_setup &&
616 (error = n->parms->neigh_setup(n)) < 0) {
618 goto out_neigh_release;
621 n->confirmed = jiffies - (NEIGH_VAR(n->parms, BASE_REACHABLE_TIME) << 1);
623 write_lock_bh(&tbl->lock);
624 nht = rcu_dereference_protected(tbl->nht,
625 lockdep_is_held(&tbl->lock));
627 if (atomic_read(&tbl->entries) > (1 << nht->hash_shift))
628 nht = neigh_hash_grow(tbl, nht->hash_shift + 1);
630 hash_val = tbl->hash(n->primary_key, dev, nht->hash_rnd) >> (32 - nht->hash_shift);
632 if (n->parms->dead) {
633 rc = ERR_PTR(-EINVAL);
637 for (n1 = rcu_dereference_protected(nht->hash_buckets[hash_val],
638 lockdep_is_held(&tbl->lock));
640 n1 = rcu_dereference_protected(n1->next,
641 lockdep_is_held(&tbl->lock))) {
642 if (dev == n1->dev && !memcmp(n1->primary_key, n->primary_key, key_len)) {
652 list_add_tail(&n->gc_list, &n->tbl->gc_list);
656 rcu_assign_pointer(n->next,
657 rcu_dereference_protected(nht->hash_buckets[hash_val],
658 lockdep_is_held(&tbl->lock)));
659 rcu_assign_pointer(nht->hash_buckets[hash_val], n);
660 write_unlock_bh(&tbl->lock);
661 neigh_dbg(2, "neigh %p is created\n", n);
666 write_unlock_bh(&tbl->lock);
669 atomic_dec(&tbl->gc_entries);
674 struct neighbour *__neigh_create(struct neigh_table *tbl, const void *pkey,
675 struct net_device *dev, bool want_ref)
677 return ___neigh_create(tbl, pkey, dev, 0, false, want_ref);
679 EXPORT_SYMBOL(__neigh_create);
681 static u32 pneigh_hash(const void *pkey, unsigned int key_len)
683 u32 hash_val = *(u32 *)(pkey + key_len - 4);
684 hash_val ^= (hash_val >> 16);
685 hash_val ^= hash_val >> 8;
686 hash_val ^= hash_val >> 4;
687 hash_val &= PNEIGH_HASHMASK;
691 static struct pneigh_entry *__pneigh_lookup_1(struct pneigh_entry *n,
694 unsigned int key_len,
695 struct net_device *dev)
698 if (!memcmp(n->key, pkey, key_len) &&
699 net_eq(pneigh_net(n), net) &&
700 (n->dev == dev || !n->dev))
707 struct pneigh_entry *__pneigh_lookup(struct neigh_table *tbl,
708 struct net *net, const void *pkey, struct net_device *dev)
710 unsigned int key_len = tbl->key_len;
711 u32 hash_val = pneigh_hash(pkey, key_len);
713 return __pneigh_lookup_1(tbl->phash_buckets[hash_val],
714 net, pkey, key_len, dev);
716 EXPORT_SYMBOL_GPL(__pneigh_lookup);
718 struct pneigh_entry * pneigh_lookup(struct neigh_table *tbl,
719 struct net *net, const void *pkey,
720 struct net_device *dev, int creat)
722 struct pneigh_entry *n;
723 unsigned int key_len = tbl->key_len;
724 u32 hash_val = pneigh_hash(pkey, key_len);
726 read_lock_bh(&tbl->lock);
727 n = __pneigh_lookup_1(tbl->phash_buckets[hash_val],
728 net, pkey, key_len, dev);
729 read_unlock_bh(&tbl->lock);
736 n = kmalloc(sizeof(*n) + key_len, GFP_KERNEL);
741 write_pnet(&n->net, net);
742 memcpy(n->key, pkey, key_len);
746 if (tbl->pconstructor && tbl->pconstructor(n)) {
753 write_lock_bh(&tbl->lock);
754 n->next = tbl->phash_buckets[hash_val];
755 tbl->phash_buckets[hash_val] = n;
756 write_unlock_bh(&tbl->lock);
760 EXPORT_SYMBOL(pneigh_lookup);
763 int pneigh_delete(struct neigh_table *tbl, struct net *net, const void *pkey,
764 struct net_device *dev)
766 struct pneigh_entry *n, **np;
767 unsigned int key_len = tbl->key_len;
768 u32 hash_val = pneigh_hash(pkey, key_len);
770 write_lock_bh(&tbl->lock);
771 for (np = &tbl->phash_buckets[hash_val]; (n = *np) != NULL;
773 if (!memcmp(n->key, pkey, key_len) && n->dev == dev &&
774 net_eq(pneigh_net(n), net)) {
776 write_unlock_bh(&tbl->lock);
777 if (tbl->pdestructor)
784 write_unlock_bh(&tbl->lock);
788 static int pneigh_ifdown_and_unlock(struct neigh_table *tbl,
789 struct net_device *dev)
791 struct pneigh_entry *n, **np, *freelist = NULL;
794 for (h = 0; h <= PNEIGH_HASHMASK; h++) {
795 np = &tbl->phash_buckets[h];
796 while ((n = *np) != NULL) {
797 if (!dev || n->dev == dev) {
806 write_unlock_bh(&tbl->lock);
807 while ((n = freelist)) {
810 if (tbl->pdestructor)
818 static void neigh_parms_destroy(struct neigh_parms *parms);
820 static inline void neigh_parms_put(struct neigh_parms *parms)
822 if (refcount_dec_and_test(&parms->refcnt))
823 neigh_parms_destroy(parms);
827 * neighbour must already be out of the table;
830 void neigh_destroy(struct neighbour *neigh)
832 struct net_device *dev = neigh->dev;
834 NEIGH_CACHE_STAT_INC(neigh->tbl, destroys);
837 pr_warn("Destroying alive neighbour %p\n", neigh);
842 if (neigh_del_timer(neigh))
843 pr_warn("Impossible event\n");
845 write_lock_bh(&neigh->lock);
846 __skb_queue_purge(&neigh->arp_queue);
847 write_unlock_bh(&neigh->lock);
848 neigh->arp_queue_len_bytes = 0;
850 if (dev->netdev_ops->ndo_neigh_destroy)
851 dev->netdev_ops->ndo_neigh_destroy(dev, neigh);
854 neigh_parms_put(neigh->parms);
856 neigh_dbg(2, "neigh %p is destroyed\n", neigh);
858 atomic_dec(&neigh->tbl->entries);
859 kfree_rcu(neigh, rcu);
861 EXPORT_SYMBOL(neigh_destroy);
863 /* Neighbour state is suspicious;
866 Called with write_locked neigh.
868 static void neigh_suspect(struct neighbour *neigh)
870 neigh_dbg(2, "neigh %p is suspected\n", neigh);
872 neigh->output = neigh->ops->output;
875 /* Neighbour state is OK;
878 Called with write_locked neigh.
880 static void neigh_connect(struct neighbour *neigh)
882 neigh_dbg(2, "neigh %p is connected\n", neigh);
884 neigh->output = neigh->ops->connected_output;
887 static void neigh_periodic_work(struct work_struct *work)
889 struct neigh_table *tbl = container_of(work, struct neigh_table, gc_work.work);
891 struct neighbour __rcu **np;
893 struct neigh_hash_table *nht;
895 NEIGH_CACHE_STAT_INC(tbl, periodic_gc_runs);
897 write_lock_bh(&tbl->lock);
898 nht = rcu_dereference_protected(tbl->nht,
899 lockdep_is_held(&tbl->lock));
902 * periodically recompute ReachableTime from random function
905 if (time_after(jiffies, tbl->last_rand + 300 * HZ)) {
906 struct neigh_parms *p;
907 tbl->last_rand = jiffies;
908 list_for_each_entry(p, &tbl->parms_list, list)
910 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
913 if (atomic_read(&tbl->entries) < tbl->gc_thresh1)
916 for (i = 0 ; i < (1 << nht->hash_shift); i++) {
917 np = &nht->hash_buckets[i];
919 while ((n = rcu_dereference_protected(*np,
920 lockdep_is_held(&tbl->lock))) != NULL) {
923 write_lock(&n->lock);
925 state = n->nud_state;
926 if ((state & (NUD_PERMANENT | NUD_IN_TIMER)) ||
927 (n->flags & NTF_EXT_LEARNED)) {
928 write_unlock(&n->lock);
932 if (time_before(n->used, n->confirmed))
933 n->used = n->confirmed;
935 if (refcount_read(&n->refcnt) == 1 &&
936 (state == NUD_FAILED ||
937 time_after(jiffies, n->used + NEIGH_VAR(n->parms, GC_STALETIME)))) {
940 write_unlock(&n->lock);
941 neigh_cleanup_and_release(n);
944 write_unlock(&n->lock);
950 * It's fine to release lock here, even if hash table
951 * grows while we are preempted.
953 write_unlock_bh(&tbl->lock);
955 write_lock_bh(&tbl->lock);
956 nht = rcu_dereference_protected(tbl->nht,
957 lockdep_is_held(&tbl->lock));
960 /* Cycle through all hash buckets every BASE_REACHABLE_TIME/2 ticks.
961 * ARP entry timeouts range from 1/2 BASE_REACHABLE_TIME to 3/2
962 * BASE_REACHABLE_TIME.
964 queue_delayed_work(system_power_efficient_wq, &tbl->gc_work,
965 NEIGH_VAR(&tbl->parms, BASE_REACHABLE_TIME) >> 1);
966 write_unlock_bh(&tbl->lock);
969 static __inline__ int neigh_max_probes(struct neighbour *n)
971 struct neigh_parms *p = n->parms;
972 return NEIGH_VAR(p, UCAST_PROBES) + NEIGH_VAR(p, APP_PROBES) +
973 (n->nud_state & NUD_PROBE ? NEIGH_VAR(p, MCAST_REPROBES) :
974 NEIGH_VAR(p, MCAST_PROBES));
977 static void neigh_invalidate(struct neighbour *neigh)
978 __releases(neigh->lock)
979 __acquires(neigh->lock)
983 NEIGH_CACHE_STAT_INC(neigh->tbl, res_failed);
984 neigh_dbg(2, "neigh %p is failed\n", neigh);
985 neigh->updated = jiffies;
987 /* It is very thin place. report_unreachable is very complicated
988 routine. Particularly, it can hit the same neighbour entry!
990 So that, we try to be accurate and avoid dead loop. --ANK
992 while (neigh->nud_state == NUD_FAILED &&
993 (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
994 write_unlock(&neigh->lock);
995 neigh->ops->error_report(neigh, skb);
996 write_lock(&neigh->lock);
998 __skb_queue_purge(&neigh->arp_queue);
999 neigh->arp_queue_len_bytes = 0;
1002 static void neigh_probe(struct neighbour *neigh)
1003 __releases(neigh->lock)
1005 struct sk_buff *skb = skb_peek_tail(&neigh->arp_queue);
1006 /* keep skb alive even if arp_queue overflows */
1008 skb = skb_clone(skb, GFP_ATOMIC);
1009 write_unlock(&neigh->lock);
1010 if (neigh->ops->solicit)
1011 neigh->ops->solicit(neigh, skb);
1012 atomic_inc(&neigh->probes);
1016 /* Called when a timer expires for a neighbour entry. */
1018 static void neigh_timer_handler(struct timer_list *t)
1020 unsigned long now, next;
1021 struct neighbour *neigh = from_timer(neigh, t, timer);
1025 write_lock(&neigh->lock);
1027 state = neigh->nud_state;
1031 if (!(state & NUD_IN_TIMER))
1034 if (state & NUD_REACHABLE) {
1035 if (time_before_eq(now,
1036 neigh->confirmed + neigh->parms->reachable_time)) {
1037 neigh_dbg(2, "neigh %p is still alive\n", neigh);
1038 next = neigh->confirmed + neigh->parms->reachable_time;
1039 } else if (time_before_eq(now,
1041 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME))) {
1042 neigh_dbg(2, "neigh %p is delayed\n", neigh);
1043 neigh->nud_state = NUD_DELAY;
1044 neigh->updated = jiffies;
1045 neigh_suspect(neigh);
1046 next = now + NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME);
1048 neigh_dbg(2, "neigh %p is suspected\n", neigh);
1049 neigh->nud_state = NUD_STALE;
1050 neigh->updated = jiffies;
1051 neigh_suspect(neigh);
1054 } else if (state & NUD_DELAY) {
1055 if (time_before_eq(now,
1057 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME))) {
1058 neigh_dbg(2, "neigh %p is now reachable\n", neigh);
1059 neigh->nud_state = NUD_REACHABLE;
1060 neigh->updated = jiffies;
1061 neigh_connect(neigh);
1063 next = neigh->confirmed + neigh->parms->reachable_time;
1065 neigh_dbg(2, "neigh %p is probed\n", neigh);
1066 neigh->nud_state = NUD_PROBE;
1067 neigh->updated = jiffies;
1068 atomic_set(&neigh->probes, 0);
1070 next = now + max(NEIGH_VAR(neigh->parms, RETRANS_TIME),
1074 /* NUD_PROBE|NUD_INCOMPLETE */
1075 next = now + max(NEIGH_VAR(neigh->parms, RETRANS_TIME), HZ/100);
1078 if ((neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) &&
1079 atomic_read(&neigh->probes) >= neigh_max_probes(neigh)) {
1080 neigh->nud_state = NUD_FAILED;
1082 neigh_invalidate(neigh);
1086 if (neigh->nud_state & NUD_IN_TIMER) {
1087 if (time_before(next, jiffies + HZ/100))
1088 next = jiffies + HZ/100;
1089 if (!mod_timer(&neigh->timer, next))
1092 if (neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) {
1096 write_unlock(&neigh->lock);
1100 neigh_update_notify(neigh, 0);
1102 trace_neigh_timer_handler(neigh, 0);
1104 neigh_release(neigh);
1107 int __neigh_event_send(struct neighbour *neigh, struct sk_buff *skb)
1110 bool immediate_probe = false;
1112 write_lock_bh(&neigh->lock);
1115 if (neigh->nud_state & (NUD_CONNECTED | NUD_DELAY | NUD_PROBE))
1120 if (!(neigh->nud_state & (NUD_STALE | NUD_INCOMPLETE))) {
1121 if (NEIGH_VAR(neigh->parms, MCAST_PROBES) +
1122 NEIGH_VAR(neigh->parms, APP_PROBES)) {
1123 unsigned long next, now = jiffies;
1125 atomic_set(&neigh->probes,
1126 NEIGH_VAR(neigh->parms, UCAST_PROBES));
1127 neigh_del_timer(neigh);
1128 neigh->nud_state = NUD_INCOMPLETE;
1129 neigh->updated = now;
1130 next = now + max(NEIGH_VAR(neigh->parms, RETRANS_TIME),
1132 neigh_add_timer(neigh, next);
1133 immediate_probe = true;
1135 neigh->nud_state = NUD_FAILED;
1136 neigh->updated = jiffies;
1137 write_unlock_bh(&neigh->lock);
1142 } else if (neigh->nud_state & NUD_STALE) {
1143 neigh_dbg(2, "neigh %p is delayed\n", neigh);
1144 neigh_del_timer(neigh);
1145 neigh->nud_state = NUD_DELAY;
1146 neigh->updated = jiffies;
1147 neigh_add_timer(neigh, jiffies +
1148 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME));
1151 if (neigh->nud_state == NUD_INCOMPLETE) {
1153 while (neigh->arp_queue_len_bytes + skb->truesize >
1154 NEIGH_VAR(neigh->parms, QUEUE_LEN_BYTES)) {
1155 struct sk_buff *buff;
1157 buff = __skb_dequeue(&neigh->arp_queue);
1160 neigh->arp_queue_len_bytes -= buff->truesize;
1162 NEIGH_CACHE_STAT_INC(neigh->tbl, unres_discards);
1165 __skb_queue_tail(&neigh->arp_queue, skb);
1166 neigh->arp_queue_len_bytes += skb->truesize;
1171 if (immediate_probe)
1174 write_unlock(&neigh->lock);
1176 trace_neigh_event_send_done(neigh, rc);
1180 if (neigh->nud_state & NUD_STALE)
1182 write_unlock_bh(&neigh->lock);
1184 trace_neigh_event_send_dead(neigh, 1);
1187 EXPORT_SYMBOL(__neigh_event_send);
1189 static void neigh_update_hhs(struct neighbour *neigh)
1191 struct hh_cache *hh;
1192 void (*update)(struct hh_cache*, const struct net_device*, const unsigned char *)
1195 if (neigh->dev->header_ops)
1196 update = neigh->dev->header_ops->cache_update;
1200 if (READ_ONCE(hh->hh_len)) {
1201 write_seqlock_bh(&hh->hh_lock);
1202 update(hh, neigh->dev, neigh->ha);
1203 write_sequnlock_bh(&hh->hh_lock);
1210 /* Generic update routine.
1211 -- lladdr is new lladdr or NULL, if it is not supplied.
1212 -- new is new state.
1214 NEIGH_UPDATE_F_OVERRIDE allows to override existing lladdr,
1216 NEIGH_UPDATE_F_WEAK_OVERRIDE will suspect existing "connected"
1217 lladdr instead of overriding it
1219 NEIGH_UPDATE_F_ADMIN means that the change is administrative.
1221 NEIGH_UPDATE_F_OVERRIDE_ISROUTER allows to override existing
1223 NEIGH_UPDATE_F_ISROUTER indicates if the neighbour is known as
1226 Caller MUST hold reference count on the entry.
1229 static int __neigh_update(struct neighbour *neigh, const u8 *lladdr,
1230 u8 new, u32 flags, u32 nlmsg_pid,
1231 struct netlink_ext_ack *extack)
1233 bool ext_learn_change = false;
1237 struct net_device *dev;
1238 int update_isrouter = 0;
1240 trace_neigh_update(neigh, lladdr, new, flags, nlmsg_pid);
1242 write_lock_bh(&neigh->lock);
1245 old = neigh->nud_state;
1249 NL_SET_ERR_MSG(extack, "Neighbor entry is now dead");
1253 if (!(flags & NEIGH_UPDATE_F_ADMIN) &&
1254 (old & (NUD_NOARP | NUD_PERMANENT)))
1257 ext_learn_change = neigh_update_ext_learned(neigh, flags, ¬ify);
1259 if (!(new & NUD_VALID)) {
1260 neigh_del_timer(neigh);
1261 if (old & NUD_CONNECTED)
1262 neigh_suspect(neigh);
1263 neigh->nud_state = new;
1265 notify = old & NUD_VALID;
1266 if ((old & (NUD_INCOMPLETE | NUD_PROBE)) &&
1267 (new & NUD_FAILED)) {
1268 neigh_invalidate(neigh);
1274 /* Compare new lladdr with cached one */
1275 if (!dev->addr_len) {
1276 /* First case: device needs no address. */
1278 } else if (lladdr) {
1279 /* The second case: if something is already cached
1280 and a new address is proposed:
1282 - if they are different, check override flag
1284 if ((old & NUD_VALID) &&
1285 !memcmp(lladdr, neigh->ha, dev->addr_len))
1288 /* No address is supplied; if we know something,
1289 use it, otherwise discard the request.
1292 if (!(old & NUD_VALID)) {
1293 NL_SET_ERR_MSG(extack, "No link layer address given");
1299 /* Update confirmed timestamp for neighbour entry after we
1300 * received ARP packet even if it doesn't change IP to MAC binding.
1302 if (new & NUD_CONNECTED)
1303 neigh->confirmed = jiffies;
1305 /* If entry was valid and address is not changed,
1306 do not change entry state, if new one is STALE.
1309 update_isrouter = flags & NEIGH_UPDATE_F_OVERRIDE_ISROUTER;
1310 if (old & NUD_VALID) {
1311 if (lladdr != neigh->ha && !(flags & NEIGH_UPDATE_F_OVERRIDE)) {
1312 update_isrouter = 0;
1313 if ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) &&
1314 (old & NUD_CONNECTED)) {
1320 if (lladdr == neigh->ha && new == NUD_STALE &&
1321 !(flags & NEIGH_UPDATE_F_ADMIN))
1326 /* Update timestamp only once we know we will make a change to the
1327 * neighbour entry. Otherwise we risk to move the locktime window with
1328 * noop updates and ignore relevant ARP updates.
1330 if (new != old || lladdr != neigh->ha)
1331 neigh->updated = jiffies;
1334 neigh_del_timer(neigh);
1335 if (new & NUD_PROBE)
1336 atomic_set(&neigh->probes, 0);
1337 if (new & NUD_IN_TIMER)
1338 neigh_add_timer(neigh, (jiffies +
1339 ((new & NUD_REACHABLE) ?
1340 neigh->parms->reachable_time :
1342 neigh->nud_state = new;
1346 if (lladdr != neigh->ha) {
1347 write_seqlock(&neigh->ha_lock);
1348 memcpy(&neigh->ha, lladdr, dev->addr_len);
1349 write_sequnlock(&neigh->ha_lock);
1350 neigh_update_hhs(neigh);
1351 if (!(new & NUD_CONNECTED))
1352 neigh->confirmed = jiffies -
1353 (NEIGH_VAR(neigh->parms, BASE_REACHABLE_TIME) << 1);
1358 if (new & NUD_CONNECTED)
1359 neigh_connect(neigh);
1361 neigh_suspect(neigh);
1362 if (!(old & NUD_VALID)) {
1363 struct sk_buff *skb;
1365 /* Again: avoid dead loop if something went wrong */
1367 while (neigh->nud_state & NUD_VALID &&
1368 (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
1369 struct dst_entry *dst = skb_dst(skb);
1370 struct neighbour *n2, *n1 = neigh;
1371 write_unlock_bh(&neigh->lock);
1375 /* Why not just use 'neigh' as-is? The problem is that
1376 * things such as shaper, eql, and sch_teql can end up
1377 * using alternative, different, neigh objects to output
1378 * the packet in the output path. So what we need to do
1379 * here is re-lookup the top-level neigh in the path so
1380 * we can reinject the packet there.
1383 if (dst && dst->obsolete != DST_OBSOLETE_DEAD) {
1384 n2 = dst_neigh_lookup_skb(dst, skb);
1388 n1->output(n1, skb);
1393 write_lock_bh(&neigh->lock);
1395 __skb_queue_purge(&neigh->arp_queue);
1396 neigh->arp_queue_len_bytes = 0;
1399 if (update_isrouter)
1400 neigh_update_is_router(neigh, flags, ¬ify);
1401 write_unlock_bh(&neigh->lock);
1403 if (((new ^ old) & NUD_PERMANENT) || ext_learn_change)
1404 neigh_update_gc_list(neigh);
1407 neigh_update_notify(neigh, nlmsg_pid);
1409 trace_neigh_update_done(neigh, err);
1414 int neigh_update(struct neighbour *neigh, const u8 *lladdr, u8 new,
1415 u32 flags, u32 nlmsg_pid)
1417 return __neigh_update(neigh, lladdr, new, flags, nlmsg_pid, NULL);
1419 EXPORT_SYMBOL(neigh_update);
1421 /* Update the neigh to listen temporarily for probe responses, even if it is
1422 * in a NUD_FAILED state. The caller has to hold neigh->lock for writing.
1424 void __neigh_set_probe_once(struct neighbour *neigh)
1428 neigh->updated = jiffies;
1429 if (!(neigh->nud_state & NUD_FAILED))
1431 neigh->nud_state = NUD_INCOMPLETE;
1432 atomic_set(&neigh->probes, neigh_max_probes(neigh));
1433 neigh_add_timer(neigh,
1434 jiffies + max(NEIGH_VAR(neigh->parms, RETRANS_TIME),
1437 EXPORT_SYMBOL(__neigh_set_probe_once);
1439 struct neighbour *neigh_event_ns(struct neigh_table *tbl,
1440 u8 *lladdr, void *saddr,
1441 struct net_device *dev)
1443 struct neighbour *neigh = __neigh_lookup(tbl, saddr, dev,
1444 lladdr || !dev->addr_len);
1446 neigh_update(neigh, lladdr, NUD_STALE,
1447 NEIGH_UPDATE_F_OVERRIDE, 0);
1450 EXPORT_SYMBOL(neigh_event_ns);
1452 /* called with read_lock_bh(&n->lock); */
1453 static void neigh_hh_init(struct neighbour *n)
1455 struct net_device *dev = n->dev;
1456 __be16 prot = n->tbl->protocol;
1457 struct hh_cache *hh = &n->hh;
1459 write_lock_bh(&n->lock);
1461 /* Only one thread can come in here and initialize the
1465 dev->header_ops->cache(n, hh, prot);
1467 write_unlock_bh(&n->lock);
1470 /* Slow and careful. */
1472 int neigh_resolve_output(struct neighbour *neigh, struct sk_buff *skb)
1476 if (!neigh_event_send(neigh, skb)) {
1478 struct net_device *dev = neigh->dev;
1481 if (dev->header_ops->cache && !READ_ONCE(neigh->hh.hh_len))
1482 neigh_hh_init(neigh);
1485 __skb_pull(skb, skb_network_offset(skb));
1486 seq = read_seqbegin(&neigh->ha_lock);
1487 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1488 neigh->ha, NULL, skb->len);
1489 } while (read_seqretry(&neigh->ha_lock, seq));
1492 rc = dev_queue_xmit(skb);
1503 EXPORT_SYMBOL(neigh_resolve_output);
1505 /* As fast as possible without hh cache */
1507 int neigh_connected_output(struct neighbour *neigh, struct sk_buff *skb)
1509 struct net_device *dev = neigh->dev;
1514 __skb_pull(skb, skb_network_offset(skb));
1515 seq = read_seqbegin(&neigh->ha_lock);
1516 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1517 neigh->ha, NULL, skb->len);
1518 } while (read_seqretry(&neigh->ha_lock, seq));
1521 err = dev_queue_xmit(skb);
1528 EXPORT_SYMBOL(neigh_connected_output);
1530 int neigh_direct_output(struct neighbour *neigh, struct sk_buff *skb)
1532 return dev_queue_xmit(skb);
1534 EXPORT_SYMBOL(neigh_direct_output);
1536 static void neigh_proxy_process(struct timer_list *t)
1538 struct neigh_table *tbl = from_timer(tbl, t, proxy_timer);
1539 long sched_next = 0;
1540 unsigned long now = jiffies;
1541 struct sk_buff *skb, *n;
1543 spin_lock(&tbl->proxy_queue.lock);
1545 skb_queue_walk_safe(&tbl->proxy_queue, skb, n) {
1546 long tdif = NEIGH_CB(skb)->sched_next - now;
1549 struct net_device *dev = skb->dev;
1551 __skb_unlink(skb, &tbl->proxy_queue);
1552 if (tbl->proxy_redo && netif_running(dev)) {
1554 tbl->proxy_redo(skb);
1561 } else if (!sched_next || tdif < sched_next)
1564 del_timer(&tbl->proxy_timer);
1566 mod_timer(&tbl->proxy_timer, jiffies + sched_next);
1567 spin_unlock(&tbl->proxy_queue.lock);
1570 void pneigh_enqueue(struct neigh_table *tbl, struct neigh_parms *p,
1571 struct sk_buff *skb)
1573 unsigned long sched_next = jiffies +
1574 prandom_u32_max(NEIGH_VAR(p, PROXY_DELAY));
1576 if (tbl->proxy_queue.qlen > NEIGH_VAR(p, PROXY_QLEN)) {
1581 NEIGH_CB(skb)->sched_next = sched_next;
1582 NEIGH_CB(skb)->flags |= LOCALLY_ENQUEUED;
1584 spin_lock(&tbl->proxy_queue.lock);
1585 if (del_timer(&tbl->proxy_timer)) {
1586 if (time_before(tbl->proxy_timer.expires, sched_next))
1587 sched_next = tbl->proxy_timer.expires;
1591 __skb_queue_tail(&tbl->proxy_queue, skb);
1592 mod_timer(&tbl->proxy_timer, sched_next);
1593 spin_unlock(&tbl->proxy_queue.lock);
1595 EXPORT_SYMBOL(pneigh_enqueue);
1597 static inline struct neigh_parms *lookup_neigh_parms(struct neigh_table *tbl,
1598 struct net *net, int ifindex)
1600 struct neigh_parms *p;
1602 list_for_each_entry(p, &tbl->parms_list, list) {
1603 if ((p->dev && p->dev->ifindex == ifindex && net_eq(neigh_parms_net(p), net)) ||
1604 (!p->dev && !ifindex && net_eq(net, &init_net)))
1611 struct neigh_parms *neigh_parms_alloc(struct net_device *dev,
1612 struct neigh_table *tbl)
1614 struct neigh_parms *p;
1615 struct net *net = dev_net(dev);
1616 const struct net_device_ops *ops = dev->netdev_ops;
1618 p = kmemdup(&tbl->parms, sizeof(*p), GFP_KERNEL);
1621 refcount_set(&p->refcnt, 1);
1623 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
1626 write_pnet(&p->net, net);
1627 p->sysctl_table = NULL;
1629 if (ops->ndo_neigh_setup && ops->ndo_neigh_setup(dev, p)) {
1635 write_lock_bh(&tbl->lock);
1636 list_add(&p->list, &tbl->parms.list);
1637 write_unlock_bh(&tbl->lock);
1639 neigh_parms_data_state_cleanall(p);
1643 EXPORT_SYMBOL(neigh_parms_alloc);
1645 static void neigh_rcu_free_parms(struct rcu_head *head)
1647 struct neigh_parms *parms =
1648 container_of(head, struct neigh_parms, rcu_head);
1650 neigh_parms_put(parms);
1653 void neigh_parms_release(struct neigh_table *tbl, struct neigh_parms *parms)
1655 if (!parms || parms == &tbl->parms)
1657 write_lock_bh(&tbl->lock);
1658 list_del(&parms->list);
1660 write_unlock_bh(&tbl->lock);
1661 dev_put(parms->dev);
1662 call_rcu(&parms->rcu_head, neigh_rcu_free_parms);
1664 EXPORT_SYMBOL(neigh_parms_release);
1666 static void neigh_parms_destroy(struct neigh_parms *parms)
1671 static struct lock_class_key neigh_table_proxy_queue_class;
1673 static struct neigh_table *neigh_tables[NEIGH_NR_TABLES] __read_mostly;
1675 void neigh_table_init(int index, struct neigh_table *tbl)
1677 unsigned long now = jiffies;
1678 unsigned long phsize;
1680 INIT_LIST_HEAD(&tbl->parms_list);
1681 INIT_LIST_HEAD(&tbl->gc_list);
1682 list_add(&tbl->parms.list, &tbl->parms_list);
1683 write_pnet(&tbl->parms.net, &init_net);
1684 refcount_set(&tbl->parms.refcnt, 1);
1685 tbl->parms.reachable_time =
1686 neigh_rand_reach_time(NEIGH_VAR(&tbl->parms, BASE_REACHABLE_TIME));
1688 tbl->stats = alloc_percpu(struct neigh_statistics);
1690 panic("cannot create neighbour cache statistics");
1692 #ifdef CONFIG_PROC_FS
1693 if (!proc_create_seq_data(tbl->id, 0, init_net.proc_net_stat,
1694 &neigh_stat_seq_ops, tbl))
1695 panic("cannot create neighbour proc dir entry");
1698 RCU_INIT_POINTER(tbl->nht, neigh_hash_alloc(3));
1700 phsize = (PNEIGH_HASHMASK + 1) * sizeof(struct pneigh_entry *);
1701 tbl->phash_buckets = kzalloc(phsize, GFP_KERNEL);
1703 if (!tbl->nht || !tbl->phash_buckets)
1704 panic("cannot allocate neighbour cache hashes");
1706 if (!tbl->entry_size)
1707 tbl->entry_size = ALIGN(offsetof(struct neighbour, primary_key) +
1708 tbl->key_len, NEIGH_PRIV_ALIGN);
1710 WARN_ON(tbl->entry_size % NEIGH_PRIV_ALIGN);
1712 rwlock_init(&tbl->lock);
1713 INIT_DEFERRABLE_WORK(&tbl->gc_work, neigh_periodic_work);
1714 queue_delayed_work(system_power_efficient_wq, &tbl->gc_work,
1715 tbl->parms.reachable_time);
1716 timer_setup(&tbl->proxy_timer, neigh_proxy_process, 0);
1717 skb_queue_head_init_class(&tbl->proxy_queue,
1718 &neigh_table_proxy_queue_class);
1720 tbl->last_flush = now;
1721 tbl->last_rand = now + tbl->parms.reachable_time * 20;
1723 neigh_tables[index] = tbl;
1725 EXPORT_SYMBOL(neigh_table_init);
1727 int neigh_table_clear(int index, struct neigh_table *tbl)
1729 neigh_tables[index] = NULL;
1730 /* It is not clean... Fix it to unload IPv6 module safely */
1731 cancel_delayed_work_sync(&tbl->gc_work);
1732 del_timer_sync(&tbl->proxy_timer);
1733 pneigh_queue_purge(&tbl->proxy_queue);
1734 neigh_ifdown(tbl, NULL);
1735 if (atomic_read(&tbl->entries))
1736 pr_crit("neighbour leakage\n");
1738 call_rcu(&rcu_dereference_protected(tbl->nht, 1)->rcu,
1739 neigh_hash_free_rcu);
1742 kfree(tbl->phash_buckets);
1743 tbl->phash_buckets = NULL;
1745 remove_proc_entry(tbl->id, init_net.proc_net_stat);
1747 free_percpu(tbl->stats);
1752 EXPORT_SYMBOL(neigh_table_clear);
1754 static struct neigh_table *neigh_find_table(int family)
1756 struct neigh_table *tbl = NULL;
1760 tbl = neigh_tables[NEIGH_ARP_TABLE];
1763 tbl = neigh_tables[NEIGH_ND_TABLE];
1766 tbl = neigh_tables[NEIGH_DN_TABLE];
1773 const struct nla_policy nda_policy[NDA_MAX+1] = {
1774 [NDA_UNSPEC] = { .strict_start_type = NDA_NH_ID },
1775 [NDA_DST] = { .type = NLA_BINARY, .len = MAX_ADDR_LEN },
1776 [NDA_LLADDR] = { .type = NLA_BINARY, .len = MAX_ADDR_LEN },
1777 [NDA_CACHEINFO] = { .len = sizeof(struct nda_cacheinfo) },
1778 [NDA_PROBES] = { .type = NLA_U32 },
1779 [NDA_VLAN] = { .type = NLA_U16 },
1780 [NDA_PORT] = { .type = NLA_U16 },
1781 [NDA_VNI] = { .type = NLA_U32 },
1782 [NDA_IFINDEX] = { .type = NLA_U32 },
1783 [NDA_MASTER] = { .type = NLA_U32 },
1784 [NDA_PROTOCOL] = { .type = NLA_U8 },
1785 [NDA_NH_ID] = { .type = NLA_U32 },
1786 [NDA_FDB_EXT_ATTRS] = { .type = NLA_NESTED },
1789 static int neigh_delete(struct sk_buff *skb, struct nlmsghdr *nlh,
1790 struct netlink_ext_ack *extack)
1792 struct net *net = sock_net(skb->sk);
1794 struct nlattr *dst_attr;
1795 struct neigh_table *tbl;
1796 struct neighbour *neigh;
1797 struct net_device *dev = NULL;
1801 if (nlmsg_len(nlh) < sizeof(*ndm))
1804 dst_attr = nlmsg_find_attr(nlh, sizeof(*ndm), NDA_DST);
1806 NL_SET_ERR_MSG(extack, "Network address not specified");
1810 ndm = nlmsg_data(nlh);
1811 if (ndm->ndm_ifindex) {
1812 dev = __dev_get_by_index(net, ndm->ndm_ifindex);
1819 tbl = neigh_find_table(ndm->ndm_family);
1821 return -EAFNOSUPPORT;
1823 if (nla_len(dst_attr) < (int)tbl->key_len) {
1824 NL_SET_ERR_MSG(extack, "Invalid network address");
1828 if (ndm->ndm_flags & NTF_PROXY) {
1829 err = pneigh_delete(tbl, net, nla_data(dst_attr), dev);
1836 neigh = neigh_lookup(tbl, nla_data(dst_attr), dev);
1837 if (neigh == NULL) {
1842 err = __neigh_update(neigh, NULL, NUD_FAILED,
1843 NEIGH_UPDATE_F_OVERRIDE | NEIGH_UPDATE_F_ADMIN,
1844 NETLINK_CB(skb).portid, extack);
1845 write_lock_bh(&tbl->lock);
1846 neigh_release(neigh);
1847 neigh_remove_one(neigh, tbl);
1848 write_unlock_bh(&tbl->lock);
1854 static int neigh_add(struct sk_buff *skb, struct nlmsghdr *nlh,
1855 struct netlink_ext_ack *extack)
1857 int flags = NEIGH_UPDATE_F_ADMIN | NEIGH_UPDATE_F_OVERRIDE |
1858 NEIGH_UPDATE_F_OVERRIDE_ISROUTER;
1859 struct net *net = sock_net(skb->sk);
1861 struct nlattr *tb[NDA_MAX+1];
1862 struct neigh_table *tbl;
1863 struct net_device *dev = NULL;
1864 struct neighbour *neigh;
1870 err = nlmsg_parse_deprecated(nlh, sizeof(*ndm), tb, NDA_MAX,
1871 nda_policy, extack);
1877 NL_SET_ERR_MSG(extack, "Network address not specified");
1881 ndm = nlmsg_data(nlh);
1882 if (ndm->ndm_ifindex) {
1883 dev = __dev_get_by_index(net, ndm->ndm_ifindex);
1889 if (tb[NDA_LLADDR] && nla_len(tb[NDA_LLADDR]) < dev->addr_len) {
1890 NL_SET_ERR_MSG(extack, "Invalid link address");
1895 tbl = neigh_find_table(ndm->ndm_family);
1897 return -EAFNOSUPPORT;
1899 if (nla_len(tb[NDA_DST]) < (int)tbl->key_len) {
1900 NL_SET_ERR_MSG(extack, "Invalid network address");
1904 dst = nla_data(tb[NDA_DST]);
1905 lladdr = tb[NDA_LLADDR] ? nla_data(tb[NDA_LLADDR]) : NULL;
1907 if (tb[NDA_PROTOCOL])
1908 protocol = nla_get_u8(tb[NDA_PROTOCOL]);
1910 if (ndm->ndm_flags & NTF_PROXY) {
1911 struct pneigh_entry *pn;
1914 pn = pneigh_lookup(tbl, net, dst, dev, 1);
1916 pn->flags = ndm->ndm_flags;
1918 pn->protocol = protocol;
1925 NL_SET_ERR_MSG(extack, "Device not specified");
1929 if (tbl->allow_add && !tbl->allow_add(dev, extack)) {
1934 neigh = neigh_lookup(tbl, dst, dev);
1935 if (neigh == NULL) {
1936 bool exempt_from_gc;
1938 if (!(nlh->nlmsg_flags & NLM_F_CREATE)) {
1943 exempt_from_gc = ndm->ndm_state & NUD_PERMANENT ||
1944 ndm->ndm_flags & NTF_EXT_LEARNED;
1945 neigh = ___neigh_create(tbl, dst, dev,
1946 ndm->ndm_flags & NTF_EXT_LEARNED,
1947 exempt_from_gc, true);
1948 if (IS_ERR(neigh)) {
1949 err = PTR_ERR(neigh);
1953 if (nlh->nlmsg_flags & NLM_F_EXCL) {
1955 neigh_release(neigh);
1959 if (!(nlh->nlmsg_flags & NLM_F_REPLACE))
1960 flags &= ~(NEIGH_UPDATE_F_OVERRIDE |
1961 NEIGH_UPDATE_F_OVERRIDE_ISROUTER);
1965 neigh->protocol = protocol;
1967 if (ndm->ndm_flags & NTF_EXT_LEARNED)
1968 flags |= NEIGH_UPDATE_F_EXT_LEARNED;
1970 if (ndm->ndm_flags & NTF_ROUTER)
1971 flags |= NEIGH_UPDATE_F_ISROUTER;
1973 if (ndm->ndm_flags & NTF_USE) {
1974 neigh_event_send(neigh, NULL);
1977 err = __neigh_update(neigh, lladdr, ndm->ndm_state, flags,
1978 NETLINK_CB(skb).portid, extack);
1980 neigh_release(neigh);
1986 static int neightbl_fill_parms(struct sk_buff *skb, struct neigh_parms *parms)
1988 struct nlattr *nest;
1990 nest = nla_nest_start_noflag(skb, NDTA_PARMS);
1995 nla_put_u32(skb, NDTPA_IFINDEX, parms->dev->ifindex)) ||
1996 nla_put_u32(skb, NDTPA_REFCNT, refcount_read(&parms->refcnt)) ||
1997 nla_put_u32(skb, NDTPA_QUEUE_LENBYTES,
1998 NEIGH_VAR(parms, QUEUE_LEN_BYTES)) ||
1999 /* approximative value for deprecated QUEUE_LEN (in packets) */
2000 nla_put_u32(skb, NDTPA_QUEUE_LEN,
2001 NEIGH_VAR(parms, QUEUE_LEN_BYTES) / SKB_TRUESIZE(ETH_FRAME_LEN)) ||
2002 nla_put_u32(skb, NDTPA_PROXY_QLEN, NEIGH_VAR(parms, PROXY_QLEN)) ||
2003 nla_put_u32(skb, NDTPA_APP_PROBES, NEIGH_VAR(parms, APP_PROBES)) ||
2004 nla_put_u32(skb, NDTPA_UCAST_PROBES,
2005 NEIGH_VAR(parms, UCAST_PROBES)) ||
2006 nla_put_u32(skb, NDTPA_MCAST_PROBES,
2007 NEIGH_VAR(parms, MCAST_PROBES)) ||
2008 nla_put_u32(skb, NDTPA_MCAST_REPROBES,
2009 NEIGH_VAR(parms, MCAST_REPROBES)) ||
2010 nla_put_msecs(skb, NDTPA_REACHABLE_TIME, parms->reachable_time,
2012 nla_put_msecs(skb, NDTPA_BASE_REACHABLE_TIME,
2013 NEIGH_VAR(parms, BASE_REACHABLE_TIME), NDTPA_PAD) ||
2014 nla_put_msecs(skb, NDTPA_GC_STALETIME,
2015 NEIGH_VAR(parms, GC_STALETIME), NDTPA_PAD) ||
2016 nla_put_msecs(skb, NDTPA_DELAY_PROBE_TIME,
2017 NEIGH_VAR(parms, DELAY_PROBE_TIME), NDTPA_PAD) ||
2018 nla_put_msecs(skb, NDTPA_RETRANS_TIME,
2019 NEIGH_VAR(parms, RETRANS_TIME), NDTPA_PAD) ||
2020 nla_put_msecs(skb, NDTPA_ANYCAST_DELAY,
2021 NEIGH_VAR(parms, ANYCAST_DELAY), NDTPA_PAD) ||
2022 nla_put_msecs(skb, NDTPA_PROXY_DELAY,
2023 NEIGH_VAR(parms, PROXY_DELAY), NDTPA_PAD) ||
2024 nla_put_msecs(skb, NDTPA_LOCKTIME,
2025 NEIGH_VAR(parms, LOCKTIME), NDTPA_PAD))
2026 goto nla_put_failure;
2027 return nla_nest_end(skb, nest);
2030 nla_nest_cancel(skb, nest);
2034 static int neightbl_fill_info(struct sk_buff *skb, struct neigh_table *tbl,
2035 u32 pid, u32 seq, int type, int flags)
2037 struct nlmsghdr *nlh;
2038 struct ndtmsg *ndtmsg;
2040 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
2044 ndtmsg = nlmsg_data(nlh);
2046 read_lock_bh(&tbl->lock);
2047 ndtmsg->ndtm_family = tbl->family;
2048 ndtmsg->ndtm_pad1 = 0;
2049 ndtmsg->ndtm_pad2 = 0;
2051 if (nla_put_string(skb, NDTA_NAME, tbl->id) ||
2052 nla_put_msecs(skb, NDTA_GC_INTERVAL, tbl->gc_interval, NDTA_PAD) ||
2053 nla_put_u32(skb, NDTA_THRESH1, tbl->gc_thresh1) ||
2054 nla_put_u32(skb, NDTA_THRESH2, tbl->gc_thresh2) ||
2055 nla_put_u32(skb, NDTA_THRESH3, tbl->gc_thresh3))
2056 goto nla_put_failure;
2058 unsigned long now = jiffies;
2059 long flush_delta = now - tbl->last_flush;
2060 long rand_delta = now - tbl->last_rand;
2061 struct neigh_hash_table *nht;
2062 struct ndt_config ndc = {
2063 .ndtc_key_len = tbl->key_len,
2064 .ndtc_entry_size = tbl->entry_size,
2065 .ndtc_entries = atomic_read(&tbl->entries),
2066 .ndtc_last_flush = jiffies_to_msecs(flush_delta),
2067 .ndtc_last_rand = jiffies_to_msecs(rand_delta),
2068 .ndtc_proxy_qlen = tbl->proxy_queue.qlen,
2072 nht = rcu_dereference_bh(tbl->nht);
2073 ndc.ndtc_hash_rnd = nht->hash_rnd[0];
2074 ndc.ndtc_hash_mask = ((1 << nht->hash_shift) - 1);
2075 rcu_read_unlock_bh();
2077 if (nla_put(skb, NDTA_CONFIG, sizeof(ndc), &ndc))
2078 goto nla_put_failure;
2083 struct ndt_stats ndst;
2085 memset(&ndst, 0, sizeof(ndst));
2087 for_each_possible_cpu(cpu) {
2088 struct neigh_statistics *st;
2090 st = per_cpu_ptr(tbl->stats, cpu);
2091 ndst.ndts_allocs += st->allocs;
2092 ndst.ndts_destroys += st->destroys;
2093 ndst.ndts_hash_grows += st->hash_grows;
2094 ndst.ndts_res_failed += st->res_failed;
2095 ndst.ndts_lookups += st->lookups;
2096 ndst.ndts_hits += st->hits;
2097 ndst.ndts_rcv_probes_mcast += st->rcv_probes_mcast;
2098 ndst.ndts_rcv_probes_ucast += st->rcv_probes_ucast;
2099 ndst.ndts_periodic_gc_runs += st->periodic_gc_runs;
2100 ndst.ndts_forced_gc_runs += st->forced_gc_runs;
2101 ndst.ndts_table_fulls += st->table_fulls;
2104 if (nla_put_64bit(skb, NDTA_STATS, sizeof(ndst), &ndst,
2106 goto nla_put_failure;
2109 BUG_ON(tbl->parms.dev);
2110 if (neightbl_fill_parms(skb, &tbl->parms) < 0)
2111 goto nla_put_failure;
2113 read_unlock_bh(&tbl->lock);
2114 nlmsg_end(skb, nlh);
2118 read_unlock_bh(&tbl->lock);
2119 nlmsg_cancel(skb, nlh);
2123 static int neightbl_fill_param_info(struct sk_buff *skb,
2124 struct neigh_table *tbl,
2125 struct neigh_parms *parms,
2126 u32 pid, u32 seq, int type,
2129 struct ndtmsg *ndtmsg;
2130 struct nlmsghdr *nlh;
2132 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
2136 ndtmsg = nlmsg_data(nlh);
2138 read_lock_bh(&tbl->lock);
2139 ndtmsg->ndtm_family = tbl->family;
2140 ndtmsg->ndtm_pad1 = 0;
2141 ndtmsg->ndtm_pad2 = 0;
2143 if (nla_put_string(skb, NDTA_NAME, tbl->id) < 0 ||
2144 neightbl_fill_parms(skb, parms) < 0)
2147 read_unlock_bh(&tbl->lock);
2148 nlmsg_end(skb, nlh);
2151 read_unlock_bh(&tbl->lock);
2152 nlmsg_cancel(skb, nlh);
2156 static const struct nla_policy nl_neightbl_policy[NDTA_MAX+1] = {
2157 [NDTA_NAME] = { .type = NLA_STRING },
2158 [NDTA_THRESH1] = { .type = NLA_U32 },
2159 [NDTA_THRESH2] = { .type = NLA_U32 },
2160 [NDTA_THRESH3] = { .type = NLA_U32 },
2161 [NDTA_GC_INTERVAL] = { .type = NLA_U64 },
2162 [NDTA_PARMS] = { .type = NLA_NESTED },
2165 static const struct nla_policy nl_ntbl_parm_policy[NDTPA_MAX+1] = {
2166 [NDTPA_IFINDEX] = { .type = NLA_U32 },
2167 [NDTPA_QUEUE_LEN] = { .type = NLA_U32 },
2168 [NDTPA_PROXY_QLEN] = { .type = NLA_U32 },
2169 [NDTPA_APP_PROBES] = { .type = NLA_U32 },
2170 [NDTPA_UCAST_PROBES] = { .type = NLA_U32 },
2171 [NDTPA_MCAST_PROBES] = { .type = NLA_U32 },
2172 [NDTPA_MCAST_REPROBES] = { .type = NLA_U32 },
2173 [NDTPA_BASE_REACHABLE_TIME] = { .type = NLA_U64 },
2174 [NDTPA_GC_STALETIME] = { .type = NLA_U64 },
2175 [NDTPA_DELAY_PROBE_TIME] = { .type = NLA_U64 },
2176 [NDTPA_RETRANS_TIME] = { .type = NLA_U64 },
2177 [NDTPA_ANYCAST_DELAY] = { .type = NLA_U64 },
2178 [NDTPA_PROXY_DELAY] = { .type = NLA_U64 },
2179 [NDTPA_LOCKTIME] = { .type = NLA_U64 },
2182 static int neightbl_set(struct sk_buff *skb, struct nlmsghdr *nlh,
2183 struct netlink_ext_ack *extack)
2185 struct net *net = sock_net(skb->sk);
2186 struct neigh_table *tbl;
2187 struct ndtmsg *ndtmsg;
2188 struct nlattr *tb[NDTA_MAX+1];
2192 err = nlmsg_parse_deprecated(nlh, sizeof(*ndtmsg), tb, NDTA_MAX,
2193 nl_neightbl_policy, extack);
2197 if (tb[NDTA_NAME] == NULL) {
2202 ndtmsg = nlmsg_data(nlh);
2204 for (tidx = 0; tidx < NEIGH_NR_TABLES; tidx++) {
2205 tbl = neigh_tables[tidx];
2208 if (ndtmsg->ndtm_family && tbl->family != ndtmsg->ndtm_family)
2210 if (nla_strcmp(tb[NDTA_NAME], tbl->id) == 0) {
2220 * We acquire tbl->lock to be nice to the periodic timers and
2221 * make sure they always see a consistent set of values.
2223 write_lock_bh(&tbl->lock);
2225 if (tb[NDTA_PARMS]) {
2226 struct nlattr *tbp[NDTPA_MAX+1];
2227 struct neigh_parms *p;
2230 err = nla_parse_nested_deprecated(tbp, NDTPA_MAX,
2232 nl_ntbl_parm_policy, extack);
2234 goto errout_tbl_lock;
2236 if (tbp[NDTPA_IFINDEX])
2237 ifindex = nla_get_u32(tbp[NDTPA_IFINDEX]);
2239 p = lookup_neigh_parms(tbl, net, ifindex);
2242 goto errout_tbl_lock;
2245 for (i = 1; i <= NDTPA_MAX; i++) {
2250 case NDTPA_QUEUE_LEN:
2251 NEIGH_VAR_SET(p, QUEUE_LEN_BYTES,
2252 nla_get_u32(tbp[i]) *
2253 SKB_TRUESIZE(ETH_FRAME_LEN));
2255 case NDTPA_QUEUE_LENBYTES:
2256 NEIGH_VAR_SET(p, QUEUE_LEN_BYTES,
2257 nla_get_u32(tbp[i]));
2259 case NDTPA_PROXY_QLEN:
2260 NEIGH_VAR_SET(p, PROXY_QLEN,
2261 nla_get_u32(tbp[i]));
2263 case NDTPA_APP_PROBES:
2264 NEIGH_VAR_SET(p, APP_PROBES,
2265 nla_get_u32(tbp[i]));
2267 case NDTPA_UCAST_PROBES:
2268 NEIGH_VAR_SET(p, UCAST_PROBES,
2269 nla_get_u32(tbp[i]));
2271 case NDTPA_MCAST_PROBES:
2272 NEIGH_VAR_SET(p, MCAST_PROBES,
2273 nla_get_u32(tbp[i]));
2275 case NDTPA_MCAST_REPROBES:
2276 NEIGH_VAR_SET(p, MCAST_REPROBES,
2277 nla_get_u32(tbp[i]));
2279 case NDTPA_BASE_REACHABLE_TIME:
2280 NEIGH_VAR_SET(p, BASE_REACHABLE_TIME,
2281 nla_get_msecs(tbp[i]));
2282 /* update reachable_time as well, otherwise, the change will
2283 * only be effective after the next time neigh_periodic_work
2284 * decides to recompute it (can be multiple minutes)
2287 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
2289 case NDTPA_GC_STALETIME:
2290 NEIGH_VAR_SET(p, GC_STALETIME,
2291 nla_get_msecs(tbp[i]));
2293 case NDTPA_DELAY_PROBE_TIME:
2294 NEIGH_VAR_SET(p, DELAY_PROBE_TIME,
2295 nla_get_msecs(tbp[i]));
2296 call_netevent_notifiers(NETEVENT_DELAY_PROBE_TIME_UPDATE, p);
2298 case NDTPA_RETRANS_TIME:
2299 NEIGH_VAR_SET(p, RETRANS_TIME,
2300 nla_get_msecs(tbp[i]));
2302 case NDTPA_ANYCAST_DELAY:
2303 NEIGH_VAR_SET(p, ANYCAST_DELAY,
2304 nla_get_msecs(tbp[i]));
2306 case NDTPA_PROXY_DELAY:
2307 NEIGH_VAR_SET(p, PROXY_DELAY,
2308 nla_get_msecs(tbp[i]));
2310 case NDTPA_LOCKTIME:
2311 NEIGH_VAR_SET(p, LOCKTIME,
2312 nla_get_msecs(tbp[i]));
2319 if ((tb[NDTA_THRESH1] || tb[NDTA_THRESH2] ||
2320 tb[NDTA_THRESH3] || tb[NDTA_GC_INTERVAL]) &&
2321 !net_eq(net, &init_net))
2322 goto errout_tbl_lock;
2324 if (tb[NDTA_THRESH1])
2325 tbl->gc_thresh1 = nla_get_u32(tb[NDTA_THRESH1]);
2327 if (tb[NDTA_THRESH2])
2328 tbl->gc_thresh2 = nla_get_u32(tb[NDTA_THRESH2]);
2330 if (tb[NDTA_THRESH3])
2331 tbl->gc_thresh3 = nla_get_u32(tb[NDTA_THRESH3]);
2333 if (tb[NDTA_GC_INTERVAL])
2334 tbl->gc_interval = nla_get_msecs(tb[NDTA_GC_INTERVAL]);
2339 write_unlock_bh(&tbl->lock);
2344 static int neightbl_valid_dump_info(const struct nlmsghdr *nlh,
2345 struct netlink_ext_ack *extack)
2347 struct ndtmsg *ndtm;
2349 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ndtm))) {
2350 NL_SET_ERR_MSG(extack, "Invalid header for neighbor table dump request");
2354 ndtm = nlmsg_data(nlh);
2355 if (ndtm->ndtm_pad1 || ndtm->ndtm_pad2) {
2356 NL_SET_ERR_MSG(extack, "Invalid values in header for neighbor table dump request");
2360 if (nlmsg_attrlen(nlh, sizeof(*ndtm))) {
2361 NL_SET_ERR_MSG(extack, "Invalid data after header in neighbor table dump request");
2368 static int neightbl_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2370 const struct nlmsghdr *nlh = cb->nlh;
2371 struct net *net = sock_net(skb->sk);
2372 int family, tidx, nidx = 0;
2373 int tbl_skip = cb->args[0];
2374 int neigh_skip = cb->args[1];
2375 struct neigh_table *tbl;
2377 if (cb->strict_check) {
2378 int err = neightbl_valid_dump_info(nlh, cb->extack);
2384 family = ((struct rtgenmsg *)nlmsg_data(nlh))->rtgen_family;
2386 for (tidx = 0; tidx < NEIGH_NR_TABLES; tidx++) {
2387 struct neigh_parms *p;
2389 tbl = neigh_tables[tidx];
2393 if (tidx < tbl_skip || (family && tbl->family != family))
2396 if (neightbl_fill_info(skb, tbl, NETLINK_CB(cb->skb).portid,
2397 nlh->nlmsg_seq, RTM_NEWNEIGHTBL,
2402 p = list_next_entry(&tbl->parms, list);
2403 list_for_each_entry_from(p, &tbl->parms_list, list) {
2404 if (!net_eq(neigh_parms_net(p), net))
2407 if (nidx < neigh_skip)
2410 if (neightbl_fill_param_info(skb, tbl, p,
2411 NETLINK_CB(cb->skb).portid,
2429 static int neigh_fill_info(struct sk_buff *skb, struct neighbour *neigh,
2430 u32 pid, u32 seq, int type, unsigned int flags)
2432 unsigned long now = jiffies;
2433 struct nda_cacheinfo ci;
2434 struct nlmsghdr *nlh;
2437 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2441 ndm = nlmsg_data(nlh);
2442 ndm->ndm_family = neigh->ops->family;
2445 ndm->ndm_flags = neigh->flags;
2446 ndm->ndm_type = neigh->type;
2447 ndm->ndm_ifindex = neigh->dev->ifindex;
2449 if (nla_put(skb, NDA_DST, neigh->tbl->key_len, neigh->primary_key))
2450 goto nla_put_failure;
2452 read_lock_bh(&neigh->lock);
2453 ndm->ndm_state = neigh->nud_state;
2454 if (neigh->nud_state & NUD_VALID) {
2455 char haddr[MAX_ADDR_LEN];
2457 neigh_ha_snapshot(haddr, neigh, neigh->dev);
2458 if (nla_put(skb, NDA_LLADDR, neigh->dev->addr_len, haddr) < 0) {
2459 read_unlock_bh(&neigh->lock);
2460 goto nla_put_failure;
2464 ci.ndm_used = jiffies_to_clock_t(now - neigh->used);
2465 ci.ndm_confirmed = jiffies_to_clock_t(now - neigh->confirmed);
2466 ci.ndm_updated = jiffies_to_clock_t(now - neigh->updated);
2467 ci.ndm_refcnt = refcount_read(&neigh->refcnt) - 1;
2468 read_unlock_bh(&neigh->lock);
2470 if (nla_put_u32(skb, NDA_PROBES, atomic_read(&neigh->probes)) ||
2471 nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
2472 goto nla_put_failure;
2474 if (neigh->protocol && nla_put_u8(skb, NDA_PROTOCOL, neigh->protocol))
2475 goto nla_put_failure;
2477 nlmsg_end(skb, nlh);
2481 nlmsg_cancel(skb, nlh);
2485 static int pneigh_fill_info(struct sk_buff *skb, struct pneigh_entry *pn,
2486 u32 pid, u32 seq, int type, unsigned int flags,
2487 struct neigh_table *tbl)
2489 struct nlmsghdr *nlh;
2492 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2496 ndm = nlmsg_data(nlh);
2497 ndm->ndm_family = tbl->family;
2500 ndm->ndm_flags = pn->flags | NTF_PROXY;
2501 ndm->ndm_type = RTN_UNICAST;
2502 ndm->ndm_ifindex = pn->dev ? pn->dev->ifindex : 0;
2503 ndm->ndm_state = NUD_NONE;
2505 if (nla_put(skb, NDA_DST, tbl->key_len, pn->key))
2506 goto nla_put_failure;
2508 if (pn->protocol && nla_put_u8(skb, NDA_PROTOCOL, pn->protocol))
2509 goto nla_put_failure;
2511 nlmsg_end(skb, nlh);
2515 nlmsg_cancel(skb, nlh);
2519 static void neigh_update_notify(struct neighbour *neigh, u32 nlmsg_pid)
2521 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, neigh);
2522 __neigh_notify(neigh, RTM_NEWNEIGH, 0, nlmsg_pid);
2525 static bool neigh_master_filtered(struct net_device *dev, int master_idx)
2527 struct net_device *master;
2532 master = dev ? netdev_master_upper_dev_get(dev) : NULL;
2534 /* 0 is already used to denote NDA_MASTER wasn't passed, therefore need another
2535 * invalid value for ifindex to denote "no master".
2537 if (master_idx == -1)
2540 if (!master || master->ifindex != master_idx)
2546 static bool neigh_ifindex_filtered(struct net_device *dev, int filter_idx)
2548 if (filter_idx && (!dev || dev->ifindex != filter_idx))
2554 struct neigh_dump_filter {
2559 static int neigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2560 struct netlink_callback *cb,
2561 struct neigh_dump_filter *filter)
2563 struct net *net = sock_net(skb->sk);
2564 struct neighbour *n;
2565 int rc, h, s_h = cb->args[1];
2566 int idx, s_idx = idx = cb->args[2];
2567 struct neigh_hash_table *nht;
2568 unsigned int flags = NLM_F_MULTI;
2570 if (filter->dev_idx || filter->master_idx)
2571 flags |= NLM_F_DUMP_FILTERED;
2574 nht = rcu_dereference_bh(tbl->nht);
2576 for (h = s_h; h < (1 << nht->hash_shift); h++) {
2579 for (n = rcu_dereference_bh(nht->hash_buckets[h]), idx = 0;
2581 n = rcu_dereference_bh(n->next)) {
2582 if (idx < s_idx || !net_eq(dev_net(n->dev), net))
2584 if (neigh_ifindex_filtered(n->dev, filter->dev_idx) ||
2585 neigh_master_filtered(n->dev, filter->master_idx))
2587 if (neigh_fill_info(skb, n, NETLINK_CB(cb->skb).portid,
2600 rcu_read_unlock_bh();
2606 static int pneigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2607 struct netlink_callback *cb,
2608 struct neigh_dump_filter *filter)
2610 struct pneigh_entry *n;
2611 struct net *net = sock_net(skb->sk);
2612 int rc, h, s_h = cb->args[3];
2613 int idx, s_idx = idx = cb->args[4];
2614 unsigned int flags = NLM_F_MULTI;
2616 if (filter->dev_idx || filter->master_idx)
2617 flags |= NLM_F_DUMP_FILTERED;
2619 read_lock_bh(&tbl->lock);
2621 for (h = s_h; h <= PNEIGH_HASHMASK; h++) {
2624 for (n = tbl->phash_buckets[h], idx = 0; n; n = n->next) {
2625 if (idx < s_idx || pneigh_net(n) != net)
2627 if (neigh_ifindex_filtered(n->dev, filter->dev_idx) ||
2628 neigh_master_filtered(n->dev, filter->master_idx))
2630 if (pneigh_fill_info(skb, n, NETLINK_CB(cb->skb).portid,
2632 RTM_NEWNEIGH, flags, tbl) < 0) {
2633 read_unlock_bh(&tbl->lock);
2642 read_unlock_bh(&tbl->lock);
2651 static int neigh_valid_dump_req(const struct nlmsghdr *nlh,
2653 struct neigh_dump_filter *filter,
2654 struct netlink_ext_ack *extack)
2656 struct nlattr *tb[NDA_MAX + 1];
2662 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ndm))) {
2663 NL_SET_ERR_MSG(extack, "Invalid header for neighbor dump request");
2667 ndm = nlmsg_data(nlh);
2668 if (ndm->ndm_pad1 || ndm->ndm_pad2 || ndm->ndm_ifindex ||
2669 ndm->ndm_state || ndm->ndm_type) {
2670 NL_SET_ERR_MSG(extack, "Invalid values in header for neighbor dump request");
2674 if (ndm->ndm_flags & ~NTF_PROXY) {
2675 NL_SET_ERR_MSG(extack, "Invalid flags in header for neighbor dump request");
2679 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct ndmsg),
2680 tb, NDA_MAX, nda_policy,
2683 err = nlmsg_parse_deprecated(nlh, sizeof(struct ndmsg), tb,
2684 NDA_MAX, nda_policy, extack);
2689 for (i = 0; i <= NDA_MAX; ++i) {
2693 /* all new attributes should require strict_check */
2696 filter->dev_idx = nla_get_u32(tb[i]);
2699 filter->master_idx = nla_get_u32(tb[i]);
2703 NL_SET_ERR_MSG(extack, "Unsupported attribute in neighbor dump request");
2712 static int neigh_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2714 const struct nlmsghdr *nlh = cb->nlh;
2715 struct neigh_dump_filter filter = {};
2716 struct neigh_table *tbl;
2721 family = ((struct rtgenmsg *)nlmsg_data(nlh))->rtgen_family;
2723 /* check for full ndmsg structure presence, family member is
2724 * the same for both structures
2726 if (nlmsg_len(nlh) >= sizeof(struct ndmsg) &&
2727 ((struct ndmsg *)nlmsg_data(nlh))->ndm_flags == NTF_PROXY)
2730 err = neigh_valid_dump_req(nlh, cb->strict_check, &filter, cb->extack);
2731 if (err < 0 && cb->strict_check)
2736 for (t = 0; t < NEIGH_NR_TABLES; t++) {
2737 tbl = neigh_tables[t];
2741 if (t < s_t || (family && tbl->family != family))
2744 memset(&cb->args[1], 0, sizeof(cb->args) -
2745 sizeof(cb->args[0]));
2747 err = pneigh_dump_table(tbl, skb, cb, &filter);
2749 err = neigh_dump_table(tbl, skb, cb, &filter);
2758 static int neigh_valid_get_req(const struct nlmsghdr *nlh,
2759 struct neigh_table **tbl,
2760 void **dst, int *dev_idx, u8 *ndm_flags,
2761 struct netlink_ext_ack *extack)
2763 struct nlattr *tb[NDA_MAX + 1];
2767 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ndm))) {
2768 NL_SET_ERR_MSG(extack, "Invalid header for neighbor get request");
2772 ndm = nlmsg_data(nlh);
2773 if (ndm->ndm_pad1 || ndm->ndm_pad2 || ndm->ndm_state ||
2775 NL_SET_ERR_MSG(extack, "Invalid values in header for neighbor get request");
2779 if (ndm->ndm_flags & ~NTF_PROXY) {
2780 NL_SET_ERR_MSG(extack, "Invalid flags in header for neighbor get request");
2784 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct ndmsg), tb,
2785 NDA_MAX, nda_policy, extack);
2789 *ndm_flags = ndm->ndm_flags;
2790 *dev_idx = ndm->ndm_ifindex;
2791 *tbl = neigh_find_table(ndm->ndm_family);
2793 NL_SET_ERR_MSG(extack, "Unsupported family in header for neighbor get request");
2794 return -EAFNOSUPPORT;
2797 for (i = 0; i <= NDA_MAX; ++i) {
2803 if (nla_len(tb[i]) != (int)(*tbl)->key_len) {
2804 NL_SET_ERR_MSG(extack, "Invalid network address in neighbor get request");
2807 *dst = nla_data(tb[i]);
2810 NL_SET_ERR_MSG(extack, "Unsupported attribute in neighbor get request");
2818 static inline size_t neigh_nlmsg_size(void)
2820 return NLMSG_ALIGN(sizeof(struct ndmsg))
2821 + nla_total_size(MAX_ADDR_LEN) /* NDA_DST */
2822 + nla_total_size(MAX_ADDR_LEN) /* NDA_LLADDR */
2823 + nla_total_size(sizeof(struct nda_cacheinfo))
2824 + nla_total_size(4) /* NDA_PROBES */
2825 + nla_total_size(1); /* NDA_PROTOCOL */
2828 static int neigh_get_reply(struct net *net, struct neighbour *neigh,
2831 struct sk_buff *skb;
2834 skb = nlmsg_new(neigh_nlmsg_size(), GFP_KERNEL);
2838 err = neigh_fill_info(skb, neigh, pid, seq, RTM_NEWNEIGH, 0);
2844 err = rtnl_unicast(skb, net, pid);
2849 static inline size_t pneigh_nlmsg_size(void)
2851 return NLMSG_ALIGN(sizeof(struct ndmsg))
2852 + nla_total_size(MAX_ADDR_LEN) /* NDA_DST */
2853 + nla_total_size(1); /* NDA_PROTOCOL */
2856 static int pneigh_get_reply(struct net *net, struct pneigh_entry *neigh,
2857 u32 pid, u32 seq, struct neigh_table *tbl)
2859 struct sk_buff *skb;
2862 skb = nlmsg_new(pneigh_nlmsg_size(), GFP_KERNEL);
2866 err = pneigh_fill_info(skb, neigh, pid, seq, RTM_NEWNEIGH, 0, tbl);
2872 err = rtnl_unicast(skb, net, pid);
2877 static int neigh_get(struct sk_buff *in_skb, struct nlmsghdr *nlh,
2878 struct netlink_ext_ack *extack)
2880 struct net *net = sock_net(in_skb->sk);
2881 struct net_device *dev = NULL;
2882 struct neigh_table *tbl = NULL;
2883 struct neighbour *neigh;
2889 err = neigh_valid_get_req(nlh, &tbl, &dst, &dev_idx, &ndm_flags,
2895 dev = __dev_get_by_index(net, dev_idx);
2897 NL_SET_ERR_MSG(extack, "Unknown device ifindex");
2903 NL_SET_ERR_MSG(extack, "Network address not specified");
2907 if (ndm_flags & NTF_PROXY) {
2908 struct pneigh_entry *pn;
2910 pn = pneigh_lookup(tbl, net, dst, dev, 0);
2912 NL_SET_ERR_MSG(extack, "Proxy neighbour entry not found");
2915 return pneigh_get_reply(net, pn, NETLINK_CB(in_skb).portid,
2916 nlh->nlmsg_seq, tbl);
2920 NL_SET_ERR_MSG(extack, "No device specified");
2924 neigh = neigh_lookup(tbl, dst, dev);
2926 NL_SET_ERR_MSG(extack, "Neighbour entry not found");
2930 err = neigh_get_reply(net, neigh, NETLINK_CB(in_skb).portid,
2933 neigh_release(neigh);
2938 void neigh_for_each(struct neigh_table *tbl, void (*cb)(struct neighbour *, void *), void *cookie)
2941 struct neigh_hash_table *nht;
2944 nht = rcu_dereference_bh(tbl->nht);
2946 read_lock(&tbl->lock); /* avoid resizes */
2947 for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
2948 struct neighbour *n;
2950 for (n = rcu_dereference_bh(nht->hash_buckets[chain]);
2952 n = rcu_dereference_bh(n->next))
2955 read_unlock(&tbl->lock);
2956 rcu_read_unlock_bh();
2958 EXPORT_SYMBOL(neigh_for_each);
2960 /* The tbl->lock must be held as a writer and BH disabled. */
2961 void __neigh_for_each_release(struct neigh_table *tbl,
2962 int (*cb)(struct neighbour *))
2965 struct neigh_hash_table *nht;
2967 nht = rcu_dereference_protected(tbl->nht,
2968 lockdep_is_held(&tbl->lock));
2969 for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
2970 struct neighbour *n;
2971 struct neighbour __rcu **np;
2973 np = &nht->hash_buckets[chain];
2974 while ((n = rcu_dereference_protected(*np,
2975 lockdep_is_held(&tbl->lock))) != NULL) {
2978 write_lock(&n->lock);
2981 rcu_assign_pointer(*np,
2982 rcu_dereference_protected(n->next,
2983 lockdep_is_held(&tbl->lock)));
2987 write_unlock(&n->lock);
2989 neigh_cleanup_and_release(n);
2993 EXPORT_SYMBOL(__neigh_for_each_release);
2995 int neigh_xmit(int index, struct net_device *dev,
2996 const void *addr, struct sk_buff *skb)
2998 int err = -EAFNOSUPPORT;
2999 if (likely(index < NEIGH_NR_TABLES)) {
3000 struct neigh_table *tbl;
3001 struct neighbour *neigh;
3003 tbl = neigh_tables[index];
3007 if (index == NEIGH_ARP_TABLE) {
3008 u32 key = *((u32 *)addr);
3010 neigh = __ipv4_neigh_lookup_noref(dev, key);
3012 neigh = __neigh_lookup_noref(tbl, addr, dev);
3015 neigh = __neigh_create(tbl, addr, dev, false);
3016 err = PTR_ERR(neigh);
3017 if (IS_ERR(neigh)) {
3018 rcu_read_unlock_bh();
3021 err = neigh->output(neigh, skb);
3022 rcu_read_unlock_bh();
3024 else if (index == NEIGH_LINK_TABLE) {
3025 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
3026 addr, NULL, skb->len);
3029 err = dev_queue_xmit(skb);
3037 EXPORT_SYMBOL(neigh_xmit);
3039 #ifdef CONFIG_PROC_FS
3041 static struct neighbour *neigh_get_first(struct seq_file *seq)
3043 struct neigh_seq_state *state = seq->private;
3044 struct net *net = seq_file_net(seq);
3045 struct neigh_hash_table *nht = state->nht;
3046 struct neighbour *n = NULL;
3049 state->flags &= ~NEIGH_SEQ_IS_PNEIGH;
3050 for (bucket = 0; bucket < (1 << nht->hash_shift); bucket++) {
3051 n = rcu_dereference_bh(nht->hash_buckets[bucket]);
3054 if (!net_eq(dev_net(n->dev), net))
3056 if (state->neigh_sub_iter) {
3060 v = state->neigh_sub_iter(state, n, &fakep);
3064 if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
3066 if (n->nud_state & ~NUD_NOARP)
3069 n = rcu_dereference_bh(n->next);
3075 state->bucket = bucket;
3080 static struct neighbour *neigh_get_next(struct seq_file *seq,
3081 struct neighbour *n,
3084 struct neigh_seq_state *state = seq->private;
3085 struct net *net = seq_file_net(seq);
3086 struct neigh_hash_table *nht = state->nht;
3088 if (state->neigh_sub_iter) {
3089 void *v = state->neigh_sub_iter(state, n, pos);
3093 n = rcu_dereference_bh(n->next);
3097 if (!net_eq(dev_net(n->dev), net))
3099 if (state->neigh_sub_iter) {
3100 void *v = state->neigh_sub_iter(state, n, pos);
3105 if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
3108 if (n->nud_state & ~NUD_NOARP)
3111 n = rcu_dereference_bh(n->next);
3117 if (++state->bucket >= (1 << nht->hash_shift))
3120 n = rcu_dereference_bh(nht->hash_buckets[state->bucket]);
3128 static struct neighbour *neigh_get_idx(struct seq_file *seq, loff_t *pos)
3130 struct neighbour *n = neigh_get_first(seq);
3135 n = neigh_get_next(seq, n, pos);
3140 return *pos ? NULL : n;
3143 static struct pneigh_entry *pneigh_get_first(struct seq_file *seq)
3145 struct neigh_seq_state *state = seq->private;
3146 struct net *net = seq_file_net(seq);
3147 struct neigh_table *tbl = state->tbl;
3148 struct pneigh_entry *pn = NULL;
3151 state->flags |= NEIGH_SEQ_IS_PNEIGH;
3152 for (bucket = 0; bucket <= PNEIGH_HASHMASK; bucket++) {
3153 pn = tbl->phash_buckets[bucket];
3154 while (pn && !net_eq(pneigh_net(pn), net))
3159 state->bucket = bucket;
3164 static struct pneigh_entry *pneigh_get_next(struct seq_file *seq,
3165 struct pneigh_entry *pn,
3168 struct neigh_seq_state *state = seq->private;
3169 struct net *net = seq_file_net(seq);
3170 struct neigh_table *tbl = state->tbl;
3174 } while (pn && !net_eq(pneigh_net(pn), net));
3177 if (++state->bucket > PNEIGH_HASHMASK)
3179 pn = tbl->phash_buckets[state->bucket];
3180 while (pn && !net_eq(pneigh_net(pn), net))
3192 static struct pneigh_entry *pneigh_get_idx(struct seq_file *seq, loff_t *pos)
3194 struct pneigh_entry *pn = pneigh_get_first(seq);
3199 pn = pneigh_get_next(seq, pn, pos);
3204 return *pos ? NULL : pn;
3207 static void *neigh_get_idx_any(struct seq_file *seq, loff_t *pos)
3209 struct neigh_seq_state *state = seq->private;
3211 loff_t idxpos = *pos;
3213 rc = neigh_get_idx(seq, &idxpos);
3214 if (!rc && !(state->flags & NEIGH_SEQ_NEIGH_ONLY))
3215 rc = pneigh_get_idx(seq, &idxpos);
3220 void *neigh_seq_start(struct seq_file *seq, loff_t *pos, struct neigh_table *tbl, unsigned int neigh_seq_flags)
3221 __acquires(tbl->lock)
3224 struct neigh_seq_state *state = seq->private;
3228 state->flags = (neigh_seq_flags & ~NEIGH_SEQ_IS_PNEIGH);
3231 state->nht = rcu_dereference_bh(tbl->nht);
3232 read_lock(&tbl->lock);
3234 return *pos ? neigh_get_idx_any(seq, pos) : SEQ_START_TOKEN;
3236 EXPORT_SYMBOL(neigh_seq_start);
3238 void *neigh_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3240 struct neigh_seq_state *state;
3243 if (v == SEQ_START_TOKEN) {
3244 rc = neigh_get_first(seq);
3248 state = seq->private;
3249 if (!(state->flags & NEIGH_SEQ_IS_PNEIGH)) {
3250 rc = neigh_get_next(seq, v, NULL);
3253 if (!(state->flags & NEIGH_SEQ_NEIGH_ONLY))
3254 rc = pneigh_get_first(seq);
3256 BUG_ON(state->flags & NEIGH_SEQ_NEIGH_ONLY);
3257 rc = pneigh_get_next(seq, v, NULL);
3263 EXPORT_SYMBOL(neigh_seq_next);
3265 void neigh_seq_stop(struct seq_file *seq, void *v)
3266 __releases(tbl->lock)
3269 struct neigh_seq_state *state = seq->private;
3270 struct neigh_table *tbl = state->tbl;
3272 read_unlock(&tbl->lock);
3273 rcu_read_unlock_bh();
3275 EXPORT_SYMBOL(neigh_seq_stop);
3277 /* statistics via seq_file */
3279 static void *neigh_stat_seq_start(struct seq_file *seq, loff_t *pos)
3281 struct neigh_table *tbl = PDE_DATA(file_inode(seq->file));
3285 return SEQ_START_TOKEN;
3287 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
3288 if (!cpu_possible(cpu))
3291 return per_cpu_ptr(tbl->stats, cpu);
3296 static void *neigh_stat_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3298 struct neigh_table *tbl = PDE_DATA(file_inode(seq->file));
3301 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
3302 if (!cpu_possible(cpu))
3305 return per_cpu_ptr(tbl->stats, cpu);
3311 static void neigh_stat_seq_stop(struct seq_file *seq, void *v)
3316 static int neigh_stat_seq_show(struct seq_file *seq, void *v)
3318 struct neigh_table *tbl = PDE_DATA(file_inode(seq->file));
3319 struct neigh_statistics *st = v;
3321 if (v == SEQ_START_TOKEN) {
3322 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");
3326 seq_printf(seq, "%08x %08lx %08lx %08lx %08lx %08lx %08lx "
3327 "%08lx %08lx %08lx "
3328 "%08lx %08lx %08lx\n",
3329 atomic_read(&tbl->entries),
3340 st->rcv_probes_mcast,
3341 st->rcv_probes_ucast,
3343 st->periodic_gc_runs,
3352 static const struct seq_operations neigh_stat_seq_ops = {
3353 .start = neigh_stat_seq_start,
3354 .next = neigh_stat_seq_next,
3355 .stop = neigh_stat_seq_stop,
3356 .show = neigh_stat_seq_show,
3358 #endif /* CONFIG_PROC_FS */
3360 static void __neigh_notify(struct neighbour *n, int type, int flags,
3363 struct net *net = dev_net(n->dev);
3364 struct sk_buff *skb;
3367 skb = nlmsg_new(neigh_nlmsg_size(), GFP_ATOMIC);
3371 err = neigh_fill_info(skb, n, pid, 0, type, flags);
3373 /* -EMSGSIZE implies BUG in neigh_nlmsg_size() */
3374 WARN_ON(err == -EMSGSIZE);
3378 rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
3382 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
3385 void neigh_app_ns(struct neighbour *n)
3387 __neigh_notify(n, RTM_GETNEIGH, NLM_F_REQUEST, 0);
3389 EXPORT_SYMBOL(neigh_app_ns);
3391 #ifdef CONFIG_SYSCTL
3392 static int unres_qlen_max = INT_MAX / SKB_TRUESIZE(ETH_FRAME_LEN);
3394 static int proc_unres_qlen(struct ctl_table *ctl, int write,
3395 void *buffer, size_t *lenp, loff_t *ppos)
3398 struct ctl_table tmp = *ctl;
3400 tmp.extra1 = SYSCTL_ZERO;
3401 tmp.extra2 = &unres_qlen_max;
3404 size = *(int *)ctl->data / SKB_TRUESIZE(ETH_FRAME_LEN);
3405 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
3408 *(int *)ctl->data = size * SKB_TRUESIZE(ETH_FRAME_LEN);
3412 static struct neigh_parms *neigh_get_dev_parms_rcu(struct net_device *dev,
3417 return __in_dev_arp_parms_get_rcu(dev);
3419 return __in6_dev_nd_parms_get_rcu(dev);
3424 static void neigh_copy_dflt_parms(struct net *net, struct neigh_parms *p,
3427 struct net_device *dev;
3428 int family = neigh_parms_family(p);
3431 for_each_netdev_rcu(net, dev) {
3432 struct neigh_parms *dst_p =
3433 neigh_get_dev_parms_rcu(dev, family);
3435 if (dst_p && !test_bit(index, dst_p->data_state))
3436 dst_p->data[index] = p->data[index];
3441 static void neigh_proc_update(struct ctl_table *ctl, int write)
3443 struct net_device *dev = ctl->extra1;
3444 struct neigh_parms *p = ctl->extra2;
3445 struct net *net = neigh_parms_net(p);
3446 int index = (int *) ctl->data - p->data;
3451 set_bit(index, p->data_state);
3452 if (index == NEIGH_VAR_DELAY_PROBE_TIME)
3453 call_netevent_notifiers(NETEVENT_DELAY_PROBE_TIME_UPDATE, p);
3454 if (!dev) /* NULL dev means this is default value */
3455 neigh_copy_dflt_parms(net, p, index);
3458 static int neigh_proc_dointvec_zero_intmax(struct ctl_table *ctl, int write,
3459 void *buffer, size_t *lenp,
3462 struct ctl_table tmp = *ctl;
3465 tmp.extra1 = SYSCTL_ZERO;
3466 tmp.extra2 = SYSCTL_INT_MAX;
3468 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
3469 neigh_proc_update(ctl, write);
3473 int neigh_proc_dointvec(struct ctl_table *ctl, int write, void *buffer,
3474 size_t *lenp, loff_t *ppos)
3476 int ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
3478 neigh_proc_update(ctl, write);
3481 EXPORT_SYMBOL(neigh_proc_dointvec);
3483 int neigh_proc_dointvec_jiffies(struct ctl_table *ctl, int write, void *buffer,
3484 size_t *lenp, loff_t *ppos)
3486 int ret = proc_dointvec_jiffies(ctl, write, buffer, lenp, ppos);
3488 neigh_proc_update(ctl, write);
3491 EXPORT_SYMBOL(neigh_proc_dointvec_jiffies);
3493 static int neigh_proc_dointvec_userhz_jiffies(struct ctl_table *ctl, int write,
3494 void *buffer, size_t *lenp,
3497 int ret = proc_dointvec_userhz_jiffies(ctl, write, buffer, lenp, ppos);
3499 neigh_proc_update(ctl, write);
3503 int neigh_proc_dointvec_ms_jiffies(struct ctl_table *ctl, int write,
3504 void *buffer, size_t *lenp, loff_t *ppos)
3506 int ret = proc_dointvec_ms_jiffies(ctl, write, buffer, lenp, ppos);
3508 neigh_proc_update(ctl, write);
3511 EXPORT_SYMBOL(neigh_proc_dointvec_ms_jiffies);
3513 static int neigh_proc_dointvec_unres_qlen(struct ctl_table *ctl, int write,
3514 void *buffer, size_t *lenp,
3517 int ret = proc_unres_qlen(ctl, write, buffer, lenp, ppos);
3519 neigh_proc_update(ctl, write);
3523 static int neigh_proc_base_reachable_time(struct ctl_table *ctl, int write,
3524 void *buffer, size_t *lenp,
3527 struct neigh_parms *p = ctl->extra2;
3530 if (strcmp(ctl->procname, "base_reachable_time") == 0)
3531 ret = neigh_proc_dointvec_jiffies(ctl, write, buffer, lenp, ppos);
3532 else if (strcmp(ctl->procname, "base_reachable_time_ms") == 0)
3533 ret = neigh_proc_dointvec_ms_jiffies(ctl, write, buffer, lenp, ppos);
3537 if (write && ret == 0) {
3538 /* update reachable_time as well, otherwise, the change will
3539 * only be effective after the next time neigh_periodic_work
3540 * decides to recompute it
3543 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
3548 #define NEIGH_PARMS_DATA_OFFSET(index) \
3549 (&((struct neigh_parms *) 0)->data[index])
3551 #define NEIGH_SYSCTL_ENTRY(attr, data_attr, name, mval, proc) \
3552 [NEIGH_VAR_ ## attr] = { \
3554 .data = NEIGH_PARMS_DATA_OFFSET(NEIGH_VAR_ ## data_attr), \
3555 .maxlen = sizeof(int), \
3557 .proc_handler = proc, \
3560 #define NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(attr, name) \
3561 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_zero_intmax)
3563 #define NEIGH_SYSCTL_JIFFIES_ENTRY(attr, name) \
3564 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_jiffies)
3566 #define NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(attr, name) \
3567 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_userhz_jiffies)
3569 #define NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(attr, data_attr, name) \
3570 NEIGH_SYSCTL_ENTRY(attr, data_attr, name, 0644, neigh_proc_dointvec_ms_jiffies)
3572 #define NEIGH_SYSCTL_UNRES_QLEN_REUSED_ENTRY(attr, data_attr, name) \
3573 NEIGH_SYSCTL_ENTRY(attr, data_attr, name, 0644, neigh_proc_dointvec_unres_qlen)
3575 static struct neigh_sysctl_table {
3576 struct ctl_table_header *sysctl_header;
3577 struct ctl_table neigh_vars[NEIGH_VAR_MAX + 1];
3578 } neigh_sysctl_template __read_mostly = {
3580 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(MCAST_PROBES, "mcast_solicit"),
3581 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(UCAST_PROBES, "ucast_solicit"),
3582 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(APP_PROBES, "app_solicit"),
3583 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(MCAST_REPROBES, "mcast_resolicit"),
3584 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(RETRANS_TIME, "retrans_time"),
3585 NEIGH_SYSCTL_JIFFIES_ENTRY(BASE_REACHABLE_TIME, "base_reachable_time"),
3586 NEIGH_SYSCTL_JIFFIES_ENTRY(DELAY_PROBE_TIME, "delay_first_probe_time"),
3587 NEIGH_SYSCTL_JIFFIES_ENTRY(GC_STALETIME, "gc_stale_time"),
3588 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(QUEUE_LEN_BYTES, "unres_qlen_bytes"),
3589 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(PROXY_QLEN, "proxy_qlen"),
3590 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(ANYCAST_DELAY, "anycast_delay"),
3591 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(PROXY_DELAY, "proxy_delay"),
3592 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(LOCKTIME, "locktime"),
3593 NEIGH_SYSCTL_UNRES_QLEN_REUSED_ENTRY(QUEUE_LEN, QUEUE_LEN_BYTES, "unres_qlen"),
3594 NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(RETRANS_TIME_MS, RETRANS_TIME, "retrans_time_ms"),
3595 NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(BASE_REACHABLE_TIME_MS, BASE_REACHABLE_TIME, "base_reachable_time_ms"),
3596 [NEIGH_VAR_GC_INTERVAL] = {
3597 .procname = "gc_interval",
3598 .maxlen = sizeof(int),
3600 .proc_handler = proc_dointvec_jiffies,
3602 [NEIGH_VAR_GC_THRESH1] = {
3603 .procname = "gc_thresh1",
3604 .maxlen = sizeof(int),
3606 .extra1 = SYSCTL_ZERO,
3607 .extra2 = SYSCTL_INT_MAX,
3608 .proc_handler = proc_dointvec_minmax,
3610 [NEIGH_VAR_GC_THRESH2] = {
3611 .procname = "gc_thresh2",
3612 .maxlen = sizeof(int),
3614 .extra1 = SYSCTL_ZERO,
3615 .extra2 = SYSCTL_INT_MAX,
3616 .proc_handler = proc_dointvec_minmax,
3618 [NEIGH_VAR_GC_THRESH3] = {
3619 .procname = "gc_thresh3",
3620 .maxlen = sizeof(int),
3622 .extra1 = SYSCTL_ZERO,
3623 .extra2 = SYSCTL_INT_MAX,
3624 .proc_handler = proc_dointvec_minmax,
3630 int neigh_sysctl_register(struct net_device *dev, struct neigh_parms *p,
3631 proc_handler *handler)
3634 struct neigh_sysctl_table *t;
3635 const char *dev_name_source;
3636 char neigh_path[ sizeof("net//neigh/") + IFNAMSIZ + IFNAMSIZ ];
3639 t = kmemdup(&neigh_sysctl_template, sizeof(*t), GFP_KERNEL);
3643 for (i = 0; i < NEIGH_VAR_GC_INTERVAL; i++) {
3644 t->neigh_vars[i].data += (long) p;
3645 t->neigh_vars[i].extra1 = dev;
3646 t->neigh_vars[i].extra2 = p;
3650 dev_name_source = dev->name;
3651 /* Terminate the table early */
3652 memset(&t->neigh_vars[NEIGH_VAR_GC_INTERVAL], 0,
3653 sizeof(t->neigh_vars[NEIGH_VAR_GC_INTERVAL]));
3655 struct neigh_table *tbl = p->tbl;
3656 dev_name_source = "default";
3657 t->neigh_vars[NEIGH_VAR_GC_INTERVAL].data = &tbl->gc_interval;
3658 t->neigh_vars[NEIGH_VAR_GC_THRESH1].data = &tbl->gc_thresh1;
3659 t->neigh_vars[NEIGH_VAR_GC_THRESH2].data = &tbl->gc_thresh2;
3660 t->neigh_vars[NEIGH_VAR_GC_THRESH3].data = &tbl->gc_thresh3;
3665 t->neigh_vars[NEIGH_VAR_RETRANS_TIME].proc_handler = handler;
3667 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].proc_handler = handler;
3668 /* RetransTime (in milliseconds)*/
3669 t->neigh_vars[NEIGH_VAR_RETRANS_TIME_MS].proc_handler = handler;
3670 /* ReachableTime (in milliseconds) */
3671 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].proc_handler = handler;
3673 /* Those handlers will update p->reachable_time after
3674 * base_reachable_time(_ms) is set to ensure the new timer starts being
3675 * applied after the next neighbour update instead of waiting for
3676 * neigh_periodic_work to update its value (can be multiple minutes)
3677 * So any handler that replaces them should do this as well
3680 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].proc_handler =
3681 neigh_proc_base_reachable_time;
3682 /* ReachableTime (in milliseconds) */
3683 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].proc_handler =
3684 neigh_proc_base_reachable_time;
3687 /* Don't export sysctls to unprivileged users */
3688 if (neigh_parms_net(p)->user_ns != &init_user_ns)
3689 t->neigh_vars[0].procname = NULL;
3691 switch (neigh_parms_family(p)) {
3702 snprintf(neigh_path, sizeof(neigh_path), "net/%s/neigh/%s",
3703 p_name, dev_name_source);
3705 register_net_sysctl(neigh_parms_net(p), neigh_path, t->neigh_vars);
3706 if (!t->sysctl_header)
3709 p->sysctl_table = t;
3717 EXPORT_SYMBOL(neigh_sysctl_register);
3719 void neigh_sysctl_unregister(struct neigh_parms *p)
3721 if (p->sysctl_table) {
3722 struct neigh_sysctl_table *t = p->sysctl_table;
3723 p->sysctl_table = NULL;
3724 unregister_net_sysctl_table(t->sysctl_header);
3728 EXPORT_SYMBOL(neigh_sysctl_unregister);
3730 #endif /* CONFIG_SYSCTL */
3732 static int __init neigh_init(void)
3734 rtnl_register(PF_UNSPEC, RTM_NEWNEIGH, neigh_add, NULL, 0);
3735 rtnl_register(PF_UNSPEC, RTM_DELNEIGH, neigh_delete, NULL, 0);
3736 rtnl_register(PF_UNSPEC, RTM_GETNEIGH, neigh_get, neigh_dump_info, 0);
3738 rtnl_register(PF_UNSPEC, RTM_GETNEIGHTBL, NULL, neightbl_dump_info,
3740 rtnl_register(PF_UNSPEC, RTM_SETNEIGHTBL, neightbl_set, NULL, 0);
3745 subsys_initcall(neigh_init);