2 * Generic address resolution entity
5 * Pedro Roque <roque@di.fc.ul.pt>
6 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License
10 * as published by the Free Software Foundation; either version
11 * 2 of the License, or (at your option) any later version.
14 * Vitaly E. Lavrov releasing NULL neighbor in neigh_add.
15 * Harald Welte Add neighbour cache statistics like rtstat
18 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
20 #include <linux/slab.h>
21 #include <linux/types.h>
22 #include <linux/kernel.h>
23 #include <linux/module.h>
24 #include <linux/socket.h>
25 #include <linux/netdevice.h>
26 #include <linux/proc_fs.h>
28 #include <linux/sysctl.h>
30 #include <linux/times.h>
31 #include <net/net_namespace.h>
32 #include <net/neighbour.h>
35 #include <net/netevent.h>
36 #include <net/netlink.h>
37 #include <linux/rtnetlink.h>
38 #include <linux/random.h>
39 #include <linux/string.h>
40 #include <linux/log2.h>
41 #include <linux/inetdevice.h>
42 #include <net/addrconf.h>
46 #define neigh_dbg(level, fmt, ...) \
48 if (level <= NEIGH_DEBUG) \
49 pr_debug(fmt, ##__VA_ARGS__); \
52 #define PNEIGH_HASHMASK 0xF
54 static void neigh_timer_handler(unsigned long arg);
55 static void __neigh_notify(struct neighbour *n, int type, int flags);
56 static void neigh_update_notify(struct neighbour *neigh);
57 static int pneigh_ifdown(struct neigh_table *tbl, struct net_device *dev);
60 static const struct file_operations neigh_stat_seq_fops;
64 Neighbour hash table buckets are protected with rwlock tbl->lock.
66 - All the scans/updates to hash buckets MUST be made under this lock.
67 - NOTHING clever should be made under this lock: no callbacks
68 to protocol backends, no attempts to send something to network.
69 It will result in deadlocks, if backend/driver wants to use neighbour
71 - If the entry requires some non-trivial actions, increase
72 its reference count and release table lock.
74 Neighbour entries are protected:
75 - with reference count.
76 - with rwlock neigh->lock
78 Reference count prevents destruction.
80 neigh->lock mainly serializes ll address data and its validity state.
81 However, the same lock is used to protect another entry fields:
85 Again, nothing clever shall be made under neigh->lock,
86 the most complicated procedure, which we allow is dev->hard_header.
87 It is supposed, that dev->hard_header is simplistic and does
88 not make callbacks to neighbour tables.
91 static int neigh_blackhole(struct neighbour *neigh, struct sk_buff *skb)
97 static void neigh_cleanup_and_release(struct neighbour *neigh)
99 if (neigh->parms->neigh_cleanup)
100 neigh->parms->neigh_cleanup(neigh);
102 __neigh_notify(neigh, RTM_DELNEIGH, 0);
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);
119 static int neigh_forced_gc(struct neigh_table *tbl)
123 struct neigh_hash_table *nht;
125 NEIGH_CACHE_STAT_INC(tbl, forced_gc_runs);
127 write_lock_bh(&tbl->lock);
128 nht = rcu_dereference_protected(tbl->nht,
129 lockdep_is_held(&tbl->lock));
130 for (i = 0; i < (1 << nht->hash_shift); i++) {
132 struct neighbour __rcu **np;
134 np = &nht->hash_buckets[i];
135 while ((n = rcu_dereference_protected(*np,
136 lockdep_is_held(&tbl->lock))) != NULL) {
137 /* Neighbour record may be discarded if:
138 * - nobody refers to it.
139 * - it is not permanent
141 write_lock(&n->lock);
142 if (atomic_read(&n->refcnt) == 1 &&
143 !(n->nud_state & NUD_PERMANENT)) {
144 rcu_assign_pointer(*np,
145 rcu_dereference_protected(n->next,
146 lockdep_is_held(&tbl->lock)));
149 write_unlock(&n->lock);
150 neigh_cleanup_and_release(n);
153 write_unlock(&n->lock);
158 tbl->last_flush = jiffies;
160 write_unlock_bh(&tbl->lock);
165 static void neigh_add_timer(struct neighbour *n, unsigned long when)
168 if (unlikely(mod_timer(&n->timer, when))) {
169 printk("NEIGH: BUG, double timer add, state is %x\n",
175 static int neigh_del_timer(struct neighbour *n)
177 if ((n->nud_state & NUD_IN_TIMER) &&
178 del_timer(&n->timer)) {
185 static void pneigh_queue_purge(struct sk_buff_head *list)
189 while ((skb = skb_dequeue(list)) != NULL) {
195 static void neigh_flush_dev(struct neigh_table *tbl, struct net_device *dev)
198 struct neigh_hash_table *nht;
200 nht = rcu_dereference_protected(tbl->nht,
201 lockdep_is_held(&tbl->lock));
203 for (i = 0; i < (1 << nht->hash_shift); i++) {
205 struct neighbour __rcu **np = &nht->hash_buckets[i];
207 while ((n = rcu_dereference_protected(*np,
208 lockdep_is_held(&tbl->lock))) != NULL) {
209 if (dev && n->dev != dev) {
213 rcu_assign_pointer(*np,
214 rcu_dereference_protected(n->next,
215 lockdep_is_held(&tbl->lock)));
216 write_lock(&n->lock);
220 if (atomic_read(&n->refcnt) != 1) {
221 /* The most unpleasant situation.
222 We must destroy neighbour entry,
223 but someone still uses it.
225 The destroy will be delayed until
226 the last user releases us, but
227 we must kill timers etc. and move
230 __skb_queue_purge(&n->arp_queue);
231 n->arp_queue_len_bytes = 0;
232 n->output = neigh_blackhole;
233 if (n->nud_state & NUD_VALID)
234 n->nud_state = NUD_NOARP;
236 n->nud_state = NUD_NONE;
237 neigh_dbg(2, "neigh %p is stray\n", n);
239 write_unlock(&n->lock);
240 neigh_cleanup_and_release(n);
245 void neigh_changeaddr(struct neigh_table *tbl, struct net_device *dev)
247 write_lock_bh(&tbl->lock);
248 neigh_flush_dev(tbl, dev);
249 write_unlock_bh(&tbl->lock);
251 EXPORT_SYMBOL(neigh_changeaddr);
253 int neigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
255 write_lock_bh(&tbl->lock);
256 neigh_flush_dev(tbl, dev);
257 pneigh_ifdown(tbl, dev);
258 write_unlock_bh(&tbl->lock);
260 del_timer_sync(&tbl->proxy_timer);
261 pneigh_queue_purge(&tbl->proxy_queue);
264 EXPORT_SYMBOL(neigh_ifdown);
266 static struct neighbour *neigh_alloc(struct neigh_table *tbl, struct net_device *dev)
268 struct neighbour *n = NULL;
269 unsigned long now = jiffies;
272 entries = atomic_inc_return(&tbl->entries) - 1;
273 if (entries >= tbl->gc_thresh3 ||
274 (entries >= tbl->gc_thresh2 &&
275 time_after(now, tbl->last_flush + 5 * HZ))) {
276 if (!neigh_forced_gc(tbl) &&
277 entries >= tbl->gc_thresh3) {
278 net_info_ratelimited("%s: neighbor table overflow!\n",
280 NEIGH_CACHE_STAT_INC(tbl, table_fulls);
285 n = kzalloc(tbl->entry_size + dev->neigh_priv_len, GFP_ATOMIC);
289 __skb_queue_head_init(&n->arp_queue);
290 rwlock_init(&n->lock);
291 seqlock_init(&n->ha_lock);
292 n->updated = n->used = now;
293 n->nud_state = NUD_NONE;
294 n->output = neigh_blackhole;
295 seqlock_init(&n->hh.hh_lock);
296 n->parms = neigh_parms_clone(&tbl->parms);
297 setup_timer(&n->timer, neigh_timer_handler, (unsigned long)n);
299 NEIGH_CACHE_STAT_INC(tbl, allocs);
301 atomic_set(&n->refcnt, 1);
307 atomic_dec(&tbl->entries);
311 static void neigh_get_hash_rnd(u32 *x)
313 get_random_bytes(x, sizeof(*x));
317 static struct neigh_hash_table *neigh_hash_alloc(unsigned int shift)
319 size_t size = (1 << shift) * sizeof(struct neighbour *);
320 struct neigh_hash_table *ret;
321 struct neighbour __rcu **buckets;
324 ret = kmalloc(sizeof(*ret), GFP_ATOMIC);
327 if (size <= PAGE_SIZE)
328 buckets = kzalloc(size, GFP_ATOMIC);
330 buckets = (struct neighbour __rcu **)
331 __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
337 ret->hash_buckets = buckets;
338 ret->hash_shift = shift;
339 for (i = 0; i < NEIGH_NUM_HASH_RND; i++)
340 neigh_get_hash_rnd(&ret->hash_rnd[i]);
344 static void neigh_hash_free_rcu(struct rcu_head *head)
346 struct neigh_hash_table *nht = container_of(head,
347 struct neigh_hash_table,
349 size_t size = (1 << nht->hash_shift) * sizeof(struct neighbour *);
350 struct neighbour __rcu **buckets = nht->hash_buckets;
352 if (size <= PAGE_SIZE)
355 free_pages((unsigned long)buckets, get_order(size));
359 static struct neigh_hash_table *neigh_hash_grow(struct neigh_table *tbl,
360 unsigned long new_shift)
362 unsigned int i, hash;
363 struct neigh_hash_table *new_nht, *old_nht;
365 NEIGH_CACHE_STAT_INC(tbl, hash_grows);
367 old_nht = rcu_dereference_protected(tbl->nht,
368 lockdep_is_held(&tbl->lock));
369 new_nht = neigh_hash_alloc(new_shift);
373 for (i = 0; i < (1 << old_nht->hash_shift); i++) {
374 struct neighbour *n, *next;
376 for (n = rcu_dereference_protected(old_nht->hash_buckets[i],
377 lockdep_is_held(&tbl->lock));
380 hash = tbl->hash(n->primary_key, n->dev,
383 hash >>= (32 - new_nht->hash_shift);
384 next = rcu_dereference_protected(n->next,
385 lockdep_is_held(&tbl->lock));
387 rcu_assign_pointer(n->next,
388 rcu_dereference_protected(
389 new_nht->hash_buckets[hash],
390 lockdep_is_held(&tbl->lock)));
391 rcu_assign_pointer(new_nht->hash_buckets[hash], n);
395 rcu_assign_pointer(tbl->nht, new_nht);
396 call_rcu(&old_nht->rcu, neigh_hash_free_rcu);
400 struct neighbour *neigh_lookup(struct neigh_table *tbl, const void *pkey,
401 struct net_device *dev)
405 NEIGH_CACHE_STAT_INC(tbl, lookups);
408 n = __neigh_lookup_noref(tbl, pkey, dev);
410 if (!atomic_inc_not_zero(&n->refcnt))
412 NEIGH_CACHE_STAT_INC(tbl, hits);
415 rcu_read_unlock_bh();
418 EXPORT_SYMBOL(neigh_lookup);
420 struct neighbour *neigh_lookup_nodev(struct neigh_table *tbl, struct net *net,
424 int key_len = tbl->key_len;
426 struct neigh_hash_table *nht;
428 NEIGH_CACHE_STAT_INC(tbl, lookups);
431 nht = rcu_dereference_bh(tbl->nht);
432 hash_val = tbl->hash(pkey, NULL, nht->hash_rnd) >> (32 - nht->hash_shift);
434 for (n = rcu_dereference_bh(nht->hash_buckets[hash_val]);
436 n = rcu_dereference_bh(n->next)) {
437 if (!memcmp(n->primary_key, pkey, key_len) &&
438 net_eq(dev_net(n->dev), net)) {
439 if (!atomic_inc_not_zero(&n->refcnt))
441 NEIGH_CACHE_STAT_INC(tbl, hits);
446 rcu_read_unlock_bh();
449 EXPORT_SYMBOL(neigh_lookup_nodev);
451 struct neighbour *__neigh_create(struct neigh_table *tbl, const void *pkey,
452 struct net_device *dev, bool want_ref)
455 int key_len = tbl->key_len;
457 struct neighbour *n1, *rc, *n = neigh_alloc(tbl, dev);
458 struct neigh_hash_table *nht;
461 rc = ERR_PTR(-ENOBUFS);
465 memcpy(n->primary_key, pkey, key_len);
469 /* Protocol specific setup. */
470 if (tbl->constructor && (error = tbl->constructor(n)) < 0) {
472 goto out_neigh_release;
475 if (dev->netdev_ops->ndo_neigh_construct) {
476 error = dev->netdev_ops->ndo_neigh_construct(n);
479 goto out_neigh_release;
483 /* Device specific setup. */
484 if (n->parms->neigh_setup &&
485 (error = n->parms->neigh_setup(n)) < 0) {
487 goto out_neigh_release;
490 n->confirmed = jiffies - (NEIGH_VAR(n->parms, BASE_REACHABLE_TIME) << 1);
492 write_lock_bh(&tbl->lock);
493 nht = rcu_dereference_protected(tbl->nht,
494 lockdep_is_held(&tbl->lock));
496 if (atomic_read(&tbl->entries) > (1 << nht->hash_shift))
497 nht = neigh_hash_grow(tbl, nht->hash_shift + 1);
499 hash_val = tbl->hash(pkey, dev, nht->hash_rnd) >> (32 - nht->hash_shift);
501 if (n->parms->dead) {
502 rc = ERR_PTR(-EINVAL);
506 for (n1 = rcu_dereference_protected(nht->hash_buckets[hash_val],
507 lockdep_is_held(&tbl->lock));
509 n1 = rcu_dereference_protected(n1->next,
510 lockdep_is_held(&tbl->lock))) {
511 if (dev == n1->dev && !memcmp(n1->primary_key, pkey, key_len)) {
522 rcu_assign_pointer(n->next,
523 rcu_dereference_protected(nht->hash_buckets[hash_val],
524 lockdep_is_held(&tbl->lock)));
525 rcu_assign_pointer(nht->hash_buckets[hash_val], n);
526 write_unlock_bh(&tbl->lock);
527 neigh_dbg(2, "neigh %p is created\n", n);
532 write_unlock_bh(&tbl->lock);
537 EXPORT_SYMBOL(__neigh_create);
539 static u32 pneigh_hash(const void *pkey, int key_len)
541 u32 hash_val = *(u32 *)(pkey + key_len - 4);
542 hash_val ^= (hash_val >> 16);
543 hash_val ^= hash_val >> 8;
544 hash_val ^= hash_val >> 4;
545 hash_val &= PNEIGH_HASHMASK;
549 static struct pneigh_entry *__pneigh_lookup_1(struct pneigh_entry *n,
553 struct net_device *dev)
556 if (!memcmp(n->key, pkey, key_len) &&
557 net_eq(pneigh_net(n), net) &&
558 (n->dev == dev || !n->dev))
565 struct pneigh_entry *__pneigh_lookup(struct neigh_table *tbl,
566 struct net *net, const void *pkey, struct net_device *dev)
568 int key_len = tbl->key_len;
569 u32 hash_val = pneigh_hash(pkey, key_len);
571 return __pneigh_lookup_1(tbl->phash_buckets[hash_val],
572 net, pkey, key_len, dev);
574 EXPORT_SYMBOL_GPL(__pneigh_lookup);
576 struct pneigh_entry * pneigh_lookup(struct neigh_table *tbl,
577 struct net *net, const void *pkey,
578 struct net_device *dev, int creat)
580 struct pneigh_entry *n;
581 int key_len = tbl->key_len;
582 u32 hash_val = pneigh_hash(pkey, key_len);
584 read_lock_bh(&tbl->lock);
585 n = __pneigh_lookup_1(tbl->phash_buckets[hash_val],
586 net, pkey, key_len, dev);
587 read_unlock_bh(&tbl->lock);
594 n = kmalloc(sizeof(*n) + key_len, GFP_KERNEL);
598 write_pnet(&n->net, net);
599 memcpy(n->key, pkey, key_len);
604 if (tbl->pconstructor && tbl->pconstructor(n)) {
612 write_lock_bh(&tbl->lock);
613 n->next = tbl->phash_buckets[hash_val];
614 tbl->phash_buckets[hash_val] = n;
615 write_unlock_bh(&tbl->lock);
619 EXPORT_SYMBOL(pneigh_lookup);
622 int pneigh_delete(struct neigh_table *tbl, struct net *net, const void *pkey,
623 struct net_device *dev)
625 struct pneigh_entry *n, **np;
626 int key_len = tbl->key_len;
627 u32 hash_val = pneigh_hash(pkey, key_len);
629 write_lock_bh(&tbl->lock);
630 for (np = &tbl->phash_buckets[hash_val]; (n = *np) != NULL;
632 if (!memcmp(n->key, pkey, key_len) && n->dev == dev &&
633 net_eq(pneigh_net(n), net)) {
635 write_unlock_bh(&tbl->lock);
636 if (tbl->pdestructor)
644 write_unlock_bh(&tbl->lock);
648 static int pneigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
650 struct pneigh_entry *n, **np;
653 for (h = 0; h <= PNEIGH_HASHMASK; h++) {
654 np = &tbl->phash_buckets[h];
655 while ((n = *np) != NULL) {
656 if (!dev || n->dev == dev) {
658 if (tbl->pdestructor)
671 static void neigh_parms_destroy(struct neigh_parms *parms);
673 static inline void neigh_parms_put(struct neigh_parms *parms)
675 if (atomic_dec_and_test(&parms->refcnt))
676 neigh_parms_destroy(parms);
680 * neighbour must already be out of the table;
683 void neigh_destroy(struct neighbour *neigh)
685 struct net_device *dev = neigh->dev;
687 NEIGH_CACHE_STAT_INC(neigh->tbl, destroys);
690 pr_warn("Destroying alive neighbour %p\n", neigh);
695 if (neigh_del_timer(neigh))
696 pr_warn("Impossible event\n");
698 write_lock_bh(&neigh->lock);
699 __skb_queue_purge(&neigh->arp_queue);
700 write_unlock_bh(&neigh->lock);
701 neigh->arp_queue_len_bytes = 0;
703 if (dev->netdev_ops->ndo_neigh_destroy)
704 dev->netdev_ops->ndo_neigh_destroy(neigh);
707 neigh_parms_put(neigh->parms);
709 neigh_dbg(2, "neigh %p is destroyed\n", neigh);
711 atomic_dec(&neigh->tbl->entries);
712 kfree_rcu(neigh, rcu);
714 EXPORT_SYMBOL(neigh_destroy);
716 /* Neighbour state is suspicious;
719 Called with write_locked neigh.
721 static void neigh_suspect(struct neighbour *neigh)
723 neigh_dbg(2, "neigh %p is suspected\n", neigh);
725 neigh->output = neigh->ops->output;
728 /* Neighbour state is OK;
731 Called with write_locked neigh.
733 static void neigh_connect(struct neighbour *neigh)
735 neigh_dbg(2, "neigh %p is connected\n", neigh);
737 neigh->output = neigh->ops->connected_output;
740 static void neigh_periodic_work(struct work_struct *work)
742 struct neigh_table *tbl = container_of(work, struct neigh_table, gc_work.work);
744 struct neighbour __rcu **np;
746 struct neigh_hash_table *nht;
748 NEIGH_CACHE_STAT_INC(tbl, periodic_gc_runs);
750 write_lock_bh(&tbl->lock);
751 nht = rcu_dereference_protected(tbl->nht,
752 lockdep_is_held(&tbl->lock));
755 * periodically recompute ReachableTime from random function
758 if (time_after(jiffies, tbl->last_rand + 300 * HZ)) {
759 struct neigh_parms *p;
760 tbl->last_rand = jiffies;
761 list_for_each_entry(p, &tbl->parms_list, list)
763 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
766 if (atomic_read(&tbl->entries) < tbl->gc_thresh1)
769 for (i = 0 ; i < (1 << nht->hash_shift); i++) {
770 np = &nht->hash_buckets[i];
772 while ((n = rcu_dereference_protected(*np,
773 lockdep_is_held(&tbl->lock))) != NULL) {
776 write_lock(&n->lock);
778 state = n->nud_state;
779 if (state & (NUD_PERMANENT | NUD_IN_TIMER)) {
780 write_unlock(&n->lock);
784 if (time_before(n->used, n->confirmed))
785 n->used = n->confirmed;
787 if (atomic_read(&n->refcnt) == 1 &&
788 (state == NUD_FAILED ||
789 time_after(jiffies, n->used + NEIGH_VAR(n->parms, GC_STALETIME)))) {
792 write_unlock(&n->lock);
793 neigh_cleanup_and_release(n);
796 write_unlock(&n->lock);
802 * It's fine to release lock here, even if hash table
803 * grows while we are preempted.
805 write_unlock_bh(&tbl->lock);
807 write_lock_bh(&tbl->lock);
808 nht = rcu_dereference_protected(tbl->nht,
809 lockdep_is_held(&tbl->lock));
812 /* Cycle through all hash buckets every BASE_REACHABLE_TIME/2 ticks.
813 * ARP entry timeouts range from 1/2 BASE_REACHABLE_TIME to 3/2
814 * BASE_REACHABLE_TIME.
816 queue_delayed_work(system_power_efficient_wq, &tbl->gc_work,
817 NEIGH_VAR(&tbl->parms, BASE_REACHABLE_TIME) >> 1);
818 write_unlock_bh(&tbl->lock);
821 static __inline__ int neigh_max_probes(struct neighbour *n)
823 struct neigh_parms *p = n->parms;
824 return NEIGH_VAR(p, UCAST_PROBES) + NEIGH_VAR(p, APP_PROBES) +
825 (n->nud_state & NUD_PROBE ? NEIGH_VAR(p, MCAST_REPROBES) :
826 NEIGH_VAR(p, MCAST_PROBES));
829 static void neigh_invalidate(struct neighbour *neigh)
830 __releases(neigh->lock)
831 __acquires(neigh->lock)
835 NEIGH_CACHE_STAT_INC(neigh->tbl, res_failed);
836 neigh_dbg(2, "neigh %p is failed\n", neigh);
837 neigh->updated = jiffies;
839 /* It is very thin place. report_unreachable is very complicated
840 routine. Particularly, it can hit the same neighbour entry!
842 So that, we try to be accurate and avoid dead loop. --ANK
844 while (neigh->nud_state == NUD_FAILED &&
845 (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
846 write_unlock(&neigh->lock);
847 neigh->ops->error_report(neigh, skb);
848 write_lock(&neigh->lock);
850 __skb_queue_purge(&neigh->arp_queue);
851 neigh->arp_queue_len_bytes = 0;
854 static void neigh_probe(struct neighbour *neigh)
855 __releases(neigh->lock)
857 struct sk_buff *skb = skb_peek_tail(&neigh->arp_queue);
858 /* keep skb alive even if arp_queue overflows */
860 skb = skb_clone(skb, GFP_ATOMIC);
861 write_unlock(&neigh->lock);
862 if (neigh->ops->solicit)
863 neigh->ops->solicit(neigh, skb);
864 atomic_inc(&neigh->probes);
868 /* Called when a timer expires for a neighbour entry. */
870 static void neigh_timer_handler(unsigned long arg)
872 unsigned long now, next;
873 struct neighbour *neigh = (struct neighbour *)arg;
877 write_lock(&neigh->lock);
879 state = neigh->nud_state;
883 if (!(state & NUD_IN_TIMER))
886 if (state & NUD_REACHABLE) {
887 if (time_before_eq(now,
888 neigh->confirmed + neigh->parms->reachable_time)) {
889 neigh_dbg(2, "neigh %p is still alive\n", neigh);
890 next = neigh->confirmed + neigh->parms->reachable_time;
891 } else if (time_before_eq(now,
893 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME))) {
894 neigh_dbg(2, "neigh %p is delayed\n", neigh);
895 neigh->nud_state = NUD_DELAY;
896 neigh->updated = jiffies;
897 neigh_suspect(neigh);
898 next = now + NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME);
900 neigh_dbg(2, "neigh %p is suspected\n", neigh);
901 neigh->nud_state = NUD_STALE;
902 neigh->updated = jiffies;
903 neigh_suspect(neigh);
906 } else if (state & NUD_DELAY) {
907 if (time_before_eq(now,
909 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME))) {
910 neigh_dbg(2, "neigh %p is now reachable\n", neigh);
911 neigh->nud_state = NUD_REACHABLE;
912 neigh->updated = jiffies;
913 neigh_connect(neigh);
915 next = neigh->confirmed + neigh->parms->reachable_time;
917 neigh_dbg(2, "neigh %p is probed\n", neigh);
918 neigh->nud_state = NUD_PROBE;
919 neigh->updated = jiffies;
920 atomic_set(&neigh->probes, 0);
922 next = now + NEIGH_VAR(neigh->parms, RETRANS_TIME);
925 /* NUD_PROBE|NUD_INCOMPLETE */
926 next = now + NEIGH_VAR(neigh->parms, RETRANS_TIME);
929 if ((neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) &&
930 atomic_read(&neigh->probes) >= neigh_max_probes(neigh)) {
931 neigh->nud_state = NUD_FAILED;
933 neigh_invalidate(neigh);
937 if (neigh->nud_state & NUD_IN_TIMER) {
938 if (time_before(next, jiffies + HZ/2))
939 next = jiffies + HZ/2;
940 if (!mod_timer(&neigh->timer, next))
943 if (neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) {
947 write_unlock(&neigh->lock);
951 neigh_update_notify(neigh);
953 neigh_release(neigh);
956 int __neigh_event_send(struct neighbour *neigh, struct sk_buff *skb)
959 bool immediate_probe = false;
961 write_lock_bh(&neigh->lock);
964 if (neigh->nud_state & (NUD_CONNECTED | NUD_DELAY | NUD_PROBE))
969 if (!(neigh->nud_state & (NUD_STALE | NUD_INCOMPLETE))) {
970 if (NEIGH_VAR(neigh->parms, MCAST_PROBES) +
971 NEIGH_VAR(neigh->parms, APP_PROBES)) {
972 unsigned long next, now = jiffies;
974 atomic_set(&neigh->probes,
975 NEIGH_VAR(neigh->parms, UCAST_PROBES));
976 neigh->nud_state = NUD_INCOMPLETE;
977 neigh->updated = now;
978 next = now + max(NEIGH_VAR(neigh->parms, RETRANS_TIME),
980 neigh_add_timer(neigh, next);
981 immediate_probe = true;
983 neigh->nud_state = NUD_FAILED;
984 neigh->updated = jiffies;
985 write_unlock_bh(&neigh->lock);
990 } else if (neigh->nud_state & NUD_STALE) {
991 neigh_dbg(2, "neigh %p is delayed\n", neigh);
992 neigh->nud_state = NUD_DELAY;
993 neigh->updated = jiffies;
994 neigh_add_timer(neigh, jiffies +
995 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME));
998 if (neigh->nud_state == NUD_INCOMPLETE) {
1000 while (neigh->arp_queue_len_bytes + skb->truesize >
1001 NEIGH_VAR(neigh->parms, QUEUE_LEN_BYTES)) {
1002 struct sk_buff *buff;
1004 buff = __skb_dequeue(&neigh->arp_queue);
1007 neigh->arp_queue_len_bytes -= buff->truesize;
1009 NEIGH_CACHE_STAT_INC(neigh->tbl, unres_discards);
1012 __skb_queue_tail(&neigh->arp_queue, skb);
1013 neigh->arp_queue_len_bytes += skb->truesize;
1018 if (immediate_probe)
1021 write_unlock(&neigh->lock);
1026 if (neigh->nud_state & NUD_STALE)
1028 write_unlock_bh(&neigh->lock);
1032 EXPORT_SYMBOL(__neigh_event_send);
1034 static void neigh_update_hhs(struct neighbour *neigh)
1036 struct hh_cache *hh;
1037 void (*update)(struct hh_cache*, const struct net_device*, const unsigned char *)
1040 if (neigh->dev->header_ops)
1041 update = neigh->dev->header_ops->cache_update;
1046 write_seqlock_bh(&hh->hh_lock);
1047 update(hh, neigh->dev, neigh->ha);
1048 write_sequnlock_bh(&hh->hh_lock);
1055 /* Generic update routine.
1056 -- lladdr is new lladdr or NULL, if it is not supplied.
1057 -- new is new state.
1059 NEIGH_UPDATE_F_OVERRIDE allows to override existing lladdr,
1061 NEIGH_UPDATE_F_WEAK_OVERRIDE will suspect existing "connected"
1062 lladdr instead of overriding it
1064 It also allows to retain current state
1065 if lladdr is unchanged.
1066 NEIGH_UPDATE_F_ADMIN means that the change is administrative.
1068 NEIGH_UPDATE_F_OVERRIDE_ISROUTER allows to override existing
1070 NEIGH_UPDATE_F_ISROUTER indicates if the neighbour is known as
1073 Caller MUST hold reference count on the entry.
1076 int neigh_update(struct neighbour *neigh, const u8 *lladdr, u8 new,
1082 struct net_device *dev;
1083 int update_isrouter = 0;
1085 write_lock_bh(&neigh->lock);
1088 old = neigh->nud_state;
1091 if (!(flags & NEIGH_UPDATE_F_ADMIN) &&
1092 (old & (NUD_NOARP | NUD_PERMANENT)))
1097 if (!(new & NUD_VALID)) {
1098 neigh_del_timer(neigh);
1099 if (old & NUD_CONNECTED)
1100 neigh_suspect(neigh);
1101 neigh->nud_state = new;
1103 notify = old & NUD_VALID;
1104 if ((old & (NUD_INCOMPLETE | NUD_PROBE)) &&
1105 (new & NUD_FAILED)) {
1106 neigh_invalidate(neigh);
1112 /* Compare new lladdr with cached one */
1113 if (!dev->addr_len) {
1114 /* First case: device needs no address. */
1116 } else if (lladdr) {
1117 /* The second case: if something is already cached
1118 and a new address is proposed:
1120 - if they are different, check override flag
1122 if ((old & NUD_VALID) &&
1123 !memcmp(lladdr, neigh->ha, dev->addr_len))
1126 /* No address is supplied; if we know something,
1127 use it, otherwise discard the request.
1130 if (!(old & NUD_VALID))
1135 if (new & NUD_CONNECTED)
1136 neigh->confirmed = jiffies;
1137 neigh->updated = jiffies;
1139 /* If entry was valid and address is not changed,
1140 do not change entry state, if new one is STALE.
1143 update_isrouter = flags & NEIGH_UPDATE_F_OVERRIDE_ISROUTER;
1144 if (old & NUD_VALID) {
1145 if (lladdr != neigh->ha && !(flags & NEIGH_UPDATE_F_OVERRIDE)) {
1146 update_isrouter = 0;
1147 if ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) &&
1148 (old & NUD_CONNECTED)) {
1154 if (lladdr == neigh->ha && new == NUD_STALE &&
1155 ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) ||
1156 (old & NUD_CONNECTED))
1163 neigh_del_timer(neigh);
1164 if (new & NUD_PROBE)
1165 atomic_set(&neigh->probes, 0);
1166 if (new & NUD_IN_TIMER)
1167 neigh_add_timer(neigh, (jiffies +
1168 ((new & NUD_REACHABLE) ?
1169 neigh->parms->reachable_time :
1171 neigh->nud_state = new;
1175 if (lladdr != neigh->ha) {
1176 write_seqlock(&neigh->ha_lock);
1177 memcpy(&neigh->ha, lladdr, dev->addr_len);
1178 write_sequnlock(&neigh->ha_lock);
1179 neigh_update_hhs(neigh);
1180 if (!(new & NUD_CONNECTED))
1181 neigh->confirmed = jiffies -
1182 (NEIGH_VAR(neigh->parms, BASE_REACHABLE_TIME) << 1);
1187 if (new & NUD_CONNECTED)
1188 neigh_connect(neigh);
1190 neigh_suspect(neigh);
1191 if (!(old & NUD_VALID)) {
1192 struct sk_buff *skb;
1194 /* Again: avoid dead loop if something went wrong */
1196 while (neigh->nud_state & NUD_VALID &&
1197 (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
1198 struct dst_entry *dst = skb_dst(skb);
1199 struct neighbour *n2, *n1 = neigh;
1200 write_unlock_bh(&neigh->lock);
1204 /* Why not just use 'neigh' as-is? The problem is that
1205 * things such as shaper, eql, and sch_teql can end up
1206 * using alternative, different, neigh objects to output
1207 * the packet in the output path. So what we need to do
1208 * here is re-lookup the top-level neigh in the path so
1209 * we can reinject the packet there.
1213 n2 = dst_neigh_lookup_skb(dst, skb);
1217 n1->output(n1, skb);
1222 write_lock_bh(&neigh->lock);
1224 __skb_queue_purge(&neigh->arp_queue);
1225 neigh->arp_queue_len_bytes = 0;
1228 if (update_isrouter) {
1229 neigh->flags = (flags & NEIGH_UPDATE_F_ISROUTER) ?
1230 (neigh->flags | NTF_ROUTER) :
1231 (neigh->flags & ~NTF_ROUTER);
1233 write_unlock_bh(&neigh->lock);
1236 neigh_update_notify(neigh);
1240 EXPORT_SYMBOL(neigh_update);
1242 /* Update the neigh to listen temporarily for probe responses, even if it is
1243 * in a NUD_FAILED state. The caller has to hold neigh->lock for writing.
1245 void __neigh_set_probe_once(struct neighbour *neigh)
1249 neigh->updated = jiffies;
1250 if (!(neigh->nud_state & NUD_FAILED))
1252 neigh->nud_state = NUD_INCOMPLETE;
1253 atomic_set(&neigh->probes, neigh_max_probes(neigh));
1254 neigh_add_timer(neigh,
1255 jiffies + NEIGH_VAR(neigh->parms, RETRANS_TIME));
1257 EXPORT_SYMBOL(__neigh_set_probe_once);
1259 struct neighbour *neigh_event_ns(struct neigh_table *tbl,
1260 u8 *lladdr, void *saddr,
1261 struct net_device *dev)
1263 struct neighbour *neigh = __neigh_lookup(tbl, saddr, dev,
1264 lladdr || !dev->addr_len);
1266 neigh_update(neigh, lladdr, NUD_STALE,
1267 NEIGH_UPDATE_F_OVERRIDE);
1270 EXPORT_SYMBOL(neigh_event_ns);
1272 /* called with read_lock_bh(&n->lock); */
1273 static void neigh_hh_init(struct neighbour *n)
1275 struct net_device *dev = n->dev;
1276 __be16 prot = n->tbl->protocol;
1277 struct hh_cache *hh = &n->hh;
1279 write_lock_bh(&n->lock);
1281 /* Only one thread can come in here and initialize the
1285 dev->header_ops->cache(n, hh, prot);
1287 write_unlock_bh(&n->lock);
1290 /* Slow and careful. */
1292 int neigh_resolve_output(struct neighbour *neigh, struct sk_buff *skb)
1296 if (!neigh_event_send(neigh, skb)) {
1298 struct net_device *dev = neigh->dev;
1301 if (dev->header_ops->cache && !neigh->hh.hh_len)
1302 neigh_hh_init(neigh);
1305 __skb_pull(skb, skb_network_offset(skb));
1306 seq = read_seqbegin(&neigh->ha_lock);
1307 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1308 neigh->ha, NULL, skb->len);
1309 } while (read_seqretry(&neigh->ha_lock, seq));
1312 rc = dev_queue_xmit(skb);
1323 EXPORT_SYMBOL(neigh_resolve_output);
1325 /* As fast as possible without hh cache */
1327 int neigh_connected_output(struct neighbour *neigh, struct sk_buff *skb)
1329 struct net_device *dev = neigh->dev;
1334 __skb_pull(skb, skb_network_offset(skb));
1335 seq = read_seqbegin(&neigh->ha_lock);
1336 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1337 neigh->ha, NULL, skb->len);
1338 } while (read_seqretry(&neigh->ha_lock, seq));
1341 err = dev_queue_xmit(skb);
1348 EXPORT_SYMBOL(neigh_connected_output);
1350 int neigh_direct_output(struct neighbour *neigh, struct sk_buff *skb)
1352 return dev_queue_xmit(skb);
1354 EXPORT_SYMBOL(neigh_direct_output);
1356 static void neigh_proxy_process(unsigned long arg)
1358 struct neigh_table *tbl = (struct neigh_table *)arg;
1359 long sched_next = 0;
1360 unsigned long now = jiffies;
1361 struct sk_buff *skb, *n;
1363 spin_lock(&tbl->proxy_queue.lock);
1365 skb_queue_walk_safe(&tbl->proxy_queue, skb, n) {
1366 long tdif = NEIGH_CB(skb)->sched_next - now;
1369 struct net_device *dev = skb->dev;
1371 __skb_unlink(skb, &tbl->proxy_queue);
1372 if (tbl->proxy_redo && netif_running(dev)) {
1374 tbl->proxy_redo(skb);
1381 } else if (!sched_next || tdif < sched_next)
1384 del_timer(&tbl->proxy_timer);
1386 mod_timer(&tbl->proxy_timer, jiffies + sched_next);
1387 spin_unlock(&tbl->proxy_queue.lock);
1390 void pneigh_enqueue(struct neigh_table *tbl, struct neigh_parms *p,
1391 struct sk_buff *skb)
1393 unsigned long now = jiffies;
1395 unsigned long sched_next = now + (prandom_u32() %
1396 NEIGH_VAR(p, PROXY_DELAY));
1398 if (tbl->proxy_queue.qlen > NEIGH_VAR(p, PROXY_QLEN)) {
1403 NEIGH_CB(skb)->sched_next = sched_next;
1404 NEIGH_CB(skb)->flags |= LOCALLY_ENQUEUED;
1406 spin_lock(&tbl->proxy_queue.lock);
1407 if (del_timer(&tbl->proxy_timer)) {
1408 if (time_before(tbl->proxy_timer.expires, sched_next))
1409 sched_next = tbl->proxy_timer.expires;
1413 __skb_queue_tail(&tbl->proxy_queue, skb);
1414 mod_timer(&tbl->proxy_timer, sched_next);
1415 spin_unlock(&tbl->proxy_queue.lock);
1417 EXPORT_SYMBOL(pneigh_enqueue);
1419 static inline struct neigh_parms *lookup_neigh_parms(struct neigh_table *tbl,
1420 struct net *net, int ifindex)
1422 struct neigh_parms *p;
1424 list_for_each_entry(p, &tbl->parms_list, list) {
1425 if ((p->dev && p->dev->ifindex == ifindex && net_eq(neigh_parms_net(p), net)) ||
1426 (!p->dev && !ifindex && net_eq(net, &init_net)))
1433 struct neigh_parms *neigh_parms_alloc(struct net_device *dev,
1434 struct neigh_table *tbl)
1436 struct neigh_parms *p;
1437 struct net *net = dev_net(dev);
1438 const struct net_device_ops *ops = dev->netdev_ops;
1440 p = kmemdup(&tbl->parms, sizeof(*p), GFP_KERNEL);
1443 atomic_set(&p->refcnt, 1);
1445 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
1448 write_pnet(&p->net, net);
1449 p->sysctl_table = NULL;
1451 if (ops->ndo_neigh_setup && ops->ndo_neigh_setup(dev, p)) {
1457 write_lock_bh(&tbl->lock);
1458 list_add(&p->list, &tbl->parms.list);
1459 write_unlock_bh(&tbl->lock);
1461 neigh_parms_data_state_cleanall(p);
1465 EXPORT_SYMBOL(neigh_parms_alloc);
1467 static void neigh_rcu_free_parms(struct rcu_head *head)
1469 struct neigh_parms *parms =
1470 container_of(head, struct neigh_parms, rcu_head);
1472 neigh_parms_put(parms);
1475 void neigh_parms_release(struct neigh_table *tbl, struct neigh_parms *parms)
1477 if (!parms || parms == &tbl->parms)
1479 write_lock_bh(&tbl->lock);
1480 list_del(&parms->list);
1482 write_unlock_bh(&tbl->lock);
1484 dev_put(parms->dev);
1485 call_rcu(&parms->rcu_head, neigh_rcu_free_parms);
1487 EXPORT_SYMBOL(neigh_parms_release);
1489 static void neigh_parms_destroy(struct neigh_parms *parms)
1494 static struct lock_class_key neigh_table_proxy_queue_class;
1496 static struct neigh_table *neigh_tables[NEIGH_NR_TABLES] __read_mostly;
1498 void neigh_table_init(int index, struct neigh_table *tbl)
1500 unsigned long now = jiffies;
1501 unsigned long phsize;
1503 INIT_LIST_HEAD(&tbl->parms_list);
1504 list_add(&tbl->parms.list, &tbl->parms_list);
1505 write_pnet(&tbl->parms.net, &init_net);
1506 atomic_set(&tbl->parms.refcnt, 1);
1507 tbl->parms.reachable_time =
1508 neigh_rand_reach_time(NEIGH_VAR(&tbl->parms, BASE_REACHABLE_TIME));
1510 tbl->stats = alloc_percpu(struct neigh_statistics);
1512 panic("cannot create neighbour cache statistics");
1514 #ifdef CONFIG_PROC_FS
1515 if (!proc_create_data(tbl->id, 0, init_net.proc_net_stat,
1516 &neigh_stat_seq_fops, tbl))
1517 panic("cannot create neighbour proc dir entry");
1520 RCU_INIT_POINTER(tbl->nht, neigh_hash_alloc(3));
1522 phsize = (PNEIGH_HASHMASK + 1) * sizeof(struct pneigh_entry *);
1523 tbl->phash_buckets = kzalloc(phsize, GFP_KERNEL);
1525 if (!tbl->nht || !tbl->phash_buckets)
1526 panic("cannot allocate neighbour cache hashes");
1528 if (!tbl->entry_size)
1529 tbl->entry_size = ALIGN(offsetof(struct neighbour, primary_key) +
1530 tbl->key_len, NEIGH_PRIV_ALIGN);
1532 WARN_ON(tbl->entry_size % NEIGH_PRIV_ALIGN);
1534 rwlock_init(&tbl->lock);
1535 INIT_DEFERRABLE_WORK(&tbl->gc_work, neigh_periodic_work);
1536 queue_delayed_work(system_power_efficient_wq, &tbl->gc_work,
1537 tbl->parms.reachable_time);
1538 setup_timer(&tbl->proxy_timer, neigh_proxy_process, (unsigned long)tbl);
1539 skb_queue_head_init_class(&tbl->proxy_queue,
1540 &neigh_table_proxy_queue_class);
1542 tbl->last_flush = now;
1543 tbl->last_rand = now + tbl->parms.reachable_time * 20;
1545 neigh_tables[index] = tbl;
1547 EXPORT_SYMBOL(neigh_table_init);
1549 int neigh_table_clear(int index, struct neigh_table *tbl)
1551 neigh_tables[index] = NULL;
1552 /* It is not clean... Fix it to unload IPv6 module safely */
1553 cancel_delayed_work_sync(&tbl->gc_work);
1554 del_timer_sync(&tbl->proxy_timer);
1555 pneigh_queue_purge(&tbl->proxy_queue);
1556 neigh_ifdown(tbl, NULL);
1557 if (atomic_read(&tbl->entries))
1558 pr_crit("neighbour leakage\n");
1560 call_rcu(&rcu_dereference_protected(tbl->nht, 1)->rcu,
1561 neigh_hash_free_rcu);
1564 kfree(tbl->phash_buckets);
1565 tbl->phash_buckets = NULL;
1567 remove_proc_entry(tbl->id, init_net.proc_net_stat);
1569 free_percpu(tbl->stats);
1574 EXPORT_SYMBOL(neigh_table_clear);
1576 static struct neigh_table *neigh_find_table(int family)
1578 struct neigh_table *tbl = NULL;
1582 tbl = neigh_tables[NEIGH_ARP_TABLE];
1585 tbl = neigh_tables[NEIGH_ND_TABLE];
1588 tbl = neigh_tables[NEIGH_DN_TABLE];
1595 static int neigh_delete(struct sk_buff *skb, struct nlmsghdr *nlh)
1597 struct net *net = sock_net(skb->sk);
1599 struct nlattr *dst_attr;
1600 struct neigh_table *tbl;
1601 struct neighbour *neigh;
1602 struct net_device *dev = NULL;
1606 if (nlmsg_len(nlh) < sizeof(*ndm))
1609 dst_attr = nlmsg_find_attr(nlh, sizeof(*ndm), NDA_DST);
1610 if (dst_attr == NULL)
1613 ndm = nlmsg_data(nlh);
1614 if (ndm->ndm_ifindex) {
1615 dev = __dev_get_by_index(net, ndm->ndm_ifindex);
1622 tbl = neigh_find_table(ndm->ndm_family);
1624 return -EAFNOSUPPORT;
1626 if (nla_len(dst_attr) < tbl->key_len)
1629 if (ndm->ndm_flags & NTF_PROXY) {
1630 err = pneigh_delete(tbl, net, nla_data(dst_attr), dev);
1637 neigh = neigh_lookup(tbl, nla_data(dst_attr), dev);
1638 if (neigh == NULL) {
1643 err = neigh_update(neigh, NULL, NUD_FAILED,
1644 NEIGH_UPDATE_F_OVERRIDE |
1645 NEIGH_UPDATE_F_ADMIN);
1646 neigh_release(neigh);
1652 static int neigh_add(struct sk_buff *skb, struct nlmsghdr *nlh)
1654 int flags = NEIGH_UPDATE_F_ADMIN | NEIGH_UPDATE_F_OVERRIDE;
1655 struct net *net = sock_net(skb->sk);
1657 struct nlattr *tb[NDA_MAX+1];
1658 struct neigh_table *tbl;
1659 struct net_device *dev = NULL;
1660 struct neighbour *neigh;
1665 err = nlmsg_parse(nlh, sizeof(*ndm), tb, NDA_MAX, NULL);
1670 if (tb[NDA_DST] == NULL)
1673 ndm = nlmsg_data(nlh);
1674 if (ndm->ndm_ifindex) {
1675 dev = __dev_get_by_index(net, ndm->ndm_ifindex);
1681 if (tb[NDA_LLADDR] && nla_len(tb[NDA_LLADDR]) < dev->addr_len)
1685 tbl = neigh_find_table(ndm->ndm_family);
1687 return -EAFNOSUPPORT;
1689 if (nla_len(tb[NDA_DST]) < tbl->key_len)
1691 dst = nla_data(tb[NDA_DST]);
1692 lladdr = tb[NDA_LLADDR] ? nla_data(tb[NDA_LLADDR]) : NULL;
1694 if (ndm->ndm_flags & NTF_PROXY) {
1695 struct pneigh_entry *pn;
1698 pn = pneigh_lookup(tbl, net, dst, dev, 1);
1700 pn->flags = ndm->ndm_flags;
1709 neigh = neigh_lookup(tbl, dst, dev);
1710 if (neigh == NULL) {
1711 if (!(nlh->nlmsg_flags & NLM_F_CREATE)) {
1716 neigh = __neigh_lookup_errno(tbl, dst, dev);
1717 if (IS_ERR(neigh)) {
1718 err = PTR_ERR(neigh);
1722 if (nlh->nlmsg_flags & NLM_F_EXCL) {
1724 neigh_release(neigh);
1728 if (!(nlh->nlmsg_flags & NLM_F_REPLACE))
1729 flags &= ~NEIGH_UPDATE_F_OVERRIDE;
1732 if (ndm->ndm_flags & NTF_USE) {
1733 neigh_event_send(neigh, NULL);
1736 err = neigh_update(neigh, lladdr, ndm->ndm_state, flags);
1737 neigh_release(neigh);
1743 static int neightbl_fill_parms(struct sk_buff *skb, struct neigh_parms *parms)
1745 struct nlattr *nest;
1747 nest = nla_nest_start(skb, NDTA_PARMS);
1752 nla_put_u32(skb, NDTPA_IFINDEX, parms->dev->ifindex)) ||
1753 nla_put_u32(skb, NDTPA_REFCNT, atomic_read(&parms->refcnt)) ||
1754 nla_put_u32(skb, NDTPA_QUEUE_LENBYTES,
1755 NEIGH_VAR(parms, QUEUE_LEN_BYTES)) ||
1756 /* approximative value for deprecated QUEUE_LEN (in packets) */
1757 nla_put_u32(skb, NDTPA_QUEUE_LEN,
1758 NEIGH_VAR(parms, QUEUE_LEN_BYTES) / SKB_TRUESIZE(ETH_FRAME_LEN)) ||
1759 nla_put_u32(skb, NDTPA_PROXY_QLEN, NEIGH_VAR(parms, PROXY_QLEN)) ||
1760 nla_put_u32(skb, NDTPA_APP_PROBES, NEIGH_VAR(parms, APP_PROBES)) ||
1761 nla_put_u32(skb, NDTPA_UCAST_PROBES,
1762 NEIGH_VAR(parms, UCAST_PROBES)) ||
1763 nla_put_u32(skb, NDTPA_MCAST_PROBES,
1764 NEIGH_VAR(parms, MCAST_PROBES)) ||
1765 nla_put_u32(skb, NDTPA_MCAST_REPROBES,
1766 NEIGH_VAR(parms, MCAST_REPROBES)) ||
1767 nla_put_msecs(skb, NDTPA_REACHABLE_TIME, parms->reachable_time) ||
1768 nla_put_msecs(skb, NDTPA_BASE_REACHABLE_TIME,
1769 NEIGH_VAR(parms, BASE_REACHABLE_TIME)) ||
1770 nla_put_msecs(skb, NDTPA_GC_STALETIME,
1771 NEIGH_VAR(parms, GC_STALETIME)) ||
1772 nla_put_msecs(skb, NDTPA_DELAY_PROBE_TIME,
1773 NEIGH_VAR(parms, DELAY_PROBE_TIME)) ||
1774 nla_put_msecs(skb, NDTPA_RETRANS_TIME,
1775 NEIGH_VAR(parms, RETRANS_TIME)) ||
1776 nla_put_msecs(skb, NDTPA_ANYCAST_DELAY,
1777 NEIGH_VAR(parms, ANYCAST_DELAY)) ||
1778 nla_put_msecs(skb, NDTPA_PROXY_DELAY,
1779 NEIGH_VAR(parms, PROXY_DELAY)) ||
1780 nla_put_msecs(skb, NDTPA_LOCKTIME,
1781 NEIGH_VAR(parms, LOCKTIME)))
1782 goto nla_put_failure;
1783 return nla_nest_end(skb, nest);
1786 nla_nest_cancel(skb, nest);
1790 static int neightbl_fill_info(struct sk_buff *skb, struct neigh_table *tbl,
1791 u32 pid, u32 seq, int type, int flags)
1793 struct nlmsghdr *nlh;
1794 struct ndtmsg *ndtmsg;
1796 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
1800 ndtmsg = nlmsg_data(nlh);
1802 read_lock_bh(&tbl->lock);
1803 ndtmsg->ndtm_family = tbl->family;
1804 ndtmsg->ndtm_pad1 = 0;
1805 ndtmsg->ndtm_pad2 = 0;
1807 if (nla_put_string(skb, NDTA_NAME, tbl->id) ||
1808 nla_put_msecs(skb, NDTA_GC_INTERVAL, tbl->gc_interval) ||
1809 nla_put_u32(skb, NDTA_THRESH1, tbl->gc_thresh1) ||
1810 nla_put_u32(skb, NDTA_THRESH2, tbl->gc_thresh2) ||
1811 nla_put_u32(skb, NDTA_THRESH3, tbl->gc_thresh3))
1812 goto nla_put_failure;
1814 unsigned long now = jiffies;
1815 unsigned int flush_delta = now - tbl->last_flush;
1816 unsigned int rand_delta = now - tbl->last_rand;
1817 struct neigh_hash_table *nht;
1818 struct ndt_config ndc = {
1819 .ndtc_key_len = tbl->key_len,
1820 .ndtc_entry_size = tbl->entry_size,
1821 .ndtc_entries = atomic_read(&tbl->entries),
1822 .ndtc_last_flush = jiffies_to_msecs(flush_delta),
1823 .ndtc_last_rand = jiffies_to_msecs(rand_delta),
1824 .ndtc_proxy_qlen = tbl->proxy_queue.qlen,
1828 nht = rcu_dereference_bh(tbl->nht);
1829 ndc.ndtc_hash_rnd = nht->hash_rnd[0];
1830 ndc.ndtc_hash_mask = ((1 << nht->hash_shift) - 1);
1831 rcu_read_unlock_bh();
1833 if (nla_put(skb, NDTA_CONFIG, sizeof(ndc), &ndc))
1834 goto nla_put_failure;
1839 struct ndt_stats ndst;
1841 memset(&ndst, 0, sizeof(ndst));
1843 for_each_possible_cpu(cpu) {
1844 struct neigh_statistics *st;
1846 st = per_cpu_ptr(tbl->stats, cpu);
1847 ndst.ndts_allocs += st->allocs;
1848 ndst.ndts_destroys += st->destroys;
1849 ndst.ndts_hash_grows += st->hash_grows;
1850 ndst.ndts_res_failed += st->res_failed;
1851 ndst.ndts_lookups += st->lookups;
1852 ndst.ndts_hits += st->hits;
1853 ndst.ndts_rcv_probes_mcast += st->rcv_probes_mcast;
1854 ndst.ndts_rcv_probes_ucast += st->rcv_probes_ucast;
1855 ndst.ndts_periodic_gc_runs += st->periodic_gc_runs;
1856 ndst.ndts_forced_gc_runs += st->forced_gc_runs;
1857 ndst.ndts_table_fulls += st->table_fulls;
1860 if (nla_put(skb, NDTA_STATS, sizeof(ndst), &ndst))
1861 goto nla_put_failure;
1864 BUG_ON(tbl->parms.dev);
1865 if (neightbl_fill_parms(skb, &tbl->parms) < 0)
1866 goto nla_put_failure;
1868 read_unlock_bh(&tbl->lock);
1869 nlmsg_end(skb, nlh);
1873 read_unlock_bh(&tbl->lock);
1874 nlmsg_cancel(skb, nlh);
1878 static int neightbl_fill_param_info(struct sk_buff *skb,
1879 struct neigh_table *tbl,
1880 struct neigh_parms *parms,
1881 u32 pid, u32 seq, int type,
1884 struct ndtmsg *ndtmsg;
1885 struct nlmsghdr *nlh;
1887 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
1891 ndtmsg = nlmsg_data(nlh);
1893 read_lock_bh(&tbl->lock);
1894 ndtmsg->ndtm_family = tbl->family;
1895 ndtmsg->ndtm_pad1 = 0;
1896 ndtmsg->ndtm_pad2 = 0;
1898 if (nla_put_string(skb, NDTA_NAME, tbl->id) < 0 ||
1899 neightbl_fill_parms(skb, parms) < 0)
1902 read_unlock_bh(&tbl->lock);
1903 nlmsg_end(skb, nlh);
1906 read_unlock_bh(&tbl->lock);
1907 nlmsg_cancel(skb, nlh);
1911 static const struct nla_policy nl_neightbl_policy[NDTA_MAX+1] = {
1912 [NDTA_NAME] = { .type = NLA_STRING },
1913 [NDTA_THRESH1] = { .type = NLA_U32 },
1914 [NDTA_THRESH2] = { .type = NLA_U32 },
1915 [NDTA_THRESH3] = { .type = NLA_U32 },
1916 [NDTA_GC_INTERVAL] = { .type = NLA_U64 },
1917 [NDTA_PARMS] = { .type = NLA_NESTED },
1920 static const struct nla_policy nl_ntbl_parm_policy[NDTPA_MAX+1] = {
1921 [NDTPA_IFINDEX] = { .type = NLA_U32 },
1922 [NDTPA_QUEUE_LEN] = { .type = NLA_U32 },
1923 [NDTPA_PROXY_QLEN] = { .type = NLA_U32 },
1924 [NDTPA_APP_PROBES] = { .type = NLA_U32 },
1925 [NDTPA_UCAST_PROBES] = { .type = NLA_U32 },
1926 [NDTPA_MCAST_PROBES] = { .type = NLA_U32 },
1927 [NDTPA_MCAST_REPROBES] = { .type = NLA_U32 },
1928 [NDTPA_BASE_REACHABLE_TIME] = { .type = NLA_U64 },
1929 [NDTPA_GC_STALETIME] = { .type = NLA_U64 },
1930 [NDTPA_DELAY_PROBE_TIME] = { .type = NLA_U64 },
1931 [NDTPA_RETRANS_TIME] = { .type = NLA_U64 },
1932 [NDTPA_ANYCAST_DELAY] = { .type = NLA_U64 },
1933 [NDTPA_PROXY_DELAY] = { .type = NLA_U64 },
1934 [NDTPA_LOCKTIME] = { .type = NLA_U64 },
1937 static int neightbl_set(struct sk_buff *skb, struct nlmsghdr *nlh)
1939 struct net *net = sock_net(skb->sk);
1940 struct neigh_table *tbl;
1941 struct ndtmsg *ndtmsg;
1942 struct nlattr *tb[NDTA_MAX+1];
1946 err = nlmsg_parse(nlh, sizeof(*ndtmsg), tb, NDTA_MAX,
1947 nl_neightbl_policy);
1951 if (tb[NDTA_NAME] == NULL) {
1956 ndtmsg = nlmsg_data(nlh);
1958 for (tidx = 0; tidx < NEIGH_NR_TABLES; tidx++) {
1959 tbl = neigh_tables[tidx];
1962 if (ndtmsg->ndtm_family && tbl->family != ndtmsg->ndtm_family)
1964 if (nla_strcmp(tb[NDTA_NAME], tbl->id) == 0) {
1974 * We acquire tbl->lock to be nice to the periodic timers and
1975 * make sure they always see a consistent set of values.
1977 write_lock_bh(&tbl->lock);
1979 if (tb[NDTA_PARMS]) {
1980 struct nlattr *tbp[NDTPA_MAX+1];
1981 struct neigh_parms *p;
1984 err = nla_parse_nested(tbp, NDTPA_MAX, tb[NDTA_PARMS],
1985 nl_ntbl_parm_policy);
1987 goto errout_tbl_lock;
1989 if (tbp[NDTPA_IFINDEX])
1990 ifindex = nla_get_u32(tbp[NDTPA_IFINDEX]);
1992 p = lookup_neigh_parms(tbl, net, ifindex);
1995 goto errout_tbl_lock;
1998 for (i = 1; i <= NDTPA_MAX; i++) {
2003 case NDTPA_QUEUE_LEN:
2004 NEIGH_VAR_SET(p, QUEUE_LEN_BYTES,
2005 nla_get_u32(tbp[i]) *
2006 SKB_TRUESIZE(ETH_FRAME_LEN));
2008 case NDTPA_QUEUE_LENBYTES:
2009 NEIGH_VAR_SET(p, QUEUE_LEN_BYTES,
2010 nla_get_u32(tbp[i]));
2012 case NDTPA_PROXY_QLEN:
2013 NEIGH_VAR_SET(p, PROXY_QLEN,
2014 nla_get_u32(tbp[i]));
2016 case NDTPA_APP_PROBES:
2017 NEIGH_VAR_SET(p, APP_PROBES,
2018 nla_get_u32(tbp[i]));
2020 case NDTPA_UCAST_PROBES:
2021 NEIGH_VAR_SET(p, UCAST_PROBES,
2022 nla_get_u32(tbp[i]));
2024 case NDTPA_MCAST_PROBES:
2025 NEIGH_VAR_SET(p, MCAST_PROBES,
2026 nla_get_u32(tbp[i]));
2028 case NDTPA_MCAST_REPROBES:
2029 NEIGH_VAR_SET(p, MCAST_REPROBES,
2030 nla_get_u32(tbp[i]));
2032 case NDTPA_BASE_REACHABLE_TIME:
2033 NEIGH_VAR_SET(p, BASE_REACHABLE_TIME,
2034 nla_get_msecs(tbp[i]));
2035 /* update reachable_time as well, otherwise, the change will
2036 * only be effective after the next time neigh_periodic_work
2037 * decides to recompute it (can be multiple minutes)
2040 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
2042 case NDTPA_GC_STALETIME:
2043 NEIGH_VAR_SET(p, GC_STALETIME,
2044 nla_get_msecs(tbp[i]));
2046 case NDTPA_DELAY_PROBE_TIME:
2047 NEIGH_VAR_SET(p, DELAY_PROBE_TIME,
2048 nla_get_msecs(tbp[i]));
2050 case NDTPA_RETRANS_TIME:
2051 NEIGH_VAR_SET(p, RETRANS_TIME,
2052 nla_get_msecs(tbp[i]));
2054 case NDTPA_ANYCAST_DELAY:
2055 NEIGH_VAR_SET(p, ANYCAST_DELAY,
2056 nla_get_msecs(tbp[i]));
2058 case NDTPA_PROXY_DELAY:
2059 NEIGH_VAR_SET(p, PROXY_DELAY,
2060 nla_get_msecs(tbp[i]));
2062 case NDTPA_LOCKTIME:
2063 NEIGH_VAR_SET(p, LOCKTIME,
2064 nla_get_msecs(tbp[i]));
2071 if ((tb[NDTA_THRESH1] || tb[NDTA_THRESH2] ||
2072 tb[NDTA_THRESH3] || tb[NDTA_GC_INTERVAL]) &&
2073 !net_eq(net, &init_net))
2074 goto errout_tbl_lock;
2076 if (tb[NDTA_THRESH1])
2077 tbl->gc_thresh1 = nla_get_u32(tb[NDTA_THRESH1]);
2079 if (tb[NDTA_THRESH2])
2080 tbl->gc_thresh2 = nla_get_u32(tb[NDTA_THRESH2]);
2082 if (tb[NDTA_THRESH3])
2083 tbl->gc_thresh3 = nla_get_u32(tb[NDTA_THRESH3]);
2085 if (tb[NDTA_GC_INTERVAL])
2086 tbl->gc_interval = nla_get_msecs(tb[NDTA_GC_INTERVAL]);
2091 write_unlock_bh(&tbl->lock);
2096 static int neightbl_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2098 struct net *net = sock_net(skb->sk);
2099 int family, tidx, nidx = 0;
2100 int tbl_skip = cb->args[0];
2101 int neigh_skip = cb->args[1];
2102 struct neigh_table *tbl;
2104 family = ((struct rtgenmsg *) nlmsg_data(cb->nlh))->rtgen_family;
2106 for (tidx = 0; tidx < NEIGH_NR_TABLES; tidx++) {
2107 struct neigh_parms *p;
2109 tbl = neigh_tables[tidx];
2113 if (tidx < tbl_skip || (family && tbl->family != family))
2116 if (neightbl_fill_info(skb, tbl, NETLINK_CB(cb->skb).portid,
2117 cb->nlh->nlmsg_seq, RTM_NEWNEIGHTBL,
2122 p = list_next_entry(&tbl->parms, list);
2123 list_for_each_entry_from(p, &tbl->parms_list, list) {
2124 if (!net_eq(neigh_parms_net(p), net))
2127 if (nidx < neigh_skip)
2130 if (neightbl_fill_param_info(skb, tbl, p,
2131 NETLINK_CB(cb->skb).portid,
2149 static int neigh_fill_info(struct sk_buff *skb, struct neighbour *neigh,
2150 u32 pid, u32 seq, int type, unsigned int flags)
2152 unsigned long now = jiffies;
2153 struct nda_cacheinfo ci;
2154 struct nlmsghdr *nlh;
2157 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2161 ndm = nlmsg_data(nlh);
2162 ndm->ndm_family = neigh->ops->family;
2165 ndm->ndm_flags = neigh->flags;
2166 ndm->ndm_type = neigh->type;
2167 ndm->ndm_ifindex = neigh->dev->ifindex;
2169 if (nla_put(skb, NDA_DST, neigh->tbl->key_len, neigh->primary_key))
2170 goto nla_put_failure;
2172 read_lock_bh(&neigh->lock);
2173 ndm->ndm_state = neigh->nud_state;
2174 if (neigh->nud_state & NUD_VALID) {
2175 char haddr[MAX_ADDR_LEN];
2177 neigh_ha_snapshot(haddr, neigh, neigh->dev);
2178 if (nla_put(skb, NDA_LLADDR, neigh->dev->addr_len, haddr) < 0) {
2179 read_unlock_bh(&neigh->lock);
2180 goto nla_put_failure;
2184 ci.ndm_used = jiffies_to_clock_t(now - neigh->used);
2185 ci.ndm_confirmed = jiffies_to_clock_t(now - neigh->confirmed);
2186 ci.ndm_updated = jiffies_to_clock_t(now - neigh->updated);
2187 ci.ndm_refcnt = atomic_read(&neigh->refcnt) - 1;
2188 read_unlock_bh(&neigh->lock);
2190 if (nla_put_u32(skb, NDA_PROBES, atomic_read(&neigh->probes)) ||
2191 nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
2192 goto nla_put_failure;
2194 nlmsg_end(skb, nlh);
2198 nlmsg_cancel(skb, nlh);
2202 static int pneigh_fill_info(struct sk_buff *skb, struct pneigh_entry *pn,
2203 u32 pid, u32 seq, int type, unsigned int flags,
2204 struct neigh_table *tbl)
2206 struct nlmsghdr *nlh;
2209 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2213 ndm = nlmsg_data(nlh);
2214 ndm->ndm_family = tbl->family;
2217 ndm->ndm_flags = pn->flags | NTF_PROXY;
2218 ndm->ndm_type = RTN_UNICAST;
2219 ndm->ndm_ifindex = pn->dev ? pn->dev->ifindex : 0;
2220 ndm->ndm_state = NUD_NONE;
2222 if (nla_put(skb, NDA_DST, tbl->key_len, pn->key))
2223 goto nla_put_failure;
2225 nlmsg_end(skb, nlh);
2229 nlmsg_cancel(skb, nlh);
2233 static void neigh_update_notify(struct neighbour *neigh)
2235 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, neigh);
2236 __neigh_notify(neigh, RTM_NEWNEIGH, 0);
2239 static bool neigh_master_filtered(struct net_device *dev, int master_idx)
2241 struct net_device *master;
2246 master = netdev_master_upper_dev_get(dev);
2247 if (!master || master->ifindex != master_idx)
2253 static bool neigh_ifindex_filtered(struct net_device *dev, int filter_idx)
2255 if (filter_idx && dev->ifindex != filter_idx)
2261 static int neigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2262 struct netlink_callback *cb)
2264 struct net *net = sock_net(skb->sk);
2265 const struct nlmsghdr *nlh = cb->nlh;
2266 struct nlattr *tb[NDA_MAX + 1];
2267 struct neighbour *n;
2268 int rc, h, s_h = cb->args[1];
2269 int idx, s_idx = idx = cb->args[2];
2270 struct neigh_hash_table *nht;
2271 int filter_master_idx = 0, filter_idx = 0;
2272 unsigned int flags = NLM_F_MULTI;
2275 err = nlmsg_parse(nlh, sizeof(struct ndmsg), tb, NDA_MAX, NULL);
2277 if (tb[NDA_IFINDEX])
2278 filter_idx = nla_get_u32(tb[NDA_IFINDEX]);
2281 filter_master_idx = nla_get_u32(tb[NDA_MASTER]);
2283 if (filter_idx || filter_master_idx)
2284 flags |= NLM_F_DUMP_FILTERED;
2288 nht = rcu_dereference_bh(tbl->nht);
2290 for (h = s_h; h < (1 << nht->hash_shift); h++) {
2293 for (n = rcu_dereference_bh(nht->hash_buckets[h]), idx = 0;
2295 n = rcu_dereference_bh(n->next)) {
2296 if (!net_eq(dev_net(n->dev), net))
2298 if (neigh_ifindex_filtered(n->dev, filter_idx))
2300 if (neigh_master_filtered(n->dev, filter_master_idx))
2304 if (neigh_fill_info(skb, n, NETLINK_CB(cb->skb).portid,
2317 rcu_read_unlock_bh();
2323 static int pneigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2324 struct netlink_callback *cb)
2326 struct pneigh_entry *n;
2327 struct net *net = sock_net(skb->sk);
2328 int rc, h, s_h = cb->args[3];
2329 int idx, s_idx = idx = cb->args[4];
2331 read_lock_bh(&tbl->lock);
2333 for (h = s_h; h <= PNEIGH_HASHMASK; h++) {
2336 for (n = tbl->phash_buckets[h], idx = 0; n; n = n->next) {
2337 if (pneigh_net(n) != net)
2341 if (pneigh_fill_info(skb, n, NETLINK_CB(cb->skb).portid,
2344 NLM_F_MULTI, tbl) < 0) {
2345 read_unlock_bh(&tbl->lock);
2354 read_unlock_bh(&tbl->lock);
2363 static int neigh_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2365 struct neigh_table *tbl;
2370 family = ((struct rtgenmsg *) nlmsg_data(cb->nlh))->rtgen_family;
2372 /* check for full ndmsg structure presence, family member is
2373 * the same for both structures
2375 if (nlmsg_len(cb->nlh) >= sizeof(struct ndmsg) &&
2376 ((struct ndmsg *) nlmsg_data(cb->nlh))->ndm_flags == NTF_PROXY)
2381 for (t = 0; t < NEIGH_NR_TABLES; t++) {
2382 tbl = neigh_tables[t];
2386 if (t < s_t || (family && tbl->family != family))
2389 memset(&cb->args[1], 0, sizeof(cb->args) -
2390 sizeof(cb->args[0]));
2392 err = pneigh_dump_table(tbl, skb, cb);
2394 err = neigh_dump_table(tbl, skb, cb);
2403 void neigh_for_each(struct neigh_table *tbl, void (*cb)(struct neighbour *, void *), void *cookie)
2406 struct neigh_hash_table *nht;
2409 nht = rcu_dereference_bh(tbl->nht);
2411 read_lock(&tbl->lock); /* avoid resizes */
2412 for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
2413 struct neighbour *n;
2415 for (n = rcu_dereference_bh(nht->hash_buckets[chain]);
2417 n = rcu_dereference_bh(n->next))
2420 read_unlock(&tbl->lock);
2421 rcu_read_unlock_bh();
2423 EXPORT_SYMBOL(neigh_for_each);
2425 /* The tbl->lock must be held as a writer and BH disabled. */
2426 void __neigh_for_each_release(struct neigh_table *tbl,
2427 int (*cb)(struct neighbour *))
2430 struct neigh_hash_table *nht;
2432 nht = rcu_dereference_protected(tbl->nht,
2433 lockdep_is_held(&tbl->lock));
2434 for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
2435 struct neighbour *n;
2436 struct neighbour __rcu **np;
2438 np = &nht->hash_buckets[chain];
2439 while ((n = rcu_dereference_protected(*np,
2440 lockdep_is_held(&tbl->lock))) != NULL) {
2443 write_lock(&n->lock);
2446 rcu_assign_pointer(*np,
2447 rcu_dereference_protected(n->next,
2448 lockdep_is_held(&tbl->lock)));
2452 write_unlock(&n->lock);
2454 neigh_cleanup_and_release(n);
2458 EXPORT_SYMBOL(__neigh_for_each_release);
2460 int neigh_xmit(int index, struct net_device *dev,
2461 const void *addr, struct sk_buff *skb)
2463 int err = -EAFNOSUPPORT;
2464 if (likely(index < NEIGH_NR_TABLES)) {
2465 struct neigh_table *tbl;
2466 struct neighbour *neigh;
2468 tbl = neigh_tables[index];
2472 neigh = __neigh_lookup_noref(tbl, addr, dev);
2474 neigh = __neigh_create(tbl, addr, dev, false);
2475 err = PTR_ERR(neigh);
2476 if (IS_ERR(neigh)) {
2477 rcu_read_unlock_bh();
2480 err = neigh->output(neigh, skb);
2481 rcu_read_unlock_bh();
2483 else if (index == NEIGH_LINK_TABLE) {
2484 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
2485 addr, NULL, skb->len);
2488 err = dev_queue_xmit(skb);
2496 EXPORT_SYMBOL(neigh_xmit);
2498 #ifdef CONFIG_PROC_FS
2500 static struct neighbour *neigh_get_first(struct seq_file *seq)
2502 struct neigh_seq_state *state = seq->private;
2503 struct net *net = seq_file_net(seq);
2504 struct neigh_hash_table *nht = state->nht;
2505 struct neighbour *n = NULL;
2506 int bucket = state->bucket;
2508 state->flags &= ~NEIGH_SEQ_IS_PNEIGH;
2509 for (bucket = 0; bucket < (1 << nht->hash_shift); bucket++) {
2510 n = rcu_dereference_bh(nht->hash_buckets[bucket]);
2513 if (!net_eq(dev_net(n->dev), net))
2515 if (state->neigh_sub_iter) {
2519 v = state->neigh_sub_iter(state, n, &fakep);
2523 if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
2525 if (n->nud_state & ~NUD_NOARP)
2528 n = rcu_dereference_bh(n->next);
2534 state->bucket = bucket;
2539 static struct neighbour *neigh_get_next(struct seq_file *seq,
2540 struct neighbour *n,
2543 struct neigh_seq_state *state = seq->private;
2544 struct net *net = seq_file_net(seq);
2545 struct neigh_hash_table *nht = state->nht;
2547 if (state->neigh_sub_iter) {
2548 void *v = state->neigh_sub_iter(state, n, pos);
2552 n = rcu_dereference_bh(n->next);
2556 if (!net_eq(dev_net(n->dev), net))
2558 if (state->neigh_sub_iter) {
2559 void *v = state->neigh_sub_iter(state, n, pos);
2564 if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
2567 if (n->nud_state & ~NUD_NOARP)
2570 n = rcu_dereference_bh(n->next);
2576 if (++state->bucket >= (1 << nht->hash_shift))
2579 n = rcu_dereference_bh(nht->hash_buckets[state->bucket]);
2587 static struct neighbour *neigh_get_idx(struct seq_file *seq, loff_t *pos)
2589 struct neighbour *n = neigh_get_first(seq);
2594 n = neigh_get_next(seq, n, pos);
2599 return *pos ? NULL : n;
2602 static struct pneigh_entry *pneigh_get_first(struct seq_file *seq)
2604 struct neigh_seq_state *state = seq->private;
2605 struct net *net = seq_file_net(seq);
2606 struct neigh_table *tbl = state->tbl;
2607 struct pneigh_entry *pn = NULL;
2608 int bucket = state->bucket;
2610 state->flags |= NEIGH_SEQ_IS_PNEIGH;
2611 for (bucket = 0; bucket <= PNEIGH_HASHMASK; bucket++) {
2612 pn = tbl->phash_buckets[bucket];
2613 while (pn && !net_eq(pneigh_net(pn), net))
2618 state->bucket = bucket;
2623 static struct pneigh_entry *pneigh_get_next(struct seq_file *seq,
2624 struct pneigh_entry *pn,
2627 struct neigh_seq_state *state = seq->private;
2628 struct net *net = seq_file_net(seq);
2629 struct neigh_table *tbl = state->tbl;
2633 } while (pn && !net_eq(pneigh_net(pn), net));
2636 if (++state->bucket > PNEIGH_HASHMASK)
2638 pn = tbl->phash_buckets[state->bucket];
2639 while (pn && !net_eq(pneigh_net(pn), net))
2651 static struct pneigh_entry *pneigh_get_idx(struct seq_file *seq, loff_t *pos)
2653 struct pneigh_entry *pn = pneigh_get_first(seq);
2658 pn = pneigh_get_next(seq, pn, pos);
2663 return *pos ? NULL : pn;
2666 static void *neigh_get_idx_any(struct seq_file *seq, loff_t *pos)
2668 struct neigh_seq_state *state = seq->private;
2670 loff_t idxpos = *pos;
2672 rc = neigh_get_idx(seq, &idxpos);
2673 if (!rc && !(state->flags & NEIGH_SEQ_NEIGH_ONLY))
2674 rc = pneigh_get_idx(seq, &idxpos);
2679 void *neigh_seq_start(struct seq_file *seq, loff_t *pos, struct neigh_table *tbl, unsigned int neigh_seq_flags)
2682 struct neigh_seq_state *state = seq->private;
2686 state->flags = (neigh_seq_flags & ~NEIGH_SEQ_IS_PNEIGH);
2689 state->nht = rcu_dereference_bh(tbl->nht);
2691 return *pos ? neigh_get_idx_any(seq, pos) : SEQ_START_TOKEN;
2693 EXPORT_SYMBOL(neigh_seq_start);
2695 void *neigh_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2697 struct neigh_seq_state *state;
2700 if (v == SEQ_START_TOKEN) {
2701 rc = neigh_get_first(seq);
2705 state = seq->private;
2706 if (!(state->flags & NEIGH_SEQ_IS_PNEIGH)) {
2707 rc = neigh_get_next(seq, v, NULL);
2710 if (!(state->flags & NEIGH_SEQ_NEIGH_ONLY))
2711 rc = pneigh_get_first(seq);
2713 BUG_ON(state->flags & NEIGH_SEQ_NEIGH_ONLY);
2714 rc = pneigh_get_next(seq, v, NULL);
2720 EXPORT_SYMBOL(neigh_seq_next);
2722 void neigh_seq_stop(struct seq_file *seq, void *v)
2725 rcu_read_unlock_bh();
2727 EXPORT_SYMBOL(neigh_seq_stop);
2729 /* statistics via seq_file */
2731 static void *neigh_stat_seq_start(struct seq_file *seq, loff_t *pos)
2733 struct neigh_table *tbl = seq->private;
2737 return SEQ_START_TOKEN;
2739 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
2740 if (!cpu_possible(cpu))
2743 return per_cpu_ptr(tbl->stats, cpu);
2748 static void *neigh_stat_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2750 struct neigh_table *tbl = seq->private;
2753 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
2754 if (!cpu_possible(cpu))
2757 return per_cpu_ptr(tbl->stats, cpu);
2762 static void neigh_stat_seq_stop(struct seq_file *seq, void *v)
2767 static int neigh_stat_seq_show(struct seq_file *seq, void *v)
2769 struct neigh_table *tbl = seq->private;
2770 struct neigh_statistics *st = v;
2772 if (v == SEQ_START_TOKEN) {
2773 seq_printf(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");
2777 seq_printf(seq, "%08x %08lx %08lx %08lx %08lx %08lx %08lx "
2778 "%08lx %08lx %08lx %08lx %08lx %08lx\n",
2779 atomic_read(&tbl->entries),
2790 st->rcv_probes_mcast,
2791 st->rcv_probes_ucast,
2793 st->periodic_gc_runs,
2802 static const struct seq_operations neigh_stat_seq_ops = {
2803 .start = neigh_stat_seq_start,
2804 .next = neigh_stat_seq_next,
2805 .stop = neigh_stat_seq_stop,
2806 .show = neigh_stat_seq_show,
2809 static int neigh_stat_seq_open(struct inode *inode, struct file *file)
2811 int ret = seq_open(file, &neigh_stat_seq_ops);
2814 struct seq_file *sf = file->private_data;
2815 sf->private = PDE_DATA(inode);
2820 static const struct file_operations neigh_stat_seq_fops = {
2821 .owner = THIS_MODULE,
2822 .open = neigh_stat_seq_open,
2824 .llseek = seq_lseek,
2825 .release = seq_release,
2828 #endif /* CONFIG_PROC_FS */
2830 static inline size_t neigh_nlmsg_size(void)
2832 return NLMSG_ALIGN(sizeof(struct ndmsg))
2833 + nla_total_size(MAX_ADDR_LEN) /* NDA_DST */
2834 + nla_total_size(MAX_ADDR_LEN) /* NDA_LLADDR */
2835 + nla_total_size(sizeof(struct nda_cacheinfo))
2836 + nla_total_size(4); /* NDA_PROBES */
2839 static void __neigh_notify(struct neighbour *n, int type, int flags)
2841 struct net *net = dev_net(n->dev);
2842 struct sk_buff *skb;
2845 skb = nlmsg_new(neigh_nlmsg_size(), GFP_ATOMIC);
2849 err = neigh_fill_info(skb, n, 0, 0, type, flags);
2851 /* -EMSGSIZE implies BUG in neigh_nlmsg_size() */
2852 WARN_ON(err == -EMSGSIZE);
2856 rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
2860 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
2863 void neigh_app_ns(struct neighbour *n)
2865 __neigh_notify(n, RTM_GETNEIGH, NLM_F_REQUEST);
2867 EXPORT_SYMBOL(neigh_app_ns);
2869 #ifdef CONFIG_SYSCTL
2871 static int int_max = INT_MAX;
2872 static int unres_qlen_max = INT_MAX / SKB_TRUESIZE(ETH_FRAME_LEN);
2874 static int proc_unres_qlen(struct ctl_table *ctl, int write,
2875 void __user *buffer, size_t *lenp, loff_t *ppos)
2878 struct ctl_table tmp = *ctl;
2881 tmp.extra2 = &unres_qlen_max;
2884 size = *(int *)ctl->data / SKB_TRUESIZE(ETH_FRAME_LEN);
2885 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
2888 *(int *)ctl->data = size * SKB_TRUESIZE(ETH_FRAME_LEN);
2892 static struct neigh_parms *neigh_get_dev_parms_rcu(struct net_device *dev,
2897 return __in_dev_arp_parms_get_rcu(dev);
2899 return __in6_dev_nd_parms_get_rcu(dev);
2904 static void neigh_copy_dflt_parms(struct net *net, struct neigh_parms *p,
2907 struct net_device *dev;
2908 int family = neigh_parms_family(p);
2911 for_each_netdev_rcu(net, dev) {
2912 struct neigh_parms *dst_p =
2913 neigh_get_dev_parms_rcu(dev, family);
2915 if (dst_p && !test_bit(index, dst_p->data_state))
2916 dst_p->data[index] = p->data[index];
2921 static void neigh_proc_update(struct ctl_table *ctl, int write)
2923 struct net_device *dev = ctl->extra1;
2924 struct neigh_parms *p = ctl->extra2;
2925 struct net *net = neigh_parms_net(p);
2926 int index = (int *) ctl->data - p->data;
2931 set_bit(index, p->data_state);
2932 if (!dev) /* NULL dev means this is default value */
2933 neigh_copy_dflt_parms(net, p, index);
2936 static int neigh_proc_dointvec_zero_intmax(struct ctl_table *ctl, int write,
2937 void __user *buffer,
2938 size_t *lenp, loff_t *ppos)
2940 struct ctl_table tmp = *ctl;
2944 tmp.extra2 = &int_max;
2946 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
2947 neigh_proc_update(ctl, write);
2951 int neigh_proc_dointvec(struct ctl_table *ctl, int write,
2952 void __user *buffer, size_t *lenp, loff_t *ppos)
2954 int ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
2956 neigh_proc_update(ctl, write);
2959 EXPORT_SYMBOL(neigh_proc_dointvec);
2961 int neigh_proc_dointvec_jiffies(struct ctl_table *ctl, int write,
2962 void __user *buffer,
2963 size_t *lenp, loff_t *ppos)
2965 int ret = proc_dointvec_jiffies(ctl, write, buffer, lenp, ppos);
2967 neigh_proc_update(ctl, write);
2970 EXPORT_SYMBOL(neigh_proc_dointvec_jiffies);
2972 static int neigh_proc_dointvec_userhz_jiffies(struct ctl_table *ctl, int write,
2973 void __user *buffer,
2974 size_t *lenp, loff_t *ppos)
2976 int ret = proc_dointvec_userhz_jiffies(ctl, write, buffer, lenp, ppos);
2978 neigh_proc_update(ctl, write);
2982 int neigh_proc_dointvec_ms_jiffies(struct ctl_table *ctl, int write,
2983 void __user *buffer,
2984 size_t *lenp, loff_t *ppos)
2986 int ret = proc_dointvec_ms_jiffies(ctl, write, buffer, lenp, ppos);
2988 neigh_proc_update(ctl, write);
2991 EXPORT_SYMBOL(neigh_proc_dointvec_ms_jiffies);
2993 static int neigh_proc_dointvec_unres_qlen(struct ctl_table *ctl, int write,
2994 void __user *buffer,
2995 size_t *lenp, loff_t *ppos)
2997 int ret = proc_unres_qlen(ctl, write, buffer, lenp, ppos);
2999 neigh_proc_update(ctl, write);
3003 static int neigh_proc_base_reachable_time(struct ctl_table *ctl, int write,
3004 void __user *buffer,
3005 size_t *lenp, loff_t *ppos)
3007 struct neigh_parms *p = ctl->extra2;
3010 if (strcmp(ctl->procname, "base_reachable_time") == 0)
3011 ret = neigh_proc_dointvec_jiffies(ctl, write, buffer, lenp, ppos);
3012 else if (strcmp(ctl->procname, "base_reachable_time_ms") == 0)
3013 ret = neigh_proc_dointvec_ms_jiffies(ctl, write, buffer, lenp, ppos);
3017 if (write && ret == 0) {
3018 /* update reachable_time as well, otherwise, the change will
3019 * only be effective after the next time neigh_periodic_work
3020 * decides to recompute it
3023 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
3028 #define NEIGH_PARMS_DATA_OFFSET(index) \
3029 (&((struct neigh_parms *) 0)->data[index])
3031 #define NEIGH_SYSCTL_ENTRY(attr, data_attr, name, mval, proc) \
3032 [NEIGH_VAR_ ## attr] = { \
3034 .data = NEIGH_PARMS_DATA_OFFSET(NEIGH_VAR_ ## data_attr), \
3035 .maxlen = sizeof(int), \
3037 .proc_handler = proc, \
3040 #define NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(attr, name) \
3041 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_zero_intmax)
3043 #define NEIGH_SYSCTL_JIFFIES_ENTRY(attr, name) \
3044 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_jiffies)
3046 #define NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(attr, name) \
3047 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_userhz_jiffies)
3049 #define NEIGH_SYSCTL_MS_JIFFIES_ENTRY(attr, name) \
3050 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_ms_jiffies)
3052 #define NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(attr, data_attr, name) \
3053 NEIGH_SYSCTL_ENTRY(attr, data_attr, name, 0644, neigh_proc_dointvec_ms_jiffies)
3055 #define NEIGH_SYSCTL_UNRES_QLEN_REUSED_ENTRY(attr, data_attr, name) \
3056 NEIGH_SYSCTL_ENTRY(attr, data_attr, name, 0644, neigh_proc_dointvec_unres_qlen)
3058 static struct neigh_sysctl_table {
3059 struct ctl_table_header *sysctl_header;
3060 struct ctl_table neigh_vars[NEIGH_VAR_MAX + 1];
3061 } neigh_sysctl_template __read_mostly = {
3063 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(MCAST_PROBES, "mcast_solicit"),
3064 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(UCAST_PROBES, "ucast_solicit"),
3065 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(APP_PROBES, "app_solicit"),
3066 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(MCAST_REPROBES, "mcast_resolicit"),
3067 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(RETRANS_TIME, "retrans_time"),
3068 NEIGH_SYSCTL_JIFFIES_ENTRY(BASE_REACHABLE_TIME, "base_reachable_time"),
3069 NEIGH_SYSCTL_JIFFIES_ENTRY(DELAY_PROBE_TIME, "delay_first_probe_time"),
3070 NEIGH_SYSCTL_JIFFIES_ENTRY(GC_STALETIME, "gc_stale_time"),
3071 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(QUEUE_LEN_BYTES, "unres_qlen_bytes"),
3072 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(PROXY_QLEN, "proxy_qlen"),
3073 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(ANYCAST_DELAY, "anycast_delay"),
3074 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(PROXY_DELAY, "proxy_delay"),
3075 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(LOCKTIME, "locktime"),
3076 NEIGH_SYSCTL_UNRES_QLEN_REUSED_ENTRY(QUEUE_LEN, QUEUE_LEN_BYTES, "unres_qlen"),
3077 NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(RETRANS_TIME_MS, RETRANS_TIME, "retrans_time_ms"),
3078 NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(BASE_REACHABLE_TIME_MS, BASE_REACHABLE_TIME, "base_reachable_time_ms"),
3079 [NEIGH_VAR_GC_INTERVAL] = {
3080 .procname = "gc_interval",
3081 .maxlen = sizeof(int),
3083 .proc_handler = proc_dointvec_jiffies,
3085 [NEIGH_VAR_GC_THRESH1] = {
3086 .procname = "gc_thresh1",
3087 .maxlen = sizeof(int),
3091 .proc_handler = proc_dointvec_minmax,
3093 [NEIGH_VAR_GC_THRESH2] = {
3094 .procname = "gc_thresh2",
3095 .maxlen = sizeof(int),
3099 .proc_handler = proc_dointvec_minmax,
3101 [NEIGH_VAR_GC_THRESH3] = {
3102 .procname = "gc_thresh3",
3103 .maxlen = sizeof(int),
3107 .proc_handler = proc_dointvec_minmax,
3113 int neigh_sysctl_register(struct net_device *dev, struct neigh_parms *p,
3114 proc_handler *handler)
3117 struct neigh_sysctl_table *t;
3118 const char *dev_name_source;
3119 char neigh_path[ sizeof("net//neigh/") + IFNAMSIZ + IFNAMSIZ ];
3122 t = kmemdup(&neigh_sysctl_template, sizeof(*t), GFP_KERNEL);
3126 for (i = 0; i < NEIGH_VAR_GC_INTERVAL; i++) {
3127 t->neigh_vars[i].data += (long) p;
3128 t->neigh_vars[i].extra1 = dev;
3129 t->neigh_vars[i].extra2 = p;
3133 dev_name_source = dev->name;
3134 /* Terminate the table early */
3135 memset(&t->neigh_vars[NEIGH_VAR_GC_INTERVAL], 0,
3136 sizeof(t->neigh_vars[NEIGH_VAR_GC_INTERVAL]));
3138 struct neigh_table *tbl = p->tbl;
3139 dev_name_source = "default";
3140 t->neigh_vars[NEIGH_VAR_GC_INTERVAL].data = &tbl->gc_interval;
3141 t->neigh_vars[NEIGH_VAR_GC_THRESH1].data = &tbl->gc_thresh1;
3142 t->neigh_vars[NEIGH_VAR_GC_THRESH2].data = &tbl->gc_thresh2;
3143 t->neigh_vars[NEIGH_VAR_GC_THRESH3].data = &tbl->gc_thresh3;
3148 t->neigh_vars[NEIGH_VAR_RETRANS_TIME].proc_handler = handler;
3150 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].proc_handler = handler;
3151 /* RetransTime (in milliseconds)*/
3152 t->neigh_vars[NEIGH_VAR_RETRANS_TIME_MS].proc_handler = handler;
3153 /* ReachableTime (in milliseconds) */
3154 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].proc_handler = handler;
3156 /* Those handlers will update p->reachable_time after
3157 * base_reachable_time(_ms) is set to ensure the new timer starts being
3158 * applied after the next neighbour update instead of waiting for
3159 * neigh_periodic_work to update its value (can be multiple minutes)
3160 * So any handler that replaces them should do this as well
3163 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].proc_handler =
3164 neigh_proc_base_reachable_time;
3165 /* ReachableTime (in milliseconds) */
3166 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].proc_handler =
3167 neigh_proc_base_reachable_time;
3170 /* Don't export sysctls to unprivileged users */
3171 if (neigh_parms_net(p)->user_ns != &init_user_ns)
3172 t->neigh_vars[0].procname = NULL;
3174 switch (neigh_parms_family(p)) {
3185 snprintf(neigh_path, sizeof(neigh_path), "net/%s/neigh/%s",
3186 p_name, dev_name_source);
3188 register_net_sysctl(neigh_parms_net(p), neigh_path, t->neigh_vars);
3189 if (!t->sysctl_header)
3192 p->sysctl_table = t;
3200 EXPORT_SYMBOL(neigh_sysctl_register);
3202 void neigh_sysctl_unregister(struct neigh_parms *p)
3204 if (p->sysctl_table) {
3205 struct neigh_sysctl_table *t = p->sysctl_table;
3206 p->sysctl_table = NULL;
3207 unregister_net_sysctl_table(t->sysctl_header);
3211 EXPORT_SYMBOL(neigh_sysctl_unregister);
3213 #endif /* CONFIG_SYSCTL */
3215 static int __init neigh_init(void)
3217 rtnl_register(PF_UNSPEC, RTM_NEWNEIGH, neigh_add, NULL, NULL);
3218 rtnl_register(PF_UNSPEC, RTM_DELNEIGH, neigh_delete, NULL, NULL);
3219 rtnl_register(PF_UNSPEC, RTM_GETNEIGH, NULL, neigh_dump_info, NULL);
3221 rtnl_register(PF_UNSPEC, RTM_GETNEIGHTBL, NULL, neightbl_dump_info,
3223 rtnl_register(PF_UNSPEC, RTM_SETNEIGHTBL, neightbl_set, NULL, NULL);
3228 subsys_initcall(neigh_init);