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)
281 n = kzalloc(tbl->entry_size + dev->neigh_priv_len, GFP_ATOMIC);
285 __skb_queue_head_init(&n->arp_queue);
286 rwlock_init(&n->lock);
287 seqlock_init(&n->ha_lock);
288 n->updated = n->used = now;
289 n->nud_state = NUD_NONE;
290 n->output = neigh_blackhole;
291 seqlock_init(&n->hh.hh_lock);
292 n->parms = neigh_parms_clone(&tbl->parms);
293 setup_timer(&n->timer, neigh_timer_handler, (unsigned long)n);
295 NEIGH_CACHE_STAT_INC(tbl, allocs);
297 atomic_set(&n->refcnt, 1);
303 atomic_dec(&tbl->entries);
307 static void neigh_get_hash_rnd(u32 *x)
309 get_random_bytes(x, sizeof(*x));
313 static struct neigh_hash_table *neigh_hash_alloc(unsigned int shift)
315 size_t size = (1 << shift) * sizeof(struct neighbour *);
316 struct neigh_hash_table *ret;
317 struct neighbour __rcu **buckets;
320 ret = kmalloc(sizeof(*ret), GFP_ATOMIC);
323 if (size <= PAGE_SIZE)
324 buckets = kzalloc(size, GFP_ATOMIC);
326 buckets = (struct neighbour __rcu **)
327 __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
333 ret->hash_buckets = buckets;
334 ret->hash_shift = shift;
335 for (i = 0; i < NEIGH_NUM_HASH_RND; i++)
336 neigh_get_hash_rnd(&ret->hash_rnd[i]);
340 static void neigh_hash_free_rcu(struct rcu_head *head)
342 struct neigh_hash_table *nht = container_of(head,
343 struct neigh_hash_table,
345 size_t size = (1 << nht->hash_shift) * sizeof(struct neighbour *);
346 struct neighbour __rcu **buckets = nht->hash_buckets;
348 if (size <= PAGE_SIZE)
351 free_pages((unsigned long)buckets, get_order(size));
355 static struct neigh_hash_table *neigh_hash_grow(struct neigh_table *tbl,
356 unsigned long new_shift)
358 unsigned int i, hash;
359 struct neigh_hash_table *new_nht, *old_nht;
361 NEIGH_CACHE_STAT_INC(tbl, hash_grows);
363 old_nht = rcu_dereference_protected(tbl->nht,
364 lockdep_is_held(&tbl->lock));
365 new_nht = neigh_hash_alloc(new_shift);
369 for (i = 0; i < (1 << old_nht->hash_shift); i++) {
370 struct neighbour *n, *next;
372 for (n = rcu_dereference_protected(old_nht->hash_buckets[i],
373 lockdep_is_held(&tbl->lock));
376 hash = tbl->hash(n->primary_key, n->dev,
379 hash >>= (32 - new_nht->hash_shift);
380 next = rcu_dereference_protected(n->next,
381 lockdep_is_held(&tbl->lock));
383 rcu_assign_pointer(n->next,
384 rcu_dereference_protected(
385 new_nht->hash_buckets[hash],
386 lockdep_is_held(&tbl->lock)));
387 rcu_assign_pointer(new_nht->hash_buckets[hash], n);
391 rcu_assign_pointer(tbl->nht, new_nht);
392 call_rcu(&old_nht->rcu, neigh_hash_free_rcu);
396 struct neighbour *neigh_lookup(struct neigh_table *tbl, const void *pkey,
397 struct net_device *dev)
400 int key_len = tbl->key_len;
402 struct neigh_hash_table *nht;
404 NEIGH_CACHE_STAT_INC(tbl, lookups);
407 nht = rcu_dereference_bh(tbl->nht);
408 hash_val = tbl->hash(pkey, dev, nht->hash_rnd) >> (32 - nht->hash_shift);
410 for (n = rcu_dereference_bh(nht->hash_buckets[hash_val]);
412 n = rcu_dereference_bh(n->next)) {
413 if (dev == n->dev && !memcmp(n->primary_key, pkey, key_len)) {
414 if (!atomic_inc_not_zero(&n->refcnt))
416 NEIGH_CACHE_STAT_INC(tbl, hits);
421 rcu_read_unlock_bh();
424 EXPORT_SYMBOL(neigh_lookup);
426 struct neighbour *neigh_lookup_nodev(struct neigh_table *tbl, struct net *net,
430 int key_len = tbl->key_len;
432 struct neigh_hash_table *nht;
434 NEIGH_CACHE_STAT_INC(tbl, lookups);
437 nht = rcu_dereference_bh(tbl->nht);
438 hash_val = tbl->hash(pkey, NULL, nht->hash_rnd) >> (32 - nht->hash_shift);
440 for (n = rcu_dereference_bh(nht->hash_buckets[hash_val]);
442 n = rcu_dereference_bh(n->next)) {
443 if (!memcmp(n->primary_key, pkey, key_len) &&
444 net_eq(dev_net(n->dev), net)) {
445 if (!atomic_inc_not_zero(&n->refcnt))
447 NEIGH_CACHE_STAT_INC(tbl, hits);
452 rcu_read_unlock_bh();
455 EXPORT_SYMBOL(neigh_lookup_nodev);
457 struct neighbour *__neigh_create(struct neigh_table *tbl, const void *pkey,
458 struct net_device *dev, bool want_ref)
461 int key_len = tbl->key_len;
463 struct neighbour *n1, *rc, *n = neigh_alloc(tbl, dev);
464 struct neigh_hash_table *nht;
467 rc = ERR_PTR(-ENOBUFS);
471 memcpy(n->primary_key, pkey, key_len);
475 /* Protocol specific setup. */
476 if (tbl->constructor && (error = tbl->constructor(n)) < 0) {
478 goto out_neigh_release;
481 if (dev->netdev_ops->ndo_neigh_construct) {
482 error = dev->netdev_ops->ndo_neigh_construct(n);
485 goto out_neigh_release;
489 /* Device specific setup. */
490 if (n->parms->neigh_setup &&
491 (error = n->parms->neigh_setup(n)) < 0) {
493 goto out_neigh_release;
496 n->confirmed = jiffies - (NEIGH_VAR(n->parms, BASE_REACHABLE_TIME) << 1);
498 write_lock_bh(&tbl->lock);
499 nht = rcu_dereference_protected(tbl->nht,
500 lockdep_is_held(&tbl->lock));
502 if (atomic_read(&tbl->entries) > (1 << nht->hash_shift))
503 nht = neigh_hash_grow(tbl, nht->hash_shift + 1);
505 hash_val = tbl->hash(pkey, dev, nht->hash_rnd) >> (32 - nht->hash_shift);
507 if (n->parms->dead) {
508 rc = ERR_PTR(-EINVAL);
512 for (n1 = rcu_dereference_protected(nht->hash_buckets[hash_val],
513 lockdep_is_held(&tbl->lock));
515 n1 = rcu_dereference_protected(n1->next,
516 lockdep_is_held(&tbl->lock))) {
517 if (dev == n1->dev && !memcmp(n1->primary_key, pkey, key_len)) {
528 rcu_assign_pointer(n->next,
529 rcu_dereference_protected(nht->hash_buckets[hash_val],
530 lockdep_is_held(&tbl->lock)));
531 rcu_assign_pointer(nht->hash_buckets[hash_val], n);
532 write_unlock_bh(&tbl->lock);
533 neigh_dbg(2, "neigh %p is created\n", n);
538 write_unlock_bh(&tbl->lock);
543 EXPORT_SYMBOL(__neigh_create);
545 static u32 pneigh_hash(const void *pkey, int key_len)
547 u32 hash_val = *(u32 *)(pkey + key_len - 4);
548 hash_val ^= (hash_val >> 16);
549 hash_val ^= hash_val >> 8;
550 hash_val ^= hash_val >> 4;
551 hash_val &= PNEIGH_HASHMASK;
555 static struct pneigh_entry *__pneigh_lookup_1(struct pneigh_entry *n,
559 struct net_device *dev)
562 if (!memcmp(n->key, pkey, key_len) &&
563 net_eq(pneigh_net(n), net) &&
564 (n->dev == dev || !n->dev))
571 struct pneigh_entry *__pneigh_lookup(struct neigh_table *tbl,
572 struct net *net, const void *pkey, struct net_device *dev)
574 int key_len = tbl->key_len;
575 u32 hash_val = pneigh_hash(pkey, key_len);
577 return __pneigh_lookup_1(tbl->phash_buckets[hash_val],
578 net, pkey, key_len, dev);
580 EXPORT_SYMBOL_GPL(__pneigh_lookup);
582 struct pneigh_entry * pneigh_lookup(struct neigh_table *tbl,
583 struct net *net, const void *pkey,
584 struct net_device *dev, int creat)
586 struct pneigh_entry *n;
587 int key_len = tbl->key_len;
588 u32 hash_val = pneigh_hash(pkey, key_len);
590 read_lock_bh(&tbl->lock);
591 n = __pneigh_lookup_1(tbl->phash_buckets[hash_val],
592 net, pkey, key_len, dev);
593 read_unlock_bh(&tbl->lock);
600 n = kmalloc(sizeof(*n) + key_len, GFP_KERNEL);
604 write_pnet(&n->net, hold_net(net));
605 memcpy(n->key, pkey, key_len);
610 if (tbl->pconstructor && tbl->pconstructor(n)) {
619 write_lock_bh(&tbl->lock);
620 n->next = tbl->phash_buckets[hash_val];
621 tbl->phash_buckets[hash_val] = n;
622 write_unlock_bh(&tbl->lock);
626 EXPORT_SYMBOL(pneigh_lookup);
629 int pneigh_delete(struct neigh_table *tbl, struct net *net, const void *pkey,
630 struct net_device *dev)
632 struct pneigh_entry *n, **np;
633 int key_len = tbl->key_len;
634 u32 hash_val = pneigh_hash(pkey, key_len);
636 write_lock_bh(&tbl->lock);
637 for (np = &tbl->phash_buckets[hash_val]; (n = *np) != NULL;
639 if (!memcmp(n->key, pkey, key_len) && n->dev == dev &&
640 net_eq(pneigh_net(n), net)) {
642 write_unlock_bh(&tbl->lock);
643 if (tbl->pdestructor)
647 release_net(pneigh_net(n));
652 write_unlock_bh(&tbl->lock);
656 static int pneigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
658 struct pneigh_entry *n, **np;
661 for (h = 0; h <= PNEIGH_HASHMASK; h++) {
662 np = &tbl->phash_buckets[h];
663 while ((n = *np) != NULL) {
664 if (!dev || n->dev == dev) {
666 if (tbl->pdestructor)
670 release_net(pneigh_net(n));
680 static void neigh_parms_destroy(struct neigh_parms *parms);
682 static inline void neigh_parms_put(struct neigh_parms *parms)
684 if (atomic_dec_and_test(&parms->refcnt))
685 neigh_parms_destroy(parms);
689 * neighbour must already be out of the table;
692 void neigh_destroy(struct neighbour *neigh)
694 struct net_device *dev = neigh->dev;
696 NEIGH_CACHE_STAT_INC(neigh->tbl, destroys);
699 pr_warn("Destroying alive neighbour %p\n", neigh);
704 if (neigh_del_timer(neigh))
705 pr_warn("Impossible event\n");
707 write_lock_bh(&neigh->lock);
708 __skb_queue_purge(&neigh->arp_queue);
709 write_unlock_bh(&neigh->lock);
710 neigh->arp_queue_len_bytes = 0;
712 if (dev->netdev_ops->ndo_neigh_destroy)
713 dev->netdev_ops->ndo_neigh_destroy(neigh);
716 neigh_parms_put(neigh->parms);
718 neigh_dbg(2, "neigh %p is destroyed\n", neigh);
720 atomic_dec(&neigh->tbl->entries);
721 kfree_rcu(neigh, rcu);
723 EXPORT_SYMBOL(neigh_destroy);
725 /* Neighbour state is suspicious;
728 Called with write_locked neigh.
730 static void neigh_suspect(struct neighbour *neigh)
732 neigh_dbg(2, "neigh %p is suspected\n", neigh);
734 neigh->output = neigh->ops->output;
737 /* Neighbour state is OK;
740 Called with write_locked neigh.
742 static void neigh_connect(struct neighbour *neigh)
744 neigh_dbg(2, "neigh %p is connected\n", neigh);
746 neigh->output = neigh->ops->connected_output;
749 static void neigh_periodic_work(struct work_struct *work)
751 struct neigh_table *tbl = container_of(work, struct neigh_table, gc_work.work);
753 struct neighbour __rcu **np;
755 struct neigh_hash_table *nht;
757 NEIGH_CACHE_STAT_INC(tbl, periodic_gc_runs);
759 write_lock_bh(&tbl->lock);
760 nht = rcu_dereference_protected(tbl->nht,
761 lockdep_is_held(&tbl->lock));
764 * periodically recompute ReachableTime from random function
767 if (time_after(jiffies, tbl->last_rand + 300 * HZ)) {
768 struct neigh_parms *p;
769 tbl->last_rand = jiffies;
770 list_for_each_entry(p, &tbl->parms_list, list)
772 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
775 if (atomic_read(&tbl->entries) < tbl->gc_thresh1)
778 for (i = 0 ; i < (1 << nht->hash_shift); i++) {
779 np = &nht->hash_buckets[i];
781 while ((n = rcu_dereference_protected(*np,
782 lockdep_is_held(&tbl->lock))) != NULL) {
785 write_lock(&n->lock);
787 state = n->nud_state;
788 if (state & (NUD_PERMANENT | NUD_IN_TIMER)) {
789 write_unlock(&n->lock);
793 if (time_before(n->used, n->confirmed))
794 n->used = n->confirmed;
796 if (atomic_read(&n->refcnt) == 1 &&
797 (state == NUD_FAILED ||
798 time_after(jiffies, n->used + NEIGH_VAR(n->parms, GC_STALETIME)))) {
801 write_unlock(&n->lock);
802 neigh_cleanup_and_release(n);
805 write_unlock(&n->lock);
811 * It's fine to release lock here, even if hash table
812 * grows while we are preempted.
814 write_unlock_bh(&tbl->lock);
816 write_lock_bh(&tbl->lock);
817 nht = rcu_dereference_protected(tbl->nht,
818 lockdep_is_held(&tbl->lock));
821 /* Cycle through all hash buckets every BASE_REACHABLE_TIME/2 ticks.
822 * ARP entry timeouts range from 1/2 BASE_REACHABLE_TIME to 3/2
823 * BASE_REACHABLE_TIME.
825 queue_delayed_work(system_power_efficient_wq, &tbl->gc_work,
826 NEIGH_VAR(&tbl->parms, BASE_REACHABLE_TIME) >> 1);
827 write_unlock_bh(&tbl->lock);
830 static __inline__ int neigh_max_probes(struct neighbour *n)
832 struct neigh_parms *p = n->parms;
833 int max_probes = NEIGH_VAR(p, UCAST_PROBES) + NEIGH_VAR(p, APP_PROBES);
834 if (!(n->nud_state & NUD_PROBE))
835 max_probes += NEIGH_VAR(p, MCAST_PROBES);
839 static void neigh_invalidate(struct neighbour *neigh)
840 __releases(neigh->lock)
841 __acquires(neigh->lock)
845 NEIGH_CACHE_STAT_INC(neigh->tbl, res_failed);
846 neigh_dbg(2, "neigh %p is failed\n", neigh);
847 neigh->updated = jiffies;
849 /* It is very thin place. report_unreachable is very complicated
850 routine. Particularly, it can hit the same neighbour entry!
852 So that, we try to be accurate and avoid dead loop. --ANK
854 while (neigh->nud_state == NUD_FAILED &&
855 (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
856 write_unlock(&neigh->lock);
857 neigh->ops->error_report(neigh, skb);
858 write_lock(&neigh->lock);
860 __skb_queue_purge(&neigh->arp_queue);
861 neigh->arp_queue_len_bytes = 0;
864 static void neigh_probe(struct neighbour *neigh)
865 __releases(neigh->lock)
867 struct sk_buff *skb = skb_peek_tail(&neigh->arp_queue);
868 /* keep skb alive even if arp_queue overflows */
870 skb = skb_copy(skb, GFP_ATOMIC);
871 write_unlock(&neigh->lock);
872 neigh->ops->solicit(neigh, skb);
873 atomic_inc(&neigh->probes);
877 /* Called when a timer expires for a neighbour entry. */
879 static void neigh_timer_handler(unsigned long arg)
881 unsigned long now, next;
882 struct neighbour *neigh = (struct neighbour *)arg;
886 write_lock(&neigh->lock);
888 state = neigh->nud_state;
892 if (!(state & NUD_IN_TIMER))
895 if (state & NUD_REACHABLE) {
896 if (time_before_eq(now,
897 neigh->confirmed + neigh->parms->reachable_time)) {
898 neigh_dbg(2, "neigh %p is still alive\n", neigh);
899 next = neigh->confirmed + neigh->parms->reachable_time;
900 } else if (time_before_eq(now,
902 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME))) {
903 neigh_dbg(2, "neigh %p is delayed\n", neigh);
904 neigh->nud_state = NUD_DELAY;
905 neigh->updated = jiffies;
906 neigh_suspect(neigh);
907 next = now + NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME);
909 neigh_dbg(2, "neigh %p is suspected\n", neigh);
910 neigh->nud_state = NUD_STALE;
911 neigh->updated = jiffies;
912 neigh_suspect(neigh);
915 } else if (state & NUD_DELAY) {
916 if (time_before_eq(now,
918 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME))) {
919 neigh_dbg(2, "neigh %p is now reachable\n", neigh);
920 neigh->nud_state = NUD_REACHABLE;
921 neigh->updated = jiffies;
922 neigh_connect(neigh);
924 next = neigh->confirmed + neigh->parms->reachable_time;
926 neigh_dbg(2, "neigh %p is probed\n", neigh);
927 neigh->nud_state = NUD_PROBE;
928 neigh->updated = jiffies;
929 atomic_set(&neigh->probes, 0);
930 next = now + NEIGH_VAR(neigh->parms, RETRANS_TIME);
933 /* NUD_PROBE|NUD_INCOMPLETE */
934 next = now + NEIGH_VAR(neigh->parms, RETRANS_TIME);
937 if ((neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) &&
938 atomic_read(&neigh->probes) >= neigh_max_probes(neigh)) {
939 neigh->nud_state = NUD_FAILED;
941 neigh_invalidate(neigh);
945 if (neigh->nud_state & NUD_IN_TIMER) {
946 if (time_before(next, jiffies + HZ/2))
947 next = jiffies + HZ/2;
948 if (!mod_timer(&neigh->timer, next))
951 if (neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) {
955 write_unlock(&neigh->lock);
959 neigh_update_notify(neigh);
961 neigh_release(neigh);
964 int __neigh_event_send(struct neighbour *neigh, struct sk_buff *skb)
967 bool immediate_probe = false;
969 write_lock_bh(&neigh->lock);
972 if (neigh->nud_state & (NUD_CONNECTED | NUD_DELAY | NUD_PROBE))
975 if (!(neigh->nud_state & (NUD_STALE | NUD_INCOMPLETE))) {
976 if (NEIGH_VAR(neigh->parms, MCAST_PROBES) +
977 NEIGH_VAR(neigh->parms, APP_PROBES)) {
978 unsigned long next, now = jiffies;
980 atomic_set(&neigh->probes,
981 NEIGH_VAR(neigh->parms, UCAST_PROBES));
982 neigh->nud_state = NUD_INCOMPLETE;
983 neigh->updated = now;
984 next = now + max(NEIGH_VAR(neigh->parms, RETRANS_TIME),
986 neigh_add_timer(neigh, next);
987 immediate_probe = true;
989 neigh->nud_state = NUD_FAILED;
990 neigh->updated = jiffies;
991 write_unlock_bh(&neigh->lock);
996 } else if (neigh->nud_state & NUD_STALE) {
997 neigh_dbg(2, "neigh %p is delayed\n", neigh);
998 neigh->nud_state = NUD_DELAY;
999 neigh->updated = jiffies;
1000 neigh_add_timer(neigh, jiffies +
1001 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME));
1004 if (neigh->nud_state == NUD_INCOMPLETE) {
1006 while (neigh->arp_queue_len_bytes + skb->truesize >
1007 NEIGH_VAR(neigh->parms, QUEUE_LEN_BYTES)) {
1008 struct sk_buff *buff;
1010 buff = __skb_dequeue(&neigh->arp_queue);
1013 neigh->arp_queue_len_bytes -= buff->truesize;
1015 NEIGH_CACHE_STAT_INC(neigh->tbl, unres_discards);
1018 __skb_queue_tail(&neigh->arp_queue, skb);
1019 neigh->arp_queue_len_bytes += skb->truesize;
1024 if (immediate_probe)
1027 write_unlock(&neigh->lock);
1031 EXPORT_SYMBOL(__neigh_event_send);
1033 static void neigh_update_hhs(struct neighbour *neigh)
1035 struct hh_cache *hh;
1036 void (*update)(struct hh_cache*, const struct net_device*, const unsigned char *)
1039 if (neigh->dev->header_ops)
1040 update = neigh->dev->header_ops->cache_update;
1045 write_seqlock_bh(&hh->hh_lock);
1046 update(hh, neigh->dev, neigh->ha);
1047 write_sequnlock_bh(&hh->hh_lock);
1054 /* Generic update routine.
1055 -- lladdr is new lladdr or NULL, if it is not supplied.
1056 -- new is new state.
1058 NEIGH_UPDATE_F_OVERRIDE allows to override existing lladdr,
1060 NEIGH_UPDATE_F_WEAK_OVERRIDE will suspect existing "connected"
1061 lladdr instead of overriding it
1063 It also allows to retain current state
1064 if lladdr is unchanged.
1065 NEIGH_UPDATE_F_ADMIN means that the change is administrative.
1067 NEIGH_UPDATE_F_OVERRIDE_ISROUTER allows to override existing
1069 NEIGH_UPDATE_F_ISROUTER indicates if the neighbour is known as
1072 Caller MUST hold reference count on the entry.
1075 int neigh_update(struct neighbour *neigh, const u8 *lladdr, u8 new,
1081 struct net_device *dev;
1082 int update_isrouter = 0;
1084 write_lock_bh(&neigh->lock);
1087 old = neigh->nud_state;
1090 if (!(flags & NEIGH_UPDATE_F_ADMIN) &&
1091 (old & (NUD_NOARP | NUD_PERMANENT)))
1094 if (!(new & NUD_VALID)) {
1095 neigh_del_timer(neigh);
1096 if (old & NUD_CONNECTED)
1097 neigh_suspect(neigh);
1098 neigh->nud_state = new;
1100 notify = old & NUD_VALID;
1101 if ((old & (NUD_INCOMPLETE | NUD_PROBE)) &&
1102 (new & NUD_FAILED)) {
1103 neigh_invalidate(neigh);
1109 /* Compare new lladdr with cached one */
1110 if (!dev->addr_len) {
1111 /* First case: device needs no address. */
1113 } else if (lladdr) {
1114 /* The second case: if something is already cached
1115 and a new address is proposed:
1117 - if they are different, check override flag
1119 if ((old & NUD_VALID) &&
1120 !memcmp(lladdr, neigh->ha, dev->addr_len))
1123 /* No address is supplied; if we know something,
1124 use it, otherwise discard the request.
1127 if (!(old & NUD_VALID))
1132 if (new & NUD_CONNECTED)
1133 neigh->confirmed = jiffies;
1134 neigh->updated = jiffies;
1136 /* If entry was valid and address is not changed,
1137 do not change entry state, if new one is STALE.
1140 update_isrouter = flags & NEIGH_UPDATE_F_OVERRIDE_ISROUTER;
1141 if (old & NUD_VALID) {
1142 if (lladdr != neigh->ha && !(flags & NEIGH_UPDATE_F_OVERRIDE)) {
1143 update_isrouter = 0;
1144 if ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) &&
1145 (old & NUD_CONNECTED)) {
1151 if (lladdr == neigh->ha && new == NUD_STALE &&
1152 ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) ||
1153 (old & NUD_CONNECTED))
1160 neigh_del_timer(neigh);
1161 if (new & NUD_IN_TIMER)
1162 neigh_add_timer(neigh, (jiffies +
1163 ((new & NUD_REACHABLE) ?
1164 neigh->parms->reachable_time :
1166 neigh->nud_state = new;
1170 if (lladdr != neigh->ha) {
1171 write_seqlock(&neigh->ha_lock);
1172 memcpy(&neigh->ha, lladdr, dev->addr_len);
1173 write_sequnlock(&neigh->ha_lock);
1174 neigh_update_hhs(neigh);
1175 if (!(new & NUD_CONNECTED))
1176 neigh->confirmed = jiffies -
1177 (NEIGH_VAR(neigh->parms, BASE_REACHABLE_TIME) << 1);
1182 if (new & NUD_CONNECTED)
1183 neigh_connect(neigh);
1185 neigh_suspect(neigh);
1186 if (!(old & NUD_VALID)) {
1187 struct sk_buff *skb;
1189 /* Again: avoid dead loop if something went wrong */
1191 while (neigh->nud_state & NUD_VALID &&
1192 (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
1193 struct dst_entry *dst = skb_dst(skb);
1194 struct neighbour *n2, *n1 = neigh;
1195 write_unlock_bh(&neigh->lock);
1199 /* Why not just use 'neigh' as-is? The problem is that
1200 * things such as shaper, eql, and sch_teql can end up
1201 * using alternative, different, neigh objects to output
1202 * the packet in the output path. So what we need to do
1203 * here is re-lookup the top-level neigh in the path so
1204 * we can reinject the packet there.
1208 n2 = dst_neigh_lookup_skb(dst, skb);
1212 n1->output(n1, skb);
1217 write_lock_bh(&neigh->lock);
1219 __skb_queue_purge(&neigh->arp_queue);
1220 neigh->arp_queue_len_bytes = 0;
1223 if (update_isrouter) {
1224 neigh->flags = (flags & NEIGH_UPDATE_F_ISROUTER) ?
1225 (neigh->flags | NTF_ROUTER) :
1226 (neigh->flags & ~NTF_ROUTER);
1228 write_unlock_bh(&neigh->lock);
1231 neigh_update_notify(neigh);
1235 EXPORT_SYMBOL(neigh_update);
1237 /* Update the neigh to listen temporarily for probe responses, even if it is
1238 * in a NUD_FAILED state. The caller has to hold neigh->lock for writing.
1240 void __neigh_set_probe_once(struct neighbour *neigh)
1242 neigh->updated = jiffies;
1243 if (!(neigh->nud_state & NUD_FAILED))
1245 neigh->nud_state = NUD_INCOMPLETE;
1246 atomic_set(&neigh->probes, neigh_max_probes(neigh));
1247 neigh_add_timer(neigh,
1248 jiffies + NEIGH_VAR(neigh->parms, RETRANS_TIME));
1250 EXPORT_SYMBOL(__neigh_set_probe_once);
1252 struct neighbour *neigh_event_ns(struct neigh_table *tbl,
1253 u8 *lladdr, void *saddr,
1254 struct net_device *dev)
1256 struct neighbour *neigh = __neigh_lookup(tbl, saddr, dev,
1257 lladdr || !dev->addr_len);
1259 neigh_update(neigh, lladdr, NUD_STALE,
1260 NEIGH_UPDATE_F_OVERRIDE);
1263 EXPORT_SYMBOL(neigh_event_ns);
1265 /* called with read_lock_bh(&n->lock); */
1266 static void neigh_hh_init(struct neighbour *n, struct dst_entry *dst)
1268 struct net_device *dev = dst->dev;
1269 __be16 prot = dst->ops->protocol;
1270 struct hh_cache *hh = &n->hh;
1272 write_lock_bh(&n->lock);
1274 /* Only one thread can come in here and initialize the
1278 dev->header_ops->cache(n, hh, prot);
1280 write_unlock_bh(&n->lock);
1283 /* This function can be used in contexts, where only old dev_queue_xmit
1284 * worked, f.e. if you want to override normal output path (eql, shaper),
1285 * but resolution is not made yet.
1288 int neigh_compat_output(struct neighbour *neigh, struct sk_buff *skb)
1290 struct net_device *dev = skb->dev;
1292 __skb_pull(skb, skb_network_offset(skb));
1294 if (dev_hard_header(skb, dev, ntohs(skb->protocol), NULL, NULL,
1296 dev_rebuild_header(skb))
1299 return dev_queue_xmit(skb);
1301 EXPORT_SYMBOL(neigh_compat_output);
1303 /* Slow and careful. */
1305 int neigh_resolve_output(struct neighbour *neigh, struct sk_buff *skb)
1307 struct dst_entry *dst = skb_dst(skb);
1313 if (!neigh_event_send(neigh, skb)) {
1315 struct net_device *dev = neigh->dev;
1318 if (dev->header_ops->cache && !neigh->hh.hh_len)
1319 neigh_hh_init(neigh, dst);
1322 __skb_pull(skb, skb_network_offset(skb));
1323 seq = read_seqbegin(&neigh->ha_lock);
1324 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1325 neigh->ha, NULL, skb->len);
1326 } while (read_seqretry(&neigh->ha_lock, seq));
1329 rc = dev_queue_xmit(skb);
1336 neigh_dbg(1, "%s: dst=%p neigh=%p\n", __func__, dst, neigh);
1342 EXPORT_SYMBOL(neigh_resolve_output);
1344 /* As fast as possible without hh cache */
1346 int neigh_connected_output(struct neighbour *neigh, struct sk_buff *skb)
1348 struct net_device *dev = neigh->dev;
1353 __skb_pull(skb, skb_network_offset(skb));
1354 seq = read_seqbegin(&neigh->ha_lock);
1355 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1356 neigh->ha, NULL, skb->len);
1357 } while (read_seqretry(&neigh->ha_lock, seq));
1360 err = dev_queue_xmit(skb);
1367 EXPORT_SYMBOL(neigh_connected_output);
1369 int neigh_direct_output(struct neighbour *neigh, struct sk_buff *skb)
1371 return dev_queue_xmit(skb);
1373 EXPORT_SYMBOL(neigh_direct_output);
1375 static void neigh_proxy_process(unsigned long arg)
1377 struct neigh_table *tbl = (struct neigh_table *)arg;
1378 long sched_next = 0;
1379 unsigned long now = jiffies;
1380 struct sk_buff *skb, *n;
1382 spin_lock(&tbl->proxy_queue.lock);
1384 skb_queue_walk_safe(&tbl->proxy_queue, skb, n) {
1385 long tdif = NEIGH_CB(skb)->sched_next - now;
1388 struct net_device *dev = skb->dev;
1390 __skb_unlink(skb, &tbl->proxy_queue);
1391 if (tbl->proxy_redo && netif_running(dev)) {
1393 tbl->proxy_redo(skb);
1400 } else if (!sched_next || tdif < sched_next)
1403 del_timer(&tbl->proxy_timer);
1405 mod_timer(&tbl->proxy_timer, jiffies + sched_next);
1406 spin_unlock(&tbl->proxy_queue.lock);
1409 void pneigh_enqueue(struct neigh_table *tbl, struct neigh_parms *p,
1410 struct sk_buff *skb)
1412 unsigned long now = jiffies;
1414 unsigned long sched_next = now + (prandom_u32() %
1415 NEIGH_VAR(p, PROXY_DELAY));
1417 if (tbl->proxy_queue.qlen > NEIGH_VAR(p, PROXY_QLEN)) {
1422 NEIGH_CB(skb)->sched_next = sched_next;
1423 NEIGH_CB(skb)->flags |= LOCALLY_ENQUEUED;
1425 spin_lock(&tbl->proxy_queue.lock);
1426 if (del_timer(&tbl->proxy_timer)) {
1427 if (time_before(tbl->proxy_timer.expires, sched_next))
1428 sched_next = tbl->proxy_timer.expires;
1432 __skb_queue_tail(&tbl->proxy_queue, skb);
1433 mod_timer(&tbl->proxy_timer, sched_next);
1434 spin_unlock(&tbl->proxy_queue.lock);
1436 EXPORT_SYMBOL(pneigh_enqueue);
1438 static inline struct neigh_parms *lookup_neigh_parms(struct neigh_table *tbl,
1439 struct net *net, int ifindex)
1441 struct neigh_parms *p;
1443 list_for_each_entry(p, &tbl->parms_list, list) {
1444 if ((p->dev && p->dev->ifindex == ifindex && net_eq(neigh_parms_net(p), net)) ||
1445 (!p->dev && !ifindex && net_eq(net, &init_net)))
1452 struct neigh_parms *neigh_parms_alloc(struct net_device *dev,
1453 struct neigh_table *tbl)
1455 struct neigh_parms *p;
1456 struct net *net = dev_net(dev);
1457 const struct net_device_ops *ops = dev->netdev_ops;
1459 p = kmemdup(&tbl->parms, sizeof(*p), GFP_KERNEL);
1462 atomic_set(&p->refcnt, 1);
1464 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
1467 write_pnet(&p->net, hold_net(net));
1468 p->sysctl_table = NULL;
1470 if (ops->ndo_neigh_setup && ops->ndo_neigh_setup(dev, p)) {
1477 write_lock_bh(&tbl->lock);
1478 list_add(&p->list, &tbl->parms.list);
1479 write_unlock_bh(&tbl->lock);
1481 neigh_parms_data_state_cleanall(p);
1485 EXPORT_SYMBOL(neigh_parms_alloc);
1487 static void neigh_rcu_free_parms(struct rcu_head *head)
1489 struct neigh_parms *parms =
1490 container_of(head, struct neigh_parms, rcu_head);
1492 neigh_parms_put(parms);
1495 void neigh_parms_release(struct neigh_table *tbl, struct neigh_parms *parms)
1497 if (!parms || parms == &tbl->parms)
1499 write_lock_bh(&tbl->lock);
1500 list_del(&parms->list);
1502 write_unlock_bh(&tbl->lock);
1504 dev_put(parms->dev);
1505 call_rcu(&parms->rcu_head, neigh_rcu_free_parms);
1507 EXPORT_SYMBOL(neigh_parms_release);
1509 static void neigh_parms_destroy(struct neigh_parms *parms)
1511 release_net(neigh_parms_net(parms));
1515 static struct lock_class_key neigh_table_proxy_queue_class;
1517 static struct neigh_table *neigh_tables[NEIGH_NR_TABLES] __read_mostly;
1519 void neigh_table_init(int index, struct neigh_table *tbl)
1521 unsigned long now = jiffies;
1522 unsigned long phsize;
1524 INIT_LIST_HEAD(&tbl->parms_list);
1525 list_add(&tbl->parms.list, &tbl->parms_list);
1526 write_pnet(&tbl->parms.net, &init_net);
1527 atomic_set(&tbl->parms.refcnt, 1);
1528 tbl->parms.reachable_time =
1529 neigh_rand_reach_time(NEIGH_VAR(&tbl->parms, BASE_REACHABLE_TIME));
1531 tbl->stats = alloc_percpu(struct neigh_statistics);
1533 panic("cannot create neighbour cache statistics");
1535 #ifdef CONFIG_PROC_FS
1536 if (!proc_create_data(tbl->id, 0, init_net.proc_net_stat,
1537 &neigh_stat_seq_fops, tbl))
1538 panic("cannot create neighbour proc dir entry");
1541 RCU_INIT_POINTER(tbl->nht, neigh_hash_alloc(3));
1543 phsize = (PNEIGH_HASHMASK + 1) * sizeof(struct pneigh_entry *);
1544 tbl->phash_buckets = kzalloc(phsize, GFP_KERNEL);
1546 if (!tbl->nht || !tbl->phash_buckets)
1547 panic("cannot allocate neighbour cache hashes");
1549 if (!tbl->entry_size)
1550 tbl->entry_size = ALIGN(offsetof(struct neighbour, primary_key) +
1551 tbl->key_len, NEIGH_PRIV_ALIGN);
1553 WARN_ON(tbl->entry_size % NEIGH_PRIV_ALIGN);
1555 rwlock_init(&tbl->lock);
1556 INIT_DEFERRABLE_WORK(&tbl->gc_work, neigh_periodic_work);
1557 queue_delayed_work(system_power_efficient_wq, &tbl->gc_work,
1558 tbl->parms.reachable_time);
1559 setup_timer(&tbl->proxy_timer, neigh_proxy_process, (unsigned long)tbl);
1560 skb_queue_head_init_class(&tbl->proxy_queue,
1561 &neigh_table_proxy_queue_class);
1563 tbl->last_flush = now;
1564 tbl->last_rand = now + tbl->parms.reachable_time * 20;
1566 neigh_tables[index] = tbl;
1568 EXPORT_SYMBOL(neigh_table_init);
1570 int neigh_table_clear(int index, struct neigh_table *tbl)
1572 neigh_tables[index] = NULL;
1573 /* It is not clean... Fix it to unload IPv6 module safely */
1574 cancel_delayed_work_sync(&tbl->gc_work);
1575 del_timer_sync(&tbl->proxy_timer);
1576 pneigh_queue_purge(&tbl->proxy_queue);
1577 neigh_ifdown(tbl, NULL);
1578 if (atomic_read(&tbl->entries))
1579 pr_crit("neighbour leakage\n");
1581 call_rcu(&rcu_dereference_protected(tbl->nht, 1)->rcu,
1582 neigh_hash_free_rcu);
1585 kfree(tbl->phash_buckets);
1586 tbl->phash_buckets = NULL;
1588 remove_proc_entry(tbl->id, init_net.proc_net_stat);
1590 free_percpu(tbl->stats);
1595 EXPORT_SYMBOL(neigh_table_clear);
1597 static struct neigh_table *neigh_find_table(int family)
1599 struct neigh_table *tbl = NULL;
1603 tbl = neigh_tables[NEIGH_ARP_TABLE];
1606 tbl = neigh_tables[NEIGH_ND_TABLE];
1609 tbl = neigh_tables[NEIGH_DN_TABLE];
1616 static int neigh_delete(struct sk_buff *skb, struct nlmsghdr *nlh)
1618 struct net *net = sock_net(skb->sk);
1620 struct nlattr *dst_attr;
1621 struct neigh_table *tbl;
1622 struct neighbour *neigh;
1623 struct net_device *dev = NULL;
1627 if (nlmsg_len(nlh) < sizeof(*ndm))
1630 dst_attr = nlmsg_find_attr(nlh, sizeof(*ndm), NDA_DST);
1631 if (dst_attr == NULL)
1634 ndm = nlmsg_data(nlh);
1635 if (ndm->ndm_ifindex) {
1636 dev = __dev_get_by_index(net, ndm->ndm_ifindex);
1643 tbl = neigh_find_table(ndm->ndm_family);
1645 return -EAFNOSUPPORT;
1647 if (nla_len(dst_attr) < tbl->key_len)
1650 if (ndm->ndm_flags & NTF_PROXY) {
1651 err = pneigh_delete(tbl, net, nla_data(dst_attr), dev);
1658 neigh = neigh_lookup(tbl, nla_data(dst_attr), dev);
1659 if (neigh == NULL) {
1664 err = neigh_update(neigh, NULL, NUD_FAILED,
1665 NEIGH_UPDATE_F_OVERRIDE |
1666 NEIGH_UPDATE_F_ADMIN);
1667 neigh_release(neigh);
1673 static int neigh_add(struct sk_buff *skb, struct nlmsghdr *nlh)
1675 int flags = NEIGH_UPDATE_F_ADMIN | NEIGH_UPDATE_F_OVERRIDE;
1676 struct net *net = sock_net(skb->sk);
1678 struct nlattr *tb[NDA_MAX+1];
1679 struct neigh_table *tbl;
1680 struct net_device *dev = NULL;
1681 struct neighbour *neigh;
1686 err = nlmsg_parse(nlh, sizeof(*ndm), tb, NDA_MAX, NULL);
1691 if (tb[NDA_DST] == NULL)
1694 ndm = nlmsg_data(nlh);
1695 if (ndm->ndm_ifindex) {
1696 dev = __dev_get_by_index(net, ndm->ndm_ifindex);
1702 if (tb[NDA_LLADDR] && nla_len(tb[NDA_LLADDR]) < dev->addr_len)
1706 tbl = neigh_find_table(ndm->ndm_family);
1708 return -EAFNOSUPPORT;
1710 if (nla_len(tb[NDA_DST]) < tbl->key_len)
1712 dst = nla_data(tb[NDA_DST]);
1713 lladdr = tb[NDA_LLADDR] ? nla_data(tb[NDA_LLADDR]) : NULL;
1715 if (ndm->ndm_flags & NTF_PROXY) {
1716 struct pneigh_entry *pn;
1719 pn = pneigh_lookup(tbl, net, dst, dev, 1);
1721 pn->flags = ndm->ndm_flags;
1730 neigh = neigh_lookup(tbl, dst, dev);
1731 if (neigh == NULL) {
1732 if (!(nlh->nlmsg_flags & NLM_F_CREATE)) {
1737 neigh = __neigh_lookup_errno(tbl, dst, dev);
1738 if (IS_ERR(neigh)) {
1739 err = PTR_ERR(neigh);
1743 if (nlh->nlmsg_flags & NLM_F_EXCL) {
1745 neigh_release(neigh);
1749 if (!(nlh->nlmsg_flags & NLM_F_REPLACE))
1750 flags &= ~NEIGH_UPDATE_F_OVERRIDE;
1753 if (ndm->ndm_flags & NTF_USE) {
1754 neigh_event_send(neigh, NULL);
1757 err = neigh_update(neigh, lladdr, ndm->ndm_state, flags);
1758 neigh_release(neigh);
1764 static int neightbl_fill_parms(struct sk_buff *skb, struct neigh_parms *parms)
1766 struct nlattr *nest;
1768 nest = nla_nest_start(skb, NDTA_PARMS);
1773 nla_put_u32(skb, NDTPA_IFINDEX, parms->dev->ifindex)) ||
1774 nla_put_u32(skb, NDTPA_REFCNT, atomic_read(&parms->refcnt)) ||
1775 nla_put_u32(skb, NDTPA_QUEUE_LENBYTES,
1776 NEIGH_VAR(parms, QUEUE_LEN_BYTES)) ||
1777 /* approximative value for deprecated QUEUE_LEN (in packets) */
1778 nla_put_u32(skb, NDTPA_QUEUE_LEN,
1779 NEIGH_VAR(parms, QUEUE_LEN_BYTES) / SKB_TRUESIZE(ETH_FRAME_LEN)) ||
1780 nla_put_u32(skb, NDTPA_PROXY_QLEN, NEIGH_VAR(parms, PROXY_QLEN)) ||
1781 nla_put_u32(skb, NDTPA_APP_PROBES, NEIGH_VAR(parms, APP_PROBES)) ||
1782 nla_put_u32(skb, NDTPA_UCAST_PROBES,
1783 NEIGH_VAR(parms, UCAST_PROBES)) ||
1784 nla_put_u32(skb, NDTPA_MCAST_PROBES,
1785 NEIGH_VAR(parms, MCAST_PROBES)) ||
1786 nla_put_msecs(skb, NDTPA_REACHABLE_TIME, parms->reachable_time) ||
1787 nla_put_msecs(skb, NDTPA_BASE_REACHABLE_TIME,
1788 NEIGH_VAR(parms, BASE_REACHABLE_TIME)) ||
1789 nla_put_msecs(skb, NDTPA_GC_STALETIME,
1790 NEIGH_VAR(parms, GC_STALETIME)) ||
1791 nla_put_msecs(skb, NDTPA_DELAY_PROBE_TIME,
1792 NEIGH_VAR(parms, DELAY_PROBE_TIME)) ||
1793 nla_put_msecs(skb, NDTPA_RETRANS_TIME,
1794 NEIGH_VAR(parms, RETRANS_TIME)) ||
1795 nla_put_msecs(skb, NDTPA_ANYCAST_DELAY,
1796 NEIGH_VAR(parms, ANYCAST_DELAY)) ||
1797 nla_put_msecs(skb, NDTPA_PROXY_DELAY,
1798 NEIGH_VAR(parms, PROXY_DELAY)) ||
1799 nla_put_msecs(skb, NDTPA_LOCKTIME,
1800 NEIGH_VAR(parms, LOCKTIME)))
1801 goto nla_put_failure;
1802 return nla_nest_end(skb, nest);
1805 nla_nest_cancel(skb, nest);
1809 static int neightbl_fill_info(struct sk_buff *skb, struct neigh_table *tbl,
1810 u32 pid, u32 seq, int type, int flags)
1812 struct nlmsghdr *nlh;
1813 struct ndtmsg *ndtmsg;
1815 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
1819 ndtmsg = nlmsg_data(nlh);
1821 read_lock_bh(&tbl->lock);
1822 ndtmsg->ndtm_family = tbl->family;
1823 ndtmsg->ndtm_pad1 = 0;
1824 ndtmsg->ndtm_pad2 = 0;
1826 if (nla_put_string(skb, NDTA_NAME, tbl->id) ||
1827 nla_put_msecs(skb, NDTA_GC_INTERVAL, tbl->gc_interval) ||
1828 nla_put_u32(skb, NDTA_THRESH1, tbl->gc_thresh1) ||
1829 nla_put_u32(skb, NDTA_THRESH2, tbl->gc_thresh2) ||
1830 nla_put_u32(skb, NDTA_THRESH3, tbl->gc_thresh3))
1831 goto nla_put_failure;
1833 unsigned long now = jiffies;
1834 unsigned int flush_delta = now - tbl->last_flush;
1835 unsigned int rand_delta = now - tbl->last_rand;
1836 struct neigh_hash_table *nht;
1837 struct ndt_config ndc = {
1838 .ndtc_key_len = tbl->key_len,
1839 .ndtc_entry_size = tbl->entry_size,
1840 .ndtc_entries = atomic_read(&tbl->entries),
1841 .ndtc_last_flush = jiffies_to_msecs(flush_delta),
1842 .ndtc_last_rand = jiffies_to_msecs(rand_delta),
1843 .ndtc_proxy_qlen = tbl->proxy_queue.qlen,
1847 nht = rcu_dereference_bh(tbl->nht);
1848 ndc.ndtc_hash_rnd = nht->hash_rnd[0];
1849 ndc.ndtc_hash_mask = ((1 << nht->hash_shift) - 1);
1850 rcu_read_unlock_bh();
1852 if (nla_put(skb, NDTA_CONFIG, sizeof(ndc), &ndc))
1853 goto nla_put_failure;
1858 struct ndt_stats ndst;
1860 memset(&ndst, 0, sizeof(ndst));
1862 for_each_possible_cpu(cpu) {
1863 struct neigh_statistics *st;
1865 st = per_cpu_ptr(tbl->stats, cpu);
1866 ndst.ndts_allocs += st->allocs;
1867 ndst.ndts_destroys += st->destroys;
1868 ndst.ndts_hash_grows += st->hash_grows;
1869 ndst.ndts_res_failed += st->res_failed;
1870 ndst.ndts_lookups += st->lookups;
1871 ndst.ndts_hits += st->hits;
1872 ndst.ndts_rcv_probes_mcast += st->rcv_probes_mcast;
1873 ndst.ndts_rcv_probes_ucast += st->rcv_probes_ucast;
1874 ndst.ndts_periodic_gc_runs += st->periodic_gc_runs;
1875 ndst.ndts_forced_gc_runs += st->forced_gc_runs;
1878 if (nla_put(skb, NDTA_STATS, sizeof(ndst), &ndst))
1879 goto nla_put_failure;
1882 BUG_ON(tbl->parms.dev);
1883 if (neightbl_fill_parms(skb, &tbl->parms) < 0)
1884 goto nla_put_failure;
1886 read_unlock_bh(&tbl->lock);
1887 return nlmsg_end(skb, nlh);
1890 read_unlock_bh(&tbl->lock);
1891 nlmsg_cancel(skb, nlh);
1895 static int neightbl_fill_param_info(struct sk_buff *skb,
1896 struct neigh_table *tbl,
1897 struct neigh_parms *parms,
1898 u32 pid, u32 seq, int type,
1901 struct ndtmsg *ndtmsg;
1902 struct nlmsghdr *nlh;
1904 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
1908 ndtmsg = nlmsg_data(nlh);
1910 read_lock_bh(&tbl->lock);
1911 ndtmsg->ndtm_family = tbl->family;
1912 ndtmsg->ndtm_pad1 = 0;
1913 ndtmsg->ndtm_pad2 = 0;
1915 if (nla_put_string(skb, NDTA_NAME, tbl->id) < 0 ||
1916 neightbl_fill_parms(skb, parms) < 0)
1919 read_unlock_bh(&tbl->lock);
1920 return nlmsg_end(skb, nlh);
1922 read_unlock_bh(&tbl->lock);
1923 nlmsg_cancel(skb, nlh);
1927 static const struct nla_policy nl_neightbl_policy[NDTA_MAX+1] = {
1928 [NDTA_NAME] = { .type = NLA_STRING },
1929 [NDTA_THRESH1] = { .type = NLA_U32 },
1930 [NDTA_THRESH2] = { .type = NLA_U32 },
1931 [NDTA_THRESH3] = { .type = NLA_U32 },
1932 [NDTA_GC_INTERVAL] = { .type = NLA_U64 },
1933 [NDTA_PARMS] = { .type = NLA_NESTED },
1936 static const struct nla_policy nl_ntbl_parm_policy[NDTPA_MAX+1] = {
1937 [NDTPA_IFINDEX] = { .type = NLA_U32 },
1938 [NDTPA_QUEUE_LEN] = { .type = NLA_U32 },
1939 [NDTPA_PROXY_QLEN] = { .type = NLA_U32 },
1940 [NDTPA_APP_PROBES] = { .type = NLA_U32 },
1941 [NDTPA_UCAST_PROBES] = { .type = NLA_U32 },
1942 [NDTPA_MCAST_PROBES] = { .type = NLA_U32 },
1943 [NDTPA_BASE_REACHABLE_TIME] = { .type = NLA_U64 },
1944 [NDTPA_GC_STALETIME] = { .type = NLA_U64 },
1945 [NDTPA_DELAY_PROBE_TIME] = { .type = NLA_U64 },
1946 [NDTPA_RETRANS_TIME] = { .type = NLA_U64 },
1947 [NDTPA_ANYCAST_DELAY] = { .type = NLA_U64 },
1948 [NDTPA_PROXY_DELAY] = { .type = NLA_U64 },
1949 [NDTPA_LOCKTIME] = { .type = NLA_U64 },
1952 static int neightbl_set(struct sk_buff *skb, struct nlmsghdr *nlh)
1954 struct net *net = sock_net(skb->sk);
1955 struct neigh_table *tbl;
1956 struct ndtmsg *ndtmsg;
1957 struct nlattr *tb[NDTA_MAX+1];
1961 err = nlmsg_parse(nlh, sizeof(*ndtmsg), tb, NDTA_MAX,
1962 nl_neightbl_policy);
1966 if (tb[NDTA_NAME] == NULL) {
1971 ndtmsg = nlmsg_data(nlh);
1973 for (tidx = 0; tidx < NEIGH_NR_TABLES; tidx++) {
1974 tbl = neigh_tables[tidx];
1977 if (ndtmsg->ndtm_family && tbl->family != ndtmsg->ndtm_family)
1979 if (nla_strcmp(tb[NDTA_NAME], tbl->id) == 0) {
1989 * We acquire tbl->lock to be nice to the periodic timers and
1990 * make sure they always see a consistent set of values.
1992 write_lock_bh(&tbl->lock);
1994 if (tb[NDTA_PARMS]) {
1995 struct nlattr *tbp[NDTPA_MAX+1];
1996 struct neigh_parms *p;
1999 err = nla_parse_nested(tbp, NDTPA_MAX, tb[NDTA_PARMS],
2000 nl_ntbl_parm_policy);
2002 goto errout_tbl_lock;
2004 if (tbp[NDTPA_IFINDEX])
2005 ifindex = nla_get_u32(tbp[NDTPA_IFINDEX]);
2007 p = lookup_neigh_parms(tbl, net, ifindex);
2010 goto errout_tbl_lock;
2013 for (i = 1; i <= NDTPA_MAX; i++) {
2018 case NDTPA_QUEUE_LEN:
2019 NEIGH_VAR_SET(p, QUEUE_LEN_BYTES,
2020 nla_get_u32(tbp[i]) *
2021 SKB_TRUESIZE(ETH_FRAME_LEN));
2023 case NDTPA_QUEUE_LENBYTES:
2024 NEIGH_VAR_SET(p, QUEUE_LEN_BYTES,
2025 nla_get_u32(tbp[i]));
2027 case NDTPA_PROXY_QLEN:
2028 NEIGH_VAR_SET(p, PROXY_QLEN,
2029 nla_get_u32(tbp[i]));
2031 case NDTPA_APP_PROBES:
2032 NEIGH_VAR_SET(p, APP_PROBES,
2033 nla_get_u32(tbp[i]));
2035 case NDTPA_UCAST_PROBES:
2036 NEIGH_VAR_SET(p, UCAST_PROBES,
2037 nla_get_u32(tbp[i]));
2039 case NDTPA_MCAST_PROBES:
2040 NEIGH_VAR_SET(p, MCAST_PROBES,
2041 nla_get_u32(tbp[i]));
2043 case NDTPA_BASE_REACHABLE_TIME:
2044 NEIGH_VAR_SET(p, BASE_REACHABLE_TIME,
2045 nla_get_msecs(tbp[i]));
2046 /* update reachable_time as well, otherwise, the change will
2047 * only be effective after the next time neigh_periodic_work
2048 * decides to recompute it (can be multiple minutes)
2051 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
2053 case NDTPA_GC_STALETIME:
2054 NEIGH_VAR_SET(p, GC_STALETIME,
2055 nla_get_msecs(tbp[i]));
2057 case NDTPA_DELAY_PROBE_TIME:
2058 NEIGH_VAR_SET(p, DELAY_PROBE_TIME,
2059 nla_get_msecs(tbp[i]));
2061 case NDTPA_RETRANS_TIME:
2062 NEIGH_VAR_SET(p, RETRANS_TIME,
2063 nla_get_msecs(tbp[i]));
2065 case NDTPA_ANYCAST_DELAY:
2066 NEIGH_VAR_SET(p, ANYCAST_DELAY,
2067 nla_get_msecs(tbp[i]));
2069 case NDTPA_PROXY_DELAY:
2070 NEIGH_VAR_SET(p, PROXY_DELAY,
2071 nla_get_msecs(tbp[i]));
2073 case NDTPA_LOCKTIME:
2074 NEIGH_VAR_SET(p, LOCKTIME,
2075 nla_get_msecs(tbp[i]));
2082 if ((tb[NDTA_THRESH1] || tb[NDTA_THRESH2] ||
2083 tb[NDTA_THRESH3] || tb[NDTA_GC_INTERVAL]) &&
2084 !net_eq(net, &init_net))
2085 goto errout_tbl_lock;
2087 if (tb[NDTA_THRESH1])
2088 tbl->gc_thresh1 = nla_get_u32(tb[NDTA_THRESH1]);
2090 if (tb[NDTA_THRESH2])
2091 tbl->gc_thresh2 = nla_get_u32(tb[NDTA_THRESH2]);
2093 if (tb[NDTA_THRESH3])
2094 tbl->gc_thresh3 = nla_get_u32(tb[NDTA_THRESH3]);
2096 if (tb[NDTA_GC_INTERVAL])
2097 tbl->gc_interval = nla_get_msecs(tb[NDTA_GC_INTERVAL]);
2102 write_unlock_bh(&tbl->lock);
2107 static int neightbl_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2109 struct net *net = sock_net(skb->sk);
2110 int family, tidx, nidx = 0;
2111 int tbl_skip = cb->args[0];
2112 int neigh_skip = cb->args[1];
2113 struct neigh_table *tbl;
2115 family = ((struct rtgenmsg *) nlmsg_data(cb->nlh))->rtgen_family;
2117 for (tidx = 0; tidx < NEIGH_NR_TABLES; tidx++) {
2118 struct neigh_parms *p;
2120 tbl = neigh_tables[tidx];
2124 if (tidx < tbl_skip || (family && tbl->family != family))
2127 if (neightbl_fill_info(skb, tbl, NETLINK_CB(cb->skb).portid,
2128 cb->nlh->nlmsg_seq, RTM_NEWNEIGHTBL,
2133 p = list_next_entry(&tbl->parms, list);
2134 list_for_each_entry_from(p, &tbl->parms_list, list) {
2135 if (!net_eq(neigh_parms_net(p), net))
2138 if (nidx < neigh_skip)
2141 if (neightbl_fill_param_info(skb, tbl, p,
2142 NETLINK_CB(cb->skb).portid,
2160 static int neigh_fill_info(struct sk_buff *skb, struct neighbour *neigh,
2161 u32 pid, u32 seq, int type, unsigned int flags)
2163 unsigned long now = jiffies;
2164 struct nda_cacheinfo ci;
2165 struct nlmsghdr *nlh;
2168 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2172 ndm = nlmsg_data(nlh);
2173 ndm->ndm_family = neigh->ops->family;
2176 ndm->ndm_flags = neigh->flags;
2177 ndm->ndm_type = neigh->type;
2178 ndm->ndm_ifindex = neigh->dev->ifindex;
2180 if (nla_put(skb, NDA_DST, neigh->tbl->key_len, neigh->primary_key))
2181 goto nla_put_failure;
2183 read_lock_bh(&neigh->lock);
2184 ndm->ndm_state = neigh->nud_state;
2185 if (neigh->nud_state & NUD_VALID) {
2186 char haddr[MAX_ADDR_LEN];
2188 neigh_ha_snapshot(haddr, neigh, neigh->dev);
2189 if (nla_put(skb, NDA_LLADDR, neigh->dev->addr_len, haddr) < 0) {
2190 read_unlock_bh(&neigh->lock);
2191 goto nla_put_failure;
2195 ci.ndm_used = jiffies_to_clock_t(now - neigh->used);
2196 ci.ndm_confirmed = jiffies_to_clock_t(now - neigh->confirmed);
2197 ci.ndm_updated = jiffies_to_clock_t(now - neigh->updated);
2198 ci.ndm_refcnt = atomic_read(&neigh->refcnt) - 1;
2199 read_unlock_bh(&neigh->lock);
2201 if (nla_put_u32(skb, NDA_PROBES, atomic_read(&neigh->probes)) ||
2202 nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
2203 goto nla_put_failure;
2205 return nlmsg_end(skb, nlh);
2208 nlmsg_cancel(skb, nlh);
2212 static int pneigh_fill_info(struct sk_buff *skb, struct pneigh_entry *pn,
2213 u32 pid, u32 seq, int type, unsigned int flags,
2214 struct neigh_table *tbl)
2216 struct nlmsghdr *nlh;
2219 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2223 ndm = nlmsg_data(nlh);
2224 ndm->ndm_family = tbl->family;
2227 ndm->ndm_flags = pn->flags | NTF_PROXY;
2228 ndm->ndm_type = RTN_UNICAST;
2229 ndm->ndm_ifindex = pn->dev->ifindex;
2230 ndm->ndm_state = NUD_NONE;
2232 if (nla_put(skb, NDA_DST, tbl->key_len, pn->key))
2233 goto nla_put_failure;
2235 return nlmsg_end(skb, nlh);
2238 nlmsg_cancel(skb, nlh);
2242 static void neigh_update_notify(struct neighbour *neigh)
2244 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, neigh);
2245 __neigh_notify(neigh, RTM_NEWNEIGH, 0);
2248 static int neigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2249 struct netlink_callback *cb)
2251 struct net *net = sock_net(skb->sk);
2252 struct neighbour *n;
2253 int rc, h, s_h = cb->args[1];
2254 int idx, s_idx = idx = cb->args[2];
2255 struct neigh_hash_table *nht;
2258 nht = rcu_dereference_bh(tbl->nht);
2260 for (h = s_h; h < (1 << nht->hash_shift); h++) {
2263 for (n = rcu_dereference_bh(nht->hash_buckets[h]), idx = 0;
2265 n = rcu_dereference_bh(n->next)) {
2266 if (!net_eq(dev_net(n->dev), net))
2270 if (neigh_fill_info(skb, n, NETLINK_CB(cb->skb).portid,
2273 NLM_F_MULTI) <= 0) {
2283 rcu_read_unlock_bh();
2289 static int pneigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2290 struct netlink_callback *cb)
2292 struct pneigh_entry *n;
2293 struct net *net = sock_net(skb->sk);
2294 int rc, h, s_h = cb->args[3];
2295 int idx, s_idx = idx = cb->args[4];
2297 read_lock_bh(&tbl->lock);
2299 for (h = s_h; h <= PNEIGH_HASHMASK; h++) {
2302 for (n = tbl->phash_buckets[h], idx = 0; n; n = n->next) {
2303 if (dev_net(n->dev) != net)
2307 if (pneigh_fill_info(skb, n, NETLINK_CB(cb->skb).portid,
2310 NLM_F_MULTI, tbl) <= 0) {
2311 read_unlock_bh(&tbl->lock);
2320 read_unlock_bh(&tbl->lock);
2329 static int neigh_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2331 struct neigh_table *tbl;
2336 family = ((struct rtgenmsg *) nlmsg_data(cb->nlh))->rtgen_family;
2338 /* check for full ndmsg structure presence, family member is
2339 * the same for both structures
2341 if (nlmsg_len(cb->nlh) >= sizeof(struct ndmsg) &&
2342 ((struct ndmsg *) nlmsg_data(cb->nlh))->ndm_flags == NTF_PROXY)
2347 for (t = 0; t < NEIGH_NR_TABLES; t++) {
2348 tbl = neigh_tables[t];
2352 if (t < s_t || (family && tbl->family != family))
2355 memset(&cb->args[1], 0, sizeof(cb->args) -
2356 sizeof(cb->args[0]));
2358 err = pneigh_dump_table(tbl, skb, cb);
2360 err = neigh_dump_table(tbl, skb, cb);
2369 void neigh_for_each(struct neigh_table *tbl, void (*cb)(struct neighbour *, void *), void *cookie)
2372 struct neigh_hash_table *nht;
2375 nht = rcu_dereference_bh(tbl->nht);
2377 read_lock(&tbl->lock); /* avoid resizes */
2378 for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
2379 struct neighbour *n;
2381 for (n = rcu_dereference_bh(nht->hash_buckets[chain]);
2383 n = rcu_dereference_bh(n->next))
2386 read_unlock(&tbl->lock);
2387 rcu_read_unlock_bh();
2389 EXPORT_SYMBOL(neigh_for_each);
2391 /* The tbl->lock must be held as a writer and BH disabled. */
2392 void __neigh_for_each_release(struct neigh_table *tbl,
2393 int (*cb)(struct neighbour *))
2396 struct neigh_hash_table *nht;
2398 nht = rcu_dereference_protected(tbl->nht,
2399 lockdep_is_held(&tbl->lock));
2400 for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
2401 struct neighbour *n;
2402 struct neighbour __rcu **np;
2404 np = &nht->hash_buckets[chain];
2405 while ((n = rcu_dereference_protected(*np,
2406 lockdep_is_held(&tbl->lock))) != NULL) {
2409 write_lock(&n->lock);
2412 rcu_assign_pointer(*np,
2413 rcu_dereference_protected(n->next,
2414 lockdep_is_held(&tbl->lock)));
2418 write_unlock(&n->lock);
2420 neigh_cleanup_and_release(n);
2424 EXPORT_SYMBOL(__neigh_for_each_release);
2426 #ifdef CONFIG_PROC_FS
2428 static struct neighbour *neigh_get_first(struct seq_file *seq)
2430 struct neigh_seq_state *state = seq->private;
2431 struct net *net = seq_file_net(seq);
2432 struct neigh_hash_table *nht = state->nht;
2433 struct neighbour *n = NULL;
2434 int bucket = state->bucket;
2436 state->flags &= ~NEIGH_SEQ_IS_PNEIGH;
2437 for (bucket = 0; bucket < (1 << nht->hash_shift); bucket++) {
2438 n = rcu_dereference_bh(nht->hash_buckets[bucket]);
2441 if (!net_eq(dev_net(n->dev), net))
2443 if (state->neigh_sub_iter) {
2447 v = state->neigh_sub_iter(state, n, &fakep);
2451 if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
2453 if (n->nud_state & ~NUD_NOARP)
2456 n = rcu_dereference_bh(n->next);
2462 state->bucket = bucket;
2467 static struct neighbour *neigh_get_next(struct seq_file *seq,
2468 struct neighbour *n,
2471 struct neigh_seq_state *state = seq->private;
2472 struct net *net = seq_file_net(seq);
2473 struct neigh_hash_table *nht = state->nht;
2475 if (state->neigh_sub_iter) {
2476 void *v = state->neigh_sub_iter(state, n, pos);
2480 n = rcu_dereference_bh(n->next);
2484 if (!net_eq(dev_net(n->dev), net))
2486 if (state->neigh_sub_iter) {
2487 void *v = state->neigh_sub_iter(state, n, pos);
2492 if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
2495 if (n->nud_state & ~NUD_NOARP)
2498 n = rcu_dereference_bh(n->next);
2504 if (++state->bucket >= (1 << nht->hash_shift))
2507 n = rcu_dereference_bh(nht->hash_buckets[state->bucket]);
2515 static struct neighbour *neigh_get_idx(struct seq_file *seq, loff_t *pos)
2517 struct neighbour *n = neigh_get_first(seq);
2522 n = neigh_get_next(seq, n, pos);
2527 return *pos ? NULL : n;
2530 static struct pneigh_entry *pneigh_get_first(struct seq_file *seq)
2532 struct neigh_seq_state *state = seq->private;
2533 struct net *net = seq_file_net(seq);
2534 struct neigh_table *tbl = state->tbl;
2535 struct pneigh_entry *pn = NULL;
2536 int bucket = state->bucket;
2538 state->flags |= NEIGH_SEQ_IS_PNEIGH;
2539 for (bucket = 0; bucket <= PNEIGH_HASHMASK; bucket++) {
2540 pn = tbl->phash_buckets[bucket];
2541 while (pn && !net_eq(pneigh_net(pn), net))
2546 state->bucket = bucket;
2551 static struct pneigh_entry *pneigh_get_next(struct seq_file *seq,
2552 struct pneigh_entry *pn,
2555 struct neigh_seq_state *state = seq->private;
2556 struct net *net = seq_file_net(seq);
2557 struct neigh_table *tbl = state->tbl;
2561 } while (pn && !net_eq(pneigh_net(pn), net));
2564 if (++state->bucket > PNEIGH_HASHMASK)
2566 pn = tbl->phash_buckets[state->bucket];
2567 while (pn && !net_eq(pneigh_net(pn), net))
2579 static struct pneigh_entry *pneigh_get_idx(struct seq_file *seq, loff_t *pos)
2581 struct pneigh_entry *pn = pneigh_get_first(seq);
2586 pn = pneigh_get_next(seq, pn, pos);
2591 return *pos ? NULL : pn;
2594 static void *neigh_get_idx_any(struct seq_file *seq, loff_t *pos)
2596 struct neigh_seq_state *state = seq->private;
2598 loff_t idxpos = *pos;
2600 rc = neigh_get_idx(seq, &idxpos);
2601 if (!rc && !(state->flags & NEIGH_SEQ_NEIGH_ONLY))
2602 rc = pneigh_get_idx(seq, &idxpos);
2607 void *neigh_seq_start(struct seq_file *seq, loff_t *pos, struct neigh_table *tbl, unsigned int neigh_seq_flags)
2610 struct neigh_seq_state *state = seq->private;
2614 state->flags = (neigh_seq_flags & ~NEIGH_SEQ_IS_PNEIGH);
2617 state->nht = rcu_dereference_bh(tbl->nht);
2619 return *pos ? neigh_get_idx_any(seq, pos) : SEQ_START_TOKEN;
2621 EXPORT_SYMBOL(neigh_seq_start);
2623 void *neigh_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2625 struct neigh_seq_state *state;
2628 if (v == SEQ_START_TOKEN) {
2629 rc = neigh_get_first(seq);
2633 state = seq->private;
2634 if (!(state->flags & NEIGH_SEQ_IS_PNEIGH)) {
2635 rc = neigh_get_next(seq, v, NULL);
2638 if (!(state->flags & NEIGH_SEQ_NEIGH_ONLY))
2639 rc = pneigh_get_first(seq);
2641 BUG_ON(state->flags & NEIGH_SEQ_NEIGH_ONLY);
2642 rc = pneigh_get_next(seq, v, NULL);
2648 EXPORT_SYMBOL(neigh_seq_next);
2650 void neigh_seq_stop(struct seq_file *seq, void *v)
2653 rcu_read_unlock_bh();
2655 EXPORT_SYMBOL(neigh_seq_stop);
2657 /* statistics via seq_file */
2659 static void *neigh_stat_seq_start(struct seq_file *seq, loff_t *pos)
2661 struct neigh_table *tbl = seq->private;
2665 return SEQ_START_TOKEN;
2667 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
2668 if (!cpu_possible(cpu))
2671 return per_cpu_ptr(tbl->stats, cpu);
2676 static void *neigh_stat_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2678 struct neigh_table *tbl = seq->private;
2681 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
2682 if (!cpu_possible(cpu))
2685 return per_cpu_ptr(tbl->stats, cpu);
2690 static void neigh_stat_seq_stop(struct seq_file *seq, void *v)
2695 static int neigh_stat_seq_show(struct seq_file *seq, void *v)
2697 struct neigh_table *tbl = seq->private;
2698 struct neigh_statistics *st = v;
2700 if (v == SEQ_START_TOKEN) {
2701 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\n");
2705 seq_printf(seq, "%08x %08lx %08lx %08lx %08lx %08lx %08lx "
2706 "%08lx %08lx %08lx %08lx %08lx\n",
2707 atomic_read(&tbl->entries),
2718 st->rcv_probes_mcast,
2719 st->rcv_probes_ucast,
2721 st->periodic_gc_runs,
2729 static const struct seq_operations neigh_stat_seq_ops = {
2730 .start = neigh_stat_seq_start,
2731 .next = neigh_stat_seq_next,
2732 .stop = neigh_stat_seq_stop,
2733 .show = neigh_stat_seq_show,
2736 static int neigh_stat_seq_open(struct inode *inode, struct file *file)
2738 int ret = seq_open(file, &neigh_stat_seq_ops);
2741 struct seq_file *sf = file->private_data;
2742 sf->private = PDE_DATA(inode);
2747 static const struct file_operations neigh_stat_seq_fops = {
2748 .owner = THIS_MODULE,
2749 .open = neigh_stat_seq_open,
2751 .llseek = seq_lseek,
2752 .release = seq_release,
2755 #endif /* CONFIG_PROC_FS */
2757 static inline size_t neigh_nlmsg_size(void)
2759 return NLMSG_ALIGN(sizeof(struct ndmsg))
2760 + nla_total_size(MAX_ADDR_LEN) /* NDA_DST */
2761 + nla_total_size(MAX_ADDR_LEN) /* NDA_LLADDR */
2762 + nla_total_size(sizeof(struct nda_cacheinfo))
2763 + nla_total_size(4); /* NDA_PROBES */
2766 static void __neigh_notify(struct neighbour *n, int type, int flags)
2768 struct net *net = dev_net(n->dev);
2769 struct sk_buff *skb;
2772 skb = nlmsg_new(neigh_nlmsg_size(), GFP_ATOMIC);
2776 err = neigh_fill_info(skb, n, 0, 0, type, flags);
2778 /* -EMSGSIZE implies BUG in neigh_nlmsg_size() */
2779 WARN_ON(err == -EMSGSIZE);
2783 rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
2787 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
2790 void neigh_app_ns(struct neighbour *n)
2792 __neigh_notify(n, RTM_GETNEIGH, NLM_F_REQUEST);
2794 EXPORT_SYMBOL(neigh_app_ns);
2796 #ifdef CONFIG_SYSCTL
2798 static int int_max = INT_MAX;
2799 static int unres_qlen_max = INT_MAX / SKB_TRUESIZE(ETH_FRAME_LEN);
2801 static int proc_unres_qlen(struct ctl_table *ctl, int write,
2802 void __user *buffer, size_t *lenp, loff_t *ppos)
2805 struct ctl_table tmp = *ctl;
2808 tmp.extra2 = &unres_qlen_max;
2811 size = *(int *)ctl->data / SKB_TRUESIZE(ETH_FRAME_LEN);
2812 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
2815 *(int *)ctl->data = size * SKB_TRUESIZE(ETH_FRAME_LEN);
2819 static struct neigh_parms *neigh_get_dev_parms_rcu(struct net_device *dev,
2824 return __in_dev_arp_parms_get_rcu(dev);
2826 return __in6_dev_nd_parms_get_rcu(dev);
2831 static void neigh_copy_dflt_parms(struct net *net, struct neigh_parms *p,
2834 struct net_device *dev;
2835 int family = neigh_parms_family(p);
2838 for_each_netdev_rcu(net, dev) {
2839 struct neigh_parms *dst_p =
2840 neigh_get_dev_parms_rcu(dev, family);
2842 if (dst_p && !test_bit(index, dst_p->data_state))
2843 dst_p->data[index] = p->data[index];
2848 static void neigh_proc_update(struct ctl_table *ctl, int write)
2850 struct net_device *dev = ctl->extra1;
2851 struct neigh_parms *p = ctl->extra2;
2852 struct net *net = neigh_parms_net(p);
2853 int index = (int *) ctl->data - p->data;
2858 set_bit(index, p->data_state);
2859 if (!dev) /* NULL dev means this is default value */
2860 neigh_copy_dflt_parms(net, p, index);
2863 static int neigh_proc_dointvec_zero_intmax(struct ctl_table *ctl, int write,
2864 void __user *buffer,
2865 size_t *lenp, loff_t *ppos)
2867 struct ctl_table tmp = *ctl;
2871 tmp.extra2 = &int_max;
2873 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
2874 neigh_proc_update(ctl, write);
2878 int neigh_proc_dointvec(struct ctl_table *ctl, int write,
2879 void __user *buffer, size_t *lenp, loff_t *ppos)
2881 int ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
2883 neigh_proc_update(ctl, write);
2886 EXPORT_SYMBOL(neigh_proc_dointvec);
2888 int neigh_proc_dointvec_jiffies(struct ctl_table *ctl, int write,
2889 void __user *buffer,
2890 size_t *lenp, loff_t *ppos)
2892 int ret = proc_dointvec_jiffies(ctl, write, buffer, lenp, ppos);
2894 neigh_proc_update(ctl, write);
2897 EXPORT_SYMBOL(neigh_proc_dointvec_jiffies);
2899 static int neigh_proc_dointvec_userhz_jiffies(struct ctl_table *ctl, int write,
2900 void __user *buffer,
2901 size_t *lenp, loff_t *ppos)
2903 int ret = proc_dointvec_userhz_jiffies(ctl, write, buffer, lenp, ppos);
2905 neigh_proc_update(ctl, write);
2909 int neigh_proc_dointvec_ms_jiffies(struct ctl_table *ctl, int write,
2910 void __user *buffer,
2911 size_t *lenp, loff_t *ppos)
2913 int ret = proc_dointvec_ms_jiffies(ctl, write, buffer, lenp, ppos);
2915 neigh_proc_update(ctl, write);
2918 EXPORT_SYMBOL(neigh_proc_dointvec_ms_jiffies);
2920 static int neigh_proc_dointvec_unres_qlen(struct ctl_table *ctl, int write,
2921 void __user *buffer,
2922 size_t *lenp, loff_t *ppos)
2924 int ret = proc_unres_qlen(ctl, write, buffer, lenp, ppos);
2926 neigh_proc_update(ctl, write);
2930 static int neigh_proc_base_reachable_time(struct ctl_table *ctl, int write,
2931 void __user *buffer,
2932 size_t *lenp, loff_t *ppos)
2934 struct neigh_parms *p = ctl->extra2;
2937 if (strcmp(ctl->procname, "base_reachable_time") == 0)
2938 ret = neigh_proc_dointvec_jiffies(ctl, write, buffer, lenp, ppos);
2939 else if (strcmp(ctl->procname, "base_reachable_time_ms") == 0)
2940 ret = neigh_proc_dointvec_ms_jiffies(ctl, write, buffer, lenp, ppos);
2944 if (write && ret == 0) {
2945 /* update reachable_time as well, otherwise, the change will
2946 * only be effective after the next time neigh_periodic_work
2947 * decides to recompute it
2950 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
2955 #define NEIGH_PARMS_DATA_OFFSET(index) \
2956 (&((struct neigh_parms *) 0)->data[index])
2958 #define NEIGH_SYSCTL_ENTRY(attr, data_attr, name, mval, proc) \
2959 [NEIGH_VAR_ ## attr] = { \
2961 .data = NEIGH_PARMS_DATA_OFFSET(NEIGH_VAR_ ## data_attr), \
2962 .maxlen = sizeof(int), \
2964 .proc_handler = proc, \
2967 #define NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(attr, name) \
2968 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_zero_intmax)
2970 #define NEIGH_SYSCTL_JIFFIES_ENTRY(attr, name) \
2971 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_jiffies)
2973 #define NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(attr, name) \
2974 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_userhz_jiffies)
2976 #define NEIGH_SYSCTL_MS_JIFFIES_ENTRY(attr, name) \
2977 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_ms_jiffies)
2979 #define NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(attr, data_attr, name) \
2980 NEIGH_SYSCTL_ENTRY(attr, data_attr, name, 0644, neigh_proc_dointvec_ms_jiffies)
2982 #define NEIGH_SYSCTL_UNRES_QLEN_REUSED_ENTRY(attr, data_attr, name) \
2983 NEIGH_SYSCTL_ENTRY(attr, data_attr, name, 0644, neigh_proc_dointvec_unres_qlen)
2985 static struct neigh_sysctl_table {
2986 struct ctl_table_header *sysctl_header;
2987 struct ctl_table neigh_vars[NEIGH_VAR_MAX + 1];
2988 } neigh_sysctl_template __read_mostly = {
2990 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(MCAST_PROBES, "mcast_solicit"),
2991 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(UCAST_PROBES, "ucast_solicit"),
2992 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(APP_PROBES, "app_solicit"),
2993 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(RETRANS_TIME, "retrans_time"),
2994 NEIGH_SYSCTL_JIFFIES_ENTRY(BASE_REACHABLE_TIME, "base_reachable_time"),
2995 NEIGH_SYSCTL_JIFFIES_ENTRY(DELAY_PROBE_TIME, "delay_first_probe_time"),
2996 NEIGH_SYSCTL_JIFFIES_ENTRY(GC_STALETIME, "gc_stale_time"),
2997 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(QUEUE_LEN_BYTES, "unres_qlen_bytes"),
2998 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(PROXY_QLEN, "proxy_qlen"),
2999 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(ANYCAST_DELAY, "anycast_delay"),
3000 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(PROXY_DELAY, "proxy_delay"),
3001 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(LOCKTIME, "locktime"),
3002 NEIGH_SYSCTL_UNRES_QLEN_REUSED_ENTRY(QUEUE_LEN, QUEUE_LEN_BYTES, "unres_qlen"),
3003 NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(RETRANS_TIME_MS, RETRANS_TIME, "retrans_time_ms"),
3004 NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(BASE_REACHABLE_TIME_MS, BASE_REACHABLE_TIME, "base_reachable_time_ms"),
3005 [NEIGH_VAR_GC_INTERVAL] = {
3006 .procname = "gc_interval",
3007 .maxlen = sizeof(int),
3009 .proc_handler = proc_dointvec_jiffies,
3011 [NEIGH_VAR_GC_THRESH1] = {
3012 .procname = "gc_thresh1",
3013 .maxlen = sizeof(int),
3017 .proc_handler = proc_dointvec_minmax,
3019 [NEIGH_VAR_GC_THRESH2] = {
3020 .procname = "gc_thresh2",
3021 .maxlen = sizeof(int),
3025 .proc_handler = proc_dointvec_minmax,
3027 [NEIGH_VAR_GC_THRESH3] = {
3028 .procname = "gc_thresh3",
3029 .maxlen = sizeof(int),
3033 .proc_handler = proc_dointvec_minmax,
3039 int neigh_sysctl_register(struct net_device *dev, struct neigh_parms *p,
3040 proc_handler *handler)
3043 struct neigh_sysctl_table *t;
3044 const char *dev_name_source;
3045 char neigh_path[ sizeof("net//neigh/") + IFNAMSIZ + IFNAMSIZ ];
3048 t = kmemdup(&neigh_sysctl_template, sizeof(*t), GFP_KERNEL);
3052 for (i = 0; i < NEIGH_VAR_GC_INTERVAL; i++) {
3053 t->neigh_vars[i].data += (long) p;
3054 t->neigh_vars[i].extra1 = dev;
3055 t->neigh_vars[i].extra2 = p;
3059 dev_name_source = dev->name;
3060 /* Terminate the table early */
3061 memset(&t->neigh_vars[NEIGH_VAR_GC_INTERVAL], 0,
3062 sizeof(t->neigh_vars[NEIGH_VAR_GC_INTERVAL]));
3064 struct neigh_table *tbl = p->tbl;
3065 dev_name_source = "default";
3066 t->neigh_vars[NEIGH_VAR_GC_INTERVAL].data = &tbl->gc_interval;
3067 t->neigh_vars[NEIGH_VAR_GC_THRESH1].data = &tbl->gc_thresh1;
3068 t->neigh_vars[NEIGH_VAR_GC_THRESH2].data = &tbl->gc_thresh2;
3069 t->neigh_vars[NEIGH_VAR_GC_THRESH3].data = &tbl->gc_thresh3;
3074 t->neigh_vars[NEIGH_VAR_RETRANS_TIME].proc_handler = handler;
3076 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].proc_handler = handler;
3077 /* RetransTime (in milliseconds)*/
3078 t->neigh_vars[NEIGH_VAR_RETRANS_TIME_MS].proc_handler = handler;
3079 /* ReachableTime (in milliseconds) */
3080 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].proc_handler = handler;
3082 /* Those handlers will update p->reachable_time after
3083 * base_reachable_time(_ms) is set to ensure the new timer starts being
3084 * applied after the next neighbour update instead of waiting for
3085 * neigh_periodic_work to update its value (can be multiple minutes)
3086 * So any handler that replaces them should do this as well
3089 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].proc_handler =
3090 neigh_proc_base_reachable_time;
3091 /* ReachableTime (in milliseconds) */
3092 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].proc_handler =
3093 neigh_proc_base_reachable_time;
3096 /* Don't export sysctls to unprivileged users */
3097 if (neigh_parms_net(p)->user_ns != &init_user_ns)
3098 t->neigh_vars[0].procname = NULL;
3100 switch (neigh_parms_family(p)) {
3111 snprintf(neigh_path, sizeof(neigh_path), "net/%s/neigh/%s",
3112 p_name, dev_name_source);
3114 register_net_sysctl(neigh_parms_net(p), neigh_path, t->neigh_vars);
3115 if (!t->sysctl_header)
3118 p->sysctl_table = t;
3126 EXPORT_SYMBOL(neigh_sysctl_register);
3128 void neigh_sysctl_unregister(struct neigh_parms *p)
3130 if (p->sysctl_table) {
3131 struct neigh_sysctl_table *t = p->sysctl_table;
3132 p->sysctl_table = NULL;
3133 unregister_net_sysctl_table(t->sysctl_header);
3137 EXPORT_SYMBOL(neigh_sysctl_unregister);
3139 #endif /* CONFIG_SYSCTL */
3141 static int __init neigh_init(void)
3143 rtnl_register(PF_UNSPEC, RTM_NEWNEIGH, neigh_add, NULL, NULL);
3144 rtnl_register(PF_UNSPEC, RTM_DELNEIGH, neigh_delete, NULL, NULL);
3145 rtnl_register(PF_UNSPEC, RTM_GETNEIGH, NULL, neigh_dump_info, NULL);
3147 rtnl_register(PF_UNSPEC, RTM_GETNEIGHTBL, NULL, neightbl_dump_info,
3149 rtnl_register(PF_UNSPEC, RTM_SETNEIGHTBL, neightbl_set, NULL, NULL);
3154 subsys_initcall(neigh_init);