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)
401 NEIGH_CACHE_STAT_INC(tbl, lookups);
404 n = __neigh_lookup_noref(tbl, pkey, dev);
406 if (!atomic_inc_not_zero(&n->refcnt))
408 NEIGH_CACHE_STAT_INC(tbl, hits);
411 rcu_read_unlock_bh();
414 EXPORT_SYMBOL(neigh_lookup);
416 struct neighbour *neigh_lookup_nodev(struct neigh_table *tbl, struct net *net,
420 int key_len = tbl->key_len;
422 struct neigh_hash_table *nht;
424 NEIGH_CACHE_STAT_INC(tbl, lookups);
427 nht = rcu_dereference_bh(tbl->nht);
428 hash_val = tbl->hash(pkey, NULL, nht->hash_rnd) >> (32 - nht->hash_shift);
430 for (n = rcu_dereference_bh(nht->hash_buckets[hash_val]);
432 n = rcu_dereference_bh(n->next)) {
433 if (!memcmp(n->primary_key, pkey, key_len) &&
434 net_eq(dev_net(n->dev), net)) {
435 if (!atomic_inc_not_zero(&n->refcnt))
437 NEIGH_CACHE_STAT_INC(tbl, hits);
442 rcu_read_unlock_bh();
445 EXPORT_SYMBOL(neigh_lookup_nodev);
447 struct neighbour *__neigh_create(struct neigh_table *tbl, const void *pkey,
448 struct net_device *dev, bool want_ref)
451 int key_len = tbl->key_len;
453 struct neighbour *n1, *rc, *n = neigh_alloc(tbl, dev);
454 struct neigh_hash_table *nht;
457 rc = ERR_PTR(-ENOBUFS);
461 memcpy(n->primary_key, pkey, key_len);
465 /* Protocol specific setup. */
466 if (tbl->constructor && (error = tbl->constructor(n)) < 0) {
468 goto out_neigh_release;
471 if (dev->netdev_ops->ndo_neigh_construct) {
472 error = dev->netdev_ops->ndo_neigh_construct(n);
475 goto out_neigh_release;
479 /* Device specific setup. */
480 if (n->parms->neigh_setup &&
481 (error = n->parms->neigh_setup(n)) < 0) {
483 goto out_neigh_release;
486 n->confirmed = jiffies - (NEIGH_VAR(n->parms, BASE_REACHABLE_TIME) << 1);
488 write_lock_bh(&tbl->lock);
489 nht = rcu_dereference_protected(tbl->nht,
490 lockdep_is_held(&tbl->lock));
492 if (atomic_read(&tbl->entries) > (1 << nht->hash_shift))
493 nht = neigh_hash_grow(tbl, nht->hash_shift + 1);
495 hash_val = tbl->hash(pkey, dev, nht->hash_rnd) >> (32 - nht->hash_shift);
497 if (n->parms->dead) {
498 rc = ERR_PTR(-EINVAL);
502 for (n1 = rcu_dereference_protected(nht->hash_buckets[hash_val],
503 lockdep_is_held(&tbl->lock));
505 n1 = rcu_dereference_protected(n1->next,
506 lockdep_is_held(&tbl->lock))) {
507 if (dev == n1->dev && !memcmp(n1->primary_key, pkey, key_len)) {
518 rcu_assign_pointer(n->next,
519 rcu_dereference_protected(nht->hash_buckets[hash_val],
520 lockdep_is_held(&tbl->lock)));
521 rcu_assign_pointer(nht->hash_buckets[hash_val], n);
522 write_unlock_bh(&tbl->lock);
523 neigh_dbg(2, "neigh %p is created\n", n);
528 write_unlock_bh(&tbl->lock);
533 EXPORT_SYMBOL(__neigh_create);
535 static u32 pneigh_hash(const void *pkey, int key_len)
537 u32 hash_val = *(u32 *)(pkey + key_len - 4);
538 hash_val ^= (hash_val >> 16);
539 hash_val ^= hash_val >> 8;
540 hash_val ^= hash_val >> 4;
541 hash_val &= PNEIGH_HASHMASK;
545 static struct pneigh_entry *__pneigh_lookup_1(struct pneigh_entry *n,
549 struct net_device *dev)
552 if (!memcmp(n->key, pkey, key_len) &&
553 net_eq(pneigh_net(n), net) &&
554 (n->dev == dev || !n->dev))
561 struct pneigh_entry *__pneigh_lookup(struct neigh_table *tbl,
562 struct net *net, const void *pkey, struct net_device *dev)
564 int key_len = tbl->key_len;
565 u32 hash_val = pneigh_hash(pkey, key_len);
567 return __pneigh_lookup_1(tbl->phash_buckets[hash_val],
568 net, pkey, key_len, dev);
570 EXPORT_SYMBOL_GPL(__pneigh_lookup);
572 struct pneigh_entry * pneigh_lookup(struct neigh_table *tbl,
573 struct net *net, const void *pkey,
574 struct net_device *dev, int creat)
576 struct pneigh_entry *n;
577 int key_len = tbl->key_len;
578 u32 hash_val = pneigh_hash(pkey, key_len);
580 read_lock_bh(&tbl->lock);
581 n = __pneigh_lookup_1(tbl->phash_buckets[hash_val],
582 net, pkey, key_len, dev);
583 read_unlock_bh(&tbl->lock);
590 n = kmalloc(sizeof(*n) + key_len, GFP_KERNEL);
594 write_pnet(&n->net, net);
595 memcpy(n->key, pkey, key_len);
600 if (tbl->pconstructor && tbl->pconstructor(n)) {
608 write_lock_bh(&tbl->lock);
609 n->next = tbl->phash_buckets[hash_val];
610 tbl->phash_buckets[hash_val] = n;
611 write_unlock_bh(&tbl->lock);
615 EXPORT_SYMBOL(pneigh_lookup);
618 int pneigh_delete(struct neigh_table *tbl, struct net *net, const void *pkey,
619 struct net_device *dev)
621 struct pneigh_entry *n, **np;
622 int key_len = tbl->key_len;
623 u32 hash_val = pneigh_hash(pkey, key_len);
625 write_lock_bh(&tbl->lock);
626 for (np = &tbl->phash_buckets[hash_val]; (n = *np) != NULL;
628 if (!memcmp(n->key, pkey, key_len) && n->dev == dev &&
629 net_eq(pneigh_net(n), net)) {
631 write_unlock_bh(&tbl->lock);
632 if (tbl->pdestructor)
640 write_unlock_bh(&tbl->lock);
644 static int pneigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
646 struct pneigh_entry *n, **np;
649 for (h = 0; h <= PNEIGH_HASHMASK; h++) {
650 np = &tbl->phash_buckets[h];
651 while ((n = *np) != NULL) {
652 if (!dev || n->dev == dev) {
654 if (tbl->pdestructor)
667 static void neigh_parms_destroy(struct neigh_parms *parms);
669 static inline void neigh_parms_put(struct neigh_parms *parms)
671 if (atomic_dec_and_test(&parms->refcnt))
672 neigh_parms_destroy(parms);
676 * neighbour must already be out of the table;
679 void neigh_destroy(struct neighbour *neigh)
681 struct net_device *dev = neigh->dev;
683 NEIGH_CACHE_STAT_INC(neigh->tbl, destroys);
686 pr_warn("Destroying alive neighbour %p\n", neigh);
691 if (neigh_del_timer(neigh))
692 pr_warn("Impossible event\n");
694 write_lock_bh(&neigh->lock);
695 __skb_queue_purge(&neigh->arp_queue);
696 write_unlock_bh(&neigh->lock);
697 neigh->arp_queue_len_bytes = 0;
699 if (dev->netdev_ops->ndo_neigh_destroy)
700 dev->netdev_ops->ndo_neigh_destroy(neigh);
703 neigh_parms_put(neigh->parms);
705 neigh_dbg(2, "neigh %p is destroyed\n", neigh);
707 atomic_dec(&neigh->tbl->entries);
708 kfree_rcu(neigh, rcu);
710 EXPORT_SYMBOL(neigh_destroy);
712 /* Neighbour state is suspicious;
715 Called with write_locked neigh.
717 static void neigh_suspect(struct neighbour *neigh)
719 neigh_dbg(2, "neigh %p is suspected\n", neigh);
721 neigh->output = neigh->ops->output;
724 /* Neighbour state is OK;
727 Called with write_locked neigh.
729 static void neigh_connect(struct neighbour *neigh)
731 neigh_dbg(2, "neigh %p is connected\n", neigh);
733 neigh->output = neigh->ops->connected_output;
736 static void neigh_periodic_work(struct work_struct *work)
738 struct neigh_table *tbl = container_of(work, struct neigh_table, gc_work.work);
740 struct neighbour __rcu **np;
742 struct neigh_hash_table *nht;
744 NEIGH_CACHE_STAT_INC(tbl, periodic_gc_runs);
746 write_lock_bh(&tbl->lock);
747 nht = rcu_dereference_protected(tbl->nht,
748 lockdep_is_held(&tbl->lock));
751 * periodically recompute ReachableTime from random function
754 if (time_after(jiffies, tbl->last_rand + 300 * HZ)) {
755 struct neigh_parms *p;
756 tbl->last_rand = jiffies;
757 list_for_each_entry(p, &tbl->parms_list, list)
759 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
762 if (atomic_read(&tbl->entries) < tbl->gc_thresh1)
765 for (i = 0 ; i < (1 << nht->hash_shift); i++) {
766 np = &nht->hash_buckets[i];
768 while ((n = rcu_dereference_protected(*np,
769 lockdep_is_held(&tbl->lock))) != NULL) {
772 write_lock(&n->lock);
774 state = n->nud_state;
775 if (state & (NUD_PERMANENT | NUD_IN_TIMER)) {
776 write_unlock(&n->lock);
780 if (time_before(n->used, n->confirmed))
781 n->used = n->confirmed;
783 if (atomic_read(&n->refcnt) == 1 &&
784 (state == NUD_FAILED ||
785 time_after(jiffies, n->used + NEIGH_VAR(n->parms, GC_STALETIME)))) {
788 write_unlock(&n->lock);
789 neigh_cleanup_and_release(n);
792 write_unlock(&n->lock);
798 * It's fine to release lock here, even if hash table
799 * grows while we are preempted.
801 write_unlock_bh(&tbl->lock);
803 write_lock_bh(&tbl->lock);
804 nht = rcu_dereference_protected(tbl->nht,
805 lockdep_is_held(&tbl->lock));
808 /* Cycle through all hash buckets every BASE_REACHABLE_TIME/2 ticks.
809 * ARP entry timeouts range from 1/2 BASE_REACHABLE_TIME to 3/2
810 * BASE_REACHABLE_TIME.
812 queue_delayed_work(system_power_efficient_wq, &tbl->gc_work,
813 NEIGH_VAR(&tbl->parms, BASE_REACHABLE_TIME) >> 1);
814 write_unlock_bh(&tbl->lock);
817 static __inline__ int neigh_max_probes(struct neighbour *n)
819 struct neigh_parms *p = n->parms;
820 return NEIGH_VAR(p, UCAST_PROBES) + NEIGH_VAR(p, APP_PROBES) +
821 (n->nud_state & NUD_PROBE ? NEIGH_VAR(p, MCAST_REPROBES) :
822 NEIGH_VAR(p, MCAST_PROBES));
825 static void neigh_invalidate(struct neighbour *neigh)
826 __releases(neigh->lock)
827 __acquires(neigh->lock)
831 NEIGH_CACHE_STAT_INC(neigh->tbl, res_failed);
832 neigh_dbg(2, "neigh %p is failed\n", neigh);
833 neigh->updated = jiffies;
835 /* It is very thin place. report_unreachable is very complicated
836 routine. Particularly, it can hit the same neighbour entry!
838 So that, we try to be accurate and avoid dead loop. --ANK
840 while (neigh->nud_state == NUD_FAILED &&
841 (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
842 write_unlock(&neigh->lock);
843 neigh->ops->error_report(neigh, skb);
844 write_lock(&neigh->lock);
846 __skb_queue_purge(&neigh->arp_queue);
847 neigh->arp_queue_len_bytes = 0;
850 static void neigh_probe(struct neighbour *neigh)
851 __releases(neigh->lock)
853 struct sk_buff *skb = skb_peek_tail(&neigh->arp_queue);
854 /* keep skb alive even if arp_queue overflows */
856 skb = skb_copy(skb, GFP_ATOMIC);
857 write_unlock(&neigh->lock);
858 neigh->ops->solicit(neigh, skb);
859 atomic_inc(&neigh->probes);
863 /* Called when a timer expires for a neighbour entry. */
865 static void neigh_timer_handler(unsigned long arg)
867 unsigned long now, next;
868 struct neighbour *neigh = (struct neighbour *)arg;
872 write_lock(&neigh->lock);
874 state = neigh->nud_state;
878 if (!(state & NUD_IN_TIMER))
881 if (state & NUD_REACHABLE) {
882 if (time_before_eq(now,
883 neigh->confirmed + neigh->parms->reachable_time)) {
884 neigh_dbg(2, "neigh %p is still alive\n", neigh);
885 next = neigh->confirmed + neigh->parms->reachable_time;
886 } else if (time_before_eq(now,
888 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME))) {
889 neigh_dbg(2, "neigh %p is delayed\n", neigh);
890 neigh->nud_state = NUD_DELAY;
891 neigh->updated = jiffies;
892 neigh_suspect(neigh);
893 next = now + NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME);
895 neigh_dbg(2, "neigh %p is suspected\n", neigh);
896 neigh->nud_state = NUD_STALE;
897 neigh->updated = jiffies;
898 neigh_suspect(neigh);
901 } else if (state & NUD_DELAY) {
902 if (time_before_eq(now,
904 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME))) {
905 neigh_dbg(2, "neigh %p is now reachable\n", neigh);
906 neigh->nud_state = NUD_REACHABLE;
907 neigh->updated = jiffies;
908 neigh_connect(neigh);
910 next = neigh->confirmed + neigh->parms->reachable_time;
912 neigh_dbg(2, "neigh %p is probed\n", neigh);
913 neigh->nud_state = NUD_PROBE;
914 neigh->updated = jiffies;
915 atomic_set(&neigh->probes, 0);
916 next = now + NEIGH_VAR(neigh->parms, RETRANS_TIME);
919 /* NUD_PROBE|NUD_INCOMPLETE */
920 next = now + NEIGH_VAR(neigh->parms, RETRANS_TIME);
923 if ((neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) &&
924 atomic_read(&neigh->probes) >= neigh_max_probes(neigh)) {
925 neigh->nud_state = NUD_FAILED;
927 neigh_invalidate(neigh);
931 if (neigh->nud_state & NUD_IN_TIMER) {
932 if (time_before(next, jiffies + HZ/2))
933 next = jiffies + HZ/2;
934 if (!mod_timer(&neigh->timer, next))
937 if (neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) {
941 write_unlock(&neigh->lock);
945 neigh_update_notify(neigh);
947 neigh_release(neigh);
950 int __neigh_event_send(struct neighbour *neigh, struct sk_buff *skb)
953 bool immediate_probe = false;
955 write_lock_bh(&neigh->lock);
958 if (neigh->nud_state & (NUD_CONNECTED | NUD_DELAY | NUD_PROBE))
963 if (!(neigh->nud_state & (NUD_STALE | NUD_INCOMPLETE))) {
964 if (NEIGH_VAR(neigh->parms, MCAST_PROBES) +
965 NEIGH_VAR(neigh->parms, APP_PROBES)) {
966 unsigned long next, now = jiffies;
968 atomic_set(&neigh->probes,
969 NEIGH_VAR(neigh->parms, UCAST_PROBES));
970 neigh->nud_state = NUD_INCOMPLETE;
971 neigh->updated = now;
972 next = now + max(NEIGH_VAR(neigh->parms, RETRANS_TIME),
974 neigh_add_timer(neigh, next);
975 immediate_probe = true;
977 neigh->nud_state = NUD_FAILED;
978 neigh->updated = jiffies;
979 write_unlock_bh(&neigh->lock);
984 } else if (neigh->nud_state & NUD_STALE) {
985 neigh_dbg(2, "neigh %p is delayed\n", neigh);
986 neigh->nud_state = NUD_DELAY;
987 neigh->updated = jiffies;
988 neigh_add_timer(neigh, jiffies +
989 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME));
992 if (neigh->nud_state == NUD_INCOMPLETE) {
994 while (neigh->arp_queue_len_bytes + skb->truesize >
995 NEIGH_VAR(neigh->parms, QUEUE_LEN_BYTES)) {
996 struct sk_buff *buff;
998 buff = __skb_dequeue(&neigh->arp_queue);
1001 neigh->arp_queue_len_bytes -= buff->truesize;
1003 NEIGH_CACHE_STAT_INC(neigh->tbl, unres_discards);
1006 __skb_queue_tail(&neigh->arp_queue, skb);
1007 neigh->arp_queue_len_bytes += skb->truesize;
1012 if (immediate_probe)
1015 write_unlock(&neigh->lock);
1020 if (neigh->nud_state & NUD_STALE)
1022 write_unlock_bh(&neigh->lock);
1026 EXPORT_SYMBOL(__neigh_event_send);
1028 static void neigh_update_hhs(struct neighbour *neigh)
1030 struct hh_cache *hh;
1031 void (*update)(struct hh_cache*, const struct net_device*, const unsigned char *)
1034 if (neigh->dev->header_ops)
1035 update = neigh->dev->header_ops->cache_update;
1040 write_seqlock_bh(&hh->hh_lock);
1041 update(hh, neigh->dev, neigh->ha);
1042 write_sequnlock_bh(&hh->hh_lock);
1049 /* Generic update routine.
1050 -- lladdr is new lladdr or NULL, if it is not supplied.
1051 -- new is new state.
1053 NEIGH_UPDATE_F_OVERRIDE allows to override existing lladdr,
1055 NEIGH_UPDATE_F_WEAK_OVERRIDE will suspect existing "connected"
1056 lladdr instead of overriding it
1058 It also allows to retain current state
1059 if lladdr is unchanged.
1060 NEIGH_UPDATE_F_ADMIN means that the change is administrative.
1062 NEIGH_UPDATE_F_OVERRIDE_ISROUTER allows to override existing
1064 NEIGH_UPDATE_F_ISROUTER indicates if the neighbour is known as
1067 Caller MUST hold reference count on the entry.
1070 int neigh_update(struct neighbour *neigh, const u8 *lladdr, u8 new,
1076 struct net_device *dev;
1077 int update_isrouter = 0;
1079 write_lock_bh(&neigh->lock);
1082 old = neigh->nud_state;
1085 if (!(flags & NEIGH_UPDATE_F_ADMIN) &&
1086 (old & (NUD_NOARP | NUD_PERMANENT)))
1091 if (!(new & NUD_VALID)) {
1092 neigh_del_timer(neigh);
1093 if (old & NUD_CONNECTED)
1094 neigh_suspect(neigh);
1095 neigh->nud_state = new;
1097 notify = old & NUD_VALID;
1098 if ((old & (NUD_INCOMPLETE | NUD_PROBE)) &&
1099 (new & NUD_FAILED)) {
1100 neigh_invalidate(neigh);
1106 /* Compare new lladdr with cached one */
1107 if (!dev->addr_len) {
1108 /* First case: device needs no address. */
1110 } else if (lladdr) {
1111 /* The second case: if something is already cached
1112 and a new address is proposed:
1114 - if they are different, check override flag
1116 if ((old & NUD_VALID) &&
1117 !memcmp(lladdr, neigh->ha, dev->addr_len))
1120 /* No address is supplied; if we know something,
1121 use it, otherwise discard the request.
1124 if (!(old & NUD_VALID))
1129 if (new & NUD_CONNECTED)
1130 neigh->confirmed = jiffies;
1131 neigh->updated = jiffies;
1133 /* If entry was valid and address is not changed,
1134 do not change entry state, if new one is STALE.
1137 update_isrouter = flags & NEIGH_UPDATE_F_OVERRIDE_ISROUTER;
1138 if (old & NUD_VALID) {
1139 if (lladdr != neigh->ha && !(flags & NEIGH_UPDATE_F_OVERRIDE)) {
1140 update_isrouter = 0;
1141 if ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) &&
1142 (old & NUD_CONNECTED)) {
1148 if (lladdr == neigh->ha && new == NUD_STALE &&
1149 ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) ||
1150 (old & NUD_CONNECTED))
1157 neigh_del_timer(neigh);
1158 if (new & NUD_IN_TIMER)
1159 neigh_add_timer(neigh, (jiffies +
1160 ((new & NUD_REACHABLE) ?
1161 neigh->parms->reachable_time :
1163 neigh->nud_state = new;
1167 if (lladdr != neigh->ha) {
1168 write_seqlock(&neigh->ha_lock);
1169 memcpy(&neigh->ha, lladdr, dev->addr_len);
1170 write_sequnlock(&neigh->ha_lock);
1171 neigh_update_hhs(neigh);
1172 if (!(new & NUD_CONNECTED))
1173 neigh->confirmed = jiffies -
1174 (NEIGH_VAR(neigh->parms, BASE_REACHABLE_TIME) << 1);
1179 if (new & NUD_CONNECTED)
1180 neigh_connect(neigh);
1182 neigh_suspect(neigh);
1183 if (!(old & NUD_VALID)) {
1184 struct sk_buff *skb;
1186 /* Again: avoid dead loop if something went wrong */
1188 while (neigh->nud_state & NUD_VALID &&
1189 (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
1190 struct dst_entry *dst = skb_dst(skb);
1191 struct neighbour *n2, *n1 = neigh;
1192 write_unlock_bh(&neigh->lock);
1196 /* Why not just use 'neigh' as-is? The problem is that
1197 * things such as shaper, eql, and sch_teql can end up
1198 * using alternative, different, neigh objects to output
1199 * the packet in the output path. So what we need to do
1200 * here is re-lookup the top-level neigh in the path so
1201 * we can reinject the packet there.
1205 n2 = dst_neigh_lookup_skb(dst, skb);
1209 n1->output(n1, skb);
1214 write_lock_bh(&neigh->lock);
1216 __skb_queue_purge(&neigh->arp_queue);
1217 neigh->arp_queue_len_bytes = 0;
1220 if (update_isrouter) {
1221 neigh->flags = (flags & NEIGH_UPDATE_F_ISROUTER) ?
1222 (neigh->flags | NTF_ROUTER) :
1223 (neigh->flags & ~NTF_ROUTER);
1225 write_unlock_bh(&neigh->lock);
1228 neigh_update_notify(neigh);
1232 EXPORT_SYMBOL(neigh_update);
1234 /* Update the neigh to listen temporarily for probe responses, even if it is
1235 * in a NUD_FAILED state. The caller has to hold neigh->lock for writing.
1237 void __neigh_set_probe_once(struct neighbour *neigh)
1241 neigh->updated = jiffies;
1242 if (!(neigh->nud_state & NUD_FAILED))
1244 neigh->nud_state = NUD_INCOMPLETE;
1245 atomic_set(&neigh->probes, neigh_max_probes(neigh));
1246 neigh_add_timer(neigh,
1247 jiffies + NEIGH_VAR(neigh->parms, RETRANS_TIME));
1249 EXPORT_SYMBOL(__neigh_set_probe_once);
1251 struct neighbour *neigh_event_ns(struct neigh_table *tbl,
1252 u8 *lladdr, void *saddr,
1253 struct net_device *dev)
1255 struct neighbour *neigh = __neigh_lookup(tbl, saddr, dev,
1256 lladdr || !dev->addr_len);
1258 neigh_update(neigh, lladdr, NUD_STALE,
1259 NEIGH_UPDATE_F_OVERRIDE);
1262 EXPORT_SYMBOL(neigh_event_ns);
1264 /* called with read_lock_bh(&n->lock); */
1265 static void neigh_hh_init(struct neighbour *n)
1267 struct net_device *dev = n->dev;
1268 __be16 prot = n->tbl->protocol;
1269 struct hh_cache *hh = &n->hh;
1271 write_lock_bh(&n->lock);
1273 /* Only one thread can come in here and initialize the
1277 dev->header_ops->cache(n, hh, prot);
1279 write_unlock_bh(&n->lock);
1282 /* Slow and careful. */
1284 int neigh_resolve_output(struct neighbour *neigh, struct sk_buff *skb)
1288 if (!neigh_event_send(neigh, skb)) {
1290 struct net_device *dev = neigh->dev;
1293 if (dev->header_ops->cache && !neigh->hh.hh_len)
1294 neigh_hh_init(neigh);
1297 __skb_pull(skb, skb_network_offset(skb));
1298 seq = read_seqbegin(&neigh->ha_lock);
1299 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1300 neigh->ha, NULL, skb->len);
1301 } while (read_seqretry(&neigh->ha_lock, seq));
1304 rc = dev_queue_xmit(skb);
1315 EXPORT_SYMBOL(neigh_resolve_output);
1317 /* As fast as possible without hh cache */
1319 int neigh_connected_output(struct neighbour *neigh, struct sk_buff *skb)
1321 struct net_device *dev = neigh->dev;
1326 __skb_pull(skb, skb_network_offset(skb));
1327 seq = read_seqbegin(&neigh->ha_lock);
1328 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1329 neigh->ha, NULL, skb->len);
1330 } while (read_seqretry(&neigh->ha_lock, seq));
1333 err = dev_queue_xmit(skb);
1340 EXPORT_SYMBOL(neigh_connected_output);
1342 int neigh_direct_output(struct neighbour *neigh, struct sk_buff *skb)
1344 return dev_queue_xmit(skb);
1346 EXPORT_SYMBOL(neigh_direct_output);
1348 static void neigh_proxy_process(unsigned long arg)
1350 struct neigh_table *tbl = (struct neigh_table *)arg;
1351 long sched_next = 0;
1352 unsigned long now = jiffies;
1353 struct sk_buff *skb, *n;
1355 spin_lock(&tbl->proxy_queue.lock);
1357 skb_queue_walk_safe(&tbl->proxy_queue, skb, n) {
1358 long tdif = NEIGH_CB(skb)->sched_next - now;
1361 struct net_device *dev = skb->dev;
1363 __skb_unlink(skb, &tbl->proxy_queue);
1364 if (tbl->proxy_redo && netif_running(dev)) {
1366 tbl->proxy_redo(skb);
1373 } else if (!sched_next || tdif < sched_next)
1376 del_timer(&tbl->proxy_timer);
1378 mod_timer(&tbl->proxy_timer, jiffies + sched_next);
1379 spin_unlock(&tbl->proxy_queue.lock);
1382 void pneigh_enqueue(struct neigh_table *tbl, struct neigh_parms *p,
1383 struct sk_buff *skb)
1385 unsigned long now = jiffies;
1387 unsigned long sched_next = now + (prandom_u32() %
1388 NEIGH_VAR(p, PROXY_DELAY));
1390 if (tbl->proxy_queue.qlen > NEIGH_VAR(p, PROXY_QLEN)) {
1395 NEIGH_CB(skb)->sched_next = sched_next;
1396 NEIGH_CB(skb)->flags |= LOCALLY_ENQUEUED;
1398 spin_lock(&tbl->proxy_queue.lock);
1399 if (del_timer(&tbl->proxy_timer)) {
1400 if (time_before(tbl->proxy_timer.expires, sched_next))
1401 sched_next = tbl->proxy_timer.expires;
1405 __skb_queue_tail(&tbl->proxy_queue, skb);
1406 mod_timer(&tbl->proxy_timer, sched_next);
1407 spin_unlock(&tbl->proxy_queue.lock);
1409 EXPORT_SYMBOL(pneigh_enqueue);
1411 static inline struct neigh_parms *lookup_neigh_parms(struct neigh_table *tbl,
1412 struct net *net, int ifindex)
1414 struct neigh_parms *p;
1416 list_for_each_entry(p, &tbl->parms_list, list) {
1417 if ((p->dev && p->dev->ifindex == ifindex && net_eq(neigh_parms_net(p), net)) ||
1418 (!p->dev && !ifindex && net_eq(net, &init_net)))
1425 struct neigh_parms *neigh_parms_alloc(struct net_device *dev,
1426 struct neigh_table *tbl)
1428 struct neigh_parms *p;
1429 struct net *net = dev_net(dev);
1430 const struct net_device_ops *ops = dev->netdev_ops;
1432 p = kmemdup(&tbl->parms, sizeof(*p), GFP_KERNEL);
1435 atomic_set(&p->refcnt, 1);
1437 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
1440 write_pnet(&p->net, net);
1441 p->sysctl_table = NULL;
1443 if (ops->ndo_neigh_setup && ops->ndo_neigh_setup(dev, p)) {
1449 write_lock_bh(&tbl->lock);
1450 list_add(&p->list, &tbl->parms.list);
1451 write_unlock_bh(&tbl->lock);
1453 neigh_parms_data_state_cleanall(p);
1457 EXPORT_SYMBOL(neigh_parms_alloc);
1459 static void neigh_rcu_free_parms(struct rcu_head *head)
1461 struct neigh_parms *parms =
1462 container_of(head, struct neigh_parms, rcu_head);
1464 neigh_parms_put(parms);
1467 void neigh_parms_release(struct neigh_table *tbl, struct neigh_parms *parms)
1469 if (!parms || parms == &tbl->parms)
1471 write_lock_bh(&tbl->lock);
1472 list_del(&parms->list);
1474 write_unlock_bh(&tbl->lock);
1476 dev_put(parms->dev);
1477 call_rcu(&parms->rcu_head, neigh_rcu_free_parms);
1479 EXPORT_SYMBOL(neigh_parms_release);
1481 static void neigh_parms_destroy(struct neigh_parms *parms)
1486 static struct lock_class_key neigh_table_proxy_queue_class;
1488 static struct neigh_table *neigh_tables[NEIGH_NR_TABLES] __read_mostly;
1490 void neigh_table_init(int index, struct neigh_table *tbl)
1492 unsigned long now = jiffies;
1493 unsigned long phsize;
1495 INIT_LIST_HEAD(&tbl->parms_list);
1496 list_add(&tbl->parms.list, &tbl->parms_list);
1497 write_pnet(&tbl->parms.net, &init_net);
1498 atomic_set(&tbl->parms.refcnt, 1);
1499 tbl->parms.reachable_time =
1500 neigh_rand_reach_time(NEIGH_VAR(&tbl->parms, BASE_REACHABLE_TIME));
1502 tbl->stats = alloc_percpu(struct neigh_statistics);
1504 panic("cannot create neighbour cache statistics");
1506 #ifdef CONFIG_PROC_FS
1507 if (!proc_create_data(tbl->id, 0, init_net.proc_net_stat,
1508 &neigh_stat_seq_fops, tbl))
1509 panic("cannot create neighbour proc dir entry");
1512 RCU_INIT_POINTER(tbl->nht, neigh_hash_alloc(3));
1514 phsize = (PNEIGH_HASHMASK + 1) * sizeof(struct pneigh_entry *);
1515 tbl->phash_buckets = kzalloc(phsize, GFP_KERNEL);
1517 if (!tbl->nht || !tbl->phash_buckets)
1518 panic("cannot allocate neighbour cache hashes");
1520 if (!tbl->entry_size)
1521 tbl->entry_size = ALIGN(offsetof(struct neighbour, primary_key) +
1522 tbl->key_len, NEIGH_PRIV_ALIGN);
1524 WARN_ON(tbl->entry_size % NEIGH_PRIV_ALIGN);
1526 rwlock_init(&tbl->lock);
1527 INIT_DEFERRABLE_WORK(&tbl->gc_work, neigh_periodic_work);
1528 queue_delayed_work(system_power_efficient_wq, &tbl->gc_work,
1529 tbl->parms.reachable_time);
1530 setup_timer(&tbl->proxy_timer, neigh_proxy_process, (unsigned long)tbl);
1531 skb_queue_head_init_class(&tbl->proxy_queue,
1532 &neigh_table_proxy_queue_class);
1534 tbl->last_flush = now;
1535 tbl->last_rand = now + tbl->parms.reachable_time * 20;
1537 neigh_tables[index] = tbl;
1539 EXPORT_SYMBOL(neigh_table_init);
1541 int neigh_table_clear(int index, struct neigh_table *tbl)
1543 neigh_tables[index] = NULL;
1544 /* It is not clean... Fix it to unload IPv6 module safely */
1545 cancel_delayed_work_sync(&tbl->gc_work);
1546 del_timer_sync(&tbl->proxy_timer);
1547 pneigh_queue_purge(&tbl->proxy_queue);
1548 neigh_ifdown(tbl, NULL);
1549 if (atomic_read(&tbl->entries))
1550 pr_crit("neighbour leakage\n");
1552 call_rcu(&rcu_dereference_protected(tbl->nht, 1)->rcu,
1553 neigh_hash_free_rcu);
1556 kfree(tbl->phash_buckets);
1557 tbl->phash_buckets = NULL;
1559 remove_proc_entry(tbl->id, init_net.proc_net_stat);
1561 free_percpu(tbl->stats);
1566 EXPORT_SYMBOL(neigh_table_clear);
1568 static struct neigh_table *neigh_find_table(int family)
1570 struct neigh_table *tbl = NULL;
1574 tbl = neigh_tables[NEIGH_ARP_TABLE];
1577 tbl = neigh_tables[NEIGH_ND_TABLE];
1580 tbl = neigh_tables[NEIGH_DN_TABLE];
1587 static int neigh_delete(struct sk_buff *skb, struct nlmsghdr *nlh)
1589 struct net *net = sock_net(skb->sk);
1591 struct nlattr *dst_attr;
1592 struct neigh_table *tbl;
1593 struct neighbour *neigh;
1594 struct net_device *dev = NULL;
1598 if (nlmsg_len(nlh) < sizeof(*ndm))
1601 dst_attr = nlmsg_find_attr(nlh, sizeof(*ndm), NDA_DST);
1602 if (dst_attr == NULL)
1605 ndm = nlmsg_data(nlh);
1606 if (ndm->ndm_ifindex) {
1607 dev = __dev_get_by_index(net, ndm->ndm_ifindex);
1614 tbl = neigh_find_table(ndm->ndm_family);
1616 return -EAFNOSUPPORT;
1618 if (nla_len(dst_attr) < tbl->key_len)
1621 if (ndm->ndm_flags & NTF_PROXY) {
1622 err = pneigh_delete(tbl, net, nla_data(dst_attr), dev);
1629 neigh = neigh_lookup(tbl, nla_data(dst_attr), dev);
1630 if (neigh == NULL) {
1635 err = neigh_update(neigh, NULL, NUD_FAILED,
1636 NEIGH_UPDATE_F_OVERRIDE |
1637 NEIGH_UPDATE_F_ADMIN);
1638 neigh_release(neigh);
1644 static int neigh_add(struct sk_buff *skb, struct nlmsghdr *nlh)
1646 int flags = NEIGH_UPDATE_F_ADMIN | NEIGH_UPDATE_F_OVERRIDE;
1647 struct net *net = sock_net(skb->sk);
1649 struct nlattr *tb[NDA_MAX+1];
1650 struct neigh_table *tbl;
1651 struct net_device *dev = NULL;
1652 struct neighbour *neigh;
1657 err = nlmsg_parse(nlh, sizeof(*ndm), tb, NDA_MAX, NULL);
1662 if (tb[NDA_DST] == NULL)
1665 ndm = nlmsg_data(nlh);
1666 if (ndm->ndm_ifindex) {
1667 dev = __dev_get_by_index(net, ndm->ndm_ifindex);
1673 if (tb[NDA_LLADDR] && nla_len(tb[NDA_LLADDR]) < dev->addr_len)
1677 tbl = neigh_find_table(ndm->ndm_family);
1679 return -EAFNOSUPPORT;
1681 if (nla_len(tb[NDA_DST]) < tbl->key_len)
1683 dst = nla_data(tb[NDA_DST]);
1684 lladdr = tb[NDA_LLADDR] ? nla_data(tb[NDA_LLADDR]) : NULL;
1686 if (ndm->ndm_flags & NTF_PROXY) {
1687 struct pneigh_entry *pn;
1690 pn = pneigh_lookup(tbl, net, dst, dev, 1);
1692 pn->flags = ndm->ndm_flags;
1701 neigh = neigh_lookup(tbl, dst, dev);
1702 if (neigh == NULL) {
1703 if (!(nlh->nlmsg_flags & NLM_F_CREATE)) {
1708 neigh = __neigh_lookup_errno(tbl, dst, dev);
1709 if (IS_ERR(neigh)) {
1710 err = PTR_ERR(neigh);
1714 if (nlh->nlmsg_flags & NLM_F_EXCL) {
1716 neigh_release(neigh);
1720 if (!(nlh->nlmsg_flags & NLM_F_REPLACE))
1721 flags &= ~NEIGH_UPDATE_F_OVERRIDE;
1724 if (ndm->ndm_flags & NTF_USE) {
1725 neigh_event_send(neigh, NULL);
1728 err = neigh_update(neigh, lladdr, ndm->ndm_state, flags);
1729 neigh_release(neigh);
1735 static int neightbl_fill_parms(struct sk_buff *skb, struct neigh_parms *parms)
1737 struct nlattr *nest;
1739 nest = nla_nest_start(skb, NDTA_PARMS);
1744 nla_put_u32(skb, NDTPA_IFINDEX, parms->dev->ifindex)) ||
1745 nla_put_u32(skb, NDTPA_REFCNT, atomic_read(&parms->refcnt)) ||
1746 nla_put_u32(skb, NDTPA_QUEUE_LENBYTES,
1747 NEIGH_VAR(parms, QUEUE_LEN_BYTES)) ||
1748 /* approximative value for deprecated QUEUE_LEN (in packets) */
1749 nla_put_u32(skb, NDTPA_QUEUE_LEN,
1750 NEIGH_VAR(parms, QUEUE_LEN_BYTES) / SKB_TRUESIZE(ETH_FRAME_LEN)) ||
1751 nla_put_u32(skb, NDTPA_PROXY_QLEN, NEIGH_VAR(parms, PROXY_QLEN)) ||
1752 nla_put_u32(skb, NDTPA_APP_PROBES, NEIGH_VAR(parms, APP_PROBES)) ||
1753 nla_put_u32(skb, NDTPA_UCAST_PROBES,
1754 NEIGH_VAR(parms, UCAST_PROBES)) ||
1755 nla_put_u32(skb, NDTPA_MCAST_PROBES,
1756 NEIGH_VAR(parms, MCAST_PROBES)) ||
1757 nla_put_u32(skb, NDTPA_MCAST_REPROBES,
1758 NEIGH_VAR(parms, MCAST_REPROBES)) ||
1759 nla_put_msecs(skb, NDTPA_REACHABLE_TIME, parms->reachable_time) ||
1760 nla_put_msecs(skb, NDTPA_BASE_REACHABLE_TIME,
1761 NEIGH_VAR(parms, BASE_REACHABLE_TIME)) ||
1762 nla_put_msecs(skb, NDTPA_GC_STALETIME,
1763 NEIGH_VAR(parms, GC_STALETIME)) ||
1764 nla_put_msecs(skb, NDTPA_DELAY_PROBE_TIME,
1765 NEIGH_VAR(parms, DELAY_PROBE_TIME)) ||
1766 nla_put_msecs(skb, NDTPA_RETRANS_TIME,
1767 NEIGH_VAR(parms, RETRANS_TIME)) ||
1768 nla_put_msecs(skb, NDTPA_ANYCAST_DELAY,
1769 NEIGH_VAR(parms, ANYCAST_DELAY)) ||
1770 nla_put_msecs(skb, NDTPA_PROXY_DELAY,
1771 NEIGH_VAR(parms, PROXY_DELAY)) ||
1772 nla_put_msecs(skb, NDTPA_LOCKTIME,
1773 NEIGH_VAR(parms, LOCKTIME)))
1774 goto nla_put_failure;
1775 return nla_nest_end(skb, nest);
1778 nla_nest_cancel(skb, nest);
1782 static int neightbl_fill_info(struct sk_buff *skb, struct neigh_table *tbl,
1783 u32 pid, u32 seq, int type, int flags)
1785 struct nlmsghdr *nlh;
1786 struct ndtmsg *ndtmsg;
1788 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
1792 ndtmsg = nlmsg_data(nlh);
1794 read_lock_bh(&tbl->lock);
1795 ndtmsg->ndtm_family = tbl->family;
1796 ndtmsg->ndtm_pad1 = 0;
1797 ndtmsg->ndtm_pad2 = 0;
1799 if (nla_put_string(skb, NDTA_NAME, tbl->id) ||
1800 nla_put_msecs(skb, NDTA_GC_INTERVAL, tbl->gc_interval) ||
1801 nla_put_u32(skb, NDTA_THRESH1, tbl->gc_thresh1) ||
1802 nla_put_u32(skb, NDTA_THRESH2, tbl->gc_thresh2) ||
1803 nla_put_u32(skb, NDTA_THRESH3, tbl->gc_thresh3))
1804 goto nla_put_failure;
1806 unsigned long now = jiffies;
1807 unsigned int flush_delta = now - tbl->last_flush;
1808 unsigned int rand_delta = now - tbl->last_rand;
1809 struct neigh_hash_table *nht;
1810 struct ndt_config ndc = {
1811 .ndtc_key_len = tbl->key_len,
1812 .ndtc_entry_size = tbl->entry_size,
1813 .ndtc_entries = atomic_read(&tbl->entries),
1814 .ndtc_last_flush = jiffies_to_msecs(flush_delta),
1815 .ndtc_last_rand = jiffies_to_msecs(rand_delta),
1816 .ndtc_proxy_qlen = tbl->proxy_queue.qlen,
1820 nht = rcu_dereference_bh(tbl->nht);
1821 ndc.ndtc_hash_rnd = nht->hash_rnd[0];
1822 ndc.ndtc_hash_mask = ((1 << nht->hash_shift) - 1);
1823 rcu_read_unlock_bh();
1825 if (nla_put(skb, NDTA_CONFIG, sizeof(ndc), &ndc))
1826 goto nla_put_failure;
1831 struct ndt_stats ndst;
1833 memset(&ndst, 0, sizeof(ndst));
1835 for_each_possible_cpu(cpu) {
1836 struct neigh_statistics *st;
1838 st = per_cpu_ptr(tbl->stats, cpu);
1839 ndst.ndts_allocs += st->allocs;
1840 ndst.ndts_destroys += st->destroys;
1841 ndst.ndts_hash_grows += st->hash_grows;
1842 ndst.ndts_res_failed += st->res_failed;
1843 ndst.ndts_lookups += st->lookups;
1844 ndst.ndts_hits += st->hits;
1845 ndst.ndts_rcv_probes_mcast += st->rcv_probes_mcast;
1846 ndst.ndts_rcv_probes_ucast += st->rcv_probes_ucast;
1847 ndst.ndts_periodic_gc_runs += st->periodic_gc_runs;
1848 ndst.ndts_forced_gc_runs += st->forced_gc_runs;
1851 if (nla_put(skb, NDTA_STATS, sizeof(ndst), &ndst))
1852 goto nla_put_failure;
1855 BUG_ON(tbl->parms.dev);
1856 if (neightbl_fill_parms(skb, &tbl->parms) < 0)
1857 goto nla_put_failure;
1859 read_unlock_bh(&tbl->lock);
1860 nlmsg_end(skb, nlh);
1864 read_unlock_bh(&tbl->lock);
1865 nlmsg_cancel(skb, nlh);
1869 static int neightbl_fill_param_info(struct sk_buff *skb,
1870 struct neigh_table *tbl,
1871 struct neigh_parms *parms,
1872 u32 pid, u32 seq, int type,
1875 struct ndtmsg *ndtmsg;
1876 struct nlmsghdr *nlh;
1878 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
1882 ndtmsg = nlmsg_data(nlh);
1884 read_lock_bh(&tbl->lock);
1885 ndtmsg->ndtm_family = tbl->family;
1886 ndtmsg->ndtm_pad1 = 0;
1887 ndtmsg->ndtm_pad2 = 0;
1889 if (nla_put_string(skb, NDTA_NAME, tbl->id) < 0 ||
1890 neightbl_fill_parms(skb, parms) < 0)
1893 read_unlock_bh(&tbl->lock);
1894 nlmsg_end(skb, nlh);
1897 read_unlock_bh(&tbl->lock);
1898 nlmsg_cancel(skb, nlh);
1902 static const struct nla_policy nl_neightbl_policy[NDTA_MAX+1] = {
1903 [NDTA_NAME] = { .type = NLA_STRING },
1904 [NDTA_THRESH1] = { .type = NLA_U32 },
1905 [NDTA_THRESH2] = { .type = NLA_U32 },
1906 [NDTA_THRESH3] = { .type = NLA_U32 },
1907 [NDTA_GC_INTERVAL] = { .type = NLA_U64 },
1908 [NDTA_PARMS] = { .type = NLA_NESTED },
1911 static const struct nla_policy nl_ntbl_parm_policy[NDTPA_MAX+1] = {
1912 [NDTPA_IFINDEX] = { .type = NLA_U32 },
1913 [NDTPA_QUEUE_LEN] = { .type = NLA_U32 },
1914 [NDTPA_PROXY_QLEN] = { .type = NLA_U32 },
1915 [NDTPA_APP_PROBES] = { .type = NLA_U32 },
1916 [NDTPA_UCAST_PROBES] = { .type = NLA_U32 },
1917 [NDTPA_MCAST_PROBES] = { .type = NLA_U32 },
1918 [NDTPA_MCAST_REPROBES] = { .type = NLA_U32 },
1919 [NDTPA_BASE_REACHABLE_TIME] = { .type = NLA_U64 },
1920 [NDTPA_GC_STALETIME] = { .type = NLA_U64 },
1921 [NDTPA_DELAY_PROBE_TIME] = { .type = NLA_U64 },
1922 [NDTPA_RETRANS_TIME] = { .type = NLA_U64 },
1923 [NDTPA_ANYCAST_DELAY] = { .type = NLA_U64 },
1924 [NDTPA_PROXY_DELAY] = { .type = NLA_U64 },
1925 [NDTPA_LOCKTIME] = { .type = NLA_U64 },
1928 static int neightbl_set(struct sk_buff *skb, struct nlmsghdr *nlh)
1930 struct net *net = sock_net(skb->sk);
1931 struct neigh_table *tbl;
1932 struct ndtmsg *ndtmsg;
1933 struct nlattr *tb[NDTA_MAX+1];
1937 err = nlmsg_parse(nlh, sizeof(*ndtmsg), tb, NDTA_MAX,
1938 nl_neightbl_policy);
1942 if (tb[NDTA_NAME] == NULL) {
1947 ndtmsg = nlmsg_data(nlh);
1949 for (tidx = 0; tidx < NEIGH_NR_TABLES; tidx++) {
1950 tbl = neigh_tables[tidx];
1953 if (ndtmsg->ndtm_family && tbl->family != ndtmsg->ndtm_family)
1955 if (nla_strcmp(tb[NDTA_NAME], tbl->id) == 0) {
1965 * We acquire tbl->lock to be nice to the periodic timers and
1966 * make sure they always see a consistent set of values.
1968 write_lock_bh(&tbl->lock);
1970 if (tb[NDTA_PARMS]) {
1971 struct nlattr *tbp[NDTPA_MAX+1];
1972 struct neigh_parms *p;
1975 err = nla_parse_nested(tbp, NDTPA_MAX, tb[NDTA_PARMS],
1976 nl_ntbl_parm_policy);
1978 goto errout_tbl_lock;
1980 if (tbp[NDTPA_IFINDEX])
1981 ifindex = nla_get_u32(tbp[NDTPA_IFINDEX]);
1983 p = lookup_neigh_parms(tbl, net, ifindex);
1986 goto errout_tbl_lock;
1989 for (i = 1; i <= NDTPA_MAX; i++) {
1994 case NDTPA_QUEUE_LEN:
1995 NEIGH_VAR_SET(p, QUEUE_LEN_BYTES,
1996 nla_get_u32(tbp[i]) *
1997 SKB_TRUESIZE(ETH_FRAME_LEN));
1999 case NDTPA_QUEUE_LENBYTES:
2000 NEIGH_VAR_SET(p, QUEUE_LEN_BYTES,
2001 nla_get_u32(tbp[i]));
2003 case NDTPA_PROXY_QLEN:
2004 NEIGH_VAR_SET(p, PROXY_QLEN,
2005 nla_get_u32(tbp[i]));
2007 case NDTPA_APP_PROBES:
2008 NEIGH_VAR_SET(p, APP_PROBES,
2009 nla_get_u32(tbp[i]));
2011 case NDTPA_UCAST_PROBES:
2012 NEIGH_VAR_SET(p, UCAST_PROBES,
2013 nla_get_u32(tbp[i]));
2015 case NDTPA_MCAST_PROBES:
2016 NEIGH_VAR_SET(p, MCAST_PROBES,
2017 nla_get_u32(tbp[i]));
2019 case NDTPA_MCAST_REPROBES:
2020 NEIGH_VAR_SET(p, MCAST_REPROBES,
2021 nla_get_u32(tbp[i]));
2023 case NDTPA_BASE_REACHABLE_TIME:
2024 NEIGH_VAR_SET(p, BASE_REACHABLE_TIME,
2025 nla_get_msecs(tbp[i]));
2026 /* update reachable_time as well, otherwise, the change will
2027 * only be effective after the next time neigh_periodic_work
2028 * decides to recompute it (can be multiple minutes)
2031 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
2033 case NDTPA_GC_STALETIME:
2034 NEIGH_VAR_SET(p, GC_STALETIME,
2035 nla_get_msecs(tbp[i]));
2037 case NDTPA_DELAY_PROBE_TIME:
2038 NEIGH_VAR_SET(p, DELAY_PROBE_TIME,
2039 nla_get_msecs(tbp[i]));
2041 case NDTPA_RETRANS_TIME:
2042 NEIGH_VAR_SET(p, RETRANS_TIME,
2043 nla_get_msecs(tbp[i]));
2045 case NDTPA_ANYCAST_DELAY:
2046 NEIGH_VAR_SET(p, ANYCAST_DELAY,
2047 nla_get_msecs(tbp[i]));
2049 case NDTPA_PROXY_DELAY:
2050 NEIGH_VAR_SET(p, PROXY_DELAY,
2051 nla_get_msecs(tbp[i]));
2053 case NDTPA_LOCKTIME:
2054 NEIGH_VAR_SET(p, LOCKTIME,
2055 nla_get_msecs(tbp[i]));
2062 if ((tb[NDTA_THRESH1] || tb[NDTA_THRESH2] ||
2063 tb[NDTA_THRESH3] || tb[NDTA_GC_INTERVAL]) &&
2064 !net_eq(net, &init_net))
2065 goto errout_tbl_lock;
2067 if (tb[NDTA_THRESH1])
2068 tbl->gc_thresh1 = nla_get_u32(tb[NDTA_THRESH1]);
2070 if (tb[NDTA_THRESH2])
2071 tbl->gc_thresh2 = nla_get_u32(tb[NDTA_THRESH2]);
2073 if (tb[NDTA_THRESH3])
2074 tbl->gc_thresh3 = nla_get_u32(tb[NDTA_THRESH3]);
2076 if (tb[NDTA_GC_INTERVAL])
2077 tbl->gc_interval = nla_get_msecs(tb[NDTA_GC_INTERVAL]);
2082 write_unlock_bh(&tbl->lock);
2087 static int neightbl_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2089 struct net *net = sock_net(skb->sk);
2090 int family, tidx, nidx = 0;
2091 int tbl_skip = cb->args[0];
2092 int neigh_skip = cb->args[1];
2093 struct neigh_table *tbl;
2095 family = ((struct rtgenmsg *) nlmsg_data(cb->nlh))->rtgen_family;
2097 for (tidx = 0; tidx < NEIGH_NR_TABLES; tidx++) {
2098 struct neigh_parms *p;
2100 tbl = neigh_tables[tidx];
2104 if (tidx < tbl_skip || (family && tbl->family != family))
2107 if (neightbl_fill_info(skb, tbl, NETLINK_CB(cb->skb).portid,
2108 cb->nlh->nlmsg_seq, RTM_NEWNEIGHTBL,
2113 p = list_next_entry(&tbl->parms, list);
2114 list_for_each_entry_from(p, &tbl->parms_list, list) {
2115 if (!net_eq(neigh_parms_net(p), net))
2118 if (nidx < neigh_skip)
2121 if (neightbl_fill_param_info(skb, tbl, p,
2122 NETLINK_CB(cb->skb).portid,
2140 static int neigh_fill_info(struct sk_buff *skb, struct neighbour *neigh,
2141 u32 pid, u32 seq, int type, unsigned int flags)
2143 unsigned long now = jiffies;
2144 struct nda_cacheinfo ci;
2145 struct nlmsghdr *nlh;
2148 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2152 ndm = nlmsg_data(nlh);
2153 ndm->ndm_family = neigh->ops->family;
2156 ndm->ndm_flags = neigh->flags;
2157 ndm->ndm_type = neigh->type;
2158 ndm->ndm_ifindex = neigh->dev->ifindex;
2160 if (nla_put(skb, NDA_DST, neigh->tbl->key_len, neigh->primary_key))
2161 goto nla_put_failure;
2163 read_lock_bh(&neigh->lock);
2164 ndm->ndm_state = neigh->nud_state;
2165 if (neigh->nud_state & NUD_VALID) {
2166 char haddr[MAX_ADDR_LEN];
2168 neigh_ha_snapshot(haddr, neigh, neigh->dev);
2169 if (nla_put(skb, NDA_LLADDR, neigh->dev->addr_len, haddr) < 0) {
2170 read_unlock_bh(&neigh->lock);
2171 goto nla_put_failure;
2175 ci.ndm_used = jiffies_to_clock_t(now - neigh->used);
2176 ci.ndm_confirmed = jiffies_to_clock_t(now - neigh->confirmed);
2177 ci.ndm_updated = jiffies_to_clock_t(now - neigh->updated);
2178 ci.ndm_refcnt = atomic_read(&neigh->refcnt) - 1;
2179 read_unlock_bh(&neigh->lock);
2181 if (nla_put_u32(skb, NDA_PROBES, atomic_read(&neigh->probes)) ||
2182 nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
2183 goto nla_put_failure;
2185 nlmsg_end(skb, nlh);
2189 nlmsg_cancel(skb, nlh);
2193 static int pneigh_fill_info(struct sk_buff *skb, struct pneigh_entry *pn,
2194 u32 pid, u32 seq, int type, unsigned int flags,
2195 struct neigh_table *tbl)
2197 struct nlmsghdr *nlh;
2200 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2204 ndm = nlmsg_data(nlh);
2205 ndm->ndm_family = tbl->family;
2208 ndm->ndm_flags = pn->flags | NTF_PROXY;
2209 ndm->ndm_type = RTN_UNICAST;
2210 ndm->ndm_ifindex = pn->dev->ifindex;
2211 ndm->ndm_state = NUD_NONE;
2213 if (nla_put(skb, NDA_DST, tbl->key_len, pn->key))
2214 goto nla_put_failure;
2216 nlmsg_end(skb, nlh);
2220 nlmsg_cancel(skb, nlh);
2224 static void neigh_update_notify(struct neighbour *neigh)
2226 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, neigh);
2227 __neigh_notify(neigh, RTM_NEWNEIGH, 0);
2230 static int neigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2231 struct netlink_callback *cb)
2233 struct net *net = sock_net(skb->sk);
2234 struct neighbour *n;
2235 int rc, h, s_h = cb->args[1];
2236 int idx, s_idx = idx = cb->args[2];
2237 struct neigh_hash_table *nht;
2240 nht = rcu_dereference_bh(tbl->nht);
2242 for (h = s_h; h < (1 << nht->hash_shift); h++) {
2245 for (n = rcu_dereference_bh(nht->hash_buckets[h]), idx = 0;
2247 n = rcu_dereference_bh(n->next)) {
2248 if (!net_eq(dev_net(n->dev), net))
2252 if (neigh_fill_info(skb, n, NETLINK_CB(cb->skb).portid,
2265 rcu_read_unlock_bh();
2271 static int pneigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2272 struct netlink_callback *cb)
2274 struct pneigh_entry *n;
2275 struct net *net = sock_net(skb->sk);
2276 int rc, h, s_h = cb->args[3];
2277 int idx, s_idx = idx = cb->args[4];
2279 read_lock_bh(&tbl->lock);
2281 for (h = s_h; h <= PNEIGH_HASHMASK; h++) {
2284 for (n = tbl->phash_buckets[h], idx = 0; n; n = n->next) {
2285 if (dev_net(n->dev) != net)
2289 if (pneigh_fill_info(skb, n, NETLINK_CB(cb->skb).portid,
2292 NLM_F_MULTI, tbl) < 0) {
2293 read_unlock_bh(&tbl->lock);
2302 read_unlock_bh(&tbl->lock);
2311 static int neigh_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2313 struct neigh_table *tbl;
2318 family = ((struct rtgenmsg *) nlmsg_data(cb->nlh))->rtgen_family;
2320 /* check for full ndmsg structure presence, family member is
2321 * the same for both structures
2323 if (nlmsg_len(cb->nlh) >= sizeof(struct ndmsg) &&
2324 ((struct ndmsg *) nlmsg_data(cb->nlh))->ndm_flags == NTF_PROXY)
2329 for (t = 0; t < NEIGH_NR_TABLES; t++) {
2330 tbl = neigh_tables[t];
2334 if (t < s_t || (family && tbl->family != family))
2337 memset(&cb->args[1], 0, sizeof(cb->args) -
2338 sizeof(cb->args[0]));
2340 err = pneigh_dump_table(tbl, skb, cb);
2342 err = neigh_dump_table(tbl, skb, cb);
2351 void neigh_for_each(struct neigh_table *tbl, void (*cb)(struct neighbour *, void *), void *cookie)
2354 struct neigh_hash_table *nht;
2357 nht = rcu_dereference_bh(tbl->nht);
2359 read_lock(&tbl->lock); /* avoid resizes */
2360 for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
2361 struct neighbour *n;
2363 for (n = rcu_dereference_bh(nht->hash_buckets[chain]);
2365 n = rcu_dereference_bh(n->next))
2368 read_unlock(&tbl->lock);
2369 rcu_read_unlock_bh();
2371 EXPORT_SYMBOL(neigh_for_each);
2373 /* The tbl->lock must be held as a writer and BH disabled. */
2374 void __neigh_for_each_release(struct neigh_table *tbl,
2375 int (*cb)(struct neighbour *))
2378 struct neigh_hash_table *nht;
2380 nht = rcu_dereference_protected(tbl->nht,
2381 lockdep_is_held(&tbl->lock));
2382 for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
2383 struct neighbour *n;
2384 struct neighbour __rcu **np;
2386 np = &nht->hash_buckets[chain];
2387 while ((n = rcu_dereference_protected(*np,
2388 lockdep_is_held(&tbl->lock))) != NULL) {
2391 write_lock(&n->lock);
2394 rcu_assign_pointer(*np,
2395 rcu_dereference_protected(n->next,
2396 lockdep_is_held(&tbl->lock)));
2400 write_unlock(&n->lock);
2402 neigh_cleanup_and_release(n);
2406 EXPORT_SYMBOL(__neigh_for_each_release);
2408 int neigh_xmit(int index, struct net_device *dev,
2409 const void *addr, struct sk_buff *skb)
2411 int err = -EAFNOSUPPORT;
2412 if (likely(index < NEIGH_NR_TABLES)) {
2413 struct neigh_table *tbl;
2414 struct neighbour *neigh;
2416 tbl = neigh_tables[index];
2419 neigh = __neigh_lookup_noref(tbl, addr, dev);
2421 neigh = __neigh_create(tbl, addr, dev, false);
2422 err = PTR_ERR(neigh);
2425 err = neigh->output(neigh, skb);
2427 else if (index == NEIGH_LINK_TABLE) {
2428 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
2429 addr, NULL, skb->len);
2432 err = dev_queue_xmit(skb);
2440 EXPORT_SYMBOL(neigh_xmit);
2442 #ifdef CONFIG_PROC_FS
2444 static struct neighbour *neigh_get_first(struct seq_file *seq)
2446 struct neigh_seq_state *state = seq->private;
2447 struct net *net = seq_file_net(seq);
2448 struct neigh_hash_table *nht = state->nht;
2449 struct neighbour *n = NULL;
2450 int bucket = state->bucket;
2452 state->flags &= ~NEIGH_SEQ_IS_PNEIGH;
2453 for (bucket = 0; bucket < (1 << nht->hash_shift); bucket++) {
2454 n = rcu_dereference_bh(nht->hash_buckets[bucket]);
2457 if (!net_eq(dev_net(n->dev), net))
2459 if (state->neigh_sub_iter) {
2463 v = state->neigh_sub_iter(state, n, &fakep);
2467 if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
2469 if (n->nud_state & ~NUD_NOARP)
2472 n = rcu_dereference_bh(n->next);
2478 state->bucket = bucket;
2483 static struct neighbour *neigh_get_next(struct seq_file *seq,
2484 struct neighbour *n,
2487 struct neigh_seq_state *state = seq->private;
2488 struct net *net = seq_file_net(seq);
2489 struct neigh_hash_table *nht = state->nht;
2491 if (state->neigh_sub_iter) {
2492 void *v = state->neigh_sub_iter(state, n, pos);
2496 n = rcu_dereference_bh(n->next);
2500 if (!net_eq(dev_net(n->dev), net))
2502 if (state->neigh_sub_iter) {
2503 void *v = state->neigh_sub_iter(state, n, pos);
2508 if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
2511 if (n->nud_state & ~NUD_NOARP)
2514 n = rcu_dereference_bh(n->next);
2520 if (++state->bucket >= (1 << nht->hash_shift))
2523 n = rcu_dereference_bh(nht->hash_buckets[state->bucket]);
2531 static struct neighbour *neigh_get_idx(struct seq_file *seq, loff_t *pos)
2533 struct neighbour *n = neigh_get_first(seq);
2538 n = neigh_get_next(seq, n, pos);
2543 return *pos ? NULL : n;
2546 static struct pneigh_entry *pneigh_get_first(struct seq_file *seq)
2548 struct neigh_seq_state *state = seq->private;
2549 struct net *net = seq_file_net(seq);
2550 struct neigh_table *tbl = state->tbl;
2551 struct pneigh_entry *pn = NULL;
2552 int bucket = state->bucket;
2554 state->flags |= NEIGH_SEQ_IS_PNEIGH;
2555 for (bucket = 0; bucket <= PNEIGH_HASHMASK; bucket++) {
2556 pn = tbl->phash_buckets[bucket];
2557 while (pn && !net_eq(pneigh_net(pn), net))
2562 state->bucket = bucket;
2567 static struct pneigh_entry *pneigh_get_next(struct seq_file *seq,
2568 struct pneigh_entry *pn,
2571 struct neigh_seq_state *state = seq->private;
2572 struct net *net = seq_file_net(seq);
2573 struct neigh_table *tbl = state->tbl;
2577 } while (pn && !net_eq(pneigh_net(pn), net));
2580 if (++state->bucket > PNEIGH_HASHMASK)
2582 pn = tbl->phash_buckets[state->bucket];
2583 while (pn && !net_eq(pneigh_net(pn), net))
2595 static struct pneigh_entry *pneigh_get_idx(struct seq_file *seq, loff_t *pos)
2597 struct pneigh_entry *pn = pneigh_get_first(seq);
2602 pn = pneigh_get_next(seq, pn, pos);
2607 return *pos ? NULL : pn;
2610 static void *neigh_get_idx_any(struct seq_file *seq, loff_t *pos)
2612 struct neigh_seq_state *state = seq->private;
2614 loff_t idxpos = *pos;
2616 rc = neigh_get_idx(seq, &idxpos);
2617 if (!rc && !(state->flags & NEIGH_SEQ_NEIGH_ONLY))
2618 rc = pneigh_get_idx(seq, &idxpos);
2623 void *neigh_seq_start(struct seq_file *seq, loff_t *pos, struct neigh_table *tbl, unsigned int neigh_seq_flags)
2626 struct neigh_seq_state *state = seq->private;
2630 state->flags = (neigh_seq_flags & ~NEIGH_SEQ_IS_PNEIGH);
2633 state->nht = rcu_dereference_bh(tbl->nht);
2635 return *pos ? neigh_get_idx_any(seq, pos) : SEQ_START_TOKEN;
2637 EXPORT_SYMBOL(neigh_seq_start);
2639 void *neigh_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2641 struct neigh_seq_state *state;
2644 if (v == SEQ_START_TOKEN) {
2645 rc = neigh_get_first(seq);
2649 state = seq->private;
2650 if (!(state->flags & NEIGH_SEQ_IS_PNEIGH)) {
2651 rc = neigh_get_next(seq, v, NULL);
2654 if (!(state->flags & NEIGH_SEQ_NEIGH_ONLY))
2655 rc = pneigh_get_first(seq);
2657 BUG_ON(state->flags & NEIGH_SEQ_NEIGH_ONLY);
2658 rc = pneigh_get_next(seq, v, NULL);
2664 EXPORT_SYMBOL(neigh_seq_next);
2666 void neigh_seq_stop(struct seq_file *seq, void *v)
2669 rcu_read_unlock_bh();
2671 EXPORT_SYMBOL(neigh_seq_stop);
2673 /* statistics via seq_file */
2675 static void *neigh_stat_seq_start(struct seq_file *seq, loff_t *pos)
2677 struct neigh_table *tbl = seq->private;
2681 return SEQ_START_TOKEN;
2683 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
2684 if (!cpu_possible(cpu))
2687 return per_cpu_ptr(tbl->stats, cpu);
2692 static void *neigh_stat_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2694 struct neigh_table *tbl = seq->private;
2697 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
2698 if (!cpu_possible(cpu))
2701 return per_cpu_ptr(tbl->stats, cpu);
2706 static void neigh_stat_seq_stop(struct seq_file *seq, void *v)
2711 static int neigh_stat_seq_show(struct seq_file *seq, void *v)
2713 struct neigh_table *tbl = seq->private;
2714 struct neigh_statistics *st = v;
2716 if (v == SEQ_START_TOKEN) {
2717 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");
2721 seq_printf(seq, "%08x %08lx %08lx %08lx %08lx %08lx %08lx "
2722 "%08lx %08lx %08lx %08lx %08lx\n",
2723 atomic_read(&tbl->entries),
2734 st->rcv_probes_mcast,
2735 st->rcv_probes_ucast,
2737 st->periodic_gc_runs,
2745 static const struct seq_operations neigh_stat_seq_ops = {
2746 .start = neigh_stat_seq_start,
2747 .next = neigh_stat_seq_next,
2748 .stop = neigh_stat_seq_stop,
2749 .show = neigh_stat_seq_show,
2752 static int neigh_stat_seq_open(struct inode *inode, struct file *file)
2754 int ret = seq_open(file, &neigh_stat_seq_ops);
2757 struct seq_file *sf = file->private_data;
2758 sf->private = PDE_DATA(inode);
2763 static const struct file_operations neigh_stat_seq_fops = {
2764 .owner = THIS_MODULE,
2765 .open = neigh_stat_seq_open,
2767 .llseek = seq_lseek,
2768 .release = seq_release,
2771 #endif /* CONFIG_PROC_FS */
2773 static inline size_t neigh_nlmsg_size(void)
2775 return NLMSG_ALIGN(sizeof(struct ndmsg))
2776 + nla_total_size(MAX_ADDR_LEN) /* NDA_DST */
2777 + nla_total_size(MAX_ADDR_LEN) /* NDA_LLADDR */
2778 + nla_total_size(sizeof(struct nda_cacheinfo))
2779 + nla_total_size(4); /* NDA_PROBES */
2782 static void __neigh_notify(struct neighbour *n, int type, int flags)
2784 struct net *net = dev_net(n->dev);
2785 struct sk_buff *skb;
2788 skb = nlmsg_new(neigh_nlmsg_size(), GFP_ATOMIC);
2792 err = neigh_fill_info(skb, n, 0, 0, type, flags);
2794 /* -EMSGSIZE implies BUG in neigh_nlmsg_size() */
2795 WARN_ON(err == -EMSGSIZE);
2799 rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
2803 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
2806 void neigh_app_ns(struct neighbour *n)
2808 __neigh_notify(n, RTM_GETNEIGH, NLM_F_REQUEST);
2810 EXPORT_SYMBOL(neigh_app_ns);
2812 #ifdef CONFIG_SYSCTL
2814 static int int_max = INT_MAX;
2815 static int unres_qlen_max = INT_MAX / SKB_TRUESIZE(ETH_FRAME_LEN);
2817 static int proc_unres_qlen(struct ctl_table *ctl, int write,
2818 void __user *buffer, size_t *lenp, loff_t *ppos)
2821 struct ctl_table tmp = *ctl;
2824 tmp.extra2 = &unres_qlen_max;
2827 size = *(int *)ctl->data / SKB_TRUESIZE(ETH_FRAME_LEN);
2828 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
2831 *(int *)ctl->data = size * SKB_TRUESIZE(ETH_FRAME_LEN);
2835 static struct neigh_parms *neigh_get_dev_parms_rcu(struct net_device *dev,
2840 return __in_dev_arp_parms_get_rcu(dev);
2842 return __in6_dev_nd_parms_get_rcu(dev);
2847 static void neigh_copy_dflt_parms(struct net *net, struct neigh_parms *p,
2850 struct net_device *dev;
2851 int family = neigh_parms_family(p);
2854 for_each_netdev_rcu(net, dev) {
2855 struct neigh_parms *dst_p =
2856 neigh_get_dev_parms_rcu(dev, family);
2858 if (dst_p && !test_bit(index, dst_p->data_state))
2859 dst_p->data[index] = p->data[index];
2864 static void neigh_proc_update(struct ctl_table *ctl, int write)
2866 struct net_device *dev = ctl->extra1;
2867 struct neigh_parms *p = ctl->extra2;
2868 struct net *net = neigh_parms_net(p);
2869 int index = (int *) ctl->data - p->data;
2874 set_bit(index, p->data_state);
2875 if (!dev) /* NULL dev means this is default value */
2876 neigh_copy_dflt_parms(net, p, index);
2879 static int neigh_proc_dointvec_zero_intmax(struct ctl_table *ctl, int write,
2880 void __user *buffer,
2881 size_t *lenp, loff_t *ppos)
2883 struct ctl_table tmp = *ctl;
2887 tmp.extra2 = &int_max;
2889 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
2890 neigh_proc_update(ctl, write);
2894 int neigh_proc_dointvec(struct ctl_table *ctl, int write,
2895 void __user *buffer, size_t *lenp, loff_t *ppos)
2897 int ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
2899 neigh_proc_update(ctl, write);
2902 EXPORT_SYMBOL(neigh_proc_dointvec);
2904 int neigh_proc_dointvec_jiffies(struct ctl_table *ctl, int write,
2905 void __user *buffer,
2906 size_t *lenp, loff_t *ppos)
2908 int ret = proc_dointvec_jiffies(ctl, write, buffer, lenp, ppos);
2910 neigh_proc_update(ctl, write);
2913 EXPORT_SYMBOL(neigh_proc_dointvec_jiffies);
2915 static int neigh_proc_dointvec_userhz_jiffies(struct ctl_table *ctl, int write,
2916 void __user *buffer,
2917 size_t *lenp, loff_t *ppos)
2919 int ret = proc_dointvec_userhz_jiffies(ctl, write, buffer, lenp, ppos);
2921 neigh_proc_update(ctl, write);
2925 int neigh_proc_dointvec_ms_jiffies(struct ctl_table *ctl, int write,
2926 void __user *buffer,
2927 size_t *lenp, loff_t *ppos)
2929 int ret = proc_dointvec_ms_jiffies(ctl, write, buffer, lenp, ppos);
2931 neigh_proc_update(ctl, write);
2934 EXPORT_SYMBOL(neigh_proc_dointvec_ms_jiffies);
2936 static int neigh_proc_dointvec_unres_qlen(struct ctl_table *ctl, int write,
2937 void __user *buffer,
2938 size_t *lenp, loff_t *ppos)
2940 int ret = proc_unres_qlen(ctl, write, buffer, lenp, ppos);
2942 neigh_proc_update(ctl, write);
2946 static int neigh_proc_base_reachable_time(struct ctl_table *ctl, int write,
2947 void __user *buffer,
2948 size_t *lenp, loff_t *ppos)
2950 struct neigh_parms *p = ctl->extra2;
2953 if (strcmp(ctl->procname, "base_reachable_time") == 0)
2954 ret = neigh_proc_dointvec_jiffies(ctl, write, buffer, lenp, ppos);
2955 else if (strcmp(ctl->procname, "base_reachable_time_ms") == 0)
2956 ret = neigh_proc_dointvec_ms_jiffies(ctl, write, buffer, lenp, ppos);
2960 if (write && ret == 0) {
2961 /* update reachable_time as well, otherwise, the change will
2962 * only be effective after the next time neigh_periodic_work
2963 * decides to recompute it
2966 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
2971 #define NEIGH_PARMS_DATA_OFFSET(index) \
2972 (&((struct neigh_parms *) 0)->data[index])
2974 #define NEIGH_SYSCTL_ENTRY(attr, data_attr, name, mval, proc) \
2975 [NEIGH_VAR_ ## attr] = { \
2977 .data = NEIGH_PARMS_DATA_OFFSET(NEIGH_VAR_ ## data_attr), \
2978 .maxlen = sizeof(int), \
2980 .proc_handler = proc, \
2983 #define NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(attr, name) \
2984 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_zero_intmax)
2986 #define NEIGH_SYSCTL_JIFFIES_ENTRY(attr, name) \
2987 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_jiffies)
2989 #define NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(attr, name) \
2990 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_userhz_jiffies)
2992 #define NEIGH_SYSCTL_MS_JIFFIES_ENTRY(attr, name) \
2993 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_ms_jiffies)
2995 #define NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(attr, data_attr, name) \
2996 NEIGH_SYSCTL_ENTRY(attr, data_attr, name, 0644, neigh_proc_dointvec_ms_jiffies)
2998 #define NEIGH_SYSCTL_UNRES_QLEN_REUSED_ENTRY(attr, data_attr, name) \
2999 NEIGH_SYSCTL_ENTRY(attr, data_attr, name, 0644, neigh_proc_dointvec_unres_qlen)
3001 static struct neigh_sysctl_table {
3002 struct ctl_table_header *sysctl_header;
3003 struct ctl_table neigh_vars[NEIGH_VAR_MAX + 1];
3004 } neigh_sysctl_template __read_mostly = {
3006 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(MCAST_PROBES, "mcast_solicit"),
3007 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(UCAST_PROBES, "ucast_solicit"),
3008 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(APP_PROBES, "app_solicit"),
3009 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(MCAST_REPROBES, "mcast_resolicit"),
3010 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(RETRANS_TIME, "retrans_time"),
3011 NEIGH_SYSCTL_JIFFIES_ENTRY(BASE_REACHABLE_TIME, "base_reachable_time"),
3012 NEIGH_SYSCTL_JIFFIES_ENTRY(DELAY_PROBE_TIME, "delay_first_probe_time"),
3013 NEIGH_SYSCTL_JIFFIES_ENTRY(GC_STALETIME, "gc_stale_time"),
3014 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(QUEUE_LEN_BYTES, "unres_qlen_bytes"),
3015 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(PROXY_QLEN, "proxy_qlen"),
3016 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(ANYCAST_DELAY, "anycast_delay"),
3017 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(PROXY_DELAY, "proxy_delay"),
3018 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(LOCKTIME, "locktime"),
3019 NEIGH_SYSCTL_UNRES_QLEN_REUSED_ENTRY(QUEUE_LEN, QUEUE_LEN_BYTES, "unres_qlen"),
3020 NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(RETRANS_TIME_MS, RETRANS_TIME, "retrans_time_ms"),
3021 NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(BASE_REACHABLE_TIME_MS, BASE_REACHABLE_TIME, "base_reachable_time_ms"),
3022 [NEIGH_VAR_GC_INTERVAL] = {
3023 .procname = "gc_interval",
3024 .maxlen = sizeof(int),
3026 .proc_handler = proc_dointvec_jiffies,
3028 [NEIGH_VAR_GC_THRESH1] = {
3029 .procname = "gc_thresh1",
3030 .maxlen = sizeof(int),
3034 .proc_handler = proc_dointvec_minmax,
3036 [NEIGH_VAR_GC_THRESH2] = {
3037 .procname = "gc_thresh2",
3038 .maxlen = sizeof(int),
3042 .proc_handler = proc_dointvec_minmax,
3044 [NEIGH_VAR_GC_THRESH3] = {
3045 .procname = "gc_thresh3",
3046 .maxlen = sizeof(int),
3050 .proc_handler = proc_dointvec_minmax,
3056 int neigh_sysctl_register(struct net_device *dev, struct neigh_parms *p,
3057 proc_handler *handler)
3060 struct neigh_sysctl_table *t;
3061 const char *dev_name_source;
3062 char neigh_path[ sizeof("net//neigh/") + IFNAMSIZ + IFNAMSIZ ];
3065 t = kmemdup(&neigh_sysctl_template, sizeof(*t), GFP_KERNEL);
3069 for (i = 0; i < NEIGH_VAR_GC_INTERVAL; i++) {
3070 t->neigh_vars[i].data += (long) p;
3071 t->neigh_vars[i].extra1 = dev;
3072 t->neigh_vars[i].extra2 = p;
3076 dev_name_source = dev->name;
3077 /* Terminate the table early */
3078 memset(&t->neigh_vars[NEIGH_VAR_GC_INTERVAL], 0,
3079 sizeof(t->neigh_vars[NEIGH_VAR_GC_INTERVAL]));
3081 struct neigh_table *tbl = p->tbl;
3082 dev_name_source = "default";
3083 t->neigh_vars[NEIGH_VAR_GC_INTERVAL].data = &tbl->gc_interval;
3084 t->neigh_vars[NEIGH_VAR_GC_THRESH1].data = &tbl->gc_thresh1;
3085 t->neigh_vars[NEIGH_VAR_GC_THRESH2].data = &tbl->gc_thresh2;
3086 t->neigh_vars[NEIGH_VAR_GC_THRESH3].data = &tbl->gc_thresh3;
3091 t->neigh_vars[NEIGH_VAR_RETRANS_TIME].proc_handler = handler;
3093 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].proc_handler = handler;
3094 /* RetransTime (in milliseconds)*/
3095 t->neigh_vars[NEIGH_VAR_RETRANS_TIME_MS].proc_handler = handler;
3096 /* ReachableTime (in milliseconds) */
3097 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].proc_handler = handler;
3099 /* Those handlers will update p->reachable_time after
3100 * base_reachable_time(_ms) is set to ensure the new timer starts being
3101 * applied after the next neighbour update instead of waiting for
3102 * neigh_periodic_work to update its value (can be multiple minutes)
3103 * So any handler that replaces them should do this as well
3106 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].proc_handler =
3107 neigh_proc_base_reachable_time;
3108 /* ReachableTime (in milliseconds) */
3109 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].proc_handler =
3110 neigh_proc_base_reachable_time;
3113 /* Don't export sysctls to unprivileged users */
3114 if (neigh_parms_net(p)->user_ns != &init_user_ns)
3115 t->neigh_vars[0].procname = NULL;
3117 switch (neigh_parms_family(p)) {
3128 snprintf(neigh_path, sizeof(neigh_path), "net/%s/neigh/%s",
3129 p_name, dev_name_source);
3131 register_net_sysctl(neigh_parms_net(p), neigh_path, t->neigh_vars);
3132 if (!t->sysctl_header)
3135 p->sysctl_table = t;
3143 EXPORT_SYMBOL(neigh_sysctl_register);
3145 void neigh_sysctl_unregister(struct neigh_parms *p)
3147 if (p->sysctl_table) {
3148 struct neigh_sysctl_table *t = p->sysctl_table;
3149 p->sysctl_table = NULL;
3150 unregister_net_sysctl_table(t->sysctl_header);
3154 EXPORT_SYMBOL(neigh_sysctl_unregister);
3156 #endif /* CONFIG_SYSCTL */
3158 static int __init neigh_init(void)
3160 rtnl_register(PF_UNSPEC, RTM_NEWNEIGH, neigh_add, NULL, NULL);
3161 rtnl_register(PF_UNSPEC, RTM_DELNEIGH, neigh_delete, NULL, NULL);
3162 rtnl_register(PF_UNSPEC, RTM_GETNEIGH, NULL, neigh_dump_info, NULL);
3164 rtnl_register(PF_UNSPEC, RTM_GETNEIGHTBL, NULL, neightbl_dump_info,
3166 rtnl_register(PF_UNSPEC, RTM_SETNEIGHTBL, neightbl_set, NULL, NULL);
3171 subsys_initcall(neigh_init);